]>
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> | |
08e9897d | 82 | #include <linux/hash.h> |
1da177e4 | 83 | #include <linux/sched.h> |
4a3e2f71 | 84 | #include <linux/mutex.h> |
1da177e4 LT |
85 | #include <linux/string.h> |
86 | #include <linux/mm.h> | |
87 | #include <linux/socket.h> | |
88 | #include <linux/sockios.h> | |
89 | #include <linux/errno.h> | |
90 | #include <linux/interrupt.h> | |
91 | #include <linux/if_ether.h> | |
92 | #include <linux/netdevice.h> | |
93 | #include <linux/etherdevice.h> | |
0187bdfb | 94 | #include <linux/ethtool.h> |
1da177e4 LT |
95 | #include <linux/notifier.h> |
96 | #include <linux/skbuff.h> | |
457c4cbc | 97 | #include <net/net_namespace.h> |
1da177e4 LT |
98 | #include <net/sock.h> |
99 | #include <linux/rtnetlink.h> | |
100 | #include <linux/proc_fs.h> | |
101 | #include <linux/seq_file.h> | |
102 | #include <linux/stat.h> | |
103 | #include <linux/if_bridge.h> | |
b863ceb7 | 104 | #include <linux/if_macvlan.h> |
1da177e4 LT |
105 | #include <net/dst.h> |
106 | #include <net/pkt_sched.h> | |
107 | #include <net/checksum.h> | |
44540960 | 108 | #include <net/xfrm.h> |
1da177e4 LT |
109 | #include <linux/highmem.h> |
110 | #include <linux/init.h> | |
111 | #include <linux/kmod.h> | |
112 | #include <linux/module.h> | |
1da177e4 LT |
113 | #include <linux/netpoll.h> |
114 | #include <linux/rcupdate.h> | |
115 | #include <linux/delay.h> | |
295f4a1f | 116 | #include <net/wext.h> |
1da177e4 | 117 | #include <net/iw_handler.h> |
1da177e4 | 118 | #include <asm/current.h> |
5bdb9886 | 119 | #include <linux/audit.h> |
db217334 | 120 | #include <linux/dmaengine.h> |
f6a78bfc | 121 | #include <linux/err.h> |
c7fa9d18 | 122 | #include <linux/ctype.h> |
723e98b7 | 123 | #include <linux/if_arp.h> |
6de329e2 | 124 | #include <linux/if_vlan.h> |
8f0f2223 | 125 | #include <linux/ip.h> |
ad55dcaf | 126 | #include <net/ip.h> |
8f0f2223 DM |
127 | #include <linux/ipv6.h> |
128 | #include <linux/in.h> | |
b6b2fed1 DM |
129 | #include <linux/jhash.h> |
130 | #include <linux/random.h> | |
9cbc1cb8 | 131 | #include <trace/events/napi.h> |
1da177e4 | 132 | |
342709ef PE |
133 | #include "net-sysfs.h" |
134 | ||
d565b0a1 HX |
135 | /* Instead of increasing this, you should create a hash table. */ |
136 | #define MAX_GRO_SKBS 8 | |
137 | ||
5d38a079 HX |
138 | /* This should be increased if a protocol with a bigger head is added. */ |
139 | #define GRO_MAX_HEAD (MAX_HEADER + 128) | |
140 | ||
1da177e4 LT |
141 | /* |
142 | * The list of packet types we will receive (as opposed to discard) | |
143 | * and the routines to invoke. | |
144 | * | |
145 | * Why 16. Because with 16 the only overlap we get on a hash of the | |
146 | * low nibble of the protocol value is RARP/SNAP/X.25. | |
147 | * | |
148 | * NOTE: That is no longer true with the addition of VLAN tags. Not | |
149 | * sure which should go first, but I bet it won't make much | |
150 | * difference if we are running VLANs. The good news is that | |
151 | * this protocol won't be in the list unless compiled in, so | |
3041a069 | 152 | * the average user (w/out VLANs) will not be adversely affected. |
1da177e4 LT |
153 | * --BLG |
154 | * | |
155 | * 0800 IP | |
156 | * 8100 802.1Q VLAN | |
157 | * 0001 802.3 | |
158 | * 0002 AX.25 | |
159 | * 0004 802.2 | |
160 | * 8035 RARP | |
161 | * 0005 SNAP | |
162 | * 0805 X.25 | |
163 | * 0806 ARP | |
164 | * 8137 IPX | |
165 | * 0009 Localtalk | |
166 | * 86DD IPv6 | |
167 | */ | |
168 | ||
82d8a867 PE |
169 | #define PTYPE_HASH_SIZE (16) |
170 | #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1) | |
171 | ||
1da177e4 | 172 | static DEFINE_SPINLOCK(ptype_lock); |
82d8a867 | 173 | static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
6b2bedc3 | 174 | static struct list_head ptype_all __read_mostly; /* Taps */ |
1da177e4 | 175 | |
1da177e4 | 176 | /* |
7562f876 | 177 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
178 | * semaphore. |
179 | * | |
c6d14c84 | 180 | * Pure readers hold dev_base_lock for reading, or rcu_read_lock() |
1da177e4 LT |
181 | * |
182 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 183 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
184 | * actual updates. This allows pure readers to access the list even |
185 | * while a writer is preparing to update it. | |
186 | * | |
187 | * To put it another way, dev_base_lock is held for writing only to | |
188 | * protect against pure readers; the rtnl semaphore provides the | |
189 | * protection against other writers. | |
190 | * | |
191 | * See, for example usages, register_netdevice() and | |
192 | * unregister_netdevice(), which must be called with the rtnl | |
193 | * semaphore held. | |
194 | */ | |
1da177e4 | 195 | DEFINE_RWLOCK(dev_base_lock); |
1da177e4 LT |
196 | EXPORT_SYMBOL(dev_base_lock); |
197 | ||
881d966b | 198 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 LT |
199 | { |
200 | unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ)); | |
08e9897d | 201 | return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; |
1da177e4 LT |
202 | } |
203 | ||
881d966b | 204 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 205 | { |
7c28bd0b | 206 | return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; |
1da177e4 LT |
207 | } |
208 | ||
ce286d32 EB |
209 | /* Device list insertion */ |
210 | static int list_netdevice(struct net_device *dev) | |
211 | { | |
c346dca1 | 212 | struct net *net = dev_net(dev); |
ce286d32 EB |
213 | |
214 | ASSERT_RTNL(); | |
215 | ||
216 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 217 | list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); |
72c9528b | 218 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); |
fb699dfd ED |
219 | hlist_add_head_rcu(&dev->index_hlist, |
220 | dev_index_hash(net, dev->ifindex)); | |
ce286d32 EB |
221 | write_unlock_bh(&dev_base_lock); |
222 | return 0; | |
223 | } | |
224 | ||
fb699dfd ED |
225 | /* Device list removal |
226 | * caller must respect a RCU grace period before freeing/reusing dev | |
227 | */ | |
ce286d32 EB |
228 | static void unlist_netdevice(struct net_device *dev) |
229 | { | |
230 | ASSERT_RTNL(); | |
231 | ||
232 | /* Unlink dev from the device chain */ | |
233 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 234 | list_del_rcu(&dev->dev_list); |
72c9528b | 235 | hlist_del_rcu(&dev->name_hlist); |
fb699dfd | 236 | hlist_del_rcu(&dev->index_hlist); |
ce286d32 EB |
237 | write_unlock_bh(&dev_base_lock); |
238 | } | |
239 | ||
1da177e4 LT |
240 | /* |
241 | * Our notifier list | |
242 | */ | |
243 | ||
f07d5b94 | 244 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
245 | |
246 | /* | |
247 | * Device drivers call our routines to queue packets here. We empty the | |
248 | * queue in the local softnet handler. | |
249 | */ | |
bea3348e SH |
250 | |
251 | DEFINE_PER_CPU(struct softnet_data, softnet_data); | |
d1b19dff | 252 | EXPORT_PER_CPU_SYMBOL(softnet_data); |
1da177e4 | 253 | |
cf508b12 | 254 | #ifdef CONFIG_LOCKDEP |
723e98b7 | 255 | /* |
c773e847 | 256 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class |
723e98b7 JP |
257 | * according to dev->type |
258 | */ | |
259 | static const unsigned short netdev_lock_type[] = | |
260 | {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
261 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, | |
262 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
263 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
264 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
265 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
266 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
267 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
268 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
269 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
270 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
271 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
272 | ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211, | |
2d91d78b | 273 | ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, |
929122cd | 274 | ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154, |
fcb94e42 | 275 | ARPHRD_VOID, ARPHRD_NONE}; |
723e98b7 | 276 | |
36cbd3dc | 277 | static const char *const netdev_lock_name[] = |
723e98b7 JP |
278 | {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", |
279 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
280 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
281 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
282 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
283 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
284 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
285 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
286 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
287 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
288 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
289 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
290 | "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211", | |
2d91d78b | 291 | "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", |
929122cd | 292 | "_xmit_PHONET_PIPE", "_xmit_IEEE802154", |
fcb94e42 | 293 | "_xmit_VOID", "_xmit_NONE"}; |
723e98b7 JP |
294 | |
295 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
cf508b12 | 296 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; |
723e98b7 JP |
297 | |
298 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
299 | { | |
300 | int i; | |
301 | ||
302 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
303 | if (netdev_lock_type[i] == dev_type) | |
304 | return i; | |
305 | /* the last key is used by default */ | |
306 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
307 | } | |
308 | ||
cf508b12 DM |
309 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
310 | unsigned short dev_type) | |
723e98b7 JP |
311 | { |
312 | int i; | |
313 | ||
314 | i = netdev_lock_pos(dev_type); | |
315 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
316 | netdev_lock_name[i]); | |
317 | } | |
cf508b12 DM |
318 | |
319 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
320 | { | |
321 | int i; | |
322 | ||
323 | i = netdev_lock_pos(dev->type); | |
324 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
325 | &netdev_addr_lock_key[i], | |
326 | netdev_lock_name[i]); | |
327 | } | |
723e98b7 | 328 | #else |
cf508b12 DM |
329 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
330 | unsigned short dev_type) | |
331 | { | |
332 | } | |
333 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
723e98b7 JP |
334 | { |
335 | } | |
336 | #endif | |
1da177e4 LT |
337 | |
338 | /******************************************************************************* | |
339 | ||
340 | Protocol management and registration routines | |
341 | ||
342 | *******************************************************************************/ | |
343 | ||
1da177e4 LT |
344 | /* |
345 | * Add a protocol ID to the list. Now that the input handler is | |
346 | * smarter we can dispense with all the messy stuff that used to be | |
347 | * here. | |
348 | * | |
349 | * BEWARE!!! Protocol handlers, mangling input packets, | |
350 | * MUST BE last in hash buckets and checking protocol handlers | |
351 | * MUST start from promiscuous ptype_all chain in net_bh. | |
352 | * It is true now, do not change it. | |
353 | * Explanation follows: if protocol handler, mangling packet, will | |
354 | * be the first on list, it is not able to sense, that packet | |
355 | * is cloned and should be copied-on-write, so that it will | |
356 | * change it and subsequent readers will get broken packet. | |
357 | * --ANK (980803) | |
358 | */ | |
359 | ||
360 | /** | |
361 | * dev_add_pack - add packet handler | |
362 | * @pt: packet type declaration | |
363 | * | |
364 | * Add a protocol handler to the networking stack. The passed &packet_type | |
365 | * is linked into kernel lists and may not be freed until it has been | |
366 | * removed from the kernel lists. | |
367 | * | |
4ec93edb | 368 | * This call does not sleep therefore it can not |
1da177e4 LT |
369 | * guarantee all CPU's that are in middle of receiving packets |
370 | * will see the new packet type (until the next received packet). | |
371 | */ | |
372 | ||
373 | void dev_add_pack(struct packet_type *pt) | |
374 | { | |
375 | int hash; | |
376 | ||
377 | spin_lock_bh(&ptype_lock); | |
9be9a6b9 | 378 | if (pt->type == htons(ETH_P_ALL)) |
1da177e4 | 379 | list_add_rcu(&pt->list, &ptype_all); |
9be9a6b9 | 380 | else { |
82d8a867 | 381 | hash = ntohs(pt->type) & PTYPE_HASH_MASK; |
1da177e4 LT |
382 | list_add_rcu(&pt->list, &ptype_base[hash]); |
383 | } | |
384 | spin_unlock_bh(&ptype_lock); | |
385 | } | |
d1b19dff | 386 | EXPORT_SYMBOL(dev_add_pack); |
1da177e4 | 387 | |
1da177e4 LT |
388 | /** |
389 | * __dev_remove_pack - remove packet handler | |
390 | * @pt: packet type declaration | |
391 | * | |
392 | * Remove a protocol handler that was previously added to the kernel | |
393 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
394 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 395 | * returns. |
1da177e4 LT |
396 | * |
397 | * The packet type might still be in use by receivers | |
398 | * and must not be freed until after all the CPU's have gone | |
399 | * through a quiescent state. | |
400 | */ | |
401 | void __dev_remove_pack(struct packet_type *pt) | |
402 | { | |
403 | struct list_head *head; | |
404 | struct packet_type *pt1; | |
405 | ||
406 | spin_lock_bh(&ptype_lock); | |
407 | ||
9be9a6b9 | 408 | if (pt->type == htons(ETH_P_ALL)) |
1da177e4 | 409 | head = &ptype_all; |
9be9a6b9 | 410 | else |
82d8a867 | 411 | head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; |
1da177e4 LT |
412 | |
413 | list_for_each_entry(pt1, head, list) { | |
414 | if (pt == pt1) { | |
415 | list_del_rcu(&pt->list); | |
416 | goto out; | |
417 | } | |
418 | } | |
419 | ||
420 | printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt); | |
421 | out: | |
422 | spin_unlock_bh(&ptype_lock); | |
423 | } | |
d1b19dff ED |
424 | EXPORT_SYMBOL(__dev_remove_pack); |
425 | ||
1da177e4 LT |
426 | /** |
427 | * dev_remove_pack - remove packet handler | |
428 | * @pt: packet type declaration | |
429 | * | |
430 | * Remove a protocol handler that was previously added to the kernel | |
431 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
432 | * from the kernel lists and can be freed or reused once this function | |
433 | * returns. | |
434 | * | |
435 | * This call sleeps to guarantee that no CPU is looking at the packet | |
436 | * type after return. | |
437 | */ | |
438 | void dev_remove_pack(struct packet_type *pt) | |
439 | { | |
440 | __dev_remove_pack(pt); | |
4ec93edb | 441 | |
1da177e4 LT |
442 | synchronize_net(); |
443 | } | |
d1b19dff | 444 | EXPORT_SYMBOL(dev_remove_pack); |
1da177e4 LT |
445 | |
446 | /****************************************************************************** | |
447 | ||
448 | Device Boot-time Settings Routines | |
449 | ||
450 | *******************************************************************************/ | |
451 | ||
452 | /* Boot time configuration table */ | |
453 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
454 | ||
455 | /** | |
456 | * netdev_boot_setup_add - add new setup entry | |
457 | * @name: name of the device | |
458 | * @map: configured settings for the device | |
459 | * | |
460 | * Adds new setup entry to the dev_boot_setup list. The function | |
461 | * returns 0 on error and 1 on success. This is a generic routine to | |
462 | * all netdevices. | |
463 | */ | |
464 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
465 | { | |
466 | struct netdev_boot_setup *s; | |
467 | int i; | |
468 | ||
469 | s = dev_boot_setup; | |
470 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
471 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
472 | memset(s[i].name, 0, sizeof(s[i].name)); | |
93b3cff9 | 473 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
474 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
475 | break; | |
476 | } | |
477 | } | |
478 | ||
479 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
480 | } | |
481 | ||
482 | /** | |
483 | * netdev_boot_setup_check - check boot time settings | |
484 | * @dev: the netdevice | |
485 | * | |
486 | * Check boot time settings for the device. | |
487 | * The found settings are set for the device to be used | |
488 | * later in the device probing. | |
489 | * Returns 0 if no settings found, 1 if they are. | |
490 | */ | |
491 | int netdev_boot_setup_check(struct net_device *dev) | |
492 | { | |
493 | struct netdev_boot_setup *s = dev_boot_setup; | |
494 | int i; | |
495 | ||
496 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
497 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
93b3cff9 | 498 | !strcmp(dev->name, s[i].name)) { |
1da177e4 LT |
499 | dev->irq = s[i].map.irq; |
500 | dev->base_addr = s[i].map.base_addr; | |
501 | dev->mem_start = s[i].map.mem_start; | |
502 | dev->mem_end = s[i].map.mem_end; | |
503 | return 1; | |
504 | } | |
505 | } | |
506 | return 0; | |
507 | } | |
d1b19dff | 508 | EXPORT_SYMBOL(netdev_boot_setup_check); |
1da177e4 LT |
509 | |
510 | ||
511 | /** | |
512 | * netdev_boot_base - get address from boot time settings | |
513 | * @prefix: prefix for network device | |
514 | * @unit: id for network device | |
515 | * | |
516 | * Check boot time settings for the base address of device. | |
517 | * The found settings are set for the device to be used | |
518 | * later in the device probing. | |
519 | * Returns 0 if no settings found. | |
520 | */ | |
521 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
522 | { | |
523 | const struct netdev_boot_setup *s = dev_boot_setup; | |
524 | char name[IFNAMSIZ]; | |
525 | int i; | |
526 | ||
527 | sprintf(name, "%s%d", prefix, unit); | |
528 | ||
529 | /* | |
530 | * If device already registered then return base of 1 | |
531 | * to indicate not to probe for this interface | |
532 | */ | |
881d966b | 533 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
534 | return 1; |
535 | ||
536 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
537 | if (!strcmp(name, s[i].name)) | |
538 | return s[i].map.base_addr; | |
539 | return 0; | |
540 | } | |
541 | ||
542 | /* | |
543 | * Saves at boot time configured settings for any netdevice. | |
544 | */ | |
545 | int __init netdev_boot_setup(char *str) | |
546 | { | |
547 | int ints[5]; | |
548 | struct ifmap map; | |
549 | ||
550 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
551 | if (!str || !*str) | |
552 | return 0; | |
553 | ||
554 | /* Save settings */ | |
555 | memset(&map, 0, sizeof(map)); | |
556 | if (ints[0] > 0) | |
557 | map.irq = ints[1]; | |
558 | if (ints[0] > 1) | |
559 | map.base_addr = ints[2]; | |
560 | if (ints[0] > 2) | |
561 | map.mem_start = ints[3]; | |
562 | if (ints[0] > 3) | |
563 | map.mem_end = ints[4]; | |
564 | ||
565 | /* Add new entry to the list */ | |
566 | return netdev_boot_setup_add(str, &map); | |
567 | } | |
568 | ||
569 | __setup("netdev=", netdev_boot_setup); | |
570 | ||
571 | /******************************************************************************* | |
572 | ||
573 | Device Interface Subroutines | |
574 | ||
575 | *******************************************************************************/ | |
576 | ||
577 | /** | |
578 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 579 | * @net: the applicable net namespace |
1da177e4 LT |
580 | * @name: name to find |
581 | * | |
582 | * Find an interface by name. Must be called under RTNL semaphore | |
583 | * or @dev_base_lock. If the name is found a pointer to the device | |
584 | * is returned. If the name is not found then %NULL is returned. The | |
585 | * reference counters are not incremented so the caller must be | |
586 | * careful with locks. | |
587 | */ | |
588 | ||
881d966b | 589 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
590 | { |
591 | struct hlist_node *p; | |
0bd8d536 ED |
592 | struct net_device *dev; |
593 | struct hlist_head *head = dev_name_hash(net, name); | |
1da177e4 | 594 | |
0bd8d536 | 595 | hlist_for_each_entry(dev, p, head, name_hlist) |
1da177e4 LT |
596 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
597 | return dev; | |
0bd8d536 | 598 | |
1da177e4 LT |
599 | return NULL; |
600 | } | |
d1b19dff | 601 | EXPORT_SYMBOL(__dev_get_by_name); |
1da177e4 | 602 | |
72c9528b ED |
603 | /** |
604 | * dev_get_by_name_rcu - find a device by its name | |
605 | * @net: the applicable net namespace | |
606 | * @name: name to find | |
607 | * | |
608 | * Find an interface by name. | |
609 | * If the name is found a pointer to the device is returned. | |
610 | * If the name is not found then %NULL is returned. | |
611 | * The reference counters are not incremented so the caller must be | |
612 | * careful with locks. The caller must hold RCU lock. | |
613 | */ | |
614 | ||
615 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
616 | { | |
617 | struct hlist_node *p; | |
618 | struct net_device *dev; | |
619 | struct hlist_head *head = dev_name_hash(net, name); | |
620 | ||
621 | hlist_for_each_entry_rcu(dev, p, head, name_hlist) | |
622 | if (!strncmp(dev->name, name, IFNAMSIZ)) | |
623 | return dev; | |
624 | ||
625 | return NULL; | |
626 | } | |
627 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
628 | ||
1da177e4 LT |
629 | /** |
630 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 631 | * @net: the applicable net namespace |
1da177e4 LT |
632 | * @name: name to find |
633 | * | |
634 | * Find an interface by name. This can be called from any | |
635 | * context and does its own locking. The returned handle has | |
636 | * the usage count incremented and the caller must use dev_put() to | |
637 | * release it when it is no longer needed. %NULL is returned if no | |
638 | * matching device is found. | |
639 | */ | |
640 | ||
881d966b | 641 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
642 | { |
643 | struct net_device *dev; | |
644 | ||
72c9528b ED |
645 | rcu_read_lock(); |
646 | dev = dev_get_by_name_rcu(net, name); | |
1da177e4 LT |
647 | if (dev) |
648 | dev_hold(dev); | |
72c9528b | 649 | rcu_read_unlock(); |
1da177e4 LT |
650 | return dev; |
651 | } | |
d1b19dff | 652 | EXPORT_SYMBOL(dev_get_by_name); |
1da177e4 LT |
653 | |
654 | /** | |
655 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 656 | * @net: the applicable net namespace |
1da177e4 LT |
657 | * @ifindex: index of device |
658 | * | |
659 | * Search for an interface by index. Returns %NULL if the device | |
660 | * is not found or a pointer to the device. The device has not | |
661 | * had its reference counter increased so the caller must be careful | |
662 | * about locking. The caller must hold either the RTNL semaphore | |
663 | * or @dev_base_lock. | |
664 | */ | |
665 | ||
881d966b | 666 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
667 | { |
668 | struct hlist_node *p; | |
0bd8d536 ED |
669 | struct net_device *dev; |
670 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
1da177e4 | 671 | |
0bd8d536 | 672 | hlist_for_each_entry(dev, p, head, index_hlist) |
1da177e4 LT |
673 | if (dev->ifindex == ifindex) |
674 | return dev; | |
0bd8d536 | 675 | |
1da177e4 LT |
676 | return NULL; |
677 | } | |
d1b19dff | 678 | EXPORT_SYMBOL(__dev_get_by_index); |
1da177e4 | 679 | |
fb699dfd ED |
680 | /** |
681 | * dev_get_by_index_rcu - find a device by its ifindex | |
682 | * @net: the applicable net namespace | |
683 | * @ifindex: index of device | |
684 | * | |
685 | * Search for an interface by index. Returns %NULL if the device | |
686 | * is not found or a pointer to the device. The device has not | |
687 | * had its reference counter increased so the caller must be careful | |
688 | * about locking. The caller must hold RCU lock. | |
689 | */ | |
690 | ||
691 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
692 | { | |
693 | struct hlist_node *p; | |
694 | struct net_device *dev; | |
695 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
696 | ||
697 | hlist_for_each_entry_rcu(dev, p, head, index_hlist) | |
698 | if (dev->ifindex == ifindex) | |
699 | return dev; | |
700 | ||
701 | return NULL; | |
702 | } | |
703 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
704 | ||
1da177e4 LT |
705 | |
706 | /** | |
707 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 708 | * @net: the applicable net namespace |
1da177e4 LT |
709 | * @ifindex: index of device |
710 | * | |
711 | * Search for an interface by index. Returns NULL if the device | |
712 | * is not found or a pointer to the device. The device returned has | |
713 | * had a reference added and the pointer is safe until the user calls | |
714 | * dev_put to indicate they have finished with it. | |
715 | */ | |
716 | ||
881d966b | 717 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
718 | { |
719 | struct net_device *dev; | |
720 | ||
fb699dfd ED |
721 | rcu_read_lock(); |
722 | dev = dev_get_by_index_rcu(net, ifindex); | |
1da177e4 LT |
723 | if (dev) |
724 | dev_hold(dev); | |
fb699dfd | 725 | rcu_read_unlock(); |
1da177e4 LT |
726 | return dev; |
727 | } | |
d1b19dff | 728 | EXPORT_SYMBOL(dev_get_by_index); |
1da177e4 LT |
729 | |
730 | /** | |
731 | * dev_getbyhwaddr - find a device by its hardware address | |
c4ea43c5 | 732 | * @net: the applicable net namespace |
1da177e4 LT |
733 | * @type: media type of device |
734 | * @ha: hardware address | |
735 | * | |
736 | * Search for an interface by MAC address. Returns NULL if the device | |
737 | * is not found or a pointer to the device. The caller must hold the | |
738 | * rtnl semaphore. The returned device has not had its ref count increased | |
739 | * and the caller must therefore be careful about locking | |
740 | * | |
741 | * BUGS: | |
742 | * If the API was consistent this would be __dev_get_by_hwaddr | |
743 | */ | |
744 | ||
881d966b | 745 | struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha) |
1da177e4 LT |
746 | { |
747 | struct net_device *dev; | |
748 | ||
749 | ASSERT_RTNL(); | |
750 | ||
81103a52 | 751 | for_each_netdev(net, dev) |
1da177e4 LT |
752 | if (dev->type == type && |
753 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
754 | return dev; |
755 | ||
756 | return NULL; | |
1da177e4 | 757 | } |
cf309e3f JF |
758 | EXPORT_SYMBOL(dev_getbyhwaddr); |
759 | ||
881d966b | 760 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
761 | { |
762 | struct net_device *dev; | |
763 | ||
4e9cac2b | 764 | ASSERT_RTNL(); |
881d966b | 765 | for_each_netdev(net, dev) |
4e9cac2b | 766 | if (dev->type == type) |
7562f876 PE |
767 | return dev; |
768 | ||
769 | return NULL; | |
4e9cac2b | 770 | } |
4e9cac2b PM |
771 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); |
772 | ||
881d966b | 773 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b PM |
774 | { |
775 | struct net_device *dev; | |
776 | ||
777 | rtnl_lock(); | |
881d966b | 778 | dev = __dev_getfirstbyhwtype(net, type); |
4e9cac2b PM |
779 | if (dev) |
780 | dev_hold(dev); | |
1da177e4 LT |
781 | rtnl_unlock(); |
782 | return dev; | |
783 | } | |
1da177e4 LT |
784 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
785 | ||
786 | /** | |
787 | * dev_get_by_flags - find any device with given flags | |
c4ea43c5 | 788 | * @net: the applicable net namespace |
1da177e4 LT |
789 | * @if_flags: IFF_* values |
790 | * @mask: bitmask of bits in if_flags to check | |
791 | * | |
792 | * Search for any interface with the given flags. Returns NULL if a device | |
4ec93edb | 793 | * is not found or a pointer to the device. The device returned has |
1da177e4 LT |
794 | * had a reference added and the pointer is safe until the user calls |
795 | * dev_put to indicate they have finished with it. | |
796 | */ | |
797 | ||
d1b19dff ED |
798 | struct net_device *dev_get_by_flags(struct net *net, unsigned short if_flags, |
799 | unsigned short mask) | |
1da177e4 | 800 | { |
7562f876 | 801 | struct net_device *dev, *ret; |
1da177e4 | 802 | |
7562f876 | 803 | ret = NULL; |
c6d14c84 ED |
804 | rcu_read_lock(); |
805 | for_each_netdev_rcu(net, dev) { | |
1da177e4 LT |
806 | if (((dev->flags ^ if_flags) & mask) == 0) { |
807 | dev_hold(dev); | |
7562f876 | 808 | ret = dev; |
1da177e4 LT |
809 | break; |
810 | } | |
811 | } | |
c6d14c84 | 812 | rcu_read_unlock(); |
7562f876 | 813 | return ret; |
1da177e4 | 814 | } |
d1b19dff | 815 | EXPORT_SYMBOL(dev_get_by_flags); |
1da177e4 LT |
816 | |
817 | /** | |
818 | * dev_valid_name - check if name is okay for network device | |
819 | * @name: name string | |
820 | * | |
821 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
822 | * to allow sysfs to work. We also disallow any kind of |
823 | * whitespace. | |
1da177e4 | 824 | */ |
c2373ee9 | 825 | int dev_valid_name(const char *name) |
1da177e4 | 826 | { |
c7fa9d18 DM |
827 | if (*name == '\0') |
828 | return 0; | |
b6fe17d6 SH |
829 | if (strlen(name) >= IFNAMSIZ) |
830 | return 0; | |
c7fa9d18 DM |
831 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
832 | return 0; | |
833 | ||
834 | while (*name) { | |
835 | if (*name == '/' || isspace(*name)) | |
836 | return 0; | |
837 | name++; | |
838 | } | |
839 | return 1; | |
1da177e4 | 840 | } |
d1b19dff | 841 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
842 | |
843 | /** | |
b267b179 EB |
844 | * __dev_alloc_name - allocate a name for a device |
845 | * @net: network namespace to allocate the device name in | |
1da177e4 | 846 | * @name: name format string |
b267b179 | 847 | * @buf: scratch buffer and result name string |
1da177e4 LT |
848 | * |
849 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
850 | * id. It scans list of devices to build up a free map, then chooses |
851 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
852 | * while allocating the name and adding the device in order to avoid | |
853 | * duplicates. | |
854 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
855 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
856 | */ |
857 | ||
b267b179 | 858 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
859 | { |
860 | int i = 0; | |
1da177e4 LT |
861 | const char *p; |
862 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 863 | unsigned long *inuse; |
1da177e4 LT |
864 | struct net_device *d; |
865 | ||
866 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
867 | if (p) { | |
868 | /* | |
869 | * Verify the string as this thing may have come from | |
870 | * the user. There must be either one "%d" and no other "%" | |
871 | * characters. | |
872 | */ | |
873 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
874 | return -EINVAL; | |
875 | ||
876 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 877 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
878 | if (!inuse) |
879 | return -ENOMEM; | |
880 | ||
881d966b | 881 | for_each_netdev(net, d) { |
1da177e4 LT |
882 | if (!sscanf(d->name, name, &i)) |
883 | continue; | |
884 | if (i < 0 || i >= max_netdevices) | |
885 | continue; | |
886 | ||
887 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 888 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
889 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
890 | set_bit(i, inuse); | |
891 | } | |
892 | ||
893 | i = find_first_zero_bit(inuse, max_netdevices); | |
894 | free_page((unsigned long) inuse); | |
895 | } | |
896 | ||
d9031024 OP |
897 | if (buf != name) |
898 | snprintf(buf, IFNAMSIZ, name, i); | |
b267b179 | 899 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 900 | return i; |
1da177e4 LT |
901 | |
902 | /* It is possible to run out of possible slots | |
903 | * when the name is long and there isn't enough space left | |
904 | * for the digits, or if all bits are used. | |
905 | */ | |
906 | return -ENFILE; | |
907 | } | |
908 | ||
b267b179 EB |
909 | /** |
910 | * dev_alloc_name - allocate a name for a device | |
911 | * @dev: device | |
912 | * @name: name format string | |
913 | * | |
914 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
915 | * id. It scans list of devices to build up a free map, then chooses | |
916 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
917 | * while allocating the name and adding the device in order to avoid | |
918 | * duplicates. | |
919 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
920 | * Returns the number of the unit assigned or a negative errno code. | |
921 | */ | |
922 | ||
923 | int dev_alloc_name(struct net_device *dev, const char *name) | |
924 | { | |
925 | char buf[IFNAMSIZ]; | |
926 | struct net *net; | |
927 | int ret; | |
928 | ||
c346dca1 YH |
929 | BUG_ON(!dev_net(dev)); |
930 | net = dev_net(dev); | |
b267b179 EB |
931 | ret = __dev_alloc_name(net, name, buf); |
932 | if (ret >= 0) | |
933 | strlcpy(dev->name, buf, IFNAMSIZ); | |
934 | return ret; | |
935 | } | |
d1b19dff | 936 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 937 | |
d9031024 OP |
938 | static int dev_get_valid_name(struct net *net, const char *name, char *buf, |
939 | bool fmt) | |
940 | { | |
941 | if (!dev_valid_name(name)) | |
942 | return -EINVAL; | |
943 | ||
944 | if (fmt && strchr(name, '%')) | |
945 | return __dev_alloc_name(net, name, buf); | |
946 | else if (__dev_get_by_name(net, name)) | |
947 | return -EEXIST; | |
948 | else if (buf != name) | |
949 | strlcpy(buf, name, IFNAMSIZ); | |
950 | ||
951 | return 0; | |
952 | } | |
1da177e4 LT |
953 | |
954 | /** | |
955 | * dev_change_name - change name of a device | |
956 | * @dev: device | |
957 | * @newname: name (or format string) must be at least IFNAMSIZ | |
958 | * | |
959 | * Change name of a device, can pass format strings "eth%d". | |
960 | * for wildcarding. | |
961 | */ | |
cf04a4c7 | 962 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 963 | { |
fcc5a03a | 964 | char oldname[IFNAMSIZ]; |
1da177e4 | 965 | int err = 0; |
fcc5a03a | 966 | int ret; |
881d966b | 967 | struct net *net; |
1da177e4 LT |
968 | |
969 | ASSERT_RTNL(); | |
c346dca1 | 970 | BUG_ON(!dev_net(dev)); |
1da177e4 | 971 | |
c346dca1 | 972 | net = dev_net(dev); |
1da177e4 LT |
973 | if (dev->flags & IFF_UP) |
974 | return -EBUSY; | |
975 | ||
c8d90dca SH |
976 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) |
977 | return 0; | |
978 | ||
fcc5a03a HX |
979 | memcpy(oldname, dev->name, IFNAMSIZ); |
980 | ||
d9031024 OP |
981 | err = dev_get_valid_name(net, newname, dev->name, 1); |
982 | if (err < 0) | |
983 | return err; | |
1da177e4 | 984 | |
fcc5a03a | 985 | rollback: |
3891845e EB |
986 | /* For now only devices in the initial network namespace |
987 | * are in sysfs. | |
988 | */ | |
09ad9bc7 | 989 | if (net_eq(net, &init_net)) { |
3891845e EB |
990 | ret = device_rename(&dev->dev, dev->name); |
991 | if (ret) { | |
992 | memcpy(dev->name, oldname, IFNAMSIZ); | |
993 | return ret; | |
994 | } | |
dcc99773 | 995 | } |
7f988eab HX |
996 | |
997 | write_lock_bh(&dev_base_lock); | |
92749821 | 998 | hlist_del(&dev->name_hlist); |
72c9528b ED |
999 | write_unlock_bh(&dev_base_lock); |
1000 | ||
1001 | synchronize_rcu(); | |
1002 | ||
1003 | write_lock_bh(&dev_base_lock); | |
1004 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
7f988eab HX |
1005 | write_unlock_bh(&dev_base_lock); |
1006 | ||
056925ab | 1007 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1008 | ret = notifier_to_errno(ret); |
1009 | ||
1010 | if (ret) { | |
91e9c07b ED |
1011 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1012 | if (err >= 0) { | |
fcc5a03a HX |
1013 | err = ret; |
1014 | memcpy(dev->name, oldname, IFNAMSIZ); | |
1015 | goto rollback; | |
91e9c07b ED |
1016 | } else { |
1017 | printk(KERN_ERR | |
1018 | "%s: name change rollback failed: %d.\n", | |
1019 | dev->name, ret); | |
fcc5a03a HX |
1020 | } |
1021 | } | |
1da177e4 LT |
1022 | |
1023 | return err; | |
1024 | } | |
1025 | ||
0b815a1a SH |
1026 | /** |
1027 | * dev_set_alias - change ifalias of a device | |
1028 | * @dev: device | |
1029 | * @alias: name up to IFALIASZ | |
f0db275a | 1030 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1031 | * |
1032 | * Set ifalias for a device, | |
1033 | */ | |
1034 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1035 | { | |
1036 | ASSERT_RTNL(); | |
1037 | ||
1038 | if (len >= IFALIASZ) | |
1039 | return -EINVAL; | |
1040 | ||
96ca4a2c OH |
1041 | if (!len) { |
1042 | if (dev->ifalias) { | |
1043 | kfree(dev->ifalias); | |
1044 | dev->ifalias = NULL; | |
1045 | } | |
1046 | return 0; | |
1047 | } | |
1048 | ||
d1b19dff | 1049 | dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL); |
0b815a1a SH |
1050 | if (!dev->ifalias) |
1051 | return -ENOMEM; | |
1052 | ||
1053 | strlcpy(dev->ifalias, alias, len+1); | |
1054 | return len; | |
1055 | } | |
1056 | ||
1057 | ||
d8a33ac4 | 1058 | /** |
3041a069 | 1059 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1060 | * @dev: device to cause notification |
1061 | * | |
1062 | * Called to indicate a device has changed features. | |
1063 | */ | |
1064 | void netdev_features_change(struct net_device *dev) | |
1065 | { | |
056925ab | 1066 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1067 | } |
1068 | EXPORT_SYMBOL(netdev_features_change); | |
1069 | ||
1da177e4 LT |
1070 | /** |
1071 | * netdev_state_change - device changes state | |
1072 | * @dev: device to cause notification | |
1073 | * | |
1074 | * Called to indicate a device has changed state. This function calls | |
1075 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1076 | * to the routing socket. | |
1077 | */ | |
1078 | void netdev_state_change(struct net_device *dev) | |
1079 | { | |
1080 | if (dev->flags & IFF_UP) { | |
056925ab | 1081 | call_netdevice_notifiers(NETDEV_CHANGE, dev); |
1da177e4 LT |
1082 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0); |
1083 | } | |
1084 | } | |
d1b19dff | 1085 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1086 | |
75c78500 | 1087 | void netdev_bonding_change(struct net_device *dev, unsigned long event) |
c1da4ac7 | 1088 | { |
75c78500 | 1089 | call_netdevice_notifiers(event, dev); |
c1da4ac7 OG |
1090 | } |
1091 | EXPORT_SYMBOL(netdev_bonding_change); | |
1092 | ||
1da177e4 LT |
1093 | /** |
1094 | * dev_load - load a network module | |
c4ea43c5 | 1095 | * @net: the applicable net namespace |
1da177e4 LT |
1096 | * @name: name of interface |
1097 | * | |
1098 | * If a network interface is not present and the process has suitable | |
1099 | * privileges this function loads the module. If module loading is not | |
1100 | * available in this kernel then it becomes a nop. | |
1101 | */ | |
1102 | ||
881d966b | 1103 | void dev_load(struct net *net, const char *name) |
1da177e4 | 1104 | { |
4ec93edb | 1105 | struct net_device *dev; |
1da177e4 | 1106 | |
72c9528b ED |
1107 | rcu_read_lock(); |
1108 | dev = dev_get_by_name_rcu(net, name); | |
1109 | rcu_read_unlock(); | |
1da177e4 | 1110 | |
a8f80e8f | 1111 | if (!dev && capable(CAP_NET_ADMIN)) |
1da177e4 LT |
1112 | request_module("%s", name); |
1113 | } | |
d1b19dff | 1114 | EXPORT_SYMBOL(dev_load); |
1da177e4 | 1115 | |
1da177e4 LT |
1116 | /** |
1117 | * dev_open - prepare an interface for use. | |
1118 | * @dev: device to open | |
1119 | * | |
1120 | * Takes a device from down to up state. The device's private open | |
1121 | * function is invoked and then the multicast lists are loaded. Finally | |
1122 | * the device is moved into the up state and a %NETDEV_UP message is | |
1123 | * sent to the netdev notifier chain. | |
1124 | * | |
1125 | * Calling this function on an active interface is a nop. On a failure | |
1126 | * a negative errno code is returned. | |
1127 | */ | |
1128 | int dev_open(struct net_device *dev) | |
1129 | { | |
d314774c | 1130 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1131 | int ret; |
1da177e4 | 1132 | |
e46b66bc BH |
1133 | ASSERT_RTNL(); |
1134 | ||
1da177e4 LT |
1135 | /* |
1136 | * Is it already up? | |
1137 | */ | |
1138 | ||
1139 | if (dev->flags & IFF_UP) | |
1140 | return 0; | |
1141 | ||
1142 | /* | |
1143 | * Is it even present? | |
1144 | */ | |
1145 | if (!netif_device_present(dev)) | |
1146 | return -ENODEV; | |
1147 | ||
3b8bcfd5 JB |
1148 | ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); |
1149 | ret = notifier_to_errno(ret); | |
1150 | if (ret) | |
1151 | return ret; | |
1152 | ||
1da177e4 LT |
1153 | /* |
1154 | * Call device private open method | |
1155 | */ | |
1156 | set_bit(__LINK_STATE_START, &dev->state); | |
bada339b | 1157 | |
d314774c SH |
1158 | if (ops->ndo_validate_addr) |
1159 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1160 | |
d314774c SH |
1161 | if (!ret && ops->ndo_open) |
1162 | ret = ops->ndo_open(dev); | |
1da177e4 | 1163 | |
4ec93edb | 1164 | /* |
1da177e4 LT |
1165 | * If it went open OK then: |
1166 | */ | |
1167 | ||
bada339b JG |
1168 | if (ret) |
1169 | clear_bit(__LINK_STATE_START, &dev->state); | |
1170 | else { | |
1da177e4 LT |
1171 | /* |
1172 | * Set the flags. | |
1173 | */ | |
1174 | dev->flags |= IFF_UP; | |
1175 | ||
649274d9 DW |
1176 | /* |
1177 | * Enable NET_DMA | |
1178 | */ | |
b4bd07c2 | 1179 | net_dmaengine_get(); |
649274d9 | 1180 | |
1da177e4 LT |
1181 | /* |
1182 | * Initialize multicasting status | |
1183 | */ | |
4417da66 | 1184 | dev_set_rx_mode(dev); |
1da177e4 LT |
1185 | |
1186 | /* | |
1187 | * Wakeup transmit queue engine | |
1188 | */ | |
1189 | dev_activate(dev); | |
1190 | ||
1191 | /* | |
1192 | * ... and announce new interface. | |
1193 | */ | |
056925ab | 1194 | call_netdevice_notifiers(NETDEV_UP, dev); |
1da177e4 | 1195 | } |
bada339b | 1196 | |
1da177e4 LT |
1197 | return ret; |
1198 | } | |
d1b19dff | 1199 | EXPORT_SYMBOL(dev_open); |
1da177e4 LT |
1200 | |
1201 | /** | |
1202 | * dev_close - shutdown an interface. | |
1203 | * @dev: device to shutdown | |
1204 | * | |
1205 | * This function moves an active device into down state. A | |
1206 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1207 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1208 | * chain. | |
1209 | */ | |
1210 | int dev_close(struct net_device *dev) | |
1211 | { | |
d314774c | 1212 | const struct net_device_ops *ops = dev->netdev_ops; |
e46b66bc BH |
1213 | ASSERT_RTNL(); |
1214 | ||
9d5010db DM |
1215 | might_sleep(); |
1216 | ||
1da177e4 LT |
1217 | if (!(dev->flags & IFF_UP)) |
1218 | return 0; | |
1219 | ||
1220 | /* | |
1221 | * Tell people we are going down, so that they can | |
1222 | * prepare to death, when device is still operating. | |
1223 | */ | |
056925ab | 1224 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1225 | |
1da177e4 LT |
1226 | clear_bit(__LINK_STATE_START, &dev->state); |
1227 | ||
1228 | /* Synchronize to scheduled poll. We cannot touch poll list, | |
bea3348e SH |
1229 | * it can be even on different cpu. So just clear netif_running(). |
1230 | * | |
1231 | * dev->stop() will invoke napi_disable() on all of it's | |
1232 | * napi_struct instances on this device. | |
1233 | */ | |
1da177e4 | 1234 | smp_mb__after_clear_bit(); /* Commit netif_running(). */ |
1da177e4 | 1235 | |
d8b2a4d2 ML |
1236 | dev_deactivate(dev); |
1237 | ||
1da177e4 LT |
1238 | /* |
1239 | * Call the device specific close. This cannot fail. | |
1240 | * Only if device is UP | |
1241 | * | |
1242 | * We allow it to be called even after a DETACH hot-plug | |
1243 | * event. | |
1244 | */ | |
d314774c SH |
1245 | if (ops->ndo_stop) |
1246 | ops->ndo_stop(dev); | |
1da177e4 LT |
1247 | |
1248 | /* | |
1249 | * Device is now down. | |
1250 | */ | |
1251 | ||
1252 | dev->flags &= ~IFF_UP; | |
1253 | ||
1254 | /* | |
1255 | * Tell people we are down | |
1256 | */ | |
056925ab | 1257 | call_netdevice_notifiers(NETDEV_DOWN, dev); |
1da177e4 | 1258 | |
649274d9 DW |
1259 | /* |
1260 | * Shutdown NET_DMA | |
1261 | */ | |
b4bd07c2 | 1262 | net_dmaengine_put(); |
649274d9 | 1263 | |
1da177e4 LT |
1264 | return 0; |
1265 | } | |
d1b19dff | 1266 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1267 | |
1268 | ||
0187bdfb BH |
1269 | /** |
1270 | * dev_disable_lro - disable Large Receive Offload on a device | |
1271 | * @dev: device | |
1272 | * | |
1273 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1274 | * called under RTNL. This is needed if received packets may be | |
1275 | * forwarded to another interface. | |
1276 | */ | |
1277 | void dev_disable_lro(struct net_device *dev) | |
1278 | { | |
1279 | if (dev->ethtool_ops && dev->ethtool_ops->get_flags && | |
1280 | dev->ethtool_ops->set_flags) { | |
1281 | u32 flags = dev->ethtool_ops->get_flags(dev); | |
1282 | if (flags & ETH_FLAG_LRO) { | |
1283 | flags &= ~ETH_FLAG_LRO; | |
1284 | dev->ethtool_ops->set_flags(dev, flags); | |
1285 | } | |
1286 | } | |
1287 | WARN_ON(dev->features & NETIF_F_LRO); | |
1288 | } | |
1289 | EXPORT_SYMBOL(dev_disable_lro); | |
1290 | ||
1291 | ||
881d966b EB |
1292 | static int dev_boot_phase = 1; |
1293 | ||
1da177e4 LT |
1294 | /* |
1295 | * Device change register/unregister. These are not inline or static | |
1296 | * as we export them to the world. | |
1297 | */ | |
1298 | ||
1299 | /** | |
1300 | * register_netdevice_notifier - register a network notifier block | |
1301 | * @nb: notifier | |
1302 | * | |
1303 | * Register a notifier to be called when network device events occur. | |
1304 | * The notifier passed is linked into the kernel structures and must | |
1305 | * not be reused until it has been unregistered. A negative errno code | |
1306 | * is returned on a failure. | |
1307 | * | |
1308 | * When registered all registration and up events are replayed | |
4ec93edb | 1309 | * to the new notifier to allow device to have a race free |
1da177e4 LT |
1310 | * view of the network device list. |
1311 | */ | |
1312 | ||
1313 | int register_netdevice_notifier(struct notifier_block *nb) | |
1314 | { | |
1315 | struct net_device *dev; | |
fcc5a03a | 1316 | struct net_device *last; |
881d966b | 1317 | struct net *net; |
1da177e4 LT |
1318 | int err; |
1319 | ||
1320 | rtnl_lock(); | |
f07d5b94 | 1321 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1322 | if (err) |
1323 | goto unlock; | |
881d966b EB |
1324 | if (dev_boot_phase) |
1325 | goto unlock; | |
1326 | for_each_net(net) { | |
1327 | for_each_netdev(net, dev) { | |
1328 | err = nb->notifier_call(nb, NETDEV_REGISTER, dev); | |
1329 | err = notifier_to_errno(err); | |
1330 | if (err) | |
1331 | goto rollback; | |
1332 | ||
1333 | if (!(dev->flags & IFF_UP)) | |
1334 | continue; | |
1da177e4 | 1335 | |
881d966b EB |
1336 | nb->notifier_call(nb, NETDEV_UP, dev); |
1337 | } | |
1da177e4 | 1338 | } |
fcc5a03a HX |
1339 | |
1340 | unlock: | |
1da177e4 LT |
1341 | rtnl_unlock(); |
1342 | return err; | |
fcc5a03a HX |
1343 | |
1344 | rollback: | |
1345 | last = dev; | |
881d966b EB |
1346 | for_each_net(net) { |
1347 | for_each_netdev(net, dev) { | |
1348 | if (dev == last) | |
1349 | break; | |
fcc5a03a | 1350 | |
881d966b EB |
1351 | if (dev->flags & IFF_UP) { |
1352 | nb->notifier_call(nb, NETDEV_GOING_DOWN, dev); | |
1353 | nb->notifier_call(nb, NETDEV_DOWN, dev); | |
1354 | } | |
1355 | nb->notifier_call(nb, NETDEV_UNREGISTER, dev); | |
a5ee1551 | 1356 | nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev); |
fcc5a03a | 1357 | } |
fcc5a03a | 1358 | } |
c67625a1 PE |
1359 | |
1360 | raw_notifier_chain_unregister(&netdev_chain, nb); | |
fcc5a03a | 1361 | goto unlock; |
1da177e4 | 1362 | } |
d1b19dff | 1363 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1364 | |
1365 | /** | |
1366 | * unregister_netdevice_notifier - unregister a network notifier block | |
1367 | * @nb: notifier | |
1368 | * | |
1369 | * Unregister a notifier previously registered by | |
1370 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1371 | * kernel structures and may then be reused. A negative errno code | |
1372 | * is returned on a failure. | |
1373 | */ | |
1374 | ||
1375 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1376 | { | |
9f514950 HX |
1377 | int err; |
1378 | ||
1379 | rtnl_lock(); | |
f07d5b94 | 1380 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
9f514950 HX |
1381 | rtnl_unlock(); |
1382 | return err; | |
1da177e4 | 1383 | } |
d1b19dff | 1384 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 LT |
1385 | |
1386 | /** | |
1387 | * call_netdevice_notifiers - call all network notifier blocks | |
1388 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1389 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1390 | * |
1391 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1392 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1393 | */ |
1394 | ||
ad7379d4 | 1395 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1396 | { |
ad7379d4 | 1397 | return raw_notifier_call_chain(&netdev_chain, val, dev); |
1da177e4 LT |
1398 | } |
1399 | ||
1400 | /* When > 0 there are consumers of rx skb time stamps */ | |
1401 | static atomic_t netstamp_needed = ATOMIC_INIT(0); | |
1402 | ||
1403 | void net_enable_timestamp(void) | |
1404 | { | |
1405 | atomic_inc(&netstamp_needed); | |
1406 | } | |
d1b19dff | 1407 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
1408 | |
1409 | void net_disable_timestamp(void) | |
1410 | { | |
1411 | atomic_dec(&netstamp_needed); | |
1412 | } | |
d1b19dff | 1413 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 1414 | |
a61bbcf2 | 1415 | static inline void net_timestamp(struct sk_buff *skb) |
1da177e4 LT |
1416 | { |
1417 | if (atomic_read(&netstamp_needed)) | |
a61bbcf2 | 1418 | __net_timestamp(skb); |
b7aa0bf7 ED |
1419 | else |
1420 | skb->tstamp.tv64 = 0; | |
1da177e4 LT |
1421 | } |
1422 | ||
44540960 AB |
1423 | /** |
1424 | * dev_forward_skb - loopback an skb to another netif | |
1425 | * | |
1426 | * @dev: destination network device | |
1427 | * @skb: buffer to forward | |
1428 | * | |
1429 | * return values: | |
1430 | * NET_RX_SUCCESS (no congestion) | |
1431 | * NET_RX_DROP (packet was dropped) | |
1432 | * | |
1433 | * dev_forward_skb can be used for injecting an skb from the | |
1434 | * start_xmit function of one device into the receive queue | |
1435 | * of another device. | |
1436 | * | |
1437 | * The receiving device may be in another namespace, so | |
1438 | * we have to clear all information in the skb that could | |
1439 | * impact namespace isolation. | |
1440 | */ | |
1441 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
1442 | { | |
1443 | skb_orphan(skb); | |
1444 | ||
1445 | if (!(dev->flags & IFF_UP)) | |
1446 | return NET_RX_DROP; | |
1447 | ||
1448 | if (skb->len > (dev->mtu + dev->hard_header_len)) | |
1449 | return NET_RX_DROP; | |
1450 | ||
1451 | skb_dst_drop(skb); | |
1452 | skb->tstamp.tv64 = 0; | |
1453 | skb->pkt_type = PACKET_HOST; | |
1454 | skb->protocol = eth_type_trans(skb, dev); | |
1455 | skb->mark = 0; | |
1456 | secpath_reset(skb); | |
1457 | nf_reset(skb); | |
1458 | return netif_rx(skb); | |
1459 | } | |
1460 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
1461 | ||
1da177e4 LT |
1462 | /* |
1463 | * Support routine. Sends outgoing frames to any network | |
1464 | * taps currently in use. | |
1465 | */ | |
1466 | ||
f6a78bfc | 1467 | static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
1468 | { |
1469 | struct packet_type *ptype; | |
a61bbcf2 | 1470 | |
8caf1539 JP |
1471 | #ifdef CONFIG_NET_CLS_ACT |
1472 | if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS))) | |
1473 | net_timestamp(skb); | |
1474 | #else | |
a61bbcf2 | 1475 | net_timestamp(skb); |
8caf1539 | 1476 | #endif |
1da177e4 LT |
1477 | |
1478 | rcu_read_lock(); | |
1479 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
1480 | /* Never send packets back to the socket | |
1481 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1482 | */ | |
1483 | if ((ptype->dev == dev || !ptype->dev) && | |
1484 | (ptype->af_packet_priv == NULL || | |
1485 | (struct sock *)ptype->af_packet_priv != skb->sk)) { | |
d1b19dff | 1486 | struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC); |
1da177e4 LT |
1487 | if (!skb2) |
1488 | break; | |
1489 | ||
1490 | /* skb->nh should be correctly | |
1491 | set by sender, so that the second statement is | |
1492 | just protection against buggy protocols. | |
1493 | */ | |
459a98ed | 1494 | skb_reset_mac_header(skb2); |
1da177e4 | 1495 | |
d56f90a7 | 1496 | if (skb_network_header(skb2) < skb2->data || |
27a884dc | 1497 | skb2->network_header > skb2->tail) { |
1da177e4 LT |
1498 | if (net_ratelimit()) |
1499 | printk(KERN_CRIT "protocol %04x is " | |
1500 | "buggy, dev %s\n", | |
1501 | skb2->protocol, dev->name); | |
c1d2bbe1 | 1502 | skb_reset_network_header(skb2); |
1da177e4 LT |
1503 | } |
1504 | ||
b0e380b1 | 1505 | skb2->transport_header = skb2->network_header; |
1da177e4 | 1506 | skb2->pkt_type = PACKET_OUTGOING; |
f2ccd8fa | 1507 | ptype->func(skb2, skb->dev, ptype, skb->dev); |
1da177e4 LT |
1508 | } |
1509 | } | |
1510 | rcu_read_unlock(); | |
1511 | } | |
1512 | ||
56079431 | 1513 | |
def82a1d | 1514 | static inline void __netif_reschedule(struct Qdisc *q) |
56079431 | 1515 | { |
def82a1d JP |
1516 | struct softnet_data *sd; |
1517 | unsigned long flags; | |
56079431 | 1518 | |
def82a1d JP |
1519 | local_irq_save(flags); |
1520 | sd = &__get_cpu_var(softnet_data); | |
1521 | q->next_sched = sd->output_queue; | |
1522 | sd->output_queue = q; | |
1523 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
1524 | local_irq_restore(flags); | |
1525 | } | |
1526 | ||
1527 | void __netif_schedule(struct Qdisc *q) | |
1528 | { | |
1529 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
1530 | __netif_reschedule(q); | |
56079431 DV |
1531 | } |
1532 | EXPORT_SYMBOL(__netif_schedule); | |
1533 | ||
bea3348e | 1534 | void dev_kfree_skb_irq(struct sk_buff *skb) |
56079431 | 1535 | { |
bea3348e SH |
1536 | if (atomic_dec_and_test(&skb->users)) { |
1537 | struct softnet_data *sd; | |
1538 | unsigned long flags; | |
56079431 | 1539 | |
bea3348e SH |
1540 | local_irq_save(flags); |
1541 | sd = &__get_cpu_var(softnet_data); | |
1542 | skb->next = sd->completion_queue; | |
1543 | sd->completion_queue = skb; | |
1544 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
1545 | local_irq_restore(flags); | |
1546 | } | |
56079431 | 1547 | } |
bea3348e | 1548 | EXPORT_SYMBOL(dev_kfree_skb_irq); |
56079431 DV |
1549 | |
1550 | void dev_kfree_skb_any(struct sk_buff *skb) | |
1551 | { | |
1552 | if (in_irq() || irqs_disabled()) | |
1553 | dev_kfree_skb_irq(skb); | |
1554 | else | |
1555 | dev_kfree_skb(skb); | |
1556 | } | |
1557 | EXPORT_SYMBOL(dev_kfree_skb_any); | |
1558 | ||
1559 | ||
bea3348e SH |
1560 | /** |
1561 | * netif_device_detach - mark device as removed | |
1562 | * @dev: network device | |
1563 | * | |
1564 | * Mark device as removed from system and therefore no longer available. | |
1565 | */ | |
56079431 DV |
1566 | void netif_device_detach(struct net_device *dev) |
1567 | { | |
1568 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
1569 | netif_running(dev)) { | |
d543103a | 1570 | netif_tx_stop_all_queues(dev); |
56079431 DV |
1571 | } |
1572 | } | |
1573 | EXPORT_SYMBOL(netif_device_detach); | |
1574 | ||
bea3348e SH |
1575 | /** |
1576 | * netif_device_attach - mark device as attached | |
1577 | * @dev: network device | |
1578 | * | |
1579 | * Mark device as attached from system and restart if needed. | |
1580 | */ | |
56079431 DV |
1581 | void netif_device_attach(struct net_device *dev) |
1582 | { | |
1583 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
1584 | netif_running(dev)) { | |
d543103a | 1585 | netif_tx_wake_all_queues(dev); |
4ec93edb | 1586 | __netdev_watchdog_up(dev); |
56079431 DV |
1587 | } |
1588 | } | |
1589 | EXPORT_SYMBOL(netif_device_attach); | |
1590 | ||
6de329e2 BH |
1591 | static bool can_checksum_protocol(unsigned long features, __be16 protocol) |
1592 | { | |
1593 | return ((features & NETIF_F_GEN_CSUM) || | |
1594 | ((features & NETIF_F_IP_CSUM) && | |
1595 | protocol == htons(ETH_P_IP)) || | |
1596 | ((features & NETIF_F_IPV6_CSUM) && | |
1c8dbcf6 YZ |
1597 | protocol == htons(ETH_P_IPV6)) || |
1598 | ((features & NETIF_F_FCOE_CRC) && | |
1599 | protocol == htons(ETH_P_FCOE))); | |
6de329e2 BH |
1600 | } |
1601 | ||
1602 | static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb) | |
1603 | { | |
1604 | if (can_checksum_protocol(dev->features, skb->protocol)) | |
1605 | return true; | |
1606 | ||
1607 | if (skb->protocol == htons(ETH_P_8021Q)) { | |
1608 | struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data; | |
1609 | if (can_checksum_protocol(dev->features & dev->vlan_features, | |
1610 | veh->h_vlan_encapsulated_proto)) | |
1611 | return true; | |
1612 | } | |
1613 | ||
1614 | return false; | |
1615 | } | |
56079431 | 1616 | |
1da177e4 LT |
1617 | /* |
1618 | * Invalidate hardware checksum when packet is to be mangled, and | |
1619 | * complete checksum manually on outgoing path. | |
1620 | */ | |
84fa7933 | 1621 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 1622 | { |
d3bc23e7 | 1623 | __wsum csum; |
663ead3b | 1624 | int ret = 0, offset; |
1da177e4 | 1625 | |
84fa7933 | 1626 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
1627 | goto out_set_summed; |
1628 | ||
1629 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
a430a43d HX |
1630 | /* Let GSO fix up the checksum. */ |
1631 | goto out_set_summed; | |
1da177e4 LT |
1632 | } |
1633 | ||
a030847e HX |
1634 | offset = skb->csum_start - skb_headroom(skb); |
1635 | BUG_ON(offset >= skb_headlen(skb)); | |
1636 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
1637 | ||
1638 | offset += skb->csum_offset; | |
1639 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
1640 | ||
1641 | if (skb_cloned(skb) && | |
1642 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1da177e4 LT |
1643 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
1644 | if (ret) | |
1645 | goto out; | |
1646 | } | |
1647 | ||
a030847e | 1648 | *(__sum16 *)(skb->data + offset) = csum_fold(csum); |
a430a43d | 1649 | out_set_summed: |
1da177e4 | 1650 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 1651 | out: |
1da177e4 LT |
1652 | return ret; |
1653 | } | |
d1b19dff | 1654 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 1655 | |
f6a78bfc HX |
1656 | /** |
1657 | * skb_gso_segment - Perform segmentation on skb. | |
1658 | * @skb: buffer to segment | |
576a30eb | 1659 | * @features: features for the output path (see dev->features) |
f6a78bfc HX |
1660 | * |
1661 | * This function segments the given skb and returns a list of segments. | |
576a30eb HX |
1662 | * |
1663 | * It may return NULL if the skb requires no segmentation. This is | |
1664 | * only possible when GSO is used for verifying header integrity. | |
f6a78bfc | 1665 | */ |
576a30eb | 1666 | struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features) |
f6a78bfc HX |
1667 | { |
1668 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
1669 | struct packet_type *ptype; | |
252e3346 | 1670 | __be16 type = skb->protocol; |
a430a43d | 1671 | int err; |
f6a78bfc | 1672 | |
459a98ed | 1673 | skb_reset_mac_header(skb); |
b0e380b1 | 1674 | skb->mac_len = skb->network_header - skb->mac_header; |
f6a78bfc HX |
1675 | __skb_pull(skb, skb->mac_len); |
1676 | ||
67fd1a73 HX |
1677 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { |
1678 | struct net_device *dev = skb->dev; | |
1679 | struct ethtool_drvinfo info = {}; | |
1680 | ||
1681 | if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo) | |
1682 | dev->ethtool_ops->get_drvinfo(dev, &info); | |
1683 | ||
1684 | WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d " | |
1685 | "ip_summed=%d", | |
1686 | info.driver, dev ? dev->features : 0L, | |
1687 | skb->sk ? skb->sk->sk_route_caps : 0L, | |
1688 | skb->len, skb->data_len, skb->ip_summed); | |
1689 | ||
a430a43d HX |
1690 | if (skb_header_cloned(skb) && |
1691 | (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))) | |
1692 | return ERR_PTR(err); | |
1693 | } | |
1694 | ||
f6a78bfc | 1695 | rcu_read_lock(); |
82d8a867 PE |
1696 | list_for_each_entry_rcu(ptype, |
1697 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
f6a78bfc | 1698 | if (ptype->type == type && !ptype->dev && ptype->gso_segment) { |
84fa7933 | 1699 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { |
a430a43d HX |
1700 | err = ptype->gso_send_check(skb); |
1701 | segs = ERR_PTR(err); | |
1702 | if (err || skb_gso_ok(skb, features)) | |
1703 | break; | |
d56f90a7 ACM |
1704 | __skb_push(skb, (skb->data - |
1705 | skb_network_header(skb))); | |
a430a43d | 1706 | } |
576a30eb | 1707 | segs = ptype->gso_segment(skb, features); |
f6a78bfc HX |
1708 | break; |
1709 | } | |
1710 | } | |
1711 | rcu_read_unlock(); | |
1712 | ||
98e399f8 | 1713 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 1714 | |
f6a78bfc HX |
1715 | return segs; |
1716 | } | |
f6a78bfc HX |
1717 | EXPORT_SYMBOL(skb_gso_segment); |
1718 | ||
fb286bb2 HX |
1719 | /* Take action when hardware reception checksum errors are detected. */ |
1720 | #ifdef CONFIG_BUG | |
1721 | void netdev_rx_csum_fault(struct net_device *dev) | |
1722 | { | |
1723 | if (net_ratelimit()) { | |
4ec93edb | 1724 | printk(KERN_ERR "%s: hw csum failure.\n", |
246a4212 | 1725 | dev ? dev->name : "<unknown>"); |
fb286bb2 HX |
1726 | dump_stack(); |
1727 | } | |
1728 | } | |
1729 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
1730 | #endif | |
1731 | ||
1da177e4 LT |
1732 | /* Actually, we should eliminate this check as soon as we know, that: |
1733 | * 1. IOMMU is present and allows to map all the memory. | |
1734 | * 2. No high memory really exists on this machine. | |
1735 | */ | |
1736 | ||
1737 | static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb) | |
1738 | { | |
3d3a8533 | 1739 | #ifdef CONFIG_HIGHMEM |
1da177e4 LT |
1740 | int i; |
1741 | ||
1742 | if (dev->features & NETIF_F_HIGHDMA) | |
1743 | return 0; | |
1744 | ||
1745 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) | |
1746 | if (PageHighMem(skb_shinfo(skb)->frags[i].page)) | |
1747 | return 1; | |
1748 | ||
3d3a8533 | 1749 | #endif |
1da177e4 LT |
1750 | return 0; |
1751 | } | |
1da177e4 | 1752 | |
f6a78bfc HX |
1753 | struct dev_gso_cb { |
1754 | void (*destructor)(struct sk_buff *skb); | |
1755 | }; | |
1756 | ||
1757 | #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb) | |
1758 | ||
1759 | static void dev_gso_skb_destructor(struct sk_buff *skb) | |
1760 | { | |
1761 | struct dev_gso_cb *cb; | |
1762 | ||
1763 | do { | |
1764 | struct sk_buff *nskb = skb->next; | |
1765 | ||
1766 | skb->next = nskb->next; | |
1767 | nskb->next = NULL; | |
1768 | kfree_skb(nskb); | |
1769 | } while (skb->next); | |
1770 | ||
1771 | cb = DEV_GSO_CB(skb); | |
1772 | if (cb->destructor) | |
1773 | cb->destructor(skb); | |
1774 | } | |
1775 | ||
1776 | /** | |
1777 | * dev_gso_segment - Perform emulated hardware segmentation on skb. | |
1778 | * @skb: buffer to segment | |
1779 | * | |
1780 | * This function segments the given skb and stores the list of segments | |
1781 | * in skb->next. | |
1782 | */ | |
1783 | static int dev_gso_segment(struct sk_buff *skb) | |
1784 | { | |
1785 | struct net_device *dev = skb->dev; | |
1786 | struct sk_buff *segs; | |
576a30eb HX |
1787 | int features = dev->features & ~(illegal_highdma(dev, skb) ? |
1788 | NETIF_F_SG : 0); | |
1789 | ||
1790 | segs = skb_gso_segment(skb, features); | |
1791 | ||
1792 | /* Verifying header integrity only. */ | |
1793 | if (!segs) | |
1794 | return 0; | |
f6a78bfc | 1795 | |
801678c5 | 1796 | if (IS_ERR(segs)) |
f6a78bfc HX |
1797 | return PTR_ERR(segs); |
1798 | ||
1799 | skb->next = segs; | |
1800 | DEV_GSO_CB(skb)->destructor = skb->destructor; | |
1801 | skb->destructor = dev_gso_skb_destructor; | |
1802 | ||
1803 | return 0; | |
1804 | } | |
1805 | ||
fd2ea0a7 DM |
1806 | int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev, |
1807 | struct netdev_queue *txq) | |
f6a78bfc | 1808 | { |
00829823 | 1809 | const struct net_device_ops *ops = dev->netdev_ops; |
572a9d7b | 1810 | int rc = NETDEV_TX_OK; |
00829823 | 1811 | |
f6a78bfc | 1812 | if (likely(!skb->next)) { |
9be9a6b9 | 1813 | if (!list_empty(&ptype_all)) |
f6a78bfc HX |
1814 | dev_queue_xmit_nit(skb, dev); |
1815 | ||
576a30eb HX |
1816 | if (netif_needs_gso(dev, skb)) { |
1817 | if (unlikely(dev_gso_segment(skb))) | |
1818 | goto out_kfree_skb; | |
1819 | if (skb->next) | |
1820 | goto gso; | |
1821 | } | |
f6a78bfc | 1822 | |
93f154b5 ED |
1823 | /* |
1824 | * If device doesnt need skb->dst, release it right now while | |
1825 | * its hot in this cpu cache | |
1826 | */ | |
adf30907 ED |
1827 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) |
1828 | skb_dst_drop(skb); | |
1829 | ||
ac45f602 | 1830 | rc = ops->ndo_start_xmit(skb, dev); |
ec634fe3 | 1831 | if (rc == NETDEV_TX_OK) |
08baf561 | 1832 | txq_trans_update(txq); |
ac45f602 PO |
1833 | /* |
1834 | * TODO: if skb_orphan() was called by | |
1835 | * dev->hard_start_xmit() (for example, the unmodified | |
1836 | * igb driver does that; bnx2 doesn't), then | |
1837 | * skb_tx_software_timestamp() will be unable to send | |
1838 | * back the time stamp. | |
1839 | * | |
1840 | * How can this be prevented? Always create another | |
1841 | * reference to the socket before calling | |
1842 | * dev->hard_start_xmit()? Prevent that skb_orphan() | |
1843 | * does anything in dev->hard_start_xmit() by clearing | |
1844 | * the skb destructor before the call and restoring it | |
1845 | * afterwards, then doing the skb_orphan() ourselves? | |
1846 | */ | |
ac45f602 | 1847 | return rc; |
f6a78bfc HX |
1848 | } |
1849 | ||
576a30eb | 1850 | gso: |
f6a78bfc HX |
1851 | do { |
1852 | struct sk_buff *nskb = skb->next; | |
f6a78bfc HX |
1853 | |
1854 | skb->next = nskb->next; | |
1855 | nskb->next = NULL; | |
068a2de5 KK |
1856 | |
1857 | /* | |
1858 | * If device doesnt need nskb->dst, release it right now while | |
1859 | * its hot in this cpu cache | |
1860 | */ | |
1861 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
1862 | skb_dst_drop(nskb); | |
1863 | ||
00829823 | 1864 | rc = ops->ndo_start_xmit(nskb, dev); |
ec634fe3 | 1865 | if (unlikely(rc != NETDEV_TX_OK)) { |
572a9d7b PM |
1866 | if (rc & ~NETDEV_TX_MASK) |
1867 | goto out_kfree_gso_skb; | |
f54d9e8d | 1868 | nskb->next = skb->next; |
f6a78bfc HX |
1869 | skb->next = nskb; |
1870 | return rc; | |
1871 | } | |
08baf561 | 1872 | txq_trans_update(txq); |
fd2ea0a7 | 1873 | if (unlikely(netif_tx_queue_stopped(txq) && skb->next)) |
f54d9e8d | 1874 | return NETDEV_TX_BUSY; |
f6a78bfc | 1875 | } while (skb->next); |
4ec93edb | 1876 | |
572a9d7b PM |
1877 | out_kfree_gso_skb: |
1878 | if (likely(skb->next == NULL)) | |
1879 | skb->destructor = DEV_GSO_CB(skb)->destructor; | |
f6a78bfc HX |
1880 | out_kfree_skb: |
1881 | kfree_skb(skb); | |
572a9d7b | 1882 | return rc; |
f6a78bfc HX |
1883 | } |
1884 | ||
7019298a | 1885 | static u32 skb_tx_hashrnd; |
b6b2fed1 | 1886 | |
9247744e | 1887 | u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb) |
8f0f2223 | 1888 | { |
7019298a | 1889 | u32 hash; |
b6b2fed1 | 1890 | |
513de11b DM |
1891 | if (skb_rx_queue_recorded(skb)) { |
1892 | hash = skb_get_rx_queue(skb); | |
d1b19dff | 1893 | while (unlikely(hash >= dev->real_num_tx_queues)) |
513de11b DM |
1894 | hash -= dev->real_num_tx_queues; |
1895 | return hash; | |
1896 | } | |
ec581f6a ED |
1897 | |
1898 | if (skb->sk && skb->sk->sk_hash) | |
7019298a | 1899 | hash = skb->sk->sk_hash; |
ec581f6a | 1900 | else |
7019298a | 1901 | hash = skb->protocol; |
d5a9e24a | 1902 | |
7019298a | 1903 | hash = jhash_1word(hash, skb_tx_hashrnd); |
b6b2fed1 DM |
1904 | |
1905 | return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32); | |
8f0f2223 | 1906 | } |
9247744e | 1907 | EXPORT_SYMBOL(skb_tx_hash); |
8f0f2223 | 1908 | |
ed04642f ED |
1909 | static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index) |
1910 | { | |
1911 | if (unlikely(queue_index >= dev->real_num_tx_queues)) { | |
1912 | if (net_ratelimit()) { | |
1913 | WARN(1, "%s selects TX queue %d, but " | |
1914 | "real number of TX queues is %d\n", | |
1915 | dev->name, queue_index, | |
1916 | dev->real_num_tx_queues); | |
1917 | } | |
1918 | return 0; | |
1919 | } | |
1920 | return queue_index; | |
1921 | } | |
1922 | ||
e8a0464c DM |
1923 | static struct netdev_queue *dev_pick_tx(struct net_device *dev, |
1924 | struct sk_buff *skb) | |
1925 | { | |
a4ee3ce3 KK |
1926 | u16 queue_index; |
1927 | struct sock *sk = skb->sk; | |
1928 | ||
1929 | if (sk_tx_queue_recorded(sk)) { | |
1930 | queue_index = sk_tx_queue_get(sk); | |
1931 | } else { | |
1932 | const struct net_device_ops *ops = dev->netdev_ops; | |
1933 | ||
1934 | if (ops->ndo_select_queue) { | |
1935 | queue_index = ops->ndo_select_queue(dev, skb); | |
ed04642f | 1936 | queue_index = dev_cap_txqueue(dev, queue_index); |
a4ee3ce3 KK |
1937 | } else { |
1938 | queue_index = 0; | |
1939 | if (dev->real_num_tx_queues > 1) | |
1940 | queue_index = skb_tx_hash(dev, skb); | |
fd2ea0a7 | 1941 | |
a4ee3ce3 KK |
1942 | if (sk && sk->sk_dst_cache) |
1943 | sk_tx_queue_set(sk, queue_index); | |
1944 | } | |
1945 | } | |
eae792b7 | 1946 | |
fd2ea0a7 DM |
1947 | skb_set_queue_mapping(skb, queue_index); |
1948 | return netdev_get_tx_queue(dev, queue_index); | |
e8a0464c DM |
1949 | } |
1950 | ||
bbd8a0d3 KK |
1951 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
1952 | struct net_device *dev, | |
1953 | struct netdev_queue *txq) | |
1954 | { | |
1955 | spinlock_t *root_lock = qdisc_lock(q); | |
1956 | int rc; | |
1957 | ||
1958 | spin_lock(root_lock); | |
1959 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
1960 | kfree_skb(skb); | |
1961 | rc = NET_XMIT_DROP; | |
1962 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
1963 | !test_and_set_bit(__QDISC_STATE_RUNNING, &q->state)) { | |
1964 | /* | |
1965 | * This is a work-conserving queue; there are no old skbs | |
1966 | * waiting to be sent out; and the qdisc is not running - | |
1967 | * xmit the skb directly. | |
1968 | */ | |
1969 | __qdisc_update_bstats(q, skb->len); | |
1970 | if (sch_direct_xmit(skb, q, dev, txq, root_lock)) | |
1971 | __qdisc_run(q); | |
1972 | else | |
1973 | clear_bit(__QDISC_STATE_RUNNING, &q->state); | |
1974 | ||
1975 | rc = NET_XMIT_SUCCESS; | |
1976 | } else { | |
1977 | rc = qdisc_enqueue_root(skb, q); | |
1978 | qdisc_run(q); | |
1979 | } | |
1980 | spin_unlock(root_lock); | |
1981 | ||
1982 | return rc; | |
1983 | } | |
1984 | ||
d29f749e DJ |
1985 | /** |
1986 | * dev_queue_xmit - transmit a buffer | |
1987 | * @skb: buffer to transmit | |
1988 | * | |
1989 | * Queue a buffer for transmission to a network device. The caller must | |
1990 | * have set the device and priority and built the buffer before calling | |
1991 | * this function. The function can be called from an interrupt. | |
1992 | * | |
1993 | * A negative errno code is returned on a failure. A success does not | |
1994 | * guarantee the frame will be transmitted as it may be dropped due | |
1995 | * to congestion or traffic shaping. | |
1996 | * | |
1997 | * ----------------------------------------------------------------------------------- | |
1998 | * I notice this method can also return errors from the queue disciplines, | |
1999 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
2000 | * be positive. | |
2001 | * | |
2002 | * Regardless of the return value, the skb is consumed, so it is currently | |
2003 | * difficult to retry a send to this method. (You can bump the ref count | |
2004 | * before sending to hold a reference for retry if you are careful.) | |
2005 | * | |
2006 | * When calling this method, interrupts MUST be enabled. This is because | |
2007 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
2008 | * --BLG | |
2009 | */ | |
1da177e4 LT |
2010 | int dev_queue_xmit(struct sk_buff *skb) |
2011 | { | |
2012 | struct net_device *dev = skb->dev; | |
dc2b4847 | 2013 | struct netdev_queue *txq; |
1da177e4 LT |
2014 | struct Qdisc *q; |
2015 | int rc = -ENOMEM; | |
2016 | ||
f6a78bfc HX |
2017 | /* GSO will handle the following emulations directly. */ |
2018 | if (netif_needs_gso(dev, skb)) | |
2019 | goto gso; | |
2020 | ||
4cf704fb | 2021 | if (skb_has_frags(skb) && |
1da177e4 | 2022 | !(dev->features & NETIF_F_FRAGLIST) && |
364c6bad | 2023 | __skb_linearize(skb)) |
1da177e4 LT |
2024 | goto out_kfree_skb; |
2025 | ||
2026 | /* Fragmented skb is linearized if device does not support SG, | |
2027 | * or if at least one of fragments is in highmem and device | |
2028 | * does not support DMA from it. | |
2029 | */ | |
2030 | if (skb_shinfo(skb)->nr_frags && | |
2031 | (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) && | |
364c6bad | 2032 | __skb_linearize(skb)) |
1da177e4 LT |
2033 | goto out_kfree_skb; |
2034 | ||
2035 | /* If packet is not checksummed and device does not support | |
2036 | * checksumming for this protocol, complete checksumming here. | |
2037 | */ | |
663ead3b HX |
2038 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
2039 | skb_set_transport_header(skb, skb->csum_start - | |
2040 | skb_headroom(skb)); | |
6de329e2 BH |
2041 | if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb)) |
2042 | goto out_kfree_skb; | |
663ead3b | 2043 | } |
1da177e4 | 2044 | |
f6a78bfc | 2045 | gso: |
4ec93edb YH |
2046 | /* Disable soft irqs for various locks below. Also |
2047 | * stops preemption for RCU. | |
1da177e4 | 2048 | */ |
4ec93edb | 2049 | rcu_read_lock_bh(); |
1da177e4 | 2050 | |
eae792b7 | 2051 | txq = dev_pick_tx(dev, skb); |
b0e1e646 | 2052 | q = rcu_dereference(txq->qdisc); |
37437bb2 | 2053 | |
1da177e4 | 2054 | #ifdef CONFIG_NET_CLS_ACT |
d1b19dff | 2055 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS); |
1da177e4 LT |
2056 | #endif |
2057 | if (q->enqueue) { | |
bbd8a0d3 | 2058 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 2059 | goto out; |
1da177e4 LT |
2060 | } |
2061 | ||
2062 | /* The device has no queue. Common case for software devices: | |
2063 | loopback, all the sorts of tunnels... | |
2064 | ||
932ff279 HX |
2065 | Really, it is unlikely that netif_tx_lock protection is necessary |
2066 | here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
1da177e4 LT |
2067 | counters.) |
2068 | However, it is possible, that they rely on protection | |
2069 | made by us here. | |
2070 | ||
2071 | Check this and shot the lock. It is not prone from deadlocks. | |
2072 | Either shot noqueue qdisc, it is even simpler 8) | |
2073 | */ | |
2074 | if (dev->flags & IFF_UP) { | |
2075 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
2076 | ||
c773e847 | 2077 | if (txq->xmit_lock_owner != cpu) { |
1da177e4 | 2078 | |
c773e847 | 2079 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 2080 | |
fd2ea0a7 | 2081 | if (!netif_tx_queue_stopped(txq)) { |
572a9d7b PM |
2082 | rc = dev_hard_start_xmit(skb, dev, txq); |
2083 | if (dev_xmit_complete(rc)) { | |
c773e847 | 2084 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
2085 | goto out; |
2086 | } | |
2087 | } | |
c773e847 | 2088 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
2089 | if (net_ratelimit()) |
2090 | printk(KERN_CRIT "Virtual device %s asks to " | |
2091 | "queue packet!\n", dev->name); | |
2092 | } else { | |
2093 | /* Recursion is detected! It is possible, | |
2094 | * unfortunately */ | |
2095 | if (net_ratelimit()) | |
2096 | printk(KERN_CRIT "Dead loop on virtual device " | |
2097 | "%s, fix it urgently!\n", dev->name); | |
2098 | } | |
2099 | } | |
2100 | ||
2101 | rc = -ENETDOWN; | |
d4828d85 | 2102 | rcu_read_unlock_bh(); |
1da177e4 LT |
2103 | |
2104 | out_kfree_skb: | |
2105 | kfree_skb(skb); | |
2106 | return rc; | |
2107 | out: | |
d4828d85 | 2108 | rcu_read_unlock_bh(); |
1da177e4 LT |
2109 | return rc; |
2110 | } | |
d1b19dff | 2111 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 LT |
2112 | |
2113 | ||
2114 | /*======================================================================= | |
2115 | Receiver routines | |
2116 | =======================================================================*/ | |
2117 | ||
6b2bedc3 SH |
2118 | int netdev_max_backlog __read_mostly = 1000; |
2119 | int netdev_budget __read_mostly = 300; | |
2120 | int weight_p __read_mostly = 64; /* old backlog weight */ | |
1da177e4 LT |
2121 | |
2122 | DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, }; | |
2123 | ||
2124 | ||
1da177e4 LT |
2125 | /** |
2126 | * netif_rx - post buffer to the network code | |
2127 | * @skb: buffer to post | |
2128 | * | |
2129 | * This function receives a packet from a device driver and queues it for | |
2130 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
2131 | * may be dropped during processing for congestion control or by the | |
2132 | * protocol layers. | |
2133 | * | |
2134 | * return values: | |
2135 | * NET_RX_SUCCESS (no congestion) | |
1da177e4 LT |
2136 | * NET_RX_DROP (packet was dropped) |
2137 | * | |
2138 | */ | |
2139 | ||
2140 | int netif_rx(struct sk_buff *skb) | |
2141 | { | |
1da177e4 LT |
2142 | struct softnet_data *queue; |
2143 | unsigned long flags; | |
2144 | ||
2145 | /* if netpoll wants it, pretend we never saw it */ | |
2146 | if (netpoll_rx(skb)) | |
2147 | return NET_RX_DROP; | |
2148 | ||
b7aa0bf7 | 2149 | if (!skb->tstamp.tv64) |
a61bbcf2 | 2150 | net_timestamp(skb); |
1da177e4 LT |
2151 | |
2152 | /* | |
2153 | * The code is rearranged so that the path is the most | |
2154 | * short when CPU is congested, but is still operating. | |
2155 | */ | |
2156 | local_irq_save(flags); | |
1da177e4 LT |
2157 | queue = &__get_cpu_var(softnet_data); |
2158 | ||
2159 | __get_cpu_var(netdev_rx_stat).total++; | |
2160 | if (queue->input_pkt_queue.qlen <= netdev_max_backlog) { | |
2161 | if (queue->input_pkt_queue.qlen) { | |
1da177e4 | 2162 | enqueue: |
1da177e4 | 2163 | __skb_queue_tail(&queue->input_pkt_queue, skb); |
1da177e4 | 2164 | local_irq_restore(flags); |
34008d8c | 2165 | return NET_RX_SUCCESS; |
1da177e4 LT |
2166 | } |
2167 | ||
bea3348e | 2168 | napi_schedule(&queue->backlog); |
1da177e4 LT |
2169 | goto enqueue; |
2170 | } | |
2171 | ||
1da177e4 LT |
2172 | __get_cpu_var(netdev_rx_stat).dropped++; |
2173 | local_irq_restore(flags); | |
2174 | ||
2175 | kfree_skb(skb); | |
2176 | return NET_RX_DROP; | |
2177 | } | |
d1b19dff | 2178 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
2179 | |
2180 | int netif_rx_ni(struct sk_buff *skb) | |
2181 | { | |
2182 | int err; | |
2183 | ||
2184 | preempt_disable(); | |
2185 | err = netif_rx(skb); | |
2186 | if (local_softirq_pending()) | |
2187 | do_softirq(); | |
2188 | preempt_enable(); | |
2189 | ||
2190 | return err; | |
2191 | } | |
1da177e4 LT |
2192 | EXPORT_SYMBOL(netif_rx_ni); |
2193 | ||
1da177e4 LT |
2194 | static void net_tx_action(struct softirq_action *h) |
2195 | { | |
2196 | struct softnet_data *sd = &__get_cpu_var(softnet_data); | |
2197 | ||
2198 | if (sd->completion_queue) { | |
2199 | struct sk_buff *clist; | |
2200 | ||
2201 | local_irq_disable(); | |
2202 | clist = sd->completion_queue; | |
2203 | sd->completion_queue = NULL; | |
2204 | local_irq_enable(); | |
2205 | ||
2206 | while (clist) { | |
2207 | struct sk_buff *skb = clist; | |
2208 | clist = clist->next; | |
2209 | ||
547b792c | 2210 | WARN_ON(atomic_read(&skb->users)); |
1da177e4 LT |
2211 | __kfree_skb(skb); |
2212 | } | |
2213 | } | |
2214 | ||
2215 | if (sd->output_queue) { | |
37437bb2 | 2216 | struct Qdisc *head; |
1da177e4 LT |
2217 | |
2218 | local_irq_disable(); | |
2219 | head = sd->output_queue; | |
2220 | sd->output_queue = NULL; | |
2221 | local_irq_enable(); | |
2222 | ||
2223 | while (head) { | |
37437bb2 DM |
2224 | struct Qdisc *q = head; |
2225 | spinlock_t *root_lock; | |
2226 | ||
1da177e4 LT |
2227 | head = head->next_sched; |
2228 | ||
5fb66229 | 2229 | root_lock = qdisc_lock(q); |
37437bb2 | 2230 | if (spin_trylock(root_lock)) { |
def82a1d JP |
2231 | smp_mb__before_clear_bit(); |
2232 | clear_bit(__QDISC_STATE_SCHED, | |
2233 | &q->state); | |
37437bb2 DM |
2234 | qdisc_run(q); |
2235 | spin_unlock(root_lock); | |
1da177e4 | 2236 | } else { |
195648bb | 2237 | if (!test_bit(__QDISC_STATE_DEACTIVATED, |
e8a83e10 | 2238 | &q->state)) { |
195648bb | 2239 | __netif_reschedule(q); |
e8a83e10 JP |
2240 | } else { |
2241 | smp_mb__before_clear_bit(); | |
2242 | clear_bit(__QDISC_STATE_SCHED, | |
2243 | &q->state); | |
2244 | } | |
1da177e4 LT |
2245 | } |
2246 | } | |
2247 | } | |
2248 | } | |
2249 | ||
6f05f629 SH |
2250 | static inline int deliver_skb(struct sk_buff *skb, |
2251 | struct packet_type *pt_prev, | |
2252 | struct net_device *orig_dev) | |
1da177e4 LT |
2253 | { |
2254 | atomic_inc(&skb->users); | |
f2ccd8fa | 2255 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
1da177e4 LT |
2256 | } |
2257 | ||
2258 | #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE) | |
da678292 MM |
2259 | |
2260 | #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE) | |
2261 | /* This hook is defined here for ATM LANE */ | |
2262 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
2263 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 2264 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 2265 | #endif |
1da177e4 | 2266 | |
6229e362 SH |
2267 | /* |
2268 | * If bridge module is loaded call bridging hook. | |
2269 | * returns NULL if packet was consumed. | |
2270 | */ | |
2271 | struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p, | |
2272 | struct sk_buff *skb) __read_mostly; | |
4fb019a0 | 2273 | EXPORT_SYMBOL_GPL(br_handle_frame_hook); |
da678292 | 2274 | |
6229e362 SH |
2275 | static inline struct sk_buff *handle_bridge(struct sk_buff *skb, |
2276 | struct packet_type **pt_prev, int *ret, | |
2277 | struct net_device *orig_dev) | |
1da177e4 LT |
2278 | { |
2279 | struct net_bridge_port *port; | |
2280 | ||
6229e362 SH |
2281 | if (skb->pkt_type == PACKET_LOOPBACK || |
2282 | (port = rcu_dereference(skb->dev->br_port)) == NULL) | |
2283 | return skb; | |
1da177e4 LT |
2284 | |
2285 | if (*pt_prev) { | |
6229e362 | 2286 | *ret = deliver_skb(skb, *pt_prev, orig_dev); |
1da177e4 | 2287 | *pt_prev = NULL; |
4ec93edb YH |
2288 | } |
2289 | ||
6229e362 | 2290 | return br_handle_frame_hook(port, skb); |
1da177e4 LT |
2291 | } |
2292 | #else | |
6229e362 | 2293 | #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb) |
1da177e4 LT |
2294 | #endif |
2295 | ||
b863ceb7 PM |
2296 | #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE) |
2297 | struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly; | |
2298 | EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook); | |
2299 | ||
2300 | static inline struct sk_buff *handle_macvlan(struct sk_buff *skb, | |
2301 | struct packet_type **pt_prev, | |
2302 | int *ret, | |
2303 | struct net_device *orig_dev) | |
2304 | { | |
2305 | if (skb->dev->macvlan_port == NULL) | |
2306 | return skb; | |
2307 | ||
2308 | if (*pt_prev) { | |
2309 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
2310 | *pt_prev = NULL; | |
2311 | } | |
2312 | return macvlan_handle_frame_hook(skb); | |
2313 | } | |
2314 | #else | |
2315 | #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb) | |
2316 | #endif | |
2317 | ||
1da177e4 LT |
2318 | #ifdef CONFIG_NET_CLS_ACT |
2319 | /* TODO: Maybe we should just force sch_ingress to be compiled in | |
2320 | * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions | |
2321 | * a compare and 2 stores extra right now if we dont have it on | |
2322 | * but have CONFIG_NET_CLS_ACT | |
4ec93edb | 2323 | * NOTE: This doesnt stop any functionality; if you dont have |
1da177e4 LT |
2324 | * the ingress scheduler, you just cant add policies on ingress. |
2325 | * | |
2326 | */ | |
4ec93edb | 2327 | static int ing_filter(struct sk_buff *skb) |
1da177e4 | 2328 | { |
1da177e4 | 2329 | struct net_device *dev = skb->dev; |
f697c3e8 | 2330 | u32 ttl = G_TC_RTTL(skb->tc_verd); |
555353cf DM |
2331 | struct netdev_queue *rxq; |
2332 | int result = TC_ACT_OK; | |
2333 | struct Qdisc *q; | |
4ec93edb | 2334 | |
f697c3e8 HX |
2335 | if (MAX_RED_LOOP < ttl++) { |
2336 | printk(KERN_WARNING | |
2337 | "Redir loop detected Dropping packet (%d->%d)\n", | |
8964be4a | 2338 | skb->skb_iif, dev->ifindex); |
f697c3e8 HX |
2339 | return TC_ACT_SHOT; |
2340 | } | |
1da177e4 | 2341 | |
f697c3e8 HX |
2342 | skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl); |
2343 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); | |
1da177e4 | 2344 | |
555353cf DM |
2345 | rxq = &dev->rx_queue; |
2346 | ||
83874000 | 2347 | q = rxq->qdisc; |
8d50b53d | 2348 | if (q != &noop_qdisc) { |
83874000 | 2349 | spin_lock(qdisc_lock(q)); |
a9312ae8 DM |
2350 | if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) |
2351 | result = qdisc_enqueue_root(skb, q); | |
83874000 DM |
2352 | spin_unlock(qdisc_lock(q)); |
2353 | } | |
f697c3e8 HX |
2354 | |
2355 | return result; | |
2356 | } | |
86e65da9 | 2357 | |
f697c3e8 HX |
2358 | static inline struct sk_buff *handle_ing(struct sk_buff *skb, |
2359 | struct packet_type **pt_prev, | |
2360 | int *ret, struct net_device *orig_dev) | |
2361 | { | |
8d50b53d | 2362 | if (skb->dev->rx_queue.qdisc == &noop_qdisc) |
f697c3e8 | 2363 | goto out; |
1da177e4 | 2364 | |
f697c3e8 HX |
2365 | if (*pt_prev) { |
2366 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
2367 | *pt_prev = NULL; | |
2368 | } else { | |
2369 | /* Huh? Why does turning on AF_PACKET affect this? */ | |
2370 | skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd); | |
1da177e4 LT |
2371 | } |
2372 | ||
f697c3e8 HX |
2373 | switch (ing_filter(skb)) { |
2374 | case TC_ACT_SHOT: | |
2375 | case TC_ACT_STOLEN: | |
2376 | kfree_skb(skb); | |
2377 | return NULL; | |
2378 | } | |
2379 | ||
2380 | out: | |
2381 | skb->tc_verd = 0; | |
2382 | return skb; | |
1da177e4 LT |
2383 | } |
2384 | #endif | |
2385 | ||
bc1d0411 PM |
2386 | /* |
2387 | * netif_nit_deliver - deliver received packets to network taps | |
2388 | * @skb: buffer | |
2389 | * | |
2390 | * This function is used to deliver incoming packets to network | |
2391 | * taps. It should be used when the normal netif_receive_skb path | |
2392 | * is bypassed, for example because of VLAN acceleration. | |
2393 | */ | |
2394 | void netif_nit_deliver(struct sk_buff *skb) | |
2395 | { | |
2396 | struct packet_type *ptype; | |
2397 | ||
2398 | if (list_empty(&ptype_all)) | |
2399 | return; | |
2400 | ||
2401 | skb_reset_network_header(skb); | |
2402 | skb_reset_transport_header(skb); | |
2403 | skb->mac_len = skb->network_header - skb->mac_header; | |
2404 | ||
2405 | rcu_read_lock(); | |
2406 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
2407 | if (!ptype->dev || ptype->dev == skb->dev) | |
2408 | deliver_skb(skb, ptype, skb->dev); | |
2409 | } | |
2410 | rcu_read_unlock(); | |
2411 | } | |
2412 | ||
3b582cc1 SH |
2413 | /** |
2414 | * netif_receive_skb - process receive buffer from network | |
2415 | * @skb: buffer to process | |
2416 | * | |
2417 | * netif_receive_skb() is the main receive data processing function. | |
2418 | * It always succeeds. The buffer may be dropped during processing | |
2419 | * for congestion control or by the protocol layers. | |
2420 | * | |
2421 | * This function may only be called from softirq context and interrupts | |
2422 | * should be enabled. | |
2423 | * | |
2424 | * Return values (usually ignored): | |
2425 | * NET_RX_SUCCESS: no congestion | |
2426 | * NET_RX_DROP: packet was dropped | |
2427 | */ | |
1da177e4 LT |
2428 | int netif_receive_skb(struct sk_buff *skb) |
2429 | { | |
2430 | struct packet_type *ptype, *pt_prev; | |
f2ccd8fa | 2431 | struct net_device *orig_dev; |
0d7a3681 | 2432 | struct net_device *null_or_orig; |
ca8d9ea3 | 2433 | struct net_device *null_or_bond; |
1da177e4 | 2434 | int ret = NET_RX_DROP; |
252e3346 | 2435 | __be16 type; |
1da177e4 | 2436 | |
81bbb3d4 ED |
2437 | if (!skb->tstamp.tv64) |
2438 | net_timestamp(skb); | |
2439 | ||
05423b24 | 2440 | if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb)) |
9b22ea56 PM |
2441 | return NET_RX_SUCCESS; |
2442 | ||
1da177e4 | 2443 | /* if we've gotten here through NAPI, check netpoll */ |
bea3348e | 2444 | if (netpoll_receive_skb(skb)) |
1da177e4 LT |
2445 | return NET_RX_DROP; |
2446 | ||
8964be4a ED |
2447 | if (!skb->skb_iif) |
2448 | skb->skb_iif = skb->dev->ifindex; | |
86e65da9 | 2449 | |
0d7a3681 | 2450 | null_or_orig = NULL; |
cc9bd5ce JE |
2451 | orig_dev = skb->dev; |
2452 | if (orig_dev->master) { | |
0d7a3681 JE |
2453 | if (skb_bond_should_drop(skb)) |
2454 | null_or_orig = orig_dev; /* deliver only exact match */ | |
2455 | else | |
2456 | skb->dev = orig_dev->master; | |
cc9bd5ce | 2457 | } |
8f903c70 | 2458 | |
1da177e4 LT |
2459 | __get_cpu_var(netdev_rx_stat).total++; |
2460 | ||
c1d2bbe1 | 2461 | skb_reset_network_header(skb); |
badff6d0 | 2462 | skb_reset_transport_header(skb); |
b0e380b1 | 2463 | skb->mac_len = skb->network_header - skb->mac_header; |
1da177e4 LT |
2464 | |
2465 | pt_prev = NULL; | |
2466 | ||
2467 | rcu_read_lock(); | |
2468 | ||
2469 | #ifdef CONFIG_NET_CLS_ACT | |
2470 | if (skb->tc_verd & TC_NCLS) { | |
2471 | skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); | |
2472 | goto ncls; | |
2473 | } | |
2474 | #endif | |
2475 | ||
2476 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
f982307f JE |
2477 | if (ptype->dev == null_or_orig || ptype->dev == skb->dev || |
2478 | ptype->dev == orig_dev) { | |
4ec93edb | 2479 | if (pt_prev) |
f2ccd8fa | 2480 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
2481 | pt_prev = ptype; |
2482 | } | |
2483 | } | |
2484 | ||
2485 | #ifdef CONFIG_NET_CLS_ACT | |
f697c3e8 HX |
2486 | skb = handle_ing(skb, &pt_prev, &ret, orig_dev); |
2487 | if (!skb) | |
1da177e4 | 2488 | goto out; |
1da177e4 LT |
2489 | ncls: |
2490 | #endif | |
2491 | ||
6229e362 | 2492 | skb = handle_bridge(skb, &pt_prev, &ret, orig_dev); |
b863ceb7 PM |
2493 | if (!skb) |
2494 | goto out; | |
2495 | skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev); | |
6229e362 | 2496 | if (!skb) |
1da177e4 LT |
2497 | goto out; |
2498 | ||
1f3c8804 AG |
2499 | /* |
2500 | * Make sure frames received on VLAN interfaces stacked on | |
2501 | * bonding interfaces still make their way to any base bonding | |
2502 | * device that may have registered for a specific ptype. The | |
2503 | * handler may have to adjust skb->dev and orig_dev. | |
1f3c8804 | 2504 | */ |
ca8d9ea3 | 2505 | null_or_bond = NULL; |
1f3c8804 AG |
2506 | if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) && |
2507 | (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) { | |
ca8d9ea3 | 2508 | null_or_bond = vlan_dev_real_dev(skb->dev); |
1f3c8804 AG |
2509 | } |
2510 | ||
1da177e4 | 2511 | type = skb->protocol; |
82d8a867 PE |
2512 | list_for_each_entry_rcu(ptype, |
2513 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
1f3c8804 | 2514 | if (ptype->type == type && (ptype->dev == null_or_orig || |
ca8d9ea3 AG |
2515 | ptype->dev == skb->dev || ptype->dev == orig_dev || |
2516 | ptype->dev == null_or_bond)) { | |
4ec93edb | 2517 | if (pt_prev) |
f2ccd8fa | 2518 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
2519 | pt_prev = ptype; |
2520 | } | |
2521 | } | |
2522 | ||
2523 | if (pt_prev) { | |
f2ccd8fa | 2524 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
1da177e4 LT |
2525 | } else { |
2526 | kfree_skb(skb); | |
2527 | /* Jamal, now you will not able to escape explaining | |
2528 | * me how you were going to use this. :-) | |
2529 | */ | |
2530 | ret = NET_RX_DROP; | |
2531 | } | |
2532 | ||
2533 | out: | |
2534 | rcu_read_unlock(); | |
2535 | return ret; | |
2536 | } | |
d1b19dff | 2537 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 2538 | |
6e583ce5 SH |
2539 | /* Network device is going away, flush any packets still pending */ |
2540 | static void flush_backlog(void *arg) | |
2541 | { | |
2542 | struct net_device *dev = arg; | |
2543 | struct softnet_data *queue = &__get_cpu_var(softnet_data); | |
2544 | struct sk_buff *skb, *tmp; | |
2545 | ||
2546 | skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp) | |
2547 | if (skb->dev == dev) { | |
2548 | __skb_unlink(skb, &queue->input_pkt_queue); | |
2549 | kfree_skb(skb); | |
2550 | } | |
2551 | } | |
2552 | ||
d565b0a1 HX |
2553 | static int napi_gro_complete(struct sk_buff *skb) |
2554 | { | |
2555 | struct packet_type *ptype; | |
2556 | __be16 type = skb->protocol; | |
2557 | struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK]; | |
2558 | int err = -ENOENT; | |
2559 | ||
fc59f9a3 HX |
2560 | if (NAPI_GRO_CB(skb)->count == 1) { |
2561 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 2562 | goto out; |
fc59f9a3 | 2563 | } |
d565b0a1 HX |
2564 | |
2565 | rcu_read_lock(); | |
2566 | list_for_each_entry_rcu(ptype, head, list) { | |
2567 | if (ptype->type != type || ptype->dev || !ptype->gro_complete) | |
2568 | continue; | |
2569 | ||
2570 | err = ptype->gro_complete(skb); | |
2571 | break; | |
2572 | } | |
2573 | rcu_read_unlock(); | |
2574 | ||
2575 | if (err) { | |
2576 | WARN_ON(&ptype->list == head); | |
2577 | kfree_skb(skb); | |
2578 | return NET_RX_SUCCESS; | |
2579 | } | |
2580 | ||
2581 | out: | |
d565b0a1 HX |
2582 | return netif_receive_skb(skb); |
2583 | } | |
2584 | ||
2585 | void napi_gro_flush(struct napi_struct *napi) | |
2586 | { | |
2587 | struct sk_buff *skb, *next; | |
2588 | ||
2589 | for (skb = napi->gro_list; skb; skb = next) { | |
2590 | next = skb->next; | |
2591 | skb->next = NULL; | |
2592 | napi_gro_complete(skb); | |
2593 | } | |
2594 | ||
4ae5544f | 2595 | napi->gro_count = 0; |
d565b0a1 HX |
2596 | napi->gro_list = NULL; |
2597 | } | |
2598 | EXPORT_SYMBOL(napi_gro_flush); | |
2599 | ||
5b252f0c | 2600 | enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 HX |
2601 | { |
2602 | struct sk_buff **pp = NULL; | |
2603 | struct packet_type *ptype; | |
2604 | __be16 type = skb->protocol; | |
2605 | struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK]; | |
0da2afd5 | 2606 | int same_flow; |
d565b0a1 | 2607 | int mac_len; |
5b252f0c | 2608 | enum gro_result ret; |
d565b0a1 HX |
2609 | |
2610 | if (!(skb->dev->features & NETIF_F_GRO)) | |
2611 | goto normal; | |
2612 | ||
4cf704fb | 2613 | if (skb_is_gso(skb) || skb_has_frags(skb)) |
f17f5c91 HX |
2614 | goto normal; |
2615 | ||
d565b0a1 HX |
2616 | rcu_read_lock(); |
2617 | list_for_each_entry_rcu(ptype, head, list) { | |
d565b0a1 HX |
2618 | if (ptype->type != type || ptype->dev || !ptype->gro_receive) |
2619 | continue; | |
2620 | ||
86911732 | 2621 | skb_set_network_header(skb, skb_gro_offset(skb)); |
d565b0a1 HX |
2622 | mac_len = skb->network_header - skb->mac_header; |
2623 | skb->mac_len = mac_len; | |
2624 | NAPI_GRO_CB(skb)->same_flow = 0; | |
2625 | NAPI_GRO_CB(skb)->flush = 0; | |
5d38a079 | 2626 | NAPI_GRO_CB(skb)->free = 0; |
d565b0a1 | 2627 | |
d565b0a1 HX |
2628 | pp = ptype->gro_receive(&napi->gro_list, skb); |
2629 | break; | |
2630 | } | |
2631 | rcu_read_unlock(); | |
2632 | ||
2633 | if (&ptype->list == head) | |
2634 | goto normal; | |
2635 | ||
0da2afd5 | 2636 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 2637 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 2638 | |
d565b0a1 HX |
2639 | if (pp) { |
2640 | struct sk_buff *nskb = *pp; | |
2641 | ||
2642 | *pp = nskb->next; | |
2643 | nskb->next = NULL; | |
2644 | napi_gro_complete(nskb); | |
4ae5544f | 2645 | napi->gro_count--; |
d565b0a1 HX |
2646 | } |
2647 | ||
0da2afd5 | 2648 | if (same_flow) |
d565b0a1 HX |
2649 | goto ok; |
2650 | ||
4ae5544f | 2651 | if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS) |
d565b0a1 | 2652 | goto normal; |
d565b0a1 | 2653 | |
4ae5544f | 2654 | napi->gro_count++; |
d565b0a1 | 2655 | NAPI_GRO_CB(skb)->count = 1; |
86911732 | 2656 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
d565b0a1 HX |
2657 | skb->next = napi->gro_list; |
2658 | napi->gro_list = skb; | |
5d0d9be8 | 2659 | ret = GRO_HELD; |
d565b0a1 | 2660 | |
ad0f9904 | 2661 | pull: |
cb18978c HX |
2662 | if (skb_headlen(skb) < skb_gro_offset(skb)) { |
2663 | int grow = skb_gro_offset(skb) - skb_headlen(skb); | |
2664 | ||
2665 | BUG_ON(skb->end - skb->tail < grow); | |
2666 | ||
2667 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
2668 | ||
2669 | skb->tail += grow; | |
2670 | skb->data_len -= grow; | |
2671 | ||
2672 | skb_shinfo(skb)->frags[0].page_offset += grow; | |
2673 | skb_shinfo(skb)->frags[0].size -= grow; | |
2674 | ||
2675 | if (unlikely(!skb_shinfo(skb)->frags[0].size)) { | |
2676 | put_page(skb_shinfo(skb)->frags[0].page); | |
2677 | memmove(skb_shinfo(skb)->frags, | |
2678 | skb_shinfo(skb)->frags + 1, | |
2679 | --skb_shinfo(skb)->nr_frags); | |
2680 | } | |
ad0f9904 HX |
2681 | } |
2682 | ||
d565b0a1 | 2683 | ok: |
5d0d9be8 | 2684 | return ret; |
d565b0a1 HX |
2685 | |
2686 | normal: | |
ad0f9904 HX |
2687 | ret = GRO_NORMAL; |
2688 | goto pull; | |
5d38a079 | 2689 | } |
96e93eab HX |
2690 | EXPORT_SYMBOL(dev_gro_receive); |
2691 | ||
5b252f0c BH |
2692 | static gro_result_t |
2693 | __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) | |
96e93eab HX |
2694 | { |
2695 | struct sk_buff *p; | |
2696 | ||
d1c76af9 HX |
2697 | if (netpoll_rx_on(skb)) |
2698 | return GRO_NORMAL; | |
2699 | ||
96e93eab | 2700 | for (p = napi->gro_list; p; p = p->next) { |
f64f9e71 JP |
2701 | NAPI_GRO_CB(p)->same_flow = |
2702 | (p->dev == skb->dev) && | |
2703 | !compare_ether_header(skb_mac_header(p), | |
2704 | skb_gro_mac_header(skb)); | |
96e93eab HX |
2705 | NAPI_GRO_CB(p)->flush = 0; |
2706 | } | |
2707 | ||
2708 | return dev_gro_receive(napi, skb); | |
2709 | } | |
5d38a079 | 2710 | |
c7c4b3b6 | 2711 | gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) |
5d38a079 | 2712 | { |
5d0d9be8 HX |
2713 | switch (ret) { |
2714 | case GRO_NORMAL: | |
c7c4b3b6 BH |
2715 | if (netif_receive_skb(skb)) |
2716 | ret = GRO_DROP; | |
2717 | break; | |
5d38a079 | 2718 | |
5d0d9be8 | 2719 | case GRO_DROP: |
5d0d9be8 | 2720 | case GRO_MERGED_FREE: |
5d38a079 HX |
2721 | kfree_skb(skb); |
2722 | break; | |
5b252f0c BH |
2723 | |
2724 | case GRO_HELD: | |
2725 | case GRO_MERGED: | |
2726 | break; | |
5d38a079 HX |
2727 | } |
2728 | ||
c7c4b3b6 | 2729 | return ret; |
5d0d9be8 HX |
2730 | } |
2731 | EXPORT_SYMBOL(napi_skb_finish); | |
2732 | ||
78a478d0 HX |
2733 | void skb_gro_reset_offset(struct sk_buff *skb) |
2734 | { | |
2735 | NAPI_GRO_CB(skb)->data_offset = 0; | |
2736 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
7489594c | 2737 | NAPI_GRO_CB(skb)->frag0_len = 0; |
78a478d0 | 2738 | |
78d3fd0b | 2739 | if (skb->mac_header == skb->tail && |
7489594c | 2740 | !PageHighMem(skb_shinfo(skb)->frags[0].page)) { |
78a478d0 HX |
2741 | NAPI_GRO_CB(skb)->frag0 = |
2742 | page_address(skb_shinfo(skb)->frags[0].page) + | |
2743 | skb_shinfo(skb)->frags[0].page_offset; | |
7489594c HX |
2744 | NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size; |
2745 | } | |
78a478d0 HX |
2746 | } |
2747 | EXPORT_SYMBOL(skb_gro_reset_offset); | |
2748 | ||
c7c4b3b6 | 2749 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 2750 | { |
86911732 HX |
2751 | skb_gro_reset_offset(skb); |
2752 | ||
5d0d9be8 | 2753 | return napi_skb_finish(__napi_gro_receive(napi, skb), skb); |
d565b0a1 HX |
2754 | } |
2755 | EXPORT_SYMBOL(napi_gro_receive); | |
2756 | ||
96e93eab HX |
2757 | void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
2758 | { | |
96e93eab HX |
2759 | __skb_pull(skb, skb_headlen(skb)); |
2760 | skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb)); | |
2761 | ||
2762 | napi->skb = skb; | |
2763 | } | |
2764 | EXPORT_SYMBOL(napi_reuse_skb); | |
2765 | ||
76620aaf | 2766 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 2767 | { |
5d38a079 | 2768 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
2769 | |
2770 | if (!skb) { | |
89d71a66 ED |
2771 | skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD); |
2772 | if (skb) | |
2773 | napi->skb = skb; | |
80595d59 | 2774 | } |
96e93eab HX |
2775 | return skb; |
2776 | } | |
76620aaf | 2777 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 2778 | |
c7c4b3b6 BH |
2779 | gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb, |
2780 | gro_result_t ret) | |
96e93eab | 2781 | { |
5d0d9be8 HX |
2782 | switch (ret) { |
2783 | case GRO_NORMAL: | |
86911732 | 2784 | case GRO_HELD: |
86911732 HX |
2785 | skb->protocol = eth_type_trans(skb, napi->dev); |
2786 | ||
c7c4b3b6 BH |
2787 | if (ret == GRO_HELD) |
2788 | skb_gro_pull(skb, -ETH_HLEN); | |
2789 | else if (netif_receive_skb(skb)) | |
2790 | ret = GRO_DROP; | |
86911732 | 2791 | break; |
5d38a079 | 2792 | |
5d0d9be8 | 2793 | case GRO_DROP: |
5d0d9be8 HX |
2794 | case GRO_MERGED_FREE: |
2795 | napi_reuse_skb(napi, skb); | |
2796 | break; | |
5b252f0c BH |
2797 | |
2798 | case GRO_MERGED: | |
2799 | break; | |
5d0d9be8 | 2800 | } |
5d38a079 | 2801 | |
c7c4b3b6 | 2802 | return ret; |
5d38a079 | 2803 | } |
5d0d9be8 HX |
2804 | EXPORT_SYMBOL(napi_frags_finish); |
2805 | ||
76620aaf HX |
2806 | struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
2807 | { | |
2808 | struct sk_buff *skb = napi->skb; | |
2809 | struct ethhdr *eth; | |
a5b1cf28 HX |
2810 | unsigned int hlen; |
2811 | unsigned int off; | |
76620aaf HX |
2812 | |
2813 | napi->skb = NULL; | |
2814 | ||
2815 | skb_reset_mac_header(skb); | |
2816 | skb_gro_reset_offset(skb); | |
2817 | ||
a5b1cf28 HX |
2818 | off = skb_gro_offset(skb); |
2819 | hlen = off + sizeof(*eth); | |
2820 | eth = skb_gro_header_fast(skb, off); | |
2821 | if (skb_gro_header_hard(skb, hlen)) { | |
2822 | eth = skb_gro_header_slow(skb, hlen, off); | |
2823 | if (unlikely(!eth)) { | |
2824 | napi_reuse_skb(napi, skb); | |
2825 | skb = NULL; | |
2826 | goto out; | |
2827 | } | |
76620aaf HX |
2828 | } |
2829 | ||
2830 | skb_gro_pull(skb, sizeof(*eth)); | |
2831 | ||
2832 | /* | |
2833 | * This works because the only protocols we care about don't require | |
2834 | * special handling. We'll fix it up properly at the end. | |
2835 | */ | |
2836 | skb->protocol = eth->h_proto; | |
2837 | ||
2838 | out: | |
2839 | return skb; | |
2840 | } | |
2841 | EXPORT_SYMBOL(napi_frags_skb); | |
2842 | ||
c7c4b3b6 | 2843 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 2844 | { |
76620aaf | 2845 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 HX |
2846 | |
2847 | if (!skb) | |
c7c4b3b6 | 2848 | return GRO_DROP; |
5d0d9be8 HX |
2849 | |
2850 | return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb)); | |
2851 | } | |
5d38a079 HX |
2852 | EXPORT_SYMBOL(napi_gro_frags); |
2853 | ||
bea3348e | 2854 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 LT |
2855 | { |
2856 | int work = 0; | |
1da177e4 LT |
2857 | struct softnet_data *queue = &__get_cpu_var(softnet_data); |
2858 | unsigned long start_time = jiffies; | |
2859 | ||
bea3348e SH |
2860 | napi->weight = weight_p; |
2861 | do { | |
1da177e4 | 2862 | struct sk_buff *skb; |
1da177e4 LT |
2863 | |
2864 | local_irq_disable(); | |
2865 | skb = __skb_dequeue(&queue->input_pkt_queue); | |
bea3348e | 2866 | if (!skb) { |
8f1ead2d | 2867 | __napi_complete(napi); |
bea3348e | 2868 | local_irq_enable(); |
8f1ead2d | 2869 | break; |
bea3348e | 2870 | } |
1da177e4 LT |
2871 | local_irq_enable(); |
2872 | ||
8f1ead2d | 2873 | netif_receive_skb(skb); |
bea3348e | 2874 | } while (++work < quota && jiffies == start_time); |
1da177e4 | 2875 | |
bea3348e SH |
2876 | return work; |
2877 | } | |
1da177e4 | 2878 | |
bea3348e SH |
2879 | /** |
2880 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 2881 | * @n: entry to schedule |
bea3348e SH |
2882 | * |
2883 | * The entry's receive function will be scheduled to run | |
2884 | */ | |
b5606c2d | 2885 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
2886 | { |
2887 | unsigned long flags; | |
1da177e4 | 2888 | |
bea3348e SH |
2889 | local_irq_save(flags); |
2890 | list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list); | |
2891 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2892 | local_irq_restore(flags); | |
1da177e4 | 2893 | } |
bea3348e SH |
2894 | EXPORT_SYMBOL(__napi_schedule); |
2895 | ||
d565b0a1 HX |
2896 | void __napi_complete(struct napi_struct *n) |
2897 | { | |
2898 | BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); | |
2899 | BUG_ON(n->gro_list); | |
2900 | ||
2901 | list_del(&n->poll_list); | |
2902 | smp_mb__before_clear_bit(); | |
2903 | clear_bit(NAPI_STATE_SCHED, &n->state); | |
2904 | } | |
2905 | EXPORT_SYMBOL(__napi_complete); | |
2906 | ||
2907 | void napi_complete(struct napi_struct *n) | |
2908 | { | |
2909 | unsigned long flags; | |
2910 | ||
2911 | /* | |
2912 | * don't let napi dequeue from the cpu poll list | |
2913 | * just in case its running on a different cpu | |
2914 | */ | |
2915 | if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state))) | |
2916 | return; | |
2917 | ||
2918 | napi_gro_flush(n); | |
2919 | local_irq_save(flags); | |
2920 | __napi_complete(n); | |
2921 | local_irq_restore(flags); | |
2922 | } | |
2923 | EXPORT_SYMBOL(napi_complete); | |
2924 | ||
2925 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, | |
2926 | int (*poll)(struct napi_struct *, int), int weight) | |
2927 | { | |
2928 | INIT_LIST_HEAD(&napi->poll_list); | |
4ae5544f | 2929 | napi->gro_count = 0; |
d565b0a1 | 2930 | napi->gro_list = NULL; |
5d38a079 | 2931 | napi->skb = NULL; |
d565b0a1 HX |
2932 | napi->poll = poll; |
2933 | napi->weight = weight; | |
2934 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 2935 | napi->dev = dev; |
5d38a079 | 2936 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
2937 | spin_lock_init(&napi->poll_lock); |
2938 | napi->poll_owner = -1; | |
2939 | #endif | |
2940 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
2941 | } | |
2942 | EXPORT_SYMBOL(netif_napi_add); | |
2943 | ||
2944 | void netif_napi_del(struct napi_struct *napi) | |
2945 | { | |
2946 | struct sk_buff *skb, *next; | |
2947 | ||
d7b06636 | 2948 | list_del_init(&napi->dev_list); |
76620aaf | 2949 | napi_free_frags(napi); |
d565b0a1 HX |
2950 | |
2951 | for (skb = napi->gro_list; skb; skb = next) { | |
2952 | next = skb->next; | |
2953 | skb->next = NULL; | |
2954 | kfree_skb(skb); | |
2955 | } | |
2956 | ||
2957 | napi->gro_list = NULL; | |
4ae5544f | 2958 | napi->gro_count = 0; |
d565b0a1 HX |
2959 | } |
2960 | EXPORT_SYMBOL(netif_napi_del); | |
2961 | ||
1da177e4 LT |
2962 | |
2963 | static void net_rx_action(struct softirq_action *h) | |
2964 | { | |
bea3348e | 2965 | struct list_head *list = &__get_cpu_var(softnet_data).poll_list; |
24f8b238 | 2966 | unsigned long time_limit = jiffies + 2; |
51b0bded | 2967 | int budget = netdev_budget; |
53fb95d3 MM |
2968 | void *have; |
2969 | ||
1da177e4 LT |
2970 | local_irq_disable(); |
2971 | ||
bea3348e SH |
2972 | while (!list_empty(list)) { |
2973 | struct napi_struct *n; | |
2974 | int work, weight; | |
1da177e4 | 2975 | |
bea3348e | 2976 | /* If softirq window is exhuasted then punt. |
24f8b238 SH |
2977 | * Allow this to run for 2 jiffies since which will allow |
2978 | * an average latency of 1.5/HZ. | |
bea3348e | 2979 | */ |
24f8b238 | 2980 | if (unlikely(budget <= 0 || time_after(jiffies, time_limit))) |
1da177e4 LT |
2981 | goto softnet_break; |
2982 | ||
2983 | local_irq_enable(); | |
2984 | ||
bea3348e SH |
2985 | /* Even though interrupts have been re-enabled, this |
2986 | * access is safe because interrupts can only add new | |
2987 | * entries to the tail of this list, and only ->poll() | |
2988 | * calls can remove this head entry from the list. | |
2989 | */ | |
2990 | n = list_entry(list->next, struct napi_struct, poll_list); | |
1da177e4 | 2991 | |
bea3348e SH |
2992 | have = netpoll_poll_lock(n); |
2993 | ||
2994 | weight = n->weight; | |
2995 | ||
0a7606c1 DM |
2996 | /* This NAPI_STATE_SCHED test is for avoiding a race |
2997 | * with netpoll's poll_napi(). Only the entity which | |
2998 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
2999 | * actually make the ->poll() call. Therefore we avoid | |
3000 | * accidently calling ->poll() when NAPI is not scheduled. | |
3001 | */ | |
3002 | work = 0; | |
4ea7e386 | 3003 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { |
0a7606c1 | 3004 | work = n->poll(n, weight); |
4ea7e386 NH |
3005 | trace_napi_poll(n); |
3006 | } | |
bea3348e SH |
3007 | |
3008 | WARN_ON_ONCE(work > weight); | |
3009 | ||
3010 | budget -= work; | |
3011 | ||
3012 | local_irq_disable(); | |
3013 | ||
3014 | /* Drivers must not modify the NAPI state if they | |
3015 | * consume the entire weight. In such cases this code | |
3016 | * still "owns" the NAPI instance and therefore can | |
3017 | * move the instance around on the list at-will. | |
3018 | */ | |
fed17f30 | 3019 | if (unlikely(work == weight)) { |
ff780cd8 HX |
3020 | if (unlikely(napi_disable_pending(n))) { |
3021 | local_irq_enable(); | |
3022 | napi_complete(n); | |
3023 | local_irq_disable(); | |
3024 | } else | |
fed17f30 DM |
3025 | list_move_tail(&n->poll_list, list); |
3026 | } | |
bea3348e SH |
3027 | |
3028 | netpoll_poll_unlock(have); | |
1da177e4 LT |
3029 | } |
3030 | out: | |
515e06c4 | 3031 | local_irq_enable(); |
bea3348e | 3032 | |
db217334 CL |
3033 | #ifdef CONFIG_NET_DMA |
3034 | /* | |
3035 | * There may not be any more sk_buffs coming right now, so push | |
3036 | * any pending DMA copies to hardware | |
3037 | */ | |
2ba05622 | 3038 | dma_issue_pending_all(); |
db217334 | 3039 | #endif |
bea3348e | 3040 | |
1da177e4 LT |
3041 | return; |
3042 | ||
3043 | softnet_break: | |
3044 | __get_cpu_var(netdev_rx_stat).time_squeeze++; | |
3045 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3046 | goto out; | |
3047 | } | |
3048 | ||
d1b19dff | 3049 | static gifconf_func_t *gifconf_list[NPROTO]; |
1da177e4 LT |
3050 | |
3051 | /** | |
3052 | * register_gifconf - register a SIOCGIF handler | |
3053 | * @family: Address family | |
3054 | * @gifconf: Function handler | |
3055 | * | |
3056 | * Register protocol dependent address dumping routines. The handler | |
3057 | * that is passed must not be freed or reused until it has been replaced | |
3058 | * by another handler. | |
3059 | */ | |
d1b19dff | 3060 | int register_gifconf(unsigned int family, gifconf_func_t *gifconf) |
1da177e4 LT |
3061 | { |
3062 | if (family >= NPROTO) | |
3063 | return -EINVAL; | |
3064 | gifconf_list[family] = gifconf; | |
3065 | return 0; | |
3066 | } | |
d1b19dff | 3067 | EXPORT_SYMBOL(register_gifconf); |
1da177e4 LT |
3068 | |
3069 | ||
3070 | /* | |
3071 | * Map an interface index to its name (SIOCGIFNAME) | |
3072 | */ | |
3073 | ||
3074 | /* | |
3075 | * We need this ioctl for efficient implementation of the | |
3076 | * if_indextoname() function required by the IPv6 API. Without | |
3077 | * it, we would have to search all the interfaces to find a | |
3078 | * match. --pb | |
3079 | */ | |
3080 | ||
881d966b | 3081 | static int dev_ifname(struct net *net, struct ifreq __user *arg) |
1da177e4 LT |
3082 | { |
3083 | struct net_device *dev; | |
3084 | struct ifreq ifr; | |
3085 | ||
3086 | /* | |
3087 | * Fetch the caller's info block. | |
3088 | */ | |
3089 | ||
3090 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
3091 | return -EFAULT; | |
3092 | ||
fb699dfd ED |
3093 | rcu_read_lock(); |
3094 | dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex); | |
1da177e4 | 3095 | if (!dev) { |
fb699dfd | 3096 | rcu_read_unlock(); |
1da177e4 LT |
3097 | return -ENODEV; |
3098 | } | |
3099 | ||
3100 | strcpy(ifr.ifr_name, dev->name); | |
fb699dfd | 3101 | rcu_read_unlock(); |
1da177e4 LT |
3102 | |
3103 | if (copy_to_user(arg, &ifr, sizeof(struct ifreq))) | |
3104 | return -EFAULT; | |
3105 | return 0; | |
3106 | } | |
3107 | ||
3108 | /* | |
3109 | * Perform a SIOCGIFCONF call. This structure will change | |
3110 | * size eventually, and there is nothing I can do about it. | |
3111 | * Thus we will need a 'compatibility mode'. | |
3112 | */ | |
3113 | ||
881d966b | 3114 | static int dev_ifconf(struct net *net, char __user *arg) |
1da177e4 LT |
3115 | { |
3116 | struct ifconf ifc; | |
3117 | struct net_device *dev; | |
3118 | char __user *pos; | |
3119 | int len; | |
3120 | int total; | |
3121 | int i; | |
3122 | ||
3123 | /* | |
3124 | * Fetch the caller's info block. | |
3125 | */ | |
3126 | ||
3127 | if (copy_from_user(&ifc, arg, sizeof(struct ifconf))) | |
3128 | return -EFAULT; | |
3129 | ||
3130 | pos = ifc.ifc_buf; | |
3131 | len = ifc.ifc_len; | |
3132 | ||
3133 | /* | |
3134 | * Loop over the interfaces, and write an info block for each. | |
3135 | */ | |
3136 | ||
3137 | total = 0; | |
881d966b | 3138 | for_each_netdev(net, dev) { |
1da177e4 LT |
3139 | for (i = 0; i < NPROTO; i++) { |
3140 | if (gifconf_list[i]) { | |
3141 | int done; | |
3142 | if (!pos) | |
3143 | done = gifconf_list[i](dev, NULL, 0); | |
3144 | else | |
3145 | done = gifconf_list[i](dev, pos + total, | |
3146 | len - total); | |
3147 | if (done < 0) | |
3148 | return -EFAULT; | |
3149 | total += done; | |
3150 | } | |
3151 | } | |
4ec93edb | 3152 | } |
1da177e4 LT |
3153 | |
3154 | /* | |
3155 | * All done. Write the updated control block back to the caller. | |
3156 | */ | |
3157 | ifc.ifc_len = total; | |
3158 | ||
3159 | /* | |
3160 | * Both BSD and Solaris return 0 here, so we do too. | |
3161 | */ | |
3162 | return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0; | |
3163 | } | |
3164 | ||
3165 | #ifdef CONFIG_PROC_FS | |
3166 | /* | |
3167 | * This is invoked by the /proc filesystem handler to display a device | |
3168 | * in detail. | |
3169 | */ | |
7562f876 | 3170 | void *dev_seq_start(struct seq_file *seq, loff_t *pos) |
c6d14c84 | 3171 | __acquires(RCU) |
1da177e4 | 3172 | { |
e372c414 | 3173 | struct net *net = seq_file_net(seq); |
7562f876 | 3174 | loff_t off; |
1da177e4 | 3175 | struct net_device *dev; |
1da177e4 | 3176 | |
c6d14c84 | 3177 | rcu_read_lock(); |
7562f876 PE |
3178 | if (!*pos) |
3179 | return SEQ_START_TOKEN; | |
1da177e4 | 3180 | |
7562f876 | 3181 | off = 1; |
c6d14c84 | 3182 | for_each_netdev_rcu(net, dev) |
7562f876 PE |
3183 | if (off++ == *pos) |
3184 | return dev; | |
1da177e4 | 3185 | |
7562f876 | 3186 | return NULL; |
1da177e4 LT |
3187 | } |
3188 | ||
3189 | void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
3190 | { | |
c6d14c84 ED |
3191 | struct net_device *dev = (v == SEQ_START_TOKEN) ? |
3192 | first_net_device(seq_file_net(seq)) : | |
3193 | next_net_device((struct net_device *)v); | |
3194 | ||
1da177e4 | 3195 | ++*pos; |
c6d14c84 | 3196 | return rcu_dereference(dev); |
1da177e4 LT |
3197 | } |
3198 | ||
3199 | void dev_seq_stop(struct seq_file *seq, void *v) | |
c6d14c84 | 3200 | __releases(RCU) |
1da177e4 | 3201 | { |
c6d14c84 | 3202 | rcu_read_unlock(); |
1da177e4 LT |
3203 | } |
3204 | ||
3205 | static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev) | |
3206 | { | |
eeda3fd6 | 3207 | const struct net_device_stats *stats = dev_get_stats(dev); |
1da177e4 | 3208 | |
5a1b5898 RR |
3209 | seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu " |
3210 | "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n", | |
3211 | dev->name, stats->rx_bytes, stats->rx_packets, | |
3212 | stats->rx_errors, | |
3213 | stats->rx_dropped + stats->rx_missed_errors, | |
3214 | stats->rx_fifo_errors, | |
3215 | stats->rx_length_errors + stats->rx_over_errors + | |
3216 | stats->rx_crc_errors + stats->rx_frame_errors, | |
3217 | stats->rx_compressed, stats->multicast, | |
3218 | stats->tx_bytes, stats->tx_packets, | |
3219 | stats->tx_errors, stats->tx_dropped, | |
3220 | stats->tx_fifo_errors, stats->collisions, | |
3221 | stats->tx_carrier_errors + | |
3222 | stats->tx_aborted_errors + | |
3223 | stats->tx_window_errors + | |
3224 | stats->tx_heartbeat_errors, | |
3225 | stats->tx_compressed); | |
1da177e4 LT |
3226 | } |
3227 | ||
3228 | /* | |
3229 | * Called from the PROCfs module. This now uses the new arbitrary sized | |
3230 | * /proc/net interface to create /proc/net/dev | |
3231 | */ | |
3232 | static int dev_seq_show(struct seq_file *seq, void *v) | |
3233 | { | |
3234 | if (v == SEQ_START_TOKEN) | |
3235 | seq_puts(seq, "Inter-| Receive " | |
3236 | " | Transmit\n" | |
3237 | " face |bytes packets errs drop fifo frame " | |
3238 | "compressed multicast|bytes packets errs " | |
3239 | "drop fifo colls carrier compressed\n"); | |
3240 | else | |
3241 | dev_seq_printf_stats(seq, v); | |
3242 | return 0; | |
3243 | } | |
3244 | ||
3245 | static struct netif_rx_stats *softnet_get_online(loff_t *pos) | |
3246 | { | |
3247 | struct netif_rx_stats *rc = NULL; | |
3248 | ||
0c0b0aca | 3249 | while (*pos < nr_cpu_ids) |
4ec93edb | 3250 | if (cpu_online(*pos)) { |
1da177e4 LT |
3251 | rc = &per_cpu(netdev_rx_stat, *pos); |
3252 | break; | |
3253 | } else | |
3254 | ++*pos; | |
3255 | return rc; | |
3256 | } | |
3257 | ||
3258 | static void *softnet_seq_start(struct seq_file *seq, loff_t *pos) | |
3259 | { | |
3260 | return softnet_get_online(pos); | |
3261 | } | |
3262 | ||
3263 | static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
3264 | { | |
3265 | ++*pos; | |
3266 | return softnet_get_online(pos); | |
3267 | } | |
3268 | ||
3269 | static void softnet_seq_stop(struct seq_file *seq, void *v) | |
3270 | { | |
3271 | } | |
3272 | ||
3273 | static int softnet_seq_show(struct seq_file *seq, void *v) | |
3274 | { | |
3275 | struct netif_rx_stats *s = v; | |
3276 | ||
3277 | seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n", | |
31aa02c5 | 3278 | s->total, s->dropped, s->time_squeeze, 0, |
c1ebcdb8 | 3279 | 0, 0, 0, 0, /* was fastroute */ |
d1b19dff | 3280 | s->cpu_collision); |
1da177e4 LT |
3281 | return 0; |
3282 | } | |
3283 | ||
f690808e | 3284 | static const struct seq_operations dev_seq_ops = { |
1da177e4 LT |
3285 | .start = dev_seq_start, |
3286 | .next = dev_seq_next, | |
3287 | .stop = dev_seq_stop, | |
3288 | .show = dev_seq_show, | |
3289 | }; | |
3290 | ||
3291 | static int dev_seq_open(struct inode *inode, struct file *file) | |
3292 | { | |
e372c414 DL |
3293 | return seq_open_net(inode, file, &dev_seq_ops, |
3294 | sizeof(struct seq_net_private)); | |
1da177e4 LT |
3295 | } |
3296 | ||
9a32144e | 3297 | static const struct file_operations dev_seq_fops = { |
1da177e4 LT |
3298 | .owner = THIS_MODULE, |
3299 | .open = dev_seq_open, | |
3300 | .read = seq_read, | |
3301 | .llseek = seq_lseek, | |
e372c414 | 3302 | .release = seq_release_net, |
1da177e4 LT |
3303 | }; |
3304 | ||
f690808e | 3305 | static const struct seq_operations softnet_seq_ops = { |
1da177e4 LT |
3306 | .start = softnet_seq_start, |
3307 | .next = softnet_seq_next, | |
3308 | .stop = softnet_seq_stop, | |
3309 | .show = softnet_seq_show, | |
3310 | }; | |
3311 | ||
3312 | static int softnet_seq_open(struct inode *inode, struct file *file) | |
3313 | { | |
3314 | return seq_open(file, &softnet_seq_ops); | |
3315 | } | |
3316 | ||
9a32144e | 3317 | static const struct file_operations softnet_seq_fops = { |
1da177e4 LT |
3318 | .owner = THIS_MODULE, |
3319 | .open = softnet_seq_open, | |
3320 | .read = seq_read, | |
3321 | .llseek = seq_lseek, | |
3322 | .release = seq_release, | |
3323 | }; | |
3324 | ||
0e1256ff SH |
3325 | static void *ptype_get_idx(loff_t pos) |
3326 | { | |
3327 | struct packet_type *pt = NULL; | |
3328 | loff_t i = 0; | |
3329 | int t; | |
3330 | ||
3331 | list_for_each_entry_rcu(pt, &ptype_all, list) { | |
3332 | if (i == pos) | |
3333 | return pt; | |
3334 | ++i; | |
3335 | } | |
3336 | ||
82d8a867 | 3337 | for (t = 0; t < PTYPE_HASH_SIZE; t++) { |
0e1256ff SH |
3338 | list_for_each_entry_rcu(pt, &ptype_base[t], list) { |
3339 | if (i == pos) | |
3340 | return pt; | |
3341 | ++i; | |
3342 | } | |
3343 | } | |
3344 | return NULL; | |
3345 | } | |
3346 | ||
3347 | static void *ptype_seq_start(struct seq_file *seq, loff_t *pos) | |
72348a42 | 3348 | __acquires(RCU) |
0e1256ff SH |
3349 | { |
3350 | rcu_read_lock(); | |
3351 | return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN; | |
3352 | } | |
3353 | ||
3354 | static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
3355 | { | |
3356 | struct packet_type *pt; | |
3357 | struct list_head *nxt; | |
3358 | int hash; | |
3359 | ||
3360 | ++*pos; | |
3361 | if (v == SEQ_START_TOKEN) | |
3362 | return ptype_get_idx(0); | |
3363 | ||
3364 | pt = v; | |
3365 | nxt = pt->list.next; | |
3366 | if (pt->type == htons(ETH_P_ALL)) { | |
3367 | if (nxt != &ptype_all) | |
3368 | goto found; | |
3369 | hash = 0; | |
3370 | nxt = ptype_base[0].next; | |
3371 | } else | |
82d8a867 | 3372 | hash = ntohs(pt->type) & PTYPE_HASH_MASK; |
0e1256ff SH |
3373 | |
3374 | while (nxt == &ptype_base[hash]) { | |
82d8a867 | 3375 | if (++hash >= PTYPE_HASH_SIZE) |
0e1256ff SH |
3376 | return NULL; |
3377 | nxt = ptype_base[hash].next; | |
3378 | } | |
3379 | found: | |
3380 | return list_entry(nxt, struct packet_type, list); | |
3381 | } | |
3382 | ||
3383 | static void ptype_seq_stop(struct seq_file *seq, void *v) | |
72348a42 | 3384 | __releases(RCU) |
0e1256ff SH |
3385 | { |
3386 | rcu_read_unlock(); | |
3387 | } | |
3388 | ||
0e1256ff SH |
3389 | static int ptype_seq_show(struct seq_file *seq, void *v) |
3390 | { | |
3391 | struct packet_type *pt = v; | |
3392 | ||
3393 | if (v == SEQ_START_TOKEN) | |
3394 | seq_puts(seq, "Type Device Function\n"); | |
c346dca1 | 3395 | else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) { |
0e1256ff SH |
3396 | if (pt->type == htons(ETH_P_ALL)) |
3397 | seq_puts(seq, "ALL "); | |
3398 | else | |
3399 | seq_printf(seq, "%04x", ntohs(pt->type)); | |
3400 | ||
908cd2da AD |
3401 | seq_printf(seq, " %-8s %pF\n", |
3402 | pt->dev ? pt->dev->name : "", pt->func); | |
0e1256ff SH |
3403 | } |
3404 | ||
3405 | return 0; | |
3406 | } | |
3407 | ||
3408 | static const struct seq_operations ptype_seq_ops = { | |
3409 | .start = ptype_seq_start, | |
3410 | .next = ptype_seq_next, | |
3411 | .stop = ptype_seq_stop, | |
3412 | .show = ptype_seq_show, | |
3413 | }; | |
3414 | ||
3415 | static int ptype_seq_open(struct inode *inode, struct file *file) | |
3416 | { | |
2feb27db PE |
3417 | return seq_open_net(inode, file, &ptype_seq_ops, |
3418 | sizeof(struct seq_net_private)); | |
0e1256ff SH |
3419 | } |
3420 | ||
3421 | static const struct file_operations ptype_seq_fops = { | |
3422 | .owner = THIS_MODULE, | |
3423 | .open = ptype_seq_open, | |
3424 | .read = seq_read, | |
3425 | .llseek = seq_lseek, | |
2feb27db | 3426 | .release = seq_release_net, |
0e1256ff SH |
3427 | }; |
3428 | ||
3429 | ||
4665079c | 3430 | static int __net_init dev_proc_net_init(struct net *net) |
1da177e4 LT |
3431 | { |
3432 | int rc = -ENOMEM; | |
3433 | ||
881d966b | 3434 | if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops)) |
1da177e4 | 3435 | goto out; |
881d966b | 3436 | if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops)) |
1da177e4 | 3437 | goto out_dev; |
881d966b | 3438 | if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops)) |
457c4cbc | 3439 | goto out_softnet; |
0e1256ff | 3440 | |
881d966b | 3441 | if (wext_proc_init(net)) |
457c4cbc | 3442 | goto out_ptype; |
1da177e4 LT |
3443 | rc = 0; |
3444 | out: | |
3445 | return rc; | |
457c4cbc | 3446 | out_ptype: |
881d966b | 3447 | proc_net_remove(net, "ptype"); |
1da177e4 | 3448 | out_softnet: |
881d966b | 3449 | proc_net_remove(net, "softnet_stat"); |
1da177e4 | 3450 | out_dev: |
881d966b | 3451 | proc_net_remove(net, "dev"); |
1da177e4 LT |
3452 | goto out; |
3453 | } | |
881d966b | 3454 | |
4665079c | 3455 | static void __net_exit dev_proc_net_exit(struct net *net) |
881d966b EB |
3456 | { |
3457 | wext_proc_exit(net); | |
3458 | ||
3459 | proc_net_remove(net, "ptype"); | |
3460 | proc_net_remove(net, "softnet_stat"); | |
3461 | proc_net_remove(net, "dev"); | |
3462 | } | |
3463 | ||
022cbae6 | 3464 | static struct pernet_operations __net_initdata dev_proc_ops = { |
881d966b EB |
3465 | .init = dev_proc_net_init, |
3466 | .exit = dev_proc_net_exit, | |
3467 | }; | |
3468 | ||
3469 | static int __init dev_proc_init(void) | |
3470 | { | |
3471 | return register_pernet_subsys(&dev_proc_ops); | |
3472 | } | |
1da177e4 LT |
3473 | #else |
3474 | #define dev_proc_init() 0 | |
3475 | #endif /* CONFIG_PROC_FS */ | |
3476 | ||
3477 | ||
3478 | /** | |
3479 | * netdev_set_master - set up master/slave pair | |
3480 | * @slave: slave device | |
3481 | * @master: new master device | |
3482 | * | |
3483 | * Changes the master device of the slave. Pass %NULL to break the | |
3484 | * bonding. The caller must hold the RTNL semaphore. On a failure | |
3485 | * a negative errno code is returned. On success the reference counts | |
3486 | * are adjusted, %RTM_NEWLINK is sent to the routing socket and the | |
3487 | * function returns zero. | |
3488 | */ | |
3489 | int netdev_set_master(struct net_device *slave, struct net_device *master) | |
3490 | { | |
3491 | struct net_device *old = slave->master; | |
3492 | ||
3493 | ASSERT_RTNL(); | |
3494 | ||
3495 | if (master) { | |
3496 | if (old) | |
3497 | return -EBUSY; | |
3498 | dev_hold(master); | |
3499 | } | |
3500 | ||
3501 | slave->master = master; | |
4ec93edb | 3502 | |
1da177e4 LT |
3503 | synchronize_net(); |
3504 | ||
3505 | if (old) | |
3506 | dev_put(old); | |
3507 | ||
3508 | if (master) | |
3509 | slave->flags |= IFF_SLAVE; | |
3510 | else | |
3511 | slave->flags &= ~IFF_SLAVE; | |
3512 | ||
3513 | rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE); | |
3514 | return 0; | |
3515 | } | |
d1b19dff | 3516 | EXPORT_SYMBOL(netdev_set_master); |
1da177e4 | 3517 | |
b6c40d68 PM |
3518 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
3519 | { | |
d314774c SH |
3520 | const struct net_device_ops *ops = dev->netdev_ops; |
3521 | ||
3522 | if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags) | |
3523 | ops->ndo_change_rx_flags(dev, flags); | |
b6c40d68 PM |
3524 | } |
3525 | ||
dad9b335 | 3526 | static int __dev_set_promiscuity(struct net_device *dev, int inc) |
1da177e4 LT |
3527 | { |
3528 | unsigned short old_flags = dev->flags; | |
8192b0c4 DH |
3529 | uid_t uid; |
3530 | gid_t gid; | |
1da177e4 | 3531 | |
24023451 PM |
3532 | ASSERT_RTNL(); |
3533 | ||
dad9b335 WC |
3534 | dev->flags |= IFF_PROMISC; |
3535 | dev->promiscuity += inc; | |
3536 | if (dev->promiscuity == 0) { | |
3537 | /* | |
3538 | * Avoid overflow. | |
3539 | * If inc causes overflow, untouch promisc and return error. | |
3540 | */ | |
3541 | if (inc < 0) | |
3542 | dev->flags &= ~IFF_PROMISC; | |
3543 | else { | |
3544 | dev->promiscuity -= inc; | |
3545 | printk(KERN_WARNING "%s: promiscuity touches roof, " | |
3546 | "set promiscuity failed, promiscuity feature " | |
3547 | "of device might be broken.\n", dev->name); | |
3548 | return -EOVERFLOW; | |
3549 | } | |
3550 | } | |
52609c0b | 3551 | if (dev->flags != old_flags) { |
1da177e4 LT |
3552 | printk(KERN_INFO "device %s %s promiscuous mode\n", |
3553 | dev->name, (dev->flags & IFF_PROMISC) ? "entered" : | |
4ec93edb | 3554 | "left"); |
8192b0c4 DH |
3555 | if (audit_enabled) { |
3556 | current_uid_gid(&uid, &gid); | |
7759db82 KHK |
3557 | audit_log(current->audit_context, GFP_ATOMIC, |
3558 | AUDIT_ANOM_PROMISCUOUS, | |
3559 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
3560 | dev->name, (dev->flags & IFF_PROMISC), | |
3561 | (old_flags & IFF_PROMISC), | |
3562 | audit_get_loginuid(current), | |
8192b0c4 | 3563 | uid, gid, |
7759db82 | 3564 | audit_get_sessionid(current)); |
8192b0c4 | 3565 | } |
24023451 | 3566 | |
b6c40d68 | 3567 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 3568 | } |
dad9b335 | 3569 | return 0; |
1da177e4 LT |
3570 | } |
3571 | ||
4417da66 PM |
3572 | /** |
3573 | * dev_set_promiscuity - update promiscuity count on a device | |
3574 | * @dev: device | |
3575 | * @inc: modifier | |
3576 | * | |
3577 | * Add or remove promiscuity from a device. While the count in the device | |
3578 | * remains above zero the interface remains promiscuous. Once it hits zero | |
3579 | * the device reverts back to normal filtering operation. A negative inc | |
3580 | * value is used to drop promiscuity on the device. | |
dad9b335 | 3581 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 3582 | */ |
dad9b335 | 3583 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 PM |
3584 | { |
3585 | unsigned short old_flags = dev->flags; | |
dad9b335 | 3586 | int err; |
4417da66 | 3587 | |
dad9b335 | 3588 | err = __dev_set_promiscuity(dev, inc); |
4b5a698e | 3589 | if (err < 0) |
dad9b335 | 3590 | return err; |
4417da66 PM |
3591 | if (dev->flags != old_flags) |
3592 | dev_set_rx_mode(dev); | |
dad9b335 | 3593 | return err; |
4417da66 | 3594 | } |
d1b19dff | 3595 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 3596 | |
1da177e4 LT |
3597 | /** |
3598 | * dev_set_allmulti - update allmulti count on a device | |
3599 | * @dev: device | |
3600 | * @inc: modifier | |
3601 | * | |
3602 | * Add or remove reception of all multicast frames to a device. While the | |
3603 | * count in the device remains above zero the interface remains listening | |
3604 | * to all interfaces. Once it hits zero the device reverts back to normal | |
3605 | * filtering operation. A negative @inc value is used to drop the counter | |
3606 | * when releasing a resource needing all multicasts. | |
dad9b335 | 3607 | * Return 0 if successful or a negative errno code on error. |
1da177e4 LT |
3608 | */ |
3609 | ||
dad9b335 | 3610 | int dev_set_allmulti(struct net_device *dev, int inc) |
1da177e4 LT |
3611 | { |
3612 | unsigned short old_flags = dev->flags; | |
3613 | ||
24023451 PM |
3614 | ASSERT_RTNL(); |
3615 | ||
1da177e4 | 3616 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
3617 | dev->allmulti += inc; |
3618 | if (dev->allmulti == 0) { | |
3619 | /* | |
3620 | * Avoid overflow. | |
3621 | * If inc causes overflow, untouch allmulti and return error. | |
3622 | */ | |
3623 | if (inc < 0) | |
3624 | dev->flags &= ~IFF_ALLMULTI; | |
3625 | else { | |
3626 | dev->allmulti -= inc; | |
3627 | printk(KERN_WARNING "%s: allmulti touches roof, " | |
3628 | "set allmulti failed, allmulti feature of " | |
3629 | "device might be broken.\n", dev->name); | |
3630 | return -EOVERFLOW; | |
3631 | } | |
3632 | } | |
24023451 | 3633 | if (dev->flags ^ old_flags) { |
b6c40d68 | 3634 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 3635 | dev_set_rx_mode(dev); |
24023451 | 3636 | } |
dad9b335 | 3637 | return 0; |
4417da66 | 3638 | } |
d1b19dff | 3639 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
3640 | |
3641 | /* | |
3642 | * Upload unicast and multicast address lists to device and | |
3643 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 3644 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
3645 | * are present. |
3646 | */ | |
3647 | void __dev_set_rx_mode(struct net_device *dev) | |
3648 | { | |
d314774c SH |
3649 | const struct net_device_ops *ops = dev->netdev_ops; |
3650 | ||
4417da66 PM |
3651 | /* dev_open will call this function so the list will stay sane. */ |
3652 | if (!(dev->flags&IFF_UP)) | |
3653 | return; | |
3654 | ||
3655 | if (!netif_device_present(dev)) | |
40b77c94 | 3656 | return; |
4417da66 | 3657 | |
d314774c SH |
3658 | if (ops->ndo_set_rx_mode) |
3659 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
3660 | else { |
3661 | /* Unicast addresses changes may only happen under the rtnl, | |
3662 | * therefore calling __dev_set_promiscuity here is safe. | |
3663 | */ | |
31278e71 | 3664 | if (dev->uc.count > 0 && !dev->uc_promisc) { |
4417da66 PM |
3665 | __dev_set_promiscuity(dev, 1); |
3666 | dev->uc_promisc = 1; | |
31278e71 | 3667 | } else if (dev->uc.count == 0 && dev->uc_promisc) { |
4417da66 PM |
3668 | __dev_set_promiscuity(dev, -1); |
3669 | dev->uc_promisc = 0; | |
3670 | } | |
3671 | ||
d314774c SH |
3672 | if (ops->ndo_set_multicast_list) |
3673 | ops->ndo_set_multicast_list(dev); | |
4417da66 PM |
3674 | } |
3675 | } | |
3676 | ||
3677 | void dev_set_rx_mode(struct net_device *dev) | |
3678 | { | |
b9e40857 | 3679 | netif_addr_lock_bh(dev); |
4417da66 | 3680 | __dev_set_rx_mode(dev); |
b9e40857 | 3681 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
3682 | } |
3683 | ||
f001fde5 JP |
3684 | /* hw addresses list handling functions */ |
3685 | ||
31278e71 JP |
3686 | static int __hw_addr_add(struct netdev_hw_addr_list *list, unsigned char *addr, |
3687 | int addr_len, unsigned char addr_type) | |
f001fde5 JP |
3688 | { |
3689 | struct netdev_hw_addr *ha; | |
3690 | int alloc_size; | |
3691 | ||
3692 | if (addr_len > MAX_ADDR_LEN) | |
3693 | return -EINVAL; | |
3694 | ||
31278e71 | 3695 | list_for_each_entry(ha, &list->list, list) { |
ccffad25 JP |
3696 | if (!memcmp(ha->addr, addr, addr_len) && |
3697 | ha->type == addr_type) { | |
3698 | ha->refcount++; | |
3699 | return 0; | |
3700 | } | |
3701 | } | |
3702 | ||
3703 | ||
f001fde5 JP |
3704 | alloc_size = sizeof(*ha); |
3705 | if (alloc_size < L1_CACHE_BYTES) | |
3706 | alloc_size = L1_CACHE_BYTES; | |
3707 | ha = kmalloc(alloc_size, GFP_ATOMIC); | |
3708 | if (!ha) | |
3709 | return -ENOMEM; | |
3710 | memcpy(ha->addr, addr, addr_len); | |
3711 | ha->type = addr_type; | |
ccffad25 JP |
3712 | ha->refcount = 1; |
3713 | ha->synced = false; | |
31278e71 JP |
3714 | list_add_tail_rcu(&ha->list, &list->list); |
3715 | list->count++; | |
f001fde5 JP |
3716 | return 0; |
3717 | } | |
3718 | ||
3719 | static void ha_rcu_free(struct rcu_head *head) | |
3720 | { | |
3721 | struct netdev_hw_addr *ha; | |
3722 | ||
3723 | ha = container_of(head, struct netdev_hw_addr, rcu_head); | |
3724 | kfree(ha); | |
3725 | } | |
3726 | ||
31278e71 JP |
3727 | static int __hw_addr_del(struct netdev_hw_addr_list *list, unsigned char *addr, |
3728 | int addr_len, unsigned char addr_type) | |
f001fde5 JP |
3729 | { |
3730 | struct netdev_hw_addr *ha; | |
f001fde5 | 3731 | |
31278e71 | 3732 | list_for_each_entry(ha, &list->list, list) { |
ccffad25 | 3733 | if (!memcmp(ha->addr, addr, addr_len) && |
f001fde5 | 3734 | (ha->type == addr_type || !addr_type)) { |
ccffad25 JP |
3735 | if (--ha->refcount) |
3736 | return 0; | |
f001fde5 JP |
3737 | list_del_rcu(&ha->list); |
3738 | call_rcu(&ha->rcu_head, ha_rcu_free); | |
31278e71 | 3739 | list->count--; |
f001fde5 JP |
3740 | return 0; |
3741 | } | |
3742 | } | |
3743 | return -ENOENT; | |
3744 | } | |
3745 | ||
31278e71 JP |
3746 | static int __hw_addr_add_multiple(struct netdev_hw_addr_list *to_list, |
3747 | struct netdev_hw_addr_list *from_list, | |
3748 | int addr_len, | |
ccffad25 | 3749 | unsigned char addr_type) |
f001fde5 JP |
3750 | { |
3751 | int err; | |
3752 | struct netdev_hw_addr *ha, *ha2; | |
3753 | unsigned char type; | |
3754 | ||
31278e71 | 3755 | list_for_each_entry(ha, &from_list->list, list) { |
f001fde5 | 3756 | type = addr_type ? addr_type : ha->type; |
31278e71 | 3757 | err = __hw_addr_add(to_list, ha->addr, addr_len, type); |
f001fde5 JP |
3758 | if (err) |
3759 | goto unroll; | |
3760 | } | |
3761 | return 0; | |
3762 | ||
3763 | unroll: | |
31278e71 | 3764 | list_for_each_entry(ha2, &from_list->list, list) { |
f001fde5 JP |
3765 | if (ha2 == ha) |
3766 | break; | |
3767 | type = addr_type ? addr_type : ha2->type; | |
31278e71 | 3768 | __hw_addr_del(to_list, ha2->addr, addr_len, type); |
f001fde5 JP |
3769 | } |
3770 | return err; | |
3771 | } | |
3772 | ||
31278e71 JP |
3773 | static void __hw_addr_del_multiple(struct netdev_hw_addr_list *to_list, |
3774 | struct netdev_hw_addr_list *from_list, | |
3775 | int addr_len, | |
ccffad25 | 3776 | unsigned char addr_type) |
f001fde5 JP |
3777 | { |
3778 | struct netdev_hw_addr *ha; | |
3779 | unsigned char type; | |
3780 | ||
31278e71 | 3781 | list_for_each_entry(ha, &from_list->list, list) { |
f001fde5 | 3782 | type = addr_type ? addr_type : ha->type; |
31278e71 | 3783 | __hw_addr_del(to_list, ha->addr, addr_len, addr_type); |
ccffad25 JP |
3784 | } |
3785 | } | |
3786 | ||
31278e71 JP |
3787 | static int __hw_addr_sync(struct netdev_hw_addr_list *to_list, |
3788 | struct netdev_hw_addr_list *from_list, | |
ccffad25 JP |
3789 | int addr_len) |
3790 | { | |
3791 | int err = 0; | |
3792 | struct netdev_hw_addr *ha, *tmp; | |
3793 | ||
31278e71 | 3794 | list_for_each_entry_safe(ha, tmp, &from_list->list, list) { |
ccffad25 | 3795 | if (!ha->synced) { |
31278e71 | 3796 | err = __hw_addr_add(to_list, ha->addr, |
ccffad25 JP |
3797 | addr_len, ha->type); |
3798 | if (err) | |
3799 | break; | |
3800 | ha->synced = true; | |
3801 | ha->refcount++; | |
3802 | } else if (ha->refcount == 1) { | |
31278e71 JP |
3803 | __hw_addr_del(to_list, ha->addr, addr_len, ha->type); |
3804 | __hw_addr_del(from_list, ha->addr, addr_len, ha->type); | |
ccffad25 | 3805 | } |
f001fde5 | 3806 | } |
ccffad25 | 3807 | return err; |
f001fde5 JP |
3808 | } |
3809 | ||
31278e71 JP |
3810 | static void __hw_addr_unsync(struct netdev_hw_addr_list *to_list, |
3811 | struct netdev_hw_addr_list *from_list, | |
ccffad25 JP |
3812 | int addr_len) |
3813 | { | |
3814 | struct netdev_hw_addr *ha, *tmp; | |
3815 | ||
31278e71 | 3816 | list_for_each_entry_safe(ha, tmp, &from_list->list, list) { |
ccffad25 | 3817 | if (ha->synced) { |
31278e71 | 3818 | __hw_addr_del(to_list, ha->addr, |
ccffad25 JP |
3819 | addr_len, ha->type); |
3820 | ha->synced = false; | |
31278e71 | 3821 | __hw_addr_del(from_list, ha->addr, |
ccffad25 JP |
3822 | addr_len, ha->type); |
3823 | } | |
3824 | } | |
3825 | } | |
3826 | ||
31278e71 | 3827 | static void __hw_addr_flush(struct netdev_hw_addr_list *list) |
f001fde5 JP |
3828 | { |
3829 | struct netdev_hw_addr *ha, *tmp; | |
3830 | ||
31278e71 | 3831 | list_for_each_entry_safe(ha, tmp, &list->list, list) { |
f001fde5 JP |
3832 | list_del_rcu(&ha->list); |
3833 | call_rcu(&ha->rcu_head, ha_rcu_free); | |
3834 | } | |
31278e71 JP |
3835 | list->count = 0; |
3836 | } | |
3837 | ||
3838 | static void __hw_addr_init(struct netdev_hw_addr_list *list) | |
3839 | { | |
3840 | INIT_LIST_HEAD(&list->list); | |
3841 | list->count = 0; | |
f001fde5 JP |
3842 | } |
3843 | ||
3844 | /* Device addresses handling functions */ | |
3845 | ||
3846 | static void dev_addr_flush(struct net_device *dev) | |
3847 | { | |
3848 | /* rtnl_mutex must be held here */ | |
3849 | ||
31278e71 | 3850 | __hw_addr_flush(&dev->dev_addrs); |
f001fde5 JP |
3851 | dev->dev_addr = NULL; |
3852 | } | |
3853 | ||
3854 | static int dev_addr_init(struct net_device *dev) | |
3855 | { | |
3856 | unsigned char addr[MAX_ADDR_LEN]; | |
3857 | struct netdev_hw_addr *ha; | |
3858 | int err; | |
3859 | ||
3860 | /* rtnl_mutex must be held here */ | |
3861 | ||
31278e71 | 3862 | __hw_addr_init(&dev->dev_addrs); |
0c27922e | 3863 | memset(addr, 0, sizeof(addr)); |
31278e71 | 3864 | err = __hw_addr_add(&dev->dev_addrs, addr, sizeof(addr), |
f001fde5 JP |
3865 | NETDEV_HW_ADDR_T_LAN); |
3866 | if (!err) { | |
3867 | /* | |
3868 | * Get the first (previously created) address from the list | |
3869 | * and set dev_addr pointer to this location. | |
3870 | */ | |
31278e71 | 3871 | ha = list_first_entry(&dev->dev_addrs.list, |
f001fde5 JP |
3872 | struct netdev_hw_addr, list); |
3873 | dev->dev_addr = ha->addr; | |
3874 | } | |
3875 | return err; | |
3876 | } | |
3877 | ||
3878 | /** | |
3879 | * dev_addr_add - Add a device address | |
3880 | * @dev: device | |
3881 | * @addr: address to add | |
3882 | * @addr_type: address type | |
3883 | * | |
3884 | * Add a device address to the device or increase the reference count if | |
3885 | * it already exists. | |
3886 | * | |
3887 | * The caller must hold the rtnl_mutex. | |
3888 | */ | |
3889 | int dev_addr_add(struct net_device *dev, unsigned char *addr, | |
3890 | unsigned char addr_type) | |
3891 | { | |
3892 | int err; | |
3893 | ||
3894 | ASSERT_RTNL(); | |
3895 | ||
31278e71 | 3896 | err = __hw_addr_add(&dev->dev_addrs, addr, dev->addr_len, addr_type); |
f001fde5 JP |
3897 | if (!err) |
3898 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); | |
3899 | return err; | |
3900 | } | |
3901 | EXPORT_SYMBOL(dev_addr_add); | |
3902 | ||
3903 | /** | |
3904 | * dev_addr_del - Release a device address. | |
3905 | * @dev: device | |
3906 | * @addr: address to delete | |
3907 | * @addr_type: address type | |
3908 | * | |
3909 | * Release reference to a device address and remove it from the device | |
3910 | * if the reference count drops to zero. | |
3911 | * | |
3912 | * The caller must hold the rtnl_mutex. | |
3913 | */ | |
3914 | int dev_addr_del(struct net_device *dev, unsigned char *addr, | |
3915 | unsigned char addr_type) | |
3916 | { | |
3917 | int err; | |
ccffad25 | 3918 | struct netdev_hw_addr *ha; |
f001fde5 JP |
3919 | |
3920 | ASSERT_RTNL(); | |
3921 | ||
ccffad25 JP |
3922 | /* |
3923 | * We can not remove the first address from the list because | |
3924 | * dev->dev_addr points to that. | |
3925 | */ | |
31278e71 JP |
3926 | ha = list_first_entry(&dev->dev_addrs.list, |
3927 | struct netdev_hw_addr, list); | |
ccffad25 JP |
3928 | if (ha->addr == dev->dev_addr && ha->refcount == 1) |
3929 | return -ENOENT; | |
3930 | ||
31278e71 | 3931 | err = __hw_addr_del(&dev->dev_addrs, addr, dev->addr_len, |
ccffad25 | 3932 | addr_type); |
f001fde5 JP |
3933 | if (!err) |
3934 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); | |
3935 | return err; | |
3936 | } | |
3937 | EXPORT_SYMBOL(dev_addr_del); | |
3938 | ||
3939 | /** | |
3940 | * dev_addr_add_multiple - Add device addresses from another device | |
3941 | * @to_dev: device to which addresses will be added | |
3942 | * @from_dev: device from which addresses will be added | |
3943 | * @addr_type: address type - 0 means type will be used from from_dev | |
3944 | * | |
3945 | * Add device addresses of the one device to another. | |
3946 | ** | |
3947 | * The caller must hold the rtnl_mutex. | |
3948 | */ | |
3949 | int dev_addr_add_multiple(struct net_device *to_dev, | |
3950 | struct net_device *from_dev, | |
3951 | unsigned char addr_type) | |
3952 | { | |
3953 | int err; | |
3954 | ||
3955 | ASSERT_RTNL(); | |
3956 | ||
3957 | if (from_dev->addr_len != to_dev->addr_len) | |
3958 | return -EINVAL; | |
31278e71 | 3959 | err = __hw_addr_add_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs, |
ccffad25 | 3960 | to_dev->addr_len, addr_type); |
f001fde5 JP |
3961 | if (!err) |
3962 | call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev); | |
3963 | return err; | |
3964 | } | |
3965 | EXPORT_SYMBOL(dev_addr_add_multiple); | |
3966 | ||
3967 | /** | |
3968 | * dev_addr_del_multiple - Delete device addresses by another device | |
3969 | * @to_dev: device where the addresses will be deleted | |
3970 | * @from_dev: device by which addresses the addresses will be deleted | |
3971 | * @addr_type: address type - 0 means type will used from from_dev | |
3972 | * | |
3973 | * Deletes addresses in to device by the list of addresses in from device. | |
3974 | * | |
3975 | * The caller must hold the rtnl_mutex. | |
3976 | */ | |
3977 | int dev_addr_del_multiple(struct net_device *to_dev, | |
3978 | struct net_device *from_dev, | |
3979 | unsigned char addr_type) | |
3980 | { | |
3981 | ASSERT_RTNL(); | |
3982 | ||
3983 | if (from_dev->addr_len != to_dev->addr_len) | |
3984 | return -EINVAL; | |
31278e71 | 3985 | __hw_addr_del_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs, |
ccffad25 | 3986 | to_dev->addr_len, addr_type); |
f001fde5 JP |
3987 | call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev); |
3988 | return 0; | |
3989 | } | |
3990 | EXPORT_SYMBOL(dev_addr_del_multiple); | |
3991 | ||
31278e71 | 3992 | /* multicast addresses handling functions */ |
f001fde5 | 3993 | |
61cbc2fc PM |
3994 | int __dev_addr_delete(struct dev_addr_list **list, int *count, |
3995 | void *addr, int alen, int glbl) | |
bf742482 PM |
3996 | { |
3997 | struct dev_addr_list *da; | |
3998 | ||
3999 | for (; (da = *list) != NULL; list = &da->next) { | |
4000 | if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 && | |
4001 | alen == da->da_addrlen) { | |
4002 | if (glbl) { | |
4003 | int old_glbl = da->da_gusers; | |
4004 | da->da_gusers = 0; | |
4005 | if (old_glbl == 0) | |
4006 | break; | |
4007 | } | |
4008 | if (--da->da_users) | |
4009 | return 0; | |
4010 | ||
4011 | *list = da->next; | |
4012 | kfree(da); | |
61cbc2fc | 4013 | (*count)--; |
bf742482 PM |
4014 | return 0; |
4015 | } | |
4016 | } | |
4017 | return -ENOENT; | |
4018 | } | |
4019 | ||
61cbc2fc PM |
4020 | int __dev_addr_add(struct dev_addr_list **list, int *count, |
4021 | void *addr, int alen, int glbl) | |
bf742482 PM |
4022 | { |
4023 | struct dev_addr_list *da; | |
4024 | ||
4025 | for (da = *list; da != NULL; da = da->next) { | |
4026 | if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 && | |
4027 | da->da_addrlen == alen) { | |
4028 | if (glbl) { | |
4029 | int old_glbl = da->da_gusers; | |
4030 | da->da_gusers = 1; | |
4031 | if (old_glbl) | |
4032 | return 0; | |
4033 | } | |
4034 | da->da_users++; | |
4035 | return 0; | |
4036 | } | |
4037 | } | |
4038 | ||
12aa343a | 4039 | da = kzalloc(sizeof(*da), GFP_ATOMIC); |
bf742482 PM |
4040 | if (da == NULL) |
4041 | return -ENOMEM; | |
4042 | memcpy(da->da_addr, addr, alen); | |
4043 | da->da_addrlen = alen; | |
4044 | da->da_users = 1; | |
4045 | da->da_gusers = glbl ? 1 : 0; | |
4046 | da->next = *list; | |
4047 | *list = da; | |
61cbc2fc | 4048 | (*count)++; |
bf742482 PM |
4049 | return 0; |
4050 | } | |
4051 | ||
4417da66 PM |
4052 | /** |
4053 | * dev_unicast_delete - Release secondary unicast address. | |
4054 | * @dev: device | |
0ed72ec4 | 4055 | * @addr: address to delete |
4417da66 PM |
4056 | * |
4057 | * Release reference to a secondary unicast address and remove it | |
0ed72ec4 | 4058 | * from the device if the reference count drops to zero. |
4417da66 PM |
4059 | * |
4060 | * The caller must hold the rtnl_mutex. | |
4061 | */ | |
ccffad25 | 4062 | int dev_unicast_delete(struct net_device *dev, void *addr) |
4417da66 PM |
4063 | { |
4064 | int err; | |
4065 | ||
4066 | ASSERT_RTNL(); | |
4067 | ||
a6ac65db | 4068 | netif_addr_lock_bh(dev); |
31278e71 JP |
4069 | err = __hw_addr_del(&dev->uc, addr, dev->addr_len, |
4070 | NETDEV_HW_ADDR_T_UNICAST); | |
61cbc2fc | 4071 | if (!err) |
4417da66 | 4072 | __dev_set_rx_mode(dev); |
a6ac65db | 4073 | netif_addr_unlock_bh(dev); |
4417da66 PM |
4074 | return err; |
4075 | } | |
4076 | EXPORT_SYMBOL(dev_unicast_delete); | |
4077 | ||
4078 | /** | |
4079 | * dev_unicast_add - add a secondary unicast address | |
4080 | * @dev: device | |
5dbaec5d | 4081 | * @addr: address to add |
4417da66 PM |
4082 | * |
4083 | * Add a secondary unicast address to the device or increase | |
4084 | * the reference count if it already exists. | |
4085 | * | |
4086 | * The caller must hold the rtnl_mutex. | |
4087 | */ | |
ccffad25 | 4088 | int dev_unicast_add(struct net_device *dev, void *addr) |
4417da66 PM |
4089 | { |
4090 | int err; | |
4091 | ||
4092 | ASSERT_RTNL(); | |
4093 | ||
a6ac65db | 4094 | netif_addr_lock_bh(dev); |
31278e71 JP |
4095 | err = __hw_addr_add(&dev->uc, addr, dev->addr_len, |
4096 | NETDEV_HW_ADDR_T_UNICAST); | |
61cbc2fc | 4097 | if (!err) |
4417da66 | 4098 | __dev_set_rx_mode(dev); |
a6ac65db | 4099 | netif_addr_unlock_bh(dev); |
4417da66 PM |
4100 | return err; |
4101 | } | |
4102 | EXPORT_SYMBOL(dev_unicast_add); | |
4103 | ||
e83a2ea8 CL |
4104 | int __dev_addr_sync(struct dev_addr_list **to, int *to_count, |
4105 | struct dev_addr_list **from, int *from_count) | |
4106 | { | |
4107 | struct dev_addr_list *da, *next; | |
4108 | int err = 0; | |
4109 | ||
4110 | da = *from; | |
4111 | while (da != NULL) { | |
4112 | next = da->next; | |
4113 | if (!da->da_synced) { | |
4114 | err = __dev_addr_add(to, to_count, | |
4115 | da->da_addr, da->da_addrlen, 0); | |
4116 | if (err < 0) | |
4117 | break; | |
4118 | da->da_synced = 1; | |
4119 | da->da_users++; | |
4120 | } else if (da->da_users == 1) { | |
4121 | __dev_addr_delete(to, to_count, | |
4122 | da->da_addr, da->da_addrlen, 0); | |
4123 | __dev_addr_delete(from, from_count, | |
4124 | da->da_addr, da->da_addrlen, 0); | |
4125 | } | |
4126 | da = next; | |
4127 | } | |
4128 | return err; | |
4129 | } | |
c4029083 | 4130 | EXPORT_SYMBOL_GPL(__dev_addr_sync); |
e83a2ea8 CL |
4131 | |
4132 | void __dev_addr_unsync(struct dev_addr_list **to, int *to_count, | |
4133 | struct dev_addr_list **from, int *from_count) | |
4134 | { | |
4135 | struct dev_addr_list *da, *next; | |
4136 | ||
4137 | da = *from; | |
4138 | while (da != NULL) { | |
4139 | next = da->next; | |
4140 | if (da->da_synced) { | |
4141 | __dev_addr_delete(to, to_count, | |
4142 | da->da_addr, da->da_addrlen, 0); | |
4143 | da->da_synced = 0; | |
4144 | __dev_addr_delete(from, from_count, | |
4145 | da->da_addr, da->da_addrlen, 0); | |
4146 | } | |
4147 | da = next; | |
4148 | } | |
4149 | } | |
c4029083 | 4150 | EXPORT_SYMBOL_GPL(__dev_addr_unsync); |
e83a2ea8 CL |
4151 | |
4152 | /** | |
4153 | * dev_unicast_sync - Synchronize device's unicast list to another device | |
4154 | * @to: destination device | |
4155 | * @from: source device | |
4156 | * | |
4157 | * Add newly added addresses to the destination device and release | |
a6ac65db JP |
4158 | * addresses that have no users left. The source device must be |
4159 | * locked by netif_tx_lock_bh. | |
e83a2ea8 CL |
4160 | * |
4161 | * This function is intended to be called from the dev->set_rx_mode | |
4162 | * function of layered software devices. | |
4163 | */ | |
4164 | int dev_unicast_sync(struct net_device *to, struct net_device *from) | |
4165 | { | |
4166 | int err = 0; | |
4167 | ||
ccffad25 JP |
4168 | if (to->addr_len != from->addr_len) |
4169 | return -EINVAL; | |
4170 | ||
a6ac65db | 4171 | netif_addr_lock_bh(to); |
31278e71 | 4172 | err = __hw_addr_sync(&to->uc, &from->uc, to->addr_len); |
e83a2ea8 CL |
4173 | if (!err) |
4174 | __dev_set_rx_mode(to); | |
a6ac65db | 4175 | netif_addr_unlock_bh(to); |
e83a2ea8 CL |
4176 | return err; |
4177 | } | |
4178 | EXPORT_SYMBOL(dev_unicast_sync); | |
4179 | ||
4180 | /** | |
bc2cda1e | 4181 | * dev_unicast_unsync - Remove synchronized addresses from the destination device |
e83a2ea8 CL |
4182 | * @to: destination device |
4183 | * @from: source device | |
4184 | * | |
4185 | * Remove all addresses that were added to the destination device by | |
4186 | * dev_unicast_sync(). This function is intended to be called from the | |
4187 | * dev->stop function of layered software devices. | |
4188 | */ | |
4189 | void dev_unicast_unsync(struct net_device *to, struct net_device *from) | |
4190 | { | |
ccffad25 JP |
4191 | if (to->addr_len != from->addr_len) |
4192 | return; | |
e83a2ea8 | 4193 | |
a6ac65db JP |
4194 | netif_addr_lock_bh(from); |
4195 | netif_addr_lock(to); | |
31278e71 | 4196 | __hw_addr_unsync(&to->uc, &from->uc, to->addr_len); |
ccffad25 | 4197 | __dev_set_rx_mode(to); |
a6ac65db JP |
4198 | netif_addr_unlock(to); |
4199 | netif_addr_unlock_bh(from); | |
e83a2ea8 CL |
4200 | } |
4201 | EXPORT_SYMBOL(dev_unicast_unsync); | |
4202 | ||
ccffad25 JP |
4203 | static void dev_unicast_flush(struct net_device *dev) |
4204 | { | |
a6ac65db | 4205 | netif_addr_lock_bh(dev); |
31278e71 | 4206 | __hw_addr_flush(&dev->uc); |
a6ac65db | 4207 | netif_addr_unlock_bh(dev); |
ccffad25 JP |
4208 | } |
4209 | ||
4210 | static void dev_unicast_init(struct net_device *dev) | |
4211 | { | |
31278e71 | 4212 | __hw_addr_init(&dev->uc); |
ccffad25 JP |
4213 | } |
4214 | ||
4215 | ||
12972621 DC |
4216 | static void __dev_addr_discard(struct dev_addr_list **list) |
4217 | { | |
4218 | struct dev_addr_list *tmp; | |
4219 | ||
4220 | while (*list != NULL) { | |
4221 | tmp = *list; | |
4222 | *list = tmp->next; | |
4223 | if (tmp->da_users > tmp->da_gusers) | |
4224 | printk("__dev_addr_discard: address leakage! " | |
4225 | "da_users=%d\n", tmp->da_users); | |
4226 | kfree(tmp); | |
4227 | } | |
4228 | } | |
4229 | ||
26cc2522 | 4230 | static void dev_addr_discard(struct net_device *dev) |
4417da66 | 4231 | { |
b9e40857 | 4232 | netif_addr_lock_bh(dev); |
26cc2522 | 4233 | |
456ad75c DC |
4234 | __dev_addr_discard(&dev->mc_list); |
4235 | dev->mc_count = 0; | |
26cc2522 | 4236 | |
b9e40857 | 4237 | netif_addr_unlock_bh(dev); |
456ad75c DC |
4238 | } |
4239 | ||
f0db275a SH |
4240 | /** |
4241 | * dev_get_flags - get flags reported to userspace | |
4242 | * @dev: device | |
4243 | * | |
4244 | * Get the combination of flag bits exported through APIs to userspace. | |
4245 | */ | |
1da177e4 LT |
4246 | unsigned dev_get_flags(const struct net_device *dev) |
4247 | { | |
4248 | unsigned flags; | |
4249 | ||
4250 | flags = (dev->flags & ~(IFF_PROMISC | | |
4251 | IFF_ALLMULTI | | |
b00055aa SR |
4252 | IFF_RUNNING | |
4253 | IFF_LOWER_UP | | |
4254 | IFF_DORMANT)) | | |
1da177e4 LT |
4255 | (dev->gflags & (IFF_PROMISC | |
4256 | IFF_ALLMULTI)); | |
4257 | ||
b00055aa SR |
4258 | if (netif_running(dev)) { |
4259 | if (netif_oper_up(dev)) | |
4260 | flags |= IFF_RUNNING; | |
4261 | if (netif_carrier_ok(dev)) | |
4262 | flags |= IFF_LOWER_UP; | |
4263 | if (netif_dormant(dev)) | |
4264 | flags |= IFF_DORMANT; | |
4265 | } | |
1da177e4 LT |
4266 | |
4267 | return flags; | |
4268 | } | |
d1b19dff | 4269 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 4270 | |
f0db275a SH |
4271 | /** |
4272 | * dev_change_flags - change device settings | |
4273 | * @dev: device | |
4274 | * @flags: device state flags | |
4275 | * | |
4276 | * Change settings on device based state flags. The flags are | |
4277 | * in the userspace exported format. | |
4278 | */ | |
1da177e4 LT |
4279 | int dev_change_flags(struct net_device *dev, unsigned flags) |
4280 | { | |
7c355f53 | 4281 | int ret, changes; |
1da177e4 LT |
4282 | int old_flags = dev->flags; |
4283 | ||
24023451 PM |
4284 | ASSERT_RTNL(); |
4285 | ||
1da177e4 LT |
4286 | /* |
4287 | * Set the flags on our device. | |
4288 | */ | |
4289 | ||
4290 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
4291 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
4292 | IFF_AUTOMEDIA)) | | |
4293 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
4294 | IFF_ALLMULTI)); | |
4295 | ||
4296 | /* | |
4297 | * Load in the correct multicast list now the flags have changed. | |
4298 | */ | |
4299 | ||
b6c40d68 PM |
4300 | if ((old_flags ^ flags) & IFF_MULTICAST) |
4301 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 4302 | |
4417da66 | 4303 | dev_set_rx_mode(dev); |
1da177e4 LT |
4304 | |
4305 | /* | |
4306 | * Have we downed the interface. We handle IFF_UP ourselves | |
4307 | * according to user attempts to set it, rather than blindly | |
4308 | * setting it. | |
4309 | */ | |
4310 | ||
4311 | ret = 0; | |
4312 | if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */ | |
4313 | ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev); | |
4314 | ||
4315 | if (!ret) | |
4417da66 | 4316 | dev_set_rx_mode(dev); |
1da177e4 LT |
4317 | } |
4318 | ||
4319 | if (dev->flags & IFF_UP && | |
d1b19dff | 4320 | ((old_flags ^ dev->flags) & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | |
1da177e4 | 4321 | IFF_VOLATILE))) |
056925ab | 4322 | call_netdevice_notifiers(NETDEV_CHANGE, dev); |
1da177e4 LT |
4323 | |
4324 | if ((flags ^ dev->gflags) & IFF_PROMISC) { | |
d1b19dff ED |
4325 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
4326 | ||
1da177e4 LT |
4327 | dev->gflags ^= IFF_PROMISC; |
4328 | dev_set_promiscuity(dev, inc); | |
4329 | } | |
4330 | ||
4331 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
4332 | is important. Some (broken) drivers set IFF_PROMISC, when | |
4333 | IFF_ALLMULTI is requested not asking us and not reporting. | |
4334 | */ | |
4335 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
4336 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
4337 | ||
1da177e4 LT |
4338 | dev->gflags ^= IFF_ALLMULTI; |
4339 | dev_set_allmulti(dev, inc); | |
4340 | } | |
4341 | ||
7c355f53 TG |
4342 | /* Exclude state transition flags, already notified */ |
4343 | changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING); | |
4344 | if (changes) | |
4345 | rtmsg_ifinfo(RTM_NEWLINK, dev, changes); | |
1da177e4 LT |
4346 | |
4347 | return ret; | |
4348 | } | |
d1b19dff | 4349 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 4350 | |
f0db275a SH |
4351 | /** |
4352 | * dev_set_mtu - Change maximum transfer unit | |
4353 | * @dev: device | |
4354 | * @new_mtu: new transfer unit | |
4355 | * | |
4356 | * Change the maximum transfer size of the network device. | |
4357 | */ | |
1da177e4 LT |
4358 | int dev_set_mtu(struct net_device *dev, int new_mtu) |
4359 | { | |
d314774c | 4360 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
4361 | int err; |
4362 | ||
4363 | if (new_mtu == dev->mtu) | |
4364 | return 0; | |
4365 | ||
4366 | /* MTU must be positive. */ | |
4367 | if (new_mtu < 0) | |
4368 | return -EINVAL; | |
4369 | ||
4370 | if (!netif_device_present(dev)) | |
4371 | return -ENODEV; | |
4372 | ||
4373 | err = 0; | |
d314774c SH |
4374 | if (ops->ndo_change_mtu) |
4375 | err = ops->ndo_change_mtu(dev, new_mtu); | |
1da177e4 LT |
4376 | else |
4377 | dev->mtu = new_mtu; | |
d314774c | 4378 | |
1da177e4 | 4379 | if (!err && dev->flags & IFF_UP) |
056925ab | 4380 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); |
1da177e4 LT |
4381 | return err; |
4382 | } | |
d1b19dff | 4383 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 4384 | |
f0db275a SH |
4385 | /** |
4386 | * dev_set_mac_address - Change Media Access Control Address | |
4387 | * @dev: device | |
4388 | * @sa: new address | |
4389 | * | |
4390 | * Change the hardware (MAC) address of the device | |
4391 | */ | |
1da177e4 LT |
4392 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) |
4393 | { | |
d314774c | 4394 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
4395 | int err; |
4396 | ||
d314774c | 4397 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
4398 | return -EOPNOTSUPP; |
4399 | if (sa->sa_family != dev->type) | |
4400 | return -EINVAL; | |
4401 | if (!netif_device_present(dev)) | |
4402 | return -ENODEV; | |
d314774c | 4403 | err = ops->ndo_set_mac_address(dev, sa); |
1da177e4 | 4404 | if (!err) |
056925ab | 4405 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
1da177e4 LT |
4406 | return err; |
4407 | } | |
d1b19dff | 4408 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 LT |
4409 | |
4410 | /* | |
3710becf | 4411 | * Perform the SIOCxIFxxx calls, inside rcu_read_lock() |
1da177e4 | 4412 | */ |
14e3e079 | 4413 | static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd) |
1da177e4 LT |
4414 | { |
4415 | int err; | |
3710becf | 4416 | struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name); |
1da177e4 LT |
4417 | |
4418 | if (!dev) | |
4419 | return -ENODEV; | |
4420 | ||
4421 | switch (cmd) { | |
d1b19dff ED |
4422 | case SIOCGIFFLAGS: /* Get interface flags */ |
4423 | ifr->ifr_flags = (short) dev_get_flags(dev); | |
4424 | return 0; | |
1da177e4 | 4425 | |
d1b19dff ED |
4426 | case SIOCGIFMETRIC: /* Get the metric on the interface |
4427 | (currently unused) */ | |
4428 | ifr->ifr_metric = 0; | |
4429 | return 0; | |
1da177e4 | 4430 | |
d1b19dff ED |
4431 | case SIOCGIFMTU: /* Get the MTU of a device */ |
4432 | ifr->ifr_mtu = dev->mtu; | |
4433 | return 0; | |
1da177e4 | 4434 | |
d1b19dff ED |
4435 | case SIOCGIFHWADDR: |
4436 | if (!dev->addr_len) | |
4437 | memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data); | |
4438 | else | |
4439 | memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr, | |
4440 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
4441 | ifr->ifr_hwaddr.sa_family = dev->type; | |
4442 | return 0; | |
1da177e4 | 4443 | |
d1b19dff ED |
4444 | case SIOCGIFSLAVE: |
4445 | err = -EINVAL; | |
4446 | break; | |
14e3e079 | 4447 | |
d1b19dff ED |
4448 | case SIOCGIFMAP: |
4449 | ifr->ifr_map.mem_start = dev->mem_start; | |
4450 | ifr->ifr_map.mem_end = dev->mem_end; | |
4451 | ifr->ifr_map.base_addr = dev->base_addr; | |
4452 | ifr->ifr_map.irq = dev->irq; | |
4453 | ifr->ifr_map.dma = dev->dma; | |
4454 | ifr->ifr_map.port = dev->if_port; | |
4455 | return 0; | |
14e3e079 | 4456 | |
d1b19dff ED |
4457 | case SIOCGIFINDEX: |
4458 | ifr->ifr_ifindex = dev->ifindex; | |
4459 | return 0; | |
14e3e079 | 4460 | |
d1b19dff ED |
4461 | case SIOCGIFTXQLEN: |
4462 | ifr->ifr_qlen = dev->tx_queue_len; | |
4463 | return 0; | |
14e3e079 | 4464 | |
d1b19dff ED |
4465 | default: |
4466 | /* dev_ioctl() should ensure this case | |
4467 | * is never reached | |
4468 | */ | |
4469 | WARN_ON(1); | |
4470 | err = -EINVAL; | |
4471 | break; | |
14e3e079 JG |
4472 | |
4473 | } | |
4474 | return err; | |
4475 | } | |
4476 | ||
4477 | /* | |
4478 | * Perform the SIOCxIFxxx calls, inside rtnl_lock() | |
4479 | */ | |
4480 | static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd) | |
4481 | { | |
4482 | int err; | |
4483 | struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name); | |
5f2f6da7 | 4484 | const struct net_device_ops *ops; |
14e3e079 JG |
4485 | |
4486 | if (!dev) | |
4487 | return -ENODEV; | |
4488 | ||
5f2f6da7 JP |
4489 | ops = dev->netdev_ops; |
4490 | ||
14e3e079 | 4491 | switch (cmd) { |
d1b19dff ED |
4492 | case SIOCSIFFLAGS: /* Set interface flags */ |
4493 | return dev_change_flags(dev, ifr->ifr_flags); | |
14e3e079 | 4494 | |
d1b19dff ED |
4495 | case SIOCSIFMETRIC: /* Set the metric on the interface |
4496 | (currently unused) */ | |
4497 | return -EOPNOTSUPP; | |
14e3e079 | 4498 | |
d1b19dff ED |
4499 | case SIOCSIFMTU: /* Set the MTU of a device */ |
4500 | return dev_set_mtu(dev, ifr->ifr_mtu); | |
1da177e4 | 4501 | |
d1b19dff ED |
4502 | case SIOCSIFHWADDR: |
4503 | return dev_set_mac_address(dev, &ifr->ifr_hwaddr); | |
1da177e4 | 4504 | |
d1b19dff ED |
4505 | case SIOCSIFHWBROADCAST: |
4506 | if (ifr->ifr_hwaddr.sa_family != dev->type) | |
4507 | return -EINVAL; | |
4508 | memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data, | |
4509 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
4510 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); | |
4511 | return 0; | |
1da177e4 | 4512 | |
d1b19dff ED |
4513 | case SIOCSIFMAP: |
4514 | if (ops->ndo_set_config) { | |
1da177e4 LT |
4515 | if (!netif_device_present(dev)) |
4516 | return -ENODEV; | |
d1b19dff ED |
4517 | return ops->ndo_set_config(dev, &ifr->ifr_map); |
4518 | } | |
4519 | return -EOPNOTSUPP; | |
1da177e4 | 4520 | |
d1b19dff ED |
4521 | case SIOCADDMULTI: |
4522 | if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) || | |
4523 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) | |
4524 | return -EINVAL; | |
4525 | if (!netif_device_present(dev)) | |
4526 | return -ENODEV; | |
4527 | return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data, | |
4528 | dev->addr_len, 1); | |
4529 | ||
4530 | case SIOCDELMULTI: | |
4531 | if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) || | |
4532 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) | |
4533 | return -EINVAL; | |
4534 | if (!netif_device_present(dev)) | |
4535 | return -ENODEV; | |
4536 | return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data, | |
4537 | dev->addr_len, 1); | |
1da177e4 | 4538 | |
d1b19dff ED |
4539 | case SIOCSIFTXQLEN: |
4540 | if (ifr->ifr_qlen < 0) | |
4541 | return -EINVAL; | |
4542 | dev->tx_queue_len = ifr->ifr_qlen; | |
4543 | return 0; | |
1da177e4 | 4544 | |
d1b19dff ED |
4545 | case SIOCSIFNAME: |
4546 | ifr->ifr_newname[IFNAMSIZ-1] = '\0'; | |
4547 | return dev_change_name(dev, ifr->ifr_newname); | |
1da177e4 | 4548 | |
d1b19dff ED |
4549 | /* |
4550 | * Unknown or private ioctl | |
4551 | */ | |
4552 | default: | |
4553 | if ((cmd >= SIOCDEVPRIVATE && | |
4554 | cmd <= SIOCDEVPRIVATE + 15) || | |
4555 | cmd == SIOCBONDENSLAVE || | |
4556 | cmd == SIOCBONDRELEASE || | |
4557 | cmd == SIOCBONDSETHWADDR || | |
4558 | cmd == SIOCBONDSLAVEINFOQUERY || | |
4559 | cmd == SIOCBONDINFOQUERY || | |
4560 | cmd == SIOCBONDCHANGEACTIVE || | |
4561 | cmd == SIOCGMIIPHY || | |
4562 | cmd == SIOCGMIIREG || | |
4563 | cmd == SIOCSMIIREG || | |
4564 | cmd == SIOCBRADDIF || | |
4565 | cmd == SIOCBRDELIF || | |
4566 | cmd == SIOCSHWTSTAMP || | |
4567 | cmd == SIOCWANDEV) { | |
4568 | err = -EOPNOTSUPP; | |
4569 | if (ops->ndo_do_ioctl) { | |
4570 | if (netif_device_present(dev)) | |
4571 | err = ops->ndo_do_ioctl(dev, ifr, cmd); | |
4572 | else | |
4573 | err = -ENODEV; | |
4574 | } | |
4575 | } else | |
4576 | err = -EINVAL; | |
1da177e4 LT |
4577 | |
4578 | } | |
4579 | return err; | |
4580 | } | |
4581 | ||
4582 | /* | |
4583 | * This function handles all "interface"-type I/O control requests. The actual | |
4584 | * 'doing' part of this is dev_ifsioc above. | |
4585 | */ | |
4586 | ||
4587 | /** | |
4588 | * dev_ioctl - network device ioctl | |
c4ea43c5 | 4589 | * @net: the applicable net namespace |
1da177e4 LT |
4590 | * @cmd: command to issue |
4591 | * @arg: pointer to a struct ifreq in user space | |
4592 | * | |
4593 | * Issue ioctl functions to devices. This is normally called by the | |
4594 | * user space syscall interfaces but can sometimes be useful for | |
4595 | * other purposes. The return value is the return from the syscall if | |
4596 | * positive or a negative errno code on error. | |
4597 | */ | |
4598 | ||
881d966b | 4599 | int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg) |
1da177e4 LT |
4600 | { |
4601 | struct ifreq ifr; | |
4602 | int ret; | |
4603 | char *colon; | |
4604 | ||
4605 | /* One special case: SIOCGIFCONF takes ifconf argument | |
4606 | and requires shared lock, because it sleeps writing | |
4607 | to user space. | |
4608 | */ | |
4609 | ||
4610 | if (cmd == SIOCGIFCONF) { | |
6756ae4b | 4611 | rtnl_lock(); |
881d966b | 4612 | ret = dev_ifconf(net, (char __user *) arg); |
6756ae4b | 4613 | rtnl_unlock(); |
1da177e4 LT |
4614 | return ret; |
4615 | } | |
4616 | if (cmd == SIOCGIFNAME) | |
881d966b | 4617 | return dev_ifname(net, (struct ifreq __user *)arg); |
1da177e4 LT |
4618 | |
4619 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
4620 | return -EFAULT; | |
4621 | ||
4622 | ifr.ifr_name[IFNAMSIZ-1] = 0; | |
4623 | ||
4624 | colon = strchr(ifr.ifr_name, ':'); | |
4625 | if (colon) | |
4626 | *colon = 0; | |
4627 | ||
4628 | /* | |
4629 | * See which interface the caller is talking about. | |
4630 | */ | |
4631 | ||
4632 | switch (cmd) { | |
d1b19dff ED |
4633 | /* |
4634 | * These ioctl calls: | |
4635 | * - can be done by all. | |
4636 | * - atomic and do not require locking. | |
4637 | * - return a value | |
4638 | */ | |
4639 | case SIOCGIFFLAGS: | |
4640 | case SIOCGIFMETRIC: | |
4641 | case SIOCGIFMTU: | |
4642 | case SIOCGIFHWADDR: | |
4643 | case SIOCGIFSLAVE: | |
4644 | case SIOCGIFMAP: | |
4645 | case SIOCGIFINDEX: | |
4646 | case SIOCGIFTXQLEN: | |
4647 | dev_load(net, ifr.ifr_name); | |
3710becf | 4648 | rcu_read_lock(); |
d1b19dff | 4649 | ret = dev_ifsioc_locked(net, &ifr, cmd); |
3710becf | 4650 | rcu_read_unlock(); |
d1b19dff ED |
4651 | if (!ret) { |
4652 | if (colon) | |
4653 | *colon = ':'; | |
4654 | if (copy_to_user(arg, &ifr, | |
4655 | sizeof(struct ifreq))) | |
4656 | ret = -EFAULT; | |
4657 | } | |
4658 | return ret; | |
1da177e4 | 4659 | |
d1b19dff ED |
4660 | case SIOCETHTOOL: |
4661 | dev_load(net, ifr.ifr_name); | |
4662 | rtnl_lock(); | |
4663 | ret = dev_ethtool(net, &ifr); | |
4664 | rtnl_unlock(); | |
4665 | if (!ret) { | |
4666 | if (colon) | |
4667 | *colon = ':'; | |
4668 | if (copy_to_user(arg, &ifr, | |
4669 | sizeof(struct ifreq))) | |
4670 | ret = -EFAULT; | |
4671 | } | |
4672 | return ret; | |
1da177e4 | 4673 | |
d1b19dff ED |
4674 | /* |
4675 | * These ioctl calls: | |
4676 | * - require superuser power. | |
4677 | * - require strict serialization. | |
4678 | * - return a value | |
4679 | */ | |
4680 | case SIOCGMIIPHY: | |
4681 | case SIOCGMIIREG: | |
4682 | case SIOCSIFNAME: | |
4683 | if (!capable(CAP_NET_ADMIN)) | |
4684 | return -EPERM; | |
4685 | dev_load(net, ifr.ifr_name); | |
4686 | rtnl_lock(); | |
4687 | ret = dev_ifsioc(net, &ifr, cmd); | |
4688 | rtnl_unlock(); | |
4689 | if (!ret) { | |
4690 | if (colon) | |
4691 | *colon = ':'; | |
4692 | if (copy_to_user(arg, &ifr, | |
4693 | sizeof(struct ifreq))) | |
4694 | ret = -EFAULT; | |
4695 | } | |
4696 | return ret; | |
1da177e4 | 4697 | |
d1b19dff ED |
4698 | /* |
4699 | * These ioctl calls: | |
4700 | * - require superuser power. | |
4701 | * - require strict serialization. | |
4702 | * - do not return a value | |
4703 | */ | |
4704 | case SIOCSIFFLAGS: | |
4705 | case SIOCSIFMETRIC: | |
4706 | case SIOCSIFMTU: | |
4707 | case SIOCSIFMAP: | |
4708 | case SIOCSIFHWADDR: | |
4709 | case SIOCSIFSLAVE: | |
4710 | case SIOCADDMULTI: | |
4711 | case SIOCDELMULTI: | |
4712 | case SIOCSIFHWBROADCAST: | |
4713 | case SIOCSIFTXQLEN: | |
4714 | case SIOCSMIIREG: | |
4715 | case SIOCBONDENSLAVE: | |
4716 | case SIOCBONDRELEASE: | |
4717 | case SIOCBONDSETHWADDR: | |
4718 | case SIOCBONDCHANGEACTIVE: | |
4719 | case SIOCBRADDIF: | |
4720 | case SIOCBRDELIF: | |
4721 | case SIOCSHWTSTAMP: | |
4722 | if (!capable(CAP_NET_ADMIN)) | |
4723 | return -EPERM; | |
4724 | /* fall through */ | |
4725 | case SIOCBONDSLAVEINFOQUERY: | |
4726 | case SIOCBONDINFOQUERY: | |
4727 | dev_load(net, ifr.ifr_name); | |
4728 | rtnl_lock(); | |
4729 | ret = dev_ifsioc(net, &ifr, cmd); | |
4730 | rtnl_unlock(); | |
4731 | return ret; | |
4732 | ||
4733 | case SIOCGIFMEM: | |
4734 | /* Get the per device memory space. We can add this but | |
4735 | * currently do not support it */ | |
4736 | case SIOCSIFMEM: | |
4737 | /* Set the per device memory buffer space. | |
4738 | * Not applicable in our case */ | |
4739 | case SIOCSIFLINK: | |
4740 | return -EINVAL; | |
4741 | ||
4742 | /* | |
4743 | * Unknown or private ioctl. | |
4744 | */ | |
4745 | default: | |
4746 | if (cmd == SIOCWANDEV || | |
4747 | (cmd >= SIOCDEVPRIVATE && | |
4748 | cmd <= SIOCDEVPRIVATE + 15)) { | |
881d966b | 4749 | dev_load(net, ifr.ifr_name); |
1da177e4 | 4750 | rtnl_lock(); |
881d966b | 4751 | ret = dev_ifsioc(net, &ifr, cmd); |
1da177e4 | 4752 | rtnl_unlock(); |
d1b19dff ED |
4753 | if (!ret && copy_to_user(arg, &ifr, |
4754 | sizeof(struct ifreq))) | |
4755 | ret = -EFAULT; | |
1da177e4 | 4756 | return ret; |
d1b19dff ED |
4757 | } |
4758 | /* Take care of Wireless Extensions */ | |
4759 | if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) | |
4760 | return wext_handle_ioctl(net, &ifr, cmd, arg); | |
4761 | return -EINVAL; | |
1da177e4 LT |
4762 | } |
4763 | } | |
4764 | ||
4765 | ||
4766 | /** | |
4767 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 4768 | * @net: the applicable net namespace |
1da177e4 LT |
4769 | * |
4770 | * Returns a suitable unique value for a new device interface | |
4771 | * number. The caller must hold the rtnl semaphore or the | |
4772 | * dev_base_lock to be sure it remains unique. | |
4773 | */ | |
881d966b | 4774 | static int dev_new_index(struct net *net) |
1da177e4 LT |
4775 | { |
4776 | static int ifindex; | |
4777 | for (;;) { | |
4778 | if (++ifindex <= 0) | |
4779 | ifindex = 1; | |
881d966b | 4780 | if (!__dev_get_by_index(net, ifindex)) |
1da177e4 LT |
4781 | return ifindex; |
4782 | } | |
4783 | } | |
4784 | ||
1da177e4 | 4785 | /* Delayed registration/unregisteration */ |
3b5b34fd | 4786 | static LIST_HEAD(net_todo_list); |
1da177e4 | 4787 | |
6f05f629 | 4788 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 4789 | { |
1da177e4 | 4790 | list_add_tail(&dev->todo_list, &net_todo_list); |
1da177e4 LT |
4791 | } |
4792 | ||
9b5e383c | 4793 | static void rollback_registered_many(struct list_head *head) |
93ee31f1 | 4794 | { |
e93737b0 | 4795 | struct net_device *dev, *tmp; |
9b5e383c | 4796 | |
93ee31f1 DL |
4797 | BUG_ON(dev_boot_phase); |
4798 | ASSERT_RTNL(); | |
4799 | ||
e93737b0 | 4800 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
9b5e383c | 4801 | /* Some devices call without registering |
e93737b0 KK |
4802 | * for initialization unwind. Remove those |
4803 | * devices and proceed with the remaining. | |
9b5e383c ED |
4804 | */ |
4805 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
4806 | pr_debug("unregister_netdevice: device %s/%p never " | |
4807 | "was registered\n", dev->name, dev); | |
93ee31f1 | 4808 | |
9b5e383c | 4809 | WARN_ON(1); |
e93737b0 KK |
4810 | list_del(&dev->unreg_list); |
4811 | continue; | |
9b5e383c | 4812 | } |
93ee31f1 | 4813 | |
9b5e383c | 4814 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
93ee31f1 | 4815 | |
9b5e383c ED |
4816 | /* If device is running, close it first. */ |
4817 | dev_close(dev); | |
93ee31f1 | 4818 | |
9b5e383c ED |
4819 | /* And unlink it from device chain. */ |
4820 | unlist_netdevice(dev); | |
93ee31f1 | 4821 | |
9b5e383c ED |
4822 | dev->reg_state = NETREG_UNREGISTERING; |
4823 | } | |
93ee31f1 DL |
4824 | |
4825 | synchronize_net(); | |
4826 | ||
9b5e383c ED |
4827 | list_for_each_entry(dev, head, unreg_list) { |
4828 | /* Shutdown queueing discipline. */ | |
4829 | dev_shutdown(dev); | |
93ee31f1 DL |
4830 | |
4831 | ||
9b5e383c ED |
4832 | /* Notify protocols, that we are about to destroy |
4833 | this device. They should clean all the things. | |
4834 | */ | |
4835 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
93ee31f1 | 4836 | |
9b5e383c ED |
4837 | /* |
4838 | * Flush the unicast and multicast chains | |
4839 | */ | |
4840 | dev_unicast_flush(dev); | |
4841 | dev_addr_discard(dev); | |
93ee31f1 | 4842 | |
9b5e383c ED |
4843 | if (dev->netdev_ops->ndo_uninit) |
4844 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 | 4845 | |
9b5e383c ED |
4846 | /* Notifier chain MUST detach us from master device. */ |
4847 | WARN_ON(dev->master); | |
93ee31f1 | 4848 | |
9b5e383c ED |
4849 | /* Remove entries from kobject tree */ |
4850 | netdev_unregister_kobject(dev); | |
4851 | } | |
93ee31f1 | 4852 | |
a5ee1551 EB |
4853 | /* Process any work delayed until the end of the batch */ |
4854 | dev = list_entry(head->next, struct net_device, unreg_list); | |
4855 | call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev); | |
93ee31f1 | 4856 | |
a5ee1551 | 4857 | synchronize_net(); |
395264d5 | 4858 | |
a5ee1551 | 4859 | list_for_each_entry(dev, head, unreg_list) |
9b5e383c ED |
4860 | dev_put(dev); |
4861 | } | |
4862 | ||
4863 | static void rollback_registered(struct net_device *dev) | |
4864 | { | |
4865 | LIST_HEAD(single); | |
4866 | ||
4867 | list_add(&dev->unreg_list, &single); | |
4868 | rollback_registered_many(&single); | |
93ee31f1 DL |
4869 | } |
4870 | ||
e8a0464c DM |
4871 | static void __netdev_init_queue_locks_one(struct net_device *dev, |
4872 | struct netdev_queue *dev_queue, | |
4873 | void *_unused) | |
c773e847 DM |
4874 | { |
4875 | spin_lock_init(&dev_queue->_xmit_lock); | |
cf508b12 | 4876 | netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type); |
c773e847 DM |
4877 | dev_queue->xmit_lock_owner = -1; |
4878 | } | |
4879 | ||
4880 | static void netdev_init_queue_locks(struct net_device *dev) | |
4881 | { | |
e8a0464c DM |
4882 | netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL); |
4883 | __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL); | |
c773e847 DM |
4884 | } |
4885 | ||
b63365a2 HX |
4886 | unsigned long netdev_fix_features(unsigned long features, const char *name) |
4887 | { | |
4888 | /* Fix illegal SG+CSUM combinations. */ | |
4889 | if ((features & NETIF_F_SG) && | |
4890 | !(features & NETIF_F_ALL_CSUM)) { | |
4891 | if (name) | |
4892 | printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no " | |
4893 | "checksum feature.\n", name); | |
4894 | features &= ~NETIF_F_SG; | |
4895 | } | |
4896 | ||
4897 | /* TSO requires that SG is present as well. */ | |
4898 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) { | |
4899 | if (name) | |
4900 | printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no " | |
4901 | "SG feature.\n", name); | |
4902 | features &= ~NETIF_F_TSO; | |
4903 | } | |
4904 | ||
4905 | if (features & NETIF_F_UFO) { | |
4906 | if (!(features & NETIF_F_GEN_CSUM)) { | |
4907 | if (name) | |
4908 | printk(KERN_ERR "%s: Dropping NETIF_F_UFO " | |
4909 | "since no NETIF_F_HW_CSUM feature.\n", | |
4910 | name); | |
4911 | features &= ~NETIF_F_UFO; | |
4912 | } | |
4913 | ||
4914 | if (!(features & NETIF_F_SG)) { | |
4915 | if (name) | |
4916 | printk(KERN_ERR "%s: Dropping NETIF_F_UFO " | |
4917 | "since no NETIF_F_SG feature.\n", name); | |
4918 | features &= ~NETIF_F_UFO; | |
4919 | } | |
4920 | } | |
4921 | ||
4922 | return features; | |
4923 | } | |
4924 | EXPORT_SYMBOL(netdev_fix_features); | |
4925 | ||
fc4a7489 PM |
4926 | /** |
4927 | * netif_stacked_transfer_operstate - transfer operstate | |
4928 | * @rootdev: the root or lower level device to transfer state from | |
4929 | * @dev: the device to transfer operstate to | |
4930 | * | |
4931 | * Transfer operational state from root to device. This is normally | |
4932 | * called when a stacking relationship exists between the root | |
4933 | * device and the device(a leaf device). | |
4934 | */ | |
4935 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
4936 | struct net_device *dev) | |
4937 | { | |
4938 | if (rootdev->operstate == IF_OPER_DORMANT) | |
4939 | netif_dormant_on(dev); | |
4940 | else | |
4941 | netif_dormant_off(dev); | |
4942 | ||
4943 | if (netif_carrier_ok(rootdev)) { | |
4944 | if (!netif_carrier_ok(dev)) | |
4945 | netif_carrier_on(dev); | |
4946 | } else { | |
4947 | if (netif_carrier_ok(dev)) | |
4948 | netif_carrier_off(dev); | |
4949 | } | |
4950 | } | |
4951 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
4952 | ||
1da177e4 LT |
4953 | /** |
4954 | * register_netdevice - register a network device | |
4955 | * @dev: device to register | |
4956 | * | |
4957 | * Take a completed network device structure and add it to the kernel | |
4958 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
4959 | * chain. 0 is returned on success. A negative errno code is returned | |
4960 | * on a failure to set up the device, or if the name is a duplicate. | |
4961 | * | |
4962 | * Callers must hold the rtnl semaphore. You may want | |
4963 | * register_netdev() instead of this. | |
4964 | * | |
4965 | * BUGS: | |
4966 | * The locking appears insufficient to guarantee two parallel registers | |
4967 | * will not get the same name. | |
4968 | */ | |
4969 | ||
4970 | int register_netdevice(struct net_device *dev) | |
4971 | { | |
1da177e4 | 4972 | int ret; |
d314774c | 4973 | struct net *net = dev_net(dev); |
1da177e4 LT |
4974 | |
4975 | BUG_ON(dev_boot_phase); | |
4976 | ASSERT_RTNL(); | |
4977 | ||
b17a7c17 SH |
4978 | might_sleep(); |
4979 | ||
1da177e4 LT |
4980 | /* When net_device's are persistent, this will be fatal. */ |
4981 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 4982 | BUG_ON(!net); |
1da177e4 | 4983 | |
f1f28aa3 | 4984 | spin_lock_init(&dev->addr_list_lock); |
cf508b12 | 4985 | netdev_set_addr_lockdep_class(dev); |
c773e847 | 4986 | netdev_init_queue_locks(dev); |
1da177e4 | 4987 | |
1da177e4 LT |
4988 | dev->iflink = -1; |
4989 | ||
4990 | /* Init, if this function is available */ | |
d314774c SH |
4991 | if (dev->netdev_ops->ndo_init) { |
4992 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
4993 | if (ret) { |
4994 | if (ret > 0) | |
4995 | ret = -EIO; | |
90833aa4 | 4996 | goto out; |
1da177e4 LT |
4997 | } |
4998 | } | |
4ec93edb | 4999 | |
d9031024 OP |
5000 | ret = dev_get_valid_name(net, dev->name, dev->name, 0); |
5001 | if (ret) | |
7ce1b0ed | 5002 | goto err_uninit; |
1da177e4 | 5003 | |
881d966b | 5004 | dev->ifindex = dev_new_index(net); |
1da177e4 LT |
5005 | if (dev->iflink == -1) |
5006 | dev->iflink = dev->ifindex; | |
5007 | ||
d212f87b SH |
5008 | /* Fix illegal checksum combinations */ |
5009 | if ((dev->features & NETIF_F_HW_CSUM) && | |
5010 | (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
5011 | printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n", | |
5012 | dev->name); | |
5013 | dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); | |
5014 | } | |
5015 | ||
5016 | if ((dev->features & NETIF_F_NO_CSUM) && | |
5017 | (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
5018 | printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n", | |
5019 | dev->name); | |
5020 | dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM); | |
5021 | } | |
5022 | ||
b63365a2 | 5023 | dev->features = netdev_fix_features(dev->features, dev->name); |
1da177e4 | 5024 | |
e5a4a72d LB |
5025 | /* Enable software GSO if SG is supported. */ |
5026 | if (dev->features & NETIF_F_SG) | |
5027 | dev->features |= NETIF_F_GSO; | |
5028 | ||
aaf8cdc3 | 5029 | netdev_initialize_kobject(dev); |
7ffbe3fd JB |
5030 | |
5031 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); | |
5032 | ret = notifier_to_errno(ret); | |
5033 | if (ret) | |
5034 | goto err_uninit; | |
5035 | ||
8b41d188 | 5036 | ret = netdev_register_kobject(dev); |
b17a7c17 | 5037 | if (ret) |
7ce1b0ed | 5038 | goto err_uninit; |
b17a7c17 SH |
5039 | dev->reg_state = NETREG_REGISTERED; |
5040 | ||
1da177e4 LT |
5041 | /* |
5042 | * Default initial state at registry is that the | |
5043 | * device is present. | |
5044 | */ | |
5045 | ||
5046 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
5047 | ||
1da177e4 | 5048 | dev_init_scheduler(dev); |
1da177e4 | 5049 | dev_hold(dev); |
ce286d32 | 5050 | list_netdevice(dev); |
1da177e4 LT |
5051 | |
5052 | /* Notify protocols, that a new device appeared. */ | |
056925ab | 5053 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 5054 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
5055 | if (ret) { |
5056 | rollback_registered(dev); | |
5057 | dev->reg_state = NETREG_UNREGISTERED; | |
5058 | } | |
d90a909e EB |
5059 | /* |
5060 | * Prevent userspace races by waiting until the network | |
5061 | * device is fully setup before sending notifications. | |
5062 | */ | |
5063 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U); | |
1da177e4 LT |
5064 | |
5065 | out: | |
5066 | return ret; | |
7ce1b0ed HX |
5067 | |
5068 | err_uninit: | |
d314774c SH |
5069 | if (dev->netdev_ops->ndo_uninit) |
5070 | dev->netdev_ops->ndo_uninit(dev); | |
7ce1b0ed | 5071 | goto out; |
1da177e4 | 5072 | } |
d1b19dff | 5073 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 5074 | |
937f1ba5 BH |
5075 | /** |
5076 | * init_dummy_netdev - init a dummy network device for NAPI | |
5077 | * @dev: device to init | |
5078 | * | |
5079 | * This takes a network device structure and initialize the minimum | |
5080 | * amount of fields so it can be used to schedule NAPI polls without | |
5081 | * registering a full blown interface. This is to be used by drivers | |
5082 | * that need to tie several hardware interfaces to a single NAPI | |
5083 | * poll scheduler due to HW limitations. | |
5084 | */ | |
5085 | int init_dummy_netdev(struct net_device *dev) | |
5086 | { | |
5087 | /* Clear everything. Note we don't initialize spinlocks | |
5088 | * are they aren't supposed to be taken by any of the | |
5089 | * NAPI code and this dummy netdev is supposed to be | |
5090 | * only ever used for NAPI polls | |
5091 | */ | |
5092 | memset(dev, 0, sizeof(struct net_device)); | |
5093 | ||
5094 | /* make sure we BUG if trying to hit standard | |
5095 | * register/unregister code path | |
5096 | */ | |
5097 | dev->reg_state = NETREG_DUMMY; | |
5098 | ||
5099 | /* initialize the ref count */ | |
5100 | atomic_set(&dev->refcnt, 1); | |
5101 | ||
5102 | /* NAPI wants this */ | |
5103 | INIT_LIST_HEAD(&dev->napi_list); | |
5104 | ||
5105 | /* a dummy interface is started by default */ | |
5106 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
5107 | set_bit(__LINK_STATE_START, &dev->state); | |
5108 | ||
5109 | return 0; | |
5110 | } | |
5111 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
5112 | ||
5113 | ||
1da177e4 LT |
5114 | /** |
5115 | * register_netdev - register a network device | |
5116 | * @dev: device to register | |
5117 | * | |
5118 | * Take a completed network device structure and add it to the kernel | |
5119 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
5120 | * chain. 0 is returned on success. A negative errno code is returned | |
5121 | * on a failure to set up the device, or if the name is a duplicate. | |
5122 | * | |
38b4da38 | 5123 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
5124 | * and expands the device name if you passed a format string to |
5125 | * alloc_netdev. | |
5126 | */ | |
5127 | int register_netdev(struct net_device *dev) | |
5128 | { | |
5129 | int err; | |
5130 | ||
5131 | rtnl_lock(); | |
5132 | ||
5133 | /* | |
5134 | * If the name is a format string the caller wants us to do a | |
5135 | * name allocation. | |
5136 | */ | |
5137 | if (strchr(dev->name, '%')) { | |
5138 | err = dev_alloc_name(dev, dev->name); | |
5139 | if (err < 0) | |
5140 | goto out; | |
5141 | } | |
4ec93edb | 5142 | |
1da177e4 LT |
5143 | err = register_netdevice(dev); |
5144 | out: | |
5145 | rtnl_unlock(); | |
5146 | return err; | |
5147 | } | |
5148 | EXPORT_SYMBOL(register_netdev); | |
5149 | ||
5150 | /* | |
5151 | * netdev_wait_allrefs - wait until all references are gone. | |
5152 | * | |
5153 | * This is called when unregistering network devices. | |
5154 | * | |
5155 | * Any protocol or device that holds a reference should register | |
5156 | * for netdevice notification, and cleanup and put back the | |
5157 | * reference if they receive an UNREGISTER event. | |
5158 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 5159 | * call dev_put. |
1da177e4 LT |
5160 | */ |
5161 | static void netdev_wait_allrefs(struct net_device *dev) | |
5162 | { | |
5163 | unsigned long rebroadcast_time, warning_time; | |
5164 | ||
e014debe ED |
5165 | linkwatch_forget_dev(dev); |
5166 | ||
1da177e4 LT |
5167 | rebroadcast_time = warning_time = jiffies; |
5168 | while (atomic_read(&dev->refcnt) != 0) { | |
5169 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { | |
6756ae4b | 5170 | rtnl_lock(); |
1da177e4 LT |
5171 | |
5172 | /* Rebroadcast unregister notification */ | |
056925ab | 5173 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
a5ee1551 | 5174 | /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users |
395264d5 | 5175 | * should have already handle it the first time */ |
1da177e4 LT |
5176 | |
5177 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, | |
5178 | &dev->state)) { | |
5179 | /* We must not have linkwatch events | |
5180 | * pending on unregister. If this | |
5181 | * happens, we simply run the queue | |
5182 | * unscheduled, resulting in a noop | |
5183 | * for this device. | |
5184 | */ | |
5185 | linkwatch_run_queue(); | |
5186 | } | |
5187 | ||
6756ae4b | 5188 | __rtnl_unlock(); |
1da177e4 LT |
5189 | |
5190 | rebroadcast_time = jiffies; | |
5191 | } | |
5192 | ||
5193 | msleep(250); | |
5194 | ||
5195 | if (time_after(jiffies, warning_time + 10 * HZ)) { | |
5196 | printk(KERN_EMERG "unregister_netdevice: " | |
5197 | "waiting for %s to become free. Usage " | |
5198 | "count = %d\n", | |
5199 | dev->name, atomic_read(&dev->refcnt)); | |
5200 | warning_time = jiffies; | |
5201 | } | |
5202 | } | |
5203 | } | |
5204 | ||
5205 | /* The sequence is: | |
5206 | * | |
5207 | * rtnl_lock(); | |
5208 | * ... | |
5209 | * register_netdevice(x1); | |
5210 | * register_netdevice(x2); | |
5211 | * ... | |
5212 | * unregister_netdevice(y1); | |
5213 | * unregister_netdevice(y2); | |
5214 | * ... | |
5215 | * rtnl_unlock(); | |
5216 | * free_netdev(y1); | |
5217 | * free_netdev(y2); | |
5218 | * | |
58ec3b4d | 5219 | * We are invoked by rtnl_unlock(). |
1da177e4 | 5220 | * This allows us to deal with problems: |
b17a7c17 | 5221 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
5222 | * without deadlocking with linkwatch via keventd. |
5223 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
5224 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
5225 | * |
5226 | * We must not return until all unregister events added during | |
5227 | * the interval the lock was held have been completed. | |
1da177e4 | 5228 | */ |
1da177e4 LT |
5229 | void netdev_run_todo(void) |
5230 | { | |
626ab0e6 | 5231 | struct list_head list; |
1da177e4 | 5232 | |
1da177e4 | 5233 | /* Snapshot list, allow later requests */ |
626ab0e6 | 5234 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
5235 | |
5236 | __rtnl_unlock(); | |
626ab0e6 | 5237 | |
1da177e4 LT |
5238 | while (!list_empty(&list)) { |
5239 | struct net_device *dev | |
5240 | = list_entry(list.next, struct net_device, todo_list); | |
5241 | list_del(&dev->todo_list); | |
5242 | ||
b17a7c17 SH |
5243 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
5244 | printk(KERN_ERR "network todo '%s' but state %d\n", | |
5245 | dev->name, dev->reg_state); | |
5246 | dump_stack(); | |
5247 | continue; | |
5248 | } | |
1da177e4 | 5249 | |
b17a7c17 | 5250 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 5251 | |
6e583ce5 SH |
5252 | on_each_cpu(flush_backlog, dev, 1); |
5253 | ||
b17a7c17 | 5254 | netdev_wait_allrefs(dev); |
1da177e4 | 5255 | |
b17a7c17 SH |
5256 | /* paranoia */ |
5257 | BUG_ON(atomic_read(&dev->refcnt)); | |
547b792c IJ |
5258 | WARN_ON(dev->ip_ptr); |
5259 | WARN_ON(dev->ip6_ptr); | |
5260 | WARN_ON(dev->dn_ptr); | |
1da177e4 | 5261 | |
b17a7c17 SH |
5262 | if (dev->destructor) |
5263 | dev->destructor(dev); | |
9093bbb2 SH |
5264 | |
5265 | /* Free network device */ | |
5266 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 5267 | } |
1da177e4 LT |
5268 | } |
5269 | ||
d83345ad ED |
5270 | /** |
5271 | * dev_txq_stats_fold - fold tx_queues stats | |
5272 | * @dev: device to get statistics from | |
5273 | * @stats: struct net_device_stats to hold results | |
5274 | */ | |
5275 | void dev_txq_stats_fold(const struct net_device *dev, | |
5276 | struct net_device_stats *stats) | |
5277 | { | |
5278 | unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0; | |
5279 | unsigned int i; | |
5280 | struct netdev_queue *txq; | |
5281 | ||
5282 | for (i = 0; i < dev->num_tx_queues; i++) { | |
5283 | txq = netdev_get_tx_queue(dev, i); | |
5284 | tx_bytes += txq->tx_bytes; | |
5285 | tx_packets += txq->tx_packets; | |
5286 | tx_dropped += txq->tx_dropped; | |
5287 | } | |
5288 | if (tx_bytes || tx_packets || tx_dropped) { | |
5289 | stats->tx_bytes = tx_bytes; | |
5290 | stats->tx_packets = tx_packets; | |
5291 | stats->tx_dropped = tx_dropped; | |
5292 | } | |
5293 | } | |
5294 | EXPORT_SYMBOL(dev_txq_stats_fold); | |
5295 | ||
eeda3fd6 SH |
5296 | /** |
5297 | * dev_get_stats - get network device statistics | |
5298 | * @dev: device to get statistics from | |
5299 | * | |
5300 | * Get network statistics from device. The device driver may provide | |
5301 | * its own method by setting dev->netdev_ops->get_stats; otherwise | |
5302 | * the internal statistics structure is used. | |
5303 | */ | |
5304 | const struct net_device_stats *dev_get_stats(struct net_device *dev) | |
7004bf25 | 5305 | { |
eeda3fd6 SH |
5306 | const struct net_device_ops *ops = dev->netdev_ops; |
5307 | ||
5308 | if (ops->ndo_get_stats) | |
5309 | return ops->ndo_get_stats(dev); | |
d83345ad ED |
5310 | |
5311 | dev_txq_stats_fold(dev, &dev->stats); | |
5312 | return &dev->stats; | |
c45d286e | 5313 | } |
eeda3fd6 | 5314 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 5315 | |
dc2b4847 | 5316 | static void netdev_init_one_queue(struct net_device *dev, |
e8a0464c DM |
5317 | struct netdev_queue *queue, |
5318 | void *_unused) | |
dc2b4847 | 5319 | { |
dc2b4847 DM |
5320 | queue->dev = dev; |
5321 | } | |
5322 | ||
bb949fbd DM |
5323 | static void netdev_init_queues(struct net_device *dev) |
5324 | { | |
e8a0464c DM |
5325 | netdev_init_one_queue(dev, &dev->rx_queue, NULL); |
5326 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); | |
c3f26a26 | 5327 | spin_lock_init(&dev->tx_global_lock); |
bb949fbd DM |
5328 | } |
5329 | ||
1da177e4 | 5330 | /** |
f25f4e44 | 5331 | * alloc_netdev_mq - allocate network device |
1da177e4 LT |
5332 | * @sizeof_priv: size of private data to allocate space for |
5333 | * @name: device name format string | |
5334 | * @setup: callback to initialize device | |
f25f4e44 | 5335 | * @queue_count: the number of subqueues to allocate |
1da177e4 LT |
5336 | * |
5337 | * Allocates a struct net_device with private data area for driver use | |
f25f4e44 PWJ |
5338 | * and performs basic initialization. Also allocates subquue structs |
5339 | * for each queue on the device at the end of the netdevice. | |
1da177e4 | 5340 | */ |
f25f4e44 PWJ |
5341 | struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name, |
5342 | void (*setup)(struct net_device *), unsigned int queue_count) | |
1da177e4 | 5343 | { |
e8a0464c | 5344 | struct netdev_queue *tx; |
1da177e4 | 5345 | struct net_device *dev; |
7943986c | 5346 | size_t alloc_size; |
1ce8e7b5 | 5347 | struct net_device *p; |
1da177e4 | 5348 | |
b6fe17d6 SH |
5349 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
5350 | ||
fd2ea0a7 | 5351 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
5352 | if (sizeof_priv) { |
5353 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 5354 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
5355 | alloc_size += sizeof_priv; |
5356 | } | |
5357 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 5358 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 5359 | |
31380de9 | 5360 | p = kzalloc(alloc_size, GFP_KERNEL); |
1da177e4 | 5361 | if (!p) { |
b6fe17d6 | 5362 | printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n"); |
1da177e4 LT |
5363 | return NULL; |
5364 | } | |
1da177e4 | 5365 | |
7943986c | 5366 | tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL); |
e8a0464c DM |
5367 | if (!tx) { |
5368 | printk(KERN_ERR "alloc_netdev: Unable to allocate " | |
5369 | "tx qdiscs.\n"); | |
ab9c73cc | 5370 | goto free_p; |
e8a0464c DM |
5371 | } |
5372 | ||
1ce8e7b5 | 5373 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 5374 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc JP |
5375 | |
5376 | if (dev_addr_init(dev)) | |
5377 | goto free_tx; | |
5378 | ||
ccffad25 JP |
5379 | dev_unicast_init(dev); |
5380 | ||
c346dca1 | 5381 | dev_net_set(dev, &init_net); |
1da177e4 | 5382 | |
e8a0464c DM |
5383 | dev->_tx = tx; |
5384 | dev->num_tx_queues = queue_count; | |
fd2ea0a7 | 5385 | dev->real_num_tx_queues = queue_count; |
e8a0464c | 5386 | |
82cc1a7a | 5387 | dev->gso_max_size = GSO_MAX_SIZE; |
1da177e4 | 5388 | |
bb949fbd DM |
5389 | netdev_init_queues(dev); |
5390 | ||
d565b0a1 | 5391 | INIT_LIST_HEAD(&dev->napi_list); |
9fdce099 | 5392 | INIT_LIST_HEAD(&dev->unreg_list); |
e014debe | 5393 | INIT_LIST_HEAD(&dev->link_watch_list); |
93f154b5 | 5394 | dev->priv_flags = IFF_XMIT_DST_RELEASE; |
1da177e4 LT |
5395 | setup(dev); |
5396 | strcpy(dev->name, name); | |
5397 | return dev; | |
ab9c73cc JP |
5398 | |
5399 | free_tx: | |
5400 | kfree(tx); | |
5401 | ||
5402 | free_p: | |
5403 | kfree(p); | |
5404 | return NULL; | |
1da177e4 | 5405 | } |
f25f4e44 | 5406 | EXPORT_SYMBOL(alloc_netdev_mq); |
1da177e4 LT |
5407 | |
5408 | /** | |
5409 | * free_netdev - free network device | |
5410 | * @dev: device | |
5411 | * | |
4ec93edb YH |
5412 | * This function does the last stage of destroying an allocated device |
5413 | * interface. The reference to the device object is released. | |
1da177e4 LT |
5414 | * If this is the last reference then it will be freed. |
5415 | */ | |
5416 | void free_netdev(struct net_device *dev) | |
5417 | { | |
d565b0a1 HX |
5418 | struct napi_struct *p, *n; |
5419 | ||
f3005d7f DL |
5420 | release_net(dev_net(dev)); |
5421 | ||
e8a0464c DM |
5422 | kfree(dev->_tx); |
5423 | ||
f001fde5 JP |
5424 | /* Flush device addresses */ |
5425 | dev_addr_flush(dev); | |
5426 | ||
d565b0a1 HX |
5427 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
5428 | netif_napi_del(p); | |
5429 | ||
3041a069 | 5430 | /* Compatibility with error handling in drivers */ |
1da177e4 LT |
5431 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
5432 | kfree((char *)dev - dev->padded); | |
5433 | return; | |
5434 | } | |
5435 | ||
5436 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
5437 | dev->reg_state = NETREG_RELEASED; | |
5438 | ||
43cb76d9 GKH |
5439 | /* will free via device release */ |
5440 | put_device(&dev->dev); | |
1da177e4 | 5441 | } |
d1b19dff | 5442 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 5443 | |
f0db275a SH |
5444 | /** |
5445 | * synchronize_net - Synchronize with packet receive processing | |
5446 | * | |
5447 | * Wait for packets currently being received to be done. | |
5448 | * Does not block later packets from starting. | |
5449 | */ | |
4ec93edb | 5450 | void synchronize_net(void) |
1da177e4 LT |
5451 | { |
5452 | might_sleep(); | |
fbd568a3 | 5453 | synchronize_rcu(); |
1da177e4 | 5454 | } |
d1b19dff | 5455 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
5456 | |
5457 | /** | |
44a0873d | 5458 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 5459 | * @dev: device |
44a0873d | 5460 | * @head: list |
6ebfbc06 | 5461 | * |
1da177e4 | 5462 | * This function shuts down a device interface and removes it |
d59b54b1 | 5463 | * from the kernel tables. |
44a0873d | 5464 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
5465 | * |
5466 | * Callers must hold the rtnl semaphore. You may want | |
5467 | * unregister_netdev() instead of this. | |
5468 | */ | |
5469 | ||
44a0873d | 5470 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 5471 | { |
a6620712 HX |
5472 | ASSERT_RTNL(); |
5473 | ||
44a0873d | 5474 | if (head) { |
9fdce099 | 5475 | list_move_tail(&dev->unreg_list, head); |
44a0873d ED |
5476 | } else { |
5477 | rollback_registered(dev); | |
5478 | /* Finish processing unregister after unlock */ | |
5479 | net_set_todo(dev); | |
5480 | } | |
1da177e4 | 5481 | } |
44a0873d | 5482 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 5483 | |
9b5e383c ED |
5484 | /** |
5485 | * unregister_netdevice_many - unregister many devices | |
5486 | * @head: list of devices | |
9b5e383c ED |
5487 | */ |
5488 | void unregister_netdevice_many(struct list_head *head) | |
5489 | { | |
5490 | struct net_device *dev; | |
5491 | ||
5492 | if (!list_empty(head)) { | |
5493 | rollback_registered_many(head); | |
5494 | list_for_each_entry(dev, head, unreg_list) | |
5495 | net_set_todo(dev); | |
5496 | } | |
5497 | } | |
63c8099d | 5498 | EXPORT_SYMBOL(unregister_netdevice_many); |
9b5e383c | 5499 | |
1da177e4 LT |
5500 | /** |
5501 | * unregister_netdev - remove device from the kernel | |
5502 | * @dev: device | |
5503 | * | |
5504 | * This function shuts down a device interface and removes it | |
d59b54b1 | 5505 | * from the kernel tables. |
1da177e4 LT |
5506 | * |
5507 | * This is just a wrapper for unregister_netdevice that takes | |
5508 | * the rtnl semaphore. In general you want to use this and not | |
5509 | * unregister_netdevice. | |
5510 | */ | |
5511 | void unregister_netdev(struct net_device *dev) | |
5512 | { | |
5513 | rtnl_lock(); | |
5514 | unregister_netdevice(dev); | |
5515 | rtnl_unlock(); | |
5516 | } | |
1da177e4 LT |
5517 | EXPORT_SYMBOL(unregister_netdev); |
5518 | ||
ce286d32 EB |
5519 | /** |
5520 | * dev_change_net_namespace - move device to different nethost namespace | |
5521 | * @dev: device | |
5522 | * @net: network namespace | |
5523 | * @pat: If not NULL name pattern to try if the current device name | |
5524 | * is already taken in the destination network namespace. | |
5525 | * | |
5526 | * This function shuts down a device interface and moves it | |
5527 | * to a new network namespace. On success 0 is returned, on | |
5528 | * a failure a netagive errno code is returned. | |
5529 | * | |
5530 | * Callers must hold the rtnl semaphore. | |
5531 | */ | |
5532 | ||
5533 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
5534 | { | |
ce286d32 EB |
5535 | int err; |
5536 | ||
5537 | ASSERT_RTNL(); | |
5538 | ||
5539 | /* Don't allow namespace local devices to be moved. */ | |
5540 | err = -EINVAL; | |
5541 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
5542 | goto out; | |
5543 | ||
3891845e EB |
5544 | #ifdef CONFIG_SYSFS |
5545 | /* Don't allow real devices to be moved when sysfs | |
5546 | * is enabled. | |
5547 | */ | |
5548 | err = -EINVAL; | |
5549 | if (dev->dev.parent) | |
5550 | goto out; | |
5551 | #endif | |
5552 | ||
ce286d32 EB |
5553 | /* Ensure the device has been registrered */ |
5554 | err = -EINVAL; | |
5555 | if (dev->reg_state != NETREG_REGISTERED) | |
5556 | goto out; | |
5557 | ||
5558 | /* Get out if there is nothing todo */ | |
5559 | err = 0; | |
878628fb | 5560 | if (net_eq(dev_net(dev), net)) |
ce286d32 EB |
5561 | goto out; |
5562 | ||
5563 | /* Pick the destination device name, and ensure | |
5564 | * we can use it in the destination network namespace. | |
5565 | */ | |
5566 | err = -EEXIST; | |
d9031024 | 5567 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
5568 | /* We get here if we can't use the current device name */ |
5569 | if (!pat) | |
5570 | goto out; | |
d9031024 | 5571 | if (dev_get_valid_name(net, pat, dev->name, 1)) |
ce286d32 EB |
5572 | goto out; |
5573 | } | |
5574 | ||
5575 | /* | |
5576 | * And now a mini version of register_netdevice unregister_netdevice. | |
5577 | */ | |
5578 | ||
5579 | /* If device is running close it first. */ | |
9b772652 | 5580 | dev_close(dev); |
ce286d32 EB |
5581 | |
5582 | /* And unlink it from device chain */ | |
5583 | err = -ENODEV; | |
5584 | unlist_netdevice(dev); | |
5585 | ||
5586 | synchronize_net(); | |
5587 | ||
5588 | /* Shutdown queueing discipline. */ | |
5589 | dev_shutdown(dev); | |
5590 | ||
5591 | /* Notify protocols, that we are about to destroy | |
5592 | this device. They should clean all the things. | |
5593 | */ | |
5594 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
a5ee1551 | 5595 | call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev); |
ce286d32 EB |
5596 | |
5597 | /* | |
5598 | * Flush the unicast and multicast chains | |
5599 | */ | |
ccffad25 | 5600 | dev_unicast_flush(dev); |
ce286d32 EB |
5601 | dev_addr_discard(dev); |
5602 | ||
3891845e EB |
5603 | netdev_unregister_kobject(dev); |
5604 | ||
ce286d32 | 5605 | /* Actually switch the network namespace */ |
c346dca1 | 5606 | dev_net_set(dev, net); |
ce286d32 | 5607 | |
ce286d32 EB |
5608 | /* If there is an ifindex conflict assign a new one */ |
5609 | if (__dev_get_by_index(net, dev->ifindex)) { | |
5610 | int iflink = (dev->iflink == dev->ifindex); | |
5611 | dev->ifindex = dev_new_index(net); | |
5612 | if (iflink) | |
5613 | dev->iflink = dev->ifindex; | |
5614 | } | |
5615 | ||
8b41d188 | 5616 | /* Fixup kobjects */ |
aaf8cdc3 | 5617 | err = netdev_register_kobject(dev); |
8b41d188 | 5618 | WARN_ON(err); |
ce286d32 EB |
5619 | |
5620 | /* Add the device back in the hashes */ | |
5621 | list_netdevice(dev); | |
5622 | ||
5623 | /* Notify protocols, that a new device appeared. */ | |
5624 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
5625 | ||
d90a909e EB |
5626 | /* |
5627 | * Prevent userspace races by waiting until the network | |
5628 | * device is fully setup before sending notifications. | |
5629 | */ | |
5630 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U); | |
5631 | ||
ce286d32 EB |
5632 | synchronize_net(); |
5633 | err = 0; | |
5634 | out: | |
5635 | return err; | |
5636 | } | |
463d0183 | 5637 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 5638 | |
1da177e4 LT |
5639 | static int dev_cpu_callback(struct notifier_block *nfb, |
5640 | unsigned long action, | |
5641 | void *ocpu) | |
5642 | { | |
5643 | struct sk_buff **list_skb; | |
37437bb2 | 5644 | struct Qdisc **list_net; |
1da177e4 LT |
5645 | struct sk_buff *skb; |
5646 | unsigned int cpu, oldcpu = (unsigned long)ocpu; | |
5647 | struct softnet_data *sd, *oldsd; | |
5648 | ||
8bb78442 | 5649 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
1da177e4 LT |
5650 | return NOTIFY_OK; |
5651 | ||
5652 | local_irq_disable(); | |
5653 | cpu = smp_processor_id(); | |
5654 | sd = &per_cpu(softnet_data, cpu); | |
5655 | oldsd = &per_cpu(softnet_data, oldcpu); | |
5656 | ||
5657 | /* Find end of our completion_queue. */ | |
5658 | list_skb = &sd->completion_queue; | |
5659 | while (*list_skb) | |
5660 | list_skb = &(*list_skb)->next; | |
5661 | /* Append completion queue from offline CPU. */ | |
5662 | *list_skb = oldsd->completion_queue; | |
5663 | oldsd->completion_queue = NULL; | |
5664 | ||
5665 | /* Find end of our output_queue. */ | |
5666 | list_net = &sd->output_queue; | |
5667 | while (*list_net) | |
5668 | list_net = &(*list_net)->next_sched; | |
5669 | /* Append output queue from offline CPU. */ | |
5670 | *list_net = oldsd->output_queue; | |
5671 | oldsd->output_queue = NULL; | |
5672 | ||
5673 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
5674 | local_irq_enable(); | |
5675 | ||
5676 | /* Process offline CPU's input_pkt_queue */ | |
5677 | while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) | |
5678 | netif_rx(skb); | |
5679 | ||
5680 | return NOTIFY_OK; | |
5681 | } | |
1da177e4 LT |
5682 | |
5683 | ||
7f353bf2 | 5684 | /** |
b63365a2 HX |
5685 | * netdev_increment_features - increment feature set by one |
5686 | * @all: current feature set | |
5687 | * @one: new feature set | |
5688 | * @mask: mask feature set | |
7f353bf2 HX |
5689 | * |
5690 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
5691 | * @one to the master device with current feature set @all. Will not |
5692 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 5693 | */ |
b63365a2 HX |
5694 | unsigned long netdev_increment_features(unsigned long all, unsigned long one, |
5695 | unsigned long mask) | |
5696 | { | |
5697 | /* If device needs checksumming, downgrade to it. */ | |
d1b19dff | 5698 | if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM)) |
b63365a2 HX |
5699 | all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM); |
5700 | else if (mask & NETIF_F_ALL_CSUM) { | |
5701 | /* If one device supports v4/v6 checksumming, set for all. */ | |
5702 | if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) && | |
5703 | !(all & NETIF_F_GEN_CSUM)) { | |
5704 | all &= ~NETIF_F_ALL_CSUM; | |
5705 | all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); | |
5706 | } | |
e2a6b852 | 5707 | |
b63365a2 HX |
5708 | /* If one device supports hw checksumming, set for all. */ |
5709 | if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) { | |
5710 | all &= ~NETIF_F_ALL_CSUM; | |
5711 | all |= NETIF_F_HW_CSUM; | |
5712 | } | |
5713 | } | |
7f353bf2 | 5714 | |
b63365a2 | 5715 | one |= NETIF_F_ALL_CSUM; |
7f353bf2 | 5716 | |
b63365a2 | 5717 | one |= all & NETIF_F_ONE_FOR_ALL; |
d9f5950f | 5718 | all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO; |
b63365a2 | 5719 | all |= one & mask & NETIF_F_ONE_FOR_ALL; |
7f353bf2 HX |
5720 | |
5721 | return all; | |
5722 | } | |
b63365a2 | 5723 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 5724 | |
30d97d35 PE |
5725 | static struct hlist_head *netdev_create_hash(void) |
5726 | { | |
5727 | int i; | |
5728 | struct hlist_head *hash; | |
5729 | ||
5730 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
5731 | if (hash != NULL) | |
5732 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
5733 | INIT_HLIST_HEAD(&hash[i]); | |
5734 | ||
5735 | return hash; | |
5736 | } | |
5737 | ||
881d966b | 5738 | /* Initialize per network namespace state */ |
4665079c | 5739 | static int __net_init netdev_init(struct net *net) |
881d966b | 5740 | { |
881d966b | 5741 | INIT_LIST_HEAD(&net->dev_base_head); |
881d966b | 5742 | |
30d97d35 PE |
5743 | net->dev_name_head = netdev_create_hash(); |
5744 | if (net->dev_name_head == NULL) | |
5745 | goto err_name; | |
881d966b | 5746 | |
30d97d35 PE |
5747 | net->dev_index_head = netdev_create_hash(); |
5748 | if (net->dev_index_head == NULL) | |
5749 | goto err_idx; | |
881d966b EB |
5750 | |
5751 | return 0; | |
30d97d35 PE |
5752 | |
5753 | err_idx: | |
5754 | kfree(net->dev_name_head); | |
5755 | err_name: | |
5756 | return -ENOMEM; | |
881d966b EB |
5757 | } |
5758 | ||
f0db275a SH |
5759 | /** |
5760 | * netdev_drivername - network driver for the device | |
5761 | * @dev: network device | |
5762 | * @buffer: buffer for resulting name | |
5763 | * @len: size of buffer | |
5764 | * | |
5765 | * Determine network driver for device. | |
5766 | */ | |
cf04a4c7 | 5767 | char *netdev_drivername(const struct net_device *dev, char *buffer, int len) |
6579e57b | 5768 | { |
cf04a4c7 SH |
5769 | const struct device_driver *driver; |
5770 | const struct device *parent; | |
6579e57b AV |
5771 | |
5772 | if (len <= 0 || !buffer) | |
5773 | return buffer; | |
5774 | buffer[0] = 0; | |
5775 | ||
5776 | parent = dev->dev.parent; | |
5777 | ||
5778 | if (!parent) | |
5779 | return buffer; | |
5780 | ||
5781 | driver = parent->driver; | |
5782 | if (driver && driver->name) | |
5783 | strlcpy(buffer, driver->name, len); | |
5784 | return buffer; | |
5785 | } | |
5786 | ||
4665079c | 5787 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
5788 | { |
5789 | kfree(net->dev_name_head); | |
5790 | kfree(net->dev_index_head); | |
5791 | } | |
5792 | ||
022cbae6 | 5793 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
5794 | .init = netdev_init, |
5795 | .exit = netdev_exit, | |
5796 | }; | |
5797 | ||
4665079c | 5798 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 5799 | { |
e008b5fc | 5800 | struct net_device *dev, *aux; |
ce286d32 | 5801 | /* |
e008b5fc | 5802 | * Push all migratable network devices back to the |
ce286d32 EB |
5803 | * initial network namespace |
5804 | */ | |
5805 | rtnl_lock(); | |
e008b5fc | 5806 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 5807 | int err; |
aca51397 | 5808 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
5809 | |
5810 | /* Ignore unmoveable devices (i.e. loopback) */ | |
5811 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
5812 | continue; | |
5813 | ||
e008b5fc EB |
5814 | /* Leave virtual devices for the generic cleanup */ |
5815 | if (dev->rtnl_link_ops) | |
5816 | continue; | |
d0c082ce | 5817 | |
ce286d32 | 5818 | /* Push remaing network devices to init_net */ |
aca51397 PE |
5819 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
5820 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
ce286d32 | 5821 | if (err) { |
aca51397 | 5822 | printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n", |
ce286d32 | 5823 | __func__, dev->name, err); |
aca51397 | 5824 | BUG(); |
ce286d32 EB |
5825 | } |
5826 | } | |
5827 | rtnl_unlock(); | |
5828 | } | |
5829 | ||
04dc7f6b EB |
5830 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
5831 | { | |
5832 | /* At exit all network devices most be removed from a network | |
5833 | * namespace. Do this in the reverse order of registeration. | |
5834 | * Do this across as many network namespaces as possible to | |
5835 | * improve batching efficiency. | |
5836 | */ | |
5837 | struct net_device *dev; | |
5838 | struct net *net; | |
5839 | LIST_HEAD(dev_kill_list); | |
5840 | ||
5841 | rtnl_lock(); | |
5842 | list_for_each_entry(net, net_list, exit_list) { | |
5843 | for_each_netdev_reverse(net, dev) { | |
5844 | if (dev->rtnl_link_ops) | |
5845 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); | |
5846 | else | |
5847 | unregister_netdevice_queue(dev, &dev_kill_list); | |
5848 | } | |
5849 | } | |
5850 | unregister_netdevice_many(&dev_kill_list); | |
5851 | rtnl_unlock(); | |
5852 | } | |
5853 | ||
022cbae6 | 5854 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 5855 | .exit = default_device_exit, |
04dc7f6b | 5856 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
5857 | }; |
5858 | ||
1da177e4 LT |
5859 | /* |
5860 | * Initialize the DEV module. At boot time this walks the device list and | |
5861 | * unhooks any devices that fail to initialise (normally hardware not | |
5862 | * present) and leaves us with a valid list of present and active devices. | |
5863 | * | |
5864 | */ | |
5865 | ||
5866 | /* | |
5867 | * This is called single threaded during boot, so no need | |
5868 | * to take the rtnl semaphore. | |
5869 | */ | |
5870 | static int __init net_dev_init(void) | |
5871 | { | |
5872 | int i, rc = -ENOMEM; | |
5873 | ||
5874 | BUG_ON(!dev_boot_phase); | |
5875 | ||
1da177e4 LT |
5876 | if (dev_proc_init()) |
5877 | goto out; | |
5878 | ||
8b41d188 | 5879 | if (netdev_kobject_init()) |
1da177e4 LT |
5880 | goto out; |
5881 | ||
5882 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 5883 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
5884 | INIT_LIST_HEAD(&ptype_base[i]); |
5885 | ||
881d966b EB |
5886 | if (register_pernet_subsys(&netdev_net_ops)) |
5887 | goto out; | |
1da177e4 LT |
5888 | |
5889 | /* | |
5890 | * Initialise the packet receive queues. | |
5891 | */ | |
5892 | ||
6f912042 | 5893 | for_each_possible_cpu(i) { |
1da177e4 LT |
5894 | struct softnet_data *queue; |
5895 | ||
5896 | queue = &per_cpu(softnet_data, i); | |
5897 | skb_queue_head_init(&queue->input_pkt_queue); | |
1da177e4 LT |
5898 | queue->completion_queue = NULL; |
5899 | INIT_LIST_HEAD(&queue->poll_list); | |
bea3348e SH |
5900 | |
5901 | queue->backlog.poll = process_backlog; | |
5902 | queue->backlog.weight = weight_p; | |
d565b0a1 | 5903 | queue->backlog.gro_list = NULL; |
4ae5544f | 5904 | queue->backlog.gro_count = 0; |
1da177e4 LT |
5905 | } |
5906 | ||
1da177e4 LT |
5907 | dev_boot_phase = 0; |
5908 | ||
505d4f73 EB |
5909 | /* The loopback device is special if any other network devices |
5910 | * is present in a network namespace the loopback device must | |
5911 | * be present. Since we now dynamically allocate and free the | |
5912 | * loopback device ensure this invariant is maintained by | |
5913 | * keeping the loopback device as the first device on the | |
5914 | * list of network devices. Ensuring the loopback devices | |
5915 | * is the first device that appears and the last network device | |
5916 | * that disappears. | |
5917 | */ | |
5918 | if (register_pernet_device(&loopback_net_ops)) | |
5919 | goto out; | |
5920 | ||
5921 | if (register_pernet_device(&default_device_ops)) | |
5922 | goto out; | |
5923 | ||
962cf36c CM |
5924 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
5925 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 LT |
5926 | |
5927 | hotcpu_notifier(dev_cpu_callback, 0); | |
5928 | dst_init(); | |
5929 | dev_mcast_init(); | |
5930 | rc = 0; | |
5931 | out: | |
5932 | return rc; | |
5933 | } | |
5934 | ||
5935 | subsys_initcall(net_dev_init); | |
5936 | ||
e88721f8 KK |
5937 | static int __init initialize_hashrnd(void) |
5938 | { | |
5939 | get_random_bytes(&skb_tx_hashrnd, sizeof(skb_tx_hashrnd)); | |
5940 | return 0; | |
5941 | } | |
5942 | ||
5943 | late_initcall_sync(initialize_hashrnd); | |
5944 |