Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 | |
02c30a84 | 10 | * Authors: Ross Biro |
1da177e4 LT |
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> | |
4fc268d2 | 78 | #include <linux/capability.h> |
1da177e4 LT |
79 | #include <linux/cpu.h> |
80 | #include <linux/types.h> | |
81 | #include <linux/kernel.h> | |
82 | #include <linux/sched.h> | |
4a3e2f71 | 83 | #include <linux/mutex.h> |
1da177e4 LT |
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> | |
0187bdfb | 93 | #include <linux/ethtool.h> |
1da177e4 LT |
94 | #include <linux/notifier.h> |
95 | #include <linux/skbuff.h> | |
457c4cbc | 96 | #include <net/net_namespace.h> |
1da177e4 LT |
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> | |
b863ceb7 | 103 | #include <linux/if_macvlan.h> |
1da177e4 LT |
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> | |
295f4a1f | 115 | #include <net/wext.h> |
1da177e4 | 116 | #include <net/iw_handler.h> |
1da177e4 | 117 | #include <asm/current.h> |
5bdb9886 | 118 | #include <linux/audit.h> |
db217334 | 119 | #include <linux/dmaengine.h> |
f6a78bfc | 120 | #include <linux/err.h> |
c7fa9d18 | 121 | #include <linux/ctype.h> |
723e98b7 | 122 | #include <linux/if_arp.h> |
6de329e2 | 123 | #include <linux/if_vlan.h> |
8f0f2223 DM |
124 | #include <linux/ip.h> |
125 | #include <linux/ipv6.h> | |
126 | #include <linux/in.h> | |
b6b2fed1 DM |
127 | #include <linux/jhash.h> |
128 | #include <linux/random.h> | |
1da177e4 | 129 | |
342709ef PE |
130 | #include "net-sysfs.h" |
131 | ||
1da177e4 LT |
132 | /* |
133 | * The list of packet types we will receive (as opposed to discard) | |
134 | * and the routines to invoke. | |
135 | * | |
136 | * Why 16. Because with 16 the only overlap we get on a hash of the | |
137 | * low nibble of the protocol value is RARP/SNAP/X.25. | |
138 | * | |
139 | * NOTE: That is no longer true with the addition of VLAN tags. Not | |
140 | * sure which should go first, but I bet it won't make much | |
141 | * difference if we are running VLANs. The good news is that | |
142 | * this protocol won't be in the list unless compiled in, so | |
3041a069 | 143 | * the average user (w/out VLANs) will not be adversely affected. |
1da177e4 LT |
144 | * --BLG |
145 | * | |
146 | * 0800 IP | |
147 | * 8100 802.1Q VLAN | |
148 | * 0001 802.3 | |
149 | * 0002 AX.25 | |
150 | * 0004 802.2 | |
151 | * 8035 RARP | |
152 | * 0005 SNAP | |
153 | * 0805 X.25 | |
154 | * 0806 ARP | |
155 | * 8137 IPX | |
156 | * 0009 Localtalk | |
157 | * 86DD IPv6 | |
158 | */ | |
159 | ||
82d8a867 PE |
160 | #define PTYPE_HASH_SIZE (16) |
161 | #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1) | |
162 | ||
1da177e4 | 163 | static DEFINE_SPINLOCK(ptype_lock); |
82d8a867 | 164 | static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
6b2bedc3 | 165 | static struct list_head ptype_all __read_mostly; /* Taps */ |
1da177e4 | 166 | |
db217334 | 167 | #ifdef CONFIG_NET_DMA |
d379b01e DW |
168 | struct net_dma { |
169 | struct dma_client client; | |
170 | spinlock_t lock; | |
171 | cpumask_t channel_mask; | |
0c0b0aca | 172 | struct dma_chan **channels; |
d379b01e DW |
173 | }; |
174 | ||
175 | static enum dma_state_client | |
176 | netdev_dma_event(struct dma_client *client, struct dma_chan *chan, | |
177 | enum dma_state state); | |
178 | ||
179 | static struct net_dma net_dma = { | |
180 | .client = { | |
181 | .event_callback = netdev_dma_event, | |
182 | }, | |
183 | }; | |
db217334 CL |
184 | #endif |
185 | ||
1da177e4 | 186 | /* |
7562f876 | 187 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
188 | * semaphore. |
189 | * | |
190 | * Pure readers hold dev_base_lock for reading. | |
191 | * | |
192 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 193 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
194 | * actual updates. This allows pure readers to access the list even |
195 | * while a writer is preparing to update it. | |
196 | * | |
197 | * To put it another way, dev_base_lock is held for writing only to | |
198 | * protect against pure readers; the rtnl semaphore provides the | |
199 | * protection against other writers. | |
200 | * | |
201 | * See, for example usages, register_netdevice() and | |
202 | * unregister_netdevice(), which must be called with the rtnl | |
203 | * semaphore held. | |
204 | */ | |
1da177e4 LT |
205 | DEFINE_RWLOCK(dev_base_lock); |
206 | ||
1da177e4 LT |
207 | EXPORT_SYMBOL(dev_base_lock); |
208 | ||
209 | #define NETDEV_HASHBITS 8 | |
881d966b | 210 | #define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS) |
1da177e4 | 211 | |
881d966b | 212 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 LT |
213 | { |
214 | unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ)); | |
881d966b | 215 | return &net->dev_name_head[hash & ((1 << NETDEV_HASHBITS) - 1)]; |
1da177e4 LT |
216 | } |
217 | ||
881d966b | 218 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 219 | { |
881d966b | 220 | return &net->dev_index_head[ifindex & ((1 << NETDEV_HASHBITS) - 1)]; |
1da177e4 LT |
221 | } |
222 | ||
ce286d32 EB |
223 | /* Device list insertion */ |
224 | static int list_netdevice(struct net_device *dev) | |
225 | { | |
c346dca1 | 226 | struct net *net = dev_net(dev); |
ce286d32 EB |
227 | |
228 | ASSERT_RTNL(); | |
229 | ||
230 | write_lock_bh(&dev_base_lock); | |
231 | list_add_tail(&dev->dev_list, &net->dev_base_head); | |
232 | hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
233 | hlist_add_head(&dev->index_hlist, dev_index_hash(net, dev->ifindex)); | |
234 | write_unlock_bh(&dev_base_lock); | |
235 | return 0; | |
236 | } | |
237 | ||
238 | /* Device list removal */ | |
239 | static void unlist_netdevice(struct net_device *dev) | |
240 | { | |
241 | ASSERT_RTNL(); | |
242 | ||
243 | /* Unlink dev from the device chain */ | |
244 | write_lock_bh(&dev_base_lock); | |
245 | list_del(&dev->dev_list); | |
246 | hlist_del(&dev->name_hlist); | |
247 | hlist_del(&dev->index_hlist); | |
248 | write_unlock_bh(&dev_base_lock); | |
249 | } | |
250 | ||
1da177e4 LT |
251 | /* |
252 | * Our notifier list | |
253 | */ | |
254 | ||
f07d5b94 | 255 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
256 | |
257 | /* | |
258 | * Device drivers call our routines to queue packets here. We empty the | |
259 | * queue in the local softnet handler. | |
260 | */ | |
bea3348e SH |
261 | |
262 | DEFINE_PER_CPU(struct softnet_data, softnet_data); | |
1da177e4 | 263 | |
cf508b12 | 264 | #ifdef CONFIG_LOCKDEP |
723e98b7 | 265 | /* |
c773e847 | 266 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class |
723e98b7 JP |
267 | * according to dev->type |
268 | */ | |
269 | static const unsigned short netdev_lock_type[] = | |
270 | {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
271 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, | |
272 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
273 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
274 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
275 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
276 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
277 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
278 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
279 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
280 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
281 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
282 | ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211, | |
283 | ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_VOID, | |
284 | ARPHRD_NONE}; | |
285 | ||
286 | static const char *netdev_lock_name[] = | |
287 | {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", | |
288 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
289 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
290 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
291 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
292 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
293 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
294 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
295 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
296 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
297 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
298 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
299 | "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211", | |
300 | "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_VOID", | |
301 | "_xmit_NONE"}; | |
302 | ||
303 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
cf508b12 | 304 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; |
723e98b7 JP |
305 | |
306 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
307 | { | |
308 | int i; | |
309 | ||
310 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
311 | if (netdev_lock_type[i] == dev_type) | |
312 | return i; | |
313 | /* the last key is used by default */ | |
314 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
315 | } | |
316 | ||
cf508b12 DM |
317 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
318 | unsigned short dev_type) | |
723e98b7 JP |
319 | { |
320 | int i; | |
321 | ||
322 | i = netdev_lock_pos(dev_type); | |
323 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
324 | netdev_lock_name[i]); | |
325 | } | |
cf508b12 DM |
326 | |
327 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
328 | { | |
329 | int i; | |
330 | ||
331 | i = netdev_lock_pos(dev->type); | |
332 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
333 | &netdev_addr_lock_key[i], | |
334 | netdev_lock_name[i]); | |
335 | } | |
723e98b7 | 336 | #else |
cf508b12 DM |
337 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
338 | unsigned short dev_type) | |
339 | { | |
340 | } | |
341 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
723e98b7 JP |
342 | { |
343 | } | |
344 | #endif | |
1da177e4 LT |
345 | |
346 | /******************************************************************************* | |
347 | ||
348 | Protocol management and registration routines | |
349 | ||
350 | *******************************************************************************/ | |
351 | ||
1da177e4 LT |
352 | /* |
353 | * Add a protocol ID to the list. Now that the input handler is | |
354 | * smarter we can dispense with all the messy stuff that used to be | |
355 | * here. | |
356 | * | |
357 | * BEWARE!!! Protocol handlers, mangling input packets, | |
358 | * MUST BE last in hash buckets and checking protocol handlers | |
359 | * MUST start from promiscuous ptype_all chain in net_bh. | |
360 | * It is true now, do not change it. | |
361 | * Explanation follows: if protocol handler, mangling packet, will | |
362 | * be the first on list, it is not able to sense, that packet | |
363 | * is cloned and should be copied-on-write, so that it will | |
364 | * change it and subsequent readers will get broken packet. | |
365 | * --ANK (980803) | |
366 | */ | |
367 | ||
368 | /** | |
369 | * dev_add_pack - add packet handler | |
370 | * @pt: packet type declaration | |
371 | * | |
372 | * Add a protocol handler to the networking stack. The passed &packet_type | |
373 | * is linked into kernel lists and may not be freed until it has been | |
374 | * removed from the kernel lists. | |
375 | * | |
4ec93edb | 376 | * This call does not sleep therefore it can not |
1da177e4 LT |
377 | * guarantee all CPU's that are in middle of receiving packets |
378 | * will see the new packet type (until the next received packet). | |
379 | */ | |
380 | ||
381 | void dev_add_pack(struct packet_type *pt) | |
382 | { | |
383 | int hash; | |
384 | ||
385 | spin_lock_bh(&ptype_lock); | |
9be9a6b9 | 386 | if (pt->type == htons(ETH_P_ALL)) |
1da177e4 | 387 | list_add_rcu(&pt->list, &ptype_all); |
9be9a6b9 | 388 | else { |
82d8a867 | 389 | hash = ntohs(pt->type) & PTYPE_HASH_MASK; |
1da177e4 LT |
390 | list_add_rcu(&pt->list, &ptype_base[hash]); |
391 | } | |
392 | spin_unlock_bh(&ptype_lock); | |
393 | } | |
394 | ||
1da177e4 LT |
395 | /** |
396 | * __dev_remove_pack - remove packet handler | |
397 | * @pt: packet type declaration | |
398 | * | |
399 | * Remove a protocol handler that was previously added to the kernel | |
400 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
401 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 402 | * returns. |
1da177e4 LT |
403 | * |
404 | * The packet type might still be in use by receivers | |
405 | * and must not be freed until after all the CPU's have gone | |
406 | * through a quiescent state. | |
407 | */ | |
408 | void __dev_remove_pack(struct packet_type *pt) | |
409 | { | |
410 | struct list_head *head; | |
411 | struct packet_type *pt1; | |
412 | ||
413 | spin_lock_bh(&ptype_lock); | |
414 | ||
9be9a6b9 | 415 | if (pt->type == htons(ETH_P_ALL)) |
1da177e4 | 416 | head = &ptype_all; |
9be9a6b9 | 417 | else |
82d8a867 | 418 | head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; |
1da177e4 LT |
419 | |
420 | list_for_each_entry(pt1, head, list) { | |
421 | if (pt == pt1) { | |
422 | list_del_rcu(&pt->list); | |
423 | goto out; | |
424 | } | |
425 | } | |
426 | ||
427 | printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt); | |
428 | out: | |
429 | spin_unlock_bh(&ptype_lock); | |
430 | } | |
431 | /** | |
432 | * dev_remove_pack - remove packet handler | |
433 | * @pt: packet type declaration | |
434 | * | |
435 | * Remove a protocol handler that was previously added to the kernel | |
436 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
437 | * from the kernel lists and can be freed or reused once this function | |
438 | * returns. | |
439 | * | |
440 | * This call sleeps to guarantee that no CPU is looking at the packet | |
441 | * type after return. | |
442 | */ | |
443 | void dev_remove_pack(struct packet_type *pt) | |
444 | { | |
445 | __dev_remove_pack(pt); | |
4ec93edb | 446 | |
1da177e4 LT |
447 | synchronize_net(); |
448 | } | |
449 | ||
450 | /****************************************************************************** | |
451 | ||
452 | Device Boot-time Settings Routines | |
453 | ||
454 | *******************************************************************************/ | |
455 | ||
456 | /* Boot time configuration table */ | |
457 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
458 | ||
459 | /** | |
460 | * netdev_boot_setup_add - add new setup entry | |
461 | * @name: name of the device | |
462 | * @map: configured settings for the device | |
463 | * | |
464 | * Adds new setup entry to the dev_boot_setup list. The function | |
465 | * returns 0 on error and 1 on success. This is a generic routine to | |
466 | * all netdevices. | |
467 | */ | |
468 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
469 | { | |
470 | struct netdev_boot_setup *s; | |
471 | int i; | |
472 | ||
473 | s = dev_boot_setup; | |
474 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
475 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
476 | memset(s[i].name, 0, sizeof(s[i].name)); | |
93b3cff9 | 477 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
478 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
479 | break; | |
480 | } | |
481 | } | |
482 | ||
483 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
484 | } | |
485 | ||
486 | /** | |
487 | * netdev_boot_setup_check - check boot time settings | |
488 | * @dev: the netdevice | |
489 | * | |
490 | * Check boot time settings for the device. | |
491 | * The found settings are set for the device to be used | |
492 | * later in the device probing. | |
493 | * Returns 0 if no settings found, 1 if they are. | |
494 | */ | |
495 | int netdev_boot_setup_check(struct net_device *dev) | |
496 | { | |
497 | struct netdev_boot_setup *s = dev_boot_setup; | |
498 | int i; | |
499 | ||
500 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
501 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
93b3cff9 | 502 | !strcmp(dev->name, s[i].name)) { |
1da177e4 LT |
503 | dev->irq = s[i].map.irq; |
504 | dev->base_addr = s[i].map.base_addr; | |
505 | dev->mem_start = s[i].map.mem_start; | |
506 | dev->mem_end = s[i].map.mem_end; | |
507 | return 1; | |
508 | } | |
509 | } | |
510 | return 0; | |
511 | } | |
512 | ||
513 | ||
514 | /** | |
515 | * netdev_boot_base - get address from boot time settings | |
516 | * @prefix: prefix for network device | |
517 | * @unit: id for network device | |
518 | * | |
519 | * Check boot time settings for the base address of device. | |
520 | * The found settings are set for the device to be used | |
521 | * later in the device probing. | |
522 | * Returns 0 if no settings found. | |
523 | */ | |
524 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
525 | { | |
526 | const struct netdev_boot_setup *s = dev_boot_setup; | |
527 | char name[IFNAMSIZ]; | |
528 | int i; | |
529 | ||
530 | sprintf(name, "%s%d", prefix, unit); | |
531 | ||
532 | /* | |
533 | * If device already registered then return base of 1 | |
534 | * to indicate not to probe for this interface | |
535 | */ | |
881d966b | 536 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
537 | return 1; |
538 | ||
539 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
540 | if (!strcmp(name, s[i].name)) | |
541 | return s[i].map.base_addr; | |
542 | return 0; | |
543 | } | |
544 | ||
545 | /* | |
546 | * Saves at boot time configured settings for any netdevice. | |
547 | */ | |
548 | int __init netdev_boot_setup(char *str) | |
549 | { | |
550 | int ints[5]; | |
551 | struct ifmap map; | |
552 | ||
553 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
554 | if (!str || !*str) | |
555 | return 0; | |
556 | ||
557 | /* Save settings */ | |
558 | memset(&map, 0, sizeof(map)); | |
559 | if (ints[0] > 0) | |
560 | map.irq = ints[1]; | |
561 | if (ints[0] > 1) | |
562 | map.base_addr = ints[2]; | |
563 | if (ints[0] > 2) | |
564 | map.mem_start = ints[3]; | |
565 | if (ints[0] > 3) | |
566 | map.mem_end = ints[4]; | |
567 | ||
568 | /* Add new entry to the list */ | |
569 | return netdev_boot_setup_add(str, &map); | |
570 | } | |
571 | ||
572 | __setup("netdev=", netdev_boot_setup); | |
573 | ||
574 | /******************************************************************************* | |
575 | ||
576 | Device Interface Subroutines | |
577 | ||
578 | *******************************************************************************/ | |
579 | ||
580 | /** | |
581 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 582 | * @net: the applicable net namespace |
1da177e4 LT |
583 | * @name: name to find |
584 | * | |
585 | * Find an interface by name. Must be called under RTNL semaphore | |
586 | * or @dev_base_lock. If the name is found a pointer to the device | |
587 | * is returned. If the name is not found then %NULL is returned. The | |
588 | * reference counters are not incremented so the caller must be | |
589 | * careful with locks. | |
590 | */ | |
591 | ||
881d966b | 592 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
593 | { |
594 | struct hlist_node *p; | |
595 | ||
881d966b | 596 | hlist_for_each(p, dev_name_hash(net, name)) { |
1da177e4 LT |
597 | struct net_device *dev |
598 | = hlist_entry(p, struct net_device, name_hlist); | |
599 | if (!strncmp(dev->name, name, IFNAMSIZ)) | |
600 | return dev; | |
601 | } | |
602 | return NULL; | |
603 | } | |
604 | ||
605 | /** | |
606 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 607 | * @net: the applicable net namespace |
1da177e4 LT |
608 | * @name: name to find |
609 | * | |
610 | * Find an interface by name. This can be called from any | |
611 | * context and does its own locking. The returned handle has | |
612 | * the usage count incremented and the caller must use dev_put() to | |
613 | * release it when it is no longer needed. %NULL is returned if no | |
614 | * matching device is found. | |
615 | */ | |
616 | ||
881d966b | 617 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
618 | { |
619 | struct net_device *dev; | |
620 | ||
621 | read_lock(&dev_base_lock); | |
881d966b | 622 | dev = __dev_get_by_name(net, name); |
1da177e4 LT |
623 | if (dev) |
624 | dev_hold(dev); | |
625 | read_unlock(&dev_base_lock); | |
626 | return dev; | |
627 | } | |
628 | ||
629 | /** | |
630 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 631 | * @net: the applicable net namespace |
1da177e4 LT |
632 | * @ifindex: index of device |
633 | * | |
634 | * Search for an interface by index. Returns %NULL if the device | |
635 | * is not found or a pointer to the device. The device has not | |
636 | * had its reference counter increased so the caller must be careful | |
637 | * about locking. The caller must hold either the RTNL semaphore | |
638 | * or @dev_base_lock. | |
639 | */ | |
640 | ||
881d966b | 641 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
642 | { |
643 | struct hlist_node *p; | |
644 | ||
881d966b | 645 | hlist_for_each(p, dev_index_hash(net, ifindex)) { |
1da177e4 LT |
646 | struct net_device *dev |
647 | = hlist_entry(p, struct net_device, index_hlist); | |
648 | if (dev->ifindex == ifindex) | |
649 | return dev; | |
650 | } | |
651 | return NULL; | |
652 | } | |
653 | ||
654 | ||
655 | /** | |
656 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 657 | * @net: the applicable net namespace |
1da177e4 LT |
658 | * @ifindex: index of device |
659 | * | |
660 | * Search for an interface by index. Returns NULL if the device | |
661 | * is not found or a pointer to the device. The device returned has | |
662 | * had a reference added and the pointer is safe until the user calls | |
663 | * dev_put to indicate they have finished with it. | |
664 | */ | |
665 | ||
881d966b | 666 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
667 | { |
668 | struct net_device *dev; | |
669 | ||
670 | read_lock(&dev_base_lock); | |
881d966b | 671 | dev = __dev_get_by_index(net, ifindex); |
1da177e4 LT |
672 | if (dev) |
673 | dev_hold(dev); | |
674 | read_unlock(&dev_base_lock); | |
675 | return dev; | |
676 | } | |
677 | ||
678 | /** | |
679 | * dev_getbyhwaddr - find a device by its hardware address | |
c4ea43c5 | 680 | * @net: the applicable net namespace |
1da177e4 LT |
681 | * @type: media type of device |
682 | * @ha: hardware address | |
683 | * | |
684 | * Search for an interface by MAC address. Returns NULL if the device | |
685 | * is not found or a pointer to the device. The caller must hold the | |
686 | * rtnl semaphore. The returned device has not had its ref count increased | |
687 | * and the caller must therefore be careful about locking | |
688 | * | |
689 | * BUGS: | |
690 | * If the API was consistent this would be __dev_get_by_hwaddr | |
691 | */ | |
692 | ||
881d966b | 693 | struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha) |
1da177e4 LT |
694 | { |
695 | struct net_device *dev; | |
696 | ||
697 | ASSERT_RTNL(); | |
698 | ||
81103a52 | 699 | for_each_netdev(net, dev) |
1da177e4 LT |
700 | if (dev->type == type && |
701 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
702 | return dev; |
703 | ||
704 | return NULL; | |
1da177e4 LT |
705 | } |
706 | ||
cf309e3f JF |
707 | EXPORT_SYMBOL(dev_getbyhwaddr); |
708 | ||
881d966b | 709 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
710 | { |
711 | struct net_device *dev; | |
712 | ||
4e9cac2b | 713 | ASSERT_RTNL(); |
881d966b | 714 | for_each_netdev(net, dev) |
4e9cac2b | 715 | if (dev->type == type) |
7562f876 PE |
716 | return dev; |
717 | ||
718 | return NULL; | |
4e9cac2b PM |
719 | } |
720 | ||
721 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); | |
722 | ||
881d966b | 723 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b PM |
724 | { |
725 | struct net_device *dev; | |
726 | ||
727 | rtnl_lock(); | |
881d966b | 728 | dev = __dev_getfirstbyhwtype(net, type); |
4e9cac2b PM |
729 | if (dev) |
730 | dev_hold(dev); | |
1da177e4 LT |
731 | rtnl_unlock(); |
732 | return dev; | |
733 | } | |
734 | ||
735 | EXPORT_SYMBOL(dev_getfirstbyhwtype); | |
736 | ||
737 | /** | |
738 | * dev_get_by_flags - find any device with given flags | |
c4ea43c5 | 739 | * @net: the applicable net namespace |
1da177e4 LT |
740 | * @if_flags: IFF_* values |
741 | * @mask: bitmask of bits in if_flags to check | |
742 | * | |
743 | * Search for any interface with the given flags. Returns NULL if a device | |
4ec93edb | 744 | * is not found or a pointer to the device. The device returned has |
1da177e4 LT |
745 | * had a reference added and the pointer is safe until the user calls |
746 | * dev_put to indicate they have finished with it. | |
747 | */ | |
748 | ||
881d966b | 749 | struct net_device * dev_get_by_flags(struct net *net, unsigned short if_flags, unsigned short mask) |
1da177e4 | 750 | { |
7562f876 | 751 | struct net_device *dev, *ret; |
1da177e4 | 752 | |
7562f876 | 753 | ret = NULL; |
1da177e4 | 754 | read_lock(&dev_base_lock); |
881d966b | 755 | for_each_netdev(net, dev) { |
1da177e4 LT |
756 | if (((dev->flags ^ if_flags) & mask) == 0) { |
757 | dev_hold(dev); | |
7562f876 | 758 | ret = dev; |
1da177e4 LT |
759 | break; |
760 | } | |
761 | } | |
762 | read_unlock(&dev_base_lock); | |
7562f876 | 763 | return ret; |
1da177e4 LT |
764 | } |
765 | ||
766 | /** | |
767 | * dev_valid_name - check if name is okay for network device | |
768 | * @name: name string | |
769 | * | |
770 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
771 | * to allow sysfs to work. We also disallow any kind of |
772 | * whitespace. | |
1da177e4 | 773 | */ |
c2373ee9 | 774 | int dev_valid_name(const char *name) |
1da177e4 | 775 | { |
c7fa9d18 DM |
776 | if (*name == '\0') |
777 | return 0; | |
b6fe17d6 SH |
778 | if (strlen(name) >= IFNAMSIZ) |
779 | return 0; | |
c7fa9d18 DM |
780 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
781 | return 0; | |
782 | ||
783 | while (*name) { | |
784 | if (*name == '/' || isspace(*name)) | |
785 | return 0; | |
786 | name++; | |
787 | } | |
788 | return 1; | |
1da177e4 LT |
789 | } |
790 | ||
791 | /** | |
b267b179 EB |
792 | * __dev_alloc_name - allocate a name for a device |
793 | * @net: network namespace to allocate the device name in | |
1da177e4 | 794 | * @name: name format string |
b267b179 | 795 | * @buf: scratch buffer and result name string |
1da177e4 LT |
796 | * |
797 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
798 | * id. It scans list of devices to build up a free map, then chooses |
799 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
800 | * while allocating the name and adding the device in order to avoid | |
801 | * duplicates. | |
802 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
803 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
804 | */ |
805 | ||
b267b179 | 806 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
807 | { |
808 | int i = 0; | |
1da177e4 LT |
809 | const char *p; |
810 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 811 | unsigned long *inuse; |
1da177e4 LT |
812 | struct net_device *d; |
813 | ||
814 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
815 | if (p) { | |
816 | /* | |
817 | * Verify the string as this thing may have come from | |
818 | * the user. There must be either one "%d" and no other "%" | |
819 | * characters. | |
820 | */ | |
821 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
822 | return -EINVAL; | |
823 | ||
824 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 825 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
826 | if (!inuse) |
827 | return -ENOMEM; | |
828 | ||
881d966b | 829 | for_each_netdev(net, d) { |
1da177e4 LT |
830 | if (!sscanf(d->name, name, &i)) |
831 | continue; | |
832 | if (i < 0 || i >= max_netdevices) | |
833 | continue; | |
834 | ||
835 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 836 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
837 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
838 | set_bit(i, inuse); | |
839 | } | |
840 | ||
841 | i = find_first_zero_bit(inuse, max_netdevices); | |
842 | free_page((unsigned long) inuse); | |
843 | } | |
844 | ||
b267b179 EB |
845 | snprintf(buf, IFNAMSIZ, name, i); |
846 | if (!__dev_get_by_name(net, buf)) | |
1da177e4 | 847 | return i; |
1da177e4 LT |
848 | |
849 | /* It is possible to run out of possible slots | |
850 | * when the name is long and there isn't enough space left | |
851 | * for the digits, or if all bits are used. | |
852 | */ | |
853 | return -ENFILE; | |
854 | } | |
855 | ||
b267b179 EB |
856 | /** |
857 | * dev_alloc_name - allocate a name for a device | |
858 | * @dev: device | |
859 | * @name: name format string | |
860 | * | |
861 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
862 | * id. It scans list of devices to build up a free map, then chooses | |
863 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
864 | * while allocating the name and adding the device in order to avoid | |
865 | * duplicates. | |
866 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
867 | * Returns the number of the unit assigned or a negative errno code. | |
868 | */ | |
869 | ||
870 | int dev_alloc_name(struct net_device *dev, const char *name) | |
871 | { | |
872 | char buf[IFNAMSIZ]; | |
873 | struct net *net; | |
874 | int ret; | |
875 | ||
c346dca1 YH |
876 | BUG_ON(!dev_net(dev)); |
877 | net = dev_net(dev); | |
b267b179 EB |
878 | ret = __dev_alloc_name(net, name, buf); |
879 | if (ret >= 0) | |
880 | strlcpy(dev->name, buf, IFNAMSIZ); | |
881 | return ret; | |
882 | } | |
883 | ||
1da177e4 LT |
884 | |
885 | /** | |
886 | * dev_change_name - change name of a device | |
887 | * @dev: device | |
888 | * @newname: name (or format string) must be at least IFNAMSIZ | |
889 | * | |
890 | * Change name of a device, can pass format strings "eth%d". | |
891 | * for wildcarding. | |
892 | */ | |
cf04a4c7 | 893 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 894 | { |
fcc5a03a | 895 | char oldname[IFNAMSIZ]; |
1da177e4 | 896 | int err = 0; |
fcc5a03a | 897 | int ret; |
881d966b | 898 | struct net *net; |
1da177e4 LT |
899 | |
900 | ASSERT_RTNL(); | |
c346dca1 | 901 | BUG_ON(!dev_net(dev)); |
1da177e4 | 902 | |
c346dca1 | 903 | net = dev_net(dev); |
1da177e4 LT |
904 | if (dev->flags & IFF_UP) |
905 | return -EBUSY; | |
906 | ||
907 | if (!dev_valid_name(newname)) | |
908 | return -EINVAL; | |
909 | ||
c8d90dca SH |
910 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) |
911 | return 0; | |
912 | ||
fcc5a03a HX |
913 | memcpy(oldname, dev->name, IFNAMSIZ); |
914 | ||
1da177e4 LT |
915 | if (strchr(newname, '%')) { |
916 | err = dev_alloc_name(dev, newname); | |
917 | if (err < 0) | |
918 | return err; | |
1da177e4 | 919 | } |
881d966b | 920 | else if (__dev_get_by_name(net, newname)) |
1da177e4 LT |
921 | return -EEXIST; |
922 | else | |
923 | strlcpy(dev->name, newname, IFNAMSIZ); | |
924 | ||
fcc5a03a | 925 | rollback: |
dcc99773 SH |
926 | err = device_rename(&dev->dev, dev->name); |
927 | if (err) { | |
928 | memcpy(dev->name, oldname, IFNAMSIZ); | |
929 | return err; | |
930 | } | |
7f988eab HX |
931 | |
932 | write_lock_bh(&dev_base_lock); | |
92749821 | 933 | hlist_del(&dev->name_hlist); |
881d966b | 934 | hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name)); |
7f988eab HX |
935 | write_unlock_bh(&dev_base_lock); |
936 | ||
056925ab | 937 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
938 | ret = notifier_to_errno(ret); |
939 | ||
940 | if (ret) { | |
941 | if (err) { | |
942 | printk(KERN_ERR | |
943 | "%s: name change rollback failed: %d.\n", | |
944 | dev->name, ret); | |
945 | } else { | |
946 | err = ret; | |
947 | memcpy(dev->name, oldname, IFNAMSIZ); | |
948 | goto rollback; | |
949 | } | |
950 | } | |
1da177e4 LT |
951 | |
952 | return err; | |
953 | } | |
954 | ||
0b815a1a SH |
955 | /** |
956 | * dev_set_alias - change ifalias of a device | |
957 | * @dev: device | |
958 | * @alias: name up to IFALIASZ | |
959 | * | |
960 | * Set ifalias for a device, | |
961 | */ | |
962 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
963 | { | |
964 | ASSERT_RTNL(); | |
965 | ||
966 | if (len >= IFALIASZ) | |
967 | return -EINVAL; | |
968 | ||
96ca4a2c OH |
969 | if (!len) { |
970 | if (dev->ifalias) { | |
971 | kfree(dev->ifalias); | |
972 | dev->ifalias = NULL; | |
973 | } | |
974 | return 0; | |
975 | } | |
976 | ||
0b815a1a SH |
977 | dev->ifalias = krealloc(dev->ifalias, len+1, GFP_KERNEL); |
978 | if (!dev->ifalias) | |
979 | return -ENOMEM; | |
980 | ||
981 | strlcpy(dev->ifalias, alias, len+1); | |
982 | return len; | |
983 | } | |
984 | ||
985 | ||
d8a33ac4 | 986 | /** |
3041a069 | 987 | * netdev_features_change - device changes features |
d8a33ac4 SH |
988 | * @dev: device to cause notification |
989 | * | |
990 | * Called to indicate a device has changed features. | |
991 | */ | |
992 | void netdev_features_change(struct net_device *dev) | |
993 | { | |
056925ab | 994 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
995 | } |
996 | EXPORT_SYMBOL(netdev_features_change); | |
997 | ||
1da177e4 LT |
998 | /** |
999 | * netdev_state_change - device changes state | |
1000 | * @dev: device to cause notification | |
1001 | * | |
1002 | * Called to indicate a device has changed state. This function calls | |
1003 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1004 | * to the routing socket. | |
1005 | */ | |
1006 | void netdev_state_change(struct net_device *dev) | |
1007 | { | |
1008 | if (dev->flags & IFF_UP) { | |
056925ab | 1009 | call_netdevice_notifiers(NETDEV_CHANGE, dev); |
1da177e4 LT |
1010 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0); |
1011 | } | |
1012 | } | |
1013 | ||
c1da4ac7 OG |
1014 | void netdev_bonding_change(struct net_device *dev) |
1015 | { | |
1016 | call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, dev); | |
1017 | } | |
1018 | EXPORT_SYMBOL(netdev_bonding_change); | |
1019 | ||
1da177e4 LT |
1020 | /** |
1021 | * dev_load - load a network module | |
c4ea43c5 | 1022 | * @net: the applicable net namespace |
1da177e4 LT |
1023 | * @name: name of interface |
1024 | * | |
1025 | * If a network interface is not present and the process has suitable | |
1026 | * privileges this function loads the module. If module loading is not | |
1027 | * available in this kernel then it becomes a nop. | |
1028 | */ | |
1029 | ||
881d966b | 1030 | void dev_load(struct net *net, const char *name) |
1da177e4 | 1031 | { |
4ec93edb | 1032 | struct net_device *dev; |
1da177e4 LT |
1033 | |
1034 | read_lock(&dev_base_lock); | |
881d966b | 1035 | dev = __dev_get_by_name(net, name); |
1da177e4 LT |
1036 | read_unlock(&dev_base_lock); |
1037 | ||
1038 | if (!dev && capable(CAP_SYS_MODULE)) | |
1039 | request_module("%s", name); | |
1040 | } | |
1041 | ||
1da177e4 LT |
1042 | /** |
1043 | * dev_open - prepare an interface for use. | |
1044 | * @dev: device to open | |
1045 | * | |
1046 | * Takes a device from down to up state. The device's private open | |
1047 | * function is invoked and then the multicast lists are loaded. Finally | |
1048 | * the device is moved into the up state and a %NETDEV_UP message is | |
1049 | * sent to the netdev notifier chain. | |
1050 | * | |
1051 | * Calling this function on an active interface is a nop. On a failure | |
1052 | * a negative errno code is returned. | |
1053 | */ | |
1054 | int dev_open(struct net_device *dev) | |
1055 | { | |
1056 | int ret = 0; | |
1057 | ||
e46b66bc BH |
1058 | ASSERT_RTNL(); |
1059 | ||
1da177e4 LT |
1060 | /* |
1061 | * Is it already up? | |
1062 | */ | |
1063 | ||
1064 | if (dev->flags & IFF_UP) | |
1065 | return 0; | |
1066 | ||
1067 | /* | |
1068 | * Is it even present? | |
1069 | */ | |
1070 | if (!netif_device_present(dev)) | |
1071 | return -ENODEV; | |
1072 | ||
1073 | /* | |
1074 | * Call device private open method | |
1075 | */ | |
1076 | set_bit(__LINK_STATE_START, &dev->state); | |
bada339b JG |
1077 | |
1078 | if (dev->validate_addr) | |
1079 | ret = dev->validate_addr(dev); | |
1080 | ||
1081 | if (!ret && dev->open) | |
1da177e4 | 1082 | ret = dev->open(dev); |
1da177e4 | 1083 | |
4ec93edb | 1084 | /* |
1da177e4 LT |
1085 | * If it went open OK then: |
1086 | */ | |
1087 | ||
bada339b JG |
1088 | if (ret) |
1089 | clear_bit(__LINK_STATE_START, &dev->state); | |
1090 | else { | |
1da177e4 LT |
1091 | /* |
1092 | * Set the flags. | |
1093 | */ | |
1094 | dev->flags |= IFF_UP; | |
1095 | ||
1096 | /* | |
1097 | * Initialize multicasting status | |
1098 | */ | |
4417da66 | 1099 | dev_set_rx_mode(dev); |
1da177e4 LT |
1100 | |
1101 | /* | |
1102 | * Wakeup transmit queue engine | |
1103 | */ | |
1104 | dev_activate(dev); | |
1105 | ||
1106 | /* | |
1107 | * ... and announce new interface. | |
1108 | */ | |
056925ab | 1109 | call_netdevice_notifiers(NETDEV_UP, dev); |
1da177e4 | 1110 | } |
bada339b | 1111 | |
1da177e4 LT |
1112 | return ret; |
1113 | } | |
1114 | ||
1115 | /** | |
1116 | * dev_close - shutdown an interface. | |
1117 | * @dev: device to shutdown | |
1118 | * | |
1119 | * This function moves an active device into down state. A | |
1120 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1121 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1122 | * chain. | |
1123 | */ | |
1124 | int dev_close(struct net_device *dev) | |
1125 | { | |
e46b66bc BH |
1126 | ASSERT_RTNL(); |
1127 | ||
9d5010db DM |
1128 | might_sleep(); |
1129 | ||
1da177e4 LT |
1130 | if (!(dev->flags & IFF_UP)) |
1131 | return 0; | |
1132 | ||
1133 | /* | |
1134 | * Tell people we are going down, so that they can | |
1135 | * prepare to death, when device is still operating. | |
1136 | */ | |
056925ab | 1137 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1138 | |
1da177e4 LT |
1139 | clear_bit(__LINK_STATE_START, &dev->state); |
1140 | ||
1141 | /* Synchronize to scheduled poll. We cannot touch poll list, | |
bea3348e SH |
1142 | * it can be even on different cpu. So just clear netif_running(). |
1143 | * | |
1144 | * dev->stop() will invoke napi_disable() on all of it's | |
1145 | * napi_struct instances on this device. | |
1146 | */ | |
1da177e4 | 1147 | smp_mb__after_clear_bit(); /* Commit netif_running(). */ |
1da177e4 | 1148 | |
d8b2a4d2 ML |
1149 | dev_deactivate(dev); |
1150 | ||
1da177e4 LT |
1151 | /* |
1152 | * Call the device specific close. This cannot fail. | |
1153 | * Only if device is UP | |
1154 | * | |
1155 | * We allow it to be called even after a DETACH hot-plug | |
1156 | * event. | |
1157 | */ | |
1158 | if (dev->stop) | |
1159 | dev->stop(dev); | |
1160 | ||
1161 | /* | |
1162 | * Device is now down. | |
1163 | */ | |
1164 | ||
1165 | dev->flags &= ~IFF_UP; | |
1166 | ||
1167 | /* | |
1168 | * Tell people we are down | |
1169 | */ | |
056925ab | 1170 | call_netdevice_notifiers(NETDEV_DOWN, dev); |
1da177e4 LT |
1171 | |
1172 | return 0; | |
1173 | } | |
1174 | ||
1175 | ||
0187bdfb BH |
1176 | /** |
1177 | * dev_disable_lro - disable Large Receive Offload on a device | |
1178 | * @dev: device | |
1179 | * | |
1180 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1181 | * called under RTNL. This is needed if received packets may be | |
1182 | * forwarded to another interface. | |
1183 | */ | |
1184 | void dev_disable_lro(struct net_device *dev) | |
1185 | { | |
1186 | if (dev->ethtool_ops && dev->ethtool_ops->get_flags && | |
1187 | dev->ethtool_ops->set_flags) { | |
1188 | u32 flags = dev->ethtool_ops->get_flags(dev); | |
1189 | if (flags & ETH_FLAG_LRO) { | |
1190 | flags &= ~ETH_FLAG_LRO; | |
1191 | dev->ethtool_ops->set_flags(dev, flags); | |
1192 | } | |
1193 | } | |
1194 | WARN_ON(dev->features & NETIF_F_LRO); | |
1195 | } | |
1196 | EXPORT_SYMBOL(dev_disable_lro); | |
1197 | ||
1198 | ||
881d966b EB |
1199 | static int dev_boot_phase = 1; |
1200 | ||
1da177e4 LT |
1201 | /* |
1202 | * Device change register/unregister. These are not inline or static | |
1203 | * as we export them to the world. | |
1204 | */ | |
1205 | ||
1206 | /** | |
1207 | * register_netdevice_notifier - register a network notifier block | |
1208 | * @nb: notifier | |
1209 | * | |
1210 | * Register a notifier to be called when network device events occur. | |
1211 | * The notifier passed is linked into the kernel structures and must | |
1212 | * not be reused until it has been unregistered. A negative errno code | |
1213 | * is returned on a failure. | |
1214 | * | |
1215 | * When registered all registration and up events are replayed | |
4ec93edb | 1216 | * to the new notifier to allow device to have a race free |
1da177e4 LT |
1217 | * view of the network device list. |
1218 | */ | |
1219 | ||
1220 | int register_netdevice_notifier(struct notifier_block *nb) | |
1221 | { | |
1222 | struct net_device *dev; | |
fcc5a03a | 1223 | struct net_device *last; |
881d966b | 1224 | struct net *net; |
1da177e4 LT |
1225 | int err; |
1226 | ||
1227 | rtnl_lock(); | |
f07d5b94 | 1228 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1229 | if (err) |
1230 | goto unlock; | |
881d966b EB |
1231 | if (dev_boot_phase) |
1232 | goto unlock; | |
1233 | for_each_net(net) { | |
1234 | for_each_netdev(net, dev) { | |
1235 | err = nb->notifier_call(nb, NETDEV_REGISTER, dev); | |
1236 | err = notifier_to_errno(err); | |
1237 | if (err) | |
1238 | goto rollback; | |
1239 | ||
1240 | if (!(dev->flags & IFF_UP)) | |
1241 | continue; | |
1da177e4 | 1242 | |
881d966b EB |
1243 | nb->notifier_call(nb, NETDEV_UP, dev); |
1244 | } | |
1da177e4 | 1245 | } |
fcc5a03a HX |
1246 | |
1247 | unlock: | |
1da177e4 LT |
1248 | rtnl_unlock(); |
1249 | return err; | |
fcc5a03a HX |
1250 | |
1251 | rollback: | |
1252 | last = dev; | |
881d966b EB |
1253 | for_each_net(net) { |
1254 | for_each_netdev(net, dev) { | |
1255 | if (dev == last) | |
1256 | break; | |
fcc5a03a | 1257 | |
881d966b EB |
1258 | if (dev->flags & IFF_UP) { |
1259 | nb->notifier_call(nb, NETDEV_GOING_DOWN, dev); | |
1260 | nb->notifier_call(nb, NETDEV_DOWN, dev); | |
1261 | } | |
1262 | nb->notifier_call(nb, NETDEV_UNREGISTER, dev); | |
fcc5a03a | 1263 | } |
fcc5a03a | 1264 | } |
c67625a1 PE |
1265 | |
1266 | raw_notifier_chain_unregister(&netdev_chain, nb); | |
fcc5a03a | 1267 | goto unlock; |
1da177e4 LT |
1268 | } |
1269 | ||
1270 | /** | |
1271 | * unregister_netdevice_notifier - unregister a network notifier block | |
1272 | * @nb: notifier | |
1273 | * | |
1274 | * Unregister a notifier previously registered by | |
1275 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1276 | * kernel structures and may then be reused. A negative errno code | |
1277 | * is returned on a failure. | |
1278 | */ | |
1279 | ||
1280 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1281 | { | |
9f514950 HX |
1282 | int err; |
1283 | ||
1284 | rtnl_lock(); | |
f07d5b94 | 1285 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
9f514950 HX |
1286 | rtnl_unlock(); |
1287 | return err; | |
1da177e4 LT |
1288 | } |
1289 | ||
1290 | /** | |
1291 | * call_netdevice_notifiers - call all network notifier blocks | |
1292 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1293 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1294 | * |
1295 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1296 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1297 | */ |
1298 | ||
ad7379d4 | 1299 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1300 | { |
ad7379d4 | 1301 | return raw_notifier_call_chain(&netdev_chain, val, dev); |
1da177e4 LT |
1302 | } |
1303 | ||
1304 | /* When > 0 there are consumers of rx skb time stamps */ | |
1305 | static atomic_t netstamp_needed = ATOMIC_INIT(0); | |
1306 | ||
1307 | void net_enable_timestamp(void) | |
1308 | { | |
1309 | atomic_inc(&netstamp_needed); | |
1310 | } | |
1311 | ||
1312 | void net_disable_timestamp(void) | |
1313 | { | |
1314 | atomic_dec(&netstamp_needed); | |
1315 | } | |
1316 | ||
a61bbcf2 | 1317 | static inline void net_timestamp(struct sk_buff *skb) |
1da177e4 LT |
1318 | { |
1319 | if (atomic_read(&netstamp_needed)) | |
a61bbcf2 | 1320 | __net_timestamp(skb); |
b7aa0bf7 ED |
1321 | else |
1322 | skb->tstamp.tv64 = 0; | |
1da177e4 LT |
1323 | } |
1324 | ||
1325 | /* | |
1326 | * Support routine. Sends outgoing frames to any network | |
1327 | * taps currently in use. | |
1328 | */ | |
1329 | ||
f6a78bfc | 1330 | static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
1331 | { |
1332 | struct packet_type *ptype; | |
a61bbcf2 PM |
1333 | |
1334 | net_timestamp(skb); | |
1da177e4 LT |
1335 | |
1336 | rcu_read_lock(); | |
1337 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
1338 | /* Never send packets back to the socket | |
1339 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1340 | */ | |
1341 | if ((ptype->dev == dev || !ptype->dev) && | |
1342 | (ptype->af_packet_priv == NULL || | |
1343 | (struct sock *)ptype->af_packet_priv != skb->sk)) { | |
1344 | struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC); | |
1345 | if (!skb2) | |
1346 | break; | |
1347 | ||
1348 | /* skb->nh should be correctly | |
1349 | set by sender, so that the second statement is | |
1350 | just protection against buggy protocols. | |
1351 | */ | |
459a98ed | 1352 | skb_reset_mac_header(skb2); |
1da177e4 | 1353 | |
d56f90a7 | 1354 | if (skb_network_header(skb2) < skb2->data || |
27a884dc | 1355 | skb2->network_header > skb2->tail) { |
1da177e4 LT |
1356 | if (net_ratelimit()) |
1357 | printk(KERN_CRIT "protocol %04x is " | |
1358 | "buggy, dev %s\n", | |
1359 | skb2->protocol, dev->name); | |
c1d2bbe1 | 1360 | skb_reset_network_header(skb2); |
1da177e4 LT |
1361 | } |
1362 | ||
b0e380b1 | 1363 | skb2->transport_header = skb2->network_header; |
1da177e4 | 1364 | skb2->pkt_type = PACKET_OUTGOING; |
f2ccd8fa | 1365 | ptype->func(skb2, skb->dev, ptype, skb->dev); |
1da177e4 LT |
1366 | } |
1367 | } | |
1368 | rcu_read_unlock(); | |
1369 | } | |
1370 | ||
56079431 | 1371 | |
def82a1d | 1372 | static inline void __netif_reschedule(struct Qdisc *q) |
56079431 | 1373 | { |
def82a1d JP |
1374 | struct softnet_data *sd; |
1375 | unsigned long flags; | |
56079431 | 1376 | |
def82a1d JP |
1377 | local_irq_save(flags); |
1378 | sd = &__get_cpu_var(softnet_data); | |
1379 | q->next_sched = sd->output_queue; | |
1380 | sd->output_queue = q; | |
1381 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
1382 | local_irq_restore(flags); | |
1383 | } | |
1384 | ||
1385 | void __netif_schedule(struct Qdisc *q) | |
1386 | { | |
1387 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
1388 | __netif_reschedule(q); | |
56079431 DV |
1389 | } |
1390 | EXPORT_SYMBOL(__netif_schedule); | |
1391 | ||
bea3348e | 1392 | void dev_kfree_skb_irq(struct sk_buff *skb) |
56079431 | 1393 | { |
bea3348e SH |
1394 | if (atomic_dec_and_test(&skb->users)) { |
1395 | struct softnet_data *sd; | |
1396 | unsigned long flags; | |
56079431 | 1397 | |
bea3348e SH |
1398 | local_irq_save(flags); |
1399 | sd = &__get_cpu_var(softnet_data); | |
1400 | skb->next = sd->completion_queue; | |
1401 | sd->completion_queue = skb; | |
1402 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
1403 | local_irq_restore(flags); | |
1404 | } | |
56079431 | 1405 | } |
bea3348e | 1406 | EXPORT_SYMBOL(dev_kfree_skb_irq); |
56079431 DV |
1407 | |
1408 | void dev_kfree_skb_any(struct sk_buff *skb) | |
1409 | { | |
1410 | if (in_irq() || irqs_disabled()) | |
1411 | dev_kfree_skb_irq(skb); | |
1412 | else | |
1413 | dev_kfree_skb(skb); | |
1414 | } | |
1415 | EXPORT_SYMBOL(dev_kfree_skb_any); | |
1416 | ||
1417 | ||
bea3348e SH |
1418 | /** |
1419 | * netif_device_detach - mark device as removed | |
1420 | * @dev: network device | |
1421 | * | |
1422 | * Mark device as removed from system and therefore no longer available. | |
1423 | */ | |
56079431 DV |
1424 | void netif_device_detach(struct net_device *dev) |
1425 | { | |
1426 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
1427 | netif_running(dev)) { | |
1428 | netif_stop_queue(dev); | |
1429 | } | |
1430 | } | |
1431 | EXPORT_SYMBOL(netif_device_detach); | |
1432 | ||
bea3348e SH |
1433 | /** |
1434 | * netif_device_attach - mark device as attached | |
1435 | * @dev: network device | |
1436 | * | |
1437 | * Mark device as attached from system and restart if needed. | |
1438 | */ | |
56079431 DV |
1439 | void netif_device_attach(struct net_device *dev) |
1440 | { | |
1441 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
1442 | netif_running(dev)) { | |
1443 | netif_wake_queue(dev); | |
4ec93edb | 1444 | __netdev_watchdog_up(dev); |
56079431 DV |
1445 | } |
1446 | } | |
1447 | EXPORT_SYMBOL(netif_device_attach); | |
1448 | ||
6de329e2 BH |
1449 | static bool can_checksum_protocol(unsigned long features, __be16 protocol) |
1450 | { | |
1451 | return ((features & NETIF_F_GEN_CSUM) || | |
1452 | ((features & NETIF_F_IP_CSUM) && | |
1453 | protocol == htons(ETH_P_IP)) || | |
1454 | ((features & NETIF_F_IPV6_CSUM) && | |
1455 | protocol == htons(ETH_P_IPV6))); | |
1456 | } | |
1457 | ||
1458 | static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb) | |
1459 | { | |
1460 | if (can_checksum_protocol(dev->features, skb->protocol)) | |
1461 | return true; | |
1462 | ||
1463 | if (skb->protocol == htons(ETH_P_8021Q)) { | |
1464 | struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data; | |
1465 | if (can_checksum_protocol(dev->features & dev->vlan_features, | |
1466 | veh->h_vlan_encapsulated_proto)) | |
1467 | return true; | |
1468 | } | |
1469 | ||
1470 | return false; | |
1471 | } | |
56079431 | 1472 | |
1da177e4 LT |
1473 | /* |
1474 | * Invalidate hardware checksum when packet is to be mangled, and | |
1475 | * complete checksum manually on outgoing path. | |
1476 | */ | |
84fa7933 | 1477 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 1478 | { |
d3bc23e7 | 1479 | __wsum csum; |
663ead3b | 1480 | int ret = 0, offset; |
1da177e4 | 1481 | |
84fa7933 | 1482 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
1483 | goto out_set_summed; |
1484 | ||
1485 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
a430a43d HX |
1486 | /* Let GSO fix up the checksum. */ |
1487 | goto out_set_summed; | |
1da177e4 LT |
1488 | } |
1489 | ||
a030847e HX |
1490 | offset = skb->csum_start - skb_headroom(skb); |
1491 | BUG_ON(offset >= skb_headlen(skb)); | |
1492 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
1493 | ||
1494 | offset += skb->csum_offset; | |
1495 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
1496 | ||
1497 | if (skb_cloned(skb) && | |
1498 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1da177e4 LT |
1499 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
1500 | if (ret) | |
1501 | goto out; | |
1502 | } | |
1503 | ||
a030847e | 1504 | *(__sum16 *)(skb->data + offset) = csum_fold(csum); |
a430a43d | 1505 | out_set_summed: |
1da177e4 | 1506 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 1507 | out: |
1da177e4 LT |
1508 | return ret; |
1509 | } | |
1510 | ||
f6a78bfc HX |
1511 | /** |
1512 | * skb_gso_segment - Perform segmentation on skb. | |
1513 | * @skb: buffer to segment | |
576a30eb | 1514 | * @features: features for the output path (see dev->features) |
f6a78bfc HX |
1515 | * |
1516 | * This function segments the given skb and returns a list of segments. | |
576a30eb HX |
1517 | * |
1518 | * It may return NULL if the skb requires no segmentation. This is | |
1519 | * only possible when GSO is used for verifying header integrity. | |
f6a78bfc | 1520 | */ |
576a30eb | 1521 | struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features) |
f6a78bfc HX |
1522 | { |
1523 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
1524 | struct packet_type *ptype; | |
252e3346 | 1525 | __be16 type = skb->protocol; |
a430a43d | 1526 | int err; |
f6a78bfc HX |
1527 | |
1528 | BUG_ON(skb_shinfo(skb)->frag_list); | |
f6a78bfc | 1529 | |
459a98ed | 1530 | skb_reset_mac_header(skb); |
b0e380b1 | 1531 | skb->mac_len = skb->network_header - skb->mac_header; |
f6a78bfc HX |
1532 | __skb_pull(skb, skb->mac_len); |
1533 | ||
f9d106a6 | 1534 | if (WARN_ON(skb->ip_summed != CHECKSUM_PARTIAL)) { |
a430a43d HX |
1535 | if (skb_header_cloned(skb) && |
1536 | (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))) | |
1537 | return ERR_PTR(err); | |
1538 | } | |
1539 | ||
f6a78bfc | 1540 | rcu_read_lock(); |
82d8a867 PE |
1541 | list_for_each_entry_rcu(ptype, |
1542 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
f6a78bfc | 1543 | if (ptype->type == type && !ptype->dev && ptype->gso_segment) { |
84fa7933 | 1544 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { |
a430a43d HX |
1545 | err = ptype->gso_send_check(skb); |
1546 | segs = ERR_PTR(err); | |
1547 | if (err || skb_gso_ok(skb, features)) | |
1548 | break; | |
d56f90a7 ACM |
1549 | __skb_push(skb, (skb->data - |
1550 | skb_network_header(skb))); | |
a430a43d | 1551 | } |
576a30eb | 1552 | segs = ptype->gso_segment(skb, features); |
f6a78bfc HX |
1553 | break; |
1554 | } | |
1555 | } | |
1556 | rcu_read_unlock(); | |
1557 | ||
98e399f8 | 1558 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 1559 | |
f6a78bfc HX |
1560 | return segs; |
1561 | } | |
1562 | ||
1563 | EXPORT_SYMBOL(skb_gso_segment); | |
1564 | ||
fb286bb2 HX |
1565 | /* Take action when hardware reception checksum errors are detected. */ |
1566 | #ifdef CONFIG_BUG | |
1567 | void netdev_rx_csum_fault(struct net_device *dev) | |
1568 | { | |
1569 | if (net_ratelimit()) { | |
4ec93edb | 1570 | printk(KERN_ERR "%s: hw csum failure.\n", |
246a4212 | 1571 | dev ? dev->name : "<unknown>"); |
fb286bb2 HX |
1572 | dump_stack(); |
1573 | } | |
1574 | } | |
1575 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
1576 | #endif | |
1577 | ||
1da177e4 LT |
1578 | /* Actually, we should eliminate this check as soon as we know, that: |
1579 | * 1. IOMMU is present and allows to map all the memory. | |
1580 | * 2. No high memory really exists on this machine. | |
1581 | */ | |
1582 | ||
1583 | static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb) | |
1584 | { | |
3d3a8533 | 1585 | #ifdef CONFIG_HIGHMEM |
1da177e4 LT |
1586 | int i; |
1587 | ||
1588 | if (dev->features & NETIF_F_HIGHDMA) | |
1589 | return 0; | |
1590 | ||
1591 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) | |
1592 | if (PageHighMem(skb_shinfo(skb)->frags[i].page)) | |
1593 | return 1; | |
1594 | ||
3d3a8533 | 1595 | #endif |
1da177e4 LT |
1596 | return 0; |
1597 | } | |
1da177e4 | 1598 | |
f6a78bfc HX |
1599 | struct dev_gso_cb { |
1600 | void (*destructor)(struct sk_buff *skb); | |
1601 | }; | |
1602 | ||
1603 | #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb) | |
1604 | ||
1605 | static void dev_gso_skb_destructor(struct sk_buff *skb) | |
1606 | { | |
1607 | struct dev_gso_cb *cb; | |
1608 | ||
1609 | do { | |
1610 | struct sk_buff *nskb = skb->next; | |
1611 | ||
1612 | skb->next = nskb->next; | |
1613 | nskb->next = NULL; | |
1614 | kfree_skb(nskb); | |
1615 | } while (skb->next); | |
1616 | ||
1617 | cb = DEV_GSO_CB(skb); | |
1618 | if (cb->destructor) | |
1619 | cb->destructor(skb); | |
1620 | } | |
1621 | ||
1622 | /** | |
1623 | * dev_gso_segment - Perform emulated hardware segmentation on skb. | |
1624 | * @skb: buffer to segment | |
1625 | * | |
1626 | * This function segments the given skb and stores the list of segments | |
1627 | * in skb->next. | |
1628 | */ | |
1629 | static int dev_gso_segment(struct sk_buff *skb) | |
1630 | { | |
1631 | struct net_device *dev = skb->dev; | |
1632 | struct sk_buff *segs; | |
576a30eb HX |
1633 | int features = dev->features & ~(illegal_highdma(dev, skb) ? |
1634 | NETIF_F_SG : 0); | |
1635 | ||
1636 | segs = skb_gso_segment(skb, features); | |
1637 | ||
1638 | /* Verifying header integrity only. */ | |
1639 | if (!segs) | |
1640 | return 0; | |
f6a78bfc | 1641 | |
801678c5 | 1642 | if (IS_ERR(segs)) |
f6a78bfc HX |
1643 | return PTR_ERR(segs); |
1644 | ||
1645 | skb->next = segs; | |
1646 | DEV_GSO_CB(skb)->destructor = skb->destructor; | |
1647 | skb->destructor = dev_gso_skb_destructor; | |
1648 | ||
1649 | return 0; | |
1650 | } | |
1651 | ||
fd2ea0a7 DM |
1652 | int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev, |
1653 | struct netdev_queue *txq) | |
f6a78bfc HX |
1654 | { |
1655 | if (likely(!skb->next)) { | |
9be9a6b9 | 1656 | if (!list_empty(&ptype_all)) |
f6a78bfc HX |
1657 | dev_queue_xmit_nit(skb, dev); |
1658 | ||
576a30eb HX |
1659 | if (netif_needs_gso(dev, skb)) { |
1660 | if (unlikely(dev_gso_segment(skb))) | |
1661 | goto out_kfree_skb; | |
1662 | if (skb->next) | |
1663 | goto gso; | |
1664 | } | |
f6a78bfc | 1665 | |
576a30eb | 1666 | return dev->hard_start_xmit(skb, dev); |
f6a78bfc HX |
1667 | } |
1668 | ||
576a30eb | 1669 | gso: |
f6a78bfc HX |
1670 | do { |
1671 | struct sk_buff *nskb = skb->next; | |
1672 | int rc; | |
1673 | ||
1674 | skb->next = nskb->next; | |
1675 | nskb->next = NULL; | |
1676 | rc = dev->hard_start_xmit(nskb, dev); | |
1677 | if (unlikely(rc)) { | |
f54d9e8d | 1678 | nskb->next = skb->next; |
f6a78bfc HX |
1679 | skb->next = nskb; |
1680 | return rc; | |
1681 | } | |
fd2ea0a7 | 1682 | if (unlikely(netif_tx_queue_stopped(txq) && skb->next)) |
f54d9e8d | 1683 | return NETDEV_TX_BUSY; |
f6a78bfc | 1684 | } while (skb->next); |
4ec93edb | 1685 | |
f6a78bfc HX |
1686 | skb->destructor = DEV_GSO_CB(skb)->destructor; |
1687 | ||
1688 | out_kfree_skb: | |
1689 | kfree_skb(skb); | |
1690 | return 0; | |
1691 | } | |
1692 | ||
b6b2fed1 DM |
1693 | static u32 simple_tx_hashrnd; |
1694 | static int simple_tx_hashrnd_initialized = 0; | |
1695 | ||
8f0f2223 DM |
1696 | static u16 simple_tx_hash(struct net_device *dev, struct sk_buff *skb) |
1697 | { | |
b6b2fed1 DM |
1698 | u32 addr1, addr2, ports; |
1699 | u32 hash, ihl; | |
8f0f2223 | 1700 | u8 ip_proto; |
b6b2fed1 DM |
1701 | |
1702 | if (unlikely(!simple_tx_hashrnd_initialized)) { | |
1703 | get_random_bytes(&simple_tx_hashrnd, 4); | |
1704 | simple_tx_hashrnd_initialized = 1; | |
1705 | } | |
8f0f2223 DM |
1706 | |
1707 | switch (skb->protocol) { | |
60678040 | 1708 | case htons(ETH_P_IP): |
8f0f2223 | 1709 | ip_proto = ip_hdr(skb)->protocol; |
b6b2fed1 DM |
1710 | addr1 = ip_hdr(skb)->saddr; |
1711 | addr2 = ip_hdr(skb)->daddr; | |
8f0f2223 | 1712 | ihl = ip_hdr(skb)->ihl; |
8f0f2223 | 1713 | break; |
60678040 | 1714 | case htons(ETH_P_IPV6): |
8f0f2223 | 1715 | ip_proto = ipv6_hdr(skb)->nexthdr; |
b6b2fed1 DM |
1716 | addr1 = ipv6_hdr(skb)->saddr.s6_addr32[3]; |
1717 | addr2 = ipv6_hdr(skb)->daddr.s6_addr32[3]; | |
8f0f2223 | 1718 | ihl = (40 >> 2); |
8f0f2223 DM |
1719 | break; |
1720 | default: | |
1721 | return 0; | |
1722 | } | |
1723 | ||
8f0f2223 DM |
1724 | |
1725 | switch (ip_proto) { | |
1726 | case IPPROTO_TCP: | |
1727 | case IPPROTO_UDP: | |
1728 | case IPPROTO_DCCP: | |
1729 | case IPPROTO_ESP: | |
1730 | case IPPROTO_AH: | |
1731 | case IPPROTO_SCTP: | |
1732 | case IPPROTO_UDPLITE: | |
b6b2fed1 | 1733 | ports = *((u32 *) (skb_network_header(skb) + (ihl * 4))); |
8f0f2223 DM |
1734 | break; |
1735 | ||
1736 | default: | |
b6b2fed1 | 1737 | ports = 0; |
8f0f2223 DM |
1738 | break; |
1739 | } | |
1740 | ||
b6b2fed1 DM |
1741 | hash = jhash_3words(addr1, addr2, ports, simple_tx_hashrnd); |
1742 | ||
1743 | return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32); | |
8f0f2223 DM |
1744 | } |
1745 | ||
e8a0464c DM |
1746 | static struct netdev_queue *dev_pick_tx(struct net_device *dev, |
1747 | struct sk_buff *skb) | |
1748 | { | |
fd2ea0a7 DM |
1749 | u16 queue_index = 0; |
1750 | ||
eae792b7 DM |
1751 | if (dev->select_queue) |
1752 | queue_index = dev->select_queue(dev, skb); | |
8f0f2223 DM |
1753 | else if (dev->real_num_tx_queues > 1) |
1754 | queue_index = simple_tx_hash(dev, skb); | |
eae792b7 | 1755 | |
fd2ea0a7 DM |
1756 | skb_set_queue_mapping(skb, queue_index); |
1757 | return netdev_get_tx_queue(dev, queue_index); | |
e8a0464c DM |
1758 | } |
1759 | ||
d29f749e DJ |
1760 | /** |
1761 | * dev_queue_xmit - transmit a buffer | |
1762 | * @skb: buffer to transmit | |
1763 | * | |
1764 | * Queue a buffer for transmission to a network device. The caller must | |
1765 | * have set the device and priority and built the buffer before calling | |
1766 | * this function. The function can be called from an interrupt. | |
1767 | * | |
1768 | * A negative errno code is returned on a failure. A success does not | |
1769 | * guarantee the frame will be transmitted as it may be dropped due | |
1770 | * to congestion or traffic shaping. | |
1771 | * | |
1772 | * ----------------------------------------------------------------------------------- | |
1773 | * I notice this method can also return errors from the queue disciplines, | |
1774 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
1775 | * be positive. | |
1776 | * | |
1777 | * Regardless of the return value, the skb is consumed, so it is currently | |
1778 | * difficult to retry a send to this method. (You can bump the ref count | |
1779 | * before sending to hold a reference for retry if you are careful.) | |
1780 | * | |
1781 | * When calling this method, interrupts MUST be enabled. This is because | |
1782 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
1783 | * --BLG | |
1784 | */ | |
1da177e4 LT |
1785 | int dev_queue_xmit(struct sk_buff *skb) |
1786 | { | |
1787 | struct net_device *dev = skb->dev; | |
dc2b4847 | 1788 | struct netdev_queue *txq; |
1da177e4 LT |
1789 | struct Qdisc *q; |
1790 | int rc = -ENOMEM; | |
1791 | ||
f6a78bfc HX |
1792 | /* GSO will handle the following emulations directly. */ |
1793 | if (netif_needs_gso(dev, skb)) | |
1794 | goto gso; | |
1795 | ||
1da177e4 LT |
1796 | if (skb_shinfo(skb)->frag_list && |
1797 | !(dev->features & NETIF_F_FRAGLIST) && | |
364c6bad | 1798 | __skb_linearize(skb)) |
1da177e4 LT |
1799 | goto out_kfree_skb; |
1800 | ||
1801 | /* Fragmented skb is linearized if device does not support SG, | |
1802 | * or if at least one of fragments is in highmem and device | |
1803 | * does not support DMA from it. | |
1804 | */ | |
1805 | if (skb_shinfo(skb)->nr_frags && | |
1806 | (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) && | |
364c6bad | 1807 | __skb_linearize(skb)) |
1da177e4 LT |
1808 | goto out_kfree_skb; |
1809 | ||
1810 | /* If packet is not checksummed and device does not support | |
1811 | * checksumming for this protocol, complete checksumming here. | |
1812 | */ | |
663ead3b HX |
1813 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
1814 | skb_set_transport_header(skb, skb->csum_start - | |
1815 | skb_headroom(skb)); | |
6de329e2 BH |
1816 | if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb)) |
1817 | goto out_kfree_skb; | |
663ead3b | 1818 | } |
1da177e4 | 1819 | |
f6a78bfc | 1820 | gso: |
4ec93edb YH |
1821 | /* Disable soft irqs for various locks below. Also |
1822 | * stops preemption for RCU. | |
1da177e4 | 1823 | */ |
4ec93edb | 1824 | rcu_read_lock_bh(); |
1da177e4 | 1825 | |
eae792b7 | 1826 | txq = dev_pick_tx(dev, skb); |
b0e1e646 | 1827 | q = rcu_dereference(txq->qdisc); |
37437bb2 | 1828 | |
1da177e4 LT |
1829 | #ifdef CONFIG_NET_CLS_ACT |
1830 | skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS); | |
1831 | #endif | |
1832 | if (q->enqueue) { | |
5fb66229 | 1833 | spinlock_t *root_lock = qdisc_lock(q); |
37437bb2 DM |
1834 | |
1835 | spin_lock(root_lock); | |
1836 | ||
a9312ae8 | 1837 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { |
96d20316 | 1838 | kfree_skb(skb); |
a9312ae8 | 1839 | rc = NET_XMIT_DROP; |
96d20316 DM |
1840 | } else { |
1841 | rc = qdisc_enqueue_root(skb, q); | |
1842 | qdisc_run(q); | |
a9312ae8 | 1843 | } |
37437bb2 DM |
1844 | spin_unlock(root_lock); |
1845 | ||
37437bb2 | 1846 | goto out; |
1da177e4 LT |
1847 | } |
1848 | ||
1849 | /* The device has no queue. Common case for software devices: | |
1850 | loopback, all the sorts of tunnels... | |
1851 | ||
932ff279 HX |
1852 | Really, it is unlikely that netif_tx_lock protection is necessary |
1853 | here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
1da177e4 LT |
1854 | counters.) |
1855 | However, it is possible, that they rely on protection | |
1856 | made by us here. | |
1857 | ||
1858 | Check this and shot the lock. It is not prone from deadlocks. | |
1859 | Either shot noqueue qdisc, it is even simpler 8) | |
1860 | */ | |
1861 | if (dev->flags & IFF_UP) { | |
1862 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
1863 | ||
c773e847 | 1864 | if (txq->xmit_lock_owner != cpu) { |
1da177e4 | 1865 | |
c773e847 | 1866 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 1867 | |
fd2ea0a7 | 1868 | if (!netif_tx_queue_stopped(txq)) { |
1da177e4 | 1869 | rc = 0; |
fd2ea0a7 | 1870 | if (!dev_hard_start_xmit(skb, dev, txq)) { |
c773e847 | 1871 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
1872 | goto out; |
1873 | } | |
1874 | } | |
c773e847 | 1875 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
1876 | if (net_ratelimit()) |
1877 | printk(KERN_CRIT "Virtual device %s asks to " | |
1878 | "queue packet!\n", dev->name); | |
1879 | } else { | |
1880 | /* Recursion is detected! It is possible, | |
1881 | * unfortunately */ | |
1882 | if (net_ratelimit()) | |
1883 | printk(KERN_CRIT "Dead loop on virtual device " | |
1884 | "%s, fix it urgently!\n", dev->name); | |
1885 | } | |
1886 | } | |
1887 | ||
1888 | rc = -ENETDOWN; | |
d4828d85 | 1889 | rcu_read_unlock_bh(); |
1da177e4 LT |
1890 | |
1891 | out_kfree_skb: | |
1892 | kfree_skb(skb); | |
1893 | return rc; | |
1894 | out: | |
d4828d85 | 1895 | rcu_read_unlock_bh(); |
1da177e4 LT |
1896 | return rc; |
1897 | } | |
1898 | ||
1899 | ||
1900 | /*======================================================================= | |
1901 | Receiver routines | |
1902 | =======================================================================*/ | |
1903 | ||
6b2bedc3 SH |
1904 | int netdev_max_backlog __read_mostly = 1000; |
1905 | int netdev_budget __read_mostly = 300; | |
1906 | int weight_p __read_mostly = 64; /* old backlog weight */ | |
1da177e4 LT |
1907 | |
1908 | DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, }; | |
1909 | ||
1910 | ||
1da177e4 LT |
1911 | /** |
1912 | * netif_rx - post buffer to the network code | |
1913 | * @skb: buffer to post | |
1914 | * | |
1915 | * This function receives a packet from a device driver and queues it for | |
1916 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
1917 | * may be dropped during processing for congestion control or by the | |
1918 | * protocol layers. | |
1919 | * | |
1920 | * return values: | |
1921 | * NET_RX_SUCCESS (no congestion) | |
1da177e4 LT |
1922 | * NET_RX_DROP (packet was dropped) |
1923 | * | |
1924 | */ | |
1925 | ||
1926 | int netif_rx(struct sk_buff *skb) | |
1927 | { | |
1da177e4 LT |
1928 | struct softnet_data *queue; |
1929 | unsigned long flags; | |
1930 | ||
1931 | /* if netpoll wants it, pretend we never saw it */ | |
1932 | if (netpoll_rx(skb)) | |
1933 | return NET_RX_DROP; | |
1934 | ||
b7aa0bf7 | 1935 | if (!skb->tstamp.tv64) |
a61bbcf2 | 1936 | net_timestamp(skb); |
1da177e4 LT |
1937 | |
1938 | /* | |
1939 | * The code is rearranged so that the path is the most | |
1940 | * short when CPU is congested, but is still operating. | |
1941 | */ | |
1942 | local_irq_save(flags); | |
1da177e4 LT |
1943 | queue = &__get_cpu_var(softnet_data); |
1944 | ||
1945 | __get_cpu_var(netdev_rx_stat).total++; | |
1946 | if (queue->input_pkt_queue.qlen <= netdev_max_backlog) { | |
1947 | if (queue->input_pkt_queue.qlen) { | |
1da177e4 | 1948 | enqueue: |
1da177e4 | 1949 | __skb_queue_tail(&queue->input_pkt_queue, skb); |
1da177e4 | 1950 | local_irq_restore(flags); |
34008d8c | 1951 | return NET_RX_SUCCESS; |
1da177e4 LT |
1952 | } |
1953 | ||
bea3348e | 1954 | napi_schedule(&queue->backlog); |
1da177e4 LT |
1955 | goto enqueue; |
1956 | } | |
1957 | ||
1da177e4 LT |
1958 | __get_cpu_var(netdev_rx_stat).dropped++; |
1959 | local_irq_restore(flags); | |
1960 | ||
1961 | kfree_skb(skb); | |
1962 | return NET_RX_DROP; | |
1963 | } | |
1964 | ||
1965 | int netif_rx_ni(struct sk_buff *skb) | |
1966 | { | |
1967 | int err; | |
1968 | ||
1969 | preempt_disable(); | |
1970 | err = netif_rx(skb); | |
1971 | if (local_softirq_pending()) | |
1972 | do_softirq(); | |
1973 | preempt_enable(); | |
1974 | ||
1975 | return err; | |
1976 | } | |
1977 | ||
1978 | EXPORT_SYMBOL(netif_rx_ni); | |
1979 | ||
1da177e4 LT |
1980 | static void net_tx_action(struct softirq_action *h) |
1981 | { | |
1982 | struct softnet_data *sd = &__get_cpu_var(softnet_data); | |
1983 | ||
1984 | if (sd->completion_queue) { | |
1985 | struct sk_buff *clist; | |
1986 | ||
1987 | local_irq_disable(); | |
1988 | clist = sd->completion_queue; | |
1989 | sd->completion_queue = NULL; | |
1990 | local_irq_enable(); | |
1991 | ||
1992 | while (clist) { | |
1993 | struct sk_buff *skb = clist; | |
1994 | clist = clist->next; | |
1995 | ||
547b792c | 1996 | WARN_ON(atomic_read(&skb->users)); |
1da177e4 LT |
1997 | __kfree_skb(skb); |
1998 | } | |
1999 | } | |
2000 | ||
2001 | if (sd->output_queue) { | |
37437bb2 | 2002 | struct Qdisc *head; |
1da177e4 LT |
2003 | |
2004 | local_irq_disable(); | |
2005 | head = sd->output_queue; | |
2006 | sd->output_queue = NULL; | |
2007 | local_irq_enable(); | |
2008 | ||
2009 | while (head) { | |
37437bb2 DM |
2010 | struct Qdisc *q = head; |
2011 | spinlock_t *root_lock; | |
2012 | ||
1da177e4 LT |
2013 | head = head->next_sched; |
2014 | ||
5fb66229 | 2015 | root_lock = qdisc_lock(q); |
37437bb2 | 2016 | if (spin_trylock(root_lock)) { |
def82a1d JP |
2017 | smp_mb__before_clear_bit(); |
2018 | clear_bit(__QDISC_STATE_SCHED, | |
2019 | &q->state); | |
37437bb2 DM |
2020 | qdisc_run(q); |
2021 | spin_unlock(root_lock); | |
1da177e4 | 2022 | } else { |
195648bb | 2023 | if (!test_bit(__QDISC_STATE_DEACTIVATED, |
e8a83e10 | 2024 | &q->state)) { |
195648bb | 2025 | __netif_reschedule(q); |
e8a83e10 JP |
2026 | } else { |
2027 | smp_mb__before_clear_bit(); | |
2028 | clear_bit(__QDISC_STATE_SCHED, | |
2029 | &q->state); | |
2030 | } | |
1da177e4 LT |
2031 | } |
2032 | } | |
2033 | } | |
2034 | } | |
2035 | ||
6f05f629 SH |
2036 | static inline int deliver_skb(struct sk_buff *skb, |
2037 | struct packet_type *pt_prev, | |
2038 | struct net_device *orig_dev) | |
1da177e4 LT |
2039 | { |
2040 | atomic_inc(&skb->users); | |
f2ccd8fa | 2041 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
1da177e4 LT |
2042 | } |
2043 | ||
2044 | #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE) | |
6229e362 | 2045 | /* These hooks defined here for ATM */ |
1da177e4 LT |
2046 | struct net_bridge; |
2047 | struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br, | |
2048 | unsigned char *addr); | |
6229e362 | 2049 | void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent) __read_mostly; |
1da177e4 | 2050 | |
6229e362 SH |
2051 | /* |
2052 | * If bridge module is loaded call bridging hook. | |
2053 | * returns NULL if packet was consumed. | |
2054 | */ | |
2055 | struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p, | |
2056 | struct sk_buff *skb) __read_mostly; | |
2057 | static inline struct sk_buff *handle_bridge(struct sk_buff *skb, | |
2058 | struct packet_type **pt_prev, int *ret, | |
2059 | struct net_device *orig_dev) | |
1da177e4 LT |
2060 | { |
2061 | struct net_bridge_port *port; | |
2062 | ||
6229e362 SH |
2063 | if (skb->pkt_type == PACKET_LOOPBACK || |
2064 | (port = rcu_dereference(skb->dev->br_port)) == NULL) | |
2065 | return skb; | |
1da177e4 LT |
2066 | |
2067 | if (*pt_prev) { | |
6229e362 | 2068 | *ret = deliver_skb(skb, *pt_prev, orig_dev); |
1da177e4 | 2069 | *pt_prev = NULL; |
4ec93edb YH |
2070 | } |
2071 | ||
6229e362 | 2072 | return br_handle_frame_hook(port, skb); |
1da177e4 LT |
2073 | } |
2074 | #else | |
6229e362 | 2075 | #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb) |
1da177e4 LT |
2076 | #endif |
2077 | ||
b863ceb7 PM |
2078 | #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE) |
2079 | struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly; | |
2080 | EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook); | |
2081 | ||
2082 | static inline struct sk_buff *handle_macvlan(struct sk_buff *skb, | |
2083 | struct packet_type **pt_prev, | |
2084 | int *ret, | |
2085 | struct net_device *orig_dev) | |
2086 | { | |
2087 | if (skb->dev->macvlan_port == NULL) | |
2088 | return skb; | |
2089 | ||
2090 | if (*pt_prev) { | |
2091 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
2092 | *pt_prev = NULL; | |
2093 | } | |
2094 | return macvlan_handle_frame_hook(skb); | |
2095 | } | |
2096 | #else | |
2097 | #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb) | |
2098 | #endif | |
2099 | ||
1da177e4 LT |
2100 | #ifdef CONFIG_NET_CLS_ACT |
2101 | /* TODO: Maybe we should just force sch_ingress to be compiled in | |
2102 | * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions | |
2103 | * a compare and 2 stores extra right now if we dont have it on | |
2104 | * but have CONFIG_NET_CLS_ACT | |
4ec93edb | 2105 | * NOTE: This doesnt stop any functionality; if you dont have |
1da177e4 LT |
2106 | * the ingress scheduler, you just cant add policies on ingress. |
2107 | * | |
2108 | */ | |
4ec93edb | 2109 | static int ing_filter(struct sk_buff *skb) |
1da177e4 | 2110 | { |
1da177e4 | 2111 | struct net_device *dev = skb->dev; |
f697c3e8 | 2112 | u32 ttl = G_TC_RTTL(skb->tc_verd); |
555353cf DM |
2113 | struct netdev_queue *rxq; |
2114 | int result = TC_ACT_OK; | |
2115 | struct Qdisc *q; | |
4ec93edb | 2116 | |
f697c3e8 HX |
2117 | if (MAX_RED_LOOP < ttl++) { |
2118 | printk(KERN_WARNING | |
2119 | "Redir loop detected Dropping packet (%d->%d)\n", | |
2120 | skb->iif, dev->ifindex); | |
2121 | return TC_ACT_SHOT; | |
2122 | } | |
1da177e4 | 2123 | |
f697c3e8 HX |
2124 | skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl); |
2125 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); | |
1da177e4 | 2126 | |
555353cf DM |
2127 | rxq = &dev->rx_queue; |
2128 | ||
83874000 | 2129 | q = rxq->qdisc; |
8d50b53d | 2130 | if (q != &noop_qdisc) { |
83874000 | 2131 | spin_lock(qdisc_lock(q)); |
a9312ae8 DM |
2132 | if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) |
2133 | result = qdisc_enqueue_root(skb, q); | |
83874000 DM |
2134 | spin_unlock(qdisc_lock(q)); |
2135 | } | |
f697c3e8 HX |
2136 | |
2137 | return result; | |
2138 | } | |
86e65da9 | 2139 | |
f697c3e8 HX |
2140 | static inline struct sk_buff *handle_ing(struct sk_buff *skb, |
2141 | struct packet_type **pt_prev, | |
2142 | int *ret, struct net_device *orig_dev) | |
2143 | { | |
8d50b53d | 2144 | if (skb->dev->rx_queue.qdisc == &noop_qdisc) |
f697c3e8 | 2145 | goto out; |
1da177e4 | 2146 | |
f697c3e8 HX |
2147 | if (*pt_prev) { |
2148 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
2149 | *pt_prev = NULL; | |
2150 | } else { | |
2151 | /* Huh? Why does turning on AF_PACKET affect this? */ | |
2152 | skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd); | |
1da177e4 LT |
2153 | } |
2154 | ||
f697c3e8 HX |
2155 | switch (ing_filter(skb)) { |
2156 | case TC_ACT_SHOT: | |
2157 | case TC_ACT_STOLEN: | |
2158 | kfree_skb(skb); | |
2159 | return NULL; | |
2160 | } | |
2161 | ||
2162 | out: | |
2163 | skb->tc_verd = 0; | |
2164 | return skb; | |
1da177e4 LT |
2165 | } |
2166 | #endif | |
2167 | ||
bc1d0411 PM |
2168 | /* |
2169 | * netif_nit_deliver - deliver received packets to network taps | |
2170 | * @skb: buffer | |
2171 | * | |
2172 | * This function is used to deliver incoming packets to network | |
2173 | * taps. It should be used when the normal netif_receive_skb path | |
2174 | * is bypassed, for example because of VLAN acceleration. | |
2175 | */ | |
2176 | void netif_nit_deliver(struct sk_buff *skb) | |
2177 | { | |
2178 | struct packet_type *ptype; | |
2179 | ||
2180 | if (list_empty(&ptype_all)) | |
2181 | return; | |
2182 | ||
2183 | skb_reset_network_header(skb); | |
2184 | skb_reset_transport_header(skb); | |
2185 | skb->mac_len = skb->network_header - skb->mac_header; | |
2186 | ||
2187 | rcu_read_lock(); | |
2188 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
2189 | if (!ptype->dev || ptype->dev == skb->dev) | |
2190 | deliver_skb(skb, ptype, skb->dev); | |
2191 | } | |
2192 | rcu_read_unlock(); | |
2193 | } | |
2194 | ||
3b582cc1 SH |
2195 | /** |
2196 | * netif_receive_skb - process receive buffer from network | |
2197 | * @skb: buffer to process | |
2198 | * | |
2199 | * netif_receive_skb() is the main receive data processing function. | |
2200 | * It always succeeds. The buffer may be dropped during processing | |
2201 | * for congestion control or by the protocol layers. | |
2202 | * | |
2203 | * This function may only be called from softirq context and interrupts | |
2204 | * should be enabled. | |
2205 | * | |
2206 | * Return values (usually ignored): | |
2207 | * NET_RX_SUCCESS: no congestion | |
2208 | * NET_RX_DROP: packet was dropped | |
2209 | */ | |
1da177e4 LT |
2210 | int netif_receive_skb(struct sk_buff *skb) |
2211 | { | |
2212 | struct packet_type *ptype, *pt_prev; | |
f2ccd8fa | 2213 | struct net_device *orig_dev; |
0d7a3681 | 2214 | struct net_device *null_or_orig; |
1da177e4 | 2215 | int ret = NET_RX_DROP; |
252e3346 | 2216 | __be16 type; |
1da177e4 LT |
2217 | |
2218 | /* if we've gotten here through NAPI, check netpoll */ | |
bea3348e | 2219 | if (netpoll_receive_skb(skb)) |
1da177e4 LT |
2220 | return NET_RX_DROP; |
2221 | ||
b7aa0bf7 | 2222 | if (!skb->tstamp.tv64) |
a61bbcf2 | 2223 | net_timestamp(skb); |
1da177e4 | 2224 | |
c01003c2 PM |
2225 | if (!skb->iif) |
2226 | skb->iif = skb->dev->ifindex; | |
86e65da9 | 2227 | |
0d7a3681 | 2228 | null_or_orig = NULL; |
cc9bd5ce JE |
2229 | orig_dev = skb->dev; |
2230 | if (orig_dev->master) { | |
0d7a3681 JE |
2231 | if (skb_bond_should_drop(skb)) |
2232 | null_or_orig = orig_dev; /* deliver only exact match */ | |
2233 | else | |
2234 | skb->dev = orig_dev->master; | |
cc9bd5ce | 2235 | } |
8f903c70 | 2236 | |
1da177e4 LT |
2237 | __get_cpu_var(netdev_rx_stat).total++; |
2238 | ||
c1d2bbe1 | 2239 | skb_reset_network_header(skb); |
badff6d0 | 2240 | skb_reset_transport_header(skb); |
b0e380b1 | 2241 | skb->mac_len = skb->network_header - skb->mac_header; |
1da177e4 LT |
2242 | |
2243 | pt_prev = NULL; | |
2244 | ||
2245 | rcu_read_lock(); | |
2246 | ||
b9f75f45 EB |
2247 | /* Don't receive packets in an exiting network namespace */ |
2248 | if (!net_alive(dev_net(skb->dev))) | |
2249 | goto out; | |
2250 | ||
1da177e4 LT |
2251 | #ifdef CONFIG_NET_CLS_ACT |
2252 | if (skb->tc_verd & TC_NCLS) { | |
2253 | skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); | |
2254 | goto ncls; | |
2255 | } | |
2256 | #endif | |
2257 | ||
2258 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
f982307f JE |
2259 | if (ptype->dev == null_or_orig || ptype->dev == skb->dev || |
2260 | ptype->dev == orig_dev) { | |
4ec93edb | 2261 | if (pt_prev) |
f2ccd8fa | 2262 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
2263 | pt_prev = ptype; |
2264 | } | |
2265 | } | |
2266 | ||
2267 | #ifdef CONFIG_NET_CLS_ACT | |
f697c3e8 HX |
2268 | skb = handle_ing(skb, &pt_prev, &ret, orig_dev); |
2269 | if (!skb) | |
1da177e4 | 2270 | goto out; |
1da177e4 LT |
2271 | ncls: |
2272 | #endif | |
2273 | ||
6229e362 | 2274 | skb = handle_bridge(skb, &pt_prev, &ret, orig_dev); |
b863ceb7 PM |
2275 | if (!skb) |
2276 | goto out; | |
2277 | skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev); | |
6229e362 | 2278 | if (!skb) |
1da177e4 LT |
2279 | goto out; |
2280 | ||
2281 | type = skb->protocol; | |
82d8a867 PE |
2282 | list_for_each_entry_rcu(ptype, |
2283 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
1da177e4 | 2284 | if (ptype->type == type && |
f982307f JE |
2285 | (ptype->dev == null_or_orig || ptype->dev == skb->dev || |
2286 | ptype->dev == orig_dev)) { | |
4ec93edb | 2287 | if (pt_prev) |
f2ccd8fa | 2288 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
2289 | pt_prev = ptype; |
2290 | } | |
2291 | } | |
2292 | ||
2293 | if (pt_prev) { | |
f2ccd8fa | 2294 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
1da177e4 LT |
2295 | } else { |
2296 | kfree_skb(skb); | |
2297 | /* Jamal, now you will not able to escape explaining | |
2298 | * me how you were going to use this. :-) | |
2299 | */ | |
2300 | ret = NET_RX_DROP; | |
2301 | } | |
2302 | ||
2303 | out: | |
2304 | rcu_read_unlock(); | |
2305 | return ret; | |
2306 | } | |
2307 | ||
6e583ce5 SH |
2308 | /* Network device is going away, flush any packets still pending */ |
2309 | static void flush_backlog(void *arg) | |
2310 | { | |
2311 | struct net_device *dev = arg; | |
2312 | struct softnet_data *queue = &__get_cpu_var(softnet_data); | |
2313 | struct sk_buff *skb, *tmp; | |
2314 | ||
2315 | skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp) | |
2316 | if (skb->dev == dev) { | |
2317 | __skb_unlink(skb, &queue->input_pkt_queue); | |
2318 | kfree_skb(skb); | |
2319 | } | |
2320 | } | |
2321 | ||
bea3348e | 2322 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 LT |
2323 | { |
2324 | int work = 0; | |
1da177e4 LT |
2325 | struct softnet_data *queue = &__get_cpu_var(softnet_data); |
2326 | unsigned long start_time = jiffies; | |
2327 | ||
bea3348e SH |
2328 | napi->weight = weight_p; |
2329 | do { | |
1da177e4 | 2330 | struct sk_buff *skb; |
1da177e4 LT |
2331 | |
2332 | local_irq_disable(); | |
2333 | skb = __skb_dequeue(&queue->input_pkt_queue); | |
bea3348e SH |
2334 | if (!skb) { |
2335 | __napi_complete(napi); | |
2336 | local_irq_enable(); | |
2337 | break; | |
2338 | } | |
1da177e4 LT |
2339 | local_irq_enable(); |
2340 | ||
1da177e4 | 2341 | netif_receive_skb(skb); |
bea3348e | 2342 | } while (++work < quota && jiffies == start_time); |
1da177e4 | 2343 | |
bea3348e SH |
2344 | return work; |
2345 | } | |
1da177e4 | 2346 | |
bea3348e SH |
2347 | /** |
2348 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 2349 | * @n: entry to schedule |
bea3348e SH |
2350 | * |
2351 | * The entry's receive function will be scheduled to run | |
2352 | */ | |
b5606c2d | 2353 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
2354 | { |
2355 | unsigned long flags; | |
1da177e4 | 2356 | |
bea3348e SH |
2357 | local_irq_save(flags); |
2358 | list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list); | |
2359 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2360 | local_irq_restore(flags); | |
1da177e4 | 2361 | } |
bea3348e SH |
2362 | EXPORT_SYMBOL(__napi_schedule); |
2363 | ||
1da177e4 LT |
2364 | |
2365 | static void net_rx_action(struct softirq_action *h) | |
2366 | { | |
bea3348e | 2367 | struct list_head *list = &__get_cpu_var(softnet_data).poll_list; |
1da177e4 | 2368 | unsigned long start_time = jiffies; |
51b0bded | 2369 | int budget = netdev_budget; |
53fb95d3 MM |
2370 | void *have; |
2371 | ||
1da177e4 LT |
2372 | local_irq_disable(); |
2373 | ||
bea3348e SH |
2374 | while (!list_empty(list)) { |
2375 | struct napi_struct *n; | |
2376 | int work, weight; | |
1da177e4 | 2377 | |
bea3348e SH |
2378 | /* If softirq window is exhuasted then punt. |
2379 | * | |
2380 | * Note that this is a slight policy change from the | |
2381 | * previous NAPI code, which would allow up to 2 | |
2382 | * jiffies to pass before breaking out. The test | |
2383 | * used to be "jiffies - start_time > 1". | |
2384 | */ | |
2385 | if (unlikely(budget <= 0 || jiffies != start_time)) | |
1da177e4 LT |
2386 | goto softnet_break; |
2387 | ||
2388 | local_irq_enable(); | |
2389 | ||
bea3348e SH |
2390 | /* Even though interrupts have been re-enabled, this |
2391 | * access is safe because interrupts can only add new | |
2392 | * entries to the tail of this list, and only ->poll() | |
2393 | * calls can remove this head entry from the list. | |
2394 | */ | |
2395 | n = list_entry(list->next, struct napi_struct, poll_list); | |
1da177e4 | 2396 | |
bea3348e SH |
2397 | have = netpoll_poll_lock(n); |
2398 | ||
2399 | weight = n->weight; | |
2400 | ||
0a7606c1 DM |
2401 | /* This NAPI_STATE_SCHED test is for avoiding a race |
2402 | * with netpoll's poll_napi(). Only the entity which | |
2403 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
2404 | * actually make the ->poll() call. Therefore we avoid | |
2405 | * accidently calling ->poll() when NAPI is not scheduled. | |
2406 | */ | |
2407 | work = 0; | |
2408 | if (test_bit(NAPI_STATE_SCHED, &n->state)) | |
2409 | work = n->poll(n, weight); | |
bea3348e SH |
2410 | |
2411 | WARN_ON_ONCE(work > weight); | |
2412 | ||
2413 | budget -= work; | |
2414 | ||
2415 | local_irq_disable(); | |
2416 | ||
2417 | /* Drivers must not modify the NAPI state if they | |
2418 | * consume the entire weight. In such cases this code | |
2419 | * still "owns" the NAPI instance and therefore can | |
2420 | * move the instance around on the list at-will. | |
2421 | */ | |
fed17f30 DM |
2422 | if (unlikely(work == weight)) { |
2423 | if (unlikely(napi_disable_pending(n))) | |
2424 | __napi_complete(n); | |
2425 | else | |
2426 | list_move_tail(&n->poll_list, list); | |
2427 | } | |
bea3348e SH |
2428 | |
2429 | netpoll_poll_unlock(have); | |
1da177e4 LT |
2430 | } |
2431 | out: | |
515e06c4 | 2432 | local_irq_enable(); |
bea3348e | 2433 | |
db217334 CL |
2434 | #ifdef CONFIG_NET_DMA |
2435 | /* | |
2436 | * There may not be any more sk_buffs coming right now, so push | |
2437 | * any pending DMA copies to hardware | |
2438 | */ | |
d379b01e DW |
2439 | if (!cpus_empty(net_dma.channel_mask)) { |
2440 | int chan_idx; | |
0e12f848 | 2441 | for_each_cpu_mask_nr(chan_idx, net_dma.channel_mask) { |
d379b01e DW |
2442 | struct dma_chan *chan = net_dma.channels[chan_idx]; |
2443 | if (chan) | |
2444 | dma_async_memcpy_issue_pending(chan); | |
2445 | } | |
db217334 CL |
2446 | } |
2447 | #endif | |
bea3348e | 2448 | |
1da177e4 LT |
2449 | return; |
2450 | ||
2451 | softnet_break: | |
2452 | __get_cpu_var(netdev_rx_stat).time_squeeze++; | |
2453 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2454 | goto out; | |
2455 | } | |
2456 | ||
2457 | static gifconf_func_t * gifconf_list [NPROTO]; | |
2458 | ||
2459 | /** | |
2460 | * register_gifconf - register a SIOCGIF handler | |
2461 | * @family: Address family | |
2462 | * @gifconf: Function handler | |
2463 | * | |
2464 | * Register protocol dependent address dumping routines. The handler | |
2465 | * that is passed must not be freed or reused until it has been replaced | |
2466 | * by another handler. | |
2467 | */ | |
2468 | int register_gifconf(unsigned int family, gifconf_func_t * gifconf) | |
2469 | { | |
2470 | if (family >= NPROTO) | |
2471 | return -EINVAL; | |
2472 | gifconf_list[family] = gifconf; | |
2473 | return 0; | |
2474 | } | |
2475 | ||
2476 | ||
2477 | /* | |
2478 | * Map an interface index to its name (SIOCGIFNAME) | |
2479 | */ | |
2480 | ||
2481 | /* | |
2482 | * We need this ioctl for efficient implementation of the | |
2483 | * if_indextoname() function required by the IPv6 API. Without | |
2484 | * it, we would have to search all the interfaces to find a | |
2485 | * match. --pb | |
2486 | */ | |
2487 | ||
881d966b | 2488 | static int dev_ifname(struct net *net, struct ifreq __user *arg) |
1da177e4 LT |
2489 | { |
2490 | struct net_device *dev; | |
2491 | struct ifreq ifr; | |
2492 | ||
2493 | /* | |
2494 | * Fetch the caller's info block. | |
2495 | */ | |
2496 | ||
2497 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
2498 | return -EFAULT; | |
2499 | ||
2500 | read_lock(&dev_base_lock); | |
881d966b | 2501 | dev = __dev_get_by_index(net, ifr.ifr_ifindex); |
1da177e4 LT |
2502 | if (!dev) { |
2503 | read_unlock(&dev_base_lock); | |
2504 | return -ENODEV; | |
2505 | } | |
2506 | ||
2507 | strcpy(ifr.ifr_name, dev->name); | |
2508 | read_unlock(&dev_base_lock); | |
2509 | ||
2510 | if (copy_to_user(arg, &ifr, sizeof(struct ifreq))) | |
2511 | return -EFAULT; | |
2512 | return 0; | |
2513 | } | |
2514 | ||
2515 | /* | |
2516 | * Perform a SIOCGIFCONF call. This structure will change | |
2517 | * size eventually, and there is nothing I can do about it. | |
2518 | * Thus we will need a 'compatibility mode'. | |
2519 | */ | |
2520 | ||
881d966b | 2521 | static int dev_ifconf(struct net *net, char __user *arg) |
1da177e4 LT |
2522 | { |
2523 | struct ifconf ifc; | |
2524 | struct net_device *dev; | |
2525 | char __user *pos; | |
2526 | int len; | |
2527 | int total; | |
2528 | int i; | |
2529 | ||
2530 | /* | |
2531 | * Fetch the caller's info block. | |
2532 | */ | |
2533 | ||
2534 | if (copy_from_user(&ifc, arg, sizeof(struct ifconf))) | |
2535 | return -EFAULT; | |
2536 | ||
2537 | pos = ifc.ifc_buf; | |
2538 | len = ifc.ifc_len; | |
2539 | ||
2540 | /* | |
2541 | * Loop over the interfaces, and write an info block for each. | |
2542 | */ | |
2543 | ||
2544 | total = 0; | |
881d966b | 2545 | for_each_netdev(net, dev) { |
1da177e4 LT |
2546 | for (i = 0; i < NPROTO; i++) { |
2547 | if (gifconf_list[i]) { | |
2548 | int done; | |
2549 | if (!pos) | |
2550 | done = gifconf_list[i](dev, NULL, 0); | |
2551 | else | |
2552 | done = gifconf_list[i](dev, pos + total, | |
2553 | len - total); | |
2554 | if (done < 0) | |
2555 | return -EFAULT; | |
2556 | total += done; | |
2557 | } | |
2558 | } | |
4ec93edb | 2559 | } |
1da177e4 LT |
2560 | |
2561 | /* | |
2562 | * All done. Write the updated control block back to the caller. | |
2563 | */ | |
2564 | ifc.ifc_len = total; | |
2565 | ||
2566 | /* | |
2567 | * Both BSD and Solaris return 0 here, so we do too. | |
2568 | */ | |
2569 | return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0; | |
2570 | } | |
2571 | ||
2572 | #ifdef CONFIG_PROC_FS | |
2573 | /* | |
2574 | * This is invoked by the /proc filesystem handler to display a device | |
2575 | * in detail. | |
2576 | */ | |
7562f876 | 2577 | void *dev_seq_start(struct seq_file *seq, loff_t *pos) |
9a429c49 | 2578 | __acquires(dev_base_lock) |
1da177e4 | 2579 | { |
e372c414 | 2580 | struct net *net = seq_file_net(seq); |
7562f876 | 2581 | loff_t off; |
1da177e4 | 2582 | struct net_device *dev; |
1da177e4 | 2583 | |
7562f876 PE |
2584 | read_lock(&dev_base_lock); |
2585 | if (!*pos) | |
2586 | return SEQ_START_TOKEN; | |
1da177e4 | 2587 | |
7562f876 | 2588 | off = 1; |
881d966b | 2589 | for_each_netdev(net, dev) |
7562f876 PE |
2590 | if (off++ == *pos) |
2591 | return dev; | |
1da177e4 | 2592 | |
7562f876 | 2593 | return NULL; |
1da177e4 LT |
2594 | } |
2595 | ||
2596 | void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
2597 | { | |
e372c414 | 2598 | struct net *net = seq_file_net(seq); |
1da177e4 | 2599 | ++*pos; |
7562f876 | 2600 | return v == SEQ_START_TOKEN ? |
881d966b | 2601 | first_net_device(net) : next_net_device((struct net_device *)v); |
1da177e4 LT |
2602 | } |
2603 | ||
2604 | void dev_seq_stop(struct seq_file *seq, void *v) | |
9a429c49 | 2605 | __releases(dev_base_lock) |
1da177e4 LT |
2606 | { |
2607 | read_unlock(&dev_base_lock); | |
2608 | } | |
2609 | ||
2610 | static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev) | |
2611 | { | |
c45d286e | 2612 | struct net_device_stats *stats = dev->get_stats(dev); |
1da177e4 | 2613 | |
5a1b5898 RR |
2614 | seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu " |
2615 | "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n", | |
2616 | dev->name, stats->rx_bytes, stats->rx_packets, | |
2617 | stats->rx_errors, | |
2618 | stats->rx_dropped + stats->rx_missed_errors, | |
2619 | stats->rx_fifo_errors, | |
2620 | stats->rx_length_errors + stats->rx_over_errors + | |
2621 | stats->rx_crc_errors + stats->rx_frame_errors, | |
2622 | stats->rx_compressed, stats->multicast, | |
2623 | stats->tx_bytes, stats->tx_packets, | |
2624 | stats->tx_errors, stats->tx_dropped, | |
2625 | stats->tx_fifo_errors, stats->collisions, | |
2626 | stats->tx_carrier_errors + | |
2627 | stats->tx_aborted_errors + | |
2628 | stats->tx_window_errors + | |
2629 | stats->tx_heartbeat_errors, | |
2630 | stats->tx_compressed); | |
1da177e4 LT |
2631 | } |
2632 | ||
2633 | /* | |
2634 | * Called from the PROCfs module. This now uses the new arbitrary sized | |
2635 | * /proc/net interface to create /proc/net/dev | |
2636 | */ | |
2637 | static int dev_seq_show(struct seq_file *seq, void *v) | |
2638 | { | |
2639 | if (v == SEQ_START_TOKEN) | |
2640 | seq_puts(seq, "Inter-| Receive " | |
2641 | " | Transmit\n" | |
2642 | " face |bytes packets errs drop fifo frame " | |
2643 | "compressed multicast|bytes packets errs " | |
2644 | "drop fifo colls carrier compressed\n"); | |
2645 | else | |
2646 | dev_seq_printf_stats(seq, v); | |
2647 | return 0; | |
2648 | } | |
2649 | ||
2650 | static struct netif_rx_stats *softnet_get_online(loff_t *pos) | |
2651 | { | |
2652 | struct netif_rx_stats *rc = NULL; | |
2653 | ||
0c0b0aca | 2654 | while (*pos < nr_cpu_ids) |
4ec93edb | 2655 | if (cpu_online(*pos)) { |
1da177e4 LT |
2656 | rc = &per_cpu(netdev_rx_stat, *pos); |
2657 | break; | |
2658 | } else | |
2659 | ++*pos; | |
2660 | return rc; | |
2661 | } | |
2662 | ||
2663 | static void *softnet_seq_start(struct seq_file *seq, loff_t *pos) | |
2664 | { | |
2665 | return softnet_get_online(pos); | |
2666 | } | |
2667 | ||
2668 | static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
2669 | { | |
2670 | ++*pos; | |
2671 | return softnet_get_online(pos); | |
2672 | } | |
2673 | ||
2674 | static void softnet_seq_stop(struct seq_file *seq, void *v) | |
2675 | { | |
2676 | } | |
2677 | ||
2678 | static int softnet_seq_show(struct seq_file *seq, void *v) | |
2679 | { | |
2680 | struct netif_rx_stats *s = v; | |
2681 | ||
2682 | seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n", | |
31aa02c5 | 2683 | s->total, s->dropped, s->time_squeeze, 0, |
c1ebcdb8 SH |
2684 | 0, 0, 0, 0, /* was fastroute */ |
2685 | s->cpu_collision ); | |
1da177e4 LT |
2686 | return 0; |
2687 | } | |
2688 | ||
f690808e | 2689 | static const struct seq_operations dev_seq_ops = { |
1da177e4 LT |
2690 | .start = dev_seq_start, |
2691 | .next = dev_seq_next, | |
2692 | .stop = dev_seq_stop, | |
2693 | .show = dev_seq_show, | |
2694 | }; | |
2695 | ||
2696 | static int dev_seq_open(struct inode *inode, struct file *file) | |
2697 | { | |
e372c414 DL |
2698 | return seq_open_net(inode, file, &dev_seq_ops, |
2699 | sizeof(struct seq_net_private)); | |
1da177e4 LT |
2700 | } |
2701 | ||
9a32144e | 2702 | static const struct file_operations dev_seq_fops = { |
1da177e4 LT |
2703 | .owner = THIS_MODULE, |
2704 | .open = dev_seq_open, | |
2705 | .read = seq_read, | |
2706 | .llseek = seq_lseek, | |
e372c414 | 2707 | .release = seq_release_net, |
1da177e4 LT |
2708 | }; |
2709 | ||
f690808e | 2710 | static const struct seq_operations softnet_seq_ops = { |
1da177e4 LT |
2711 | .start = softnet_seq_start, |
2712 | .next = softnet_seq_next, | |
2713 | .stop = softnet_seq_stop, | |
2714 | .show = softnet_seq_show, | |
2715 | }; | |
2716 | ||
2717 | static int softnet_seq_open(struct inode *inode, struct file *file) | |
2718 | { | |
2719 | return seq_open(file, &softnet_seq_ops); | |
2720 | } | |
2721 | ||
9a32144e | 2722 | static const struct file_operations softnet_seq_fops = { |
1da177e4 LT |
2723 | .owner = THIS_MODULE, |
2724 | .open = softnet_seq_open, | |
2725 | .read = seq_read, | |
2726 | .llseek = seq_lseek, | |
2727 | .release = seq_release, | |
2728 | }; | |
2729 | ||
0e1256ff SH |
2730 | static void *ptype_get_idx(loff_t pos) |
2731 | { | |
2732 | struct packet_type *pt = NULL; | |
2733 | loff_t i = 0; | |
2734 | int t; | |
2735 | ||
2736 | list_for_each_entry_rcu(pt, &ptype_all, list) { | |
2737 | if (i == pos) | |
2738 | return pt; | |
2739 | ++i; | |
2740 | } | |
2741 | ||
82d8a867 | 2742 | for (t = 0; t < PTYPE_HASH_SIZE; t++) { |
0e1256ff SH |
2743 | list_for_each_entry_rcu(pt, &ptype_base[t], list) { |
2744 | if (i == pos) | |
2745 | return pt; | |
2746 | ++i; | |
2747 | } | |
2748 | } | |
2749 | return NULL; | |
2750 | } | |
2751 | ||
2752 | static void *ptype_seq_start(struct seq_file *seq, loff_t *pos) | |
72348a42 | 2753 | __acquires(RCU) |
0e1256ff SH |
2754 | { |
2755 | rcu_read_lock(); | |
2756 | return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN; | |
2757 | } | |
2758 | ||
2759 | static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
2760 | { | |
2761 | struct packet_type *pt; | |
2762 | struct list_head *nxt; | |
2763 | int hash; | |
2764 | ||
2765 | ++*pos; | |
2766 | if (v == SEQ_START_TOKEN) | |
2767 | return ptype_get_idx(0); | |
2768 | ||
2769 | pt = v; | |
2770 | nxt = pt->list.next; | |
2771 | if (pt->type == htons(ETH_P_ALL)) { | |
2772 | if (nxt != &ptype_all) | |
2773 | goto found; | |
2774 | hash = 0; | |
2775 | nxt = ptype_base[0].next; | |
2776 | } else | |
82d8a867 | 2777 | hash = ntohs(pt->type) & PTYPE_HASH_MASK; |
0e1256ff SH |
2778 | |
2779 | while (nxt == &ptype_base[hash]) { | |
82d8a867 | 2780 | if (++hash >= PTYPE_HASH_SIZE) |
0e1256ff SH |
2781 | return NULL; |
2782 | nxt = ptype_base[hash].next; | |
2783 | } | |
2784 | found: | |
2785 | return list_entry(nxt, struct packet_type, list); | |
2786 | } | |
2787 | ||
2788 | static void ptype_seq_stop(struct seq_file *seq, void *v) | |
72348a42 | 2789 | __releases(RCU) |
0e1256ff SH |
2790 | { |
2791 | rcu_read_unlock(); | |
2792 | } | |
2793 | ||
2794 | static void ptype_seq_decode(struct seq_file *seq, void *sym) | |
2795 | { | |
2796 | #ifdef CONFIG_KALLSYMS | |
2797 | unsigned long offset = 0, symsize; | |
2798 | const char *symname; | |
2799 | char *modname; | |
2800 | char namebuf[128]; | |
2801 | ||
2802 | symname = kallsyms_lookup((unsigned long)sym, &symsize, &offset, | |
2803 | &modname, namebuf); | |
2804 | ||
2805 | if (symname) { | |
2806 | char *delim = ":"; | |
2807 | ||
2808 | if (!modname) | |
2809 | modname = delim = ""; | |
2810 | seq_printf(seq, "%s%s%s%s+0x%lx", delim, modname, delim, | |
2811 | symname, offset); | |
2812 | return; | |
2813 | } | |
2814 | #endif | |
2815 | ||
2816 | seq_printf(seq, "[%p]", sym); | |
2817 | } | |
2818 | ||
2819 | static int ptype_seq_show(struct seq_file *seq, void *v) | |
2820 | { | |
2821 | struct packet_type *pt = v; | |
2822 | ||
2823 | if (v == SEQ_START_TOKEN) | |
2824 | seq_puts(seq, "Type Device Function\n"); | |
c346dca1 | 2825 | else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) { |
0e1256ff SH |
2826 | if (pt->type == htons(ETH_P_ALL)) |
2827 | seq_puts(seq, "ALL "); | |
2828 | else | |
2829 | seq_printf(seq, "%04x", ntohs(pt->type)); | |
2830 | ||
2831 | seq_printf(seq, " %-8s ", | |
2832 | pt->dev ? pt->dev->name : ""); | |
2833 | ptype_seq_decode(seq, pt->func); | |
2834 | seq_putc(seq, '\n'); | |
2835 | } | |
2836 | ||
2837 | return 0; | |
2838 | } | |
2839 | ||
2840 | static const struct seq_operations ptype_seq_ops = { | |
2841 | .start = ptype_seq_start, | |
2842 | .next = ptype_seq_next, | |
2843 | .stop = ptype_seq_stop, | |
2844 | .show = ptype_seq_show, | |
2845 | }; | |
2846 | ||
2847 | static int ptype_seq_open(struct inode *inode, struct file *file) | |
2848 | { | |
2feb27db PE |
2849 | return seq_open_net(inode, file, &ptype_seq_ops, |
2850 | sizeof(struct seq_net_private)); | |
0e1256ff SH |
2851 | } |
2852 | ||
2853 | static const struct file_operations ptype_seq_fops = { | |
2854 | .owner = THIS_MODULE, | |
2855 | .open = ptype_seq_open, | |
2856 | .read = seq_read, | |
2857 | .llseek = seq_lseek, | |
2feb27db | 2858 | .release = seq_release_net, |
0e1256ff SH |
2859 | }; |
2860 | ||
2861 | ||
4665079c | 2862 | static int __net_init dev_proc_net_init(struct net *net) |
1da177e4 LT |
2863 | { |
2864 | int rc = -ENOMEM; | |
2865 | ||
881d966b | 2866 | if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops)) |
1da177e4 | 2867 | goto out; |
881d966b | 2868 | if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops)) |
1da177e4 | 2869 | goto out_dev; |
881d966b | 2870 | if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops)) |
457c4cbc | 2871 | goto out_softnet; |
0e1256ff | 2872 | |
881d966b | 2873 | if (wext_proc_init(net)) |
457c4cbc | 2874 | goto out_ptype; |
1da177e4 LT |
2875 | rc = 0; |
2876 | out: | |
2877 | return rc; | |
457c4cbc | 2878 | out_ptype: |
881d966b | 2879 | proc_net_remove(net, "ptype"); |
1da177e4 | 2880 | out_softnet: |
881d966b | 2881 | proc_net_remove(net, "softnet_stat"); |
1da177e4 | 2882 | out_dev: |
881d966b | 2883 | proc_net_remove(net, "dev"); |
1da177e4 LT |
2884 | goto out; |
2885 | } | |
881d966b | 2886 | |
4665079c | 2887 | static void __net_exit dev_proc_net_exit(struct net *net) |
881d966b EB |
2888 | { |
2889 | wext_proc_exit(net); | |
2890 | ||
2891 | proc_net_remove(net, "ptype"); | |
2892 | proc_net_remove(net, "softnet_stat"); | |
2893 | proc_net_remove(net, "dev"); | |
2894 | } | |
2895 | ||
022cbae6 | 2896 | static struct pernet_operations __net_initdata dev_proc_ops = { |
881d966b EB |
2897 | .init = dev_proc_net_init, |
2898 | .exit = dev_proc_net_exit, | |
2899 | }; | |
2900 | ||
2901 | static int __init dev_proc_init(void) | |
2902 | { | |
2903 | return register_pernet_subsys(&dev_proc_ops); | |
2904 | } | |
1da177e4 LT |
2905 | #else |
2906 | #define dev_proc_init() 0 | |
2907 | #endif /* CONFIG_PROC_FS */ | |
2908 | ||
2909 | ||
2910 | /** | |
2911 | * netdev_set_master - set up master/slave pair | |
2912 | * @slave: slave device | |
2913 | * @master: new master device | |
2914 | * | |
2915 | * Changes the master device of the slave. Pass %NULL to break the | |
2916 | * bonding. The caller must hold the RTNL semaphore. On a failure | |
2917 | * a negative errno code is returned. On success the reference counts | |
2918 | * are adjusted, %RTM_NEWLINK is sent to the routing socket and the | |
2919 | * function returns zero. | |
2920 | */ | |
2921 | int netdev_set_master(struct net_device *slave, struct net_device *master) | |
2922 | { | |
2923 | struct net_device *old = slave->master; | |
2924 | ||
2925 | ASSERT_RTNL(); | |
2926 | ||
2927 | if (master) { | |
2928 | if (old) | |
2929 | return -EBUSY; | |
2930 | dev_hold(master); | |
2931 | } | |
2932 | ||
2933 | slave->master = master; | |
4ec93edb | 2934 | |
1da177e4 LT |
2935 | synchronize_net(); |
2936 | ||
2937 | if (old) | |
2938 | dev_put(old); | |
2939 | ||
2940 | if (master) | |
2941 | slave->flags |= IFF_SLAVE; | |
2942 | else | |
2943 | slave->flags &= ~IFF_SLAVE; | |
2944 | ||
2945 | rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE); | |
2946 | return 0; | |
2947 | } | |
2948 | ||
dad9b335 | 2949 | static int __dev_set_promiscuity(struct net_device *dev, int inc) |
1da177e4 LT |
2950 | { |
2951 | unsigned short old_flags = dev->flags; | |
2952 | ||
24023451 PM |
2953 | ASSERT_RTNL(); |
2954 | ||
dad9b335 WC |
2955 | dev->flags |= IFF_PROMISC; |
2956 | dev->promiscuity += inc; | |
2957 | if (dev->promiscuity == 0) { | |
2958 | /* | |
2959 | * Avoid overflow. | |
2960 | * If inc causes overflow, untouch promisc and return error. | |
2961 | */ | |
2962 | if (inc < 0) | |
2963 | dev->flags &= ~IFF_PROMISC; | |
2964 | else { | |
2965 | dev->promiscuity -= inc; | |
2966 | printk(KERN_WARNING "%s: promiscuity touches roof, " | |
2967 | "set promiscuity failed, promiscuity feature " | |
2968 | "of device might be broken.\n", dev->name); | |
2969 | return -EOVERFLOW; | |
2970 | } | |
2971 | } | |
52609c0b | 2972 | if (dev->flags != old_flags) { |
1da177e4 LT |
2973 | printk(KERN_INFO "device %s %s promiscuous mode\n", |
2974 | dev->name, (dev->flags & IFF_PROMISC) ? "entered" : | |
4ec93edb | 2975 | "left"); |
7759db82 KHK |
2976 | if (audit_enabled) |
2977 | audit_log(current->audit_context, GFP_ATOMIC, | |
2978 | AUDIT_ANOM_PROMISCUOUS, | |
2979 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
2980 | dev->name, (dev->flags & IFF_PROMISC), | |
2981 | (old_flags & IFF_PROMISC), | |
2982 | audit_get_loginuid(current), | |
2983 | current->uid, current->gid, | |
2984 | audit_get_sessionid(current)); | |
24023451 PM |
2985 | |
2986 | if (dev->change_rx_flags) | |
2987 | dev->change_rx_flags(dev, IFF_PROMISC); | |
1da177e4 | 2988 | } |
dad9b335 | 2989 | return 0; |
1da177e4 LT |
2990 | } |
2991 | ||
4417da66 PM |
2992 | /** |
2993 | * dev_set_promiscuity - update promiscuity count on a device | |
2994 | * @dev: device | |
2995 | * @inc: modifier | |
2996 | * | |
2997 | * Add or remove promiscuity from a device. While the count in the device | |
2998 | * remains above zero the interface remains promiscuous. Once it hits zero | |
2999 | * the device reverts back to normal filtering operation. A negative inc | |
3000 | * value is used to drop promiscuity on the device. | |
dad9b335 | 3001 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 3002 | */ |
dad9b335 | 3003 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 PM |
3004 | { |
3005 | unsigned short old_flags = dev->flags; | |
dad9b335 | 3006 | int err; |
4417da66 | 3007 | |
dad9b335 | 3008 | err = __dev_set_promiscuity(dev, inc); |
4b5a698e | 3009 | if (err < 0) |
dad9b335 | 3010 | return err; |
4417da66 PM |
3011 | if (dev->flags != old_flags) |
3012 | dev_set_rx_mode(dev); | |
dad9b335 | 3013 | return err; |
4417da66 PM |
3014 | } |
3015 | ||
1da177e4 LT |
3016 | /** |
3017 | * dev_set_allmulti - update allmulti count on a device | |
3018 | * @dev: device | |
3019 | * @inc: modifier | |
3020 | * | |
3021 | * Add or remove reception of all multicast frames to a device. While the | |
3022 | * count in the device remains above zero the interface remains listening | |
3023 | * to all interfaces. Once it hits zero the device reverts back to normal | |
3024 | * filtering operation. A negative @inc value is used to drop the counter | |
3025 | * when releasing a resource needing all multicasts. | |
dad9b335 | 3026 | * Return 0 if successful or a negative errno code on error. |
1da177e4 LT |
3027 | */ |
3028 | ||
dad9b335 | 3029 | int dev_set_allmulti(struct net_device *dev, int inc) |
1da177e4 LT |
3030 | { |
3031 | unsigned short old_flags = dev->flags; | |
3032 | ||
24023451 PM |
3033 | ASSERT_RTNL(); |
3034 | ||
1da177e4 | 3035 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
3036 | dev->allmulti += inc; |
3037 | if (dev->allmulti == 0) { | |
3038 | /* | |
3039 | * Avoid overflow. | |
3040 | * If inc causes overflow, untouch allmulti and return error. | |
3041 | */ | |
3042 | if (inc < 0) | |
3043 | dev->flags &= ~IFF_ALLMULTI; | |
3044 | else { | |
3045 | dev->allmulti -= inc; | |
3046 | printk(KERN_WARNING "%s: allmulti touches roof, " | |
3047 | "set allmulti failed, allmulti feature of " | |
3048 | "device might be broken.\n", dev->name); | |
3049 | return -EOVERFLOW; | |
3050 | } | |
3051 | } | |
24023451 PM |
3052 | if (dev->flags ^ old_flags) { |
3053 | if (dev->change_rx_flags) | |
3054 | dev->change_rx_flags(dev, IFF_ALLMULTI); | |
4417da66 | 3055 | dev_set_rx_mode(dev); |
24023451 | 3056 | } |
dad9b335 | 3057 | return 0; |
4417da66 PM |
3058 | } |
3059 | ||
3060 | /* | |
3061 | * Upload unicast and multicast address lists to device and | |
3062 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 3063 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
3064 | * are present. |
3065 | */ | |
3066 | void __dev_set_rx_mode(struct net_device *dev) | |
3067 | { | |
3068 | /* dev_open will call this function so the list will stay sane. */ | |
3069 | if (!(dev->flags&IFF_UP)) | |
3070 | return; | |
3071 | ||
3072 | if (!netif_device_present(dev)) | |
40b77c94 | 3073 | return; |
4417da66 PM |
3074 | |
3075 | if (dev->set_rx_mode) | |
3076 | dev->set_rx_mode(dev); | |
3077 | else { | |
3078 | /* Unicast addresses changes may only happen under the rtnl, | |
3079 | * therefore calling __dev_set_promiscuity here is safe. | |
3080 | */ | |
3081 | if (dev->uc_count > 0 && !dev->uc_promisc) { | |
3082 | __dev_set_promiscuity(dev, 1); | |
3083 | dev->uc_promisc = 1; | |
3084 | } else if (dev->uc_count == 0 && dev->uc_promisc) { | |
3085 | __dev_set_promiscuity(dev, -1); | |
3086 | dev->uc_promisc = 0; | |
3087 | } | |
3088 | ||
3089 | if (dev->set_multicast_list) | |
3090 | dev->set_multicast_list(dev); | |
3091 | } | |
3092 | } | |
3093 | ||
3094 | void dev_set_rx_mode(struct net_device *dev) | |
3095 | { | |
b9e40857 | 3096 | netif_addr_lock_bh(dev); |
4417da66 | 3097 | __dev_set_rx_mode(dev); |
b9e40857 | 3098 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
3099 | } |
3100 | ||
61cbc2fc PM |
3101 | int __dev_addr_delete(struct dev_addr_list **list, int *count, |
3102 | void *addr, int alen, int glbl) | |
bf742482 PM |
3103 | { |
3104 | struct dev_addr_list *da; | |
3105 | ||
3106 | for (; (da = *list) != NULL; list = &da->next) { | |
3107 | if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 && | |
3108 | alen == da->da_addrlen) { | |
3109 | if (glbl) { | |
3110 | int old_glbl = da->da_gusers; | |
3111 | da->da_gusers = 0; | |
3112 | if (old_glbl == 0) | |
3113 | break; | |
3114 | } | |
3115 | if (--da->da_users) | |
3116 | return 0; | |
3117 | ||
3118 | *list = da->next; | |
3119 | kfree(da); | |
61cbc2fc | 3120 | (*count)--; |
bf742482 PM |
3121 | return 0; |
3122 | } | |
3123 | } | |
3124 | return -ENOENT; | |
3125 | } | |
3126 | ||
61cbc2fc PM |
3127 | int __dev_addr_add(struct dev_addr_list **list, int *count, |
3128 | void *addr, int alen, int glbl) | |
bf742482 PM |
3129 | { |
3130 | struct dev_addr_list *da; | |
3131 | ||
3132 | for (da = *list; da != NULL; da = da->next) { | |
3133 | if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 && | |
3134 | da->da_addrlen == alen) { | |
3135 | if (glbl) { | |
3136 | int old_glbl = da->da_gusers; | |
3137 | da->da_gusers = 1; | |
3138 | if (old_glbl) | |
3139 | return 0; | |
3140 | } | |
3141 | da->da_users++; | |
3142 | return 0; | |
3143 | } | |
3144 | } | |
3145 | ||
12aa343a | 3146 | da = kzalloc(sizeof(*da), GFP_ATOMIC); |
bf742482 PM |
3147 | if (da == NULL) |
3148 | return -ENOMEM; | |
3149 | memcpy(da->da_addr, addr, alen); | |
3150 | da->da_addrlen = alen; | |
3151 | da->da_users = 1; | |
3152 | da->da_gusers = glbl ? 1 : 0; | |
3153 | da->next = *list; | |
3154 | *list = da; | |
61cbc2fc | 3155 | (*count)++; |
bf742482 PM |
3156 | return 0; |
3157 | } | |
3158 | ||
4417da66 PM |
3159 | /** |
3160 | * dev_unicast_delete - Release secondary unicast address. | |
3161 | * @dev: device | |
0ed72ec4 RD |
3162 | * @addr: address to delete |
3163 | * @alen: length of @addr | |
4417da66 PM |
3164 | * |
3165 | * Release reference to a secondary unicast address and remove it | |
0ed72ec4 | 3166 | * from the device if the reference count drops to zero. |
4417da66 PM |
3167 | * |
3168 | * The caller must hold the rtnl_mutex. | |
3169 | */ | |
3170 | int dev_unicast_delete(struct net_device *dev, void *addr, int alen) | |
3171 | { | |
3172 | int err; | |
3173 | ||
3174 | ASSERT_RTNL(); | |
3175 | ||
b9e40857 | 3176 | netif_addr_lock_bh(dev); |
61cbc2fc PM |
3177 | err = __dev_addr_delete(&dev->uc_list, &dev->uc_count, addr, alen, 0); |
3178 | if (!err) | |
4417da66 | 3179 | __dev_set_rx_mode(dev); |
b9e40857 | 3180 | netif_addr_unlock_bh(dev); |
4417da66 PM |
3181 | return err; |
3182 | } | |
3183 | EXPORT_SYMBOL(dev_unicast_delete); | |
3184 | ||
3185 | /** | |
3186 | * dev_unicast_add - add a secondary unicast address | |
3187 | * @dev: device | |
5dbaec5d | 3188 | * @addr: address to add |
0ed72ec4 | 3189 | * @alen: length of @addr |
4417da66 PM |
3190 | * |
3191 | * Add a secondary unicast address to the device or increase | |
3192 | * the reference count if it already exists. | |
3193 | * | |
3194 | * The caller must hold the rtnl_mutex. | |
3195 | */ | |
3196 | int dev_unicast_add(struct net_device *dev, void *addr, int alen) | |
3197 | { | |
3198 | int err; | |
3199 | ||
3200 | ASSERT_RTNL(); | |
3201 | ||
b9e40857 | 3202 | netif_addr_lock_bh(dev); |
61cbc2fc PM |
3203 | err = __dev_addr_add(&dev->uc_list, &dev->uc_count, addr, alen, 0); |
3204 | if (!err) | |
4417da66 | 3205 | __dev_set_rx_mode(dev); |
b9e40857 | 3206 | netif_addr_unlock_bh(dev); |
4417da66 PM |
3207 | return err; |
3208 | } | |
3209 | EXPORT_SYMBOL(dev_unicast_add); | |
3210 | ||
e83a2ea8 CL |
3211 | int __dev_addr_sync(struct dev_addr_list **to, int *to_count, |
3212 | struct dev_addr_list **from, int *from_count) | |
3213 | { | |
3214 | struct dev_addr_list *da, *next; | |
3215 | int err = 0; | |
3216 | ||
3217 | da = *from; | |
3218 | while (da != NULL) { | |
3219 | next = da->next; | |
3220 | if (!da->da_synced) { | |
3221 | err = __dev_addr_add(to, to_count, | |
3222 | da->da_addr, da->da_addrlen, 0); | |
3223 | if (err < 0) | |
3224 | break; | |
3225 | da->da_synced = 1; | |
3226 | da->da_users++; | |
3227 | } else if (da->da_users == 1) { | |
3228 | __dev_addr_delete(to, to_count, | |
3229 | da->da_addr, da->da_addrlen, 0); | |
3230 | __dev_addr_delete(from, from_count, | |
3231 | da->da_addr, da->da_addrlen, 0); | |
3232 | } | |
3233 | da = next; | |
3234 | } | |
3235 | return err; | |
3236 | } | |
3237 | ||
3238 | void __dev_addr_unsync(struct dev_addr_list **to, int *to_count, | |
3239 | struct dev_addr_list **from, int *from_count) | |
3240 | { | |
3241 | struct dev_addr_list *da, *next; | |
3242 | ||
3243 | da = *from; | |
3244 | while (da != NULL) { | |
3245 | next = da->next; | |
3246 | if (da->da_synced) { | |
3247 | __dev_addr_delete(to, to_count, | |
3248 | da->da_addr, da->da_addrlen, 0); | |
3249 | da->da_synced = 0; | |
3250 | __dev_addr_delete(from, from_count, | |
3251 | da->da_addr, da->da_addrlen, 0); | |
3252 | } | |
3253 | da = next; | |
3254 | } | |
3255 | } | |
3256 | ||
3257 | /** | |
3258 | * dev_unicast_sync - Synchronize device's unicast list to another device | |
3259 | * @to: destination device | |
3260 | * @from: source device | |
3261 | * | |
3262 | * Add newly added addresses to the destination device and release | |
3263 | * addresses that have no users left. The source device must be | |
3264 | * locked by netif_tx_lock_bh. | |
3265 | * | |
3266 | * This function is intended to be called from the dev->set_rx_mode | |
3267 | * function of layered software devices. | |
3268 | */ | |
3269 | int dev_unicast_sync(struct net_device *to, struct net_device *from) | |
3270 | { | |
3271 | int err = 0; | |
3272 | ||
b9e40857 | 3273 | netif_addr_lock_bh(to); |
e83a2ea8 CL |
3274 | err = __dev_addr_sync(&to->uc_list, &to->uc_count, |
3275 | &from->uc_list, &from->uc_count); | |
3276 | if (!err) | |
3277 | __dev_set_rx_mode(to); | |
b9e40857 | 3278 | netif_addr_unlock_bh(to); |
e83a2ea8 CL |
3279 | return err; |
3280 | } | |
3281 | EXPORT_SYMBOL(dev_unicast_sync); | |
3282 | ||
3283 | /** | |
bc2cda1e | 3284 | * dev_unicast_unsync - Remove synchronized addresses from the destination device |
e83a2ea8 CL |
3285 | * @to: destination device |
3286 | * @from: source device | |
3287 | * | |
3288 | * Remove all addresses that were added to the destination device by | |
3289 | * dev_unicast_sync(). This function is intended to be called from the | |
3290 | * dev->stop function of layered software devices. | |
3291 | */ | |
3292 | void dev_unicast_unsync(struct net_device *to, struct net_device *from) | |
3293 | { | |
b9e40857 | 3294 | netif_addr_lock_bh(from); |
e308a5d8 | 3295 | netif_addr_lock(to); |
e83a2ea8 CL |
3296 | |
3297 | __dev_addr_unsync(&to->uc_list, &to->uc_count, | |
3298 | &from->uc_list, &from->uc_count); | |
3299 | __dev_set_rx_mode(to); | |
3300 | ||
e308a5d8 | 3301 | netif_addr_unlock(to); |
b9e40857 | 3302 | netif_addr_unlock_bh(from); |
e83a2ea8 CL |
3303 | } |
3304 | EXPORT_SYMBOL(dev_unicast_unsync); | |
3305 | ||
12972621 DC |
3306 | static void __dev_addr_discard(struct dev_addr_list **list) |
3307 | { | |
3308 | struct dev_addr_list *tmp; | |
3309 | ||
3310 | while (*list != NULL) { | |
3311 | tmp = *list; | |
3312 | *list = tmp->next; | |
3313 | if (tmp->da_users > tmp->da_gusers) | |
3314 | printk("__dev_addr_discard: address leakage! " | |
3315 | "da_users=%d\n", tmp->da_users); | |
3316 | kfree(tmp); | |
3317 | } | |
3318 | } | |
3319 | ||
26cc2522 | 3320 | static void dev_addr_discard(struct net_device *dev) |
4417da66 | 3321 | { |
b9e40857 | 3322 | netif_addr_lock_bh(dev); |
26cc2522 | 3323 | |
4417da66 PM |
3324 | __dev_addr_discard(&dev->uc_list); |
3325 | dev->uc_count = 0; | |
4417da66 | 3326 | |
456ad75c DC |
3327 | __dev_addr_discard(&dev->mc_list); |
3328 | dev->mc_count = 0; | |
26cc2522 | 3329 | |
b9e40857 | 3330 | netif_addr_unlock_bh(dev); |
456ad75c DC |
3331 | } |
3332 | ||
1da177e4 LT |
3333 | unsigned dev_get_flags(const struct net_device *dev) |
3334 | { | |
3335 | unsigned flags; | |
3336 | ||
3337 | flags = (dev->flags & ~(IFF_PROMISC | | |
3338 | IFF_ALLMULTI | | |
b00055aa SR |
3339 | IFF_RUNNING | |
3340 | IFF_LOWER_UP | | |
3341 | IFF_DORMANT)) | | |
1da177e4 LT |
3342 | (dev->gflags & (IFF_PROMISC | |
3343 | IFF_ALLMULTI)); | |
3344 | ||
b00055aa SR |
3345 | if (netif_running(dev)) { |
3346 | if (netif_oper_up(dev)) | |
3347 | flags |= IFF_RUNNING; | |
3348 | if (netif_carrier_ok(dev)) | |
3349 | flags |= IFF_LOWER_UP; | |
3350 | if (netif_dormant(dev)) | |
3351 | flags |= IFF_DORMANT; | |
3352 | } | |
1da177e4 LT |
3353 | |
3354 | return flags; | |
3355 | } | |
3356 | ||
3357 | int dev_change_flags(struct net_device *dev, unsigned flags) | |
3358 | { | |
7c355f53 | 3359 | int ret, changes; |
1da177e4 LT |
3360 | int old_flags = dev->flags; |
3361 | ||
24023451 PM |
3362 | ASSERT_RTNL(); |
3363 | ||
1da177e4 LT |
3364 | /* |
3365 | * Set the flags on our device. | |
3366 | */ | |
3367 | ||
3368 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
3369 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
3370 | IFF_AUTOMEDIA)) | | |
3371 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
3372 | IFF_ALLMULTI)); | |
3373 | ||
3374 | /* | |
3375 | * Load in the correct multicast list now the flags have changed. | |
3376 | */ | |
3377 | ||
0e91796e | 3378 | if (dev->change_rx_flags && (old_flags ^ flags) & IFF_MULTICAST) |
24023451 PM |
3379 | dev->change_rx_flags(dev, IFF_MULTICAST); |
3380 | ||
4417da66 | 3381 | dev_set_rx_mode(dev); |
1da177e4 LT |
3382 | |
3383 | /* | |
3384 | * Have we downed the interface. We handle IFF_UP ourselves | |
3385 | * according to user attempts to set it, rather than blindly | |
3386 | * setting it. | |
3387 | */ | |
3388 | ||
3389 | ret = 0; | |
3390 | if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */ | |
3391 | ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev); | |
3392 | ||
3393 | if (!ret) | |
4417da66 | 3394 | dev_set_rx_mode(dev); |
1da177e4 LT |
3395 | } |
3396 | ||
3397 | if (dev->flags & IFF_UP && | |
3398 | ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI | | |
3399 | IFF_VOLATILE))) | |
056925ab | 3400 | call_netdevice_notifiers(NETDEV_CHANGE, dev); |
1da177e4 LT |
3401 | |
3402 | if ((flags ^ dev->gflags) & IFF_PROMISC) { | |
3403 | int inc = (flags & IFF_PROMISC) ? +1 : -1; | |
3404 | dev->gflags ^= IFF_PROMISC; | |
3405 | dev_set_promiscuity(dev, inc); | |
3406 | } | |
3407 | ||
3408 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
3409 | is important. Some (broken) drivers set IFF_PROMISC, when | |
3410 | IFF_ALLMULTI is requested not asking us and not reporting. | |
3411 | */ | |
3412 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
3413 | int inc = (flags & IFF_ALLMULTI) ? +1 : -1; | |
3414 | dev->gflags ^= IFF_ALLMULTI; | |
3415 | dev_set_allmulti(dev, inc); | |
3416 | } | |
3417 | ||
7c355f53 TG |
3418 | /* Exclude state transition flags, already notified */ |
3419 | changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING); | |
3420 | if (changes) | |
3421 | rtmsg_ifinfo(RTM_NEWLINK, dev, changes); | |
1da177e4 LT |
3422 | |
3423 | return ret; | |
3424 | } | |
3425 | ||
3426 | int dev_set_mtu(struct net_device *dev, int new_mtu) | |
3427 | { | |
3428 | int err; | |
3429 | ||
3430 | if (new_mtu == dev->mtu) | |
3431 | return 0; | |
3432 | ||
3433 | /* MTU must be positive. */ | |
3434 | if (new_mtu < 0) | |
3435 | return -EINVAL; | |
3436 | ||
3437 | if (!netif_device_present(dev)) | |
3438 | return -ENODEV; | |
3439 | ||
3440 | err = 0; | |
3441 | if (dev->change_mtu) | |
3442 | err = dev->change_mtu(dev, new_mtu); | |
3443 | else | |
3444 | dev->mtu = new_mtu; | |
3445 | if (!err && dev->flags & IFF_UP) | |
056925ab | 3446 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); |
1da177e4 LT |
3447 | return err; |
3448 | } | |
3449 | ||
3450 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) | |
3451 | { | |
3452 | int err; | |
3453 | ||
3454 | if (!dev->set_mac_address) | |
3455 | return -EOPNOTSUPP; | |
3456 | if (sa->sa_family != dev->type) | |
3457 | return -EINVAL; | |
3458 | if (!netif_device_present(dev)) | |
3459 | return -ENODEV; | |
3460 | err = dev->set_mac_address(dev, sa); | |
3461 | if (!err) | |
056925ab | 3462 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
1da177e4 LT |
3463 | return err; |
3464 | } | |
3465 | ||
3466 | /* | |
14e3e079 | 3467 | * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock) |
1da177e4 | 3468 | */ |
14e3e079 | 3469 | static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd) |
1da177e4 LT |
3470 | { |
3471 | int err; | |
881d966b | 3472 | struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name); |
1da177e4 LT |
3473 | |
3474 | if (!dev) | |
3475 | return -ENODEV; | |
3476 | ||
3477 | switch (cmd) { | |
3478 | case SIOCGIFFLAGS: /* Get interface flags */ | |
3479 | ifr->ifr_flags = dev_get_flags(dev); | |
3480 | return 0; | |
3481 | ||
1da177e4 LT |
3482 | case SIOCGIFMETRIC: /* Get the metric on the interface |
3483 | (currently unused) */ | |
3484 | ifr->ifr_metric = 0; | |
3485 | return 0; | |
3486 | ||
1da177e4 LT |
3487 | case SIOCGIFMTU: /* Get the MTU of a device */ |
3488 | ifr->ifr_mtu = dev->mtu; | |
3489 | return 0; | |
3490 | ||
1da177e4 LT |
3491 | case SIOCGIFHWADDR: |
3492 | if (!dev->addr_len) | |
3493 | memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data); | |
3494 | else | |
3495 | memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr, | |
3496 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
3497 | ifr->ifr_hwaddr.sa_family = dev->type; | |
3498 | return 0; | |
3499 | ||
14e3e079 JG |
3500 | case SIOCGIFSLAVE: |
3501 | err = -EINVAL; | |
3502 | break; | |
3503 | ||
3504 | case SIOCGIFMAP: | |
3505 | ifr->ifr_map.mem_start = dev->mem_start; | |
3506 | ifr->ifr_map.mem_end = dev->mem_end; | |
3507 | ifr->ifr_map.base_addr = dev->base_addr; | |
3508 | ifr->ifr_map.irq = dev->irq; | |
3509 | ifr->ifr_map.dma = dev->dma; | |
3510 | ifr->ifr_map.port = dev->if_port; | |
3511 | return 0; | |
3512 | ||
3513 | case SIOCGIFINDEX: | |
3514 | ifr->ifr_ifindex = dev->ifindex; | |
3515 | return 0; | |
3516 | ||
3517 | case SIOCGIFTXQLEN: | |
3518 | ifr->ifr_qlen = dev->tx_queue_len; | |
3519 | return 0; | |
3520 | ||
3521 | default: | |
3522 | /* dev_ioctl() should ensure this case | |
3523 | * is never reached | |
3524 | */ | |
3525 | WARN_ON(1); | |
3526 | err = -EINVAL; | |
3527 | break; | |
3528 | ||
3529 | } | |
3530 | return err; | |
3531 | } | |
3532 | ||
3533 | /* | |
3534 | * Perform the SIOCxIFxxx calls, inside rtnl_lock() | |
3535 | */ | |
3536 | static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd) | |
3537 | { | |
3538 | int err; | |
3539 | struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name); | |
3540 | ||
3541 | if (!dev) | |
3542 | return -ENODEV; | |
3543 | ||
3544 | switch (cmd) { | |
3545 | case SIOCSIFFLAGS: /* Set interface flags */ | |
3546 | return dev_change_flags(dev, ifr->ifr_flags); | |
3547 | ||
3548 | case SIOCSIFMETRIC: /* Set the metric on the interface | |
3549 | (currently unused) */ | |
3550 | return -EOPNOTSUPP; | |
3551 | ||
3552 | case SIOCSIFMTU: /* Set the MTU of a device */ | |
3553 | return dev_set_mtu(dev, ifr->ifr_mtu); | |
3554 | ||
1da177e4 LT |
3555 | case SIOCSIFHWADDR: |
3556 | return dev_set_mac_address(dev, &ifr->ifr_hwaddr); | |
3557 | ||
3558 | case SIOCSIFHWBROADCAST: | |
3559 | if (ifr->ifr_hwaddr.sa_family != dev->type) | |
3560 | return -EINVAL; | |
3561 | memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data, | |
3562 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
056925ab | 3563 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
1da177e4 LT |
3564 | return 0; |
3565 | ||
1da177e4 LT |
3566 | case SIOCSIFMAP: |
3567 | if (dev->set_config) { | |
3568 | if (!netif_device_present(dev)) | |
3569 | return -ENODEV; | |
3570 | return dev->set_config(dev, &ifr->ifr_map); | |
3571 | } | |
3572 | return -EOPNOTSUPP; | |
3573 | ||
3574 | case SIOCADDMULTI: | |
61ee6bd4 | 3575 | if ((!dev->set_multicast_list && !dev->set_rx_mode) || |
1da177e4 LT |
3576 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) |
3577 | return -EINVAL; | |
3578 | if (!netif_device_present(dev)) | |
3579 | return -ENODEV; | |
3580 | return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data, | |
3581 | dev->addr_len, 1); | |
3582 | ||
3583 | case SIOCDELMULTI: | |
61ee6bd4 | 3584 | if ((!dev->set_multicast_list && !dev->set_rx_mode) || |
1da177e4 LT |
3585 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) |
3586 | return -EINVAL; | |
3587 | if (!netif_device_present(dev)) | |
3588 | return -ENODEV; | |
3589 | return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data, | |
3590 | dev->addr_len, 1); | |
3591 | ||
1da177e4 LT |
3592 | case SIOCSIFTXQLEN: |
3593 | if (ifr->ifr_qlen < 0) | |
3594 | return -EINVAL; | |
3595 | dev->tx_queue_len = ifr->ifr_qlen; | |
3596 | return 0; | |
3597 | ||
3598 | case SIOCSIFNAME: | |
3599 | ifr->ifr_newname[IFNAMSIZ-1] = '\0'; | |
3600 | return dev_change_name(dev, ifr->ifr_newname); | |
3601 | ||
3602 | /* | |
3603 | * Unknown or private ioctl | |
3604 | */ | |
3605 | ||
3606 | default: | |
3607 | if ((cmd >= SIOCDEVPRIVATE && | |
3608 | cmd <= SIOCDEVPRIVATE + 15) || | |
3609 | cmd == SIOCBONDENSLAVE || | |
3610 | cmd == SIOCBONDRELEASE || | |
3611 | cmd == SIOCBONDSETHWADDR || | |
3612 | cmd == SIOCBONDSLAVEINFOQUERY || | |
3613 | cmd == SIOCBONDINFOQUERY || | |
3614 | cmd == SIOCBONDCHANGEACTIVE || | |
3615 | cmd == SIOCGMIIPHY || | |
3616 | cmd == SIOCGMIIREG || | |
3617 | cmd == SIOCSMIIREG || | |
3618 | cmd == SIOCBRADDIF || | |
3619 | cmd == SIOCBRDELIF || | |
3620 | cmd == SIOCWANDEV) { | |
3621 | err = -EOPNOTSUPP; | |
3622 | if (dev->do_ioctl) { | |
3623 | if (netif_device_present(dev)) | |
3624 | err = dev->do_ioctl(dev, ifr, | |
3625 | cmd); | |
3626 | else | |
3627 | err = -ENODEV; | |
3628 | } | |
3629 | } else | |
3630 | err = -EINVAL; | |
3631 | ||
3632 | } | |
3633 | return err; | |
3634 | } | |
3635 | ||
3636 | /* | |
3637 | * This function handles all "interface"-type I/O control requests. The actual | |
3638 | * 'doing' part of this is dev_ifsioc above. | |
3639 | */ | |
3640 | ||
3641 | /** | |
3642 | * dev_ioctl - network device ioctl | |
c4ea43c5 | 3643 | * @net: the applicable net namespace |
1da177e4 LT |
3644 | * @cmd: command to issue |
3645 | * @arg: pointer to a struct ifreq in user space | |
3646 | * | |
3647 | * Issue ioctl functions to devices. This is normally called by the | |
3648 | * user space syscall interfaces but can sometimes be useful for | |
3649 | * other purposes. The return value is the return from the syscall if | |
3650 | * positive or a negative errno code on error. | |
3651 | */ | |
3652 | ||
881d966b | 3653 | int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg) |
1da177e4 LT |
3654 | { |
3655 | struct ifreq ifr; | |
3656 | int ret; | |
3657 | char *colon; | |
3658 | ||
3659 | /* One special case: SIOCGIFCONF takes ifconf argument | |
3660 | and requires shared lock, because it sleeps writing | |
3661 | to user space. | |
3662 | */ | |
3663 | ||
3664 | if (cmd == SIOCGIFCONF) { | |
6756ae4b | 3665 | rtnl_lock(); |
881d966b | 3666 | ret = dev_ifconf(net, (char __user *) arg); |
6756ae4b | 3667 | rtnl_unlock(); |
1da177e4 LT |
3668 | return ret; |
3669 | } | |
3670 | if (cmd == SIOCGIFNAME) | |
881d966b | 3671 | return dev_ifname(net, (struct ifreq __user *)arg); |
1da177e4 LT |
3672 | |
3673 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
3674 | return -EFAULT; | |
3675 | ||
3676 | ifr.ifr_name[IFNAMSIZ-1] = 0; | |
3677 | ||
3678 | colon = strchr(ifr.ifr_name, ':'); | |
3679 | if (colon) | |
3680 | *colon = 0; | |
3681 | ||
3682 | /* | |
3683 | * See which interface the caller is talking about. | |
3684 | */ | |
3685 | ||
3686 | switch (cmd) { | |
3687 | /* | |
3688 | * These ioctl calls: | |
3689 | * - can be done by all. | |
3690 | * - atomic and do not require locking. | |
3691 | * - return a value | |
3692 | */ | |
3693 | case SIOCGIFFLAGS: | |
3694 | case SIOCGIFMETRIC: | |
3695 | case SIOCGIFMTU: | |
3696 | case SIOCGIFHWADDR: | |
3697 | case SIOCGIFSLAVE: | |
3698 | case SIOCGIFMAP: | |
3699 | case SIOCGIFINDEX: | |
3700 | case SIOCGIFTXQLEN: | |
881d966b | 3701 | dev_load(net, ifr.ifr_name); |
1da177e4 | 3702 | read_lock(&dev_base_lock); |
14e3e079 | 3703 | ret = dev_ifsioc_locked(net, &ifr, cmd); |
1da177e4 LT |
3704 | read_unlock(&dev_base_lock); |
3705 | if (!ret) { | |
3706 | if (colon) | |
3707 | *colon = ':'; | |
3708 | if (copy_to_user(arg, &ifr, | |
3709 | sizeof(struct ifreq))) | |
3710 | ret = -EFAULT; | |
3711 | } | |
3712 | return ret; | |
3713 | ||
3714 | case SIOCETHTOOL: | |
881d966b | 3715 | dev_load(net, ifr.ifr_name); |
1da177e4 | 3716 | rtnl_lock(); |
881d966b | 3717 | ret = dev_ethtool(net, &ifr); |
1da177e4 LT |
3718 | rtnl_unlock(); |
3719 | if (!ret) { | |
3720 | if (colon) | |
3721 | *colon = ':'; | |
3722 | if (copy_to_user(arg, &ifr, | |
3723 | sizeof(struct ifreq))) | |
3724 | ret = -EFAULT; | |
3725 | } | |
3726 | return ret; | |
3727 | ||
3728 | /* | |
3729 | * These ioctl calls: | |
3730 | * - require superuser power. | |
3731 | * - require strict serialization. | |
3732 | * - return a value | |
3733 | */ | |
3734 | case SIOCGMIIPHY: | |
3735 | case SIOCGMIIREG: | |
3736 | case SIOCSIFNAME: | |
3737 | if (!capable(CAP_NET_ADMIN)) | |
3738 | return -EPERM; | |
881d966b | 3739 | dev_load(net, ifr.ifr_name); |
1da177e4 | 3740 | rtnl_lock(); |
881d966b | 3741 | ret = dev_ifsioc(net, &ifr, cmd); |
1da177e4 LT |
3742 | rtnl_unlock(); |
3743 | if (!ret) { | |
3744 | if (colon) | |
3745 | *colon = ':'; | |
3746 | if (copy_to_user(arg, &ifr, | |
3747 | sizeof(struct ifreq))) | |
3748 | ret = -EFAULT; | |
3749 | } | |
3750 | return ret; | |
3751 | ||
3752 | /* | |
3753 | * These ioctl calls: | |
3754 | * - require superuser power. | |
3755 | * - require strict serialization. | |
3756 | * - do not return a value | |
3757 | */ | |
3758 | case SIOCSIFFLAGS: | |
3759 | case SIOCSIFMETRIC: | |
3760 | case SIOCSIFMTU: | |
3761 | case SIOCSIFMAP: | |
3762 | case SIOCSIFHWADDR: | |
3763 | case SIOCSIFSLAVE: | |
3764 | case SIOCADDMULTI: | |
3765 | case SIOCDELMULTI: | |
3766 | case SIOCSIFHWBROADCAST: | |
3767 | case SIOCSIFTXQLEN: | |
3768 | case SIOCSMIIREG: | |
3769 | case SIOCBONDENSLAVE: | |
3770 | case SIOCBONDRELEASE: | |
3771 | case SIOCBONDSETHWADDR: | |
1da177e4 LT |
3772 | case SIOCBONDCHANGEACTIVE: |
3773 | case SIOCBRADDIF: | |
3774 | case SIOCBRDELIF: | |
3775 | if (!capable(CAP_NET_ADMIN)) | |
3776 | return -EPERM; | |
cabcac0b TG |
3777 | /* fall through */ |
3778 | case SIOCBONDSLAVEINFOQUERY: | |
3779 | case SIOCBONDINFOQUERY: | |
881d966b | 3780 | dev_load(net, ifr.ifr_name); |
1da177e4 | 3781 | rtnl_lock(); |
881d966b | 3782 | ret = dev_ifsioc(net, &ifr, cmd); |
1da177e4 LT |
3783 | rtnl_unlock(); |
3784 | return ret; | |
3785 | ||
3786 | case SIOCGIFMEM: | |
3787 | /* Get the per device memory space. We can add this but | |
3788 | * currently do not support it */ | |
3789 | case SIOCSIFMEM: | |
3790 | /* Set the per device memory buffer space. | |
3791 | * Not applicable in our case */ | |
3792 | case SIOCSIFLINK: | |
3793 | return -EINVAL; | |
3794 | ||
3795 | /* | |
3796 | * Unknown or private ioctl. | |
3797 | */ | |
3798 | default: | |
3799 | if (cmd == SIOCWANDEV || | |
3800 | (cmd >= SIOCDEVPRIVATE && | |
3801 | cmd <= SIOCDEVPRIVATE + 15)) { | |
881d966b | 3802 | dev_load(net, ifr.ifr_name); |
1da177e4 | 3803 | rtnl_lock(); |
881d966b | 3804 | ret = dev_ifsioc(net, &ifr, cmd); |
1da177e4 LT |
3805 | rtnl_unlock(); |
3806 | if (!ret && copy_to_user(arg, &ifr, | |
3807 | sizeof(struct ifreq))) | |
3808 | ret = -EFAULT; | |
3809 | return ret; | |
3810 | } | |
1da177e4 | 3811 | /* Take care of Wireless Extensions */ |
295f4a1f | 3812 | if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) |
881d966b | 3813 | return wext_handle_ioctl(net, &ifr, cmd, arg); |
1da177e4 LT |
3814 | return -EINVAL; |
3815 | } | |
3816 | } | |
3817 | ||
3818 | ||
3819 | /** | |
3820 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 3821 | * @net: the applicable net namespace |
1da177e4 LT |
3822 | * |
3823 | * Returns a suitable unique value for a new device interface | |
3824 | * number. The caller must hold the rtnl semaphore or the | |
3825 | * dev_base_lock to be sure it remains unique. | |
3826 | */ | |
881d966b | 3827 | static int dev_new_index(struct net *net) |
1da177e4 LT |
3828 | { |
3829 | static int ifindex; | |
3830 | for (;;) { | |
3831 | if (++ifindex <= 0) | |
3832 | ifindex = 1; | |
881d966b | 3833 | if (!__dev_get_by_index(net, ifindex)) |
1da177e4 LT |
3834 | return ifindex; |
3835 | } | |
3836 | } | |
3837 | ||
1da177e4 LT |
3838 | /* Delayed registration/unregisteration */ |
3839 | static DEFINE_SPINLOCK(net_todo_list_lock); | |
3b5b34fd | 3840 | static LIST_HEAD(net_todo_list); |
1da177e4 | 3841 | |
6f05f629 | 3842 | static void net_set_todo(struct net_device *dev) |
1da177e4 LT |
3843 | { |
3844 | spin_lock(&net_todo_list_lock); | |
3845 | list_add_tail(&dev->todo_list, &net_todo_list); | |
3846 | spin_unlock(&net_todo_list_lock); | |
3847 | } | |
3848 | ||
93ee31f1 DL |
3849 | static void rollback_registered(struct net_device *dev) |
3850 | { | |
3851 | BUG_ON(dev_boot_phase); | |
3852 | ASSERT_RTNL(); | |
3853 | ||
3854 | /* Some devices call without registering for initialization unwind. */ | |
3855 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
3856 | printk(KERN_DEBUG "unregister_netdevice: device %s/%p never " | |
3857 | "was registered\n", dev->name, dev); | |
3858 | ||
3859 | WARN_ON(1); | |
3860 | return; | |
3861 | } | |
3862 | ||
3863 | BUG_ON(dev->reg_state != NETREG_REGISTERED); | |
3864 | ||
3865 | /* If device is running, close it first. */ | |
3866 | dev_close(dev); | |
3867 | ||
3868 | /* And unlink it from device chain. */ | |
3869 | unlist_netdevice(dev); | |
3870 | ||
3871 | dev->reg_state = NETREG_UNREGISTERING; | |
3872 | ||
3873 | synchronize_net(); | |
3874 | ||
3875 | /* Shutdown queueing discipline. */ | |
3876 | dev_shutdown(dev); | |
3877 | ||
3878 | ||
3879 | /* Notify protocols, that we are about to destroy | |
3880 | this device. They should clean all the things. | |
3881 | */ | |
3882 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
3883 | ||
3884 | /* | |
3885 | * Flush the unicast and multicast chains | |
3886 | */ | |
3887 | dev_addr_discard(dev); | |
3888 | ||
3889 | if (dev->uninit) | |
3890 | dev->uninit(dev); | |
3891 | ||
3892 | /* Notifier chain MUST detach us from master device. */ | |
547b792c | 3893 | WARN_ON(dev->master); |
93ee31f1 DL |
3894 | |
3895 | /* Remove entries from kobject tree */ | |
3896 | netdev_unregister_kobject(dev); | |
3897 | ||
3898 | synchronize_net(); | |
3899 | ||
3900 | dev_put(dev); | |
3901 | } | |
3902 | ||
e8a0464c DM |
3903 | static void __netdev_init_queue_locks_one(struct net_device *dev, |
3904 | struct netdev_queue *dev_queue, | |
3905 | void *_unused) | |
c773e847 DM |
3906 | { |
3907 | spin_lock_init(&dev_queue->_xmit_lock); | |
cf508b12 | 3908 | netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type); |
c773e847 DM |
3909 | dev_queue->xmit_lock_owner = -1; |
3910 | } | |
3911 | ||
3912 | static void netdev_init_queue_locks(struct net_device *dev) | |
3913 | { | |
e8a0464c DM |
3914 | netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL); |
3915 | __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL); | |
c773e847 DM |
3916 | } |
3917 | ||
1da177e4 LT |
3918 | /** |
3919 | * register_netdevice - register a network device | |
3920 | * @dev: device to register | |
3921 | * | |
3922 | * Take a completed network device structure and add it to the kernel | |
3923 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
3924 | * chain. 0 is returned on success. A negative errno code is returned | |
3925 | * on a failure to set up the device, or if the name is a duplicate. | |
3926 | * | |
3927 | * Callers must hold the rtnl semaphore. You may want | |
3928 | * register_netdev() instead of this. | |
3929 | * | |
3930 | * BUGS: | |
3931 | * The locking appears insufficient to guarantee two parallel registers | |
3932 | * will not get the same name. | |
3933 | */ | |
3934 | ||
3935 | int register_netdevice(struct net_device *dev) | |
3936 | { | |
3937 | struct hlist_head *head; | |
3938 | struct hlist_node *p; | |
3939 | int ret; | |
881d966b | 3940 | struct net *net; |
1da177e4 LT |
3941 | |
3942 | BUG_ON(dev_boot_phase); | |
3943 | ASSERT_RTNL(); | |
3944 | ||
b17a7c17 SH |
3945 | might_sleep(); |
3946 | ||
1da177e4 LT |
3947 | /* When net_device's are persistent, this will be fatal. */ |
3948 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
c346dca1 YH |
3949 | BUG_ON(!dev_net(dev)); |
3950 | net = dev_net(dev); | |
1da177e4 | 3951 | |
f1f28aa3 | 3952 | spin_lock_init(&dev->addr_list_lock); |
cf508b12 | 3953 | netdev_set_addr_lockdep_class(dev); |
c773e847 | 3954 | netdev_init_queue_locks(dev); |
1da177e4 | 3955 | |
1da177e4 LT |
3956 | dev->iflink = -1; |
3957 | ||
3958 | /* Init, if this function is available */ | |
3959 | if (dev->init) { | |
3960 | ret = dev->init(dev); | |
3961 | if (ret) { | |
3962 | if (ret > 0) | |
3963 | ret = -EIO; | |
90833aa4 | 3964 | goto out; |
1da177e4 LT |
3965 | } |
3966 | } | |
4ec93edb | 3967 | |
1da177e4 LT |
3968 | if (!dev_valid_name(dev->name)) { |
3969 | ret = -EINVAL; | |
7ce1b0ed | 3970 | goto err_uninit; |
1da177e4 LT |
3971 | } |
3972 | ||
881d966b | 3973 | dev->ifindex = dev_new_index(net); |
1da177e4 LT |
3974 | if (dev->iflink == -1) |
3975 | dev->iflink = dev->ifindex; | |
3976 | ||
3977 | /* Check for existence of name */ | |
881d966b | 3978 | head = dev_name_hash(net, dev->name); |
1da177e4 LT |
3979 | hlist_for_each(p, head) { |
3980 | struct net_device *d | |
3981 | = hlist_entry(p, struct net_device, name_hlist); | |
3982 | if (!strncmp(d->name, dev->name, IFNAMSIZ)) { | |
3983 | ret = -EEXIST; | |
7ce1b0ed | 3984 | goto err_uninit; |
1da177e4 | 3985 | } |
4ec93edb | 3986 | } |
1da177e4 | 3987 | |
d212f87b SH |
3988 | /* Fix illegal checksum combinations */ |
3989 | if ((dev->features & NETIF_F_HW_CSUM) && | |
3990 | (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
3991 | printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n", | |
3992 | dev->name); | |
3993 | dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); | |
3994 | } | |
3995 | ||
3996 | if ((dev->features & NETIF_F_NO_CSUM) && | |
3997 | (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
3998 | printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n", | |
3999 | dev->name); | |
4000 | dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM); | |
4001 | } | |
4002 | ||
4003 | ||
1da177e4 LT |
4004 | /* Fix illegal SG+CSUM combinations. */ |
4005 | if ((dev->features & NETIF_F_SG) && | |
8648b305 | 4006 | !(dev->features & NETIF_F_ALL_CSUM)) { |
5a8da02b | 4007 | printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no checksum feature.\n", |
1da177e4 LT |
4008 | dev->name); |
4009 | dev->features &= ~NETIF_F_SG; | |
4010 | } | |
4011 | ||
4012 | /* TSO requires that SG is present as well. */ | |
4013 | if ((dev->features & NETIF_F_TSO) && | |
4014 | !(dev->features & NETIF_F_SG)) { | |
5a8da02b | 4015 | printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no SG feature.\n", |
1da177e4 LT |
4016 | dev->name); |
4017 | dev->features &= ~NETIF_F_TSO; | |
4018 | } | |
e89e9cf5 AR |
4019 | if (dev->features & NETIF_F_UFO) { |
4020 | if (!(dev->features & NETIF_F_HW_CSUM)) { | |
4021 | printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no " | |
4022 | "NETIF_F_HW_CSUM feature.\n", | |
4023 | dev->name); | |
4024 | dev->features &= ~NETIF_F_UFO; | |
4025 | } | |
4026 | if (!(dev->features & NETIF_F_SG)) { | |
4027 | printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no " | |
4028 | "NETIF_F_SG feature.\n", | |
4029 | dev->name); | |
4030 | dev->features &= ~NETIF_F_UFO; | |
4031 | } | |
4032 | } | |
1da177e4 | 4033 | |
e5a4a72d LB |
4034 | /* Enable software GSO if SG is supported. */ |
4035 | if (dev->features & NETIF_F_SG) | |
4036 | dev->features |= NETIF_F_GSO; | |
4037 | ||
aaf8cdc3 | 4038 | netdev_initialize_kobject(dev); |
8b41d188 | 4039 | ret = netdev_register_kobject(dev); |
b17a7c17 | 4040 | if (ret) |
7ce1b0ed | 4041 | goto err_uninit; |
b17a7c17 SH |
4042 | dev->reg_state = NETREG_REGISTERED; |
4043 | ||
1da177e4 LT |
4044 | /* |
4045 | * Default initial state at registry is that the | |
4046 | * device is present. | |
4047 | */ | |
4048 | ||
4049 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
4050 | ||
1da177e4 | 4051 | dev_init_scheduler(dev); |
1da177e4 | 4052 | dev_hold(dev); |
ce286d32 | 4053 | list_netdevice(dev); |
1da177e4 LT |
4054 | |
4055 | /* Notify protocols, that a new device appeared. */ | |
056925ab | 4056 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 4057 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
4058 | if (ret) { |
4059 | rollback_registered(dev); | |
4060 | dev->reg_state = NETREG_UNREGISTERED; | |
4061 | } | |
1da177e4 LT |
4062 | |
4063 | out: | |
4064 | return ret; | |
7ce1b0ed HX |
4065 | |
4066 | err_uninit: | |
4067 | if (dev->uninit) | |
4068 | dev->uninit(dev); | |
4069 | goto out; | |
1da177e4 LT |
4070 | } |
4071 | ||
4072 | /** | |
4073 | * register_netdev - register a network device | |
4074 | * @dev: device to register | |
4075 | * | |
4076 | * Take a completed network device structure and add it to the kernel | |
4077 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
4078 | * chain. 0 is returned on success. A negative errno code is returned | |
4079 | * on a failure to set up the device, or if the name is a duplicate. | |
4080 | * | |
38b4da38 | 4081 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
4082 | * and expands the device name if you passed a format string to |
4083 | * alloc_netdev. | |
4084 | */ | |
4085 | int register_netdev(struct net_device *dev) | |
4086 | { | |
4087 | int err; | |
4088 | ||
4089 | rtnl_lock(); | |
4090 | ||
4091 | /* | |
4092 | * If the name is a format string the caller wants us to do a | |
4093 | * name allocation. | |
4094 | */ | |
4095 | if (strchr(dev->name, '%')) { | |
4096 | err = dev_alloc_name(dev, dev->name); | |
4097 | if (err < 0) | |
4098 | goto out; | |
4099 | } | |
4ec93edb | 4100 | |
1da177e4 LT |
4101 | err = register_netdevice(dev); |
4102 | out: | |
4103 | rtnl_unlock(); | |
4104 | return err; | |
4105 | } | |
4106 | EXPORT_SYMBOL(register_netdev); | |
4107 | ||
4108 | /* | |
4109 | * netdev_wait_allrefs - wait until all references are gone. | |
4110 | * | |
4111 | * This is called when unregistering network devices. | |
4112 | * | |
4113 | * Any protocol or device that holds a reference should register | |
4114 | * for netdevice notification, and cleanup and put back the | |
4115 | * reference if they receive an UNREGISTER event. | |
4116 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 4117 | * call dev_put. |
1da177e4 LT |
4118 | */ |
4119 | static void netdev_wait_allrefs(struct net_device *dev) | |
4120 | { | |
4121 | unsigned long rebroadcast_time, warning_time; | |
4122 | ||
4123 | rebroadcast_time = warning_time = jiffies; | |
4124 | while (atomic_read(&dev->refcnt) != 0) { | |
4125 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { | |
6756ae4b | 4126 | rtnl_lock(); |
1da177e4 LT |
4127 | |
4128 | /* Rebroadcast unregister notification */ | |
056925ab | 4129 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
1da177e4 LT |
4130 | |
4131 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, | |
4132 | &dev->state)) { | |
4133 | /* We must not have linkwatch events | |
4134 | * pending on unregister. If this | |
4135 | * happens, we simply run the queue | |
4136 | * unscheduled, resulting in a noop | |
4137 | * for this device. | |
4138 | */ | |
4139 | linkwatch_run_queue(); | |
4140 | } | |
4141 | ||
6756ae4b | 4142 | __rtnl_unlock(); |
1da177e4 LT |
4143 | |
4144 | rebroadcast_time = jiffies; | |
4145 | } | |
4146 | ||
4147 | msleep(250); | |
4148 | ||
4149 | if (time_after(jiffies, warning_time + 10 * HZ)) { | |
4150 | printk(KERN_EMERG "unregister_netdevice: " | |
4151 | "waiting for %s to become free. Usage " | |
4152 | "count = %d\n", | |
4153 | dev->name, atomic_read(&dev->refcnt)); | |
4154 | warning_time = jiffies; | |
4155 | } | |
4156 | } | |
4157 | } | |
4158 | ||
4159 | /* The sequence is: | |
4160 | * | |
4161 | * rtnl_lock(); | |
4162 | * ... | |
4163 | * register_netdevice(x1); | |
4164 | * register_netdevice(x2); | |
4165 | * ... | |
4166 | * unregister_netdevice(y1); | |
4167 | * unregister_netdevice(y2); | |
4168 | * ... | |
4169 | * rtnl_unlock(); | |
4170 | * free_netdev(y1); | |
4171 | * free_netdev(y2); | |
4172 | * | |
4173 | * We are invoked by rtnl_unlock() after it drops the semaphore. | |
4174 | * This allows us to deal with problems: | |
b17a7c17 | 4175 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
4176 | * without deadlocking with linkwatch via keventd. |
4177 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
4178 | * safely in order to wait for the netdev refcnt to drop to zero. | |
4179 | */ | |
4a3e2f71 | 4180 | static DEFINE_MUTEX(net_todo_run_mutex); |
1da177e4 LT |
4181 | void netdev_run_todo(void) |
4182 | { | |
626ab0e6 | 4183 | struct list_head list; |
1da177e4 LT |
4184 | |
4185 | /* Need to guard against multiple cpu's getting out of order. */ | |
4a3e2f71 | 4186 | mutex_lock(&net_todo_run_mutex); |
1da177e4 LT |
4187 | |
4188 | /* Not safe to do outside the semaphore. We must not return | |
4189 | * until all unregister events invoked by the local processor | |
4190 | * have been completed (either by this todo run, or one on | |
4191 | * another cpu). | |
4192 | */ | |
4193 | if (list_empty(&net_todo_list)) | |
4194 | goto out; | |
4195 | ||
4196 | /* Snapshot list, allow later requests */ | |
4197 | spin_lock(&net_todo_list_lock); | |
626ab0e6 | 4198 | list_replace_init(&net_todo_list, &list); |
1da177e4 | 4199 | spin_unlock(&net_todo_list_lock); |
626ab0e6 | 4200 | |
1da177e4 LT |
4201 | while (!list_empty(&list)) { |
4202 | struct net_device *dev | |
4203 | = list_entry(list.next, struct net_device, todo_list); | |
4204 | list_del(&dev->todo_list); | |
4205 | ||
b17a7c17 SH |
4206 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
4207 | printk(KERN_ERR "network todo '%s' but state %d\n", | |
4208 | dev->name, dev->reg_state); | |
4209 | dump_stack(); | |
4210 | continue; | |
4211 | } | |
1da177e4 | 4212 | |
b17a7c17 | 4213 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 4214 | |
6e583ce5 SH |
4215 | on_each_cpu(flush_backlog, dev, 1); |
4216 | ||
b17a7c17 | 4217 | netdev_wait_allrefs(dev); |
1da177e4 | 4218 | |
b17a7c17 SH |
4219 | /* paranoia */ |
4220 | BUG_ON(atomic_read(&dev->refcnt)); | |
547b792c IJ |
4221 | WARN_ON(dev->ip_ptr); |
4222 | WARN_ON(dev->ip6_ptr); | |
4223 | WARN_ON(dev->dn_ptr); | |
1da177e4 | 4224 | |
b17a7c17 SH |
4225 | if (dev->destructor) |
4226 | dev->destructor(dev); | |
9093bbb2 SH |
4227 | |
4228 | /* Free network device */ | |
4229 | kobject_put(&dev->dev.kobj); | |
1da177e4 LT |
4230 | } |
4231 | ||
4232 | out: | |
4a3e2f71 | 4233 | mutex_unlock(&net_todo_run_mutex); |
1da177e4 LT |
4234 | } |
4235 | ||
5a1b5898 | 4236 | static struct net_device_stats *internal_stats(struct net_device *dev) |
c45d286e | 4237 | { |
5a1b5898 | 4238 | return &dev->stats; |
c45d286e RR |
4239 | } |
4240 | ||
dc2b4847 | 4241 | static void netdev_init_one_queue(struct net_device *dev, |
e8a0464c DM |
4242 | struct netdev_queue *queue, |
4243 | void *_unused) | |
dc2b4847 | 4244 | { |
dc2b4847 DM |
4245 | queue->dev = dev; |
4246 | } | |
4247 | ||
bb949fbd DM |
4248 | static void netdev_init_queues(struct net_device *dev) |
4249 | { | |
e8a0464c DM |
4250 | netdev_init_one_queue(dev, &dev->rx_queue, NULL); |
4251 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); | |
c3f26a26 | 4252 | spin_lock_init(&dev->tx_global_lock); |
bb949fbd DM |
4253 | } |
4254 | ||
1da177e4 | 4255 | /** |
f25f4e44 | 4256 | * alloc_netdev_mq - allocate network device |
1da177e4 LT |
4257 | * @sizeof_priv: size of private data to allocate space for |
4258 | * @name: device name format string | |
4259 | * @setup: callback to initialize device | |
f25f4e44 | 4260 | * @queue_count: the number of subqueues to allocate |
1da177e4 LT |
4261 | * |
4262 | * Allocates a struct net_device with private data area for driver use | |
f25f4e44 PWJ |
4263 | * and performs basic initialization. Also allocates subquue structs |
4264 | * for each queue on the device at the end of the netdevice. | |
1da177e4 | 4265 | */ |
f25f4e44 PWJ |
4266 | struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name, |
4267 | void (*setup)(struct net_device *), unsigned int queue_count) | |
1da177e4 | 4268 | { |
e8a0464c | 4269 | struct netdev_queue *tx; |
1da177e4 | 4270 | struct net_device *dev; |
7943986c | 4271 | size_t alloc_size; |
e8a0464c | 4272 | void *p; |
1da177e4 | 4273 | |
b6fe17d6 SH |
4274 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
4275 | ||
fd2ea0a7 | 4276 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
4277 | if (sizeof_priv) { |
4278 | /* ensure 32-byte alignment of private area */ | |
4279 | alloc_size = (alloc_size + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST; | |
4280 | alloc_size += sizeof_priv; | |
4281 | } | |
4282 | /* ensure 32-byte alignment of whole construct */ | |
4283 | alloc_size += NETDEV_ALIGN_CONST; | |
1da177e4 | 4284 | |
31380de9 | 4285 | p = kzalloc(alloc_size, GFP_KERNEL); |
1da177e4 | 4286 | if (!p) { |
b6fe17d6 | 4287 | printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n"); |
1da177e4 LT |
4288 | return NULL; |
4289 | } | |
1da177e4 | 4290 | |
7943986c | 4291 | tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL); |
e8a0464c DM |
4292 | if (!tx) { |
4293 | printk(KERN_ERR "alloc_netdev: Unable to allocate " | |
4294 | "tx qdiscs.\n"); | |
4295 | kfree(p); | |
4296 | return NULL; | |
4297 | } | |
4298 | ||
1da177e4 LT |
4299 | dev = (struct net_device *) |
4300 | (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST); | |
4301 | dev->padded = (char *)dev - (char *)p; | |
c346dca1 | 4302 | dev_net_set(dev, &init_net); |
1da177e4 | 4303 | |
e8a0464c DM |
4304 | dev->_tx = tx; |
4305 | dev->num_tx_queues = queue_count; | |
fd2ea0a7 | 4306 | dev->real_num_tx_queues = queue_count; |
e8a0464c | 4307 | |
f25f4e44 PWJ |
4308 | if (sizeof_priv) { |
4309 | dev->priv = ((char *)dev + | |
fd2ea0a7 | 4310 | ((sizeof(struct net_device) + NETDEV_ALIGN_CONST) |
f25f4e44 PWJ |
4311 | & ~NETDEV_ALIGN_CONST)); |
4312 | } | |
4313 | ||
82cc1a7a | 4314 | dev->gso_max_size = GSO_MAX_SIZE; |
1da177e4 | 4315 | |
bb949fbd DM |
4316 | netdev_init_queues(dev); |
4317 | ||
5a1b5898 | 4318 | dev->get_stats = internal_stats; |
bea3348e | 4319 | netpoll_netdev_init(dev); |
1da177e4 LT |
4320 | setup(dev); |
4321 | strcpy(dev->name, name); | |
4322 | return dev; | |
4323 | } | |
f25f4e44 | 4324 | EXPORT_SYMBOL(alloc_netdev_mq); |
1da177e4 LT |
4325 | |
4326 | /** | |
4327 | * free_netdev - free network device | |
4328 | * @dev: device | |
4329 | * | |
4ec93edb YH |
4330 | * This function does the last stage of destroying an allocated device |
4331 | * interface. The reference to the device object is released. | |
1da177e4 LT |
4332 | * If this is the last reference then it will be freed. |
4333 | */ | |
4334 | void free_netdev(struct net_device *dev) | |
4335 | { | |
f3005d7f DL |
4336 | release_net(dev_net(dev)); |
4337 | ||
e8a0464c DM |
4338 | kfree(dev->_tx); |
4339 | ||
3041a069 | 4340 | /* Compatibility with error handling in drivers */ |
1da177e4 LT |
4341 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
4342 | kfree((char *)dev - dev->padded); | |
4343 | return; | |
4344 | } | |
4345 | ||
4346 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
4347 | dev->reg_state = NETREG_RELEASED; | |
4348 | ||
43cb76d9 GKH |
4349 | /* will free via device release */ |
4350 | put_device(&dev->dev); | |
1da177e4 | 4351 | } |
4ec93edb | 4352 | |
1da177e4 | 4353 | /* Synchronize with packet receive processing. */ |
4ec93edb | 4354 | void synchronize_net(void) |
1da177e4 LT |
4355 | { |
4356 | might_sleep(); | |
fbd568a3 | 4357 | synchronize_rcu(); |
1da177e4 LT |
4358 | } |
4359 | ||
4360 | /** | |
4361 | * unregister_netdevice - remove device from the kernel | |
4362 | * @dev: device | |
4363 | * | |
4364 | * This function shuts down a device interface and removes it | |
d59b54b1 | 4365 | * from the kernel tables. |
1da177e4 LT |
4366 | * |
4367 | * Callers must hold the rtnl semaphore. You may want | |
4368 | * unregister_netdev() instead of this. | |
4369 | */ | |
4370 | ||
22f8cde5 | 4371 | void unregister_netdevice(struct net_device *dev) |
1da177e4 | 4372 | { |
a6620712 HX |
4373 | ASSERT_RTNL(); |
4374 | ||
93ee31f1 | 4375 | rollback_registered(dev); |
1da177e4 LT |
4376 | /* Finish processing unregister after unlock */ |
4377 | net_set_todo(dev); | |
1da177e4 LT |
4378 | } |
4379 | ||
4380 | /** | |
4381 | * unregister_netdev - remove device from the kernel | |
4382 | * @dev: device | |
4383 | * | |
4384 | * This function shuts down a device interface and removes it | |
d59b54b1 | 4385 | * from the kernel tables. |
1da177e4 LT |
4386 | * |
4387 | * This is just a wrapper for unregister_netdevice that takes | |
4388 | * the rtnl semaphore. In general you want to use this and not | |
4389 | * unregister_netdevice. | |
4390 | */ | |
4391 | void unregister_netdev(struct net_device *dev) | |
4392 | { | |
4393 | rtnl_lock(); | |
4394 | unregister_netdevice(dev); | |
4395 | rtnl_unlock(); | |
4396 | } | |
4397 | ||
4398 | EXPORT_SYMBOL(unregister_netdev); | |
4399 | ||
ce286d32 EB |
4400 | /** |
4401 | * dev_change_net_namespace - move device to different nethost namespace | |
4402 | * @dev: device | |
4403 | * @net: network namespace | |
4404 | * @pat: If not NULL name pattern to try if the current device name | |
4405 | * is already taken in the destination network namespace. | |
4406 | * | |
4407 | * This function shuts down a device interface and moves it | |
4408 | * to a new network namespace. On success 0 is returned, on | |
4409 | * a failure a netagive errno code is returned. | |
4410 | * | |
4411 | * Callers must hold the rtnl semaphore. | |
4412 | */ | |
4413 | ||
4414 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
4415 | { | |
4416 | char buf[IFNAMSIZ]; | |
4417 | const char *destname; | |
4418 | int err; | |
4419 | ||
4420 | ASSERT_RTNL(); | |
4421 | ||
4422 | /* Don't allow namespace local devices to be moved. */ | |
4423 | err = -EINVAL; | |
4424 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
4425 | goto out; | |
4426 | ||
4427 | /* Ensure the device has been registrered */ | |
4428 | err = -EINVAL; | |
4429 | if (dev->reg_state != NETREG_REGISTERED) | |
4430 | goto out; | |
4431 | ||
4432 | /* Get out if there is nothing todo */ | |
4433 | err = 0; | |
878628fb | 4434 | if (net_eq(dev_net(dev), net)) |
ce286d32 EB |
4435 | goto out; |
4436 | ||
4437 | /* Pick the destination device name, and ensure | |
4438 | * we can use it in the destination network namespace. | |
4439 | */ | |
4440 | err = -EEXIST; | |
4441 | destname = dev->name; | |
4442 | if (__dev_get_by_name(net, destname)) { | |
4443 | /* We get here if we can't use the current device name */ | |
4444 | if (!pat) | |
4445 | goto out; | |
4446 | if (!dev_valid_name(pat)) | |
4447 | goto out; | |
4448 | if (strchr(pat, '%')) { | |
4449 | if (__dev_alloc_name(net, pat, buf) < 0) | |
4450 | goto out; | |
4451 | destname = buf; | |
4452 | } else | |
4453 | destname = pat; | |
4454 | if (__dev_get_by_name(net, destname)) | |
4455 | goto out; | |
4456 | } | |
4457 | ||
4458 | /* | |
4459 | * And now a mini version of register_netdevice unregister_netdevice. | |
4460 | */ | |
4461 | ||
4462 | /* If device is running close it first. */ | |
9b772652 | 4463 | dev_close(dev); |
ce286d32 EB |
4464 | |
4465 | /* And unlink it from device chain */ | |
4466 | err = -ENODEV; | |
4467 | unlist_netdevice(dev); | |
4468 | ||
4469 | synchronize_net(); | |
4470 | ||
4471 | /* Shutdown queueing discipline. */ | |
4472 | dev_shutdown(dev); | |
4473 | ||
4474 | /* Notify protocols, that we are about to destroy | |
4475 | this device. They should clean all the things. | |
4476 | */ | |
4477 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
4478 | ||
4479 | /* | |
4480 | * Flush the unicast and multicast chains | |
4481 | */ | |
4482 | dev_addr_discard(dev); | |
4483 | ||
4484 | /* Actually switch the network namespace */ | |
c346dca1 | 4485 | dev_net_set(dev, net); |
ce286d32 EB |
4486 | |
4487 | /* Assign the new device name */ | |
4488 | if (destname != dev->name) | |
4489 | strcpy(dev->name, destname); | |
4490 | ||
4491 | /* If there is an ifindex conflict assign a new one */ | |
4492 | if (__dev_get_by_index(net, dev->ifindex)) { | |
4493 | int iflink = (dev->iflink == dev->ifindex); | |
4494 | dev->ifindex = dev_new_index(net); | |
4495 | if (iflink) | |
4496 | dev->iflink = dev->ifindex; | |
4497 | } | |
4498 | ||
8b41d188 | 4499 | /* Fixup kobjects */ |
aaf8cdc3 DL |
4500 | netdev_unregister_kobject(dev); |
4501 | err = netdev_register_kobject(dev); | |
8b41d188 | 4502 | WARN_ON(err); |
ce286d32 EB |
4503 | |
4504 | /* Add the device back in the hashes */ | |
4505 | list_netdevice(dev); | |
4506 | ||
4507 | /* Notify protocols, that a new device appeared. */ | |
4508 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
4509 | ||
4510 | synchronize_net(); | |
4511 | err = 0; | |
4512 | out: | |
4513 | return err; | |
4514 | } | |
4515 | ||
1da177e4 LT |
4516 | static int dev_cpu_callback(struct notifier_block *nfb, |
4517 | unsigned long action, | |
4518 | void *ocpu) | |
4519 | { | |
4520 | struct sk_buff **list_skb; | |
37437bb2 | 4521 | struct Qdisc **list_net; |
1da177e4 LT |
4522 | struct sk_buff *skb; |
4523 | unsigned int cpu, oldcpu = (unsigned long)ocpu; | |
4524 | struct softnet_data *sd, *oldsd; | |
4525 | ||
8bb78442 | 4526 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
1da177e4 LT |
4527 | return NOTIFY_OK; |
4528 | ||
4529 | local_irq_disable(); | |
4530 | cpu = smp_processor_id(); | |
4531 | sd = &per_cpu(softnet_data, cpu); | |
4532 | oldsd = &per_cpu(softnet_data, oldcpu); | |
4533 | ||
4534 | /* Find end of our completion_queue. */ | |
4535 | list_skb = &sd->completion_queue; | |
4536 | while (*list_skb) | |
4537 | list_skb = &(*list_skb)->next; | |
4538 | /* Append completion queue from offline CPU. */ | |
4539 | *list_skb = oldsd->completion_queue; | |
4540 | oldsd->completion_queue = NULL; | |
4541 | ||
4542 | /* Find end of our output_queue. */ | |
4543 | list_net = &sd->output_queue; | |
4544 | while (*list_net) | |
4545 | list_net = &(*list_net)->next_sched; | |
4546 | /* Append output queue from offline CPU. */ | |
4547 | *list_net = oldsd->output_queue; | |
4548 | oldsd->output_queue = NULL; | |
4549 | ||
4550 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
4551 | local_irq_enable(); | |
4552 | ||
4553 | /* Process offline CPU's input_pkt_queue */ | |
4554 | while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) | |
4555 | netif_rx(skb); | |
4556 | ||
4557 | return NOTIFY_OK; | |
4558 | } | |
1da177e4 | 4559 | |
db217334 CL |
4560 | #ifdef CONFIG_NET_DMA |
4561 | /** | |
0ed72ec4 RD |
4562 | * net_dma_rebalance - try to maintain one DMA channel per CPU |
4563 | * @net_dma: DMA client and associated data (lock, channels, channel_mask) | |
4564 | * | |
4565 | * This is called when the number of channels allocated to the net_dma client | |
4566 | * changes. The net_dma client tries to have one DMA channel per CPU. | |
db217334 | 4567 | */ |
d379b01e DW |
4568 | |
4569 | static void net_dma_rebalance(struct net_dma *net_dma) | |
db217334 | 4570 | { |
d379b01e | 4571 | unsigned int cpu, i, n, chan_idx; |
db217334 CL |
4572 | struct dma_chan *chan; |
4573 | ||
d379b01e | 4574 | if (cpus_empty(net_dma->channel_mask)) { |
db217334 | 4575 | for_each_online_cpu(cpu) |
29bbd72d | 4576 | rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL); |
db217334 CL |
4577 | return; |
4578 | } | |
4579 | ||
4580 | i = 0; | |
4581 | cpu = first_cpu(cpu_online_map); | |
4582 | ||
0e12f848 | 4583 | for_each_cpu_mask_nr(chan_idx, net_dma->channel_mask) { |
d379b01e DW |
4584 | chan = net_dma->channels[chan_idx]; |
4585 | ||
4586 | n = ((num_online_cpus() / cpus_weight(net_dma->channel_mask)) | |
4587 | + (i < (num_online_cpus() % | |
4588 | cpus_weight(net_dma->channel_mask)) ? 1 : 0)); | |
db217334 CL |
4589 | |
4590 | while(n) { | |
29bbd72d | 4591 | per_cpu(softnet_data, cpu).net_dma = chan; |
db217334 CL |
4592 | cpu = next_cpu(cpu, cpu_online_map); |
4593 | n--; | |
4594 | } | |
4595 | i++; | |
4596 | } | |
db217334 CL |
4597 | } |
4598 | ||
4599 | /** | |
4600 | * netdev_dma_event - event callback for the net_dma_client | |
4601 | * @client: should always be net_dma_client | |
f4b8ea78 | 4602 | * @chan: DMA channel for the event |
0ed72ec4 | 4603 | * @state: DMA state to be handled |
db217334 | 4604 | */ |
d379b01e DW |
4605 | static enum dma_state_client |
4606 | netdev_dma_event(struct dma_client *client, struct dma_chan *chan, | |
4607 | enum dma_state state) | |
4608 | { | |
4609 | int i, found = 0, pos = -1; | |
4610 | struct net_dma *net_dma = | |
4611 | container_of(client, struct net_dma, client); | |
4612 | enum dma_state_client ack = DMA_DUP; /* default: take no action */ | |
4613 | ||
4614 | spin_lock(&net_dma->lock); | |
4615 | switch (state) { | |
4616 | case DMA_RESOURCE_AVAILABLE: | |
0c0b0aca | 4617 | for (i = 0; i < nr_cpu_ids; i++) |
d379b01e DW |
4618 | if (net_dma->channels[i] == chan) { |
4619 | found = 1; | |
4620 | break; | |
4621 | } else if (net_dma->channels[i] == NULL && pos < 0) | |
4622 | pos = i; | |
4623 | ||
4624 | if (!found && pos >= 0) { | |
4625 | ack = DMA_ACK; | |
4626 | net_dma->channels[pos] = chan; | |
4627 | cpu_set(pos, net_dma->channel_mask); | |
4628 | net_dma_rebalance(net_dma); | |
4629 | } | |
db217334 CL |
4630 | break; |
4631 | case DMA_RESOURCE_REMOVED: | |
0c0b0aca | 4632 | for (i = 0; i < nr_cpu_ids; i++) |
d379b01e DW |
4633 | if (net_dma->channels[i] == chan) { |
4634 | found = 1; | |
4635 | pos = i; | |
4636 | break; | |
4637 | } | |
4638 | ||
4639 | if (found) { | |
4640 | ack = DMA_ACK; | |
4641 | cpu_clear(pos, net_dma->channel_mask); | |
4642 | net_dma->channels[i] = NULL; | |
4643 | net_dma_rebalance(net_dma); | |
4644 | } | |
db217334 CL |
4645 | break; |
4646 | default: | |
4647 | break; | |
4648 | } | |
d379b01e DW |
4649 | spin_unlock(&net_dma->lock); |
4650 | ||
4651 | return ack; | |
db217334 CL |
4652 | } |
4653 | ||
4654 | /** | |
4655 | * netdev_dma_regiser - register the networking subsystem as a DMA client | |
4656 | */ | |
4657 | static int __init netdev_dma_register(void) | |
4658 | { | |
0c0b0aca MT |
4659 | net_dma.channels = kzalloc(nr_cpu_ids * sizeof(struct net_dma), |
4660 | GFP_KERNEL); | |
4661 | if (unlikely(!net_dma.channels)) { | |
4662 | printk(KERN_NOTICE | |
4663 | "netdev_dma: no memory for net_dma.channels\n"); | |
4664 | return -ENOMEM; | |
4665 | } | |
d379b01e DW |
4666 | spin_lock_init(&net_dma.lock); |
4667 | dma_cap_set(DMA_MEMCPY, net_dma.client.cap_mask); | |
4668 | dma_async_client_register(&net_dma.client); | |
4669 | dma_async_client_chan_request(&net_dma.client); | |
db217334 CL |
4670 | return 0; |
4671 | } | |
4672 | ||
4673 | #else | |
4674 | static int __init netdev_dma_register(void) { return -ENODEV; } | |
4675 | #endif /* CONFIG_NET_DMA */ | |
1da177e4 | 4676 | |
7f353bf2 HX |
4677 | /** |
4678 | * netdev_compute_feature - compute conjunction of two feature sets | |
4679 | * @all: first feature set | |
4680 | * @one: second feature set | |
4681 | * | |
4682 | * Computes a new feature set after adding a device with feature set | |
4683 | * @one to the master device with current feature set @all. Returns | |
4684 | * the new feature set. | |
4685 | */ | |
4686 | int netdev_compute_features(unsigned long all, unsigned long one) | |
4687 | { | |
4688 | /* if device needs checksumming, downgrade to hw checksumming */ | |
4689 | if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM)) | |
4690 | all ^= NETIF_F_NO_CSUM | NETIF_F_HW_CSUM; | |
4691 | ||
4692 | /* if device can't do all checksum, downgrade to ipv4/ipv6 */ | |
4693 | if (all & NETIF_F_HW_CSUM && !(one & NETIF_F_HW_CSUM)) | |
4694 | all ^= NETIF_F_HW_CSUM | |
4695 | | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM; | |
4696 | ||
4697 | if (one & NETIF_F_GSO) | |
4698 | one |= NETIF_F_GSO_SOFTWARE; | |
4699 | one |= NETIF_F_GSO; | |
4700 | ||
e2a6b852 HX |
4701 | /* |
4702 | * If even one device supports a GSO protocol with software fallback, | |
4703 | * enable it for all. | |
4704 | */ | |
4705 | all |= one & NETIF_F_GSO_SOFTWARE; | |
4706 | ||
7f353bf2 HX |
4707 | /* If even one device supports robust GSO, enable it for all. */ |
4708 | if (one & NETIF_F_GSO_ROBUST) | |
4709 | all |= NETIF_F_GSO_ROBUST; | |
4710 | ||
4711 | all &= one | NETIF_F_LLTX; | |
4712 | ||
4713 | if (!(all & NETIF_F_ALL_CSUM)) | |
4714 | all &= ~NETIF_F_SG; | |
4715 | if (!(all & NETIF_F_SG)) | |
4716 | all &= ~NETIF_F_GSO_MASK; | |
4717 | ||
4718 | return all; | |
4719 | } | |
4720 | EXPORT_SYMBOL(netdev_compute_features); | |
4721 | ||
30d97d35 PE |
4722 | static struct hlist_head *netdev_create_hash(void) |
4723 | { | |
4724 | int i; | |
4725 | struct hlist_head *hash; | |
4726 | ||
4727 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
4728 | if (hash != NULL) | |
4729 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
4730 | INIT_HLIST_HEAD(&hash[i]); | |
4731 | ||
4732 | return hash; | |
4733 | } | |
4734 | ||
881d966b | 4735 | /* Initialize per network namespace state */ |
4665079c | 4736 | static int __net_init netdev_init(struct net *net) |
881d966b | 4737 | { |
881d966b | 4738 | INIT_LIST_HEAD(&net->dev_base_head); |
881d966b | 4739 | |
30d97d35 PE |
4740 | net->dev_name_head = netdev_create_hash(); |
4741 | if (net->dev_name_head == NULL) | |
4742 | goto err_name; | |
881d966b | 4743 | |
30d97d35 PE |
4744 | net->dev_index_head = netdev_create_hash(); |
4745 | if (net->dev_index_head == NULL) | |
4746 | goto err_idx; | |
881d966b EB |
4747 | |
4748 | return 0; | |
30d97d35 PE |
4749 | |
4750 | err_idx: | |
4751 | kfree(net->dev_name_head); | |
4752 | err_name: | |
4753 | return -ENOMEM; | |
881d966b EB |
4754 | } |
4755 | ||
cf04a4c7 | 4756 | char *netdev_drivername(const struct net_device *dev, char *buffer, int len) |
6579e57b | 4757 | { |
cf04a4c7 SH |
4758 | const struct device_driver *driver; |
4759 | const struct device *parent; | |
6579e57b AV |
4760 | |
4761 | if (len <= 0 || !buffer) | |
4762 | return buffer; | |
4763 | buffer[0] = 0; | |
4764 | ||
4765 | parent = dev->dev.parent; | |
4766 | ||
4767 | if (!parent) | |
4768 | return buffer; | |
4769 | ||
4770 | driver = parent->driver; | |
4771 | if (driver && driver->name) | |
4772 | strlcpy(buffer, driver->name, len); | |
4773 | return buffer; | |
4774 | } | |
4775 | ||
4665079c | 4776 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
4777 | { |
4778 | kfree(net->dev_name_head); | |
4779 | kfree(net->dev_index_head); | |
4780 | } | |
4781 | ||
022cbae6 | 4782 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
4783 | .init = netdev_init, |
4784 | .exit = netdev_exit, | |
4785 | }; | |
4786 | ||
4665079c | 4787 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 EB |
4788 | { |
4789 | struct net_device *dev, *next; | |
4790 | /* | |
4791 | * Push all migratable of the network devices back to the | |
4792 | * initial network namespace | |
4793 | */ | |
4794 | rtnl_lock(); | |
4795 | for_each_netdev_safe(net, dev, next) { | |
4796 | int err; | |
aca51397 | 4797 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
4798 | |
4799 | /* Ignore unmoveable devices (i.e. loopback) */ | |
4800 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
4801 | continue; | |
4802 | ||
4803 | /* Push remaing network devices to init_net */ | |
aca51397 PE |
4804 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
4805 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
ce286d32 | 4806 | if (err) { |
aca51397 | 4807 | printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n", |
ce286d32 | 4808 | __func__, dev->name, err); |
aca51397 | 4809 | BUG(); |
ce286d32 EB |
4810 | } |
4811 | } | |
4812 | rtnl_unlock(); | |
4813 | } | |
4814 | ||
022cbae6 | 4815 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 EB |
4816 | .exit = default_device_exit, |
4817 | }; | |
4818 | ||
1da177e4 LT |
4819 | /* |
4820 | * Initialize the DEV module. At boot time this walks the device list and | |
4821 | * unhooks any devices that fail to initialise (normally hardware not | |
4822 | * present) and leaves us with a valid list of present and active devices. | |
4823 | * | |
4824 | */ | |
4825 | ||
4826 | /* | |
4827 | * This is called single threaded during boot, so no need | |
4828 | * to take the rtnl semaphore. | |
4829 | */ | |
4830 | static int __init net_dev_init(void) | |
4831 | { | |
4832 | int i, rc = -ENOMEM; | |
4833 | ||
4834 | BUG_ON(!dev_boot_phase); | |
4835 | ||
1da177e4 LT |
4836 | if (dev_proc_init()) |
4837 | goto out; | |
4838 | ||
8b41d188 | 4839 | if (netdev_kobject_init()) |
1da177e4 LT |
4840 | goto out; |
4841 | ||
4842 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 4843 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
4844 | INIT_LIST_HEAD(&ptype_base[i]); |
4845 | ||
881d966b EB |
4846 | if (register_pernet_subsys(&netdev_net_ops)) |
4847 | goto out; | |
1da177e4 | 4848 | |
ce286d32 EB |
4849 | if (register_pernet_device(&default_device_ops)) |
4850 | goto out; | |
4851 | ||
1da177e4 LT |
4852 | /* |
4853 | * Initialise the packet receive queues. | |
4854 | */ | |
4855 | ||
6f912042 | 4856 | for_each_possible_cpu(i) { |
1da177e4 LT |
4857 | struct softnet_data *queue; |
4858 | ||
4859 | queue = &per_cpu(softnet_data, i); | |
4860 | skb_queue_head_init(&queue->input_pkt_queue); | |
1da177e4 LT |
4861 | queue->completion_queue = NULL; |
4862 | INIT_LIST_HEAD(&queue->poll_list); | |
bea3348e SH |
4863 | |
4864 | queue->backlog.poll = process_backlog; | |
4865 | queue->backlog.weight = weight_p; | |
1da177e4 LT |
4866 | } |
4867 | ||
db217334 CL |
4868 | netdev_dma_register(); |
4869 | ||
1da177e4 LT |
4870 | dev_boot_phase = 0; |
4871 | ||
962cf36c CM |
4872 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
4873 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 LT |
4874 | |
4875 | hotcpu_notifier(dev_cpu_callback, 0); | |
4876 | dst_init(); | |
4877 | dev_mcast_init(); | |
4878 | rc = 0; | |
4879 | out: | |
4880 | return rc; | |
4881 | } | |
4882 | ||
4883 | subsys_initcall(net_dev_init); | |
4884 | ||
4885 | EXPORT_SYMBOL(__dev_get_by_index); | |
4886 | EXPORT_SYMBOL(__dev_get_by_name); | |
4887 | EXPORT_SYMBOL(__dev_remove_pack); | |
c2373ee9 | 4888 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
4889 | EXPORT_SYMBOL(dev_add_pack); |
4890 | EXPORT_SYMBOL(dev_alloc_name); | |
4891 | EXPORT_SYMBOL(dev_close); | |
4892 | EXPORT_SYMBOL(dev_get_by_flags); | |
4893 | EXPORT_SYMBOL(dev_get_by_index); | |
4894 | EXPORT_SYMBOL(dev_get_by_name); | |
1da177e4 LT |
4895 | EXPORT_SYMBOL(dev_open); |
4896 | EXPORT_SYMBOL(dev_queue_xmit); | |
4897 | EXPORT_SYMBOL(dev_remove_pack); | |
4898 | EXPORT_SYMBOL(dev_set_allmulti); | |
4899 | EXPORT_SYMBOL(dev_set_promiscuity); | |
4900 | EXPORT_SYMBOL(dev_change_flags); | |
4901 | EXPORT_SYMBOL(dev_set_mtu); | |
4902 | EXPORT_SYMBOL(dev_set_mac_address); | |
4903 | EXPORT_SYMBOL(free_netdev); | |
4904 | EXPORT_SYMBOL(netdev_boot_setup_check); | |
4905 | EXPORT_SYMBOL(netdev_set_master); | |
4906 | EXPORT_SYMBOL(netdev_state_change); | |
4907 | EXPORT_SYMBOL(netif_receive_skb); | |
4908 | EXPORT_SYMBOL(netif_rx); | |
4909 | EXPORT_SYMBOL(register_gifconf); | |
4910 | EXPORT_SYMBOL(register_netdevice); | |
4911 | EXPORT_SYMBOL(register_netdevice_notifier); | |
4912 | EXPORT_SYMBOL(skb_checksum_help); | |
4913 | EXPORT_SYMBOL(synchronize_net); | |
4914 | EXPORT_SYMBOL(unregister_netdevice); | |
4915 | EXPORT_SYMBOL(unregister_netdevice_notifier); | |
4916 | EXPORT_SYMBOL(net_enable_timestamp); | |
4917 | EXPORT_SYMBOL(net_disable_timestamp); | |
4918 | EXPORT_SYMBOL(dev_get_flags); | |
4919 | ||
4920 | #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE) | |
4921 | EXPORT_SYMBOL(br_handle_frame_hook); | |
4922 | EXPORT_SYMBOL(br_fdb_get_hook); | |
4923 | EXPORT_SYMBOL(br_fdb_put_hook); | |
4924 | #endif | |
4925 | ||
4926 | #ifdef CONFIG_KMOD | |
4927 | EXPORT_SYMBOL(dev_load); | |
4928 | #endif | |
4929 | ||
4930 | EXPORT_PER_CPU_SYMBOL(softnet_data); |