Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * NET3 Protocol independent device support routines. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public License | |
6 | * as published by the Free Software Foundation; either version | |
7 | * 2 of the License, or (at your option) any later version. | |
8 | * | |
9 | * Derived from the non IP parts of dev.c 1.0.19 | |
02c30a84 | 10 | * Authors: Ross Biro |
1da177e4 LT |
11 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
12 | * Mark Evans, <evansmp@uhura.aston.ac.uk> | |
13 | * | |
14 | * Additional Authors: | |
15 | * Florian la Roche <rzsfl@rz.uni-sb.de> | |
16 | * Alan Cox <gw4pts@gw4pts.ampr.org> | |
17 | * David Hinds <dahinds@users.sourceforge.net> | |
18 | * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> | |
19 | * Adam Sulmicki <adam@cfar.umd.edu> | |
20 | * Pekka Riikonen <priikone@poesidon.pspt.fi> | |
21 | * | |
22 | * Changes: | |
23 | * D.J. Barrow : Fixed bug where dev->refcnt gets set | |
24 | * to 2 if register_netdev gets called | |
25 | * before net_dev_init & also removed a | |
26 | * few lines of code in the process. | |
27 | * Alan Cox : device private ioctl copies fields back. | |
28 | * Alan Cox : Transmit queue code does relevant | |
29 | * stunts to keep the queue safe. | |
30 | * Alan Cox : Fixed double lock. | |
31 | * Alan Cox : Fixed promisc NULL pointer trap | |
32 | * ???????? : Support the full private ioctl range | |
33 | * Alan Cox : Moved ioctl permission check into | |
34 | * drivers | |
35 | * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI | |
36 | * Alan Cox : 100 backlog just doesn't cut it when | |
37 | * you start doing multicast video 8) | |
38 | * Alan Cox : Rewrote net_bh and list manager. | |
39 | * Alan Cox : Fix ETH_P_ALL echoback lengths. | |
40 | * Alan Cox : Took out transmit every packet pass | |
41 | * Saved a few bytes in the ioctl handler | |
42 | * Alan Cox : Network driver sets packet type before | |
43 | * calling netif_rx. Saves a function | |
44 | * call a packet. | |
45 | * Alan Cox : Hashed net_bh() | |
46 | * Richard Kooijman: Timestamp fixes. | |
47 | * Alan Cox : Wrong field in SIOCGIFDSTADDR | |
48 | * Alan Cox : Device lock protection. | |
49 | * Alan Cox : Fixed nasty side effect of device close | |
50 | * changes. | |
51 | * Rudi Cilibrasi : Pass the right thing to | |
52 | * set_mac_address() | |
53 | * Dave Miller : 32bit quantity for the device lock to | |
54 | * make it work out on a Sparc. | |
55 | * Bjorn Ekwall : Added KERNELD hack. | |
56 | * Alan Cox : Cleaned up the backlog initialise. | |
57 | * Craig Metz : SIOCGIFCONF fix if space for under | |
58 | * 1 device. | |
59 | * Thomas Bogendoerfer : Return ENODEV for dev_open, if there | |
60 | * is no device open function. | |
61 | * Andi Kleen : Fix error reporting for SIOCGIFCONF | |
62 | * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF | |
63 | * Cyrus Durgin : Cleaned for KMOD | |
64 | * Adam Sulmicki : Bug Fix : Network Device Unload | |
65 | * A network device unload needs to purge | |
66 | * the backlog queue. | |
67 | * Paul Rusty Russell : SIOCSIFNAME | |
68 | * Pekka Riikonen : Netdev boot-time settings code | |
69 | * Andrew Morton : Make unregister_netdevice wait | |
70 | * indefinitely on dev->refcnt | |
71 | * J Hadi Salim : - Backlog queue sampling | |
72 | * - netif_rx() feedback | |
73 | */ | |
74 | ||
75 | #include <asm/uaccess.h> | |
76 | #include <asm/system.h> | |
77 | #include <linux/bitops.h> | |
4fc268d2 | 78 | #include <linux/capability.h> |
1da177e4 LT |
79 | #include <linux/cpu.h> |
80 | #include <linux/types.h> | |
81 | #include <linux/kernel.h> | |
08e9897d | 82 | #include <linux/hash.h> |
5a0e3ad6 | 83 | #include <linux/slab.h> |
1da177e4 | 84 | #include <linux/sched.h> |
4a3e2f71 | 85 | #include <linux/mutex.h> |
1da177e4 LT |
86 | #include <linux/string.h> |
87 | #include <linux/mm.h> | |
88 | #include <linux/socket.h> | |
89 | #include <linux/sockios.h> | |
90 | #include <linux/errno.h> | |
91 | #include <linux/interrupt.h> | |
92 | #include <linux/if_ether.h> | |
93 | #include <linux/netdevice.h> | |
94 | #include <linux/etherdevice.h> | |
0187bdfb | 95 | #include <linux/ethtool.h> |
1da177e4 LT |
96 | #include <linux/notifier.h> |
97 | #include <linux/skbuff.h> | |
457c4cbc | 98 | #include <net/net_namespace.h> |
1da177e4 LT |
99 | #include <net/sock.h> |
100 | #include <linux/rtnetlink.h> | |
101 | #include <linux/proc_fs.h> | |
102 | #include <linux/seq_file.h> | |
103 | #include <linux/stat.h> | |
1da177e4 LT |
104 | #include <net/dst.h> |
105 | #include <net/pkt_sched.h> | |
106 | #include <net/checksum.h> | |
44540960 | 107 | #include <net/xfrm.h> |
1da177e4 LT |
108 | #include <linux/highmem.h> |
109 | #include <linux/init.h> | |
110 | #include <linux/kmod.h> | |
111 | #include <linux/module.h> | |
1da177e4 LT |
112 | #include <linux/netpoll.h> |
113 | #include <linux/rcupdate.h> | |
114 | #include <linux/delay.h> | |
295f4a1f | 115 | #include <net/wext.h> |
1da177e4 | 116 | #include <net/iw_handler.h> |
1da177e4 | 117 | #include <asm/current.h> |
5bdb9886 | 118 | #include <linux/audit.h> |
db217334 | 119 | #include <linux/dmaengine.h> |
f6a78bfc | 120 | #include <linux/err.h> |
c7fa9d18 | 121 | #include <linux/ctype.h> |
723e98b7 | 122 | #include <linux/if_arp.h> |
6de329e2 | 123 | #include <linux/if_vlan.h> |
8f0f2223 | 124 | #include <linux/ip.h> |
ad55dcaf | 125 | #include <net/ip.h> |
8f0f2223 DM |
126 | #include <linux/ipv6.h> |
127 | #include <linux/in.h> | |
b6b2fed1 DM |
128 | #include <linux/jhash.h> |
129 | #include <linux/random.h> | |
9cbc1cb8 | 130 | #include <trace/events/napi.h> |
cf66ba58 | 131 | #include <trace/events/net.h> |
07dc22e7 | 132 | #include <trace/events/skb.h> |
5acbbd42 | 133 | #include <linux/pci.h> |
caeda9b9 | 134 | #include <linux/inetdevice.h> |
1da177e4 | 135 | |
342709ef PE |
136 | #include "net-sysfs.h" |
137 | ||
d565b0a1 HX |
138 | /* Instead of increasing this, you should create a hash table. */ |
139 | #define MAX_GRO_SKBS 8 | |
140 | ||
5d38a079 HX |
141 | /* This should be increased if a protocol with a bigger head is added. */ |
142 | #define GRO_MAX_HEAD (MAX_HEADER + 128) | |
143 | ||
1da177e4 LT |
144 | /* |
145 | * The list of packet types we will receive (as opposed to discard) | |
146 | * and the routines to invoke. | |
147 | * | |
148 | * Why 16. Because with 16 the only overlap we get on a hash of the | |
149 | * low nibble of the protocol value is RARP/SNAP/X.25. | |
150 | * | |
151 | * NOTE: That is no longer true with the addition of VLAN tags. Not | |
152 | * sure which should go first, but I bet it won't make much | |
153 | * difference if we are running VLANs. The good news is that | |
154 | * this protocol won't be in the list unless compiled in, so | |
3041a069 | 155 | * the average user (w/out VLANs) will not be adversely affected. |
1da177e4 LT |
156 | * --BLG |
157 | * | |
158 | * 0800 IP | |
159 | * 8100 802.1Q VLAN | |
160 | * 0001 802.3 | |
161 | * 0002 AX.25 | |
162 | * 0004 802.2 | |
163 | * 8035 RARP | |
164 | * 0005 SNAP | |
165 | * 0805 X.25 | |
166 | * 0806 ARP | |
167 | * 8137 IPX | |
168 | * 0009 Localtalk | |
169 | * 86DD IPv6 | |
170 | */ | |
171 | ||
82d8a867 PE |
172 | #define PTYPE_HASH_SIZE (16) |
173 | #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1) | |
174 | ||
1da177e4 | 175 | static DEFINE_SPINLOCK(ptype_lock); |
82d8a867 | 176 | static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
6b2bedc3 | 177 | static struct list_head ptype_all __read_mostly; /* Taps */ |
1da177e4 | 178 | |
1da177e4 | 179 | /* |
7562f876 | 180 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
181 | * semaphore. |
182 | * | |
c6d14c84 | 183 | * Pure readers hold dev_base_lock for reading, or rcu_read_lock() |
1da177e4 LT |
184 | * |
185 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 186 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
187 | * actual updates. This allows pure readers to access the list even |
188 | * while a writer is preparing to update it. | |
189 | * | |
190 | * To put it another way, dev_base_lock is held for writing only to | |
191 | * protect against pure readers; the rtnl semaphore provides the | |
192 | * protection against other writers. | |
193 | * | |
194 | * See, for example usages, register_netdevice() and | |
195 | * unregister_netdevice(), which must be called with the rtnl | |
196 | * semaphore held. | |
197 | */ | |
1da177e4 | 198 | DEFINE_RWLOCK(dev_base_lock); |
1da177e4 LT |
199 | EXPORT_SYMBOL(dev_base_lock); |
200 | ||
881d966b | 201 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 LT |
202 | { |
203 | unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ)); | |
08e9897d | 204 | return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; |
1da177e4 LT |
205 | } |
206 | ||
881d966b | 207 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 208 | { |
7c28bd0b | 209 | return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; |
1da177e4 LT |
210 | } |
211 | ||
e36fa2f7 | 212 | static inline void rps_lock(struct softnet_data *sd) |
152102c7 CG |
213 | { |
214 | #ifdef CONFIG_RPS | |
e36fa2f7 | 215 | spin_lock(&sd->input_pkt_queue.lock); |
152102c7 CG |
216 | #endif |
217 | } | |
218 | ||
e36fa2f7 | 219 | static inline void rps_unlock(struct softnet_data *sd) |
152102c7 CG |
220 | { |
221 | #ifdef CONFIG_RPS | |
e36fa2f7 | 222 | spin_unlock(&sd->input_pkt_queue.lock); |
152102c7 CG |
223 | #endif |
224 | } | |
225 | ||
ce286d32 EB |
226 | /* Device list insertion */ |
227 | static int list_netdevice(struct net_device *dev) | |
228 | { | |
c346dca1 | 229 | struct net *net = dev_net(dev); |
ce286d32 EB |
230 | |
231 | ASSERT_RTNL(); | |
232 | ||
233 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 234 | list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); |
72c9528b | 235 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); |
fb699dfd ED |
236 | hlist_add_head_rcu(&dev->index_hlist, |
237 | dev_index_hash(net, dev->ifindex)); | |
ce286d32 EB |
238 | write_unlock_bh(&dev_base_lock); |
239 | return 0; | |
240 | } | |
241 | ||
fb699dfd ED |
242 | /* Device list removal |
243 | * caller must respect a RCU grace period before freeing/reusing dev | |
244 | */ | |
ce286d32 EB |
245 | static void unlist_netdevice(struct net_device *dev) |
246 | { | |
247 | ASSERT_RTNL(); | |
248 | ||
249 | /* Unlink dev from the device chain */ | |
250 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 251 | list_del_rcu(&dev->dev_list); |
72c9528b | 252 | hlist_del_rcu(&dev->name_hlist); |
fb699dfd | 253 | hlist_del_rcu(&dev->index_hlist); |
ce286d32 EB |
254 | write_unlock_bh(&dev_base_lock); |
255 | } | |
256 | ||
1da177e4 LT |
257 | /* |
258 | * Our notifier list | |
259 | */ | |
260 | ||
f07d5b94 | 261 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
262 | |
263 | /* | |
264 | * Device drivers call our routines to queue packets here. We empty the | |
265 | * queue in the local softnet handler. | |
266 | */ | |
bea3348e | 267 | |
9958da05 | 268 | DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); |
d1b19dff | 269 | EXPORT_PER_CPU_SYMBOL(softnet_data); |
1da177e4 | 270 | |
cf508b12 | 271 | #ifdef CONFIG_LOCKDEP |
723e98b7 | 272 | /* |
c773e847 | 273 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class |
723e98b7 JP |
274 | * according to dev->type |
275 | */ | |
276 | static const unsigned short netdev_lock_type[] = | |
277 | {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
278 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, | |
279 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
280 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
281 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
282 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
283 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
284 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
285 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
286 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
287 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
288 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
289 | ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211, | |
2d91d78b | 290 | ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, |
929122cd | 291 | ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154, |
fcb94e42 | 292 | ARPHRD_VOID, ARPHRD_NONE}; |
723e98b7 | 293 | |
36cbd3dc | 294 | static const char *const netdev_lock_name[] = |
723e98b7 JP |
295 | {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", |
296 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
297 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
298 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
299 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
300 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
301 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
302 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
303 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
304 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
305 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
306 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
307 | "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211", | |
2d91d78b | 308 | "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", |
929122cd | 309 | "_xmit_PHONET_PIPE", "_xmit_IEEE802154", |
fcb94e42 | 310 | "_xmit_VOID", "_xmit_NONE"}; |
723e98b7 JP |
311 | |
312 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
cf508b12 | 313 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; |
723e98b7 JP |
314 | |
315 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
316 | { | |
317 | int i; | |
318 | ||
319 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
320 | if (netdev_lock_type[i] == dev_type) | |
321 | return i; | |
322 | /* the last key is used by default */ | |
323 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
324 | } | |
325 | ||
cf508b12 DM |
326 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
327 | unsigned short dev_type) | |
723e98b7 JP |
328 | { |
329 | int i; | |
330 | ||
331 | i = netdev_lock_pos(dev_type); | |
332 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
333 | netdev_lock_name[i]); | |
334 | } | |
cf508b12 DM |
335 | |
336 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
337 | { | |
338 | int i; | |
339 | ||
340 | i = netdev_lock_pos(dev->type); | |
341 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
342 | &netdev_addr_lock_key[i], | |
343 | netdev_lock_name[i]); | |
344 | } | |
723e98b7 | 345 | #else |
cf508b12 DM |
346 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
347 | unsigned short dev_type) | |
348 | { | |
349 | } | |
350 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
723e98b7 JP |
351 | { |
352 | } | |
353 | #endif | |
1da177e4 LT |
354 | |
355 | /******************************************************************************* | |
356 | ||
357 | Protocol management and registration routines | |
358 | ||
359 | *******************************************************************************/ | |
360 | ||
1da177e4 LT |
361 | /* |
362 | * Add a protocol ID to the list. Now that the input handler is | |
363 | * smarter we can dispense with all the messy stuff that used to be | |
364 | * here. | |
365 | * | |
366 | * BEWARE!!! Protocol handlers, mangling input packets, | |
367 | * MUST BE last in hash buckets and checking protocol handlers | |
368 | * MUST start from promiscuous ptype_all chain in net_bh. | |
369 | * It is true now, do not change it. | |
370 | * Explanation follows: if protocol handler, mangling packet, will | |
371 | * be the first on list, it is not able to sense, that packet | |
372 | * is cloned and should be copied-on-write, so that it will | |
373 | * change it and subsequent readers will get broken packet. | |
374 | * --ANK (980803) | |
375 | */ | |
376 | ||
c07b68e8 ED |
377 | static inline struct list_head *ptype_head(const struct packet_type *pt) |
378 | { | |
379 | if (pt->type == htons(ETH_P_ALL)) | |
380 | return &ptype_all; | |
381 | else | |
382 | return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
383 | } | |
384 | ||
1da177e4 LT |
385 | /** |
386 | * dev_add_pack - add packet handler | |
387 | * @pt: packet type declaration | |
388 | * | |
389 | * Add a protocol handler to the networking stack. The passed &packet_type | |
390 | * is linked into kernel lists and may not be freed until it has been | |
391 | * removed from the kernel lists. | |
392 | * | |
4ec93edb | 393 | * This call does not sleep therefore it can not |
1da177e4 LT |
394 | * guarantee all CPU's that are in middle of receiving packets |
395 | * will see the new packet type (until the next received packet). | |
396 | */ | |
397 | ||
398 | void dev_add_pack(struct packet_type *pt) | |
399 | { | |
c07b68e8 | 400 | struct list_head *head = ptype_head(pt); |
1da177e4 | 401 | |
c07b68e8 ED |
402 | spin_lock(&ptype_lock); |
403 | list_add_rcu(&pt->list, head); | |
404 | spin_unlock(&ptype_lock); | |
1da177e4 | 405 | } |
d1b19dff | 406 | EXPORT_SYMBOL(dev_add_pack); |
1da177e4 | 407 | |
1da177e4 LT |
408 | /** |
409 | * __dev_remove_pack - remove packet handler | |
410 | * @pt: packet type declaration | |
411 | * | |
412 | * Remove a protocol handler that was previously added to the kernel | |
413 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
414 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 415 | * returns. |
1da177e4 LT |
416 | * |
417 | * The packet type might still be in use by receivers | |
418 | * and must not be freed until after all the CPU's have gone | |
419 | * through a quiescent state. | |
420 | */ | |
421 | void __dev_remove_pack(struct packet_type *pt) | |
422 | { | |
c07b68e8 | 423 | struct list_head *head = ptype_head(pt); |
1da177e4 LT |
424 | struct packet_type *pt1; |
425 | ||
c07b68e8 | 426 | spin_lock(&ptype_lock); |
1da177e4 LT |
427 | |
428 | list_for_each_entry(pt1, head, list) { | |
429 | if (pt == pt1) { | |
430 | list_del_rcu(&pt->list); | |
431 | goto out; | |
432 | } | |
433 | } | |
434 | ||
435 | printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt); | |
436 | out: | |
c07b68e8 | 437 | spin_unlock(&ptype_lock); |
1da177e4 | 438 | } |
d1b19dff ED |
439 | EXPORT_SYMBOL(__dev_remove_pack); |
440 | ||
1da177e4 LT |
441 | /** |
442 | * dev_remove_pack - remove packet handler | |
443 | * @pt: packet type declaration | |
444 | * | |
445 | * Remove a protocol handler that was previously added to the kernel | |
446 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
447 | * from the kernel lists and can be freed or reused once this function | |
448 | * returns. | |
449 | * | |
450 | * This call sleeps to guarantee that no CPU is looking at the packet | |
451 | * type after return. | |
452 | */ | |
453 | void dev_remove_pack(struct packet_type *pt) | |
454 | { | |
455 | __dev_remove_pack(pt); | |
4ec93edb | 456 | |
1da177e4 LT |
457 | synchronize_net(); |
458 | } | |
d1b19dff | 459 | EXPORT_SYMBOL(dev_remove_pack); |
1da177e4 LT |
460 | |
461 | /****************************************************************************** | |
462 | ||
463 | Device Boot-time Settings Routines | |
464 | ||
465 | *******************************************************************************/ | |
466 | ||
467 | /* Boot time configuration table */ | |
468 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
469 | ||
470 | /** | |
471 | * netdev_boot_setup_add - add new setup entry | |
472 | * @name: name of the device | |
473 | * @map: configured settings for the device | |
474 | * | |
475 | * Adds new setup entry to the dev_boot_setup list. The function | |
476 | * returns 0 on error and 1 on success. This is a generic routine to | |
477 | * all netdevices. | |
478 | */ | |
479 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
480 | { | |
481 | struct netdev_boot_setup *s; | |
482 | int i; | |
483 | ||
484 | s = dev_boot_setup; | |
485 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
486 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
487 | memset(s[i].name, 0, sizeof(s[i].name)); | |
93b3cff9 | 488 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
489 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
490 | break; | |
491 | } | |
492 | } | |
493 | ||
494 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
495 | } | |
496 | ||
497 | /** | |
498 | * netdev_boot_setup_check - check boot time settings | |
499 | * @dev: the netdevice | |
500 | * | |
501 | * Check boot time settings for the device. | |
502 | * The found settings are set for the device to be used | |
503 | * later in the device probing. | |
504 | * Returns 0 if no settings found, 1 if they are. | |
505 | */ | |
506 | int netdev_boot_setup_check(struct net_device *dev) | |
507 | { | |
508 | struct netdev_boot_setup *s = dev_boot_setup; | |
509 | int i; | |
510 | ||
511 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
512 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
93b3cff9 | 513 | !strcmp(dev->name, s[i].name)) { |
1da177e4 LT |
514 | dev->irq = s[i].map.irq; |
515 | dev->base_addr = s[i].map.base_addr; | |
516 | dev->mem_start = s[i].map.mem_start; | |
517 | dev->mem_end = s[i].map.mem_end; | |
518 | return 1; | |
519 | } | |
520 | } | |
521 | return 0; | |
522 | } | |
d1b19dff | 523 | EXPORT_SYMBOL(netdev_boot_setup_check); |
1da177e4 LT |
524 | |
525 | ||
526 | /** | |
527 | * netdev_boot_base - get address from boot time settings | |
528 | * @prefix: prefix for network device | |
529 | * @unit: id for network device | |
530 | * | |
531 | * Check boot time settings for the base address of device. | |
532 | * The found settings are set for the device to be used | |
533 | * later in the device probing. | |
534 | * Returns 0 if no settings found. | |
535 | */ | |
536 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
537 | { | |
538 | const struct netdev_boot_setup *s = dev_boot_setup; | |
539 | char name[IFNAMSIZ]; | |
540 | int i; | |
541 | ||
542 | sprintf(name, "%s%d", prefix, unit); | |
543 | ||
544 | /* | |
545 | * If device already registered then return base of 1 | |
546 | * to indicate not to probe for this interface | |
547 | */ | |
881d966b | 548 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
549 | return 1; |
550 | ||
551 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
552 | if (!strcmp(name, s[i].name)) | |
553 | return s[i].map.base_addr; | |
554 | return 0; | |
555 | } | |
556 | ||
557 | /* | |
558 | * Saves at boot time configured settings for any netdevice. | |
559 | */ | |
560 | int __init netdev_boot_setup(char *str) | |
561 | { | |
562 | int ints[5]; | |
563 | struct ifmap map; | |
564 | ||
565 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
566 | if (!str || !*str) | |
567 | return 0; | |
568 | ||
569 | /* Save settings */ | |
570 | memset(&map, 0, sizeof(map)); | |
571 | if (ints[0] > 0) | |
572 | map.irq = ints[1]; | |
573 | if (ints[0] > 1) | |
574 | map.base_addr = ints[2]; | |
575 | if (ints[0] > 2) | |
576 | map.mem_start = ints[3]; | |
577 | if (ints[0] > 3) | |
578 | map.mem_end = ints[4]; | |
579 | ||
580 | /* Add new entry to the list */ | |
581 | return netdev_boot_setup_add(str, &map); | |
582 | } | |
583 | ||
584 | __setup("netdev=", netdev_boot_setup); | |
585 | ||
586 | /******************************************************************************* | |
587 | ||
588 | Device Interface Subroutines | |
589 | ||
590 | *******************************************************************************/ | |
591 | ||
592 | /** | |
593 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 594 | * @net: the applicable net namespace |
1da177e4 LT |
595 | * @name: name to find |
596 | * | |
597 | * Find an interface by name. Must be called under RTNL semaphore | |
598 | * or @dev_base_lock. If the name is found a pointer to the device | |
599 | * is returned. If the name is not found then %NULL is returned. The | |
600 | * reference counters are not incremented so the caller must be | |
601 | * careful with locks. | |
602 | */ | |
603 | ||
881d966b | 604 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
605 | { |
606 | struct hlist_node *p; | |
0bd8d536 ED |
607 | struct net_device *dev; |
608 | struct hlist_head *head = dev_name_hash(net, name); | |
1da177e4 | 609 | |
0bd8d536 | 610 | hlist_for_each_entry(dev, p, head, name_hlist) |
1da177e4 LT |
611 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
612 | return dev; | |
0bd8d536 | 613 | |
1da177e4 LT |
614 | return NULL; |
615 | } | |
d1b19dff | 616 | EXPORT_SYMBOL(__dev_get_by_name); |
1da177e4 | 617 | |
72c9528b ED |
618 | /** |
619 | * dev_get_by_name_rcu - find a device by its name | |
620 | * @net: the applicable net namespace | |
621 | * @name: name to find | |
622 | * | |
623 | * Find an interface by name. | |
624 | * If the name is found a pointer to the device is returned. | |
625 | * If the name is not found then %NULL is returned. | |
626 | * The reference counters are not incremented so the caller must be | |
627 | * careful with locks. The caller must hold RCU lock. | |
628 | */ | |
629 | ||
630 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
631 | { | |
632 | struct hlist_node *p; | |
633 | struct net_device *dev; | |
634 | struct hlist_head *head = dev_name_hash(net, name); | |
635 | ||
636 | hlist_for_each_entry_rcu(dev, p, head, name_hlist) | |
637 | if (!strncmp(dev->name, name, IFNAMSIZ)) | |
638 | return dev; | |
639 | ||
640 | return NULL; | |
641 | } | |
642 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
643 | ||
1da177e4 LT |
644 | /** |
645 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 646 | * @net: the applicable net namespace |
1da177e4 LT |
647 | * @name: name to find |
648 | * | |
649 | * Find an interface by name. This can be called from any | |
650 | * context and does its own locking. The returned handle has | |
651 | * the usage count incremented and the caller must use dev_put() to | |
652 | * release it when it is no longer needed. %NULL is returned if no | |
653 | * matching device is found. | |
654 | */ | |
655 | ||
881d966b | 656 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
657 | { |
658 | struct net_device *dev; | |
659 | ||
72c9528b ED |
660 | rcu_read_lock(); |
661 | dev = dev_get_by_name_rcu(net, name); | |
1da177e4 LT |
662 | if (dev) |
663 | dev_hold(dev); | |
72c9528b | 664 | rcu_read_unlock(); |
1da177e4 LT |
665 | return dev; |
666 | } | |
d1b19dff | 667 | EXPORT_SYMBOL(dev_get_by_name); |
1da177e4 LT |
668 | |
669 | /** | |
670 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 671 | * @net: the applicable net namespace |
1da177e4 LT |
672 | * @ifindex: index of device |
673 | * | |
674 | * Search for an interface by index. Returns %NULL if the device | |
675 | * is not found or a pointer to the device. The device has not | |
676 | * had its reference counter increased so the caller must be careful | |
677 | * about locking. The caller must hold either the RTNL semaphore | |
678 | * or @dev_base_lock. | |
679 | */ | |
680 | ||
881d966b | 681 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
682 | { |
683 | struct hlist_node *p; | |
0bd8d536 ED |
684 | struct net_device *dev; |
685 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
1da177e4 | 686 | |
0bd8d536 | 687 | hlist_for_each_entry(dev, p, head, index_hlist) |
1da177e4 LT |
688 | if (dev->ifindex == ifindex) |
689 | return dev; | |
0bd8d536 | 690 | |
1da177e4 LT |
691 | return NULL; |
692 | } | |
d1b19dff | 693 | EXPORT_SYMBOL(__dev_get_by_index); |
1da177e4 | 694 | |
fb699dfd ED |
695 | /** |
696 | * dev_get_by_index_rcu - find a device by its ifindex | |
697 | * @net: the applicable net namespace | |
698 | * @ifindex: index of device | |
699 | * | |
700 | * Search for an interface by index. Returns %NULL if the device | |
701 | * is not found or a pointer to the device. The device has not | |
702 | * had its reference counter increased so the caller must be careful | |
703 | * about locking. The caller must hold RCU lock. | |
704 | */ | |
705 | ||
706 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
707 | { | |
708 | struct hlist_node *p; | |
709 | struct net_device *dev; | |
710 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
711 | ||
712 | hlist_for_each_entry_rcu(dev, p, head, index_hlist) | |
713 | if (dev->ifindex == ifindex) | |
714 | return dev; | |
715 | ||
716 | return NULL; | |
717 | } | |
718 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
719 | ||
1da177e4 LT |
720 | |
721 | /** | |
722 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 723 | * @net: the applicable net namespace |
1da177e4 LT |
724 | * @ifindex: index of device |
725 | * | |
726 | * Search for an interface by index. Returns NULL if the device | |
727 | * is not found or a pointer to the device. The device returned has | |
728 | * had a reference added and the pointer is safe until the user calls | |
729 | * dev_put to indicate they have finished with it. | |
730 | */ | |
731 | ||
881d966b | 732 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
733 | { |
734 | struct net_device *dev; | |
735 | ||
fb699dfd ED |
736 | rcu_read_lock(); |
737 | dev = dev_get_by_index_rcu(net, ifindex); | |
1da177e4 LT |
738 | if (dev) |
739 | dev_hold(dev); | |
fb699dfd | 740 | rcu_read_unlock(); |
1da177e4 LT |
741 | return dev; |
742 | } | |
d1b19dff | 743 | EXPORT_SYMBOL(dev_get_by_index); |
1da177e4 LT |
744 | |
745 | /** | |
941666c2 | 746 | * dev_getbyhwaddr_rcu - find a device by its hardware address |
c4ea43c5 | 747 | * @net: the applicable net namespace |
1da177e4 LT |
748 | * @type: media type of device |
749 | * @ha: hardware address | |
750 | * | |
751 | * Search for an interface by MAC address. Returns NULL if the device | |
c506653d ED |
752 | * is not found or a pointer to the device. |
753 | * The caller must hold RCU or RTNL. | |
941666c2 | 754 | * The returned device has not had its ref count increased |
1da177e4 LT |
755 | * and the caller must therefore be careful about locking |
756 | * | |
1da177e4 LT |
757 | */ |
758 | ||
941666c2 ED |
759 | struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, |
760 | const char *ha) | |
1da177e4 LT |
761 | { |
762 | struct net_device *dev; | |
763 | ||
941666c2 | 764 | for_each_netdev_rcu(net, dev) |
1da177e4 LT |
765 | if (dev->type == type && |
766 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
767 | return dev; |
768 | ||
769 | return NULL; | |
1da177e4 | 770 | } |
941666c2 | 771 | EXPORT_SYMBOL(dev_getbyhwaddr_rcu); |
cf309e3f | 772 | |
881d966b | 773 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
774 | { |
775 | struct net_device *dev; | |
776 | ||
4e9cac2b | 777 | ASSERT_RTNL(); |
881d966b | 778 | for_each_netdev(net, dev) |
4e9cac2b | 779 | if (dev->type == type) |
7562f876 PE |
780 | return dev; |
781 | ||
782 | return NULL; | |
4e9cac2b | 783 | } |
4e9cac2b PM |
784 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); |
785 | ||
881d966b | 786 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b | 787 | { |
99fe3c39 | 788 | struct net_device *dev, *ret = NULL; |
4e9cac2b | 789 | |
99fe3c39 ED |
790 | rcu_read_lock(); |
791 | for_each_netdev_rcu(net, dev) | |
792 | if (dev->type == type) { | |
793 | dev_hold(dev); | |
794 | ret = dev; | |
795 | break; | |
796 | } | |
797 | rcu_read_unlock(); | |
798 | return ret; | |
1da177e4 | 799 | } |
1da177e4 LT |
800 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
801 | ||
802 | /** | |
bb69ae04 | 803 | * dev_get_by_flags_rcu - find any device with given flags |
c4ea43c5 | 804 | * @net: the applicable net namespace |
1da177e4 LT |
805 | * @if_flags: IFF_* values |
806 | * @mask: bitmask of bits in if_flags to check | |
807 | * | |
808 | * Search for any interface with the given flags. Returns NULL if a device | |
bb69ae04 ED |
809 | * is not found or a pointer to the device. Must be called inside |
810 | * rcu_read_lock(), and result refcount is unchanged. | |
1da177e4 LT |
811 | */ |
812 | ||
bb69ae04 | 813 | struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags, |
d1b19dff | 814 | unsigned short mask) |
1da177e4 | 815 | { |
7562f876 | 816 | struct net_device *dev, *ret; |
1da177e4 | 817 | |
7562f876 | 818 | ret = NULL; |
c6d14c84 | 819 | for_each_netdev_rcu(net, dev) { |
1da177e4 | 820 | if (((dev->flags ^ if_flags) & mask) == 0) { |
7562f876 | 821 | ret = dev; |
1da177e4 LT |
822 | break; |
823 | } | |
824 | } | |
7562f876 | 825 | return ret; |
1da177e4 | 826 | } |
bb69ae04 | 827 | EXPORT_SYMBOL(dev_get_by_flags_rcu); |
1da177e4 LT |
828 | |
829 | /** | |
830 | * dev_valid_name - check if name is okay for network device | |
831 | * @name: name string | |
832 | * | |
833 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
834 | * to allow sysfs to work. We also disallow any kind of |
835 | * whitespace. | |
1da177e4 | 836 | */ |
c2373ee9 | 837 | int dev_valid_name(const char *name) |
1da177e4 | 838 | { |
c7fa9d18 DM |
839 | if (*name == '\0') |
840 | return 0; | |
b6fe17d6 SH |
841 | if (strlen(name) >= IFNAMSIZ) |
842 | return 0; | |
c7fa9d18 DM |
843 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
844 | return 0; | |
845 | ||
846 | while (*name) { | |
847 | if (*name == '/' || isspace(*name)) | |
848 | return 0; | |
849 | name++; | |
850 | } | |
851 | return 1; | |
1da177e4 | 852 | } |
d1b19dff | 853 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
854 | |
855 | /** | |
b267b179 EB |
856 | * __dev_alloc_name - allocate a name for a device |
857 | * @net: network namespace to allocate the device name in | |
1da177e4 | 858 | * @name: name format string |
b267b179 | 859 | * @buf: scratch buffer and result name string |
1da177e4 LT |
860 | * |
861 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
862 | * id. It scans list of devices to build up a free map, then chooses |
863 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
864 | * while allocating the name and adding the device in order to avoid | |
865 | * duplicates. | |
866 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
867 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
868 | */ |
869 | ||
b267b179 | 870 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
871 | { |
872 | int i = 0; | |
1da177e4 LT |
873 | const char *p; |
874 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 875 | unsigned long *inuse; |
1da177e4 LT |
876 | struct net_device *d; |
877 | ||
878 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
879 | if (p) { | |
880 | /* | |
881 | * Verify the string as this thing may have come from | |
882 | * the user. There must be either one "%d" and no other "%" | |
883 | * characters. | |
884 | */ | |
885 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
886 | return -EINVAL; | |
887 | ||
888 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 889 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
890 | if (!inuse) |
891 | return -ENOMEM; | |
892 | ||
881d966b | 893 | for_each_netdev(net, d) { |
1da177e4 LT |
894 | if (!sscanf(d->name, name, &i)) |
895 | continue; | |
896 | if (i < 0 || i >= max_netdevices) | |
897 | continue; | |
898 | ||
899 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 900 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
901 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
902 | set_bit(i, inuse); | |
903 | } | |
904 | ||
905 | i = find_first_zero_bit(inuse, max_netdevices); | |
906 | free_page((unsigned long) inuse); | |
907 | } | |
908 | ||
d9031024 OP |
909 | if (buf != name) |
910 | snprintf(buf, IFNAMSIZ, name, i); | |
b267b179 | 911 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 912 | return i; |
1da177e4 LT |
913 | |
914 | /* It is possible to run out of possible slots | |
915 | * when the name is long and there isn't enough space left | |
916 | * for the digits, or if all bits are used. | |
917 | */ | |
918 | return -ENFILE; | |
919 | } | |
920 | ||
b267b179 EB |
921 | /** |
922 | * dev_alloc_name - allocate a name for a device | |
923 | * @dev: device | |
924 | * @name: name format string | |
925 | * | |
926 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
927 | * id. It scans list of devices to build up a free map, then chooses | |
928 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
929 | * while allocating the name and adding the device in order to avoid | |
930 | * duplicates. | |
931 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
932 | * Returns the number of the unit assigned or a negative errno code. | |
933 | */ | |
934 | ||
935 | int dev_alloc_name(struct net_device *dev, const char *name) | |
936 | { | |
937 | char buf[IFNAMSIZ]; | |
938 | struct net *net; | |
939 | int ret; | |
940 | ||
c346dca1 YH |
941 | BUG_ON(!dev_net(dev)); |
942 | net = dev_net(dev); | |
b267b179 EB |
943 | ret = __dev_alloc_name(net, name, buf); |
944 | if (ret >= 0) | |
945 | strlcpy(dev->name, buf, IFNAMSIZ); | |
946 | return ret; | |
947 | } | |
d1b19dff | 948 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 949 | |
8ce6cebc | 950 | static int dev_get_valid_name(struct net_device *dev, const char *name, bool fmt) |
d9031024 | 951 | { |
8ce6cebc DL |
952 | struct net *net; |
953 | ||
954 | BUG_ON(!dev_net(dev)); | |
955 | net = dev_net(dev); | |
956 | ||
d9031024 OP |
957 | if (!dev_valid_name(name)) |
958 | return -EINVAL; | |
959 | ||
960 | if (fmt && strchr(name, '%')) | |
8ce6cebc | 961 | return dev_alloc_name(dev, name); |
d9031024 OP |
962 | else if (__dev_get_by_name(net, name)) |
963 | return -EEXIST; | |
8ce6cebc DL |
964 | else if (dev->name != name) |
965 | strlcpy(dev->name, name, IFNAMSIZ); | |
d9031024 OP |
966 | |
967 | return 0; | |
968 | } | |
1da177e4 LT |
969 | |
970 | /** | |
971 | * dev_change_name - change name of a device | |
972 | * @dev: device | |
973 | * @newname: name (or format string) must be at least IFNAMSIZ | |
974 | * | |
975 | * Change name of a device, can pass format strings "eth%d". | |
976 | * for wildcarding. | |
977 | */ | |
cf04a4c7 | 978 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 979 | { |
fcc5a03a | 980 | char oldname[IFNAMSIZ]; |
1da177e4 | 981 | int err = 0; |
fcc5a03a | 982 | int ret; |
881d966b | 983 | struct net *net; |
1da177e4 LT |
984 | |
985 | ASSERT_RTNL(); | |
c346dca1 | 986 | BUG_ON(!dev_net(dev)); |
1da177e4 | 987 | |
c346dca1 | 988 | net = dev_net(dev); |
1da177e4 LT |
989 | if (dev->flags & IFF_UP) |
990 | return -EBUSY; | |
991 | ||
c8d90dca SH |
992 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) |
993 | return 0; | |
994 | ||
fcc5a03a HX |
995 | memcpy(oldname, dev->name, IFNAMSIZ); |
996 | ||
8ce6cebc | 997 | err = dev_get_valid_name(dev, newname, 1); |
d9031024 OP |
998 | if (err < 0) |
999 | return err; | |
1da177e4 | 1000 | |
fcc5a03a | 1001 | rollback: |
a1b3f594 EB |
1002 | ret = device_rename(&dev->dev, dev->name); |
1003 | if (ret) { | |
1004 | memcpy(dev->name, oldname, IFNAMSIZ); | |
1005 | return ret; | |
dcc99773 | 1006 | } |
7f988eab HX |
1007 | |
1008 | write_lock_bh(&dev_base_lock); | |
92749821 | 1009 | hlist_del(&dev->name_hlist); |
72c9528b ED |
1010 | write_unlock_bh(&dev_base_lock); |
1011 | ||
1012 | synchronize_rcu(); | |
1013 | ||
1014 | write_lock_bh(&dev_base_lock); | |
1015 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
7f988eab HX |
1016 | write_unlock_bh(&dev_base_lock); |
1017 | ||
056925ab | 1018 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1019 | ret = notifier_to_errno(ret); |
1020 | ||
1021 | if (ret) { | |
91e9c07b ED |
1022 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1023 | if (err >= 0) { | |
fcc5a03a HX |
1024 | err = ret; |
1025 | memcpy(dev->name, oldname, IFNAMSIZ); | |
1026 | goto rollback; | |
91e9c07b ED |
1027 | } else { |
1028 | printk(KERN_ERR | |
1029 | "%s: name change rollback failed: %d.\n", | |
1030 | dev->name, ret); | |
fcc5a03a HX |
1031 | } |
1032 | } | |
1da177e4 LT |
1033 | |
1034 | return err; | |
1035 | } | |
1036 | ||
0b815a1a SH |
1037 | /** |
1038 | * dev_set_alias - change ifalias of a device | |
1039 | * @dev: device | |
1040 | * @alias: name up to IFALIASZ | |
f0db275a | 1041 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1042 | * |
1043 | * Set ifalias for a device, | |
1044 | */ | |
1045 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1046 | { | |
1047 | ASSERT_RTNL(); | |
1048 | ||
1049 | if (len >= IFALIASZ) | |
1050 | return -EINVAL; | |
1051 | ||
96ca4a2c OH |
1052 | if (!len) { |
1053 | if (dev->ifalias) { | |
1054 | kfree(dev->ifalias); | |
1055 | dev->ifalias = NULL; | |
1056 | } | |
1057 | return 0; | |
1058 | } | |
1059 | ||
d1b19dff | 1060 | dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL); |
0b815a1a SH |
1061 | if (!dev->ifalias) |
1062 | return -ENOMEM; | |
1063 | ||
1064 | strlcpy(dev->ifalias, alias, len+1); | |
1065 | return len; | |
1066 | } | |
1067 | ||
1068 | ||
d8a33ac4 | 1069 | /** |
3041a069 | 1070 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1071 | * @dev: device to cause notification |
1072 | * | |
1073 | * Called to indicate a device has changed features. | |
1074 | */ | |
1075 | void netdev_features_change(struct net_device *dev) | |
1076 | { | |
056925ab | 1077 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1078 | } |
1079 | EXPORT_SYMBOL(netdev_features_change); | |
1080 | ||
1da177e4 LT |
1081 | /** |
1082 | * netdev_state_change - device changes state | |
1083 | * @dev: device to cause notification | |
1084 | * | |
1085 | * Called to indicate a device has changed state. This function calls | |
1086 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1087 | * to the routing socket. | |
1088 | */ | |
1089 | void netdev_state_change(struct net_device *dev) | |
1090 | { | |
1091 | if (dev->flags & IFF_UP) { | |
056925ab | 1092 | call_netdevice_notifiers(NETDEV_CHANGE, dev); |
1da177e4 LT |
1093 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0); |
1094 | } | |
1095 | } | |
d1b19dff | 1096 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1097 | |
3ca5b404 | 1098 | int netdev_bonding_change(struct net_device *dev, unsigned long event) |
c1da4ac7 | 1099 | { |
3ca5b404 | 1100 | return call_netdevice_notifiers(event, dev); |
c1da4ac7 OG |
1101 | } |
1102 | EXPORT_SYMBOL(netdev_bonding_change); | |
1103 | ||
1da177e4 LT |
1104 | /** |
1105 | * dev_load - load a network module | |
c4ea43c5 | 1106 | * @net: the applicable net namespace |
1da177e4 LT |
1107 | * @name: name of interface |
1108 | * | |
1109 | * If a network interface is not present and the process has suitable | |
1110 | * privileges this function loads the module. If module loading is not | |
1111 | * available in this kernel then it becomes a nop. | |
1112 | */ | |
1113 | ||
881d966b | 1114 | void dev_load(struct net *net, const char *name) |
1da177e4 | 1115 | { |
4ec93edb | 1116 | struct net_device *dev; |
1da177e4 | 1117 | |
72c9528b ED |
1118 | rcu_read_lock(); |
1119 | dev = dev_get_by_name_rcu(net, name); | |
1120 | rcu_read_unlock(); | |
1da177e4 | 1121 | |
a8f80e8f | 1122 | if (!dev && capable(CAP_NET_ADMIN)) |
1da177e4 LT |
1123 | request_module("%s", name); |
1124 | } | |
d1b19dff | 1125 | EXPORT_SYMBOL(dev_load); |
1da177e4 | 1126 | |
bd380811 | 1127 | static int __dev_open(struct net_device *dev) |
1da177e4 | 1128 | { |
d314774c | 1129 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1130 | int ret; |
1da177e4 | 1131 | |
e46b66bc BH |
1132 | ASSERT_RTNL(); |
1133 | ||
1da177e4 LT |
1134 | /* |
1135 | * Is it even present? | |
1136 | */ | |
1137 | if (!netif_device_present(dev)) | |
1138 | return -ENODEV; | |
1139 | ||
3b8bcfd5 JB |
1140 | ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); |
1141 | ret = notifier_to_errno(ret); | |
1142 | if (ret) | |
1143 | return ret; | |
1144 | ||
1da177e4 LT |
1145 | /* |
1146 | * Call device private open method | |
1147 | */ | |
1148 | set_bit(__LINK_STATE_START, &dev->state); | |
bada339b | 1149 | |
d314774c SH |
1150 | if (ops->ndo_validate_addr) |
1151 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1152 | |
d314774c SH |
1153 | if (!ret && ops->ndo_open) |
1154 | ret = ops->ndo_open(dev); | |
1da177e4 | 1155 | |
4ec93edb | 1156 | /* |
1da177e4 LT |
1157 | * If it went open OK then: |
1158 | */ | |
1159 | ||
bada339b JG |
1160 | if (ret) |
1161 | clear_bit(__LINK_STATE_START, &dev->state); | |
1162 | else { | |
1da177e4 LT |
1163 | /* |
1164 | * Set the flags. | |
1165 | */ | |
1166 | dev->flags |= IFF_UP; | |
1167 | ||
649274d9 DW |
1168 | /* |
1169 | * Enable NET_DMA | |
1170 | */ | |
b4bd07c2 | 1171 | net_dmaengine_get(); |
649274d9 | 1172 | |
1da177e4 LT |
1173 | /* |
1174 | * Initialize multicasting status | |
1175 | */ | |
4417da66 | 1176 | dev_set_rx_mode(dev); |
1da177e4 LT |
1177 | |
1178 | /* | |
1179 | * Wakeup transmit queue engine | |
1180 | */ | |
1181 | dev_activate(dev); | |
1da177e4 | 1182 | } |
bada339b | 1183 | |
1da177e4 LT |
1184 | return ret; |
1185 | } | |
1186 | ||
1187 | /** | |
bd380811 PM |
1188 | * dev_open - prepare an interface for use. |
1189 | * @dev: device to open | |
1da177e4 | 1190 | * |
bd380811 PM |
1191 | * Takes a device from down to up state. The device's private open |
1192 | * function is invoked and then the multicast lists are loaded. Finally | |
1193 | * the device is moved into the up state and a %NETDEV_UP message is | |
1194 | * sent to the netdev notifier chain. | |
1195 | * | |
1196 | * Calling this function on an active interface is a nop. On a failure | |
1197 | * a negative errno code is returned. | |
1da177e4 | 1198 | */ |
bd380811 PM |
1199 | int dev_open(struct net_device *dev) |
1200 | { | |
1201 | int ret; | |
1202 | ||
1203 | /* | |
1204 | * Is it already up? | |
1205 | */ | |
1206 | if (dev->flags & IFF_UP) | |
1207 | return 0; | |
1208 | ||
1209 | /* | |
1210 | * Open device | |
1211 | */ | |
1212 | ret = __dev_open(dev); | |
1213 | if (ret < 0) | |
1214 | return ret; | |
1215 | ||
1216 | /* | |
1217 | * ... and announce new interface. | |
1218 | */ | |
1219 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING); | |
1220 | call_netdevice_notifiers(NETDEV_UP, dev); | |
1221 | ||
1222 | return ret; | |
1223 | } | |
1224 | EXPORT_SYMBOL(dev_open); | |
1225 | ||
44345724 | 1226 | static int __dev_close_many(struct list_head *head) |
1da177e4 | 1227 | { |
44345724 | 1228 | struct net_device *dev; |
e46b66bc | 1229 | |
bd380811 | 1230 | ASSERT_RTNL(); |
9d5010db DM |
1231 | might_sleep(); |
1232 | ||
44345724 OP |
1233 | list_for_each_entry(dev, head, unreg_list) { |
1234 | /* | |
1235 | * Tell people we are going down, so that they can | |
1236 | * prepare to death, when device is still operating. | |
1237 | */ | |
1238 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); | |
1da177e4 | 1239 | |
44345724 | 1240 | clear_bit(__LINK_STATE_START, &dev->state); |
1da177e4 | 1241 | |
44345724 OP |
1242 | /* Synchronize to scheduled poll. We cannot touch poll list, it |
1243 | * can be even on different cpu. So just clear netif_running(). | |
1244 | * | |
1245 | * dev->stop() will invoke napi_disable() on all of it's | |
1246 | * napi_struct instances on this device. | |
1247 | */ | |
1248 | smp_mb__after_clear_bit(); /* Commit netif_running(). */ | |
1249 | } | |
1da177e4 | 1250 | |
44345724 | 1251 | dev_deactivate_many(head); |
d8b2a4d2 | 1252 | |
44345724 OP |
1253 | list_for_each_entry(dev, head, unreg_list) { |
1254 | const struct net_device_ops *ops = dev->netdev_ops; | |
1da177e4 | 1255 | |
44345724 OP |
1256 | /* |
1257 | * Call the device specific close. This cannot fail. | |
1258 | * Only if device is UP | |
1259 | * | |
1260 | * We allow it to be called even after a DETACH hot-plug | |
1261 | * event. | |
1262 | */ | |
1263 | if (ops->ndo_stop) | |
1264 | ops->ndo_stop(dev); | |
1265 | ||
1266 | /* | |
1267 | * Device is now down. | |
1268 | */ | |
1269 | ||
1270 | dev->flags &= ~IFF_UP; | |
1271 | ||
1272 | /* | |
1273 | * Shutdown NET_DMA | |
1274 | */ | |
1275 | net_dmaengine_put(); | |
1276 | } | |
1277 | ||
1278 | return 0; | |
1279 | } | |
1280 | ||
1281 | static int __dev_close(struct net_device *dev) | |
1282 | { | |
f87e6f47 | 1283 | int retval; |
44345724 OP |
1284 | LIST_HEAD(single); |
1285 | ||
1286 | list_add(&dev->unreg_list, &single); | |
f87e6f47 LT |
1287 | retval = __dev_close_many(&single); |
1288 | list_del(&single); | |
1289 | return retval; | |
44345724 OP |
1290 | } |
1291 | ||
1292 | int dev_close_many(struct list_head *head) | |
1293 | { | |
1294 | struct net_device *dev, *tmp; | |
1295 | LIST_HEAD(tmp_list); | |
1da177e4 | 1296 | |
44345724 OP |
1297 | list_for_each_entry_safe(dev, tmp, head, unreg_list) |
1298 | if (!(dev->flags & IFF_UP)) | |
1299 | list_move(&dev->unreg_list, &tmp_list); | |
1300 | ||
1301 | __dev_close_many(head); | |
1da177e4 LT |
1302 | |
1303 | /* | |
44345724 | 1304 | * Tell people we are down |
1da177e4 | 1305 | */ |
44345724 OP |
1306 | list_for_each_entry(dev, head, unreg_list) { |
1307 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING); | |
1308 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
1309 | } | |
bd380811 | 1310 | |
44345724 OP |
1311 | /* rollback_registered_many needs the complete original list */ |
1312 | list_splice(&tmp_list, head); | |
bd380811 PM |
1313 | return 0; |
1314 | } | |
1315 | ||
1316 | /** | |
1317 | * dev_close - shutdown an interface. | |
1318 | * @dev: device to shutdown | |
1319 | * | |
1320 | * This function moves an active device into down state. A | |
1321 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1322 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1323 | * chain. | |
1324 | */ | |
1325 | int dev_close(struct net_device *dev) | |
1326 | { | |
44345724 | 1327 | LIST_HEAD(single); |
1da177e4 | 1328 | |
44345724 OP |
1329 | list_add(&dev->unreg_list, &single); |
1330 | dev_close_many(&single); | |
f87e6f47 | 1331 | list_del(&single); |
1da177e4 LT |
1332 | return 0; |
1333 | } | |
d1b19dff | 1334 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1335 | |
1336 | ||
0187bdfb BH |
1337 | /** |
1338 | * dev_disable_lro - disable Large Receive Offload on a device | |
1339 | * @dev: device | |
1340 | * | |
1341 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1342 | * called under RTNL. This is needed if received packets may be | |
1343 | * forwarded to another interface. | |
1344 | */ | |
1345 | void dev_disable_lro(struct net_device *dev) | |
1346 | { | |
1347 | if (dev->ethtool_ops && dev->ethtool_ops->get_flags && | |
1348 | dev->ethtool_ops->set_flags) { | |
1349 | u32 flags = dev->ethtool_ops->get_flags(dev); | |
1350 | if (flags & ETH_FLAG_LRO) { | |
1351 | flags &= ~ETH_FLAG_LRO; | |
1352 | dev->ethtool_ops->set_flags(dev, flags); | |
1353 | } | |
1354 | } | |
1355 | WARN_ON(dev->features & NETIF_F_LRO); | |
1356 | } | |
1357 | EXPORT_SYMBOL(dev_disable_lro); | |
1358 | ||
1359 | ||
881d966b EB |
1360 | static int dev_boot_phase = 1; |
1361 | ||
1da177e4 LT |
1362 | /* |
1363 | * Device change register/unregister. These are not inline or static | |
1364 | * as we export them to the world. | |
1365 | */ | |
1366 | ||
1367 | /** | |
1368 | * register_netdevice_notifier - register a network notifier block | |
1369 | * @nb: notifier | |
1370 | * | |
1371 | * Register a notifier to be called when network device events occur. | |
1372 | * The notifier passed is linked into the kernel structures and must | |
1373 | * not be reused until it has been unregistered. A negative errno code | |
1374 | * is returned on a failure. | |
1375 | * | |
1376 | * When registered all registration and up events are replayed | |
4ec93edb | 1377 | * to the new notifier to allow device to have a race free |
1da177e4 LT |
1378 | * view of the network device list. |
1379 | */ | |
1380 | ||
1381 | int register_netdevice_notifier(struct notifier_block *nb) | |
1382 | { | |
1383 | struct net_device *dev; | |
fcc5a03a | 1384 | struct net_device *last; |
881d966b | 1385 | struct net *net; |
1da177e4 LT |
1386 | int err; |
1387 | ||
1388 | rtnl_lock(); | |
f07d5b94 | 1389 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1390 | if (err) |
1391 | goto unlock; | |
881d966b EB |
1392 | if (dev_boot_phase) |
1393 | goto unlock; | |
1394 | for_each_net(net) { | |
1395 | for_each_netdev(net, dev) { | |
1396 | err = nb->notifier_call(nb, NETDEV_REGISTER, dev); | |
1397 | err = notifier_to_errno(err); | |
1398 | if (err) | |
1399 | goto rollback; | |
1400 | ||
1401 | if (!(dev->flags & IFF_UP)) | |
1402 | continue; | |
1da177e4 | 1403 | |
881d966b EB |
1404 | nb->notifier_call(nb, NETDEV_UP, dev); |
1405 | } | |
1da177e4 | 1406 | } |
fcc5a03a HX |
1407 | |
1408 | unlock: | |
1da177e4 LT |
1409 | rtnl_unlock(); |
1410 | return err; | |
fcc5a03a HX |
1411 | |
1412 | rollback: | |
1413 | last = dev; | |
881d966b EB |
1414 | for_each_net(net) { |
1415 | for_each_netdev(net, dev) { | |
1416 | if (dev == last) | |
1417 | break; | |
fcc5a03a | 1418 | |
881d966b EB |
1419 | if (dev->flags & IFF_UP) { |
1420 | nb->notifier_call(nb, NETDEV_GOING_DOWN, dev); | |
1421 | nb->notifier_call(nb, NETDEV_DOWN, dev); | |
1422 | } | |
1423 | nb->notifier_call(nb, NETDEV_UNREGISTER, dev); | |
a5ee1551 | 1424 | nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev); |
fcc5a03a | 1425 | } |
fcc5a03a | 1426 | } |
c67625a1 PE |
1427 | |
1428 | raw_notifier_chain_unregister(&netdev_chain, nb); | |
fcc5a03a | 1429 | goto unlock; |
1da177e4 | 1430 | } |
d1b19dff | 1431 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1432 | |
1433 | /** | |
1434 | * unregister_netdevice_notifier - unregister a network notifier block | |
1435 | * @nb: notifier | |
1436 | * | |
1437 | * Unregister a notifier previously registered by | |
1438 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1439 | * kernel structures and may then be reused. A negative errno code | |
1440 | * is returned on a failure. | |
1441 | */ | |
1442 | ||
1443 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1444 | { | |
9f514950 HX |
1445 | int err; |
1446 | ||
1447 | rtnl_lock(); | |
f07d5b94 | 1448 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
9f514950 HX |
1449 | rtnl_unlock(); |
1450 | return err; | |
1da177e4 | 1451 | } |
d1b19dff | 1452 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 LT |
1453 | |
1454 | /** | |
1455 | * call_netdevice_notifiers - call all network notifier blocks | |
1456 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1457 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1458 | * |
1459 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1460 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1461 | */ |
1462 | ||
ad7379d4 | 1463 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1464 | { |
ab930471 | 1465 | ASSERT_RTNL(); |
ad7379d4 | 1466 | return raw_notifier_call_chain(&netdev_chain, val, dev); |
1da177e4 LT |
1467 | } |
1468 | ||
1469 | /* When > 0 there are consumers of rx skb time stamps */ | |
1470 | static atomic_t netstamp_needed = ATOMIC_INIT(0); | |
1471 | ||
1472 | void net_enable_timestamp(void) | |
1473 | { | |
1474 | atomic_inc(&netstamp_needed); | |
1475 | } | |
d1b19dff | 1476 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
1477 | |
1478 | void net_disable_timestamp(void) | |
1479 | { | |
1480 | atomic_dec(&netstamp_needed); | |
1481 | } | |
d1b19dff | 1482 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 1483 | |
3b098e2d | 1484 | static inline void net_timestamp_set(struct sk_buff *skb) |
1da177e4 LT |
1485 | { |
1486 | if (atomic_read(&netstamp_needed)) | |
a61bbcf2 | 1487 | __net_timestamp(skb); |
b7aa0bf7 ED |
1488 | else |
1489 | skb->tstamp.tv64 = 0; | |
1da177e4 LT |
1490 | } |
1491 | ||
3b098e2d ED |
1492 | static inline void net_timestamp_check(struct sk_buff *skb) |
1493 | { | |
1494 | if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed)) | |
1495 | __net_timestamp(skb); | |
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 | { | |
1518 | skb_orphan(skb); | |
c736eefa | 1519 | nf_reset(skb); |
44540960 | 1520 | |
caf586e5 | 1521 | if (unlikely(!(dev->flags & IFF_UP) || |
2198a10b | 1522 | (skb->len > (dev->mtu + dev->hard_header_len + VLAN_HLEN)))) { |
caf586e5 | 1523 | atomic_long_inc(&dev->rx_dropped); |
6ec82562 | 1524 | kfree_skb(skb); |
44540960 | 1525 | return NET_RX_DROP; |
6ec82562 | 1526 | } |
8a83a00b | 1527 | skb_set_dev(skb, dev); |
44540960 AB |
1528 | skb->tstamp.tv64 = 0; |
1529 | skb->pkt_type = PACKET_HOST; | |
1530 | skb->protocol = eth_type_trans(skb, dev); | |
44540960 AB |
1531 | return netif_rx(skb); |
1532 | } | |
1533 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
1534 | ||
71d9dec2 CG |
1535 | static inline int deliver_skb(struct sk_buff *skb, |
1536 | struct packet_type *pt_prev, | |
1537 | struct net_device *orig_dev) | |
1538 | { | |
1539 | atomic_inc(&skb->users); | |
1540 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1541 | } | |
1542 | ||
1da177e4 LT |
1543 | /* |
1544 | * Support routine. Sends outgoing frames to any network | |
1545 | * taps currently in use. | |
1546 | */ | |
1547 | ||
f6a78bfc | 1548 | static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
1549 | { |
1550 | struct packet_type *ptype; | |
71d9dec2 CG |
1551 | struct sk_buff *skb2 = NULL; |
1552 | struct packet_type *pt_prev = NULL; | |
a61bbcf2 | 1553 | |
1da177e4 LT |
1554 | rcu_read_lock(); |
1555 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
1556 | /* Never send packets back to the socket | |
1557 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1558 | */ | |
1559 | if ((ptype->dev == dev || !ptype->dev) && | |
1560 | (ptype->af_packet_priv == NULL || | |
1561 | (struct sock *)ptype->af_packet_priv != skb->sk)) { | |
71d9dec2 CG |
1562 | if (pt_prev) { |
1563 | deliver_skb(skb2, pt_prev, skb->dev); | |
1564 | pt_prev = ptype; | |
1565 | continue; | |
1566 | } | |
1567 | ||
1568 | skb2 = skb_clone(skb, GFP_ATOMIC); | |
1da177e4 LT |
1569 | if (!skb2) |
1570 | break; | |
1571 | ||
70978182 ED |
1572 | net_timestamp_set(skb2); |
1573 | ||
1da177e4 LT |
1574 | /* skb->nh should be correctly |
1575 | set by sender, so that the second statement is | |
1576 | just protection against buggy protocols. | |
1577 | */ | |
459a98ed | 1578 | skb_reset_mac_header(skb2); |
1da177e4 | 1579 | |
d56f90a7 | 1580 | if (skb_network_header(skb2) < skb2->data || |
27a884dc | 1581 | skb2->network_header > skb2->tail) { |
1da177e4 LT |
1582 | if (net_ratelimit()) |
1583 | printk(KERN_CRIT "protocol %04x is " | |
1584 | "buggy, dev %s\n", | |
70777d03 SAS |
1585 | ntohs(skb2->protocol), |
1586 | dev->name); | |
c1d2bbe1 | 1587 | skb_reset_network_header(skb2); |
1da177e4 LT |
1588 | } |
1589 | ||
b0e380b1 | 1590 | skb2->transport_header = skb2->network_header; |
1da177e4 | 1591 | skb2->pkt_type = PACKET_OUTGOING; |
71d9dec2 | 1592 | pt_prev = ptype; |
1da177e4 LT |
1593 | } |
1594 | } | |
71d9dec2 CG |
1595 | if (pt_prev) |
1596 | pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); | |
1da177e4 LT |
1597 | rcu_read_unlock(); |
1598 | } | |
1599 | ||
f0796d5c JF |
1600 | /* |
1601 | * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues | |
1602 | * greater then real_num_tx_queues stale skbs on the qdisc must be flushed. | |
1603 | */ | |
e6484930 | 1604 | int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) |
f0796d5c | 1605 | { |
1d24eb48 TH |
1606 | int rc; |
1607 | ||
e6484930 TH |
1608 | if (txq < 1 || txq > dev->num_tx_queues) |
1609 | return -EINVAL; | |
f0796d5c | 1610 | |
e6484930 TH |
1611 | if (dev->reg_state == NETREG_REGISTERED) { |
1612 | ASSERT_RTNL(); | |
1613 | ||
1d24eb48 TH |
1614 | rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, |
1615 | txq); | |
bf264145 TH |
1616 | if (rc) |
1617 | return rc; | |
1618 | ||
e6484930 TH |
1619 | if (txq < dev->real_num_tx_queues) |
1620 | qdisc_reset_all_tx_gt(dev, txq); | |
f0796d5c | 1621 | } |
e6484930 TH |
1622 | |
1623 | dev->real_num_tx_queues = txq; | |
1624 | return 0; | |
f0796d5c JF |
1625 | } |
1626 | EXPORT_SYMBOL(netif_set_real_num_tx_queues); | |
56079431 | 1627 | |
62fe0b40 BH |
1628 | #ifdef CONFIG_RPS |
1629 | /** | |
1630 | * netif_set_real_num_rx_queues - set actual number of RX queues used | |
1631 | * @dev: Network device | |
1632 | * @rxq: Actual number of RX queues | |
1633 | * | |
1634 | * This must be called either with the rtnl_lock held or before | |
1635 | * registration of the net device. Returns 0 on success, or a | |
4e7f7951 BH |
1636 | * negative error code. If called before registration, it always |
1637 | * succeeds. | |
62fe0b40 BH |
1638 | */ |
1639 | int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) | |
1640 | { | |
1641 | int rc; | |
1642 | ||
bd25fa7b TH |
1643 | if (rxq < 1 || rxq > dev->num_rx_queues) |
1644 | return -EINVAL; | |
1645 | ||
62fe0b40 BH |
1646 | if (dev->reg_state == NETREG_REGISTERED) { |
1647 | ASSERT_RTNL(); | |
1648 | ||
62fe0b40 BH |
1649 | rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, |
1650 | rxq); | |
1651 | if (rc) | |
1652 | return rc; | |
62fe0b40 BH |
1653 | } |
1654 | ||
1655 | dev->real_num_rx_queues = rxq; | |
1656 | return 0; | |
1657 | } | |
1658 | EXPORT_SYMBOL(netif_set_real_num_rx_queues); | |
1659 | #endif | |
1660 | ||
def82a1d | 1661 | static inline void __netif_reschedule(struct Qdisc *q) |
56079431 | 1662 | { |
def82a1d JP |
1663 | struct softnet_data *sd; |
1664 | unsigned long flags; | |
56079431 | 1665 | |
def82a1d JP |
1666 | local_irq_save(flags); |
1667 | sd = &__get_cpu_var(softnet_data); | |
a9cbd588 CG |
1668 | q->next_sched = NULL; |
1669 | *sd->output_queue_tailp = q; | |
1670 | sd->output_queue_tailp = &q->next_sched; | |
def82a1d JP |
1671 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
1672 | local_irq_restore(flags); | |
1673 | } | |
1674 | ||
1675 | void __netif_schedule(struct Qdisc *q) | |
1676 | { | |
1677 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
1678 | __netif_reschedule(q); | |
56079431 DV |
1679 | } |
1680 | EXPORT_SYMBOL(__netif_schedule); | |
1681 | ||
bea3348e | 1682 | void dev_kfree_skb_irq(struct sk_buff *skb) |
56079431 | 1683 | { |
3578b0c8 | 1684 | if (atomic_dec_and_test(&skb->users)) { |
bea3348e SH |
1685 | struct softnet_data *sd; |
1686 | unsigned long flags; | |
56079431 | 1687 | |
bea3348e SH |
1688 | local_irq_save(flags); |
1689 | sd = &__get_cpu_var(softnet_data); | |
1690 | skb->next = sd->completion_queue; | |
1691 | sd->completion_queue = skb; | |
1692 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
1693 | local_irq_restore(flags); | |
1694 | } | |
56079431 | 1695 | } |
bea3348e | 1696 | EXPORT_SYMBOL(dev_kfree_skb_irq); |
56079431 DV |
1697 | |
1698 | void dev_kfree_skb_any(struct sk_buff *skb) | |
1699 | { | |
1700 | if (in_irq() || irqs_disabled()) | |
1701 | dev_kfree_skb_irq(skb); | |
1702 | else | |
1703 | dev_kfree_skb(skb); | |
1704 | } | |
1705 | EXPORT_SYMBOL(dev_kfree_skb_any); | |
1706 | ||
1707 | ||
bea3348e SH |
1708 | /** |
1709 | * netif_device_detach - mark device as removed | |
1710 | * @dev: network device | |
1711 | * | |
1712 | * Mark device as removed from system and therefore no longer available. | |
1713 | */ | |
56079431 DV |
1714 | void netif_device_detach(struct net_device *dev) |
1715 | { | |
1716 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
1717 | netif_running(dev)) { | |
d543103a | 1718 | netif_tx_stop_all_queues(dev); |
56079431 DV |
1719 | } |
1720 | } | |
1721 | EXPORT_SYMBOL(netif_device_detach); | |
1722 | ||
bea3348e SH |
1723 | /** |
1724 | * netif_device_attach - mark device as attached | |
1725 | * @dev: network device | |
1726 | * | |
1727 | * Mark device as attached from system and restart if needed. | |
1728 | */ | |
56079431 DV |
1729 | void netif_device_attach(struct net_device *dev) |
1730 | { | |
1731 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
1732 | netif_running(dev)) { | |
d543103a | 1733 | netif_tx_wake_all_queues(dev); |
4ec93edb | 1734 | __netdev_watchdog_up(dev); |
56079431 DV |
1735 | } |
1736 | } | |
1737 | EXPORT_SYMBOL(netif_device_attach); | |
1738 | ||
8a83a00b AB |
1739 | /** |
1740 | * skb_dev_set -- assign a new device to a buffer | |
1741 | * @skb: buffer for the new device | |
1742 | * @dev: network device | |
1743 | * | |
1744 | * If an skb is owned by a device already, we have to reset | |
1745 | * all data private to the namespace a device belongs to | |
1746 | * before assigning it a new device. | |
1747 | */ | |
1748 | #ifdef CONFIG_NET_NS | |
1749 | void skb_set_dev(struct sk_buff *skb, struct net_device *dev) | |
1750 | { | |
1751 | skb_dst_drop(skb); | |
1752 | if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) { | |
1753 | secpath_reset(skb); | |
1754 | nf_reset(skb); | |
1755 | skb_init_secmark(skb); | |
1756 | skb->mark = 0; | |
1757 | skb->priority = 0; | |
1758 | skb->nf_trace = 0; | |
1759 | skb->ipvs_property = 0; | |
1760 | #ifdef CONFIG_NET_SCHED | |
1761 | skb->tc_index = 0; | |
1762 | #endif | |
1763 | } | |
1764 | skb->dev = dev; | |
1765 | } | |
1766 | EXPORT_SYMBOL(skb_set_dev); | |
1767 | #endif /* CONFIG_NET_NS */ | |
1768 | ||
1da177e4 LT |
1769 | /* |
1770 | * Invalidate hardware checksum when packet is to be mangled, and | |
1771 | * complete checksum manually on outgoing path. | |
1772 | */ | |
84fa7933 | 1773 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 1774 | { |
d3bc23e7 | 1775 | __wsum csum; |
663ead3b | 1776 | int ret = 0, offset; |
1da177e4 | 1777 | |
84fa7933 | 1778 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
1779 | goto out_set_summed; |
1780 | ||
1781 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
a430a43d HX |
1782 | /* Let GSO fix up the checksum. */ |
1783 | goto out_set_summed; | |
1da177e4 LT |
1784 | } |
1785 | ||
55508d60 | 1786 | offset = skb_checksum_start_offset(skb); |
a030847e HX |
1787 | BUG_ON(offset >= skb_headlen(skb)); |
1788 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
1789 | ||
1790 | offset += skb->csum_offset; | |
1791 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
1792 | ||
1793 | if (skb_cloned(skb) && | |
1794 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1da177e4 LT |
1795 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
1796 | if (ret) | |
1797 | goto out; | |
1798 | } | |
1799 | ||
a030847e | 1800 | *(__sum16 *)(skb->data + offset) = csum_fold(csum); |
a430a43d | 1801 | out_set_summed: |
1da177e4 | 1802 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 1803 | out: |
1da177e4 LT |
1804 | return ret; |
1805 | } | |
d1b19dff | 1806 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 1807 | |
f6a78bfc HX |
1808 | /** |
1809 | * skb_gso_segment - Perform segmentation on skb. | |
1810 | * @skb: buffer to segment | |
576a30eb | 1811 | * @features: features for the output path (see dev->features) |
f6a78bfc HX |
1812 | * |
1813 | * This function segments the given skb and returns a list of segments. | |
576a30eb HX |
1814 | * |
1815 | * It may return NULL if the skb requires no segmentation. This is | |
1816 | * only possible when GSO is used for verifying header integrity. | |
f6a78bfc | 1817 | */ |
576a30eb | 1818 | struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features) |
f6a78bfc HX |
1819 | { |
1820 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
1821 | struct packet_type *ptype; | |
252e3346 | 1822 | __be16 type = skb->protocol; |
c8d5bcd1 | 1823 | int vlan_depth = ETH_HLEN; |
a430a43d | 1824 | int err; |
f6a78bfc | 1825 | |
c8d5bcd1 JG |
1826 | while (type == htons(ETH_P_8021Q)) { |
1827 | struct vlan_hdr *vh; | |
7b9c6090 | 1828 | |
c8d5bcd1 | 1829 | if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN))) |
7b9c6090 JG |
1830 | return ERR_PTR(-EINVAL); |
1831 | ||
c8d5bcd1 JG |
1832 | vh = (struct vlan_hdr *)(skb->data + vlan_depth); |
1833 | type = vh->h_vlan_encapsulated_proto; | |
1834 | vlan_depth += VLAN_HLEN; | |
7b9c6090 JG |
1835 | } |
1836 | ||
459a98ed | 1837 | skb_reset_mac_header(skb); |
b0e380b1 | 1838 | skb->mac_len = skb->network_header - skb->mac_header; |
f6a78bfc HX |
1839 | __skb_pull(skb, skb->mac_len); |
1840 | ||
67fd1a73 HX |
1841 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { |
1842 | struct net_device *dev = skb->dev; | |
1843 | struct ethtool_drvinfo info = {}; | |
1844 | ||
1845 | if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo) | |
1846 | dev->ethtool_ops->get_drvinfo(dev, &info); | |
1847 | ||
b194a367 | 1848 | WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n", |
67fd1a73 HX |
1849 | info.driver, dev ? dev->features : 0L, |
1850 | skb->sk ? skb->sk->sk_route_caps : 0L, | |
1851 | skb->len, skb->data_len, skb->ip_summed); | |
1852 | ||
a430a43d HX |
1853 | if (skb_header_cloned(skb) && |
1854 | (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))) | |
1855 | return ERR_PTR(err); | |
1856 | } | |
1857 | ||
f6a78bfc | 1858 | rcu_read_lock(); |
82d8a867 PE |
1859 | list_for_each_entry_rcu(ptype, |
1860 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
f6a78bfc | 1861 | if (ptype->type == type && !ptype->dev && ptype->gso_segment) { |
84fa7933 | 1862 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { |
a430a43d HX |
1863 | err = ptype->gso_send_check(skb); |
1864 | segs = ERR_PTR(err); | |
1865 | if (err || skb_gso_ok(skb, features)) | |
1866 | break; | |
d56f90a7 ACM |
1867 | __skb_push(skb, (skb->data - |
1868 | skb_network_header(skb))); | |
a430a43d | 1869 | } |
576a30eb | 1870 | segs = ptype->gso_segment(skb, features); |
f6a78bfc HX |
1871 | break; |
1872 | } | |
1873 | } | |
1874 | rcu_read_unlock(); | |
1875 | ||
98e399f8 | 1876 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 1877 | |
f6a78bfc HX |
1878 | return segs; |
1879 | } | |
f6a78bfc HX |
1880 | EXPORT_SYMBOL(skb_gso_segment); |
1881 | ||
fb286bb2 HX |
1882 | /* Take action when hardware reception checksum errors are detected. */ |
1883 | #ifdef CONFIG_BUG | |
1884 | void netdev_rx_csum_fault(struct net_device *dev) | |
1885 | { | |
1886 | if (net_ratelimit()) { | |
4ec93edb | 1887 | printk(KERN_ERR "%s: hw csum failure.\n", |
246a4212 | 1888 | dev ? dev->name : "<unknown>"); |
fb286bb2 HX |
1889 | dump_stack(); |
1890 | } | |
1891 | } | |
1892 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
1893 | #endif | |
1894 | ||
1da177e4 LT |
1895 | /* Actually, we should eliminate this check as soon as we know, that: |
1896 | * 1. IOMMU is present and allows to map all the memory. | |
1897 | * 2. No high memory really exists on this machine. | |
1898 | */ | |
1899 | ||
9092c658 | 1900 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
1da177e4 | 1901 | { |
3d3a8533 | 1902 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 1903 | int i; |
5acbbd42 FT |
1904 | if (!(dev->features & NETIF_F_HIGHDMA)) { |
1905 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) | |
1906 | if (PageHighMem(skb_shinfo(skb)->frags[i].page)) | |
1907 | return 1; | |
1908 | } | |
1da177e4 | 1909 | |
5acbbd42 FT |
1910 | if (PCI_DMA_BUS_IS_PHYS) { |
1911 | struct device *pdev = dev->dev.parent; | |
1da177e4 | 1912 | |
9092c658 ED |
1913 | if (!pdev) |
1914 | return 0; | |
5acbbd42 FT |
1915 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
1916 | dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page); | |
1917 | if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask) | |
1918 | return 1; | |
1919 | } | |
1920 | } | |
3d3a8533 | 1921 | #endif |
1da177e4 LT |
1922 | return 0; |
1923 | } | |
1da177e4 | 1924 | |
f6a78bfc HX |
1925 | struct dev_gso_cb { |
1926 | void (*destructor)(struct sk_buff *skb); | |
1927 | }; | |
1928 | ||
1929 | #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb) | |
1930 | ||
1931 | static void dev_gso_skb_destructor(struct sk_buff *skb) | |
1932 | { | |
1933 | struct dev_gso_cb *cb; | |
1934 | ||
1935 | do { | |
1936 | struct sk_buff *nskb = skb->next; | |
1937 | ||
1938 | skb->next = nskb->next; | |
1939 | nskb->next = NULL; | |
1940 | kfree_skb(nskb); | |
1941 | } while (skb->next); | |
1942 | ||
1943 | cb = DEV_GSO_CB(skb); | |
1944 | if (cb->destructor) | |
1945 | cb->destructor(skb); | |
1946 | } | |
1947 | ||
1948 | /** | |
1949 | * dev_gso_segment - Perform emulated hardware segmentation on skb. | |
1950 | * @skb: buffer to segment | |
91ecb63c | 1951 | * @features: device features as applicable to this skb |
f6a78bfc HX |
1952 | * |
1953 | * This function segments the given skb and stores the list of segments | |
1954 | * in skb->next. | |
1955 | */ | |
91ecb63c | 1956 | static int dev_gso_segment(struct sk_buff *skb, int features) |
f6a78bfc | 1957 | { |
f6a78bfc | 1958 | struct sk_buff *segs; |
576a30eb HX |
1959 | |
1960 | segs = skb_gso_segment(skb, features); | |
1961 | ||
1962 | /* Verifying header integrity only. */ | |
1963 | if (!segs) | |
1964 | return 0; | |
f6a78bfc | 1965 | |
801678c5 | 1966 | if (IS_ERR(segs)) |
f6a78bfc HX |
1967 | return PTR_ERR(segs); |
1968 | ||
1969 | skb->next = segs; | |
1970 | DEV_GSO_CB(skb)->destructor = skb->destructor; | |
1971 | skb->destructor = dev_gso_skb_destructor; | |
1972 | ||
1973 | return 0; | |
1974 | } | |
1975 | ||
fc6055a5 ED |
1976 | /* |
1977 | * Try to orphan skb early, right before transmission by the device. | |
2244d07b OH |
1978 | * We cannot orphan skb if tx timestamp is requested or the sk-reference |
1979 | * is needed on driver level for other reasons, e.g. see net/can/raw.c | |
fc6055a5 ED |
1980 | */ |
1981 | static inline void skb_orphan_try(struct sk_buff *skb) | |
1982 | { | |
87fd308c ED |
1983 | struct sock *sk = skb->sk; |
1984 | ||
2244d07b | 1985 | if (sk && !skb_shinfo(skb)->tx_flags) { |
87fd308c ED |
1986 | /* skb_tx_hash() wont be able to get sk. |
1987 | * We copy sk_hash into skb->rxhash | |
1988 | */ | |
1989 | if (!skb->rxhash) | |
1990 | skb->rxhash = sk->sk_hash; | |
fc6055a5 | 1991 | skb_orphan(skb); |
87fd308c | 1992 | } |
fc6055a5 ED |
1993 | } |
1994 | ||
03634668 JG |
1995 | static bool can_checksum_protocol(unsigned long features, __be16 protocol) |
1996 | { | |
1997 | return ((features & NETIF_F_GEN_CSUM) || | |
1998 | ((features & NETIF_F_V4_CSUM) && | |
1999 | protocol == htons(ETH_P_IP)) || | |
2000 | ((features & NETIF_F_V6_CSUM) && | |
2001 | protocol == htons(ETH_P_IPV6)) || | |
2002 | ((features & NETIF_F_FCOE_CRC) && | |
2003 | protocol == htons(ETH_P_FCOE))); | |
2004 | } | |
2005 | ||
f01a5236 JG |
2006 | static int harmonize_features(struct sk_buff *skb, __be16 protocol, int features) |
2007 | { | |
d402786e | 2008 | if (!can_checksum_protocol(features, protocol)) { |
f01a5236 JG |
2009 | features &= ~NETIF_F_ALL_CSUM; |
2010 | features &= ~NETIF_F_SG; | |
2011 | } else if (illegal_highdma(skb->dev, skb)) { | |
2012 | features &= ~NETIF_F_SG; | |
2013 | } | |
2014 | ||
2015 | return features; | |
2016 | } | |
2017 | ||
2018 | int netif_skb_features(struct sk_buff *skb) | |
58e998c6 JG |
2019 | { |
2020 | __be16 protocol = skb->protocol; | |
f01a5236 | 2021 | int features = skb->dev->features; |
58e998c6 JG |
2022 | |
2023 | if (protocol == htons(ETH_P_8021Q)) { | |
2024 | struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data; | |
2025 | protocol = veh->h_vlan_encapsulated_proto; | |
f01a5236 JG |
2026 | } else if (!vlan_tx_tag_present(skb)) { |
2027 | return harmonize_features(skb, protocol, features); | |
2028 | } | |
58e998c6 | 2029 | |
6ee400aa | 2030 | features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX); |
f01a5236 JG |
2031 | |
2032 | if (protocol != htons(ETH_P_8021Q)) { | |
2033 | return harmonize_features(skb, protocol, features); | |
2034 | } else { | |
2035 | features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | | |
6ee400aa | 2036 | NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX; |
f01a5236 JG |
2037 | return harmonize_features(skb, protocol, features); |
2038 | } | |
58e998c6 | 2039 | } |
f01a5236 | 2040 | EXPORT_SYMBOL(netif_skb_features); |
58e998c6 | 2041 | |
6afff0ca JF |
2042 | /* |
2043 | * Returns true if either: | |
2044 | * 1. skb has frag_list and the device doesn't support FRAGLIST, or | |
2045 | * 2. skb is fragmented and the device does not support SG, or if | |
2046 | * at least one of fragments is in highmem and device does not | |
2047 | * support DMA from it. | |
2048 | */ | |
2049 | static inline int skb_needs_linearize(struct sk_buff *skb, | |
02932ce9 | 2050 | int features) |
6afff0ca | 2051 | { |
02932ce9 JG |
2052 | return skb_is_nonlinear(skb) && |
2053 | ((skb_has_frag_list(skb) && | |
2054 | !(features & NETIF_F_FRAGLIST)) || | |
e1e78db6 | 2055 | (skb_shinfo(skb)->nr_frags && |
02932ce9 | 2056 | !(features & NETIF_F_SG))); |
6afff0ca JF |
2057 | } |
2058 | ||
fd2ea0a7 DM |
2059 | int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev, |
2060 | struct netdev_queue *txq) | |
f6a78bfc | 2061 | { |
00829823 | 2062 | const struct net_device_ops *ops = dev->netdev_ops; |
572a9d7b | 2063 | int rc = NETDEV_TX_OK; |
00829823 | 2064 | |
f6a78bfc | 2065 | if (likely(!skb->next)) { |
fc741216 JG |
2066 | int features; |
2067 | ||
93f154b5 ED |
2068 | /* |
2069 | * If device doesnt need skb->dst, release it right now while | |
2070 | * its hot in this cpu cache | |
2071 | */ | |
adf30907 ED |
2072 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) |
2073 | skb_dst_drop(skb); | |
2074 | ||
15c2d75f ED |
2075 | if (!list_empty(&ptype_all)) |
2076 | dev_queue_xmit_nit(skb, dev); | |
2077 | ||
fc6055a5 | 2078 | skb_orphan_try(skb); |
9ccb8975 | 2079 | |
fc741216 JG |
2080 | features = netif_skb_features(skb); |
2081 | ||
7b9c6090 | 2082 | if (vlan_tx_tag_present(skb) && |
fc741216 | 2083 | !(features & NETIF_F_HW_VLAN_TX)) { |
7b9c6090 JG |
2084 | skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb)); |
2085 | if (unlikely(!skb)) | |
2086 | goto out; | |
2087 | ||
2088 | skb->vlan_tci = 0; | |
2089 | } | |
2090 | ||
fc741216 | 2091 | if (netif_needs_gso(skb, features)) { |
91ecb63c | 2092 | if (unlikely(dev_gso_segment(skb, features))) |
9ccb8975 DM |
2093 | goto out_kfree_skb; |
2094 | if (skb->next) | |
2095 | goto gso; | |
6afff0ca | 2096 | } else { |
02932ce9 | 2097 | if (skb_needs_linearize(skb, features) && |
6afff0ca JF |
2098 | __skb_linearize(skb)) |
2099 | goto out_kfree_skb; | |
2100 | ||
2101 | /* If packet is not checksummed and device does not | |
2102 | * support checksumming for this protocol, complete | |
2103 | * checksumming here. | |
2104 | */ | |
2105 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
55508d60 MM |
2106 | skb_set_transport_header(skb, |
2107 | skb_checksum_start_offset(skb)); | |
03634668 | 2108 | if (!(features & NETIF_F_ALL_CSUM) && |
6afff0ca JF |
2109 | skb_checksum_help(skb)) |
2110 | goto out_kfree_skb; | |
2111 | } | |
9ccb8975 DM |
2112 | } |
2113 | ||
ac45f602 | 2114 | rc = ops->ndo_start_xmit(skb, dev); |
cf66ba58 | 2115 | trace_net_dev_xmit(skb, rc); |
ec634fe3 | 2116 | if (rc == NETDEV_TX_OK) |
08baf561 | 2117 | txq_trans_update(txq); |
ac45f602 | 2118 | return rc; |
f6a78bfc HX |
2119 | } |
2120 | ||
576a30eb | 2121 | gso: |
f6a78bfc HX |
2122 | do { |
2123 | struct sk_buff *nskb = skb->next; | |
f6a78bfc HX |
2124 | |
2125 | skb->next = nskb->next; | |
2126 | nskb->next = NULL; | |
068a2de5 KK |
2127 | |
2128 | /* | |
2129 | * If device doesnt need nskb->dst, release it right now while | |
2130 | * its hot in this cpu cache | |
2131 | */ | |
2132 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
2133 | skb_dst_drop(nskb); | |
2134 | ||
00829823 | 2135 | rc = ops->ndo_start_xmit(nskb, dev); |
cf66ba58 | 2136 | trace_net_dev_xmit(nskb, rc); |
ec634fe3 | 2137 | if (unlikely(rc != NETDEV_TX_OK)) { |
572a9d7b PM |
2138 | if (rc & ~NETDEV_TX_MASK) |
2139 | goto out_kfree_gso_skb; | |
f54d9e8d | 2140 | nskb->next = skb->next; |
f6a78bfc HX |
2141 | skb->next = nskb; |
2142 | return rc; | |
2143 | } | |
08baf561 | 2144 | txq_trans_update(txq); |
fd2ea0a7 | 2145 | if (unlikely(netif_tx_queue_stopped(txq) && skb->next)) |
f54d9e8d | 2146 | return NETDEV_TX_BUSY; |
f6a78bfc | 2147 | } while (skb->next); |
4ec93edb | 2148 | |
572a9d7b PM |
2149 | out_kfree_gso_skb: |
2150 | if (likely(skb->next == NULL)) | |
2151 | skb->destructor = DEV_GSO_CB(skb)->destructor; | |
f6a78bfc HX |
2152 | out_kfree_skb: |
2153 | kfree_skb(skb); | |
7b9c6090 | 2154 | out: |
572a9d7b | 2155 | return rc; |
f6a78bfc HX |
2156 | } |
2157 | ||
0a9627f2 | 2158 | static u32 hashrnd __read_mostly; |
b6b2fed1 | 2159 | |
a3d22a68 VZ |
2160 | /* |
2161 | * Returns a Tx hash based on the given packet descriptor a Tx queues' number | |
2162 | * to be used as a distribution range. | |
2163 | */ | |
2164 | u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb, | |
2165 | unsigned int num_tx_queues) | |
8f0f2223 | 2166 | { |
7019298a | 2167 | u32 hash; |
b6b2fed1 | 2168 | |
513de11b DM |
2169 | if (skb_rx_queue_recorded(skb)) { |
2170 | hash = skb_get_rx_queue(skb); | |
a3d22a68 VZ |
2171 | while (unlikely(hash >= num_tx_queues)) |
2172 | hash -= num_tx_queues; | |
513de11b DM |
2173 | return hash; |
2174 | } | |
ec581f6a ED |
2175 | |
2176 | if (skb->sk && skb->sk->sk_hash) | |
7019298a | 2177 | hash = skb->sk->sk_hash; |
ec581f6a | 2178 | else |
87fd308c | 2179 | hash = (__force u16) skb->protocol ^ skb->rxhash; |
0a9627f2 | 2180 | hash = jhash_1word(hash, hashrnd); |
b6b2fed1 | 2181 | |
a3d22a68 | 2182 | return (u16) (((u64) hash * num_tx_queues) >> 32); |
8f0f2223 | 2183 | } |
a3d22a68 | 2184 | EXPORT_SYMBOL(__skb_tx_hash); |
8f0f2223 | 2185 | |
ed04642f ED |
2186 | static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index) |
2187 | { | |
2188 | if (unlikely(queue_index >= dev->real_num_tx_queues)) { | |
2189 | if (net_ratelimit()) { | |
7a161ea9 ED |
2190 | pr_warning("%s selects TX queue %d, but " |
2191 | "real number of TX queues is %d\n", | |
2192 | dev->name, queue_index, dev->real_num_tx_queues); | |
ed04642f ED |
2193 | } |
2194 | return 0; | |
2195 | } | |
2196 | return queue_index; | |
2197 | } | |
2198 | ||
1d24eb48 TH |
2199 | static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb) |
2200 | { | |
bf264145 | 2201 | #ifdef CONFIG_XPS |
1d24eb48 TH |
2202 | struct xps_dev_maps *dev_maps; |
2203 | struct xps_map *map; | |
2204 | int queue_index = -1; | |
2205 | ||
2206 | rcu_read_lock(); | |
2207 | dev_maps = rcu_dereference(dev->xps_maps); | |
2208 | if (dev_maps) { | |
2209 | map = rcu_dereference( | |
2210 | dev_maps->cpu_map[raw_smp_processor_id()]); | |
2211 | if (map) { | |
2212 | if (map->len == 1) | |
2213 | queue_index = map->queues[0]; | |
2214 | else { | |
2215 | u32 hash; | |
2216 | if (skb->sk && skb->sk->sk_hash) | |
2217 | hash = skb->sk->sk_hash; | |
2218 | else | |
2219 | hash = (__force u16) skb->protocol ^ | |
2220 | skb->rxhash; | |
2221 | hash = jhash_1word(hash, hashrnd); | |
2222 | queue_index = map->queues[ | |
2223 | ((u64)hash * map->len) >> 32]; | |
2224 | } | |
2225 | if (unlikely(queue_index >= dev->real_num_tx_queues)) | |
2226 | queue_index = -1; | |
2227 | } | |
2228 | } | |
2229 | rcu_read_unlock(); | |
2230 | ||
2231 | return queue_index; | |
2232 | #else | |
2233 | return -1; | |
2234 | #endif | |
2235 | } | |
2236 | ||
e8a0464c DM |
2237 | static struct netdev_queue *dev_pick_tx(struct net_device *dev, |
2238 | struct sk_buff *skb) | |
2239 | { | |
b0f77d0e | 2240 | int queue_index; |
deabc772 | 2241 | const struct net_device_ops *ops = dev->netdev_ops; |
a4ee3ce3 | 2242 | |
3853b584 TH |
2243 | if (dev->real_num_tx_queues == 1) |
2244 | queue_index = 0; | |
2245 | else if (ops->ndo_select_queue) { | |
deabc772 HS |
2246 | queue_index = ops->ndo_select_queue(dev, skb); |
2247 | queue_index = dev_cap_txqueue(dev, queue_index); | |
2248 | } else { | |
2249 | struct sock *sk = skb->sk; | |
2250 | queue_index = sk_tx_queue_get(sk); | |
a4ee3ce3 | 2251 | |
3853b584 TH |
2252 | if (queue_index < 0 || skb->ooo_okay || |
2253 | queue_index >= dev->real_num_tx_queues) { | |
2254 | int old_index = queue_index; | |
fd2ea0a7 | 2255 | |
1d24eb48 TH |
2256 | queue_index = get_xps_queue(dev, skb); |
2257 | if (queue_index < 0) | |
2258 | queue_index = skb_tx_hash(dev, skb); | |
3853b584 TH |
2259 | |
2260 | if (queue_index != old_index && sk) { | |
2261 | struct dst_entry *dst = | |
2262 | rcu_dereference_check(sk->sk_dst_cache, 1); | |
8728c544 ED |
2263 | |
2264 | if (dst && skb_dst(skb) == dst) | |
2265 | sk_tx_queue_set(sk, queue_index); | |
2266 | } | |
a4ee3ce3 KK |
2267 | } |
2268 | } | |
eae792b7 | 2269 | |
fd2ea0a7 DM |
2270 | skb_set_queue_mapping(skb, queue_index); |
2271 | return netdev_get_tx_queue(dev, queue_index); | |
e8a0464c DM |
2272 | } |
2273 | ||
bbd8a0d3 KK |
2274 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
2275 | struct net_device *dev, | |
2276 | struct netdev_queue *txq) | |
2277 | { | |
2278 | spinlock_t *root_lock = qdisc_lock(q); | |
79640a4c | 2279 | bool contended = qdisc_is_running(q); |
bbd8a0d3 KK |
2280 | int rc; |
2281 | ||
79640a4c ED |
2282 | /* |
2283 | * Heuristic to force contended enqueues to serialize on a | |
2284 | * separate lock before trying to get qdisc main lock. | |
2285 | * This permits __QDISC_STATE_RUNNING owner to get the lock more often | |
2286 | * and dequeue packets faster. | |
2287 | */ | |
2288 | if (unlikely(contended)) | |
2289 | spin_lock(&q->busylock); | |
2290 | ||
bbd8a0d3 KK |
2291 | spin_lock(root_lock); |
2292 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
2293 | kfree_skb(skb); | |
2294 | rc = NET_XMIT_DROP; | |
2295 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
bc135b23 | 2296 | qdisc_run_begin(q)) { |
bbd8a0d3 KK |
2297 | /* |
2298 | * This is a work-conserving queue; there are no old skbs | |
2299 | * waiting to be sent out; and the qdisc is not running - | |
2300 | * xmit the skb directly. | |
2301 | */ | |
7fee226a ED |
2302 | if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE)) |
2303 | skb_dst_force(skb); | |
bfe0d029 ED |
2304 | |
2305 | qdisc_skb_cb(skb)->pkt_len = skb->len; | |
2306 | qdisc_bstats_update(q, skb); | |
2307 | ||
79640a4c ED |
2308 | if (sch_direct_xmit(skb, q, dev, txq, root_lock)) { |
2309 | if (unlikely(contended)) { | |
2310 | spin_unlock(&q->busylock); | |
2311 | contended = false; | |
2312 | } | |
bbd8a0d3 | 2313 | __qdisc_run(q); |
79640a4c | 2314 | } else |
bc135b23 | 2315 | qdisc_run_end(q); |
bbd8a0d3 KK |
2316 | |
2317 | rc = NET_XMIT_SUCCESS; | |
2318 | } else { | |
7fee226a | 2319 | skb_dst_force(skb); |
bbd8a0d3 | 2320 | rc = qdisc_enqueue_root(skb, q); |
79640a4c ED |
2321 | if (qdisc_run_begin(q)) { |
2322 | if (unlikely(contended)) { | |
2323 | spin_unlock(&q->busylock); | |
2324 | contended = false; | |
2325 | } | |
2326 | __qdisc_run(q); | |
2327 | } | |
bbd8a0d3 KK |
2328 | } |
2329 | spin_unlock(root_lock); | |
79640a4c ED |
2330 | if (unlikely(contended)) |
2331 | spin_unlock(&q->busylock); | |
bbd8a0d3 KK |
2332 | return rc; |
2333 | } | |
2334 | ||
745e20f1 | 2335 | static DEFINE_PER_CPU(int, xmit_recursion); |
11a766ce | 2336 | #define RECURSION_LIMIT 10 |
745e20f1 | 2337 | |
d29f749e DJ |
2338 | /** |
2339 | * dev_queue_xmit - transmit a buffer | |
2340 | * @skb: buffer to transmit | |
2341 | * | |
2342 | * Queue a buffer for transmission to a network device. The caller must | |
2343 | * have set the device and priority and built the buffer before calling | |
2344 | * this function. The function can be called from an interrupt. | |
2345 | * | |
2346 | * A negative errno code is returned on a failure. A success does not | |
2347 | * guarantee the frame will be transmitted as it may be dropped due | |
2348 | * to congestion or traffic shaping. | |
2349 | * | |
2350 | * ----------------------------------------------------------------------------------- | |
2351 | * I notice this method can also return errors from the queue disciplines, | |
2352 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
2353 | * be positive. | |
2354 | * | |
2355 | * Regardless of the return value, the skb is consumed, so it is currently | |
2356 | * difficult to retry a send to this method. (You can bump the ref count | |
2357 | * before sending to hold a reference for retry if you are careful.) | |
2358 | * | |
2359 | * When calling this method, interrupts MUST be enabled. This is because | |
2360 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
2361 | * --BLG | |
2362 | */ | |
1da177e4 LT |
2363 | int dev_queue_xmit(struct sk_buff *skb) |
2364 | { | |
2365 | struct net_device *dev = skb->dev; | |
dc2b4847 | 2366 | struct netdev_queue *txq; |
1da177e4 LT |
2367 | struct Qdisc *q; |
2368 | int rc = -ENOMEM; | |
2369 | ||
4ec93edb YH |
2370 | /* Disable soft irqs for various locks below. Also |
2371 | * stops preemption for RCU. | |
1da177e4 | 2372 | */ |
4ec93edb | 2373 | rcu_read_lock_bh(); |
1da177e4 | 2374 | |
eae792b7 | 2375 | txq = dev_pick_tx(dev, skb); |
a898def2 | 2376 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 2377 | |
1da177e4 | 2378 | #ifdef CONFIG_NET_CLS_ACT |
d1b19dff | 2379 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS); |
1da177e4 | 2380 | #endif |
cf66ba58 | 2381 | trace_net_dev_queue(skb); |
1da177e4 | 2382 | if (q->enqueue) { |
bbd8a0d3 | 2383 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 2384 | goto out; |
1da177e4 LT |
2385 | } |
2386 | ||
2387 | /* The device has no queue. Common case for software devices: | |
2388 | loopback, all the sorts of tunnels... | |
2389 | ||
932ff279 HX |
2390 | Really, it is unlikely that netif_tx_lock protection is necessary |
2391 | here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
1da177e4 LT |
2392 | counters.) |
2393 | However, it is possible, that they rely on protection | |
2394 | made by us here. | |
2395 | ||
2396 | Check this and shot the lock. It is not prone from deadlocks. | |
2397 | Either shot noqueue qdisc, it is even simpler 8) | |
2398 | */ | |
2399 | if (dev->flags & IFF_UP) { | |
2400 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
2401 | ||
c773e847 | 2402 | if (txq->xmit_lock_owner != cpu) { |
1da177e4 | 2403 | |
745e20f1 ED |
2404 | if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT) |
2405 | goto recursion_alert; | |
2406 | ||
c773e847 | 2407 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 2408 | |
fd2ea0a7 | 2409 | if (!netif_tx_queue_stopped(txq)) { |
745e20f1 | 2410 | __this_cpu_inc(xmit_recursion); |
572a9d7b | 2411 | rc = dev_hard_start_xmit(skb, dev, txq); |
745e20f1 | 2412 | __this_cpu_dec(xmit_recursion); |
572a9d7b | 2413 | if (dev_xmit_complete(rc)) { |
c773e847 | 2414 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
2415 | goto out; |
2416 | } | |
2417 | } | |
c773e847 | 2418 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
2419 | if (net_ratelimit()) |
2420 | printk(KERN_CRIT "Virtual device %s asks to " | |
2421 | "queue packet!\n", dev->name); | |
2422 | } else { | |
2423 | /* Recursion is detected! It is possible, | |
745e20f1 ED |
2424 | * unfortunately |
2425 | */ | |
2426 | recursion_alert: | |
1da177e4 LT |
2427 | if (net_ratelimit()) |
2428 | printk(KERN_CRIT "Dead loop on virtual device " | |
2429 | "%s, fix it urgently!\n", dev->name); | |
2430 | } | |
2431 | } | |
2432 | ||
2433 | rc = -ENETDOWN; | |
d4828d85 | 2434 | rcu_read_unlock_bh(); |
1da177e4 | 2435 | |
1da177e4 LT |
2436 | kfree_skb(skb); |
2437 | return rc; | |
2438 | out: | |
d4828d85 | 2439 | rcu_read_unlock_bh(); |
1da177e4 LT |
2440 | return rc; |
2441 | } | |
d1b19dff | 2442 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 LT |
2443 | |
2444 | ||
2445 | /*======================================================================= | |
2446 | Receiver routines | |
2447 | =======================================================================*/ | |
2448 | ||
6b2bedc3 | 2449 | int netdev_max_backlog __read_mostly = 1000; |
3b098e2d | 2450 | int netdev_tstamp_prequeue __read_mostly = 1; |
6b2bedc3 SH |
2451 | int netdev_budget __read_mostly = 300; |
2452 | int weight_p __read_mostly = 64; /* old backlog weight */ | |
1da177e4 | 2453 | |
eecfd7c4 ED |
2454 | /* Called with irq disabled */ |
2455 | static inline void ____napi_schedule(struct softnet_data *sd, | |
2456 | struct napi_struct *napi) | |
2457 | { | |
2458 | list_add_tail(&napi->poll_list, &sd->poll_list); | |
2459 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2460 | } | |
2461 | ||
0a9627f2 | 2462 | /* |
bfb564e7 KK |
2463 | * __skb_get_rxhash: calculate a flow hash based on src/dst addresses |
2464 | * and src/dst port numbers. Returns a non-zero hash number on success | |
2465 | * and 0 on failure. | |
0a9627f2 | 2466 | */ |
bfb564e7 | 2467 | __u32 __skb_get_rxhash(struct sk_buff *skb) |
0a9627f2 | 2468 | { |
12fcdefb | 2469 | int nhoff, hash = 0, poff; |
0a9627f2 TH |
2470 | struct ipv6hdr *ip6; |
2471 | struct iphdr *ip; | |
0a9627f2 | 2472 | u8 ip_proto; |
8c52d509 CG |
2473 | u32 addr1, addr2, ihl; |
2474 | union { | |
2475 | u32 v32; | |
2476 | u16 v16[2]; | |
2477 | } ports; | |
0a9627f2 | 2478 | |
bfb564e7 | 2479 | nhoff = skb_network_offset(skb); |
0a9627f2 TH |
2480 | |
2481 | switch (skb->protocol) { | |
2482 | case __constant_htons(ETH_P_IP): | |
bfb564e7 | 2483 | if (!pskb_may_pull(skb, sizeof(*ip) + nhoff)) |
0a9627f2 TH |
2484 | goto done; |
2485 | ||
1003489e | 2486 | ip = (struct iphdr *) (skb->data + nhoff); |
dbe5775b CG |
2487 | if (ip->frag_off & htons(IP_MF | IP_OFFSET)) |
2488 | ip_proto = 0; | |
2489 | else | |
2490 | ip_proto = ip->protocol; | |
b249dcb8 ED |
2491 | addr1 = (__force u32) ip->saddr; |
2492 | addr2 = (__force u32) ip->daddr; | |
0a9627f2 TH |
2493 | ihl = ip->ihl; |
2494 | break; | |
2495 | case __constant_htons(ETH_P_IPV6): | |
bfb564e7 | 2496 | if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff)) |
0a9627f2 TH |
2497 | goto done; |
2498 | ||
1003489e | 2499 | ip6 = (struct ipv6hdr *) (skb->data + nhoff); |
0a9627f2 | 2500 | ip_proto = ip6->nexthdr; |
b249dcb8 ED |
2501 | addr1 = (__force u32) ip6->saddr.s6_addr32[3]; |
2502 | addr2 = (__force u32) ip6->daddr.s6_addr32[3]; | |
0a9627f2 TH |
2503 | ihl = (40 >> 2); |
2504 | break; | |
2505 | default: | |
2506 | goto done; | |
2507 | } | |
bfb564e7 | 2508 | |
12fcdefb CG |
2509 | ports.v32 = 0; |
2510 | poff = proto_ports_offset(ip_proto); | |
2511 | if (poff >= 0) { | |
2512 | nhoff += ihl * 4 + poff; | |
2513 | if (pskb_may_pull(skb, nhoff + 4)) { | |
2514 | ports.v32 = * (__force u32 *) (skb->data + nhoff); | |
8c52d509 CG |
2515 | if (ports.v16[1] < ports.v16[0]) |
2516 | swap(ports.v16[0], ports.v16[1]); | |
b249dcb8 | 2517 | } |
0a9627f2 TH |
2518 | } |
2519 | ||
b249dcb8 ED |
2520 | /* get a consistent hash (same value on both flow directions) */ |
2521 | if (addr2 < addr1) | |
2522 | swap(addr1, addr2); | |
0a9627f2 | 2523 | |
bfb564e7 KK |
2524 | hash = jhash_3words(addr1, addr2, ports.v32, hashrnd); |
2525 | if (!hash) | |
2526 | hash = 1; | |
2527 | ||
2528 | done: | |
2529 | return hash; | |
2530 | } | |
2531 | EXPORT_SYMBOL(__skb_get_rxhash); | |
2532 | ||
2533 | #ifdef CONFIG_RPS | |
2534 | ||
2535 | /* One global table that all flow-based protocols share. */ | |
6e3f7faf | 2536 | struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; |
bfb564e7 KK |
2537 | EXPORT_SYMBOL(rps_sock_flow_table); |
2538 | ||
2539 | /* | |
2540 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
2541 | * CPU from the RPS map of the receiving queue for a given skb. | |
2542 | * rcu_read_lock must be held on entry. | |
2543 | */ | |
2544 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
2545 | struct rps_dev_flow **rflowp) | |
2546 | { | |
2547 | struct netdev_rx_queue *rxqueue; | |
6e3f7faf | 2548 | struct rps_map *map; |
bfb564e7 KK |
2549 | struct rps_dev_flow_table *flow_table; |
2550 | struct rps_sock_flow_table *sock_flow_table; | |
2551 | int cpu = -1; | |
2552 | u16 tcpu; | |
2553 | ||
2554 | if (skb_rx_queue_recorded(skb)) { | |
2555 | u16 index = skb_get_rx_queue(skb); | |
62fe0b40 BH |
2556 | if (unlikely(index >= dev->real_num_rx_queues)) { |
2557 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
2558 | "%s received packet on queue %u, but number " | |
2559 | "of RX queues is %u\n", | |
2560 | dev->name, index, dev->real_num_rx_queues); | |
bfb564e7 KK |
2561 | goto done; |
2562 | } | |
2563 | rxqueue = dev->_rx + index; | |
2564 | } else | |
2565 | rxqueue = dev->_rx; | |
2566 | ||
6e3f7faf ED |
2567 | map = rcu_dereference(rxqueue->rps_map); |
2568 | if (map) { | |
85875236 TH |
2569 | if (map->len == 1 && |
2570 | !rcu_dereference_raw(rxqueue->rps_flow_table)) { | |
6febfca9 CG |
2571 | tcpu = map->cpus[0]; |
2572 | if (cpu_online(tcpu)) | |
2573 | cpu = tcpu; | |
2574 | goto done; | |
2575 | } | |
6e3f7faf | 2576 | } else if (!rcu_dereference_raw(rxqueue->rps_flow_table)) { |
bfb564e7 | 2577 | goto done; |
6febfca9 | 2578 | } |
bfb564e7 | 2579 | |
2d47b459 | 2580 | skb_reset_network_header(skb); |
bfb564e7 KK |
2581 | if (!skb_get_rxhash(skb)) |
2582 | goto done; | |
2583 | ||
fec5e652 TH |
2584 | flow_table = rcu_dereference(rxqueue->rps_flow_table); |
2585 | sock_flow_table = rcu_dereference(rps_sock_flow_table); | |
2586 | if (flow_table && sock_flow_table) { | |
2587 | u16 next_cpu; | |
2588 | struct rps_dev_flow *rflow; | |
2589 | ||
2590 | rflow = &flow_table->flows[skb->rxhash & flow_table->mask]; | |
2591 | tcpu = rflow->cpu; | |
2592 | ||
2593 | next_cpu = sock_flow_table->ents[skb->rxhash & | |
2594 | sock_flow_table->mask]; | |
2595 | ||
2596 | /* | |
2597 | * If the desired CPU (where last recvmsg was done) is | |
2598 | * different from current CPU (one in the rx-queue flow | |
2599 | * table entry), switch if one of the following holds: | |
2600 | * - Current CPU is unset (equal to RPS_NO_CPU). | |
2601 | * - Current CPU is offline. | |
2602 | * - The current CPU's queue tail has advanced beyond the | |
2603 | * last packet that was enqueued using this table entry. | |
2604 | * This guarantees that all previous packets for the flow | |
2605 | * have been dequeued, thus preserving in order delivery. | |
2606 | */ | |
2607 | if (unlikely(tcpu != next_cpu) && | |
2608 | (tcpu == RPS_NO_CPU || !cpu_online(tcpu) || | |
2609 | ((int)(per_cpu(softnet_data, tcpu).input_queue_head - | |
2610 | rflow->last_qtail)) >= 0)) { | |
2611 | tcpu = rflow->cpu = next_cpu; | |
2612 | if (tcpu != RPS_NO_CPU) | |
2613 | rflow->last_qtail = per_cpu(softnet_data, | |
2614 | tcpu).input_queue_head; | |
2615 | } | |
2616 | if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) { | |
2617 | *rflowp = rflow; | |
2618 | cpu = tcpu; | |
2619 | goto done; | |
2620 | } | |
2621 | } | |
2622 | ||
0a9627f2 | 2623 | if (map) { |
fec5e652 | 2624 | tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32]; |
0a9627f2 TH |
2625 | |
2626 | if (cpu_online(tcpu)) { | |
2627 | cpu = tcpu; | |
2628 | goto done; | |
2629 | } | |
2630 | } | |
2631 | ||
2632 | done: | |
0a9627f2 TH |
2633 | return cpu; |
2634 | } | |
2635 | ||
0a9627f2 | 2636 | /* Called from hardirq (IPI) context */ |
e36fa2f7 | 2637 | static void rps_trigger_softirq(void *data) |
0a9627f2 | 2638 | { |
e36fa2f7 ED |
2639 | struct softnet_data *sd = data; |
2640 | ||
eecfd7c4 | 2641 | ____napi_schedule(sd, &sd->backlog); |
dee42870 | 2642 | sd->received_rps++; |
0a9627f2 | 2643 | } |
e36fa2f7 | 2644 | |
fec5e652 | 2645 | #endif /* CONFIG_RPS */ |
0a9627f2 | 2646 | |
e36fa2f7 ED |
2647 | /* |
2648 | * Check if this softnet_data structure is another cpu one | |
2649 | * If yes, queue it to our IPI list and return 1 | |
2650 | * If no, return 0 | |
2651 | */ | |
2652 | static int rps_ipi_queued(struct softnet_data *sd) | |
2653 | { | |
2654 | #ifdef CONFIG_RPS | |
2655 | struct softnet_data *mysd = &__get_cpu_var(softnet_data); | |
2656 | ||
2657 | if (sd != mysd) { | |
2658 | sd->rps_ipi_next = mysd->rps_ipi_list; | |
2659 | mysd->rps_ipi_list = sd; | |
2660 | ||
2661 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2662 | return 1; | |
2663 | } | |
2664 | #endif /* CONFIG_RPS */ | |
2665 | return 0; | |
2666 | } | |
2667 | ||
0a9627f2 TH |
2668 | /* |
2669 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
2670 | * queue (may be a remote CPU queue). | |
2671 | */ | |
fec5e652 TH |
2672 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, |
2673 | unsigned int *qtail) | |
0a9627f2 | 2674 | { |
e36fa2f7 | 2675 | struct softnet_data *sd; |
0a9627f2 TH |
2676 | unsigned long flags; |
2677 | ||
e36fa2f7 | 2678 | sd = &per_cpu(softnet_data, cpu); |
0a9627f2 TH |
2679 | |
2680 | local_irq_save(flags); | |
0a9627f2 | 2681 | |
e36fa2f7 | 2682 | rps_lock(sd); |
6e7676c1 CG |
2683 | if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) { |
2684 | if (skb_queue_len(&sd->input_pkt_queue)) { | |
0a9627f2 | 2685 | enqueue: |
e36fa2f7 | 2686 | __skb_queue_tail(&sd->input_pkt_queue, skb); |
76cc8b13 | 2687 | input_queue_tail_incr_save(sd, qtail); |
e36fa2f7 | 2688 | rps_unlock(sd); |
152102c7 | 2689 | local_irq_restore(flags); |
0a9627f2 TH |
2690 | return NET_RX_SUCCESS; |
2691 | } | |
2692 | ||
ebda37c2 ED |
2693 | /* Schedule NAPI for backlog device |
2694 | * We can use non atomic operation since we own the queue lock | |
2695 | */ | |
2696 | if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { | |
e36fa2f7 | 2697 | if (!rps_ipi_queued(sd)) |
eecfd7c4 | 2698 | ____napi_schedule(sd, &sd->backlog); |
0a9627f2 TH |
2699 | } |
2700 | goto enqueue; | |
2701 | } | |
2702 | ||
dee42870 | 2703 | sd->dropped++; |
e36fa2f7 | 2704 | rps_unlock(sd); |
0a9627f2 | 2705 | |
0a9627f2 TH |
2706 | local_irq_restore(flags); |
2707 | ||
caf586e5 | 2708 | atomic_long_inc(&skb->dev->rx_dropped); |
0a9627f2 TH |
2709 | kfree_skb(skb); |
2710 | return NET_RX_DROP; | |
2711 | } | |
1da177e4 | 2712 | |
1da177e4 LT |
2713 | /** |
2714 | * netif_rx - post buffer to the network code | |
2715 | * @skb: buffer to post | |
2716 | * | |
2717 | * This function receives a packet from a device driver and queues it for | |
2718 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
2719 | * may be dropped during processing for congestion control or by the | |
2720 | * protocol layers. | |
2721 | * | |
2722 | * return values: | |
2723 | * NET_RX_SUCCESS (no congestion) | |
1da177e4 LT |
2724 | * NET_RX_DROP (packet was dropped) |
2725 | * | |
2726 | */ | |
2727 | ||
2728 | int netif_rx(struct sk_buff *skb) | |
2729 | { | |
b0e28f1e | 2730 | int ret; |
1da177e4 LT |
2731 | |
2732 | /* if netpoll wants it, pretend we never saw it */ | |
2733 | if (netpoll_rx(skb)) | |
2734 | return NET_RX_DROP; | |
2735 | ||
3b098e2d ED |
2736 | if (netdev_tstamp_prequeue) |
2737 | net_timestamp_check(skb); | |
1da177e4 | 2738 | |
cf66ba58 | 2739 | trace_netif_rx(skb); |
df334545 | 2740 | #ifdef CONFIG_RPS |
b0e28f1e | 2741 | { |
fec5e652 | 2742 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
b0e28f1e ED |
2743 | int cpu; |
2744 | ||
cece1945 | 2745 | preempt_disable(); |
b0e28f1e | 2746 | rcu_read_lock(); |
fec5e652 TH |
2747 | |
2748 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
b0e28f1e ED |
2749 | if (cpu < 0) |
2750 | cpu = smp_processor_id(); | |
fec5e652 TH |
2751 | |
2752 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
2753 | ||
b0e28f1e | 2754 | rcu_read_unlock(); |
cece1945 | 2755 | preempt_enable(); |
b0e28f1e | 2756 | } |
1e94d72f | 2757 | #else |
fec5e652 TH |
2758 | { |
2759 | unsigned int qtail; | |
2760 | ret = enqueue_to_backlog(skb, get_cpu(), &qtail); | |
2761 | put_cpu(); | |
2762 | } | |
1e94d72f | 2763 | #endif |
b0e28f1e | 2764 | return ret; |
1da177e4 | 2765 | } |
d1b19dff | 2766 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
2767 | |
2768 | int netif_rx_ni(struct sk_buff *skb) | |
2769 | { | |
2770 | int err; | |
2771 | ||
2772 | preempt_disable(); | |
2773 | err = netif_rx(skb); | |
2774 | if (local_softirq_pending()) | |
2775 | do_softirq(); | |
2776 | preempt_enable(); | |
2777 | ||
2778 | return err; | |
2779 | } | |
1da177e4 LT |
2780 | EXPORT_SYMBOL(netif_rx_ni); |
2781 | ||
1da177e4 LT |
2782 | static void net_tx_action(struct softirq_action *h) |
2783 | { | |
2784 | struct softnet_data *sd = &__get_cpu_var(softnet_data); | |
2785 | ||
2786 | if (sd->completion_queue) { | |
2787 | struct sk_buff *clist; | |
2788 | ||
2789 | local_irq_disable(); | |
2790 | clist = sd->completion_queue; | |
2791 | sd->completion_queue = NULL; | |
2792 | local_irq_enable(); | |
2793 | ||
2794 | while (clist) { | |
2795 | struct sk_buff *skb = clist; | |
2796 | clist = clist->next; | |
2797 | ||
547b792c | 2798 | WARN_ON(atomic_read(&skb->users)); |
07dc22e7 | 2799 | trace_kfree_skb(skb, net_tx_action); |
1da177e4 LT |
2800 | __kfree_skb(skb); |
2801 | } | |
2802 | } | |
2803 | ||
2804 | if (sd->output_queue) { | |
37437bb2 | 2805 | struct Qdisc *head; |
1da177e4 LT |
2806 | |
2807 | local_irq_disable(); | |
2808 | head = sd->output_queue; | |
2809 | sd->output_queue = NULL; | |
a9cbd588 | 2810 | sd->output_queue_tailp = &sd->output_queue; |
1da177e4 LT |
2811 | local_irq_enable(); |
2812 | ||
2813 | while (head) { | |
37437bb2 DM |
2814 | struct Qdisc *q = head; |
2815 | spinlock_t *root_lock; | |
2816 | ||
1da177e4 LT |
2817 | head = head->next_sched; |
2818 | ||
5fb66229 | 2819 | root_lock = qdisc_lock(q); |
37437bb2 | 2820 | if (spin_trylock(root_lock)) { |
def82a1d JP |
2821 | smp_mb__before_clear_bit(); |
2822 | clear_bit(__QDISC_STATE_SCHED, | |
2823 | &q->state); | |
37437bb2 DM |
2824 | qdisc_run(q); |
2825 | spin_unlock(root_lock); | |
1da177e4 | 2826 | } else { |
195648bb | 2827 | if (!test_bit(__QDISC_STATE_DEACTIVATED, |
e8a83e10 | 2828 | &q->state)) { |
195648bb | 2829 | __netif_reschedule(q); |
e8a83e10 JP |
2830 | } else { |
2831 | smp_mb__before_clear_bit(); | |
2832 | clear_bit(__QDISC_STATE_SCHED, | |
2833 | &q->state); | |
2834 | } | |
1da177e4 LT |
2835 | } |
2836 | } | |
2837 | } | |
2838 | } | |
2839 | ||
ab95bfe0 JP |
2840 | #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \ |
2841 | (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)) | |
da678292 MM |
2842 | /* This hook is defined here for ATM LANE */ |
2843 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
2844 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 2845 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 2846 | #endif |
1da177e4 | 2847 | |
1da177e4 LT |
2848 | #ifdef CONFIG_NET_CLS_ACT |
2849 | /* TODO: Maybe we should just force sch_ingress to be compiled in | |
2850 | * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions | |
2851 | * a compare and 2 stores extra right now if we dont have it on | |
2852 | * but have CONFIG_NET_CLS_ACT | |
4ec93edb | 2853 | * NOTE: This doesnt stop any functionality; if you dont have |
1da177e4 LT |
2854 | * the ingress scheduler, you just cant add policies on ingress. |
2855 | * | |
2856 | */ | |
24824a09 | 2857 | static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq) |
1da177e4 | 2858 | { |
1da177e4 | 2859 | struct net_device *dev = skb->dev; |
f697c3e8 | 2860 | u32 ttl = G_TC_RTTL(skb->tc_verd); |
555353cf DM |
2861 | int result = TC_ACT_OK; |
2862 | struct Qdisc *q; | |
4ec93edb | 2863 | |
de384830 SH |
2864 | if (unlikely(MAX_RED_LOOP < ttl++)) { |
2865 | if (net_ratelimit()) | |
2866 | pr_warning( "Redir loop detected Dropping packet (%d->%d)\n", | |
2867 | skb->skb_iif, dev->ifindex); | |
f697c3e8 HX |
2868 | return TC_ACT_SHOT; |
2869 | } | |
1da177e4 | 2870 | |
f697c3e8 HX |
2871 | skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl); |
2872 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); | |
1da177e4 | 2873 | |
83874000 | 2874 | q = rxq->qdisc; |
8d50b53d | 2875 | if (q != &noop_qdisc) { |
83874000 | 2876 | spin_lock(qdisc_lock(q)); |
a9312ae8 DM |
2877 | if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) |
2878 | result = qdisc_enqueue_root(skb, q); | |
83874000 DM |
2879 | spin_unlock(qdisc_lock(q)); |
2880 | } | |
f697c3e8 HX |
2881 | |
2882 | return result; | |
2883 | } | |
86e65da9 | 2884 | |
f697c3e8 HX |
2885 | static inline struct sk_buff *handle_ing(struct sk_buff *skb, |
2886 | struct packet_type **pt_prev, | |
2887 | int *ret, struct net_device *orig_dev) | |
2888 | { | |
24824a09 ED |
2889 | struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue); |
2890 | ||
2891 | if (!rxq || rxq->qdisc == &noop_qdisc) | |
f697c3e8 | 2892 | goto out; |
1da177e4 | 2893 | |
f697c3e8 HX |
2894 | if (*pt_prev) { |
2895 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
2896 | *pt_prev = NULL; | |
1da177e4 LT |
2897 | } |
2898 | ||
24824a09 | 2899 | switch (ing_filter(skb, rxq)) { |
f697c3e8 HX |
2900 | case TC_ACT_SHOT: |
2901 | case TC_ACT_STOLEN: | |
2902 | kfree_skb(skb); | |
2903 | return NULL; | |
2904 | } | |
2905 | ||
2906 | out: | |
2907 | skb->tc_verd = 0; | |
2908 | return skb; | |
1da177e4 LT |
2909 | } |
2910 | #endif | |
2911 | ||
ab95bfe0 JP |
2912 | /** |
2913 | * netdev_rx_handler_register - register receive handler | |
2914 | * @dev: device to register a handler for | |
2915 | * @rx_handler: receive handler to register | |
93e2c32b | 2916 | * @rx_handler_data: data pointer that is used by rx handler |
ab95bfe0 JP |
2917 | * |
2918 | * Register a receive hander for a device. This handler will then be | |
2919 | * called from __netif_receive_skb. A negative errno code is returned | |
2920 | * on a failure. | |
2921 | * | |
2922 | * The caller must hold the rtnl_mutex. | |
2923 | */ | |
2924 | int netdev_rx_handler_register(struct net_device *dev, | |
93e2c32b JP |
2925 | rx_handler_func_t *rx_handler, |
2926 | void *rx_handler_data) | |
ab95bfe0 JP |
2927 | { |
2928 | ASSERT_RTNL(); | |
2929 | ||
2930 | if (dev->rx_handler) | |
2931 | return -EBUSY; | |
2932 | ||
93e2c32b | 2933 | rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); |
ab95bfe0 JP |
2934 | rcu_assign_pointer(dev->rx_handler, rx_handler); |
2935 | ||
2936 | return 0; | |
2937 | } | |
2938 | EXPORT_SYMBOL_GPL(netdev_rx_handler_register); | |
2939 | ||
2940 | /** | |
2941 | * netdev_rx_handler_unregister - unregister receive handler | |
2942 | * @dev: device to unregister a handler from | |
2943 | * | |
2944 | * Unregister a receive hander from a device. | |
2945 | * | |
2946 | * The caller must hold the rtnl_mutex. | |
2947 | */ | |
2948 | void netdev_rx_handler_unregister(struct net_device *dev) | |
2949 | { | |
2950 | ||
2951 | ASSERT_RTNL(); | |
2952 | rcu_assign_pointer(dev->rx_handler, NULL); | |
93e2c32b | 2953 | rcu_assign_pointer(dev->rx_handler_data, NULL); |
ab95bfe0 JP |
2954 | } |
2955 | EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); | |
2956 | ||
acbbc071 ED |
2957 | static inline void skb_bond_set_mac_by_master(struct sk_buff *skb, |
2958 | struct net_device *master) | |
2959 | { | |
2960 | if (skb->pkt_type == PACKET_HOST) { | |
2961 | u16 *dest = (u16 *) eth_hdr(skb)->h_dest; | |
2962 | ||
2963 | memcpy(dest, master->dev_addr, ETH_ALEN); | |
2964 | } | |
2965 | } | |
2966 | ||
2967 | /* On bonding slaves other than the currently active slave, suppress | |
2968 | * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and | |
2969 | * ARP on active-backup slaves with arp_validate enabled. | |
2970 | */ | |
2971 | int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master) | |
2972 | { | |
2973 | struct net_device *dev = skb->dev; | |
2974 | ||
2975 | if (master->priv_flags & IFF_MASTER_ARPMON) | |
2976 | dev->last_rx = jiffies; | |
2977 | ||
f350a0a8 JP |
2978 | if ((master->priv_flags & IFF_MASTER_ALB) && |
2979 | (master->priv_flags & IFF_BRIDGE_PORT)) { | |
acbbc071 ED |
2980 | /* Do address unmangle. The local destination address |
2981 | * will be always the one master has. Provides the right | |
2982 | * functionality in a bridge. | |
2983 | */ | |
2984 | skb_bond_set_mac_by_master(skb, master); | |
2985 | } | |
2986 | ||
2987 | if (dev->priv_flags & IFF_SLAVE_INACTIVE) { | |
2988 | if ((dev->priv_flags & IFF_SLAVE_NEEDARP) && | |
2989 | skb->protocol == __cpu_to_be16(ETH_P_ARP)) | |
2990 | return 0; | |
2991 | ||
2992 | if (master->priv_flags & IFF_MASTER_ALB) { | |
2993 | if (skb->pkt_type != PACKET_BROADCAST && | |
2994 | skb->pkt_type != PACKET_MULTICAST) | |
2995 | return 0; | |
2996 | } | |
2997 | if (master->priv_flags & IFF_MASTER_8023AD && | |
2998 | skb->protocol == __cpu_to_be16(ETH_P_SLOW)) | |
2999 | return 0; | |
3000 | ||
3001 | return 1; | |
3002 | } | |
3003 | return 0; | |
3004 | } | |
3005 | EXPORT_SYMBOL(__skb_bond_should_drop); | |
3006 | ||
10f744d2 | 3007 | static int __netif_receive_skb(struct sk_buff *skb) |
1da177e4 LT |
3008 | { |
3009 | struct packet_type *ptype, *pt_prev; | |
ab95bfe0 | 3010 | rx_handler_func_t *rx_handler; |
f2ccd8fa | 3011 | struct net_device *orig_dev; |
0641e4fb | 3012 | struct net_device *master; |
0d7a3681 | 3013 | struct net_device *null_or_orig; |
2df4a0fa | 3014 | struct net_device *orig_or_bond; |
1da177e4 | 3015 | int ret = NET_RX_DROP; |
252e3346 | 3016 | __be16 type; |
1da177e4 | 3017 | |
3b098e2d ED |
3018 | if (!netdev_tstamp_prequeue) |
3019 | net_timestamp_check(skb); | |
81bbb3d4 | 3020 | |
cf66ba58 | 3021 | trace_netif_receive_skb(skb); |
9b22ea56 | 3022 | |
1da177e4 | 3023 | /* if we've gotten here through NAPI, check netpoll */ |
bea3348e | 3024 | if (netpoll_receive_skb(skb)) |
1da177e4 LT |
3025 | return NET_RX_DROP; |
3026 | ||
8964be4a ED |
3027 | if (!skb->skb_iif) |
3028 | skb->skb_iif = skb->dev->ifindex; | |
86e65da9 | 3029 | |
597a264b JF |
3030 | /* |
3031 | * bonding note: skbs received on inactive slaves should only | |
3032 | * be delivered to pkt handlers that are exact matches. Also | |
3033 | * the deliver_no_wcard flag will be set. If packet handlers | |
3034 | * are sensitive to duplicate packets these skbs will need to | |
3701e513 | 3035 | * be dropped at the handler. |
597a264b | 3036 | */ |
0d7a3681 | 3037 | null_or_orig = NULL; |
cc9bd5ce | 3038 | orig_dev = skb->dev; |
0641e4fb | 3039 | master = ACCESS_ONCE(orig_dev->master); |
597a264b JF |
3040 | if (skb->deliver_no_wcard) |
3041 | null_or_orig = orig_dev; | |
3042 | else if (master) { | |
3043 | if (skb_bond_should_drop(skb, master)) { | |
3044 | skb->deliver_no_wcard = 1; | |
0d7a3681 | 3045 | null_or_orig = orig_dev; /* deliver only exact match */ |
597a264b | 3046 | } else |
0641e4fb | 3047 | skb->dev = master; |
cc9bd5ce | 3048 | } |
8f903c70 | 3049 | |
27f39c73 | 3050 | __this_cpu_inc(softnet_data.processed); |
c1d2bbe1 | 3051 | skb_reset_network_header(skb); |
badff6d0 | 3052 | skb_reset_transport_header(skb); |
b0e380b1 | 3053 | skb->mac_len = skb->network_header - skb->mac_header; |
1da177e4 LT |
3054 | |
3055 | pt_prev = NULL; | |
3056 | ||
3057 | rcu_read_lock(); | |
3058 | ||
3059 | #ifdef CONFIG_NET_CLS_ACT | |
3060 | if (skb->tc_verd & TC_NCLS) { | |
3061 | skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); | |
3062 | goto ncls; | |
3063 | } | |
3064 | #endif | |
3065 | ||
3066 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
f982307f JE |
3067 | if (ptype->dev == null_or_orig || ptype->dev == skb->dev || |
3068 | ptype->dev == orig_dev) { | |
4ec93edb | 3069 | if (pt_prev) |
f2ccd8fa | 3070 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
3071 | pt_prev = ptype; |
3072 | } | |
3073 | } | |
3074 | ||
3075 | #ifdef CONFIG_NET_CLS_ACT | |
f697c3e8 HX |
3076 | skb = handle_ing(skb, &pt_prev, &ret, orig_dev); |
3077 | if (!skb) | |
1da177e4 | 3078 | goto out; |
1da177e4 LT |
3079 | ncls: |
3080 | #endif | |
3081 | ||
ab95bfe0 JP |
3082 | /* Handle special case of bridge or macvlan */ |
3083 | rx_handler = rcu_dereference(skb->dev->rx_handler); | |
3084 | if (rx_handler) { | |
3085 | if (pt_prev) { | |
3086 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3087 | pt_prev = NULL; | |
3088 | } | |
3089 | skb = rx_handler(skb); | |
3090 | if (!skb) | |
3091 | goto out; | |
3092 | } | |
1da177e4 | 3093 | |
3701e513 JG |
3094 | if (vlan_tx_tag_present(skb)) { |
3095 | if (pt_prev) { | |
3096 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3097 | pt_prev = NULL; | |
3098 | } | |
3099 | if (vlan_hwaccel_do_receive(&skb)) { | |
3100 | ret = __netif_receive_skb(skb); | |
3101 | goto out; | |
3102 | } else if (unlikely(!skb)) | |
3103 | goto out; | |
3104 | } | |
3105 | ||
1f3c8804 AG |
3106 | /* |
3107 | * Make sure frames received on VLAN interfaces stacked on | |
3108 | * bonding interfaces still make their way to any base bonding | |
3109 | * device that may have registered for a specific ptype. The | |
3110 | * handler may have to adjust skb->dev and orig_dev. | |
1f3c8804 | 3111 | */ |
2df4a0fa | 3112 | orig_or_bond = orig_dev; |
1f3c8804 AG |
3113 | if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) && |
3114 | (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) { | |
2df4a0fa | 3115 | orig_or_bond = vlan_dev_real_dev(skb->dev); |
1f3c8804 AG |
3116 | } |
3117 | ||
1da177e4 | 3118 | type = skb->protocol; |
82d8a867 PE |
3119 | list_for_each_entry_rcu(ptype, |
3120 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
1f3c8804 | 3121 | if (ptype->type == type && (ptype->dev == null_or_orig || |
ca8d9ea3 | 3122 | ptype->dev == skb->dev || ptype->dev == orig_dev || |
2df4a0fa | 3123 | ptype->dev == orig_or_bond)) { |
4ec93edb | 3124 | if (pt_prev) |
f2ccd8fa | 3125 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
3126 | pt_prev = ptype; |
3127 | } | |
3128 | } | |
3129 | ||
3130 | if (pt_prev) { | |
f2ccd8fa | 3131 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
1da177e4 | 3132 | } else { |
caf586e5 | 3133 | atomic_long_inc(&skb->dev->rx_dropped); |
1da177e4 LT |
3134 | kfree_skb(skb); |
3135 | /* Jamal, now you will not able to escape explaining | |
3136 | * me how you were going to use this. :-) | |
3137 | */ | |
3138 | ret = NET_RX_DROP; | |
3139 | } | |
3140 | ||
3141 | out: | |
3142 | rcu_read_unlock(); | |
3143 | return ret; | |
3144 | } | |
0a9627f2 TH |
3145 | |
3146 | /** | |
3147 | * netif_receive_skb - process receive buffer from network | |
3148 | * @skb: buffer to process | |
3149 | * | |
3150 | * netif_receive_skb() is the main receive data processing function. | |
3151 | * It always succeeds. The buffer may be dropped during processing | |
3152 | * for congestion control or by the protocol layers. | |
3153 | * | |
3154 | * This function may only be called from softirq context and interrupts | |
3155 | * should be enabled. | |
3156 | * | |
3157 | * Return values (usually ignored): | |
3158 | * NET_RX_SUCCESS: no congestion | |
3159 | * NET_RX_DROP: packet was dropped | |
3160 | */ | |
3161 | int netif_receive_skb(struct sk_buff *skb) | |
3162 | { | |
3b098e2d ED |
3163 | if (netdev_tstamp_prequeue) |
3164 | net_timestamp_check(skb); | |
3165 | ||
c1f19b51 RC |
3166 | if (skb_defer_rx_timestamp(skb)) |
3167 | return NET_RX_SUCCESS; | |
3168 | ||
df334545 | 3169 | #ifdef CONFIG_RPS |
3b098e2d ED |
3170 | { |
3171 | struct rps_dev_flow voidflow, *rflow = &voidflow; | |
3172 | int cpu, ret; | |
fec5e652 | 3173 | |
3b098e2d ED |
3174 | rcu_read_lock(); |
3175 | ||
3176 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
0a9627f2 | 3177 | |
3b098e2d ED |
3178 | if (cpu >= 0) { |
3179 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
3180 | rcu_read_unlock(); | |
3181 | } else { | |
3182 | rcu_read_unlock(); | |
3183 | ret = __netif_receive_skb(skb); | |
3184 | } | |
0a9627f2 | 3185 | |
3b098e2d | 3186 | return ret; |
fec5e652 | 3187 | } |
1e94d72f TH |
3188 | #else |
3189 | return __netif_receive_skb(skb); | |
3190 | #endif | |
0a9627f2 | 3191 | } |
d1b19dff | 3192 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 3193 | |
88751275 ED |
3194 | /* Network device is going away, flush any packets still pending |
3195 | * Called with irqs disabled. | |
3196 | */ | |
152102c7 | 3197 | static void flush_backlog(void *arg) |
6e583ce5 | 3198 | { |
152102c7 | 3199 | struct net_device *dev = arg; |
e36fa2f7 | 3200 | struct softnet_data *sd = &__get_cpu_var(softnet_data); |
6e583ce5 SH |
3201 | struct sk_buff *skb, *tmp; |
3202 | ||
e36fa2f7 | 3203 | rps_lock(sd); |
6e7676c1 | 3204 | skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { |
6e583ce5 | 3205 | if (skb->dev == dev) { |
e36fa2f7 | 3206 | __skb_unlink(skb, &sd->input_pkt_queue); |
6e583ce5 | 3207 | kfree_skb(skb); |
76cc8b13 | 3208 | input_queue_head_incr(sd); |
6e583ce5 | 3209 | } |
6e7676c1 | 3210 | } |
e36fa2f7 | 3211 | rps_unlock(sd); |
6e7676c1 CG |
3212 | |
3213 | skb_queue_walk_safe(&sd->process_queue, skb, tmp) { | |
3214 | if (skb->dev == dev) { | |
3215 | __skb_unlink(skb, &sd->process_queue); | |
3216 | kfree_skb(skb); | |
76cc8b13 | 3217 | input_queue_head_incr(sd); |
6e7676c1 CG |
3218 | } |
3219 | } | |
6e583ce5 SH |
3220 | } |
3221 | ||
d565b0a1 HX |
3222 | static int napi_gro_complete(struct sk_buff *skb) |
3223 | { | |
3224 | struct packet_type *ptype; | |
3225 | __be16 type = skb->protocol; | |
3226 | struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK]; | |
3227 | int err = -ENOENT; | |
3228 | ||
fc59f9a3 HX |
3229 | if (NAPI_GRO_CB(skb)->count == 1) { |
3230 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 3231 | goto out; |
fc59f9a3 | 3232 | } |
d565b0a1 HX |
3233 | |
3234 | rcu_read_lock(); | |
3235 | list_for_each_entry_rcu(ptype, head, list) { | |
3236 | if (ptype->type != type || ptype->dev || !ptype->gro_complete) | |
3237 | continue; | |
3238 | ||
3239 | err = ptype->gro_complete(skb); | |
3240 | break; | |
3241 | } | |
3242 | rcu_read_unlock(); | |
3243 | ||
3244 | if (err) { | |
3245 | WARN_ON(&ptype->list == head); | |
3246 | kfree_skb(skb); | |
3247 | return NET_RX_SUCCESS; | |
3248 | } | |
3249 | ||
3250 | out: | |
d565b0a1 HX |
3251 | return netif_receive_skb(skb); |
3252 | } | |
3253 | ||
86cac58b | 3254 | inline void napi_gro_flush(struct napi_struct *napi) |
d565b0a1 HX |
3255 | { |
3256 | struct sk_buff *skb, *next; | |
3257 | ||
3258 | for (skb = napi->gro_list; skb; skb = next) { | |
3259 | next = skb->next; | |
3260 | skb->next = NULL; | |
3261 | napi_gro_complete(skb); | |
3262 | } | |
3263 | ||
4ae5544f | 3264 | napi->gro_count = 0; |
d565b0a1 HX |
3265 | napi->gro_list = NULL; |
3266 | } | |
86cac58b | 3267 | EXPORT_SYMBOL(napi_gro_flush); |
d565b0a1 | 3268 | |
5b252f0c | 3269 | enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 HX |
3270 | { |
3271 | struct sk_buff **pp = NULL; | |
3272 | struct packet_type *ptype; | |
3273 | __be16 type = skb->protocol; | |
3274 | struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK]; | |
0da2afd5 | 3275 | int same_flow; |
d565b0a1 | 3276 | int mac_len; |
5b252f0c | 3277 | enum gro_result ret; |
d565b0a1 | 3278 | |
ce9e76c8 | 3279 | if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb)) |
d565b0a1 HX |
3280 | goto normal; |
3281 | ||
21dc3301 | 3282 | if (skb_is_gso(skb) || skb_has_frag_list(skb)) |
f17f5c91 HX |
3283 | goto normal; |
3284 | ||
d565b0a1 HX |
3285 | rcu_read_lock(); |
3286 | list_for_each_entry_rcu(ptype, head, list) { | |
d565b0a1 HX |
3287 | if (ptype->type != type || ptype->dev || !ptype->gro_receive) |
3288 | continue; | |
3289 | ||
86911732 | 3290 | skb_set_network_header(skb, skb_gro_offset(skb)); |
d565b0a1 HX |
3291 | mac_len = skb->network_header - skb->mac_header; |
3292 | skb->mac_len = mac_len; | |
3293 | NAPI_GRO_CB(skb)->same_flow = 0; | |
3294 | NAPI_GRO_CB(skb)->flush = 0; | |
5d38a079 | 3295 | NAPI_GRO_CB(skb)->free = 0; |
d565b0a1 | 3296 | |
d565b0a1 HX |
3297 | pp = ptype->gro_receive(&napi->gro_list, skb); |
3298 | break; | |
3299 | } | |
3300 | rcu_read_unlock(); | |
3301 | ||
3302 | if (&ptype->list == head) | |
3303 | goto normal; | |
3304 | ||
0da2afd5 | 3305 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 3306 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 3307 | |
d565b0a1 HX |
3308 | if (pp) { |
3309 | struct sk_buff *nskb = *pp; | |
3310 | ||
3311 | *pp = nskb->next; | |
3312 | nskb->next = NULL; | |
3313 | napi_gro_complete(nskb); | |
4ae5544f | 3314 | napi->gro_count--; |
d565b0a1 HX |
3315 | } |
3316 | ||
0da2afd5 | 3317 | if (same_flow) |
d565b0a1 HX |
3318 | goto ok; |
3319 | ||
4ae5544f | 3320 | if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS) |
d565b0a1 | 3321 | goto normal; |
d565b0a1 | 3322 | |
4ae5544f | 3323 | napi->gro_count++; |
d565b0a1 | 3324 | NAPI_GRO_CB(skb)->count = 1; |
86911732 | 3325 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
d565b0a1 HX |
3326 | skb->next = napi->gro_list; |
3327 | napi->gro_list = skb; | |
5d0d9be8 | 3328 | ret = GRO_HELD; |
d565b0a1 | 3329 | |
ad0f9904 | 3330 | pull: |
cb18978c HX |
3331 | if (skb_headlen(skb) < skb_gro_offset(skb)) { |
3332 | int grow = skb_gro_offset(skb) - skb_headlen(skb); | |
3333 | ||
3334 | BUG_ON(skb->end - skb->tail < grow); | |
3335 | ||
3336 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
3337 | ||
3338 | skb->tail += grow; | |
3339 | skb->data_len -= grow; | |
3340 | ||
3341 | skb_shinfo(skb)->frags[0].page_offset += grow; | |
3342 | skb_shinfo(skb)->frags[0].size -= grow; | |
3343 | ||
3344 | if (unlikely(!skb_shinfo(skb)->frags[0].size)) { | |
3345 | put_page(skb_shinfo(skb)->frags[0].page); | |
3346 | memmove(skb_shinfo(skb)->frags, | |
3347 | skb_shinfo(skb)->frags + 1, | |
e5093aec | 3348 | --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t)); |
cb18978c | 3349 | } |
ad0f9904 HX |
3350 | } |
3351 | ||
d565b0a1 | 3352 | ok: |
5d0d9be8 | 3353 | return ret; |
d565b0a1 HX |
3354 | |
3355 | normal: | |
ad0f9904 HX |
3356 | ret = GRO_NORMAL; |
3357 | goto pull; | |
5d38a079 | 3358 | } |
96e93eab HX |
3359 | EXPORT_SYMBOL(dev_gro_receive); |
3360 | ||
40d0802b | 3361 | static inline gro_result_t |
5b252f0c | 3362 | __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab HX |
3363 | { |
3364 | struct sk_buff *p; | |
3365 | ||
3366 | for (p = napi->gro_list; p; p = p->next) { | |
40d0802b ED |
3367 | unsigned long diffs; |
3368 | ||
3369 | diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; | |
3701e513 | 3370 | diffs |= p->vlan_tci ^ skb->vlan_tci; |
40d0802b | 3371 | diffs |= compare_ether_header(skb_mac_header(p), |
f64f9e71 | 3372 | skb_gro_mac_header(skb)); |
40d0802b | 3373 | NAPI_GRO_CB(p)->same_flow = !diffs; |
96e93eab HX |
3374 | NAPI_GRO_CB(p)->flush = 0; |
3375 | } | |
3376 | ||
3377 | return dev_gro_receive(napi, skb); | |
3378 | } | |
5d38a079 | 3379 | |
c7c4b3b6 | 3380 | gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) |
5d38a079 | 3381 | { |
5d0d9be8 HX |
3382 | switch (ret) { |
3383 | case GRO_NORMAL: | |
c7c4b3b6 BH |
3384 | if (netif_receive_skb(skb)) |
3385 | ret = GRO_DROP; | |
3386 | break; | |
5d38a079 | 3387 | |
5d0d9be8 | 3388 | case GRO_DROP: |
5d0d9be8 | 3389 | case GRO_MERGED_FREE: |
5d38a079 HX |
3390 | kfree_skb(skb); |
3391 | break; | |
5b252f0c BH |
3392 | |
3393 | case GRO_HELD: | |
3394 | case GRO_MERGED: | |
3395 | break; | |
5d38a079 HX |
3396 | } |
3397 | ||
c7c4b3b6 | 3398 | return ret; |
5d0d9be8 HX |
3399 | } |
3400 | EXPORT_SYMBOL(napi_skb_finish); | |
3401 | ||
78a478d0 HX |
3402 | void skb_gro_reset_offset(struct sk_buff *skb) |
3403 | { | |
3404 | NAPI_GRO_CB(skb)->data_offset = 0; | |
3405 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
7489594c | 3406 | NAPI_GRO_CB(skb)->frag0_len = 0; |
78a478d0 | 3407 | |
78d3fd0b | 3408 | if (skb->mac_header == skb->tail && |
7489594c | 3409 | !PageHighMem(skb_shinfo(skb)->frags[0].page)) { |
78a478d0 HX |
3410 | NAPI_GRO_CB(skb)->frag0 = |
3411 | page_address(skb_shinfo(skb)->frags[0].page) + | |
3412 | skb_shinfo(skb)->frags[0].page_offset; | |
7489594c HX |
3413 | NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size; |
3414 | } | |
78a478d0 HX |
3415 | } |
3416 | EXPORT_SYMBOL(skb_gro_reset_offset); | |
3417 | ||
c7c4b3b6 | 3418 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 3419 | { |
86911732 HX |
3420 | skb_gro_reset_offset(skb); |
3421 | ||
5d0d9be8 | 3422 | return napi_skb_finish(__napi_gro_receive(napi, skb), skb); |
d565b0a1 HX |
3423 | } |
3424 | EXPORT_SYMBOL(napi_gro_receive); | |
3425 | ||
d0c2b0d2 | 3426 | static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab | 3427 | { |
96e93eab HX |
3428 | __skb_pull(skb, skb_headlen(skb)); |
3429 | skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb)); | |
3701e513 | 3430 | skb->vlan_tci = 0; |
66c46d74 | 3431 | skb->dev = napi->dev; |
6d152e23 | 3432 | skb->skb_iif = 0; |
96e93eab HX |
3433 | |
3434 | napi->skb = skb; | |
3435 | } | |
96e93eab | 3436 | |
76620aaf | 3437 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 3438 | { |
5d38a079 | 3439 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
3440 | |
3441 | if (!skb) { | |
89d71a66 ED |
3442 | skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD); |
3443 | if (skb) | |
3444 | napi->skb = skb; | |
80595d59 | 3445 | } |
96e93eab HX |
3446 | return skb; |
3447 | } | |
76620aaf | 3448 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 3449 | |
c7c4b3b6 BH |
3450 | gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb, |
3451 | gro_result_t ret) | |
96e93eab | 3452 | { |
5d0d9be8 HX |
3453 | switch (ret) { |
3454 | case GRO_NORMAL: | |
86911732 | 3455 | case GRO_HELD: |
e76b69cc | 3456 | skb->protocol = eth_type_trans(skb, skb->dev); |
86911732 | 3457 | |
c7c4b3b6 BH |
3458 | if (ret == GRO_HELD) |
3459 | skb_gro_pull(skb, -ETH_HLEN); | |
3460 | else if (netif_receive_skb(skb)) | |
3461 | ret = GRO_DROP; | |
86911732 | 3462 | break; |
5d38a079 | 3463 | |
5d0d9be8 | 3464 | case GRO_DROP: |
5d0d9be8 HX |
3465 | case GRO_MERGED_FREE: |
3466 | napi_reuse_skb(napi, skb); | |
3467 | break; | |
5b252f0c BH |
3468 | |
3469 | case GRO_MERGED: | |
3470 | break; | |
5d0d9be8 | 3471 | } |
5d38a079 | 3472 | |
c7c4b3b6 | 3473 | return ret; |
5d38a079 | 3474 | } |
5d0d9be8 HX |
3475 | EXPORT_SYMBOL(napi_frags_finish); |
3476 | ||
76620aaf HX |
3477 | struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
3478 | { | |
3479 | struct sk_buff *skb = napi->skb; | |
3480 | struct ethhdr *eth; | |
a5b1cf28 HX |
3481 | unsigned int hlen; |
3482 | unsigned int off; | |
76620aaf HX |
3483 | |
3484 | napi->skb = NULL; | |
3485 | ||
3486 | skb_reset_mac_header(skb); | |
3487 | skb_gro_reset_offset(skb); | |
3488 | ||
a5b1cf28 HX |
3489 | off = skb_gro_offset(skb); |
3490 | hlen = off + sizeof(*eth); | |
3491 | eth = skb_gro_header_fast(skb, off); | |
3492 | if (skb_gro_header_hard(skb, hlen)) { | |
3493 | eth = skb_gro_header_slow(skb, hlen, off); | |
3494 | if (unlikely(!eth)) { | |
3495 | napi_reuse_skb(napi, skb); | |
3496 | skb = NULL; | |
3497 | goto out; | |
3498 | } | |
76620aaf HX |
3499 | } |
3500 | ||
3501 | skb_gro_pull(skb, sizeof(*eth)); | |
3502 | ||
3503 | /* | |
3504 | * This works because the only protocols we care about don't require | |
3505 | * special handling. We'll fix it up properly at the end. | |
3506 | */ | |
3507 | skb->protocol = eth->h_proto; | |
3508 | ||
3509 | out: | |
3510 | return skb; | |
3511 | } | |
3512 | EXPORT_SYMBOL(napi_frags_skb); | |
3513 | ||
c7c4b3b6 | 3514 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 3515 | { |
76620aaf | 3516 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 HX |
3517 | |
3518 | if (!skb) | |
c7c4b3b6 | 3519 | return GRO_DROP; |
5d0d9be8 HX |
3520 | |
3521 | return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb)); | |
3522 | } | |
5d38a079 HX |
3523 | EXPORT_SYMBOL(napi_gro_frags); |
3524 | ||
e326bed2 ED |
3525 | /* |
3526 | * net_rps_action sends any pending IPI's for rps. | |
3527 | * Note: called with local irq disabled, but exits with local irq enabled. | |
3528 | */ | |
3529 | static void net_rps_action_and_irq_enable(struct softnet_data *sd) | |
3530 | { | |
3531 | #ifdef CONFIG_RPS | |
3532 | struct softnet_data *remsd = sd->rps_ipi_list; | |
3533 | ||
3534 | if (remsd) { | |
3535 | sd->rps_ipi_list = NULL; | |
3536 | ||
3537 | local_irq_enable(); | |
3538 | ||
3539 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
3540 | while (remsd) { | |
3541 | struct softnet_data *next = remsd->rps_ipi_next; | |
3542 | ||
3543 | if (cpu_online(remsd->cpu)) | |
3544 | __smp_call_function_single(remsd->cpu, | |
3545 | &remsd->csd, 0); | |
3546 | remsd = next; | |
3547 | } | |
3548 | } else | |
3549 | #endif | |
3550 | local_irq_enable(); | |
3551 | } | |
3552 | ||
bea3348e | 3553 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 LT |
3554 | { |
3555 | int work = 0; | |
eecfd7c4 | 3556 | struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); |
1da177e4 | 3557 | |
e326bed2 ED |
3558 | #ifdef CONFIG_RPS |
3559 | /* Check if we have pending ipi, its better to send them now, | |
3560 | * not waiting net_rx_action() end. | |
3561 | */ | |
3562 | if (sd->rps_ipi_list) { | |
3563 | local_irq_disable(); | |
3564 | net_rps_action_and_irq_enable(sd); | |
3565 | } | |
3566 | #endif | |
bea3348e | 3567 | napi->weight = weight_p; |
6e7676c1 CG |
3568 | local_irq_disable(); |
3569 | while (work < quota) { | |
1da177e4 | 3570 | struct sk_buff *skb; |
6e7676c1 CG |
3571 | unsigned int qlen; |
3572 | ||
3573 | while ((skb = __skb_dequeue(&sd->process_queue))) { | |
3574 | local_irq_enable(); | |
3575 | __netif_receive_skb(skb); | |
6e7676c1 | 3576 | local_irq_disable(); |
76cc8b13 TH |
3577 | input_queue_head_incr(sd); |
3578 | if (++work >= quota) { | |
3579 | local_irq_enable(); | |
3580 | return work; | |
3581 | } | |
6e7676c1 | 3582 | } |
1da177e4 | 3583 | |
e36fa2f7 | 3584 | rps_lock(sd); |
6e7676c1 | 3585 | qlen = skb_queue_len(&sd->input_pkt_queue); |
76cc8b13 | 3586 | if (qlen) |
6e7676c1 CG |
3587 | skb_queue_splice_tail_init(&sd->input_pkt_queue, |
3588 | &sd->process_queue); | |
76cc8b13 | 3589 | |
6e7676c1 | 3590 | if (qlen < quota - work) { |
eecfd7c4 ED |
3591 | /* |
3592 | * Inline a custom version of __napi_complete(). | |
3593 | * only current cpu owns and manipulates this napi, | |
3594 | * and NAPI_STATE_SCHED is the only possible flag set on backlog. | |
3595 | * we can use a plain write instead of clear_bit(), | |
3596 | * and we dont need an smp_mb() memory barrier. | |
3597 | */ | |
3598 | list_del(&napi->poll_list); | |
3599 | napi->state = 0; | |
3600 | ||
6e7676c1 | 3601 | quota = work + qlen; |
bea3348e | 3602 | } |
e36fa2f7 | 3603 | rps_unlock(sd); |
6e7676c1 CG |
3604 | } |
3605 | local_irq_enable(); | |
1da177e4 | 3606 | |
bea3348e SH |
3607 | return work; |
3608 | } | |
1da177e4 | 3609 | |
bea3348e SH |
3610 | /** |
3611 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 3612 | * @n: entry to schedule |
bea3348e SH |
3613 | * |
3614 | * The entry's receive function will be scheduled to run | |
3615 | */ | |
b5606c2d | 3616 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
3617 | { |
3618 | unsigned long flags; | |
1da177e4 | 3619 | |
bea3348e | 3620 | local_irq_save(flags); |
eecfd7c4 | 3621 | ____napi_schedule(&__get_cpu_var(softnet_data), n); |
bea3348e | 3622 | local_irq_restore(flags); |
1da177e4 | 3623 | } |
bea3348e SH |
3624 | EXPORT_SYMBOL(__napi_schedule); |
3625 | ||
d565b0a1 HX |
3626 | void __napi_complete(struct napi_struct *n) |
3627 | { | |
3628 | BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); | |
3629 | BUG_ON(n->gro_list); | |
3630 | ||
3631 | list_del(&n->poll_list); | |
3632 | smp_mb__before_clear_bit(); | |
3633 | clear_bit(NAPI_STATE_SCHED, &n->state); | |
3634 | } | |
3635 | EXPORT_SYMBOL(__napi_complete); | |
3636 | ||
3637 | void napi_complete(struct napi_struct *n) | |
3638 | { | |
3639 | unsigned long flags; | |
3640 | ||
3641 | /* | |
3642 | * don't let napi dequeue from the cpu poll list | |
3643 | * just in case its running on a different cpu | |
3644 | */ | |
3645 | if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state))) | |
3646 | return; | |
3647 | ||
3648 | napi_gro_flush(n); | |
3649 | local_irq_save(flags); | |
3650 | __napi_complete(n); | |
3651 | local_irq_restore(flags); | |
3652 | } | |
3653 | EXPORT_SYMBOL(napi_complete); | |
3654 | ||
3655 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, | |
3656 | int (*poll)(struct napi_struct *, int), int weight) | |
3657 | { | |
3658 | INIT_LIST_HEAD(&napi->poll_list); | |
4ae5544f | 3659 | napi->gro_count = 0; |
d565b0a1 | 3660 | napi->gro_list = NULL; |
5d38a079 | 3661 | napi->skb = NULL; |
d565b0a1 HX |
3662 | napi->poll = poll; |
3663 | napi->weight = weight; | |
3664 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 3665 | napi->dev = dev; |
5d38a079 | 3666 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
3667 | spin_lock_init(&napi->poll_lock); |
3668 | napi->poll_owner = -1; | |
3669 | #endif | |
3670 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
3671 | } | |
3672 | EXPORT_SYMBOL(netif_napi_add); | |
3673 | ||
3674 | void netif_napi_del(struct napi_struct *napi) | |
3675 | { | |
3676 | struct sk_buff *skb, *next; | |
3677 | ||
d7b06636 | 3678 | list_del_init(&napi->dev_list); |
76620aaf | 3679 | napi_free_frags(napi); |
d565b0a1 HX |
3680 | |
3681 | for (skb = napi->gro_list; skb; skb = next) { | |
3682 | next = skb->next; | |
3683 | skb->next = NULL; | |
3684 | kfree_skb(skb); | |
3685 | } | |
3686 | ||
3687 | napi->gro_list = NULL; | |
4ae5544f | 3688 | napi->gro_count = 0; |
d565b0a1 HX |
3689 | } |
3690 | EXPORT_SYMBOL(netif_napi_del); | |
3691 | ||
1da177e4 LT |
3692 | static void net_rx_action(struct softirq_action *h) |
3693 | { | |
e326bed2 | 3694 | struct softnet_data *sd = &__get_cpu_var(softnet_data); |
24f8b238 | 3695 | unsigned long time_limit = jiffies + 2; |
51b0bded | 3696 | int budget = netdev_budget; |
53fb95d3 MM |
3697 | void *have; |
3698 | ||
1da177e4 LT |
3699 | local_irq_disable(); |
3700 | ||
e326bed2 | 3701 | while (!list_empty(&sd->poll_list)) { |
bea3348e SH |
3702 | struct napi_struct *n; |
3703 | int work, weight; | |
1da177e4 | 3704 | |
bea3348e | 3705 | /* If softirq window is exhuasted then punt. |
24f8b238 SH |
3706 | * Allow this to run for 2 jiffies since which will allow |
3707 | * an average latency of 1.5/HZ. | |
bea3348e | 3708 | */ |
24f8b238 | 3709 | if (unlikely(budget <= 0 || time_after(jiffies, time_limit))) |
1da177e4 LT |
3710 | goto softnet_break; |
3711 | ||
3712 | local_irq_enable(); | |
3713 | ||
bea3348e SH |
3714 | /* Even though interrupts have been re-enabled, this |
3715 | * access is safe because interrupts can only add new | |
3716 | * entries to the tail of this list, and only ->poll() | |
3717 | * calls can remove this head entry from the list. | |
3718 | */ | |
e326bed2 | 3719 | n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list); |
1da177e4 | 3720 | |
bea3348e SH |
3721 | have = netpoll_poll_lock(n); |
3722 | ||
3723 | weight = n->weight; | |
3724 | ||
0a7606c1 DM |
3725 | /* This NAPI_STATE_SCHED test is for avoiding a race |
3726 | * with netpoll's poll_napi(). Only the entity which | |
3727 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
3728 | * actually make the ->poll() call. Therefore we avoid | |
3729 | * accidently calling ->poll() when NAPI is not scheduled. | |
3730 | */ | |
3731 | work = 0; | |
4ea7e386 | 3732 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { |
0a7606c1 | 3733 | work = n->poll(n, weight); |
4ea7e386 NH |
3734 | trace_napi_poll(n); |
3735 | } | |
bea3348e SH |
3736 | |
3737 | WARN_ON_ONCE(work > weight); | |
3738 | ||
3739 | budget -= work; | |
3740 | ||
3741 | local_irq_disable(); | |
3742 | ||
3743 | /* Drivers must not modify the NAPI state if they | |
3744 | * consume the entire weight. In such cases this code | |
3745 | * still "owns" the NAPI instance and therefore can | |
3746 | * move the instance around on the list at-will. | |
3747 | */ | |
fed17f30 | 3748 | if (unlikely(work == weight)) { |
ff780cd8 HX |
3749 | if (unlikely(napi_disable_pending(n))) { |
3750 | local_irq_enable(); | |
3751 | napi_complete(n); | |
3752 | local_irq_disable(); | |
3753 | } else | |
e326bed2 | 3754 | list_move_tail(&n->poll_list, &sd->poll_list); |
fed17f30 | 3755 | } |
bea3348e SH |
3756 | |
3757 | netpoll_poll_unlock(have); | |
1da177e4 LT |
3758 | } |
3759 | out: | |
e326bed2 | 3760 | net_rps_action_and_irq_enable(sd); |
0a9627f2 | 3761 | |
db217334 CL |
3762 | #ifdef CONFIG_NET_DMA |
3763 | /* | |
3764 | * There may not be any more sk_buffs coming right now, so push | |
3765 | * any pending DMA copies to hardware | |
3766 | */ | |
2ba05622 | 3767 | dma_issue_pending_all(); |
db217334 | 3768 | #endif |
bea3348e | 3769 | |
1da177e4 LT |
3770 | return; |
3771 | ||
3772 | softnet_break: | |
dee42870 | 3773 | sd->time_squeeze++; |
1da177e4 LT |
3774 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); |
3775 | goto out; | |
3776 | } | |
3777 | ||
d1b19dff | 3778 | static gifconf_func_t *gifconf_list[NPROTO]; |
1da177e4 LT |
3779 | |
3780 | /** | |
3781 | * register_gifconf - register a SIOCGIF handler | |
3782 | * @family: Address family | |
3783 | * @gifconf: Function handler | |
3784 | * | |
3785 | * Register protocol dependent address dumping routines. The handler | |
3786 | * that is passed must not be freed or reused until it has been replaced | |
3787 | * by another handler. | |
3788 | */ | |
d1b19dff | 3789 | int register_gifconf(unsigned int family, gifconf_func_t *gifconf) |
1da177e4 LT |
3790 | { |
3791 | if (family >= NPROTO) | |
3792 | return -EINVAL; | |
3793 | gifconf_list[family] = gifconf; | |
3794 | return 0; | |
3795 | } | |
d1b19dff | 3796 | EXPORT_SYMBOL(register_gifconf); |
1da177e4 LT |
3797 | |
3798 | ||
3799 | /* | |
3800 | * Map an interface index to its name (SIOCGIFNAME) | |
3801 | */ | |
3802 | ||
3803 | /* | |
3804 | * We need this ioctl for efficient implementation of the | |
3805 | * if_indextoname() function required by the IPv6 API. Without | |
3806 | * it, we would have to search all the interfaces to find a | |
3807 | * match. --pb | |
3808 | */ | |
3809 | ||
881d966b | 3810 | static int dev_ifname(struct net *net, struct ifreq __user *arg) |
1da177e4 LT |
3811 | { |
3812 | struct net_device *dev; | |
3813 | struct ifreq ifr; | |
3814 | ||
3815 | /* | |
3816 | * Fetch the caller's info block. | |
3817 | */ | |
3818 | ||
3819 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
3820 | return -EFAULT; | |
3821 | ||
fb699dfd ED |
3822 | rcu_read_lock(); |
3823 | dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex); | |
1da177e4 | 3824 | if (!dev) { |
fb699dfd | 3825 | rcu_read_unlock(); |
1da177e4 LT |
3826 | return -ENODEV; |
3827 | } | |
3828 | ||
3829 | strcpy(ifr.ifr_name, dev->name); | |
fb699dfd | 3830 | rcu_read_unlock(); |
1da177e4 LT |
3831 | |
3832 | if (copy_to_user(arg, &ifr, sizeof(struct ifreq))) | |
3833 | return -EFAULT; | |
3834 | return 0; | |
3835 | } | |
3836 | ||
3837 | /* | |
3838 | * Perform a SIOCGIFCONF call. This structure will change | |
3839 | * size eventually, and there is nothing I can do about it. | |
3840 | * Thus we will need a 'compatibility mode'. | |
3841 | */ | |
3842 | ||
881d966b | 3843 | static int dev_ifconf(struct net *net, char __user *arg) |
1da177e4 LT |
3844 | { |
3845 | struct ifconf ifc; | |
3846 | struct net_device *dev; | |
3847 | char __user *pos; | |
3848 | int len; | |
3849 | int total; | |
3850 | int i; | |
3851 | ||
3852 | /* | |
3853 | * Fetch the caller's info block. | |
3854 | */ | |
3855 | ||
3856 | if (copy_from_user(&ifc, arg, sizeof(struct ifconf))) | |
3857 | return -EFAULT; | |
3858 | ||
3859 | pos = ifc.ifc_buf; | |
3860 | len = ifc.ifc_len; | |
3861 | ||
3862 | /* | |
3863 | * Loop over the interfaces, and write an info block for each. | |
3864 | */ | |
3865 | ||
3866 | total = 0; | |
881d966b | 3867 | for_each_netdev(net, dev) { |
1da177e4 LT |
3868 | for (i = 0; i < NPROTO; i++) { |
3869 | if (gifconf_list[i]) { | |
3870 | int done; | |
3871 | if (!pos) | |
3872 | done = gifconf_list[i](dev, NULL, 0); | |
3873 | else | |
3874 | done = gifconf_list[i](dev, pos + total, | |
3875 | len - total); | |
3876 | if (done < 0) | |
3877 | return -EFAULT; | |
3878 | total += done; | |
3879 | } | |
3880 | } | |
4ec93edb | 3881 | } |
1da177e4 LT |
3882 | |
3883 | /* | |
3884 | * All done. Write the updated control block back to the caller. | |
3885 | */ | |
3886 | ifc.ifc_len = total; | |
3887 | ||
3888 | /* | |
3889 | * Both BSD and Solaris return 0 here, so we do too. | |
3890 | */ | |
3891 | return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0; | |
3892 | } | |
3893 | ||
3894 | #ifdef CONFIG_PROC_FS | |
3895 | /* | |
3896 | * This is invoked by the /proc filesystem handler to display a device | |
3897 | * in detail. | |
3898 | */ | |
7562f876 | 3899 | void *dev_seq_start(struct seq_file *seq, loff_t *pos) |
c6d14c84 | 3900 | __acquires(RCU) |
1da177e4 | 3901 | { |
e372c414 | 3902 | struct net *net = seq_file_net(seq); |
7562f876 | 3903 | loff_t off; |
1da177e4 | 3904 | struct net_device *dev; |
1da177e4 | 3905 | |
c6d14c84 | 3906 | rcu_read_lock(); |
7562f876 PE |
3907 | if (!*pos) |
3908 | return SEQ_START_TOKEN; | |
1da177e4 | 3909 | |
7562f876 | 3910 | off = 1; |
c6d14c84 | 3911 | for_each_netdev_rcu(net, dev) |
7562f876 PE |
3912 | if (off++ == *pos) |
3913 | return dev; | |
1da177e4 | 3914 | |
7562f876 | 3915 | return NULL; |
1da177e4 LT |
3916 | } |
3917 | ||
3918 | void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
3919 | { | |
c6d14c84 ED |
3920 | struct net_device *dev = (v == SEQ_START_TOKEN) ? |
3921 | first_net_device(seq_file_net(seq)) : | |
3922 | next_net_device((struct net_device *)v); | |
3923 | ||
1da177e4 | 3924 | ++*pos; |
c6d14c84 | 3925 | return rcu_dereference(dev); |
1da177e4 LT |
3926 | } |
3927 | ||
3928 | void dev_seq_stop(struct seq_file *seq, void *v) | |
c6d14c84 | 3929 | __releases(RCU) |
1da177e4 | 3930 | { |
c6d14c84 | 3931 | rcu_read_unlock(); |
1da177e4 LT |
3932 | } |
3933 | ||
3934 | static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev) | |
3935 | { | |
28172739 ED |
3936 | struct rtnl_link_stats64 temp; |
3937 | const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp); | |
1da177e4 | 3938 | |
be1f3c2c BH |
3939 | seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu " |
3940 | "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n", | |
5a1b5898 RR |
3941 | dev->name, stats->rx_bytes, stats->rx_packets, |
3942 | stats->rx_errors, | |
3943 | stats->rx_dropped + stats->rx_missed_errors, | |
3944 | stats->rx_fifo_errors, | |
3945 | stats->rx_length_errors + stats->rx_over_errors + | |
3946 | stats->rx_crc_errors + stats->rx_frame_errors, | |
3947 | stats->rx_compressed, stats->multicast, | |
3948 | stats->tx_bytes, stats->tx_packets, | |
3949 | stats->tx_errors, stats->tx_dropped, | |
3950 | stats->tx_fifo_errors, stats->collisions, | |
3951 | stats->tx_carrier_errors + | |
3952 | stats->tx_aborted_errors + | |
3953 | stats->tx_window_errors + | |
3954 | stats->tx_heartbeat_errors, | |
3955 | stats->tx_compressed); | |
1da177e4 LT |
3956 | } |
3957 | ||
3958 | /* | |
3959 | * Called from the PROCfs module. This now uses the new arbitrary sized | |
3960 | * /proc/net interface to create /proc/net/dev | |
3961 | */ | |
3962 | static int dev_seq_show(struct seq_file *seq, void *v) | |
3963 | { | |
3964 | if (v == SEQ_START_TOKEN) | |
3965 | seq_puts(seq, "Inter-| Receive " | |
3966 | " | Transmit\n" | |
3967 | " face |bytes packets errs drop fifo frame " | |
3968 | "compressed multicast|bytes packets errs " | |
3969 | "drop fifo colls carrier compressed\n"); | |
3970 | else | |
3971 | dev_seq_printf_stats(seq, v); | |
3972 | return 0; | |
3973 | } | |
3974 | ||
dee42870 | 3975 | static struct softnet_data *softnet_get_online(loff_t *pos) |
1da177e4 | 3976 | { |
dee42870 | 3977 | struct softnet_data *sd = NULL; |
1da177e4 | 3978 | |
0c0b0aca | 3979 | while (*pos < nr_cpu_ids) |
4ec93edb | 3980 | if (cpu_online(*pos)) { |
dee42870 | 3981 | sd = &per_cpu(softnet_data, *pos); |
1da177e4 LT |
3982 | break; |
3983 | } else | |
3984 | ++*pos; | |
dee42870 | 3985 | return sd; |
1da177e4 LT |
3986 | } |
3987 | ||
3988 | static void *softnet_seq_start(struct seq_file *seq, loff_t *pos) | |
3989 | { | |
3990 | return softnet_get_online(pos); | |
3991 | } | |
3992 | ||
3993 | static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
3994 | { | |
3995 | ++*pos; | |
3996 | return softnet_get_online(pos); | |
3997 | } | |
3998 | ||
3999 | static void softnet_seq_stop(struct seq_file *seq, void *v) | |
4000 | { | |
4001 | } | |
4002 | ||
4003 | static int softnet_seq_show(struct seq_file *seq, void *v) | |
4004 | { | |
dee42870 | 4005 | struct softnet_data *sd = v; |
1da177e4 | 4006 | |
0a9627f2 | 4007 | seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n", |
dee42870 | 4008 | sd->processed, sd->dropped, sd->time_squeeze, 0, |
c1ebcdb8 | 4009 | 0, 0, 0, 0, /* was fastroute */ |
dee42870 | 4010 | sd->cpu_collision, sd->received_rps); |
1da177e4 LT |
4011 | return 0; |
4012 | } | |
4013 | ||
f690808e | 4014 | static const struct seq_operations dev_seq_ops = { |
1da177e4 LT |
4015 | .start = dev_seq_start, |
4016 | .next = dev_seq_next, | |
4017 | .stop = dev_seq_stop, | |
4018 | .show = dev_seq_show, | |
4019 | }; | |
4020 | ||
4021 | static int dev_seq_open(struct inode *inode, struct file *file) | |
4022 | { | |
e372c414 DL |
4023 | return seq_open_net(inode, file, &dev_seq_ops, |
4024 | sizeof(struct seq_net_private)); | |
1da177e4 LT |
4025 | } |
4026 | ||
9a32144e | 4027 | static const struct file_operations dev_seq_fops = { |
1da177e4 LT |
4028 | .owner = THIS_MODULE, |
4029 | .open = dev_seq_open, | |
4030 | .read = seq_read, | |
4031 | .llseek = seq_lseek, | |
e372c414 | 4032 | .release = seq_release_net, |
1da177e4 LT |
4033 | }; |
4034 | ||
f690808e | 4035 | static const struct seq_operations softnet_seq_ops = { |
1da177e4 LT |
4036 | .start = softnet_seq_start, |
4037 | .next = softnet_seq_next, | |
4038 | .stop = softnet_seq_stop, | |
4039 | .show = softnet_seq_show, | |
4040 | }; | |
4041 | ||
4042 | static int softnet_seq_open(struct inode *inode, struct file *file) | |
4043 | { | |
4044 | return seq_open(file, &softnet_seq_ops); | |
4045 | } | |
4046 | ||
9a32144e | 4047 | static const struct file_operations softnet_seq_fops = { |
1da177e4 LT |
4048 | .owner = THIS_MODULE, |
4049 | .open = softnet_seq_open, | |
4050 | .read = seq_read, | |
4051 | .llseek = seq_lseek, | |
4052 | .release = seq_release, | |
4053 | }; | |
4054 | ||
0e1256ff SH |
4055 | static void *ptype_get_idx(loff_t pos) |
4056 | { | |
4057 | struct packet_type *pt = NULL; | |
4058 | loff_t i = 0; | |
4059 | int t; | |
4060 | ||
4061 | list_for_each_entry_rcu(pt, &ptype_all, list) { | |
4062 | if (i == pos) | |
4063 | return pt; | |
4064 | ++i; | |
4065 | } | |
4066 | ||
82d8a867 | 4067 | for (t = 0; t < PTYPE_HASH_SIZE; t++) { |
0e1256ff SH |
4068 | list_for_each_entry_rcu(pt, &ptype_base[t], list) { |
4069 | if (i == pos) | |
4070 | return pt; | |
4071 | ++i; | |
4072 | } | |
4073 | } | |
4074 | return NULL; | |
4075 | } | |
4076 | ||
4077 | static void *ptype_seq_start(struct seq_file *seq, loff_t *pos) | |
72348a42 | 4078 | __acquires(RCU) |
0e1256ff SH |
4079 | { |
4080 | rcu_read_lock(); | |
4081 | return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN; | |
4082 | } | |
4083 | ||
4084 | static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
4085 | { | |
4086 | struct packet_type *pt; | |
4087 | struct list_head *nxt; | |
4088 | int hash; | |
4089 | ||
4090 | ++*pos; | |
4091 | if (v == SEQ_START_TOKEN) | |
4092 | return ptype_get_idx(0); | |
4093 | ||
4094 | pt = v; | |
4095 | nxt = pt->list.next; | |
4096 | if (pt->type == htons(ETH_P_ALL)) { | |
4097 | if (nxt != &ptype_all) | |
4098 | goto found; | |
4099 | hash = 0; | |
4100 | nxt = ptype_base[0].next; | |
4101 | } else | |
82d8a867 | 4102 | hash = ntohs(pt->type) & PTYPE_HASH_MASK; |
0e1256ff SH |
4103 | |
4104 | while (nxt == &ptype_base[hash]) { | |
82d8a867 | 4105 | if (++hash >= PTYPE_HASH_SIZE) |
0e1256ff SH |
4106 | return NULL; |
4107 | nxt = ptype_base[hash].next; | |
4108 | } | |
4109 | found: | |
4110 | return list_entry(nxt, struct packet_type, list); | |
4111 | } | |
4112 | ||
4113 | static void ptype_seq_stop(struct seq_file *seq, void *v) | |
72348a42 | 4114 | __releases(RCU) |
0e1256ff SH |
4115 | { |
4116 | rcu_read_unlock(); | |
4117 | } | |
4118 | ||
0e1256ff SH |
4119 | static int ptype_seq_show(struct seq_file *seq, void *v) |
4120 | { | |
4121 | struct packet_type *pt = v; | |
4122 | ||
4123 | if (v == SEQ_START_TOKEN) | |
4124 | seq_puts(seq, "Type Device Function\n"); | |
c346dca1 | 4125 | else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) { |
0e1256ff SH |
4126 | if (pt->type == htons(ETH_P_ALL)) |
4127 | seq_puts(seq, "ALL "); | |
4128 | else | |
4129 | seq_printf(seq, "%04x", ntohs(pt->type)); | |
4130 | ||
908cd2da AD |
4131 | seq_printf(seq, " %-8s %pF\n", |
4132 | pt->dev ? pt->dev->name : "", pt->func); | |
0e1256ff SH |
4133 | } |
4134 | ||
4135 | return 0; | |
4136 | } | |
4137 | ||
4138 | static const struct seq_operations ptype_seq_ops = { | |
4139 | .start = ptype_seq_start, | |
4140 | .next = ptype_seq_next, | |
4141 | .stop = ptype_seq_stop, | |
4142 | .show = ptype_seq_show, | |
4143 | }; | |
4144 | ||
4145 | static int ptype_seq_open(struct inode *inode, struct file *file) | |
4146 | { | |
2feb27db PE |
4147 | return seq_open_net(inode, file, &ptype_seq_ops, |
4148 | sizeof(struct seq_net_private)); | |
0e1256ff SH |
4149 | } |
4150 | ||
4151 | static const struct file_operations ptype_seq_fops = { | |
4152 | .owner = THIS_MODULE, | |
4153 | .open = ptype_seq_open, | |
4154 | .read = seq_read, | |
4155 | .llseek = seq_lseek, | |
2feb27db | 4156 | .release = seq_release_net, |
0e1256ff SH |
4157 | }; |
4158 | ||
4159 | ||
4665079c | 4160 | static int __net_init dev_proc_net_init(struct net *net) |
1da177e4 LT |
4161 | { |
4162 | int rc = -ENOMEM; | |
4163 | ||
881d966b | 4164 | if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops)) |
1da177e4 | 4165 | goto out; |
881d966b | 4166 | if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops)) |
1da177e4 | 4167 | goto out_dev; |
881d966b | 4168 | if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops)) |
457c4cbc | 4169 | goto out_softnet; |
0e1256ff | 4170 | |
881d966b | 4171 | if (wext_proc_init(net)) |
457c4cbc | 4172 | goto out_ptype; |
1da177e4 LT |
4173 | rc = 0; |
4174 | out: | |
4175 | return rc; | |
457c4cbc | 4176 | out_ptype: |
881d966b | 4177 | proc_net_remove(net, "ptype"); |
1da177e4 | 4178 | out_softnet: |
881d966b | 4179 | proc_net_remove(net, "softnet_stat"); |
1da177e4 | 4180 | out_dev: |
881d966b | 4181 | proc_net_remove(net, "dev"); |
1da177e4 LT |
4182 | goto out; |
4183 | } | |
881d966b | 4184 | |
4665079c | 4185 | static void __net_exit dev_proc_net_exit(struct net *net) |
881d966b EB |
4186 | { |
4187 | wext_proc_exit(net); | |
4188 | ||
4189 | proc_net_remove(net, "ptype"); | |
4190 | proc_net_remove(net, "softnet_stat"); | |
4191 | proc_net_remove(net, "dev"); | |
4192 | } | |
4193 | ||
022cbae6 | 4194 | static struct pernet_operations __net_initdata dev_proc_ops = { |
881d966b EB |
4195 | .init = dev_proc_net_init, |
4196 | .exit = dev_proc_net_exit, | |
4197 | }; | |
4198 | ||
4199 | static int __init dev_proc_init(void) | |
4200 | { | |
4201 | return register_pernet_subsys(&dev_proc_ops); | |
4202 | } | |
1da177e4 LT |
4203 | #else |
4204 | #define dev_proc_init() 0 | |
4205 | #endif /* CONFIG_PROC_FS */ | |
4206 | ||
4207 | ||
4208 | /** | |
4209 | * netdev_set_master - set up master/slave pair | |
4210 | * @slave: slave device | |
4211 | * @master: new master device | |
4212 | * | |
4213 | * Changes the master device of the slave. Pass %NULL to break the | |
4214 | * bonding. The caller must hold the RTNL semaphore. On a failure | |
4215 | * a negative errno code is returned. On success the reference counts | |
4216 | * are adjusted, %RTM_NEWLINK is sent to the routing socket and the | |
4217 | * function returns zero. | |
4218 | */ | |
4219 | int netdev_set_master(struct net_device *slave, struct net_device *master) | |
4220 | { | |
4221 | struct net_device *old = slave->master; | |
4222 | ||
4223 | ASSERT_RTNL(); | |
4224 | ||
4225 | if (master) { | |
4226 | if (old) | |
4227 | return -EBUSY; | |
4228 | dev_hold(master); | |
4229 | } | |
4230 | ||
4231 | slave->master = master; | |
4ec93edb | 4232 | |
283f2fe8 ED |
4233 | if (old) { |
4234 | synchronize_net(); | |
1da177e4 | 4235 | dev_put(old); |
283f2fe8 | 4236 | } |
1da177e4 LT |
4237 | if (master) |
4238 | slave->flags |= IFF_SLAVE; | |
4239 | else | |
4240 | slave->flags &= ~IFF_SLAVE; | |
4241 | ||
4242 | rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE); | |
4243 | return 0; | |
4244 | } | |
d1b19dff | 4245 | EXPORT_SYMBOL(netdev_set_master); |
1da177e4 | 4246 | |
b6c40d68 PM |
4247 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
4248 | { | |
d314774c SH |
4249 | const struct net_device_ops *ops = dev->netdev_ops; |
4250 | ||
4251 | if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags) | |
4252 | ops->ndo_change_rx_flags(dev, flags); | |
b6c40d68 PM |
4253 | } |
4254 | ||
dad9b335 | 4255 | static int __dev_set_promiscuity(struct net_device *dev, int inc) |
1da177e4 LT |
4256 | { |
4257 | unsigned short old_flags = dev->flags; | |
8192b0c4 DH |
4258 | uid_t uid; |
4259 | gid_t gid; | |
1da177e4 | 4260 | |
24023451 PM |
4261 | ASSERT_RTNL(); |
4262 | ||
dad9b335 WC |
4263 | dev->flags |= IFF_PROMISC; |
4264 | dev->promiscuity += inc; | |
4265 | if (dev->promiscuity == 0) { | |
4266 | /* | |
4267 | * Avoid overflow. | |
4268 | * If inc causes overflow, untouch promisc and return error. | |
4269 | */ | |
4270 | if (inc < 0) | |
4271 | dev->flags &= ~IFF_PROMISC; | |
4272 | else { | |
4273 | dev->promiscuity -= inc; | |
4274 | printk(KERN_WARNING "%s: promiscuity touches roof, " | |
4275 | "set promiscuity failed, promiscuity feature " | |
4276 | "of device might be broken.\n", dev->name); | |
4277 | return -EOVERFLOW; | |
4278 | } | |
4279 | } | |
52609c0b | 4280 | if (dev->flags != old_flags) { |
1da177e4 LT |
4281 | printk(KERN_INFO "device %s %s promiscuous mode\n", |
4282 | dev->name, (dev->flags & IFF_PROMISC) ? "entered" : | |
4ec93edb | 4283 | "left"); |
8192b0c4 DH |
4284 | if (audit_enabled) { |
4285 | current_uid_gid(&uid, &gid); | |
7759db82 KHK |
4286 | audit_log(current->audit_context, GFP_ATOMIC, |
4287 | AUDIT_ANOM_PROMISCUOUS, | |
4288 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
4289 | dev->name, (dev->flags & IFF_PROMISC), | |
4290 | (old_flags & IFF_PROMISC), | |
4291 | audit_get_loginuid(current), | |
8192b0c4 | 4292 | uid, gid, |
7759db82 | 4293 | audit_get_sessionid(current)); |
8192b0c4 | 4294 | } |
24023451 | 4295 | |
b6c40d68 | 4296 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 4297 | } |
dad9b335 | 4298 | return 0; |
1da177e4 LT |
4299 | } |
4300 | ||
4417da66 PM |
4301 | /** |
4302 | * dev_set_promiscuity - update promiscuity count on a device | |
4303 | * @dev: device | |
4304 | * @inc: modifier | |
4305 | * | |
4306 | * Add or remove promiscuity from a device. While the count in the device | |
4307 | * remains above zero the interface remains promiscuous. Once it hits zero | |
4308 | * the device reverts back to normal filtering operation. A negative inc | |
4309 | * value is used to drop promiscuity on the device. | |
dad9b335 | 4310 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 4311 | */ |
dad9b335 | 4312 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 PM |
4313 | { |
4314 | unsigned short old_flags = dev->flags; | |
dad9b335 | 4315 | int err; |
4417da66 | 4316 | |
dad9b335 | 4317 | err = __dev_set_promiscuity(dev, inc); |
4b5a698e | 4318 | if (err < 0) |
dad9b335 | 4319 | return err; |
4417da66 PM |
4320 | if (dev->flags != old_flags) |
4321 | dev_set_rx_mode(dev); | |
dad9b335 | 4322 | return err; |
4417da66 | 4323 | } |
d1b19dff | 4324 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 4325 | |
1da177e4 LT |
4326 | /** |
4327 | * dev_set_allmulti - update allmulti count on a device | |
4328 | * @dev: device | |
4329 | * @inc: modifier | |
4330 | * | |
4331 | * Add or remove reception of all multicast frames to a device. While the | |
4332 | * count in the device remains above zero the interface remains listening | |
4333 | * to all interfaces. Once it hits zero the device reverts back to normal | |
4334 | * filtering operation. A negative @inc value is used to drop the counter | |
4335 | * when releasing a resource needing all multicasts. | |
dad9b335 | 4336 | * Return 0 if successful or a negative errno code on error. |
1da177e4 LT |
4337 | */ |
4338 | ||
dad9b335 | 4339 | int dev_set_allmulti(struct net_device *dev, int inc) |
1da177e4 LT |
4340 | { |
4341 | unsigned short old_flags = dev->flags; | |
4342 | ||
24023451 PM |
4343 | ASSERT_RTNL(); |
4344 | ||
1da177e4 | 4345 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
4346 | dev->allmulti += inc; |
4347 | if (dev->allmulti == 0) { | |
4348 | /* | |
4349 | * Avoid overflow. | |
4350 | * If inc causes overflow, untouch allmulti and return error. | |
4351 | */ | |
4352 | if (inc < 0) | |
4353 | dev->flags &= ~IFF_ALLMULTI; | |
4354 | else { | |
4355 | dev->allmulti -= inc; | |
4356 | printk(KERN_WARNING "%s: allmulti touches roof, " | |
4357 | "set allmulti failed, allmulti feature of " | |
4358 | "device might be broken.\n", dev->name); | |
4359 | return -EOVERFLOW; | |
4360 | } | |
4361 | } | |
24023451 | 4362 | if (dev->flags ^ old_flags) { |
b6c40d68 | 4363 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 4364 | dev_set_rx_mode(dev); |
24023451 | 4365 | } |
dad9b335 | 4366 | return 0; |
4417da66 | 4367 | } |
d1b19dff | 4368 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
4369 | |
4370 | /* | |
4371 | * Upload unicast and multicast address lists to device and | |
4372 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 4373 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
4374 | * are present. |
4375 | */ | |
4376 | void __dev_set_rx_mode(struct net_device *dev) | |
4377 | { | |
d314774c SH |
4378 | const struct net_device_ops *ops = dev->netdev_ops; |
4379 | ||
4417da66 PM |
4380 | /* dev_open will call this function so the list will stay sane. */ |
4381 | if (!(dev->flags&IFF_UP)) | |
4382 | return; | |
4383 | ||
4384 | if (!netif_device_present(dev)) | |
40b77c94 | 4385 | return; |
4417da66 | 4386 | |
d314774c SH |
4387 | if (ops->ndo_set_rx_mode) |
4388 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
4389 | else { |
4390 | /* Unicast addresses changes may only happen under the rtnl, | |
4391 | * therefore calling __dev_set_promiscuity here is safe. | |
4392 | */ | |
32e7bfc4 | 4393 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
4417da66 PM |
4394 | __dev_set_promiscuity(dev, 1); |
4395 | dev->uc_promisc = 1; | |
32e7bfc4 | 4396 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
4417da66 PM |
4397 | __dev_set_promiscuity(dev, -1); |
4398 | dev->uc_promisc = 0; | |
4399 | } | |
4400 | ||
d314774c SH |
4401 | if (ops->ndo_set_multicast_list) |
4402 | ops->ndo_set_multicast_list(dev); | |
4417da66 PM |
4403 | } |
4404 | } | |
4405 | ||
4406 | void dev_set_rx_mode(struct net_device *dev) | |
4407 | { | |
b9e40857 | 4408 | netif_addr_lock_bh(dev); |
4417da66 | 4409 | __dev_set_rx_mode(dev); |
b9e40857 | 4410 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
4411 | } |
4412 | ||
f0db275a SH |
4413 | /** |
4414 | * dev_get_flags - get flags reported to userspace | |
4415 | * @dev: device | |
4416 | * | |
4417 | * Get the combination of flag bits exported through APIs to userspace. | |
4418 | */ | |
1da177e4 LT |
4419 | unsigned dev_get_flags(const struct net_device *dev) |
4420 | { | |
4421 | unsigned flags; | |
4422 | ||
4423 | flags = (dev->flags & ~(IFF_PROMISC | | |
4424 | IFF_ALLMULTI | | |
b00055aa SR |
4425 | IFF_RUNNING | |
4426 | IFF_LOWER_UP | | |
4427 | IFF_DORMANT)) | | |
1da177e4 LT |
4428 | (dev->gflags & (IFF_PROMISC | |
4429 | IFF_ALLMULTI)); | |
4430 | ||
b00055aa SR |
4431 | if (netif_running(dev)) { |
4432 | if (netif_oper_up(dev)) | |
4433 | flags |= IFF_RUNNING; | |
4434 | if (netif_carrier_ok(dev)) | |
4435 | flags |= IFF_LOWER_UP; | |
4436 | if (netif_dormant(dev)) | |
4437 | flags |= IFF_DORMANT; | |
4438 | } | |
1da177e4 LT |
4439 | |
4440 | return flags; | |
4441 | } | |
d1b19dff | 4442 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 4443 | |
bd380811 | 4444 | int __dev_change_flags(struct net_device *dev, unsigned int flags) |
1da177e4 | 4445 | { |
1da177e4 | 4446 | int old_flags = dev->flags; |
bd380811 | 4447 | int ret; |
1da177e4 | 4448 | |
24023451 PM |
4449 | ASSERT_RTNL(); |
4450 | ||
1da177e4 LT |
4451 | /* |
4452 | * Set the flags on our device. | |
4453 | */ | |
4454 | ||
4455 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
4456 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
4457 | IFF_AUTOMEDIA)) | | |
4458 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
4459 | IFF_ALLMULTI)); | |
4460 | ||
4461 | /* | |
4462 | * Load in the correct multicast list now the flags have changed. | |
4463 | */ | |
4464 | ||
b6c40d68 PM |
4465 | if ((old_flags ^ flags) & IFF_MULTICAST) |
4466 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 4467 | |
4417da66 | 4468 | dev_set_rx_mode(dev); |
1da177e4 LT |
4469 | |
4470 | /* | |
4471 | * Have we downed the interface. We handle IFF_UP ourselves | |
4472 | * according to user attempts to set it, rather than blindly | |
4473 | * setting it. | |
4474 | */ | |
4475 | ||
4476 | ret = 0; | |
4477 | if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */ | |
bd380811 | 4478 | ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev); |
1da177e4 LT |
4479 | |
4480 | if (!ret) | |
4417da66 | 4481 | dev_set_rx_mode(dev); |
1da177e4 LT |
4482 | } |
4483 | ||
1da177e4 | 4484 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff ED |
4485 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
4486 | ||
1da177e4 LT |
4487 | dev->gflags ^= IFF_PROMISC; |
4488 | dev_set_promiscuity(dev, inc); | |
4489 | } | |
4490 | ||
4491 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
4492 | is important. Some (broken) drivers set IFF_PROMISC, when | |
4493 | IFF_ALLMULTI is requested not asking us and not reporting. | |
4494 | */ | |
4495 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
4496 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
4497 | ||
1da177e4 LT |
4498 | dev->gflags ^= IFF_ALLMULTI; |
4499 | dev_set_allmulti(dev, inc); | |
4500 | } | |
4501 | ||
bd380811 PM |
4502 | return ret; |
4503 | } | |
4504 | ||
4505 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags) | |
4506 | { | |
4507 | unsigned int changes = dev->flags ^ old_flags; | |
4508 | ||
4509 | if (changes & IFF_UP) { | |
4510 | if (dev->flags & IFF_UP) | |
4511 | call_netdevice_notifiers(NETDEV_UP, dev); | |
4512 | else | |
4513 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
4514 | } | |
4515 | ||
4516 | if (dev->flags & IFF_UP && | |
4517 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) | |
4518 | call_netdevice_notifiers(NETDEV_CHANGE, dev); | |
4519 | } | |
4520 | ||
4521 | /** | |
4522 | * dev_change_flags - change device settings | |
4523 | * @dev: device | |
4524 | * @flags: device state flags | |
4525 | * | |
4526 | * Change settings on device based state flags. The flags are | |
4527 | * in the userspace exported format. | |
4528 | */ | |
4529 | int dev_change_flags(struct net_device *dev, unsigned flags) | |
4530 | { | |
4531 | int ret, changes; | |
4532 | int old_flags = dev->flags; | |
4533 | ||
4534 | ret = __dev_change_flags(dev, flags); | |
4535 | if (ret < 0) | |
4536 | return ret; | |
4537 | ||
4538 | changes = old_flags ^ dev->flags; | |
7c355f53 TG |
4539 | if (changes) |
4540 | rtmsg_ifinfo(RTM_NEWLINK, dev, changes); | |
1da177e4 | 4541 | |
bd380811 | 4542 | __dev_notify_flags(dev, old_flags); |
1da177e4 LT |
4543 | return ret; |
4544 | } | |
d1b19dff | 4545 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 4546 | |
f0db275a SH |
4547 | /** |
4548 | * dev_set_mtu - Change maximum transfer unit | |
4549 | * @dev: device | |
4550 | * @new_mtu: new transfer unit | |
4551 | * | |
4552 | * Change the maximum transfer size of the network device. | |
4553 | */ | |
1da177e4 LT |
4554 | int dev_set_mtu(struct net_device *dev, int new_mtu) |
4555 | { | |
d314774c | 4556 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
4557 | int err; |
4558 | ||
4559 | if (new_mtu == dev->mtu) | |
4560 | return 0; | |
4561 | ||
4562 | /* MTU must be positive. */ | |
4563 | if (new_mtu < 0) | |
4564 | return -EINVAL; | |
4565 | ||
4566 | if (!netif_device_present(dev)) | |
4567 | return -ENODEV; | |
4568 | ||
4569 | err = 0; | |
d314774c SH |
4570 | if (ops->ndo_change_mtu) |
4571 | err = ops->ndo_change_mtu(dev, new_mtu); | |
1da177e4 LT |
4572 | else |
4573 | dev->mtu = new_mtu; | |
d314774c | 4574 | |
1da177e4 | 4575 | if (!err && dev->flags & IFF_UP) |
056925ab | 4576 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); |
1da177e4 LT |
4577 | return err; |
4578 | } | |
d1b19dff | 4579 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 4580 | |
f0db275a SH |
4581 | /** |
4582 | * dev_set_mac_address - Change Media Access Control Address | |
4583 | * @dev: device | |
4584 | * @sa: new address | |
4585 | * | |
4586 | * Change the hardware (MAC) address of the device | |
4587 | */ | |
1da177e4 LT |
4588 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) |
4589 | { | |
d314774c | 4590 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
4591 | int err; |
4592 | ||
d314774c | 4593 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
4594 | return -EOPNOTSUPP; |
4595 | if (sa->sa_family != dev->type) | |
4596 | return -EINVAL; | |
4597 | if (!netif_device_present(dev)) | |
4598 | return -ENODEV; | |
d314774c | 4599 | err = ops->ndo_set_mac_address(dev, sa); |
1da177e4 | 4600 | if (!err) |
056925ab | 4601 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
1da177e4 LT |
4602 | return err; |
4603 | } | |
d1b19dff | 4604 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 LT |
4605 | |
4606 | /* | |
3710becf | 4607 | * Perform the SIOCxIFxxx calls, inside rcu_read_lock() |
1da177e4 | 4608 | */ |
14e3e079 | 4609 | static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd) |
1da177e4 LT |
4610 | { |
4611 | int err; | |
3710becf | 4612 | struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name); |
1da177e4 LT |
4613 | |
4614 | if (!dev) | |
4615 | return -ENODEV; | |
4616 | ||
4617 | switch (cmd) { | |
d1b19dff ED |
4618 | case SIOCGIFFLAGS: /* Get interface flags */ |
4619 | ifr->ifr_flags = (short) dev_get_flags(dev); | |
4620 | return 0; | |
1da177e4 | 4621 | |
d1b19dff ED |
4622 | case SIOCGIFMETRIC: /* Get the metric on the interface |
4623 | (currently unused) */ | |
4624 | ifr->ifr_metric = 0; | |
4625 | return 0; | |
1da177e4 | 4626 | |
d1b19dff ED |
4627 | case SIOCGIFMTU: /* Get the MTU of a device */ |
4628 | ifr->ifr_mtu = dev->mtu; | |
4629 | return 0; | |
1da177e4 | 4630 | |
d1b19dff ED |
4631 | case SIOCGIFHWADDR: |
4632 | if (!dev->addr_len) | |
4633 | memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data); | |
4634 | else | |
4635 | memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr, | |
4636 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
4637 | ifr->ifr_hwaddr.sa_family = dev->type; | |
4638 | return 0; | |
1da177e4 | 4639 | |
d1b19dff ED |
4640 | case SIOCGIFSLAVE: |
4641 | err = -EINVAL; | |
4642 | break; | |
14e3e079 | 4643 | |
d1b19dff ED |
4644 | case SIOCGIFMAP: |
4645 | ifr->ifr_map.mem_start = dev->mem_start; | |
4646 | ifr->ifr_map.mem_end = dev->mem_end; | |
4647 | ifr->ifr_map.base_addr = dev->base_addr; | |
4648 | ifr->ifr_map.irq = dev->irq; | |
4649 | ifr->ifr_map.dma = dev->dma; | |
4650 | ifr->ifr_map.port = dev->if_port; | |
4651 | return 0; | |
14e3e079 | 4652 | |
d1b19dff ED |
4653 | case SIOCGIFINDEX: |
4654 | ifr->ifr_ifindex = dev->ifindex; | |
4655 | return 0; | |
14e3e079 | 4656 | |
d1b19dff ED |
4657 | case SIOCGIFTXQLEN: |
4658 | ifr->ifr_qlen = dev->tx_queue_len; | |
4659 | return 0; | |
14e3e079 | 4660 | |
d1b19dff ED |
4661 | default: |
4662 | /* dev_ioctl() should ensure this case | |
4663 | * is never reached | |
4664 | */ | |
4665 | WARN_ON(1); | |
4666 | err = -EINVAL; | |
4667 | break; | |
14e3e079 JG |
4668 | |
4669 | } | |
4670 | return err; | |
4671 | } | |
4672 | ||
4673 | /* | |
4674 | * Perform the SIOCxIFxxx calls, inside rtnl_lock() | |
4675 | */ | |
4676 | static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd) | |
4677 | { | |
4678 | int err; | |
4679 | struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name); | |
5f2f6da7 | 4680 | const struct net_device_ops *ops; |
14e3e079 JG |
4681 | |
4682 | if (!dev) | |
4683 | return -ENODEV; | |
4684 | ||
5f2f6da7 JP |
4685 | ops = dev->netdev_ops; |
4686 | ||
14e3e079 | 4687 | switch (cmd) { |
d1b19dff ED |
4688 | case SIOCSIFFLAGS: /* Set interface flags */ |
4689 | return dev_change_flags(dev, ifr->ifr_flags); | |
14e3e079 | 4690 | |
d1b19dff ED |
4691 | case SIOCSIFMETRIC: /* Set the metric on the interface |
4692 | (currently unused) */ | |
4693 | return -EOPNOTSUPP; | |
14e3e079 | 4694 | |
d1b19dff ED |
4695 | case SIOCSIFMTU: /* Set the MTU of a device */ |
4696 | return dev_set_mtu(dev, ifr->ifr_mtu); | |
1da177e4 | 4697 | |
d1b19dff ED |
4698 | case SIOCSIFHWADDR: |
4699 | return dev_set_mac_address(dev, &ifr->ifr_hwaddr); | |
1da177e4 | 4700 | |
d1b19dff ED |
4701 | case SIOCSIFHWBROADCAST: |
4702 | if (ifr->ifr_hwaddr.sa_family != dev->type) | |
4703 | return -EINVAL; | |
4704 | memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data, | |
4705 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
4706 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); | |
4707 | return 0; | |
1da177e4 | 4708 | |
d1b19dff ED |
4709 | case SIOCSIFMAP: |
4710 | if (ops->ndo_set_config) { | |
1da177e4 LT |
4711 | if (!netif_device_present(dev)) |
4712 | return -ENODEV; | |
d1b19dff ED |
4713 | return ops->ndo_set_config(dev, &ifr->ifr_map); |
4714 | } | |
4715 | return -EOPNOTSUPP; | |
1da177e4 | 4716 | |
d1b19dff ED |
4717 | case SIOCADDMULTI: |
4718 | if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) || | |
4719 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) | |
4720 | return -EINVAL; | |
4721 | if (!netif_device_present(dev)) | |
4722 | return -ENODEV; | |
22bedad3 | 4723 | return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data); |
d1b19dff ED |
4724 | |
4725 | case SIOCDELMULTI: | |
4726 | if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) || | |
4727 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) | |
4728 | return -EINVAL; | |
4729 | if (!netif_device_present(dev)) | |
4730 | return -ENODEV; | |
22bedad3 | 4731 | return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data); |
1da177e4 | 4732 | |
d1b19dff ED |
4733 | case SIOCSIFTXQLEN: |
4734 | if (ifr->ifr_qlen < 0) | |
4735 | return -EINVAL; | |
4736 | dev->tx_queue_len = ifr->ifr_qlen; | |
4737 | return 0; | |
1da177e4 | 4738 | |
d1b19dff ED |
4739 | case SIOCSIFNAME: |
4740 | ifr->ifr_newname[IFNAMSIZ-1] = '\0'; | |
4741 | return dev_change_name(dev, ifr->ifr_newname); | |
1da177e4 | 4742 | |
d1b19dff ED |
4743 | /* |
4744 | * Unknown or private ioctl | |
4745 | */ | |
4746 | default: | |
4747 | if ((cmd >= SIOCDEVPRIVATE && | |
4748 | cmd <= SIOCDEVPRIVATE + 15) || | |
4749 | cmd == SIOCBONDENSLAVE || | |
4750 | cmd == SIOCBONDRELEASE || | |
4751 | cmd == SIOCBONDSETHWADDR || | |
4752 | cmd == SIOCBONDSLAVEINFOQUERY || | |
4753 | cmd == SIOCBONDINFOQUERY || | |
4754 | cmd == SIOCBONDCHANGEACTIVE || | |
4755 | cmd == SIOCGMIIPHY || | |
4756 | cmd == SIOCGMIIREG || | |
4757 | cmd == SIOCSMIIREG || | |
4758 | cmd == SIOCBRADDIF || | |
4759 | cmd == SIOCBRDELIF || | |
4760 | cmd == SIOCSHWTSTAMP || | |
4761 | cmd == SIOCWANDEV) { | |
4762 | err = -EOPNOTSUPP; | |
4763 | if (ops->ndo_do_ioctl) { | |
4764 | if (netif_device_present(dev)) | |
4765 | err = ops->ndo_do_ioctl(dev, ifr, cmd); | |
4766 | else | |
4767 | err = -ENODEV; | |
4768 | } | |
4769 | } else | |
4770 | err = -EINVAL; | |
1da177e4 LT |
4771 | |
4772 | } | |
4773 | return err; | |
4774 | } | |
4775 | ||
4776 | /* | |
4777 | * This function handles all "interface"-type I/O control requests. The actual | |
4778 | * 'doing' part of this is dev_ifsioc above. | |
4779 | */ | |
4780 | ||
4781 | /** | |
4782 | * dev_ioctl - network device ioctl | |
c4ea43c5 | 4783 | * @net: the applicable net namespace |
1da177e4 LT |
4784 | * @cmd: command to issue |
4785 | * @arg: pointer to a struct ifreq in user space | |
4786 | * | |
4787 | * Issue ioctl functions to devices. This is normally called by the | |
4788 | * user space syscall interfaces but can sometimes be useful for | |
4789 | * other purposes. The return value is the return from the syscall if | |
4790 | * positive or a negative errno code on error. | |
4791 | */ | |
4792 | ||
881d966b | 4793 | int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg) |
1da177e4 LT |
4794 | { |
4795 | struct ifreq ifr; | |
4796 | int ret; | |
4797 | char *colon; | |
4798 | ||
4799 | /* One special case: SIOCGIFCONF takes ifconf argument | |
4800 | and requires shared lock, because it sleeps writing | |
4801 | to user space. | |
4802 | */ | |
4803 | ||
4804 | if (cmd == SIOCGIFCONF) { | |
6756ae4b | 4805 | rtnl_lock(); |
881d966b | 4806 | ret = dev_ifconf(net, (char __user *) arg); |
6756ae4b | 4807 | rtnl_unlock(); |
1da177e4 LT |
4808 | return ret; |
4809 | } | |
4810 | if (cmd == SIOCGIFNAME) | |
881d966b | 4811 | return dev_ifname(net, (struct ifreq __user *)arg); |
1da177e4 LT |
4812 | |
4813 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
4814 | return -EFAULT; | |
4815 | ||
4816 | ifr.ifr_name[IFNAMSIZ-1] = 0; | |
4817 | ||
4818 | colon = strchr(ifr.ifr_name, ':'); | |
4819 | if (colon) | |
4820 | *colon = 0; | |
4821 | ||
4822 | /* | |
4823 | * See which interface the caller is talking about. | |
4824 | */ | |
4825 | ||
4826 | switch (cmd) { | |
d1b19dff ED |
4827 | /* |
4828 | * These ioctl calls: | |
4829 | * - can be done by all. | |
4830 | * - atomic and do not require locking. | |
4831 | * - return a value | |
4832 | */ | |
4833 | case SIOCGIFFLAGS: | |
4834 | case SIOCGIFMETRIC: | |
4835 | case SIOCGIFMTU: | |
4836 | case SIOCGIFHWADDR: | |
4837 | case SIOCGIFSLAVE: | |
4838 | case SIOCGIFMAP: | |
4839 | case SIOCGIFINDEX: | |
4840 | case SIOCGIFTXQLEN: | |
4841 | dev_load(net, ifr.ifr_name); | |
3710becf | 4842 | rcu_read_lock(); |
d1b19dff | 4843 | ret = dev_ifsioc_locked(net, &ifr, cmd); |
3710becf | 4844 | rcu_read_unlock(); |
d1b19dff ED |
4845 | if (!ret) { |
4846 | if (colon) | |
4847 | *colon = ':'; | |
4848 | if (copy_to_user(arg, &ifr, | |
4849 | sizeof(struct ifreq))) | |
4850 | ret = -EFAULT; | |
4851 | } | |
4852 | return ret; | |
1da177e4 | 4853 | |
d1b19dff ED |
4854 | case SIOCETHTOOL: |
4855 | dev_load(net, ifr.ifr_name); | |
4856 | rtnl_lock(); | |
4857 | ret = dev_ethtool(net, &ifr); | |
4858 | rtnl_unlock(); | |
4859 | if (!ret) { | |
4860 | if (colon) | |
4861 | *colon = ':'; | |
4862 | if (copy_to_user(arg, &ifr, | |
4863 | sizeof(struct ifreq))) | |
4864 | ret = -EFAULT; | |
4865 | } | |
4866 | return ret; | |
1da177e4 | 4867 | |
d1b19dff ED |
4868 | /* |
4869 | * These ioctl calls: | |
4870 | * - require superuser power. | |
4871 | * - require strict serialization. | |
4872 | * - return a value | |
4873 | */ | |
4874 | case SIOCGMIIPHY: | |
4875 | case SIOCGMIIREG: | |
4876 | case SIOCSIFNAME: | |
4877 | if (!capable(CAP_NET_ADMIN)) | |
4878 | return -EPERM; | |
4879 | dev_load(net, ifr.ifr_name); | |
4880 | rtnl_lock(); | |
4881 | ret = dev_ifsioc(net, &ifr, cmd); | |
4882 | rtnl_unlock(); | |
4883 | if (!ret) { | |
4884 | if (colon) | |
4885 | *colon = ':'; | |
4886 | if (copy_to_user(arg, &ifr, | |
4887 | sizeof(struct ifreq))) | |
4888 | ret = -EFAULT; | |
4889 | } | |
4890 | return ret; | |
1da177e4 | 4891 | |
d1b19dff ED |
4892 | /* |
4893 | * These ioctl calls: | |
4894 | * - require superuser power. | |
4895 | * - require strict serialization. | |
4896 | * - do not return a value | |
4897 | */ | |
4898 | case SIOCSIFFLAGS: | |
4899 | case SIOCSIFMETRIC: | |
4900 | case SIOCSIFMTU: | |
4901 | case SIOCSIFMAP: | |
4902 | case SIOCSIFHWADDR: | |
4903 | case SIOCSIFSLAVE: | |
4904 | case SIOCADDMULTI: | |
4905 | case SIOCDELMULTI: | |
4906 | case SIOCSIFHWBROADCAST: | |
4907 | case SIOCSIFTXQLEN: | |
4908 | case SIOCSMIIREG: | |
4909 | case SIOCBONDENSLAVE: | |
4910 | case SIOCBONDRELEASE: | |
4911 | case SIOCBONDSETHWADDR: | |
4912 | case SIOCBONDCHANGEACTIVE: | |
4913 | case SIOCBRADDIF: | |
4914 | case SIOCBRDELIF: | |
4915 | case SIOCSHWTSTAMP: | |
4916 | if (!capable(CAP_NET_ADMIN)) | |
4917 | return -EPERM; | |
4918 | /* fall through */ | |
4919 | case SIOCBONDSLAVEINFOQUERY: | |
4920 | case SIOCBONDINFOQUERY: | |
4921 | dev_load(net, ifr.ifr_name); | |
4922 | rtnl_lock(); | |
4923 | ret = dev_ifsioc(net, &ifr, cmd); | |
4924 | rtnl_unlock(); | |
4925 | return ret; | |
4926 | ||
4927 | case SIOCGIFMEM: | |
4928 | /* Get the per device memory space. We can add this but | |
4929 | * currently do not support it */ | |
4930 | case SIOCSIFMEM: | |
4931 | /* Set the per device memory buffer space. | |
4932 | * Not applicable in our case */ | |
4933 | case SIOCSIFLINK: | |
4934 | return -EINVAL; | |
4935 | ||
4936 | /* | |
4937 | * Unknown or private ioctl. | |
4938 | */ | |
4939 | default: | |
4940 | if (cmd == SIOCWANDEV || | |
4941 | (cmd >= SIOCDEVPRIVATE && | |
4942 | cmd <= SIOCDEVPRIVATE + 15)) { | |
881d966b | 4943 | dev_load(net, ifr.ifr_name); |
1da177e4 | 4944 | rtnl_lock(); |
881d966b | 4945 | ret = dev_ifsioc(net, &ifr, cmd); |
1da177e4 | 4946 | rtnl_unlock(); |
d1b19dff ED |
4947 | if (!ret && copy_to_user(arg, &ifr, |
4948 | sizeof(struct ifreq))) | |
4949 | ret = -EFAULT; | |
1da177e4 | 4950 | return ret; |
d1b19dff ED |
4951 | } |
4952 | /* Take care of Wireless Extensions */ | |
4953 | if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) | |
4954 | return wext_handle_ioctl(net, &ifr, cmd, arg); | |
4955 | return -EINVAL; | |
1da177e4 LT |
4956 | } |
4957 | } | |
4958 | ||
4959 | ||
4960 | /** | |
4961 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 4962 | * @net: the applicable net namespace |
1da177e4 LT |
4963 | * |
4964 | * Returns a suitable unique value for a new device interface | |
4965 | * number. The caller must hold the rtnl semaphore or the | |
4966 | * dev_base_lock to be sure it remains unique. | |
4967 | */ | |
881d966b | 4968 | static int dev_new_index(struct net *net) |
1da177e4 LT |
4969 | { |
4970 | static int ifindex; | |
4971 | for (;;) { | |
4972 | if (++ifindex <= 0) | |
4973 | ifindex = 1; | |
881d966b | 4974 | if (!__dev_get_by_index(net, ifindex)) |
1da177e4 LT |
4975 | return ifindex; |
4976 | } | |
4977 | } | |
4978 | ||
1da177e4 | 4979 | /* Delayed registration/unregisteration */ |
3b5b34fd | 4980 | static LIST_HEAD(net_todo_list); |
1da177e4 | 4981 | |
6f05f629 | 4982 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 4983 | { |
1da177e4 | 4984 | list_add_tail(&dev->todo_list, &net_todo_list); |
1da177e4 LT |
4985 | } |
4986 | ||
9b5e383c | 4987 | static void rollback_registered_many(struct list_head *head) |
93ee31f1 | 4988 | { |
e93737b0 | 4989 | struct net_device *dev, *tmp; |
9b5e383c | 4990 | |
93ee31f1 DL |
4991 | BUG_ON(dev_boot_phase); |
4992 | ASSERT_RTNL(); | |
4993 | ||
e93737b0 | 4994 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
9b5e383c | 4995 | /* Some devices call without registering |
e93737b0 KK |
4996 | * for initialization unwind. Remove those |
4997 | * devices and proceed with the remaining. | |
9b5e383c ED |
4998 | */ |
4999 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
5000 | pr_debug("unregister_netdevice: device %s/%p never " | |
5001 | "was registered\n", dev->name, dev); | |
93ee31f1 | 5002 | |
9b5e383c | 5003 | WARN_ON(1); |
e93737b0 KK |
5004 | list_del(&dev->unreg_list); |
5005 | continue; | |
9b5e383c | 5006 | } |
93ee31f1 | 5007 | |
9b5e383c | 5008 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
44345724 | 5009 | } |
93ee31f1 | 5010 | |
44345724 OP |
5011 | /* If device is running, close it first. */ |
5012 | dev_close_many(head); | |
93ee31f1 | 5013 | |
44345724 | 5014 | list_for_each_entry(dev, head, unreg_list) { |
9b5e383c ED |
5015 | /* And unlink it from device chain. */ |
5016 | unlist_netdevice(dev); | |
93ee31f1 | 5017 | |
9b5e383c ED |
5018 | dev->reg_state = NETREG_UNREGISTERING; |
5019 | } | |
93ee31f1 DL |
5020 | |
5021 | synchronize_net(); | |
5022 | ||
9b5e383c ED |
5023 | list_for_each_entry(dev, head, unreg_list) { |
5024 | /* Shutdown queueing discipline. */ | |
5025 | dev_shutdown(dev); | |
93ee31f1 DL |
5026 | |
5027 | ||
9b5e383c ED |
5028 | /* Notify protocols, that we are about to destroy |
5029 | this device. They should clean all the things. | |
5030 | */ | |
5031 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
93ee31f1 | 5032 | |
a2835763 PM |
5033 | if (!dev->rtnl_link_ops || |
5034 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
5035 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U); | |
5036 | ||
9b5e383c ED |
5037 | /* |
5038 | * Flush the unicast and multicast chains | |
5039 | */ | |
a748ee24 | 5040 | dev_uc_flush(dev); |
22bedad3 | 5041 | dev_mc_flush(dev); |
93ee31f1 | 5042 | |
9b5e383c ED |
5043 | if (dev->netdev_ops->ndo_uninit) |
5044 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 | 5045 | |
9b5e383c ED |
5046 | /* Notifier chain MUST detach us from master device. */ |
5047 | WARN_ON(dev->master); | |
93ee31f1 | 5048 | |
9b5e383c ED |
5049 | /* Remove entries from kobject tree */ |
5050 | netdev_unregister_kobject(dev); | |
5051 | } | |
93ee31f1 | 5052 | |
a5ee1551 | 5053 | /* Process any work delayed until the end of the batch */ |
e5e26d75 | 5054 | dev = list_first_entry(head, struct net_device, unreg_list); |
a5ee1551 | 5055 | call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev); |
93ee31f1 | 5056 | |
ef885afb | 5057 | rcu_barrier(); |
395264d5 | 5058 | |
a5ee1551 | 5059 | list_for_each_entry(dev, head, unreg_list) |
9b5e383c ED |
5060 | dev_put(dev); |
5061 | } | |
5062 | ||
5063 | static void rollback_registered(struct net_device *dev) | |
5064 | { | |
5065 | LIST_HEAD(single); | |
5066 | ||
5067 | list_add(&dev->unreg_list, &single); | |
5068 | rollback_registered_many(&single); | |
ceaaec98 | 5069 | list_del(&single); |
93ee31f1 DL |
5070 | } |
5071 | ||
b63365a2 HX |
5072 | unsigned long netdev_fix_features(unsigned long features, const char *name) |
5073 | { | |
5074 | /* Fix illegal SG+CSUM combinations. */ | |
5075 | if ((features & NETIF_F_SG) && | |
5076 | !(features & NETIF_F_ALL_CSUM)) { | |
5077 | if (name) | |
5078 | printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no " | |
5079 | "checksum feature.\n", name); | |
5080 | features &= ~NETIF_F_SG; | |
5081 | } | |
5082 | ||
5083 | /* TSO requires that SG is present as well. */ | |
5084 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) { | |
5085 | if (name) | |
5086 | printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no " | |
5087 | "SG feature.\n", name); | |
5088 | features &= ~NETIF_F_TSO; | |
5089 | } | |
5090 | ||
5091 | if (features & NETIF_F_UFO) { | |
79032644 MM |
5092 | /* maybe split UFO into V4 and V6? */ |
5093 | if (!((features & NETIF_F_GEN_CSUM) || | |
5094 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM)) | |
5095 | == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
b63365a2 HX |
5096 | if (name) |
5097 | printk(KERN_ERR "%s: Dropping NETIF_F_UFO " | |
79032644 | 5098 | "since no checksum offload features.\n", |
b63365a2 HX |
5099 | name); |
5100 | features &= ~NETIF_F_UFO; | |
5101 | } | |
5102 | ||
5103 | if (!(features & NETIF_F_SG)) { | |
5104 | if (name) | |
5105 | printk(KERN_ERR "%s: Dropping NETIF_F_UFO " | |
5106 | "since no NETIF_F_SG feature.\n", name); | |
5107 | features &= ~NETIF_F_UFO; | |
5108 | } | |
5109 | } | |
5110 | ||
5111 | return features; | |
5112 | } | |
5113 | EXPORT_SYMBOL(netdev_fix_features); | |
5114 | ||
fc4a7489 PM |
5115 | /** |
5116 | * netif_stacked_transfer_operstate - transfer operstate | |
5117 | * @rootdev: the root or lower level device to transfer state from | |
5118 | * @dev: the device to transfer operstate to | |
5119 | * | |
5120 | * Transfer operational state from root to device. This is normally | |
5121 | * called when a stacking relationship exists between the root | |
5122 | * device and the device(a leaf device). | |
5123 | */ | |
5124 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
5125 | struct net_device *dev) | |
5126 | { | |
5127 | if (rootdev->operstate == IF_OPER_DORMANT) | |
5128 | netif_dormant_on(dev); | |
5129 | else | |
5130 | netif_dormant_off(dev); | |
5131 | ||
5132 | if (netif_carrier_ok(rootdev)) { | |
5133 | if (!netif_carrier_ok(dev)) | |
5134 | netif_carrier_on(dev); | |
5135 | } else { | |
5136 | if (netif_carrier_ok(dev)) | |
5137 | netif_carrier_off(dev); | |
5138 | } | |
5139 | } | |
5140 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
5141 | ||
bf264145 | 5142 | #ifdef CONFIG_RPS |
1b4bf461 ED |
5143 | static int netif_alloc_rx_queues(struct net_device *dev) |
5144 | { | |
1b4bf461 | 5145 | unsigned int i, count = dev->num_rx_queues; |
bd25fa7b | 5146 | struct netdev_rx_queue *rx; |
1b4bf461 | 5147 | |
bd25fa7b | 5148 | BUG_ON(count < 1); |
1b4bf461 | 5149 | |
bd25fa7b TH |
5150 | rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL); |
5151 | if (!rx) { | |
5152 | pr_err("netdev: Unable to allocate %u rx queues.\n", count); | |
5153 | return -ENOMEM; | |
1b4bf461 | 5154 | } |
bd25fa7b TH |
5155 | dev->_rx = rx; |
5156 | ||
bd25fa7b | 5157 | for (i = 0; i < count; i++) |
fe822240 | 5158 | rx[i].dev = dev; |
1b4bf461 ED |
5159 | return 0; |
5160 | } | |
bf264145 | 5161 | #endif |
1b4bf461 | 5162 | |
aa942104 CG |
5163 | static void netdev_init_one_queue(struct net_device *dev, |
5164 | struct netdev_queue *queue, void *_unused) | |
5165 | { | |
5166 | /* Initialize queue lock */ | |
5167 | spin_lock_init(&queue->_xmit_lock); | |
5168 | netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); | |
5169 | queue->xmit_lock_owner = -1; | |
b236da69 | 5170 | netdev_queue_numa_node_write(queue, NUMA_NO_NODE); |
aa942104 CG |
5171 | queue->dev = dev; |
5172 | } | |
5173 | ||
e6484930 TH |
5174 | static int netif_alloc_netdev_queues(struct net_device *dev) |
5175 | { | |
5176 | unsigned int count = dev->num_tx_queues; | |
5177 | struct netdev_queue *tx; | |
5178 | ||
5179 | BUG_ON(count < 1); | |
5180 | ||
5181 | tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL); | |
5182 | if (!tx) { | |
5183 | pr_err("netdev: Unable to allocate %u tx queues.\n", | |
5184 | count); | |
5185 | return -ENOMEM; | |
5186 | } | |
5187 | dev->_tx = tx; | |
1d24eb48 | 5188 | |
e6484930 TH |
5189 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); |
5190 | spin_lock_init(&dev->tx_global_lock); | |
aa942104 CG |
5191 | |
5192 | return 0; | |
e6484930 TH |
5193 | } |
5194 | ||
1da177e4 LT |
5195 | /** |
5196 | * register_netdevice - register a network device | |
5197 | * @dev: device to register | |
5198 | * | |
5199 | * Take a completed network device structure and add it to the kernel | |
5200 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
5201 | * chain. 0 is returned on success. A negative errno code is returned | |
5202 | * on a failure to set up the device, or if the name is a duplicate. | |
5203 | * | |
5204 | * Callers must hold the rtnl semaphore. You may want | |
5205 | * register_netdev() instead of this. | |
5206 | * | |
5207 | * BUGS: | |
5208 | * The locking appears insufficient to guarantee two parallel registers | |
5209 | * will not get the same name. | |
5210 | */ | |
5211 | ||
5212 | int register_netdevice(struct net_device *dev) | |
5213 | { | |
1da177e4 | 5214 | int ret; |
d314774c | 5215 | struct net *net = dev_net(dev); |
1da177e4 LT |
5216 | |
5217 | BUG_ON(dev_boot_phase); | |
5218 | ASSERT_RTNL(); | |
5219 | ||
b17a7c17 SH |
5220 | might_sleep(); |
5221 | ||
1da177e4 LT |
5222 | /* When net_device's are persistent, this will be fatal. */ |
5223 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 5224 | BUG_ON(!net); |
1da177e4 | 5225 | |
f1f28aa3 | 5226 | spin_lock_init(&dev->addr_list_lock); |
cf508b12 | 5227 | netdev_set_addr_lockdep_class(dev); |
1da177e4 | 5228 | |
1da177e4 LT |
5229 | dev->iflink = -1; |
5230 | ||
5231 | /* Init, if this function is available */ | |
d314774c SH |
5232 | if (dev->netdev_ops->ndo_init) { |
5233 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
5234 | if (ret) { |
5235 | if (ret > 0) | |
5236 | ret = -EIO; | |
90833aa4 | 5237 | goto out; |
1da177e4 LT |
5238 | } |
5239 | } | |
4ec93edb | 5240 | |
8ce6cebc | 5241 | ret = dev_get_valid_name(dev, dev->name, 0); |
d9031024 | 5242 | if (ret) |
7ce1b0ed | 5243 | goto err_uninit; |
1da177e4 | 5244 | |
881d966b | 5245 | dev->ifindex = dev_new_index(net); |
1da177e4 LT |
5246 | if (dev->iflink == -1) |
5247 | dev->iflink = dev->ifindex; | |
5248 | ||
d212f87b SH |
5249 | /* Fix illegal checksum combinations */ |
5250 | if ((dev->features & NETIF_F_HW_CSUM) && | |
5251 | (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
5252 | printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n", | |
5253 | dev->name); | |
5254 | dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); | |
5255 | } | |
5256 | ||
5257 | if ((dev->features & NETIF_F_NO_CSUM) && | |
5258 | (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
5259 | printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n", | |
5260 | dev->name); | |
5261 | dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM); | |
5262 | } | |
5263 | ||
b63365a2 | 5264 | dev->features = netdev_fix_features(dev->features, dev->name); |
1da177e4 | 5265 | |
e5a4a72d LB |
5266 | /* Enable software GSO if SG is supported. */ |
5267 | if (dev->features & NETIF_F_SG) | |
5268 | dev->features |= NETIF_F_GSO; | |
5269 | ||
c5256c51 ED |
5270 | /* Enable GRO and NETIF_F_HIGHDMA for vlans by default, |
5271 | * vlan_dev_init() will do the dev->features check, so these features | |
5272 | * are enabled only if supported by underlying device. | |
16c3ea78 | 5273 | */ |
c5256c51 | 5274 | dev->vlan_features |= (NETIF_F_GRO | NETIF_F_HIGHDMA); |
16c3ea78 | 5275 | |
7ffbe3fd JB |
5276 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); |
5277 | ret = notifier_to_errno(ret); | |
5278 | if (ret) | |
5279 | goto err_uninit; | |
5280 | ||
8b41d188 | 5281 | ret = netdev_register_kobject(dev); |
b17a7c17 | 5282 | if (ret) |
7ce1b0ed | 5283 | goto err_uninit; |
b17a7c17 SH |
5284 | dev->reg_state = NETREG_REGISTERED; |
5285 | ||
1da177e4 LT |
5286 | /* |
5287 | * Default initial state at registry is that the | |
5288 | * device is present. | |
5289 | */ | |
5290 | ||
5291 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
5292 | ||
1da177e4 | 5293 | dev_init_scheduler(dev); |
1da177e4 | 5294 | dev_hold(dev); |
ce286d32 | 5295 | list_netdevice(dev); |
1da177e4 LT |
5296 | |
5297 | /* Notify protocols, that a new device appeared. */ | |
056925ab | 5298 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 5299 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
5300 | if (ret) { |
5301 | rollback_registered(dev); | |
5302 | dev->reg_state = NETREG_UNREGISTERED; | |
5303 | } | |
d90a909e EB |
5304 | /* |
5305 | * Prevent userspace races by waiting until the network | |
5306 | * device is fully setup before sending notifications. | |
5307 | */ | |
a2835763 PM |
5308 | if (!dev->rtnl_link_ops || |
5309 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
5310 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U); | |
1da177e4 LT |
5311 | |
5312 | out: | |
5313 | return ret; | |
7ce1b0ed HX |
5314 | |
5315 | err_uninit: | |
d314774c SH |
5316 | if (dev->netdev_ops->ndo_uninit) |
5317 | dev->netdev_ops->ndo_uninit(dev); | |
7ce1b0ed | 5318 | goto out; |
1da177e4 | 5319 | } |
d1b19dff | 5320 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 5321 | |
937f1ba5 BH |
5322 | /** |
5323 | * init_dummy_netdev - init a dummy network device for NAPI | |
5324 | * @dev: device to init | |
5325 | * | |
5326 | * This takes a network device structure and initialize the minimum | |
5327 | * amount of fields so it can be used to schedule NAPI polls without | |
5328 | * registering a full blown interface. This is to be used by drivers | |
5329 | * that need to tie several hardware interfaces to a single NAPI | |
5330 | * poll scheduler due to HW limitations. | |
5331 | */ | |
5332 | int init_dummy_netdev(struct net_device *dev) | |
5333 | { | |
5334 | /* Clear everything. Note we don't initialize spinlocks | |
5335 | * are they aren't supposed to be taken by any of the | |
5336 | * NAPI code and this dummy netdev is supposed to be | |
5337 | * only ever used for NAPI polls | |
5338 | */ | |
5339 | memset(dev, 0, sizeof(struct net_device)); | |
5340 | ||
5341 | /* make sure we BUG if trying to hit standard | |
5342 | * register/unregister code path | |
5343 | */ | |
5344 | dev->reg_state = NETREG_DUMMY; | |
5345 | ||
937f1ba5 BH |
5346 | /* NAPI wants this */ |
5347 | INIT_LIST_HEAD(&dev->napi_list); | |
5348 | ||
5349 | /* a dummy interface is started by default */ | |
5350 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
5351 | set_bit(__LINK_STATE_START, &dev->state); | |
5352 | ||
29b4433d ED |
5353 | /* Note : We dont allocate pcpu_refcnt for dummy devices, |
5354 | * because users of this 'device' dont need to change | |
5355 | * its refcount. | |
5356 | */ | |
5357 | ||
937f1ba5 BH |
5358 | return 0; |
5359 | } | |
5360 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
5361 | ||
5362 | ||
1da177e4 LT |
5363 | /** |
5364 | * register_netdev - register a network device | |
5365 | * @dev: device to register | |
5366 | * | |
5367 | * Take a completed network device structure and add it to the kernel | |
5368 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
5369 | * chain. 0 is returned on success. A negative errno code is returned | |
5370 | * on a failure to set up the device, or if the name is a duplicate. | |
5371 | * | |
38b4da38 | 5372 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
5373 | * and expands the device name if you passed a format string to |
5374 | * alloc_netdev. | |
5375 | */ | |
5376 | int register_netdev(struct net_device *dev) | |
5377 | { | |
5378 | int err; | |
5379 | ||
5380 | rtnl_lock(); | |
5381 | ||
5382 | /* | |
5383 | * If the name is a format string the caller wants us to do a | |
5384 | * name allocation. | |
5385 | */ | |
5386 | if (strchr(dev->name, '%')) { | |
5387 | err = dev_alloc_name(dev, dev->name); | |
5388 | if (err < 0) | |
5389 | goto out; | |
5390 | } | |
4ec93edb | 5391 | |
1da177e4 LT |
5392 | err = register_netdevice(dev); |
5393 | out: | |
5394 | rtnl_unlock(); | |
5395 | return err; | |
5396 | } | |
5397 | EXPORT_SYMBOL(register_netdev); | |
5398 | ||
29b4433d ED |
5399 | int netdev_refcnt_read(const struct net_device *dev) |
5400 | { | |
5401 | int i, refcnt = 0; | |
5402 | ||
5403 | for_each_possible_cpu(i) | |
5404 | refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); | |
5405 | return refcnt; | |
5406 | } | |
5407 | EXPORT_SYMBOL(netdev_refcnt_read); | |
5408 | ||
1da177e4 LT |
5409 | /* |
5410 | * netdev_wait_allrefs - wait until all references are gone. | |
5411 | * | |
5412 | * This is called when unregistering network devices. | |
5413 | * | |
5414 | * Any protocol or device that holds a reference should register | |
5415 | * for netdevice notification, and cleanup and put back the | |
5416 | * reference if they receive an UNREGISTER event. | |
5417 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 5418 | * call dev_put. |
1da177e4 LT |
5419 | */ |
5420 | static void netdev_wait_allrefs(struct net_device *dev) | |
5421 | { | |
5422 | unsigned long rebroadcast_time, warning_time; | |
29b4433d | 5423 | int refcnt; |
1da177e4 | 5424 | |
e014debe ED |
5425 | linkwatch_forget_dev(dev); |
5426 | ||
1da177e4 | 5427 | rebroadcast_time = warning_time = jiffies; |
29b4433d ED |
5428 | refcnt = netdev_refcnt_read(dev); |
5429 | ||
5430 | while (refcnt != 0) { | |
1da177e4 | 5431 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { |
6756ae4b | 5432 | rtnl_lock(); |
1da177e4 LT |
5433 | |
5434 | /* Rebroadcast unregister notification */ | |
056925ab | 5435 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
a5ee1551 | 5436 | /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users |
395264d5 | 5437 | * should have already handle it the first time */ |
1da177e4 LT |
5438 | |
5439 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, | |
5440 | &dev->state)) { | |
5441 | /* We must not have linkwatch events | |
5442 | * pending on unregister. If this | |
5443 | * happens, we simply run the queue | |
5444 | * unscheduled, resulting in a noop | |
5445 | * for this device. | |
5446 | */ | |
5447 | linkwatch_run_queue(); | |
5448 | } | |
5449 | ||
6756ae4b | 5450 | __rtnl_unlock(); |
1da177e4 LT |
5451 | |
5452 | rebroadcast_time = jiffies; | |
5453 | } | |
5454 | ||
5455 | msleep(250); | |
5456 | ||
29b4433d ED |
5457 | refcnt = netdev_refcnt_read(dev); |
5458 | ||
1da177e4 LT |
5459 | if (time_after(jiffies, warning_time + 10 * HZ)) { |
5460 | printk(KERN_EMERG "unregister_netdevice: " | |
5461 | "waiting for %s to become free. Usage " | |
5462 | "count = %d\n", | |
29b4433d | 5463 | dev->name, refcnt); |
1da177e4 LT |
5464 | warning_time = jiffies; |
5465 | } | |
5466 | } | |
5467 | } | |
5468 | ||
5469 | /* The sequence is: | |
5470 | * | |
5471 | * rtnl_lock(); | |
5472 | * ... | |
5473 | * register_netdevice(x1); | |
5474 | * register_netdevice(x2); | |
5475 | * ... | |
5476 | * unregister_netdevice(y1); | |
5477 | * unregister_netdevice(y2); | |
5478 | * ... | |
5479 | * rtnl_unlock(); | |
5480 | * free_netdev(y1); | |
5481 | * free_netdev(y2); | |
5482 | * | |
58ec3b4d | 5483 | * We are invoked by rtnl_unlock(). |
1da177e4 | 5484 | * This allows us to deal with problems: |
b17a7c17 | 5485 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
5486 | * without deadlocking with linkwatch via keventd. |
5487 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
5488 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
5489 | * |
5490 | * We must not return until all unregister events added during | |
5491 | * the interval the lock was held have been completed. | |
1da177e4 | 5492 | */ |
1da177e4 LT |
5493 | void netdev_run_todo(void) |
5494 | { | |
626ab0e6 | 5495 | struct list_head list; |
1da177e4 | 5496 | |
1da177e4 | 5497 | /* Snapshot list, allow later requests */ |
626ab0e6 | 5498 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
5499 | |
5500 | __rtnl_unlock(); | |
626ab0e6 | 5501 | |
1da177e4 LT |
5502 | while (!list_empty(&list)) { |
5503 | struct net_device *dev | |
e5e26d75 | 5504 | = list_first_entry(&list, struct net_device, todo_list); |
1da177e4 LT |
5505 | list_del(&dev->todo_list); |
5506 | ||
b17a7c17 SH |
5507 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
5508 | printk(KERN_ERR "network todo '%s' but state %d\n", | |
5509 | dev->name, dev->reg_state); | |
5510 | dump_stack(); | |
5511 | continue; | |
5512 | } | |
1da177e4 | 5513 | |
b17a7c17 | 5514 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 5515 | |
152102c7 | 5516 | on_each_cpu(flush_backlog, dev, 1); |
6e583ce5 | 5517 | |
b17a7c17 | 5518 | netdev_wait_allrefs(dev); |
1da177e4 | 5519 | |
b17a7c17 | 5520 | /* paranoia */ |
29b4433d | 5521 | BUG_ON(netdev_refcnt_read(dev)); |
95ae6b22 | 5522 | WARN_ON(rcu_dereference_raw(dev->ip_ptr)); |
198caeca | 5523 | WARN_ON(rcu_dereference_raw(dev->ip6_ptr)); |
547b792c | 5524 | WARN_ON(dev->dn_ptr); |
1da177e4 | 5525 | |
b17a7c17 SH |
5526 | if (dev->destructor) |
5527 | dev->destructor(dev); | |
9093bbb2 SH |
5528 | |
5529 | /* Free network device */ | |
5530 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 5531 | } |
1da177e4 LT |
5532 | } |
5533 | ||
3cfde79c BH |
5534 | /* Convert net_device_stats to rtnl_link_stats64. They have the same |
5535 | * fields in the same order, with only the type differing. | |
5536 | */ | |
5537 | static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, | |
5538 | const struct net_device_stats *netdev_stats) | |
5539 | { | |
5540 | #if BITS_PER_LONG == 64 | |
5541 | BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats)); | |
5542 | memcpy(stats64, netdev_stats, sizeof(*stats64)); | |
5543 | #else | |
5544 | size_t i, n = sizeof(*stats64) / sizeof(u64); | |
5545 | const unsigned long *src = (const unsigned long *)netdev_stats; | |
5546 | u64 *dst = (u64 *)stats64; | |
5547 | ||
5548 | BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) != | |
5549 | sizeof(*stats64) / sizeof(u64)); | |
5550 | for (i = 0; i < n; i++) | |
5551 | dst[i] = src[i]; | |
5552 | #endif | |
5553 | } | |
5554 | ||
eeda3fd6 SH |
5555 | /** |
5556 | * dev_get_stats - get network device statistics | |
5557 | * @dev: device to get statistics from | |
28172739 | 5558 | * @storage: place to store stats |
eeda3fd6 | 5559 | * |
d7753516 BH |
5560 | * Get network statistics from device. Return @storage. |
5561 | * The device driver may provide its own method by setting | |
5562 | * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; | |
5563 | * otherwise the internal statistics structure is used. | |
eeda3fd6 | 5564 | */ |
d7753516 BH |
5565 | struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, |
5566 | struct rtnl_link_stats64 *storage) | |
7004bf25 | 5567 | { |
eeda3fd6 SH |
5568 | const struct net_device_ops *ops = dev->netdev_ops; |
5569 | ||
28172739 ED |
5570 | if (ops->ndo_get_stats64) { |
5571 | memset(storage, 0, sizeof(*storage)); | |
caf586e5 ED |
5572 | ops->ndo_get_stats64(dev, storage); |
5573 | } else if (ops->ndo_get_stats) { | |
3cfde79c | 5574 | netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); |
caf586e5 ED |
5575 | } else { |
5576 | netdev_stats_to_stats64(storage, &dev->stats); | |
28172739 | 5577 | } |
caf586e5 | 5578 | storage->rx_dropped += atomic_long_read(&dev->rx_dropped); |
28172739 | 5579 | return storage; |
c45d286e | 5580 | } |
eeda3fd6 | 5581 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 5582 | |
24824a09 | 5583 | struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) |
dc2b4847 | 5584 | { |
24824a09 | 5585 | struct netdev_queue *queue = dev_ingress_queue(dev); |
dc2b4847 | 5586 | |
24824a09 ED |
5587 | #ifdef CONFIG_NET_CLS_ACT |
5588 | if (queue) | |
5589 | return queue; | |
5590 | queue = kzalloc(sizeof(*queue), GFP_KERNEL); | |
5591 | if (!queue) | |
5592 | return NULL; | |
5593 | netdev_init_one_queue(dev, queue, NULL); | |
24824a09 ED |
5594 | queue->qdisc = &noop_qdisc; |
5595 | queue->qdisc_sleeping = &noop_qdisc; | |
5596 | rcu_assign_pointer(dev->ingress_queue, queue); | |
5597 | #endif | |
5598 | return queue; | |
bb949fbd DM |
5599 | } |
5600 | ||
1da177e4 | 5601 | /** |
36909ea4 | 5602 | * alloc_netdev_mqs - allocate network device |
1da177e4 LT |
5603 | * @sizeof_priv: size of private data to allocate space for |
5604 | * @name: device name format string | |
5605 | * @setup: callback to initialize device | |
36909ea4 TH |
5606 | * @txqs: the number of TX subqueues to allocate |
5607 | * @rxqs: the number of RX subqueues to allocate | |
1da177e4 LT |
5608 | * |
5609 | * Allocates a struct net_device with private data area for driver use | |
f25f4e44 | 5610 | * and performs basic initialization. Also allocates subquue structs |
36909ea4 | 5611 | * for each queue on the device. |
1da177e4 | 5612 | */ |
36909ea4 TH |
5613 | struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, |
5614 | void (*setup)(struct net_device *), | |
5615 | unsigned int txqs, unsigned int rxqs) | |
1da177e4 | 5616 | { |
1da177e4 | 5617 | struct net_device *dev; |
7943986c | 5618 | size_t alloc_size; |
1ce8e7b5 | 5619 | struct net_device *p; |
1da177e4 | 5620 | |
b6fe17d6 SH |
5621 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
5622 | ||
36909ea4 | 5623 | if (txqs < 1) { |
55513fb4 TH |
5624 | pr_err("alloc_netdev: Unable to allocate device " |
5625 | "with zero queues.\n"); | |
5626 | return NULL; | |
5627 | } | |
5628 | ||
36909ea4 TH |
5629 | #ifdef CONFIG_RPS |
5630 | if (rxqs < 1) { | |
5631 | pr_err("alloc_netdev: Unable to allocate device " | |
5632 | "with zero RX queues.\n"); | |
5633 | return NULL; | |
5634 | } | |
5635 | #endif | |
5636 | ||
fd2ea0a7 | 5637 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
5638 | if (sizeof_priv) { |
5639 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 5640 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
5641 | alloc_size += sizeof_priv; |
5642 | } | |
5643 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 5644 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 5645 | |
31380de9 | 5646 | p = kzalloc(alloc_size, GFP_KERNEL); |
1da177e4 | 5647 | if (!p) { |
b6fe17d6 | 5648 | printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n"); |
1da177e4 LT |
5649 | return NULL; |
5650 | } | |
1da177e4 | 5651 | |
1ce8e7b5 | 5652 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 5653 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc | 5654 | |
29b4433d ED |
5655 | dev->pcpu_refcnt = alloc_percpu(int); |
5656 | if (!dev->pcpu_refcnt) | |
e6484930 | 5657 | goto free_p; |
ab9c73cc | 5658 | |
ab9c73cc | 5659 | if (dev_addr_init(dev)) |
29b4433d | 5660 | goto free_pcpu; |
ab9c73cc | 5661 | |
22bedad3 | 5662 | dev_mc_init(dev); |
a748ee24 | 5663 | dev_uc_init(dev); |
ccffad25 | 5664 | |
c346dca1 | 5665 | dev_net_set(dev, &init_net); |
1da177e4 | 5666 | |
8d3bdbd5 DM |
5667 | dev->gso_max_size = GSO_MAX_SIZE; |
5668 | ||
5669 | INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list); | |
5670 | dev->ethtool_ntuple_list.count = 0; | |
5671 | INIT_LIST_HEAD(&dev->napi_list); | |
5672 | INIT_LIST_HEAD(&dev->unreg_list); | |
5673 | INIT_LIST_HEAD(&dev->link_watch_list); | |
5674 | dev->priv_flags = IFF_XMIT_DST_RELEASE; | |
5675 | setup(dev); | |
5676 | ||
36909ea4 TH |
5677 | dev->num_tx_queues = txqs; |
5678 | dev->real_num_tx_queues = txqs; | |
ed9af2e8 | 5679 | if (netif_alloc_netdev_queues(dev)) |
8d3bdbd5 | 5680 | goto free_all; |
e8a0464c | 5681 | |
df334545 | 5682 | #ifdef CONFIG_RPS |
36909ea4 TH |
5683 | dev->num_rx_queues = rxqs; |
5684 | dev->real_num_rx_queues = rxqs; | |
fe822240 | 5685 | if (netif_alloc_rx_queues(dev)) |
8d3bdbd5 | 5686 | goto free_all; |
df334545 | 5687 | #endif |
0a9627f2 | 5688 | |
1da177e4 LT |
5689 | strcpy(dev->name, name); |
5690 | return dev; | |
ab9c73cc | 5691 | |
8d3bdbd5 DM |
5692 | free_all: |
5693 | free_netdev(dev); | |
5694 | return NULL; | |
5695 | ||
29b4433d ED |
5696 | free_pcpu: |
5697 | free_percpu(dev->pcpu_refcnt); | |
ed9af2e8 | 5698 | kfree(dev->_tx); |
fe822240 TH |
5699 | #ifdef CONFIG_RPS |
5700 | kfree(dev->_rx); | |
5701 | #endif | |
5702 | ||
ab9c73cc JP |
5703 | free_p: |
5704 | kfree(p); | |
5705 | return NULL; | |
1da177e4 | 5706 | } |
36909ea4 | 5707 | EXPORT_SYMBOL(alloc_netdev_mqs); |
1da177e4 LT |
5708 | |
5709 | /** | |
5710 | * free_netdev - free network device | |
5711 | * @dev: device | |
5712 | * | |
4ec93edb YH |
5713 | * This function does the last stage of destroying an allocated device |
5714 | * interface. The reference to the device object is released. | |
1da177e4 LT |
5715 | * If this is the last reference then it will be freed. |
5716 | */ | |
5717 | void free_netdev(struct net_device *dev) | |
5718 | { | |
d565b0a1 HX |
5719 | struct napi_struct *p, *n; |
5720 | ||
f3005d7f DL |
5721 | release_net(dev_net(dev)); |
5722 | ||
e8a0464c | 5723 | kfree(dev->_tx); |
fe822240 TH |
5724 | #ifdef CONFIG_RPS |
5725 | kfree(dev->_rx); | |
5726 | #endif | |
e8a0464c | 5727 | |
24824a09 ED |
5728 | kfree(rcu_dereference_raw(dev->ingress_queue)); |
5729 | ||
f001fde5 JP |
5730 | /* Flush device addresses */ |
5731 | dev_addr_flush(dev); | |
5732 | ||
15682bc4 PWJ |
5733 | /* Clear ethtool n-tuple list */ |
5734 | ethtool_ntuple_flush(dev); | |
5735 | ||
d565b0a1 HX |
5736 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
5737 | netif_napi_del(p); | |
5738 | ||
29b4433d ED |
5739 | free_percpu(dev->pcpu_refcnt); |
5740 | dev->pcpu_refcnt = NULL; | |
5741 | ||
3041a069 | 5742 | /* Compatibility with error handling in drivers */ |
1da177e4 LT |
5743 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
5744 | kfree((char *)dev - dev->padded); | |
5745 | return; | |
5746 | } | |
5747 | ||
5748 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
5749 | dev->reg_state = NETREG_RELEASED; | |
5750 | ||
43cb76d9 GKH |
5751 | /* will free via device release */ |
5752 | put_device(&dev->dev); | |
1da177e4 | 5753 | } |
d1b19dff | 5754 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 5755 | |
f0db275a SH |
5756 | /** |
5757 | * synchronize_net - Synchronize with packet receive processing | |
5758 | * | |
5759 | * Wait for packets currently being received to be done. | |
5760 | * Does not block later packets from starting. | |
5761 | */ | |
4ec93edb | 5762 | void synchronize_net(void) |
1da177e4 LT |
5763 | { |
5764 | might_sleep(); | |
fbd568a3 | 5765 | synchronize_rcu(); |
1da177e4 | 5766 | } |
d1b19dff | 5767 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
5768 | |
5769 | /** | |
44a0873d | 5770 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 5771 | * @dev: device |
44a0873d | 5772 | * @head: list |
6ebfbc06 | 5773 | * |
1da177e4 | 5774 | * This function shuts down a device interface and removes it |
d59b54b1 | 5775 | * from the kernel tables. |
44a0873d | 5776 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
5777 | * |
5778 | * Callers must hold the rtnl semaphore. You may want | |
5779 | * unregister_netdev() instead of this. | |
5780 | */ | |
5781 | ||
44a0873d | 5782 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 5783 | { |
a6620712 HX |
5784 | ASSERT_RTNL(); |
5785 | ||
44a0873d | 5786 | if (head) { |
9fdce099 | 5787 | list_move_tail(&dev->unreg_list, head); |
44a0873d ED |
5788 | } else { |
5789 | rollback_registered(dev); | |
5790 | /* Finish processing unregister after unlock */ | |
5791 | net_set_todo(dev); | |
5792 | } | |
1da177e4 | 5793 | } |
44a0873d | 5794 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 5795 | |
9b5e383c ED |
5796 | /** |
5797 | * unregister_netdevice_many - unregister many devices | |
5798 | * @head: list of devices | |
9b5e383c ED |
5799 | */ |
5800 | void unregister_netdevice_many(struct list_head *head) | |
5801 | { | |
5802 | struct net_device *dev; | |
5803 | ||
5804 | if (!list_empty(head)) { | |
5805 | rollback_registered_many(head); | |
5806 | list_for_each_entry(dev, head, unreg_list) | |
5807 | net_set_todo(dev); | |
5808 | } | |
5809 | } | |
63c8099d | 5810 | EXPORT_SYMBOL(unregister_netdevice_many); |
9b5e383c | 5811 | |
1da177e4 LT |
5812 | /** |
5813 | * unregister_netdev - remove device from the kernel | |
5814 | * @dev: device | |
5815 | * | |
5816 | * This function shuts down a device interface and removes it | |
d59b54b1 | 5817 | * from the kernel tables. |
1da177e4 LT |
5818 | * |
5819 | * This is just a wrapper for unregister_netdevice that takes | |
5820 | * the rtnl semaphore. In general you want to use this and not | |
5821 | * unregister_netdevice. | |
5822 | */ | |
5823 | void unregister_netdev(struct net_device *dev) | |
5824 | { | |
5825 | rtnl_lock(); | |
5826 | unregister_netdevice(dev); | |
5827 | rtnl_unlock(); | |
5828 | } | |
1da177e4 LT |
5829 | EXPORT_SYMBOL(unregister_netdev); |
5830 | ||
ce286d32 EB |
5831 | /** |
5832 | * dev_change_net_namespace - move device to different nethost namespace | |
5833 | * @dev: device | |
5834 | * @net: network namespace | |
5835 | * @pat: If not NULL name pattern to try if the current device name | |
5836 | * is already taken in the destination network namespace. | |
5837 | * | |
5838 | * This function shuts down a device interface and moves it | |
5839 | * to a new network namespace. On success 0 is returned, on | |
5840 | * a failure a netagive errno code is returned. | |
5841 | * | |
5842 | * Callers must hold the rtnl semaphore. | |
5843 | */ | |
5844 | ||
5845 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
5846 | { | |
ce286d32 EB |
5847 | int err; |
5848 | ||
5849 | ASSERT_RTNL(); | |
5850 | ||
5851 | /* Don't allow namespace local devices to be moved. */ | |
5852 | err = -EINVAL; | |
5853 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
5854 | goto out; | |
5855 | ||
5856 | /* Ensure the device has been registrered */ | |
5857 | err = -EINVAL; | |
5858 | if (dev->reg_state != NETREG_REGISTERED) | |
5859 | goto out; | |
5860 | ||
5861 | /* Get out if there is nothing todo */ | |
5862 | err = 0; | |
878628fb | 5863 | if (net_eq(dev_net(dev), net)) |
ce286d32 EB |
5864 | goto out; |
5865 | ||
5866 | /* Pick the destination device name, and ensure | |
5867 | * we can use it in the destination network namespace. | |
5868 | */ | |
5869 | err = -EEXIST; | |
d9031024 | 5870 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
5871 | /* We get here if we can't use the current device name */ |
5872 | if (!pat) | |
5873 | goto out; | |
8ce6cebc | 5874 | if (dev_get_valid_name(dev, pat, 1)) |
ce286d32 EB |
5875 | goto out; |
5876 | } | |
5877 | ||
5878 | /* | |
5879 | * And now a mini version of register_netdevice unregister_netdevice. | |
5880 | */ | |
5881 | ||
5882 | /* If device is running close it first. */ | |
9b772652 | 5883 | dev_close(dev); |
ce286d32 EB |
5884 | |
5885 | /* And unlink it from device chain */ | |
5886 | err = -ENODEV; | |
5887 | unlist_netdevice(dev); | |
5888 | ||
5889 | synchronize_net(); | |
5890 | ||
5891 | /* Shutdown queueing discipline. */ | |
5892 | dev_shutdown(dev); | |
5893 | ||
5894 | /* Notify protocols, that we are about to destroy | |
5895 | this device. They should clean all the things. | |
3b27e105 DL |
5896 | |
5897 | Note that dev->reg_state stays at NETREG_REGISTERED. | |
5898 | This is wanted because this way 8021q and macvlan know | |
5899 | the device is just moving and can keep their slaves up. | |
ce286d32 EB |
5900 | */ |
5901 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
a5ee1551 | 5902 | call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev); |
ce286d32 EB |
5903 | |
5904 | /* | |
5905 | * Flush the unicast and multicast chains | |
5906 | */ | |
a748ee24 | 5907 | dev_uc_flush(dev); |
22bedad3 | 5908 | dev_mc_flush(dev); |
ce286d32 EB |
5909 | |
5910 | /* Actually switch the network namespace */ | |
c346dca1 | 5911 | dev_net_set(dev, net); |
ce286d32 | 5912 | |
ce286d32 EB |
5913 | /* If there is an ifindex conflict assign a new one */ |
5914 | if (__dev_get_by_index(net, dev->ifindex)) { | |
5915 | int iflink = (dev->iflink == dev->ifindex); | |
5916 | dev->ifindex = dev_new_index(net); | |
5917 | if (iflink) | |
5918 | dev->iflink = dev->ifindex; | |
5919 | } | |
5920 | ||
8b41d188 | 5921 | /* Fixup kobjects */ |
a1b3f594 | 5922 | err = device_rename(&dev->dev, dev->name); |
8b41d188 | 5923 | WARN_ON(err); |
ce286d32 EB |
5924 | |
5925 | /* Add the device back in the hashes */ | |
5926 | list_netdevice(dev); | |
5927 | ||
5928 | /* Notify protocols, that a new device appeared. */ | |
5929 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
5930 | ||
d90a909e EB |
5931 | /* |
5932 | * Prevent userspace races by waiting until the network | |
5933 | * device is fully setup before sending notifications. | |
5934 | */ | |
5935 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U); | |
5936 | ||
ce286d32 EB |
5937 | synchronize_net(); |
5938 | err = 0; | |
5939 | out: | |
5940 | return err; | |
5941 | } | |
463d0183 | 5942 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 5943 | |
1da177e4 LT |
5944 | static int dev_cpu_callback(struct notifier_block *nfb, |
5945 | unsigned long action, | |
5946 | void *ocpu) | |
5947 | { | |
5948 | struct sk_buff **list_skb; | |
1da177e4 LT |
5949 | struct sk_buff *skb; |
5950 | unsigned int cpu, oldcpu = (unsigned long)ocpu; | |
5951 | struct softnet_data *sd, *oldsd; | |
5952 | ||
8bb78442 | 5953 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
1da177e4 LT |
5954 | return NOTIFY_OK; |
5955 | ||
5956 | local_irq_disable(); | |
5957 | cpu = smp_processor_id(); | |
5958 | sd = &per_cpu(softnet_data, cpu); | |
5959 | oldsd = &per_cpu(softnet_data, oldcpu); | |
5960 | ||
5961 | /* Find end of our completion_queue. */ | |
5962 | list_skb = &sd->completion_queue; | |
5963 | while (*list_skb) | |
5964 | list_skb = &(*list_skb)->next; | |
5965 | /* Append completion queue from offline CPU. */ | |
5966 | *list_skb = oldsd->completion_queue; | |
5967 | oldsd->completion_queue = NULL; | |
5968 | ||
1da177e4 | 5969 | /* Append output queue from offline CPU. */ |
a9cbd588 CG |
5970 | if (oldsd->output_queue) { |
5971 | *sd->output_queue_tailp = oldsd->output_queue; | |
5972 | sd->output_queue_tailp = oldsd->output_queue_tailp; | |
5973 | oldsd->output_queue = NULL; | |
5974 | oldsd->output_queue_tailp = &oldsd->output_queue; | |
5975 | } | |
1da177e4 LT |
5976 | |
5977 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
5978 | local_irq_enable(); | |
5979 | ||
5980 | /* Process offline CPU's input_pkt_queue */ | |
76cc8b13 | 5981 | while ((skb = __skb_dequeue(&oldsd->process_queue))) { |
1da177e4 | 5982 | netif_rx(skb); |
76cc8b13 | 5983 | input_queue_head_incr(oldsd); |
fec5e652 | 5984 | } |
76cc8b13 | 5985 | while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) { |
6e7676c1 | 5986 | netif_rx(skb); |
76cc8b13 TH |
5987 | input_queue_head_incr(oldsd); |
5988 | } | |
1da177e4 LT |
5989 | |
5990 | return NOTIFY_OK; | |
5991 | } | |
1da177e4 LT |
5992 | |
5993 | ||
7f353bf2 | 5994 | /** |
b63365a2 HX |
5995 | * netdev_increment_features - increment feature set by one |
5996 | * @all: current feature set | |
5997 | * @one: new feature set | |
5998 | * @mask: mask feature set | |
7f353bf2 HX |
5999 | * |
6000 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
6001 | * @one to the master device with current feature set @all. Will not |
6002 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 6003 | */ |
b63365a2 HX |
6004 | unsigned long netdev_increment_features(unsigned long all, unsigned long one, |
6005 | unsigned long mask) | |
6006 | { | |
6007 | /* If device needs checksumming, downgrade to it. */ | |
d1b19dff | 6008 | if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM)) |
b63365a2 HX |
6009 | all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM); |
6010 | else if (mask & NETIF_F_ALL_CSUM) { | |
6011 | /* If one device supports v4/v6 checksumming, set for all. */ | |
6012 | if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) && | |
6013 | !(all & NETIF_F_GEN_CSUM)) { | |
6014 | all &= ~NETIF_F_ALL_CSUM; | |
6015 | all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); | |
6016 | } | |
e2a6b852 | 6017 | |
b63365a2 HX |
6018 | /* If one device supports hw checksumming, set for all. */ |
6019 | if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) { | |
6020 | all &= ~NETIF_F_ALL_CSUM; | |
6021 | all |= NETIF_F_HW_CSUM; | |
6022 | } | |
6023 | } | |
7f353bf2 | 6024 | |
b63365a2 | 6025 | one |= NETIF_F_ALL_CSUM; |
7f353bf2 | 6026 | |
b63365a2 | 6027 | one |= all & NETIF_F_ONE_FOR_ALL; |
d9f5950f | 6028 | all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO; |
b63365a2 | 6029 | all |= one & mask & NETIF_F_ONE_FOR_ALL; |
7f353bf2 HX |
6030 | |
6031 | return all; | |
6032 | } | |
b63365a2 | 6033 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 6034 | |
30d97d35 PE |
6035 | static struct hlist_head *netdev_create_hash(void) |
6036 | { | |
6037 | int i; | |
6038 | struct hlist_head *hash; | |
6039 | ||
6040 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
6041 | if (hash != NULL) | |
6042 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
6043 | INIT_HLIST_HEAD(&hash[i]); | |
6044 | ||
6045 | return hash; | |
6046 | } | |
6047 | ||
881d966b | 6048 | /* Initialize per network namespace state */ |
4665079c | 6049 | static int __net_init netdev_init(struct net *net) |
881d966b | 6050 | { |
881d966b | 6051 | INIT_LIST_HEAD(&net->dev_base_head); |
881d966b | 6052 | |
30d97d35 PE |
6053 | net->dev_name_head = netdev_create_hash(); |
6054 | if (net->dev_name_head == NULL) | |
6055 | goto err_name; | |
881d966b | 6056 | |
30d97d35 PE |
6057 | net->dev_index_head = netdev_create_hash(); |
6058 | if (net->dev_index_head == NULL) | |
6059 | goto err_idx; | |
881d966b EB |
6060 | |
6061 | return 0; | |
30d97d35 PE |
6062 | |
6063 | err_idx: | |
6064 | kfree(net->dev_name_head); | |
6065 | err_name: | |
6066 | return -ENOMEM; | |
881d966b EB |
6067 | } |
6068 | ||
f0db275a SH |
6069 | /** |
6070 | * netdev_drivername - network driver for the device | |
6071 | * @dev: network device | |
6072 | * @buffer: buffer for resulting name | |
6073 | * @len: size of buffer | |
6074 | * | |
6075 | * Determine network driver for device. | |
6076 | */ | |
cf04a4c7 | 6077 | char *netdev_drivername(const struct net_device *dev, char *buffer, int len) |
6579e57b | 6078 | { |
cf04a4c7 SH |
6079 | const struct device_driver *driver; |
6080 | const struct device *parent; | |
6579e57b AV |
6081 | |
6082 | if (len <= 0 || !buffer) | |
6083 | return buffer; | |
6084 | buffer[0] = 0; | |
6085 | ||
6086 | parent = dev->dev.parent; | |
6087 | ||
6088 | if (!parent) | |
6089 | return buffer; | |
6090 | ||
6091 | driver = parent->driver; | |
6092 | if (driver && driver->name) | |
6093 | strlcpy(buffer, driver->name, len); | |
6094 | return buffer; | |
6095 | } | |
6096 | ||
256df2f3 JP |
6097 | static int __netdev_printk(const char *level, const struct net_device *dev, |
6098 | struct va_format *vaf) | |
6099 | { | |
6100 | int r; | |
6101 | ||
6102 | if (dev && dev->dev.parent) | |
6103 | r = dev_printk(level, dev->dev.parent, "%s: %pV", | |
6104 | netdev_name(dev), vaf); | |
6105 | else if (dev) | |
6106 | r = printk("%s%s: %pV", level, netdev_name(dev), vaf); | |
6107 | else | |
6108 | r = printk("%s(NULL net_device): %pV", level, vaf); | |
6109 | ||
6110 | return r; | |
6111 | } | |
6112 | ||
6113 | int netdev_printk(const char *level, const struct net_device *dev, | |
6114 | const char *format, ...) | |
6115 | { | |
6116 | struct va_format vaf; | |
6117 | va_list args; | |
6118 | int r; | |
6119 | ||
6120 | va_start(args, format); | |
6121 | ||
6122 | vaf.fmt = format; | |
6123 | vaf.va = &args; | |
6124 | ||
6125 | r = __netdev_printk(level, dev, &vaf); | |
6126 | va_end(args); | |
6127 | ||
6128 | return r; | |
6129 | } | |
6130 | EXPORT_SYMBOL(netdev_printk); | |
6131 | ||
6132 | #define define_netdev_printk_level(func, level) \ | |
6133 | int func(const struct net_device *dev, const char *fmt, ...) \ | |
6134 | { \ | |
6135 | int r; \ | |
6136 | struct va_format vaf; \ | |
6137 | va_list args; \ | |
6138 | \ | |
6139 | va_start(args, fmt); \ | |
6140 | \ | |
6141 | vaf.fmt = fmt; \ | |
6142 | vaf.va = &args; \ | |
6143 | \ | |
6144 | r = __netdev_printk(level, dev, &vaf); \ | |
6145 | va_end(args); \ | |
6146 | \ | |
6147 | return r; \ | |
6148 | } \ | |
6149 | EXPORT_SYMBOL(func); | |
6150 | ||
6151 | define_netdev_printk_level(netdev_emerg, KERN_EMERG); | |
6152 | define_netdev_printk_level(netdev_alert, KERN_ALERT); | |
6153 | define_netdev_printk_level(netdev_crit, KERN_CRIT); | |
6154 | define_netdev_printk_level(netdev_err, KERN_ERR); | |
6155 | define_netdev_printk_level(netdev_warn, KERN_WARNING); | |
6156 | define_netdev_printk_level(netdev_notice, KERN_NOTICE); | |
6157 | define_netdev_printk_level(netdev_info, KERN_INFO); | |
6158 | ||
4665079c | 6159 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
6160 | { |
6161 | kfree(net->dev_name_head); | |
6162 | kfree(net->dev_index_head); | |
6163 | } | |
6164 | ||
022cbae6 | 6165 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
6166 | .init = netdev_init, |
6167 | .exit = netdev_exit, | |
6168 | }; | |
6169 | ||
4665079c | 6170 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 6171 | { |
e008b5fc | 6172 | struct net_device *dev, *aux; |
ce286d32 | 6173 | /* |
e008b5fc | 6174 | * Push all migratable network devices back to the |
ce286d32 EB |
6175 | * initial network namespace |
6176 | */ | |
6177 | rtnl_lock(); | |
e008b5fc | 6178 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 6179 | int err; |
aca51397 | 6180 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
6181 | |
6182 | /* Ignore unmoveable devices (i.e. loopback) */ | |
6183 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
6184 | continue; | |
6185 | ||
e008b5fc EB |
6186 | /* Leave virtual devices for the generic cleanup */ |
6187 | if (dev->rtnl_link_ops) | |
6188 | continue; | |
d0c082ce | 6189 | |
ce286d32 | 6190 | /* Push remaing network devices to init_net */ |
aca51397 PE |
6191 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
6192 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
ce286d32 | 6193 | if (err) { |
aca51397 | 6194 | printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n", |
ce286d32 | 6195 | __func__, dev->name, err); |
aca51397 | 6196 | BUG(); |
ce286d32 EB |
6197 | } |
6198 | } | |
6199 | rtnl_unlock(); | |
6200 | } | |
6201 | ||
04dc7f6b EB |
6202 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
6203 | { | |
6204 | /* At exit all network devices most be removed from a network | |
b595076a | 6205 | * namespace. Do this in the reverse order of registration. |
04dc7f6b EB |
6206 | * Do this across as many network namespaces as possible to |
6207 | * improve batching efficiency. | |
6208 | */ | |
6209 | struct net_device *dev; | |
6210 | struct net *net; | |
6211 | LIST_HEAD(dev_kill_list); | |
6212 | ||
6213 | rtnl_lock(); | |
6214 | list_for_each_entry(net, net_list, exit_list) { | |
6215 | for_each_netdev_reverse(net, dev) { | |
6216 | if (dev->rtnl_link_ops) | |
6217 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); | |
6218 | else | |
6219 | unregister_netdevice_queue(dev, &dev_kill_list); | |
6220 | } | |
6221 | } | |
6222 | unregister_netdevice_many(&dev_kill_list); | |
ceaaec98 | 6223 | list_del(&dev_kill_list); |
04dc7f6b EB |
6224 | rtnl_unlock(); |
6225 | } | |
6226 | ||
022cbae6 | 6227 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 6228 | .exit = default_device_exit, |
04dc7f6b | 6229 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
6230 | }; |
6231 | ||
1da177e4 LT |
6232 | /* |
6233 | * Initialize the DEV module. At boot time this walks the device list and | |
6234 | * unhooks any devices that fail to initialise (normally hardware not | |
6235 | * present) and leaves us with a valid list of present and active devices. | |
6236 | * | |
6237 | */ | |
6238 | ||
6239 | /* | |
6240 | * This is called single threaded during boot, so no need | |
6241 | * to take the rtnl semaphore. | |
6242 | */ | |
6243 | static int __init net_dev_init(void) | |
6244 | { | |
6245 | int i, rc = -ENOMEM; | |
6246 | ||
6247 | BUG_ON(!dev_boot_phase); | |
6248 | ||
1da177e4 LT |
6249 | if (dev_proc_init()) |
6250 | goto out; | |
6251 | ||
8b41d188 | 6252 | if (netdev_kobject_init()) |
1da177e4 LT |
6253 | goto out; |
6254 | ||
6255 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 6256 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
6257 | INIT_LIST_HEAD(&ptype_base[i]); |
6258 | ||
881d966b EB |
6259 | if (register_pernet_subsys(&netdev_net_ops)) |
6260 | goto out; | |
1da177e4 LT |
6261 | |
6262 | /* | |
6263 | * Initialise the packet receive queues. | |
6264 | */ | |
6265 | ||
6f912042 | 6266 | for_each_possible_cpu(i) { |
e36fa2f7 | 6267 | struct softnet_data *sd = &per_cpu(softnet_data, i); |
1da177e4 | 6268 | |
dee42870 | 6269 | memset(sd, 0, sizeof(*sd)); |
e36fa2f7 | 6270 | skb_queue_head_init(&sd->input_pkt_queue); |
6e7676c1 | 6271 | skb_queue_head_init(&sd->process_queue); |
e36fa2f7 ED |
6272 | sd->completion_queue = NULL; |
6273 | INIT_LIST_HEAD(&sd->poll_list); | |
a9cbd588 CG |
6274 | sd->output_queue = NULL; |
6275 | sd->output_queue_tailp = &sd->output_queue; | |
df334545 | 6276 | #ifdef CONFIG_RPS |
e36fa2f7 ED |
6277 | sd->csd.func = rps_trigger_softirq; |
6278 | sd->csd.info = sd; | |
6279 | sd->csd.flags = 0; | |
6280 | sd->cpu = i; | |
1e94d72f | 6281 | #endif |
0a9627f2 | 6282 | |
e36fa2f7 ED |
6283 | sd->backlog.poll = process_backlog; |
6284 | sd->backlog.weight = weight_p; | |
6285 | sd->backlog.gro_list = NULL; | |
6286 | sd->backlog.gro_count = 0; | |
1da177e4 LT |
6287 | } |
6288 | ||
1da177e4 LT |
6289 | dev_boot_phase = 0; |
6290 | ||
505d4f73 EB |
6291 | /* The loopback device is special if any other network devices |
6292 | * is present in a network namespace the loopback device must | |
6293 | * be present. Since we now dynamically allocate and free the | |
6294 | * loopback device ensure this invariant is maintained by | |
6295 | * keeping the loopback device as the first device on the | |
6296 | * list of network devices. Ensuring the loopback devices | |
6297 | * is the first device that appears and the last network device | |
6298 | * that disappears. | |
6299 | */ | |
6300 | if (register_pernet_device(&loopback_net_ops)) | |
6301 | goto out; | |
6302 | ||
6303 | if (register_pernet_device(&default_device_ops)) | |
6304 | goto out; | |
6305 | ||
962cf36c CM |
6306 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
6307 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 LT |
6308 | |
6309 | hotcpu_notifier(dev_cpu_callback, 0); | |
6310 | dst_init(); | |
6311 | dev_mcast_init(); | |
6312 | rc = 0; | |
6313 | out: | |
6314 | return rc; | |
6315 | } | |
6316 | ||
6317 | subsys_initcall(net_dev_init); | |
6318 | ||
e88721f8 KK |
6319 | static int __init initialize_hashrnd(void) |
6320 | { | |
0a9627f2 | 6321 | get_random_bytes(&hashrnd, sizeof(hashrnd)); |
e88721f8 KK |
6322 | return 0; |
6323 | } | |
6324 | ||
6325 | late_initcall_sync(initialize_hashrnd); | |
6326 |