]>
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> |
5a0e3ad6 | 83 | #include <linux/slab.h> |
1da177e4 | 84 | #include <linux/sched.h> |
4a3e2f71 | 85 | #include <linux/mutex.h> |
1da177e4 LT |
86 | #include <linux/string.h> |
87 | #include <linux/mm.h> | |
88 | #include <linux/socket.h> | |
89 | #include <linux/sockios.h> | |
90 | #include <linux/errno.h> | |
91 | #include <linux/interrupt.h> | |
92 | #include <linux/if_ether.h> | |
93 | #include <linux/netdevice.h> | |
94 | #include <linux/etherdevice.h> | |
0187bdfb | 95 | #include <linux/ethtool.h> |
1da177e4 LT |
96 | #include <linux/notifier.h> |
97 | #include <linux/skbuff.h> | |
457c4cbc | 98 | #include <net/net_namespace.h> |
1da177e4 LT |
99 | #include <net/sock.h> |
100 | #include <linux/rtnetlink.h> | |
101 | #include <linux/proc_fs.h> | |
102 | #include <linux/seq_file.h> | |
103 | #include <linux/stat.h> | |
1da177e4 LT |
104 | #include <net/dst.h> |
105 | #include <net/pkt_sched.h> | |
106 | #include <net/checksum.h> | |
44540960 | 107 | #include <net/xfrm.h> |
1da177e4 LT |
108 | #include <linux/highmem.h> |
109 | #include <linux/init.h> | |
110 | #include <linux/kmod.h> | |
111 | #include <linux/module.h> | |
1da177e4 LT |
112 | #include <linux/netpoll.h> |
113 | #include <linux/rcupdate.h> | |
114 | #include <linux/delay.h> | |
295f4a1f | 115 | #include <net/wext.h> |
1da177e4 | 116 | #include <net/iw_handler.h> |
1da177e4 | 117 | #include <asm/current.h> |
5bdb9886 | 118 | #include <linux/audit.h> |
db217334 | 119 | #include <linux/dmaengine.h> |
f6a78bfc | 120 | #include <linux/err.h> |
c7fa9d18 | 121 | #include <linux/ctype.h> |
723e98b7 | 122 | #include <linux/if_arp.h> |
6de329e2 | 123 | #include <linux/if_vlan.h> |
8f0f2223 | 124 | #include <linux/ip.h> |
ad55dcaf | 125 | #include <net/ip.h> |
8f0f2223 DM |
126 | #include <linux/ipv6.h> |
127 | #include <linux/in.h> | |
b6b2fed1 DM |
128 | #include <linux/jhash.h> |
129 | #include <linux/random.h> | |
9cbc1cb8 | 130 | #include <trace/events/napi.h> |
cf66ba58 | 131 | #include <trace/events/net.h> |
07dc22e7 | 132 | #include <trace/events/skb.h> |
5acbbd42 | 133 | #include <linux/pci.h> |
caeda9b9 | 134 | #include <linux/inetdevice.h> |
1da177e4 | 135 | |
342709ef PE |
136 | #include "net-sysfs.h" |
137 | ||
d565b0a1 HX |
138 | /* Instead of increasing this, you should create a hash table. */ |
139 | #define MAX_GRO_SKBS 8 | |
140 | ||
5d38a079 HX |
141 | /* This should be increased if a protocol with a bigger head is added. */ |
142 | #define GRO_MAX_HEAD (MAX_HEADER + 128) | |
143 | ||
1da177e4 LT |
144 | /* |
145 | * The list of packet types we will receive (as opposed to discard) | |
146 | * and the routines to invoke. | |
147 | * | |
148 | * Why 16. Because with 16 the only overlap we get on a hash of the | |
149 | * low nibble of the protocol value is RARP/SNAP/X.25. | |
150 | * | |
151 | * NOTE: That is no longer true with the addition of VLAN tags. Not | |
152 | * sure which should go first, but I bet it won't make much | |
153 | * difference if we are running VLANs. The good news is that | |
154 | * this protocol won't be in the list unless compiled in, so | |
3041a069 | 155 | * the average user (w/out VLANs) will not be adversely affected. |
1da177e4 LT |
156 | * --BLG |
157 | * | |
158 | * 0800 IP | |
159 | * 8100 802.1Q VLAN | |
160 | * 0001 802.3 | |
161 | * 0002 AX.25 | |
162 | * 0004 802.2 | |
163 | * 8035 RARP | |
164 | * 0005 SNAP | |
165 | * 0805 X.25 | |
166 | * 0806 ARP | |
167 | * 8137 IPX | |
168 | * 0009 Localtalk | |
169 | * 86DD IPv6 | |
170 | */ | |
171 | ||
82d8a867 PE |
172 | #define PTYPE_HASH_SIZE (16) |
173 | #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1) | |
174 | ||
1da177e4 | 175 | static DEFINE_SPINLOCK(ptype_lock); |
82d8a867 | 176 | static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
6b2bedc3 | 177 | static struct list_head ptype_all __read_mostly; /* Taps */ |
1da177e4 | 178 | |
1da177e4 | 179 | /* |
7562f876 | 180 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
181 | * semaphore. |
182 | * | |
c6d14c84 | 183 | * Pure readers hold dev_base_lock for reading, or rcu_read_lock() |
1da177e4 LT |
184 | * |
185 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 186 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
187 | * actual updates. This allows pure readers to access the list even |
188 | * while a writer is preparing to update it. | |
189 | * | |
190 | * To put it another way, dev_base_lock is held for writing only to | |
191 | * protect against pure readers; the rtnl semaphore provides the | |
192 | * protection against other writers. | |
193 | * | |
194 | * See, for example usages, register_netdevice() and | |
195 | * unregister_netdevice(), which must be called with the rtnl | |
196 | * semaphore held. | |
197 | */ | |
1da177e4 | 198 | DEFINE_RWLOCK(dev_base_lock); |
1da177e4 LT |
199 | EXPORT_SYMBOL(dev_base_lock); |
200 | ||
881d966b | 201 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 LT |
202 | { |
203 | unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ)); | |
08e9897d | 204 | return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; |
1da177e4 LT |
205 | } |
206 | ||
881d966b | 207 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 208 | { |
7c28bd0b | 209 | return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; |
1da177e4 LT |
210 | } |
211 | ||
e36fa2f7 | 212 | static inline void rps_lock(struct softnet_data *sd) |
152102c7 CG |
213 | { |
214 | #ifdef CONFIG_RPS | |
e36fa2f7 | 215 | spin_lock(&sd->input_pkt_queue.lock); |
152102c7 CG |
216 | #endif |
217 | } | |
218 | ||
e36fa2f7 | 219 | static inline void rps_unlock(struct softnet_data *sd) |
152102c7 CG |
220 | { |
221 | #ifdef CONFIG_RPS | |
e36fa2f7 | 222 | spin_unlock(&sd->input_pkt_queue.lock); |
152102c7 CG |
223 | #endif |
224 | } | |
225 | ||
ce286d32 EB |
226 | /* Device list insertion */ |
227 | static int list_netdevice(struct net_device *dev) | |
228 | { | |
c346dca1 | 229 | struct net *net = dev_net(dev); |
ce286d32 EB |
230 | |
231 | ASSERT_RTNL(); | |
232 | ||
233 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 234 | list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); |
72c9528b | 235 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); |
fb699dfd ED |
236 | hlist_add_head_rcu(&dev->index_hlist, |
237 | dev_index_hash(net, dev->ifindex)); | |
ce286d32 EB |
238 | write_unlock_bh(&dev_base_lock); |
239 | return 0; | |
240 | } | |
241 | ||
fb699dfd ED |
242 | /* Device list removal |
243 | * caller must respect a RCU grace period before freeing/reusing dev | |
244 | */ | |
ce286d32 EB |
245 | static void unlist_netdevice(struct net_device *dev) |
246 | { | |
247 | ASSERT_RTNL(); | |
248 | ||
249 | /* Unlink dev from the device chain */ | |
250 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 251 | list_del_rcu(&dev->dev_list); |
72c9528b | 252 | hlist_del_rcu(&dev->name_hlist); |
fb699dfd | 253 | hlist_del_rcu(&dev->index_hlist); |
ce286d32 EB |
254 | write_unlock_bh(&dev_base_lock); |
255 | } | |
256 | ||
1da177e4 LT |
257 | /* |
258 | * Our notifier list | |
259 | */ | |
260 | ||
f07d5b94 | 261 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
262 | |
263 | /* | |
264 | * Device drivers call our routines to queue packets here. We empty the | |
265 | * queue in the local softnet handler. | |
266 | */ | |
bea3348e | 267 | |
9958da05 | 268 | DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); |
d1b19dff | 269 | EXPORT_PER_CPU_SYMBOL(softnet_data); |
1da177e4 | 270 | |
cf508b12 | 271 | #ifdef CONFIG_LOCKDEP |
723e98b7 | 272 | /* |
c773e847 | 273 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class |
723e98b7 JP |
274 | * according to dev->type |
275 | */ | |
276 | static const unsigned short netdev_lock_type[] = | |
277 | {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
278 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, | |
279 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
280 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
281 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
282 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
283 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
284 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
285 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
286 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
287 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
288 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
289 | ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211, | |
2d91d78b | 290 | ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, |
929122cd | 291 | ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154, |
fcb94e42 | 292 | ARPHRD_VOID, ARPHRD_NONE}; |
723e98b7 | 293 | |
36cbd3dc | 294 | static const char *const netdev_lock_name[] = |
723e98b7 JP |
295 | {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", |
296 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
297 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
298 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
299 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
300 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
301 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
302 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
303 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
304 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
305 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
306 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
307 | "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211", | |
2d91d78b | 308 | "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", |
929122cd | 309 | "_xmit_PHONET_PIPE", "_xmit_IEEE802154", |
fcb94e42 | 310 | "_xmit_VOID", "_xmit_NONE"}; |
723e98b7 JP |
311 | |
312 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
cf508b12 | 313 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; |
723e98b7 JP |
314 | |
315 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
316 | { | |
317 | int i; | |
318 | ||
319 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
320 | if (netdev_lock_type[i] == dev_type) | |
321 | return i; | |
322 | /* the last key is used by default */ | |
323 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
324 | } | |
325 | ||
cf508b12 DM |
326 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
327 | unsigned short dev_type) | |
723e98b7 JP |
328 | { |
329 | int i; | |
330 | ||
331 | i = netdev_lock_pos(dev_type); | |
332 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
333 | netdev_lock_name[i]); | |
334 | } | |
cf508b12 DM |
335 | |
336 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
337 | { | |
338 | int i; | |
339 | ||
340 | i = netdev_lock_pos(dev->type); | |
341 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
342 | &netdev_addr_lock_key[i], | |
343 | netdev_lock_name[i]); | |
344 | } | |
723e98b7 | 345 | #else |
cf508b12 DM |
346 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
347 | unsigned short dev_type) | |
348 | { | |
349 | } | |
350 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
723e98b7 JP |
351 | { |
352 | } | |
353 | #endif | |
1da177e4 LT |
354 | |
355 | /******************************************************************************* | |
356 | ||
357 | Protocol management and registration routines | |
358 | ||
359 | *******************************************************************************/ | |
360 | ||
1da177e4 LT |
361 | /* |
362 | * Add a protocol ID to the list. Now that the input handler is | |
363 | * smarter we can dispense with all the messy stuff that used to be | |
364 | * here. | |
365 | * | |
366 | * BEWARE!!! Protocol handlers, mangling input packets, | |
367 | * MUST BE last in hash buckets and checking protocol handlers | |
368 | * MUST start from promiscuous ptype_all chain in net_bh. | |
369 | * It is true now, do not change it. | |
370 | * Explanation follows: if protocol handler, mangling packet, will | |
371 | * be the first on list, it is not able to sense, that packet | |
372 | * is cloned and should be copied-on-write, so that it will | |
373 | * change it and subsequent readers will get broken packet. | |
374 | * --ANK (980803) | |
375 | */ | |
376 | ||
c07b68e8 ED |
377 | static inline struct list_head *ptype_head(const struct packet_type *pt) |
378 | { | |
379 | if (pt->type == htons(ETH_P_ALL)) | |
380 | return &ptype_all; | |
381 | else | |
382 | return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
383 | } | |
384 | ||
1da177e4 LT |
385 | /** |
386 | * dev_add_pack - add packet handler | |
387 | * @pt: packet type declaration | |
388 | * | |
389 | * Add a protocol handler to the networking stack. The passed &packet_type | |
390 | * is linked into kernel lists and may not be freed until it has been | |
391 | * removed from the kernel lists. | |
392 | * | |
4ec93edb | 393 | * This call does not sleep therefore it can not |
1da177e4 LT |
394 | * guarantee all CPU's that are in middle of receiving packets |
395 | * will see the new packet type (until the next received packet). | |
396 | */ | |
397 | ||
398 | void dev_add_pack(struct packet_type *pt) | |
399 | { | |
c07b68e8 | 400 | struct list_head *head = ptype_head(pt); |
1da177e4 | 401 | |
c07b68e8 ED |
402 | spin_lock(&ptype_lock); |
403 | list_add_rcu(&pt->list, head); | |
404 | spin_unlock(&ptype_lock); | |
1da177e4 | 405 | } |
d1b19dff | 406 | EXPORT_SYMBOL(dev_add_pack); |
1da177e4 | 407 | |
1da177e4 LT |
408 | /** |
409 | * __dev_remove_pack - remove packet handler | |
410 | * @pt: packet type declaration | |
411 | * | |
412 | * Remove a protocol handler that was previously added to the kernel | |
413 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
414 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 415 | * returns. |
1da177e4 LT |
416 | * |
417 | * The packet type might still be in use by receivers | |
418 | * and must not be freed until after all the CPU's have gone | |
419 | * through a quiescent state. | |
420 | */ | |
421 | void __dev_remove_pack(struct packet_type *pt) | |
422 | { | |
c07b68e8 | 423 | struct list_head *head = ptype_head(pt); |
1da177e4 LT |
424 | struct packet_type *pt1; |
425 | ||
c07b68e8 | 426 | spin_lock(&ptype_lock); |
1da177e4 LT |
427 | |
428 | list_for_each_entry(pt1, head, list) { | |
429 | if (pt == pt1) { | |
430 | list_del_rcu(&pt->list); | |
431 | goto out; | |
432 | } | |
433 | } | |
434 | ||
435 | printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt); | |
436 | out: | |
c07b68e8 | 437 | spin_unlock(&ptype_lock); |
1da177e4 | 438 | } |
d1b19dff ED |
439 | EXPORT_SYMBOL(__dev_remove_pack); |
440 | ||
1da177e4 LT |
441 | /** |
442 | * dev_remove_pack - remove packet handler | |
443 | * @pt: packet type declaration | |
444 | * | |
445 | * Remove a protocol handler that was previously added to the kernel | |
446 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
447 | * from the kernel lists and can be freed or reused once this function | |
448 | * returns. | |
449 | * | |
450 | * This call sleeps to guarantee that no CPU is looking at the packet | |
451 | * type after return. | |
452 | */ | |
453 | void dev_remove_pack(struct packet_type *pt) | |
454 | { | |
455 | __dev_remove_pack(pt); | |
4ec93edb | 456 | |
1da177e4 LT |
457 | synchronize_net(); |
458 | } | |
d1b19dff | 459 | EXPORT_SYMBOL(dev_remove_pack); |
1da177e4 LT |
460 | |
461 | /****************************************************************************** | |
462 | ||
463 | Device Boot-time Settings Routines | |
464 | ||
465 | *******************************************************************************/ | |
466 | ||
467 | /* Boot time configuration table */ | |
468 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
469 | ||
470 | /** | |
471 | * netdev_boot_setup_add - add new setup entry | |
472 | * @name: name of the device | |
473 | * @map: configured settings for the device | |
474 | * | |
475 | * Adds new setup entry to the dev_boot_setup list. The function | |
476 | * returns 0 on error and 1 on success. This is a generic routine to | |
477 | * all netdevices. | |
478 | */ | |
479 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
480 | { | |
481 | struct netdev_boot_setup *s; | |
482 | int i; | |
483 | ||
484 | s = dev_boot_setup; | |
485 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
486 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
487 | memset(s[i].name, 0, sizeof(s[i].name)); | |
93b3cff9 | 488 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
489 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
490 | break; | |
491 | } | |
492 | } | |
493 | ||
494 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
495 | } | |
496 | ||
497 | /** | |
498 | * netdev_boot_setup_check - check boot time settings | |
499 | * @dev: the netdevice | |
500 | * | |
501 | * Check boot time settings for the device. | |
502 | * The found settings are set for the device to be used | |
503 | * later in the device probing. | |
504 | * Returns 0 if no settings found, 1 if they are. | |
505 | */ | |
506 | int netdev_boot_setup_check(struct net_device *dev) | |
507 | { | |
508 | struct netdev_boot_setup *s = dev_boot_setup; | |
509 | int i; | |
510 | ||
511 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
512 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
93b3cff9 | 513 | !strcmp(dev->name, s[i].name)) { |
1da177e4 LT |
514 | dev->irq = s[i].map.irq; |
515 | dev->base_addr = s[i].map.base_addr; | |
516 | dev->mem_start = s[i].map.mem_start; | |
517 | dev->mem_end = s[i].map.mem_end; | |
518 | return 1; | |
519 | } | |
520 | } | |
521 | return 0; | |
522 | } | |
d1b19dff | 523 | EXPORT_SYMBOL(netdev_boot_setup_check); |
1da177e4 LT |
524 | |
525 | ||
526 | /** | |
527 | * netdev_boot_base - get address from boot time settings | |
528 | * @prefix: prefix for network device | |
529 | * @unit: id for network device | |
530 | * | |
531 | * Check boot time settings for the base address of device. | |
532 | * The found settings are set for the device to be used | |
533 | * later in the device probing. | |
534 | * Returns 0 if no settings found. | |
535 | */ | |
536 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
537 | { | |
538 | const struct netdev_boot_setup *s = dev_boot_setup; | |
539 | char name[IFNAMSIZ]; | |
540 | int i; | |
541 | ||
542 | sprintf(name, "%s%d", prefix, unit); | |
543 | ||
544 | /* | |
545 | * If device already registered then return base of 1 | |
546 | * to indicate not to probe for this interface | |
547 | */ | |
881d966b | 548 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
549 | return 1; |
550 | ||
551 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
552 | if (!strcmp(name, s[i].name)) | |
553 | return s[i].map.base_addr; | |
554 | return 0; | |
555 | } | |
556 | ||
557 | /* | |
558 | * Saves at boot time configured settings for any netdevice. | |
559 | */ | |
560 | int __init netdev_boot_setup(char *str) | |
561 | { | |
562 | int ints[5]; | |
563 | struct ifmap map; | |
564 | ||
565 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
566 | if (!str || !*str) | |
567 | return 0; | |
568 | ||
569 | /* Save settings */ | |
570 | memset(&map, 0, sizeof(map)); | |
571 | if (ints[0] > 0) | |
572 | map.irq = ints[1]; | |
573 | if (ints[0] > 1) | |
574 | map.base_addr = ints[2]; | |
575 | if (ints[0] > 2) | |
576 | map.mem_start = ints[3]; | |
577 | if (ints[0] > 3) | |
578 | map.mem_end = ints[4]; | |
579 | ||
580 | /* Add new entry to the list */ | |
581 | return netdev_boot_setup_add(str, &map); | |
582 | } | |
583 | ||
584 | __setup("netdev=", netdev_boot_setup); | |
585 | ||
586 | /******************************************************************************* | |
587 | ||
588 | Device Interface Subroutines | |
589 | ||
590 | *******************************************************************************/ | |
591 | ||
592 | /** | |
593 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 594 | * @net: the applicable net namespace |
1da177e4 LT |
595 | * @name: name to find |
596 | * | |
597 | * Find an interface by name. Must be called under RTNL semaphore | |
598 | * or @dev_base_lock. If the name is found a pointer to the device | |
599 | * is returned. If the name is not found then %NULL is returned. The | |
600 | * reference counters are not incremented so the caller must be | |
601 | * careful with locks. | |
602 | */ | |
603 | ||
881d966b | 604 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
605 | { |
606 | struct hlist_node *p; | |
0bd8d536 ED |
607 | struct net_device *dev; |
608 | struct hlist_head *head = dev_name_hash(net, name); | |
1da177e4 | 609 | |
0bd8d536 | 610 | hlist_for_each_entry(dev, p, head, name_hlist) |
1da177e4 LT |
611 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
612 | return dev; | |
0bd8d536 | 613 | |
1da177e4 LT |
614 | return NULL; |
615 | } | |
d1b19dff | 616 | EXPORT_SYMBOL(__dev_get_by_name); |
1da177e4 | 617 | |
72c9528b ED |
618 | /** |
619 | * dev_get_by_name_rcu - find a device by its name | |
620 | * @net: the applicable net namespace | |
621 | * @name: name to find | |
622 | * | |
623 | * Find an interface by name. | |
624 | * If the name is found a pointer to the device is returned. | |
625 | * If the name is not found then %NULL is returned. | |
626 | * The reference counters are not incremented so the caller must be | |
627 | * careful with locks. The caller must hold RCU lock. | |
628 | */ | |
629 | ||
630 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
631 | { | |
632 | struct hlist_node *p; | |
633 | struct net_device *dev; | |
634 | struct hlist_head *head = dev_name_hash(net, name); | |
635 | ||
636 | hlist_for_each_entry_rcu(dev, p, head, name_hlist) | |
637 | if (!strncmp(dev->name, name, IFNAMSIZ)) | |
638 | return dev; | |
639 | ||
640 | return NULL; | |
641 | } | |
642 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
643 | ||
1da177e4 LT |
644 | /** |
645 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 646 | * @net: the applicable net namespace |
1da177e4 LT |
647 | * @name: name to find |
648 | * | |
649 | * Find an interface by name. This can be called from any | |
650 | * context and does its own locking. The returned handle has | |
651 | * the usage count incremented and the caller must use dev_put() to | |
652 | * release it when it is no longer needed. %NULL is returned if no | |
653 | * matching device is found. | |
654 | */ | |
655 | ||
881d966b | 656 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
657 | { |
658 | struct net_device *dev; | |
659 | ||
72c9528b ED |
660 | rcu_read_lock(); |
661 | dev = dev_get_by_name_rcu(net, name); | |
1da177e4 LT |
662 | if (dev) |
663 | dev_hold(dev); | |
72c9528b | 664 | rcu_read_unlock(); |
1da177e4 LT |
665 | return dev; |
666 | } | |
d1b19dff | 667 | EXPORT_SYMBOL(dev_get_by_name); |
1da177e4 LT |
668 | |
669 | /** | |
670 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 671 | * @net: the applicable net namespace |
1da177e4 LT |
672 | * @ifindex: index of device |
673 | * | |
674 | * Search for an interface by index. Returns %NULL if the device | |
675 | * is not found or a pointer to the device. The device has not | |
676 | * had its reference counter increased so the caller must be careful | |
677 | * about locking. The caller must hold either the RTNL semaphore | |
678 | * or @dev_base_lock. | |
679 | */ | |
680 | ||
881d966b | 681 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
682 | { |
683 | struct hlist_node *p; | |
0bd8d536 ED |
684 | struct net_device *dev; |
685 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
1da177e4 | 686 | |
0bd8d536 | 687 | hlist_for_each_entry(dev, p, head, index_hlist) |
1da177e4 LT |
688 | if (dev->ifindex == ifindex) |
689 | return dev; | |
0bd8d536 | 690 | |
1da177e4 LT |
691 | return NULL; |
692 | } | |
d1b19dff | 693 | EXPORT_SYMBOL(__dev_get_by_index); |
1da177e4 | 694 | |
fb699dfd ED |
695 | /** |
696 | * dev_get_by_index_rcu - find a device by its ifindex | |
697 | * @net: the applicable net namespace | |
698 | * @ifindex: index of device | |
699 | * | |
700 | * Search for an interface by index. Returns %NULL if the device | |
701 | * is not found or a pointer to the device. The device has not | |
702 | * had its reference counter increased so the caller must be careful | |
703 | * about locking. The caller must hold RCU lock. | |
704 | */ | |
705 | ||
706 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
707 | { | |
708 | struct hlist_node *p; | |
709 | struct net_device *dev; | |
710 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
711 | ||
712 | hlist_for_each_entry_rcu(dev, p, head, index_hlist) | |
713 | if (dev->ifindex == ifindex) | |
714 | return dev; | |
715 | ||
716 | return NULL; | |
717 | } | |
718 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
719 | ||
1da177e4 LT |
720 | |
721 | /** | |
722 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 723 | * @net: the applicable net namespace |
1da177e4 LT |
724 | * @ifindex: index of device |
725 | * | |
726 | * Search for an interface by index. Returns NULL if the device | |
727 | * is not found or a pointer to the device. The device returned has | |
728 | * had a reference added and the pointer is safe until the user calls | |
729 | * dev_put to indicate they have finished with it. | |
730 | */ | |
731 | ||
881d966b | 732 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
733 | { |
734 | struct net_device *dev; | |
735 | ||
fb699dfd ED |
736 | rcu_read_lock(); |
737 | dev = dev_get_by_index_rcu(net, ifindex); | |
1da177e4 LT |
738 | if (dev) |
739 | dev_hold(dev); | |
fb699dfd | 740 | rcu_read_unlock(); |
1da177e4 LT |
741 | return dev; |
742 | } | |
d1b19dff | 743 | EXPORT_SYMBOL(dev_get_by_index); |
1da177e4 LT |
744 | |
745 | /** | |
941666c2 | 746 | * dev_getbyhwaddr_rcu - find a device by its hardware address |
c4ea43c5 | 747 | * @net: the applicable net namespace |
1da177e4 LT |
748 | * @type: media type of device |
749 | * @ha: hardware address | |
750 | * | |
751 | * Search for an interface by MAC address. Returns NULL if the device | |
941666c2 ED |
752 | * is not found or a pointer to the device. The caller must hold RCU |
753 | * The returned device has not had its ref count increased | |
1da177e4 LT |
754 | * and the caller must therefore be careful about locking |
755 | * | |
1da177e4 LT |
756 | */ |
757 | ||
941666c2 ED |
758 | struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, |
759 | const char *ha) | |
1da177e4 LT |
760 | { |
761 | struct net_device *dev; | |
762 | ||
941666c2 | 763 | for_each_netdev_rcu(net, dev) |
1da177e4 LT |
764 | if (dev->type == type && |
765 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
766 | return dev; |
767 | ||
768 | return NULL; | |
1da177e4 | 769 | } |
941666c2 | 770 | EXPORT_SYMBOL(dev_getbyhwaddr_rcu); |
cf309e3f | 771 | |
881d966b | 772 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
773 | { |
774 | struct net_device *dev; | |
775 | ||
4e9cac2b | 776 | ASSERT_RTNL(); |
881d966b | 777 | for_each_netdev(net, dev) |
4e9cac2b | 778 | if (dev->type == type) |
7562f876 PE |
779 | return dev; |
780 | ||
781 | return NULL; | |
4e9cac2b | 782 | } |
4e9cac2b PM |
783 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); |
784 | ||
881d966b | 785 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b | 786 | { |
99fe3c39 | 787 | struct net_device *dev, *ret = NULL; |
4e9cac2b | 788 | |
99fe3c39 ED |
789 | rcu_read_lock(); |
790 | for_each_netdev_rcu(net, dev) | |
791 | if (dev->type == type) { | |
792 | dev_hold(dev); | |
793 | ret = dev; | |
794 | break; | |
795 | } | |
796 | rcu_read_unlock(); | |
797 | return ret; | |
1da177e4 | 798 | } |
1da177e4 LT |
799 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
800 | ||
801 | /** | |
bb69ae04 | 802 | * dev_get_by_flags_rcu - find any device with given flags |
c4ea43c5 | 803 | * @net: the applicable net namespace |
1da177e4 LT |
804 | * @if_flags: IFF_* values |
805 | * @mask: bitmask of bits in if_flags to check | |
806 | * | |
807 | * Search for any interface with the given flags. Returns NULL if a device | |
bb69ae04 ED |
808 | * is not found or a pointer to the device. Must be called inside |
809 | * rcu_read_lock(), and result refcount is unchanged. | |
1da177e4 LT |
810 | */ |
811 | ||
bb69ae04 | 812 | struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags, |
d1b19dff | 813 | unsigned short mask) |
1da177e4 | 814 | { |
7562f876 | 815 | struct net_device *dev, *ret; |
1da177e4 | 816 | |
7562f876 | 817 | ret = NULL; |
c6d14c84 | 818 | for_each_netdev_rcu(net, dev) { |
1da177e4 | 819 | if (((dev->flags ^ if_flags) & mask) == 0) { |
7562f876 | 820 | ret = dev; |
1da177e4 LT |
821 | break; |
822 | } | |
823 | } | |
7562f876 | 824 | return ret; |
1da177e4 | 825 | } |
bb69ae04 | 826 | EXPORT_SYMBOL(dev_get_by_flags_rcu); |
1da177e4 LT |
827 | |
828 | /** | |
829 | * dev_valid_name - check if name is okay for network device | |
830 | * @name: name string | |
831 | * | |
832 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
833 | * to allow sysfs to work. We also disallow any kind of |
834 | * whitespace. | |
1da177e4 | 835 | */ |
c2373ee9 | 836 | int dev_valid_name(const char *name) |
1da177e4 | 837 | { |
c7fa9d18 DM |
838 | if (*name == '\0') |
839 | return 0; | |
b6fe17d6 SH |
840 | if (strlen(name) >= IFNAMSIZ) |
841 | return 0; | |
c7fa9d18 DM |
842 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
843 | return 0; | |
844 | ||
845 | while (*name) { | |
846 | if (*name == '/' || isspace(*name)) | |
847 | return 0; | |
848 | name++; | |
849 | } | |
850 | return 1; | |
1da177e4 | 851 | } |
d1b19dff | 852 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
853 | |
854 | /** | |
b267b179 EB |
855 | * __dev_alloc_name - allocate a name for a device |
856 | * @net: network namespace to allocate the device name in | |
1da177e4 | 857 | * @name: name format string |
b267b179 | 858 | * @buf: scratch buffer and result name string |
1da177e4 LT |
859 | * |
860 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
861 | * id. It scans list of devices to build up a free map, then chooses |
862 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
863 | * while allocating the name and adding the device in order to avoid | |
864 | * duplicates. | |
865 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
866 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
867 | */ |
868 | ||
b267b179 | 869 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
870 | { |
871 | int i = 0; | |
1da177e4 LT |
872 | const char *p; |
873 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 874 | unsigned long *inuse; |
1da177e4 LT |
875 | struct net_device *d; |
876 | ||
877 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
878 | if (p) { | |
879 | /* | |
880 | * Verify the string as this thing may have come from | |
881 | * the user. There must be either one "%d" and no other "%" | |
882 | * characters. | |
883 | */ | |
884 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
885 | return -EINVAL; | |
886 | ||
887 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 888 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
889 | if (!inuse) |
890 | return -ENOMEM; | |
891 | ||
881d966b | 892 | for_each_netdev(net, d) { |
1da177e4 LT |
893 | if (!sscanf(d->name, name, &i)) |
894 | continue; | |
895 | if (i < 0 || i >= max_netdevices) | |
896 | continue; | |
897 | ||
898 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 899 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
900 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
901 | set_bit(i, inuse); | |
902 | } | |
903 | ||
904 | i = find_first_zero_bit(inuse, max_netdevices); | |
905 | free_page((unsigned long) inuse); | |
906 | } | |
907 | ||
d9031024 OP |
908 | if (buf != name) |
909 | snprintf(buf, IFNAMSIZ, name, i); | |
b267b179 | 910 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 911 | return i; |
1da177e4 LT |
912 | |
913 | /* It is possible to run out of possible slots | |
914 | * when the name is long and there isn't enough space left | |
915 | * for the digits, or if all bits are used. | |
916 | */ | |
917 | return -ENFILE; | |
918 | } | |
919 | ||
b267b179 EB |
920 | /** |
921 | * dev_alloc_name - allocate a name for a device | |
922 | * @dev: device | |
923 | * @name: name format string | |
924 | * | |
925 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
926 | * id. It scans list of devices to build up a free map, then chooses | |
927 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
928 | * while allocating the name and adding the device in order to avoid | |
929 | * duplicates. | |
930 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
931 | * Returns the number of the unit assigned or a negative errno code. | |
932 | */ | |
933 | ||
934 | int dev_alloc_name(struct net_device *dev, const char *name) | |
935 | { | |
936 | char buf[IFNAMSIZ]; | |
937 | struct net *net; | |
938 | int ret; | |
939 | ||
c346dca1 YH |
940 | BUG_ON(!dev_net(dev)); |
941 | net = dev_net(dev); | |
b267b179 EB |
942 | ret = __dev_alloc_name(net, name, buf); |
943 | if (ret >= 0) | |
944 | strlcpy(dev->name, buf, IFNAMSIZ); | |
945 | return ret; | |
946 | } | |
d1b19dff | 947 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 948 | |
8ce6cebc | 949 | static int dev_get_valid_name(struct net_device *dev, const char *name, bool fmt) |
d9031024 | 950 | { |
8ce6cebc DL |
951 | struct net *net; |
952 | ||
953 | BUG_ON(!dev_net(dev)); | |
954 | net = dev_net(dev); | |
955 | ||
d9031024 OP |
956 | if (!dev_valid_name(name)) |
957 | return -EINVAL; | |
958 | ||
959 | if (fmt && strchr(name, '%')) | |
8ce6cebc | 960 | return dev_alloc_name(dev, name); |
d9031024 OP |
961 | else if (__dev_get_by_name(net, name)) |
962 | return -EEXIST; | |
8ce6cebc DL |
963 | else if (dev->name != name) |
964 | strlcpy(dev->name, name, IFNAMSIZ); | |
d9031024 OP |
965 | |
966 | return 0; | |
967 | } | |
1da177e4 LT |
968 | |
969 | /** | |
970 | * dev_change_name - change name of a device | |
971 | * @dev: device | |
972 | * @newname: name (or format string) must be at least IFNAMSIZ | |
973 | * | |
974 | * Change name of a device, can pass format strings "eth%d". | |
975 | * for wildcarding. | |
976 | */ | |
cf04a4c7 | 977 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 978 | { |
fcc5a03a | 979 | char oldname[IFNAMSIZ]; |
1da177e4 | 980 | int err = 0; |
fcc5a03a | 981 | int ret; |
881d966b | 982 | struct net *net; |
1da177e4 LT |
983 | |
984 | ASSERT_RTNL(); | |
c346dca1 | 985 | BUG_ON(!dev_net(dev)); |
1da177e4 | 986 | |
c346dca1 | 987 | net = dev_net(dev); |
1da177e4 LT |
988 | if (dev->flags & IFF_UP) |
989 | return -EBUSY; | |
990 | ||
c8d90dca SH |
991 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) |
992 | return 0; | |
993 | ||
fcc5a03a HX |
994 | memcpy(oldname, dev->name, IFNAMSIZ); |
995 | ||
8ce6cebc | 996 | err = dev_get_valid_name(dev, newname, 1); |
d9031024 OP |
997 | if (err < 0) |
998 | return err; | |
1da177e4 | 999 | |
fcc5a03a | 1000 | rollback: |
a1b3f594 EB |
1001 | ret = device_rename(&dev->dev, dev->name); |
1002 | if (ret) { | |
1003 | memcpy(dev->name, oldname, IFNAMSIZ); | |
1004 | return ret; | |
dcc99773 | 1005 | } |
7f988eab HX |
1006 | |
1007 | write_lock_bh(&dev_base_lock); | |
92749821 | 1008 | hlist_del(&dev->name_hlist); |
72c9528b ED |
1009 | write_unlock_bh(&dev_base_lock); |
1010 | ||
1011 | synchronize_rcu(); | |
1012 | ||
1013 | write_lock_bh(&dev_base_lock); | |
1014 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
7f988eab HX |
1015 | write_unlock_bh(&dev_base_lock); |
1016 | ||
056925ab | 1017 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1018 | ret = notifier_to_errno(ret); |
1019 | ||
1020 | if (ret) { | |
91e9c07b ED |
1021 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1022 | if (err >= 0) { | |
fcc5a03a HX |
1023 | err = ret; |
1024 | memcpy(dev->name, oldname, IFNAMSIZ); | |
1025 | goto rollback; | |
91e9c07b ED |
1026 | } else { |
1027 | printk(KERN_ERR | |
1028 | "%s: name change rollback failed: %d.\n", | |
1029 | dev->name, ret); | |
fcc5a03a HX |
1030 | } |
1031 | } | |
1da177e4 LT |
1032 | |
1033 | return err; | |
1034 | } | |
1035 | ||
0b815a1a SH |
1036 | /** |
1037 | * dev_set_alias - change ifalias of a device | |
1038 | * @dev: device | |
1039 | * @alias: name up to IFALIASZ | |
f0db275a | 1040 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1041 | * |
1042 | * Set ifalias for a device, | |
1043 | */ | |
1044 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1045 | { | |
1046 | ASSERT_RTNL(); | |
1047 | ||
1048 | if (len >= IFALIASZ) | |
1049 | return -EINVAL; | |
1050 | ||
96ca4a2c OH |
1051 | if (!len) { |
1052 | if (dev->ifalias) { | |
1053 | kfree(dev->ifalias); | |
1054 | dev->ifalias = NULL; | |
1055 | } | |
1056 | return 0; | |
1057 | } | |
1058 | ||
d1b19dff | 1059 | dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL); |
0b815a1a SH |
1060 | if (!dev->ifalias) |
1061 | return -ENOMEM; | |
1062 | ||
1063 | strlcpy(dev->ifalias, alias, len+1); | |
1064 | return len; | |
1065 | } | |
1066 | ||
1067 | ||
d8a33ac4 | 1068 | /** |
3041a069 | 1069 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1070 | * @dev: device to cause notification |
1071 | * | |
1072 | * Called to indicate a device has changed features. | |
1073 | */ | |
1074 | void netdev_features_change(struct net_device *dev) | |
1075 | { | |
056925ab | 1076 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1077 | } |
1078 | EXPORT_SYMBOL(netdev_features_change); | |
1079 | ||
1da177e4 LT |
1080 | /** |
1081 | * netdev_state_change - device changes state | |
1082 | * @dev: device to cause notification | |
1083 | * | |
1084 | * Called to indicate a device has changed state. This function calls | |
1085 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1086 | * to the routing socket. | |
1087 | */ | |
1088 | void netdev_state_change(struct net_device *dev) | |
1089 | { | |
1090 | if (dev->flags & IFF_UP) { | |
056925ab | 1091 | call_netdevice_notifiers(NETDEV_CHANGE, dev); |
1da177e4 LT |
1092 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0); |
1093 | } | |
1094 | } | |
d1b19dff | 1095 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1096 | |
3ca5b404 | 1097 | int netdev_bonding_change(struct net_device *dev, unsigned long event) |
c1da4ac7 | 1098 | { |
3ca5b404 | 1099 | return call_netdevice_notifiers(event, dev); |
c1da4ac7 OG |
1100 | } |
1101 | EXPORT_SYMBOL(netdev_bonding_change); | |
1102 | ||
1da177e4 LT |
1103 | /** |
1104 | * dev_load - load a network module | |
c4ea43c5 | 1105 | * @net: the applicable net namespace |
1da177e4 LT |
1106 | * @name: name of interface |
1107 | * | |
1108 | * If a network interface is not present and the process has suitable | |
1109 | * privileges this function loads the module. If module loading is not | |
1110 | * available in this kernel then it becomes a nop. | |
1111 | */ | |
1112 | ||
881d966b | 1113 | void dev_load(struct net *net, const char *name) |
1da177e4 | 1114 | { |
4ec93edb | 1115 | struct net_device *dev; |
1da177e4 | 1116 | |
72c9528b ED |
1117 | rcu_read_lock(); |
1118 | dev = dev_get_by_name_rcu(net, name); | |
1119 | rcu_read_unlock(); | |
1da177e4 | 1120 | |
a8f80e8f | 1121 | if (!dev && capable(CAP_NET_ADMIN)) |
1da177e4 LT |
1122 | request_module("%s", name); |
1123 | } | |
d1b19dff | 1124 | EXPORT_SYMBOL(dev_load); |
1da177e4 | 1125 | |
bd380811 | 1126 | static int __dev_open(struct net_device *dev) |
1da177e4 | 1127 | { |
d314774c | 1128 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1129 | int ret; |
1da177e4 | 1130 | |
e46b66bc BH |
1131 | ASSERT_RTNL(); |
1132 | ||
1da177e4 LT |
1133 | /* |
1134 | * Is it even present? | |
1135 | */ | |
1136 | if (!netif_device_present(dev)) | |
1137 | return -ENODEV; | |
1138 | ||
3b8bcfd5 JB |
1139 | ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); |
1140 | ret = notifier_to_errno(ret); | |
1141 | if (ret) | |
1142 | return ret; | |
1143 | ||
1da177e4 LT |
1144 | /* |
1145 | * Call device private open method | |
1146 | */ | |
1147 | set_bit(__LINK_STATE_START, &dev->state); | |
bada339b | 1148 | |
d314774c SH |
1149 | if (ops->ndo_validate_addr) |
1150 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1151 | |
d314774c SH |
1152 | if (!ret && ops->ndo_open) |
1153 | ret = ops->ndo_open(dev); | |
1da177e4 | 1154 | |
4ec93edb | 1155 | /* |
1da177e4 LT |
1156 | * If it went open OK then: |
1157 | */ | |
1158 | ||
bada339b JG |
1159 | if (ret) |
1160 | clear_bit(__LINK_STATE_START, &dev->state); | |
1161 | else { | |
1da177e4 LT |
1162 | /* |
1163 | * Set the flags. | |
1164 | */ | |
1165 | dev->flags |= IFF_UP; | |
1166 | ||
649274d9 DW |
1167 | /* |
1168 | * Enable NET_DMA | |
1169 | */ | |
b4bd07c2 | 1170 | net_dmaengine_get(); |
649274d9 | 1171 | |
1da177e4 LT |
1172 | /* |
1173 | * Initialize multicasting status | |
1174 | */ | |
4417da66 | 1175 | dev_set_rx_mode(dev); |
1da177e4 LT |
1176 | |
1177 | /* | |
1178 | * Wakeup transmit queue engine | |
1179 | */ | |
1180 | dev_activate(dev); | |
1da177e4 | 1181 | } |
bada339b | 1182 | |
1da177e4 LT |
1183 | return ret; |
1184 | } | |
1185 | ||
1186 | /** | |
bd380811 PM |
1187 | * dev_open - prepare an interface for use. |
1188 | * @dev: device to open | |
1da177e4 | 1189 | * |
bd380811 PM |
1190 | * Takes a device from down to up state. The device's private open |
1191 | * function is invoked and then the multicast lists are loaded. Finally | |
1192 | * the device is moved into the up state and a %NETDEV_UP message is | |
1193 | * sent to the netdev notifier chain. | |
1194 | * | |
1195 | * Calling this function on an active interface is a nop. On a failure | |
1196 | * a negative errno code is returned. | |
1da177e4 | 1197 | */ |
bd380811 PM |
1198 | int dev_open(struct net_device *dev) |
1199 | { | |
1200 | int ret; | |
1201 | ||
1202 | /* | |
1203 | * Is it already up? | |
1204 | */ | |
1205 | if (dev->flags & IFF_UP) | |
1206 | return 0; | |
1207 | ||
1208 | /* | |
1209 | * Open device | |
1210 | */ | |
1211 | ret = __dev_open(dev); | |
1212 | if (ret < 0) | |
1213 | return ret; | |
1214 | ||
1215 | /* | |
1216 | * ... and announce new interface. | |
1217 | */ | |
1218 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING); | |
1219 | call_netdevice_notifiers(NETDEV_UP, dev); | |
1220 | ||
1221 | return ret; | |
1222 | } | |
1223 | EXPORT_SYMBOL(dev_open); | |
1224 | ||
44345724 | 1225 | static int __dev_close_many(struct list_head *head) |
1da177e4 | 1226 | { |
44345724 | 1227 | struct net_device *dev; |
e46b66bc | 1228 | |
bd380811 | 1229 | ASSERT_RTNL(); |
9d5010db DM |
1230 | might_sleep(); |
1231 | ||
44345724 OP |
1232 | list_for_each_entry(dev, head, unreg_list) { |
1233 | /* | |
1234 | * Tell people we are going down, so that they can | |
1235 | * prepare to death, when device is still operating. | |
1236 | */ | |
1237 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); | |
1da177e4 | 1238 | |
44345724 | 1239 | clear_bit(__LINK_STATE_START, &dev->state); |
1da177e4 | 1240 | |
44345724 OP |
1241 | /* Synchronize to scheduled poll. We cannot touch poll list, it |
1242 | * can be even on different cpu. So just clear netif_running(). | |
1243 | * | |
1244 | * dev->stop() will invoke napi_disable() on all of it's | |
1245 | * napi_struct instances on this device. | |
1246 | */ | |
1247 | smp_mb__after_clear_bit(); /* Commit netif_running(). */ | |
1248 | } | |
1da177e4 | 1249 | |
44345724 | 1250 | dev_deactivate_many(head); |
d8b2a4d2 | 1251 | |
44345724 OP |
1252 | list_for_each_entry(dev, head, unreg_list) { |
1253 | const struct net_device_ops *ops = dev->netdev_ops; | |
1da177e4 | 1254 | |
44345724 OP |
1255 | /* |
1256 | * Call the device specific close. This cannot fail. | |
1257 | * Only if device is UP | |
1258 | * | |
1259 | * We allow it to be called even after a DETACH hot-plug | |
1260 | * event. | |
1261 | */ | |
1262 | if (ops->ndo_stop) | |
1263 | ops->ndo_stop(dev); | |
1264 | ||
1265 | /* | |
1266 | * Device is now down. | |
1267 | */ | |
1268 | ||
1269 | dev->flags &= ~IFF_UP; | |
1270 | ||
1271 | /* | |
1272 | * Shutdown NET_DMA | |
1273 | */ | |
1274 | net_dmaengine_put(); | |
1275 | } | |
1276 | ||
1277 | return 0; | |
1278 | } | |
1279 | ||
1280 | static int __dev_close(struct net_device *dev) | |
1281 | { | |
1282 | LIST_HEAD(single); | |
1283 | ||
1284 | list_add(&dev->unreg_list, &single); | |
1285 | return __dev_close_many(&single); | |
1286 | } | |
1287 | ||
1288 | int dev_close_many(struct list_head *head) | |
1289 | { | |
1290 | struct net_device *dev, *tmp; | |
1291 | LIST_HEAD(tmp_list); | |
1da177e4 | 1292 | |
44345724 OP |
1293 | list_for_each_entry_safe(dev, tmp, head, unreg_list) |
1294 | if (!(dev->flags & IFF_UP)) | |
1295 | list_move(&dev->unreg_list, &tmp_list); | |
1296 | ||
1297 | __dev_close_many(head); | |
1da177e4 LT |
1298 | |
1299 | /* | |
44345724 | 1300 | * Tell people we are down |
1da177e4 | 1301 | */ |
44345724 OP |
1302 | list_for_each_entry(dev, head, unreg_list) { |
1303 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING); | |
1304 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
1305 | } | |
bd380811 | 1306 | |
44345724 OP |
1307 | /* rollback_registered_many needs the complete original list */ |
1308 | list_splice(&tmp_list, head); | |
bd380811 PM |
1309 | return 0; |
1310 | } | |
1311 | ||
1312 | /** | |
1313 | * dev_close - shutdown an interface. | |
1314 | * @dev: device to shutdown | |
1315 | * | |
1316 | * This function moves an active device into down state. A | |
1317 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1318 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1319 | * chain. | |
1320 | */ | |
1321 | int dev_close(struct net_device *dev) | |
1322 | { | |
44345724 | 1323 | LIST_HEAD(single); |
1da177e4 | 1324 | |
44345724 OP |
1325 | list_add(&dev->unreg_list, &single); |
1326 | dev_close_many(&single); | |
649274d9 | 1327 | |
1da177e4 LT |
1328 | return 0; |
1329 | } | |
d1b19dff | 1330 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1331 | |
1332 | ||
0187bdfb BH |
1333 | /** |
1334 | * dev_disable_lro - disable Large Receive Offload on a device | |
1335 | * @dev: device | |
1336 | * | |
1337 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1338 | * called under RTNL. This is needed if received packets may be | |
1339 | * forwarded to another interface. | |
1340 | */ | |
1341 | void dev_disable_lro(struct net_device *dev) | |
1342 | { | |
1343 | if (dev->ethtool_ops && dev->ethtool_ops->get_flags && | |
1344 | dev->ethtool_ops->set_flags) { | |
1345 | u32 flags = dev->ethtool_ops->get_flags(dev); | |
1346 | if (flags & ETH_FLAG_LRO) { | |
1347 | flags &= ~ETH_FLAG_LRO; | |
1348 | dev->ethtool_ops->set_flags(dev, flags); | |
1349 | } | |
1350 | } | |
1351 | WARN_ON(dev->features & NETIF_F_LRO); | |
1352 | } | |
1353 | EXPORT_SYMBOL(dev_disable_lro); | |
1354 | ||
1355 | ||
881d966b EB |
1356 | static int dev_boot_phase = 1; |
1357 | ||
1da177e4 LT |
1358 | /* |
1359 | * Device change register/unregister. These are not inline or static | |
1360 | * as we export them to the world. | |
1361 | */ | |
1362 | ||
1363 | /** | |
1364 | * register_netdevice_notifier - register a network notifier block | |
1365 | * @nb: notifier | |
1366 | * | |
1367 | * Register a notifier to be called when network device events occur. | |
1368 | * The notifier passed is linked into the kernel structures and must | |
1369 | * not be reused until it has been unregistered. A negative errno code | |
1370 | * is returned on a failure. | |
1371 | * | |
1372 | * When registered all registration and up events are replayed | |
4ec93edb | 1373 | * to the new notifier to allow device to have a race free |
1da177e4 LT |
1374 | * view of the network device list. |
1375 | */ | |
1376 | ||
1377 | int register_netdevice_notifier(struct notifier_block *nb) | |
1378 | { | |
1379 | struct net_device *dev; | |
fcc5a03a | 1380 | struct net_device *last; |
881d966b | 1381 | struct net *net; |
1da177e4 LT |
1382 | int err; |
1383 | ||
1384 | rtnl_lock(); | |
f07d5b94 | 1385 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1386 | if (err) |
1387 | goto unlock; | |
881d966b EB |
1388 | if (dev_boot_phase) |
1389 | goto unlock; | |
1390 | for_each_net(net) { | |
1391 | for_each_netdev(net, dev) { | |
1392 | err = nb->notifier_call(nb, NETDEV_REGISTER, dev); | |
1393 | err = notifier_to_errno(err); | |
1394 | if (err) | |
1395 | goto rollback; | |
1396 | ||
1397 | if (!(dev->flags & IFF_UP)) | |
1398 | continue; | |
1da177e4 | 1399 | |
881d966b EB |
1400 | nb->notifier_call(nb, NETDEV_UP, dev); |
1401 | } | |
1da177e4 | 1402 | } |
fcc5a03a HX |
1403 | |
1404 | unlock: | |
1da177e4 LT |
1405 | rtnl_unlock(); |
1406 | return err; | |
fcc5a03a HX |
1407 | |
1408 | rollback: | |
1409 | last = dev; | |
881d966b EB |
1410 | for_each_net(net) { |
1411 | for_each_netdev(net, dev) { | |
1412 | if (dev == last) | |
1413 | break; | |
fcc5a03a | 1414 | |
881d966b EB |
1415 | if (dev->flags & IFF_UP) { |
1416 | nb->notifier_call(nb, NETDEV_GOING_DOWN, dev); | |
1417 | nb->notifier_call(nb, NETDEV_DOWN, dev); | |
1418 | } | |
1419 | nb->notifier_call(nb, NETDEV_UNREGISTER, dev); | |
a5ee1551 | 1420 | nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev); |
fcc5a03a | 1421 | } |
fcc5a03a | 1422 | } |
c67625a1 PE |
1423 | |
1424 | raw_notifier_chain_unregister(&netdev_chain, nb); | |
fcc5a03a | 1425 | goto unlock; |
1da177e4 | 1426 | } |
d1b19dff | 1427 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1428 | |
1429 | /** | |
1430 | * unregister_netdevice_notifier - unregister a network notifier block | |
1431 | * @nb: notifier | |
1432 | * | |
1433 | * Unregister a notifier previously registered by | |
1434 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1435 | * kernel structures and may then be reused. A negative errno code | |
1436 | * is returned on a failure. | |
1437 | */ | |
1438 | ||
1439 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1440 | { | |
9f514950 HX |
1441 | int err; |
1442 | ||
1443 | rtnl_lock(); | |
f07d5b94 | 1444 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
9f514950 HX |
1445 | rtnl_unlock(); |
1446 | return err; | |
1da177e4 | 1447 | } |
d1b19dff | 1448 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 LT |
1449 | |
1450 | /** | |
1451 | * call_netdevice_notifiers - call all network notifier blocks | |
1452 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1453 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1454 | * |
1455 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1456 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1457 | */ |
1458 | ||
ad7379d4 | 1459 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1460 | { |
ab930471 | 1461 | ASSERT_RTNL(); |
ad7379d4 | 1462 | return raw_notifier_call_chain(&netdev_chain, val, dev); |
1da177e4 LT |
1463 | } |
1464 | ||
1465 | /* When > 0 there are consumers of rx skb time stamps */ | |
1466 | static atomic_t netstamp_needed = ATOMIC_INIT(0); | |
1467 | ||
1468 | void net_enable_timestamp(void) | |
1469 | { | |
1470 | atomic_inc(&netstamp_needed); | |
1471 | } | |
d1b19dff | 1472 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
1473 | |
1474 | void net_disable_timestamp(void) | |
1475 | { | |
1476 | atomic_dec(&netstamp_needed); | |
1477 | } | |
d1b19dff | 1478 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 1479 | |
3b098e2d | 1480 | static inline void net_timestamp_set(struct sk_buff *skb) |
1da177e4 LT |
1481 | { |
1482 | if (atomic_read(&netstamp_needed)) | |
a61bbcf2 | 1483 | __net_timestamp(skb); |
b7aa0bf7 ED |
1484 | else |
1485 | skb->tstamp.tv64 = 0; | |
1da177e4 LT |
1486 | } |
1487 | ||
3b098e2d ED |
1488 | static inline void net_timestamp_check(struct sk_buff *skb) |
1489 | { | |
1490 | if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed)) | |
1491 | __net_timestamp(skb); | |
1492 | } | |
1493 | ||
44540960 AB |
1494 | /** |
1495 | * dev_forward_skb - loopback an skb to another netif | |
1496 | * | |
1497 | * @dev: destination network device | |
1498 | * @skb: buffer to forward | |
1499 | * | |
1500 | * return values: | |
1501 | * NET_RX_SUCCESS (no congestion) | |
6ec82562 | 1502 | * NET_RX_DROP (packet was dropped, but freed) |
44540960 AB |
1503 | * |
1504 | * dev_forward_skb can be used for injecting an skb from the | |
1505 | * start_xmit function of one device into the receive queue | |
1506 | * of another device. | |
1507 | * | |
1508 | * The receiving device may be in another namespace, so | |
1509 | * we have to clear all information in the skb that could | |
1510 | * impact namespace isolation. | |
1511 | */ | |
1512 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
1513 | { | |
1514 | skb_orphan(skb); | |
c736eefa | 1515 | nf_reset(skb); |
44540960 | 1516 | |
caf586e5 | 1517 | if (unlikely(!(dev->flags & IFF_UP) || |
2198a10b | 1518 | (skb->len > (dev->mtu + dev->hard_header_len + VLAN_HLEN)))) { |
caf586e5 | 1519 | atomic_long_inc(&dev->rx_dropped); |
6ec82562 | 1520 | kfree_skb(skb); |
44540960 | 1521 | return NET_RX_DROP; |
6ec82562 | 1522 | } |
8a83a00b | 1523 | skb_set_dev(skb, dev); |
44540960 AB |
1524 | skb->tstamp.tv64 = 0; |
1525 | skb->pkt_type = PACKET_HOST; | |
1526 | skb->protocol = eth_type_trans(skb, dev); | |
44540960 AB |
1527 | return netif_rx(skb); |
1528 | } | |
1529 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
1530 | ||
71d9dec2 CG |
1531 | static inline int deliver_skb(struct sk_buff *skb, |
1532 | struct packet_type *pt_prev, | |
1533 | struct net_device *orig_dev) | |
1534 | { | |
1535 | atomic_inc(&skb->users); | |
1536 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1537 | } | |
1538 | ||
1da177e4 LT |
1539 | /* |
1540 | * Support routine. Sends outgoing frames to any network | |
1541 | * taps currently in use. | |
1542 | */ | |
1543 | ||
f6a78bfc | 1544 | static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
1545 | { |
1546 | struct packet_type *ptype; | |
71d9dec2 CG |
1547 | struct sk_buff *skb2 = NULL; |
1548 | struct packet_type *pt_prev = NULL; | |
a61bbcf2 | 1549 | |
1da177e4 LT |
1550 | rcu_read_lock(); |
1551 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
1552 | /* Never send packets back to the socket | |
1553 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1554 | */ | |
1555 | if ((ptype->dev == dev || !ptype->dev) && | |
1556 | (ptype->af_packet_priv == NULL || | |
1557 | (struct sock *)ptype->af_packet_priv != skb->sk)) { | |
71d9dec2 CG |
1558 | if (pt_prev) { |
1559 | deliver_skb(skb2, pt_prev, skb->dev); | |
1560 | pt_prev = ptype; | |
1561 | continue; | |
1562 | } | |
1563 | ||
1564 | skb2 = skb_clone(skb, GFP_ATOMIC); | |
1da177e4 LT |
1565 | if (!skb2) |
1566 | break; | |
1567 | ||
70978182 ED |
1568 | net_timestamp_set(skb2); |
1569 | ||
1da177e4 LT |
1570 | /* skb->nh should be correctly |
1571 | set by sender, so that the second statement is | |
1572 | just protection against buggy protocols. | |
1573 | */ | |
459a98ed | 1574 | skb_reset_mac_header(skb2); |
1da177e4 | 1575 | |
d56f90a7 | 1576 | if (skb_network_header(skb2) < skb2->data || |
27a884dc | 1577 | skb2->network_header > skb2->tail) { |
1da177e4 LT |
1578 | if (net_ratelimit()) |
1579 | printk(KERN_CRIT "protocol %04x is " | |
1580 | "buggy, dev %s\n", | |
70777d03 SAS |
1581 | ntohs(skb2->protocol), |
1582 | dev->name); | |
c1d2bbe1 | 1583 | skb_reset_network_header(skb2); |
1da177e4 LT |
1584 | } |
1585 | ||
b0e380b1 | 1586 | skb2->transport_header = skb2->network_header; |
1da177e4 | 1587 | skb2->pkt_type = PACKET_OUTGOING; |
71d9dec2 | 1588 | pt_prev = ptype; |
1da177e4 LT |
1589 | } |
1590 | } | |
71d9dec2 CG |
1591 | if (pt_prev) |
1592 | pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); | |
1da177e4 LT |
1593 | rcu_read_unlock(); |
1594 | } | |
1595 | ||
4f57c087 JF |
1596 | /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change |
1597 | * @dev: Network device | |
1598 | * @txq: number of queues available | |
1599 | * | |
1600 | * If real_num_tx_queues is changed the tc mappings may no longer be | |
1601 | * valid. To resolve this verify the tc mapping remains valid and if | |
1602 | * not NULL the mapping. With no priorities mapping to this | |
1603 | * offset/count pair it will no longer be used. In the worst case TC0 | |
1604 | * is invalid nothing can be done so disable priority mappings. If is | |
1605 | * expected that drivers will fix this mapping if they can before | |
1606 | * calling netif_set_real_num_tx_queues. | |
1607 | */ | |
1608 | void netif_setup_tc(struct net_device *dev, unsigned int txq) | |
1609 | { | |
1610 | int i; | |
1611 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
1612 | ||
1613 | /* If TC0 is invalidated disable TC mapping */ | |
1614 | if (tc->offset + tc->count > txq) { | |
1615 | pr_warning("Number of in use tx queues changed " | |
1616 | "invalidating tc mappings. Priority " | |
1617 | "traffic classification disabled!\n"); | |
1618 | dev->num_tc = 0; | |
1619 | return; | |
1620 | } | |
1621 | ||
1622 | /* Invalidated prio to tc mappings set to TC0 */ | |
1623 | for (i = 1; i < TC_BITMASK + 1; i++) { | |
1624 | int q = netdev_get_prio_tc_map(dev, i); | |
1625 | ||
1626 | tc = &dev->tc_to_txq[q]; | |
1627 | if (tc->offset + tc->count > txq) { | |
1628 | pr_warning("Number of in use tx queues " | |
1629 | "changed. Priority %i to tc " | |
1630 | "mapping %i is no longer valid " | |
1631 | "setting map to 0\n", | |
1632 | i, q); | |
1633 | netdev_set_prio_tc_map(dev, i, 0); | |
1634 | } | |
1635 | } | |
1636 | } | |
1637 | ||
f0796d5c JF |
1638 | /* |
1639 | * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues | |
1640 | * greater then real_num_tx_queues stale skbs on the qdisc must be flushed. | |
1641 | */ | |
e6484930 | 1642 | int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) |
f0796d5c | 1643 | { |
1d24eb48 TH |
1644 | int rc; |
1645 | ||
e6484930 TH |
1646 | if (txq < 1 || txq > dev->num_tx_queues) |
1647 | return -EINVAL; | |
f0796d5c | 1648 | |
e6484930 TH |
1649 | if (dev->reg_state == NETREG_REGISTERED) { |
1650 | ASSERT_RTNL(); | |
1651 | ||
1d24eb48 TH |
1652 | rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, |
1653 | txq); | |
bf264145 TH |
1654 | if (rc) |
1655 | return rc; | |
1656 | ||
4f57c087 JF |
1657 | if (dev->num_tc) |
1658 | netif_setup_tc(dev, txq); | |
1659 | ||
e6484930 TH |
1660 | if (txq < dev->real_num_tx_queues) |
1661 | qdisc_reset_all_tx_gt(dev, txq); | |
f0796d5c | 1662 | } |
e6484930 TH |
1663 | |
1664 | dev->real_num_tx_queues = txq; | |
1665 | return 0; | |
f0796d5c JF |
1666 | } |
1667 | EXPORT_SYMBOL(netif_set_real_num_tx_queues); | |
56079431 | 1668 | |
62fe0b40 BH |
1669 | #ifdef CONFIG_RPS |
1670 | /** | |
1671 | * netif_set_real_num_rx_queues - set actual number of RX queues used | |
1672 | * @dev: Network device | |
1673 | * @rxq: Actual number of RX queues | |
1674 | * | |
1675 | * This must be called either with the rtnl_lock held or before | |
1676 | * registration of the net device. Returns 0 on success, or a | |
4e7f7951 BH |
1677 | * negative error code. If called before registration, it always |
1678 | * succeeds. | |
62fe0b40 BH |
1679 | */ |
1680 | int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) | |
1681 | { | |
1682 | int rc; | |
1683 | ||
bd25fa7b TH |
1684 | if (rxq < 1 || rxq > dev->num_rx_queues) |
1685 | return -EINVAL; | |
1686 | ||
62fe0b40 BH |
1687 | if (dev->reg_state == NETREG_REGISTERED) { |
1688 | ASSERT_RTNL(); | |
1689 | ||
62fe0b40 BH |
1690 | rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, |
1691 | rxq); | |
1692 | if (rc) | |
1693 | return rc; | |
62fe0b40 BH |
1694 | } |
1695 | ||
1696 | dev->real_num_rx_queues = rxq; | |
1697 | return 0; | |
1698 | } | |
1699 | EXPORT_SYMBOL(netif_set_real_num_rx_queues); | |
1700 | #endif | |
1701 | ||
def82a1d | 1702 | static inline void __netif_reschedule(struct Qdisc *q) |
56079431 | 1703 | { |
def82a1d JP |
1704 | struct softnet_data *sd; |
1705 | unsigned long flags; | |
56079431 | 1706 | |
def82a1d JP |
1707 | local_irq_save(flags); |
1708 | sd = &__get_cpu_var(softnet_data); | |
a9cbd588 CG |
1709 | q->next_sched = NULL; |
1710 | *sd->output_queue_tailp = q; | |
1711 | sd->output_queue_tailp = &q->next_sched; | |
def82a1d JP |
1712 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
1713 | local_irq_restore(flags); | |
1714 | } | |
1715 | ||
1716 | void __netif_schedule(struct Qdisc *q) | |
1717 | { | |
1718 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
1719 | __netif_reschedule(q); | |
56079431 DV |
1720 | } |
1721 | EXPORT_SYMBOL(__netif_schedule); | |
1722 | ||
bea3348e | 1723 | void dev_kfree_skb_irq(struct sk_buff *skb) |
56079431 | 1724 | { |
3578b0c8 | 1725 | if (atomic_dec_and_test(&skb->users)) { |
bea3348e SH |
1726 | struct softnet_data *sd; |
1727 | unsigned long flags; | |
56079431 | 1728 | |
bea3348e SH |
1729 | local_irq_save(flags); |
1730 | sd = &__get_cpu_var(softnet_data); | |
1731 | skb->next = sd->completion_queue; | |
1732 | sd->completion_queue = skb; | |
1733 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
1734 | local_irq_restore(flags); | |
1735 | } | |
56079431 | 1736 | } |
bea3348e | 1737 | EXPORT_SYMBOL(dev_kfree_skb_irq); |
56079431 DV |
1738 | |
1739 | void dev_kfree_skb_any(struct sk_buff *skb) | |
1740 | { | |
1741 | if (in_irq() || irqs_disabled()) | |
1742 | dev_kfree_skb_irq(skb); | |
1743 | else | |
1744 | dev_kfree_skb(skb); | |
1745 | } | |
1746 | EXPORT_SYMBOL(dev_kfree_skb_any); | |
1747 | ||
1748 | ||
bea3348e SH |
1749 | /** |
1750 | * netif_device_detach - mark device as removed | |
1751 | * @dev: network device | |
1752 | * | |
1753 | * Mark device as removed from system and therefore no longer available. | |
1754 | */ | |
56079431 DV |
1755 | void netif_device_detach(struct net_device *dev) |
1756 | { | |
1757 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
1758 | netif_running(dev)) { | |
d543103a | 1759 | netif_tx_stop_all_queues(dev); |
56079431 DV |
1760 | } |
1761 | } | |
1762 | EXPORT_SYMBOL(netif_device_detach); | |
1763 | ||
bea3348e SH |
1764 | /** |
1765 | * netif_device_attach - mark device as attached | |
1766 | * @dev: network device | |
1767 | * | |
1768 | * Mark device as attached from system and restart if needed. | |
1769 | */ | |
56079431 DV |
1770 | void netif_device_attach(struct net_device *dev) |
1771 | { | |
1772 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
1773 | netif_running(dev)) { | |
d543103a | 1774 | netif_tx_wake_all_queues(dev); |
4ec93edb | 1775 | __netdev_watchdog_up(dev); |
56079431 DV |
1776 | } |
1777 | } | |
1778 | EXPORT_SYMBOL(netif_device_attach); | |
1779 | ||
8a83a00b AB |
1780 | /** |
1781 | * skb_dev_set -- assign a new device to a buffer | |
1782 | * @skb: buffer for the new device | |
1783 | * @dev: network device | |
1784 | * | |
1785 | * If an skb is owned by a device already, we have to reset | |
1786 | * all data private to the namespace a device belongs to | |
1787 | * before assigning it a new device. | |
1788 | */ | |
1789 | #ifdef CONFIG_NET_NS | |
1790 | void skb_set_dev(struct sk_buff *skb, struct net_device *dev) | |
1791 | { | |
1792 | skb_dst_drop(skb); | |
1793 | if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) { | |
1794 | secpath_reset(skb); | |
1795 | nf_reset(skb); | |
1796 | skb_init_secmark(skb); | |
1797 | skb->mark = 0; | |
1798 | skb->priority = 0; | |
1799 | skb->nf_trace = 0; | |
1800 | skb->ipvs_property = 0; | |
1801 | #ifdef CONFIG_NET_SCHED | |
1802 | skb->tc_index = 0; | |
1803 | #endif | |
1804 | } | |
1805 | skb->dev = dev; | |
1806 | } | |
1807 | EXPORT_SYMBOL(skb_set_dev); | |
1808 | #endif /* CONFIG_NET_NS */ | |
1809 | ||
1da177e4 LT |
1810 | /* |
1811 | * Invalidate hardware checksum when packet is to be mangled, and | |
1812 | * complete checksum manually on outgoing path. | |
1813 | */ | |
84fa7933 | 1814 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 1815 | { |
d3bc23e7 | 1816 | __wsum csum; |
663ead3b | 1817 | int ret = 0, offset; |
1da177e4 | 1818 | |
84fa7933 | 1819 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
1820 | goto out_set_summed; |
1821 | ||
1822 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
a430a43d HX |
1823 | /* Let GSO fix up the checksum. */ |
1824 | goto out_set_summed; | |
1da177e4 LT |
1825 | } |
1826 | ||
55508d60 | 1827 | offset = skb_checksum_start_offset(skb); |
a030847e HX |
1828 | BUG_ON(offset >= skb_headlen(skb)); |
1829 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
1830 | ||
1831 | offset += skb->csum_offset; | |
1832 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
1833 | ||
1834 | if (skb_cloned(skb) && | |
1835 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1da177e4 LT |
1836 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
1837 | if (ret) | |
1838 | goto out; | |
1839 | } | |
1840 | ||
a030847e | 1841 | *(__sum16 *)(skb->data + offset) = csum_fold(csum); |
a430a43d | 1842 | out_set_summed: |
1da177e4 | 1843 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 1844 | out: |
1da177e4 LT |
1845 | return ret; |
1846 | } | |
d1b19dff | 1847 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 1848 | |
f6a78bfc HX |
1849 | /** |
1850 | * skb_gso_segment - Perform segmentation on skb. | |
1851 | * @skb: buffer to segment | |
576a30eb | 1852 | * @features: features for the output path (see dev->features) |
f6a78bfc HX |
1853 | * |
1854 | * This function segments the given skb and returns a list of segments. | |
576a30eb HX |
1855 | * |
1856 | * It may return NULL if the skb requires no segmentation. This is | |
1857 | * only possible when GSO is used for verifying header integrity. | |
f6a78bfc | 1858 | */ |
576a30eb | 1859 | struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features) |
f6a78bfc HX |
1860 | { |
1861 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
1862 | struct packet_type *ptype; | |
252e3346 | 1863 | __be16 type = skb->protocol; |
c8d5bcd1 | 1864 | int vlan_depth = ETH_HLEN; |
a430a43d | 1865 | int err; |
f6a78bfc | 1866 | |
c8d5bcd1 JG |
1867 | while (type == htons(ETH_P_8021Q)) { |
1868 | struct vlan_hdr *vh; | |
7b9c6090 | 1869 | |
c8d5bcd1 | 1870 | if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN))) |
7b9c6090 JG |
1871 | return ERR_PTR(-EINVAL); |
1872 | ||
c8d5bcd1 JG |
1873 | vh = (struct vlan_hdr *)(skb->data + vlan_depth); |
1874 | type = vh->h_vlan_encapsulated_proto; | |
1875 | vlan_depth += VLAN_HLEN; | |
7b9c6090 JG |
1876 | } |
1877 | ||
459a98ed | 1878 | skb_reset_mac_header(skb); |
b0e380b1 | 1879 | skb->mac_len = skb->network_header - skb->mac_header; |
f6a78bfc HX |
1880 | __skb_pull(skb, skb->mac_len); |
1881 | ||
67fd1a73 HX |
1882 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { |
1883 | struct net_device *dev = skb->dev; | |
1884 | struct ethtool_drvinfo info = {}; | |
1885 | ||
1886 | if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo) | |
1887 | dev->ethtool_ops->get_drvinfo(dev, &info); | |
1888 | ||
b194a367 | 1889 | WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n", |
67fd1a73 HX |
1890 | info.driver, dev ? dev->features : 0L, |
1891 | skb->sk ? skb->sk->sk_route_caps : 0L, | |
1892 | skb->len, skb->data_len, skb->ip_summed); | |
1893 | ||
a430a43d HX |
1894 | if (skb_header_cloned(skb) && |
1895 | (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))) | |
1896 | return ERR_PTR(err); | |
1897 | } | |
1898 | ||
f6a78bfc | 1899 | rcu_read_lock(); |
82d8a867 PE |
1900 | list_for_each_entry_rcu(ptype, |
1901 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
f6a78bfc | 1902 | if (ptype->type == type && !ptype->dev && ptype->gso_segment) { |
84fa7933 | 1903 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { |
a430a43d HX |
1904 | err = ptype->gso_send_check(skb); |
1905 | segs = ERR_PTR(err); | |
1906 | if (err || skb_gso_ok(skb, features)) | |
1907 | break; | |
d56f90a7 ACM |
1908 | __skb_push(skb, (skb->data - |
1909 | skb_network_header(skb))); | |
a430a43d | 1910 | } |
576a30eb | 1911 | segs = ptype->gso_segment(skb, features); |
f6a78bfc HX |
1912 | break; |
1913 | } | |
1914 | } | |
1915 | rcu_read_unlock(); | |
1916 | ||
98e399f8 | 1917 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 1918 | |
f6a78bfc HX |
1919 | return segs; |
1920 | } | |
f6a78bfc HX |
1921 | EXPORT_SYMBOL(skb_gso_segment); |
1922 | ||
fb286bb2 HX |
1923 | /* Take action when hardware reception checksum errors are detected. */ |
1924 | #ifdef CONFIG_BUG | |
1925 | void netdev_rx_csum_fault(struct net_device *dev) | |
1926 | { | |
1927 | if (net_ratelimit()) { | |
4ec93edb | 1928 | printk(KERN_ERR "%s: hw csum failure.\n", |
246a4212 | 1929 | dev ? dev->name : "<unknown>"); |
fb286bb2 HX |
1930 | dump_stack(); |
1931 | } | |
1932 | } | |
1933 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
1934 | #endif | |
1935 | ||
1da177e4 LT |
1936 | /* Actually, we should eliminate this check as soon as we know, that: |
1937 | * 1. IOMMU is present and allows to map all the memory. | |
1938 | * 2. No high memory really exists on this machine. | |
1939 | */ | |
1940 | ||
9092c658 | 1941 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
1da177e4 | 1942 | { |
3d3a8533 | 1943 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 1944 | int i; |
5acbbd42 FT |
1945 | if (!(dev->features & NETIF_F_HIGHDMA)) { |
1946 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) | |
1947 | if (PageHighMem(skb_shinfo(skb)->frags[i].page)) | |
1948 | return 1; | |
1949 | } | |
1da177e4 | 1950 | |
5acbbd42 FT |
1951 | if (PCI_DMA_BUS_IS_PHYS) { |
1952 | struct device *pdev = dev->dev.parent; | |
1da177e4 | 1953 | |
9092c658 ED |
1954 | if (!pdev) |
1955 | return 0; | |
5acbbd42 FT |
1956 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
1957 | dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page); | |
1958 | if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask) | |
1959 | return 1; | |
1960 | } | |
1961 | } | |
3d3a8533 | 1962 | #endif |
1da177e4 LT |
1963 | return 0; |
1964 | } | |
1da177e4 | 1965 | |
f6a78bfc HX |
1966 | struct dev_gso_cb { |
1967 | void (*destructor)(struct sk_buff *skb); | |
1968 | }; | |
1969 | ||
1970 | #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb) | |
1971 | ||
1972 | static void dev_gso_skb_destructor(struct sk_buff *skb) | |
1973 | { | |
1974 | struct dev_gso_cb *cb; | |
1975 | ||
1976 | do { | |
1977 | struct sk_buff *nskb = skb->next; | |
1978 | ||
1979 | skb->next = nskb->next; | |
1980 | nskb->next = NULL; | |
1981 | kfree_skb(nskb); | |
1982 | } while (skb->next); | |
1983 | ||
1984 | cb = DEV_GSO_CB(skb); | |
1985 | if (cb->destructor) | |
1986 | cb->destructor(skb); | |
1987 | } | |
1988 | ||
1989 | /** | |
1990 | * dev_gso_segment - Perform emulated hardware segmentation on skb. | |
1991 | * @skb: buffer to segment | |
91ecb63c | 1992 | * @features: device features as applicable to this skb |
f6a78bfc HX |
1993 | * |
1994 | * This function segments the given skb and stores the list of segments | |
1995 | * in skb->next. | |
1996 | */ | |
91ecb63c | 1997 | static int dev_gso_segment(struct sk_buff *skb, int features) |
f6a78bfc | 1998 | { |
f6a78bfc | 1999 | struct sk_buff *segs; |
576a30eb HX |
2000 | |
2001 | segs = skb_gso_segment(skb, features); | |
2002 | ||
2003 | /* Verifying header integrity only. */ | |
2004 | if (!segs) | |
2005 | return 0; | |
f6a78bfc | 2006 | |
801678c5 | 2007 | if (IS_ERR(segs)) |
f6a78bfc HX |
2008 | return PTR_ERR(segs); |
2009 | ||
2010 | skb->next = segs; | |
2011 | DEV_GSO_CB(skb)->destructor = skb->destructor; | |
2012 | skb->destructor = dev_gso_skb_destructor; | |
2013 | ||
2014 | return 0; | |
2015 | } | |
2016 | ||
fc6055a5 ED |
2017 | /* |
2018 | * Try to orphan skb early, right before transmission by the device. | |
2244d07b OH |
2019 | * We cannot orphan skb if tx timestamp is requested or the sk-reference |
2020 | * is needed on driver level for other reasons, e.g. see net/can/raw.c | |
fc6055a5 ED |
2021 | */ |
2022 | static inline void skb_orphan_try(struct sk_buff *skb) | |
2023 | { | |
87fd308c ED |
2024 | struct sock *sk = skb->sk; |
2025 | ||
2244d07b | 2026 | if (sk && !skb_shinfo(skb)->tx_flags) { |
87fd308c ED |
2027 | /* skb_tx_hash() wont be able to get sk. |
2028 | * We copy sk_hash into skb->rxhash | |
2029 | */ | |
2030 | if (!skb->rxhash) | |
2031 | skb->rxhash = sk->sk_hash; | |
fc6055a5 | 2032 | skb_orphan(skb); |
87fd308c | 2033 | } |
fc6055a5 ED |
2034 | } |
2035 | ||
03634668 JG |
2036 | static bool can_checksum_protocol(unsigned long features, __be16 protocol) |
2037 | { | |
2038 | return ((features & NETIF_F_GEN_CSUM) || | |
2039 | ((features & NETIF_F_V4_CSUM) && | |
2040 | protocol == htons(ETH_P_IP)) || | |
2041 | ((features & NETIF_F_V6_CSUM) && | |
2042 | protocol == htons(ETH_P_IPV6)) || | |
2043 | ((features & NETIF_F_FCOE_CRC) && | |
2044 | protocol == htons(ETH_P_FCOE))); | |
2045 | } | |
2046 | ||
f01a5236 JG |
2047 | static int harmonize_features(struct sk_buff *skb, __be16 protocol, int features) |
2048 | { | |
2049 | if (!can_checksum_protocol(protocol, features)) { | |
2050 | features &= ~NETIF_F_ALL_CSUM; | |
2051 | features &= ~NETIF_F_SG; | |
2052 | } else if (illegal_highdma(skb->dev, skb)) { | |
2053 | features &= ~NETIF_F_SG; | |
2054 | } | |
2055 | ||
2056 | return features; | |
2057 | } | |
2058 | ||
2059 | int netif_skb_features(struct sk_buff *skb) | |
58e998c6 JG |
2060 | { |
2061 | __be16 protocol = skb->protocol; | |
f01a5236 | 2062 | int features = skb->dev->features; |
58e998c6 JG |
2063 | |
2064 | if (protocol == htons(ETH_P_8021Q)) { | |
2065 | struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data; | |
2066 | protocol = veh->h_vlan_encapsulated_proto; | |
f01a5236 JG |
2067 | } else if (!vlan_tx_tag_present(skb)) { |
2068 | return harmonize_features(skb, protocol, features); | |
2069 | } | |
58e998c6 | 2070 | |
6ee400aa | 2071 | features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX); |
f01a5236 JG |
2072 | |
2073 | if (protocol != htons(ETH_P_8021Q)) { | |
2074 | return harmonize_features(skb, protocol, features); | |
2075 | } else { | |
2076 | features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | | |
6ee400aa | 2077 | NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX; |
f01a5236 JG |
2078 | return harmonize_features(skb, protocol, features); |
2079 | } | |
58e998c6 | 2080 | } |
f01a5236 | 2081 | EXPORT_SYMBOL(netif_skb_features); |
58e998c6 | 2082 | |
6afff0ca JF |
2083 | /* |
2084 | * Returns true if either: | |
2085 | * 1. skb has frag_list and the device doesn't support FRAGLIST, or | |
2086 | * 2. skb is fragmented and the device does not support SG, or if | |
2087 | * at least one of fragments is in highmem and device does not | |
2088 | * support DMA from it. | |
2089 | */ | |
2090 | static inline int skb_needs_linearize(struct sk_buff *skb, | |
02932ce9 | 2091 | int features) |
6afff0ca | 2092 | { |
02932ce9 JG |
2093 | return skb_is_nonlinear(skb) && |
2094 | ((skb_has_frag_list(skb) && | |
2095 | !(features & NETIF_F_FRAGLIST)) || | |
e1e78db6 | 2096 | (skb_shinfo(skb)->nr_frags && |
02932ce9 | 2097 | !(features & NETIF_F_SG))); |
6afff0ca JF |
2098 | } |
2099 | ||
fd2ea0a7 DM |
2100 | int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev, |
2101 | struct netdev_queue *txq) | |
f6a78bfc | 2102 | { |
00829823 | 2103 | const struct net_device_ops *ops = dev->netdev_ops; |
572a9d7b | 2104 | int rc = NETDEV_TX_OK; |
00829823 | 2105 | |
f6a78bfc | 2106 | if (likely(!skb->next)) { |
fc741216 JG |
2107 | int features; |
2108 | ||
93f154b5 ED |
2109 | /* |
2110 | * If device doesnt need skb->dst, release it right now while | |
2111 | * its hot in this cpu cache | |
2112 | */ | |
adf30907 ED |
2113 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) |
2114 | skb_dst_drop(skb); | |
2115 | ||
15c2d75f ED |
2116 | if (!list_empty(&ptype_all)) |
2117 | dev_queue_xmit_nit(skb, dev); | |
2118 | ||
fc6055a5 | 2119 | skb_orphan_try(skb); |
9ccb8975 | 2120 | |
fc741216 JG |
2121 | features = netif_skb_features(skb); |
2122 | ||
7b9c6090 | 2123 | if (vlan_tx_tag_present(skb) && |
fc741216 | 2124 | !(features & NETIF_F_HW_VLAN_TX)) { |
7b9c6090 JG |
2125 | skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb)); |
2126 | if (unlikely(!skb)) | |
2127 | goto out; | |
2128 | ||
2129 | skb->vlan_tci = 0; | |
2130 | } | |
2131 | ||
fc741216 | 2132 | if (netif_needs_gso(skb, features)) { |
91ecb63c | 2133 | if (unlikely(dev_gso_segment(skb, features))) |
9ccb8975 DM |
2134 | goto out_kfree_skb; |
2135 | if (skb->next) | |
2136 | goto gso; | |
6afff0ca | 2137 | } else { |
02932ce9 | 2138 | if (skb_needs_linearize(skb, features) && |
6afff0ca JF |
2139 | __skb_linearize(skb)) |
2140 | goto out_kfree_skb; | |
2141 | ||
2142 | /* If packet is not checksummed and device does not | |
2143 | * support checksumming for this protocol, complete | |
2144 | * checksumming here. | |
2145 | */ | |
2146 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
55508d60 MM |
2147 | skb_set_transport_header(skb, |
2148 | skb_checksum_start_offset(skb)); | |
03634668 | 2149 | if (!(features & NETIF_F_ALL_CSUM) && |
6afff0ca JF |
2150 | skb_checksum_help(skb)) |
2151 | goto out_kfree_skb; | |
2152 | } | |
9ccb8975 DM |
2153 | } |
2154 | ||
ac45f602 | 2155 | rc = ops->ndo_start_xmit(skb, dev); |
cf66ba58 | 2156 | trace_net_dev_xmit(skb, rc); |
ec634fe3 | 2157 | if (rc == NETDEV_TX_OK) |
08baf561 | 2158 | txq_trans_update(txq); |
ac45f602 | 2159 | return rc; |
f6a78bfc HX |
2160 | } |
2161 | ||
576a30eb | 2162 | gso: |
f6a78bfc HX |
2163 | do { |
2164 | struct sk_buff *nskb = skb->next; | |
f6a78bfc HX |
2165 | |
2166 | skb->next = nskb->next; | |
2167 | nskb->next = NULL; | |
068a2de5 KK |
2168 | |
2169 | /* | |
2170 | * If device doesnt need nskb->dst, release it right now while | |
2171 | * its hot in this cpu cache | |
2172 | */ | |
2173 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
2174 | skb_dst_drop(nskb); | |
2175 | ||
00829823 | 2176 | rc = ops->ndo_start_xmit(nskb, dev); |
cf66ba58 | 2177 | trace_net_dev_xmit(nskb, rc); |
ec634fe3 | 2178 | if (unlikely(rc != NETDEV_TX_OK)) { |
572a9d7b PM |
2179 | if (rc & ~NETDEV_TX_MASK) |
2180 | goto out_kfree_gso_skb; | |
f54d9e8d | 2181 | nskb->next = skb->next; |
f6a78bfc HX |
2182 | skb->next = nskb; |
2183 | return rc; | |
2184 | } | |
08baf561 | 2185 | txq_trans_update(txq); |
fd2ea0a7 | 2186 | if (unlikely(netif_tx_queue_stopped(txq) && skb->next)) |
f54d9e8d | 2187 | return NETDEV_TX_BUSY; |
f6a78bfc | 2188 | } while (skb->next); |
4ec93edb | 2189 | |
572a9d7b PM |
2190 | out_kfree_gso_skb: |
2191 | if (likely(skb->next == NULL)) | |
2192 | skb->destructor = DEV_GSO_CB(skb)->destructor; | |
f6a78bfc HX |
2193 | out_kfree_skb: |
2194 | kfree_skb(skb); | |
7b9c6090 | 2195 | out: |
572a9d7b | 2196 | return rc; |
f6a78bfc HX |
2197 | } |
2198 | ||
0a9627f2 | 2199 | static u32 hashrnd __read_mostly; |
b6b2fed1 | 2200 | |
a3d22a68 VZ |
2201 | /* |
2202 | * Returns a Tx hash based on the given packet descriptor a Tx queues' number | |
2203 | * to be used as a distribution range. | |
2204 | */ | |
2205 | u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb, | |
2206 | unsigned int num_tx_queues) | |
8f0f2223 | 2207 | { |
7019298a | 2208 | u32 hash; |
4f57c087 JF |
2209 | u16 qoffset = 0; |
2210 | u16 qcount = num_tx_queues; | |
b6b2fed1 | 2211 | |
513de11b DM |
2212 | if (skb_rx_queue_recorded(skb)) { |
2213 | hash = skb_get_rx_queue(skb); | |
a3d22a68 VZ |
2214 | while (unlikely(hash >= num_tx_queues)) |
2215 | hash -= num_tx_queues; | |
513de11b DM |
2216 | return hash; |
2217 | } | |
ec581f6a | 2218 | |
4f57c087 JF |
2219 | if (dev->num_tc) { |
2220 | u8 tc = netdev_get_prio_tc_map(dev, skb->priority); | |
2221 | qoffset = dev->tc_to_txq[tc].offset; | |
2222 | qcount = dev->tc_to_txq[tc].count; | |
2223 | } | |
2224 | ||
ec581f6a | 2225 | if (skb->sk && skb->sk->sk_hash) |
7019298a | 2226 | hash = skb->sk->sk_hash; |
ec581f6a | 2227 | else |
87fd308c | 2228 | hash = (__force u16) skb->protocol ^ skb->rxhash; |
0a9627f2 | 2229 | hash = jhash_1word(hash, hashrnd); |
b6b2fed1 | 2230 | |
4f57c087 | 2231 | return (u16) (((u64) hash * qcount) >> 32) + qoffset; |
8f0f2223 | 2232 | } |
a3d22a68 | 2233 | EXPORT_SYMBOL(__skb_tx_hash); |
8f0f2223 | 2234 | |
ed04642f ED |
2235 | static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index) |
2236 | { | |
2237 | if (unlikely(queue_index >= dev->real_num_tx_queues)) { | |
2238 | if (net_ratelimit()) { | |
7a161ea9 ED |
2239 | pr_warning("%s selects TX queue %d, but " |
2240 | "real number of TX queues is %d\n", | |
2241 | dev->name, queue_index, dev->real_num_tx_queues); | |
ed04642f ED |
2242 | } |
2243 | return 0; | |
2244 | } | |
2245 | return queue_index; | |
2246 | } | |
2247 | ||
1d24eb48 TH |
2248 | static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb) |
2249 | { | |
bf264145 | 2250 | #ifdef CONFIG_XPS |
1d24eb48 TH |
2251 | struct xps_dev_maps *dev_maps; |
2252 | struct xps_map *map; | |
2253 | int queue_index = -1; | |
2254 | ||
2255 | rcu_read_lock(); | |
2256 | dev_maps = rcu_dereference(dev->xps_maps); | |
2257 | if (dev_maps) { | |
2258 | map = rcu_dereference( | |
2259 | dev_maps->cpu_map[raw_smp_processor_id()]); | |
2260 | if (map) { | |
2261 | if (map->len == 1) | |
2262 | queue_index = map->queues[0]; | |
2263 | else { | |
2264 | u32 hash; | |
2265 | if (skb->sk && skb->sk->sk_hash) | |
2266 | hash = skb->sk->sk_hash; | |
2267 | else | |
2268 | hash = (__force u16) skb->protocol ^ | |
2269 | skb->rxhash; | |
2270 | hash = jhash_1word(hash, hashrnd); | |
2271 | queue_index = map->queues[ | |
2272 | ((u64)hash * map->len) >> 32]; | |
2273 | } | |
2274 | if (unlikely(queue_index >= dev->real_num_tx_queues)) | |
2275 | queue_index = -1; | |
2276 | } | |
2277 | } | |
2278 | rcu_read_unlock(); | |
2279 | ||
2280 | return queue_index; | |
2281 | #else | |
2282 | return -1; | |
2283 | #endif | |
2284 | } | |
2285 | ||
e8a0464c DM |
2286 | static struct netdev_queue *dev_pick_tx(struct net_device *dev, |
2287 | struct sk_buff *skb) | |
2288 | { | |
b0f77d0e | 2289 | int queue_index; |
deabc772 | 2290 | const struct net_device_ops *ops = dev->netdev_ops; |
a4ee3ce3 | 2291 | |
3853b584 TH |
2292 | if (dev->real_num_tx_queues == 1) |
2293 | queue_index = 0; | |
2294 | else if (ops->ndo_select_queue) { | |
deabc772 HS |
2295 | queue_index = ops->ndo_select_queue(dev, skb); |
2296 | queue_index = dev_cap_txqueue(dev, queue_index); | |
2297 | } else { | |
2298 | struct sock *sk = skb->sk; | |
2299 | queue_index = sk_tx_queue_get(sk); | |
a4ee3ce3 | 2300 | |
3853b584 TH |
2301 | if (queue_index < 0 || skb->ooo_okay || |
2302 | queue_index >= dev->real_num_tx_queues) { | |
2303 | int old_index = queue_index; | |
fd2ea0a7 | 2304 | |
1d24eb48 TH |
2305 | queue_index = get_xps_queue(dev, skb); |
2306 | if (queue_index < 0) | |
2307 | queue_index = skb_tx_hash(dev, skb); | |
3853b584 TH |
2308 | |
2309 | if (queue_index != old_index && sk) { | |
2310 | struct dst_entry *dst = | |
2311 | rcu_dereference_check(sk->sk_dst_cache, 1); | |
8728c544 ED |
2312 | |
2313 | if (dst && skb_dst(skb) == dst) | |
2314 | sk_tx_queue_set(sk, queue_index); | |
2315 | } | |
a4ee3ce3 KK |
2316 | } |
2317 | } | |
eae792b7 | 2318 | |
fd2ea0a7 DM |
2319 | skb_set_queue_mapping(skb, queue_index); |
2320 | return netdev_get_tx_queue(dev, queue_index); | |
e8a0464c DM |
2321 | } |
2322 | ||
bbd8a0d3 KK |
2323 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
2324 | struct net_device *dev, | |
2325 | struct netdev_queue *txq) | |
2326 | { | |
2327 | spinlock_t *root_lock = qdisc_lock(q); | |
79640a4c | 2328 | bool contended = qdisc_is_running(q); |
bbd8a0d3 KK |
2329 | int rc; |
2330 | ||
79640a4c ED |
2331 | /* |
2332 | * Heuristic to force contended enqueues to serialize on a | |
2333 | * separate lock before trying to get qdisc main lock. | |
2334 | * This permits __QDISC_STATE_RUNNING owner to get the lock more often | |
2335 | * and dequeue packets faster. | |
2336 | */ | |
2337 | if (unlikely(contended)) | |
2338 | spin_lock(&q->busylock); | |
2339 | ||
bbd8a0d3 KK |
2340 | spin_lock(root_lock); |
2341 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
2342 | kfree_skb(skb); | |
2343 | rc = NET_XMIT_DROP; | |
2344 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
bc135b23 | 2345 | qdisc_run_begin(q)) { |
bbd8a0d3 KK |
2346 | /* |
2347 | * This is a work-conserving queue; there are no old skbs | |
2348 | * waiting to be sent out; and the qdisc is not running - | |
2349 | * xmit the skb directly. | |
2350 | */ | |
7fee226a ED |
2351 | if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE)) |
2352 | skb_dst_force(skb); | |
bfe0d029 ED |
2353 | |
2354 | qdisc_skb_cb(skb)->pkt_len = skb->len; | |
2355 | qdisc_bstats_update(q, skb); | |
2356 | ||
79640a4c ED |
2357 | if (sch_direct_xmit(skb, q, dev, txq, root_lock)) { |
2358 | if (unlikely(contended)) { | |
2359 | spin_unlock(&q->busylock); | |
2360 | contended = false; | |
2361 | } | |
bbd8a0d3 | 2362 | __qdisc_run(q); |
79640a4c | 2363 | } else |
bc135b23 | 2364 | qdisc_run_end(q); |
bbd8a0d3 KK |
2365 | |
2366 | rc = NET_XMIT_SUCCESS; | |
2367 | } else { | |
7fee226a | 2368 | skb_dst_force(skb); |
bbd8a0d3 | 2369 | rc = qdisc_enqueue_root(skb, q); |
79640a4c ED |
2370 | if (qdisc_run_begin(q)) { |
2371 | if (unlikely(contended)) { | |
2372 | spin_unlock(&q->busylock); | |
2373 | contended = false; | |
2374 | } | |
2375 | __qdisc_run(q); | |
2376 | } | |
bbd8a0d3 KK |
2377 | } |
2378 | spin_unlock(root_lock); | |
79640a4c ED |
2379 | if (unlikely(contended)) |
2380 | spin_unlock(&q->busylock); | |
bbd8a0d3 KK |
2381 | return rc; |
2382 | } | |
2383 | ||
745e20f1 | 2384 | static DEFINE_PER_CPU(int, xmit_recursion); |
11a766ce | 2385 | #define RECURSION_LIMIT 10 |
745e20f1 | 2386 | |
d29f749e DJ |
2387 | /** |
2388 | * dev_queue_xmit - transmit a buffer | |
2389 | * @skb: buffer to transmit | |
2390 | * | |
2391 | * Queue a buffer for transmission to a network device. The caller must | |
2392 | * have set the device and priority and built the buffer before calling | |
2393 | * this function. The function can be called from an interrupt. | |
2394 | * | |
2395 | * A negative errno code is returned on a failure. A success does not | |
2396 | * guarantee the frame will be transmitted as it may be dropped due | |
2397 | * to congestion or traffic shaping. | |
2398 | * | |
2399 | * ----------------------------------------------------------------------------------- | |
2400 | * I notice this method can also return errors from the queue disciplines, | |
2401 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
2402 | * be positive. | |
2403 | * | |
2404 | * Regardless of the return value, the skb is consumed, so it is currently | |
2405 | * difficult to retry a send to this method. (You can bump the ref count | |
2406 | * before sending to hold a reference for retry if you are careful.) | |
2407 | * | |
2408 | * When calling this method, interrupts MUST be enabled. This is because | |
2409 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
2410 | * --BLG | |
2411 | */ | |
1da177e4 LT |
2412 | int dev_queue_xmit(struct sk_buff *skb) |
2413 | { | |
2414 | struct net_device *dev = skb->dev; | |
dc2b4847 | 2415 | struct netdev_queue *txq; |
1da177e4 LT |
2416 | struct Qdisc *q; |
2417 | int rc = -ENOMEM; | |
2418 | ||
4ec93edb YH |
2419 | /* Disable soft irqs for various locks below. Also |
2420 | * stops preemption for RCU. | |
1da177e4 | 2421 | */ |
4ec93edb | 2422 | rcu_read_lock_bh(); |
1da177e4 | 2423 | |
eae792b7 | 2424 | txq = dev_pick_tx(dev, skb); |
a898def2 | 2425 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 2426 | |
1da177e4 | 2427 | #ifdef CONFIG_NET_CLS_ACT |
d1b19dff | 2428 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS); |
1da177e4 | 2429 | #endif |
cf66ba58 | 2430 | trace_net_dev_queue(skb); |
1da177e4 | 2431 | if (q->enqueue) { |
bbd8a0d3 | 2432 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 2433 | goto out; |
1da177e4 LT |
2434 | } |
2435 | ||
2436 | /* The device has no queue. Common case for software devices: | |
2437 | loopback, all the sorts of tunnels... | |
2438 | ||
932ff279 HX |
2439 | Really, it is unlikely that netif_tx_lock protection is necessary |
2440 | here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
1da177e4 LT |
2441 | counters.) |
2442 | However, it is possible, that they rely on protection | |
2443 | made by us here. | |
2444 | ||
2445 | Check this and shot the lock. It is not prone from deadlocks. | |
2446 | Either shot noqueue qdisc, it is even simpler 8) | |
2447 | */ | |
2448 | if (dev->flags & IFF_UP) { | |
2449 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
2450 | ||
c773e847 | 2451 | if (txq->xmit_lock_owner != cpu) { |
1da177e4 | 2452 | |
745e20f1 ED |
2453 | if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT) |
2454 | goto recursion_alert; | |
2455 | ||
c773e847 | 2456 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 2457 | |
fd2ea0a7 | 2458 | if (!netif_tx_queue_stopped(txq)) { |
745e20f1 | 2459 | __this_cpu_inc(xmit_recursion); |
572a9d7b | 2460 | rc = dev_hard_start_xmit(skb, dev, txq); |
745e20f1 | 2461 | __this_cpu_dec(xmit_recursion); |
572a9d7b | 2462 | if (dev_xmit_complete(rc)) { |
c773e847 | 2463 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
2464 | goto out; |
2465 | } | |
2466 | } | |
c773e847 | 2467 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
2468 | if (net_ratelimit()) |
2469 | printk(KERN_CRIT "Virtual device %s asks to " | |
2470 | "queue packet!\n", dev->name); | |
2471 | } else { | |
2472 | /* Recursion is detected! It is possible, | |
745e20f1 ED |
2473 | * unfortunately |
2474 | */ | |
2475 | recursion_alert: | |
1da177e4 LT |
2476 | if (net_ratelimit()) |
2477 | printk(KERN_CRIT "Dead loop on virtual device " | |
2478 | "%s, fix it urgently!\n", dev->name); | |
2479 | } | |
2480 | } | |
2481 | ||
2482 | rc = -ENETDOWN; | |
d4828d85 | 2483 | rcu_read_unlock_bh(); |
1da177e4 | 2484 | |
1da177e4 LT |
2485 | kfree_skb(skb); |
2486 | return rc; | |
2487 | out: | |
d4828d85 | 2488 | rcu_read_unlock_bh(); |
1da177e4 LT |
2489 | return rc; |
2490 | } | |
d1b19dff | 2491 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 LT |
2492 | |
2493 | ||
2494 | /*======================================================================= | |
2495 | Receiver routines | |
2496 | =======================================================================*/ | |
2497 | ||
6b2bedc3 | 2498 | int netdev_max_backlog __read_mostly = 1000; |
3b098e2d | 2499 | int netdev_tstamp_prequeue __read_mostly = 1; |
6b2bedc3 SH |
2500 | int netdev_budget __read_mostly = 300; |
2501 | int weight_p __read_mostly = 64; /* old backlog weight */ | |
1da177e4 | 2502 | |
eecfd7c4 ED |
2503 | /* Called with irq disabled */ |
2504 | static inline void ____napi_schedule(struct softnet_data *sd, | |
2505 | struct napi_struct *napi) | |
2506 | { | |
2507 | list_add_tail(&napi->poll_list, &sd->poll_list); | |
2508 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2509 | } | |
2510 | ||
0a9627f2 | 2511 | /* |
bfb564e7 KK |
2512 | * __skb_get_rxhash: calculate a flow hash based on src/dst addresses |
2513 | * and src/dst port numbers. Returns a non-zero hash number on success | |
2514 | * and 0 on failure. | |
0a9627f2 | 2515 | */ |
bfb564e7 | 2516 | __u32 __skb_get_rxhash(struct sk_buff *skb) |
0a9627f2 | 2517 | { |
12fcdefb | 2518 | int nhoff, hash = 0, poff; |
0a9627f2 TH |
2519 | struct ipv6hdr *ip6; |
2520 | struct iphdr *ip; | |
0a9627f2 | 2521 | u8 ip_proto; |
8c52d509 CG |
2522 | u32 addr1, addr2, ihl; |
2523 | union { | |
2524 | u32 v32; | |
2525 | u16 v16[2]; | |
2526 | } ports; | |
0a9627f2 | 2527 | |
bfb564e7 | 2528 | nhoff = skb_network_offset(skb); |
0a9627f2 TH |
2529 | |
2530 | switch (skb->protocol) { | |
2531 | case __constant_htons(ETH_P_IP): | |
bfb564e7 | 2532 | if (!pskb_may_pull(skb, sizeof(*ip) + nhoff)) |
0a9627f2 TH |
2533 | goto done; |
2534 | ||
1003489e | 2535 | ip = (struct iphdr *) (skb->data + nhoff); |
dbe5775b CG |
2536 | if (ip->frag_off & htons(IP_MF | IP_OFFSET)) |
2537 | ip_proto = 0; | |
2538 | else | |
2539 | ip_proto = ip->protocol; | |
b249dcb8 ED |
2540 | addr1 = (__force u32) ip->saddr; |
2541 | addr2 = (__force u32) ip->daddr; | |
0a9627f2 TH |
2542 | ihl = ip->ihl; |
2543 | break; | |
2544 | case __constant_htons(ETH_P_IPV6): | |
bfb564e7 | 2545 | if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff)) |
0a9627f2 TH |
2546 | goto done; |
2547 | ||
1003489e | 2548 | ip6 = (struct ipv6hdr *) (skb->data + nhoff); |
0a9627f2 | 2549 | ip_proto = ip6->nexthdr; |
b249dcb8 ED |
2550 | addr1 = (__force u32) ip6->saddr.s6_addr32[3]; |
2551 | addr2 = (__force u32) ip6->daddr.s6_addr32[3]; | |
0a9627f2 TH |
2552 | ihl = (40 >> 2); |
2553 | break; | |
2554 | default: | |
2555 | goto done; | |
2556 | } | |
bfb564e7 | 2557 | |
12fcdefb CG |
2558 | ports.v32 = 0; |
2559 | poff = proto_ports_offset(ip_proto); | |
2560 | if (poff >= 0) { | |
2561 | nhoff += ihl * 4 + poff; | |
2562 | if (pskb_may_pull(skb, nhoff + 4)) { | |
2563 | ports.v32 = * (__force u32 *) (skb->data + nhoff); | |
8c52d509 CG |
2564 | if (ports.v16[1] < ports.v16[0]) |
2565 | swap(ports.v16[0], ports.v16[1]); | |
b249dcb8 | 2566 | } |
0a9627f2 TH |
2567 | } |
2568 | ||
b249dcb8 ED |
2569 | /* get a consistent hash (same value on both flow directions) */ |
2570 | if (addr2 < addr1) | |
2571 | swap(addr1, addr2); | |
0a9627f2 | 2572 | |
bfb564e7 KK |
2573 | hash = jhash_3words(addr1, addr2, ports.v32, hashrnd); |
2574 | if (!hash) | |
2575 | hash = 1; | |
2576 | ||
2577 | done: | |
2578 | return hash; | |
2579 | } | |
2580 | EXPORT_SYMBOL(__skb_get_rxhash); | |
2581 | ||
2582 | #ifdef CONFIG_RPS | |
2583 | ||
2584 | /* One global table that all flow-based protocols share. */ | |
6e3f7faf | 2585 | struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; |
bfb564e7 KK |
2586 | EXPORT_SYMBOL(rps_sock_flow_table); |
2587 | ||
2588 | /* | |
2589 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
2590 | * CPU from the RPS map of the receiving queue for a given skb. | |
2591 | * rcu_read_lock must be held on entry. | |
2592 | */ | |
2593 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
2594 | struct rps_dev_flow **rflowp) | |
2595 | { | |
2596 | struct netdev_rx_queue *rxqueue; | |
6e3f7faf | 2597 | struct rps_map *map; |
bfb564e7 KK |
2598 | struct rps_dev_flow_table *flow_table; |
2599 | struct rps_sock_flow_table *sock_flow_table; | |
2600 | int cpu = -1; | |
2601 | u16 tcpu; | |
2602 | ||
2603 | if (skb_rx_queue_recorded(skb)) { | |
2604 | u16 index = skb_get_rx_queue(skb); | |
62fe0b40 BH |
2605 | if (unlikely(index >= dev->real_num_rx_queues)) { |
2606 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
2607 | "%s received packet on queue %u, but number " | |
2608 | "of RX queues is %u\n", | |
2609 | dev->name, index, dev->real_num_rx_queues); | |
bfb564e7 KK |
2610 | goto done; |
2611 | } | |
2612 | rxqueue = dev->_rx + index; | |
2613 | } else | |
2614 | rxqueue = dev->_rx; | |
2615 | ||
6e3f7faf ED |
2616 | map = rcu_dereference(rxqueue->rps_map); |
2617 | if (map) { | |
2618 | if (map->len == 1) { | |
6febfca9 CG |
2619 | tcpu = map->cpus[0]; |
2620 | if (cpu_online(tcpu)) | |
2621 | cpu = tcpu; | |
2622 | goto done; | |
2623 | } | |
6e3f7faf | 2624 | } else if (!rcu_dereference_raw(rxqueue->rps_flow_table)) { |
bfb564e7 | 2625 | goto done; |
6febfca9 | 2626 | } |
bfb564e7 | 2627 | |
2d47b459 | 2628 | skb_reset_network_header(skb); |
bfb564e7 KK |
2629 | if (!skb_get_rxhash(skb)) |
2630 | goto done; | |
2631 | ||
fec5e652 TH |
2632 | flow_table = rcu_dereference(rxqueue->rps_flow_table); |
2633 | sock_flow_table = rcu_dereference(rps_sock_flow_table); | |
2634 | if (flow_table && sock_flow_table) { | |
2635 | u16 next_cpu; | |
2636 | struct rps_dev_flow *rflow; | |
2637 | ||
2638 | rflow = &flow_table->flows[skb->rxhash & flow_table->mask]; | |
2639 | tcpu = rflow->cpu; | |
2640 | ||
2641 | next_cpu = sock_flow_table->ents[skb->rxhash & | |
2642 | sock_flow_table->mask]; | |
2643 | ||
2644 | /* | |
2645 | * If the desired CPU (where last recvmsg was done) is | |
2646 | * different from current CPU (one in the rx-queue flow | |
2647 | * table entry), switch if one of the following holds: | |
2648 | * - Current CPU is unset (equal to RPS_NO_CPU). | |
2649 | * - Current CPU is offline. | |
2650 | * - The current CPU's queue tail has advanced beyond the | |
2651 | * last packet that was enqueued using this table entry. | |
2652 | * This guarantees that all previous packets for the flow | |
2653 | * have been dequeued, thus preserving in order delivery. | |
2654 | */ | |
2655 | if (unlikely(tcpu != next_cpu) && | |
2656 | (tcpu == RPS_NO_CPU || !cpu_online(tcpu) || | |
2657 | ((int)(per_cpu(softnet_data, tcpu).input_queue_head - | |
2658 | rflow->last_qtail)) >= 0)) { | |
2659 | tcpu = rflow->cpu = next_cpu; | |
2660 | if (tcpu != RPS_NO_CPU) | |
2661 | rflow->last_qtail = per_cpu(softnet_data, | |
2662 | tcpu).input_queue_head; | |
2663 | } | |
2664 | if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) { | |
2665 | *rflowp = rflow; | |
2666 | cpu = tcpu; | |
2667 | goto done; | |
2668 | } | |
2669 | } | |
2670 | ||
0a9627f2 | 2671 | if (map) { |
fec5e652 | 2672 | tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32]; |
0a9627f2 TH |
2673 | |
2674 | if (cpu_online(tcpu)) { | |
2675 | cpu = tcpu; | |
2676 | goto done; | |
2677 | } | |
2678 | } | |
2679 | ||
2680 | done: | |
0a9627f2 TH |
2681 | return cpu; |
2682 | } | |
2683 | ||
0a9627f2 | 2684 | /* Called from hardirq (IPI) context */ |
e36fa2f7 | 2685 | static void rps_trigger_softirq(void *data) |
0a9627f2 | 2686 | { |
e36fa2f7 ED |
2687 | struct softnet_data *sd = data; |
2688 | ||
eecfd7c4 | 2689 | ____napi_schedule(sd, &sd->backlog); |
dee42870 | 2690 | sd->received_rps++; |
0a9627f2 | 2691 | } |
e36fa2f7 | 2692 | |
fec5e652 | 2693 | #endif /* CONFIG_RPS */ |
0a9627f2 | 2694 | |
e36fa2f7 ED |
2695 | /* |
2696 | * Check if this softnet_data structure is another cpu one | |
2697 | * If yes, queue it to our IPI list and return 1 | |
2698 | * If no, return 0 | |
2699 | */ | |
2700 | static int rps_ipi_queued(struct softnet_data *sd) | |
2701 | { | |
2702 | #ifdef CONFIG_RPS | |
2703 | struct softnet_data *mysd = &__get_cpu_var(softnet_data); | |
2704 | ||
2705 | if (sd != mysd) { | |
2706 | sd->rps_ipi_next = mysd->rps_ipi_list; | |
2707 | mysd->rps_ipi_list = sd; | |
2708 | ||
2709 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2710 | return 1; | |
2711 | } | |
2712 | #endif /* CONFIG_RPS */ | |
2713 | return 0; | |
2714 | } | |
2715 | ||
0a9627f2 TH |
2716 | /* |
2717 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
2718 | * queue (may be a remote CPU queue). | |
2719 | */ | |
fec5e652 TH |
2720 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, |
2721 | unsigned int *qtail) | |
0a9627f2 | 2722 | { |
e36fa2f7 | 2723 | struct softnet_data *sd; |
0a9627f2 TH |
2724 | unsigned long flags; |
2725 | ||
e36fa2f7 | 2726 | sd = &per_cpu(softnet_data, cpu); |
0a9627f2 TH |
2727 | |
2728 | local_irq_save(flags); | |
0a9627f2 | 2729 | |
e36fa2f7 | 2730 | rps_lock(sd); |
6e7676c1 CG |
2731 | if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) { |
2732 | if (skb_queue_len(&sd->input_pkt_queue)) { | |
0a9627f2 | 2733 | enqueue: |
e36fa2f7 | 2734 | __skb_queue_tail(&sd->input_pkt_queue, skb); |
76cc8b13 | 2735 | input_queue_tail_incr_save(sd, qtail); |
e36fa2f7 | 2736 | rps_unlock(sd); |
152102c7 | 2737 | local_irq_restore(flags); |
0a9627f2 TH |
2738 | return NET_RX_SUCCESS; |
2739 | } | |
2740 | ||
ebda37c2 ED |
2741 | /* Schedule NAPI for backlog device |
2742 | * We can use non atomic operation since we own the queue lock | |
2743 | */ | |
2744 | if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { | |
e36fa2f7 | 2745 | if (!rps_ipi_queued(sd)) |
eecfd7c4 | 2746 | ____napi_schedule(sd, &sd->backlog); |
0a9627f2 TH |
2747 | } |
2748 | goto enqueue; | |
2749 | } | |
2750 | ||
dee42870 | 2751 | sd->dropped++; |
e36fa2f7 | 2752 | rps_unlock(sd); |
0a9627f2 | 2753 | |
0a9627f2 TH |
2754 | local_irq_restore(flags); |
2755 | ||
caf586e5 | 2756 | atomic_long_inc(&skb->dev->rx_dropped); |
0a9627f2 TH |
2757 | kfree_skb(skb); |
2758 | return NET_RX_DROP; | |
2759 | } | |
1da177e4 | 2760 | |
1da177e4 LT |
2761 | /** |
2762 | * netif_rx - post buffer to the network code | |
2763 | * @skb: buffer to post | |
2764 | * | |
2765 | * This function receives a packet from a device driver and queues it for | |
2766 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
2767 | * may be dropped during processing for congestion control or by the | |
2768 | * protocol layers. | |
2769 | * | |
2770 | * return values: | |
2771 | * NET_RX_SUCCESS (no congestion) | |
1da177e4 LT |
2772 | * NET_RX_DROP (packet was dropped) |
2773 | * | |
2774 | */ | |
2775 | ||
2776 | int netif_rx(struct sk_buff *skb) | |
2777 | { | |
b0e28f1e | 2778 | int ret; |
1da177e4 LT |
2779 | |
2780 | /* if netpoll wants it, pretend we never saw it */ | |
2781 | if (netpoll_rx(skb)) | |
2782 | return NET_RX_DROP; | |
2783 | ||
3b098e2d ED |
2784 | if (netdev_tstamp_prequeue) |
2785 | net_timestamp_check(skb); | |
1da177e4 | 2786 | |
cf66ba58 | 2787 | trace_netif_rx(skb); |
df334545 | 2788 | #ifdef CONFIG_RPS |
b0e28f1e | 2789 | { |
fec5e652 | 2790 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
b0e28f1e ED |
2791 | int cpu; |
2792 | ||
cece1945 | 2793 | preempt_disable(); |
b0e28f1e | 2794 | rcu_read_lock(); |
fec5e652 TH |
2795 | |
2796 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
b0e28f1e ED |
2797 | if (cpu < 0) |
2798 | cpu = smp_processor_id(); | |
fec5e652 TH |
2799 | |
2800 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
2801 | ||
b0e28f1e | 2802 | rcu_read_unlock(); |
cece1945 | 2803 | preempt_enable(); |
b0e28f1e | 2804 | } |
1e94d72f | 2805 | #else |
fec5e652 TH |
2806 | { |
2807 | unsigned int qtail; | |
2808 | ret = enqueue_to_backlog(skb, get_cpu(), &qtail); | |
2809 | put_cpu(); | |
2810 | } | |
1e94d72f | 2811 | #endif |
b0e28f1e | 2812 | return ret; |
1da177e4 | 2813 | } |
d1b19dff | 2814 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
2815 | |
2816 | int netif_rx_ni(struct sk_buff *skb) | |
2817 | { | |
2818 | int err; | |
2819 | ||
2820 | preempt_disable(); | |
2821 | err = netif_rx(skb); | |
2822 | if (local_softirq_pending()) | |
2823 | do_softirq(); | |
2824 | preempt_enable(); | |
2825 | ||
2826 | return err; | |
2827 | } | |
1da177e4 LT |
2828 | EXPORT_SYMBOL(netif_rx_ni); |
2829 | ||
1da177e4 LT |
2830 | static void net_tx_action(struct softirq_action *h) |
2831 | { | |
2832 | struct softnet_data *sd = &__get_cpu_var(softnet_data); | |
2833 | ||
2834 | if (sd->completion_queue) { | |
2835 | struct sk_buff *clist; | |
2836 | ||
2837 | local_irq_disable(); | |
2838 | clist = sd->completion_queue; | |
2839 | sd->completion_queue = NULL; | |
2840 | local_irq_enable(); | |
2841 | ||
2842 | while (clist) { | |
2843 | struct sk_buff *skb = clist; | |
2844 | clist = clist->next; | |
2845 | ||
547b792c | 2846 | WARN_ON(atomic_read(&skb->users)); |
07dc22e7 | 2847 | trace_kfree_skb(skb, net_tx_action); |
1da177e4 LT |
2848 | __kfree_skb(skb); |
2849 | } | |
2850 | } | |
2851 | ||
2852 | if (sd->output_queue) { | |
37437bb2 | 2853 | struct Qdisc *head; |
1da177e4 LT |
2854 | |
2855 | local_irq_disable(); | |
2856 | head = sd->output_queue; | |
2857 | sd->output_queue = NULL; | |
a9cbd588 | 2858 | sd->output_queue_tailp = &sd->output_queue; |
1da177e4 LT |
2859 | local_irq_enable(); |
2860 | ||
2861 | while (head) { | |
37437bb2 DM |
2862 | struct Qdisc *q = head; |
2863 | spinlock_t *root_lock; | |
2864 | ||
1da177e4 LT |
2865 | head = head->next_sched; |
2866 | ||
5fb66229 | 2867 | root_lock = qdisc_lock(q); |
37437bb2 | 2868 | if (spin_trylock(root_lock)) { |
def82a1d JP |
2869 | smp_mb__before_clear_bit(); |
2870 | clear_bit(__QDISC_STATE_SCHED, | |
2871 | &q->state); | |
37437bb2 DM |
2872 | qdisc_run(q); |
2873 | spin_unlock(root_lock); | |
1da177e4 | 2874 | } else { |
195648bb | 2875 | if (!test_bit(__QDISC_STATE_DEACTIVATED, |
e8a83e10 | 2876 | &q->state)) { |
195648bb | 2877 | __netif_reschedule(q); |
e8a83e10 JP |
2878 | } else { |
2879 | smp_mb__before_clear_bit(); | |
2880 | clear_bit(__QDISC_STATE_SCHED, | |
2881 | &q->state); | |
2882 | } | |
1da177e4 LT |
2883 | } |
2884 | } | |
2885 | } | |
2886 | } | |
2887 | ||
ab95bfe0 JP |
2888 | #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \ |
2889 | (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)) | |
da678292 MM |
2890 | /* This hook is defined here for ATM LANE */ |
2891 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
2892 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 2893 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 2894 | #endif |
1da177e4 | 2895 | |
1da177e4 LT |
2896 | #ifdef CONFIG_NET_CLS_ACT |
2897 | /* TODO: Maybe we should just force sch_ingress to be compiled in | |
2898 | * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions | |
2899 | * a compare and 2 stores extra right now if we dont have it on | |
2900 | * but have CONFIG_NET_CLS_ACT | |
4ec93edb | 2901 | * NOTE: This doesnt stop any functionality; if you dont have |
1da177e4 LT |
2902 | * the ingress scheduler, you just cant add policies on ingress. |
2903 | * | |
2904 | */ | |
24824a09 | 2905 | static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq) |
1da177e4 | 2906 | { |
1da177e4 | 2907 | struct net_device *dev = skb->dev; |
f697c3e8 | 2908 | u32 ttl = G_TC_RTTL(skb->tc_verd); |
555353cf DM |
2909 | int result = TC_ACT_OK; |
2910 | struct Qdisc *q; | |
4ec93edb | 2911 | |
de384830 SH |
2912 | if (unlikely(MAX_RED_LOOP < ttl++)) { |
2913 | if (net_ratelimit()) | |
2914 | pr_warning( "Redir loop detected Dropping packet (%d->%d)\n", | |
2915 | skb->skb_iif, dev->ifindex); | |
f697c3e8 HX |
2916 | return TC_ACT_SHOT; |
2917 | } | |
1da177e4 | 2918 | |
f697c3e8 HX |
2919 | skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl); |
2920 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); | |
1da177e4 | 2921 | |
83874000 | 2922 | q = rxq->qdisc; |
8d50b53d | 2923 | if (q != &noop_qdisc) { |
83874000 | 2924 | spin_lock(qdisc_lock(q)); |
a9312ae8 DM |
2925 | if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) |
2926 | result = qdisc_enqueue_root(skb, q); | |
83874000 DM |
2927 | spin_unlock(qdisc_lock(q)); |
2928 | } | |
f697c3e8 HX |
2929 | |
2930 | return result; | |
2931 | } | |
86e65da9 | 2932 | |
f697c3e8 HX |
2933 | static inline struct sk_buff *handle_ing(struct sk_buff *skb, |
2934 | struct packet_type **pt_prev, | |
2935 | int *ret, struct net_device *orig_dev) | |
2936 | { | |
24824a09 ED |
2937 | struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue); |
2938 | ||
2939 | if (!rxq || rxq->qdisc == &noop_qdisc) | |
f697c3e8 | 2940 | goto out; |
1da177e4 | 2941 | |
f697c3e8 HX |
2942 | if (*pt_prev) { |
2943 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
2944 | *pt_prev = NULL; | |
1da177e4 LT |
2945 | } |
2946 | ||
24824a09 | 2947 | switch (ing_filter(skb, rxq)) { |
f697c3e8 HX |
2948 | case TC_ACT_SHOT: |
2949 | case TC_ACT_STOLEN: | |
2950 | kfree_skb(skb); | |
2951 | return NULL; | |
2952 | } | |
2953 | ||
2954 | out: | |
2955 | skb->tc_verd = 0; | |
2956 | return skb; | |
1da177e4 LT |
2957 | } |
2958 | #endif | |
2959 | ||
ab95bfe0 JP |
2960 | /** |
2961 | * netdev_rx_handler_register - register receive handler | |
2962 | * @dev: device to register a handler for | |
2963 | * @rx_handler: receive handler to register | |
93e2c32b | 2964 | * @rx_handler_data: data pointer that is used by rx handler |
ab95bfe0 JP |
2965 | * |
2966 | * Register a receive hander for a device. This handler will then be | |
2967 | * called from __netif_receive_skb. A negative errno code is returned | |
2968 | * on a failure. | |
2969 | * | |
2970 | * The caller must hold the rtnl_mutex. | |
2971 | */ | |
2972 | int netdev_rx_handler_register(struct net_device *dev, | |
93e2c32b JP |
2973 | rx_handler_func_t *rx_handler, |
2974 | void *rx_handler_data) | |
ab95bfe0 JP |
2975 | { |
2976 | ASSERT_RTNL(); | |
2977 | ||
2978 | if (dev->rx_handler) | |
2979 | return -EBUSY; | |
2980 | ||
93e2c32b | 2981 | rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); |
ab95bfe0 JP |
2982 | rcu_assign_pointer(dev->rx_handler, rx_handler); |
2983 | ||
2984 | return 0; | |
2985 | } | |
2986 | EXPORT_SYMBOL_GPL(netdev_rx_handler_register); | |
2987 | ||
2988 | /** | |
2989 | * netdev_rx_handler_unregister - unregister receive handler | |
2990 | * @dev: device to unregister a handler from | |
2991 | * | |
2992 | * Unregister a receive hander from a device. | |
2993 | * | |
2994 | * The caller must hold the rtnl_mutex. | |
2995 | */ | |
2996 | void netdev_rx_handler_unregister(struct net_device *dev) | |
2997 | { | |
2998 | ||
2999 | ASSERT_RTNL(); | |
3000 | rcu_assign_pointer(dev->rx_handler, NULL); | |
93e2c32b | 3001 | rcu_assign_pointer(dev->rx_handler_data, NULL); |
ab95bfe0 JP |
3002 | } |
3003 | EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); | |
3004 | ||
acbbc071 ED |
3005 | static inline void skb_bond_set_mac_by_master(struct sk_buff *skb, |
3006 | struct net_device *master) | |
3007 | { | |
3008 | if (skb->pkt_type == PACKET_HOST) { | |
3009 | u16 *dest = (u16 *) eth_hdr(skb)->h_dest; | |
3010 | ||
3011 | memcpy(dest, master->dev_addr, ETH_ALEN); | |
3012 | } | |
3013 | } | |
3014 | ||
3015 | /* On bonding slaves other than the currently active slave, suppress | |
3016 | * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and | |
3017 | * ARP on active-backup slaves with arp_validate enabled. | |
3018 | */ | |
3019 | int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master) | |
3020 | { | |
3021 | struct net_device *dev = skb->dev; | |
3022 | ||
3023 | if (master->priv_flags & IFF_MASTER_ARPMON) | |
3024 | dev->last_rx = jiffies; | |
3025 | ||
f350a0a8 JP |
3026 | if ((master->priv_flags & IFF_MASTER_ALB) && |
3027 | (master->priv_flags & IFF_BRIDGE_PORT)) { | |
acbbc071 ED |
3028 | /* Do address unmangle. The local destination address |
3029 | * will be always the one master has. Provides the right | |
3030 | * functionality in a bridge. | |
3031 | */ | |
3032 | skb_bond_set_mac_by_master(skb, master); | |
3033 | } | |
3034 | ||
3035 | if (dev->priv_flags & IFF_SLAVE_INACTIVE) { | |
3036 | if ((dev->priv_flags & IFF_SLAVE_NEEDARP) && | |
3037 | skb->protocol == __cpu_to_be16(ETH_P_ARP)) | |
3038 | return 0; | |
3039 | ||
3040 | if (master->priv_flags & IFF_MASTER_ALB) { | |
3041 | if (skb->pkt_type != PACKET_BROADCAST && | |
3042 | skb->pkt_type != PACKET_MULTICAST) | |
3043 | return 0; | |
3044 | } | |
3045 | if (master->priv_flags & IFF_MASTER_8023AD && | |
3046 | skb->protocol == __cpu_to_be16(ETH_P_SLOW)) | |
3047 | return 0; | |
3048 | ||
3049 | return 1; | |
3050 | } | |
3051 | return 0; | |
3052 | } | |
3053 | EXPORT_SYMBOL(__skb_bond_should_drop); | |
3054 | ||
10f744d2 | 3055 | static int __netif_receive_skb(struct sk_buff *skb) |
1da177e4 LT |
3056 | { |
3057 | struct packet_type *ptype, *pt_prev; | |
ab95bfe0 | 3058 | rx_handler_func_t *rx_handler; |
f2ccd8fa | 3059 | struct net_device *orig_dev; |
0641e4fb | 3060 | struct net_device *master; |
0d7a3681 | 3061 | struct net_device *null_or_orig; |
2df4a0fa | 3062 | struct net_device *orig_or_bond; |
1da177e4 | 3063 | int ret = NET_RX_DROP; |
252e3346 | 3064 | __be16 type; |
1da177e4 | 3065 | |
3b098e2d ED |
3066 | if (!netdev_tstamp_prequeue) |
3067 | net_timestamp_check(skb); | |
81bbb3d4 | 3068 | |
cf66ba58 | 3069 | trace_netif_receive_skb(skb); |
9b22ea56 | 3070 | |
1da177e4 | 3071 | /* if we've gotten here through NAPI, check netpoll */ |
bea3348e | 3072 | if (netpoll_receive_skb(skb)) |
1da177e4 LT |
3073 | return NET_RX_DROP; |
3074 | ||
8964be4a ED |
3075 | if (!skb->skb_iif) |
3076 | skb->skb_iif = skb->dev->ifindex; | |
86e65da9 | 3077 | |
597a264b JF |
3078 | /* |
3079 | * bonding note: skbs received on inactive slaves should only | |
3080 | * be delivered to pkt handlers that are exact matches. Also | |
3081 | * the deliver_no_wcard flag will be set. If packet handlers | |
3082 | * are sensitive to duplicate packets these skbs will need to | |
3701e513 | 3083 | * be dropped at the handler. |
597a264b | 3084 | */ |
0d7a3681 | 3085 | null_or_orig = NULL; |
cc9bd5ce | 3086 | orig_dev = skb->dev; |
0641e4fb | 3087 | master = ACCESS_ONCE(orig_dev->master); |
597a264b JF |
3088 | if (skb->deliver_no_wcard) |
3089 | null_or_orig = orig_dev; | |
3090 | else if (master) { | |
3091 | if (skb_bond_should_drop(skb, master)) { | |
3092 | skb->deliver_no_wcard = 1; | |
0d7a3681 | 3093 | null_or_orig = orig_dev; /* deliver only exact match */ |
597a264b | 3094 | } else |
0641e4fb | 3095 | skb->dev = master; |
cc9bd5ce | 3096 | } |
8f903c70 | 3097 | |
27f39c73 | 3098 | __this_cpu_inc(softnet_data.processed); |
c1d2bbe1 | 3099 | skb_reset_network_header(skb); |
badff6d0 | 3100 | skb_reset_transport_header(skb); |
b0e380b1 | 3101 | skb->mac_len = skb->network_header - skb->mac_header; |
1da177e4 LT |
3102 | |
3103 | pt_prev = NULL; | |
3104 | ||
3105 | rcu_read_lock(); | |
3106 | ||
3107 | #ifdef CONFIG_NET_CLS_ACT | |
3108 | if (skb->tc_verd & TC_NCLS) { | |
3109 | skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); | |
3110 | goto ncls; | |
3111 | } | |
3112 | #endif | |
3113 | ||
3114 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
f982307f JE |
3115 | if (ptype->dev == null_or_orig || ptype->dev == skb->dev || |
3116 | ptype->dev == orig_dev) { | |
4ec93edb | 3117 | if (pt_prev) |
f2ccd8fa | 3118 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
3119 | pt_prev = ptype; |
3120 | } | |
3121 | } | |
3122 | ||
3123 | #ifdef CONFIG_NET_CLS_ACT | |
f697c3e8 HX |
3124 | skb = handle_ing(skb, &pt_prev, &ret, orig_dev); |
3125 | if (!skb) | |
1da177e4 | 3126 | goto out; |
1da177e4 LT |
3127 | ncls: |
3128 | #endif | |
3129 | ||
ab95bfe0 JP |
3130 | /* Handle special case of bridge or macvlan */ |
3131 | rx_handler = rcu_dereference(skb->dev->rx_handler); | |
3132 | if (rx_handler) { | |
3133 | if (pt_prev) { | |
3134 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3135 | pt_prev = NULL; | |
3136 | } | |
3137 | skb = rx_handler(skb); | |
3138 | if (!skb) | |
3139 | goto out; | |
3140 | } | |
1da177e4 | 3141 | |
3701e513 JG |
3142 | if (vlan_tx_tag_present(skb)) { |
3143 | if (pt_prev) { | |
3144 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3145 | pt_prev = NULL; | |
3146 | } | |
3147 | if (vlan_hwaccel_do_receive(&skb)) { | |
3148 | ret = __netif_receive_skb(skb); | |
3149 | goto out; | |
3150 | } else if (unlikely(!skb)) | |
3151 | goto out; | |
3152 | } | |
3153 | ||
1f3c8804 AG |
3154 | /* |
3155 | * Make sure frames received on VLAN interfaces stacked on | |
3156 | * bonding interfaces still make their way to any base bonding | |
3157 | * device that may have registered for a specific ptype. The | |
3158 | * handler may have to adjust skb->dev and orig_dev. | |
1f3c8804 | 3159 | */ |
2df4a0fa | 3160 | orig_or_bond = orig_dev; |
1f3c8804 AG |
3161 | if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) && |
3162 | (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) { | |
2df4a0fa | 3163 | orig_or_bond = vlan_dev_real_dev(skb->dev); |
1f3c8804 AG |
3164 | } |
3165 | ||
1da177e4 | 3166 | type = skb->protocol; |
82d8a867 PE |
3167 | list_for_each_entry_rcu(ptype, |
3168 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
1f3c8804 | 3169 | if (ptype->type == type && (ptype->dev == null_or_orig || |
ca8d9ea3 | 3170 | ptype->dev == skb->dev || ptype->dev == orig_dev || |
2df4a0fa | 3171 | ptype->dev == orig_or_bond)) { |
4ec93edb | 3172 | if (pt_prev) |
f2ccd8fa | 3173 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
3174 | pt_prev = ptype; |
3175 | } | |
3176 | } | |
3177 | ||
3178 | if (pt_prev) { | |
f2ccd8fa | 3179 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
1da177e4 | 3180 | } else { |
caf586e5 | 3181 | atomic_long_inc(&skb->dev->rx_dropped); |
1da177e4 LT |
3182 | kfree_skb(skb); |
3183 | /* Jamal, now you will not able to escape explaining | |
3184 | * me how you were going to use this. :-) | |
3185 | */ | |
3186 | ret = NET_RX_DROP; | |
3187 | } | |
3188 | ||
3189 | out: | |
3190 | rcu_read_unlock(); | |
3191 | return ret; | |
3192 | } | |
0a9627f2 TH |
3193 | |
3194 | /** | |
3195 | * netif_receive_skb - process receive buffer from network | |
3196 | * @skb: buffer to process | |
3197 | * | |
3198 | * netif_receive_skb() is the main receive data processing function. | |
3199 | * It always succeeds. The buffer may be dropped during processing | |
3200 | * for congestion control or by the protocol layers. | |
3201 | * | |
3202 | * This function may only be called from softirq context and interrupts | |
3203 | * should be enabled. | |
3204 | * | |
3205 | * Return values (usually ignored): | |
3206 | * NET_RX_SUCCESS: no congestion | |
3207 | * NET_RX_DROP: packet was dropped | |
3208 | */ | |
3209 | int netif_receive_skb(struct sk_buff *skb) | |
3210 | { | |
3b098e2d ED |
3211 | if (netdev_tstamp_prequeue) |
3212 | net_timestamp_check(skb); | |
3213 | ||
c1f19b51 RC |
3214 | if (skb_defer_rx_timestamp(skb)) |
3215 | return NET_RX_SUCCESS; | |
3216 | ||
df334545 | 3217 | #ifdef CONFIG_RPS |
3b098e2d ED |
3218 | { |
3219 | struct rps_dev_flow voidflow, *rflow = &voidflow; | |
3220 | int cpu, ret; | |
fec5e652 | 3221 | |
3b098e2d ED |
3222 | rcu_read_lock(); |
3223 | ||
3224 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
0a9627f2 | 3225 | |
3b098e2d ED |
3226 | if (cpu >= 0) { |
3227 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
3228 | rcu_read_unlock(); | |
3229 | } else { | |
3230 | rcu_read_unlock(); | |
3231 | ret = __netif_receive_skb(skb); | |
3232 | } | |
0a9627f2 | 3233 | |
3b098e2d | 3234 | return ret; |
fec5e652 | 3235 | } |
1e94d72f TH |
3236 | #else |
3237 | return __netif_receive_skb(skb); | |
3238 | #endif | |
0a9627f2 | 3239 | } |
d1b19dff | 3240 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 3241 | |
88751275 ED |
3242 | /* Network device is going away, flush any packets still pending |
3243 | * Called with irqs disabled. | |
3244 | */ | |
152102c7 | 3245 | static void flush_backlog(void *arg) |
6e583ce5 | 3246 | { |
152102c7 | 3247 | struct net_device *dev = arg; |
e36fa2f7 | 3248 | struct softnet_data *sd = &__get_cpu_var(softnet_data); |
6e583ce5 SH |
3249 | struct sk_buff *skb, *tmp; |
3250 | ||
e36fa2f7 | 3251 | rps_lock(sd); |
6e7676c1 | 3252 | skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { |
6e583ce5 | 3253 | if (skb->dev == dev) { |
e36fa2f7 | 3254 | __skb_unlink(skb, &sd->input_pkt_queue); |
6e583ce5 | 3255 | kfree_skb(skb); |
76cc8b13 | 3256 | input_queue_head_incr(sd); |
6e583ce5 | 3257 | } |
6e7676c1 | 3258 | } |
e36fa2f7 | 3259 | rps_unlock(sd); |
6e7676c1 CG |
3260 | |
3261 | skb_queue_walk_safe(&sd->process_queue, skb, tmp) { | |
3262 | if (skb->dev == dev) { | |
3263 | __skb_unlink(skb, &sd->process_queue); | |
3264 | kfree_skb(skb); | |
76cc8b13 | 3265 | input_queue_head_incr(sd); |
6e7676c1 CG |
3266 | } |
3267 | } | |
6e583ce5 SH |
3268 | } |
3269 | ||
d565b0a1 HX |
3270 | static int napi_gro_complete(struct sk_buff *skb) |
3271 | { | |
3272 | struct packet_type *ptype; | |
3273 | __be16 type = skb->protocol; | |
3274 | struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK]; | |
3275 | int err = -ENOENT; | |
3276 | ||
fc59f9a3 HX |
3277 | if (NAPI_GRO_CB(skb)->count == 1) { |
3278 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 3279 | goto out; |
fc59f9a3 | 3280 | } |
d565b0a1 HX |
3281 | |
3282 | rcu_read_lock(); | |
3283 | list_for_each_entry_rcu(ptype, head, list) { | |
3284 | if (ptype->type != type || ptype->dev || !ptype->gro_complete) | |
3285 | continue; | |
3286 | ||
3287 | err = ptype->gro_complete(skb); | |
3288 | break; | |
3289 | } | |
3290 | rcu_read_unlock(); | |
3291 | ||
3292 | if (err) { | |
3293 | WARN_ON(&ptype->list == head); | |
3294 | kfree_skb(skb); | |
3295 | return NET_RX_SUCCESS; | |
3296 | } | |
3297 | ||
3298 | out: | |
d565b0a1 HX |
3299 | return netif_receive_skb(skb); |
3300 | } | |
3301 | ||
86cac58b | 3302 | inline void napi_gro_flush(struct napi_struct *napi) |
d565b0a1 HX |
3303 | { |
3304 | struct sk_buff *skb, *next; | |
3305 | ||
3306 | for (skb = napi->gro_list; skb; skb = next) { | |
3307 | next = skb->next; | |
3308 | skb->next = NULL; | |
3309 | napi_gro_complete(skb); | |
3310 | } | |
3311 | ||
4ae5544f | 3312 | napi->gro_count = 0; |
d565b0a1 HX |
3313 | napi->gro_list = NULL; |
3314 | } | |
86cac58b | 3315 | EXPORT_SYMBOL(napi_gro_flush); |
d565b0a1 | 3316 | |
5b252f0c | 3317 | enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 HX |
3318 | { |
3319 | struct sk_buff **pp = NULL; | |
3320 | struct packet_type *ptype; | |
3321 | __be16 type = skb->protocol; | |
3322 | struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK]; | |
0da2afd5 | 3323 | int same_flow; |
d565b0a1 | 3324 | int mac_len; |
5b252f0c | 3325 | enum gro_result ret; |
d565b0a1 | 3326 | |
ce9e76c8 | 3327 | if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb)) |
d565b0a1 HX |
3328 | goto normal; |
3329 | ||
21dc3301 | 3330 | if (skb_is_gso(skb) || skb_has_frag_list(skb)) |
f17f5c91 HX |
3331 | goto normal; |
3332 | ||
d565b0a1 HX |
3333 | rcu_read_lock(); |
3334 | list_for_each_entry_rcu(ptype, head, list) { | |
d565b0a1 HX |
3335 | if (ptype->type != type || ptype->dev || !ptype->gro_receive) |
3336 | continue; | |
3337 | ||
86911732 | 3338 | skb_set_network_header(skb, skb_gro_offset(skb)); |
d565b0a1 HX |
3339 | mac_len = skb->network_header - skb->mac_header; |
3340 | skb->mac_len = mac_len; | |
3341 | NAPI_GRO_CB(skb)->same_flow = 0; | |
3342 | NAPI_GRO_CB(skb)->flush = 0; | |
5d38a079 | 3343 | NAPI_GRO_CB(skb)->free = 0; |
d565b0a1 | 3344 | |
d565b0a1 HX |
3345 | pp = ptype->gro_receive(&napi->gro_list, skb); |
3346 | break; | |
3347 | } | |
3348 | rcu_read_unlock(); | |
3349 | ||
3350 | if (&ptype->list == head) | |
3351 | goto normal; | |
3352 | ||
0da2afd5 | 3353 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 3354 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 3355 | |
d565b0a1 HX |
3356 | if (pp) { |
3357 | struct sk_buff *nskb = *pp; | |
3358 | ||
3359 | *pp = nskb->next; | |
3360 | nskb->next = NULL; | |
3361 | napi_gro_complete(nskb); | |
4ae5544f | 3362 | napi->gro_count--; |
d565b0a1 HX |
3363 | } |
3364 | ||
0da2afd5 | 3365 | if (same_flow) |
d565b0a1 HX |
3366 | goto ok; |
3367 | ||
4ae5544f | 3368 | if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS) |
d565b0a1 | 3369 | goto normal; |
d565b0a1 | 3370 | |
4ae5544f | 3371 | napi->gro_count++; |
d565b0a1 | 3372 | NAPI_GRO_CB(skb)->count = 1; |
86911732 | 3373 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
d565b0a1 HX |
3374 | skb->next = napi->gro_list; |
3375 | napi->gro_list = skb; | |
5d0d9be8 | 3376 | ret = GRO_HELD; |
d565b0a1 | 3377 | |
ad0f9904 | 3378 | pull: |
cb18978c HX |
3379 | if (skb_headlen(skb) < skb_gro_offset(skb)) { |
3380 | int grow = skb_gro_offset(skb) - skb_headlen(skb); | |
3381 | ||
3382 | BUG_ON(skb->end - skb->tail < grow); | |
3383 | ||
3384 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
3385 | ||
3386 | skb->tail += grow; | |
3387 | skb->data_len -= grow; | |
3388 | ||
3389 | skb_shinfo(skb)->frags[0].page_offset += grow; | |
3390 | skb_shinfo(skb)->frags[0].size -= grow; | |
3391 | ||
3392 | if (unlikely(!skb_shinfo(skb)->frags[0].size)) { | |
3393 | put_page(skb_shinfo(skb)->frags[0].page); | |
3394 | memmove(skb_shinfo(skb)->frags, | |
3395 | skb_shinfo(skb)->frags + 1, | |
e5093aec | 3396 | --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t)); |
cb18978c | 3397 | } |
ad0f9904 HX |
3398 | } |
3399 | ||
d565b0a1 | 3400 | ok: |
5d0d9be8 | 3401 | return ret; |
d565b0a1 HX |
3402 | |
3403 | normal: | |
ad0f9904 HX |
3404 | ret = GRO_NORMAL; |
3405 | goto pull; | |
5d38a079 | 3406 | } |
96e93eab HX |
3407 | EXPORT_SYMBOL(dev_gro_receive); |
3408 | ||
40d0802b | 3409 | static inline gro_result_t |
5b252f0c | 3410 | __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab HX |
3411 | { |
3412 | struct sk_buff *p; | |
3413 | ||
3414 | for (p = napi->gro_list; p; p = p->next) { | |
40d0802b ED |
3415 | unsigned long diffs; |
3416 | ||
3417 | diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; | |
3701e513 | 3418 | diffs |= p->vlan_tci ^ skb->vlan_tci; |
40d0802b | 3419 | diffs |= compare_ether_header(skb_mac_header(p), |
f64f9e71 | 3420 | skb_gro_mac_header(skb)); |
40d0802b | 3421 | NAPI_GRO_CB(p)->same_flow = !diffs; |
96e93eab HX |
3422 | NAPI_GRO_CB(p)->flush = 0; |
3423 | } | |
3424 | ||
3425 | return dev_gro_receive(napi, skb); | |
3426 | } | |
5d38a079 | 3427 | |
c7c4b3b6 | 3428 | gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) |
5d38a079 | 3429 | { |
5d0d9be8 HX |
3430 | switch (ret) { |
3431 | case GRO_NORMAL: | |
c7c4b3b6 BH |
3432 | if (netif_receive_skb(skb)) |
3433 | ret = GRO_DROP; | |
3434 | break; | |
5d38a079 | 3435 | |
5d0d9be8 | 3436 | case GRO_DROP: |
5d0d9be8 | 3437 | case GRO_MERGED_FREE: |
5d38a079 HX |
3438 | kfree_skb(skb); |
3439 | break; | |
5b252f0c BH |
3440 | |
3441 | case GRO_HELD: | |
3442 | case GRO_MERGED: | |
3443 | break; | |
5d38a079 HX |
3444 | } |
3445 | ||
c7c4b3b6 | 3446 | return ret; |
5d0d9be8 HX |
3447 | } |
3448 | EXPORT_SYMBOL(napi_skb_finish); | |
3449 | ||
78a478d0 HX |
3450 | void skb_gro_reset_offset(struct sk_buff *skb) |
3451 | { | |
3452 | NAPI_GRO_CB(skb)->data_offset = 0; | |
3453 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
7489594c | 3454 | NAPI_GRO_CB(skb)->frag0_len = 0; |
78a478d0 | 3455 | |
78d3fd0b | 3456 | if (skb->mac_header == skb->tail && |
7489594c | 3457 | !PageHighMem(skb_shinfo(skb)->frags[0].page)) { |
78a478d0 HX |
3458 | NAPI_GRO_CB(skb)->frag0 = |
3459 | page_address(skb_shinfo(skb)->frags[0].page) + | |
3460 | skb_shinfo(skb)->frags[0].page_offset; | |
7489594c HX |
3461 | NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size; |
3462 | } | |
78a478d0 HX |
3463 | } |
3464 | EXPORT_SYMBOL(skb_gro_reset_offset); | |
3465 | ||
c7c4b3b6 | 3466 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 3467 | { |
86911732 HX |
3468 | skb_gro_reset_offset(skb); |
3469 | ||
5d0d9be8 | 3470 | return napi_skb_finish(__napi_gro_receive(napi, skb), skb); |
d565b0a1 HX |
3471 | } |
3472 | EXPORT_SYMBOL(napi_gro_receive); | |
3473 | ||
d0c2b0d2 | 3474 | static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab | 3475 | { |
96e93eab HX |
3476 | __skb_pull(skb, skb_headlen(skb)); |
3477 | skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb)); | |
3701e513 | 3478 | skb->vlan_tci = 0; |
96e93eab HX |
3479 | |
3480 | napi->skb = skb; | |
3481 | } | |
96e93eab | 3482 | |
76620aaf | 3483 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 3484 | { |
5d38a079 | 3485 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
3486 | |
3487 | if (!skb) { | |
89d71a66 ED |
3488 | skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD); |
3489 | if (skb) | |
3490 | napi->skb = skb; | |
80595d59 | 3491 | } |
96e93eab HX |
3492 | return skb; |
3493 | } | |
76620aaf | 3494 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 3495 | |
c7c4b3b6 BH |
3496 | gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb, |
3497 | gro_result_t ret) | |
96e93eab | 3498 | { |
5d0d9be8 HX |
3499 | switch (ret) { |
3500 | case GRO_NORMAL: | |
86911732 | 3501 | case GRO_HELD: |
e76b69cc | 3502 | skb->protocol = eth_type_trans(skb, skb->dev); |
86911732 | 3503 | |
c7c4b3b6 BH |
3504 | if (ret == GRO_HELD) |
3505 | skb_gro_pull(skb, -ETH_HLEN); | |
3506 | else if (netif_receive_skb(skb)) | |
3507 | ret = GRO_DROP; | |
86911732 | 3508 | break; |
5d38a079 | 3509 | |
5d0d9be8 | 3510 | case GRO_DROP: |
5d0d9be8 HX |
3511 | case GRO_MERGED_FREE: |
3512 | napi_reuse_skb(napi, skb); | |
3513 | break; | |
5b252f0c BH |
3514 | |
3515 | case GRO_MERGED: | |
3516 | break; | |
5d0d9be8 | 3517 | } |
5d38a079 | 3518 | |
c7c4b3b6 | 3519 | return ret; |
5d38a079 | 3520 | } |
5d0d9be8 HX |
3521 | EXPORT_SYMBOL(napi_frags_finish); |
3522 | ||
76620aaf HX |
3523 | struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
3524 | { | |
3525 | struct sk_buff *skb = napi->skb; | |
3526 | struct ethhdr *eth; | |
a5b1cf28 HX |
3527 | unsigned int hlen; |
3528 | unsigned int off; | |
76620aaf HX |
3529 | |
3530 | napi->skb = NULL; | |
3531 | ||
3532 | skb_reset_mac_header(skb); | |
3533 | skb_gro_reset_offset(skb); | |
3534 | ||
a5b1cf28 HX |
3535 | off = skb_gro_offset(skb); |
3536 | hlen = off + sizeof(*eth); | |
3537 | eth = skb_gro_header_fast(skb, off); | |
3538 | if (skb_gro_header_hard(skb, hlen)) { | |
3539 | eth = skb_gro_header_slow(skb, hlen, off); | |
3540 | if (unlikely(!eth)) { | |
3541 | napi_reuse_skb(napi, skb); | |
3542 | skb = NULL; | |
3543 | goto out; | |
3544 | } | |
76620aaf HX |
3545 | } |
3546 | ||
3547 | skb_gro_pull(skb, sizeof(*eth)); | |
3548 | ||
3549 | /* | |
3550 | * This works because the only protocols we care about don't require | |
3551 | * special handling. We'll fix it up properly at the end. | |
3552 | */ | |
3553 | skb->protocol = eth->h_proto; | |
3554 | ||
3555 | out: | |
3556 | return skb; | |
3557 | } | |
3558 | EXPORT_SYMBOL(napi_frags_skb); | |
3559 | ||
c7c4b3b6 | 3560 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 3561 | { |
76620aaf | 3562 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 HX |
3563 | |
3564 | if (!skb) | |
c7c4b3b6 | 3565 | return GRO_DROP; |
5d0d9be8 HX |
3566 | |
3567 | return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb)); | |
3568 | } | |
5d38a079 HX |
3569 | EXPORT_SYMBOL(napi_gro_frags); |
3570 | ||
e326bed2 ED |
3571 | /* |
3572 | * net_rps_action sends any pending IPI's for rps. | |
3573 | * Note: called with local irq disabled, but exits with local irq enabled. | |
3574 | */ | |
3575 | static void net_rps_action_and_irq_enable(struct softnet_data *sd) | |
3576 | { | |
3577 | #ifdef CONFIG_RPS | |
3578 | struct softnet_data *remsd = sd->rps_ipi_list; | |
3579 | ||
3580 | if (remsd) { | |
3581 | sd->rps_ipi_list = NULL; | |
3582 | ||
3583 | local_irq_enable(); | |
3584 | ||
3585 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
3586 | while (remsd) { | |
3587 | struct softnet_data *next = remsd->rps_ipi_next; | |
3588 | ||
3589 | if (cpu_online(remsd->cpu)) | |
3590 | __smp_call_function_single(remsd->cpu, | |
3591 | &remsd->csd, 0); | |
3592 | remsd = next; | |
3593 | } | |
3594 | } else | |
3595 | #endif | |
3596 | local_irq_enable(); | |
3597 | } | |
3598 | ||
bea3348e | 3599 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 LT |
3600 | { |
3601 | int work = 0; | |
eecfd7c4 | 3602 | struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); |
1da177e4 | 3603 | |
e326bed2 ED |
3604 | #ifdef CONFIG_RPS |
3605 | /* Check if we have pending ipi, its better to send them now, | |
3606 | * not waiting net_rx_action() end. | |
3607 | */ | |
3608 | if (sd->rps_ipi_list) { | |
3609 | local_irq_disable(); | |
3610 | net_rps_action_and_irq_enable(sd); | |
3611 | } | |
3612 | #endif | |
bea3348e | 3613 | napi->weight = weight_p; |
6e7676c1 CG |
3614 | local_irq_disable(); |
3615 | while (work < quota) { | |
1da177e4 | 3616 | struct sk_buff *skb; |
6e7676c1 CG |
3617 | unsigned int qlen; |
3618 | ||
3619 | while ((skb = __skb_dequeue(&sd->process_queue))) { | |
3620 | local_irq_enable(); | |
3621 | __netif_receive_skb(skb); | |
6e7676c1 | 3622 | local_irq_disable(); |
76cc8b13 TH |
3623 | input_queue_head_incr(sd); |
3624 | if (++work >= quota) { | |
3625 | local_irq_enable(); | |
3626 | return work; | |
3627 | } | |
6e7676c1 | 3628 | } |
1da177e4 | 3629 | |
e36fa2f7 | 3630 | rps_lock(sd); |
6e7676c1 | 3631 | qlen = skb_queue_len(&sd->input_pkt_queue); |
76cc8b13 | 3632 | if (qlen) |
6e7676c1 CG |
3633 | skb_queue_splice_tail_init(&sd->input_pkt_queue, |
3634 | &sd->process_queue); | |
76cc8b13 | 3635 | |
6e7676c1 | 3636 | if (qlen < quota - work) { |
eecfd7c4 ED |
3637 | /* |
3638 | * Inline a custom version of __napi_complete(). | |
3639 | * only current cpu owns and manipulates this napi, | |
3640 | * and NAPI_STATE_SCHED is the only possible flag set on backlog. | |
3641 | * we can use a plain write instead of clear_bit(), | |
3642 | * and we dont need an smp_mb() memory barrier. | |
3643 | */ | |
3644 | list_del(&napi->poll_list); | |
3645 | napi->state = 0; | |
3646 | ||
6e7676c1 | 3647 | quota = work + qlen; |
bea3348e | 3648 | } |
e36fa2f7 | 3649 | rps_unlock(sd); |
6e7676c1 CG |
3650 | } |
3651 | local_irq_enable(); | |
1da177e4 | 3652 | |
bea3348e SH |
3653 | return work; |
3654 | } | |
1da177e4 | 3655 | |
bea3348e SH |
3656 | /** |
3657 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 3658 | * @n: entry to schedule |
bea3348e SH |
3659 | * |
3660 | * The entry's receive function will be scheduled to run | |
3661 | */ | |
b5606c2d | 3662 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
3663 | { |
3664 | unsigned long flags; | |
1da177e4 | 3665 | |
bea3348e | 3666 | local_irq_save(flags); |
eecfd7c4 | 3667 | ____napi_schedule(&__get_cpu_var(softnet_data), n); |
bea3348e | 3668 | local_irq_restore(flags); |
1da177e4 | 3669 | } |
bea3348e SH |
3670 | EXPORT_SYMBOL(__napi_schedule); |
3671 | ||
d565b0a1 HX |
3672 | void __napi_complete(struct napi_struct *n) |
3673 | { | |
3674 | BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); | |
3675 | BUG_ON(n->gro_list); | |
3676 | ||
3677 | list_del(&n->poll_list); | |
3678 | smp_mb__before_clear_bit(); | |
3679 | clear_bit(NAPI_STATE_SCHED, &n->state); | |
3680 | } | |
3681 | EXPORT_SYMBOL(__napi_complete); | |
3682 | ||
3683 | void napi_complete(struct napi_struct *n) | |
3684 | { | |
3685 | unsigned long flags; | |
3686 | ||
3687 | /* | |
3688 | * don't let napi dequeue from the cpu poll list | |
3689 | * just in case its running on a different cpu | |
3690 | */ | |
3691 | if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state))) | |
3692 | return; | |
3693 | ||
3694 | napi_gro_flush(n); | |
3695 | local_irq_save(flags); | |
3696 | __napi_complete(n); | |
3697 | local_irq_restore(flags); | |
3698 | } | |
3699 | EXPORT_SYMBOL(napi_complete); | |
3700 | ||
3701 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, | |
3702 | int (*poll)(struct napi_struct *, int), int weight) | |
3703 | { | |
3704 | INIT_LIST_HEAD(&napi->poll_list); | |
4ae5544f | 3705 | napi->gro_count = 0; |
d565b0a1 | 3706 | napi->gro_list = NULL; |
5d38a079 | 3707 | napi->skb = NULL; |
d565b0a1 HX |
3708 | napi->poll = poll; |
3709 | napi->weight = weight; | |
3710 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 3711 | napi->dev = dev; |
5d38a079 | 3712 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
3713 | spin_lock_init(&napi->poll_lock); |
3714 | napi->poll_owner = -1; | |
3715 | #endif | |
3716 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
3717 | } | |
3718 | EXPORT_SYMBOL(netif_napi_add); | |
3719 | ||
3720 | void netif_napi_del(struct napi_struct *napi) | |
3721 | { | |
3722 | struct sk_buff *skb, *next; | |
3723 | ||
d7b06636 | 3724 | list_del_init(&napi->dev_list); |
76620aaf | 3725 | napi_free_frags(napi); |
d565b0a1 HX |
3726 | |
3727 | for (skb = napi->gro_list; skb; skb = next) { | |
3728 | next = skb->next; | |
3729 | skb->next = NULL; | |
3730 | kfree_skb(skb); | |
3731 | } | |
3732 | ||
3733 | napi->gro_list = NULL; | |
4ae5544f | 3734 | napi->gro_count = 0; |
d565b0a1 HX |
3735 | } |
3736 | EXPORT_SYMBOL(netif_napi_del); | |
3737 | ||
1da177e4 LT |
3738 | static void net_rx_action(struct softirq_action *h) |
3739 | { | |
e326bed2 | 3740 | struct softnet_data *sd = &__get_cpu_var(softnet_data); |
24f8b238 | 3741 | unsigned long time_limit = jiffies + 2; |
51b0bded | 3742 | int budget = netdev_budget; |
53fb95d3 MM |
3743 | void *have; |
3744 | ||
1da177e4 LT |
3745 | local_irq_disable(); |
3746 | ||
e326bed2 | 3747 | while (!list_empty(&sd->poll_list)) { |
bea3348e SH |
3748 | struct napi_struct *n; |
3749 | int work, weight; | |
1da177e4 | 3750 | |
bea3348e | 3751 | /* If softirq window is exhuasted then punt. |
24f8b238 SH |
3752 | * Allow this to run for 2 jiffies since which will allow |
3753 | * an average latency of 1.5/HZ. | |
bea3348e | 3754 | */ |
24f8b238 | 3755 | if (unlikely(budget <= 0 || time_after(jiffies, time_limit))) |
1da177e4 LT |
3756 | goto softnet_break; |
3757 | ||
3758 | local_irq_enable(); | |
3759 | ||
bea3348e SH |
3760 | /* Even though interrupts have been re-enabled, this |
3761 | * access is safe because interrupts can only add new | |
3762 | * entries to the tail of this list, and only ->poll() | |
3763 | * calls can remove this head entry from the list. | |
3764 | */ | |
e326bed2 | 3765 | n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list); |
1da177e4 | 3766 | |
bea3348e SH |
3767 | have = netpoll_poll_lock(n); |
3768 | ||
3769 | weight = n->weight; | |
3770 | ||
0a7606c1 DM |
3771 | /* This NAPI_STATE_SCHED test is for avoiding a race |
3772 | * with netpoll's poll_napi(). Only the entity which | |
3773 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
3774 | * actually make the ->poll() call. Therefore we avoid | |
3775 | * accidently calling ->poll() when NAPI is not scheduled. | |
3776 | */ | |
3777 | work = 0; | |
4ea7e386 | 3778 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { |
0a7606c1 | 3779 | work = n->poll(n, weight); |
4ea7e386 NH |
3780 | trace_napi_poll(n); |
3781 | } | |
bea3348e SH |
3782 | |
3783 | WARN_ON_ONCE(work > weight); | |
3784 | ||
3785 | budget -= work; | |
3786 | ||
3787 | local_irq_disable(); | |
3788 | ||
3789 | /* Drivers must not modify the NAPI state if they | |
3790 | * consume the entire weight. In such cases this code | |
3791 | * still "owns" the NAPI instance and therefore can | |
3792 | * move the instance around on the list at-will. | |
3793 | */ | |
fed17f30 | 3794 | if (unlikely(work == weight)) { |
ff780cd8 HX |
3795 | if (unlikely(napi_disable_pending(n))) { |
3796 | local_irq_enable(); | |
3797 | napi_complete(n); | |
3798 | local_irq_disable(); | |
3799 | } else | |
e326bed2 | 3800 | list_move_tail(&n->poll_list, &sd->poll_list); |
fed17f30 | 3801 | } |
bea3348e SH |
3802 | |
3803 | netpoll_poll_unlock(have); | |
1da177e4 LT |
3804 | } |
3805 | out: | |
e326bed2 | 3806 | net_rps_action_and_irq_enable(sd); |
0a9627f2 | 3807 | |
db217334 CL |
3808 | #ifdef CONFIG_NET_DMA |
3809 | /* | |
3810 | * There may not be any more sk_buffs coming right now, so push | |
3811 | * any pending DMA copies to hardware | |
3812 | */ | |
2ba05622 | 3813 | dma_issue_pending_all(); |
db217334 | 3814 | #endif |
bea3348e | 3815 | |
1da177e4 LT |
3816 | return; |
3817 | ||
3818 | softnet_break: | |
dee42870 | 3819 | sd->time_squeeze++; |
1da177e4 LT |
3820 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); |
3821 | goto out; | |
3822 | } | |
3823 | ||
d1b19dff | 3824 | static gifconf_func_t *gifconf_list[NPROTO]; |
1da177e4 LT |
3825 | |
3826 | /** | |
3827 | * register_gifconf - register a SIOCGIF handler | |
3828 | * @family: Address family | |
3829 | * @gifconf: Function handler | |
3830 | * | |
3831 | * Register protocol dependent address dumping routines. The handler | |
3832 | * that is passed must not be freed or reused until it has been replaced | |
3833 | * by another handler. | |
3834 | */ | |
d1b19dff | 3835 | int register_gifconf(unsigned int family, gifconf_func_t *gifconf) |
1da177e4 LT |
3836 | { |
3837 | if (family >= NPROTO) | |
3838 | return -EINVAL; | |
3839 | gifconf_list[family] = gifconf; | |
3840 | return 0; | |
3841 | } | |
d1b19dff | 3842 | EXPORT_SYMBOL(register_gifconf); |
1da177e4 LT |
3843 | |
3844 | ||
3845 | /* | |
3846 | * Map an interface index to its name (SIOCGIFNAME) | |
3847 | */ | |
3848 | ||
3849 | /* | |
3850 | * We need this ioctl for efficient implementation of the | |
3851 | * if_indextoname() function required by the IPv6 API. Without | |
3852 | * it, we would have to search all the interfaces to find a | |
3853 | * match. --pb | |
3854 | */ | |
3855 | ||
881d966b | 3856 | static int dev_ifname(struct net *net, struct ifreq __user *arg) |
1da177e4 LT |
3857 | { |
3858 | struct net_device *dev; | |
3859 | struct ifreq ifr; | |
3860 | ||
3861 | /* | |
3862 | * Fetch the caller's info block. | |
3863 | */ | |
3864 | ||
3865 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
3866 | return -EFAULT; | |
3867 | ||
fb699dfd ED |
3868 | rcu_read_lock(); |
3869 | dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex); | |
1da177e4 | 3870 | if (!dev) { |
fb699dfd | 3871 | rcu_read_unlock(); |
1da177e4 LT |
3872 | return -ENODEV; |
3873 | } | |
3874 | ||
3875 | strcpy(ifr.ifr_name, dev->name); | |
fb699dfd | 3876 | rcu_read_unlock(); |
1da177e4 LT |
3877 | |
3878 | if (copy_to_user(arg, &ifr, sizeof(struct ifreq))) | |
3879 | return -EFAULT; | |
3880 | return 0; | |
3881 | } | |
3882 | ||
3883 | /* | |
3884 | * Perform a SIOCGIFCONF call. This structure will change | |
3885 | * size eventually, and there is nothing I can do about it. | |
3886 | * Thus we will need a 'compatibility mode'. | |
3887 | */ | |
3888 | ||
881d966b | 3889 | static int dev_ifconf(struct net *net, char __user *arg) |
1da177e4 LT |
3890 | { |
3891 | struct ifconf ifc; | |
3892 | struct net_device *dev; | |
3893 | char __user *pos; | |
3894 | int len; | |
3895 | int total; | |
3896 | int i; | |
3897 | ||
3898 | /* | |
3899 | * Fetch the caller's info block. | |
3900 | */ | |
3901 | ||
3902 | if (copy_from_user(&ifc, arg, sizeof(struct ifconf))) | |
3903 | return -EFAULT; | |
3904 | ||
3905 | pos = ifc.ifc_buf; | |
3906 | len = ifc.ifc_len; | |
3907 | ||
3908 | /* | |
3909 | * Loop over the interfaces, and write an info block for each. | |
3910 | */ | |
3911 | ||
3912 | total = 0; | |
881d966b | 3913 | for_each_netdev(net, dev) { |
1da177e4 LT |
3914 | for (i = 0; i < NPROTO; i++) { |
3915 | if (gifconf_list[i]) { | |
3916 | int done; | |
3917 | if (!pos) | |
3918 | done = gifconf_list[i](dev, NULL, 0); | |
3919 | else | |
3920 | done = gifconf_list[i](dev, pos + total, | |
3921 | len - total); | |
3922 | if (done < 0) | |
3923 | return -EFAULT; | |
3924 | total += done; | |
3925 | } | |
3926 | } | |
4ec93edb | 3927 | } |
1da177e4 LT |
3928 | |
3929 | /* | |
3930 | * All done. Write the updated control block back to the caller. | |
3931 | */ | |
3932 | ifc.ifc_len = total; | |
3933 | ||
3934 | /* | |
3935 | * Both BSD and Solaris return 0 here, so we do too. | |
3936 | */ | |
3937 | return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0; | |
3938 | } | |
3939 | ||
3940 | #ifdef CONFIG_PROC_FS | |
3941 | /* | |
3942 | * This is invoked by the /proc filesystem handler to display a device | |
3943 | * in detail. | |
3944 | */ | |
7562f876 | 3945 | void *dev_seq_start(struct seq_file *seq, loff_t *pos) |
c6d14c84 | 3946 | __acquires(RCU) |
1da177e4 | 3947 | { |
e372c414 | 3948 | struct net *net = seq_file_net(seq); |
7562f876 | 3949 | loff_t off; |
1da177e4 | 3950 | struct net_device *dev; |
1da177e4 | 3951 | |
c6d14c84 | 3952 | rcu_read_lock(); |
7562f876 PE |
3953 | if (!*pos) |
3954 | return SEQ_START_TOKEN; | |
1da177e4 | 3955 | |
7562f876 | 3956 | off = 1; |
c6d14c84 | 3957 | for_each_netdev_rcu(net, dev) |
7562f876 PE |
3958 | if (off++ == *pos) |
3959 | return dev; | |
1da177e4 | 3960 | |
7562f876 | 3961 | return NULL; |
1da177e4 LT |
3962 | } |
3963 | ||
3964 | void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
3965 | { | |
c6d14c84 ED |
3966 | struct net_device *dev = (v == SEQ_START_TOKEN) ? |
3967 | first_net_device(seq_file_net(seq)) : | |
3968 | next_net_device((struct net_device *)v); | |
3969 | ||
1da177e4 | 3970 | ++*pos; |
c6d14c84 | 3971 | return rcu_dereference(dev); |
1da177e4 LT |
3972 | } |
3973 | ||
3974 | void dev_seq_stop(struct seq_file *seq, void *v) | |
c6d14c84 | 3975 | __releases(RCU) |
1da177e4 | 3976 | { |
c6d14c84 | 3977 | rcu_read_unlock(); |
1da177e4 LT |
3978 | } |
3979 | ||
3980 | static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev) | |
3981 | { | |
28172739 ED |
3982 | struct rtnl_link_stats64 temp; |
3983 | const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp); | |
1da177e4 | 3984 | |
be1f3c2c BH |
3985 | seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu " |
3986 | "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n", | |
5a1b5898 RR |
3987 | dev->name, stats->rx_bytes, stats->rx_packets, |
3988 | stats->rx_errors, | |
3989 | stats->rx_dropped + stats->rx_missed_errors, | |
3990 | stats->rx_fifo_errors, | |
3991 | stats->rx_length_errors + stats->rx_over_errors + | |
3992 | stats->rx_crc_errors + stats->rx_frame_errors, | |
3993 | stats->rx_compressed, stats->multicast, | |
3994 | stats->tx_bytes, stats->tx_packets, | |
3995 | stats->tx_errors, stats->tx_dropped, | |
3996 | stats->tx_fifo_errors, stats->collisions, | |
3997 | stats->tx_carrier_errors + | |
3998 | stats->tx_aborted_errors + | |
3999 | stats->tx_window_errors + | |
4000 | stats->tx_heartbeat_errors, | |
4001 | stats->tx_compressed); | |
1da177e4 LT |
4002 | } |
4003 | ||
4004 | /* | |
4005 | * Called from the PROCfs module. This now uses the new arbitrary sized | |
4006 | * /proc/net interface to create /proc/net/dev | |
4007 | */ | |
4008 | static int dev_seq_show(struct seq_file *seq, void *v) | |
4009 | { | |
4010 | if (v == SEQ_START_TOKEN) | |
4011 | seq_puts(seq, "Inter-| Receive " | |
4012 | " | Transmit\n" | |
4013 | " face |bytes packets errs drop fifo frame " | |
4014 | "compressed multicast|bytes packets errs " | |
4015 | "drop fifo colls carrier compressed\n"); | |
4016 | else | |
4017 | dev_seq_printf_stats(seq, v); | |
4018 | return 0; | |
4019 | } | |
4020 | ||
dee42870 | 4021 | static struct softnet_data *softnet_get_online(loff_t *pos) |
1da177e4 | 4022 | { |
dee42870 | 4023 | struct softnet_data *sd = NULL; |
1da177e4 | 4024 | |
0c0b0aca | 4025 | while (*pos < nr_cpu_ids) |
4ec93edb | 4026 | if (cpu_online(*pos)) { |
dee42870 | 4027 | sd = &per_cpu(softnet_data, *pos); |
1da177e4 LT |
4028 | break; |
4029 | } else | |
4030 | ++*pos; | |
dee42870 | 4031 | return sd; |
1da177e4 LT |
4032 | } |
4033 | ||
4034 | static void *softnet_seq_start(struct seq_file *seq, loff_t *pos) | |
4035 | { | |
4036 | return softnet_get_online(pos); | |
4037 | } | |
4038 | ||
4039 | static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
4040 | { | |
4041 | ++*pos; | |
4042 | return softnet_get_online(pos); | |
4043 | } | |
4044 | ||
4045 | static void softnet_seq_stop(struct seq_file *seq, void *v) | |
4046 | { | |
4047 | } | |
4048 | ||
4049 | static int softnet_seq_show(struct seq_file *seq, void *v) | |
4050 | { | |
dee42870 | 4051 | struct softnet_data *sd = v; |
1da177e4 | 4052 | |
0a9627f2 | 4053 | seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n", |
dee42870 | 4054 | sd->processed, sd->dropped, sd->time_squeeze, 0, |
c1ebcdb8 | 4055 | 0, 0, 0, 0, /* was fastroute */ |
dee42870 | 4056 | sd->cpu_collision, sd->received_rps); |
1da177e4 LT |
4057 | return 0; |
4058 | } | |
4059 | ||
f690808e | 4060 | static const struct seq_operations dev_seq_ops = { |
1da177e4 LT |
4061 | .start = dev_seq_start, |
4062 | .next = dev_seq_next, | |
4063 | .stop = dev_seq_stop, | |
4064 | .show = dev_seq_show, | |
4065 | }; | |
4066 | ||
4067 | static int dev_seq_open(struct inode *inode, struct file *file) | |
4068 | { | |
e372c414 DL |
4069 | return seq_open_net(inode, file, &dev_seq_ops, |
4070 | sizeof(struct seq_net_private)); | |
1da177e4 LT |
4071 | } |
4072 | ||
9a32144e | 4073 | static const struct file_operations dev_seq_fops = { |
1da177e4 LT |
4074 | .owner = THIS_MODULE, |
4075 | .open = dev_seq_open, | |
4076 | .read = seq_read, | |
4077 | .llseek = seq_lseek, | |
e372c414 | 4078 | .release = seq_release_net, |
1da177e4 LT |
4079 | }; |
4080 | ||
f690808e | 4081 | static const struct seq_operations softnet_seq_ops = { |
1da177e4 LT |
4082 | .start = softnet_seq_start, |
4083 | .next = softnet_seq_next, | |
4084 | .stop = softnet_seq_stop, | |
4085 | .show = softnet_seq_show, | |
4086 | }; | |
4087 | ||
4088 | static int softnet_seq_open(struct inode *inode, struct file *file) | |
4089 | { | |
4090 | return seq_open(file, &softnet_seq_ops); | |
4091 | } | |
4092 | ||
9a32144e | 4093 | static const struct file_operations softnet_seq_fops = { |
1da177e4 LT |
4094 | .owner = THIS_MODULE, |
4095 | .open = softnet_seq_open, | |
4096 | .read = seq_read, | |
4097 | .llseek = seq_lseek, | |
4098 | .release = seq_release, | |
4099 | }; | |
4100 | ||
0e1256ff SH |
4101 | static void *ptype_get_idx(loff_t pos) |
4102 | { | |
4103 | struct packet_type *pt = NULL; | |
4104 | loff_t i = 0; | |
4105 | int t; | |
4106 | ||
4107 | list_for_each_entry_rcu(pt, &ptype_all, list) { | |
4108 | if (i == pos) | |
4109 | return pt; | |
4110 | ++i; | |
4111 | } | |
4112 | ||
82d8a867 | 4113 | for (t = 0; t < PTYPE_HASH_SIZE; t++) { |
0e1256ff SH |
4114 | list_for_each_entry_rcu(pt, &ptype_base[t], list) { |
4115 | if (i == pos) | |
4116 | return pt; | |
4117 | ++i; | |
4118 | } | |
4119 | } | |
4120 | return NULL; | |
4121 | } | |
4122 | ||
4123 | static void *ptype_seq_start(struct seq_file *seq, loff_t *pos) | |
72348a42 | 4124 | __acquires(RCU) |
0e1256ff SH |
4125 | { |
4126 | rcu_read_lock(); | |
4127 | return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN; | |
4128 | } | |
4129 | ||
4130 | static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
4131 | { | |
4132 | struct packet_type *pt; | |
4133 | struct list_head *nxt; | |
4134 | int hash; | |
4135 | ||
4136 | ++*pos; | |
4137 | if (v == SEQ_START_TOKEN) | |
4138 | return ptype_get_idx(0); | |
4139 | ||
4140 | pt = v; | |
4141 | nxt = pt->list.next; | |
4142 | if (pt->type == htons(ETH_P_ALL)) { | |
4143 | if (nxt != &ptype_all) | |
4144 | goto found; | |
4145 | hash = 0; | |
4146 | nxt = ptype_base[0].next; | |
4147 | } else | |
82d8a867 | 4148 | hash = ntohs(pt->type) & PTYPE_HASH_MASK; |
0e1256ff SH |
4149 | |
4150 | while (nxt == &ptype_base[hash]) { | |
82d8a867 | 4151 | if (++hash >= PTYPE_HASH_SIZE) |
0e1256ff SH |
4152 | return NULL; |
4153 | nxt = ptype_base[hash].next; | |
4154 | } | |
4155 | found: | |
4156 | return list_entry(nxt, struct packet_type, list); | |
4157 | } | |
4158 | ||
4159 | static void ptype_seq_stop(struct seq_file *seq, void *v) | |
72348a42 | 4160 | __releases(RCU) |
0e1256ff SH |
4161 | { |
4162 | rcu_read_unlock(); | |
4163 | } | |
4164 | ||
0e1256ff SH |
4165 | static int ptype_seq_show(struct seq_file *seq, void *v) |
4166 | { | |
4167 | struct packet_type *pt = v; | |
4168 | ||
4169 | if (v == SEQ_START_TOKEN) | |
4170 | seq_puts(seq, "Type Device Function\n"); | |
c346dca1 | 4171 | else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) { |
0e1256ff SH |
4172 | if (pt->type == htons(ETH_P_ALL)) |
4173 | seq_puts(seq, "ALL "); | |
4174 | else | |
4175 | seq_printf(seq, "%04x", ntohs(pt->type)); | |
4176 | ||
908cd2da AD |
4177 | seq_printf(seq, " %-8s %pF\n", |
4178 | pt->dev ? pt->dev->name : "", pt->func); | |
0e1256ff SH |
4179 | } |
4180 | ||
4181 | return 0; | |
4182 | } | |
4183 | ||
4184 | static const struct seq_operations ptype_seq_ops = { | |
4185 | .start = ptype_seq_start, | |
4186 | .next = ptype_seq_next, | |
4187 | .stop = ptype_seq_stop, | |
4188 | .show = ptype_seq_show, | |
4189 | }; | |
4190 | ||
4191 | static int ptype_seq_open(struct inode *inode, struct file *file) | |
4192 | { | |
2feb27db PE |
4193 | return seq_open_net(inode, file, &ptype_seq_ops, |
4194 | sizeof(struct seq_net_private)); | |
0e1256ff SH |
4195 | } |
4196 | ||
4197 | static const struct file_operations ptype_seq_fops = { | |
4198 | .owner = THIS_MODULE, | |
4199 | .open = ptype_seq_open, | |
4200 | .read = seq_read, | |
4201 | .llseek = seq_lseek, | |
2feb27db | 4202 | .release = seq_release_net, |
0e1256ff SH |
4203 | }; |
4204 | ||
4205 | ||
4665079c | 4206 | static int __net_init dev_proc_net_init(struct net *net) |
1da177e4 LT |
4207 | { |
4208 | int rc = -ENOMEM; | |
4209 | ||
881d966b | 4210 | if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops)) |
1da177e4 | 4211 | goto out; |
881d966b | 4212 | if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops)) |
1da177e4 | 4213 | goto out_dev; |
881d966b | 4214 | if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops)) |
457c4cbc | 4215 | goto out_softnet; |
0e1256ff | 4216 | |
881d966b | 4217 | if (wext_proc_init(net)) |
457c4cbc | 4218 | goto out_ptype; |
1da177e4 LT |
4219 | rc = 0; |
4220 | out: | |
4221 | return rc; | |
457c4cbc | 4222 | out_ptype: |
881d966b | 4223 | proc_net_remove(net, "ptype"); |
1da177e4 | 4224 | out_softnet: |
881d966b | 4225 | proc_net_remove(net, "softnet_stat"); |
1da177e4 | 4226 | out_dev: |
881d966b | 4227 | proc_net_remove(net, "dev"); |
1da177e4 LT |
4228 | goto out; |
4229 | } | |
881d966b | 4230 | |
4665079c | 4231 | static void __net_exit dev_proc_net_exit(struct net *net) |
881d966b EB |
4232 | { |
4233 | wext_proc_exit(net); | |
4234 | ||
4235 | proc_net_remove(net, "ptype"); | |
4236 | proc_net_remove(net, "softnet_stat"); | |
4237 | proc_net_remove(net, "dev"); | |
4238 | } | |
4239 | ||
022cbae6 | 4240 | static struct pernet_operations __net_initdata dev_proc_ops = { |
881d966b EB |
4241 | .init = dev_proc_net_init, |
4242 | .exit = dev_proc_net_exit, | |
4243 | }; | |
4244 | ||
4245 | static int __init dev_proc_init(void) | |
4246 | { | |
4247 | return register_pernet_subsys(&dev_proc_ops); | |
4248 | } | |
1da177e4 LT |
4249 | #else |
4250 | #define dev_proc_init() 0 | |
4251 | #endif /* CONFIG_PROC_FS */ | |
4252 | ||
4253 | ||
4254 | /** | |
4255 | * netdev_set_master - set up master/slave pair | |
4256 | * @slave: slave device | |
4257 | * @master: new master device | |
4258 | * | |
4259 | * Changes the master device of the slave. Pass %NULL to break the | |
4260 | * bonding. The caller must hold the RTNL semaphore. On a failure | |
4261 | * a negative errno code is returned. On success the reference counts | |
4262 | * are adjusted, %RTM_NEWLINK is sent to the routing socket and the | |
4263 | * function returns zero. | |
4264 | */ | |
4265 | int netdev_set_master(struct net_device *slave, struct net_device *master) | |
4266 | { | |
4267 | struct net_device *old = slave->master; | |
4268 | ||
4269 | ASSERT_RTNL(); | |
4270 | ||
4271 | if (master) { | |
4272 | if (old) | |
4273 | return -EBUSY; | |
4274 | dev_hold(master); | |
4275 | } | |
4276 | ||
4277 | slave->master = master; | |
4ec93edb | 4278 | |
283f2fe8 ED |
4279 | if (old) { |
4280 | synchronize_net(); | |
1da177e4 | 4281 | dev_put(old); |
283f2fe8 | 4282 | } |
1da177e4 LT |
4283 | if (master) |
4284 | slave->flags |= IFF_SLAVE; | |
4285 | else | |
4286 | slave->flags &= ~IFF_SLAVE; | |
4287 | ||
4288 | rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE); | |
4289 | return 0; | |
4290 | } | |
d1b19dff | 4291 | EXPORT_SYMBOL(netdev_set_master); |
1da177e4 | 4292 | |
b6c40d68 PM |
4293 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
4294 | { | |
d314774c SH |
4295 | const struct net_device_ops *ops = dev->netdev_ops; |
4296 | ||
4297 | if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags) | |
4298 | ops->ndo_change_rx_flags(dev, flags); | |
b6c40d68 PM |
4299 | } |
4300 | ||
dad9b335 | 4301 | static int __dev_set_promiscuity(struct net_device *dev, int inc) |
1da177e4 LT |
4302 | { |
4303 | unsigned short old_flags = dev->flags; | |
8192b0c4 DH |
4304 | uid_t uid; |
4305 | gid_t gid; | |
1da177e4 | 4306 | |
24023451 PM |
4307 | ASSERT_RTNL(); |
4308 | ||
dad9b335 WC |
4309 | dev->flags |= IFF_PROMISC; |
4310 | dev->promiscuity += inc; | |
4311 | if (dev->promiscuity == 0) { | |
4312 | /* | |
4313 | * Avoid overflow. | |
4314 | * If inc causes overflow, untouch promisc and return error. | |
4315 | */ | |
4316 | if (inc < 0) | |
4317 | dev->flags &= ~IFF_PROMISC; | |
4318 | else { | |
4319 | dev->promiscuity -= inc; | |
4320 | printk(KERN_WARNING "%s: promiscuity touches roof, " | |
4321 | "set promiscuity failed, promiscuity feature " | |
4322 | "of device might be broken.\n", dev->name); | |
4323 | return -EOVERFLOW; | |
4324 | } | |
4325 | } | |
52609c0b | 4326 | if (dev->flags != old_flags) { |
1da177e4 LT |
4327 | printk(KERN_INFO "device %s %s promiscuous mode\n", |
4328 | dev->name, (dev->flags & IFF_PROMISC) ? "entered" : | |
4ec93edb | 4329 | "left"); |
8192b0c4 DH |
4330 | if (audit_enabled) { |
4331 | current_uid_gid(&uid, &gid); | |
7759db82 KHK |
4332 | audit_log(current->audit_context, GFP_ATOMIC, |
4333 | AUDIT_ANOM_PROMISCUOUS, | |
4334 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
4335 | dev->name, (dev->flags & IFF_PROMISC), | |
4336 | (old_flags & IFF_PROMISC), | |
4337 | audit_get_loginuid(current), | |
8192b0c4 | 4338 | uid, gid, |
7759db82 | 4339 | audit_get_sessionid(current)); |
8192b0c4 | 4340 | } |
24023451 | 4341 | |
b6c40d68 | 4342 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 4343 | } |
dad9b335 | 4344 | return 0; |
1da177e4 LT |
4345 | } |
4346 | ||
4417da66 PM |
4347 | /** |
4348 | * dev_set_promiscuity - update promiscuity count on a device | |
4349 | * @dev: device | |
4350 | * @inc: modifier | |
4351 | * | |
4352 | * Add or remove promiscuity from a device. While the count in the device | |
4353 | * remains above zero the interface remains promiscuous. Once it hits zero | |
4354 | * the device reverts back to normal filtering operation. A negative inc | |
4355 | * value is used to drop promiscuity on the device. | |
dad9b335 | 4356 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 4357 | */ |
dad9b335 | 4358 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 PM |
4359 | { |
4360 | unsigned short old_flags = dev->flags; | |
dad9b335 | 4361 | int err; |
4417da66 | 4362 | |
dad9b335 | 4363 | err = __dev_set_promiscuity(dev, inc); |
4b5a698e | 4364 | if (err < 0) |
dad9b335 | 4365 | return err; |
4417da66 PM |
4366 | if (dev->flags != old_flags) |
4367 | dev_set_rx_mode(dev); | |
dad9b335 | 4368 | return err; |
4417da66 | 4369 | } |
d1b19dff | 4370 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 4371 | |
1da177e4 LT |
4372 | /** |
4373 | * dev_set_allmulti - update allmulti count on a device | |
4374 | * @dev: device | |
4375 | * @inc: modifier | |
4376 | * | |
4377 | * Add or remove reception of all multicast frames to a device. While the | |
4378 | * count in the device remains above zero the interface remains listening | |
4379 | * to all interfaces. Once it hits zero the device reverts back to normal | |
4380 | * filtering operation. A negative @inc value is used to drop the counter | |
4381 | * when releasing a resource needing all multicasts. | |
dad9b335 | 4382 | * Return 0 if successful or a negative errno code on error. |
1da177e4 LT |
4383 | */ |
4384 | ||
dad9b335 | 4385 | int dev_set_allmulti(struct net_device *dev, int inc) |
1da177e4 LT |
4386 | { |
4387 | unsigned short old_flags = dev->flags; | |
4388 | ||
24023451 PM |
4389 | ASSERT_RTNL(); |
4390 | ||
1da177e4 | 4391 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
4392 | dev->allmulti += inc; |
4393 | if (dev->allmulti == 0) { | |
4394 | /* | |
4395 | * Avoid overflow. | |
4396 | * If inc causes overflow, untouch allmulti and return error. | |
4397 | */ | |
4398 | if (inc < 0) | |
4399 | dev->flags &= ~IFF_ALLMULTI; | |
4400 | else { | |
4401 | dev->allmulti -= inc; | |
4402 | printk(KERN_WARNING "%s: allmulti touches roof, " | |
4403 | "set allmulti failed, allmulti feature of " | |
4404 | "device might be broken.\n", dev->name); | |
4405 | return -EOVERFLOW; | |
4406 | } | |
4407 | } | |
24023451 | 4408 | if (dev->flags ^ old_flags) { |
b6c40d68 | 4409 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 4410 | dev_set_rx_mode(dev); |
24023451 | 4411 | } |
dad9b335 | 4412 | return 0; |
4417da66 | 4413 | } |
d1b19dff | 4414 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
4415 | |
4416 | /* | |
4417 | * Upload unicast and multicast address lists to device and | |
4418 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 4419 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
4420 | * are present. |
4421 | */ | |
4422 | void __dev_set_rx_mode(struct net_device *dev) | |
4423 | { | |
d314774c SH |
4424 | const struct net_device_ops *ops = dev->netdev_ops; |
4425 | ||
4417da66 PM |
4426 | /* dev_open will call this function so the list will stay sane. */ |
4427 | if (!(dev->flags&IFF_UP)) | |
4428 | return; | |
4429 | ||
4430 | if (!netif_device_present(dev)) | |
40b77c94 | 4431 | return; |
4417da66 | 4432 | |
d314774c SH |
4433 | if (ops->ndo_set_rx_mode) |
4434 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
4435 | else { |
4436 | /* Unicast addresses changes may only happen under the rtnl, | |
4437 | * therefore calling __dev_set_promiscuity here is safe. | |
4438 | */ | |
32e7bfc4 | 4439 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
4417da66 PM |
4440 | __dev_set_promiscuity(dev, 1); |
4441 | dev->uc_promisc = 1; | |
32e7bfc4 | 4442 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
4417da66 PM |
4443 | __dev_set_promiscuity(dev, -1); |
4444 | dev->uc_promisc = 0; | |
4445 | } | |
4446 | ||
d314774c SH |
4447 | if (ops->ndo_set_multicast_list) |
4448 | ops->ndo_set_multicast_list(dev); | |
4417da66 PM |
4449 | } |
4450 | } | |
4451 | ||
4452 | void dev_set_rx_mode(struct net_device *dev) | |
4453 | { | |
b9e40857 | 4454 | netif_addr_lock_bh(dev); |
4417da66 | 4455 | __dev_set_rx_mode(dev); |
b9e40857 | 4456 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
4457 | } |
4458 | ||
f0db275a SH |
4459 | /** |
4460 | * dev_get_flags - get flags reported to userspace | |
4461 | * @dev: device | |
4462 | * | |
4463 | * Get the combination of flag bits exported through APIs to userspace. | |
4464 | */ | |
1da177e4 LT |
4465 | unsigned dev_get_flags(const struct net_device *dev) |
4466 | { | |
4467 | unsigned flags; | |
4468 | ||
4469 | flags = (dev->flags & ~(IFF_PROMISC | | |
4470 | IFF_ALLMULTI | | |
b00055aa SR |
4471 | IFF_RUNNING | |
4472 | IFF_LOWER_UP | | |
4473 | IFF_DORMANT)) | | |
1da177e4 LT |
4474 | (dev->gflags & (IFF_PROMISC | |
4475 | IFF_ALLMULTI)); | |
4476 | ||
b00055aa SR |
4477 | if (netif_running(dev)) { |
4478 | if (netif_oper_up(dev)) | |
4479 | flags |= IFF_RUNNING; | |
4480 | if (netif_carrier_ok(dev)) | |
4481 | flags |= IFF_LOWER_UP; | |
4482 | if (netif_dormant(dev)) | |
4483 | flags |= IFF_DORMANT; | |
4484 | } | |
1da177e4 LT |
4485 | |
4486 | return flags; | |
4487 | } | |
d1b19dff | 4488 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 4489 | |
bd380811 | 4490 | int __dev_change_flags(struct net_device *dev, unsigned int flags) |
1da177e4 | 4491 | { |
1da177e4 | 4492 | int old_flags = dev->flags; |
bd380811 | 4493 | int ret; |
1da177e4 | 4494 | |
24023451 PM |
4495 | ASSERT_RTNL(); |
4496 | ||
1da177e4 LT |
4497 | /* |
4498 | * Set the flags on our device. | |
4499 | */ | |
4500 | ||
4501 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
4502 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
4503 | IFF_AUTOMEDIA)) | | |
4504 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
4505 | IFF_ALLMULTI)); | |
4506 | ||
4507 | /* | |
4508 | * Load in the correct multicast list now the flags have changed. | |
4509 | */ | |
4510 | ||
b6c40d68 PM |
4511 | if ((old_flags ^ flags) & IFF_MULTICAST) |
4512 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 4513 | |
4417da66 | 4514 | dev_set_rx_mode(dev); |
1da177e4 LT |
4515 | |
4516 | /* | |
4517 | * Have we downed the interface. We handle IFF_UP ourselves | |
4518 | * according to user attempts to set it, rather than blindly | |
4519 | * setting it. | |
4520 | */ | |
4521 | ||
4522 | ret = 0; | |
4523 | if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */ | |
bd380811 | 4524 | ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev); |
1da177e4 LT |
4525 | |
4526 | if (!ret) | |
4417da66 | 4527 | dev_set_rx_mode(dev); |
1da177e4 LT |
4528 | } |
4529 | ||
1da177e4 | 4530 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff ED |
4531 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
4532 | ||
1da177e4 LT |
4533 | dev->gflags ^= IFF_PROMISC; |
4534 | dev_set_promiscuity(dev, inc); | |
4535 | } | |
4536 | ||
4537 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
4538 | is important. Some (broken) drivers set IFF_PROMISC, when | |
4539 | IFF_ALLMULTI is requested not asking us and not reporting. | |
4540 | */ | |
4541 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
4542 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
4543 | ||
1da177e4 LT |
4544 | dev->gflags ^= IFF_ALLMULTI; |
4545 | dev_set_allmulti(dev, inc); | |
4546 | } | |
4547 | ||
bd380811 PM |
4548 | return ret; |
4549 | } | |
4550 | ||
4551 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags) | |
4552 | { | |
4553 | unsigned int changes = dev->flags ^ old_flags; | |
4554 | ||
4555 | if (changes & IFF_UP) { | |
4556 | if (dev->flags & IFF_UP) | |
4557 | call_netdevice_notifiers(NETDEV_UP, dev); | |
4558 | else | |
4559 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
4560 | } | |
4561 | ||
4562 | if (dev->flags & IFF_UP && | |
4563 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) | |
4564 | call_netdevice_notifiers(NETDEV_CHANGE, dev); | |
4565 | } | |
4566 | ||
4567 | /** | |
4568 | * dev_change_flags - change device settings | |
4569 | * @dev: device | |
4570 | * @flags: device state flags | |
4571 | * | |
4572 | * Change settings on device based state flags. The flags are | |
4573 | * in the userspace exported format. | |
4574 | */ | |
4575 | int dev_change_flags(struct net_device *dev, unsigned flags) | |
4576 | { | |
4577 | int ret, changes; | |
4578 | int old_flags = dev->flags; | |
4579 | ||
4580 | ret = __dev_change_flags(dev, flags); | |
4581 | if (ret < 0) | |
4582 | return ret; | |
4583 | ||
4584 | changes = old_flags ^ dev->flags; | |
7c355f53 TG |
4585 | if (changes) |
4586 | rtmsg_ifinfo(RTM_NEWLINK, dev, changes); | |
1da177e4 | 4587 | |
bd380811 | 4588 | __dev_notify_flags(dev, old_flags); |
1da177e4 LT |
4589 | return ret; |
4590 | } | |
d1b19dff | 4591 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 4592 | |
f0db275a SH |
4593 | /** |
4594 | * dev_set_mtu - Change maximum transfer unit | |
4595 | * @dev: device | |
4596 | * @new_mtu: new transfer unit | |
4597 | * | |
4598 | * Change the maximum transfer size of the network device. | |
4599 | */ | |
1da177e4 LT |
4600 | int dev_set_mtu(struct net_device *dev, int new_mtu) |
4601 | { | |
d314774c | 4602 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
4603 | int err; |
4604 | ||
4605 | if (new_mtu == dev->mtu) | |
4606 | return 0; | |
4607 | ||
4608 | /* MTU must be positive. */ | |
4609 | if (new_mtu < 0) | |
4610 | return -EINVAL; | |
4611 | ||
4612 | if (!netif_device_present(dev)) | |
4613 | return -ENODEV; | |
4614 | ||
4615 | err = 0; | |
d314774c SH |
4616 | if (ops->ndo_change_mtu) |
4617 | err = ops->ndo_change_mtu(dev, new_mtu); | |
1da177e4 LT |
4618 | else |
4619 | dev->mtu = new_mtu; | |
d314774c | 4620 | |
1da177e4 | 4621 | if (!err && dev->flags & IFF_UP) |
056925ab | 4622 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); |
1da177e4 LT |
4623 | return err; |
4624 | } | |
d1b19dff | 4625 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 4626 | |
cbda10fa VD |
4627 | /** |
4628 | * dev_set_group - Change group this device belongs to | |
4629 | * @dev: device | |
4630 | * @new_group: group this device should belong to | |
4631 | */ | |
4632 | void dev_set_group(struct net_device *dev, int new_group) | |
4633 | { | |
4634 | dev->group = new_group; | |
4635 | } | |
4636 | EXPORT_SYMBOL(dev_set_group); | |
4637 | ||
f0db275a SH |
4638 | /** |
4639 | * dev_set_mac_address - Change Media Access Control Address | |
4640 | * @dev: device | |
4641 | * @sa: new address | |
4642 | * | |
4643 | * Change the hardware (MAC) address of the device | |
4644 | */ | |
1da177e4 LT |
4645 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) |
4646 | { | |
d314774c | 4647 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
4648 | int err; |
4649 | ||
d314774c | 4650 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
4651 | return -EOPNOTSUPP; |
4652 | if (sa->sa_family != dev->type) | |
4653 | return -EINVAL; | |
4654 | if (!netif_device_present(dev)) | |
4655 | return -ENODEV; | |
d314774c | 4656 | err = ops->ndo_set_mac_address(dev, sa); |
1da177e4 | 4657 | if (!err) |
056925ab | 4658 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
1da177e4 LT |
4659 | return err; |
4660 | } | |
d1b19dff | 4661 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 LT |
4662 | |
4663 | /* | |
3710becf | 4664 | * Perform the SIOCxIFxxx calls, inside rcu_read_lock() |
1da177e4 | 4665 | */ |
14e3e079 | 4666 | static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd) |
1da177e4 LT |
4667 | { |
4668 | int err; | |
3710becf | 4669 | struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name); |
1da177e4 LT |
4670 | |
4671 | if (!dev) | |
4672 | return -ENODEV; | |
4673 | ||
4674 | switch (cmd) { | |
d1b19dff ED |
4675 | case SIOCGIFFLAGS: /* Get interface flags */ |
4676 | ifr->ifr_flags = (short) dev_get_flags(dev); | |
4677 | return 0; | |
1da177e4 | 4678 | |
d1b19dff ED |
4679 | case SIOCGIFMETRIC: /* Get the metric on the interface |
4680 | (currently unused) */ | |
4681 | ifr->ifr_metric = 0; | |
4682 | return 0; | |
1da177e4 | 4683 | |
d1b19dff ED |
4684 | case SIOCGIFMTU: /* Get the MTU of a device */ |
4685 | ifr->ifr_mtu = dev->mtu; | |
4686 | return 0; | |
1da177e4 | 4687 | |
d1b19dff ED |
4688 | case SIOCGIFHWADDR: |
4689 | if (!dev->addr_len) | |
4690 | memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data); | |
4691 | else | |
4692 | memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr, | |
4693 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
4694 | ifr->ifr_hwaddr.sa_family = dev->type; | |
4695 | return 0; | |
1da177e4 | 4696 | |
d1b19dff ED |
4697 | case SIOCGIFSLAVE: |
4698 | err = -EINVAL; | |
4699 | break; | |
14e3e079 | 4700 | |
d1b19dff ED |
4701 | case SIOCGIFMAP: |
4702 | ifr->ifr_map.mem_start = dev->mem_start; | |
4703 | ifr->ifr_map.mem_end = dev->mem_end; | |
4704 | ifr->ifr_map.base_addr = dev->base_addr; | |
4705 | ifr->ifr_map.irq = dev->irq; | |
4706 | ifr->ifr_map.dma = dev->dma; | |
4707 | ifr->ifr_map.port = dev->if_port; | |
4708 | return 0; | |
14e3e079 | 4709 | |
d1b19dff ED |
4710 | case SIOCGIFINDEX: |
4711 | ifr->ifr_ifindex = dev->ifindex; | |
4712 | return 0; | |
14e3e079 | 4713 | |
d1b19dff ED |
4714 | case SIOCGIFTXQLEN: |
4715 | ifr->ifr_qlen = dev->tx_queue_len; | |
4716 | return 0; | |
14e3e079 | 4717 | |
d1b19dff ED |
4718 | default: |
4719 | /* dev_ioctl() should ensure this case | |
4720 | * is never reached | |
4721 | */ | |
4722 | WARN_ON(1); | |
4723 | err = -EINVAL; | |
4724 | break; | |
14e3e079 JG |
4725 | |
4726 | } | |
4727 | return err; | |
4728 | } | |
4729 | ||
4730 | /* | |
4731 | * Perform the SIOCxIFxxx calls, inside rtnl_lock() | |
4732 | */ | |
4733 | static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd) | |
4734 | { | |
4735 | int err; | |
4736 | struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name); | |
5f2f6da7 | 4737 | const struct net_device_ops *ops; |
14e3e079 JG |
4738 | |
4739 | if (!dev) | |
4740 | return -ENODEV; | |
4741 | ||
5f2f6da7 JP |
4742 | ops = dev->netdev_ops; |
4743 | ||
14e3e079 | 4744 | switch (cmd) { |
d1b19dff ED |
4745 | case SIOCSIFFLAGS: /* Set interface flags */ |
4746 | return dev_change_flags(dev, ifr->ifr_flags); | |
14e3e079 | 4747 | |
d1b19dff ED |
4748 | case SIOCSIFMETRIC: /* Set the metric on the interface |
4749 | (currently unused) */ | |
4750 | return -EOPNOTSUPP; | |
14e3e079 | 4751 | |
d1b19dff ED |
4752 | case SIOCSIFMTU: /* Set the MTU of a device */ |
4753 | return dev_set_mtu(dev, ifr->ifr_mtu); | |
1da177e4 | 4754 | |
d1b19dff ED |
4755 | case SIOCSIFHWADDR: |
4756 | return dev_set_mac_address(dev, &ifr->ifr_hwaddr); | |
1da177e4 | 4757 | |
d1b19dff ED |
4758 | case SIOCSIFHWBROADCAST: |
4759 | if (ifr->ifr_hwaddr.sa_family != dev->type) | |
4760 | return -EINVAL; | |
4761 | memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data, | |
4762 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
4763 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); | |
4764 | return 0; | |
1da177e4 | 4765 | |
d1b19dff ED |
4766 | case SIOCSIFMAP: |
4767 | if (ops->ndo_set_config) { | |
1da177e4 LT |
4768 | if (!netif_device_present(dev)) |
4769 | return -ENODEV; | |
d1b19dff ED |
4770 | return ops->ndo_set_config(dev, &ifr->ifr_map); |
4771 | } | |
4772 | return -EOPNOTSUPP; | |
1da177e4 | 4773 | |
d1b19dff ED |
4774 | case SIOCADDMULTI: |
4775 | if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) || | |
4776 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) | |
4777 | return -EINVAL; | |
4778 | if (!netif_device_present(dev)) | |
4779 | return -ENODEV; | |
22bedad3 | 4780 | return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data); |
d1b19dff ED |
4781 | |
4782 | case SIOCDELMULTI: | |
4783 | if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) || | |
4784 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) | |
4785 | return -EINVAL; | |
4786 | if (!netif_device_present(dev)) | |
4787 | return -ENODEV; | |
22bedad3 | 4788 | return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data); |
1da177e4 | 4789 | |
d1b19dff ED |
4790 | case SIOCSIFTXQLEN: |
4791 | if (ifr->ifr_qlen < 0) | |
4792 | return -EINVAL; | |
4793 | dev->tx_queue_len = ifr->ifr_qlen; | |
4794 | return 0; | |
1da177e4 | 4795 | |
d1b19dff ED |
4796 | case SIOCSIFNAME: |
4797 | ifr->ifr_newname[IFNAMSIZ-1] = '\0'; | |
4798 | return dev_change_name(dev, ifr->ifr_newname); | |
1da177e4 | 4799 | |
d1b19dff ED |
4800 | /* |
4801 | * Unknown or private ioctl | |
4802 | */ | |
4803 | default: | |
4804 | if ((cmd >= SIOCDEVPRIVATE && | |
4805 | cmd <= SIOCDEVPRIVATE + 15) || | |
4806 | cmd == SIOCBONDENSLAVE || | |
4807 | cmd == SIOCBONDRELEASE || | |
4808 | cmd == SIOCBONDSETHWADDR || | |
4809 | cmd == SIOCBONDSLAVEINFOQUERY || | |
4810 | cmd == SIOCBONDINFOQUERY || | |
4811 | cmd == SIOCBONDCHANGEACTIVE || | |
4812 | cmd == SIOCGMIIPHY || | |
4813 | cmd == SIOCGMIIREG || | |
4814 | cmd == SIOCSMIIREG || | |
4815 | cmd == SIOCBRADDIF || | |
4816 | cmd == SIOCBRDELIF || | |
4817 | cmd == SIOCSHWTSTAMP || | |
4818 | cmd == SIOCWANDEV) { | |
4819 | err = -EOPNOTSUPP; | |
4820 | if (ops->ndo_do_ioctl) { | |
4821 | if (netif_device_present(dev)) | |
4822 | err = ops->ndo_do_ioctl(dev, ifr, cmd); | |
4823 | else | |
4824 | err = -ENODEV; | |
4825 | } | |
4826 | } else | |
4827 | err = -EINVAL; | |
1da177e4 LT |
4828 | |
4829 | } | |
4830 | return err; | |
4831 | } | |
4832 | ||
4833 | /* | |
4834 | * This function handles all "interface"-type I/O control requests. The actual | |
4835 | * 'doing' part of this is dev_ifsioc above. | |
4836 | */ | |
4837 | ||
4838 | /** | |
4839 | * dev_ioctl - network device ioctl | |
c4ea43c5 | 4840 | * @net: the applicable net namespace |
1da177e4 LT |
4841 | * @cmd: command to issue |
4842 | * @arg: pointer to a struct ifreq in user space | |
4843 | * | |
4844 | * Issue ioctl functions to devices. This is normally called by the | |
4845 | * user space syscall interfaces but can sometimes be useful for | |
4846 | * other purposes. The return value is the return from the syscall if | |
4847 | * positive or a negative errno code on error. | |
4848 | */ | |
4849 | ||
881d966b | 4850 | int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg) |
1da177e4 LT |
4851 | { |
4852 | struct ifreq ifr; | |
4853 | int ret; | |
4854 | char *colon; | |
4855 | ||
4856 | /* One special case: SIOCGIFCONF takes ifconf argument | |
4857 | and requires shared lock, because it sleeps writing | |
4858 | to user space. | |
4859 | */ | |
4860 | ||
4861 | if (cmd == SIOCGIFCONF) { | |
6756ae4b | 4862 | rtnl_lock(); |
881d966b | 4863 | ret = dev_ifconf(net, (char __user *) arg); |
6756ae4b | 4864 | rtnl_unlock(); |
1da177e4 LT |
4865 | return ret; |
4866 | } | |
4867 | if (cmd == SIOCGIFNAME) | |
881d966b | 4868 | return dev_ifname(net, (struct ifreq __user *)arg); |
1da177e4 LT |
4869 | |
4870 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
4871 | return -EFAULT; | |
4872 | ||
4873 | ifr.ifr_name[IFNAMSIZ-1] = 0; | |
4874 | ||
4875 | colon = strchr(ifr.ifr_name, ':'); | |
4876 | if (colon) | |
4877 | *colon = 0; | |
4878 | ||
4879 | /* | |
4880 | * See which interface the caller is talking about. | |
4881 | */ | |
4882 | ||
4883 | switch (cmd) { | |
d1b19dff ED |
4884 | /* |
4885 | * These ioctl calls: | |
4886 | * - can be done by all. | |
4887 | * - atomic and do not require locking. | |
4888 | * - return a value | |
4889 | */ | |
4890 | case SIOCGIFFLAGS: | |
4891 | case SIOCGIFMETRIC: | |
4892 | case SIOCGIFMTU: | |
4893 | case SIOCGIFHWADDR: | |
4894 | case SIOCGIFSLAVE: | |
4895 | case SIOCGIFMAP: | |
4896 | case SIOCGIFINDEX: | |
4897 | case SIOCGIFTXQLEN: | |
4898 | dev_load(net, ifr.ifr_name); | |
3710becf | 4899 | rcu_read_lock(); |
d1b19dff | 4900 | ret = dev_ifsioc_locked(net, &ifr, cmd); |
3710becf | 4901 | rcu_read_unlock(); |
d1b19dff ED |
4902 | if (!ret) { |
4903 | if (colon) | |
4904 | *colon = ':'; | |
4905 | if (copy_to_user(arg, &ifr, | |
4906 | sizeof(struct ifreq))) | |
4907 | ret = -EFAULT; | |
4908 | } | |
4909 | return ret; | |
1da177e4 | 4910 | |
d1b19dff ED |
4911 | case SIOCETHTOOL: |
4912 | dev_load(net, ifr.ifr_name); | |
4913 | rtnl_lock(); | |
4914 | ret = dev_ethtool(net, &ifr); | |
4915 | rtnl_unlock(); | |
4916 | if (!ret) { | |
4917 | if (colon) | |
4918 | *colon = ':'; | |
4919 | if (copy_to_user(arg, &ifr, | |
4920 | sizeof(struct ifreq))) | |
4921 | ret = -EFAULT; | |
4922 | } | |
4923 | return ret; | |
1da177e4 | 4924 | |
d1b19dff ED |
4925 | /* |
4926 | * These ioctl calls: | |
4927 | * - require superuser power. | |
4928 | * - require strict serialization. | |
4929 | * - return a value | |
4930 | */ | |
4931 | case SIOCGMIIPHY: | |
4932 | case SIOCGMIIREG: | |
4933 | case SIOCSIFNAME: | |
4934 | if (!capable(CAP_NET_ADMIN)) | |
4935 | return -EPERM; | |
4936 | dev_load(net, ifr.ifr_name); | |
4937 | rtnl_lock(); | |
4938 | ret = dev_ifsioc(net, &ifr, cmd); | |
4939 | rtnl_unlock(); | |
4940 | if (!ret) { | |
4941 | if (colon) | |
4942 | *colon = ':'; | |
4943 | if (copy_to_user(arg, &ifr, | |
4944 | sizeof(struct ifreq))) | |
4945 | ret = -EFAULT; | |
4946 | } | |
4947 | return ret; | |
1da177e4 | 4948 | |
d1b19dff ED |
4949 | /* |
4950 | * These ioctl calls: | |
4951 | * - require superuser power. | |
4952 | * - require strict serialization. | |
4953 | * - do not return a value | |
4954 | */ | |
4955 | case SIOCSIFFLAGS: | |
4956 | case SIOCSIFMETRIC: | |
4957 | case SIOCSIFMTU: | |
4958 | case SIOCSIFMAP: | |
4959 | case SIOCSIFHWADDR: | |
4960 | case SIOCSIFSLAVE: | |
4961 | case SIOCADDMULTI: | |
4962 | case SIOCDELMULTI: | |
4963 | case SIOCSIFHWBROADCAST: | |
4964 | case SIOCSIFTXQLEN: | |
4965 | case SIOCSMIIREG: | |
4966 | case SIOCBONDENSLAVE: | |
4967 | case SIOCBONDRELEASE: | |
4968 | case SIOCBONDSETHWADDR: | |
4969 | case SIOCBONDCHANGEACTIVE: | |
4970 | case SIOCBRADDIF: | |
4971 | case SIOCBRDELIF: | |
4972 | case SIOCSHWTSTAMP: | |
4973 | if (!capable(CAP_NET_ADMIN)) | |
4974 | return -EPERM; | |
4975 | /* fall through */ | |
4976 | case SIOCBONDSLAVEINFOQUERY: | |
4977 | case SIOCBONDINFOQUERY: | |
4978 | dev_load(net, ifr.ifr_name); | |
4979 | rtnl_lock(); | |
4980 | ret = dev_ifsioc(net, &ifr, cmd); | |
4981 | rtnl_unlock(); | |
4982 | return ret; | |
4983 | ||
4984 | case SIOCGIFMEM: | |
4985 | /* Get the per device memory space. We can add this but | |
4986 | * currently do not support it */ | |
4987 | case SIOCSIFMEM: | |
4988 | /* Set the per device memory buffer space. | |
4989 | * Not applicable in our case */ | |
4990 | case SIOCSIFLINK: | |
4991 | return -EINVAL; | |
4992 | ||
4993 | /* | |
4994 | * Unknown or private ioctl. | |
4995 | */ | |
4996 | default: | |
4997 | if (cmd == SIOCWANDEV || | |
4998 | (cmd >= SIOCDEVPRIVATE && | |
4999 | cmd <= SIOCDEVPRIVATE + 15)) { | |
881d966b | 5000 | dev_load(net, ifr.ifr_name); |
1da177e4 | 5001 | rtnl_lock(); |
881d966b | 5002 | ret = dev_ifsioc(net, &ifr, cmd); |
1da177e4 | 5003 | rtnl_unlock(); |
d1b19dff ED |
5004 | if (!ret && copy_to_user(arg, &ifr, |
5005 | sizeof(struct ifreq))) | |
5006 | ret = -EFAULT; | |
1da177e4 | 5007 | return ret; |
d1b19dff ED |
5008 | } |
5009 | /* Take care of Wireless Extensions */ | |
5010 | if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) | |
5011 | return wext_handle_ioctl(net, &ifr, cmd, arg); | |
5012 | return -EINVAL; | |
1da177e4 LT |
5013 | } |
5014 | } | |
5015 | ||
5016 | ||
5017 | /** | |
5018 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 5019 | * @net: the applicable net namespace |
1da177e4 LT |
5020 | * |
5021 | * Returns a suitable unique value for a new device interface | |
5022 | * number. The caller must hold the rtnl semaphore or the | |
5023 | * dev_base_lock to be sure it remains unique. | |
5024 | */ | |
881d966b | 5025 | static int dev_new_index(struct net *net) |
1da177e4 LT |
5026 | { |
5027 | static int ifindex; | |
5028 | for (;;) { | |
5029 | if (++ifindex <= 0) | |
5030 | ifindex = 1; | |
881d966b | 5031 | if (!__dev_get_by_index(net, ifindex)) |
1da177e4 LT |
5032 | return ifindex; |
5033 | } | |
5034 | } | |
5035 | ||
1da177e4 | 5036 | /* Delayed registration/unregisteration */ |
3b5b34fd | 5037 | static LIST_HEAD(net_todo_list); |
1da177e4 | 5038 | |
6f05f629 | 5039 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 5040 | { |
1da177e4 | 5041 | list_add_tail(&dev->todo_list, &net_todo_list); |
1da177e4 LT |
5042 | } |
5043 | ||
9b5e383c | 5044 | static void rollback_registered_many(struct list_head *head) |
93ee31f1 | 5045 | { |
e93737b0 | 5046 | struct net_device *dev, *tmp; |
9b5e383c | 5047 | |
93ee31f1 DL |
5048 | BUG_ON(dev_boot_phase); |
5049 | ASSERT_RTNL(); | |
5050 | ||
e93737b0 | 5051 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
9b5e383c | 5052 | /* Some devices call without registering |
e93737b0 KK |
5053 | * for initialization unwind. Remove those |
5054 | * devices and proceed with the remaining. | |
9b5e383c ED |
5055 | */ |
5056 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
5057 | pr_debug("unregister_netdevice: device %s/%p never " | |
5058 | "was registered\n", dev->name, dev); | |
93ee31f1 | 5059 | |
9b5e383c | 5060 | WARN_ON(1); |
e93737b0 KK |
5061 | list_del(&dev->unreg_list); |
5062 | continue; | |
9b5e383c | 5063 | } |
93ee31f1 | 5064 | |
9b5e383c | 5065 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
44345724 | 5066 | } |
93ee31f1 | 5067 | |
44345724 OP |
5068 | /* If device is running, close it first. */ |
5069 | dev_close_many(head); | |
93ee31f1 | 5070 | |
44345724 | 5071 | list_for_each_entry(dev, head, unreg_list) { |
9b5e383c ED |
5072 | /* And unlink it from device chain. */ |
5073 | unlist_netdevice(dev); | |
93ee31f1 | 5074 | |
9b5e383c ED |
5075 | dev->reg_state = NETREG_UNREGISTERING; |
5076 | } | |
93ee31f1 DL |
5077 | |
5078 | synchronize_net(); | |
5079 | ||
9b5e383c ED |
5080 | list_for_each_entry(dev, head, unreg_list) { |
5081 | /* Shutdown queueing discipline. */ | |
5082 | dev_shutdown(dev); | |
93ee31f1 DL |
5083 | |
5084 | ||
9b5e383c ED |
5085 | /* Notify protocols, that we are about to destroy |
5086 | this device. They should clean all the things. | |
5087 | */ | |
5088 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
93ee31f1 | 5089 | |
a2835763 PM |
5090 | if (!dev->rtnl_link_ops || |
5091 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
5092 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U); | |
5093 | ||
9b5e383c ED |
5094 | /* |
5095 | * Flush the unicast and multicast chains | |
5096 | */ | |
a748ee24 | 5097 | dev_uc_flush(dev); |
22bedad3 | 5098 | dev_mc_flush(dev); |
93ee31f1 | 5099 | |
9b5e383c ED |
5100 | if (dev->netdev_ops->ndo_uninit) |
5101 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 | 5102 | |
9b5e383c ED |
5103 | /* Notifier chain MUST detach us from master device. */ |
5104 | WARN_ON(dev->master); | |
93ee31f1 | 5105 | |
9b5e383c ED |
5106 | /* Remove entries from kobject tree */ |
5107 | netdev_unregister_kobject(dev); | |
5108 | } | |
93ee31f1 | 5109 | |
a5ee1551 | 5110 | /* Process any work delayed until the end of the batch */ |
e5e26d75 | 5111 | dev = list_first_entry(head, struct net_device, unreg_list); |
a5ee1551 | 5112 | call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev); |
93ee31f1 | 5113 | |
ef885afb | 5114 | rcu_barrier(); |
395264d5 | 5115 | |
a5ee1551 | 5116 | list_for_each_entry(dev, head, unreg_list) |
9b5e383c ED |
5117 | dev_put(dev); |
5118 | } | |
5119 | ||
5120 | static void rollback_registered(struct net_device *dev) | |
5121 | { | |
5122 | LIST_HEAD(single); | |
5123 | ||
5124 | list_add(&dev->unreg_list, &single); | |
5125 | rollback_registered_many(&single); | |
93ee31f1 DL |
5126 | } |
5127 | ||
b63365a2 HX |
5128 | unsigned long netdev_fix_features(unsigned long features, const char *name) |
5129 | { | |
5130 | /* Fix illegal SG+CSUM combinations. */ | |
5131 | if ((features & NETIF_F_SG) && | |
5132 | !(features & NETIF_F_ALL_CSUM)) { | |
5133 | if (name) | |
5134 | printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no " | |
5135 | "checksum feature.\n", name); | |
5136 | features &= ~NETIF_F_SG; | |
5137 | } | |
5138 | ||
5139 | /* TSO requires that SG is present as well. */ | |
5140 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) { | |
5141 | if (name) | |
5142 | printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no " | |
5143 | "SG feature.\n", name); | |
5144 | features &= ~NETIF_F_TSO; | |
5145 | } | |
5146 | ||
5147 | if (features & NETIF_F_UFO) { | |
79032644 MM |
5148 | /* maybe split UFO into V4 and V6? */ |
5149 | if (!((features & NETIF_F_GEN_CSUM) || | |
5150 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM)) | |
5151 | == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
b63365a2 HX |
5152 | if (name) |
5153 | printk(KERN_ERR "%s: Dropping NETIF_F_UFO " | |
79032644 | 5154 | "since no checksum offload features.\n", |
b63365a2 HX |
5155 | name); |
5156 | features &= ~NETIF_F_UFO; | |
5157 | } | |
5158 | ||
5159 | if (!(features & NETIF_F_SG)) { | |
5160 | if (name) | |
5161 | printk(KERN_ERR "%s: Dropping NETIF_F_UFO " | |
5162 | "since no NETIF_F_SG feature.\n", name); | |
5163 | features &= ~NETIF_F_UFO; | |
5164 | } | |
5165 | } | |
5166 | ||
5167 | return features; | |
5168 | } | |
5169 | EXPORT_SYMBOL(netdev_fix_features); | |
5170 | ||
fc4a7489 PM |
5171 | /** |
5172 | * netif_stacked_transfer_operstate - transfer operstate | |
5173 | * @rootdev: the root or lower level device to transfer state from | |
5174 | * @dev: the device to transfer operstate to | |
5175 | * | |
5176 | * Transfer operational state from root to device. This is normally | |
5177 | * called when a stacking relationship exists between the root | |
5178 | * device and the device(a leaf device). | |
5179 | */ | |
5180 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
5181 | struct net_device *dev) | |
5182 | { | |
5183 | if (rootdev->operstate == IF_OPER_DORMANT) | |
5184 | netif_dormant_on(dev); | |
5185 | else | |
5186 | netif_dormant_off(dev); | |
5187 | ||
5188 | if (netif_carrier_ok(rootdev)) { | |
5189 | if (!netif_carrier_ok(dev)) | |
5190 | netif_carrier_on(dev); | |
5191 | } else { | |
5192 | if (netif_carrier_ok(dev)) | |
5193 | netif_carrier_off(dev); | |
5194 | } | |
5195 | } | |
5196 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
5197 | ||
bf264145 | 5198 | #ifdef CONFIG_RPS |
1b4bf461 ED |
5199 | static int netif_alloc_rx_queues(struct net_device *dev) |
5200 | { | |
1b4bf461 | 5201 | unsigned int i, count = dev->num_rx_queues; |
bd25fa7b | 5202 | struct netdev_rx_queue *rx; |
1b4bf461 | 5203 | |
bd25fa7b | 5204 | BUG_ON(count < 1); |
1b4bf461 | 5205 | |
bd25fa7b TH |
5206 | rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL); |
5207 | if (!rx) { | |
5208 | pr_err("netdev: Unable to allocate %u rx queues.\n", count); | |
5209 | return -ENOMEM; | |
1b4bf461 | 5210 | } |
bd25fa7b TH |
5211 | dev->_rx = rx; |
5212 | ||
bd25fa7b | 5213 | for (i = 0; i < count; i++) |
fe822240 | 5214 | rx[i].dev = dev; |
1b4bf461 ED |
5215 | return 0; |
5216 | } | |
bf264145 | 5217 | #endif |
1b4bf461 | 5218 | |
aa942104 CG |
5219 | static void netdev_init_one_queue(struct net_device *dev, |
5220 | struct netdev_queue *queue, void *_unused) | |
5221 | { | |
5222 | /* Initialize queue lock */ | |
5223 | spin_lock_init(&queue->_xmit_lock); | |
5224 | netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); | |
5225 | queue->xmit_lock_owner = -1; | |
b236da69 | 5226 | netdev_queue_numa_node_write(queue, NUMA_NO_NODE); |
aa942104 CG |
5227 | queue->dev = dev; |
5228 | } | |
5229 | ||
e6484930 TH |
5230 | static int netif_alloc_netdev_queues(struct net_device *dev) |
5231 | { | |
5232 | unsigned int count = dev->num_tx_queues; | |
5233 | struct netdev_queue *tx; | |
5234 | ||
5235 | BUG_ON(count < 1); | |
5236 | ||
5237 | tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL); | |
5238 | if (!tx) { | |
5239 | pr_err("netdev: Unable to allocate %u tx queues.\n", | |
5240 | count); | |
5241 | return -ENOMEM; | |
5242 | } | |
5243 | dev->_tx = tx; | |
1d24eb48 | 5244 | |
e6484930 TH |
5245 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); |
5246 | spin_lock_init(&dev->tx_global_lock); | |
aa942104 CG |
5247 | |
5248 | return 0; | |
e6484930 TH |
5249 | } |
5250 | ||
1da177e4 LT |
5251 | /** |
5252 | * register_netdevice - register a network device | |
5253 | * @dev: device to register | |
5254 | * | |
5255 | * Take a completed network device structure and add it to the kernel | |
5256 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
5257 | * chain. 0 is returned on success. A negative errno code is returned | |
5258 | * on a failure to set up the device, or if the name is a duplicate. | |
5259 | * | |
5260 | * Callers must hold the rtnl semaphore. You may want | |
5261 | * register_netdev() instead of this. | |
5262 | * | |
5263 | * BUGS: | |
5264 | * The locking appears insufficient to guarantee two parallel registers | |
5265 | * will not get the same name. | |
5266 | */ | |
5267 | ||
5268 | int register_netdevice(struct net_device *dev) | |
5269 | { | |
1da177e4 | 5270 | int ret; |
d314774c | 5271 | struct net *net = dev_net(dev); |
1da177e4 LT |
5272 | |
5273 | BUG_ON(dev_boot_phase); | |
5274 | ASSERT_RTNL(); | |
5275 | ||
b17a7c17 SH |
5276 | might_sleep(); |
5277 | ||
1da177e4 LT |
5278 | /* When net_device's are persistent, this will be fatal. */ |
5279 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 5280 | BUG_ON(!net); |
1da177e4 | 5281 | |
f1f28aa3 | 5282 | spin_lock_init(&dev->addr_list_lock); |
cf508b12 | 5283 | netdev_set_addr_lockdep_class(dev); |
1da177e4 | 5284 | |
1da177e4 LT |
5285 | dev->iflink = -1; |
5286 | ||
5287 | /* Init, if this function is available */ | |
d314774c SH |
5288 | if (dev->netdev_ops->ndo_init) { |
5289 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
5290 | if (ret) { |
5291 | if (ret > 0) | |
5292 | ret = -EIO; | |
90833aa4 | 5293 | goto out; |
1da177e4 LT |
5294 | } |
5295 | } | |
4ec93edb | 5296 | |
8ce6cebc | 5297 | ret = dev_get_valid_name(dev, dev->name, 0); |
d9031024 | 5298 | if (ret) |
7ce1b0ed | 5299 | goto err_uninit; |
1da177e4 | 5300 | |
881d966b | 5301 | dev->ifindex = dev_new_index(net); |
1da177e4 LT |
5302 | if (dev->iflink == -1) |
5303 | dev->iflink = dev->ifindex; | |
5304 | ||
d212f87b SH |
5305 | /* Fix illegal checksum combinations */ |
5306 | if ((dev->features & NETIF_F_HW_CSUM) && | |
5307 | (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
5308 | printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n", | |
5309 | dev->name); | |
5310 | dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); | |
5311 | } | |
5312 | ||
5313 | if ((dev->features & NETIF_F_NO_CSUM) && | |
5314 | (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
5315 | printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n", | |
5316 | dev->name); | |
5317 | dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM); | |
5318 | } | |
5319 | ||
b63365a2 | 5320 | dev->features = netdev_fix_features(dev->features, dev->name); |
1da177e4 | 5321 | |
e5a4a72d LB |
5322 | /* Enable software GSO if SG is supported. */ |
5323 | if (dev->features & NETIF_F_SG) | |
5324 | dev->features |= NETIF_F_GSO; | |
5325 | ||
c5256c51 ED |
5326 | /* Enable GRO and NETIF_F_HIGHDMA for vlans by default, |
5327 | * vlan_dev_init() will do the dev->features check, so these features | |
5328 | * are enabled only if supported by underlying device. | |
16c3ea78 | 5329 | */ |
c5256c51 | 5330 | dev->vlan_features |= (NETIF_F_GRO | NETIF_F_HIGHDMA); |
16c3ea78 | 5331 | |
7ffbe3fd JB |
5332 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); |
5333 | ret = notifier_to_errno(ret); | |
5334 | if (ret) | |
5335 | goto err_uninit; | |
5336 | ||
8b41d188 | 5337 | ret = netdev_register_kobject(dev); |
b17a7c17 | 5338 | if (ret) |
7ce1b0ed | 5339 | goto err_uninit; |
b17a7c17 SH |
5340 | dev->reg_state = NETREG_REGISTERED; |
5341 | ||
1da177e4 LT |
5342 | /* |
5343 | * Default initial state at registry is that the | |
5344 | * device is present. | |
5345 | */ | |
5346 | ||
5347 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
5348 | ||
1da177e4 | 5349 | dev_init_scheduler(dev); |
1da177e4 | 5350 | dev_hold(dev); |
ce286d32 | 5351 | list_netdevice(dev); |
1da177e4 LT |
5352 | |
5353 | /* Notify protocols, that a new device appeared. */ | |
056925ab | 5354 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 5355 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
5356 | if (ret) { |
5357 | rollback_registered(dev); | |
5358 | dev->reg_state = NETREG_UNREGISTERED; | |
5359 | } | |
d90a909e EB |
5360 | /* |
5361 | * Prevent userspace races by waiting until the network | |
5362 | * device is fully setup before sending notifications. | |
5363 | */ | |
a2835763 PM |
5364 | if (!dev->rtnl_link_ops || |
5365 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
5366 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U); | |
1da177e4 LT |
5367 | |
5368 | out: | |
5369 | return ret; | |
7ce1b0ed HX |
5370 | |
5371 | err_uninit: | |
d314774c SH |
5372 | if (dev->netdev_ops->ndo_uninit) |
5373 | dev->netdev_ops->ndo_uninit(dev); | |
7ce1b0ed | 5374 | goto out; |
1da177e4 | 5375 | } |
d1b19dff | 5376 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 5377 | |
937f1ba5 BH |
5378 | /** |
5379 | * init_dummy_netdev - init a dummy network device for NAPI | |
5380 | * @dev: device to init | |
5381 | * | |
5382 | * This takes a network device structure and initialize the minimum | |
5383 | * amount of fields so it can be used to schedule NAPI polls without | |
5384 | * registering a full blown interface. This is to be used by drivers | |
5385 | * that need to tie several hardware interfaces to a single NAPI | |
5386 | * poll scheduler due to HW limitations. | |
5387 | */ | |
5388 | int init_dummy_netdev(struct net_device *dev) | |
5389 | { | |
5390 | /* Clear everything. Note we don't initialize spinlocks | |
5391 | * are they aren't supposed to be taken by any of the | |
5392 | * NAPI code and this dummy netdev is supposed to be | |
5393 | * only ever used for NAPI polls | |
5394 | */ | |
5395 | memset(dev, 0, sizeof(struct net_device)); | |
5396 | ||
5397 | /* make sure we BUG if trying to hit standard | |
5398 | * register/unregister code path | |
5399 | */ | |
5400 | dev->reg_state = NETREG_DUMMY; | |
5401 | ||
937f1ba5 BH |
5402 | /* NAPI wants this */ |
5403 | INIT_LIST_HEAD(&dev->napi_list); | |
5404 | ||
5405 | /* a dummy interface is started by default */ | |
5406 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
5407 | set_bit(__LINK_STATE_START, &dev->state); | |
5408 | ||
29b4433d ED |
5409 | /* Note : We dont allocate pcpu_refcnt for dummy devices, |
5410 | * because users of this 'device' dont need to change | |
5411 | * its refcount. | |
5412 | */ | |
5413 | ||
937f1ba5 BH |
5414 | return 0; |
5415 | } | |
5416 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
5417 | ||
5418 | ||
1da177e4 LT |
5419 | /** |
5420 | * register_netdev - register a network device | |
5421 | * @dev: device to register | |
5422 | * | |
5423 | * Take a completed network device structure and add it to the kernel | |
5424 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
5425 | * chain. 0 is returned on success. A negative errno code is returned | |
5426 | * on a failure to set up the device, or if the name is a duplicate. | |
5427 | * | |
38b4da38 | 5428 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
5429 | * and expands the device name if you passed a format string to |
5430 | * alloc_netdev. | |
5431 | */ | |
5432 | int register_netdev(struct net_device *dev) | |
5433 | { | |
5434 | int err; | |
5435 | ||
5436 | rtnl_lock(); | |
5437 | ||
5438 | /* | |
5439 | * If the name is a format string the caller wants us to do a | |
5440 | * name allocation. | |
5441 | */ | |
5442 | if (strchr(dev->name, '%')) { | |
5443 | err = dev_alloc_name(dev, dev->name); | |
5444 | if (err < 0) | |
5445 | goto out; | |
5446 | } | |
4ec93edb | 5447 | |
1da177e4 LT |
5448 | err = register_netdevice(dev); |
5449 | out: | |
5450 | rtnl_unlock(); | |
5451 | return err; | |
5452 | } | |
5453 | EXPORT_SYMBOL(register_netdev); | |
5454 | ||
29b4433d ED |
5455 | int netdev_refcnt_read(const struct net_device *dev) |
5456 | { | |
5457 | int i, refcnt = 0; | |
5458 | ||
5459 | for_each_possible_cpu(i) | |
5460 | refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); | |
5461 | return refcnt; | |
5462 | } | |
5463 | EXPORT_SYMBOL(netdev_refcnt_read); | |
5464 | ||
1da177e4 LT |
5465 | /* |
5466 | * netdev_wait_allrefs - wait until all references are gone. | |
5467 | * | |
5468 | * This is called when unregistering network devices. | |
5469 | * | |
5470 | * Any protocol or device that holds a reference should register | |
5471 | * for netdevice notification, and cleanup and put back the | |
5472 | * reference if they receive an UNREGISTER event. | |
5473 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 5474 | * call dev_put. |
1da177e4 LT |
5475 | */ |
5476 | static void netdev_wait_allrefs(struct net_device *dev) | |
5477 | { | |
5478 | unsigned long rebroadcast_time, warning_time; | |
29b4433d | 5479 | int refcnt; |
1da177e4 | 5480 | |
e014debe ED |
5481 | linkwatch_forget_dev(dev); |
5482 | ||
1da177e4 | 5483 | rebroadcast_time = warning_time = jiffies; |
29b4433d ED |
5484 | refcnt = netdev_refcnt_read(dev); |
5485 | ||
5486 | while (refcnt != 0) { | |
1da177e4 | 5487 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { |
6756ae4b | 5488 | rtnl_lock(); |
1da177e4 LT |
5489 | |
5490 | /* Rebroadcast unregister notification */ | |
056925ab | 5491 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
a5ee1551 | 5492 | /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users |
395264d5 | 5493 | * should have already handle it the first time */ |
1da177e4 LT |
5494 | |
5495 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, | |
5496 | &dev->state)) { | |
5497 | /* We must not have linkwatch events | |
5498 | * pending on unregister. If this | |
5499 | * happens, we simply run the queue | |
5500 | * unscheduled, resulting in a noop | |
5501 | * for this device. | |
5502 | */ | |
5503 | linkwatch_run_queue(); | |
5504 | } | |
5505 | ||
6756ae4b | 5506 | __rtnl_unlock(); |
1da177e4 LT |
5507 | |
5508 | rebroadcast_time = jiffies; | |
5509 | } | |
5510 | ||
5511 | msleep(250); | |
5512 | ||
29b4433d ED |
5513 | refcnt = netdev_refcnt_read(dev); |
5514 | ||
1da177e4 LT |
5515 | if (time_after(jiffies, warning_time + 10 * HZ)) { |
5516 | printk(KERN_EMERG "unregister_netdevice: " | |
5517 | "waiting for %s to become free. Usage " | |
5518 | "count = %d\n", | |
29b4433d | 5519 | dev->name, refcnt); |
1da177e4 LT |
5520 | warning_time = jiffies; |
5521 | } | |
5522 | } | |
5523 | } | |
5524 | ||
5525 | /* The sequence is: | |
5526 | * | |
5527 | * rtnl_lock(); | |
5528 | * ... | |
5529 | * register_netdevice(x1); | |
5530 | * register_netdevice(x2); | |
5531 | * ... | |
5532 | * unregister_netdevice(y1); | |
5533 | * unregister_netdevice(y2); | |
5534 | * ... | |
5535 | * rtnl_unlock(); | |
5536 | * free_netdev(y1); | |
5537 | * free_netdev(y2); | |
5538 | * | |
58ec3b4d | 5539 | * We are invoked by rtnl_unlock(). |
1da177e4 | 5540 | * This allows us to deal with problems: |
b17a7c17 | 5541 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
5542 | * without deadlocking with linkwatch via keventd. |
5543 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
5544 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
5545 | * |
5546 | * We must not return until all unregister events added during | |
5547 | * the interval the lock was held have been completed. | |
1da177e4 | 5548 | */ |
1da177e4 LT |
5549 | void netdev_run_todo(void) |
5550 | { | |
626ab0e6 | 5551 | struct list_head list; |
1da177e4 | 5552 | |
1da177e4 | 5553 | /* Snapshot list, allow later requests */ |
626ab0e6 | 5554 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
5555 | |
5556 | __rtnl_unlock(); | |
626ab0e6 | 5557 | |
1da177e4 LT |
5558 | while (!list_empty(&list)) { |
5559 | struct net_device *dev | |
e5e26d75 | 5560 | = list_first_entry(&list, struct net_device, todo_list); |
1da177e4 LT |
5561 | list_del(&dev->todo_list); |
5562 | ||
b17a7c17 SH |
5563 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
5564 | printk(KERN_ERR "network todo '%s' but state %d\n", | |
5565 | dev->name, dev->reg_state); | |
5566 | dump_stack(); | |
5567 | continue; | |
5568 | } | |
1da177e4 | 5569 | |
b17a7c17 | 5570 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 5571 | |
152102c7 | 5572 | on_each_cpu(flush_backlog, dev, 1); |
6e583ce5 | 5573 | |
b17a7c17 | 5574 | netdev_wait_allrefs(dev); |
1da177e4 | 5575 | |
b17a7c17 | 5576 | /* paranoia */ |
29b4433d | 5577 | BUG_ON(netdev_refcnt_read(dev)); |
95ae6b22 | 5578 | WARN_ON(rcu_dereference_raw(dev->ip_ptr)); |
198caeca | 5579 | WARN_ON(rcu_dereference_raw(dev->ip6_ptr)); |
547b792c | 5580 | WARN_ON(dev->dn_ptr); |
1da177e4 | 5581 | |
b17a7c17 SH |
5582 | if (dev->destructor) |
5583 | dev->destructor(dev); | |
9093bbb2 SH |
5584 | |
5585 | /* Free network device */ | |
5586 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 5587 | } |
1da177e4 LT |
5588 | } |
5589 | ||
3cfde79c BH |
5590 | /* Convert net_device_stats to rtnl_link_stats64. They have the same |
5591 | * fields in the same order, with only the type differing. | |
5592 | */ | |
5593 | static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, | |
5594 | const struct net_device_stats *netdev_stats) | |
5595 | { | |
5596 | #if BITS_PER_LONG == 64 | |
5597 | BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats)); | |
5598 | memcpy(stats64, netdev_stats, sizeof(*stats64)); | |
5599 | #else | |
5600 | size_t i, n = sizeof(*stats64) / sizeof(u64); | |
5601 | const unsigned long *src = (const unsigned long *)netdev_stats; | |
5602 | u64 *dst = (u64 *)stats64; | |
5603 | ||
5604 | BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) != | |
5605 | sizeof(*stats64) / sizeof(u64)); | |
5606 | for (i = 0; i < n; i++) | |
5607 | dst[i] = src[i]; | |
5608 | #endif | |
5609 | } | |
5610 | ||
eeda3fd6 SH |
5611 | /** |
5612 | * dev_get_stats - get network device statistics | |
5613 | * @dev: device to get statistics from | |
28172739 | 5614 | * @storage: place to store stats |
eeda3fd6 | 5615 | * |
d7753516 BH |
5616 | * Get network statistics from device. Return @storage. |
5617 | * The device driver may provide its own method by setting | |
5618 | * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; | |
5619 | * otherwise the internal statistics structure is used. | |
eeda3fd6 | 5620 | */ |
d7753516 BH |
5621 | struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, |
5622 | struct rtnl_link_stats64 *storage) | |
7004bf25 | 5623 | { |
eeda3fd6 SH |
5624 | const struct net_device_ops *ops = dev->netdev_ops; |
5625 | ||
28172739 ED |
5626 | if (ops->ndo_get_stats64) { |
5627 | memset(storage, 0, sizeof(*storage)); | |
caf586e5 ED |
5628 | ops->ndo_get_stats64(dev, storage); |
5629 | } else if (ops->ndo_get_stats) { | |
3cfde79c | 5630 | netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); |
caf586e5 ED |
5631 | } else { |
5632 | netdev_stats_to_stats64(storage, &dev->stats); | |
28172739 | 5633 | } |
caf586e5 | 5634 | storage->rx_dropped += atomic_long_read(&dev->rx_dropped); |
28172739 | 5635 | return storage; |
c45d286e | 5636 | } |
eeda3fd6 | 5637 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 5638 | |
24824a09 | 5639 | struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) |
dc2b4847 | 5640 | { |
24824a09 | 5641 | struct netdev_queue *queue = dev_ingress_queue(dev); |
dc2b4847 | 5642 | |
24824a09 ED |
5643 | #ifdef CONFIG_NET_CLS_ACT |
5644 | if (queue) | |
5645 | return queue; | |
5646 | queue = kzalloc(sizeof(*queue), GFP_KERNEL); | |
5647 | if (!queue) | |
5648 | return NULL; | |
5649 | netdev_init_one_queue(dev, queue, NULL); | |
24824a09 ED |
5650 | queue->qdisc = &noop_qdisc; |
5651 | queue->qdisc_sleeping = &noop_qdisc; | |
5652 | rcu_assign_pointer(dev->ingress_queue, queue); | |
5653 | #endif | |
5654 | return queue; | |
bb949fbd DM |
5655 | } |
5656 | ||
1da177e4 | 5657 | /** |
36909ea4 | 5658 | * alloc_netdev_mqs - allocate network device |
1da177e4 LT |
5659 | * @sizeof_priv: size of private data to allocate space for |
5660 | * @name: device name format string | |
5661 | * @setup: callback to initialize device | |
36909ea4 TH |
5662 | * @txqs: the number of TX subqueues to allocate |
5663 | * @rxqs: the number of RX subqueues to allocate | |
1da177e4 LT |
5664 | * |
5665 | * Allocates a struct net_device with private data area for driver use | |
f25f4e44 | 5666 | * and performs basic initialization. Also allocates subquue structs |
36909ea4 | 5667 | * for each queue on the device. |
1da177e4 | 5668 | */ |
36909ea4 TH |
5669 | struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, |
5670 | void (*setup)(struct net_device *), | |
5671 | unsigned int txqs, unsigned int rxqs) | |
1da177e4 | 5672 | { |
1da177e4 | 5673 | struct net_device *dev; |
7943986c | 5674 | size_t alloc_size; |
1ce8e7b5 | 5675 | struct net_device *p; |
1da177e4 | 5676 | |
b6fe17d6 SH |
5677 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
5678 | ||
36909ea4 | 5679 | if (txqs < 1) { |
55513fb4 TH |
5680 | pr_err("alloc_netdev: Unable to allocate device " |
5681 | "with zero queues.\n"); | |
5682 | return NULL; | |
5683 | } | |
5684 | ||
36909ea4 TH |
5685 | #ifdef CONFIG_RPS |
5686 | if (rxqs < 1) { | |
5687 | pr_err("alloc_netdev: Unable to allocate device " | |
5688 | "with zero RX queues.\n"); | |
5689 | return NULL; | |
5690 | } | |
5691 | #endif | |
5692 | ||
fd2ea0a7 | 5693 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
5694 | if (sizeof_priv) { |
5695 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 5696 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
5697 | alloc_size += sizeof_priv; |
5698 | } | |
5699 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 5700 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 5701 | |
31380de9 | 5702 | p = kzalloc(alloc_size, GFP_KERNEL); |
1da177e4 | 5703 | if (!p) { |
b6fe17d6 | 5704 | printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n"); |
1da177e4 LT |
5705 | return NULL; |
5706 | } | |
1da177e4 | 5707 | |
1ce8e7b5 | 5708 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 5709 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc | 5710 | |
29b4433d ED |
5711 | dev->pcpu_refcnt = alloc_percpu(int); |
5712 | if (!dev->pcpu_refcnt) | |
e6484930 | 5713 | goto free_p; |
ab9c73cc | 5714 | |
ab9c73cc | 5715 | if (dev_addr_init(dev)) |
29b4433d | 5716 | goto free_pcpu; |
ab9c73cc | 5717 | |
22bedad3 | 5718 | dev_mc_init(dev); |
a748ee24 | 5719 | dev_uc_init(dev); |
ccffad25 | 5720 | |
c346dca1 | 5721 | dev_net_set(dev, &init_net); |
1da177e4 | 5722 | |
36909ea4 TH |
5723 | dev->num_tx_queues = txqs; |
5724 | dev->real_num_tx_queues = txqs; | |
ed9af2e8 TH |
5725 | if (netif_alloc_netdev_queues(dev)) |
5726 | goto free_pcpu; | |
e8a0464c | 5727 | |
df334545 | 5728 | #ifdef CONFIG_RPS |
36909ea4 TH |
5729 | dev->num_rx_queues = rxqs; |
5730 | dev->real_num_rx_queues = rxqs; | |
fe822240 TH |
5731 | if (netif_alloc_rx_queues(dev)) |
5732 | goto free_pcpu; | |
df334545 | 5733 | #endif |
0a9627f2 | 5734 | |
82cc1a7a | 5735 | dev->gso_max_size = GSO_MAX_SIZE; |
1da177e4 | 5736 | |
15682bc4 PWJ |
5737 | INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list); |
5738 | dev->ethtool_ntuple_list.count = 0; | |
d565b0a1 | 5739 | INIT_LIST_HEAD(&dev->napi_list); |
9fdce099 | 5740 | INIT_LIST_HEAD(&dev->unreg_list); |
e014debe | 5741 | INIT_LIST_HEAD(&dev->link_watch_list); |
93f154b5 | 5742 | dev->priv_flags = IFF_XMIT_DST_RELEASE; |
1da177e4 LT |
5743 | setup(dev); |
5744 | strcpy(dev->name, name); | |
cbda10fa | 5745 | dev->group = INIT_NETDEV_GROUP; |
1da177e4 | 5746 | return dev; |
ab9c73cc | 5747 | |
29b4433d ED |
5748 | free_pcpu: |
5749 | free_percpu(dev->pcpu_refcnt); | |
ed9af2e8 | 5750 | kfree(dev->_tx); |
fe822240 TH |
5751 | #ifdef CONFIG_RPS |
5752 | kfree(dev->_rx); | |
5753 | #endif | |
5754 | ||
ab9c73cc JP |
5755 | free_p: |
5756 | kfree(p); | |
5757 | return NULL; | |
1da177e4 | 5758 | } |
36909ea4 | 5759 | EXPORT_SYMBOL(alloc_netdev_mqs); |
1da177e4 LT |
5760 | |
5761 | /** | |
5762 | * free_netdev - free network device | |
5763 | * @dev: device | |
5764 | * | |
4ec93edb YH |
5765 | * This function does the last stage of destroying an allocated device |
5766 | * interface. The reference to the device object is released. | |
1da177e4 LT |
5767 | * If this is the last reference then it will be freed. |
5768 | */ | |
5769 | void free_netdev(struct net_device *dev) | |
5770 | { | |
d565b0a1 HX |
5771 | struct napi_struct *p, *n; |
5772 | ||
f3005d7f DL |
5773 | release_net(dev_net(dev)); |
5774 | ||
e8a0464c | 5775 | kfree(dev->_tx); |
fe822240 TH |
5776 | #ifdef CONFIG_RPS |
5777 | kfree(dev->_rx); | |
5778 | #endif | |
e8a0464c | 5779 | |
24824a09 ED |
5780 | kfree(rcu_dereference_raw(dev->ingress_queue)); |
5781 | ||
f001fde5 JP |
5782 | /* Flush device addresses */ |
5783 | dev_addr_flush(dev); | |
5784 | ||
15682bc4 PWJ |
5785 | /* Clear ethtool n-tuple list */ |
5786 | ethtool_ntuple_flush(dev); | |
5787 | ||
d565b0a1 HX |
5788 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
5789 | netif_napi_del(p); | |
5790 | ||
29b4433d ED |
5791 | free_percpu(dev->pcpu_refcnt); |
5792 | dev->pcpu_refcnt = NULL; | |
5793 | ||
3041a069 | 5794 | /* Compatibility with error handling in drivers */ |
1da177e4 LT |
5795 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
5796 | kfree((char *)dev - dev->padded); | |
5797 | return; | |
5798 | } | |
5799 | ||
5800 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
5801 | dev->reg_state = NETREG_RELEASED; | |
5802 | ||
43cb76d9 GKH |
5803 | /* will free via device release */ |
5804 | put_device(&dev->dev); | |
1da177e4 | 5805 | } |
d1b19dff | 5806 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 5807 | |
f0db275a SH |
5808 | /** |
5809 | * synchronize_net - Synchronize with packet receive processing | |
5810 | * | |
5811 | * Wait for packets currently being received to be done. | |
5812 | * Does not block later packets from starting. | |
5813 | */ | |
4ec93edb | 5814 | void synchronize_net(void) |
1da177e4 LT |
5815 | { |
5816 | might_sleep(); | |
fbd568a3 | 5817 | synchronize_rcu(); |
1da177e4 | 5818 | } |
d1b19dff | 5819 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
5820 | |
5821 | /** | |
44a0873d | 5822 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 5823 | * @dev: device |
44a0873d | 5824 | * @head: list |
6ebfbc06 | 5825 | * |
1da177e4 | 5826 | * This function shuts down a device interface and removes it |
d59b54b1 | 5827 | * from the kernel tables. |
44a0873d | 5828 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
5829 | * |
5830 | * Callers must hold the rtnl semaphore. You may want | |
5831 | * unregister_netdev() instead of this. | |
5832 | */ | |
5833 | ||
44a0873d | 5834 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 5835 | { |
a6620712 HX |
5836 | ASSERT_RTNL(); |
5837 | ||
44a0873d | 5838 | if (head) { |
9fdce099 | 5839 | list_move_tail(&dev->unreg_list, head); |
44a0873d ED |
5840 | } else { |
5841 | rollback_registered(dev); | |
5842 | /* Finish processing unregister after unlock */ | |
5843 | net_set_todo(dev); | |
5844 | } | |
1da177e4 | 5845 | } |
44a0873d | 5846 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 5847 | |
9b5e383c ED |
5848 | /** |
5849 | * unregister_netdevice_many - unregister many devices | |
5850 | * @head: list of devices | |
9b5e383c ED |
5851 | */ |
5852 | void unregister_netdevice_many(struct list_head *head) | |
5853 | { | |
5854 | struct net_device *dev; | |
5855 | ||
5856 | if (!list_empty(head)) { | |
5857 | rollback_registered_many(head); | |
5858 | list_for_each_entry(dev, head, unreg_list) | |
5859 | net_set_todo(dev); | |
5860 | } | |
5861 | } | |
63c8099d | 5862 | EXPORT_SYMBOL(unregister_netdevice_many); |
9b5e383c | 5863 | |
1da177e4 LT |
5864 | /** |
5865 | * unregister_netdev - remove device from the kernel | |
5866 | * @dev: device | |
5867 | * | |
5868 | * This function shuts down a device interface and removes it | |
d59b54b1 | 5869 | * from the kernel tables. |
1da177e4 LT |
5870 | * |
5871 | * This is just a wrapper for unregister_netdevice that takes | |
5872 | * the rtnl semaphore. In general you want to use this and not | |
5873 | * unregister_netdevice. | |
5874 | */ | |
5875 | void unregister_netdev(struct net_device *dev) | |
5876 | { | |
5877 | rtnl_lock(); | |
5878 | unregister_netdevice(dev); | |
5879 | rtnl_unlock(); | |
5880 | } | |
1da177e4 LT |
5881 | EXPORT_SYMBOL(unregister_netdev); |
5882 | ||
ce286d32 EB |
5883 | /** |
5884 | * dev_change_net_namespace - move device to different nethost namespace | |
5885 | * @dev: device | |
5886 | * @net: network namespace | |
5887 | * @pat: If not NULL name pattern to try if the current device name | |
5888 | * is already taken in the destination network namespace. | |
5889 | * | |
5890 | * This function shuts down a device interface and moves it | |
5891 | * to a new network namespace. On success 0 is returned, on | |
5892 | * a failure a netagive errno code is returned. | |
5893 | * | |
5894 | * Callers must hold the rtnl semaphore. | |
5895 | */ | |
5896 | ||
5897 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
5898 | { | |
ce286d32 EB |
5899 | int err; |
5900 | ||
5901 | ASSERT_RTNL(); | |
5902 | ||
5903 | /* Don't allow namespace local devices to be moved. */ | |
5904 | err = -EINVAL; | |
5905 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
5906 | goto out; | |
5907 | ||
5908 | /* Ensure the device has been registrered */ | |
5909 | err = -EINVAL; | |
5910 | if (dev->reg_state != NETREG_REGISTERED) | |
5911 | goto out; | |
5912 | ||
5913 | /* Get out if there is nothing todo */ | |
5914 | err = 0; | |
878628fb | 5915 | if (net_eq(dev_net(dev), net)) |
ce286d32 EB |
5916 | goto out; |
5917 | ||
5918 | /* Pick the destination device name, and ensure | |
5919 | * we can use it in the destination network namespace. | |
5920 | */ | |
5921 | err = -EEXIST; | |
d9031024 | 5922 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
5923 | /* We get here if we can't use the current device name */ |
5924 | if (!pat) | |
5925 | goto out; | |
8ce6cebc | 5926 | if (dev_get_valid_name(dev, pat, 1)) |
ce286d32 EB |
5927 | goto out; |
5928 | } | |
5929 | ||
5930 | /* | |
5931 | * And now a mini version of register_netdevice unregister_netdevice. | |
5932 | */ | |
5933 | ||
5934 | /* If device is running close it first. */ | |
9b772652 | 5935 | dev_close(dev); |
ce286d32 EB |
5936 | |
5937 | /* And unlink it from device chain */ | |
5938 | err = -ENODEV; | |
5939 | unlist_netdevice(dev); | |
5940 | ||
5941 | synchronize_net(); | |
5942 | ||
5943 | /* Shutdown queueing discipline. */ | |
5944 | dev_shutdown(dev); | |
5945 | ||
5946 | /* Notify protocols, that we are about to destroy | |
5947 | this device. They should clean all the things. | |
3b27e105 DL |
5948 | |
5949 | Note that dev->reg_state stays at NETREG_REGISTERED. | |
5950 | This is wanted because this way 8021q and macvlan know | |
5951 | the device is just moving and can keep their slaves up. | |
ce286d32 EB |
5952 | */ |
5953 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
a5ee1551 | 5954 | call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev); |
ce286d32 EB |
5955 | |
5956 | /* | |
5957 | * Flush the unicast and multicast chains | |
5958 | */ | |
a748ee24 | 5959 | dev_uc_flush(dev); |
22bedad3 | 5960 | dev_mc_flush(dev); |
ce286d32 EB |
5961 | |
5962 | /* Actually switch the network namespace */ | |
c346dca1 | 5963 | dev_net_set(dev, net); |
ce286d32 | 5964 | |
ce286d32 EB |
5965 | /* If there is an ifindex conflict assign a new one */ |
5966 | if (__dev_get_by_index(net, dev->ifindex)) { | |
5967 | int iflink = (dev->iflink == dev->ifindex); | |
5968 | dev->ifindex = dev_new_index(net); | |
5969 | if (iflink) | |
5970 | dev->iflink = dev->ifindex; | |
5971 | } | |
5972 | ||
8b41d188 | 5973 | /* Fixup kobjects */ |
a1b3f594 | 5974 | err = device_rename(&dev->dev, dev->name); |
8b41d188 | 5975 | WARN_ON(err); |
ce286d32 EB |
5976 | |
5977 | /* Add the device back in the hashes */ | |
5978 | list_netdevice(dev); | |
5979 | ||
5980 | /* Notify protocols, that a new device appeared. */ | |
5981 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
5982 | ||
d90a909e EB |
5983 | /* |
5984 | * Prevent userspace races by waiting until the network | |
5985 | * device is fully setup before sending notifications. | |
5986 | */ | |
5987 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U); | |
5988 | ||
ce286d32 EB |
5989 | synchronize_net(); |
5990 | err = 0; | |
5991 | out: | |
5992 | return err; | |
5993 | } | |
463d0183 | 5994 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 5995 | |
1da177e4 LT |
5996 | static int dev_cpu_callback(struct notifier_block *nfb, |
5997 | unsigned long action, | |
5998 | void *ocpu) | |
5999 | { | |
6000 | struct sk_buff **list_skb; | |
1da177e4 LT |
6001 | struct sk_buff *skb; |
6002 | unsigned int cpu, oldcpu = (unsigned long)ocpu; | |
6003 | struct softnet_data *sd, *oldsd; | |
6004 | ||
8bb78442 | 6005 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
1da177e4 LT |
6006 | return NOTIFY_OK; |
6007 | ||
6008 | local_irq_disable(); | |
6009 | cpu = smp_processor_id(); | |
6010 | sd = &per_cpu(softnet_data, cpu); | |
6011 | oldsd = &per_cpu(softnet_data, oldcpu); | |
6012 | ||
6013 | /* Find end of our completion_queue. */ | |
6014 | list_skb = &sd->completion_queue; | |
6015 | while (*list_skb) | |
6016 | list_skb = &(*list_skb)->next; | |
6017 | /* Append completion queue from offline CPU. */ | |
6018 | *list_skb = oldsd->completion_queue; | |
6019 | oldsd->completion_queue = NULL; | |
6020 | ||
1da177e4 | 6021 | /* Append output queue from offline CPU. */ |
a9cbd588 CG |
6022 | if (oldsd->output_queue) { |
6023 | *sd->output_queue_tailp = oldsd->output_queue; | |
6024 | sd->output_queue_tailp = oldsd->output_queue_tailp; | |
6025 | oldsd->output_queue = NULL; | |
6026 | oldsd->output_queue_tailp = &oldsd->output_queue; | |
6027 | } | |
1da177e4 LT |
6028 | |
6029 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
6030 | local_irq_enable(); | |
6031 | ||
6032 | /* Process offline CPU's input_pkt_queue */ | |
76cc8b13 | 6033 | while ((skb = __skb_dequeue(&oldsd->process_queue))) { |
1da177e4 | 6034 | netif_rx(skb); |
76cc8b13 | 6035 | input_queue_head_incr(oldsd); |
fec5e652 | 6036 | } |
76cc8b13 | 6037 | while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) { |
6e7676c1 | 6038 | netif_rx(skb); |
76cc8b13 TH |
6039 | input_queue_head_incr(oldsd); |
6040 | } | |
1da177e4 LT |
6041 | |
6042 | return NOTIFY_OK; | |
6043 | } | |
1da177e4 LT |
6044 | |
6045 | ||
7f353bf2 | 6046 | /** |
b63365a2 HX |
6047 | * netdev_increment_features - increment feature set by one |
6048 | * @all: current feature set | |
6049 | * @one: new feature set | |
6050 | * @mask: mask feature set | |
7f353bf2 HX |
6051 | * |
6052 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
6053 | * @one to the master device with current feature set @all. Will not |
6054 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 6055 | */ |
b63365a2 HX |
6056 | unsigned long netdev_increment_features(unsigned long all, unsigned long one, |
6057 | unsigned long mask) | |
6058 | { | |
6059 | /* If device needs checksumming, downgrade to it. */ | |
d1b19dff | 6060 | if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM)) |
b63365a2 HX |
6061 | all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM); |
6062 | else if (mask & NETIF_F_ALL_CSUM) { | |
6063 | /* If one device supports v4/v6 checksumming, set for all. */ | |
6064 | if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) && | |
6065 | !(all & NETIF_F_GEN_CSUM)) { | |
6066 | all &= ~NETIF_F_ALL_CSUM; | |
6067 | all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); | |
6068 | } | |
e2a6b852 | 6069 | |
b63365a2 HX |
6070 | /* If one device supports hw checksumming, set for all. */ |
6071 | if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) { | |
6072 | all &= ~NETIF_F_ALL_CSUM; | |
6073 | all |= NETIF_F_HW_CSUM; | |
6074 | } | |
6075 | } | |
7f353bf2 | 6076 | |
b63365a2 | 6077 | one |= NETIF_F_ALL_CSUM; |
7f353bf2 | 6078 | |
b63365a2 | 6079 | one |= all & NETIF_F_ONE_FOR_ALL; |
d9f5950f | 6080 | all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO; |
b63365a2 | 6081 | all |= one & mask & NETIF_F_ONE_FOR_ALL; |
7f353bf2 HX |
6082 | |
6083 | return all; | |
6084 | } | |
b63365a2 | 6085 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 6086 | |
30d97d35 PE |
6087 | static struct hlist_head *netdev_create_hash(void) |
6088 | { | |
6089 | int i; | |
6090 | struct hlist_head *hash; | |
6091 | ||
6092 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
6093 | if (hash != NULL) | |
6094 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
6095 | INIT_HLIST_HEAD(&hash[i]); | |
6096 | ||
6097 | return hash; | |
6098 | } | |
6099 | ||
881d966b | 6100 | /* Initialize per network namespace state */ |
4665079c | 6101 | static int __net_init netdev_init(struct net *net) |
881d966b | 6102 | { |
881d966b | 6103 | INIT_LIST_HEAD(&net->dev_base_head); |
881d966b | 6104 | |
30d97d35 PE |
6105 | net->dev_name_head = netdev_create_hash(); |
6106 | if (net->dev_name_head == NULL) | |
6107 | goto err_name; | |
881d966b | 6108 | |
30d97d35 PE |
6109 | net->dev_index_head = netdev_create_hash(); |
6110 | if (net->dev_index_head == NULL) | |
6111 | goto err_idx; | |
881d966b EB |
6112 | |
6113 | return 0; | |
30d97d35 PE |
6114 | |
6115 | err_idx: | |
6116 | kfree(net->dev_name_head); | |
6117 | err_name: | |
6118 | return -ENOMEM; | |
881d966b EB |
6119 | } |
6120 | ||
f0db275a SH |
6121 | /** |
6122 | * netdev_drivername - network driver for the device | |
6123 | * @dev: network device | |
6124 | * @buffer: buffer for resulting name | |
6125 | * @len: size of buffer | |
6126 | * | |
6127 | * Determine network driver for device. | |
6128 | */ | |
cf04a4c7 | 6129 | char *netdev_drivername(const struct net_device *dev, char *buffer, int len) |
6579e57b | 6130 | { |
cf04a4c7 SH |
6131 | const struct device_driver *driver; |
6132 | const struct device *parent; | |
6579e57b AV |
6133 | |
6134 | if (len <= 0 || !buffer) | |
6135 | return buffer; | |
6136 | buffer[0] = 0; | |
6137 | ||
6138 | parent = dev->dev.parent; | |
6139 | ||
6140 | if (!parent) | |
6141 | return buffer; | |
6142 | ||
6143 | driver = parent->driver; | |
6144 | if (driver && driver->name) | |
6145 | strlcpy(buffer, driver->name, len); | |
6146 | return buffer; | |
6147 | } | |
6148 | ||
256df2f3 JP |
6149 | static int __netdev_printk(const char *level, const struct net_device *dev, |
6150 | struct va_format *vaf) | |
6151 | { | |
6152 | int r; | |
6153 | ||
6154 | if (dev && dev->dev.parent) | |
6155 | r = dev_printk(level, dev->dev.parent, "%s: %pV", | |
6156 | netdev_name(dev), vaf); | |
6157 | else if (dev) | |
6158 | r = printk("%s%s: %pV", level, netdev_name(dev), vaf); | |
6159 | else | |
6160 | r = printk("%s(NULL net_device): %pV", level, vaf); | |
6161 | ||
6162 | return r; | |
6163 | } | |
6164 | ||
6165 | int netdev_printk(const char *level, const struct net_device *dev, | |
6166 | const char *format, ...) | |
6167 | { | |
6168 | struct va_format vaf; | |
6169 | va_list args; | |
6170 | int r; | |
6171 | ||
6172 | va_start(args, format); | |
6173 | ||
6174 | vaf.fmt = format; | |
6175 | vaf.va = &args; | |
6176 | ||
6177 | r = __netdev_printk(level, dev, &vaf); | |
6178 | va_end(args); | |
6179 | ||
6180 | return r; | |
6181 | } | |
6182 | EXPORT_SYMBOL(netdev_printk); | |
6183 | ||
6184 | #define define_netdev_printk_level(func, level) \ | |
6185 | int func(const struct net_device *dev, const char *fmt, ...) \ | |
6186 | { \ | |
6187 | int r; \ | |
6188 | struct va_format vaf; \ | |
6189 | va_list args; \ | |
6190 | \ | |
6191 | va_start(args, fmt); \ | |
6192 | \ | |
6193 | vaf.fmt = fmt; \ | |
6194 | vaf.va = &args; \ | |
6195 | \ | |
6196 | r = __netdev_printk(level, dev, &vaf); \ | |
6197 | va_end(args); \ | |
6198 | \ | |
6199 | return r; \ | |
6200 | } \ | |
6201 | EXPORT_SYMBOL(func); | |
6202 | ||
6203 | define_netdev_printk_level(netdev_emerg, KERN_EMERG); | |
6204 | define_netdev_printk_level(netdev_alert, KERN_ALERT); | |
6205 | define_netdev_printk_level(netdev_crit, KERN_CRIT); | |
6206 | define_netdev_printk_level(netdev_err, KERN_ERR); | |
6207 | define_netdev_printk_level(netdev_warn, KERN_WARNING); | |
6208 | define_netdev_printk_level(netdev_notice, KERN_NOTICE); | |
6209 | define_netdev_printk_level(netdev_info, KERN_INFO); | |
6210 | ||
4665079c | 6211 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
6212 | { |
6213 | kfree(net->dev_name_head); | |
6214 | kfree(net->dev_index_head); | |
6215 | } | |
6216 | ||
022cbae6 | 6217 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
6218 | .init = netdev_init, |
6219 | .exit = netdev_exit, | |
6220 | }; | |
6221 | ||
4665079c | 6222 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 6223 | { |
e008b5fc | 6224 | struct net_device *dev, *aux; |
ce286d32 | 6225 | /* |
e008b5fc | 6226 | * Push all migratable network devices back to the |
ce286d32 EB |
6227 | * initial network namespace |
6228 | */ | |
6229 | rtnl_lock(); | |
e008b5fc | 6230 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 6231 | int err; |
aca51397 | 6232 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
6233 | |
6234 | /* Ignore unmoveable devices (i.e. loopback) */ | |
6235 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
6236 | continue; | |
6237 | ||
e008b5fc EB |
6238 | /* Leave virtual devices for the generic cleanup */ |
6239 | if (dev->rtnl_link_ops) | |
6240 | continue; | |
d0c082ce | 6241 | |
ce286d32 | 6242 | /* Push remaing network devices to init_net */ |
aca51397 PE |
6243 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
6244 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
ce286d32 | 6245 | if (err) { |
aca51397 | 6246 | printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n", |
ce286d32 | 6247 | __func__, dev->name, err); |
aca51397 | 6248 | BUG(); |
ce286d32 EB |
6249 | } |
6250 | } | |
6251 | rtnl_unlock(); | |
6252 | } | |
6253 | ||
04dc7f6b EB |
6254 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
6255 | { | |
6256 | /* At exit all network devices most be removed from a network | |
b595076a | 6257 | * namespace. Do this in the reverse order of registration. |
04dc7f6b EB |
6258 | * Do this across as many network namespaces as possible to |
6259 | * improve batching efficiency. | |
6260 | */ | |
6261 | struct net_device *dev; | |
6262 | struct net *net; | |
6263 | LIST_HEAD(dev_kill_list); | |
6264 | ||
6265 | rtnl_lock(); | |
6266 | list_for_each_entry(net, net_list, exit_list) { | |
6267 | for_each_netdev_reverse(net, dev) { | |
6268 | if (dev->rtnl_link_ops) | |
6269 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); | |
6270 | else | |
6271 | unregister_netdevice_queue(dev, &dev_kill_list); | |
6272 | } | |
6273 | } | |
6274 | unregister_netdevice_many(&dev_kill_list); | |
6275 | rtnl_unlock(); | |
6276 | } | |
6277 | ||
022cbae6 | 6278 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 6279 | .exit = default_device_exit, |
04dc7f6b | 6280 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
6281 | }; |
6282 | ||
1da177e4 LT |
6283 | /* |
6284 | * Initialize the DEV module. At boot time this walks the device list and | |
6285 | * unhooks any devices that fail to initialise (normally hardware not | |
6286 | * present) and leaves us with a valid list of present and active devices. | |
6287 | * | |
6288 | */ | |
6289 | ||
6290 | /* | |
6291 | * This is called single threaded during boot, so no need | |
6292 | * to take the rtnl semaphore. | |
6293 | */ | |
6294 | static int __init net_dev_init(void) | |
6295 | { | |
6296 | int i, rc = -ENOMEM; | |
6297 | ||
6298 | BUG_ON(!dev_boot_phase); | |
6299 | ||
1da177e4 LT |
6300 | if (dev_proc_init()) |
6301 | goto out; | |
6302 | ||
8b41d188 | 6303 | if (netdev_kobject_init()) |
1da177e4 LT |
6304 | goto out; |
6305 | ||
6306 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 6307 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
6308 | INIT_LIST_HEAD(&ptype_base[i]); |
6309 | ||
881d966b EB |
6310 | if (register_pernet_subsys(&netdev_net_ops)) |
6311 | goto out; | |
1da177e4 LT |
6312 | |
6313 | /* | |
6314 | * Initialise the packet receive queues. | |
6315 | */ | |
6316 | ||
6f912042 | 6317 | for_each_possible_cpu(i) { |
e36fa2f7 | 6318 | struct softnet_data *sd = &per_cpu(softnet_data, i); |
1da177e4 | 6319 | |
dee42870 | 6320 | memset(sd, 0, sizeof(*sd)); |
e36fa2f7 | 6321 | skb_queue_head_init(&sd->input_pkt_queue); |
6e7676c1 | 6322 | skb_queue_head_init(&sd->process_queue); |
e36fa2f7 ED |
6323 | sd->completion_queue = NULL; |
6324 | INIT_LIST_HEAD(&sd->poll_list); | |
a9cbd588 CG |
6325 | sd->output_queue = NULL; |
6326 | sd->output_queue_tailp = &sd->output_queue; | |
df334545 | 6327 | #ifdef CONFIG_RPS |
e36fa2f7 ED |
6328 | sd->csd.func = rps_trigger_softirq; |
6329 | sd->csd.info = sd; | |
6330 | sd->csd.flags = 0; | |
6331 | sd->cpu = i; | |
1e94d72f | 6332 | #endif |
0a9627f2 | 6333 | |
e36fa2f7 ED |
6334 | sd->backlog.poll = process_backlog; |
6335 | sd->backlog.weight = weight_p; | |
6336 | sd->backlog.gro_list = NULL; | |
6337 | sd->backlog.gro_count = 0; | |
1da177e4 LT |
6338 | } |
6339 | ||
1da177e4 LT |
6340 | dev_boot_phase = 0; |
6341 | ||
505d4f73 EB |
6342 | /* The loopback device is special if any other network devices |
6343 | * is present in a network namespace the loopback device must | |
6344 | * be present. Since we now dynamically allocate and free the | |
6345 | * loopback device ensure this invariant is maintained by | |
6346 | * keeping the loopback device as the first device on the | |
6347 | * list of network devices. Ensuring the loopback devices | |
6348 | * is the first device that appears and the last network device | |
6349 | * that disappears. | |
6350 | */ | |
6351 | if (register_pernet_device(&loopback_net_ops)) | |
6352 | goto out; | |
6353 | ||
6354 | if (register_pernet_device(&default_device_ops)) | |
6355 | goto out; | |
6356 | ||
962cf36c CM |
6357 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
6358 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 LT |
6359 | |
6360 | hotcpu_notifier(dev_cpu_callback, 0); | |
6361 | dst_init(); | |
6362 | dev_mcast_init(); | |
6363 | rc = 0; | |
6364 | out: | |
6365 | return rc; | |
6366 | } | |
6367 | ||
6368 | subsys_initcall(net_dev_init); | |
6369 | ||
e88721f8 KK |
6370 | static int __init initialize_hashrnd(void) |
6371 | { | |
0a9627f2 | 6372 | get_random_bytes(&hashrnd, sizeof(hashrnd)); |
e88721f8 KK |
6373 | return 0; |
6374 | } | |
6375 | ||
6376 | late_initcall_sync(initialize_hashrnd); | |
6377 |