CRED: Wrap task credential accesses in the networking subsystem
[deliverable/linux.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/ethtool.h>
94 #include <linux/notifier.h>
95 #include <linux/skbuff.h>
96 #include <net/net_namespace.h>
97 #include <net/sock.h>
98 #include <linux/rtnetlink.h>
99 #include <linux/proc_fs.h>
100 #include <linux/seq_file.h>
101 #include <linux/stat.h>
102 #include <linux/if_bridge.h>
103 #include <linux/if_macvlan.h>
104 #include <net/dst.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <linux/highmem.h>
108 #include <linux/init.h>
109 #include <linux/kmod.h>
110 #include <linux/module.h>
111 #include <linux/kallsyms.h>
112 #include <linux/netpoll.h>
113 #include <linux/rcupdate.h>
114 #include <linux/delay.h>
115 #include <net/wext.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
125 #include <net/ip.h>
126 #include <linux/ipv6.h>
127 #include <linux/in.h>
128 #include <linux/jhash.h>
129 #include <linux/random.h>
130
131 #include "net-sysfs.h"
132
133 /*
134 * The list of packet types we will receive (as opposed to discard)
135 * and the routines to invoke.
136 *
137 * Why 16. Because with 16 the only overlap we get on a hash of the
138 * low nibble of the protocol value is RARP/SNAP/X.25.
139 *
140 * NOTE: That is no longer true with the addition of VLAN tags. Not
141 * sure which should go first, but I bet it won't make much
142 * difference if we are running VLANs. The good news is that
143 * this protocol won't be in the list unless compiled in, so
144 * the average user (w/out VLANs) will not be adversely affected.
145 * --BLG
146 *
147 * 0800 IP
148 * 8100 802.1Q VLAN
149 * 0001 802.3
150 * 0002 AX.25
151 * 0004 802.2
152 * 8035 RARP
153 * 0005 SNAP
154 * 0805 X.25
155 * 0806 ARP
156 * 8137 IPX
157 * 0009 Localtalk
158 * 86DD IPv6
159 */
160
161 #define PTYPE_HASH_SIZE (16)
162 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
163
164 static DEFINE_SPINLOCK(ptype_lock);
165 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
166 static struct list_head ptype_all __read_mostly; /* Taps */
167
168 #ifdef CONFIG_NET_DMA
169 struct net_dma {
170 struct dma_client client;
171 spinlock_t lock;
172 cpumask_t channel_mask;
173 struct dma_chan **channels;
174 };
175
176 static enum dma_state_client
177 netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
178 enum dma_state state);
179
180 static struct net_dma net_dma = {
181 .client = {
182 .event_callback = netdev_dma_event,
183 },
184 };
185 #endif
186
187 /*
188 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
189 * semaphore.
190 *
191 * Pure readers hold dev_base_lock for reading.
192 *
193 * Writers must hold the rtnl semaphore while they loop through the
194 * dev_base_head list, and hold dev_base_lock for writing when they do the
195 * actual updates. This allows pure readers to access the list even
196 * while a writer is preparing to update it.
197 *
198 * To put it another way, dev_base_lock is held for writing only to
199 * protect against pure readers; the rtnl semaphore provides the
200 * protection against other writers.
201 *
202 * See, for example usages, register_netdevice() and
203 * unregister_netdevice(), which must be called with the rtnl
204 * semaphore held.
205 */
206 DEFINE_RWLOCK(dev_base_lock);
207
208 EXPORT_SYMBOL(dev_base_lock);
209
210 #define NETDEV_HASHBITS 8
211 #define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS)
212
213 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
214 {
215 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
216 return &net->dev_name_head[hash & ((1 << NETDEV_HASHBITS) - 1)];
217 }
218
219 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
220 {
221 return &net->dev_index_head[ifindex & ((1 << NETDEV_HASHBITS) - 1)];
222 }
223
224 /* Device list insertion */
225 static int list_netdevice(struct net_device *dev)
226 {
227 struct net *net = dev_net(dev);
228
229 ASSERT_RTNL();
230
231 write_lock_bh(&dev_base_lock);
232 list_add_tail(&dev->dev_list, &net->dev_base_head);
233 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
234 hlist_add_head(&dev->index_hlist, dev_index_hash(net, dev->ifindex));
235 write_unlock_bh(&dev_base_lock);
236 return 0;
237 }
238
239 /* Device list removal */
240 static void unlist_netdevice(struct net_device *dev)
241 {
242 ASSERT_RTNL();
243
244 /* Unlink dev from the device chain */
245 write_lock_bh(&dev_base_lock);
246 list_del(&dev->dev_list);
247 hlist_del(&dev->name_hlist);
248 hlist_del(&dev->index_hlist);
249 write_unlock_bh(&dev_base_lock);
250 }
251
252 /*
253 * Our notifier list
254 */
255
256 static RAW_NOTIFIER_HEAD(netdev_chain);
257
258 /*
259 * Device drivers call our routines to queue packets here. We empty the
260 * queue in the local softnet handler.
261 */
262
263 DEFINE_PER_CPU(struct softnet_data, softnet_data);
264
265 #ifdef CONFIG_LOCKDEP
266 /*
267 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
268 * according to dev->type
269 */
270 static const unsigned short netdev_lock_type[] =
271 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
272 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
273 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
274 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
275 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
276 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
277 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
278 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
279 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
280 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
281 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
282 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
283 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
284 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_VOID,
285 ARPHRD_NONE};
286
287 static const char *netdev_lock_name[] =
288 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
289 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
290 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
291 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
292 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
293 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
294 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
295 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
296 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
297 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
298 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
299 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
300 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
301 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_VOID",
302 "_xmit_NONE"};
303
304 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
305 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
306
307 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
308 {
309 int i;
310
311 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
312 if (netdev_lock_type[i] == dev_type)
313 return i;
314 /* the last key is used by default */
315 return ARRAY_SIZE(netdev_lock_type) - 1;
316 }
317
318 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
319 unsigned short dev_type)
320 {
321 int i;
322
323 i = netdev_lock_pos(dev_type);
324 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
325 netdev_lock_name[i]);
326 }
327
328 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
329 {
330 int i;
331
332 i = netdev_lock_pos(dev->type);
333 lockdep_set_class_and_name(&dev->addr_list_lock,
334 &netdev_addr_lock_key[i],
335 netdev_lock_name[i]);
336 }
337 #else
338 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
339 unsigned short dev_type)
340 {
341 }
342 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
343 {
344 }
345 #endif
346
347 /*******************************************************************************
348
349 Protocol management and registration routines
350
351 *******************************************************************************/
352
353 /*
354 * Add a protocol ID to the list. Now that the input handler is
355 * smarter we can dispense with all the messy stuff that used to be
356 * here.
357 *
358 * BEWARE!!! Protocol handlers, mangling input packets,
359 * MUST BE last in hash buckets and checking protocol handlers
360 * MUST start from promiscuous ptype_all chain in net_bh.
361 * It is true now, do not change it.
362 * Explanation follows: if protocol handler, mangling packet, will
363 * be the first on list, it is not able to sense, that packet
364 * is cloned and should be copied-on-write, so that it will
365 * change it and subsequent readers will get broken packet.
366 * --ANK (980803)
367 */
368
369 /**
370 * dev_add_pack - add packet handler
371 * @pt: packet type declaration
372 *
373 * Add a protocol handler to the networking stack. The passed &packet_type
374 * is linked into kernel lists and may not be freed until it has been
375 * removed from the kernel lists.
376 *
377 * This call does not sleep therefore it can not
378 * guarantee all CPU's that are in middle of receiving packets
379 * will see the new packet type (until the next received packet).
380 */
381
382 void dev_add_pack(struct packet_type *pt)
383 {
384 int hash;
385
386 spin_lock_bh(&ptype_lock);
387 if (pt->type == htons(ETH_P_ALL))
388 list_add_rcu(&pt->list, &ptype_all);
389 else {
390 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
391 list_add_rcu(&pt->list, &ptype_base[hash]);
392 }
393 spin_unlock_bh(&ptype_lock);
394 }
395
396 /**
397 * __dev_remove_pack - remove packet handler
398 * @pt: packet type declaration
399 *
400 * Remove a protocol handler that was previously added to the kernel
401 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
402 * from the kernel lists and can be freed or reused once this function
403 * returns.
404 *
405 * The packet type might still be in use by receivers
406 * and must not be freed until after all the CPU's have gone
407 * through a quiescent state.
408 */
409 void __dev_remove_pack(struct packet_type *pt)
410 {
411 struct list_head *head;
412 struct packet_type *pt1;
413
414 spin_lock_bh(&ptype_lock);
415
416 if (pt->type == htons(ETH_P_ALL))
417 head = &ptype_all;
418 else
419 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
420
421 list_for_each_entry(pt1, head, list) {
422 if (pt == pt1) {
423 list_del_rcu(&pt->list);
424 goto out;
425 }
426 }
427
428 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
429 out:
430 spin_unlock_bh(&ptype_lock);
431 }
432 /**
433 * dev_remove_pack - remove packet handler
434 * @pt: packet type declaration
435 *
436 * Remove a protocol handler that was previously added to the kernel
437 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
438 * from the kernel lists and can be freed or reused once this function
439 * returns.
440 *
441 * This call sleeps to guarantee that no CPU is looking at the packet
442 * type after return.
443 */
444 void dev_remove_pack(struct packet_type *pt)
445 {
446 __dev_remove_pack(pt);
447
448 synchronize_net();
449 }
450
451 /******************************************************************************
452
453 Device Boot-time Settings Routines
454
455 *******************************************************************************/
456
457 /* Boot time configuration table */
458 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
459
460 /**
461 * netdev_boot_setup_add - add new setup entry
462 * @name: name of the device
463 * @map: configured settings for the device
464 *
465 * Adds new setup entry to the dev_boot_setup list. The function
466 * returns 0 on error and 1 on success. This is a generic routine to
467 * all netdevices.
468 */
469 static int netdev_boot_setup_add(char *name, struct ifmap *map)
470 {
471 struct netdev_boot_setup *s;
472 int i;
473
474 s = dev_boot_setup;
475 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
476 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
477 memset(s[i].name, 0, sizeof(s[i].name));
478 strlcpy(s[i].name, name, IFNAMSIZ);
479 memcpy(&s[i].map, map, sizeof(s[i].map));
480 break;
481 }
482 }
483
484 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
485 }
486
487 /**
488 * netdev_boot_setup_check - check boot time settings
489 * @dev: the netdevice
490 *
491 * Check boot time settings for the device.
492 * The found settings are set for the device to be used
493 * later in the device probing.
494 * Returns 0 if no settings found, 1 if they are.
495 */
496 int netdev_boot_setup_check(struct net_device *dev)
497 {
498 struct netdev_boot_setup *s = dev_boot_setup;
499 int i;
500
501 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
502 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
503 !strcmp(dev->name, s[i].name)) {
504 dev->irq = s[i].map.irq;
505 dev->base_addr = s[i].map.base_addr;
506 dev->mem_start = s[i].map.mem_start;
507 dev->mem_end = s[i].map.mem_end;
508 return 1;
509 }
510 }
511 return 0;
512 }
513
514
515 /**
516 * netdev_boot_base - get address from boot time settings
517 * @prefix: prefix for network device
518 * @unit: id for network device
519 *
520 * Check boot time settings for the base address of device.
521 * The found settings are set for the device to be used
522 * later in the device probing.
523 * Returns 0 if no settings found.
524 */
525 unsigned long netdev_boot_base(const char *prefix, int unit)
526 {
527 const struct netdev_boot_setup *s = dev_boot_setup;
528 char name[IFNAMSIZ];
529 int i;
530
531 sprintf(name, "%s%d", prefix, unit);
532
533 /*
534 * If device already registered then return base of 1
535 * to indicate not to probe for this interface
536 */
537 if (__dev_get_by_name(&init_net, name))
538 return 1;
539
540 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
541 if (!strcmp(name, s[i].name))
542 return s[i].map.base_addr;
543 return 0;
544 }
545
546 /*
547 * Saves at boot time configured settings for any netdevice.
548 */
549 int __init netdev_boot_setup(char *str)
550 {
551 int ints[5];
552 struct ifmap map;
553
554 str = get_options(str, ARRAY_SIZE(ints), ints);
555 if (!str || !*str)
556 return 0;
557
558 /* Save settings */
559 memset(&map, 0, sizeof(map));
560 if (ints[0] > 0)
561 map.irq = ints[1];
562 if (ints[0] > 1)
563 map.base_addr = ints[2];
564 if (ints[0] > 2)
565 map.mem_start = ints[3];
566 if (ints[0] > 3)
567 map.mem_end = ints[4];
568
569 /* Add new entry to the list */
570 return netdev_boot_setup_add(str, &map);
571 }
572
573 __setup("netdev=", netdev_boot_setup);
574
575 /*******************************************************************************
576
577 Device Interface Subroutines
578
579 *******************************************************************************/
580
581 /**
582 * __dev_get_by_name - find a device by its name
583 * @net: the applicable net namespace
584 * @name: name to find
585 *
586 * Find an interface by name. Must be called under RTNL semaphore
587 * or @dev_base_lock. If the name is found a pointer to the device
588 * is returned. If the name is not found then %NULL is returned. The
589 * reference counters are not incremented so the caller must be
590 * careful with locks.
591 */
592
593 struct net_device *__dev_get_by_name(struct net *net, const char *name)
594 {
595 struct hlist_node *p;
596
597 hlist_for_each(p, dev_name_hash(net, name)) {
598 struct net_device *dev
599 = hlist_entry(p, struct net_device, name_hlist);
600 if (!strncmp(dev->name, name, IFNAMSIZ))
601 return dev;
602 }
603 return NULL;
604 }
605
606 /**
607 * dev_get_by_name - find a device by its name
608 * @net: the applicable net namespace
609 * @name: name to find
610 *
611 * Find an interface by name. This can be called from any
612 * context and does its own locking. The returned handle has
613 * the usage count incremented and the caller must use dev_put() to
614 * release it when it is no longer needed. %NULL is returned if no
615 * matching device is found.
616 */
617
618 struct net_device *dev_get_by_name(struct net *net, const char *name)
619 {
620 struct net_device *dev;
621
622 read_lock(&dev_base_lock);
623 dev = __dev_get_by_name(net, name);
624 if (dev)
625 dev_hold(dev);
626 read_unlock(&dev_base_lock);
627 return dev;
628 }
629
630 /**
631 * __dev_get_by_index - find a device by its ifindex
632 * @net: the applicable net namespace
633 * @ifindex: index of device
634 *
635 * Search for an interface by index. Returns %NULL if the device
636 * is not found or a pointer to the device. The device has not
637 * had its reference counter increased so the caller must be careful
638 * about locking. The caller must hold either the RTNL semaphore
639 * or @dev_base_lock.
640 */
641
642 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
643 {
644 struct hlist_node *p;
645
646 hlist_for_each(p, dev_index_hash(net, ifindex)) {
647 struct net_device *dev
648 = hlist_entry(p, struct net_device, index_hlist);
649 if (dev->ifindex == ifindex)
650 return dev;
651 }
652 return NULL;
653 }
654
655
656 /**
657 * dev_get_by_index - find a device by its ifindex
658 * @net: the applicable net namespace
659 * @ifindex: index of device
660 *
661 * Search for an interface by index. Returns NULL if the device
662 * is not found or a pointer to the device. The device returned has
663 * had a reference added and the pointer is safe until the user calls
664 * dev_put to indicate they have finished with it.
665 */
666
667 struct net_device *dev_get_by_index(struct net *net, int ifindex)
668 {
669 struct net_device *dev;
670
671 read_lock(&dev_base_lock);
672 dev = __dev_get_by_index(net, ifindex);
673 if (dev)
674 dev_hold(dev);
675 read_unlock(&dev_base_lock);
676 return dev;
677 }
678
679 /**
680 * dev_getbyhwaddr - find a device by its hardware address
681 * @net: the applicable net namespace
682 * @type: media type of device
683 * @ha: hardware address
684 *
685 * Search for an interface by MAC address. Returns NULL if the device
686 * is not found or a pointer to the device. The caller must hold the
687 * rtnl semaphore. The returned device has not had its ref count increased
688 * and the caller must therefore be careful about locking
689 *
690 * BUGS:
691 * If the API was consistent this would be __dev_get_by_hwaddr
692 */
693
694 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
695 {
696 struct net_device *dev;
697
698 ASSERT_RTNL();
699
700 for_each_netdev(net, dev)
701 if (dev->type == type &&
702 !memcmp(dev->dev_addr, ha, dev->addr_len))
703 return dev;
704
705 return NULL;
706 }
707
708 EXPORT_SYMBOL(dev_getbyhwaddr);
709
710 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
711 {
712 struct net_device *dev;
713
714 ASSERT_RTNL();
715 for_each_netdev(net, dev)
716 if (dev->type == type)
717 return dev;
718
719 return NULL;
720 }
721
722 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
723
724 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
725 {
726 struct net_device *dev;
727
728 rtnl_lock();
729 dev = __dev_getfirstbyhwtype(net, type);
730 if (dev)
731 dev_hold(dev);
732 rtnl_unlock();
733 return dev;
734 }
735
736 EXPORT_SYMBOL(dev_getfirstbyhwtype);
737
738 /**
739 * dev_get_by_flags - find any device with given flags
740 * @net: the applicable net namespace
741 * @if_flags: IFF_* values
742 * @mask: bitmask of bits in if_flags to check
743 *
744 * Search for any interface with the given flags. Returns NULL if a device
745 * is not found or a pointer to the device. The device returned has
746 * had a reference added and the pointer is safe until the user calls
747 * dev_put to indicate they have finished with it.
748 */
749
750 struct net_device * dev_get_by_flags(struct net *net, unsigned short if_flags, unsigned short mask)
751 {
752 struct net_device *dev, *ret;
753
754 ret = NULL;
755 read_lock(&dev_base_lock);
756 for_each_netdev(net, dev) {
757 if (((dev->flags ^ if_flags) & mask) == 0) {
758 dev_hold(dev);
759 ret = dev;
760 break;
761 }
762 }
763 read_unlock(&dev_base_lock);
764 return ret;
765 }
766
767 /**
768 * dev_valid_name - check if name is okay for network device
769 * @name: name string
770 *
771 * Network device names need to be valid file names to
772 * to allow sysfs to work. We also disallow any kind of
773 * whitespace.
774 */
775 int dev_valid_name(const char *name)
776 {
777 if (*name == '\0')
778 return 0;
779 if (strlen(name) >= IFNAMSIZ)
780 return 0;
781 if (!strcmp(name, ".") || !strcmp(name, ".."))
782 return 0;
783
784 while (*name) {
785 if (*name == '/' || isspace(*name))
786 return 0;
787 name++;
788 }
789 return 1;
790 }
791
792 /**
793 * __dev_alloc_name - allocate a name for a device
794 * @net: network namespace to allocate the device name in
795 * @name: name format string
796 * @buf: scratch buffer and result name string
797 *
798 * Passed a format string - eg "lt%d" it will try and find a suitable
799 * id. It scans list of devices to build up a free map, then chooses
800 * the first empty slot. The caller must hold the dev_base or rtnl lock
801 * while allocating the name and adding the device in order to avoid
802 * duplicates.
803 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
804 * Returns the number of the unit assigned or a negative errno code.
805 */
806
807 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
808 {
809 int i = 0;
810 const char *p;
811 const int max_netdevices = 8*PAGE_SIZE;
812 unsigned long *inuse;
813 struct net_device *d;
814
815 p = strnchr(name, IFNAMSIZ-1, '%');
816 if (p) {
817 /*
818 * Verify the string as this thing may have come from
819 * the user. There must be either one "%d" and no other "%"
820 * characters.
821 */
822 if (p[1] != 'd' || strchr(p + 2, '%'))
823 return -EINVAL;
824
825 /* Use one page as a bit array of possible slots */
826 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
827 if (!inuse)
828 return -ENOMEM;
829
830 for_each_netdev(net, d) {
831 if (!sscanf(d->name, name, &i))
832 continue;
833 if (i < 0 || i >= max_netdevices)
834 continue;
835
836 /* avoid cases where sscanf is not exact inverse of printf */
837 snprintf(buf, IFNAMSIZ, name, i);
838 if (!strncmp(buf, d->name, IFNAMSIZ))
839 set_bit(i, inuse);
840 }
841
842 i = find_first_zero_bit(inuse, max_netdevices);
843 free_page((unsigned long) inuse);
844 }
845
846 snprintf(buf, IFNAMSIZ, name, i);
847 if (!__dev_get_by_name(net, buf))
848 return i;
849
850 /* It is possible to run out of possible slots
851 * when the name is long and there isn't enough space left
852 * for the digits, or if all bits are used.
853 */
854 return -ENFILE;
855 }
856
857 /**
858 * dev_alloc_name - allocate a name for a device
859 * @dev: device
860 * @name: name format string
861 *
862 * Passed a format string - eg "lt%d" it will try and find a suitable
863 * id. It scans list of devices to build up a free map, then chooses
864 * the first empty slot. The caller must hold the dev_base or rtnl lock
865 * while allocating the name and adding the device in order to avoid
866 * duplicates.
867 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
868 * Returns the number of the unit assigned or a negative errno code.
869 */
870
871 int dev_alloc_name(struct net_device *dev, const char *name)
872 {
873 char buf[IFNAMSIZ];
874 struct net *net;
875 int ret;
876
877 BUG_ON(!dev_net(dev));
878 net = dev_net(dev);
879 ret = __dev_alloc_name(net, name, buf);
880 if (ret >= 0)
881 strlcpy(dev->name, buf, IFNAMSIZ);
882 return ret;
883 }
884
885
886 /**
887 * dev_change_name - change name of a device
888 * @dev: device
889 * @newname: name (or format string) must be at least IFNAMSIZ
890 *
891 * Change name of a device, can pass format strings "eth%d".
892 * for wildcarding.
893 */
894 int dev_change_name(struct net_device *dev, const char *newname)
895 {
896 char oldname[IFNAMSIZ];
897 int err = 0;
898 int ret;
899 struct net *net;
900
901 ASSERT_RTNL();
902 BUG_ON(!dev_net(dev));
903
904 net = dev_net(dev);
905 if (dev->flags & IFF_UP)
906 return -EBUSY;
907
908 if (!dev_valid_name(newname))
909 return -EINVAL;
910
911 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
912 return 0;
913
914 memcpy(oldname, dev->name, IFNAMSIZ);
915
916 if (strchr(newname, '%')) {
917 err = dev_alloc_name(dev, newname);
918 if (err < 0)
919 return err;
920 }
921 else if (__dev_get_by_name(net, newname))
922 return -EEXIST;
923 else
924 strlcpy(dev->name, newname, IFNAMSIZ);
925
926 rollback:
927 ret = device_rename(&dev->dev, dev->name);
928 if (ret) {
929 memcpy(dev->name, oldname, IFNAMSIZ);
930 return ret;
931 }
932
933 write_lock_bh(&dev_base_lock);
934 hlist_del(&dev->name_hlist);
935 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
936 write_unlock_bh(&dev_base_lock);
937
938 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
939 ret = notifier_to_errno(ret);
940
941 if (ret) {
942 if (err) {
943 printk(KERN_ERR
944 "%s: name change rollback failed: %d.\n",
945 dev->name, ret);
946 } else {
947 err = ret;
948 memcpy(dev->name, oldname, IFNAMSIZ);
949 goto rollback;
950 }
951 }
952
953 return err;
954 }
955
956 /**
957 * dev_set_alias - change ifalias of a device
958 * @dev: device
959 * @alias: name up to IFALIASZ
960 * @len: limit of bytes to copy from info
961 *
962 * Set ifalias for a device,
963 */
964 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
965 {
966 ASSERT_RTNL();
967
968 if (len >= IFALIASZ)
969 return -EINVAL;
970
971 if (!len) {
972 if (dev->ifalias) {
973 kfree(dev->ifalias);
974 dev->ifalias = NULL;
975 }
976 return 0;
977 }
978
979 dev->ifalias = krealloc(dev->ifalias, len+1, GFP_KERNEL);
980 if (!dev->ifalias)
981 return -ENOMEM;
982
983 strlcpy(dev->ifalias, alias, len+1);
984 return len;
985 }
986
987
988 /**
989 * netdev_features_change - device changes features
990 * @dev: device to cause notification
991 *
992 * Called to indicate a device has changed features.
993 */
994 void netdev_features_change(struct net_device *dev)
995 {
996 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
997 }
998 EXPORT_SYMBOL(netdev_features_change);
999
1000 /**
1001 * netdev_state_change - device changes state
1002 * @dev: device to cause notification
1003 *
1004 * Called to indicate a device has changed state. This function calls
1005 * the notifier chains for netdev_chain and sends a NEWLINK message
1006 * to the routing socket.
1007 */
1008 void netdev_state_change(struct net_device *dev)
1009 {
1010 if (dev->flags & IFF_UP) {
1011 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1012 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1013 }
1014 }
1015
1016 void netdev_bonding_change(struct net_device *dev)
1017 {
1018 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, dev);
1019 }
1020 EXPORT_SYMBOL(netdev_bonding_change);
1021
1022 /**
1023 * dev_load - load a network module
1024 * @net: the applicable net namespace
1025 * @name: name of interface
1026 *
1027 * If a network interface is not present and the process has suitable
1028 * privileges this function loads the module. If module loading is not
1029 * available in this kernel then it becomes a nop.
1030 */
1031
1032 void dev_load(struct net *net, const char *name)
1033 {
1034 struct net_device *dev;
1035
1036 read_lock(&dev_base_lock);
1037 dev = __dev_get_by_name(net, name);
1038 read_unlock(&dev_base_lock);
1039
1040 if (!dev && capable(CAP_SYS_MODULE))
1041 request_module("%s", name);
1042 }
1043
1044 /**
1045 * dev_open - prepare an interface for use.
1046 * @dev: device to open
1047 *
1048 * Takes a device from down to up state. The device's private open
1049 * function is invoked and then the multicast lists are loaded. Finally
1050 * the device is moved into the up state and a %NETDEV_UP message is
1051 * sent to the netdev notifier chain.
1052 *
1053 * Calling this function on an active interface is a nop. On a failure
1054 * a negative errno code is returned.
1055 */
1056 int dev_open(struct net_device *dev)
1057 {
1058 int ret = 0;
1059
1060 ASSERT_RTNL();
1061
1062 /*
1063 * Is it already up?
1064 */
1065
1066 if (dev->flags & IFF_UP)
1067 return 0;
1068
1069 /*
1070 * Is it even present?
1071 */
1072 if (!netif_device_present(dev))
1073 return -ENODEV;
1074
1075 /*
1076 * Call device private open method
1077 */
1078 set_bit(__LINK_STATE_START, &dev->state);
1079
1080 if (dev->validate_addr)
1081 ret = dev->validate_addr(dev);
1082
1083 if (!ret && dev->open)
1084 ret = dev->open(dev);
1085
1086 /*
1087 * If it went open OK then:
1088 */
1089
1090 if (ret)
1091 clear_bit(__LINK_STATE_START, &dev->state);
1092 else {
1093 /*
1094 * Set the flags.
1095 */
1096 dev->flags |= IFF_UP;
1097
1098 /*
1099 * Initialize multicasting status
1100 */
1101 dev_set_rx_mode(dev);
1102
1103 /*
1104 * Wakeup transmit queue engine
1105 */
1106 dev_activate(dev);
1107
1108 /*
1109 * ... and announce new interface.
1110 */
1111 call_netdevice_notifiers(NETDEV_UP, dev);
1112 }
1113
1114 return ret;
1115 }
1116
1117 /**
1118 * dev_close - shutdown an interface.
1119 * @dev: device to shutdown
1120 *
1121 * This function moves an active device into down state. A
1122 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1123 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1124 * chain.
1125 */
1126 int dev_close(struct net_device *dev)
1127 {
1128 ASSERT_RTNL();
1129
1130 might_sleep();
1131
1132 if (!(dev->flags & IFF_UP))
1133 return 0;
1134
1135 /*
1136 * Tell people we are going down, so that they can
1137 * prepare to death, when device is still operating.
1138 */
1139 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1140
1141 clear_bit(__LINK_STATE_START, &dev->state);
1142
1143 /* Synchronize to scheduled poll. We cannot touch poll list,
1144 * it can be even on different cpu. So just clear netif_running().
1145 *
1146 * dev->stop() will invoke napi_disable() on all of it's
1147 * napi_struct instances on this device.
1148 */
1149 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1150
1151 dev_deactivate(dev);
1152
1153 /*
1154 * Call the device specific close. This cannot fail.
1155 * Only if device is UP
1156 *
1157 * We allow it to be called even after a DETACH hot-plug
1158 * event.
1159 */
1160 if (dev->stop)
1161 dev->stop(dev);
1162
1163 /*
1164 * Device is now down.
1165 */
1166
1167 dev->flags &= ~IFF_UP;
1168
1169 /*
1170 * Tell people we are down
1171 */
1172 call_netdevice_notifiers(NETDEV_DOWN, dev);
1173
1174 return 0;
1175 }
1176
1177
1178 /**
1179 * dev_disable_lro - disable Large Receive Offload on a device
1180 * @dev: device
1181 *
1182 * Disable Large Receive Offload (LRO) on a net device. Must be
1183 * called under RTNL. This is needed if received packets may be
1184 * forwarded to another interface.
1185 */
1186 void dev_disable_lro(struct net_device *dev)
1187 {
1188 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1189 dev->ethtool_ops->set_flags) {
1190 u32 flags = dev->ethtool_ops->get_flags(dev);
1191 if (flags & ETH_FLAG_LRO) {
1192 flags &= ~ETH_FLAG_LRO;
1193 dev->ethtool_ops->set_flags(dev, flags);
1194 }
1195 }
1196 WARN_ON(dev->features & NETIF_F_LRO);
1197 }
1198 EXPORT_SYMBOL(dev_disable_lro);
1199
1200
1201 static int dev_boot_phase = 1;
1202
1203 /*
1204 * Device change register/unregister. These are not inline or static
1205 * as we export them to the world.
1206 */
1207
1208 /**
1209 * register_netdevice_notifier - register a network notifier block
1210 * @nb: notifier
1211 *
1212 * Register a notifier to be called when network device events occur.
1213 * The notifier passed is linked into the kernel structures and must
1214 * not be reused until it has been unregistered. A negative errno code
1215 * is returned on a failure.
1216 *
1217 * When registered all registration and up events are replayed
1218 * to the new notifier to allow device to have a race free
1219 * view of the network device list.
1220 */
1221
1222 int register_netdevice_notifier(struct notifier_block *nb)
1223 {
1224 struct net_device *dev;
1225 struct net_device *last;
1226 struct net *net;
1227 int err;
1228
1229 rtnl_lock();
1230 err = raw_notifier_chain_register(&netdev_chain, nb);
1231 if (err)
1232 goto unlock;
1233 if (dev_boot_phase)
1234 goto unlock;
1235 for_each_net(net) {
1236 for_each_netdev(net, dev) {
1237 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1238 err = notifier_to_errno(err);
1239 if (err)
1240 goto rollback;
1241
1242 if (!(dev->flags & IFF_UP))
1243 continue;
1244
1245 nb->notifier_call(nb, NETDEV_UP, dev);
1246 }
1247 }
1248
1249 unlock:
1250 rtnl_unlock();
1251 return err;
1252
1253 rollback:
1254 last = dev;
1255 for_each_net(net) {
1256 for_each_netdev(net, dev) {
1257 if (dev == last)
1258 break;
1259
1260 if (dev->flags & IFF_UP) {
1261 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1262 nb->notifier_call(nb, NETDEV_DOWN, dev);
1263 }
1264 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1265 }
1266 }
1267
1268 raw_notifier_chain_unregister(&netdev_chain, nb);
1269 goto unlock;
1270 }
1271
1272 /**
1273 * unregister_netdevice_notifier - unregister a network notifier block
1274 * @nb: notifier
1275 *
1276 * Unregister a notifier previously registered by
1277 * register_netdevice_notifier(). The notifier is unlinked into the
1278 * kernel structures and may then be reused. A negative errno code
1279 * is returned on a failure.
1280 */
1281
1282 int unregister_netdevice_notifier(struct notifier_block *nb)
1283 {
1284 int err;
1285
1286 rtnl_lock();
1287 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1288 rtnl_unlock();
1289 return err;
1290 }
1291
1292 /**
1293 * call_netdevice_notifiers - call all network notifier blocks
1294 * @val: value passed unmodified to notifier function
1295 * @dev: net_device pointer passed unmodified to notifier function
1296 *
1297 * Call all network notifier blocks. Parameters and return value
1298 * are as for raw_notifier_call_chain().
1299 */
1300
1301 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1302 {
1303 return raw_notifier_call_chain(&netdev_chain, val, dev);
1304 }
1305
1306 /* When > 0 there are consumers of rx skb time stamps */
1307 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1308
1309 void net_enable_timestamp(void)
1310 {
1311 atomic_inc(&netstamp_needed);
1312 }
1313
1314 void net_disable_timestamp(void)
1315 {
1316 atomic_dec(&netstamp_needed);
1317 }
1318
1319 static inline void net_timestamp(struct sk_buff *skb)
1320 {
1321 if (atomic_read(&netstamp_needed))
1322 __net_timestamp(skb);
1323 else
1324 skb->tstamp.tv64 = 0;
1325 }
1326
1327 /*
1328 * Support routine. Sends outgoing frames to any network
1329 * taps currently in use.
1330 */
1331
1332 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1333 {
1334 struct packet_type *ptype;
1335
1336 net_timestamp(skb);
1337
1338 rcu_read_lock();
1339 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1340 /* Never send packets back to the socket
1341 * they originated from - MvS (miquels@drinkel.ow.org)
1342 */
1343 if ((ptype->dev == dev || !ptype->dev) &&
1344 (ptype->af_packet_priv == NULL ||
1345 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1346 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1347 if (!skb2)
1348 break;
1349
1350 /* skb->nh should be correctly
1351 set by sender, so that the second statement is
1352 just protection against buggy protocols.
1353 */
1354 skb_reset_mac_header(skb2);
1355
1356 if (skb_network_header(skb2) < skb2->data ||
1357 skb2->network_header > skb2->tail) {
1358 if (net_ratelimit())
1359 printk(KERN_CRIT "protocol %04x is "
1360 "buggy, dev %s\n",
1361 skb2->protocol, dev->name);
1362 skb_reset_network_header(skb2);
1363 }
1364
1365 skb2->transport_header = skb2->network_header;
1366 skb2->pkt_type = PACKET_OUTGOING;
1367 ptype->func(skb2, skb->dev, ptype, skb->dev);
1368 }
1369 }
1370 rcu_read_unlock();
1371 }
1372
1373
1374 static inline void __netif_reschedule(struct Qdisc *q)
1375 {
1376 struct softnet_data *sd;
1377 unsigned long flags;
1378
1379 local_irq_save(flags);
1380 sd = &__get_cpu_var(softnet_data);
1381 q->next_sched = sd->output_queue;
1382 sd->output_queue = q;
1383 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1384 local_irq_restore(flags);
1385 }
1386
1387 void __netif_schedule(struct Qdisc *q)
1388 {
1389 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1390 __netif_reschedule(q);
1391 }
1392 EXPORT_SYMBOL(__netif_schedule);
1393
1394 void dev_kfree_skb_irq(struct sk_buff *skb)
1395 {
1396 if (atomic_dec_and_test(&skb->users)) {
1397 struct softnet_data *sd;
1398 unsigned long flags;
1399
1400 local_irq_save(flags);
1401 sd = &__get_cpu_var(softnet_data);
1402 skb->next = sd->completion_queue;
1403 sd->completion_queue = skb;
1404 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1405 local_irq_restore(flags);
1406 }
1407 }
1408 EXPORT_SYMBOL(dev_kfree_skb_irq);
1409
1410 void dev_kfree_skb_any(struct sk_buff *skb)
1411 {
1412 if (in_irq() || irqs_disabled())
1413 dev_kfree_skb_irq(skb);
1414 else
1415 dev_kfree_skb(skb);
1416 }
1417 EXPORT_SYMBOL(dev_kfree_skb_any);
1418
1419
1420 /**
1421 * netif_device_detach - mark device as removed
1422 * @dev: network device
1423 *
1424 * Mark device as removed from system and therefore no longer available.
1425 */
1426 void netif_device_detach(struct net_device *dev)
1427 {
1428 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1429 netif_running(dev)) {
1430 netif_stop_queue(dev);
1431 }
1432 }
1433 EXPORT_SYMBOL(netif_device_detach);
1434
1435 /**
1436 * netif_device_attach - mark device as attached
1437 * @dev: network device
1438 *
1439 * Mark device as attached from system and restart if needed.
1440 */
1441 void netif_device_attach(struct net_device *dev)
1442 {
1443 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1444 netif_running(dev)) {
1445 netif_wake_queue(dev);
1446 __netdev_watchdog_up(dev);
1447 }
1448 }
1449 EXPORT_SYMBOL(netif_device_attach);
1450
1451 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1452 {
1453 return ((features & NETIF_F_GEN_CSUM) ||
1454 ((features & NETIF_F_IP_CSUM) &&
1455 protocol == htons(ETH_P_IP)) ||
1456 ((features & NETIF_F_IPV6_CSUM) &&
1457 protocol == htons(ETH_P_IPV6)));
1458 }
1459
1460 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1461 {
1462 if (can_checksum_protocol(dev->features, skb->protocol))
1463 return true;
1464
1465 if (skb->protocol == htons(ETH_P_8021Q)) {
1466 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1467 if (can_checksum_protocol(dev->features & dev->vlan_features,
1468 veh->h_vlan_encapsulated_proto))
1469 return true;
1470 }
1471
1472 return false;
1473 }
1474
1475 /*
1476 * Invalidate hardware checksum when packet is to be mangled, and
1477 * complete checksum manually on outgoing path.
1478 */
1479 int skb_checksum_help(struct sk_buff *skb)
1480 {
1481 __wsum csum;
1482 int ret = 0, offset;
1483
1484 if (skb->ip_summed == CHECKSUM_COMPLETE)
1485 goto out_set_summed;
1486
1487 if (unlikely(skb_shinfo(skb)->gso_size)) {
1488 /* Let GSO fix up the checksum. */
1489 goto out_set_summed;
1490 }
1491
1492 offset = skb->csum_start - skb_headroom(skb);
1493 BUG_ON(offset >= skb_headlen(skb));
1494 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1495
1496 offset += skb->csum_offset;
1497 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1498
1499 if (skb_cloned(skb) &&
1500 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1501 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1502 if (ret)
1503 goto out;
1504 }
1505
1506 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1507 out_set_summed:
1508 skb->ip_summed = CHECKSUM_NONE;
1509 out:
1510 return ret;
1511 }
1512
1513 /**
1514 * skb_gso_segment - Perform segmentation on skb.
1515 * @skb: buffer to segment
1516 * @features: features for the output path (see dev->features)
1517 *
1518 * This function segments the given skb and returns a list of segments.
1519 *
1520 * It may return NULL if the skb requires no segmentation. This is
1521 * only possible when GSO is used for verifying header integrity.
1522 */
1523 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1524 {
1525 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1526 struct packet_type *ptype;
1527 __be16 type = skb->protocol;
1528 int err;
1529
1530 BUG_ON(skb_shinfo(skb)->frag_list);
1531
1532 skb_reset_mac_header(skb);
1533 skb->mac_len = skb->network_header - skb->mac_header;
1534 __skb_pull(skb, skb->mac_len);
1535
1536 if (WARN_ON(skb->ip_summed != CHECKSUM_PARTIAL)) {
1537 if (skb_header_cloned(skb) &&
1538 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1539 return ERR_PTR(err);
1540 }
1541
1542 rcu_read_lock();
1543 list_for_each_entry_rcu(ptype,
1544 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1545 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1546 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1547 err = ptype->gso_send_check(skb);
1548 segs = ERR_PTR(err);
1549 if (err || skb_gso_ok(skb, features))
1550 break;
1551 __skb_push(skb, (skb->data -
1552 skb_network_header(skb)));
1553 }
1554 segs = ptype->gso_segment(skb, features);
1555 break;
1556 }
1557 }
1558 rcu_read_unlock();
1559
1560 __skb_push(skb, skb->data - skb_mac_header(skb));
1561
1562 return segs;
1563 }
1564
1565 EXPORT_SYMBOL(skb_gso_segment);
1566
1567 /* Take action when hardware reception checksum errors are detected. */
1568 #ifdef CONFIG_BUG
1569 void netdev_rx_csum_fault(struct net_device *dev)
1570 {
1571 if (net_ratelimit()) {
1572 printk(KERN_ERR "%s: hw csum failure.\n",
1573 dev ? dev->name : "<unknown>");
1574 dump_stack();
1575 }
1576 }
1577 EXPORT_SYMBOL(netdev_rx_csum_fault);
1578 #endif
1579
1580 /* Actually, we should eliminate this check as soon as we know, that:
1581 * 1. IOMMU is present and allows to map all the memory.
1582 * 2. No high memory really exists on this machine.
1583 */
1584
1585 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1586 {
1587 #ifdef CONFIG_HIGHMEM
1588 int i;
1589
1590 if (dev->features & NETIF_F_HIGHDMA)
1591 return 0;
1592
1593 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1594 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1595 return 1;
1596
1597 #endif
1598 return 0;
1599 }
1600
1601 struct dev_gso_cb {
1602 void (*destructor)(struct sk_buff *skb);
1603 };
1604
1605 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1606
1607 static void dev_gso_skb_destructor(struct sk_buff *skb)
1608 {
1609 struct dev_gso_cb *cb;
1610
1611 do {
1612 struct sk_buff *nskb = skb->next;
1613
1614 skb->next = nskb->next;
1615 nskb->next = NULL;
1616 kfree_skb(nskb);
1617 } while (skb->next);
1618
1619 cb = DEV_GSO_CB(skb);
1620 if (cb->destructor)
1621 cb->destructor(skb);
1622 }
1623
1624 /**
1625 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1626 * @skb: buffer to segment
1627 *
1628 * This function segments the given skb and stores the list of segments
1629 * in skb->next.
1630 */
1631 static int dev_gso_segment(struct sk_buff *skb)
1632 {
1633 struct net_device *dev = skb->dev;
1634 struct sk_buff *segs;
1635 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1636 NETIF_F_SG : 0);
1637
1638 segs = skb_gso_segment(skb, features);
1639
1640 /* Verifying header integrity only. */
1641 if (!segs)
1642 return 0;
1643
1644 if (IS_ERR(segs))
1645 return PTR_ERR(segs);
1646
1647 skb->next = segs;
1648 DEV_GSO_CB(skb)->destructor = skb->destructor;
1649 skb->destructor = dev_gso_skb_destructor;
1650
1651 return 0;
1652 }
1653
1654 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1655 struct netdev_queue *txq)
1656 {
1657 if (likely(!skb->next)) {
1658 if (!list_empty(&ptype_all))
1659 dev_queue_xmit_nit(skb, dev);
1660
1661 if (netif_needs_gso(dev, skb)) {
1662 if (unlikely(dev_gso_segment(skb)))
1663 goto out_kfree_skb;
1664 if (skb->next)
1665 goto gso;
1666 }
1667
1668 return dev->hard_start_xmit(skb, dev);
1669 }
1670
1671 gso:
1672 do {
1673 struct sk_buff *nskb = skb->next;
1674 int rc;
1675
1676 skb->next = nskb->next;
1677 nskb->next = NULL;
1678 rc = dev->hard_start_xmit(nskb, dev);
1679 if (unlikely(rc)) {
1680 nskb->next = skb->next;
1681 skb->next = nskb;
1682 return rc;
1683 }
1684 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1685 return NETDEV_TX_BUSY;
1686 } while (skb->next);
1687
1688 skb->destructor = DEV_GSO_CB(skb)->destructor;
1689
1690 out_kfree_skb:
1691 kfree_skb(skb);
1692 return 0;
1693 }
1694
1695 static u32 simple_tx_hashrnd;
1696 static int simple_tx_hashrnd_initialized = 0;
1697
1698 static u16 simple_tx_hash(struct net_device *dev, struct sk_buff *skb)
1699 {
1700 u32 addr1, addr2, ports;
1701 u32 hash, ihl;
1702 u8 ip_proto = 0;
1703
1704 if (unlikely(!simple_tx_hashrnd_initialized)) {
1705 get_random_bytes(&simple_tx_hashrnd, 4);
1706 simple_tx_hashrnd_initialized = 1;
1707 }
1708
1709 switch (skb->protocol) {
1710 case htons(ETH_P_IP):
1711 if (!(ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)))
1712 ip_proto = ip_hdr(skb)->protocol;
1713 addr1 = ip_hdr(skb)->saddr;
1714 addr2 = ip_hdr(skb)->daddr;
1715 ihl = ip_hdr(skb)->ihl;
1716 break;
1717 case htons(ETH_P_IPV6):
1718 ip_proto = ipv6_hdr(skb)->nexthdr;
1719 addr1 = ipv6_hdr(skb)->saddr.s6_addr32[3];
1720 addr2 = ipv6_hdr(skb)->daddr.s6_addr32[3];
1721 ihl = (40 >> 2);
1722 break;
1723 default:
1724 return 0;
1725 }
1726
1727
1728 switch (ip_proto) {
1729 case IPPROTO_TCP:
1730 case IPPROTO_UDP:
1731 case IPPROTO_DCCP:
1732 case IPPROTO_ESP:
1733 case IPPROTO_AH:
1734 case IPPROTO_SCTP:
1735 case IPPROTO_UDPLITE:
1736 ports = *((u32 *) (skb_network_header(skb) + (ihl * 4)));
1737 break;
1738
1739 default:
1740 ports = 0;
1741 break;
1742 }
1743
1744 hash = jhash_3words(addr1, addr2, ports, simple_tx_hashrnd);
1745
1746 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1747 }
1748
1749 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1750 struct sk_buff *skb)
1751 {
1752 u16 queue_index = 0;
1753
1754 if (dev->select_queue)
1755 queue_index = dev->select_queue(dev, skb);
1756 else if (dev->real_num_tx_queues > 1)
1757 queue_index = simple_tx_hash(dev, skb);
1758
1759 skb_set_queue_mapping(skb, queue_index);
1760 return netdev_get_tx_queue(dev, queue_index);
1761 }
1762
1763 /**
1764 * dev_queue_xmit - transmit a buffer
1765 * @skb: buffer to transmit
1766 *
1767 * Queue a buffer for transmission to a network device. The caller must
1768 * have set the device and priority and built the buffer before calling
1769 * this function. The function can be called from an interrupt.
1770 *
1771 * A negative errno code is returned on a failure. A success does not
1772 * guarantee the frame will be transmitted as it may be dropped due
1773 * to congestion or traffic shaping.
1774 *
1775 * -----------------------------------------------------------------------------------
1776 * I notice this method can also return errors from the queue disciplines,
1777 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1778 * be positive.
1779 *
1780 * Regardless of the return value, the skb is consumed, so it is currently
1781 * difficult to retry a send to this method. (You can bump the ref count
1782 * before sending to hold a reference for retry if you are careful.)
1783 *
1784 * When calling this method, interrupts MUST be enabled. This is because
1785 * the BH enable code must have IRQs enabled so that it will not deadlock.
1786 * --BLG
1787 */
1788 int dev_queue_xmit(struct sk_buff *skb)
1789 {
1790 struct net_device *dev = skb->dev;
1791 struct netdev_queue *txq;
1792 struct Qdisc *q;
1793 int rc = -ENOMEM;
1794
1795 /* GSO will handle the following emulations directly. */
1796 if (netif_needs_gso(dev, skb))
1797 goto gso;
1798
1799 if (skb_shinfo(skb)->frag_list &&
1800 !(dev->features & NETIF_F_FRAGLIST) &&
1801 __skb_linearize(skb))
1802 goto out_kfree_skb;
1803
1804 /* Fragmented skb is linearized if device does not support SG,
1805 * or if at least one of fragments is in highmem and device
1806 * does not support DMA from it.
1807 */
1808 if (skb_shinfo(skb)->nr_frags &&
1809 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1810 __skb_linearize(skb))
1811 goto out_kfree_skb;
1812
1813 /* If packet is not checksummed and device does not support
1814 * checksumming for this protocol, complete checksumming here.
1815 */
1816 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1817 skb_set_transport_header(skb, skb->csum_start -
1818 skb_headroom(skb));
1819 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
1820 goto out_kfree_skb;
1821 }
1822
1823 gso:
1824 /* Disable soft irqs for various locks below. Also
1825 * stops preemption for RCU.
1826 */
1827 rcu_read_lock_bh();
1828
1829 txq = dev_pick_tx(dev, skb);
1830 q = rcu_dereference(txq->qdisc);
1831
1832 #ifdef CONFIG_NET_CLS_ACT
1833 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1834 #endif
1835 if (q->enqueue) {
1836 spinlock_t *root_lock = qdisc_lock(q);
1837
1838 spin_lock(root_lock);
1839
1840 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
1841 kfree_skb(skb);
1842 rc = NET_XMIT_DROP;
1843 } else {
1844 rc = qdisc_enqueue_root(skb, q);
1845 qdisc_run(q);
1846 }
1847 spin_unlock(root_lock);
1848
1849 goto out;
1850 }
1851
1852 /* The device has no queue. Common case for software devices:
1853 loopback, all the sorts of tunnels...
1854
1855 Really, it is unlikely that netif_tx_lock protection is necessary
1856 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1857 counters.)
1858 However, it is possible, that they rely on protection
1859 made by us here.
1860
1861 Check this and shot the lock. It is not prone from deadlocks.
1862 Either shot noqueue qdisc, it is even simpler 8)
1863 */
1864 if (dev->flags & IFF_UP) {
1865 int cpu = smp_processor_id(); /* ok because BHs are off */
1866
1867 if (txq->xmit_lock_owner != cpu) {
1868
1869 HARD_TX_LOCK(dev, txq, cpu);
1870
1871 if (!netif_tx_queue_stopped(txq)) {
1872 rc = 0;
1873 if (!dev_hard_start_xmit(skb, dev, txq)) {
1874 HARD_TX_UNLOCK(dev, txq);
1875 goto out;
1876 }
1877 }
1878 HARD_TX_UNLOCK(dev, txq);
1879 if (net_ratelimit())
1880 printk(KERN_CRIT "Virtual device %s asks to "
1881 "queue packet!\n", dev->name);
1882 } else {
1883 /* Recursion is detected! It is possible,
1884 * unfortunately */
1885 if (net_ratelimit())
1886 printk(KERN_CRIT "Dead loop on virtual device "
1887 "%s, fix it urgently!\n", dev->name);
1888 }
1889 }
1890
1891 rc = -ENETDOWN;
1892 rcu_read_unlock_bh();
1893
1894 out_kfree_skb:
1895 kfree_skb(skb);
1896 return rc;
1897 out:
1898 rcu_read_unlock_bh();
1899 return rc;
1900 }
1901
1902
1903 /*=======================================================================
1904 Receiver routines
1905 =======================================================================*/
1906
1907 int netdev_max_backlog __read_mostly = 1000;
1908 int netdev_budget __read_mostly = 300;
1909 int weight_p __read_mostly = 64; /* old backlog weight */
1910
1911 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1912
1913
1914 /**
1915 * netif_rx - post buffer to the network code
1916 * @skb: buffer to post
1917 *
1918 * This function receives a packet from a device driver and queues it for
1919 * the upper (protocol) levels to process. It always succeeds. The buffer
1920 * may be dropped during processing for congestion control or by the
1921 * protocol layers.
1922 *
1923 * return values:
1924 * NET_RX_SUCCESS (no congestion)
1925 * NET_RX_DROP (packet was dropped)
1926 *
1927 */
1928
1929 int netif_rx(struct sk_buff *skb)
1930 {
1931 struct softnet_data *queue;
1932 unsigned long flags;
1933
1934 /* if netpoll wants it, pretend we never saw it */
1935 if (netpoll_rx(skb))
1936 return NET_RX_DROP;
1937
1938 if (!skb->tstamp.tv64)
1939 net_timestamp(skb);
1940
1941 /*
1942 * The code is rearranged so that the path is the most
1943 * short when CPU is congested, but is still operating.
1944 */
1945 local_irq_save(flags);
1946 queue = &__get_cpu_var(softnet_data);
1947
1948 __get_cpu_var(netdev_rx_stat).total++;
1949 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1950 if (queue->input_pkt_queue.qlen) {
1951 enqueue:
1952 __skb_queue_tail(&queue->input_pkt_queue, skb);
1953 local_irq_restore(flags);
1954 return NET_RX_SUCCESS;
1955 }
1956
1957 napi_schedule(&queue->backlog);
1958 goto enqueue;
1959 }
1960
1961 __get_cpu_var(netdev_rx_stat).dropped++;
1962 local_irq_restore(flags);
1963
1964 kfree_skb(skb);
1965 return NET_RX_DROP;
1966 }
1967
1968 int netif_rx_ni(struct sk_buff *skb)
1969 {
1970 int err;
1971
1972 preempt_disable();
1973 err = netif_rx(skb);
1974 if (local_softirq_pending())
1975 do_softirq();
1976 preempt_enable();
1977
1978 return err;
1979 }
1980
1981 EXPORT_SYMBOL(netif_rx_ni);
1982
1983 static void net_tx_action(struct softirq_action *h)
1984 {
1985 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1986
1987 if (sd->completion_queue) {
1988 struct sk_buff *clist;
1989
1990 local_irq_disable();
1991 clist = sd->completion_queue;
1992 sd->completion_queue = NULL;
1993 local_irq_enable();
1994
1995 while (clist) {
1996 struct sk_buff *skb = clist;
1997 clist = clist->next;
1998
1999 WARN_ON(atomic_read(&skb->users));
2000 __kfree_skb(skb);
2001 }
2002 }
2003
2004 if (sd->output_queue) {
2005 struct Qdisc *head;
2006
2007 local_irq_disable();
2008 head = sd->output_queue;
2009 sd->output_queue = NULL;
2010 local_irq_enable();
2011
2012 while (head) {
2013 struct Qdisc *q = head;
2014 spinlock_t *root_lock;
2015
2016 head = head->next_sched;
2017
2018 root_lock = qdisc_lock(q);
2019 if (spin_trylock(root_lock)) {
2020 smp_mb__before_clear_bit();
2021 clear_bit(__QDISC_STATE_SCHED,
2022 &q->state);
2023 qdisc_run(q);
2024 spin_unlock(root_lock);
2025 } else {
2026 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2027 &q->state)) {
2028 __netif_reschedule(q);
2029 } else {
2030 smp_mb__before_clear_bit();
2031 clear_bit(__QDISC_STATE_SCHED,
2032 &q->state);
2033 }
2034 }
2035 }
2036 }
2037 }
2038
2039 static inline int deliver_skb(struct sk_buff *skb,
2040 struct packet_type *pt_prev,
2041 struct net_device *orig_dev)
2042 {
2043 atomic_inc(&skb->users);
2044 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2045 }
2046
2047 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2048 /* These hooks defined here for ATM */
2049 struct net_bridge;
2050 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
2051 unsigned char *addr);
2052 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent) __read_mostly;
2053
2054 /*
2055 * If bridge module is loaded call bridging hook.
2056 * returns NULL if packet was consumed.
2057 */
2058 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2059 struct sk_buff *skb) __read_mostly;
2060 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2061 struct packet_type **pt_prev, int *ret,
2062 struct net_device *orig_dev)
2063 {
2064 struct net_bridge_port *port;
2065
2066 if (skb->pkt_type == PACKET_LOOPBACK ||
2067 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2068 return skb;
2069
2070 if (*pt_prev) {
2071 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2072 *pt_prev = NULL;
2073 }
2074
2075 return br_handle_frame_hook(port, skb);
2076 }
2077 #else
2078 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2079 #endif
2080
2081 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2082 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2083 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2084
2085 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2086 struct packet_type **pt_prev,
2087 int *ret,
2088 struct net_device *orig_dev)
2089 {
2090 if (skb->dev->macvlan_port == NULL)
2091 return skb;
2092
2093 if (*pt_prev) {
2094 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2095 *pt_prev = NULL;
2096 }
2097 return macvlan_handle_frame_hook(skb);
2098 }
2099 #else
2100 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2101 #endif
2102
2103 #ifdef CONFIG_NET_CLS_ACT
2104 /* TODO: Maybe we should just force sch_ingress to be compiled in
2105 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2106 * a compare and 2 stores extra right now if we dont have it on
2107 * but have CONFIG_NET_CLS_ACT
2108 * NOTE: This doesnt stop any functionality; if you dont have
2109 * the ingress scheduler, you just cant add policies on ingress.
2110 *
2111 */
2112 static int ing_filter(struct sk_buff *skb)
2113 {
2114 struct net_device *dev = skb->dev;
2115 u32 ttl = G_TC_RTTL(skb->tc_verd);
2116 struct netdev_queue *rxq;
2117 int result = TC_ACT_OK;
2118 struct Qdisc *q;
2119
2120 if (MAX_RED_LOOP < ttl++) {
2121 printk(KERN_WARNING
2122 "Redir loop detected Dropping packet (%d->%d)\n",
2123 skb->iif, dev->ifindex);
2124 return TC_ACT_SHOT;
2125 }
2126
2127 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2128 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2129
2130 rxq = &dev->rx_queue;
2131
2132 q = rxq->qdisc;
2133 if (q != &noop_qdisc) {
2134 spin_lock(qdisc_lock(q));
2135 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2136 result = qdisc_enqueue_root(skb, q);
2137 spin_unlock(qdisc_lock(q));
2138 }
2139
2140 return result;
2141 }
2142
2143 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2144 struct packet_type **pt_prev,
2145 int *ret, struct net_device *orig_dev)
2146 {
2147 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2148 goto out;
2149
2150 if (*pt_prev) {
2151 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2152 *pt_prev = NULL;
2153 } else {
2154 /* Huh? Why does turning on AF_PACKET affect this? */
2155 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2156 }
2157
2158 switch (ing_filter(skb)) {
2159 case TC_ACT_SHOT:
2160 case TC_ACT_STOLEN:
2161 kfree_skb(skb);
2162 return NULL;
2163 }
2164
2165 out:
2166 skb->tc_verd = 0;
2167 return skb;
2168 }
2169 #endif
2170
2171 /*
2172 * netif_nit_deliver - deliver received packets to network taps
2173 * @skb: buffer
2174 *
2175 * This function is used to deliver incoming packets to network
2176 * taps. It should be used when the normal netif_receive_skb path
2177 * is bypassed, for example because of VLAN acceleration.
2178 */
2179 void netif_nit_deliver(struct sk_buff *skb)
2180 {
2181 struct packet_type *ptype;
2182
2183 if (list_empty(&ptype_all))
2184 return;
2185
2186 skb_reset_network_header(skb);
2187 skb_reset_transport_header(skb);
2188 skb->mac_len = skb->network_header - skb->mac_header;
2189
2190 rcu_read_lock();
2191 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2192 if (!ptype->dev || ptype->dev == skb->dev)
2193 deliver_skb(skb, ptype, skb->dev);
2194 }
2195 rcu_read_unlock();
2196 }
2197
2198 /**
2199 * netif_receive_skb - process receive buffer from network
2200 * @skb: buffer to process
2201 *
2202 * netif_receive_skb() is the main receive data processing function.
2203 * It always succeeds. The buffer may be dropped during processing
2204 * for congestion control or by the protocol layers.
2205 *
2206 * This function may only be called from softirq context and interrupts
2207 * should be enabled.
2208 *
2209 * Return values (usually ignored):
2210 * NET_RX_SUCCESS: no congestion
2211 * NET_RX_DROP: packet was dropped
2212 */
2213 int netif_receive_skb(struct sk_buff *skb)
2214 {
2215 struct packet_type *ptype, *pt_prev;
2216 struct net_device *orig_dev;
2217 struct net_device *null_or_orig;
2218 int ret = NET_RX_DROP;
2219 __be16 type;
2220
2221 /* if we've gotten here through NAPI, check netpoll */
2222 if (netpoll_receive_skb(skb))
2223 return NET_RX_DROP;
2224
2225 if (!skb->tstamp.tv64)
2226 net_timestamp(skb);
2227
2228 if (!skb->iif)
2229 skb->iif = skb->dev->ifindex;
2230
2231 null_or_orig = NULL;
2232 orig_dev = skb->dev;
2233 if (orig_dev->master) {
2234 if (skb_bond_should_drop(skb))
2235 null_or_orig = orig_dev; /* deliver only exact match */
2236 else
2237 skb->dev = orig_dev->master;
2238 }
2239
2240 __get_cpu_var(netdev_rx_stat).total++;
2241
2242 skb_reset_network_header(skb);
2243 skb_reset_transport_header(skb);
2244 skb->mac_len = skb->network_header - skb->mac_header;
2245
2246 pt_prev = NULL;
2247
2248 rcu_read_lock();
2249
2250 /* Don't receive packets in an exiting network namespace */
2251 if (!net_alive(dev_net(skb->dev)))
2252 goto out;
2253
2254 #ifdef CONFIG_NET_CLS_ACT
2255 if (skb->tc_verd & TC_NCLS) {
2256 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2257 goto ncls;
2258 }
2259 #endif
2260
2261 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2262 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2263 ptype->dev == orig_dev) {
2264 if (pt_prev)
2265 ret = deliver_skb(skb, pt_prev, orig_dev);
2266 pt_prev = ptype;
2267 }
2268 }
2269
2270 #ifdef CONFIG_NET_CLS_ACT
2271 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2272 if (!skb)
2273 goto out;
2274 ncls:
2275 #endif
2276
2277 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2278 if (!skb)
2279 goto out;
2280 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2281 if (!skb)
2282 goto out;
2283
2284 type = skb->protocol;
2285 list_for_each_entry_rcu(ptype,
2286 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2287 if (ptype->type == type &&
2288 (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2289 ptype->dev == orig_dev)) {
2290 if (pt_prev)
2291 ret = deliver_skb(skb, pt_prev, orig_dev);
2292 pt_prev = ptype;
2293 }
2294 }
2295
2296 if (pt_prev) {
2297 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2298 } else {
2299 kfree_skb(skb);
2300 /* Jamal, now you will not able to escape explaining
2301 * me how you were going to use this. :-)
2302 */
2303 ret = NET_RX_DROP;
2304 }
2305
2306 out:
2307 rcu_read_unlock();
2308 return ret;
2309 }
2310
2311 /* Network device is going away, flush any packets still pending */
2312 static void flush_backlog(void *arg)
2313 {
2314 struct net_device *dev = arg;
2315 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2316 struct sk_buff *skb, *tmp;
2317
2318 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2319 if (skb->dev == dev) {
2320 __skb_unlink(skb, &queue->input_pkt_queue);
2321 kfree_skb(skb);
2322 }
2323 }
2324
2325 static int process_backlog(struct napi_struct *napi, int quota)
2326 {
2327 int work = 0;
2328 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2329 unsigned long start_time = jiffies;
2330
2331 napi->weight = weight_p;
2332 do {
2333 struct sk_buff *skb;
2334
2335 local_irq_disable();
2336 skb = __skb_dequeue(&queue->input_pkt_queue);
2337 if (!skb) {
2338 __napi_complete(napi);
2339 local_irq_enable();
2340 break;
2341 }
2342 local_irq_enable();
2343
2344 netif_receive_skb(skb);
2345 } while (++work < quota && jiffies == start_time);
2346
2347 return work;
2348 }
2349
2350 /**
2351 * __napi_schedule - schedule for receive
2352 * @n: entry to schedule
2353 *
2354 * The entry's receive function will be scheduled to run
2355 */
2356 void __napi_schedule(struct napi_struct *n)
2357 {
2358 unsigned long flags;
2359
2360 local_irq_save(flags);
2361 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2362 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2363 local_irq_restore(flags);
2364 }
2365 EXPORT_SYMBOL(__napi_schedule);
2366
2367
2368 static void net_rx_action(struct softirq_action *h)
2369 {
2370 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2371 unsigned long start_time = jiffies;
2372 int budget = netdev_budget;
2373 void *have;
2374
2375 local_irq_disable();
2376
2377 while (!list_empty(list)) {
2378 struct napi_struct *n;
2379 int work, weight;
2380
2381 /* If softirq window is exhuasted then punt.
2382 *
2383 * Note that this is a slight policy change from the
2384 * previous NAPI code, which would allow up to 2
2385 * jiffies to pass before breaking out. The test
2386 * used to be "jiffies - start_time > 1".
2387 */
2388 if (unlikely(budget <= 0 || jiffies != start_time))
2389 goto softnet_break;
2390
2391 local_irq_enable();
2392
2393 /* Even though interrupts have been re-enabled, this
2394 * access is safe because interrupts can only add new
2395 * entries to the tail of this list, and only ->poll()
2396 * calls can remove this head entry from the list.
2397 */
2398 n = list_entry(list->next, struct napi_struct, poll_list);
2399
2400 have = netpoll_poll_lock(n);
2401
2402 weight = n->weight;
2403
2404 /* This NAPI_STATE_SCHED test is for avoiding a race
2405 * with netpoll's poll_napi(). Only the entity which
2406 * obtains the lock and sees NAPI_STATE_SCHED set will
2407 * actually make the ->poll() call. Therefore we avoid
2408 * accidently calling ->poll() when NAPI is not scheduled.
2409 */
2410 work = 0;
2411 if (test_bit(NAPI_STATE_SCHED, &n->state))
2412 work = n->poll(n, weight);
2413
2414 WARN_ON_ONCE(work > weight);
2415
2416 budget -= work;
2417
2418 local_irq_disable();
2419
2420 /* Drivers must not modify the NAPI state if they
2421 * consume the entire weight. In such cases this code
2422 * still "owns" the NAPI instance and therefore can
2423 * move the instance around on the list at-will.
2424 */
2425 if (unlikely(work == weight)) {
2426 if (unlikely(napi_disable_pending(n)))
2427 __napi_complete(n);
2428 else
2429 list_move_tail(&n->poll_list, list);
2430 }
2431
2432 netpoll_poll_unlock(have);
2433 }
2434 out:
2435 local_irq_enable();
2436
2437 #ifdef CONFIG_NET_DMA
2438 /*
2439 * There may not be any more sk_buffs coming right now, so push
2440 * any pending DMA copies to hardware
2441 */
2442 if (!cpus_empty(net_dma.channel_mask)) {
2443 int chan_idx;
2444 for_each_cpu_mask_nr(chan_idx, net_dma.channel_mask) {
2445 struct dma_chan *chan = net_dma.channels[chan_idx];
2446 if (chan)
2447 dma_async_memcpy_issue_pending(chan);
2448 }
2449 }
2450 #endif
2451
2452 return;
2453
2454 softnet_break:
2455 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2456 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2457 goto out;
2458 }
2459
2460 static gifconf_func_t * gifconf_list [NPROTO];
2461
2462 /**
2463 * register_gifconf - register a SIOCGIF handler
2464 * @family: Address family
2465 * @gifconf: Function handler
2466 *
2467 * Register protocol dependent address dumping routines. The handler
2468 * that is passed must not be freed or reused until it has been replaced
2469 * by another handler.
2470 */
2471 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2472 {
2473 if (family >= NPROTO)
2474 return -EINVAL;
2475 gifconf_list[family] = gifconf;
2476 return 0;
2477 }
2478
2479
2480 /*
2481 * Map an interface index to its name (SIOCGIFNAME)
2482 */
2483
2484 /*
2485 * We need this ioctl for efficient implementation of the
2486 * if_indextoname() function required by the IPv6 API. Without
2487 * it, we would have to search all the interfaces to find a
2488 * match. --pb
2489 */
2490
2491 static int dev_ifname(struct net *net, struct ifreq __user *arg)
2492 {
2493 struct net_device *dev;
2494 struct ifreq ifr;
2495
2496 /*
2497 * Fetch the caller's info block.
2498 */
2499
2500 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2501 return -EFAULT;
2502
2503 read_lock(&dev_base_lock);
2504 dev = __dev_get_by_index(net, ifr.ifr_ifindex);
2505 if (!dev) {
2506 read_unlock(&dev_base_lock);
2507 return -ENODEV;
2508 }
2509
2510 strcpy(ifr.ifr_name, dev->name);
2511 read_unlock(&dev_base_lock);
2512
2513 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2514 return -EFAULT;
2515 return 0;
2516 }
2517
2518 /*
2519 * Perform a SIOCGIFCONF call. This structure will change
2520 * size eventually, and there is nothing I can do about it.
2521 * Thus we will need a 'compatibility mode'.
2522 */
2523
2524 static int dev_ifconf(struct net *net, char __user *arg)
2525 {
2526 struct ifconf ifc;
2527 struct net_device *dev;
2528 char __user *pos;
2529 int len;
2530 int total;
2531 int i;
2532
2533 /*
2534 * Fetch the caller's info block.
2535 */
2536
2537 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2538 return -EFAULT;
2539
2540 pos = ifc.ifc_buf;
2541 len = ifc.ifc_len;
2542
2543 /*
2544 * Loop over the interfaces, and write an info block for each.
2545 */
2546
2547 total = 0;
2548 for_each_netdev(net, dev) {
2549 for (i = 0; i < NPROTO; i++) {
2550 if (gifconf_list[i]) {
2551 int done;
2552 if (!pos)
2553 done = gifconf_list[i](dev, NULL, 0);
2554 else
2555 done = gifconf_list[i](dev, pos + total,
2556 len - total);
2557 if (done < 0)
2558 return -EFAULT;
2559 total += done;
2560 }
2561 }
2562 }
2563
2564 /*
2565 * All done. Write the updated control block back to the caller.
2566 */
2567 ifc.ifc_len = total;
2568
2569 /*
2570 * Both BSD and Solaris return 0 here, so we do too.
2571 */
2572 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2573 }
2574
2575 #ifdef CONFIG_PROC_FS
2576 /*
2577 * This is invoked by the /proc filesystem handler to display a device
2578 * in detail.
2579 */
2580 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2581 __acquires(dev_base_lock)
2582 {
2583 struct net *net = seq_file_net(seq);
2584 loff_t off;
2585 struct net_device *dev;
2586
2587 read_lock(&dev_base_lock);
2588 if (!*pos)
2589 return SEQ_START_TOKEN;
2590
2591 off = 1;
2592 for_each_netdev(net, dev)
2593 if (off++ == *pos)
2594 return dev;
2595
2596 return NULL;
2597 }
2598
2599 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2600 {
2601 struct net *net = seq_file_net(seq);
2602 ++*pos;
2603 return v == SEQ_START_TOKEN ?
2604 first_net_device(net) : next_net_device((struct net_device *)v);
2605 }
2606
2607 void dev_seq_stop(struct seq_file *seq, void *v)
2608 __releases(dev_base_lock)
2609 {
2610 read_unlock(&dev_base_lock);
2611 }
2612
2613 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2614 {
2615 struct net_device_stats *stats = dev->get_stats(dev);
2616
2617 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2618 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2619 dev->name, stats->rx_bytes, stats->rx_packets,
2620 stats->rx_errors,
2621 stats->rx_dropped + stats->rx_missed_errors,
2622 stats->rx_fifo_errors,
2623 stats->rx_length_errors + stats->rx_over_errors +
2624 stats->rx_crc_errors + stats->rx_frame_errors,
2625 stats->rx_compressed, stats->multicast,
2626 stats->tx_bytes, stats->tx_packets,
2627 stats->tx_errors, stats->tx_dropped,
2628 stats->tx_fifo_errors, stats->collisions,
2629 stats->tx_carrier_errors +
2630 stats->tx_aborted_errors +
2631 stats->tx_window_errors +
2632 stats->tx_heartbeat_errors,
2633 stats->tx_compressed);
2634 }
2635
2636 /*
2637 * Called from the PROCfs module. This now uses the new arbitrary sized
2638 * /proc/net interface to create /proc/net/dev
2639 */
2640 static int dev_seq_show(struct seq_file *seq, void *v)
2641 {
2642 if (v == SEQ_START_TOKEN)
2643 seq_puts(seq, "Inter-| Receive "
2644 " | Transmit\n"
2645 " face |bytes packets errs drop fifo frame "
2646 "compressed multicast|bytes packets errs "
2647 "drop fifo colls carrier compressed\n");
2648 else
2649 dev_seq_printf_stats(seq, v);
2650 return 0;
2651 }
2652
2653 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2654 {
2655 struct netif_rx_stats *rc = NULL;
2656
2657 while (*pos < nr_cpu_ids)
2658 if (cpu_online(*pos)) {
2659 rc = &per_cpu(netdev_rx_stat, *pos);
2660 break;
2661 } else
2662 ++*pos;
2663 return rc;
2664 }
2665
2666 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2667 {
2668 return softnet_get_online(pos);
2669 }
2670
2671 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2672 {
2673 ++*pos;
2674 return softnet_get_online(pos);
2675 }
2676
2677 static void softnet_seq_stop(struct seq_file *seq, void *v)
2678 {
2679 }
2680
2681 static int softnet_seq_show(struct seq_file *seq, void *v)
2682 {
2683 struct netif_rx_stats *s = v;
2684
2685 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2686 s->total, s->dropped, s->time_squeeze, 0,
2687 0, 0, 0, 0, /* was fastroute */
2688 s->cpu_collision );
2689 return 0;
2690 }
2691
2692 static const struct seq_operations dev_seq_ops = {
2693 .start = dev_seq_start,
2694 .next = dev_seq_next,
2695 .stop = dev_seq_stop,
2696 .show = dev_seq_show,
2697 };
2698
2699 static int dev_seq_open(struct inode *inode, struct file *file)
2700 {
2701 return seq_open_net(inode, file, &dev_seq_ops,
2702 sizeof(struct seq_net_private));
2703 }
2704
2705 static const struct file_operations dev_seq_fops = {
2706 .owner = THIS_MODULE,
2707 .open = dev_seq_open,
2708 .read = seq_read,
2709 .llseek = seq_lseek,
2710 .release = seq_release_net,
2711 };
2712
2713 static const struct seq_operations softnet_seq_ops = {
2714 .start = softnet_seq_start,
2715 .next = softnet_seq_next,
2716 .stop = softnet_seq_stop,
2717 .show = softnet_seq_show,
2718 };
2719
2720 static int softnet_seq_open(struct inode *inode, struct file *file)
2721 {
2722 return seq_open(file, &softnet_seq_ops);
2723 }
2724
2725 static const struct file_operations softnet_seq_fops = {
2726 .owner = THIS_MODULE,
2727 .open = softnet_seq_open,
2728 .read = seq_read,
2729 .llseek = seq_lseek,
2730 .release = seq_release,
2731 };
2732
2733 static void *ptype_get_idx(loff_t pos)
2734 {
2735 struct packet_type *pt = NULL;
2736 loff_t i = 0;
2737 int t;
2738
2739 list_for_each_entry_rcu(pt, &ptype_all, list) {
2740 if (i == pos)
2741 return pt;
2742 ++i;
2743 }
2744
2745 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
2746 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
2747 if (i == pos)
2748 return pt;
2749 ++i;
2750 }
2751 }
2752 return NULL;
2753 }
2754
2755 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
2756 __acquires(RCU)
2757 {
2758 rcu_read_lock();
2759 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
2760 }
2761
2762 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2763 {
2764 struct packet_type *pt;
2765 struct list_head *nxt;
2766 int hash;
2767
2768 ++*pos;
2769 if (v == SEQ_START_TOKEN)
2770 return ptype_get_idx(0);
2771
2772 pt = v;
2773 nxt = pt->list.next;
2774 if (pt->type == htons(ETH_P_ALL)) {
2775 if (nxt != &ptype_all)
2776 goto found;
2777 hash = 0;
2778 nxt = ptype_base[0].next;
2779 } else
2780 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
2781
2782 while (nxt == &ptype_base[hash]) {
2783 if (++hash >= PTYPE_HASH_SIZE)
2784 return NULL;
2785 nxt = ptype_base[hash].next;
2786 }
2787 found:
2788 return list_entry(nxt, struct packet_type, list);
2789 }
2790
2791 static void ptype_seq_stop(struct seq_file *seq, void *v)
2792 __releases(RCU)
2793 {
2794 rcu_read_unlock();
2795 }
2796
2797 static void ptype_seq_decode(struct seq_file *seq, void *sym)
2798 {
2799 #ifdef CONFIG_KALLSYMS
2800 unsigned long offset = 0, symsize;
2801 const char *symname;
2802 char *modname;
2803 char namebuf[128];
2804
2805 symname = kallsyms_lookup((unsigned long)sym, &symsize, &offset,
2806 &modname, namebuf);
2807
2808 if (symname) {
2809 char *delim = ":";
2810
2811 if (!modname)
2812 modname = delim = "";
2813 seq_printf(seq, "%s%s%s%s+0x%lx", delim, modname, delim,
2814 symname, offset);
2815 return;
2816 }
2817 #endif
2818
2819 seq_printf(seq, "[%p]", sym);
2820 }
2821
2822 static int ptype_seq_show(struct seq_file *seq, void *v)
2823 {
2824 struct packet_type *pt = v;
2825
2826 if (v == SEQ_START_TOKEN)
2827 seq_puts(seq, "Type Device Function\n");
2828 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
2829 if (pt->type == htons(ETH_P_ALL))
2830 seq_puts(seq, "ALL ");
2831 else
2832 seq_printf(seq, "%04x", ntohs(pt->type));
2833
2834 seq_printf(seq, " %-8s ",
2835 pt->dev ? pt->dev->name : "");
2836 ptype_seq_decode(seq, pt->func);
2837 seq_putc(seq, '\n');
2838 }
2839
2840 return 0;
2841 }
2842
2843 static const struct seq_operations ptype_seq_ops = {
2844 .start = ptype_seq_start,
2845 .next = ptype_seq_next,
2846 .stop = ptype_seq_stop,
2847 .show = ptype_seq_show,
2848 };
2849
2850 static int ptype_seq_open(struct inode *inode, struct file *file)
2851 {
2852 return seq_open_net(inode, file, &ptype_seq_ops,
2853 sizeof(struct seq_net_private));
2854 }
2855
2856 static const struct file_operations ptype_seq_fops = {
2857 .owner = THIS_MODULE,
2858 .open = ptype_seq_open,
2859 .read = seq_read,
2860 .llseek = seq_lseek,
2861 .release = seq_release_net,
2862 };
2863
2864
2865 static int __net_init dev_proc_net_init(struct net *net)
2866 {
2867 int rc = -ENOMEM;
2868
2869 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
2870 goto out;
2871 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
2872 goto out_dev;
2873 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
2874 goto out_softnet;
2875
2876 if (wext_proc_init(net))
2877 goto out_ptype;
2878 rc = 0;
2879 out:
2880 return rc;
2881 out_ptype:
2882 proc_net_remove(net, "ptype");
2883 out_softnet:
2884 proc_net_remove(net, "softnet_stat");
2885 out_dev:
2886 proc_net_remove(net, "dev");
2887 goto out;
2888 }
2889
2890 static void __net_exit dev_proc_net_exit(struct net *net)
2891 {
2892 wext_proc_exit(net);
2893
2894 proc_net_remove(net, "ptype");
2895 proc_net_remove(net, "softnet_stat");
2896 proc_net_remove(net, "dev");
2897 }
2898
2899 static struct pernet_operations __net_initdata dev_proc_ops = {
2900 .init = dev_proc_net_init,
2901 .exit = dev_proc_net_exit,
2902 };
2903
2904 static int __init dev_proc_init(void)
2905 {
2906 return register_pernet_subsys(&dev_proc_ops);
2907 }
2908 #else
2909 #define dev_proc_init() 0
2910 #endif /* CONFIG_PROC_FS */
2911
2912
2913 /**
2914 * netdev_set_master - set up master/slave pair
2915 * @slave: slave device
2916 * @master: new master device
2917 *
2918 * Changes the master device of the slave. Pass %NULL to break the
2919 * bonding. The caller must hold the RTNL semaphore. On a failure
2920 * a negative errno code is returned. On success the reference counts
2921 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2922 * function returns zero.
2923 */
2924 int netdev_set_master(struct net_device *slave, struct net_device *master)
2925 {
2926 struct net_device *old = slave->master;
2927
2928 ASSERT_RTNL();
2929
2930 if (master) {
2931 if (old)
2932 return -EBUSY;
2933 dev_hold(master);
2934 }
2935
2936 slave->master = master;
2937
2938 synchronize_net();
2939
2940 if (old)
2941 dev_put(old);
2942
2943 if (master)
2944 slave->flags |= IFF_SLAVE;
2945 else
2946 slave->flags &= ~IFF_SLAVE;
2947
2948 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2949 return 0;
2950 }
2951
2952 static void dev_change_rx_flags(struct net_device *dev, int flags)
2953 {
2954 if (dev->flags & IFF_UP && dev->change_rx_flags)
2955 dev->change_rx_flags(dev, flags);
2956 }
2957
2958 static int __dev_set_promiscuity(struct net_device *dev, int inc)
2959 {
2960 unsigned short old_flags = dev->flags;
2961 uid_t uid;
2962 gid_t gid;
2963
2964 ASSERT_RTNL();
2965
2966 dev->flags |= IFF_PROMISC;
2967 dev->promiscuity += inc;
2968 if (dev->promiscuity == 0) {
2969 /*
2970 * Avoid overflow.
2971 * If inc causes overflow, untouch promisc and return error.
2972 */
2973 if (inc < 0)
2974 dev->flags &= ~IFF_PROMISC;
2975 else {
2976 dev->promiscuity -= inc;
2977 printk(KERN_WARNING "%s: promiscuity touches roof, "
2978 "set promiscuity failed, promiscuity feature "
2979 "of device might be broken.\n", dev->name);
2980 return -EOVERFLOW;
2981 }
2982 }
2983 if (dev->flags != old_flags) {
2984 printk(KERN_INFO "device %s %s promiscuous mode\n",
2985 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2986 "left");
2987 if (audit_enabled) {
2988 current_uid_gid(&uid, &gid);
2989 audit_log(current->audit_context, GFP_ATOMIC,
2990 AUDIT_ANOM_PROMISCUOUS,
2991 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
2992 dev->name, (dev->flags & IFF_PROMISC),
2993 (old_flags & IFF_PROMISC),
2994 audit_get_loginuid(current),
2995 uid, gid,
2996 audit_get_sessionid(current));
2997 }
2998
2999 dev_change_rx_flags(dev, IFF_PROMISC);
3000 }
3001 return 0;
3002 }
3003
3004 /**
3005 * dev_set_promiscuity - update promiscuity count on a device
3006 * @dev: device
3007 * @inc: modifier
3008 *
3009 * Add or remove promiscuity from a device. While the count in the device
3010 * remains above zero the interface remains promiscuous. Once it hits zero
3011 * the device reverts back to normal filtering operation. A negative inc
3012 * value is used to drop promiscuity on the device.
3013 * Return 0 if successful or a negative errno code on error.
3014 */
3015 int dev_set_promiscuity(struct net_device *dev, int inc)
3016 {
3017 unsigned short old_flags = dev->flags;
3018 int err;
3019
3020 err = __dev_set_promiscuity(dev, inc);
3021 if (err < 0)
3022 return err;
3023 if (dev->flags != old_flags)
3024 dev_set_rx_mode(dev);
3025 return err;
3026 }
3027
3028 /**
3029 * dev_set_allmulti - update allmulti count on a device
3030 * @dev: device
3031 * @inc: modifier
3032 *
3033 * Add or remove reception of all multicast frames to a device. While the
3034 * count in the device remains above zero the interface remains listening
3035 * to all interfaces. Once it hits zero the device reverts back to normal
3036 * filtering operation. A negative @inc value is used to drop the counter
3037 * when releasing a resource needing all multicasts.
3038 * Return 0 if successful or a negative errno code on error.
3039 */
3040
3041 int dev_set_allmulti(struct net_device *dev, int inc)
3042 {
3043 unsigned short old_flags = dev->flags;
3044
3045 ASSERT_RTNL();
3046
3047 dev->flags |= IFF_ALLMULTI;
3048 dev->allmulti += inc;
3049 if (dev->allmulti == 0) {
3050 /*
3051 * Avoid overflow.
3052 * If inc causes overflow, untouch allmulti and return error.
3053 */
3054 if (inc < 0)
3055 dev->flags &= ~IFF_ALLMULTI;
3056 else {
3057 dev->allmulti -= inc;
3058 printk(KERN_WARNING "%s: allmulti touches roof, "
3059 "set allmulti failed, allmulti feature of "
3060 "device might be broken.\n", dev->name);
3061 return -EOVERFLOW;
3062 }
3063 }
3064 if (dev->flags ^ old_flags) {
3065 dev_change_rx_flags(dev, IFF_ALLMULTI);
3066 dev_set_rx_mode(dev);
3067 }
3068 return 0;
3069 }
3070
3071 /*
3072 * Upload unicast and multicast address lists to device and
3073 * configure RX filtering. When the device doesn't support unicast
3074 * filtering it is put in promiscuous mode while unicast addresses
3075 * are present.
3076 */
3077 void __dev_set_rx_mode(struct net_device *dev)
3078 {
3079 /* dev_open will call this function so the list will stay sane. */
3080 if (!(dev->flags&IFF_UP))
3081 return;
3082
3083 if (!netif_device_present(dev))
3084 return;
3085
3086 if (dev->set_rx_mode)
3087 dev->set_rx_mode(dev);
3088 else {
3089 /* Unicast addresses changes may only happen under the rtnl,
3090 * therefore calling __dev_set_promiscuity here is safe.
3091 */
3092 if (dev->uc_count > 0 && !dev->uc_promisc) {
3093 __dev_set_promiscuity(dev, 1);
3094 dev->uc_promisc = 1;
3095 } else if (dev->uc_count == 0 && dev->uc_promisc) {
3096 __dev_set_promiscuity(dev, -1);
3097 dev->uc_promisc = 0;
3098 }
3099
3100 if (dev->set_multicast_list)
3101 dev->set_multicast_list(dev);
3102 }
3103 }
3104
3105 void dev_set_rx_mode(struct net_device *dev)
3106 {
3107 netif_addr_lock_bh(dev);
3108 __dev_set_rx_mode(dev);
3109 netif_addr_unlock_bh(dev);
3110 }
3111
3112 int __dev_addr_delete(struct dev_addr_list **list, int *count,
3113 void *addr, int alen, int glbl)
3114 {
3115 struct dev_addr_list *da;
3116
3117 for (; (da = *list) != NULL; list = &da->next) {
3118 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3119 alen == da->da_addrlen) {
3120 if (glbl) {
3121 int old_glbl = da->da_gusers;
3122 da->da_gusers = 0;
3123 if (old_glbl == 0)
3124 break;
3125 }
3126 if (--da->da_users)
3127 return 0;
3128
3129 *list = da->next;
3130 kfree(da);
3131 (*count)--;
3132 return 0;
3133 }
3134 }
3135 return -ENOENT;
3136 }
3137
3138 int __dev_addr_add(struct dev_addr_list **list, int *count,
3139 void *addr, int alen, int glbl)
3140 {
3141 struct dev_addr_list *da;
3142
3143 for (da = *list; da != NULL; da = da->next) {
3144 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3145 da->da_addrlen == alen) {
3146 if (glbl) {
3147 int old_glbl = da->da_gusers;
3148 da->da_gusers = 1;
3149 if (old_glbl)
3150 return 0;
3151 }
3152 da->da_users++;
3153 return 0;
3154 }
3155 }
3156
3157 da = kzalloc(sizeof(*da), GFP_ATOMIC);
3158 if (da == NULL)
3159 return -ENOMEM;
3160 memcpy(da->da_addr, addr, alen);
3161 da->da_addrlen = alen;
3162 da->da_users = 1;
3163 da->da_gusers = glbl ? 1 : 0;
3164 da->next = *list;
3165 *list = da;
3166 (*count)++;
3167 return 0;
3168 }
3169
3170 /**
3171 * dev_unicast_delete - Release secondary unicast address.
3172 * @dev: device
3173 * @addr: address to delete
3174 * @alen: length of @addr
3175 *
3176 * Release reference to a secondary unicast address and remove it
3177 * from the device if the reference count drops to zero.
3178 *
3179 * The caller must hold the rtnl_mutex.
3180 */
3181 int dev_unicast_delete(struct net_device *dev, void *addr, int alen)
3182 {
3183 int err;
3184
3185 ASSERT_RTNL();
3186
3187 netif_addr_lock_bh(dev);
3188 err = __dev_addr_delete(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3189 if (!err)
3190 __dev_set_rx_mode(dev);
3191 netif_addr_unlock_bh(dev);
3192 return err;
3193 }
3194 EXPORT_SYMBOL(dev_unicast_delete);
3195
3196 /**
3197 * dev_unicast_add - add a secondary unicast address
3198 * @dev: device
3199 * @addr: address to add
3200 * @alen: length of @addr
3201 *
3202 * Add a secondary unicast address to the device or increase
3203 * the reference count if it already exists.
3204 *
3205 * The caller must hold the rtnl_mutex.
3206 */
3207 int dev_unicast_add(struct net_device *dev, void *addr, int alen)
3208 {
3209 int err;
3210
3211 ASSERT_RTNL();
3212
3213 netif_addr_lock_bh(dev);
3214 err = __dev_addr_add(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3215 if (!err)
3216 __dev_set_rx_mode(dev);
3217 netif_addr_unlock_bh(dev);
3218 return err;
3219 }
3220 EXPORT_SYMBOL(dev_unicast_add);
3221
3222 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
3223 struct dev_addr_list **from, int *from_count)
3224 {
3225 struct dev_addr_list *da, *next;
3226 int err = 0;
3227
3228 da = *from;
3229 while (da != NULL) {
3230 next = da->next;
3231 if (!da->da_synced) {
3232 err = __dev_addr_add(to, to_count,
3233 da->da_addr, da->da_addrlen, 0);
3234 if (err < 0)
3235 break;
3236 da->da_synced = 1;
3237 da->da_users++;
3238 } else if (da->da_users == 1) {
3239 __dev_addr_delete(to, to_count,
3240 da->da_addr, da->da_addrlen, 0);
3241 __dev_addr_delete(from, from_count,
3242 da->da_addr, da->da_addrlen, 0);
3243 }
3244 da = next;
3245 }
3246 return err;
3247 }
3248
3249 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
3250 struct dev_addr_list **from, int *from_count)
3251 {
3252 struct dev_addr_list *da, *next;
3253
3254 da = *from;
3255 while (da != NULL) {
3256 next = da->next;
3257 if (da->da_synced) {
3258 __dev_addr_delete(to, to_count,
3259 da->da_addr, da->da_addrlen, 0);
3260 da->da_synced = 0;
3261 __dev_addr_delete(from, from_count,
3262 da->da_addr, da->da_addrlen, 0);
3263 }
3264 da = next;
3265 }
3266 }
3267
3268 /**
3269 * dev_unicast_sync - Synchronize device's unicast list to another device
3270 * @to: destination device
3271 * @from: source device
3272 *
3273 * Add newly added addresses to the destination device and release
3274 * addresses that have no users left. The source device must be
3275 * locked by netif_tx_lock_bh.
3276 *
3277 * This function is intended to be called from the dev->set_rx_mode
3278 * function of layered software devices.
3279 */
3280 int dev_unicast_sync(struct net_device *to, struct net_device *from)
3281 {
3282 int err = 0;
3283
3284 netif_addr_lock_bh(to);
3285 err = __dev_addr_sync(&to->uc_list, &to->uc_count,
3286 &from->uc_list, &from->uc_count);
3287 if (!err)
3288 __dev_set_rx_mode(to);
3289 netif_addr_unlock_bh(to);
3290 return err;
3291 }
3292 EXPORT_SYMBOL(dev_unicast_sync);
3293
3294 /**
3295 * dev_unicast_unsync - Remove synchronized addresses from the destination device
3296 * @to: destination device
3297 * @from: source device
3298 *
3299 * Remove all addresses that were added to the destination device by
3300 * dev_unicast_sync(). This function is intended to be called from the
3301 * dev->stop function of layered software devices.
3302 */
3303 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
3304 {
3305 netif_addr_lock_bh(from);
3306 netif_addr_lock(to);
3307
3308 __dev_addr_unsync(&to->uc_list, &to->uc_count,
3309 &from->uc_list, &from->uc_count);
3310 __dev_set_rx_mode(to);
3311
3312 netif_addr_unlock(to);
3313 netif_addr_unlock_bh(from);
3314 }
3315 EXPORT_SYMBOL(dev_unicast_unsync);
3316
3317 static void __dev_addr_discard(struct dev_addr_list **list)
3318 {
3319 struct dev_addr_list *tmp;
3320
3321 while (*list != NULL) {
3322 tmp = *list;
3323 *list = tmp->next;
3324 if (tmp->da_users > tmp->da_gusers)
3325 printk("__dev_addr_discard: address leakage! "
3326 "da_users=%d\n", tmp->da_users);
3327 kfree(tmp);
3328 }
3329 }
3330
3331 static void dev_addr_discard(struct net_device *dev)
3332 {
3333 netif_addr_lock_bh(dev);
3334
3335 __dev_addr_discard(&dev->uc_list);
3336 dev->uc_count = 0;
3337
3338 __dev_addr_discard(&dev->mc_list);
3339 dev->mc_count = 0;
3340
3341 netif_addr_unlock_bh(dev);
3342 }
3343
3344 /**
3345 * dev_get_flags - get flags reported to userspace
3346 * @dev: device
3347 *
3348 * Get the combination of flag bits exported through APIs to userspace.
3349 */
3350 unsigned dev_get_flags(const struct net_device *dev)
3351 {
3352 unsigned flags;
3353
3354 flags = (dev->flags & ~(IFF_PROMISC |
3355 IFF_ALLMULTI |
3356 IFF_RUNNING |
3357 IFF_LOWER_UP |
3358 IFF_DORMANT)) |
3359 (dev->gflags & (IFF_PROMISC |
3360 IFF_ALLMULTI));
3361
3362 if (netif_running(dev)) {
3363 if (netif_oper_up(dev))
3364 flags |= IFF_RUNNING;
3365 if (netif_carrier_ok(dev))
3366 flags |= IFF_LOWER_UP;
3367 if (netif_dormant(dev))
3368 flags |= IFF_DORMANT;
3369 }
3370
3371 return flags;
3372 }
3373
3374 /**
3375 * dev_change_flags - change device settings
3376 * @dev: device
3377 * @flags: device state flags
3378 *
3379 * Change settings on device based state flags. The flags are
3380 * in the userspace exported format.
3381 */
3382 int dev_change_flags(struct net_device *dev, unsigned flags)
3383 {
3384 int ret, changes;
3385 int old_flags = dev->flags;
3386
3387 ASSERT_RTNL();
3388
3389 /*
3390 * Set the flags on our device.
3391 */
3392
3393 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
3394 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
3395 IFF_AUTOMEDIA)) |
3396 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
3397 IFF_ALLMULTI));
3398
3399 /*
3400 * Load in the correct multicast list now the flags have changed.
3401 */
3402
3403 if ((old_flags ^ flags) & IFF_MULTICAST)
3404 dev_change_rx_flags(dev, IFF_MULTICAST);
3405
3406 dev_set_rx_mode(dev);
3407
3408 /*
3409 * Have we downed the interface. We handle IFF_UP ourselves
3410 * according to user attempts to set it, rather than blindly
3411 * setting it.
3412 */
3413
3414 ret = 0;
3415 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
3416 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
3417
3418 if (!ret)
3419 dev_set_rx_mode(dev);
3420 }
3421
3422 if (dev->flags & IFF_UP &&
3423 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
3424 IFF_VOLATILE)))
3425 call_netdevice_notifiers(NETDEV_CHANGE, dev);
3426
3427 if ((flags ^ dev->gflags) & IFF_PROMISC) {
3428 int inc = (flags & IFF_PROMISC) ? +1 : -1;
3429 dev->gflags ^= IFF_PROMISC;
3430 dev_set_promiscuity(dev, inc);
3431 }
3432
3433 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
3434 is important. Some (broken) drivers set IFF_PROMISC, when
3435 IFF_ALLMULTI is requested not asking us and not reporting.
3436 */
3437 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
3438 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
3439 dev->gflags ^= IFF_ALLMULTI;
3440 dev_set_allmulti(dev, inc);
3441 }
3442
3443 /* Exclude state transition flags, already notified */
3444 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
3445 if (changes)
3446 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
3447
3448 return ret;
3449 }
3450
3451 /**
3452 * dev_set_mtu - Change maximum transfer unit
3453 * @dev: device
3454 * @new_mtu: new transfer unit
3455 *
3456 * Change the maximum transfer size of the network device.
3457 */
3458 int dev_set_mtu(struct net_device *dev, int new_mtu)
3459 {
3460 int err;
3461
3462 if (new_mtu == dev->mtu)
3463 return 0;
3464
3465 /* MTU must be positive. */
3466 if (new_mtu < 0)
3467 return -EINVAL;
3468
3469 if (!netif_device_present(dev))
3470 return -ENODEV;
3471
3472 err = 0;
3473 if (dev->change_mtu)
3474 err = dev->change_mtu(dev, new_mtu);
3475 else
3476 dev->mtu = new_mtu;
3477 if (!err && dev->flags & IFF_UP)
3478 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
3479 return err;
3480 }
3481
3482 /**
3483 * dev_set_mac_address - Change Media Access Control Address
3484 * @dev: device
3485 * @sa: new address
3486 *
3487 * Change the hardware (MAC) address of the device
3488 */
3489 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
3490 {
3491 int err;
3492
3493 if (!dev->set_mac_address)
3494 return -EOPNOTSUPP;
3495 if (sa->sa_family != dev->type)
3496 return -EINVAL;
3497 if (!netif_device_present(dev))
3498 return -ENODEV;
3499 err = dev->set_mac_address(dev, sa);
3500 if (!err)
3501 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3502 return err;
3503 }
3504
3505 /*
3506 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
3507 */
3508 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
3509 {
3510 int err;
3511 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3512
3513 if (!dev)
3514 return -ENODEV;
3515
3516 switch (cmd) {
3517 case SIOCGIFFLAGS: /* Get interface flags */
3518 ifr->ifr_flags = dev_get_flags(dev);
3519 return 0;
3520
3521 case SIOCGIFMETRIC: /* Get the metric on the interface
3522 (currently unused) */
3523 ifr->ifr_metric = 0;
3524 return 0;
3525
3526 case SIOCGIFMTU: /* Get the MTU of a device */
3527 ifr->ifr_mtu = dev->mtu;
3528 return 0;
3529
3530 case SIOCGIFHWADDR:
3531 if (!dev->addr_len)
3532 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
3533 else
3534 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
3535 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3536 ifr->ifr_hwaddr.sa_family = dev->type;
3537 return 0;
3538
3539 case SIOCGIFSLAVE:
3540 err = -EINVAL;
3541 break;
3542
3543 case SIOCGIFMAP:
3544 ifr->ifr_map.mem_start = dev->mem_start;
3545 ifr->ifr_map.mem_end = dev->mem_end;
3546 ifr->ifr_map.base_addr = dev->base_addr;
3547 ifr->ifr_map.irq = dev->irq;
3548 ifr->ifr_map.dma = dev->dma;
3549 ifr->ifr_map.port = dev->if_port;
3550 return 0;
3551
3552 case SIOCGIFINDEX:
3553 ifr->ifr_ifindex = dev->ifindex;
3554 return 0;
3555
3556 case SIOCGIFTXQLEN:
3557 ifr->ifr_qlen = dev->tx_queue_len;
3558 return 0;
3559
3560 default:
3561 /* dev_ioctl() should ensure this case
3562 * is never reached
3563 */
3564 WARN_ON(1);
3565 err = -EINVAL;
3566 break;
3567
3568 }
3569 return err;
3570 }
3571
3572 /*
3573 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
3574 */
3575 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
3576 {
3577 int err;
3578 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3579
3580 if (!dev)
3581 return -ENODEV;
3582
3583 switch (cmd) {
3584 case SIOCSIFFLAGS: /* Set interface flags */
3585 return dev_change_flags(dev, ifr->ifr_flags);
3586
3587 case SIOCSIFMETRIC: /* Set the metric on the interface
3588 (currently unused) */
3589 return -EOPNOTSUPP;
3590
3591 case SIOCSIFMTU: /* Set the MTU of a device */
3592 return dev_set_mtu(dev, ifr->ifr_mtu);
3593
3594 case SIOCSIFHWADDR:
3595 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
3596
3597 case SIOCSIFHWBROADCAST:
3598 if (ifr->ifr_hwaddr.sa_family != dev->type)
3599 return -EINVAL;
3600 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
3601 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3602 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3603 return 0;
3604
3605 case SIOCSIFMAP:
3606 if (dev->set_config) {
3607 if (!netif_device_present(dev))
3608 return -ENODEV;
3609 return dev->set_config(dev, &ifr->ifr_map);
3610 }
3611 return -EOPNOTSUPP;
3612
3613 case SIOCADDMULTI:
3614 if ((!dev->set_multicast_list && !dev->set_rx_mode) ||
3615 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3616 return -EINVAL;
3617 if (!netif_device_present(dev))
3618 return -ENODEV;
3619 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
3620 dev->addr_len, 1);
3621
3622 case SIOCDELMULTI:
3623 if ((!dev->set_multicast_list && !dev->set_rx_mode) ||
3624 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3625 return -EINVAL;
3626 if (!netif_device_present(dev))
3627 return -ENODEV;
3628 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
3629 dev->addr_len, 1);
3630
3631 case SIOCSIFTXQLEN:
3632 if (ifr->ifr_qlen < 0)
3633 return -EINVAL;
3634 dev->tx_queue_len = ifr->ifr_qlen;
3635 return 0;
3636
3637 case SIOCSIFNAME:
3638 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
3639 return dev_change_name(dev, ifr->ifr_newname);
3640
3641 /*
3642 * Unknown or private ioctl
3643 */
3644
3645 default:
3646 if ((cmd >= SIOCDEVPRIVATE &&
3647 cmd <= SIOCDEVPRIVATE + 15) ||
3648 cmd == SIOCBONDENSLAVE ||
3649 cmd == SIOCBONDRELEASE ||
3650 cmd == SIOCBONDSETHWADDR ||
3651 cmd == SIOCBONDSLAVEINFOQUERY ||
3652 cmd == SIOCBONDINFOQUERY ||
3653 cmd == SIOCBONDCHANGEACTIVE ||
3654 cmd == SIOCGMIIPHY ||
3655 cmd == SIOCGMIIREG ||
3656 cmd == SIOCSMIIREG ||
3657 cmd == SIOCBRADDIF ||
3658 cmd == SIOCBRDELIF ||
3659 cmd == SIOCWANDEV) {
3660 err = -EOPNOTSUPP;
3661 if (dev->do_ioctl) {
3662 if (netif_device_present(dev))
3663 err = dev->do_ioctl(dev, ifr,
3664 cmd);
3665 else
3666 err = -ENODEV;
3667 }
3668 } else
3669 err = -EINVAL;
3670
3671 }
3672 return err;
3673 }
3674
3675 /*
3676 * This function handles all "interface"-type I/O control requests. The actual
3677 * 'doing' part of this is dev_ifsioc above.
3678 */
3679
3680 /**
3681 * dev_ioctl - network device ioctl
3682 * @net: the applicable net namespace
3683 * @cmd: command to issue
3684 * @arg: pointer to a struct ifreq in user space
3685 *
3686 * Issue ioctl functions to devices. This is normally called by the
3687 * user space syscall interfaces but can sometimes be useful for
3688 * other purposes. The return value is the return from the syscall if
3689 * positive or a negative errno code on error.
3690 */
3691
3692 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
3693 {
3694 struct ifreq ifr;
3695 int ret;
3696 char *colon;
3697
3698 /* One special case: SIOCGIFCONF takes ifconf argument
3699 and requires shared lock, because it sleeps writing
3700 to user space.
3701 */
3702
3703 if (cmd == SIOCGIFCONF) {
3704 rtnl_lock();
3705 ret = dev_ifconf(net, (char __user *) arg);
3706 rtnl_unlock();
3707 return ret;
3708 }
3709 if (cmd == SIOCGIFNAME)
3710 return dev_ifname(net, (struct ifreq __user *)arg);
3711
3712 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3713 return -EFAULT;
3714
3715 ifr.ifr_name[IFNAMSIZ-1] = 0;
3716
3717 colon = strchr(ifr.ifr_name, ':');
3718 if (colon)
3719 *colon = 0;
3720
3721 /*
3722 * See which interface the caller is talking about.
3723 */
3724
3725 switch (cmd) {
3726 /*
3727 * These ioctl calls:
3728 * - can be done by all.
3729 * - atomic and do not require locking.
3730 * - return a value
3731 */
3732 case SIOCGIFFLAGS:
3733 case SIOCGIFMETRIC:
3734 case SIOCGIFMTU:
3735 case SIOCGIFHWADDR:
3736 case SIOCGIFSLAVE:
3737 case SIOCGIFMAP:
3738 case SIOCGIFINDEX:
3739 case SIOCGIFTXQLEN:
3740 dev_load(net, ifr.ifr_name);
3741 read_lock(&dev_base_lock);
3742 ret = dev_ifsioc_locked(net, &ifr, cmd);
3743 read_unlock(&dev_base_lock);
3744 if (!ret) {
3745 if (colon)
3746 *colon = ':';
3747 if (copy_to_user(arg, &ifr,
3748 sizeof(struct ifreq)))
3749 ret = -EFAULT;
3750 }
3751 return ret;
3752
3753 case SIOCETHTOOL:
3754 dev_load(net, ifr.ifr_name);
3755 rtnl_lock();
3756 ret = dev_ethtool(net, &ifr);
3757 rtnl_unlock();
3758 if (!ret) {
3759 if (colon)
3760 *colon = ':';
3761 if (copy_to_user(arg, &ifr,
3762 sizeof(struct ifreq)))
3763 ret = -EFAULT;
3764 }
3765 return ret;
3766
3767 /*
3768 * These ioctl calls:
3769 * - require superuser power.
3770 * - require strict serialization.
3771 * - return a value
3772 */
3773 case SIOCGMIIPHY:
3774 case SIOCGMIIREG:
3775 case SIOCSIFNAME:
3776 if (!capable(CAP_NET_ADMIN))
3777 return -EPERM;
3778 dev_load(net, ifr.ifr_name);
3779 rtnl_lock();
3780 ret = dev_ifsioc(net, &ifr, cmd);
3781 rtnl_unlock();
3782 if (!ret) {
3783 if (colon)
3784 *colon = ':';
3785 if (copy_to_user(arg, &ifr,
3786 sizeof(struct ifreq)))
3787 ret = -EFAULT;
3788 }
3789 return ret;
3790
3791 /*
3792 * These ioctl calls:
3793 * - require superuser power.
3794 * - require strict serialization.
3795 * - do not return a value
3796 */
3797 case SIOCSIFFLAGS:
3798 case SIOCSIFMETRIC:
3799 case SIOCSIFMTU:
3800 case SIOCSIFMAP:
3801 case SIOCSIFHWADDR:
3802 case SIOCSIFSLAVE:
3803 case SIOCADDMULTI:
3804 case SIOCDELMULTI:
3805 case SIOCSIFHWBROADCAST:
3806 case SIOCSIFTXQLEN:
3807 case SIOCSMIIREG:
3808 case SIOCBONDENSLAVE:
3809 case SIOCBONDRELEASE:
3810 case SIOCBONDSETHWADDR:
3811 case SIOCBONDCHANGEACTIVE:
3812 case SIOCBRADDIF:
3813 case SIOCBRDELIF:
3814 if (!capable(CAP_NET_ADMIN))
3815 return -EPERM;
3816 /* fall through */
3817 case SIOCBONDSLAVEINFOQUERY:
3818 case SIOCBONDINFOQUERY:
3819 dev_load(net, ifr.ifr_name);
3820 rtnl_lock();
3821 ret = dev_ifsioc(net, &ifr, cmd);
3822 rtnl_unlock();
3823 return ret;
3824
3825 case SIOCGIFMEM:
3826 /* Get the per device memory space. We can add this but
3827 * currently do not support it */
3828 case SIOCSIFMEM:
3829 /* Set the per device memory buffer space.
3830 * Not applicable in our case */
3831 case SIOCSIFLINK:
3832 return -EINVAL;
3833
3834 /*
3835 * Unknown or private ioctl.
3836 */
3837 default:
3838 if (cmd == SIOCWANDEV ||
3839 (cmd >= SIOCDEVPRIVATE &&
3840 cmd <= SIOCDEVPRIVATE + 15)) {
3841 dev_load(net, ifr.ifr_name);
3842 rtnl_lock();
3843 ret = dev_ifsioc(net, &ifr, cmd);
3844 rtnl_unlock();
3845 if (!ret && copy_to_user(arg, &ifr,
3846 sizeof(struct ifreq)))
3847 ret = -EFAULT;
3848 return ret;
3849 }
3850 /* Take care of Wireless Extensions */
3851 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
3852 return wext_handle_ioctl(net, &ifr, cmd, arg);
3853 return -EINVAL;
3854 }
3855 }
3856
3857
3858 /**
3859 * dev_new_index - allocate an ifindex
3860 * @net: the applicable net namespace
3861 *
3862 * Returns a suitable unique value for a new device interface
3863 * number. The caller must hold the rtnl semaphore or the
3864 * dev_base_lock to be sure it remains unique.
3865 */
3866 static int dev_new_index(struct net *net)
3867 {
3868 static int ifindex;
3869 for (;;) {
3870 if (++ifindex <= 0)
3871 ifindex = 1;
3872 if (!__dev_get_by_index(net, ifindex))
3873 return ifindex;
3874 }
3875 }
3876
3877 /* Delayed registration/unregisteration */
3878 static LIST_HEAD(net_todo_list);
3879
3880 static void net_set_todo(struct net_device *dev)
3881 {
3882 list_add_tail(&dev->todo_list, &net_todo_list);
3883 }
3884
3885 static void rollback_registered(struct net_device *dev)
3886 {
3887 BUG_ON(dev_boot_phase);
3888 ASSERT_RTNL();
3889
3890 /* Some devices call without registering for initialization unwind. */
3891 if (dev->reg_state == NETREG_UNINITIALIZED) {
3892 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3893 "was registered\n", dev->name, dev);
3894
3895 WARN_ON(1);
3896 return;
3897 }
3898
3899 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3900
3901 /* If device is running, close it first. */
3902 dev_close(dev);
3903
3904 /* And unlink it from device chain. */
3905 unlist_netdevice(dev);
3906
3907 dev->reg_state = NETREG_UNREGISTERING;
3908
3909 synchronize_net();
3910
3911 /* Shutdown queueing discipline. */
3912 dev_shutdown(dev);
3913
3914
3915 /* Notify protocols, that we are about to destroy
3916 this device. They should clean all the things.
3917 */
3918 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
3919
3920 /*
3921 * Flush the unicast and multicast chains
3922 */
3923 dev_addr_discard(dev);
3924
3925 if (dev->uninit)
3926 dev->uninit(dev);
3927
3928 /* Notifier chain MUST detach us from master device. */
3929 WARN_ON(dev->master);
3930
3931 /* Remove entries from kobject tree */
3932 netdev_unregister_kobject(dev);
3933
3934 synchronize_net();
3935
3936 dev_put(dev);
3937 }
3938
3939 static void __netdev_init_queue_locks_one(struct net_device *dev,
3940 struct netdev_queue *dev_queue,
3941 void *_unused)
3942 {
3943 spin_lock_init(&dev_queue->_xmit_lock);
3944 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
3945 dev_queue->xmit_lock_owner = -1;
3946 }
3947
3948 static void netdev_init_queue_locks(struct net_device *dev)
3949 {
3950 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
3951 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
3952 }
3953
3954 unsigned long netdev_fix_features(unsigned long features, const char *name)
3955 {
3956 /* Fix illegal SG+CSUM combinations. */
3957 if ((features & NETIF_F_SG) &&
3958 !(features & NETIF_F_ALL_CSUM)) {
3959 if (name)
3960 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
3961 "checksum feature.\n", name);
3962 features &= ~NETIF_F_SG;
3963 }
3964
3965 /* TSO requires that SG is present as well. */
3966 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
3967 if (name)
3968 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
3969 "SG feature.\n", name);
3970 features &= ~NETIF_F_TSO;
3971 }
3972
3973 if (features & NETIF_F_UFO) {
3974 if (!(features & NETIF_F_GEN_CSUM)) {
3975 if (name)
3976 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
3977 "since no NETIF_F_HW_CSUM feature.\n",
3978 name);
3979 features &= ~NETIF_F_UFO;
3980 }
3981
3982 if (!(features & NETIF_F_SG)) {
3983 if (name)
3984 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
3985 "since no NETIF_F_SG feature.\n", name);
3986 features &= ~NETIF_F_UFO;
3987 }
3988 }
3989
3990 return features;
3991 }
3992 EXPORT_SYMBOL(netdev_fix_features);
3993
3994 /**
3995 * register_netdevice - register a network device
3996 * @dev: device to register
3997 *
3998 * Take a completed network device structure and add it to the kernel
3999 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4000 * chain. 0 is returned on success. A negative errno code is returned
4001 * on a failure to set up the device, or if the name is a duplicate.
4002 *
4003 * Callers must hold the rtnl semaphore. You may want
4004 * register_netdev() instead of this.
4005 *
4006 * BUGS:
4007 * The locking appears insufficient to guarantee two parallel registers
4008 * will not get the same name.
4009 */
4010
4011 int register_netdevice(struct net_device *dev)
4012 {
4013 struct hlist_head *head;
4014 struct hlist_node *p;
4015 int ret;
4016 struct net *net;
4017
4018 BUG_ON(dev_boot_phase);
4019 ASSERT_RTNL();
4020
4021 might_sleep();
4022
4023 /* When net_device's are persistent, this will be fatal. */
4024 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4025 BUG_ON(!dev_net(dev));
4026 net = dev_net(dev);
4027
4028 spin_lock_init(&dev->addr_list_lock);
4029 netdev_set_addr_lockdep_class(dev);
4030 netdev_init_queue_locks(dev);
4031
4032 dev->iflink = -1;
4033
4034 /* Init, if this function is available */
4035 if (dev->init) {
4036 ret = dev->init(dev);
4037 if (ret) {
4038 if (ret > 0)
4039 ret = -EIO;
4040 goto out;
4041 }
4042 }
4043
4044 if (!dev_valid_name(dev->name)) {
4045 ret = -EINVAL;
4046 goto err_uninit;
4047 }
4048
4049 dev->ifindex = dev_new_index(net);
4050 if (dev->iflink == -1)
4051 dev->iflink = dev->ifindex;
4052
4053 /* Check for existence of name */
4054 head = dev_name_hash(net, dev->name);
4055 hlist_for_each(p, head) {
4056 struct net_device *d
4057 = hlist_entry(p, struct net_device, name_hlist);
4058 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
4059 ret = -EEXIST;
4060 goto err_uninit;
4061 }
4062 }
4063
4064 /* Fix illegal checksum combinations */
4065 if ((dev->features & NETIF_F_HW_CSUM) &&
4066 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4067 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
4068 dev->name);
4069 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4070 }
4071
4072 if ((dev->features & NETIF_F_NO_CSUM) &&
4073 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4074 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
4075 dev->name);
4076 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
4077 }
4078
4079 dev->features = netdev_fix_features(dev->features, dev->name);
4080
4081 /* Enable software GSO if SG is supported. */
4082 if (dev->features & NETIF_F_SG)
4083 dev->features |= NETIF_F_GSO;
4084
4085 netdev_initialize_kobject(dev);
4086 ret = netdev_register_kobject(dev);
4087 if (ret)
4088 goto err_uninit;
4089 dev->reg_state = NETREG_REGISTERED;
4090
4091 /*
4092 * Default initial state at registry is that the
4093 * device is present.
4094 */
4095
4096 set_bit(__LINK_STATE_PRESENT, &dev->state);
4097
4098 dev_init_scheduler(dev);
4099 dev_hold(dev);
4100 list_netdevice(dev);
4101
4102 /* Notify protocols, that a new device appeared. */
4103 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
4104 ret = notifier_to_errno(ret);
4105 if (ret) {
4106 rollback_registered(dev);
4107 dev->reg_state = NETREG_UNREGISTERED;
4108 }
4109
4110 out:
4111 return ret;
4112
4113 err_uninit:
4114 if (dev->uninit)
4115 dev->uninit(dev);
4116 goto out;
4117 }
4118
4119 /**
4120 * register_netdev - register a network device
4121 * @dev: device to register
4122 *
4123 * Take a completed network device structure and add it to the kernel
4124 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4125 * chain. 0 is returned on success. A negative errno code is returned
4126 * on a failure to set up the device, or if the name is a duplicate.
4127 *
4128 * This is a wrapper around register_netdevice that takes the rtnl semaphore
4129 * and expands the device name if you passed a format string to
4130 * alloc_netdev.
4131 */
4132 int register_netdev(struct net_device *dev)
4133 {
4134 int err;
4135
4136 rtnl_lock();
4137
4138 /*
4139 * If the name is a format string the caller wants us to do a
4140 * name allocation.
4141 */
4142 if (strchr(dev->name, '%')) {
4143 err = dev_alloc_name(dev, dev->name);
4144 if (err < 0)
4145 goto out;
4146 }
4147
4148 err = register_netdevice(dev);
4149 out:
4150 rtnl_unlock();
4151 return err;
4152 }
4153 EXPORT_SYMBOL(register_netdev);
4154
4155 /*
4156 * netdev_wait_allrefs - wait until all references are gone.
4157 *
4158 * This is called when unregistering network devices.
4159 *
4160 * Any protocol or device that holds a reference should register
4161 * for netdevice notification, and cleanup and put back the
4162 * reference if they receive an UNREGISTER event.
4163 * We can get stuck here if buggy protocols don't correctly
4164 * call dev_put.
4165 */
4166 static void netdev_wait_allrefs(struct net_device *dev)
4167 {
4168 unsigned long rebroadcast_time, warning_time;
4169
4170 rebroadcast_time = warning_time = jiffies;
4171 while (atomic_read(&dev->refcnt) != 0) {
4172 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
4173 rtnl_lock();
4174
4175 /* Rebroadcast unregister notification */
4176 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4177
4178 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
4179 &dev->state)) {
4180 /* We must not have linkwatch events
4181 * pending on unregister. If this
4182 * happens, we simply run the queue
4183 * unscheduled, resulting in a noop
4184 * for this device.
4185 */
4186 linkwatch_run_queue();
4187 }
4188
4189 __rtnl_unlock();
4190
4191 rebroadcast_time = jiffies;
4192 }
4193
4194 msleep(250);
4195
4196 if (time_after(jiffies, warning_time + 10 * HZ)) {
4197 printk(KERN_EMERG "unregister_netdevice: "
4198 "waiting for %s to become free. Usage "
4199 "count = %d\n",
4200 dev->name, atomic_read(&dev->refcnt));
4201 warning_time = jiffies;
4202 }
4203 }
4204 }
4205
4206 /* The sequence is:
4207 *
4208 * rtnl_lock();
4209 * ...
4210 * register_netdevice(x1);
4211 * register_netdevice(x2);
4212 * ...
4213 * unregister_netdevice(y1);
4214 * unregister_netdevice(y2);
4215 * ...
4216 * rtnl_unlock();
4217 * free_netdev(y1);
4218 * free_netdev(y2);
4219 *
4220 * We are invoked by rtnl_unlock().
4221 * This allows us to deal with problems:
4222 * 1) We can delete sysfs objects which invoke hotplug
4223 * without deadlocking with linkwatch via keventd.
4224 * 2) Since we run with the RTNL semaphore not held, we can sleep
4225 * safely in order to wait for the netdev refcnt to drop to zero.
4226 *
4227 * We must not return until all unregister events added during
4228 * the interval the lock was held have been completed.
4229 */
4230 void netdev_run_todo(void)
4231 {
4232 struct list_head list;
4233
4234 /* Snapshot list, allow later requests */
4235 list_replace_init(&net_todo_list, &list);
4236
4237 __rtnl_unlock();
4238
4239 while (!list_empty(&list)) {
4240 struct net_device *dev
4241 = list_entry(list.next, struct net_device, todo_list);
4242 list_del(&dev->todo_list);
4243
4244 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
4245 printk(KERN_ERR "network todo '%s' but state %d\n",
4246 dev->name, dev->reg_state);
4247 dump_stack();
4248 continue;
4249 }
4250
4251 dev->reg_state = NETREG_UNREGISTERED;
4252
4253 on_each_cpu(flush_backlog, dev, 1);
4254
4255 netdev_wait_allrefs(dev);
4256
4257 /* paranoia */
4258 BUG_ON(atomic_read(&dev->refcnt));
4259 WARN_ON(dev->ip_ptr);
4260 WARN_ON(dev->ip6_ptr);
4261 WARN_ON(dev->dn_ptr);
4262
4263 if (dev->destructor)
4264 dev->destructor(dev);
4265
4266 /* Free network device */
4267 kobject_put(&dev->dev.kobj);
4268 }
4269 }
4270
4271 static struct net_device_stats *internal_stats(struct net_device *dev)
4272 {
4273 return &dev->stats;
4274 }
4275
4276 static void netdev_init_one_queue(struct net_device *dev,
4277 struct netdev_queue *queue,
4278 void *_unused)
4279 {
4280 queue->dev = dev;
4281 }
4282
4283 static void netdev_init_queues(struct net_device *dev)
4284 {
4285 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
4286 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
4287 spin_lock_init(&dev->tx_global_lock);
4288 }
4289
4290 /**
4291 * alloc_netdev_mq - allocate network device
4292 * @sizeof_priv: size of private data to allocate space for
4293 * @name: device name format string
4294 * @setup: callback to initialize device
4295 * @queue_count: the number of subqueues to allocate
4296 *
4297 * Allocates a struct net_device with private data area for driver use
4298 * and performs basic initialization. Also allocates subquue structs
4299 * for each queue on the device at the end of the netdevice.
4300 */
4301 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
4302 void (*setup)(struct net_device *), unsigned int queue_count)
4303 {
4304 struct netdev_queue *tx;
4305 struct net_device *dev;
4306 size_t alloc_size;
4307 void *p;
4308
4309 BUG_ON(strlen(name) >= sizeof(dev->name));
4310
4311 alloc_size = sizeof(struct net_device);
4312 if (sizeof_priv) {
4313 /* ensure 32-byte alignment of private area */
4314 alloc_size = (alloc_size + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
4315 alloc_size += sizeof_priv;
4316 }
4317 /* ensure 32-byte alignment of whole construct */
4318 alloc_size += NETDEV_ALIGN_CONST;
4319
4320 p = kzalloc(alloc_size, GFP_KERNEL);
4321 if (!p) {
4322 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
4323 return NULL;
4324 }
4325
4326 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
4327 if (!tx) {
4328 printk(KERN_ERR "alloc_netdev: Unable to allocate "
4329 "tx qdiscs.\n");
4330 kfree(p);
4331 return NULL;
4332 }
4333
4334 dev = (struct net_device *)
4335 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
4336 dev->padded = (char *)dev - (char *)p;
4337 dev_net_set(dev, &init_net);
4338
4339 dev->_tx = tx;
4340 dev->num_tx_queues = queue_count;
4341 dev->real_num_tx_queues = queue_count;
4342
4343 if (sizeof_priv) {
4344 dev->priv = ((char *)dev +
4345 ((sizeof(struct net_device) + NETDEV_ALIGN_CONST)
4346 & ~NETDEV_ALIGN_CONST));
4347 }
4348
4349 dev->gso_max_size = GSO_MAX_SIZE;
4350
4351 netdev_init_queues(dev);
4352
4353 dev->get_stats = internal_stats;
4354 netpoll_netdev_init(dev);
4355 setup(dev);
4356 strcpy(dev->name, name);
4357 return dev;
4358 }
4359 EXPORT_SYMBOL(alloc_netdev_mq);
4360
4361 /**
4362 * free_netdev - free network device
4363 * @dev: device
4364 *
4365 * This function does the last stage of destroying an allocated device
4366 * interface. The reference to the device object is released.
4367 * If this is the last reference then it will be freed.
4368 */
4369 void free_netdev(struct net_device *dev)
4370 {
4371 release_net(dev_net(dev));
4372
4373 kfree(dev->_tx);
4374
4375 /* Compatibility with error handling in drivers */
4376 if (dev->reg_state == NETREG_UNINITIALIZED) {
4377 kfree((char *)dev - dev->padded);
4378 return;
4379 }
4380
4381 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
4382 dev->reg_state = NETREG_RELEASED;
4383
4384 /* will free via device release */
4385 put_device(&dev->dev);
4386 }
4387
4388 /**
4389 * synchronize_net - Synchronize with packet receive processing
4390 *
4391 * Wait for packets currently being received to be done.
4392 * Does not block later packets from starting.
4393 */
4394 void synchronize_net(void)
4395 {
4396 might_sleep();
4397 synchronize_rcu();
4398 }
4399
4400 /**
4401 * unregister_netdevice - remove device from the kernel
4402 * @dev: device
4403 *
4404 * This function shuts down a device interface and removes it
4405 * from the kernel tables.
4406 *
4407 * Callers must hold the rtnl semaphore. You may want
4408 * unregister_netdev() instead of this.
4409 */
4410
4411 void unregister_netdevice(struct net_device *dev)
4412 {
4413 ASSERT_RTNL();
4414
4415 rollback_registered(dev);
4416 /* Finish processing unregister after unlock */
4417 net_set_todo(dev);
4418 }
4419
4420 /**
4421 * unregister_netdev - remove device from the kernel
4422 * @dev: device
4423 *
4424 * This function shuts down a device interface and removes it
4425 * from the kernel tables.
4426 *
4427 * This is just a wrapper for unregister_netdevice that takes
4428 * the rtnl semaphore. In general you want to use this and not
4429 * unregister_netdevice.
4430 */
4431 void unregister_netdev(struct net_device *dev)
4432 {
4433 rtnl_lock();
4434 unregister_netdevice(dev);
4435 rtnl_unlock();
4436 }
4437
4438 EXPORT_SYMBOL(unregister_netdev);
4439
4440 /**
4441 * dev_change_net_namespace - move device to different nethost namespace
4442 * @dev: device
4443 * @net: network namespace
4444 * @pat: If not NULL name pattern to try if the current device name
4445 * is already taken in the destination network namespace.
4446 *
4447 * This function shuts down a device interface and moves it
4448 * to a new network namespace. On success 0 is returned, on
4449 * a failure a netagive errno code is returned.
4450 *
4451 * Callers must hold the rtnl semaphore.
4452 */
4453
4454 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
4455 {
4456 char buf[IFNAMSIZ];
4457 const char *destname;
4458 int err;
4459
4460 ASSERT_RTNL();
4461
4462 /* Don't allow namespace local devices to be moved. */
4463 err = -EINVAL;
4464 if (dev->features & NETIF_F_NETNS_LOCAL)
4465 goto out;
4466
4467 /* Ensure the device has been registrered */
4468 err = -EINVAL;
4469 if (dev->reg_state != NETREG_REGISTERED)
4470 goto out;
4471
4472 /* Get out if there is nothing todo */
4473 err = 0;
4474 if (net_eq(dev_net(dev), net))
4475 goto out;
4476
4477 /* Pick the destination device name, and ensure
4478 * we can use it in the destination network namespace.
4479 */
4480 err = -EEXIST;
4481 destname = dev->name;
4482 if (__dev_get_by_name(net, destname)) {
4483 /* We get here if we can't use the current device name */
4484 if (!pat)
4485 goto out;
4486 if (!dev_valid_name(pat))
4487 goto out;
4488 if (strchr(pat, '%')) {
4489 if (__dev_alloc_name(net, pat, buf) < 0)
4490 goto out;
4491 destname = buf;
4492 } else
4493 destname = pat;
4494 if (__dev_get_by_name(net, destname))
4495 goto out;
4496 }
4497
4498 /*
4499 * And now a mini version of register_netdevice unregister_netdevice.
4500 */
4501
4502 /* If device is running close it first. */
4503 dev_close(dev);
4504
4505 /* And unlink it from device chain */
4506 err = -ENODEV;
4507 unlist_netdevice(dev);
4508
4509 synchronize_net();
4510
4511 /* Shutdown queueing discipline. */
4512 dev_shutdown(dev);
4513
4514 /* Notify protocols, that we are about to destroy
4515 this device. They should clean all the things.
4516 */
4517 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4518
4519 /*
4520 * Flush the unicast and multicast chains
4521 */
4522 dev_addr_discard(dev);
4523
4524 /* Actually switch the network namespace */
4525 dev_net_set(dev, net);
4526
4527 /* Assign the new device name */
4528 if (destname != dev->name)
4529 strcpy(dev->name, destname);
4530
4531 /* If there is an ifindex conflict assign a new one */
4532 if (__dev_get_by_index(net, dev->ifindex)) {
4533 int iflink = (dev->iflink == dev->ifindex);
4534 dev->ifindex = dev_new_index(net);
4535 if (iflink)
4536 dev->iflink = dev->ifindex;
4537 }
4538
4539 /* Fixup kobjects */
4540 netdev_unregister_kobject(dev);
4541 err = netdev_register_kobject(dev);
4542 WARN_ON(err);
4543
4544 /* Add the device back in the hashes */
4545 list_netdevice(dev);
4546
4547 /* Notify protocols, that a new device appeared. */
4548 call_netdevice_notifiers(NETDEV_REGISTER, dev);
4549
4550 synchronize_net();
4551 err = 0;
4552 out:
4553 return err;
4554 }
4555
4556 static int dev_cpu_callback(struct notifier_block *nfb,
4557 unsigned long action,
4558 void *ocpu)
4559 {
4560 struct sk_buff **list_skb;
4561 struct Qdisc **list_net;
4562 struct sk_buff *skb;
4563 unsigned int cpu, oldcpu = (unsigned long)ocpu;
4564 struct softnet_data *sd, *oldsd;
4565
4566 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
4567 return NOTIFY_OK;
4568
4569 local_irq_disable();
4570 cpu = smp_processor_id();
4571 sd = &per_cpu(softnet_data, cpu);
4572 oldsd = &per_cpu(softnet_data, oldcpu);
4573
4574 /* Find end of our completion_queue. */
4575 list_skb = &sd->completion_queue;
4576 while (*list_skb)
4577 list_skb = &(*list_skb)->next;
4578 /* Append completion queue from offline CPU. */
4579 *list_skb = oldsd->completion_queue;
4580 oldsd->completion_queue = NULL;
4581
4582 /* Find end of our output_queue. */
4583 list_net = &sd->output_queue;
4584 while (*list_net)
4585 list_net = &(*list_net)->next_sched;
4586 /* Append output queue from offline CPU. */
4587 *list_net = oldsd->output_queue;
4588 oldsd->output_queue = NULL;
4589
4590 raise_softirq_irqoff(NET_TX_SOFTIRQ);
4591 local_irq_enable();
4592
4593 /* Process offline CPU's input_pkt_queue */
4594 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
4595 netif_rx(skb);
4596
4597 return NOTIFY_OK;
4598 }
4599
4600 #ifdef CONFIG_NET_DMA
4601 /**
4602 * net_dma_rebalance - try to maintain one DMA channel per CPU
4603 * @net_dma: DMA client and associated data (lock, channels, channel_mask)
4604 *
4605 * This is called when the number of channels allocated to the net_dma client
4606 * changes. The net_dma client tries to have one DMA channel per CPU.
4607 */
4608
4609 static void net_dma_rebalance(struct net_dma *net_dma)
4610 {
4611 unsigned int cpu, i, n, chan_idx;
4612 struct dma_chan *chan;
4613
4614 if (cpus_empty(net_dma->channel_mask)) {
4615 for_each_online_cpu(cpu)
4616 rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL);
4617 return;
4618 }
4619
4620 i = 0;
4621 cpu = first_cpu(cpu_online_map);
4622
4623 for_each_cpu_mask_nr(chan_idx, net_dma->channel_mask) {
4624 chan = net_dma->channels[chan_idx];
4625
4626 n = ((num_online_cpus() / cpus_weight(net_dma->channel_mask))
4627 + (i < (num_online_cpus() %
4628 cpus_weight(net_dma->channel_mask)) ? 1 : 0));
4629
4630 while(n) {
4631 per_cpu(softnet_data, cpu).net_dma = chan;
4632 cpu = next_cpu(cpu, cpu_online_map);
4633 n--;
4634 }
4635 i++;
4636 }
4637 }
4638
4639 /**
4640 * netdev_dma_event - event callback for the net_dma_client
4641 * @client: should always be net_dma_client
4642 * @chan: DMA channel for the event
4643 * @state: DMA state to be handled
4644 */
4645 static enum dma_state_client
4646 netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
4647 enum dma_state state)
4648 {
4649 int i, found = 0, pos = -1;
4650 struct net_dma *net_dma =
4651 container_of(client, struct net_dma, client);
4652 enum dma_state_client ack = DMA_DUP; /* default: take no action */
4653
4654 spin_lock(&net_dma->lock);
4655 switch (state) {
4656 case DMA_RESOURCE_AVAILABLE:
4657 for (i = 0; i < nr_cpu_ids; i++)
4658 if (net_dma->channels[i] == chan) {
4659 found = 1;
4660 break;
4661 } else if (net_dma->channels[i] == NULL && pos < 0)
4662 pos = i;
4663
4664 if (!found && pos >= 0) {
4665 ack = DMA_ACK;
4666 net_dma->channels[pos] = chan;
4667 cpu_set(pos, net_dma->channel_mask);
4668 net_dma_rebalance(net_dma);
4669 }
4670 break;
4671 case DMA_RESOURCE_REMOVED:
4672 for (i = 0; i < nr_cpu_ids; i++)
4673 if (net_dma->channels[i] == chan) {
4674 found = 1;
4675 pos = i;
4676 break;
4677 }
4678
4679 if (found) {
4680 ack = DMA_ACK;
4681 cpu_clear(pos, net_dma->channel_mask);
4682 net_dma->channels[i] = NULL;
4683 net_dma_rebalance(net_dma);
4684 }
4685 break;
4686 default:
4687 break;
4688 }
4689 spin_unlock(&net_dma->lock);
4690
4691 return ack;
4692 }
4693
4694 /**
4695 * netdev_dma_register - register the networking subsystem as a DMA client
4696 */
4697 static int __init netdev_dma_register(void)
4698 {
4699 net_dma.channels = kzalloc(nr_cpu_ids * sizeof(struct net_dma),
4700 GFP_KERNEL);
4701 if (unlikely(!net_dma.channels)) {
4702 printk(KERN_NOTICE
4703 "netdev_dma: no memory for net_dma.channels\n");
4704 return -ENOMEM;
4705 }
4706 spin_lock_init(&net_dma.lock);
4707 dma_cap_set(DMA_MEMCPY, net_dma.client.cap_mask);
4708 dma_async_client_register(&net_dma.client);
4709 dma_async_client_chan_request(&net_dma.client);
4710 return 0;
4711 }
4712
4713 #else
4714 static int __init netdev_dma_register(void) { return -ENODEV; }
4715 #endif /* CONFIG_NET_DMA */
4716
4717 /**
4718 * netdev_increment_features - increment feature set by one
4719 * @all: current feature set
4720 * @one: new feature set
4721 * @mask: mask feature set
4722 *
4723 * Computes a new feature set after adding a device with feature set
4724 * @one to the master device with current feature set @all. Will not
4725 * enable anything that is off in @mask. Returns the new feature set.
4726 */
4727 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
4728 unsigned long mask)
4729 {
4730 /* If device needs checksumming, downgrade to it. */
4731 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
4732 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
4733 else if (mask & NETIF_F_ALL_CSUM) {
4734 /* If one device supports v4/v6 checksumming, set for all. */
4735 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
4736 !(all & NETIF_F_GEN_CSUM)) {
4737 all &= ~NETIF_F_ALL_CSUM;
4738 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
4739 }
4740
4741 /* If one device supports hw checksumming, set for all. */
4742 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
4743 all &= ~NETIF_F_ALL_CSUM;
4744 all |= NETIF_F_HW_CSUM;
4745 }
4746 }
4747
4748 one |= NETIF_F_ALL_CSUM;
4749
4750 one |= all & NETIF_F_ONE_FOR_ALL;
4751 all &= one | NETIF_F_LLTX | NETIF_F_GSO;
4752 all |= one & mask & NETIF_F_ONE_FOR_ALL;
4753
4754 return all;
4755 }
4756 EXPORT_SYMBOL(netdev_increment_features);
4757
4758 static struct hlist_head *netdev_create_hash(void)
4759 {
4760 int i;
4761 struct hlist_head *hash;
4762
4763 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
4764 if (hash != NULL)
4765 for (i = 0; i < NETDEV_HASHENTRIES; i++)
4766 INIT_HLIST_HEAD(&hash[i]);
4767
4768 return hash;
4769 }
4770
4771 /* Initialize per network namespace state */
4772 static int __net_init netdev_init(struct net *net)
4773 {
4774 INIT_LIST_HEAD(&net->dev_base_head);
4775
4776 net->dev_name_head = netdev_create_hash();
4777 if (net->dev_name_head == NULL)
4778 goto err_name;
4779
4780 net->dev_index_head = netdev_create_hash();
4781 if (net->dev_index_head == NULL)
4782 goto err_idx;
4783
4784 return 0;
4785
4786 err_idx:
4787 kfree(net->dev_name_head);
4788 err_name:
4789 return -ENOMEM;
4790 }
4791
4792 /**
4793 * netdev_drivername - network driver for the device
4794 * @dev: network device
4795 * @buffer: buffer for resulting name
4796 * @len: size of buffer
4797 *
4798 * Determine network driver for device.
4799 */
4800 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
4801 {
4802 const struct device_driver *driver;
4803 const struct device *parent;
4804
4805 if (len <= 0 || !buffer)
4806 return buffer;
4807 buffer[0] = 0;
4808
4809 parent = dev->dev.parent;
4810
4811 if (!parent)
4812 return buffer;
4813
4814 driver = parent->driver;
4815 if (driver && driver->name)
4816 strlcpy(buffer, driver->name, len);
4817 return buffer;
4818 }
4819
4820 static void __net_exit netdev_exit(struct net *net)
4821 {
4822 kfree(net->dev_name_head);
4823 kfree(net->dev_index_head);
4824 }
4825
4826 static struct pernet_operations __net_initdata netdev_net_ops = {
4827 .init = netdev_init,
4828 .exit = netdev_exit,
4829 };
4830
4831 static void __net_exit default_device_exit(struct net *net)
4832 {
4833 struct net_device *dev, *next;
4834 /*
4835 * Push all migratable of the network devices back to the
4836 * initial network namespace
4837 */
4838 rtnl_lock();
4839 for_each_netdev_safe(net, dev, next) {
4840 int err;
4841 char fb_name[IFNAMSIZ];
4842
4843 /* Ignore unmoveable devices (i.e. loopback) */
4844 if (dev->features & NETIF_F_NETNS_LOCAL)
4845 continue;
4846
4847 /* Push remaing network devices to init_net */
4848 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
4849 err = dev_change_net_namespace(dev, &init_net, fb_name);
4850 if (err) {
4851 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
4852 __func__, dev->name, err);
4853 BUG();
4854 }
4855 }
4856 rtnl_unlock();
4857 }
4858
4859 static struct pernet_operations __net_initdata default_device_ops = {
4860 .exit = default_device_exit,
4861 };
4862
4863 /*
4864 * Initialize the DEV module. At boot time this walks the device list and
4865 * unhooks any devices that fail to initialise (normally hardware not
4866 * present) and leaves us with a valid list of present and active devices.
4867 *
4868 */
4869
4870 /*
4871 * This is called single threaded during boot, so no need
4872 * to take the rtnl semaphore.
4873 */
4874 static int __init net_dev_init(void)
4875 {
4876 int i, rc = -ENOMEM;
4877
4878 BUG_ON(!dev_boot_phase);
4879
4880 if (dev_proc_init())
4881 goto out;
4882
4883 if (netdev_kobject_init())
4884 goto out;
4885
4886 INIT_LIST_HEAD(&ptype_all);
4887 for (i = 0; i < PTYPE_HASH_SIZE; i++)
4888 INIT_LIST_HEAD(&ptype_base[i]);
4889
4890 if (register_pernet_subsys(&netdev_net_ops))
4891 goto out;
4892
4893 if (register_pernet_device(&default_device_ops))
4894 goto out;
4895
4896 /*
4897 * Initialise the packet receive queues.
4898 */
4899
4900 for_each_possible_cpu(i) {
4901 struct softnet_data *queue;
4902
4903 queue = &per_cpu(softnet_data, i);
4904 skb_queue_head_init(&queue->input_pkt_queue);
4905 queue->completion_queue = NULL;
4906 INIT_LIST_HEAD(&queue->poll_list);
4907
4908 queue->backlog.poll = process_backlog;
4909 queue->backlog.weight = weight_p;
4910 }
4911
4912 netdev_dma_register();
4913
4914 dev_boot_phase = 0;
4915
4916 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
4917 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
4918
4919 hotcpu_notifier(dev_cpu_callback, 0);
4920 dst_init();
4921 dev_mcast_init();
4922 rc = 0;
4923 out:
4924 return rc;
4925 }
4926
4927 subsys_initcall(net_dev_init);
4928
4929 EXPORT_SYMBOL(__dev_get_by_index);
4930 EXPORT_SYMBOL(__dev_get_by_name);
4931 EXPORT_SYMBOL(__dev_remove_pack);
4932 EXPORT_SYMBOL(dev_valid_name);
4933 EXPORT_SYMBOL(dev_add_pack);
4934 EXPORT_SYMBOL(dev_alloc_name);
4935 EXPORT_SYMBOL(dev_close);
4936 EXPORT_SYMBOL(dev_get_by_flags);
4937 EXPORT_SYMBOL(dev_get_by_index);
4938 EXPORT_SYMBOL(dev_get_by_name);
4939 EXPORT_SYMBOL(dev_open);
4940 EXPORT_SYMBOL(dev_queue_xmit);
4941 EXPORT_SYMBOL(dev_remove_pack);
4942 EXPORT_SYMBOL(dev_set_allmulti);
4943 EXPORT_SYMBOL(dev_set_promiscuity);
4944 EXPORT_SYMBOL(dev_change_flags);
4945 EXPORT_SYMBOL(dev_set_mtu);
4946 EXPORT_SYMBOL(dev_set_mac_address);
4947 EXPORT_SYMBOL(free_netdev);
4948 EXPORT_SYMBOL(netdev_boot_setup_check);
4949 EXPORT_SYMBOL(netdev_set_master);
4950 EXPORT_SYMBOL(netdev_state_change);
4951 EXPORT_SYMBOL(netif_receive_skb);
4952 EXPORT_SYMBOL(netif_rx);
4953 EXPORT_SYMBOL(register_gifconf);
4954 EXPORT_SYMBOL(register_netdevice);
4955 EXPORT_SYMBOL(register_netdevice_notifier);
4956 EXPORT_SYMBOL(skb_checksum_help);
4957 EXPORT_SYMBOL(synchronize_net);
4958 EXPORT_SYMBOL(unregister_netdevice);
4959 EXPORT_SYMBOL(unregister_netdevice_notifier);
4960 EXPORT_SYMBOL(net_enable_timestamp);
4961 EXPORT_SYMBOL(net_disable_timestamp);
4962 EXPORT_SYMBOL(dev_get_flags);
4963
4964 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
4965 EXPORT_SYMBOL(br_handle_frame_hook);
4966 EXPORT_SYMBOL(br_fdb_get_hook);
4967 EXPORT_SYMBOL(br_fdb_put_hook);
4968 #endif
4969
4970 EXPORT_SYMBOL(dev_load);
4971
4972 EXPORT_PER_CPU_SYMBOL(softnet_data);
This page took 0.137865 seconds and 5 git commands to generate.