]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - net/core/dev.c
Merge master.kernel.org:/pub/scm/linux/kernel/git/lethal/sh64-2.6
[mirror_ubuntu-bionic-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 char oldname[IFNAMSIZ];
821 int err = 0;
822 int ret;
823
824 ASSERT_RTNL();
825
826 if (dev->flags & IFF_UP)
827 return -EBUSY;
828
829 if (!dev_valid_name(newname))
830 return -EINVAL;
831
832 memcpy(oldname, dev->name, IFNAMSIZ);
833
834 if (strchr(newname, '%')) {
835 err = dev_alloc_name(dev, newname);
836 if (err < 0)
837 return err;
838 strcpy(newname, dev->name);
839 }
840 else if (__dev_get_by_name(newname))
841 return -EEXIST;
842 else
843 strlcpy(dev->name, newname, IFNAMSIZ);
844
845 rollback:
846 device_rename(&dev->dev, dev->name);
847
848 write_lock_bh(&dev_base_lock);
849 hlist_del(&dev->name_hlist);
850 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
851 write_unlock_bh(&dev_base_lock);
852
853 ret = raw_notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
854 ret = notifier_to_errno(ret);
855
856 if (ret) {
857 if (err) {
858 printk(KERN_ERR
859 "%s: name change rollback failed: %d.\n",
860 dev->name, ret);
861 } else {
862 err = ret;
863 memcpy(dev->name, oldname, IFNAMSIZ);
864 goto rollback;
865 }
866 }
867
868 return err;
869 }
870
871 /**
872 * netdev_features_change - device changes features
873 * @dev: device to cause notification
874 *
875 * Called to indicate a device has changed features.
876 */
877 void netdev_features_change(struct net_device *dev)
878 {
879 raw_notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
880 }
881 EXPORT_SYMBOL(netdev_features_change);
882
883 /**
884 * netdev_state_change - device changes state
885 * @dev: device to cause notification
886 *
887 * Called to indicate a device has changed state. This function calls
888 * the notifier chains for netdev_chain and sends a NEWLINK message
889 * to the routing socket.
890 */
891 void netdev_state_change(struct net_device *dev)
892 {
893 if (dev->flags & IFF_UP) {
894 raw_notifier_call_chain(&netdev_chain,
895 NETDEV_CHANGE, dev);
896 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
897 }
898 }
899
900 /**
901 * dev_load - load a network module
902 * @name: name of interface
903 *
904 * If a network interface is not present and the process has suitable
905 * privileges this function loads the module. If module loading is not
906 * available in this kernel then it becomes a nop.
907 */
908
909 void dev_load(const char *name)
910 {
911 struct net_device *dev;
912
913 read_lock(&dev_base_lock);
914 dev = __dev_get_by_name(name);
915 read_unlock(&dev_base_lock);
916
917 if (!dev && capable(CAP_SYS_MODULE))
918 request_module("%s", name);
919 }
920
921 static int default_rebuild_header(struct sk_buff *skb)
922 {
923 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
924 skb->dev ? skb->dev->name : "NULL!!!");
925 kfree_skb(skb);
926 return 1;
927 }
928
929 /**
930 * dev_open - prepare an interface for use.
931 * @dev: device to open
932 *
933 * Takes a device from down to up state. The device's private open
934 * function is invoked and then the multicast lists are loaded. Finally
935 * the device is moved into the up state and a %NETDEV_UP message is
936 * sent to the netdev notifier chain.
937 *
938 * Calling this function on an active interface is a nop. On a failure
939 * a negative errno code is returned.
940 */
941 int dev_open(struct net_device *dev)
942 {
943 int ret = 0;
944
945 /*
946 * Is it already up?
947 */
948
949 if (dev->flags & IFF_UP)
950 return 0;
951
952 /*
953 * Is it even present?
954 */
955 if (!netif_device_present(dev))
956 return -ENODEV;
957
958 /*
959 * Call device private open method
960 */
961 set_bit(__LINK_STATE_START, &dev->state);
962 if (dev->open) {
963 ret = dev->open(dev);
964 if (ret)
965 clear_bit(__LINK_STATE_START, &dev->state);
966 }
967
968 /*
969 * If it went open OK then:
970 */
971
972 if (!ret) {
973 /*
974 * Set the flags.
975 */
976 dev->flags |= IFF_UP;
977
978 /*
979 * Initialize multicasting status
980 */
981 dev_set_rx_mode(dev);
982
983 /*
984 * Wakeup transmit queue engine
985 */
986 dev_activate(dev);
987
988 /*
989 * ... and announce new interface.
990 */
991 raw_notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
992 }
993 return ret;
994 }
995
996 /**
997 * dev_close - shutdown an interface.
998 * @dev: device to shutdown
999 *
1000 * This function moves an active device into down state. A
1001 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1002 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1003 * chain.
1004 */
1005 int dev_close(struct net_device *dev)
1006 {
1007 if (!(dev->flags & IFF_UP))
1008 return 0;
1009
1010 /*
1011 * Tell people we are going down, so that they can
1012 * prepare to death, when device is still operating.
1013 */
1014 raw_notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
1015
1016 dev_deactivate(dev);
1017
1018 clear_bit(__LINK_STATE_START, &dev->state);
1019
1020 /* Synchronize to scheduled poll. We cannot touch poll list,
1021 * it can be even on different cpu. So just clear netif_running(),
1022 * and wait when poll really will happen. Actually, the best place
1023 * for this is inside dev->stop() after device stopped its irq
1024 * engine, but this requires more changes in devices. */
1025
1026 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1027 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
1028 /* No hurry. */
1029 msleep(1);
1030 }
1031
1032 /*
1033 * Call the device specific close. This cannot fail.
1034 * Only if device is UP
1035 *
1036 * We allow it to be called even after a DETACH hot-plug
1037 * event.
1038 */
1039 if (dev->stop)
1040 dev->stop(dev);
1041
1042 /*
1043 * Device is now down.
1044 */
1045
1046 dev->flags &= ~IFF_UP;
1047
1048 /*
1049 * Tell people we are down
1050 */
1051 raw_notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
1052
1053 return 0;
1054 }
1055
1056
1057 /*
1058 * Device change register/unregister. These are not inline or static
1059 * as we export them to the world.
1060 */
1061
1062 /**
1063 * register_netdevice_notifier - register a network notifier block
1064 * @nb: notifier
1065 *
1066 * Register a notifier to be called when network device events occur.
1067 * The notifier passed is linked into the kernel structures and must
1068 * not be reused until it has been unregistered. A negative errno code
1069 * is returned on a failure.
1070 *
1071 * When registered all registration and up events are replayed
1072 * to the new notifier to allow device to have a race free
1073 * view of the network device list.
1074 */
1075
1076 int register_netdevice_notifier(struct notifier_block *nb)
1077 {
1078 struct net_device *dev;
1079 struct net_device *last;
1080 int err;
1081
1082 rtnl_lock();
1083 err = raw_notifier_chain_register(&netdev_chain, nb);
1084 if (err)
1085 goto unlock;
1086
1087 for_each_netdev(dev) {
1088 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1089 err = notifier_to_errno(err);
1090 if (err)
1091 goto rollback;
1092
1093 if (!(dev->flags & IFF_UP))
1094 continue;
1095
1096 nb->notifier_call(nb, NETDEV_UP, dev);
1097 }
1098
1099 unlock:
1100 rtnl_unlock();
1101 return err;
1102
1103 rollback:
1104 last = dev;
1105 for_each_netdev(dev) {
1106 if (dev == last)
1107 break;
1108
1109 if (dev->flags & IFF_UP) {
1110 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1111 nb->notifier_call(nb, NETDEV_DOWN, dev);
1112 }
1113 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1114 }
1115 goto unlock;
1116 }
1117
1118 /**
1119 * unregister_netdevice_notifier - unregister a network notifier block
1120 * @nb: notifier
1121 *
1122 * Unregister a notifier previously registered by
1123 * register_netdevice_notifier(). The notifier is unlinked into the
1124 * kernel structures and may then be reused. A negative errno code
1125 * is returned on a failure.
1126 */
1127
1128 int unregister_netdevice_notifier(struct notifier_block *nb)
1129 {
1130 int err;
1131
1132 rtnl_lock();
1133 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1134 rtnl_unlock();
1135 return err;
1136 }
1137
1138 /**
1139 * call_netdevice_notifiers - call all network notifier blocks
1140 * @val: value passed unmodified to notifier function
1141 * @v: pointer passed unmodified to notifier function
1142 *
1143 * Call all network notifier blocks. Parameters and return value
1144 * are as for raw_notifier_call_chain().
1145 */
1146
1147 int call_netdevice_notifiers(unsigned long val, void *v)
1148 {
1149 return raw_notifier_call_chain(&netdev_chain, val, v);
1150 }
1151
1152 /* When > 0 there are consumers of rx skb time stamps */
1153 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1154
1155 void net_enable_timestamp(void)
1156 {
1157 atomic_inc(&netstamp_needed);
1158 }
1159
1160 void net_disable_timestamp(void)
1161 {
1162 atomic_dec(&netstamp_needed);
1163 }
1164
1165 static inline void net_timestamp(struct sk_buff *skb)
1166 {
1167 if (atomic_read(&netstamp_needed))
1168 __net_timestamp(skb);
1169 else
1170 skb->tstamp.tv64 = 0;
1171 }
1172
1173 /*
1174 * Support routine. Sends outgoing frames to any network
1175 * taps currently in use.
1176 */
1177
1178 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1179 {
1180 struct packet_type *ptype;
1181
1182 net_timestamp(skb);
1183
1184 rcu_read_lock();
1185 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1186 /* Never send packets back to the socket
1187 * they originated from - MvS (miquels@drinkel.ow.org)
1188 */
1189 if ((ptype->dev == dev || !ptype->dev) &&
1190 (ptype->af_packet_priv == NULL ||
1191 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1192 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1193 if (!skb2)
1194 break;
1195
1196 /* skb->nh should be correctly
1197 set by sender, so that the second statement is
1198 just protection against buggy protocols.
1199 */
1200 skb_reset_mac_header(skb2);
1201
1202 if (skb_network_header(skb2) < skb2->data ||
1203 skb2->network_header > skb2->tail) {
1204 if (net_ratelimit())
1205 printk(KERN_CRIT "protocol %04x is "
1206 "buggy, dev %s\n",
1207 skb2->protocol, dev->name);
1208 skb_reset_network_header(skb2);
1209 }
1210
1211 skb2->transport_header = skb2->network_header;
1212 skb2->pkt_type = PACKET_OUTGOING;
1213 ptype->func(skb2, skb->dev, ptype, skb->dev);
1214 }
1215 }
1216 rcu_read_unlock();
1217 }
1218
1219
1220 void __netif_schedule(struct net_device *dev)
1221 {
1222 if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) {
1223 unsigned long flags;
1224 struct softnet_data *sd;
1225
1226 local_irq_save(flags);
1227 sd = &__get_cpu_var(softnet_data);
1228 dev->next_sched = sd->output_queue;
1229 sd->output_queue = dev;
1230 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1231 local_irq_restore(flags);
1232 }
1233 }
1234 EXPORT_SYMBOL(__netif_schedule);
1235
1236 void __netif_rx_schedule(struct net_device *dev)
1237 {
1238 unsigned long flags;
1239
1240 local_irq_save(flags);
1241 dev_hold(dev);
1242 list_add_tail(&dev->poll_list, &__get_cpu_var(softnet_data).poll_list);
1243 if (dev->quota < 0)
1244 dev->quota += dev->weight;
1245 else
1246 dev->quota = dev->weight;
1247 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1248 local_irq_restore(flags);
1249 }
1250 EXPORT_SYMBOL(__netif_rx_schedule);
1251
1252 void dev_kfree_skb_any(struct sk_buff *skb)
1253 {
1254 if (in_irq() || irqs_disabled())
1255 dev_kfree_skb_irq(skb);
1256 else
1257 dev_kfree_skb(skb);
1258 }
1259 EXPORT_SYMBOL(dev_kfree_skb_any);
1260
1261
1262 /* Hot-plugging. */
1263 void netif_device_detach(struct net_device *dev)
1264 {
1265 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1266 netif_running(dev)) {
1267 netif_stop_queue(dev);
1268 }
1269 }
1270 EXPORT_SYMBOL(netif_device_detach);
1271
1272 void netif_device_attach(struct net_device *dev)
1273 {
1274 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1275 netif_running(dev)) {
1276 netif_wake_queue(dev);
1277 __netdev_watchdog_up(dev);
1278 }
1279 }
1280 EXPORT_SYMBOL(netif_device_attach);
1281
1282
1283 /*
1284 * Invalidate hardware checksum when packet is to be mangled, and
1285 * complete checksum manually on outgoing path.
1286 */
1287 int skb_checksum_help(struct sk_buff *skb)
1288 {
1289 __wsum csum;
1290 int ret = 0, offset;
1291
1292 if (skb->ip_summed == CHECKSUM_COMPLETE)
1293 goto out_set_summed;
1294
1295 if (unlikely(skb_shinfo(skb)->gso_size)) {
1296 /* Let GSO fix up the checksum. */
1297 goto out_set_summed;
1298 }
1299
1300 if (skb_cloned(skb)) {
1301 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1302 if (ret)
1303 goto out;
1304 }
1305
1306 offset = skb->csum_start - skb_headroom(skb);
1307 BUG_ON(offset > (int)skb->len);
1308 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1309
1310 offset = skb_headlen(skb) - offset;
1311 BUG_ON(offset <= 0);
1312 BUG_ON(skb->csum_offset + 2 > offset);
1313
1314 *(__sum16 *)(skb->head + skb->csum_start + skb->csum_offset) =
1315 csum_fold(csum);
1316 out_set_summed:
1317 skb->ip_summed = CHECKSUM_NONE;
1318 out:
1319 return ret;
1320 }
1321
1322 /**
1323 * skb_gso_segment - Perform segmentation on skb.
1324 * @skb: buffer to segment
1325 * @features: features for the output path (see dev->features)
1326 *
1327 * This function segments the given skb and returns a list of segments.
1328 *
1329 * It may return NULL if the skb requires no segmentation. This is
1330 * only possible when GSO is used for verifying header integrity.
1331 */
1332 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1333 {
1334 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1335 struct packet_type *ptype;
1336 __be16 type = skb->protocol;
1337 int err;
1338
1339 BUG_ON(skb_shinfo(skb)->frag_list);
1340
1341 skb_reset_mac_header(skb);
1342 skb->mac_len = skb->network_header - skb->mac_header;
1343 __skb_pull(skb, skb->mac_len);
1344
1345 if (WARN_ON(skb->ip_summed != CHECKSUM_PARTIAL)) {
1346 if (skb_header_cloned(skb) &&
1347 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1348 return ERR_PTR(err);
1349 }
1350
1351 rcu_read_lock();
1352 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type) & 15], list) {
1353 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1354 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1355 err = ptype->gso_send_check(skb);
1356 segs = ERR_PTR(err);
1357 if (err || skb_gso_ok(skb, features))
1358 break;
1359 __skb_push(skb, (skb->data -
1360 skb_network_header(skb)));
1361 }
1362 segs = ptype->gso_segment(skb, features);
1363 break;
1364 }
1365 }
1366 rcu_read_unlock();
1367
1368 __skb_push(skb, skb->data - skb_mac_header(skb));
1369
1370 return segs;
1371 }
1372
1373 EXPORT_SYMBOL(skb_gso_segment);
1374
1375 /* Take action when hardware reception checksum errors are detected. */
1376 #ifdef CONFIG_BUG
1377 void netdev_rx_csum_fault(struct net_device *dev)
1378 {
1379 if (net_ratelimit()) {
1380 printk(KERN_ERR "%s: hw csum failure.\n",
1381 dev ? dev->name : "<unknown>");
1382 dump_stack();
1383 }
1384 }
1385 EXPORT_SYMBOL(netdev_rx_csum_fault);
1386 #endif
1387
1388 /* Actually, we should eliminate this check as soon as we know, that:
1389 * 1. IOMMU is present and allows to map all the memory.
1390 * 2. No high memory really exists on this machine.
1391 */
1392
1393 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1394 {
1395 #ifdef CONFIG_HIGHMEM
1396 int i;
1397
1398 if (dev->features & NETIF_F_HIGHDMA)
1399 return 0;
1400
1401 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1402 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1403 return 1;
1404
1405 #endif
1406 return 0;
1407 }
1408
1409 struct dev_gso_cb {
1410 void (*destructor)(struct sk_buff *skb);
1411 };
1412
1413 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1414
1415 static void dev_gso_skb_destructor(struct sk_buff *skb)
1416 {
1417 struct dev_gso_cb *cb;
1418
1419 do {
1420 struct sk_buff *nskb = skb->next;
1421
1422 skb->next = nskb->next;
1423 nskb->next = NULL;
1424 kfree_skb(nskb);
1425 } while (skb->next);
1426
1427 cb = DEV_GSO_CB(skb);
1428 if (cb->destructor)
1429 cb->destructor(skb);
1430 }
1431
1432 /**
1433 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1434 * @skb: buffer to segment
1435 *
1436 * This function segments the given skb and stores the list of segments
1437 * in skb->next.
1438 */
1439 static int dev_gso_segment(struct sk_buff *skb)
1440 {
1441 struct net_device *dev = skb->dev;
1442 struct sk_buff *segs;
1443 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1444 NETIF_F_SG : 0);
1445
1446 segs = skb_gso_segment(skb, features);
1447
1448 /* Verifying header integrity only. */
1449 if (!segs)
1450 return 0;
1451
1452 if (unlikely(IS_ERR(segs)))
1453 return PTR_ERR(segs);
1454
1455 skb->next = segs;
1456 DEV_GSO_CB(skb)->destructor = skb->destructor;
1457 skb->destructor = dev_gso_skb_destructor;
1458
1459 return 0;
1460 }
1461
1462 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
1463 {
1464 if (likely(!skb->next)) {
1465 if (!list_empty(&ptype_all))
1466 dev_queue_xmit_nit(skb, dev);
1467
1468 if (netif_needs_gso(dev, skb)) {
1469 if (unlikely(dev_gso_segment(skb)))
1470 goto out_kfree_skb;
1471 if (skb->next)
1472 goto gso;
1473 }
1474
1475 return dev->hard_start_xmit(skb, dev);
1476 }
1477
1478 gso:
1479 do {
1480 struct sk_buff *nskb = skb->next;
1481 int rc;
1482
1483 skb->next = nskb->next;
1484 nskb->next = NULL;
1485 rc = dev->hard_start_xmit(nskb, dev);
1486 if (unlikely(rc)) {
1487 nskb->next = skb->next;
1488 skb->next = nskb;
1489 return rc;
1490 }
1491 if (unlikely((netif_queue_stopped(dev) ||
1492 netif_subqueue_stopped(dev, skb->queue_mapping)) &&
1493 skb->next))
1494 return NETDEV_TX_BUSY;
1495 } while (skb->next);
1496
1497 skb->destructor = DEV_GSO_CB(skb)->destructor;
1498
1499 out_kfree_skb:
1500 kfree_skb(skb);
1501 return 0;
1502 }
1503
1504 #define HARD_TX_LOCK(dev, cpu) { \
1505 if ((dev->features & NETIF_F_LLTX) == 0) { \
1506 netif_tx_lock(dev); \
1507 } \
1508 }
1509
1510 #define HARD_TX_UNLOCK(dev) { \
1511 if ((dev->features & NETIF_F_LLTX) == 0) { \
1512 netif_tx_unlock(dev); \
1513 } \
1514 }
1515
1516 /**
1517 * dev_queue_xmit - transmit a buffer
1518 * @skb: buffer to transmit
1519 *
1520 * Queue a buffer for transmission to a network device. The caller must
1521 * have set the device and priority and built the buffer before calling
1522 * this function. The function can be called from an interrupt.
1523 *
1524 * A negative errno code is returned on a failure. A success does not
1525 * guarantee the frame will be transmitted as it may be dropped due
1526 * to congestion or traffic shaping.
1527 *
1528 * -----------------------------------------------------------------------------------
1529 * I notice this method can also return errors from the queue disciplines,
1530 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1531 * be positive.
1532 *
1533 * Regardless of the return value, the skb is consumed, so it is currently
1534 * difficult to retry a send to this method. (You can bump the ref count
1535 * before sending to hold a reference for retry if you are careful.)
1536 *
1537 * When calling this method, interrupts MUST be enabled. This is because
1538 * the BH enable code must have IRQs enabled so that it will not deadlock.
1539 * --BLG
1540 */
1541
1542 int dev_queue_xmit(struct sk_buff *skb)
1543 {
1544 struct net_device *dev = skb->dev;
1545 struct Qdisc *q;
1546 int rc = -ENOMEM;
1547
1548 /* GSO will handle the following emulations directly. */
1549 if (netif_needs_gso(dev, skb))
1550 goto gso;
1551
1552 if (skb_shinfo(skb)->frag_list &&
1553 !(dev->features & NETIF_F_FRAGLIST) &&
1554 __skb_linearize(skb))
1555 goto out_kfree_skb;
1556
1557 /* Fragmented skb is linearized if device does not support SG,
1558 * or if at least one of fragments is in highmem and device
1559 * does not support DMA from it.
1560 */
1561 if (skb_shinfo(skb)->nr_frags &&
1562 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1563 __skb_linearize(skb))
1564 goto out_kfree_skb;
1565
1566 /* If packet is not checksummed and device does not support
1567 * checksumming for this protocol, complete checksumming here.
1568 */
1569 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1570 skb_set_transport_header(skb, skb->csum_start -
1571 skb_headroom(skb));
1572
1573 if (!(dev->features & NETIF_F_GEN_CSUM) &&
1574 !((dev->features & NETIF_F_IP_CSUM) &&
1575 skb->protocol == htons(ETH_P_IP)) &&
1576 !((dev->features & NETIF_F_IPV6_CSUM) &&
1577 skb->protocol == htons(ETH_P_IPV6)))
1578 if (skb_checksum_help(skb))
1579 goto out_kfree_skb;
1580 }
1581
1582 gso:
1583 spin_lock_prefetch(&dev->queue_lock);
1584
1585 /* Disable soft irqs for various locks below. Also
1586 * stops preemption for RCU.
1587 */
1588 rcu_read_lock_bh();
1589
1590 /* Updates of qdisc are serialized by queue_lock.
1591 * The struct Qdisc which is pointed to by qdisc is now a
1592 * rcu structure - it may be accessed without acquiring
1593 * a lock (but the structure may be stale.) The freeing of the
1594 * qdisc will be deferred until it's known that there are no
1595 * more references to it.
1596 *
1597 * If the qdisc has an enqueue function, we still need to
1598 * hold the queue_lock before calling it, since queue_lock
1599 * also serializes access to the device queue.
1600 */
1601
1602 q = rcu_dereference(dev->qdisc);
1603 #ifdef CONFIG_NET_CLS_ACT
1604 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1605 #endif
1606 if (q->enqueue) {
1607 /* Grab device queue */
1608 spin_lock(&dev->queue_lock);
1609 q = dev->qdisc;
1610 if (q->enqueue) {
1611 /* reset queue_mapping to zero */
1612 skb->queue_mapping = 0;
1613 rc = q->enqueue(skb, q);
1614 qdisc_run(dev);
1615 spin_unlock(&dev->queue_lock);
1616
1617 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1618 goto out;
1619 }
1620 spin_unlock(&dev->queue_lock);
1621 }
1622
1623 /* The device has no queue. Common case for software devices:
1624 loopback, all the sorts of tunnels...
1625
1626 Really, it is unlikely that netif_tx_lock protection is necessary
1627 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1628 counters.)
1629 However, it is possible, that they rely on protection
1630 made by us here.
1631
1632 Check this and shot the lock. It is not prone from deadlocks.
1633 Either shot noqueue qdisc, it is even simpler 8)
1634 */
1635 if (dev->flags & IFF_UP) {
1636 int cpu = smp_processor_id(); /* ok because BHs are off */
1637
1638 if (dev->xmit_lock_owner != cpu) {
1639
1640 HARD_TX_LOCK(dev, cpu);
1641
1642 if (!netif_queue_stopped(dev) &&
1643 !netif_subqueue_stopped(dev, skb->queue_mapping)) {
1644 rc = 0;
1645 if (!dev_hard_start_xmit(skb, dev)) {
1646 HARD_TX_UNLOCK(dev);
1647 goto out;
1648 }
1649 }
1650 HARD_TX_UNLOCK(dev);
1651 if (net_ratelimit())
1652 printk(KERN_CRIT "Virtual device %s asks to "
1653 "queue packet!\n", dev->name);
1654 } else {
1655 /* Recursion is detected! It is possible,
1656 * unfortunately */
1657 if (net_ratelimit())
1658 printk(KERN_CRIT "Dead loop on virtual device "
1659 "%s, fix it urgently!\n", dev->name);
1660 }
1661 }
1662
1663 rc = -ENETDOWN;
1664 rcu_read_unlock_bh();
1665
1666 out_kfree_skb:
1667 kfree_skb(skb);
1668 return rc;
1669 out:
1670 rcu_read_unlock_bh();
1671 return rc;
1672 }
1673
1674
1675 /*=======================================================================
1676 Receiver routines
1677 =======================================================================*/
1678
1679 int netdev_max_backlog __read_mostly = 1000;
1680 int netdev_budget __read_mostly = 300;
1681 int weight_p __read_mostly = 64; /* old backlog weight */
1682
1683 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1684
1685
1686 /**
1687 * netif_rx - post buffer to the network code
1688 * @skb: buffer to post
1689 *
1690 * This function receives a packet from a device driver and queues it for
1691 * the upper (protocol) levels to process. It always succeeds. The buffer
1692 * may be dropped during processing for congestion control or by the
1693 * protocol layers.
1694 *
1695 * return values:
1696 * NET_RX_SUCCESS (no congestion)
1697 * NET_RX_CN_LOW (low congestion)
1698 * NET_RX_CN_MOD (moderate congestion)
1699 * NET_RX_CN_HIGH (high congestion)
1700 * NET_RX_DROP (packet was dropped)
1701 *
1702 */
1703
1704 int netif_rx(struct sk_buff *skb)
1705 {
1706 struct softnet_data *queue;
1707 unsigned long flags;
1708
1709 /* if netpoll wants it, pretend we never saw it */
1710 if (netpoll_rx(skb))
1711 return NET_RX_DROP;
1712
1713 if (!skb->tstamp.tv64)
1714 net_timestamp(skb);
1715
1716 /*
1717 * The code is rearranged so that the path is the most
1718 * short when CPU is congested, but is still operating.
1719 */
1720 local_irq_save(flags);
1721 queue = &__get_cpu_var(softnet_data);
1722
1723 __get_cpu_var(netdev_rx_stat).total++;
1724 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1725 if (queue->input_pkt_queue.qlen) {
1726 enqueue:
1727 dev_hold(skb->dev);
1728 __skb_queue_tail(&queue->input_pkt_queue, skb);
1729 local_irq_restore(flags);
1730 return NET_RX_SUCCESS;
1731 }
1732
1733 netif_rx_schedule(&queue->backlog_dev);
1734 goto enqueue;
1735 }
1736
1737 __get_cpu_var(netdev_rx_stat).dropped++;
1738 local_irq_restore(flags);
1739
1740 kfree_skb(skb);
1741 return NET_RX_DROP;
1742 }
1743
1744 int netif_rx_ni(struct sk_buff *skb)
1745 {
1746 int err;
1747
1748 preempt_disable();
1749 err = netif_rx(skb);
1750 if (local_softirq_pending())
1751 do_softirq();
1752 preempt_enable();
1753
1754 return err;
1755 }
1756
1757 EXPORT_SYMBOL(netif_rx_ni);
1758
1759 static inline struct net_device *skb_bond(struct sk_buff *skb)
1760 {
1761 struct net_device *dev = skb->dev;
1762
1763 if (dev->master) {
1764 if (skb_bond_should_drop(skb)) {
1765 kfree_skb(skb);
1766 return NULL;
1767 }
1768 skb->dev = dev->master;
1769 }
1770
1771 return dev;
1772 }
1773
1774 static void net_tx_action(struct softirq_action *h)
1775 {
1776 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1777
1778 if (sd->completion_queue) {
1779 struct sk_buff *clist;
1780
1781 local_irq_disable();
1782 clist = sd->completion_queue;
1783 sd->completion_queue = NULL;
1784 local_irq_enable();
1785
1786 while (clist) {
1787 struct sk_buff *skb = clist;
1788 clist = clist->next;
1789
1790 BUG_TRAP(!atomic_read(&skb->users));
1791 __kfree_skb(skb);
1792 }
1793 }
1794
1795 if (sd->output_queue) {
1796 struct net_device *head;
1797
1798 local_irq_disable();
1799 head = sd->output_queue;
1800 sd->output_queue = NULL;
1801 local_irq_enable();
1802
1803 while (head) {
1804 struct net_device *dev = head;
1805 head = head->next_sched;
1806
1807 smp_mb__before_clear_bit();
1808 clear_bit(__LINK_STATE_SCHED, &dev->state);
1809
1810 if (spin_trylock(&dev->queue_lock)) {
1811 qdisc_run(dev);
1812 spin_unlock(&dev->queue_lock);
1813 } else {
1814 netif_schedule(dev);
1815 }
1816 }
1817 }
1818 }
1819
1820 static inline int deliver_skb(struct sk_buff *skb,
1821 struct packet_type *pt_prev,
1822 struct net_device *orig_dev)
1823 {
1824 atomic_inc(&skb->users);
1825 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1826 }
1827
1828 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1829 /* These hooks defined here for ATM */
1830 struct net_bridge;
1831 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1832 unsigned char *addr);
1833 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent) __read_mostly;
1834
1835 /*
1836 * If bridge module is loaded call bridging hook.
1837 * returns NULL if packet was consumed.
1838 */
1839 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
1840 struct sk_buff *skb) __read_mostly;
1841 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
1842 struct packet_type **pt_prev, int *ret,
1843 struct net_device *orig_dev)
1844 {
1845 struct net_bridge_port *port;
1846
1847 if (skb->pkt_type == PACKET_LOOPBACK ||
1848 (port = rcu_dereference(skb->dev->br_port)) == NULL)
1849 return skb;
1850
1851 if (*pt_prev) {
1852 *ret = deliver_skb(skb, *pt_prev, orig_dev);
1853 *pt_prev = NULL;
1854 }
1855
1856 return br_handle_frame_hook(port, skb);
1857 }
1858 #else
1859 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
1860 #endif
1861
1862 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
1863 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
1864 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
1865
1866 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
1867 struct packet_type **pt_prev,
1868 int *ret,
1869 struct net_device *orig_dev)
1870 {
1871 if (skb->dev->macvlan_port == NULL)
1872 return skb;
1873
1874 if (*pt_prev) {
1875 *ret = deliver_skb(skb, *pt_prev, orig_dev);
1876 *pt_prev = NULL;
1877 }
1878 return macvlan_handle_frame_hook(skb);
1879 }
1880 #else
1881 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
1882 #endif
1883
1884 #ifdef CONFIG_NET_CLS_ACT
1885 /* TODO: Maybe we should just force sch_ingress to be compiled in
1886 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1887 * a compare and 2 stores extra right now if we dont have it on
1888 * but have CONFIG_NET_CLS_ACT
1889 * NOTE: This doesnt stop any functionality; if you dont have
1890 * the ingress scheduler, you just cant add policies on ingress.
1891 *
1892 */
1893 static int ing_filter(struct sk_buff *skb)
1894 {
1895 struct Qdisc *q;
1896 struct net_device *dev = skb->dev;
1897 int result = TC_ACT_OK;
1898
1899 if (dev->qdisc_ingress) {
1900 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1901 if (MAX_RED_LOOP < ttl++) {
1902 printk(KERN_WARNING "Redir loop detected Dropping packet (%d->%d)\n",
1903 skb->iif, skb->dev->ifindex);
1904 return TC_ACT_SHOT;
1905 }
1906
1907 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1908
1909 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1910
1911 spin_lock(&dev->ingress_lock);
1912 if ((q = dev->qdisc_ingress) != NULL)
1913 result = q->enqueue(skb, q);
1914 spin_unlock(&dev->ingress_lock);
1915
1916 }
1917
1918 return result;
1919 }
1920 #endif
1921
1922 int netif_receive_skb(struct sk_buff *skb)
1923 {
1924 struct packet_type *ptype, *pt_prev;
1925 struct net_device *orig_dev;
1926 int ret = NET_RX_DROP;
1927 __be16 type;
1928
1929 /* if we've gotten here through NAPI, check netpoll */
1930 if (skb->dev->poll && netpoll_rx(skb))
1931 return NET_RX_DROP;
1932
1933 if (!skb->tstamp.tv64)
1934 net_timestamp(skb);
1935
1936 if (!skb->iif)
1937 skb->iif = skb->dev->ifindex;
1938
1939 orig_dev = skb_bond(skb);
1940
1941 if (!orig_dev)
1942 return NET_RX_DROP;
1943
1944 __get_cpu_var(netdev_rx_stat).total++;
1945
1946 skb_reset_network_header(skb);
1947 skb_reset_transport_header(skb);
1948 skb->mac_len = skb->network_header - skb->mac_header;
1949
1950 pt_prev = NULL;
1951
1952 rcu_read_lock();
1953
1954 #ifdef CONFIG_NET_CLS_ACT
1955 if (skb->tc_verd & TC_NCLS) {
1956 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1957 goto ncls;
1958 }
1959 #endif
1960
1961 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1962 if (!ptype->dev || ptype->dev == skb->dev) {
1963 if (pt_prev)
1964 ret = deliver_skb(skb, pt_prev, orig_dev);
1965 pt_prev = ptype;
1966 }
1967 }
1968
1969 #ifdef CONFIG_NET_CLS_ACT
1970 if (pt_prev) {
1971 ret = deliver_skb(skb, pt_prev, orig_dev);
1972 pt_prev = NULL; /* noone else should process this after*/
1973 } else {
1974 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1975 }
1976
1977 ret = ing_filter(skb);
1978
1979 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1980 kfree_skb(skb);
1981 goto out;
1982 }
1983
1984 skb->tc_verd = 0;
1985 ncls:
1986 #endif
1987
1988 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
1989 if (!skb)
1990 goto out;
1991 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
1992 if (!skb)
1993 goto out;
1994
1995 type = skb->protocol;
1996 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1997 if (ptype->type == type &&
1998 (!ptype->dev || ptype->dev == skb->dev)) {
1999 if (pt_prev)
2000 ret = deliver_skb(skb, pt_prev, orig_dev);
2001 pt_prev = ptype;
2002 }
2003 }
2004
2005 if (pt_prev) {
2006 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2007 } else {
2008 kfree_skb(skb);
2009 /* Jamal, now you will not able to escape explaining
2010 * me how you were going to use this. :-)
2011 */
2012 ret = NET_RX_DROP;
2013 }
2014
2015 out:
2016 rcu_read_unlock();
2017 return ret;
2018 }
2019
2020 static int process_backlog(struct net_device *backlog_dev, int *budget)
2021 {
2022 int work = 0;
2023 int quota = min(backlog_dev->quota, *budget);
2024 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2025 unsigned long start_time = jiffies;
2026
2027 backlog_dev->weight = weight_p;
2028 for (;;) {
2029 struct sk_buff *skb;
2030 struct net_device *dev;
2031
2032 local_irq_disable();
2033 skb = __skb_dequeue(&queue->input_pkt_queue);
2034 if (!skb)
2035 goto job_done;
2036 local_irq_enable();
2037
2038 dev = skb->dev;
2039
2040 netif_receive_skb(skb);
2041
2042 dev_put(dev);
2043
2044 work++;
2045
2046 if (work >= quota || jiffies - start_time > 1)
2047 break;
2048
2049 }
2050
2051 backlog_dev->quota -= work;
2052 *budget -= work;
2053 return -1;
2054
2055 job_done:
2056 backlog_dev->quota -= work;
2057 *budget -= work;
2058
2059 list_del(&backlog_dev->poll_list);
2060 smp_mb__before_clear_bit();
2061 netif_poll_enable(backlog_dev);
2062
2063 local_irq_enable();
2064 return 0;
2065 }
2066
2067 static void net_rx_action(struct softirq_action *h)
2068 {
2069 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2070 unsigned long start_time = jiffies;
2071 int budget = netdev_budget;
2072 void *have;
2073
2074 local_irq_disable();
2075
2076 while (!list_empty(&queue->poll_list)) {
2077 struct net_device *dev;
2078
2079 if (budget <= 0 || jiffies - start_time > 1)
2080 goto softnet_break;
2081
2082 local_irq_enable();
2083
2084 dev = list_entry(queue->poll_list.next,
2085 struct net_device, poll_list);
2086 have = netpoll_poll_lock(dev);
2087
2088 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
2089 netpoll_poll_unlock(have);
2090 local_irq_disable();
2091 list_move_tail(&dev->poll_list, &queue->poll_list);
2092 if (dev->quota < 0)
2093 dev->quota += dev->weight;
2094 else
2095 dev->quota = dev->weight;
2096 } else {
2097 netpoll_poll_unlock(have);
2098 dev_put(dev);
2099 local_irq_disable();
2100 }
2101 }
2102 out:
2103 local_irq_enable();
2104 #ifdef CONFIG_NET_DMA
2105 /*
2106 * There may not be any more sk_buffs coming right now, so push
2107 * any pending DMA copies to hardware
2108 */
2109 if (!cpus_empty(net_dma.channel_mask)) {
2110 int chan_idx;
2111 for_each_cpu_mask(chan_idx, net_dma.channel_mask) {
2112 struct dma_chan *chan = net_dma.channels[chan_idx];
2113 if (chan)
2114 dma_async_memcpy_issue_pending(chan);
2115 }
2116 }
2117 #endif
2118 return;
2119
2120 softnet_break:
2121 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2122 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2123 goto out;
2124 }
2125
2126 static gifconf_func_t * gifconf_list [NPROTO];
2127
2128 /**
2129 * register_gifconf - register a SIOCGIF handler
2130 * @family: Address family
2131 * @gifconf: Function handler
2132 *
2133 * Register protocol dependent address dumping routines. The handler
2134 * that is passed must not be freed or reused until it has been replaced
2135 * by another handler.
2136 */
2137 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2138 {
2139 if (family >= NPROTO)
2140 return -EINVAL;
2141 gifconf_list[family] = gifconf;
2142 return 0;
2143 }
2144
2145
2146 /*
2147 * Map an interface index to its name (SIOCGIFNAME)
2148 */
2149
2150 /*
2151 * We need this ioctl for efficient implementation of the
2152 * if_indextoname() function required by the IPv6 API. Without
2153 * it, we would have to search all the interfaces to find a
2154 * match. --pb
2155 */
2156
2157 static int dev_ifname(struct ifreq __user *arg)
2158 {
2159 struct net_device *dev;
2160 struct ifreq ifr;
2161
2162 /*
2163 * Fetch the caller's info block.
2164 */
2165
2166 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2167 return -EFAULT;
2168
2169 read_lock(&dev_base_lock);
2170 dev = __dev_get_by_index(ifr.ifr_ifindex);
2171 if (!dev) {
2172 read_unlock(&dev_base_lock);
2173 return -ENODEV;
2174 }
2175
2176 strcpy(ifr.ifr_name, dev->name);
2177 read_unlock(&dev_base_lock);
2178
2179 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2180 return -EFAULT;
2181 return 0;
2182 }
2183
2184 /*
2185 * Perform a SIOCGIFCONF call. This structure will change
2186 * size eventually, and there is nothing I can do about it.
2187 * Thus we will need a 'compatibility mode'.
2188 */
2189
2190 static int dev_ifconf(char __user *arg)
2191 {
2192 struct ifconf ifc;
2193 struct net_device *dev;
2194 char __user *pos;
2195 int len;
2196 int total;
2197 int i;
2198
2199 /*
2200 * Fetch the caller's info block.
2201 */
2202
2203 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2204 return -EFAULT;
2205
2206 pos = ifc.ifc_buf;
2207 len = ifc.ifc_len;
2208
2209 /*
2210 * Loop over the interfaces, and write an info block for each.
2211 */
2212
2213 total = 0;
2214 for_each_netdev(dev) {
2215 for (i = 0; i < NPROTO; i++) {
2216 if (gifconf_list[i]) {
2217 int done;
2218 if (!pos)
2219 done = gifconf_list[i](dev, NULL, 0);
2220 else
2221 done = gifconf_list[i](dev, pos + total,
2222 len - total);
2223 if (done < 0)
2224 return -EFAULT;
2225 total += done;
2226 }
2227 }
2228 }
2229
2230 /*
2231 * All done. Write the updated control block back to the caller.
2232 */
2233 ifc.ifc_len = total;
2234
2235 /*
2236 * Both BSD and Solaris return 0 here, so we do too.
2237 */
2238 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2239 }
2240
2241 #ifdef CONFIG_PROC_FS
2242 /*
2243 * This is invoked by the /proc filesystem handler to display a device
2244 * in detail.
2245 */
2246 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2247 {
2248 loff_t off;
2249 struct net_device *dev;
2250
2251 read_lock(&dev_base_lock);
2252 if (!*pos)
2253 return SEQ_START_TOKEN;
2254
2255 off = 1;
2256 for_each_netdev(dev)
2257 if (off++ == *pos)
2258 return dev;
2259
2260 return NULL;
2261 }
2262
2263 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2264 {
2265 ++*pos;
2266 return v == SEQ_START_TOKEN ?
2267 first_net_device() : next_net_device((struct net_device *)v);
2268 }
2269
2270 void dev_seq_stop(struct seq_file *seq, void *v)
2271 {
2272 read_unlock(&dev_base_lock);
2273 }
2274
2275 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2276 {
2277 struct net_device_stats *stats = dev->get_stats(dev);
2278
2279 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2280 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2281 dev->name, stats->rx_bytes, stats->rx_packets,
2282 stats->rx_errors,
2283 stats->rx_dropped + stats->rx_missed_errors,
2284 stats->rx_fifo_errors,
2285 stats->rx_length_errors + stats->rx_over_errors +
2286 stats->rx_crc_errors + stats->rx_frame_errors,
2287 stats->rx_compressed, stats->multicast,
2288 stats->tx_bytes, stats->tx_packets,
2289 stats->tx_errors, stats->tx_dropped,
2290 stats->tx_fifo_errors, stats->collisions,
2291 stats->tx_carrier_errors +
2292 stats->tx_aborted_errors +
2293 stats->tx_window_errors +
2294 stats->tx_heartbeat_errors,
2295 stats->tx_compressed);
2296 }
2297
2298 /*
2299 * Called from the PROCfs module. This now uses the new arbitrary sized
2300 * /proc/net interface to create /proc/net/dev
2301 */
2302 static int dev_seq_show(struct seq_file *seq, void *v)
2303 {
2304 if (v == SEQ_START_TOKEN)
2305 seq_puts(seq, "Inter-| Receive "
2306 " | Transmit\n"
2307 " face |bytes packets errs drop fifo frame "
2308 "compressed multicast|bytes packets errs "
2309 "drop fifo colls carrier compressed\n");
2310 else
2311 dev_seq_printf_stats(seq, v);
2312 return 0;
2313 }
2314
2315 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2316 {
2317 struct netif_rx_stats *rc = NULL;
2318
2319 while (*pos < NR_CPUS)
2320 if (cpu_online(*pos)) {
2321 rc = &per_cpu(netdev_rx_stat, *pos);
2322 break;
2323 } else
2324 ++*pos;
2325 return rc;
2326 }
2327
2328 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2329 {
2330 return softnet_get_online(pos);
2331 }
2332
2333 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2334 {
2335 ++*pos;
2336 return softnet_get_online(pos);
2337 }
2338
2339 static void softnet_seq_stop(struct seq_file *seq, void *v)
2340 {
2341 }
2342
2343 static int softnet_seq_show(struct seq_file *seq, void *v)
2344 {
2345 struct netif_rx_stats *s = v;
2346
2347 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2348 s->total, s->dropped, s->time_squeeze, 0,
2349 0, 0, 0, 0, /* was fastroute */
2350 s->cpu_collision );
2351 return 0;
2352 }
2353
2354 static const struct seq_operations dev_seq_ops = {
2355 .start = dev_seq_start,
2356 .next = dev_seq_next,
2357 .stop = dev_seq_stop,
2358 .show = dev_seq_show,
2359 };
2360
2361 static int dev_seq_open(struct inode *inode, struct file *file)
2362 {
2363 return seq_open(file, &dev_seq_ops);
2364 }
2365
2366 static const struct file_operations dev_seq_fops = {
2367 .owner = THIS_MODULE,
2368 .open = dev_seq_open,
2369 .read = seq_read,
2370 .llseek = seq_lseek,
2371 .release = seq_release,
2372 };
2373
2374 static const struct seq_operations softnet_seq_ops = {
2375 .start = softnet_seq_start,
2376 .next = softnet_seq_next,
2377 .stop = softnet_seq_stop,
2378 .show = softnet_seq_show,
2379 };
2380
2381 static int softnet_seq_open(struct inode *inode, struct file *file)
2382 {
2383 return seq_open(file, &softnet_seq_ops);
2384 }
2385
2386 static const struct file_operations softnet_seq_fops = {
2387 .owner = THIS_MODULE,
2388 .open = softnet_seq_open,
2389 .read = seq_read,
2390 .llseek = seq_lseek,
2391 .release = seq_release,
2392 };
2393
2394 static void *ptype_get_idx(loff_t pos)
2395 {
2396 struct packet_type *pt = NULL;
2397 loff_t i = 0;
2398 int t;
2399
2400 list_for_each_entry_rcu(pt, &ptype_all, list) {
2401 if (i == pos)
2402 return pt;
2403 ++i;
2404 }
2405
2406 for (t = 0; t < 16; t++) {
2407 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
2408 if (i == pos)
2409 return pt;
2410 ++i;
2411 }
2412 }
2413 return NULL;
2414 }
2415
2416 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
2417 {
2418 rcu_read_lock();
2419 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
2420 }
2421
2422 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2423 {
2424 struct packet_type *pt;
2425 struct list_head *nxt;
2426 int hash;
2427
2428 ++*pos;
2429 if (v == SEQ_START_TOKEN)
2430 return ptype_get_idx(0);
2431
2432 pt = v;
2433 nxt = pt->list.next;
2434 if (pt->type == htons(ETH_P_ALL)) {
2435 if (nxt != &ptype_all)
2436 goto found;
2437 hash = 0;
2438 nxt = ptype_base[0].next;
2439 } else
2440 hash = ntohs(pt->type) & 15;
2441
2442 while (nxt == &ptype_base[hash]) {
2443 if (++hash >= 16)
2444 return NULL;
2445 nxt = ptype_base[hash].next;
2446 }
2447 found:
2448 return list_entry(nxt, struct packet_type, list);
2449 }
2450
2451 static void ptype_seq_stop(struct seq_file *seq, void *v)
2452 {
2453 rcu_read_unlock();
2454 }
2455
2456 static void ptype_seq_decode(struct seq_file *seq, void *sym)
2457 {
2458 #ifdef CONFIG_KALLSYMS
2459 unsigned long offset = 0, symsize;
2460 const char *symname;
2461 char *modname;
2462 char namebuf[128];
2463
2464 symname = kallsyms_lookup((unsigned long)sym, &symsize, &offset,
2465 &modname, namebuf);
2466
2467 if (symname) {
2468 char *delim = ":";
2469
2470 if (!modname)
2471 modname = delim = "";
2472 seq_printf(seq, "%s%s%s%s+0x%lx", delim, modname, delim,
2473 symname, offset);
2474 return;
2475 }
2476 #endif
2477
2478 seq_printf(seq, "[%p]", sym);
2479 }
2480
2481 static int ptype_seq_show(struct seq_file *seq, void *v)
2482 {
2483 struct packet_type *pt = v;
2484
2485 if (v == SEQ_START_TOKEN)
2486 seq_puts(seq, "Type Device Function\n");
2487 else {
2488 if (pt->type == htons(ETH_P_ALL))
2489 seq_puts(seq, "ALL ");
2490 else
2491 seq_printf(seq, "%04x", ntohs(pt->type));
2492
2493 seq_printf(seq, " %-8s ",
2494 pt->dev ? pt->dev->name : "");
2495 ptype_seq_decode(seq, pt->func);
2496 seq_putc(seq, '\n');
2497 }
2498
2499 return 0;
2500 }
2501
2502 static const struct seq_operations ptype_seq_ops = {
2503 .start = ptype_seq_start,
2504 .next = ptype_seq_next,
2505 .stop = ptype_seq_stop,
2506 .show = ptype_seq_show,
2507 };
2508
2509 static int ptype_seq_open(struct inode *inode, struct file *file)
2510 {
2511 return seq_open(file, &ptype_seq_ops);
2512 }
2513
2514 static const struct file_operations ptype_seq_fops = {
2515 .owner = THIS_MODULE,
2516 .open = ptype_seq_open,
2517 .read = seq_read,
2518 .llseek = seq_lseek,
2519 .release = seq_release,
2520 };
2521
2522
2523 static int __init dev_proc_init(void)
2524 {
2525 int rc = -ENOMEM;
2526
2527 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2528 goto out;
2529 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2530 goto out_dev;
2531 if (!proc_net_fops_create("ptype", S_IRUGO, &ptype_seq_fops))
2532 goto out_dev2;
2533
2534 if (wext_proc_init())
2535 goto out_softnet;
2536 rc = 0;
2537 out:
2538 return rc;
2539 out_softnet:
2540 proc_net_remove("ptype");
2541 out_dev2:
2542 proc_net_remove("softnet_stat");
2543 out_dev:
2544 proc_net_remove("dev");
2545 goto out;
2546 }
2547 #else
2548 #define dev_proc_init() 0
2549 #endif /* CONFIG_PROC_FS */
2550
2551
2552 /**
2553 * netdev_set_master - set up master/slave pair
2554 * @slave: slave device
2555 * @master: new master device
2556 *
2557 * Changes the master device of the slave. Pass %NULL to break the
2558 * bonding. The caller must hold the RTNL semaphore. On a failure
2559 * a negative errno code is returned. On success the reference counts
2560 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2561 * function returns zero.
2562 */
2563 int netdev_set_master(struct net_device *slave, struct net_device *master)
2564 {
2565 struct net_device *old = slave->master;
2566
2567 ASSERT_RTNL();
2568
2569 if (master) {
2570 if (old)
2571 return -EBUSY;
2572 dev_hold(master);
2573 }
2574
2575 slave->master = master;
2576
2577 synchronize_net();
2578
2579 if (old)
2580 dev_put(old);
2581
2582 if (master)
2583 slave->flags |= IFF_SLAVE;
2584 else
2585 slave->flags &= ~IFF_SLAVE;
2586
2587 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2588 return 0;
2589 }
2590
2591 static void __dev_set_promiscuity(struct net_device *dev, int inc)
2592 {
2593 unsigned short old_flags = dev->flags;
2594
2595 ASSERT_RTNL();
2596
2597 if ((dev->promiscuity += inc) == 0)
2598 dev->flags &= ~IFF_PROMISC;
2599 else
2600 dev->flags |= IFF_PROMISC;
2601 if (dev->flags != old_flags) {
2602 printk(KERN_INFO "device %s %s promiscuous mode\n",
2603 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2604 "left");
2605 audit_log(current->audit_context, GFP_ATOMIC,
2606 AUDIT_ANOM_PROMISCUOUS,
2607 "dev=%s prom=%d old_prom=%d auid=%u",
2608 dev->name, (dev->flags & IFF_PROMISC),
2609 (old_flags & IFF_PROMISC),
2610 audit_get_loginuid(current->audit_context));
2611
2612 if (dev->change_rx_flags)
2613 dev->change_rx_flags(dev, IFF_PROMISC);
2614 }
2615 }
2616
2617 /**
2618 * dev_set_promiscuity - update promiscuity count on a device
2619 * @dev: device
2620 * @inc: modifier
2621 *
2622 * Add or remove promiscuity from a device. While the count in the device
2623 * remains above zero the interface remains promiscuous. Once it hits zero
2624 * the device reverts back to normal filtering operation. A negative inc
2625 * value is used to drop promiscuity on the device.
2626 */
2627 void dev_set_promiscuity(struct net_device *dev, int inc)
2628 {
2629 unsigned short old_flags = dev->flags;
2630
2631 __dev_set_promiscuity(dev, inc);
2632 if (dev->flags != old_flags)
2633 dev_set_rx_mode(dev);
2634 }
2635
2636 /**
2637 * dev_set_allmulti - update allmulti count on a device
2638 * @dev: device
2639 * @inc: modifier
2640 *
2641 * Add or remove reception of all multicast frames to a device. While the
2642 * count in the device remains above zero the interface remains listening
2643 * to all interfaces. Once it hits zero the device reverts back to normal
2644 * filtering operation. A negative @inc value is used to drop the counter
2645 * when releasing a resource needing all multicasts.
2646 */
2647
2648 void dev_set_allmulti(struct net_device *dev, int inc)
2649 {
2650 unsigned short old_flags = dev->flags;
2651
2652 ASSERT_RTNL();
2653
2654 dev->flags |= IFF_ALLMULTI;
2655 if ((dev->allmulti += inc) == 0)
2656 dev->flags &= ~IFF_ALLMULTI;
2657 if (dev->flags ^ old_flags) {
2658 if (dev->change_rx_flags)
2659 dev->change_rx_flags(dev, IFF_ALLMULTI);
2660 dev_set_rx_mode(dev);
2661 }
2662 }
2663
2664 /*
2665 * Upload unicast and multicast address lists to device and
2666 * configure RX filtering. When the device doesn't support unicast
2667 * filtering it is put in promiscous mode while unicast addresses
2668 * are present.
2669 */
2670 void __dev_set_rx_mode(struct net_device *dev)
2671 {
2672 /* dev_open will call this function so the list will stay sane. */
2673 if (!(dev->flags&IFF_UP))
2674 return;
2675
2676 if (!netif_device_present(dev))
2677 return;
2678
2679 if (dev->set_rx_mode)
2680 dev->set_rx_mode(dev);
2681 else {
2682 /* Unicast addresses changes may only happen under the rtnl,
2683 * therefore calling __dev_set_promiscuity here is safe.
2684 */
2685 if (dev->uc_count > 0 && !dev->uc_promisc) {
2686 __dev_set_promiscuity(dev, 1);
2687 dev->uc_promisc = 1;
2688 } else if (dev->uc_count == 0 && dev->uc_promisc) {
2689 __dev_set_promiscuity(dev, -1);
2690 dev->uc_promisc = 0;
2691 }
2692
2693 if (dev->set_multicast_list)
2694 dev->set_multicast_list(dev);
2695 }
2696 }
2697
2698 void dev_set_rx_mode(struct net_device *dev)
2699 {
2700 netif_tx_lock_bh(dev);
2701 __dev_set_rx_mode(dev);
2702 netif_tx_unlock_bh(dev);
2703 }
2704
2705 int __dev_addr_delete(struct dev_addr_list **list, int *count,
2706 void *addr, int alen, int glbl)
2707 {
2708 struct dev_addr_list *da;
2709
2710 for (; (da = *list) != NULL; list = &da->next) {
2711 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
2712 alen == da->da_addrlen) {
2713 if (glbl) {
2714 int old_glbl = da->da_gusers;
2715 da->da_gusers = 0;
2716 if (old_glbl == 0)
2717 break;
2718 }
2719 if (--da->da_users)
2720 return 0;
2721
2722 *list = da->next;
2723 kfree(da);
2724 (*count)--;
2725 return 0;
2726 }
2727 }
2728 return -ENOENT;
2729 }
2730
2731 int __dev_addr_add(struct dev_addr_list **list, int *count,
2732 void *addr, int alen, int glbl)
2733 {
2734 struct dev_addr_list *da;
2735
2736 for (da = *list; da != NULL; da = da->next) {
2737 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
2738 da->da_addrlen == alen) {
2739 if (glbl) {
2740 int old_glbl = da->da_gusers;
2741 da->da_gusers = 1;
2742 if (old_glbl)
2743 return 0;
2744 }
2745 da->da_users++;
2746 return 0;
2747 }
2748 }
2749
2750 da = kmalloc(sizeof(*da), GFP_ATOMIC);
2751 if (da == NULL)
2752 return -ENOMEM;
2753 memcpy(da->da_addr, addr, alen);
2754 da->da_addrlen = alen;
2755 da->da_users = 1;
2756 da->da_gusers = glbl ? 1 : 0;
2757 da->next = *list;
2758 *list = da;
2759 (*count)++;
2760 return 0;
2761 }
2762
2763 /**
2764 * dev_unicast_delete - Release secondary unicast address.
2765 * @dev: device
2766 * @addr: address to delete
2767 * @alen: length of @addr
2768 *
2769 * Release reference to a secondary unicast address and remove it
2770 * from the device if the reference count drops to zero.
2771 *
2772 * The caller must hold the rtnl_mutex.
2773 */
2774 int dev_unicast_delete(struct net_device *dev, void *addr, int alen)
2775 {
2776 int err;
2777
2778 ASSERT_RTNL();
2779
2780 netif_tx_lock_bh(dev);
2781 err = __dev_addr_delete(&dev->uc_list, &dev->uc_count, addr, alen, 0);
2782 if (!err)
2783 __dev_set_rx_mode(dev);
2784 netif_tx_unlock_bh(dev);
2785 return err;
2786 }
2787 EXPORT_SYMBOL(dev_unicast_delete);
2788
2789 /**
2790 * dev_unicast_add - add a secondary unicast address
2791 * @dev: device
2792 * @addr: address to delete
2793 * @alen: length of @addr
2794 *
2795 * Add a secondary unicast address to the device or increase
2796 * the reference count if it already exists.
2797 *
2798 * The caller must hold the rtnl_mutex.
2799 */
2800 int dev_unicast_add(struct net_device *dev, void *addr, int alen)
2801 {
2802 int err;
2803
2804 ASSERT_RTNL();
2805
2806 netif_tx_lock_bh(dev);
2807 err = __dev_addr_add(&dev->uc_list, &dev->uc_count, addr, alen, 0);
2808 if (!err)
2809 __dev_set_rx_mode(dev);
2810 netif_tx_unlock_bh(dev);
2811 return err;
2812 }
2813 EXPORT_SYMBOL(dev_unicast_add);
2814
2815 static void __dev_addr_discard(struct dev_addr_list **list)
2816 {
2817 struct dev_addr_list *tmp;
2818
2819 while (*list != NULL) {
2820 tmp = *list;
2821 *list = tmp->next;
2822 if (tmp->da_users > tmp->da_gusers)
2823 printk("__dev_addr_discard: address leakage! "
2824 "da_users=%d\n", tmp->da_users);
2825 kfree(tmp);
2826 }
2827 }
2828
2829 static void dev_addr_discard(struct net_device *dev)
2830 {
2831 netif_tx_lock_bh(dev);
2832
2833 __dev_addr_discard(&dev->uc_list);
2834 dev->uc_count = 0;
2835
2836 __dev_addr_discard(&dev->mc_list);
2837 dev->mc_count = 0;
2838
2839 netif_tx_unlock_bh(dev);
2840 }
2841
2842 unsigned dev_get_flags(const struct net_device *dev)
2843 {
2844 unsigned flags;
2845
2846 flags = (dev->flags & ~(IFF_PROMISC |
2847 IFF_ALLMULTI |
2848 IFF_RUNNING |
2849 IFF_LOWER_UP |
2850 IFF_DORMANT)) |
2851 (dev->gflags & (IFF_PROMISC |
2852 IFF_ALLMULTI));
2853
2854 if (netif_running(dev)) {
2855 if (netif_oper_up(dev))
2856 flags |= IFF_RUNNING;
2857 if (netif_carrier_ok(dev))
2858 flags |= IFF_LOWER_UP;
2859 if (netif_dormant(dev))
2860 flags |= IFF_DORMANT;
2861 }
2862
2863 return flags;
2864 }
2865
2866 int dev_change_flags(struct net_device *dev, unsigned flags)
2867 {
2868 int ret, changes;
2869 int old_flags = dev->flags;
2870
2871 ASSERT_RTNL();
2872
2873 /*
2874 * Set the flags on our device.
2875 */
2876
2877 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2878 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2879 IFF_AUTOMEDIA)) |
2880 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2881 IFF_ALLMULTI));
2882
2883 /*
2884 * Load in the correct multicast list now the flags have changed.
2885 */
2886
2887 if (dev->change_rx_flags && (dev->flags ^ flags) & IFF_MULTICAST)
2888 dev->change_rx_flags(dev, IFF_MULTICAST);
2889
2890 dev_set_rx_mode(dev);
2891
2892 /*
2893 * Have we downed the interface. We handle IFF_UP ourselves
2894 * according to user attempts to set it, rather than blindly
2895 * setting it.
2896 */
2897
2898 ret = 0;
2899 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2900 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2901
2902 if (!ret)
2903 dev_set_rx_mode(dev);
2904 }
2905
2906 if (dev->flags & IFF_UP &&
2907 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2908 IFF_VOLATILE)))
2909 raw_notifier_call_chain(&netdev_chain,
2910 NETDEV_CHANGE, dev);
2911
2912 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2913 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2914 dev->gflags ^= IFF_PROMISC;
2915 dev_set_promiscuity(dev, inc);
2916 }
2917
2918 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2919 is important. Some (broken) drivers set IFF_PROMISC, when
2920 IFF_ALLMULTI is requested not asking us and not reporting.
2921 */
2922 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2923 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2924 dev->gflags ^= IFF_ALLMULTI;
2925 dev_set_allmulti(dev, inc);
2926 }
2927
2928 /* Exclude state transition flags, already notified */
2929 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
2930 if (changes)
2931 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
2932
2933 return ret;
2934 }
2935
2936 int dev_set_mtu(struct net_device *dev, int new_mtu)
2937 {
2938 int err;
2939
2940 if (new_mtu == dev->mtu)
2941 return 0;
2942
2943 /* MTU must be positive. */
2944 if (new_mtu < 0)
2945 return -EINVAL;
2946
2947 if (!netif_device_present(dev))
2948 return -ENODEV;
2949
2950 err = 0;
2951 if (dev->change_mtu)
2952 err = dev->change_mtu(dev, new_mtu);
2953 else
2954 dev->mtu = new_mtu;
2955 if (!err && dev->flags & IFF_UP)
2956 raw_notifier_call_chain(&netdev_chain,
2957 NETDEV_CHANGEMTU, dev);
2958 return err;
2959 }
2960
2961 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2962 {
2963 int err;
2964
2965 if (!dev->set_mac_address)
2966 return -EOPNOTSUPP;
2967 if (sa->sa_family != dev->type)
2968 return -EINVAL;
2969 if (!netif_device_present(dev))
2970 return -ENODEV;
2971 err = dev->set_mac_address(dev, sa);
2972 if (!err)
2973 raw_notifier_call_chain(&netdev_chain,
2974 NETDEV_CHANGEADDR, dev);
2975 return err;
2976 }
2977
2978 /*
2979 * Perform the SIOCxIFxxx calls.
2980 */
2981 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2982 {
2983 int err;
2984 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2985
2986 if (!dev)
2987 return -ENODEV;
2988
2989 switch (cmd) {
2990 case SIOCGIFFLAGS: /* Get interface flags */
2991 ifr->ifr_flags = dev_get_flags(dev);
2992 return 0;
2993
2994 case SIOCSIFFLAGS: /* Set interface flags */
2995 return dev_change_flags(dev, ifr->ifr_flags);
2996
2997 case SIOCGIFMETRIC: /* Get the metric on the interface
2998 (currently unused) */
2999 ifr->ifr_metric = 0;
3000 return 0;
3001
3002 case SIOCSIFMETRIC: /* Set the metric on the interface
3003 (currently unused) */
3004 return -EOPNOTSUPP;
3005
3006 case SIOCGIFMTU: /* Get the MTU of a device */
3007 ifr->ifr_mtu = dev->mtu;
3008 return 0;
3009
3010 case SIOCSIFMTU: /* Set the MTU of a device */
3011 return dev_set_mtu(dev, ifr->ifr_mtu);
3012
3013 case SIOCGIFHWADDR:
3014 if (!dev->addr_len)
3015 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
3016 else
3017 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
3018 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3019 ifr->ifr_hwaddr.sa_family = dev->type;
3020 return 0;
3021
3022 case SIOCSIFHWADDR:
3023 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
3024
3025 case SIOCSIFHWBROADCAST:
3026 if (ifr->ifr_hwaddr.sa_family != dev->type)
3027 return -EINVAL;
3028 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
3029 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3030 raw_notifier_call_chain(&netdev_chain,
3031 NETDEV_CHANGEADDR, dev);
3032 return 0;
3033
3034 case SIOCGIFMAP:
3035 ifr->ifr_map.mem_start = dev->mem_start;
3036 ifr->ifr_map.mem_end = dev->mem_end;
3037 ifr->ifr_map.base_addr = dev->base_addr;
3038 ifr->ifr_map.irq = dev->irq;
3039 ifr->ifr_map.dma = dev->dma;
3040 ifr->ifr_map.port = dev->if_port;
3041 return 0;
3042
3043 case SIOCSIFMAP:
3044 if (dev->set_config) {
3045 if (!netif_device_present(dev))
3046 return -ENODEV;
3047 return dev->set_config(dev, &ifr->ifr_map);
3048 }
3049 return -EOPNOTSUPP;
3050
3051 case SIOCADDMULTI:
3052 if (!dev->set_multicast_list ||
3053 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3054 return -EINVAL;
3055 if (!netif_device_present(dev))
3056 return -ENODEV;
3057 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
3058 dev->addr_len, 1);
3059
3060 case SIOCDELMULTI:
3061 if (!dev->set_multicast_list ||
3062 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3063 return -EINVAL;
3064 if (!netif_device_present(dev))
3065 return -ENODEV;
3066 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
3067 dev->addr_len, 1);
3068
3069 case SIOCGIFINDEX:
3070 ifr->ifr_ifindex = dev->ifindex;
3071 return 0;
3072
3073 case SIOCGIFTXQLEN:
3074 ifr->ifr_qlen = dev->tx_queue_len;
3075 return 0;
3076
3077 case SIOCSIFTXQLEN:
3078 if (ifr->ifr_qlen < 0)
3079 return -EINVAL;
3080 dev->tx_queue_len = ifr->ifr_qlen;
3081 return 0;
3082
3083 case SIOCSIFNAME:
3084 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
3085 return dev_change_name(dev, ifr->ifr_newname);
3086
3087 /*
3088 * Unknown or private ioctl
3089 */
3090
3091 default:
3092 if ((cmd >= SIOCDEVPRIVATE &&
3093 cmd <= SIOCDEVPRIVATE + 15) ||
3094 cmd == SIOCBONDENSLAVE ||
3095 cmd == SIOCBONDRELEASE ||
3096 cmd == SIOCBONDSETHWADDR ||
3097 cmd == SIOCBONDSLAVEINFOQUERY ||
3098 cmd == SIOCBONDINFOQUERY ||
3099 cmd == SIOCBONDCHANGEACTIVE ||
3100 cmd == SIOCGMIIPHY ||
3101 cmd == SIOCGMIIREG ||
3102 cmd == SIOCSMIIREG ||
3103 cmd == SIOCBRADDIF ||
3104 cmd == SIOCBRDELIF ||
3105 cmd == SIOCWANDEV) {
3106 err = -EOPNOTSUPP;
3107 if (dev->do_ioctl) {
3108 if (netif_device_present(dev))
3109 err = dev->do_ioctl(dev, ifr,
3110 cmd);
3111 else
3112 err = -ENODEV;
3113 }
3114 } else
3115 err = -EINVAL;
3116
3117 }
3118 return err;
3119 }
3120
3121 /*
3122 * This function handles all "interface"-type I/O control requests. The actual
3123 * 'doing' part of this is dev_ifsioc above.
3124 */
3125
3126 /**
3127 * dev_ioctl - network device ioctl
3128 * @cmd: command to issue
3129 * @arg: pointer to a struct ifreq in user space
3130 *
3131 * Issue ioctl functions to devices. This is normally called by the
3132 * user space syscall interfaces but can sometimes be useful for
3133 * other purposes. The return value is the return from the syscall if
3134 * positive or a negative errno code on error.
3135 */
3136
3137 int dev_ioctl(unsigned int cmd, void __user *arg)
3138 {
3139 struct ifreq ifr;
3140 int ret;
3141 char *colon;
3142
3143 /* One special case: SIOCGIFCONF takes ifconf argument
3144 and requires shared lock, because it sleeps writing
3145 to user space.
3146 */
3147
3148 if (cmd == SIOCGIFCONF) {
3149 rtnl_lock();
3150 ret = dev_ifconf((char __user *) arg);
3151 rtnl_unlock();
3152 return ret;
3153 }
3154 if (cmd == SIOCGIFNAME)
3155 return dev_ifname((struct ifreq __user *)arg);
3156
3157 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3158 return -EFAULT;
3159
3160 ifr.ifr_name[IFNAMSIZ-1] = 0;
3161
3162 colon = strchr(ifr.ifr_name, ':');
3163 if (colon)
3164 *colon = 0;
3165
3166 /*
3167 * See which interface the caller is talking about.
3168 */
3169
3170 switch (cmd) {
3171 /*
3172 * These ioctl calls:
3173 * - can be done by all.
3174 * - atomic and do not require locking.
3175 * - return a value
3176 */
3177 case SIOCGIFFLAGS:
3178 case SIOCGIFMETRIC:
3179 case SIOCGIFMTU:
3180 case SIOCGIFHWADDR:
3181 case SIOCGIFSLAVE:
3182 case SIOCGIFMAP:
3183 case SIOCGIFINDEX:
3184 case SIOCGIFTXQLEN:
3185 dev_load(ifr.ifr_name);
3186 read_lock(&dev_base_lock);
3187 ret = dev_ifsioc(&ifr, cmd);
3188 read_unlock(&dev_base_lock);
3189 if (!ret) {
3190 if (colon)
3191 *colon = ':';
3192 if (copy_to_user(arg, &ifr,
3193 sizeof(struct ifreq)))
3194 ret = -EFAULT;
3195 }
3196 return ret;
3197
3198 case SIOCETHTOOL:
3199 dev_load(ifr.ifr_name);
3200 rtnl_lock();
3201 ret = dev_ethtool(&ifr);
3202 rtnl_unlock();
3203 if (!ret) {
3204 if (colon)
3205 *colon = ':';
3206 if (copy_to_user(arg, &ifr,
3207 sizeof(struct ifreq)))
3208 ret = -EFAULT;
3209 }
3210 return ret;
3211
3212 /*
3213 * These ioctl calls:
3214 * - require superuser power.
3215 * - require strict serialization.
3216 * - return a value
3217 */
3218 case SIOCGMIIPHY:
3219 case SIOCGMIIREG:
3220 case SIOCSIFNAME:
3221 if (!capable(CAP_NET_ADMIN))
3222 return -EPERM;
3223 dev_load(ifr.ifr_name);
3224 rtnl_lock();
3225 ret = dev_ifsioc(&ifr, cmd);
3226 rtnl_unlock();
3227 if (!ret) {
3228 if (colon)
3229 *colon = ':';
3230 if (copy_to_user(arg, &ifr,
3231 sizeof(struct ifreq)))
3232 ret = -EFAULT;
3233 }
3234 return ret;
3235
3236 /*
3237 * These ioctl calls:
3238 * - require superuser power.
3239 * - require strict serialization.
3240 * - do not return a value
3241 */
3242 case SIOCSIFFLAGS:
3243 case SIOCSIFMETRIC:
3244 case SIOCSIFMTU:
3245 case SIOCSIFMAP:
3246 case SIOCSIFHWADDR:
3247 case SIOCSIFSLAVE:
3248 case SIOCADDMULTI:
3249 case SIOCDELMULTI:
3250 case SIOCSIFHWBROADCAST:
3251 case SIOCSIFTXQLEN:
3252 case SIOCSMIIREG:
3253 case SIOCBONDENSLAVE:
3254 case SIOCBONDRELEASE:
3255 case SIOCBONDSETHWADDR:
3256 case SIOCBONDCHANGEACTIVE:
3257 case SIOCBRADDIF:
3258 case SIOCBRDELIF:
3259 if (!capable(CAP_NET_ADMIN))
3260 return -EPERM;
3261 /* fall through */
3262 case SIOCBONDSLAVEINFOQUERY:
3263 case SIOCBONDINFOQUERY:
3264 dev_load(ifr.ifr_name);
3265 rtnl_lock();
3266 ret = dev_ifsioc(&ifr, cmd);
3267 rtnl_unlock();
3268 return ret;
3269
3270 case SIOCGIFMEM:
3271 /* Get the per device memory space. We can add this but
3272 * currently do not support it */
3273 case SIOCSIFMEM:
3274 /* Set the per device memory buffer space.
3275 * Not applicable in our case */
3276 case SIOCSIFLINK:
3277 return -EINVAL;
3278
3279 /*
3280 * Unknown or private ioctl.
3281 */
3282 default:
3283 if (cmd == SIOCWANDEV ||
3284 (cmd >= SIOCDEVPRIVATE &&
3285 cmd <= SIOCDEVPRIVATE + 15)) {
3286 dev_load(ifr.ifr_name);
3287 rtnl_lock();
3288 ret = dev_ifsioc(&ifr, cmd);
3289 rtnl_unlock();
3290 if (!ret && copy_to_user(arg, &ifr,
3291 sizeof(struct ifreq)))
3292 ret = -EFAULT;
3293 return ret;
3294 }
3295 /* Take care of Wireless Extensions */
3296 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
3297 return wext_handle_ioctl(&ifr, cmd, arg);
3298 return -EINVAL;
3299 }
3300 }
3301
3302
3303 /**
3304 * dev_new_index - allocate an ifindex
3305 *
3306 * Returns a suitable unique value for a new device interface
3307 * number. The caller must hold the rtnl semaphore or the
3308 * dev_base_lock to be sure it remains unique.
3309 */
3310 static int dev_new_index(void)
3311 {
3312 static int ifindex;
3313 for (;;) {
3314 if (++ifindex <= 0)
3315 ifindex = 1;
3316 if (!__dev_get_by_index(ifindex))
3317 return ifindex;
3318 }
3319 }
3320
3321 static int dev_boot_phase = 1;
3322
3323 /* Delayed registration/unregisteration */
3324 static DEFINE_SPINLOCK(net_todo_list_lock);
3325 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
3326
3327 static void net_set_todo(struct net_device *dev)
3328 {
3329 spin_lock(&net_todo_list_lock);
3330 list_add_tail(&dev->todo_list, &net_todo_list);
3331 spin_unlock(&net_todo_list_lock);
3332 }
3333
3334 /**
3335 * register_netdevice - register a network device
3336 * @dev: device to register
3337 *
3338 * Take a completed network device structure and add it to the kernel
3339 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3340 * chain. 0 is returned on success. A negative errno code is returned
3341 * on a failure to set up the device, or if the name is a duplicate.
3342 *
3343 * Callers must hold the rtnl semaphore. You may want
3344 * register_netdev() instead of this.
3345 *
3346 * BUGS:
3347 * The locking appears insufficient to guarantee two parallel registers
3348 * will not get the same name.
3349 */
3350
3351 int register_netdevice(struct net_device *dev)
3352 {
3353 struct hlist_head *head;
3354 struct hlist_node *p;
3355 int ret;
3356
3357 BUG_ON(dev_boot_phase);
3358 ASSERT_RTNL();
3359
3360 might_sleep();
3361
3362 /* When net_device's are persistent, this will be fatal. */
3363 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
3364
3365 spin_lock_init(&dev->queue_lock);
3366 spin_lock_init(&dev->_xmit_lock);
3367 netdev_set_lockdep_class(&dev->_xmit_lock, dev->type);
3368 dev->xmit_lock_owner = -1;
3369 spin_lock_init(&dev->ingress_lock);
3370
3371 dev->iflink = -1;
3372
3373 /* Init, if this function is available */
3374 if (dev->init) {
3375 ret = dev->init(dev);
3376 if (ret) {
3377 if (ret > 0)
3378 ret = -EIO;
3379 goto out;
3380 }
3381 }
3382
3383 if (!dev_valid_name(dev->name)) {
3384 ret = -EINVAL;
3385 goto err_uninit;
3386 }
3387
3388 dev->ifindex = dev_new_index();
3389 if (dev->iflink == -1)
3390 dev->iflink = dev->ifindex;
3391
3392 /* Check for existence of name */
3393 head = dev_name_hash(dev->name);
3394 hlist_for_each(p, head) {
3395 struct net_device *d
3396 = hlist_entry(p, struct net_device, name_hlist);
3397 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
3398 ret = -EEXIST;
3399 goto err_uninit;
3400 }
3401 }
3402
3403 /* Fix illegal checksum combinations */
3404 if ((dev->features & NETIF_F_HW_CSUM) &&
3405 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
3406 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
3407 dev->name);
3408 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
3409 }
3410
3411 if ((dev->features & NETIF_F_NO_CSUM) &&
3412 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
3413 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
3414 dev->name);
3415 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
3416 }
3417
3418
3419 /* Fix illegal SG+CSUM combinations. */
3420 if ((dev->features & NETIF_F_SG) &&
3421 !(dev->features & NETIF_F_ALL_CSUM)) {
3422 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no checksum feature.\n",
3423 dev->name);
3424 dev->features &= ~NETIF_F_SG;
3425 }
3426
3427 /* TSO requires that SG is present as well. */
3428 if ((dev->features & NETIF_F_TSO) &&
3429 !(dev->features & NETIF_F_SG)) {
3430 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no SG feature.\n",
3431 dev->name);
3432 dev->features &= ~NETIF_F_TSO;
3433 }
3434 if (dev->features & NETIF_F_UFO) {
3435 if (!(dev->features & NETIF_F_HW_CSUM)) {
3436 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3437 "NETIF_F_HW_CSUM feature.\n",
3438 dev->name);
3439 dev->features &= ~NETIF_F_UFO;
3440 }
3441 if (!(dev->features & NETIF_F_SG)) {
3442 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3443 "NETIF_F_SG feature.\n",
3444 dev->name);
3445 dev->features &= ~NETIF_F_UFO;
3446 }
3447 }
3448
3449 /*
3450 * nil rebuild_header routine,
3451 * that should be never called and used as just bug trap.
3452 */
3453
3454 if (!dev->rebuild_header)
3455 dev->rebuild_header = default_rebuild_header;
3456
3457 ret = netdev_register_sysfs(dev);
3458 if (ret)
3459 goto err_uninit;
3460 dev->reg_state = NETREG_REGISTERED;
3461
3462 /*
3463 * Default initial state at registry is that the
3464 * device is present.
3465 */
3466
3467 set_bit(__LINK_STATE_PRESENT, &dev->state);
3468
3469 dev_init_scheduler(dev);
3470 write_lock_bh(&dev_base_lock);
3471 list_add_tail(&dev->dev_list, &dev_base_head);
3472 hlist_add_head(&dev->name_hlist, head);
3473 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
3474 dev_hold(dev);
3475 write_unlock_bh(&dev_base_lock);
3476
3477 /* Notify protocols, that a new device appeared. */
3478 ret = raw_notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
3479 ret = notifier_to_errno(ret);
3480 if (ret)
3481 unregister_netdevice(dev);
3482
3483 out:
3484 return ret;
3485
3486 err_uninit:
3487 if (dev->uninit)
3488 dev->uninit(dev);
3489 goto out;
3490 }
3491
3492 /**
3493 * register_netdev - register a network device
3494 * @dev: device to register
3495 *
3496 * Take a completed network device structure and add it to the kernel
3497 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3498 * chain. 0 is returned on success. A negative errno code is returned
3499 * on a failure to set up the device, or if the name is a duplicate.
3500 *
3501 * This is a wrapper around register_netdevice that takes the rtnl semaphore
3502 * and expands the device name if you passed a format string to
3503 * alloc_netdev.
3504 */
3505 int register_netdev(struct net_device *dev)
3506 {
3507 int err;
3508
3509 rtnl_lock();
3510
3511 /*
3512 * If the name is a format string the caller wants us to do a
3513 * name allocation.
3514 */
3515 if (strchr(dev->name, '%')) {
3516 err = dev_alloc_name(dev, dev->name);
3517 if (err < 0)
3518 goto out;
3519 }
3520
3521 err = register_netdevice(dev);
3522 out:
3523 rtnl_unlock();
3524 return err;
3525 }
3526 EXPORT_SYMBOL(register_netdev);
3527
3528 /*
3529 * netdev_wait_allrefs - wait until all references are gone.
3530 *
3531 * This is called when unregistering network devices.
3532 *
3533 * Any protocol or device that holds a reference should register
3534 * for netdevice notification, and cleanup and put back the
3535 * reference if they receive an UNREGISTER event.
3536 * We can get stuck here if buggy protocols don't correctly
3537 * call dev_put.
3538 */
3539 static void netdev_wait_allrefs(struct net_device *dev)
3540 {
3541 unsigned long rebroadcast_time, warning_time;
3542
3543 rebroadcast_time = warning_time = jiffies;
3544 while (atomic_read(&dev->refcnt) != 0) {
3545 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
3546 rtnl_lock();
3547
3548 /* Rebroadcast unregister notification */
3549 raw_notifier_call_chain(&netdev_chain,
3550 NETDEV_UNREGISTER, dev);
3551
3552 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
3553 &dev->state)) {
3554 /* We must not have linkwatch events
3555 * pending on unregister. If this
3556 * happens, we simply run the queue
3557 * unscheduled, resulting in a noop
3558 * for this device.
3559 */
3560 linkwatch_run_queue();
3561 }
3562
3563 __rtnl_unlock();
3564
3565 rebroadcast_time = jiffies;
3566 }
3567
3568 msleep(250);
3569
3570 if (time_after(jiffies, warning_time + 10 * HZ)) {
3571 printk(KERN_EMERG "unregister_netdevice: "
3572 "waiting for %s to become free. Usage "
3573 "count = %d\n",
3574 dev->name, atomic_read(&dev->refcnt));
3575 warning_time = jiffies;
3576 }
3577 }
3578 }
3579
3580 /* The sequence is:
3581 *
3582 * rtnl_lock();
3583 * ...
3584 * register_netdevice(x1);
3585 * register_netdevice(x2);
3586 * ...
3587 * unregister_netdevice(y1);
3588 * unregister_netdevice(y2);
3589 * ...
3590 * rtnl_unlock();
3591 * free_netdev(y1);
3592 * free_netdev(y2);
3593 *
3594 * We are invoked by rtnl_unlock() after it drops the semaphore.
3595 * This allows us to deal with problems:
3596 * 1) We can delete sysfs objects which invoke hotplug
3597 * without deadlocking with linkwatch via keventd.
3598 * 2) Since we run with the RTNL semaphore not held, we can sleep
3599 * safely in order to wait for the netdev refcnt to drop to zero.
3600 */
3601 static DEFINE_MUTEX(net_todo_run_mutex);
3602 void netdev_run_todo(void)
3603 {
3604 struct list_head list;
3605
3606 /* Need to guard against multiple cpu's getting out of order. */
3607 mutex_lock(&net_todo_run_mutex);
3608
3609 /* Not safe to do outside the semaphore. We must not return
3610 * until all unregister events invoked by the local processor
3611 * have been completed (either by this todo run, or one on
3612 * another cpu).
3613 */
3614 if (list_empty(&net_todo_list))
3615 goto out;
3616
3617 /* Snapshot list, allow later requests */
3618 spin_lock(&net_todo_list_lock);
3619 list_replace_init(&net_todo_list, &list);
3620 spin_unlock(&net_todo_list_lock);
3621
3622 while (!list_empty(&list)) {
3623 struct net_device *dev
3624 = list_entry(list.next, struct net_device, todo_list);
3625 list_del(&dev->todo_list);
3626
3627 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
3628 printk(KERN_ERR "network todo '%s' but state %d\n",
3629 dev->name, dev->reg_state);
3630 dump_stack();
3631 continue;
3632 }
3633
3634 dev->reg_state = NETREG_UNREGISTERED;
3635
3636 netdev_wait_allrefs(dev);
3637
3638 /* paranoia */
3639 BUG_ON(atomic_read(&dev->refcnt));
3640 BUG_TRAP(!dev->ip_ptr);
3641 BUG_TRAP(!dev->ip6_ptr);
3642 BUG_TRAP(!dev->dn_ptr);
3643
3644 if (dev->destructor)
3645 dev->destructor(dev);
3646
3647 /* Free network device */
3648 kobject_put(&dev->dev.kobj);
3649 }
3650
3651 out:
3652 mutex_unlock(&net_todo_run_mutex);
3653 }
3654
3655 static struct net_device_stats *internal_stats(struct net_device *dev)
3656 {
3657 return &dev->stats;
3658 }
3659
3660 /**
3661 * alloc_netdev_mq - allocate network device
3662 * @sizeof_priv: size of private data to allocate space for
3663 * @name: device name format string
3664 * @setup: callback to initialize device
3665 * @queue_count: the number of subqueues to allocate
3666 *
3667 * Allocates a struct net_device with private data area for driver use
3668 * and performs basic initialization. Also allocates subquue structs
3669 * for each queue on the device at the end of the netdevice.
3670 */
3671 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
3672 void (*setup)(struct net_device *), unsigned int queue_count)
3673 {
3674 void *p;
3675 struct net_device *dev;
3676 int alloc_size;
3677
3678 BUG_ON(strlen(name) >= sizeof(dev->name));
3679
3680 /* ensure 32-byte alignment of both the device and private area */
3681 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST +
3682 (sizeof(struct net_device_subqueue) * (queue_count - 1))) &
3683 ~NETDEV_ALIGN_CONST;
3684 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3685
3686 p = kzalloc(alloc_size, GFP_KERNEL);
3687 if (!p) {
3688 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
3689 return NULL;
3690 }
3691
3692 dev = (struct net_device *)
3693 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3694 dev->padded = (char *)dev - (char *)p;
3695
3696 if (sizeof_priv) {
3697 dev->priv = ((char *)dev +
3698 ((sizeof(struct net_device) +
3699 (sizeof(struct net_device_subqueue) *
3700 (queue_count - 1)) + NETDEV_ALIGN_CONST)
3701 & ~NETDEV_ALIGN_CONST));
3702 }
3703
3704 dev->egress_subqueue_count = queue_count;
3705
3706 dev->get_stats = internal_stats;
3707 setup(dev);
3708 strcpy(dev->name, name);
3709 return dev;
3710 }
3711 EXPORT_SYMBOL(alloc_netdev_mq);
3712
3713 /**
3714 * free_netdev - free network device
3715 * @dev: device
3716 *
3717 * This function does the last stage of destroying an allocated device
3718 * interface. The reference to the device object is released.
3719 * If this is the last reference then it will be freed.
3720 */
3721 void free_netdev(struct net_device *dev)
3722 {
3723 #ifdef CONFIG_SYSFS
3724 /* Compatibility with error handling in drivers */
3725 if (dev->reg_state == NETREG_UNINITIALIZED) {
3726 kfree((char *)dev - dev->padded);
3727 return;
3728 }
3729
3730 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3731 dev->reg_state = NETREG_RELEASED;
3732
3733 /* will free via device release */
3734 put_device(&dev->dev);
3735 #else
3736 kfree((char *)dev - dev->padded);
3737 #endif
3738 }
3739
3740 /* Synchronize with packet receive processing. */
3741 void synchronize_net(void)
3742 {
3743 might_sleep();
3744 synchronize_rcu();
3745 }
3746
3747 /**
3748 * unregister_netdevice - remove device from the kernel
3749 * @dev: device
3750 *
3751 * This function shuts down a device interface and removes it
3752 * from the kernel tables. On success 0 is returned, on a failure
3753 * a negative errno code is returned.
3754 *
3755 * Callers must hold the rtnl semaphore. You may want
3756 * unregister_netdev() instead of this.
3757 */
3758
3759 void unregister_netdevice(struct net_device *dev)
3760 {
3761 BUG_ON(dev_boot_phase);
3762 ASSERT_RTNL();
3763
3764 /* Some devices call without registering for initialization unwind. */
3765 if (dev->reg_state == NETREG_UNINITIALIZED) {
3766 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3767 "was registered\n", dev->name, dev);
3768
3769 WARN_ON(1);
3770 return;
3771 }
3772
3773 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3774
3775 /* If device is running, close it first. */
3776 if (dev->flags & IFF_UP)
3777 dev_close(dev);
3778
3779 /* And unlink it from device chain. */
3780 write_lock_bh(&dev_base_lock);
3781 list_del(&dev->dev_list);
3782 hlist_del(&dev->name_hlist);
3783 hlist_del(&dev->index_hlist);
3784 write_unlock_bh(&dev_base_lock);
3785
3786 dev->reg_state = NETREG_UNREGISTERING;
3787
3788 synchronize_net();
3789
3790 /* Shutdown queueing discipline. */
3791 dev_shutdown(dev);
3792
3793
3794 /* Notify protocols, that we are about to destroy
3795 this device. They should clean all the things.
3796 */
3797 raw_notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3798
3799 /*
3800 * Flush the unicast and multicast chains
3801 */
3802 dev_addr_discard(dev);
3803
3804 if (dev->uninit)
3805 dev->uninit(dev);
3806
3807 /* Notifier chain MUST detach us from master device. */
3808 BUG_TRAP(!dev->master);
3809
3810 /* Remove entries from sysfs */
3811 netdev_unregister_sysfs(dev);
3812
3813 /* Finish processing unregister after unlock */
3814 net_set_todo(dev);
3815
3816 synchronize_net();
3817
3818 dev_put(dev);
3819 }
3820
3821 /**
3822 * unregister_netdev - remove device from the kernel
3823 * @dev: device
3824 *
3825 * This function shuts down a device interface and removes it
3826 * from the kernel tables. On success 0 is returned, on a failure
3827 * a negative errno code is returned.
3828 *
3829 * This is just a wrapper for unregister_netdevice that takes
3830 * the rtnl semaphore. In general you want to use this and not
3831 * unregister_netdevice.
3832 */
3833 void unregister_netdev(struct net_device *dev)
3834 {
3835 rtnl_lock();
3836 unregister_netdevice(dev);
3837 rtnl_unlock();
3838 }
3839
3840 EXPORT_SYMBOL(unregister_netdev);
3841
3842 static int dev_cpu_callback(struct notifier_block *nfb,
3843 unsigned long action,
3844 void *ocpu)
3845 {
3846 struct sk_buff **list_skb;
3847 struct net_device **list_net;
3848 struct sk_buff *skb;
3849 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3850 struct softnet_data *sd, *oldsd;
3851
3852 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
3853 return NOTIFY_OK;
3854
3855 local_irq_disable();
3856 cpu = smp_processor_id();
3857 sd = &per_cpu(softnet_data, cpu);
3858 oldsd = &per_cpu(softnet_data, oldcpu);
3859
3860 /* Find end of our completion_queue. */
3861 list_skb = &sd->completion_queue;
3862 while (*list_skb)
3863 list_skb = &(*list_skb)->next;
3864 /* Append completion queue from offline CPU. */
3865 *list_skb = oldsd->completion_queue;
3866 oldsd->completion_queue = NULL;
3867
3868 /* Find end of our output_queue. */
3869 list_net = &sd->output_queue;
3870 while (*list_net)
3871 list_net = &(*list_net)->next_sched;
3872 /* Append output queue from offline CPU. */
3873 *list_net = oldsd->output_queue;
3874 oldsd->output_queue = NULL;
3875
3876 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3877 local_irq_enable();
3878
3879 /* Process offline CPU's input_pkt_queue */
3880 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3881 netif_rx(skb);
3882
3883 return NOTIFY_OK;
3884 }
3885
3886 #ifdef CONFIG_NET_DMA
3887 /**
3888 * net_dma_rebalance - try to maintain one DMA channel per CPU
3889 * @net_dma: DMA client and associated data (lock, channels, channel_mask)
3890 *
3891 * This is called when the number of channels allocated to the net_dma client
3892 * changes. The net_dma client tries to have one DMA channel per CPU.
3893 */
3894
3895 static void net_dma_rebalance(struct net_dma *net_dma)
3896 {
3897 unsigned int cpu, i, n, chan_idx;
3898 struct dma_chan *chan;
3899
3900 if (cpus_empty(net_dma->channel_mask)) {
3901 for_each_online_cpu(cpu)
3902 rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL);
3903 return;
3904 }
3905
3906 i = 0;
3907 cpu = first_cpu(cpu_online_map);
3908
3909 for_each_cpu_mask(chan_idx, net_dma->channel_mask) {
3910 chan = net_dma->channels[chan_idx];
3911
3912 n = ((num_online_cpus() / cpus_weight(net_dma->channel_mask))
3913 + (i < (num_online_cpus() %
3914 cpus_weight(net_dma->channel_mask)) ? 1 : 0));
3915
3916 while(n) {
3917 per_cpu(softnet_data, cpu).net_dma = chan;
3918 cpu = next_cpu(cpu, cpu_online_map);
3919 n--;
3920 }
3921 i++;
3922 }
3923 }
3924
3925 /**
3926 * netdev_dma_event - event callback for the net_dma_client
3927 * @client: should always be net_dma_client
3928 * @chan: DMA channel for the event
3929 * @state: DMA state to be handled
3930 */
3931 static enum dma_state_client
3932 netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
3933 enum dma_state state)
3934 {
3935 int i, found = 0, pos = -1;
3936 struct net_dma *net_dma =
3937 container_of(client, struct net_dma, client);
3938 enum dma_state_client ack = DMA_DUP; /* default: take no action */
3939
3940 spin_lock(&net_dma->lock);
3941 switch (state) {
3942 case DMA_RESOURCE_AVAILABLE:
3943 for (i = 0; i < NR_CPUS; i++)
3944 if (net_dma->channels[i] == chan) {
3945 found = 1;
3946 break;
3947 } else if (net_dma->channels[i] == NULL && pos < 0)
3948 pos = i;
3949
3950 if (!found && pos >= 0) {
3951 ack = DMA_ACK;
3952 net_dma->channels[pos] = chan;
3953 cpu_set(pos, net_dma->channel_mask);
3954 net_dma_rebalance(net_dma);
3955 }
3956 break;
3957 case DMA_RESOURCE_REMOVED:
3958 for (i = 0; i < NR_CPUS; i++)
3959 if (net_dma->channels[i] == chan) {
3960 found = 1;
3961 pos = i;
3962 break;
3963 }
3964
3965 if (found) {
3966 ack = DMA_ACK;
3967 cpu_clear(pos, net_dma->channel_mask);
3968 net_dma->channels[i] = NULL;
3969 net_dma_rebalance(net_dma);
3970 }
3971 break;
3972 default:
3973 break;
3974 }
3975 spin_unlock(&net_dma->lock);
3976
3977 return ack;
3978 }
3979
3980 /**
3981 * netdev_dma_regiser - register the networking subsystem as a DMA client
3982 */
3983 static int __init netdev_dma_register(void)
3984 {
3985 spin_lock_init(&net_dma.lock);
3986 dma_cap_set(DMA_MEMCPY, net_dma.client.cap_mask);
3987 dma_async_client_register(&net_dma.client);
3988 dma_async_client_chan_request(&net_dma.client);
3989 return 0;
3990 }
3991
3992 #else
3993 static int __init netdev_dma_register(void) { return -ENODEV; }
3994 #endif /* CONFIG_NET_DMA */
3995
3996 /**
3997 * netdev_compute_feature - compute conjunction of two feature sets
3998 * @all: first feature set
3999 * @one: second feature set
4000 *
4001 * Computes a new feature set after adding a device with feature set
4002 * @one to the master device with current feature set @all. Returns
4003 * the new feature set.
4004 */
4005 int netdev_compute_features(unsigned long all, unsigned long one)
4006 {
4007 /* if device needs checksumming, downgrade to hw checksumming */
4008 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
4009 all ^= NETIF_F_NO_CSUM | NETIF_F_HW_CSUM;
4010
4011 /* if device can't do all checksum, downgrade to ipv4/ipv6 */
4012 if (all & NETIF_F_HW_CSUM && !(one & NETIF_F_HW_CSUM))
4013 all ^= NETIF_F_HW_CSUM
4014 | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
4015
4016 if (one & NETIF_F_GSO)
4017 one |= NETIF_F_GSO_SOFTWARE;
4018 one |= NETIF_F_GSO;
4019
4020 /* If even one device supports robust GSO, enable it for all. */
4021 if (one & NETIF_F_GSO_ROBUST)
4022 all |= NETIF_F_GSO_ROBUST;
4023
4024 all &= one | NETIF_F_LLTX;
4025
4026 if (!(all & NETIF_F_ALL_CSUM))
4027 all &= ~NETIF_F_SG;
4028 if (!(all & NETIF_F_SG))
4029 all &= ~NETIF_F_GSO_MASK;
4030
4031 return all;
4032 }
4033 EXPORT_SYMBOL(netdev_compute_features);
4034
4035 /*
4036 * Initialize the DEV module. At boot time this walks the device list and
4037 * unhooks any devices that fail to initialise (normally hardware not
4038 * present) and leaves us with a valid list of present and active devices.
4039 *
4040 */
4041
4042 /*
4043 * This is called single threaded during boot, so no need
4044 * to take the rtnl semaphore.
4045 */
4046 static int __init net_dev_init(void)
4047 {
4048 int i, rc = -ENOMEM;
4049
4050 BUG_ON(!dev_boot_phase);
4051
4052 if (dev_proc_init())
4053 goto out;
4054
4055 if (netdev_sysfs_init())
4056 goto out;
4057
4058 INIT_LIST_HEAD(&ptype_all);
4059 for (i = 0; i < 16; i++)
4060 INIT_LIST_HEAD(&ptype_base[i]);
4061
4062 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
4063 INIT_HLIST_HEAD(&dev_name_head[i]);
4064
4065 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
4066 INIT_HLIST_HEAD(&dev_index_head[i]);
4067
4068 /*
4069 * Initialise the packet receive queues.
4070 */
4071
4072 for_each_possible_cpu(i) {
4073 struct softnet_data *queue;
4074
4075 queue = &per_cpu(softnet_data, i);
4076 skb_queue_head_init(&queue->input_pkt_queue);
4077 queue->completion_queue = NULL;
4078 INIT_LIST_HEAD(&queue->poll_list);
4079 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
4080 queue->backlog_dev.weight = weight_p;
4081 queue->backlog_dev.poll = process_backlog;
4082 atomic_set(&queue->backlog_dev.refcnt, 1);
4083 }
4084
4085 netdev_dma_register();
4086
4087 dev_boot_phase = 0;
4088
4089 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
4090 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
4091
4092 hotcpu_notifier(dev_cpu_callback, 0);
4093 dst_init();
4094 dev_mcast_init();
4095 rc = 0;
4096 out:
4097 return rc;
4098 }
4099
4100 subsys_initcall(net_dev_init);
4101
4102 EXPORT_SYMBOL(__dev_get_by_index);
4103 EXPORT_SYMBOL(__dev_get_by_name);
4104 EXPORT_SYMBOL(__dev_remove_pack);
4105 EXPORT_SYMBOL(dev_valid_name);
4106 EXPORT_SYMBOL(dev_add_pack);
4107 EXPORT_SYMBOL(dev_alloc_name);
4108 EXPORT_SYMBOL(dev_close);
4109 EXPORT_SYMBOL(dev_get_by_flags);
4110 EXPORT_SYMBOL(dev_get_by_index);
4111 EXPORT_SYMBOL(dev_get_by_name);
4112 EXPORT_SYMBOL(dev_open);
4113 EXPORT_SYMBOL(dev_queue_xmit);
4114 EXPORT_SYMBOL(dev_remove_pack);
4115 EXPORT_SYMBOL(dev_set_allmulti);
4116 EXPORT_SYMBOL(dev_set_promiscuity);
4117 EXPORT_SYMBOL(dev_change_flags);
4118 EXPORT_SYMBOL(dev_set_mtu);
4119 EXPORT_SYMBOL(dev_set_mac_address);
4120 EXPORT_SYMBOL(free_netdev);
4121 EXPORT_SYMBOL(netdev_boot_setup_check);
4122 EXPORT_SYMBOL(netdev_set_master);
4123 EXPORT_SYMBOL(netdev_state_change);
4124 EXPORT_SYMBOL(netif_receive_skb);
4125 EXPORT_SYMBOL(netif_rx);
4126 EXPORT_SYMBOL(register_gifconf);
4127 EXPORT_SYMBOL(register_netdevice);
4128 EXPORT_SYMBOL(register_netdevice_notifier);
4129 EXPORT_SYMBOL(skb_checksum_help);
4130 EXPORT_SYMBOL(synchronize_net);
4131 EXPORT_SYMBOL(unregister_netdevice);
4132 EXPORT_SYMBOL(unregister_netdevice_notifier);
4133 EXPORT_SYMBOL(net_enable_timestamp);
4134 EXPORT_SYMBOL(net_disable_timestamp);
4135 EXPORT_SYMBOL(dev_get_flags);
4136
4137 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
4138 EXPORT_SYMBOL(br_handle_frame_hook);
4139 EXPORT_SYMBOL(br_fdb_get_hook);
4140 EXPORT_SYMBOL(br_fdb_put_hook);
4141 #endif
4142
4143 #ifdef CONFIG_KMOD
4144 EXPORT_SYMBOL(dev_load);
4145 #endif
4146
4147 EXPORT_PER_CPU_SYMBOL(softnet_data);