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> |
c445477d | 135 | #include <linux/cpu_rmap.h> |
1da177e4 | 136 | |
342709ef PE |
137 | #include "net-sysfs.h" |
138 | ||
d565b0a1 HX |
139 | /* Instead of increasing this, you should create a hash table. */ |
140 | #define MAX_GRO_SKBS 8 | |
141 | ||
5d38a079 HX |
142 | /* This should be increased if a protocol with a bigger head is added. */ |
143 | #define GRO_MAX_HEAD (MAX_HEADER + 128) | |
144 | ||
1da177e4 LT |
145 | /* |
146 | * The list of packet types we will receive (as opposed to discard) | |
147 | * and the routines to invoke. | |
148 | * | |
149 | * Why 16. Because with 16 the only overlap we get on a hash of the | |
150 | * low nibble of the protocol value is RARP/SNAP/X.25. | |
151 | * | |
152 | * NOTE: That is no longer true with the addition of VLAN tags. Not | |
153 | * sure which should go first, but I bet it won't make much | |
154 | * difference if we are running VLANs. The good news is that | |
155 | * this protocol won't be in the list unless compiled in, so | |
3041a069 | 156 | * the average user (w/out VLANs) will not be adversely affected. |
1da177e4 LT |
157 | * --BLG |
158 | * | |
159 | * 0800 IP | |
160 | * 8100 802.1Q VLAN | |
161 | * 0001 802.3 | |
162 | * 0002 AX.25 | |
163 | * 0004 802.2 | |
164 | * 8035 RARP | |
165 | * 0005 SNAP | |
166 | * 0805 X.25 | |
167 | * 0806 ARP | |
168 | * 8137 IPX | |
169 | * 0009 Localtalk | |
170 | * 86DD IPv6 | |
171 | */ | |
172 | ||
82d8a867 PE |
173 | #define PTYPE_HASH_SIZE (16) |
174 | #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1) | |
175 | ||
1da177e4 | 176 | static DEFINE_SPINLOCK(ptype_lock); |
82d8a867 | 177 | static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
6b2bedc3 | 178 | static struct list_head ptype_all __read_mostly; /* Taps */ |
1da177e4 | 179 | |
1da177e4 | 180 | /* |
7562f876 | 181 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
182 | * semaphore. |
183 | * | |
c6d14c84 | 184 | * Pure readers hold dev_base_lock for reading, or rcu_read_lock() |
1da177e4 LT |
185 | * |
186 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 187 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
188 | * actual updates. This allows pure readers to access the list even |
189 | * while a writer is preparing to update it. | |
190 | * | |
191 | * To put it another way, dev_base_lock is held for writing only to | |
192 | * protect against pure readers; the rtnl semaphore provides the | |
193 | * protection against other writers. | |
194 | * | |
195 | * See, for example usages, register_netdevice() and | |
196 | * unregister_netdevice(), which must be called with the rtnl | |
197 | * semaphore held. | |
198 | */ | |
1da177e4 | 199 | DEFINE_RWLOCK(dev_base_lock); |
1da177e4 LT |
200 | EXPORT_SYMBOL(dev_base_lock); |
201 | ||
4e985ada TG |
202 | static inline void dev_base_seq_inc(struct net *net) |
203 | { | |
204 | while (++net->dev_base_seq == 0); | |
205 | } | |
206 | ||
881d966b | 207 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 LT |
208 | { |
209 | unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ)); | |
08e9897d | 210 | return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; |
1da177e4 LT |
211 | } |
212 | ||
881d966b | 213 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 214 | { |
7c28bd0b | 215 | return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; |
1da177e4 LT |
216 | } |
217 | ||
e36fa2f7 | 218 | static inline void rps_lock(struct softnet_data *sd) |
152102c7 CG |
219 | { |
220 | #ifdef CONFIG_RPS | |
e36fa2f7 | 221 | spin_lock(&sd->input_pkt_queue.lock); |
152102c7 CG |
222 | #endif |
223 | } | |
224 | ||
e36fa2f7 | 225 | static inline void rps_unlock(struct softnet_data *sd) |
152102c7 CG |
226 | { |
227 | #ifdef CONFIG_RPS | |
e36fa2f7 | 228 | spin_unlock(&sd->input_pkt_queue.lock); |
152102c7 CG |
229 | #endif |
230 | } | |
231 | ||
ce286d32 EB |
232 | /* Device list insertion */ |
233 | static int list_netdevice(struct net_device *dev) | |
234 | { | |
c346dca1 | 235 | struct net *net = dev_net(dev); |
ce286d32 EB |
236 | |
237 | ASSERT_RTNL(); | |
238 | ||
239 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 240 | list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); |
72c9528b | 241 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); |
fb699dfd ED |
242 | hlist_add_head_rcu(&dev->index_hlist, |
243 | dev_index_hash(net, dev->ifindex)); | |
ce286d32 | 244 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
245 | |
246 | dev_base_seq_inc(net); | |
247 | ||
ce286d32 EB |
248 | return 0; |
249 | } | |
250 | ||
fb699dfd ED |
251 | /* Device list removal |
252 | * caller must respect a RCU grace period before freeing/reusing dev | |
253 | */ | |
ce286d32 EB |
254 | static void unlist_netdevice(struct net_device *dev) |
255 | { | |
256 | ASSERT_RTNL(); | |
257 | ||
258 | /* Unlink dev from the device chain */ | |
259 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 260 | list_del_rcu(&dev->dev_list); |
72c9528b | 261 | hlist_del_rcu(&dev->name_hlist); |
fb699dfd | 262 | hlist_del_rcu(&dev->index_hlist); |
ce286d32 | 263 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
264 | |
265 | dev_base_seq_inc(dev_net(dev)); | |
ce286d32 EB |
266 | } |
267 | ||
1da177e4 LT |
268 | /* |
269 | * Our notifier list | |
270 | */ | |
271 | ||
f07d5b94 | 272 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
273 | |
274 | /* | |
275 | * Device drivers call our routines to queue packets here. We empty the | |
276 | * queue in the local softnet handler. | |
277 | */ | |
bea3348e | 278 | |
9958da05 | 279 | DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); |
d1b19dff | 280 | EXPORT_PER_CPU_SYMBOL(softnet_data); |
1da177e4 | 281 | |
cf508b12 | 282 | #ifdef CONFIG_LOCKDEP |
723e98b7 | 283 | /* |
c773e847 | 284 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class |
723e98b7 JP |
285 | * according to dev->type |
286 | */ | |
287 | static const unsigned short netdev_lock_type[] = | |
288 | {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
289 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, | |
290 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
291 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
292 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
293 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
294 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
295 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
296 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
297 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
298 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
299 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
300 | ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211, | |
2d91d78b | 301 | ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, |
929122cd | 302 | ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154, |
fcb94e42 | 303 | ARPHRD_VOID, ARPHRD_NONE}; |
723e98b7 | 304 | |
36cbd3dc | 305 | static const char *const netdev_lock_name[] = |
723e98b7 JP |
306 | {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", |
307 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
308 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
309 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
310 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
311 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
312 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
313 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
314 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
315 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
316 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
317 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
318 | "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211", | |
2d91d78b | 319 | "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", |
929122cd | 320 | "_xmit_PHONET_PIPE", "_xmit_IEEE802154", |
fcb94e42 | 321 | "_xmit_VOID", "_xmit_NONE"}; |
723e98b7 JP |
322 | |
323 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
cf508b12 | 324 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; |
723e98b7 JP |
325 | |
326 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
327 | { | |
328 | int i; | |
329 | ||
330 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
331 | if (netdev_lock_type[i] == dev_type) | |
332 | return i; | |
333 | /* the last key is used by default */ | |
334 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
335 | } | |
336 | ||
cf508b12 DM |
337 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
338 | unsigned short dev_type) | |
723e98b7 JP |
339 | { |
340 | int i; | |
341 | ||
342 | i = netdev_lock_pos(dev_type); | |
343 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
344 | netdev_lock_name[i]); | |
345 | } | |
cf508b12 DM |
346 | |
347 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
348 | { | |
349 | int i; | |
350 | ||
351 | i = netdev_lock_pos(dev->type); | |
352 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
353 | &netdev_addr_lock_key[i], | |
354 | netdev_lock_name[i]); | |
355 | } | |
723e98b7 | 356 | #else |
cf508b12 DM |
357 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
358 | unsigned short dev_type) | |
359 | { | |
360 | } | |
361 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
723e98b7 JP |
362 | { |
363 | } | |
364 | #endif | |
1da177e4 LT |
365 | |
366 | /******************************************************************************* | |
367 | ||
368 | Protocol management and registration routines | |
369 | ||
370 | *******************************************************************************/ | |
371 | ||
1da177e4 LT |
372 | /* |
373 | * Add a protocol ID to the list. Now that the input handler is | |
374 | * smarter we can dispense with all the messy stuff that used to be | |
375 | * here. | |
376 | * | |
377 | * BEWARE!!! Protocol handlers, mangling input packets, | |
378 | * MUST BE last in hash buckets and checking protocol handlers | |
379 | * MUST start from promiscuous ptype_all chain in net_bh. | |
380 | * It is true now, do not change it. | |
381 | * Explanation follows: if protocol handler, mangling packet, will | |
382 | * be the first on list, it is not able to sense, that packet | |
383 | * is cloned and should be copied-on-write, so that it will | |
384 | * change it and subsequent readers will get broken packet. | |
385 | * --ANK (980803) | |
386 | */ | |
387 | ||
c07b68e8 ED |
388 | static inline struct list_head *ptype_head(const struct packet_type *pt) |
389 | { | |
390 | if (pt->type == htons(ETH_P_ALL)) | |
391 | return &ptype_all; | |
392 | else | |
393 | return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
394 | } | |
395 | ||
1da177e4 LT |
396 | /** |
397 | * dev_add_pack - add packet handler | |
398 | * @pt: packet type declaration | |
399 | * | |
400 | * Add a protocol handler to the networking stack. The passed &packet_type | |
401 | * is linked into kernel lists and may not be freed until it has been | |
402 | * removed from the kernel lists. | |
403 | * | |
4ec93edb | 404 | * This call does not sleep therefore it can not |
1da177e4 LT |
405 | * guarantee all CPU's that are in middle of receiving packets |
406 | * will see the new packet type (until the next received packet). | |
407 | */ | |
408 | ||
409 | void dev_add_pack(struct packet_type *pt) | |
410 | { | |
c07b68e8 | 411 | struct list_head *head = ptype_head(pt); |
1da177e4 | 412 | |
c07b68e8 ED |
413 | spin_lock(&ptype_lock); |
414 | list_add_rcu(&pt->list, head); | |
415 | spin_unlock(&ptype_lock); | |
1da177e4 | 416 | } |
d1b19dff | 417 | EXPORT_SYMBOL(dev_add_pack); |
1da177e4 | 418 | |
1da177e4 LT |
419 | /** |
420 | * __dev_remove_pack - remove packet handler | |
421 | * @pt: packet type declaration | |
422 | * | |
423 | * Remove a protocol handler that was previously added to the kernel | |
424 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
425 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 426 | * returns. |
1da177e4 LT |
427 | * |
428 | * The packet type might still be in use by receivers | |
429 | * and must not be freed until after all the CPU's have gone | |
430 | * through a quiescent state. | |
431 | */ | |
432 | void __dev_remove_pack(struct packet_type *pt) | |
433 | { | |
c07b68e8 | 434 | struct list_head *head = ptype_head(pt); |
1da177e4 LT |
435 | struct packet_type *pt1; |
436 | ||
c07b68e8 | 437 | spin_lock(&ptype_lock); |
1da177e4 LT |
438 | |
439 | list_for_each_entry(pt1, head, list) { | |
440 | if (pt == pt1) { | |
441 | list_del_rcu(&pt->list); | |
442 | goto out; | |
443 | } | |
444 | } | |
445 | ||
446 | printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt); | |
447 | out: | |
c07b68e8 | 448 | spin_unlock(&ptype_lock); |
1da177e4 | 449 | } |
d1b19dff ED |
450 | EXPORT_SYMBOL(__dev_remove_pack); |
451 | ||
1da177e4 LT |
452 | /** |
453 | * dev_remove_pack - remove packet handler | |
454 | * @pt: packet type declaration | |
455 | * | |
456 | * Remove a protocol handler that was previously added to the kernel | |
457 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
458 | * from the kernel lists and can be freed or reused once this function | |
459 | * returns. | |
460 | * | |
461 | * This call sleeps to guarantee that no CPU is looking at the packet | |
462 | * type after return. | |
463 | */ | |
464 | void dev_remove_pack(struct packet_type *pt) | |
465 | { | |
466 | __dev_remove_pack(pt); | |
4ec93edb | 467 | |
1da177e4 LT |
468 | synchronize_net(); |
469 | } | |
d1b19dff | 470 | EXPORT_SYMBOL(dev_remove_pack); |
1da177e4 LT |
471 | |
472 | /****************************************************************************** | |
473 | ||
474 | Device Boot-time Settings Routines | |
475 | ||
476 | *******************************************************************************/ | |
477 | ||
478 | /* Boot time configuration table */ | |
479 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
480 | ||
481 | /** | |
482 | * netdev_boot_setup_add - add new setup entry | |
483 | * @name: name of the device | |
484 | * @map: configured settings for the device | |
485 | * | |
486 | * Adds new setup entry to the dev_boot_setup list. The function | |
487 | * returns 0 on error and 1 on success. This is a generic routine to | |
488 | * all netdevices. | |
489 | */ | |
490 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
491 | { | |
492 | struct netdev_boot_setup *s; | |
493 | int i; | |
494 | ||
495 | s = dev_boot_setup; | |
496 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
497 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
498 | memset(s[i].name, 0, sizeof(s[i].name)); | |
93b3cff9 | 499 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
500 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
501 | break; | |
502 | } | |
503 | } | |
504 | ||
505 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
506 | } | |
507 | ||
508 | /** | |
509 | * netdev_boot_setup_check - check boot time settings | |
510 | * @dev: the netdevice | |
511 | * | |
512 | * Check boot time settings for the device. | |
513 | * The found settings are set for the device to be used | |
514 | * later in the device probing. | |
515 | * Returns 0 if no settings found, 1 if they are. | |
516 | */ | |
517 | int netdev_boot_setup_check(struct net_device *dev) | |
518 | { | |
519 | struct netdev_boot_setup *s = dev_boot_setup; | |
520 | int i; | |
521 | ||
522 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
523 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
93b3cff9 | 524 | !strcmp(dev->name, s[i].name)) { |
1da177e4 LT |
525 | dev->irq = s[i].map.irq; |
526 | dev->base_addr = s[i].map.base_addr; | |
527 | dev->mem_start = s[i].map.mem_start; | |
528 | dev->mem_end = s[i].map.mem_end; | |
529 | return 1; | |
530 | } | |
531 | } | |
532 | return 0; | |
533 | } | |
d1b19dff | 534 | EXPORT_SYMBOL(netdev_boot_setup_check); |
1da177e4 LT |
535 | |
536 | ||
537 | /** | |
538 | * netdev_boot_base - get address from boot time settings | |
539 | * @prefix: prefix for network device | |
540 | * @unit: id for network device | |
541 | * | |
542 | * Check boot time settings for the base address of device. | |
543 | * The found settings are set for the device to be used | |
544 | * later in the device probing. | |
545 | * Returns 0 if no settings found. | |
546 | */ | |
547 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
548 | { | |
549 | const struct netdev_boot_setup *s = dev_boot_setup; | |
550 | char name[IFNAMSIZ]; | |
551 | int i; | |
552 | ||
553 | sprintf(name, "%s%d", prefix, unit); | |
554 | ||
555 | /* | |
556 | * If device already registered then return base of 1 | |
557 | * to indicate not to probe for this interface | |
558 | */ | |
881d966b | 559 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
560 | return 1; |
561 | ||
562 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
563 | if (!strcmp(name, s[i].name)) | |
564 | return s[i].map.base_addr; | |
565 | return 0; | |
566 | } | |
567 | ||
568 | /* | |
569 | * Saves at boot time configured settings for any netdevice. | |
570 | */ | |
571 | int __init netdev_boot_setup(char *str) | |
572 | { | |
573 | int ints[5]; | |
574 | struct ifmap map; | |
575 | ||
576 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
577 | if (!str || !*str) | |
578 | return 0; | |
579 | ||
580 | /* Save settings */ | |
581 | memset(&map, 0, sizeof(map)); | |
582 | if (ints[0] > 0) | |
583 | map.irq = ints[1]; | |
584 | if (ints[0] > 1) | |
585 | map.base_addr = ints[2]; | |
586 | if (ints[0] > 2) | |
587 | map.mem_start = ints[3]; | |
588 | if (ints[0] > 3) | |
589 | map.mem_end = ints[4]; | |
590 | ||
591 | /* Add new entry to the list */ | |
592 | return netdev_boot_setup_add(str, &map); | |
593 | } | |
594 | ||
595 | __setup("netdev=", netdev_boot_setup); | |
596 | ||
597 | /******************************************************************************* | |
598 | ||
599 | Device Interface Subroutines | |
600 | ||
601 | *******************************************************************************/ | |
602 | ||
603 | /** | |
604 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 605 | * @net: the applicable net namespace |
1da177e4 LT |
606 | * @name: name to find |
607 | * | |
608 | * Find an interface by name. Must be called under RTNL semaphore | |
609 | * or @dev_base_lock. If the name is found a pointer to the device | |
610 | * is returned. If the name is not found then %NULL is returned. The | |
611 | * reference counters are not incremented so the caller must be | |
612 | * careful with locks. | |
613 | */ | |
614 | ||
881d966b | 615 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
616 | { |
617 | struct hlist_node *p; | |
0bd8d536 ED |
618 | struct net_device *dev; |
619 | struct hlist_head *head = dev_name_hash(net, name); | |
1da177e4 | 620 | |
0bd8d536 | 621 | hlist_for_each_entry(dev, p, head, name_hlist) |
1da177e4 LT |
622 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
623 | return dev; | |
0bd8d536 | 624 | |
1da177e4 LT |
625 | return NULL; |
626 | } | |
d1b19dff | 627 | EXPORT_SYMBOL(__dev_get_by_name); |
1da177e4 | 628 | |
72c9528b ED |
629 | /** |
630 | * dev_get_by_name_rcu - find a device by its name | |
631 | * @net: the applicable net namespace | |
632 | * @name: name to find | |
633 | * | |
634 | * Find an interface by name. | |
635 | * If the name is found a pointer to the device is returned. | |
636 | * If the name is not found then %NULL is returned. | |
637 | * The reference counters are not incremented so the caller must be | |
638 | * careful with locks. The caller must hold RCU lock. | |
639 | */ | |
640 | ||
641 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
642 | { | |
643 | struct hlist_node *p; | |
644 | struct net_device *dev; | |
645 | struct hlist_head *head = dev_name_hash(net, name); | |
646 | ||
647 | hlist_for_each_entry_rcu(dev, p, head, name_hlist) | |
648 | if (!strncmp(dev->name, name, IFNAMSIZ)) | |
649 | return dev; | |
650 | ||
651 | return NULL; | |
652 | } | |
653 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
654 | ||
1da177e4 LT |
655 | /** |
656 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 657 | * @net: the applicable net namespace |
1da177e4 LT |
658 | * @name: name to find |
659 | * | |
660 | * Find an interface by name. This can be called from any | |
661 | * context and does its own locking. The returned handle has | |
662 | * the usage count incremented and the caller must use dev_put() to | |
663 | * release it when it is no longer needed. %NULL is returned if no | |
664 | * matching device is found. | |
665 | */ | |
666 | ||
881d966b | 667 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
668 | { |
669 | struct net_device *dev; | |
670 | ||
72c9528b ED |
671 | rcu_read_lock(); |
672 | dev = dev_get_by_name_rcu(net, name); | |
1da177e4 LT |
673 | if (dev) |
674 | dev_hold(dev); | |
72c9528b | 675 | rcu_read_unlock(); |
1da177e4 LT |
676 | return dev; |
677 | } | |
d1b19dff | 678 | EXPORT_SYMBOL(dev_get_by_name); |
1da177e4 LT |
679 | |
680 | /** | |
681 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 682 | * @net: the applicable net namespace |
1da177e4 LT |
683 | * @ifindex: index of device |
684 | * | |
685 | * Search for an interface by index. Returns %NULL if the device | |
686 | * is not found or a pointer to the device. The device has not | |
687 | * had its reference counter increased so the caller must be careful | |
688 | * about locking. The caller must hold either the RTNL semaphore | |
689 | * or @dev_base_lock. | |
690 | */ | |
691 | ||
881d966b | 692 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
693 | { |
694 | struct hlist_node *p; | |
0bd8d536 ED |
695 | struct net_device *dev; |
696 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
1da177e4 | 697 | |
0bd8d536 | 698 | hlist_for_each_entry(dev, p, head, index_hlist) |
1da177e4 LT |
699 | if (dev->ifindex == ifindex) |
700 | return dev; | |
0bd8d536 | 701 | |
1da177e4 LT |
702 | return NULL; |
703 | } | |
d1b19dff | 704 | EXPORT_SYMBOL(__dev_get_by_index); |
1da177e4 | 705 | |
fb699dfd ED |
706 | /** |
707 | * dev_get_by_index_rcu - find a device by its ifindex | |
708 | * @net: the applicable net namespace | |
709 | * @ifindex: index of device | |
710 | * | |
711 | * Search for an interface by index. Returns %NULL if the device | |
712 | * is not found or a pointer to the device. The device has not | |
713 | * had its reference counter increased so the caller must be careful | |
714 | * about locking. The caller must hold RCU lock. | |
715 | */ | |
716 | ||
717 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
718 | { | |
719 | struct hlist_node *p; | |
720 | struct net_device *dev; | |
721 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
722 | ||
723 | hlist_for_each_entry_rcu(dev, p, head, index_hlist) | |
724 | if (dev->ifindex == ifindex) | |
725 | return dev; | |
726 | ||
727 | return NULL; | |
728 | } | |
729 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
730 | ||
1da177e4 LT |
731 | |
732 | /** | |
733 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 734 | * @net: the applicable net namespace |
1da177e4 LT |
735 | * @ifindex: index of device |
736 | * | |
737 | * Search for an interface by index. Returns NULL if the device | |
738 | * is not found or a pointer to the device. The device returned has | |
739 | * had a reference added and the pointer is safe until the user calls | |
740 | * dev_put to indicate they have finished with it. | |
741 | */ | |
742 | ||
881d966b | 743 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
744 | { |
745 | struct net_device *dev; | |
746 | ||
fb699dfd ED |
747 | rcu_read_lock(); |
748 | dev = dev_get_by_index_rcu(net, ifindex); | |
1da177e4 LT |
749 | if (dev) |
750 | dev_hold(dev); | |
fb699dfd | 751 | rcu_read_unlock(); |
1da177e4 LT |
752 | return dev; |
753 | } | |
d1b19dff | 754 | EXPORT_SYMBOL(dev_get_by_index); |
1da177e4 LT |
755 | |
756 | /** | |
941666c2 | 757 | * dev_getbyhwaddr_rcu - find a device by its hardware address |
c4ea43c5 | 758 | * @net: the applicable net namespace |
1da177e4 LT |
759 | * @type: media type of device |
760 | * @ha: hardware address | |
761 | * | |
762 | * Search for an interface by MAC address. Returns NULL if the device | |
c506653d ED |
763 | * is not found or a pointer to the device. |
764 | * The caller must hold RCU or RTNL. | |
941666c2 | 765 | * The returned device has not had its ref count increased |
1da177e4 LT |
766 | * and the caller must therefore be careful about locking |
767 | * | |
1da177e4 LT |
768 | */ |
769 | ||
941666c2 ED |
770 | struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, |
771 | const char *ha) | |
1da177e4 LT |
772 | { |
773 | struct net_device *dev; | |
774 | ||
941666c2 | 775 | for_each_netdev_rcu(net, dev) |
1da177e4 LT |
776 | if (dev->type == type && |
777 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
778 | return dev; |
779 | ||
780 | return NULL; | |
1da177e4 | 781 | } |
941666c2 | 782 | EXPORT_SYMBOL(dev_getbyhwaddr_rcu); |
cf309e3f | 783 | |
881d966b | 784 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
785 | { |
786 | struct net_device *dev; | |
787 | ||
4e9cac2b | 788 | ASSERT_RTNL(); |
881d966b | 789 | for_each_netdev(net, dev) |
4e9cac2b | 790 | if (dev->type == type) |
7562f876 PE |
791 | return dev; |
792 | ||
793 | return NULL; | |
4e9cac2b | 794 | } |
4e9cac2b PM |
795 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); |
796 | ||
881d966b | 797 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b | 798 | { |
99fe3c39 | 799 | struct net_device *dev, *ret = NULL; |
4e9cac2b | 800 | |
99fe3c39 ED |
801 | rcu_read_lock(); |
802 | for_each_netdev_rcu(net, dev) | |
803 | if (dev->type == type) { | |
804 | dev_hold(dev); | |
805 | ret = dev; | |
806 | break; | |
807 | } | |
808 | rcu_read_unlock(); | |
809 | return ret; | |
1da177e4 | 810 | } |
1da177e4 LT |
811 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
812 | ||
813 | /** | |
bb69ae04 | 814 | * dev_get_by_flags_rcu - find any device with given flags |
c4ea43c5 | 815 | * @net: the applicable net namespace |
1da177e4 LT |
816 | * @if_flags: IFF_* values |
817 | * @mask: bitmask of bits in if_flags to check | |
818 | * | |
819 | * Search for any interface with the given flags. Returns NULL if a device | |
bb69ae04 ED |
820 | * is not found or a pointer to the device. Must be called inside |
821 | * rcu_read_lock(), and result refcount is unchanged. | |
1da177e4 LT |
822 | */ |
823 | ||
bb69ae04 | 824 | struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags, |
d1b19dff | 825 | unsigned short mask) |
1da177e4 | 826 | { |
7562f876 | 827 | struct net_device *dev, *ret; |
1da177e4 | 828 | |
7562f876 | 829 | ret = NULL; |
c6d14c84 | 830 | for_each_netdev_rcu(net, dev) { |
1da177e4 | 831 | if (((dev->flags ^ if_flags) & mask) == 0) { |
7562f876 | 832 | ret = dev; |
1da177e4 LT |
833 | break; |
834 | } | |
835 | } | |
7562f876 | 836 | return ret; |
1da177e4 | 837 | } |
bb69ae04 | 838 | EXPORT_SYMBOL(dev_get_by_flags_rcu); |
1da177e4 LT |
839 | |
840 | /** | |
841 | * dev_valid_name - check if name is okay for network device | |
842 | * @name: name string | |
843 | * | |
844 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
845 | * to allow sysfs to work. We also disallow any kind of |
846 | * whitespace. | |
1da177e4 | 847 | */ |
c2373ee9 | 848 | int dev_valid_name(const char *name) |
1da177e4 | 849 | { |
c7fa9d18 DM |
850 | if (*name == '\0') |
851 | return 0; | |
b6fe17d6 SH |
852 | if (strlen(name) >= IFNAMSIZ) |
853 | return 0; | |
c7fa9d18 DM |
854 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
855 | return 0; | |
856 | ||
857 | while (*name) { | |
858 | if (*name == '/' || isspace(*name)) | |
859 | return 0; | |
860 | name++; | |
861 | } | |
862 | return 1; | |
1da177e4 | 863 | } |
d1b19dff | 864 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
865 | |
866 | /** | |
b267b179 EB |
867 | * __dev_alloc_name - allocate a name for a device |
868 | * @net: network namespace to allocate the device name in | |
1da177e4 | 869 | * @name: name format string |
b267b179 | 870 | * @buf: scratch buffer and result name string |
1da177e4 LT |
871 | * |
872 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
873 | * id. It scans list of devices to build up a free map, then chooses |
874 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
875 | * while allocating the name and adding the device in order to avoid | |
876 | * duplicates. | |
877 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
878 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
879 | */ |
880 | ||
b267b179 | 881 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
882 | { |
883 | int i = 0; | |
1da177e4 LT |
884 | const char *p; |
885 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 886 | unsigned long *inuse; |
1da177e4 LT |
887 | struct net_device *d; |
888 | ||
889 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
890 | if (p) { | |
891 | /* | |
892 | * Verify the string as this thing may have come from | |
893 | * the user. There must be either one "%d" and no other "%" | |
894 | * characters. | |
895 | */ | |
896 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
897 | return -EINVAL; | |
898 | ||
899 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 900 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
901 | if (!inuse) |
902 | return -ENOMEM; | |
903 | ||
881d966b | 904 | for_each_netdev(net, d) { |
1da177e4 LT |
905 | if (!sscanf(d->name, name, &i)) |
906 | continue; | |
907 | if (i < 0 || i >= max_netdevices) | |
908 | continue; | |
909 | ||
910 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 911 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
912 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
913 | set_bit(i, inuse); | |
914 | } | |
915 | ||
916 | i = find_first_zero_bit(inuse, max_netdevices); | |
917 | free_page((unsigned long) inuse); | |
918 | } | |
919 | ||
d9031024 OP |
920 | if (buf != name) |
921 | snprintf(buf, IFNAMSIZ, name, i); | |
b267b179 | 922 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 923 | return i; |
1da177e4 LT |
924 | |
925 | /* It is possible to run out of possible slots | |
926 | * when the name is long and there isn't enough space left | |
927 | * for the digits, or if all bits are used. | |
928 | */ | |
929 | return -ENFILE; | |
930 | } | |
931 | ||
b267b179 EB |
932 | /** |
933 | * dev_alloc_name - allocate a name for a device | |
934 | * @dev: device | |
935 | * @name: name format string | |
936 | * | |
937 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
938 | * id. It scans list of devices to build up a free map, then chooses | |
939 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
940 | * while allocating the name and adding the device in order to avoid | |
941 | * duplicates. | |
942 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
943 | * Returns the number of the unit assigned or a negative errno code. | |
944 | */ | |
945 | ||
946 | int dev_alloc_name(struct net_device *dev, const char *name) | |
947 | { | |
948 | char buf[IFNAMSIZ]; | |
949 | struct net *net; | |
950 | int ret; | |
951 | ||
c346dca1 YH |
952 | BUG_ON(!dev_net(dev)); |
953 | net = dev_net(dev); | |
b267b179 EB |
954 | ret = __dev_alloc_name(net, name, buf); |
955 | if (ret >= 0) | |
956 | strlcpy(dev->name, buf, IFNAMSIZ); | |
957 | return ret; | |
958 | } | |
d1b19dff | 959 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 960 | |
1c5cae81 | 961 | static int dev_get_valid_name(struct net_device *dev, const char *name) |
d9031024 | 962 | { |
8ce6cebc DL |
963 | struct net *net; |
964 | ||
965 | BUG_ON(!dev_net(dev)); | |
966 | net = dev_net(dev); | |
967 | ||
d9031024 OP |
968 | if (!dev_valid_name(name)) |
969 | return -EINVAL; | |
970 | ||
1c5cae81 | 971 | if (strchr(name, '%')) |
8ce6cebc | 972 | return dev_alloc_name(dev, name); |
d9031024 OP |
973 | else if (__dev_get_by_name(net, name)) |
974 | return -EEXIST; | |
8ce6cebc DL |
975 | else if (dev->name != name) |
976 | strlcpy(dev->name, name, IFNAMSIZ); | |
d9031024 OP |
977 | |
978 | return 0; | |
979 | } | |
1da177e4 LT |
980 | |
981 | /** | |
982 | * dev_change_name - change name of a device | |
983 | * @dev: device | |
984 | * @newname: name (or format string) must be at least IFNAMSIZ | |
985 | * | |
986 | * Change name of a device, can pass format strings "eth%d". | |
987 | * for wildcarding. | |
988 | */ | |
cf04a4c7 | 989 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 990 | { |
fcc5a03a | 991 | char oldname[IFNAMSIZ]; |
1da177e4 | 992 | int err = 0; |
fcc5a03a | 993 | int ret; |
881d966b | 994 | struct net *net; |
1da177e4 LT |
995 | |
996 | ASSERT_RTNL(); | |
c346dca1 | 997 | BUG_ON(!dev_net(dev)); |
1da177e4 | 998 | |
c346dca1 | 999 | net = dev_net(dev); |
1da177e4 LT |
1000 | if (dev->flags & IFF_UP) |
1001 | return -EBUSY; | |
1002 | ||
c8d90dca SH |
1003 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) |
1004 | return 0; | |
1005 | ||
fcc5a03a HX |
1006 | memcpy(oldname, dev->name, IFNAMSIZ); |
1007 | ||
1c5cae81 | 1008 | err = dev_get_valid_name(dev, newname); |
d9031024 OP |
1009 | if (err < 0) |
1010 | return err; | |
1da177e4 | 1011 | |
fcc5a03a | 1012 | rollback: |
a1b3f594 EB |
1013 | ret = device_rename(&dev->dev, dev->name); |
1014 | if (ret) { | |
1015 | memcpy(dev->name, oldname, IFNAMSIZ); | |
1016 | return ret; | |
dcc99773 | 1017 | } |
7f988eab HX |
1018 | |
1019 | write_lock_bh(&dev_base_lock); | |
372b2312 | 1020 | hlist_del_rcu(&dev->name_hlist); |
72c9528b ED |
1021 | write_unlock_bh(&dev_base_lock); |
1022 | ||
1023 | synchronize_rcu(); | |
1024 | ||
1025 | write_lock_bh(&dev_base_lock); | |
1026 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
7f988eab HX |
1027 | write_unlock_bh(&dev_base_lock); |
1028 | ||
056925ab | 1029 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1030 | ret = notifier_to_errno(ret); |
1031 | ||
1032 | if (ret) { | |
91e9c07b ED |
1033 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1034 | if (err >= 0) { | |
fcc5a03a HX |
1035 | err = ret; |
1036 | memcpy(dev->name, oldname, IFNAMSIZ); | |
1037 | goto rollback; | |
91e9c07b ED |
1038 | } else { |
1039 | printk(KERN_ERR | |
1040 | "%s: name change rollback failed: %d.\n", | |
1041 | dev->name, ret); | |
fcc5a03a HX |
1042 | } |
1043 | } | |
1da177e4 LT |
1044 | |
1045 | return err; | |
1046 | } | |
1047 | ||
0b815a1a SH |
1048 | /** |
1049 | * dev_set_alias - change ifalias of a device | |
1050 | * @dev: device | |
1051 | * @alias: name up to IFALIASZ | |
f0db275a | 1052 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1053 | * |
1054 | * Set ifalias for a device, | |
1055 | */ | |
1056 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1057 | { | |
1058 | ASSERT_RTNL(); | |
1059 | ||
1060 | if (len >= IFALIASZ) | |
1061 | return -EINVAL; | |
1062 | ||
96ca4a2c OH |
1063 | if (!len) { |
1064 | if (dev->ifalias) { | |
1065 | kfree(dev->ifalias); | |
1066 | dev->ifalias = NULL; | |
1067 | } | |
1068 | return 0; | |
1069 | } | |
1070 | ||
d1b19dff | 1071 | dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL); |
0b815a1a SH |
1072 | if (!dev->ifalias) |
1073 | return -ENOMEM; | |
1074 | ||
1075 | strlcpy(dev->ifalias, alias, len+1); | |
1076 | return len; | |
1077 | } | |
1078 | ||
1079 | ||
d8a33ac4 | 1080 | /** |
3041a069 | 1081 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1082 | * @dev: device to cause notification |
1083 | * | |
1084 | * Called to indicate a device has changed features. | |
1085 | */ | |
1086 | void netdev_features_change(struct net_device *dev) | |
1087 | { | |
056925ab | 1088 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1089 | } |
1090 | EXPORT_SYMBOL(netdev_features_change); | |
1091 | ||
1da177e4 LT |
1092 | /** |
1093 | * netdev_state_change - device changes state | |
1094 | * @dev: device to cause notification | |
1095 | * | |
1096 | * Called to indicate a device has changed state. This function calls | |
1097 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1098 | * to the routing socket. | |
1099 | */ | |
1100 | void netdev_state_change(struct net_device *dev) | |
1101 | { | |
1102 | if (dev->flags & IFF_UP) { | |
056925ab | 1103 | call_netdevice_notifiers(NETDEV_CHANGE, dev); |
1da177e4 LT |
1104 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0); |
1105 | } | |
1106 | } | |
d1b19dff | 1107 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1108 | |
3ca5b404 | 1109 | int netdev_bonding_change(struct net_device *dev, unsigned long event) |
c1da4ac7 | 1110 | { |
3ca5b404 | 1111 | return call_netdevice_notifiers(event, dev); |
c1da4ac7 OG |
1112 | } |
1113 | EXPORT_SYMBOL(netdev_bonding_change); | |
1114 | ||
1da177e4 LT |
1115 | /** |
1116 | * dev_load - load a network module | |
c4ea43c5 | 1117 | * @net: the applicable net namespace |
1da177e4 LT |
1118 | * @name: name of interface |
1119 | * | |
1120 | * If a network interface is not present and the process has suitable | |
1121 | * privileges this function loads the module. If module loading is not | |
1122 | * available in this kernel then it becomes a nop. | |
1123 | */ | |
1124 | ||
881d966b | 1125 | void dev_load(struct net *net, const char *name) |
1da177e4 | 1126 | { |
4ec93edb | 1127 | struct net_device *dev; |
8909c9ad | 1128 | int no_module; |
1da177e4 | 1129 | |
72c9528b ED |
1130 | rcu_read_lock(); |
1131 | dev = dev_get_by_name_rcu(net, name); | |
1132 | rcu_read_unlock(); | |
1da177e4 | 1133 | |
8909c9ad VK |
1134 | no_module = !dev; |
1135 | if (no_module && capable(CAP_NET_ADMIN)) | |
1136 | no_module = request_module("netdev-%s", name); | |
1137 | if (no_module && capable(CAP_SYS_MODULE)) { | |
1138 | if (!request_module("%s", name)) | |
1139 | pr_err("Loading kernel module for a network device " | |
1140 | "with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s " | |
1141 | "instead\n", name); | |
1142 | } | |
1da177e4 | 1143 | } |
d1b19dff | 1144 | EXPORT_SYMBOL(dev_load); |
1da177e4 | 1145 | |
bd380811 | 1146 | static int __dev_open(struct net_device *dev) |
1da177e4 | 1147 | { |
d314774c | 1148 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1149 | int ret; |
1da177e4 | 1150 | |
e46b66bc BH |
1151 | ASSERT_RTNL(); |
1152 | ||
1da177e4 LT |
1153 | if (!netif_device_present(dev)) |
1154 | return -ENODEV; | |
1155 | ||
3b8bcfd5 JB |
1156 | ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); |
1157 | ret = notifier_to_errno(ret); | |
1158 | if (ret) | |
1159 | return ret; | |
1160 | ||
1da177e4 | 1161 | set_bit(__LINK_STATE_START, &dev->state); |
bada339b | 1162 | |
d314774c SH |
1163 | if (ops->ndo_validate_addr) |
1164 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1165 | |
d314774c SH |
1166 | if (!ret && ops->ndo_open) |
1167 | ret = ops->ndo_open(dev); | |
1da177e4 | 1168 | |
bada339b JG |
1169 | if (ret) |
1170 | clear_bit(__LINK_STATE_START, &dev->state); | |
1171 | else { | |
1da177e4 | 1172 | dev->flags |= IFF_UP; |
b4bd07c2 | 1173 | net_dmaengine_get(); |
4417da66 | 1174 | dev_set_rx_mode(dev); |
1da177e4 | 1175 | dev_activate(dev); |
1da177e4 | 1176 | } |
bada339b | 1177 | |
1da177e4 LT |
1178 | return ret; |
1179 | } | |
1180 | ||
1181 | /** | |
bd380811 PM |
1182 | * dev_open - prepare an interface for use. |
1183 | * @dev: device to open | |
1da177e4 | 1184 | * |
bd380811 PM |
1185 | * Takes a device from down to up state. The device's private open |
1186 | * function is invoked and then the multicast lists are loaded. Finally | |
1187 | * the device is moved into the up state and a %NETDEV_UP message is | |
1188 | * sent to the netdev notifier chain. | |
1189 | * | |
1190 | * Calling this function on an active interface is a nop. On a failure | |
1191 | * a negative errno code is returned. | |
1da177e4 | 1192 | */ |
bd380811 PM |
1193 | int dev_open(struct net_device *dev) |
1194 | { | |
1195 | int ret; | |
1196 | ||
bd380811 PM |
1197 | if (dev->flags & IFF_UP) |
1198 | return 0; | |
1199 | ||
bd380811 PM |
1200 | ret = __dev_open(dev); |
1201 | if (ret < 0) | |
1202 | return ret; | |
1203 | ||
bd380811 PM |
1204 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING); |
1205 | call_netdevice_notifiers(NETDEV_UP, dev); | |
1206 | ||
1207 | return ret; | |
1208 | } | |
1209 | EXPORT_SYMBOL(dev_open); | |
1210 | ||
44345724 | 1211 | static int __dev_close_many(struct list_head *head) |
1da177e4 | 1212 | { |
44345724 | 1213 | struct net_device *dev; |
e46b66bc | 1214 | |
bd380811 | 1215 | ASSERT_RTNL(); |
9d5010db DM |
1216 | might_sleep(); |
1217 | ||
44345724 | 1218 | list_for_each_entry(dev, head, unreg_list) { |
44345724 | 1219 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1220 | |
44345724 | 1221 | clear_bit(__LINK_STATE_START, &dev->state); |
1da177e4 | 1222 | |
44345724 OP |
1223 | /* Synchronize to scheduled poll. We cannot touch poll list, it |
1224 | * can be even on different cpu. So just clear netif_running(). | |
1225 | * | |
1226 | * dev->stop() will invoke napi_disable() on all of it's | |
1227 | * napi_struct instances on this device. | |
1228 | */ | |
1229 | smp_mb__after_clear_bit(); /* Commit netif_running(). */ | |
1230 | } | |
1da177e4 | 1231 | |
44345724 | 1232 | dev_deactivate_many(head); |
d8b2a4d2 | 1233 | |
44345724 OP |
1234 | list_for_each_entry(dev, head, unreg_list) { |
1235 | const struct net_device_ops *ops = dev->netdev_ops; | |
1da177e4 | 1236 | |
44345724 OP |
1237 | /* |
1238 | * Call the device specific close. This cannot fail. | |
1239 | * Only if device is UP | |
1240 | * | |
1241 | * We allow it to be called even after a DETACH hot-plug | |
1242 | * event. | |
1243 | */ | |
1244 | if (ops->ndo_stop) | |
1245 | ops->ndo_stop(dev); | |
1246 | ||
44345724 | 1247 | dev->flags &= ~IFF_UP; |
44345724 OP |
1248 | net_dmaengine_put(); |
1249 | } | |
1250 | ||
1251 | return 0; | |
1252 | } | |
1253 | ||
1254 | static int __dev_close(struct net_device *dev) | |
1255 | { | |
f87e6f47 | 1256 | int retval; |
44345724 OP |
1257 | LIST_HEAD(single); |
1258 | ||
1259 | list_add(&dev->unreg_list, &single); | |
f87e6f47 LT |
1260 | retval = __dev_close_many(&single); |
1261 | list_del(&single); | |
1262 | return retval; | |
44345724 OP |
1263 | } |
1264 | ||
3fbd8758 | 1265 | static int dev_close_many(struct list_head *head) |
44345724 OP |
1266 | { |
1267 | struct net_device *dev, *tmp; | |
1268 | LIST_HEAD(tmp_list); | |
1da177e4 | 1269 | |
44345724 OP |
1270 | list_for_each_entry_safe(dev, tmp, head, unreg_list) |
1271 | if (!(dev->flags & IFF_UP)) | |
1272 | list_move(&dev->unreg_list, &tmp_list); | |
1273 | ||
1274 | __dev_close_many(head); | |
1da177e4 | 1275 | |
44345724 OP |
1276 | list_for_each_entry(dev, head, unreg_list) { |
1277 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING); | |
1278 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
1279 | } | |
bd380811 | 1280 | |
44345724 OP |
1281 | /* rollback_registered_many needs the complete original list */ |
1282 | list_splice(&tmp_list, head); | |
bd380811 PM |
1283 | return 0; |
1284 | } | |
1285 | ||
1286 | /** | |
1287 | * dev_close - shutdown an interface. | |
1288 | * @dev: device to shutdown | |
1289 | * | |
1290 | * This function moves an active device into down state. A | |
1291 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1292 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1293 | * chain. | |
1294 | */ | |
1295 | int dev_close(struct net_device *dev) | |
1296 | { | |
e14a5993 ED |
1297 | if (dev->flags & IFF_UP) { |
1298 | LIST_HEAD(single); | |
1da177e4 | 1299 | |
e14a5993 ED |
1300 | list_add(&dev->unreg_list, &single); |
1301 | dev_close_many(&single); | |
1302 | list_del(&single); | |
1303 | } | |
1da177e4 LT |
1304 | return 0; |
1305 | } | |
d1b19dff | 1306 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1307 | |
1308 | ||
0187bdfb BH |
1309 | /** |
1310 | * dev_disable_lro - disable Large Receive Offload on a device | |
1311 | * @dev: device | |
1312 | * | |
1313 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1314 | * called under RTNL. This is needed if received packets may be | |
1315 | * forwarded to another interface. | |
1316 | */ | |
1317 | void dev_disable_lro(struct net_device *dev) | |
1318 | { | |
27660515 MM |
1319 | u32 flags; |
1320 | ||
f11970e3 NH |
1321 | /* |
1322 | * If we're trying to disable lro on a vlan device | |
1323 | * use the underlying physical device instead | |
1324 | */ | |
1325 | if (is_vlan_dev(dev)) | |
1326 | dev = vlan_dev_real_dev(dev); | |
1327 | ||
27660515 MM |
1328 | if (dev->ethtool_ops && dev->ethtool_ops->get_flags) |
1329 | flags = dev->ethtool_ops->get_flags(dev); | |
1330 | else | |
1331 | flags = ethtool_op_get_flags(dev); | |
1332 | ||
1333 | if (!(flags & ETH_FLAG_LRO)) | |
1334 | return; | |
1335 | ||
1336 | __ethtool_set_flags(dev, flags & ~ETH_FLAG_LRO); | |
22d5969f MM |
1337 | if (unlikely(dev->features & NETIF_F_LRO)) |
1338 | netdev_WARN(dev, "failed to disable LRO!\n"); | |
0187bdfb BH |
1339 | } |
1340 | EXPORT_SYMBOL(dev_disable_lro); | |
1341 | ||
1342 | ||
881d966b EB |
1343 | static int dev_boot_phase = 1; |
1344 | ||
1da177e4 LT |
1345 | /** |
1346 | * register_netdevice_notifier - register a network notifier block | |
1347 | * @nb: notifier | |
1348 | * | |
1349 | * Register a notifier to be called when network device events occur. | |
1350 | * The notifier passed is linked into the kernel structures and must | |
1351 | * not be reused until it has been unregistered. A negative errno code | |
1352 | * is returned on a failure. | |
1353 | * | |
1354 | * When registered all registration and up events are replayed | |
4ec93edb | 1355 | * to the new notifier to allow device to have a race free |
1da177e4 LT |
1356 | * view of the network device list. |
1357 | */ | |
1358 | ||
1359 | int register_netdevice_notifier(struct notifier_block *nb) | |
1360 | { | |
1361 | struct net_device *dev; | |
fcc5a03a | 1362 | struct net_device *last; |
881d966b | 1363 | struct net *net; |
1da177e4 LT |
1364 | int err; |
1365 | ||
1366 | rtnl_lock(); | |
f07d5b94 | 1367 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1368 | if (err) |
1369 | goto unlock; | |
881d966b EB |
1370 | if (dev_boot_phase) |
1371 | goto unlock; | |
1372 | for_each_net(net) { | |
1373 | for_each_netdev(net, dev) { | |
1374 | err = nb->notifier_call(nb, NETDEV_REGISTER, dev); | |
1375 | err = notifier_to_errno(err); | |
1376 | if (err) | |
1377 | goto rollback; | |
1378 | ||
1379 | if (!(dev->flags & IFF_UP)) | |
1380 | continue; | |
1da177e4 | 1381 | |
881d966b EB |
1382 | nb->notifier_call(nb, NETDEV_UP, dev); |
1383 | } | |
1da177e4 | 1384 | } |
fcc5a03a HX |
1385 | |
1386 | unlock: | |
1da177e4 LT |
1387 | rtnl_unlock(); |
1388 | return err; | |
fcc5a03a HX |
1389 | |
1390 | rollback: | |
1391 | last = dev; | |
881d966b EB |
1392 | for_each_net(net) { |
1393 | for_each_netdev(net, dev) { | |
1394 | if (dev == last) | |
1395 | break; | |
fcc5a03a | 1396 | |
881d966b EB |
1397 | if (dev->flags & IFF_UP) { |
1398 | nb->notifier_call(nb, NETDEV_GOING_DOWN, dev); | |
1399 | nb->notifier_call(nb, NETDEV_DOWN, dev); | |
1400 | } | |
1401 | nb->notifier_call(nb, NETDEV_UNREGISTER, dev); | |
a5ee1551 | 1402 | nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev); |
fcc5a03a | 1403 | } |
fcc5a03a | 1404 | } |
c67625a1 PE |
1405 | |
1406 | raw_notifier_chain_unregister(&netdev_chain, nb); | |
fcc5a03a | 1407 | goto unlock; |
1da177e4 | 1408 | } |
d1b19dff | 1409 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1410 | |
1411 | /** | |
1412 | * unregister_netdevice_notifier - unregister a network notifier block | |
1413 | * @nb: notifier | |
1414 | * | |
1415 | * Unregister a notifier previously registered by | |
1416 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1417 | * kernel structures and may then be reused. A negative errno code | |
1418 | * is returned on a failure. | |
1419 | */ | |
1420 | ||
1421 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1422 | { | |
9f514950 HX |
1423 | int err; |
1424 | ||
1425 | rtnl_lock(); | |
f07d5b94 | 1426 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
9f514950 HX |
1427 | rtnl_unlock(); |
1428 | return err; | |
1da177e4 | 1429 | } |
d1b19dff | 1430 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 LT |
1431 | |
1432 | /** | |
1433 | * call_netdevice_notifiers - call all network notifier blocks | |
1434 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1435 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1436 | * |
1437 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1438 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1439 | */ |
1440 | ||
ad7379d4 | 1441 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1442 | { |
ab930471 | 1443 | ASSERT_RTNL(); |
ad7379d4 | 1444 | return raw_notifier_call_chain(&netdev_chain, val, dev); |
1da177e4 | 1445 | } |
edf947f1 | 1446 | EXPORT_SYMBOL(call_netdevice_notifiers); |
1da177e4 LT |
1447 | |
1448 | /* When > 0 there are consumers of rx skb time stamps */ | |
1449 | static atomic_t netstamp_needed = ATOMIC_INIT(0); | |
1450 | ||
1451 | void net_enable_timestamp(void) | |
1452 | { | |
1453 | atomic_inc(&netstamp_needed); | |
1454 | } | |
d1b19dff | 1455 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
1456 | |
1457 | void net_disable_timestamp(void) | |
1458 | { | |
1459 | atomic_dec(&netstamp_needed); | |
1460 | } | |
d1b19dff | 1461 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 1462 | |
3b098e2d | 1463 | static inline void net_timestamp_set(struct sk_buff *skb) |
1da177e4 LT |
1464 | { |
1465 | if (atomic_read(&netstamp_needed)) | |
a61bbcf2 | 1466 | __net_timestamp(skb); |
b7aa0bf7 ED |
1467 | else |
1468 | skb->tstamp.tv64 = 0; | |
1da177e4 LT |
1469 | } |
1470 | ||
3b098e2d ED |
1471 | static inline void net_timestamp_check(struct sk_buff *skb) |
1472 | { | |
1473 | if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed)) | |
1474 | __net_timestamp(skb); | |
1475 | } | |
1476 | ||
79b569f0 DL |
1477 | static inline bool is_skb_forwardable(struct net_device *dev, |
1478 | struct sk_buff *skb) | |
1479 | { | |
1480 | unsigned int len; | |
1481 | ||
1482 | if (!(dev->flags & IFF_UP)) | |
1483 | return false; | |
1484 | ||
1485 | len = dev->mtu + dev->hard_header_len + VLAN_HLEN; | |
1486 | if (skb->len <= len) | |
1487 | return true; | |
1488 | ||
1489 | /* if TSO is enabled, we don't care about the length as the packet | |
1490 | * could be forwarded without being segmented before | |
1491 | */ | |
1492 | if (skb_is_gso(skb)) | |
1493 | return true; | |
1494 | ||
1495 | return false; | |
1496 | } | |
1497 | ||
44540960 AB |
1498 | /** |
1499 | * dev_forward_skb - loopback an skb to another netif | |
1500 | * | |
1501 | * @dev: destination network device | |
1502 | * @skb: buffer to forward | |
1503 | * | |
1504 | * return values: | |
1505 | * NET_RX_SUCCESS (no congestion) | |
6ec82562 | 1506 | * NET_RX_DROP (packet was dropped, but freed) |
44540960 AB |
1507 | * |
1508 | * dev_forward_skb can be used for injecting an skb from the | |
1509 | * start_xmit function of one device into the receive queue | |
1510 | * of another device. | |
1511 | * | |
1512 | * The receiving device may be in another namespace, so | |
1513 | * we have to clear all information in the skb that could | |
1514 | * impact namespace isolation. | |
1515 | */ | |
1516 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
1517 | { | |
48c83012 MT |
1518 | if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) { |
1519 | if (skb_copy_ubufs(skb, GFP_ATOMIC)) { | |
1520 | atomic_long_inc(&dev->rx_dropped); | |
1521 | kfree_skb(skb); | |
1522 | return NET_RX_DROP; | |
1523 | } | |
1524 | } | |
1525 | ||
44540960 | 1526 | skb_orphan(skb); |
c736eefa | 1527 | nf_reset(skb); |
44540960 | 1528 | |
79b569f0 | 1529 | if (unlikely(!is_skb_forwardable(dev, skb))) { |
caf586e5 | 1530 | atomic_long_inc(&dev->rx_dropped); |
6ec82562 | 1531 | kfree_skb(skb); |
44540960 | 1532 | return NET_RX_DROP; |
6ec82562 | 1533 | } |
8a83a00b | 1534 | skb_set_dev(skb, dev); |
44540960 AB |
1535 | skb->tstamp.tv64 = 0; |
1536 | skb->pkt_type = PACKET_HOST; | |
1537 | skb->protocol = eth_type_trans(skb, dev); | |
44540960 AB |
1538 | return netif_rx(skb); |
1539 | } | |
1540 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
1541 | ||
71d9dec2 CG |
1542 | static inline int deliver_skb(struct sk_buff *skb, |
1543 | struct packet_type *pt_prev, | |
1544 | struct net_device *orig_dev) | |
1545 | { | |
1546 | atomic_inc(&skb->users); | |
1547 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1548 | } | |
1549 | ||
1da177e4 LT |
1550 | /* |
1551 | * Support routine. Sends outgoing frames to any network | |
1552 | * taps currently in use. | |
1553 | */ | |
1554 | ||
f6a78bfc | 1555 | static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
1556 | { |
1557 | struct packet_type *ptype; | |
71d9dec2 CG |
1558 | struct sk_buff *skb2 = NULL; |
1559 | struct packet_type *pt_prev = NULL; | |
a61bbcf2 | 1560 | |
1da177e4 LT |
1561 | rcu_read_lock(); |
1562 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
1563 | /* Never send packets back to the socket | |
1564 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1565 | */ | |
1566 | if ((ptype->dev == dev || !ptype->dev) && | |
1567 | (ptype->af_packet_priv == NULL || | |
1568 | (struct sock *)ptype->af_packet_priv != skb->sk)) { | |
71d9dec2 CG |
1569 | if (pt_prev) { |
1570 | deliver_skb(skb2, pt_prev, skb->dev); | |
1571 | pt_prev = ptype; | |
1572 | continue; | |
1573 | } | |
1574 | ||
1575 | skb2 = skb_clone(skb, GFP_ATOMIC); | |
1da177e4 LT |
1576 | if (!skb2) |
1577 | break; | |
1578 | ||
70978182 ED |
1579 | net_timestamp_set(skb2); |
1580 | ||
1da177e4 LT |
1581 | /* skb->nh should be correctly |
1582 | set by sender, so that the second statement is | |
1583 | just protection against buggy protocols. | |
1584 | */ | |
459a98ed | 1585 | skb_reset_mac_header(skb2); |
1da177e4 | 1586 | |
d56f90a7 | 1587 | if (skb_network_header(skb2) < skb2->data || |
27a884dc | 1588 | skb2->network_header > skb2->tail) { |
1da177e4 LT |
1589 | if (net_ratelimit()) |
1590 | printk(KERN_CRIT "protocol %04x is " | |
1591 | "buggy, dev %s\n", | |
70777d03 SAS |
1592 | ntohs(skb2->protocol), |
1593 | dev->name); | |
c1d2bbe1 | 1594 | skb_reset_network_header(skb2); |
1da177e4 LT |
1595 | } |
1596 | ||
b0e380b1 | 1597 | skb2->transport_header = skb2->network_header; |
1da177e4 | 1598 | skb2->pkt_type = PACKET_OUTGOING; |
71d9dec2 | 1599 | pt_prev = ptype; |
1da177e4 LT |
1600 | } |
1601 | } | |
71d9dec2 CG |
1602 | if (pt_prev) |
1603 | pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); | |
1da177e4 LT |
1604 | rcu_read_unlock(); |
1605 | } | |
1606 | ||
4f57c087 JF |
1607 | /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change |
1608 | * @dev: Network device | |
1609 | * @txq: number of queues available | |
1610 | * | |
1611 | * If real_num_tx_queues is changed the tc mappings may no longer be | |
1612 | * valid. To resolve this verify the tc mapping remains valid and if | |
1613 | * not NULL the mapping. With no priorities mapping to this | |
1614 | * offset/count pair it will no longer be used. In the worst case TC0 | |
1615 | * is invalid nothing can be done so disable priority mappings. If is | |
1616 | * expected that drivers will fix this mapping if they can before | |
1617 | * calling netif_set_real_num_tx_queues. | |
1618 | */ | |
bb134d22 | 1619 | static void netif_setup_tc(struct net_device *dev, unsigned int txq) |
4f57c087 JF |
1620 | { |
1621 | int i; | |
1622 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
1623 | ||
1624 | /* If TC0 is invalidated disable TC mapping */ | |
1625 | if (tc->offset + tc->count > txq) { | |
1626 | pr_warning("Number of in use tx queues changed " | |
1627 | "invalidating tc mappings. Priority " | |
1628 | "traffic classification disabled!\n"); | |
1629 | dev->num_tc = 0; | |
1630 | return; | |
1631 | } | |
1632 | ||
1633 | /* Invalidated prio to tc mappings set to TC0 */ | |
1634 | for (i = 1; i < TC_BITMASK + 1; i++) { | |
1635 | int q = netdev_get_prio_tc_map(dev, i); | |
1636 | ||
1637 | tc = &dev->tc_to_txq[q]; | |
1638 | if (tc->offset + tc->count > txq) { | |
1639 | pr_warning("Number of in use tx queues " | |
1640 | "changed. Priority %i to tc " | |
1641 | "mapping %i is no longer valid " | |
1642 | "setting map to 0\n", | |
1643 | i, q); | |
1644 | netdev_set_prio_tc_map(dev, i, 0); | |
1645 | } | |
1646 | } | |
1647 | } | |
1648 | ||
f0796d5c JF |
1649 | /* |
1650 | * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues | |
1651 | * greater then real_num_tx_queues stale skbs on the qdisc must be flushed. | |
1652 | */ | |
e6484930 | 1653 | int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) |
f0796d5c | 1654 | { |
1d24eb48 TH |
1655 | int rc; |
1656 | ||
e6484930 TH |
1657 | if (txq < 1 || txq > dev->num_tx_queues) |
1658 | return -EINVAL; | |
f0796d5c | 1659 | |
5c56580b BH |
1660 | if (dev->reg_state == NETREG_REGISTERED || |
1661 | dev->reg_state == NETREG_UNREGISTERING) { | |
e6484930 TH |
1662 | ASSERT_RTNL(); |
1663 | ||
1d24eb48 TH |
1664 | rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, |
1665 | txq); | |
bf264145 TH |
1666 | if (rc) |
1667 | return rc; | |
1668 | ||
4f57c087 JF |
1669 | if (dev->num_tc) |
1670 | netif_setup_tc(dev, txq); | |
1671 | ||
e6484930 TH |
1672 | if (txq < dev->real_num_tx_queues) |
1673 | qdisc_reset_all_tx_gt(dev, txq); | |
f0796d5c | 1674 | } |
e6484930 TH |
1675 | |
1676 | dev->real_num_tx_queues = txq; | |
1677 | return 0; | |
f0796d5c JF |
1678 | } |
1679 | EXPORT_SYMBOL(netif_set_real_num_tx_queues); | |
56079431 | 1680 | |
62fe0b40 BH |
1681 | #ifdef CONFIG_RPS |
1682 | /** | |
1683 | * netif_set_real_num_rx_queues - set actual number of RX queues used | |
1684 | * @dev: Network device | |
1685 | * @rxq: Actual number of RX queues | |
1686 | * | |
1687 | * This must be called either with the rtnl_lock held or before | |
1688 | * registration of the net device. Returns 0 on success, or a | |
4e7f7951 BH |
1689 | * negative error code. If called before registration, it always |
1690 | * succeeds. | |
62fe0b40 BH |
1691 | */ |
1692 | int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) | |
1693 | { | |
1694 | int rc; | |
1695 | ||
bd25fa7b TH |
1696 | if (rxq < 1 || rxq > dev->num_rx_queues) |
1697 | return -EINVAL; | |
1698 | ||
62fe0b40 BH |
1699 | if (dev->reg_state == NETREG_REGISTERED) { |
1700 | ASSERT_RTNL(); | |
1701 | ||
62fe0b40 BH |
1702 | rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, |
1703 | rxq); | |
1704 | if (rc) | |
1705 | return rc; | |
62fe0b40 BH |
1706 | } |
1707 | ||
1708 | dev->real_num_rx_queues = rxq; | |
1709 | return 0; | |
1710 | } | |
1711 | EXPORT_SYMBOL(netif_set_real_num_rx_queues); | |
1712 | #endif | |
1713 | ||
def82a1d | 1714 | static inline void __netif_reschedule(struct Qdisc *q) |
56079431 | 1715 | { |
def82a1d JP |
1716 | struct softnet_data *sd; |
1717 | unsigned long flags; | |
56079431 | 1718 | |
def82a1d JP |
1719 | local_irq_save(flags); |
1720 | sd = &__get_cpu_var(softnet_data); | |
a9cbd588 CG |
1721 | q->next_sched = NULL; |
1722 | *sd->output_queue_tailp = q; | |
1723 | sd->output_queue_tailp = &q->next_sched; | |
def82a1d JP |
1724 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
1725 | local_irq_restore(flags); | |
1726 | } | |
1727 | ||
1728 | void __netif_schedule(struct Qdisc *q) | |
1729 | { | |
1730 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
1731 | __netif_reschedule(q); | |
56079431 DV |
1732 | } |
1733 | EXPORT_SYMBOL(__netif_schedule); | |
1734 | ||
bea3348e | 1735 | void dev_kfree_skb_irq(struct sk_buff *skb) |
56079431 | 1736 | { |
3578b0c8 | 1737 | if (atomic_dec_and_test(&skb->users)) { |
bea3348e SH |
1738 | struct softnet_data *sd; |
1739 | unsigned long flags; | |
56079431 | 1740 | |
bea3348e SH |
1741 | local_irq_save(flags); |
1742 | sd = &__get_cpu_var(softnet_data); | |
1743 | skb->next = sd->completion_queue; | |
1744 | sd->completion_queue = skb; | |
1745 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
1746 | local_irq_restore(flags); | |
1747 | } | |
56079431 | 1748 | } |
bea3348e | 1749 | EXPORT_SYMBOL(dev_kfree_skb_irq); |
56079431 DV |
1750 | |
1751 | void dev_kfree_skb_any(struct sk_buff *skb) | |
1752 | { | |
1753 | if (in_irq() || irqs_disabled()) | |
1754 | dev_kfree_skb_irq(skb); | |
1755 | else | |
1756 | dev_kfree_skb(skb); | |
1757 | } | |
1758 | EXPORT_SYMBOL(dev_kfree_skb_any); | |
1759 | ||
1760 | ||
bea3348e SH |
1761 | /** |
1762 | * netif_device_detach - mark device as removed | |
1763 | * @dev: network device | |
1764 | * | |
1765 | * Mark device as removed from system and therefore no longer available. | |
1766 | */ | |
56079431 DV |
1767 | void netif_device_detach(struct net_device *dev) |
1768 | { | |
1769 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
1770 | netif_running(dev)) { | |
d543103a | 1771 | netif_tx_stop_all_queues(dev); |
56079431 DV |
1772 | } |
1773 | } | |
1774 | EXPORT_SYMBOL(netif_device_detach); | |
1775 | ||
bea3348e SH |
1776 | /** |
1777 | * netif_device_attach - mark device as attached | |
1778 | * @dev: network device | |
1779 | * | |
1780 | * Mark device as attached from system and restart if needed. | |
1781 | */ | |
56079431 DV |
1782 | void netif_device_attach(struct net_device *dev) |
1783 | { | |
1784 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
1785 | netif_running(dev)) { | |
d543103a | 1786 | netif_tx_wake_all_queues(dev); |
4ec93edb | 1787 | __netdev_watchdog_up(dev); |
56079431 DV |
1788 | } |
1789 | } | |
1790 | EXPORT_SYMBOL(netif_device_attach); | |
1791 | ||
8a83a00b AB |
1792 | /** |
1793 | * skb_dev_set -- assign a new device to a buffer | |
1794 | * @skb: buffer for the new device | |
1795 | * @dev: network device | |
1796 | * | |
1797 | * If an skb is owned by a device already, we have to reset | |
1798 | * all data private to the namespace a device belongs to | |
1799 | * before assigning it a new device. | |
1800 | */ | |
1801 | #ifdef CONFIG_NET_NS | |
1802 | void skb_set_dev(struct sk_buff *skb, struct net_device *dev) | |
1803 | { | |
1804 | skb_dst_drop(skb); | |
1805 | if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) { | |
1806 | secpath_reset(skb); | |
1807 | nf_reset(skb); | |
1808 | skb_init_secmark(skb); | |
1809 | skb->mark = 0; | |
1810 | skb->priority = 0; | |
1811 | skb->nf_trace = 0; | |
1812 | skb->ipvs_property = 0; | |
1813 | #ifdef CONFIG_NET_SCHED | |
1814 | skb->tc_index = 0; | |
1815 | #endif | |
1816 | } | |
1817 | skb->dev = dev; | |
1818 | } | |
1819 | EXPORT_SYMBOL(skb_set_dev); | |
1820 | #endif /* CONFIG_NET_NS */ | |
1821 | ||
1da177e4 LT |
1822 | /* |
1823 | * Invalidate hardware checksum when packet is to be mangled, and | |
1824 | * complete checksum manually on outgoing path. | |
1825 | */ | |
84fa7933 | 1826 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 1827 | { |
d3bc23e7 | 1828 | __wsum csum; |
663ead3b | 1829 | int ret = 0, offset; |
1da177e4 | 1830 | |
84fa7933 | 1831 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
1832 | goto out_set_summed; |
1833 | ||
1834 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
a430a43d HX |
1835 | /* Let GSO fix up the checksum. */ |
1836 | goto out_set_summed; | |
1da177e4 LT |
1837 | } |
1838 | ||
55508d60 | 1839 | offset = skb_checksum_start_offset(skb); |
a030847e HX |
1840 | BUG_ON(offset >= skb_headlen(skb)); |
1841 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
1842 | ||
1843 | offset += skb->csum_offset; | |
1844 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
1845 | ||
1846 | if (skb_cloned(skb) && | |
1847 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1da177e4 LT |
1848 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
1849 | if (ret) | |
1850 | goto out; | |
1851 | } | |
1852 | ||
a030847e | 1853 | *(__sum16 *)(skb->data + offset) = csum_fold(csum); |
a430a43d | 1854 | out_set_summed: |
1da177e4 | 1855 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 1856 | out: |
1da177e4 LT |
1857 | return ret; |
1858 | } | |
d1b19dff | 1859 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 1860 | |
f6a78bfc HX |
1861 | /** |
1862 | * skb_gso_segment - Perform segmentation on skb. | |
1863 | * @skb: buffer to segment | |
576a30eb | 1864 | * @features: features for the output path (see dev->features) |
f6a78bfc HX |
1865 | * |
1866 | * This function segments the given skb and returns a list of segments. | |
576a30eb HX |
1867 | * |
1868 | * It may return NULL if the skb requires no segmentation. This is | |
1869 | * only possible when GSO is used for verifying header integrity. | |
f6a78bfc | 1870 | */ |
04ed3e74 | 1871 | struct sk_buff *skb_gso_segment(struct sk_buff *skb, u32 features) |
f6a78bfc HX |
1872 | { |
1873 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
1874 | struct packet_type *ptype; | |
252e3346 | 1875 | __be16 type = skb->protocol; |
c8d5bcd1 | 1876 | int vlan_depth = ETH_HLEN; |
a430a43d | 1877 | int err; |
f6a78bfc | 1878 | |
c8d5bcd1 JG |
1879 | while (type == htons(ETH_P_8021Q)) { |
1880 | struct vlan_hdr *vh; | |
7b9c6090 | 1881 | |
c8d5bcd1 | 1882 | if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN))) |
7b9c6090 JG |
1883 | return ERR_PTR(-EINVAL); |
1884 | ||
c8d5bcd1 JG |
1885 | vh = (struct vlan_hdr *)(skb->data + vlan_depth); |
1886 | type = vh->h_vlan_encapsulated_proto; | |
1887 | vlan_depth += VLAN_HLEN; | |
7b9c6090 JG |
1888 | } |
1889 | ||
459a98ed | 1890 | skb_reset_mac_header(skb); |
b0e380b1 | 1891 | skb->mac_len = skb->network_header - skb->mac_header; |
f6a78bfc HX |
1892 | __skb_pull(skb, skb->mac_len); |
1893 | ||
67fd1a73 HX |
1894 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { |
1895 | struct net_device *dev = skb->dev; | |
1896 | struct ethtool_drvinfo info = {}; | |
1897 | ||
1898 | if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo) | |
1899 | dev->ethtool_ops->get_drvinfo(dev, &info); | |
1900 | ||
b194a367 | 1901 | WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n", |
67fd1a73 HX |
1902 | info.driver, dev ? dev->features : 0L, |
1903 | skb->sk ? skb->sk->sk_route_caps : 0L, | |
1904 | skb->len, skb->data_len, skb->ip_summed); | |
1905 | ||
a430a43d HX |
1906 | if (skb_header_cloned(skb) && |
1907 | (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))) | |
1908 | return ERR_PTR(err); | |
1909 | } | |
1910 | ||
f6a78bfc | 1911 | rcu_read_lock(); |
82d8a867 PE |
1912 | list_for_each_entry_rcu(ptype, |
1913 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
f6a78bfc | 1914 | if (ptype->type == type && !ptype->dev && ptype->gso_segment) { |
84fa7933 | 1915 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { |
a430a43d HX |
1916 | err = ptype->gso_send_check(skb); |
1917 | segs = ERR_PTR(err); | |
1918 | if (err || skb_gso_ok(skb, features)) | |
1919 | break; | |
d56f90a7 ACM |
1920 | __skb_push(skb, (skb->data - |
1921 | skb_network_header(skb))); | |
a430a43d | 1922 | } |
576a30eb | 1923 | segs = ptype->gso_segment(skb, features); |
f6a78bfc HX |
1924 | break; |
1925 | } | |
1926 | } | |
1927 | rcu_read_unlock(); | |
1928 | ||
98e399f8 | 1929 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 1930 | |
f6a78bfc HX |
1931 | return segs; |
1932 | } | |
f6a78bfc HX |
1933 | EXPORT_SYMBOL(skb_gso_segment); |
1934 | ||
fb286bb2 HX |
1935 | /* Take action when hardware reception checksum errors are detected. */ |
1936 | #ifdef CONFIG_BUG | |
1937 | void netdev_rx_csum_fault(struct net_device *dev) | |
1938 | { | |
1939 | if (net_ratelimit()) { | |
4ec93edb | 1940 | printk(KERN_ERR "%s: hw csum failure.\n", |
246a4212 | 1941 | dev ? dev->name : "<unknown>"); |
fb286bb2 HX |
1942 | dump_stack(); |
1943 | } | |
1944 | } | |
1945 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
1946 | #endif | |
1947 | ||
1da177e4 LT |
1948 | /* Actually, we should eliminate this check as soon as we know, that: |
1949 | * 1. IOMMU is present and allows to map all the memory. | |
1950 | * 2. No high memory really exists on this machine. | |
1951 | */ | |
1952 | ||
9092c658 | 1953 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
1da177e4 | 1954 | { |
3d3a8533 | 1955 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 1956 | int i; |
5acbbd42 FT |
1957 | if (!(dev->features & NETIF_F_HIGHDMA)) { |
1958 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) | |
1959 | if (PageHighMem(skb_shinfo(skb)->frags[i].page)) | |
1960 | return 1; | |
1961 | } | |
1da177e4 | 1962 | |
5acbbd42 FT |
1963 | if (PCI_DMA_BUS_IS_PHYS) { |
1964 | struct device *pdev = dev->dev.parent; | |
1da177e4 | 1965 | |
9092c658 ED |
1966 | if (!pdev) |
1967 | return 0; | |
5acbbd42 FT |
1968 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
1969 | dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page); | |
1970 | if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask) | |
1971 | return 1; | |
1972 | } | |
1973 | } | |
3d3a8533 | 1974 | #endif |
1da177e4 LT |
1975 | return 0; |
1976 | } | |
1da177e4 | 1977 | |
f6a78bfc HX |
1978 | struct dev_gso_cb { |
1979 | void (*destructor)(struct sk_buff *skb); | |
1980 | }; | |
1981 | ||
1982 | #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb) | |
1983 | ||
1984 | static void dev_gso_skb_destructor(struct sk_buff *skb) | |
1985 | { | |
1986 | struct dev_gso_cb *cb; | |
1987 | ||
1988 | do { | |
1989 | struct sk_buff *nskb = skb->next; | |
1990 | ||
1991 | skb->next = nskb->next; | |
1992 | nskb->next = NULL; | |
1993 | kfree_skb(nskb); | |
1994 | } while (skb->next); | |
1995 | ||
1996 | cb = DEV_GSO_CB(skb); | |
1997 | if (cb->destructor) | |
1998 | cb->destructor(skb); | |
1999 | } | |
2000 | ||
2001 | /** | |
2002 | * dev_gso_segment - Perform emulated hardware segmentation on skb. | |
2003 | * @skb: buffer to segment | |
91ecb63c | 2004 | * @features: device features as applicable to this skb |
f6a78bfc HX |
2005 | * |
2006 | * This function segments the given skb and stores the list of segments | |
2007 | * in skb->next. | |
2008 | */ | |
91ecb63c | 2009 | static int dev_gso_segment(struct sk_buff *skb, int features) |
f6a78bfc | 2010 | { |
f6a78bfc | 2011 | struct sk_buff *segs; |
576a30eb HX |
2012 | |
2013 | segs = skb_gso_segment(skb, features); | |
2014 | ||
2015 | /* Verifying header integrity only. */ | |
2016 | if (!segs) | |
2017 | return 0; | |
f6a78bfc | 2018 | |
801678c5 | 2019 | if (IS_ERR(segs)) |
f6a78bfc HX |
2020 | return PTR_ERR(segs); |
2021 | ||
2022 | skb->next = segs; | |
2023 | DEV_GSO_CB(skb)->destructor = skb->destructor; | |
2024 | skb->destructor = dev_gso_skb_destructor; | |
2025 | ||
2026 | return 0; | |
2027 | } | |
2028 | ||
fc6055a5 ED |
2029 | /* |
2030 | * Try to orphan skb early, right before transmission by the device. | |
2244d07b OH |
2031 | * We cannot orphan skb if tx timestamp is requested or the sk-reference |
2032 | * is needed on driver level for other reasons, e.g. see net/can/raw.c | |
fc6055a5 ED |
2033 | */ |
2034 | static inline void skb_orphan_try(struct sk_buff *skb) | |
2035 | { | |
87fd308c ED |
2036 | struct sock *sk = skb->sk; |
2037 | ||
2244d07b | 2038 | if (sk && !skb_shinfo(skb)->tx_flags) { |
87fd308c ED |
2039 | /* skb_tx_hash() wont be able to get sk. |
2040 | * We copy sk_hash into skb->rxhash | |
2041 | */ | |
2042 | if (!skb->rxhash) | |
2043 | skb->rxhash = sk->sk_hash; | |
fc6055a5 | 2044 | skb_orphan(skb); |
87fd308c | 2045 | } |
fc6055a5 ED |
2046 | } |
2047 | ||
03634668 JG |
2048 | static bool can_checksum_protocol(unsigned long features, __be16 protocol) |
2049 | { | |
2050 | return ((features & NETIF_F_GEN_CSUM) || | |
2051 | ((features & NETIF_F_V4_CSUM) && | |
2052 | protocol == htons(ETH_P_IP)) || | |
2053 | ((features & NETIF_F_V6_CSUM) && | |
2054 | protocol == htons(ETH_P_IPV6)) || | |
2055 | ((features & NETIF_F_FCOE_CRC) && | |
2056 | protocol == htons(ETH_P_FCOE))); | |
2057 | } | |
2058 | ||
04ed3e74 | 2059 | static u32 harmonize_features(struct sk_buff *skb, __be16 protocol, u32 features) |
f01a5236 | 2060 | { |
d402786e | 2061 | if (!can_checksum_protocol(features, protocol)) { |
f01a5236 JG |
2062 | features &= ~NETIF_F_ALL_CSUM; |
2063 | features &= ~NETIF_F_SG; | |
2064 | } else if (illegal_highdma(skb->dev, skb)) { | |
2065 | features &= ~NETIF_F_SG; | |
2066 | } | |
2067 | ||
2068 | return features; | |
2069 | } | |
2070 | ||
04ed3e74 | 2071 | u32 netif_skb_features(struct sk_buff *skb) |
58e998c6 JG |
2072 | { |
2073 | __be16 protocol = skb->protocol; | |
04ed3e74 | 2074 | u32 features = skb->dev->features; |
58e998c6 JG |
2075 | |
2076 | if (protocol == htons(ETH_P_8021Q)) { | |
2077 | struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data; | |
2078 | protocol = veh->h_vlan_encapsulated_proto; | |
f01a5236 JG |
2079 | } else if (!vlan_tx_tag_present(skb)) { |
2080 | return harmonize_features(skb, protocol, features); | |
2081 | } | |
58e998c6 | 2082 | |
6ee400aa | 2083 | features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX); |
f01a5236 JG |
2084 | |
2085 | if (protocol != htons(ETH_P_8021Q)) { | |
2086 | return harmonize_features(skb, protocol, features); | |
2087 | } else { | |
2088 | features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | | |
6ee400aa | 2089 | NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX; |
f01a5236 JG |
2090 | return harmonize_features(skb, protocol, features); |
2091 | } | |
58e998c6 | 2092 | } |
f01a5236 | 2093 | EXPORT_SYMBOL(netif_skb_features); |
58e998c6 | 2094 | |
6afff0ca JF |
2095 | /* |
2096 | * Returns true if either: | |
2097 | * 1. skb has frag_list and the device doesn't support FRAGLIST, or | |
2098 | * 2. skb is fragmented and the device does not support SG, or if | |
2099 | * at least one of fragments is in highmem and device does not | |
2100 | * support DMA from it. | |
2101 | */ | |
2102 | static inline int skb_needs_linearize(struct sk_buff *skb, | |
02932ce9 | 2103 | int features) |
6afff0ca | 2104 | { |
02932ce9 JG |
2105 | return skb_is_nonlinear(skb) && |
2106 | ((skb_has_frag_list(skb) && | |
2107 | !(features & NETIF_F_FRAGLIST)) || | |
e1e78db6 | 2108 | (skb_shinfo(skb)->nr_frags && |
02932ce9 | 2109 | !(features & NETIF_F_SG))); |
6afff0ca JF |
2110 | } |
2111 | ||
fd2ea0a7 DM |
2112 | int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev, |
2113 | struct netdev_queue *txq) | |
f6a78bfc | 2114 | { |
00829823 | 2115 | const struct net_device_ops *ops = dev->netdev_ops; |
572a9d7b | 2116 | int rc = NETDEV_TX_OK; |
ec764bf0 | 2117 | unsigned int skb_len; |
00829823 | 2118 | |
f6a78bfc | 2119 | if (likely(!skb->next)) { |
04ed3e74 | 2120 | u32 features; |
fc741216 | 2121 | |
93f154b5 | 2122 | /* |
25985edc | 2123 | * If device doesn't need skb->dst, release it right now while |
93f154b5 ED |
2124 | * its hot in this cpu cache |
2125 | */ | |
adf30907 ED |
2126 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) |
2127 | skb_dst_drop(skb); | |
2128 | ||
15c2d75f ED |
2129 | if (!list_empty(&ptype_all)) |
2130 | dev_queue_xmit_nit(skb, dev); | |
2131 | ||
fc6055a5 | 2132 | skb_orphan_try(skb); |
9ccb8975 | 2133 | |
fc741216 JG |
2134 | features = netif_skb_features(skb); |
2135 | ||
7b9c6090 | 2136 | if (vlan_tx_tag_present(skb) && |
fc741216 | 2137 | !(features & NETIF_F_HW_VLAN_TX)) { |
7b9c6090 JG |
2138 | skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb)); |
2139 | if (unlikely(!skb)) | |
2140 | goto out; | |
2141 | ||
2142 | skb->vlan_tci = 0; | |
2143 | } | |
2144 | ||
fc741216 | 2145 | if (netif_needs_gso(skb, features)) { |
91ecb63c | 2146 | if (unlikely(dev_gso_segment(skb, features))) |
9ccb8975 DM |
2147 | goto out_kfree_skb; |
2148 | if (skb->next) | |
2149 | goto gso; | |
6afff0ca | 2150 | } else { |
02932ce9 | 2151 | if (skb_needs_linearize(skb, features) && |
6afff0ca JF |
2152 | __skb_linearize(skb)) |
2153 | goto out_kfree_skb; | |
2154 | ||
2155 | /* If packet is not checksummed and device does not | |
2156 | * support checksumming for this protocol, complete | |
2157 | * checksumming here. | |
2158 | */ | |
2159 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
55508d60 MM |
2160 | skb_set_transport_header(skb, |
2161 | skb_checksum_start_offset(skb)); | |
03634668 | 2162 | if (!(features & NETIF_F_ALL_CSUM) && |
6afff0ca JF |
2163 | skb_checksum_help(skb)) |
2164 | goto out_kfree_skb; | |
2165 | } | |
9ccb8975 DM |
2166 | } |
2167 | ||
ec764bf0 | 2168 | skb_len = skb->len; |
ac45f602 | 2169 | rc = ops->ndo_start_xmit(skb, dev); |
ec764bf0 | 2170 | trace_net_dev_xmit(skb, rc, dev, skb_len); |
ec634fe3 | 2171 | if (rc == NETDEV_TX_OK) |
08baf561 | 2172 | txq_trans_update(txq); |
ac45f602 | 2173 | return rc; |
f6a78bfc HX |
2174 | } |
2175 | ||
576a30eb | 2176 | gso: |
f6a78bfc HX |
2177 | do { |
2178 | struct sk_buff *nskb = skb->next; | |
f6a78bfc HX |
2179 | |
2180 | skb->next = nskb->next; | |
2181 | nskb->next = NULL; | |
068a2de5 KK |
2182 | |
2183 | /* | |
25985edc | 2184 | * If device doesn't need nskb->dst, release it right now while |
068a2de5 KK |
2185 | * its hot in this cpu cache |
2186 | */ | |
2187 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
2188 | skb_dst_drop(nskb); | |
2189 | ||
ec764bf0 | 2190 | skb_len = nskb->len; |
00829823 | 2191 | rc = ops->ndo_start_xmit(nskb, dev); |
ec764bf0 | 2192 | trace_net_dev_xmit(nskb, rc, dev, skb_len); |
ec634fe3 | 2193 | if (unlikely(rc != NETDEV_TX_OK)) { |
572a9d7b PM |
2194 | if (rc & ~NETDEV_TX_MASK) |
2195 | goto out_kfree_gso_skb; | |
f54d9e8d | 2196 | nskb->next = skb->next; |
f6a78bfc HX |
2197 | skb->next = nskb; |
2198 | return rc; | |
2199 | } | |
08baf561 | 2200 | txq_trans_update(txq); |
fd2ea0a7 | 2201 | if (unlikely(netif_tx_queue_stopped(txq) && skb->next)) |
f54d9e8d | 2202 | return NETDEV_TX_BUSY; |
f6a78bfc | 2203 | } while (skb->next); |
4ec93edb | 2204 | |
572a9d7b PM |
2205 | out_kfree_gso_skb: |
2206 | if (likely(skb->next == NULL)) | |
2207 | skb->destructor = DEV_GSO_CB(skb)->destructor; | |
f6a78bfc HX |
2208 | out_kfree_skb: |
2209 | kfree_skb(skb); | |
7b9c6090 | 2210 | out: |
572a9d7b | 2211 | return rc; |
f6a78bfc HX |
2212 | } |
2213 | ||
0a9627f2 | 2214 | static u32 hashrnd __read_mostly; |
b6b2fed1 | 2215 | |
a3d22a68 VZ |
2216 | /* |
2217 | * Returns a Tx hash based on the given packet descriptor a Tx queues' number | |
2218 | * to be used as a distribution range. | |
2219 | */ | |
2220 | u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb, | |
2221 | unsigned int num_tx_queues) | |
8f0f2223 | 2222 | { |
7019298a | 2223 | u32 hash; |
4f57c087 JF |
2224 | u16 qoffset = 0; |
2225 | u16 qcount = num_tx_queues; | |
b6b2fed1 | 2226 | |
513de11b DM |
2227 | if (skb_rx_queue_recorded(skb)) { |
2228 | hash = skb_get_rx_queue(skb); | |
a3d22a68 VZ |
2229 | while (unlikely(hash >= num_tx_queues)) |
2230 | hash -= num_tx_queues; | |
513de11b DM |
2231 | return hash; |
2232 | } | |
ec581f6a | 2233 | |
4f57c087 JF |
2234 | if (dev->num_tc) { |
2235 | u8 tc = netdev_get_prio_tc_map(dev, skb->priority); | |
2236 | qoffset = dev->tc_to_txq[tc].offset; | |
2237 | qcount = dev->tc_to_txq[tc].count; | |
2238 | } | |
2239 | ||
ec581f6a | 2240 | if (skb->sk && skb->sk->sk_hash) |
7019298a | 2241 | hash = skb->sk->sk_hash; |
ec581f6a | 2242 | else |
87fd308c | 2243 | hash = (__force u16) skb->protocol ^ skb->rxhash; |
0a9627f2 | 2244 | hash = jhash_1word(hash, hashrnd); |
b6b2fed1 | 2245 | |
4f57c087 | 2246 | return (u16) (((u64) hash * qcount) >> 32) + qoffset; |
8f0f2223 | 2247 | } |
a3d22a68 | 2248 | EXPORT_SYMBOL(__skb_tx_hash); |
8f0f2223 | 2249 | |
ed04642f ED |
2250 | static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index) |
2251 | { | |
2252 | if (unlikely(queue_index >= dev->real_num_tx_queues)) { | |
2253 | if (net_ratelimit()) { | |
7a161ea9 ED |
2254 | pr_warning("%s selects TX queue %d, but " |
2255 | "real number of TX queues is %d\n", | |
2256 | dev->name, queue_index, dev->real_num_tx_queues); | |
ed04642f ED |
2257 | } |
2258 | return 0; | |
2259 | } | |
2260 | return queue_index; | |
2261 | } | |
2262 | ||
1d24eb48 TH |
2263 | static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb) |
2264 | { | |
bf264145 | 2265 | #ifdef CONFIG_XPS |
1d24eb48 TH |
2266 | struct xps_dev_maps *dev_maps; |
2267 | struct xps_map *map; | |
2268 | int queue_index = -1; | |
2269 | ||
2270 | rcu_read_lock(); | |
2271 | dev_maps = rcu_dereference(dev->xps_maps); | |
2272 | if (dev_maps) { | |
2273 | map = rcu_dereference( | |
2274 | dev_maps->cpu_map[raw_smp_processor_id()]); | |
2275 | if (map) { | |
2276 | if (map->len == 1) | |
2277 | queue_index = map->queues[0]; | |
2278 | else { | |
2279 | u32 hash; | |
2280 | if (skb->sk && skb->sk->sk_hash) | |
2281 | hash = skb->sk->sk_hash; | |
2282 | else | |
2283 | hash = (__force u16) skb->protocol ^ | |
2284 | skb->rxhash; | |
2285 | hash = jhash_1word(hash, hashrnd); | |
2286 | queue_index = map->queues[ | |
2287 | ((u64)hash * map->len) >> 32]; | |
2288 | } | |
2289 | if (unlikely(queue_index >= dev->real_num_tx_queues)) | |
2290 | queue_index = -1; | |
2291 | } | |
2292 | } | |
2293 | rcu_read_unlock(); | |
2294 | ||
2295 | return queue_index; | |
2296 | #else | |
2297 | return -1; | |
2298 | #endif | |
2299 | } | |
2300 | ||
e8a0464c DM |
2301 | static struct netdev_queue *dev_pick_tx(struct net_device *dev, |
2302 | struct sk_buff *skb) | |
2303 | { | |
b0f77d0e | 2304 | int queue_index; |
deabc772 | 2305 | const struct net_device_ops *ops = dev->netdev_ops; |
a4ee3ce3 | 2306 | |
3853b584 TH |
2307 | if (dev->real_num_tx_queues == 1) |
2308 | queue_index = 0; | |
2309 | else if (ops->ndo_select_queue) { | |
deabc772 HS |
2310 | queue_index = ops->ndo_select_queue(dev, skb); |
2311 | queue_index = dev_cap_txqueue(dev, queue_index); | |
2312 | } else { | |
2313 | struct sock *sk = skb->sk; | |
2314 | queue_index = sk_tx_queue_get(sk); | |
a4ee3ce3 | 2315 | |
3853b584 TH |
2316 | if (queue_index < 0 || skb->ooo_okay || |
2317 | queue_index >= dev->real_num_tx_queues) { | |
2318 | int old_index = queue_index; | |
fd2ea0a7 | 2319 | |
1d24eb48 TH |
2320 | queue_index = get_xps_queue(dev, skb); |
2321 | if (queue_index < 0) | |
2322 | queue_index = skb_tx_hash(dev, skb); | |
3853b584 TH |
2323 | |
2324 | if (queue_index != old_index && sk) { | |
2325 | struct dst_entry *dst = | |
2326 | rcu_dereference_check(sk->sk_dst_cache, 1); | |
8728c544 ED |
2327 | |
2328 | if (dst && skb_dst(skb) == dst) | |
2329 | sk_tx_queue_set(sk, queue_index); | |
2330 | } | |
a4ee3ce3 KK |
2331 | } |
2332 | } | |
eae792b7 | 2333 | |
fd2ea0a7 DM |
2334 | skb_set_queue_mapping(skb, queue_index); |
2335 | return netdev_get_tx_queue(dev, queue_index); | |
e8a0464c DM |
2336 | } |
2337 | ||
bbd8a0d3 KK |
2338 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
2339 | struct net_device *dev, | |
2340 | struct netdev_queue *txq) | |
2341 | { | |
2342 | spinlock_t *root_lock = qdisc_lock(q); | |
a2da570d | 2343 | bool contended; |
bbd8a0d3 KK |
2344 | int rc; |
2345 | ||
a2da570d ED |
2346 | qdisc_skb_cb(skb)->pkt_len = skb->len; |
2347 | qdisc_calculate_pkt_len(skb, q); | |
79640a4c ED |
2348 | /* |
2349 | * Heuristic to force contended enqueues to serialize on a | |
2350 | * separate lock before trying to get qdisc main lock. | |
2351 | * This permits __QDISC_STATE_RUNNING owner to get the lock more often | |
2352 | * and dequeue packets faster. | |
2353 | */ | |
a2da570d | 2354 | contended = qdisc_is_running(q); |
79640a4c ED |
2355 | if (unlikely(contended)) |
2356 | spin_lock(&q->busylock); | |
2357 | ||
bbd8a0d3 KK |
2358 | spin_lock(root_lock); |
2359 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
2360 | kfree_skb(skb); | |
2361 | rc = NET_XMIT_DROP; | |
2362 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
bc135b23 | 2363 | qdisc_run_begin(q)) { |
bbd8a0d3 KK |
2364 | /* |
2365 | * This is a work-conserving queue; there are no old skbs | |
2366 | * waiting to be sent out; and the qdisc is not running - | |
2367 | * xmit the skb directly. | |
2368 | */ | |
7fee226a ED |
2369 | if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE)) |
2370 | skb_dst_force(skb); | |
bfe0d029 | 2371 | |
bfe0d029 ED |
2372 | qdisc_bstats_update(q, skb); |
2373 | ||
79640a4c ED |
2374 | if (sch_direct_xmit(skb, q, dev, txq, root_lock)) { |
2375 | if (unlikely(contended)) { | |
2376 | spin_unlock(&q->busylock); | |
2377 | contended = false; | |
2378 | } | |
bbd8a0d3 | 2379 | __qdisc_run(q); |
79640a4c | 2380 | } else |
bc135b23 | 2381 | qdisc_run_end(q); |
bbd8a0d3 KK |
2382 | |
2383 | rc = NET_XMIT_SUCCESS; | |
2384 | } else { | |
7fee226a | 2385 | skb_dst_force(skb); |
a2da570d | 2386 | rc = q->enqueue(skb, q) & NET_XMIT_MASK; |
79640a4c ED |
2387 | if (qdisc_run_begin(q)) { |
2388 | if (unlikely(contended)) { | |
2389 | spin_unlock(&q->busylock); | |
2390 | contended = false; | |
2391 | } | |
2392 | __qdisc_run(q); | |
2393 | } | |
bbd8a0d3 KK |
2394 | } |
2395 | spin_unlock(root_lock); | |
79640a4c ED |
2396 | if (unlikely(contended)) |
2397 | spin_unlock(&q->busylock); | |
bbd8a0d3 KK |
2398 | return rc; |
2399 | } | |
2400 | ||
745e20f1 | 2401 | static DEFINE_PER_CPU(int, xmit_recursion); |
11a766ce | 2402 | #define RECURSION_LIMIT 10 |
745e20f1 | 2403 | |
d29f749e DJ |
2404 | /** |
2405 | * dev_queue_xmit - transmit a buffer | |
2406 | * @skb: buffer to transmit | |
2407 | * | |
2408 | * Queue a buffer for transmission to a network device. The caller must | |
2409 | * have set the device and priority and built the buffer before calling | |
2410 | * this function. The function can be called from an interrupt. | |
2411 | * | |
2412 | * A negative errno code is returned on a failure. A success does not | |
2413 | * guarantee the frame will be transmitted as it may be dropped due | |
2414 | * to congestion or traffic shaping. | |
2415 | * | |
2416 | * ----------------------------------------------------------------------------------- | |
2417 | * I notice this method can also return errors from the queue disciplines, | |
2418 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
2419 | * be positive. | |
2420 | * | |
2421 | * Regardless of the return value, the skb is consumed, so it is currently | |
2422 | * difficult to retry a send to this method. (You can bump the ref count | |
2423 | * before sending to hold a reference for retry if you are careful.) | |
2424 | * | |
2425 | * When calling this method, interrupts MUST be enabled. This is because | |
2426 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
2427 | * --BLG | |
2428 | */ | |
1da177e4 LT |
2429 | int dev_queue_xmit(struct sk_buff *skb) |
2430 | { | |
2431 | struct net_device *dev = skb->dev; | |
dc2b4847 | 2432 | struct netdev_queue *txq; |
1da177e4 LT |
2433 | struct Qdisc *q; |
2434 | int rc = -ENOMEM; | |
2435 | ||
4ec93edb YH |
2436 | /* Disable soft irqs for various locks below. Also |
2437 | * stops preemption for RCU. | |
1da177e4 | 2438 | */ |
4ec93edb | 2439 | rcu_read_lock_bh(); |
1da177e4 | 2440 | |
eae792b7 | 2441 | txq = dev_pick_tx(dev, skb); |
a898def2 | 2442 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 2443 | |
1da177e4 | 2444 | #ifdef CONFIG_NET_CLS_ACT |
d1b19dff | 2445 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS); |
1da177e4 | 2446 | #endif |
cf66ba58 | 2447 | trace_net_dev_queue(skb); |
1da177e4 | 2448 | if (q->enqueue) { |
bbd8a0d3 | 2449 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 2450 | goto out; |
1da177e4 LT |
2451 | } |
2452 | ||
2453 | /* The device has no queue. Common case for software devices: | |
2454 | loopback, all the sorts of tunnels... | |
2455 | ||
932ff279 HX |
2456 | Really, it is unlikely that netif_tx_lock protection is necessary |
2457 | here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
1da177e4 LT |
2458 | counters.) |
2459 | However, it is possible, that they rely on protection | |
2460 | made by us here. | |
2461 | ||
2462 | Check this and shot the lock. It is not prone from deadlocks. | |
2463 | Either shot noqueue qdisc, it is even simpler 8) | |
2464 | */ | |
2465 | if (dev->flags & IFF_UP) { | |
2466 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
2467 | ||
c773e847 | 2468 | if (txq->xmit_lock_owner != cpu) { |
1da177e4 | 2469 | |
745e20f1 ED |
2470 | if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT) |
2471 | goto recursion_alert; | |
2472 | ||
c773e847 | 2473 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 2474 | |
fd2ea0a7 | 2475 | if (!netif_tx_queue_stopped(txq)) { |
745e20f1 | 2476 | __this_cpu_inc(xmit_recursion); |
572a9d7b | 2477 | rc = dev_hard_start_xmit(skb, dev, txq); |
745e20f1 | 2478 | __this_cpu_dec(xmit_recursion); |
572a9d7b | 2479 | if (dev_xmit_complete(rc)) { |
c773e847 | 2480 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
2481 | goto out; |
2482 | } | |
2483 | } | |
c773e847 | 2484 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
2485 | if (net_ratelimit()) |
2486 | printk(KERN_CRIT "Virtual device %s asks to " | |
2487 | "queue packet!\n", dev->name); | |
2488 | } else { | |
2489 | /* Recursion is detected! It is possible, | |
745e20f1 ED |
2490 | * unfortunately |
2491 | */ | |
2492 | recursion_alert: | |
1da177e4 LT |
2493 | if (net_ratelimit()) |
2494 | printk(KERN_CRIT "Dead loop on virtual device " | |
2495 | "%s, fix it urgently!\n", dev->name); | |
2496 | } | |
2497 | } | |
2498 | ||
2499 | rc = -ENETDOWN; | |
d4828d85 | 2500 | rcu_read_unlock_bh(); |
1da177e4 | 2501 | |
1da177e4 LT |
2502 | kfree_skb(skb); |
2503 | return rc; | |
2504 | out: | |
d4828d85 | 2505 | rcu_read_unlock_bh(); |
1da177e4 LT |
2506 | return rc; |
2507 | } | |
d1b19dff | 2508 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 LT |
2509 | |
2510 | ||
2511 | /*======================================================================= | |
2512 | Receiver routines | |
2513 | =======================================================================*/ | |
2514 | ||
6b2bedc3 | 2515 | int netdev_max_backlog __read_mostly = 1000; |
3b098e2d | 2516 | int netdev_tstamp_prequeue __read_mostly = 1; |
6b2bedc3 SH |
2517 | int netdev_budget __read_mostly = 300; |
2518 | int weight_p __read_mostly = 64; /* old backlog weight */ | |
1da177e4 | 2519 | |
eecfd7c4 ED |
2520 | /* Called with irq disabled */ |
2521 | static inline void ____napi_schedule(struct softnet_data *sd, | |
2522 | struct napi_struct *napi) | |
2523 | { | |
2524 | list_add_tail(&napi->poll_list, &sd->poll_list); | |
2525 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2526 | } | |
2527 | ||
0a9627f2 | 2528 | /* |
bfb564e7 KK |
2529 | * __skb_get_rxhash: calculate a flow hash based on src/dst addresses |
2530 | * and src/dst port numbers. Returns a non-zero hash number on success | |
2531 | * and 0 on failure. | |
0a9627f2 | 2532 | */ |
bfb564e7 | 2533 | __u32 __skb_get_rxhash(struct sk_buff *skb) |
0a9627f2 | 2534 | { |
12fcdefb | 2535 | int nhoff, hash = 0, poff; |
b71d1d42 ED |
2536 | const struct ipv6hdr *ip6; |
2537 | const struct iphdr *ip; | |
0a9627f2 | 2538 | u8 ip_proto; |
8c52d509 CG |
2539 | u32 addr1, addr2, ihl; |
2540 | union { | |
2541 | u32 v32; | |
2542 | u16 v16[2]; | |
2543 | } ports; | |
0a9627f2 | 2544 | |
bfb564e7 | 2545 | nhoff = skb_network_offset(skb); |
0a9627f2 TH |
2546 | |
2547 | switch (skb->protocol) { | |
2548 | case __constant_htons(ETH_P_IP): | |
bfb564e7 | 2549 | if (!pskb_may_pull(skb, sizeof(*ip) + nhoff)) |
0a9627f2 TH |
2550 | goto done; |
2551 | ||
b71d1d42 | 2552 | ip = (const struct iphdr *) (skb->data + nhoff); |
56f8a75c | 2553 | if (ip_is_fragment(ip)) |
dbe5775b CG |
2554 | ip_proto = 0; |
2555 | else | |
2556 | ip_proto = ip->protocol; | |
b249dcb8 ED |
2557 | addr1 = (__force u32) ip->saddr; |
2558 | addr2 = (__force u32) ip->daddr; | |
0a9627f2 TH |
2559 | ihl = ip->ihl; |
2560 | break; | |
2561 | case __constant_htons(ETH_P_IPV6): | |
bfb564e7 | 2562 | if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff)) |
0a9627f2 TH |
2563 | goto done; |
2564 | ||
b71d1d42 | 2565 | ip6 = (const struct ipv6hdr *) (skb->data + nhoff); |
0a9627f2 | 2566 | ip_proto = ip6->nexthdr; |
b249dcb8 ED |
2567 | addr1 = (__force u32) ip6->saddr.s6_addr32[3]; |
2568 | addr2 = (__force u32) ip6->daddr.s6_addr32[3]; | |
0a9627f2 TH |
2569 | ihl = (40 >> 2); |
2570 | break; | |
2571 | default: | |
2572 | goto done; | |
2573 | } | |
bfb564e7 | 2574 | |
12fcdefb CG |
2575 | ports.v32 = 0; |
2576 | poff = proto_ports_offset(ip_proto); | |
2577 | if (poff >= 0) { | |
2578 | nhoff += ihl * 4 + poff; | |
2579 | if (pskb_may_pull(skb, nhoff + 4)) { | |
2580 | ports.v32 = * (__force u32 *) (skb->data + nhoff); | |
8c52d509 CG |
2581 | if (ports.v16[1] < ports.v16[0]) |
2582 | swap(ports.v16[0], ports.v16[1]); | |
b249dcb8 | 2583 | } |
0a9627f2 TH |
2584 | } |
2585 | ||
b249dcb8 ED |
2586 | /* get a consistent hash (same value on both flow directions) */ |
2587 | if (addr2 < addr1) | |
2588 | swap(addr1, addr2); | |
0a9627f2 | 2589 | |
bfb564e7 KK |
2590 | hash = jhash_3words(addr1, addr2, ports.v32, hashrnd); |
2591 | if (!hash) | |
2592 | hash = 1; | |
2593 | ||
2594 | done: | |
2595 | return hash; | |
2596 | } | |
2597 | EXPORT_SYMBOL(__skb_get_rxhash); | |
2598 | ||
2599 | #ifdef CONFIG_RPS | |
2600 | ||
2601 | /* One global table that all flow-based protocols share. */ | |
6e3f7faf | 2602 | struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; |
bfb564e7 KK |
2603 | EXPORT_SYMBOL(rps_sock_flow_table); |
2604 | ||
c445477d BH |
2605 | static struct rps_dev_flow * |
2606 | set_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
2607 | struct rps_dev_flow *rflow, u16 next_cpu) | |
2608 | { | |
2609 | u16 tcpu; | |
2610 | ||
2611 | tcpu = rflow->cpu = next_cpu; | |
2612 | if (tcpu != RPS_NO_CPU) { | |
2613 | #ifdef CONFIG_RFS_ACCEL | |
2614 | struct netdev_rx_queue *rxqueue; | |
2615 | struct rps_dev_flow_table *flow_table; | |
2616 | struct rps_dev_flow *old_rflow; | |
2617 | u32 flow_id; | |
2618 | u16 rxq_index; | |
2619 | int rc; | |
2620 | ||
2621 | /* Should we steer this flow to a different hardware queue? */ | |
69a19ee6 BH |
2622 | if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || |
2623 | !(dev->features & NETIF_F_NTUPLE)) | |
c445477d BH |
2624 | goto out; |
2625 | rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); | |
2626 | if (rxq_index == skb_get_rx_queue(skb)) | |
2627 | goto out; | |
2628 | ||
2629 | rxqueue = dev->_rx + rxq_index; | |
2630 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
2631 | if (!flow_table) | |
2632 | goto out; | |
2633 | flow_id = skb->rxhash & flow_table->mask; | |
2634 | rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, | |
2635 | rxq_index, flow_id); | |
2636 | if (rc < 0) | |
2637 | goto out; | |
2638 | old_rflow = rflow; | |
2639 | rflow = &flow_table->flows[flow_id]; | |
2640 | rflow->cpu = next_cpu; | |
2641 | rflow->filter = rc; | |
2642 | if (old_rflow->filter == rflow->filter) | |
2643 | old_rflow->filter = RPS_NO_FILTER; | |
2644 | out: | |
2645 | #endif | |
2646 | rflow->last_qtail = | |
2647 | per_cpu(softnet_data, tcpu).input_queue_head; | |
2648 | } | |
2649 | ||
2650 | return rflow; | |
2651 | } | |
2652 | ||
bfb564e7 KK |
2653 | /* |
2654 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
2655 | * CPU from the RPS map of the receiving queue for a given skb. | |
2656 | * rcu_read_lock must be held on entry. | |
2657 | */ | |
2658 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
2659 | struct rps_dev_flow **rflowp) | |
2660 | { | |
2661 | struct netdev_rx_queue *rxqueue; | |
6e3f7faf | 2662 | struct rps_map *map; |
bfb564e7 KK |
2663 | struct rps_dev_flow_table *flow_table; |
2664 | struct rps_sock_flow_table *sock_flow_table; | |
2665 | int cpu = -1; | |
2666 | u16 tcpu; | |
2667 | ||
2668 | if (skb_rx_queue_recorded(skb)) { | |
2669 | u16 index = skb_get_rx_queue(skb); | |
62fe0b40 BH |
2670 | if (unlikely(index >= dev->real_num_rx_queues)) { |
2671 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
2672 | "%s received packet on queue %u, but number " | |
2673 | "of RX queues is %u\n", | |
2674 | dev->name, index, dev->real_num_rx_queues); | |
bfb564e7 KK |
2675 | goto done; |
2676 | } | |
2677 | rxqueue = dev->_rx + index; | |
2678 | } else | |
2679 | rxqueue = dev->_rx; | |
2680 | ||
6e3f7faf ED |
2681 | map = rcu_dereference(rxqueue->rps_map); |
2682 | if (map) { | |
85875236 TH |
2683 | if (map->len == 1 && |
2684 | !rcu_dereference_raw(rxqueue->rps_flow_table)) { | |
6febfca9 CG |
2685 | tcpu = map->cpus[0]; |
2686 | if (cpu_online(tcpu)) | |
2687 | cpu = tcpu; | |
2688 | goto done; | |
2689 | } | |
6e3f7faf | 2690 | } else if (!rcu_dereference_raw(rxqueue->rps_flow_table)) { |
bfb564e7 | 2691 | goto done; |
6febfca9 | 2692 | } |
bfb564e7 | 2693 | |
2d47b459 | 2694 | skb_reset_network_header(skb); |
bfb564e7 KK |
2695 | if (!skb_get_rxhash(skb)) |
2696 | goto done; | |
2697 | ||
fec5e652 TH |
2698 | flow_table = rcu_dereference(rxqueue->rps_flow_table); |
2699 | sock_flow_table = rcu_dereference(rps_sock_flow_table); | |
2700 | if (flow_table && sock_flow_table) { | |
2701 | u16 next_cpu; | |
2702 | struct rps_dev_flow *rflow; | |
2703 | ||
2704 | rflow = &flow_table->flows[skb->rxhash & flow_table->mask]; | |
2705 | tcpu = rflow->cpu; | |
2706 | ||
2707 | next_cpu = sock_flow_table->ents[skb->rxhash & | |
2708 | sock_flow_table->mask]; | |
2709 | ||
2710 | /* | |
2711 | * If the desired CPU (where last recvmsg was done) is | |
2712 | * different from current CPU (one in the rx-queue flow | |
2713 | * table entry), switch if one of the following holds: | |
2714 | * - Current CPU is unset (equal to RPS_NO_CPU). | |
2715 | * - Current CPU is offline. | |
2716 | * - The current CPU's queue tail has advanced beyond the | |
2717 | * last packet that was enqueued using this table entry. | |
2718 | * This guarantees that all previous packets for the flow | |
2719 | * have been dequeued, thus preserving in order delivery. | |
2720 | */ | |
2721 | if (unlikely(tcpu != next_cpu) && | |
2722 | (tcpu == RPS_NO_CPU || !cpu_online(tcpu) || | |
2723 | ((int)(per_cpu(softnet_data, tcpu).input_queue_head - | |
c445477d BH |
2724 | rflow->last_qtail)) >= 0)) |
2725 | rflow = set_rps_cpu(dev, skb, rflow, next_cpu); | |
2726 | ||
fec5e652 TH |
2727 | if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) { |
2728 | *rflowp = rflow; | |
2729 | cpu = tcpu; | |
2730 | goto done; | |
2731 | } | |
2732 | } | |
2733 | ||
0a9627f2 | 2734 | if (map) { |
fec5e652 | 2735 | tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32]; |
0a9627f2 TH |
2736 | |
2737 | if (cpu_online(tcpu)) { | |
2738 | cpu = tcpu; | |
2739 | goto done; | |
2740 | } | |
2741 | } | |
2742 | ||
2743 | done: | |
0a9627f2 TH |
2744 | return cpu; |
2745 | } | |
2746 | ||
c445477d BH |
2747 | #ifdef CONFIG_RFS_ACCEL |
2748 | ||
2749 | /** | |
2750 | * rps_may_expire_flow - check whether an RFS hardware filter may be removed | |
2751 | * @dev: Device on which the filter was set | |
2752 | * @rxq_index: RX queue index | |
2753 | * @flow_id: Flow ID passed to ndo_rx_flow_steer() | |
2754 | * @filter_id: Filter ID returned by ndo_rx_flow_steer() | |
2755 | * | |
2756 | * Drivers that implement ndo_rx_flow_steer() should periodically call | |
2757 | * this function for each installed filter and remove the filters for | |
2758 | * which it returns %true. | |
2759 | */ | |
2760 | bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, | |
2761 | u32 flow_id, u16 filter_id) | |
2762 | { | |
2763 | struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; | |
2764 | struct rps_dev_flow_table *flow_table; | |
2765 | struct rps_dev_flow *rflow; | |
2766 | bool expire = true; | |
2767 | int cpu; | |
2768 | ||
2769 | rcu_read_lock(); | |
2770 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
2771 | if (flow_table && flow_id <= flow_table->mask) { | |
2772 | rflow = &flow_table->flows[flow_id]; | |
2773 | cpu = ACCESS_ONCE(rflow->cpu); | |
2774 | if (rflow->filter == filter_id && cpu != RPS_NO_CPU && | |
2775 | ((int)(per_cpu(softnet_data, cpu).input_queue_head - | |
2776 | rflow->last_qtail) < | |
2777 | (int)(10 * flow_table->mask))) | |
2778 | expire = false; | |
2779 | } | |
2780 | rcu_read_unlock(); | |
2781 | return expire; | |
2782 | } | |
2783 | EXPORT_SYMBOL(rps_may_expire_flow); | |
2784 | ||
2785 | #endif /* CONFIG_RFS_ACCEL */ | |
2786 | ||
0a9627f2 | 2787 | /* Called from hardirq (IPI) context */ |
e36fa2f7 | 2788 | static void rps_trigger_softirq(void *data) |
0a9627f2 | 2789 | { |
e36fa2f7 ED |
2790 | struct softnet_data *sd = data; |
2791 | ||
eecfd7c4 | 2792 | ____napi_schedule(sd, &sd->backlog); |
dee42870 | 2793 | sd->received_rps++; |
0a9627f2 | 2794 | } |
e36fa2f7 | 2795 | |
fec5e652 | 2796 | #endif /* CONFIG_RPS */ |
0a9627f2 | 2797 | |
e36fa2f7 ED |
2798 | /* |
2799 | * Check if this softnet_data structure is another cpu one | |
2800 | * If yes, queue it to our IPI list and return 1 | |
2801 | * If no, return 0 | |
2802 | */ | |
2803 | static int rps_ipi_queued(struct softnet_data *sd) | |
2804 | { | |
2805 | #ifdef CONFIG_RPS | |
2806 | struct softnet_data *mysd = &__get_cpu_var(softnet_data); | |
2807 | ||
2808 | if (sd != mysd) { | |
2809 | sd->rps_ipi_next = mysd->rps_ipi_list; | |
2810 | mysd->rps_ipi_list = sd; | |
2811 | ||
2812 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2813 | return 1; | |
2814 | } | |
2815 | #endif /* CONFIG_RPS */ | |
2816 | return 0; | |
2817 | } | |
2818 | ||
0a9627f2 TH |
2819 | /* |
2820 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
2821 | * queue (may be a remote CPU queue). | |
2822 | */ | |
fec5e652 TH |
2823 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, |
2824 | unsigned int *qtail) | |
0a9627f2 | 2825 | { |
e36fa2f7 | 2826 | struct softnet_data *sd; |
0a9627f2 TH |
2827 | unsigned long flags; |
2828 | ||
e36fa2f7 | 2829 | sd = &per_cpu(softnet_data, cpu); |
0a9627f2 TH |
2830 | |
2831 | local_irq_save(flags); | |
0a9627f2 | 2832 | |
e36fa2f7 | 2833 | rps_lock(sd); |
6e7676c1 CG |
2834 | if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) { |
2835 | if (skb_queue_len(&sd->input_pkt_queue)) { | |
0a9627f2 | 2836 | enqueue: |
e36fa2f7 | 2837 | __skb_queue_tail(&sd->input_pkt_queue, skb); |
76cc8b13 | 2838 | input_queue_tail_incr_save(sd, qtail); |
e36fa2f7 | 2839 | rps_unlock(sd); |
152102c7 | 2840 | local_irq_restore(flags); |
0a9627f2 TH |
2841 | return NET_RX_SUCCESS; |
2842 | } | |
2843 | ||
ebda37c2 ED |
2844 | /* Schedule NAPI for backlog device |
2845 | * We can use non atomic operation since we own the queue lock | |
2846 | */ | |
2847 | if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { | |
e36fa2f7 | 2848 | if (!rps_ipi_queued(sd)) |
eecfd7c4 | 2849 | ____napi_schedule(sd, &sd->backlog); |
0a9627f2 TH |
2850 | } |
2851 | goto enqueue; | |
2852 | } | |
2853 | ||
dee42870 | 2854 | sd->dropped++; |
e36fa2f7 | 2855 | rps_unlock(sd); |
0a9627f2 | 2856 | |
0a9627f2 TH |
2857 | local_irq_restore(flags); |
2858 | ||
caf586e5 | 2859 | atomic_long_inc(&skb->dev->rx_dropped); |
0a9627f2 TH |
2860 | kfree_skb(skb); |
2861 | return NET_RX_DROP; | |
2862 | } | |
1da177e4 | 2863 | |
1da177e4 LT |
2864 | /** |
2865 | * netif_rx - post buffer to the network code | |
2866 | * @skb: buffer to post | |
2867 | * | |
2868 | * This function receives a packet from a device driver and queues it for | |
2869 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
2870 | * may be dropped during processing for congestion control or by the | |
2871 | * protocol layers. | |
2872 | * | |
2873 | * return values: | |
2874 | * NET_RX_SUCCESS (no congestion) | |
1da177e4 LT |
2875 | * NET_RX_DROP (packet was dropped) |
2876 | * | |
2877 | */ | |
2878 | ||
2879 | int netif_rx(struct sk_buff *skb) | |
2880 | { | |
b0e28f1e | 2881 | int ret; |
1da177e4 LT |
2882 | |
2883 | /* if netpoll wants it, pretend we never saw it */ | |
2884 | if (netpoll_rx(skb)) | |
2885 | return NET_RX_DROP; | |
2886 | ||
3b098e2d ED |
2887 | if (netdev_tstamp_prequeue) |
2888 | net_timestamp_check(skb); | |
1da177e4 | 2889 | |
cf66ba58 | 2890 | trace_netif_rx(skb); |
df334545 | 2891 | #ifdef CONFIG_RPS |
b0e28f1e | 2892 | { |
fec5e652 | 2893 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
b0e28f1e ED |
2894 | int cpu; |
2895 | ||
cece1945 | 2896 | preempt_disable(); |
b0e28f1e | 2897 | rcu_read_lock(); |
fec5e652 TH |
2898 | |
2899 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
b0e28f1e ED |
2900 | if (cpu < 0) |
2901 | cpu = smp_processor_id(); | |
fec5e652 TH |
2902 | |
2903 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
2904 | ||
b0e28f1e | 2905 | rcu_read_unlock(); |
cece1945 | 2906 | preempt_enable(); |
b0e28f1e | 2907 | } |
1e94d72f | 2908 | #else |
fec5e652 TH |
2909 | { |
2910 | unsigned int qtail; | |
2911 | ret = enqueue_to_backlog(skb, get_cpu(), &qtail); | |
2912 | put_cpu(); | |
2913 | } | |
1e94d72f | 2914 | #endif |
b0e28f1e | 2915 | return ret; |
1da177e4 | 2916 | } |
d1b19dff | 2917 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
2918 | |
2919 | int netif_rx_ni(struct sk_buff *skb) | |
2920 | { | |
2921 | int err; | |
2922 | ||
2923 | preempt_disable(); | |
2924 | err = netif_rx(skb); | |
2925 | if (local_softirq_pending()) | |
2926 | do_softirq(); | |
2927 | preempt_enable(); | |
2928 | ||
2929 | return err; | |
2930 | } | |
1da177e4 LT |
2931 | EXPORT_SYMBOL(netif_rx_ni); |
2932 | ||
1da177e4 LT |
2933 | static void net_tx_action(struct softirq_action *h) |
2934 | { | |
2935 | struct softnet_data *sd = &__get_cpu_var(softnet_data); | |
2936 | ||
2937 | if (sd->completion_queue) { | |
2938 | struct sk_buff *clist; | |
2939 | ||
2940 | local_irq_disable(); | |
2941 | clist = sd->completion_queue; | |
2942 | sd->completion_queue = NULL; | |
2943 | local_irq_enable(); | |
2944 | ||
2945 | while (clist) { | |
2946 | struct sk_buff *skb = clist; | |
2947 | clist = clist->next; | |
2948 | ||
547b792c | 2949 | WARN_ON(atomic_read(&skb->users)); |
07dc22e7 | 2950 | trace_kfree_skb(skb, net_tx_action); |
1da177e4 LT |
2951 | __kfree_skb(skb); |
2952 | } | |
2953 | } | |
2954 | ||
2955 | if (sd->output_queue) { | |
37437bb2 | 2956 | struct Qdisc *head; |
1da177e4 LT |
2957 | |
2958 | local_irq_disable(); | |
2959 | head = sd->output_queue; | |
2960 | sd->output_queue = NULL; | |
a9cbd588 | 2961 | sd->output_queue_tailp = &sd->output_queue; |
1da177e4 LT |
2962 | local_irq_enable(); |
2963 | ||
2964 | while (head) { | |
37437bb2 DM |
2965 | struct Qdisc *q = head; |
2966 | spinlock_t *root_lock; | |
2967 | ||
1da177e4 LT |
2968 | head = head->next_sched; |
2969 | ||
5fb66229 | 2970 | root_lock = qdisc_lock(q); |
37437bb2 | 2971 | if (spin_trylock(root_lock)) { |
def82a1d JP |
2972 | smp_mb__before_clear_bit(); |
2973 | clear_bit(__QDISC_STATE_SCHED, | |
2974 | &q->state); | |
37437bb2 DM |
2975 | qdisc_run(q); |
2976 | spin_unlock(root_lock); | |
1da177e4 | 2977 | } else { |
195648bb | 2978 | if (!test_bit(__QDISC_STATE_DEACTIVATED, |
e8a83e10 | 2979 | &q->state)) { |
195648bb | 2980 | __netif_reschedule(q); |
e8a83e10 JP |
2981 | } else { |
2982 | smp_mb__before_clear_bit(); | |
2983 | clear_bit(__QDISC_STATE_SCHED, | |
2984 | &q->state); | |
2985 | } | |
1da177e4 LT |
2986 | } |
2987 | } | |
2988 | } | |
2989 | } | |
2990 | ||
ab95bfe0 JP |
2991 | #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \ |
2992 | (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)) | |
da678292 MM |
2993 | /* This hook is defined here for ATM LANE */ |
2994 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
2995 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 2996 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 2997 | #endif |
1da177e4 | 2998 | |
1da177e4 LT |
2999 | #ifdef CONFIG_NET_CLS_ACT |
3000 | /* TODO: Maybe we should just force sch_ingress to be compiled in | |
3001 | * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions | |
3002 | * a compare and 2 stores extra right now if we dont have it on | |
3003 | * but have CONFIG_NET_CLS_ACT | |
25985edc LDM |
3004 | * NOTE: This doesn't stop any functionality; if you dont have |
3005 | * the ingress scheduler, you just can't add policies on ingress. | |
1da177e4 LT |
3006 | * |
3007 | */ | |
24824a09 | 3008 | static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq) |
1da177e4 | 3009 | { |
1da177e4 | 3010 | struct net_device *dev = skb->dev; |
f697c3e8 | 3011 | u32 ttl = G_TC_RTTL(skb->tc_verd); |
555353cf DM |
3012 | int result = TC_ACT_OK; |
3013 | struct Qdisc *q; | |
4ec93edb | 3014 | |
de384830 SH |
3015 | if (unlikely(MAX_RED_LOOP < ttl++)) { |
3016 | if (net_ratelimit()) | |
3017 | pr_warning( "Redir loop detected Dropping packet (%d->%d)\n", | |
3018 | skb->skb_iif, dev->ifindex); | |
f697c3e8 HX |
3019 | return TC_ACT_SHOT; |
3020 | } | |
1da177e4 | 3021 | |
f697c3e8 HX |
3022 | skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl); |
3023 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); | |
1da177e4 | 3024 | |
83874000 | 3025 | q = rxq->qdisc; |
8d50b53d | 3026 | if (q != &noop_qdisc) { |
83874000 | 3027 | spin_lock(qdisc_lock(q)); |
a9312ae8 DM |
3028 | if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) |
3029 | result = qdisc_enqueue_root(skb, q); | |
83874000 DM |
3030 | spin_unlock(qdisc_lock(q)); |
3031 | } | |
f697c3e8 HX |
3032 | |
3033 | return result; | |
3034 | } | |
86e65da9 | 3035 | |
f697c3e8 HX |
3036 | static inline struct sk_buff *handle_ing(struct sk_buff *skb, |
3037 | struct packet_type **pt_prev, | |
3038 | int *ret, struct net_device *orig_dev) | |
3039 | { | |
24824a09 ED |
3040 | struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue); |
3041 | ||
3042 | if (!rxq || rxq->qdisc == &noop_qdisc) | |
f697c3e8 | 3043 | goto out; |
1da177e4 | 3044 | |
f697c3e8 HX |
3045 | if (*pt_prev) { |
3046 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
3047 | *pt_prev = NULL; | |
1da177e4 LT |
3048 | } |
3049 | ||
24824a09 | 3050 | switch (ing_filter(skb, rxq)) { |
f697c3e8 HX |
3051 | case TC_ACT_SHOT: |
3052 | case TC_ACT_STOLEN: | |
3053 | kfree_skb(skb); | |
3054 | return NULL; | |
3055 | } | |
3056 | ||
3057 | out: | |
3058 | skb->tc_verd = 0; | |
3059 | return skb; | |
1da177e4 LT |
3060 | } |
3061 | #endif | |
3062 | ||
ab95bfe0 JP |
3063 | /** |
3064 | * netdev_rx_handler_register - register receive handler | |
3065 | * @dev: device to register a handler for | |
3066 | * @rx_handler: receive handler to register | |
93e2c32b | 3067 | * @rx_handler_data: data pointer that is used by rx handler |
ab95bfe0 JP |
3068 | * |
3069 | * Register a receive hander for a device. This handler will then be | |
3070 | * called from __netif_receive_skb. A negative errno code is returned | |
3071 | * on a failure. | |
3072 | * | |
3073 | * The caller must hold the rtnl_mutex. | |
8a4eb573 JP |
3074 | * |
3075 | * For a general description of rx_handler, see enum rx_handler_result. | |
ab95bfe0 JP |
3076 | */ |
3077 | int netdev_rx_handler_register(struct net_device *dev, | |
93e2c32b JP |
3078 | rx_handler_func_t *rx_handler, |
3079 | void *rx_handler_data) | |
ab95bfe0 JP |
3080 | { |
3081 | ASSERT_RTNL(); | |
3082 | ||
3083 | if (dev->rx_handler) | |
3084 | return -EBUSY; | |
3085 | ||
93e2c32b | 3086 | rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); |
ab95bfe0 JP |
3087 | rcu_assign_pointer(dev->rx_handler, rx_handler); |
3088 | ||
3089 | return 0; | |
3090 | } | |
3091 | EXPORT_SYMBOL_GPL(netdev_rx_handler_register); | |
3092 | ||
3093 | /** | |
3094 | * netdev_rx_handler_unregister - unregister receive handler | |
3095 | * @dev: device to unregister a handler from | |
3096 | * | |
3097 | * Unregister a receive hander from a device. | |
3098 | * | |
3099 | * The caller must hold the rtnl_mutex. | |
3100 | */ | |
3101 | void netdev_rx_handler_unregister(struct net_device *dev) | |
3102 | { | |
3103 | ||
3104 | ASSERT_RTNL(); | |
3105 | rcu_assign_pointer(dev->rx_handler, NULL); | |
93e2c32b | 3106 | rcu_assign_pointer(dev->rx_handler_data, NULL); |
ab95bfe0 JP |
3107 | } |
3108 | EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); | |
3109 | ||
10f744d2 | 3110 | static int __netif_receive_skb(struct sk_buff *skb) |
1da177e4 LT |
3111 | { |
3112 | struct packet_type *ptype, *pt_prev; | |
ab95bfe0 | 3113 | rx_handler_func_t *rx_handler; |
f2ccd8fa | 3114 | struct net_device *orig_dev; |
63d8ea7f | 3115 | struct net_device *null_or_dev; |
8a4eb573 | 3116 | bool deliver_exact = false; |
1da177e4 | 3117 | int ret = NET_RX_DROP; |
252e3346 | 3118 | __be16 type; |
1da177e4 | 3119 | |
3b098e2d ED |
3120 | if (!netdev_tstamp_prequeue) |
3121 | net_timestamp_check(skb); | |
81bbb3d4 | 3122 | |
cf66ba58 | 3123 | trace_netif_receive_skb(skb); |
9b22ea56 | 3124 | |
1da177e4 | 3125 | /* if we've gotten here through NAPI, check netpoll */ |
bea3348e | 3126 | if (netpoll_receive_skb(skb)) |
1da177e4 LT |
3127 | return NET_RX_DROP; |
3128 | ||
8964be4a ED |
3129 | if (!skb->skb_iif) |
3130 | skb->skb_iif = skb->dev->ifindex; | |
cc9bd5ce | 3131 | orig_dev = skb->dev; |
8f903c70 | 3132 | |
c1d2bbe1 | 3133 | skb_reset_network_header(skb); |
badff6d0 | 3134 | skb_reset_transport_header(skb); |
0b5c9db1 | 3135 | skb_reset_mac_len(skb); |
1da177e4 LT |
3136 | |
3137 | pt_prev = NULL; | |
3138 | ||
3139 | rcu_read_lock(); | |
3140 | ||
63d8ea7f DM |
3141 | another_round: |
3142 | ||
3143 | __this_cpu_inc(softnet_data.processed); | |
3144 | ||
bcc6d479 JP |
3145 | if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) { |
3146 | skb = vlan_untag(skb); | |
3147 | if (unlikely(!skb)) | |
3148 | goto out; | |
3149 | } | |
3150 | ||
1da177e4 LT |
3151 | #ifdef CONFIG_NET_CLS_ACT |
3152 | if (skb->tc_verd & TC_NCLS) { | |
3153 | skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); | |
3154 | goto ncls; | |
3155 | } | |
3156 | #endif | |
3157 | ||
3158 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
63d8ea7f | 3159 | if (!ptype->dev || ptype->dev == skb->dev) { |
4ec93edb | 3160 | if (pt_prev) |
f2ccd8fa | 3161 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
3162 | pt_prev = ptype; |
3163 | } | |
3164 | } | |
3165 | ||
3166 | #ifdef CONFIG_NET_CLS_ACT | |
f697c3e8 HX |
3167 | skb = handle_ing(skb, &pt_prev, &ret, orig_dev); |
3168 | if (!skb) | |
1da177e4 | 3169 | goto out; |
1da177e4 LT |
3170 | ncls: |
3171 | #endif | |
3172 | ||
ab95bfe0 JP |
3173 | rx_handler = rcu_dereference(skb->dev->rx_handler); |
3174 | if (rx_handler) { | |
3175 | if (pt_prev) { | |
3176 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3177 | pt_prev = NULL; | |
3178 | } | |
8a4eb573 JP |
3179 | switch (rx_handler(&skb)) { |
3180 | case RX_HANDLER_CONSUMED: | |
ab95bfe0 | 3181 | goto out; |
8a4eb573 | 3182 | case RX_HANDLER_ANOTHER: |
63d8ea7f | 3183 | goto another_round; |
8a4eb573 JP |
3184 | case RX_HANDLER_EXACT: |
3185 | deliver_exact = true; | |
3186 | case RX_HANDLER_PASS: | |
3187 | break; | |
3188 | default: | |
3189 | BUG(); | |
3190 | } | |
ab95bfe0 | 3191 | } |
1da177e4 | 3192 | |
3701e513 JG |
3193 | if (vlan_tx_tag_present(skb)) { |
3194 | if (pt_prev) { | |
3195 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3196 | pt_prev = NULL; | |
3197 | } | |
bcc6d479 | 3198 | if (vlan_do_receive(&skb)) { |
3701e513 JG |
3199 | ret = __netif_receive_skb(skb); |
3200 | goto out; | |
3201 | } else if (unlikely(!skb)) | |
3202 | goto out; | |
3203 | } | |
3204 | ||
63d8ea7f | 3205 | /* deliver only exact match when indicated */ |
8a4eb573 | 3206 | null_or_dev = deliver_exact ? skb->dev : NULL; |
1f3c8804 | 3207 | |
1da177e4 | 3208 | type = skb->protocol; |
82d8a867 PE |
3209 | list_for_each_entry_rcu(ptype, |
3210 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
63d8ea7f | 3211 | if (ptype->type == type && |
e3f48d37 JP |
3212 | (ptype->dev == null_or_dev || ptype->dev == skb->dev || |
3213 | ptype->dev == orig_dev)) { | |
4ec93edb | 3214 | if (pt_prev) |
f2ccd8fa | 3215 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
3216 | pt_prev = ptype; |
3217 | } | |
3218 | } | |
3219 | ||
3220 | if (pt_prev) { | |
f2ccd8fa | 3221 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
1da177e4 | 3222 | } else { |
caf586e5 | 3223 | atomic_long_inc(&skb->dev->rx_dropped); |
1da177e4 LT |
3224 | kfree_skb(skb); |
3225 | /* Jamal, now you will not able to escape explaining | |
3226 | * me how you were going to use this. :-) | |
3227 | */ | |
3228 | ret = NET_RX_DROP; | |
3229 | } | |
3230 | ||
3231 | out: | |
3232 | rcu_read_unlock(); | |
3233 | return ret; | |
3234 | } | |
0a9627f2 TH |
3235 | |
3236 | /** | |
3237 | * netif_receive_skb - process receive buffer from network | |
3238 | * @skb: buffer to process | |
3239 | * | |
3240 | * netif_receive_skb() is the main receive data processing function. | |
3241 | * It always succeeds. The buffer may be dropped during processing | |
3242 | * for congestion control or by the protocol layers. | |
3243 | * | |
3244 | * This function may only be called from softirq context and interrupts | |
3245 | * should be enabled. | |
3246 | * | |
3247 | * Return values (usually ignored): | |
3248 | * NET_RX_SUCCESS: no congestion | |
3249 | * NET_RX_DROP: packet was dropped | |
3250 | */ | |
3251 | int netif_receive_skb(struct sk_buff *skb) | |
3252 | { | |
3b098e2d ED |
3253 | if (netdev_tstamp_prequeue) |
3254 | net_timestamp_check(skb); | |
3255 | ||
c1f19b51 RC |
3256 | if (skb_defer_rx_timestamp(skb)) |
3257 | return NET_RX_SUCCESS; | |
3258 | ||
df334545 | 3259 | #ifdef CONFIG_RPS |
3b098e2d ED |
3260 | { |
3261 | struct rps_dev_flow voidflow, *rflow = &voidflow; | |
3262 | int cpu, ret; | |
fec5e652 | 3263 | |
3b098e2d ED |
3264 | rcu_read_lock(); |
3265 | ||
3266 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
0a9627f2 | 3267 | |
3b098e2d ED |
3268 | if (cpu >= 0) { |
3269 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
3270 | rcu_read_unlock(); | |
3271 | } else { | |
3272 | rcu_read_unlock(); | |
3273 | ret = __netif_receive_skb(skb); | |
3274 | } | |
0a9627f2 | 3275 | |
3b098e2d | 3276 | return ret; |
fec5e652 | 3277 | } |
1e94d72f TH |
3278 | #else |
3279 | return __netif_receive_skb(skb); | |
3280 | #endif | |
0a9627f2 | 3281 | } |
d1b19dff | 3282 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 3283 | |
88751275 ED |
3284 | /* Network device is going away, flush any packets still pending |
3285 | * Called with irqs disabled. | |
3286 | */ | |
152102c7 | 3287 | static void flush_backlog(void *arg) |
6e583ce5 | 3288 | { |
152102c7 | 3289 | struct net_device *dev = arg; |
e36fa2f7 | 3290 | struct softnet_data *sd = &__get_cpu_var(softnet_data); |
6e583ce5 SH |
3291 | struct sk_buff *skb, *tmp; |
3292 | ||
e36fa2f7 | 3293 | rps_lock(sd); |
6e7676c1 | 3294 | skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { |
6e583ce5 | 3295 | if (skb->dev == dev) { |
e36fa2f7 | 3296 | __skb_unlink(skb, &sd->input_pkt_queue); |
6e583ce5 | 3297 | kfree_skb(skb); |
76cc8b13 | 3298 | input_queue_head_incr(sd); |
6e583ce5 | 3299 | } |
6e7676c1 | 3300 | } |
e36fa2f7 | 3301 | rps_unlock(sd); |
6e7676c1 CG |
3302 | |
3303 | skb_queue_walk_safe(&sd->process_queue, skb, tmp) { | |
3304 | if (skb->dev == dev) { | |
3305 | __skb_unlink(skb, &sd->process_queue); | |
3306 | kfree_skb(skb); | |
76cc8b13 | 3307 | input_queue_head_incr(sd); |
6e7676c1 CG |
3308 | } |
3309 | } | |
6e583ce5 SH |
3310 | } |
3311 | ||
d565b0a1 HX |
3312 | static int napi_gro_complete(struct sk_buff *skb) |
3313 | { | |
3314 | struct packet_type *ptype; | |
3315 | __be16 type = skb->protocol; | |
3316 | struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK]; | |
3317 | int err = -ENOENT; | |
3318 | ||
fc59f9a3 HX |
3319 | if (NAPI_GRO_CB(skb)->count == 1) { |
3320 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 3321 | goto out; |
fc59f9a3 | 3322 | } |
d565b0a1 HX |
3323 | |
3324 | rcu_read_lock(); | |
3325 | list_for_each_entry_rcu(ptype, head, list) { | |
3326 | if (ptype->type != type || ptype->dev || !ptype->gro_complete) | |
3327 | continue; | |
3328 | ||
3329 | err = ptype->gro_complete(skb); | |
3330 | break; | |
3331 | } | |
3332 | rcu_read_unlock(); | |
3333 | ||
3334 | if (err) { | |
3335 | WARN_ON(&ptype->list == head); | |
3336 | kfree_skb(skb); | |
3337 | return NET_RX_SUCCESS; | |
3338 | } | |
3339 | ||
3340 | out: | |
d565b0a1 HX |
3341 | return netif_receive_skb(skb); |
3342 | } | |
3343 | ||
86cac58b | 3344 | inline void napi_gro_flush(struct napi_struct *napi) |
d565b0a1 HX |
3345 | { |
3346 | struct sk_buff *skb, *next; | |
3347 | ||
3348 | for (skb = napi->gro_list; skb; skb = next) { | |
3349 | next = skb->next; | |
3350 | skb->next = NULL; | |
3351 | napi_gro_complete(skb); | |
3352 | } | |
3353 | ||
4ae5544f | 3354 | napi->gro_count = 0; |
d565b0a1 HX |
3355 | napi->gro_list = NULL; |
3356 | } | |
86cac58b | 3357 | EXPORT_SYMBOL(napi_gro_flush); |
d565b0a1 | 3358 | |
5b252f0c | 3359 | enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 HX |
3360 | { |
3361 | struct sk_buff **pp = NULL; | |
3362 | struct packet_type *ptype; | |
3363 | __be16 type = skb->protocol; | |
3364 | struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK]; | |
0da2afd5 | 3365 | int same_flow; |
d565b0a1 | 3366 | int mac_len; |
5b252f0c | 3367 | enum gro_result ret; |
d565b0a1 | 3368 | |
ce9e76c8 | 3369 | if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb)) |
d565b0a1 HX |
3370 | goto normal; |
3371 | ||
21dc3301 | 3372 | if (skb_is_gso(skb) || skb_has_frag_list(skb)) |
f17f5c91 HX |
3373 | goto normal; |
3374 | ||
d565b0a1 HX |
3375 | rcu_read_lock(); |
3376 | list_for_each_entry_rcu(ptype, head, list) { | |
d565b0a1 HX |
3377 | if (ptype->type != type || ptype->dev || !ptype->gro_receive) |
3378 | continue; | |
3379 | ||
86911732 | 3380 | skb_set_network_header(skb, skb_gro_offset(skb)); |
d565b0a1 HX |
3381 | mac_len = skb->network_header - skb->mac_header; |
3382 | skb->mac_len = mac_len; | |
3383 | NAPI_GRO_CB(skb)->same_flow = 0; | |
3384 | NAPI_GRO_CB(skb)->flush = 0; | |
5d38a079 | 3385 | NAPI_GRO_CB(skb)->free = 0; |
d565b0a1 | 3386 | |
d565b0a1 HX |
3387 | pp = ptype->gro_receive(&napi->gro_list, skb); |
3388 | break; | |
3389 | } | |
3390 | rcu_read_unlock(); | |
3391 | ||
3392 | if (&ptype->list == head) | |
3393 | goto normal; | |
3394 | ||
0da2afd5 | 3395 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 3396 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 3397 | |
d565b0a1 HX |
3398 | if (pp) { |
3399 | struct sk_buff *nskb = *pp; | |
3400 | ||
3401 | *pp = nskb->next; | |
3402 | nskb->next = NULL; | |
3403 | napi_gro_complete(nskb); | |
4ae5544f | 3404 | napi->gro_count--; |
d565b0a1 HX |
3405 | } |
3406 | ||
0da2afd5 | 3407 | if (same_flow) |
d565b0a1 HX |
3408 | goto ok; |
3409 | ||
4ae5544f | 3410 | if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS) |
d565b0a1 | 3411 | goto normal; |
d565b0a1 | 3412 | |
4ae5544f | 3413 | napi->gro_count++; |
d565b0a1 | 3414 | NAPI_GRO_CB(skb)->count = 1; |
86911732 | 3415 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
d565b0a1 HX |
3416 | skb->next = napi->gro_list; |
3417 | napi->gro_list = skb; | |
5d0d9be8 | 3418 | ret = GRO_HELD; |
d565b0a1 | 3419 | |
ad0f9904 | 3420 | pull: |
cb18978c HX |
3421 | if (skb_headlen(skb) < skb_gro_offset(skb)) { |
3422 | int grow = skb_gro_offset(skb) - skb_headlen(skb); | |
3423 | ||
3424 | BUG_ON(skb->end - skb->tail < grow); | |
3425 | ||
3426 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
3427 | ||
3428 | skb->tail += grow; | |
3429 | skb->data_len -= grow; | |
3430 | ||
3431 | skb_shinfo(skb)->frags[0].page_offset += grow; | |
3432 | skb_shinfo(skb)->frags[0].size -= grow; | |
3433 | ||
3434 | if (unlikely(!skb_shinfo(skb)->frags[0].size)) { | |
3435 | put_page(skb_shinfo(skb)->frags[0].page); | |
3436 | memmove(skb_shinfo(skb)->frags, | |
3437 | skb_shinfo(skb)->frags + 1, | |
e5093aec | 3438 | --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t)); |
cb18978c | 3439 | } |
ad0f9904 HX |
3440 | } |
3441 | ||
d565b0a1 | 3442 | ok: |
5d0d9be8 | 3443 | return ret; |
d565b0a1 HX |
3444 | |
3445 | normal: | |
ad0f9904 HX |
3446 | ret = GRO_NORMAL; |
3447 | goto pull; | |
5d38a079 | 3448 | } |
96e93eab HX |
3449 | EXPORT_SYMBOL(dev_gro_receive); |
3450 | ||
40d0802b | 3451 | static inline gro_result_t |
5b252f0c | 3452 | __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab HX |
3453 | { |
3454 | struct sk_buff *p; | |
3455 | ||
3456 | for (p = napi->gro_list; p; p = p->next) { | |
40d0802b ED |
3457 | unsigned long diffs; |
3458 | ||
3459 | diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; | |
3701e513 | 3460 | diffs |= p->vlan_tci ^ skb->vlan_tci; |
40d0802b | 3461 | diffs |= compare_ether_header(skb_mac_header(p), |
f64f9e71 | 3462 | skb_gro_mac_header(skb)); |
40d0802b | 3463 | NAPI_GRO_CB(p)->same_flow = !diffs; |
96e93eab HX |
3464 | NAPI_GRO_CB(p)->flush = 0; |
3465 | } | |
3466 | ||
3467 | return dev_gro_receive(napi, skb); | |
3468 | } | |
5d38a079 | 3469 | |
c7c4b3b6 | 3470 | gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) |
5d38a079 | 3471 | { |
5d0d9be8 HX |
3472 | switch (ret) { |
3473 | case GRO_NORMAL: | |
c7c4b3b6 BH |
3474 | if (netif_receive_skb(skb)) |
3475 | ret = GRO_DROP; | |
3476 | break; | |
5d38a079 | 3477 | |
5d0d9be8 | 3478 | case GRO_DROP: |
5d0d9be8 | 3479 | case GRO_MERGED_FREE: |
5d38a079 HX |
3480 | kfree_skb(skb); |
3481 | break; | |
5b252f0c BH |
3482 | |
3483 | case GRO_HELD: | |
3484 | case GRO_MERGED: | |
3485 | break; | |
5d38a079 HX |
3486 | } |
3487 | ||
c7c4b3b6 | 3488 | return ret; |
5d0d9be8 HX |
3489 | } |
3490 | EXPORT_SYMBOL(napi_skb_finish); | |
3491 | ||
78a478d0 HX |
3492 | void skb_gro_reset_offset(struct sk_buff *skb) |
3493 | { | |
3494 | NAPI_GRO_CB(skb)->data_offset = 0; | |
3495 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
7489594c | 3496 | NAPI_GRO_CB(skb)->frag0_len = 0; |
78a478d0 | 3497 | |
78d3fd0b | 3498 | if (skb->mac_header == skb->tail && |
7489594c | 3499 | !PageHighMem(skb_shinfo(skb)->frags[0].page)) { |
78a478d0 HX |
3500 | NAPI_GRO_CB(skb)->frag0 = |
3501 | page_address(skb_shinfo(skb)->frags[0].page) + | |
3502 | skb_shinfo(skb)->frags[0].page_offset; | |
7489594c HX |
3503 | NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size; |
3504 | } | |
78a478d0 HX |
3505 | } |
3506 | EXPORT_SYMBOL(skb_gro_reset_offset); | |
3507 | ||
c7c4b3b6 | 3508 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 3509 | { |
86911732 HX |
3510 | skb_gro_reset_offset(skb); |
3511 | ||
5d0d9be8 | 3512 | return napi_skb_finish(__napi_gro_receive(napi, skb), skb); |
d565b0a1 HX |
3513 | } |
3514 | EXPORT_SYMBOL(napi_gro_receive); | |
3515 | ||
d0c2b0d2 | 3516 | static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab | 3517 | { |
96e93eab HX |
3518 | __skb_pull(skb, skb_headlen(skb)); |
3519 | skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb)); | |
3701e513 | 3520 | skb->vlan_tci = 0; |
66c46d74 | 3521 | skb->dev = napi->dev; |
6d152e23 | 3522 | skb->skb_iif = 0; |
96e93eab HX |
3523 | |
3524 | napi->skb = skb; | |
3525 | } | |
96e93eab | 3526 | |
76620aaf | 3527 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 3528 | { |
5d38a079 | 3529 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
3530 | |
3531 | if (!skb) { | |
89d71a66 ED |
3532 | skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD); |
3533 | if (skb) | |
3534 | napi->skb = skb; | |
80595d59 | 3535 | } |
96e93eab HX |
3536 | return skb; |
3537 | } | |
76620aaf | 3538 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 3539 | |
c7c4b3b6 BH |
3540 | gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb, |
3541 | gro_result_t ret) | |
96e93eab | 3542 | { |
5d0d9be8 HX |
3543 | switch (ret) { |
3544 | case GRO_NORMAL: | |
86911732 | 3545 | case GRO_HELD: |
e76b69cc | 3546 | skb->protocol = eth_type_trans(skb, skb->dev); |
86911732 | 3547 | |
c7c4b3b6 BH |
3548 | if (ret == GRO_HELD) |
3549 | skb_gro_pull(skb, -ETH_HLEN); | |
3550 | else if (netif_receive_skb(skb)) | |
3551 | ret = GRO_DROP; | |
86911732 | 3552 | break; |
5d38a079 | 3553 | |
5d0d9be8 | 3554 | case GRO_DROP: |
5d0d9be8 HX |
3555 | case GRO_MERGED_FREE: |
3556 | napi_reuse_skb(napi, skb); | |
3557 | break; | |
5b252f0c BH |
3558 | |
3559 | case GRO_MERGED: | |
3560 | break; | |
5d0d9be8 | 3561 | } |
5d38a079 | 3562 | |
c7c4b3b6 | 3563 | return ret; |
5d38a079 | 3564 | } |
5d0d9be8 HX |
3565 | EXPORT_SYMBOL(napi_frags_finish); |
3566 | ||
76620aaf HX |
3567 | struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
3568 | { | |
3569 | struct sk_buff *skb = napi->skb; | |
3570 | struct ethhdr *eth; | |
a5b1cf28 HX |
3571 | unsigned int hlen; |
3572 | unsigned int off; | |
76620aaf HX |
3573 | |
3574 | napi->skb = NULL; | |
3575 | ||
3576 | skb_reset_mac_header(skb); | |
3577 | skb_gro_reset_offset(skb); | |
3578 | ||
a5b1cf28 HX |
3579 | off = skb_gro_offset(skb); |
3580 | hlen = off + sizeof(*eth); | |
3581 | eth = skb_gro_header_fast(skb, off); | |
3582 | if (skb_gro_header_hard(skb, hlen)) { | |
3583 | eth = skb_gro_header_slow(skb, hlen, off); | |
3584 | if (unlikely(!eth)) { | |
3585 | napi_reuse_skb(napi, skb); | |
3586 | skb = NULL; | |
3587 | goto out; | |
3588 | } | |
76620aaf HX |
3589 | } |
3590 | ||
3591 | skb_gro_pull(skb, sizeof(*eth)); | |
3592 | ||
3593 | /* | |
3594 | * This works because the only protocols we care about don't require | |
3595 | * special handling. We'll fix it up properly at the end. | |
3596 | */ | |
3597 | skb->protocol = eth->h_proto; | |
3598 | ||
3599 | out: | |
3600 | return skb; | |
3601 | } | |
3602 | EXPORT_SYMBOL(napi_frags_skb); | |
3603 | ||
c7c4b3b6 | 3604 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 3605 | { |
76620aaf | 3606 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 HX |
3607 | |
3608 | if (!skb) | |
c7c4b3b6 | 3609 | return GRO_DROP; |
5d0d9be8 HX |
3610 | |
3611 | return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb)); | |
3612 | } | |
5d38a079 HX |
3613 | EXPORT_SYMBOL(napi_gro_frags); |
3614 | ||
e326bed2 ED |
3615 | /* |
3616 | * net_rps_action sends any pending IPI's for rps. | |
3617 | * Note: called with local irq disabled, but exits with local irq enabled. | |
3618 | */ | |
3619 | static void net_rps_action_and_irq_enable(struct softnet_data *sd) | |
3620 | { | |
3621 | #ifdef CONFIG_RPS | |
3622 | struct softnet_data *remsd = sd->rps_ipi_list; | |
3623 | ||
3624 | if (remsd) { | |
3625 | sd->rps_ipi_list = NULL; | |
3626 | ||
3627 | local_irq_enable(); | |
3628 | ||
3629 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
3630 | while (remsd) { | |
3631 | struct softnet_data *next = remsd->rps_ipi_next; | |
3632 | ||
3633 | if (cpu_online(remsd->cpu)) | |
3634 | __smp_call_function_single(remsd->cpu, | |
3635 | &remsd->csd, 0); | |
3636 | remsd = next; | |
3637 | } | |
3638 | } else | |
3639 | #endif | |
3640 | local_irq_enable(); | |
3641 | } | |
3642 | ||
bea3348e | 3643 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 LT |
3644 | { |
3645 | int work = 0; | |
eecfd7c4 | 3646 | struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); |
1da177e4 | 3647 | |
e326bed2 ED |
3648 | #ifdef CONFIG_RPS |
3649 | /* Check if we have pending ipi, its better to send them now, | |
3650 | * not waiting net_rx_action() end. | |
3651 | */ | |
3652 | if (sd->rps_ipi_list) { | |
3653 | local_irq_disable(); | |
3654 | net_rps_action_and_irq_enable(sd); | |
3655 | } | |
3656 | #endif | |
bea3348e | 3657 | napi->weight = weight_p; |
6e7676c1 CG |
3658 | local_irq_disable(); |
3659 | while (work < quota) { | |
1da177e4 | 3660 | struct sk_buff *skb; |
6e7676c1 CG |
3661 | unsigned int qlen; |
3662 | ||
3663 | while ((skb = __skb_dequeue(&sd->process_queue))) { | |
3664 | local_irq_enable(); | |
3665 | __netif_receive_skb(skb); | |
6e7676c1 | 3666 | local_irq_disable(); |
76cc8b13 TH |
3667 | input_queue_head_incr(sd); |
3668 | if (++work >= quota) { | |
3669 | local_irq_enable(); | |
3670 | return work; | |
3671 | } | |
6e7676c1 | 3672 | } |
1da177e4 | 3673 | |
e36fa2f7 | 3674 | rps_lock(sd); |
6e7676c1 | 3675 | qlen = skb_queue_len(&sd->input_pkt_queue); |
76cc8b13 | 3676 | if (qlen) |
6e7676c1 CG |
3677 | skb_queue_splice_tail_init(&sd->input_pkt_queue, |
3678 | &sd->process_queue); | |
76cc8b13 | 3679 | |
6e7676c1 | 3680 | if (qlen < quota - work) { |
eecfd7c4 ED |
3681 | /* |
3682 | * Inline a custom version of __napi_complete(). | |
3683 | * only current cpu owns and manipulates this napi, | |
3684 | * and NAPI_STATE_SCHED is the only possible flag set on backlog. | |
3685 | * we can use a plain write instead of clear_bit(), | |
3686 | * and we dont need an smp_mb() memory barrier. | |
3687 | */ | |
3688 | list_del(&napi->poll_list); | |
3689 | napi->state = 0; | |
3690 | ||
6e7676c1 | 3691 | quota = work + qlen; |
bea3348e | 3692 | } |
e36fa2f7 | 3693 | rps_unlock(sd); |
6e7676c1 CG |
3694 | } |
3695 | local_irq_enable(); | |
1da177e4 | 3696 | |
bea3348e SH |
3697 | return work; |
3698 | } | |
1da177e4 | 3699 | |
bea3348e SH |
3700 | /** |
3701 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 3702 | * @n: entry to schedule |
bea3348e SH |
3703 | * |
3704 | * The entry's receive function will be scheduled to run | |
3705 | */ | |
b5606c2d | 3706 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
3707 | { |
3708 | unsigned long flags; | |
1da177e4 | 3709 | |
bea3348e | 3710 | local_irq_save(flags); |
eecfd7c4 | 3711 | ____napi_schedule(&__get_cpu_var(softnet_data), n); |
bea3348e | 3712 | local_irq_restore(flags); |
1da177e4 | 3713 | } |
bea3348e SH |
3714 | EXPORT_SYMBOL(__napi_schedule); |
3715 | ||
d565b0a1 HX |
3716 | void __napi_complete(struct napi_struct *n) |
3717 | { | |
3718 | BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); | |
3719 | BUG_ON(n->gro_list); | |
3720 | ||
3721 | list_del(&n->poll_list); | |
3722 | smp_mb__before_clear_bit(); | |
3723 | clear_bit(NAPI_STATE_SCHED, &n->state); | |
3724 | } | |
3725 | EXPORT_SYMBOL(__napi_complete); | |
3726 | ||
3727 | void napi_complete(struct napi_struct *n) | |
3728 | { | |
3729 | unsigned long flags; | |
3730 | ||
3731 | /* | |
3732 | * don't let napi dequeue from the cpu poll list | |
3733 | * just in case its running on a different cpu | |
3734 | */ | |
3735 | if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state))) | |
3736 | return; | |
3737 | ||
3738 | napi_gro_flush(n); | |
3739 | local_irq_save(flags); | |
3740 | __napi_complete(n); | |
3741 | local_irq_restore(flags); | |
3742 | } | |
3743 | EXPORT_SYMBOL(napi_complete); | |
3744 | ||
3745 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, | |
3746 | int (*poll)(struct napi_struct *, int), int weight) | |
3747 | { | |
3748 | INIT_LIST_HEAD(&napi->poll_list); | |
4ae5544f | 3749 | napi->gro_count = 0; |
d565b0a1 | 3750 | napi->gro_list = NULL; |
5d38a079 | 3751 | napi->skb = NULL; |
d565b0a1 HX |
3752 | napi->poll = poll; |
3753 | napi->weight = weight; | |
3754 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 3755 | napi->dev = dev; |
5d38a079 | 3756 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
3757 | spin_lock_init(&napi->poll_lock); |
3758 | napi->poll_owner = -1; | |
3759 | #endif | |
3760 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
3761 | } | |
3762 | EXPORT_SYMBOL(netif_napi_add); | |
3763 | ||
3764 | void netif_napi_del(struct napi_struct *napi) | |
3765 | { | |
3766 | struct sk_buff *skb, *next; | |
3767 | ||
d7b06636 | 3768 | list_del_init(&napi->dev_list); |
76620aaf | 3769 | napi_free_frags(napi); |
d565b0a1 HX |
3770 | |
3771 | for (skb = napi->gro_list; skb; skb = next) { | |
3772 | next = skb->next; | |
3773 | skb->next = NULL; | |
3774 | kfree_skb(skb); | |
3775 | } | |
3776 | ||
3777 | napi->gro_list = NULL; | |
4ae5544f | 3778 | napi->gro_count = 0; |
d565b0a1 HX |
3779 | } |
3780 | EXPORT_SYMBOL(netif_napi_del); | |
3781 | ||
1da177e4 LT |
3782 | static void net_rx_action(struct softirq_action *h) |
3783 | { | |
e326bed2 | 3784 | struct softnet_data *sd = &__get_cpu_var(softnet_data); |
24f8b238 | 3785 | unsigned long time_limit = jiffies + 2; |
51b0bded | 3786 | int budget = netdev_budget; |
53fb95d3 MM |
3787 | void *have; |
3788 | ||
1da177e4 LT |
3789 | local_irq_disable(); |
3790 | ||
e326bed2 | 3791 | while (!list_empty(&sd->poll_list)) { |
bea3348e SH |
3792 | struct napi_struct *n; |
3793 | int work, weight; | |
1da177e4 | 3794 | |
bea3348e | 3795 | /* If softirq window is exhuasted then punt. |
24f8b238 SH |
3796 | * Allow this to run for 2 jiffies since which will allow |
3797 | * an average latency of 1.5/HZ. | |
bea3348e | 3798 | */ |
24f8b238 | 3799 | if (unlikely(budget <= 0 || time_after(jiffies, time_limit))) |
1da177e4 LT |
3800 | goto softnet_break; |
3801 | ||
3802 | local_irq_enable(); | |
3803 | ||
bea3348e SH |
3804 | /* Even though interrupts have been re-enabled, this |
3805 | * access is safe because interrupts can only add new | |
3806 | * entries to the tail of this list, and only ->poll() | |
3807 | * calls can remove this head entry from the list. | |
3808 | */ | |
e326bed2 | 3809 | n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list); |
1da177e4 | 3810 | |
bea3348e SH |
3811 | have = netpoll_poll_lock(n); |
3812 | ||
3813 | weight = n->weight; | |
3814 | ||
0a7606c1 DM |
3815 | /* This NAPI_STATE_SCHED test is for avoiding a race |
3816 | * with netpoll's poll_napi(). Only the entity which | |
3817 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
3818 | * actually make the ->poll() call. Therefore we avoid | |
25985edc | 3819 | * accidentally calling ->poll() when NAPI is not scheduled. |
0a7606c1 DM |
3820 | */ |
3821 | work = 0; | |
4ea7e386 | 3822 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { |
0a7606c1 | 3823 | work = n->poll(n, weight); |
4ea7e386 NH |
3824 | trace_napi_poll(n); |
3825 | } | |
bea3348e SH |
3826 | |
3827 | WARN_ON_ONCE(work > weight); | |
3828 | ||
3829 | budget -= work; | |
3830 | ||
3831 | local_irq_disable(); | |
3832 | ||
3833 | /* Drivers must not modify the NAPI state if they | |
3834 | * consume the entire weight. In such cases this code | |
3835 | * still "owns" the NAPI instance and therefore can | |
3836 | * move the instance around on the list at-will. | |
3837 | */ | |
fed17f30 | 3838 | if (unlikely(work == weight)) { |
ff780cd8 HX |
3839 | if (unlikely(napi_disable_pending(n))) { |
3840 | local_irq_enable(); | |
3841 | napi_complete(n); | |
3842 | local_irq_disable(); | |
3843 | } else | |
e326bed2 | 3844 | list_move_tail(&n->poll_list, &sd->poll_list); |
fed17f30 | 3845 | } |
bea3348e SH |
3846 | |
3847 | netpoll_poll_unlock(have); | |
1da177e4 LT |
3848 | } |
3849 | out: | |
e326bed2 | 3850 | net_rps_action_and_irq_enable(sd); |
0a9627f2 | 3851 | |
db217334 CL |
3852 | #ifdef CONFIG_NET_DMA |
3853 | /* | |
3854 | * There may not be any more sk_buffs coming right now, so push | |
3855 | * any pending DMA copies to hardware | |
3856 | */ | |
2ba05622 | 3857 | dma_issue_pending_all(); |
db217334 | 3858 | #endif |
bea3348e | 3859 | |
1da177e4 LT |
3860 | return; |
3861 | ||
3862 | softnet_break: | |
dee42870 | 3863 | sd->time_squeeze++; |
1da177e4 LT |
3864 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); |
3865 | goto out; | |
3866 | } | |
3867 | ||
d1b19dff | 3868 | static gifconf_func_t *gifconf_list[NPROTO]; |
1da177e4 LT |
3869 | |
3870 | /** | |
3871 | * register_gifconf - register a SIOCGIF handler | |
3872 | * @family: Address family | |
3873 | * @gifconf: Function handler | |
3874 | * | |
3875 | * Register protocol dependent address dumping routines. The handler | |
3876 | * that is passed must not be freed or reused until it has been replaced | |
3877 | * by another handler. | |
3878 | */ | |
d1b19dff | 3879 | int register_gifconf(unsigned int family, gifconf_func_t *gifconf) |
1da177e4 LT |
3880 | { |
3881 | if (family >= NPROTO) | |
3882 | return -EINVAL; | |
3883 | gifconf_list[family] = gifconf; | |
3884 | return 0; | |
3885 | } | |
d1b19dff | 3886 | EXPORT_SYMBOL(register_gifconf); |
1da177e4 LT |
3887 | |
3888 | ||
3889 | /* | |
3890 | * Map an interface index to its name (SIOCGIFNAME) | |
3891 | */ | |
3892 | ||
3893 | /* | |
3894 | * We need this ioctl for efficient implementation of the | |
3895 | * if_indextoname() function required by the IPv6 API. Without | |
3896 | * it, we would have to search all the interfaces to find a | |
3897 | * match. --pb | |
3898 | */ | |
3899 | ||
881d966b | 3900 | static int dev_ifname(struct net *net, struct ifreq __user *arg) |
1da177e4 LT |
3901 | { |
3902 | struct net_device *dev; | |
3903 | struct ifreq ifr; | |
3904 | ||
3905 | /* | |
3906 | * Fetch the caller's info block. | |
3907 | */ | |
3908 | ||
3909 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
3910 | return -EFAULT; | |
3911 | ||
fb699dfd ED |
3912 | rcu_read_lock(); |
3913 | dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex); | |
1da177e4 | 3914 | if (!dev) { |
fb699dfd | 3915 | rcu_read_unlock(); |
1da177e4 LT |
3916 | return -ENODEV; |
3917 | } | |
3918 | ||
3919 | strcpy(ifr.ifr_name, dev->name); | |
fb699dfd | 3920 | rcu_read_unlock(); |
1da177e4 LT |
3921 | |
3922 | if (copy_to_user(arg, &ifr, sizeof(struct ifreq))) | |
3923 | return -EFAULT; | |
3924 | return 0; | |
3925 | } | |
3926 | ||
3927 | /* | |
3928 | * Perform a SIOCGIFCONF call. This structure will change | |
3929 | * size eventually, and there is nothing I can do about it. | |
3930 | * Thus we will need a 'compatibility mode'. | |
3931 | */ | |
3932 | ||
881d966b | 3933 | static int dev_ifconf(struct net *net, char __user *arg) |
1da177e4 LT |
3934 | { |
3935 | struct ifconf ifc; | |
3936 | struct net_device *dev; | |
3937 | char __user *pos; | |
3938 | int len; | |
3939 | int total; | |
3940 | int i; | |
3941 | ||
3942 | /* | |
3943 | * Fetch the caller's info block. | |
3944 | */ | |
3945 | ||
3946 | if (copy_from_user(&ifc, arg, sizeof(struct ifconf))) | |
3947 | return -EFAULT; | |
3948 | ||
3949 | pos = ifc.ifc_buf; | |
3950 | len = ifc.ifc_len; | |
3951 | ||
3952 | /* | |
3953 | * Loop over the interfaces, and write an info block for each. | |
3954 | */ | |
3955 | ||
3956 | total = 0; | |
881d966b | 3957 | for_each_netdev(net, dev) { |
1da177e4 LT |
3958 | for (i = 0; i < NPROTO; i++) { |
3959 | if (gifconf_list[i]) { | |
3960 | int done; | |
3961 | if (!pos) | |
3962 | done = gifconf_list[i](dev, NULL, 0); | |
3963 | else | |
3964 | done = gifconf_list[i](dev, pos + total, | |
3965 | len - total); | |
3966 | if (done < 0) | |
3967 | return -EFAULT; | |
3968 | total += done; | |
3969 | } | |
3970 | } | |
4ec93edb | 3971 | } |
1da177e4 LT |
3972 | |
3973 | /* | |
3974 | * All done. Write the updated control block back to the caller. | |
3975 | */ | |
3976 | ifc.ifc_len = total; | |
3977 | ||
3978 | /* | |
3979 | * Both BSD and Solaris return 0 here, so we do too. | |
3980 | */ | |
3981 | return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0; | |
3982 | } | |
3983 | ||
3984 | #ifdef CONFIG_PROC_FS | |
3985 | /* | |
3986 | * This is invoked by the /proc filesystem handler to display a device | |
3987 | * in detail. | |
3988 | */ | |
7562f876 | 3989 | void *dev_seq_start(struct seq_file *seq, loff_t *pos) |
c6d14c84 | 3990 | __acquires(RCU) |
1da177e4 | 3991 | { |
e372c414 | 3992 | struct net *net = seq_file_net(seq); |
7562f876 | 3993 | loff_t off; |
1da177e4 | 3994 | struct net_device *dev; |
1da177e4 | 3995 | |
c6d14c84 | 3996 | rcu_read_lock(); |
7562f876 PE |
3997 | if (!*pos) |
3998 | return SEQ_START_TOKEN; | |
1da177e4 | 3999 | |
7562f876 | 4000 | off = 1; |
c6d14c84 | 4001 | for_each_netdev_rcu(net, dev) |
7562f876 PE |
4002 | if (off++ == *pos) |
4003 | return dev; | |
1da177e4 | 4004 | |
7562f876 | 4005 | return NULL; |
1da177e4 LT |
4006 | } |
4007 | ||
4008 | void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
4009 | { | |
ccf43438 ED |
4010 | struct net_device *dev = v; |
4011 | ||
4012 | if (v == SEQ_START_TOKEN) | |
4013 | dev = first_net_device_rcu(seq_file_net(seq)); | |
4014 | else | |
4015 | dev = next_net_device_rcu(dev); | |
c6d14c84 | 4016 | |
1da177e4 | 4017 | ++*pos; |
ccf43438 | 4018 | return dev; |
1da177e4 LT |
4019 | } |
4020 | ||
4021 | void dev_seq_stop(struct seq_file *seq, void *v) | |
c6d14c84 | 4022 | __releases(RCU) |
1da177e4 | 4023 | { |
c6d14c84 | 4024 | rcu_read_unlock(); |
1da177e4 LT |
4025 | } |
4026 | ||
4027 | static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev) | |
4028 | { | |
28172739 ED |
4029 | struct rtnl_link_stats64 temp; |
4030 | const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp); | |
1da177e4 | 4031 | |
be1f3c2c BH |
4032 | seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu " |
4033 | "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n", | |
5a1b5898 RR |
4034 | dev->name, stats->rx_bytes, stats->rx_packets, |
4035 | stats->rx_errors, | |
4036 | stats->rx_dropped + stats->rx_missed_errors, | |
4037 | stats->rx_fifo_errors, | |
4038 | stats->rx_length_errors + stats->rx_over_errors + | |
4039 | stats->rx_crc_errors + stats->rx_frame_errors, | |
4040 | stats->rx_compressed, stats->multicast, | |
4041 | stats->tx_bytes, stats->tx_packets, | |
4042 | stats->tx_errors, stats->tx_dropped, | |
4043 | stats->tx_fifo_errors, stats->collisions, | |
4044 | stats->tx_carrier_errors + | |
4045 | stats->tx_aborted_errors + | |
4046 | stats->tx_window_errors + | |
4047 | stats->tx_heartbeat_errors, | |
4048 | stats->tx_compressed); | |
1da177e4 LT |
4049 | } |
4050 | ||
4051 | /* | |
4052 | * Called from the PROCfs module. This now uses the new arbitrary sized | |
4053 | * /proc/net interface to create /proc/net/dev | |
4054 | */ | |
4055 | static int dev_seq_show(struct seq_file *seq, void *v) | |
4056 | { | |
4057 | if (v == SEQ_START_TOKEN) | |
4058 | seq_puts(seq, "Inter-| Receive " | |
4059 | " | Transmit\n" | |
4060 | " face |bytes packets errs drop fifo frame " | |
4061 | "compressed multicast|bytes packets errs " | |
4062 | "drop fifo colls carrier compressed\n"); | |
4063 | else | |
4064 | dev_seq_printf_stats(seq, v); | |
4065 | return 0; | |
4066 | } | |
4067 | ||
dee42870 | 4068 | static struct softnet_data *softnet_get_online(loff_t *pos) |
1da177e4 | 4069 | { |
dee42870 | 4070 | struct softnet_data *sd = NULL; |
1da177e4 | 4071 | |
0c0b0aca | 4072 | while (*pos < nr_cpu_ids) |
4ec93edb | 4073 | if (cpu_online(*pos)) { |
dee42870 | 4074 | sd = &per_cpu(softnet_data, *pos); |
1da177e4 LT |
4075 | break; |
4076 | } else | |
4077 | ++*pos; | |
dee42870 | 4078 | return sd; |
1da177e4 LT |
4079 | } |
4080 | ||
4081 | static void *softnet_seq_start(struct seq_file *seq, loff_t *pos) | |
4082 | { | |
4083 | return softnet_get_online(pos); | |
4084 | } | |
4085 | ||
4086 | static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
4087 | { | |
4088 | ++*pos; | |
4089 | return softnet_get_online(pos); | |
4090 | } | |
4091 | ||
4092 | static void softnet_seq_stop(struct seq_file *seq, void *v) | |
4093 | { | |
4094 | } | |
4095 | ||
4096 | static int softnet_seq_show(struct seq_file *seq, void *v) | |
4097 | { | |
dee42870 | 4098 | struct softnet_data *sd = v; |
1da177e4 | 4099 | |
0a9627f2 | 4100 | seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n", |
dee42870 | 4101 | sd->processed, sd->dropped, sd->time_squeeze, 0, |
c1ebcdb8 | 4102 | 0, 0, 0, 0, /* was fastroute */ |
dee42870 | 4103 | sd->cpu_collision, sd->received_rps); |
1da177e4 LT |
4104 | return 0; |
4105 | } | |
4106 | ||
f690808e | 4107 | static const struct seq_operations dev_seq_ops = { |
1da177e4 LT |
4108 | .start = dev_seq_start, |
4109 | .next = dev_seq_next, | |
4110 | .stop = dev_seq_stop, | |
4111 | .show = dev_seq_show, | |
4112 | }; | |
4113 | ||
4114 | static int dev_seq_open(struct inode *inode, struct file *file) | |
4115 | { | |
e372c414 DL |
4116 | return seq_open_net(inode, file, &dev_seq_ops, |
4117 | sizeof(struct seq_net_private)); | |
1da177e4 LT |
4118 | } |
4119 | ||
9a32144e | 4120 | static const struct file_operations dev_seq_fops = { |
1da177e4 LT |
4121 | .owner = THIS_MODULE, |
4122 | .open = dev_seq_open, | |
4123 | .read = seq_read, | |
4124 | .llseek = seq_lseek, | |
e372c414 | 4125 | .release = seq_release_net, |
1da177e4 LT |
4126 | }; |
4127 | ||
f690808e | 4128 | static const struct seq_operations softnet_seq_ops = { |
1da177e4 LT |
4129 | .start = softnet_seq_start, |
4130 | .next = softnet_seq_next, | |
4131 | .stop = softnet_seq_stop, | |
4132 | .show = softnet_seq_show, | |
4133 | }; | |
4134 | ||
4135 | static int softnet_seq_open(struct inode *inode, struct file *file) | |
4136 | { | |
4137 | return seq_open(file, &softnet_seq_ops); | |
4138 | } | |
4139 | ||
9a32144e | 4140 | static const struct file_operations softnet_seq_fops = { |
1da177e4 LT |
4141 | .owner = THIS_MODULE, |
4142 | .open = softnet_seq_open, | |
4143 | .read = seq_read, | |
4144 | .llseek = seq_lseek, | |
4145 | .release = seq_release, | |
4146 | }; | |
4147 | ||
0e1256ff SH |
4148 | static void *ptype_get_idx(loff_t pos) |
4149 | { | |
4150 | struct packet_type *pt = NULL; | |
4151 | loff_t i = 0; | |
4152 | int t; | |
4153 | ||
4154 | list_for_each_entry_rcu(pt, &ptype_all, list) { | |
4155 | if (i == pos) | |
4156 | return pt; | |
4157 | ++i; | |
4158 | } | |
4159 | ||
82d8a867 | 4160 | for (t = 0; t < PTYPE_HASH_SIZE; t++) { |
0e1256ff SH |
4161 | list_for_each_entry_rcu(pt, &ptype_base[t], list) { |
4162 | if (i == pos) | |
4163 | return pt; | |
4164 | ++i; | |
4165 | } | |
4166 | } | |
4167 | return NULL; | |
4168 | } | |
4169 | ||
4170 | static void *ptype_seq_start(struct seq_file *seq, loff_t *pos) | |
72348a42 | 4171 | __acquires(RCU) |
0e1256ff SH |
4172 | { |
4173 | rcu_read_lock(); | |
4174 | return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN; | |
4175 | } | |
4176 | ||
4177 | static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
4178 | { | |
4179 | struct packet_type *pt; | |
4180 | struct list_head *nxt; | |
4181 | int hash; | |
4182 | ||
4183 | ++*pos; | |
4184 | if (v == SEQ_START_TOKEN) | |
4185 | return ptype_get_idx(0); | |
4186 | ||
4187 | pt = v; | |
4188 | nxt = pt->list.next; | |
4189 | if (pt->type == htons(ETH_P_ALL)) { | |
4190 | if (nxt != &ptype_all) | |
4191 | goto found; | |
4192 | hash = 0; | |
4193 | nxt = ptype_base[0].next; | |
4194 | } else | |
82d8a867 | 4195 | hash = ntohs(pt->type) & PTYPE_HASH_MASK; |
0e1256ff SH |
4196 | |
4197 | while (nxt == &ptype_base[hash]) { | |
82d8a867 | 4198 | if (++hash >= PTYPE_HASH_SIZE) |
0e1256ff SH |
4199 | return NULL; |
4200 | nxt = ptype_base[hash].next; | |
4201 | } | |
4202 | found: | |
4203 | return list_entry(nxt, struct packet_type, list); | |
4204 | } | |
4205 | ||
4206 | static void ptype_seq_stop(struct seq_file *seq, void *v) | |
72348a42 | 4207 | __releases(RCU) |
0e1256ff SH |
4208 | { |
4209 | rcu_read_unlock(); | |
4210 | } | |
4211 | ||
0e1256ff SH |
4212 | static int ptype_seq_show(struct seq_file *seq, void *v) |
4213 | { | |
4214 | struct packet_type *pt = v; | |
4215 | ||
4216 | if (v == SEQ_START_TOKEN) | |
4217 | seq_puts(seq, "Type Device Function\n"); | |
c346dca1 | 4218 | else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) { |
0e1256ff SH |
4219 | if (pt->type == htons(ETH_P_ALL)) |
4220 | seq_puts(seq, "ALL "); | |
4221 | else | |
4222 | seq_printf(seq, "%04x", ntohs(pt->type)); | |
4223 | ||
908cd2da AD |
4224 | seq_printf(seq, " %-8s %pF\n", |
4225 | pt->dev ? pt->dev->name : "", pt->func); | |
0e1256ff SH |
4226 | } |
4227 | ||
4228 | return 0; | |
4229 | } | |
4230 | ||
4231 | static const struct seq_operations ptype_seq_ops = { | |
4232 | .start = ptype_seq_start, | |
4233 | .next = ptype_seq_next, | |
4234 | .stop = ptype_seq_stop, | |
4235 | .show = ptype_seq_show, | |
4236 | }; | |
4237 | ||
4238 | static int ptype_seq_open(struct inode *inode, struct file *file) | |
4239 | { | |
2feb27db PE |
4240 | return seq_open_net(inode, file, &ptype_seq_ops, |
4241 | sizeof(struct seq_net_private)); | |
0e1256ff SH |
4242 | } |
4243 | ||
4244 | static const struct file_operations ptype_seq_fops = { | |
4245 | .owner = THIS_MODULE, | |
4246 | .open = ptype_seq_open, | |
4247 | .read = seq_read, | |
4248 | .llseek = seq_lseek, | |
2feb27db | 4249 | .release = seq_release_net, |
0e1256ff SH |
4250 | }; |
4251 | ||
4252 | ||
4665079c | 4253 | static int __net_init dev_proc_net_init(struct net *net) |
1da177e4 LT |
4254 | { |
4255 | int rc = -ENOMEM; | |
4256 | ||
881d966b | 4257 | if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops)) |
1da177e4 | 4258 | goto out; |
881d966b | 4259 | if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops)) |
1da177e4 | 4260 | goto out_dev; |
881d966b | 4261 | if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops)) |
457c4cbc | 4262 | goto out_softnet; |
0e1256ff | 4263 | |
881d966b | 4264 | if (wext_proc_init(net)) |
457c4cbc | 4265 | goto out_ptype; |
1da177e4 LT |
4266 | rc = 0; |
4267 | out: | |
4268 | return rc; | |
457c4cbc | 4269 | out_ptype: |
881d966b | 4270 | proc_net_remove(net, "ptype"); |
1da177e4 | 4271 | out_softnet: |
881d966b | 4272 | proc_net_remove(net, "softnet_stat"); |
1da177e4 | 4273 | out_dev: |
881d966b | 4274 | proc_net_remove(net, "dev"); |
1da177e4 LT |
4275 | goto out; |
4276 | } | |
881d966b | 4277 | |
4665079c | 4278 | static void __net_exit dev_proc_net_exit(struct net *net) |
881d966b EB |
4279 | { |
4280 | wext_proc_exit(net); | |
4281 | ||
4282 | proc_net_remove(net, "ptype"); | |
4283 | proc_net_remove(net, "softnet_stat"); | |
4284 | proc_net_remove(net, "dev"); | |
4285 | } | |
4286 | ||
022cbae6 | 4287 | static struct pernet_operations __net_initdata dev_proc_ops = { |
881d966b EB |
4288 | .init = dev_proc_net_init, |
4289 | .exit = dev_proc_net_exit, | |
4290 | }; | |
4291 | ||
4292 | static int __init dev_proc_init(void) | |
4293 | { | |
4294 | return register_pernet_subsys(&dev_proc_ops); | |
4295 | } | |
1da177e4 LT |
4296 | #else |
4297 | #define dev_proc_init() 0 | |
4298 | #endif /* CONFIG_PROC_FS */ | |
4299 | ||
4300 | ||
4301 | /** | |
1765a575 | 4302 | * netdev_set_master - set up master pointer |
1da177e4 LT |
4303 | * @slave: slave device |
4304 | * @master: new master device | |
4305 | * | |
4306 | * Changes the master device of the slave. Pass %NULL to break the | |
4307 | * bonding. The caller must hold the RTNL semaphore. On a failure | |
4308 | * a negative errno code is returned. On success the reference counts | |
1765a575 | 4309 | * are adjusted and the function returns zero. |
1da177e4 LT |
4310 | */ |
4311 | int netdev_set_master(struct net_device *slave, struct net_device *master) | |
4312 | { | |
4313 | struct net_device *old = slave->master; | |
4314 | ||
4315 | ASSERT_RTNL(); | |
4316 | ||
4317 | if (master) { | |
4318 | if (old) | |
4319 | return -EBUSY; | |
4320 | dev_hold(master); | |
4321 | } | |
4322 | ||
4323 | slave->master = master; | |
4ec93edb | 4324 | |
6df427fe | 4325 | if (old) |
1da177e4 | 4326 | dev_put(old); |
1765a575 JP |
4327 | return 0; |
4328 | } | |
4329 | EXPORT_SYMBOL(netdev_set_master); | |
4330 | ||
4331 | /** | |
4332 | * netdev_set_bond_master - set up bonding master/slave pair | |
4333 | * @slave: slave device | |
4334 | * @master: new master device | |
4335 | * | |
4336 | * Changes the master device of the slave. Pass %NULL to break the | |
4337 | * bonding. The caller must hold the RTNL semaphore. On a failure | |
4338 | * a negative errno code is returned. On success %RTM_NEWLINK is sent | |
4339 | * to the routing socket and the function returns zero. | |
4340 | */ | |
4341 | int netdev_set_bond_master(struct net_device *slave, struct net_device *master) | |
4342 | { | |
4343 | int err; | |
4344 | ||
4345 | ASSERT_RTNL(); | |
4346 | ||
4347 | err = netdev_set_master(slave, master); | |
4348 | if (err) | |
4349 | return err; | |
1da177e4 LT |
4350 | if (master) |
4351 | slave->flags |= IFF_SLAVE; | |
4352 | else | |
4353 | slave->flags &= ~IFF_SLAVE; | |
4354 | ||
4355 | rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE); | |
4356 | return 0; | |
4357 | } | |
1765a575 | 4358 | EXPORT_SYMBOL(netdev_set_bond_master); |
1da177e4 | 4359 | |
b6c40d68 PM |
4360 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
4361 | { | |
d314774c SH |
4362 | const struct net_device_ops *ops = dev->netdev_ops; |
4363 | ||
4364 | if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags) | |
4365 | ops->ndo_change_rx_flags(dev, flags); | |
b6c40d68 PM |
4366 | } |
4367 | ||
dad9b335 | 4368 | static int __dev_set_promiscuity(struct net_device *dev, int inc) |
1da177e4 LT |
4369 | { |
4370 | unsigned short old_flags = dev->flags; | |
8192b0c4 DH |
4371 | uid_t uid; |
4372 | gid_t gid; | |
1da177e4 | 4373 | |
24023451 PM |
4374 | ASSERT_RTNL(); |
4375 | ||
dad9b335 WC |
4376 | dev->flags |= IFF_PROMISC; |
4377 | dev->promiscuity += inc; | |
4378 | if (dev->promiscuity == 0) { | |
4379 | /* | |
4380 | * Avoid overflow. | |
4381 | * If inc causes overflow, untouch promisc and return error. | |
4382 | */ | |
4383 | if (inc < 0) | |
4384 | dev->flags &= ~IFF_PROMISC; | |
4385 | else { | |
4386 | dev->promiscuity -= inc; | |
4387 | printk(KERN_WARNING "%s: promiscuity touches roof, " | |
4388 | "set promiscuity failed, promiscuity feature " | |
4389 | "of device might be broken.\n", dev->name); | |
4390 | return -EOVERFLOW; | |
4391 | } | |
4392 | } | |
52609c0b | 4393 | if (dev->flags != old_flags) { |
1da177e4 LT |
4394 | printk(KERN_INFO "device %s %s promiscuous mode\n", |
4395 | dev->name, (dev->flags & IFF_PROMISC) ? "entered" : | |
4ec93edb | 4396 | "left"); |
8192b0c4 DH |
4397 | if (audit_enabled) { |
4398 | current_uid_gid(&uid, &gid); | |
7759db82 KHK |
4399 | audit_log(current->audit_context, GFP_ATOMIC, |
4400 | AUDIT_ANOM_PROMISCUOUS, | |
4401 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
4402 | dev->name, (dev->flags & IFF_PROMISC), | |
4403 | (old_flags & IFF_PROMISC), | |
4404 | audit_get_loginuid(current), | |
8192b0c4 | 4405 | uid, gid, |
7759db82 | 4406 | audit_get_sessionid(current)); |
8192b0c4 | 4407 | } |
24023451 | 4408 | |
b6c40d68 | 4409 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 4410 | } |
dad9b335 | 4411 | return 0; |
1da177e4 LT |
4412 | } |
4413 | ||
4417da66 PM |
4414 | /** |
4415 | * dev_set_promiscuity - update promiscuity count on a device | |
4416 | * @dev: device | |
4417 | * @inc: modifier | |
4418 | * | |
4419 | * Add or remove promiscuity from a device. While the count in the device | |
4420 | * remains above zero the interface remains promiscuous. Once it hits zero | |
4421 | * the device reverts back to normal filtering operation. A negative inc | |
4422 | * value is used to drop promiscuity on the device. | |
dad9b335 | 4423 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 4424 | */ |
dad9b335 | 4425 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 PM |
4426 | { |
4427 | unsigned short old_flags = dev->flags; | |
dad9b335 | 4428 | int err; |
4417da66 | 4429 | |
dad9b335 | 4430 | err = __dev_set_promiscuity(dev, inc); |
4b5a698e | 4431 | if (err < 0) |
dad9b335 | 4432 | return err; |
4417da66 PM |
4433 | if (dev->flags != old_flags) |
4434 | dev_set_rx_mode(dev); | |
dad9b335 | 4435 | return err; |
4417da66 | 4436 | } |
d1b19dff | 4437 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 4438 | |
1da177e4 LT |
4439 | /** |
4440 | * dev_set_allmulti - update allmulti count on a device | |
4441 | * @dev: device | |
4442 | * @inc: modifier | |
4443 | * | |
4444 | * Add or remove reception of all multicast frames to a device. While the | |
4445 | * count in the device remains above zero the interface remains listening | |
4446 | * to all interfaces. Once it hits zero the device reverts back to normal | |
4447 | * filtering operation. A negative @inc value is used to drop the counter | |
4448 | * when releasing a resource needing all multicasts. | |
dad9b335 | 4449 | * Return 0 if successful or a negative errno code on error. |
1da177e4 LT |
4450 | */ |
4451 | ||
dad9b335 | 4452 | int dev_set_allmulti(struct net_device *dev, int inc) |
1da177e4 LT |
4453 | { |
4454 | unsigned short old_flags = dev->flags; | |
4455 | ||
24023451 PM |
4456 | ASSERT_RTNL(); |
4457 | ||
1da177e4 | 4458 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
4459 | dev->allmulti += inc; |
4460 | if (dev->allmulti == 0) { | |
4461 | /* | |
4462 | * Avoid overflow. | |
4463 | * If inc causes overflow, untouch allmulti and return error. | |
4464 | */ | |
4465 | if (inc < 0) | |
4466 | dev->flags &= ~IFF_ALLMULTI; | |
4467 | else { | |
4468 | dev->allmulti -= inc; | |
4469 | printk(KERN_WARNING "%s: allmulti touches roof, " | |
4470 | "set allmulti failed, allmulti feature of " | |
4471 | "device might be broken.\n", dev->name); | |
4472 | return -EOVERFLOW; | |
4473 | } | |
4474 | } | |
24023451 | 4475 | if (dev->flags ^ old_flags) { |
b6c40d68 | 4476 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 4477 | dev_set_rx_mode(dev); |
24023451 | 4478 | } |
dad9b335 | 4479 | return 0; |
4417da66 | 4480 | } |
d1b19dff | 4481 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
4482 | |
4483 | /* | |
4484 | * Upload unicast and multicast address lists to device and | |
4485 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 4486 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
4487 | * are present. |
4488 | */ | |
4489 | void __dev_set_rx_mode(struct net_device *dev) | |
4490 | { | |
d314774c SH |
4491 | const struct net_device_ops *ops = dev->netdev_ops; |
4492 | ||
4417da66 PM |
4493 | /* dev_open will call this function so the list will stay sane. */ |
4494 | if (!(dev->flags&IFF_UP)) | |
4495 | return; | |
4496 | ||
4497 | if (!netif_device_present(dev)) | |
40b77c94 | 4498 | return; |
4417da66 | 4499 | |
d314774c SH |
4500 | if (ops->ndo_set_rx_mode) |
4501 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
4502 | else { |
4503 | /* Unicast addresses changes may only happen under the rtnl, | |
4504 | * therefore calling __dev_set_promiscuity here is safe. | |
4505 | */ | |
32e7bfc4 | 4506 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
4417da66 | 4507 | __dev_set_promiscuity(dev, 1); |
2d348d1f | 4508 | dev->uc_promisc = true; |
32e7bfc4 | 4509 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
4417da66 | 4510 | __dev_set_promiscuity(dev, -1); |
2d348d1f | 4511 | dev->uc_promisc = false; |
4417da66 PM |
4512 | } |
4513 | ||
d314774c SH |
4514 | if (ops->ndo_set_multicast_list) |
4515 | ops->ndo_set_multicast_list(dev); | |
4417da66 PM |
4516 | } |
4517 | } | |
4518 | ||
4519 | void dev_set_rx_mode(struct net_device *dev) | |
4520 | { | |
b9e40857 | 4521 | netif_addr_lock_bh(dev); |
4417da66 | 4522 | __dev_set_rx_mode(dev); |
b9e40857 | 4523 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
4524 | } |
4525 | ||
8ae6daca DD |
4526 | /** |
4527 | * dev_ethtool_get_settings - call device's ethtool_ops::get_settings() | |
4528 | * @dev: device | |
4529 | * @cmd: memory area for ethtool_ops::get_settings() result | |
4530 | * | |
4531 | * The cmd arg is initialized properly (cleared and | |
4532 | * ethtool_cmd::cmd field set to ETHTOOL_GSET). | |
4533 | * | |
4534 | * Return device's ethtool_ops::get_settings() result value or | |
4535 | * -EOPNOTSUPP when device doesn't expose | |
4536 | * ethtool_ops::get_settings() operation. | |
4537 | */ | |
4538 | int dev_ethtool_get_settings(struct net_device *dev, | |
4539 | struct ethtool_cmd *cmd) | |
4540 | { | |
4541 | if (!dev->ethtool_ops || !dev->ethtool_ops->get_settings) | |
4542 | return -EOPNOTSUPP; | |
4543 | ||
4544 | memset(cmd, 0, sizeof(struct ethtool_cmd)); | |
4545 | cmd->cmd = ETHTOOL_GSET; | |
4546 | return dev->ethtool_ops->get_settings(dev, cmd); | |
4547 | } | |
4548 | EXPORT_SYMBOL(dev_ethtool_get_settings); | |
4549 | ||
f0db275a SH |
4550 | /** |
4551 | * dev_get_flags - get flags reported to userspace | |
4552 | * @dev: device | |
4553 | * | |
4554 | * Get the combination of flag bits exported through APIs to userspace. | |
4555 | */ | |
1da177e4 LT |
4556 | unsigned dev_get_flags(const struct net_device *dev) |
4557 | { | |
4558 | unsigned flags; | |
4559 | ||
4560 | flags = (dev->flags & ~(IFF_PROMISC | | |
4561 | IFF_ALLMULTI | | |
b00055aa SR |
4562 | IFF_RUNNING | |
4563 | IFF_LOWER_UP | | |
4564 | IFF_DORMANT)) | | |
1da177e4 LT |
4565 | (dev->gflags & (IFF_PROMISC | |
4566 | IFF_ALLMULTI)); | |
4567 | ||
b00055aa SR |
4568 | if (netif_running(dev)) { |
4569 | if (netif_oper_up(dev)) | |
4570 | flags |= IFF_RUNNING; | |
4571 | if (netif_carrier_ok(dev)) | |
4572 | flags |= IFF_LOWER_UP; | |
4573 | if (netif_dormant(dev)) | |
4574 | flags |= IFF_DORMANT; | |
4575 | } | |
1da177e4 LT |
4576 | |
4577 | return flags; | |
4578 | } | |
d1b19dff | 4579 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 4580 | |
bd380811 | 4581 | int __dev_change_flags(struct net_device *dev, unsigned int flags) |
1da177e4 | 4582 | { |
1da177e4 | 4583 | int old_flags = dev->flags; |
bd380811 | 4584 | int ret; |
1da177e4 | 4585 | |
24023451 PM |
4586 | ASSERT_RTNL(); |
4587 | ||
1da177e4 LT |
4588 | /* |
4589 | * Set the flags on our device. | |
4590 | */ | |
4591 | ||
4592 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
4593 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
4594 | IFF_AUTOMEDIA)) | | |
4595 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
4596 | IFF_ALLMULTI)); | |
4597 | ||
4598 | /* | |
4599 | * Load in the correct multicast list now the flags have changed. | |
4600 | */ | |
4601 | ||
b6c40d68 PM |
4602 | if ((old_flags ^ flags) & IFF_MULTICAST) |
4603 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 4604 | |
4417da66 | 4605 | dev_set_rx_mode(dev); |
1da177e4 LT |
4606 | |
4607 | /* | |
4608 | * Have we downed the interface. We handle IFF_UP ourselves | |
4609 | * according to user attempts to set it, rather than blindly | |
4610 | * setting it. | |
4611 | */ | |
4612 | ||
4613 | ret = 0; | |
4614 | if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */ | |
bd380811 | 4615 | ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev); |
1da177e4 LT |
4616 | |
4617 | if (!ret) | |
4417da66 | 4618 | dev_set_rx_mode(dev); |
1da177e4 LT |
4619 | } |
4620 | ||
1da177e4 | 4621 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff ED |
4622 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
4623 | ||
1da177e4 LT |
4624 | dev->gflags ^= IFF_PROMISC; |
4625 | dev_set_promiscuity(dev, inc); | |
4626 | } | |
4627 | ||
4628 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
4629 | is important. Some (broken) drivers set IFF_PROMISC, when | |
4630 | IFF_ALLMULTI is requested not asking us and not reporting. | |
4631 | */ | |
4632 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
4633 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
4634 | ||
1da177e4 LT |
4635 | dev->gflags ^= IFF_ALLMULTI; |
4636 | dev_set_allmulti(dev, inc); | |
4637 | } | |
4638 | ||
bd380811 PM |
4639 | return ret; |
4640 | } | |
4641 | ||
4642 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags) | |
4643 | { | |
4644 | unsigned int changes = dev->flags ^ old_flags; | |
4645 | ||
4646 | if (changes & IFF_UP) { | |
4647 | if (dev->flags & IFF_UP) | |
4648 | call_netdevice_notifiers(NETDEV_UP, dev); | |
4649 | else | |
4650 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
4651 | } | |
4652 | ||
4653 | if (dev->flags & IFF_UP && | |
4654 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) | |
4655 | call_netdevice_notifiers(NETDEV_CHANGE, dev); | |
4656 | } | |
4657 | ||
4658 | /** | |
4659 | * dev_change_flags - change device settings | |
4660 | * @dev: device | |
4661 | * @flags: device state flags | |
4662 | * | |
4663 | * Change settings on device based state flags. The flags are | |
4664 | * in the userspace exported format. | |
4665 | */ | |
4666 | int dev_change_flags(struct net_device *dev, unsigned flags) | |
4667 | { | |
4668 | int ret, changes; | |
4669 | int old_flags = dev->flags; | |
4670 | ||
4671 | ret = __dev_change_flags(dev, flags); | |
4672 | if (ret < 0) | |
4673 | return ret; | |
4674 | ||
4675 | changes = old_flags ^ dev->flags; | |
7c355f53 TG |
4676 | if (changes) |
4677 | rtmsg_ifinfo(RTM_NEWLINK, dev, changes); | |
1da177e4 | 4678 | |
bd380811 | 4679 | __dev_notify_flags(dev, old_flags); |
1da177e4 LT |
4680 | return ret; |
4681 | } | |
d1b19dff | 4682 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 4683 | |
f0db275a SH |
4684 | /** |
4685 | * dev_set_mtu - Change maximum transfer unit | |
4686 | * @dev: device | |
4687 | * @new_mtu: new transfer unit | |
4688 | * | |
4689 | * Change the maximum transfer size of the network device. | |
4690 | */ | |
1da177e4 LT |
4691 | int dev_set_mtu(struct net_device *dev, int new_mtu) |
4692 | { | |
d314774c | 4693 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
4694 | int err; |
4695 | ||
4696 | if (new_mtu == dev->mtu) | |
4697 | return 0; | |
4698 | ||
4699 | /* MTU must be positive. */ | |
4700 | if (new_mtu < 0) | |
4701 | return -EINVAL; | |
4702 | ||
4703 | if (!netif_device_present(dev)) | |
4704 | return -ENODEV; | |
4705 | ||
4706 | err = 0; | |
d314774c SH |
4707 | if (ops->ndo_change_mtu) |
4708 | err = ops->ndo_change_mtu(dev, new_mtu); | |
1da177e4 LT |
4709 | else |
4710 | dev->mtu = new_mtu; | |
d314774c | 4711 | |
1da177e4 | 4712 | if (!err && dev->flags & IFF_UP) |
056925ab | 4713 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); |
1da177e4 LT |
4714 | return err; |
4715 | } | |
d1b19dff | 4716 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 4717 | |
cbda10fa VD |
4718 | /** |
4719 | * dev_set_group - Change group this device belongs to | |
4720 | * @dev: device | |
4721 | * @new_group: group this device should belong to | |
4722 | */ | |
4723 | void dev_set_group(struct net_device *dev, int new_group) | |
4724 | { | |
4725 | dev->group = new_group; | |
4726 | } | |
4727 | EXPORT_SYMBOL(dev_set_group); | |
4728 | ||
f0db275a SH |
4729 | /** |
4730 | * dev_set_mac_address - Change Media Access Control Address | |
4731 | * @dev: device | |
4732 | * @sa: new address | |
4733 | * | |
4734 | * Change the hardware (MAC) address of the device | |
4735 | */ | |
1da177e4 LT |
4736 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) |
4737 | { | |
d314774c | 4738 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
4739 | int err; |
4740 | ||
d314774c | 4741 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
4742 | return -EOPNOTSUPP; |
4743 | if (sa->sa_family != dev->type) | |
4744 | return -EINVAL; | |
4745 | if (!netif_device_present(dev)) | |
4746 | return -ENODEV; | |
d314774c | 4747 | err = ops->ndo_set_mac_address(dev, sa); |
1da177e4 | 4748 | if (!err) |
056925ab | 4749 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
1da177e4 LT |
4750 | return err; |
4751 | } | |
d1b19dff | 4752 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 LT |
4753 | |
4754 | /* | |
3710becf | 4755 | * Perform the SIOCxIFxxx calls, inside rcu_read_lock() |
1da177e4 | 4756 | */ |
14e3e079 | 4757 | static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd) |
1da177e4 LT |
4758 | { |
4759 | int err; | |
3710becf | 4760 | struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name); |
1da177e4 LT |
4761 | |
4762 | if (!dev) | |
4763 | return -ENODEV; | |
4764 | ||
4765 | switch (cmd) { | |
d1b19dff ED |
4766 | case SIOCGIFFLAGS: /* Get interface flags */ |
4767 | ifr->ifr_flags = (short) dev_get_flags(dev); | |
4768 | return 0; | |
1da177e4 | 4769 | |
d1b19dff ED |
4770 | case SIOCGIFMETRIC: /* Get the metric on the interface |
4771 | (currently unused) */ | |
4772 | ifr->ifr_metric = 0; | |
4773 | return 0; | |
1da177e4 | 4774 | |
d1b19dff ED |
4775 | case SIOCGIFMTU: /* Get the MTU of a device */ |
4776 | ifr->ifr_mtu = dev->mtu; | |
4777 | return 0; | |
1da177e4 | 4778 | |
d1b19dff ED |
4779 | case SIOCGIFHWADDR: |
4780 | if (!dev->addr_len) | |
4781 | memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data); | |
4782 | else | |
4783 | memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr, | |
4784 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
4785 | ifr->ifr_hwaddr.sa_family = dev->type; | |
4786 | return 0; | |
1da177e4 | 4787 | |
d1b19dff ED |
4788 | case SIOCGIFSLAVE: |
4789 | err = -EINVAL; | |
4790 | break; | |
14e3e079 | 4791 | |
d1b19dff ED |
4792 | case SIOCGIFMAP: |
4793 | ifr->ifr_map.mem_start = dev->mem_start; | |
4794 | ifr->ifr_map.mem_end = dev->mem_end; | |
4795 | ifr->ifr_map.base_addr = dev->base_addr; | |
4796 | ifr->ifr_map.irq = dev->irq; | |
4797 | ifr->ifr_map.dma = dev->dma; | |
4798 | ifr->ifr_map.port = dev->if_port; | |
4799 | return 0; | |
14e3e079 | 4800 | |
d1b19dff ED |
4801 | case SIOCGIFINDEX: |
4802 | ifr->ifr_ifindex = dev->ifindex; | |
4803 | return 0; | |
14e3e079 | 4804 | |
d1b19dff ED |
4805 | case SIOCGIFTXQLEN: |
4806 | ifr->ifr_qlen = dev->tx_queue_len; | |
4807 | return 0; | |
14e3e079 | 4808 | |
d1b19dff ED |
4809 | default: |
4810 | /* dev_ioctl() should ensure this case | |
4811 | * is never reached | |
4812 | */ | |
4813 | WARN_ON(1); | |
41c31f31 | 4814 | err = -ENOTTY; |
d1b19dff | 4815 | break; |
14e3e079 JG |
4816 | |
4817 | } | |
4818 | return err; | |
4819 | } | |
4820 | ||
4821 | /* | |
4822 | * Perform the SIOCxIFxxx calls, inside rtnl_lock() | |
4823 | */ | |
4824 | static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd) | |
4825 | { | |
4826 | int err; | |
4827 | struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name); | |
5f2f6da7 | 4828 | const struct net_device_ops *ops; |
14e3e079 JG |
4829 | |
4830 | if (!dev) | |
4831 | return -ENODEV; | |
4832 | ||
5f2f6da7 JP |
4833 | ops = dev->netdev_ops; |
4834 | ||
14e3e079 | 4835 | switch (cmd) { |
d1b19dff ED |
4836 | case SIOCSIFFLAGS: /* Set interface flags */ |
4837 | return dev_change_flags(dev, ifr->ifr_flags); | |
14e3e079 | 4838 | |
d1b19dff ED |
4839 | case SIOCSIFMETRIC: /* Set the metric on the interface |
4840 | (currently unused) */ | |
4841 | return -EOPNOTSUPP; | |
14e3e079 | 4842 | |
d1b19dff ED |
4843 | case SIOCSIFMTU: /* Set the MTU of a device */ |
4844 | return dev_set_mtu(dev, ifr->ifr_mtu); | |
1da177e4 | 4845 | |
d1b19dff ED |
4846 | case SIOCSIFHWADDR: |
4847 | return dev_set_mac_address(dev, &ifr->ifr_hwaddr); | |
1da177e4 | 4848 | |
d1b19dff ED |
4849 | case SIOCSIFHWBROADCAST: |
4850 | if (ifr->ifr_hwaddr.sa_family != dev->type) | |
4851 | return -EINVAL; | |
4852 | memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data, | |
4853 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
4854 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); | |
4855 | return 0; | |
1da177e4 | 4856 | |
d1b19dff ED |
4857 | case SIOCSIFMAP: |
4858 | if (ops->ndo_set_config) { | |
1da177e4 LT |
4859 | if (!netif_device_present(dev)) |
4860 | return -ENODEV; | |
d1b19dff ED |
4861 | return ops->ndo_set_config(dev, &ifr->ifr_map); |
4862 | } | |
4863 | return -EOPNOTSUPP; | |
1da177e4 | 4864 | |
d1b19dff ED |
4865 | case SIOCADDMULTI: |
4866 | if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) || | |
4867 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) | |
4868 | return -EINVAL; | |
4869 | if (!netif_device_present(dev)) | |
4870 | return -ENODEV; | |
22bedad3 | 4871 | return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data); |
d1b19dff ED |
4872 | |
4873 | case SIOCDELMULTI: | |
4874 | if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) || | |
4875 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) | |
4876 | return -EINVAL; | |
4877 | if (!netif_device_present(dev)) | |
4878 | return -ENODEV; | |
22bedad3 | 4879 | return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data); |
1da177e4 | 4880 | |
d1b19dff ED |
4881 | case SIOCSIFTXQLEN: |
4882 | if (ifr->ifr_qlen < 0) | |
4883 | return -EINVAL; | |
4884 | dev->tx_queue_len = ifr->ifr_qlen; | |
4885 | return 0; | |
1da177e4 | 4886 | |
d1b19dff ED |
4887 | case SIOCSIFNAME: |
4888 | ifr->ifr_newname[IFNAMSIZ-1] = '\0'; | |
4889 | return dev_change_name(dev, ifr->ifr_newname); | |
1da177e4 | 4890 | |
d1b19dff ED |
4891 | /* |
4892 | * Unknown or private ioctl | |
4893 | */ | |
4894 | default: | |
4895 | if ((cmd >= SIOCDEVPRIVATE && | |
4896 | cmd <= SIOCDEVPRIVATE + 15) || | |
4897 | cmd == SIOCBONDENSLAVE || | |
4898 | cmd == SIOCBONDRELEASE || | |
4899 | cmd == SIOCBONDSETHWADDR || | |
4900 | cmd == SIOCBONDSLAVEINFOQUERY || | |
4901 | cmd == SIOCBONDINFOQUERY || | |
4902 | cmd == SIOCBONDCHANGEACTIVE || | |
4903 | cmd == SIOCGMIIPHY || | |
4904 | cmd == SIOCGMIIREG || | |
4905 | cmd == SIOCSMIIREG || | |
4906 | cmd == SIOCBRADDIF || | |
4907 | cmd == SIOCBRDELIF || | |
4908 | cmd == SIOCSHWTSTAMP || | |
4909 | cmd == SIOCWANDEV) { | |
4910 | err = -EOPNOTSUPP; | |
4911 | if (ops->ndo_do_ioctl) { | |
4912 | if (netif_device_present(dev)) | |
4913 | err = ops->ndo_do_ioctl(dev, ifr, cmd); | |
4914 | else | |
4915 | err = -ENODEV; | |
4916 | } | |
4917 | } else | |
4918 | err = -EINVAL; | |
1da177e4 LT |
4919 | |
4920 | } | |
4921 | return err; | |
4922 | } | |
4923 | ||
4924 | /* | |
4925 | * This function handles all "interface"-type I/O control requests. The actual | |
4926 | * 'doing' part of this is dev_ifsioc above. | |
4927 | */ | |
4928 | ||
4929 | /** | |
4930 | * dev_ioctl - network device ioctl | |
c4ea43c5 | 4931 | * @net: the applicable net namespace |
1da177e4 LT |
4932 | * @cmd: command to issue |
4933 | * @arg: pointer to a struct ifreq in user space | |
4934 | * | |
4935 | * Issue ioctl functions to devices. This is normally called by the | |
4936 | * user space syscall interfaces but can sometimes be useful for | |
4937 | * other purposes. The return value is the return from the syscall if | |
4938 | * positive or a negative errno code on error. | |
4939 | */ | |
4940 | ||
881d966b | 4941 | int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg) |
1da177e4 LT |
4942 | { |
4943 | struct ifreq ifr; | |
4944 | int ret; | |
4945 | char *colon; | |
4946 | ||
4947 | /* One special case: SIOCGIFCONF takes ifconf argument | |
4948 | and requires shared lock, because it sleeps writing | |
4949 | to user space. | |
4950 | */ | |
4951 | ||
4952 | if (cmd == SIOCGIFCONF) { | |
6756ae4b | 4953 | rtnl_lock(); |
881d966b | 4954 | ret = dev_ifconf(net, (char __user *) arg); |
6756ae4b | 4955 | rtnl_unlock(); |
1da177e4 LT |
4956 | return ret; |
4957 | } | |
4958 | if (cmd == SIOCGIFNAME) | |
881d966b | 4959 | return dev_ifname(net, (struct ifreq __user *)arg); |
1da177e4 LT |
4960 | |
4961 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
4962 | return -EFAULT; | |
4963 | ||
4964 | ifr.ifr_name[IFNAMSIZ-1] = 0; | |
4965 | ||
4966 | colon = strchr(ifr.ifr_name, ':'); | |
4967 | if (colon) | |
4968 | *colon = 0; | |
4969 | ||
4970 | /* | |
4971 | * See which interface the caller is talking about. | |
4972 | */ | |
4973 | ||
4974 | switch (cmd) { | |
d1b19dff ED |
4975 | /* |
4976 | * These ioctl calls: | |
4977 | * - can be done by all. | |
4978 | * - atomic and do not require locking. | |
4979 | * - return a value | |
4980 | */ | |
4981 | case SIOCGIFFLAGS: | |
4982 | case SIOCGIFMETRIC: | |
4983 | case SIOCGIFMTU: | |
4984 | case SIOCGIFHWADDR: | |
4985 | case SIOCGIFSLAVE: | |
4986 | case SIOCGIFMAP: | |
4987 | case SIOCGIFINDEX: | |
4988 | case SIOCGIFTXQLEN: | |
4989 | dev_load(net, ifr.ifr_name); | |
3710becf | 4990 | rcu_read_lock(); |
d1b19dff | 4991 | ret = dev_ifsioc_locked(net, &ifr, cmd); |
3710becf | 4992 | rcu_read_unlock(); |
d1b19dff ED |
4993 | if (!ret) { |
4994 | if (colon) | |
4995 | *colon = ':'; | |
4996 | if (copy_to_user(arg, &ifr, | |
4997 | sizeof(struct ifreq))) | |
4998 | ret = -EFAULT; | |
4999 | } | |
5000 | return ret; | |
1da177e4 | 5001 | |
d1b19dff ED |
5002 | case SIOCETHTOOL: |
5003 | dev_load(net, ifr.ifr_name); | |
5004 | rtnl_lock(); | |
5005 | ret = dev_ethtool(net, &ifr); | |
5006 | rtnl_unlock(); | |
5007 | if (!ret) { | |
5008 | if (colon) | |
5009 | *colon = ':'; | |
5010 | if (copy_to_user(arg, &ifr, | |
5011 | sizeof(struct ifreq))) | |
5012 | ret = -EFAULT; | |
5013 | } | |
5014 | return ret; | |
1da177e4 | 5015 | |
d1b19dff ED |
5016 | /* |
5017 | * These ioctl calls: | |
5018 | * - require superuser power. | |
5019 | * - require strict serialization. | |
5020 | * - return a value | |
5021 | */ | |
5022 | case SIOCGMIIPHY: | |
5023 | case SIOCGMIIREG: | |
5024 | case SIOCSIFNAME: | |
5025 | if (!capable(CAP_NET_ADMIN)) | |
5026 | return -EPERM; | |
5027 | dev_load(net, ifr.ifr_name); | |
5028 | rtnl_lock(); | |
5029 | ret = dev_ifsioc(net, &ifr, cmd); | |
5030 | rtnl_unlock(); | |
5031 | if (!ret) { | |
5032 | if (colon) | |
5033 | *colon = ':'; | |
5034 | if (copy_to_user(arg, &ifr, | |
5035 | sizeof(struct ifreq))) | |
5036 | ret = -EFAULT; | |
5037 | } | |
5038 | return ret; | |
1da177e4 | 5039 | |
d1b19dff ED |
5040 | /* |
5041 | * These ioctl calls: | |
5042 | * - require superuser power. | |
5043 | * - require strict serialization. | |
5044 | * - do not return a value | |
5045 | */ | |
5046 | case SIOCSIFFLAGS: | |
5047 | case SIOCSIFMETRIC: | |
5048 | case SIOCSIFMTU: | |
5049 | case SIOCSIFMAP: | |
5050 | case SIOCSIFHWADDR: | |
5051 | case SIOCSIFSLAVE: | |
5052 | case SIOCADDMULTI: | |
5053 | case SIOCDELMULTI: | |
5054 | case SIOCSIFHWBROADCAST: | |
5055 | case SIOCSIFTXQLEN: | |
5056 | case SIOCSMIIREG: | |
5057 | case SIOCBONDENSLAVE: | |
5058 | case SIOCBONDRELEASE: | |
5059 | case SIOCBONDSETHWADDR: | |
5060 | case SIOCBONDCHANGEACTIVE: | |
5061 | case SIOCBRADDIF: | |
5062 | case SIOCBRDELIF: | |
5063 | case SIOCSHWTSTAMP: | |
5064 | if (!capable(CAP_NET_ADMIN)) | |
5065 | return -EPERM; | |
5066 | /* fall through */ | |
5067 | case SIOCBONDSLAVEINFOQUERY: | |
5068 | case SIOCBONDINFOQUERY: | |
5069 | dev_load(net, ifr.ifr_name); | |
5070 | rtnl_lock(); | |
5071 | ret = dev_ifsioc(net, &ifr, cmd); | |
5072 | rtnl_unlock(); | |
5073 | return ret; | |
5074 | ||
5075 | case SIOCGIFMEM: | |
5076 | /* Get the per device memory space. We can add this but | |
5077 | * currently do not support it */ | |
5078 | case SIOCSIFMEM: | |
5079 | /* Set the per device memory buffer space. | |
5080 | * Not applicable in our case */ | |
5081 | case SIOCSIFLINK: | |
41c31f31 | 5082 | return -ENOTTY; |
d1b19dff ED |
5083 | |
5084 | /* | |
5085 | * Unknown or private ioctl. | |
5086 | */ | |
5087 | default: | |
5088 | if (cmd == SIOCWANDEV || | |
5089 | (cmd >= SIOCDEVPRIVATE && | |
5090 | cmd <= SIOCDEVPRIVATE + 15)) { | |
881d966b | 5091 | dev_load(net, ifr.ifr_name); |
1da177e4 | 5092 | rtnl_lock(); |
881d966b | 5093 | ret = dev_ifsioc(net, &ifr, cmd); |
1da177e4 | 5094 | rtnl_unlock(); |
d1b19dff ED |
5095 | if (!ret && copy_to_user(arg, &ifr, |
5096 | sizeof(struct ifreq))) | |
5097 | ret = -EFAULT; | |
1da177e4 | 5098 | return ret; |
d1b19dff ED |
5099 | } |
5100 | /* Take care of Wireless Extensions */ | |
5101 | if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) | |
5102 | return wext_handle_ioctl(net, &ifr, cmd, arg); | |
41c31f31 | 5103 | return -ENOTTY; |
1da177e4 LT |
5104 | } |
5105 | } | |
5106 | ||
5107 | ||
5108 | /** | |
5109 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 5110 | * @net: the applicable net namespace |
1da177e4 LT |
5111 | * |
5112 | * Returns a suitable unique value for a new device interface | |
5113 | * number. The caller must hold the rtnl semaphore or the | |
5114 | * dev_base_lock to be sure it remains unique. | |
5115 | */ | |
881d966b | 5116 | static int dev_new_index(struct net *net) |
1da177e4 LT |
5117 | { |
5118 | static int ifindex; | |
5119 | for (;;) { | |
5120 | if (++ifindex <= 0) | |
5121 | ifindex = 1; | |
881d966b | 5122 | if (!__dev_get_by_index(net, ifindex)) |
1da177e4 LT |
5123 | return ifindex; |
5124 | } | |
5125 | } | |
5126 | ||
1da177e4 | 5127 | /* Delayed registration/unregisteration */ |
3b5b34fd | 5128 | static LIST_HEAD(net_todo_list); |
1da177e4 | 5129 | |
6f05f629 | 5130 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 5131 | { |
1da177e4 | 5132 | list_add_tail(&dev->todo_list, &net_todo_list); |
1da177e4 LT |
5133 | } |
5134 | ||
9b5e383c | 5135 | static void rollback_registered_many(struct list_head *head) |
93ee31f1 | 5136 | { |
e93737b0 | 5137 | struct net_device *dev, *tmp; |
9b5e383c | 5138 | |
93ee31f1 DL |
5139 | BUG_ON(dev_boot_phase); |
5140 | ASSERT_RTNL(); | |
5141 | ||
e93737b0 | 5142 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
9b5e383c | 5143 | /* Some devices call without registering |
e93737b0 KK |
5144 | * for initialization unwind. Remove those |
5145 | * devices and proceed with the remaining. | |
9b5e383c ED |
5146 | */ |
5147 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
5148 | pr_debug("unregister_netdevice: device %s/%p never " | |
5149 | "was registered\n", dev->name, dev); | |
93ee31f1 | 5150 | |
9b5e383c | 5151 | WARN_ON(1); |
e93737b0 KK |
5152 | list_del(&dev->unreg_list); |
5153 | continue; | |
9b5e383c | 5154 | } |
449f4544 | 5155 | dev->dismantle = true; |
9b5e383c | 5156 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
44345724 | 5157 | } |
93ee31f1 | 5158 | |
44345724 OP |
5159 | /* If device is running, close it first. */ |
5160 | dev_close_many(head); | |
93ee31f1 | 5161 | |
44345724 | 5162 | list_for_each_entry(dev, head, unreg_list) { |
9b5e383c ED |
5163 | /* And unlink it from device chain. */ |
5164 | unlist_netdevice(dev); | |
93ee31f1 | 5165 | |
9b5e383c ED |
5166 | dev->reg_state = NETREG_UNREGISTERING; |
5167 | } | |
93ee31f1 DL |
5168 | |
5169 | synchronize_net(); | |
5170 | ||
9b5e383c ED |
5171 | list_for_each_entry(dev, head, unreg_list) { |
5172 | /* Shutdown queueing discipline. */ | |
5173 | dev_shutdown(dev); | |
93ee31f1 DL |
5174 | |
5175 | ||
9b5e383c ED |
5176 | /* Notify protocols, that we are about to destroy |
5177 | this device. They should clean all the things. | |
5178 | */ | |
5179 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
93ee31f1 | 5180 | |
a2835763 PM |
5181 | if (!dev->rtnl_link_ops || |
5182 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
5183 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U); | |
5184 | ||
9b5e383c ED |
5185 | /* |
5186 | * Flush the unicast and multicast chains | |
5187 | */ | |
a748ee24 | 5188 | dev_uc_flush(dev); |
22bedad3 | 5189 | dev_mc_flush(dev); |
93ee31f1 | 5190 | |
9b5e383c ED |
5191 | if (dev->netdev_ops->ndo_uninit) |
5192 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 | 5193 | |
9b5e383c ED |
5194 | /* Notifier chain MUST detach us from master device. */ |
5195 | WARN_ON(dev->master); | |
93ee31f1 | 5196 | |
9b5e383c ED |
5197 | /* Remove entries from kobject tree */ |
5198 | netdev_unregister_kobject(dev); | |
5199 | } | |
93ee31f1 | 5200 | |
a5ee1551 | 5201 | /* Process any work delayed until the end of the batch */ |
e5e26d75 | 5202 | dev = list_first_entry(head, struct net_device, unreg_list); |
a5ee1551 | 5203 | call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev); |
93ee31f1 | 5204 | |
ef885afb | 5205 | rcu_barrier(); |
395264d5 | 5206 | |
a5ee1551 | 5207 | list_for_each_entry(dev, head, unreg_list) |
9b5e383c ED |
5208 | dev_put(dev); |
5209 | } | |
5210 | ||
5211 | static void rollback_registered(struct net_device *dev) | |
5212 | { | |
5213 | LIST_HEAD(single); | |
5214 | ||
5215 | list_add(&dev->unreg_list, &single); | |
5216 | rollback_registered_many(&single); | |
ceaaec98 | 5217 | list_del(&single); |
93ee31f1 DL |
5218 | } |
5219 | ||
fec30c33 | 5220 | static u32 netdev_fix_features(struct net_device *dev, u32 features) |
b63365a2 | 5221 | { |
57422dc5 MM |
5222 | /* Fix illegal checksum combinations */ |
5223 | if ((features & NETIF_F_HW_CSUM) && | |
5224 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 5225 | netdev_warn(dev, "mixed HW and IP checksum settings.\n"); |
57422dc5 MM |
5226 | features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); |
5227 | } | |
5228 | ||
5229 | if ((features & NETIF_F_NO_CSUM) && | |
5230 | (features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 5231 | netdev_warn(dev, "mixed no checksumming and other settings.\n"); |
57422dc5 MM |
5232 | features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM); |
5233 | } | |
5234 | ||
b63365a2 HX |
5235 | /* Fix illegal SG+CSUM combinations. */ |
5236 | if ((features & NETIF_F_SG) && | |
5237 | !(features & NETIF_F_ALL_CSUM)) { | |
6f404e44 MM |
5238 | netdev_dbg(dev, |
5239 | "Dropping NETIF_F_SG since no checksum feature.\n"); | |
b63365a2 HX |
5240 | features &= ~NETIF_F_SG; |
5241 | } | |
5242 | ||
5243 | /* TSO requires that SG is present as well. */ | |
ea2d3688 | 5244 | if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { |
6f404e44 | 5245 | netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); |
ea2d3688 | 5246 | features &= ~NETIF_F_ALL_TSO; |
b63365a2 HX |
5247 | } |
5248 | ||
31d8b9e0 BH |
5249 | /* TSO ECN requires that TSO is present as well. */ |
5250 | if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) | |
5251 | features &= ~NETIF_F_TSO_ECN; | |
5252 | ||
212b573f MM |
5253 | /* Software GSO depends on SG. */ |
5254 | if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { | |
6f404e44 | 5255 | netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); |
212b573f MM |
5256 | features &= ~NETIF_F_GSO; |
5257 | } | |
5258 | ||
acd1130e | 5259 | /* UFO needs SG and checksumming */ |
b63365a2 | 5260 | if (features & NETIF_F_UFO) { |
79032644 MM |
5261 | /* maybe split UFO into V4 and V6? */ |
5262 | if (!((features & NETIF_F_GEN_CSUM) || | |
5263 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM)) | |
5264 | == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 5265 | netdev_dbg(dev, |
acd1130e | 5266 | "Dropping NETIF_F_UFO since no checksum offload features.\n"); |
b63365a2 HX |
5267 | features &= ~NETIF_F_UFO; |
5268 | } | |
5269 | ||
5270 | if (!(features & NETIF_F_SG)) { | |
6f404e44 | 5271 | netdev_dbg(dev, |
acd1130e | 5272 | "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n"); |
b63365a2 HX |
5273 | features &= ~NETIF_F_UFO; |
5274 | } | |
5275 | } | |
5276 | ||
5277 | return features; | |
5278 | } | |
b63365a2 | 5279 | |
6cb6a27c | 5280 | int __netdev_update_features(struct net_device *dev) |
5455c699 MM |
5281 | { |
5282 | u32 features; | |
5283 | int err = 0; | |
5284 | ||
87267485 MM |
5285 | ASSERT_RTNL(); |
5286 | ||
5455c699 MM |
5287 | features = netdev_get_wanted_features(dev); |
5288 | ||
5289 | if (dev->netdev_ops->ndo_fix_features) | |
5290 | features = dev->netdev_ops->ndo_fix_features(dev, features); | |
5291 | ||
5292 | /* driver might be less strict about feature dependencies */ | |
5293 | features = netdev_fix_features(dev, features); | |
5294 | ||
5295 | if (dev->features == features) | |
6cb6a27c | 5296 | return 0; |
5455c699 | 5297 | |
604ae14f | 5298 | netdev_dbg(dev, "Features changed: 0x%08x -> 0x%08x\n", |
5455c699 MM |
5299 | dev->features, features); |
5300 | ||
5301 | if (dev->netdev_ops->ndo_set_features) | |
5302 | err = dev->netdev_ops->ndo_set_features(dev, features); | |
5303 | ||
6cb6a27c | 5304 | if (unlikely(err < 0)) { |
5455c699 MM |
5305 | netdev_err(dev, |
5306 | "set_features() failed (%d); wanted 0x%08x, left 0x%08x\n", | |
5307 | err, features, dev->features); | |
6cb6a27c MM |
5308 | return -1; |
5309 | } | |
5310 | ||
5311 | if (!err) | |
5312 | dev->features = features; | |
5313 | ||
5314 | return 1; | |
5315 | } | |
5316 | ||
afe12cc8 MM |
5317 | /** |
5318 | * netdev_update_features - recalculate device features | |
5319 | * @dev: the device to check | |
5320 | * | |
5321 | * Recalculate dev->features set and send notifications if it | |
5322 | * has changed. Should be called after driver or hardware dependent | |
5323 | * conditions might have changed that influence the features. | |
5324 | */ | |
6cb6a27c MM |
5325 | void netdev_update_features(struct net_device *dev) |
5326 | { | |
5327 | if (__netdev_update_features(dev)) | |
5328 | netdev_features_change(dev); | |
5455c699 MM |
5329 | } |
5330 | EXPORT_SYMBOL(netdev_update_features); | |
5331 | ||
afe12cc8 MM |
5332 | /** |
5333 | * netdev_change_features - recalculate device features | |
5334 | * @dev: the device to check | |
5335 | * | |
5336 | * Recalculate dev->features set and send notifications even | |
5337 | * if they have not changed. Should be called instead of | |
5338 | * netdev_update_features() if also dev->vlan_features might | |
5339 | * have changed to allow the changes to be propagated to stacked | |
5340 | * VLAN devices. | |
5341 | */ | |
5342 | void netdev_change_features(struct net_device *dev) | |
5343 | { | |
5344 | __netdev_update_features(dev); | |
5345 | netdev_features_change(dev); | |
5346 | } | |
5347 | EXPORT_SYMBOL(netdev_change_features); | |
5348 | ||
fc4a7489 PM |
5349 | /** |
5350 | * netif_stacked_transfer_operstate - transfer operstate | |
5351 | * @rootdev: the root or lower level device to transfer state from | |
5352 | * @dev: the device to transfer operstate to | |
5353 | * | |
5354 | * Transfer operational state from root to device. This is normally | |
5355 | * called when a stacking relationship exists between the root | |
5356 | * device and the device(a leaf device). | |
5357 | */ | |
5358 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
5359 | struct net_device *dev) | |
5360 | { | |
5361 | if (rootdev->operstate == IF_OPER_DORMANT) | |
5362 | netif_dormant_on(dev); | |
5363 | else | |
5364 | netif_dormant_off(dev); | |
5365 | ||
5366 | if (netif_carrier_ok(rootdev)) { | |
5367 | if (!netif_carrier_ok(dev)) | |
5368 | netif_carrier_on(dev); | |
5369 | } else { | |
5370 | if (netif_carrier_ok(dev)) | |
5371 | netif_carrier_off(dev); | |
5372 | } | |
5373 | } | |
5374 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
5375 | ||
bf264145 | 5376 | #ifdef CONFIG_RPS |
1b4bf461 ED |
5377 | static int netif_alloc_rx_queues(struct net_device *dev) |
5378 | { | |
1b4bf461 | 5379 | unsigned int i, count = dev->num_rx_queues; |
bd25fa7b | 5380 | struct netdev_rx_queue *rx; |
1b4bf461 | 5381 | |
bd25fa7b | 5382 | BUG_ON(count < 1); |
1b4bf461 | 5383 | |
bd25fa7b TH |
5384 | rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL); |
5385 | if (!rx) { | |
5386 | pr_err("netdev: Unable to allocate %u rx queues.\n", count); | |
5387 | return -ENOMEM; | |
1b4bf461 | 5388 | } |
bd25fa7b TH |
5389 | dev->_rx = rx; |
5390 | ||
bd25fa7b | 5391 | for (i = 0; i < count; i++) |
fe822240 | 5392 | rx[i].dev = dev; |
1b4bf461 ED |
5393 | return 0; |
5394 | } | |
bf264145 | 5395 | #endif |
1b4bf461 | 5396 | |
aa942104 CG |
5397 | static void netdev_init_one_queue(struct net_device *dev, |
5398 | struct netdev_queue *queue, void *_unused) | |
5399 | { | |
5400 | /* Initialize queue lock */ | |
5401 | spin_lock_init(&queue->_xmit_lock); | |
5402 | netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); | |
5403 | queue->xmit_lock_owner = -1; | |
b236da69 | 5404 | netdev_queue_numa_node_write(queue, NUMA_NO_NODE); |
aa942104 CG |
5405 | queue->dev = dev; |
5406 | } | |
5407 | ||
e6484930 TH |
5408 | static int netif_alloc_netdev_queues(struct net_device *dev) |
5409 | { | |
5410 | unsigned int count = dev->num_tx_queues; | |
5411 | struct netdev_queue *tx; | |
5412 | ||
5413 | BUG_ON(count < 1); | |
5414 | ||
5415 | tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL); | |
5416 | if (!tx) { | |
5417 | pr_err("netdev: Unable to allocate %u tx queues.\n", | |
5418 | count); | |
5419 | return -ENOMEM; | |
5420 | } | |
5421 | dev->_tx = tx; | |
1d24eb48 | 5422 | |
e6484930 TH |
5423 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); |
5424 | spin_lock_init(&dev->tx_global_lock); | |
aa942104 CG |
5425 | |
5426 | return 0; | |
e6484930 TH |
5427 | } |
5428 | ||
1da177e4 LT |
5429 | /** |
5430 | * register_netdevice - register a network device | |
5431 | * @dev: device to register | |
5432 | * | |
5433 | * Take a completed network device structure and add it to the kernel | |
5434 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
5435 | * chain. 0 is returned on success. A negative errno code is returned | |
5436 | * on a failure to set up the device, or if the name is a duplicate. | |
5437 | * | |
5438 | * Callers must hold the rtnl semaphore. You may want | |
5439 | * register_netdev() instead of this. | |
5440 | * | |
5441 | * BUGS: | |
5442 | * The locking appears insufficient to guarantee two parallel registers | |
5443 | * will not get the same name. | |
5444 | */ | |
5445 | ||
5446 | int register_netdevice(struct net_device *dev) | |
5447 | { | |
1da177e4 | 5448 | int ret; |
d314774c | 5449 | struct net *net = dev_net(dev); |
1da177e4 LT |
5450 | |
5451 | BUG_ON(dev_boot_phase); | |
5452 | ASSERT_RTNL(); | |
5453 | ||
b17a7c17 SH |
5454 | might_sleep(); |
5455 | ||
1da177e4 LT |
5456 | /* When net_device's are persistent, this will be fatal. */ |
5457 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 5458 | BUG_ON(!net); |
1da177e4 | 5459 | |
f1f28aa3 | 5460 | spin_lock_init(&dev->addr_list_lock); |
cf508b12 | 5461 | netdev_set_addr_lockdep_class(dev); |
1da177e4 | 5462 | |
1da177e4 LT |
5463 | dev->iflink = -1; |
5464 | ||
0696c3a8 PP |
5465 | ret = dev_get_valid_name(dev, dev->name); |
5466 | if (ret < 0) | |
5467 | goto out; | |
5468 | ||
1da177e4 | 5469 | /* Init, if this function is available */ |
d314774c SH |
5470 | if (dev->netdev_ops->ndo_init) { |
5471 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
5472 | if (ret) { |
5473 | if (ret > 0) | |
5474 | ret = -EIO; | |
90833aa4 | 5475 | goto out; |
1da177e4 LT |
5476 | } |
5477 | } | |
4ec93edb | 5478 | |
881d966b | 5479 | dev->ifindex = dev_new_index(net); |
1da177e4 LT |
5480 | if (dev->iflink == -1) |
5481 | dev->iflink = dev->ifindex; | |
5482 | ||
5455c699 MM |
5483 | /* Transfer changeable features to wanted_features and enable |
5484 | * software offloads (GSO and GRO). | |
5485 | */ | |
5486 | dev->hw_features |= NETIF_F_SOFT_FEATURES; | |
14d1232f MM |
5487 | dev->features |= NETIF_F_SOFT_FEATURES; |
5488 | dev->wanted_features = dev->features & dev->hw_features; | |
1da177e4 | 5489 | |
c6e1a0d1 TH |
5490 | /* Turn on no cache copy if HW is doing checksum */ |
5491 | dev->hw_features |= NETIF_F_NOCACHE_COPY; | |
5492 | if ((dev->features & NETIF_F_ALL_CSUM) && | |
5493 | !(dev->features & NETIF_F_NO_CSUM)) { | |
5494 | dev->wanted_features |= NETIF_F_NOCACHE_COPY; | |
5495 | dev->features |= NETIF_F_NOCACHE_COPY; | |
5496 | } | |
5497 | ||
1180e7d6 | 5498 | /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. |
16c3ea78 | 5499 | */ |
1180e7d6 | 5500 | dev->vlan_features |= NETIF_F_HIGHDMA; |
16c3ea78 | 5501 | |
7ffbe3fd JB |
5502 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); |
5503 | ret = notifier_to_errno(ret); | |
5504 | if (ret) | |
5505 | goto err_uninit; | |
5506 | ||
8b41d188 | 5507 | ret = netdev_register_kobject(dev); |
b17a7c17 | 5508 | if (ret) |
7ce1b0ed | 5509 | goto err_uninit; |
b17a7c17 SH |
5510 | dev->reg_state = NETREG_REGISTERED; |
5511 | ||
6cb6a27c | 5512 | __netdev_update_features(dev); |
8e9b59b2 | 5513 | |
1da177e4 LT |
5514 | /* |
5515 | * Default initial state at registry is that the | |
5516 | * device is present. | |
5517 | */ | |
5518 | ||
5519 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
5520 | ||
1da177e4 | 5521 | dev_init_scheduler(dev); |
1da177e4 | 5522 | dev_hold(dev); |
ce286d32 | 5523 | list_netdevice(dev); |
1da177e4 LT |
5524 | |
5525 | /* Notify protocols, that a new device appeared. */ | |
056925ab | 5526 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 5527 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
5528 | if (ret) { |
5529 | rollback_registered(dev); | |
5530 | dev->reg_state = NETREG_UNREGISTERED; | |
5531 | } | |
d90a909e EB |
5532 | /* |
5533 | * Prevent userspace races by waiting until the network | |
5534 | * device is fully setup before sending notifications. | |
5535 | */ | |
a2835763 PM |
5536 | if (!dev->rtnl_link_ops || |
5537 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
5538 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U); | |
1da177e4 LT |
5539 | |
5540 | out: | |
5541 | return ret; | |
7ce1b0ed HX |
5542 | |
5543 | err_uninit: | |
d314774c SH |
5544 | if (dev->netdev_ops->ndo_uninit) |
5545 | dev->netdev_ops->ndo_uninit(dev); | |
7ce1b0ed | 5546 | goto out; |
1da177e4 | 5547 | } |
d1b19dff | 5548 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 5549 | |
937f1ba5 BH |
5550 | /** |
5551 | * init_dummy_netdev - init a dummy network device for NAPI | |
5552 | * @dev: device to init | |
5553 | * | |
5554 | * This takes a network device structure and initialize the minimum | |
5555 | * amount of fields so it can be used to schedule NAPI polls without | |
5556 | * registering a full blown interface. This is to be used by drivers | |
5557 | * that need to tie several hardware interfaces to a single NAPI | |
5558 | * poll scheduler due to HW limitations. | |
5559 | */ | |
5560 | int init_dummy_netdev(struct net_device *dev) | |
5561 | { | |
5562 | /* Clear everything. Note we don't initialize spinlocks | |
5563 | * are they aren't supposed to be taken by any of the | |
5564 | * NAPI code and this dummy netdev is supposed to be | |
5565 | * only ever used for NAPI polls | |
5566 | */ | |
5567 | memset(dev, 0, sizeof(struct net_device)); | |
5568 | ||
5569 | /* make sure we BUG if trying to hit standard | |
5570 | * register/unregister code path | |
5571 | */ | |
5572 | dev->reg_state = NETREG_DUMMY; | |
5573 | ||
937f1ba5 BH |
5574 | /* NAPI wants this */ |
5575 | INIT_LIST_HEAD(&dev->napi_list); | |
5576 | ||
5577 | /* a dummy interface is started by default */ | |
5578 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
5579 | set_bit(__LINK_STATE_START, &dev->state); | |
5580 | ||
29b4433d ED |
5581 | /* Note : We dont allocate pcpu_refcnt for dummy devices, |
5582 | * because users of this 'device' dont need to change | |
5583 | * its refcount. | |
5584 | */ | |
5585 | ||
937f1ba5 BH |
5586 | return 0; |
5587 | } | |
5588 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
5589 | ||
5590 | ||
1da177e4 LT |
5591 | /** |
5592 | * register_netdev - register a network device | |
5593 | * @dev: device to register | |
5594 | * | |
5595 | * Take a completed network device structure and add it to the kernel | |
5596 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
5597 | * chain. 0 is returned on success. A negative errno code is returned | |
5598 | * on a failure to set up the device, or if the name is a duplicate. | |
5599 | * | |
38b4da38 | 5600 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
5601 | * and expands the device name if you passed a format string to |
5602 | * alloc_netdev. | |
5603 | */ | |
5604 | int register_netdev(struct net_device *dev) | |
5605 | { | |
5606 | int err; | |
5607 | ||
5608 | rtnl_lock(); | |
1da177e4 | 5609 | err = register_netdevice(dev); |
1da177e4 LT |
5610 | rtnl_unlock(); |
5611 | return err; | |
5612 | } | |
5613 | EXPORT_SYMBOL(register_netdev); | |
5614 | ||
29b4433d ED |
5615 | int netdev_refcnt_read(const struct net_device *dev) |
5616 | { | |
5617 | int i, refcnt = 0; | |
5618 | ||
5619 | for_each_possible_cpu(i) | |
5620 | refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); | |
5621 | return refcnt; | |
5622 | } | |
5623 | EXPORT_SYMBOL(netdev_refcnt_read); | |
5624 | ||
1da177e4 LT |
5625 | /* |
5626 | * netdev_wait_allrefs - wait until all references are gone. | |
5627 | * | |
5628 | * This is called when unregistering network devices. | |
5629 | * | |
5630 | * Any protocol or device that holds a reference should register | |
5631 | * for netdevice notification, and cleanup and put back the | |
5632 | * reference if they receive an UNREGISTER event. | |
5633 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 5634 | * call dev_put. |
1da177e4 LT |
5635 | */ |
5636 | static void netdev_wait_allrefs(struct net_device *dev) | |
5637 | { | |
5638 | unsigned long rebroadcast_time, warning_time; | |
29b4433d | 5639 | int refcnt; |
1da177e4 | 5640 | |
e014debe ED |
5641 | linkwatch_forget_dev(dev); |
5642 | ||
1da177e4 | 5643 | rebroadcast_time = warning_time = jiffies; |
29b4433d ED |
5644 | refcnt = netdev_refcnt_read(dev); |
5645 | ||
5646 | while (refcnt != 0) { | |
1da177e4 | 5647 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { |
6756ae4b | 5648 | rtnl_lock(); |
1da177e4 LT |
5649 | |
5650 | /* Rebroadcast unregister notification */ | |
056925ab | 5651 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
a5ee1551 | 5652 | /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users |
395264d5 | 5653 | * should have already handle it the first time */ |
1da177e4 LT |
5654 | |
5655 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, | |
5656 | &dev->state)) { | |
5657 | /* We must not have linkwatch events | |
5658 | * pending on unregister. If this | |
5659 | * happens, we simply run the queue | |
5660 | * unscheduled, resulting in a noop | |
5661 | * for this device. | |
5662 | */ | |
5663 | linkwatch_run_queue(); | |
5664 | } | |
5665 | ||
6756ae4b | 5666 | __rtnl_unlock(); |
1da177e4 LT |
5667 | |
5668 | rebroadcast_time = jiffies; | |
5669 | } | |
5670 | ||
5671 | msleep(250); | |
5672 | ||
29b4433d ED |
5673 | refcnt = netdev_refcnt_read(dev); |
5674 | ||
1da177e4 LT |
5675 | if (time_after(jiffies, warning_time + 10 * HZ)) { |
5676 | printk(KERN_EMERG "unregister_netdevice: " | |
5677 | "waiting for %s to become free. Usage " | |
5678 | "count = %d\n", | |
29b4433d | 5679 | dev->name, refcnt); |
1da177e4 LT |
5680 | warning_time = jiffies; |
5681 | } | |
5682 | } | |
5683 | } | |
5684 | ||
5685 | /* The sequence is: | |
5686 | * | |
5687 | * rtnl_lock(); | |
5688 | * ... | |
5689 | * register_netdevice(x1); | |
5690 | * register_netdevice(x2); | |
5691 | * ... | |
5692 | * unregister_netdevice(y1); | |
5693 | * unregister_netdevice(y2); | |
5694 | * ... | |
5695 | * rtnl_unlock(); | |
5696 | * free_netdev(y1); | |
5697 | * free_netdev(y2); | |
5698 | * | |
58ec3b4d | 5699 | * We are invoked by rtnl_unlock(). |
1da177e4 | 5700 | * This allows us to deal with problems: |
b17a7c17 | 5701 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
5702 | * without deadlocking with linkwatch via keventd. |
5703 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
5704 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
5705 | * |
5706 | * We must not return until all unregister events added during | |
5707 | * the interval the lock was held have been completed. | |
1da177e4 | 5708 | */ |
1da177e4 LT |
5709 | void netdev_run_todo(void) |
5710 | { | |
626ab0e6 | 5711 | struct list_head list; |
1da177e4 | 5712 | |
1da177e4 | 5713 | /* Snapshot list, allow later requests */ |
626ab0e6 | 5714 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
5715 | |
5716 | __rtnl_unlock(); | |
626ab0e6 | 5717 | |
1da177e4 LT |
5718 | while (!list_empty(&list)) { |
5719 | struct net_device *dev | |
e5e26d75 | 5720 | = list_first_entry(&list, struct net_device, todo_list); |
1da177e4 LT |
5721 | list_del(&dev->todo_list); |
5722 | ||
b17a7c17 SH |
5723 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
5724 | printk(KERN_ERR "network todo '%s' but state %d\n", | |
5725 | dev->name, dev->reg_state); | |
5726 | dump_stack(); | |
5727 | continue; | |
5728 | } | |
1da177e4 | 5729 | |
b17a7c17 | 5730 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 5731 | |
152102c7 | 5732 | on_each_cpu(flush_backlog, dev, 1); |
6e583ce5 | 5733 | |
b17a7c17 | 5734 | netdev_wait_allrefs(dev); |
1da177e4 | 5735 | |
b17a7c17 | 5736 | /* paranoia */ |
29b4433d | 5737 | BUG_ON(netdev_refcnt_read(dev)); |
95ae6b22 | 5738 | WARN_ON(rcu_dereference_raw(dev->ip_ptr)); |
198caeca | 5739 | WARN_ON(rcu_dereference_raw(dev->ip6_ptr)); |
547b792c | 5740 | WARN_ON(dev->dn_ptr); |
1da177e4 | 5741 | |
b17a7c17 SH |
5742 | if (dev->destructor) |
5743 | dev->destructor(dev); | |
9093bbb2 SH |
5744 | |
5745 | /* Free network device */ | |
5746 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 5747 | } |
1da177e4 LT |
5748 | } |
5749 | ||
3cfde79c BH |
5750 | /* Convert net_device_stats to rtnl_link_stats64. They have the same |
5751 | * fields in the same order, with only the type differing. | |
5752 | */ | |
5753 | static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, | |
5754 | const struct net_device_stats *netdev_stats) | |
5755 | { | |
5756 | #if BITS_PER_LONG == 64 | |
5757 | BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats)); | |
5758 | memcpy(stats64, netdev_stats, sizeof(*stats64)); | |
5759 | #else | |
5760 | size_t i, n = sizeof(*stats64) / sizeof(u64); | |
5761 | const unsigned long *src = (const unsigned long *)netdev_stats; | |
5762 | u64 *dst = (u64 *)stats64; | |
5763 | ||
5764 | BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) != | |
5765 | sizeof(*stats64) / sizeof(u64)); | |
5766 | for (i = 0; i < n; i++) | |
5767 | dst[i] = src[i]; | |
5768 | #endif | |
5769 | } | |
5770 | ||
eeda3fd6 SH |
5771 | /** |
5772 | * dev_get_stats - get network device statistics | |
5773 | * @dev: device to get statistics from | |
28172739 | 5774 | * @storage: place to store stats |
eeda3fd6 | 5775 | * |
d7753516 BH |
5776 | * Get network statistics from device. Return @storage. |
5777 | * The device driver may provide its own method by setting | |
5778 | * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; | |
5779 | * otherwise the internal statistics structure is used. | |
eeda3fd6 | 5780 | */ |
d7753516 BH |
5781 | struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, |
5782 | struct rtnl_link_stats64 *storage) | |
7004bf25 | 5783 | { |
eeda3fd6 SH |
5784 | const struct net_device_ops *ops = dev->netdev_ops; |
5785 | ||
28172739 ED |
5786 | if (ops->ndo_get_stats64) { |
5787 | memset(storage, 0, sizeof(*storage)); | |
caf586e5 ED |
5788 | ops->ndo_get_stats64(dev, storage); |
5789 | } else if (ops->ndo_get_stats) { | |
3cfde79c | 5790 | netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); |
caf586e5 ED |
5791 | } else { |
5792 | netdev_stats_to_stats64(storage, &dev->stats); | |
28172739 | 5793 | } |
caf586e5 | 5794 | storage->rx_dropped += atomic_long_read(&dev->rx_dropped); |
28172739 | 5795 | return storage; |
c45d286e | 5796 | } |
eeda3fd6 | 5797 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 5798 | |
24824a09 | 5799 | struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) |
dc2b4847 | 5800 | { |
24824a09 | 5801 | struct netdev_queue *queue = dev_ingress_queue(dev); |
dc2b4847 | 5802 | |
24824a09 ED |
5803 | #ifdef CONFIG_NET_CLS_ACT |
5804 | if (queue) | |
5805 | return queue; | |
5806 | queue = kzalloc(sizeof(*queue), GFP_KERNEL); | |
5807 | if (!queue) | |
5808 | return NULL; | |
5809 | netdev_init_one_queue(dev, queue, NULL); | |
24824a09 ED |
5810 | queue->qdisc = &noop_qdisc; |
5811 | queue->qdisc_sleeping = &noop_qdisc; | |
5812 | rcu_assign_pointer(dev->ingress_queue, queue); | |
5813 | #endif | |
5814 | return queue; | |
bb949fbd DM |
5815 | } |
5816 | ||
1da177e4 | 5817 | /** |
36909ea4 | 5818 | * alloc_netdev_mqs - allocate network device |
1da177e4 LT |
5819 | * @sizeof_priv: size of private data to allocate space for |
5820 | * @name: device name format string | |
5821 | * @setup: callback to initialize device | |
36909ea4 TH |
5822 | * @txqs: the number of TX subqueues to allocate |
5823 | * @rxqs: the number of RX subqueues to allocate | |
1da177e4 LT |
5824 | * |
5825 | * Allocates a struct net_device with private data area for driver use | |
f25f4e44 | 5826 | * and performs basic initialization. Also allocates subquue structs |
36909ea4 | 5827 | * for each queue on the device. |
1da177e4 | 5828 | */ |
36909ea4 TH |
5829 | struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, |
5830 | void (*setup)(struct net_device *), | |
5831 | unsigned int txqs, unsigned int rxqs) | |
1da177e4 | 5832 | { |
1da177e4 | 5833 | struct net_device *dev; |
7943986c | 5834 | size_t alloc_size; |
1ce8e7b5 | 5835 | struct net_device *p; |
1da177e4 | 5836 | |
b6fe17d6 SH |
5837 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
5838 | ||
36909ea4 | 5839 | if (txqs < 1) { |
55513fb4 TH |
5840 | pr_err("alloc_netdev: Unable to allocate device " |
5841 | "with zero queues.\n"); | |
5842 | return NULL; | |
5843 | } | |
5844 | ||
36909ea4 TH |
5845 | #ifdef CONFIG_RPS |
5846 | if (rxqs < 1) { | |
5847 | pr_err("alloc_netdev: Unable to allocate device " | |
5848 | "with zero RX queues.\n"); | |
5849 | return NULL; | |
5850 | } | |
5851 | #endif | |
5852 | ||
fd2ea0a7 | 5853 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
5854 | if (sizeof_priv) { |
5855 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 5856 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
5857 | alloc_size += sizeof_priv; |
5858 | } | |
5859 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 5860 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 5861 | |
31380de9 | 5862 | p = kzalloc(alloc_size, GFP_KERNEL); |
1da177e4 | 5863 | if (!p) { |
b6fe17d6 | 5864 | printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n"); |
1da177e4 LT |
5865 | return NULL; |
5866 | } | |
1da177e4 | 5867 | |
1ce8e7b5 | 5868 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 5869 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc | 5870 | |
29b4433d ED |
5871 | dev->pcpu_refcnt = alloc_percpu(int); |
5872 | if (!dev->pcpu_refcnt) | |
e6484930 | 5873 | goto free_p; |
ab9c73cc | 5874 | |
ab9c73cc | 5875 | if (dev_addr_init(dev)) |
29b4433d | 5876 | goto free_pcpu; |
ab9c73cc | 5877 | |
22bedad3 | 5878 | dev_mc_init(dev); |
a748ee24 | 5879 | dev_uc_init(dev); |
ccffad25 | 5880 | |
c346dca1 | 5881 | dev_net_set(dev, &init_net); |
1da177e4 | 5882 | |
8d3bdbd5 DM |
5883 | dev->gso_max_size = GSO_MAX_SIZE; |
5884 | ||
8d3bdbd5 DM |
5885 | INIT_LIST_HEAD(&dev->napi_list); |
5886 | INIT_LIST_HEAD(&dev->unreg_list); | |
5887 | INIT_LIST_HEAD(&dev->link_watch_list); | |
5888 | dev->priv_flags = IFF_XMIT_DST_RELEASE; | |
5889 | setup(dev); | |
5890 | ||
36909ea4 TH |
5891 | dev->num_tx_queues = txqs; |
5892 | dev->real_num_tx_queues = txqs; | |
ed9af2e8 | 5893 | if (netif_alloc_netdev_queues(dev)) |
8d3bdbd5 | 5894 | goto free_all; |
e8a0464c | 5895 | |
df334545 | 5896 | #ifdef CONFIG_RPS |
36909ea4 TH |
5897 | dev->num_rx_queues = rxqs; |
5898 | dev->real_num_rx_queues = rxqs; | |
fe822240 | 5899 | if (netif_alloc_rx_queues(dev)) |
8d3bdbd5 | 5900 | goto free_all; |
df334545 | 5901 | #endif |
0a9627f2 | 5902 | |
1da177e4 | 5903 | strcpy(dev->name, name); |
cbda10fa | 5904 | dev->group = INIT_NETDEV_GROUP; |
1da177e4 | 5905 | return dev; |
ab9c73cc | 5906 | |
8d3bdbd5 DM |
5907 | free_all: |
5908 | free_netdev(dev); | |
5909 | return NULL; | |
5910 | ||
29b4433d ED |
5911 | free_pcpu: |
5912 | free_percpu(dev->pcpu_refcnt); | |
ed9af2e8 | 5913 | kfree(dev->_tx); |
fe822240 TH |
5914 | #ifdef CONFIG_RPS |
5915 | kfree(dev->_rx); | |
5916 | #endif | |
5917 | ||
ab9c73cc JP |
5918 | free_p: |
5919 | kfree(p); | |
5920 | return NULL; | |
1da177e4 | 5921 | } |
36909ea4 | 5922 | EXPORT_SYMBOL(alloc_netdev_mqs); |
1da177e4 LT |
5923 | |
5924 | /** | |
5925 | * free_netdev - free network device | |
5926 | * @dev: device | |
5927 | * | |
4ec93edb YH |
5928 | * This function does the last stage of destroying an allocated device |
5929 | * interface. The reference to the device object is released. | |
1da177e4 LT |
5930 | * If this is the last reference then it will be freed. |
5931 | */ | |
5932 | void free_netdev(struct net_device *dev) | |
5933 | { | |
d565b0a1 HX |
5934 | struct napi_struct *p, *n; |
5935 | ||
f3005d7f DL |
5936 | release_net(dev_net(dev)); |
5937 | ||
e8a0464c | 5938 | kfree(dev->_tx); |
fe822240 TH |
5939 | #ifdef CONFIG_RPS |
5940 | kfree(dev->_rx); | |
5941 | #endif | |
e8a0464c | 5942 | |
24824a09 ED |
5943 | kfree(rcu_dereference_raw(dev->ingress_queue)); |
5944 | ||
f001fde5 JP |
5945 | /* Flush device addresses */ |
5946 | dev_addr_flush(dev); | |
5947 | ||
d565b0a1 HX |
5948 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
5949 | netif_napi_del(p); | |
5950 | ||
29b4433d ED |
5951 | free_percpu(dev->pcpu_refcnt); |
5952 | dev->pcpu_refcnt = NULL; | |
5953 | ||
3041a069 | 5954 | /* Compatibility with error handling in drivers */ |
1da177e4 LT |
5955 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
5956 | kfree((char *)dev - dev->padded); | |
5957 | return; | |
5958 | } | |
5959 | ||
5960 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
5961 | dev->reg_state = NETREG_RELEASED; | |
5962 | ||
43cb76d9 GKH |
5963 | /* will free via device release */ |
5964 | put_device(&dev->dev); | |
1da177e4 | 5965 | } |
d1b19dff | 5966 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 5967 | |
f0db275a SH |
5968 | /** |
5969 | * synchronize_net - Synchronize with packet receive processing | |
5970 | * | |
5971 | * Wait for packets currently being received to be done. | |
5972 | * Does not block later packets from starting. | |
5973 | */ | |
4ec93edb | 5974 | void synchronize_net(void) |
1da177e4 LT |
5975 | { |
5976 | might_sleep(); | |
be3fc413 ED |
5977 | if (rtnl_is_locked()) |
5978 | synchronize_rcu_expedited(); | |
5979 | else | |
5980 | synchronize_rcu(); | |
1da177e4 | 5981 | } |
d1b19dff | 5982 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
5983 | |
5984 | /** | |
44a0873d | 5985 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 5986 | * @dev: device |
44a0873d | 5987 | * @head: list |
6ebfbc06 | 5988 | * |
1da177e4 | 5989 | * This function shuts down a device interface and removes it |
d59b54b1 | 5990 | * from the kernel tables. |
44a0873d | 5991 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
5992 | * |
5993 | * Callers must hold the rtnl semaphore. You may want | |
5994 | * unregister_netdev() instead of this. | |
5995 | */ | |
5996 | ||
44a0873d | 5997 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 5998 | { |
a6620712 HX |
5999 | ASSERT_RTNL(); |
6000 | ||
44a0873d | 6001 | if (head) { |
9fdce099 | 6002 | list_move_tail(&dev->unreg_list, head); |
44a0873d ED |
6003 | } else { |
6004 | rollback_registered(dev); | |
6005 | /* Finish processing unregister after unlock */ | |
6006 | net_set_todo(dev); | |
6007 | } | |
1da177e4 | 6008 | } |
44a0873d | 6009 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 6010 | |
9b5e383c ED |
6011 | /** |
6012 | * unregister_netdevice_many - unregister many devices | |
6013 | * @head: list of devices | |
9b5e383c ED |
6014 | */ |
6015 | void unregister_netdevice_many(struct list_head *head) | |
6016 | { | |
6017 | struct net_device *dev; | |
6018 | ||
6019 | if (!list_empty(head)) { | |
6020 | rollback_registered_many(head); | |
6021 | list_for_each_entry(dev, head, unreg_list) | |
6022 | net_set_todo(dev); | |
6023 | } | |
6024 | } | |
63c8099d | 6025 | EXPORT_SYMBOL(unregister_netdevice_many); |
9b5e383c | 6026 | |
1da177e4 LT |
6027 | /** |
6028 | * unregister_netdev - remove device from the kernel | |
6029 | * @dev: device | |
6030 | * | |
6031 | * This function shuts down a device interface and removes it | |
d59b54b1 | 6032 | * from the kernel tables. |
1da177e4 LT |
6033 | * |
6034 | * This is just a wrapper for unregister_netdevice that takes | |
6035 | * the rtnl semaphore. In general you want to use this and not | |
6036 | * unregister_netdevice. | |
6037 | */ | |
6038 | void unregister_netdev(struct net_device *dev) | |
6039 | { | |
6040 | rtnl_lock(); | |
6041 | unregister_netdevice(dev); | |
6042 | rtnl_unlock(); | |
6043 | } | |
1da177e4 LT |
6044 | EXPORT_SYMBOL(unregister_netdev); |
6045 | ||
ce286d32 EB |
6046 | /** |
6047 | * dev_change_net_namespace - move device to different nethost namespace | |
6048 | * @dev: device | |
6049 | * @net: network namespace | |
6050 | * @pat: If not NULL name pattern to try if the current device name | |
6051 | * is already taken in the destination network namespace. | |
6052 | * | |
6053 | * This function shuts down a device interface and moves it | |
6054 | * to a new network namespace. On success 0 is returned, on | |
6055 | * a failure a netagive errno code is returned. | |
6056 | * | |
6057 | * Callers must hold the rtnl semaphore. | |
6058 | */ | |
6059 | ||
6060 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
6061 | { | |
ce286d32 EB |
6062 | int err; |
6063 | ||
6064 | ASSERT_RTNL(); | |
6065 | ||
6066 | /* Don't allow namespace local devices to be moved. */ | |
6067 | err = -EINVAL; | |
6068 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
6069 | goto out; | |
6070 | ||
6071 | /* Ensure the device has been registrered */ | |
6072 | err = -EINVAL; | |
6073 | if (dev->reg_state != NETREG_REGISTERED) | |
6074 | goto out; | |
6075 | ||
6076 | /* Get out if there is nothing todo */ | |
6077 | err = 0; | |
878628fb | 6078 | if (net_eq(dev_net(dev), net)) |
ce286d32 EB |
6079 | goto out; |
6080 | ||
6081 | /* Pick the destination device name, and ensure | |
6082 | * we can use it in the destination network namespace. | |
6083 | */ | |
6084 | err = -EEXIST; | |
d9031024 | 6085 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
6086 | /* We get here if we can't use the current device name */ |
6087 | if (!pat) | |
6088 | goto out; | |
1c5cae81 | 6089 | if (dev_get_valid_name(dev, pat) < 0) |
ce286d32 EB |
6090 | goto out; |
6091 | } | |
6092 | ||
6093 | /* | |
6094 | * And now a mini version of register_netdevice unregister_netdevice. | |
6095 | */ | |
6096 | ||
6097 | /* If device is running close it first. */ | |
9b772652 | 6098 | dev_close(dev); |
ce286d32 EB |
6099 | |
6100 | /* And unlink it from device chain */ | |
6101 | err = -ENODEV; | |
6102 | unlist_netdevice(dev); | |
6103 | ||
6104 | synchronize_net(); | |
6105 | ||
6106 | /* Shutdown queueing discipline. */ | |
6107 | dev_shutdown(dev); | |
6108 | ||
6109 | /* Notify protocols, that we are about to destroy | |
6110 | this device. They should clean all the things. | |
3b27e105 DL |
6111 | |
6112 | Note that dev->reg_state stays at NETREG_REGISTERED. | |
6113 | This is wanted because this way 8021q and macvlan know | |
6114 | the device is just moving and can keep their slaves up. | |
ce286d32 EB |
6115 | */ |
6116 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
a5ee1551 | 6117 | call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev); |
d2237d35 | 6118 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U); |
ce286d32 EB |
6119 | |
6120 | /* | |
6121 | * Flush the unicast and multicast chains | |
6122 | */ | |
a748ee24 | 6123 | dev_uc_flush(dev); |
22bedad3 | 6124 | dev_mc_flush(dev); |
ce286d32 EB |
6125 | |
6126 | /* Actually switch the network namespace */ | |
c346dca1 | 6127 | dev_net_set(dev, net); |
ce286d32 | 6128 | |
ce286d32 EB |
6129 | /* If there is an ifindex conflict assign a new one */ |
6130 | if (__dev_get_by_index(net, dev->ifindex)) { | |
6131 | int iflink = (dev->iflink == dev->ifindex); | |
6132 | dev->ifindex = dev_new_index(net); | |
6133 | if (iflink) | |
6134 | dev->iflink = dev->ifindex; | |
6135 | } | |
6136 | ||
8b41d188 | 6137 | /* Fixup kobjects */ |
a1b3f594 | 6138 | err = device_rename(&dev->dev, dev->name); |
8b41d188 | 6139 | WARN_ON(err); |
ce286d32 EB |
6140 | |
6141 | /* Add the device back in the hashes */ | |
6142 | list_netdevice(dev); | |
6143 | ||
6144 | /* Notify protocols, that a new device appeared. */ | |
6145 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
6146 | ||
d90a909e EB |
6147 | /* |
6148 | * Prevent userspace races by waiting until the network | |
6149 | * device is fully setup before sending notifications. | |
6150 | */ | |
6151 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U); | |
6152 | ||
ce286d32 EB |
6153 | synchronize_net(); |
6154 | err = 0; | |
6155 | out: | |
6156 | return err; | |
6157 | } | |
463d0183 | 6158 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 6159 | |
1da177e4 LT |
6160 | static int dev_cpu_callback(struct notifier_block *nfb, |
6161 | unsigned long action, | |
6162 | void *ocpu) | |
6163 | { | |
6164 | struct sk_buff **list_skb; | |
1da177e4 LT |
6165 | struct sk_buff *skb; |
6166 | unsigned int cpu, oldcpu = (unsigned long)ocpu; | |
6167 | struct softnet_data *sd, *oldsd; | |
6168 | ||
8bb78442 | 6169 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
1da177e4 LT |
6170 | return NOTIFY_OK; |
6171 | ||
6172 | local_irq_disable(); | |
6173 | cpu = smp_processor_id(); | |
6174 | sd = &per_cpu(softnet_data, cpu); | |
6175 | oldsd = &per_cpu(softnet_data, oldcpu); | |
6176 | ||
6177 | /* Find end of our completion_queue. */ | |
6178 | list_skb = &sd->completion_queue; | |
6179 | while (*list_skb) | |
6180 | list_skb = &(*list_skb)->next; | |
6181 | /* Append completion queue from offline CPU. */ | |
6182 | *list_skb = oldsd->completion_queue; | |
6183 | oldsd->completion_queue = NULL; | |
6184 | ||
1da177e4 | 6185 | /* Append output queue from offline CPU. */ |
a9cbd588 CG |
6186 | if (oldsd->output_queue) { |
6187 | *sd->output_queue_tailp = oldsd->output_queue; | |
6188 | sd->output_queue_tailp = oldsd->output_queue_tailp; | |
6189 | oldsd->output_queue = NULL; | |
6190 | oldsd->output_queue_tailp = &oldsd->output_queue; | |
6191 | } | |
264524d5 HC |
6192 | /* Append NAPI poll list from offline CPU. */ |
6193 | if (!list_empty(&oldsd->poll_list)) { | |
6194 | list_splice_init(&oldsd->poll_list, &sd->poll_list); | |
6195 | raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
6196 | } | |
1da177e4 LT |
6197 | |
6198 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
6199 | local_irq_enable(); | |
6200 | ||
6201 | /* Process offline CPU's input_pkt_queue */ | |
76cc8b13 | 6202 | while ((skb = __skb_dequeue(&oldsd->process_queue))) { |
1da177e4 | 6203 | netif_rx(skb); |
76cc8b13 | 6204 | input_queue_head_incr(oldsd); |
fec5e652 | 6205 | } |
76cc8b13 | 6206 | while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) { |
6e7676c1 | 6207 | netif_rx(skb); |
76cc8b13 TH |
6208 | input_queue_head_incr(oldsd); |
6209 | } | |
1da177e4 LT |
6210 | |
6211 | return NOTIFY_OK; | |
6212 | } | |
1da177e4 LT |
6213 | |
6214 | ||
7f353bf2 | 6215 | /** |
b63365a2 HX |
6216 | * netdev_increment_features - increment feature set by one |
6217 | * @all: current feature set | |
6218 | * @one: new feature set | |
6219 | * @mask: mask feature set | |
7f353bf2 HX |
6220 | * |
6221 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
6222 | * @one to the master device with current feature set @all. Will not |
6223 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 6224 | */ |
04ed3e74 | 6225 | u32 netdev_increment_features(u32 all, u32 one, u32 mask) |
b63365a2 | 6226 | { |
1742f183 MM |
6227 | if (mask & NETIF_F_GEN_CSUM) |
6228 | mask |= NETIF_F_ALL_CSUM; | |
6229 | mask |= NETIF_F_VLAN_CHALLENGED; | |
7f353bf2 | 6230 | |
1742f183 MM |
6231 | all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask; |
6232 | all &= one | ~NETIF_F_ALL_FOR_ALL; | |
c6e1a0d1 | 6233 | |
1742f183 MM |
6234 | /* If device needs checksumming, downgrade to it. */ |
6235 | if (all & (NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM)) | |
6236 | all &= ~NETIF_F_NO_CSUM; | |
7f353bf2 | 6237 | |
1742f183 MM |
6238 | /* If one device supports hw checksumming, set for all. */ |
6239 | if (all & NETIF_F_GEN_CSUM) | |
6240 | all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM); | |
7f353bf2 HX |
6241 | |
6242 | return all; | |
6243 | } | |
b63365a2 | 6244 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 6245 | |
30d97d35 PE |
6246 | static struct hlist_head *netdev_create_hash(void) |
6247 | { | |
6248 | int i; | |
6249 | struct hlist_head *hash; | |
6250 | ||
6251 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
6252 | if (hash != NULL) | |
6253 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
6254 | INIT_HLIST_HEAD(&hash[i]); | |
6255 | ||
6256 | return hash; | |
6257 | } | |
6258 | ||
881d966b | 6259 | /* Initialize per network namespace state */ |
4665079c | 6260 | static int __net_init netdev_init(struct net *net) |
881d966b | 6261 | { |
881d966b | 6262 | INIT_LIST_HEAD(&net->dev_base_head); |
881d966b | 6263 | |
30d97d35 PE |
6264 | net->dev_name_head = netdev_create_hash(); |
6265 | if (net->dev_name_head == NULL) | |
6266 | goto err_name; | |
881d966b | 6267 | |
30d97d35 PE |
6268 | net->dev_index_head = netdev_create_hash(); |
6269 | if (net->dev_index_head == NULL) | |
6270 | goto err_idx; | |
881d966b EB |
6271 | |
6272 | return 0; | |
30d97d35 PE |
6273 | |
6274 | err_idx: | |
6275 | kfree(net->dev_name_head); | |
6276 | err_name: | |
6277 | return -ENOMEM; | |
881d966b EB |
6278 | } |
6279 | ||
f0db275a SH |
6280 | /** |
6281 | * netdev_drivername - network driver for the device | |
6282 | * @dev: network device | |
f0db275a SH |
6283 | * |
6284 | * Determine network driver for device. | |
6285 | */ | |
3019de12 | 6286 | const char *netdev_drivername(const struct net_device *dev) |
6579e57b | 6287 | { |
cf04a4c7 SH |
6288 | const struct device_driver *driver; |
6289 | const struct device *parent; | |
3019de12 | 6290 | const char *empty = ""; |
6579e57b AV |
6291 | |
6292 | parent = dev->dev.parent; | |
6579e57b | 6293 | if (!parent) |
3019de12 | 6294 | return empty; |
6579e57b AV |
6295 | |
6296 | driver = parent->driver; | |
6297 | if (driver && driver->name) | |
3019de12 DM |
6298 | return driver->name; |
6299 | return empty; | |
6579e57b AV |
6300 | } |
6301 | ||
256df2f3 JP |
6302 | static int __netdev_printk(const char *level, const struct net_device *dev, |
6303 | struct va_format *vaf) | |
6304 | { | |
6305 | int r; | |
6306 | ||
6307 | if (dev && dev->dev.parent) | |
6308 | r = dev_printk(level, dev->dev.parent, "%s: %pV", | |
6309 | netdev_name(dev), vaf); | |
6310 | else if (dev) | |
6311 | r = printk("%s%s: %pV", level, netdev_name(dev), vaf); | |
6312 | else | |
6313 | r = printk("%s(NULL net_device): %pV", level, vaf); | |
6314 | ||
6315 | return r; | |
6316 | } | |
6317 | ||
6318 | int netdev_printk(const char *level, const struct net_device *dev, | |
6319 | const char *format, ...) | |
6320 | { | |
6321 | struct va_format vaf; | |
6322 | va_list args; | |
6323 | int r; | |
6324 | ||
6325 | va_start(args, format); | |
6326 | ||
6327 | vaf.fmt = format; | |
6328 | vaf.va = &args; | |
6329 | ||
6330 | r = __netdev_printk(level, dev, &vaf); | |
6331 | va_end(args); | |
6332 | ||
6333 | return r; | |
6334 | } | |
6335 | EXPORT_SYMBOL(netdev_printk); | |
6336 | ||
6337 | #define define_netdev_printk_level(func, level) \ | |
6338 | int func(const struct net_device *dev, const char *fmt, ...) \ | |
6339 | { \ | |
6340 | int r; \ | |
6341 | struct va_format vaf; \ | |
6342 | va_list args; \ | |
6343 | \ | |
6344 | va_start(args, fmt); \ | |
6345 | \ | |
6346 | vaf.fmt = fmt; \ | |
6347 | vaf.va = &args; \ | |
6348 | \ | |
6349 | r = __netdev_printk(level, dev, &vaf); \ | |
6350 | va_end(args); \ | |
6351 | \ | |
6352 | return r; \ | |
6353 | } \ | |
6354 | EXPORT_SYMBOL(func); | |
6355 | ||
6356 | define_netdev_printk_level(netdev_emerg, KERN_EMERG); | |
6357 | define_netdev_printk_level(netdev_alert, KERN_ALERT); | |
6358 | define_netdev_printk_level(netdev_crit, KERN_CRIT); | |
6359 | define_netdev_printk_level(netdev_err, KERN_ERR); | |
6360 | define_netdev_printk_level(netdev_warn, KERN_WARNING); | |
6361 | define_netdev_printk_level(netdev_notice, KERN_NOTICE); | |
6362 | define_netdev_printk_level(netdev_info, KERN_INFO); | |
6363 | ||
4665079c | 6364 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
6365 | { |
6366 | kfree(net->dev_name_head); | |
6367 | kfree(net->dev_index_head); | |
6368 | } | |
6369 | ||
022cbae6 | 6370 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
6371 | .init = netdev_init, |
6372 | .exit = netdev_exit, | |
6373 | }; | |
6374 | ||
4665079c | 6375 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 6376 | { |
e008b5fc | 6377 | struct net_device *dev, *aux; |
ce286d32 | 6378 | /* |
e008b5fc | 6379 | * Push all migratable network devices back to the |
ce286d32 EB |
6380 | * initial network namespace |
6381 | */ | |
6382 | rtnl_lock(); | |
e008b5fc | 6383 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 6384 | int err; |
aca51397 | 6385 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
6386 | |
6387 | /* Ignore unmoveable devices (i.e. loopback) */ | |
6388 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
6389 | continue; | |
6390 | ||
e008b5fc EB |
6391 | /* Leave virtual devices for the generic cleanup */ |
6392 | if (dev->rtnl_link_ops) | |
6393 | continue; | |
d0c082ce | 6394 | |
25985edc | 6395 | /* Push remaining network devices to init_net */ |
aca51397 PE |
6396 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
6397 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
ce286d32 | 6398 | if (err) { |
aca51397 | 6399 | printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n", |
ce286d32 | 6400 | __func__, dev->name, err); |
aca51397 | 6401 | BUG(); |
ce286d32 EB |
6402 | } |
6403 | } | |
6404 | rtnl_unlock(); | |
6405 | } | |
6406 | ||
04dc7f6b EB |
6407 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
6408 | { | |
6409 | /* At exit all network devices most be removed from a network | |
b595076a | 6410 | * namespace. Do this in the reverse order of registration. |
04dc7f6b EB |
6411 | * Do this across as many network namespaces as possible to |
6412 | * improve batching efficiency. | |
6413 | */ | |
6414 | struct net_device *dev; | |
6415 | struct net *net; | |
6416 | LIST_HEAD(dev_kill_list); | |
6417 | ||
6418 | rtnl_lock(); | |
6419 | list_for_each_entry(net, net_list, exit_list) { | |
6420 | for_each_netdev_reverse(net, dev) { | |
6421 | if (dev->rtnl_link_ops) | |
6422 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); | |
6423 | else | |
6424 | unregister_netdevice_queue(dev, &dev_kill_list); | |
6425 | } | |
6426 | } | |
6427 | unregister_netdevice_many(&dev_kill_list); | |
ceaaec98 | 6428 | list_del(&dev_kill_list); |
04dc7f6b EB |
6429 | rtnl_unlock(); |
6430 | } | |
6431 | ||
022cbae6 | 6432 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 6433 | .exit = default_device_exit, |
04dc7f6b | 6434 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
6435 | }; |
6436 | ||
1da177e4 LT |
6437 | /* |
6438 | * Initialize the DEV module. At boot time this walks the device list and | |
6439 | * unhooks any devices that fail to initialise (normally hardware not | |
6440 | * present) and leaves us with a valid list of present and active devices. | |
6441 | * | |
6442 | */ | |
6443 | ||
6444 | /* | |
6445 | * This is called single threaded during boot, so no need | |
6446 | * to take the rtnl semaphore. | |
6447 | */ | |
6448 | static int __init net_dev_init(void) | |
6449 | { | |
6450 | int i, rc = -ENOMEM; | |
6451 | ||
6452 | BUG_ON(!dev_boot_phase); | |
6453 | ||
1da177e4 LT |
6454 | if (dev_proc_init()) |
6455 | goto out; | |
6456 | ||
8b41d188 | 6457 | if (netdev_kobject_init()) |
1da177e4 LT |
6458 | goto out; |
6459 | ||
6460 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 6461 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
6462 | INIT_LIST_HEAD(&ptype_base[i]); |
6463 | ||
881d966b EB |
6464 | if (register_pernet_subsys(&netdev_net_ops)) |
6465 | goto out; | |
1da177e4 LT |
6466 | |
6467 | /* | |
6468 | * Initialise the packet receive queues. | |
6469 | */ | |
6470 | ||
6f912042 | 6471 | for_each_possible_cpu(i) { |
e36fa2f7 | 6472 | struct softnet_data *sd = &per_cpu(softnet_data, i); |
1da177e4 | 6473 | |
dee42870 | 6474 | memset(sd, 0, sizeof(*sd)); |
e36fa2f7 | 6475 | skb_queue_head_init(&sd->input_pkt_queue); |
6e7676c1 | 6476 | skb_queue_head_init(&sd->process_queue); |
e36fa2f7 ED |
6477 | sd->completion_queue = NULL; |
6478 | INIT_LIST_HEAD(&sd->poll_list); | |
a9cbd588 CG |
6479 | sd->output_queue = NULL; |
6480 | sd->output_queue_tailp = &sd->output_queue; | |
df334545 | 6481 | #ifdef CONFIG_RPS |
e36fa2f7 ED |
6482 | sd->csd.func = rps_trigger_softirq; |
6483 | sd->csd.info = sd; | |
6484 | sd->csd.flags = 0; | |
6485 | sd->cpu = i; | |
1e94d72f | 6486 | #endif |
0a9627f2 | 6487 | |
e36fa2f7 ED |
6488 | sd->backlog.poll = process_backlog; |
6489 | sd->backlog.weight = weight_p; | |
6490 | sd->backlog.gro_list = NULL; | |
6491 | sd->backlog.gro_count = 0; | |
1da177e4 LT |
6492 | } |
6493 | ||
1da177e4 LT |
6494 | dev_boot_phase = 0; |
6495 | ||
505d4f73 EB |
6496 | /* The loopback device is special if any other network devices |
6497 | * is present in a network namespace the loopback device must | |
6498 | * be present. Since we now dynamically allocate and free the | |
6499 | * loopback device ensure this invariant is maintained by | |
6500 | * keeping the loopback device as the first device on the | |
6501 | * list of network devices. Ensuring the loopback devices | |
6502 | * is the first device that appears and the last network device | |
6503 | * that disappears. | |
6504 | */ | |
6505 | if (register_pernet_device(&loopback_net_ops)) | |
6506 | goto out; | |
6507 | ||
6508 | if (register_pernet_device(&default_device_ops)) | |
6509 | goto out; | |
6510 | ||
962cf36c CM |
6511 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
6512 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 LT |
6513 | |
6514 | hotcpu_notifier(dev_cpu_callback, 0); | |
6515 | dst_init(); | |
6516 | dev_mcast_init(); | |
6517 | rc = 0; | |
6518 | out: | |
6519 | return rc; | |
6520 | } | |
6521 | ||
6522 | subsys_initcall(net_dev_init); | |
6523 | ||
e88721f8 KK |
6524 | static int __init initialize_hashrnd(void) |
6525 | { | |
0a9627f2 | 6526 | get_random_bytes(&hashrnd, sizeof(hashrnd)); |
e88721f8 KK |
6527 | return 0; |
6528 | } | |
6529 | ||
6530 | late_initcall_sync(initialize_hashrnd); | |
6531 |