Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * NET3 Protocol independent device support routines. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public License | |
6 | * as published by the Free Software Foundation; either version | |
7 | * 2 of the License, or (at your option) any later version. | |
8 | * | |
9 | * Derived from the non IP parts of dev.c 1.0.19 | |
02c30a84 | 10 | * Authors: Ross Biro |
1da177e4 LT |
11 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
12 | * Mark Evans, <evansmp@uhura.aston.ac.uk> | |
13 | * | |
14 | * Additional Authors: | |
15 | * Florian la Roche <rzsfl@rz.uni-sb.de> | |
16 | * Alan Cox <gw4pts@gw4pts.ampr.org> | |
17 | * David Hinds <dahinds@users.sourceforge.net> | |
18 | * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> | |
19 | * Adam Sulmicki <adam@cfar.umd.edu> | |
20 | * Pekka Riikonen <priikone@poesidon.pspt.fi> | |
21 | * | |
22 | * Changes: | |
23 | * D.J. Barrow : Fixed bug where dev->refcnt gets set | |
24 | * to 2 if register_netdev gets called | |
25 | * before net_dev_init & also removed a | |
26 | * few lines of code in the process. | |
27 | * Alan Cox : device private ioctl copies fields back. | |
28 | * Alan Cox : Transmit queue code does relevant | |
29 | * stunts to keep the queue safe. | |
30 | * Alan Cox : Fixed double lock. | |
31 | * Alan Cox : Fixed promisc NULL pointer trap | |
32 | * ???????? : Support the full private ioctl range | |
33 | * Alan Cox : Moved ioctl permission check into | |
34 | * drivers | |
35 | * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI | |
36 | * Alan Cox : 100 backlog just doesn't cut it when | |
37 | * you start doing multicast video 8) | |
38 | * Alan Cox : Rewrote net_bh and list manager. | |
39 | * Alan Cox : Fix ETH_P_ALL echoback lengths. | |
40 | * Alan Cox : Took out transmit every packet pass | |
41 | * Saved a few bytes in the ioctl handler | |
42 | * Alan Cox : Network driver sets packet type before | |
43 | * calling netif_rx. Saves a function | |
44 | * call a packet. | |
45 | * Alan Cox : Hashed net_bh() | |
46 | * Richard Kooijman: Timestamp fixes. | |
47 | * Alan Cox : Wrong field in SIOCGIFDSTADDR | |
48 | * Alan Cox : Device lock protection. | |
49 | * Alan Cox : Fixed nasty side effect of device close | |
50 | * changes. | |
51 | * Rudi Cilibrasi : Pass the right thing to | |
52 | * set_mac_address() | |
53 | * Dave Miller : 32bit quantity for the device lock to | |
54 | * make it work out on a Sparc. | |
55 | * Bjorn Ekwall : Added KERNELD hack. | |
56 | * Alan Cox : Cleaned up the backlog initialise. | |
57 | * Craig Metz : SIOCGIFCONF fix if space for under | |
58 | * 1 device. | |
59 | * Thomas Bogendoerfer : Return ENODEV for dev_open, if there | |
60 | * is no device open function. | |
61 | * Andi Kleen : Fix error reporting for SIOCGIFCONF | |
62 | * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF | |
63 | * Cyrus Durgin : Cleaned for KMOD | |
64 | * Adam Sulmicki : Bug Fix : Network Device Unload | |
65 | * A network device unload needs to purge | |
66 | * the backlog queue. | |
67 | * Paul Rusty Russell : SIOCSIFNAME | |
68 | * Pekka Riikonen : Netdev boot-time settings code | |
69 | * Andrew Morton : Make unregister_netdevice wait | |
70 | * indefinitely on dev->refcnt | |
71 | * J Hadi Salim : - Backlog queue sampling | |
72 | * - netif_rx() feedback | |
73 | */ | |
74 | ||
75 | #include <asm/uaccess.h> | |
76 | #include <asm/system.h> | |
77 | #include <linux/bitops.h> | |
4fc268d2 | 78 | #include <linux/capability.h> |
1da177e4 LT |
79 | #include <linux/cpu.h> |
80 | #include <linux/types.h> | |
81 | #include <linux/kernel.h> | |
08e9897d | 82 | #include <linux/hash.h> |
5a0e3ad6 | 83 | #include <linux/slab.h> |
1da177e4 | 84 | #include <linux/sched.h> |
4a3e2f71 | 85 | #include <linux/mutex.h> |
1da177e4 LT |
86 | #include <linux/string.h> |
87 | #include <linux/mm.h> | |
88 | #include <linux/socket.h> | |
89 | #include <linux/sockios.h> | |
90 | #include <linux/errno.h> | |
91 | #include <linux/interrupt.h> | |
92 | #include <linux/if_ether.h> | |
93 | #include <linux/netdevice.h> | |
94 | #include <linux/etherdevice.h> | |
0187bdfb | 95 | #include <linux/ethtool.h> |
1da177e4 LT |
96 | #include <linux/notifier.h> |
97 | #include <linux/skbuff.h> | |
457c4cbc | 98 | #include <net/net_namespace.h> |
1da177e4 LT |
99 | #include <net/sock.h> |
100 | #include <linux/rtnetlink.h> | |
101 | #include <linux/proc_fs.h> | |
102 | #include <linux/seq_file.h> | |
103 | #include <linux/stat.h> | |
1da177e4 LT |
104 | #include <net/dst.h> |
105 | #include <net/pkt_sched.h> | |
106 | #include <net/checksum.h> | |
44540960 | 107 | #include <net/xfrm.h> |
1da177e4 LT |
108 | #include <linux/highmem.h> |
109 | #include <linux/init.h> | |
110 | #include <linux/kmod.h> | |
111 | #include <linux/module.h> | |
1da177e4 LT |
112 | #include <linux/netpoll.h> |
113 | #include <linux/rcupdate.h> | |
114 | #include <linux/delay.h> | |
295f4a1f | 115 | #include <net/wext.h> |
1da177e4 | 116 | #include <net/iw_handler.h> |
1da177e4 | 117 | #include <asm/current.h> |
5bdb9886 | 118 | #include <linux/audit.h> |
db217334 | 119 | #include <linux/dmaengine.h> |
f6a78bfc | 120 | #include <linux/err.h> |
c7fa9d18 | 121 | #include <linux/ctype.h> |
723e98b7 | 122 | #include <linux/if_arp.h> |
6de329e2 | 123 | #include <linux/if_vlan.h> |
8f0f2223 | 124 | #include <linux/ip.h> |
ad55dcaf | 125 | #include <net/ip.h> |
8f0f2223 DM |
126 | #include <linux/ipv6.h> |
127 | #include <linux/in.h> | |
b6b2fed1 DM |
128 | #include <linux/jhash.h> |
129 | #include <linux/random.h> | |
9cbc1cb8 | 130 | #include <trace/events/napi.h> |
cf66ba58 | 131 | #include <trace/events/net.h> |
07dc22e7 | 132 | #include <trace/events/skb.h> |
5acbbd42 | 133 | #include <linux/pci.h> |
caeda9b9 | 134 | #include <linux/inetdevice.h> |
c445477d | 135 | #include <linux/cpu_rmap.h> |
e971b722 | 136 | #include <linux/if_tunnel.h> |
ae1511bf | 137 | #include <linux/if_pppox.h> |
5dd17e08 | 138 | #include <linux/ppp_defs.h> |
1da177e4 | 139 | |
342709ef PE |
140 | #include "net-sysfs.h" |
141 | ||
d565b0a1 HX |
142 | /* Instead of increasing this, you should create a hash table. */ |
143 | #define MAX_GRO_SKBS 8 | |
144 | ||
5d38a079 HX |
145 | /* This should be increased if a protocol with a bigger head is added. */ |
146 | #define GRO_MAX_HEAD (MAX_HEADER + 128) | |
147 | ||
1da177e4 LT |
148 | /* |
149 | * The list of packet types we will receive (as opposed to discard) | |
150 | * and the routines to invoke. | |
151 | * | |
152 | * Why 16. Because with 16 the only overlap we get on a hash of the | |
153 | * low nibble of the protocol value is RARP/SNAP/X.25. | |
154 | * | |
155 | * NOTE: That is no longer true with the addition of VLAN tags. Not | |
156 | * sure which should go first, but I bet it won't make much | |
157 | * difference if we are running VLANs. The good news is that | |
158 | * this protocol won't be in the list unless compiled in, so | |
3041a069 | 159 | * the average user (w/out VLANs) will not be adversely affected. |
1da177e4 LT |
160 | * --BLG |
161 | * | |
162 | * 0800 IP | |
163 | * 8100 802.1Q VLAN | |
164 | * 0001 802.3 | |
165 | * 0002 AX.25 | |
166 | * 0004 802.2 | |
167 | * 8035 RARP | |
168 | * 0005 SNAP | |
169 | * 0805 X.25 | |
170 | * 0806 ARP | |
171 | * 8137 IPX | |
172 | * 0009 Localtalk | |
173 | * 86DD IPv6 | |
174 | */ | |
175 | ||
82d8a867 PE |
176 | #define PTYPE_HASH_SIZE (16) |
177 | #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1) | |
178 | ||
1da177e4 | 179 | static DEFINE_SPINLOCK(ptype_lock); |
82d8a867 | 180 | static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
6b2bedc3 | 181 | static struct list_head ptype_all __read_mostly; /* Taps */ |
1da177e4 | 182 | |
1da177e4 | 183 | /* |
7562f876 | 184 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
185 | * semaphore. |
186 | * | |
c6d14c84 | 187 | * Pure readers hold dev_base_lock for reading, or rcu_read_lock() |
1da177e4 LT |
188 | * |
189 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 190 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
191 | * actual updates. This allows pure readers to access the list even |
192 | * while a writer is preparing to update it. | |
193 | * | |
194 | * To put it another way, dev_base_lock is held for writing only to | |
195 | * protect against pure readers; the rtnl semaphore provides the | |
196 | * protection against other writers. | |
197 | * | |
198 | * See, for example usages, register_netdevice() and | |
199 | * unregister_netdevice(), which must be called with the rtnl | |
200 | * semaphore held. | |
201 | */ | |
1da177e4 | 202 | DEFINE_RWLOCK(dev_base_lock); |
1da177e4 LT |
203 | EXPORT_SYMBOL(dev_base_lock); |
204 | ||
4e985ada TG |
205 | static inline void dev_base_seq_inc(struct net *net) |
206 | { | |
207 | while (++net->dev_base_seq == 0); | |
208 | } | |
209 | ||
881d966b | 210 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 LT |
211 | { |
212 | unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ)); | |
08e9897d | 213 | return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; |
1da177e4 LT |
214 | } |
215 | ||
881d966b | 216 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 217 | { |
7c28bd0b | 218 | return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; |
1da177e4 LT |
219 | } |
220 | ||
e36fa2f7 | 221 | static inline void rps_lock(struct softnet_data *sd) |
152102c7 CG |
222 | { |
223 | #ifdef CONFIG_RPS | |
e36fa2f7 | 224 | spin_lock(&sd->input_pkt_queue.lock); |
152102c7 CG |
225 | #endif |
226 | } | |
227 | ||
e36fa2f7 | 228 | static inline void rps_unlock(struct softnet_data *sd) |
152102c7 CG |
229 | { |
230 | #ifdef CONFIG_RPS | |
e36fa2f7 | 231 | spin_unlock(&sd->input_pkt_queue.lock); |
152102c7 CG |
232 | #endif |
233 | } | |
234 | ||
ce286d32 EB |
235 | /* Device list insertion */ |
236 | static int list_netdevice(struct net_device *dev) | |
237 | { | |
c346dca1 | 238 | struct net *net = dev_net(dev); |
ce286d32 EB |
239 | |
240 | ASSERT_RTNL(); | |
241 | ||
242 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 243 | list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); |
72c9528b | 244 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); |
fb699dfd ED |
245 | hlist_add_head_rcu(&dev->index_hlist, |
246 | dev_index_hash(net, dev->ifindex)); | |
ce286d32 | 247 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
248 | |
249 | dev_base_seq_inc(net); | |
250 | ||
ce286d32 EB |
251 | return 0; |
252 | } | |
253 | ||
fb699dfd ED |
254 | /* Device list removal |
255 | * caller must respect a RCU grace period before freeing/reusing dev | |
256 | */ | |
ce286d32 EB |
257 | static void unlist_netdevice(struct net_device *dev) |
258 | { | |
259 | ASSERT_RTNL(); | |
260 | ||
261 | /* Unlink dev from the device chain */ | |
262 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 263 | list_del_rcu(&dev->dev_list); |
72c9528b | 264 | hlist_del_rcu(&dev->name_hlist); |
fb699dfd | 265 | hlist_del_rcu(&dev->index_hlist); |
ce286d32 | 266 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
267 | |
268 | dev_base_seq_inc(dev_net(dev)); | |
ce286d32 EB |
269 | } |
270 | ||
1da177e4 LT |
271 | /* |
272 | * Our notifier list | |
273 | */ | |
274 | ||
f07d5b94 | 275 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
276 | |
277 | /* | |
278 | * Device drivers call our routines to queue packets here. We empty the | |
279 | * queue in the local softnet handler. | |
280 | */ | |
bea3348e | 281 | |
9958da05 | 282 | DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); |
d1b19dff | 283 | EXPORT_PER_CPU_SYMBOL(softnet_data); |
1da177e4 | 284 | |
cf508b12 | 285 | #ifdef CONFIG_LOCKDEP |
723e98b7 | 286 | /* |
c773e847 | 287 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class |
723e98b7 JP |
288 | * according to dev->type |
289 | */ | |
290 | static const unsigned short netdev_lock_type[] = | |
291 | {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
292 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, | |
293 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
294 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
295 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
296 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
297 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
298 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
299 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
300 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
301 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
302 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
303 | ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211, | |
2d91d78b | 304 | ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, |
929122cd | 305 | ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154, |
fcb94e42 | 306 | ARPHRD_VOID, ARPHRD_NONE}; |
723e98b7 | 307 | |
36cbd3dc | 308 | static const char *const netdev_lock_name[] = |
723e98b7 JP |
309 | {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", |
310 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
311 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
312 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
313 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
314 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
315 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
316 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
317 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
318 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
319 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
320 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
321 | "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211", | |
2d91d78b | 322 | "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", |
929122cd | 323 | "_xmit_PHONET_PIPE", "_xmit_IEEE802154", |
fcb94e42 | 324 | "_xmit_VOID", "_xmit_NONE"}; |
723e98b7 JP |
325 | |
326 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
cf508b12 | 327 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; |
723e98b7 JP |
328 | |
329 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
330 | { | |
331 | int i; | |
332 | ||
333 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
334 | if (netdev_lock_type[i] == dev_type) | |
335 | return i; | |
336 | /* the last key is used by default */ | |
337 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
338 | } | |
339 | ||
cf508b12 DM |
340 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
341 | unsigned short dev_type) | |
723e98b7 JP |
342 | { |
343 | int i; | |
344 | ||
345 | i = netdev_lock_pos(dev_type); | |
346 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
347 | netdev_lock_name[i]); | |
348 | } | |
cf508b12 DM |
349 | |
350 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
351 | { | |
352 | int i; | |
353 | ||
354 | i = netdev_lock_pos(dev->type); | |
355 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
356 | &netdev_addr_lock_key[i], | |
357 | netdev_lock_name[i]); | |
358 | } | |
723e98b7 | 359 | #else |
cf508b12 DM |
360 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
361 | unsigned short dev_type) | |
362 | { | |
363 | } | |
364 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
723e98b7 JP |
365 | { |
366 | } | |
367 | #endif | |
1da177e4 LT |
368 | |
369 | /******************************************************************************* | |
370 | ||
371 | Protocol management and registration routines | |
372 | ||
373 | *******************************************************************************/ | |
374 | ||
1da177e4 LT |
375 | /* |
376 | * Add a protocol ID to the list. Now that the input handler is | |
377 | * smarter we can dispense with all the messy stuff that used to be | |
378 | * here. | |
379 | * | |
380 | * BEWARE!!! Protocol handlers, mangling input packets, | |
381 | * MUST BE last in hash buckets and checking protocol handlers | |
382 | * MUST start from promiscuous ptype_all chain in net_bh. | |
383 | * It is true now, do not change it. | |
384 | * Explanation follows: if protocol handler, mangling packet, will | |
385 | * be the first on list, it is not able to sense, that packet | |
386 | * is cloned and should be copied-on-write, so that it will | |
387 | * change it and subsequent readers will get broken packet. | |
388 | * --ANK (980803) | |
389 | */ | |
390 | ||
c07b68e8 ED |
391 | static inline struct list_head *ptype_head(const struct packet_type *pt) |
392 | { | |
393 | if (pt->type == htons(ETH_P_ALL)) | |
394 | return &ptype_all; | |
395 | else | |
396 | return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
397 | } | |
398 | ||
1da177e4 LT |
399 | /** |
400 | * dev_add_pack - add packet handler | |
401 | * @pt: packet type declaration | |
402 | * | |
403 | * Add a protocol handler to the networking stack. The passed &packet_type | |
404 | * is linked into kernel lists and may not be freed until it has been | |
405 | * removed from the kernel lists. | |
406 | * | |
4ec93edb | 407 | * This call does not sleep therefore it can not |
1da177e4 LT |
408 | * guarantee all CPU's that are in middle of receiving packets |
409 | * will see the new packet type (until the next received packet). | |
410 | */ | |
411 | ||
412 | void dev_add_pack(struct packet_type *pt) | |
413 | { | |
c07b68e8 | 414 | struct list_head *head = ptype_head(pt); |
1da177e4 | 415 | |
c07b68e8 ED |
416 | spin_lock(&ptype_lock); |
417 | list_add_rcu(&pt->list, head); | |
418 | spin_unlock(&ptype_lock); | |
1da177e4 | 419 | } |
d1b19dff | 420 | EXPORT_SYMBOL(dev_add_pack); |
1da177e4 | 421 | |
1da177e4 LT |
422 | /** |
423 | * __dev_remove_pack - remove packet handler | |
424 | * @pt: packet type declaration | |
425 | * | |
426 | * Remove a protocol handler that was previously added to the kernel | |
427 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
428 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 429 | * returns. |
1da177e4 LT |
430 | * |
431 | * The packet type might still be in use by receivers | |
432 | * and must not be freed until after all the CPU's have gone | |
433 | * through a quiescent state. | |
434 | */ | |
435 | void __dev_remove_pack(struct packet_type *pt) | |
436 | { | |
c07b68e8 | 437 | struct list_head *head = ptype_head(pt); |
1da177e4 LT |
438 | struct packet_type *pt1; |
439 | ||
c07b68e8 | 440 | spin_lock(&ptype_lock); |
1da177e4 LT |
441 | |
442 | list_for_each_entry(pt1, head, list) { | |
443 | if (pt == pt1) { | |
444 | list_del_rcu(&pt->list); | |
445 | goto out; | |
446 | } | |
447 | } | |
448 | ||
449 | printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt); | |
450 | out: | |
c07b68e8 | 451 | spin_unlock(&ptype_lock); |
1da177e4 | 452 | } |
d1b19dff ED |
453 | EXPORT_SYMBOL(__dev_remove_pack); |
454 | ||
1da177e4 LT |
455 | /** |
456 | * dev_remove_pack - remove packet handler | |
457 | * @pt: packet type declaration | |
458 | * | |
459 | * Remove a protocol handler that was previously added to the kernel | |
460 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
461 | * from the kernel lists and can be freed or reused once this function | |
462 | * returns. | |
463 | * | |
464 | * This call sleeps to guarantee that no CPU is looking at the packet | |
465 | * type after return. | |
466 | */ | |
467 | void dev_remove_pack(struct packet_type *pt) | |
468 | { | |
469 | __dev_remove_pack(pt); | |
4ec93edb | 470 | |
1da177e4 LT |
471 | synchronize_net(); |
472 | } | |
d1b19dff | 473 | EXPORT_SYMBOL(dev_remove_pack); |
1da177e4 LT |
474 | |
475 | /****************************************************************************** | |
476 | ||
477 | Device Boot-time Settings Routines | |
478 | ||
479 | *******************************************************************************/ | |
480 | ||
481 | /* Boot time configuration table */ | |
482 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
483 | ||
484 | /** | |
485 | * netdev_boot_setup_add - add new setup entry | |
486 | * @name: name of the device | |
487 | * @map: configured settings for the device | |
488 | * | |
489 | * Adds new setup entry to the dev_boot_setup list. The function | |
490 | * returns 0 on error and 1 on success. This is a generic routine to | |
491 | * all netdevices. | |
492 | */ | |
493 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
494 | { | |
495 | struct netdev_boot_setup *s; | |
496 | int i; | |
497 | ||
498 | s = dev_boot_setup; | |
499 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
500 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
501 | memset(s[i].name, 0, sizeof(s[i].name)); | |
93b3cff9 | 502 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
503 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
504 | break; | |
505 | } | |
506 | } | |
507 | ||
508 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
509 | } | |
510 | ||
511 | /** | |
512 | * netdev_boot_setup_check - check boot time settings | |
513 | * @dev: the netdevice | |
514 | * | |
515 | * Check boot time settings for the device. | |
516 | * The found settings are set for the device to be used | |
517 | * later in the device probing. | |
518 | * Returns 0 if no settings found, 1 if they are. | |
519 | */ | |
520 | int netdev_boot_setup_check(struct net_device *dev) | |
521 | { | |
522 | struct netdev_boot_setup *s = dev_boot_setup; | |
523 | int i; | |
524 | ||
525 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
526 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
93b3cff9 | 527 | !strcmp(dev->name, s[i].name)) { |
1da177e4 LT |
528 | dev->irq = s[i].map.irq; |
529 | dev->base_addr = s[i].map.base_addr; | |
530 | dev->mem_start = s[i].map.mem_start; | |
531 | dev->mem_end = s[i].map.mem_end; | |
532 | return 1; | |
533 | } | |
534 | } | |
535 | return 0; | |
536 | } | |
d1b19dff | 537 | EXPORT_SYMBOL(netdev_boot_setup_check); |
1da177e4 LT |
538 | |
539 | ||
540 | /** | |
541 | * netdev_boot_base - get address from boot time settings | |
542 | * @prefix: prefix for network device | |
543 | * @unit: id for network device | |
544 | * | |
545 | * Check boot time settings for the base address of device. | |
546 | * The found settings are set for the device to be used | |
547 | * later in the device probing. | |
548 | * Returns 0 if no settings found. | |
549 | */ | |
550 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
551 | { | |
552 | const struct netdev_boot_setup *s = dev_boot_setup; | |
553 | char name[IFNAMSIZ]; | |
554 | int i; | |
555 | ||
556 | sprintf(name, "%s%d", prefix, unit); | |
557 | ||
558 | /* | |
559 | * If device already registered then return base of 1 | |
560 | * to indicate not to probe for this interface | |
561 | */ | |
881d966b | 562 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
563 | return 1; |
564 | ||
565 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
566 | if (!strcmp(name, s[i].name)) | |
567 | return s[i].map.base_addr; | |
568 | return 0; | |
569 | } | |
570 | ||
571 | /* | |
572 | * Saves at boot time configured settings for any netdevice. | |
573 | */ | |
574 | int __init netdev_boot_setup(char *str) | |
575 | { | |
576 | int ints[5]; | |
577 | struct ifmap map; | |
578 | ||
579 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
580 | if (!str || !*str) | |
581 | return 0; | |
582 | ||
583 | /* Save settings */ | |
584 | memset(&map, 0, sizeof(map)); | |
585 | if (ints[0] > 0) | |
586 | map.irq = ints[1]; | |
587 | if (ints[0] > 1) | |
588 | map.base_addr = ints[2]; | |
589 | if (ints[0] > 2) | |
590 | map.mem_start = ints[3]; | |
591 | if (ints[0] > 3) | |
592 | map.mem_end = ints[4]; | |
593 | ||
594 | /* Add new entry to the list */ | |
595 | return netdev_boot_setup_add(str, &map); | |
596 | } | |
597 | ||
598 | __setup("netdev=", netdev_boot_setup); | |
599 | ||
600 | /******************************************************************************* | |
601 | ||
602 | Device Interface Subroutines | |
603 | ||
604 | *******************************************************************************/ | |
605 | ||
606 | /** | |
607 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 608 | * @net: the applicable net namespace |
1da177e4 LT |
609 | * @name: name to find |
610 | * | |
611 | * Find an interface by name. Must be called under RTNL semaphore | |
612 | * or @dev_base_lock. If the name is found a pointer to the device | |
613 | * is returned. If the name is not found then %NULL is returned. The | |
614 | * reference counters are not incremented so the caller must be | |
615 | * careful with locks. | |
616 | */ | |
617 | ||
881d966b | 618 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
619 | { |
620 | struct hlist_node *p; | |
0bd8d536 ED |
621 | struct net_device *dev; |
622 | struct hlist_head *head = dev_name_hash(net, name); | |
1da177e4 | 623 | |
0bd8d536 | 624 | hlist_for_each_entry(dev, p, head, name_hlist) |
1da177e4 LT |
625 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
626 | return dev; | |
0bd8d536 | 627 | |
1da177e4 LT |
628 | return NULL; |
629 | } | |
d1b19dff | 630 | EXPORT_SYMBOL(__dev_get_by_name); |
1da177e4 | 631 | |
72c9528b ED |
632 | /** |
633 | * dev_get_by_name_rcu - find a device by its name | |
634 | * @net: the applicable net namespace | |
635 | * @name: name to find | |
636 | * | |
637 | * Find an interface by name. | |
638 | * If the name is found a pointer to the device is returned. | |
639 | * If the name is not found then %NULL is returned. | |
640 | * The reference counters are not incremented so the caller must be | |
641 | * careful with locks. The caller must hold RCU lock. | |
642 | */ | |
643 | ||
644 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
645 | { | |
646 | struct hlist_node *p; | |
647 | struct net_device *dev; | |
648 | struct hlist_head *head = dev_name_hash(net, name); | |
649 | ||
650 | hlist_for_each_entry_rcu(dev, p, head, name_hlist) | |
651 | if (!strncmp(dev->name, name, IFNAMSIZ)) | |
652 | return dev; | |
653 | ||
654 | return NULL; | |
655 | } | |
656 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
657 | ||
1da177e4 LT |
658 | /** |
659 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 660 | * @net: the applicable net namespace |
1da177e4 LT |
661 | * @name: name to find |
662 | * | |
663 | * Find an interface by name. This can be called from any | |
664 | * context and does its own locking. The returned handle has | |
665 | * the usage count incremented and the caller must use dev_put() to | |
666 | * release it when it is no longer needed. %NULL is returned if no | |
667 | * matching device is found. | |
668 | */ | |
669 | ||
881d966b | 670 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
671 | { |
672 | struct net_device *dev; | |
673 | ||
72c9528b ED |
674 | rcu_read_lock(); |
675 | dev = dev_get_by_name_rcu(net, name); | |
1da177e4 LT |
676 | if (dev) |
677 | dev_hold(dev); | |
72c9528b | 678 | rcu_read_unlock(); |
1da177e4 LT |
679 | return dev; |
680 | } | |
d1b19dff | 681 | EXPORT_SYMBOL(dev_get_by_name); |
1da177e4 LT |
682 | |
683 | /** | |
684 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 685 | * @net: the applicable net namespace |
1da177e4 LT |
686 | * @ifindex: index of device |
687 | * | |
688 | * Search for an interface by index. Returns %NULL if the device | |
689 | * is not found or a pointer to the device. The device has not | |
690 | * had its reference counter increased so the caller must be careful | |
691 | * about locking. The caller must hold either the RTNL semaphore | |
692 | * or @dev_base_lock. | |
693 | */ | |
694 | ||
881d966b | 695 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
696 | { |
697 | struct hlist_node *p; | |
0bd8d536 ED |
698 | struct net_device *dev; |
699 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
1da177e4 | 700 | |
0bd8d536 | 701 | hlist_for_each_entry(dev, p, head, index_hlist) |
1da177e4 LT |
702 | if (dev->ifindex == ifindex) |
703 | return dev; | |
0bd8d536 | 704 | |
1da177e4 LT |
705 | return NULL; |
706 | } | |
d1b19dff | 707 | EXPORT_SYMBOL(__dev_get_by_index); |
1da177e4 | 708 | |
fb699dfd ED |
709 | /** |
710 | * dev_get_by_index_rcu - find a device by its ifindex | |
711 | * @net: the applicable net namespace | |
712 | * @ifindex: index of device | |
713 | * | |
714 | * Search for an interface by index. Returns %NULL if the device | |
715 | * is not found or a pointer to the device. The device has not | |
716 | * had its reference counter increased so the caller must be careful | |
717 | * about locking. The caller must hold RCU lock. | |
718 | */ | |
719 | ||
720 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
721 | { | |
722 | struct hlist_node *p; | |
723 | struct net_device *dev; | |
724 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
725 | ||
726 | hlist_for_each_entry_rcu(dev, p, head, index_hlist) | |
727 | if (dev->ifindex == ifindex) | |
728 | return dev; | |
729 | ||
730 | return NULL; | |
731 | } | |
732 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
733 | ||
1da177e4 LT |
734 | |
735 | /** | |
736 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 737 | * @net: the applicable net namespace |
1da177e4 LT |
738 | * @ifindex: index of device |
739 | * | |
740 | * Search for an interface by index. Returns NULL if the device | |
741 | * is not found or a pointer to the device. The device returned has | |
742 | * had a reference added and the pointer is safe until the user calls | |
743 | * dev_put to indicate they have finished with it. | |
744 | */ | |
745 | ||
881d966b | 746 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
747 | { |
748 | struct net_device *dev; | |
749 | ||
fb699dfd ED |
750 | rcu_read_lock(); |
751 | dev = dev_get_by_index_rcu(net, ifindex); | |
1da177e4 LT |
752 | if (dev) |
753 | dev_hold(dev); | |
fb699dfd | 754 | rcu_read_unlock(); |
1da177e4 LT |
755 | return dev; |
756 | } | |
d1b19dff | 757 | EXPORT_SYMBOL(dev_get_by_index); |
1da177e4 LT |
758 | |
759 | /** | |
941666c2 | 760 | * dev_getbyhwaddr_rcu - find a device by its hardware address |
c4ea43c5 | 761 | * @net: the applicable net namespace |
1da177e4 LT |
762 | * @type: media type of device |
763 | * @ha: hardware address | |
764 | * | |
765 | * Search for an interface by MAC address. Returns NULL if the device | |
c506653d ED |
766 | * is not found or a pointer to the device. |
767 | * The caller must hold RCU or RTNL. | |
941666c2 | 768 | * The returned device has not had its ref count increased |
1da177e4 LT |
769 | * and the caller must therefore be careful about locking |
770 | * | |
1da177e4 LT |
771 | */ |
772 | ||
941666c2 ED |
773 | struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, |
774 | const char *ha) | |
1da177e4 LT |
775 | { |
776 | struct net_device *dev; | |
777 | ||
941666c2 | 778 | for_each_netdev_rcu(net, dev) |
1da177e4 LT |
779 | if (dev->type == type && |
780 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
781 | return dev; |
782 | ||
783 | return NULL; | |
1da177e4 | 784 | } |
941666c2 | 785 | EXPORT_SYMBOL(dev_getbyhwaddr_rcu); |
cf309e3f | 786 | |
881d966b | 787 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
788 | { |
789 | struct net_device *dev; | |
790 | ||
4e9cac2b | 791 | ASSERT_RTNL(); |
881d966b | 792 | for_each_netdev(net, dev) |
4e9cac2b | 793 | if (dev->type == type) |
7562f876 PE |
794 | return dev; |
795 | ||
796 | return NULL; | |
4e9cac2b | 797 | } |
4e9cac2b PM |
798 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); |
799 | ||
881d966b | 800 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b | 801 | { |
99fe3c39 | 802 | struct net_device *dev, *ret = NULL; |
4e9cac2b | 803 | |
99fe3c39 ED |
804 | rcu_read_lock(); |
805 | for_each_netdev_rcu(net, dev) | |
806 | if (dev->type == type) { | |
807 | dev_hold(dev); | |
808 | ret = dev; | |
809 | break; | |
810 | } | |
811 | rcu_read_unlock(); | |
812 | return ret; | |
1da177e4 | 813 | } |
1da177e4 LT |
814 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
815 | ||
816 | /** | |
bb69ae04 | 817 | * dev_get_by_flags_rcu - find any device with given flags |
c4ea43c5 | 818 | * @net: the applicable net namespace |
1da177e4 LT |
819 | * @if_flags: IFF_* values |
820 | * @mask: bitmask of bits in if_flags to check | |
821 | * | |
822 | * Search for any interface with the given flags. Returns NULL if a device | |
bb69ae04 ED |
823 | * is not found or a pointer to the device. Must be called inside |
824 | * rcu_read_lock(), and result refcount is unchanged. | |
1da177e4 LT |
825 | */ |
826 | ||
bb69ae04 | 827 | struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags, |
d1b19dff | 828 | unsigned short mask) |
1da177e4 | 829 | { |
7562f876 | 830 | struct net_device *dev, *ret; |
1da177e4 | 831 | |
7562f876 | 832 | ret = NULL; |
c6d14c84 | 833 | for_each_netdev_rcu(net, dev) { |
1da177e4 | 834 | if (((dev->flags ^ if_flags) & mask) == 0) { |
7562f876 | 835 | ret = dev; |
1da177e4 LT |
836 | break; |
837 | } | |
838 | } | |
7562f876 | 839 | return ret; |
1da177e4 | 840 | } |
bb69ae04 | 841 | EXPORT_SYMBOL(dev_get_by_flags_rcu); |
1da177e4 LT |
842 | |
843 | /** | |
844 | * dev_valid_name - check if name is okay for network device | |
845 | * @name: name string | |
846 | * | |
847 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
848 | * to allow sysfs to work. We also disallow any kind of |
849 | * whitespace. | |
1da177e4 | 850 | */ |
c2373ee9 | 851 | int dev_valid_name(const char *name) |
1da177e4 | 852 | { |
c7fa9d18 DM |
853 | if (*name == '\0') |
854 | return 0; | |
b6fe17d6 SH |
855 | if (strlen(name) >= IFNAMSIZ) |
856 | return 0; | |
c7fa9d18 DM |
857 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
858 | return 0; | |
859 | ||
860 | while (*name) { | |
861 | if (*name == '/' || isspace(*name)) | |
862 | return 0; | |
863 | name++; | |
864 | } | |
865 | return 1; | |
1da177e4 | 866 | } |
d1b19dff | 867 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
868 | |
869 | /** | |
b267b179 EB |
870 | * __dev_alloc_name - allocate a name for a device |
871 | * @net: network namespace to allocate the device name in | |
1da177e4 | 872 | * @name: name format string |
b267b179 | 873 | * @buf: scratch buffer and result name string |
1da177e4 LT |
874 | * |
875 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
876 | * id. It scans list of devices to build up a free map, then chooses |
877 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
878 | * while allocating the name and adding the device in order to avoid | |
879 | * duplicates. | |
880 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
881 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
882 | */ |
883 | ||
b267b179 | 884 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
885 | { |
886 | int i = 0; | |
1da177e4 LT |
887 | const char *p; |
888 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 889 | unsigned long *inuse; |
1da177e4 LT |
890 | struct net_device *d; |
891 | ||
892 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
893 | if (p) { | |
894 | /* | |
895 | * Verify the string as this thing may have come from | |
896 | * the user. There must be either one "%d" and no other "%" | |
897 | * characters. | |
898 | */ | |
899 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
900 | return -EINVAL; | |
901 | ||
902 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 903 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
904 | if (!inuse) |
905 | return -ENOMEM; | |
906 | ||
881d966b | 907 | for_each_netdev(net, d) { |
1da177e4 LT |
908 | if (!sscanf(d->name, name, &i)) |
909 | continue; | |
910 | if (i < 0 || i >= max_netdevices) | |
911 | continue; | |
912 | ||
913 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 914 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
915 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
916 | set_bit(i, inuse); | |
917 | } | |
918 | ||
919 | i = find_first_zero_bit(inuse, max_netdevices); | |
920 | free_page((unsigned long) inuse); | |
921 | } | |
922 | ||
d9031024 OP |
923 | if (buf != name) |
924 | snprintf(buf, IFNAMSIZ, name, i); | |
b267b179 | 925 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 926 | return i; |
1da177e4 LT |
927 | |
928 | /* It is possible to run out of possible slots | |
929 | * when the name is long and there isn't enough space left | |
930 | * for the digits, or if all bits are used. | |
931 | */ | |
932 | return -ENFILE; | |
933 | } | |
934 | ||
b267b179 EB |
935 | /** |
936 | * dev_alloc_name - allocate a name for a device | |
937 | * @dev: device | |
938 | * @name: name format string | |
939 | * | |
940 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
941 | * id. It scans list of devices to build up a free map, then chooses | |
942 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
943 | * while allocating the name and adding the device in order to avoid | |
944 | * duplicates. | |
945 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
946 | * Returns the number of the unit assigned or a negative errno code. | |
947 | */ | |
948 | ||
949 | int dev_alloc_name(struct net_device *dev, const char *name) | |
950 | { | |
951 | char buf[IFNAMSIZ]; | |
952 | struct net *net; | |
953 | int ret; | |
954 | ||
c346dca1 YH |
955 | BUG_ON(!dev_net(dev)); |
956 | net = dev_net(dev); | |
b267b179 EB |
957 | ret = __dev_alloc_name(net, name, buf); |
958 | if (ret >= 0) | |
959 | strlcpy(dev->name, buf, IFNAMSIZ); | |
960 | return ret; | |
961 | } | |
d1b19dff | 962 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 963 | |
1c5cae81 | 964 | static int dev_get_valid_name(struct net_device *dev, const char *name) |
d9031024 | 965 | { |
8ce6cebc DL |
966 | struct net *net; |
967 | ||
968 | BUG_ON(!dev_net(dev)); | |
969 | net = dev_net(dev); | |
970 | ||
d9031024 OP |
971 | if (!dev_valid_name(name)) |
972 | return -EINVAL; | |
973 | ||
1c5cae81 | 974 | if (strchr(name, '%')) |
8ce6cebc | 975 | return dev_alloc_name(dev, name); |
d9031024 OP |
976 | else if (__dev_get_by_name(net, name)) |
977 | return -EEXIST; | |
8ce6cebc DL |
978 | else if (dev->name != name) |
979 | strlcpy(dev->name, name, IFNAMSIZ); | |
d9031024 OP |
980 | |
981 | return 0; | |
982 | } | |
1da177e4 LT |
983 | |
984 | /** | |
985 | * dev_change_name - change name of a device | |
986 | * @dev: device | |
987 | * @newname: name (or format string) must be at least IFNAMSIZ | |
988 | * | |
989 | * Change name of a device, can pass format strings "eth%d". | |
990 | * for wildcarding. | |
991 | */ | |
cf04a4c7 | 992 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 993 | { |
fcc5a03a | 994 | char oldname[IFNAMSIZ]; |
1da177e4 | 995 | int err = 0; |
fcc5a03a | 996 | int ret; |
881d966b | 997 | struct net *net; |
1da177e4 LT |
998 | |
999 | ASSERT_RTNL(); | |
c346dca1 | 1000 | BUG_ON(!dev_net(dev)); |
1da177e4 | 1001 | |
c346dca1 | 1002 | net = dev_net(dev); |
1da177e4 LT |
1003 | if (dev->flags & IFF_UP) |
1004 | return -EBUSY; | |
1005 | ||
c8d90dca SH |
1006 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) |
1007 | return 0; | |
1008 | ||
fcc5a03a HX |
1009 | memcpy(oldname, dev->name, IFNAMSIZ); |
1010 | ||
1c5cae81 | 1011 | err = dev_get_valid_name(dev, newname); |
d9031024 OP |
1012 | if (err < 0) |
1013 | return err; | |
1da177e4 | 1014 | |
fcc5a03a | 1015 | rollback: |
a1b3f594 EB |
1016 | ret = device_rename(&dev->dev, dev->name); |
1017 | if (ret) { | |
1018 | memcpy(dev->name, oldname, IFNAMSIZ); | |
1019 | return ret; | |
dcc99773 | 1020 | } |
7f988eab HX |
1021 | |
1022 | write_lock_bh(&dev_base_lock); | |
372b2312 | 1023 | hlist_del_rcu(&dev->name_hlist); |
72c9528b ED |
1024 | write_unlock_bh(&dev_base_lock); |
1025 | ||
1026 | synchronize_rcu(); | |
1027 | ||
1028 | write_lock_bh(&dev_base_lock); | |
1029 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
7f988eab HX |
1030 | write_unlock_bh(&dev_base_lock); |
1031 | ||
056925ab | 1032 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1033 | ret = notifier_to_errno(ret); |
1034 | ||
1035 | if (ret) { | |
91e9c07b ED |
1036 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1037 | if (err >= 0) { | |
fcc5a03a HX |
1038 | err = ret; |
1039 | memcpy(dev->name, oldname, IFNAMSIZ); | |
1040 | goto rollback; | |
91e9c07b ED |
1041 | } else { |
1042 | printk(KERN_ERR | |
1043 | "%s: name change rollback failed: %d.\n", | |
1044 | dev->name, ret); | |
fcc5a03a HX |
1045 | } |
1046 | } | |
1da177e4 LT |
1047 | |
1048 | return err; | |
1049 | } | |
1050 | ||
0b815a1a SH |
1051 | /** |
1052 | * dev_set_alias - change ifalias of a device | |
1053 | * @dev: device | |
1054 | * @alias: name up to IFALIASZ | |
f0db275a | 1055 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1056 | * |
1057 | * Set ifalias for a device, | |
1058 | */ | |
1059 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1060 | { | |
1061 | ASSERT_RTNL(); | |
1062 | ||
1063 | if (len >= IFALIASZ) | |
1064 | return -EINVAL; | |
1065 | ||
96ca4a2c OH |
1066 | if (!len) { |
1067 | if (dev->ifalias) { | |
1068 | kfree(dev->ifalias); | |
1069 | dev->ifalias = NULL; | |
1070 | } | |
1071 | return 0; | |
1072 | } | |
1073 | ||
d1b19dff | 1074 | dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL); |
0b815a1a SH |
1075 | if (!dev->ifalias) |
1076 | return -ENOMEM; | |
1077 | ||
1078 | strlcpy(dev->ifalias, alias, len+1); | |
1079 | return len; | |
1080 | } | |
1081 | ||
1082 | ||
d8a33ac4 | 1083 | /** |
3041a069 | 1084 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1085 | * @dev: device to cause notification |
1086 | * | |
1087 | * Called to indicate a device has changed features. | |
1088 | */ | |
1089 | void netdev_features_change(struct net_device *dev) | |
1090 | { | |
056925ab | 1091 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1092 | } |
1093 | EXPORT_SYMBOL(netdev_features_change); | |
1094 | ||
1da177e4 LT |
1095 | /** |
1096 | * netdev_state_change - device changes state | |
1097 | * @dev: device to cause notification | |
1098 | * | |
1099 | * Called to indicate a device has changed state. This function calls | |
1100 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1101 | * to the routing socket. | |
1102 | */ | |
1103 | void netdev_state_change(struct net_device *dev) | |
1104 | { | |
1105 | if (dev->flags & IFF_UP) { | |
056925ab | 1106 | call_netdevice_notifiers(NETDEV_CHANGE, dev); |
1da177e4 LT |
1107 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0); |
1108 | } | |
1109 | } | |
d1b19dff | 1110 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1111 | |
3ca5b404 | 1112 | int netdev_bonding_change(struct net_device *dev, unsigned long event) |
c1da4ac7 | 1113 | { |
3ca5b404 | 1114 | return call_netdevice_notifiers(event, dev); |
c1da4ac7 OG |
1115 | } |
1116 | EXPORT_SYMBOL(netdev_bonding_change); | |
1117 | ||
1da177e4 LT |
1118 | /** |
1119 | * dev_load - load a network module | |
c4ea43c5 | 1120 | * @net: the applicable net namespace |
1da177e4 LT |
1121 | * @name: name of interface |
1122 | * | |
1123 | * If a network interface is not present and the process has suitable | |
1124 | * privileges this function loads the module. If module loading is not | |
1125 | * available in this kernel then it becomes a nop. | |
1126 | */ | |
1127 | ||
881d966b | 1128 | void dev_load(struct net *net, const char *name) |
1da177e4 | 1129 | { |
4ec93edb | 1130 | struct net_device *dev; |
8909c9ad | 1131 | int no_module; |
1da177e4 | 1132 | |
72c9528b ED |
1133 | rcu_read_lock(); |
1134 | dev = dev_get_by_name_rcu(net, name); | |
1135 | rcu_read_unlock(); | |
1da177e4 | 1136 | |
8909c9ad VK |
1137 | no_module = !dev; |
1138 | if (no_module && capable(CAP_NET_ADMIN)) | |
1139 | no_module = request_module("netdev-%s", name); | |
1140 | if (no_module && capable(CAP_SYS_MODULE)) { | |
1141 | if (!request_module("%s", name)) | |
1142 | pr_err("Loading kernel module for a network device " | |
1143 | "with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s " | |
1144 | "instead\n", name); | |
1145 | } | |
1da177e4 | 1146 | } |
d1b19dff | 1147 | EXPORT_SYMBOL(dev_load); |
1da177e4 | 1148 | |
bd380811 | 1149 | static int __dev_open(struct net_device *dev) |
1da177e4 | 1150 | { |
d314774c | 1151 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1152 | int ret; |
1da177e4 | 1153 | |
e46b66bc BH |
1154 | ASSERT_RTNL(); |
1155 | ||
1da177e4 LT |
1156 | if (!netif_device_present(dev)) |
1157 | return -ENODEV; | |
1158 | ||
3b8bcfd5 JB |
1159 | ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); |
1160 | ret = notifier_to_errno(ret); | |
1161 | if (ret) | |
1162 | return ret; | |
1163 | ||
1da177e4 | 1164 | set_bit(__LINK_STATE_START, &dev->state); |
bada339b | 1165 | |
d314774c SH |
1166 | if (ops->ndo_validate_addr) |
1167 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1168 | |
d314774c SH |
1169 | if (!ret && ops->ndo_open) |
1170 | ret = ops->ndo_open(dev); | |
1da177e4 | 1171 | |
bada339b JG |
1172 | if (ret) |
1173 | clear_bit(__LINK_STATE_START, &dev->state); | |
1174 | else { | |
1da177e4 | 1175 | dev->flags |= IFF_UP; |
b4bd07c2 | 1176 | net_dmaengine_get(); |
4417da66 | 1177 | dev_set_rx_mode(dev); |
1da177e4 | 1178 | dev_activate(dev); |
1da177e4 | 1179 | } |
bada339b | 1180 | |
1da177e4 LT |
1181 | return ret; |
1182 | } | |
1183 | ||
1184 | /** | |
bd380811 PM |
1185 | * dev_open - prepare an interface for use. |
1186 | * @dev: device to open | |
1da177e4 | 1187 | * |
bd380811 PM |
1188 | * Takes a device from down to up state. The device's private open |
1189 | * function is invoked and then the multicast lists are loaded. Finally | |
1190 | * the device is moved into the up state and a %NETDEV_UP message is | |
1191 | * sent to the netdev notifier chain. | |
1192 | * | |
1193 | * Calling this function on an active interface is a nop. On a failure | |
1194 | * a negative errno code is returned. | |
1da177e4 | 1195 | */ |
bd380811 PM |
1196 | int dev_open(struct net_device *dev) |
1197 | { | |
1198 | int ret; | |
1199 | ||
bd380811 PM |
1200 | if (dev->flags & IFF_UP) |
1201 | return 0; | |
1202 | ||
bd380811 PM |
1203 | ret = __dev_open(dev); |
1204 | if (ret < 0) | |
1205 | return ret; | |
1206 | ||
bd380811 PM |
1207 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING); |
1208 | call_netdevice_notifiers(NETDEV_UP, dev); | |
1209 | ||
1210 | return ret; | |
1211 | } | |
1212 | EXPORT_SYMBOL(dev_open); | |
1213 | ||
44345724 | 1214 | static int __dev_close_many(struct list_head *head) |
1da177e4 | 1215 | { |
44345724 | 1216 | struct net_device *dev; |
e46b66bc | 1217 | |
bd380811 | 1218 | ASSERT_RTNL(); |
9d5010db DM |
1219 | might_sleep(); |
1220 | ||
44345724 | 1221 | list_for_each_entry(dev, head, unreg_list) { |
44345724 | 1222 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1223 | |
44345724 | 1224 | clear_bit(__LINK_STATE_START, &dev->state); |
1da177e4 | 1225 | |
44345724 OP |
1226 | /* Synchronize to scheduled poll. We cannot touch poll list, it |
1227 | * can be even on different cpu. So just clear netif_running(). | |
1228 | * | |
1229 | * dev->stop() will invoke napi_disable() on all of it's | |
1230 | * napi_struct instances on this device. | |
1231 | */ | |
1232 | smp_mb__after_clear_bit(); /* Commit netif_running(). */ | |
1233 | } | |
1da177e4 | 1234 | |
44345724 | 1235 | dev_deactivate_many(head); |
d8b2a4d2 | 1236 | |
44345724 OP |
1237 | list_for_each_entry(dev, head, unreg_list) { |
1238 | const struct net_device_ops *ops = dev->netdev_ops; | |
1da177e4 | 1239 | |
44345724 OP |
1240 | /* |
1241 | * Call the device specific close. This cannot fail. | |
1242 | * Only if device is UP | |
1243 | * | |
1244 | * We allow it to be called even after a DETACH hot-plug | |
1245 | * event. | |
1246 | */ | |
1247 | if (ops->ndo_stop) | |
1248 | ops->ndo_stop(dev); | |
1249 | ||
44345724 | 1250 | dev->flags &= ~IFF_UP; |
44345724 OP |
1251 | net_dmaengine_put(); |
1252 | } | |
1253 | ||
1254 | return 0; | |
1255 | } | |
1256 | ||
1257 | static int __dev_close(struct net_device *dev) | |
1258 | { | |
f87e6f47 | 1259 | int retval; |
44345724 OP |
1260 | LIST_HEAD(single); |
1261 | ||
1262 | list_add(&dev->unreg_list, &single); | |
f87e6f47 LT |
1263 | retval = __dev_close_many(&single); |
1264 | list_del(&single); | |
1265 | return retval; | |
44345724 OP |
1266 | } |
1267 | ||
3fbd8758 | 1268 | static int dev_close_many(struct list_head *head) |
44345724 OP |
1269 | { |
1270 | struct net_device *dev, *tmp; | |
1271 | LIST_HEAD(tmp_list); | |
1da177e4 | 1272 | |
44345724 OP |
1273 | list_for_each_entry_safe(dev, tmp, head, unreg_list) |
1274 | if (!(dev->flags & IFF_UP)) | |
1275 | list_move(&dev->unreg_list, &tmp_list); | |
1276 | ||
1277 | __dev_close_many(head); | |
1da177e4 | 1278 | |
44345724 OP |
1279 | list_for_each_entry(dev, head, unreg_list) { |
1280 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING); | |
1281 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
1282 | } | |
bd380811 | 1283 | |
44345724 OP |
1284 | /* rollback_registered_many needs the complete original list */ |
1285 | list_splice(&tmp_list, head); | |
bd380811 PM |
1286 | return 0; |
1287 | } | |
1288 | ||
1289 | /** | |
1290 | * dev_close - shutdown an interface. | |
1291 | * @dev: device to shutdown | |
1292 | * | |
1293 | * This function moves an active device into down state. A | |
1294 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1295 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1296 | * chain. | |
1297 | */ | |
1298 | int dev_close(struct net_device *dev) | |
1299 | { | |
e14a5993 ED |
1300 | if (dev->flags & IFF_UP) { |
1301 | LIST_HEAD(single); | |
1da177e4 | 1302 | |
e14a5993 ED |
1303 | list_add(&dev->unreg_list, &single); |
1304 | dev_close_many(&single); | |
1305 | list_del(&single); | |
1306 | } | |
1da177e4 LT |
1307 | return 0; |
1308 | } | |
d1b19dff | 1309 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1310 | |
1311 | ||
0187bdfb BH |
1312 | /** |
1313 | * dev_disable_lro - disable Large Receive Offload on a device | |
1314 | * @dev: device | |
1315 | * | |
1316 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1317 | * called under RTNL. This is needed if received packets may be | |
1318 | * forwarded to another interface. | |
1319 | */ | |
1320 | void dev_disable_lro(struct net_device *dev) | |
1321 | { | |
27660515 MM |
1322 | u32 flags; |
1323 | ||
f11970e3 NH |
1324 | /* |
1325 | * If we're trying to disable lro on a vlan device | |
1326 | * use the underlying physical device instead | |
1327 | */ | |
1328 | if (is_vlan_dev(dev)) | |
1329 | dev = vlan_dev_real_dev(dev); | |
1330 | ||
27660515 MM |
1331 | if (dev->ethtool_ops && dev->ethtool_ops->get_flags) |
1332 | flags = dev->ethtool_ops->get_flags(dev); | |
1333 | else | |
1334 | flags = ethtool_op_get_flags(dev); | |
1335 | ||
1336 | if (!(flags & ETH_FLAG_LRO)) | |
1337 | return; | |
1338 | ||
1339 | __ethtool_set_flags(dev, flags & ~ETH_FLAG_LRO); | |
22d5969f MM |
1340 | if (unlikely(dev->features & NETIF_F_LRO)) |
1341 | netdev_WARN(dev, "failed to disable LRO!\n"); | |
0187bdfb BH |
1342 | } |
1343 | EXPORT_SYMBOL(dev_disable_lro); | |
1344 | ||
1345 | ||
881d966b EB |
1346 | static int dev_boot_phase = 1; |
1347 | ||
1da177e4 LT |
1348 | /** |
1349 | * register_netdevice_notifier - register a network notifier block | |
1350 | * @nb: notifier | |
1351 | * | |
1352 | * Register a notifier to be called when network device events occur. | |
1353 | * The notifier passed is linked into the kernel structures and must | |
1354 | * not be reused until it has been unregistered. A negative errno code | |
1355 | * is returned on a failure. | |
1356 | * | |
1357 | * When registered all registration and up events are replayed | |
4ec93edb | 1358 | * to the new notifier to allow device to have a race free |
1da177e4 LT |
1359 | * view of the network device list. |
1360 | */ | |
1361 | ||
1362 | int register_netdevice_notifier(struct notifier_block *nb) | |
1363 | { | |
1364 | struct net_device *dev; | |
fcc5a03a | 1365 | struct net_device *last; |
881d966b | 1366 | struct net *net; |
1da177e4 LT |
1367 | int err; |
1368 | ||
1369 | rtnl_lock(); | |
f07d5b94 | 1370 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1371 | if (err) |
1372 | goto unlock; | |
881d966b EB |
1373 | if (dev_boot_phase) |
1374 | goto unlock; | |
1375 | for_each_net(net) { | |
1376 | for_each_netdev(net, dev) { | |
1377 | err = nb->notifier_call(nb, NETDEV_REGISTER, dev); | |
1378 | err = notifier_to_errno(err); | |
1379 | if (err) | |
1380 | goto rollback; | |
1381 | ||
1382 | if (!(dev->flags & IFF_UP)) | |
1383 | continue; | |
1da177e4 | 1384 | |
881d966b EB |
1385 | nb->notifier_call(nb, NETDEV_UP, dev); |
1386 | } | |
1da177e4 | 1387 | } |
fcc5a03a HX |
1388 | |
1389 | unlock: | |
1da177e4 LT |
1390 | rtnl_unlock(); |
1391 | return err; | |
fcc5a03a HX |
1392 | |
1393 | rollback: | |
1394 | last = dev; | |
881d966b EB |
1395 | for_each_net(net) { |
1396 | for_each_netdev(net, dev) { | |
1397 | if (dev == last) | |
1398 | break; | |
fcc5a03a | 1399 | |
881d966b EB |
1400 | if (dev->flags & IFF_UP) { |
1401 | nb->notifier_call(nb, NETDEV_GOING_DOWN, dev); | |
1402 | nb->notifier_call(nb, NETDEV_DOWN, dev); | |
1403 | } | |
1404 | nb->notifier_call(nb, NETDEV_UNREGISTER, dev); | |
a5ee1551 | 1405 | nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev); |
fcc5a03a | 1406 | } |
fcc5a03a | 1407 | } |
c67625a1 PE |
1408 | |
1409 | raw_notifier_chain_unregister(&netdev_chain, nb); | |
fcc5a03a | 1410 | goto unlock; |
1da177e4 | 1411 | } |
d1b19dff | 1412 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1413 | |
1414 | /** | |
1415 | * unregister_netdevice_notifier - unregister a network notifier block | |
1416 | * @nb: notifier | |
1417 | * | |
1418 | * Unregister a notifier previously registered by | |
1419 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1420 | * kernel structures and may then be reused. A negative errno code | |
1421 | * is returned on a failure. | |
1422 | */ | |
1423 | ||
1424 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1425 | { | |
9f514950 HX |
1426 | int err; |
1427 | ||
1428 | rtnl_lock(); | |
f07d5b94 | 1429 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
9f514950 HX |
1430 | rtnl_unlock(); |
1431 | return err; | |
1da177e4 | 1432 | } |
d1b19dff | 1433 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 LT |
1434 | |
1435 | /** | |
1436 | * call_netdevice_notifiers - call all network notifier blocks | |
1437 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1438 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1439 | * |
1440 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1441 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1442 | */ |
1443 | ||
ad7379d4 | 1444 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1445 | { |
ab930471 | 1446 | ASSERT_RTNL(); |
ad7379d4 | 1447 | return raw_notifier_call_chain(&netdev_chain, val, dev); |
1da177e4 | 1448 | } |
edf947f1 | 1449 | EXPORT_SYMBOL(call_netdevice_notifiers); |
1da177e4 LT |
1450 | |
1451 | /* When > 0 there are consumers of rx skb time stamps */ | |
1452 | static atomic_t netstamp_needed = ATOMIC_INIT(0); | |
1453 | ||
1454 | void net_enable_timestamp(void) | |
1455 | { | |
1456 | atomic_inc(&netstamp_needed); | |
1457 | } | |
d1b19dff | 1458 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
1459 | |
1460 | void net_disable_timestamp(void) | |
1461 | { | |
1462 | atomic_dec(&netstamp_needed); | |
1463 | } | |
d1b19dff | 1464 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 1465 | |
3b098e2d | 1466 | static inline void net_timestamp_set(struct sk_buff *skb) |
1da177e4 LT |
1467 | { |
1468 | if (atomic_read(&netstamp_needed)) | |
a61bbcf2 | 1469 | __net_timestamp(skb); |
b7aa0bf7 ED |
1470 | else |
1471 | skb->tstamp.tv64 = 0; | |
1da177e4 LT |
1472 | } |
1473 | ||
3b098e2d ED |
1474 | static inline void net_timestamp_check(struct sk_buff *skb) |
1475 | { | |
1476 | if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed)) | |
1477 | __net_timestamp(skb); | |
1478 | } | |
1479 | ||
79b569f0 DL |
1480 | static inline bool is_skb_forwardable(struct net_device *dev, |
1481 | struct sk_buff *skb) | |
1482 | { | |
1483 | unsigned int len; | |
1484 | ||
1485 | if (!(dev->flags & IFF_UP)) | |
1486 | return false; | |
1487 | ||
1488 | len = dev->mtu + dev->hard_header_len + VLAN_HLEN; | |
1489 | if (skb->len <= len) | |
1490 | return true; | |
1491 | ||
1492 | /* if TSO is enabled, we don't care about the length as the packet | |
1493 | * could be forwarded without being segmented before | |
1494 | */ | |
1495 | if (skb_is_gso(skb)) | |
1496 | return true; | |
1497 | ||
1498 | return false; | |
1499 | } | |
1500 | ||
44540960 AB |
1501 | /** |
1502 | * dev_forward_skb - loopback an skb to another netif | |
1503 | * | |
1504 | * @dev: destination network device | |
1505 | * @skb: buffer to forward | |
1506 | * | |
1507 | * return values: | |
1508 | * NET_RX_SUCCESS (no congestion) | |
6ec82562 | 1509 | * NET_RX_DROP (packet was dropped, but freed) |
44540960 AB |
1510 | * |
1511 | * dev_forward_skb can be used for injecting an skb from the | |
1512 | * start_xmit function of one device into the receive queue | |
1513 | * of another device. | |
1514 | * | |
1515 | * The receiving device may be in another namespace, so | |
1516 | * we have to clear all information in the skb that could | |
1517 | * impact namespace isolation. | |
1518 | */ | |
1519 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
1520 | { | |
48c83012 MT |
1521 | if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) { |
1522 | if (skb_copy_ubufs(skb, GFP_ATOMIC)) { | |
1523 | atomic_long_inc(&dev->rx_dropped); | |
1524 | kfree_skb(skb); | |
1525 | return NET_RX_DROP; | |
1526 | } | |
1527 | } | |
1528 | ||
44540960 | 1529 | skb_orphan(skb); |
c736eefa | 1530 | nf_reset(skb); |
44540960 | 1531 | |
79b569f0 | 1532 | if (unlikely(!is_skb_forwardable(dev, skb))) { |
caf586e5 | 1533 | atomic_long_inc(&dev->rx_dropped); |
6ec82562 | 1534 | kfree_skb(skb); |
44540960 | 1535 | return NET_RX_DROP; |
6ec82562 | 1536 | } |
8a83a00b | 1537 | skb_set_dev(skb, dev); |
44540960 AB |
1538 | skb->tstamp.tv64 = 0; |
1539 | skb->pkt_type = PACKET_HOST; | |
1540 | skb->protocol = eth_type_trans(skb, dev); | |
44540960 AB |
1541 | return netif_rx(skb); |
1542 | } | |
1543 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
1544 | ||
71d9dec2 CG |
1545 | static inline int deliver_skb(struct sk_buff *skb, |
1546 | struct packet_type *pt_prev, | |
1547 | struct net_device *orig_dev) | |
1548 | { | |
1549 | atomic_inc(&skb->users); | |
1550 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1551 | } | |
1552 | ||
1da177e4 LT |
1553 | /* |
1554 | * Support routine. Sends outgoing frames to any network | |
1555 | * taps currently in use. | |
1556 | */ | |
1557 | ||
f6a78bfc | 1558 | static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
1559 | { |
1560 | struct packet_type *ptype; | |
71d9dec2 CG |
1561 | struct sk_buff *skb2 = NULL; |
1562 | struct packet_type *pt_prev = NULL; | |
a61bbcf2 | 1563 | |
1da177e4 LT |
1564 | rcu_read_lock(); |
1565 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
1566 | /* Never send packets back to the socket | |
1567 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1568 | */ | |
1569 | if ((ptype->dev == dev || !ptype->dev) && | |
1570 | (ptype->af_packet_priv == NULL || | |
1571 | (struct sock *)ptype->af_packet_priv != skb->sk)) { | |
71d9dec2 CG |
1572 | if (pt_prev) { |
1573 | deliver_skb(skb2, pt_prev, skb->dev); | |
1574 | pt_prev = ptype; | |
1575 | continue; | |
1576 | } | |
1577 | ||
1578 | skb2 = skb_clone(skb, GFP_ATOMIC); | |
1da177e4 LT |
1579 | if (!skb2) |
1580 | break; | |
1581 | ||
70978182 ED |
1582 | net_timestamp_set(skb2); |
1583 | ||
1da177e4 LT |
1584 | /* skb->nh should be correctly |
1585 | set by sender, so that the second statement is | |
1586 | just protection against buggy protocols. | |
1587 | */ | |
459a98ed | 1588 | skb_reset_mac_header(skb2); |
1da177e4 | 1589 | |
d56f90a7 | 1590 | if (skb_network_header(skb2) < skb2->data || |
27a884dc | 1591 | skb2->network_header > skb2->tail) { |
1da177e4 LT |
1592 | if (net_ratelimit()) |
1593 | printk(KERN_CRIT "protocol %04x is " | |
1594 | "buggy, dev %s\n", | |
70777d03 SAS |
1595 | ntohs(skb2->protocol), |
1596 | dev->name); | |
c1d2bbe1 | 1597 | skb_reset_network_header(skb2); |
1da177e4 LT |
1598 | } |
1599 | ||
b0e380b1 | 1600 | skb2->transport_header = skb2->network_header; |
1da177e4 | 1601 | skb2->pkt_type = PACKET_OUTGOING; |
71d9dec2 | 1602 | pt_prev = ptype; |
1da177e4 LT |
1603 | } |
1604 | } | |
71d9dec2 CG |
1605 | if (pt_prev) |
1606 | pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); | |
1da177e4 LT |
1607 | rcu_read_unlock(); |
1608 | } | |
1609 | ||
4f57c087 JF |
1610 | /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change |
1611 | * @dev: Network device | |
1612 | * @txq: number of queues available | |
1613 | * | |
1614 | * If real_num_tx_queues is changed the tc mappings may no longer be | |
1615 | * valid. To resolve this verify the tc mapping remains valid and if | |
1616 | * not NULL the mapping. With no priorities mapping to this | |
1617 | * offset/count pair it will no longer be used. In the worst case TC0 | |
1618 | * is invalid nothing can be done so disable priority mappings. If is | |
1619 | * expected that drivers will fix this mapping if they can before | |
1620 | * calling netif_set_real_num_tx_queues. | |
1621 | */ | |
bb134d22 | 1622 | static void netif_setup_tc(struct net_device *dev, unsigned int txq) |
4f57c087 JF |
1623 | { |
1624 | int i; | |
1625 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
1626 | ||
1627 | /* If TC0 is invalidated disable TC mapping */ | |
1628 | if (tc->offset + tc->count > txq) { | |
1629 | pr_warning("Number of in use tx queues changed " | |
1630 | "invalidating tc mappings. Priority " | |
1631 | "traffic classification disabled!\n"); | |
1632 | dev->num_tc = 0; | |
1633 | return; | |
1634 | } | |
1635 | ||
1636 | /* Invalidated prio to tc mappings set to TC0 */ | |
1637 | for (i = 1; i < TC_BITMASK + 1; i++) { | |
1638 | int q = netdev_get_prio_tc_map(dev, i); | |
1639 | ||
1640 | tc = &dev->tc_to_txq[q]; | |
1641 | if (tc->offset + tc->count > txq) { | |
1642 | pr_warning("Number of in use tx queues " | |
1643 | "changed. Priority %i to tc " | |
1644 | "mapping %i is no longer valid " | |
1645 | "setting map to 0\n", | |
1646 | i, q); | |
1647 | netdev_set_prio_tc_map(dev, i, 0); | |
1648 | } | |
1649 | } | |
1650 | } | |
1651 | ||
f0796d5c JF |
1652 | /* |
1653 | * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues | |
1654 | * greater then real_num_tx_queues stale skbs on the qdisc must be flushed. | |
1655 | */ | |
e6484930 | 1656 | int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) |
f0796d5c | 1657 | { |
1d24eb48 TH |
1658 | int rc; |
1659 | ||
e6484930 TH |
1660 | if (txq < 1 || txq > dev->num_tx_queues) |
1661 | return -EINVAL; | |
f0796d5c | 1662 | |
5c56580b BH |
1663 | if (dev->reg_state == NETREG_REGISTERED || |
1664 | dev->reg_state == NETREG_UNREGISTERING) { | |
e6484930 TH |
1665 | ASSERT_RTNL(); |
1666 | ||
1d24eb48 TH |
1667 | rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, |
1668 | txq); | |
bf264145 TH |
1669 | if (rc) |
1670 | return rc; | |
1671 | ||
4f57c087 JF |
1672 | if (dev->num_tc) |
1673 | netif_setup_tc(dev, txq); | |
1674 | ||
e6484930 TH |
1675 | if (txq < dev->real_num_tx_queues) |
1676 | qdisc_reset_all_tx_gt(dev, txq); | |
f0796d5c | 1677 | } |
e6484930 TH |
1678 | |
1679 | dev->real_num_tx_queues = txq; | |
1680 | return 0; | |
f0796d5c JF |
1681 | } |
1682 | EXPORT_SYMBOL(netif_set_real_num_tx_queues); | |
56079431 | 1683 | |
62fe0b40 BH |
1684 | #ifdef CONFIG_RPS |
1685 | /** | |
1686 | * netif_set_real_num_rx_queues - set actual number of RX queues used | |
1687 | * @dev: Network device | |
1688 | * @rxq: Actual number of RX queues | |
1689 | * | |
1690 | * This must be called either with the rtnl_lock held or before | |
1691 | * registration of the net device. Returns 0 on success, or a | |
4e7f7951 BH |
1692 | * negative error code. If called before registration, it always |
1693 | * succeeds. | |
62fe0b40 BH |
1694 | */ |
1695 | int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) | |
1696 | { | |
1697 | int rc; | |
1698 | ||
bd25fa7b TH |
1699 | if (rxq < 1 || rxq > dev->num_rx_queues) |
1700 | return -EINVAL; | |
1701 | ||
62fe0b40 BH |
1702 | if (dev->reg_state == NETREG_REGISTERED) { |
1703 | ASSERT_RTNL(); | |
1704 | ||
62fe0b40 BH |
1705 | rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, |
1706 | rxq); | |
1707 | if (rc) | |
1708 | return rc; | |
62fe0b40 BH |
1709 | } |
1710 | ||
1711 | dev->real_num_rx_queues = rxq; | |
1712 | return 0; | |
1713 | } | |
1714 | EXPORT_SYMBOL(netif_set_real_num_rx_queues); | |
1715 | #endif | |
1716 | ||
def82a1d | 1717 | static inline void __netif_reschedule(struct Qdisc *q) |
56079431 | 1718 | { |
def82a1d JP |
1719 | struct softnet_data *sd; |
1720 | unsigned long flags; | |
56079431 | 1721 | |
def82a1d JP |
1722 | local_irq_save(flags); |
1723 | sd = &__get_cpu_var(softnet_data); | |
a9cbd588 CG |
1724 | q->next_sched = NULL; |
1725 | *sd->output_queue_tailp = q; | |
1726 | sd->output_queue_tailp = &q->next_sched; | |
def82a1d JP |
1727 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
1728 | local_irq_restore(flags); | |
1729 | } | |
1730 | ||
1731 | void __netif_schedule(struct Qdisc *q) | |
1732 | { | |
1733 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
1734 | __netif_reschedule(q); | |
56079431 DV |
1735 | } |
1736 | EXPORT_SYMBOL(__netif_schedule); | |
1737 | ||
bea3348e | 1738 | void dev_kfree_skb_irq(struct sk_buff *skb) |
56079431 | 1739 | { |
3578b0c8 | 1740 | if (atomic_dec_and_test(&skb->users)) { |
bea3348e SH |
1741 | struct softnet_data *sd; |
1742 | unsigned long flags; | |
56079431 | 1743 | |
bea3348e SH |
1744 | local_irq_save(flags); |
1745 | sd = &__get_cpu_var(softnet_data); | |
1746 | skb->next = sd->completion_queue; | |
1747 | sd->completion_queue = skb; | |
1748 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
1749 | local_irq_restore(flags); | |
1750 | } | |
56079431 | 1751 | } |
bea3348e | 1752 | EXPORT_SYMBOL(dev_kfree_skb_irq); |
56079431 DV |
1753 | |
1754 | void dev_kfree_skb_any(struct sk_buff *skb) | |
1755 | { | |
1756 | if (in_irq() || irqs_disabled()) | |
1757 | dev_kfree_skb_irq(skb); | |
1758 | else | |
1759 | dev_kfree_skb(skb); | |
1760 | } | |
1761 | EXPORT_SYMBOL(dev_kfree_skb_any); | |
1762 | ||
1763 | ||
bea3348e SH |
1764 | /** |
1765 | * netif_device_detach - mark device as removed | |
1766 | * @dev: network device | |
1767 | * | |
1768 | * Mark device as removed from system and therefore no longer available. | |
1769 | */ | |
56079431 DV |
1770 | void netif_device_detach(struct net_device *dev) |
1771 | { | |
1772 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
1773 | netif_running(dev)) { | |
d543103a | 1774 | netif_tx_stop_all_queues(dev); |
56079431 DV |
1775 | } |
1776 | } | |
1777 | EXPORT_SYMBOL(netif_device_detach); | |
1778 | ||
bea3348e SH |
1779 | /** |
1780 | * netif_device_attach - mark device as attached | |
1781 | * @dev: network device | |
1782 | * | |
1783 | * Mark device as attached from system and restart if needed. | |
1784 | */ | |
56079431 DV |
1785 | void netif_device_attach(struct net_device *dev) |
1786 | { | |
1787 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
1788 | netif_running(dev)) { | |
d543103a | 1789 | netif_tx_wake_all_queues(dev); |
4ec93edb | 1790 | __netdev_watchdog_up(dev); |
56079431 DV |
1791 | } |
1792 | } | |
1793 | EXPORT_SYMBOL(netif_device_attach); | |
1794 | ||
8a83a00b AB |
1795 | /** |
1796 | * skb_dev_set -- assign a new device to a buffer | |
1797 | * @skb: buffer for the new device | |
1798 | * @dev: network device | |
1799 | * | |
1800 | * If an skb is owned by a device already, we have to reset | |
1801 | * all data private to the namespace a device belongs to | |
1802 | * before assigning it a new device. | |
1803 | */ | |
1804 | #ifdef CONFIG_NET_NS | |
1805 | void skb_set_dev(struct sk_buff *skb, struct net_device *dev) | |
1806 | { | |
1807 | skb_dst_drop(skb); | |
1808 | if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) { | |
1809 | secpath_reset(skb); | |
1810 | nf_reset(skb); | |
1811 | skb_init_secmark(skb); | |
1812 | skb->mark = 0; | |
1813 | skb->priority = 0; | |
1814 | skb->nf_trace = 0; | |
1815 | skb->ipvs_property = 0; | |
1816 | #ifdef CONFIG_NET_SCHED | |
1817 | skb->tc_index = 0; | |
1818 | #endif | |
1819 | } | |
1820 | skb->dev = dev; | |
1821 | } | |
1822 | EXPORT_SYMBOL(skb_set_dev); | |
1823 | #endif /* CONFIG_NET_NS */ | |
1824 | ||
1da177e4 LT |
1825 | /* |
1826 | * Invalidate hardware checksum when packet is to be mangled, and | |
1827 | * complete checksum manually on outgoing path. | |
1828 | */ | |
84fa7933 | 1829 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 1830 | { |
d3bc23e7 | 1831 | __wsum csum; |
663ead3b | 1832 | int ret = 0, offset; |
1da177e4 | 1833 | |
84fa7933 | 1834 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
1835 | goto out_set_summed; |
1836 | ||
1837 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
a430a43d HX |
1838 | /* Let GSO fix up the checksum. */ |
1839 | goto out_set_summed; | |
1da177e4 LT |
1840 | } |
1841 | ||
55508d60 | 1842 | offset = skb_checksum_start_offset(skb); |
a030847e HX |
1843 | BUG_ON(offset >= skb_headlen(skb)); |
1844 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
1845 | ||
1846 | offset += skb->csum_offset; | |
1847 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
1848 | ||
1849 | if (skb_cloned(skb) && | |
1850 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1da177e4 LT |
1851 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
1852 | if (ret) | |
1853 | goto out; | |
1854 | } | |
1855 | ||
a030847e | 1856 | *(__sum16 *)(skb->data + offset) = csum_fold(csum); |
a430a43d | 1857 | out_set_summed: |
1da177e4 | 1858 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 1859 | out: |
1da177e4 LT |
1860 | return ret; |
1861 | } | |
d1b19dff | 1862 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 1863 | |
f6a78bfc HX |
1864 | /** |
1865 | * skb_gso_segment - Perform segmentation on skb. | |
1866 | * @skb: buffer to segment | |
576a30eb | 1867 | * @features: features for the output path (see dev->features) |
f6a78bfc HX |
1868 | * |
1869 | * This function segments the given skb and returns a list of segments. | |
576a30eb HX |
1870 | * |
1871 | * It may return NULL if the skb requires no segmentation. This is | |
1872 | * only possible when GSO is used for verifying header integrity. | |
f6a78bfc | 1873 | */ |
04ed3e74 | 1874 | struct sk_buff *skb_gso_segment(struct sk_buff *skb, u32 features) |
f6a78bfc HX |
1875 | { |
1876 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
1877 | struct packet_type *ptype; | |
252e3346 | 1878 | __be16 type = skb->protocol; |
c8d5bcd1 | 1879 | int vlan_depth = ETH_HLEN; |
a430a43d | 1880 | int err; |
f6a78bfc | 1881 | |
c8d5bcd1 JG |
1882 | while (type == htons(ETH_P_8021Q)) { |
1883 | struct vlan_hdr *vh; | |
7b9c6090 | 1884 | |
c8d5bcd1 | 1885 | if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN))) |
7b9c6090 JG |
1886 | return ERR_PTR(-EINVAL); |
1887 | ||
c8d5bcd1 JG |
1888 | vh = (struct vlan_hdr *)(skb->data + vlan_depth); |
1889 | type = vh->h_vlan_encapsulated_proto; | |
1890 | vlan_depth += VLAN_HLEN; | |
7b9c6090 JG |
1891 | } |
1892 | ||
459a98ed | 1893 | skb_reset_mac_header(skb); |
b0e380b1 | 1894 | skb->mac_len = skb->network_header - skb->mac_header; |
f6a78bfc HX |
1895 | __skb_pull(skb, skb->mac_len); |
1896 | ||
67fd1a73 HX |
1897 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { |
1898 | struct net_device *dev = skb->dev; | |
1899 | struct ethtool_drvinfo info = {}; | |
1900 | ||
1901 | if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo) | |
1902 | dev->ethtool_ops->get_drvinfo(dev, &info); | |
1903 | ||
b194a367 | 1904 | WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n", |
67fd1a73 HX |
1905 | info.driver, dev ? dev->features : 0L, |
1906 | skb->sk ? skb->sk->sk_route_caps : 0L, | |
1907 | skb->len, skb->data_len, skb->ip_summed); | |
1908 | ||
a430a43d HX |
1909 | if (skb_header_cloned(skb) && |
1910 | (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))) | |
1911 | return ERR_PTR(err); | |
1912 | } | |
1913 | ||
f6a78bfc | 1914 | rcu_read_lock(); |
82d8a867 PE |
1915 | list_for_each_entry_rcu(ptype, |
1916 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
f6a78bfc | 1917 | if (ptype->type == type && !ptype->dev && ptype->gso_segment) { |
84fa7933 | 1918 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { |
a430a43d HX |
1919 | err = ptype->gso_send_check(skb); |
1920 | segs = ERR_PTR(err); | |
1921 | if (err || skb_gso_ok(skb, features)) | |
1922 | break; | |
d56f90a7 ACM |
1923 | __skb_push(skb, (skb->data - |
1924 | skb_network_header(skb))); | |
a430a43d | 1925 | } |
576a30eb | 1926 | segs = ptype->gso_segment(skb, features); |
f6a78bfc HX |
1927 | break; |
1928 | } | |
1929 | } | |
1930 | rcu_read_unlock(); | |
1931 | ||
98e399f8 | 1932 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 1933 | |
f6a78bfc HX |
1934 | return segs; |
1935 | } | |
f6a78bfc HX |
1936 | EXPORT_SYMBOL(skb_gso_segment); |
1937 | ||
fb286bb2 HX |
1938 | /* Take action when hardware reception checksum errors are detected. */ |
1939 | #ifdef CONFIG_BUG | |
1940 | void netdev_rx_csum_fault(struct net_device *dev) | |
1941 | { | |
1942 | if (net_ratelimit()) { | |
4ec93edb | 1943 | printk(KERN_ERR "%s: hw csum failure.\n", |
246a4212 | 1944 | dev ? dev->name : "<unknown>"); |
fb286bb2 HX |
1945 | dump_stack(); |
1946 | } | |
1947 | } | |
1948 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
1949 | #endif | |
1950 | ||
1da177e4 LT |
1951 | /* Actually, we should eliminate this check as soon as we know, that: |
1952 | * 1. IOMMU is present and allows to map all the memory. | |
1953 | * 2. No high memory really exists on this machine. | |
1954 | */ | |
1955 | ||
9092c658 | 1956 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
1da177e4 | 1957 | { |
3d3a8533 | 1958 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 1959 | int i; |
5acbbd42 | 1960 | if (!(dev->features & NETIF_F_HIGHDMA)) { |
ea2ab693 IC |
1961 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
1962 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
1963 | if (PageHighMem(skb_frag_page(frag))) | |
5acbbd42 | 1964 | return 1; |
ea2ab693 | 1965 | } |
5acbbd42 | 1966 | } |
1da177e4 | 1967 | |
5acbbd42 FT |
1968 | if (PCI_DMA_BUS_IS_PHYS) { |
1969 | struct device *pdev = dev->dev.parent; | |
1da177e4 | 1970 | |
9092c658 ED |
1971 | if (!pdev) |
1972 | return 0; | |
5acbbd42 | 1973 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
ea2ab693 IC |
1974 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
1975 | dma_addr_t addr = page_to_phys(skb_frag_page(frag)); | |
5acbbd42 FT |
1976 | if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask) |
1977 | return 1; | |
1978 | } | |
1979 | } | |
3d3a8533 | 1980 | #endif |
1da177e4 LT |
1981 | return 0; |
1982 | } | |
1da177e4 | 1983 | |
f6a78bfc HX |
1984 | struct dev_gso_cb { |
1985 | void (*destructor)(struct sk_buff *skb); | |
1986 | }; | |
1987 | ||
1988 | #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb) | |
1989 | ||
1990 | static void dev_gso_skb_destructor(struct sk_buff *skb) | |
1991 | { | |
1992 | struct dev_gso_cb *cb; | |
1993 | ||
1994 | do { | |
1995 | struct sk_buff *nskb = skb->next; | |
1996 | ||
1997 | skb->next = nskb->next; | |
1998 | nskb->next = NULL; | |
1999 | kfree_skb(nskb); | |
2000 | } while (skb->next); | |
2001 | ||
2002 | cb = DEV_GSO_CB(skb); | |
2003 | if (cb->destructor) | |
2004 | cb->destructor(skb); | |
2005 | } | |
2006 | ||
2007 | /** | |
2008 | * dev_gso_segment - Perform emulated hardware segmentation on skb. | |
2009 | * @skb: buffer to segment | |
91ecb63c | 2010 | * @features: device features as applicable to this skb |
f6a78bfc HX |
2011 | * |
2012 | * This function segments the given skb and stores the list of segments | |
2013 | * in skb->next. | |
2014 | */ | |
91ecb63c | 2015 | static int dev_gso_segment(struct sk_buff *skb, int features) |
f6a78bfc | 2016 | { |
f6a78bfc | 2017 | struct sk_buff *segs; |
576a30eb HX |
2018 | |
2019 | segs = skb_gso_segment(skb, features); | |
2020 | ||
2021 | /* Verifying header integrity only. */ | |
2022 | if (!segs) | |
2023 | return 0; | |
f6a78bfc | 2024 | |
801678c5 | 2025 | if (IS_ERR(segs)) |
f6a78bfc HX |
2026 | return PTR_ERR(segs); |
2027 | ||
2028 | skb->next = segs; | |
2029 | DEV_GSO_CB(skb)->destructor = skb->destructor; | |
2030 | skb->destructor = dev_gso_skb_destructor; | |
2031 | ||
2032 | return 0; | |
2033 | } | |
2034 | ||
fc6055a5 ED |
2035 | /* |
2036 | * Try to orphan skb early, right before transmission by the device. | |
2244d07b OH |
2037 | * We cannot orphan skb if tx timestamp is requested or the sk-reference |
2038 | * is needed on driver level for other reasons, e.g. see net/can/raw.c | |
fc6055a5 ED |
2039 | */ |
2040 | static inline void skb_orphan_try(struct sk_buff *skb) | |
2041 | { | |
87fd308c ED |
2042 | struct sock *sk = skb->sk; |
2043 | ||
2244d07b | 2044 | if (sk && !skb_shinfo(skb)->tx_flags) { |
87fd308c ED |
2045 | /* skb_tx_hash() wont be able to get sk. |
2046 | * We copy sk_hash into skb->rxhash | |
2047 | */ | |
2048 | if (!skb->rxhash) | |
2049 | skb->rxhash = sk->sk_hash; | |
fc6055a5 | 2050 | skb_orphan(skb); |
87fd308c | 2051 | } |
fc6055a5 ED |
2052 | } |
2053 | ||
03634668 JG |
2054 | static bool can_checksum_protocol(unsigned long features, __be16 protocol) |
2055 | { | |
2056 | return ((features & NETIF_F_GEN_CSUM) || | |
2057 | ((features & NETIF_F_V4_CSUM) && | |
2058 | protocol == htons(ETH_P_IP)) || | |
2059 | ((features & NETIF_F_V6_CSUM) && | |
2060 | protocol == htons(ETH_P_IPV6)) || | |
2061 | ((features & NETIF_F_FCOE_CRC) && | |
2062 | protocol == htons(ETH_P_FCOE))); | |
2063 | } | |
2064 | ||
04ed3e74 | 2065 | static u32 harmonize_features(struct sk_buff *skb, __be16 protocol, u32 features) |
f01a5236 | 2066 | { |
d402786e | 2067 | if (!can_checksum_protocol(features, protocol)) { |
f01a5236 JG |
2068 | features &= ~NETIF_F_ALL_CSUM; |
2069 | features &= ~NETIF_F_SG; | |
2070 | } else if (illegal_highdma(skb->dev, skb)) { | |
2071 | features &= ~NETIF_F_SG; | |
2072 | } | |
2073 | ||
2074 | return features; | |
2075 | } | |
2076 | ||
04ed3e74 | 2077 | u32 netif_skb_features(struct sk_buff *skb) |
58e998c6 JG |
2078 | { |
2079 | __be16 protocol = skb->protocol; | |
04ed3e74 | 2080 | u32 features = skb->dev->features; |
58e998c6 JG |
2081 | |
2082 | if (protocol == htons(ETH_P_8021Q)) { | |
2083 | struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data; | |
2084 | protocol = veh->h_vlan_encapsulated_proto; | |
f01a5236 JG |
2085 | } else if (!vlan_tx_tag_present(skb)) { |
2086 | return harmonize_features(skb, protocol, features); | |
2087 | } | |
58e998c6 | 2088 | |
6ee400aa | 2089 | features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX); |
f01a5236 JG |
2090 | |
2091 | if (protocol != htons(ETH_P_8021Q)) { | |
2092 | return harmonize_features(skb, protocol, features); | |
2093 | } else { | |
2094 | features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | | |
6ee400aa | 2095 | NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX; |
f01a5236 JG |
2096 | return harmonize_features(skb, protocol, features); |
2097 | } | |
58e998c6 | 2098 | } |
f01a5236 | 2099 | EXPORT_SYMBOL(netif_skb_features); |
58e998c6 | 2100 | |
6afff0ca JF |
2101 | /* |
2102 | * Returns true if either: | |
2103 | * 1. skb has frag_list and the device doesn't support FRAGLIST, or | |
2104 | * 2. skb is fragmented and the device does not support SG, or if | |
2105 | * at least one of fragments is in highmem and device does not | |
2106 | * support DMA from it. | |
2107 | */ | |
2108 | static inline int skb_needs_linearize(struct sk_buff *skb, | |
02932ce9 | 2109 | int features) |
6afff0ca | 2110 | { |
02932ce9 JG |
2111 | return skb_is_nonlinear(skb) && |
2112 | ((skb_has_frag_list(skb) && | |
2113 | !(features & NETIF_F_FRAGLIST)) || | |
e1e78db6 | 2114 | (skb_shinfo(skb)->nr_frags && |
02932ce9 | 2115 | !(features & NETIF_F_SG))); |
6afff0ca JF |
2116 | } |
2117 | ||
fd2ea0a7 DM |
2118 | int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev, |
2119 | struct netdev_queue *txq) | |
f6a78bfc | 2120 | { |
00829823 | 2121 | const struct net_device_ops *ops = dev->netdev_ops; |
572a9d7b | 2122 | int rc = NETDEV_TX_OK; |
ec764bf0 | 2123 | unsigned int skb_len; |
00829823 | 2124 | |
f6a78bfc | 2125 | if (likely(!skb->next)) { |
04ed3e74 | 2126 | u32 features; |
fc741216 | 2127 | |
93f154b5 | 2128 | /* |
25985edc | 2129 | * If device doesn't need skb->dst, release it right now while |
93f154b5 ED |
2130 | * its hot in this cpu cache |
2131 | */ | |
adf30907 ED |
2132 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) |
2133 | skb_dst_drop(skb); | |
2134 | ||
15c2d75f ED |
2135 | if (!list_empty(&ptype_all)) |
2136 | dev_queue_xmit_nit(skb, dev); | |
2137 | ||
fc6055a5 | 2138 | skb_orphan_try(skb); |
9ccb8975 | 2139 | |
fc741216 JG |
2140 | features = netif_skb_features(skb); |
2141 | ||
7b9c6090 | 2142 | if (vlan_tx_tag_present(skb) && |
fc741216 | 2143 | !(features & NETIF_F_HW_VLAN_TX)) { |
7b9c6090 JG |
2144 | skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb)); |
2145 | if (unlikely(!skb)) | |
2146 | goto out; | |
2147 | ||
2148 | skb->vlan_tci = 0; | |
2149 | } | |
2150 | ||
fc741216 | 2151 | if (netif_needs_gso(skb, features)) { |
91ecb63c | 2152 | if (unlikely(dev_gso_segment(skb, features))) |
9ccb8975 DM |
2153 | goto out_kfree_skb; |
2154 | if (skb->next) | |
2155 | goto gso; | |
6afff0ca | 2156 | } else { |
02932ce9 | 2157 | if (skb_needs_linearize(skb, features) && |
6afff0ca JF |
2158 | __skb_linearize(skb)) |
2159 | goto out_kfree_skb; | |
2160 | ||
2161 | /* If packet is not checksummed and device does not | |
2162 | * support checksumming for this protocol, complete | |
2163 | * checksumming here. | |
2164 | */ | |
2165 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
55508d60 MM |
2166 | skb_set_transport_header(skb, |
2167 | skb_checksum_start_offset(skb)); | |
03634668 | 2168 | if (!(features & NETIF_F_ALL_CSUM) && |
6afff0ca JF |
2169 | skb_checksum_help(skb)) |
2170 | goto out_kfree_skb; | |
2171 | } | |
9ccb8975 DM |
2172 | } |
2173 | ||
ec764bf0 | 2174 | skb_len = skb->len; |
ac45f602 | 2175 | rc = ops->ndo_start_xmit(skb, dev); |
ec764bf0 | 2176 | trace_net_dev_xmit(skb, rc, dev, skb_len); |
ec634fe3 | 2177 | if (rc == NETDEV_TX_OK) |
08baf561 | 2178 | txq_trans_update(txq); |
ac45f602 | 2179 | return rc; |
f6a78bfc HX |
2180 | } |
2181 | ||
576a30eb | 2182 | gso: |
f6a78bfc HX |
2183 | do { |
2184 | struct sk_buff *nskb = skb->next; | |
f6a78bfc HX |
2185 | |
2186 | skb->next = nskb->next; | |
2187 | nskb->next = NULL; | |
068a2de5 KK |
2188 | |
2189 | /* | |
25985edc | 2190 | * If device doesn't need nskb->dst, release it right now while |
068a2de5 KK |
2191 | * its hot in this cpu cache |
2192 | */ | |
2193 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
2194 | skb_dst_drop(nskb); | |
2195 | ||
ec764bf0 | 2196 | skb_len = nskb->len; |
00829823 | 2197 | rc = ops->ndo_start_xmit(nskb, dev); |
ec764bf0 | 2198 | trace_net_dev_xmit(nskb, rc, dev, skb_len); |
ec634fe3 | 2199 | if (unlikely(rc != NETDEV_TX_OK)) { |
572a9d7b PM |
2200 | if (rc & ~NETDEV_TX_MASK) |
2201 | goto out_kfree_gso_skb; | |
f54d9e8d | 2202 | nskb->next = skb->next; |
f6a78bfc HX |
2203 | skb->next = nskb; |
2204 | return rc; | |
2205 | } | |
08baf561 | 2206 | txq_trans_update(txq); |
fd2ea0a7 | 2207 | if (unlikely(netif_tx_queue_stopped(txq) && skb->next)) |
f54d9e8d | 2208 | return NETDEV_TX_BUSY; |
f6a78bfc | 2209 | } while (skb->next); |
4ec93edb | 2210 | |
572a9d7b PM |
2211 | out_kfree_gso_skb: |
2212 | if (likely(skb->next == NULL)) | |
2213 | skb->destructor = DEV_GSO_CB(skb)->destructor; | |
f6a78bfc HX |
2214 | out_kfree_skb: |
2215 | kfree_skb(skb); | |
7b9c6090 | 2216 | out: |
572a9d7b | 2217 | return rc; |
f6a78bfc HX |
2218 | } |
2219 | ||
0a9627f2 | 2220 | static u32 hashrnd __read_mostly; |
b6b2fed1 | 2221 | |
a3d22a68 VZ |
2222 | /* |
2223 | * Returns a Tx hash based on the given packet descriptor a Tx queues' number | |
2224 | * to be used as a distribution range. | |
2225 | */ | |
2226 | u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb, | |
2227 | unsigned int num_tx_queues) | |
8f0f2223 | 2228 | { |
7019298a | 2229 | u32 hash; |
4f57c087 JF |
2230 | u16 qoffset = 0; |
2231 | u16 qcount = num_tx_queues; | |
b6b2fed1 | 2232 | |
513de11b DM |
2233 | if (skb_rx_queue_recorded(skb)) { |
2234 | hash = skb_get_rx_queue(skb); | |
a3d22a68 VZ |
2235 | while (unlikely(hash >= num_tx_queues)) |
2236 | hash -= num_tx_queues; | |
513de11b DM |
2237 | return hash; |
2238 | } | |
ec581f6a | 2239 | |
4f57c087 JF |
2240 | if (dev->num_tc) { |
2241 | u8 tc = netdev_get_prio_tc_map(dev, skb->priority); | |
2242 | qoffset = dev->tc_to_txq[tc].offset; | |
2243 | qcount = dev->tc_to_txq[tc].count; | |
2244 | } | |
2245 | ||
ec581f6a | 2246 | if (skb->sk && skb->sk->sk_hash) |
7019298a | 2247 | hash = skb->sk->sk_hash; |
ec581f6a | 2248 | else |
87fd308c | 2249 | hash = (__force u16) skb->protocol ^ skb->rxhash; |
0a9627f2 | 2250 | hash = jhash_1word(hash, hashrnd); |
b6b2fed1 | 2251 | |
4f57c087 | 2252 | return (u16) (((u64) hash * qcount) >> 32) + qoffset; |
8f0f2223 | 2253 | } |
a3d22a68 | 2254 | EXPORT_SYMBOL(__skb_tx_hash); |
8f0f2223 | 2255 | |
ed04642f ED |
2256 | static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index) |
2257 | { | |
2258 | if (unlikely(queue_index >= dev->real_num_tx_queues)) { | |
2259 | if (net_ratelimit()) { | |
7a161ea9 ED |
2260 | pr_warning("%s selects TX queue %d, but " |
2261 | "real number of TX queues is %d\n", | |
2262 | dev->name, queue_index, dev->real_num_tx_queues); | |
ed04642f ED |
2263 | } |
2264 | return 0; | |
2265 | } | |
2266 | return queue_index; | |
2267 | } | |
2268 | ||
1d24eb48 TH |
2269 | static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb) |
2270 | { | |
bf264145 | 2271 | #ifdef CONFIG_XPS |
1d24eb48 TH |
2272 | struct xps_dev_maps *dev_maps; |
2273 | struct xps_map *map; | |
2274 | int queue_index = -1; | |
2275 | ||
2276 | rcu_read_lock(); | |
2277 | dev_maps = rcu_dereference(dev->xps_maps); | |
2278 | if (dev_maps) { | |
2279 | map = rcu_dereference( | |
2280 | dev_maps->cpu_map[raw_smp_processor_id()]); | |
2281 | if (map) { | |
2282 | if (map->len == 1) | |
2283 | queue_index = map->queues[0]; | |
2284 | else { | |
2285 | u32 hash; | |
2286 | if (skb->sk && skb->sk->sk_hash) | |
2287 | hash = skb->sk->sk_hash; | |
2288 | else | |
2289 | hash = (__force u16) skb->protocol ^ | |
2290 | skb->rxhash; | |
2291 | hash = jhash_1word(hash, hashrnd); | |
2292 | queue_index = map->queues[ | |
2293 | ((u64)hash * map->len) >> 32]; | |
2294 | } | |
2295 | if (unlikely(queue_index >= dev->real_num_tx_queues)) | |
2296 | queue_index = -1; | |
2297 | } | |
2298 | } | |
2299 | rcu_read_unlock(); | |
2300 | ||
2301 | return queue_index; | |
2302 | #else | |
2303 | return -1; | |
2304 | #endif | |
2305 | } | |
2306 | ||
e8a0464c DM |
2307 | static struct netdev_queue *dev_pick_tx(struct net_device *dev, |
2308 | struct sk_buff *skb) | |
2309 | { | |
b0f77d0e | 2310 | int queue_index; |
deabc772 | 2311 | const struct net_device_ops *ops = dev->netdev_ops; |
a4ee3ce3 | 2312 | |
3853b584 TH |
2313 | if (dev->real_num_tx_queues == 1) |
2314 | queue_index = 0; | |
2315 | else if (ops->ndo_select_queue) { | |
deabc772 HS |
2316 | queue_index = ops->ndo_select_queue(dev, skb); |
2317 | queue_index = dev_cap_txqueue(dev, queue_index); | |
2318 | } else { | |
2319 | struct sock *sk = skb->sk; | |
2320 | queue_index = sk_tx_queue_get(sk); | |
a4ee3ce3 | 2321 | |
3853b584 TH |
2322 | if (queue_index < 0 || skb->ooo_okay || |
2323 | queue_index >= dev->real_num_tx_queues) { | |
2324 | int old_index = queue_index; | |
fd2ea0a7 | 2325 | |
1d24eb48 TH |
2326 | queue_index = get_xps_queue(dev, skb); |
2327 | if (queue_index < 0) | |
2328 | queue_index = skb_tx_hash(dev, skb); | |
3853b584 TH |
2329 | |
2330 | if (queue_index != old_index && sk) { | |
2331 | struct dst_entry *dst = | |
2332 | rcu_dereference_check(sk->sk_dst_cache, 1); | |
8728c544 ED |
2333 | |
2334 | if (dst && skb_dst(skb) == dst) | |
2335 | sk_tx_queue_set(sk, queue_index); | |
2336 | } | |
a4ee3ce3 KK |
2337 | } |
2338 | } | |
eae792b7 | 2339 | |
fd2ea0a7 DM |
2340 | skb_set_queue_mapping(skb, queue_index); |
2341 | return netdev_get_tx_queue(dev, queue_index); | |
e8a0464c DM |
2342 | } |
2343 | ||
bbd8a0d3 KK |
2344 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
2345 | struct net_device *dev, | |
2346 | struct netdev_queue *txq) | |
2347 | { | |
2348 | spinlock_t *root_lock = qdisc_lock(q); | |
a2da570d | 2349 | bool contended; |
bbd8a0d3 KK |
2350 | int rc; |
2351 | ||
a2da570d ED |
2352 | qdisc_skb_cb(skb)->pkt_len = skb->len; |
2353 | qdisc_calculate_pkt_len(skb, q); | |
79640a4c ED |
2354 | /* |
2355 | * Heuristic to force contended enqueues to serialize on a | |
2356 | * separate lock before trying to get qdisc main lock. | |
2357 | * This permits __QDISC_STATE_RUNNING owner to get the lock more often | |
2358 | * and dequeue packets faster. | |
2359 | */ | |
a2da570d | 2360 | contended = qdisc_is_running(q); |
79640a4c ED |
2361 | if (unlikely(contended)) |
2362 | spin_lock(&q->busylock); | |
2363 | ||
bbd8a0d3 KK |
2364 | spin_lock(root_lock); |
2365 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
2366 | kfree_skb(skb); | |
2367 | rc = NET_XMIT_DROP; | |
2368 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
bc135b23 | 2369 | qdisc_run_begin(q)) { |
bbd8a0d3 KK |
2370 | /* |
2371 | * This is a work-conserving queue; there are no old skbs | |
2372 | * waiting to be sent out; and the qdisc is not running - | |
2373 | * xmit the skb directly. | |
2374 | */ | |
7fee226a ED |
2375 | if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE)) |
2376 | skb_dst_force(skb); | |
bfe0d029 | 2377 | |
bfe0d029 ED |
2378 | qdisc_bstats_update(q, skb); |
2379 | ||
79640a4c ED |
2380 | if (sch_direct_xmit(skb, q, dev, txq, root_lock)) { |
2381 | if (unlikely(contended)) { | |
2382 | spin_unlock(&q->busylock); | |
2383 | contended = false; | |
2384 | } | |
bbd8a0d3 | 2385 | __qdisc_run(q); |
79640a4c | 2386 | } else |
bc135b23 | 2387 | qdisc_run_end(q); |
bbd8a0d3 KK |
2388 | |
2389 | rc = NET_XMIT_SUCCESS; | |
2390 | } else { | |
7fee226a | 2391 | skb_dst_force(skb); |
a2da570d | 2392 | rc = q->enqueue(skb, q) & NET_XMIT_MASK; |
79640a4c ED |
2393 | if (qdisc_run_begin(q)) { |
2394 | if (unlikely(contended)) { | |
2395 | spin_unlock(&q->busylock); | |
2396 | contended = false; | |
2397 | } | |
2398 | __qdisc_run(q); | |
2399 | } | |
bbd8a0d3 KK |
2400 | } |
2401 | spin_unlock(root_lock); | |
79640a4c ED |
2402 | if (unlikely(contended)) |
2403 | spin_unlock(&q->busylock); | |
bbd8a0d3 KK |
2404 | return rc; |
2405 | } | |
2406 | ||
745e20f1 | 2407 | static DEFINE_PER_CPU(int, xmit_recursion); |
11a766ce | 2408 | #define RECURSION_LIMIT 10 |
745e20f1 | 2409 | |
d29f749e DJ |
2410 | /** |
2411 | * dev_queue_xmit - transmit a buffer | |
2412 | * @skb: buffer to transmit | |
2413 | * | |
2414 | * Queue a buffer for transmission to a network device. The caller must | |
2415 | * have set the device and priority and built the buffer before calling | |
2416 | * this function. The function can be called from an interrupt. | |
2417 | * | |
2418 | * A negative errno code is returned on a failure. A success does not | |
2419 | * guarantee the frame will be transmitted as it may be dropped due | |
2420 | * to congestion or traffic shaping. | |
2421 | * | |
2422 | * ----------------------------------------------------------------------------------- | |
2423 | * I notice this method can also return errors from the queue disciplines, | |
2424 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
2425 | * be positive. | |
2426 | * | |
2427 | * Regardless of the return value, the skb is consumed, so it is currently | |
2428 | * difficult to retry a send to this method. (You can bump the ref count | |
2429 | * before sending to hold a reference for retry if you are careful.) | |
2430 | * | |
2431 | * When calling this method, interrupts MUST be enabled. This is because | |
2432 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
2433 | * --BLG | |
2434 | */ | |
1da177e4 LT |
2435 | int dev_queue_xmit(struct sk_buff *skb) |
2436 | { | |
2437 | struct net_device *dev = skb->dev; | |
dc2b4847 | 2438 | struct netdev_queue *txq; |
1da177e4 LT |
2439 | struct Qdisc *q; |
2440 | int rc = -ENOMEM; | |
2441 | ||
4ec93edb YH |
2442 | /* Disable soft irqs for various locks below. Also |
2443 | * stops preemption for RCU. | |
1da177e4 | 2444 | */ |
4ec93edb | 2445 | rcu_read_lock_bh(); |
1da177e4 | 2446 | |
eae792b7 | 2447 | txq = dev_pick_tx(dev, skb); |
a898def2 | 2448 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 2449 | |
1da177e4 | 2450 | #ifdef CONFIG_NET_CLS_ACT |
d1b19dff | 2451 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS); |
1da177e4 | 2452 | #endif |
cf66ba58 | 2453 | trace_net_dev_queue(skb); |
1da177e4 | 2454 | if (q->enqueue) { |
bbd8a0d3 | 2455 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 2456 | goto out; |
1da177e4 LT |
2457 | } |
2458 | ||
2459 | /* The device has no queue. Common case for software devices: | |
2460 | loopback, all the sorts of tunnels... | |
2461 | ||
932ff279 HX |
2462 | Really, it is unlikely that netif_tx_lock protection is necessary |
2463 | here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
1da177e4 LT |
2464 | counters.) |
2465 | However, it is possible, that they rely on protection | |
2466 | made by us here. | |
2467 | ||
2468 | Check this and shot the lock. It is not prone from deadlocks. | |
2469 | Either shot noqueue qdisc, it is even simpler 8) | |
2470 | */ | |
2471 | if (dev->flags & IFF_UP) { | |
2472 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
2473 | ||
c773e847 | 2474 | if (txq->xmit_lock_owner != cpu) { |
1da177e4 | 2475 | |
745e20f1 ED |
2476 | if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT) |
2477 | goto recursion_alert; | |
2478 | ||
c773e847 | 2479 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 2480 | |
fd2ea0a7 | 2481 | if (!netif_tx_queue_stopped(txq)) { |
745e20f1 | 2482 | __this_cpu_inc(xmit_recursion); |
572a9d7b | 2483 | rc = dev_hard_start_xmit(skb, dev, txq); |
745e20f1 | 2484 | __this_cpu_dec(xmit_recursion); |
572a9d7b | 2485 | if (dev_xmit_complete(rc)) { |
c773e847 | 2486 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
2487 | goto out; |
2488 | } | |
2489 | } | |
c773e847 | 2490 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
2491 | if (net_ratelimit()) |
2492 | printk(KERN_CRIT "Virtual device %s asks to " | |
2493 | "queue packet!\n", dev->name); | |
2494 | } else { | |
2495 | /* Recursion is detected! It is possible, | |
745e20f1 ED |
2496 | * unfortunately |
2497 | */ | |
2498 | recursion_alert: | |
1da177e4 LT |
2499 | if (net_ratelimit()) |
2500 | printk(KERN_CRIT "Dead loop on virtual device " | |
2501 | "%s, fix it urgently!\n", dev->name); | |
2502 | } | |
2503 | } | |
2504 | ||
2505 | rc = -ENETDOWN; | |
d4828d85 | 2506 | rcu_read_unlock_bh(); |
1da177e4 | 2507 | |
1da177e4 LT |
2508 | kfree_skb(skb); |
2509 | return rc; | |
2510 | out: | |
d4828d85 | 2511 | rcu_read_unlock_bh(); |
1da177e4 LT |
2512 | return rc; |
2513 | } | |
d1b19dff | 2514 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 LT |
2515 | |
2516 | ||
2517 | /*======================================================================= | |
2518 | Receiver routines | |
2519 | =======================================================================*/ | |
2520 | ||
6b2bedc3 | 2521 | int netdev_max_backlog __read_mostly = 1000; |
3b098e2d | 2522 | int netdev_tstamp_prequeue __read_mostly = 1; |
6b2bedc3 SH |
2523 | int netdev_budget __read_mostly = 300; |
2524 | int weight_p __read_mostly = 64; /* old backlog weight */ | |
1da177e4 | 2525 | |
eecfd7c4 ED |
2526 | /* Called with irq disabled */ |
2527 | static inline void ____napi_schedule(struct softnet_data *sd, | |
2528 | struct napi_struct *napi) | |
2529 | { | |
2530 | list_add_tail(&napi->poll_list, &sd->poll_list); | |
2531 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2532 | } | |
2533 | ||
0a9627f2 | 2534 | /* |
bfb564e7 | 2535 | * __skb_get_rxhash: calculate a flow hash based on src/dst addresses |
bdeab991 TH |
2536 | * and src/dst port numbers. Sets rxhash in skb to non-zero hash value |
2537 | * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb | |
2538 | * if hash is a canonical 4-tuple hash over transport ports. | |
0a9627f2 | 2539 | */ |
bdeab991 | 2540 | void __skb_get_rxhash(struct sk_buff *skb) |
0a9627f2 | 2541 | { |
12fcdefb | 2542 | int nhoff, hash = 0, poff; |
b71d1d42 ED |
2543 | const struct ipv6hdr *ip6; |
2544 | const struct iphdr *ip; | |
1ff1986f | 2545 | const struct vlan_hdr *vlan; |
0a9627f2 | 2546 | u8 ip_proto; |
792df22c TH |
2547 | u32 addr1, addr2; |
2548 | u16 proto; | |
8c52d509 CG |
2549 | union { |
2550 | u32 v32; | |
2551 | u16 v16[2]; | |
2552 | } ports; | |
0a9627f2 | 2553 | |
bfb564e7 | 2554 | nhoff = skb_network_offset(skb); |
792df22c | 2555 | proto = skb->protocol; |
0a9627f2 | 2556 | |
e971b722 | 2557 | again: |
792df22c | 2558 | switch (proto) { |
0a9627f2 | 2559 | case __constant_htons(ETH_P_IP): |
5dd17e08 | 2560 | ip: |
bfb564e7 | 2561 | if (!pskb_may_pull(skb, sizeof(*ip) + nhoff)) |
0a9627f2 TH |
2562 | goto done; |
2563 | ||
b71d1d42 | 2564 | ip = (const struct iphdr *) (skb->data + nhoff); |
56f8a75c | 2565 | if (ip_is_fragment(ip)) |
dbe5775b CG |
2566 | ip_proto = 0; |
2567 | else | |
2568 | ip_proto = ip->protocol; | |
b249dcb8 ED |
2569 | addr1 = (__force u32) ip->saddr; |
2570 | addr2 = (__force u32) ip->daddr; | |
792df22c | 2571 | nhoff += ip->ihl * 4; |
0a9627f2 TH |
2572 | break; |
2573 | case __constant_htons(ETH_P_IPV6): | |
5dd17e08 | 2574 | ipv6: |
bfb564e7 | 2575 | if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff)) |
0a9627f2 TH |
2576 | goto done; |
2577 | ||
b71d1d42 | 2578 | ip6 = (const struct ipv6hdr *) (skb->data + nhoff); |
0a9627f2 | 2579 | ip_proto = ip6->nexthdr; |
b249dcb8 ED |
2580 | addr1 = (__force u32) ip6->saddr.s6_addr32[3]; |
2581 | addr2 = (__force u32) ip6->daddr.s6_addr32[3]; | |
792df22c | 2582 | nhoff += 40; |
0a9627f2 | 2583 | break; |
1ff1986f CG |
2584 | case __constant_htons(ETH_P_8021Q): |
2585 | if (!pskb_may_pull(skb, sizeof(*vlan) + nhoff)) | |
2586 | goto done; | |
2587 | vlan = (const struct vlan_hdr *) (skb->data + nhoff); | |
2588 | proto = vlan->h_vlan_encapsulated_proto; | |
2589 | nhoff += sizeof(*vlan); | |
2590 | goto again; | |
ae1511bf CG |
2591 | case __constant_htons(ETH_P_PPP_SES): |
2592 | if (!pskb_may_pull(skb, PPPOE_SES_HLEN + nhoff)) | |
2593 | goto done; | |
2594 | proto = *((__be16 *) (skb->data + nhoff + | |
2595 | sizeof(struct pppoe_hdr))); | |
2596 | nhoff += PPPOE_SES_HLEN; | |
5dd17e08 CG |
2597 | switch (proto) { |
2598 | case __constant_htons(PPP_IP): | |
2599 | goto ip; | |
2600 | case __constant_htons(PPP_IPV6): | |
2601 | goto ipv6; | |
2602 | default: | |
2603 | goto done; | |
2604 | } | |
0a9627f2 TH |
2605 | default: |
2606 | goto done; | |
2607 | } | |
bfb564e7 | 2608 | |
e971b722 | 2609 | switch (ip_proto) { |
c6865cb3 TH |
2610 | case IPPROTO_GRE: |
2611 | if (pskb_may_pull(skb, nhoff + 16)) { | |
2612 | u8 *h = skb->data + nhoff; | |
2613 | __be16 flags = *(__be16 *)h; | |
2614 | ||
2615 | /* | |
2616 | * Only look inside GRE if version zero and no | |
2617 | * routing | |
2618 | */ | |
2619 | if (!(flags & (GRE_VERSION|GRE_ROUTING))) { | |
2620 | proto = *(__be16 *)(h + 2); | |
2621 | nhoff += 4; | |
2622 | if (flags & GRE_CSUM) | |
2623 | nhoff += 4; | |
2624 | if (flags & GRE_KEY) | |
2625 | nhoff += 4; | |
2626 | if (flags & GRE_SEQ) | |
2627 | nhoff += 4; | |
2628 | goto again; | |
2629 | } | |
2630 | } | |
2631 | break; | |
ec5efe79 ED |
2632 | case IPPROTO_IPIP: |
2633 | goto again; | |
e971b722 TH |
2634 | default: |
2635 | break; | |
2636 | } | |
2637 | ||
12fcdefb CG |
2638 | ports.v32 = 0; |
2639 | poff = proto_ports_offset(ip_proto); | |
2640 | if (poff >= 0) { | |
792df22c | 2641 | nhoff += poff; |
12fcdefb CG |
2642 | if (pskb_may_pull(skb, nhoff + 4)) { |
2643 | ports.v32 = * (__force u32 *) (skb->data + nhoff); | |
8c52d509 CG |
2644 | if (ports.v16[1] < ports.v16[0]) |
2645 | swap(ports.v16[0], ports.v16[1]); | |
bdeab991 | 2646 | skb->l4_rxhash = 1; |
b249dcb8 | 2647 | } |
0a9627f2 TH |
2648 | } |
2649 | ||
b249dcb8 ED |
2650 | /* get a consistent hash (same value on both flow directions) */ |
2651 | if (addr2 < addr1) | |
2652 | swap(addr1, addr2); | |
0a9627f2 | 2653 | |
bfb564e7 KK |
2654 | hash = jhash_3words(addr1, addr2, ports.v32, hashrnd); |
2655 | if (!hash) | |
2656 | hash = 1; | |
2657 | ||
2658 | done: | |
bdeab991 | 2659 | skb->rxhash = hash; |
bfb564e7 KK |
2660 | } |
2661 | EXPORT_SYMBOL(__skb_get_rxhash); | |
2662 | ||
2663 | #ifdef CONFIG_RPS | |
2664 | ||
2665 | /* One global table that all flow-based protocols share. */ | |
6e3f7faf | 2666 | struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; |
bfb564e7 KK |
2667 | EXPORT_SYMBOL(rps_sock_flow_table); |
2668 | ||
c445477d BH |
2669 | static struct rps_dev_flow * |
2670 | set_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
2671 | struct rps_dev_flow *rflow, u16 next_cpu) | |
2672 | { | |
09994d1b | 2673 | if (next_cpu != RPS_NO_CPU) { |
c445477d BH |
2674 | #ifdef CONFIG_RFS_ACCEL |
2675 | struct netdev_rx_queue *rxqueue; | |
2676 | struct rps_dev_flow_table *flow_table; | |
2677 | struct rps_dev_flow *old_rflow; | |
2678 | u32 flow_id; | |
2679 | u16 rxq_index; | |
2680 | int rc; | |
2681 | ||
2682 | /* Should we steer this flow to a different hardware queue? */ | |
69a19ee6 BH |
2683 | if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || |
2684 | !(dev->features & NETIF_F_NTUPLE)) | |
c445477d BH |
2685 | goto out; |
2686 | rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); | |
2687 | if (rxq_index == skb_get_rx_queue(skb)) | |
2688 | goto out; | |
2689 | ||
2690 | rxqueue = dev->_rx + rxq_index; | |
2691 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
2692 | if (!flow_table) | |
2693 | goto out; | |
2694 | flow_id = skb->rxhash & flow_table->mask; | |
2695 | rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, | |
2696 | rxq_index, flow_id); | |
2697 | if (rc < 0) | |
2698 | goto out; | |
2699 | old_rflow = rflow; | |
2700 | rflow = &flow_table->flows[flow_id]; | |
c445477d BH |
2701 | rflow->filter = rc; |
2702 | if (old_rflow->filter == rflow->filter) | |
2703 | old_rflow->filter = RPS_NO_FILTER; | |
2704 | out: | |
2705 | #endif | |
2706 | rflow->last_qtail = | |
09994d1b | 2707 | per_cpu(softnet_data, next_cpu).input_queue_head; |
c445477d BH |
2708 | } |
2709 | ||
09994d1b | 2710 | rflow->cpu = next_cpu; |
c445477d BH |
2711 | return rflow; |
2712 | } | |
2713 | ||
bfb564e7 KK |
2714 | /* |
2715 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
2716 | * CPU from the RPS map of the receiving queue for a given skb. | |
2717 | * rcu_read_lock must be held on entry. | |
2718 | */ | |
2719 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
2720 | struct rps_dev_flow **rflowp) | |
2721 | { | |
2722 | struct netdev_rx_queue *rxqueue; | |
6e3f7faf | 2723 | struct rps_map *map; |
bfb564e7 KK |
2724 | struct rps_dev_flow_table *flow_table; |
2725 | struct rps_sock_flow_table *sock_flow_table; | |
2726 | int cpu = -1; | |
2727 | u16 tcpu; | |
2728 | ||
2729 | if (skb_rx_queue_recorded(skb)) { | |
2730 | u16 index = skb_get_rx_queue(skb); | |
62fe0b40 BH |
2731 | if (unlikely(index >= dev->real_num_rx_queues)) { |
2732 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
2733 | "%s received packet on queue %u, but number " | |
2734 | "of RX queues is %u\n", | |
2735 | dev->name, index, dev->real_num_rx_queues); | |
bfb564e7 KK |
2736 | goto done; |
2737 | } | |
2738 | rxqueue = dev->_rx + index; | |
2739 | } else | |
2740 | rxqueue = dev->_rx; | |
2741 | ||
6e3f7faf ED |
2742 | map = rcu_dereference(rxqueue->rps_map); |
2743 | if (map) { | |
85875236 | 2744 | if (map->len == 1 && |
33d480ce | 2745 | !rcu_access_pointer(rxqueue->rps_flow_table)) { |
6febfca9 CG |
2746 | tcpu = map->cpus[0]; |
2747 | if (cpu_online(tcpu)) | |
2748 | cpu = tcpu; | |
2749 | goto done; | |
2750 | } | |
33d480ce | 2751 | } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) { |
bfb564e7 | 2752 | goto done; |
6febfca9 | 2753 | } |
bfb564e7 | 2754 | |
2d47b459 | 2755 | skb_reset_network_header(skb); |
bfb564e7 KK |
2756 | if (!skb_get_rxhash(skb)) |
2757 | goto done; | |
2758 | ||
fec5e652 TH |
2759 | flow_table = rcu_dereference(rxqueue->rps_flow_table); |
2760 | sock_flow_table = rcu_dereference(rps_sock_flow_table); | |
2761 | if (flow_table && sock_flow_table) { | |
2762 | u16 next_cpu; | |
2763 | struct rps_dev_flow *rflow; | |
2764 | ||
2765 | rflow = &flow_table->flows[skb->rxhash & flow_table->mask]; | |
2766 | tcpu = rflow->cpu; | |
2767 | ||
2768 | next_cpu = sock_flow_table->ents[skb->rxhash & | |
2769 | sock_flow_table->mask]; | |
2770 | ||
2771 | /* | |
2772 | * If the desired CPU (where last recvmsg was done) is | |
2773 | * different from current CPU (one in the rx-queue flow | |
2774 | * table entry), switch if one of the following holds: | |
2775 | * - Current CPU is unset (equal to RPS_NO_CPU). | |
2776 | * - Current CPU is offline. | |
2777 | * - The current CPU's queue tail has advanced beyond the | |
2778 | * last packet that was enqueued using this table entry. | |
2779 | * This guarantees that all previous packets for the flow | |
2780 | * have been dequeued, thus preserving in order delivery. | |
2781 | */ | |
2782 | if (unlikely(tcpu != next_cpu) && | |
2783 | (tcpu == RPS_NO_CPU || !cpu_online(tcpu) || | |
2784 | ((int)(per_cpu(softnet_data, tcpu).input_queue_head - | |
c445477d BH |
2785 | rflow->last_qtail)) >= 0)) |
2786 | rflow = set_rps_cpu(dev, skb, rflow, next_cpu); | |
2787 | ||
fec5e652 TH |
2788 | if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) { |
2789 | *rflowp = rflow; | |
2790 | cpu = tcpu; | |
2791 | goto done; | |
2792 | } | |
2793 | } | |
2794 | ||
0a9627f2 | 2795 | if (map) { |
fec5e652 | 2796 | tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32]; |
0a9627f2 TH |
2797 | |
2798 | if (cpu_online(tcpu)) { | |
2799 | cpu = tcpu; | |
2800 | goto done; | |
2801 | } | |
2802 | } | |
2803 | ||
2804 | done: | |
0a9627f2 TH |
2805 | return cpu; |
2806 | } | |
2807 | ||
c445477d BH |
2808 | #ifdef CONFIG_RFS_ACCEL |
2809 | ||
2810 | /** | |
2811 | * rps_may_expire_flow - check whether an RFS hardware filter may be removed | |
2812 | * @dev: Device on which the filter was set | |
2813 | * @rxq_index: RX queue index | |
2814 | * @flow_id: Flow ID passed to ndo_rx_flow_steer() | |
2815 | * @filter_id: Filter ID returned by ndo_rx_flow_steer() | |
2816 | * | |
2817 | * Drivers that implement ndo_rx_flow_steer() should periodically call | |
2818 | * this function for each installed filter and remove the filters for | |
2819 | * which it returns %true. | |
2820 | */ | |
2821 | bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, | |
2822 | u32 flow_id, u16 filter_id) | |
2823 | { | |
2824 | struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; | |
2825 | struct rps_dev_flow_table *flow_table; | |
2826 | struct rps_dev_flow *rflow; | |
2827 | bool expire = true; | |
2828 | int cpu; | |
2829 | ||
2830 | rcu_read_lock(); | |
2831 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
2832 | if (flow_table && flow_id <= flow_table->mask) { | |
2833 | rflow = &flow_table->flows[flow_id]; | |
2834 | cpu = ACCESS_ONCE(rflow->cpu); | |
2835 | if (rflow->filter == filter_id && cpu != RPS_NO_CPU && | |
2836 | ((int)(per_cpu(softnet_data, cpu).input_queue_head - | |
2837 | rflow->last_qtail) < | |
2838 | (int)(10 * flow_table->mask))) | |
2839 | expire = false; | |
2840 | } | |
2841 | rcu_read_unlock(); | |
2842 | return expire; | |
2843 | } | |
2844 | EXPORT_SYMBOL(rps_may_expire_flow); | |
2845 | ||
2846 | #endif /* CONFIG_RFS_ACCEL */ | |
2847 | ||
0a9627f2 | 2848 | /* Called from hardirq (IPI) context */ |
e36fa2f7 | 2849 | static void rps_trigger_softirq(void *data) |
0a9627f2 | 2850 | { |
e36fa2f7 ED |
2851 | struct softnet_data *sd = data; |
2852 | ||
eecfd7c4 | 2853 | ____napi_schedule(sd, &sd->backlog); |
dee42870 | 2854 | sd->received_rps++; |
0a9627f2 | 2855 | } |
e36fa2f7 | 2856 | |
fec5e652 | 2857 | #endif /* CONFIG_RPS */ |
0a9627f2 | 2858 | |
e36fa2f7 ED |
2859 | /* |
2860 | * Check if this softnet_data structure is another cpu one | |
2861 | * If yes, queue it to our IPI list and return 1 | |
2862 | * If no, return 0 | |
2863 | */ | |
2864 | static int rps_ipi_queued(struct softnet_data *sd) | |
2865 | { | |
2866 | #ifdef CONFIG_RPS | |
2867 | struct softnet_data *mysd = &__get_cpu_var(softnet_data); | |
2868 | ||
2869 | if (sd != mysd) { | |
2870 | sd->rps_ipi_next = mysd->rps_ipi_list; | |
2871 | mysd->rps_ipi_list = sd; | |
2872 | ||
2873 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2874 | return 1; | |
2875 | } | |
2876 | #endif /* CONFIG_RPS */ | |
2877 | return 0; | |
2878 | } | |
2879 | ||
0a9627f2 TH |
2880 | /* |
2881 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
2882 | * queue (may be a remote CPU queue). | |
2883 | */ | |
fec5e652 TH |
2884 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, |
2885 | unsigned int *qtail) | |
0a9627f2 | 2886 | { |
e36fa2f7 | 2887 | struct softnet_data *sd; |
0a9627f2 TH |
2888 | unsigned long flags; |
2889 | ||
e36fa2f7 | 2890 | sd = &per_cpu(softnet_data, cpu); |
0a9627f2 TH |
2891 | |
2892 | local_irq_save(flags); | |
0a9627f2 | 2893 | |
e36fa2f7 | 2894 | rps_lock(sd); |
6e7676c1 CG |
2895 | if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) { |
2896 | if (skb_queue_len(&sd->input_pkt_queue)) { | |
0a9627f2 | 2897 | enqueue: |
e36fa2f7 | 2898 | __skb_queue_tail(&sd->input_pkt_queue, skb); |
76cc8b13 | 2899 | input_queue_tail_incr_save(sd, qtail); |
e36fa2f7 | 2900 | rps_unlock(sd); |
152102c7 | 2901 | local_irq_restore(flags); |
0a9627f2 TH |
2902 | return NET_RX_SUCCESS; |
2903 | } | |
2904 | ||
ebda37c2 ED |
2905 | /* Schedule NAPI for backlog device |
2906 | * We can use non atomic operation since we own the queue lock | |
2907 | */ | |
2908 | if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { | |
e36fa2f7 | 2909 | if (!rps_ipi_queued(sd)) |
eecfd7c4 | 2910 | ____napi_schedule(sd, &sd->backlog); |
0a9627f2 TH |
2911 | } |
2912 | goto enqueue; | |
2913 | } | |
2914 | ||
dee42870 | 2915 | sd->dropped++; |
e36fa2f7 | 2916 | rps_unlock(sd); |
0a9627f2 | 2917 | |
0a9627f2 TH |
2918 | local_irq_restore(flags); |
2919 | ||
caf586e5 | 2920 | atomic_long_inc(&skb->dev->rx_dropped); |
0a9627f2 TH |
2921 | kfree_skb(skb); |
2922 | return NET_RX_DROP; | |
2923 | } | |
1da177e4 | 2924 | |
1da177e4 LT |
2925 | /** |
2926 | * netif_rx - post buffer to the network code | |
2927 | * @skb: buffer to post | |
2928 | * | |
2929 | * This function receives a packet from a device driver and queues it for | |
2930 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
2931 | * may be dropped during processing for congestion control or by the | |
2932 | * protocol layers. | |
2933 | * | |
2934 | * return values: | |
2935 | * NET_RX_SUCCESS (no congestion) | |
1da177e4 LT |
2936 | * NET_RX_DROP (packet was dropped) |
2937 | * | |
2938 | */ | |
2939 | ||
2940 | int netif_rx(struct sk_buff *skb) | |
2941 | { | |
b0e28f1e | 2942 | int ret; |
1da177e4 LT |
2943 | |
2944 | /* if netpoll wants it, pretend we never saw it */ | |
2945 | if (netpoll_rx(skb)) | |
2946 | return NET_RX_DROP; | |
2947 | ||
3b098e2d ED |
2948 | if (netdev_tstamp_prequeue) |
2949 | net_timestamp_check(skb); | |
1da177e4 | 2950 | |
cf66ba58 | 2951 | trace_netif_rx(skb); |
df334545 | 2952 | #ifdef CONFIG_RPS |
b0e28f1e | 2953 | { |
fec5e652 | 2954 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
b0e28f1e ED |
2955 | int cpu; |
2956 | ||
cece1945 | 2957 | preempt_disable(); |
b0e28f1e | 2958 | rcu_read_lock(); |
fec5e652 TH |
2959 | |
2960 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
b0e28f1e ED |
2961 | if (cpu < 0) |
2962 | cpu = smp_processor_id(); | |
fec5e652 TH |
2963 | |
2964 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
2965 | ||
b0e28f1e | 2966 | rcu_read_unlock(); |
cece1945 | 2967 | preempt_enable(); |
b0e28f1e | 2968 | } |
1e94d72f | 2969 | #else |
fec5e652 TH |
2970 | { |
2971 | unsigned int qtail; | |
2972 | ret = enqueue_to_backlog(skb, get_cpu(), &qtail); | |
2973 | put_cpu(); | |
2974 | } | |
1e94d72f | 2975 | #endif |
b0e28f1e | 2976 | return ret; |
1da177e4 | 2977 | } |
d1b19dff | 2978 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
2979 | |
2980 | int netif_rx_ni(struct sk_buff *skb) | |
2981 | { | |
2982 | int err; | |
2983 | ||
2984 | preempt_disable(); | |
2985 | err = netif_rx(skb); | |
2986 | if (local_softirq_pending()) | |
2987 | do_softirq(); | |
2988 | preempt_enable(); | |
2989 | ||
2990 | return err; | |
2991 | } | |
1da177e4 LT |
2992 | EXPORT_SYMBOL(netif_rx_ni); |
2993 | ||
1da177e4 LT |
2994 | static void net_tx_action(struct softirq_action *h) |
2995 | { | |
2996 | struct softnet_data *sd = &__get_cpu_var(softnet_data); | |
2997 | ||
2998 | if (sd->completion_queue) { | |
2999 | struct sk_buff *clist; | |
3000 | ||
3001 | local_irq_disable(); | |
3002 | clist = sd->completion_queue; | |
3003 | sd->completion_queue = NULL; | |
3004 | local_irq_enable(); | |
3005 | ||
3006 | while (clist) { | |
3007 | struct sk_buff *skb = clist; | |
3008 | clist = clist->next; | |
3009 | ||
547b792c | 3010 | WARN_ON(atomic_read(&skb->users)); |
07dc22e7 | 3011 | trace_kfree_skb(skb, net_tx_action); |
1da177e4 LT |
3012 | __kfree_skb(skb); |
3013 | } | |
3014 | } | |
3015 | ||
3016 | if (sd->output_queue) { | |
37437bb2 | 3017 | struct Qdisc *head; |
1da177e4 LT |
3018 | |
3019 | local_irq_disable(); | |
3020 | head = sd->output_queue; | |
3021 | sd->output_queue = NULL; | |
a9cbd588 | 3022 | sd->output_queue_tailp = &sd->output_queue; |
1da177e4 LT |
3023 | local_irq_enable(); |
3024 | ||
3025 | while (head) { | |
37437bb2 DM |
3026 | struct Qdisc *q = head; |
3027 | spinlock_t *root_lock; | |
3028 | ||
1da177e4 LT |
3029 | head = head->next_sched; |
3030 | ||
5fb66229 | 3031 | root_lock = qdisc_lock(q); |
37437bb2 | 3032 | if (spin_trylock(root_lock)) { |
def82a1d JP |
3033 | smp_mb__before_clear_bit(); |
3034 | clear_bit(__QDISC_STATE_SCHED, | |
3035 | &q->state); | |
37437bb2 DM |
3036 | qdisc_run(q); |
3037 | spin_unlock(root_lock); | |
1da177e4 | 3038 | } else { |
195648bb | 3039 | if (!test_bit(__QDISC_STATE_DEACTIVATED, |
e8a83e10 | 3040 | &q->state)) { |
195648bb | 3041 | __netif_reschedule(q); |
e8a83e10 JP |
3042 | } else { |
3043 | smp_mb__before_clear_bit(); | |
3044 | clear_bit(__QDISC_STATE_SCHED, | |
3045 | &q->state); | |
3046 | } | |
1da177e4 LT |
3047 | } |
3048 | } | |
3049 | } | |
3050 | } | |
3051 | ||
ab95bfe0 JP |
3052 | #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \ |
3053 | (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)) | |
da678292 MM |
3054 | /* This hook is defined here for ATM LANE */ |
3055 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
3056 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 3057 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 3058 | #endif |
1da177e4 | 3059 | |
1da177e4 LT |
3060 | #ifdef CONFIG_NET_CLS_ACT |
3061 | /* TODO: Maybe we should just force sch_ingress to be compiled in | |
3062 | * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions | |
3063 | * a compare and 2 stores extra right now if we dont have it on | |
3064 | * but have CONFIG_NET_CLS_ACT | |
25985edc LDM |
3065 | * NOTE: This doesn't stop any functionality; if you dont have |
3066 | * the ingress scheduler, you just can't add policies on ingress. | |
1da177e4 LT |
3067 | * |
3068 | */ | |
24824a09 | 3069 | static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq) |
1da177e4 | 3070 | { |
1da177e4 | 3071 | struct net_device *dev = skb->dev; |
f697c3e8 | 3072 | u32 ttl = G_TC_RTTL(skb->tc_verd); |
555353cf DM |
3073 | int result = TC_ACT_OK; |
3074 | struct Qdisc *q; | |
4ec93edb | 3075 | |
de384830 SH |
3076 | if (unlikely(MAX_RED_LOOP < ttl++)) { |
3077 | if (net_ratelimit()) | |
3078 | pr_warning( "Redir loop detected Dropping packet (%d->%d)\n", | |
3079 | skb->skb_iif, dev->ifindex); | |
f697c3e8 HX |
3080 | return TC_ACT_SHOT; |
3081 | } | |
1da177e4 | 3082 | |
f697c3e8 HX |
3083 | skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl); |
3084 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); | |
1da177e4 | 3085 | |
83874000 | 3086 | q = rxq->qdisc; |
8d50b53d | 3087 | if (q != &noop_qdisc) { |
83874000 | 3088 | spin_lock(qdisc_lock(q)); |
a9312ae8 DM |
3089 | if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) |
3090 | result = qdisc_enqueue_root(skb, q); | |
83874000 DM |
3091 | spin_unlock(qdisc_lock(q)); |
3092 | } | |
f697c3e8 HX |
3093 | |
3094 | return result; | |
3095 | } | |
86e65da9 | 3096 | |
f697c3e8 HX |
3097 | static inline struct sk_buff *handle_ing(struct sk_buff *skb, |
3098 | struct packet_type **pt_prev, | |
3099 | int *ret, struct net_device *orig_dev) | |
3100 | { | |
24824a09 ED |
3101 | struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue); |
3102 | ||
3103 | if (!rxq || rxq->qdisc == &noop_qdisc) | |
f697c3e8 | 3104 | goto out; |
1da177e4 | 3105 | |
f697c3e8 HX |
3106 | if (*pt_prev) { |
3107 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
3108 | *pt_prev = NULL; | |
1da177e4 LT |
3109 | } |
3110 | ||
24824a09 | 3111 | switch (ing_filter(skb, rxq)) { |
f697c3e8 HX |
3112 | case TC_ACT_SHOT: |
3113 | case TC_ACT_STOLEN: | |
3114 | kfree_skb(skb); | |
3115 | return NULL; | |
3116 | } | |
3117 | ||
3118 | out: | |
3119 | skb->tc_verd = 0; | |
3120 | return skb; | |
1da177e4 LT |
3121 | } |
3122 | #endif | |
3123 | ||
ab95bfe0 JP |
3124 | /** |
3125 | * netdev_rx_handler_register - register receive handler | |
3126 | * @dev: device to register a handler for | |
3127 | * @rx_handler: receive handler to register | |
93e2c32b | 3128 | * @rx_handler_data: data pointer that is used by rx handler |
ab95bfe0 JP |
3129 | * |
3130 | * Register a receive hander for a device. This handler will then be | |
3131 | * called from __netif_receive_skb. A negative errno code is returned | |
3132 | * on a failure. | |
3133 | * | |
3134 | * The caller must hold the rtnl_mutex. | |
8a4eb573 JP |
3135 | * |
3136 | * For a general description of rx_handler, see enum rx_handler_result. | |
ab95bfe0 JP |
3137 | */ |
3138 | int netdev_rx_handler_register(struct net_device *dev, | |
93e2c32b JP |
3139 | rx_handler_func_t *rx_handler, |
3140 | void *rx_handler_data) | |
ab95bfe0 JP |
3141 | { |
3142 | ASSERT_RTNL(); | |
3143 | ||
3144 | if (dev->rx_handler) | |
3145 | return -EBUSY; | |
3146 | ||
93e2c32b | 3147 | rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); |
ab95bfe0 JP |
3148 | rcu_assign_pointer(dev->rx_handler, rx_handler); |
3149 | ||
3150 | return 0; | |
3151 | } | |
3152 | EXPORT_SYMBOL_GPL(netdev_rx_handler_register); | |
3153 | ||
3154 | /** | |
3155 | * netdev_rx_handler_unregister - unregister receive handler | |
3156 | * @dev: device to unregister a handler from | |
3157 | * | |
3158 | * Unregister a receive hander from a device. | |
3159 | * | |
3160 | * The caller must hold the rtnl_mutex. | |
3161 | */ | |
3162 | void netdev_rx_handler_unregister(struct net_device *dev) | |
3163 | { | |
3164 | ||
3165 | ASSERT_RTNL(); | |
a9b3cd7f SH |
3166 | RCU_INIT_POINTER(dev->rx_handler, NULL); |
3167 | RCU_INIT_POINTER(dev->rx_handler_data, NULL); | |
ab95bfe0 JP |
3168 | } |
3169 | EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); | |
3170 | ||
10f744d2 | 3171 | static int __netif_receive_skb(struct sk_buff *skb) |
1da177e4 LT |
3172 | { |
3173 | struct packet_type *ptype, *pt_prev; | |
ab95bfe0 | 3174 | rx_handler_func_t *rx_handler; |
f2ccd8fa | 3175 | struct net_device *orig_dev; |
63d8ea7f | 3176 | struct net_device *null_or_dev; |
8a4eb573 | 3177 | bool deliver_exact = false; |
1da177e4 | 3178 | int ret = NET_RX_DROP; |
252e3346 | 3179 | __be16 type; |
1da177e4 | 3180 | |
3b098e2d ED |
3181 | if (!netdev_tstamp_prequeue) |
3182 | net_timestamp_check(skb); | |
81bbb3d4 | 3183 | |
cf66ba58 | 3184 | trace_netif_receive_skb(skb); |
9b22ea56 | 3185 | |
1da177e4 | 3186 | /* if we've gotten here through NAPI, check netpoll */ |
bea3348e | 3187 | if (netpoll_receive_skb(skb)) |
1da177e4 LT |
3188 | return NET_RX_DROP; |
3189 | ||
8964be4a ED |
3190 | if (!skb->skb_iif) |
3191 | skb->skb_iif = skb->dev->ifindex; | |
cc9bd5ce | 3192 | orig_dev = skb->dev; |
8f903c70 | 3193 | |
c1d2bbe1 | 3194 | skb_reset_network_header(skb); |
badff6d0 | 3195 | skb_reset_transport_header(skb); |
0b5c9db1 | 3196 | skb_reset_mac_len(skb); |
1da177e4 LT |
3197 | |
3198 | pt_prev = NULL; | |
3199 | ||
3200 | rcu_read_lock(); | |
3201 | ||
63d8ea7f DM |
3202 | another_round: |
3203 | ||
3204 | __this_cpu_inc(softnet_data.processed); | |
3205 | ||
bcc6d479 JP |
3206 | if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) { |
3207 | skb = vlan_untag(skb); | |
3208 | if (unlikely(!skb)) | |
3209 | goto out; | |
3210 | } | |
3211 | ||
1da177e4 LT |
3212 | #ifdef CONFIG_NET_CLS_ACT |
3213 | if (skb->tc_verd & TC_NCLS) { | |
3214 | skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); | |
3215 | goto ncls; | |
3216 | } | |
3217 | #endif | |
3218 | ||
3219 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
63d8ea7f | 3220 | if (!ptype->dev || ptype->dev == skb->dev) { |
4ec93edb | 3221 | if (pt_prev) |
f2ccd8fa | 3222 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
3223 | pt_prev = ptype; |
3224 | } | |
3225 | } | |
3226 | ||
3227 | #ifdef CONFIG_NET_CLS_ACT | |
f697c3e8 HX |
3228 | skb = handle_ing(skb, &pt_prev, &ret, orig_dev); |
3229 | if (!skb) | |
1da177e4 | 3230 | goto out; |
1da177e4 LT |
3231 | ncls: |
3232 | #endif | |
3233 | ||
2425717b JF |
3234 | if (vlan_tx_tag_present(skb)) { |
3235 | if (pt_prev) { | |
3236 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3237 | pt_prev = NULL; | |
3238 | } | |
3239 | if (vlan_do_receive(&skb)) | |
3240 | goto another_round; | |
3241 | else if (unlikely(!skb)) | |
3242 | goto out; | |
3243 | } | |
3244 | ||
ab95bfe0 JP |
3245 | rx_handler = rcu_dereference(skb->dev->rx_handler); |
3246 | if (rx_handler) { | |
3247 | if (pt_prev) { | |
3248 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3249 | pt_prev = NULL; | |
3250 | } | |
8a4eb573 JP |
3251 | switch (rx_handler(&skb)) { |
3252 | case RX_HANDLER_CONSUMED: | |
ab95bfe0 | 3253 | goto out; |
8a4eb573 | 3254 | case RX_HANDLER_ANOTHER: |
63d8ea7f | 3255 | goto another_round; |
8a4eb573 JP |
3256 | case RX_HANDLER_EXACT: |
3257 | deliver_exact = true; | |
3258 | case RX_HANDLER_PASS: | |
3259 | break; | |
3260 | default: | |
3261 | BUG(); | |
3262 | } | |
ab95bfe0 | 3263 | } |
1da177e4 | 3264 | |
63d8ea7f | 3265 | /* deliver only exact match when indicated */ |
8a4eb573 | 3266 | null_or_dev = deliver_exact ? skb->dev : NULL; |
1f3c8804 | 3267 | |
1da177e4 | 3268 | type = skb->protocol; |
82d8a867 PE |
3269 | list_for_each_entry_rcu(ptype, |
3270 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
63d8ea7f | 3271 | if (ptype->type == type && |
e3f48d37 JP |
3272 | (ptype->dev == null_or_dev || ptype->dev == skb->dev || |
3273 | ptype->dev == orig_dev)) { | |
4ec93edb | 3274 | if (pt_prev) |
f2ccd8fa | 3275 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
3276 | pt_prev = ptype; |
3277 | } | |
3278 | } | |
3279 | ||
3280 | if (pt_prev) { | |
f2ccd8fa | 3281 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
1da177e4 | 3282 | } else { |
caf586e5 | 3283 | atomic_long_inc(&skb->dev->rx_dropped); |
1da177e4 LT |
3284 | kfree_skb(skb); |
3285 | /* Jamal, now you will not able to escape explaining | |
3286 | * me how you were going to use this. :-) | |
3287 | */ | |
3288 | ret = NET_RX_DROP; | |
3289 | } | |
3290 | ||
3291 | out: | |
3292 | rcu_read_unlock(); | |
3293 | return ret; | |
3294 | } | |
0a9627f2 TH |
3295 | |
3296 | /** | |
3297 | * netif_receive_skb - process receive buffer from network | |
3298 | * @skb: buffer to process | |
3299 | * | |
3300 | * netif_receive_skb() is the main receive data processing function. | |
3301 | * It always succeeds. The buffer may be dropped during processing | |
3302 | * for congestion control or by the protocol layers. | |
3303 | * | |
3304 | * This function may only be called from softirq context and interrupts | |
3305 | * should be enabled. | |
3306 | * | |
3307 | * Return values (usually ignored): | |
3308 | * NET_RX_SUCCESS: no congestion | |
3309 | * NET_RX_DROP: packet was dropped | |
3310 | */ | |
3311 | int netif_receive_skb(struct sk_buff *skb) | |
3312 | { | |
3b098e2d ED |
3313 | if (netdev_tstamp_prequeue) |
3314 | net_timestamp_check(skb); | |
3315 | ||
c1f19b51 RC |
3316 | if (skb_defer_rx_timestamp(skb)) |
3317 | return NET_RX_SUCCESS; | |
3318 | ||
df334545 | 3319 | #ifdef CONFIG_RPS |
3b098e2d ED |
3320 | { |
3321 | struct rps_dev_flow voidflow, *rflow = &voidflow; | |
3322 | int cpu, ret; | |
fec5e652 | 3323 | |
3b098e2d ED |
3324 | rcu_read_lock(); |
3325 | ||
3326 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
0a9627f2 | 3327 | |
3b098e2d ED |
3328 | if (cpu >= 0) { |
3329 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
3330 | rcu_read_unlock(); | |
3331 | } else { | |
3332 | rcu_read_unlock(); | |
3333 | ret = __netif_receive_skb(skb); | |
3334 | } | |
0a9627f2 | 3335 | |
3b098e2d | 3336 | return ret; |
fec5e652 | 3337 | } |
1e94d72f TH |
3338 | #else |
3339 | return __netif_receive_skb(skb); | |
3340 | #endif | |
0a9627f2 | 3341 | } |
d1b19dff | 3342 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 3343 | |
88751275 ED |
3344 | /* Network device is going away, flush any packets still pending |
3345 | * Called with irqs disabled. | |
3346 | */ | |
152102c7 | 3347 | static void flush_backlog(void *arg) |
6e583ce5 | 3348 | { |
152102c7 | 3349 | struct net_device *dev = arg; |
e36fa2f7 | 3350 | struct softnet_data *sd = &__get_cpu_var(softnet_data); |
6e583ce5 SH |
3351 | struct sk_buff *skb, *tmp; |
3352 | ||
e36fa2f7 | 3353 | rps_lock(sd); |
6e7676c1 | 3354 | skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { |
6e583ce5 | 3355 | if (skb->dev == dev) { |
e36fa2f7 | 3356 | __skb_unlink(skb, &sd->input_pkt_queue); |
6e583ce5 | 3357 | kfree_skb(skb); |
76cc8b13 | 3358 | input_queue_head_incr(sd); |
6e583ce5 | 3359 | } |
6e7676c1 | 3360 | } |
e36fa2f7 | 3361 | rps_unlock(sd); |
6e7676c1 CG |
3362 | |
3363 | skb_queue_walk_safe(&sd->process_queue, skb, tmp) { | |
3364 | if (skb->dev == dev) { | |
3365 | __skb_unlink(skb, &sd->process_queue); | |
3366 | kfree_skb(skb); | |
76cc8b13 | 3367 | input_queue_head_incr(sd); |
6e7676c1 CG |
3368 | } |
3369 | } | |
6e583ce5 SH |
3370 | } |
3371 | ||
d565b0a1 HX |
3372 | static int napi_gro_complete(struct sk_buff *skb) |
3373 | { | |
3374 | struct packet_type *ptype; | |
3375 | __be16 type = skb->protocol; | |
3376 | struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK]; | |
3377 | int err = -ENOENT; | |
3378 | ||
fc59f9a3 HX |
3379 | if (NAPI_GRO_CB(skb)->count == 1) { |
3380 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 3381 | goto out; |
fc59f9a3 | 3382 | } |
d565b0a1 HX |
3383 | |
3384 | rcu_read_lock(); | |
3385 | list_for_each_entry_rcu(ptype, head, list) { | |
3386 | if (ptype->type != type || ptype->dev || !ptype->gro_complete) | |
3387 | continue; | |
3388 | ||
3389 | err = ptype->gro_complete(skb); | |
3390 | break; | |
3391 | } | |
3392 | rcu_read_unlock(); | |
3393 | ||
3394 | if (err) { | |
3395 | WARN_ON(&ptype->list == head); | |
3396 | kfree_skb(skb); | |
3397 | return NET_RX_SUCCESS; | |
3398 | } | |
3399 | ||
3400 | out: | |
d565b0a1 HX |
3401 | return netif_receive_skb(skb); |
3402 | } | |
3403 | ||
86cac58b | 3404 | inline void napi_gro_flush(struct napi_struct *napi) |
d565b0a1 HX |
3405 | { |
3406 | struct sk_buff *skb, *next; | |
3407 | ||
3408 | for (skb = napi->gro_list; skb; skb = next) { | |
3409 | next = skb->next; | |
3410 | skb->next = NULL; | |
3411 | napi_gro_complete(skb); | |
3412 | } | |
3413 | ||
4ae5544f | 3414 | napi->gro_count = 0; |
d565b0a1 HX |
3415 | napi->gro_list = NULL; |
3416 | } | |
86cac58b | 3417 | EXPORT_SYMBOL(napi_gro_flush); |
d565b0a1 | 3418 | |
5b252f0c | 3419 | enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 HX |
3420 | { |
3421 | struct sk_buff **pp = NULL; | |
3422 | struct packet_type *ptype; | |
3423 | __be16 type = skb->protocol; | |
3424 | struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK]; | |
0da2afd5 | 3425 | int same_flow; |
d565b0a1 | 3426 | int mac_len; |
5b252f0c | 3427 | enum gro_result ret; |
d565b0a1 | 3428 | |
ce9e76c8 | 3429 | if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb)) |
d565b0a1 HX |
3430 | goto normal; |
3431 | ||
21dc3301 | 3432 | if (skb_is_gso(skb) || skb_has_frag_list(skb)) |
f17f5c91 HX |
3433 | goto normal; |
3434 | ||
d565b0a1 HX |
3435 | rcu_read_lock(); |
3436 | list_for_each_entry_rcu(ptype, head, list) { | |
d565b0a1 HX |
3437 | if (ptype->type != type || ptype->dev || !ptype->gro_receive) |
3438 | continue; | |
3439 | ||
86911732 | 3440 | skb_set_network_header(skb, skb_gro_offset(skb)); |
d565b0a1 HX |
3441 | mac_len = skb->network_header - skb->mac_header; |
3442 | skb->mac_len = mac_len; | |
3443 | NAPI_GRO_CB(skb)->same_flow = 0; | |
3444 | NAPI_GRO_CB(skb)->flush = 0; | |
5d38a079 | 3445 | NAPI_GRO_CB(skb)->free = 0; |
d565b0a1 | 3446 | |
d565b0a1 HX |
3447 | pp = ptype->gro_receive(&napi->gro_list, skb); |
3448 | break; | |
3449 | } | |
3450 | rcu_read_unlock(); | |
3451 | ||
3452 | if (&ptype->list == head) | |
3453 | goto normal; | |
3454 | ||
0da2afd5 | 3455 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 3456 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 3457 | |
d565b0a1 HX |
3458 | if (pp) { |
3459 | struct sk_buff *nskb = *pp; | |
3460 | ||
3461 | *pp = nskb->next; | |
3462 | nskb->next = NULL; | |
3463 | napi_gro_complete(nskb); | |
4ae5544f | 3464 | napi->gro_count--; |
d565b0a1 HX |
3465 | } |
3466 | ||
0da2afd5 | 3467 | if (same_flow) |
d565b0a1 HX |
3468 | goto ok; |
3469 | ||
4ae5544f | 3470 | if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS) |
d565b0a1 | 3471 | goto normal; |
d565b0a1 | 3472 | |
4ae5544f | 3473 | napi->gro_count++; |
d565b0a1 | 3474 | NAPI_GRO_CB(skb)->count = 1; |
86911732 | 3475 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
d565b0a1 HX |
3476 | skb->next = napi->gro_list; |
3477 | napi->gro_list = skb; | |
5d0d9be8 | 3478 | ret = GRO_HELD; |
d565b0a1 | 3479 | |
ad0f9904 | 3480 | pull: |
cb18978c HX |
3481 | if (skb_headlen(skb) < skb_gro_offset(skb)) { |
3482 | int grow = skb_gro_offset(skb) - skb_headlen(skb); | |
3483 | ||
3484 | BUG_ON(skb->end - skb->tail < grow); | |
3485 | ||
3486 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
3487 | ||
3488 | skb->tail += grow; | |
3489 | skb->data_len -= grow; | |
3490 | ||
3491 | skb_shinfo(skb)->frags[0].page_offset += grow; | |
9e903e08 | 3492 | skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow); |
cb18978c | 3493 | |
9e903e08 | 3494 | if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) { |
ea2ab693 | 3495 | skb_frag_unref(skb, 0); |
cb18978c HX |
3496 | memmove(skb_shinfo(skb)->frags, |
3497 | skb_shinfo(skb)->frags + 1, | |
e5093aec | 3498 | --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t)); |
cb18978c | 3499 | } |
ad0f9904 HX |
3500 | } |
3501 | ||
d565b0a1 | 3502 | ok: |
5d0d9be8 | 3503 | return ret; |
d565b0a1 HX |
3504 | |
3505 | normal: | |
ad0f9904 HX |
3506 | ret = GRO_NORMAL; |
3507 | goto pull; | |
5d38a079 | 3508 | } |
96e93eab HX |
3509 | EXPORT_SYMBOL(dev_gro_receive); |
3510 | ||
40d0802b | 3511 | static inline gro_result_t |
5b252f0c | 3512 | __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab HX |
3513 | { |
3514 | struct sk_buff *p; | |
3515 | ||
3516 | for (p = napi->gro_list; p; p = p->next) { | |
40d0802b ED |
3517 | unsigned long diffs; |
3518 | ||
3519 | diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; | |
3701e513 | 3520 | diffs |= p->vlan_tci ^ skb->vlan_tci; |
40d0802b | 3521 | diffs |= compare_ether_header(skb_mac_header(p), |
f64f9e71 | 3522 | skb_gro_mac_header(skb)); |
40d0802b | 3523 | NAPI_GRO_CB(p)->same_flow = !diffs; |
96e93eab HX |
3524 | NAPI_GRO_CB(p)->flush = 0; |
3525 | } | |
3526 | ||
3527 | return dev_gro_receive(napi, skb); | |
3528 | } | |
5d38a079 | 3529 | |
c7c4b3b6 | 3530 | gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) |
5d38a079 | 3531 | { |
5d0d9be8 HX |
3532 | switch (ret) { |
3533 | case GRO_NORMAL: | |
c7c4b3b6 BH |
3534 | if (netif_receive_skb(skb)) |
3535 | ret = GRO_DROP; | |
3536 | break; | |
5d38a079 | 3537 | |
5d0d9be8 | 3538 | case GRO_DROP: |
5d0d9be8 | 3539 | case GRO_MERGED_FREE: |
5d38a079 HX |
3540 | kfree_skb(skb); |
3541 | break; | |
5b252f0c BH |
3542 | |
3543 | case GRO_HELD: | |
3544 | case GRO_MERGED: | |
3545 | break; | |
5d38a079 HX |
3546 | } |
3547 | ||
c7c4b3b6 | 3548 | return ret; |
5d0d9be8 HX |
3549 | } |
3550 | EXPORT_SYMBOL(napi_skb_finish); | |
3551 | ||
78a478d0 HX |
3552 | void skb_gro_reset_offset(struct sk_buff *skb) |
3553 | { | |
3554 | NAPI_GRO_CB(skb)->data_offset = 0; | |
3555 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
7489594c | 3556 | NAPI_GRO_CB(skb)->frag0_len = 0; |
78a478d0 | 3557 | |
78d3fd0b | 3558 | if (skb->mac_header == skb->tail && |
ea2ab693 | 3559 | !PageHighMem(skb_frag_page(&skb_shinfo(skb)->frags[0]))) { |
78a478d0 | 3560 | NAPI_GRO_CB(skb)->frag0 = |
ea2ab693 | 3561 | skb_frag_address(&skb_shinfo(skb)->frags[0]); |
9e903e08 | 3562 | NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(&skb_shinfo(skb)->frags[0]); |
7489594c | 3563 | } |
78a478d0 HX |
3564 | } |
3565 | EXPORT_SYMBOL(skb_gro_reset_offset); | |
3566 | ||
c7c4b3b6 | 3567 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 3568 | { |
86911732 HX |
3569 | skb_gro_reset_offset(skb); |
3570 | ||
5d0d9be8 | 3571 | return napi_skb_finish(__napi_gro_receive(napi, skb), skb); |
d565b0a1 HX |
3572 | } |
3573 | EXPORT_SYMBOL(napi_gro_receive); | |
3574 | ||
d0c2b0d2 | 3575 | static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab | 3576 | { |
96e93eab HX |
3577 | __skb_pull(skb, skb_headlen(skb)); |
3578 | skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb)); | |
3701e513 | 3579 | skb->vlan_tci = 0; |
66c46d74 | 3580 | skb->dev = napi->dev; |
6d152e23 | 3581 | skb->skb_iif = 0; |
96e93eab HX |
3582 | |
3583 | napi->skb = skb; | |
3584 | } | |
96e93eab | 3585 | |
76620aaf | 3586 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 3587 | { |
5d38a079 | 3588 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
3589 | |
3590 | if (!skb) { | |
89d71a66 ED |
3591 | skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD); |
3592 | if (skb) | |
3593 | napi->skb = skb; | |
80595d59 | 3594 | } |
96e93eab HX |
3595 | return skb; |
3596 | } | |
76620aaf | 3597 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 3598 | |
c7c4b3b6 BH |
3599 | gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb, |
3600 | gro_result_t ret) | |
96e93eab | 3601 | { |
5d0d9be8 HX |
3602 | switch (ret) { |
3603 | case GRO_NORMAL: | |
86911732 | 3604 | case GRO_HELD: |
e76b69cc | 3605 | skb->protocol = eth_type_trans(skb, skb->dev); |
86911732 | 3606 | |
c7c4b3b6 BH |
3607 | if (ret == GRO_HELD) |
3608 | skb_gro_pull(skb, -ETH_HLEN); | |
3609 | else if (netif_receive_skb(skb)) | |
3610 | ret = GRO_DROP; | |
86911732 | 3611 | break; |
5d38a079 | 3612 | |
5d0d9be8 | 3613 | case GRO_DROP: |
5d0d9be8 HX |
3614 | case GRO_MERGED_FREE: |
3615 | napi_reuse_skb(napi, skb); | |
3616 | break; | |
5b252f0c BH |
3617 | |
3618 | case GRO_MERGED: | |
3619 | break; | |
5d0d9be8 | 3620 | } |
5d38a079 | 3621 | |
c7c4b3b6 | 3622 | return ret; |
5d38a079 | 3623 | } |
5d0d9be8 HX |
3624 | EXPORT_SYMBOL(napi_frags_finish); |
3625 | ||
76620aaf HX |
3626 | struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
3627 | { | |
3628 | struct sk_buff *skb = napi->skb; | |
3629 | struct ethhdr *eth; | |
a5b1cf28 HX |
3630 | unsigned int hlen; |
3631 | unsigned int off; | |
76620aaf HX |
3632 | |
3633 | napi->skb = NULL; | |
3634 | ||
3635 | skb_reset_mac_header(skb); | |
3636 | skb_gro_reset_offset(skb); | |
3637 | ||
a5b1cf28 HX |
3638 | off = skb_gro_offset(skb); |
3639 | hlen = off + sizeof(*eth); | |
3640 | eth = skb_gro_header_fast(skb, off); | |
3641 | if (skb_gro_header_hard(skb, hlen)) { | |
3642 | eth = skb_gro_header_slow(skb, hlen, off); | |
3643 | if (unlikely(!eth)) { | |
3644 | napi_reuse_skb(napi, skb); | |
3645 | skb = NULL; | |
3646 | goto out; | |
3647 | } | |
76620aaf HX |
3648 | } |
3649 | ||
3650 | skb_gro_pull(skb, sizeof(*eth)); | |
3651 | ||
3652 | /* | |
3653 | * This works because the only protocols we care about don't require | |
3654 | * special handling. We'll fix it up properly at the end. | |
3655 | */ | |
3656 | skb->protocol = eth->h_proto; | |
3657 | ||
3658 | out: | |
3659 | return skb; | |
3660 | } | |
3661 | EXPORT_SYMBOL(napi_frags_skb); | |
3662 | ||
c7c4b3b6 | 3663 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 3664 | { |
76620aaf | 3665 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 HX |
3666 | |
3667 | if (!skb) | |
c7c4b3b6 | 3668 | return GRO_DROP; |
5d0d9be8 HX |
3669 | |
3670 | return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb)); | |
3671 | } | |
5d38a079 HX |
3672 | EXPORT_SYMBOL(napi_gro_frags); |
3673 | ||
e326bed2 ED |
3674 | /* |
3675 | * net_rps_action sends any pending IPI's for rps. | |
3676 | * Note: called with local irq disabled, but exits with local irq enabled. | |
3677 | */ | |
3678 | static void net_rps_action_and_irq_enable(struct softnet_data *sd) | |
3679 | { | |
3680 | #ifdef CONFIG_RPS | |
3681 | struct softnet_data *remsd = sd->rps_ipi_list; | |
3682 | ||
3683 | if (remsd) { | |
3684 | sd->rps_ipi_list = NULL; | |
3685 | ||
3686 | local_irq_enable(); | |
3687 | ||
3688 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
3689 | while (remsd) { | |
3690 | struct softnet_data *next = remsd->rps_ipi_next; | |
3691 | ||
3692 | if (cpu_online(remsd->cpu)) | |
3693 | __smp_call_function_single(remsd->cpu, | |
3694 | &remsd->csd, 0); | |
3695 | remsd = next; | |
3696 | } | |
3697 | } else | |
3698 | #endif | |
3699 | local_irq_enable(); | |
3700 | } | |
3701 | ||
bea3348e | 3702 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 LT |
3703 | { |
3704 | int work = 0; | |
eecfd7c4 | 3705 | struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); |
1da177e4 | 3706 | |
e326bed2 ED |
3707 | #ifdef CONFIG_RPS |
3708 | /* Check if we have pending ipi, its better to send them now, | |
3709 | * not waiting net_rx_action() end. | |
3710 | */ | |
3711 | if (sd->rps_ipi_list) { | |
3712 | local_irq_disable(); | |
3713 | net_rps_action_and_irq_enable(sd); | |
3714 | } | |
3715 | #endif | |
bea3348e | 3716 | napi->weight = weight_p; |
6e7676c1 CG |
3717 | local_irq_disable(); |
3718 | while (work < quota) { | |
1da177e4 | 3719 | struct sk_buff *skb; |
6e7676c1 CG |
3720 | unsigned int qlen; |
3721 | ||
3722 | while ((skb = __skb_dequeue(&sd->process_queue))) { | |
3723 | local_irq_enable(); | |
3724 | __netif_receive_skb(skb); | |
6e7676c1 | 3725 | local_irq_disable(); |
76cc8b13 TH |
3726 | input_queue_head_incr(sd); |
3727 | if (++work >= quota) { | |
3728 | local_irq_enable(); | |
3729 | return work; | |
3730 | } | |
6e7676c1 | 3731 | } |
1da177e4 | 3732 | |
e36fa2f7 | 3733 | rps_lock(sd); |
6e7676c1 | 3734 | qlen = skb_queue_len(&sd->input_pkt_queue); |
76cc8b13 | 3735 | if (qlen) |
6e7676c1 CG |
3736 | skb_queue_splice_tail_init(&sd->input_pkt_queue, |
3737 | &sd->process_queue); | |
76cc8b13 | 3738 | |
6e7676c1 | 3739 | if (qlen < quota - work) { |
eecfd7c4 ED |
3740 | /* |
3741 | * Inline a custom version of __napi_complete(). | |
3742 | * only current cpu owns and manipulates this napi, | |
3743 | * and NAPI_STATE_SCHED is the only possible flag set on backlog. | |
3744 | * we can use a plain write instead of clear_bit(), | |
3745 | * and we dont need an smp_mb() memory barrier. | |
3746 | */ | |
3747 | list_del(&napi->poll_list); | |
3748 | napi->state = 0; | |
3749 | ||
6e7676c1 | 3750 | quota = work + qlen; |
bea3348e | 3751 | } |
e36fa2f7 | 3752 | rps_unlock(sd); |
6e7676c1 CG |
3753 | } |
3754 | local_irq_enable(); | |
1da177e4 | 3755 | |
bea3348e SH |
3756 | return work; |
3757 | } | |
1da177e4 | 3758 | |
bea3348e SH |
3759 | /** |
3760 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 3761 | * @n: entry to schedule |
bea3348e SH |
3762 | * |
3763 | * The entry's receive function will be scheduled to run | |
3764 | */ | |
b5606c2d | 3765 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
3766 | { |
3767 | unsigned long flags; | |
1da177e4 | 3768 | |
bea3348e | 3769 | local_irq_save(flags); |
eecfd7c4 | 3770 | ____napi_schedule(&__get_cpu_var(softnet_data), n); |
bea3348e | 3771 | local_irq_restore(flags); |
1da177e4 | 3772 | } |
bea3348e SH |
3773 | EXPORT_SYMBOL(__napi_schedule); |
3774 | ||
d565b0a1 HX |
3775 | void __napi_complete(struct napi_struct *n) |
3776 | { | |
3777 | BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); | |
3778 | BUG_ON(n->gro_list); | |
3779 | ||
3780 | list_del(&n->poll_list); | |
3781 | smp_mb__before_clear_bit(); | |
3782 | clear_bit(NAPI_STATE_SCHED, &n->state); | |
3783 | } | |
3784 | EXPORT_SYMBOL(__napi_complete); | |
3785 | ||
3786 | void napi_complete(struct napi_struct *n) | |
3787 | { | |
3788 | unsigned long flags; | |
3789 | ||
3790 | /* | |
3791 | * don't let napi dequeue from the cpu poll list | |
3792 | * just in case its running on a different cpu | |
3793 | */ | |
3794 | if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state))) | |
3795 | return; | |
3796 | ||
3797 | napi_gro_flush(n); | |
3798 | local_irq_save(flags); | |
3799 | __napi_complete(n); | |
3800 | local_irq_restore(flags); | |
3801 | } | |
3802 | EXPORT_SYMBOL(napi_complete); | |
3803 | ||
3804 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, | |
3805 | int (*poll)(struct napi_struct *, int), int weight) | |
3806 | { | |
3807 | INIT_LIST_HEAD(&napi->poll_list); | |
4ae5544f | 3808 | napi->gro_count = 0; |
d565b0a1 | 3809 | napi->gro_list = NULL; |
5d38a079 | 3810 | napi->skb = NULL; |
d565b0a1 HX |
3811 | napi->poll = poll; |
3812 | napi->weight = weight; | |
3813 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 3814 | napi->dev = dev; |
5d38a079 | 3815 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
3816 | spin_lock_init(&napi->poll_lock); |
3817 | napi->poll_owner = -1; | |
3818 | #endif | |
3819 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
3820 | } | |
3821 | EXPORT_SYMBOL(netif_napi_add); | |
3822 | ||
3823 | void netif_napi_del(struct napi_struct *napi) | |
3824 | { | |
3825 | struct sk_buff *skb, *next; | |
3826 | ||
d7b06636 | 3827 | list_del_init(&napi->dev_list); |
76620aaf | 3828 | napi_free_frags(napi); |
d565b0a1 HX |
3829 | |
3830 | for (skb = napi->gro_list; skb; skb = next) { | |
3831 | next = skb->next; | |
3832 | skb->next = NULL; | |
3833 | kfree_skb(skb); | |
3834 | } | |
3835 | ||
3836 | napi->gro_list = NULL; | |
4ae5544f | 3837 | napi->gro_count = 0; |
d565b0a1 HX |
3838 | } |
3839 | EXPORT_SYMBOL(netif_napi_del); | |
3840 | ||
1da177e4 LT |
3841 | static void net_rx_action(struct softirq_action *h) |
3842 | { | |
e326bed2 | 3843 | struct softnet_data *sd = &__get_cpu_var(softnet_data); |
24f8b238 | 3844 | unsigned long time_limit = jiffies + 2; |
51b0bded | 3845 | int budget = netdev_budget; |
53fb95d3 MM |
3846 | void *have; |
3847 | ||
1da177e4 LT |
3848 | local_irq_disable(); |
3849 | ||
e326bed2 | 3850 | while (!list_empty(&sd->poll_list)) { |
bea3348e SH |
3851 | struct napi_struct *n; |
3852 | int work, weight; | |
1da177e4 | 3853 | |
bea3348e | 3854 | /* If softirq window is exhuasted then punt. |
24f8b238 SH |
3855 | * Allow this to run for 2 jiffies since which will allow |
3856 | * an average latency of 1.5/HZ. | |
bea3348e | 3857 | */ |
24f8b238 | 3858 | if (unlikely(budget <= 0 || time_after(jiffies, time_limit))) |
1da177e4 LT |
3859 | goto softnet_break; |
3860 | ||
3861 | local_irq_enable(); | |
3862 | ||
bea3348e SH |
3863 | /* Even though interrupts have been re-enabled, this |
3864 | * access is safe because interrupts can only add new | |
3865 | * entries to the tail of this list, and only ->poll() | |
3866 | * calls can remove this head entry from the list. | |
3867 | */ | |
e326bed2 | 3868 | n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list); |
1da177e4 | 3869 | |
bea3348e SH |
3870 | have = netpoll_poll_lock(n); |
3871 | ||
3872 | weight = n->weight; | |
3873 | ||
0a7606c1 DM |
3874 | /* This NAPI_STATE_SCHED test is for avoiding a race |
3875 | * with netpoll's poll_napi(). Only the entity which | |
3876 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
3877 | * actually make the ->poll() call. Therefore we avoid | |
25985edc | 3878 | * accidentally calling ->poll() when NAPI is not scheduled. |
0a7606c1 DM |
3879 | */ |
3880 | work = 0; | |
4ea7e386 | 3881 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { |
0a7606c1 | 3882 | work = n->poll(n, weight); |
4ea7e386 NH |
3883 | trace_napi_poll(n); |
3884 | } | |
bea3348e SH |
3885 | |
3886 | WARN_ON_ONCE(work > weight); | |
3887 | ||
3888 | budget -= work; | |
3889 | ||
3890 | local_irq_disable(); | |
3891 | ||
3892 | /* Drivers must not modify the NAPI state if they | |
3893 | * consume the entire weight. In such cases this code | |
3894 | * still "owns" the NAPI instance and therefore can | |
3895 | * move the instance around on the list at-will. | |
3896 | */ | |
fed17f30 | 3897 | if (unlikely(work == weight)) { |
ff780cd8 HX |
3898 | if (unlikely(napi_disable_pending(n))) { |
3899 | local_irq_enable(); | |
3900 | napi_complete(n); | |
3901 | local_irq_disable(); | |
3902 | } else | |
e326bed2 | 3903 | list_move_tail(&n->poll_list, &sd->poll_list); |
fed17f30 | 3904 | } |
bea3348e SH |
3905 | |
3906 | netpoll_poll_unlock(have); | |
1da177e4 LT |
3907 | } |
3908 | out: | |
e326bed2 | 3909 | net_rps_action_and_irq_enable(sd); |
0a9627f2 | 3910 | |
db217334 CL |
3911 | #ifdef CONFIG_NET_DMA |
3912 | /* | |
3913 | * There may not be any more sk_buffs coming right now, so push | |
3914 | * any pending DMA copies to hardware | |
3915 | */ | |
2ba05622 | 3916 | dma_issue_pending_all(); |
db217334 | 3917 | #endif |
bea3348e | 3918 | |
1da177e4 LT |
3919 | return; |
3920 | ||
3921 | softnet_break: | |
dee42870 | 3922 | sd->time_squeeze++; |
1da177e4 LT |
3923 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); |
3924 | goto out; | |
3925 | } | |
3926 | ||
d1b19dff | 3927 | static gifconf_func_t *gifconf_list[NPROTO]; |
1da177e4 LT |
3928 | |
3929 | /** | |
3930 | * register_gifconf - register a SIOCGIF handler | |
3931 | * @family: Address family | |
3932 | * @gifconf: Function handler | |
3933 | * | |
3934 | * Register protocol dependent address dumping routines. The handler | |
3935 | * that is passed must not be freed or reused until it has been replaced | |
3936 | * by another handler. | |
3937 | */ | |
d1b19dff | 3938 | int register_gifconf(unsigned int family, gifconf_func_t *gifconf) |
1da177e4 LT |
3939 | { |
3940 | if (family >= NPROTO) | |
3941 | return -EINVAL; | |
3942 | gifconf_list[family] = gifconf; | |
3943 | return 0; | |
3944 | } | |
d1b19dff | 3945 | EXPORT_SYMBOL(register_gifconf); |
1da177e4 LT |
3946 | |
3947 | ||
3948 | /* | |
3949 | * Map an interface index to its name (SIOCGIFNAME) | |
3950 | */ | |
3951 | ||
3952 | /* | |
3953 | * We need this ioctl for efficient implementation of the | |
3954 | * if_indextoname() function required by the IPv6 API. Without | |
3955 | * it, we would have to search all the interfaces to find a | |
3956 | * match. --pb | |
3957 | */ | |
3958 | ||
881d966b | 3959 | static int dev_ifname(struct net *net, struct ifreq __user *arg) |
1da177e4 LT |
3960 | { |
3961 | struct net_device *dev; | |
3962 | struct ifreq ifr; | |
3963 | ||
3964 | /* | |
3965 | * Fetch the caller's info block. | |
3966 | */ | |
3967 | ||
3968 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
3969 | return -EFAULT; | |
3970 | ||
fb699dfd ED |
3971 | rcu_read_lock(); |
3972 | dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex); | |
1da177e4 | 3973 | if (!dev) { |
fb699dfd | 3974 | rcu_read_unlock(); |
1da177e4 LT |
3975 | return -ENODEV; |
3976 | } | |
3977 | ||
3978 | strcpy(ifr.ifr_name, dev->name); | |
fb699dfd | 3979 | rcu_read_unlock(); |
1da177e4 LT |
3980 | |
3981 | if (copy_to_user(arg, &ifr, sizeof(struct ifreq))) | |
3982 | return -EFAULT; | |
3983 | return 0; | |
3984 | } | |
3985 | ||
3986 | /* | |
3987 | * Perform a SIOCGIFCONF call. This structure will change | |
3988 | * size eventually, and there is nothing I can do about it. | |
3989 | * Thus we will need a 'compatibility mode'. | |
3990 | */ | |
3991 | ||
881d966b | 3992 | static int dev_ifconf(struct net *net, char __user *arg) |
1da177e4 LT |
3993 | { |
3994 | struct ifconf ifc; | |
3995 | struct net_device *dev; | |
3996 | char __user *pos; | |
3997 | int len; | |
3998 | int total; | |
3999 | int i; | |
4000 | ||
4001 | /* | |
4002 | * Fetch the caller's info block. | |
4003 | */ | |
4004 | ||
4005 | if (copy_from_user(&ifc, arg, sizeof(struct ifconf))) | |
4006 | return -EFAULT; | |
4007 | ||
4008 | pos = ifc.ifc_buf; | |
4009 | len = ifc.ifc_len; | |
4010 | ||
4011 | /* | |
4012 | * Loop over the interfaces, and write an info block for each. | |
4013 | */ | |
4014 | ||
4015 | total = 0; | |
881d966b | 4016 | for_each_netdev(net, dev) { |
1da177e4 LT |
4017 | for (i = 0; i < NPROTO; i++) { |
4018 | if (gifconf_list[i]) { | |
4019 | int done; | |
4020 | if (!pos) | |
4021 | done = gifconf_list[i](dev, NULL, 0); | |
4022 | else | |
4023 | done = gifconf_list[i](dev, pos + total, | |
4024 | len - total); | |
4025 | if (done < 0) | |
4026 | return -EFAULT; | |
4027 | total += done; | |
4028 | } | |
4029 | } | |
4ec93edb | 4030 | } |
1da177e4 LT |
4031 | |
4032 | /* | |
4033 | * All done. Write the updated control block back to the caller. | |
4034 | */ | |
4035 | ifc.ifc_len = total; | |
4036 | ||
4037 | /* | |
4038 | * Both BSD and Solaris return 0 here, so we do too. | |
4039 | */ | |
4040 | return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0; | |
4041 | } | |
4042 | ||
4043 | #ifdef CONFIG_PROC_FS | |
4044 | /* | |
4045 | * This is invoked by the /proc filesystem handler to display a device | |
4046 | * in detail. | |
4047 | */ | |
7562f876 | 4048 | void *dev_seq_start(struct seq_file *seq, loff_t *pos) |
c6d14c84 | 4049 | __acquires(RCU) |
1da177e4 | 4050 | { |
e372c414 | 4051 | struct net *net = seq_file_net(seq); |
7562f876 | 4052 | loff_t off; |
1da177e4 | 4053 | struct net_device *dev; |
1da177e4 | 4054 | |
c6d14c84 | 4055 | rcu_read_lock(); |
7562f876 PE |
4056 | if (!*pos) |
4057 | return SEQ_START_TOKEN; | |
1da177e4 | 4058 | |
7562f876 | 4059 | off = 1; |
c6d14c84 | 4060 | for_each_netdev_rcu(net, dev) |
7562f876 PE |
4061 | if (off++ == *pos) |
4062 | return dev; | |
1da177e4 | 4063 | |
7562f876 | 4064 | return NULL; |
1da177e4 LT |
4065 | } |
4066 | ||
4067 | void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
4068 | { | |
ccf43438 ED |
4069 | struct net_device *dev = v; |
4070 | ||
4071 | if (v == SEQ_START_TOKEN) | |
4072 | dev = first_net_device_rcu(seq_file_net(seq)); | |
4073 | else | |
4074 | dev = next_net_device_rcu(dev); | |
c6d14c84 | 4075 | |
1da177e4 | 4076 | ++*pos; |
ccf43438 | 4077 | return dev; |
1da177e4 LT |
4078 | } |
4079 | ||
4080 | void dev_seq_stop(struct seq_file *seq, void *v) | |
c6d14c84 | 4081 | __releases(RCU) |
1da177e4 | 4082 | { |
c6d14c84 | 4083 | rcu_read_unlock(); |
1da177e4 LT |
4084 | } |
4085 | ||
4086 | static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev) | |
4087 | { | |
28172739 ED |
4088 | struct rtnl_link_stats64 temp; |
4089 | const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp); | |
1da177e4 | 4090 | |
be1f3c2c BH |
4091 | seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu " |
4092 | "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n", | |
5a1b5898 RR |
4093 | dev->name, stats->rx_bytes, stats->rx_packets, |
4094 | stats->rx_errors, | |
4095 | stats->rx_dropped + stats->rx_missed_errors, | |
4096 | stats->rx_fifo_errors, | |
4097 | stats->rx_length_errors + stats->rx_over_errors + | |
4098 | stats->rx_crc_errors + stats->rx_frame_errors, | |
4099 | stats->rx_compressed, stats->multicast, | |
4100 | stats->tx_bytes, stats->tx_packets, | |
4101 | stats->tx_errors, stats->tx_dropped, | |
4102 | stats->tx_fifo_errors, stats->collisions, | |
4103 | stats->tx_carrier_errors + | |
4104 | stats->tx_aborted_errors + | |
4105 | stats->tx_window_errors + | |
4106 | stats->tx_heartbeat_errors, | |
4107 | stats->tx_compressed); | |
1da177e4 LT |
4108 | } |
4109 | ||
4110 | /* | |
4111 | * Called from the PROCfs module. This now uses the new arbitrary sized | |
4112 | * /proc/net interface to create /proc/net/dev | |
4113 | */ | |
4114 | static int dev_seq_show(struct seq_file *seq, void *v) | |
4115 | { | |
4116 | if (v == SEQ_START_TOKEN) | |
4117 | seq_puts(seq, "Inter-| Receive " | |
4118 | " | Transmit\n" | |
4119 | " face |bytes packets errs drop fifo frame " | |
4120 | "compressed multicast|bytes packets errs " | |
4121 | "drop fifo colls carrier compressed\n"); | |
4122 | else | |
4123 | dev_seq_printf_stats(seq, v); | |
4124 | return 0; | |
4125 | } | |
4126 | ||
dee42870 | 4127 | static struct softnet_data *softnet_get_online(loff_t *pos) |
1da177e4 | 4128 | { |
dee42870 | 4129 | struct softnet_data *sd = NULL; |
1da177e4 | 4130 | |
0c0b0aca | 4131 | while (*pos < nr_cpu_ids) |
4ec93edb | 4132 | if (cpu_online(*pos)) { |
dee42870 | 4133 | sd = &per_cpu(softnet_data, *pos); |
1da177e4 LT |
4134 | break; |
4135 | } else | |
4136 | ++*pos; | |
dee42870 | 4137 | return sd; |
1da177e4 LT |
4138 | } |
4139 | ||
4140 | static void *softnet_seq_start(struct seq_file *seq, loff_t *pos) | |
4141 | { | |
4142 | return softnet_get_online(pos); | |
4143 | } | |
4144 | ||
4145 | static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
4146 | { | |
4147 | ++*pos; | |
4148 | return softnet_get_online(pos); | |
4149 | } | |
4150 | ||
4151 | static void softnet_seq_stop(struct seq_file *seq, void *v) | |
4152 | { | |
4153 | } | |
4154 | ||
4155 | static int softnet_seq_show(struct seq_file *seq, void *v) | |
4156 | { | |
dee42870 | 4157 | struct softnet_data *sd = v; |
1da177e4 | 4158 | |
0a9627f2 | 4159 | seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n", |
dee42870 | 4160 | sd->processed, sd->dropped, sd->time_squeeze, 0, |
c1ebcdb8 | 4161 | 0, 0, 0, 0, /* was fastroute */ |
dee42870 | 4162 | sd->cpu_collision, sd->received_rps); |
1da177e4 LT |
4163 | return 0; |
4164 | } | |
4165 | ||
f690808e | 4166 | static const struct seq_operations dev_seq_ops = { |
1da177e4 LT |
4167 | .start = dev_seq_start, |
4168 | .next = dev_seq_next, | |
4169 | .stop = dev_seq_stop, | |
4170 | .show = dev_seq_show, | |
4171 | }; | |
4172 | ||
4173 | static int dev_seq_open(struct inode *inode, struct file *file) | |
4174 | { | |
e372c414 DL |
4175 | return seq_open_net(inode, file, &dev_seq_ops, |
4176 | sizeof(struct seq_net_private)); | |
1da177e4 LT |
4177 | } |
4178 | ||
9a32144e | 4179 | static const struct file_operations dev_seq_fops = { |
1da177e4 LT |
4180 | .owner = THIS_MODULE, |
4181 | .open = dev_seq_open, | |
4182 | .read = seq_read, | |
4183 | .llseek = seq_lseek, | |
e372c414 | 4184 | .release = seq_release_net, |
1da177e4 LT |
4185 | }; |
4186 | ||
f690808e | 4187 | static const struct seq_operations softnet_seq_ops = { |
1da177e4 LT |
4188 | .start = softnet_seq_start, |
4189 | .next = softnet_seq_next, | |
4190 | .stop = softnet_seq_stop, | |
4191 | .show = softnet_seq_show, | |
4192 | }; | |
4193 | ||
4194 | static int softnet_seq_open(struct inode *inode, struct file *file) | |
4195 | { | |
4196 | return seq_open(file, &softnet_seq_ops); | |
4197 | } | |
4198 | ||
9a32144e | 4199 | static const struct file_operations softnet_seq_fops = { |
1da177e4 LT |
4200 | .owner = THIS_MODULE, |
4201 | .open = softnet_seq_open, | |
4202 | .read = seq_read, | |
4203 | .llseek = seq_lseek, | |
4204 | .release = seq_release, | |
4205 | }; | |
4206 | ||
0e1256ff SH |
4207 | static void *ptype_get_idx(loff_t pos) |
4208 | { | |
4209 | struct packet_type *pt = NULL; | |
4210 | loff_t i = 0; | |
4211 | int t; | |
4212 | ||
4213 | list_for_each_entry_rcu(pt, &ptype_all, list) { | |
4214 | if (i == pos) | |
4215 | return pt; | |
4216 | ++i; | |
4217 | } | |
4218 | ||
82d8a867 | 4219 | for (t = 0; t < PTYPE_HASH_SIZE; t++) { |
0e1256ff SH |
4220 | list_for_each_entry_rcu(pt, &ptype_base[t], list) { |
4221 | if (i == pos) | |
4222 | return pt; | |
4223 | ++i; | |
4224 | } | |
4225 | } | |
4226 | return NULL; | |
4227 | } | |
4228 | ||
4229 | static void *ptype_seq_start(struct seq_file *seq, loff_t *pos) | |
72348a42 | 4230 | __acquires(RCU) |
0e1256ff SH |
4231 | { |
4232 | rcu_read_lock(); | |
4233 | return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN; | |
4234 | } | |
4235 | ||
4236 | static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
4237 | { | |
4238 | struct packet_type *pt; | |
4239 | struct list_head *nxt; | |
4240 | int hash; | |
4241 | ||
4242 | ++*pos; | |
4243 | if (v == SEQ_START_TOKEN) | |
4244 | return ptype_get_idx(0); | |
4245 | ||
4246 | pt = v; | |
4247 | nxt = pt->list.next; | |
4248 | if (pt->type == htons(ETH_P_ALL)) { | |
4249 | if (nxt != &ptype_all) | |
4250 | goto found; | |
4251 | hash = 0; | |
4252 | nxt = ptype_base[0].next; | |
4253 | } else | |
82d8a867 | 4254 | hash = ntohs(pt->type) & PTYPE_HASH_MASK; |
0e1256ff SH |
4255 | |
4256 | while (nxt == &ptype_base[hash]) { | |
82d8a867 | 4257 | if (++hash >= PTYPE_HASH_SIZE) |
0e1256ff SH |
4258 | return NULL; |
4259 | nxt = ptype_base[hash].next; | |
4260 | } | |
4261 | found: | |
4262 | return list_entry(nxt, struct packet_type, list); | |
4263 | } | |
4264 | ||
4265 | static void ptype_seq_stop(struct seq_file *seq, void *v) | |
72348a42 | 4266 | __releases(RCU) |
0e1256ff SH |
4267 | { |
4268 | rcu_read_unlock(); | |
4269 | } | |
4270 | ||
0e1256ff SH |
4271 | static int ptype_seq_show(struct seq_file *seq, void *v) |
4272 | { | |
4273 | struct packet_type *pt = v; | |
4274 | ||
4275 | if (v == SEQ_START_TOKEN) | |
4276 | seq_puts(seq, "Type Device Function\n"); | |
c346dca1 | 4277 | else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) { |
0e1256ff SH |
4278 | if (pt->type == htons(ETH_P_ALL)) |
4279 | seq_puts(seq, "ALL "); | |
4280 | else | |
4281 | seq_printf(seq, "%04x", ntohs(pt->type)); | |
4282 | ||
908cd2da AD |
4283 | seq_printf(seq, " %-8s %pF\n", |
4284 | pt->dev ? pt->dev->name : "", pt->func); | |
0e1256ff SH |
4285 | } |
4286 | ||
4287 | return 0; | |
4288 | } | |
4289 | ||
4290 | static const struct seq_operations ptype_seq_ops = { | |
4291 | .start = ptype_seq_start, | |
4292 | .next = ptype_seq_next, | |
4293 | .stop = ptype_seq_stop, | |
4294 | .show = ptype_seq_show, | |
4295 | }; | |
4296 | ||
4297 | static int ptype_seq_open(struct inode *inode, struct file *file) | |
4298 | { | |
2feb27db PE |
4299 | return seq_open_net(inode, file, &ptype_seq_ops, |
4300 | sizeof(struct seq_net_private)); | |
0e1256ff SH |
4301 | } |
4302 | ||
4303 | static const struct file_operations ptype_seq_fops = { | |
4304 | .owner = THIS_MODULE, | |
4305 | .open = ptype_seq_open, | |
4306 | .read = seq_read, | |
4307 | .llseek = seq_lseek, | |
2feb27db | 4308 | .release = seq_release_net, |
0e1256ff SH |
4309 | }; |
4310 | ||
4311 | ||
4665079c | 4312 | static int __net_init dev_proc_net_init(struct net *net) |
1da177e4 LT |
4313 | { |
4314 | int rc = -ENOMEM; | |
4315 | ||
881d966b | 4316 | if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops)) |
1da177e4 | 4317 | goto out; |
881d966b | 4318 | if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops)) |
1da177e4 | 4319 | goto out_dev; |
881d966b | 4320 | if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops)) |
457c4cbc | 4321 | goto out_softnet; |
0e1256ff | 4322 | |
881d966b | 4323 | if (wext_proc_init(net)) |
457c4cbc | 4324 | goto out_ptype; |
1da177e4 LT |
4325 | rc = 0; |
4326 | out: | |
4327 | return rc; | |
457c4cbc | 4328 | out_ptype: |
881d966b | 4329 | proc_net_remove(net, "ptype"); |
1da177e4 | 4330 | out_softnet: |
881d966b | 4331 | proc_net_remove(net, "softnet_stat"); |
1da177e4 | 4332 | out_dev: |
881d966b | 4333 | proc_net_remove(net, "dev"); |
1da177e4 LT |
4334 | goto out; |
4335 | } | |
881d966b | 4336 | |
4665079c | 4337 | static void __net_exit dev_proc_net_exit(struct net *net) |
881d966b EB |
4338 | { |
4339 | wext_proc_exit(net); | |
4340 | ||
4341 | proc_net_remove(net, "ptype"); | |
4342 | proc_net_remove(net, "softnet_stat"); | |
4343 | proc_net_remove(net, "dev"); | |
4344 | } | |
4345 | ||
022cbae6 | 4346 | static struct pernet_operations __net_initdata dev_proc_ops = { |
881d966b EB |
4347 | .init = dev_proc_net_init, |
4348 | .exit = dev_proc_net_exit, | |
4349 | }; | |
4350 | ||
4351 | static int __init dev_proc_init(void) | |
4352 | { | |
4353 | return register_pernet_subsys(&dev_proc_ops); | |
4354 | } | |
1da177e4 LT |
4355 | #else |
4356 | #define dev_proc_init() 0 | |
4357 | #endif /* CONFIG_PROC_FS */ | |
4358 | ||
4359 | ||
4360 | /** | |
1765a575 | 4361 | * netdev_set_master - set up master pointer |
1da177e4 LT |
4362 | * @slave: slave device |
4363 | * @master: new master device | |
4364 | * | |
4365 | * Changes the master device of the slave. Pass %NULL to break the | |
4366 | * bonding. The caller must hold the RTNL semaphore. On a failure | |
4367 | * a negative errno code is returned. On success the reference counts | |
1765a575 | 4368 | * are adjusted and the function returns zero. |
1da177e4 LT |
4369 | */ |
4370 | int netdev_set_master(struct net_device *slave, struct net_device *master) | |
4371 | { | |
4372 | struct net_device *old = slave->master; | |
4373 | ||
4374 | ASSERT_RTNL(); | |
4375 | ||
4376 | if (master) { | |
4377 | if (old) | |
4378 | return -EBUSY; | |
4379 | dev_hold(master); | |
4380 | } | |
4381 | ||
4382 | slave->master = master; | |
4ec93edb | 4383 | |
6df427fe | 4384 | if (old) |
1da177e4 | 4385 | dev_put(old); |
1765a575 JP |
4386 | return 0; |
4387 | } | |
4388 | EXPORT_SYMBOL(netdev_set_master); | |
4389 | ||
4390 | /** | |
4391 | * netdev_set_bond_master - set up bonding master/slave pair | |
4392 | * @slave: slave device | |
4393 | * @master: new master device | |
4394 | * | |
4395 | * Changes the master device of the slave. Pass %NULL to break the | |
4396 | * bonding. The caller must hold the RTNL semaphore. On a failure | |
4397 | * a negative errno code is returned. On success %RTM_NEWLINK is sent | |
4398 | * to the routing socket and the function returns zero. | |
4399 | */ | |
4400 | int netdev_set_bond_master(struct net_device *slave, struct net_device *master) | |
4401 | { | |
4402 | int err; | |
4403 | ||
4404 | ASSERT_RTNL(); | |
4405 | ||
4406 | err = netdev_set_master(slave, master); | |
4407 | if (err) | |
4408 | return err; | |
1da177e4 LT |
4409 | if (master) |
4410 | slave->flags |= IFF_SLAVE; | |
4411 | else | |
4412 | slave->flags &= ~IFF_SLAVE; | |
4413 | ||
4414 | rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE); | |
4415 | return 0; | |
4416 | } | |
1765a575 | 4417 | EXPORT_SYMBOL(netdev_set_bond_master); |
1da177e4 | 4418 | |
b6c40d68 PM |
4419 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
4420 | { | |
d314774c SH |
4421 | const struct net_device_ops *ops = dev->netdev_ops; |
4422 | ||
4423 | if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags) | |
4424 | ops->ndo_change_rx_flags(dev, flags); | |
b6c40d68 PM |
4425 | } |
4426 | ||
dad9b335 | 4427 | static int __dev_set_promiscuity(struct net_device *dev, int inc) |
1da177e4 LT |
4428 | { |
4429 | unsigned short old_flags = dev->flags; | |
8192b0c4 DH |
4430 | uid_t uid; |
4431 | gid_t gid; | |
1da177e4 | 4432 | |
24023451 PM |
4433 | ASSERT_RTNL(); |
4434 | ||
dad9b335 WC |
4435 | dev->flags |= IFF_PROMISC; |
4436 | dev->promiscuity += inc; | |
4437 | if (dev->promiscuity == 0) { | |
4438 | /* | |
4439 | * Avoid overflow. | |
4440 | * If inc causes overflow, untouch promisc and return error. | |
4441 | */ | |
4442 | if (inc < 0) | |
4443 | dev->flags &= ~IFF_PROMISC; | |
4444 | else { | |
4445 | dev->promiscuity -= inc; | |
4446 | printk(KERN_WARNING "%s: promiscuity touches roof, " | |
4447 | "set promiscuity failed, promiscuity feature " | |
4448 | "of device might be broken.\n", dev->name); | |
4449 | return -EOVERFLOW; | |
4450 | } | |
4451 | } | |
52609c0b | 4452 | if (dev->flags != old_flags) { |
1da177e4 LT |
4453 | printk(KERN_INFO "device %s %s promiscuous mode\n", |
4454 | dev->name, (dev->flags & IFF_PROMISC) ? "entered" : | |
4ec93edb | 4455 | "left"); |
8192b0c4 DH |
4456 | if (audit_enabled) { |
4457 | current_uid_gid(&uid, &gid); | |
7759db82 KHK |
4458 | audit_log(current->audit_context, GFP_ATOMIC, |
4459 | AUDIT_ANOM_PROMISCUOUS, | |
4460 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
4461 | dev->name, (dev->flags & IFF_PROMISC), | |
4462 | (old_flags & IFF_PROMISC), | |
4463 | audit_get_loginuid(current), | |
8192b0c4 | 4464 | uid, gid, |
7759db82 | 4465 | audit_get_sessionid(current)); |
8192b0c4 | 4466 | } |
24023451 | 4467 | |
b6c40d68 | 4468 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 4469 | } |
dad9b335 | 4470 | return 0; |
1da177e4 LT |
4471 | } |
4472 | ||
4417da66 PM |
4473 | /** |
4474 | * dev_set_promiscuity - update promiscuity count on a device | |
4475 | * @dev: device | |
4476 | * @inc: modifier | |
4477 | * | |
4478 | * Add or remove promiscuity from a device. While the count in the device | |
4479 | * remains above zero the interface remains promiscuous. Once it hits zero | |
4480 | * the device reverts back to normal filtering operation. A negative inc | |
4481 | * value is used to drop promiscuity on the device. | |
dad9b335 | 4482 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 4483 | */ |
dad9b335 | 4484 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 PM |
4485 | { |
4486 | unsigned short old_flags = dev->flags; | |
dad9b335 | 4487 | int err; |
4417da66 | 4488 | |
dad9b335 | 4489 | err = __dev_set_promiscuity(dev, inc); |
4b5a698e | 4490 | if (err < 0) |
dad9b335 | 4491 | return err; |
4417da66 PM |
4492 | if (dev->flags != old_flags) |
4493 | dev_set_rx_mode(dev); | |
dad9b335 | 4494 | return err; |
4417da66 | 4495 | } |
d1b19dff | 4496 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 4497 | |
1da177e4 LT |
4498 | /** |
4499 | * dev_set_allmulti - update allmulti count on a device | |
4500 | * @dev: device | |
4501 | * @inc: modifier | |
4502 | * | |
4503 | * Add or remove reception of all multicast frames to a device. While the | |
4504 | * count in the device remains above zero the interface remains listening | |
4505 | * to all interfaces. Once it hits zero the device reverts back to normal | |
4506 | * filtering operation. A negative @inc value is used to drop the counter | |
4507 | * when releasing a resource needing all multicasts. | |
dad9b335 | 4508 | * Return 0 if successful or a negative errno code on error. |
1da177e4 LT |
4509 | */ |
4510 | ||
dad9b335 | 4511 | int dev_set_allmulti(struct net_device *dev, int inc) |
1da177e4 LT |
4512 | { |
4513 | unsigned short old_flags = dev->flags; | |
4514 | ||
24023451 PM |
4515 | ASSERT_RTNL(); |
4516 | ||
1da177e4 | 4517 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
4518 | dev->allmulti += inc; |
4519 | if (dev->allmulti == 0) { | |
4520 | /* | |
4521 | * Avoid overflow. | |
4522 | * If inc causes overflow, untouch allmulti and return error. | |
4523 | */ | |
4524 | if (inc < 0) | |
4525 | dev->flags &= ~IFF_ALLMULTI; | |
4526 | else { | |
4527 | dev->allmulti -= inc; | |
4528 | printk(KERN_WARNING "%s: allmulti touches roof, " | |
4529 | "set allmulti failed, allmulti feature of " | |
4530 | "device might be broken.\n", dev->name); | |
4531 | return -EOVERFLOW; | |
4532 | } | |
4533 | } | |
24023451 | 4534 | if (dev->flags ^ old_flags) { |
b6c40d68 | 4535 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 4536 | dev_set_rx_mode(dev); |
24023451 | 4537 | } |
dad9b335 | 4538 | return 0; |
4417da66 | 4539 | } |
d1b19dff | 4540 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
4541 | |
4542 | /* | |
4543 | * Upload unicast and multicast address lists to device and | |
4544 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 4545 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
4546 | * are present. |
4547 | */ | |
4548 | void __dev_set_rx_mode(struct net_device *dev) | |
4549 | { | |
d314774c SH |
4550 | const struct net_device_ops *ops = dev->netdev_ops; |
4551 | ||
4417da66 PM |
4552 | /* dev_open will call this function so the list will stay sane. */ |
4553 | if (!(dev->flags&IFF_UP)) | |
4554 | return; | |
4555 | ||
4556 | if (!netif_device_present(dev)) | |
40b77c94 | 4557 | return; |
4417da66 | 4558 | |
01789349 | 4559 | if (!(dev->priv_flags & IFF_UNICAST_FLT)) { |
4417da66 PM |
4560 | /* Unicast addresses changes may only happen under the rtnl, |
4561 | * therefore calling __dev_set_promiscuity here is safe. | |
4562 | */ | |
32e7bfc4 | 4563 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
4417da66 | 4564 | __dev_set_promiscuity(dev, 1); |
2d348d1f | 4565 | dev->uc_promisc = true; |
32e7bfc4 | 4566 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
4417da66 | 4567 | __dev_set_promiscuity(dev, -1); |
2d348d1f | 4568 | dev->uc_promisc = false; |
4417da66 | 4569 | } |
4417da66 | 4570 | } |
01789349 JP |
4571 | |
4572 | if (ops->ndo_set_rx_mode) | |
4573 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
4574 | } |
4575 | ||
4576 | void dev_set_rx_mode(struct net_device *dev) | |
4577 | { | |
b9e40857 | 4578 | netif_addr_lock_bh(dev); |
4417da66 | 4579 | __dev_set_rx_mode(dev); |
b9e40857 | 4580 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
4581 | } |
4582 | ||
f0db275a SH |
4583 | /** |
4584 | * dev_get_flags - get flags reported to userspace | |
4585 | * @dev: device | |
4586 | * | |
4587 | * Get the combination of flag bits exported through APIs to userspace. | |
4588 | */ | |
1da177e4 LT |
4589 | unsigned dev_get_flags(const struct net_device *dev) |
4590 | { | |
4591 | unsigned flags; | |
4592 | ||
4593 | flags = (dev->flags & ~(IFF_PROMISC | | |
4594 | IFF_ALLMULTI | | |
b00055aa SR |
4595 | IFF_RUNNING | |
4596 | IFF_LOWER_UP | | |
4597 | IFF_DORMANT)) | | |
1da177e4 LT |
4598 | (dev->gflags & (IFF_PROMISC | |
4599 | IFF_ALLMULTI)); | |
4600 | ||
b00055aa SR |
4601 | if (netif_running(dev)) { |
4602 | if (netif_oper_up(dev)) | |
4603 | flags |= IFF_RUNNING; | |
4604 | if (netif_carrier_ok(dev)) | |
4605 | flags |= IFF_LOWER_UP; | |
4606 | if (netif_dormant(dev)) | |
4607 | flags |= IFF_DORMANT; | |
4608 | } | |
1da177e4 LT |
4609 | |
4610 | return flags; | |
4611 | } | |
d1b19dff | 4612 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 4613 | |
bd380811 | 4614 | int __dev_change_flags(struct net_device *dev, unsigned int flags) |
1da177e4 | 4615 | { |
1da177e4 | 4616 | int old_flags = dev->flags; |
bd380811 | 4617 | int ret; |
1da177e4 | 4618 | |
24023451 PM |
4619 | ASSERT_RTNL(); |
4620 | ||
1da177e4 LT |
4621 | /* |
4622 | * Set the flags on our device. | |
4623 | */ | |
4624 | ||
4625 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
4626 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
4627 | IFF_AUTOMEDIA)) | | |
4628 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
4629 | IFF_ALLMULTI)); | |
4630 | ||
4631 | /* | |
4632 | * Load in the correct multicast list now the flags have changed. | |
4633 | */ | |
4634 | ||
b6c40d68 PM |
4635 | if ((old_flags ^ flags) & IFF_MULTICAST) |
4636 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 4637 | |
4417da66 | 4638 | dev_set_rx_mode(dev); |
1da177e4 LT |
4639 | |
4640 | /* | |
4641 | * Have we downed the interface. We handle IFF_UP ourselves | |
4642 | * according to user attempts to set it, rather than blindly | |
4643 | * setting it. | |
4644 | */ | |
4645 | ||
4646 | ret = 0; | |
4647 | if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */ | |
bd380811 | 4648 | ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev); |
1da177e4 LT |
4649 | |
4650 | if (!ret) | |
4417da66 | 4651 | dev_set_rx_mode(dev); |
1da177e4 LT |
4652 | } |
4653 | ||
1da177e4 | 4654 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff ED |
4655 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
4656 | ||
1da177e4 LT |
4657 | dev->gflags ^= IFF_PROMISC; |
4658 | dev_set_promiscuity(dev, inc); | |
4659 | } | |
4660 | ||
4661 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
4662 | is important. Some (broken) drivers set IFF_PROMISC, when | |
4663 | IFF_ALLMULTI is requested not asking us and not reporting. | |
4664 | */ | |
4665 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
4666 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
4667 | ||
1da177e4 LT |
4668 | dev->gflags ^= IFF_ALLMULTI; |
4669 | dev_set_allmulti(dev, inc); | |
4670 | } | |
4671 | ||
bd380811 PM |
4672 | return ret; |
4673 | } | |
4674 | ||
4675 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags) | |
4676 | { | |
4677 | unsigned int changes = dev->flags ^ old_flags; | |
4678 | ||
4679 | if (changes & IFF_UP) { | |
4680 | if (dev->flags & IFF_UP) | |
4681 | call_netdevice_notifiers(NETDEV_UP, dev); | |
4682 | else | |
4683 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
4684 | } | |
4685 | ||
4686 | if (dev->flags & IFF_UP && | |
4687 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) | |
4688 | call_netdevice_notifiers(NETDEV_CHANGE, dev); | |
4689 | } | |
4690 | ||
4691 | /** | |
4692 | * dev_change_flags - change device settings | |
4693 | * @dev: device | |
4694 | * @flags: device state flags | |
4695 | * | |
4696 | * Change settings on device based state flags. The flags are | |
4697 | * in the userspace exported format. | |
4698 | */ | |
4699 | int dev_change_flags(struct net_device *dev, unsigned flags) | |
4700 | { | |
4701 | int ret, changes; | |
4702 | int old_flags = dev->flags; | |
4703 | ||
4704 | ret = __dev_change_flags(dev, flags); | |
4705 | if (ret < 0) | |
4706 | return ret; | |
4707 | ||
4708 | changes = old_flags ^ dev->flags; | |
7c355f53 TG |
4709 | if (changes) |
4710 | rtmsg_ifinfo(RTM_NEWLINK, dev, changes); | |
1da177e4 | 4711 | |
bd380811 | 4712 | __dev_notify_flags(dev, old_flags); |
1da177e4 LT |
4713 | return ret; |
4714 | } | |
d1b19dff | 4715 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 4716 | |
f0db275a SH |
4717 | /** |
4718 | * dev_set_mtu - Change maximum transfer unit | |
4719 | * @dev: device | |
4720 | * @new_mtu: new transfer unit | |
4721 | * | |
4722 | * Change the maximum transfer size of the network device. | |
4723 | */ | |
1da177e4 LT |
4724 | int dev_set_mtu(struct net_device *dev, int new_mtu) |
4725 | { | |
d314774c | 4726 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
4727 | int err; |
4728 | ||
4729 | if (new_mtu == dev->mtu) | |
4730 | return 0; | |
4731 | ||
4732 | /* MTU must be positive. */ | |
4733 | if (new_mtu < 0) | |
4734 | return -EINVAL; | |
4735 | ||
4736 | if (!netif_device_present(dev)) | |
4737 | return -ENODEV; | |
4738 | ||
4739 | err = 0; | |
d314774c SH |
4740 | if (ops->ndo_change_mtu) |
4741 | err = ops->ndo_change_mtu(dev, new_mtu); | |
1da177e4 LT |
4742 | else |
4743 | dev->mtu = new_mtu; | |
d314774c | 4744 | |
1da177e4 | 4745 | if (!err && dev->flags & IFF_UP) |
056925ab | 4746 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); |
1da177e4 LT |
4747 | return err; |
4748 | } | |
d1b19dff | 4749 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 4750 | |
cbda10fa VD |
4751 | /** |
4752 | * dev_set_group - Change group this device belongs to | |
4753 | * @dev: device | |
4754 | * @new_group: group this device should belong to | |
4755 | */ | |
4756 | void dev_set_group(struct net_device *dev, int new_group) | |
4757 | { | |
4758 | dev->group = new_group; | |
4759 | } | |
4760 | EXPORT_SYMBOL(dev_set_group); | |
4761 | ||
f0db275a SH |
4762 | /** |
4763 | * dev_set_mac_address - Change Media Access Control Address | |
4764 | * @dev: device | |
4765 | * @sa: new address | |
4766 | * | |
4767 | * Change the hardware (MAC) address of the device | |
4768 | */ | |
1da177e4 LT |
4769 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) |
4770 | { | |
d314774c | 4771 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
4772 | int err; |
4773 | ||
d314774c | 4774 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
4775 | return -EOPNOTSUPP; |
4776 | if (sa->sa_family != dev->type) | |
4777 | return -EINVAL; | |
4778 | if (!netif_device_present(dev)) | |
4779 | return -ENODEV; | |
d314774c | 4780 | err = ops->ndo_set_mac_address(dev, sa); |
1da177e4 | 4781 | if (!err) |
056925ab | 4782 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
1da177e4 LT |
4783 | return err; |
4784 | } | |
d1b19dff | 4785 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 LT |
4786 | |
4787 | /* | |
3710becf | 4788 | * Perform the SIOCxIFxxx calls, inside rcu_read_lock() |
1da177e4 | 4789 | */ |
14e3e079 | 4790 | static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd) |
1da177e4 LT |
4791 | { |
4792 | int err; | |
3710becf | 4793 | struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name); |
1da177e4 LT |
4794 | |
4795 | if (!dev) | |
4796 | return -ENODEV; | |
4797 | ||
4798 | switch (cmd) { | |
d1b19dff ED |
4799 | case SIOCGIFFLAGS: /* Get interface flags */ |
4800 | ifr->ifr_flags = (short) dev_get_flags(dev); | |
4801 | return 0; | |
1da177e4 | 4802 | |
d1b19dff ED |
4803 | case SIOCGIFMETRIC: /* Get the metric on the interface |
4804 | (currently unused) */ | |
4805 | ifr->ifr_metric = 0; | |
4806 | return 0; | |
1da177e4 | 4807 | |
d1b19dff ED |
4808 | case SIOCGIFMTU: /* Get the MTU of a device */ |
4809 | ifr->ifr_mtu = dev->mtu; | |
4810 | return 0; | |
1da177e4 | 4811 | |
d1b19dff ED |
4812 | case SIOCGIFHWADDR: |
4813 | if (!dev->addr_len) | |
4814 | memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data); | |
4815 | else | |
4816 | memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr, | |
4817 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
4818 | ifr->ifr_hwaddr.sa_family = dev->type; | |
4819 | return 0; | |
1da177e4 | 4820 | |
d1b19dff ED |
4821 | case SIOCGIFSLAVE: |
4822 | err = -EINVAL; | |
4823 | break; | |
14e3e079 | 4824 | |
d1b19dff ED |
4825 | case SIOCGIFMAP: |
4826 | ifr->ifr_map.mem_start = dev->mem_start; | |
4827 | ifr->ifr_map.mem_end = dev->mem_end; | |
4828 | ifr->ifr_map.base_addr = dev->base_addr; | |
4829 | ifr->ifr_map.irq = dev->irq; | |
4830 | ifr->ifr_map.dma = dev->dma; | |
4831 | ifr->ifr_map.port = dev->if_port; | |
4832 | return 0; | |
14e3e079 | 4833 | |
d1b19dff ED |
4834 | case SIOCGIFINDEX: |
4835 | ifr->ifr_ifindex = dev->ifindex; | |
4836 | return 0; | |
14e3e079 | 4837 | |
d1b19dff ED |
4838 | case SIOCGIFTXQLEN: |
4839 | ifr->ifr_qlen = dev->tx_queue_len; | |
4840 | return 0; | |
14e3e079 | 4841 | |
d1b19dff ED |
4842 | default: |
4843 | /* dev_ioctl() should ensure this case | |
4844 | * is never reached | |
4845 | */ | |
4846 | WARN_ON(1); | |
41c31f31 | 4847 | err = -ENOTTY; |
d1b19dff | 4848 | break; |
14e3e079 JG |
4849 | |
4850 | } | |
4851 | return err; | |
4852 | } | |
4853 | ||
4854 | /* | |
4855 | * Perform the SIOCxIFxxx calls, inside rtnl_lock() | |
4856 | */ | |
4857 | static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd) | |
4858 | { | |
4859 | int err; | |
4860 | struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name); | |
5f2f6da7 | 4861 | const struct net_device_ops *ops; |
14e3e079 JG |
4862 | |
4863 | if (!dev) | |
4864 | return -ENODEV; | |
4865 | ||
5f2f6da7 JP |
4866 | ops = dev->netdev_ops; |
4867 | ||
14e3e079 | 4868 | switch (cmd) { |
d1b19dff ED |
4869 | case SIOCSIFFLAGS: /* Set interface flags */ |
4870 | return dev_change_flags(dev, ifr->ifr_flags); | |
14e3e079 | 4871 | |
d1b19dff ED |
4872 | case SIOCSIFMETRIC: /* Set the metric on the interface |
4873 | (currently unused) */ | |
4874 | return -EOPNOTSUPP; | |
14e3e079 | 4875 | |
d1b19dff ED |
4876 | case SIOCSIFMTU: /* Set the MTU of a device */ |
4877 | return dev_set_mtu(dev, ifr->ifr_mtu); | |
1da177e4 | 4878 | |
d1b19dff ED |
4879 | case SIOCSIFHWADDR: |
4880 | return dev_set_mac_address(dev, &ifr->ifr_hwaddr); | |
1da177e4 | 4881 | |
d1b19dff ED |
4882 | case SIOCSIFHWBROADCAST: |
4883 | if (ifr->ifr_hwaddr.sa_family != dev->type) | |
4884 | return -EINVAL; | |
4885 | memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data, | |
4886 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
4887 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); | |
4888 | return 0; | |
1da177e4 | 4889 | |
d1b19dff ED |
4890 | case SIOCSIFMAP: |
4891 | if (ops->ndo_set_config) { | |
1da177e4 LT |
4892 | if (!netif_device_present(dev)) |
4893 | return -ENODEV; | |
d1b19dff ED |
4894 | return ops->ndo_set_config(dev, &ifr->ifr_map); |
4895 | } | |
4896 | return -EOPNOTSUPP; | |
1da177e4 | 4897 | |
d1b19dff | 4898 | case SIOCADDMULTI: |
b81693d9 | 4899 | if (!ops->ndo_set_rx_mode || |
d1b19dff ED |
4900 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) |
4901 | return -EINVAL; | |
4902 | if (!netif_device_present(dev)) | |
4903 | return -ENODEV; | |
22bedad3 | 4904 | return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data); |
d1b19dff ED |
4905 | |
4906 | case SIOCDELMULTI: | |
b81693d9 | 4907 | if (!ops->ndo_set_rx_mode || |
d1b19dff ED |
4908 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) |
4909 | return -EINVAL; | |
4910 | if (!netif_device_present(dev)) | |
4911 | return -ENODEV; | |
22bedad3 | 4912 | return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data); |
1da177e4 | 4913 | |
d1b19dff ED |
4914 | case SIOCSIFTXQLEN: |
4915 | if (ifr->ifr_qlen < 0) | |
4916 | return -EINVAL; | |
4917 | dev->tx_queue_len = ifr->ifr_qlen; | |
4918 | return 0; | |
1da177e4 | 4919 | |
d1b19dff ED |
4920 | case SIOCSIFNAME: |
4921 | ifr->ifr_newname[IFNAMSIZ-1] = '\0'; | |
4922 | return dev_change_name(dev, ifr->ifr_newname); | |
1da177e4 | 4923 | |
d1b19dff ED |
4924 | /* |
4925 | * Unknown or private ioctl | |
4926 | */ | |
4927 | default: | |
4928 | if ((cmd >= SIOCDEVPRIVATE && | |
4929 | cmd <= SIOCDEVPRIVATE + 15) || | |
4930 | cmd == SIOCBONDENSLAVE || | |
4931 | cmd == SIOCBONDRELEASE || | |
4932 | cmd == SIOCBONDSETHWADDR || | |
4933 | cmd == SIOCBONDSLAVEINFOQUERY || | |
4934 | cmd == SIOCBONDINFOQUERY || | |
4935 | cmd == SIOCBONDCHANGEACTIVE || | |
4936 | cmd == SIOCGMIIPHY || | |
4937 | cmd == SIOCGMIIREG || | |
4938 | cmd == SIOCSMIIREG || | |
4939 | cmd == SIOCBRADDIF || | |
4940 | cmd == SIOCBRDELIF || | |
4941 | cmd == SIOCSHWTSTAMP || | |
4942 | cmd == SIOCWANDEV) { | |
4943 | err = -EOPNOTSUPP; | |
4944 | if (ops->ndo_do_ioctl) { | |
4945 | if (netif_device_present(dev)) | |
4946 | err = ops->ndo_do_ioctl(dev, ifr, cmd); | |
4947 | else | |
4948 | err = -ENODEV; | |
4949 | } | |
4950 | } else | |
4951 | err = -EINVAL; | |
1da177e4 LT |
4952 | |
4953 | } | |
4954 | return err; | |
4955 | } | |
4956 | ||
4957 | /* | |
4958 | * This function handles all "interface"-type I/O control requests. The actual | |
4959 | * 'doing' part of this is dev_ifsioc above. | |
4960 | */ | |
4961 | ||
4962 | /** | |
4963 | * dev_ioctl - network device ioctl | |
c4ea43c5 | 4964 | * @net: the applicable net namespace |
1da177e4 LT |
4965 | * @cmd: command to issue |
4966 | * @arg: pointer to a struct ifreq in user space | |
4967 | * | |
4968 | * Issue ioctl functions to devices. This is normally called by the | |
4969 | * user space syscall interfaces but can sometimes be useful for | |
4970 | * other purposes. The return value is the return from the syscall if | |
4971 | * positive or a negative errno code on error. | |
4972 | */ | |
4973 | ||
881d966b | 4974 | int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg) |
1da177e4 LT |
4975 | { |
4976 | struct ifreq ifr; | |
4977 | int ret; | |
4978 | char *colon; | |
4979 | ||
4980 | /* One special case: SIOCGIFCONF takes ifconf argument | |
4981 | and requires shared lock, because it sleeps writing | |
4982 | to user space. | |
4983 | */ | |
4984 | ||
4985 | if (cmd == SIOCGIFCONF) { | |
6756ae4b | 4986 | rtnl_lock(); |
881d966b | 4987 | ret = dev_ifconf(net, (char __user *) arg); |
6756ae4b | 4988 | rtnl_unlock(); |
1da177e4 LT |
4989 | return ret; |
4990 | } | |
4991 | if (cmd == SIOCGIFNAME) | |
881d966b | 4992 | return dev_ifname(net, (struct ifreq __user *)arg); |
1da177e4 LT |
4993 | |
4994 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
4995 | return -EFAULT; | |
4996 | ||
4997 | ifr.ifr_name[IFNAMSIZ-1] = 0; | |
4998 | ||
4999 | colon = strchr(ifr.ifr_name, ':'); | |
5000 | if (colon) | |
5001 | *colon = 0; | |
5002 | ||
5003 | /* | |
5004 | * See which interface the caller is talking about. | |
5005 | */ | |
5006 | ||
5007 | switch (cmd) { | |
d1b19dff ED |
5008 | /* |
5009 | * These ioctl calls: | |
5010 | * - can be done by all. | |
5011 | * - atomic and do not require locking. | |
5012 | * - return a value | |
5013 | */ | |
5014 | case SIOCGIFFLAGS: | |
5015 | case SIOCGIFMETRIC: | |
5016 | case SIOCGIFMTU: | |
5017 | case SIOCGIFHWADDR: | |
5018 | case SIOCGIFSLAVE: | |
5019 | case SIOCGIFMAP: | |
5020 | case SIOCGIFINDEX: | |
5021 | case SIOCGIFTXQLEN: | |
5022 | dev_load(net, ifr.ifr_name); | |
3710becf | 5023 | rcu_read_lock(); |
d1b19dff | 5024 | ret = dev_ifsioc_locked(net, &ifr, cmd); |
3710becf | 5025 | rcu_read_unlock(); |
d1b19dff ED |
5026 | if (!ret) { |
5027 | if (colon) | |
5028 | *colon = ':'; | |
5029 | if (copy_to_user(arg, &ifr, | |
5030 | sizeof(struct ifreq))) | |
5031 | ret = -EFAULT; | |
5032 | } | |
5033 | return ret; | |
1da177e4 | 5034 | |
d1b19dff ED |
5035 | case SIOCETHTOOL: |
5036 | dev_load(net, ifr.ifr_name); | |
5037 | rtnl_lock(); | |
5038 | ret = dev_ethtool(net, &ifr); | |
5039 | rtnl_unlock(); | |
5040 | if (!ret) { | |
5041 | if (colon) | |
5042 | *colon = ':'; | |
5043 | if (copy_to_user(arg, &ifr, | |
5044 | sizeof(struct ifreq))) | |
5045 | ret = -EFAULT; | |
5046 | } | |
5047 | return ret; | |
1da177e4 | 5048 | |
d1b19dff ED |
5049 | /* |
5050 | * These ioctl calls: | |
5051 | * - require superuser power. | |
5052 | * - require strict serialization. | |
5053 | * - return a value | |
5054 | */ | |
5055 | case SIOCGMIIPHY: | |
5056 | case SIOCGMIIREG: | |
5057 | case SIOCSIFNAME: | |
5058 | if (!capable(CAP_NET_ADMIN)) | |
5059 | return -EPERM; | |
5060 | dev_load(net, ifr.ifr_name); | |
5061 | rtnl_lock(); | |
5062 | ret = dev_ifsioc(net, &ifr, cmd); | |
5063 | rtnl_unlock(); | |
5064 | if (!ret) { | |
5065 | if (colon) | |
5066 | *colon = ':'; | |
5067 | if (copy_to_user(arg, &ifr, | |
5068 | sizeof(struct ifreq))) | |
5069 | ret = -EFAULT; | |
5070 | } | |
5071 | return ret; | |
1da177e4 | 5072 | |
d1b19dff ED |
5073 | /* |
5074 | * These ioctl calls: | |
5075 | * - require superuser power. | |
5076 | * - require strict serialization. | |
5077 | * - do not return a value | |
5078 | */ | |
5079 | case SIOCSIFFLAGS: | |
5080 | case SIOCSIFMETRIC: | |
5081 | case SIOCSIFMTU: | |
5082 | case SIOCSIFMAP: | |
5083 | case SIOCSIFHWADDR: | |
5084 | case SIOCSIFSLAVE: | |
5085 | case SIOCADDMULTI: | |
5086 | case SIOCDELMULTI: | |
5087 | case SIOCSIFHWBROADCAST: | |
5088 | case SIOCSIFTXQLEN: | |
5089 | case SIOCSMIIREG: | |
5090 | case SIOCBONDENSLAVE: | |
5091 | case SIOCBONDRELEASE: | |
5092 | case SIOCBONDSETHWADDR: | |
5093 | case SIOCBONDCHANGEACTIVE: | |
5094 | case SIOCBRADDIF: | |
5095 | case SIOCBRDELIF: | |
5096 | case SIOCSHWTSTAMP: | |
5097 | if (!capable(CAP_NET_ADMIN)) | |
5098 | return -EPERM; | |
5099 | /* fall through */ | |
5100 | case SIOCBONDSLAVEINFOQUERY: | |
5101 | case SIOCBONDINFOQUERY: | |
5102 | dev_load(net, ifr.ifr_name); | |
5103 | rtnl_lock(); | |
5104 | ret = dev_ifsioc(net, &ifr, cmd); | |
5105 | rtnl_unlock(); | |
5106 | return ret; | |
5107 | ||
5108 | case SIOCGIFMEM: | |
5109 | /* Get the per device memory space. We can add this but | |
5110 | * currently do not support it */ | |
5111 | case SIOCSIFMEM: | |
5112 | /* Set the per device memory buffer space. | |
5113 | * Not applicable in our case */ | |
5114 | case SIOCSIFLINK: | |
41c31f31 | 5115 | return -ENOTTY; |
d1b19dff ED |
5116 | |
5117 | /* | |
5118 | * Unknown or private ioctl. | |
5119 | */ | |
5120 | default: | |
5121 | if (cmd == SIOCWANDEV || | |
5122 | (cmd >= SIOCDEVPRIVATE && | |
5123 | cmd <= SIOCDEVPRIVATE + 15)) { | |
881d966b | 5124 | dev_load(net, ifr.ifr_name); |
1da177e4 | 5125 | rtnl_lock(); |
881d966b | 5126 | ret = dev_ifsioc(net, &ifr, cmd); |
1da177e4 | 5127 | rtnl_unlock(); |
d1b19dff ED |
5128 | if (!ret && copy_to_user(arg, &ifr, |
5129 | sizeof(struct ifreq))) | |
5130 | ret = -EFAULT; | |
1da177e4 | 5131 | return ret; |
d1b19dff ED |
5132 | } |
5133 | /* Take care of Wireless Extensions */ | |
5134 | if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) | |
5135 | return wext_handle_ioctl(net, &ifr, cmd, arg); | |
41c31f31 | 5136 | return -ENOTTY; |
1da177e4 LT |
5137 | } |
5138 | } | |
5139 | ||
5140 | ||
5141 | /** | |
5142 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 5143 | * @net: the applicable net namespace |
1da177e4 LT |
5144 | * |
5145 | * Returns a suitable unique value for a new device interface | |
5146 | * number. The caller must hold the rtnl semaphore or the | |
5147 | * dev_base_lock to be sure it remains unique. | |
5148 | */ | |
881d966b | 5149 | static int dev_new_index(struct net *net) |
1da177e4 LT |
5150 | { |
5151 | static int ifindex; | |
5152 | for (;;) { | |
5153 | if (++ifindex <= 0) | |
5154 | ifindex = 1; | |
881d966b | 5155 | if (!__dev_get_by_index(net, ifindex)) |
1da177e4 LT |
5156 | return ifindex; |
5157 | } | |
5158 | } | |
5159 | ||
1da177e4 | 5160 | /* Delayed registration/unregisteration */ |
3b5b34fd | 5161 | static LIST_HEAD(net_todo_list); |
1da177e4 | 5162 | |
6f05f629 | 5163 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 5164 | { |
1da177e4 | 5165 | list_add_tail(&dev->todo_list, &net_todo_list); |
1da177e4 LT |
5166 | } |
5167 | ||
9b5e383c | 5168 | static void rollback_registered_many(struct list_head *head) |
93ee31f1 | 5169 | { |
e93737b0 | 5170 | struct net_device *dev, *tmp; |
9b5e383c | 5171 | |
93ee31f1 DL |
5172 | BUG_ON(dev_boot_phase); |
5173 | ASSERT_RTNL(); | |
5174 | ||
e93737b0 | 5175 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
9b5e383c | 5176 | /* Some devices call without registering |
e93737b0 KK |
5177 | * for initialization unwind. Remove those |
5178 | * devices and proceed with the remaining. | |
9b5e383c ED |
5179 | */ |
5180 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
5181 | pr_debug("unregister_netdevice: device %s/%p never " | |
5182 | "was registered\n", dev->name, dev); | |
93ee31f1 | 5183 | |
9b5e383c | 5184 | WARN_ON(1); |
e93737b0 KK |
5185 | list_del(&dev->unreg_list); |
5186 | continue; | |
9b5e383c | 5187 | } |
449f4544 | 5188 | dev->dismantle = true; |
9b5e383c | 5189 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
44345724 | 5190 | } |
93ee31f1 | 5191 | |
44345724 OP |
5192 | /* If device is running, close it first. */ |
5193 | dev_close_many(head); | |
93ee31f1 | 5194 | |
44345724 | 5195 | list_for_each_entry(dev, head, unreg_list) { |
9b5e383c ED |
5196 | /* And unlink it from device chain. */ |
5197 | unlist_netdevice(dev); | |
93ee31f1 | 5198 | |
9b5e383c ED |
5199 | dev->reg_state = NETREG_UNREGISTERING; |
5200 | } | |
93ee31f1 DL |
5201 | |
5202 | synchronize_net(); | |
5203 | ||
9b5e383c ED |
5204 | list_for_each_entry(dev, head, unreg_list) { |
5205 | /* Shutdown queueing discipline. */ | |
5206 | dev_shutdown(dev); | |
93ee31f1 DL |
5207 | |
5208 | ||
9b5e383c ED |
5209 | /* Notify protocols, that we are about to destroy |
5210 | this device. They should clean all the things. | |
5211 | */ | |
5212 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
93ee31f1 | 5213 | |
a2835763 PM |
5214 | if (!dev->rtnl_link_ops || |
5215 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
5216 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U); | |
5217 | ||
9b5e383c ED |
5218 | /* |
5219 | * Flush the unicast and multicast chains | |
5220 | */ | |
a748ee24 | 5221 | dev_uc_flush(dev); |
22bedad3 | 5222 | dev_mc_flush(dev); |
93ee31f1 | 5223 | |
9b5e383c ED |
5224 | if (dev->netdev_ops->ndo_uninit) |
5225 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 | 5226 | |
9b5e383c ED |
5227 | /* Notifier chain MUST detach us from master device. */ |
5228 | WARN_ON(dev->master); | |
93ee31f1 | 5229 | |
9b5e383c ED |
5230 | /* Remove entries from kobject tree */ |
5231 | netdev_unregister_kobject(dev); | |
5232 | } | |
93ee31f1 | 5233 | |
a5ee1551 | 5234 | /* Process any work delayed until the end of the batch */ |
e5e26d75 | 5235 | dev = list_first_entry(head, struct net_device, unreg_list); |
a5ee1551 | 5236 | call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev); |
93ee31f1 | 5237 | |
850a545b | 5238 | synchronize_net(); |
395264d5 | 5239 | |
a5ee1551 | 5240 | list_for_each_entry(dev, head, unreg_list) |
9b5e383c ED |
5241 | dev_put(dev); |
5242 | } | |
5243 | ||
5244 | static void rollback_registered(struct net_device *dev) | |
5245 | { | |
5246 | LIST_HEAD(single); | |
5247 | ||
5248 | list_add(&dev->unreg_list, &single); | |
5249 | rollback_registered_many(&single); | |
ceaaec98 | 5250 | list_del(&single); |
93ee31f1 DL |
5251 | } |
5252 | ||
fec30c33 | 5253 | static u32 netdev_fix_features(struct net_device *dev, u32 features) |
b63365a2 | 5254 | { |
57422dc5 MM |
5255 | /* Fix illegal checksum combinations */ |
5256 | if ((features & NETIF_F_HW_CSUM) && | |
5257 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 5258 | netdev_warn(dev, "mixed HW and IP checksum settings.\n"); |
57422dc5 MM |
5259 | features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); |
5260 | } | |
5261 | ||
5262 | if ((features & NETIF_F_NO_CSUM) && | |
5263 | (features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 5264 | netdev_warn(dev, "mixed no checksumming and other settings.\n"); |
57422dc5 MM |
5265 | features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM); |
5266 | } | |
5267 | ||
b63365a2 HX |
5268 | /* Fix illegal SG+CSUM combinations. */ |
5269 | if ((features & NETIF_F_SG) && | |
5270 | !(features & NETIF_F_ALL_CSUM)) { | |
6f404e44 MM |
5271 | netdev_dbg(dev, |
5272 | "Dropping NETIF_F_SG since no checksum feature.\n"); | |
b63365a2 HX |
5273 | features &= ~NETIF_F_SG; |
5274 | } | |
5275 | ||
5276 | /* TSO requires that SG is present as well. */ | |
ea2d3688 | 5277 | if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { |
6f404e44 | 5278 | netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); |
ea2d3688 | 5279 | features &= ~NETIF_F_ALL_TSO; |
b63365a2 HX |
5280 | } |
5281 | ||
31d8b9e0 BH |
5282 | /* TSO ECN requires that TSO is present as well. */ |
5283 | if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) | |
5284 | features &= ~NETIF_F_TSO_ECN; | |
5285 | ||
212b573f MM |
5286 | /* Software GSO depends on SG. */ |
5287 | if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { | |
6f404e44 | 5288 | netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); |
212b573f MM |
5289 | features &= ~NETIF_F_GSO; |
5290 | } | |
5291 | ||
acd1130e | 5292 | /* UFO needs SG and checksumming */ |
b63365a2 | 5293 | if (features & NETIF_F_UFO) { |
79032644 MM |
5294 | /* maybe split UFO into V4 and V6? */ |
5295 | if (!((features & NETIF_F_GEN_CSUM) || | |
5296 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM)) | |
5297 | == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 5298 | netdev_dbg(dev, |
acd1130e | 5299 | "Dropping NETIF_F_UFO since no checksum offload features.\n"); |
b63365a2 HX |
5300 | features &= ~NETIF_F_UFO; |
5301 | } | |
5302 | ||
5303 | if (!(features & NETIF_F_SG)) { | |
6f404e44 | 5304 | netdev_dbg(dev, |
acd1130e | 5305 | "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n"); |
b63365a2 HX |
5306 | features &= ~NETIF_F_UFO; |
5307 | } | |
5308 | } | |
5309 | ||
5310 | return features; | |
5311 | } | |
b63365a2 | 5312 | |
6cb6a27c | 5313 | int __netdev_update_features(struct net_device *dev) |
5455c699 MM |
5314 | { |
5315 | u32 features; | |
5316 | int err = 0; | |
5317 | ||
87267485 MM |
5318 | ASSERT_RTNL(); |
5319 | ||
5455c699 MM |
5320 | features = netdev_get_wanted_features(dev); |
5321 | ||
5322 | if (dev->netdev_ops->ndo_fix_features) | |
5323 | features = dev->netdev_ops->ndo_fix_features(dev, features); | |
5324 | ||
5325 | /* driver might be less strict about feature dependencies */ | |
5326 | features = netdev_fix_features(dev, features); | |
5327 | ||
5328 | if (dev->features == features) | |
6cb6a27c | 5329 | return 0; |
5455c699 | 5330 | |
604ae14f | 5331 | netdev_dbg(dev, "Features changed: 0x%08x -> 0x%08x\n", |
5455c699 MM |
5332 | dev->features, features); |
5333 | ||
5334 | if (dev->netdev_ops->ndo_set_features) | |
5335 | err = dev->netdev_ops->ndo_set_features(dev, features); | |
5336 | ||
6cb6a27c | 5337 | if (unlikely(err < 0)) { |
5455c699 MM |
5338 | netdev_err(dev, |
5339 | "set_features() failed (%d); wanted 0x%08x, left 0x%08x\n", | |
5340 | err, features, dev->features); | |
6cb6a27c MM |
5341 | return -1; |
5342 | } | |
5343 | ||
5344 | if (!err) | |
5345 | dev->features = features; | |
5346 | ||
5347 | return 1; | |
5348 | } | |
5349 | ||
afe12cc8 MM |
5350 | /** |
5351 | * netdev_update_features - recalculate device features | |
5352 | * @dev: the device to check | |
5353 | * | |
5354 | * Recalculate dev->features set and send notifications if it | |
5355 | * has changed. Should be called after driver or hardware dependent | |
5356 | * conditions might have changed that influence the features. | |
5357 | */ | |
6cb6a27c MM |
5358 | void netdev_update_features(struct net_device *dev) |
5359 | { | |
5360 | if (__netdev_update_features(dev)) | |
5361 | netdev_features_change(dev); | |
5455c699 MM |
5362 | } |
5363 | EXPORT_SYMBOL(netdev_update_features); | |
5364 | ||
afe12cc8 MM |
5365 | /** |
5366 | * netdev_change_features - recalculate device features | |
5367 | * @dev: the device to check | |
5368 | * | |
5369 | * Recalculate dev->features set and send notifications even | |
5370 | * if they have not changed. Should be called instead of | |
5371 | * netdev_update_features() if also dev->vlan_features might | |
5372 | * have changed to allow the changes to be propagated to stacked | |
5373 | * VLAN devices. | |
5374 | */ | |
5375 | void netdev_change_features(struct net_device *dev) | |
5376 | { | |
5377 | __netdev_update_features(dev); | |
5378 | netdev_features_change(dev); | |
5379 | } | |
5380 | EXPORT_SYMBOL(netdev_change_features); | |
5381 | ||
fc4a7489 PM |
5382 | /** |
5383 | * netif_stacked_transfer_operstate - transfer operstate | |
5384 | * @rootdev: the root or lower level device to transfer state from | |
5385 | * @dev: the device to transfer operstate to | |
5386 | * | |
5387 | * Transfer operational state from root to device. This is normally | |
5388 | * called when a stacking relationship exists between the root | |
5389 | * device and the device(a leaf device). | |
5390 | */ | |
5391 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
5392 | struct net_device *dev) | |
5393 | { | |
5394 | if (rootdev->operstate == IF_OPER_DORMANT) | |
5395 | netif_dormant_on(dev); | |
5396 | else | |
5397 | netif_dormant_off(dev); | |
5398 | ||
5399 | if (netif_carrier_ok(rootdev)) { | |
5400 | if (!netif_carrier_ok(dev)) | |
5401 | netif_carrier_on(dev); | |
5402 | } else { | |
5403 | if (netif_carrier_ok(dev)) | |
5404 | netif_carrier_off(dev); | |
5405 | } | |
5406 | } | |
5407 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
5408 | ||
bf264145 | 5409 | #ifdef CONFIG_RPS |
1b4bf461 ED |
5410 | static int netif_alloc_rx_queues(struct net_device *dev) |
5411 | { | |
1b4bf461 | 5412 | unsigned int i, count = dev->num_rx_queues; |
bd25fa7b | 5413 | struct netdev_rx_queue *rx; |
1b4bf461 | 5414 | |
bd25fa7b | 5415 | BUG_ON(count < 1); |
1b4bf461 | 5416 | |
bd25fa7b TH |
5417 | rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL); |
5418 | if (!rx) { | |
5419 | pr_err("netdev: Unable to allocate %u rx queues.\n", count); | |
5420 | return -ENOMEM; | |
1b4bf461 | 5421 | } |
bd25fa7b TH |
5422 | dev->_rx = rx; |
5423 | ||
bd25fa7b | 5424 | for (i = 0; i < count; i++) |
fe822240 | 5425 | rx[i].dev = dev; |
1b4bf461 ED |
5426 | return 0; |
5427 | } | |
bf264145 | 5428 | #endif |
1b4bf461 | 5429 | |
aa942104 CG |
5430 | static void netdev_init_one_queue(struct net_device *dev, |
5431 | struct netdev_queue *queue, void *_unused) | |
5432 | { | |
5433 | /* Initialize queue lock */ | |
5434 | spin_lock_init(&queue->_xmit_lock); | |
5435 | netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); | |
5436 | queue->xmit_lock_owner = -1; | |
b236da69 | 5437 | netdev_queue_numa_node_write(queue, NUMA_NO_NODE); |
aa942104 CG |
5438 | queue->dev = dev; |
5439 | } | |
5440 | ||
e6484930 TH |
5441 | static int netif_alloc_netdev_queues(struct net_device *dev) |
5442 | { | |
5443 | unsigned int count = dev->num_tx_queues; | |
5444 | struct netdev_queue *tx; | |
5445 | ||
5446 | BUG_ON(count < 1); | |
5447 | ||
5448 | tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL); | |
5449 | if (!tx) { | |
5450 | pr_err("netdev: Unable to allocate %u tx queues.\n", | |
5451 | count); | |
5452 | return -ENOMEM; | |
5453 | } | |
5454 | dev->_tx = tx; | |
1d24eb48 | 5455 | |
e6484930 TH |
5456 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); |
5457 | spin_lock_init(&dev->tx_global_lock); | |
aa942104 CG |
5458 | |
5459 | return 0; | |
e6484930 TH |
5460 | } |
5461 | ||
1da177e4 LT |
5462 | /** |
5463 | * register_netdevice - register a network device | |
5464 | * @dev: device to register | |
5465 | * | |
5466 | * Take a completed network device structure and add it to the kernel | |
5467 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
5468 | * chain. 0 is returned on success. A negative errno code is returned | |
5469 | * on a failure to set up the device, or if the name is a duplicate. | |
5470 | * | |
5471 | * Callers must hold the rtnl semaphore. You may want | |
5472 | * register_netdev() instead of this. | |
5473 | * | |
5474 | * BUGS: | |
5475 | * The locking appears insufficient to guarantee two parallel registers | |
5476 | * will not get the same name. | |
5477 | */ | |
5478 | ||
5479 | int register_netdevice(struct net_device *dev) | |
5480 | { | |
1da177e4 | 5481 | int ret; |
d314774c | 5482 | struct net *net = dev_net(dev); |
1da177e4 LT |
5483 | |
5484 | BUG_ON(dev_boot_phase); | |
5485 | ASSERT_RTNL(); | |
5486 | ||
b17a7c17 SH |
5487 | might_sleep(); |
5488 | ||
1da177e4 LT |
5489 | /* When net_device's are persistent, this will be fatal. */ |
5490 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 5491 | BUG_ON(!net); |
1da177e4 | 5492 | |
f1f28aa3 | 5493 | spin_lock_init(&dev->addr_list_lock); |
cf508b12 | 5494 | netdev_set_addr_lockdep_class(dev); |
1da177e4 | 5495 | |
1da177e4 LT |
5496 | dev->iflink = -1; |
5497 | ||
0696c3a8 PP |
5498 | ret = dev_get_valid_name(dev, dev->name); |
5499 | if (ret < 0) | |
5500 | goto out; | |
5501 | ||
1da177e4 | 5502 | /* Init, if this function is available */ |
d314774c SH |
5503 | if (dev->netdev_ops->ndo_init) { |
5504 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
5505 | if (ret) { |
5506 | if (ret > 0) | |
5507 | ret = -EIO; | |
90833aa4 | 5508 | goto out; |
1da177e4 LT |
5509 | } |
5510 | } | |
4ec93edb | 5511 | |
881d966b | 5512 | dev->ifindex = dev_new_index(net); |
1da177e4 LT |
5513 | if (dev->iflink == -1) |
5514 | dev->iflink = dev->ifindex; | |
5515 | ||
5455c699 MM |
5516 | /* Transfer changeable features to wanted_features and enable |
5517 | * software offloads (GSO and GRO). | |
5518 | */ | |
5519 | dev->hw_features |= NETIF_F_SOFT_FEATURES; | |
14d1232f MM |
5520 | dev->features |= NETIF_F_SOFT_FEATURES; |
5521 | dev->wanted_features = dev->features & dev->hw_features; | |
1da177e4 | 5522 | |
c6e1a0d1 TH |
5523 | /* Turn on no cache copy if HW is doing checksum */ |
5524 | dev->hw_features |= NETIF_F_NOCACHE_COPY; | |
5525 | if ((dev->features & NETIF_F_ALL_CSUM) && | |
5526 | !(dev->features & NETIF_F_NO_CSUM)) { | |
5527 | dev->wanted_features |= NETIF_F_NOCACHE_COPY; | |
5528 | dev->features |= NETIF_F_NOCACHE_COPY; | |
5529 | } | |
5530 | ||
1180e7d6 | 5531 | /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. |
16c3ea78 | 5532 | */ |
1180e7d6 | 5533 | dev->vlan_features |= NETIF_F_HIGHDMA; |
16c3ea78 | 5534 | |
7ffbe3fd JB |
5535 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); |
5536 | ret = notifier_to_errno(ret); | |
5537 | if (ret) | |
5538 | goto err_uninit; | |
5539 | ||
8b41d188 | 5540 | ret = netdev_register_kobject(dev); |
b17a7c17 | 5541 | if (ret) |
7ce1b0ed | 5542 | goto err_uninit; |
b17a7c17 SH |
5543 | dev->reg_state = NETREG_REGISTERED; |
5544 | ||
6cb6a27c | 5545 | __netdev_update_features(dev); |
8e9b59b2 | 5546 | |
1da177e4 LT |
5547 | /* |
5548 | * Default initial state at registry is that the | |
5549 | * device is present. | |
5550 | */ | |
5551 | ||
5552 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
5553 | ||
1da177e4 | 5554 | dev_init_scheduler(dev); |
1da177e4 | 5555 | dev_hold(dev); |
ce286d32 | 5556 | list_netdevice(dev); |
1da177e4 LT |
5557 | |
5558 | /* Notify protocols, that a new device appeared. */ | |
056925ab | 5559 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 5560 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
5561 | if (ret) { |
5562 | rollback_registered(dev); | |
5563 | dev->reg_state = NETREG_UNREGISTERED; | |
5564 | } | |
d90a909e EB |
5565 | /* |
5566 | * Prevent userspace races by waiting until the network | |
5567 | * device is fully setup before sending notifications. | |
5568 | */ | |
a2835763 PM |
5569 | if (!dev->rtnl_link_ops || |
5570 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
5571 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U); | |
1da177e4 LT |
5572 | |
5573 | out: | |
5574 | return ret; | |
7ce1b0ed HX |
5575 | |
5576 | err_uninit: | |
d314774c SH |
5577 | if (dev->netdev_ops->ndo_uninit) |
5578 | dev->netdev_ops->ndo_uninit(dev); | |
7ce1b0ed | 5579 | goto out; |
1da177e4 | 5580 | } |
d1b19dff | 5581 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 5582 | |
937f1ba5 BH |
5583 | /** |
5584 | * init_dummy_netdev - init a dummy network device for NAPI | |
5585 | * @dev: device to init | |
5586 | * | |
5587 | * This takes a network device structure and initialize the minimum | |
5588 | * amount of fields so it can be used to schedule NAPI polls without | |
5589 | * registering a full blown interface. This is to be used by drivers | |
5590 | * that need to tie several hardware interfaces to a single NAPI | |
5591 | * poll scheduler due to HW limitations. | |
5592 | */ | |
5593 | int init_dummy_netdev(struct net_device *dev) | |
5594 | { | |
5595 | /* Clear everything. Note we don't initialize spinlocks | |
5596 | * are they aren't supposed to be taken by any of the | |
5597 | * NAPI code and this dummy netdev is supposed to be | |
5598 | * only ever used for NAPI polls | |
5599 | */ | |
5600 | memset(dev, 0, sizeof(struct net_device)); | |
5601 | ||
5602 | /* make sure we BUG if trying to hit standard | |
5603 | * register/unregister code path | |
5604 | */ | |
5605 | dev->reg_state = NETREG_DUMMY; | |
5606 | ||
937f1ba5 BH |
5607 | /* NAPI wants this */ |
5608 | INIT_LIST_HEAD(&dev->napi_list); | |
5609 | ||
5610 | /* a dummy interface is started by default */ | |
5611 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
5612 | set_bit(__LINK_STATE_START, &dev->state); | |
5613 | ||
29b4433d ED |
5614 | /* Note : We dont allocate pcpu_refcnt for dummy devices, |
5615 | * because users of this 'device' dont need to change | |
5616 | * its refcount. | |
5617 | */ | |
5618 | ||
937f1ba5 BH |
5619 | return 0; |
5620 | } | |
5621 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
5622 | ||
5623 | ||
1da177e4 LT |
5624 | /** |
5625 | * register_netdev - register a network device | |
5626 | * @dev: device to register | |
5627 | * | |
5628 | * Take a completed network device structure and add it to the kernel | |
5629 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
5630 | * chain. 0 is returned on success. A negative errno code is returned | |
5631 | * on a failure to set up the device, or if the name is a duplicate. | |
5632 | * | |
38b4da38 | 5633 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
5634 | * and expands the device name if you passed a format string to |
5635 | * alloc_netdev. | |
5636 | */ | |
5637 | int register_netdev(struct net_device *dev) | |
5638 | { | |
5639 | int err; | |
5640 | ||
5641 | rtnl_lock(); | |
1da177e4 | 5642 | err = register_netdevice(dev); |
1da177e4 LT |
5643 | rtnl_unlock(); |
5644 | return err; | |
5645 | } | |
5646 | EXPORT_SYMBOL(register_netdev); | |
5647 | ||
29b4433d ED |
5648 | int netdev_refcnt_read(const struct net_device *dev) |
5649 | { | |
5650 | int i, refcnt = 0; | |
5651 | ||
5652 | for_each_possible_cpu(i) | |
5653 | refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); | |
5654 | return refcnt; | |
5655 | } | |
5656 | EXPORT_SYMBOL(netdev_refcnt_read); | |
5657 | ||
1da177e4 LT |
5658 | /* |
5659 | * netdev_wait_allrefs - wait until all references are gone. | |
5660 | * | |
5661 | * This is called when unregistering network devices. | |
5662 | * | |
5663 | * Any protocol or device that holds a reference should register | |
5664 | * for netdevice notification, and cleanup and put back the | |
5665 | * reference if they receive an UNREGISTER event. | |
5666 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 5667 | * call dev_put. |
1da177e4 LT |
5668 | */ |
5669 | static void netdev_wait_allrefs(struct net_device *dev) | |
5670 | { | |
5671 | unsigned long rebroadcast_time, warning_time; | |
29b4433d | 5672 | int refcnt; |
1da177e4 | 5673 | |
e014debe ED |
5674 | linkwatch_forget_dev(dev); |
5675 | ||
1da177e4 | 5676 | rebroadcast_time = warning_time = jiffies; |
29b4433d ED |
5677 | refcnt = netdev_refcnt_read(dev); |
5678 | ||
5679 | while (refcnt != 0) { | |
1da177e4 | 5680 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { |
6756ae4b | 5681 | rtnl_lock(); |
1da177e4 LT |
5682 | |
5683 | /* Rebroadcast unregister notification */ | |
056925ab | 5684 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
a5ee1551 | 5685 | /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users |
395264d5 | 5686 | * should have already handle it the first time */ |
1da177e4 LT |
5687 | |
5688 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, | |
5689 | &dev->state)) { | |
5690 | /* We must not have linkwatch events | |
5691 | * pending on unregister. If this | |
5692 | * happens, we simply run the queue | |
5693 | * unscheduled, resulting in a noop | |
5694 | * for this device. | |
5695 | */ | |
5696 | linkwatch_run_queue(); | |
5697 | } | |
5698 | ||
6756ae4b | 5699 | __rtnl_unlock(); |
1da177e4 LT |
5700 | |
5701 | rebroadcast_time = jiffies; | |
5702 | } | |
5703 | ||
5704 | msleep(250); | |
5705 | ||
29b4433d ED |
5706 | refcnt = netdev_refcnt_read(dev); |
5707 | ||
1da177e4 LT |
5708 | if (time_after(jiffies, warning_time + 10 * HZ)) { |
5709 | printk(KERN_EMERG "unregister_netdevice: " | |
5710 | "waiting for %s to become free. Usage " | |
5711 | "count = %d\n", | |
29b4433d | 5712 | dev->name, refcnt); |
1da177e4 LT |
5713 | warning_time = jiffies; |
5714 | } | |
5715 | } | |
5716 | } | |
5717 | ||
5718 | /* The sequence is: | |
5719 | * | |
5720 | * rtnl_lock(); | |
5721 | * ... | |
5722 | * register_netdevice(x1); | |
5723 | * register_netdevice(x2); | |
5724 | * ... | |
5725 | * unregister_netdevice(y1); | |
5726 | * unregister_netdevice(y2); | |
5727 | * ... | |
5728 | * rtnl_unlock(); | |
5729 | * free_netdev(y1); | |
5730 | * free_netdev(y2); | |
5731 | * | |
58ec3b4d | 5732 | * We are invoked by rtnl_unlock(). |
1da177e4 | 5733 | * This allows us to deal with problems: |
b17a7c17 | 5734 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
5735 | * without deadlocking with linkwatch via keventd. |
5736 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
5737 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
5738 | * |
5739 | * We must not return until all unregister events added during | |
5740 | * the interval the lock was held have been completed. | |
1da177e4 | 5741 | */ |
1da177e4 LT |
5742 | void netdev_run_todo(void) |
5743 | { | |
626ab0e6 | 5744 | struct list_head list; |
1da177e4 | 5745 | |
1da177e4 | 5746 | /* Snapshot list, allow later requests */ |
626ab0e6 | 5747 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
5748 | |
5749 | __rtnl_unlock(); | |
626ab0e6 | 5750 | |
850a545b EB |
5751 | /* Wait for rcu callbacks to finish before attempting to drain |
5752 | * the device list. This usually avoids a 250ms wait. | |
5753 | */ | |
5754 | if (!list_empty(&list)) | |
5755 | rcu_barrier(); | |
5756 | ||
1da177e4 LT |
5757 | while (!list_empty(&list)) { |
5758 | struct net_device *dev | |
e5e26d75 | 5759 | = list_first_entry(&list, struct net_device, todo_list); |
1da177e4 LT |
5760 | list_del(&dev->todo_list); |
5761 | ||
b17a7c17 SH |
5762 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
5763 | printk(KERN_ERR "network todo '%s' but state %d\n", | |
5764 | dev->name, dev->reg_state); | |
5765 | dump_stack(); | |
5766 | continue; | |
5767 | } | |
1da177e4 | 5768 | |
b17a7c17 | 5769 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 5770 | |
152102c7 | 5771 | on_each_cpu(flush_backlog, dev, 1); |
6e583ce5 | 5772 | |
b17a7c17 | 5773 | netdev_wait_allrefs(dev); |
1da177e4 | 5774 | |
b17a7c17 | 5775 | /* paranoia */ |
29b4433d | 5776 | BUG_ON(netdev_refcnt_read(dev)); |
33d480ce ED |
5777 | WARN_ON(rcu_access_pointer(dev->ip_ptr)); |
5778 | WARN_ON(rcu_access_pointer(dev->ip6_ptr)); | |
547b792c | 5779 | WARN_ON(dev->dn_ptr); |
1da177e4 | 5780 | |
b17a7c17 SH |
5781 | if (dev->destructor) |
5782 | dev->destructor(dev); | |
9093bbb2 SH |
5783 | |
5784 | /* Free network device */ | |
5785 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 5786 | } |
1da177e4 LT |
5787 | } |
5788 | ||
3cfde79c BH |
5789 | /* Convert net_device_stats to rtnl_link_stats64. They have the same |
5790 | * fields in the same order, with only the type differing. | |
5791 | */ | |
5792 | static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, | |
5793 | const struct net_device_stats *netdev_stats) | |
5794 | { | |
5795 | #if BITS_PER_LONG == 64 | |
5796 | BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats)); | |
5797 | memcpy(stats64, netdev_stats, sizeof(*stats64)); | |
5798 | #else | |
5799 | size_t i, n = sizeof(*stats64) / sizeof(u64); | |
5800 | const unsigned long *src = (const unsigned long *)netdev_stats; | |
5801 | u64 *dst = (u64 *)stats64; | |
5802 | ||
5803 | BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) != | |
5804 | sizeof(*stats64) / sizeof(u64)); | |
5805 | for (i = 0; i < n; i++) | |
5806 | dst[i] = src[i]; | |
5807 | #endif | |
5808 | } | |
5809 | ||
eeda3fd6 SH |
5810 | /** |
5811 | * dev_get_stats - get network device statistics | |
5812 | * @dev: device to get statistics from | |
28172739 | 5813 | * @storage: place to store stats |
eeda3fd6 | 5814 | * |
d7753516 BH |
5815 | * Get network statistics from device. Return @storage. |
5816 | * The device driver may provide its own method by setting | |
5817 | * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; | |
5818 | * otherwise the internal statistics structure is used. | |
eeda3fd6 | 5819 | */ |
d7753516 BH |
5820 | struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, |
5821 | struct rtnl_link_stats64 *storage) | |
7004bf25 | 5822 | { |
eeda3fd6 SH |
5823 | const struct net_device_ops *ops = dev->netdev_ops; |
5824 | ||
28172739 ED |
5825 | if (ops->ndo_get_stats64) { |
5826 | memset(storage, 0, sizeof(*storage)); | |
caf586e5 ED |
5827 | ops->ndo_get_stats64(dev, storage); |
5828 | } else if (ops->ndo_get_stats) { | |
3cfde79c | 5829 | netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); |
caf586e5 ED |
5830 | } else { |
5831 | netdev_stats_to_stats64(storage, &dev->stats); | |
28172739 | 5832 | } |
caf586e5 | 5833 | storage->rx_dropped += atomic_long_read(&dev->rx_dropped); |
28172739 | 5834 | return storage; |
c45d286e | 5835 | } |
eeda3fd6 | 5836 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 5837 | |
24824a09 | 5838 | struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) |
dc2b4847 | 5839 | { |
24824a09 | 5840 | struct netdev_queue *queue = dev_ingress_queue(dev); |
dc2b4847 | 5841 | |
24824a09 ED |
5842 | #ifdef CONFIG_NET_CLS_ACT |
5843 | if (queue) | |
5844 | return queue; | |
5845 | queue = kzalloc(sizeof(*queue), GFP_KERNEL); | |
5846 | if (!queue) | |
5847 | return NULL; | |
5848 | netdev_init_one_queue(dev, queue, NULL); | |
24824a09 ED |
5849 | queue->qdisc = &noop_qdisc; |
5850 | queue->qdisc_sleeping = &noop_qdisc; | |
5851 | rcu_assign_pointer(dev->ingress_queue, queue); | |
5852 | #endif | |
5853 | return queue; | |
bb949fbd DM |
5854 | } |
5855 | ||
1da177e4 | 5856 | /** |
36909ea4 | 5857 | * alloc_netdev_mqs - allocate network device |
1da177e4 LT |
5858 | * @sizeof_priv: size of private data to allocate space for |
5859 | * @name: device name format string | |
5860 | * @setup: callback to initialize device | |
36909ea4 TH |
5861 | * @txqs: the number of TX subqueues to allocate |
5862 | * @rxqs: the number of RX subqueues to allocate | |
1da177e4 LT |
5863 | * |
5864 | * Allocates a struct net_device with private data area for driver use | |
f25f4e44 | 5865 | * and performs basic initialization. Also allocates subquue structs |
36909ea4 | 5866 | * for each queue on the device. |
1da177e4 | 5867 | */ |
36909ea4 TH |
5868 | struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, |
5869 | void (*setup)(struct net_device *), | |
5870 | unsigned int txqs, unsigned int rxqs) | |
1da177e4 | 5871 | { |
1da177e4 | 5872 | struct net_device *dev; |
7943986c | 5873 | size_t alloc_size; |
1ce8e7b5 | 5874 | struct net_device *p; |
1da177e4 | 5875 | |
b6fe17d6 SH |
5876 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
5877 | ||
36909ea4 | 5878 | if (txqs < 1) { |
55513fb4 TH |
5879 | pr_err("alloc_netdev: Unable to allocate device " |
5880 | "with zero queues.\n"); | |
5881 | return NULL; | |
5882 | } | |
5883 | ||
36909ea4 TH |
5884 | #ifdef CONFIG_RPS |
5885 | if (rxqs < 1) { | |
5886 | pr_err("alloc_netdev: Unable to allocate device " | |
5887 | "with zero RX queues.\n"); | |
5888 | return NULL; | |
5889 | } | |
5890 | #endif | |
5891 | ||
fd2ea0a7 | 5892 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
5893 | if (sizeof_priv) { |
5894 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 5895 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
5896 | alloc_size += sizeof_priv; |
5897 | } | |
5898 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 5899 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 5900 | |
31380de9 | 5901 | p = kzalloc(alloc_size, GFP_KERNEL); |
1da177e4 | 5902 | if (!p) { |
b6fe17d6 | 5903 | printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n"); |
1da177e4 LT |
5904 | return NULL; |
5905 | } | |
1da177e4 | 5906 | |
1ce8e7b5 | 5907 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 5908 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc | 5909 | |
29b4433d ED |
5910 | dev->pcpu_refcnt = alloc_percpu(int); |
5911 | if (!dev->pcpu_refcnt) | |
e6484930 | 5912 | goto free_p; |
ab9c73cc | 5913 | |
ab9c73cc | 5914 | if (dev_addr_init(dev)) |
29b4433d | 5915 | goto free_pcpu; |
ab9c73cc | 5916 | |
22bedad3 | 5917 | dev_mc_init(dev); |
a748ee24 | 5918 | dev_uc_init(dev); |
ccffad25 | 5919 | |
c346dca1 | 5920 | dev_net_set(dev, &init_net); |
1da177e4 | 5921 | |
8d3bdbd5 DM |
5922 | dev->gso_max_size = GSO_MAX_SIZE; |
5923 | ||
8d3bdbd5 DM |
5924 | INIT_LIST_HEAD(&dev->napi_list); |
5925 | INIT_LIST_HEAD(&dev->unreg_list); | |
5926 | INIT_LIST_HEAD(&dev->link_watch_list); | |
5927 | dev->priv_flags = IFF_XMIT_DST_RELEASE; | |
5928 | setup(dev); | |
5929 | ||
36909ea4 TH |
5930 | dev->num_tx_queues = txqs; |
5931 | dev->real_num_tx_queues = txqs; | |
ed9af2e8 | 5932 | if (netif_alloc_netdev_queues(dev)) |
8d3bdbd5 | 5933 | goto free_all; |
e8a0464c | 5934 | |
df334545 | 5935 | #ifdef CONFIG_RPS |
36909ea4 TH |
5936 | dev->num_rx_queues = rxqs; |
5937 | dev->real_num_rx_queues = rxqs; | |
fe822240 | 5938 | if (netif_alloc_rx_queues(dev)) |
8d3bdbd5 | 5939 | goto free_all; |
df334545 | 5940 | #endif |
0a9627f2 | 5941 | |
1da177e4 | 5942 | strcpy(dev->name, name); |
cbda10fa | 5943 | dev->group = INIT_NETDEV_GROUP; |
1da177e4 | 5944 | return dev; |
ab9c73cc | 5945 | |
8d3bdbd5 DM |
5946 | free_all: |
5947 | free_netdev(dev); | |
5948 | return NULL; | |
5949 | ||
29b4433d ED |
5950 | free_pcpu: |
5951 | free_percpu(dev->pcpu_refcnt); | |
ed9af2e8 | 5952 | kfree(dev->_tx); |
fe822240 TH |
5953 | #ifdef CONFIG_RPS |
5954 | kfree(dev->_rx); | |
5955 | #endif | |
5956 | ||
ab9c73cc JP |
5957 | free_p: |
5958 | kfree(p); | |
5959 | return NULL; | |
1da177e4 | 5960 | } |
36909ea4 | 5961 | EXPORT_SYMBOL(alloc_netdev_mqs); |
1da177e4 LT |
5962 | |
5963 | /** | |
5964 | * free_netdev - free network device | |
5965 | * @dev: device | |
5966 | * | |
4ec93edb YH |
5967 | * This function does the last stage of destroying an allocated device |
5968 | * interface. The reference to the device object is released. | |
1da177e4 LT |
5969 | * If this is the last reference then it will be freed. |
5970 | */ | |
5971 | void free_netdev(struct net_device *dev) | |
5972 | { | |
d565b0a1 HX |
5973 | struct napi_struct *p, *n; |
5974 | ||
f3005d7f DL |
5975 | release_net(dev_net(dev)); |
5976 | ||
e8a0464c | 5977 | kfree(dev->_tx); |
fe822240 TH |
5978 | #ifdef CONFIG_RPS |
5979 | kfree(dev->_rx); | |
5980 | #endif | |
e8a0464c | 5981 | |
33d480ce | 5982 | kfree(rcu_dereference_protected(dev->ingress_queue, 1)); |
24824a09 | 5983 | |
f001fde5 JP |
5984 | /* Flush device addresses */ |
5985 | dev_addr_flush(dev); | |
5986 | ||
d565b0a1 HX |
5987 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
5988 | netif_napi_del(p); | |
5989 | ||
29b4433d ED |
5990 | free_percpu(dev->pcpu_refcnt); |
5991 | dev->pcpu_refcnt = NULL; | |
5992 | ||
3041a069 | 5993 | /* Compatibility with error handling in drivers */ |
1da177e4 LT |
5994 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
5995 | kfree((char *)dev - dev->padded); | |
5996 | return; | |
5997 | } | |
5998 | ||
5999 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
6000 | dev->reg_state = NETREG_RELEASED; | |
6001 | ||
43cb76d9 GKH |
6002 | /* will free via device release */ |
6003 | put_device(&dev->dev); | |
1da177e4 | 6004 | } |
d1b19dff | 6005 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 6006 | |
f0db275a SH |
6007 | /** |
6008 | * synchronize_net - Synchronize with packet receive processing | |
6009 | * | |
6010 | * Wait for packets currently being received to be done. | |
6011 | * Does not block later packets from starting. | |
6012 | */ | |
4ec93edb | 6013 | void synchronize_net(void) |
1da177e4 LT |
6014 | { |
6015 | might_sleep(); | |
be3fc413 ED |
6016 | if (rtnl_is_locked()) |
6017 | synchronize_rcu_expedited(); | |
6018 | else | |
6019 | synchronize_rcu(); | |
1da177e4 | 6020 | } |
d1b19dff | 6021 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
6022 | |
6023 | /** | |
44a0873d | 6024 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 6025 | * @dev: device |
44a0873d | 6026 | * @head: list |
6ebfbc06 | 6027 | * |
1da177e4 | 6028 | * This function shuts down a device interface and removes it |
d59b54b1 | 6029 | * from the kernel tables. |
44a0873d | 6030 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
6031 | * |
6032 | * Callers must hold the rtnl semaphore. You may want | |
6033 | * unregister_netdev() instead of this. | |
6034 | */ | |
6035 | ||
44a0873d | 6036 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 6037 | { |
a6620712 HX |
6038 | ASSERT_RTNL(); |
6039 | ||
44a0873d | 6040 | if (head) { |
9fdce099 | 6041 | list_move_tail(&dev->unreg_list, head); |
44a0873d ED |
6042 | } else { |
6043 | rollback_registered(dev); | |
6044 | /* Finish processing unregister after unlock */ | |
6045 | net_set_todo(dev); | |
6046 | } | |
1da177e4 | 6047 | } |
44a0873d | 6048 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 6049 | |
9b5e383c ED |
6050 | /** |
6051 | * unregister_netdevice_many - unregister many devices | |
6052 | * @head: list of devices | |
9b5e383c ED |
6053 | */ |
6054 | void unregister_netdevice_many(struct list_head *head) | |
6055 | { | |
6056 | struct net_device *dev; | |
6057 | ||
6058 | if (!list_empty(head)) { | |
6059 | rollback_registered_many(head); | |
6060 | list_for_each_entry(dev, head, unreg_list) | |
6061 | net_set_todo(dev); | |
6062 | } | |
6063 | } | |
63c8099d | 6064 | EXPORT_SYMBOL(unregister_netdevice_many); |
9b5e383c | 6065 | |
1da177e4 LT |
6066 | /** |
6067 | * unregister_netdev - remove device from the kernel | |
6068 | * @dev: device | |
6069 | * | |
6070 | * This function shuts down a device interface and removes it | |
d59b54b1 | 6071 | * from the kernel tables. |
1da177e4 LT |
6072 | * |
6073 | * This is just a wrapper for unregister_netdevice that takes | |
6074 | * the rtnl semaphore. In general you want to use this and not | |
6075 | * unregister_netdevice. | |
6076 | */ | |
6077 | void unregister_netdev(struct net_device *dev) | |
6078 | { | |
6079 | rtnl_lock(); | |
6080 | unregister_netdevice(dev); | |
6081 | rtnl_unlock(); | |
6082 | } | |
1da177e4 LT |
6083 | EXPORT_SYMBOL(unregister_netdev); |
6084 | ||
ce286d32 EB |
6085 | /** |
6086 | * dev_change_net_namespace - move device to different nethost namespace | |
6087 | * @dev: device | |
6088 | * @net: network namespace | |
6089 | * @pat: If not NULL name pattern to try if the current device name | |
6090 | * is already taken in the destination network namespace. | |
6091 | * | |
6092 | * This function shuts down a device interface and moves it | |
6093 | * to a new network namespace. On success 0 is returned, on | |
6094 | * a failure a netagive errno code is returned. | |
6095 | * | |
6096 | * Callers must hold the rtnl semaphore. | |
6097 | */ | |
6098 | ||
6099 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
6100 | { | |
ce286d32 EB |
6101 | int err; |
6102 | ||
6103 | ASSERT_RTNL(); | |
6104 | ||
6105 | /* Don't allow namespace local devices to be moved. */ | |
6106 | err = -EINVAL; | |
6107 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
6108 | goto out; | |
6109 | ||
6110 | /* Ensure the device has been registrered */ | |
6111 | err = -EINVAL; | |
6112 | if (dev->reg_state != NETREG_REGISTERED) | |
6113 | goto out; | |
6114 | ||
6115 | /* Get out if there is nothing todo */ | |
6116 | err = 0; | |
878628fb | 6117 | if (net_eq(dev_net(dev), net)) |
ce286d32 EB |
6118 | goto out; |
6119 | ||
6120 | /* Pick the destination device name, and ensure | |
6121 | * we can use it in the destination network namespace. | |
6122 | */ | |
6123 | err = -EEXIST; | |
d9031024 | 6124 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
6125 | /* We get here if we can't use the current device name */ |
6126 | if (!pat) | |
6127 | goto out; | |
1c5cae81 | 6128 | if (dev_get_valid_name(dev, pat) < 0) |
ce286d32 EB |
6129 | goto out; |
6130 | } | |
6131 | ||
6132 | /* | |
6133 | * And now a mini version of register_netdevice unregister_netdevice. | |
6134 | */ | |
6135 | ||
6136 | /* If device is running close it first. */ | |
9b772652 | 6137 | dev_close(dev); |
ce286d32 EB |
6138 | |
6139 | /* And unlink it from device chain */ | |
6140 | err = -ENODEV; | |
6141 | unlist_netdevice(dev); | |
6142 | ||
6143 | synchronize_net(); | |
6144 | ||
6145 | /* Shutdown queueing discipline. */ | |
6146 | dev_shutdown(dev); | |
6147 | ||
6148 | /* Notify protocols, that we are about to destroy | |
6149 | this device. They should clean all the things. | |
3b27e105 DL |
6150 | |
6151 | Note that dev->reg_state stays at NETREG_REGISTERED. | |
6152 | This is wanted because this way 8021q and macvlan know | |
6153 | the device is just moving and can keep their slaves up. | |
ce286d32 EB |
6154 | */ |
6155 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
a5ee1551 | 6156 | call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev); |
ce286d32 EB |
6157 | |
6158 | /* | |
6159 | * Flush the unicast and multicast chains | |
6160 | */ | |
a748ee24 | 6161 | dev_uc_flush(dev); |
22bedad3 | 6162 | dev_mc_flush(dev); |
ce286d32 EB |
6163 | |
6164 | /* Actually switch the network namespace */ | |
c346dca1 | 6165 | dev_net_set(dev, net); |
ce286d32 | 6166 | |
ce286d32 EB |
6167 | /* If there is an ifindex conflict assign a new one */ |
6168 | if (__dev_get_by_index(net, dev->ifindex)) { | |
6169 | int iflink = (dev->iflink == dev->ifindex); | |
6170 | dev->ifindex = dev_new_index(net); | |
6171 | if (iflink) | |
6172 | dev->iflink = dev->ifindex; | |
6173 | } | |
6174 | ||
8b41d188 | 6175 | /* Fixup kobjects */ |
a1b3f594 | 6176 | err = device_rename(&dev->dev, dev->name); |
8b41d188 | 6177 | WARN_ON(err); |
ce286d32 EB |
6178 | |
6179 | /* Add the device back in the hashes */ | |
6180 | list_netdevice(dev); | |
6181 | ||
6182 | /* Notify protocols, that a new device appeared. */ | |
6183 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
6184 | ||
d90a909e EB |
6185 | /* |
6186 | * Prevent userspace races by waiting until the network | |
6187 | * device is fully setup before sending notifications. | |
6188 | */ | |
6189 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U); | |
6190 | ||
ce286d32 EB |
6191 | synchronize_net(); |
6192 | err = 0; | |
6193 | out: | |
6194 | return err; | |
6195 | } | |
463d0183 | 6196 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 6197 | |
1da177e4 LT |
6198 | static int dev_cpu_callback(struct notifier_block *nfb, |
6199 | unsigned long action, | |
6200 | void *ocpu) | |
6201 | { | |
6202 | struct sk_buff **list_skb; | |
1da177e4 LT |
6203 | struct sk_buff *skb; |
6204 | unsigned int cpu, oldcpu = (unsigned long)ocpu; | |
6205 | struct softnet_data *sd, *oldsd; | |
6206 | ||
8bb78442 | 6207 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
1da177e4 LT |
6208 | return NOTIFY_OK; |
6209 | ||
6210 | local_irq_disable(); | |
6211 | cpu = smp_processor_id(); | |
6212 | sd = &per_cpu(softnet_data, cpu); | |
6213 | oldsd = &per_cpu(softnet_data, oldcpu); | |
6214 | ||
6215 | /* Find end of our completion_queue. */ | |
6216 | list_skb = &sd->completion_queue; | |
6217 | while (*list_skb) | |
6218 | list_skb = &(*list_skb)->next; | |
6219 | /* Append completion queue from offline CPU. */ | |
6220 | *list_skb = oldsd->completion_queue; | |
6221 | oldsd->completion_queue = NULL; | |
6222 | ||
1da177e4 | 6223 | /* Append output queue from offline CPU. */ |
a9cbd588 CG |
6224 | if (oldsd->output_queue) { |
6225 | *sd->output_queue_tailp = oldsd->output_queue; | |
6226 | sd->output_queue_tailp = oldsd->output_queue_tailp; | |
6227 | oldsd->output_queue = NULL; | |
6228 | oldsd->output_queue_tailp = &oldsd->output_queue; | |
6229 | } | |
264524d5 HC |
6230 | /* Append NAPI poll list from offline CPU. */ |
6231 | if (!list_empty(&oldsd->poll_list)) { | |
6232 | list_splice_init(&oldsd->poll_list, &sd->poll_list); | |
6233 | raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
6234 | } | |
1da177e4 LT |
6235 | |
6236 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
6237 | local_irq_enable(); | |
6238 | ||
6239 | /* Process offline CPU's input_pkt_queue */ | |
76cc8b13 | 6240 | while ((skb = __skb_dequeue(&oldsd->process_queue))) { |
1da177e4 | 6241 | netif_rx(skb); |
76cc8b13 | 6242 | input_queue_head_incr(oldsd); |
fec5e652 | 6243 | } |
76cc8b13 | 6244 | while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) { |
6e7676c1 | 6245 | netif_rx(skb); |
76cc8b13 TH |
6246 | input_queue_head_incr(oldsd); |
6247 | } | |
1da177e4 LT |
6248 | |
6249 | return NOTIFY_OK; | |
6250 | } | |
1da177e4 LT |
6251 | |
6252 | ||
7f353bf2 | 6253 | /** |
b63365a2 HX |
6254 | * netdev_increment_features - increment feature set by one |
6255 | * @all: current feature set | |
6256 | * @one: new feature set | |
6257 | * @mask: mask feature set | |
7f353bf2 HX |
6258 | * |
6259 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
6260 | * @one to the master device with current feature set @all. Will not |
6261 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 6262 | */ |
04ed3e74 | 6263 | u32 netdev_increment_features(u32 all, u32 one, u32 mask) |
b63365a2 | 6264 | { |
1742f183 MM |
6265 | if (mask & NETIF_F_GEN_CSUM) |
6266 | mask |= NETIF_F_ALL_CSUM; | |
6267 | mask |= NETIF_F_VLAN_CHALLENGED; | |
7f353bf2 | 6268 | |
1742f183 MM |
6269 | all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask; |
6270 | all &= one | ~NETIF_F_ALL_FOR_ALL; | |
c6e1a0d1 | 6271 | |
1742f183 MM |
6272 | /* If device needs checksumming, downgrade to it. */ |
6273 | if (all & (NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM)) | |
6274 | all &= ~NETIF_F_NO_CSUM; | |
7f353bf2 | 6275 | |
1742f183 MM |
6276 | /* If one device supports hw checksumming, set for all. */ |
6277 | if (all & NETIF_F_GEN_CSUM) | |
6278 | all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM); | |
7f353bf2 HX |
6279 | |
6280 | return all; | |
6281 | } | |
b63365a2 | 6282 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 6283 | |
30d97d35 PE |
6284 | static struct hlist_head *netdev_create_hash(void) |
6285 | { | |
6286 | int i; | |
6287 | struct hlist_head *hash; | |
6288 | ||
6289 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
6290 | if (hash != NULL) | |
6291 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
6292 | INIT_HLIST_HEAD(&hash[i]); | |
6293 | ||
6294 | return hash; | |
6295 | } | |
6296 | ||
881d966b | 6297 | /* Initialize per network namespace state */ |
4665079c | 6298 | static int __net_init netdev_init(struct net *net) |
881d966b | 6299 | { |
881d966b | 6300 | INIT_LIST_HEAD(&net->dev_base_head); |
881d966b | 6301 | |
30d97d35 PE |
6302 | net->dev_name_head = netdev_create_hash(); |
6303 | if (net->dev_name_head == NULL) | |
6304 | goto err_name; | |
881d966b | 6305 | |
30d97d35 PE |
6306 | net->dev_index_head = netdev_create_hash(); |
6307 | if (net->dev_index_head == NULL) | |
6308 | goto err_idx; | |
881d966b EB |
6309 | |
6310 | return 0; | |
30d97d35 PE |
6311 | |
6312 | err_idx: | |
6313 | kfree(net->dev_name_head); | |
6314 | err_name: | |
6315 | return -ENOMEM; | |
881d966b EB |
6316 | } |
6317 | ||
f0db275a SH |
6318 | /** |
6319 | * netdev_drivername - network driver for the device | |
6320 | * @dev: network device | |
f0db275a SH |
6321 | * |
6322 | * Determine network driver for device. | |
6323 | */ | |
3019de12 | 6324 | const char *netdev_drivername(const struct net_device *dev) |
6579e57b | 6325 | { |
cf04a4c7 SH |
6326 | const struct device_driver *driver; |
6327 | const struct device *parent; | |
3019de12 | 6328 | const char *empty = ""; |
6579e57b AV |
6329 | |
6330 | parent = dev->dev.parent; | |
6579e57b | 6331 | if (!parent) |
3019de12 | 6332 | return empty; |
6579e57b AV |
6333 | |
6334 | driver = parent->driver; | |
6335 | if (driver && driver->name) | |
3019de12 DM |
6336 | return driver->name; |
6337 | return empty; | |
6579e57b AV |
6338 | } |
6339 | ||
256df2f3 JP |
6340 | static int __netdev_printk(const char *level, const struct net_device *dev, |
6341 | struct va_format *vaf) | |
6342 | { | |
6343 | int r; | |
6344 | ||
6345 | if (dev && dev->dev.parent) | |
6346 | r = dev_printk(level, dev->dev.parent, "%s: %pV", | |
6347 | netdev_name(dev), vaf); | |
6348 | else if (dev) | |
6349 | r = printk("%s%s: %pV", level, netdev_name(dev), vaf); | |
6350 | else | |
6351 | r = printk("%s(NULL net_device): %pV", level, vaf); | |
6352 | ||
6353 | return r; | |
6354 | } | |
6355 | ||
6356 | int netdev_printk(const char *level, const struct net_device *dev, | |
6357 | const char *format, ...) | |
6358 | { | |
6359 | struct va_format vaf; | |
6360 | va_list args; | |
6361 | int r; | |
6362 | ||
6363 | va_start(args, format); | |
6364 | ||
6365 | vaf.fmt = format; | |
6366 | vaf.va = &args; | |
6367 | ||
6368 | r = __netdev_printk(level, dev, &vaf); | |
6369 | va_end(args); | |
6370 | ||
6371 | return r; | |
6372 | } | |
6373 | EXPORT_SYMBOL(netdev_printk); | |
6374 | ||
6375 | #define define_netdev_printk_level(func, level) \ | |
6376 | int func(const struct net_device *dev, const char *fmt, ...) \ | |
6377 | { \ | |
6378 | int r; \ | |
6379 | struct va_format vaf; \ | |
6380 | va_list args; \ | |
6381 | \ | |
6382 | va_start(args, fmt); \ | |
6383 | \ | |
6384 | vaf.fmt = fmt; \ | |
6385 | vaf.va = &args; \ | |
6386 | \ | |
6387 | r = __netdev_printk(level, dev, &vaf); \ | |
6388 | va_end(args); \ | |
6389 | \ | |
6390 | return r; \ | |
6391 | } \ | |
6392 | EXPORT_SYMBOL(func); | |
6393 | ||
6394 | define_netdev_printk_level(netdev_emerg, KERN_EMERG); | |
6395 | define_netdev_printk_level(netdev_alert, KERN_ALERT); | |
6396 | define_netdev_printk_level(netdev_crit, KERN_CRIT); | |
6397 | define_netdev_printk_level(netdev_err, KERN_ERR); | |
6398 | define_netdev_printk_level(netdev_warn, KERN_WARNING); | |
6399 | define_netdev_printk_level(netdev_notice, KERN_NOTICE); | |
6400 | define_netdev_printk_level(netdev_info, KERN_INFO); | |
6401 | ||
4665079c | 6402 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
6403 | { |
6404 | kfree(net->dev_name_head); | |
6405 | kfree(net->dev_index_head); | |
6406 | } | |
6407 | ||
022cbae6 | 6408 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
6409 | .init = netdev_init, |
6410 | .exit = netdev_exit, | |
6411 | }; | |
6412 | ||
4665079c | 6413 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 6414 | { |
e008b5fc | 6415 | struct net_device *dev, *aux; |
ce286d32 | 6416 | /* |
e008b5fc | 6417 | * Push all migratable network devices back to the |
ce286d32 EB |
6418 | * initial network namespace |
6419 | */ | |
6420 | rtnl_lock(); | |
e008b5fc | 6421 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 6422 | int err; |
aca51397 | 6423 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
6424 | |
6425 | /* Ignore unmoveable devices (i.e. loopback) */ | |
6426 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
6427 | continue; | |
6428 | ||
e008b5fc EB |
6429 | /* Leave virtual devices for the generic cleanup */ |
6430 | if (dev->rtnl_link_ops) | |
6431 | continue; | |
d0c082ce | 6432 | |
25985edc | 6433 | /* Push remaining network devices to init_net */ |
aca51397 PE |
6434 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
6435 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
ce286d32 | 6436 | if (err) { |
aca51397 | 6437 | printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n", |
ce286d32 | 6438 | __func__, dev->name, err); |
aca51397 | 6439 | BUG(); |
ce286d32 EB |
6440 | } |
6441 | } | |
6442 | rtnl_unlock(); | |
6443 | } | |
6444 | ||
04dc7f6b EB |
6445 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
6446 | { | |
6447 | /* At exit all network devices most be removed from a network | |
b595076a | 6448 | * namespace. Do this in the reverse order of registration. |
04dc7f6b EB |
6449 | * Do this across as many network namespaces as possible to |
6450 | * improve batching efficiency. | |
6451 | */ | |
6452 | struct net_device *dev; | |
6453 | struct net *net; | |
6454 | LIST_HEAD(dev_kill_list); | |
6455 | ||
6456 | rtnl_lock(); | |
6457 | list_for_each_entry(net, net_list, exit_list) { | |
6458 | for_each_netdev_reverse(net, dev) { | |
6459 | if (dev->rtnl_link_ops) | |
6460 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); | |
6461 | else | |
6462 | unregister_netdevice_queue(dev, &dev_kill_list); | |
6463 | } | |
6464 | } | |
6465 | unregister_netdevice_many(&dev_kill_list); | |
ceaaec98 | 6466 | list_del(&dev_kill_list); |
04dc7f6b EB |
6467 | rtnl_unlock(); |
6468 | } | |
6469 | ||
022cbae6 | 6470 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 6471 | .exit = default_device_exit, |
04dc7f6b | 6472 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
6473 | }; |
6474 | ||
1da177e4 LT |
6475 | /* |
6476 | * Initialize the DEV module. At boot time this walks the device list and | |
6477 | * unhooks any devices that fail to initialise (normally hardware not | |
6478 | * present) and leaves us with a valid list of present and active devices. | |
6479 | * | |
6480 | */ | |
6481 | ||
6482 | /* | |
6483 | * This is called single threaded during boot, so no need | |
6484 | * to take the rtnl semaphore. | |
6485 | */ | |
6486 | static int __init net_dev_init(void) | |
6487 | { | |
6488 | int i, rc = -ENOMEM; | |
6489 | ||
6490 | BUG_ON(!dev_boot_phase); | |
6491 | ||
1da177e4 LT |
6492 | if (dev_proc_init()) |
6493 | goto out; | |
6494 | ||
8b41d188 | 6495 | if (netdev_kobject_init()) |
1da177e4 LT |
6496 | goto out; |
6497 | ||
6498 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 6499 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
6500 | INIT_LIST_HEAD(&ptype_base[i]); |
6501 | ||
881d966b EB |
6502 | if (register_pernet_subsys(&netdev_net_ops)) |
6503 | goto out; | |
1da177e4 LT |
6504 | |
6505 | /* | |
6506 | * Initialise the packet receive queues. | |
6507 | */ | |
6508 | ||
6f912042 | 6509 | for_each_possible_cpu(i) { |
e36fa2f7 | 6510 | struct softnet_data *sd = &per_cpu(softnet_data, i); |
1da177e4 | 6511 | |
dee42870 | 6512 | memset(sd, 0, sizeof(*sd)); |
e36fa2f7 | 6513 | skb_queue_head_init(&sd->input_pkt_queue); |
6e7676c1 | 6514 | skb_queue_head_init(&sd->process_queue); |
e36fa2f7 ED |
6515 | sd->completion_queue = NULL; |
6516 | INIT_LIST_HEAD(&sd->poll_list); | |
a9cbd588 CG |
6517 | sd->output_queue = NULL; |
6518 | sd->output_queue_tailp = &sd->output_queue; | |
df334545 | 6519 | #ifdef CONFIG_RPS |
e36fa2f7 ED |
6520 | sd->csd.func = rps_trigger_softirq; |
6521 | sd->csd.info = sd; | |
6522 | sd->csd.flags = 0; | |
6523 | sd->cpu = i; | |
1e94d72f | 6524 | #endif |
0a9627f2 | 6525 | |
e36fa2f7 ED |
6526 | sd->backlog.poll = process_backlog; |
6527 | sd->backlog.weight = weight_p; | |
6528 | sd->backlog.gro_list = NULL; | |
6529 | sd->backlog.gro_count = 0; | |
1da177e4 LT |
6530 | } |
6531 | ||
1da177e4 LT |
6532 | dev_boot_phase = 0; |
6533 | ||
505d4f73 EB |
6534 | /* The loopback device is special if any other network devices |
6535 | * is present in a network namespace the loopback device must | |
6536 | * be present. Since we now dynamically allocate and free the | |
6537 | * loopback device ensure this invariant is maintained by | |
6538 | * keeping the loopback device as the first device on the | |
6539 | * list of network devices. Ensuring the loopback devices | |
6540 | * is the first device that appears and the last network device | |
6541 | * that disappears. | |
6542 | */ | |
6543 | if (register_pernet_device(&loopback_net_ops)) | |
6544 | goto out; | |
6545 | ||
6546 | if (register_pernet_device(&default_device_ops)) | |
6547 | goto out; | |
6548 | ||
962cf36c CM |
6549 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
6550 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 LT |
6551 | |
6552 | hotcpu_notifier(dev_cpu_callback, 0); | |
6553 | dst_init(); | |
6554 | dev_mcast_init(); | |
6555 | rc = 0; | |
6556 | out: | |
6557 | return rc; | |
6558 | } | |
6559 | ||
6560 | subsys_initcall(net_dev_init); | |
6561 | ||
e88721f8 KK |
6562 | static int __init initialize_hashrnd(void) |
6563 | { | |
0a9627f2 | 6564 | get_random_bytes(&hashrnd, sizeof(hashrnd)); |
e88721f8 KK |
6565 | return 0; |
6566 | } | |
6567 | ||
6568 | late_initcall_sync(initialize_hashrnd); | |
6569 |