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 | 774 | { |
99fe3c39 | 775 | struct net_device *dev, *ret = NULL; |
4e9cac2b | 776 | |
99fe3c39 ED |
777 | rcu_read_lock(); |
778 | for_each_netdev_rcu(net, dev) | |
779 | if (dev->type == type) { | |
780 | dev_hold(dev); | |
781 | ret = dev; | |
782 | break; | |
783 | } | |
784 | rcu_read_unlock(); | |
785 | return ret; | |
1da177e4 | 786 | } |
1da177e4 LT |
787 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
788 | ||
789 | /** | |
790 | * dev_get_by_flags - find any device with given flags | |
c4ea43c5 | 791 | * @net: the applicable net namespace |
1da177e4 LT |
792 | * @if_flags: IFF_* values |
793 | * @mask: bitmask of bits in if_flags to check | |
794 | * | |
795 | * Search for any interface with the given flags. Returns NULL if a device | |
4ec93edb | 796 | * is not found or a pointer to the device. The device returned has |
1da177e4 LT |
797 | * had a reference added and the pointer is safe until the user calls |
798 | * dev_put to indicate they have finished with it. | |
799 | */ | |
800 | ||
d1b19dff ED |
801 | struct net_device *dev_get_by_flags(struct net *net, unsigned short if_flags, |
802 | unsigned short mask) | |
1da177e4 | 803 | { |
7562f876 | 804 | struct net_device *dev, *ret; |
1da177e4 | 805 | |
7562f876 | 806 | ret = NULL; |
c6d14c84 ED |
807 | rcu_read_lock(); |
808 | for_each_netdev_rcu(net, dev) { | |
1da177e4 LT |
809 | if (((dev->flags ^ if_flags) & mask) == 0) { |
810 | dev_hold(dev); | |
7562f876 | 811 | ret = dev; |
1da177e4 LT |
812 | break; |
813 | } | |
814 | } | |
c6d14c84 | 815 | rcu_read_unlock(); |
7562f876 | 816 | return ret; |
1da177e4 | 817 | } |
d1b19dff | 818 | EXPORT_SYMBOL(dev_get_by_flags); |
1da177e4 LT |
819 | |
820 | /** | |
821 | * dev_valid_name - check if name is okay for network device | |
822 | * @name: name string | |
823 | * | |
824 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
825 | * to allow sysfs to work. We also disallow any kind of |
826 | * whitespace. | |
1da177e4 | 827 | */ |
c2373ee9 | 828 | int dev_valid_name(const char *name) |
1da177e4 | 829 | { |
c7fa9d18 DM |
830 | if (*name == '\0') |
831 | return 0; | |
b6fe17d6 SH |
832 | if (strlen(name) >= IFNAMSIZ) |
833 | return 0; | |
c7fa9d18 DM |
834 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
835 | return 0; | |
836 | ||
837 | while (*name) { | |
838 | if (*name == '/' || isspace(*name)) | |
839 | return 0; | |
840 | name++; | |
841 | } | |
842 | return 1; | |
1da177e4 | 843 | } |
d1b19dff | 844 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
845 | |
846 | /** | |
b267b179 EB |
847 | * __dev_alloc_name - allocate a name for a device |
848 | * @net: network namespace to allocate the device name in | |
1da177e4 | 849 | * @name: name format string |
b267b179 | 850 | * @buf: scratch buffer and result name string |
1da177e4 LT |
851 | * |
852 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
853 | * id. It scans list of devices to build up a free map, then chooses |
854 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
855 | * while allocating the name and adding the device in order to avoid | |
856 | * duplicates. | |
857 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
858 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
859 | */ |
860 | ||
b267b179 | 861 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
862 | { |
863 | int i = 0; | |
1da177e4 LT |
864 | const char *p; |
865 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 866 | unsigned long *inuse; |
1da177e4 LT |
867 | struct net_device *d; |
868 | ||
869 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
870 | if (p) { | |
871 | /* | |
872 | * Verify the string as this thing may have come from | |
873 | * the user. There must be either one "%d" and no other "%" | |
874 | * characters. | |
875 | */ | |
876 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
877 | return -EINVAL; | |
878 | ||
879 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 880 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
881 | if (!inuse) |
882 | return -ENOMEM; | |
883 | ||
881d966b | 884 | for_each_netdev(net, d) { |
1da177e4 LT |
885 | if (!sscanf(d->name, name, &i)) |
886 | continue; | |
887 | if (i < 0 || i >= max_netdevices) | |
888 | continue; | |
889 | ||
890 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 891 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
892 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
893 | set_bit(i, inuse); | |
894 | } | |
895 | ||
896 | i = find_first_zero_bit(inuse, max_netdevices); | |
897 | free_page((unsigned long) inuse); | |
898 | } | |
899 | ||
d9031024 OP |
900 | if (buf != name) |
901 | snprintf(buf, IFNAMSIZ, name, i); | |
b267b179 | 902 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 903 | return i; |
1da177e4 LT |
904 | |
905 | /* It is possible to run out of possible slots | |
906 | * when the name is long and there isn't enough space left | |
907 | * for the digits, or if all bits are used. | |
908 | */ | |
909 | return -ENFILE; | |
910 | } | |
911 | ||
b267b179 EB |
912 | /** |
913 | * dev_alloc_name - allocate a name for a device | |
914 | * @dev: device | |
915 | * @name: name format string | |
916 | * | |
917 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
918 | * id. It scans list of devices to build up a free map, then chooses | |
919 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
920 | * while allocating the name and adding the device in order to avoid | |
921 | * duplicates. | |
922 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
923 | * Returns the number of the unit assigned or a negative errno code. | |
924 | */ | |
925 | ||
926 | int dev_alloc_name(struct net_device *dev, const char *name) | |
927 | { | |
928 | char buf[IFNAMSIZ]; | |
929 | struct net *net; | |
930 | int ret; | |
931 | ||
c346dca1 YH |
932 | BUG_ON(!dev_net(dev)); |
933 | net = dev_net(dev); | |
b267b179 EB |
934 | ret = __dev_alloc_name(net, name, buf); |
935 | if (ret >= 0) | |
936 | strlcpy(dev->name, buf, IFNAMSIZ); | |
937 | return ret; | |
938 | } | |
d1b19dff | 939 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 940 | |
d9031024 OP |
941 | static int dev_get_valid_name(struct net *net, const char *name, char *buf, |
942 | bool fmt) | |
943 | { | |
944 | if (!dev_valid_name(name)) | |
945 | return -EINVAL; | |
946 | ||
947 | if (fmt && strchr(name, '%')) | |
948 | return __dev_alloc_name(net, name, buf); | |
949 | else if (__dev_get_by_name(net, name)) | |
950 | return -EEXIST; | |
951 | else if (buf != name) | |
952 | strlcpy(buf, name, IFNAMSIZ); | |
953 | ||
954 | return 0; | |
955 | } | |
1da177e4 LT |
956 | |
957 | /** | |
958 | * dev_change_name - change name of a device | |
959 | * @dev: device | |
960 | * @newname: name (or format string) must be at least IFNAMSIZ | |
961 | * | |
962 | * Change name of a device, can pass format strings "eth%d". | |
963 | * for wildcarding. | |
964 | */ | |
cf04a4c7 | 965 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 966 | { |
fcc5a03a | 967 | char oldname[IFNAMSIZ]; |
1da177e4 | 968 | int err = 0; |
fcc5a03a | 969 | int ret; |
881d966b | 970 | struct net *net; |
1da177e4 LT |
971 | |
972 | ASSERT_RTNL(); | |
c346dca1 | 973 | BUG_ON(!dev_net(dev)); |
1da177e4 | 974 | |
c346dca1 | 975 | net = dev_net(dev); |
1da177e4 LT |
976 | if (dev->flags & IFF_UP) |
977 | return -EBUSY; | |
978 | ||
c8d90dca SH |
979 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) |
980 | return 0; | |
981 | ||
fcc5a03a HX |
982 | memcpy(oldname, dev->name, IFNAMSIZ); |
983 | ||
d9031024 OP |
984 | err = dev_get_valid_name(net, newname, dev->name, 1); |
985 | if (err < 0) | |
986 | return err; | |
1da177e4 | 987 | |
fcc5a03a | 988 | rollback: |
3891845e EB |
989 | /* For now only devices in the initial network namespace |
990 | * are in sysfs. | |
991 | */ | |
09ad9bc7 | 992 | if (net_eq(net, &init_net)) { |
3891845e EB |
993 | ret = device_rename(&dev->dev, dev->name); |
994 | if (ret) { | |
995 | memcpy(dev->name, oldname, IFNAMSIZ); | |
996 | return ret; | |
997 | } | |
dcc99773 | 998 | } |
7f988eab HX |
999 | |
1000 | write_lock_bh(&dev_base_lock); | |
92749821 | 1001 | hlist_del(&dev->name_hlist); |
72c9528b ED |
1002 | write_unlock_bh(&dev_base_lock); |
1003 | ||
1004 | synchronize_rcu(); | |
1005 | ||
1006 | write_lock_bh(&dev_base_lock); | |
1007 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
7f988eab HX |
1008 | write_unlock_bh(&dev_base_lock); |
1009 | ||
056925ab | 1010 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1011 | ret = notifier_to_errno(ret); |
1012 | ||
1013 | if (ret) { | |
91e9c07b ED |
1014 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1015 | if (err >= 0) { | |
fcc5a03a HX |
1016 | err = ret; |
1017 | memcpy(dev->name, oldname, IFNAMSIZ); | |
1018 | goto rollback; | |
91e9c07b ED |
1019 | } else { |
1020 | printk(KERN_ERR | |
1021 | "%s: name change rollback failed: %d.\n", | |
1022 | dev->name, ret); | |
fcc5a03a HX |
1023 | } |
1024 | } | |
1da177e4 LT |
1025 | |
1026 | return err; | |
1027 | } | |
1028 | ||
0b815a1a SH |
1029 | /** |
1030 | * dev_set_alias - change ifalias of a device | |
1031 | * @dev: device | |
1032 | * @alias: name up to IFALIASZ | |
f0db275a | 1033 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1034 | * |
1035 | * Set ifalias for a device, | |
1036 | */ | |
1037 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1038 | { | |
1039 | ASSERT_RTNL(); | |
1040 | ||
1041 | if (len >= IFALIASZ) | |
1042 | return -EINVAL; | |
1043 | ||
96ca4a2c OH |
1044 | if (!len) { |
1045 | if (dev->ifalias) { | |
1046 | kfree(dev->ifalias); | |
1047 | dev->ifalias = NULL; | |
1048 | } | |
1049 | return 0; | |
1050 | } | |
1051 | ||
d1b19dff | 1052 | dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL); |
0b815a1a SH |
1053 | if (!dev->ifalias) |
1054 | return -ENOMEM; | |
1055 | ||
1056 | strlcpy(dev->ifalias, alias, len+1); | |
1057 | return len; | |
1058 | } | |
1059 | ||
1060 | ||
d8a33ac4 | 1061 | /** |
3041a069 | 1062 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1063 | * @dev: device to cause notification |
1064 | * | |
1065 | * Called to indicate a device has changed features. | |
1066 | */ | |
1067 | void netdev_features_change(struct net_device *dev) | |
1068 | { | |
056925ab | 1069 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1070 | } |
1071 | EXPORT_SYMBOL(netdev_features_change); | |
1072 | ||
1da177e4 LT |
1073 | /** |
1074 | * netdev_state_change - device changes state | |
1075 | * @dev: device to cause notification | |
1076 | * | |
1077 | * Called to indicate a device has changed state. This function calls | |
1078 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1079 | * to the routing socket. | |
1080 | */ | |
1081 | void netdev_state_change(struct net_device *dev) | |
1082 | { | |
1083 | if (dev->flags & IFF_UP) { | |
056925ab | 1084 | call_netdevice_notifiers(NETDEV_CHANGE, dev); |
1da177e4 LT |
1085 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0); |
1086 | } | |
1087 | } | |
d1b19dff | 1088 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1089 | |
3ca5b404 | 1090 | int netdev_bonding_change(struct net_device *dev, unsigned long event) |
c1da4ac7 | 1091 | { |
3ca5b404 | 1092 | return call_netdevice_notifiers(event, dev); |
c1da4ac7 OG |
1093 | } |
1094 | EXPORT_SYMBOL(netdev_bonding_change); | |
1095 | ||
1da177e4 LT |
1096 | /** |
1097 | * dev_load - load a network module | |
c4ea43c5 | 1098 | * @net: the applicable net namespace |
1da177e4 LT |
1099 | * @name: name of interface |
1100 | * | |
1101 | * If a network interface is not present and the process has suitable | |
1102 | * privileges this function loads the module. If module loading is not | |
1103 | * available in this kernel then it becomes a nop. | |
1104 | */ | |
1105 | ||
881d966b | 1106 | void dev_load(struct net *net, const char *name) |
1da177e4 | 1107 | { |
4ec93edb | 1108 | struct net_device *dev; |
1da177e4 | 1109 | |
72c9528b ED |
1110 | rcu_read_lock(); |
1111 | dev = dev_get_by_name_rcu(net, name); | |
1112 | rcu_read_unlock(); | |
1da177e4 | 1113 | |
a8f80e8f | 1114 | if (!dev && capable(CAP_NET_ADMIN)) |
1da177e4 LT |
1115 | request_module("%s", name); |
1116 | } | |
d1b19dff | 1117 | EXPORT_SYMBOL(dev_load); |
1da177e4 | 1118 | |
bd380811 | 1119 | static int __dev_open(struct net_device *dev) |
1da177e4 | 1120 | { |
d314774c | 1121 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1122 | int ret; |
1da177e4 | 1123 | |
e46b66bc BH |
1124 | ASSERT_RTNL(); |
1125 | ||
1da177e4 LT |
1126 | /* |
1127 | * Is it even present? | |
1128 | */ | |
1129 | if (!netif_device_present(dev)) | |
1130 | return -ENODEV; | |
1131 | ||
3b8bcfd5 JB |
1132 | ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); |
1133 | ret = notifier_to_errno(ret); | |
1134 | if (ret) | |
1135 | return ret; | |
1136 | ||
1da177e4 LT |
1137 | /* |
1138 | * Call device private open method | |
1139 | */ | |
1140 | set_bit(__LINK_STATE_START, &dev->state); | |
bada339b | 1141 | |
d314774c SH |
1142 | if (ops->ndo_validate_addr) |
1143 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1144 | |
d314774c SH |
1145 | if (!ret && ops->ndo_open) |
1146 | ret = ops->ndo_open(dev); | |
1da177e4 | 1147 | |
4ec93edb | 1148 | /* |
1da177e4 LT |
1149 | * If it went open OK then: |
1150 | */ | |
1151 | ||
bada339b JG |
1152 | if (ret) |
1153 | clear_bit(__LINK_STATE_START, &dev->state); | |
1154 | else { | |
1da177e4 LT |
1155 | /* |
1156 | * Set the flags. | |
1157 | */ | |
1158 | dev->flags |= IFF_UP; | |
1159 | ||
649274d9 DW |
1160 | /* |
1161 | * Enable NET_DMA | |
1162 | */ | |
b4bd07c2 | 1163 | net_dmaengine_get(); |
649274d9 | 1164 | |
1da177e4 LT |
1165 | /* |
1166 | * Initialize multicasting status | |
1167 | */ | |
4417da66 | 1168 | dev_set_rx_mode(dev); |
1da177e4 LT |
1169 | |
1170 | /* | |
1171 | * Wakeup transmit queue engine | |
1172 | */ | |
1173 | dev_activate(dev); | |
1da177e4 | 1174 | } |
bada339b | 1175 | |
1da177e4 LT |
1176 | return ret; |
1177 | } | |
1178 | ||
1179 | /** | |
bd380811 PM |
1180 | * dev_open - prepare an interface for use. |
1181 | * @dev: device to open | |
1da177e4 | 1182 | * |
bd380811 PM |
1183 | * Takes a device from down to up state. The device's private open |
1184 | * function is invoked and then the multicast lists are loaded. Finally | |
1185 | * the device is moved into the up state and a %NETDEV_UP message is | |
1186 | * sent to the netdev notifier chain. | |
1187 | * | |
1188 | * Calling this function on an active interface is a nop. On a failure | |
1189 | * a negative errno code is returned. | |
1da177e4 | 1190 | */ |
bd380811 PM |
1191 | int dev_open(struct net_device *dev) |
1192 | { | |
1193 | int ret; | |
1194 | ||
1195 | /* | |
1196 | * Is it already up? | |
1197 | */ | |
1198 | if (dev->flags & IFF_UP) | |
1199 | return 0; | |
1200 | ||
1201 | /* | |
1202 | * Open device | |
1203 | */ | |
1204 | ret = __dev_open(dev); | |
1205 | if (ret < 0) | |
1206 | return ret; | |
1207 | ||
1208 | /* | |
1209 | * ... and announce new interface. | |
1210 | */ | |
1211 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING); | |
1212 | call_netdevice_notifiers(NETDEV_UP, dev); | |
1213 | ||
1214 | return ret; | |
1215 | } | |
1216 | EXPORT_SYMBOL(dev_open); | |
1217 | ||
1218 | static int __dev_close(struct net_device *dev) | |
1da177e4 | 1219 | { |
d314774c | 1220 | const struct net_device_ops *ops = dev->netdev_ops; |
e46b66bc | 1221 | |
bd380811 | 1222 | ASSERT_RTNL(); |
9d5010db DM |
1223 | might_sleep(); |
1224 | ||
1da177e4 LT |
1225 | /* |
1226 | * Tell people we are going down, so that they can | |
1227 | * prepare to death, when device is still operating. | |
1228 | */ | |
056925ab | 1229 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1230 | |
1da177e4 LT |
1231 | clear_bit(__LINK_STATE_START, &dev->state); |
1232 | ||
1233 | /* Synchronize to scheduled poll. We cannot touch poll list, | |
bea3348e SH |
1234 | * it can be even on different cpu. So just clear netif_running(). |
1235 | * | |
1236 | * dev->stop() will invoke napi_disable() on all of it's | |
1237 | * napi_struct instances on this device. | |
1238 | */ | |
1da177e4 | 1239 | smp_mb__after_clear_bit(); /* Commit netif_running(). */ |
1da177e4 | 1240 | |
d8b2a4d2 ML |
1241 | dev_deactivate(dev); |
1242 | ||
1da177e4 LT |
1243 | /* |
1244 | * Call the device specific close. This cannot fail. | |
1245 | * Only if device is UP | |
1246 | * | |
1247 | * We allow it to be called even after a DETACH hot-plug | |
1248 | * event. | |
1249 | */ | |
d314774c SH |
1250 | if (ops->ndo_stop) |
1251 | ops->ndo_stop(dev); | |
1da177e4 LT |
1252 | |
1253 | /* | |
1254 | * Device is now down. | |
1255 | */ | |
1256 | ||
1257 | dev->flags &= ~IFF_UP; | |
1258 | ||
1259 | /* | |
bd380811 | 1260 | * Shutdown NET_DMA |
1da177e4 | 1261 | */ |
bd380811 PM |
1262 | net_dmaengine_put(); |
1263 | ||
1264 | return 0; | |
1265 | } | |
1266 | ||
1267 | /** | |
1268 | * dev_close - shutdown an interface. | |
1269 | * @dev: device to shutdown | |
1270 | * | |
1271 | * This function moves an active device into down state. A | |
1272 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1273 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1274 | * chain. | |
1275 | */ | |
1276 | int dev_close(struct net_device *dev) | |
1277 | { | |
1278 | if (!(dev->flags & IFF_UP)) | |
1279 | return 0; | |
1280 | ||
1281 | __dev_close(dev); | |
1da177e4 | 1282 | |
649274d9 | 1283 | /* |
bd380811 | 1284 | * Tell people we are down |
649274d9 | 1285 | */ |
bd380811 PM |
1286 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING); |
1287 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
649274d9 | 1288 | |
1da177e4 LT |
1289 | return 0; |
1290 | } | |
d1b19dff | 1291 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1292 | |
1293 | ||
0187bdfb BH |
1294 | /** |
1295 | * dev_disable_lro - disable Large Receive Offload on a device | |
1296 | * @dev: device | |
1297 | * | |
1298 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1299 | * called under RTNL. This is needed if received packets may be | |
1300 | * forwarded to another interface. | |
1301 | */ | |
1302 | void dev_disable_lro(struct net_device *dev) | |
1303 | { | |
1304 | if (dev->ethtool_ops && dev->ethtool_ops->get_flags && | |
1305 | dev->ethtool_ops->set_flags) { | |
1306 | u32 flags = dev->ethtool_ops->get_flags(dev); | |
1307 | if (flags & ETH_FLAG_LRO) { | |
1308 | flags &= ~ETH_FLAG_LRO; | |
1309 | dev->ethtool_ops->set_flags(dev, flags); | |
1310 | } | |
1311 | } | |
1312 | WARN_ON(dev->features & NETIF_F_LRO); | |
1313 | } | |
1314 | EXPORT_SYMBOL(dev_disable_lro); | |
1315 | ||
1316 | ||
881d966b EB |
1317 | static int dev_boot_phase = 1; |
1318 | ||
1da177e4 LT |
1319 | /* |
1320 | * Device change register/unregister. These are not inline or static | |
1321 | * as we export them to the world. | |
1322 | */ | |
1323 | ||
1324 | /** | |
1325 | * register_netdevice_notifier - register a network notifier block | |
1326 | * @nb: notifier | |
1327 | * | |
1328 | * Register a notifier to be called when network device events occur. | |
1329 | * The notifier passed is linked into the kernel structures and must | |
1330 | * not be reused until it has been unregistered. A negative errno code | |
1331 | * is returned on a failure. | |
1332 | * | |
1333 | * When registered all registration and up events are replayed | |
4ec93edb | 1334 | * to the new notifier to allow device to have a race free |
1da177e4 LT |
1335 | * view of the network device list. |
1336 | */ | |
1337 | ||
1338 | int register_netdevice_notifier(struct notifier_block *nb) | |
1339 | { | |
1340 | struct net_device *dev; | |
fcc5a03a | 1341 | struct net_device *last; |
881d966b | 1342 | struct net *net; |
1da177e4 LT |
1343 | int err; |
1344 | ||
1345 | rtnl_lock(); | |
f07d5b94 | 1346 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1347 | if (err) |
1348 | goto unlock; | |
881d966b EB |
1349 | if (dev_boot_phase) |
1350 | goto unlock; | |
1351 | for_each_net(net) { | |
1352 | for_each_netdev(net, dev) { | |
1353 | err = nb->notifier_call(nb, NETDEV_REGISTER, dev); | |
1354 | err = notifier_to_errno(err); | |
1355 | if (err) | |
1356 | goto rollback; | |
1357 | ||
1358 | if (!(dev->flags & IFF_UP)) | |
1359 | continue; | |
1da177e4 | 1360 | |
881d966b EB |
1361 | nb->notifier_call(nb, NETDEV_UP, dev); |
1362 | } | |
1da177e4 | 1363 | } |
fcc5a03a HX |
1364 | |
1365 | unlock: | |
1da177e4 LT |
1366 | rtnl_unlock(); |
1367 | return err; | |
fcc5a03a HX |
1368 | |
1369 | rollback: | |
1370 | last = dev; | |
881d966b EB |
1371 | for_each_net(net) { |
1372 | for_each_netdev(net, dev) { | |
1373 | if (dev == last) | |
1374 | break; | |
fcc5a03a | 1375 | |
881d966b EB |
1376 | if (dev->flags & IFF_UP) { |
1377 | nb->notifier_call(nb, NETDEV_GOING_DOWN, dev); | |
1378 | nb->notifier_call(nb, NETDEV_DOWN, dev); | |
1379 | } | |
1380 | nb->notifier_call(nb, NETDEV_UNREGISTER, dev); | |
a5ee1551 | 1381 | nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev); |
fcc5a03a | 1382 | } |
fcc5a03a | 1383 | } |
c67625a1 PE |
1384 | |
1385 | raw_notifier_chain_unregister(&netdev_chain, nb); | |
fcc5a03a | 1386 | goto unlock; |
1da177e4 | 1387 | } |
d1b19dff | 1388 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1389 | |
1390 | /** | |
1391 | * unregister_netdevice_notifier - unregister a network notifier block | |
1392 | * @nb: notifier | |
1393 | * | |
1394 | * Unregister a notifier previously registered by | |
1395 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1396 | * kernel structures and may then be reused. A negative errno code | |
1397 | * is returned on a failure. | |
1398 | */ | |
1399 | ||
1400 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1401 | { | |
9f514950 HX |
1402 | int err; |
1403 | ||
1404 | rtnl_lock(); | |
f07d5b94 | 1405 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
9f514950 HX |
1406 | rtnl_unlock(); |
1407 | return err; | |
1da177e4 | 1408 | } |
d1b19dff | 1409 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 LT |
1410 | |
1411 | /** | |
1412 | * call_netdevice_notifiers - call all network notifier blocks | |
1413 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1414 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1415 | * |
1416 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1417 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1418 | */ |
1419 | ||
ad7379d4 | 1420 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1421 | { |
ad7379d4 | 1422 | return raw_notifier_call_chain(&netdev_chain, val, dev); |
1da177e4 LT |
1423 | } |
1424 | ||
1425 | /* When > 0 there are consumers of rx skb time stamps */ | |
1426 | static atomic_t netstamp_needed = ATOMIC_INIT(0); | |
1427 | ||
1428 | void net_enable_timestamp(void) | |
1429 | { | |
1430 | atomic_inc(&netstamp_needed); | |
1431 | } | |
d1b19dff | 1432 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
1433 | |
1434 | void net_disable_timestamp(void) | |
1435 | { | |
1436 | atomic_dec(&netstamp_needed); | |
1437 | } | |
d1b19dff | 1438 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 1439 | |
a61bbcf2 | 1440 | static inline void net_timestamp(struct sk_buff *skb) |
1da177e4 LT |
1441 | { |
1442 | if (atomic_read(&netstamp_needed)) | |
a61bbcf2 | 1443 | __net_timestamp(skb); |
b7aa0bf7 ED |
1444 | else |
1445 | skb->tstamp.tv64 = 0; | |
1da177e4 LT |
1446 | } |
1447 | ||
44540960 AB |
1448 | /** |
1449 | * dev_forward_skb - loopback an skb to another netif | |
1450 | * | |
1451 | * @dev: destination network device | |
1452 | * @skb: buffer to forward | |
1453 | * | |
1454 | * return values: | |
1455 | * NET_RX_SUCCESS (no congestion) | |
1456 | * NET_RX_DROP (packet was dropped) | |
1457 | * | |
1458 | * dev_forward_skb can be used for injecting an skb from the | |
1459 | * start_xmit function of one device into the receive queue | |
1460 | * of another device. | |
1461 | * | |
1462 | * The receiving device may be in another namespace, so | |
1463 | * we have to clear all information in the skb that could | |
1464 | * impact namespace isolation. | |
1465 | */ | |
1466 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
1467 | { | |
1468 | skb_orphan(skb); | |
1469 | ||
1470 | if (!(dev->flags & IFF_UP)) | |
1471 | return NET_RX_DROP; | |
1472 | ||
1473 | if (skb->len > (dev->mtu + dev->hard_header_len)) | |
1474 | return NET_RX_DROP; | |
1475 | ||
8a83a00b | 1476 | skb_set_dev(skb, dev); |
44540960 AB |
1477 | skb->tstamp.tv64 = 0; |
1478 | skb->pkt_type = PACKET_HOST; | |
1479 | skb->protocol = eth_type_trans(skb, dev); | |
44540960 AB |
1480 | return netif_rx(skb); |
1481 | } | |
1482 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
1483 | ||
1da177e4 LT |
1484 | /* |
1485 | * Support routine. Sends outgoing frames to any network | |
1486 | * taps currently in use. | |
1487 | */ | |
1488 | ||
f6a78bfc | 1489 | static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
1490 | { |
1491 | struct packet_type *ptype; | |
a61bbcf2 | 1492 | |
8caf1539 JP |
1493 | #ifdef CONFIG_NET_CLS_ACT |
1494 | if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS))) | |
1495 | net_timestamp(skb); | |
1496 | #else | |
a61bbcf2 | 1497 | net_timestamp(skb); |
8caf1539 | 1498 | #endif |
1da177e4 LT |
1499 | |
1500 | rcu_read_lock(); | |
1501 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
1502 | /* Never send packets back to the socket | |
1503 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1504 | */ | |
1505 | if ((ptype->dev == dev || !ptype->dev) && | |
1506 | (ptype->af_packet_priv == NULL || | |
1507 | (struct sock *)ptype->af_packet_priv != skb->sk)) { | |
d1b19dff | 1508 | struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC); |
1da177e4 LT |
1509 | if (!skb2) |
1510 | break; | |
1511 | ||
1512 | /* skb->nh should be correctly | |
1513 | set by sender, so that the second statement is | |
1514 | just protection against buggy protocols. | |
1515 | */ | |
459a98ed | 1516 | skb_reset_mac_header(skb2); |
1da177e4 | 1517 | |
d56f90a7 | 1518 | if (skb_network_header(skb2) < skb2->data || |
27a884dc | 1519 | skb2->network_header > skb2->tail) { |
1da177e4 LT |
1520 | if (net_ratelimit()) |
1521 | printk(KERN_CRIT "protocol %04x is " | |
1522 | "buggy, dev %s\n", | |
1523 | skb2->protocol, dev->name); | |
c1d2bbe1 | 1524 | skb_reset_network_header(skb2); |
1da177e4 LT |
1525 | } |
1526 | ||
b0e380b1 | 1527 | skb2->transport_header = skb2->network_header; |
1da177e4 | 1528 | skb2->pkt_type = PACKET_OUTGOING; |
f2ccd8fa | 1529 | ptype->func(skb2, skb->dev, ptype, skb->dev); |
1da177e4 LT |
1530 | } |
1531 | } | |
1532 | rcu_read_unlock(); | |
1533 | } | |
1534 | ||
56079431 | 1535 | |
def82a1d | 1536 | static inline void __netif_reschedule(struct Qdisc *q) |
56079431 | 1537 | { |
def82a1d JP |
1538 | struct softnet_data *sd; |
1539 | unsigned long flags; | |
56079431 | 1540 | |
def82a1d JP |
1541 | local_irq_save(flags); |
1542 | sd = &__get_cpu_var(softnet_data); | |
1543 | q->next_sched = sd->output_queue; | |
1544 | sd->output_queue = q; | |
1545 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
1546 | local_irq_restore(flags); | |
1547 | } | |
1548 | ||
1549 | void __netif_schedule(struct Qdisc *q) | |
1550 | { | |
1551 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
1552 | __netif_reschedule(q); | |
56079431 DV |
1553 | } |
1554 | EXPORT_SYMBOL(__netif_schedule); | |
1555 | ||
bea3348e | 1556 | void dev_kfree_skb_irq(struct sk_buff *skb) |
56079431 | 1557 | { |
bea3348e SH |
1558 | if (atomic_dec_and_test(&skb->users)) { |
1559 | struct softnet_data *sd; | |
1560 | unsigned long flags; | |
56079431 | 1561 | |
bea3348e SH |
1562 | local_irq_save(flags); |
1563 | sd = &__get_cpu_var(softnet_data); | |
1564 | skb->next = sd->completion_queue; | |
1565 | sd->completion_queue = skb; | |
1566 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
1567 | local_irq_restore(flags); | |
1568 | } | |
56079431 | 1569 | } |
bea3348e | 1570 | EXPORT_SYMBOL(dev_kfree_skb_irq); |
56079431 DV |
1571 | |
1572 | void dev_kfree_skb_any(struct sk_buff *skb) | |
1573 | { | |
1574 | if (in_irq() || irqs_disabled()) | |
1575 | dev_kfree_skb_irq(skb); | |
1576 | else | |
1577 | dev_kfree_skb(skb); | |
1578 | } | |
1579 | EXPORT_SYMBOL(dev_kfree_skb_any); | |
1580 | ||
1581 | ||
bea3348e SH |
1582 | /** |
1583 | * netif_device_detach - mark device as removed | |
1584 | * @dev: network device | |
1585 | * | |
1586 | * Mark device as removed from system and therefore no longer available. | |
1587 | */ | |
56079431 DV |
1588 | void netif_device_detach(struct net_device *dev) |
1589 | { | |
1590 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
1591 | netif_running(dev)) { | |
d543103a | 1592 | netif_tx_stop_all_queues(dev); |
56079431 DV |
1593 | } |
1594 | } | |
1595 | EXPORT_SYMBOL(netif_device_detach); | |
1596 | ||
bea3348e SH |
1597 | /** |
1598 | * netif_device_attach - mark device as attached | |
1599 | * @dev: network device | |
1600 | * | |
1601 | * Mark device as attached from system and restart if needed. | |
1602 | */ | |
56079431 DV |
1603 | void netif_device_attach(struct net_device *dev) |
1604 | { | |
1605 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
1606 | netif_running(dev)) { | |
d543103a | 1607 | netif_tx_wake_all_queues(dev); |
4ec93edb | 1608 | __netdev_watchdog_up(dev); |
56079431 DV |
1609 | } |
1610 | } | |
1611 | EXPORT_SYMBOL(netif_device_attach); | |
1612 | ||
6de329e2 BH |
1613 | static bool can_checksum_protocol(unsigned long features, __be16 protocol) |
1614 | { | |
1615 | return ((features & NETIF_F_GEN_CSUM) || | |
1616 | ((features & NETIF_F_IP_CSUM) && | |
1617 | protocol == htons(ETH_P_IP)) || | |
1618 | ((features & NETIF_F_IPV6_CSUM) && | |
1c8dbcf6 YZ |
1619 | protocol == htons(ETH_P_IPV6)) || |
1620 | ((features & NETIF_F_FCOE_CRC) && | |
1621 | protocol == htons(ETH_P_FCOE))); | |
6de329e2 BH |
1622 | } |
1623 | ||
1624 | static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb) | |
1625 | { | |
1626 | if (can_checksum_protocol(dev->features, skb->protocol)) | |
1627 | return true; | |
1628 | ||
1629 | if (skb->protocol == htons(ETH_P_8021Q)) { | |
1630 | struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data; | |
1631 | if (can_checksum_protocol(dev->features & dev->vlan_features, | |
1632 | veh->h_vlan_encapsulated_proto)) | |
1633 | return true; | |
1634 | } | |
1635 | ||
1636 | return false; | |
1637 | } | |
56079431 | 1638 | |
8a83a00b AB |
1639 | /** |
1640 | * skb_dev_set -- assign a new device to a buffer | |
1641 | * @skb: buffer for the new device | |
1642 | * @dev: network device | |
1643 | * | |
1644 | * If an skb is owned by a device already, we have to reset | |
1645 | * all data private to the namespace a device belongs to | |
1646 | * before assigning it a new device. | |
1647 | */ | |
1648 | #ifdef CONFIG_NET_NS | |
1649 | void skb_set_dev(struct sk_buff *skb, struct net_device *dev) | |
1650 | { | |
1651 | skb_dst_drop(skb); | |
1652 | if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) { | |
1653 | secpath_reset(skb); | |
1654 | nf_reset(skb); | |
1655 | skb_init_secmark(skb); | |
1656 | skb->mark = 0; | |
1657 | skb->priority = 0; | |
1658 | skb->nf_trace = 0; | |
1659 | skb->ipvs_property = 0; | |
1660 | #ifdef CONFIG_NET_SCHED | |
1661 | skb->tc_index = 0; | |
1662 | #endif | |
1663 | } | |
1664 | skb->dev = dev; | |
1665 | } | |
1666 | EXPORT_SYMBOL(skb_set_dev); | |
1667 | #endif /* CONFIG_NET_NS */ | |
1668 | ||
1da177e4 LT |
1669 | /* |
1670 | * Invalidate hardware checksum when packet is to be mangled, and | |
1671 | * complete checksum manually on outgoing path. | |
1672 | */ | |
84fa7933 | 1673 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 1674 | { |
d3bc23e7 | 1675 | __wsum csum; |
663ead3b | 1676 | int ret = 0, offset; |
1da177e4 | 1677 | |
84fa7933 | 1678 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
1679 | goto out_set_summed; |
1680 | ||
1681 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
a430a43d HX |
1682 | /* Let GSO fix up the checksum. */ |
1683 | goto out_set_summed; | |
1da177e4 LT |
1684 | } |
1685 | ||
a030847e HX |
1686 | offset = skb->csum_start - skb_headroom(skb); |
1687 | BUG_ON(offset >= skb_headlen(skb)); | |
1688 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
1689 | ||
1690 | offset += skb->csum_offset; | |
1691 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
1692 | ||
1693 | if (skb_cloned(skb) && | |
1694 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1da177e4 LT |
1695 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
1696 | if (ret) | |
1697 | goto out; | |
1698 | } | |
1699 | ||
a030847e | 1700 | *(__sum16 *)(skb->data + offset) = csum_fold(csum); |
a430a43d | 1701 | out_set_summed: |
1da177e4 | 1702 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 1703 | out: |
1da177e4 LT |
1704 | return ret; |
1705 | } | |
d1b19dff | 1706 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 1707 | |
f6a78bfc HX |
1708 | /** |
1709 | * skb_gso_segment - Perform segmentation on skb. | |
1710 | * @skb: buffer to segment | |
576a30eb | 1711 | * @features: features for the output path (see dev->features) |
f6a78bfc HX |
1712 | * |
1713 | * This function segments the given skb and returns a list of segments. | |
576a30eb HX |
1714 | * |
1715 | * It may return NULL if the skb requires no segmentation. This is | |
1716 | * only possible when GSO is used for verifying header integrity. | |
f6a78bfc | 1717 | */ |
576a30eb | 1718 | struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features) |
f6a78bfc HX |
1719 | { |
1720 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
1721 | struct packet_type *ptype; | |
252e3346 | 1722 | __be16 type = skb->protocol; |
a430a43d | 1723 | int err; |
f6a78bfc | 1724 | |
459a98ed | 1725 | skb_reset_mac_header(skb); |
b0e380b1 | 1726 | skb->mac_len = skb->network_header - skb->mac_header; |
f6a78bfc HX |
1727 | __skb_pull(skb, skb->mac_len); |
1728 | ||
67fd1a73 HX |
1729 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { |
1730 | struct net_device *dev = skb->dev; | |
1731 | struct ethtool_drvinfo info = {}; | |
1732 | ||
1733 | if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo) | |
1734 | dev->ethtool_ops->get_drvinfo(dev, &info); | |
1735 | ||
1736 | WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d " | |
1737 | "ip_summed=%d", | |
1738 | info.driver, dev ? dev->features : 0L, | |
1739 | skb->sk ? skb->sk->sk_route_caps : 0L, | |
1740 | skb->len, skb->data_len, skb->ip_summed); | |
1741 | ||
a430a43d HX |
1742 | if (skb_header_cloned(skb) && |
1743 | (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))) | |
1744 | return ERR_PTR(err); | |
1745 | } | |
1746 | ||
f6a78bfc | 1747 | rcu_read_lock(); |
82d8a867 PE |
1748 | list_for_each_entry_rcu(ptype, |
1749 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
f6a78bfc | 1750 | if (ptype->type == type && !ptype->dev && ptype->gso_segment) { |
84fa7933 | 1751 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { |
a430a43d HX |
1752 | err = ptype->gso_send_check(skb); |
1753 | segs = ERR_PTR(err); | |
1754 | if (err || skb_gso_ok(skb, features)) | |
1755 | break; | |
d56f90a7 ACM |
1756 | __skb_push(skb, (skb->data - |
1757 | skb_network_header(skb))); | |
a430a43d | 1758 | } |
576a30eb | 1759 | segs = ptype->gso_segment(skb, features); |
f6a78bfc HX |
1760 | break; |
1761 | } | |
1762 | } | |
1763 | rcu_read_unlock(); | |
1764 | ||
98e399f8 | 1765 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 1766 | |
f6a78bfc HX |
1767 | return segs; |
1768 | } | |
f6a78bfc HX |
1769 | EXPORT_SYMBOL(skb_gso_segment); |
1770 | ||
fb286bb2 HX |
1771 | /* Take action when hardware reception checksum errors are detected. */ |
1772 | #ifdef CONFIG_BUG | |
1773 | void netdev_rx_csum_fault(struct net_device *dev) | |
1774 | { | |
1775 | if (net_ratelimit()) { | |
4ec93edb | 1776 | printk(KERN_ERR "%s: hw csum failure.\n", |
246a4212 | 1777 | dev ? dev->name : "<unknown>"); |
fb286bb2 HX |
1778 | dump_stack(); |
1779 | } | |
1780 | } | |
1781 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
1782 | #endif | |
1783 | ||
1da177e4 LT |
1784 | /* Actually, we should eliminate this check as soon as we know, that: |
1785 | * 1. IOMMU is present and allows to map all the memory. | |
1786 | * 2. No high memory really exists on this machine. | |
1787 | */ | |
1788 | ||
1789 | static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb) | |
1790 | { | |
3d3a8533 | 1791 | #ifdef CONFIG_HIGHMEM |
1da177e4 LT |
1792 | int i; |
1793 | ||
1794 | if (dev->features & NETIF_F_HIGHDMA) | |
1795 | return 0; | |
1796 | ||
1797 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) | |
1798 | if (PageHighMem(skb_shinfo(skb)->frags[i].page)) | |
1799 | return 1; | |
1800 | ||
3d3a8533 | 1801 | #endif |
1da177e4 LT |
1802 | return 0; |
1803 | } | |
1da177e4 | 1804 | |
f6a78bfc HX |
1805 | struct dev_gso_cb { |
1806 | void (*destructor)(struct sk_buff *skb); | |
1807 | }; | |
1808 | ||
1809 | #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb) | |
1810 | ||
1811 | static void dev_gso_skb_destructor(struct sk_buff *skb) | |
1812 | { | |
1813 | struct dev_gso_cb *cb; | |
1814 | ||
1815 | do { | |
1816 | struct sk_buff *nskb = skb->next; | |
1817 | ||
1818 | skb->next = nskb->next; | |
1819 | nskb->next = NULL; | |
1820 | kfree_skb(nskb); | |
1821 | } while (skb->next); | |
1822 | ||
1823 | cb = DEV_GSO_CB(skb); | |
1824 | if (cb->destructor) | |
1825 | cb->destructor(skb); | |
1826 | } | |
1827 | ||
1828 | /** | |
1829 | * dev_gso_segment - Perform emulated hardware segmentation on skb. | |
1830 | * @skb: buffer to segment | |
1831 | * | |
1832 | * This function segments the given skb and stores the list of segments | |
1833 | * in skb->next. | |
1834 | */ | |
1835 | static int dev_gso_segment(struct sk_buff *skb) | |
1836 | { | |
1837 | struct net_device *dev = skb->dev; | |
1838 | struct sk_buff *segs; | |
576a30eb HX |
1839 | int features = dev->features & ~(illegal_highdma(dev, skb) ? |
1840 | NETIF_F_SG : 0); | |
1841 | ||
1842 | segs = skb_gso_segment(skb, features); | |
1843 | ||
1844 | /* Verifying header integrity only. */ | |
1845 | if (!segs) | |
1846 | return 0; | |
f6a78bfc | 1847 | |
801678c5 | 1848 | if (IS_ERR(segs)) |
f6a78bfc HX |
1849 | return PTR_ERR(segs); |
1850 | ||
1851 | skb->next = segs; | |
1852 | DEV_GSO_CB(skb)->destructor = skb->destructor; | |
1853 | skb->destructor = dev_gso_skb_destructor; | |
1854 | ||
1855 | return 0; | |
1856 | } | |
1857 | ||
fd2ea0a7 DM |
1858 | int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev, |
1859 | struct netdev_queue *txq) | |
f6a78bfc | 1860 | { |
00829823 | 1861 | const struct net_device_ops *ops = dev->netdev_ops; |
572a9d7b | 1862 | int rc = NETDEV_TX_OK; |
00829823 | 1863 | |
f6a78bfc | 1864 | if (likely(!skb->next)) { |
9be9a6b9 | 1865 | if (!list_empty(&ptype_all)) |
f6a78bfc HX |
1866 | dev_queue_xmit_nit(skb, dev); |
1867 | ||
576a30eb HX |
1868 | if (netif_needs_gso(dev, skb)) { |
1869 | if (unlikely(dev_gso_segment(skb))) | |
1870 | goto out_kfree_skb; | |
1871 | if (skb->next) | |
1872 | goto gso; | |
1873 | } | |
f6a78bfc | 1874 | |
93f154b5 ED |
1875 | /* |
1876 | * If device doesnt need skb->dst, release it right now while | |
1877 | * its hot in this cpu cache | |
1878 | */ | |
adf30907 ED |
1879 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) |
1880 | skb_dst_drop(skb); | |
1881 | ||
ac45f602 | 1882 | rc = ops->ndo_start_xmit(skb, dev); |
ec634fe3 | 1883 | if (rc == NETDEV_TX_OK) |
08baf561 | 1884 | txq_trans_update(txq); |
ac45f602 PO |
1885 | /* |
1886 | * TODO: if skb_orphan() was called by | |
1887 | * dev->hard_start_xmit() (for example, the unmodified | |
1888 | * igb driver does that; bnx2 doesn't), then | |
1889 | * skb_tx_software_timestamp() will be unable to send | |
1890 | * back the time stamp. | |
1891 | * | |
1892 | * How can this be prevented? Always create another | |
1893 | * reference to the socket before calling | |
1894 | * dev->hard_start_xmit()? Prevent that skb_orphan() | |
1895 | * does anything in dev->hard_start_xmit() by clearing | |
1896 | * the skb destructor before the call and restoring it | |
1897 | * afterwards, then doing the skb_orphan() ourselves? | |
1898 | */ | |
ac45f602 | 1899 | return rc; |
f6a78bfc HX |
1900 | } |
1901 | ||
576a30eb | 1902 | gso: |
f6a78bfc HX |
1903 | do { |
1904 | struct sk_buff *nskb = skb->next; | |
f6a78bfc HX |
1905 | |
1906 | skb->next = nskb->next; | |
1907 | nskb->next = NULL; | |
068a2de5 KK |
1908 | |
1909 | /* | |
1910 | * If device doesnt need nskb->dst, release it right now while | |
1911 | * its hot in this cpu cache | |
1912 | */ | |
1913 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
1914 | skb_dst_drop(nskb); | |
1915 | ||
00829823 | 1916 | rc = ops->ndo_start_xmit(nskb, dev); |
ec634fe3 | 1917 | if (unlikely(rc != NETDEV_TX_OK)) { |
572a9d7b PM |
1918 | if (rc & ~NETDEV_TX_MASK) |
1919 | goto out_kfree_gso_skb; | |
f54d9e8d | 1920 | nskb->next = skb->next; |
f6a78bfc HX |
1921 | skb->next = nskb; |
1922 | return rc; | |
1923 | } | |
08baf561 | 1924 | txq_trans_update(txq); |
fd2ea0a7 | 1925 | if (unlikely(netif_tx_queue_stopped(txq) && skb->next)) |
f54d9e8d | 1926 | return NETDEV_TX_BUSY; |
f6a78bfc | 1927 | } while (skb->next); |
4ec93edb | 1928 | |
572a9d7b PM |
1929 | out_kfree_gso_skb: |
1930 | if (likely(skb->next == NULL)) | |
1931 | skb->destructor = DEV_GSO_CB(skb)->destructor; | |
f6a78bfc HX |
1932 | out_kfree_skb: |
1933 | kfree_skb(skb); | |
572a9d7b | 1934 | return rc; |
f6a78bfc HX |
1935 | } |
1936 | ||
0a9627f2 | 1937 | static u32 hashrnd __read_mostly; |
b6b2fed1 | 1938 | |
9247744e | 1939 | u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb) |
8f0f2223 | 1940 | { |
7019298a | 1941 | u32 hash; |
b6b2fed1 | 1942 | |
513de11b DM |
1943 | if (skb_rx_queue_recorded(skb)) { |
1944 | hash = skb_get_rx_queue(skb); | |
d1b19dff | 1945 | while (unlikely(hash >= dev->real_num_tx_queues)) |
513de11b DM |
1946 | hash -= dev->real_num_tx_queues; |
1947 | return hash; | |
1948 | } | |
ec581f6a ED |
1949 | |
1950 | if (skb->sk && skb->sk->sk_hash) | |
7019298a | 1951 | hash = skb->sk->sk_hash; |
ec581f6a | 1952 | else |
7019298a | 1953 | hash = skb->protocol; |
d5a9e24a | 1954 | |
0a9627f2 | 1955 | hash = jhash_1word(hash, hashrnd); |
b6b2fed1 DM |
1956 | |
1957 | return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32); | |
8f0f2223 | 1958 | } |
9247744e | 1959 | EXPORT_SYMBOL(skb_tx_hash); |
8f0f2223 | 1960 | |
ed04642f ED |
1961 | static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index) |
1962 | { | |
1963 | if (unlikely(queue_index >= dev->real_num_tx_queues)) { | |
1964 | if (net_ratelimit()) { | |
0a9627f2 | 1965 | netdev_warn(dev, "selects TX queue %d, but " |
ed04642f | 1966 | "real number of TX queues is %d\n", |
0a9627f2 | 1967 | queue_index, dev->real_num_tx_queues); |
ed04642f ED |
1968 | } |
1969 | return 0; | |
1970 | } | |
1971 | return queue_index; | |
1972 | } | |
1973 | ||
e8a0464c DM |
1974 | static struct netdev_queue *dev_pick_tx(struct net_device *dev, |
1975 | struct sk_buff *skb) | |
1976 | { | |
a4ee3ce3 KK |
1977 | u16 queue_index; |
1978 | struct sock *sk = skb->sk; | |
1979 | ||
1980 | if (sk_tx_queue_recorded(sk)) { | |
1981 | queue_index = sk_tx_queue_get(sk); | |
1982 | } else { | |
1983 | const struct net_device_ops *ops = dev->netdev_ops; | |
1984 | ||
1985 | if (ops->ndo_select_queue) { | |
1986 | queue_index = ops->ndo_select_queue(dev, skb); | |
ed04642f | 1987 | queue_index = dev_cap_txqueue(dev, queue_index); |
a4ee3ce3 KK |
1988 | } else { |
1989 | queue_index = 0; | |
1990 | if (dev->real_num_tx_queues > 1) | |
1991 | queue_index = skb_tx_hash(dev, skb); | |
fd2ea0a7 | 1992 | |
a4ee3ce3 KK |
1993 | if (sk && sk->sk_dst_cache) |
1994 | sk_tx_queue_set(sk, queue_index); | |
1995 | } | |
1996 | } | |
eae792b7 | 1997 | |
fd2ea0a7 DM |
1998 | skb_set_queue_mapping(skb, queue_index); |
1999 | return netdev_get_tx_queue(dev, queue_index); | |
e8a0464c DM |
2000 | } |
2001 | ||
bbd8a0d3 KK |
2002 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
2003 | struct net_device *dev, | |
2004 | struct netdev_queue *txq) | |
2005 | { | |
2006 | spinlock_t *root_lock = qdisc_lock(q); | |
2007 | int rc; | |
2008 | ||
2009 | spin_lock(root_lock); | |
2010 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
2011 | kfree_skb(skb); | |
2012 | rc = NET_XMIT_DROP; | |
2013 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
2014 | !test_and_set_bit(__QDISC_STATE_RUNNING, &q->state)) { | |
2015 | /* | |
2016 | * This is a work-conserving queue; there are no old skbs | |
2017 | * waiting to be sent out; and the qdisc is not running - | |
2018 | * xmit the skb directly. | |
2019 | */ | |
2020 | __qdisc_update_bstats(q, skb->len); | |
2021 | if (sch_direct_xmit(skb, q, dev, txq, root_lock)) | |
2022 | __qdisc_run(q); | |
2023 | else | |
2024 | clear_bit(__QDISC_STATE_RUNNING, &q->state); | |
2025 | ||
2026 | rc = NET_XMIT_SUCCESS; | |
2027 | } else { | |
2028 | rc = qdisc_enqueue_root(skb, q); | |
2029 | qdisc_run(q); | |
2030 | } | |
2031 | spin_unlock(root_lock); | |
2032 | ||
2033 | return rc; | |
2034 | } | |
2035 | ||
4b258461 KK |
2036 | /* |
2037 | * Returns true if either: | |
2038 | * 1. skb has frag_list and the device doesn't support FRAGLIST, or | |
2039 | * 2. skb is fragmented and the device does not support SG, or if | |
2040 | * at least one of fragments is in highmem and device does not | |
2041 | * support DMA from it. | |
2042 | */ | |
2043 | static inline int skb_needs_linearize(struct sk_buff *skb, | |
2044 | struct net_device *dev) | |
2045 | { | |
2046 | return (skb_has_frags(skb) && !(dev->features & NETIF_F_FRAGLIST)) || | |
2047 | (skb_shinfo(skb)->nr_frags && (!(dev->features & NETIF_F_SG) || | |
2048 | illegal_highdma(dev, skb))); | |
2049 | } | |
2050 | ||
d29f749e DJ |
2051 | /** |
2052 | * dev_queue_xmit - transmit a buffer | |
2053 | * @skb: buffer to transmit | |
2054 | * | |
2055 | * Queue a buffer for transmission to a network device. The caller must | |
2056 | * have set the device and priority and built the buffer before calling | |
2057 | * this function. The function can be called from an interrupt. | |
2058 | * | |
2059 | * A negative errno code is returned on a failure. A success does not | |
2060 | * guarantee the frame will be transmitted as it may be dropped due | |
2061 | * to congestion or traffic shaping. | |
2062 | * | |
2063 | * ----------------------------------------------------------------------------------- | |
2064 | * I notice this method can also return errors from the queue disciplines, | |
2065 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
2066 | * be positive. | |
2067 | * | |
2068 | * Regardless of the return value, the skb is consumed, so it is currently | |
2069 | * difficult to retry a send to this method. (You can bump the ref count | |
2070 | * before sending to hold a reference for retry if you are careful.) | |
2071 | * | |
2072 | * When calling this method, interrupts MUST be enabled. This is because | |
2073 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
2074 | * --BLG | |
2075 | */ | |
1da177e4 LT |
2076 | int dev_queue_xmit(struct sk_buff *skb) |
2077 | { | |
2078 | struct net_device *dev = skb->dev; | |
dc2b4847 | 2079 | struct netdev_queue *txq; |
1da177e4 LT |
2080 | struct Qdisc *q; |
2081 | int rc = -ENOMEM; | |
2082 | ||
f6a78bfc HX |
2083 | /* GSO will handle the following emulations directly. */ |
2084 | if (netif_needs_gso(dev, skb)) | |
2085 | goto gso; | |
2086 | ||
4b258461 KK |
2087 | /* Convert a paged skb to linear, if required */ |
2088 | if (skb_needs_linearize(skb, dev) && __skb_linearize(skb)) | |
1da177e4 LT |
2089 | goto out_kfree_skb; |
2090 | ||
2091 | /* If packet is not checksummed and device does not support | |
2092 | * checksumming for this protocol, complete checksumming here. | |
2093 | */ | |
663ead3b HX |
2094 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
2095 | skb_set_transport_header(skb, skb->csum_start - | |
2096 | skb_headroom(skb)); | |
6de329e2 BH |
2097 | if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb)) |
2098 | goto out_kfree_skb; | |
663ead3b | 2099 | } |
1da177e4 | 2100 | |
f6a78bfc | 2101 | gso: |
4ec93edb YH |
2102 | /* Disable soft irqs for various locks below. Also |
2103 | * stops preemption for RCU. | |
1da177e4 | 2104 | */ |
4ec93edb | 2105 | rcu_read_lock_bh(); |
1da177e4 | 2106 | |
eae792b7 | 2107 | txq = dev_pick_tx(dev, skb); |
a898def2 | 2108 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 2109 | |
1da177e4 | 2110 | #ifdef CONFIG_NET_CLS_ACT |
d1b19dff | 2111 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS); |
1da177e4 LT |
2112 | #endif |
2113 | if (q->enqueue) { | |
bbd8a0d3 | 2114 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 2115 | goto out; |
1da177e4 LT |
2116 | } |
2117 | ||
2118 | /* The device has no queue. Common case for software devices: | |
2119 | loopback, all the sorts of tunnels... | |
2120 | ||
932ff279 HX |
2121 | Really, it is unlikely that netif_tx_lock protection is necessary |
2122 | here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
1da177e4 LT |
2123 | counters.) |
2124 | However, it is possible, that they rely on protection | |
2125 | made by us here. | |
2126 | ||
2127 | Check this and shot the lock. It is not prone from deadlocks. | |
2128 | Either shot noqueue qdisc, it is even simpler 8) | |
2129 | */ | |
2130 | if (dev->flags & IFF_UP) { | |
2131 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
2132 | ||
c773e847 | 2133 | if (txq->xmit_lock_owner != cpu) { |
1da177e4 | 2134 | |
c773e847 | 2135 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 2136 | |
fd2ea0a7 | 2137 | if (!netif_tx_queue_stopped(txq)) { |
572a9d7b PM |
2138 | rc = dev_hard_start_xmit(skb, dev, txq); |
2139 | if (dev_xmit_complete(rc)) { | |
c773e847 | 2140 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
2141 | goto out; |
2142 | } | |
2143 | } | |
c773e847 | 2144 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
2145 | if (net_ratelimit()) |
2146 | printk(KERN_CRIT "Virtual device %s asks to " | |
2147 | "queue packet!\n", dev->name); | |
2148 | } else { | |
2149 | /* Recursion is detected! It is possible, | |
2150 | * unfortunately */ | |
2151 | if (net_ratelimit()) | |
2152 | printk(KERN_CRIT "Dead loop on virtual device " | |
2153 | "%s, fix it urgently!\n", dev->name); | |
2154 | } | |
2155 | } | |
2156 | ||
2157 | rc = -ENETDOWN; | |
d4828d85 | 2158 | rcu_read_unlock_bh(); |
1da177e4 LT |
2159 | |
2160 | out_kfree_skb: | |
2161 | kfree_skb(skb); | |
2162 | return rc; | |
2163 | out: | |
d4828d85 | 2164 | rcu_read_unlock_bh(); |
1da177e4 LT |
2165 | return rc; |
2166 | } | |
d1b19dff | 2167 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 LT |
2168 | |
2169 | ||
2170 | /*======================================================================= | |
2171 | Receiver routines | |
2172 | =======================================================================*/ | |
2173 | ||
6b2bedc3 SH |
2174 | int netdev_max_backlog __read_mostly = 1000; |
2175 | int netdev_budget __read_mostly = 300; | |
2176 | int weight_p __read_mostly = 64; /* old backlog weight */ | |
1da177e4 LT |
2177 | |
2178 | DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, }; | |
2179 | ||
df334545 | 2180 | #ifdef CONFIG_RPS |
0a9627f2 TH |
2181 | /* |
2182 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
2183 | * CPU from the RPS map of the receiving queue for a given skb. | |
2184 | */ | |
2185 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb) | |
2186 | { | |
2187 | struct ipv6hdr *ip6; | |
2188 | struct iphdr *ip; | |
2189 | struct netdev_rx_queue *rxqueue; | |
2190 | struct rps_map *map; | |
2191 | int cpu = -1; | |
2192 | u8 ip_proto; | |
2193 | u32 addr1, addr2, ports, ihl; | |
2194 | ||
2195 | rcu_read_lock(); | |
2196 | ||
2197 | if (skb_rx_queue_recorded(skb)) { | |
2198 | u16 index = skb_get_rx_queue(skb); | |
2199 | if (unlikely(index >= dev->num_rx_queues)) { | |
2200 | if (net_ratelimit()) { | |
2201 | netdev_warn(dev, "received packet on queue " | |
2202 | "%u, but number of RX queues is %u\n", | |
2203 | index, dev->num_rx_queues); | |
2204 | } | |
2205 | goto done; | |
2206 | } | |
2207 | rxqueue = dev->_rx + index; | |
2208 | } else | |
2209 | rxqueue = dev->_rx; | |
2210 | ||
2211 | if (!rxqueue->rps_map) | |
2212 | goto done; | |
2213 | ||
2214 | if (skb->rxhash) | |
2215 | goto got_hash; /* Skip hash computation on packet header */ | |
2216 | ||
2217 | switch (skb->protocol) { | |
2218 | case __constant_htons(ETH_P_IP): | |
2219 | if (!pskb_may_pull(skb, sizeof(*ip))) | |
2220 | goto done; | |
2221 | ||
2222 | ip = (struct iphdr *) skb->data; | |
2223 | ip_proto = ip->protocol; | |
2224 | addr1 = ip->saddr; | |
2225 | addr2 = ip->daddr; | |
2226 | ihl = ip->ihl; | |
2227 | break; | |
2228 | case __constant_htons(ETH_P_IPV6): | |
2229 | if (!pskb_may_pull(skb, sizeof(*ip6))) | |
2230 | goto done; | |
2231 | ||
2232 | ip6 = (struct ipv6hdr *) skb->data; | |
2233 | ip_proto = ip6->nexthdr; | |
2234 | addr1 = ip6->saddr.s6_addr32[3]; | |
2235 | addr2 = ip6->daddr.s6_addr32[3]; | |
2236 | ihl = (40 >> 2); | |
2237 | break; | |
2238 | default: | |
2239 | goto done; | |
2240 | } | |
2241 | ports = 0; | |
2242 | switch (ip_proto) { | |
2243 | case IPPROTO_TCP: | |
2244 | case IPPROTO_UDP: | |
2245 | case IPPROTO_DCCP: | |
2246 | case IPPROTO_ESP: | |
2247 | case IPPROTO_AH: | |
2248 | case IPPROTO_SCTP: | |
2249 | case IPPROTO_UDPLITE: | |
2250 | if (pskb_may_pull(skb, (ihl * 4) + 4)) | |
2251 | ports = *((u32 *) (skb->data + (ihl * 4))); | |
2252 | break; | |
2253 | ||
2254 | default: | |
2255 | break; | |
2256 | } | |
2257 | ||
2258 | skb->rxhash = jhash_3words(addr1, addr2, ports, hashrnd); | |
2259 | if (!skb->rxhash) | |
2260 | skb->rxhash = 1; | |
2261 | ||
2262 | got_hash: | |
2263 | map = rcu_dereference(rxqueue->rps_map); | |
2264 | if (map) { | |
2265 | u16 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32]; | |
2266 | ||
2267 | if (cpu_online(tcpu)) { | |
2268 | cpu = tcpu; | |
2269 | goto done; | |
2270 | } | |
2271 | } | |
2272 | ||
2273 | done: | |
2274 | rcu_read_unlock(); | |
2275 | return cpu; | |
2276 | } | |
2277 | ||
2278 | /* | |
2279 | * This structure holds the per-CPU mask of CPUs for which IPIs are scheduled | |
2280 | * to be sent to kick remote softirq processing. There are two masks since | |
2281 | * the sending of IPIs must be done with interrupts enabled. The select field | |
2282 | * indicates the current mask that enqueue_backlog uses to schedule IPIs. | |
2283 | * select is flipped before net_rps_action is called while still under lock, | |
2284 | * net_rps_action then uses the non-selected mask to send the IPIs and clears | |
2285 | * it without conflicting with enqueue_backlog operation. | |
2286 | */ | |
2287 | struct rps_remote_softirq_cpus { | |
2288 | cpumask_t mask[2]; | |
2289 | int select; | |
2290 | }; | |
2291 | static DEFINE_PER_CPU(struct rps_remote_softirq_cpus, rps_remote_softirq_cpus); | |
2292 | ||
2293 | /* Called from hardirq (IPI) context */ | |
2294 | static void trigger_softirq(void *data) | |
2295 | { | |
2296 | struct softnet_data *queue = data; | |
2297 | __napi_schedule(&queue->backlog); | |
2298 | __get_cpu_var(netdev_rx_stat).received_rps++; | |
2299 | } | |
1e94d72f | 2300 | #endif /* CONFIG_SMP */ |
0a9627f2 TH |
2301 | |
2302 | /* | |
2303 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
2304 | * queue (may be a remote CPU queue). | |
2305 | */ | |
2306 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu) | |
2307 | { | |
2308 | struct softnet_data *queue; | |
2309 | unsigned long flags; | |
2310 | ||
2311 | queue = &per_cpu(softnet_data, cpu); | |
2312 | ||
2313 | local_irq_save(flags); | |
2314 | __get_cpu_var(netdev_rx_stat).total++; | |
2315 | ||
2316 | spin_lock(&queue->input_pkt_queue.lock); | |
2317 | if (queue->input_pkt_queue.qlen <= netdev_max_backlog) { | |
2318 | if (queue->input_pkt_queue.qlen) { | |
2319 | enqueue: | |
2320 | __skb_queue_tail(&queue->input_pkt_queue, skb); | |
2321 | spin_unlock_irqrestore(&queue->input_pkt_queue.lock, | |
2322 | flags); | |
2323 | return NET_RX_SUCCESS; | |
2324 | } | |
2325 | ||
2326 | /* Schedule NAPI for backlog device */ | |
2327 | if (napi_schedule_prep(&queue->backlog)) { | |
df334545 | 2328 | #ifdef CONFIG_RPS |
0a9627f2 TH |
2329 | if (cpu != smp_processor_id()) { |
2330 | struct rps_remote_softirq_cpus *rcpus = | |
2331 | &__get_cpu_var(rps_remote_softirq_cpus); | |
2332 | ||
2333 | cpu_set(cpu, rcpus->mask[rcpus->select]); | |
2334 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2335 | } else | |
2336 | __napi_schedule(&queue->backlog); | |
1e94d72f TH |
2337 | #else |
2338 | __napi_schedule(&queue->backlog); | |
2339 | #endif | |
0a9627f2 TH |
2340 | } |
2341 | goto enqueue; | |
2342 | } | |
2343 | ||
2344 | spin_unlock(&queue->input_pkt_queue.lock); | |
2345 | ||
2346 | __get_cpu_var(netdev_rx_stat).dropped++; | |
2347 | local_irq_restore(flags); | |
2348 | ||
2349 | kfree_skb(skb); | |
2350 | return NET_RX_DROP; | |
2351 | } | |
1da177e4 | 2352 | |
1da177e4 LT |
2353 | /** |
2354 | * netif_rx - post buffer to the network code | |
2355 | * @skb: buffer to post | |
2356 | * | |
2357 | * This function receives a packet from a device driver and queues it for | |
2358 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
2359 | * may be dropped during processing for congestion control or by the | |
2360 | * protocol layers. | |
2361 | * | |
2362 | * return values: | |
2363 | * NET_RX_SUCCESS (no congestion) | |
1da177e4 LT |
2364 | * NET_RX_DROP (packet was dropped) |
2365 | * | |
2366 | */ | |
2367 | ||
2368 | int netif_rx(struct sk_buff *skb) | |
2369 | { | |
0a9627f2 | 2370 | int cpu; |
1da177e4 LT |
2371 | |
2372 | /* if netpoll wants it, pretend we never saw it */ | |
2373 | if (netpoll_rx(skb)) | |
2374 | return NET_RX_DROP; | |
2375 | ||
b7aa0bf7 | 2376 | if (!skb->tstamp.tv64) |
a61bbcf2 | 2377 | net_timestamp(skb); |
1da177e4 | 2378 | |
df334545 | 2379 | #ifdef CONFIG_RPS |
0a9627f2 TH |
2380 | cpu = get_rps_cpu(skb->dev, skb); |
2381 | if (cpu < 0) | |
2382 | cpu = smp_processor_id(); | |
1e94d72f TH |
2383 | #else |
2384 | cpu = smp_processor_id(); | |
2385 | #endif | |
1da177e4 | 2386 | |
0a9627f2 | 2387 | return enqueue_to_backlog(skb, cpu); |
1da177e4 | 2388 | } |
d1b19dff | 2389 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
2390 | |
2391 | int netif_rx_ni(struct sk_buff *skb) | |
2392 | { | |
2393 | int err; | |
2394 | ||
2395 | preempt_disable(); | |
2396 | err = netif_rx(skb); | |
2397 | if (local_softirq_pending()) | |
2398 | do_softirq(); | |
2399 | preempt_enable(); | |
2400 | ||
2401 | return err; | |
2402 | } | |
1da177e4 LT |
2403 | EXPORT_SYMBOL(netif_rx_ni); |
2404 | ||
1da177e4 LT |
2405 | static void net_tx_action(struct softirq_action *h) |
2406 | { | |
2407 | struct softnet_data *sd = &__get_cpu_var(softnet_data); | |
2408 | ||
2409 | if (sd->completion_queue) { | |
2410 | struct sk_buff *clist; | |
2411 | ||
2412 | local_irq_disable(); | |
2413 | clist = sd->completion_queue; | |
2414 | sd->completion_queue = NULL; | |
2415 | local_irq_enable(); | |
2416 | ||
2417 | while (clist) { | |
2418 | struct sk_buff *skb = clist; | |
2419 | clist = clist->next; | |
2420 | ||
547b792c | 2421 | WARN_ON(atomic_read(&skb->users)); |
1da177e4 LT |
2422 | __kfree_skb(skb); |
2423 | } | |
2424 | } | |
2425 | ||
2426 | if (sd->output_queue) { | |
37437bb2 | 2427 | struct Qdisc *head; |
1da177e4 LT |
2428 | |
2429 | local_irq_disable(); | |
2430 | head = sd->output_queue; | |
2431 | sd->output_queue = NULL; | |
2432 | local_irq_enable(); | |
2433 | ||
2434 | while (head) { | |
37437bb2 DM |
2435 | struct Qdisc *q = head; |
2436 | spinlock_t *root_lock; | |
2437 | ||
1da177e4 LT |
2438 | head = head->next_sched; |
2439 | ||
5fb66229 | 2440 | root_lock = qdisc_lock(q); |
37437bb2 | 2441 | if (spin_trylock(root_lock)) { |
def82a1d JP |
2442 | smp_mb__before_clear_bit(); |
2443 | clear_bit(__QDISC_STATE_SCHED, | |
2444 | &q->state); | |
37437bb2 DM |
2445 | qdisc_run(q); |
2446 | spin_unlock(root_lock); | |
1da177e4 | 2447 | } else { |
195648bb | 2448 | if (!test_bit(__QDISC_STATE_DEACTIVATED, |
e8a83e10 | 2449 | &q->state)) { |
195648bb | 2450 | __netif_reschedule(q); |
e8a83e10 JP |
2451 | } else { |
2452 | smp_mb__before_clear_bit(); | |
2453 | clear_bit(__QDISC_STATE_SCHED, | |
2454 | &q->state); | |
2455 | } | |
1da177e4 LT |
2456 | } |
2457 | } | |
2458 | } | |
2459 | } | |
2460 | ||
6f05f629 SH |
2461 | static inline int deliver_skb(struct sk_buff *skb, |
2462 | struct packet_type *pt_prev, | |
2463 | struct net_device *orig_dev) | |
1da177e4 LT |
2464 | { |
2465 | atomic_inc(&skb->users); | |
f2ccd8fa | 2466 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
1da177e4 LT |
2467 | } |
2468 | ||
2469 | #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE) | |
da678292 MM |
2470 | |
2471 | #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE) | |
2472 | /* This hook is defined here for ATM LANE */ | |
2473 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
2474 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 2475 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 2476 | #endif |
1da177e4 | 2477 | |
6229e362 SH |
2478 | /* |
2479 | * If bridge module is loaded call bridging hook. | |
2480 | * returns NULL if packet was consumed. | |
2481 | */ | |
2482 | struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p, | |
2483 | struct sk_buff *skb) __read_mostly; | |
4fb019a0 | 2484 | EXPORT_SYMBOL_GPL(br_handle_frame_hook); |
da678292 | 2485 | |
6229e362 SH |
2486 | static inline struct sk_buff *handle_bridge(struct sk_buff *skb, |
2487 | struct packet_type **pt_prev, int *ret, | |
2488 | struct net_device *orig_dev) | |
1da177e4 LT |
2489 | { |
2490 | struct net_bridge_port *port; | |
2491 | ||
6229e362 SH |
2492 | if (skb->pkt_type == PACKET_LOOPBACK || |
2493 | (port = rcu_dereference(skb->dev->br_port)) == NULL) | |
2494 | return skb; | |
1da177e4 LT |
2495 | |
2496 | if (*pt_prev) { | |
6229e362 | 2497 | *ret = deliver_skb(skb, *pt_prev, orig_dev); |
1da177e4 | 2498 | *pt_prev = NULL; |
4ec93edb YH |
2499 | } |
2500 | ||
6229e362 | 2501 | return br_handle_frame_hook(port, skb); |
1da177e4 LT |
2502 | } |
2503 | #else | |
6229e362 | 2504 | #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb) |
1da177e4 LT |
2505 | #endif |
2506 | ||
b863ceb7 PM |
2507 | #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE) |
2508 | struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly; | |
2509 | EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook); | |
2510 | ||
2511 | static inline struct sk_buff *handle_macvlan(struct sk_buff *skb, | |
2512 | struct packet_type **pt_prev, | |
2513 | int *ret, | |
2514 | struct net_device *orig_dev) | |
2515 | { | |
2516 | if (skb->dev->macvlan_port == NULL) | |
2517 | return skb; | |
2518 | ||
2519 | if (*pt_prev) { | |
2520 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
2521 | *pt_prev = NULL; | |
2522 | } | |
2523 | return macvlan_handle_frame_hook(skb); | |
2524 | } | |
2525 | #else | |
2526 | #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb) | |
2527 | #endif | |
2528 | ||
1da177e4 LT |
2529 | #ifdef CONFIG_NET_CLS_ACT |
2530 | /* TODO: Maybe we should just force sch_ingress to be compiled in | |
2531 | * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions | |
2532 | * a compare and 2 stores extra right now if we dont have it on | |
2533 | * but have CONFIG_NET_CLS_ACT | |
4ec93edb | 2534 | * NOTE: This doesnt stop any functionality; if you dont have |
1da177e4 LT |
2535 | * the ingress scheduler, you just cant add policies on ingress. |
2536 | * | |
2537 | */ | |
4ec93edb | 2538 | static int ing_filter(struct sk_buff *skb) |
1da177e4 | 2539 | { |
1da177e4 | 2540 | struct net_device *dev = skb->dev; |
f697c3e8 | 2541 | u32 ttl = G_TC_RTTL(skb->tc_verd); |
555353cf DM |
2542 | struct netdev_queue *rxq; |
2543 | int result = TC_ACT_OK; | |
2544 | struct Qdisc *q; | |
4ec93edb | 2545 | |
f697c3e8 HX |
2546 | if (MAX_RED_LOOP < ttl++) { |
2547 | printk(KERN_WARNING | |
2548 | "Redir loop detected Dropping packet (%d->%d)\n", | |
8964be4a | 2549 | skb->skb_iif, dev->ifindex); |
f697c3e8 HX |
2550 | return TC_ACT_SHOT; |
2551 | } | |
1da177e4 | 2552 | |
f697c3e8 HX |
2553 | skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl); |
2554 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); | |
1da177e4 | 2555 | |
555353cf DM |
2556 | rxq = &dev->rx_queue; |
2557 | ||
83874000 | 2558 | q = rxq->qdisc; |
8d50b53d | 2559 | if (q != &noop_qdisc) { |
83874000 | 2560 | spin_lock(qdisc_lock(q)); |
a9312ae8 DM |
2561 | if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) |
2562 | result = qdisc_enqueue_root(skb, q); | |
83874000 DM |
2563 | spin_unlock(qdisc_lock(q)); |
2564 | } | |
f697c3e8 HX |
2565 | |
2566 | return result; | |
2567 | } | |
86e65da9 | 2568 | |
f697c3e8 HX |
2569 | static inline struct sk_buff *handle_ing(struct sk_buff *skb, |
2570 | struct packet_type **pt_prev, | |
2571 | int *ret, struct net_device *orig_dev) | |
2572 | { | |
8d50b53d | 2573 | if (skb->dev->rx_queue.qdisc == &noop_qdisc) |
f697c3e8 | 2574 | goto out; |
1da177e4 | 2575 | |
f697c3e8 HX |
2576 | if (*pt_prev) { |
2577 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
2578 | *pt_prev = NULL; | |
2579 | } else { | |
2580 | /* Huh? Why does turning on AF_PACKET affect this? */ | |
2581 | skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd); | |
1da177e4 LT |
2582 | } |
2583 | ||
f697c3e8 HX |
2584 | switch (ing_filter(skb)) { |
2585 | case TC_ACT_SHOT: | |
2586 | case TC_ACT_STOLEN: | |
2587 | kfree_skb(skb); | |
2588 | return NULL; | |
2589 | } | |
2590 | ||
2591 | out: | |
2592 | skb->tc_verd = 0; | |
2593 | return skb; | |
1da177e4 LT |
2594 | } |
2595 | #endif | |
2596 | ||
bc1d0411 PM |
2597 | /* |
2598 | * netif_nit_deliver - deliver received packets to network taps | |
2599 | * @skb: buffer | |
2600 | * | |
2601 | * This function is used to deliver incoming packets to network | |
2602 | * taps. It should be used when the normal netif_receive_skb path | |
2603 | * is bypassed, for example because of VLAN acceleration. | |
2604 | */ | |
2605 | void netif_nit_deliver(struct sk_buff *skb) | |
2606 | { | |
2607 | struct packet_type *ptype; | |
2608 | ||
2609 | if (list_empty(&ptype_all)) | |
2610 | return; | |
2611 | ||
2612 | skb_reset_network_header(skb); | |
2613 | skb_reset_transport_header(skb); | |
2614 | skb->mac_len = skb->network_header - skb->mac_header; | |
2615 | ||
2616 | rcu_read_lock(); | |
2617 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
2618 | if (!ptype->dev || ptype->dev == skb->dev) | |
2619 | deliver_skb(skb, ptype, skb->dev); | |
2620 | } | |
2621 | rcu_read_unlock(); | |
2622 | } | |
2623 | ||
0a9627f2 | 2624 | int __netif_receive_skb(struct sk_buff *skb) |
1da177e4 LT |
2625 | { |
2626 | struct packet_type *ptype, *pt_prev; | |
f2ccd8fa | 2627 | struct net_device *orig_dev; |
0641e4fb | 2628 | struct net_device *master; |
0d7a3681 | 2629 | struct net_device *null_or_orig; |
ca8d9ea3 | 2630 | struct net_device *null_or_bond; |
1da177e4 | 2631 | int ret = NET_RX_DROP; |
252e3346 | 2632 | __be16 type; |
1da177e4 | 2633 | |
81bbb3d4 ED |
2634 | if (!skb->tstamp.tv64) |
2635 | net_timestamp(skb); | |
2636 | ||
05423b24 | 2637 | if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb)) |
9b22ea56 PM |
2638 | return NET_RX_SUCCESS; |
2639 | ||
1da177e4 | 2640 | /* if we've gotten here through NAPI, check netpoll */ |
bea3348e | 2641 | if (netpoll_receive_skb(skb)) |
1da177e4 LT |
2642 | return NET_RX_DROP; |
2643 | ||
8964be4a ED |
2644 | if (!skb->skb_iif) |
2645 | skb->skb_iif = skb->dev->ifindex; | |
86e65da9 | 2646 | |
0d7a3681 | 2647 | null_or_orig = NULL; |
cc9bd5ce | 2648 | orig_dev = skb->dev; |
0641e4fb ED |
2649 | master = ACCESS_ONCE(orig_dev->master); |
2650 | if (master) { | |
2651 | if (skb_bond_should_drop(skb, master)) | |
0d7a3681 JE |
2652 | null_or_orig = orig_dev; /* deliver only exact match */ |
2653 | else | |
0641e4fb | 2654 | skb->dev = master; |
cc9bd5ce | 2655 | } |
8f903c70 | 2656 | |
1da177e4 LT |
2657 | __get_cpu_var(netdev_rx_stat).total++; |
2658 | ||
c1d2bbe1 | 2659 | skb_reset_network_header(skb); |
badff6d0 | 2660 | skb_reset_transport_header(skb); |
b0e380b1 | 2661 | skb->mac_len = skb->network_header - skb->mac_header; |
1da177e4 LT |
2662 | |
2663 | pt_prev = NULL; | |
2664 | ||
2665 | rcu_read_lock(); | |
2666 | ||
2667 | #ifdef CONFIG_NET_CLS_ACT | |
2668 | if (skb->tc_verd & TC_NCLS) { | |
2669 | skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); | |
2670 | goto ncls; | |
2671 | } | |
2672 | #endif | |
2673 | ||
2674 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
f982307f JE |
2675 | if (ptype->dev == null_or_orig || ptype->dev == skb->dev || |
2676 | ptype->dev == orig_dev) { | |
4ec93edb | 2677 | if (pt_prev) |
f2ccd8fa | 2678 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
2679 | pt_prev = ptype; |
2680 | } | |
2681 | } | |
2682 | ||
2683 | #ifdef CONFIG_NET_CLS_ACT | |
f697c3e8 HX |
2684 | skb = handle_ing(skb, &pt_prev, &ret, orig_dev); |
2685 | if (!skb) | |
1da177e4 | 2686 | goto out; |
1da177e4 LT |
2687 | ncls: |
2688 | #endif | |
2689 | ||
6229e362 | 2690 | skb = handle_bridge(skb, &pt_prev, &ret, orig_dev); |
b863ceb7 PM |
2691 | if (!skb) |
2692 | goto out; | |
2693 | skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev); | |
6229e362 | 2694 | if (!skb) |
1da177e4 LT |
2695 | goto out; |
2696 | ||
1f3c8804 AG |
2697 | /* |
2698 | * Make sure frames received on VLAN interfaces stacked on | |
2699 | * bonding interfaces still make their way to any base bonding | |
2700 | * device that may have registered for a specific ptype. The | |
2701 | * handler may have to adjust skb->dev and orig_dev. | |
1f3c8804 | 2702 | */ |
ca8d9ea3 | 2703 | null_or_bond = NULL; |
1f3c8804 AG |
2704 | if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) && |
2705 | (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) { | |
ca8d9ea3 | 2706 | null_or_bond = vlan_dev_real_dev(skb->dev); |
1f3c8804 AG |
2707 | } |
2708 | ||
1da177e4 | 2709 | type = skb->protocol; |
82d8a867 PE |
2710 | list_for_each_entry_rcu(ptype, |
2711 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
1f3c8804 | 2712 | if (ptype->type == type && (ptype->dev == null_or_orig || |
ca8d9ea3 AG |
2713 | ptype->dev == skb->dev || ptype->dev == orig_dev || |
2714 | ptype->dev == null_or_bond)) { | |
4ec93edb | 2715 | if (pt_prev) |
f2ccd8fa | 2716 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
2717 | pt_prev = ptype; |
2718 | } | |
2719 | } | |
2720 | ||
2721 | if (pt_prev) { | |
f2ccd8fa | 2722 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
1da177e4 LT |
2723 | } else { |
2724 | kfree_skb(skb); | |
2725 | /* Jamal, now you will not able to escape explaining | |
2726 | * me how you were going to use this. :-) | |
2727 | */ | |
2728 | ret = NET_RX_DROP; | |
2729 | } | |
2730 | ||
2731 | out: | |
2732 | rcu_read_unlock(); | |
2733 | return ret; | |
2734 | } | |
0a9627f2 TH |
2735 | |
2736 | /** | |
2737 | * netif_receive_skb - process receive buffer from network | |
2738 | * @skb: buffer to process | |
2739 | * | |
2740 | * netif_receive_skb() is the main receive data processing function. | |
2741 | * It always succeeds. The buffer may be dropped during processing | |
2742 | * for congestion control or by the protocol layers. | |
2743 | * | |
2744 | * This function may only be called from softirq context and interrupts | |
2745 | * should be enabled. | |
2746 | * | |
2747 | * Return values (usually ignored): | |
2748 | * NET_RX_SUCCESS: no congestion | |
2749 | * NET_RX_DROP: packet was dropped | |
2750 | */ | |
2751 | int netif_receive_skb(struct sk_buff *skb) | |
2752 | { | |
df334545 | 2753 | #ifdef CONFIG_RPS |
0a9627f2 TH |
2754 | int cpu; |
2755 | ||
2756 | cpu = get_rps_cpu(skb->dev, skb); | |
2757 | ||
2758 | if (cpu < 0) | |
2759 | return __netif_receive_skb(skb); | |
2760 | else | |
2761 | return enqueue_to_backlog(skb, cpu); | |
1e94d72f TH |
2762 | #else |
2763 | return __netif_receive_skb(skb); | |
2764 | #endif | |
0a9627f2 | 2765 | } |
d1b19dff | 2766 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 2767 | |
6e583ce5 | 2768 | /* Network device is going away, flush any packets still pending */ |
e51d739a | 2769 | static void flush_backlog(struct net_device *dev, int cpu) |
6e583ce5 | 2770 | { |
e51d739a | 2771 | struct softnet_data *queue = &per_cpu(softnet_data, cpu); |
6e583ce5 | 2772 | struct sk_buff *skb, *tmp; |
e51d739a | 2773 | unsigned long flags; |
6e583ce5 | 2774 | |
e51d739a | 2775 | spin_lock_irqsave(&queue->input_pkt_queue.lock, flags); |
6e583ce5 SH |
2776 | skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp) |
2777 | if (skb->dev == dev) { | |
2778 | __skb_unlink(skb, &queue->input_pkt_queue); | |
2779 | kfree_skb(skb); | |
2780 | } | |
e51d739a | 2781 | spin_unlock_irqrestore(&queue->input_pkt_queue.lock, flags); |
6e583ce5 SH |
2782 | } |
2783 | ||
d565b0a1 HX |
2784 | static int napi_gro_complete(struct sk_buff *skb) |
2785 | { | |
2786 | struct packet_type *ptype; | |
2787 | __be16 type = skb->protocol; | |
2788 | struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK]; | |
2789 | int err = -ENOENT; | |
2790 | ||
fc59f9a3 HX |
2791 | if (NAPI_GRO_CB(skb)->count == 1) { |
2792 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 2793 | goto out; |
fc59f9a3 | 2794 | } |
d565b0a1 HX |
2795 | |
2796 | rcu_read_lock(); | |
2797 | list_for_each_entry_rcu(ptype, head, list) { | |
2798 | if (ptype->type != type || ptype->dev || !ptype->gro_complete) | |
2799 | continue; | |
2800 | ||
2801 | err = ptype->gro_complete(skb); | |
2802 | break; | |
2803 | } | |
2804 | rcu_read_unlock(); | |
2805 | ||
2806 | if (err) { | |
2807 | WARN_ON(&ptype->list == head); | |
2808 | kfree_skb(skb); | |
2809 | return NET_RX_SUCCESS; | |
2810 | } | |
2811 | ||
2812 | out: | |
d565b0a1 HX |
2813 | return netif_receive_skb(skb); |
2814 | } | |
2815 | ||
11380a4b | 2816 | static void napi_gro_flush(struct napi_struct *napi) |
d565b0a1 HX |
2817 | { |
2818 | struct sk_buff *skb, *next; | |
2819 | ||
2820 | for (skb = napi->gro_list; skb; skb = next) { | |
2821 | next = skb->next; | |
2822 | skb->next = NULL; | |
2823 | napi_gro_complete(skb); | |
2824 | } | |
2825 | ||
4ae5544f | 2826 | napi->gro_count = 0; |
d565b0a1 HX |
2827 | napi->gro_list = NULL; |
2828 | } | |
d565b0a1 | 2829 | |
5b252f0c | 2830 | enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 HX |
2831 | { |
2832 | struct sk_buff **pp = NULL; | |
2833 | struct packet_type *ptype; | |
2834 | __be16 type = skb->protocol; | |
2835 | struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK]; | |
0da2afd5 | 2836 | int same_flow; |
d565b0a1 | 2837 | int mac_len; |
5b252f0c | 2838 | enum gro_result ret; |
d565b0a1 HX |
2839 | |
2840 | if (!(skb->dev->features & NETIF_F_GRO)) | |
2841 | goto normal; | |
2842 | ||
4cf704fb | 2843 | if (skb_is_gso(skb) || skb_has_frags(skb)) |
f17f5c91 HX |
2844 | goto normal; |
2845 | ||
d565b0a1 HX |
2846 | rcu_read_lock(); |
2847 | list_for_each_entry_rcu(ptype, head, list) { | |
d565b0a1 HX |
2848 | if (ptype->type != type || ptype->dev || !ptype->gro_receive) |
2849 | continue; | |
2850 | ||
86911732 | 2851 | skb_set_network_header(skb, skb_gro_offset(skb)); |
d565b0a1 HX |
2852 | mac_len = skb->network_header - skb->mac_header; |
2853 | skb->mac_len = mac_len; | |
2854 | NAPI_GRO_CB(skb)->same_flow = 0; | |
2855 | NAPI_GRO_CB(skb)->flush = 0; | |
5d38a079 | 2856 | NAPI_GRO_CB(skb)->free = 0; |
d565b0a1 | 2857 | |
d565b0a1 HX |
2858 | pp = ptype->gro_receive(&napi->gro_list, skb); |
2859 | break; | |
2860 | } | |
2861 | rcu_read_unlock(); | |
2862 | ||
2863 | if (&ptype->list == head) | |
2864 | goto normal; | |
2865 | ||
0da2afd5 | 2866 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 2867 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 2868 | |
d565b0a1 HX |
2869 | if (pp) { |
2870 | struct sk_buff *nskb = *pp; | |
2871 | ||
2872 | *pp = nskb->next; | |
2873 | nskb->next = NULL; | |
2874 | napi_gro_complete(nskb); | |
4ae5544f | 2875 | napi->gro_count--; |
d565b0a1 HX |
2876 | } |
2877 | ||
0da2afd5 | 2878 | if (same_flow) |
d565b0a1 HX |
2879 | goto ok; |
2880 | ||
4ae5544f | 2881 | if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS) |
d565b0a1 | 2882 | goto normal; |
d565b0a1 | 2883 | |
4ae5544f | 2884 | napi->gro_count++; |
d565b0a1 | 2885 | NAPI_GRO_CB(skb)->count = 1; |
86911732 | 2886 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
d565b0a1 HX |
2887 | skb->next = napi->gro_list; |
2888 | napi->gro_list = skb; | |
5d0d9be8 | 2889 | ret = GRO_HELD; |
d565b0a1 | 2890 | |
ad0f9904 | 2891 | pull: |
cb18978c HX |
2892 | if (skb_headlen(skb) < skb_gro_offset(skb)) { |
2893 | int grow = skb_gro_offset(skb) - skb_headlen(skb); | |
2894 | ||
2895 | BUG_ON(skb->end - skb->tail < grow); | |
2896 | ||
2897 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
2898 | ||
2899 | skb->tail += grow; | |
2900 | skb->data_len -= grow; | |
2901 | ||
2902 | skb_shinfo(skb)->frags[0].page_offset += grow; | |
2903 | skb_shinfo(skb)->frags[0].size -= grow; | |
2904 | ||
2905 | if (unlikely(!skb_shinfo(skb)->frags[0].size)) { | |
2906 | put_page(skb_shinfo(skb)->frags[0].page); | |
2907 | memmove(skb_shinfo(skb)->frags, | |
2908 | skb_shinfo(skb)->frags + 1, | |
2909 | --skb_shinfo(skb)->nr_frags); | |
2910 | } | |
ad0f9904 HX |
2911 | } |
2912 | ||
d565b0a1 | 2913 | ok: |
5d0d9be8 | 2914 | return ret; |
d565b0a1 HX |
2915 | |
2916 | normal: | |
ad0f9904 HX |
2917 | ret = GRO_NORMAL; |
2918 | goto pull; | |
5d38a079 | 2919 | } |
96e93eab HX |
2920 | EXPORT_SYMBOL(dev_gro_receive); |
2921 | ||
5b252f0c BH |
2922 | static gro_result_t |
2923 | __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) | |
96e93eab HX |
2924 | { |
2925 | struct sk_buff *p; | |
2926 | ||
d1c76af9 HX |
2927 | if (netpoll_rx_on(skb)) |
2928 | return GRO_NORMAL; | |
2929 | ||
96e93eab | 2930 | for (p = napi->gro_list; p; p = p->next) { |
f64f9e71 JP |
2931 | NAPI_GRO_CB(p)->same_flow = |
2932 | (p->dev == skb->dev) && | |
2933 | !compare_ether_header(skb_mac_header(p), | |
2934 | skb_gro_mac_header(skb)); | |
96e93eab HX |
2935 | NAPI_GRO_CB(p)->flush = 0; |
2936 | } | |
2937 | ||
2938 | return dev_gro_receive(napi, skb); | |
2939 | } | |
5d38a079 | 2940 | |
c7c4b3b6 | 2941 | gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) |
5d38a079 | 2942 | { |
5d0d9be8 HX |
2943 | switch (ret) { |
2944 | case GRO_NORMAL: | |
c7c4b3b6 BH |
2945 | if (netif_receive_skb(skb)) |
2946 | ret = GRO_DROP; | |
2947 | break; | |
5d38a079 | 2948 | |
5d0d9be8 | 2949 | case GRO_DROP: |
5d0d9be8 | 2950 | case GRO_MERGED_FREE: |
5d38a079 HX |
2951 | kfree_skb(skb); |
2952 | break; | |
5b252f0c BH |
2953 | |
2954 | case GRO_HELD: | |
2955 | case GRO_MERGED: | |
2956 | break; | |
5d38a079 HX |
2957 | } |
2958 | ||
c7c4b3b6 | 2959 | return ret; |
5d0d9be8 HX |
2960 | } |
2961 | EXPORT_SYMBOL(napi_skb_finish); | |
2962 | ||
78a478d0 HX |
2963 | void skb_gro_reset_offset(struct sk_buff *skb) |
2964 | { | |
2965 | NAPI_GRO_CB(skb)->data_offset = 0; | |
2966 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
7489594c | 2967 | NAPI_GRO_CB(skb)->frag0_len = 0; |
78a478d0 | 2968 | |
78d3fd0b | 2969 | if (skb->mac_header == skb->tail && |
7489594c | 2970 | !PageHighMem(skb_shinfo(skb)->frags[0].page)) { |
78a478d0 HX |
2971 | NAPI_GRO_CB(skb)->frag0 = |
2972 | page_address(skb_shinfo(skb)->frags[0].page) + | |
2973 | skb_shinfo(skb)->frags[0].page_offset; | |
7489594c HX |
2974 | NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size; |
2975 | } | |
78a478d0 HX |
2976 | } |
2977 | EXPORT_SYMBOL(skb_gro_reset_offset); | |
2978 | ||
c7c4b3b6 | 2979 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 2980 | { |
86911732 HX |
2981 | skb_gro_reset_offset(skb); |
2982 | ||
5d0d9be8 | 2983 | return napi_skb_finish(__napi_gro_receive(napi, skb), skb); |
d565b0a1 HX |
2984 | } |
2985 | EXPORT_SYMBOL(napi_gro_receive); | |
2986 | ||
96e93eab HX |
2987 | void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
2988 | { | |
96e93eab HX |
2989 | __skb_pull(skb, skb_headlen(skb)); |
2990 | skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb)); | |
2991 | ||
2992 | napi->skb = skb; | |
2993 | } | |
2994 | EXPORT_SYMBOL(napi_reuse_skb); | |
2995 | ||
76620aaf | 2996 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 2997 | { |
5d38a079 | 2998 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
2999 | |
3000 | if (!skb) { | |
89d71a66 ED |
3001 | skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD); |
3002 | if (skb) | |
3003 | napi->skb = skb; | |
80595d59 | 3004 | } |
96e93eab HX |
3005 | return skb; |
3006 | } | |
76620aaf | 3007 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 3008 | |
c7c4b3b6 BH |
3009 | gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb, |
3010 | gro_result_t ret) | |
96e93eab | 3011 | { |
5d0d9be8 HX |
3012 | switch (ret) { |
3013 | case GRO_NORMAL: | |
86911732 | 3014 | case GRO_HELD: |
e76b69cc | 3015 | skb->protocol = eth_type_trans(skb, skb->dev); |
86911732 | 3016 | |
c7c4b3b6 BH |
3017 | if (ret == GRO_HELD) |
3018 | skb_gro_pull(skb, -ETH_HLEN); | |
3019 | else if (netif_receive_skb(skb)) | |
3020 | ret = GRO_DROP; | |
86911732 | 3021 | break; |
5d38a079 | 3022 | |
5d0d9be8 | 3023 | case GRO_DROP: |
5d0d9be8 HX |
3024 | case GRO_MERGED_FREE: |
3025 | napi_reuse_skb(napi, skb); | |
3026 | break; | |
5b252f0c BH |
3027 | |
3028 | case GRO_MERGED: | |
3029 | break; | |
5d0d9be8 | 3030 | } |
5d38a079 | 3031 | |
c7c4b3b6 | 3032 | return ret; |
5d38a079 | 3033 | } |
5d0d9be8 HX |
3034 | EXPORT_SYMBOL(napi_frags_finish); |
3035 | ||
76620aaf HX |
3036 | struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
3037 | { | |
3038 | struct sk_buff *skb = napi->skb; | |
3039 | struct ethhdr *eth; | |
a5b1cf28 HX |
3040 | unsigned int hlen; |
3041 | unsigned int off; | |
76620aaf HX |
3042 | |
3043 | napi->skb = NULL; | |
3044 | ||
3045 | skb_reset_mac_header(skb); | |
3046 | skb_gro_reset_offset(skb); | |
3047 | ||
a5b1cf28 HX |
3048 | off = skb_gro_offset(skb); |
3049 | hlen = off + sizeof(*eth); | |
3050 | eth = skb_gro_header_fast(skb, off); | |
3051 | if (skb_gro_header_hard(skb, hlen)) { | |
3052 | eth = skb_gro_header_slow(skb, hlen, off); | |
3053 | if (unlikely(!eth)) { | |
3054 | napi_reuse_skb(napi, skb); | |
3055 | skb = NULL; | |
3056 | goto out; | |
3057 | } | |
76620aaf HX |
3058 | } |
3059 | ||
3060 | skb_gro_pull(skb, sizeof(*eth)); | |
3061 | ||
3062 | /* | |
3063 | * This works because the only protocols we care about don't require | |
3064 | * special handling. We'll fix it up properly at the end. | |
3065 | */ | |
3066 | skb->protocol = eth->h_proto; | |
3067 | ||
3068 | out: | |
3069 | return skb; | |
3070 | } | |
3071 | EXPORT_SYMBOL(napi_frags_skb); | |
3072 | ||
c7c4b3b6 | 3073 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 3074 | { |
76620aaf | 3075 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 HX |
3076 | |
3077 | if (!skb) | |
c7c4b3b6 | 3078 | return GRO_DROP; |
5d0d9be8 HX |
3079 | |
3080 | return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb)); | |
3081 | } | |
5d38a079 HX |
3082 | EXPORT_SYMBOL(napi_gro_frags); |
3083 | ||
bea3348e | 3084 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 LT |
3085 | { |
3086 | int work = 0; | |
1da177e4 LT |
3087 | struct softnet_data *queue = &__get_cpu_var(softnet_data); |
3088 | unsigned long start_time = jiffies; | |
3089 | ||
bea3348e SH |
3090 | napi->weight = weight_p; |
3091 | do { | |
1da177e4 | 3092 | struct sk_buff *skb; |
1da177e4 | 3093 | |
0a9627f2 | 3094 | spin_lock_irq(&queue->input_pkt_queue.lock); |
1da177e4 | 3095 | skb = __skb_dequeue(&queue->input_pkt_queue); |
bea3348e | 3096 | if (!skb) { |
8f1ead2d | 3097 | __napi_complete(napi); |
0a9627f2 | 3098 | spin_unlock_irq(&queue->input_pkt_queue.lock); |
8f1ead2d | 3099 | break; |
bea3348e | 3100 | } |
0a9627f2 | 3101 | spin_unlock_irq(&queue->input_pkt_queue.lock); |
1da177e4 | 3102 | |
0a9627f2 | 3103 | __netif_receive_skb(skb); |
bea3348e | 3104 | } while (++work < quota && jiffies == start_time); |
1da177e4 | 3105 | |
bea3348e SH |
3106 | return work; |
3107 | } | |
1da177e4 | 3108 | |
bea3348e SH |
3109 | /** |
3110 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 3111 | * @n: entry to schedule |
bea3348e SH |
3112 | * |
3113 | * The entry's receive function will be scheduled to run | |
3114 | */ | |
b5606c2d | 3115 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
3116 | { |
3117 | unsigned long flags; | |
1da177e4 | 3118 | |
bea3348e SH |
3119 | local_irq_save(flags); |
3120 | list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list); | |
3121 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3122 | local_irq_restore(flags); | |
1da177e4 | 3123 | } |
bea3348e SH |
3124 | EXPORT_SYMBOL(__napi_schedule); |
3125 | ||
d565b0a1 HX |
3126 | void __napi_complete(struct napi_struct *n) |
3127 | { | |
3128 | BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); | |
3129 | BUG_ON(n->gro_list); | |
3130 | ||
3131 | list_del(&n->poll_list); | |
3132 | smp_mb__before_clear_bit(); | |
3133 | clear_bit(NAPI_STATE_SCHED, &n->state); | |
3134 | } | |
3135 | EXPORT_SYMBOL(__napi_complete); | |
3136 | ||
3137 | void napi_complete(struct napi_struct *n) | |
3138 | { | |
3139 | unsigned long flags; | |
3140 | ||
3141 | /* | |
3142 | * don't let napi dequeue from the cpu poll list | |
3143 | * just in case its running on a different cpu | |
3144 | */ | |
3145 | if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state))) | |
3146 | return; | |
3147 | ||
3148 | napi_gro_flush(n); | |
3149 | local_irq_save(flags); | |
3150 | __napi_complete(n); | |
3151 | local_irq_restore(flags); | |
3152 | } | |
3153 | EXPORT_SYMBOL(napi_complete); | |
3154 | ||
3155 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, | |
3156 | int (*poll)(struct napi_struct *, int), int weight) | |
3157 | { | |
3158 | INIT_LIST_HEAD(&napi->poll_list); | |
4ae5544f | 3159 | napi->gro_count = 0; |
d565b0a1 | 3160 | napi->gro_list = NULL; |
5d38a079 | 3161 | napi->skb = NULL; |
d565b0a1 HX |
3162 | napi->poll = poll; |
3163 | napi->weight = weight; | |
3164 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 3165 | napi->dev = dev; |
5d38a079 | 3166 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
3167 | spin_lock_init(&napi->poll_lock); |
3168 | napi->poll_owner = -1; | |
3169 | #endif | |
3170 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
3171 | } | |
3172 | EXPORT_SYMBOL(netif_napi_add); | |
3173 | ||
3174 | void netif_napi_del(struct napi_struct *napi) | |
3175 | { | |
3176 | struct sk_buff *skb, *next; | |
3177 | ||
d7b06636 | 3178 | list_del_init(&napi->dev_list); |
76620aaf | 3179 | napi_free_frags(napi); |
d565b0a1 HX |
3180 | |
3181 | for (skb = napi->gro_list; skb; skb = next) { | |
3182 | next = skb->next; | |
3183 | skb->next = NULL; | |
3184 | kfree_skb(skb); | |
3185 | } | |
3186 | ||
3187 | napi->gro_list = NULL; | |
4ae5544f | 3188 | napi->gro_count = 0; |
d565b0a1 HX |
3189 | } |
3190 | EXPORT_SYMBOL(netif_napi_del); | |
3191 | ||
df334545 | 3192 | #ifdef CONFIG_RPS |
0a9627f2 TH |
3193 | /* |
3194 | * net_rps_action sends any pending IPI's for rps. This is only called from | |
3195 | * softirq and interrupts must be enabled. | |
3196 | */ | |
3197 | static void net_rps_action(cpumask_t *mask) | |
3198 | { | |
3199 | int cpu; | |
3200 | ||
3201 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
3202 | for_each_cpu_mask_nr(cpu, *mask) { | |
3203 | struct softnet_data *queue = &per_cpu(softnet_data, cpu); | |
3204 | if (cpu_online(cpu)) | |
3205 | __smp_call_function_single(cpu, &queue->csd, 0); | |
3206 | } | |
3207 | cpus_clear(*mask); | |
3208 | } | |
1e94d72f | 3209 | #endif |
1da177e4 LT |
3210 | |
3211 | static void net_rx_action(struct softirq_action *h) | |
3212 | { | |
bea3348e | 3213 | struct list_head *list = &__get_cpu_var(softnet_data).poll_list; |
24f8b238 | 3214 | unsigned long time_limit = jiffies + 2; |
51b0bded | 3215 | int budget = netdev_budget; |
53fb95d3 | 3216 | void *have; |
df334545 | 3217 | #ifdef CONFIG_RPS |
0a9627f2 TH |
3218 | int select; |
3219 | struct rps_remote_softirq_cpus *rcpus; | |
1e94d72f | 3220 | #endif |
53fb95d3 | 3221 | |
1da177e4 LT |
3222 | local_irq_disable(); |
3223 | ||
bea3348e SH |
3224 | while (!list_empty(list)) { |
3225 | struct napi_struct *n; | |
3226 | int work, weight; | |
1da177e4 | 3227 | |
bea3348e | 3228 | /* If softirq window is exhuasted then punt. |
24f8b238 SH |
3229 | * Allow this to run for 2 jiffies since which will allow |
3230 | * an average latency of 1.5/HZ. | |
bea3348e | 3231 | */ |
24f8b238 | 3232 | if (unlikely(budget <= 0 || time_after(jiffies, time_limit))) |
1da177e4 LT |
3233 | goto softnet_break; |
3234 | ||
3235 | local_irq_enable(); | |
3236 | ||
bea3348e SH |
3237 | /* Even though interrupts have been re-enabled, this |
3238 | * access is safe because interrupts can only add new | |
3239 | * entries to the tail of this list, and only ->poll() | |
3240 | * calls can remove this head entry from the list. | |
3241 | */ | |
e5e26d75 | 3242 | n = list_first_entry(list, struct napi_struct, poll_list); |
1da177e4 | 3243 | |
bea3348e SH |
3244 | have = netpoll_poll_lock(n); |
3245 | ||
3246 | weight = n->weight; | |
3247 | ||
0a7606c1 DM |
3248 | /* This NAPI_STATE_SCHED test is for avoiding a race |
3249 | * with netpoll's poll_napi(). Only the entity which | |
3250 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
3251 | * actually make the ->poll() call. Therefore we avoid | |
3252 | * accidently calling ->poll() when NAPI is not scheduled. | |
3253 | */ | |
3254 | work = 0; | |
4ea7e386 | 3255 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { |
0a7606c1 | 3256 | work = n->poll(n, weight); |
4ea7e386 NH |
3257 | trace_napi_poll(n); |
3258 | } | |
bea3348e SH |
3259 | |
3260 | WARN_ON_ONCE(work > weight); | |
3261 | ||
3262 | budget -= work; | |
3263 | ||
3264 | local_irq_disable(); | |
3265 | ||
3266 | /* Drivers must not modify the NAPI state if they | |
3267 | * consume the entire weight. In such cases this code | |
3268 | * still "owns" the NAPI instance and therefore can | |
3269 | * move the instance around on the list at-will. | |
3270 | */ | |
fed17f30 | 3271 | if (unlikely(work == weight)) { |
ff780cd8 HX |
3272 | if (unlikely(napi_disable_pending(n))) { |
3273 | local_irq_enable(); | |
3274 | napi_complete(n); | |
3275 | local_irq_disable(); | |
3276 | } else | |
fed17f30 DM |
3277 | list_move_tail(&n->poll_list, list); |
3278 | } | |
bea3348e SH |
3279 | |
3280 | netpoll_poll_unlock(have); | |
1da177e4 LT |
3281 | } |
3282 | out: | |
df334545 | 3283 | #ifdef CONFIG_RPS |
0a9627f2 TH |
3284 | rcpus = &__get_cpu_var(rps_remote_softirq_cpus); |
3285 | select = rcpus->select; | |
3286 | rcpus->select ^= 1; | |
3287 | ||
515e06c4 | 3288 | local_irq_enable(); |
bea3348e | 3289 | |
0a9627f2 | 3290 | net_rps_action(&rcpus->mask[select]); |
1e94d72f TH |
3291 | #else |
3292 | local_irq_enable(); | |
3293 | #endif | |
0a9627f2 | 3294 | |
db217334 CL |
3295 | #ifdef CONFIG_NET_DMA |
3296 | /* | |
3297 | * There may not be any more sk_buffs coming right now, so push | |
3298 | * any pending DMA copies to hardware | |
3299 | */ | |
2ba05622 | 3300 | dma_issue_pending_all(); |
db217334 | 3301 | #endif |
bea3348e | 3302 | |
1da177e4 LT |
3303 | return; |
3304 | ||
3305 | softnet_break: | |
3306 | __get_cpu_var(netdev_rx_stat).time_squeeze++; | |
3307 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3308 | goto out; | |
3309 | } | |
3310 | ||
d1b19dff | 3311 | static gifconf_func_t *gifconf_list[NPROTO]; |
1da177e4 LT |
3312 | |
3313 | /** | |
3314 | * register_gifconf - register a SIOCGIF handler | |
3315 | * @family: Address family | |
3316 | * @gifconf: Function handler | |
3317 | * | |
3318 | * Register protocol dependent address dumping routines. The handler | |
3319 | * that is passed must not be freed or reused until it has been replaced | |
3320 | * by another handler. | |
3321 | */ | |
d1b19dff | 3322 | int register_gifconf(unsigned int family, gifconf_func_t *gifconf) |
1da177e4 LT |
3323 | { |
3324 | if (family >= NPROTO) | |
3325 | return -EINVAL; | |
3326 | gifconf_list[family] = gifconf; | |
3327 | return 0; | |
3328 | } | |
d1b19dff | 3329 | EXPORT_SYMBOL(register_gifconf); |
1da177e4 LT |
3330 | |
3331 | ||
3332 | /* | |
3333 | * Map an interface index to its name (SIOCGIFNAME) | |
3334 | */ | |
3335 | ||
3336 | /* | |
3337 | * We need this ioctl for efficient implementation of the | |
3338 | * if_indextoname() function required by the IPv6 API. Without | |
3339 | * it, we would have to search all the interfaces to find a | |
3340 | * match. --pb | |
3341 | */ | |
3342 | ||
881d966b | 3343 | static int dev_ifname(struct net *net, struct ifreq __user *arg) |
1da177e4 LT |
3344 | { |
3345 | struct net_device *dev; | |
3346 | struct ifreq ifr; | |
3347 | ||
3348 | /* | |
3349 | * Fetch the caller's info block. | |
3350 | */ | |
3351 | ||
3352 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
3353 | return -EFAULT; | |
3354 | ||
fb699dfd ED |
3355 | rcu_read_lock(); |
3356 | dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex); | |
1da177e4 | 3357 | if (!dev) { |
fb699dfd | 3358 | rcu_read_unlock(); |
1da177e4 LT |
3359 | return -ENODEV; |
3360 | } | |
3361 | ||
3362 | strcpy(ifr.ifr_name, dev->name); | |
fb699dfd | 3363 | rcu_read_unlock(); |
1da177e4 LT |
3364 | |
3365 | if (copy_to_user(arg, &ifr, sizeof(struct ifreq))) | |
3366 | return -EFAULT; | |
3367 | return 0; | |
3368 | } | |
3369 | ||
3370 | /* | |
3371 | * Perform a SIOCGIFCONF call. This structure will change | |
3372 | * size eventually, and there is nothing I can do about it. | |
3373 | * Thus we will need a 'compatibility mode'. | |
3374 | */ | |
3375 | ||
881d966b | 3376 | static int dev_ifconf(struct net *net, char __user *arg) |
1da177e4 LT |
3377 | { |
3378 | struct ifconf ifc; | |
3379 | struct net_device *dev; | |
3380 | char __user *pos; | |
3381 | int len; | |
3382 | int total; | |
3383 | int i; | |
3384 | ||
3385 | /* | |
3386 | * Fetch the caller's info block. | |
3387 | */ | |
3388 | ||
3389 | if (copy_from_user(&ifc, arg, sizeof(struct ifconf))) | |
3390 | return -EFAULT; | |
3391 | ||
3392 | pos = ifc.ifc_buf; | |
3393 | len = ifc.ifc_len; | |
3394 | ||
3395 | /* | |
3396 | * Loop over the interfaces, and write an info block for each. | |
3397 | */ | |
3398 | ||
3399 | total = 0; | |
881d966b | 3400 | for_each_netdev(net, dev) { |
1da177e4 LT |
3401 | for (i = 0; i < NPROTO; i++) { |
3402 | if (gifconf_list[i]) { | |
3403 | int done; | |
3404 | if (!pos) | |
3405 | done = gifconf_list[i](dev, NULL, 0); | |
3406 | else | |
3407 | done = gifconf_list[i](dev, pos + total, | |
3408 | len - total); | |
3409 | if (done < 0) | |
3410 | return -EFAULT; | |
3411 | total += done; | |
3412 | } | |
3413 | } | |
4ec93edb | 3414 | } |
1da177e4 LT |
3415 | |
3416 | /* | |
3417 | * All done. Write the updated control block back to the caller. | |
3418 | */ | |
3419 | ifc.ifc_len = total; | |
3420 | ||
3421 | /* | |
3422 | * Both BSD and Solaris return 0 here, so we do too. | |
3423 | */ | |
3424 | return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0; | |
3425 | } | |
3426 | ||
3427 | #ifdef CONFIG_PROC_FS | |
3428 | /* | |
3429 | * This is invoked by the /proc filesystem handler to display a device | |
3430 | * in detail. | |
3431 | */ | |
7562f876 | 3432 | void *dev_seq_start(struct seq_file *seq, loff_t *pos) |
c6d14c84 | 3433 | __acquires(RCU) |
1da177e4 | 3434 | { |
e372c414 | 3435 | struct net *net = seq_file_net(seq); |
7562f876 | 3436 | loff_t off; |
1da177e4 | 3437 | struct net_device *dev; |
1da177e4 | 3438 | |
c6d14c84 | 3439 | rcu_read_lock(); |
7562f876 PE |
3440 | if (!*pos) |
3441 | return SEQ_START_TOKEN; | |
1da177e4 | 3442 | |
7562f876 | 3443 | off = 1; |
c6d14c84 | 3444 | for_each_netdev_rcu(net, dev) |
7562f876 PE |
3445 | if (off++ == *pos) |
3446 | return dev; | |
1da177e4 | 3447 | |
7562f876 | 3448 | return NULL; |
1da177e4 LT |
3449 | } |
3450 | ||
3451 | void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
3452 | { | |
c6d14c84 ED |
3453 | struct net_device *dev = (v == SEQ_START_TOKEN) ? |
3454 | first_net_device(seq_file_net(seq)) : | |
3455 | next_net_device((struct net_device *)v); | |
3456 | ||
1da177e4 | 3457 | ++*pos; |
c6d14c84 | 3458 | return rcu_dereference(dev); |
1da177e4 LT |
3459 | } |
3460 | ||
3461 | void dev_seq_stop(struct seq_file *seq, void *v) | |
c6d14c84 | 3462 | __releases(RCU) |
1da177e4 | 3463 | { |
c6d14c84 | 3464 | rcu_read_unlock(); |
1da177e4 LT |
3465 | } |
3466 | ||
3467 | static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev) | |
3468 | { | |
eeda3fd6 | 3469 | const struct net_device_stats *stats = dev_get_stats(dev); |
1da177e4 | 3470 | |
2d13bafe | 3471 | seq_printf(seq, "%6s: %7lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu " |
5a1b5898 RR |
3472 | "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n", |
3473 | dev->name, stats->rx_bytes, stats->rx_packets, | |
3474 | stats->rx_errors, | |
3475 | stats->rx_dropped + stats->rx_missed_errors, | |
3476 | stats->rx_fifo_errors, | |
3477 | stats->rx_length_errors + stats->rx_over_errors + | |
3478 | stats->rx_crc_errors + stats->rx_frame_errors, | |
3479 | stats->rx_compressed, stats->multicast, | |
3480 | stats->tx_bytes, stats->tx_packets, | |
3481 | stats->tx_errors, stats->tx_dropped, | |
3482 | stats->tx_fifo_errors, stats->collisions, | |
3483 | stats->tx_carrier_errors + | |
3484 | stats->tx_aborted_errors + | |
3485 | stats->tx_window_errors + | |
3486 | stats->tx_heartbeat_errors, | |
3487 | stats->tx_compressed); | |
1da177e4 LT |
3488 | } |
3489 | ||
3490 | /* | |
3491 | * Called from the PROCfs module. This now uses the new arbitrary sized | |
3492 | * /proc/net interface to create /proc/net/dev | |
3493 | */ | |
3494 | static int dev_seq_show(struct seq_file *seq, void *v) | |
3495 | { | |
3496 | if (v == SEQ_START_TOKEN) | |
3497 | seq_puts(seq, "Inter-| Receive " | |
3498 | " | Transmit\n" | |
3499 | " face |bytes packets errs drop fifo frame " | |
3500 | "compressed multicast|bytes packets errs " | |
3501 | "drop fifo colls carrier compressed\n"); | |
3502 | else | |
3503 | dev_seq_printf_stats(seq, v); | |
3504 | return 0; | |
3505 | } | |
3506 | ||
3507 | static struct netif_rx_stats *softnet_get_online(loff_t *pos) | |
3508 | { | |
3509 | struct netif_rx_stats *rc = NULL; | |
3510 | ||
0c0b0aca | 3511 | while (*pos < nr_cpu_ids) |
4ec93edb | 3512 | if (cpu_online(*pos)) { |
1da177e4 LT |
3513 | rc = &per_cpu(netdev_rx_stat, *pos); |
3514 | break; | |
3515 | } else | |
3516 | ++*pos; | |
3517 | return rc; | |
3518 | } | |
3519 | ||
3520 | static void *softnet_seq_start(struct seq_file *seq, loff_t *pos) | |
3521 | { | |
3522 | return softnet_get_online(pos); | |
3523 | } | |
3524 | ||
3525 | static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
3526 | { | |
3527 | ++*pos; | |
3528 | return softnet_get_online(pos); | |
3529 | } | |
3530 | ||
3531 | static void softnet_seq_stop(struct seq_file *seq, void *v) | |
3532 | { | |
3533 | } | |
3534 | ||
3535 | static int softnet_seq_show(struct seq_file *seq, void *v) | |
3536 | { | |
3537 | struct netif_rx_stats *s = v; | |
3538 | ||
0a9627f2 | 3539 | seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n", |
31aa02c5 | 3540 | s->total, s->dropped, s->time_squeeze, 0, |
c1ebcdb8 | 3541 | 0, 0, 0, 0, /* was fastroute */ |
0a9627f2 | 3542 | s->cpu_collision, s->received_rps); |
1da177e4 LT |
3543 | return 0; |
3544 | } | |
3545 | ||
f690808e | 3546 | static const struct seq_operations dev_seq_ops = { |
1da177e4 LT |
3547 | .start = dev_seq_start, |
3548 | .next = dev_seq_next, | |
3549 | .stop = dev_seq_stop, | |
3550 | .show = dev_seq_show, | |
3551 | }; | |
3552 | ||
3553 | static int dev_seq_open(struct inode *inode, struct file *file) | |
3554 | { | |
e372c414 DL |
3555 | return seq_open_net(inode, file, &dev_seq_ops, |
3556 | sizeof(struct seq_net_private)); | |
1da177e4 LT |
3557 | } |
3558 | ||
9a32144e | 3559 | static const struct file_operations dev_seq_fops = { |
1da177e4 LT |
3560 | .owner = THIS_MODULE, |
3561 | .open = dev_seq_open, | |
3562 | .read = seq_read, | |
3563 | .llseek = seq_lseek, | |
e372c414 | 3564 | .release = seq_release_net, |
1da177e4 LT |
3565 | }; |
3566 | ||
f690808e | 3567 | static const struct seq_operations softnet_seq_ops = { |
1da177e4 LT |
3568 | .start = softnet_seq_start, |
3569 | .next = softnet_seq_next, | |
3570 | .stop = softnet_seq_stop, | |
3571 | .show = softnet_seq_show, | |
3572 | }; | |
3573 | ||
3574 | static int softnet_seq_open(struct inode *inode, struct file *file) | |
3575 | { | |
3576 | return seq_open(file, &softnet_seq_ops); | |
3577 | } | |
3578 | ||
9a32144e | 3579 | static const struct file_operations softnet_seq_fops = { |
1da177e4 LT |
3580 | .owner = THIS_MODULE, |
3581 | .open = softnet_seq_open, | |
3582 | .read = seq_read, | |
3583 | .llseek = seq_lseek, | |
3584 | .release = seq_release, | |
3585 | }; | |
3586 | ||
0e1256ff SH |
3587 | static void *ptype_get_idx(loff_t pos) |
3588 | { | |
3589 | struct packet_type *pt = NULL; | |
3590 | loff_t i = 0; | |
3591 | int t; | |
3592 | ||
3593 | list_for_each_entry_rcu(pt, &ptype_all, list) { | |
3594 | if (i == pos) | |
3595 | return pt; | |
3596 | ++i; | |
3597 | } | |
3598 | ||
82d8a867 | 3599 | for (t = 0; t < PTYPE_HASH_SIZE; t++) { |
0e1256ff SH |
3600 | list_for_each_entry_rcu(pt, &ptype_base[t], list) { |
3601 | if (i == pos) | |
3602 | return pt; | |
3603 | ++i; | |
3604 | } | |
3605 | } | |
3606 | return NULL; | |
3607 | } | |
3608 | ||
3609 | static void *ptype_seq_start(struct seq_file *seq, loff_t *pos) | |
72348a42 | 3610 | __acquires(RCU) |
0e1256ff SH |
3611 | { |
3612 | rcu_read_lock(); | |
3613 | return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN; | |
3614 | } | |
3615 | ||
3616 | static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
3617 | { | |
3618 | struct packet_type *pt; | |
3619 | struct list_head *nxt; | |
3620 | int hash; | |
3621 | ||
3622 | ++*pos; | |
3623 | if (v == SEQ_START_TOKEN) | |
3624 | return ptype_get_idx(0); | |
3625 | ||
3626 | pt = v; | |
3627 | nxt = pt->list.next; | |
3628 | if (pt->type == htons(ETH_P_ALL)) { | |
3629 | if (nxt != &ptype_all) | |
3630 | goto found; | |
3631 | hash = 0; | |
3632 | nxt = ptype_base[0].next; | |
3633 | } else | |
82d8a867 | 3634 | hash = ntohs(pt->type) & PTYPE_HASH_MASK; |
0e1256ff SH |
3635 | |
3636 | while (nxt == &ptype_base[hash]) { | |
82d8a867 | 3637 | if (++hash >= PTYPE_HASH_SIZE) |
0e1256ff SH |
3638 | return NULL; |
3639 | nxt = ptype_base[hash].next; | |
3640 | } | |
3641 | found: | |
3642 | return list_entry(nxt, struct packet_type, list); | |
3643 | } | |
3644 | ||
3645 | static void ptype_seq_stop(struct seq_file *seq, void *v) | |
72348a42 | 3646 | __releases(RCU) |
0e1256ff SH |
3647 | { |
3648 | rcu_read_unlock(); | |
3649 | } | |
3650 | ||
0e1256ff SH |
3651 | static int ptype_seq_show(struct seq_file *seq, void *v) |
3652 | { | |
3653 | struct packet_type *pt = v; | |
3654 | ||
3655 | if (v == SEQ_START_TOKEN) | |
3656 | seq_puts(seq, "Type Device Function\n"); | |
c346dca1 | 3657 | else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) { |
0e1256ff SH |
3658 | if (pt->type == htons(ETH_P_ALL)) |
3659 | seq_puts(seq, "ALL "); | |
3660 | else | |
3661 | seq_printf(seq, "%04x", ntohs(pt->type)); | |
3662 | ||
908cd2da AD |
3663 | seq_printf(seq, " %-8s %pF\n", |
3664 | pt->dev ? pt->dev->name : "", pt->func); | |
0e1256ff SH |
3665 | } |
3666 | ||
3667 | return 0; | |
3668 | } | |
3669 | ||
3670 | static const struct seq_operations ptype_seq_ops = { | |
3671 | .start = ptype_seq_start, | |
3672 | .next = ptype_seq_next, | |
3673 | .stop = ptype_seq_stop, | |
3674 | .show = ptype_seq_show, | |
3675 | }; | |
3676 | ||
3677 | static int ptype_seq_open(struct inode *inode, struct file *file) | |
3678 | { | |
2feb27db PE |
3679 | return seq_open_net(inode, file, &ptype_seq_ops, |
3680 | sizeof(struct seq_net_private)); | |
0e1256ff SH |
3681 | } |
3682 | ||
3683 | static const struct file_operations ptype_seq_fops = { | |
3684 | .owner = THIS_MODULE, | |
3685 | .open = ptype_seq_open, | |
3686 | .read = seq_read, | |
3687 | .llseek = seq_lseek, | |
2feb27db | 3688 | .release = seq_release_net, |
0e1256ff SH |
3689 | }; |
3690 | ||
3691 | ||
4665079c | 3692 | static int __net_init dev_proc_net_init(struct net *net) |
1da177e4 LT |
3693 | { |
3694 | int rc = -ENOMEM; | |
3695 | ||
881d966b | 3696 | if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops)) |
1da177e4 | 3697 | goto out; |
881d966b | 3698 | if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops)) |
1da177e4 | 3699 | goto out_dev; |
881d966b | 3700 | if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops)) |
457c4cbc | 3701 | goto out_softnet; |
0e1256ff | 3702 | |
881d966b | 3703 | if (wext_proc_init(net)) |
457c4cbc | 3704 | goto out_ptype; |
1da177e4 LT |
3705 | rc = 0; |
3706 | out: | |
3707 | return rc; | |
457c4cbc | 3708 | out_ptype: |
881d966b | 3709 | proc_net_remove(net, "ptype"); |
1da177e4 | 3710 | out_softnet: |
881d966b | 3711 | proc_net_remove(net, "softnet_stat"); |
1da177e4 | 3712 | out_dev: |
881d966b | 3713 | proc_net_remove(net, "dev"); |
1da177e4 LT |
3714 | goto out; |
3715 | } | |
881d966b | 3716 | |
4665079c | 3717 | static void __net_exit dev_proc_net_exit(struct net *net) |
881d966b EB |
3718 | { |
3719 | wext_proc_exit(net); | |
3720 | ||
3721 | proc_net_remove(net, "ptype"); | |
3722 | proc_net_remove(net, "softnet_stat"); | |
3723 | proc_net_remove(net, "dev"); | |
3724 | } | |
3725 | ||
022cbae6 | 3726 | static struct pernet_operations __net_initdata dev_proc_ops = { |
881d966b EB |
3727 | .init = dev_proc_net_init, |
3728 | .exit = dev_proc_net_exit, | |
3729 | }; | |
3730 | ||
3731 | static int __init dev_proc_init(void) | |
3732 | { | |
3733 | return register_pernet_subsys(&dev_proc_ops); | |
3734 | } | |
1da177e4 LT |
3735 | #else |
3736 | #define dev_proc_init() 0 | |
3737 | #endif /* CONFIG_PROC_FS */ | |
3738 | ||
3739 | ||
3740 | /** | |
3741 | * netdev_set_master - set up master/slave pair | |
3742 | * @slave: slave device | |
3743 | * @master: new master device | |
3744 | * | |
3745 | * Changes the master device of the slave. Pass %NULL to break the | |
3746 | * bonding. The caller must hold the RTNL semaphore. On a failure | |
3747 | * a negative errno code is returned. On success the reference counts | |
3748 | * are adjusted, %RTM_NEWLINK is sent to the routing socket and the | |
3749 | * function returns zero. | |
3750 | */ | |
3751 | int netdev_set_master(struct net_device *slave, struct net_device *master) | |
3752 | { | |
3753 | struct net_device *old = slave->master; | |
3754 | ||
3755 | ASSERT_RTNL(); | |
3756 | ||
3757 | if (master) { | |
3758 | if (old) | |
3759 | return -EBUSY; | |
3760 | dev_hold(master); | |
3761 | } | |
3762 | ||
3763 | slave->master = master; | |
4ec93edb | 3764 | |
283f2fe8 ED |
3765 | if (old) { |
3766 | synchronize_net(); | |
1da177e4 | 3767 | dev_put(old); |
283f2fe8 | 3768 | } |
1da177e4 LT |
3769 | if (master) |
3770 | slave->flags |= IFF_SLAVE; | |
3771 | else | |
3772 | slave->flags &= ~IFF_SLAVE; | |
3773 | ||
3774 | rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE); | |
3775 | return 0; | |
3776 | } | |
d1b19dff | 3777 | EXPORT_SYMBOL(netdev_set_master); |
1da177e4 | 3778 | |
b6c40d68 PM |
3779 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
3780 | { | |
d314774c SH |
3781 | const struct net_device_ops *ops = dev->netdev_ops; |
3782 | ||
3783 | if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags) | |
3784 | ops->ndo_change_rx_flags(dev, flags); | |
b6c40d68 PM |
3785 | } |
3786 | ||
dad9b335 | 3787 | static int __dev_set_promiscuity(struct net_device *dev, int inc) |
1da177e4 LT |
3788 | { |
3789 | unsigned short old_flags = dev->flags; | |
8192b0c4 DH |
3790 | uid_t uid; |
3791 | gid_t gid; | |
1da177e4 | 3792 | |
24023451 PM |
3793 | ASSERT_RTNL(); |
3794 | ||
dad9b335 WC |
3795 | dev->flags |= IFF_PROMISC; |
3796 | dev->promiscuity += inc; | |
3797 | if (dev->promiscuity == 0) { | |
3798 | /* | |
3799 | * Avoid overflow. | |
3800 | * If inc causes overflow, untouch promisc and return error. | |
3801 | */ | |
3802 | if (inc < 0) | |
3803 | dev->flags &= ~IFF_PROMISC; | |
3804 | else { | |
3805 | dev->promiscuity -= inc; | |
3806 | printk(KERN_WARNING "%s: promiscuity touches roof, " | |
3807 | "set promiscuity failed, promiscuity feature " | |
3808 | "of device might be broken.\n", dev->name); | |
3809 | return -EOVERFLOW; | |
3810 | } | |
3811 | } | |
52609c0b | 3812 | if (dev->flags != old_flags) { |
1da177e4 LT |
3813 | printk(KERN_INFO "device %s %s promiscuous mode\n", |
3814 | dev->name, (dev->flags & IFF_PROMISC) ? "entered" : | |
4ec93edb | 3815 | "left"); |
8192b0c4 DH |
3816 | if (audit_enabled) { |
3817 | current_uid_gid(&uid, &gid); | |
7759db82 KHK |
3818 | audit_log(current->audit_context, GFP_ATOMIC, |
3819 | AUDIT_ANOM_PROMISCUOUS, | |
3820 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
3821 | dev->name, (dev->flags & IFF_PROMISC), | |
3822 | (old_flags & IFF_PROMISC), | |
3823 | audit_get_loginuid(current), | |
8192b0c4 | 3824 | uid, gid, |
7759db82 | 3825 | audit_get_sessionid(current)); |
8192b0c4 | 3826 | } |
24023451 | 3827 | |
b6c40d68 | 3828 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 3829 | } |
dad9b335 | 3830 | return 0; |
1da177e4 LT |
3831 | } |
3832 | ||
4417da66 PM |
3833 | /** |
3834 | * dev_set_promiscuity - update promiscuity count on a device | |
3835 | * @dev: device | |
3836 | * @inc: modifier | |
3837 | * | |
3838 | * Add or remove promiscuity from a device. While the count in the device | |
3839 | * remains above zero the interface remains promiscuous. Once it hits zero | |
3840 | * the device reverts back to normal filtering operation. A negative inc | |
3841 | * value is used to drop promiscuity on the device. | |
dad9b335 | 3842 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 3843 | */ |
dad9b335 | 3844 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 PM |
3845 | { |
3846 | unsigned short old_flags = dev->flags; | |
dad9b335 | 3847 | int err; |
4417da66 | 3848 | |
dad9b335 | 3849 | err = __dev_set_promiscuity(dev, inc); |
4b5a698e | 3850 | if (err < 0) |
dad9b335 | 3851 | return err; |
4417da66 PM |
3852 | if (dev->flags != old_flags) |
3853 | dev_set_rx_mode(dev); | |
dad9b335 | 3854 | return err; |
4417da66 | 3855 | } |
d1b19dff | 3856 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 3857 | |
1da177e4 LT |
3858 | /** |
3859 | * dev_set_allmulti - update allmulti count on a device | |
3860 | * @dev: device | |
3861 | * @inc: modifier | |
3862 | * | |
3863 | * Add or remove reception of all multicast frames to a device. While the | |
3864 | * count in the device remains above zero the interface remains listening | |
3865 | * to all interfaces. Once it hits zero the device reverts back to normal | |
3866 | * filtering operation. A negative @inc value is used to drop the counter | |
3867 | * when releasing a resource needing all multicasts. | |
dad9b335 | 3868 | * Return 0 if successful or a negative errno code on error. |
1da177e4 LT |
3869 | */ |
3870 | ||
dad9b335 | 3871 | int dev_set_allmulti(struct net_device *dev, int inc) |
1da177e4 LT |
3872 | { |
3873 | unsigned short old_flags = dev->flags; | |
3874 | ||
24023451 PM |
3875 | ASSERT_RTNL(); |
3876 | ||
1da177e4 | 3877 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
3878 | dev->allmulti += inc; |
3879 | if (dev->allmulti == 0) { | |
3880 | /* | |
3881 | * Avoid overflow. | |
3882 | * If inc causes overflow, untouch allmulti and return error. | |
3883 | */ | |
3884 | if (inc < 0) | |
3885 | dev->flags &= ~IFF_ALLMULTI; | |
3886 | else { | |
3887 | dev->allmulti -= inc; | |
3888 | printk(KERN_WARNING "%s: allmulti touches roof, " | |
3889 | "set allmulti failed, allmulti feature of " | |
3890 | "device might be broken.\n", dev->name); | |
3891 | return -EOVERFLOW; | |
3892 | } | |
3893 | } | |
24023451 | 3894 | if (dev->flags ^ old_flags) { |
b6c40d68 | 3895 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 3896 | dev_set_rx_mode(dev); |
24023451 | 3897 | } |
dad9b335 | 3898 | return 0; |
4417da66 | 3899 | } |
d1b19dff | 3900 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
3901 | |
3902 | /* | |
3903 | * Upload unicast and multicast address lists to device and | |
3904 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 3905 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
3906 | * are present. |
3907 | */ | |
3908 | void __dev_set_rx_mode(struct net_device *dev) | |
3909 | { | |
d314774c SH |
3910 | const struct net_device_ops *ops = dev->netdev_ops; |
3911 | ||
4417da66 PM |
3912 | /* dev_open will call this function so the list will stay sane. */ |
3913 | if (!(dev->flags&IFF_UP)) | |
3914 | return; | |
3915 | ||
3916 | if (!netif_device_present(dev)) | |
40b77c94 | 3917 | return; |
4417da66 | 3918 | |
d314774c SH |
3919 | if (ops->ndo_set_rx_mode) |
3920 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
3921 | else { |
3922 | /* Unicast addresses changes may only happen under the rtnl, | |
3923 | * therefore calling __dev_set_promiscuity here is safe. | |
3924 | */ | |
32e7bfc4 | 3925 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
4417da66 PM |
3926 | __dev_set_promiscuity(dev, 1); |
3927 | dev->uc_promisc = 1; | |
32e7bfc4 | 3928 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
4417da66 PM |
3929 | __dev_set_promiscuity(dev, -1); |
3930 | dev->uc_promisc = 0; | |
3931 | } | |
3932 | ||
d314774c SH |
3933 | if (ops->ndo_set_multicast_list) |
3934 | ops->ndo_set_multicast_list(dev); | |
4417da66 PM |
3935 | } |
3936 | } | |
3937 | ||
3938 | void dev_set_rx_mode(struct net_device *dev) | |
3939 | { | |
b9e40857 | 3940 | netif_addr_lock_bh(dev); |
4417da66 | 3941 | __dev_set_rx_mode(dev); |
b9e40857 | 3942 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
3943 | } |
3944 | ||
f001fde5 JP |
3945 | /* hw addresses list handling functions */ |
3946 | ||
31278e71 JP |
3947 | static int __hw_addr_add(struct netdev_hw_addr_list *list, unsigned char *addr, |
3948 | int addr_len, unsigned char addr_type) | |
f001fde5 JP |
3949 | { |
3950 | struct netdev_hw_addr *ha; | |
3951 | int alloc_size; | |
3952 | ||
3953 | if (addr_len > MAX_ADDR_LEN) | |
3954 | return -EINVAL; | |
3955 | ||
31278e71 | 3956 | list_for_each_entry(ha, &list->list, list) { |
ccffad25 JP |
3957 | if (!memcmp(ha->addr, addr, addr_len) && |
3958 | ha->type == addr_type) { | |
3959 | ha->refcount++; | |
3960 | return 0; | |
3961 | } | |
3962 | } | |
3963 | ||
3964 | ||
f001fde5 JP |
3965 | alloc_size = sizeof(*ha); |
3966 | if (alloc_size < L1_CACHE_BYTES) | |
3967 | alloc_size = L1_CACHE_BYTES; | |
3968 | ha = kmalloc(alloc_size, GFP_ATOMIC); | |
3969 | if (!ha) | |
3970 | return -ENOMEM; | |
3971 | memcpy(ha->addr, addr, addr_len); | |
3972 | ha->type = addr_type; | |
ccffad25 JP |
3973 | ha->refcount = 1; |
3974 | ha->synced = false; | |
31278e71 JP |
3975 | list_add_tail_rcu(&ha->list, &list->list); |
3976 | list->count++; | |
f001fde5 JP |
3977 | return 0; |
3978 | } | |
3979 | ||
3980 | static void ha_rcu_free(struct rcu_head *head) | |
3981 | { | |
3982 | struct netdev_hw_addr *ha; | |
3983 | ||
3984 | ha = container_of(head, struct netdev_hw_addr, rcu_head); | |
3985 | kfree(ha); | |
3986 | } | |
3987 | ||
31278e71 JP |
3988 | static int __hw_addr_del(struct netdev_hw_addr_list *list, unsigned char *addr, |
3989 | int addr_len, unsigned char addr_type) | |
f001fde5 JP |
3990 | { |
3991 | struct netdev_hw_addr *ha; | |
f001fde5 | 3992 | |
31278e71 | 3993 | list_for_each_entry(ha, &list->list, list) { |
ccffad25 | 3994 | if (!memcmp(ha->addr, addr, addr_len) && |
f001fde5 | 3995 | (ha->type == addr_type || !addr_type)) { |
ccffad25 JP |
3996 | if (--ha->refcount) |
3997 | return 0; | |
f001fde5 JP |
3998 | list_del_rcu(&ha->list); |
3999 | call_rcu(&ha->rcu_head, ha_rcu_free); | |
31278e71 | 4000 | list->count--; |
f001fde5 JP |
4001 | return 0; |
4002 | } | |
4003 | } | |
4004 | return -ENOENT; | |
4005 | } | |
4006 | ||
31278e71 JP |
4007 | static int __hw_addr_add_multiple(struct netdev_hw_addr_list *to_list, |
4008 | struct netdev_hw_addr_list *from_list, | |
4009 | int addr_len, | |
ccffad25 | 4010 | unsigned char addr_type) |
f001fde5 JP |
4011 | { |
4012 | int err; | |
4013 | struct netdev_hw_addr *ha, *ha2; | |
4014 | unsigned char type; | |
4015 | ||
31278e71 | 4016 | list_for_each_entry(ha, &from_list->list, list) { |
f001fde5 | 4017 | type = addr_type ? addr_type : ha->type; |
31278e71 | 4018 | err = __hw_addr_add(to_list, ha->addr, addr_len, type); |
f001fde5 JP |
4019 | if (err) |
4020 | goto unroll; | |
4021 | } | |
4022 | return 0; | |
4023 | ||
4024 | unroll: | |
31278e71 | 4025 | list_for_each_entry(ha2, &from_list->list, list) { |
f001fde5 JP |
4026 | if (ha2 == ha) |
4027 | break; | |
4028 | type = addr_type ? addr_type : ha2->type; | |
31278e71 | 4029 | __hw_addr_del(to_list, ha2->addr, addr_len, type); |
f001fde5 JP |
4030 | } |
4031 | return err; | |
4032 | } | |
4033 | ||
31278e71 JP |
4034 | static void __hw_addr_del_multiple(struct netdev_hw_addr_list *to_list, |
4035 | struct netdev_hw_addr_list *from_list, | |
4036 | int addr_len, | |
ccffad25 | 4037 | unsigned char addr_type) |
f001fde5 JP |
4038 | { |
4039 | struct netdev_hw_addr *ha; | |
4040 | unsigned char type; | |
4041 | ||
31278e71 | 4042 | list_for_each_entry(ha, &from_list->list, list) { |
f001fde5 | 4043 | type = addr_type ? addr_type : ha->type; |
31278e71 | 4044 | __hw_addr_del(to_list, ha->addr, addr_len, addr_type); |
ccffad25 JP |
4045 | } |
4046 | } | |
4047 | ||
31278e71 JP |
4048 | static int __hw_addr_sync(struct netdev_hw_addr_list *to_list, |
4049 | struct netdev_hw_addr_list *from_list, | |
ccffad25 JP |
4050 | int addr_len) |
4051 | { | |
4052 | int err = 0; | |
4053 | struct netdev_hw_addr *ha, *tmp; | |
4054 | ||
31278e71 | 4055 | list_for_each_entry_safe(ha, tmp, &from_list->list, list) { |
ccffad25 | 4056 | if (!ha->synced) { |
31278e71 | 4057 | err = __hw_addr_add(to_list, ha->addr, |
ccffad25 JP |
4058 | addr_len, ha->type); |
4059 | if (err) | |
4060 | break; | |
4061 | ha->synced = true; | |
4062 | ha->refcount++; | |
4063 | } else if (ha->refcount == 1) { | |
31278e71 JP |
4064 | __hw_addr_del(to_list, ha->addr, addr_len, ha->type); |
4065 | __hw_addr_del(from_list, ha->addr, addr_len, ha->type); | |
ccffad25 | 4066 | } |
f001fde5 | 4067 | } |
ccffad25 | 4068 | return err; |
f001fde5 JP |
4069 | } |
4070 | ||
31278e71 JP |
4071 | static void __hw_addr_unsync(struct netdev_hw_addr_list *to_list, |
4072 | struct netdev_hw_addr_list *from_list, | |
ccffad25 JP |
4073 | int addr_len) |
4074 | { | |
4075 | struct netdev_hw_addr *ha, *tmp; | |
4076 | ||
31278e71 | 4077 | list_for_each_entry_safe(ha, tmp, &from_list->list, list) { |
ccffad25 | 4078 | if (ha->synced) { |
31278e71 | 4079 | __hw_addr_del(to_list, ha->addr, |
ccffad25 JP |
4080 | addr_len, ha->type); |
4081 | ha->synced = false; | |
31278e71 | 4082 | __hw_addr_del(from_list, ha->addr, |
ccffad25 JP |
4083 | addr_len, ha->type); |
4084 | } | |
4085 | } | |
4086 | } | |
4087 | ||
31278e71 | 4088 | static void __hw_addr_flush(struct netdev_hw_addr_list *list) |
f001fde5 JP |
4089 | { |
4090 | struct netdev_hw_addr *ha, *tmp; | |
4091 | ||
31278e71 | 4092 | list_for_each_entry_safe(ha, tmp, &list->list, list) { |
f001fde5 JP |
4093 | list_del_rcu(&ha->list); |
4094 | call_rcu(&ha->rcu_head, ha_rcu_free); | |
4095 | } | |
31278e71 JP |
4096 | list->count = 0; |
4097 | } | |
4098 | ||
4099 | static void __hw_addr_init(struct netdev_hw_addr_list *list) | |
4100 | { | |
4101 | INIT_LIST_HEAD(&list->list); | |
4102 | list->count = 0; | |
f001fde5 JP |
4103 | } |
4104 | ||
4105 | /* Device addresses handling functions */ | |
4106 | ||
4107 | static void dev_addr_flush(struct net_device *dev) | |
4108 | { | |
4109 | /* rtnl_mutex must be held here */ | |
4110 | ||
31278e71 | 4111 | __hw_addr_flush(&dev->dev_addrs); |
f001fde5 JP |
4112 | dev->dev_addr = NULL; |
4113 | } | |
4114 | ||
4115 | static int dev_addr_init(struct net_device *dev) | |
4116 | { | |
4117 | unsigned char addr[MAX_ADDR_LEN]; | |
4118 | struct netdev_hw_addr *ha; | |
4119 | int err; | |
4120 | ||
4121 | /* rtnl_mutex must be held here */ | |
4122 | ||
31278e71 | 4123 | __hw_addr_init(&dev->dev_addrs); |
0c27922e | 4124 | memset(addr, 0, sizeof(addr)); |
31278e71 | 4125 | err = __hw_addr_add(&dev->dev_addrs, addr, sizeof(addr), |
f001fde5 JP |
4126 | NETDEV_HW_ADDR_T_LAN); |
4127 | if (!err) { | |
4128 | /* | |
4129 | * Get the first (previously created) address from the list | |
4130 | * and set dev_addr pointer to this location. | |
4131 | */ | |
31278e71 | 4132 | ha = list_first_entry(&dev->dev_addrs.list, |
f001fde5 JP |
4133 | struct netdev_hw_addr, list); |
4134 | dev->dev_addr = ha->addr; | |
4135 | } | |
4136 | return err; | |
4137 | } | |
4138 | ||
4139 | /** | |
4140 | * dev_addr_add - Add a device address | |
4141 | * @dev: device | |
4142 | * @addr: address to add | |
4143 | * @addr_type: address type | |
4144 | * | |
4145 | * Add a device address to the device or increase the reference count if | |
4146 | * it already exists. | |
4147 | * | |
4148 | * The caller must hold the rtnl_mutex. | |
4149 | */ | |
4150 | int dev_addr_add(struct net_device *dev, unsigned char *addr, | |
4151 | unsigned char addr_type) | |
4152 | { | |
4153 | int err; | |
4154 | ||
4155 | ASSERT_RTNL(); | |
4156 | ||
31278e71 | 4157 | err = __hw_addr_add(&dev->dev_addrs, addr, dev->addr_len, addr_type); |
f001fde5 JP |
4158 | if (!err) |
4159 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); | |
4160 | return err; | |
4161 | } | |
4162 | EXPORT_SYMBOL(dev_addr_add); | |
4163 | ||
4164 | /** | |
4165 | * dev_addr_del - Release a device address. | |
4166 | * @dev: device | |
4167 | * @addr: address to delete | |
4168 | * @addr_type: address type | |
4169 | * | |
4170 | * Release reference to a device address and remove it from the device | |
4171 | * if the reference count drops to zero. | |
4172 | * | |
4173 | * The caller must hold the rtnl_mutex. | |
4174 | */ | |
4175 | int dev_addr_del(struct net_device *dev, unsigned char *addr, | |
4176 | unsigned char addr_type) | |
4177 | { | |
4178 | int err; | |
ccffad25 | 4179 | struct netdev_hw_addr *ha; |
f001fde5 JP |
4180 | |
4181 | ASSERT_RTNL(); | |
4182 | ||
ccffad25 JP |
4183 | /* |
4184 | * We can not remove the first address from the list because | |
4185 | * dev->dev_addr points to that. | |
4186 | */ | |
31278e71 JP |
4187 | ha = list_first_entry(&dev->dev_addrs.list, |
4188 | struct netdev_hw_addr, list); | |
ccffad25 JP |
4189 | if (ha->addr == dev->dev_addr && ha->refcount == 1) |
4190 | return -ENOENT; | |
4191 | ||
31278e71 | 4192 | err = __hw_addr_del(&dev->dev_addrs, addr, dev->addr_len, |
ccffad25 | 4193 | addr_type); |
f001fde5 JP |
4194 | if (!err) |
4195 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); | |
4196 | return err; | |
4197 | } | |
4198 | EXPORT_SYMBOL(dev_addr_del); | |
4199 | ||
4200 | /** | |
4201 | * dev_addr_add_multiple - Add device addresses from another device | |
4202 | * @to_dev: device to which addresses will be added | |
4203 | * @from_dev: device from which addresses will be added | |
4204 | * @addr_type: address type - 0 means type will be used from from_dev | |
4205 | * | |
4206 | * Add device addresses of the one device to another. | |
4207 | ** | |
4208 | * The caller must hold the rtnl_mutex. | |
4209 | */ | |
4210 | int dev_addr_add_multiple(struct net_device *to_dev, | |
4211 | struct net_device *from_dev, | |
4212 | unsigned char addr_type) | |
4213 | { | |
4214 | int err; | |
4215 | ||
4216 | ASSERT_RTNL(); | |
4217 | ||
4218 | if (from_dev->addr_len != to_dev->addr_len) | |
4219 | return -EINVAL; | |
31278e71 | 4220 | err = __hw_addr_add_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs, |
ccffad25 | 4221 | to_dev->addr_len, addr_type); |
f001fde5 JP |
4222 | if (!err) |
4223 | call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev); | |
4224 | return err; | |
4225 | } | |
4226 | EXPORT_SYMBOL(dev_addr_add_multiple); | |
4227 | ||
4228 | /** | |
4229 | * dev_addr_del_multiple - Delete device addresses by another device | |
4230 | * @to_dev: device where the addresses will be deleted | |
4231 | * @from_dev: device by which addresses the addresses will be deleted | |
4232 | * @addr_type: address type - 0 means type will used from from_dev | |
4233 | * | |
4234 | * Deletes addresses in to device by the list of addresses in from device. | |
4235 | * | |
4236 | * The caller must hold the rtnl_mutex. | |
4237 | */ | |
4238 | int dev_addr_del_multiple(struct net_device *to_dev, | |
4239 | struct net_device *from_dev, | |
4240 | unsigned char addr_type) | |
4241 | { | |
4242 | ASSERT_RTNL(); | |
4243 | ||
4244 | if (from_dev->addr_len != to_dev->addr_len) | |
4245 | return -EINVAL; | |
31278e71 | 4246 | __hw_addr_del_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs, |
ccffad25 | 4247 | to_dev->addr_len, addr_type); |
f001fde5 JP |
4248 | call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev); |
4249 | return 0; | |
4250 | } | |
4251 | EXPORT_SYMBOL(dev_addr_del_multiple); | |
4252 | ||
31278e71 | 4253 | /* multicast addresses handling functions */ |
f001fde5 | 4254 | |
61cbc2fc PM |
4255 | int __dev_addr_delete(struct dev_addr_list **list, int *count, |
4256 | void *addr, int alen, int glbl) | |
bf742482 PM |
4257 | { |
4258 | struct dev_addr_list *da; | |
4259 | ||
4260 | for (; (da = *list) != NULL; list = &da->next) { | |
4261 | if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 && | |
4262 | alen == da->da_addrlen) { | |
4263 | if (glbl) { | |
4264 | int old_glbl = da->da_gusers; | |
4265 | da->da_gusers = 0; | |
4266 | if (old_glbl == 0) | |
4267 | break; | |
4268 | } | |
4269 | if (--da->da_users) | |
4270 | return 0; | |
4271 | ||
4272 | *list = da->next; | |
4273 | kfree(da); | |
61cbc2fc | 4274 | (*count)--; |
bf742482 PM |
4275 | return 0; |
4276 | } | |
4277 | } | |
4278 | return -ENOENT; | |
4279 | } | |
4280 | ||
61cbc2fc PM |
4281 | int __dev_addr_add(struct dev_addr_list **list, int *count, |
4282 | void *addr, int alen, int glbl) | |
bf742482 PM |
4283 | { |
4284 | struct dev_addr_list *da; | |
4285 | ||
4286 | for (da = *list; da != NULL; da = da->next) { | |
4287 | if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 && | |
4288 | da->da_addrlen == alen) { | |
4289 | if (glbl) { | |
4290 | int old_glbl = da->da_gusers; | |
4291 | da->da_gusers = 1; | |
4292 | if (old_glbl) | |
4293 | return 0; | |
4294 | } | |
4295 | da->da_users++; | |
4296 | return 0; | |
4297 | } | |
4298 | } | |
4299 | ||
12aa343a | 4300 | da = kzalloc(sizeof(*da), GFP_ATOMIC); |
bf742482 PM |
4301 | if (da == NULL) |
4302 | return -ENOMEM; | |
4303 | memcpy(da->da_addr, addr, alen); | |
4304 | da->da_addrlen = alen; | |
4305 | da->da_users = 1; | |
4306 | da->da_gusers = glbl ? 1 : 0; | |
4307 | da->next = *list; | |
4308 | *list = da; | |
61cbc2fc | 4309 | (*count)++; |
bf742482 PM |
4310 | return 0; |
4311 | } | |
4312 | ||
4417da66 PM |
4313 | /** |
4314 | * dev_unicast_delete - Release secondary unicast address. | |
4315 | * @dev: device | |
0ed72ec4 | 4316 | * @addr: address to delete |
4417da66 PM |
4317 | * |
4318 | * Release reference to a secondary unicast address and remove it | |
0ed72ec4 | 4319 | * from the device if the reference count drops to zero. |
4417da66 PM |
4320 | * |
4321 | * The caller must hold the rtnl_mutex. | |
4322 | */ | |
ccffad25 | 4323 | int dev_unicast_delete(struct net_device *dev, void *addr) |
4417da66 PM |
4324 | { |
4325 | int err; | |
4326 | ||
4327 | ASSERT_RTNL(); | |
4328 | ||
a6ac65db | 4329 | netif_addr_lock_bh(dev); |
31278e71 JP |
4330 | err = __hw_addr_del(&dev->uc, addr, dev->addr_len, |
4331 | NETDEV_HW_ADDR_T_UNICAST); | |
61cbc2fc | 4332 | if (!err) |
4417da66 | 4333 | __dev_set_rx_mode(dev); |
a6ac65db | 4334 | netif_addr_unlock_bh(dev); |
4417da66 PM |
4335 | return err; |
4336 | } | |
4337 | EXPORT_SYMBOL(dev_unicast_delete); | |
4338 | ||
4339 | /** | |
4340 | * dev_unicast_add - add a secondary unicast address | |
4341 | * @dev: device | |
5dbaec5d | 4342 | * @addr: address to add |
4417da66 PM |
4343 | * |
4344 | * Add a secondary unicast address to the device or increase | |
4345 | * the reference count if it already exists. | |
4346 | * | |
4347 | * The caller must hold the rtnl_mutex. | |
4348 | */ | |
ccffad25 | 4349 | int dev_unicast_add(struct net_device *dev, void *addr) |
4417da66 PM |
4350 | { |
4351 | int err; | |
4352 | ||
4353 | ASSERT_RTNL(); | |
4354 | ||
a6ac65db | 4355 | netif_addr_lock_bh(dev); |
31278e71 JP |
4356 | err = __hw_addr_add(&dev->uc, addr, dev->addr_len, |
4357 | NETDEV_HW_ADDR_T_UNICAST); | |
61cbc2fc | 4358 | if (!err) |
4417da66 | 4359 | __dev_set_rx_mode(dev); |
a6ac65db | 4360 | netif_addr_unlock_bh(dev); |
4417da66 PM |
4361 | return err; |
4362 | } | |
4363 | EXPORT_SYMBOL(dev_unicast_add); | |
4364 | ||
e83a2ea8 CL |
4365 | int __dev_addr_sync(struct dev_addr_list **to, int *to_count, |
4366 | struct dev_addr_list **from, int *from_count) | |
4367 | { | |
4368 | struct dev_addr_list *da, *next; | |
4369 | int err = 0; | |
4370 | ||
4371 | da = *from; | |
4372 | while (da != NULL) { | |
4373 | next = da->next; | |
4374 | if (!da->da_synced) { | |
4375 | err = __dev_addr_add(to, to_count, | |
4376 | da->da_addr, da->da_addrlen, 0); | |
4377 | if (err < 0) | |
4378 | break; | |
4379 | da->da_synced = 1; | |
4380 | da->da_users++; | |
4381 | } else if (da->da_users == 1) { | |
4382 | __dev_addr_delete(to, to_count, | |
4383 | da->da_addr, da->da_addrlen, 0); | |
4384 | __dev_addr_delete(from, from_count, | |
4385 | da->da_addr, da->da_addrlen, 0); | |
4386 | } | |
4387 | da = next; | |
4388 | } | |
4389 | return err; | |
4390 | } | |
c4029083 | 4391 | EXPORT_SYMBOL_GPL(__dev_addr_sync); |
e83a2ea8 CL |
4392 | |
4393 | void __dev_addr_unsync(struct dev_addr_list **to, int *to_count, | |
4394 | struct dev_addr_list **from, int *from_count) | |
4395 | { | |
4396 | struct dev_addr_list *da, *next; | |
4397 | ||
4398 | da = *from; | |
4399 | while (da != NULL) { | |
4400 | next = da->next; | |
4401 | if (da->da_synced) { | |
4402 | __dev_addr_delete(to, to_count, | |
4403 | da->da_addr, da->da_addrlen, 0); | |
4404 | da->da_synced = 0; | |
4405 | __dev_addr_delete(from, from_count, | |
4406 | da->da_addr, da->da_addrlen, 0); | |
4407 | } | |
4408 | da = next; | |
4409 | } | |
4410 | } | |
c4029083 | 4411 | EXPORT_SYMBOL_GPL(__dev_addr_unsync); |
e83a2ea8 CL |
4412 | |
4413 | /** | |
4414 | * dev_unicast_sync - Synchronize device's unicast list to another device | |
4415 | * @to: destination device | |
4416 | * @from: source device | |
4417 | * | |
4418 | * Add newly added addresses to the destination device and release | |
a6ac65db JP |
4419 | * addresses that have no users left. The source device must be |
4420 | * locked by netif_tx_lock_bh. | |
e83a2ea8 CL |
4421 | * |
4422 | * This function is intended to be called from the dev->set_rx_mode | |
4423 | * function of layered software devices. | |
4424 | */ | |
4425 | int dev_unicast_sync(struct net_device *to, struct net_device *from) | |
4426 | { | |
4427 | int err = 0; | |
4428 | ||
ccffad25 JP |
4429 | if (to->addr_len != from->addr_len) |
4430 | return -EINVAL; | |
4431 | ||
a6ac65db | 4432 | netif_addr_lock_bh(to); |
31278e71 | 4433 | err = __hw_addr_sync(&to->uc, &from->uc, to->addr_len); |
e83a2ea8 CL |
4434 | if (!err) |
4435 | __dev_set_rx_mode(to); | |
a6ac65db | 4436 | netif_addr_unlock_bh(to); |
e83a2ea8 CL |
4437 | return err; |
4438 | } | |
4439 | EXPORT_SYMBOL(dev_unicast_sync); | |
4440 | ||
4441 | /** | |
bc2cda1e | 4442 | * dev_unicast_unsync - Remove synchronized addresses from the destination device |
e83a2ea8 CL |
4443 | * @to: destination device |
4444 | * @from: source device | |
4445 | * | |
4446 | * Remove all addresses that were added to the destination device by | |
4447 | * dev_unicast_sync(). This function is intended to be called from the | |
4448 | * dev->stop function of layered software devices. | |
4449 | */ | |
4450 | void dev_unicast_unsync(struct net_device *to, struct net_device *from) | |
4451 | { | |
ccffad25 JP |
4452 | if (to->addr_len != from->addr_len) |
4453 | return; | |
e83a2ea8 | 4454 | |
a6ac65db JP |
4455 | netif_addr_lock_bh(from); |
4456 | netif_addr_lock(to); | |
31278e71 | 4457 | __hw_addr_unsync(&to->uc, &from->uc, to->addr_len); |
ccffad25 | 4458 | __dev_set_rx_mode(to); |
a6ac65db JP |
4459 | netif_addr_unlock(to); |
4460 | netif_addr_unlock_bh(from); | |
e83a2ea8 CL |
4461 | } |
4462 | EXPORT_SYMBOL(dev_unicast_unsync); | |
4463 | ||
32a806c1 | 4464 | void dev_unicast_flush(struct net_device *dev) |
ccffad25 | 4465 | { |
a6ac65db | 4466 | netif_addr_lock_bh(dev); |
31278e71 | 4467 | __hw_addr_flush(&dev->uc); |
a6ac65db | 4468 | netif_addr_unlock_bh(dev); |
ccffad25 | 4469 | } |
32a806c1 | 4470 | EXPORT_SYMBOL(dev_unicast_flush); |
ccffad25 JP |
4471 | |
4472 | static void dev_unicast_init(struct net_device *dev) | |
4473 | { | |
31278e71 | 4474 | __hw_addr_init(&dev->uc); |
ccffad25 JP |
4475 | } |
4476 | ||
4477 | ||
12972621 DC |
4478 | static void __dev_addr_discard(struct dev_addr_list **list) |
4479 | { | |
4480 | struct dev_addr_list *tmp; | |
4481 | ||
4482 | while (*list != NULL) { | |
4483 | tmp = *list; | |
4484 | *list = tmp->next; | |
4485 | if (tmp->da_users > tmp->da_gusers) | |
4486 | printk("__dev_addr_discard: address leakage! " | |
4487 | "da_users=%d\n", tmp->da_users); | |
4488 | kfree(tmp); | |
4489 | } | |
4490 | } | |
4491 | ||
32a806c1 | 4492 | void dev_addr_discard(struct net_device *dev) |
4417da66 | 4493 | { |
b9e40857 | 4494 | netif_addr_lock_bh(dev); |
26cc2522 | 4495 | |
456ad75c | 4496 | __dev_addr_discard(&dev->mc_list); |
4cd24eaf | 4497 | netdev_mc_count(dev) = 0; |
26cc2522 | 4498 | |
b9e40857 | 4499 | netif_addr_unlock_bh(dev); |
456ad75c | 4500 | } |
32a806c1 | 4501 | EXPORT_SYMBOL(dev_addr_discard); |
456ad75c | 4502 | |
f0db275a SH |
4503 | /** |
4504 | * dev_get_flags - get flags reported to userspace | |
4505 | * @dev: device | |
4506 | * | |
4507 | * Get the combination of flag bits exported through APIs to userspace. | |
4508 | */ | |
1da177e4 LT |
4509 | unsigned dev_get_flags(const struct net_device *dev) |
4510 | { | |
4511 | unsigned flags; | |
4512 | ||
4513 | flags = (dev->flags & ~(IFF_PROMISC | | |
4514 | IFF_ALLMULTI | | |
b00055aa SR |
4515 | IFF_RUNNING | |
4516 | IFF_LOWER_UP | | |
4517 | IFF_DORMANT)) | | |
1da177e4 LT |
4518 | (dev->gflags & (IFF_PROMISC | |
4519 | IFF_ALLMULTI)); | |
4520 | ||
b00055aa SR |
4521 | if (netif_running(dev)) { |
4522 | if (netif_oper_up(dev)) | |
4523 | flags |= IFF_RUNNING; | |
4524 | if (netif_carrier_ok(dev)) | |
4525 | flags |= IFF_LOWER_UP; | |
4526 | if (netif_dormant(dev)) | |
4527 | flags |= IFF_DORMANT; | |
4528 | } | |
1da177e4 LT |
4529 | |
4530 | return flags; | |
4531 | } | |
d1b19dff | 4532 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 4533 | |
bd380811 | 4534 | int __dev_change_flags(struct net_device *dev, unsigned int flags) |
1da177e4 | 4535 | { |
1da177e4 | 4536 | int old_flags = dev->flags; |
bd380811 | 4537 | int ret; |
1da177e4 | 4538 | |
24023451 PM |
4539 | ASSERT_RTNL(); |
4540 | ||
1da177e4 LT |
4541 | /* |
4542 | * Set the flags on our device. | |
4543 | */ | |
4544 | ||
4545 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
4546 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
4547 | IFF_AUTOMEDIA)) | | |
4548 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
4549 | IFF_ALLMULTI)); | |
4550 | ||
4551 | /* | |
4552 | * Load in the correct multicast list now the flags have changed. | |
4553 | */ | |
4554 | ||
b6c40d68 PM |
4555 | if ((old_flags ^ flags) & IFF_MULTICAST) |
4556 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 4557 | |
4417da66 | 4558 | dev_set_rx_mode(dev); |
1da177e4 LT |
4559 | |
4560 | /* | |
4561 | * Have we downed the interface. We handle IFF_UP ourselves | |
4562 | * according to user attempts to set it, rather than blindly | |
4563 | * setting it. | |
4564 | */ | |
4565 | ||
4566 | ret = 0; | |
4567 | if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */ | |
bd380811 | 4568 | ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev); |
1da177e4 LT |
4569 | |
4570 | if (!ret) | |
4417da66 | 4571 | dev_set_rx_mode(dev); |
1da177e4 LT |
4572 | } |
4573 | ||
1da177e4 | 4574 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff ED |
4575 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
4576 | ||
1da177e4 LT |
4577 | dev->gflags ^= IFF_PROMISC; |
4578 | dev_set_promiscuity(dev, inc); | |
4579 | } | |
4580 | ||
4581 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
4582 | is important. Some (broken) drivers set IFF_PROMISC, when | |
4583 | IFF_ALLMULTI is requested not asking us and not reporting. | |
4584 | */ | |
4585 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
4586 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
4587 | ||
1da177e4 LT |
4588 | dev->gflags ^= IFF_ALLMULTI; |
4589 | dev_set_allmulti(dev, inc); | |
4590 | } | |
4591 | ||
bd380811 PM |
4592 | return ret; |
4593 | } | |
4594 | ||
4595 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags) | |
4596 | { | |
4597 | unsigned int changes = dev->flags ^ old_flags; | |
4598 | ||
4599 | if (changes & IFF_UP) { | |
4600 | if (dev->flags & IFF_UP) | |
4601 | call_netdevice_notifiers(NETDEV_UP, dev); | |
4602 | else | |
4603 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
4604 | } | |
4605 | ||
4606 | if (dev->flags & IFF_UP && | |
4607 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) | |
4608 | call_netdevice_notifiers(NETDEV_CHANGE, dev); | |
4609 | } | |
4610 | ||
4611 | /** | |
4612 | * dev_change_flags - change device settings | |
4613 | * @dev: device | |
4614 | * @flags: device state flags | |
4615 | * | |
4616 | * Change settings on device based state flags. The flags are | |
4617 | * in the userspace exported format. | |
4618 | */ | |
4619 | int dev_change_flags(struct net_device *dev, unsigned flags) | |
4620 | { | |
4621 | int ret, changes; | |
4622 | int old_flags = dev->flags; | |
4623 | ||
4624 | ret = __dev_change_flags(dev, flags); | |
4625 | if (ret < 0) | |
4626 | return ret; | |
4627 | ||
4628 | changes = old_flags ^ dev->flags; | |
7c355f53 TG |
4629 | if (changes) |
4630 | rtmsg_ifinfo(RTM_NEWLINK, dev, changes); | |
1da177e4 | 4631 | |
bd380811 | 4632 | __dev_notify_flags(dev, old_flags); |
1da177e4 LT |
4633 | return ret; |
4634 | } | |
d1b19dff | 4635 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 4636 | |
f0db275a SH |
4637 | /** |
4638 | * dev_set_mtu - Change maximum transfer unit | |
4639 | * @dev: device | |
4640 | * @new_mtu: new transfer unit | |
4641 | * | |
4642 | * Change the maximum transfer size of the network device. | |
4643 | */ | |
1da177e4 LT |
4644 | int dev_set_mtu(struct net_device *dev, int new_mtu) |
4645 | { | |
d314774c | 4646 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
4647 | int err; |
4648 | ||
4649 | if (new_mtu == dev->mtu) | |
4650 | return 0; | |
4651 | ||
4652 | /* MTU must be positive. */ | |
4653 | if (new_mtu < 0) | |
4654 | return -EINVAL; | |
4655 | ||
4656 | if (!netif_device_present(dev)) | |
4657 | return -ENODEV; | |
4658 | ||
4659 | err = 0; | |
d314774c SH |
4660 | if (ops->ndo_change_mtu) |
4661 | err = ops->ndo_change_mtu(dev, new_mtu); | |
1da177e4 LT |
4662 | else |
4663 | dev->mtu = new_mtu; | |
d314774c | 4664 | |
1da177e4 | 4665 | if (!err && dev->flags & IFF_UP) |
056925ab | 4666 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); |
1da177e4 LT |
4667 | return err; |
4668 | } | |
d1b19dff | 4669 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 4670 | |
f0db275a SH |
4671 | /** |
4672 | * dev_set_mac_address - Change Media Access Control Address | |
4673 | * @dev: device | |
4674 | * @sa: new address | |
4675 | * | |
4676 | * Change the hardware (MAC) address of the device | |
4677 | */ | |
1da177e4 LT |
4678 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) |
4679 | { | |
d314774c | 4680 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
4681 | int err; |
4682 | ||
d314774c | 4683 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
4684 | return -EOPNOTSUPP; |
4685 | if (sa->sa_family != dev->type) | |
4686 | return -EINVAL; | |
4687 | if (!netif_device_present(dev)) | |
4688 | return -ENODEV; | |
d314774c | 4689 | err = ops->ndo_set_mac_address(dev, sa); |
1da177e4 | 4690 | if (!err) |
056925ab | 4691 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
1da177e4 LT |
4692 | return err; |
4693 | } | |
d1b19dff | 4694 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 LT |
4695 | |
4696 | /* | |
3710becf | 4697 | * Perform the SIOCxIFxxx calls, inside rcu_read_lock() |
1da177e4 | 4698 | */ |
14e3e079 | 4699 | static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd) |
1da177e4 LT |
4700 | { |
4701 | int err; | |
3710becf | 4702 | struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name); |
1da177e4 LT |
4703 | |
4704 | if (!dev) | |
4705 | return -ENODEV; | |
4706 | ||
4707 | switch (cmd) { | |
d1b19dff ED |
4708 | case SIOCGIFFLAGS: /* Get interface flags */ |
4709 | ifr->ifr_flags = (short) dev_get_flags(dev); | |
4710 | return 0; | |
1da177e4 | 4711 | |
d1b19dff ED |
4712 | case SIOCGIFMETRIC: /* Get the metric on the interface |
4713 | (currently unused) */ | |
4714 | ifr->ifr_metric = 0; | |
4715 | return 0; | |
1da177e4 | 4716 | |
d1b19dff ED |
4717 | case SIOCGIFMTU: /* Get the MTU of a device */ |
4718 | ifr->ifr_mtu = dev->mtu; | |
4719 | return 0; | |
1da177e4 | 4720 | |
d1b19dff ED |
4721 | case SIOCGIFHWADDR: |
4722 | if (!dev->addr_len) | |
4723 | memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data); | |
4724 | else | |
4725 | memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr, | |
4726 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
4727 | ifr->ifr_hwaddr.sa_family = dev->type; | |
4728 | return 0; | |
1da177e4 | 4729 | |
d1b19dff ED |
4730 | case SIOCGIFSLAVE: |
4731 | err = -EINVAL; | |
4732 | break; | |
14e3e079 | 4733 | |
d1b19dff ED |
4734 | case SIOCGIFMAP: |
4735 | ifr->ifr_map.mem_start = dev->mem_start; | |
4736 | ifr->ifr_map.mem_end = dev->mem_end; | |
4737 | ifr->ifr_map.base_addr = dev->base_addr; | |
4738 | ifr->ifr_map.irq = dev->irq; | |
4739 | ifr->ifr_map.dma = dev->dma; | |
4740 | ifr->ifr_map.port = dev->if_port; | |
4741 | return 0; | |
14e3e079 | 4742 | |
d1b19dff ED |
4743 | case SIOCGIFINDEX: |
4744 | ifr->ifr_ifindex = dev->ifindex; | |
4745 | return 0; | |
14e3e079 | 4746 | |
d1b19dff ED |
4747 | case SIOCGIFTXQLEN: |
4748 | ifr->ifr_qlen = dev->tx_queue_len; | |
4749 | return 0; | |
14e3e079 | 4750 | |
d1b19dff ED |
4751 | default: |
4752 | /* dev_ioctl() should ensure this case | |
4753 | * is never reached | |
4754 | */ | |
4755 | WARN_ON(1); | |
4756 | err = -EINVAL; | |
4757 | break; | |
14e3e079 JG |
4758 | |
4759 | } | |
4760 | return err; | |
4761 | } | |
4762 | ||
4763 | /* | |
4764 | * Perform the SIOCxIFxxx calls, inside rtnl_lock() | |
4765 | */ | |
4766 | static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd) | |
4767 | { | |
4768 | int err; | |
4769 | struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name); | |
5f2f6da7 | 4770 | const struct net_device_ops *ops; |
14e3e079 JG |
4771 | |
4772 | if (!dev) | |
4773 | return -ENODEV; | |
4774 | ||
5f2f6da7 JP |
4775 | ops = dev->netdev_ops; |
4776 | ||
14e3e079 | 4777 | switch (cmd) { |
d1b19dff ED |
4778 | case SIOCSIFFLAGS: /* Set interface flags */ |
4779 | return dev_change_flags(dev, ifr->ifr_flags); | |
14e3e079 | 4780 | |
d1b19dff ED |
4781 | case SIOCSIFMETRIC: /* Set the metric on the interface |
4782 | (currently unused) */ | |
4783 | return -EOPNOTSUPP; | |
14e3e079 | 4784 | |
d1b19dff ED |
4785 | case SIOCSIFMTU: /* Set the MTU of a device */ |
4786 | return dev_set_mtu(dev, ifr->ifr_mtu); | |
1da177e4 | 4787 | |
d1b19dff ED |
4788 | case SIOCSIFHWADDR: |
4789 | return dev_set_mac_address(dev, &ifr->ifr_hwaddr); | |
1da177e4 | 4790 | |
d1b19dff ED |
4791 | case SIOCSIFHWBROADCAST: |
4792 | if (ifr->ifr_hwaddr.sa_family != dev->type) | |
4793 | return -EINVAL; | |
4794 | memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data, | |
4795 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
4796 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); | |
4797 | return 0; | |
1da177e4 | 4798 | |
d1b19dff ED |
4799 | case SIOCSIFMAP: |
4800 | if (ops->ndo_set_config) { | |
1da177e4 LT |
4801 | if (!netif_device_present(dev)) |
4802 | return -ENODEV; | |
d1b19dff ED |
4803 | return ops->ndo_set_config(dev, &ifr->ifr_map); |
4804 | } | |
4805 | return -EOPNOTSUPP; | |
1da177e4 | 4806 | |
d1b19dff ED |
4807 | case SIOCADDMULTI: |
4808 | if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) || | |
4809 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) | |
4810 | return -EINVAL; | |
4811 | if (!netif_device_present(dev)) | |
4812 | return -ENODEV; | |
4813 | return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data, | |
4814 | dev->addr_len, 1); | |
4815 | ||
4816 | case SIOCDELMULTI: | |
4817 | if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) || | |
4818 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) | |
4819 | return -EINVAL; | |
4820 | if (!netif_device_present(dev)) | |
4821 | return -ENODEV; | |
4822 | return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data, | |
4823 | dev->addr_len, 1); | |
1da177e4 | 4824 | |
d1b19dff ED |
4825 | case SIOCSIFTXQLEN: |
4826 | if (ifr->ifr_qlen < 0) | |
4827 | return -EINVAL; | |
4828 | dev->tx_queue_len = ifr->ifr_qlen; | |
4829 | return 0; | |
1da177e4 | 4830 | |
d1b19dff ED |
4831 | case SIOCSIFNAME: |
4832 | ifr->ifr_newname[IFNAMSIZ-1] = '\0'; | |
4833 | return dev_change_name(dev, ifr->ifr_newname); | |
1da177e4 | 4834 | |
d1b19dff ED |
4835 | /* |
4836 | * Unknown or private ioctl | |
4837 | */ | |
4838 | default: | |
4839 | if ((cmd >= SIOCDEVPRIVATE && | |
4840 | cmd <= SIOCDEVPRIVATE + 15) || | |
4841 | cmd == SIOCBONDENSLAVE || | |
4842 | cmd == SIOCBONDRELEASE || | |
4843 | cmd == SIOCBONDSETHWADDR || | |
4844 | cmd == SIOCBONDSLAVEINFOQUERY || | |
4845 | cmd == SIOCBONDINFOQUERY || | |
4846 | cmd == SIOCBONDCHANGEACTIVE || | |
4847 | cmd == SIOCGMIIPHY || | |
4848 | cmd == SIOCGMIIREG || | |
4849 | cmd == SIOCSMIIREG || | |
4850 | cmd == SIOCBRADDIF || | |
4851 | cmd == SIOCBRDELIF || | |
4852 | cmd == SIOCSHWTSTAMP || | |
4853 | cmd == SIOCWANDEV) { | |
4854 | err = -EOPNOTSUPP; | |
4855 | if (ops->ndo_do_ioctl) { | |
4856 | if (netif_device_present(dev)) | |
4857 | err = ops->ndo_do_ioctl(dev, ifr, cmd); | |
4858 | else | |
4859 | err = -ENODEV; | |
4860 | } | |
4861 | } else | |
4862 | err = -EINVAL; | |
1da177e4 LT |
4863 | |
4864 | } | |
4865 | return err; | |
4866 | } | |
4867 | ||
4868 | /* | |
4869 | * This function handles all "interface"-type I/O control requests. The actual | |
4870 | * 'doing' part of this is dev_ifsioc above. | |
4871 | */ | |
4872 | ||
4873 | /** | |
4874 | * dev_ioctl - network device ioctl | |
c4ea43c5 | 4875 | * @net: the applicable net namespace |
1da177e4 LT |
4876 | * @cmd: command to issue |
4877 | * @arg: pointer to a struct ifreq in user space | |
4878 | * | |
4879 | * Issue ioctl functions to devices. This is normally called by the | |
4880 | * user space syscall interfaces but can sometimes be useful for | |
4881 | * other purposes. The return value is the return from the syscall if | |
4882 | * positive or a negative errno code on error. | |
4883 | */ | |
4884 | ||
881d966b | 4885 | int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg) |
1da177e4 LT |
4886 | { |
4887 | struct ifreq ifr; | |
4888 | int ret; | |
4889 | char *colon; | |
4890 | ||
4891 | /* One special case: SIOCGIFCONF takes ifconf argument | |
4892 | and requires shared lock, because it sleeps writing | |
4893 | to user space. | |
4894 | */ | |
4895 | ||
4896 | if (cmd == SIOCGIFCONF) { | |
6756ae4b | 4897 | rtnl_lock(); |
881d966b | 4898 | ret = dev_ifconf(net, (char __user *) arg); |
6756ae4b | 4899 | rtnl_unlock(); |
1da177e4 LT |
4900 | return ret; |
4901 | } | |
4902 | if (cmd == SIOCGIFNAME) | |
881d966b | 4903 | return dev_ifname(net, (struct ifreq __user *)arg); |
1da177e4 LT |
4904 | |
4905 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
4906 | return -EFAULT; | |
4907 | ||
4908 | ifr.ifr_name[IFNAMSIZ-1] = 0; | |
4909 | ||
4910 | colon = strchr(ifr.ifr_name, ':'); | |
4911 | if (colon) | |
4912 | *colon = 0; | |
4913 | ||
4914 | /* | |
4915 | * See which interface the caller is talking about. | |
4916 | */ | |
4917 | ||
4918 | switch (cmd) { | |
d1b19dff ED |
4919 | /* |
4920 | * These ioctl calls: | |
4921 | * - can be done by all. | |
4922 | * - atomic and do not require locking. | |
4923 | * - return a value | |
4924 | */ | |
4925 | case SIOCGIFFLAGS: | |
4926 | case SIOCGIFMETRIC: | |
4927 | case SIOCGIFMTU: | |
4928 | case SIOCGIFHWADDR: | |
4929 | case SIOCGIFSLAVE: | |
4930 | case SIOCGIFMAP: | |
4931 | case SIOCGIFINDEX: | |
4932 | case SIOCGIFTXQLEN: | |
4933 | dev_load(net, ifr.ifr_name); | |
3710becf | 4934 | rcu_read_lock(); |
d1b19dff | 4935 | ret = dev_ifsioc_locked(net, &ifr, cmd); |
3710becf | 4936 | rcu_read_unlock(); |
d1b19dff ED |
4937 | if (!ret) { |
4938 | if (colon) | |
4939 | *colon = ':'; | |
4940 | if (copy_to_user(arg, &ifr, | |
4941 | sizeof(struct ifreq))) | |
4942 | ret = -EFAULT; | |
4943 | } | |
4944 | return ret; | |
1da177e4 | 4945 | |
d1b19dff ED |
4946 | case SIOCETHTOOL: |
4947 | dev_load(net, ifr.ifr_name); | |
4948 | rtnl_lock(); | |
4949 | ret = dev_ethtool(net, &ifr); | |
4950 | rtnl_unlock(); | |
4951 | if (!ret) { | |
4952 | if (colon) | |
4953 | *colon = ':'; | |
4954 | if (copy_to_user(arg, &ifr, | |
4955 | sizeof(struct ifreq))) | |
4956 | ret = -EFAULT; | |
4957 | } | |
4958 | return ret; | |
1da177e4 | 4959 | |
d1b19dff ED |
4960 | /* |
4961 | * These ioctl calls: | |
4962 | * - require superuser power. | |
4963 | * - require strict serialization. | |
4964 | * - return a value | |
4965 | */ | |
4966 | case SIOCGMIIPHY: | |
4967 | case SIOCGMIIREG: | |
4968 | case SIOCSIFNAME: | |
4969 | if (!capable(CAP_NET_ADMIN)) | |
4970 | return -EPERM; | |
4971 | dev_load(net, ifr.ifr_name); | |
4972 | rtnl_lock(); | |
4973 | ret = dev_ifsioc(net, &ifr, cmd); | |
4974 | rtnl_unlock(); | |
4975 | if (!ret) { | |
4976 | if (colon) | |
4977 | *colon = ':'; | |
4978 | if (copy_to_user(arg, &ifr, | |
4979 | sizeof(struct ifreq))) | |
4980 | ret = -EFAULT; | |
4981 | } | |
4982 | return ret; | |
1da177e4 | 4983 | |
d1b19dff ED |
4984 | /* |
4985 | * These ioctl calls: | |
4986 | * - require superuser power. | |
4987 | * - require strict serialization. | |
4988 | * - do not return a value | |
4989 | */ | |
4990 | case SIOCSIFFLAGS: | |
4991 | case SIOCSIFMETRIC: | |
4992 | case SIOCSIFMTU: | |
4993 | case SIOCSIFMAP: | |
4994 | case SIOCSIFHWADDR: | |
4995 | case SIOCSIFSLAVE: | |
4996 | case SIOCADDMULTI: | |
4997 | case SIOCDELMULTI: | |
4998 | case SIOCSIFHWBROADCAST: | |
4999 | case SIOCSIFTXQLEN: | |
5000 | case SIOCSMIIREG: | |
5001 | case SIOCBONDENSLAVE: | |
5002 | case SIOCBONDRELEASE: | |
5003 | case SIOCBONDSETHWADDR: | |
5004 | case SIOCBONDCHANGEACTIVE: | |
5005 | case SIOCBRADDIF: | |
5006 | case SIOCBRDELIF: | |
5007 | case SIOCSHWTSTAMP: | |
5008 | if (!capable(CAP_NET_ADMIN)) | |
5009 | return -EPERM; | |
5010 | /* fall through */ | |
5011 | case SIOCBONDSLAVEINFOQUERY: | |
5012 | case SIOCBONDINFOQUERY: | |
5013 | dev_load(net, ifr.ifr_name); | |
5014 | rtnl_lock(); | |
5015 | ret = dev_ifsioc(net, &ifr, cmd); | |
5016 | rtnl_unlock(); | |
5017 | return ret; | |
5018 | ||
5019 | case SIOCGIFMEM: | |
5020 | /* Get the per device memory space. We can add this but | |
5021 | * currently do not support it */ | |
5022 | case SIOCSIFMEM: | |
5023 | /* Set the per device memory buffer space. | |
5024 | * Not applicable in our case */ | |
5025 | case SIOCSIFLINK: | |
5026 | return -EINVAL; | |
5027 | ||
5028 | /* | |
5029 | * Unknown or private ioctl. | |
5030 | */ | |
5031 | default: | |
5032 | if (cmd == SIOCWANDEV || | |
5033 | (cmd >= SIOCDEVPRIVATE && | |
5034 | cmd <= SIOCDEVPRIVATE + 15)) { | |
881d966b | 5035 | dev_load(net, ifr.ifr_name); |
1da177e4 | 5036 | rtnl_lock(); |
881d966b | 5037 | ret = dev_ifsioc(net, &ifr, cmd); |
1da177e4 | 5038 | rtnl_unlock(); |
d1b19dff ED |
5039 | if (!ret && copy_to_user(arg, &ifr, |
5040 | sizeof(struct ifreq))) | |
5041 | ret = -EFAULT; | |
1da177e4 | 5042 | return ret; |
d1b19dff ED |
5043 | } |
5044 | /* Take care of Wireless Extensions */ | |
5045 | if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) | |
5046 | return wext_handle_ioctl(net, &ifr, cmd, arg); | |
5047 | return -EINVAL; | |
1da177e4 LT |
5048 | } |
5049 | } | |
5050 | ||
5051 | ||
5052 | /** | |
5053 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 5054 | * @net: the applicable net namespace |
1da177e4 LT |
5055 | * |
5056 | * Returns a suitable unique value for a new device interface | |
5057 | * number. The caller must hold the rtnl semaphore or the | |
5058 | * dev_base_lock to be sure it remains unique. | |
5059 | */ | |
881d966b | 5060 | static int dev_new_index(struct net *net) |
1da177e4 LT |
5061 | { |
5062 | static int ifindex; | |
5063 | for (;;) { | |
5064 | if (++ifindex <= 0) | |
5065 | ifindex = 1; | |
881d966b | 5066 | if (!__dev_get_by_index(net, ifindex)) |
1da177e4 LT |
5067 | return ifindex; |
5068 | } | |
5069 | } | |
5070 | ||
1da177e4 | 5071 | /* Delayed registration/unregisteration */ |
3b5b34fd | 5072 | static LIST_HEAD(net_todo_list); |
1da177e4 | 5073 | |
6f05f629 | 5074 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 5075 | { |
1da177e4 | 5076 | list_add_tail(&dev->todo_list, &net_todo_list); |
1da177e4 LT |
5077 | } |
5078 | ||
9b5e383c | 5079 | static void rollback_registered_many(struct list_head *head) |
93ee31f1 | 5080 | { |
e93737b0 | 5081 | struct net_device *dev, *tmp; |
9b5e383c | 5082 | |
93ee31f1 DL |
5083 | BUG_ON(dev_boot_phase); |
5084 | ASSERT_RTNL(); | |
5085 | ||
e93737b0 | 5086 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
9b5e383c | 5087 | /* Some devices call without registering |
e93737b0 KK |
5088 | * for initialization unwind. Remove those |
5089 | * devices and proceed with the remaining. | |
9b5e383c ED |
5090 | */ |
5091 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
5092 | pr_debug("unregister_netdevice: device %s/%p never " | |
5093 | "was registered\n", dev->name, dev); | |
93ee31f1 | 5094 | |
9b5e383c | 5095 | WARN_ON(1); |
e93737b0 KK |
5096 | list_del(&dev->unreg_list); |
5097 | continue; | |
9b5e383c | 5098 | } |
93ee31f1 | 5099 | |
9b5e383c | 5100 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
93ee31f1 | 5101 | |
9b5e383c ED |
5102 | /* If device is running, close it first. */ |
5103 | dev_close(dev); | |
93ee31f1 | 5104 | |
9b5e383c ED |
5105 | /* And unlink it from device chain. */ |
5106 | unlist_netdevice(dev); | |
93ee31f1 | 5107 | |
9b5e383c ED |
5108 | dev->reg_state = NETREG_UNREGISTERING; |
5109 | } | |
93ee31f1 DL |
5110 | |
5111 | synchronize_net(); | |
5112 | ||
9b5e383c ED |
5113 | list_for_each_entry(dev, head, unreg_list) { |
5114 | /* Shutdown queueing discipline. */ | |
5115 | dev_shutdown(dev); | |
93ee31f1 DL |
5116 | |
5117 | ||
9b5e383c ED |
5118 | /* Notify protocols, that we are about to destroy |
5119 | this device. They should clean all the things. | |
5120 | */ | |
5121 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
93ee31f1 | 5122 | |
a2835763 PM |
5123 | if (!dev->rtnl_link_ops || |
5124 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
5125 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U); | |
5126 | ||
9b5e383c ED |
5127 | /* |
5128 | * Flush the unicast and multicast chains | |
5129 | */ | |
5130 | dev_unicast_flush(dev); | |
5131 | dev_addr_discard(dev); | |
93ee31f1 | 5132 | |
9b5e383c ED |
5133 | if (dev->netdev_ops->ndo_uninit) |
5134 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 | 5135 | |
9b5e383c ED |
5136 | /* Notifier chain MUST detach us from master device. */ |
5137 | WARN_ON(dev->master); | |
93ee31f1 | 5138 | |
9b5e383c ED |
5139 | /* Remove entries from kobject tree */ |
5140 | netdev_unregister_kobject(dev); | |
5141 | } | |
93ee31f1 | 5142 | |
a5ee1551 | 5143 | /* Process any work delayed until the end of the batch */ |
e5e26d75 | 5144 | dev = list_first_entry(head, struct net_device, unreg_list); |
a5ee1551 | 5145 | call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev); |
93ee31f1 | 5146 | |
a5ee1551 | 5147 | synchronize_net(); |
395264d5 | 5148 | |
a5ee1551 | 5149 | list_for_each_entry(dev, head, unreg_list) |
9b5e383c ED |
5150 | dev_put(dev); |
5151 | } | |
5152 | ||
5153 | static void rollback_registered(struct net_device *dev) | |
5154 | { | |
5155 | LIST_HEAD(single); | |
5156 | ||
5157 | list_add(&dev->unreg_list, &single); | |
5158 | rollback_registered_many(&single); | |
93ee31f1 DL |
5159 | } |
5160 | ||
e8a0464c DM |
5161 | static void __netdev_init_queue_locks_one(struct net_device *dev, |
5162 | struct netdev_queue *dev_queue, | |
5163 | void *_unused) | |
c773e847 DM |
5164 | { |
5165 | spin_lock_init(&dev_queue->_xmit_lock); | |
cf508b12 | 5166 | netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type); |
c773e847 DM |
5167 | dev_queue->xmit_lock_owner = -1; |
5168 | } | |
5169 | ||
5170 | static void netdev_init_queue_locks(struct net_device *dev) | |
5171 | { | |
e8a0464c DM |
5172 | netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL); |
5173 | __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL); | |
c773e847 DM |
5174 | } |
5175 | ||
b63365a2 HX |
5176 | unsigned long netdev_fix_features(unsigned long features, const char *name) |
5177 | { | |
5178 | /* Fix illegal SG+CSUM combinations. */ | |
5179 | if ((features & NETIF_F_SG) && | |
5180 | !(features & NETIF_F_ALL_CSUM)) { | |
5181 | if (name) | |
5182 | printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no " | |
5183 | "checksum feature.\n", name); | |
5184 | features &= ~NETIF_F_SG; | |
5185 | } | |
5186 | ||
5187 | /* TSO requires that SG is present as well. */ | |
5188 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) { | |
5189 | if (name) | |
5190 | printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no " | |
5191 | "SG feature.\n", name); | |
5192 | features &= ~NETIF_F_TSO; | |
5193 | } | |
5194 | ||
5195 | if (features & NETIF_F_UFO) { | |
5196 | if (!(features & NETIF_F_GEN_CSUM)) { | |
5197 | if (name) | |
5198 | printk(KERN_ERR "%s: Dropping NETIF_F_UFO " | |
5199 | "since no NETIF_F_HW_CSUM feature.\n", | |
5200 | name); | |
5201 | features &= ~NETIF_F_UFO; | |
5202 | } | |
5203 | ||
5204 | if (!(features & NETIF_F_SG)) { | |
5205 | if (name) | |
5206 | printk(KERN_ERR "%s: Dropping NETIF_F_UFO " | |
5207 | "since no NETIF_F_SG feature.\n", name); | |
5208 | features &= ~NETIF_F_UFO; | |
5209 | } | |
5210 | } | |
5211 | ||
5212 | return features; | |
5213 | } | |
5214 | EXPORT_SYMBOL(netdev_fix_features); | |
5215 | ||
fc4a7489 PM |
5216 | /** |
5217 | * netif_stacked_transfer_operstate - transfer operstate | |
5218 | * @rootdev: the root or lower level device to transfer state from | |
5219 | * @dev: the device to transfer operstate to | |
5220 | * | |
5221 | * Transfer operational state from root to device. This is normally | |
5222 | * called when a stacking relationship exists between the root | |
5223 | * device and the device(a leaf device). | |
5224 | */ | |
5225 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
5226 | struct net_device *dev) | |
5227 | { | |
5228 | if (rootdev->operstate == IF_OPER_DORMANT) | |
5229 | netif_dormant_on(dev); | |
5230 | else | |
5231 | netif_dormant_off(dev); | |
5232 | ||
5233 | if (netif_carrier_ok(rootdev)) { | |
5234 | if (!netif_carrier_ok(dev)) | |
5235 | netif_carrier_on(dev); | |
5236 | } else { | |
5237 | if (netif_carrier_ok(dev)) | |
5238 | netif_carrier_off(dev); | |
5239 | } | |
5240 | } | |
5241 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
5242 | ||
1da177e4 LT |
5243 | /** |
5244 | * register_netdevice - register a network device | |
5245 | * @dev: device to register | |
5246 | * | |
5247 | * Take a completed network device structure and add it to the kernel | |
5248 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
5249 | * chain. 0 is returned on success. A negative errno code is returned | |
5250 | * on a failure to set up the device, or if the name is a duplicate. | |
5251 | * | |
5252 | * Callers must hold the rtnl semaphore. You may want | |
5253 | * register_netdev() instead of this. | |
5254 | * | |
5255 | * BUGS: | |
5256 | * The locking appears insufficient to guarantee two parallel registers | |
5257 | * will not get the same name. | |
5258 | */ | |
5259 | ||
5260 | int register_netdevice(struct net_device *dev) | |
5261 | { | |
1da177e4 | 5262 | int ret; |
d314774c | 5263 | struct net *net = dev_net(dev); |
1da177e4 LT |
5264 | |
5265 | BUG_ON(dev_boot_phase); | |
5266 | ASSERT_RTNL(); | |
5267 | ||
b17a7c17 SH |
5268 | might_sleep(); |
5269 | ||
1da177e4 LT |
5270 | /* When net_device's are persistent, this will be fatal. */ |
5271 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 5272 | BUG_ON(!net); |
1da177e4 | 5273 | |
f1f28aa3 | 5274 | spin_lock_init(&dev->addr_list_lock); |
cf508b12 | 5275 | netdev_set_addr_lockdep_class(dev); |
c773e847 | 5276 | netdev_init_queue_locks(dev); |
1da177e4 | 5277 | |
1da177e4 LT |
5278 | dev->iflink = -1; |
5279 | ||
df334545 | 5280 | #ifdef CONFIG_RPS |
0a9627f2 TH |
5281 | if (!dev->num_rx_queues) { |
5282 | /* | |
5283 | * Allocate a single RX queue if driver never called | |
5284 | * alloc_netdev_mq | |
5285 | */ | |
5286 | ||
5287 | dev->_rx = kzalloc(sizeof(struct netdev_rx_queue), GFP_KERNEL); | |
5288 | if (!dev->_rx) { | |
5289 | ret = -ENOMEM; | |
5290 | goto out; | |
5291 | } | |
5292 | ||
5293 | dev->_rx->first = dev->_rx; | |
5294 | atomic_set(&dev->_rx->count, 1); | |
5295 | dev->num_rx_queues = 1; | |
5296 | } | |
df334545 | 5297 | #endif |
1da177e4 | 5298 | /* Init, if this function is available */ |
d314774c SH |
5299 | if (dev->netdev_ops->ndo_init) { |
5300 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
5301 | if (ret) { |
5302 | if (ret > 0) | |
5303 | ret = -EIO; | |
90833aa4 | 5304 | goto out; |
1da177e4 LT |
5305 | } |
5306 | } | |
4ec93edb | 5307 | |
d9031024 OP |
5308 | ret = dev_get_valid_name(net, dev->name, dev->name, 0); |
5309 | if (ret) | |
7ce1b0ed | 5310 | goto err_uninit; |
1da177e4 | 5311 | |
881d966b | 5312 | dev->ifindex = dev_new_index(net); |
1da177e4 LT |
5313 | if (dev->iflink == -1) |
5314 | dev->iflink = dev->ifindex; | |
5315 | ||
d212f87b SH |
5316 | /* Fix illegal checksum combinations */ |
5317 | if ((dev->features & NETIF_F_HW_CSUM) && | |
5318 | (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
5319 | printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n", | |
5320 | dev->name); | |
5321 | dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); | |
5322 | } | |
5323 | ||
5324 | if ((dev->features & NETIF_F_NO_CSUM) && | |
5325 | (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
5326 | printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n", | |
5327 | dev->name); | |
5328 | dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM); | |
5329 | } | |
5330 | ||
b63365a2 | 5331 | dev->features = netdev_fix_features(dev->features, dev->name); |
1da177e4 | 5332 | |
e5a4a72d LB |
5333 | /* Enable software GSO if SG is supported. */ |
5334 | if (dev->features & NETIF_F_SG) | |
5335 | dev->features |= NETIF_F_GSO; | |
5336 | ||
aaf8cdc3 | 5337 | netdev_initialize_kobject(dev); |
7ffbe3fd JB |
5338 | |
5339 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); | |
5340 | ret = notifier_to_errno(ret); | |
5341 | if (ret) | |
5342 | goto err_uninit; | |
5343 | ||
8b41d188 | 5344 | ret = netdev_register_kobject(dev); |
b17a7c17 | 5345 | if (ret) |
7ce1b0ed | 5346 | goto err_uninit; |
b17a7c17 SH |
5347 | dev->reg_state = NETREG_REGISTERED; |
5348 | ||
1da177e4 LT |
5349 | /* |
5350 | * Default initial state at registry is that the | |
5351 | * device is present. | |
5352 | */ | |
5353 | ||
5354 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
5355 | ||
1da177e4 | 5356 | dev_init_scheduler(dev); |
1da177e4 | 5357 | dev_hold(dev); |
ce286d32 | 5358 | list_netdevice(dev); |
1da177e4 LT |
5359 | |
5360 | /* Notify protocols, that a new device appeared. */ | |
056925ab | 5361 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 5362 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
5363 | if (ret) { |
5364 | rollback_registered(dev); | |
5365 | dev->reg_state = NETREG_UNREGISTERED; | |
5366 | } | |
d90a909e EB |
5367 | /* |
5368 | * Prevent userspace races by waiting until the network | |
5369 | * device is fully setup before sending notifications. | |
5370 | */ | |
a2835763 PM |
5371 | if (!dev->rtnl_link_ops || |
5372 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
5373 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U); | |
1da177e4 LT |
5374 | |
5375 | out: | |
5376 | return ret; | |
7ce1b0ed HX |
5377 | |
5378 | err_uninit: | |
d314774c SH |
5379 | if (dev->netdev_ops->ndo_uninit) |
5380 | dev->netdev_ops->ndo_uninit(dev); | |
7ce1b0ed | 5381 | goto out; |
1da177e4 | 5382 | } |
d1b19dff | 5383 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 5384 | |
937f1ba5 BH |
5385 | /** |
5386 | * init_dummy_netdev - init a dummy network device for NAPI | |
5387 | * @dev: device to init | |
5388 | * | |
5389 | * This takes a network device structure and initialize the minimum | |
5390 | * amount of fields so it can be used to schedule NAPI polls without | |
5391 | * registering a full blown interface. This is to be used by drivers | |
5392 | * that need to tie several hardware interfaces to a single NAPI | |
5393 | * poll scheduler due to HW limitations. | |
5394 | */ | |
5395 | int init_dummy_netdev(struct net_device *dev) | |
5396 | { | |
5397 | /* Clear everything. Note we don't initialize spinlocks | |
5398 | * are they aren't supposed to be taken by any of the | |
5399 | * NAPI code and this dummy netdev is supposed to be | |
5400 | * only ever used for NAPI polls | |
5401 | */ | |
5402 | memset(dev, 0, sizeof(struct net_device)); | |
5403 | ||
5404 | /* make sure we BUG if trying to hit standard | |
5405 | * register/unregister code path | |
5406 | */ | |
5407 | dev->reg_state = NETREG_DUMMY; | |
5408 | ||
5409 | /* initialize the ref count */ | |
5410 | atomic_set(&dev->refcnt, 1); | |
5411 | ||
5412 | /* NAPI wants this */ | |
5413 | INIT_LIST_HEAD(&dev->napi_list); | |
5414 | ||
5415 | /* a dummy interface is started by default */ | |
5416 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
5417 | set_bit(__LINK_STATE_START, &dev->state); | |
5418 | ||
5419 | return 0; | |
5420 | } | |
5421 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
5422 | ||
5423 | ||
1da177e4 LT |
5424 | /** |
5425 | * register_netdev - register a network device | |
5426 | * @dev: device to register | |
5427 | * | |
5428 | * Take a completed network device structure and add it to the kernel | |
5429 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
5430 | * chain. 0 is returned on success. A negative errno code is returned | |
5431 | * on a failure to set up the device, or if the name is a duplicate. | |
5432 | * | |
38b4da38 | 5433 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
5434 | * and expands the device name if you passed a format string to |
5435 | * alloc_netdev. | |
5436 | */ | |
5437 | int register_netdev(struct net_device *dev) | |
5438 | { | |
5439 | int err; | |
5440 | ||
5441 | rtnl_lock(); | |
5442 | ||
5443 | /* | |
5444 | * If the name is a format string the caller wants us to do a | |
5445 | * name allocation. | |
5446 | */ | |
5447 | if (strchr(dev->name, '%')) { | |
5448 | err = dev_alloc_name(dev, dev->name); | |
5449 | if (err < 0) | |
5450 | goto out; | |
5451 | } | |
4ec93edb | 5452 | |
1da177e4 LT |
5453 | err = register_netdevice(dev); |
5454 | out: | |
5455 | rtnl_unlock(); | |
5456 | return err; | |
5457 | } | |
5458 | EXPORT_SYMBOL(register_netdev); | |
5459 | ||
5460 | /* | |
5461 | * netdev_wait_allrefs - wait until all references are gone. | |
5462 | * | |
5463 | * This is called when unregistering network devices. | |
5464 | * | |
5465 | * Any protocol or device that holds a reference should register | |
5466 | * for netdevice notification, and cleanup and put back the | |
5467 | * reference if they receive an UNREGISTER event. | |
5468 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 5469 | * call dev_put. |
1da177e4 LT |
5470 | */ |
5471 | static void netdev_wait_allrefs(struct net_device *dev) | |
5472 | { | |
5473 | unsigned long rebroadcast_time, warning_time; | |
5474 | ||
e014debe ED |
5475 | linkwatch_forget_dev(dev); |
5476 | ||
1da177e4 LT |
5477 | rebroadcast_time = warning_time = jiffies; |
5478 | while (atomic_read(&dev->refcnt) != 0) { | |
5479 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { | |
6756ae4b | 5480 | rtnl_lock(); |
1da177e4 LT |
5481 | |
5482 | /* Rebroadcast unregister notification */ | |
056925ab | 5483 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
a5ee1551 | 5484 | /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users |
395264d5 | 5485 | * should have already handle it the first time */ |
1da177e4 LT |
5486 | |
5487 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, | |
5488 | &dev->state)) { | |
5489 | /* We must not have linkwatch events | |
5490 | * pending on unregister. If this | |
5491 | * happens, we simply run the queue | |
5492 | * unscheduled, resulting in a noop | |
5493 | * for this device. | |
5494 | */ | |
5495 | linkwatch_run_queue(); | |
5496 | } | |
5497 | ||
6756ae4b | 5498 | __rtnl_unlock(); |
1da177e4 LT |
5499 | |
5500 | rebroadcast_time = jiffies; | |
5501 | } | |
5502 | ||
5503 | msleep(250); | |
5504 | ||
5505 | if (time_after(jiffies, warning_time + 10 * HZ)) { | |
5506 | printk(KERN_EMERG "unregister_netdevice: " | |
5507 | "waiting for %s to become free. Usage " | |
5508 | "count = %d\n", | |
5509 | dev->name, atomic_read(&dev->refcnt)); | |
5510 | warning_time = jiffies; | |
5511 | } | |
5512 | } | |
5513 | } | |
5514 | ||
5515 | /* The sequence is: | |
5516 | * | |
5517 | * rtnl_lock(); | |
5518 | * ... | |
5519 | * register_netdevice(x1); | |
5520 | * register_netdevice(x2); | |
5521 | * ... | |
5522 | * unregister_netdevice(y1); | |
5523 | * unregister_netdevice(y2); | |
5524 | * ... | |
5525 | * rtnl_unlock(); | |
5526 | * free_netdev(y1); | |
5527 | * free_netdev(y2); | |
5528 | * | |
58ec3b4d | 5529 | * We are invoked by rtnl_unlock(). |
1da177e4 | 5530 | * This allows us to deal with problems: |
b17a7c17 | 5531 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
5532 | * without deadlocking with linkwatch via keventd. |
5533 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
5534 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
5535 | * |
5536 | * We must not return until all unregister events added during | |
5537 | * the interval the lock was held have been completed. | |
1da177e4 | 5538 | */ |
1da177e4 LT |
5539 | void netdev_run_todo(void) |
5540 | { | |
626ab0e6 | 5541 | struct list_head list; |
1da177e4 | 5542 | |
1da177e4 | 5543 | /* Snapshot list, allow later requests */ |
626ab0e6 | 5544 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
5545 | |
5546 | __rtnl_unlock(); | |
626ab0e6 | 5547 | |
1da177e4 LT |
5548 | while (!list_empty(&list)) { |
5549 | struct net_device *dev | |
e5e26d75 | 5550 | = list_first_entry(&list, struct net_device, todo_list); |
e51d739a | 5551 | int i; |
1da177e4 LT |
5552 | list_del(&dev->todo_list); |
5553 | ||
b17a7c17 SH |
5554 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
5555 | printk(KERN_ERR "network todo '%s' but state %d\n", | |
5556 | dev->name, dev->reg_state); | |
5557 | dump_stack(); | |
5558 | continue; | |
5559 | } | |
1da177e4 | 5560 | |
b17a7c17 | 5561 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 5562 | |
e51d739a TH |
5563 | for_each_online_cpu(i) |
5564 | flush_backlog(dev, i); | |
6e583ce5 | 5565 | |
b17a7c17 | 5566 | netdev_wait_allrefs(dev); |
1da177e4 | 5567 | |
b17a7c17 SH |
5568 | /* paranoia */ |
5569 | BUG_ON(atomic_read(&dev->refcnt)); | |
547b792c IJ |
5570 | WARN_ON(dev->ip_ptr); |
5571 | WARN_ON(dev->ip6_ptr); | |
5572 | WARN_ON(dev->dn_ptr); | |
1da177e4 | 5573 | |
b17a7c17 SH |
5574 | if (dev->destructor) |
5575 | dev->destructor(dev); | |
9093bbb2 SH |
5576 | |
5577 | /* Free network device */ | |
5578 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 5579 | } |
1da177e4 LT |
5580 | } |
5581 | ||
d83345ad ED |
5582 | /** |
5583 | * dev_txq_stats_fold - fold tx_queues stats | |
5584 | * @dev: device to get statistics from | |
5585 | * @stats: struct net_device_stats to hold results | |
5586 | */ | |
5587 | void dev_txq_stats_fold(const struct net_device *dev, | |
5588 | struct net_device_stats *stats) | |
5589 | { | |
5590 | unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0; | |
5591 | unsigned int i; | |
5592 | struct netdev_queue *txq; | |
5593 | ||
5594 | for (i = 0; i < dev->num_tx_queues; i++) { | |
5595 | txq = netdev_get_tx_queue(dev, i); | |
5596 | tx_bytes += txq->tx_bytes; | |
5597 | tx_packets += txq->tx_packets; | |
5598 | tx_dropped += txq->tx_dropped; | |
5599 | } | |
5600 | if (tx_bytes || tx_packets || tx_dropped) { | |
5601 | stats->tx_bytes = tx_bytes; | |
5602 | stats->tx_packets = tx_packets; | |
5603 | stats->tx_dropped = tx_dropped; | |
5604 | } | |
5605 | } | |
5606 | EXPORT_SYMBOL(dev_txq_stats_fold); | |
5607 | ||
eeda3fd6 SH |
5608 | /** |
5609 | * dev_get_stats - get network device statistics | |
5610 | * @dev: device to get statistics from | |
5611 | * | |
5612 | * Get network statistics from device. The device driver may provide | |
5613 | * its own method by setting dev->netdev_ops->get_stats; otherwise | |
5614 | * the internal statistics structure is used. | |
5615 | */ | |
5616 | const struct net_device_stats *dev_get_stats(struct net_device *dev) | |
7004bf25 | 5617 | { |
eeda3fd6 SH |
5618 | const struct net_device_ops *ops = dev->netdev_ops; |
5619 | ||
5620 | if (ops->ndo_get_stats) | |
5621 | return ops->ndo_get_stats(dev); | |
d83345ad ED |
5622 | |
5623 | dev_txq_stats_fold(dev, &dev->stats); | |
5624 | return &dev->stats; | |
c45d286e | 5625 | } |
eeda3fd6 | 5626 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 5627 | |
dc2b4847 | 5628 | static void netdev_init_one_queue(struct net_device *dev, |
e8a0464c DM |
5629 | struct netdev_queue *queue, |
5630 | void *_unused) | |
dc2b4847 | 5631 | { |
dc2b4847 DM |
5632 | queue->dev = dev; |
5633 | } | |
5634 | ||
bb949fbd DM |
5635 | static void netdev_init_queues(struct net_device *dev) |
5636 | { | |
e8a0464c DM |
5637 | netdev_init_one_queue(dev, &dev->rx_queue, NULL); |
5638 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); | |
c3f26a26 | 5639 | spin_lock_init(&dev->tx_global_lock); |
bb949fbd DM |
5640 | } |
5641 | ||
1da177e4 | 5642 | /** |
f25f4e44 | 5643 | * alloc_netdev_mq - allocate network device |
1da177e4 LT |
5644 | * @sizeof_priv: size of private data to allocate space for |
5645 | * @name: device name format string | |
5646 | * @setup: callback to initialize device | |
f25f4e44 | 5647 | * @queue_count: the number of subqueues to allocate |
1da177e4 LT |
5648 | * |
5649 | * Allocates a struct net_device with private data area for driver use | |
f25f4e44 PWJ |
5650 | * and performs basic initialization. Also allocates subquue structs |
5651 | * for each queue on the device at the end of the netdevice. | |
1da177e4 | 5652 | */ |
f25f4e44 PWJ |
5653 | struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name, |
5654 | void (*setup)(struct net_device *), unsigned int queue_count) | |
1da177e4 | 5655 | { |
e8a0464c | 5656 | struct netdev_queue *tx; |
1da177e4 | 5657 | struct net_device *dev; |
7943986c | 5658 | size_t alloc_size; |
1ce8e7b5 | 5659 | struct net_device *p; |
df334545 ED |
5660 | #ifdef CONFIG_RPS |
5661 | struct netdev_rx_queue *rx; | |
0a9627f2 | 5662 | int i; |
df334545 | 5663 | #endif |
1da177e4 | 5664 | |
b6fe17d6 SH |
5665 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
5666 | ||
fd2ea0a7 | 5667 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
5668 | if (sizeof_priv) { |
5669 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 5670 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
5671 | alloc_size += sizeof_priv; |
5672 | } | |
5673 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 5674 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 5675 | |
31380de9 | 5676 | p = kzalloc(alloc_size, GFP_KERNEL); |
1da177e4 | 5677 | if (!p) { |
b6fe17d6 | 5678 | printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n"); |
1da177e4 LT |
5679 | return NULL; |
5680 | } | |
1da177e4 | 5681 | |
7943986c | 5682 | tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL); |
e8a0464c DM |
5683 | if (!tx) { |
5684 | printk(KERN_ERR "alloc_netdev: Unable to allocate " | |
5685 | "tx qdiscs.\n"); | |
ab9c73cc | 5686 | goto free_p; |
e8a0464c DM |
5687 | } |
5688 | ||
df334545 | 5689 | #ifdef CONFIG_RPS |
0a9627f2 TH |
5690 | rx = kcalloc(queue_count, sizeof(struct netdev_rx_queue), GFP_KERNEL); |
5691 | if (!rx) { | |
5692 | printk(KERN_ERR "alloc_netdev: Unable to allocate " | |
5693 | "rx queues.\n"); | |
5694 | goto free_tx; | |
5695 | } | |
5696 | ||
5697 | atomic_set(&rx->count, queue_count); | |
5698 | ||
5699 | /* | |
5700 | * Set a pointer to first element in the array which holds the | |
5701 | * reference count. | |
5702 | */ | |
5703 | for (i = 0; i < queue_count; i++) | |
5704 | rx[i].first = rx; | |
df334545 | 5705 | #endif |
0a9627f2 | 5706 | |
1ce8e7b5 | 5707 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 5708 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc JP |
5709 | |
5710 | if (dev_addr_init(dev)) | |
0a9627f2 | 5711 | goto free_rx; |
ab9c73cc | 5712 | |
ccffad25 JP |
5713 | dev_unicast_init(dev); |
5714 | ||
c346dca1 | 5715 | dev_net_set(dev, &init_net); |
1da177e4 | 5716 | |
e8a0464c DM |
5717 | dev->_tx = tx; |
5718 | dev->num_tx_queues = queue_count; | |
fd2ea0a7 | 5719 | dev->real_num_tx_queues = queue_count; |
e8a0464c | 5720 | |
df334545 | 5721 | #ifdef CONFIG_RPS |
0a9627f2 TH |
5722 | dev->_rx = rx; |
5723 | dev->num_rx_queues = queue_count; | |
df334545 | 5724 | #endif |
0a9627f2 | 5725 | |
82cc1a7a | 5726 | dev->gso_max_size = GSO_MAX_SIZE; |
1da177e4 | 5727 | |
bb949fbd DM |
5728 | netdev_init_queues(dev); |
5729 | ||
15682bc4 PWJ |
5730 | INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list); |
5731 | dev->ethtool_ntuple_list.count = 0; | |
d565b0a1 | 5732 | INIT_LIST_HEAD(&dev->napi_list); |
9fdce099 | 5733 | INIT_LIST_HEAD(&dev->unreg_list); |
e014debe | 5734 | INIT_LIST_HEAD(&dev->link_watch_list); |
93f154b5 | 5735 | dev->priv_flags = IFF_XMIT_DST_RELEASE; |
1da177e4 LT |
5736 | setup(dev); |
5737 | strcpy(dev->name, name); | |
5738 | return dev; | |
ab9c73cc | 5739 | |
0a9627f2 | 5740 | free_rx: |
df334545 | 5741 | #ifdef CONFIG_RPS |
0a9627f2 | 5742 | kfree(rx); |
ab9c73cc | 5743 | free_tx: |
df334545 | 5744 | #endif |
ab9c73cc | 5745 | kfree(tx); |
ab9c73cc JP |
5746 | free_p: |
5747 | kfree(p); | |
5748 | return NULL; | |
1da177e4 | 5749 | } |
f25f4e44 | 5750 | EXPORT_SYMBOL(alloc_netdev_mq); |
1da177e4 LT |
5751 | |
5752 | /** | |
5753 | * free_netdev - free network device | |
5754 | * @dev: device | |
5755 | * | |
4ec93edb YH |
5756 | * This function does the last stage of destroying an allocated device |
5757 | * interface. The reference to the device object is released. | |
1da177e4 LT |
5758 | * If this is the last reference then it will be freed. |
5759 | */ | |
5760 | void free_netdev(struct net_device *dev) | |
5761 | { | |
d565b0a1 HX |
5762 | struct napi_struct *p, *n; |
5763 | ||
f3005d7f DL |
5764 | release_net(dev_net(dev)); |
5765 | ||
e8a0464c DM |
5766 | kfree(dev->_tx); |
5767 | ||
f001fde5 JP |
5768 | /* Flush device addresses */ |
5769 | dev_addr_flush(dev); | |
5770 | ||
15682bc4 PWJ |
5771 | /* Clear ethtool n-tuple list */ |
5772 | ethtool_ntuple_flush(dev); | |
5773 | ||
d565b0a1 HX |
5774 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
5775 | netif_napi_del(p); | |
5776 | ||
3041a069 | 5777 | /* Compatibility with error handling in drivers */ |
1da177e4 LT |
5778 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
5779 | kfree((char *)dev - dev->padded); | |
5780 | return; | |
5781 | } | |
5782 | ||
5783 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
5784 | dev->reg_state = NETREG_RELEASED; | |
5785 | ||
43cb76d9 GKH |
5786 | /* will free via device release */ |
5787 | put_device(&dev->dev); | |
1da177e4 | 5788 | } |
d1b19dff | 5789 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 5790 | |
f0db275a SH |
5791 | /** |
5792 | * synchronize_net - Synchronize with packet receive processing | |
5793 | * | |
5794 | * Wait for packets currently being received to be done. | |
5795 | * Does not block later packets from starting. | |
5796 | */ | |
4ec93edb | 5797 | void synchronize_net(void) |
1da177e4 LT |
5798 | { |
5799 | might_sleep(); | |
fbd568a3 | 5800 | synchronize_rcu(); |
1da177e4 | 5801 | } |
d1b19dff | 5802 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
5803 | |
5804 | /** | |
44a0873d | 5805 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 5806 | * @dev: device |
44a0873d | 5807 | * @head: list |
6ebfbc06 | 5808 | * |
1da177e4 | 5809 | * This function shuts down a device interface and removes it |
d59b54b1 | 5810 | * from the kernel tables. |
44a0873d | 5811 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
5812 | * |
5813 | * Callers must hold the rtnl semaphore. You may want | |
5814 | * unregister_netdev() instead of this. | |
5815 | */ | |
5816 | ||
44a0873d | 5817 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 5818 | { |
a6620712 HX |
5819 | ASSERT_RTNL(); |
5820 | ||
44a0873d | 5821 | if (head) { |
9fdce099 | 5822 | list_move_tail(&dev->unreg_list, head); |
44a0873d ED |
5823 | } else { |
5824 | rollback_registered(dev); | |
5825 | /* Finish processing unregister after unlock */ | |
5826 | net_set_todo(dev); | |
5827 | } | |
1da177e4 | 5828 | } |
44a0873d | 5829 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 5830 | |
9b5e383c ED |
5831 | /** |
5832 | * unregister_netdevice_many - unregister many devices | |
5833 | * @head: list of devices | |
9b5e383c ED |
5834 | */ |
5835 | void unregister_netdevice_many(struct list_head *head) | |
5836 | { | |
5837 | struct net_device *dev; | |
5838 | ||
5839 | if (!list_empty(head)) { | |
5840 | rollback_registered_many(head); | |
5841 | list_for_each_entry(dev, head, unreg_list) | |
5842 | net_set_todo(dev); | |
5843 | } | |
5844 | } | |
63c8099d | 5845 | EXPORT_SYMBOL(unregister_netdevice_many); |
9b5e383c | 5846 | |
1da177e4 LT |
5847 | /** |
5848 | * unregister_netdev - remove device from the kernel | |
5849 | * @dev: device | |
5850 | * | |
5851 | * This function shuts down a device interface and removes it | |
d59b54b1 | 5852 | * from the kernel tables. |
1da177e4 LT |
5853 | * |
5854 | * This is just a wrapper for unregister_netdevice that takes | |
5855 | * the rtnl semaphore. In general you want to use this and not | |
5856 | * unregister_netdevice. | |
5857 | */ | |
5858 | void unregister_netdev(struct net_device *dev) | |
5859 | { | |
5860 | rtnl_lock(); | |
5861 | unregister_netdevice(dev); | |
5862 | rtnl_unlock(); | |
5863 | } | |
1da177e4 LT |
5864 | EXPORT_SYMBOL(unregister_netdev); |
5865 | ||
ce286d32 EB |
5866 | /** |
5867 | * dev_change_net_namespace - move device to different nethost namespace | |
5868 | * @dev: device | |
5869 | * @net: network namespace | |
5870 | * @pat: If not NULL name pattern to try if the current device name | |
5871 | * is already taken in the destination network namespace. | |
5872 | * | |
5873 | * This function shuts down a device interface and moves it | |
5874 | * to a new network namespace. On success 0 is returned, on | |
5875 | * a failure a netagive errno code is returned. | |
5876 | * | |
5877 | * Callers must hold the rtnl semaphore. | |
5878 | */ | |
5879 | ||
5880 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
5881 | { | |
ce286d32 EB |
5882 | int err; |
5883 | ||
5884 | ASSERT_RTNL(); | |
5885 | ||
5886 | /* Don't allow namespace local devices to be moved. */ | |
5887 | err = -EINVAL; | |
5888 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
5889 | goto out; | |
5890 | ||
3891845e EB |
5891 | #ifdef CONFIG_SYSFS |
5892 | /* Don't allow real devices to be moved when sysfs | |
5893 | * is enabled. | |
5894 | */ | |
5895 | err = -EINVAL; | |
5896 | if (dev->dev.parent) | |
5897 | goto out; | |
5898 | #endif | |
5899 | ||
ce286d32 EB |
5900 | /* Ensure the device has been registrered */ |
5901 | err = -EINVAL; | |
5902 | if (dev->reg_state != NETREG_REGISTERED) | |
5903 | goto out; | |
5904 | ||
5905 | /* Get out if there is nothing todo */ | |
5906 | err = 0; | |
878628fb | 5907 | if (net_eq(dev_net(dev), net)) |
ce286d32 EB |
5908 | goto out; |
5909 | ||
5910 | /* Pick the destination device name, and ensure | |
5911 | * we can use it in the destination network namespace. | |
5912 | */ | |
5913 | err = -EEXIST; | |
d9031024 | 5914 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
5915 | /* We get here if we can't use the current device name */ |
5916 | if (!pat) | |
5917 | goto out; | |
d9031024 | 5918 | if (dev_get_valid_name(net, pat, dev->name, 1)) |
ce286d32 EB |
5919 | goto out; |
5920 | } | |
5921 | ||
5922 | /* | |
5923 | * And now a mini version of register_netdevice unregister_netdevice. | |
5924 | */ | |
5925 | ||
5926 | /* If device is running close it first. */ | |
9b772652 | 5927 | dev_close(dev); |
ce286d32 EB |
5928 | |
5929 | /* And unlink it from device chain */ | |
5930 | err = -ENODEV; | |
5931 | unlist_netdevice(dev); | |
5932 | ||
5933 | synchronize_net(); | |
5934 | ||
5935 | /* Shutdown queueing discipline. */ | |
5936 | dev_shutdown(dev); | |
5937 | ||
5938 | /* Notify protocols, that we are about to destroy | |
5939 | this device. They should clean all the things. | |
5940 | */ | |
5941 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
a5ee1551 | 5942 | call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev); |
ce286d32 EB |
5943 | |
5944 | /* | |
5945 | * Flush the unicast and multicast chains | |
5946 | */ | |
ccffad25 | 5947 | dev_unicast_flush(dev); |
ce286d32 EB |
5948 | dev_addr_discard(dev); |
5949 | ||
3891845e EB |
5950 | netdev_unregister_kobject(dev); |
5951 | ||
ce286d32 | 5952 | /* Actually switch the network namespace */ |
c346dca1 | 5953 | dev_net_set(dev, net); |
ce286d32 | 5954 | |
ce286d32 EB |
5955 | /* If there is an ifindex conflict assign a new one */ |
5956 | if (__dev_get_by_index(net, dev->ifindex)) { | |
5957 | int iflink = (dev->iflink == dev->ifindex); | |
5958 | dev->ifindex = dev_new_index(net); | |
5959 | if (iflink) | |
5960 | dev->iflink = dev->ifindex; | |
5961 | } | |
5962 | ||
8b41d188 | 5963 | /* Fixup kobjects */ |
aaf8cdc3 | 5964 | err = netdev_register_kobject(dev); |
8b41d188 | 5965 | WARN_ON(err); |
ce286d32 EB |
5966 | |
5967 | /* Add the device back in the hashes */ | |
5968 | list_netdevice(dev); | |
5969 | ||
5970 | /* Notify protocols, that a new device appeared. */ | |
5971 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
5972 | ||
d90a909e EB |
5973 | /* |
5974 | * Prevent userspace races by waiting until the network | |
5975 | * device is fully setup before sending notifications. | |
5976 | */ | |
5977 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U); | |
5978 | ||
ce286d32 EB |
5979 | synchronize_net(); |
5980 | err = 0; | |
5981 | out: | |
5982 | return err; | |
5983 | } | |
463d0183 | 5984 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 5985 | |
1da177e4 LT |
5986 | static int dev_cpu_callback(struct notifier_block *nfb, |
5987 | unsigned long action, | |
5988 | void *ocpu) | |
5989 | { | |
5990 | struct sk_buff **list_skb; | |
37437bb2 | 5991 | struct Qdisc **list_net; |
1da177e4 LT |
5992 | struct sk_buff *skb; |
5993 | unsigned int cpu, oldcpu = (unsigned long)ocpu; | |
5994 | struct softnet_data *sd, *oldsd; | |
5995 | ||
8bb78442 | 5996 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
1da177e4 LT |
5997 | return NOTIFY_OK; |
5998 | ||
5999 | local_irq_disable(); | |
6000 | cpu = smp_processor_id(); | |
6001 | sd = &per_cpu(softnet_data, cpu); | |
6002 | oldsd = &per_cpu(softnet_data, oldcpu); | |
6003 | ||
6004 | /* Find end of our completion_queue. */ | |
6005 | list_skb = &sd->completion_queue; | |
6006 | while (*list_skb) | |
6007 | list_skb = &(*list_skb)->next; | |
6008 | /* Append completion queue from offline CPU. */ | |
6009 | *list_skb = oldsd->completion_queue; | |
6010 | oldsd->completion_queue = NULL; | |
6011 | ||
6012 | /* Find end of our output_queue. */ | |
6013 | list_net = &sd->output_queue; | |
6014 | while (*list_net) | |
6015 | list_net = &(*list_net)->next_sched; | |
6016 | /* Append output queue from offline CPU. */ | |
6017 | *list_net = oldsd->output_queue; | |
6018 | oldsd->output_queue = NULL; | |
6019 | ||
6020 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
6021 | local_irq_enable(); | |
6022 | ||
6023 | /* Process offline CPU's input_pkt_queue */ | |
6024 | while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) | |
6025 | netif_rx(skb); | |
6026 | ||
6027 | return NOTIFY_OK; | |
6028 | } | |
1da177e4 LT |
6029 | |
6030 | ||
7f353bf2 | 6031 | /** |
b63365a2 HX |
6032 | * netdev_increment_features - increment feature set by one |
6033 | * @all: current feature set | |
6034 | * @one: new feature set | |
6035 | * @mask: mask feature set | |
7f353bf2 HX |
6036 | * |
6037 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
6038 | * @one to the master device with current feature set @all. Will not |
6039 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 6040 | */ |
b63365a2 HX |
6041 | unsigned long netdev_increment_features(unsigned long all, unsigned long one, |
6042 | unsigned long mask) | |
6043 | { | |
6044 | /* If device needs checksumming, downgrade to it. */ | |
d1b19dff | 6045 | if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM)) |
b63365a2 HX |
6046 | all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM); |
6047 | else if (mask & NETIF_F_ALL_CSUM) { | |
6048 | /* If one device supports v4/v6 checksumming, set for all. */ | |
6049 | if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) && | |
6050 | !(all & NETIF_F_GEN_CSUM)) { | |
6051 | all &= ~NETIF_F_ALL_CSUM; | |
6052 | all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); | |
6053 | } | |
e2a6b852 | 6054 | |
b63365a2 HX |
6055 | /* If one device supports hw checksumming, set for all. */ |
6056 | if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) { | |
6057 | all &= ~NETIF_F_ALL_CSUM; | |
6058 | all |= NETIF_F_HW_CSUM; | |
6059 | } | |
6060 | } | |
7f353bf2 | 6061 | |
b63365a2 | 6062 | one |= NETIF_F_ALL_CSUM; |
7f353bf2 | 6063 | |
b63365a2 | 6064 | one |= all & NETIF_F_ONE_FOR_ALL; |
d9f5950f | 6065 | all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO; |
b63365a2 | 6066 | all |= one & mask & NETIF_F_ONE_FOR_ALL; |
7f353bf2 HX |
6067 | |
6068 | return all; | |
6069 | } | |
b63365a2 | 6070 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 6071 | |
30d97d35 PE |
6072 | static struct hlist_head *netdev_create_hash(void) |
6073 | { | |
6074 | int i; | |
6075 | struct hlist_head *hash; | |
6076 | ||
6077 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
6078 | if (hash != NULL) | |
6079 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
6080 | INIT_HLIST_HEAD(&hash[i]); | |
6081 | ||
6082 | return hash; | |
6083 | } | |
6084 | ||
881d966b | 6085 | /* Initialize per network namespace state */ |
4665079c | 6086 | static int __net_init netdev_init(struct net *net) |
881d966b | 6087 | { |
881d966b | 6088 | INIT_LIST_HEAD(&net->dev_base_head); |
881d966b | 6089 | |
30d97d35 PE |
6090 | net->dev_name_head = netdev_create_hash(); |
6091 | if (net->dev_name_head == NULL) | |
6092 | goto err_name; | |
881d966b | 6093 | |
30d97d35 PE |
6094 | net->dev_index_head = netdev_create_hash(); |
6095 | if (net->dev_index_head == NULL) | |
6096 | goto err_idx; | |
881d966b EB |
6097 | |
6098 | return 0; | |
30d97d35 PE |
6099 | |
6100 | err_idx: | |
6101 | kfree(net->dev_name_head); | |
6102 | err_name: | |
6103 | return -ENOMEM; | |
881d966b EB |
6104 | } |
6105 | ||
f0db275a SH |
6106 | /** |
6107 | * netdev_drivername - network driver for the device | |
6108 | * @dev: network device | |
6109 | * @buffer: buffer for resulting name | |
6110 | * @len: size of buffer | |
6111 | * | |
6112 | * Determine network driver for device. | |
6113 | */ | |
cf04a4c7 | 6114 | char *netdev_drivername(const struct net_device *dev, char *buffer, int len) |
6579e57b | 6115 | { |
cf04a4c7 SH |
6116 | const struct device_driver *driver; |
6117 | const struct device *parent; | |
6579e57b AV |
6118 | |
6119 | if (len <= 0 || !buffer) | |
6120 | return buffer; | |
6121 | buffer[0] = 0; | |
6122 | ||
6123 | parent = dev->dev.parent; | |
6124 | ||
6125 | if (!parent) | |
6126 | return buffer; | |
6127 | ||
6128 | driver = parent->driver; | |
6129 | if (driver && driver->name) | |
6130 | strlcpy(buffer, driver->name, len); | |
6131 | return buffer; | |
6132 | } | |
6133 | ||
4665079c | 6134 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
6135 | { |
6136 | kfree(net->dev_name_head); | |
6137 | kfree(net->dev_index_head); | |
6138 | } | |
6139 | ||
022cbae6 | 6140 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
6141 | .init = netdev_init, |
6142 | .exit = netdev_exit, | |
6143 | }; | |
6144 | ||
4665079c | 6145 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 6146 | { |
e008b5fc | 6147 | struct net_device *dev, *aux; |
ce286d32 | 6148 | /* |
e008b5fc | 6149 | * Push all migratable network devices back to the |
ce286d32 EB |
6150 | * initial network namespace |
6151 | */ | |
6152 | rtnl_lock(); | |
e008b5fc | 6153 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 6154 | int err; |
aca51397 | 6155 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
6156 | |
6157 | /* Ignore unmoveable devices (i.e. loopback) */ | |
6158 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
6159 | continue; | |
6160 | ||
e008b5fc EB |
6161 | /* Leave virtual devices for the generic cleanup */ |
6162 | if (dev->rtnl_link_ops) | |
6163 | continue; | |
d0c082ce | 6164 | |
ce286d32 | 6165 | /* Push remaing network devices to init_net */ |
aca51397 PE |
6166 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
6167 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
ce286d32 | 6168 | if (err) { |
aca51397 | 6169 | printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n", |
ce286d32 | 6170 | __func__, dev->name, err); |
aca51397 | 6171 | BUG(); |
ce286d32 EB |
6172 | } |
6173 | } | |
6174 | rtnl_unlock(); | |
6175 | } | |
6176 | ||
04dc7f6b EB |
6177 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
6178 | { | |
6179 | /* At exit all network devices most be removed from a network | |
6180 | * namespace. Do this in the reverse order of registeration. | |
6181 | * Do this across as many network namespaces as possible to | |
6182 | * improve batching efficiency. | |
6183 | */ | |
6184 | struct net_device *dev; | |
6185 | struct net *net; | |
6186 | LIST_HEAD(dev_kill_list); | |
6187 | ||
6188 | rtnl_lock(); | |
6189 | list_for_each_entry(net, net_list, exit_list) { | |
6190 | for_each_netdev_reverse(net, dev) { | |
6191 | if (dev->rtnl_link_ops) | |
6192 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); | |
6193 | else | |
6194 | unregister_netdevice_queue(dev, &dev_kill_list); | |
6195 | } | |
6196 | } | |
6197 | unregister_netdevice_many(&dev_kill_list); | |
6198 | rtnl_unlock(); | |
6199 | } | |
6200 | ||
022cbae6 | 6201 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 6202 | .exit = default_device_exit, |
04dc7f6b | 6203 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
6204 | }; |
6205 | ||
1da177e4 LT |
6206 | /* |
6207 | * Initialize the DEV module. At boot time this walks the device list and | |
6208 | * unhooks any devices that fail to initialise (normally hardware not | |
6209 | * present) and leaves us with a valid list of present and active devices. | |
6210 | * | |
6211 | */ | |
6212 | ||
6213 | /* | |
6214 | * This is called single threaded during boot, so no need | |
6215 | * to take the rtnl semaphore. | |
6216 | */ | |
6217 | static int __init net_dev_init(void) | |
6218 | { | |
6219 | int i, rc = -ENOMEM; | |
6220 | ||
6221 | BUG_ON(!dev_boot_phase); | |
6222 | ||
1da177e4 LT |
6223 | if (dev_proc_init()) |
6224 | goto out; | |
6225 | ||
8b41d188 | 6226 | if (netdev_kobject_init()) |
1da177e4 LT |
6227 | goto out; |
6228 | ||
6229 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 6230 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
6231 | INIT_LIST_HEAD(&ptype_base[i]); |
6232 | ||
881d966b EB |
6233 | if (register_pernet_subsys(&netdev_net_ops)) |
6234 | goto out; | |
1da177e4 LT |
6235 | |
6236 | /* | |
6237 | * Initialise the packet receive queues. | |
6238 | */ | |
6239 | ||
6f912042 | 6240 | for_each_possible_cpu(i) { |
1da177e4 LT |
6241 | struct softnet_data *queue; |
6242 | ||
6243 | queue = &per_cpu(softnet_data, i); | |
6244 | skb_queue_head_init(&queue->input_pkt_queue); | |
1da177e4 LT |
6245 | queue->completion_queue = NULL; |
6246 | INIT_LIST_HEAD(&queue->poll_list); | |
bea3348e | 6247 | |
df334545 | 6248 | #ifdef CONFIG_RPS |
0a9627f2 TH |
6249 | queue->csd.func = trigger_softirq; |
6250 | queue->csd.info = queue; | |
6251 | queue->csd.flags = 0; | |
1e94d72f | 6252 | #endif |
0a9627f2 | 6253 | |
bea3348e SH |
6254 | queue->backlog.poll = process_backlog; |
6255 | queue->backlog.weight = weight_p; | |
d565b0a1 | 6256 | queue->backlog.gro_list = NULL; |
4ae5544f | 6257 | queue->backlog.gro_count = 0; |
1da177e4 LT |
6258 | } |
6259 | ||
1da177e4 LT |
6260 | dev_boot_phase = 0; |
6261 | ||
505d4f73 EB |
6262 | /* The loopback device is special if any other network devices |
6263 | * is present in a network namespace the loopback device must | |
6264 | * be present. Since we now dynamically allocate and free the | |
6265 | * loopback device ensure this invariant is maintained by | |
6266 | * keeping the loopback device as the first device on the | |
6267 | * list of network devices. Ensuring the loopback devices | |
6268 | * is the first device that appears and the last network device | |
6269 | * that disappears. | |
6270 | */ | |
6271 | if (register_pernet_device(&loopback_net_ops)) | |
6272 | goto out; | |
6273 | ||
6274 | if (register_pernet_device(&default_device_ops)) | |
6275 | goto out; | |
6276 | ||
962cf36c CM |
6277 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
6278 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 LT |
6279 | |
6280 | hotcpu_notifier(dev_cpu_callback, 0); | |
6281 | dst_init(); | |
6282 | dev_mcast_init(); | |
6283 | rc = 0; | |
6284 | out: | |
6285 | return rc; | |
6286 | } | |
6287 | ||
6288 | subsys_initcall(net_dev_init); | |
6289 | ||
e88721f8 KK |
6290 | static int __init initialize_hashrnd(void) |
6291 | { | |
0a9627f2 | 6292 | get_random_bytes(&hashrnd, sizeof(hashrnd)); |
e88721f8 KK |
6293 | return 0; |
6294 | } | |
6295 | ||
6296 | late_initcall_sync(initialize_hashrnd); | |
6297 |