]> git.proxmox.com Git - mirror_ubuntu-hirsute-kernel.git/blob - net/core/dev.c
[IPVS]: Use skb_forward_csum
[mirror_ubuntu-hirsute-kernel.git] / net / core / dev.c
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/notifier.h>
94 #include <linux/skbuff.h>
95 #include <net/sock.h>
96 #include <linux/rtnetlink.h>
97 #include <linux/proc_fs.h>
98 #include <linux/seq_file.h>
99 #include <linux/stat.h>
100 #include <linux/if_bridge.h>
101 #include <linux/if_macvlan.h>
102 #include <net/dst.h>
103 #include <net/pkt_sched.h>
104 #include <net/checksum.h>
105 #include <linux/highmem.h>
106 #include <linux/init.h>
107 #include <linux/kmod.h>
108 #include <linux/module.h>
109 #include <linux/kallsyms.h>
110 #include <linux/netpoll.h>
111 #include <linux/rcupdate.h>
112 #include <linux/delay.h>
113 #include <net/wext.h>
114 #include <net/iw_handler.h>
115 #include <asm/current.h>
116 #include <linux/audit.h>
117 #include <linux/dmaengine.h>
118 #include <linux/err.h>
119 #include <linux/ctype.h>
120 #include <linux/if_arp.h>
121
122 /*
123 * The list of packet types we will receive (as opposed to discard)
124 * and the routines to invoke.
125 *
126 * Why 16. Because with 16 the only overlap we get on a hash of the
127 * low nibble of the protocol value is RARP/SNAP/X.25.
128 *
129 * NOTE: That is no longer true with the addition of VLAN tags. Not
130 * sure which should go first, but I bet it won't make much
131 * difference if we are running VLANs. The good news is that
132 * this protocol won't be in the list unless compiled in, so
133 * the average user (w/out VLANs) will not be adversely affected.
134 * --BLG
135 *
136 * 0800 IP
137 * 8100 802.1Q VLAN
138 * 0001 802.3
139 * 0002 AX.25
140 * 0004 802.2
141 * 8035 RARP
142 * 0005 SNAP
143 * 0805 X.25
144 * 0806 ARP
145 * 8137 IPX
146 * 0009 Localtalk
147 * 86DD IPv6
148 */
149
150 static DEFINE_SPINLOCK(ptype_lock);
151 static struct list_head ptype_base[16] __read_mostly; /* 16 way hashed list */
152 static struct list_head ptype_all __read_mostly; /* Taps */
153
154 #ifdef CONFIG_NET_DMA
155 struct net_dma {
156 struct dma_client client;
157 spinlock_t lock;
158 cpumask_t channel_mask;
159 struct dma_chan *channels[NR_CPUS];
160 };
161
162 static enum dma_state_client
163 netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
164 enum dma_state state);
165
166 static struct net_dma net_dma = {
167 .client = {
168 .event_callback = netdev_dma_event,
169 },
170 };
171 #endif
172
173 /*
174 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
175 * semaphore.
176 *
177 * Pure readers hold dev_base_lock for reading.
178 *
179 * Writers must hold the rtnl semaphore while they loop through the
180 * dev_base_head list, and hold dev_base_lock for writing when they do the
181 * actual updates. This allows pure readers to access the list even
182 * while a writer is preparing to update it.
183 *
184 * To put it another way, dev_base_lock is held for writing only to
185 * protect against pure readers; the rtnl semaphore provides the
186 * protection against other writers.
187 *
188 * See, for example usages, register_netdevice() and
189 * unregister_netdevice(), which must be called with the rtnl
190 * semaphore held.
191 */
192 LIST_HEAD(dev_base_head);
193 DEFINE_RWLOCK(dev_base_lock);
194
195 EXPORT_SYMBOL(dev_base_head);
196 EXPORT_SYMBOL(dev_base_lock);
197
198 #define NETDEV_HASHBITS 8
199 static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
200 static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
201
202 static inline struct hlist_head *dev_name_hash(const char *name)
203 {
204 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
205 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
206 }
207
208 static inline struct hlist_head *dev_index_hash(int ifindex)
209 {
210 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
211 }
212
213 /*
214 * Our notifier list
215 */
216
217 static RAW_NOTIFIER_HEAD(netdev_chain);
218
219 /*
220 * Device drivers call our routines to queue packets here. We empty the
221 * queue in the local softnet handler.
222 */
223 DEFINE_PER_CPU(struct softnet_data, softnet_data) = { NULL };
224
225 #ifdef CONFIG_SYSFS
226 extern int netdev_sysfs_init(void);
227 extern int netdev_register_sysfs(struct net_device *);
228 extern void netdev_unregister_sysfs(struct net_device *);
229 #else
230 #define netdev_sysfs_init() (0)
231 #define netdev_register_sysfs(dev) (0)
232 #define netdev_unregister_sysfs(dev) do { } while(0)
233 #endif
234
235 #ifdef CONFIG_DEBUG_LOCK_ALLOC
236 /*
237 * register_netdevice() inits dev->_xmit_lock and sets lockdep class
238 * according to dev->type
239 */
240 static const unsigned short netdev_lock_type[] =
241 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
242 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
243 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
244 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
245 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
246 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
247 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
248 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
249 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
250 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
251 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
252 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
253 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
254 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_VOID,
255 ARPHRD_NONE};
256
257 static const char *netdev_lock_name[] =
258 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
259 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
260 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
261 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
262 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
263 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
264 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
265 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
266 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
267 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
268 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
269 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
270 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
271 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_VOID",
272 "_xmit_NONE"};
273
274 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
275
276 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
277 {
278 int i;
279
280 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
281 if (netdev_lock_type[i] == dev_type)
282 return i;
283 /* the last key is used by default */
284 return ARRAY_SIZE(netdev_lock_type) - 1;
285 }
286
287 static inline void netdev_set_lockdep_class(spinlock_t *lock,
288 unsigned short dev_type)
289 {
290 int i;
291
292 i = netdev_lock_pos(dev_type);
293 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
294 netdev_lock_name[i]);
295 }
296 #else
297 static inline void netdev_set_lockdep_class(spinlock_t *lock,
298 unsigned short dev_type)
299 {
300 }
301 #endif
302
303 /*******************************************************************************
304
305 Protocol management and registration routines
306
307 *******************************************************************************/
308
309 /*
310 * Add a protocol ID to the list. Now that the input handler is
311 * smarter we can dispense with all the messy stuff that used to be
312 * here.
313 *
314 * BEWARE!!! Protocol handlers, mangling input packets,
315 * MUST BE last in hash buckets and checking protocol handlers
316 * MUST start from promiscuous ptype_all chain in net_bh.
317 * It is true now, do not change it.
318 * Explanation follows: if protocol handler, mangling packet, will
319 * be the first on list, it is not able to sense, that packet
320 * is cloned and should be copied-on-write, so that it will
321 * change it and subsequent readers will get broken packet.
322 * --ANK (980803)
323 */
324
325 /**
326 * dev_add_pack - add packet handler
327 * @pt: packet type declaration
328 *
329 * Add a protocol handler to the networking stack. The passed &packet_type
330 * is linked into kernel lists and may not be freed until it has been
331 * removed from the kernel lists.
332 *
333 * This call does not sleep therefore it can not
334 * guarantee all CPU's that are in middle of receiving packets
335 * will see the new packet type (until the next received packet).
336 */
337
338 void dev_add_pack(struct packet_type *pt)
339 {
340 int hash;
341
342 spin_lock_bh(&ptype_lock);
343 if (pt->type == htons(ETH_P_ALL))
344 list_add_rcu(&pt->list, &ptype_all);
345 else {
346 hash = ntohs(pt->type) & 15;
347 list_add_rcu(&pt->list, &ptype_base[hash]);
348 }
349 spin_unlock_bh(&ptype_lock);
350 }
351
352 /**
353 * __dev_remove_pack - remove packet handler
354 * @pt: packet type declaration
355 *
356 * Remove a protocol handler that was previously added to the kernel
357 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
358 * from the kernel lists and can be freed or reused once this function
359 * returns.
360 *
361 * The packet type might still be in use by receivers
362 * and must not be freed until after all the CPU's have gone
363 * through a quiescent state.
364 */
365 void __dev_remove_pack(struct packet_type *pt)
366 {
367 struct list_head *head;
368 struct packet_type *pt1;
369
370 spin_lock_bh(&ptype_lock);
371
372 if (pt->type == htons(ETH_P_ALL))
373 head = &ptype_all;
374 else
375 head = &ptype_base[ntohs(pt->type) & 15];
376
377 list_for_each_entry(pt1, head, list) {
378 if (pt == pt1) {
379 list_del_rcu(&pt->list);
380 goto out;
381 }
382 }
383
384 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
385 out:
386 spin_unlock_bh(&ptype_lock);
387 }
388 /**
389 * dev_remove_pack - remove packet handler
390 * @pt: packet type declaration
391 *
392 * Remove a protocol handler that was previously added to the kernel
393 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
394 * from the kernel lists and can be freed or reused once this function
395 * returns.
396 *
397 * This call sleeps to guarantee that no CPU is looking at the packet
398 * type after return.
399 */
400 void dev_remove_pack(struct packet_type *pt)
401 {
402 __dev_remove_pack(pt);
403
404 synchronize_net();
405 }
406
407 /******************************************************************************
408
409 Device Boot-time Settings Routines
410
411 *******************************************************************************/
412
413 /* Boot time configuration table */
414 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
415
416 /**
417 * netdev_boot_setup_add - add new setup entry
418 * @name: name of the device
419 * @map: configured settings for the device
420 *
421 * Adds new setup entry to the dev_boot_setup list. The function
422 * returns 0 on error and 1 on success. This is a generic routine to
423 * all netdevices.
424 */
425 static int netdev_boot_setup_add(char *name, struct ifmap *map)
426 {
427 struct netdev_boot_setup *s;
428 int i;
429
430 s = dev_boot_setup;
431 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
432 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
433 memset(s[i].name, 0, sizeof(s[i].name));
434 strcpy(s[i].name, name);
435 memcpy(&s[i].map, map, sizeof(s[i].map));
436 break;
437 }
438 }
439
440 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
441 }
442
443 /**
444 * netdev_boot_setup_check - check boot time settings
445 * @dev: the netdevice
446 *
447 * Check boot time settings for the device.
448 * The found settings are set for the device to be used
449 * later in the device probing.
450 * Returns 0 if no settings found, 1 if they are.
451 */
452 int netdev_boot_setup_check(struct net_device *dev)
453 {
454 struct netdev_boot_setup *s = dev_boot_setup;
455 int i;
456
457 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
458 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
459 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
460 dev->irq = s[i].map.irq;
461 dev->base_addr = s[i].map.base_addr;
462 dev->mem_start = s[i].map.mem_start;
463 dev->mem_end = s[i].map.mem_end;
464 return 1;
465 }
466 }
467 return 0;
468 }
469
470
471 /**
472 * netdev_boot_base - get address from boot time settings
473 * @prefix: prefix for network device
474 * @unit: id for network device
475 *
476 * Check boot time settings for the base address of device.
477 * The found settings are set for the device to be used
478 * later in the device probing.
479 * Returns 0 if no settings found.
480 */
481 unsigned long netdev_boot_base(const char *prefix, int unit)
482 {
483 const struct netdev_boot_setup *s = dev_boot_setup;
484 char name[IFNAMSIZ];
485 int i;
486
487 sprintf(name, "%s%d", prefix, unit);
488
489 /*
490 * If device already registered then return base of 1
491 * to indicate not to probe for this interface
492 */
493 if (__dev_get_by_name(name))
494 return 1;
495
496 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
497 if (!strcmp(name, s[i].name))
498 return s[i].map.base_addr;
499 return 0;
500 }
501
502 /*
503 * Saves at boot time configured settings for any netdevice.
504 */
505 int __init netdev_boot_setup(char *str)
506 {
507 int ints[5];
508 struct ifmap map;
509
510 str = get_options(str, ARRAY_SIZE(ints), ints);
511 if (!str || !*str)
512 return 0;
513
514 /* Save settings */
515 memset(&map, 0, sizeof(map));
516 if (ints[0] > 0)
517 map.irq = ints[1];
518 if (ints[0] > 1)
519 map.base_addr = ints[2];
520 if (ints[0] > 2)
521 map.mem_start = ints[3];
522 if (ints[0] > 3)
523 map.mem_end = ints[4];
524
525 /* Add new entry to the list */
526 return netdev_boot_setup_add(str, &map);
527 }
528
529 __setup("netdev=", netdev_boot_setup);
530
531 /*******************************************************************************
532
533 Device Interface Subroutines
534
535 *******************************************************************************/
536
537 /**
538 * __dev_get_by_name - find a device by its name
539 * @name: name to find
540 *
541 * Find an interface by name. Must be called under RTNL semaphore
542 * or @dev_base_lock. If the name is found a pointer to the device
543 * is returned. If the name is not found then %NULL is returned. The
544 * reference counters are not incremented so the caller must be
545 * careful with locks.
546 */
547
548 struct net_device *__dev_get_by_name(const char *name)
549 {
550 struct hlist_node *p;
551
552 hlist_for_each(p, dev_name_hash(name)) {
553 struct net_device *dev
554 = hlist_entry(p, struct net_device, name_hlist);
555 if (!strncmp(dev->name, name, IFNAMSIZ))
556 return dev;
557 }
558 return NULL;
559 }
560
561 /**
562 * dev_get_by_name - find a device by its name
563 * @name: name to find
564 *
565 * Find an interface by name. This can be called from any
566 * context and does its own locking. The returned handle has
567 * the usage count incremented and the caller must use dev_put() to
568 * release it when it is no longer needed. %NULL is returned if no
569 * matching device is found.
570 */
571
572 struct net_device *dev_get_by_name(const char *name)
573 {
574 struct net_device *dev;
575
576 read_lock(&dev_base_lock);
577 dev = __dev_get_by_name(name);
578 if (dev)
579 dev_hold(dev);
580 read_unlock(&dev_base_lock);
581 return dev;
582 }
583
584 /**
585 * __dev_get_by_index - find a device by its ifindex
586 * @ifindex: index of device
587 *
588 * Search for an interface by index. Returns %NULL if the device
589 * is not found or a pointer to the device. The device has not
590 * had its reference counter increased so the caller must be careful
591 * about locking. The caller must hold either the RTNL semaphore
592 * or @dev_base_lock.
593 */
594
595 struct net_device *__dev_get_by_index(int ifindex)
596 {
597 struct hlist_node *p;
598
599 hlist_for_each(p, dev_index_hash(ifindex)) {
600 struct net_device *dev
601 = hlist_entry(p, struct net_device, index_hlist);
602 if (dev->ifindex == ifindex)
603 return dev;
604 }
605 return NULL;
606 }
607
608
609 /**
610 * dev_get_by_index - find a device by its ifindex
611 * @ifindex: index of device
612 *
613 * Search for an interface by index. Returns NULL if the device
614 * is not found or a pointer to the device. The device returned has
615 * had a reference added and the pointer is safe until the user calls
616 * dev_put to indicate they have finished with it.
617 */
618
619 struct net_device *dev_get_by_index(int ifindex)
620 {
621 struct net_device *dev;
622
623 read_lock(&dev_base_lock);
624 dev = __dev_get_by_index(ifindex);
625 if (dev)
626 dev_hold(dev);
627 read_unlock(&dev_base_lock);
628 return dev;
629 }
630
631 /**
632 * dev_getbyhwaddr - find a device by its hardware address
633 * @type: media type of device
634 * @ha: hardware address
635 *
636 * Search for an interface by MAC address. Returns NULL if the device
637 * is not found or a pointer to the device. The caller must hold the
638 * rtnl semaphore. The returned device has not had its ref count increased
639 * and the caller must therefore be careful about locking
640 *
641 * BUGS:
642 * If the API was consistent this would be __dev_get_by_hwaddr
643 */
644
645 struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
646 {
647 struct net_device *dev;
648
649 ASSERT_RTNL();
650
651 for_each_netdev(dev)
652 if (dev->type == type &&
653 !memcmp(dev->dev_addr, ha, dev->addr_len))
654 return dev;
655
656 return NULL;
657 }
658
659 EXPORT_SYMBOL(dev_getbyhwaddr);
660
661 struct net_device *__dev_getfirstbyhwtype(unsigned short type)
662 {
663 struct net_device *dev;
664
665 ASSERT_RTNL();
666 for_each_netdev(dev)
667 if (dev->type == type)
668 return dev;
669
670 return NULL;
671 }
672
673 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
674
675 struct net_device *dev_getfirstbyhwtype(unsigned short type)
676 {
677 struct net_device *dev;
678
679 rtnl_lock();
680 dev = __dev_getfirstbyhwtype(type);
681 if (dev)
682 dev_hold(dev);
683 rtnl_unlock();
684 return dev;
685 }
686
687 EXPORT_SYMBOL(dev_getfirstbyhwtype);
688
689 /**
690 * dev_get_by_flags - find any device with given flags
691 * @if_flags: IFF_* values
692 * @mask: bitmask of bits in if_flags to check
693 *
694 * Search for any interface with the given flags. Returns NULL if a device
695 * is not found or a pointer to the device. The device returned has
696 * had a reference added and the pointer is safe until the user calls
697 * dev_put to indicate they have finished with it.
698 */
699
700 struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
701 {
702 struct net_device *dev, *ret;
703
704 ret = NULL;
705 read_lock(&dev_base_lock);
706 for_each_netdev(dev) {
707 if (((dev->flags ^ if_flags) & mask) == 0) {
708 dev_hold(dev);
709 ret = dev;
710 break;
711 }
712 }
713 read_unlock(&dev_base_lock);
714 return ret;
715 }
716
717 /**
718 * dev_valid_name - check if name is okay for network device
719 * @name: name string
720 *
721 * Network device names need to be valid file names to
722 * to allow sysfs to work. We also disallow any kind of
723 * whitespace.
724 */
725 int dev_valid_name(const char *name)
726 {
727 if (*name == '\0')
728 return 0;
729 if (strlen(name) >= IFNAMSIZ)
730 return 0;
731 if (!strcmp(name, ".") || !strcmp(name, ".."))
732 return 0;
733
734 while (*name) {
735 if (*name == '/' || isspace(*name))
736 return 0;
737 name++;
738 }
739 return 1;
740 }
741
742 /**
743 * dev_alloc_name - allocate a name for a device
744 * @dev: device
745 * @name: name format string
746 *
747 * Passed a format string - eg "lt%d" it will try and find a suitable
748 * id. It scans list of devices to build up a free map, then chooses
749 * the first empty slot. The caller must hold the dev_base or rtnl lock
750 * while allocating the name and adding the device in order to avoid
751 * duplicates.
752 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
753 * Returns the number of the unit assigned or a negative errno code.
754 */
755
756 int dev_alloc_name(struct net_device *dev, const char *name)
757 {
758 int i = 0;
759 char buf[IFNAMSIZ];
760 const char *p;
761 const int max_netdevices = 8*PAGE_SIZE;
762 long *inuse;
763 struct net_device *d;
764
765 p = strnchr(name, IFNAMSIZ-1, '%');
766 if (p) {
767 /*
768 * Verify the string as this thing may have come from
769 * the user. There must be either one "%d" and no other "%"
770 * characters.
771 */
772 if (p[1] != 'd' || strchr(p + 2, '%'))
773 return -EINVAL;
774
775 /* Use one page as a bit array of possible slots */
776 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
777 if (!inuse)
778 return -ENOMEM;
779
780 for_each_netdev(d) {
781 if (!sscanf(d->name, name, &i))
782 continue;
783 if (i < 0 || i >= max_netdevices)
784 continue;
785
786 /* avoid cases where sscanf is not exact inverse of printf */
787 snprintf(buf, sizeof(buf), name, i);
788 if (!strncmp(buf, d->name, IFNAMSIZ))
789 set_bit(i, inuse);
790 }
791
792 i = find_first_zero_bit(inuse, max_netdevices);
793 free_page((unsigned long) inuse);
794 }
795
796 snprintf(buf, sizeof(buf), name, i);
797 if (!__dev_get_by_name(buf)) {
798 strlcpy(dev->name, buf, IFNAMSIZ);
799 return i;
800 }
801
802 /* It is possible to run out of possible slots
803 * when the name is long and there isn't enough space left
804 * for the digits, or if all bits are used.
805 */
806 return -ENFILE;
807 }
808
809
810 /**
811 * dev_change_name - change name of a device
812 * @dev: device
813 * @newname: name (or format string) must be at least IFNAMSIZ
814 *
815 * Change name of a device, can pass format strings "eth%d".
816 * for wildcarding.
817 */
818 int dev_change_name(struct net_device *dev, char *newname)
819 {
820 int err = 0;
821
822 ASSERT_RTNL();
823
824 if (dev->flags & IFF_UP)
825 return -EBUSY;
826
827 if (!dev_valid_name(newname))
828 return -EINVAL;
829
830 if (strchr(newname, '%')) {
831 err = dev_alloc_name(dev, newname);
832 if (err < 0)
833 return err;
834 strcpy(newname, dev->name);
835 }
836 else if (__dev_get_by_name(newname))
837 return -EEXIST;
838 else
839 strlcpy(dev->name, newname, IFNAMSIZ);
840
841 device_rename(&dev->dev, dev->name);
842 hlist_del(&dev->name_hlist);
843 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
844 raw_notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
845
846 return err;
847 }
848
849 /**
850 * netdev_features_change - device changes features
851 * @dev: device to cause notification
852 *
853 * Called to indicate a device has changed features.
854 */
855 void netdev_features_change(struct net_device *dev)
856 {
857 raw_notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
858 }
859 EXPORT_SYMBOL(netdev_features_change);
860
861 /**
862 * netdev_state_change - device changes state
863 * @dev: device to cause notification
864 *
865 * Called to indicate a device has changed state. This function calls
866 * the notifier chains for netdev_chain and sends a NEWLINK message
867 * to the routing socket.
868 */
869 void netdev_state_change(struct net_device *dev)
870 {
871 if (dev->flags & IFF_UP) {
872 raw_notifier_call_chain(&netdev_chain,
873 NETDEV_CHANGE, dev);
874 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
875 }
876 }
877
878 /**
879 * dev_load - load a network module
880 * @name: name of interface
881 *
882 * If a network interface is not present and the process has suitable
883 * privileges this function loads the module. If module loading is not
884 * available in this kernel then it becomes a nop.
885 */
886
887 void dev_load(const char *name)
888 {
889 struct net_device *dev;
890
891 read_lock(&dev_base_lock);
892 dev = __dev_get_by_name(name);
893 read_unlock(&dev_base_lock);
894
895 if (!dev && capable(CAP_SYS_MODULE))
896 request_module("%s", name);
897 }
898
899 static int default_rebuild_header(struct sk_buff *skb)
900 {
901 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
902 skb->dev ? skb->dev->name : "NULL!!!");
903 kfree_skb(skb);
904 return 1;
905 }
906
907 /**
908 * dev_open - prepare an interface for use.
909 * @dev: device to open
910 *
911 * Takes a device from down to up state. The device's private open
912 * function is invoked and then the multicast lists are loaded. Finally
913 * the device is moved into the up state and a %NETDEV_UP message is
914 * sent to the netdev notifier chain.
915 *
916 * Calling this function on an active interface is a nop. On a failure
917 * a negative errno code is returned.
918 */
919 int dev_open(struct net_device *dev)
920 {
921 int ret = 0;
922
923 /*
924 * Is it already up?
925 */
926
927 if (dev->flags & IFF_UP)
928 return 0;
929
930 /*
931 * Is it even present?
932 */
933 if (!netif_device_present(dev))
934 return -ENODEV;
935
936 /*
937 * Call device private open method
938 */
939 set_bit(__LINK_STATE_START, &dev->state);
940 if (dev->open) {
941 ret = dev->open(dev);
942 if (ret)
943 clear_bit(__LINK_STATE_START, &dev->state);
944 }
945
946 /*
947 * If it went open OK then:
948 */
949
950 if (!ret) {
951 /*
952 * Set the flags.
953 */
954 dev->flags |= IFF_UP;
955
956 /*
957 * Initialize multicasting status
958 */
959 dev_set_rx_mode(dev);
960
961 /*
962 * Wakeup transmit queue engine
963 */
964 dev_activate(dev);
965
966 /*
967 * ... and announce new interface.
968 */
969 raw_notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
970 }
971 return ret;
972 }
973
974 /**
975 * dev_close - shutdown an interface.
976 * @dev: device to shutdown
977 *
978 * This function moves an active device into down state. A
979 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
980 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
981 * chain.
982 */
983 int dev_close(struct net_device *dev)
984 {
985 if (!(dev->flags & IFF_UP))
986 return 0;
987
988 /*
989 * Tell people we are going down, so that they can
990 * prepare to death, when device is still operating.
991 */
992 raw_notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
993
994 dev_deactivate(dev);
995
996 clear_bit(__LINK_STATE_START, &dev->state);
997
998 /* Synchronize to scheduled poll. We cannot touch poll list,
999 * it can be even on different cpu. So just clear netif_running(),
1000 * and wait when poll really will happen. Actually, the best place
1001 * for this is inside dev->stop() after device stopped its irq
1002 * engine, but this requires more changes in devices. */
1003
1004 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1005 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
1006 /* No hurry. */
1007 msleep(1);
1008 }
1009
1010 /*
1011 * Call the device specific close. This cannot fail.
1012 * Only if device is UP
1013 *
1014 * We allow it to be called even after a DETACH hot-plug
1015 * event.
1016 */
1017 if (dev->stop)
1018 dev->stop(dev);
1019
1020 /*
1021 * Device is now down.
1022 */
1023
1024 dev->flags &= ~IFF_UP;
1025
1026 /*
1027 * Tell people we are down
1028 */
1029 raw_notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
1030
1031 return 0;
1032 }
1033
1034
1035 /*
1036 * Device change register/unregister. These are not inline or static
1037 * as we export them to the world.
1038 */
1039
1040 /**
1041 * register_netdevice_notifier - register a network notifier block
1042 * @nb: notifier
1043 *
1044 * Register a notifier to be called when network device events occur.
1045 * The notifier passed is linked into the kernel structures and must
1046 * not be reused until it has been unregistered. A negative errno code
1047 * is returned on a failure.
1048 *
1049 * When registered all registration and up events are replayed
1050 * to the new notifier to allow device to have a race free
1051 * view of the network device list.
1052 */
1053
1054 int register_netdevice_notifier(struct notifier_block *nb)
1055 {
1056 struct net_device *dev;
1057 int err;
1058
1059 rtnl_lock();
1060 err = raw_notifier_chain_register(&netdev_chain, nb);
1061 if (!err) {
1062 for_each_netdev(dev) {
1063 nb->notifier_call(nb, NETDEV_REGISTER, dev);
1064
1065 if (dev->flags & IFF_UP)
1066 nb->notifier_call(nb, NETDEV_UP, dev);
1067 }
1068 }
1069 rtnl_unlock();
1070 return err;
1071 }
1072
1073 /**
1074 * unregister_netdevice_notifier - unregister a network notifier block
1075 * @nb: notifier
1076 *
1077 * Unregister a notifier previously registered by
1078 * register_netdevice_notifier(). The notifier is unlinked into the
1079 * kernel structures and may then be reused. A negative errno code
1080 * is returned on a failure.
1081 */
1082
1083 int unregister_netdevice_notifier(struct notifier_block *nb)
1084 {
1085 int err;
1086
1087 rtnl_lock();
1088 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1089 rtnl_unlock();
1090 return err;
1091 }
1092
1093 /**
1094 * call_netdevice_notifiers - call all network notifier blocks
1095 * @val: value passed unmodified to notifier function
1096 * @v: pointer passed unmodified to notifier function
1097 *
1098 * Call all network notifier blocks. Parameters and return value
1099 * are as for raw_notifier_call_chain().
1100 */
1101
1102 int call_netdevice_notifiers(unsigned long val, void *v)
1103 {
1104 return raw_notifier_call_chain(&netdev_chain, val, v);
1105 }
1106
1107 /* When > 0 there are consumers of rx skb time stamps */
1108 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1109
1110 void net_enable_timestamp(void)
1111 {
1112 atomic_inc(&netstamp_needed);
1113 }
1114
1115 void net_disable_timestamp(void)
1116 {
1117 atomic_dec(&netstamp_needed);
1118 }
1119
1120 static inline void net_timestamp(struct sk_buff *skb)
1121 {
1122 if (atomic_read(&netstamp_needed))
1123 __net_timestamp(skb);
1124 else
1125 skb->tstamp.tv64 = 0;
1126 }
1127
1128 /*
1129 * Support routine. Sends outgoing frames to any network
1130 * taps currently in use.
1131 */
1132
1133 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1134 {
1135 struct packet_type *ptype;
1136
1137 net_timestamp(skb);
1138
1139 rcu_read_lock();
1140 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1141 /* Never send packets back to the socket
1142 * they originated from - MvS (miquels@drinkel.ow.org)
1143 */
1144 if ((ptype->dev == dev || !ptype->dev) &&
1145 (ptype->af_packet_priv == NULL ||
1146 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1147 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1148 if (!skb2)
1149 break;
1150
1151 /* skb->nh should be correctly
1152 set by sender, so that the second statement is
1153 just protection against buggy protocols.
1154 */
1155 skb_reset_mac_header(skb2);
1156
1157 if (skb_network_header(skb2) < skb2->data ||
1158 skb2->network_header > skb2->tail) {
1159 if (net_ratelimit())
1160 printk(KERN_CRIT "protocol %04x is "
1161 "buggy, dev %s\n",
1162 skb2->protocol, dev->name);
1163 skb_reset_network_header(skb2);
1164 }
1165
1166 skb2->transport_header = skb2->network_header;
1167 skb2->pkt_type = PACKET_OUTGOING;
1168 ptype->func(skb2, skb->dev, ptype, skb->dev);
1169 }
1170 }
1171 rcu_read_unlock();
1172 }
1173
1174
1175 void __netif_schedule(struct net_device *dev)
1176 {
1177 if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) {
1178 unsigned long flags;
1179 struct softnet_data *sd;
1180
1181 local_irq_save(flags);
1182 sd = &__get_cpu_var(softnet_data);
1183 dev->next_sched = sd->output_queue;
1184 sd->output_queue = dev;
1185 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1186 local_irq_restore(flags);
1187 }
1188 }
1189 EXPORT_SYMBOL(__netif_schedule);
1190
1191 void __netif_rx_schedule(struct net_device *dev)
1192 {
1193 unsigned long flags;
1194
1195 local_irq_save(flags);
1196 dev_hold(dev);
1197 list_add_tail(&dev->poll_list, &__get_cpu_var(softnet_data).poll_list);
1198 if (dev->quota < 0)
1199 dev->quota += dev->weight;
1200 else
1201 dev->quota = dev->weight;
1202 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1203 local_irq_restore(flags);
1204 }
1205 EXPORT_SYMBOL(__netif_rx_schedule);
1206
1207 void dev_kfree_skb_any(struct sk_buff *skb)
1208 {
1209 if (in_irq() || irqs_disabled())
1210 dev_kfree_skb_irq(skb);
1211 else
1212 dev_kfree_skb(skb);
1213 }
1214 EXPORT_SYMBOL(dev_kfree_skb_any);
1215
1216
1217 /* Hot-plugging. */
1218 void netif_device_detach(struct net_device *dev)
1219 {
1220 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1221 netif_running(dev)) {
1222 netif_stop_queue(dev);
1223 }
1224 }
1225 EXPORT_SYMBOL(netif_device_detach);
1226
1227 void netif_device_attach(struct net_device *dev)
1228 {
1229 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1230 netif_running(dev)) {
1231 netif_wake_queue(dev);
1232 __netdev_watchdog_up(dev);
1233 }
1234 }
1235 EXPORT_SYMBOL(netif_device_attach);
1236
1237
1238 /*
1239 * Invalidate hardware checksum when packet is to be mangled, and
1240 * complete checksum manually on outgoing path.
1241 */
1242 int skb_checksum_help(struct sk_buff *skb)
1243 {
1244 __wsum csum;
1245 int ret = 0, offset;
1246
1247 if (skb->ip_summed == CHECKSUM_COMPLETE)
1248 goto out_set_summed;
1249
1250 if (unlikely(skb_shinfo(skb)->gso_size)) {
1251 /* Let GSO fix up the checksum. */
1252 goto out_set_summed;
1253 }
1254
1255 if (skb_cloned(skb)) {
1256 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1257 if (ret)
1258 goto out;
1259 }
1260
1261 offset = skb->csum_start - skb_headroom(skb);
1262 BUG_ON(offset > (int)skb->len);
1263 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1264
1265 offset = skb_headlen(skb) - offset;
1266 BUG_ON(offset <= 0);
1267 BUG_ON(skb->csum_offset + 2 > offset);
1268
1269 *(__sum16 *)(skb->head + skb->csum_start + skb->csum_offset) =
1270 csum_fold(csum);
1271 out_set_summed:
1272 skb->ip_summed = CHECKSUM_NONE;
1273 out:
1274 return ret;
1275 }
1276
1277 /**
1278 * skb_gso_segment - Perform segmentation on skb.
1279 * @skb: buffer to segment
1280 * @features: features for the output path (see dev->features)
1281 *
1282 * This function segments the given skb and returns a list of segments.
1283 *
1284 * It may return NULL if the skb requires no segmentation. This is
1285 * only possible when GSO is used for verifying header integrity.
1286 */
1287 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1288 {
1289 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1290 struct packet_type *ptype;
1291 __be16 type = skb->protocol;
1292 int err;
1293
1294 BUG_ON(skb_shinfo(skb)->frag_list);
1295
1296 skb_reset_mac_header(skb);
1297 skb->mac_len = skb->network_header - skb->mac_header;
1298 __skb_pull(skb, skb->mac_len);
1299
1300 if (WARN_ON(skb->ip_summed != CHECKSUM_PARTIAL)) {
1301 if (skb_header_cloned(skb) &&
1302 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1303 return ERR_PTR(err);
1304 }
1305
1306 rcu_read_lock();
1307 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type) & 15], list) {
1308 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1309 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1310 err = ptype->gso_send_check(skb);
1311 segs = ERR_PTR(err);
1312 if (err || skb_gso_ok(skb, features))
1313 break;
1314 __skb_push(skb, (skb->data -
1315 skb_network_header(skb)));
1316 }
1317 segs = ptype->gso_segment(skb, features);
1318 break;
1319 }
1320 }
1321 rcu_read_unlock();
1322
1323 __skb_push(skb, skb->data - skb_mac_header(skb));
1324
1325 return segs;
1326 }
1327
1328 EXPORT_SYMBOL(skb_gso_segment);
1329
1330 /* Take action when hardware reception checksum errors are detected. */
1331 #ifdef CONFIG_BUG
1332 void netdev_rx_csum_fault(struct net_device *dev)
1333 {
1334 if (net_ratelimit()) {
1335 printk(KERN_ERR "%s: hw csum failure.\n",
1336 dev ? dev->name : "<unknown>");
1337 dump_stack();
1338 }
1339 }
1340 EXPORT_SYMBOL(netdev_rx_csum_fault);
1341 #endif
1342
1343 /* Actually, we should eliminate this check as soon as we know, that:
1344 * 1. IOMMU is present and allows to map all the memory.
1345 * 2. No high memory really exists on this machine.
1346 */
1347
1348 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1349 {
1350 #ifdef CONFIG_HIGHMEM
1351 int i;
1352
1353 if (dev->features & NETIF_F_HIGHDMA)
1354 return 0;
1355
1356 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1357 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1358 return 1;
1359
1360 #endif
1361 return 0;
1362 }
1363
1364 struct dev_gso_cb {
1365 void (*destructor)(struct sk_buff *skb);
1366 };
1367
1368 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1369
1370 static void dev_gso_skb_destructor(struct sk_buff *skb)
1371 {
1372 struct dev_gso_cb *cb;
1373
1374 do {
1375 struct sk_buff *nskb = skb->next;
1376
1377 skb->next = nskb->next;
1378 nskb->next = NULL;
1379 kfree_skb(nskb);
1380 } while (skb->next);
1381
1382 cb = DEV_GSO_CB(skb);
1383 if (cb->destructor)
1384 cb->destructor(skb);
1385 }
1386
1387 /**
1388 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1389 * @skb: buffer to segment
1390 *
1391 * This function segments the given skb and stores the list of segments
1392 * in skb->next.
1393 */
1394 static int dev_gso_segment(struct sk_buff *skb)
1395 {
1396 struct net_device *dev = skb->dev;
1397 struct sk_buff *segs;
1398 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1399 NETIF_F_SG : 0);
1400
1401 segs = skb_gso_segment(skb, features);
1402
1403 /* Verifying header integrity only. */
1404 if (!segs)
1405 return 0;
1406
1407 if (unlikely(IS_ERR(segs)))
1408 return PTR_ERR(segs);
1409
1410 skb->next = segs;
1411 DEV_GSO_CB(skb)->destructor = skb->destructor;
1412 skb->destructor = dev_gso_skb_destructor;
1413
1414 return 0;
1415 }
1416
1417 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
1418 {
1419 if (likely(!skb->next)) {
1420 if (!list_empty(&ptype_all))
1421 dev_queue_xmit_nit(skb, dev);
1422
1423 if (netif_needs_gso(dev, skb)) {
1424 if (unlikely(dev_gso_segment(skb)))
1425 goto out_kfree_skb;
1426 if (skb->next)
1427 goto gso;
1428 }
1429
1430 return dev->hard_start_xmit(skb, dev);
1431 }
1432
1433 gso:
1434 do {
1435 struct sk_buff *nskb = skb->next;
1436 int rc;
1437
1438 skb->next = nskb->next;
1439 nskb->next = NULL;
1440 rc = dev->hard_start_xmit(nskb, dev);
1441 if (unlikely(rc)) {
1442 nskb->next = skb->next;
1443 skb->next = nskb;
1444 return rc;
1445 }
1446 if (unlikely((netif_queue_stopped(dev) ||
1447 netif_subqueue_stopped(dev, skb->queue_mapping)) &&
1448 skb->next))
1449 return NETDEV_TX_BUSY;
1450 } while (skb->next);
1451
1452 skb->destructor = DEV_GSO_CB(skb)->destructor;
1453
1454 out_kfree_skb:
1455 kfree_skb(skb);
1456 return 0;
1457 }
1458
1459 #define HARD_TX_LOCK(dev, cpu) { \
1460 if ((dev->features & NETIF_F_LLTX) == 0) { \
1461 netif_tx_lock(dev); \
1462 } \
1463 }
1464
1465 #define HARD_TX_UNLOCK(dev) { \
1466 if ((dev->features & NETIF_F_LLTX) == 0) { \
1467 netif_tx_unlock(dev); \
1468 } \
1469 }
1470
1471 /**
1472 * dev_queue_xmit - transmit a buffer
1473 * @skb: buffer to transmit
1474 *
1475 * Queue a buffer for transmission to a network device. The caller must
1476 * have set the device and priority and built the buffer before calling
1477 * this function. The function can be called from an interrupt.
1478 *
1479 * A negative errno code is returned on a failure. A success does not
1480 * guarantee the frame will be transmitted as it may be dropped due
1481 * to congestion or traffic shaping.
1482 *
1483 * -----------------------------------------------------------------------------------
1484 * I notice this method can also return errors from the queue disciplines,
1485 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1486 * be positive.
1487 *
1488 * Regardless of the return value, the skb is consumed, so it is currently
1489 * difficult to retry a send to this method. (You can bump the ref count
1490 * before sending to hold a reference for retry if you are careful.)
1491 *
1492 * When calling this method, interrupts MUST be enabled. This is because
1493 * the BH enable code must have IRQs enabled so that it will not deadlock.
1494 * --BLG
1495 */
1496
1497 int dev_queue_xmit(struct sk_buff *skb)
1498 {
1499 struct net_device *dev = skb->dev;
1500 struct Qdisc *q;
1501 int rc = -ENOMEM;
1502
1503 /* GSO will handle the following emulations directly. */
1504 if (netif_needs_gso(dev, skb))
1505 goto gso;
1506
1507 if (skb_shinfo(skb)->frag_list &&
1508 !(dev->features & NETIF_F_FRAGLIST) &&
1509 __skb_linearize(skb))
1510 goto out_kfree_skb;
1511
1512 /* Fragmented skb is linearized if device does not support SG,
1513 * or if at least one of fragments is in highmem and device
1514 * does not support DMA from it.
1515 */
1516 if (skb_shinfo(skb)->nr_frags &&
1517 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1518 __skb_linearize(skb))
1519 goto out_kfree_skb;
1520
1521 /* If packet is not checksummed and device does not support
1522 * checksumming for this protocol, complete checksumming here.
1523 */
1524 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1525 skb_set_transport_header(skb, skb->csum_start -
1526 skb_headroom(skb));
1527
1528 if (!(dev->features & NETIF_F_GEN_CSUM) &&
1529 !((dev->features & NETIF_F_IP_CSUM) &&
1530 skb->protocol == htons(ETH_P_IP)) &&
1531 !((dev->features & NETIF_F_IPV6_CSUM) &&
1532 skb->protocol == htons(ETH_P_IPV6)))
1533 if (skb_checksum_help(skb))
1534 goto out_kfree_skb;
1535 }
1536
1537 gso:
1538 spin_lock_prefetch(&dev->queue_lock);
1539
1540 /* Disable soft irqs for various locks below. Also
1541 * stops preemption for RCU.
1542 */
1543 rcu_read_lock_bh();
1544
1545 /* Updates of qdisc are serialized by queue_lock.
1546 * The struct Qdisc which is pointed to by qdisc is now a
1547 * rcu structure - it may be accessed without acquiring
1548 * a lock (but the structure may be stale.) The freeing of the
1549 * qdisc will be deferred until it's known that there are no
1550 * more references to it.
1551 *
1552 * If the qdisc has an enqueue function, we still need to
1553 * hold the queue_lock before calling it, since queue_lock
1554 * also serializes access to the device queue.
1555 */
1556
1557 q = rcu_dereference(dev->qdisc);
1558 #ifdef CONFIG_NET_CLS_ACT
1559 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1560 #endif
1561 if (q->enqueue) {
1562 /* Grab device queue */
1563 spin_lock(&dev->queue_lock);
1564 q = dev->qdisc;
1565 if (q->enqueue) {
1566 /* reset queue_mapping to zero */
1567 skb->queue_mapping = 0;
1568 rc = q->enqueue(skb, q);
1569 qdisc_run(dev);
1570 spin_unlock(&dev->queue_lock);
1571
1572 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1573 goto out;
1574 }
1575 spin_unlock(&dev->queue_lock);
1576 }
1577
1578 /* The device has no queue. Common case for software devices:
1579 loopback, all the sorts of tunnels...
1580
1581 Really, it is unlikely that netif_tx_lock protection is necessary
1582 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1583 counters.)
1584 However, it is possible, that they rely on protection
1585 made by us here.
1586
1587 Check this and shot the lock. It is not prone from deadlocks.
1588 Either shot noqueue qdisc, it is even simpler 8)
1589 */
1590 if (dev->flags & IFF_UP) {
1591 int cpu = smp_processor_id(); /* ok because BHs are off */
1592
1593 if (dev->xmit_lock_owner != cpu) {
1594
1595 HARD_TX_LOCK(dev, cpu);
1596
1597 if (!netif_queue_stopped(dev) &&
1598 !netif_subqueue_stopped(dev, skb->queue_mapping)) {
1599 rc = 0;
1600 if (!dev_hard_start_xmit(skb, dev)) {
1601 HARD_TX_UNLOCK(dev);
1602 goto out;
1603 }
1604 }
1605 HARD_TX_UNLOCK(dev);
1606 if (net_ratelimit())
1607 printk(KERN_CRIT "Virtual device %s asks to "
1608 "queue packet!\n", dev->name);
1609 } else {
1610 /* Recursion is detected! It is possible,
1611 * unfortunately */
1612 if (net_ratelimit())
1613 printk(KERN_CRIT "Dead loop on virtual device "
1614 "%s, fix it urgently!\n", dev->name);
1615 }
1616 }
1617
1618 rc = -ENETDOWN;
1619 rcu_read_unlock_bh();
1620
1621 out_kfree_skb:
1622 kfree_skb(skb);
1623 return rc;
1624 out:
1625 rcu_read_unlock_bh();
1626 return rc;
1627 }
1628
1629
1630 /*=======================================================================
1631 Receiver routines
1632 =======================================================================*/
1633
1634 int netdev_max_backlog __read_mostly = 1000;
1635 int netdev_budget __read_mostly = 300;
1636 int weight_p __read_mostly = 64; /* old backlog weight */
1637
1638 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1639
1640
1641 /**
1642 * netif_rx - post buffer to the network code
1643 * @skb: buffer to post
1644 *
1645 * This function receives a packet from a device driver and queues it for
1646 * the upper (protocol) levels to process. It always succeeds. The buffer
1647 * may be dropped during processing for congestion control or by the
1648 * protocol layers.
1649 *
1650 * return values:
1651 * NET_RX_SUCCESS (no congestion)
1652 * NET_RX_CN_LOW (low congestion)
1653 * NET_RX_CN_MOD (moderate congestion)
1654 * NET_RX_CN_HIGH (high congestion)
1655 * NET_RX_DROP (packet was dropped)
1656 *
1657 */
1658
1659 int netif_rx(struct sk_buff *skb)
1660 {
1661 struct softnet_data *queue;
1662 unsigned long flags;
1663
1664 /* if netpoll wants it, pretend we never saw it */
1665 if (netpoll_rx(skb))
1666 return NET_RX_DROP;
1667
1668 if (!skb->tstamp.tv64)
1669 net_timestamp(skb);
1670
1671 /*
1672 * The code is rearranged so that the path is the most
1673 * short when CPU is congested, but is still operating.
1674 */
1675 local_irq_save(flags);
1676 queue = &__get_cpu_var(softnet_data);
1677
1678 __get_cpu_var(netdev_rx_stat).total++;
1679 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1680 if (queue->input_pkt_queue.qlen) {
1681 enqueue:
1682 dev_hold(skb->dev);
1683 __skb_queue_tail(&queue->input_pkt_queue, skb);
1684 local_irq_restore(flags);
1685 return NET_RX_SUCCESS;
1686 }
1687
1688 netif_rx_schedule(&queue->backlog_dev);
1689 goto enqueue;
1690 }
1691
1692 __get_cpu_var(netdev_rx_stat).dropped++;
1693 local_irq_restore(flags);
1694
1695 kfree_skb(skb);
1696 return NET_RX_DROP;
1697 }
1698
1699 int netif_rx_ni(struct sk_buff *skb)
1700 {
1701 int err;
1702
1703 preempt_disable();
1704 err = netif_rx(skb);
1705 if (local_softirq_pending())
1706 do_softirq();
1707 preempt_enable();
1708
1709 return err;
1710 }
1711
1712 EXPORT_SYMBOL(netif_rx_ni);
1713
1714 static inline struct net_device *skb_bond(struct sk_buff *skb)
1715 {
1716 struct net_device *dev = skb->dev;
1717
1718 if (dev->master) {
1719 if (skb_bond_should_drop(skb)) {
1720 kfree_skb(skb);
1721 return NULL;
1722 }
1723 skb->dev = dev->master;
1724 }
1725
1726 return dev;
1727 }
1728
1729 static void net_tx_action(struct softirq_action *h)
1730 {
1731 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1732
1733 if (sd->completion_queue) {
1734 struct sk_buff *clist;
1735
1736 local_irq_disable();
1737 clist = sd->completion_queue;
1738 sd->completion_queue = NULL;
1739 local_irq_enable();
1740
1741 while (clist) {
1742 struct sk_buff *skb = clist;
1743 clist = clist->next;
1744
1745 BUG_TRAP(!atomic_read(&skb->users));
1746 __kfree_skb(skb);
1747 }
1748 }
1749
1750 if (sd->output_queue) {
1751 struct net_device *head;
1752
1753 local_irq_disable();
1754 head = sd->output_queue;
1755 sd->output_queue = NULL;
1756 local_irq_enable();
1757
1758 while (head) {
1759 struct net_device *dev = head;
1760 head = head->next_sched;
1761
1762 smp_mb__before_clear_bit();
1763 clear_bit(__LINK_STATE_SCHED, &dev->state);
1764
1765 if (spin_trylock(&dev->queue_lock)) {
1766 qdisc_run(dev);
1767 spin_unlock(&dev->queue_lock);
1768 } else {
1769 netif_schedule(dev);
1770 }
1771 }
1772 }
1773 }
1774
1775 static inline int deliver_skb(struct sk_buff *skb,
1776 struct packet_type *pt_prev,
1777 struct net_device *orig_dev)
1778 {
1779 atomic_inc(&skb->users);
1780 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1781 }
1782
1783 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1784 /* These hooks defined here for ATM */
1785 struct net_bridge;
1786 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1787 unsigned char *addr);
1788 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent) __read_mostly;
1789
1790 /*
1791 * If bridge module is loaded call bridging hook.
1792 * returns NULL if packet was consumed.
1793 */
1794 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
1795 struct sk_buff *skb) __read_mostly;
1796 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
1797 struct packet_type **pt_prev, int *ret,
1798 struct net_device *orig_dev)
1799 {
1800 struct net_bridge_port *port;
1801
1802 if (skb->pkt_type == PACKET_LOOPBACK ||
1803 (port = rcu_dereference(skb->dev->br_port)) == NULL)
1804 return skb;
1805
1806 if (*pt_prev) {
1807 *ret = deliver_skb(skb, *pt_prev, orig_dev);
1808 *pt_prev = NULL;
1809 }
1810
1811 return br_handle_frame_hook(port, skb);
1812 }
1813 #else
1814 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
1815 #endif
1816
1817 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
1818 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
1819 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
1820
1821 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
1822 struct packet_type **pt_prev,
1823 int *ret,
1824 struct net_device *orig_dev)
1825 {
1826 if (skb->dev->macvlan_port == NULL)
1827 return skb;
1828
1829 if (*pt_prev) {
1830 *ret = deliver_skb(skb, *pt_prev, orig_dev);
1831 *pt_prev = NULL;
1832 }
1833 return macvlan_handle_frame_hook(skb);
1834 }
1835 #else
1836 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
1837 #endif
1838
1839 #ifdef CONFIG_NET_CLS_ACT
1840 /* TODO: Maybe we should just force sch_ingress to be compiled in
1841 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1842 * a compare and 2 stores extra right now if we dont have it on
1843 * but have CONFIG_NET_CLS_ACT
1844 * NOTE: This doesnt stop any functionality; if you dont have
1845 * the ingress scheduler, you just cant add policies on ingress.
1846 *
1847 */
1848 static int ing_filter(struct sk_buff *skb)
1849 {
1850 struct Qdisc *q;
1851 struct net_device *dev = skb->dev;
1852 int result = TC_ACT_OK;
1853
1854 if (dev->qdisc_ingress) {
1855 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1856 if (MAX_RED_LOOP < ttl++) {
1857 printk(KERN_WARNING "Redir loop detected Dropping packet (%d->%d)\n",
1858 skb->iif, skb->dev->ifindex);
1859 return TC_ACT_SHOT;
1860 }
1861
1862 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1863
1864 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1865
1866 spin_lock(&dev->ingress_lock);
1867 if ((q = dev->qdisc_ingress) != NULL)
1868 result = q->enqueue(skb, q);
1869 spin_unlock(&dev->ingress_lock);
1870
1871 }
1872
1873 return result;
1874 }
1875 #endif
1876
1877 int netif_receive_skb(struct sk_buff *skb)
1878 {
1879 struct packet_type *ptype, *pt_prev;
1880 struct net_device *orig_dev;
1881 int ret = NET_RX_DROP;
1882 __be16 type;
1883
1884 /* if we've gotten here through NAPI, check netpoll */
1885 if (skb->dev->poll && netpoll_rx(skb))
1886 return NET_RX_DROP;
1887
1888 if (!skb->tstamp.tv64)
1889 net_timestamp(skb);
1890
1891 if (!skb->iif)
1892 skb->iif = skb->dev->ifindex;
1893
1894 orig_dev = skb_bond(skb);
1895
1896 if (!orig_dev)
1897 return NET_RX_DROP;
1898
1899 __get_cpu_var(netdev_rx_stat).total++;
1900
1901 skb_reset_network_header(skb);
1902 skb_reset_transport_header(skb);
1903 skb->mac_len = skb->network_header - skb->mac_header;
1904
1905 pt_prev = NULL;
1906
1907 rcu_read_lock();
1908
1909 #ifdef CONFIG_NET_CLS_ACT
1910 if (skb->tc_verd & TC_NCLS) {
1911 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1912 goto ncls;
1913 }
1914 #endif
1915
1916 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1917 if (!ptype->dev || ptype->dev == skb->dev) {
1918 if (pt_prev)
1919 ret = deliver_skb(skb, pt_prev, orig_dev);
1920 pt_prev = ptype;
1921 }
1922 }
1923
1924 #ifdef CONFIG_NET_CLS_ACT
1925 if (pt_prev) {
1926 ret = deliver_skb(skb, pt_prev, orig_dev);
1927 pt_prev = NULL; /* noone else should process this after*/
1928 } else {
1929 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1930 }
1931
1932 ret = ing_filter(skb);
1933
1934 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1935 kfree_skb(skb);
1936 goto out;
1937 }
1938
1939 skb->tc_verd = 0;
1940 ncls:
1941 #endif
1942
1943 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
1944 if (!skb)
1945 goto out;
1946 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
1947 if (!skb)
1948 goto out;
1949
1950 type = skb->protocol;
1951 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1952 if (ptype->type == type &&
1953 (!ptype->dev || ptype->dev == skb->dev)) {
1954 if (pt_prev)
1955 ret = deliver_skb(skb, pt_prev, orig_dev);
1956 pt_prev = ptype;
1957 }
1958 }
1959
1960 if (pt_prev) {
1961 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1962 } else {
1963 kfree_skb(skb);
1964 /* Jamal, now you will not able to escape explaining
1965 * me how you were going to use this. :-)
1966 */
1967 ret = NET_RX_DROP;
1968 }
1969
1970 out:
1971 rcu_read_unlock();
1972 return ret;
1973 }
1974
1975 static int process_backlog(struct net_device *backlog_dev, int *budget)
1976 {
1977 int work = 0;
1978 int quota = min(backlog_dev->quota, *budget);
1979 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1980 unsigned long start_time = jiffies;
1981
1982 backlog_dev->weight = weight_p;
1983 for (;;) {
1984 struct sk_buff *skb;
1985 struct net_device *dev;
1986
1987 local_irq_disable();
1988 skb = __skb_dequeue(&queue->input_pkt_queue);
1989 if (!skb)
1990 goto job_done;
1991 local_irq_enable();
1992
1993 dev = skb->dev;
1994
1995 netif_receive_skb(skb);
1996
1997 dev_put(dev);
1998
1999 work++;
2000
2001 if (work >= quota || jiffies - start_time > 1)
2002 break;
2003
2004 }
2005
2006 backlog_dev->quota -= work;
2007 *budget -= work;
2008 return -1;
2009
2010 job_done:
2011 backlog_dev->quota -= work;
2012 *budget -= work;
2013
2014 list_del(&backlog_dev->poll_list);
2015 smp_mb__before_clear_bit();
2016 netif_poll_enable(backlog_dev);
2017
2018 local_irq_enable();
2019 return 0;
2020 }
2021
2022 static void net_rx_action(struct softirq_action *h)
2023 {
2024 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2025 unsigned long start_time = jiffies;
2026 int budget = netdev_budget;
2027 void *have;
2028
2029 local_irq_disable();
2030
2031 while (!list_empty(&queue->poll_list)) {
2032 struct net_device *dev;
2033
2034 if (budget <= 0 || jiffies - start_time > 1)
2035 goto softnet_break;
2036
2037 local_irq_enable();
2038
2039 dev = list_entry(queue->poll_list.next,
2040 struct net_device, poll_list);
2041 have = netpoll_poll_lock(dev);
2042
2043 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
2044 netpoll_poll_unlock(have);
2045 local_irq_disable();
2046 list_move_tail(&dev->poll_list, &queue->poll_list);
2047 if (dev->quota < 0)
2048 dev->quota += dev->weight;
2049 else
2050 dev->quota = dev->weight;
2051 } else {
2052 netpoll_poll_unlock(have);
2053 dev_put(dev);
2054 local_irq_disable();
2055 }
2056 }
2057 out:
2058 local_irq_enable();
2059 #ifdef CONFIG_NET_DMA
2060 /*
2061 * There may not be any more sk_buffs coming right now, so push
2062 * any pending DMA copies to hardware
2063 */
2064 if (!cpus_empty(net_dma.channel_mask)) {
2065 int chan_idx;
2066 for_each_cpu_mask(chan_idx, net_dma.channel_mask) {
2067 struct dma_chan *chan = net_dma.channels[chan_idx];
2068 if (chan)
2069 dma_async_memcpy_issue_pending(chan);
2070 }
2071 }
2072 #endif
2073 return;
2074
2075 softnet_break:
2076 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2077 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2078 goto out;
2079 }
2080
2081 static gifconf_func_t * gifconf_list [NPROTO];
2082
2083 /**
2084 * register_gifconf - register a SIOCGIF handler
2085 * @family: Address family
2086 * @gifconf: Function handler
2087 *
2088 * Register protocol dependent address dumping routines. The handler
2089 * that is passed must not be freed or reused until it has been replaced
2090 * by another handler.
2091 */
2092 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2093 {
2094 if (family >= NPROTO)
2095 return -EINVAL;
2096 gifconf_list[family] = gifconf;
2097 return 0;
2098 }
2099
2100
2101 /*
2102 * Map an interface index to its name (SIOCGIFNAME)
2103 */
2104
2105 /*
2106 * We need this ioctl for efficient implementation of the
2107 * if_indextoname() function required by the IPv6 API. Without
2108 * it, we would have to search all the interfaces to find a
2109 * match. --pb
2110 */
2111
2112 static int dev_ifname(struct ifreq __user *arg)
2113 {
2114 struct net_device *dev;
2115 struct ifreq ifr;
2116
2117 /*
2118 * Fetch the caller's info block.
2119 */
2120
2121 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2122 return -EFAULT;
2123
2124 read_lock(&dev_base_lock);
2125 dev = __dev_get_by_index(ifr.ifr_ifindex);
2126 if (!dev) {
2127 read_unlock(&dev_base_lock);
2128 return -ENODEV;
2129 }
2130
2131 strcpy(ifr.ifr_name, dev->name);
2132 read_unlock(&dev_base_lock);
2133
2134 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2135 return -EFAULT;
2136 return 0;
2137 }
2138
2139 /*
2140 * Perform a SIOCGIFCONF call. This structure will change
2141 * size eventually, and there is nothing I can do about it.
2142 * Thus we will need a 'compatibility mode'.
2143 */
2144
2145 static int dev_ifconf(char __user *arg)
2146 {
2147 struct ifconf ifc;
2148 struct net_device *dev;
2149 char __user *pos;
2150 int len;
2151 int total;
2152 int i;
2153
2154 /*
2155 * Fetch the caller's info block.
2156 */
2157
2158 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2159 return -EFAULT;
2160
2161 pos = ifc.ifc_buf;
2162 len = ifc.ifc_len;
2163
2164 /*
2165 * Loop over the interfaces, and write an info block for each.
2166 */
2167
2168 total = 0;
2169 for_each_netdev(dev) {
2170 for (i = 0; i < NPROTO; i++) {
2171 if (gifconf_list[i]) {
2172 int done;
2173 if (!pos)
2174 done = gifconf_list[i](dev, NULL, 0);
2175 else
2176 done = gifconf_list[i](dev, pos + total,
2177 len - total);
2178 if (done < 0)
2179 return -EFAULT;
2180 total += done;
2181 }
2182 }
2183 }
2184
2185 /*
2186 * All done. Write the updated control block back to the caller.
2187 */
2188 ifc.ifc_len = total;
2189
2190 /*
2191 * Both BSD and Solaris return 0 here, so we do too.
2192 */
2193 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2194 }
2195
2196 #ifdef CONFIG_PROC_FS
2197 /*
2198 * This is invoked by the /proc filesystem handler to display a device
2199 * in detail.
2200 */
2201 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2202 {
2203 loff_t off;
2204 struct net_device *dev;
2205
2206 read_lock(&dev_base_lock);
2207 if (!*pos)
2208 return SEQ_START_TOKEN;
2209
2210 off = 1;
2211 for_each_netdev(dev)
2212 if (off++ == *pos)
2213 return dev;
2214
2215 return NULL;
2216 }
2217
2218 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2219 {
2220 ++*pos;
2221 return v == SEQ_START_TOKEN ?
2222 first_net_device() : next_net_device((struct net_device *)v);
2223 }
2224
2225 void dev_seq_stop(struct seq_file *seq, void *v)
2226 {
2227 read_unlock(&dev_base_lock);
2228 }
2229
2230 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2231 {
2232 struct net_device_stats *stats = dev->get_stats(dev);
2233
2234 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2235 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2236 dev->name, stats->rx_bytes, stats->rx_packets,
2237 stats->rx_errors,
2238 stats->rx_dropped + stats->rx_missed_errors,
2239 stats->rx_fifo_errors,
2240 stats->rx_length_errors + stats->rx_over_errors +
2241 stats->rx_crc_errors + stats->rx_frame_errors,
2242 stats->rx_compressed, stats->multicast,
2243 stats->tx_bytes, stats->tx_packets,
2244 stats->tx_errors, stats->tx_dropped,
2245 stats->tx_fifo_errors, stats->collisions,
2246 stats->tx_carrier_errors +
2247 stats->tx_aborted_errors +
2248 stats->tx_window_errors +
2249 stats->tx_heartbeat_errors,
2250 stats->tx_compressed);
2251 }
2252
2253 /*
2254 * Called from the PROCfs module. This now uses the new arbitrary sized
2255 * /proc/net interface to create /proc/net/dev
2256 */
2257 static int dev_seq_show(struct seq_file *seq, void *v)
2258 {
2259 if (v == SEQ_START_TOKEN)
2260 seq_puts(seq, "Inter-| Receive "
2261 " | Transmit\n"
2262 " face |bytes packets errs drop fifo frame "
2263 "compressed multicast|bytes packets errs "
2264 "drop fifo colls carrier compressed\n");
2265 else
2266 dev_seq_printf_stats(seq, v);
2267 return 0;
2268 }
2269
2270 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2271 {
2272 struct netif_rx_stats *rc = NULL;
2273
2274 while (*pos < NR_CPUS)
2275 if (cpu_online(*pos)) {
2276 rc = &per_cpu(netdev_rx_stat, *pos);
2277 break;
2278 } else
2279 ++*pos;
2280 return rc;
2281 }
2282
2283 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2284 {
2285 return softnet_get_online(pos);
2286 }
2287
2288 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2289 {
2290 ++*pos;
2291 return softnet_get_online(pos);
2292 }
2293
2294 static void softnet_seq_stop(struct seq_file *seq, void *v)
2295 {
2296 }
2297
2298 static int softnet_seq_show(struct seq_file *seq, void *v)
2299 {
2300 struct netif_rx_stats *s = v;
2301
2302 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2303 s->total, s->dropped, s->time_squeeze, 0,
2304 0, 0, 0, 0, /* was fastroute */
2305 s->cpu_collision );
2306 return 0;
2307 }
2308
2309 static const struct seq_operations dev_seq_ops = {
2310 .start = dev_seq_start,
2311 .next = dev_seq_next,
2312 .stop = dev_seq_stop,
2313 .show = dev_seq_show,
2314 };
2315
2316 static int dev_seq_open(struct inode *inode, struct file *file)
2317 {
2318 return seq_open(file, &dev_seq_ops);
2319 }
2320
2321 static const struct file_operations dev_seq_fops = {
2322 .owner = THIS_MODULE,
2323 .open = dev_seq_open,
2324 .read = seq_read,
2325 .llseek = seq_lseek,
2326 .release = seq_release,
2327 };
2328
2329 static const struct seq_operations softnet_seq_ops = {
2330 .start = softnet_seq_start,
2331 .next = softnet_seq_next,
2332 .stop = softnet_seq_stop,
2333 .show = softnet_seq_show,
2334 };
2335
2336 static int softnet_seq_open(struct inode *inode, struct file *file)
2337 {
2338 return seq_open(file, &softnet_seq_ops);
2339 }
2340
2341 static const struct file_operations softnet_seq_fops = {
2342 .owner = THIS_MODULE,
2343 .open = softnet_seq_open,
2344 .read = seq_read,
2345 .llseek = seq_lseek,
2346 .release = seq_release,
2347 };
2348
2349 static void *ptype_get_idx(loff_t pos)
2350 {
2351 struct packet_type *pt = NULL;
2352 loff_t i = 0;
2353 int t;
2354
2355 list_for_each_entry_rcu(pt, &ptype_all, list) {
2356 if (i == pos)
2357 return pt;
2358 ++i;
2359 }
2360
2361 for (t = 0; t < 16; t++) {
2362 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
2363 if (i == pos)
2364 return pt;
2365 ++i;
2366 }
2367 }
2368 return NULL;
2369 }
2370
2371 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
2372 {
2373 rcu_read_lock();
2374 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
2375 }
2376
2377 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2378 {
2379 struct packet_type *pt;
2380 struct list_head *nxt;
2381 int hash;
2382
2383 ++*pos;
2384 if (v == SEQ_START_TOKEN)
2385 return ptype_get_idx(0);
2386
2387 pt = v;
2388 nxt = pt->list.next;
2389 if (pt->type == htons(ETH_P_ALL)) {
2390 if (nxt != &ptype_all)
2391 goto found;
2392 hash = 0;
2393 nxt = ptype_base[0].next;
2394 } else
2395 hash = ntohs(pt->type) & 15;
2396
2397 while (nxt == &ptype_base[hash]) {
2398 if (++hash >= 16)
2399 return NULL;
2400 nxt = ptype_base[hash].next;
2401 }
2402 found:
2403 return list_entry(nxt, struct packet_type, list);
2404 }
2405
2406 static void ptype_seq_stop(struct seq_file *seq, void *v)
2407 {
2408 rcu_read_unlock();
2409 }
2410
2411 static void ptype_seq_decode(struct seq_file *seq, void *sym)
2412 {
2413 #ifdef CONFIG_KALLSYMS
2414 unsigned long offset = 0, symsize;
2415 const char *symname;
2416 char *modname;
2417 char namebuf[128];
2418
2419 symname = kallsyms_lookup((unsigned long)sym, &symsize, &offset,
2420 &modname, namebuf);
2421
2422 if (symname) {
2423 char *delim = ":";
2424
2425 if (!modname)
2426 modname = delim = "";
2427 seq_printf(seq, "%s%s%s%s+0x%lx", delim, modname, delim,
2428 symname, offset);
2429 return;
2430 }
2431 #endif
2432
2433 seq_printf(seq, "[%p]", sym);
2434 }
2435
2436 static int ptype_seq_show(struct seq_file *seq, void *v)
2437 {
2438 struct packet_type *pt = v;
2439
2440 if (v == SEQ_START_TOKEN)
2441 seq_puts(seq, "Type Device Function\n");
2442 else {
2443 if (pt->type == htons(ETH_P_ALL))
2444 seq_puts(seq, "ALL ");
2445 else
2446 seq_printf(seq, "%04x", ntohs(pt->type));
2447
2448 seq_printf(seq, " %-8s ",
2449 pt->dev ? pt->dev->name : "");
2450 ptype_seq_decode(seq, pt->func);
2451 seq_putc(seq, '\n');
2452 }
2453
2454 return 0;
2455 }
2456
2457 static const struct seq_operations ptype_seq_ops = {
2458 .start = ptype_seq_start,
2459 .next = ptype_seq_next,
2460 .stop = ptype_seq_stop,
2461 .show = ptype_seq_show,
2462 };
2463
2464 static int ptype_seq_open(struct inode *inode, struct file *file)
2465 {
2466 return seq_open(file, &ptype_seq_ops);
2467 }
2468
2469 static const struct file_operations ptype_seq_fops = {
2470 .owner = THIS_MODULE,
2471 .open = ptype_seq_open,
2472 .read = seq_read,
2473 .llseek = seq_lseek,
2474 .release = seq_release,
2475 };
2476
2477
2478 static int __init dev_proc_init(void)
2479 {
2480 int rc = -ENOMEM;
2481
2482 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2483 goto out;
2484 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2485 goto out_dev;
2486 if (!proc_net_fops_create("ptype", S_IRUGO, &ptype_seq_fops))
2487 goto out_dev2;
2488
2489 if (wext_proc_init())
2490 goto out_softnet;
2491 rc = 0;
2492 out:
2493 return rc;
2494 out_softnet:
2495 proc_net_remove("ptype");
2496 out_dev2:
2497 proc_net_remove("softnet_stat");
2498 out_dev:
2499 proc_net_remove("dev");
2500 goto out;
2501 }
2502 #else
2503 #define dev_proc_init() 0
2504 #endif /* CONFIG_PROC_FS */
2505
2506
2507 /**
2508 * netdev_set_master - set up master/slave pair
2509 * @slave: slave device
2510 * @master: new master device
2511 *
2512 * Changes the master device of the slave. Pass %NULL to break the
2513 * bonding. The caller must hold the RTNL semaphore. On a failure
2514 * a negative errno code is returned. On success the reference counts
2515 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2516 * function returns zero.
2517 */
2518 int netdev_set_master(struct net_device *slave, struct net_device *master)
2519 {
2520 struct net_device *old = slave->master;
2521
2522 ASSERT_RTNL();
2523
2524 if (master) {
2525 if (old)
2526 return -EBUSY;
2527 dev_hold(master);
2528 }
2529
2530 slave->master = master;
2531
2532 synchronize_net();
2533
2534 if (old)
2535 dev_put(old);
2536
2537 if (master)
2538 slave->flags |= IFF_SLAVE;
2539 else
2540 slave->flags &= ~IFF_SLAVE;
2541
2542 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2543 return 0;
2544 }
2545
2546 static void __dev_set_promiscuity(struct net_device *dev, int inc)
2547 {
2548 unsigned short old_flags = dev->flags;
2549
2550 ASSERT_RTNL();
2551
2552 if ((dev->promiscuity += inc) == 0)
2553 dev->flags &= ~IFF_PROMISC;
2554 else
2555 dev->flags |= IFF_PROMISC;
2556 if (dev->flags != old_flags) {
2557 printk(KERN_INFO "device %s %s promiscuous mode\n",
2558 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2559 "left");
2560 audit_log(current->audit_context, GFP_ATOMIC,
2561 AUDIT_ANOM_PROMISCUOUS,
2562 "dev=%s prom=%d old_prom=%d auid=%u",
2563 dev->name, (dev->flags & IFF_PROMISC),
2564 (old_flags & IFF_PROMISC),
2565 audit_get_loginuid(current->audit_context));
2566
2567 if (dev->change_rx_flags)
2568 dev->change_rx_flags(dev, IFF_PROMISC);
2569 }
2570 }
2571
2572 /**
2573 * dev_set_promiscuity - update promiscuity count on a device
2574 * @dev: device
2575 * @inc: modifier
2576 *
2577 * Add or remove promiscuity from a device. While the count in the device
2578 * remains above zero the interface remains promiscuous. Once it hits zero
2579 * the device reverts back to normal filtering operation. A negative inc
2580 * value is used to drop promiscuity on the device.
2581 */
2582 void dev_set_promiscuity(struct net_device *dev, int inc)
2583 {
2584 unsigned short old_flags = dev->flags;
2585
2586 __dev_set_promiscuity(dev, inc);
2587 if (dev->flags != old_flags)
2588 dev_set_rx_mode(dev);
2589 }
2590
2591 /**
2592 * dev_set_allmulti - update allmulti count on a device
2593 * @dev: device
2594 * @inc: modifier
2595 *
2596 * Add or remove reception of all multicast frames to a device. While the
2597 * count in the device remains above zero the interface remains listening
2598 * to all interfaces. Once it hits zero the device reverts back to normal
2599 * filtering operation. A negative @inc value is used to drop the counter
2600 * when releasing a resource needing all multicasts.
2601 */
2602
2603 void dev_set_allmulti(struct net_device *dev, int inc)
2604 {
2605 unsigned short old_flags = dev->flags;
2606
2607 ASSERT_RTNL();
2608
2609 dev->flags |= IFF_ALLMULTI;
2610 if ((dev->allmulti += inc) == 0)
2611 dev->flags &= ~IFF_ALLMULTI;
2612 if (dev->flags ^ old_flags) {
2613 if (dev->change_rx_flags)
2614 dev->change_rx_flags(dev, IFF_ALLMULTI);
2615 dev_set_rx_mode(dev);
2616 }
2617 }
2618
2619 /*
2620 * Upload unicast and multicast address lists to device and
2621 * configure RX filtering. When the device doesn't support unicast
2622 * filtering it is put in promiscous mode while unicast addresses
2623 * are present.
2624 */
2625 void __dev_set_rx_mode(struct net_device *dev)
2626 {
2627 /* dev_open will call this function so the list will stay sane. */
2628 if (!(dev->flags&IFF_UP))
2629 return;
2630
2631 if (!netif_device_present(dev))
2632 return;
2633
2634 if (dev->set_rx_mode)
2635 dev->set_rx_mode(dev);
2636 else {
2637 /* Unicast addresses changes may only happen under the rtnl,
2638 * therefore calling __dev_set_promiscuity here is safe.
2639 */
2640 if (dev->uc_count > 0 && !dev->uc_promisc) {
2641 __dev_set_promiscuity(dev, 1);
2642 dev->uc_promisc = 1;
2643 } else if (dev->uc_count == 0 && dev->uc_promisc) {
2644 __dev_set_promiscuity(dev, -1);
2645 dev->uc_promisc = 0;
2646 }
2647
2648 if (dev->set_multicast_list)
2649 dev->set_multicast_list(dev);
2650 }
2651 }
2652
2653 void dev_set_rx_mode(struct net_device *dev)
2654 {
2655 netif_tx_lock_bh(dev);
2656 __dev_set_rx_mode(dev);
2657 netif_tx_unlock_bh(dev);
2658 }
2659
2660 int __dev_addr_delete(struct dev_addr_list **list, int *count,
2661 void *addr, int alen, int glbl)
2662 {
2663 struct dev_addr_list *da;
2664
2665 for (; (da = *list) != NULL; list = &da->next) {
2666 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
2667 alen == da->da_addrlen) {
2668 if (glbl) {
2669 int old_glbl = da->da_gusers;
2670 da->da_gusers = 0;
2671 if (old_glbl == 0)
2672 break;
2673 }
2674 if (--da->da_users)
2675 return 0;
2676
2677 *list = da->next;
2678 kfree(da);
2679 (*count)--;
2680 return 0;
2681 }
2682 }
2683 return -ENOENT;
2684 }
2685
2686 int __dev_addr_add(struct dev_addr_list **list, int *count,
2687 void *addr, int alen, int glbl)
2688 {
2689 struct dev_addr_list *da;
2690
2691 for (da = *list; da != NULL; da = da->next) {
2692 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
2693 da->da_addrlen == alen) {
2694 if (glbl) {
2695 int old_glbl = da->da_gusers;
2696 da->da_gusers = 1;
2697 if (old_glbl)
2698 return 0;
2699 }
2700 da->da_users++;
2701 return 0;
2702 }
2703 }
2704
2705 da = kmalloc(sizeof(*da), GFP_ATOMIC);
2706 if (da == NULL)
2707 return -ENOMEM;
2708 memcpy(da->da_addr, addr, alen);
2709 da->da_addrlen = alen;
2710 da->da_users = 1;
2711 da->da_gusers = glbl ? 1 : 0;
2712 da->next = *list;
2713 *list = da;
2714 (*count)++;
2715 return 0;
2716 }
2717
2718 /**
2719 * dev_unicast_delete - Release secondary unicast address.
2720 * @dev: device
2721 * @addr: address to delete
2722 * @alen: length of @addr
2723 *
2724 * Release reference to a secondary unicast address and remove it
2725 * from the device if the reference count drops to zero.
2726 *
2727 * The caller must hold the rtnl_mutex.
2728 */
2729 int dev_unicast_delete(struct net_device *dev, void *addr, int alen)
2730 {
2731 int err;
2732
2733 ASSERT_RTNL();
2734
2735 netif_tx_lock_bh(dev);
2736 err = __dev_addr_delete(&dev->uc_list, &dev->uc_count, addr, alen, 0);
2737 if (!err)
2738 __dev_set_rx_mode(dev);
2739 netif_tx_unlock_bh(dev);
2740 return err;
2741 }
2742 EXPORT_SYMBOL(dev_unicast_delete);
2743
2744 /**
2745 * dev_unicast_add - add a secondary unicast address
2746 * @dev: device
2747 * @addr: address to delete
2748 * @alen: length of @addr
2749 *
2750 * Add a secondary unicast address to the device or increase
2751 * the reference count if it already exists.
2752 *
2753 * The caller must hold the rtnl_mutex.
2754 */
2755 int dev_unicast_add(struct net_device *dev, void *addr, int alen)
2756 {
2757 int err;
2758
2759 ASSERT_RTNL();
2760
2761 netif_tx_lock_bh(dev);
2762 err = __dev_addr_add(&dev->uc_list, &dev->uc_count, addr, alen, 0);
2763 if (!err)
2764 __dev_set_rx_mode(dev);
2765 netif_tx_unlock_bh(dev);
2766 return err;
2767 }
2768 EXPORT_SYMBOL(dev_unicast_add);
2769
2770 static void __dev_addr_discard(struct dev_addr_list **list)
2771 {
2772 struct dev_addr_list *tmp;
2773
2774 while (*list != NULL) {
2775 tmp = *list;
2776 *list = tmp->next;
2777 if (tmp->da_users > tmp->da_gusers)
2778 printk("__dev_addr_discard: address leakage! "
2779 "da_users=%d\n", tmp->da_users);
2780 kfree(tmp);
2781 }
2782 }
2783
2784 static void dev_addr_discard(struct net_device *dev)
2785 {
2786 netif_tx_lock_bh(dev);
2787
2788 __dev_addr_discard(&dev->uc_list);
2789 dev->uc_count = 0;
2790
2791 __dev_addr_discard(&dev->mc_list);
2792 dev->mc_count = 0;
2793
2794 netif_tx_unlock_bh(dev);
2795 }
2796
2797 unsigned dev_get_flags(const struct net_device *dev)
2798 {
2799 unsigned flags;
2800
2801 flags = (dev->flags & ~(IFF_PROMISC |
2802 IFF_ALLMULTI |
2803 IFF_RUNNING |
2804 IFF_LOWER_UP |
2805 IFF_DORMANT)) |
2806 (dev->gflags & (IFF_PROMISC |
2807 IFF_ALLMULTI));
2808
2809 if (netif_running(dev)) {
2810 if (netif_oper_up(dev))
2811 flags |= IFF_RUNNING;
2812 if (netif_carrier_ok(dev))
2813 flags |= IFF_LOWER_UP;
2814 if (netif_dormant(dev))
2815 flags |= IFF_DORMANT;
2816 }
2817
2818 return flags;
2819 }
2820
2821 int dev_change_flags(struct net_device *dev, unsigned flags)
2822 {
2823 int ret, changes;
2824 int old_flags = dev->flags;
2825
2826 ASSERT_RTNL();
2827
2828 /*
2829 * Set the flags on our device.
2830 */
2831
2832 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2833 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2834 IFF_AUTOMEDIA)) |
2835 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2836 IFF_ALLMULTI));
2837
2838 /*
2839 * Load in the correct multicast list now the flags have changed.
2840 */
2841
2842 if (dev->change_rx_flags && (dev->flags ^ flags) & IFF_MULTICAST)
2843 dev->change_rx_flags(dev, IFF_MULTICAST);
2844
2845 dev_set_rx_mode(dev);
2846
2847 /*
2848 * Have we downed the interface. We handle IFF_UP ourselves
2849 * according to user attempts to set it, rather than blindly
2850 * setting it.
2851 */
2852
2853 ret = 0;
2854 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2855 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2856
2857 if (!ret)
2858 dev_set_rx_mode(dev);
2859 }
2860
2861 if (dev->flags & IFF_UP &&
2862 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2863 IFF_VOLATILE)))
2864 raw_notifier_call_chain(&netdev_chain,
2865 NETDEV_CHANGE, dev);
2866
2867 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2868 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2869 dev->gflags ^= IFF_PROMISC;
2870 dev_set_promiscuity(dev, inc);
2871 }
2872
2873 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2874 is important. Some (broken) drivers set IFF_PROMISC, when
2875 IFF_ALLMULTI is requested not asking us and not reporting.
2876 */
2877 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2878 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2879 dev->gflags ^= IFF_ALLMULTI;
2880 dev_set_allmulti(dev, inc);
2881 }
2882
2883 /* Exclude state transition flags, already notified */
2884 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
2885 if (changes)
2886 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
2887
2888 return ret;
2889 }
2890
2891 int dev_set_mtu(struct net_device *dev, int new_mtu)
2892 {
2893 int err;
2894
2895 if (new_mtu == dev->mtu)
2896 return 0;
2897
2898 /* MTU must be positive. */
2899 if (new_mtu < 0)
2900 return -EINVAL;
2901
2902 if (!netif_device_present(dev))
2903 return -ENODEV;
2904
2905 err = 0;
2906 if (dev->change_mtu)
2907 err = dev->change_mtu(dev, new_mtu);
2908 else
2909 dev->mtu = new_mtu;
2910 if (!err && dev->flags & IFF_UP)
2911 raw_notifier_call_chain(&netdev_chain,
2912 NETDEV_CHANGEMTU, dev);
2913 return err;
2914 }
2915
2916 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2917 {
2918 int err;
2919
2920 if (!dev->set_mac_address)
2921 return -EOPNOTSUPP;
2922 if (sa->sa_family != dev->type)
2923 return -EINVAL;
2924 if (!netif_device_present(dev))
2925 return -ENODEV;
2926 err = dev->set_mac_address(dev, sa);
2927 if (!err)
2928 raw_notifier_call_chain(&netdev_chain,
2929 NETDEV_CHANGEADDR, dev);
2930 return err;
2931 }
2932
2933 /*
2934 * Perform the SIOCxIFxxx calls.
2935 */
2936 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2937 {
2938 int err;
2939 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2940
2941 if (!dev)
2942 return -ENODEV;
2943
2944 switch (cmd) {
2945 case SIOCGIFFLAGS: /* Get interface flags */
2946 ifr->ifr_flags = dev_get_flags(dev);
2947 return 0;
2948
2949 case SIOCSIFFLAGS: /* Set interface flags */
2950 return dev_change_flags(dev, ifr->ifr_flags);
2951
2952 case SIOCGIFMETRIC: /* Get the metric on the interface
2953 (currently unused) */
2954 ifr->ifr_metric = 0;
2955 return 0;
2956
2957 case SIOCSIFMETRIC: /* Set the metric on the interface
2958 (currently unused) */
2959 return -EOPNOTSUPP;
2960
2961 case SIOCGIFMTU: /* Get the MTU of a device */
2962 ifr->ifr_mtu = dev->mtu;
2963 return 0;
2964
2965 case SIOCSIFMTU: /* Set the MTU of a device */
2966 return dev_set_mtu(dev, ifr->ifr_mtu);
2967
2968 case SIOCGIFHWADDR:
2969 if (!dev->addr_len)
2970 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2971 else
2972 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2973 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2974 ifr->ifr_hwaddr.sa_family = dev->type;
2975 return 0;
2976
2977 case SIOCSIFHWADDR:
2978 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2979
2980 case SIOCSIFHWBROADCAST:
2981 if (ifr->ifr_hwaddr.sa_family != dev->type)
2982 return -EINVAL;
2983 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2984 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2985 raw_notifier_call_chain(&netdev_chain,
2986 NETDEV_CHANGEADDR, dev);
2987 return 0;
2988
2989 case SIOCGIFMAP:
2990 ifr->ifr_map.mem_start = dev->mem_start;
2991 ifr->ifr_map.mem_end = dev->mem_end;
2992 ifr->ifr_map.base_addr = dev->base_addr;
2993 ifr->ifr_map.irq = dev->irq;
2994 ifr->ifr_map.dma = dev->dma;
2995 ifr->ifr_map.port = dev->if_port;
2996 return 0;
2997
2998 case SIOCSIFMAP:
2999 if (dev->set_config) {
3000 if (!netif_device_present(dev))
3001 return -ENODEV;
3002 return dev->set_config(dev, &ifr->ifr_map);
3003 }
3004 return -EOPNOTSUPP;
3005
3006 case SIOCADDMULTI:
3007 if (!dev->set_multicast_list ||
3008 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3009 return -EINVAL;
3010 if (!netif_device_present(dev))
3011 return -ENODEV;
3012 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
3013 dev->addr_len, 1);
3014
3015 case SIOCDELMULTI:
3016 if (!dev->set_multicast_list ||
3017 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3018 return -EINVAL;
3019 if (!netif_device_present(dev))
3020 return -ENODEV;
3021 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
3022 dev->addr_len, 1);
3023
3024 case SIOCGIFINDEX:
3025 ifr->ifr_ifindex = dev->ifindex;
3026 return 0;
3027
3028 case SIOCGIFTXQLEN:
3029 ifr->ifr_qlen = dev->tx_queue_len;
3030 return 0;
3031
3032 case SIOCSIFTXQLEN:
3033 if (ifr->ifr_qlen < 0)
3034 return -EINVAL;
3035 dev->tx_queue_len = ifr->ifr_qlen;
3036 return 0;
3037
3038 case SIOCSIFNAME:
3039 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
3040 return dev_change_name(dev, ifr->ifr_newname);
3041
3042 /*
3043 * Unknown or private ioctl
3044 */
3045
3046 default:
3047 if ((cmd >= SIOCDEVPRIVATE &&
3048 cmd <= SIOCDEVPRIVATE + 15) ||
3049 cmd == SIOCBONDENSLAVE ||
3050 cmd == SIOCBONDRELEASE ||
3051 cmd == SIOCBONDSETHWADDR ||
3052 cmd == SIOCBONDSLAVEINFOQUERY ||
3053 cmd == SIOCBONDINFOQUERY ||
3054 cmd == SIOCBONDCHANGEACTIVE ||
3055 cmd == SIOCGMIIPHY ||
3056 cmd == SIOCGMIIREG ||
3057 cmd == SIOCSMIIREG ||
3058 cmd == SIOCBRADDIF ||
3059 cmd == SIOCBRDELIF ||
3060 cmd == SIOCWANDEV) {
3061 err = -EOPNOTSUPP;
3062 if (dev->do_ioctl) {
3063 if (netif_device_present(dev))
3064 err = dev->do_ioctl(dev, ifr,
3065 cmd);
3066 else
3067 err = -ENODEV;
3068 }
3069 } else
3070 err = -EINVAL;
3071
3072 }
3073 return err;
3074 }
3075
3076 /*
3077 * This function handles all "interface"-type I/O control requests. The actual
3078 * 'doing' part of this is dev_ifsioc above.
3079 */
3080
3081 /**
3082 * dev_ioctl - network device ioctl
3083 * @cmd: command to issue
3084 * @arg: pointer to a struct ifreq in user space
3085 *
3086 * Issue ioctl functions to devices. This is normally called by the
3087 * user space syscall interfaces but can sometimes be useful for
3088 * other purposes. The return value is the return from the syscall if
3089 * positive or a negative errno code on error.
3090 */
3091
3092 int dev_ioctl(unsigned int cmd, void __user *arg)
3093 {
3094 struct ifreq ifr;
3095 int ret;
3096 char *colon;
3097
3098 /* One special case: SIOCGIFCONF takes ifconf argument
3099 and requires shared lock, because it sleeps writing
3100 to user space.
3101 */
3102
3103 if (cmd == SIOCGIFCONF) {
3104 rtnl_lock();
3105 ret = dev_ifconf((char __user *) arg);
3106 rtnl_unlock();
3107 return ret;
3108 }
3109 if (cmd == SIOCGIFNAME)
3110 return dev_ifname((struct ifreq __user *)arg);
3111
3112 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3113 return -EFAULT;
3114
3115 ifr.ifr_name[IFNAMSIZ-1] = 0;
3116
3117 colon = strchr(ifr.ifr_name, ':');
3118 if (colon)
3119 *colon = 0;
3120
3121 /*
3122 * See which interface the caller is talking about.
3123 */
3124
3125 switch (cmd) {
3126 /*
3127 * These ioctl calls:
3128 * - can be done by all.
3129 * - atomic and do not require locking.
3130 * - return a value
3131 */
3132 case SIOCGIFFLAGS:
3133 case SIOCGIFMETRIC:
3134 case SIOCGIFMTU:
3135 case SIOCGIFHWADDR:
3136 case SIOCGIFSLAVE:
3137 case SIOCGIFMAP:
3138 case SIOCGIFINDEX:
3139 case SIOCGIFTXQLEN:
3140 dev_load(ifr.ifr_name);
3141 read_lock(&dev_base_lock);
3142 ret = dev_ifsioc(&ifr, cmd);
3143 read_unlock(&dev_base_lock);
3144 if (!ret) {
3145 if (colon)
3146 *colon = ':';
3147 if (copy_to_user(arg, &ifr,
3148 sizeof(struct ifreq)))
3149 ret = -EFAULT;
3150 }
3151 return ret;
3152
3153 case SIOCETHTOOL:
3154 dev_load(ifr.ifr_name);
3155 rtnl_lock();
3156 ret = dev_ethtool(&ifr);
3157 rtnl_unlock();
3158 if (!ret) {
3159 if (colon)
3160 *colon = ':';
3161 if (copy_to_user(arg, &ifr,
3162 sizeof(struct ifreq)))
3163 ret = -EFAULT;
3164 }
3165 return ret;
3166
3167 /*
3168 * These ioctl calls:
3169 * - require superuser power.
3170 * - require strict serialization.
3171 * - return a value
3172 */
3173 case SIOCGMIIPHY:
3174 case SIOCGMIIREG:
3175 case SIOCSIFNAME:
3176 if (!capable(CAP_NET_ADMIN))
3177 return -EPERM;
3178 dev_load(ifr.ifr_name);
3179 rtnl_lock();
3180 ret = dev_ifsioc(&ifr, cmd);
3181 rtnl_unlock();
3182 if (!ret) {
3183 if (colon)
3184 *colon = ':';
3185 if (copy_to_user(arg, &ifr,
3186 sizeof(struct ifreq)))
3187 ret = -EFAULT;
3188 }
3189 return ret;
3190
3191 /*
3192 * These ioctl calls:
3193 * - require superuser power.
3194 * - require strict serialization.
3195 * - do not return a value
3196 */
3197 case SIOCSIFFLAGS:
3198 case SIOCSIFMETRIC:
3199 case SIOCSIFMTU:
3200 case SIOCSIFMAP:
3201 case SIOCSIFHWADDR:
3202 case SIOCSIFSLAVE:
3203 case SIOCADDMULTI:
3204 case SIOCDELMULTI:
3205 case SIOCSIFHWBROADCAST:
3206 case SIOCSIFTXQLEN:
3207 case SIOCSMIIREG:
3208 case SIOCBONDENSLAVE:
3209 case SIOCBONDRELEASE:
3210 case SIOCBONDSETHWADDR:
3211 case SIOCBONDCHANGEACTIVE:
3212 case SIOCBRADDIF:
3213 case SIOCBRDELIF:
3214 if (!capable(CAP_NET_ADMIN))
3215 return -EPERM;
3216 /* fall through */
3217 case SIOCBONDSLAVEINFOQUERY:
3218 case SIOCBONDINFOQUERY:
3219 dev_load(ifr.ifr_name);
3220 rtnl_lock();
3221 ret = dev_ifsioc(&ifr, cmd);
3222 rtnl_unlock();
3223 return ret;
3224
3225 case SIOCGIFMEM:
3226 /* Get the per device memory space. We can add this but
3227 * currently do not support it */
3228 case SIOCSIFMEM:
3229 /* Set the per device memory buffer space.
3230 * Not applicable in our case */
3231 case SIOCSIFLINK:
3232 return -EINVAL;
3233
3234 /*
3235 * Unknown or private ioctl.
3236 */
3237 default:
3238 if (cmd == SIOCWANDEV ||
3239 (cmd >= SIOCDEVPRIVATE &&
3240 cmd <= SIOCDEVPRIVATE + 15)) {
3241 dev_load(ifr.ifr_name);
3242 rtnl_lock();
3243 ret = dev_ifsioc(&ifr, cmd);
3244 rtnl_unlock();
3245 if (!ret && copy_to_user(arg, &ifr,
3246 sizeof(struct ifreq)))
3247 ret = -EFAULT;
3248 return ret;
3249 }
3250 /* Take care of Wireless Extensions */
3251 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
3252 return wext_handle_ioctl(&ifr, cmd, arg);
3253 return -EINVAL;
3254 }
3255 }
3256
3257
3258 /**
3259 * dev_new_index - allocate an ifindex
3260 *
3261 * Returns a suitable unique value for a new device interface
3262 * number. The caller must hold the rtnl semaphore or the
3263 * dev_base_lock to be sure it remains unique.
3264 */
3265 static int dev_new_index(void)
3266 {
3267 static int ifindex;
3268 for (;;) {
3269 if (++ifindex <= 0)
3270 ifindex = 1;
3271 if (!__dev_get_by_index(ifindex))
3272 return ifindex;
3273 }
3274 }
3275
3276 static int dev_boot_phase = 1;
3277
3278 /* Delayed registration/unregisteration */
3279 static DEFINE_SPINLOCK(net_todo_list_lock);
3280 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
3281
3282 static void net_set_todo(struct net_device *dev)
3283 {
3284 spin_lock(&net_todo_list_lock);
3285 list_add_tail(&dev->todo_list, &net_todo_list);
3286 spin_unlock(&net_todo_list_lock);
3287 }
3288
3289 /**
3290 * register_netdevice - register a network device
3291 * @dev: device to register
3292 *
3293 * Take a completed network device structure and add it to the kernel
3294 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3295 * chain. 0 is returned on success. A negative errno code is returned
3296 * on a failure to set up the device, or if the name is a duplicate.
3297 *
3298 * Callers must hold the rtnl semaphore. You may want
3299 * register_netdev() instead of this.
3300 *
3301 * BUGS:
3302 * The locking appears insufficient to guarantee two parallel registers
3303 * will not get the same name.
3304 */
3305
3306 int register_netdevice(struct net_device *dev)
3307 {
3308 struct hlist_head *head;
3309 struct hlist_node *p;
3310 int ret;
3311
3312 BUG_ON(dev_boot_phase);
3313 ASSERT_RTNL();
3314
3315 might_sleep();
3316
3317 /* When net_device's are persistent, this will be fatal. */
3318 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
3319
3320 spin_lock_init(&dev->queue_lock);
3321 spin_lock_init(&dev->_xmit_lock);
3322 netdev_set_lockdep_class(&dev->_xmit_lock, dev->type);
3323 dev->xmit_lock_owner = -1;
3324 spin_lock_init(&dev->ingress_lock);
3325
3326 dev->iflink = -1;
3327
3328 /* Init, if this function is available */
3329 if (dev->init) {
3330 ret = dev->init(dev);
3331 if (ret) {
3332 if (ret > 0)
3333 ret = -EIO;
3334 goto out;
3335 }
3336 }
3337
3338 if (!dev_valid_name(dev->name)) {
3339 ret = -EINVAL;
3340 goto out;
3341 }
3342
3343 dev->ifindex = dev_new_index();
3344 if (dev->iflink == -1)
3345 dev->iflink = dev->ifindex;
3346
3347 /* Check for existence of name */
3348 head = dev_name_hash(dev->name);
3349 hlist_for_each(p, head) {
3350 struct net_device *d
3351 = hlist_entry(p, struct net_device, name_hlist);
3352 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
3353 ret = -EEXIST;
3354 goto out;
3355 }
3356 }
3357
3358 /* Fix illegal checksum combinations */
3359 if ((dev->features & NETIF_F_HW_CSUM) &&
3360 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
3361 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
3362 dev->name);
3363 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
3364 }
3365
3366 if ((dev->features & NETIF_F_NO_CSUM) &&
3367 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
3368 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
3369 dev->name);
3370 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
3371 }
3372
3373
3374 /* Fix illegal SG+CSUM combinations. */
3375 if ((dev->features & NETIF_F_SG) &&
3376 !(dev->features & NETIF_F_ALL_CSUM)) {
3377 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no checksum feature.\n",
3378 dev->name);
3379 dev->features &= ~NETIF_F_SG;
3380 }
3381
3382 /* TSO requires that SG is present as well. */
3383 if ((dev->features & NETIF_F_TSO) &&
3384 !(dev->features & NETIF_F_SG)) {
3385 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no SG feature.\n",
3386 dev->name);
3387 dev->features &= ~NETIF_F_TSO;
3388 }
3389 if (dev->features & NETIF_F_UFO) {
3390 if (!(dev->features & NETIF_F_HW_CSUM)) {
3391 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3392 "NETIF_F_HW_CSUM feature.\n",
3393 dev->name);
3394 dev->features &= ~NETIF_F_UFO;
3395 }
3396 if (!(dev->features & NETIF_F_SG)) {
3397 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3398 "NETIF_F_SG feature.\n",
3399 dev->name);
3400 dev->features &= ~NETIF_F_UFO;
3401 }
3402 }
3403
3404 /*
3405 * nil rebuild_header routine,
3406 * that should be never called and used as just bug trap.
3407 */
3408
3409 if (!dev->rebuild_header)
3410 dev->rebuild_header = default_rebuild_header;
3411
3412 ret = netdev_register_sysfs(dev);
3413 if (ret)
3414 goto out;
3415 dev->reg_state = NETREG_REGISTERED;
3416
3417 /*
3418 * Default initial state at registry is that the
3419 * device is present.
3420 */
3421
3422 set_bit(__LINK_STATE_PRESENT, &dev->state);
3423
3424 dev_init_scheduler(dev);
3425 write_lock_bh(&dev_base_lock);
3426 list_add_tail(&dev->dev_list, &dev_base_head);
3427 hlist_add_head(&dev->name_hlist, head);
3428 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
3429 dev_hold(dev);
3430 write_unlock_bh(&dev_base_lock);
3431
3432 /* Notify protocols, that a new device appeared. */
3433 raw_notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
3434
3435 ret = 0;
3436
3437 out:
3438 return ret;
3439 }
3440
3441 /**
3442 * register_netdev - register a network device
3443 * @dev: device to register
3444 *
3445 * Take a completed network device structure and add it to the kernel
3446 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3447 * chain. 0 is returned on success. A negative errno code is returned
3448 * on a failure to set up the device, or if the name is a duplicate.
3449 *
3450 * This is a wrapper around register_netdevice that takes the rtnl semaphore
3451 * and expands the device name if you passed a format string to
3452 * alloc_netdev.
3453 */
3454 int register_netdev(struct net_device *dev)
3455 {
3456 int err;
3457
3458 rtnl_lock();
3459
3460 /*
3461 * If the name is a format string the caller wants us to do a
3462 * name allocation.
3463 */
3464 if (strchr(dev->name, '%')) {
3465 err = dev_alloc_name(dev, dev->name);
3466 if (err < 0)
3467 goto out;
3468 }
3469
3470 err = register_netdevice(dev);
3471 out:
3472 rtnl_unlock();
3473 return err;
3474 }
3475 EXPORT_SYMBOL(register_netdev);
3476
3477 /*
3478 * netdev_wait_allrefs - wait until all references are gone.
3479 *
3480 * This is called when unregistering network devices.
3481 *
3482 * Any protocol or device that holds a reference should register
3483 * for netdevice notification, and cleanup and put back the
3484 * reference if they receive an UNREGISTER event.
3485 * We can get stuck here if buggy protocols don't correctly
3486 * call dev_put.
3487 */
3488 static void netdev_wait_allrefs(struct net_device *dev)
3489 {
3490 unsigned long rebroadcast_time, warning_time;
3491
3492 rebroadcast_time = warning_time = jiffies;
3493 while (atomic_read(&dev->refcnt) != 0) {
3494 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
3495 rtnl_lock();
3496
3497 /* Rebroadcast unregister notification */
3498 raw_notifier_call_chain(&netdev_chain,
3499 NETDEV_UNREGISTER, dev);
3500
3501 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
3502 &dev->state)) {
3503 /* We must not have linkwatch events
3504 * pending on unregister. If this
3505 * happens, we simply run the queue
3506 * unscheduled, resulting in a noop
3507 * for this device.
3508 */
3509 linkwatch_run_queue();
3510 }
3511
3512 __rtnl_unlock();
3513
3514 rebroadcast_time = jiffies;
3515 }
3516
3517 msleep(250);
3518
3519 if (time_after(jiffies, warning_time + 10 * HZ)) {
3520 printk(KERN_EMERG "unregister_netdevice: "
3521 "waiting for %s to become free. Usage "
3522 "count = %d\n",
3523 dev->name, atomic_read(&dev->refcnt));
3524 warning_time = jiffies;
3525 }
3526 }
3527 }
3528
3529 /* The sequence is:
3530 *
3531 * rtnl_lock();
3532 * ...
3533 * register_netdevice(x1);
3534 * register_netdevice(x2);
3535 * ...
3536 * unregister_netdevice(y1);
3537 * unregister_netdevice(y2);
3538 * ...
3539 * rtnl_unlock();
3540 * free_netdev(y1);
3541 * free_netdev(y2);
3542 *
3543 * We are invoked by rtnl_unlock() after it drops the semaphore.
3544 * This allows us to deal with problems:
3545 * 1) We can delete sysfs objects which invoke hotplug
3546 * without deadlocking with linkwatch via keventd.
3547 * 2) Since we run with the RTNL semaphore not held, we can sleep
3548 * safely in order to wait for the netdev refcnt to drop to zero.
3549 */
3550 static DEFINE_MUTEX(net_todo_run_mutex);
3551 void netdev_run_todo(void)
3552 {
3553 struct list_head list;
3554
3555 /* Need to guard against multiple cpu's getting out of order. */
3556 mutex_lock(&net_todo_run_mutex);
3557
3558 /* Not safe to do outside the semaphore. We must not return
3559 * until all unregister events invoked by the local processor
3560 * have been completed (either by this todo run, or one on
3561 * another cpu).
3562 */
3563 if (list_empty(&net_todo_list))
3564 goto out;
3565
3566 /* Snapshot list, allow later requests */
3567 spin_lock(&net_todo_list_lock);
3568 list_replace_init(&net_todo_list, &list);
3569 spin_unlock(&net_todo_list_lock);
3570
3571 while (!list_empty(&list)) {
3572 struct net_device *dev
3573 = list_entry(list.next, struct net_device, todo_list);
3574 list_del(&dev->todo_list);
3575
3576 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
3577 printk(KERN_ERR "network todo '%s' but state %d\n",
3578 dev->name, dev->reg_state);
3579 dump_stack();
3580 continue;
3581 }
3582
3583 dev->reg_state = NETREG_UNREGISTERED;
3584
3585 netdev_wait_allrefs(dev);
3586
3587 /* paranoia */
3588 BUG_ON(atomic_read(&dev->refcnt));
3589 BUG_TRAP(!dev->ip_ptr);
3590 BUG_TRAP(!dev->ip6_ptr);
3591 BUG_TRAP(!dev->dn_ptr);
3592
3593 if (dev->destructor)
3594 dev->destructor(dev);
3595
3596 /* Free network device */
3597 kobject_put(&dev->dev.kobj);
3598 }
3599
3600 out:
3601 mutex_unlock(&net_todo_run_mutex);
3602 }
3603
3604 static struct net_device_stats *internal_stats(struct net_device *dev)
3605 {
3606 return &dev->stats;
3607 }
3608
3609 /**
3610 * alloc_netdev_mq - allocate network device
3611 * @sizeof_priv: size of private data to allocate space for
3612 * @name: device name format string
3613 * @setup: callback to initialize device
3614 * @queue_count: the number of subqueues to allocate
3615 *
3616 * Allocates a struct net_device with private data area for driver use
3617 * and performs basic initialization. Also allocates subquue structs
3618 * for each queue on the device at the end of the netdevice.
3619 */
3620 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
3621 void (*setup)(struct net_device *), unsigned int queue_count)
3622 {
3623 void *p;
3624 struct net_device *dev;
3625 int alloc_size;
3626
3627 BUG_ON(strlen(name) >= sizeof(dev->name));
3628
3629 /* ensure 32-byte alignment of both the device and private area */
3630 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST +
3631 (sizeof(struct net_device_subqueue) * (queue_count - 1))) &
3632 ~NETDEV_ALIGN_CONST;
3633 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3634
3635 p = kzalloc(alloc_size, GFP_KERNEL);
3636 if (!p) {
3637 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
3638 return NULL;
3639 }
3640
3641 dev = (struct net_device *)
3642 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3643 dev->padded = (char *)dev - (char *)p;
3644
3645 if (sizeof_priv) {
3646 dev->priv = ((char *)dev +
3647 ((sizeof(struct net_device) +
3648 (sizeof(struct net_device_subqueue) *
3649 (queue_count - 1)) + NETDEV_ALIGN_CONST)
3650 & ~NETDEV_ALIGN_CONST));
3651 }
3652
3653 dev->egress_subqueue_count = queue_count;
3654
3655 dev->get_stats = internal_stats;
3656 setup(dev);
3657 strcpy(dev->name, name);
3658 return dev;
3659 }
3660 EXPORT_SYMBOL(alloc_netdev_mq);
3661
3662 /**
3663 * free_netdev - free network device
3664 * @dev: device
3665 *
3666 * This function does the last stage of destroying an allocated device
3667 * interface. The reference to the device object is released.
3668 * If this is the last reference then it will be freed.
3669 */
3670 void free_netdev(struct net_device *dev)
3671 {
3672 #ifdef CONFIG_SYSFS
3673 /* Compatibility with error handling in drivers */
3674 if (dev->reg_state == NETREG_UNINITIALIZED) {
3675 kfree((char *)dev - dev->padded);
3676 return;
3677 }
3678
3679 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3680 dev->reg_state = NETREG_RELEASED;
3681
3682 /* will free via device release */
3683 put_device(&dev->dev);
3684 #else
3685 kfree((char *)dev - dev->padded);
3686 #endif
3687 }
3688
3689 /* Synchronize with packet receive processing. */
3690 void synchronize_net(void)
3691 {
3692 might_sleep();
3693 synchronize_rcu();
3694 }
3695
3696 /**
3697 * unregister_netdevice - remove device from the kernel
3698 * @dev: device
3699 *
3700 * This function shuts down a device interface and removes it
3701 * from the kernel tables. On success 0 is returned, on a failure
3702 * a negative errno code is returned.
3703 *
3704 * Callers must hold the rtnl semaphore. You may want
3705 * unregister_netdev() instead of this.
3706 */
3707
3708 void unregister_netdevice(struct net_device *dev)
3709 {
3710 BUG_ON(dev_boot_phase);
3711 ASSERT_RTNL();
3712
3713 /* Some devices call without registering for initialization unwind. */
3714 if (dev->reg_state == NETREG_UNINITIALIZED) {
3715 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3716 "was registered\n", dev->name, dev);
3717
3718 WARN_ON(1);
3719 return;
3720 }
3721
3722 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3723
3724 /* If device is running, close it first. */
3725 if (dev->flags & IFF_UP)
3726 dev_close(dev);
3727
3728 /* And unlink it from device chain. */
3729 write_lock_bh(&dev_base_lock);
3730 list_del(&dev->dev_list);
3731 hlist_del(&dev->name_hlist);
3732 hlist_del(&dev->index_hlist);
3733 write_unlock_bh(&dev_base_lock);
3734
3735 dev->reg_state = NETREG_UNREGISTERING;
3736
3737 synchronize_net();
3738
3739 /* Shutdown queueing discipline. */
3740 dev_shutdown(dev);
3741
3742
3743 /* Notify protocols, that we are about to destroy
3744 this device. They should clean all the things.
3745 */
3746 raw_notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3747
3748 /*
3749 * Flush the unicast and multicast chains
3750 */
3751 dev_addr_discard(dev);
3752
3753 if (dev->uninit)
3754 dev->uninit(dev);
3755
3756 /* Notifier chain MUST detach us from master device. */
3757 BUG_TRAP(!dev->master);
3758
3759 /* Remove entries from sysfs */
3760 netdev_unregister_sysfs(dev);
3761
3762 /* Finish processing unregister after unlock */
3763 net_set_todo(dev);
3764
3765 synchronize_net();
3766
3767 dev_put(dev);
3768 }
3769
3770 /**
3771 * unregister_netdev - remove device from the kernel
3772 * @dev: device
3773 *
3774 * This function shuts down a device interface and removes it
3775 * from the kernel tables. On success 0 is returned, on a failure
3776 * a negative errno code is returned.
3777 *
3778 * This is just a wrapper for unregister_netdevice that takes
3779 * the rtnl semaphore. In general you want to use this and not
3780 * unregister_netdevice.
3781 */
3782 void unregister_netdev(struct net_device *dev)
3783 {
3784 rtnl_lock();
3785 unregister_netdevice(dev);
3786 rtnl_unlock();
3787 }
3788
3789 EXPORT_SYMBOL(unregister_netdev);
3790
3791 static int dev_cpu_callback(struct notifier_block *nfb,
3792 unsigned long action,
3793 void *ocpu)
3794 {
3795 struct sk_buff **list_skb;
3796 struct net_device **list_net;
3797 struct sk_buff *skb;
3798 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3799 struct softnet_data *sd, *oldsd;
3800
3801 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
3802 return NOTIFY_OK;
3803
3804 local_irq_disable();
3805 cpu = smp_processor_id();
3806 sd = &per_cpu(softnet_data, cpu);
3807 oldsd = &per_cpu(softnet_data, oldcpu);
3808
3809 /* Find end of our completion_queue. */
3810 list_skb = &sd->completion_queue;
3811 while (*list_skb)
3812 list_skb = &(*list_skb)->next;
3813 /* Append completion queue from offline CPU. */
3814 *list_skb = oldsd->completion_queue;
3815 oldsd->completion_queue = NULL;
3816
3817 /* Find end of our output_queue. */
3818 list_net = &sd->output_queue;
3819 while (*list_net)
3820 list_net = &(*list_net)->next_sched;
3821 /* Append output queue from offline CPU. */
3822 *list_net = oldsd->output_queue;
3823 oldsd->output_queue = NULL;
3824
3825 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3826 local_irq_enable();
3827
3828 /* Process offline CPU's input_pkt_queue */
3829 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3830 netif_rx(skb);
3831
3832 return NOTIFY_OK;
3833 }
3834
3835 #ifdef CONFIG_NET_DMA
3836 /**
3837 * net_dma_rebalance - try to maintain one DMA channel per CPU
3838 * @net_dma: DMA client and associated data (lock, channels, channel_mask)
3839 *
3840 * This is called when the number of channels allocated to the net_dma client
3841 * changes. The net_dma client tries to have one DMA channel per CPU.
3842 */
3843
3844 static void net_dma_rebalance(struct net_dma *net_dma)
3845 {
3846 unsigned int cpu, i, n, chan_idx;
3847 struct dma_chan *chan;
3848
3849 if (cpus_empty(net_dma->channel_mask)) {
3850 for_each_online_cpu(cpu)
3851 rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL);
3852 return;
3853 }
3854
3855 i = 0;
3856 cpu = first_cpu(cpu_online_map);
3857
3858 for_each_cpu_mask(chan_idx, net_dma->channel_mask) {
3859 chan = net_dma->channels[chan_idx];
3860
3861 n = ((num_online_cpus() / cpus_weight(net_dma->channel_mask))
3862 + (i < (num_online_cpus() %
3863 cpus_weight(net_dma->channel_mask)) ? 1 : 0));
3864
3865 while(n) {
3866 per_cpu(softnet_data, cpu).net_dma = chan;
3867 cpu = next_cpu(cpu, cpu_online_map);
3868 n--;
3869 }
3870 i++;
3871 }
3872 }
3873
3874 /**
3875 * netdev_dma_event - event callback for the net_dma_client
3876 * @client: should always be net_dma_client
3877 * @chan: DMA channel for the event
3878 * @state: DMA state to be handled
3879 */
3880 static enum dma_state_client
3881 netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
3882 enum dma_state state)
3883 {
3884 int i, found = 0, pos = -1;
3885 struct net_dma *net_dma =
3886 container_of(client, struct net_dma, client);
3887 enum dma_state_client ack = DMA_DUP; /* default: take no action */
3888
3889 spin_lock(&net_dma->lock);
3890 switch (state) {
3891 case DMA_RESOURCE_AVAILABLE:
3892 for (i = 0; i < NR_CPUS; i++)
3893 if (net_dma->channels[i] == chan) {
3894 found = 1;
3895 break;
3896 } else if (net_dma->channels[i] == NULL && pos < 0)
3897 pos = i;
3898
3899 if (!found && pos >= 0) {
3900 ack = DMA_ACK;
3901 net_dma->channels[pos] = chan;
3902 cpu_set(pos, net_dma->channel_mask);
3903 net_dma_rebalance(net_dma);
3904 }
3905 break;
3906 case DMA_RESOURCE_REMOVED:
3907 for (i = 0; i < NR_CPUS; i++)
3908 if (net_dma->channels[i] == chan) {
3909 found = 1;
3910 pos = i;
3911 break;
3912 }
3913
3914 if (found) {
3915 ack = DMA_ACK;
3916 cpu_clear(pos, net_dma->channel_mask);
3917 net_dma->channels[i] = NULL;
3918 net_dma_rebalance(net_dma);
3919 }
3920 break;
3921 default:
3922 break;
3923 }
3924 spin_unlock(&net_dma->lock);
3925
3926 return ack;
3927 }
3928
3929 /**
3930 * netdev_dma_regiser - register the networking subsystem as a DMA client
3931 */
3932 static int __init netdev_dma_register(void)
3933 {
3934 spin_lock_init(&net_dma.lock);
3935 dma_cap_set(DMA_MEMCPY, net_dma.client.cap_mask);
3936 dma_async_client_register(&net_dma.client);
3937 dma_async_client_chan_request(&net_dma.client);
3938 return 0;
3939 }
3940
3941 #else
3942 static int __init netdev_dma_register(void) { return -ENODEV; }
3943 #endif /* CONFIG_NET_DMA */
3944
3945 /*
3946 * Initialize the DEV module. At boot time this walks the device list and
3947 * unhooks any devices that fail to initialise (normally hardware not
3948 * present) and leaves us with a valid list of present and active devices.
3949 *
3950 */
3951
3952 /*
3953 * This is called single threaded during boot, so no need
3954 * to take the rtnl semaphore.
3955 */
3956 static int __init net_dev_init(void)
3957 {
3958 int i, rc = -ENOMEM;
3959
3960 BUG_ON(!dev_boot_phase);
3961
3962 if (dev_proc_init())
3963 goto out;
3964
3965 if (netdev_sysfs_init())
3966 goto out;
3967
3968 INIT_LIST_HEAD(&ptype_all);
3969 for (i = 0; i < 16; i++)
3970 INIT_LIST_HEAD(&ptype_base[i]);
3971
3972 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3973 INIT_HLIST_HEAD(&dev_name_head[i]);
3974
3975 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3976 INIT_HLIST_HEAD(&dev_index_head[i]);
3977
3978 /*
3979 * Initialise the packet receive queues.
3980 */
3981
3982 for_each_possible_cpu(i) {
3983 struct softnet_data *queue;
3984
3985 queue = &per_cpu(softnet_data, i);
3986 skb_queue_head_init(&queue->input_pkt_queue);
3987 queue->completion_queue = NULL;
3988 INIT_LIST_HEAD(&queue->poll_list);
3989 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3990 queue->backlog_dev.weight = weight_p;
3991 queue->backlog_dev.poll = process_backlog;
3992 atomic_set(&queue->backlog_dev.refcnt, 1);
3993 }
3994
3995 netdev_dma_register();
3996
3997 dev_boot_phase = 0;
3998
3999 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
4000 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
4001
4002 hotcpu_notifier(dev_cpu_callback, 0);
4003 dst_init();
4004 dev_mcast_init();
4005 rc = 0;
4006 out:
4007 return rc;
4008 }
4009
4010 subsys_initcall(net_dev_init);
4011
4012 EXPORT_SYMBOL(__dev_get_by_index);
4013 EXPORT_SYMBOL(__dev_get_by_name);
4014 EXPORT_SYMBOL(__dev_remove_pack);
4015 EXPORT_SYMBOL(dev_valid_name);
4016 EXPORT_SYMBOL(dev_add_pack);
4017 EXPORT_SYMBOL(dev_alloc_name);
4018 EXPORT_SYMBOL(dev_close);
4019 EXPORT_SYMBOL(dev_get_by_flags);
4020 EXPORT_SYMBOL(dev_get_by_index);
4021 EXPORT_SYMBOL(dev_get_by_name);
4022 EXPORT_SYMBOL(dev_open);
4023 EXPORT_SYMBOL(dev_queue_xmit);
4024 EXPORT_SYMBOL(dev_remove_pack);
4025 EXPORT_SYMBOL(dev_set_allmulti);
4026 EXPORT_SYMBOL(dev_set_promiscuity);
4027 EXPORT_SYMBOL(dev_change_flags);
4028 EXPORT_SYMBOL(dev_set_mtu);
4029 EXPORT_SYMBOL(dev_set_mac_address);
4030 EXPORT_SYMBOL(free_netdev);
4031 EXPORT_SYMBOL(netdev_boot_setup_check);
4032 EXPORT_SYMBOL(netdev_set_master);
4033 EXPORT_SYMBOL(netdev_state_change);
4034 EXPORT_SYMBOL(netif_receive_skb);
4035 EXPORT_SYMBOL(netif_rx);
4036 EXPORT_SYMBOL(register_gifconf);
4037 EXPORT_SYMBOL(register_netdevice);
4038 EXPORT_SYMBOL(register_netdevice_notifier);
4039 EXPORT_SYMBOL(skb_checksum_help);
4040 EXPORT_SYMBOL(synchronize_net);
4041 EXPORT_SYMBOL(unregister_netdevice);
4042 EXPORT_SYMBOL(unregister_netdevice_notifier);
4043 EXPORT_SYMBOL(net_enable_timestamp);
4044 EXPORT_SYMBOL(net_disable_timestamp);
4045 EXPORT_SYMBOL(dev_get_flags);
4046
4047 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
4048 EXPORT_SYMBOL(br_handle_frame_hook);
4049 EXPORT_SYMBOL(br_fdb_get_hook);
4050 EXPORT_SYMBOL(br_fdb_put_hook);
4051 #endif
4052
4053 #ifdef CONFIG_KMOD
4054 EXPORT_SYMBOL(dev_load);
4055 #endif
4056
4057 EXPORT_PER_CPU_SYMBOL(softnet_data);