]>
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> | |
1da177e4 | 76 | #include <linux/bitops.h> |
4fc268d2 | 77 | #include <linux/capability.h> |
1da177e4 LT |
78 | #include <linux/cpu.h> |
79 | #include <linux/types.h> | |
80 | #include <linux/kernel.h> | |
08e9897d | 81 | #include <linux/hash.h> |
5a0e3ad6 | 82 | #include <linux/slab.h> |
1da177e4 | 83 | #include <linux/sched.h> |
4a3e2f71 | 84 | #include <linux/mutex.h> |
1da177e4 LT |
85 | #include <linux/string.h> |
86 | #include <linux/mm.h> | |
87 | #include <linux/socket.h> | |
88 | #include <linux/sockios.h> | |
89 | #include <linux/errno.h> | |
90 | #include <linux/interrupt.h> | |
91 | #include <linux/if_ether.h> | |
92 | #include <linux/netdevice.h> | |
93 | #include <linux/etherdevice.h> | |
0187bdfb | 94 | #include <linux/ethtool.h> |
1da177e4 LT |
95 | #include <linux/notifier.h> |
96 | #include <linux/skbuff.h> | |
457c4cbc | 97 | #include <net/net_namespace.h> |
1da177e4 LT |
98 | #include <net/sock.h> |
99 | #include <linux/rtnetlink.h> | |
100 | #include <linux/proc_fs.h> | |
101 | #include <linux/seq_file.h> | |
102 | #include <linux/stat.h> | |
1da177e4 LT |
103 | #include <net/dst.h> |
104 | #include <net/pkt_sched.h> | |
105 | #include <net/checksum.h> | |
44540960 | 106 | #include <net/xfrm.h> |
1da177e4 LT |
107 | #include <linux/highmem.h> |
108 | #include <linux/init.h> | |
109 | #include <linux/kmod.h> | |
110 | #include <linux/module.h> | |
1da177e4 LT |
111 | #include <linux/netpoll.h> |
112 | #include <linux/rcupdate.h> | |
113 | #include <linux/delay.h> | |
295f4a1f | 114 | #include <net/wext.h> |
1da177e4 | 115 | #include <net/iw_handler.h> |
1da177e4 | 116 | #include <asm/current.h> |
5bdb9886 | 117 | #include <linux/audit.h> |
db217334 | 118 | #include <linux/dmaengine.h> |
f6a78bfc | 119 | #include <linux/err.h> |
c7fa9d18 | 120 | #include <linux/ctype.h> |
723e98b7 | 121 | #include <linux/if_arp.h> |
6de329e2 | 122 | #include <linux/if_vlan.h> |
8f0f2223 | 123 | #include <linux/ip.h> |
ad55dcaf | 124 | #include <net/ip.h> |
8f0f2223 DM |
125 | #include <linux/ipv6.h> |
126 | #include <linux/in.h> | |
b6b2fed1 DM |
127 | #include <linux/jhash.h> |
128 | #include <linux/random.h> | |
9cbc1cb8 | 129 | #include <trace/events/napi.h> |
cf66ba58 | 130 | #include <trace/events/net.h> |
07dc22e7 | 131 | #include <trace/events/skb.h> |
5acbbd42 | 132 | #include <linux/pci.h> |
caeda9b9 | 133 | #include <linux/inetdevice.h> |
c445477d | 134 | #include <linux/cpu_rmap.h> |
4dc360c5 | 135 | #include <linux/net_tstamp.h> |
c5905afb | 136 | #include <linux/static_key.h> |
4504b861 | 137 | #include <net/flow_keys.h> |
1da177e4 | 138 | |
342709ef PE |
139 | #include "net-sysfs.h" |
140 | ||
d565b0a1 HX |
141 | /* Instead of increasing this, you should create a hash table. */ |
142 | #define MAX_GRO_SKBS 8 | |
143 | ||
5d38a079 HX |
144 | /* This should be increased if a protocol with a bigger head is added. */ |
145 | #define GRO_MAX_HEAD (MAX_HEADER + 128) | |
146 | ||
1da177e4 LT |
147 | /* |
148 | * The list of packet types we will receive (as opposed to discard) | |
149 | * and the routines to invoke. | |
150 | * | |
151 | * Why 16. Because with 16 the only overlap we get on a hash of the | |
152 | * low nibble of the protocol value is RARP/SNAP/X.25. | |
153 | * | |
154 | * NOTE: That is no longer true with the addition of VLAN tags. Not | |
155 | * sure which should go first, but I bet it won't make much | |
156 | * difference if we are running VLANs. The good news is that | |
157 | * this protocol won't be in the list unless compiled in, so | |
3041a069 | 158 | * the average user (w/out VLANs) will not be adversely affected. |
1da177e4 LT |
159 | * --BLG |
160 | * | |
161 | * 0800 IP | |
162 | * 8100 802.1Q VLAN | |
163 | * 0001 802.3 | |
164 | * 0002 AX.25 | |
165 | * 0004 802.2 | |
166 | * 8035 RARP | |
167 | * 0005 SNAP | |
168 | * 0805 X.25 | |
169 | * 0806 ARP | |
170 | * 8137 IPX | |
171 | * 0009 Localtalk | |
172 | * 86DD IPv6 | |
173 | */ | |
174 | ||
82d8a867 PE |
175 | #define PTYPE_HASH_SIZE (16) |
176 | #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1) | |
177 | ||
1da177e4 | 178 | static DEFINE_SPINLOCK(ptype_lock); |
82d8a867 | 179 | static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
6b2bedc3 | 180 | static struct list_head ptype_all __read_mostly; /* Taps */ |
1da177e4 | 181 | |
1da177e4 | 182 | /* |
7562f876 | 183 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
184 | * semaphore. |
185 | * | |
c6d14c84 | 186 | * Pure readers hold dev_base_lock for reading, or rcu_read_lock() |
1da177e4 LT |
187 | * |
188 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 189 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
190 | * actual updates. This allows pure readers to access the list even |
191 | * while a writer is preparing to update it. | |
192 | * | |
193 | * To put it another way, dev_base_lock is held for writing only to | |
194 | * protect against pure readers; the rtnl semaphore provides the | |
195 | * protection against other writers. | |
196 | * | |
197 | * See, for example usages, register_netdevice() and | |
198 | * unregister_netdevice(), which must be called with the rtnl | |
199 | * semaphore held. | |
200 | */ | |
1da177e4 | 201 | DEFINE_RWLOCK(dev_base_lock); |
1da177e4 LT |
202 | EXPORT_SYMBOL(dev_base_lock); |
203 | ||
4e985ada TG |
204 | static inline void dev_base_seq_inc(struct net *net) |
205 | { | |
206 | while (++net->dev_base_seq == 0); | |
207 | } | |
208 | ||
881d966b | 209 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 | 210 | { |
95c96174 ED |
211 | unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ)); |
212 | ||
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, | |
211ed865 PG |
303 | ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, |
304 | ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, | |
305 | ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; | |
723e98b7 | 306 | |
36cbd3dc | 307 | static const char *const netdev_lock_name[] = |
723e98b7 JP |
308 | {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", |
309 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
310 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
311 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
312 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
313 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
314 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
315 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
316 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
317 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
318 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
319 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
211ed865 PG |
320 | "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", |
321 | "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", | |
322 | "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; | |
723e98b7 JP |
323 | |
324 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
cf508b12 | 325 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; |
723e98b7 JP |
326 | |
327 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
328 | { | |
329 | int i; | |
330 | ||
331 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
332 | if (netdev_lock_type[i] == dev_type) | |
333 | return i; | |
334 | /* the last key is used by default */ | |
335 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
336 | } | |
337 | ||
cf508b12 DM |
338 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
339 | unsigned short dev_type) | |
723e98b7 JP |
340 | { |
341 | int i; | |
342 | ||
343 | i = netdev_lock_pos(dev_type); | |
344 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
345 | netdev_lock_name[i]); | |
346 | } | |
cf508b12 DM |
347 | |
348 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
349 | { | |
350 | int i; | |
351 | ||
352 | i = netdev_lock_pos(dev->type); | |
353 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
354 | &netdev_addr_lock_key[i], | |
355 | netdev_lock_name[i]); | |
356 | } | |
723e98b7 | 357 | #else |
cf508b12 DM |
358 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
359 | unsigned short dev_type) | |
360 | { | |
361 | } | |
362 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
723e98b7 JP |
363 | { |
364 | } | |
365 | #endif | |
1da177e4 LT |
366 | |
367 | /******************************************************************************* | |
368 | ||
369 | Protocol management and registration routines | |
370 | ||
371 | *******************************************************************************/ | |
372 | ||
1da177e4 LT |
373 | /* |
374 | * Add a protocol ID to the list. Now that the input handler is | |
375 | * smarter we can dispense with all the messy stuff that used to be | |
376 | * here. | |
377 | * | |
378 | * BEWARE!!! Protocol handlers, mangling input packets, | |
379 | * MUST BE last in hash buckets and checking protocol handlers | |
380 | * MUST start from promiscuous ptype_all chain in net_bh. | |
381 | * It is true now, do not change it. | |
382 | * Explanation follows: if protocol handler, mangling packet, will | |
383 | * be the first on list, it is not able to sense, that packet | |
384 | * is cloned and should be copied-on-write, so that it will | |
385 | * change it and subsequent readers will get broken packet. | |
386 | * --ANK (980803) | |
387 | */ | |
388 | ||
c07b68e8 ED |
389 | static inline struct list_head *ptype_head(const struct packet_type *pt) |
390 | { | |
391 | if (pt->type == htons(ETH_P_ALL)) | |
392 | return &ptype_all; | |
393 | else | |
394 | return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
395 | } | |
396 | ||
1da177e4 LT |
397 | /** |
398 | * dev_add_pack - add packet handler | |
399 | * @pt: packet type declaration | |
400 | * | |
401 | * Add a protocol handler to the networking stack. The passed &packet_type | |
402 | * is linked into kernel lists and may not be freed until it has been | |
403 | * removed from the kernel lists. | |
404 | * | |
4ec93edb | 405 | * This call does not sleep therefore it can not |
1da177e4 LT |
406 | * guarantee all CPU's that are in middle of receiving packets |
407 | * will see the new packet type (until the next received packet). | |
408 | */ | |
409 | ||
410 | void dev_add_pack(struct packet_type *pt) | |
411 | { | |
c07b68e8 | 412 | struct list_head *head = ptype_head(pt); |
1da177e4 | 413 | |
c07b68e8 ED |
414 | spin_lock(&ptype_lock); |
415 | list_add_rcu(&pt->list, head); | |
416 | spin_unlock(&ptype_lock); | |
1da177e4 | 417 | } |
d1b19dff | 418 | EXPORT_SYMBOL(dev_add_pack); |
1da177e4 | 419 | |
1da177e4 LT |
420 | /** |
421 | * __dev_remove_pack - remove packet handler | |
422 | * @pt: packet type declaration | |
423 | * | |
424 | * Remove a protocol handler that was previously added to the kernel | |
425 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
426 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 427 | * returns. |
1da177e4 LT |
428 | * |
429 | * The packet type might still be in use by receivers | |
430 | * and must not be freed until after all the CPU's have gone | |
431 | * through a quiescent state. | |
432 | */ | |
433 | void __dev_remove_pack(struct packet_type *pt) | |
434 | { | |
c07b68e8 | 435 | struct list_head *head = ptype_head(pt); |
1da177e4 LT |
436 | struct packet_type *pt1; |
437 | ||
c07b68e8 | 438 | spin_lock(&ptype_lock); |
1da177e4 LT |
439 | |
440 | list_for_each_entry(pt1, head, list) { | |
441 | if (pt == pt1) { | |
442 | list_del_rcu(&pt->list); | |
443 | goto out; | |
444 | } | |
445 | } | |
446 | ||
7b6cd1ce | 447 | pr_warn("dev_remove_pack: %p not found\n", pt); |
1da177e4 | 448 | out: |
c07b68e8 | 449 | spin_unlock(&ptype_lock); |
1da177e4 | 450 | } |
d1b19dff ED |
451 | EXPORT_SYMBOL(__dev_remove_pack); |
452 | ||
1da177e4 LT |
453 | /** |
454 | * dev_remove_pack - remove packet handler | |
455 | * @pt: packet type declaration | |
456 | * | |
457 | * Remove a protocol handler that was previously added to the kernel | |
458 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
459 | * from the kernel lists and can be freed or reused once this function | |
460 | * returns. | |
461 | * | |
462 | * This call sleeps to guarantee that no CPU is looking at the packet | |
463 | * type after return. | |
464 | */ | |
465 | void dev_remove_pack(struct packet_type *pt) | |
466 | { | |
467 | __dev_remove_pack(pt); | |
4ec93edb | 468 | |
1da177e4 LT |
469 | synchronize_net(); |
470 | } | |
d1b19dff | 471 | EXPORT_SYMBOL(dev_remove_pack); |
1da177e4 LT |
472 | |
473 | /****************************************************************************** | |
474 | ||
475 | Device Boot-time Settings Routines | |
476 | ||
477 | *******************************************************************************/ | |
478 | ||
479 | /* Boot time configuration table */ | |
480 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
481 | ||
482 | /** | |
483 | * netdev_boot_setup_add - add new setup entry | |
484 | * @name: name of the device | |
485 | * @map: configured settings for the device | |
486 | * | |
487 | * Adds new setup entry to the dev_boot_setup list. The function | |
488 | * returns 0 on error and 1 on success. This is a generic routine to | |
489 | * all netdevices. | |
490 | */ | |
491 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
492 | { | |
493 | struct netdev_boot_setup *s; | |
494 | int i; | |
495 | ||
496 | s = dev_boot_setup; | |
497 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
498 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
499 | memset(s[i].name, 0, sizeof(s[i].name)); | |
93b3cff9 | 500 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
501 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
502 | break; | |
503 | } | |
504 | } | |
505 | ||
506 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
507 | } | |
508 | ||
509 | /** | |
510 | * netdev_boot_setup_check - check boot time settings | |
511 | * @dev: the netdevice | |
512 | * | |
513 | * Check boot time settings for the device. | |
514 | * The found settings are set for the device to be used | |
515 | * later in the device probing. | |
516 | * Returns 0 if no settings found, 1 if they are. | |
517 | */ | |
518 | int netdev_boot_setup_check(struct net_device *dev) | |
519 | { | |
520 | struct netdev_boot_setup *s = dev_boot_setup; | |
521 | int i; | |
522 | ||
523 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
524 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
93b3cff9 | 525 | !strcmp(dev->name, s[i].name)) { |
1da177e4 LT |
526 | dev->irq = s[i].map.irq; |
527 | dev->base_addr = s[i].map.base_addr; | |
528 | dev->mem_start = s[i].map.mem_start; | |
529 | dev->mem_end = s[i].map.mem_end; | |
530 | return 1; | |
531 | } | |
532 | } | |
533 | return 0; | |
534 | } | |
d1b19dff | 535 | EXPORT_SYMBOL(netdev_boot_setup_check); |
1da177e4 LT |
536 | |
537 | ||
538 | /** | |
539 | * netdev_boot_base - get address from boot time settings | |
540 | * @prefix: prefix for network device | |
541 | * @unit: id for network device | |
542 | * | |
543 | * Check boot time settings for the base address of device. | |
544 | * The found settings are set for the device to be used | |
545 | * later in the device probing. | |
546 | * Returns 0 if no settings found. | |
547 | */ | |
548 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
549 | { | |
550 | const struct netdev_boot_setup *s = dev_boot_setup; | |
551 | char name[IFNAMSIZ]; | |
552 | int i; | |
553 | ||
554 | sprintf(name, "%s%d", prefix, unit); | |
555 | ||
556 | /* | |
557 | * If device already registered then return base of 1 | |
558 | * to indicate not to probe for this interface | |
559 | */ | |
881d966b | 560 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
561 | return 1; |
562 | ||
563 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
564 | if (!strcmp(name, s[i].name)) | |
565 | return s[i].map.base_addr; | |
566 | return 0; | |
567 | } | |
568 | ||
569 | /* | |
570 | * Saves at boot time configured settings for any netdevice. | |
571 | */ | |
572 | int __init netdev_boot_setup(char *str) | |
573 | { | |
574 | int ints[5]; | |
575 | struct ifmap map; | |
576 | ||
577 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
578 | if (!str || !*str) | |
579 | return 0; | |
580 | ||
581 | /* Save settings */ | |
582 | memset(&map, 0, sizeof(map)); | |
583 | if (ints[0] > 0) | |
584 | map.irq = ints[1]; | |
585 | if (ints[0] > 1) | |
586 | map.base_addr = ints[2]; | |
587 | if (ints[0] > 2) | |
588 | map.mem_start = ints[3]; | |
589 | if (ints[0] > 3) | |
590 | map.mem_end = ints[4]; | |
591 | ||
592 | /* Add new entry to the list */ | |
593 | return netdev_boot_setup_add(str, &map); | |
594 | } | |
595 | ||
596 | __setup("netdev=", netdev_boot_setup); | |
597 | ||
598 | /******************************************************************************* | |
599 | ||
600 | Device Interface Subroutines | |
601 | ||
602 | *******************************************************************************/ | |
603 | ||
604 | /** | |
605 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 606 | * @net: the applicable net namespace |
1da177e4 LT |
607 | * @name: name to find |
608 | * | |
609 | * Find an interface by name. Must be called under RTNL semaphore | |
610 | * or @dev_base_lock. If the name is found a pointer to the device | |
611 | * is returned. If the name is not found then %NULL is returned. The | |
612 | * reference counters are not incremented so the caller must be | |
613 | * careful with locks. | |
614 | */ | |
615 | ||
881d966b | 616 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
617 | { |
618 | struct hlist_node *p; | |
0bd8d536 ED |
619 | struct net_device *dev; |
620 | struct hlist_head *head = dev_name_hash(net, name); | |
1da177e4 | 621 | |
0bd8d536 | 622 | hlist_for_each_entry(dev, p, head, name_hlist) |
1da177e4 LT |
623 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
624 | return dev; | |
0bd8d536 | 625 | |
1da177e4 LT |
626 | return NULL; |
627 | } | |
d1b19dff | 628 | EXPORT_SYMBOL(__dev_get_by_name); |
1da177e4 | 629 | |
72c9528b ED |
630 | /** |
631 | * dev_get_by_name_rcu - find a device by its name | |
632 | * @net: the applicable net namespace | |
633 | * @name: name to find | |
634 | * | |
635 | * Find an interface by name. | |
636 | * If the name is found a pointer to the device is returned. | |
637 | * If the name is not found then %NULL is returned. | |
638 | * The reference counters are not incremented so the caller must be | |
639 | * careful with locks. The caller must hold RCU lock. | |
640 | */ | |
641 | ||
642 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
643 | { | |
644 | struct hlist_node *p; | |
645 | struct net_device *dev; | |
646 | struct hlist_head *head = dev_name_hash(net, name); | |
647 | ||
648 | hlist_for_each_entry_rcu(dev, p, head, name_hlist) | |
649 | if (!strncmp(dev->name, name, IFNAMSIZ)) | |
650 | return dev; | |
651 | ||
652 | return NULL; | |
653 | } | |
654 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
655 | ||
1da177e4 LT |
656 | /** |
657 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 658 | * @net: the applicable net namespace |
1da177e4 LT |
659 | * @name: name to find |
660 | * | |
661 | * Find an interface by name. This can be called from any | |
662 | * context and does its own locking. The returned handle has | |
663 | * the usage count incremented and the caller must use dev_put() to | |
664 | * release it when it is no longer needed. %NULL is returned if no | |
665 | * matching device is found. | |
666 | */ | |
667 | ||
881d966b | 668 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
669 | { |
670 | struct net_device *dev; | |
671 | ||
72c9528b ED |
672 | rcu_read_lock(); |
673 | dev = dev_get_by_name_rcu(net, name); | |
1da177e4 LT |
674 | if (dev) |
675 | dev_hold(dev); | |
72c9528b | 676 | rcu_read_unlock(); |
1da177e4 LT |
677 | return dev; |
678 | } | |
d1b19dff | 679 | EXPORT_SYMBOL(dev_get_by_name); |
1da177e4 LT |
680 | |
681 | /** | |
682 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 683 | * @net: the applicable net namespace |
1da177e4 LT |
684 | * @ifindex: index of device |
685 | * | |
686 | * Search for an interface by index. Returns %NULL if the device | |
687 | * is not found or a pointer to the device. The device has not | |
688 | * had its reference counter increased so the caller must be careful | |
689 | * about locking. The caller must hold either the RTNL semaphore | |
690 | * or @dev_base_lock. | |
691 | */ | |
692 | ||
881d966b | 693 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
694 | { |
695 | struct hlist_node *p; | |
0bd8d536 ED |
696 | struct net_device *dev; |
697 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
1da177e4 | 698 | |
0bd8d536 | 699 | hlist_for_each_entry(dev, p, head, index_hlist) |
1da177e4 LT |
700 | if (dev->ifindex == ifindex) |
701 | return dev; | |
0bd8d536 | 702 | |
1da177e4 LT |
703 | return NULL; |
704 | } | |
d1b19dff | 705 | EXPORT_SYMBOL(__dev_get_by_index); |
1da177e4 | 706 | |
fb699dfd ED |
707 | /** |
708 | * dev_get_by_index_rcu - find a device by its ifindex | |
709 | * @net: the applicable net namespace | |
710 | * @ifindex: index of device | |
711 | * | |
712 | * Search for an interface by index. Returns %NULL if the device | |
713 | * is not found or a pointer to the device. The device has not | |
714 | * had its reference counter increased so the caller must be careful | |
715 | * about locking. The caller must hold RCU lock. | |
716 | */ | |
717 | ||
718 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
719 | { | |
720 | struct hlist_node *p; | |
721 | struct net_device *dev; | |
722 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
723 | ||
724 | hlist_for_each_entry_rcu(dev, p, head, index_hlist) | |
725 | if (dev->ifindex == ifindex) | |
726 | return dev; | |
727 | ||
728 | return NULL; | |
729 | } | |
730 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
731 | ||
1da177e4 LT |
732 | |
733 | /** | |
734 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 735 | * @net: the applicable net namespace |
1da177e4 LT |
736 | * @ifindex: index of device |
737 | * | |
738 | * Search for an interface by index. Returns NULL if the device | |
739 | * is not found or a pointer to the device. The device returned has | |
740 | * had a reference added and the pointer is safe until the user calls | |
741 | * dev_put to indicate they have finished with it. | |
742 | */ | |
743 | ||
881d966b | 744 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
745 | { |
746 | struct net_device *dev; | |
747 | ||
fb699dfd ED |
748 | rcu_read_lock(); |
749 | dev = dev_get_by_index_rcu(net, ifindex); | |
1da177e4 LT |
750 | if (dev) |
751 | dev_hold(dev); | |
fb699dfd | 752 | rcu_read_unlock(); |
1da177e4 LT |
753 | return dev; |
754 | } | |
d1b19dff | 755 | EXPORT_SYMBOL(dev_get_by_index); |
1da177e4 LT |
756 | |
757 | /** | |
941666c2 | 758 | * dev_getbyhwaddr_rcu - find a device by its hardware address |
c4ea43c5 | 759 | * @net: the applicable net namespace |
1da177e4 LT |
760 | * @type: media type of device |
761 | * @ha: hardware address | |
762 | * | |
763 | * Search for an interface by MAC address. Returns NULL if the device | |
c506653d ED |
764 | * is not found or a pointer to the device. |
765 | * The caller must hold RCU or RTNL. | |
941666c2 | 766 | * The returned device has not had its ref count increased |
1da177e4 LT |
767 | * and the caller must therefore be careful about locking |
768 | * | |
1da177e4 LT |
769 | */ |
770 | ||
941666c2 ED |
771 | struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, |
772 | const char *ha) | |
1da177e4 LT |
773 | { |
774 | struct net_device *dev; | |
775 | ||
941666c2 | 776 | for_each_netdev_rcu(net, dev) |
1da177e4 LT |
777 | if (dev->type == type && |
778 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
779 | return dev; |
780 | ||
781 | return NULL; | |
1da177e4 | 782 | } |
941666c2 | 783 | EXPORT_SYMBOL(dev_getbyhwaddr_rcu); |
cf309e3f | 784 | |
881d966b | 785 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
786 | { |
787 | struct net_device *dev; | |
788 | ||
4e9cac2b | 789 | ASSERT_RTNL(); |
881d966b | 790 | for_each_netdev(net, dev) |
4e9cac2b | 791 | if (dev->type == type) |
7562f876 PE |
792 | return dev; |
793 | ||
794 | return NULL; | |
4e9cac2b | 795 | } |
4e9cac2b PM |
796 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); |
797 | ||
881d966b | 798 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b | 799 | { |
99fe3c39 | 800 | struct net_device *dev, *ret = NULL; |
4e9cac2b | 801 | |
99fe3c39 ED |
802 | rcu_read_lock(); |
803 | for_each_netdev_rcu(net, dev) | |
804 | if (dev->type == type) { | |
805 | dev_hold(dev); | |
806 | ret = dev; | |
807 | break; | |
808 | } | |
809 | rcu_read_unlock(); | |
810 | return ret; | |
1da177e4 | 811 | } |
1da177e4 LT |
812 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
813 | ||
814 | /** | |
bb69ae04 | 815 | * dev_get_by_flags_rcu - find any device with given flags |
c4ea43c5 | 816 | * @net: the applicable net namespace |
1da177e4 LT |
817 | * @if_flags: IFF_* values |
818 | * @mask: bitmask of bits in if_flags to check | |
819 | * | |
820 | * Search for any interface with the given flags. Returns NULL if a device | |
bb69ae04 ED |
821 | * is not found or a pointer to the device. Must be called inside |
822 | * rcu_read_lock(), and result refcount is unchanged. | |
1da177e4 LT |
823 | */ |
824 | ||
bb69ae04 | 825 | struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags, |
d1b19dff | 826 | unsigned short mask) |
1da177e4 | 827 | { |
7562f876 | 828 | struct net_device *dev, *ret; |
1da177e4 | 829 | |
7562f876 | 830 | ret = NULL; |
c6d14c84 | 831 | for_each_netdev_rcu(net, dev) { |
1da177e4 | 832 | if (((dev->flags ^ if_flags) & mask) == 0) { |
7562f876 | 833 | ret = dev; |
1da177e4 LT |
834 | break; |
835 | } | |
836 | } | |
7562f876 | 837 | return ret; |
1da177e4 | 838 | } |
bb69ae04 | 839 | EXPORT_SYMBOL(dev_get_by_flags_rcu); |
1da177e4 LT |
840 | |
841 | /** | |
842 | * dev_valid_name - check if name is okay for network device | |
843 | * @name: name string | |
844 | * | |
845 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
846 | * to allow sysfs to work. We also disallow any kind of |
847 | * whitespace. | |
1da177e4 | 848 | */ |
95f050bf | 849 | bool dev_valid_name(const char *name) |
1da177e4 | 850 | { |
c7fa9d18 | 851 | if (*name == '\0') |
95f050bf | 852 | return false; |
b6fe17d6 | 853 | if (strlen(name) >= IFNAMSIZ) |
95f050bf | 854 | return false; |
c7fa9d18 | 855 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
95f050bf | 856 | return false; |
c7fa9d18 DM |
857 | |
858 | while (*name) { | |
859 | if (*name == '/' || isspace(*name)) | |
95f050bf | 860 | return false; |
c7fa9d18 DM |
861 | name++; |
862 | } | |
95f050bf | 863 | return true; |
1da177e4 | 864 | } |
d1b19dff | 865 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
866 | |
867 | /** | |
b267b179 EB |
868 | * __dev_alloc_name - allocate a name for a device |
869 | * @net: network namespace to allocate the device name in | |
1da177e4 | 870 | * @name: name format string |
b267b179 | 871 | * @buf: scratch buffer and result name string |
1da177e4 LT |
872 | * |
873 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
874 | * id. It scans list of devices to build up a free map, then chooses |
875 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
876 | * while allocating the name and adding the device in order to avoid | |
877 | * duplicates. | |
878 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
879 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
880 | */ |
881 | ||
b267b179 | 882 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
883 | { |
884 | int i = 0; | |
1da177e4 LT |
885 | const char *p; |
886 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 887 | unsigned long *inuse; |
1da177e4 LT |
888 | struct net_device *d; |
889 | ||
890 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
891 | if (p) { | |
892 | /* | |
893 | * Verify the string as this thing may have come from | |
894 | * the user. There must be either one "%d" and no other "%" | |
895 | * characters. | |
896 | */ | |
897 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
898 | return -EINVAL; | |
899 | ||
900 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 901 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
902 | if (!inuse) |
903 | return -ENOMEM; | |
904 | ||
881d966b | 905 | for_each_netdev(net, d) { |
1da177e4 LT |
906 | if (!sscanf(d->name, name, &i)) |
907 | continue; | |
908 | if (i < 0 || i >= max_netdevices) | |
909 | continue; | |
910 | ||
911 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 912 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
913 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
914 | set_bit(i, inuse); | |
915 | } | |
916 | ||
917 | i = find_first_zero_bit(inuse, max_netdevices); | |
918 | free_page((unsigned long) inuse); | |
919 | } | |
920 | ||
d9031024 OP |
921 | if (buf != name) |
922 | snprintf(buf, IFNAMSIZ, name, i); | |
b267b179 | 923 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 924 | return i; |
1da177e4 LT |
925 | |
926 | /* It is possible to run out of possible slots | |
927 | * when the name is long and there isn't enough space left | |
928 | * for the digits, or if all bits are used. | |
929 | */ | |
930 | return -ENFILE; | |
931 | } | |
932 | ||
b267b179 EB |
933 | /** |
934 | * dev_alloc_name - allocate a name for a device | |
935 | * @dev: device | |
936 | * @name: name format string | |
937 | * | |
938 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
939 | * id. It scans list of devices to build up a free map, then chooses | |
940 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
941 | * while allocating the name and adding the device in order to avoid | |
942 | * duplicates. | |
943 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
944 | * Returns the number of the unit assigned or a negative errno code. | |
945 | */ | |
946 | ||
947 | int dev_alloc_name(struct net_device *dev, const char *name) | |
948 | { | |
949 | char buf[IFNAMSIZ]; | |
950 | struct net *net; | |
951 | int ret; | |
952 | ||
c346dca1 YH |
953 | BUG_ON(!dev_net(dev)); |
954 | net = dev_net(dev); | |
b267b179 EB |
955 | ret = __dev_alloc_name(net, name, buf); |
956 | if (ret >= 0) | |
957 | strlcpy(dev->name, buf, IFNAMSIZ); | |
958 | return ret; | |
959 | } | |
d1b19dff | 960 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 961 | |
1c5cae81 | 962 | static int dev_get_valid_name(struct net_device *dev, const char *name) |
d9031024 | 963 | { |
8ce6cebc DL |
964 | struct net *net; |
965 | ||
966 | BUG_ON(!dev_net(dev)); | |
967 | net = dev_net(dev); | |
968 | ||
d9031024 OP |
969 | if (!dev_valid_name(name)) |
970 | return -EINVAL; | |
971 | ||
1c5cae81 | 972 | if (strchr(name, '%')) |
8ce6cebc | 973 | return dev_alloc_name(dev, name); |
d9031024 OP |
974 | else if (__dev_get_by_name(net, name)) |
975 | return -EEXIST; | |
8ce6cebc DL |
976 | else if (dev->name != name) |
977 | strlcpy(dev->name, name, IFNAMSIZ); | |
d9031024 OP |
978 | |
979 | return 0; | |
980 | } | |
1da177e4 LT |
981 | |
982 | /** | |
983 | * dev_change_name - change name of a device | |
984 | * @dev: device | |
985 | * @newname: name (or format string) must be at least IFNAMSIZ | |
986 | * | |
987 | * Change name of a device, can pass format strings "eth%d". | |
988 | * for wildcarding. | |
989 | */ | |
cf04a4c7 | 990 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 991 | { |
fcc5a03a | 992 | char oldname[IFNAMSIZ]; |
1da177e4 | 993 | int err = 0; |
fcc5a03a | 994 | int ret; |
881d966b | 995 | struct net *net; |
1da177e4 LT |
996 | |
997 | ASSERT_RTNL(); | |
c346dca1 | 998 | BUG_ON(!dev_net(dev)); |
1da177e4 | 999 | |
c346dca1 | 1000 | net = dev_net(dev); |
1da177e4 LT |
1001 | if (dev->flags & IFF_UP) |
1002 | return -EBUSY; | |
1003 | ||
c8d90dca SH |
1004 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) |
1005 | return 0; | |
1006 | ||
fcc5a03a HX |
1007 | memcpy(oldname, dev->name, IFNAMSIZ); |
1008 | ||
1c5cae81 | 1009 | err = dev_get_valid_name(dev, newname); |
d9031024 OP |
1010 | if (err < 0) |
1011 | return err; | |
1da177e4 | 1012 | |
fcc5a03a | 1013 | rollback: |
a1b3f594 EB |
1014 | ret = device_rename(&dev->dev, dev->name); |
1015 | if (ret) { | |
1016 | memcpy(dev->name, oldname, IFNAMSIZ); | |
1017 | return ret; | |
dcc99773 | 1018 | } |
7f988eab HX |
1019 | |
1020 | write_lock_bh(&dev_base_lock); | |
372b2312 | 1021 | hlist_del_rcu(&dev->name_hlist); |
72c9528b ED |
1022 | write_unlock_bh(&dev_base_lock); |
1023 | ||
1024 | synchronize_rcu(); | |
1025 | ||
1026 | write_lock_bh(&dev_base_lock); | |
1027 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
7f988eab HX |
1028 | write_unlock_bh(&dev_base_lock); |
1029 | ||
056925ab | 1030 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1031 | ret = notifier_to_errno(ret); |
1032 | ||
1033 | if (ret) { | |
91e9c07b ED |
1034 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1035 | if (err >= 0) { | |
fcc5a03a HX |
1036 | err = ret; |
1037 | memcpy(dev->name, oldname, IFNAMSIZ); | |
1038 | goto rollback; | |
91e9c07b | 1039 | } else { |
7b6cd1ce | 1040 | pr_err("%s: name change rollback failed: %d\n", |
91e9c07b | 1041 | dev->name, ret); |
fcc5a03a HX |
1042 | } |
1043 | } | |
1da177e4 LT |
1044 | |
1045 | return err; | |
1046 | } | |
1047 | ||
0b815a1a SH |
1048 | /** |
1049 | * dev_set_alias - change ifalias of a device | |
1050 | * @dev: device | |
1051 | * @alias: name up to IFALIASZ | |
f0db275a | 1052 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1053 | * |
1054 | * Set ifalias for a device, | |
1055 | */ | |
1056 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1057 | { | |
1058 | ASSERT_RTNL(); | |
1059 | ||
1060 | if (len >= IFALIASZ) | |
1061 | return -EINVAL; | |
1062 | ||
96ca4a2c OH |
1063 | if (!len) { |
1064 | if (dev->ifalias) { | |
1065 | kfree(dev->ifalias); | |
1066 | dev->ifalias = NULL; | |
1067 | } | |
1068 | return 0; | |
1069 | } | |
1070 | ||
d1b19dff | 1071 | dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL); |
0b815a1a SH |
1072 | if (!dev->ifalias) |
1073 | return -ENOMEM; | |
1074 | ||
1075 | strlcpy(dev->ifalias, alias, len+1); | |
1076 | return len; | |
1077 | } | |
1078 | ||
1079 | ||
d8a33ac4 | 1080 | /** |
3041a069 | 1081 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1082 | * @dev: device to cause notification |
1083 | * | |
1084 | * Called to indicate a device has changed features. | |
1085 | */ | |
1086 | void netdev_features_change(struct net_device *dev) | |
1087 | { | |
056925ab | 1088 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1089 | } |
1090 | EXPORT_SYMBOL(netdev_features_change); | |
1091 | ||
1da177e4 LT |
1092 | /** |
1093 | * netdev_state_change - device changes state | |
1094 | * @dev: device to cause notification | |
1095 | * | |
1096 | * Called to indicate a device has changed state. This function calls | |
1097 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1098 | * to the routing socket. | |
1099 | */ | |
1100 | void netdev_state_change(struct net_device *dev) | |
1101 | { | |
1102 | if (dev->flags & IFF_UP) { | |
056925ab | 1103 | call_netdevice_notifiers(NETDEV_CHANGE, dev); |
1da177e4 LT |
1104 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0); |
1105 | } | |
1106 | } | |
d1b19dff | 1107 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1108 | |
3ca5b404 | 1109 | int netdev_bonding_change(struct net_device *dev, unsigned long event) |
c1da4ac7 | 1110 | { |
3ca5b404 | 1111 | return call_netdevice_notifiers(event, dev); |
c1da4ac7 OG |
1112 | } |
1113 | EXPORT_SYMBOL(netdev_bonding_change); | |
1114 | ||
1da177e4 LT |
1115 | /** |
1116 | * dev_load - load a network module | |
c4ea43c5 | 1117 | * @net: the applicable net namespace |
1da177e4 LT |
1118 | * @name: name of interface |
1119 | * | |
1120 | * If a network interface is not present and the process has suitable | |
1121 | * privileges this function loads the module. If module loading is not | |
1122 | * available in this kernel then it becomes a nop. | |
1123 | */ | |
1124 | ||
881d966b | 1125 | void dev_load(struct net *net, const char *name) |
1da177e4 | 1126 | { |
4ec93edb | 1127 | struct net_device *dev; |
8909c9ad | 1128 | int no_module; |
1da177e4 | 1129 | |
72c9528b ED |
1130 | rcu_read_lock(); |
1131 | dev = dev_get_by_name_rcu(net, name); | |
1132 | rcu_read_unlock(); | |
1da177e4 | 1133 | |
8909c9ad VK |
1134 | no_module = !dev; |
1135 | if (no_module && capable(CAP_NET_ADMIN)) | |
1136 | no_module = request_module("netdev-%s", name); | |
1137 | if (no_module && capable(CAP_SYS_MODULE)) { | |
1138 | if (!request_module("%s", name)) | |
7cecb523 VL |
1139 | pr_warn("Loading kernel module for a network device with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s instead.\n", |
1140 | name); | |
8909c9ad | 1141 | } |
1da177e4 | 1142 | } |
d1b19dff | 1143 | EXPORT_SYMBOL(dev_load); |
1da177e4 | 1144 | |
bd380811 | 1145 | static int __dev_open(struct net_device *dev) |
1da177e4 | 1146 | { |
d314774c | 1147 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1148 | int ret; |
1da177e4 | 1149 | |
e46b66bc BH |
1150 | ASSERT_RTNL(); |
1151 | ||
1da177e4 LT |
1152 | if (!netif_device_present(dev)) |
1153 | return -ENODEV; | |
1154 | ||
3b8bcfd5 JB |
1155 | ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); |
1156 | ret = notifier_to_errno(ret); | |
1157 | if (ret) | |
1158 | return ret; | |
1159 | ||
1da177e4 | 1160 | set_bit(__LINK_STATE_START, &dev->state); |
bada339b | 1161 | |
d314774c SH |
1162 | if (ops->ndo_validate_addr) |
1163 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1164 | |
d314774c SH |
1165 | if (!ret && ops->ndo_open) |
1166 | ret = ops->ndo_open(dev); | |
1da177e4 | 1167 | |
bada339b JG |
1168 | if (ret) |
1169 | clear_bit(__LINK_STATE_START, &dev->state); | |
1170 | else { | |
1da177e4 | 1171 | dev->flags |= IFF_UP; |
b4bd07c2 | 1172 | net_dmaengine_get(); |
4417da66 | 1173 | dev_set_rx_mode(dev); |
1da177e4 | 1174 | dev_activate(dev); |
1da177e4 | 1175 | } |
bada339b | 1176 | |
1da177e4 LT |
1177 | return ret; |
1178 | } | |
1179 | ||
1180 | /** | |
bd380811 PM |
1181 | * dev_open - prepare an interface for use. |
1182 | * @dev: device to open | |
1da177e4 | 1183 | * |
bd380811 PM |
1184 | * Takes a device from down to up state. The device's private open |
1185 | * function is invoked and then the multicast lists are loaded. Finally | |
1186 | * the device is moved into the up state and a %NETDEV_UP message is | |
1187 | * sent to the netdev notifier chain. | |
1188 | * | |
1189 | * Calling this function on an active interface is a nop. On a failure | |
1190 | * a negative errno code is returned. | |
1da177e4 | 1191 | */ |
bd380811 PM |
1192 | int dev_open(struct net_device *dev) |
1193 | { | |
1194 | int ret; | |
1195 | ||
bd380811 PM |
1196 | if (dev->flags & IFF_UP) |
1197 | return 0; | |
1198 | ||
bd380811 PM |
1199 | ret = __dev_open(dev); |
1200 | if (ret < 0) | |
1201 | return ret; | |
1202 | ||
bd380811 PM |
1203 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING); |
1204 | call_netdevice_notifiers(NETDEV_UP, dev); | |
1205 | ||
1206 | return ret; | |
1207 | } | |
1208 | EXPORT_SYMBOL(dev_open); | |
1209 | ||
44345724 | 1210 | static int __dev_close_many(struct list_head *head) |
1da177e4 | 1211 | { |
44345724 | 1212 | struct net_device *dev; |
e46b66bc | 1213 | |
bd380811 | 1214 | ASSERT_RTNL(); |
9d5010db DM |
1215 | might_sleep(); |
1216 | ||
44345724 | 1217 | list_for_each_entry(dev, head, unreg_list) { |
44345724 | 1218 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1219 | |
44345724 | 1220 | clear_bit(__LINK_STATE_START, &dev->state); |
1da177e4 | 1221 | |
44345724 OP |
1222 | /* Synchronize to scheduled poll. We cannot touch poll list, it |
1223 | * can be even on different cpu. So just clear netif_running(). | |
1224 | * | |
1225 | * dev->stop() will invoke napi_disable() on all of it's | |
1226 | * napi_struct instances on this device. | |
1227 | */ | |
1228 | smp_mb__after_clear_bit(); /* Commit netif_running(). */ | |
1229 | } | |
1da177e4 | 1230 | |
44345724 | 1231 | dev_deactivate_many(head); |
d8b2a4d2 | 1232 | |
44345724 OP |
1233 | list_for_each_entry(dev, head, unreg_list) { |
1234 | const struct net_device_ops *ops = dev->netdev_ops; | |
1da177e4 | 1235 | |
44345724 OP |
1236 | /* |
1237 | * Call the device specific close. This cannot fail. | |
1238 | * Only if device is UP | |
1239 | * | |
1240 | * We allow it to be called even after a DETACH hot-plug | |
1241 | * event. | |
1242 | */ | |
1243 | if (ops->ndo_stop) | |
1244 | ops->ndo_stop(dev); | |
1245 | ||
44345724 | 1246 | dev->flags &= ~IFF_UP; |
44345724 OP |
1247 | net_dmaengine_put(); |
1248 | } | |
1249 | ||
1250 | return 0; | |
1251 | } | |
1252 | ||
1253 | static int __dev_close(struct net_device *dev) | |
1254 | { | |
f87e6f47 | 1255 | int retval; |
44345724 OP |
1256 | LIST_HEAD(single); |
1257 | ||
1258 | list_add(&dev->unreg_list, &single); | |
f87e6f47 LT |
1259 | retval = __dev_close_many(&single); |
1260 | list_del(&single); | |
1261 | return retval; | |
44345724 OP |
1262 | } |
1263 | ||
3fbd8758 | 1264 | static int dev_close_many(struct list_head *head) |
44345724 OP |
1265 | { |
1266 | struct net_device *dev, *tmp; | |
1267 | LIST_HEAD(tmp_list); | |
1da177e4 | 1268 | |
44345724 OP |
1269 | list_for_each_entry_safe(dev, tmp, head, unreg_list) |
1270 | if (!(dev->flags & IFF_UP)) | |
1271 | list_move(&dev->unreg_list, &tmp_list); | |
1272 | ||
1273 | __dev_close_many(head); | |
1da177e4 | 1274 | |
44345724 OP |
1275 | list_for_each_entry(dev, head, unreg_list) { |
1276 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING); | |
1277 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
1278 | } | |
bd380811 | 1279 | |
44345724 OP |
1280 | /* rollback_registered_many needs the complete original list */ |
1281 | list_splice(&tmp_list, head); | |
bd380811 PM |
1282 | return 0; |
1283 | } | |
1284 | ||
1285 | /** | |
1286 | * dev_close - shutdown an interface. | |
1287 | * @dev: device to shutdown | |
1288 | * | |
1289 | * This function moves an active device into down state. A | |
1290 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1291 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1292 | * chain. | |
1293 | */ | |
1294 | int dev_close(struct net_device *dev) | |
1295 | { | |
e14a5993 ED |
1296 | if (dev->flags & IFF_UP) { |
1297 | LIST_HEAD(single); | |
1da177e4 | 1298 | |
e14a5993 ED |
1299 | list_add(&dev->unreg_list, &single); |
1300 | dev_close_many(&single); | |
1301 | list_del(&single); | |
1302 | } | |
1da177e4 LT |
1303 | return 0; |
1304 | } | |
d1b19dff | 1305 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1306 | |
1307 | ||
0187bdfb BH |
1308 | /** |
1309 | * dev_disable_lro - disable Large Receive Offload on a device | |
1310 | * @dev: device | |
1311 | * | |
1312 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1313 | * called under RTNL. This is needed if received packets may be | |
1314 | * forwarded to another interface. | |
1315 | */ | |
1316 | void dev_disable_lro(struct net_device *dev) | |
1317 | { | |
f11970e3 NH |
1318 | /* |
1319 | * If we're trying to disable lro on a vlan device | |
1320 | * use the underlying physical device instead | |
1321 | */ | |
1322 | if (is_vlan_dev(dev)) | |
1323 | dev = vlan_dev_real_dev(dev); | |
1324 | ||
bc5787c6 MM |
1325 | dev->wanted_features &= ~NETIF_F_LRO; |
1326 | netdev_update_features(dev); | |
27660515 | 1327 | |
22d5969f MM |
1328 | if (unlikely(dev->features & NETIF_F_LRO)) |
1329 | netdev_WARN(dev, "failed to disable LRO!\n"); | |
0187bdfb BH |
1330 | } |
1331 | EXPORT_SYMBOL(dev_disable_lro); | |
1332 | ||
1333 | ||
881d966b EB |
1334 | static int dev_boot_phase = 1; |
1335 | ||
1da177e4 LT |
1336 | /** |
1337 | * register_netdevice_notifier - register a network notifier block | |
1338 | * @nb: notifier | |
1339 | * | |
1340 | * Register a notifier to be called when network device events occur. | |
1341 | * The notifier passed is linked into the kernel structures and must | |
1342 | * not be reused until it has been unregistered. A negative errno code | |
1343 | * is returned on a failure. | |
1344 | * | |
1345 | * When registered all registration and up events are replayed | |
4ec93edb | 1346 | * to the new notifier to allow device to have a race free |
1da177e4 LT |
1347 | * view of the network device list. |
1348 | */ | |
1349 | ||
1350 | int register_netdevice_notifier(struct notifier_block *nb) | |
1351 | { | |
1352 | struct net_device *dev; | |
fcc5a03a | 1353 | struct net_device *last; |
881d966b | 1354 | struct net *net; |
1da177e4 LT |
1355 | int err; |
1356 | ||
1357 | rtnl_lock(); | |
f07d5b94 | 1358 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1359 | if (err) |
1360 | goto unlock; | |
881d966b EB |
1361 | if (dev_boot_phase) |
1362 | goto unlock; | |
1363 | for_each_net(net) { | |
1364 | for_each_netdev(net, dev) { | |
1365 | err = nb->notifier_call(nb, NETDEV_REGISTER, dev); | |
1366 | err = notifier_to_errno(err); | |
1367 | if (err) | |
1368 | goto rollback; | |
1369 | ||
1370 | if (!(dev->flags & IFF_UP)) | |
1371 | continue; | |
1da177e4 | 1372 | |
881d966b EB |
1373 | nb->notifier_call(nb, NETDEV_UP, dev); |
1374 | } | |
1da177e4 | 1375 | } |
fcc5a03a HX |
1376 | |
1377 | unlock: | |
1da177e4 LT |
1378 | rtnl_unlock(); |
1379 | return err; | |
fcc5a03a HX |
1380 | |
1381 | rollback: | |
1382 | last = dev; | |
881d966b EB |
1383 | for_each_net(net) { |
1384 | for_each_netdev(net, dev) { | |
1385 | if (dev == last) | |
8f891489 | 1386 | goto outroll; |
fcc5a03a | 1387 | |
881d966b EB |
1388 | if (dev->flags & IFF_UP) { |
1389 | nb->notifier_call(nb, NETDEV_GOING_DOWN, dev); | |
1390 | nb->notifier_call(nb, NETDEV_DOWN, dev); | |
1391 | } | |
1392 | nb->notifier_call(nb, NETDEV_UNREGISTER, dev); | |
a5ee1551 | 1393 | nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev); |
fcc5a03a | 1394 | } |
fcc5a03a | 1395 | } |
c67625a1 | 1396 | |
8f891489 | 1397 | outroll: |
c67625a1 | 1398 | raw_notifier_chain_unregister(&netdev_chain, nb); |
fcc5a03a | 1399 | goto unlock; |
1da177e4 | 1400 | } |
d1b19dff | 1401 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1402 | |
1403 | /** | |
1404 | * unregister_netdevice_notifier - unregister a network notifier block | |
1405 | * @nb: notifier | |
1406 | * | |
1407 | * Unregister a notifier previously registered by | |
1408 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1409 | * kernel structures and may then be reused. A negative errno code | |
1410 | * is returned on a failure. | |
7d3d43da EB |
1411 | * |
1412 | * After unregistering unregister and down device events are synthesized | |
1413 | * for all devices on the device list to the removed notifier to remove | |
1414 | * the need for special case cleanup code. | |
1da177e4 LT |
1415 | */ |
1416 | ||
1417 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1418 | { | |
7d3d43da EB |
1419 | struct net_device *dev; |
1420 | struct net *net; | |
9f514950 HX |
1421 | int err; |
1422 | ||
1423 | rtnl_lock(); | |
f07d5b94 | 1424 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
7d3d43da EB |
1425 | if (err) |
1426 | goto unlock; | |
1427 | ||
1428 | for_each_net(net) { | |
1429 | for_each_netdev(net, dev) { | |
1430 | if (dev->flags & IFF_UP) { | |
1431 | nb->notifier_call(nb, NETDEV_GOING_DOWN, dev); | |
1432 | nb->notifier_call(nb, NETDEV_DOWN, dev); | |
1433 | } | |
1434 | nb->notifier_call(nb, NETDEV_UNREGISTER, dev); | |
1435 | nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev); | |
1436 | } | |
1437 | } | |
1438 | unlock: | |
9f514950 HX |
1439 | rtnl_unlock(); |
1440 | return err; | |
1da177e4 | 1441 | } |
d1b19dff | 1442 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 LT |
1443 | |
1444 | /** | |
1445 | * call_netdevice_notifiers - call all network notifier blocks | |
1446 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1447 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1448 | * |
1449 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1450 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1451 | */ |
1452 | ||
ad7379d4 | 1453 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1454 | { |
ab930471 | 1455 | ASSERT_RTNL(); |
ad7379d4 | 1456 | return raw_notifier_call_chain(&netdev_chain, val, dev); |
1da177e4 | 1457 | } |
edf947f1 | 1458 | EXPORT_SYMBOL(call_netdevice_notifiers); |
1da177e4 | 1459 | |
c5905afb | 1460 | static struct static_key netstamp_needed __read_mostly; |
b90e5794 | 1461 | #ifdef HAVE_JUMP_LABEL |
c5905afb | 1462 | /* We are not allowed to call static_key_slow_dec() from irq context |
b90e5794 | 1463 | * If net_disable_timestamp() is called from irq context, defer the |
c5905afb | 1464 | * static_key_slow_dec() calls. |
b90e5794 ED |
1465 | */ |
1466 | static atomic_t netstamp_needed_deferred; | |
1467 | #endif | |
1da177e4 LT |
1468 | |
1469 | void net_enable_timestamp(void) | |
1470 | { | |
b90e5794 ED |
1471 | #ifdef HAVE_JUMP_LABEL |
1472 | int deferred = atomic_xchg(&netstamp_needed_deferred, 0); | |
1473 | ||
1474 | if (deferred) { | |
1475 | while (--deferred) | |
c5905afb | 1476 | static_key_slow_dec(&netstamp_needed); |
b90e5794 ED |
1477 | return; |
1478 | } | |
1479 | #endif | |
1480 | WARN_ON(in_interrupt()); | |
c5905afb | 1481 | static_key_slow_inc(&netstamp_needed); |
1da177e4 | 1482 | } |
d1b19dff | 1483 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
1484 | |
1485 | void net_disable_timestamp(void) | |
1486 | { | |
b90e5794 ED |
1487 | #ifdef HAVE_JUMP_LABEL |
1488 | if (in_interrupt()) { | |
1489 | atomic_inc(&netstamp_needed_deferred); | |
1490 | return; | |
1491 | } | |
1492 | #endif | |
c5905afb | 1493 | static_key_slow_dec(&netstamp_needed); |
1da177e4 | 1494 | } |
d1b19dff | 1495 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 1496 | |
3b098e2d | 1497 | static inline void net_timestamp_set(struct sk_buff *skb) |
1da177e4 | 1498 | { |
588f0330 | 1499 | skb->tstamp.tv64 = 0; |
c5905afb | 1500 | if (static_key_false(&netstamp_needed)) |
a61bbcf2 | 1501 | __net_timestamp(skb); |
1da177e4 LT |
1502 | } |
1503 | ||
588f0330 | 1504 | #define net_timestamp_check(COND, SKB) \ |
c5905afb | 1505 | if (static_key_false(&netstamp_needed)) { \ |
588f0330 ED |
1506 | if ((COND) && !(SKB)->tstamp.tv64) \ |
1507 | __net_timestamp(SKB); \ | |
1508 | } \ | |
3b098e2d | 1509 | |
4dc360c5 RC |
1510 | static int net_hwtstamp_validate(struct ifreq *ifr) |
1511 | { | |
1512 | struct hwtstamp_config cfg; | |
1513 | enum hwtstamp_tx_types tx_type; | |
1514 | enum hwtstamp_rx_filters rx_filter; | |
1515 | int tx_type_valid = 0; | |
1516 | int rx_filter_valid = 0; | |
1517 | ||
1518 | if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg))) | |
1519 | return -EFAULT; | |
1520 | ||
1521 | if (cfg.flags) /* reserved for future extensions */ | |
1522 | return -EINVAL; | |
1523 | ||
1524 | tx_type = cfg.tx_type; | |
1525 | rx_filter = cfg.rx_filter; | |
1526 | ||
1527 | switch (tx_type) { | |
1528 | case HWTSTAMP_TX_OFF: | |
1529 | case HWTSTAMP_TX_ON: | |
1530 | case HWTSTAMP_TX_ONESTEP_SYNC: | |
1531 | tx_type_valid = 1; | |
1532 | break; | |
1533 | } | |
1534 | ||
1535 | switch (rx_filter) { | |
1536 | case HWTSTAMP_FILTER_NONE: | |
1537 | case HWTSTAMP_FILTER_ALL: | |
1538 | case HWTSTAMP_FILTER_SOME: | |
1539 | case HWTSTAMP_FILTER_PTP_V1_L4_EVENT: | |
1540 | case HWTSTAMP_FILTER_PTP_V1_L4_SYNC: | |
1541 | case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ: | |
1542 | case HWTSTAMP_FILTER_PTP_V2_L4_EVENT: | |
1543 | case HWTSTAMP_FILTER_PTP_V2_L4_SYNC: | |
1544 | case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ: | |
1545 | case HWTSTAMP_FILTER_PTP_V2_L2_EVENT: | |
1546 | case HWTSTAMP_FILTER_PTP_V2_L2_SYNC: | |
1547 | case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ: | |
1548 | case HWTSTAMP_FILTER_PTP_V2_EVENT: | |
1549 | case HWTSTAMP_FILTER_PTP_V2_SYNC: | |
1550 | case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ: | |
1551 | rx_filter_valid = 1; | |
1552 | break; | |
1553 | } | |
1554 | ||
1555 | if (!tx_type_valid || !rx_filter_valid) | |
1556 | return -ERANGE; | |
1557 | ||
1558 | return 0; | |
1559 | } | |
1560 | ||
79b569f0 DL |
1561 | static inline bool is_skb_forwardable(struct net_device *dev, |
1562 | struct sk_buff *skb) | |
1563 | { | |
1564 | unsigned int len; | |
1565 | ||
1566 | if (!(dev->flags & IFF_UP)) | |
1567 | return false; | |
1568 | ||
1569 | len = dev->mtu + dev->hard_header_len + VLAN_HLEN; | |
1570 | if (skb->len <= len) | |
1571 | return true; | |
1572 | ||
1573 | /* if TSO is enabled, we don't care about the length as the packet | |
1574 | * could be forwarded without being segmented before | |
1575 | */ | |
1576 | if (skb_is_gso(skb)) | |
1577 | return true; | |
1578 | ||
1579 | return false; | |
1580 | } | |
1581 | ||
44540960 AB |
1582 | /** |
1583 | * dev_forward_skb - loopback an skb to another netif | |
1584 | * | |
1585 | * @dev: destination network device | |
1586 | * @skb: buffer to forward | |
1587 | * | |
1588 | * return values: | |
1589 | * NET_RX_SUCCESS (no congestion) | |
6ec82562 | 1590 | * NET_RX_DROP (packet was dropped, but freed) |
44540960 AB |
1591 | * |
1592 | * dev_forward_skb can be used for injecting an skb from the | |
1593 | * start_xmit function of one device into the receive queue | |
1594 | * of another device. | |
1595 | * | |
1596 | * The receiving device may be in another namespace, so | |
1597 | * we have to clear all information in the skb that could | |
1598 | * impact namespace isolation. | |
1599 | */ | |
1600 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
1601 | { | |
48c83012 MT |
1602 | if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) { |
1603 | if (skb_copy_ubufs(skb, GFP_ATOMIC)) { | |
1604 | atomic_long_inc(&dev->rx_dropped); | |
1605 | kfree_skb(skb); | |
1606 | return NET_RX_DROP; | |
1607 | } | |
1608 | } | |
1609 | ||
44540960 | 1610 | skb_orphan(skb); |
c736eefa | 1611 | nf_reset(skb); |
44540960 | 1612 | |
79b569f0 | 1613 | if (unlikely(!is_skb_forwardable(dev, skb))) { |
caf586e5 | 1614 | atomic_long_inc(&dev->rx_dropped); |
6ec82562 | 1615 | kfree_skb(skb); |
44540960 | 1616 | return NET_RX_DROP; |
6ec82562 | 1617 | } |
3b9785c6 | 1618 | skb->skb_iif = 0; |
59b9997b DM |
1619 | skb->dev = dev; |
1620 | skb_dst_drop(skb); | |
44540960 AB |
1621 | skb->tstamp.tv64 = 0; |
1622 | skb->pkt_type = PACKET_HOST; | |
1623 | skb->protocol = eth_type_trans(skb, dev); | |
59b9997b DM |
1624 | skb->mark = 0; |
1625 | secpath_reset(skb); | |
1626 | nf_reset(skb); | |
44540960 AB |
1627 | return netif_rx(skb); |
1628 | } | |
1629 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
1630 | ||
71d9dec2 CG |
1631 | static inline int deliver_skb(struct sk_buff *skb, |
1632 | struct packet_type *pt_prev, | |
1633 | struct net_device *orig_dev) | |
1634 | { | |
1635 | atomic_inc(&skb->users); | |
1636 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1637 | } | |
1638 | ||
1da177e4 LT |
1639 | /* |
1640 | * Support routine. Sends outgoing frames to any network | |
1641 | * taps currently in use. | |
1642 | */ | |
1643 | ||
f6a78bfc | 1644 | static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
1645 | { |
1646 | struct packet_type *ptype; | |
71d9dec2 CG |
1647 | struct sk_buff *skb2 = NULL; |
1648 | struct packet_type *pt_prev = NULL; | |
a61bbcf2 | 1649 | |
1da177e4 LT |
1650 | rcu_read_lock(); |
1651 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
1652 | /* Never send packets back to the socket | |
1653 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1654 | */ | |
1655 | if ((ptype->dev == dev || !ptype->dev) && | |
1656 | (ptype->af_packet_priv == NULL || | |
1657 | (struct sock *)ptype->af_packet_priv != skb->sk)) { | |
71d9dec2 CG |
1658 | if (pt_prev) { |
1659 | deliver_skb(skb2, pt_prev, skb->dev); | |
1660 | pt_prev = ptype; | |
1661 | continue; | |
1662 | } | |
1663 | ||
1664 | skb2 = skb_clone(skb, GFP_ATOMIC); | |
1da177e4 LT |
1665 | if (!skb2) |
1666 | break; | |
1667 | ||
70978182 ED |
1668 | net_timestamp_set(skb2); |
1669 | ||
1da177e4 LT |
1670 | /* skb->nh should be correctly |
1671 | set by sender, so that the second statement is | |
1672 | just protection against buggy protocols. | |
1673 | */ | |
459a98ed | 1674 | skb_reset_mac_header(skb2); |
1da177e4 | 1675 | |
d56f90a7 | 1676 | if (skb_network_header(skb2) < skb2->data || |
27a884dc | 1677 | skb2->network_header > skb2->tail) { |
e87cc472 JP |
1678 | net_crit_ratelimited("protocol %04x is buggy, dev %s\n", |
1679 | ntohs(skb2->protocol), | |
1680 | dev->name); | |
c1d2bbe1 | 1681 | skb_reset_network_header(skb2); |
1da177e4 LT |
1682 | } |
1683 | ||
b0e380b1 | 1684 | skb2->transport_header = skb2->network_header; |
1da177e4 | 1685 | skb2->pkt_type = PACKET_OUTGOING; |
71d9dec2 | 1686 | pt_prev = ptype; |
1da177e4 LT |
1687 | } |
1688 | } | |
71d9dec2 CG |
1689 | if (pt_prev) |
1690 | pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); | |
1da177e4 LT |
1691 | rcu_read_unlock(); |
1692 | } | |
1693 | ||
4f57c087 JF |
1694 | /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change |
1695 | * @dev: Network device | |
1696 | * @txq: number of queues available | |
1697 | * | |
1698 | * If real_num_tx_queues is changed the tc mappings may no longer be | |
1699 | * valid. To resolve this verify the tc mapping remains valid and if | |
1700 | * not NULL the mapping. With no priorities mapping to this | |
1701 | * offset/count pair it will no longer be used. In the worst case TC0 | |
1702 | * is invalid nothing can be done so disable priority mappings. If is | |
1703 | * expected that drivers will fix this mapping if they can before | |
1704 | * calling netif_set_real_num_tx_queues. | |
1705 | */ | |
bb134d22 | 1706 | static void netif_setup_tc(struct net_device *dev, unsigned int txq) |
4f57c087 JF |
1707 | { |
1708 | int i; | |
1709 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
1710 | ||
1711 | /* If TC0 is invalidated disable TC mapping */ | |
1712 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce | 1713 | pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); |
4f57c087 JF |
1714 | dev->num_tc = 0; |
1715 | return; | |
1716 | } | |
1717 | ||
1718 | /* Invalidated prio to tc mappings set to TC0 */ | |
1719 | for (i = 1; i < TC_BITMASK + 1; i++) { | |
1720 | int q = netdev_get_prio_tc_map(dev, i); | |
1721 | ||
1722 | tc = &dev->tc_to_txq[q]; | |
1723 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce JP |
1724 | pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", |
1725 | i, q); | |
4f57c087 JF |
1726 | netdev_set_prio_tc_map(dev, i, 0); |
1727 | } | |
1728 | } | |
1729 | } | |
1730 | ||
f0796d5c JF |
1731 | /* |
1732 | * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues | |
1733 | * greater then real_num_tx_queues stale skbs on the qdisc must be flushed. | |
1734 | */ | |
e6484930 | 1735 | int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) |
f0796d5c | 1736 | { |
1d24eb48 TH |
1737 | int rc; |
1738 | ||
e6484930 TH |
1739 | if (txq < 1 || txq > dev->num_tx_queues) |
1740 | return -EINVAL; | |
f0796d5c | 1741 | |
5c56580b BH |
1742 | if (dev->reg_state == NETREG_REGISTERED || |
1743 | dev->reg_state == NETREG_UNREGISTERING) { | |
e6484930 TH |
1744 | ASSERT_RTNL(); |
1745 | ||
1d24eb48 TH |
1746 | rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, |
1747 | txq); | |
bf264145 TH |
1748 | if (rc) |
1749 | return rc; | |
1750 | ||
4f57c087 JF |
1751 | if (dev->num_tc) |
1752 | netif_setup_tc(dev, txq); | |
1753 | ||
e6484930 TH |
1754 | if (txq < dev->real_num_tx_queues) |
1755 | qdisc_reset_all_tx_gt(dev, txq); | |
f0796d5c | 1756 | } |
e6484930 TH |
1757 | |
1758 | dev->real_num_tx_queues = txq; | |
1759 | return 0; | |
f0796d5c JF |
1760 | } |
1761 | EXPORT_SYMBOL(netif_set_real_num_tx_queues); | |
56079431 | 1762 | |
62fe0b40 BH |
1763 | #ifdef CONFIG_RPS |
1764 | /** | |
1765 | * netif_set_real_num_rx_queues - set actual number of RX queues used | |
1766 | * @dev: Network device | |
1767 | * @rxq: Actual number of RX queues | |
1768 | * | |
1769 | * This must be called either with the rtnl_lock held or before | |
1770 | * registration of the net device. Returns 0 on success, or a | |
4e7f7951 BH |
1771 | * negative error code. If called before registration, it always |
1772 | * succeeds. | |
62fe0b40 BH |
1773 | */ |
1774 | int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) | |
1775 | { | |
1776 | int rc; | |
1777 | ||
bd25fa7b TH |
1778 | if (rxq < 1 || rxq > dev->num_rx_queues) |
1779 | return -EINVAL; | |
1780 | ||
62fe0b40 BH |
1781 | if (dev->reg_state == NETREG_REGISTERED) { |
1782 | ASSERT_RTNL(); | |
1783 | ||
62fe0b40 BH |
1784 | rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, |
1785 | rxq); | |
1786 | if (rc) | |
1787 | return rc; | |
62fe0b40 BH |
1788 | } |
1789 | ||
1790 | dev->real_num_rx_queues = rxq; | |
1791 | return 0; | |
1792 | } | |
1793 | EXPORT_SYMBOL(netif_set_real_num_rx_queues); | |
1794 | #endif | |
1795 | ||
16917b87 YM |
1796 | /* netif_get_num_default_rss_queues - default number of RSS queues |
1797 | * | |
1798 | * This routine should set an upper limit on the number of RSS queues | |
1799 | * used by default by multiqueue devices. | |
1800 | */ | |
1801 | int netif_get_num_default_rss_queues() | |
1802 | { | |
1803 | return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); | |
1804 | } | |
1805 | EXPORT_SYMBOL(netif_get_num_default_rss_queues); | |
1806 | ||
def82a1d | 1807 | static inline void __netif_reschedule(struct Qdisc *q) |
56079431 | 1808 | { |
def82a1d JP |
1809 | struct softnet_data *sd; |
1810 | unsigned long flags; | |
56079431 | 1811 | |
def82a1d JP |
1812 | local_irq_save(flags); |
1813 | sd = &__get_cpu_var(softnet_data); | |
a9cbd588 CG |
1814 | q->next_sched = NULL; |
1815 | *sd->output_queue_tailp = q; | |
1816 | sd->output_queue_tailp = &q->next_sched; | |
def82a1d JP |
1817 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
1818 | local_irq_restore(flags); | |
1819 | } | |
1820 | ||
1821 | void __netif_schedule(struct Qdisc *q) | |
1822 | { | |
1823 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
1824 | __netif_reschedule(q); | |
56079431 DV |
1825 | } |
1826 | EXPORT_SYMBOL(__netif_schedule); | |
1827 | ||
bea3348e | 1828 | void dev_kfree_skb_irq(struct sk_buff *skb) |
56079431 | 1829 | { |
3578b0c8 | 1830 | if (atomic_dec_and_test(&skb->users)) { |
bea3348e SH |
1831 | struct softnet_data *sd; |
1832 | unsigned long flags; | |
56079431 | 1833 | |
bea3348e SH |
1834 | local_irq_save(flags); |
1835 | sd = &__get_cpu_var(softnet_data); | |
1836 | skb->next = sd->completion_queue; | |
1837 | sd->completion_queue = skb; | |
1838 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
1839 | local_irq_restore(flags); | |
1840 | } | |
56079431 | 1841 | } |
bea3348e | 1842 | EXPORT_SYMBOL(dev_kfree_skb_irq); |
56079431 DV |
1843 | |
1844 | void dev_kfree_skb_any(struct sk_buff *skb) | |
1845 | { | |
1846 | if (in_irq() || irqs_disabled()) | |
1847 | dev_kfree_skb_irq(skb); | |
1848 | else | |
1849 | dev_kfree_skb(skb); | |
1850 | } | |
1851 | EXPORT_SYMBOL(dev_kfree_skb_any); | |
1852 | ||
1853 | ||
bea3348e SH |
1854 | /** |
1855 | * netif_device_detach - mark device as removed | |
1856 | * @dev: network device | |
1857 | * | |
1858 | * Mark device as removed from system and therefore no longer available. | |
1859 | */ | |
56079431 DV |
1860 | void netif_device_detach(struct net_device *dev) |
1861 | { | |
1862 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
1863 | netif_running(dev)) { | |
d543103a | 1864 | netif_tx_stop_all_queues(dev); |
56079431 DV |
1865 | } |
1866 | } | |
1867 | EXPORT_SYMBOL(netif_device_detach); | |
1868 | ||
bea3348e SH |
1869 | /** |
1870 | * netif_device_attach - mark device as attached | |
1871 | * @dev: network device | |
1872 | * | |
1873 | * Mark device as attached from system and restart if needed. | |
1874 | */ | |
56079431 DV |
1875 | void netif_device_attach(struct net_device *dev) |
1876 | { | |
1877 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
1878 | netif_running(dev)) { | |
d543103a | 1879 | netif_tx_wake_all_queues(dev); |
4ec93edb | 1880 | __netdev_watchdog_up(dev); |
56079431 DV |
1881 | } |
1882 | } | |
1883 | EXPORT_SYMBOL(netif_device_attach); | |
1884 | ||
36c92474 BH |
1885 | static void skb_warn_bad_offload(const struct sk_buff *skb) |
1886 | { | |
65e9d2fa | 1887 | static const netdev_features_t null_features = 0; |
36c92474 BH |
1888 | struct net_device *dev = skb->dev; |
1889 | const char *driver = ""; | |
1890 | ||
1891 | if (dev && dev->dev.parent) | |
1892 | driver = dev_driver_string(dev->dev.parent); | |
1893 | ||
1894 | WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d " | |
1895 | "gso_type=%d ip_summed=%d\n", | |
65e9d2fa MM |
1896 | driver, dev ? &dev->features : &null_features, |
1897 | skb->sk ? &skb->sk->sk_route_caps : &null_features, | |
36c92474 BH |
1898 | skb->len, skb->data_len, skb_shinfo(skb)->gso_size, |
1899 | skb_shinfo(skb)->gso_type, skb->ip_summed); | |
1900 | } | |
1901 | ||
1da177e4 LT |
1902 | /* |
1903 | * Invalidate hardware checksum when packet is to be mangled, and | |
1904 | * complete checksum manually on outgoing path. | |
1905 | */ | |
84fa7933 | 1906 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 1907 | { |
d3bc23e7 | 1908 | __wsum csum; |
663ead3b | 1909 | int ret = 0, offset; |
1da177e4 | 1910 | |
84fa7933 | 1911 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
1912 | goto out_set_summed; |
1913 | ||
1914 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
36c92474 BH |
1915 | skb_warn_bad_offload(skb); |
1916 | return -EINVAL; | |
1da177e4 LT |
1917 | } |
1918 | ||
55508d60 | 1919 | offset = skb_checksum_start_offset(skb); |
a030847e HX |
1920 | BUG_ON(offset >= skb_headlen(skb)); |
1921 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
1922 | ||
1923 | offset += skb->csum_offset; | |
1924 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
1925 | ||
1926 | if (skb_cloned(skb) && | |
1927 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1da177e4 LT |
1928 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
1929 | if (ret) | |
1930 | goto out; | |
1931 | } | |
1932 | ||
a030847e | 1933 | *(__sum16 *)(skb->data + offset) = csum_fold(csum); |
a430a43d | 1934 | out_set_summed: |
1da177e4 | 1935 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 1936 | out: |
1da177e4 LT |
1937 | return ret; |
1938 | } | |
d1b19dff | 1939 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 1940 | |
f6a78bfc HX |
1941 | /** |
1942 | * skb_gso_segment - Perform segmentation on skb. | |
1943 | * @skb: buffer to segment | |
576a30eb | 1944 | * @features: features for the output path (see dev->features) |
f6a78bfc HX |
1945 | * |
1946 | * This function segments the given skb and returns a list of segments. | |
576a30eb HX |
1947 | * |
1948 | * It may return NULL if the skb requires no segmentation. This is | |
1949 | * only possible when GSO is used for verifying header integrity. | |
f6a78bfc | 1950 | */ |
c8f44aff MM |
1951 | struct sk_buff *skb_gso_segment(struct sk_buff *skb, |
1952 | netdev_features_t features) | |
f6a78bfc HX |
1953 | { |
1954 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
1955 | struct packet_type *ptype; | |
252e3346 | 1956 | __be16 type = skb->protocol; |
c8d5bcd1 | 1957 | int vlan_depth = ETH_HLEN; |
a430a43d | 1958 | int err; |
f6a78bfc | 1959 | |
c8d5bcd1 JG |
1960 | while (type == htons(ETH_P_8021Q)) { |
1961 | struct vlan_hdr *vh; | |
7b9c6090 | 1962 | |
c8d5bcd1 | 1963 | if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN))) |
7b9c6090 JG |
1964 | return ERR_PTR(-EINVAL); |
1965 | ||
c8d5bcd1 JG |
1966 | vh = (struct vlan_hdr *)(skb->data + vlan_depth); |
1967 | type = vh->h_vlan_encapsulated_proto; | |
1968 | vlan_depth += VLAN_HLEN; | |
7b9c6090 JG |
1969 | } |
1970 | ||
459a98ed | 1971 | skb_reset_mac_header(skb); |
b0e380b1 | 1972 | skb->mac_len = skb->network_header - skb->mac_header; |
f6a78bfc HX |
1973 | __skb_pull(skb, skb->mac_len); |
1974 | ||
67fd1a73 | 1975 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { |
36c92474 | 1976 | skb_warn_bad_offload(skb); |
67fd1a73 | 1977 | |
a430a43d HX |
1978 | if (skb_header_cloned(skb) && |
1979 | (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))) | |
1980 | return ERR_PTR(err); | |
1981 | } | |
1982 | ||
f6a78bfc | 1983 | rcu_read_lock(); |
82d8a867 PE |
1984 | list_for_each_entry_rcu(ptype, |
1985 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
f6a78bfc | 1986 | if (ptype->type == type && !ptype->dev && ptype->gso_segment) { |
84fa7933 | 1987 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { |
a430a43d HX |
1988 | err = ptype->gso_send_check(skb); |
1989 | segs = ERR_PTR(err); | |
1990 | if (err || skb_gso_ok(skb, features)) | |
1991 | break; | |
d56f90a7 ACM |
1992 | __skb_push(skb, (skb->data - |
1993 | skb_network_header(skb))); | |
a430a43d | 1994 | } |
576a30eb | 1995 | segs = ptype->gso_segment(skb, features); |
f6a78bfc HX |
1996 | break; |
1997 | } | |
1998 | } | |
1999 | rcu_read_unlock(); | |
2000 | ||
98e399f8 | 2001 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 2002 | |
f6a78bfc HX |
2003 | return segs; |
2004 | } | |
f6a78bfc HX |
2005 | EXPORT_SYMBOL(skb_gso_segment); |
2006 | ||
fb286bb2 HX |
2007 | /* Take action when hardware reception checksum errors are detected. */ |
2008 | #ifdef CONFIG_BUG | |
2009 | void netdev_rx_csum_fault(struct net_device *dev) | |
2010 | { | |
2011 | if (net_ratelimit()) { | |
7b6cd1ce | 2012 | pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); |
fb286bb2 HX |
2013 | dump_stack(); |
2014 | } | |
2015 | } | |
2016 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
2017 | #endif | |
2018 | ||
1da177e4 LT |
2019 | /* Actually, we should eliminate this check as soon as we know, that: |
2020 | * 1. IOMMU is present and allows to map all the memory. | |
2021 | * 2. No high memory really exists on this machine. | |
2022 | */ | |
2023 | ||
9092c658 | 2024 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
1da177e4 | 2025 | { |
3d3a8533 | 2026 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 2027 | int i; |
5acbbd42 | 2028 | if (!(dev->features & NETIF_F_HIGHDMA)) { |
ea2ab693 IC |
2029 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
2030 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
2031 | if (PageHighMem(skb_frag_page(frag))) | |
5acbbd42 | 2032 | return 1; |
ea2ab693 | 2033 | } |
5acbbd42 | 2034 | } |
1da177e4 | 2035 | |
5acbbd42 FT |
2036 | if (PCI_DMA_BUS_IS_PHYS) { |
2037 | struct device *pdev = dev->dev.parent; | |
1da177e4 | 2038 | |
9092c658 ED |
2039 | if (!pdev) |
2040 | return 0; | |
5acbbd42 | 2041 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
ea2ab693 IC |
2042 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
2043 | dma_addr_t addr = page_to_phys(skb_frag_page(frag)); | |
5acbbd42 FT |
2044 | if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask) |
2045 | return 1; | |
2046 | } | |
2047 | } | |
3d3a8533 | 2048 | #endif |
1da177e4 LT |
2049 | return 0; |
2050 | } | |
1da177e4 | 2051 | |
f6a78bfc HX |
2052 | struct dev_gso_cb { |
2053 | void (*destructor)(struct sk_buff *skb); | |
2054 | }; | |
2055 | ||
2056 | #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb) | |
2057 | ||
2058 | static void dev_gso_skb_destructor(struct sk_buff *skb) | |
2059 | { | |
2060 | struct dev_gso_cb *cb; | |
2061 | ||
2062 | do { | |
2063 | struct sk_buff *nskb = skb->next; | |
2064 | ||
2065 | skb->next = nskb->next; | |
2066 | nskb->next = NULL; | |
2067 | kfree_skb(nskb); | |
2068 | } while (skb->next); | |
2069 | ||
2070 | cb = DEV_GSO_CB(skb); | |
2071 | if (cb->destructor) | |
2072 | cb->destructor(skb); | |
2073 | } | |
2074 | ||
2075 | /** | |
2076 | * dev_gso_segment - Perform emulated hardware segmentation on skb. | |
2077 | * @skb: buffer to segment | |
91ecb63c | 2078 | * @features: device features as applicable to this skb |
f6a78bfc HX |
2079 | * |
2080 | * This function segments the given skb and stores the list of segments | |
2081 | * in skb->next. | |
2082 | */ | |
c8f44aff | 2083 | static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features) |
f6a78bfc | 2084 | { |
f6a78bfc | 2085 | struct sk_buff *segs; |
576a30eb HX |
2086 | |
2087 | segs = skb_gso_segment(skb, features); | |
2088 | ||
2089 | /* Verifying header integrity only. */ | |
2090 | if (!segs) | |
2091 | return 0; | |
f6a78bfc | 2092 | |
801678c5 | 2093 | if (IS_ERR(segs)) |
f6a78bfc HX |
2094 | return PTR_ERR(segs); |
2095 | ||
2096 | skb->next = segs; | |
2097 | DEV_GSO_CB(skb)->destructor = skb->destructor; | |
2098 | skb->destructor = dev_gso_skb_destructor; | |
2099 | ||
2100 | return 0; | |
2101 | } | |
2102 | ||
c8f44aff | 2103 | static bool can_checksum_protocol(netdev_features_t features, __be16 protocol) |
03634668 JG |
2104 | { |
2105 | return ((features & NETIF_F_GEN_CSUM) || | |
2106 | ((features & NETIF_F_V4_CSUM) && | |
2107 | protocol == htons(ETH_P_IP)) || | |
2108 | ((features & NETIF_F_V6_CSUM) && | |
2109 | protocol == htons(ETH_P_IPV6)) || | |
2110 | ((features & NETIF_F_FCOE_CRC) && | |
2111 | protocol == htons(ETH_P_FCOE))); | |
2112 | } | |
2113 | ||
c8f44aff MM |
2114 | static netdev_features_t harmonize_features(struct sk_buff *skb, |
2115 | __be16 protocol, netdev_features_t features) | |
f01a5236 | 2116 | { |
d402786e | 2117 | if (!can_checksum_protocol(features, protocol)) { |
f01a5236 JG |
2118 | features &= ~NETIF_F_ALL_CSUM; |
2119 | features &= ~NETIF_F_SG; | |
2120 | } else if (illegal_highdma(skb->dev, skb)) { | |
2121 | features &= ~NETIF_F_SG; | |
2122 | } | |
2123 | ||
2124 | return features; | |
2125 | } | |
2126 | ||
c8f44aff | 2127 | netdev_features_t netif_skb_features(struct sk_buff *skb) |
58e998c6 JG |
2128 | { |
2129 | __be16 protocol = skb->protocol; | |
c8f44aff | 2130 | netdev_features_t features = skb->dev->features; |
58e998c6 JG |
2131 | |
2132 | if (protocol == htons(ETH_P_8021Q)) { | |
2133 | struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data; | |
2134 | protocol = veh->h_vlan_encapsulated_proto; | |
f01a5236 JG |
2135 | } else if (!vlan_tx_tag_present(skb)) { |
2136 | return harmonize_features(skb, protocol, features); | |
2137 | } | |
58e998c6 | 2138 | |
6ee400aa | 2139 | features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX); |
f01a5236 JG |
2140 | |
2141 | if (protocol != htons(ETH_P_8021Q)) { | |
2142 | return harmonize_features(skb, protocol, features); | |
2143 | } else { | |
2144 | features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | | |
6ee400aa | 2145 | NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX; |
f01a5236 JG |
2146 | return harmonize_features(skb, protocol, features); |
2147 | } | |
58e998c6 | 2148 | } |
f01a5236 | 2149 | EXPORT_SYMBOL(netif_skb_features); |
58e998c6 | 2150 | |
6afff0ca JF |
2151 | /* |
2152 | * Returns true if either: | |
2153 | * 1. skb has frag_list and the device doesn't support FRAGLIST, or | |
2154 | * 2. skb is fragmented and the device does not support SG, or if | |
2155 | * at least one of fragments is in highmem and device does not | |
2156 | * support DMA from it. | |
2157 | */ | |
2158 | static inline int skb_needs_linearize(struct sk_buff *skb, | |
02932ce9 | 2159 | int features) |
6afff0ca | 2160 | { |
02932ce9 JG |
2161 | return skb_is_nonlinear(skb) && |
2162 | ((skb_has_frag_list(skb) && | |
2163 | !(features & NETIF_F_FRAGLIST)) || | |
e1e78db6 | 2164 | (skb_shinfo(skb)->nr_frags && |
02932ce9 | 2165 | !(features & NETIF_F_SG))); |
6afff0ca JF |
2166 | } |
2167 | ||
fd2ea0a7 DM |
2168 | int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev, |
2169 | struct netdev_queue *txq) | |
f6a78bfc | 2170 | { |
00829823 | 2171 | const struct net_device_ops *ops = dev->netdev_ops; |
572a9d7b | 2172 | int rc = NETDEV_TX_OK; |
ec764bf0 | 2173 | unsigned int skb_len; |
00829823 | 2174 | |
f6a78bfc | 2175 | if (likely(!skb->next)) { |
c8f44aff | 2176 | netdev_features_t features; |
fc741216 | 2177 | |
93f154b5 | 2178 | /* |
25985edc | 2179 | * If device doesn't need skb->dst, release it right now while |
93f154b5 ED |
2180 | * its hot in this cpu cache |
2181 | */ | |
adf30907 ED |
2182 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) |
2183 | skb_dst_drop(skb); | |
2184 | ||
15c2d75f ED |
2185 | if (!list_empty(&ptype_all)) |
2186 | dev_queue_xmit_nit(skb, dev); | |
2187 | ||
fc741216 JG |
2188 | features = netif_skb_features(skb); |
2189 | ||
7b9c6090 | 2190 | if (vlan_tx_tag_present(skb) && |
fc741216 | 2191 | !(features & NETIF_F_HW_VLAN_TX)) { |
7b9c6090 JG |
2192 | skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb)); |
2193 | if (unlikely(!skb)) | |
2194 | goto out; | |
2195 | ||
2196 | skb->vlan_tci = 0; | |
2197 | } | |
2198 | ||
fc741216 | 2199 | if (netif_needs_gso(skb, features)) { |
91ecb63c | 2200 | if (unlikely(dev_gso_segment(skb, features))) |
9ccb8975 DM |
2201 | goto out_kfree_skb; |
2202 | if (skb->next) | |
2203 | goto gso; | |
6afff0ca | 2204 | } else { |
02932ce9 | 2205 | if (skb_needs_linearize(skb, features) && |
6afff0ca JF |
2206 | __skb_linearize(skb)) |
2207 | goto out_kfree_skb; | |
2208 | ||
2209 | /* If packet is not checksummed and device does not | |
2210 | * support checksumming for this protocol, complete | |
2211 | * checksumming here. | |
2212 | */ | |
2213 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
55508d60 MM |
2214 | skb_set_transport_header(skb, |
2215 | skb_checksum_start_offset(skb)); | |
03634668 | 2216 | if (!(features & NETIF_F_ALL_CSUM) && |
6afff0ca JF |
2217 | skb_checksum_help(skb)) |
2218 | goto out_kfree_skb; | |
2219 | } | |
9ccb8975 DM |
2220 | } |
2221 | ||
ec764bf0 | 2222 | skb_len = skb->len; |
ac45f602 | 2223 | rc = ops->ndo_start_xmit(skb, dev); |
ec764bf0 | 2224 | trace_net_dev_xmit(skb, rc, dev, skb_len); |
ec634fe3 | 2225 | if (rc == NETDEV_TX_OK) |
08baf561 | 2226 | txq_trans_update(txq); |
ac45f602 | 2227 | return rc; |
f6a78bfc HX |
2228 | } |
2229 | ||
576a30eb | 2230 | gso: |
f6a78bfc HX |
2231 | do { |
2232 | struct sk_buff *nskb = skb->next; | |
f6a78bfc HX |
2233 | |
2234 | skb->next = nskb->next; | |
2235 | nskb->next = NULL; | |
068a2de5 KK |
2236 | |
2237 | /* | |
25985edc | 2238 | * If device doesn't need nskb->dst, release it right now while |
068a2de5 KK |
2239 | * its hot in this cpu cache |
2240 | */ | |
2241 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
2242 | skb_dst_drop(nskb); | |
2243 | ||
ec764bf0 | 2244 | skb_len = nskb->len; |
00829823 | 2245 | rc = ops->ndo_start_xmit(nskb, dev); |
ec764bf0 | 2246 | trace_net_dev_xmit(nskb, rc, dev, skb_len); |
ec634fe3 | 2247 | if (unlikely(rc != NETDEV_TX_OK)) { |
572a9d7b PM |
2248 | if (rc & ~NETDEV_TX_MASK) |
2249 | goto out_kfree_gso_skb; | |
f54d9e8d | 2250 | nskb->next = skb->next; |
f6a78bfc HX |
2251 | skb->next = nskb; |
2252 | return rc; | |
2253 | } | |
08baf561 | 2254 | txq_trans_update(txq); |
73466498 | 2255 | if (unlikely(netif_xmit_stopped(txq) && skb->next)) |
f54d9e8d | 2256 | return NETDEV_TX_BUSY; |
f6a78bfc | 2257 | } while (skb->next); |
4ec93edb | 2258 | |
572a9d7b PM |
2259 | out_kfree_gso_skb: |
2260 | if (likely(skb->next == NULL)) | |
2261 | skb->destructor = DEV_GSO_CB(skb)->destructor; | |
f6a78bfc HX |
2262 | out_kfree_skb: |
2263 | kfree_skb(skb); | |
7b9c6090 | 2264 | out: |
572a9d7b | 2265 | return rc; |
f6a78bfc HX |
2266 | } |
2267 | ||
0a9627f2 | 2268 | static u32 hashrnd __read_mostly; |
b6b2fed1 | 2269 | |
a3d22a68 VZ |
2270 | /* |
2271 | * Returns a Tx hash based on the given packet descriptor a Tx queues' number | |
2272 | * to be used as a distribution range. | |
2273 | */ | |
2274 | u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb, | |
2275 | unsigned int num_tx_queues) | |
8f0f2223 | 2276 | { |
7019298a | 2277 | u32 hash; |
4f57c087 JF |
2278 | u16 qoffset = 0; |
2279 | u16 qcount = num_tx_queues; | |
b6b2fed1 | 2280 | |
513de11b DM |
2281 | if (skb_rx_queue_recorded(skb)) { |
2282 | hash = skb_get_rx_queue(skb); | |
a3d22a68 VZ |
2283 | while (unlikely(hash >= num_tx_queues)) |
2284 | hash -= num_tx_queues; | |
513de11b DM |
2285 | return hash; |
2286 | } | |
ec581f6a | 2287 | |
4f57c087 JF |
2288 | if (dev->num_tc) { |
2289 | u8 tc = netdev_get_prio_tc_map(dev, skb->priority); | |
2290 | qoffset = dev->tc_to_txq[tc].offset; | |
2291 | qcount = dev->tc_to_txq[tc].count; | |
2292 | } | |
2293 | ||
ec581f6a | 2294 | if (skb->sk && skb->sk->sk_hash) |
7019298a | 2295 | hash = skb->sk->sk_hash; |
ec581f6a | 2296 | else |
62b1a8ab | 2297 | hash = (__force u16) skb->protocol; |
0a9627f2 | 2298 | hash = jhash_1word(hash, hashrnd); |
b6b2fed1 | 2299 | |
4f57c087 | 2300 | return (u16) (((u64) hash * qcount) >> 32) + qoffset; |
8f0f2223 | 2301 | } |
a3d22a68 | 2302 | EXPORT_SYMBOL(__skb_tx_hash); |
8f0f2223 | 2303 | |
ed04642f ED |
2304 | static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index) |
2305 | { | |
2306 | if (unlikely(queue_index >= dev->real_num_tx_queues)) { | |
e87cc472 JP |
2307 | net_warn_ratelimited("%s selects TX queue %d, but real number of TX queues is %d\n", |
2308 | dev->name, queue_index, | |
2309 | dev->real_num_tx_queues); | |
ed04642f ED |
2310 | return 0; |
2311 | } | |
2312 | return queue_index; | |
2313 | } | |
2314 | ||
1d24eb48 TH |
2315 | static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb) |
2316 | { | |
bf264145 | 2317 | #ifdef CONFIG_XPS |
1d24eb48 TH |
2318 | struct xps_dev_maps *dev_maps; |
2319 | struct xps_map *map; | |
2320 | int queue_index = -1; | |
2321 | ||
2322 | rcu_read_lock(); | |
2323 | dev_maps = rcu_dereference(dev->xps_maps); | |
2324 | if (dev_maps) { | |
2325 | map = rcu_dereference( | |
2326 | dev_maps->cpu_map[raw_smp_processor_id()]); | |
2327 | if (map) { | |
2328 | if (map->len == 1) | |
2329 | queue_index = map->queues[0]; | |
2330 | else { | |
2331 | u32 hash; | |
2332 | if (skb->sk && skb->sk->sk_hash) | |
2333 | hash = skb->sk->sk_hash; | |
2334 | else | |
2335 | hash = (__force u16) skb->protocol ^ | |
2336 | skb->rxhash; | |
2337 | hash = jhash_1word(hash, hashrnd); | |
2338 | queue_index = map->queues[ | |
2339 | ((u64)hash * map->len) >> 32]; | |
2340 | } | |
2341 | if (unlikely(queue_index >= dev->real_num_tx_queues)) | |
2342 | queue_index = -1; | |
2343 | } | |
2344 | } | |
2345 | rcu_read_unlock(); | |
2346 | ||
2347 | return queue_index; | |
2348 | #else | |
2349 | return -1; | |
2350 | #endif | |
2351 | } | |
2352 | ||
e8a0464c DM |
2353 | static struct netdev_queue *dev_pick_tx(struct net_device *dev, |
2354 | struct sk_buff *skb) | |
2355 | { | |
b0f77d0e | 2356 | int queue_index; |
deabc772 | 2357 | const struct net_device_ops *ops = dev->netdev_ops; |
a4ee3ce3 | 2358 | |
3853b584 TH |
2359 | if (dev->real_num_tx_queues == 1) |
2360 | queue_index = 0; | |
2361 | else if (ops->ndo_select_queue) { | |
deabc772 HS |
2362 | queue_index = ops->ndo_select_queue(dev, skb); |
2363 | queue_index = dev_cap_txqueue(dev, queue_index); | |
2364 | } else { | |
2365 | struct sock *sk = skb->sk; | |
2366 | queue_index = sk_tx_queue_get(sk); | |
a4ee3ce3 | 2367 | |
3853b584 TH |
2368 | if (queue_index < 0 || skb->ooo_okay || |
2369 | queue_index >= dev->real_num_tx_queues) { | |
2370 | int old_index = queue_index; | |
fd2ea0a7 | 2371 | |
1d24eb48 TH |
2372 | queue_index = get_xps_queue(dev, skb); |
2373 | if (queue_index < 0) | |
2374 | queue_index = skb_tx_hash(dev, skb); | |
3853b584 TH |
2375 | |
2376 | if (queue_index != old_index && sk) { | |
2377 | struct dst_entry *dst = | |
2378 | rcu_dereference_check(sk->sk_dst_cache, 1); | |
8728c544 ED |
2379 | |
2380 | if (dst && skb_dst(skb) == dst) | |
2381 | sk_tx_queue_set(sk, queue_index); | |
2382 | } | |
a4ee3ce3 KK |
2383 | } |
2384 | } | |
eae792b7 | 2385 | |
fd2ea0a7 DM |
2386 | skb_set_queue_mapping(skb, queue_index); |
2387 | return netdev_get_tx_queue(dev, queue_index); | |
e8a0464c DM |
2388 | } |
2389 | ||
bbd8a0d3 KK |
2390 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
2391 | struct net_device *dev, | |
2392 | struct netdev_queue *txq) | |
2393 | { | |
2394 | spinlock_t *root_lock = qdisc_lock(q); | |
a2da570d | 2395 | bool contended; |
bbd8a0d3 KK |
2396 | int rc; |
2397 | ||
a2da570d ED |
2398 | qdisc_skb_cb(skb)->pkt_len = skb->len; |
2399 | qdisc_calculate_pkt_len(skb, q); | |
79640a4c ED |
2400 | /* |
2401 | * Heuristic to force contended enqueues to serialize on a | |
2402 | * separate lock before trying to get qdisc main lock. | |
2403 | * This permits __QDISC_STATE_RUNNING owner to get the lock more often | |
2404 | * and dequeue packets faster. | |
2405 | */ | |
a2da570d | 2406 | contended = qdisc_is_running(q); |
79640a4c ED |
2407 | if (unlikely(contended)) |
2408 | spin_lock(&q->busylock); | |
2409 | ||
bbd8a0d3 KK |
2410 | spin_lock(root_lock); |
2411 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
2412 | kfree_skb(skb); | |
2413 | rc = NET_XMIT_DROP; | |
2414 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
bc135b23 | 2415 | qdisc_run_begin(q)) { |
bbd8a0d3 KK |
2416 | /* |
2417 | * This is a work-conserving queue; there are no old skbs | |
2418 | * waiting to be sent out; and the qdisc is not running - | |
2419 | * xmit the skb directly. | |
2420 | */ | |
7fee226a ED |
2421 | if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE)) |
2422 | skb_dst_force(skb); | |
bfe0d029 | 2423 | |
bfe0d029 ED |
2424 | qdisc_bstats_update(q, skb); |
2425 | ||
79640a4c ED |
2426 | if (sch_direct_xmit(skb, q, dev, txq, root_lock)) { |
2427 | if (unlikely(contended)) { | |
2428 | spin_unlock(&q->busylock); | |
2429 | contended = false; | |
2430 | } | |
bbd8a0d3 | 2431 | __qdisc_run(q); |
79640a4c | 2432 | } else |
bc135b23 | 2433 | qdisc_run_end(q); |
bbd8a0d3 KK |
2434 | |
2435 | rc = NET_XMIT_SUCCESS; | |
2436 | } else { | |
7fee226a | 2437 | skb_dst_force(skb); |
a2da570d | 2438 | rc = q->enqueue(skb, q) & NET_XMIT_MASK; |
79640a4c ED |
2439 | if (qdisc_run_begin(q)) { |
2440 | if (unlikely(contended)) { | |
2441 | spin_unlock(&q->busylock); | |
2442 | contended = false; | |
2443 | } | |
2444 | __qdisc_run(q); | |
2445 | } | |
bbd8a0d3 KK |
2446 | } |
2447 | spin_unlock(root_lock); | |
79640a4c ED |
2448 | if (unlikely(contended)) |
2449 | spin_unlock(&q->busylock); | |
bbd8a0d3 KK |
2450 | return rc; |
2451 | } | |
2452 | ||
5bc1421e NH |
2453 | #if IS_ENABLED(CONFIG_NETPRIO_CGROUP) |
2454 | static void skb_update_prio(struct sk_buff *skb) | |
2455 | { | |
6977a79d | 2456 | struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap); |
5bc1421e NH |
2457 | |
2458 | if ((!skb->priority) && (skb->sk) && map) | |
2459 | skb->priority = map->priomap[skb->sk->sk_cgrp_prioidx]; | |
2460 | } | |
2461 | #else | |
2462 | #define skb_update_prio(skb) | |
2463 | #endif | |
2464 | ||
745e20f1 | 2465 | static DEFINE_PER_CPU(int, xmit_recursion); |
11a766ce | 2466 | #define RECURSION_LIMIT 10 |
745e20f1 | 2467 | |
95603e22 MM |
2468 | /** |
2469 | * dev_loopback_xmit - loop back @skb | |
2470 | * @skb: buffer to transmit | |
2471 | */ | |
2472 | int dev_loopback_xmit(struct sk_buff *skb) | |
2473 | { | |
2474 | skb_reset_mac_header(skb); | |
2475 | __skb_pull(skb, skb_network_offset(skb)); | |
2476 | skb->pkt_type = PACKET_LOOPBACK; | |
2477 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
2478 | WARN_ON(!skb_dst(skb)); | |
2479 | skb_dst_force(skb); | |
2480 | netif_rx_ni(skb); | |
2481 | return 0; | |
2482 | } | |
2483 | EXPORT_SYMBOL(dev_loopback_xmit); | |
2484 | ||
d29f749e DJ |
2485 | /** |
2486 | * dev_queue_xmit - transmit a buffer | |
2487 | * @skb: buffer to transmit | |
2488 | * | |
2489 | * Queue a buffer for transmission to a network device. The caller must | |
2490 | * have set the device and priority and built the buffer before calling | |
2491 | * this function. The function can be called from an interrupt. | |
2492 | * | |
2493 | * A negative errno code is returned on a failure. A success does not | |
2494 | * guarantee the frame will be transmitted as it may be dropped due | |
2495 | * to congestion or traffic shaping. | |
2496 | * | |
2497 | * ----------------------------------------------------------------------------------- | |
2498 | * I notice this method can also return errors from the queue disciplines, | |
2499 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
2500 | * be positive. | |
2501 | * | |
2502 | * Regardless of the return value, the skb is consumed, so it is currently | |
2503 | * difficult to retry a send to this method. (You can bump the ref count | |
2504 | * before sending to hold a reference for retry if you are careful.) | |
2505 | * | |
2506 | * When calling this method, interrupts MUST be enabled. This is because | |
2507 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
2508 | * --BLG | |
2509 | */ | |
1da177e4 LT |
2510 | int dev_queue_xmit(struct sk_buff *skb) |
2511 | { | |
2512 | struct net_device *dev = skb->dev; | |
dc2b4847 | 2513 | struct netdev_queue *txq; |
1da177e4 LT |
2514 | struct Qdisc *q; |
2515 | int rc = -ENOMEM; | |
2516 | ||
4ec93edb YH |
2517 | /* Disable soft irqs for various locks below. Also |
2518 | * stops preemption for RCU. | |
1da177e4 | 2519 | */ |
4ec93edb | 2520 | rcu_read_lock_bh(); |
1da177e4 | 2521 | |
5bc1421e NH |
2522 | skb_update_prio(skb); |
2523 | ||
eae792b7 | 2524 | txq = dev_pick_tx(dev, skb); |
a898def2 | 2525 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 2526 | |
1da177e4 | 2527 | #ifdef CONFIG_NET_CLS_ACT |
d1b19dff | 2528 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS); |
1da177e4 | 2529 | #endif |
cf66ba58 | 2530 | trace_net_dev_queue(skb); |
1da177e4 | 2531 | if (q->enqueue) { |
bbd8a0d3 | 2532 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 2533 | goto out; |
1da177e4 LT |
2534 | } |
2535 | ||
2536 | /* The device has no queue. Common case for software devices: | |
2537 | loopback, all the sorts of tunnels... | |
2538 | ||
932ff279 HX |
2539 | Really, it is unlikely that netif_tx_lock protection is necessary |
2540 | here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
1da177e4 LT |
2541 | counters.) |
2542 | However, it is possible, that they rely on protection | |
2543 | made by us here. | |
2544 | ||
2545 | Check this and shot the lock. It is not prone from deadlocks. | |
2546 | Either shot noqueue qdisc, it is even simpler 8) | |
2547 | */ | |
2548 | if (dev->flags & IFF_UP) { | |
2549 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
2550 | ||
c773e847 | 2551 | if (txq->xmit_lock_owner != cpu) { |
1da177e4 | 2552 | |
745e20f1 ED |
2553 | if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT) |
2554 | goto recursion_alert; | |
2555 | ||
c773e847 | 2556 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 2557 | |
73466498 | 2558 | if (!netif_xmit_stopped(txq)) { |
745e20f1 | 2559 | __this_cpu_inc(xmit_recursion); |
572a9d7b | 2560 | rc = dev_hard_start_xmit(skb, dev, txq); |
745e20f1 | 2561 | __this_cpu_dec(xmit_recursion); |
572a9d7b | 2562 | if (dev_xmit_complete(rc)) { |
c773e847 | 2563 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
2564 | goto out; |
2565 | } | |
2566 | } | |
c773e847 | 2567 | HARD_TX_UNLOCK(dev, txq); |
e87cc472 JP |
2568 | net_crit_ratelimited("Virtual device %s asks to queue packet!\n", |
2569 | dev->name); | |
1da177e4 LT |
2570 | } else { |
2571 | /* Recursion is detected! It is possible, | |
745e20f1 ED |
2572 | * unfortunately |
2573 | */ | |
2574 | recursion_alert: | |
e87cc472 JP |
2575 | net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", |
2576 | dev->name); | |
1da177e4 LT |
2577 | } |
2578 | } | |
2579 | ||
2580 | rc = -ENETDOWN; | |
d4828d85 | 2581 | rcu_read_unlock_bh(); |
1da177e4 | 2582 | |
1da177e4 LT |
2583 | kfree_skb(skb); |
2584 | return rc; | |
2585 | out: | |
d4828d85 | 2586 | rcu_read_unlock_bh(); |
1da177e4 LT |
2587 | return rc; |
2588 | } | |
d1b19dff | 2589 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 LT |
2590 | |
2591 | ||
2592 | /*======================================================================= | |
2593 | Receiver routines | |
2594 | =======================================================================*/ | |
2595 | ||
6b2bedc3 | 2596 | int netdev_max_backlog __read_mostly = 1000; |
3b098e2d | 2597 | int netdev_tstamp_prequeue __read_mostly = 1; |
6b2bedc3 SH |
2598 | int netdev_budget __read_mostly = 300; |
2599 | int weight_p __read_mostly = 64; /* old backlog weight */ | |
1da177e4 | 2600 | |
eecfd7c4 ED |
2601 | /* Called with irq disabled */ |
2602 | static inline void ____napi_schedule(struct softnet_data *sd, | |
2603 | struct napi_struct *napi) | |
2604 | { | |
2605 | list_add_tail(&napi->poll_list, &sd->poll_list); | |
2606 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2607 | } | |
2608 | ||
0a9627f2 | 2609 | /* |
bfb564e7 | 2610 | * __skb_get_rxhash: calculate a flow hash based on src/dst addresses |
bdeab991 TH |
2611 | * and src/dst port numbers. Sets rxhash in skb to non-zero hash value |
2612 | * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb | |
2613 | * if hash is a canonical 4-tuple hash over transport ports. | |
0a9627f2 | 2614 | */ |
bdeab991 | 2615 | void __skb_get_rxhash(struct sk_buff *skb) |
0a9627f2 | 2616 | { |
4504b861 ED |
2617 | struct flow_keys keys; |
2618 | u32 hash; | |
c6865cb3 | 2619 | |
4504b861 ED |
2620 | if (!skb_flow_dissect(skb, &keys)) |
2621 | return; | |
e971b722 | 2622 | |
4504b861 ED |
2623 | if (keys.ports) { |
2624 | if ((__force u16)keys.port16[1] < (__force u16)keys.port16[0]) | |
2625 | swap(keys.port16[0], keys.port16[1]); | |
2626 | skb->l4_rxhash = 1; | |
0a9627f2 TH |
2627 | } |
2628 | ||
b249dcb8 | 2629 | /* get a consistent hash (same value on both flow directions) */ |
4504b861 ED |
2630 | if ((__force u32)keys.dst < (__force u32)keys.src) |
2631 | swap(keys.dst, keys.src); | |
0a9627f2 | 2632 | |
4504b861 ED |
2633 | hash = jhash_3words((__force u32)keys.dst, |
2634 | (__force u32)keys.src, | |
2635 | (__force u32)keys.ports, hashrnd); | |
bfb564e7 KK |
2636 | if (!hash) |
2637 | hash = 1; | |
2638 | ||
bdeab991 | 2639 | skb->rxhash = hash; |
bfb564e7 KK |
2640 | } |
2641 | EXPORT_SYMBOL(__skb_get_rxhash); | |
2642 | ||
2643 | #ifdef CONFIG_RPS | |
2644 | ||
2645 | /* One global table that all flow-based protocols share. */ | |
6e3f7faf | 2646 | struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; |
bfb564e7 KK |
2647 | EXPORT_SYMBOL(rps_sock_flow_table); |
2648 | ||
c5905afb | 2649 | struct static_key rps_needed __read_mostly; |
adc9300e | 2650 | |
c445477d BH |
2651 | static struct rps_dev_flow * |
2652 | set_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
2653 | struct rps_dev_flow *rflow, u16 next_cpu) | |
2654 | { | |
09994d1b | 2655 | if (next_cpu != RPS_NO_CPU) { |
c445477d BH |
2656 | #ifdef CONFIG_RFS_ACCEL |
2657 | struct netdev_rx_queue *rxqueue; | |
2658 | struct rps_dev_flow_table *flow_table; | |
2659 | struct rps_dev_flow *old_rflow; | |
2660 | u32 flow_id; | |
2661 | u16 rxq_index; | |
2662 | int rc; | |
2663 | ||
2664 | /* Should we steer this flow to a different hardware queue? */ | |
69a19ee6 BH |
2665 | if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || |
2666 | !(dev->features & NETIF_F_NTUPLE)) | |
c445477d BH |
2667 | goto out; |
2668 | rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); | |
2669 | if (rxq_index == skb_get_rx_queue(skb)) | |
2670 | goto out; | |
2671 | ||
2672 | rxqueue = dev->_rx + rxq_index; | |
2673 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
2674 | if (!flow_table) | |
2675 | goto out; | |
2676 | flow_id = skb->rxhash & flow_table->mask; | |
2677 | rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, | |
2678 | rxq_index, flow_id); | |
2679 | if (rc < 0) | |
2680 | goto out; | |
2681 | old_rflow = rflow; | |
2682 | rflow = &flow_table->flows[flow_id]; | |
c445477d BH |
2683 | rflow->filter = rc; |
2684 | if (old_rflow->filter == rflow->filter) | |
2685 | old_rflow->filter = RPS_NO_FILTER; | |
2686 | out: | |
2687 | #endif | |
2688 | rflow->last_qtail = | |
09994d1b | 2689 | per_cpu(softnet_data, next_cpu).input_queue_head; |
c445477d BH |
2690 | } |
2691 | ||
09994d1b | 2692 | rflow->cpu = next_cpu; |
c445477d BH |
2693 | return rflow; |
2694 | } | |
2695 | ||
bfb564e7 KK |
2696 | /* |
2697 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
2698 | * CPU from the RPS map of the receiving queue for a given skb. | |
2699 | * rcu_read_lock must be held on entry. | |
2700 | */ | |
2701 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
2702 | struct rps_dev_flow **rflowp) | |
2703 | { | |
2704 | struct netdev_rx_queue *rxqueue; | |
6e3f7faf | 2705 | struct rps_map *map; |
bfb564e7 KK |
2706 | struct rps_dev_flow_table *flow_table; |
2707 | struct rps_sock_flow_table *sock_flow_table; | |
2708 | int cpu = -1; | |
2709 | u16 tcpu; | |
2710 | ||
2711 | if (skb_rx_queue_recorded(skb)) { | |
2712 | u16 index = skb_get_rx_queue(skb); | |
62fe0b40 BH |
2713 | if (unlikely(index >= dev->real_num_rx_queues)) { |
2714 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
2715 | "%s received packet on queue %u, but number " | |
2716 | "of RX queues is %u\n", | |
2717 | dev->name, index, dev->real_num_rx_queues); | |
bfb564e7 KK |
2718 | goto done; |
2719 | } | |
2720 | rxqueue = dev->_rx + index; | |
2721 | } else | |
2722 | rxqueue = dev->_rx; | |
2723 | ||
6e3f7faf ED |
2724 | map = rcu_dereference(rxqueue->rps_map); |
2725 | if (map) { | |
85875236 | 2726 | if (map->len == 1 && |
33d480ce | 2727 | !rcu_access_pointer(rxqueue->rps_flow_table)) { |
6febfca9 CG |
2728 | tcpu = map->cpus[0]; |
2729 | if (cpu_online(tcpu)) | |
2730 | cpu = tcpu; | |
2731 | goto done; | |
2732 | } | |
33d480ce | 2733 | } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) { |
bfb564e7 | 2734 | goto done; |
6febfca9 | 2735 | } |
bfb564e7 | 2736 | |
2d47b459 | 2737 | skb_reset_network_header(skb); |
bfb564e7 KK |
2738 | if (!skb_get_rxhash(skb)) |
2739 | goto done; | |
2740 | ||
fec5e652 TH |
2741 | flow_table = rcu_dereference(rxqueue->rps_flow_table); |
2742 | sock_flow_table = rcu_dereference(rps_sock_flow_table); | |
2743 | if (flow_table && sock_flow_table) { | |
2744 | u16 next_cpu; | |
2745 | struct rps_dev_flow *rflow; | |
2746 | ||
2747 | rflow = &flow_table->flows[skb->rxhash & flow_table->mask]; | |
2748 | tcpu = rflow->cpu; | |
2749 | ||
2750 | next_cpu = sock_flow_table->ents[skb->rxhash & | |
2751 | sock_flow_table->mask]; | |
2752 | ||
2753 | /* | |
2754 | * If the desired CPU (where last recvmsg was done) is | |
2755 | * different from current CPU (one in the rx-queue flow | |
2756 | * table entry), switch if one of the following holds: | |
2757 | * - Current CPU is unset (equal to RPS_NO_CPU). | |
2758 | * - Current CPU is offline. | |
2759 | * - The current CPU's queue tail has advanced beyond the | |
2760 | * last packet that was enqueued using this table entry. | |
2761 | * This guarantees that all previous packets for the flow | |
2762 | * have been dequeued, thus preserving in order delivery. | |
2763 | */ | |
2764 | if (unlikely(tcpu != next_cpu) && | |
2765 | (tcpu == RPS_NO_CPU || !cpu_online(tcpu) || | |
2766 | ((int)(per_cpu(softnet_data, tcpu).input_queue_head - | |
c445477d BH |
2767 | rflow->last_qtail)) >= 0)) |
2768 | rflow = set_rps_cpu(dev, skb, rflow, next_cpu); | |
2769 | ||
fec5e652 TH |
2770 | if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) { |
2771 | *rflowp = rflow; | |
2772 | cpu = tcpu; | |
2773 | goto done; | |
2774 | } | |
2775 | } | |
2776 | ||
0a9627f2 | 2777 | if (map) { |
fec5e652 | 2778 | tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32]; |
0a9627f2 TH |
2779 | |
2780 | if (cpu_online(tcpu)) { | |
2781 | cpu = tcpu; | |
2782 | goto done; | |
2783 | } | |
2784 | } | |
2785 | ||
2786 | done: | |
0a9627f2 TH |
2787 | return cpu; |
2788 | } | |
2789 | ||
c445477d BH |
2790 | #ifdef CONFIG_RFS_ACCEL |
2791 | ||
2792 | /** | |
2793 | * rps_may_expire_flow - check whether an RFS hardware filter may be removed | |
2794 | * @dev: Device on which the filter was set | |
2795 | * @rxq_index: RX queue index | |
2796 | * @flow_id: Flow ID passed to ndo_rx_flow_steer() | |
2797 | * @filter_id: Filter ID returned by ndo_rx_flow_steer() | |
2798 | * | |
2799 | * Drivers that implement ndo_rx_flow_steer() should periodically call | |
2800 | * this function for each installed filter and remove the filters for | |
2801 | * which it returns %true. | |
2802 | */ | |
2803 | bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, | |
2804 | u32 flow_id, u16 filter_id) | |
2805 | { | |
2806 | struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; | |
2807 | struct rps_dev_flow_table *flow_table; | |
2808 | struct rps_dev_flow *rflow; | |
2809 | bool expire = true; | |
2810 | int cpu; | |
2811 | ||
2812 | rcu_read_lock(); | |
2813 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
2814 | if (flow_table && flow_id <= flow_table->mask) { | |
2815 | rflow = &flow_table->flows[flow_id]; | |
2816 | cpu = ACCESS_ONCE(rflow->cpu); | |
2817 | if (rflow->filter == filter_id && cpu != RPS_NO_CPU && | |
2818 | ((int)(per_cpu(softnet_data, cpu).input_queue_head - | |
2819 | rflow->last_qtail) < | |
2820 | (int)(10 * flow_table->mask))) | |
2821 | expire = false; | |
2822 | } | |
2823 | rcu_read_unlock(); | |
2824 | return expire; | |
2825 | } | |
2826 | EXPORT_SYMBOL(rps_may_expire_flow); | |
2827 | ||
2828 | #endif /* CONFIG_RFS_ACCEL */ | |
2829 | ||
0a9627f2 | 2830 | /* Called from hardirq (IPI) context */ |
e36fa2f7 | 2831 | static void rps_trigger_softirq(void *data) |
0a9627f2 | 2832 | { |
e36fa2f7 ED |
2833 | struct softnet_data *sd = data; |
2834 | ||
eecfd7c4 | 2835 | ____napi_schedule(sd, &sd->backlog); |
dee42870 | 2836 | sd->received_rps++; |
0a9627f2 | 2837 | } |
e36fa2f7 | 2838 | |
fec5e652 | 2839 | #endif /* CONFIG_RPS */ |
0a9627f2 | 2840 | |
e36fa2f7 ED |
2841 | /* |
2842 | * Check if this softnet_data structure is another cpu one | |
2843 | * If yes, queue it to our IPI list and return 1 | |
2844 | * If no, return 0 | |
2845 | */ | |
2846 | static int rps_ipi_queued(struct softnet_data *sd) | |
2847 | { | |
2848 | #ifdef CONFIG_RPS | |
2849 | struct softnet_data *mysd = &__get_cpu_var(softnet_data); | |
2850 | ||
2851 | if (sd != mysd) { | |
2852 | sd->rps_ipi_next = mysd->rps_ipi_list; | |
2853 | mysd->rps_ipi_list = sd; | |
2854 | ||
2855 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2856 | return 1; | |
2857 | } | |
2858 | #endif /* CONFIG_RPS */ | |
2859 | return 0; | |
2860 | } | |
2861 | ||
0a9627f2 TH |
2862 | /* |
2863 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
2864 | * queue (may be a remote CPU queue). | |
2865 | */ | |
fec5e652 TH |
2866 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, |
2867 | unsigned int *qtail) | |
0a9627f2 | 2868 | { |
e36fa2f7 | 2869 | struct softnet_data *sd; |
0a9627f2 TH |
2870 | unsigned long flags; |
2871 | ||
e36fa2f7 | 2872 | sd = &per_cpu(softnet_data, cpu); |
0a9627f2 TH |
2873 | |
2874 | local_irq_save(flags); | |
0a9627f2 | 2875 | |
e36fa2f7 | 2876 | rps_lock(sd); |
6e7676c1 CG |
2877 | if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) { |
2878 | if (skb_queue_len(&sd->input_pkt_queue)) { | |
0a9627f2 | 2879 | enqueue: |
e36fa2f7 | 2880 | __skb_queue_tail(&sd->input_pkt_queue, skb); |
76cc8b13 | 2881 | input_queue_tail_incr_save(sd, qtail); |
e36fa2f7 | 2882 | rps_unlock(sd); |
152102c7 | 2883 | local_irq_restore(flags); |
0a9627f2 TH |
2884 | return NET_RX_SUCCESS; |
2885 | } | |
2886 | ||
ebda37c2 ED |
2887 | /* Schedule NAPI for backlog device |
2888 | * We can use non atomic operation since we own the queue lock | |
2889 | */ | |
2890 | if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { | |
e36fa2f7 | 2891 | if (!rps_ipi_queued(sd)) |
eecfd7c4 | 2892 | ____napi_schedule(sd, &sd->backlog); |
0a9627f2 TH |
2893 | } |
2894 | goto enqueue; | |
2895 | } | |
2896 | ||
dee42870 | 2897 | sd->dropped++; |
e36fa2f7 | 2898 | rps_unlock(sd); |
0a9627f2 | 2899 | |
0a9627f2 TH |
2900 | local_irq_restore(flags); |
2901 | ||
caf586e5 | 2902 | atomic_long_inc(&skb->dev->rx_dropped); |
0a9627f2 TH |
2903 | kfree_skb(skb); |
2904 | return NET_RX_DROP; | |
2905 | } | |
1da177e4 | 2906 | |
1da177e4 LT |
2907 | /** |
2908 | * netif_rx - post buffer to the network code | |
2909 | * @skb: buffer to post | |
2910 | * | |
2911 | * This function receives a packet from a device driver and queues it for | |
2912 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
2913 | * may be dropped during processing for congestion control or by the | |
2914 | * protocol layers. | |
2915 | * | |
2916 | * return values: | |
2917 | * NET_RX_SUCCESS (no congestion) | |
1da177e4 LT |
2918 | * NET_RX_DROP (packet was dropped) |
2919 | * | |
2920 | */ | |
2921 | ||
2922 | int netif_rx(struct sk_buff *skb) | |
2923 | { | |
b0e28f1e | 2924 | int ret; |
1da177e4 LT |
2925 | |
2926 | /* if netpoll wants it, pretend we never saw it */ | |
2927 | if (netpoll_rx(skb)) | |
2928 | return NET_RX_DROP; | |
2929 | ||
588f0330 | 2930 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
1da177e4 | 2931 | |
cf66ba58 | 2932 | trace_netif_rx(skb); |
df334545 | 2933 | #ifdef CONFIG_RPS |
c5905afb | 2934 | if (static_key_false(&rps_needed)) { |
fec5e652 | 2935 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
b0e28f1e ED |
2936 | int cpu; |
2937 | ||
cece1945 | 2938 | preempt_disable(); |
b0e28f1e | 2939 | rcu_read_lock(); |
fec5e652 TH |
2940 | |
2941 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
b0e28f1e ED |
2942 | if (cpu < 0) |
2943 | cpu = smp_processor_id(); | |
fec5e652 TH |
2944 | |
2945 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
2946 | ||
b0e28f1e | 2947 | rcu_read_unlock(); |
cece1945 | 2948 | preempt_enable(); |
adc9300e ED |
2949 | } else |
2950 | #endif | |
fec5e652 TH |
2951 | { |
2952 | unsigned int qtail; | |
2953 | ret = enqueue_to_backlog(skb, get_cpu(), &qtail); | |
2954 | put_cpu(); | |
2955 | } | |
b0e28f1e | 2956 | return ret; |
1da177e4 | 2957 | } |
d1b19dff | 2958 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
2959 | |
2960 | int netif_rx_ni(struct sk_buff *skb) | |
2961 | { | |
2962 | int err; | |
2963 | ||
2964 | preempt_disable(); | |
2965 | err = netif_rx(skb); | |
2966 | if (local_softirq_pending()) | |
2967 | do_softirq(); | |
2968 | preempt_enable(); | |
2969 | ||
2970 | return err; | |
2971 | } | |
1da177e4 LT |
2972 | EXPORT_SYMBOL(netif_rx_ni); |
2973 | ||
1da177e4 LT |
2974 | static void net_tx_action(struct softirq_action *h) |
2975 | { | |
2976 | struct softnet_data *sd = &__get_cpu_var(softnet_data); | |
2977 | ||
2978 | if (sd->completion_queue) { | |
2979 | struct sk_buff *clist; | |
2980 | ||
2981 | local_irq_disable(); | |
2982 | clist = sd->completion_queue; | |
2983 | sd->completion_queue = NULL; | |
2984 | local_irq_enable(); | |
2985 | ||
2986 | while (clist) { | |
2987 | struct sk_buff *skb = clist; | |
2988 | clist = clist->next; | |
2989 | ||
547b792c | 2990 | WARN_ON(atomic_read(&skb->users)); |
07dc22e7 | 2991 | trace_kfree_skb(skb, net_tx_action); |
1da177e4 LT |
2992 | __kfree_skb(skb); |
2993 | } | |
2994 | } | |
2995 | ||
2996 | if (sd->output_queue) { | |
37437bb2 | 2997 | struct Qdisc *head; |
1da177e4 LT |
2998 | |
2999 | local_irq_disable(); | |
3000 | head = sd->output_queue; | |
3001 | sd->output_queue = NULL; | |
a9cbd588 | 3002 | sd->output_queue_tailp = &sd->output_queue; |
1da177e4 LT |
3003 | local_irq_enable(); |
3004 | ||
3005 | while (head) { | |
37437bb2 DM |
3006 | struct Qdisc *q = head; |
3007 | spinlock_t *root_lock; | |
3008 | ||
1da177e4 LT |
3009 | head = head->next_sched; |
3010 | ||
5fb66229 | 3011 | root_lock = qdisc_lock(q); |
37437bb2 | 3012 | if (spin_trylock(root_lock)) { |
def82a1d JP |
3013 | smp_mb__before_clear_bit(); |
3014 | clear_bit(__QDISC_STATE_SCHED, | |
3015 | &q->state); | |
37437bb2 DM |
3016 | qdisc_run(q); |
3017 | spin_unlock(root_lock); | |
1da177e4 | 3018 | } else { |
195648bb | 3019 | if (!test_bit(__QDISC_STATE_DEACTIVATED, |
e8a83e10 | 3020 | &q->state)) { |
195648bb | 3021 | __netif_reschedule(q); |
e8a83e10 JP |
3022 | } else { |
3023 | smp_mb__before_clear_bit(); | |
3024 | clear_bit(__QDISC_STATE_SCHED, | |
3025 | &q->state); | |
3026 | } | |
1da177e4 LT |
3027 | } |
3028 | } | |
3029 | } | |
3030 | } | |
3031 | ||
ab95bfe0 JP |
3032 | #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \ |
3033 | (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)) | |
da678292 MM |
3034 | /* This hook is defined here for ATM LANE */ |
3035 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
3036 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 3037 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 3038 | #endif |
1da177e4 | 3039 | |
1da177e4 LT |
3040 | #ifdef CONFIG_NET_CLS_ACT |
3041 | /* TODO: Maybe we should just force sch_ingress to be compiled in | |
3042 | * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions | |
3043 | * a compare and 2 stores extra right now if we dont have it on | |
3044 | * but have CONFIG_NET_CLS_ACT | |
25985edc LDM |
3045 | * NOTE: This doesn't stop any functionality; if you dont have |
3046 | * the ingress scheduler, you just can't add policies on ingress. | |
1da177e4 LT |
3047 | * |
3048 | */ | |
24824a09 | 3049 | static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq) |
1da177e4 | 3050 | { |
1da177e4 | 3051 | struct net_device *dev = skb->dev; |
f697c3e8 | 3052 | u32 ttl = G_TC_RTTL(skb->tc_verd); |
555353cf DM |
3053 | int result = TC_ACT_OK; |
3054 | struct Qdisc *q; | |
4ec93edb | 3055 | |
de384830 | 3056 | if (unlikely(MAX_RED_LOOP < ttl++)) { |
e87cc472 JP |
3057 | net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n", |
3058 | skb->skb_iif, dev->ifindex); | |
f697c3e8 HX |
3059 | return TC_ACT_SHOT; |
3060 | } | |
1da177e4 | 3061 | |
f697c3e8 HX |
3062 | skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl); |
3063 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); | |
1da177e4 | 3064 | |
83874000 | 3065 | q = rxq->qdisc; |
8d50b53d | 3066 | if (q != &noop_qdisc) { |
83874000 | 3067 | spin_lock(qdisc_lock(q)); |
a9312ae8 DM |
3068 | if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) |
3069 | result = qdisc_enqueue_root(skb, q); | |
83874000 DM |
3070 | spin_unlock(qdisc_lock(q)); |
3071 | } | |
f697c3e8 HX |
3072 | |
3073 | return result; | |
3074 | } | |
86e65da9 | 3075 | |
f697c3e8 HX |
3076 | static inline struct sk_buff *handle_ing(struct sk_buff *skb, |
3077 | struct packet_type **pt_prev, | |
3078 | int *ret, struct net_device *orig_dev) | |
3079 | { | |
24824a09 ED |
3080 | struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue); |
3081 | ||
3082 | if (!rxq || rxq->qdisc == &noop_qdisc) | |
f697c3e8 | 3083 | goto out; |
1da177e4 | 3084 | |
f697c3e8 HX |
3085 | if (*pt_prev) { |
3086 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
3087 | *pt_prev = NULL; | |
1da177e4 LT |
3088 | } |
3089 | ||
24824a09 | 3090 | switch (ing_filter(skb, rxq)) { |
f697c3e8 HX |
3091 | case TC_ACT_SHOT: |
3092 | case TC_ACT_STOLEN: | |
3093 | kfree_skb(skb); | |
3094 | return NULL; | |
3095 | } | |
3096 | ||
3097 | out: | |
3098 | skb->tc_verd = 0; | |
3099 | return skb; | |
1da177e4 LT |
3100 | } |
3101 | #endif | |
3102 | ||
ab95bfe0 JP |
3103 | /** |
3104 | * netdev_rx_handler_register - register receive handler | |
3105 | * @dev: device to register a handler for | |
3106 | * @rx_handler: receive handler to register | |
93e2c32b | 3107 | * @rx_handler_data: data pointer that is used by rx handler |
ab95bfe0 JP |
3108 | * |
3109 | * Register a receive hander for a device. This handler will then be | |
3110 | * called from __netif_receive_skb. A negative errno code is returned | |
3111 | * on a failure. | |
3112 | * | |
3113 | * The caller must hold the rtnl_mutex. | |
8a4eb573 JP |
3114 | * |
3115 | * For a general description of rx_handler, see enum rx_handler_result. | |
ab95bfe0 JP |
3116 | */ |
3117 | int netdev_rx_handler_register(struct net_device *dev, | |
93e2c32b JP |
3118 | rx_handler_func_t *rx_handler, |
3119 | void *rx_handler_data) | |
ab95bfe0 JP |
3120 | { |
3121 | ASSERT_RTNL(); | |
3122 | ||
3123 | if (dev->rx_handler) | |
3124 | return -EBUSY; | |
3125 | ||
93e2c32b | 3126 | rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); |
ab95bfe0 JP |
3127 | rcu_assign_pointer(dev->rx_handler, rx_handler); |
3128 | ||
3129 | return 0; | |
3130 | } | |
3131 | EXPORT_SYMBOL_GPL(netdev_rx_handler_register); | |
3132 | ||
3133 | /** | |
3134 | * netdev_rx_handler_unregister - unregister receive handler | |
3135 | * @dev: device to unregister a handler from | |
3136 | * | |
3137 | * Unregister a receive hander from a device. | |
3138 | * | |
3139 | * The caller must hold the rtnl_mutex. | |
3140 | */ | |
3141 | void netdev_rx_handler_unregister(struct net_device *dev) | |
3142 | { | |
3143 | ||
3144 | ASSERT_RTNL(); | |
a9b3cd7f SH |
3145 | RCU_INIT_POINTER(dev->rx_handler, NULL); |
3146 | RCU_INIT_POINTER(dev->rx_handler_data, NULL); | |
ab95bfe0 JP |
3147 | } |
3148 | EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); | |
3149 | ||
10f744d2 | 3150 | static int __netif_receive_skb(struct sk_buff *skb) |
1da177e4 LT |
3151 | { |
3152 | struct packet_type *ptype, *pt_prev; | |
ab95bfe0 | 3153 | rx_handler_func_t *rx_handler; |
f2ccd8fa | 3154 | struct net_device *orig_dev; |
63d8ea7f | 3155 | struct net_device *null_or_dev; |
8a4eb573 | 3156 | bool deliver_exact = false; |
1da177e4 | 3157 | int ret = NET_RX_DROP; |
252e3346 | 3158 | __be16 type; |
1da177e4 | 3159 | |
588f0330 | 3160 | net_timestamp_check(!netdev_tstamp_prequeue, skb); |
81bbb3d4 | 3161 | |
cf66ba58 | 3162 | trace_netif_receive_skb(skb); |
9b22ea56 | 3163 | |
1da177e4 | 3164 | /* if we've gotten here through NAPI, check netpoll */ |
bea3348e | 3165 | if (netpoll_receive_skb(skb)) |
1da177e4 LT |
3166 | return NET_RX_DROP; |
3167 | ||
8964be4a ED |
3168 | if (!skb->skb_iif) |
3169 | skb->skb_iif = skb->dev->ifindex; | |
cc9bd5ce | 3170 | orig_dev = skb->dev; |
8f903c70 | 3171 | |
c1d2bbe1 | 3172 | skb_reset_network_header(skb); |
badff6d0 | 3173 | skb_reset_transport_header(skb); |
0b5c9db1 | 3174 | skb_reset_mac_len(skb); |
1da177e4 LT |
3175 | |
3176 | pt_prev = NULL; | |
3177 | ||
3178 | rcu_read_lock(); | |
3179 | ||
63d8ea7f DM |
3180 | another_round: |
3181 | ||
3182 | __this_cpu_inc(softnet_data.processed); | |
3183 | ||
bcc6d479 JP |
3184 | if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) { |
3185 | skb = vlan_untag(skb); | |
3186 | if (unlikely(!skb)) | |
3187 | goto out; | |
3188 | } | |
3189 | ||
1da177e4 LT |
3190 | #ifdef CONFIG_NET_CLS_ACT |
3191 | if (skb->tc_verd & TC_NCLS) { | |
3192 | skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); | |
3193 | goto ncls; | |
3194 | } | |
3195 | #endif | |
3196 | ||
3197 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
63d8ea7f | 3198 | if (!ptype->dev || ptype->dev == skb->dev) { |
4ec93edb | 3199 | if (pt_prev) |
f2ccd8fa | 3200 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
3201 | pt_prev = ptype; |
3202 | } | |
3203 | } | |
3204 | ||
3205 | #ifdef CONFIG_NET_CLS_ACT | |
f697c3e8 HX |
3206 | skb = handle_ing(skb, &pt_prev, &ret, orig_dev); |
3207 | if (!skb) | |
1da177e4 | 3208 | goto out; |
1da177e4 LT |
3209 | ncls: |
3210 | #endif | |
3211 | ||
6a32e4f9 | 3212 | rx_handler = rcu_dereference(skb->dev->rx_handler); |
2425717b JF |
3213 | if (vlan_tx_tag_present(skb)) { |
3214 | if (pt_prev) { | |
3215 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3216 | pt_prev = NULL; | |
3217 | } | |
6a32e4f9 | 3218 | if (vlan_do_receive(&skb, !rx_handler)) |
2425717b JF |
3219 | goto another_round; |
3220 | else if (unlikely(!skb)) | |
3221 | goto out; | |
3222 | } | |
3223 | ||
ab95bfe0 JP |
3224 | if (rx_handler) { |
3225 | if (pt_prev) { | |
3226 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3227 | pt_prev = NULL; | |
3228 | } | |
8a4eb573 JP |
3229 | switch (rx_handler(&skb)) { |
3230 | case RX_HANDLER_CONSUMED: | |
ab95bfe0 | 3231 | goto out; |
8a4eb573 | 3232 | case RX_HANDLER_ANOTHER: |
63d8ea7f | 3233 | goto another_round; |
8a4eb573 JP |
3234 | case RX_HANDLER_EXACT: |
3235 | deliver_exact = true; | |
3236 | case RX_HANDLER_PASS: | |
3237 | break; | |
3238 | default: | |
3239 | BUG(); | |
3240 | } | |
ab95bfe0 | 3241 | } |
1da177e4 | 3242 | |
63d8ea7f | 3243 | /* deliver only exact match when indicated */ |
8a4eb573 | 3244 | null_or_dev = deliver_exact ? skb->dev : NULL; |
1f3c8804 | 3245 | |
1da177e4 | 3246 | type = skb->protocol; |
82d8a867 PE |
3247 | list_for_each_entry_rcu(ptype, |
3248 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
63d8ea7f | 3249 | if (ptype->type == type && |
e3f48d37 JP |
3250 | (ptype->dev == null_or_dev || ptype->dev == skb->dev || |
3251 | ptype->dev == orig_dev)) { | |
4ec93edb | 3252 | if (pt_prev) |
f2ccd8fa | 3253 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
3254 | pt_prev = ptype; |
3255 | } | |
3256 | } | |
3257 | ||
3258 | if (pt_prev) { | |
f2ccd8fa | 3259 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
1da177e4 | 3260 | } else { |
caf586e5 | 3261 | atomic_long_inc(&skb->dev->rx_dropped); |
1da177e4 LT |
3262 | kfree_skb(skb); |
3263 | /* Jamal, now you will not able to escape explaining | |
3264 | * me how you were going to use this. :-) | |
3265 | */ | |
3266 | ret = NET_RX_DROP; | |
3267 | } | |
3268 | ||
3269 | out: | |
3270 | rcu_read_unlock(); | |
3271 | return ret; | |
3272 | } | |
0a9627f2 TH |
3273 | |
3274 | /** | |
3275 | * netif_receive_skb - process receive buffer from network | |
3276 | * @skb: buffer to process | |
3277 | * | |
3278 | * netif_receive_skb() is the main receive data processing function. | |
3279 | * It always succeeds. The buffer may be dropped during processing | |
3280 | * for congestion control or by the protocol layers. | |
3281 | * | |
3282 | * This function may only be called from softirq context and interrupts | |
3283 | * should be enabled. | |
3284 | * | |
3285 | * Return values (usually ignored): | |
3286 | * NET_RX_SUCCESS: no congestion | |
3287 | * NET_RX_DROP: packet was dropped | |
3288 | */ | |
3289 | int netif_receive_skb(struct sk_buff *skb) | |
3290 | { | |
588f0330 | 3291 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
3b098e2d | 3292 | |
c1f19b51 RC |
3293 | if (skb_defer_rx_timestamp(skb)) |
3294 | return NET_RX_SUCCESS; | |
3295 | ||
df334545 | 3296 | #ifdef CONFIG_RPS |
c5905afb | 3297 | if (static_key_false(&rps_needed)) { |
3b098e2d ED |
3298 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
3299 | int cpu, ret; | |
fec5e652 | 3300 | |
3b098e2d ED |
3301 | rcu_read_lock(); |
3302 | ||
3303 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
0a9627f2 | 3304 | |
3b098e2d ED |
3305 | if (cpu >= 0) { |
3306 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
3307 | rcu_read_unlock(); | |
adc9300e | 3308 | return ret; |
3b098e2d | 3309 | } |
adc9300e | 3310 | rcu_read_unlock(); |
fec5e652 | 3311 | } |
1e94d72f | 3312 | #endif |
adc9300e | 3313 | return __netif_receive_skb(skb); |
0a9627f2 | 3314 | } |
d1b19dff | 3315 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 3316 | |
88751275 ED |
3317 | /* Network device is going away, flush any packets still pending |
3318 | * Called with irqs disabled. | |
3319 | */ | |
152102c7 | 3320 | static void flush_backlog(void *arg) |
6e583ce5 | 3321 | { |
152102c7 | 3322 | struct net_device *dev = arg; |
e36fa2f7 | 3323 | struct softnet_data *sd = &__get_cpu_var(softnet_data); |
6e583ce5 SH |
3324 | struct sk_buff *skb, *tmp; |
3325 | ||
e36fa2f7 | 3326 | rps_lock(sd); |
6e7676c1 | 3327 | skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { |
6e583ce5 | 3328 | if (skb->dev == dev) { |
e36fa2f7 | 3329 | __skb_unlink(skb, &sd->input_pkt_queue); |
6e583ce5 | 3330 | kfree_skb(skb); |
76cc8b13 | 3331 | input_queue_head_incr(sd); |
6e583ce5 | 3332 | } |
6e7676c1 | 3333 | } |
e36fa2f7 | 3334 | rps_unlock(sd); |
6e7676c1 CG |
3335 | |
3336 | skb_queue_walk_safe(&sd->process_queue, skb, tmp) { | |
3337 | if (skb->dev == dev) { | |
3338 | __skb_unlink(skb, &sd->process_queue); | |
3339 | kfree_skb(skb); | |
76cc8b13 | 3340 | input_queue_head_incr(sd); |
6e7676c1 CG |
3341 | } |
3342 | } | |
6e583ce5 SH |
3343 | } |
3344 | ||
d565b0a1 HX |
3345 | static int napi_gro_complete(struct sk_buff *skb) |
3346 | { | |
3347 | struct packet_type *ptype; | |
3348 | __be16 type = skb->protocol; | |
3349 | struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK]; | |
3350 | int err = -ENOENT; | |
3351 | ||
fc59f9a3 HX |
3352 | if (NAPI_GRO_CB(skb)->count == 1) { |
3353 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 3354 | goto out; |
fc59f9a3 | 3355 | } |
d565b0a1 HX |
3356 | |
3357 | rcu_read_lock(); | |
3358 | list_for_each_entry_rcu(ptype, head, list) { | |
3359 | if (ptype->type != type || ptype->dev || !ptype->gro_complete) | |
3360 | continue; | |
3361 | ||
3362 | err = ptype->gro_complete(skb); | |
3363 | break; | |
3364 | } | |
3365 | rcu_read_unlock(); | |
3366 | ||
3367 | if (err) { | |
3368 | WARN_ON(&ptype->list == head); | |
3369 | kfree_skb(skb); | |
3370 | return NET_RX_SUCCESS; | |
3371 | } | |
3372 | ||
3373 | out: | |
d565b0a1 HX |
3374 | return netif_receive_skb(skb); |
3375 | } | |
3376 | ||
86cac58b | 3377 | inline void napi_gro_flush(struct napi_struct *napi) |
d565b0a1 HX |
3378 | { |
3379 | struct sk_buff *skb, *next; | |
3380 | ||
3381 | for (skb = napi->gro_list; skb; skb = next) { | |
3382 | next = skb->next; | |
3383 | skb->next = NULL; | |
3384 | napi_gro_complete(skb); | |
3385 | } | |
3386 | ||
4ae5544f | 3387 | napi->gro_count = 0; |
d565b0a1 HX |
3388 | napi->gro_list = NULL; |
3389 | } | |
86cac58b | 3390 | EXPORT_SYMBOL(napi_gro_flush); |
d565b0a1 | 3391 | |
5b252f0c | 3392 | enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 HX |
3393 | { |
3394 | struct sk_buff **pp = NULL; | |
3395 | struct packet_type *ptype; | |
3396 | __be16 type = skb->protocol; | |
3397 | struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK]; | |
0da2afd5 | 3398 | int same_flow; |
d565b0a1 | 3399 | int mac_len; |
5b252f0c | 3400 | enum gro_result ret; |
d565b0a1 | 3401 | |
ce9e76c8 | 3402 | if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb)) |
d565b0a1 HX |
3403 | goto normal; |
3404 | ||
21dc3301 | 3405 | if (skb_is_gso(skb) || skb_has_frag_list(skb)) |
f17f5c91 HX |
3406 | goto normal; |
3407 | ||
d565b0a1 HX |
3408 | rcu_read_lock(); |
3409 | list_for_each_entry_rcu(ptype, head, list) { | |
d565b0a1 HX |
3410 | if (ptype->type != type || ptype->dev || !ptype->gro_receive) |
3411 | continue; | |
3412 | ||
86911732 | 3413 | skb_set_network_header(skb, skb_gro_offset(skb)); |
d565b0a1 HX |
3414 | mac_len = skb->network_header - skb->mac_header; |
3415 | skb->mac_len = mac_len; | |
3416 | NAPI_GRO_CB(skb)->same_flow = 0; | |
3417 | NAPI_GRO_CB(skb)->flush = 0; | |
5d38a079 | 3418 | NAPI_GRO_CB(skb)->free = 0; |
d565b0a1 | 3419 | |
d565b0a1 HX |
3420 | pp = ptype->gro_receive(&napi->gro_list, skb); |
3421 | break; | |
3422 | } | |
3423 | rcu_read_unlock(); | |
3424 | ||
3425 | if (&ptype->list == head) | |
3426 | goto normal; | |
3427 | ||
0da2afd5 | 3428 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 3429 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 3430 | |
d565b0a1 HX |
3431 | if (pp) { |
3432 | struct sk_buff *nskb = *pp; | |
3433 | ||
3434 | *pp = nskb->next; | |
3435 | nskb->next = NULL; | |
3436 | napi_gro_complete(nskb); | |
4ae5544f | 3437 | napi->gro_count--; |
d565b0a1 HX |
3438 | } |
3439 | ||
0da2afd5 | 3440 | if (same_flow) |
d565b0a1 HX |
3441 | goto ok; |
3442 | ||
4ae5544f | 3443 | if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS) |
d565b0a1 | 3444 | goto normal; |
d565b0a1 | 3445 | |
4ae5544f | 3446 | napi->gro_count++; |
d565b0a1 | 3447 | NAPI_GRO_CB(skb)->count = 1; |
86911732 | 3448 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
d565b0a1 HX |
3449 | skb->next = napi->gro_list; |
3450 | napi->gro_list = skb; | |
5d0d9be8 | 3451 | ret = GRO_HELD; |
d565b0a1 | 3452 | |
ad0f9904 | 3453 | pull: |
cb18978c HX |
3454 | if (skb_headlen(skb) < skb_gro_offset(skb)) { |
3455 | int grow = skb_gro_offset(skb) - skb_headlen(skb); | |
3456 | ||
3457 | BUG_ON(skb->end - skb->tail < grow); | |
3458 | ||
3459 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
3460 | ||
3461 | skb->tail += grow; | |
3462 | skb->data_len -= grow; | |
3463 | ||
3464 | skb_shinfo(skb)->frags[0].page_offset += grow; | |
9e903e08 | 3465 | skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow); |
cb18978c | 3466 | |
9e903e08 | 3467 | if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) { |
ea2ab693 | 3468 | skb_frag_unref(skb, 0); |
cb18978c HX |
3469 | memmove(skb_shinfo(skb)->frags, |
3470 | skb_shinfo(skb)->frags + 1, | |
e5093aec | 3471 | --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t)); |
cb18978c | 3472 | } |
ad0f9904 HX |
3473 | } |
3474 | ||
d565b0a1 | 3475 | ok: |
5d0d9be8 | 3476 | return ret; |
d565b0a1 HX |
3477 | |
3478 | normal: | |
ad0f9904 HX |
3479 | ret = GRO_NORMAL; |
3480 | goto pull; | |
5d38a079 | 3481 | } |
96e93eab HX |
3482 | EXPORT_SYMBOL(dev_gro_receive); |
3483 | ||
40d0802b | 3484 | static inline gro_result_t |
5b252f0c | 3485 | __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab HX |
3486 | { |
3487 | struct sk_buff *p; | |
5ca3b72c | 3488 | unsigned int maclen = skb->dev->hard_header_len; |
96e93eab HX |
3489 | |
3490 | for (p = napi->gro_list; p; p = p->next) { | |
40d0802b ED |
3491 | unsigned long diffs; |
3492 | ||
3493 | diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; | |
3701e513 | 3494 | diffs |= p->vlan_tci ^ skb->vlan_tci; |
5ca3b72c ED |
3495 | if (maclen == ETH_HLEN) |
3496 | diffs |= compare_ether_header(skb_mac_header(p), | |
3497 | skb_gro_mac_header(skb)); | |
3498 | else if (!diffs) | |
3499 | diffs = memcmp(skb_mac_header(p), | |
3500 | skb_gro_mac_header(skb), | |
3501 | maclen); | |
40d0802b | 3502 | NAPI_GRO_CB(p)->same_flow = !diffs; |
96e93eab HX |
3503 | NAPI_GRO_CB(p)->flush = 0; |
3504 | } | |
3505 | ||
3506 | return dev_gro_receive(napi, skb); | |
3507 | } | |
5d38a079 | 3508 | |
c7c4b3b6 | 3509 | gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) |
5d38a079 | 3510 | { |
5d0d9be8 HX |
3511 | switch (ret) { |
3512 | case GRO_NORMAL: | |
c7c4b3b6 BH |
3513 | if (netif_receive_skb(skb)) |
3514 | ret = GRO_DROP; | |
3515 | break; | |
5d38a079 | 3516 | |
5d0d9be8 | 3517 | case GRO_DROP: |
5d38a079 HX |
3518 | kfree_skb(skb); |
3519 | break; | |
5b252f0c | 3520 | |
daa86548 | 3521 | case GRO_MERGED_FREE: |
d7e8883c ED |
3522 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) |
3523 | kmem_cache_free(skbuff_head_cache, skb); | |
3524 | else | |
3525 | __kfree_skb(skb); | |
daa86548 ED |
3526 | break; |
3527 | ||
5b252f0c BH |
3528 | case GRO_HELD: |
3529 | case GRO_MERGED: | |
3530 | break; | |
5d38a079 HX |
3531 | } |
3532 | ||
c7c4b3b6 | 3533 | return ret; |
5d0d9be8 HX |
3534 | } |
3535 | EXPORT_SYMBOL(napi_skb_finish); | |
3536 | ||
78a478d0 HX |
3537 | void skb_gro_reset_offset(struct sk_buff *skb) |
3538 | { | |
3539 | NAPI_GRO_CB(skb)->data_offset = 0; | |
3540 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
7489594c | 3541 | NAPI_GRO_CB(skb)->frag0_len = 0; |
78a478d0 | 3542 | |
78d3fd0b | 3543 | if (skb->mac_header == skb->tail && |
ea2ab693 | 3544 | !PageHighMem(skb_frag_page(&skb_shinfo(skb)->frags[0]))) { |
78a478d0 | 3545 | NAPI_GRO_CB(skb)->frag0 = |
ea2ab693 | 3546 | skb_frag_address(&skb_shinfo(skb)->frags[0]); |
9e903e08 | 3547 | NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(&skb_shinfo(skb)->frags[0]); |
7489594c | 3548 | } |
78a478d0 HX |
3549 | } |
3550 | EXPORT_SYMBOL(skb_gro_reset_offset); | |
3551 | ||
c7c4b3b6 | 3552 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 3553 | { |
86911732 HX |
3554 | skb_gro_reset_offset(skb); |
3555 | ||
5d0d9be8 | 3556 | return napi_skb_finish(__napi_gro_receive(napi, skb), skb); |
d565b0a1 HX |
3557 | } |
3558 | EXPORT_SYMBOL(napi_gro_receive); | |
3559 | ||
d0c2b0d2 | 3560 | static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab | 3561 | { |
96e93eab | 3562 | __skb_pull(skb, skb_headlen(skb)); |
2a2a459e ED |
3563 | /* restore the reserve we had after netdev_alloc_skb_ip_align() */ |
3564 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); | |
3701e513 | 3565 | skb->vlan_tci = 0; |
66c46d74 | 3566 | skb->dev = napi->dev; |
6d152e23 | 3567 | skb->skb_iif = 0; |
96e93eab HX |
3568 | |
3569 | napi->skb = skb; | |
3570 | } | |
96e93eab | 3571 | |
76620aaf | 3572 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 3573 | { |
5d38a079 | 3574 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
3575 | |
3576 | if (!skb) { | |
89d71a66 ED |
3577 | skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD); |
3578 | if (skb) | |
3579 | napi->skb = skb; | |
80595d59 | 3580 | } |
96e93eab HX |
3581 | return skb; |
3582 | } | |
76620aaf | 3583 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 3584 | |
c7c4b3b6 BH |
3585 | gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb, |
3586 | gro_result_t ret) | |
96e93eab | 3587 | { |
5d0d9be8 HX |
3588 | switch (ret) { |
3589 | case GRO_NORMAL: | |
86911732 | 3590 | case GRO_HELD: |
e76b69cc | 3591 | skb->protocol = eth_type_trans(skb, skb->dev); |
86911732 | 3592 | |
c7c4b3b6 BH |
3593 | if (ret == GRO_HELD) |
3594 | skb_gro_pull(skb, -ETH_HLEN); | |
3595 | else if (netif_receive_skb(skb)) | |
3596 | ret = GRO_DROP; | |
86911732 | 3597 | break; |
5d38a079 | 3598 | |
5d0d9be8 | 3599 | case GRO_DROP: |
5d0d9be8 HX |
3600 | case GRO_MERGED_FREE: |
3601 | napi_reuse_skb(napi, skb); | |
3602 | break; | |
5b252f0c BH |
3603 | |
3604 | case GRO_MERGED: | |
3605 | break; | |
5d0d9be8 | 3606 | } |
5d38a079 | 3607 | |
c7c4b3b6 | 3608 | return ret; |
5d38a079 | 3609 | } |
5d0d9be8 HX |
3610 | EXPORT_SYMBOL(napi_frags_finish); |
3611 | ||
4adb9c4a | 3612 | static struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
76620aaf HX |
3613 | { |
3614 | struct sk_buff *skb = napi->skb; | |
3615 | struct ethhdr *eth; | |
a5b1cf28 HX |
3616 | unsigned int hlen; |
3617 | unsigned int off; | |
76620aaf HX |
3618 | |
3619 | napi->skb = NULL; | |
3620 | ||
3621 | skb_reset_mac_header(skb); | |
3622 | skb_gro_reset_offset(skb); | |
3623 | ||
a5b1cf28 HX |
3624 | off = skb_gro_offset(skb); |
3625 | hlen = off + sizeof(*eth); | |
3626 | eth = skb_gro_header_fast(skb, off); | |
3627 | if (skb_gro_header_hard(skb, hlen)) { | |
3628 | eth = skb_gro_header_slow(skb, hlen, off); | |
3629 | if (unlikely(!eth)) { | |
3630 | napi_reuse_skb(napi, skb); | |
3631 | skb = NULL; | |
3632 | goto out; | |
3633 | } | |
76620aaf HX |
3634 | } |
3635 | ||
3636 | skb_gro_pull(skb, sizeof(*eth)); | |
3637 | ||
3638 | /* | |
3639 | * This works because the only protocols we care about don't require | |
3640 | * special handling. We'll fix it up properly at the end. | |
3641 | */ | |
3642 | skb->protocol = eth->h_proto; | |
3643 | ||
3644 | out: | |
3645 | return skb; | |
3646 | } | |
76620aaf | 3647 | |
c7c4b3b6 | 3648 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 3649 | { |
76620aaf | 3650 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 HX |
3651 | |
3652 | if (!skb) | |
c7c4b3b6 | 3653 | return GRO_DROP; |
5d0d9be8 HX |
3654 | |
3655 | return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb)); | |
3656 | } | |
5d38a079 HX |
3657 | EXPORT_SYMBOL(napi_gro_frags); |
3658 | ||
e326bed2 ED |
3659 | /* |
3660 | * net_rps_action sends any pending IPI's for rps. | |
3661 | * Note: called with local irq disabled, but exits with local irq enabled. | |
3662 | */ | |
3663 | static void net_rps_action_and_irq_enable(struct softnet_data *sd) | |
3664 | { | |
3665 | #ifdef CONFIG_RPS | |
3666 | struct softnet_data *remsd = sd->rps_ipi_list; | |
3667 | ||
3668 | if (remsd) { | |
3669 | sd->rps_ipi_list = NULL; | |
3670 | ||
3671 | local_irq_enable(); | |
3672 | ||
3673 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
3674 | while (remsd) { | |
3675 | struct softnet_data *next = remsd->rps_ipi_next; | |
3676 | ||
3677 | if (cpu_online(remsd->cpu)) | |
3678 | __smp_call_function_single(remsd->cpu, | |
3679 | &remsd->csd, 0); | |
3680 | remsd = next; | |
3681 | } | |
3682 | } else | |
3683 | #endif | |
3684 | local_irq_enable(); | |
3685 | } | |
3686 | ||
bea3348e | 3687 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 LT |
3688 | { |
3689 | int work = 0; | |
eecfd7c4 | 3690 | struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); |
1da177e4 | 3691 | |
e326bed2 ED |
3692 | #ifdef CONFIG_RPS |
3693 | /* Check if we have pending ipi, its better to send them now, | |
3694 | * not waiting net_rx_action() end. | |
3695 | */ | |
3696 | if (sd->rps_ipi_list) { | |
3697 | local_irq_disable(); | |
3698 | net_rps_action_and_irq_enable(sd); | |
3699 | } | |
3700 | #endif | |
bea3348e | 3701 | napi->weight = weight_p; |
6e7676c1 CG |
3702 | local_irq_disable(); |
3703 | while (work < quota) { | |
1da177e4 | 3704 | struct sk_buff *skb; |
6e7676c1 CG |
3705 | unsigned int qlen; |
3706 | ||
3707 | while ((skb = __skb_dequeue(&sd->process_queue))) { | |
3708 | local_irq_enable(); | |
3709 | __netif_receive_skb(skb); | |
6e7676c1 | 3710 | local_irq_disable(); |
76cc8b13 TH |
3711 | input_queue_head_incr(sd); |
3712 | if (++work >= quota) { | |
3713 | local_irq_enable(); | |
3714 | return work; | |
3715 | } | |
6e7676c1 | 3716 | } |
1da177e4 | 3717 | |
e36fa2f7 | 3718 | rps_lock(sd); |
6e7676c1 | 3719 | qlen = skb_queue_len(&sd->input_pkt_queue); |
76cc8b13 | 3720 | if (qlen) |
6e7676c1 CG |
3721 | skb_queue_splice_tail_init(&sd->input_pkt_queue, |
3722 | &sd->process_queue); | |
76cc8b13 | 3723 | |
6e7676c1 | 3724 | if (qlen < quota - work) { |
eecfd7c4 ED |
3725 | /* |
3726 | * Inline a custom version of __napi_complete(). | |
3727 | * only current cpu owns and manipulates this napi, | |
3728 | * and NAPI_STATE_SCHED is the only possible flag set on backlog. | |
3729 | * we can use a plain write instead of clear_bit(), | |
3730 | * and we dont need an smp_mb() memory barrier. | |
3731 | */ | |
3732 | list_del(&napi->poll_list); | |
3733 | napi->state = 0; | |
3734 | ||
6e7676c1 | 3735 | quota = work + qlen; |
bea3348e | 3736 | } |
e36fa2f7 | 3737 | rps_unlock(sd); |
6e7676c1 CG |
3738 | } |
3739 | local_irq_enable(); | |
1da177e4 | 3740 | |
bea3348e SH |
3741 | return work; |
3742 | } | |
1da177e4 | 3743 | |
bea3348e SH |
3744 | /** |
3745 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 3746 | * @n: entry to schedule |
bea3348e SH |
3747 | * |
3748 | * The entry's receive function will be scheduled to run | |
3749 | */ | |
b5606c2d | 3750 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
3751 | { |
3752 | unsigned long flags; | |
1da177e4 | 3753 | |
bea3348e | 3754 | local_irq_save(flags); |
eecfd7c4 | 3755 | ____napi_schedule(&__get_cpu_var(softnet_data), n); |
bea3348e | 3756 | local_irq_restore(flags); |
1da177e4 | 3757 | } |
bea3348e SH |
3758 | EXPORT_SYMBOL(__napi_schedule); |
3759 | ||
d565b0a1 HX |
3760 | void __napi_complete(struct napi_struct *n) |
3761 | { | |
3762 | BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); | |
3763 | BUG_ON(n->gro_list); | |
3764 | ||
3765 | list_del(&n->poll_list); | |
3766 | smp_mb__before_clear_bit(); | |
3767 | clear_bit(NAPI_STATE_SCHED, &n->state); | |
3768 | } | |
3769 | EXPORT_SYMBOL(__napi_complete); | |
3770 | ||
3771 | void napi_complete(struct napi_struct *n) | |
3772 | { | |
3773 | unsigned long flags; | |
3774 | ||
3775 | /* | |
3776 | * don't let napi dequeue from the cpu poll list | |
3777 | * just in case its running on a different cpu | |
3778 | */ | |
3779 | if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state))) | |
3780 | return; | |
3781 | ||
3782 | napi_gro_flush(n); | |
3783 | local_irq_save(flags); | |
3784 | __napi_complete(n); | |
3785 | local_irq_restore(flags); | |
3786 | } | |
3787 | EXPORT_SYMBOL(napi_complete); | |
3788 | ||
3789 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, | |
3790 | int (*poll)(struct napi_struct *, int), int weight) | |
3791 | { | |
3792 | INIT_LIST_HEAD(&napi->poll_list); | |
4ae5544f | 3793 | napi->gro_count = 0; |
d565b0a1 | 3794 | napi->gro_list = NULL; |
5d38a079 | 3795 | napi->skb = NULL; |
d565b0a1 HX |
3796 | napi->poll = poll; |
3797 | napi->weight = weight; | |
3798 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 3799 | napi->dev = dev; |
5d38a079 | 3800 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
3801 | spin_lock_init(&napi->poll_lock); |
3802 | napi->poll_owner = -1; | |
3803 | #endif | |
3804 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
3805 | } | |
3806 | EXPORT_SYMBOL(netif_napi_add); | |
3807 | ||
3808 | void netif_napi_del(struct napi_struct *napi) | |
3809 | { | |
3810 | struct sk_buff *skb, *next; | |
3811 | ||
d7b06636 | 3812 | list_del_init(&napi->dev_list); |
76620aaf | 3813 | napi_free_frags(napi); |
d565b0a1 HX |
3814 | |
3815 | for (skb = napi->gro_list; skb; skb = next) { | |
3816 | next = skb->next; | |
3817 | skb->next = NULL; | |
3818 | kfree_skb(skb); | |
3819 | } | |
3820 | ||
3821 | napi->gro_list = NULL; | |
4ae5544f | 3822 | napi->gro_count = 0; |
d565b0a1 HX |
3823 | } |
3824 | EXPORT_SYMBOL(netif_napi_del); | |
3825 | ||
1da177e4 LT |
3826 | static void net_rx_action(struct softirq_action *h) |
3827 | { | |
e326bed2 | 3828 | struct softnet_data *sd = &__get_cpu_var(softnet_data); |
24f8b238 | 3829 | unsigned long time_limit = jiffies + 2; |
51b0bded | 3830 | int budget = netdev_budget; |
53fb95d3 MM |
3831 | void *have; |
3832 | ||
1da177e4 LT |
3833 | local_irq_disable(); |
3834 | ||
e326bed2 | 3835 | while (!list_empty(&sd->poll_list)) { |
bea3348e SH |
3836 | struct napi_struct *n; |
3837 | int work, weight; | |
1da177e4 | 3838 | |
bea3348e | 3839 | /* If softirq window is exhuasted then punt. |
24f8b238 SH |
3840 | * Allow this to run for 2 jiffies since which will allow |
3841 | * an average latency of 1.5/HZ. | |
bea3348e | 3842 | */ |
24f8b238 | 3843 | if (unlikely(budget <= 0 || time_after(jiffies, time_limit))) |
1da177e4 LT |
3844 | goto softnet_break; |
3845 | ||
3846 | local_irq_enable(); | |
3847 | ||
bea3348e SH |
3848 | /* Even though interrupts have been re-enabled, this |
3849 | * access is safe because interrupts can only add new | |
3850 | * entries to the tail of this list, and only ->poll() | |
3851 | * calls can remove this head entry from the list. | |
3852 | */ | |
e326bed2 | 3853 | n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list); |
1da177e4 | 3854 | |
bea3348e SH |
3855 | have = netpoll_poll_lock(n); |
3856 | ||
3857 | weight = n->weight; | |
3858 | ||
0a7606c1 DM |
3859 | /* This NAPI_STATE_SCHED test is for avoiding a race |
3860 | * with netpoll's poll_napi(). Only the entity which | |
3861 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
3862 | * actually make the ->poll() call. Therefore we avoid | |
25985edc | 3863 | * accidentally calling ->poll() when NAPI is not scheduled. |
0a7606c1 DM |
3864 | */ |
3865 | work = 0; | |
4ea7e386 | 3866 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { |
0a7606c1 | 3867 | work = n->poll(n, weight); |
4ea7e386 NH |
3868 | trace_napi_poll(n); |
3869 | } | |
bea3348e SH |
3870 | |
3871 | WARN_ON_ONCE(work > weight); | |
3872 | ||
3873 | budget -= work; | |
3874 | ||
3875 | local_irq_disable(); | |
3876 | ||
3877 | /* Drivers must not modify the NAPI state if they | |
3878 | * consume the entire weight. In such cases this code | |
3879 | * still "owns" the NAPI instance and therefore can | |
3880 | * move the instance around on the list at-will. | |
3881 | */ | |
fed17f30 | 3882 | if (unlikely(work == weight)) { |
ff780cd8 HX |
3883 | if (unlikely(napi_disable_pending(n))) { |
3884 | local_irq_enable(); | |
3885 | napi_complete(n); | |
3886 | local_irq_disable(); | |
3887 | } else | |
e326bed2 | 3888 | list_move_tail(&n->poll_list, &sd->poll_list); |
fed17f30 | 3889 | } |
bea3348e SH |
3890 | |
3891 | netpoll_poll_unlock(have); | |
1da177e4 LT |
3892 | } |
3893 | out: | |
e326bed2 | 3894 | net_rps_action_and_irq_enable(sd); |
0a9627f2 | 3895 | |
db217334 CL |
3896 | #ifdef CONFIG_NET_DMA |
3897 | /* | |
3898 | * There may not be any more sk_buffs coming right now, so push | |
3899 | * any pending DMA copies to hardware | |
3900 | */ | |
2ba05622 | 3901 | dma_issue_pending_all(); |
db217334 | 3902 | #endif |
bea3348e | 3903 | |
1da177e4 LT |
3904 | return; |
3905 | ||
3906 | softnet_break: | |
dee42870 | 3907 | sd->time_squeeze++; |
1da177e4 LT |
3908 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); |
3909 | goto out; | |
3910 | } | |
3911 | ||
d1b19dff | 3912 | static gifconf_func_t *gifconf_list[NPROTO]; |
1da177e4 LT |
3913 | |
3914 | /** | |
3915 | * register_gifconf - register a SIOCGIF handler | |
3916 | * @family: Address family | |
3917 | * @gifconf: Function handler | |
3918 | * | |
3919 | * Register protocol dependent address dumping routines. The handler | |
3920 | * that is passed must not be freed or reused until it has been replaced | |
3921 | * by another handler. | |
3922 | */ | |
d1b19dff | 3923 | int register_gifconf(unsigned int family, gifconf_func_t *gifconf) |
1da177e4 LT |
3924 | { |
3925 | if (family >= NPROTO) | |
3926 | return -EINVAL; | |
3927 | gifconf_list[family] = gifconf; | |
3928 | return 0; | |
3929 | } | |
d1b19dff | 3930 | EXPORT_SYMBOL(register_gifconf); |
1da177e4 LT |
3931 | |
3932 | ||
3933 | /* | |
3934 | * Map an interface index to its name (SIOCGIFNAME) | |
3935 | */ | |
3936 | ||
3937 | /* | |
3938 | * We need this ioctl for efficient implementation of the | |
3939 | * if_indextoname() function required by the IPv6 API. Without | |
3940 | * it, we would have to search all the interfaces to find a | |
3941 | * match. --pb | |
3942 | */ | |
3943 | ||
881d966b | 3944 | static int dev_ifname(struct net *net, struct ifreq __user *arg) |
1da177e4 LT |
3945 | { |
3946 | struct net_device *dev; | |
3947 | struct ifreq ifr; | |
3948 | ||
3949 | /* | |
3950 | * Fetch the caller's info block. | |
3951 | */ | |
3952 | ||
3953 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
3954 | return -EFAULT; | |
3955 | ||
fb699dfd ED |
3956 | rcu_read_lock(); |
3957 | dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex); | |
1da177e4 | 3958 | if (!dev) { |
fb699dfd | 3959 | rcu_read_unlock(); |
1da177e4 LT |
3960 | return -ENODEV; |
3961 | } | |
3962 | ||
3963 | strcpy(ifr.ifr_name, dev->name); | |
fb699dfd | 3964 | rcu_read_unlock(); |
1da177e4 LT |
3965 | |
3966 | if (copy_to_user(arg, &ifr, sizeof(struct ifreq))) | |
3967 | return -EFAULT; | |
3968 | return 0; | |
3969 | } | |
3970 | ||
3971 | /* | |
3972 | * Perform a SIOCGIFCONF call. This structure will change | |
3973 | * size eventually, and there is nothing I can do about it. | |
3974 | * Thus we will need a 'compatibility mode'. | |
3975 | */ | |
3976 | ||
881d966b | 3977 | static int dev_ifconf(struct net *net, char __user *arg) |
1da177e4 LT |
3978 | { |
3979 | struct ifconf ifc; | |
3980 | struct net_device *dev; | |
3981 | char __user *pos; | |
3982 | int len; | |
3983 | int total; | |
3984 | int i; | |
3985 | ||
3986 | /* | |
3987 | * Fetch the caller's info block. | |
3988 | */ | |
3989 | ||
3990 | if (copy_from_user(&ifc, arg, sizeof(struct ifconf))) | |
3991 | return -EFAULT; | |
3992 | ||
3993 | pos = ifc.ifc_buf; | |
3994 | len = ifc.ifc_len; | |
3995 | ||
3996 | /* | |
3997 | * Loop over the interfaces, and write an info block for each. | |
3998 | */ | |
3999 | ||
4000 | total = 0; | |
881d966b | 4001 | for_each_netdev(net, dev) { |
1da177e4 LT |
4002 | for (i = 0; i < NPROTO; i++) { |
4003 | if (gifconf_list[i]) { | |
4004 | int done; | |
4005 | if (!pos) | |
4006 | done = gifconf_list[i](dev, NULL, 0); | |
4007 | else | |
4008 | done = gifconf_list[i](dev, pos + total, | |
4009 | len - total); | |
4010 | if (done < 0) | |
4011 | return -EFAULT; | |
4012 | total += done; | |
4013 | } | |
4014 | } | |
4ec93edb | 4015 | } |
1da177e4 LT |
4016 | |
4017 | /* | |
4018 | * All done. Write the updated control block back to the caller. | |
4019 | */ | |
4020 | ifc.ifc_len = total; | |
4021 | ||
4022 | /* | |
4023 | * Both BSD and Solaris return 0 here, so we do too. | |
4024 | */ | |
4025 | return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0; | |
4026 | } | |
4027 | ||
4028 | #ifdef CONFIG_PROC_FS | |
f04565dd | 4029 | |
2def16ae | 4030 | #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1) |
f04565dd MM |
4031 | |
4032 | #define get_bucket(x) ((x) >> BUCKET_SPACE) | |
4033 | #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1)) | |
4034 | #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o)) | |
4035 | ||
2def16ae | 4036 | static inline struct net_device *dev_from_same_bucket(struct seq_file *seq, loff_t *pos) |
f04565dd | 4037 | { |
f04565dd MM |
4038 | struct net *net = seq_file_net(seq); |
4039 | struct net_device *dev; | |
4040 | struct hlist_node *p; | |
4041 | struct hlist_head *h; | |
2def16ae | 4042 | unsigned int count = 0, offset = get_offset(*pos); |
f04565dd | 4043 | |
2def16ae | 4044 | h = &net->dev_name_head[get_bucket(*pos)]; |
f04565dd | 4045 | hlist_for_each_entry_rcu(dev, p, h, name_hlist) { |
2def16ae | 4046 | if (++count == offset) |
f04565dd | 4047 | return dev; |
f04565dd MM |
4048 | } |
4049 | ||
4050 | return NULL; | |
4051 | } | |
4052 | ||
2def16ae | 4053 | static inline struct net_device *dev_from_bucket(struct seq_file *seq, loff_t *pos) |
f04565dd | 4054 | { |
f04565dd MM |
4055 | struct net_device *dev; |
4056 | unsigned int bucket; | |
4057 | ||
f04565dd | 4058 | do { |
2def16ae | 4059 | dev = dev_from_same_bucket(seq, pos); |
f04565dd MM |
4060 | if (dev) |
4061 | return dev; | |
4062 | ||
2def16ae ED |
4063 | bucket = get_bucket(*pos) + 1; |
4064 | *pos = set_bucket_offset(bucket, 1); | |
f04565dd MM |
4065 | } while (bucket < NETDEV_HASHENTRIES); |
4066 | ||
4067 | return NULL; | |
4068 | } | |
4069 | ||
1da177e4 LT |
4070 | /* |
4071 | * This is invoked by the /proc filesystem handler to display a device | |
4072 | * in detail. | |
4073 | */ | |
7562f876 | 4074 | void *dev_seq_start(struct seq_file *seq, loff_t *pos) |
c6d14c84 | 4075 | __acquires(RCU) |
1da177e4 | 4076 | { |
c6d14c84 | 4077 | rcu_read_lock(); |
7562f876 PE |
4078 | if (!*pos) |
4079 | return SEQ_START_TOKEN; | |
1da177e4 | 4080 | |
2def16ae | 4081 | if (get_bucket(*pos) >= NETDEV_HASHENTRIES) |
f04565dd | 4082 | return NULL; |
1da177e4 | 4083 | |
2def16ae | 4084 | return dev_from_bucket(seq, pos); |
1da177e4 LT |
4085 | } |
4086 | ||
4087 | void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
4088 | { | |
f04565dd | 4089 | ++*pos; |
2def16ae | 4090 | return dev_from_bucket(seq, pos); |
1da177e4 LT |
4091 | } |
4092 | ||
4093 | void dev_seq_stop(struct seq_file *seq, void *v) | |
c6d14c84 | 4094 | __releases(RCU) |
1da177e4 | 4095 | { |
c6d14c84 | 4096 | rcu_read_unlock(); |
1da177e4 LT |
4097 | } |
4098 | ||
4099 | static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev) | |
4100 | { | |
28172739 ED |
4101 | struct rtnl_link_stats64 temp; |
4102 | const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp); | |
1da177e4 | 4103 | |
be1f3c2c BH |
4104 | seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu " |
4105 | "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n", | |
5a1b5898 RR |
4106 | dev->name, stats->rx_bytes, stats->rx_packets, |
4107 | stats->rx_errors, | |
4108 | stats->rx_dropped + stats->rx_missed_errors, | |
4109 | stats->rx_fifo_errors, | |
4110 | stats->rx_length_errors + stats->rx_over_errors + | |
4111 | stats->rx_crc_errors + stats->rx_frame_errors, | |
4112 | stats->rx_compressed, stats->multicast, | |
4113 | stats->tx_bytes, stats->tx_packets, | |
4114 | stats->tx_errors, stats->tx_dropped, | |
4115 | stats->tx_fifo_errors, stats->collisions, | |
4116 | stats->tx_carrier_errors + | |
4117 | stats->tx_aborted_errors + | |
4118 | stats->tx_window_errors + | |
4119 | stats->tx_heartbeat_errors, | |
4120 | stats->tx_compressed); | |
1da177e4 LT |
4121 | } |
4122 | ||
4123 | /* | |
4124 | * Called from the PROCfs module. This now uses the new arbitrary sized | |
4125 | * /proc/net interface to create /proc/net/dev | |
4126 | */ | |
4127 | static int dev_seq_show(struct seq_file *seq, void *v) | |
4128 | { | |
4129 | if (v == SEQ_START_TOKEN) | |
4130 | seq_puts(seq, "Inter-| Receive " | |
4131 | " | Transmit\n" | |
4132 | " face |bytes packets errs drop fifo frame " | |
4133 | "compressed multicast|bytes packets errs " | |
4134 | "drop fifo colls carrier compressed\n"); | |
4135 | else | |
4136 | dev_seq_printf_stats(seq, v); | |
4137 | return 0; | |
4138 | } | |
4139 | ||
dee42870 | 4140 | static struct softnet_data *softnet_get_online(loff_t *pos) |
1da177e4 | 4141 | { |
dee42870 | 4142 | struct softnet_data *sd = NULL; |
1da177e4 | 4143 | |
0c0b0aca | 4144 | while (*pos < nr_cpu_ids) |
4ec93edb | 4145 | if (cpu_online(*pos)) { |
dee42870 | 4146 | sd = &per_cpu(softnet_data, *pos); |
1da177e4 LT |
4147 | break; |
4148 | } else | |
4149 | ++*pos; | |
dee42870 | 4150 | return sd; |
1da177e4 LT |
4151 | } |
4152 | ||
4153 | static void *softnet_seq_start(struct seq_file *seq, loff_t *pos) | |
4154 | { | |
4155 | return softnet_get_online(pos); | |
4156 | } | |
4157 | ||
4158 | static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
4159 | { | |
4160 | ++*pos; | |
4161 | return softnet_get_online(pos); | |
4162 | } | |
4163 | ||
4164 | static void softnet_seq_stop(struct seq_file *seq, void *v) | |
4165 | { | |
4166 | } | |
4167 | ||
4168 | static int softnet_seq_show(struct seq_file *seq, void *v) | |
4169 | { | |
dee42870 | 4170 | struct softnet_data *sd = v; |
1da177e4 | 4171 | |
0a9627f2 | 4172 | seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n", |
dee42870 | 4173 | sd->processed, sd->dropped, sd->time_squeeze, 0, |
c1ebcdb8 | 4174 | 0, 0, 0, 0, /* was fastroute */ |
dee42870 | 4175 | sd->cpu_collision, sd->received_rps); |
1da177e4 LT |
4176 | return 0; |
4177 | } | |
4178 | ||
f690808e | 4179 | static const struct seq_operations dev_seq_ops = { |
1da177e4 LT |
4180 | .start = dev_seq_start, |
4181 | .next = dev_seq_next, | |
4182 | .stop = dev_seq_stop, | |
4183 | .show = dev_seq_show, | |
4184 | }; | |
4185 | ||
4186 | static int dev_seq_open(struct inode *inode, struct file *file) | |
4187 | { | |
e372c414 | 4188 | return seq_open_net(inode, file, &dev_seq_ops, |
2def16ae | 4189 | sizeof(struct seq_net_private)); |
5cac98dd AB |
4190 | } |
4191 | ||
9a32144e | 4192 | static const struct file_operations dev_seq_fops = { |
1da177e4 LT |
4193 | .owner = THIS_MODULE, |
4194 | .open = dev_seq_open, | |
4195 | .read = seq_read, | |
4196 | .llseek = seq_lseek, | |
e372c414 | 4197 | .release = seq_release_net, |
1da177e4 LT |
4198 | }; |
4199 | ||
f690808e | 4200 | static const struct seq_operations softnet_seq_ops = { |
1da177e4 LT |
4201 | .start = softnet_seq_start, |
4202 | .next = softnet_seq_next, | |
4203 | .stop = softnet_seq_stop, | |
4204 | .show = softnet_seq_show, | |
4205 | }; | |
4206 | ||
4207 | static int softnet_seq_open(struct inode *inode, struct file *file) | |
4208 | { | |
4209 | return seq_open(file, &softnet_seq_ops); | |
4210 | } | |
4211 | ||
9a32144e | 4212 | static const struct file_operations softnet_seq_fops = { |
1da177e4 LT |
4213 | .owner = THIS_MODULE, |
4214 | .open = softnet_seq_open, | |
4215 | .read = seq_read, | |
4216 | .llseek = seq_lseek, | |
4217 | .release = seq_release, | |
4218 | }; | |
4219 | ||
0e1256ff SH |
4220 | static void *ptype_get_idx(loff_t pos) |
4221 | { | |
4222 | struct packet_type *pt = NULL; | |
4223 | loff_t i = 0; | |
4224 | int t; | |
4225 | ||
4226 | list_for_each_entry_rcu(pt, &ptype_all, list) { | |
4227 | if (i == pos) | |
4228 | return pt; | |
4229 | ++i; | |
4230 | } | |
4231 | ||
82d8a867 | 4232 | for (t = 0; t < PTYPE_HASH_SIZE; t++) { |
0e1256ff SH |
4233 | list_for_each_entry_rcu(pt, &ptype_base[t], list) { |
4234 | if (i == pos) | |
4235 | return pt; | |
4236 | ++i; | |
4237 | } | |
4238 | } | |
4239 | return NULL; | |
4240 | } | |
4241 | ||
4242 | static void *ptype_seq_start(struct seq_file *seq, loff_t *pos) | |
72348a42 | 4243 | __acquires(RCU) |
0e1256ff SH |
4244 | { |
4245 | rcu_read_lock(); | |
4246 | return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN; | |
4247 | } | |
4248 | ||
4249 | static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
4250 | { | |
4251 | struct packet_type *pt; | |
4252 | struct list_head *nxt; | |
4253 | int hash; | |
4254 | ||
4255 | ++*pos; | |
4256 | if (v == SEQ_START_TOKEN) | |
4257 | return ptype_get_idx(0); | |
4258 | ||
4259 | pt = v; | |
4260 | nxt = pt->list.next; | |
4261 | if (pt->type == htons(ETH_P_ALL)) { | |
4262 | if (nxt != &ptype_all) | |
4263 | goto found; | |
4264 | hash = 0; | |
4265 | nxt = ptype_base[0].next; | |
4266 | } else | |
82d8a867 | 4267 | hash = ntohs(pt->type) & PTYPE_HASH_MASK; |
0e1256ff SH |
4268 | |
4269 | while (nxt == &ptype_base[hash]) { | |
82d8a867 | 4270 | if (++hash >= PTYPE_HASH_SIZE) |
0e1256ff SH |
4271 | return NULL; |
4272 | nxt = ptype_base[hash].next; | |
4273 | } | |
4274 | found: | |
4275 | return list_entry(nxt, struct packet_type, list); | |
4276 | } | |
4277 | ||
4278 | static void ptype_seq_stop(struct seq_file *seq, void *v) | |
72348a42 | 4279 | __releases(RCU) |
0e1256ff SH |
4280 | { |
4281 | rcu_read_unlock(); | |
4282 | } | |
4283 | ||
0e1256ff SH |
4284 | static int ptype_seq_show(struct seq_file *seq, void *v) |
4285 | { | |
4286 | struct packet_type *pt = v; | |
4287 | ||
4288 | if (v == SEQ_START_TOKEN) | |
4289 | seq_puts(seq, "Type Device Function\n"); | |
c346dca1 | 4290 | else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) { |
0e1256ff SH |
4291 | if (pt->type == htons(ETH_P_ALL)) |
4292 | seq_puts(seq, "ALL "); | |
4293 | else | |
4294 | seq_printf(seq, "%04x", ntohs(pt->type)); | |
4295 | ||
908cd2da AD |
4296 | seq_printf(seq, " %-8s %pF\n", |
4297 | pt->dev ? pt->dev->name : "", pt->func); | |
0e1256ff SH |
4298 | } |
4299 | ||
4300 | return 0; | |
4301 | } | |
4302 | ||
4303 | static const struct seq_operations ptype_seq_ops = { | |
4304 | .start = ptype_seq_start, | |
4305 | .next = ptype_seq_next, | |
4306 | .stop = ptype_seq_stop, | |
4307 | .show = ptype_seq_show, | |
4308 | }; | |
4309 | ||
4310 | static int ptype_seq_open(struct inode *inode, struct file *file) | |
4311 | { | |
2feb27db PE |
4312 | return seq_open_net(inode, file, &ptype_seq_ops, |
4313 | sizeof(struct seq_net_private)); | |
0e1256ff SH |
4314 | } |
4315 | ||
4316 | static const struct file_operations ptype_seq_fops = { | |
4317 | .owner = THIS_MODULE, | |
4318 | .open = ptype_seq_open, | |
4319 | .read = seq_read, | |
4320 | .llseek = seq_lseek, | |
2feb27db | 4321 | .release = seq_release_net, |
0e1256ff SH |
4322 | }; |
4323 | ||
4324 | ||
4665079c | 4325 | static int __net_init dev_proc_net_init(struct net *net) |
1da177e4 LT |
4326 | { |
4327 | int rc = -ENOMEM; | |
4328 | ||
881d966b | 4329 | if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops)) |
1da177e4 | 4330 | goto out; |
881d966b | 4331 | if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops)) |
1da177e4 | 4332 | goto out_dev; |
881d966b | 4333 | if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops)) |
457c4cbc | 4334 | goto out_softnet; |
0e1256ff | 4335 | |
881d966b | 4336 | if (wext_proc_init(net)) |
457c4cbc | 4337 | goto out_ptype; |
1da177e4 LT |
4338 | rc = 0; |
4339 | out: | |
4340 | return rc; | |
457c4cbc | 4341 | out_ptype: |
881d966b | 4342 | proc_net_remove(net, "ptype"); |
1da177e4 | 4343 | out_softnet: |
881d966b | 4344 | proc_net_remove(net, "softnet_stat"); |
1da177e4 | 4345 | out_dev: |
881d966b | 4346 | proc_net_remove(net, "dev"); |
1da177e4 LT |
4347 | goto out; |
4348 | } | |
881d966b | 4349 | |
4665079c | 4350 | static void __net_exit dev_proc_net_exit(struct net *net) |
881d966b EB |
4351 | { |
4352 | wext_proc_exit(net); | |
4353 | ||
4354 | proc_net_remove(net, "ptype"); | |
4355 | proc_net_remove(net, "softnet_stat"); | |
4356 | proc_net_remove(net, "dev"); | |
4357 | } | |
4358 | ||
022cbae6 | 4359 | static struct pernet_operations __net_initdata dev_proc_ops = { |
881d966b EB |
4360 | .init = dev_proc_net_init, |
4361 | .exit = dev_proc_net_exit, | |
4362 | }; | |
4363 | ||
4364 | static int __init dev_proc_init(void) | |
4365 | { | |
4366 | return register_pernet_subsys(&dev_proc_ops); | |
4367 | } | |
1da177e4 LT |
4368 | #else |
4369 | #define dev_proc_init() 0 | |
4370 | #endif /* CONFIG_PROC_FS */ | |
4371 | ||
4372 | ||
4373 | /** | |
1765a575 | 4374 | * netdev_set_master - set up master pointer |
1da177e4 LT |
4375 | * @slave: slave device |
4376 | * @master: new master device | |
4377 | * | |
4378 | * Changes the master device of the slave. Pass %NULL to break the | |
4379 | * bonding. The caller must hold the RTNL semaphore. On a failure | |
4380 | * a negative errno code is returned. On success the reference counts | |
1765a575 | 4381 | * are adjusted and the function returns zero. |
1da177e4 LT |
4382 | */ |
4383 | int netdev_set_master(struct net_device *slave, struct net_device *master) | |
4384 | { | |
4385 | struct net_device *old = slave->master; | |
4386 | ||
4387 | ASSERT_RTNL(); | |
4388 | ||
4389 | if (master) { | |
4390 | if (old) | |
4391 | return -EBUSY; | |
4392 | dev_hold(master); | |
4393 | } | |
4394 | ||
4395 | slave->master = master; | |
4ec93edb | 4396 | |
6df427fe | 4397 | if (old) |
1da177e4 | 4398 | dev_put(old); |
1765a575 JP |
4399 | return 0; |
4400 | } | |
4401 | EXPORT_SYMBOL(netdev_set_master); | |
4402 | ||
4403 | /** | |
4404 | * netdev_set_bond_master - set up bonding master/slave pair | |
4405 | * @slave: slave device | |
4406 | * @master: new master device | |
4407 | * | |
4408 | * Changes the master device of the slave. Pass %NULL to break the | |
4409 | * bonding. The caller must hold the RTNL semaphore. On a failure | |
4410 | * a negative errno code is returned. On success %RTM_NEWLINK is sent | |
4411 | * to the routing socket and the function returns zero. | |
4412 | */ | |
4413 | int netdev_set_bond_master(struct net_device *slave, struct net_device *master) | |
4414 | { | |
4415 | int err; | |
4416 | ||
4417 | ASSERT_RTNL(); | |
4418 | ||
4419 | err = netdev_set_master(slave, master); | |
4420 | if (err) | |
4421 | return err; | |
1da177e4 LT |
4422 | if (master) |
4423 | slave->flags |= IFF_SLAVE; | |
4424 | else | |
4425 | slave->flags &= ~IFF_SLAVE; | |
4426 | ||
4427 | rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE); | |
4428 | return 0; | |
4429 | } | |
1765a575 | 4430 | EXPORT_SYMBOL(netdev_set_bond_master); |
1da177e4 | 4431 | |
b6c40d68 PM |
4432 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
4433 | { | |
d314774c SH |
4434 | const struct net_device_ops *ops = dev->netdev_ops; |
4435 | ||
4436 | if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags) | |
4437 | ops->ndo_change_rx_flags(dev, flags); | |
b6c40d68 PM |
4438 | } |
4439 | ||
dad9b335 | 4440 | static int __dev_set_promiscuity(struct net_device *dev, int inc) |
1da177e4 | 4441 | { |
b536db93 | 4442 | unsigned int old_flags = dev->flags; |
8192b0c4 DH |
4443 | uid_t uid; |
4444 | gid_t gid; | |
1da177e4 | 4445 | |
24023451 PM |
4446 | ASSERT_RTNL(); |
4447 | ||
dad9b335 WC |
4448 | dev->flags |= IFF_PROMISC; |
4449 | dev->promiscuity += inc; | |
4450 | if (dev->promiscuity == 0) { | |
4451 | /* | |
4452 | * Avoid overflow. | |
4453 | * If inc causes overflow, untouch promisc and return error. | |
4454 | */ | |
4455 | if (inc < 0) | |
4456 | dev->flags &= ~IFF_PROMISC; | |
4457 | else { | |
4458 | dev->promiscuity -= inc; | |
7b6cd1ce JP |
4459 | pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", |
4460 | dev->name); | |
dad9b335 WC |
4461 | return -EOVERFLOW; |
4462 | } | |
4463 | } | |
52609c0b | 4464 | if (dev->flags != old_flags) { |
7b6cd1ce JP |
4465 | pr_info("device %s %s promiscuous mode\n", |
4466 | dev->name, | |
4467 | dev->flags & IFF_PROMISC ? "entered" : "left"); | |
8192b0c4 DH |
4468 | if (audit_enabled) { |
4469 | current_uid_gid(&uid, &gid); | |
7759db82 KHK |
4470 | audit_log(current->audit_context, GFP_ATOMIC, |
4471 | AUDIT_ANOM_PROMISCUOUS, | |
4472 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
4473 | dev->name, (dev->flags & IFF_PROMISC), | |
4474 | (old_flags & IFF_PROMISC), | |
4475 | audit_get_loginuid(current), | |
8192b0c4 | 4476 | uid, gid, |
7759db82 | 4477 | audit_get_sessionid(current)); |
8192b0c4 | 4478 | } |
24023451 | 4479 | |
b6c40d68 | 4480 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 4481 | } |
dad9b335 | 4482 | return 0; |
1da177e4 LT |
4483 | } |
4484 | ||
4417da66 PM |
4485 | /** |
4486 | * dev_set_promiscuity - update promiscuity count on a device | |
4487 | * @dev: device | |
4488 | * @inc: modifier | |
4489 | * | |
4490 | * Add or remove promiscuity from a device. While the count in the device | |
4491 | * remains above zero the interface remains promiscuous. Once it hits zero | |
4492 | * the device reverts back to normal filtering operation. A negative inc | |
4493 | * value is used to drop promiscuity on the device. | |
dad9b335 | 4494 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 4495 | */ |
dad9b335 | 4496 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 | 4497 | { |
b536db93 | 4498 | unsigned int old_flags = dev->flags; |
dad9b335 | 4499 | int err; |
4417da66 | 4500 | |
dad9b335 | 4501 | err = __dev_set_promiscuity(dev, inc); |
4b5a698e | 4502 | if (err < 0) |
dad9b335 | 4503 | return err; |
4417da66 PM |
4504 | if (dev->flags != old_flags) |
4505 | dev_set_rx_mode(dev); | |
dad9b335 | 4506 | return err; |
4417da66 | 4507 | } |
d1b19dff | 4508 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 4509 | |
1da177e4 LT |
4510 | /** |
4511 | * dev_set_allmulti - update allmulti count on a device | |
4512 | * @dev: device | |
4513 | * @inc: modifier | |
4514 | * | |
4515 | * Add or remove reception of all multicast frames to a device. While the | |
4516 | * count in the device remains above zero the interface remains listening | |
4517 | * to all interfaces. Once it hits zero the device reverts back to normal | |
4518 | * filtering operation. A negative @inc value is used to drop the counter | |
4519 | * when releasing a resource needing all multicasts. | |
dad9b335 | 4520 | * Return 0 if successful or a negative errno code on error. |
1da177e4 LT |
4521 | */ |
4522 | ||
dad9b335 | 4523 | int dev_set_allmulti(struct net_device *dev, int inc) |
1da177e4 | 4524 | { |
b536db93 | 4525 | unsigned int old_flags = dev->flags; |
1da177e4 | 4526 | |
24023451 PM |
4527 | ASSERT_RTNL(); |
4528 | ||
1da177e4 | 4529 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
4530 | dev->allmulti += inc; |
4531 | if (dev->allmulti == 0) { | |
4532 | /* | |
4533 | * Avoid overflow. | |
4534 | * If inc causes overflow, untouch allmulti and return error. | |
4535 | */ | |
4536 | if (inc < 0) | |
4537 | dev->flags &= ~IFF_ALLMULTI; | |
4538 | else { | |
4539 | dev->allmulti -= inc; | |
7b6cd1ce JP |
4540 | pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", |
4541 | dev->name); | |
dad9b335 WC |
4542 | return -EOVERFLOW; |
4543 | } | |
4544 | } | |
24023451 | 4545 | if (dev->flags ^ old_flags) { |
b6c40d68 | 4546 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 4547 | dev_set_rx_mode(dev); |
24023451 | 4548 | } |
dad9b335 | 4549 | return 0; |
4417da66 | 4550 | } |
d1b19dff | 4551 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
4552 | |
4553 | /* | |
4554 | * Upload unicast and multicast address lists to device and | |
4555 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 4556 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
4557 | * are present. |
4558 | */ | |
4559 | void __dev_set_rx_mode(struct net_device *dev) | |
4560 | { | |
d314774c SH |
4561 | const struct net_device_ops *ops = dev->netdev_ops; |
4562 | ||
4417da66 PM |
4563 | /* dev_open will call this function so the list will stay sane. */ |
4564 | if (!(dev->flags&IFF_UP)) | |
4565 | return; | |
4566 | ||
4567 | if (!netif_device_present(dev)) | |
40b77c94 | 4568 | return; |
4417da66 | 4569 | |
01789349 | 4570 | if (!(dev->priv_flags & IFF_UNICAST_FLT)) { |
4417da66 PM |
4571 | /* Unicast addresses changes may only happen under the rtnl, |
4572 | * therefore calling __dev_set_promiscuity here is safe. | |
4573 | */ | |
32e7bfc4 | 4574 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
4417da66 | 4575 | __dev_set_promiscuity(dev, 1); |
2d348d1f | 4576 | dev->uc_promisc = true; |
32e7bfc4 | 4577 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
4417da66 | 4578 | __dev_set_promiscuity(dev, -1); |
2d348d1f | 4579 | dev->uc_promisc = false; |
4417da66 | 4580 | } |
4417da66 | 4581 | } |
01789349 JP |
4582 | |
4583 | if (ops->ndo_set_rx_mode) | |
4584 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
4585 | } |
4586 | ||
4587 | void dev_set_rx_mode(struct net_device *dev) | |
4588 | { | |
b9e40857 | 4589 | netif_addr_lock_bh(dev); |
4417da66 | 4590 | __dev_set_rx_mode(dev); |
b9e40857 | 4591 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
4592 | } |
4593 | ||
f0db275a SH |
4594 | /** |
4595 | * dev_get_flags - get flags reported to userspace | |
4596 | * @dev: device | |
4597 | * | |
4598 | * Get the combination of flag bits exported through APIs to userspace. | |
4599 | */ | |
95c96174 | 4600 | unsigned int dev_get_flags(const struct net_device *dev) |
1da177e4 | 4601 | { |
95c96174 | 4602 | unsigned int flags; |
1da177e4 LT |
4603 | |
4604 | flags = (dev->flags & ~(IFF_PROMISC | | |
4605 | IFF_ALLMULTI | | |
b00055aa SR |
4606 | IFF_RUNNING | |
4607 | IFF_LOWER_UP | | |
4608 | IFF_DORMANT)) | | |
1da177e4 LT |
4609 | (dev->gflags & (IFF_PROMISC | |
4610 | IFF_ALLMULTI)); | |
4611 | ||
b00055aa SR |
4612 | if (netif_running(dev)) { |
4613 | if (netif_oper_up(dev)) | |
4614 | flags |= IFF_RUNNING; | |
4615 | if (netif_carrier_ok(dev)) | |
4616 | flags |= IFF_LOWER_UP; | |
4617 | if (netif_dormant(dev)) | |
4618 | flags |= IFF_DORMANT; | |
4619 | } | |
1da177e4 LT |
4620 | |
4621 | return flags; | |
4622 | } | |
d1b19dff | 4623 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 4624 | |
bd380811 | 4625 | int __dev_change_flags(struct net_device *dev, unsigned int flags) |
1da177e4 | 4626 | { |
b536db93 | 4627 | unsigned int old_flags = dev->flags; |
bd380811 | 4628 | int ret; |
1da177e4 | 4629 | |
24023451 PM |
4630 | ASSERT_RTNL(); |
4631 | ||
1da177e4 LT |
4632 | /* |
4633 | * Set the flags on our device. | |
4634 | */ | |
4635 | ||
4636 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
4637 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
4638 | IFF_AUTOMEDIA)) | | |
4639 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
4640 | IFF_ALLMULTI)); | |
4641 | ||
4642 | /* | |
4643 | * Load in the correct multicast list now the flags have changed. | |
4644 | */ | |
4645 | ||
b6c40d68 PM |
4646 | if ((old_flags ^ flags) & IFF_MULTICAST) |
4647 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 4648 | |
4417da66 | 4649 | dev_set_rx_mode(dev); |
1da177e4 LT |
4650 | |
4651 | /* | |
4652 | * Have we downed the interface. We handle IFF_UP ourselves | |
4653 | * according to user attempts to set it, rather than blindly | |
4654 | * setting it. | |
4655 | */ | |
4656 | ||
4657 | ret = 0; | |
4658 | if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */ | |
bd380811 | 4659 | ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev); |
1da177e4 LT |
4660 | |
4661 | if (!ret) | |
4417da66 | 4662 | dev_set_rx_mode(dev); |
1da177e4 LT |
4663 | } |
4664 | ||
1da177e4 | 4665 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff ED |
4666 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
4667 | ||
1da177e4 LT |
4668 | dev->gflags ^= IFF_PROMISC; |
4669 | dev_set_promiscuity(dev, inc); | |
4670 | } | |
4671 | ||
4672 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
4673 | is important. Some (broken) drivers set IFF_PROMISC, when | |
4674 | IFF_ALLMULTI is requested not asking us and not reporting. | |
4675 | */ | |
4676 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
4677 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
4678 | ||
1da177e4 LT |
4679 | dev->gflags ^= IFF_ALLMULTI; |
4680 | dev_set_allmulti(dev, inc); | |
4681 | } | |
4682 | ||
bd380811 PM |
4683 | return ret; |
4684 | } | |
4685 | ||
4686 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags) | |
4687 | { | |
4688 | unsigned int changes = dev->flags ^ old_flags; | |
4689 | ||
4690 | if (changes & IFF_UP) { | |
4691 | if (dev->flags & IFF_UP) | |
4692 | call_netdevice_notifiers(NETDEV_UP, dev); | |
4693 | else | |
4694 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
4695 | } | |
4696 | ||
4697 | if (dev->flags & IFF_UP && | |
4698 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) | |
4699 | call_netdevice_notifiers(NETDEV_CHANGE, dev); | |
4700 | } | |
4701 | ||
4702 | /** | |
4703 | * dev_change_flags - change device settings | |
4704 | * @dev: device | |
4705 | * @flags: device state flags | |
4706 | * | |
4707 | * Change settings on device based state flags. The flags are | |
4708 | * in the userspace exported format. | |
4709 | */ | |
b536db93 | 4710 | int dev_change_flags(struct net_device *dev, unsigned int flags) |
bd380811 | 4711 | { |
b536db93 ED |
4712 | int ret; |
4713 | unsigned int changes, old_flags = dev->flags; | |
bd380811 PM |
4714 | |
4715 | ret = __dev_change_flags(dev, flags); | |
4716 | if (ret < 0) | |
4717 | return ret; | |
4718 | ||
4719 | changes = old_flags ^ dev->flags; | |
7c355f53 TG |
4720 | if (changes) |
4721 | rtmsg_ifinfo(RTM_NEWLINK, dev, changes); | |
1da177e4 | 4722 | |
bd380811 | 4723 | __dev_notify_flags(dev, old_flags); |
1da177e4 LT |
4724 | return ret; |
4725 | } | |
d1b19dff | 4726 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 4727 | |
f0db275a SH |
4728 | /** |
4729 | * dev_set_mtu - Change maximum transfer unit | |
4730 | * @dev: device | |
4731 | * @new_mtu: new transfer unit | |
4732 | * | |
4733 | * Change the maximum transfer size of the network device. | |
4734 | */ | |
1da177e4 LT |
4735 | int dev_set_mtu(struct net_device *dev, int new_mtu) |
4736 | { | |
d314774c | 4737 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
4738 | int err; |
4739 | ||
4740 | if (new_mtu == dev->mtu) | |
4741 | return 0; | |
4742 | ||
4743 | /* MTU must be positive. */ | |
4744 | if (new_mtu < 0) | |
4745 | return -EINVAL; | |
4746 | ||
4747 | if (!netif_device_present(dev)) | |
4748 | return -ENODEV; | |
4749 | ||
4750 | err = 0; | |
d314774c SH |
4751 | if (ops->ndo_change_mtu) |
4752 | err = ops->ndo_change_mtu(dev, new_mtu); | |
1da177e4 LT |
4753 | else |
4754 | dev->mtu = new_mtu; | |
d314774c | 4755 | |
1da177e4 | 4756 | if (!err && dev->flags & IFF_UP) |
056925ab | 4757 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); |
1da177e4 LT |
4758 | return err; |
4759 | } | |
d1b19dff | 4760 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 4761 | |
cbda10fa VD |
4762 | /** |
4763 | * dev_set_group - Change group this device belongs to | |
4764 | * @dev: device | |
4765 | * @new_group: group this device should belong to | |
4766 | */ | |
4767 | void dev_set_group(struct net_device *dev, int new_group) | |
4768 | { | |
4769 | dev->group = new_group; | |
4770 | } | |
4771 | EXPORT_SYMBOL(dev_set_group); | |
4772 | ||
f0db275a SH |
4773 | /** |
4774 | * dev_set_mac_address - Change Media Access Control Address | |
4775 | * @dev: device | |
4776 | * @sa: new address | |
4777 | * | |
4778 | * Change the hardware (MAC) address of the device | |
4779 | */ | |
1da177e4 LT |
4780 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) |
4781 | { | |
d314774c | 4782 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
4783 | int err; |
4784 | ||
d314774c | 4785 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
4786 | return -EOPNOTSUPP; |
4787 | if (sa->sa_family != dev->type) | |
4788 | return -EINVAL; | |
4789 | if (!netif_device_present(dev)) | |
4790 | return -ENODEV; | |
d314774c | 4791 | err = ops->ndo_set_mac_address(dev, sa); |
1da177e4 | 4792 | if (!err) |
056925ab | 4793 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
1da177e4 LT |
4794 | return err; |
4795 | } | |
d1b19dff | 4796 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 LT |
4797 | |
4798 | /* | |
3710becf | 4799 | * Perform the SIOCxIFxxx calls, inside rcu_read_lock() |
1da177e4 | 4800 | */ |
14e3e079 | 4801 | static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd) |
1da177e4 LT |
4802 | { |
4803 | int err; | |
3710becf | 4804 | struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name); |
1da177e4 LT |
4805 | |
4806 | if (!dev) | |
4807 | return -ENODEV; | |
4808 | ||
4809 | switch (cmd) { | |
d1b19dff ED |
4810 | case SIOCGIFFLAGS: /* Get interface flags */ |
4811 | ifr->ifr_flags = (short) dev_get_flags(dev); | |
4812 | return 0; | |
1da177e4 | 4813 | |
d1b19dff ED |
4814 | case SIOCGIFMETRIC: /* Get the metric on the interface |
4815 | (currently unused) */ | |
4816 | ifr->ifr_metric = 0; | |
4817 | return 0; | |
1da177e4 | 4818 | |
d1b19dff ED |
4819 | case SIOCGIFMTU: /* Get the MTU of a device */ |
4820 | ifr->ifr_mtu = dev->mtu; | |
4821 | return 0; | |
1da177e4 | 4822 | |
d1b19dff ED |
4823 | case SIOCGIFHWADDR: |
4824 | if (!dev->addr_len) | |
4825 | memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data); | |
4826 | else | |
4827 | memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr, | |
4828 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
4829 | ifr->ifr_hwaddr.sa_family = dev->type; | |
4830 | return 0; | |
1da177e4 | 4831 | |
d1b19dff ED |
4832 | case SIOCGIFSLAVE: |
4833 | err = -EINVAL; | |
4834 | break; | |
14e3e079 | 4835 | |
d1b19dff ED |
4836 | case SIOCGIFMAP: |
4837 | ifr->ifr_map.mem_start = dev->mem_start; | |
4838 | ifr->ifr_map.mem_end = dev->mem_end; | |
4839 | ifr->ifr_map.base_addr = dev->base_addr; | |
4840 | ifr->ifr_map.irq = dev->irq; | |
4841 | ifr->ifr_map.dma = dev->dma; | |
4842 | ifr->ifr_map.port = dev->if_port; | |
4843 | return 0; | |
14e3e079 | 4844 | |
d1b19dff ED |
4845 | case SIOCGIFINDEX: |
4846 | ifr->ifr_ifindex = dev->ifindex; | |
4847 | return 0; | |
14e3e079 | 4848 | |
d1b19dff ED |
4849 | case SIOCGIFTXQLEN: |
4850 | ifr->ifr_qlen = dev->tx_queue_len; | |
4851 | return 0; | |
14e3e079 | 4852 | |
d1b19dff ED |
4853 | default: |
4854 | /* dev_ioctl() should ensure this case | |
4855 | * is never reached | |
4856 | */ | |
4857 | WARN_ON(1); | |
41c31f31 | 4858 | err = -ENOTTY; |
d1b19dff | 4859 | break; |
14e3e079 JG |
4860 | |
4861 | } | |
4862 | return err; | |
4863 | } | |
4864 | ||
4865 | /* | |
4866 | * Perform the SIOCxIFxxx calls, inside rtnl_lock() | |
4867 | */ | |
4868 | static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd) | |
4869 | { | |
4870 | int err; | |
4871 | struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name); | |
5f2f6da7 | 4872 | const struct net_device_ops *ops; |
14e3e079 JG |
4873 | |
4874 | if (!dev) | |
4875 | return -ENODEV; | |
4876 | ||
5f2f6da7 JP |
4877 | ops = dev->netdev_ops; |
4878 | ||
14e3e079 | 4879 | switch (cmd) { |
d1b19dff ED |
4880 | case SIOCSIFFLAGS: /* Set interface flags */ |
4881 | return dev_change_flags(dev, ifr->ifr_flags); | |
14e3e079 | 4882 | |
d1b19dff ED |
4883 | case SIOCSIFMETRIC: /* Set the metric on the interface |
4884 | (currently unused) */ | |
4885 | return -EOPNOTSUPP; | |
14e3e079 | 4886 | |
d1b19dff ED |
4887 | case SIOCSIFMTU: /* Set the MTU of a device */ |
4888 | return dev_set_mtu(dev, ifr->ifr_mtu); | |
1da177e4 | 4889 | |
d1b19dff ED |
4890 | case SIOCSIFHWADDR: |
4891 | return dev_set_mac_address(dev, &ifr->ifr_hwaddr); | |
1da177e4 | 4892 | |
d1b19dff ED |
4893 | case SIOCSIFHWBROADCAST: |
4894 | if (ifr->ifr_hwaddr.sa_family != dev->type) | |
4895 | return -EINVAL; | |
4896 | memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data, | |
4897 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
4898 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); | |
4899 | return 0; | |
1da177e4 | 4900 | |
d1b19dff ED |
4901 | case SIOCSIFMAP: |
4902 | if (ops->ndo_set_config) { | |
1da177e4 LT |
4903 | if (!netif_device_present(dev)) |
4904 | return -ENODEV; | |
d1b19dff ED |
4905 | return ops->ndo_set_config(dev, &ifr->ifr_map); |
4906 | } | |
4907 | return -EOPNOTSUPP; | |
1da177e4 | 4908 | |
d1b19dff | 4909 | case SIOCADDMULTI: |
b81693d9 | 4910 | if (!ops->ndo_set_rx_mode || |
d1b19dff ED |
4911 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) |
4912 | return -EINVAL; | |
4913 | if (!netif_device_present(dev)) | |
4914 | return -ENODEV; | |
22bedad3 | 4915 | return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data); |
d1b19dff ED |
4916 | |
4917 | case SIOCDELMULTI: | |
b81693d9 | 4918 | if (!ops->ndo_set_rx_mode || |
d1b19dff ED |
4919 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) |
4920 | return -EINVAL; | |
4921 | if (!netif_device_present(dev)) | |
4922 | return -ENODEV; | |
22bedad3 | 4923 | return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data); |
1da177e4 | 4924 | |
d1b19dff ED |
4925 | case SIOCSIFTXQLEN: |
4926 | if (ifr->ifr_qlen < 0) | |
4927 | return -EINVAL; | |
4928 | dev->tx_queue_len = ifr->ifr_qlen; | |
4929 | return 0; | |
1da177e4 | 4930 | |
d1b19dff ED |
4931 | case SIOCSIFNAME: |
4932 | ifr->ifr_newname[IFNAMSIZ-1] = '\0'; | |
4933 | return dev_change_name(dev, ifr->ifr_newname); | |
1da177e4 | 4934 | |
4dc360c5 RC |
4935 | case SIOCSHWTSTAMP: |
4936 | err = net_hwtstamp_validate(ifr); | |
4937 | if (err) | |
4938 | return err; | |
4939 | /* fall through */ | |
4940 | ||
d1b19dff ED |
4941 | /* |
4942 | * Unknown or private ioctl | |
4943 | */ | |
4944 | default: | |
4945 | if ((cmd >= SIOCDEVPRIVATE && | |
4946 | cmd <= SIOCDEVPRIVATE + 15) || | |
4947 | cmd == SIOCBONDENSLAVE || | |
4948 | cmd == SIOCBONDRELEASE || | |
4949 | cmd == SIOCBONDSETHWADDR || | |
4950 | cmd == SIOCBONDSLAVEINFOQUERY || | |
4951 | cmd == SIOCBONDINFOQUERY || | |
4952 | cmd == SIOCBONDCHANGEACTIVE || | |
4953 | cmd == SIOCGMIIPHY || | |
4954 | cmd == SIOCGMIIREG || | |
4955 | cmd == SIOCSMIIREG || | |
4956 | cmd == SIOCBRADDIF || | |
4957 | cmd == SIOCBRDELIF || | |
4958 | cmd == SIOCSHWTSTAMP || | |
4959 | cmd == SIOCWANDEV) { | |
4960 | err = -EOPNOTSUPP; | |
4961 | if (ops->ndo_do_ioctl) { | |
4962 | if (netif_device_present(dev)) | |
4963 | err = ops->ndo_do_ioctl(dev, ifr, cmd); | |
4964 | else | |
4965 | err = -ENODEV; | |
4966 | } | |
4967 | } else | |
4968 | err = -EINVAL; | |
1da177e4 LT |
4969 | |
4970 | } | |
4971 | return err; | |
4972 | } | |
4973 | ||
4974 | /* | |
4975 | * This function handles all "interface"-type I/O control requests. The actual | |
4976 | * 'doing' part of this is dev_ifsioc above. | |
4977 | */ | |
4978 | ||
4979 | /** | |
4980 | * dev_ioctl - network device ioctl | |
c4ea43c5 | 4981 | * @net: the applicable net namespace |
1da177e4 LT |
4982 | * @cmd: command to issue |
4983 | * @arg: pointer to a struct ifreq in user space | |
4984 | * | |
4985 | * Issue ioctl functions to devices. This is normally called by the | |
4986 | * user space syscall interfaces but can sometimes be useful for | |
4987 | * other purposes. The return value is the return from the syscall if | |
4988 | * positive or a negative errno code on error. | |
4989 | */ | |
4990 | ||
881d966b | 4991 | int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg) |
1da177e4 LT |
4992 | { |
4993 | struct ifreq ifr; | |
4994 | int ret; | |
4995 | char *colon; | |
4996 | ||
4997 | /* One special case: SIOCGIFCONF takes ifconf argument | |
4998 | and requires shared lock, because it sleeps writing | |
4999 | to user space. | |
5000 | */ | |
5001 | ||
5002 | if (cmd == SIOCGIFCONF) { | |
6756ae4b | 5003 | rtnl_lock(); |
881d966b | 5004 | ret = dev_ifconf(net, (char __user *) arg); |
6756ae4b | 5005 | rtnl_unlock(); |
1da177e4 LT |
5006 | return ret; |
5007 | } | |
5008 | if (cmd == SIOCGIFNAME) | |
881d966b | 5009 | return dev_ifname(net, (struct ifreq __user *)arg); |
1da177e4 LT |
5010 | |
5011 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
5012 | return -EFAULT; | |
5013 | ||
5014 | ifr.ifr_name[IFNAMSIZ-1] = 0; | |
5015 | ||
5016 | colon = strchr(ifr.ifr_name, ':'); | |
5017 | if (colon) | |
5018 | *colon = 0; | |
5019 | ||
5020 | /* | |
5021 | * See which interface the caller is talking about. | |
5022 | */ | |
5023 | ||
5024 | switch (cmd) { | |
d1b19dff ED |
5025 | /* |
5026 | * These ioctl calls: | |
5027 | * - can be done by all. | |
5028 | * - atomic and do not require locking. | |
5029 | * - return a value | |
5030 | */ | |
5031 | case SIOCGIFFLAGS: | |
5032 | case SIOCGIFMETRIC: | |
5033 | case SIOCGIFMTU: | |
5034 | case SIOCGIFHWADDR: | |
5035 | case SIOCGIFSLAVE: | |
5036 | case SIOCGIFMAP: | |
5037 | case SIOCGIFINDEX: | |
5038 | case SIOCGIFTXQLEN: | |
5039 | dev_load(net, ifr.ifr_name); | |
3710becf | 5040 | rcu_read_lock(); |
d1b19dff | 5041 | ret = dev_ifsioc_locked(net, &ifr, cmd); |
3710becf | 5042 | rcu_read_unlock(); |
d1b19dff ED |
5043 | if (!ret) { |
5044 | if (colon) | |
5045 | *colon = ':'; | |
5046 | if (copy_to_user(arg, &ifr, | |
5047 | sizeof(struct ifreq))) | |
5048 | ret = -EFAULT; | |
5049 | } | |
5050 | return ret; | |
1da177e4 | 5051 | |
d1b19dff ED |
5052 | case SIOCETHTOOL: |
5053 | dev_load(net, ifr.ifr_name); | |
5054 | rtnl_lock(); | |
5055 | ret = dev_ethtool(net, &ifr); | |
5056 | rtnl_unlock(); | |
5057 | if (!ret) { | |
5058 | if (colon) | |
5059 | *colon = ':'; | |
5060 | if (copy_to_user(arg, &ifr, | |
5061 | sizeof(struct ifreq))) | |
5062 | ret = -EFAULT; | |
5063 | } | |
5064 | return ret; | |
1da177e4 | 5065 | |
d1b19dff ED |
5066 | /* |
5067 | * These ioctl calls: | |
5068 | * - require superuser power. | |
5069 | * - require strict serialization. | |
5070 | * - return a value | |
5071 | */ | |
5072 | case SIOCGMIIPHY: | |
5073 | case SIOCGMIIREG: | |
5074 | case SIOCSIFNAME: | |
5075 | if (!capable(CAP_NET_ADMIN)) | |
5076 | return -EPERM; | |
5077 | dev_load(net, ifr.ifr_name); | |
5078 | rtnl_lock(); | |
5079 | ret = dev_ifsioc(net, &ifr, cmd); | |
5080 | rtnl_unlock(); | |
5081 | if (!ret) { | |
5082 | if (colon) | |
5083 | *colon = ':'; | |
5084 | if (copy_to_user(arg, &ifr, | |
5085 | sizeof(struct ifreq))) | |
5086 | ret = -EFAULT; | |
5087 | } | |
5088 | return ret; | |
1da177e4 | 5089 | |
d1b19dff ED |
5090 | /* |
5091 | * These ioctl calls: | |
5092 | * - require superuser power. | |
5093 | * - require strict serialization. | |
5094 | * - do not return a value | |
5095 | */ | |
5096 | case SIOCSIFFLAGS: | |
5097 | case SIOCSIFMETRIC: | |
5098 | case SIOCSIFMTU: | |
5099 | case SIOCSIFMAP: | |
5100 | case SIOCSIFHWADDR: | |
5101 | case SIOCSIFSLAVE: | |
5102 | case SIOCADDMULTI: | |
5103 | case SIOCDELMULTI: | |
5104 | case SIOCSIFHWBROADCAST: | |
5105 | case SIOCSIFTXQLEN: | |
5106 | case SIOCSMIIREG: | |
5107 | case SIOCBONDENSLAVE: | |
5108 | case SIOCBONDRELEASE: | |
5109 | case SIOCBONDSETHWADDR: | |
5110 | case SIOCBONDCHANGEACTIVE: | |
5111 | case SIOCBRADDIF: | |
5112 | case SIOCBRDELIF: | |
5113 | case SIOCSHWTSTAMP: | |
5114 | if (!capable(CAP_NET_ADMIN)) | |
5115 | return -EPERM; | |
5116 | /* fall through */ | |
5117 | case SIOCBONDSLAVEINFOQUERY: | |
5118 | case SIOCBONDINFOQUERY: | |
5119 | dev_load(net, ifr.ifr_name); | |
5120 | rtnl_lock(); | |
5121 | ret = dev_ifsioc(net, &ifr, cmd); | |
5122 | rtnl_unlock(); | |
5123 | return ret; | |
5124 | ||
5125 | case SIOCGIFMEM: | |
5126 | /* Get the per device memory space. We can add this but | |
5127 | * currently do not support it */ | |
5128 | case SIOCSIFMEM: | |
5129 | /* Set the per device memory buffer space. | |
5130 | * Not applicable in our case */ | |
5131 | case SIOCSIFLINK: | |
41c31f31 | 5132 | return -ENOTTY; |
d1b19dff ED |
5133 | |
5134 | /* | |
5135 | * Unknown or private ioctl. | |
5136 | */ | |
5137 | default: | |
5138 | if (cmd == SIOCWANDEV || | |
5139 | (cmd >= SIOCDEVPRIVATE && | |
5140 | cmd <= SIOCDEVPRIVATE + 15)) { | |
881d966b | 5141 | dev_load(net, ifr.ifr_name); |
1da177e4 | 5142 | rtnl_lock(); |
881d966b | 5143 | ret = dev_ifsioc(net, &ifr, cmd); |
1da177e4 | 5144 | rtnl_unlock(); |
d1b19dff ED |
5145 | if (!ret && copy_to_user(arg, &ifr, |
5146 | sizeof(struct ifreq))) | |
5147 | ret = -EFAULT; | |
1da177e4 | 5148 | return ret; |
d1b19dff ED |
5149 | } |
5150 | /* Take care of Wireless Extensions */ | |
5151 | if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) | |
5152 | return wext_handle_ioctl(net, &ifr, cmd, arg); | |
41c31f31 | 5153 | return -ENOTTY; |
1da177e4 LT |
5154 | } |
5155 | } | |
5156 | ||
5157 | ||
5158 | /** | |
5159 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 5160 | * @net: the applicable net namespace |
1da177e4 LT |
5161 | * |
5162 | * Returns a suitable unique value for a new device interface | |
5163 | * number. The caller must hold the rtnl semaphore or the | |
5164 | * dev_base_lock to be sure it remains unique. | |
5165 | */ | |
881d966b | 5166 | static int dev_new_index(struct net *net) |
1da177e4 LT |
5167 | { |
5168 | static int ifindex; | |
5169 | for (;;) { | |
5170 | if (++ifindex <= 0) | |
5171 | ifindex = 1; | |
881d966b | 5172 | if (!__dev_get_by_index(net, ifindex)) |
1da177e4 LT |
5173 | return ifindex; |
5174 | } | |
5175 | } | |
5176 | ||
1da177e4 | 5177 | /* Delayed registration/unregisteration */ |
3b5b34fd | 5178 | static LIST_HEAD(net_todo_list); |
1da177e4 | 5179 | |
6f05f629 | 5180 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 5181 | { |
1da177e4 | 5182 | list_add_tail(&dev->todo_list, &net_todo_list); |
1da177e4 LT |
5183 | } |
5184 | ||
9b5e383c | 5185 | static void rollback_registered_many(struct list_head *head) |
93ee31f1 | 5186 | { |
e93737b0 | 5187 | struct net_device *dev, *tmp; |
9b5e383c | 5188 | |
93ee31f1 DL |
5189 | BUG_ON(dev_boot_phase); |
5190 | ASSERT_RTNL(); | |
5191 | ||
e93737b0 | 5192 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
9b5e383c | 5193 | /* Some devices call without registering |
e93737b0 KK |
5194 | * for initialization unwind. Remove those |
5195 | * devices and proceed with the remaining. | |
9b5e383c ED |
5196 | */ |
5197 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
7b6cd1ce JP |
5198 | pr_debug("unregister_netdevice: device %s/%p never was registered\n", |
5199 | dev->name, dev); | |
93ee31f1 | 5200 | |
9b5e383c | 5201 | WARN_ON(1); |
e93737b0 KK |
5202 | list_del(&dev->unreg_list); |
5203 | continue; | |
9b5e383c | 5204 | } |
449f4544 | 5205 | dev->dismantle = true; |
9b5e383c | 5206 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
44345724 | 5207 | } |
93ee31f1 | 5208 | |
44345724 OP |
5209 | /* If device is running, close it first. */ |
5210 | dev_close_many(head); | |
93ee31f1 | 5211 | |
44345724 | 5212 | list_for_each_entry(dev, head, unreg_list) { |
9b5e383c ED |
5213 | /* And unlink it from device chain. */ |
5214 | unlist_netdevice(dev); | |
93ee31f1 | 5215 | |
9b5e383c ED |
5216 | dev->reg_state = NETREG_UNREGISTERING; |
5217 | } | |
93ee31f1 DL |
5218 | |
5219 | synchronize_net(); | |
5220 | ||
9b5e383c ED |
5221 | list_for_each_entry(dev, head, unreg_list) { |
5222 | /* Shutdown queueing discipline. */ | |
5223 | dev_shutdown(dev); | |
93ee31f1 DL |
5224 | |
5225 | ||
9b5e383c ED |
5226 | /* Notify protocols, that we are about to destroy |
5227 | this device. They should clean all the things. | |
5228 | */ | |
5229 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
93ee31f1 | 5230 | |
a2835763 PM |
5231 | if (!dev->rtnl_link_ops || |
5232 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
5233 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U); | |
5234 | ||
9b5e383c ED |
5235 | /* |
5236 | * Flush the unicast and multicast chains | |
5237 | */ | |
a748ee24 | 5238 | dev_uc_flush(dev); |
22bedad3 | 5239 | dev_mc_flush(dev); |
93ee31f1 | 5240 | |
9b5e383c ED |
5241 | if (dev->netdev_ops->ndo_uninit) |
5242 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 | 5243 | |
9b5e383c ED |
5244 | /* Notifier chain MUST detach us from master device. */ |
5245 | WARN_ON(dev->master); | |
93ee31f1 | 5246 | |
9b5e383c ED |
5247 | /* Remove entries from kobject tree */ |
5248 | netdev_unregister_kobject(dev); | |
5249 | } | |
93ee31f1 | 5250 | |
a5ee1551 | 5251 | /* Process any work delayed until the end of the batch */ |
e5e26d75 | 5252 | dev = list_first_entry(head, struct net_device, unreg_list); |
a5ee1551 | 5253 | call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev); |
93ee31f1 | 5254 | |
850a545b | 5255 | synchronize_net(); |
395264d5 | 5256 | |
a5ee1551 | 5257 | list_for_each_entry(dev, head, unreg_list) |
9b5e383c ED |
5258 | dev_put(dev); |
5259 | } | |
5260 | ||
5261 | static void rollback_registered(struct net_device *dev) | |
5262 | { | |
5263 | LIST_HEAD(single); | |
5264 | ||
5265 | list_add(&dev->unreg_list, &single); | |
5266 | rollback_registered_many(&single); | |
ceaaec98 | 5267 | list_del(&single); |
93ee31f1 DL |
5268 | } |
5269 | ||
c8f44aff MM |
5270 | static netdev_features_t netdev_fix_features(struct net_device *dev, |
5271 | netdev_features_t features) | |
b63365a2 | 5272 | { |
57422dc5 MM |
5273 | /* Fix illegal checksum combinations */ |
5274 | if ((features & NETIF_F_HW_CSUM) && | |
5275 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 5276 | netdev_warn(dev, "mixed HW and IP checksum settings.\n"); |
57422dc5 MM |
5277 | features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); |
5278 | } | |
5279 | ||
b63365a2 HX |
5280 | /* Fix illegal SG+CSUM combinations. */ |
5281 | if ((features & NETIF_F_SG) && | |
5282 | !(features & NETIF_F_ALL_CSUM)) { | |
6f404e44 MM |
5283 | netdev_dbg(dev, |
5284 | "Dropping NETIF_F_SG since no checksum feature.\n"); | |
b63365a2 HX |
5285 | features &= ~NETIF_F_SG; |
5286 | } | |
5287 | ||
5288 | /* TSO requires that SG is present as well. */ | |
ea2d3688 | 5289 | if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { |
6f404e44 | 5290 | netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); |
ea2d3688 | 5291 | features &= ~NETIF_F_ALL_TSO; |
b63365a2 HX |
5292 | } |
5293 | ||
31d8b9e0 BH |
5294 | /* TSO ECN requires that TSO is present as well. */ |
5295 | if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) | |
5296 | features &= ~NETIF_F_TSO_ECN; | |
5297 | ||
212b573f MM |
5298 | /* Software GSO depends on SG. */ |
5299 | if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { | |
6f404e44 | 5300 | netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); |
212b573f MM |
5301 | features &= ~NETIF_F_GSO; |
5302 | } | |
5303 | ||
acd1130e | 5304 | /* UFO needs SG and checksumming */ |
b63365a2 | 5305 | if (features & NETIF_F_UFO) { |
79032644 MM |
5306 | /* maybe split UFO into V4 and V6? */ |
5307 | if (!((features & NETIF_F_GEN_CSUM) || | |
5308 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM)) | |
5309 | == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 5310 | netdev_dbg(dev, |
acd1130e | 5311 | "Dropping NETIF_F_UFO since no checksum offload features.\n"); |
b63365a2 HX |
5312 | features &= ~NETIF_F_UFO; |
5313 | } | |
5314 | ||
5315 | if (!(features & NETIF_F_SG)) { | |
6f404e44 | 5316 | netdev_dbg(dev, |
acd1130e | 5317 | "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n"); |
b63365a2 HX |
5318 | features &= ~NETIF_F_UFO; |
5319 | } | |
5320 | } | |
5321 | ||
5322 | return features; | |
5323 | } | |
b63365a2 | 5324 | |
6cb6a27c | 5325 | int __netdev_update_features(struct net_device *dev) |
5455c699 | 5326 | { |
c8f44aff | 5327 | netdev_features_t features; |
5455c699 MM |
5328 | int err = 0; |
5329 | ||
87267485 MM |
5330 | ASSERT_RTNL(); |
5331 | ||
5455c699 MM |
5332 | features = netdev_get_wanted_features(dev); |
5333 | ||
5334 | if (dev->netdev_ops->ndo_fix_features) | |
5335 | features = dev->netdev_ops->ndo_fix_features(dev, features); | |
5336 | ||
5337 | /* driver might be less strict about feature dependencies */ | |
5338 | features = netdev_fix_features(dev, features); | |
5339 | ||
5340 | if (dev->features == features) | |
6cb6a27c | 5341 | return 0; |
5455c699 | 5342 | |
c8f44aff MM |
5343 | netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", |
5344 | &dev->features, &features); | |
5455c699 MM |
5345 | |
5346 | if (dev->netdev_ops->ndo_set_features) | |
5347 | err = dev->netdev_ops->ndo_set_features(dev, features); | |
5348 | ||
6cb6a27c | 5349 | if (unlikely(err < 0)) { |
5455c699 | 5350 | netdev_err(dev, |
c8f44aff MM |
5351 | "set_features() failed (%d); wanted %pNF, left %pNF\n", |
5352 | err, &features, &dev->features); | |
6cb6a27c MM |
5353 | return -1; |
5354 | } | |
5355 | ||
5356 | if (!err) | |
5357 | dev->features = features; | |
5358 | ||
5359 | return 1; | |
5360 | } | |
5361 | ||
afe12cc8 MM |
5362 | /** |
5363 | * netdev_update_features - recalculate device features | |
5364 | * @dev: the device to check | |
5365 | * | |
5366 | * Recalculate dev->features set and send notifications if it | |
5367 | * has changed. Should be called after driver or hardware dependent | |
5368 | * conditions might have changed that influence the features. | |
5369 | */ | |
6cb6a27c MM |
5370 | void netdev_update_features(struct net_device *dev) |
5371 | { | |
5372 | if (__netdev_update_features(dev)) | |
5373 | netdev_features_change(dev); | |
5455c699 MM |
5374 | } |
5375 | EXPORT_SYMBOL(netdev_update_features); | |
5376 | ||
afe12cc8 MM |
5377 | /** |
5378 | * netdev_change_features - recalculate device features | |
5379 | * @dev: the device to check | |
5380 | * | |
5381 | * Recalculate dev->features set and send notifications even | |
5382 | * if they have not changed. Should be called instead of | |
5383 | * netdev_update_features() if also dev->vlan_features might | |
5384 | * have changed to allow the changes to be propagated to stacked | |
5385 | * VLAN devices. | |
5386 | */ | |
5387 | void netdev_change_features(struct net_device *dev) | |
5388 | { | |
5389 | __netdev_update_features(dev); | |
5390 | netdev_features_change(dev); | |
5391 | } | |
5392 | EXPORT_SYMBOL(netdev_change_features); | |
5393 | ||
fc4a7489 PM |
5394 | /** |
5395 | * netif_stacked_transfer_operstate - transfer operstate | |
5396 | * @rootdev: the root or lower level device to transfer state from | |
5397 | * @dev: the device to transfer operstate to | |
5398 | * | |
5399 | * Transfer operational state from root to device. This is normally | |
5400 | * called when a stacking relationship exists between the root | |
5401 | * device and the device(a leaf device). | |
5402 | */ | |
5403 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
5404 | struct net_device *dev) | |
5405 | { | |
5406 | if (rootdev->operstate == IF_OPER_DORMANT) | |
5407 | netif_dormant_on(dev); | |
5408 | else | |
5409 | netif_dormant_off(dev); | |
5410 | ||
5411 | if (netif_carrier_ok(rootdev)) { | |
5412 | if (!netif_carrier_ok(dev)) | |
5413 | netif_carrier_on(dev); | |
5414 | } else { | |
5415 | if (netif_carrier_ok(dev)) | |
5416 | netif_carrier_off(dev); | |
5417 | } | |
5418 | } | |
5419 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
5420 | ||
bf264145 | 5421 | #ifdef CONFIG_RPS |
1b4bf461 ED |
5422 | static int netif_alloc_rx_queues(struct net_device *dev) |
5423 | { | |
1b4bf461 | 5424 | unsigned int i, count = dev->num_rx_queues; |
bd25fa7b | 5425 | struct netdev_rx_queue *rx; |
1b4bf461 | 5426 | |
bd25fa7b | 5427 | BUG_ON(count < 1); |
1b4bf461 | 5428 | |
bd25fa7b TH |
5429 | rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL); |
5430 | if (!rx) { | |
7b6cd1ce | 5431 | pr_err("netdev: Unable to allocate %u rx queues\n", count); |
bd25fa7b | 5432 | return -ENOMEM; |
1b4bf461 | 5433 | } |
bd25fa7b TH |
5434 | dev->_rx = rx; |
5435 | ||
bd25fa7b | 5436 | for (i = 0; i < count; i++) |
fe822240 | 5437 | rx[i].dev = dev; |
1b4bf461 ED |
5438 | return 0; |
5439 | } | |
bf264145 | 5440 | #endif |
1b4bf461 | 5441 | |
aa942104 CG |
5442 | static void netdev_init_one_queue(struct net_device *dev, |
5443 | struct netdev_queue *queue, void *_unused) | |
5444 | { | |
5445 | /* Initialize queue lock */ | |
5446 | spin_lock_init(&queue->_xmit_lock); | |
5447 | netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); | |
5448 | queue->xmit_lock_owner = -1; | |
b236da69 | 5449 | netdev_queue_numa_node_write(queue, NUMA_NO_NODE); |
aa942104 | 5450 | queue->dev = dev; |
114cf580 TH |
5451 | #ifdef CONFIG_BQL |
5452 | dql_init(&queue->dql, HZ); | |
5453 | #endif | |
aa942104 CG |
5454 | } |
5455 | ||
e6484930 TH |
5456 | static int netif_alloc_netdev_queues(struct net_device *dev) |
5457 | { | |
5458 | unsigned int count = dev->num_tx_queues; | |
5459 | struct netdev_queue *tx; | |
5460 | ||
5461 | BUG_ON(count < 1); | |
5462 | ||
5463 | tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL); | |
5464 | if (!tx) { | |
7b6cd1ce | 5465 | pr_err("netdev: Unable to allocate %u tx queues\n", count); |
e6484930 TH |
5466 | return -ENOMEM; |
5467 | } | |
5468 | dev->_tx = tx; | |
1d24eb48 | 5469 | |
e6484930 TH |
5470 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); |
5471 | spin_lock_init(&dev->tx_global_lock); | |
aa942104 CG |
5472 | |
5473 | return 0; | |
e6484930 TH |
5474 | } |
5475 | ||
1da177e4 LT |
5476 | /** |
5477 | * register_netdevice - register a network device | |
5478 | * @dev: device to register | |
5479 | * | |
5480 | * Take a completed network device structure and add it to the kernel | |
5481 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
5482 | * chain. 0 is returned on success. A negative errno code is returned | |
5483 | * on a failure to set up the device, or if the name is a duplicate. | |
5484 | * | |
5485 | * Callers must hold the rtnl semaphore. You may want | |
5486 | * register_netdev() instead of this. | |
5487 | * | |
5488 | * BUGS: | |
5489 | * The locking appears insufficient to guarantee two parallel registers | |
5490 | * will not get the same name. | |
5491 | */ | |
5492 | ||
5493 | int register_netdevice(struct net_device *dev) | |
5494 | { | |
1da177e4 | 5495 | int ret; |
d314774c | 5496 | struct net *net = dev_net(dev); |
1da177e4 LT |
5497 | |
5498 | BUG_ON(dev_boot_phase); | |
5499 | ASSERT_RTNL(); | |
5500 | ||
b17a7c17 SH |
5501 | might_sleep(); |
5502 | ||
1da177e4 LT |
5503 | /* When net_device's are persistent, this will be fatal. */ |
5504 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 5505 | BUG_ON(!net); |
1da177e4 | 5506 | |
f1f28aa3 | 5507 | spin_lock_init(&dev->addr_list_lock); |
cf508b12 | 5508 | netdev_set_addr_lockdep_class(dev); |
1da177e4 | 5509 | |
1da177e4 LT |
5510 | dev->iflink = -1; |
5511 | ||
0696c3a8 PP |
5512 | ret = dev_get_valid_name(dev, dev->name); |
5513 | if (ret < 0) | |
5514 | goto out; | |
5515 | ||
1da177e4 | 5516 | /* Init, if this function is available */ |
d314774c SH |
5517 | if (dev->netdev_ops->ndo_init) { |
5518 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
5519 | if (ret) { |
5520 | if (ret > 0) | |
5521 | ret = -EIO; | |
90833aa4 | 5522 | goto out; |
1da177e4 LT |
5523 | } |
5524 | } | |
4ec93edb | 5525 | |
881d966b | 5526 | dev->ifindex = dev_new_index(net); |
1da177e4 LT |
5527 | if (dev->iflink == -1) |
5528 | dev->iflink = dev->ifindex; | |
5529 | ||
5455c699 MM |
5530 | /* Transfer changeable features to wanted_features and enable |
5531 | * software offloads (GSO and GRO). | |
5532 | */ | |
5533 | dev->hw_features |= NETIF_F_SOFT_FEATURES; | |
14d1232f MM |
5534 | dev->features |= NETIF_F_SOFT_FEATURES; |
5535 | dev->wanted_features = dev->features & dev->hw_features; | |
1da177e4 | 5536 | |
c6e1a0d1 | 5537 | /* Turn on no cache copy if HW is doing checksum */ |
34324dc2 MM |
5538 | if (!(dev->flags & IFF_LOOPBACK)) { |
5539 | dev->hw_features |= NETIF_F_NOCACHE_COPY; | |
5540 | if (dev->features & NETIF_F_ALL_CSUM) { | |
5541 | dev->wanted_features |= NETIF_F_NOCACHE_COPY; | |
5542 | dev->features |= NETIF_F_NOCACHE_COPY; | |
5543 | } | |
c6e1a0d1 TH |
5544 | } |
5545 | ||
1180e7d6 | 5546 | /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. |
16c3ea78 | 5547 | */ |
1180e7d6 | 5548 | dev->vlan_features |= NETIF_F_HIGHDMA; |
16c3ea78 | 5549 | |
7ffbe3fd JB |
5550 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); |
5551 | ret = notifier_to_errno(ret); | |
5552 | if (ret) | |
5553 | goto err_uninit; | |
5554 | ||
8b41d188 | 5555 | ret = netdev_register_kobject(dev); |
b17a7c17 | 5556 | if (ret) |
7ce1b0ed | 5557 | goto err_uninit; |
b17a7c17 SH |
5558 | dev->reg_state = NETREG_REGISTERED; |
5559 | ||
6cb6a27c | 5560 | __netdev_update_features(dev); |
8e9b59b2 | 5561 | |
1da177e4 LT |
5562 | /* |
5563 | * Default initial state at registry is that the | |
5564 | * device is present. | |
5565 | */ | |
5566 | ||
5567 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
5568 | ||
1da177e4 | 5569 | dev_init_scheduler(dev); |
1da177e4 | 5570 | dev_hold(dev); |
ce286d32 | 5571 | list_netdevice(dev); |
1da177e4 LT |
5572 | |
5573 | /* Notify protocols, that a new device appeared. */ | |
056925ab | 5574 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 5575 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
5576 | if (ret) { |
5577 | rollback_registered(dev); | |
5578 | dev->reg_state = NETREG_UNREGISTERED; | |
5579 | } | |
d90a909e EB |
5580 | /* |
5581 | * Prevent userspace races by waiting until the network | |
5582 | * device is fully setup before sending notifications. | |
5583 | */ | |
a2835763 PM |
5584 | if (!dev->rtnl_link_ops || |
5585 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
5586 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U); | |
1da177e4 LT |
5587 | |
5588 | out: | |
5589 | return ret; | |
7ce1b0ed HX |
5590 | |
5591 | err_uninit: | |
d314774c SH |
5592 | if (dev->netdev_ops->ndo_uninit) |
5593 | dev->netdev_ops->ndo_uninit(dev); | |
7ce1b0ed | 5594 | goto out; |
1da177e4 | 5595 | } |
d1b19dff | 5596 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 5597 | |
937f1ba5 BH |
5598 | /** |
5599 | * init_dummy_netdev - init a dummy network device for NAPI | |
5600 | * @dev: device to init | |
5601 | * | |
5602 | * This takes a network device structure and initialize the minimum | |
5603 | * amount of fields so it can be used to schedule NAPI polls without | |
5604 | * registering a full blown interface. This is to be used by drivers | |
5605 | * that need to tie several hardware interfaces to a single NAPI | |
5606 | * poll scheduler due to HW limitations. | |
5607 | */ | |
5608 | int init_dummy_netdev(struct net_device *dev) | |
5609 | { | |
5610 | /* Clear everything. Note we don't initialize spinlocks | |
5611 | * are they aren't supposed to be taken by any of the | |
5612 | * NAPI code and this dummy netdev is supposed to be | |
5613 | * only ever used for NAPI polls | |
5614 | */ | |
5615 | memset(dev, 0, sizeof(struct net_device)); | |
5616 | ||
5617 | /* make sure we BUG if trying to hit standard | |
5618 | * register/unregister code path | |
5619 | */ | |
5620 | dev->reg_state = NETREG_DUMMY; | |
5621 | ||
937f1ba5 BH |
5622 | /* NAPI wants this */ |
5623 | INIT_LIST_HEAD(&dev->napi_list); | |
5624 | ||
5625 | /* a dummy interface is started by default */ | |
5626 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
5627 | set_bit(__LINK_STATE_START, &dev->state); | |
5628 | ||
29b4433d ED |
5629 | /* Note : We dont allocate pcpu_refcnt for dummy devices, |
5630 | * because users of this 'device' dont need to change | |
5631 | * its refcount. | |
5632 | */ | |
5633 | ||
937f1ba5 BH |
5634 | return 0; |
5635 | } | |
5636 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
5637 | ||
5638 | ||
1da177e4 LT |
5639 | /** |
5640 | * register_netdev - register a network device | |
5641 | * @dev: device to register | |
5642 | * | |
5643 | * Take a completed network device structure and add it to the kernel | |
5644 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
5645 | * chain. 0 is returned on success. A negative errno code is returned | |
5646 | * on a failure to set up the device, or if the name is a duplicate. | |
5647 | * | |
38b4da38 | 5648 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
5649 | * and expands the device name if you passed a format string to |
5650 | * alloc_netdev. | |
5651 | */ | |
5652 | int register_netdev(struct net_device *dev) | |
5653 | { | |
5654 | int err; | |
5655 | ||
5656 | rtnl_lock(); | |
1da177e4 | 5657 | err = register_netdevice(dev); |
1da177e4 LT |
5658 | rtnl_unlock(); |
5659 | return err; | |
5660 | } | |
5661 | EXPORT_SYMBOL(register_netdev); | |
5662 | ||
29b4433d ED |
5663 | int netdev_refcnt_read(const struct net_device *dev) |
5664 | { | |
5665 | int i, refcnt = 0; | |
5666 | ||
5667 | for_each_possible_cpu(i) | |
5668 | refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); | |
5669 | return refcnt; | |
5670 | } | |
5671 | EXPORT_SYMBOL(netdev_refcnt_read); | |
5672 | ||
1da177e4 LT |
5673 | /* |
5674 | * netdev_wait_allrefs - wait until all references are gone. | |
5675 | * | |
5676 | * This is called when unregistering network devices. | |
5677 | * | |
5678 | * Any protocol or device that holds a reference should register | |
5679 | * for netdevice notification, and cleanup and put back the | |
5680 | * reference if they receive an UNREGISTER event. | |
5681 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 5682 | * call dev_put. |
1da177e4 LT |
5683 | */ |
5684 | static void netdev_wait_allrefs(struct net_device *dev) | |
5685 | { | |
5686 | unsigned long rebroadcast_time, warning_time; | |
29b4433d | 5687 | int refcnt; |
1da177e4 | 5688 | |
e014debe ED |
5689 | linkwatch_forget_dev(dev); |
5690 | ||
1da177e4 | 5691 | rebroadcast_time = warning_time = jiffies; |
29b4433d ED |
5692 | refcnt = netdev_refcnt_read(dev); |
5693 | ||
5694 | while (refcnt != 0) { | |
1da177e4 | 5695 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { |
6756ae4b | 5696 | rtnl_lock(); |
1da177e4 LT |
5697 | |
5698 | /* Rebroadcast unregister notification */ | |
056925ab | 5699 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
a5ee1551 | 5700 | /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users |
395264d5 | 5701 | * should have already handle it the first time */ |
1da177e4 LT |
5702 | |
5703 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, | |
5704 | &dev->state)) { | |
5705 | /* We must not have linkwatch events | |
5706 | * pending on unregister. If this | |
5707 | * happens, we simply run the queue | |
5708 | * unscheduled, resulting in a noop | |
5709 | * for this device. | |
5710 | */ | |
5711 | linkwatch_run_queue(); | |
5712 | } | |
5713 | ||
6756ae4b | 5714 | __rtnl_unlock(); |
1da177e4 LT |
5715 | |
5716 | rebroadcast_time = jiffies; | |
5717 | } | |
5718 | ||
5719 | msleep(250); | |
5720 | ||
29b4433d ED |
5721 | refcnt = netdev_refcnt_read(dev); |
5722 | ||
1da177e4 | 5723 | if (time_after(jiffies, warning_time + 10 * HZ)) { |
7b6cd1ce JP |
5724 | pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", |
5725 | dev->name, refcnt); | |
1da177e4 LT |
5726 | warning_time = jiffies; |
5727 | } | |
5728 | } | |
5729 | } | |
5730 | ||
5731 | /* The sequence is: | |
5732 | * | |
5733 | * rtnl_lock(); | |
5734 | * ... | |
5735 | * register_netdevice(x1); | |
5736 | * register_netdevice(x2); | |
5737 | * ... | |
5738 | * unregister_netdevice(y1); | |
5739 | * unregister_netdevice(y2); | |
5740 | * ... | |
5741 | * rtnl_unlock(); | |
5742 | * free_netdev(y1); | |
5743 | * free_netdev(y2); | |
5744 | * | |
58ec3b4d | 5745 | * We are invoked by rtnl_unlock(). |
1da177e4 | 5746 | * This allows us to deal with problems: |
b17a7c17 | 5747 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
5748 | * without deadlocking with linkwatch via keventd. |
5749 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
5750 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
5751 | * |
5752 | * We must not return until all unregister events added during | |
5753 | * the interval the lock was held have been completed. | |
1da177e4 | 5754 | */ |
1da177e4 LT |
5755 | void netdev_run_todo(void) |
5756 | { | |
626ab0e6 | 5757 | struct list_head list; |
1da177e4 | 5758 | |
1da177e4 | 5759 | /* Snapshot list, allow later requests */ |
626ab0e6 | 5760 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
5761 | |
5762 | __rtnl_unlock(); | |
626ab0e6 | 5763 | |
850a545b EB |
5764 | /* Wait for rcu callbacks to finish before attempting to drain |
5765 | * the device list. This usually avoids a 250ms wait. | |
5766 | */ | |
5767 | if (!list_empty(&list)) | |
5768 | rcu_barrier(); | |
5769 | ||
1da177e4 LT |
5770 | while (!list_empty(&list)) { |
5771 | struct net_device *dev | |
e5e26d75 | 5772 | = list_first_entry(&list, struct net_device, todo_list); |
1da177e4 LT |
5773 | list_del(&dev->todo_list); |
5774 | ||
b17a7c17 | 5775 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
7b6cd1ce | 5776 | pr_err("network todo '%s' but state %d\n", |
b17a7c17 SH |
5777 | dev->name, dev->reg_state); |
5778 | dump_stack(); | |
5779 | continue; | |
5780 | } | |
1da177e4 | 5781 | |
b17a7c17 | 5782 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 5783 | |
152102c7 | 5784 | on_each_cpu(flush_backlog, dev, 1); |
6e583ce5 | 5785 | |
b17a7c17 | 5786 | netdev_wait_allrefs(dev); |
1da177e4 | 5787 | |
b17a7c17 | 5788 | /* paranoia */ |
29b4433d | 5789 | BUG_ON(netdev_refcnt_read(dev)); |
33d480ce ED |
5790 | WARN_ON(rcu_access_pointer(dev->ip_ptr)); |
5791 | WARN_ON(rcu_access_pointer(dev->ip6_ptr)); | |
547b792c | 5792 | WARN_ON(dev->dn_ptr); |
1da177e4 | 5793 | |
b17a7c17 SH |
5794 | if (dev->destructor) |
5795 | dev->destructor(dev); | |
9093bbb2 SH |
5796 | |
5797 | /* Free network device */ | |
5798 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 5799 | } |
1da177e4 LT |
5800 | } |
5801 | ||
3cfde79c BH |
5802 | /* Convert net_device_stats to rtnl_link_stats64. They have the same |
5803 | * fields in the same order, with only the type differing. | |
5804 | */ | |
77a1abf5 ED |
5805 | void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, |
5806 | const struct net_device_stats *netdev_stats) | |
3cfde79c BH |
5807 | { |
5808 | #if BITS_PER_LONG == 64 | |
77a1abf5 ED |
5809 | BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats)); |
5810 | memcpy(stats64, netdev_stats, sizeof(*stats64)); | |
3cfde79c BH |
5811 | #else |
5812 | size_t i, n = sizeof(*stats64) / sizeof(u64); | |
5813 | const unsigned long *src = (const unsigned long *)netdev_stats; | |
5814 | u64 *dst = (u64 *)stats64; | |
5815 | ||
5816 | BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) != | |
5817 | sizeof(*stats64) / sizeof(u64)); | |
5818 | for (i = 0; i < n; i++) | |
5819 | dst[i] = src[i]; | |
5820 | #endif | |
5821 | } | |
77a1abf5 | 5822 | EXPORT_SYMBOL(netdev_stats_to_stats64); |
3cfde79c | 5823 | |
eeda3fd6 SH |
5824 | /** |
5825 | * dev_get_stats - get network device statistics | |
5826 | * @dev: device to get statistics from | |
28172739 | 5827 | * @storage: place to store stats |
eeda3fd6 | 5828 | * |
d7753516 BH |
5829 | * Get network statistics from device. Return @storage. |
5830 | * The device driver may provide its own method by setting | |
5831 | * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; | |
5832 | * otherwise the internal statistics structure is used. | |
eeda3fd6 | 5833 | */ |
d7753516 BH |
5834 | struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, |
5835 | struct rtnl_link_stats64 *storage) | |
7004bf25 | 5836 | { |
eeda3fd6 SH |
5837 | const struct net_device_ops *ops = dev->netdev_ops; |
5838 | ||
28172739 ED |
5839 | if (ops->ndo_get_stats64) { |
5840 | memset(storage, 0, sizeof(*storage)); | |
caf586e5 ED |
5841 | ops->ndo_get_stats64(dev, storage); |
5842 | } else if (ops->ndo_get_stats) { | |
3cfde79c | 5843 | netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); |
caf586e5 ED |
5844 | } else { |
5845 | netdev_stats_to_stats64(storage, &dev->stats); | |
28172739 | 5846 | } |
caf586e5 | 5847 | storage->rx_dropped += atomic_long_read(&dev->rx_dropped); |
28172739 | 5848 | return storage; |
c45d286e | 5849 | } |
eeda3fd6 | 5850 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 5851 | |
24824a09 | 5852 | struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) |
dc2b4847 | 5853 | { |
24824a09 | 5854 | struct netdev_queue *queue = dev_ingress_queue(dev); |
dc2b4847 | 5855 | |
24824a09 ED |
5856 | #ifdef CONFIG_NET_CLS_ACT |
5857 | if (queue) | |
5858 | return queue; | |
5859 | queue = kzalloc(sizeof(*queue), GFP_KERNEL); | |
5860 | if (!queue) | |
5861 | return NULL; | |
5862 | netdev_init_one_queue(dev, queue, NULL); | |
24824a09 ED |
5863 | queue->qdisc = &noop_qdisc; |
5864 | queue->qdisc_sleeping = &noop_qdisc; | |
5865 | rcu_assign_pointer(dev->ingress_queue, queue); | |
5866 | #endif | |
5867 | return queue; | |
bb949fbd DM |
5868 | } |
5869 | ||
1da177e4 | 5870 | /** |
36909ea4 | 5871 | * alloc_netdev_mqs - allocate network device |
1da177e4 LT |
5872 | * @sizeof_priv: size of private data to allocate space for |
5873 | * @name: device name format string | |
5874 | * @setup: callback to initialize device | |
36909ea4 TH |
5875 | * @txqs: the number of TX subqueues to allocate |
5876 | * @rxqs: the number of RX subqueues to allocate | |
1da177e4 LT |
5877 | * |
5878 | * Allocates a struct net_device with private data area for driver use | |
f25f4e44 | 5879 | * and performs basic initialization. Also allocates subquue structs |
36909ea4 | 5880 | * for each queue on the device. |
1da177e4 | 5881 | */ |
36909ea4 TH |
5882 | struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, |
5883 | void (*setup)(struct net_device *), | |
5884 | unsigned int txqs, unsigned int rxqs) | |
1da177e4 | 5885 | { |
1da177e4 | 5886 | struct net_device *dev; |
7943986c | 5887 | size_t alloc_size; |
1ce8e7b5 | 5888 | struct net_device *p; |
1da177e4 | 5889 | |
b6fe17d6 SH |
5890 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
5891 | ||
36909ea4 | 5892 | if (txqs < 1) { |
7b6cd1ce | 5893 | pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); |
55513fb4 TH |
5894 | return NULL; |
5895 | } | |
5896 | ||
36909ea4 TH |
5897 | #ifdef CONFIG_RPS |
5898 | if (rxqs < 1) { | |
7b6cd1ce | 5899 | pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); |
36909ea4 TH |
5900 | return NULL; |
5901 | } | |
5902 | #endif | |
5903 | ||
fd2ea0a7 | 5904 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
5905 | if (sizeof_priv) { |
5906 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 5907 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
5908 | alloc_size += sizeof_priv; |
5909 | } | |
5910 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 5911 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 5912 | |
31380de9 | 5913 | p = kzalloc(alloc_size, GFP_KERNEL); |
1da177e4 | 5914 | if (!p) { |
7b6cd1ce | 5915 | pr_err("alloc_netdev: Unable to allocate device\n"); |
1da177e4 LT |
5916 | return NULL; |
5917 | } | |
1da177e4 | 5918 | |
1ce8e7b5 | 5919 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 5920 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc | 5921 | |
29b4433d ED |
5922 | dev->pcpu_refcnt = alloc_percpu(int); |
5923 | if (!dev->pcpu_refcnt) | |
e6484930 | 5924 | goto free_p; |
ab9c73cc | 5925 | |
ab9c73cc | 5926 | if (dev_addr_init(dev)) |
29b4433d | 5927 | goto free_pcpu; |
ab9c73cc | 5928 | |
22bedad3 | 5929 | dev_mc_init(dev); |
a748ee24 | 5930 | dev_uc_init(dev); |
ccffad25 | 5931 | |
c346dca1 | 5932 | dev_net_set(dev, &init_net); |
1da177e4 | 5933 | |
8d3bdbd5 DM |
5934 | dev->gso_max_size = GSO_MAX_SIZE; |
5935 | ||
8d3bdbd5 DM |
5936 | INIT_LIST_HEAD(&dev->napi_list); |
5937 | INIT_LIST_HEAD(&dev->unreg_list); | |
5938 | INIT_LIST_HEAD(&dev->link_watch_list); | |
5939 | dev->priv_flags = IFF_XMIT_DST_RELEASE; | |
5940 | setup(dev); | |
5941 | ||
36909ea4 TH |
5942 | dev->num_tx_queues = txqs; |
5943 | dev->real_num_tx_queues = txqs; | |
ed9af2e8 | 5944 | if (netif_alloc_netdev_queues(dev)) |
8d3bdbd5 | 5945 | goto free_all; |
e8a0464c | 5946 | |
df334545 | 5947 | #ifdef CONFIG_RPS |
36909ea4 TH |
5948 | dev->num_rx_queues = rxqs; |
5949 | dev->real_num_rx_queues = rxqs; | |
fe822240 | 5950 | if (netif_alloc_rx_queues(dev)) |
8d3bdbd5 | 5951 | goto free_all; |
df334545 | 5952 | #endif |
0a9627f2 | 5953 | |
1da177e4 | 5954 | strcpy(dev->name, name); |
cbda10fa | 5955 | dev->group = INIT_NETDEV_GROUP; |
1da177e4 | 5956 | return dev; |
ab9c73cc | 5957 | |
8d3bdbd5 DM |
5958 | free_all: |
5959 | free_netdev(dev); | |
5960 | return NULL; | |
5961 | ||
29b4433d ED |
5962 | free_pcpu: |
5963 | free_percpu(dev->pcpu_refcnt); | |
ed9af2e8 | 5964 | kfree(dev->_tx); |
fe822240 TH |
5965 | #ifdef CONFIG_RPS |
5966 | kfree(dev->_rx); | |
5967 | #endif | |
5968 | ||
ab9c73cc JP |
5969 | free_p: |
5970 | kfree(p); | |
5971 | return NULL; | |
1da177e4 | 5972 | } |
36909ea4 | 5973 | EXPORT_SYMBOL(alloc_netdev_mqs); |
1da177e4 LT |
5974 | |
5975 | /** | |
5976 | * free_netdev - free network device | |
5977 | * @dev: device | |
5978 | * | |
4ec93edb YH |
5979 | * This function does the last stage of destroying an allocated device |
5980 | * interface. The reference to the device object is released. | |
1da177e4 LT |
5981 | * If this is the last reference then it will be freed. |
5982 | */ | |
5983 | void free_netdev(struct net_device *dev) | |
5984 | { | |
d565b0a1 HX |
5985 | struct napi_struct *p, *n; |
5986 | ||
f3005d7f DL |
5987 | release_net(dev_net(dev)); |
5988 | ||
e8a0464c | 5989 | kfree(dev->_tx); |
fe822240 TH |
5990 | #ifdef CONFIG_RPS |
5991 | kfree(dev->_rx); | |
5992 | #endif | |
e8a0464c | 5993 | |
33d480ce | 5994 | kfree(rcu_dereference_protected(dev->ingress_queue, 1)); |
24824a09 | 5995 | |
f001fde5 JP |
5996 | /* Flush device addresses */ |
5997 | dev_addr_flush(dev); | |
5998 | ||
d565b0a1 HX |
5999 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
6000 | netif_napi_del(p); | |
6001 | ||
29b4433d ED |
6002 | free_percpu(dev->pcpu_refcnt); |
6003 | dev->pcpu_refcnt = NULL; | |
6004 | ||
3041a069 | 6005 | /* Compatibility with error handling in drivers */ |
1da177e4 LT |
6006 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
6007 | kfree((char *)dev - dev->padded); | |
6008 | return; | |
6009 | } | |
6010 | ||
6011 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
6012 | dev->reg_state = NETREG_RELEASED; | |
6013 | ||
43cb76d9 GKH |
6014 | /* will free via device release */ |
6015 | put_device(&dev->dev); | |
1da177e4 | 6016 | } |
d1b19dff | 6017 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 6018 | |
f0db275a SH |
6019 | /** |
6020 | * synchronize_net - Synchronize with packet receive processing | |
6021 | * | |
6022 | * Wait for packets currently being received to be done. | |
6023 | * Does not block later packets from starting. | |
6024 | */ | |
4ec93edb | 6025 | void synchronize_net(void) |
1da177e4 LT |
6026 | { |
6027 | might_sleep(); | |
be3fc413 ED |
6028 | if (rtnl_is_locked()) |
6029 | synchronize_rcu_expedited(); | |
6030 | else | |
6031 | synchronize_rcu(); | |
1da177e4 | 6032 | } |
d1b19dff | 6033 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
6034 | |
6035 | /** | |
44a0873d | 6036 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 6037 | * @dev: device |
44a0873d | 6038 | * @head: list |
6ebfbc06 | 6039 | * |
1da177e4 | 6040 | * This function shuts down a device interface and removes it |
d59b54b1 | 6041 | * from the kernel tables. |
44a0873d | 6042 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
6043 | * |
6044 | * Callers must hold the rtnl semaphore. You may want | |
6045 | * unregister_netdev() instead of this. | |
6046 | */ | |
6047 | ||
44a0873d | 6048 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 6049 | { |
a6620712 HX |
6050 | ASSERT_RTNL(); |
6051 | ||
44a0873d | 6052 | if (head) { |
9fdce099 | 6053 | list_move_tail(&dev->unreg_list, head); |
44a0873d ED |
6054 | } else { |
6055 | rollback_registered(dev); | |
6056 | /* Finish processing unregister after unlock */ | |
6057 | net_set_todo(dev); | |
6058 | } | |
1da177e4 | 6059 | } |
44a0873d | 6060 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 6061 | |
9b5e383c ED |
6062 | /** |
6063 | * unregister_netdevice_many - unregister many devices | |
6064 | * @head: list of devices | |
9b5e383c ED |
6065 | */ |
6066 | void unregister_netdevice_many(struct list_head *head) | |
6067 | { | |
6068 | struct net_device *dev; | |
6069 | ||
6070 | if (!list_empty(head)) { | |
6071 | rollback_registered_many(head); | |
6072 | list_for_each_entry(dev, head, unreg_list) | |
6073 | net_set_todo(dev); | |
6074 | } | |
6075 | } | |
63c8099d | 6076 | EXPORT_SYMBOL(unregister_netdevice_many); |
9b5e383c | 6077 | |
1da177e4 LT |
6078 | /** |
6079 | * unregister_netdev - remove device from the kernel | |
6080 | * @dev: device | |
6081 | * | |
6082 | * This function shuts down a device interface and removes it | |
d59b54b1 | 6083 | * from the kernel tables. |
1da177e4 LT |
6084 | * |
6085 | * This is just a wrapper for unregister_netdevice that takes | |
6086 | * the rtnl semaphore. In general you want to use this and not | |
6087 | * unregister_netdevice. | |
6088 | */ | |
6089 | void unregister_netdev(struct net_device *dev) | |
6090 | { | |
6091 | rtnl_lock(); | |
6092 | unregister_netdevice(dev); | |
6093 | rtnl_unlock(); | |
6094 | } | |
1da177e4 LT |
6095 | EXPORT_SYMBOL(unregister_netdev); |
6096 | ||
ce286d32 EB |
6097 | /** |
6098 | * dev_change_net_namespace - move device to different nethost namespace | |
6099 | * @dev: device | |
6100 | * @net: network namespace | |
6101 | * @pat: If not NULL name pattern to try if the current device name | |
6102 | * is already taken in the destination network namespace. | |
6103 | * | |
6104 | * This function shuts down a device interface and moves it | |
6105 | * to a new network namespace. On success 0 is returned, on | |
6106 | * a failure a netagive errno code is returned. | |
6107 | * | |
6108 | * Callers must hold the rtnl semaphore. | |
6109 | */ | |
6110 | ||
6111 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
6112 | { | |
ce286d32 EB |
6113 | int err; |
6114 | ||
6115 | ASSERT_RTNL(); | |
6116 | ||
6117 | /* Don't allow namespace local devices to be moved. */ | |
6118 | err = -EINVAL; | |
6119 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
6120 | goto out; | |
6121 | ||
6122 | /* Ensure the device has been registrered */ | |
6123 | err = -EINVAL; | |
6124 | if (dev->reg_state != NETREG_REGISTERED) | |
6125 | goto out; | |
6126 | ||
6127 | /* Get out if there is nothing todo */ | |
6128 | err = 0; | |
878628fb | 6129 | if (net_eq(dev_net(dev), net)) |
ce286d32 EB |
6130 | goto out; |
6131 | ||
6132 | /* Pick the destination device name, and ensure | |
6133 | * we can use it in the destination network namespace. | |
6134 | */ | |
6135 | err = -EEXIST; | |
d9031024 | 6136 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
6137 | /* We get here if we can't use the current device name */ |
6138 | if (!pat) | |
6139 | goto out; | |
1c5cae81 | 6140 | if (dev_get_valid_name(dev, pat) < 0) |
ce286d32 EB |
6141 | goto out; |
6142 | } | |
6143 | ||
6144 | /* | |
6145 | * And now a mini version of register_netdevice unregister_netdevice. | |
6146 | */ | |
6147 | ||
6148 | /* If device is running close it first. */ | |
9b772652 | 6149 | dev_close(dev); |
ce286d32 EB |
6150 | |
6151 | /* And unlink it from device chain */ | |
6152 | err = -ENODEV; | |
6153 | unlist_netdevice(dev); | |
6154 | ||
6155 | synchronize_net(); | |
6156 | ||
6157 | /* Shutdown queueing discipline. */ | |
6158 | dev_shutdown(dev); | |
6159 | ||
6160 | /* Notify protocols, that we are about to destroy | |
6161 | this device. They should clean all the things. | |
3b27e105 DL |
6162 | |
6163 | Note that dev->reg_state stays at NETREG_REGISTERED. | |
6164 | This is wanted because this way 8021q and macvlan know | |
6165 | the device is just moving and can keep their slaves up. | |
ce286d32 EB |
6166 | */ |
6167 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
a5ee1551 | 6168 | call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev); |
d2237d35 | 6169 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U); |
ce286d32 EB |
6170 | |
6171 | /* | |
6172 | * Flush the unicast and multicast chains | |
6173 | */ | |
a748ee24 | 6174 | dev_uc_flush(dev); |
22bedad3 | 6175 | dev_mc_flush(dev); |
ce286d32 EB |
6176 | |
6177 | /* Actually switch the network namespace */ | |
c346dca1 | 6178 | dev_net_set(dev, net); |
ce286d32 | 6179 | |
ce286d32 EB |
6180 | /* If there is an ifindex conflict assign a new one */ |
6181 | if (__dev_get_by_index(net, dev->ifindex)) { | |
6182 | int iflink = (dev->iflink == dev->ifindex); | |
6183 | dev->ifindex = dev_new_index(net); | |
6184 | if (iflink) | |
6185 | dev->iflink = dev->ifindex; | |
6186 | } | |
6187 | ||
8b41d188 | 6188 | /* Fixup kobjects */ |
a1b3f594 | 6189 | err = device_rename(&dev->dev, dev->name); |
8b41d188 | 6190 | WARN_ON(err); |
ce286d32 EB |
6191 | |
6192 | /* Add the device back in the hashes */ | |
6193 | list_netdevice(dev); | |
6194 | ||
6195 | /* Notify protocols, that a new device appeared. */ | |
6196 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
6197 | ||
d90a909e EB |
6198 | /* |
6199 | * Prevent userspace races by waiting until the network | |
6200 | * device is fully setup before sending notifications. | |
6201 | */ | |
6202 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U); | |
6203 | ||
ce286d32 EB |
6204 | synchronize_net(); |
6205 | err = 0; | |
6206 | out: | |
6207 | return err; | |
6208 | } | |
463d0183 | 6209 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 6210 | |
1da177e4 LT |
6211 | static int dev_cpu_callback(struct notifier_block *nfb, |
6212 | unsigned long action, | |
6213 | void *ocpu) | |
6214 | { | |
6215 | struct sk_buff **list_skb; | |
1da177e4 LT |
6216 | struct sk_buff *skb; |
6217 | unsigned int cpu, oldcpu = (unsigned long)ocpu; | |
6218 | struct softnet_data *sd, *oldsd; | |
6219 | ||
8bb78442 | 6220 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
1da177e4 LT |
6221 | return NOTIFY_OK; |
6222 | ||
6223 | local_irq_disable(); | |
6224 | cpu = smp_processor_id(); | |
6225 | sd = &per_cpu(softnet_data, cpu); | |
6226 | oldsd = &per_cpu(softnet_data, oldcpu); | |
6227 | ||
6228 | /* Find end of our completion_queue. */ | |
6229 | list_skb = &sd->completion_queue; | |
6230 | while (*list_skb) | |
6231 | list_skb = &(*list_skb)->next; | |
6232 | /* Append completion queue from offline CPU. */ | |
6233 | *list_skb = oldsd->completion_queue; | |
6234 | oldsd->completion_queue = NULL; | |
6235 | ||
1da177e4 | 6236 | /* Append output queue from offline CPU. */ |
a9cbd588 CG |
6237 | if (oldsd->output_queue) { |
6238 | *sd->output_queue_tailp = oldsd->output_queue; | |
6239 | sd->output_queue_tailp = oldsd->output_queue_tailp; | |
6240 | oldsd->output_queue = NULL; | |
6241 | oldsd->output_queue_tailp = &oldsd->output_queue; | |
6242 | } | |
264524d5 HC |
6243 | /* Append NAPI poll list from offline CPU. */ |
6244 | if (!list_empty(&oldsd->poll_list)) { | |
6245 | list_splice_init(&oldsd->poll_list, &sd->poll_list); | |
6246 | raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
6247 | } | |
1da177e4 LT |
6248 | |
6249 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
6250 | local_irq_enable(); | |
6251 | ||
6252 | /* Process offline CPU's input_pkt_queue */ | |
76cc8b13 | 6253 | while ((skb = __skb_dequeue(&oldsd->process_queue))) { |
1da177e4 | 6254 | netif_rx(skb); |
76cc8b13 | 6255 | input_queue_head_incr(oldsd); |
fec5e652 | 6256 | } |
76cc8b13 | 6257 | while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) { |
6e7676c1 | 6258 | netif_rx(skb); |
76cc8b13 TH |
6259 | input_queue_head_incr(oldsd); |
6260 | } | |
1da177e4 LT |
6261 | |
6262 | return NOTIFY_OK; | |
6263 | } | |
1da177e4 LT |
6264 | |
6265 | ||
7f353bf2 | 6266 | /** |
b63365a2 HX |
6267 | * netdev_increment_features - increment feature set by one |
6268 | * @all: current feature set | |
6269 | * @one: new feature set | |
6270 | * @mask: mask feature set | |
7f353bf2 HX |
6271 | * |
6272 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
6273 | * @one to the master device with current feature set @all. Will not |
6274 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 6275 | */ |
c8f44aff MM |
6276 | netdev_features_t netdev_increment_features(netdev_features_t all, |
6277 | netdev_features_t one, netdev_features_t mask) | |
b63365a2 | 6278 | { |
1742f183 MM |
6279 | if (mask & NETIF_F_GEN_CSUM) |
6280 | mask |= NETIF_F_ALL_CSUM; | |
6281 | mask |= NETIF_F_VLAN_CHALLENGED; | |
7f353bf2 | 6282 | |
1742f183 MM |
6283 | all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask; |
6284 | all &= one | ~NETIF_F_ALL_FOR_ALL; | |
c6e1a0d1 | 6285 | |
1742f183 MM |
6286 | /* If one device supports hw checksumming, set for all. */ |
6287 | if (all & NETIF_F_GEN_CSUM) | |
6288 | all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM); | |
7f353bf2 HX |
6289 | |
6290 | return all; | |
6291 | } | |
b63365a2 | 6292 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 6293 | |
30d97d35 PE |
6294 | static struct hlist_head *netdev_create_hash(void) |
6295 | { | |
6296 | int i; | |
6297 | struct hlist_head *hash; | |
6298 | ||
6299 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
6300 | if (hash != NULL) | |
6301 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
6302 | INIT_HLIST_HEAD(&hash[i]); | |
6303 | ||
6304 | return hash; | |
6305 | } | |
6306 | ||
881d966b | 6307 | /* Initialize per network namespace state */ |
4665079c | 6308 | static int __net_init netdev_init(struct net *net) |
881d966b | 6309 | { |
881d966b | 6310 | INIT_LIST_HEAD(&net->dev_base_head); |
881d966b | 6311 | |
30d97d35 PE |
6312 | net->dev_name_head = netdev_create_hash(); |
6313 | if (net->dev_name_head == NULL) | |
6314 | goto err_name; | |
881d966b | 6315 | |
30d97d35 PE |
6316 | net->dev_index_head = netdev_create_hash(); |
6317 | if (net->dev_index_head == NULL) | |
6318 | goto err_idx; | |
881d966b EB |
6319 | |
6320 | return 0; | |
30d97d35 PE |
6321 | |
6322 | err_idx: | |
6323 | kfree(net->dev_name_head); | |
6324 | err_name: | |
6325 | return -ENOMEM; | |
881d966b EB |
6326 | } |
6327 | ||
f0db275a SH |
6328 | /** |
6329 | * netdev_drivername - network driver for the device | |
6330 | * @dev: network device | |
f0db275a SH |
6331 | * |
6332 | * Determine network driver for device. | |
6333 | */ | |
3019de12 | 6334 | const char *netdev_drivername(const struct net_device *dev) |
6579e57b | 6335 | { |
cf04a4c7 SH |
6336 | const struct device_driver *driver; |
6337 | const struct device *parent; | |
3019de12 | 6338 | const char *empty = ""; |
6579e57b AV |
6339 | |
6340 | parent = dev->dev.parent; | |
6579e57b | 6341 | if (!parent) |
3019de12 | 6342 | return empty; |
6579e57b AV |
6343 | |
6344 | driver = parent->driver; | |
6345 | if (driver && driver->name) | |
3019de12 DM |
6346 | return driver->name; |
6347 | return empty; | |
6579e57b AV |
6348 | } |
6349 | ||
ffa10cb4 | 6350 | int __netdev_printk(const char *level, const struct net_device *dev, |
256df2f3 JP |
6351 | struct va_format *vaf) |
6352 | { | |
6353 | int r; | |
6354 | ||
6355 | if (dev && dev->dev.parent) | |
6356 | r = dev_printk(level, dev->dev.parent, "%s: %pV", | |
6357 | netdev_name(dev), vaf); | |
6358 | else if (dev) | |
6359 | r = printk("%s%s: %pV", level, netdev_name(dev), vaf); | |
6360 | else | |
6361 | r = printk("%s(NULL net_device): %pV", level, vaf); | |
6362 | ||
6363 | return r; | |
6364 | } | |
ffa10cb4 | 6365 | EXPORT_SYMBOL(__netdev_printk); |
256df2f3 JP |
6366 | |
6367 | int netdev_printk(const char *level, const struct net_device *dev, | |
6368 | const char *format, ...) | |
6369 | { | |
6370 | struct va_format vaf; | |
6371 | va_list args; | |
6372 | int r; | |
6373 | ||
6374 | va_start(args, format); | |
6375 | ||
6376 | vaf.fmt = format; | |
6377 | vaf.va = &args; | |
6378 | ||
6379 | r = __netdev_printk(level, dev, &vaf); | |
6380 | va_end(args); | |
6381 | ||
6382 | return r; | |
6383 | } | |
6384 | EXPORT_SYMBOL(netdev_printk); | |
6385 | ||
6386 | #define define_netdev_printk_level(func, level) \ | |
6387 | int func(const struct net_device *dev, const char *fmt, ...) \ | |
6388 | { \ | |
6389 | int r; \ | |
6390 | struct va_format vaf; \ | |
6391 | va_list args; \ | |
6392 | \ | |
6393 | va_start(args, fmt); \ | |
6394 | \ | |
6395 | vaf.fmt = fmt; \ | |
6396 | vaf.va = &args; \ | |
6397 | \ | |
6398 | r = __netdev_printk(level, dev, &vaf); \ | |
6399 | va_end(args); \ | |
6400 | \ | |
6401 | return r; \ | |
6402 | } \ | |
6403 | EXPORT_SYMBOL(func); | |
6404 | ||
6405 | define_netdev_printk_level(netdev_emerg, KERN_EMERG); | |
6406 | define_netdev_printk_level(netdev_alert, KERN_ALERT); | |
6407 | define_netdev_printk_level(netdev_crit, KERN_CRIT); | |
6408 | define_netdev_printk_level(netdev_err, KERN_ERR); | |
6409 | define_netdev_printk_level(netdev_warn, KERN_WARNING); | |
6410 | define_netdev_printk_level(netdev_notice, KERN_NOTICE); | |
6411 | define_netdev_printk_level(netdev_info, KERN_INFO); | |
6412 | ||
4665079c | 6413 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
6414 | { |
6415 | kfree(net->dev_name_head); | |
6416 | kfree(net->dev_index_head); | |
6417 | } | |
6418 | ||
022cbae6 | 6419 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
6420 | .init = netdev_init, |
6421 | .exit = netdev_exit, | |
6422 | }; | |
6423 | ||
4665079c | 6424 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 6425 | { |
e008b5fc | 6426 | struct net_device *dev, *aux; |
ce286d32 | 6427 | /* |
e008b5fc | 6428 | * Push all migratable network devices back to the |
ce286d32 EB |
6429 | * initial network namespace |
6430 | */ | |
6431 | rtnl_lock(); | |
e008b5fc | 6432 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 6433 | int err; |
aca51397 | 6434 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
6435 | |
6436 | /* Ignore unmoveable devices (i.e. loopback) */ | |
6437 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
6438 | continue; | |
6439 | ||
e008b5fc EB |
6440 | /* Leave virtual devices for the generic cleanup */ |
6441 | if (dev->rtnl_link_ops) | |
6442 | continue; | |
d0c082ce | 6443 | |
25985edc | 6444 | /* Push remaining network devices to init_net */ |
aca51397 PE |
6445 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
6446 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
ce286d32 | 6447 | if (err) { |
7b6cd1ce JP |
6448 | pr_emerg("%s: failed to move %s to init_net: %d\n", |
6449 | __func__, dev->name, err); | |
aca51397 | 6450 | BUG(); |
ce286d32 EB |
6451 | } |
6452 | } | |
6453 | rtnl_unlock(); | |
6454 | } | |
6455 | ||
04dc7f6b EB |
6456 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
6457 | { | |
6458 | /* At exit all network devices most be removed from a network | |
b595076a | 6459 | * namespace. Do this in the reverse order of registration. |
04dc7f6b EB |
6460 | * Do this across as many network namespaces as possible to |
6461 | * improve batching efficiency. | |
6462 | */ | |
6463 | struct net_device *dev; | |
6464 | struct net *net; | |
6465 | LIST_HEAD(dev_kill_list); | |
6466 | ||
6467 | rtnl_lock(); | |
6468 | list_for_each_entry(net, net_list, exit_list) { | |
6469 | for_each_netdev_reverse(net, dev) { | |
6470 | if (dev->rtnl_link_ops) | |
6471 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); | |
6472 | else | |
6473 | unregister_netdevice_queue(dev, &dev_kill_list); | |
6474 | } | |
6475 | } | |
6476 | unregister_netdevice_many(&dev_kill_list); | |
ceaaec98 | 6477 | list_del(&dev_kill_list); |
04dc7f6b EB |
6478 | rtnl_unlock(); |
6479 | } | |
6480 | ||
022cbae6 | 6481 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 6482 | .exit = default_device_exit, |
04dc7f6b | 6483 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
6484 | }; |
6485 | ||
1da177e4 LT |
6486 | /* |
6487 | * Initialize the DEV module. At boot time this walks the device list and | |
6488 | * unhooks any devices that fail to initialise (normally hardware not | |
6489 | * present) and leaves us with a valid list of present and active devices. | |
6490 | * | |
6491 | */ | |
6492 | ||
6493 | /* | |
6494 | * This is called single threaded during boot, so no need | |
6495 | * to take the rtnl semaphore. | |
6496 | */ | |
6497 | static int __init net_dev_init(void) | |
6498 | { | |
6499 | int i, rc = -ENOMEM; | |
6500 | ||
6501 | BUG_ON(!dev_boot_phase); | |
6502 | ||
1da177e4 LT |
6503 | if (dev_proc_init()) |
6504 | goto out; | |
6505 | ||
8b41d188 | 6506 | if (netdev_kobject_init()) |
1da177e4 LT |
6507 | goto out; |
6508 | ||
6509 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 6510 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
6511 | INIT_LIST_HEAD(&ptype_base[i]); |
6512 | ||
881d966b EB |
6513 | if (register_pernet_subsys(&netdev_net_ops)) |
6514 | goto out; | |
1da177e4 LT |
6515 | |
6516 | /* | |
6517 | * Initialise the packet receive queues. | |
6518 | */ | |
6519 | ||
6f912042 | 6520 | for_each_possible_cpu(i) { |
e36fa2f7 | 6521 | struct softnet_data *sd = &per_cpu(softnet_data, i); |
1da177e4 | 6522 | |
dee42870 | 6523 | memset(sd, 0, sizeof(*sd)); |
e36fa2f7 | 6524 | skb_queue_head_init(&sd->input_pkt_queue); |
6e7676c1 | 6525 | skb_queue_head_init(&sd->process_queue); |
e36fa2f7 ED |
6526 | sd->completion_queue = NULL; |
6527 | INIT_LIST_HEAD(&sd->poll_list); | |
a9cbd588 CG |
6528 | sd->output_queue = NULL; |
6529 | sd->output_queue_tailp = &sd->output_queue; | |
df334545 | 6530 | #ifdef CONFIG_RPS |
e36fa2f7 ED |
6531 | sd->csd.func = rps_trigger_softirq; |
6532 | sd->csd.info = sd; | |
6533 | sd->csd.flags = 0; | |
6534 | sd->cpu = i; | |
1e94d72f | 6535 | #endif |
0a9627f2 | 6536 | |
e36fa2f7 ED |
6537 | sd->backlog.poll = process_backlog; |
6538 | sd->backlog.weight = weight_p; | |
6539 | sd->backlog.gro_list = NULL; | |
6540 | sd->backlog.gro_count = 0; | |
1da177e4 LT |
6541 | } |
6542 | ||
1da177e4 LT |
6543 | dev_boot_phase = 0; |
6544 | ||
505d4f73 EB |
6545 | /* The loopback device is special if any other network devices |
6546 | * is present in a network namespace the loopback device must | |
6547 | * be present. Since we now dynamically allocate and free the | |
6548 | * loopback device ensure this invariant is maintained by | |
6549 | * keeping the loopback device as the first device on the | |
6550 | * list of network devices. Ensuring the loopback devices | |
6551 | * is the first device that appears and the last network device | |
6552 | * that disappears. | |
6553 | */ | |
6554 | if (register_pernet_device(&loopback_net_ops)) | |
6555 | goto out; | |
6556 | ||
6557 | if (register_pernet_device(&default_device_ops)) | |
6558 | goto out; | |
6559 | ||
962cf36c CM |
6560 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
6561 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 LT |
6562 | |
6563 | hotcpu_notifier(dev_cpu_callback, 0); | |
6564 | dst_init(); | |
6565 | dev_mcast_init(); | |
6566 | rc = 0; | |
6567 | out: | |
6568 | return rc; | |
6569 | } | |
6570 | ||
6571 | subsys_initcall(net_dev_init); | |
6572 | ||
e88721f8 KK |
6573 | static int __init initialize_hashrnd(void) |
6574 | { | |
0a9627f2 | 6575 | get_random_bytes(&hashrnd, sizeof(hashrnd)); |
e88721f8 KK |
6576 | return 0; |
6577 | } | |
6578 | ||
6579 | late_initcall_sync(initialize_hashrnd); | |
6580 |