]>
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); |
62532da9 | 179 | static DEFINE_SPINLOCK(offload_lock); |
82d8a867 | 180 | static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
6b2bedc3 | 181 | static struct list_head ptype_all __read_mostly; /* Taps */ |
62532da9 | 182 | static struct list_head offload_base __read_mostly; |
1da177e4 | 183 | |
1da177e4 | 184 | /* |
7562f876 | 185 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
186 | * semaphore. |
187 | * | |
c6d14c84 | 188 | * Pure readers hold dev_base_lock for reading, or rcu_read_lock() |
1da177e4 LT |
189 | * |
190 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 191 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
192 | * actual updates. This allows pure readers to access the list even |
193 | * while a writer is preparing to update it. | |
194 | * | |
195 | * To put it another way, dev_base_lock is held for writing only to | |
196 | * protect against pure readers; the rtnl semaphore provides the | |
197 | * protection against other writers. | |
198 | * | |
199 | * See, for example usages, register_netdevice() and | |
200 | * unregister_netdevice(), which must be called with the rtnl | |
201 | * semaphore held. | |
202 | */ | |
1da177e4 | 203 | DEFINE_RWLOCK(dev_base_lock); |
1da177e4 LT |
204 | EXPORT_SYMBOL(dev_base_lock); |
205 | ||
30e6c9fa | 206 | seqcount_t devnet_rename_seq; |
c91f6df2 | 207 | |
4e985ada TG |
208 | static inline void dev_base_seq_inc(struct net *net) |
209 | { | |
210 | while (++net->dev_base_seq == 0); | |
211 | } | |
212 | ||
881d966b | 213 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 | 214 | { |
95c96174 ED |
215 | unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ)); |
216 | ||
08e9897d | 217 | return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; |
1da177e4 LT |
218 | } |
219 | ||
881d966b | 220 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 221 | { |
7c28bd0b | 222 | return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; |
1da177e4 LT |
223 | } |
224 | ||
e36fa2f7 | 225 | static inline void rps_lock(struct softnet_data *sd) |
152102c7 CG |
226 | { |
227 | #ifdef CONFIG_RPS | |
e36fa2f7 | 228 | spin_lock(&sd->input_pkt_queue.lock); |
152102c7 CG |
229 | #endif |
230 | } | |
231 | ||
e36fa2f7 | 232 | static inline void rps_unlock(struct softnet_data *sd) |
152102c7 CG |
233 | { |
234 | #ifdef CONFIG_RPS | |
e36fa2f7 | 235 | spin_unlock(&sd->input_pkt_queue.lock); |
152102c7 CG |
236 | #endif |
237 | } | |
238 | ||
ce286d32 EB |
239 | /* Device list insertion */ |
240 | static int list_netdevice(struct net_device *dev) | |
241 | { | |
c346dca1 | 242 | struct net *net = dev_net(dev); |
ce286d32 EB |
243 | |
244 | ASSERT_RTNL(); | |
245 | ||
246 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 247 | list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); |
72c9528b | 248 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); |
fb699dfd ED |
249 | hlist_add_head_rcu(&dev->index_hlist, |
250 | dev_index_hash(net, dev->ifindex)); | |
ce286d32 | 251 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
252 | |
253 | dev_base_seq_inc(net); | |
254 | ||
ce286d32 EB |
255 | return 0; |
256 | } | |
257 | ||
fb699dfd ED |
258 | /* Device list removal |
259 | * caller must respect a RCU grace period before freeing/reusing dev | |
260 | */ | |
ce286d32 EB |
261 | static void unlist_netdevice(struct net_device *dev) |
262 | { | |
263 | ASSERT_RTNL(); | |
264 | ||
265 | /* Unlink dev from the device chain */ | |
266 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 267 | list_del_rcu(&dev->dev_list); |
72c9528b | 268 | hlist_del_rcu(&dev->name_hlist); |
fb699dfd | 269 | hlist_del_rcu(&dev->index_hlist); |
ce286d32 | 270 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
271 | |
272 | dev_base_seq_inc(dev_net(dev)); | |
ce286d32 EB |
273 | } |
274 | ||
1da177e4 LT |
275 | /* |
276 | * Our notifier list | |
277 | */ | |
278 | ||
f07d5b94 | 279 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
280 | |
281 | /* | |
282 | * Device drivers call our routines to queue packets here. We empty the | |
283 | * queue in the local softnet handler. | |
284 | */ | |
bea3348e | 285 | |
9958da05 | 286 | DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); |
d1b19dff | 287 | EXPORT_PER_CPU_SYMBOL(softnet_data); |
1da177e4 | 288 | |
cf508b12 | 289 | #ifdef CONFIG_LOCKDEP |
723e98b7 | 290 | /* |
c773e847 | 291 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class |
723e98b7 JP |
292 | * according to dev->type |
293 | */ | |
294 | static const unsigned short netdev_lock_type[] = | |
295 | {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
296 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, | |
297 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
298 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
299 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
300 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
301 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
302 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
303 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
304 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
305 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
306 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
211ed865 PG |
307 | ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, |
308 | ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, | |
309 | ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; | |
723e98b7 | 310 | |
36cbd3dc | 311 | static const char *const netdev_lock_name[] = |
723e98b7 JP |
312 | {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", |
313 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
314 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
315 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
316 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
317 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
318 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
319 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
320 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
321 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
322 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
323 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
211ed865 PG |
324 | "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", |
325 | "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", | |
326 | "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; | |
723e98b7 JP |
327 | |
328 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
cf508b12 | 329 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; |
723e98b7 JP |
330 | |
331 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
332 | { | |
333 | int i; | |
334 | ||
335 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
336 | if (netdev_lock_type[i] == dev_type) | |
337 | return i; | |
338 | /* the last key is used by default */ | |
339 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
340 | } | |
341 | ||
cf508b12 DM |
342 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
343 | unsigned short dev_type) | |
723e98b7 JP |
344 | { |
345 | int i; | |
346 | ||
347 | i = netdev_lock_pos(dev_type); | |
348 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
349 | netdev_lock_name[i]); | |
350 | } | |
cf508b12 DM |
351 | |
352 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
353 | { | |
354 | int i; | |
355 | ||
356 | i = netdev_lock_pos(dev->type); | |
357 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
358 | &netdev_addr_lock_key[i], | |
359 | netdev_lock_name[i]); | |
360 | } | |
723e98b7 | 361 | #else |
cf508b12 DM |
362 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
363 | unsigned short dev_type) | |
364 | { | |
365 | } | |
366 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
723e98b7 JP |
367 | { |
368 | } | |
369 | #endif | |
1da177e4 LT |
370 | |
371 | /******************************************************************************* | |
372 | ||
373 | Protocol management and registration routines | |
374 | ||
375 | *******************************************************************************/ | |
376 | ||
1da177e4 LT |
377 | /* |
378 | * Add a protocol ID to the list. Now that the input handler is | |
379 | * smarter we can dispense with all the messy stuff that used to be | |
380 | * here. | |
381 | * | |
382 | * BEWARE!!! Protocol handlers, mangling input packets, | |
383 | * MUST BE last in hash buckets and checking protocol handlers | |
384 | * MUST start from promiscuous ptype_all chain in net_bh. | |
385 | * It is true now, do not change it. | |
386 | * Explanation follows: if protocol handler, mangling packet, will | |
387 | * be the first on list, it is not able to sense, that packet | |
388 | * is cloned and should be copied-on-write, so that it will | |
389 | * change it and subsequent readers will get broken packet. | |
390 | * --ANK (980803) | |
391 | */ | |
392 | ||
c07b68e8 ED |
393 | static inline struct list_head *ptype_head(const struct packet_type *pt) |
394 | { | |
395 | if (pt->type == htons(ETH_P_ALL)) | |
396 | return &ptype_all; | |
397 | else | |
398 | return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
399 | } | |
400 | ||
1da177e4 LT |
401 | /** |
402 | * dev_add_pack - add packet handler | |
403 | * @pt: packet type declaration | |
404 | * | |
405 | * Add a protocol handler to the networking stack. The passed &packet_type | |
406 | * is linked into kernel lists and may not be freed until it has been | |
407 | * removed from the kernel lists. | |
408 | * | |
4ec93edb | 409 | * This call does not sleep therefore it can not |
1da177e4 LT |
410 | * guarantee all CPU's that are in middle of receiving packets |
411 | * will see the new packet type (until the next received packet). | |
412 | */ | |
413 | ||
414 | void dev_add_pack(struct packet_type *pt) | |
415 | { | |
c07b68e8 | 416 | struct list_head *head = ptype_head(pt); |
1da177e4 | 417 | |
c07b68e8 ED |
418 | spin_lock(&ptype_lock); |
419 | list_add_rcu(&pt->list, head); | |
420 | spin_unlock(&ptype_lock); | |
1da177e4 | 421 | } |
d1b19dff | 422 | EXPORT_SYMBOL(dev_add_pack); |
1da177e4 | 423 | |
1da177e4 LT |
424 | /** |
425 | * __dev_remove_pack - remove packet handler | |
426 | * @pt: packet type declaration | |
427 | * | |
428 | * Remove a protocol handler that was previously added to the kernel | |
429 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
430 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 431 | * returns. |
1da177e4 LT |
432 | * |
433 | * The packet type might still be in use by receivers | |
434 | * and must not be freed until after all the CPU's have gone | |
435 | * through a quiescent state. | |
436 | */ | |
437 | void __dev_remove_pack(struct packet_type *pt) | |
438 | { | |
c07b68e8 | 439 | struct list_head *head = ptype_head(pt); |
1da177e4 LT |
440 | struct packet_type *pt1; |
441 | ||
c07b68e8 | 442 | spin_lock(&ptype_lock); |
1da177e4 LT |
443 | |
444 | list_for_each_entry(pt1, head, list) { | |
445 | if (pt == pt1) { | |
446 | list_del_rcu(&pt->list); | |
447 | goto out; | |
448 | } | |
449 | } | |
450 | ||
7b6cd1ce | 451 | pr_warn("dev_remove_pack: %p not found\n", pt); |
1da177e4 | 452 | out: |
c07b68e8 | 453 | spin_unlock(&ptype_lock); |
1da177e4 | 454 | } |
d1b19dff ED |
455 | EXPORT_SYMBOL(__dev_remove_pack); |
456 | ||
1da177e4 LT |
457 | /** |
458 | * dev_remove_pack - remove packet handler | |
459 | * @pt: packet type declaration | |
460 | * | |
461 | * Remove a protocol handler that was previously added to the kernel | |
462 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
463 | * from the kernel lists and can be freed or reused once this function | |
464 | * returns. | |
465 | * | |
466 | * This call sleeps to guarantee that no CPU is looking at the packet | |
467 | * type after return. | |
468 | */ | |
469 | void dev_remove_pack(struct packet_type *pt) | |
470 | { | |
471 | __dev_remove_pack(pt); | |
4ec93edb | 472 | |
1da177e4 LT |
473 | synchronize_net(); |
474 | } | |
d1b19dff | 475 | EXPORT_SYMBOL(dev_remove_pack); |
1da177e4 | 476 | |
62532da9 VY |
477 | |
478 | /** | |
479 | * dev_add_offload - register offload handlers | |
480 | * @po: protocol offload declaration | |
481 | * | |
482 | * Add protocol offload handlers to the networking stack. The passed | |
483 | * &proto_offload is linked into kernel lists and may not be freed until | |
484 | * it has been removed from the kernel lists. | |
485 | * | |
486 | * This call does not sleep therefore it can not | |
487 | * guarantee all CPU's that are in middle of receiving packets | |
488 | * will see the new offload handlers (until the next received packet). | |
489 | */ | |
490 | void dev_add_offload(struct packet_offload *po) | |
491 | { | |
492 | struct list_head *head = &offload_base; | |
493 | ||
494 | spin_lock(&offload_lock); | |
495 | list_add_rcu(&po->list, head); | |
496 | spin_unlock(&offload_lock); | |
497 | } | |
498 | EXPORT_SYMBOL(dev_add_offload); | |
499 | ||
500 | /** | |
501 | * __dev_remove_offload - remove offload handler | |
502 | * @po: packet offload declaration | |
503 | * | |
504 | * Remove a protocol offload handler that was previously added to the | |
505 | * kernel offload handlers by dev_add_offload(). The passed &offload_type | |
506 | * is removed from the kernel lists and can be freed or reused once this | |
507 | * function returns. | |
508 | * | |
509 | * The packet type might still be in use by receivers | |
510 | * and must not be freed until after all the CPU's have gone | |
511 | * through a quiescent state. | |
512 | */ | |
513 | void __dev_remove_offload(struct packet_offload *po) | |
514 | { | |
515 | struct list_head *head = &offload_base; | |
516 | struct packet_offload *po1; | |
517 | ||
c53aa505 | 518 | spin_lock(&offload_lock); |
62532da9 VY |
519 | |
520 | list_for_each_entry(po1, head, list) { | |
521 | if (po == po1) { | |
522 | list_del_rcu(&po->list); | |
523 | goto out; | |
524 | } | |
525 | } | |
526 | ||
527 | pr_warn("dev_remove_offload: %p not found\n", po); | |
528 | out: | |
c53aa505 | 529 | spin_unlock(&offload_lock); |
62532da9 VY |
530 | } |
531 | EXPORT_SYMBOL(__dev_remove_offload); | |
532 | ||
533 | /** | |
534 | * dev_remove_offload - remove packet offload handler | |
535 | * @po: packet offload declaration | |
536 | * | |
537 | * Remove a packet offload handler that was previously added to the kernel | |
538 | * offload handlers by dev_add_offload(). The passed &offload_type is | |
539 | * removed from the kernel lists and can be freed or reused once this | |
540 | * function returns. | |
541 | * | |
542 | * This call sleeps to guarantee that no CPU is looking at the packet | |
543 | * type after return. | |
544 | */ | |
545 | void dev_remove_offload(struct packet_offload *po) | |
546 | { | |
547 | __dev_remove_offload(po); | |
548 | ||
549 | synchronize_net(); | |
550 | } | |
551 | EXPORT_SYMBOL(dev_remove_offload); | |
552 | ||
1da177e4 LT |
553 | /****************************************************************************** |
554 | ||
555 | Device Boot-time Settings Routines | |
556 | ||
557 | *******************************************************************************/ | |
558 | ||
559 | /* Boot time configuration table */ | |
560 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
561 | ||
562 | /** | |
563 | * netdev_boot_setup_add - add new setup entry | |
564 | * @name: name of the device | |
565 | * @map: configured settings for the device | |
566 | * | |
567 | * Adds new setup entry to the dev_boot_setup list. The function | |
568 | * returns 0 on error and 1 on success. This is a generic routine to | |
569 | * all netdevices. | |
570 | */ | |
571 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
572 | { | |
573 | struct netdev_boot_setup *s; | |
574 | int i; | |
575 | ||
576 | s = dev_boot_setup; | |
577 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
578 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
579 | memset(s[i].name, 0, sizeof(s[i].name)); | |
93b3cff9 | 580 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
581 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
582 | break; | |
583 | } | |
584 | } | |
585 | ||
586 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
587 | } | |
588 | ||
589 | /** | |
590 | * netdev_boot_setup_check - check boot time settings | |
591 | * @dev: the netdevice | |
592 | * | |
593 | * Check boot time settings for the device. | |
594 | * The found settings are set for the device to be used | |
595 | * later in the device probing. | |
596 | * Returns 0 if no settings found, 1 if they are. | |
597 | */ | |
598 | int netdev_boot_setup_check(struct net_device *dev) | |
599 | { | |
600 | struct netdev_boot_setup *s = dev_boot_setup; | |
601 | int i; | |
602 | ||
603 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
604 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
93b3cff9 | 605 | !strcmp(dev->name, s[i].name)) { |
1da177e4 LT |
606 | dev->irq = s[i].map.irq; |
607 | dev->base_addr = s[i].map.base_addr; | |
608 | dev->mem_start = s[i].map.mem_start; | |
609 | dev->mem_end = s[i].map.mem_end; | |
610 | return 1; | |
611 | } | |
612 | } | |
613 | return 0; | |
614 | } | |
d1b19dff | 615 | EXPORT_SYMBOL(netdev_boot_setup_check); |
1da177e4 LT |
616 | |
617 | ||
618 | /** | |
619 | * netdev_boot_base - get address from boot time settings | |
620 | * @prefix: prefix for network device | |
621 | * @unit: id for network device | |
622 | * | |
623 | * Check boot time settings for the base address of device. | |
624 | * The found settings are set for the device to be used | |
625 | * later in the device probing. | |
626 | * Returns 0 if no settings found. | |
627 | */ | |
628 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
629 | { | |
630 | const struct netdev_boot_setup *s = dev_boot_setup; | |
631 | char name[IFNAMSIZ]; | |
632 | int i; | |
633 | ||
634 | sprintf(name, "%s%d", prefix, unit); | |
635 | ||
636 | /* | |
637 | * If device already registered then return base of 1 | |
638 | * to indicate not to probe for this interface | |
639 | */ | |
881d966b | 640 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
641 | return 1; |
642 | ||
643 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
644 | if (!strcmp(name, s[i].name)) | |
645 | return s[i].map.base_addr; | |
646 | return 0; | |
647 | } | |
648 | ||
649 | /* | |
650 | * Saves at boot time configured settings for any netdevice. | |
651 | */ | |
652 | int __init netdev_boot_setup(char *str) | |
653 | { | |
654 | int ints[5]; | |
655 | struct ifmap map; | |
656 | ||
657 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
658 | if (!str || !*str) | |
659 | return 0; | |
660 | ||
661 | /* Save settings */ | |
662 | memset(&map, 0, sizeof(map)); | |
663 | if (ints[0] > 0) | |
664 | map.irq = ints[1]; | |
665 | if (ints[0] > 1) | |
666 | map.base_addr = ints[2]; | |
667 | if (ints[0] > 2) | |
668 | map.mem_start = ints[3]; | |
669 | if (ints[0] > 3) | |
670 | map.mem_end = ints[4]; | |
671 | ||
672 | /* Add new entry to the list */ | |
673 | return netdev_boot_setup_add(str, &map); | |
674 | } | |
675 | ||
676 | __setup("netdev=", netdev_boot_setup); | |
677 | ||
678 | /******************************************************************************* | |
679 | ||
680 | Device Interface Subroutines | |
681 | ||
682 | *******************************************************************************/ | |
683 | ||
684 | /** | |
685 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 686 | * @net: the applicable net namespace |
1da177e4 LT |
687 | * @name: name to find |
688 | * | |
689 | * Find an interface by name. Must be called under RTNL semaphore | |
690 | * or @dev_base_lock. If the name is found a pointer to the device | |
691 | * is returned. If the name is not found then %NULL is returned. The | |
692 | * reference counters are not incremented so the caller must be | |
693 | * careful with locks. | |
694 | */ | |
695 | ||
881d966b | 696 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
697 | { |
698 | struct hlist_node *p; | |
0bd8d536 ED |
699 | struct net_device *dev; |
700 | struct hlist_head *head = dev_name_hash(net, name); | |
1da177e4 | 701 | |
0bd8d536 | 702 | hlist_for_each_entry(dev, p, head, name_hlist) |
1da177e4 LT |
703 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
704 | return dev; | |
0bd8d536 | 705 | |
1da177e4 LT |
706 | return NULL; |
707 | } | |
d1b19dff | 708 | EXPORT_SYMBOL(__dev_get_by_name); |
1da177e4 | 709 | |
72c9528b ED |
710 | /** |
711 | * dev_get_by_name_rcu - find a device by its name | |
712 | * @net: the applicable net namespace | |
713 | * @name: name to find | |
714 | * | |
715 | * Find an interface by name. | |
716 | * If the name is found a pointer to the device is returned. | |
717 | * If the name is not found then %NULL is returned. | |
718 | * The reference counters are not incremented so the caller must be | |
719 | * careful with locks. The caller must hold RCU lock. | |
720 | */ | |
721 | ||
722 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
723 | { | |
724 | struct hlist_node *p; | |
725 | struct net_device *dev; | |
726 | struct hlist_head *head = dev_name_hash(net, name); | |
727 | ||
728 | hlist_for_each_entry_rcu(dev, p, head, name_hlist) | |
729 | if (!strncmp(dev->name, name, IFNAMSIZ)) | |
730 | return dev; | |
731 | ||
732 | return NULL; | |
733 | } | |
734 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
735 | ||
1da177e4 LT |
736 | /** |
737 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 738 | * @net: the applicable net namespace |
1da177e4 LT |
739 | * @name: name to find |
740 | * | |
741 | * Find an interface by name. This can be called from any | |
742 | * context and does its own locking. The returned handle has | |
743 | * the usage count incremented and the caller must use dev_put() to | |
744 | * release it when it is no longer needed. %NULL is returned if no | |
745 | * matching device is found. | |
746 | */ | |
747 | ||
881d966b | 748 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
749 | { |
750 | struct net_device *dev; | |
751 | ||
72c9528b ED |
752 | rcu_read_lock(); |
753 | dev = dev_get_by_name_rcu(net, name); | |
1da177e4 LT |
754 | if (dev) |
755 | dev_hold(dev); | |
72c9528b | 756 | rcu_read_unlock(); |
1da177e4 LT |
757 | return dev; |
758 | } | |
d1b19dff | 759 | EXPORT_SYMBOL(dev_get_by_name); |
1da177e4 LT |
760 | |
761 | /** | |
762 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 763 | * @net: the applicable net namespace |
1da177e4 LT |
764 | * @ifindex: index of device |
765 | * | |
766 | * Search for an interface by index. Returns %NULL if the device | |
767 | * is not found or a pointer to the device. The device has not | |
768 | * had its reference counter increased so the caller must be careful | |
769 | * about locking. The caller must hold either the RTNL semaphore | |
770 | * or @dev_base_lock. | |
771 | */ | |
772 | ||
881d966b | 773 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
774 | { |
775 | struct hlist_node *p; | |
0bd8d536 ED |
776 | struct net_device *dev; |
777 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
1da177e4 | 778 | |
0bd8d536 | 779 | hlist_for_each_entry(dev, p, head, index_hlist) |
1da177e4 LT |
780 | if (dev->ifindex == ifindex) |
781 | return dev; | |
0bd8d536 | 782 | |
1da177e4 LT |
783 | return NULL; |
784 | } | |
d1b19dff | 785 | EXPORT_SYMBOL(__dev_get_by_index); |
1da177e4 | 786 | |
fb699dfd ED |
787 | /** |
788 | * dev_get_by_index_rcu - find a device by its ifindex | |
789 | * @net: the applicable net namespace | |
790 | * @ifindex: index of device | |
791 | * | |
792 | * Search for an interface by index. Returns %NULL if the device | |
793 | * is not found or a pointer to the device. The device has not | |
794 | * had its reference counter increased so the caller must be careful | |
795 | * about locking. The caller must hold RCU lock. | |
796 | */ | |
797 | ||
798 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
799 | { | |
800 | struct hlist_node *p; | |
801 | struct net_device *dev; | |
802 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
803 | ||
804 | hlist_for_each_entry_rcu(dev, p, head, index_hlist) | |
805 | if (dev->ifindex == ifindex) | |
806 | return dev; | |
807 | ||
808 | return NULL; | |
809 | } | |
810 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
811 | ||
1da177e4 LT |
812 | |
813 | /** | |
814 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 815 | * @net: the applicable net namespace |
1da177e4 LT |
816 | * @ifindex: index of device |
817 | * | |
818 | * Search for an interface by index. Returns NULL if the device | |
819 | * is not found or a pointer to the device. The device returned has | |
820 | * had a reference added and the pointer is safe until the user calls | |
821 | * dev_put to indicate they have finished with it. | |
822 | */ | |
823 | ||
881d966b | 824 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
825 | { |
826 | struct net_device *dev; | |
827 | ||
fb699dfd ED |
828 | rcu_read_lock(); |
829 | dev = dev_get_by_index_rcu(net, ifindex); | |
1da177e4 LT |
830 | if (dev) |
831 | dev_hold(dev); | |
fb699dfd | 832 | rcu_read_unlock(); |
1da177e4 LT |
833 | return dev; |
834 | } | |
d1b19dff | 835 | EXPORT_SYMBOL(dev_get_by_index); |
1da177e4 LT |
836 | |
837 | /** | |
941666c2 | 838 | * dev_getbyhwaddr_rcu - find a device by its hardware address |
c4ea43c5 | 839 | * @net: the applicable net namespace |
1da177e4 LT |
840 | * @type: media type of device |
841 | * @ha: hardware address | |
842 | * | |
843 | * Search for an interface by MAC address. Returns NULL if the device | |
c506653d ED |
844 | * is not found or a pointer to the device. |
845 | * The caller must hold RCU or RTNL. | |
941666c2 | 846 | * The returned device has not had its ref count increased |
1da177e4 LT |
847 | * and the caller must therefore be careful about locking |
848 | * | |
1da177e4 LT |
849 | */ |
850 | ||
941666c2 ED |
851 | struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, |
852 | const char *ha) | |
1da177e4 LT |
853 | { |
854 | struct net_device *dev; | |
855 | ||
941666c2 | 856 | for_each_netdev_rcu(net, dev) |
1da177e4 LT |
857 | if (dev->type == type && |
858 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
859 | return dev; |
860 | ||
861 | return NULL; | |
1da177e4 | 862 | } |
941666c2 | 863 | EXPORT_SYMBOL(dev_getbyhwaddr_rcu); |
cf309e3f | 864 | |
881d966b | 865 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
866 | { |
867 | struct net_device *dev; | |
868 | ||
4e9cac2b | 869 | ASSERT_RTNL(); |
881d966b | 870 | for_each_netdev(net, dev) |
4e9cac2b | 871 | if (dev->type == type) |
7562f876 PE |
872 | return dev; |
873 | ||
874 | return NULL; | |
4e9cac2b | 875 | } |
4e9cac2b PM |
876 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); |
877 | ||
881d966b | 878 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b | 879 | { |
99fe3c39 | 880 | struct net_device *dev, *ret = NULL; |
4e9cac2b | 881 | |
99fe3c39 ED |
882 | rcu_read_lock(); |
883 | for_each_netdev_rcu(net, dev) | |
884 | if (dev->type == type) { | |
885 | dev_hold(dev); | |
886 | ret = dev; | |
887 | break; | |
888 | } | |
889 | rcu_read_unlock(); | |
890 | return ret; | |
1da177e4 | 891 | } |
1da177e4 LT |
892 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
893 | ||
894 | /** | |
bb69ae04 | 895 | * dev_get_by_flags_rcu - find any device with given flags |
c4ea43c5 | 896 | * @net: the applicable net namespace |
1da177e4 LT |
897 | * @if_flags: IFF_* values |
898 | * @mask: bitmask of bits in if_flags to check | |
899 | * | |
900 | * Search for any interface with the given flags. Returns NULL if a device | |
bb69ae04 ED |
901 | * is not found or a pointer to the device. Must be called inside |
902 | * rcu_read_lock(), and result refcount is unchanged. | |
1da177e4 LT |
903 | */ |
904 | ||
bb69ae04 | 905 | struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags, |
d1b19dff | 906 | unsigned short mask) |
1da177e4 | 907 | { |
7562f876 | 908 | struct net_device *dev, *ret; |
1da177e4 | 909 | |
7562f876 | 910 | ret = NULL; |
c6d14c84 | 911 | for_each_netdev_rcu(net, dev) { |
1da177e4 | 912 | if (((dev->flags ^ if_flags) & mask) == 0) { |
7562f876 | 913 | ret = dev; |
1da177e4 LT |
914 | break; |
915 | } | |
916 | } | |
7562f876 | 917 | return ret; |
1da177e4 | 918 | } |
bb69ae04 | 919 | EXPORT_SYMBOL(dev_get_by_flags_rcu); |
1da177e4 LT |
920 | |
921 | /** | |
922 | * dev_valid_name - check if name is okay for network device | |
923 | * @name: name string | |
924 | * | |
925 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
926 | * to allow sysfs to work. We also disallow any kind of |
927 | * whitespace. | |
1da177e4 | 928 | */ |
95f050bf | 929 | bool dev_valid_name(const char *name) |
1da177e4 | 930 | { |
c7fa9d18 | 931 | if (*name == '\0') |
95f050bf | 932 | return false; |
b6fe17d6 | 933 | if (strlen(name) >= IFNAMSIZ) |
95f050bf | 934 | return false; |
c7fa9d18 | 935 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
95f050bf | 936 | return false; |
c7fa9d18 DM |
937 | |
938 | while (*name) { | |
939 | if (*name == '/' || isspace(*name)) | |
95f050bf | 940 | return false; |
c7fa9d18 DM |
941 | name++; |
942 | } | |
95f050bf | 943 | return true; |
1da177e4 | 944 | } |
d1b19dff | 945 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
946 | |
947 | /** | |
b267b179 EB |
948 | * __dev_alloc_name - allocate a name for a device |
949 | * @net: network namespace to allocate the device name in | |
1da177e4 | 950 | * @name: name format string |
b267b179 | 951 | * @buf: scratch buffer and result name string |
1da177e4 LT |
952 | * |
953 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
954 | * id. It scans list of devices to build up a free map, then chooses |
955 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
956 | * while allocating the name and adding the device in order to avoid | |
957 | * duplicates. | |
958 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
959 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
960 | */ |
961 | ||
b267b179 | 962 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
963 | { |
964 | int i = 0; | |
1da177e4 LT |
965 | const char *p; |
966 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 967 | unsigned long *inuse; |
1da177e4 LT |
968 | struct net_device *d; |
969 | ||
970 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
971 | if (p) { | |
972 | /* | |
973 | * Verify the string as this thing may have come from | |
974 | * the user. There must be either one "%d" and no other "%" | |
975 | * characters. | |
976 | */ | |
977 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
978 | return -EINVAL; | |
979 | ||
980 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 981 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
982 | if (!inuse) |
983 | return -ENOMEM; | |
984 | ||
881d966b | 985 | for_each_netdev(net, d) { |
1da177e4 LT |
986 | if (!sscanf(d->name, name, &i)) |
987 | continue; | |
988 | if (i < 0 || i >= max_netdevices) | |
989 | continue; | |
990 | ||
991 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 992 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
993 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
994 | set_bit(i, inuse); | |
995 | } | |
996 | ||
997 | i = find_first_zero_bit(inuse, max_netdevices); | |
998 | free_page((unsigned long) inuse); | |
999 | } | |
1000 | ||
d9031024 OP |
1001 | if (buf != name) |
1002 | snprintf(buf, IFNAMSIZ, name, i); | |
b267b179 | 1003 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 1004 | return i; |
1da177e4 LT |
1005 | |
1006 | /* It is possible to run out of possible slots | |
1007 | * when the name is long and there isn't enough space left | |
1008 | * for the digits, or if all bits are used. | |
1009 | */ | |
1010 | return -ENFILE; | |
1011 | } | |
1012 | ||
b267b179 EB |
1013 | /** |
1014 | * dev_alloc_name - allocate a name for a device | |
1015 | * @dev: device | |
1016 | * @name: name format string | |
1017 | * | |
1018 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
1019 | * id. It scans list of devices to build up a free map, then chooses | |
1020 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1021 | * while allocating the name and adding the device in order to avoid | |
1022 | * duplicates. | |
1023 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1024 | * Returns the number of the unit assigned or a negative errno code. | |
1025 | */ | |
1026 | ||
1027 | int dev_alloc_name(struct net_device *dev, const char *name) | |
1028 | { | |
1029 | char buf[IFNAMSIZ]; | |
1030 | struct net *net; | |
1031 | int ret; | |
1032 | ||
c346dca1 YH |
1033 | BUG_ON(!dev_net(dev)); |
1034 | net = dev_net(dev); | |
b267b179 EB |
1035 | ret = __dev_alloc_name(net, name, buf); |
1036 | if (ret >= 0) | |
1037 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1038 | return ret; | |
1039 | } | |
d1b19dff | 1040 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 1041 | |
828de4f6 G |
1042 | static int dev_alloc_name_ns(struct net *net, |
1043 | struct net_device *dev, | |
1044 | const char *name) | |
d9031024 | 1045 | { |
828de4f6 G |
1046 | char buf[IFNAMSIZ]; |
1047 | int ret; | |
8ce6cebc | 1048 | |
828de4f6 G |
1049 | ret = __dev_alloc_name(net, name, buf); |
1050 | if (ret >= 0) | |
1051 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1052 | return ret; | |
1053 | } | |
1054 | ||
1055 | static int dev_get_valid_name(struct net *net, | |
1056 | struct net_device *dev, | |
1057 | const char *name) | |
1058 | { | |
1059 | BUG_ON(!net); | |
8ce6cebc | 1060 | |
d9031024 OP |
1061 | if (!dev_valid_name(name)) |
1062 | return -EINVAL; | |
1063 | ||
1c5cae81 | 1064 | if (strchr(name, '%')) |
828de4f6 | 1065 | return dev_alloc_name_ns(net, dev, name); |
d9031024 OP |
1066 | else if (__dev_get_by_name(net, name)) |
1067 | return -EEXIST; | |
8ce6cebc DL |
1068 | else if (dev->name != name) |
1069 | strlcpy(dev->name, name, IFNAMSIZ); | |
d9031024 OP |
1070 | |
1071 | return 0; | |
1072 | } | |
1da177e4 LT |
1073 | |
1074 | /** | |
1075 | * dev_change_name - change name of a device | |
1076 | * @dev: device | |
1077 | * @newname: name (or format string) must be at least IFNAMSIZ | |
1078 | * | |
1079 | * Change name of a device, can pass format strings "eth%d". | |
1080 | * for wildcarding. | |
1081 | */ | |
cf04a4c7 | 1082 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 1083 | { |
fcc5a03a | 1084 | char oldname[IFNAMSIZ]; |
1da177e4 | 1085 | int err = 0; |
fcc5a03a | 1086 | int ret; |
881d966b | 1087 | struct net *net; |
1da177e4 LT |
1088 | |
1089 | ASSERT_RTNL(); | |
c346dca1 | 1090 | BUG_ON(!dev_net(dev)); |
1da177e4 | 1091 | |
c346dca1 | 1092 | net = dev_net(dev); |
1da177e4 LT |
1093 | if (dev->flags & IFF_UP) |
1094 | return -EBUSY; | |
1095 | ||
30e6c9fa | 1096 | write_seqcount_begin(&devnet_rename_seq); |
c91f6df2 BH |
1097 | |
1098 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { | |
30e6c9fa | 1099 | write_seqcount_end(&devnet_rename_seq); |
c8d90dca | 1100 | return 0; |
c91f6df2 | 1101 | } |
c8d90dca | 1102 | |
fcc5a03a HX |
1103 | memcpy(oldname, dev->name, IFNAMSIZ); |
1104 | ||
828de4f6 | 1105 | err = dev_get_valid_name(net, dev, newname); |
c91f6df2 | 1106 | if (err < 0) { |
30e6c9fa | 1107 | write_seqcount_end(&devnet_rename_seq); |
d9031024 | 1108 | return err; |
c91f6df2 | 1109 | } |
1da177e4 | 1110 | |
fcc5a03a | 1111 | rollback: |
a1b3f594 EB |
1112 | ret = device_rename(&dev->dev, dev->name); |
1113 | if (ret) { | |
1114 | memcpy(dev->name, oldname, IFNAMSIZ); | |
30e6c9fa | 1115 | write_seqcount_end(&devnet_rename_seq); |
a1b3f594 | 1116 | return ret; |
dcc99773 | 1117 | } |
7f988eab | 1118 | |
30e6c9fa | 1119 | write_seqcount_end(&devnet_rename_seq); |
c91f6df2 | 1120 | |
7f988eab | 1121 | write_lock_bh(&dev_base_lock); |
372b2312 | 1122 | hlist_del_rcu(&dev->name_hlist); |
72c9528b ED |
1123 | write_unlock_bh(&dev_base_lock); |
1124 | ||
1125 | synchronize_rcu(); | |
1126 | ||
1127 | write_lock_bh(&dev_base_lock); | |
1128 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
7f988eab HX |
1129 | write_unlock_bh(&dev_base_lock); |
1130 | ||
056925ab | 1131 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1132 | ret = notifier_to_errno(ret); |
1133 | ||
1134 | if (ret) { | |
91e9c07b ED |
1135 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1136 | if (err >= 0) { | |
fcc5a03a | 1137 | err = ret; |
30e6c9fa | 1138 | write_seqcount_begin(&devnet_rename_seq); |
fcc5a03a HX |
1139 | memcpy(dev->name, oldname, IFNAMSIZ); |
1140 | goto rollback; | |
91e9c07b | 1141 | } else { |
7b6cd1ce | 1142 | pr_err("%s: name change rollback failed: %d\n", |
91e9c07b | 1143 | dev->name, ret); |
fcc5a03a HX |
1144 | } |
1145 | } | |
1da177e4 LT |
1146 | |
1147 | return err; | |
1148 | } | |
1149 | ||
0b815a1a SH |
1150 | /** |
1151 | * dev_set_alias - change ifalias of a device | |
1152 | * @dev: device | |
1153 | * @alias: name up to IFALIASZ | |
f0db275a | 1154 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1155 | * |
1156 | * Set ifalias for a device, | |
1157 | */ | |
1158 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1159 | { | |
7364e445 AK |
1160 | char *new_ifalias; |
1161 | ||
0b815a1a SH |
1162 | ASSERT_RTNL(); |
1163 | ||
1164 | if (len >= IFALIASZ) | |
1165 | return -EINVAL; | |
1166 | ||
96ca4a2c | 1167 | if (!len) { |
388dfc2d SK |
1168 | kfree(dev->ifalias); |
1169 | dev->ifalias = NULL; | |
96ca4a2c OH |
1170 | return 0; |
1171 | } | |
1172 | ||
7364e445 AK |
1173 | new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL); |
1174 | if (!new_ifalias) | |
0b815a1a | 1175 | return -ENOMEM; |
7364e445 | 1176 | dev->ifalias = new_ifalias; |
0b815a1a SH |
1177 | |
1178 | strlcpy(dev->ifalias, alias, len+1); | |
1179 | return len; | |
1180 | } | |
1181 | ||
1182 | ||
d8a33ac4 | 1183 | /** |
3041a069 | 1184 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1185 | * @dev: device to cause notification |
1186 | * | |
1187 | * Called to indicate a device has changed features. | |
1188 | */ | |
1189 | void netdev_features_change(struct net_device *dev) | |
1190 | { | |
056925ab | 1191 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1192 | } |
1193 | EXPORT_SYMBOL(netdev_features_change); | |
1194 | ||
1da177e4 LT |
1195 | /** |
1196 | * netdev_state_change - device changes state | |
1197 | * @dev: device to cause notification | |
1198 | * | |
1199 | * Called to indicate a device has changed state. This function calls | |
1200 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1201 | * to the routing socket. | |
1202 | */ | |
1203 | void netdev_state_change(struct net_device *dev) | |
1204 | { | |
1205 | if (dev->flags & IFF_UP) { | |
056925ab | 1206 | call_netdevice_notifiers(NETDEV_CHANGE, dev); |
1da177e4 LT |
1207 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0); |
1208 | } | |
1209 | } | |
d1b19dff | 1210 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1211 | |
ee89bab1 AW |
1212 | /** |
1213 | * netdev_notify_peers - notify network peers about existence of @dev | |
1214 | * @dev: network device | |
1215 | * | |
1216 | * Generate traffic such that interested network peers are aware of | |
1217 | * @dev, such as by generating a gratuitous ARP. This may be used when | |
1218 | * a device wants to inform the rest of the network about some sort of | |
1219 | * reconfiguration such as a failover event or virtual machine | |
1220 | * migration. | |
1221 | */ | |
1222 | void netdev_notify_peers(struct net_device *dev) | |
c1da4ac7 | 1223 | { |
ee89bab1 AW |
1224 | rtnl_lock(); |
1225 | call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); | |
1226 | rtnl_unlock(); | |
c1da4ac7 | 1227 | } |
ee89bab1 | 1228 | EXPORT_SYMBOL(netdev_notify_peers); |
c1da4ac7 | 1229 | |
1da177e4 LT |
1230 | /** |
1231 | * dev_load - load a network module | |
c4ea43c5 | 1232 | * @net: the applicable net namespace |
1da177e4 LT |
1233 | * @name: name of interface |
1234 | * | |
1235 | * If a network interface is not present and the process has suitable | |
1236 | * privileges this function loads the module. If module loading is not | |
1237 | * available in this kernel then it becomes a nop. | |
1238 | */ | |
1239 | ||
881d966b | 1240 | void dev_load(struct net *net, const char *name) |
1da177e4 | 1241 | { |
4ec93edb | 1242 | struct net_device *dev; |
8909c9ad | 1243 | int no_module; |
1da177e4 | 1244 | |
72c9528b ED |
1245 | rcu_read_lock(); |
1246 | dev = dev_get_by_name_rcu(net, name); | |
1247 | rcu_read_unlock(); | |
1da177e4 | 1248 | |
8909c9ad VK |
1249 | no_module = !dev; |
1250 | if (no_module && capable(CAP_NET_ADMIN)) | |
1251 | no_module = request_module("netdev-%s", name); | |
1252 | if (no_module && capable(CAP_SYS_MODULE)) { | |
1253 | if (!request_module("%s", name)) | |
7cecb523 VL |
1254 | pr_warn("Loading kernel module for a network device with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s instead.\n", |
1255 | name); | |
8909c9ad | 1256 | } |
1da177e4 | 1257 | } |
d1b19dff | 1258 | EXPORT_SYMBOL(dev_load); |
1da177e4 | 1259 | |
bd380811 | 1260 | static int __dev_open(struct net_device *dev) |
1da177e4 | 1261 | { |
d314774c | 1262 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1263 | int ret; |
1da177e4 | 1264 | |
e46b66bc BH |
1265 | ASSERT_RTNL(); |
1266 | ||
1da177e4 LT |
1267 | if (!netif_device_present(dev)) |
1268 | return -ENODEV; | |
1269 | ||
3b8bcfd5 JB |
1270 | ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); |
1271 | ret = notifier_to_errno(ret); | |
1272 | if (ret) | |
1273 | return ret; | |
1274 | ||
1da177e4 | 1275 | set_bit(__LINK_STATE_START, &dev->state); |
bada339b | 1276 | |
d314774c SH |
1277 | if (ops->ndo_validate_addr) |
1278 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1279 | |
d314774c SH |
1280 | if (!ret && ops->ndo_open) |
1281 | ret = ops->ndo_open(dev); | |
1da177e4 | 1282 | |
bada339b JG |
1283 | if (ret) |
1284 | clear_bit(__LINK_STATE_START, &dev->state); | |
1285 | else { | |
1da177e4 | 1286 | dev->flags |= IFF_UP; |
b4bd07c2 | 1287 | net_dmaengine_get(); |
4417da66 | 1288 | dev_set_rx_mode(dev); |
1da177e4 | 1289 | dev_activate(dev); |
7bf23575 | 1290 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 1291 | } |
bada339b | 1292 | |
1da177e4 LT |
1293 | return ret; |
1294 | } | |
1295 | ||
1296 | /** | |
bd380811 PM |
1297 | * dev_open - prepare an interface for use. |
1298 | * @dev: device to open | |
1da177e4 | 1299 | * |
bd380811 PM |
1300 | * Takes a device from down to up state. The device's private open |
1301 | * function is invoked and then the multicast lists are loaded. Finally | |
1302 | * the device is moved into the up state and a %NETDEV_UP message is | |
1303 | * sent to the netdev notifier chain. | |
1304 | * | |
1305 | * Calling this function on an active interface is a nop. On a failure | |
1306 | * a negative errno code is returned. | |
1da177e4 | 1307 | */ |
bd380811 PM |
1308 | int dev_open(struct net_device *dev) |
1309 | { | |
1310 | int ret; | |
1311 | ||
bd380811 PM |
1312 | if (dev->flags & IFF_UP) |
1313 | return 0; | |
1314 | ||
bd380811 PM |
1315 | ret = __dev_open(dev); |
1316 | if (ret < 0) | |
1317 | return ret; | |
1318 | ||
bd380811 PM |
1319 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING); |
1320 | call_netdevice_notifiers(NETDEV_UP, dev); | |
1321 | ||
1322 | return ret; | |
1323 | } | |
1324 | EXPORT_SYMBOL(dev_open); | |
1325 | ||
44345724 | 1326 | static int __dev_close_many(struct list_head *head) |
1da177e4 | 1327 | { |
44345724 | 1328 | struct net_device *dev; |
e46b66bc | 1329 | |
bd380811 | 1330 | ASSERT_RTNL(); |
9d5010db DM |
1331 | might_sleep(); |
1332 | ||
44345724 | 1333 | list_for_each_entry(dev, head, unreg_list) { |
44345724 | 1334 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1335 | |
44345724 | 1336 | clear_bit(__LINK_STATE_START, &dev->state); |
1da177e4 | 1337 | |
44345724 OP |
1338 | /* Synchronize to scheduled poll. We cannot touch poll list, it |
1339 | * can be even on different cpu. So just clear netif_running(). | |
1340 | * | |
1341 | * dev->stop() will invoke napi_disable() on all of it's | |
1342 | * napi_struct instances on this device. | |
1343 | */ | |
1344 | smp_mb__after_clear_bit(); /* Commit netif_running(). */ | |
1345 | } | |
1da177e4 | 1346 | |
44345724 | 1347 | dev_deactivate_many(head); |
d8b2a4d2 | 1348 | |
44345724 OP |
1349 | list_for_each_entry(dev, head, unreg_list) { |
1350 | const struct net_device_ops *ops = dev->netdev_ops; | |
1da177e4 | 1351 | |
44345724 OP |
1352 | /* |
1353 | * Call the device specific close. This cannot fail. | |
1354 | * Only if device is UP | |
1355 | * | |
1356 | * We allow it to be called even after a DETACH hot-plug | |
1357 | * event. | |
1358 | */ | |
1359 | if (ops->ndo_stop) | |
1360 | ops->ndo_stop(dev); | |
1361 | ||
44345724 | 1362 | dev->flags &= ~IFF_UP; |
44345724 OP |
1363 | net_dmaengine_put(); |
1364 | } | |
1365 | ||
1366 | return 0; | |
1367 | } | |
1368 | ||
1369 | static int __dev_close(struct net_device *dev) | |
1370 | { | |
f87e6f47 | 1371 | int retval; |
44345724 OP |
1372 | LIST_HEAD(single); |
1373 | ||
1374 | list_add(&dev->unreg_list, &single); | |
f87e6f47 LT |
1375 | retval = __dev_close_many(&single); |
1376 | list_del(&single); | |
1377 | return retval; | |
44345724 OP |
1378 | } |
1379 | ||
3fbd8758 | 1380 | static int dev_close_many(struct list_head *head) |
44345724 OP |
1381 | { |
1382 | struct net_device *dev, *tmp; | |
1383 | LIST_HEAD(tmp_list); | |
1da177e4 | 1384 | |
44345724 OP |
1385 | list_for_each_entry_safe(dev, tmp, head, unreg_list) |
1386 | if (!(dev->flags & IFF_UP)) | |
1387 | list_move(&dev->unreg_list, &tmp_list); | |
1388 | ||
1389 | __dev_close_many(head); | |
1da177e4 | 1390 | |
44345724 OP |
1391 | list_for_each_entry(dev, head, unreg_list) { |
1392 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING); | |
1393 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
1394 | } | |
bd380811 | 1395 | |
44345724 OP |
1396 | /* rollback_registered_many needs the complete original list */ |
1397 | list_splice(&tmp_list, head); | |
bd380811 PM |
1398 | return 0; |
1399 | } | |
1400 | ||
1401 | /** | |
1402 | * dev_close - shutdown an interface. | |
1403 | * @dev: device to shutdown | |
1404 | * | |
1405 | * This function moves an active device into down state. A | |
1406 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1407 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1408 | * chain. | |
1409 | */ | |
1410 | int dev_close(struct net_device *dev) | |
1411 | { | |
e14a5993 ED |
1412 | if (dev->flags & IFF_UP) { |
1413 | LIST_HEAD(single); | |
1da177e4 | 1414 | |
e14a5993 ED |
1415 | list_add(&dev->unreg_list, &single); |
1416 | dev_close_many(&single); | |
1417 | list_del(&single); | |
1418 | } | |
1da177e4 LT |
1419 | return 0; |
1420 | } | |
d1b19dff | 1421 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1422 | |
1423 | ||
0187bdfb BH |
1424 | /** |
1425 | * dev_disable_lro - disable Large Receive Offload on a device | |
1426 | * @dev: device | |
1427 | * | |
1428 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1429 | * called under RTNL. This is needed if received packets may be | |
1430 | * forwarded to another interface. | |
1431 | */ | |
1432 | void dev_disable_lro(struct net_device *dev) | |
1433 | { | |
f11970e3 NH |
1434 | /* |
1435 | * If we're trying to disable lro on a vlan device | |
1436 | * use the underlying physical device instead | |
1437 | */ | |
1438 | if (is_vlan_dev(dev)) | |
1439 | dev = vlan_dev_real_dev(dev); | |
1440 | ||
bc5787c6 MM |
1441 | dev->wanted_features &= ~NETIF_F_LRO; |
1442 | netdev_update_features(dev); | |
27660515 | 1443 | |
22d5969f MM |
1444 | if (unlikely(dev->features & NETIF_F_LRO)) |
1445 | netdev_WARN(dev, "failed to disable LRO!\n"); | |
0187bdfb BH |
1446 | } |
1447 | EXPORT_SYMBOL(dev_disable_lro); | |
1448 | ||
1449 | ||
881d966b EB |
1450 | static int dev_boot_phase = 1; |
1451 | ||
1da177e4 LT |
1452 | /** |
1453 | * register_netdevice_notifier - register a network notifier block | |
1454 | * @nb: notifier | |
1455 | * | |
1456 | * Register a notifier to be called when network device events occur. | |
1457 | * The notifier passed is linked into the kernel structures and must | |
1458 | * not be reused until it has been unregistered. A negative errno code | |
1459 | * is returned on a failure. | |
1460 | * | |
1461 | * When registered all registration and up events are replayed | |
4ec93edb | 1462 | * to the new notifier to allow device to have a race free |
1da177e4 LT |
1463 | * view of the network device list. |
1464 | */ | |
1465 | ||
1466 | int register_netdevice_notifier(struct notifier_block *nb) | |
1467 | { | |
1468 | struct net_device *dev; | |
fcc5a03a | 1469 | struct net_device *last; |
881d966b | 1470 | struct net *net; |
1da177e4 LT |
1471 | int err; |
1472 | ||
1473 | rtnl_lock(); | |
f07d5b94 | 1474 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1475 | if (err) |
1476 | goto unlock; | |
881d966b EB |
1477 | if (dev_boot_phase) |
1478 | goto unlock; | |
1479 | for_each_net(net) { | |
1480 | for_each_netdev(net, dev) { | |
1481 | err = nb->notifier_call(nb, NETDEV_REGISTER, dev); | |
1482 | err = notifier_to_errno(err); | |
1483 | if (err) | |
1484 | goto rollback; | |
1485 | ||
1486 | if (!(dev->flags & IFF_UP)) | |
1487 | continue; | |
1da177e4 | 1488 | |
881d966b EB |
1489 | nb->notifier_call(nb, NETDEV_UP, dev); |
1490 | } | |
1da177e4 | 1491 | } |
fcc5a03a HX |
1492 | |
1493 | unlock: | |
1da177e4 LT |
1494 | rtnl_unlock(); |
1495 | return err; | |
fcc5a03a HX |
1496 | |
1497 | rollback: | |
1498 | last = dev; | |
881d966b EB |
1499 | for_each_net(net) { |
1500 | for_each_netdev(net, dev) { | |
1501 | if (dev == last) | |
8f891489 | 1502 | goto outroll; |
fcc5a03a | 1503 | |
881d966b EB |
1504 | if (dev->flags & IFF_UP) { |
1505 | nb->notifier_call(nb, NETDEV_GOING_DOWN, dev); | |
1506 | nb->notifier_call(nb, NETDEV_DOWN, dev); | |
1507 | } | |
1508 | nb->notifier_call(nb, NETDEV_UNREGISTER, dev); | |
fcc5a03a | 1509 | } |
fcc5a03a | 1510 | } |
c67625a1 | 1511 | |
8f891489 | 1512 | outroll: |
c67625a1 | 1513 | raw_notifier_chain_unregister(&netdev_chain, nb); |
fcc5a03a | 1514 | goto unlock; |
1da177e4 | 1515 | } |
d1b19dff | 1516 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1517 | |
1518 | /** | |
1519 | * unregister_netdevice_notifier - unregister a network notifier block | |
1520 | * @nb: notifier | |
1521 | * | |
1522 | * Unregister a notifier previously registered by | |
1523 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1524 | * kernel structures and may then be reused. A negative errno code | |
1525 | * is returned on a failure. | |
7d3d43da EB |
1526 | * |
1527 | * After unregistering unregister and down device events are synthesized | |
1528 | * for all devices on the device list to the removed notifier to remove | |
1529 | * the need for special case cleanup code. | |
1da177e4 LT |
1530 | */ |
1531 | ||
1532 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1533 | { | |
7d3d43da EB |
1534 | struct net_device *dev; |
1535 | struct net *net; | |
9f514950 HX |
1536 | int err; |
1537 | ||
1538 | rtnl_lock(); | |
f07d5b94 | 1539 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
7d3d43da EB |
1540 | if (err) |
1541 | goto unlock; | |
1542 | ||
1543 | for_each_net(net) { | |
1544 | for_each_netdev(net, dev) { | |
1545 | if (dev->flags & IFF_UP) { | |
1546 | nb->notifier_call(nb, NETDEV_GOING_DOWN, dev); | |
1547 | nb->notifier_call(nb, NETDEV_DOWN, dev); | |
1548 | } | |
1549 | nb->notifier_call(nb, NETDEV_UNREGISTER, dev); | |
7d3d43da EB |
1550 | } |
1551 | } | |
1552 | unlock: | |
9f514950 HX |
1553 | rtnl_unlock(); |
1554 | return err; | |
1da177e4 | 1555 | } |
d1b19dff | 1556 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 LT |
1557 | |
1558 | /** | |
1559 | * call_netdevice_notifiers - call all network notifier blocks | |
1560 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1561 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1562 | * |
1563 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1564 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1565 | */ |
1566 | ||
ad7379d4 | 1567 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1568 | { |
ab930471 | 1569 | ASSERT_RTNL(); |
ad7379d4 | 1570 | return raw_notifier_call_chain(&netdev_chain, val, dev); |
1da177e4 | 1571 | } |
edf947f1 | 1572 | EXPORT_SYMBOL(call_netdevice_notifiers); |
1da177e4 | 1573 | |
c5905afb | 1574 | static struct static_key netstamp_needed __read_mostly; |
b90e5794 | 1575 | #ifdef HAVE_JUMP_LABEL |
c5905afb | 1576 | /* We are not allowed to call static_key_slow_dec() from irq context |
b90e5794 | 1577 | * If net_disable_timestamp() is called from irq context, defer the |
c5905afb | 1578 | * static_key_slow_dec() calls. |
b90e5794 ED |
1579 | */ |
1580 | static atomic_t netstamp_needed_deferred; | |
1581 | #endif | |
1da177e4 LT |
1582 | |
1583 | void net_enable_timestamp(void) | |
1584 | { | |
b90e5794 ED |
1585 | #ifdef HAVE_JUMP_LABEL |
1586 | int deferred = atomic_xchg(&netstamp_needed_deferred, 0); | |
1587 | ||
1588 | if (deferred) { | |
1589 | while (--deferred) | |
c5905afb | 1590 | static_key_slow_dec(&netstamp_needed); |
b90e5794 ED |
1591 | return; |
1592 | } | |
1593 | #endif | |
1594 | WARN_ON(in_interrupt()); | |
c5905afb | 1595 | static_key_slow_inc(&netstamp_needed); |
1da177e4 | 1596 | } |
d1b19dff | 1597 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
1598 | |
1599 | void net_disable_timestamp(void) | |
1600 | { | |
b90e5794 ED |
1601 | #ifdef HAVE_JUMP_LABEL |
1602 | if (in_interrupt()) { | |
1603 | atomic_inc(&netstamp_needed_deferred); | |
1604 | return; | |
1605 | } | |
1606 | #endif | |
c5905afb | 1607 | static_key_slow_dec(&netstamp_needed); |
1da177e4 | 1608 | } |
d1b19dff | 1609 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 1610 | |
3b098e2d | 1611 | static inline void net_timestamp_set(struct sk_buff *skb) |
1da177e4 | 1612 | { |
588f0330 | 1613 | skb->tstamp.tv64 = 0; |
c5905afb | 1614 | if (static_key_false(&netstamp_needed)) |
a61bbcf2 | 1615 | __net_timestamp(skb); |
1da177e4 LT |
1616 | } |
1617 | ||
588f0330 | 1618 | #define net_timestamp_check(COND, SKB) \ |
c5905afb | 1619 | if (static_key_false(&netstamp_needed)) { \ |
588f0330 ED |
1620 | if ((COND) && !(SKB)->tstamp.tv64) \ |
1621 | __net_timestamp(SKB); \ | |
1622 | } \ | |
3b098e2d | 1623 | |
4dc360c5 RC |
1624 | static int net_hwtstamp_validate(struct ifreq *ifr) |
1625 | { | |
1626 | struct hwtstamp_config cfg; | |
1627 | enum hwtstamp_tx_types tx_type; | |
1628 | enum hwtstamp_rx_filters rx_filter; | |
1629 | int tx_type_valid = 0; | |
1630 | int rx_filter_valid = 0; | |
1631 | ||
1632 | if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg))) | |
1633 | return -EFAULT; | |
1634 | ||
1635 | if (cfg.flags) /* reserved for future extensions */ | |
1636 | return -EINVAL; | |
1637 | ||
1638 | tx_type = cfg.tx_type; | |
1639 | rx_filter = cfg.rx_filter; | |
1640 | ||
1641 | switch (tx_type) { | |
1642 | case HWTSTAMP_TX_OFF: | |
1643 | case HWTSTAMP_TX_ON: | |
1644 | case HWTSTAMP_TX_ONESTEP_SYNC: | |
1645 | tx_type_valid = 1; | |
1646 | break; | |
1647 | } | |
1648 | ||
1649 | switch (rx_filter) { | |
1650 | case HWTSTAMP_FILTER_NONE: | |
1651 | case HWTSTAMP_FILTER_ALL: | |
1652 | case HWTSTAMP_FILTER_SOME: | |
1653 | case HWTSTAMP_FILTER_PTP_V1_L4_EVENT: | |
1654 | case HWTSTAMP_FILTER_PTP_V1_L4_SYNC: | |
1655 | case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ: | |
1656 | case HWTSTAMP_FILTER_PTP_V2_L4_EVENT: | |
1657 | case HWTSTAMP_FILTER_PTP_V2_L4_SYNC: | |
1658 | case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ: | |
1659 | case HWTSTAMP_FILTER_PTP_V2_L2_EVENT: | |
1660 | case HWTSTAMP_FILTER_PTP_V2_L2_SYNC: | |
1661 | case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ: | |
1662 | case HWTSTAMP_FILTER_PTP_V2_EVENT: | |
1663 | case HWTSTAMP_FILTER_PTP_V2_SYNC: | |
1664 | case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ: | |
1665 | rx_filter_valid = 1; | |
1666 | break; | |
1667 | } | |
1668 | ||
1669 | if (!tx_type_valid || !rx_filter_valid) | |
1670 | return -ERANGE; | |
1671 | ||
1672 | return 0; | |
1673 | } | |
1674 | ||
79b569f0 DL |
1675 | static inline bool is_skb_forwardable(struct net_device *dev, |
1676 | struct sk_buff *skb) | |
1677 | { | |
1678 | unsigned int len; | |
1679 | ||
1680 | if (!(dev->flags & IFF_UP)) | |
1681 | return false; | |
1682 | ||
1683 | len = dev->mtu + dev->hard_header_len + VLAN_HLEN; | |
1684 | if (skb->len <= len) | |
1685 | return true; | |
1686 | ||
1687 | /* if TSO is enabled, we don't care about the length as the packet | |
1688 | * could be forwarded without being segmented before | |
1689 | */ | |
1690 | if (skb_is_gso(skb)) | |
1691 | return true; | |
1692 | ||
1693 | return false; | |
1694 | } | |
1695 | ||
44540960 AB |
1696 | /** |
1697 | * dev_forward_skb - loopback an skb to another netif | |
1698 | * | |
1699 | * @dev: destination network device | |
1700 | * @skb: buffer to forward | |
1701 | * | |
1702 | * return values: | |
1703 | * NET_RX_SUCCESS (no congestion) | |
6ec82562 | 1704 | * NET_RX_DROP (packet was dropped, but freed) |
44540960 AB |
1705 | * |
1706 | * dev_forward_skb can be used for injecting an skb from the | |
1707 | * start_xmit function of one device into the receive queue | |
1708 | * of another device. | |
1709 | * | |
1710 | * The receiving device may be in another namespace, so | |
1711 | * we have to clear all information in the skb that could | |
1712 | * impact namespace isolation. | |
1713 | */ | |
1714 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
1715 | { | |
48c83012 MT |
1716 | if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) { |
1717 | if (skb_copy_ubufs(skb, GFP_ATOMIC)) { | |
1718 | atomic_long_inc(&dev->rx_dropped); | |
1719 | kfree_skb(skb); | |
1720 | return NET_RX_DROP; | |
1721 | } | |
1722 | } | |
1723 | ||
44540960 | 1724 | skb_orphan(skb); |
c736eefa | 1725 | nf_reset(skb); |
44540960 | 1726 | |
79b569f0 | 1727 | if (unlikely(!is_skb_forwardable(dev, skb))) { |
caf586e5 | 1728 | atomic_long_inc(&dev->rx_dropped); |
6ec82562 | 1729 | kfree_skb(skb); |
44540960 | 1730 | return NET_RX_DROP; |
6ec82562 | 1731 | } |
3b9785c6 | 1732 | skb->skb_iif = 0; |
59b9997b DM |
1733 | skb->dev = dev; |
1734 | skb_dst_drop(skb); | |
44540960 AB |
1735 | skb->tstamp.tv64 = 0; |
1736 | skb->pkt_type = PACKET_HOST; | |
1737 | skb->protocol = eth_type_trans(skb, dev); | |
59b9997b DM |
1738 | skb->mark = 0; |
1739 | secpath_reset(skb); | |
1740 | nf_reset(skb); | |
44540960 AB |
1741 | return netif_rx(skb); |
1742 | } | |
1743 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
1744 | ||
71d9dec2 CG |
1745 | static inline int deliver_skb(struct sk_buff *skb, |
1746 | struct packet_type *pt_prev, | |
1747 | struct net_device *orig_dev) | |
1748 | { | |
1080e512 MT |
1749 | if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) |
1750 | return -ENOMEM; | |
71d9dec2 CG |
1751 | atomic_inc(&skb->users); |
1752 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1753 | } | |
1754 | ||
c0de08d0 EL |
1755 | static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) |
1756 | { | |
a3d744e9 | 1757 | if (!ptype->af_packet_priv || !skb->sk) |
c0de08d0 EL |
1758 | return false; |
1759 | ||
1760 | if (ptype->id_match) | |
1761 | return ptype->id_match(ptype, skb->sk); | |
1762 | else if ((struct sock *)ptype->af_packet_priv == skb->sk) | |
1763 | return true; | |
1764 | ||
1765 | return false; | |
1766 | } | |
1767 | ||
1da177e4 LT |
1768 | /* |
1769 | * Support routine. Sends outgoing frames to any network | |
1770 | * taps currently in use. | |
1771 | */ | |
1772 | ||
f6a78bfc | 1773 | static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
1774 | { |
1775 | struct packet_type *ptype; | |
71d9dec2 CG |
1776 | struct sk_buff *skb2 = NULL; |
1777 | struct packet_type *pt_prev = NULL; | |
a61bbcf2 | 1778 | |
1da177e4 LT |
1779 | rcu_read_lock(); |
1780 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
1781 | /* Never send packets back to the socket | |
1782 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1783 | */ | |
1784 | if ((ptype->dev == dev || !ptype->dev) && | |
c0de08d0 | 1785 | (!skb_loop_sk(ptype, skb))) { |
71d9dec2 CG |
1786 | if (pt_prev) { |
1787 | deliver_skb(skb2, pt_prev, skb->dev); | |
1788 | pt_prev = ptype; | |
1789 | continue; | |
1790 | } | |
1791 | ||
1792 | skb2 = skb_clone(skb, GFP_ATOMIC); | |
1da177e4 LT |
1793 | if (!skb2) |
1794 | break; | |
1795 | ||
70978182 ED |
1796 | net_timestamp_set(skb2); |
1797 | ||
1da177e4 LT |
1798 | /* skb->nh should be correctly |
1799 | set by sender, so that the second statement is | |
1800 | just protection against buggy protocols. | |
1801 | */ | |
459a98ed | 1802 | skb_reset_mac_header(skb2); |
1da177e4 | 1803 | |
d56f90a7 | 1804 | if (skb_network_header(skb2) < skb2->data || |
27a884dc | 1805 | skb2->network_header > skb2->tail) { |
e87cc472 JP |
1806 | net_crit_ratelimited("protocol %04x is buggy, dev %s\n", |
1807 | ntohs(skb2->protocol), | |
1808 | dev->name); | |
c1d2bbe1 | 1809 | skb_reset_network_header(skb2); |
1da177e4 LT |
1810 | } |
1811 | ||
b0e380b1 | 1812 | skb2->transport_header = skb2->network_header; |
1da177e4 | 1813 | skb2->pkt_type = PACKET_OUTGOING; |
71d9dec2 | 1814 | pt_prev = ptype; |
1da177e4 LT |
1815 | } |
1816 | } | |
71d9dec2 CG |
1817 | if (pt_prev) |
1818 | pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); | |
1da177e4 LT |
1819 | rcu_read_unlock(); |
1820 | } | |
1821 | ||
2c53040f BH |
1822 | /** |
1823 | * netif_setup_tc - Handle tc mappings on real_num_tx_queues change | |
4f57c087 JF |
1824 | * @dev: Network device |
1825 | * @txq: number of queues available | |
1826 | * | |
1827 | * If real_num_tx_queues is changed the tc mappings may no longer be | |
1828 | * valid. To resolve this verify the tc mapping remains valid and if | |
1829 | * not NULL the mapping. With no priorities mapping to this | |
1830 | * offset/count pair it will no longer be used. In the worst case TC0 | |
1831 | * is invalid nothing can be done so disable priority mappings. If is | |
1832 | * expected that drivers will fix this mapping if they can before | |
1833 | * calling netif_set_real_num_tx_queues. | |
1834 | */ | |
bb134d22 | 1835 | static void netif_setup_tc(struct net_device *dev, unsigned int txq) |
4f57c087 JF |
1836 | { |
1837 | int i; | |
1838 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
1839 | ||
1840 | /* If TC0 is invalidated disable TC mapping */ | |
1841 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce | 1842 | pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); |
4f57c087 JF |
1843 | dev->num_tc = 0; |
1844 | return; | |
1845 | } | |
1846 | ||
1847 | /* Invalidated prio to tc mappings set to TC0 */ | |
1848 | for (i = 1; i < TC_BITMASK + 1; i++) { | |
1849 | int q = netdev_get_prio_tc_map(dev, i); | |
1850 | ||
1851 | tc = &dev->tc_to_txq[q]; | |
1852 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce JP |
1853 | pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", |
1854 | i, q); | |
4f57c087 JF |
1855 | netdev_set_prio_tc_map(dev, i, 0); |
1856 | } | |
1857 | } | |
1858 | } | |
1859 | ||
537c00de AD |
1860 | #ifdef CONFIG_XPS |
1861 | static DEFINE_MUTEX(xps_map_mutex); | |
1862 | #define xmap_dereference(P) \ | |
1863 | rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) | |
1864 | ||
10cdc3f3 AD |
1865 | static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps, |
1866 | int cpu, u16 index) | |
537c00de | 1867 | { |
10cdc3f3 AD |
1868 | struct xps_map *map = NULL; |
1869 | int pos; | |
537c00de | 1870 | |
10cdc3f3 AD |
1871 | if (dev_maps) |
1872 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
537c00de | 1873 | |
10cdc3f3 AD |
1874 | for (pos = 0; map && pos < map->len; pos++) { |
1875 | if (map->queues[pos] == index) { | |
537c00de AD |
1876 | if (map->len > 1) { |
1877 | map->queues[pos] = map->queues[--map->len]; | |
1878 | } else { | |
10cdc3f3 | 1879 | RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL); |
537c00de AD |
1880 | kfree_rcu(map, rcu); |
1881 | map = NULL; | |
1882 | } | |
10cdc3f3 | 1883 | break; |
537c00de | 1884 | } |
537c00de AD |
1885 | } |
1886 | ||
10cdc3f3 AD |
1887 | return map; |
1888 | } | |
1889 | ||
1890 | void netif_reset_xps_queue(struct net_device *dev, u16 index) | |
1891 | { | |
1892 | struct xps_dev_maps *dev_maps; | |
1893 | int cpu; | |
1894 | bool active = false; | |
1895 | ||
1896 | mutex_lock(&xps_map_mutex); | |
1897 | dev_maps = xmap_dereference(dev->xps_maps); | |
1898 | ||
1899 | if (!dev_maps) | |
1900 | goto out_no_maps; | |
1901 | ||
1902 | for_each_possible_cpu(cpu) { | |
1903 | if (remove_xps_queue(dev_maps, cpu, index)) | |
1904 | active = true; | |
1905 | } | |
1906 | ||
1907 | if (!active) { | |
537c00de AD |
1908 | RCU_INIT_POINTER(dev->xps_maps, NULL); |
1909 | kfree_rcu(dev_maps, rcu); | |
1910 | } | |
1911 | ||
10cdc3f3 AD |
1912 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), |
1913 | NUMA_NO_NODE); | |
537c00de AD |
1914 | out_no_maps: |
1915 | mutex_unlock(&xps_map_mutex); | |
1916 | } | |
1917 | ||
01c5f864 AD |
1918 | static struct xps_map *expand_xps_map(struct xps_map *map, |
1919 | int cpu, u16 index) | |
1920 | { | |
1921 | struct xps_map *new_map; | |
1922 | int alloc_len = XPS_MIN_MAP_ALLOC; | |
1923 | int i, pos; | |
1924 | ||
1925 | for (pos = 0; map && pos < map->len; pos++) { | |
1926 | if (map->queues[pos] != index) | |
1927 | continue; | |
1928 | return map; | |
1929 | } | |
1930 | ||
1931 | /* Need to add queue to this CPU's existing map */ | |
1932 | if (map) { | |
1933 | if (pos < map->alloc_len) | |
1934 | return map; | |
1935 | ||
1936 | alloc_len = map->alloc_len * 2; | |
1937 | } | |
1938 | ||
1939 | /* Need to allocate new map to store queue on this CPU's map */ | |
1940 | new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, | |
1941 | cpu_to_node(cpu)); | |
1942 | if (!new_map) | |
1943 | return NULL; | |
1944 | ||
1945 | for (i = 0; i < pos; i++) | |
1946 | new_map->queues[i] = map->queues[i]; | |
1947 | new_map->alloc_len = alloc_len; | |
1948 | new_map->len = pos; | |
1949 | ||
1950 | return new_map; | |
1951 | } | |
1952 | ||
537c00de AD |
1953 | int netif_set_xps_queue(struct net_device *dev, struct cpumask *mask, u16 index) |
1954 | { | |
01c5f864 | 1955 | struct xps_dev_maps *dev_maps, *new_dev_maps = NULL; |
537c00de | 1956 | struct xps_map *map, *new_map; |
537c00de | 1957 | int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES); |
01c5f864 AD |
1958 | int cpu, numa_node_id = -2; |
1959 | bool active = false; | |
537c00de AD |
1960 | |
1961 | mutex_lock(&xps_map_mutex); | |
1962 | ||
1963 | dev_maps = xmap_dereference(dev->xps_maps); | |
1964 | ||
01c5f864 AD |
1965 | /* allocate memory for queue storage */ |
1966 | for_each_online_cpu(cpu) { | |
1967 | if (!cpumask_test_cpu(cpu, mask)) | |
1968 | continue; | |
1969 | ||
1970 | if (!new_dev_maps) | |
1971 | new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); | |
1972 | if (!new_dev_maps) | |
1973 | return -ENOMEM; | |
1974 | ||
1975 | map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) : | |
1976 | NULL; | |
1977 | ||
1978 | map = expand_xps_map(map, cpu, index); | |
1979 | if (!map) | |
1980 | goto error; | |
1981 | ||
1982 | RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map); | |
1983 | } | |
1984 | ||
1985 | if (!new_dev_maps) | |
1986 | goto out_no_new_maps; | |
1987 | ||
537c00de | 1988 | for_each_possible_cpu(cpu) { |
01c5f864 AD |
1989 | if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) { |
1990 | /* add queue to CPU maps */ | |
1991 | int pos = 0; | |
1992 | ||
1993 | map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
1994 | while ((pos < map->len) && (map->queues[pos] != index)) | |
1995 | pos++; | |
1996 | ||
1997 | if (pos == map->len) | |
1998 | map->queues[map->len++] = index; | |
537c00de | 1999 | #ifdef CONFIG_NUMA |
537c00de AD |
2000 | if (numa_node_id == -2) |
2001 | numa_node_id = cpu_to_node(cpu); | |
2002 | else if (numa_node_id != cpu_to_node(cpu)) | |
2003 | numa_node_id = -1; | |
537c00de | 2004 | #endif |
01c5f864 AD |
2005 | } else if (dev_maps) { |
2006 | /* fill in the new device map from the old device map */ | |
2007 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
2008 | RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map); | |
537c00de | 2009 | } |
01c5f864 | 2010 | |
537c00de AD |
2011 | } |
2012 | ||
01c5f864 AD |
2013 | rcu_assign_pointer(dev->xps_maps, new_dev_maps); |
2014 | ||
537c00de | 2015 | /* Cleanup old maps */ |
01c5f864 AD |
2016 | if (dev_maps) { |
2017 | for_each_possible_cpu(cpu) { | |
2018 | new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
2019 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
2020 | if (map && map != new_map) | |
2021 | kfree_rcu(map, rcu); | |
2022 | } | |
537c00de | 2023 | |
01c5f864 | 2024 | kfree_rcu(dev_maps, rcu); |
537c00de AD |
2025 | } |
2026 | ||
01c5f864 AD |
2027 | dev_maps = new_dev_maps; |
2028 | active = true; | |
537c00de | 2029 | |
01c5f864 AD |
2030 | out_no_new_maps: |
2031 | /* update Tx queue numa node */ | |
537c00de AD |
2032 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), |
2033 | (numa_node_id >= 0) ? numa_node_id : | |
2034 | NUMA_NO_NODE); | |
2035 | ||
01c5f864 AD |
2036 | if (!dev_maps) |
2037 | goto out_no_maps; | |
2038 | ||
2039 | /* removes queue from unused CPUs */ | |
2040 | for_each_possible_cpu(cpu) { | |
2041 | if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) | |
2042 | continue; | |
2043 | ||
2044 | if (remove_xps_queue(dev_maps, cpu, index)) | |
2045 | active = true; | |
2046 | } | |
2047 | ||
2048 | /* free map if not active */ | |
2049 | if (!active) { | |
2050 | RCU_INIT_POINTER(dev->xps_maps, NULL); | |
2051 | kfree_rcu(dev_maps, rcu); | |
2052 | } | |
2053 | ||
2054 | out_no_maps: | |
537c00de AD |
2055 | mutex_unlock(&xps_map_mutex); |
2056 | ||
2057 | return 0; | |
2058 | error: | |
01c5f864 AD |
2059 | /* remove any maps that we added */ |
2060 | for_each_possible_cpu(cpu) { | |
2061 | new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
2062 | map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) : | |
2063 | NULL; | |
2064 | if (new_map && new_map != map) | |
2065 | kfree(new_map); | |
2066 | } | |
2067 | ||
537c00de AD |
2068 | mutex_unlock(&xps_map_mutex); |
2069 | ||
537c00de AD |
2070 | kfree(new_dev_maps); |
2071 | return -ENOMEM; | |
2072 | } | |
2073 | EXPORT_SYMBOL(netif_set_xps_queue); | |
2074 | ||
2075 | #endif | |
f0796d5c JF |
2076 | /* |
2077 | * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues | |
2078 | * greater then real_num_tx_queues stale skbs on the qdisc must be flushed. | |
2079 | */ | |
e6484930 | 2080 | int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) |
f0796d5c | 2081 | { |
1d24eb48 TH |
2082 | int rc; |
2083 | ||
e6484930 TH |
2084 | if (txq < 1 || txq > dev->num_tx_queues) |
2085 | return -EINVAL; | |
f0796d5c | 2086 | |
5c56580b BH |
2087 | if (dev->reg_state == NETREG_REGISTERED || |
2088 | dev->reg_state == NETREG_UNREGISTERING) { | |
e6484930 TH |
2089 | ASSERT_RTNL(); |
2090 | ||
1d24eb48 TH |
2091 | rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, |
2092 | txq); | |
bf264145 TH |
2093 | if (rc) |
2094 | return rc; | |
2095 | ||
4f57c087 JF |
2096 | if (dev->num_tc) |
2097 | netif_setup_tc(dev, txq); | |
2098 | ||
e6484930 TH |
2099 | if (txq < dev->real_num_tx_queues) |
2100 | qdisc_reset_all_tx_gt(dev, txq); | |
f0796d5c | 2101 | } |
e6484930 TH |
2102 | |
2103 | dev->real_num_tx_queues = txq; | |
2104 | return 0; | |
f0796d5c JF |
2105 | } |
2106 | EXPORT_SYMBOL(netif_set_real_num_tx_queues); | |
56079431 | 2107 | |
62fe0b40 BH |
2108 | #ifdef CONFIG_RPS |
2109 | /** | |
2110 | * netif_set_real_num_rx_queues - set actual number of RX queues used | |
2111 | * @dev: Network device | |
2112 | * @rxq: Actual number of RX queues | |
2113 | * | |
2114 | * This must be called either with the rtnl_lock held or before | |
2115 | * registration of the net device. Returns 0 on success, or a | |
4e7f7951 BH |
2116 | * negative error code. If called before registration, it always |
2117 | * succeeds. | |
62fe0b40 BH |
2118 | */ |
2119 | int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) | |
2120 | { | |
2121 | int rc; | |
2122 | ||
bd25fa7b TH |
2123 | if (rxq < 1 || rxq > dev->num_rx_queues) |
2124 | return -EINVAL; | |
2125 | ||
62fe0b40 BH |
2126 | if (dev->reg_state == NETREG_REGISTERED) { |
2127 | ASSERT_RTNL(); | |
2128 | ||
62fe0b40 BH |
2129 | rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, |
2130 | rxq); | |
2131 | if (rc) | |
2132 | return rc; | |
62fe0b40 BH |
2133 | } |
2134 | ||
2135 | dev->real_num_rx_queues = rxq; | |
2136 | return 0; | |
2137 | } | |
2138 | EXPORT_SYMBOL(netif_set_real_num_rx_queues); | |
2139 | #endif | |
2140 | ||
2c53040f BH |
2141 | /** |
2142 | * netif_get_num_default_rss_queues - default number of RSS queues | |
16917b87 YM |
2143 | * |
2144 | * This routine should set an upper limit on the number of RSS queues | |
2145 | * used by default by multiqueue devices. | |
2146 | */ | |
a55b138b | 2147 | int netif_get_num_default_rss_queues(void) |
16917b87 YM |
2148 | { |
2149 | return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); | |
2150 | } | |
2151 | EXPORT_SYMBOL(netif_get_num_default_rss_queues); | |
2152 | ||
def82a1d | 2153 | static inline void __netif_reschedule(struct Qdisc *q) |
56079431 | 2154 | { |
def82a1d JP |
2155 | struct softnet_data *sd; |
2156 | unsigned long flags; | |
56079431 | 2157 | |
def82a1d JP |
2158 | local_irq_save(flags); |
2159 | sd = &__get_cpu_var(softnet_data); | |
a9cbd588 CG |
2160 | q->next_sched = NULL; |
2161 | *sd->output_queue_tailp = q; | |
2162 | sd->output_queue_tailp = &q->next_sched; | |
def82a1d JP |
2163 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
2164 | local_irq_restore(flags); | |
2165 | } | |
2166 | ||
2167 | void __netif_schedule(struct Qdisc *q) | |
2168 | { | |
2169 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
2170 | __netif_reschedule(q); | |
56079431 DV |
2171 | } |
2172 | EXPORT_SYMBOL(__netif_schedule); | |
2173 | ||
bea3348e | 2174 | void dev_kfree_skb_irq(struct sk_buff *skb) |
56079431 | 2175 | { |
3578b0c8 | 2176 | if (atomic_dec_and_test(&skb->users)) { |
bea3348e SH |
2177 | struct softnet_data *sd; |
2178 | unsigned long flags; | |
56079431 | 2179 | |
bea3348e SH |
2180 | local_irq_save(flags); |
2181 | sd = &__get_cpu_var(softnet_data); | |
2182 | skb->next = sd->completion_queue; | |
2183 | sd->completion_queue = skb; | |
2184 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
2185 | local_irq_restore(flags); | |
2186 | } | |
56079431 | 2187 | } |
bea3348e | 2188 | EXPORT_SYMBOL(dev_kfree_skb_irq); |
56079431 DV |
2189 | |
2190 | void dev_kfree_skb_any(struct sk_buff *skb) | |
2191 | { | |
2192 | if (in_irq() || irqs_disabled()) | |
2193 | dev_kfree_skb_irq(skb); | |
2194 | else | |
2195 | dev_kfree_skb(skb); | |
2196 | } | |
2197 | EXPORT_SYMBOL(dev_kfree_skb_any); | |
2198 | ||
2199 | ||
bea3348e SH |
2200 | /** |
2201 | * netif_device_detach - mark device as removed | |
2202 | * @dev: network device | |
2203 | * | |
2204 | * Mark device as removed from system and therefore no longer available. | |
2205 | */ | |
56079431 DV |
2206 | void netif_device_detach(struct net_device *dev) |
2207 | { | |
2208 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2209 | netif_running(dev)) { | |
d543103a | 2210 | netif_tx_stop_all_queues(dev); |
56079431 DV |
2211 | } |
2212 | } | |
2213 | EXPORT_SYMBOL(netif_device_detach); | |
2214 | ||
bea3348e SH |
2215 | /** |
2216 | * netif_device_attach - mark device as attached | |
2217 | * @dev: network device | |
2218 | * | |
2219 | * Mark device as attached from system and restart if needed. | |
2220 | */ | |
56079431 DV |
2221 | void netif_device_attach(struct net_device *dev) |
2222 | { | |
2223 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2224 | netif_running(dev)) { | |
d543103a | 2225 | netif_tx_wake_all_queues(dev); |
4ec93edb | 2226 | __netdev_watchdog_up(dev); |
56079431 DV |
2227 | } |
2228 | } | |
2229 | EXPORT_SYMBOL(netif_device_attach); | |
2230 | ||
36c92474 BH |
2231 | static void skb_warn_bad_offload(const struct sk_buff *skb) |
2232 | { | |
65e9d2fa | 2233 | static const netdev_features_t null_features = 0; |
36c92474 BH |
2234 | struct net_device *dev = skb->dev; |
2235 | const char *driver = ""; | |
2236 | ||
2237 | if (dev && dev->dev.parent) | |
2238 | driver = dev_driver_string(dev->dev.parent); | |
2239 | ||
2240 | WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d " | |
2241 | "gso_type=%d ip_summed=%d\n", | |
65e9d2fa MM |
2242 | driver, dev ? &dev->features : &null_features, |
2243 | skb->sk ? &skb->sk->sk_route_caps : &null_features, | |
36c92474 BH |
2244 | skb->len, skb->data_len, skb_shinfo(skb)->gso_size, |
2245 | skb_shinfo(skb)->gso_type, skb->ip_summed); | |
2246 | } | |
2247 | ||
1da177e4 LT |
2248 | /* |
2249 | * Invalidate hardware checksum when packet is to be mangled, and | |
2250 | * complete checksum manually on outgoing path. | |
2251 | */ | |
84fa7933 | 2252 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 2253 | { |
d3bc23e7 | 2254 | __wsum csum; |
663ead3b | 2255 | int ret = 0, offset; |
1da177e4 | 2256 | |
84fa7933 | 2257 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
2258 | goto out_set_summed; |
2259 | ||
2260 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
36c92474 BH |
2261 | skb_warn_bad_offload(skb); |
2262 | return -EINVAL; | |
1da177e4 LT |
2263 | } |
2264 | ||
55508d60 | 2265 | offset = skb_checksum_start_offset(skb); |
a030847e HX |
2266 | BUG_ON(offset >= skb_headlen(skb)); |
2267 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
2268 | ||
2269 | offset += skb->csum_offset; | |
2270 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
2271 | ||
2272 | if (skb_cloned(skb) && | |
2273 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1da177e4 LT |
2274 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
2275 | if (ret) | |
2276 | goto out; | |
2277 | } | |
2278 | ||
a030847e | 2279 | *(__sum16 *)(skb->data + offset) = csum_fold(csum); |
a430a43d | 2280 | out_set_summed: |
1da177e4 | 2281 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 2282 | out: |
1da177e4 LT |
2283 | return ret; |
2284 | } | |
d1b19dff | 2285 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 2286 | |
f6a78bfc HX |
2287 | /** |
2288 | * skb_gso_segment - Perform segmentation on skb. | |
2289 | * @skb: buffer to segment | |
576a30eb | 2290 | * @features: features for the output path (see dev->features) |
f6a78bfc HX |
2291 | * |
2292 | * This function segments the given skb and returns a list of segments. | |
576a30eb HX |
2293 | * |
2294 | * It may return NULL if the skb requires no segmentation. This is | |
2295 | * only possible when GSO is used for verifying header integrity. | |
f6a78bfc | 2296 | */ |
c8f44aff MM |
2297 | struct sk_buff *skb_gso_segment(struct sk_buff *skb, |
2298 | netdev_features_t features) | |
f6a78bfc HX |
2299 | { |
2300 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
22061d80 | 2301 | struct packet_offload *ptype; |
252e3346 | 2302 | __be16 type = skb->protocol; |
c8d5bcd1 | 2303 | int vlan_depth = ETH_HLEN; |
a430a43d | 2304 | int err; |
f6a78bfc | 2305 | |
c8d5bcd1 JG |
2306 | while (type == htons(ETH_P_8021Q)) { |
2307 | struct vlan_hdr *vh; | |
7b9c6090 | 2308 | |
c8d5bcd1 | 2309 | if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN))) |
7b9c6090 JG |
2310 | return ERR_PTR(-EINVAL); |
2311 | ||
c8d5bcd1 JG |
2312 | vh = (struct vlan_hdr *)(skb->data + vlan_depth); |
2313 | type = vh->h_vlan_encapsulated_proto; | |
2314 | vlan_depth += VLAN_HLEN; | |
7b9c6090 JG |
2315 | } |
2316 | ||
459a98ed | 2317 | skb_reset_mac_header(skb); |
b0e380b1 | 2318 | skb->mac_len = skb->network_header - skb->mac_header; |
f6a78bfc HX |
2319 | __skb_pull(skb, skb->mac_len); |
2320 | ||
67fd1a73 | 2321 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { |
36c92474 | 2322 | skb_warn_bad_offload(skb); |
67fd1a73 | 2323 | |
a430a43d HX |
2324 | if (skb_header_cloned(skb) && |
2325 | (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))) | |
2326 | return ERR_PTR(err); | |
2327 | } | |
2328 | ||
f6a78bfc | 2329 | rcu_read_lock(); |
22061d80 | 2330 | list_for_each_entry_rcu(ptype, &offload_base, list) { |
f191a1d1 | 2331 | if (ptype->type == type && ptype->callbacks.gso_segment) { |
84fa7933 | 2332 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { |
f191a1d1 | 2333 | err = ptype->callbacks.gso_send_check(skb); |
a430a43d HX |
2334 | segs = ERR_PTR(err); |
2335 | if (err || skb_gso_ok(skb, features)) | |
2336 | break; | |
d56f90a7 ACM |
2337 | __skb_push(skb, (skb->data - |
2338 | skb_network_header(skb))); | |
a430a43d | 2339 | } |
f191a1d1 | 2340 | segs = ptype->callbacks.gso_segment(skb, features); |
f6a78bfc HX |
2341 | break; |
2342 | } | |
2343 | } | |
2344 | rcu_read_unlock(); | |
2345 | ||
98e399f8 | 2346 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 2347 | |
f6a78bfc HX |
2348 | return segs; |
2349 | } | |
f6a78bfc HX |
2350 | EXPORT_SYMBOL(skb_gso_segment); |
2351 | ||
fb286bb2 HX |
2352 | /* Take action when hardware reception checksum errors are detected. */ |
2353 | #ifdef CONFIG_BUG | |
2354 | void netdev_rx_csum_fault(struct net_device *dev) | |
2355 | { | |
2356 | if (net_ratelimit()) { | |
7b6cd1ce | 2357 | pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); |
fb286bb2 HX |
2358 | dump_stack(); |
2359 | } | |
2360 | } | |
2361 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
2362 | #endif | |
2363 | ||
1da177e4 LT |
2364 | /* Actually, we should eliminate this check as soon as we know, that: |
2365 | * 1. IOMMU is present and allows to map all the memory. | |
2366 | * 2. No high memory really exists on this machine. | |
2367 | */ | |
2368 | ||
9092c658 | 2369 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
1da177e4 | 2370 | { |
3d3a8533 | 2371 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 2372 | int i; |
5acbbd42 | 2373 | if (!(dev->features & NETIF_F_HIGHDMA)) { |
ea2ab693 IC |
2374 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
2375 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
2376 | if (PageHighMem(skb_frag_page(frag))) | |
5acbbd42 | 2377 | return 1; |
ea2ab693 | 2378 | } |
5acbbd42 | 2379 | } |
1da177e4 | 2380 | |
5acbbd42 FT |
2381 | if (PCI_DMA_BUS_IS_PHYS) { |
2382 | struct device *pdev = dev->dev.parent; | |
1da177e4 | 2383 | |
9092c658 ED |
2384 | if (!pdev) |
2385 | return 0; | |
5acbbd42 | 2386 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
ea2ab693 IC |
2387 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
2388 | dma_addr_t addr = page_to_phys(skb_frag_page(frag)); | |
5acbbd42 FT |
2389 | if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask) |
2390 | return 1; | |
2391 | } | |
2392 | } | |
3d3a8533 | 2393 | #endif |
1da177e4 LT |
2394 | return 0; |
2395 | } | |
1da177e4 | 2396 | |
f6a78bfc HX |
2397 | struct dev_gso_cb { |
2398 | void (*destructor)(struct sk_buff *skb); | |
2399 | }; | |
2400 | ||
2401 | #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb) | |
2402 | ||
2403 | static void dev_gso_skb_destructor(struct sk_buff *skb) | |
2404 | { | |
2405 | struct dev_gso_cb *cb; | |
2406 | ||
2407 | do { | |
2408 | struct sk_buff *nskb = skb->next; | |
2409 | ||
2410 | skb->next = nskb->next; | |
2411 | nskb->next = NULL; | |
2412 | kfree_skb(nskb); | |
2413 | } while (skb->next); | |
2414 | ||
2415 | cb = DEV_GSO_CB(skb); | |
2416 | if (cb->destructor) | |
2417 | cb->destructor(skb); | |
2418 | } | |
2419 | ||
2420 | /** | |
2421 | * dev_gso_segment - Perform emulated hardware segmentation on skb. | |
2422 | * @skb: buffer to segment | |
91ecb63c | 2423 | * @features: device features as applicable to this skb |
f6a78bfc HX |
2424 | * |
2425 | * This function segments the given skb and stores the list of segments | |
2426 | * in skb->next. | |
2427 | */ | |
c8f44aff | 2428 | static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features) |
f6a78bfc | 2429 | { |
f6a78bfc | 2430 | struct sk_buff *segs; |
576a30eb HX |
2431 | |
2432 | segs = skb_gso_segment(skb, features); | |
2433 | ||
2434 | /* Verifying header integrity only. */ | |
2435 | if (!segs) | |
2436 | return 0; | |
f6a78bfc | 2437 | |
801678c5 | 2438 | if (IS_ERR(segs)) |
f6a78bfc HX |
2439 | return PTR_ERR(segs); |
2440 | ||
2441 | skb->next = segs; | |
2442 | DEV_GSO_CB(skb)->destructor = skb->destructor; | |
2443 | skb->destructor = dev_gso_skb_destructor; | |
2444 | ||
2445 | return 0; | |
2446 | } | |
2447 | ||
c8f44aff | 2448 | static bool can_checksum_protocol(netdev_features_t features, __be16 protocol) |
03634668 JG |
2449 | { |
2450 | return ((features & NETIF_F_GEN_CSUM) || | |
2451 | ((features & NETIF_F_V4_CSUM) && | |
2452 | protocol == htons(ETH_P_IP)) || | |
2453 | ((features & NETIF_F_V6_CSUM) && | |
2454 | protocol == htons(ETH_P_IPV6)) || | |
2455 | ((features & NETIF_F_FCOE_CRC) && | |
2456 | protocol == htons(ETH_P_FCOE))); | |
2457 | } | |
2458 | ||
c8f44aff MM |
2459 | static netdev_features_t harmonize_features(struct sk_buff *skb, |
2460 | __be16 protocol, netdev_features_t features) | |
f01a5236 | 2461 | { |
c0d680e5 EC |
2462 | if (skb->ip_summed != CHECKSUM_NONE && |
2463 | !can_checksum_protocol(features, protocol)) { | |
f01a5236 JG |
2464 | features &= ~NETIF_F_ALL_CSUM; |
2465 | features &= ~NETIF_F_SG; | |
2466 | } else if (illegal_highdma(skb->dev, skb)) { | |
2467 | features &= ~NETIF_F_SG; | |
2468 | } | |
2469 | ||
2470 | return features; | |
2471 | } | |
2472 | ||
c8f44aff | 2473 | netdev_features_t netif_skb_features(struct sk_buff *skb) |
58e998c6 JG |
2474 | { |
2475 | __be16 protocol = skb->protocol; | |
c8f44aff | 2476 | netdev_features_t features = skb->dev->features; |
58e998c6 | 2477 | |
30b678d8 BH |
2478 | if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs) |
2479 | features &= ~NETIF_F_GSO_MASK; | |
2480 | ||
58e998c6 JG |
2481 | if (protocol == htons(ETH_P_8021Q)) { |
2482 | struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data; | |
2483 | protocol = veh->h_vlan_encapsulated_proto; | |
f01a5236 JG |
2484 | } else if (!vlan_tx_tag_present(skb)) { |
2485 | return harmonize_features(skb, protocol, features); | |
2486 | } | |
58e998c6 | 2487 | |
6ee400aa | 2488 | features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX); |
f01a5236 JG |
2489 | |
2490 | if (protocol != htons(ETH_P_8021Q)) { | |
2491 | return harmonize_features(skb, protocol, features); | |
2492 | } else { | |
2493 | features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | | |
6ee400aa | 2494 | NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX; |
f01a5236 JG |
2495 | return harmonize_features(skb, protocol, features); |
2496 | } | |
58e998c6 | 2497 | } |
f01a5236 | 2498 | EXPORT_SYMBOL(netif_skb_features); |
58e998c6 | 2499 | |
6afff0ca JF |
2500 | /* |
2501 | * Returns true if either: | |
2502 | * 1. skb has frag_list and the device doesn't support FRAGLIST, or | |
d1a53dfd | 2503 | * 2. skb is fragmented and the device does not support SG. |
6afff0ca JF |
2504 | */ |
2505 | static inline int skb_needs_linearize(struct sk_buff *skb, | |
02932ce9 | 2506 | int features) |
6afff0ca | 2507 | { |
02932ce9 JG |
2508 | return skb_is_nonlinear(skb) && |
2509 | ((skb_has_frag_list(skb) && | |
2510 | !(features & NETIF_F_FRAGLIST)) || | |
e1e78db6 | 2511 | (skb_shinfo(skb)->nr_frags && |
02932ce9 | 2512 | !(features & NETIF_F_SG))); |
6afff0ca JF |
2513 | } |
2514 | ||
fd2ea0a7 DM |
2515 | int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev, |
2516 | struct netdev_queue *txq) | |
f6a78bfc | 2517 | { |
00829823 | 2518 | const struct net_device_ops *ops = dev->netdev_ops; |
572a9d7b | 2519 | int rc = NETDEV_TX_OK; |
ec764bf0 | 2520 | unsigned int skb_len; |
00829823 | 2521 | |
f6a78bfc | 2522 | if (likely(!skb->next)) { |
c8f44aff | 2523 | netdev_features_t features; |
fc741216 | 2524 | |
93f154b5 | 2525 | /* |
25985edc | 2526 | * If device doesn't need skb->dst, release it right now while |
93f154b5 ED |
2527 | * its hot in this cpu cache |
2528 | */ | |
adf30907 ED |
2529 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) |
2530 | skb_dst_drop(skb); | |
2531 | ||
fc741216 JG |
2532 | features = netif_skb_features(skb); |
2533 | ||
7b9c6090 | 2534 | if (vlan_tx_tag_present(skb) && |
fc741216 | 2535 | !(features & NETIF_F_HW_VLAN_TX)) { |
7b9c6090 JG |
2536 | skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb)); |
2537 | if (unlikely(!skb)) | |
2538 | goto out; | |
2539 | ||
2540 | skb->vlan_tci = 0; | |
2541 | } | |
2542 | ||
fc70fb64 AD |
2543 | /* If encapsulation offload request, verify we are testing |
2544 | * hardware encapsulation features instead of standard | |
2545 | * features for the netdev | |
2546 | */ | |
2547 | if (skb->encapsulation) | |
2548 | features &= dev->hw_enc_features; | |
2549 | ||
fc741216 | 2550 | if (netif_needs_gso(skb, features)) { |
91ecb63c | 2551 | if (unlikely(dev_gso_segment(skb, features))) |
9ccb8975 DM |
2552 | goto out_kfree_skb; |
2553 | if (skb->next) | |
2554 | goto gso; | |
6afff0ca | 2555 | } else { |
02932ce9 | 2556 | if (skb_needs_linearize(skb, features) && |
6afff0ca JF |
2557 | __skb_linearize(skb)) |
2558 | goto out_kfree_skb; | |
2559 | ||
2560 | /* If packet is not checksummed and device does not | |
2561 | * support checksumming for this protocol, complete | |
2562 | * checksumming here. | |
2563 | */ | |
2564 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
fc70fb64 AD |
2565 | if (skb->encapsulation) |
2566 | skb_set_inner_transport_header(skb, | |
2567 | skb_checksum_start_offset(skb)); | |
2568 | else | |
2569 | skb_set_transport_header(skb, | |
2570 | skb_checksum_start_offset(skb)); | |
03634668 | 2571 | if (!(features & NETIF_F_ALL_CSUM) && |
6afff0ca JF |
2572 | skb_checksum_help(skb)) |
2573 | goto out_kfree_skb; | |
2574 | } | |
9ccb8975 DM |
2575 | } |
2576 | ||
b40863c6 ED |
2577 | if (!list_empty(&ptype_all)) |
2578 | dev_queue_xmit_nit(skb, dev); | |
2579 | ||
ec764bf0 | 2580 | skb_len = skb->len; |
ac45f602 | 2581 | rc = ops->ndo_start_xmit(skb, dev); |
ec764bf0 | 2582 | trace_net_dev_xmit(skb, rc, dev, skb_len); |
ec634fe3 | 2583 | if (rc == NETDEV_TX_OK) |
08baf561 | 2584 | txq_trans_update(txq); |
ac45f602 | 2585 | return rc; |
f6a78bfc HX |
2586 | } |
2587 | ||
576a30eb | 2588 | gso: |
f6a78bfc HX |
2589 | do { |
2590 | struct sk_buff *nskb = skb->next; | |
f6a78bfc HX |
2591 | |
2592 | skb->next = nskb->next; | |
2593 | nskb->next = NULL; | |
068a2de5 KK |
2594 | |
2595 | /* | |
25985edc | 2596 | * If device doesn't need nskb->dst, release it right now while |
068a2de5 KK |
2597 | * its hot in this cpu cache |
2598 | */ | |
2599 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
2600 | skb_dst_drop(nskb); | |
2601 | ||
b40863c6 ED |
2602 | if (!list_empty(&ptype_all)) |
2603 | dev_queue_xmit_nit(nskb, dev); | |
2604 | ||
ec764bf0 | 2605 | skb_len = nskb->len; |
00829823 | 2606 | rc = ops->ndo_start_xmit(nskb, dev); |
ec764bf0 | 2607 | trace_net_dev_xmit(nskb, rc, dev, skb_len); |
ec634fe3 | 2608 | if (unlikely(rc != NETDEV_TX_OK)) { |
572a9d7b PM |
2609 | if (rc & ~NETDEV_TX_MASK) |
2610 | goto out_kfree_gso_skb; | |
f54d9e8d | 2611 | nskb->next = skb->next; |
f6a78bfc HX |
2612 | skb->next = nskb; |
2613 | return rc; | |
2614 | } | |
08baf561 | 2615 | txq_trans_update(txq); |
73466498 | 2616 | if (unlikely(netif_xmit_stopped(txq) && skb->next)) |
f54d9e8d | 2617 | return NETDEV_TX_BUSY; |
f6a78bfc | 2618 | } while (skb->next); |
4ec93edb | 2619 | |
572a9d7b PM |
2620 | out_kfree_gso_skb: |
2621 | if (likely(skb->next == NULL)) | |
2622 | skb->destructor = DEV_GSO_CB(skb)->destructor; | |
f6a78bfc HX |
2623 | out_kfree_skb: |
2624 | kfree_skb(skb); | |
7b9c6090 | 2625 | out: |
572a9d7b | 2626 | return rc; |
f6a78bfc HX |
2627 | } |
2628 | ||
0a9627f2 | 2629 | static u32 hashrnd __read_mostly; |
b6b2fed1 | 2630 | |
a3d22a68 VZ |
2631 | /* |
2632 | * Returns a Tx hash based on the given packet descriptor a Tx queues' number | |
2633 | * to be used as a distribution range. | |
2634 | */ | |
2635 | u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb, | |
2636 | unsigned int num_tx_queues) | |
8f0f2223 | 2637 | { |
7019298a | 2638 | u32 hash; |
4f57c087 JF |
2639 | u16 qoffset = 0; |
2640 | u16 qcount = num_tx_queues; | |
b6b2fed1 | 2641 | |
513de11b DM |
2642 | if (skb_rx_queue_recorded(skb)) { |
2643 | hash = skb_get_rx_queue(skb); | |
a3d22a68 VZ |
2644 | while (unlikely(hash >= num_tx_queues)) |
2645 | hash -= num_tx_queues; | |
513de11b DM |
2646 | return hash; |
2647 | } | |
ec581f6a | 2648 | |
4f57c087 JF |
2649 | if (dev->num_tc) { |
2650 | u8 tc = netdev_get_prio_tc_map(dev, skb->priority); | |
2651 | qoffset = dev->tc_to_txq[tc].offset; | |
2652 | qcount = dev->tc_to_txq[tc].count; | |
2653 | } | |
2654 | ||
ec581f6a | 2655 | if (skb->sk && skb->sk->sk_hash) |
7019298a | 2656 | hash = skb->sk->sk_hash; |
ec581f6a | 2657 | else |
62b1a8ab | 2658 | hash = (__force u16) skb->protocol; |
0a9627f2 | 2659 | hash = jhash_1word(hash, hashrnd); |
b6b2fed1 | 2660 | |
4f57c087 | 2661 | return (u16) (((u64) hash * qcount) >> 32) + qoffset; |
8f0f2223 | 2662 | } |
a3d22a68 | 2663 | EXPORT_SYMBOL(__skb_tx_hash); |
8f0f2223 | 2664 | |
ed04642f ED |
2665 | static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index) |
2666 | { | |
2667 | if (unlikely(queue_index >= dev->real_num_tx_queues)) { | |
e87cc472 JP |
2668 | net_warn_ratelimited("%s selects TX queue %d, but real number of TX queues is %d\n", |
2669 | dev->name, queue_index, | |
2670 | dev->real_num_tx_queues); | |
ed04642f ED |
2671 | return 0; |
2672 | } | |
2673 | return queue_index; | |
2674 | } | |
2675 | ||
1d24eb48 TH |
2676 | static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb) |
2677 | { | |
bf264145 | 2678 | #ifdef CONFIG_XPS |
1d24eb48 TH |
2679 | struct xps_dev_maps *dev_maps; |
2680 | struct xps_map *map; | |
2681 | int queue_index = -1; | |
2682 | ||
2683 | rcu_read_lock(); | |
2684 | dev_maps = rcu_dereference(dev->xps_maps); | |
2685 | if (dev_maps) { | |
2686 | map = rcu_dereference( | |
2687 | dev_maps->cpu_map[raw_smp_processor_id()]); | |
2688 | if (map) { | |
2689 | if (map->len == 1) | |
2690 | queue_index = map->queues[0]; | |
2691 | else { | |
2692 | u32 hash; | |
2693 | if (skb->sk && skb->sk->sk_hash) | |
2694 | hash = skb->sk->sk_hash; | |
2695 | else | |
2696 | hash = (__force u16) skb->protocol ^ | |
2697 | skb->rxhash; | |
2698 | hash = jhash_1word(hash, hashrnd); | |
2699 | queue_index = map->queues[ | |
2700 | ((u64)hash * map->len) >> 32]; | |
2701 | } | |
2702 | if (unlikely(queue_index >= dev->real_num_tx_queues)) | |
2703 | queue_index = -1; | |
2704 | } | |
2705 | } | |
2706 | rcu_read_unlock(); | |
2707 | ||
2708 | return queue_index; | |
2709 | #else | |
2710 | return -1; | |
2711 | #endif | |
2712 | } | |
2713 | ||
416186fb | 2714 | u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb) |
e8a0464c | 2715 | { |
416186fb AD |
2716 | struct sock *sk = skb->sk; |
2717 | int queue_index = sk_tx_queue_get(sk); | |
a4ee3ce3 | 2718 | |
416186fb AD |
2719 | if (queue_index < 0 || skb->ooo_okay || |
2720 | queue_index >= dev->real_num_tx_queues) { | |
2721 | int new_index = get_xps_queue(dev, skb); | |
2722 | if (new_index < 0) | |
2723 | new_index = skb_tx_hash(dev, skb); | |
fd2ea0a7 | 2724 | |
416186fb AD |
2725 | if (queue_index != new_index && sk) { |
2726 | struct dst_entry *dst = | |
3853b584 | 2727 | rcu_dereference_check(sk->sk_dst_cache, 1); |
8728c544 | 2728 | |
416186fb AD |
2729 | if (dst && skb_dst(skb) == dst) |
2730 | sk_tx_queue_set(sk, queue_index); | |
2731 | ||
a4ee3ce3 | 2732 | } |
416186fb AD |
2733 | |
2734 | queue_index = new_index; | |
2735 | } | |
2736 | ||
2737 | return queue_index; | |
2738 | } | |
2739 | ||
2740 | struct netdev_queue *netdev_pick_tx(struct net_device *dev, | |
2741 | struct sk_buff *skb) | |
2742 | { | |
2743 | int queue_index = 0; | |
2744 | ||
2745 | if (dev->real_num_tx_queues != 1) { | |
2746 | const struct net_device_ops *ops = dev->netdev_ops; | |
2747 | if (ops->ndo_select_queue) | |
2748 | queue_index = ops->ndo_select_queue(dev, skb); | |
2749 | else | |
2750 | queue_index = __netdev_pick_tx(dev, skb); | |
2751 | queue_index = dev_cap_txqueue(dev, queue_index); | |
a4ee3ce3 | 2752 | } |
eae792b7 | 2753 | |
fd2ea0a7 DM |
2754 | skb_set_queue_mapping(skb, queue_index); |
2755 | return netdev_get_tx_queue(dev, queue_index); | |
e8a0464c DM |
2756 | } |
2757 | ||
1def9238 ED |
2758 | static void qdisc_pkt_len_init(struct sk_buff *skb) |
2759 | { | |
2760 | const struct skb_shared_info *shinfo = skb_shinfo(skb); | |
2761 | ||
2762 | qdisc_skb_cb(skb)->pkt_len = skb->len; | |
2763 | ||
2764 | /* To get more precise estimation of bytes sent on wire, | |
2765 | * we add to pkt_len the headers size of all segments | |
2766 | */ | |
2767 | if (shinfo->gso_size) { | |
2768 | unsigned int hdr_len = skb_transport_offset(skb); | |
2769 | ||
2770 | if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) | |
2771 | hdr_len += tcp_hdrlen(skb); | |
2772 | else | |
2773 | hdr_len += sizeof(struct udphdr); | |
2774 | qdisc_skb_cb(skb)->pkt_len += (shinfo->gso_segs - 1) * hdr_len; | |
2775 | } | |
2776 | } | |
2777 | ||
bbd8a0d3 KK |
2778 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
2779 | struct net_device *dev, | |
2780 | struct netdev_queue *txq) | |
2781 | { | |
2782 | spinlock_t *root_lock = qdisc_lock(q); | |
a2da570d | 2783 | bool contended; |
bbd8a0d3 KK |
2784 | int rc; |
2785 | ||
1def9238 | 2786 | qdisc_pkt_len_init(skb); |
a2da570d | 2787 | qdisc_calculate_pkt_len(skb, q); |
79640a4c ED |
2788 | /* |
2789 | * Heuristic to force contended enqueues to serialize on a | |
2790 | * separate lock before trying to get qdisc main lock. | |
2791 | * This permits __QDISC_STATE_RUNNING owner to get the lock more often | |
2792 | * and dequeue packets faster. | |
2793 | */ | |
a2da570d | 2794 | contended = qdisc_is_running(q); |
79640a4c ED |
2795 | if (unlikely(contended)) |
2796 | spin_lock(&q->busylock); | |
2797 | ||
bbd8a0d3 KK |
2798 | spin_lock(root_lock); |
2799 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
2800 | kfree_skb(skb); | |
2801 | rc = NET_XMIT_DROP; | |
2802 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
bc135b23 | 2803 | qdisc_run_begin(q)) { |
bbd8a0d3 KK |
2804 | /* |
2805 | * This is a work-conserving queue; there are no old skbs | |
2806 | * waiting to be sent out; and the qdisc is not running - | |
2807 | * xmit the skb directly. | |
2808 | */ | |
7fee226a ED |
2809 | if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE)) |
2810 | skb_dst_force(skb); | |
bfe0d029 | 2811 | |
bfe0d029 ED |
2812 | qdisc_bstats_update(q, skb); |
2813 | ||
79640a4c ED |
2814 | if (sch_direct_xmit(skb, q, dev, txq, root_lock)) { |
2815 | if (unlikely(contended)) { | |
2816 | spin_unlock(&q->busylock); | |
2817 | contended = false; | |
2818 | } | |
bbd8a0d3 | 2819 | __qdisc_run(q); |
79640a4c | 2820 | } else |
bc135b23 | 2821 | qdisc_run_end(q); |
bbd8a0d3 KK |
2822 | |
2823 | rc = NET_XMIT_SUCCESS; | |
2824 | } else { | |
7fee226a | 2825 | skb_dst_force(skb); |
a2da570d | 2826 | rc = q->enqueue(skb, q) & NET_XMIT_MASK; |
79640a4c ED |
2827 | if (qdisc_run_begin(q)) { |
2828 | if (unlikely(contended)) { | |
2829 | spin_unlock(&q->busylock); | |
2830 | contended = false; | |
2831 | } | |
2832 | __qdisc_run(q); | |
2833 | } | |
bbd8a0d3 KK |
2834 | } |
2835 | spin_unlock(root_lock); | |
79640a4c ED |
2836 | if (unlikely(contended)) |
2837 | spin_unlock(&q->busylock); | |
bbd8a0d3 KK |
2838 | return rc; |
2839 | } | |
2840 | ||
5bc1421e NH |
2841 | #if IS_ENABLED(CONFIG_NETPRIO_CGROUP) |
2842 | static void skb_update_prio(struct sk_buff *skb) | |
2843 | { | |
6977a79d | 2844 | struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap); |
5bc1421e | 2845 | |
91c68ce2 ED |
2846 | if (!skb->priority && skb->sk && map) { |
2847 | unsigned int prioidx = skb->sk->sk_cgrp_prioidx; | |
2848 | ||
2849 | if (prioidx < map->priomap_len) | |
2850 | skb->priority = map->priomap[prioidx]; | |
2851 | } | |
5bc1421e NH |
2852 | } |
2853 | #else | |
2854 | #define skb_update_prio(skb) | |
2855 | #endif | |
2856 | ||
745e20f1 | 2857 | static DEFINE_PER_CPU(int, xmit_recursion); |
11a766ce | 2858 | #define RECURSION_LIMIT 10 |
745e20f1 | 2859 | |
95603e22 MM |
2860 | /** |
2861 | * dev_loopback_xmit - loop back @skb | |
2862 | * @skb: buffer to transmit | |
2863 | */ | |
2864 | int dev_loopback_xmit(struct sk_buff *skb) | |
2865 | { | |
2866 | skb_reset_mac_header(skb); | |
2867 | __skb_pull(skb, skb_network_offset(skb)); | |
2868 | skb->pkt_type = PACKET_LOOPBACK; | |
2869 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
2870 | WARN_ON(!skb_dst(skb)); | |
2871 | skb_dst_force(skb); | |
2872 | netif_rx_ni(skb); | |
2873 | return 0; | |
2874 | } | |
2875 | EXPORT_SYMBOL(dev_loopback_xmit); | |
2876 | ||
d29f749e DJ |
2877 | /** |
2878 | * dev_queue_xmit - transmit a buffer | |
2879 | * @skb: buffer to transmit | |
2880 | * | |
2881 | * Queue a buffer for transmission to a network device. The caller must | |
2882 | * have set the device and priority and built the buffer before calling | |
2883 | * this function. The function can be called from an interrupt. | |
2884 | * | |
2885 | * A negative errno code is returned on a failure. A success does not | |
2886 | * guarantee the frame will be transmitted as it may be dropped due | |
2887 | * to congestion or traffic shaping. | |
2888 | * | |
2889 | * ----------------------------------------------------------------------------------- | |
2890 | * I notice this method can also return errors from the queue disciplines, | |
2891 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
2892 | * be positive. | |
2893 | * | |
2894 | * Regardless of the return value, the skb is consumed, so it is currently | |
2895 | * difficult to retry a send to this method. (You can bump the ref count | |
2896 | * before sending to hold a reference for retry if you are careful.) | |
2897 | * | |
2898 | * When calling this method, interrupts MUST be enabled. This is because | |
2899 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
2900 | * --BLG | |
2901 | */ | |
1da177e4 LT |
2902 | int dev_queue_xmit(struct sk_buff *skb) |
2903 | { | |
2904 | struct net_device *dev = skb->dev; | |
dc2b4847 | 2905 | struct netdev_queue *txq; |
1da177e4 LT |
2906 | struct Qdisc *q; |
2907 | int rc = -ENOMEM; | |
2908 | ||
4ec93edb YH |
2909 | /* Disable soft irqs for various locks below. Also |
2910 | * stops preemption for RCU. | |
1da177e4 | 2911 | */ |
4ec93edb | 2912 | rcu_read_lock_bh(); |
1da177e4 | 2913 | |
5bc1421e NH |
2914 | skb_update_prio(skb); |
2915 | ||
8c4c49df | 2916 | txq = netdev_pick_tx(dev, skb); |
a898def2 | 2917 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 2918 | |
1da177e4 | 2919 | #ifdef CONFIG_NET_CLS_ACT |
d1b19dff | 2920 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS); |
1da177e4 | 2921 | #endif |
cf66ba58 | 2922 | trace_net_dev_queue(skb); |
1da177e4 | 2923 | if (q->enqueue) { |
bbd8a0d3 | 2924 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 2925 | goto out; |
1da177e4 LT |
2926 | } |
2927 | ||
2928 | /* The device has no queue. Common case for software devices: | |
2929 | loopback, all the sorts of tunnels... | |
2930 | ||
932ff279 HX |
2931 | Really, it is unlikely that netif_tx_lock protection is necessary |
2932 | here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
1da177e4 LT |
2933 | counters.) |
2934 | However, it is possible, that they rely on protection | |
2935 | made by us here. | |
2936 | ||
2937 | Check this and shot the lock. It is not prone from deadlocks. | |
2938 | Either shot noqueue qdisc, it is even simpler 8) | |
2939 | */ | |
2940 | if (dev->flags & IFF_UP) { | |
2941 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
2942 | ||
c773e847 | 2943 | if (txq->xmit_lock_owner != cpu) { |
1da177e4 | 2944 | |
745e20f1 ED |
2945 | if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT) |
2946 | goto recursion_alert; | |
2947 | ||
c773e847 | 2948 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 2949 | |
73466498 | 2950 | if (!netif_xmit_stopped(txq)) { |
745e20f1 | 2951 | __this_cpu_inc(xmit_recursion); |
572a9d7b | 2952 | rc = dev_hard_start_xmit(skb, dev, txq); |
745e20f1 | 2953 | __this_cpu_dec(xmit_recursion); |
572a9d7b | 2954 | if (dev_xmit_complete(rc)) { |
c773e847 | 2955 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
2956 | goto out; |
2957 | } | |
2958 | } | |
c773e847 | 2959 | HARD_TX_UNLOCK(dev, txq); |
e87cc472 JP |
2960 | net_crit_ratelimited("Virtual device %s asks to queue packet!\n", |
2961 | dev->name); | |
1da177e4 LT |
2962 | } else { |
2963 | /* Recursion is detected! It is possible, | |
745e20f1 ED |
2964 | * unfortunately |
2965 | */ | |
2966 | recursion_alert: | |
e87cc472 JP |
2967 | net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", |
2968 | dev->name); | |
1da177e4 LT |
2969 | } |
2970 | } | |
2971 | ||
2972 | rc = -ENETDOWN; | |
d4828d85 | 2973 | rcu_read_unlock_bh(); |
1da177e4 | 2974 | |
1da177e4 LT |
2975 | kfree_skb(skb); |
2976 | return rc; | |
2977 | out: | |
d4828d85 | 2978 | rcu_read_unlock_bh(); |
1da177e4 LT |
2979 | return rc; |
2980 | } | |
d1b19dff | 2981 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 LT |
2982 | |
2983 | ||
2984 | /*======================================================================= | |
2985 | Receiver routines | |
2986 | =======================================================================*/ | |
2987 | ||
6b2bedc3 | 2988 | int netdev_max_backlog __read_mostly = 1000; |
c9e6bc64 ED |
2989 | EXPORT_SYMBOL(netdev_max_backlog); |
2990 | ||
3b098e2d | 2991 | int netdev_tstamp_prequeue __read_mostly = 1; |
6b2bedc3 SH |
2992 | int netdev_budget __read_mostly = 300; |
2993 | int weight_p __read_mostly = 64; /* old backlog weight */ | |
1da177e4 | 2994 | |
eecfd7c4 ED |
2995 | /* Called with irq disabled */ |
2996 | static inline void ____napi_schedule(struct softnet_data *sd, | |
2997 | struct napi_struct *napi) | |
2998 | { | |
2999 | list_add_tail(&napi->poll_list, &sd->poll_list); | |
3000 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3001 | } | |
3002 | ||
0a9627f2 | 3003 | /* |
bfb564e7 | 3004 | * __skb_get_rxhash: calculate a flow hash based on src/dst addresses |
bdeab991 TH |
3005 | * and src/dst port numbers. Sets rxhash in skb to non-zero hash value |
3006 | * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb | |
3007 | * if hash is a canonical 4-tuple hash over transport ports. | |
0a9627f2 | 3008 | */ |
bdeab991 | 3009 | void __skb_get_rxhash(struct sk_buff *skb) |
0a9627f2 | 3010 | { |
4504b861 ED |
3011 | struct flow_keys keys; |
3012 | u32 hash; | |
c6865cb3 | 3013 | |
4504b861 ED |
3014 | if (!skb_flow_dissect(skb, &keys)) |
3015 | return; | |
e971b722 | 3016 | |
68622342 | 3017 | if (keys.ports) |
4504b861 | 3018 | skb->l4_rxhash = 1; |
0a9627f2 | 3019 | |
b249dcb8 | 3020 | /* get a consistent hash (same value on both flow directions) */ |
68622342 CG |
3021 | if (((__force u32)keys.dst < (__force u32)keys.src) || |
3022 | (((__force u32)keys.dst == (__force u32)keys.src) && | |
3023 | ((__force u16)keys.port16[1] < (__force u16)keys.port16[0]))) { | |
4504b861 | 3024 | swap(keys.dst, keys.src); |
68622342 CG |
3025 | swap(keys.port16[0], keys.port16[1]); |
3026 | } | |
0a9627f2 | 3027 | |
4504b861 ED |
3028 | hash = jhash_3words((__force u32)keys.dst, |
3029 | (__force u32)keys.src, | |
3030 | (__force u32)keys.ports, hashrnd); | |
bfb564e7 KK |
3031 | if (!hash) |
3032 | hash = 1; | |
3033 | ||
bdeab991 | 3034 | skb->rxhash = hash; |
bfb564e7 KK |
3035 | } |
3036 | EXPORT_SYMBOL(__skb_get_rxhash); | |
3037 | ||
3038 | #ifdef CONFIG_RPS | |
3039 | ||
3040 | /* One global table that all flow-based protocols share. */ | |
6e3f7faf | 3041 | struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; |
bfb564e7 KK |
3042 | EXPORT_SYMBOL(rps_sock_flow_table); |
3043 | ||
c5905afb | 3044 | struct static_key rps_needed __read_mostly; |
adc9300e | 3045 | |
c445477d BH |
3046 | static struct rps_dev_flow * |
3047 | set_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
3048 | struct rps_dev_flow *rflow, u16 next_cpu) | |
3049 | { | |
09994d1b | 3050 | if (next_cpu != RPS_NO_CPU) { |
c445477d BH |
3051 | #ifdef CONFIG_RFS_ACCEL |
3052 | struct netdev_rx_queue *rxqueue; | |
3053 | struct rps_dev_flow_table *flow_table; | |
3054 | struct rps_dev_flow *old_rflow; | |
3055 | u32 flow_id; | |
3056 | u16 rxq_index; | |
3057 | int rc; | |
3058 | ||
3059 | /* Should we steer this flow to a different hardware queue? */ | |
69a19ee6 BH |
3060 | if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || |
3061 | !(dev->features & NETIF_F_NTUPLE)) | |
c445477d BH |
3062 | goto out; |
3063 | rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); | |
3064 | if (rxq_index == skb_get_rx_queue(skb)) | |
3065 | goto out; | |
3066 | ||
3067 | rxqueue = dev->_rx + rxq_index; | |
3068 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
3069 | if (!flow_table) | |
3070 | goto out; | |
3071 | flow_id = skb->rxhash & flow_table->mask; | |
3072 | rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, | |
3073 | rxq_index, flow_id); | |
3074 | if (rc < 0) | |
3075 | goto out; | |
3076 | old_rflow = rflow; | |
3077 | rflow = &flow_table->flows[flow_id]; | |
c445477d BH |
3078 | rflow->filter = rc; |
3079 | if (old_rflow->filter == rflow->filter) | |
3080 | old_rflow->filter = RPS_NO_FILTER; | |
3081 | out: | |
3082 | #endif | |
3083 | rflow->last_qtail = | |
09994d1b | 3084 | per_cpu(softnet_data, next_cpu).input_queue_head; |
c445477d BH |
3085 | } |
3086 | ||
09994d1b | 3087 | rflow->cpu = next_cpu; |
c445477d BH |
3088 | return rflow; |
3089 | } | |
3090 | ||
bfb564e7 KK |
3091 | /* |
3092 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
3093 | * CPU from the RPS map of the receiving queue for a given skb. | |
3094 | * rcu_read_lock must be held on entry. | |
3095 | */ | |
3096 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
3097 | struct rps_dev_flow **rflowp) | |
3098 | { | |
3099 | struct netdev_rx_queue *rxqueue; | |
6e3f7faf | 3100 | struct rps_map *map; |
bfb564e7 KK |
3101 | struct rps_dev_flow_table *flow_table; |
3102 | struct rps_sock_flow_table *sock_flow_table; | |
3103 | int cpu = -1; | |
3104 | u16 tcpu; | |
3105 | ||
3106 | if (skb_rx_queue_recorded(skb)) { | |
3107 | u16 index = skb_get_rx_queue(skb); | |
62fe0b40 BH |
3108 | if (unlikely(index >= dev->real_num_rx_queues)) { |
3109 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
3110 | "%s received packet on queue %u, but number " | |
3111 | "of RX queues is %u\n", | |
3112 | dev->name, index, dev->real_num_rx_queues); | |
bfb564e7 KK |
3113 | goto done; |
3114 | } | |
3115 | rxqueue = dev->_rx + index; | |
3116 | } else | |
3117 | rxqueue = dev->_rx; | |
3118 | ||
6e3f7faf ED |
3119 | map = rcu_dereference(rxqueue->rps_map); |
3120 | if (map) { | |
85875236 | 3121 | if (map->len == 1 && |
33d480ce | 3122 | !rcu_access_pointer(rxqueue->rps_flow_table)) { |
6febfca9 CG |
3123 | tcpu = map->cpus[0]; |
3124 | if (cpu_online(tcpu)) | |
3125 | cpu = tcpu; | |
3126 | goto done; | |
3127 | } | |
33d480ce | 3128 | } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) { |
bfb564e7 | 3129 | goto done; |
6febfca9 | 3130 | } |
bfb564e7 | 3131 | |
2d47b459 | 3132 | skb_reset_network_header(skb); |
bfb564e7 KK |
3133 | if (!skb_get_rxhash(skb)) |
3134 | goto done; | |
3135 | ||
fec5e652 TH |
3136 | flow_table = rcu_dereference(rxqueue->rps_flow_table); |
3137 | sock_flow_table = rcu_dereference(rps_sock_flow_table); | |
3138 | if (flow_table && sock_flow_table) { | |
3139 | u16 next_cpu; | |
3140 | struct rps_dev_flow *rflow; | |
3141 | ||
3142 | rflow = &flow_table->flows[skb->rxhash & flow_table->mask]; | |
3143 | tcpu = rflow->cpu; | |
3144 | ||
3145 | next_cpu = sock_flow_table->ents[skb->rxhash & | |
3146 | sock_flow_table->mask]; | |
3147 | ||
3148 | /* | |
3149 | * If the desired CPU (where last recvmsg was done) is | |
3150 | * different from current CPU (one in the rx-queue flow | |
3151 | * table entry), switch if one of the following holds: | |
3152 | * - Current CPU is unset (equal to RPS_NO_CPU). | |
3153 | * - Current CPU is offline. | |
3154 | * - The current CPU's queue tail has advanced beyond the | |
3155 | * last packet that was enqueued using this table entry. | |
3156 | * This guarantees that all previous packets for the flow | |
3157 | * have been dequeued, thus preserving in order delivery. | |
3158 | */ | |
3159 | if (unlikely(tcpu != next_cpu) && | |
3160 | (tcpu == RPS_NO_CPU || !cpu_online(tcpu) || | |
3161 | ((int)(per_cpu(softnet_data, tcpu).input_queue_head - | |
baefa31d TH |
3162 | rflow->last_qtail)) >= 0)) { |
3163 | tcpu = next_cpu; | |
c445477d | 3164 | rflow = set_rps_cpu(dev, skb, rflow, next_cpu); |
baefa31d | 3165 | } |
c445477d | 3166 | |
fec5e652 TH |
3167 | if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) { |
3168 | *rflowp = rflow; | |
3169 | cpu = tcpu; | |
3170 | goto done; | |
3171 | } | |
3172 | } | |
3173 | ||
0a9627f2 | 3174 | if (map) { |
fec5e652 | 3175 | tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32]; |
0a9627f2 TH |
3176 | |
3177 | if (cpu_online(tcpu)) { | |
3178 | cpu = tcpu; | |
3179 | goto done; | |
3180 | } | |
3181 | } | |
3182 | ||
3183 | done: | |
0a9627f2 TH |
3184 | return cpu; |
3185 | } | |
3186 | ||
c445477d BH |
3187 | #ifdef CONFIG_RFS_ACCEL |
3188 | ||
3189 | /** | |
3190 | * rps_may_expire_flow - check whether an RFS hardware filter may be removed | |
3191 | * @dev: Device on which the filter was set | |
3192 | * @rxq_index: RX queue index | |
3193 | * @flow_id: Flow ID passed to ndo_rx_flow_steer() | |
3194 | * @filter_id: Filter ID returned by ndo_rx_flow_steer() | |
3195 | * | |
3196 | * Drivers that implement ndo_rx_flow_steer() should periodically call | |
3197 | * this function for each installed filter and remove the filters for | |
3198 | * which it returns %true. | |
3199 | */ | |
3200 | bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, | |
3201 | u32 flow_id, u16 filter_id) | |
3202 | { | |
3203 | struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; | |
3204 | struct rps_dev_flow_table *flow_table; | |
3205 | struct rps_dev_flow *rflow; | |
3206 | bool expire = true; | |
3207 | int cpu; | |
3208 | ||
3209 | rcu_read_lock(); | |
3210 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
3211 | if (flow_table && flow_id <= flow_table->mask) { | |
3212 | rflow = &flow_table->flows[flow_id]; | |
3213 | cpu = ACCESS_ONCE(rflow->cpu); | |
3214 | if (rflow->filter == filter_id && cpu != RPS_NO_CPU && | |
3215 | ((int)(per_cpu(softnet_data, cpu).input_queue_head - | |
3216 | rflow->last_qtail) < | |
3217 | (int)(10 * flow_table->mask))) | |
3218 | expire = false; | |
3219 | } | |
3220 | rcu_read_unlock(); | |
3221 | return expire; | |
3222 | } | |
3223 | EXPORT_SYMBOL(rps_may_expire_flow); | |
3224 | ||
3225 | #endif /* CONFIG_RFS_ACCEL */ | |
3226 | ||
0a9627f2 | 3227 | /* Called from hardirq (IPI) context */ |
e36fa2f7 | 3228 | static void rps_trigger_softirq(void *data) |
0a9627f2 | 3229 | { |
e36fa2f7 ED |
3230 | struct softnet_data *sd = data; |
3231 | ||
eecfd7c4 | 3232 | ____napi_schedule(sd, &sd->backlog); |
dee42870 | 3233 | sd->received_rps++; |
0a9627f2 | 3234 | } |
e36fa2f7 | 3235 | |
fec5e652 | 3236 | #endif /* CONFIG_RPS */ |
0a9627f2 | 3237 | |
e36fa2f7 ED |
3238 | /* |
3239 | * Check if this softnet_data structure is another cpu one | |
3240 | * If yes, queue it to our IPI list and return 1 | |
3241 | * If no, return 0 | |
3242 | */ | |
3243 | static int rps_ipi_queued(struct softnet_data *sd) | |
3244 | { | |
3245 | #ifdef CONFIG_RPS | |
3246 | struct softnet_data *mysd = &__get_cpu_var(softnet_data); | |
3247 | ||
3248 | if (sd != mysd) { | |
3249 | sd->rps_ipi_next = mysd->rps_ipi_list; | |
3250 | mysd->rps_ipi_list = sd; | |
3251 | ||
3252 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3253 | return 1; | |
3254 | } | |
3255 | #endif /* CONFIG_RPS */ | |
3256 | return 0; | |
3257 | } | |
3258 | ||
0a9627f2 TH |
3259 | /* |
3260 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
3261 | * queue (may be a remote CPU queue). | |
3262 | */ | |
fec5e652 TH |
3263 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, |
3264 | unsigned int *qtail) | |
0a9627f2 | 3265 | { |
e36fa2f7 | 3266 | struct softnet_data *sd; |
0a9627f2 TH |
3267 | unsigned long flags; |
3268 | ||
e36fa2f7 | 3269 | sd = &per_cpu(softnet_data, cpu); |
0a9627f2 TH |
3270 | |
3271 | local_irq_save(flags); | |
0a9627f2 | 3272 | |
e36fa2f7 | 3273 | rps_lock(sd); |
6e7676c1 CG |
3274 | if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) { |
3275 | if (skb_queue_len(&sd->input_pkt_queue)) { | |
0a9627f2 | 3276 | enqueue: |
e36fa2f7 | 3277 | __skb_queue_tail(&sd->input_pkt_queue, skb); |
76cc8b13 | 3278 | input_queue_tail_incr_save(sd, qtail); |
e36fa2f7 | 3279 | rps_unlock(sd); |
152102c7 | 3280 | local_irq_restore(flags); |
0a9627f2 TH |
3281 | return NET_RX_SUCCESS; |
3282 | } | |
3283 | ||
ebda37c2 ED |
3284 | /* Schedule NAPI for backlog device |
3285 | * We can use non atomic operation since we own the queue lock | |
3286 | */ | |
3287 | if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { | |
e36fa2f7 | 3288 | if (!rps_ipi_queued(sd)) |
eecfd7c4 | 3289 | ____napi_schedule(sd, &sd->backlog); |
0a9627f2 TH |
3290 | } |
3291 | goto enqueue; | |
3292 | } | |
3293 | ||
dee42870 | 3294 | sd->dropped++; |
e36fa2f7 | 3295 | rps_unlock(sd); |
0a9627f2 | 3296 | |
0a9627f2 TH |
3297 | local_irq_restore(flags); |
3298 | ||
caf586e5 | 3299 | atomic_long_inc(&skb->dev->rx_dropped); |
0a9627f2 TH |
3300 | kfree_skb(skb); |
3301 | return NET_RX_DROP; | |
3302 | } | |
1da177e4 | 3303 | |
1da177e4 LT |
3304 | /** |
3305 | * netif_rx - post buffer to the network code | |
3306 | * @skb: buffer to post | |
3307 | * | |
3308 | * This function receives a packet from a device driver and queues it for | |
3309 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
3310 | * may be dropped during processing for congestion control or by the | |
3311 | * protocol layers. | |
3312 | * | |
3313 | * return values: | |
3314 | * NET_RX_SUCCESS (no congestion) | |
1da177e4 LT |
3315 | * NET_RX_DROP (packet was dropped) |
3316 | * | |
3317 | */ | |
3318 | ||
3319 | int netif_rx(struct sk_buff *skb) | |
3320 | { | |
b0e28f1e | 3321 | int ret; |
1da177e4 LT |
3322 | |
3323 | /* if netpoll wants it, pretend we never saw it */ | |
3324 | if (netpoll_rx(skb)) | |
3325 | return NET_RX_DROP; | |
3326 | ||
588f0330 | 3327 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
1da177e4 | 3328 | |
cf66ba58 | 3329 | trace_netif_rx(skb); |
df334545 | 3330 | #ifdef CONFIG_RPS |
c5905afb | 3331 | if (static_key_false(&rps_needed)) { |
fec5e652 | 3332 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
b0e28f1e ED |
3333 | int cpu; |
3334 | ||
cece1945 | 3335 | preempt_disable(); |
b0e28f1e | 3336 | rcu_read_lock(); |
fec5e652 TH |
3337 | |
3338 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
b0e28f1e ED |
3339 | if (cpu < 0) |
3340 | cpu = smp_processor_id(); | |
fec5e652 TH |
3341 | |
3342 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
3343 | ||
b0e28f1e | 3344 | rcu_read_unlock(); |
cece1945 | 3345 | preempt_enable(); |
adc9300e ED |
3346 | } else |
3347 | #endif | |
fec5e652 TH |
3348 | { |
3349 | unsigned int qtail; | |
3350 | ret = enqueue_to_backlog(skb, get_cpu(), &qtail); | |
3351 | put_cpu(); | |
3352 | } | |
b0e28f1e | 3353 | return ret; |
1da177e4 | 3354 | } |
d1b19dff | 3355 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
3356 | |
3357 | int netif_rx_ni(struct sk_buff *skb) | |
3358 | { | |
3359 | int err; | |
3360 | ||
3361 | preempt_disable(); | |
3362 | err = netif_rx(skb); | |
3363 | if (local_softirq_pending()) | |
3364 | do_softirq(); | |
3365 | preempt_enable(); | |
3366 | ||
3367 | return err; | |
3368 | } | |
1da177e4 LT |
3369 | EXPORT_SYMBOL(netif_rx_ni); |
3370 | ||
1da177e4 LT |
3371 | static void net_tx_action(struct softirq_action *h) |
3372 | { | |
3373 | struct softnet_data *sd = &__get_cpu_var(softnet_data); | |
3374 | ||
3375 | if (sd->completion_queue) { | |
3376 | struct sk_buff *clist; | |
3377 | ||
3378 | local_irq_disable(); | |
3379 | clist = sd->completion_queue; | |
3380 | sd->completion_queue = NULL; | |
3381 | local_irq_enable(); | |
3382 | ||
3383 | while (clist) { | |
3384 | struct sk_buff *skb = clist; | |
3385 | clist = clist->next; | |
3386 | ||
547b792c | 3387 | WARN_ON(atomic_read(&skb->users)); |
07dc22e7 | 3388 | trace_kfree_skb(skb, net_tx_action); |
1da177e4 LT |
3389 | __kfree_skb(skb); |
3390 | } | |
3391 | } | |
3392 | ||
3393 | if (sd->output_queue) { | |
37437bb2 | 3394 | struct Qdisc *head; |
1da177e4 LT |
3395 | |
3396 | local_irq_disable(); | |
3397 | head = sd->output_queue; | |
3398 | sd->output_queue = NULL; | |
a9cbd588 | 3399 | sd->output_queue_tailp = &sd->output_queue; |
1da177e4 LT |
3400 | local_irq_enable(); |
3401 | ||
3402 | while (head) { | |
37437bb2 DM |
3403 | struct Qdisc *q = head; |
3404 | spinlock_t *root_lock; | |
3405 | ||
1da177e4 LT |
3406 | head = head->next_sched; |
3407 | ||
5fb66229 | 3408 | root_lock = qdisc_lock(q); |
37437bb2 | 3409 | if (spin_trylock(root_lock)) { |
def82a1d JP |
3410 | smp_mb__before_clear_bit(); |
3411 | clear_bit(__QDISC_STATE_SCHED, | |
3412 | &q->state); | |
37437bb2 DM |
3413 | qdisc_run(q); |
3414 | spin_unlock(root_lock); | |
1da177e4 | 3415 | } else { |
195648bb | 3416 | if (!test_bit(__QDISC_STATE_DEACTIVATED, |
e8a83e10 | 3417 | &q->state)) { |
195648bb | 3418 | __netif_reschedule(q); |
e8a83e10 JP |
3419 | } else { |
3420 | smp_mb__before_clear_bit(); | |
3421 | clear_bit(__QDISC_STATE_SCHED, | |
3422 | &q->state); | |
3423 | } | |
1da177e4 LT |
3424 | } |
3425 | } | |
3426 | } | |
3427 | } | |
3428 | ||
ab95bfe0 JP |
3429 | #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \ |
3430 | (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)) | |
da678292 MM |
3431 | /* This hook is defined here for ATM LANE */ |
3432 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
3433 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 3434 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 3435 | #endif |
1da177e4 | 3436 | |
1da177e4 LT |
3437 | #ifdef CONFIG_NET_CLS_ACT |
3438 | /* TODO: Maybe we should just force sch_ingress to be compiled in | |
3439 | * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions | |
3440 | * a compare and 2 stores extra right now if we dont have it on | |
3441 | * but have CONFIG_NET_CLS_ACT | |
25985edc LDM |
3442 | * NOTE: This doesn't stop any functionality; if you dont have |
3443 | * the ingress scheduler, you just can't add policies on ingress. | |
1da177e4 LT |
3444 | * |
3445 | */ | |
24824a09 | 3446 | static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq) |
1da177e4 | 3447 | { |
1da177e4 | 3448 | struct net_device *dev = skb->dev; |
f697c3e8 | 3449 | u32 ttl = G_TC_RTTL(skb->tc_verd); |
555353cf DM |
3450 | int result = TC_ACT_OK; |
3451 | struct Qdisc *q; | |
4ec93edb | 3452 | |
de384830 | 3453 | if (unlikely(MAX_RED_LOOP < ttl++)) { |
e87cc472 JP |
3454 | net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n", |
3455 | skb->skb_iif, dev->ifindex); | |
f697c3e8 HX |
3456 | return TC_ACT_SHOT; |
3457 | } | |
1da177e4 | 3458 | |
f697c3e8 HX |
3459 | skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl); |
3460 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); | |
1da177e4 | 3461 | |
83874000 | 3462 | q = rxq->qdisc; |
8d50b53d | 3463 | if (q != &noop_qdisc) { |
83874000 | 3464 | spin_lock(qdisc_lock(q)); |
a9312ae8 DM |
3465 | if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) |
3466 | result = qdisc_enqueue_root(skb, q); | |
83874000 DM |
3467 | spin_unlock(qdisc_lock(q)); |
3468 | } | |
f697c3e8 HX |
3469 | |
3470 | return result; | |
3471 | } | |
86e65da9 | 3472 | |
f697c3e8 HX |
3473 | static inline struct sk_buff *handle_ing(struct sk_buff *skb, |
3474 | struct packet_type **pt_prev, | |
3475 | int *ret, struct net_device *orig_dev) | |
3476 | { | |
24824a09 ED |
3477 | struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue); |
3478 | ||
3479 | if (!rxq || rxq->qdisc == &noop_qdisc) | |
f697c3e8 | 3480 | goto out; |
1da177e4 | 3481 | |
f697c3e8 HX |
3482 | if (*pt_prev) { |
3483 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
3484 | *pt_prev = NULL; | |
1da177e4 LT |
3485 | } |
3486 | ||
24824a09 | 3487 | switch (ing_filter(skb, rxq)) { |
f697c3e8 HX |
3488 | case TC_ACT_SHOT: |
3489 | case TC_ACT_STOLEN: | |
3490 | kfree_skb(skb); | |
3491 | return NULL; | |
3492 | } | |
3493 | ||
3494 | out: | |
3495 | skb->tc_verd = 0; | |
3496 | return skb; | |
1da177e4 LT |
3497 | } |
3498 | #endif | |
3499 | ||
ab95bfe0 JP |
3500 | /** |
3501 | * netdev_rx_handler_register - register receive handler | |
3502 | * @dev: device to register a handler for | |
3503 | * @rx_handler: receive handler to register | |
93e2c32b | 3504 | * @rx_handler_data: data pointer that is used by rx handler |
ab95bfe0 JP |
3505 | * |
3506 | * Register a receive hander for a device. This handler will then be | |
3507 | * called from __netif_receive_skb. A negative errno code is returned | |
3508 | * on a failure. | |
3509 | * | |
3510 | * The caller must hold the rtnl_mutex. | |
8a4eb573 JP |
3511 | * |
3512 | * For a general description of rx_handler, see enum rx_handler_result. | |
ab95bfe0 JP |
3513 | */ |
3514 | int netdev_rx_handler_register(struct net_device *dev, | |
93e2c32b JP |
3515 | rx_handler_func_t *rx_handler, |
3516 | void *rx_handler_data) | |
ab95bfe0 JP |
3517 | { |
3518 | ASSERT_RTNL(); | |
3519 | ||
3520 | if (dev->rx_handler) | |
3521 | return -EBUSY; | |
3522 | ||
93e2c32b | 3523 | rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); |
ab95bfe0 JP |
3524 | rcu_assign_pointer(dev->rx_handler, rx_handler); |
3525 | ||
3526 | return 0; | |
3527 | } | |
3528 | EXPORT_SYMBOL_GPL(netdev_rx_handler_register); | |
3529 | ||
3530 | /** | |
3531 | * netdev_rx_handler_unregister - unregister receive handler | |
3532 | * @dev: device to unregister a handler from | |
3533 | * | |
3534 | * Unregister a receive hander from a device. | |
3535 | * | |
3536 | * The caller must hold the rtnl_mutex. | |
3537 | */ | |
3538 | void netdev_rx_handler_unregister(struct net_device *dev) | |
3539 | { | |
3540 | ||
3541 | ASSERT_RTNL(); | |
a9b3cd7f SH |
3542 | RCU_INIT_POINTER(dev->rx_handler, NULL); |
3543 | RCU_INIT_POINTER(dev->rx_handler_data, NULL); | |
ab95bfe0 JP |
3544 | } |
3545 | EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); | |
3546 | ||
b4b9e355 MG |
3547 | /* |
3548 | * Limit the use of PFMEMALLOC reserves to those protocols that implement | |
3549 | * the special handling of PFMEMALLOC skbs. | |
3550 | */ | |
3551 | static bool skb_pfmemalloc_protocol(struct sk_buff *skb) | |
3552 | { | |
3553 | switch (skb->protocol) { | |
3554 | case __constant_htons(ETH_P_ARP): | |
3555 | case __constant_htons(ETH_P_IP): | |
3556 | case __constant_htons(ETH_P_IPV6): | |
3557 | case __constant_htons(ETH_P_8021Q): | |
3558 | return true; | |
3559 | default: | |
3560 | return false; | |
3561 | } | |
3562 | } | |
3563 | ||
10f744d2 | 3564 | static int __netif_receive_skb(struct sk_buff *skb) |
1da177e4 LT |
3565 | { |
3566 | struct packet_type *ptype, *pt_prev; | |
ab95bfe0 | 3567 | rx_handler_func_t *rx_handler; |
f2ccd8fa | 3568 | struct net_device *orig_dev; |
63d8ea7f | 3569 | struct net_device *null_or_dev; |
8a4eb573 | 3570 | bool deliver_exact = false; |
1da177e4 | 3571 | int ret = NET_RX_DROP; |
252e3346 | 3572 | __be16 type; |
b4b9e355 | 3573 | unsigned long pflags = current->flags; |
1da177e4 | 3574 | |
588f0330 | 3575 | net_timestamp_check(!netdev_tstamp_prequeue, skb); |
81bbb3d4 | 3576 | |
cf66ba58 | 3577 | trace_netif_receive_skb(skb); |
9b22ea56 | 3578 | |
b4b9e355 MG |
3579 | /* |
3580 | * PFMEMALLOC skbs are special, they should | |
3581 | * - be delivered to SOCK_MEMALLOC sockets only | |
3582 | * - stay away from userspace | |
3583 | * - have bounded memory usage | |
3584 | * | |
3585 | * Use PF_MEMALLOC as this saves us from propagating the allocation | |
3586 | * context down to all allocation sites. | |
3587 | */ | |
3588 | if (sk_memalloc_socks() && skb_pfmemalloc(skb)) | |
3589 | current->flags |= PF_MEMALLOC; | |
3590 | ||
1da177e4 | 3591 | /* if we've gotten here through NAPI, check netpoll */ |
bea3348e | 3592 | if (netpoll_receive_skb(skb)) |
b4b9e355 | 3593 | goto out; |
1da177e4 | 3594 | |
cc9bd5ce | 3595 | orig_dev = skb->dev; |
8f903c70 | 3596 | |
c1d2bbe1 | 3597 | skb_reset_network_header(skb); |
fda55eca ED |
3598 | if (!skb_transport_header_was_set(skb)) |
3599 | skb_reset_transport_header(skb); | |
0b5c9db1 | 3600 | skb_reset_mac_len(skb); |
1da177e4 LT |
3601 | |
3602 | pt_prev = NULL; | |
3603 | ||
3604 | rcu_read_lock(); | |
3605 | ||
63d8ea7f | 3606 | another_round: |
b6858177 | 3607 | skb->skb_iif = skb->dev->ifindex; |
63d8ea7f DM |
3608 | |
3609 | __this_cpu_inc(softnet_data.processed); | |
3610 | ||
bcc6d479 JP |
3611 | if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) { |
3612 | skb = vlan_untag(skb); | |
3613 | if (unlikely(!skb)) | |
b4b9e355 | 3614 | goto unlock; |
bcc6d479 JP |
3615 | } |
3616 | ||
1da177e4 LT |
3617 | #ifdef CONFIG_NET_CLS_ACT |
3618 | if (skb->tc_verd & TC_NCLS) { | |
3619 | skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); | |
3620 | goto ncls; | |
3621 | } | |
3622 | #endif | |
3623 | ||
b4b9e355 MG |
3624 | if (sk_memalloc_socks() && skb_pfmemalloc(skb)) |
3625 | goto skip_taps; | |
3626 | ||
1da177e4 | 3627 | list_for_each_entry_rcu(ptype, &ptype_all, list) { |
63d8ea7f | 3628 | if (!ptype->dev || ptype->dev == skb->dev) { |
4ec93edb | 3629 | if (pt_prev) |
f2ccd8fa | 3630 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
3631 | pt_prev = ptype; |
3632 | } | |
3633 | } | |
3634 | ||
b4b9e355 | 3635 | skip_taps: |
1da177e4 | 3636 | #ifdef CONFIG_NET_CLS_ACT |
f697c3e8 HX |
3637 | skb = handle_ing(skb, &pt_prev, &ret, orig_dev); |
3638 | if (!skb) | |
b4b9e355 | 3639 | goto unlock; |
1da177e4 LT |
3640 | ncls: |
3641 | #endif | |
3642 | ||
b4b9e355 MG |
3643 | if (sk_memalloc_socks() && skb_pfmemalloc(skb) |
3644 | && !skb_pfmemalloc_protocol(skb)) | |
3645 | goto drop; | |
3646 | ||
2425717b JF |
3647 | if (vlan_tx_tag_present(skb)) { |
3648 | if (pt_prev) { | |
3649 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3650 | pt_prev = NULL; | |
3651 | } | |
48cc32d3 | 3652 | if (vlan_do_receive(&skb)) |
2425717b JF |
3653 | goto another_round; |
3654 | else if (unlikely(!skb)) | |
b4b9e355 | 3655 | goto unlock; |
2425717b JF |
3656 | } |
3657 | ||
48cc32d3 | 3658 | rx_handler = rcu_dereference(skb->dev->rx_handler); |
ab95bfe0 JP |
3659 | if (rx_handler) { |
3660 | if (pt_prev) { | |
3661 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3662 | pt_prev = NULL; | |
3663 | } | |
8a4eb573 JP |
3664 | switch (rx_handler(&skb)) { |
3665 | case RX_HANDLER_CONSUMED: | |
b4b9e355 | 3666 | goto unlock; |
8a4eb573 | 3667 | case RX_HANDLER_ANOTHER: |
63d8ea7f | 3668 | goto another_round; |
8a4eb573 JP |
3669 | case RX_HANDLER_EXACT: |
3670 | deliver_exact = true; | |
3671 | case RX_HANDLER_PASS: | |
3672 | break; | |
3673 | default: | |
3674 | BUG(); | |
3675 | } | |
ab95bfe0 | 3676 | } |
1da177e4 | 3677 | |
48cc32d3 FZ |
3678 | if (vlan_tx_nonzero_tag_present(skb)) |
3679 | skb->pkt_type = PACKET_OTHERHOST; | |
3680 | ||
63d8ea7f | 3681 | /* deliver only exact match when indicated */ |
8a4eb573 | 3682 | null_or_dev = deliver_exact ? skb->dev : NULL; |
1f3c8804 | 3683 | |
1da177e4 | 3684 | type = skb->protocol; |
82d8a867 PE |
3685 | list_for_each_entry_rcu(ptype, |
3686 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
63d8ea7f | 3687 | if (ptype->type == type && |
e3f48d37 JP |
3688 | (ptype->dev == null_or_dev || ptype->dev == skb->dev || |
3689 | ptype->dev == orig_dev)) { | |
4ec93edb | 3690 | if (pt_prev) |
f2ccd8fa | 3691 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
3692 | pt_prev = ptype; |
3693 | } | |
3694 | } | |
3695 | ||
3696 | if (pt_prev) { | |
1080e512 | 3697 | if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) |
0e698bf6 | 3698 | goto drop; |
1080e512 MT |
3699 | else |
3700 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1da177e4 | 3701 | } else { |
b4b9e355 | 3702 | drop: |
caf586e5 | 3703 | atomic_long_inc(&skb->dev->rx_dropped); |
1da177e4 LT |
3704 | kfree_skb(skb); |
3705 | /* Jamal, now you will not able to escape explaining | |
3706 | * me how you were going to use this. :-) | |
3707 | */ | |
3708 | ret = NET_RX_DROP; | |
3709 | } | |
3710 | ||
b4b9e355 | 3711 | unlock: |
1da177e4 | 3712 | rcu_read_unlock(); |
b4b9e355 MG |
3713 | out: |
3714 | tsk_restore_flags(current, pflags, PF_MEMALLOC); | |
1da177e4 LT |
3715 | return ret; |
3716 | } | |
0a9627f2 TH |
3717 | |
3718 | /** | |
3719 | * netif_receive_skb - process receive buffer from network | |
3720 | * @skb: buffer to process | |
3721 | * | |
3722 | * netif_receive_skb() is the main receive data processing function. | |
3723 | * It always succeeds. The buffer may be dropped during processing | |
3724 | * for congestion control or by the protocol layers. | |
3725 | * | |
3726 | * This function may only be called from softirq context and interrupts | |
3727 | * should be enabled. | |
3728 | * | |
3729 | * Return values (usually ignored): | |
3730 | * NET_RX_SUCCESS: no congestion | |
3731 | * NET_RX_DROP: packet was dropped | |
3732 | */ | |
3733 | int netif_receive_skb(struct sk_buff *skb) | |
3734 | { | |
588f0330 | 3735 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
3b098e2d | 3736 | |
c1f19b51 RC |
3737 | if (skb_defer_rx_timestamp(skb)) |
3738 | return NET_RX_SUCCESS; | |
3739 | ||
df334545 | 3740 | #ifdef CONFIG_RPS |
c5905afb | 3741 | if (static_key_false(&rps_needed)) { |
3b098e2d ED |
3742 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
3743 | int cpu, ret; | |
fec5e652 | 3744 | |
3b098e2d ED |
3745 | rcu_read_lock(); |
3746 | ||
3747 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
0a9627f2 | 3748 | |
3b098e2d ED |
3749 | if (cpu >= 0) { |
3750 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
3751 | rcu_read_unlock(); | |
adc9300e | 3752 | return ret; |
3b098e2d | 3753 | } |
adc9300e | 3754 | rcu_read_unlock(); |
fec5e652 | 3755 | } |
1e94d72f | 3756 | #endif |
adc9300e | 3757 | return __netif_receive_skb(skb); |
0a9627f2 | 3758 | } |
d1b19dff | 3759 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 3760 | |
88751275 ED |
3761 | /* Network device is going away, flush any packets still pending |
3762 | * Called with irqs disabled. | |
3763 | */ | |
152102c7 | 3764 | static void flush_backlog(void *arg) |
6e583ce5 | 3765 | { |
152102c7 | 3766 | struct net_device *dev = arg; |
e36fa2f7 | 3767 | struct softnet_data *sd = &__get_cpu_var(softnet_data); |
6e583ce5 SH |
3768 | struct sk_buff *skb, *tmp; |
3769 | ||
e36fa2f7 | 3770 | rps_lock(sd); |
6e7676c1 | 3771 | skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { |
6e583ce5 | 3772 | if (skb->dev == dev) { |
e36fa2f7 | 3773 | __skb_unlink(skb, &sd->input_pkt_queue); |
6e583ce5 | 3774 | kfree_skb(skb); |
76cc8b13 | 3775 | input_queue_head_incr(sd); |
6e583ce5 | 3776 | } |
6e7676c1 | 3777 | } |
e36fa2f7 | 3778 | rps_unlock(sd); |
6e7676c1 CG |
3779 | |
3780 | skb_queue_walk_safe(&sd->process_queue, skb, tmp) { | |
3781 | if (skb->dev == dev) { | |
3782 | __skb_unlink(skb, &sd->process_queue); | |
3783 | kfree_skb(skb); | |
76cc8b13 | 3784 | input_queue_head_incr(sd); |
6e7676c1 CG |
3785 | } |
3786 | } | |
6e583ce5 SH |
3787 | } |
3788 | ||
d565b0a1 HX |
3789 | static int napi_gro_complete(struct sk_buff *skb) |
3790 | { | |
22061d80 | 3791 | struct packet_offload *ptype; |
d565b0a1 | 3792 | __be16 type = skb->protocol; |
22061d80 | 3793 | struct list_head *head = &offload_base; |
d565b0a1 HX |
3794 | int err = -ENOENT; |
3795 | ||
c3c7c254 ED |
3796 | BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); |
3797 | ||
fc59f9a3 HX |
3798 | if (NAPI_GRO_CB(skb)->count == 1) { |
3799 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 3800 | goto out; |
fc59f9a3 | 3801 | } |
d565b0a1 HX |
3802 | |
3803 | rcu_read_lock(); | |
3804 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 3805 | if (ptype->type != type || !ptype->callbacks.gro_complete) |
d565b0a1 HX |
3806 | continue; |
3807 | ||
f191a1d1 | 3808 | err = ptype->callbacks.gro_complete(skb); |
d565b0a1 HX |
3809 | break; |
3810 | } | |
3811 | rcu_read_unlock(); | |
3812 | ||
3813 | if (err) { | |
3814 | WARN_ON(&ptype->list == head); | |
3815 | kfree_skb(skb); | |
3816 | return NET_RX_SUCCESS; | |
3817 | } | |
3818 | ||
3819 | out: | |
d565b0a1 HX |
3820 | return netif_receive_skb(skb); |
3821 | } | |
3822 | ||
2e71a6f8 ED |
3823 | /* napi->gro_list contains packets ordered by age. |
3824 | * youngest packets at the head of it. | |
3825 | * Complete skbs in reverse order to reduce latencies. | |
3826 | */ | |
3827 | void napi_gro_flush(struct napi_struct *napi, bool flush_old) | |
d565b0a1 | 3828 | { |
2e71a6f8 | 3829 | struct sk_buff *skb, *prev = NULL; |
d565b0a1 | 3830 | |
2e71a6f8 ED |
3831 | /* scan list and build reverse chain */ |
3832 | for (skb = napi->gro_list; skb != NULL; skb = skb->next) { | |
3833 | skb->prev = prev; | |
3834 | prev = skb; | |
3835 | } | |
3836 | ||
3837 | for (skb = prev; skb; skb = prev) { | |
d565b0a1 | 3838 | skb->next = NULL; |
2e71a6f8 ED |
3839 | |
3840 | if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) | |
3841 | return; | |
3842 | ||
3843 | prev = skb->prev; | |
d565b0a1 | 3844 | napi_gro_complete(skb); |
2e71a6f8 | 3845 | napi->gro_count--; |
d565b0a1 HX |
3846 | } |
3847 | ||
3848 | napi->gro_list = NULL; | |
3849 | } | |
86cac58b | 3850 | EXPORT_SYMBOL(napi_gro_flush); |
d565b0a1 | 3851 | |
89c5fa33 ED |
3852 | static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb) |
3853 | { | |
3854 | struct sk_buff *p; | |
3855 | unsigned int maclen = skb->dev->hard_header_len; | |
3856 | ||
3857 | for (p = napi->gro_list; p; p = p->next) { | |
3858 | unsigned long diffs; | |
3859 | ||
3860 | diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; | |
3861 | diffs |= p->vlan_tci ^ skb->vlan_tci; | |
3862 | if (maclen == ETH_HLEN) | |
3863 | diffs |= compare_ether_header(skb_mac_header(p), | |
3864 | skb_gro_mac_header(skb)); | |
3865 | else if (!diffs) | |
3866 | diffs = memcmp(skb_mac_header(p), | |
3867 | skb_gro_mac_header(skb), | |
3868 | maclen); | |
3869 | NAPI_GRO_CB(p)->same_flow = !diffs; | |
3870 | NAPI_GRO_CB(p)->flush = 0; | |
3871 | } | |
3872 | } | |
3873 | ||
bb728820 | 3874 | static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 HX |
3875 | { |
3876 | struct sk_buff **pp = NULL; | |
22061d80 | 3877 | struct packet_offload *ptype; |
d565b0a1 | 3878 | __be16 type = skb->protocol; |
22061d80 | 3879 | struct list_head *head = &offload_base; |
0da2afd5 | 3880 | int same_flow; |
d565b0a1 | 3881 | int mac_len; |
5b252f0c | 3882 | enum gro_result ret; |
d565b0a1 | 3883 | |
ce9e76c8 | 3884 | if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb)) |
d565b0a1 HX |
3885 | goto normal; |
3886 | ||
21dc3301 | 3887 | if (skb_is_gso(skb) || skb_has_frag_list(skb)) |
f17f5c91 HX |
3888 | goto normal; |
3889 | ||
89c5fa33 ED |
3890 | gro_list_prepare(napi, skb); |
3891 | ||
d565b0a1 HX |
3892 | rcu_read_lock(); |
3893 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 3894 | if (ptype->type != type || !ptype->callbacks.gro_receive) |
d565b0a1 HX |
3895 | continue; |
3896 | ||
86911732 | 3897 | skb_set_network_header(skb, skb_gro_offset(skb)); |
d565b0a1 HX |
3898 | mac_len = skb->network_header - skb->mac_header; |
3899 | skb->mac_len = mac_len; | |
3900 | NAPI_GRO_CB(skb)->same_flow = 0; | |
3901 | NAPI_GRO_CB(skb)->flush = 0; | |
5d38a079 | 3902 | NAPI_GRO_CB(skb)->free = 0; |
d565b0a1 | 3903 | |
f191a1d1 | 3904 | pp = ptype->callbacks.gro_receive(&napi->gro_list, skb); |
d565b0a1 HX |
3905 | break; |
3906 | } | |
3907 | rcu_read_unlock(); | |
3908 | ||
3909 | if (&ptype->list == head) | |
3910 | goto normal; | |
3911 | ||
0da2afd5 | 3912 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 3913 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 3914 | |
d565b0a1 HX |
3915 | if (pp) { |
3916 | struct sk_buff *nskb = *pp; | |
3917 | ||
3918 | *pp = nskb->next; | |
3919 | nskb->next = NULL; | |
3920 | napi_gro_complete(nskb); | |
4ae5544f | 3921 | napi->gro_count--; |
d565b0a1 HX |
3922 | } |
3923 | ||
0da2afd5 | 3924 | if (same_flow) |
d565b0a1 HX |
3925 | goto ok; |
3926 | ||
4ae5544f | 3927 | if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS) |
d565b0a1 | 3928 | goto normal; |
d565b0a1 | 3929 | |
4ae5544f | 3930 | napi->gro_count++; |
d565b0a1 | 3931 | NAPI_GRO_CB(skb)->count = 1; |
2e71a6f8 | 3932 | NAPI_GRO_CB(skb)->age = jiffies; |
86911732 | 3933 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
d565b0a1 HX |
3934 | skb->next = napi->gro_list; |
3935 | napi->gro_list = skb; | |
5d0d9be8 | 3936 | ret = GRO_HELD; |
d565b0a1 | 3937 | |
ad0f9904 | 3938 | pull: |
cb18978c HX |
3939 | if (skb_headlen(skb) < skb_gro_offset(skb)) { |
3940 | int grow = skb_gro_offset(skb) - skb_headlen(skb); | |
3941 | ||
3942 | BUG_ON(skb->end - skb->tail < grow); | |
3943 | ||
3944 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
3945 | ||
3946 | skb->tail += grow; | |
3947 | skb->data_len -= grow; | |
3948 | ||
3949 | skb_shinfo(skb)->frags[0].page_offset += grow; | |
9e903e08 | 3950 | skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow); |
cb18978c | 3951 | |
9e903e08 | 3952 | if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) { |
ea2ab693 | 3953 | skb_frag_unref(skb, 0); |
cb18978c HX |
3954 | memmove(skb_shinfo(skb)->frags, |
3955 | skb_shinfo(skb)->frags + 1, | |
e5093aec | 3956 | --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t)); |
cb18978c | 3957 | } |
ad0f9904 HX |
3958 | } |
3959 | ||
d565b0a1 | 3960 | ok: |
5d0d9be8 | 3961 | return ret; |
d565b0a1 HX |
3962 | |
3963 | normal: | |
ad0f9904 HX |
3964 | ret = GRO_NORMAL; |
3965 | goto pull; | |
5d38a079 | 3966 | } |
96e93eab | 3967 | |
5d38a079 | 3968 | |
bb728820 | 3969 | static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) |
5d38a079 | 3970 | { |
5d0d9be8 HX |
3971 | switch (ret) { |
3972 | case GRO_NORMAL: | |
c7c4b3b6 BH |
3973 | if (netif_receive_skb(skb)) |
3974 | ret = GRO_DROP; | |
3975 | break; | |
5d38a079 | 3976 | |
5d0d9be8 | 3977 | case GRO_DROP: |
5d38a079 HX |
3978 | kfree_skb(skb); |
3979 | break; | |
5b252f0c | 3980 | |
daa86548 | 3981 | case GRO_MERGED_FREE: |
d7e8883c ED |
3982 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) |
3983 | kmem_cache_free(skbuff_head_cache, skb); | |
3984 | else | |
3985 | __kfree_skb(skb); | |
daa86548 ED |
3986 | break; |
3987 | ||
5b252f0c BH |
3988 | case GRO_HELD: |
3989 | case GRO_MERGED: | |
3990 | break; | |
5d38a079 HX |
3991 | } |
3992 | ||
c7c4b3b6 | 3993 | return ret; |
5d0d9be8 | 3994 | } |
5d0d9be8 | 3995 | |
ca07e43e | 3996 | static void skb_gro_reset_offset(struct sk_buff *skb) |
78a478d0 | 3997 | { |
ca07e43e ED |
3998 | const struct skb_shared_info *pinfo = skb_shinfo(skb); |
3999 | const skb_frag_t *frag0 = &pinfo->frags[0]; | |
4000 | ||
78a478d0 HX |
4001 | NAPI_GRO_CB(skb)->data_offset = 0; |
4002 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
7489594c | 4003 | NAPI_GRO_CB(skb)->frag0_len = 0; |
78a478d0 | 4004 | |
78d3fd0b | 4005 | if (skb->mac_header == skb->tail && |
ca07e43e ED |
4006 | pinfo->nr_frags && |
4007 | !PageHighMem(skb_frag_page(frag0))) { | |
4008 | NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); | |
4009 | NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0); | |
7489594c | 4010 | } |
78a478d0 | 4011 | } |
78a478d0 | 4012 | |
c7c4b3b6 | 4013 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 4014 | { |
86911732 HX |
4015 | skb_gro_reset_offset(skb); |
4016 | ||
89c5fa33 | 4017 | return napi_skb_finish(dev_gro_receive(napi, skb), skb); |
d565b0a1 HX |
4018 | } |
4019 | EXPORT_SYMBOL(napi_gro_receive); | |
4020 | ||
d0c2b0d2 | 4021 | static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab | 4022 | { |
96e93eab | 4023 | __skb_pull(skb, skb_headlen(skb)); |
2a2a459e ED |
4024 | /* restore the reserve we had after netdev_alloc_skb_ip_align() */ |
4025 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); | |
3701e513 | 4026 | skb->vlan_tci = 0; |
66c46d74 | 4027 | skb->dev = napi->dev; |
6d152e23 | 4028 | skb->skb_iif = 0; |
96e93eab HX |
4029 | |
4030 | napi->skb = skb; | |
4031 | } | |
96e93eab | 4032 | |
76620aaf | 4033 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 4034 | { |
5d38a079 | 4035 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
4036 | |
4037 | if (!skb) { | |
89d71a66 ED |
4038 | skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD); |
4039 | if (skb) | |
4040 | napi->skb = skb; | |
80595d59 | 4041 | } |
96e93eab HX |
4042 | return skb; |
4043 | } | |
76620aaf | 4044 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 4045 | |
bb728820 | 4046 | static gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb, |
c7c4b3b6 | 4047 | gro_result_t ret) |
96e93eab | 4048 | { |
5d0d9be8 HX |
4049 | switch (ret) { |
4050 | case GRO_NORMAL: | |
86911732 | 4051 | case GRO_HELD: |
e76b69cc | 4052 | skb->protocol = eth_type_trans(skb, skb->dev); |
86911732 | 4053 | |
c7c4b3b6 BH |
4054 | if (ret == GRO_HELD) |
4055 | skb_gro_pull(skb, -ETH_HLEN); | |
4056 | else if (netif_receive_skb(skb)) | |
4057 | ret = GRO_DROP; | |
86911732 | 4058 | break; |
5d38a079 | 4059 | |
5d0d9be8 | 4060 | case GRO_DROP: |
5d0d9be8 HX |
4061 | case GRO_MERGED_FREE: |
4062 | napi_reuse_skb(napi, skb); | |
4063 | break; | |
5b252f0c BH |
4064 | |
4065 | case GRO_MERGED: | |
4066 | break; | |
5d0d9be8 | 4067 | } |
5d38a079 | 4068 | |
c7c4b3b6 | 4069 | return ret; |
5d38a079 | 4070 | } |
5d0d9be8 | 4071 | |
4adb9c4a | 4072 | static struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
76620aaf HX |
4073 | { |
4074 | struct sk_buff *skb = napi->skb; | |
4075 | struct ethhdr *eth; | |
a5b1cf28 HX |
4076 | unsigned int hlen; |
4077 | unsigned int off; | |
76620aaf HX |
4078 | |
4079 | napi->skb = NULL; | |
4080 | ||
4081 | skb_reset_mac_header(skb); | |
4082 | skb_gro_reset_offset(skb); | |
4083 | ||
a5b1cf28 HX |
4084 | off = skb_gro_offset(skb); |
4085 | hlen = off + sizeof(*eth); | |
4086 | eth = skb_gro_header_fast(skb, off); | |
4087 | if (skb_gro_header_hard(skb, hlen)) { | |
4088 | eth = skb_gro_header_slow(skb, hlen, off); | |
4089 | if (unlikely(!eth)) { | |
4090 | napi_reuse_skb(napi, skb); | |
4091 | skb = NULL; | |
4092 | goto out; | |
4093 | } | |
76620aaf HX |
4094 | } |
4095 | ||
4096 | skb_gro_pull(skb, sizeof(*eth)); | |
4097 | ||
4098 | /* | |
4099 | * This works because the only protocols we care about don't require | |
4100 | * special handling. We'll fix it up properly at the end. | |
4101 | */ | |
4102 | skb->protocol = eth->h_proto; | |
4103 | ||
4104 | out: | |
4105 | return skb; | |
4106 | } | |
76620aaf | 4107 | |
c7c4b3b6 | 4108 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 4109 | { |
76620aaf | 4110 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 HX |
4111 | |
4112 | if (!skb) | |
c7c4b3b6 | 4113 | return GRO_DROP; |
5d0d9be8 | 4114 | |
89c5fa33 | 4115 | return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); |
5d0d9be8 | 4116 | } |
5d38a079 HX |
4117 | EXPORT_SYMBOL(napi_gro_frags); |
4118 | ||
e326bed2 ED |
4119 | /* |
4120 | * net_rps_action sends any pending IPI's for rps. | |
4121 | * Note: called with local irq disabled, but exits with local irq enabled. | |
4122 | */ | |
4123 | static void net_rps_action_and_irq_enable(struct softnet_data *sd) | |
4124 | { | |
4125 | #ifdef CONFIG_RPS | |
4126 | struct softnet_data *remsd = sd->rps_ipi_list; | |
4127 | ||
4128 | if (remsd) { | |
4129 | sd->rps_ipi_list = NULL; | |
4130 | ||
4131 | local_irq_enable(); | |
4132 | ||
4133 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
4134 | while (remsd) { | |
4135 | struct softnet_data *next = remsd->rps_ipi_next; | |
4136 | ||
4137 | if (cpu_online(remsd->cpu)) | |
4138 | __smp_call_function_single(remsd->cpu, | |
4139 | &remsd->csd, 0); | |
4140 | remsd = next; | |
4141 | } | |
4142 | } else | |
4143 | #endif | |
4144 | local_irq_enable(); | |
4145 | } | |
4146 | ||
bea3348e | 4147 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 LT |
4148 | { |
4149 | int work = 0; | |
eecfd7c4 | 4150 | struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); |
1da177e4 | 4151 | |
e326bed2 ED |
4152 | #ifdef CONFIG_RPS |
4153 | /* Check if we have pending ipi, its better to send them now, | |
4154 | * not waiting net_rx_action() end. | |
4155 | */ | |
4156 | if (sd->rps_ipi_list) { | |
4157 | local_irq_disable(); | |
4158 | net_rps_action_and_irq_enable(sd); | |
4159 | } | |
4160 | #endif | |
bea3348e | 4161 | napi->weight = weight_p; |
6e7676c1 CG |
4162 | local_irq_disable(); |
4163 | while (work < quota) { | |
1da177e4 | 4164 | struct sk_buff *skb; |
6e7676c1 CG |
4165 | unsigned int qlen; |
4166 | ||
4167 | while ((skb = __skb_dequeue(&sd->process_queue))) { | |
4168 | local_irq_enable(); | |
4169 | __netif_receive_skb(skb); | |
6e7676c1 | 4170 | local_irq_disable(); |
76cc8b13 TH |
4171 | input_queue_head_incr(sd); |
4172 | if (++work >= quota) { | |
4173 | local_irq_enable(); | |
4174 | return work; | |
4175 | } | |
6e7676c1 | 4176 | } |
1da177e4 | 4177 | |
e36fa2f7 | 4178 | rps_lock(sd); |
6e7676c1 | 4179 | qlen = skb_queue_len(&sd->input_pkt_queue); |
76cc8b13 | 4180 | if (qlen) |
6e7676c1 CG |
4181 | skb_queue_splice_tail_init(&sd->input_pkt_queue, |
4182 | &sd->process_queue); | |
76cc8b13 | 4183 | |
6e7676c1 | 4184 | if (qlen < quota - work) { |
eecfd7c4 ED |
4185 | /* |
4186 | * Inline a custom version of __napi_complete(). | |
4187 | * only current cpu owns and manipulates this napi, | |
4188 | * and NAPI_STATE_SCHED is the only possible flag set on backlog. | |
4189 | * we can use a plain write instead of clear_bit(), | |
4190 | * and we dont need an smp_mb() memory barrier. | |
4191 | */ | |
4192 | list_del(&napi->poll_list); | |
4193 | napi->state = 0; | |
4194 | ||
6e7676c1 | 4195 | quota = work + qlen; |
bea3348e | 4196 | } |
e36fa2f7 | 4197 | rps_unlock(sd); |
6e7676c1 CG |
4198 | } |
4199 | local_irq_enable(); | |
1da177e4 | 4200 | |
bea3348e SH |
4201 | return work; |
4202 | } | |
1da177e4 | 4203 | |
bea3348e SH |
4204 | /** |
4205 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 4206 | * @n: entry to schedule |
bea3348e SH |
4207 | * |
4208 | * The entry's receive function will be scheduled to run | |
4209 | */ | |
b5606c2d | 4210 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
4211 | { |
4212 | unsigned long flags; | |
1da177e4 | 4213 | |
bea3348e | 4214 | local_irq_save(flags); |
eecfd7c4 | 4215 | ____napi_schedule(&__get_cpu_var(softnet_data), n); |
bea3348e | 4216 | local_irq_restore(flags); |
1da177e4 | 4217 | } |
bea3348e SH |
4218 | EXPORT_SYMBOL(__napi_schedule); |
4219 | ||
d565b0a1 HX |
4220 | void __napi_complete(struct napi_struct *n) |
4221 | { | |
4222 | BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); | |
4223 | BUG_ON(n->gro_list); | |
4224 | ||
4225 | list_del(&n->poll_list); | |
4226 | smp_mb__before_clear_bit(); | |
4227 | clear_bit(NAPI_STATE_SCHED, &n->state); | |
4228 | } | |
4229 | EXPORT_SYMBOL(__napi_complete); | |
4230 | ||
4231 | void napi_complete(struct napi_struct *n) | |
4232 | { | |
4233 | unsigned long flags; | |
4234 | ||
4235 | /* | |
4236 | * don't let napi dequeue from the cpu poll list | |
4237 | * just in case its running on a different cpu | |
4238 | */ | |
4239 | if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state))) | |
4240 | return; | |
4241 | ||
2e71a6f8 | 4242 | napi_gro_flush(n, false); |
d565b0a1 HX |
4243 | local_irq_save(flags); |
4244 | __napi_complete(n); | |
4245 | local_irq_restore(flags); | |
4246 | } | |
4247 | EXPORT_SYMBOL(napi_complete); | |
4248 | ||
4249 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, | |
4250 | int (*poll)(struct napi_struct *, int), int weight) | |
4251 | { | |
4252 | INIT_LIST_HEAD(&napi->poll_list); | |
4ae5544f | 4253 | napi->gro_count = 0; |
d565b0a1 | 4254 | napi->gro_list = NULL; |
5d38a079 | 4255 | napi->skb = NULL; |
d565b0a1 HX |
4256 | napi->poll = poll; |
4257 | napi->weight = weight; | |
4258 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 4259 | napi->dev = dev; |
5d38a079 | 4260 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
4261 | spin_lock_init(&napi->poll_lock); |
4262 | napi->poll_owner = -1; | |
4263 | #endif | |
4264 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
4265 | } | |
4266 | EXPORT_SYMBOL(netif_napi_add); | |
4267 | ||
4268 | void netif_napi_del(struct napi_struct *napi) | |
4269 | { | |
4270 | struct sk_buff *skb, *next; | |
4271 | ||
d7b06636 | 4272 | list_del_init(&napi->dev_list); |
76620aaf | 4273 | napi_free_frags(napi); |
d565b0a1 HX |
4274 | |
4275 | for (skb = napi->gro_list; skb; skb = next) { | |
4276 | next = skb->next; | |
4277 | skb->next = NULL; | |
4278 | kfree_skb(skb); | |
4279 | } | |
4280 | ||
4281 | napi->gro_list = NULL; | |
4ae5544f | 4282 | napi->gro_count = 0; |
d565b0a1 HX |
4283 | } |
4284 | EXPORT_SYMBOL(netif_napi_del); | |
4285 | ||
1da177e4 LT |
4286 | static void net_rx_action(struct softirq_action *h) |
4287 | { | |
e326bed2 | 4288 | struct softnet_data *sd = &__get_cpu_var(softnet_data); |
24f8b238 | 4289 | unsigned long time_limit = jiffies + 2; |
51b0bded | 4290 | int budget = netdev_budget; |
53fb95d3 MM |
4291 | void *have; |
4292 | ||
1da177e4 LT |
4293 | local_irq_disable(); |
4294 | ||
e326bed2 | 4295 | while (!list_empty(&sd->poll_list)) { |
bea3348e SH |
4296 | struct napi_struct *n; |
4297 | int work, weight; | |
1da177e4 | 4298 | |
bea3348e | 4299 | /* If softirq window is exhuasted then punt. |
24f8b238 SH |
4300 | * Allow this to run for 2 jiffies since which will allow |
4301 | * an average latency of 1.5/HZ. | |
bea3348e | 4302 | */ |
24f8b238 | 4303 | if (unlikely(budget <= 0 || time_after(jiffies, time_limit))) |
1da177e4 LT |
4304 | goto softnet_break; |
4305 | ||
4306 | local_irq_enable(); | |
4307 | ||
bea3348e SH |
4308 | /* Even though interrupts have been re-enabled, this |
4309 | * access is safe because interrupts can only add new | |
4310 | * entries to the tail of this list, and only ->poll() | |
4311 | * calls can remove this head entry from the list. | |
4312 | */ | |
e326bed2 | 4313 | n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list); |
1da177e4 | 4314 | |
bea3348e SH |
4315 | have = netpoll_poll_lock(n); |
4316 | ||
4317 | weight = n->weight; | |
4318 | ||
0a7606c1 DM |
4319 | /* This NAPI_STATE_SCHED test is for avoiding a race |
4320 | * with netpoll's poll_napi(). Only the entity which | |
4321 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
4322 | * actually make the ->poll() call. Therefore we avoid | |
25985edc | 4323 | * accidentally calling ->poll() when NAPI is not scheduled. |
0a7606c1 DM |
4324 | */ |
4325 | work = 0; | |
4ea7e386 | 4326 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { |
0a7606c1 | 4327 | work = n->poll(n, weight); |
4ea7e386 NH |
4328 | trace_napi_poll(n); |
4329 | } | |
bea3348e SH |
4330 | |
4331 | WARN_ON_ONCE(work > weight); | |
4332 | ||
4333 | budget -= work; | |
4334 | ||
4335 | local_irq_disable(); | |
4336 | ||
4337 | /* Drivers must not modify the NAPI state if they | |
4338 | * consume the entire weight. In such cases this code | |
4339 | * still "owns" the NAPI instance and therefore can | |
4340 | * move the instance around on the list at-will. | |
4341 | */ | |
fed17f30 | 4342 | if (unlikely(work == weight)) { |
ff780cd8 HX |
4343 | if (unlikely(napi_disable_pending(n))) { |
4344 | local_irq_enable(); | |
4345 | napi_complete(n); | |
4346 | local_irq_disable(); | |
2e71a6f8 ED |
4347 | } else { |
4348 | if (n->gro_list) { | |
4349 | /* flush too old packets | |
4350 | * If HZ < 1000, flush all packets. | |
4351 | */ | |
4352 | local_irq_enable(); | |
4353 | napi_gro_flush(n, HZ >= 1000); | |
4354 | local_irq_disable(); | |
4355 | } | |
e326bed2 | 4356 | list_move_tail(&n->poll_list, &sd->poll_list); |
2e71a6f8 | 4357 | } |
fed17f30 | 4358 | } |
bea3348e SH |
4359 | |
4360 | netpoll_poll_unlock(have); | |
1da177e4 LT |
4361 | } |
4362 | out: | |
e326bed2 | 4363 | net_rps_action_and_irq_enable(sd); |
0a9627f2 | 4364 | |
db217334 CL |
4365 | #ifdef CONFIG_NET_DMA |
4366 | /* | |
4367 | * There may not be any more sk_buffs coming right now, so push | |
4368 | * any pending DMA copies to hardware | |
4369 | */ | |
2ba05622 | 4370 | dma_issue_pending_all(); |
db217334 | 4371 | #endif |
bea3348e | 4372 | |
1da177e4 LT |
4373 | return; |
4374 | ||
4375 | softnet_break: | |
dee42870 | 4376 | sd->time_squeeze++; |
1da177e4 LT |
4377 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); |
4378 | goto out; | |
4379 | } | |
4380 | ||
d1b19dff | 4381 | static gifconf_func_t *gifconf_list[NPROTO]; |
1da177e4 LT |
4382 | |
4383 | /** | |
4384 | * register_gifconf - register a SIOCGIF handler | |
4385 | * @family: Address family | |
4386 | * @gifconf: Function handler | |
4387 | * | |
4388 | * Register protocol dependent address dumping routines. The handler | |
4389 | * that is passed must not be freed or reused until it has been replaced | |
4390 | * by another handler. | |
4391 | */ | |
d1b19dff | 4392 | int register_gifconf(unsigned int family, gifconf_func_t *gifconf) |
1da177e4 LT |
4393 | { |
4394 | if (family >= NPROTO) | |
4395 | return -EINVAL; | |
4396 | gifconf_list[family] = gifconf; | |
4397 | return 0; | |
4398 | } | |
d1b19dff | 4399 | EXPORT_SYMBOL(register_gifconf); |
1da177e4 LT |
4400 | |
4401 | ||
4402 | /* | |
4403 | * Map an interface index to its name (SIOCGIFNAME) | |
4404 | */ | |
4405 | ||
4406 | /* | |
4407 | * We need this ioctl for efficient implementation of the | |
4408 | * if_indextoname() function required by the IPv6 API. Without | |
4409 | * it, we would have to search all the interfaces to find a | |
4410 | * match. --pb | |
4411 | */ | |
4412 | ||
881d966b | 4413 | static int dev_ifname(struct net *net, struct ifreq __user *arg) |
1da177e4 LT |
4414 | { |
4415 | struct net_device *dev; | |
4416 | struct ifreq ifr; | |
c91f6df2 | 4417 | unsigned seq; |
1da177e4 LT |
4418 | |
4419 | /* | |
4420 | * Fetch the caller's info block. | |
4421 | */ | |
4422 | ||
4423 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
4424 | return -EFAULT; | |
4425 | ||
c91f6df2 | 4426 | retry: |
30e6c9fa | 4427 | seq = read_seqcount_begin(&devnet_rename_seq); |
fb699dfd ED |
4428 | rcu_read_lock(); |
4429 | dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex); | |
1da177e4 | 4430 | if (!dev) { |
fb699dfd | 4431 | rcu_read_unlock(); |
1da177e4 LT |
4432 | return -ENODEV; |
4433 | } | |
4434 | ||
4435 | strcpy(ifr.ifr_name, dev->name); | |
fb699dfd | 4436 | rcu_read_unlock(); |
30e6c9fa | 4437 | if (read_seqcount_retry(&devnet_rename_seq, seq)) |
c91f6df2 | 4438 | goto retry; |
1da177e4 LT |
4439 | |
4440 | if (copy_to_user(arg, &ifr, sizeof(struct ifreq))) | |
4441 | return -EFAULT; | |
4442 | return 0; | |
4443 | } | |
4444 | ||
4445 | /* | |
4446 | * Perform a SIOCGIFCONF call. This structure will change | |
4447 | * size eventually, and there is nothing I can do about it. | |
4448 | * Thus we will need a 'compatibility mode'. | |
4449 | */ | |
4450 | ||
881d966b | 4451 | static int dev_ifconf(struct net *net, char __user *arg) |
1da177e4 LT |
4452 | { |
4453 | struct ifconf ifc; | |
4454 | struct net_device *dev; | |
4455 | char __user *pos; | |
4456 | int len; | |
4457 | int total; | |
4458 | int i; | |
4459 | ||
4460 | /* | |
4461 | * Fetch the caller's info block. | |
4462 | */ | |
4463 | ||
4464 | if (copy_from_user(&ifc, arg, sizeof(struct ifconf))) | |
4465 | return -EFAULT; | |
4466 | ||
4467 | pos = ifc.ifc_buf; | |
4468 | len = ifc.ifc_len; | |
4469 | ||
4470 | /* | |
4471 | * Loop over the interfaces, and write an info block for each. | |
4472 | */ | |
4473 | ||
4474 | total = 0; | |
881d966b | 4475 | for_each_netdev(net, dev) { |
1da177e4 LT |
4476 | for (i = 0; i < NPROTO; i++) { |
4477 | if (gifconf_list[i]) { | |
4478 | int done; | |
4479 | if (!pos) | |
4480 | done = gifconf_list[i](dev, NULL, 0); | |
4481 | else | |
4482 | done = gifconf_list[i](dev, pos + total, | |
4483 | len - total); | |
4484 | if (done < 0) | |
4485 | return -EFAULT; | |
4486 | total += done; | |
4487 | } | |
4488 | } | |
4ec93edb | 4489 | } |
1da177e4 LT |
4490 | |
4491 | /* | |
4492 | * All done. Write the updated control block back to the caller. | |
4493 | */ | |
4494 | ifc.ifc_len = total; | |
4495 | ||
4496 | /* | |
4497 | * Both BSD and Solaris return 0 here, so we do too. | |
4498 | */ | |
4499 | return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0; | |
4500 | } | |
4501 | ||
4502 | #ifdef CONFIG_PROC_FS | |
f04565dd | 4503 | |
2def16ae | 4504 | #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1) |
f04565dd MM |
4505 | |
4506 | #define get_bucket(x) ((x) >> BUCKET_SPACE) | |
4507 | #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1)) | |
4508 | #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o)) | |
4509 | ||
2def16ae | 4510 | static inline struct net_device *dev_from_same_bucket(struct seq_file *seq, loff_t *pos) |
f04565dd | 4511 | { |
f04565dd MM |
4512 | struct net *net = seq_file_net(seq); |
4513 | struct net_device *dev; | |
4514 | struct hlist_node *p; | |
4515 | struct hlist_head *h; | |
2def16ae | 4516 | unsigned int count = 0, offset = get_offset(*pos); |
f04565dd | 4517 | |
2def16ae | 4518 | h = &net->dev_name_head[get_bucket(*pos)]; |
f04565dd | 4519 | hlist_for_each_entry_rcu(dev, p, h, name_hlist) { |
2def16ae | 4520 | if (++count == offset) |
f04565dd | 4521 | return dev; |
f04565dd MM |
4522 | } |
4523 | ||
4524 | return NULL; | |
4525 | } | |
4526 | ||
2def16ae | 4527 | static inline struct net_device *dev_from_bucket(struct seq_file *seq, loff_t *pos) |
f04565dd | 4528 | { |
f04565dd MM |
4529 | struct net_device *dev; |
4530 | unsigned int bucket; | |
4531 | ||
f04565dd | 4532 | do { |
2def16ae | 4533 | dev = dev_from_same_bucket(seq, pos); |
f04565dd MM |
4534 | if (dev) |
4535 | return dev; | |
4536 | ||
2def16ae ED |
4537 | bucket = get_bucket(*pos) + 1; |
4538 | *pos = set_bucket_offset(bucket, 1); | |
f04565dd MM |
4539 | } while (bucket < NETDEV_HASHENTRIES); |
4540 | ||
4541 | return NULL; | |
4542 | } | |
4543 | ||
1da177e4 LT |
4544 | /* |
4545 | * This is invoked by the /proc filesystem handler to display a device | |
4546 | * in detail. | |
4547 | */ | |
7562f876 | 4548 | void *dev_seq_start(struct seq_file *seq, loff_t *pos) |
c6d14c84 | 4549 | __acquires(RCU) |
1da177e4 | 4550 | { |
c6d14c84 | 4551 | rcu_read_lock(); |
7562f876 PE |
4552 | if (!*pos) |
4553 | return SEQ_START_TOKEN; | |
1da177e4 | 4554 | |
2def16ae | 4555 | if (get_bucket(*pos) >= NETDEV_HASHENTRIES) |
f04565dd | 4556 | return NULL; |
1da177e4 | 4557 | |
2def16ae | 4558 | return dev_from_bucket(seq, pos); |
1da177e4 LT |
4559 | } |
4560 | ||
4561 | void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
4562 | { | |
f04565dd | 4563 | ++*pos; |
2def16ae | 4564 | return dev_from_bucket(seq, pos); |
1da177e4 LT |
4565 | } |
4566 | ||
4567 | void dev_seq_stop(struct seq_file *seq, void *v) | |
c6d14c84 | 4568 | __releases(RCU) |
1da177e4 | 4569 | { |
c6d14c84 | 4570 | rcu_read_unlock(); |
1da177e4 LT |
4571 | } |
4572 | ||
4573 | static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev) | |
4574 | { | |
28172739 ED |
4575 | struct rtnl_link_stats64 temp; |
4576 | const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp); | |
1da177e4 | 4577 | |
be1f3c2c BH |
4578 | seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu " |
4579 | "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n", | |
5a1b5898 RR |
4580 | dev->name, stats->rx_bytes, stats->rx_packets, |
4581 | stats->rx_errors, | |
4582 | stats->rx_dropped + stats->rx_missed_errors, | |
4583 | stats->rx_fifo_errors, | |
4584 | stats->rx_length_errors + stats->rx_over_errors + | |
4585 | stats->rx_crc_errors + stats->rx_frame_errors, | |
4586 | stats->rx_compressed, stats->multicast, | |
4587 | stats->tx_bytes, stats->tx_packets, | |
4588 | stats->tx_errors, stats->tx_dropped, | |
4589 | stats->tx_fifo_errors, stats->collisions, | |
4590 | stats->tx_carrier_errors + | |
4591 | stats->tx_aborted_errors + | |
4592 | stats->tx_window_errors + | |
4593 | stats->tx_heartbeat_errors, | |
4594 | stats->tx_compressed); | |
1da177e4 LT |
4595 | } |
4596 | ||
4597 | /* | |
4598 | * Called from the PROCfs module. This now uses the new arbitrary sized | |
4599 | * /proc/net interface to create /proc/net/dev | |
4600 | */ | |
4601 | static int dev_seq_show(struct seq_file *seq, void *v) | |
4602 | { | |
4603 | if (v == SEQ_START_TOKEN) | |
4604 | seq_puts(seq, "Inter-| Receive " | |
4605 | " | Transmit\n" | |
4606 | " face |bytes packets errs drop fifo frame " | |
4607 | "compressed multicast|bytes packets errs " | |
4608 | "drop fifo colls carrier compressed\n"); | |
4609 | else | |
4610 | dev_seq_printf_stats(seq, v); | |
4611 | return 0; | |
4612 | } | |
4613 | ||
dee42870 | 4614 | static struct softnet_data *softnet_get_online(loff_t *pos) |
1da177e4 | 4615 | { |
dee42870 | 4616 | struct softnet_data *sd = NULL; |
1da177e4 | 4617 | |
0c0b0aca | 4618 | while (*pos < nr_cpu_ids) |
4ec93edb | 4619 | if (cpu_online(*pos)) { |
dee42870 | 4620 | sd = &per_cpu(softnet_data, *pos); |
1da177e4 LT |
4621 | break; |
4622 | } else | |
4623 | ++*pos; | |
dee42870 | 4624 | return sd; |
1da177e4 LT |
4625 | } |
4626 | ||
4627 | static void *softnet_seq_start(struct seq_file *seq, loff_t *pos) | |
4628 | { | |
4629 | return softnet_get_online(pos); | |
4630 | } | |
4631 | ||
4632 | static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
4633 | { | |
4634 | ++*pos; | |
4635 | return softnet_get_online(pos); | |
4636 | } | |
4637 | ||
4638 | static void softnet_seq_stop(struct seq_file *seq, void *v) | |
4639 | { | |
4640 | } | |
4641 | ||
4642 | static int softnet_seq_show(struct seq_file *seq, void *v) | |
4643 | { | |
dee42870 | 4644 | struct softnet_data *sd = v; |
1da177e4 | 4645 | |
0a9627f2 | 4646 | seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n", |
dee42870 | 4647 | sd->processed, sd->dropped, sd->time_squeeze, 0, |
c1ebcdb8 | 4648 | 0, 0, 0, 0, /* was fastroute */ |
dee42870 | 4649 | sd->cpu_collision, sd->received_rps); |
1da177e4 LT |
4650 | return 0; |
4651 | } | |
4652 | ||
f690808e | 4653 | static const struct seq_operations dev_seq_ops = { |
1da177e4 LT |
4654 | .start = dev_seq_start, |
4655 | .next = dev_seq_next, | |
4656 | .stop = dev_seq_stop, | |
4657 | .show = dev_seq_show, | |
4658 | }; | |
4659 | ||
4660 | static int dev_seq_open(struct inode *inode, struct file *file) | |
4661 | { | |
e372c414 | 4662 | return seq_open_net(inode, file, &dev_seq_ops, |
2def16ae | 4663 | sizeof(struct seq_net_private)); |
5cac98dd AB |
4664 | } |
4665 | ||
9a32144e | 4666 | static const struct file_operations dev_seq_fops = { |
1da177e4 LT |
4667 | .owner = THIS_MODULE, |
4668 | .open = dev_seq_open, | |
4669 | .read = seq_read, | |
4670 | .llseek = seq_lseek, | |
e372c414 | 4671 | .release = seq_release_net, |
1da177e4 LT |
4672 | }; |
4673 | ||
f690808e | 4674 | static const struct seq_operations softnet_seq_ops = { |
1da177e4 LT |
4675 | .start = softnet_seq_start, |
4676 | .next = softnet_seq_next, | |
4677 | .stop = softnet_seq_stop, | |
4678 | .show = softnet_seq_show, | |
4679 | }; | |
4680 | ||
4681 | static int softnet_seq_open(struct inode *inode, struct file *file) | |
4682 | { | |
4683 | return seq_open(file, &softnet_seq_ops); | |
4684 | } | |
4685 | ||
9a32144e | 4686 | static const struct file_operations softnet_seq_fops = { |
1da177e4 LT |
4687 | .owner = THIS_MODULE, |
4688 | .open = softnet_seq_open, | |
4689 | .read = seq_read, | |
4690 | .llseek = seq_lseek, | |
4691 | .release = seq_release, | |
4692 | }; | |
4693 | ||
0e1256ff SH |
4694 | static void *ptype_get_idx(loff_t pos) |
4695 | { | |
4696 | struct packet_type *pt = NULL; | |
4697 | loff_t i = 0; | |
4698 | int t; | |
4699 | ||
4700 | list_for_each_entry_rcu(pt, &ptype_all, list) { | |
4701 | if (i == pos) | |
4702 | return pt; | |
4703 | ++i; | |
4704 | } | |
4705 | ||
82d8a867 | 4706 | for (t = 0; t < PTYPE_HASH_SIZE; t++) { |
0e1256ff SH |
4707 | list_for_each_entry_rcu(pt, &ptype_base[t], list) { |
4708 | if (i == pos) | |
4709 | return pt; | |
4710 | ++i; | |
4711 | } | |
4712 | } | |
4713 | return NULL; | |
4714 | } | |
4715 | ||
4716 | static void *ptype_seq_start(struct seq_file *seq, loff_t *pos) | |
72348a42 | 4717 | __acquires(RCU) |
0e1256ff SH |
4718 | { |
4719 | rcu_read_lock(); | |
4720 | return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN; | |
4721 | } | |
4722 | ||
4723 | static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
4724 | { | |
4725 | struct packet_type *pt; | |
4726 | struct list_head *nxt; | |
4727 | int hash; | |
4728 | ||
4729 | ++*pos; | |
4730 | if (v == SEQ_START_TOKEN) | |
4731 | return ptype_get_idx(0); | |
4732 | ||
4733 | pt = v; | |
4734 | nxt = pt->list.next; | |
4735 | if (pt->type == htons(ETH_P_ALL)) { | |
4736 | if (nxt != &ptype_all) | |
4737 | goto found; | |
4738 | hash = 0; | |
4739 | nxt = ptype_base[0].next; | |
4740 | } else | |
82d8a867 | 4741 | hash = ntohs(pt->type) & PTYPE_HASH_MASK; |
0e1256ff SH |
4742 | |
4743 | while (nxt == &ptype_base[hash]) { | |
82d8a867 | 4744 | if (++hash >= PTYPE_HASH_SIZE) |
0e1256ff SH |
4745 | return NULL; |
4746 | nxt = ptype_base[hash].next; | |
4747 | } | |
4748 | found: | |
4749 | return list_entry(nxt, struct packet_type, list); | |
4750 | } | |
4751 | ||
4752 | static void ptype_seq_stop(struct seq_file *seq, void *v) | |
72348a42 | 4753 | __releases(RCU) |
0e1256ff SH |
4754 | { |
4755 | rcu_read_unlock(); | |
4756 | } | |
4757 | ||
0e1256ff SH |
4758 | static int ptype_seq_show(struct seq_file *seq, void *v) |
4759 | { | |
4760 | struct packet_type *pt = v; | |
4761 | ||
4762 | if (v == SEQ_START_TOKEN) | |
4763 | seq_puts(seq, "Type Device Function\n"); | |
c346dca1 | 4764 | else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) { |
0e1256ff SH |
4765 | if (pt->type == htons(ETH_P_ALL)) |
4766 | seq_puts(seq, "ALL "); | |
4767 | else | |
4768 | seq_printf(seq, "%04x", ntohs(pt->type)); | |
4769 | ||
908cd2da AD |
4770 | seq_printf(seq, " %-8s %pF\n", |
4771 | pt->dev ? pt->dev->name : "", pt->func); | |
0e1256ff SH |
4772 | } |
4773 | ||
4774 | return 0; | |
4775 | } | |
4776 | ||
4777 | static const struct seq_operations ptype_seq_ops = { | |
4778 | .start = ptype_seq_start, | |
4779 | .next = ptype_seq_next, | |
4780 | .stop = ptype_seq_stop, | |
4781 | .show = ptype_seq_show, | |
4782 | }; | |
4783 | ||
4784 | static int ptype_seq_open(struct inode *inode, struct file *file) | |
4785 | { | |
2feb27db PE |
4786 | return seq_open_net(inode, file, &ptype_seq_ops, |
4787 | sizeof(struct seq_net_private)); | |
0e1256ff SH |
4788 | } |
4789 | ||
4790 | static const struct file_operations ptype_seq_fops = { | |
4791 | .owner = THIS_MODULE, | |
4792 | .open = ptype_seq_open, | |
4793 | .read = seq_read, | |
4794 | .llseek = seq_lseek, | |
2feb27db | 4795 | .release = seq_release_net, |
0e1256ff SH |
4796 | }; |
4797 | ||
4798 | ||
4665079c | 4799 | static int __net_init dev_proc_net_init(struct net *net) |
1da177e4 LT |
4800 | { |
4801 | int rc = -ENOMEM; | |
4802 | ||
881d966b | 4803 | if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops)) |
1da177e4 | 4804 | goto out; |
881d966b | 4805 | if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops)) |
1da177e4 | 4806 | goto out_dev; |
881d966b | 4807 | if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops)) |
457c4cbc | 4808 | goto out_softnet; |
0e1256ff | 4809 | |
881d966b | 4810 | if (wext_proc_init(net)) |
457c4cbc | 4811 | goto out_ptype; |
1da177e4 LT |
4812 | rc = 0; |
4813 | out: | |
4814 | return rc; | |
457c4cbc | 4815 | out_ptype: |
881d966b | 4816 | proc_net_remove(net, "ptype"); |
1da177e4 | 4817 | out_softnet: |
881d966b | 4818 | proc_net_remove(net, "softnet_stat"); |
1da177e4 | 4819 | out_dev: |
881d966b | 4820 | proc_net_remove(net, "dev"); |
1da177e4 LT |
4821 | goto out; |
4822 | } | |
881d966b | 4823 | |
4665079c | 4824 | static void __net_exit dev_proc_net_exit(struct net *net) |
881d966b EB |
4825 | { |
4826 | wext_proc_exit(net); | |
4827 | ||
4828 | proc_net_remove(net, "ptype"); | |
4829 | proc_net_remove(net, "softnet_stat"); | |
4830 | proc_net_remove(net, "dev"); | |
4831 | } | |
4832 | ||
022cbae6 | 4833 | static struct pernet_operations __net_initdata dev_proc_ops = { |
881d966b EB |
4834 | .init = dev_proc_net_init, |
4835 | .exit = dev_proc_net_exit, | |
4836 | }; | |
4837 | ||
4838 | static int __init dev_proc_init(void) | |
4839 | { | |
4840 | return register_pernet_subsys(&dev_proc_ops); | |
4841 | } | |
1da177e4 LT |
4842 | #else |
4843 | #define dev_proc_init() 0 | |
4844 | #endif /* CONFIG_PROC_FS */ | |
4845 | ||
4846 | ||
9ff162a8 JP |
4847 | struct netdev_upper { |
4848 | struct net_device *dev; | |
4849 | bool master; | |
4850 | struct list_head list; | |
4851 | struct rcu_head rcu; | |
4852 | struct list_head search_list; | |
4853 | }; | |
4854 | ||
4855 | static void __append_search_uppers(struct list_head *search_list, | |
4856 | struct net_device *dev) | |
4857 | { | |
4858 | struct netdev_upper *upper; | |
4859 | ||
4860 | list_for_each_entry(upper, &dev->upper_dev_list, list) { | |
4861 | /* check if this upper is not already in search list */ | |
4862 | if (list_empty(&upper->search_list)) | |
4863 | list_add_tail(&upper->search_list, search_list); | |
4864 | } | |
4865 | } | |
4866 | ||
4867 | static bool __netdev_search_upper_dev(struct net_device *dev, | |
4868 | struct net_device *upper_dev) | |
4869 | { | |
4870 | LIST_HEAD(search_list); | |
4871 | struct netdev_upper *upper; | |
4872 | struct netdev_upper *tmp; | |
4873 | bool ret = false; | |
4874 | ||
4875 | __append_search_uppers(&search_list, dev); | |
4876 | list_for_each_entry(upper, &search_list, search_list) { | |
4877 | if (upper->dev == upper_dev) { | |
4878 | ret = true; | |
4879 | break; | |
4880 | } | |
4881 | __append_search_uppers(&search_list, upper->dev); | |
4882 | } | |
4883 | list_for_each_entry_safe(upper, tmp, &search_list, search_list) | |
4884 | INIT_LIST_HEAD(&upper->search_list); | |
4885 | return ret; | |
4886 | } | |
4887 | ||
4888 | static struct netdev_upper *__netdev_find_upper(struct net_device *dev, | |
4889 | struct net_device *upper_dev) | |
4890 | { | |
4891 | struct netdev_upper *upper; | |
4892 | ||
4893 | list_for_each_entry(upper, &dev->upper_dev_list, list) { | |
4894 | if (upper->dev == upper_dev) | |
4895 | return upper; | |
4896 | } | |
4897 | return NULL; | |
4898 | } | |
4899 | ||
4900 | /** | |
4901 | * netdev_has_upper_dev - Check if device is linked to an upper device | |
4902 | * @dev: device | |
4903 | * @upper_dev: upper device to check | |
4904 | * | |
4905 | * Find out if a device is linked to specified upper device and return true | |
4906 | * in case it is. Note that this checks only immediate upper device, | |
4907 | * not through a complete stack of devices. The caller must hold the RTNL lock. | |
4908 | */ | |
4909 | bool netdev_has_upper_dev(struct net_device *dev, | |
4910 | struct net_device *upper_dev) | |
4911 | { | |
4912 | ASSERT_RTNL(); | |
4913 | ||
4914 | return __netdev_find_upper(dev, upper_dev); | |
4915 | } | |
4916 | EXPORT_SYMBOL(netdev_has_upper_dev); | |
4917 | ||
4918 | /** | |
4919 | * netdev_has_any_upper_dev - Check if device is linked to some device | |
4920 | * @dev: device | |
4921 | * | |
4922 | * Find out if a device is linked to an upper device and return true in case | |
4923 | * it is. The caller must hold the RTNL lock. | |
4924 | */ | |
4925 | bool netdev_has_any_upper_dev(struct net_device *dev) | |
4926 | { | |
4927 | ASSERT_RTNL(); | |
4928 | ||
4929 | return !list_empty(&dev->upper_dev_list); | |
4930 | } | |
4931 | EXPORT_SYMBOL(netdev_has_any_upper_dev); | |
4932 | ||
4933 | /** | |
4934 | * netdev_master_upper_dev_get - Get master upper device | |
4935 | * @dev: device | |
4936 | * | |
4937 | * Find a master upper device and return pointer to it or NULL in case | |
4938 | * it's not there. The caller must hold the RTNL lock. | |
4939 | */ | |
4940 | struct net_device *netdev_master_upper_dev_get(struct net_device *dev) | |
4941 | { | |
4942 | struct netdev_upper *upper; | |
4943 | ||
4944 | ASSERT_RTNL(); | |
4945 | ||
4946 | if (list_empty(&dev->upper_dev_list)) | |
4947 | return NULL; | |
4948 | ||
4949 | upper = list_first_entry(&dev->upper_dev_list, | |
4950 | struct netdev_upper, list); | |
4951 | if (likely(upper->master)) | |
4952 | return upper->dev; | |
4953 | return NULL; | |
4954 | } | |
4955 | EXPORT_SYMBOL(netdev_master_upper_dev_get); | |
4956 | ||
4957 | /** | |
4958 | * netdev_master_upper_dev_get_rcu - Get master upper device | |
4959 | * @dev: device | |
4960 | * | |
4961 | * Find a master upper device and return pointer to it or NULL in case | |
4962 | * it's not there. The caller must hold the RCU read lock. | |
4963 | */ | |
4964 | struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) | |
4965 | { | |
4966 | struct netdev_upper *upper; | |
4967 | ||
4968 | upper = list_first_or_null_rcu(&dev->upper_dev_list, | |
4969 | struct netdev_upper, list); | |
4970 | if (upper && likely(upper->master)) | |
4971 | return upper->dev; | |
4972 | return NULL; | |
4973 | } | |
4974 | EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); | |
4975 | ||
4976 | static int __netdev_upper_dev_link(struct net_device *dev, | |
4977 | struct net_device *upper_dev, bool master) | |
4978 | { | |
4979 | struct netdev_upper *upper; | |
4980 | ||
4981 | ASSERT_RTNL(); | |
4982 | ||
4983 | if (dev == upper_dev) | |
4984 | return -EBUSY; | |
4985 | ||
4986 | /* To prevent loops, check if dev is not upper device to upper_dev. */ | |
4987 | if (__netdev_search_upper_dev(upper_dev, dev)) | |
4988 | return -EBUSY; | |
4989 | ||
4990 | if (__netdev_find_upper(dev, upper_dev)) | |
4991 | return -EEXIST; | |
4992 | ||
4993 | if (master && netdev_master_upper_dev_get(dev)) | |
4994 | return -EBUSY; | |
4995 | ||
4996 | upper = kmalloc(sizeof(*upper), GFP_KERNEL); | |
4997 | if (!upper) | |
4998 | return -ENOMEM; | |
4999 | ||
5000 | upper->dev = upper_dev; | |
5001 | upper->master = master; | |
5002 | INIT_LIST_HEAD(&upper->search_list); | |
5003 | ||
5004 | /* Ensure that master upper link is always the first item in list. */ | |
5005 | if (master) | |
5006 | list_add_rcu(&upper->list, &dev->upper_dev_list); | |
5007 | else | |
5008 | list_add_tail_rcu(&upper->list, &dev->upper_dev_list); | |
5009 | dev_hold(upper_dev); | |
5010 | ||
5011 | return 0; | |
5012 | } | |
5013 | ||
5014 | /** | |
5015 | * netdev_upper_dev_link - Add a link to the upper device | |
5016 | * @dev: device | |
5017 | * @upper_dev: new upper device | |
5018 | * | |
5019 | * Adds a link to device which is upper to this one. The caller must hold | |
5020 | * the RTNL lock. On a failure a negative errno code is returned. | |
5021 | * On success the reference counts are adjusted and the function | |
5022 | * returns zero. | |
5023 | */ | |
5024 | int netdev_upper_dev_link(struct net_device *dev, | |
5025 | struct net_device *upper_dev) | |
5026 | { | |
5027 | return __netdev_upper_dev_link(dev, upper_dev, false); | |
5028 | } | |
5029 | EXPORT_SYMBOL(netdev_upper_dev_link); | |
5030 | ||
5031 | /** | |
5032 | * netdev_master_upper_dev_link - Add a master link to the upper device | |
5033 | * @dev: device | |
5034 | * @upper_dev: new upper device | |
5035 | * | |
5036 | * Adds a link to device which is upper to this one. In this case, only | |
5037 | * one master upper device can be linked, although other non-master devices | |
5038 | * might be linked as well. The caller must hold the RTNL lock. | |
5039 | * On a failure a negative errno code is returned. On success the reference | |
5040 | * counts are adjusted and the function returns zero. | |
5041 | */ | |
5042 | int netdev_master_upper_dev_link(struct net_device *dev, | |
5043 | struct net_device *upper_dev) | |
5044 | { | |
5045 | return __netdev_upper_dev_link(dev, upper_dev, true); | |
5046 | } | |
5047 | EXPORT_SYMBOL(netdev_master_upper_dev_link); | |
5048 | ||
5049 | /** | |
5050 | * netdev_upper_dev_unlink - Removes a link to upper device | |
5051 | * @dev: device | |
5052 | * @upper_dev: new upper device | |
5053 | * | |
5054 | * Removes a link to device which is upper to this one. The caller must hold | |
5055 | * the RTNL lock. | |
5056 | */ | |
5057 | void netdev_upper_dev_unlink(struct net_device *dev, | |
5058 | struct net_device *upper_dev) | |
5059 | { | |
5060 | struct netdev_upper *upper; | |
5061 | ||
5062 | ASSERT_RTNL(); | |
5063 | ||
5064 | upper = __netdev_find_upper(dev, upper_dev); | |
5065 | if (!upper) | |
5066 | return; | |
5067 | list_del_rcu(&upper->list); | |
5068 | dev_put(upper_dev); | |
5069 | kfree_rcu(upper, rcu); | |
5070 | } | |
5071 | EXPORT_SYMBOL(netdev_upper_dev_unlink); | |
5072 | ||
b6c40d68 PM |
5073 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
5074 | { | |
d314774c SH |
5075 | const struct net_device_ops *ops = dev->netdev_ops; |
5076 | ||
5077 | if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags) | |
5078 | ops->ndo_change_rx_flags(dev, flags); | |
b6c40d68 PM |
5079 | } |
5080 | ||
dad9b335 | 5081 | static int __dev_set_promiscuity(struct net_device *dev, int inc) |
1da177e4 | 5082 | { |
b536db93 | 5083 | unsigned int old_flags = dev->flags; |
d04a48b0 EB |
5084 | kuid_t uid; |
5085 | kgid_t gid; | |
1da177e4 | 5086 | |
24023451 PM |
5087 | ASSERT_RTNL(); |
5088 | ||
dad9b335 WC |
5089 | dev->flags |= IFF_PROMISC; |
5090 | dev->promiscuity += inc; | |
5091 | if (dev->promiscuity == 0) { | |
5092 | /* | |
5093 | * Avoid overflow. | |
5094 | * If inc causes overflow, untouch promisc and return error. | |
5095 | */ | |
5096 | if (inc < 0) | |
5097 | dev->flags &= ~IFF_PROMISC; | |
5098 | else { | |
5099 | dev->promiscuity -= inc; | |
7b6cd1ce JP |
5100 | pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", |
5101 | dev->name); | |
dad9b335 WC |
5102 | return -EOVERFLOW; |
5103 | } | |
5104 | } | |
52609c0b | 5105 | if (dev->flags != old_flags) { |
7b6cd1ce JP |
5106 | pr_info("device %s %s promiscuous mode\n", |
5107 | dev->name, | |
5108 | dev->flags & IFF_PROMISC ? "entered" : "left"); | |
8192b0c4 DH |
5109 | if (audit_enabled) { |
5110 | current_uid_gid(&uid, &gid); | |
7759db82 KHK |
5111 | audit_log(current->audit_context, GFP_ATOMIC, |
5112 | AUDIT_ANOM_PROMISCUOUS, | |
5113 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
5114 | dev->name, (dev->flags & IFF_PROMISC), | |
5115 | (old_flags & IFF_PROMISC), | |
e1760bd5 | 5116 | from_kuid(&init_user_ns, audit_get_loginuid(current)), |
d04a48b0 EB |
5117 | from_kuid(&init_user_ns, uid), |
5118 | from_kgid(&init_user_ns, gid), | |
7759db82 | 5119 | audit_get_sessionid(current)); |
8192b0c4 | 5120 | } |
24023451 | 5121 | |
b6c40d68 | 5122 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 5123 | } |
dad9b335 | 5124 | return 0; |
1da177e4 LT |
5125 | } |
5126 | ||
4417da66 PM |
5127 | /** |
5128 | * dev_set_promiscuity - update promiscuity count on a device | |
5129 | * @dev: device | |
5130 | * @inc: modifier | |
5131 | * | |
5132 | * Add or remove promiscuity from a device. While the count in the device | |
5133 | * remains above zero the interface remains promiscuous. Once it hits zero | |
5134 | * the device reverts back to normal filtering operation. A negative inc | |
5135 | * value is used to drop promiscuity on the device. | |
dad9b335 | 5136 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 5137 | */ |
dad9b335 | 5138 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 | 5139 | { |
b536db93 | 5140 | unsigned int old_flags = dev->flags; |
dad9b335 | 5141 | int err; |
4417da66 | 5142 | |
dad9b335 | 5143 | err = __dev_set_promiscuity(dev, inc); |
4b5a698e | 5144 | if (err < 0) |
dad9b335 | 5145 | return err; |
4417da66 PM |
5146 | if (dev->flags != old_flags) |
5147 | dev_set_rx_mode(dev); | |
dad9b335 | 5148 | return err; |
4417da66 | 5149 | } |
d1b19dff | 5150 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 5151 | |
1da177e4 LT |
5152 | /** |
5153 | * dev_set_allmulti - update allmulti count on a device | |
5154 | * @dev: device | |
5155 | * @inc: modifier | |
5156 | * | |
5157 | * Add or remove reception of all multicast frames to a device. While the | |
5158 | * count in the device remains above zero the interface remains listening | |
5159 | * to all interfaces. Once it hits zero the device reverts back to normal | |
5160 | * filtering operation. A negative @inc value is used to drop the counter | |
5161 | * when releasing a resource needing all multicasts. | |
dad9b335 | 5162 | * Return 0 if successful or a negative errno code on error. |
1da177e4 LT |
5163 | */ |
5164 | ||
dad9b335 | 5165 | int dev_set_allmulti(struct net_device *dev, int inc) |
1da177e4 | 5166 | { |
b536db93 | 5167 | unsigned int old_flags = dev->flags; |
1da177e4 | 5168 | |
24023451 PM |
5169 | ASSERT_RTNL(); |
5170 | ||
1da177e4 | 5171 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
5172 | dev->allmulti += inc; |
5173 | if (dev->allmulti == 0) { | |
5174 | /* | |
5175 | * Avoid overflow. | |
5176 | * If inc causes overflow, untouch allmulti and return error. | |
5177 | */ | |
5178 | if (inc < 0) | |
5179 | dev->flags &= ~IFF_ALLMULTI; | |
5180 | else { | |
5181 | dev->allmulti -= inc; | |
7b6cd1ce JP |
5182 | pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", |
5183 | dev->name); | |
dad9b335 WC |
5184 | return -EOVERFLOW; |
5185 | } | |
5186 | } | |
24023451 | 5187 | if (dev->flags ^ old_flags) { |
b6c40d68 | 5188 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 5189 | dev_set_rx_mode(dev); |
24023451 | 5190 | } |
dad9b335 | 5191 | return 0; |
4417da66 | 5192 | } |
d1b19dff | 5193 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
5194 | |
5195 | /* | |
5196 | * Upload unicast and multicast address lists to device and | |
5197 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 5198 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
5199 | * are present. |
5200 | */ | |
5201 | void __dev_set_rx_mode(struct net_device *dev) | |
5202 | { | |
d314774c SH |
5203 | const struct net_device_ops *ops = dev->netdev_ops; |
5204 | ||
4417da66 PM |
5205 | /* dev_open will call this function so the list will stay sane. */ |
5206 | if (!(dev->flags&IFF_UP)) | |
5207 | return; | |
5208 | ||
5209 | if (!netif_device_present(dev)) | |
40b77c94 | 5210 | return; |
4417da66 | 5211 | |
01789349 | 5212 | if (!(dev->priv_flags & IFF_UNICAST_FLT)) { |
4417da66 PM |
5213 | /* Unicast addresses changes may only happen under the rtnl, |
5214 | * therefore calling __dev_set_promiscuity here is safe. | |
5215 | */ | |
32e7bfc4 | 5216 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
4417da66 | 5217 | __dev_set_promiscuity(dev, 1); |
2d348d1f | 5218 | dev->uc_promisc = true; |
32e7bfc4 | 5219 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
4417da66 | 5220 | __dev_set_promiscuity(dev, -1); |
2d348d1f | 5221 | dev->uc_promisc = false; |
4417da66 | 5222 | } |
4417da66 | 5223 | } |
01789349 JP |
5224 | |
5225 | if (ops->ndo_set_rx_mode) | |
5226 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
5227 | } |
5228 | ||
5229 | void dev_set_rx_mode(struct net_device *dev) | |
5230 | { | |
b9e40857 | 5231 | netif_addr_lock_bh(dev); |
4417da66 | 5232 | __dev_set_rx_mode(dev); |
b9e40857 | 5233 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
5234 | } |
5235 | ||
f0db275a SH |
5236 | /** |
5237 | * dev_get_flags - get flags reported to userspace | |
5238 | * @dev: device | |
5239 | * | |
5240 | * Get the combination of flag bits exported through APIs to userspace. | |
5241 | */ | |
95c96174 | 5242 | unsigned int dev_get_flags(const struct net_device *dev) |
1da177e4 | 5243 | { |
95c96174 | 5244 | unsigned int flags; |
1da177e4 LT |
5245 | |
5246 | flags = (dev->flags & ~(IFF_PROMISC | | |
5247 | IFF_ALLMULTI | | |
b00055aa SR |
5248 | IFF_RUNNING | |
5249 | IFF_LOWER_UP | | |
5250 | IFF_DORMANT)) | | |
1da177e4 LT |
5251 | (dev->gflags & (IFF_PROMISC | |
5252 | IFF_ALLMULTI)); | |
5253 | ||
b00055aa SR |
5254 | if (netif_running(dev)) { |
5255 | if (netif_oper_up(dev)) | |
5256 | flags |= IFF_RUNNING; | |
5257 | if (netif_carrier_ok(dev)) | |
5258 | flags |= IFF_LOWER_UP; | |
5259 | if (netif_dormant(dev)) | |
5260 | flags |= IFF_DORMANT; | |
5261 | } | |
1da177e4 LT |
5262 | |
5263 | return flags; | |
5264 | } | |
d1b19dff | 5265 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 5266 | |
bd380811 | 5267 | int __dev_change_flags(struct net_device *dev, unsigned int flags) |
1da177e4 | 5268 | { |
b536db93 | 5269 | unsigned int old_flags = dev->flags; |
bd380811 | 5270 | int ret; |
1da177e4 | 5271 | |
24023451 PM |
5272 | ASSERT_RTNL(); |
5273 | ||
1da177e4 LT |
5274 | /* |
5275 | * Set the flags on our device. | |
5276 | */ | |
5277 | ||
5278 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
5279 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
5280 | IFF_AUTOMEDIA)) | | |
5281 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
5282 | IFF_ALLMULTI)); | |
5283 | ||
5284 | /* | |
5285 | * Load in the correct multicast list now the flags have changed. | |
5286 | */ | |
5287 | ||
b6c40d68 PM |
5288 | if ((old_flags ^ flags) & IFF_MULTICAST) |
5289 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 5290 | |
4417da66 | 5291 | dev_set_rx_mode(dev); |
1da177e4 LT |
5292 | |
5293 | /* | |
5294 | * Have we downed the interface. We handle IFF_UP ourselves | |
5295 | * according to user attempts to set it, rather than blindly | |
5296 | * setting it. | |
5297 | */ | |
5298 | ||
5299 | ret = 0; | |
5300 | if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */ | |
bd380811 | 5301 | ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev); |
1da177e4 LT |
5302 | |
5303 | if (!ret) | |
4417da66 | 5304 | dev_set_rx_mode(dev); |
1da177e4 LT |
5305 | } |
5306 | ||
1da177e4 | 5307 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff ED |
5308 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
5309 | ||
1da177e4 LT |
5310 | dev->gflags ^= IFF_PROMISC; |
5311 | dev_set_promiscuity(dev, inc); | |
5312 | } | |
5313 | ||
5314 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
5315 | is important. Some (broken) drivers set IFF_PROMISC, when | |
5316 | IFF_ALLMULTI is requested not asking us and not reporting. | |
5317 | */ | |
5318 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
5319 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
5320 | ||
1da177e4 LT |
5321 | dev->gflags ^= IFF_ALLMULTI; |
5322 | dev_set_allmulti(dev, inc); | |
5323 | } | |
5324 | ||
bd380811 PM |
5325 | return ret; |
5326 | } | |
5327 | ||
5328 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags) | |
5329 | { | |
5330 | unsigned int changes = dev->flags ^ old_flags; | |
5331 | ||
5332 | if (changes & IFF_UP) { | |
5333 | if (dev->flags & IFF_UP) | |
5334 | call_netdevice_notifiers(NETDEV_UP, dev); | |
5335 | else | |
5336 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
5337 | } | |
5338 | ||
5339 | if (dev->flags & IFF_UP && | |
5340 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) | |
5341 | call_netdevice_notifiers(NETDEV_CHANGE, dev); | |
5342 | } | |
5343 | ||
5344 | /** | |
5345 | * dev_change_flags - change device settings | |
5346 | * @dev: device | |
5347 | * @flags: device state flags | |
5348 | * | |
5349 | * Change settings on device based state flags. The flags are | |
5350 | * in the userspace exported format. | |
5351 | */ | |
b536db93 | 5352 | int dev_change_flags(struct net_device *dev, unsigned int flags) |
bd380811 | 5353 | { |
b536db93 ED |
5354 | int ret; |
5355 | unsigned int changes, old_flags = dev->flags; | |
bd380811 PM |
5356 | |
5357 | ret = __dev_change_flags(dev, flags); | |
5358 | if (ret < 0) | |
5359 | return ret; | |
5360 | ||
5361 | changes = old_flags ^ dev->flags; | |
7c355f53 TG |
5362 | if (changes) |
5363 | rtmsg_ifinfo(RTM_NEWLINK, dev, changes); | |
1da177e4 | 5364 | |
bd380811 | 5365 | __dev_notify_flags(dev, old_flags); |
1da177e4 LT |
5366 | return ret; |
5367 | } | |
d1b19dff | 5368 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 5369 | |
f0db275a SH |
5370 | /** |
5371 | * dev_set_mtu - Change maximum transfer unit | |
5372 | * @dev: device | |
5373 | * @new_mtu: new transfer unit | |
5374 | * | |
5375 | * Change the maximum transfer size of the network device. | |
5376 | */ | |
1da177e4 LT |
5377 | int dev_set_mtu(struct net_device *dev, int new_mtu) |
5378 | { | |
d314774c | 5379 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
5380 | int err; |
5381 | ||
5382 | if (new_mtu == dev->mtu) | |
5383 | return 0; | |
5384 | ||
5385 | /* MTU must be positive. */ | |
5386 | if (new_mtu < 0) | |
5387 | return -EINVAL; | |
5388 | ||
5389 | if (!netif_device_present(dev)) | |
5390 | return -ENODEV; | |
5391 | ||
5392 | err = 0; | |
d314774c SH |
5393 | if (ops->ndo_change_mtu) |
5394 | err = ops->ndo_change_mtu(dev, new_mtu); | |
1da177e4 LT |
5395 | else |
5396 | dev->mtu = new_mtu; | |
d314774c | 5397 | |
e3d8fabe | 5398 | if (!err) |
056925ab | 5399 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); |
1da177e4 LT |
5400 | return err; |
5401 | } | |
d1b19dff | 5402 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 5403 | |
cbda10fa VD |
5404 | /** |
5405 | * dev_set_group - Change group this device belongs to | |
5406 | * @dev: device | |
5407 | * @new_group: group this device should belong to | |
5408 | */ | |
5409 | void dev_set_group(struct net_device *dev, int new_group) | |
5410 | { | |
5411 | dev->group = new_group; | |
5412 | } | |
5413 | EXPORT_SYMBOL(dev_set_group); | |
5414 | ||
f0db275a SH |
5415 | /** |
5416 | * dev_set_mac_address - Change Media Access Control Address | |
5417 | * @dev: device | |
5418 | * @sa: new address | |
5419 | * | |
5420 | * Change the hardware (MAC) address of the device | |
5421 | */ | |
1da177e4 LT |
5422 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) |
5423 | { | |
d314774c | 5424 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
5425 | int err; |
5426 | ||
d314774c | 5427 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
5428 | return -EOPNOTSUPP; |
5429 | if (sa->sa_family != dev->type) | |
5430 | return -EINVAL; | |
5431 | if (!netif_device_present(dev)) | |
5432 | return -ENODEV; | |
d314774c | 5433 | err = ops->ndo_set_mac_address(dev, sa); |
f6521516 JP |
5434 | if (err) |
5435 | return err; | |
fbdeca2d | 5436 | dev->addr_assign_type = NET_ADDR_SET; |
f6521516 | 5437 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
7bf23575 | 5438 | add_device_randomness(dev->dev_addr, dev->addr_len); |
f6521516 | 5439 | return 0; |
1da177e4 | 5440 | } |
d1b19dff | 5441 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 | 5442 | |
4bf84c35 JP |
5443 | /** |
5444 | * dev_change_carrier - Change device carrier | |
5445 | * @dev: device | |
5446 | * @new_carries: new value | |
5447 | * | |
5448 | * Change device carrier | |
5449 | */ | |
5450 | int dev_change_carrier(struct net_device *dev, bool new_carrier) | |
5451 | { | |
5452 | const struct net_device_ops *ops = dev->netdev_ops; | |
5453 | ||
5454 | if (!ops->ndo_change_carrier) | |
5455 | return -EOPNOTSUPP; | |
5456 | if (!netif_device_present(dev)) | |
5457 | return -ENODEV; | |
5458 | return ops->ndo_change_carrier(dev, new_carrier); | |
5459 | } | |
5460 | EXPORT_SYMBOL(dev_change_carrier); | |
5461 | ||
1da177e4 | 5462 | /* |
3710becf | 5463 | * Perform the SIOCxIFxxx calls, inside rcu_read_lock() |
1da177e4 | 5464 | */ |
14e3e079 | 5465 | static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd) |
1da177e4 LT |
5466 | { |
5467 | int err; | |
3710becf | 5468 | struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name); |
1da177e4 LT |
5469 | |
5470 | if (!dev) | |
5471 | return -ENODEV; | |
5472 | ||
5473 | switch (cmd) { | |
d1b19dff ED |
5474 | case SIOCGIFFLAGS: /* Get interface flags */ |
5475 | ifr->ifr_flags = (short) dev_get_flags(dev); | |
5476 | return 0; | |
1da177e4 | 5477 | |
d1b19dff ED |
5478 | case SIOCGIFMETRIC: /* Get the metric on the interface |
5479 | (currently unused) */ | |
5480 | ifr->ifr_metric = 0; | |
5481 | return 0; | |
1da177e4 | 5482 | |
d1b19dff ED |
5483 | case SIOCGIFMTU: /* Get the MTU of a device */ |
5484 | ifr->ifr_mtu = dev->mtu; | |
5485 | return 0; | |
1da177e4 | 5486 | |
d1b19dff ED |
5487 | case SIOCGIFHWADDR: |
5488 | if (!dev->addr_len) | |
5489 | memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data); | |
5490 | else | |
5491 | memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr, | |
5492 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
5493 | ifr->ifr_hwaddr.sa_family = dev->type; | |
5494 | return 0; | |
1da177e4 | 5495 | |
d1b19dff ED |
5496 | case SIOCGIFSLAVE: |
5497 | err = -EINVAL; | |
5498 | break; | |
14e3e079 | 5499 | |
d1b19dff ED |
5500 | case SIOCGIFMAP: |
5501 | ifr->ifr_map.mem_start = dev->mem_start; | |
5502 | ifr->ifr_map.mem_end = dev->mem_end; | |
5503 | ifr->ifr_map.base_addr = dev->base_addr; | |
5504 | ifr->ifr_map.irq = dev->irq; | |
5505 | ifr->ifr_map.dma = dev->dma; | |
5506 | ifr->ifr_map.port = dev->if_port; | |
5507 | return 0; | |
14e3e079 | 5508 | |
d1b19dff ED |
5509 | case SIOCGIFINDEX: |
5510 | ifr->ifr_ifindex = dev->ifindex; | |
5511 | return 0; | |
14e3e079 | 5512 | |
d1b19dff ED |
5513 | case SIOCGIFTXQLEN: |
5514 | ifr->ifr_qlen = dev->tx_queue_len; | |
5515 | return 0; | |
14e3e079 | 5516 | |
d1b19dff ED |
5517 | default: |
5518 | /* dev_ioctl() should ensure this case | |
5519 | * is never reached | |
5520 | */ | |
5521 | WARN_ON(1); | |
41c31f31 | 5522 | err = -ENOTTY; |
d1b19dff | 5523 | break; |
14e3e079 JG |
5524 | |
5525 | } | |
5526 | return err; | |
5527 | } | |
5528 | ||
5529 | /* | |
5530 | * Perform the SIOCxIFxxx calls, inside rtnl_lock() | |
5531 | */ | |
5532 | static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd) | |
5533 | { | |
5534 | int err; | |
5535 | struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name); | |
5f2f6da7 | 5536 | const struct net_device_ops *ops; |
14e3e079 JG |
5537 | |
5538 | if (!dev) | |
5539 | return -ENODEV; | |
5540 | ||
5f2f6da7 JP |
5541 | ops = dev->netdev_ops; |
5542 | ||
14e3e079 | 5543 | switch (cmd) { |
d1b19dff ED |
5544 | case SIOCSIFFLAGS: /* Set interface flags */ |
5545 | return dev_change_flags(dev, ifr->ifr_flags); | |
14e3e079 | 5546 | |
d1b19dff ED |
5547 | case SIOCSIFMETRIC: /* Set the metric on the interface |
5548 | (currently unused) */ | |
5549 | return -EOPNOTSUPP; | |
14e3e079 | 5550 | |
d1b19dff ED |
5551 | case SIOCSIFMTU: /* Set the MTU of a device */ |
5552 | return dev_set_mtu(dev, ifr->ifr_mtu); | |
1da177e4 | 5553 | |
d1b19dff ED |
5554 | case SIOCSIFHWADDR: |
5555 | return dev_set_mac_address(dev, &ifr->ifr_hwaddr); | |
1da177e4 | 5556 | |
d1b19dff ED |
5557 | case SIOCSIFHWBROADCAST: |
5558 | if (ifr->ifr_hwaddr.sa_family != dev->type) | |
5559 | return -EINVAL; | |
5560 | memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data, | |
5561 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
5562 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); | |
5563 | return 0; | |
1da177e4 | 5564 | |
d1b19dff ED |
5565 | case SIOCSIFMAP: |
5566 | if (ops->ndo_set_config) { | |
1da177e4 LT |
5567 | if (!netif_device_present(dev)) |
5568 | return -ENODEV; | |
d1b19dff ED |
5569 | return ops->ndo_set_config(dev, &ifr->ifr_map); |
5570 | } | |
5571 | return -EOPNOTSUPP; | |
1da177e4 | 5572 | |
d1b19dff | 5573 | case SIOCADDMULTI: |
b81693d9 | 5574 | if (!ops->ndo_set_rx_mode || |
d1b19dff ED |
5575 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) |
5576 | return -EINVAL; | |
5577 | if (!netif_device_present(dev)) | |
5578 | return -ENODEV; | |
22bedad3 | 5579 | return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data); |
d1b19dff ED |
5580 | |
5581 | case SIOCDELMULTI: | |
b81693d9 | 5582 | if (!ops->ndo_set_rx_mode || |
d1b19dff ED |
5583 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) |
5584 | return -EINVAL; | |
5585 | if (!netif_device_present(dev)) | |
5586 | return -ENODEV; | |
22bedad3 | 5587 | return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data); |
1da177e4 | 5588 | |
d1b19dff ED |
5589 | case SIOCSIFTXQLEN: |
5590 | if (ifr->ifr_qlen < 0) | |
5591 | return -EINVAL; | |
5592 | dev->tx_queue_len = ifr->ifr_qlen; | |
5593 | return 0; | |
1da177e4 | 5594 | |
d1b19dff ED |
5595 | case SIOCSIFNAME: |
5596 | ifr->ifr_newname[IFNAMSIZ-1] = '\0'; | |
5597 | return dev_change_name(dev, ifr->ifr_newname); | |
1da177e4 | 5598 | |
4dc360c5 RC |
5599 | case SIOCSHWTSTAMP: |
5600 | err = net_hwtstamp_validate(ifr); | |
5601 | if (err) | |
5602 | return err; | |
5603 | /* fall through */ | |
5604 | ||
d1b19dff ED |
5605 | /* |
5606 | * Unknown or private ioctl | |
5607 | */ | |
5608 | default: | |
5609 | if ((cmd >= SIOCDEVPRIVATE && | |
5610 | cmd <= SIOCDEVPRIVATE + 15) || | |
5611 | cmd == SIOCBONDENSLAVE || | |
5612 | cmd == SIOCBONDRELEASE || | |
5613 | cmd == SIOCBONDSETHWADDR || | |
5614 | cmd == SIOCBONDSLAVEINFOQUERY || | |
5615 | cmd == SIOCBONDINFOQUERY || | |
5616 | cmd == SIOCBONDCHANGEACTIVE || | |
5617 | cmd == SIOCGMIIPHY || | |
5618 | cmd == SIOCGMIIREG || | |
5619 | cmd == SIOCSMIIREG || | |
5620 | cmd == SIOCBRADDIF || | |
5621 | cmd == SIOCBRDELIF || | |
5622 | cmd == SIOCSHWTSTAMP || | |
5623 | cmd == SIOCWANDEV) { | |
5624 | err = -EOPNOTSUPP; | |
5625 | if (ops->ndo_do_ioctl) { | |
5626 | if (netif_device_present(dev)) | |
5627 | err = ops->ndo_do_ioctl(dev, ifr, cmd); | |
5628 | else | |
5629 | err = -ENODEV; | |
5630 | } | |
5631 | } else | |
5632 | err = -EINVAL; | |
1da177e4 LT |
5633 | |
5634 | } | |
5635 | return err; | |
5636 | } | |
5637 | ||
5638 | /* | |
5639 | * This function handles all "interface"-type I/O control requests. The actual | |
5640 | * 'doing' part of this is dev_ifsioc above. | |
5641 | */ | |
5642 | ||
5643 | /** | |
5644 | * dev_ioctl - network device ioctl | |
c4ea43c5 | 5645 | * @net: the applicable net namespace |
1da177e4 LT |
5646 | * @cmd: command to issue |
5647 | * @arg: pointer to a struct ifreq in user space | |
5648 | * | |
5649 | * Issue ioctl functions to devices. This is normally called by the | |
5650 | * user space syscall interfaces but can sometimes be useful for | |
5651 | * other purposes. The return value is the return from the syscall if | |
5652 | * positive or a negative errno code on error. | |
5653 | */ | |
5654 | ||
881d966b | 5655 | int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg) |
1da177e4 LT |
5656 | { |
5657 | struct ifreq ifr; | |
5658 | int ret; | |
5659 | char *colon; | |
5660 | ||
5661 | /* One special case: SIOCGIFCONF takes ifconf argument | |
5662 | and requires shared lock, because it sleeps writing | |
5663 | to user space. | |
5664 | */ | |
5665 | ||
5666 | if (cmd == SIOCGIFCONF) { | |
6756ae4b | 5667 | rtnl_lock(); |
881d966b | 5668 | ret = dev_ifconf(net, (char __user *) arg); |
6756ae4b | 5669 | rtnl_unlock(); |
1da177e4 LT |
5670 | return ret; |
5671 | } | |
5672 | if (cmd == SIOCGIFNAME) | |
881d966b | 5673 | return dev_ifname(net, (struct ifreq __user *)arg); |
1da177e4 LT |
5674 | |
5675 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
5676 | return -EFAULT; | |
5677 | ||
5678 | ifr.ifr_name[IFNAMSIZ-1] = 0; | |
5679 | ||
5680 | colon = strchr(ifr.ifr_name, ':'); | |
5681 | if (colon) | |
5682 | *colon = 0; | |
5683 | ||
5684 | /* | |
5685 | * See which interface the caller is talking about. | |
5686 | */ | |
5687 | ||
5688 | switch (cmd) { | |
d1b19dff ED |
5689 | /* |
5690 | * These ioctl calls: | |
5691 | * - can be done by all. | |
5692 | * - atomic and do not require locking. | |
5693 | * - return a value | |
5694 | */ | |
5695 | case SIOCGIFFLAGS: | |
5696 | case SIOCGIFMETRIC: | |
5697 | case SIOCGIFMTU: | |
5698 | case SIOCGIFHWADDR: | |
5699 | case SIOCGIFSLAVE: | |
5700 | case SIOCGIFMAP: | |
5701 | case SIOCGIFINDEX: | |
5702 | case SIOCGIFTXQLEN: | |
5703 | dev_load(net, ifr.ifr_name); | |
3710becf | 5704 | rcu_read_lock(); |
d1b19dff | 5705 | ret = dev_ifsioc_locked(net, &ifr, cmd); |
3710becf | 5706 | rcu_read_unlock(); |
d1b19dff ED |
5707 | if (!ret) { |
5708 | if (colon) | |
5709 | *colon = ':'; | |
5710 | if (copy_to_user(arg, &ifr, | |
5711 | sizeof(struct ifreq))) | |
5712 | ret = -EFAULT; | |
5713 | } | |
5714 | return ret; | |
1da177e4 | 5715 | |
d1b19dff ED |
5716 | case SIOCETHTOOL: |
5717 | dev_load(net, ifr.ifr_name); | |
5718 | rtnl_lock(); | |
5719 | ret = dev_ethtool(net, &ifr); | |
5720 | rtnl_unlock(); | |
5721 | if (!ret) { | |
5722 | if (colon) | |
5723 | *colon = ':'; | |
5724 | if (copy_to_user(arg, &ifr, | |
5725 | sizeof(struct ifreq))) | |
5726 | ret = -EFAULT; | |
5727 | } | |
5728 | return ret; | |
1da177e4 | 5729 | |
d1b19dff ED |
5730 | /* |
5731 | * These ioctl calls: | |
5732 | * - require superuser power. | |
5733 | * - require strict serialization. | |
5734 | * - return a value | |
5735 | */ | |
5736 | case SIOCGMIIPHY: | |
5737 | case SIOCGMIIREG: | |
5738 | case SIOCSIFNAME: | |
5e1fccc0 | 5739 | if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) |
d1b19dff ED |
5740 | return -EPERM; |
5741 | dev_load(net, ifr.ifr_name); | |
5742 | rtnl_lock(); | |
5743 | ret = dev_ifsioc(net, &ifr, cmd); | |
5744 | rtnl_unlock(); | |
5745 | if (!ret) { | |
5746 | if (colon) | |
5747 | *colon = ':'; | |
5748 | if (copy_to_user(arg, &ifr, | |
5749 | sizeof(struct ifreq))) | |
5750 | ret = -EFAULT; | |
5751 | } | |
5752 | return ret; | |
1da177e4 | 5753 | |
d1b19dff ED |
5754 | /* |
5755 | * These ioctl calls: | |
5756 | * - require superuser power. | |
5757 | * - require strict serialization. | |
5758 | * - do not return a value | |
5759 | */ | |
5e1fccc0 EB |
5760 | case SIOCSIFMAP: |
5761 | case SIOCSIFTXQLEN: | |
5762 | if (!capable(CAP_NET_ADMIN)) | |
5763 | return -EPERM; | |
5764 | /* fall through */ | |
5765 | /* | |
5766 | * These ioctl calls: | |
5767 | * - require local superuser power. | |
5768 | * - require strict serialization. | |
5769 | * - do not return a value | |
5770 | */ | |
d1b19dff ED |
5771 | case SIOCSIFFLAGS: |
5772 | case SIOCSIFMETRIC: | |
5773 | case SIOCSIFMTU: | |
d1b19dff ED |
5774 | case SIOCSIFHWADDR: |
5775 | case SIOCSIFSLAVE: | |
5776 | case SIOCADDMULTI: | |
5777 | case SIOCDELMULTI: | |
5778 | case SIOCSIFHWBROADCAST: | |
d1b19dff ED |
5779 | case SIOCSMIIREG: |
5780 | case SIOCBONDENSLAVE: | |
5781 | case SIOCBONDRELEASE: | |
5782 | case SIOCBONDSETHWADDR: | |
5783 | case SIOCBONDCHANGEACTIVE: | |
5784 | case SIOCBRADDIF: | |
5785 | case SIOCBRDELIF: | |
5786 | case SIOCSHWTSTAMP: | |
5e1fccc0 | 5787 | if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) |
d1b19dff ED |
5788 | return -EPERM; |
5789 | /* fall through */ | |
5790 | case SIOCBONDSLAVEINFOQUERY: | |
5791 | case SIOCBONDINFOQUERY: | |
5792 | dev_load(net, ifr.ifr_name); | |
5793 | rtnl_lock(); | |
5794 | ret = dev_ifsioc(net, &ifr, cmd); | |
5795 | rtnl_unlock(); | |
5796 | return ret; | |
5797 | ||
5798 | case SIOCGIFMEM: | |
5799 | /* Get the per device memory space. We can add this but | |
5800 | * currently do not support it */ | |
5801 | case SIOCSIFMEM: | |
5802 | /* Set the per device memory buffer space. | |
5803 | * Not applicable in our case */ | |
5804 | case SIOCSIFLINK: | |
41c31f31 | 5805 | return -ENOTTY; |
d1b19dff ED |
5806 | |
5807 | /* | |
5808 | * Unknown or private ioctl. | |
5809 | */ | |
5810 | default: | |
5811 | if (cmd == SIOCWANDEV || | |
5812 | (cmd >= SIOCDEVPRIVATE && | |
5813 | cmd <= SIOCDEVPRIVATE + 15)) { | |
881d966b | 5814 | dev_load(net, ifr.ifr_name); |
1da177e4 | 5815 | rtnl_lock(); |
881d966b | 5816 | ret = dev_ifsioc(net, &ifr, cmd); |
1da177e4 | 5817 | rtnl_unlock(); |
d1b19dff ED |
5818 | if (!ret && copy_to_user(arg, &ifr, |
5819 | sizeof(struct ifreq))) | |
5820 | ret = -EFAULT; | |
1da177e4 | 5821 | return ret; |
d1b19dff ED |
5822 | } |
5823 | /* Take care of Wireless Extensions */ | |
5824 | if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) | |
5825 | return wext_handle_ioctl(net, &ifr, cmd, arg); | |
41c31f31 | 5826 | return -ENOTTY; |
1da177e4 LT |
5827 | } |
5828 | } | |
5829 | ||
5830 | ||
5831 | /** | |
5832 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 5833 | * @net: the applicable net namespace |
1da177e4 LT |
5834 | * |
5835 | * Returns a suitable unique value for a new device interface | |
5836 | * number. The caller must hold the rtnl semaphore or the | |
5837 | * dev_base_lock to be sure it remains unique. | |
5838 | */ | |
881d966b | 5839 | static int dev_new_index(struct net *net) |
1da177e4 | 5840 | { |
aa79e66e | 5841 | int ifindex = net->ifindex; |
1da177e4 LT |
5842 | for (;;) { |
5843 | if (++ifindex <= 0) | |
5844 | ifindex = 1; | |
881d966b | 5845 | if (!__dev_get_by_index(net, ifindex)) |
aa79e66e | 5846 | return net->ifindex = ifindex; |
1da177e4 LT |
5847 | } |
5848 | } | |
5849 | ||
1da177e4 | 5850 | /* Delayed registration/unregisteration */ |
3b5b34fd | 5851 | static LIST_HEAD(net_todo_list); |
1da177e4 | 5852 | |
6f05f629 | 5853 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 5854 | { |
1da177e4 | 5855 | list_add_tail(&dev->todo_list, &net_todo_list); |
1da177e4 LT |
5856 | } |
5857 | ||
9b5e383c | 5858 | static void rollback_registered_many(struct list_head *head) |
93ee31f1 | 5859 | { |
e93737b0 | 5860 | struct net_device *dev, *tmp; |
9b5e383c | 5861 | |
93ee31f1 DL |
5862 | BUG_ON(dev_boot_phase); |
5863 | ASSERT_RTNL(); | |
5864 | ||
e93737b0 | 5865 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
9b5e383c | 5866 | /* Some devices call without registering |
e93737b0 KK |
5867 | * for initialization unwind. Remove those |
5868 | * devices and proceed with the remaining. | |
9b5e383c ED |
5869 | */ |
5870 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
7b6cd1ce JP |
5871 | pr_debug("unregister_netdevice: device %s/%p never was registered\n", |
5872 | dev->name, dev); | |
93ee31f1 | 5873 | |
9b5e383c | 5874 | WARN_ON(1); |
e93737b0 KK |
5875 | list_del(&dev->unreg_list); |
5876 | continue; | |
9b5e383c | 5877 | } |
449f4544 | 5878 | dev->dismantle = true; |
9b5e383c | 5879 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
44345724 | 5880 | } |
93ee31f1 | 5881 | |
44345724 OP |
5882 | /* If device is running, close it first. */ |
5883 | dev_close_many(head); | |
93ee31f1 | 5884 | |
44345724 | 5885 | list_for_each_entry(dev, head, unreg_list) { |
9b5e383c ED |
5886 | /* And unlink it from device chain. */ |
5887 | unlist_netdevice(dev); | |
93ee31f1 | 5888 | |
9b5e383c ED |
5889 | dev->reg_state = NETREG_UNREGISTERING; |
5890 | } | |
93ee31f1 DL |
5891 | |
5892 | synchronize_net(); | |
5893 | ||
9b5e383c ED |
5894 | list_for_each_entry(dev, head, unreg_list) { |
5895 | /* Shutdown queueing discipline. */ | |
5896 | dev_shutdown(dev); | |
93ee31f1 DL |
5897 | |
5898 | ||
9b5e383c ED |
5899 | /* Notify protocols, that we are about to destroy |
5900 | this device. They should clean all the things. | |
5901 | */ | |
5902 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
93ee31f1 | 5903 | |
a2835763 PM |
5904 | if (!dev->rtnl_link_ops || |
5905 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
5906 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U); | |
5907 | ||
9b5e383c ED |
5908 | /* |
5909 | * Flush the unicast and multicast chains | |
5910 | */ | |
a748ee24 | 5911 | dev_uc_flush(dev); |
22bedad3 | 5912 | dev_mc_flush(dev); |
93ee31f1 | 5913 | |
9b5e383c ED |
5914 | if (dev->netdev_ops->ndo_uninit) |
5915 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 | 5916 | |
9ff162a8 JP |
5917 | /* Notifier chain MUST detach us all upper devices. */ |
5918 | WARN_ON(netdev_has_any_upper_dev(dev)); | |
93ee31f1 | 5919 | |
9b5e383c ED |
5920 | /* Remove entries from kobject tree */ |
5921 | netdev_unregister_kobject(dev); | |
5922 | } | |
93ee31f1 | 5923 | |
850a545b | 5924 | synchronize_net(); |
395264d5 | 5925 | |
a5ee1551 | 5926 | list_for_each_entry(dev, head, unreg_list) |
9b5e383c ED |
5927 | dev_put(dev); |
5928 | } | |
5929 | ||
5930 | static void rollback_registered(struct net_device *dev) | |
5931 | { | |
5932 | LIST_HEAD(single); | |
5933 | ||
5934 | list_add(&dev->unreg_list, &single); | |
5935 | rollback_registered_many(&single); | |
ceaaec98 | 5936 | list_del(&single); |
93ee31f1 DL |
5937 | } |
5938 | ||
c8f44aff MM |
5939 | static netdev_features_t netdev_fix_features(struct net_device *dev, |
5940 | netdev_features_t features) | |
b63365a2 | 5941 | { |
57422dc5 MM |
5942 | /* Fix illegal checksum combinations */ |
5943 | if ((features & NETIF_F_HW_CSUM) && | |
5944 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 5945 | netdev_warn(dev, "mixed HW and IP checksum settings.\n"); |
57422dc5 MM |
5946 | features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); |
5947 | } | |
5948 | ||
b63365a2 HX |
5949 | /* Fix illegal SG+CSUM combinations. */ |
5950 | if ((features & NETIF_F_SG) && | |
5951 | !(features & NETIF_F_ALL_CSUM)) { | |
6f404e44 MM |
5952 | netdev_dbg(dev, |
5953 | "Dropping NETIF_F_SG since no checksum feature.\n"); | |
b63365a2 HX |
5954 | features &= ~NETIF_F_SG; |
5955 | } | |
5956 | ||
5957 | /* TSO requires that SG is present as well. */ | |
ea2d3688 | 5958 | if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { |
6f404e44 | 5959 | netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); |
ea2d3688 | 5960 | features &= ~NETIF_F_ALL_TSO; |
b63365a2 HX |
5961 | } |
5962 | ||
31d8b9e0 BH |
5963 | /* TSO ECN requires that TSO is present as well. */ |
5964 | if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) | |
5965 | features &= ~NETIF_F_TSO_ECN; | |
5966 | ||
212b573f MM |
5967 | /* Software GSO depends on SG. */ |
5968 | if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { | |
6f404e44 | 5969 | netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); |
212b573f MM |
5970 | features &= ~NETIF_F_GSO; |
5971 | } | |
5972 | ||
acd1130e | 5973 | /* UFO needs SG and checksumming */ |
b63365a2 | 5974 | if (features & NETIF_F_UFO) { |
79032644 MM |
5975 | /* maybe split UFO into V4 and V6? */ |
5976 | if (!((features & NETIF_F_GEN_CSUM) || | |
5977 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM)) | |
5978 | == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 5979 | netdev_dbg(dev, |
acd1130e | 5980 | "Dropping NETIF_F_UFO since no checksum offload features.\n"); |
b63365a2 HX |
5981 | features &= ~NETIF_F_UFO; |
5982 | } | |
5983 | ||
5984 | if (!(features & NETIF_F_SG)) { | |
6f404e44 | 5985 | netdev_dbg(dev, |
acd1130e | 5986 | "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n"); |
b63365a2 HX |
5987 | features &= ~NETIF_F_UFO; |
5988 | } | |
5989 | } | |
5990 | ||
5991 | return features; | |
5992 | } | |
b63365a2 | 5993 | |
6cb6a27c | 5994 | int __netdev_update_features(struct net_device *dev) |
5455c699 | 5995 | { |
c8f44aff | 5996 | netdev_features_t features; |
5455c699 MM |
5997 | int err = 0; |
5998 | ||
87267485 MM |
5999 | ASSERT_RTNL(); |
6000 | ||
5455c699 MM |
6001 | features = netdev_get_wanted_features(dev); |
6002 | ||
6003 | if (dev->netdev_ops->ndo_fix_features) | |
6004 | features = dev->netdev_ops->ndo_fix_features(dev, features); | |
6005 | ||
6006 | /* driver might be less strict about feature dependencies */ | |
6007 | features = netdev_fix_features(dev, features); | |
6008 | ||
6009 | if (dev->features == features) | |
6cb6a27c | 6010 | return 0; |
5455c699 | 6011 | |
c8f44aff MM |
6012 | netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", |
6013 | &dev->features, &features); | |
5455c699 MM |
6014 | |
6015 | if (dev->netdev_ops->ndo_set_features) | |
6016 | err = dev->netdev_ops->ndo_set_features(dev, features); | |
6017 | ||
6cb6a27c | 6018 | if (unlikely(err < 0)) { |
5455c699 | 6019 | netdev_err(dev, |
c8f44aff MM |
6020 | "set_features() failed (%d); wanted %pNF, left %pNF\n", |
6021 | err, &features, &dev->features); | |
6cb6a27c MM |
6022 | return -1; |
6023 | } | |
6024 | ||
6025 | if (!err) | |
6026 | dev->features = features; | |
6027 | ||
6028 | return 1; | |
6029 | } | |
6030 | ||
afe12cc8 MM |
6031 | /** |
6032 | * netdev_update_features - recalculate device features | |
6033 | * @dev: the device to check | |
6034 | * | |
6035 | * Recalculate dev->features set and send notifications if it | |
6036 | * has changed. Should be called after driver or hardware dependent | |
6037 | * conditions might have changed that influence the features. | |
6038 | */ | |
6cb6a27c MM |
6039 | void netdev_update_features(struct net_device *dev) |
6040 | { | |
6041 | if (__netdev_update_features(dev)) | |
6042 | netdev_features_change(dev); | |
5455c699 MM |
6043 | } |
6044 | EXPORT_SYMBOL(netdev_update_features); | |
6045 | ||
afe12cc8 MM |
6046 | /** |
6047 | * netdev_change_features - recalculate device features | |
6048 | * @dev: the device to check | |
6049 | * | |
6050 | * Recalculate dev->features set and send notifications even | |
6051 | * if they have not changed. Should be called instead of | |
6052 | * netdev_update_features() if also dev->vlan_features might | |
6053 | * have changed to allow the changes to be propagated to stacked | |
6054 | * VLAN devices. | |
6055 | */ | |
6056 | void netdev_change_features(struct net_device *dev) | |
6057 | { | |
6058 | __netdev_update_features(dev); | |
6059 | netdev_features_change(dev); | |
6060 | } | |
6061 | EXPORT_SYMBOL(netdev_change_features); | |
6062 | ||
fc4a7489 PM |
6063 | /** |
6064 | * netif_stacked_transfer_operstate - transfer operstate | |
6065 | * @rootdev: the root or lower level device to transfer state from | |
6066 | * @dev: the device to transfer operstate to | |
6067 | * | |
6068 | * Transfer operational state from root to device. This is normally | |
6069 | * called when a stacking relationship exists between the root | |
6070 | * device and the device(a leaf device). | |
6071 | */ | |
6072 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
6073 | struct net_device *dev) | |
6074 | { | |
6075 | if (rootdev->operstate == IF_OPER_DORMANT) | |
6076 | netif_dormant_on(dev); | |
6077 | else | |
6078 | netif_dormant_off(dev); | |
6079 | ||
6080 | if (netif_carrier_ok(rootdev)) { | |
6081 | if (!netif_carrier_ok(dev)) | |
6082 | netif_carrier_on(dev); | |
6083 | } else { | |
6084 | if (netif_carrier_ok(dev)) | |
6085 | netif_carrier_off(dev); | |
6086 | } | |
6087 | } | |
6088 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
6089 | ||
bf264145 | 6090 | #ifdef CONFIG_RPS |
1b4bf461 ED |
6091 | static int netif_alloc_rx_queues(struct net_device *dev) |
6092 | { | |
1b4bf461 | 6093 | unsigned int i, count = dev->num_rx_queues; |
bd25fa7b | 6094 | struct netdev_rx_queue *rx; |
1b4bf461 | 6095 | |
bd25fa7b | 6096 | BUG_ON(count < 1); |
1b4bf461 | 6097 | |
bd25fa7b TH |
6098 | rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL); |
6099 | if (!rx) { | |
7b6cd1ce | 6100 | pr_err("netdev: Unable to allocate %u rx queues\n", count); |
bd25fa7b | 6101 | return -ENOMEM; |
1b4bf461 | 6102 | } |
bd25fa7b TH |
6103 | dev->_rx = rx; |
6104 | ||
bd25fa7b | 6105 | for (i = 0; i < count; i++) |
fe822240 | 6106 | rx[i].dev = dev; |
1b4bf461 ED |
6107 | return 0; |
6108 | } | |
bf264145 | 6109 | #endif |
1b4bf461 | 6110 | |
aa942104 CG |
6111 | static void netdev_init_one_queue(struct net_device *dev, |
6112 | struct netdev_queue *queue, void *_unused) | |
6113 | { | |
6114 | /* Initialize queue lock */ | |
6115 | spin_lock_init(&queue->_xmit_lock); | |
6116 | netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); | |
6117 | queue->xmit_lock_owner = -1; | |
b236da69 | 6118 | netdev_queue_numa_node_write(queue, NUMA_NO_NODE); |
aa942104 | 6119 | queue->dev = dev; |
114cf580 TH |
6120 | #ifdef CONFIG_BQL |
6121 | dql_init(&queue->dql, HZ); | |
6122 | #endif | |
aa942104 CG |
6123 | } |
6124 | ||
e6484930 TH |
6125 | static int netif_alloc_netdev_queues(struct net_device *dev) |
6126 | { | |
6127 | unsigned int count = dev->num_tx_queues; | |
6128 | struct netdev_queue *tx; | |
6129 | ||
6130 | BUG_ON(count < 1); | |
6131 | ||
6132 | tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL); | |
6133 | if (!tx) { | |
7b6cd1ce | 6134 | pr_err("netdev: Unable to allocate %u tx queues\n", count); |
e6484930 TH |
6135 | return -ENOMEM; |
6136 | } | |
6137 | dev->_tx = tx; | |
1d24eb48 | 6138 | |
e6484930 TH |
6139 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); |
6140 | spin_lock_init(&dev->tx_global_lock); | |
aa942104 CG |
6141 | |
6142 | return 0; | |
e6484930 TH |
6143 | } |
6144 | ||
1da177e4 LT |
6145 | /** |
6146 | * register_netdevice - register a network device | |
6147 | * @dev: device to register | |
6148 | * | |
6149 | * Take a completed network device structure and add it to the kernel | |
6150 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
6151 | * chain. 0 is returned on success. A negative errno code is returned | |
6152 | * on a failure to set up the device, or if the name is a duplicate. | |
6153 | * | |
6154 | * Callers must hold the rtnl semaphore. You may want | |
6155 | * register_netdev() instead of this. | |
6156 | * | |
6157 | * BUGS: | |
6158 | * The locking appears insufficient to guarantee two parallel registers | |
6159 | * will not get the same name. | |
6160 | */ | |
6161 | ||
6162 | int register_netdevice(struct net_device *dev) | |
6163 | { | |
1da177e4 | 6164 | int ret; |
d314774c | 6165 | struct net *net = dev_net(dev); |
1da177e4 LT |
6166 | |
6167 | BUG_ON(dev_boot_phase); | |
6168 | ASSERT_RTNL(); | |
6169 | ||
b17a7c17 SH |
6170 | might_sleep(); |
6171 | ||
1da177e4 LT |
6172 | /* When net_device's are persistent, this will be fatal. */ |
6173 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 6174 | BUG_ON(!net); |
1da177e4 | 6175 | |
f1f28aa3 | 6176 | spin_lock_init(&dev->addr_list_lock); |
cf508b12 | 6177 | netdev_set_addr_lockdep_class(dev); |
1da177e4 | 6178 | |
1da177e4 LT |
6179 | dev->iflink = -1; |
6180 | ||
828de4f6 | 6181 | ret = dev_get_valid_name(net, dev, dev->name); |
0696c3a8 PP |
6182 | if (ret < 0) |
6183 | goto out; | |
6184 | ||
1da177e4 | 6185 | /* Init, if this function is available */ |
d314774c SH |
6186 | if (dev->netdev_ops->ndo_init) { |
6187 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
6188 | if (ret) { |
6189 | if (ret > 0) | |
6190 | ret = -EIO; | |
90833aa4 | 6191 | goto out; |
1da177e4 LT |
6192 | } |
6193 | } | |
4ec93edb | 6194 | |
9c7dafbf PE |
6195 | ret = -EBUSY; |
6196 | if (!dev->ifindex) | |
6197 | dev->ifindex = dev_new_index(net); | |
6198 | else if (__dev_get_by_index(net, dev->ifindex)) | |
6199 | goto err_uninit; | |
6200 | ||
1da177e4 LT |
6201 | if (dev->iflink == -1) |
6202 | dev->iflink = dev->ifindex; | |
6203 | ||
5455c699 MM |
6204 | /* Transfer changeable features to wanted_features and enable |
6205 | * software offloads (GSO and GRO). | |
6206 | */ | |
6207 | dev->hw_features |= NETIF_F_SOFT_FEATURES; | |
14d1232f MM |
6208 | dev->features |= NETIF_F_SOFT_FEATURES; |
6209 | dev->wanted_features = dev->features & dev->hw_features; | |
1da177e4 | 6210 | |
c6e1a0d1 | 6211 | /* Turn on no cache copy if HW is doing checksum */ |
34324dc2 MM |
6212 | if (!(dev->flags & IFF_LOOPBACK)) { |
6213 | dev->hw_features |= NETIF_F_NOCACHE_COPY; | |
6214 | if (dev->features & NETIF_F_ALL_CSUM) { | |
6215 | dev->wanted_features |= NETIF_F_NOCACHE_COPY; | |
6216 | dev->features |= NETIF_F_NOCACHE_COPY; | |
6217 | } | |
c6e1a0d1 TH |
6218 | } |
6219 | ||
1180e7d6 | 6220 | /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. |
16c3ea78 | 6221 | */ |
1180e7d6 | 6222 | dev->vlan_features |= NETIF_F_HIGHDMA; |
16c3ea78 | 6223 | |
7ffbe3fd JB |
6224 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); |
6225 | ret = notifier_to_errno(ret); | |
6226 | if (ret) | |
6227 | goto err_uninit; | |
6228 | ||
8b41d188 | 6229 | ret = netdev_register_kobject(dev); |
b17a7c17 | 6230 | if (ret) |
7ce1b0ed | 6231 | goto err_uninit; |
b17a7c17 SH |
6232 | dev->reg_state = NETREG_REGISTERED; |
6233 | ||
6cb6a27c | 6234 | __netdev_update_features(dev); |
8e9b59b2 | 6235 | |
1da177e4 LT |
6236 | /* |
6237 | * Default initial state at registry is that the | |
6238 | * device is present. | |
6239 | */ | |
6240 | ||
6241 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
6242 | ||
8f4cccbb BH |
6243 | linkwatch_init_dev(dev); |
6244 | ||
1da177e4 | 6245 | dev_init_scheduler(dev); |
1da177e4 | 6246 | dev_hold(dev); |
ce286d32 | 6247 | list_netdevice(dev); |
7bf23575 | 6248 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 6249 | |
948b337e JP |
6250 | /* If the device has permanent device address, driver should |
6251 | * set dev_addr and also addr_assign_type should be set to | |
6252 | * NET_ADDR_PERM (default value). | |
6253 | */ | |
6254 | if (dev->addr_assign_type == NET_ADDR_PERM) | |
6255 | memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); | |
6256 | ||
1da177e4 | 6257 | /* Notify protocols, that a new device appeared. */ |
056925ab | 6258 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 6259 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
6260 | if (ret) { |
6261 | rollback_registered(dev); | |
6262 | dev->reg_state = NETREG_UNREGISTERED; | |
6263 | } | |
d90a909e EB |
6264 | /* |
6265 | * Prevent userspace races by waiting until the network | |
6266 | * device is fully setup before sending notifications. | |
6267 | */ | |
a2835763 PM |
6268 | if (!dev->rtnl_link_ops || |
6269 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
6270 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U); | |
1da177e4 LT |
6271 | |
6272 | out: | |
6273 | return ret; | |
7ce1b0ed HX |
6274 | |
6275 | err_uninit: | |
d314774c SH |
6276 | if (dev->netdev_ops->ndo_uninit) |
6277 | dev->netdev_ops->ndo_uninit(dev); | |
7ce1b0ed | 6278 | goto out; |
1da177e4 | 6279 | } |
d1b19dff | 6280 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 6281 | |
937f1ba5 BH |
6282 | /** |
6283 | * init_dummy_netdev - init a dummy network device for NAPI | |
6284 | * @dev: device to init | |
6285 | * | |
6286 | * This takes a network device structure and initialize the minimum | |
6287 | * amount of fields so it can be used to schedule NAPI polls without | |
6288 | * registering a full blown interface. This is to be used by drivers | |
6289 | * that need to tie several hardware interfaces to a single NAPI | |
6290 | * poll scheduler due to HW limitations. | |
6291 | */ | |
6292 | int init_dummy_netdev(struct net_device *dev) | |
6293 | { | |
6294 | /* Clear everything. Note we don't initialize spinlocks | |
6295 | * are they aren't supposed to be taken by any of the | |
6296 | * NAPI code and this dummy netdev is supposed to be | |
6297 | * only ever used for NAPI polls | |
6298 | */ | |
6299 | memset(dev, 0, sizeof(struct net_device)); | |
6300 | ||
6301 | /* make sure we BUG if trying to hit standard | |
6302 | * register/unregister code path | |
6303 | */ | |
6304 | dev->reg_state = NETREG_DUMMY; | |
6305 | ||
937f1ba5 BH |
6306 | /* NAPI wants this */ |
6307 | INIT_LIST_HEAD(&dev->napi_list); | |
6308 | ||
6309 | /* a dummy interface is started by default */ | |
6310 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
6311 | set_bit(__LINK_STATE_START, &dev->state); | |
6312 | ||
29b4433d ED |
6313 | /* Note : We dont allocate pcpu_refcnt for dummy devices, |
6314 | * because users of this 'device' dont need to change | |
6315 | * its refcount. | |
6316 | */ | |
6317 | ||
937f1ba5 BH |
6318 | return 0; |
6319 | } | |
6320 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
6321 | ||
6322 | ||
1da177e4 LT |
6323 | /** |
6324 | * register_netdev - register a network device | |
6325 | * @dev: device to register | |
6326 | * | |
6327 | * Take a completed network device structure and add it to the kernel | |
6328 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
6329 | * chain. 0 is returned on success. A negative errno code is returned | |
6330 | * on a failure to set up the device, or if the name is a duplicate. | |
6331 | * | |
38b4da38 | 6332 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
6333 | * and expands the device name if you passed a format string to |
6334 | * alloc_netdev. | |
6335 | */ | |
6336 | int register_netdev(struct net_device *dev) | |
6337 | { | |
6338 | int err; | |
6339 | ||
6340 | rtnl_lock(); | |
1da177e4 | 6341 | err = register_netdevice(dev); |
1da177e4 LT |
6342 | rtnl_unlock(); |
6343 | return err; | |
6344 | } | |
6345 | EXPORT_SYMBOL(register_netdev); | |
6346 | ||
29b4433d ED |
6347 | int netdev_refcnt_read(const struct net_device *dev) |
6348 | { | |
6349 | int i, refcnt = 0; | |
6350 | ||
6351 | for_each_possible_cpu(i) | |
6352 | refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); | |
6353 | return refcnt; | |
6354 | } | |
6355 | EXPORT_SYMBOL(netdev_refcnt_read); | |
6356 | ||
2c53040f | 6357 | /** |
1da177e4 | 6358 | * netdev_wait_allrefs - wait until all references are gone. |
3de7a37b | 6359 | * @dev: target net_device |
1da177e4 LT |
6360 | * |
6361 | * This is called when unregistering network devices. | |
6362 | * | |
6363 | * Any protocol or device that holds a reference should register | |
6364 | * for netdevice notification, and cleanup and put back the | |
6365 | * reference if they receive an UNREGISTER event. | |
6366 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 6367 | * call dev_put. |
1da177e4 LT |
6368 | */ |
6369 | static void netdev_wait_allrefs(struct net_device *dev) | |
6370 | { | |
6371 | unsigned long rebroadcast_time, warning_time; | |
29b4433d | 6372 | int refcnt; |
1da177e4 | 6373 | |
e014debe ED |
6374 | linkwatch_forget_dev(dev); |
6375 | ||
1da177e4 | 6376 | rebroadcast_time = warning_time = jiffies; |
29b4433d ED |
6377 | refcnt = netdev_refcnt_read(dev); |
6378 | ||
6379 | while (refcnt != 0) { | |
1da177e4 | 6380 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { |
6756ae4b | 6381 | rtnl_lock(); |
1da177e4 LT |
6382 | |
6383 | /* Rebroadcast unregister notification */ | |
056925ab | 6384 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
1da177e4 | 6385 | |
748e2d93 | 6386 | __rtnl_unlock(); |
0115e8e3 | 6387 | rcu_barrier(); |
748e2d93 ED |
6388 | rtnl_lock(); |
6389 | ||
0115e8e3 | 6390 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
1da177e4 LT |
6391 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, |
6392 | &dev->state)) { | |
6393 | /* We must not have linkwatch events | |
6394 | * pending on unregister. If this | |
6395 | * happens, we simply run the queue | |
6396 | * unscheduled, resulting in a noop | |
6397 | * for this device. | |
6398 | */ | |
6399 | linkwatch_run_queue(); | |
6400 | } | |
6401 | ||
6756ae4b | 6402 | __rtnl_unlock(); |
1da177e4 LT |
6403 | |
6404 | rebroadcast_time = jiffies; | |
6405 | } | |
6406 | ||
6407 | msleep(250); | |
6408 | ||
29b4433d ED |
6409 | refcnt = netdev_refcnt_read(dev); |
6410 | ||
1da177e4 | 6411 | if (time_after(jiffies, warning_time + 10 * HZ)) { |
7b6cd1ce JP |
6412 | pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", |
6413 | dev->name, refcnt); | |
1da177e4 LT |
6414 | warning_time = jiffies; |
6415 | } | |
6416 | } | |
6417 | } | |
6418 | ||
6419 | /* The sequence is: | |
6420 | * | |
6421 | * rtnl_lock(); | |
6422 | * ... | |
6423 | * register_netdevice(x1); | |
6424 | * register_netdevice(x2); | |
6425 | * ... | |
6426 | * unregister_netdevice(y1); | |
6427 | * unregister_netdevice(y2); | |
6428 | * ... | |
6429 | * rtnl_unlock(); | |
6430 | * free_netdev(y1); | |
6431 | * free_netdev(y2); | |
6432 | * | |
58ec3b4d | 6433 | * We are invoked by rtnl_unlock(). |
1da177e4 | 6434 | * This allows us to deal with problems: |
b17a7c17 | 6435 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
6436 | * without deadlocking with linkwatch via keventd. |
6437 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
6438 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
6439 | * |
6440 | * We must not return until all unregister events added during | |
6441 | * the interval the lock was held have been completed. | |
1da177e4 | 6442 | */ |
1da177e4 LT |
6443 | void netdev_run_todo(void) |
6444 | { | |
626ab0e6 | 6445 | struct list_head list; |
1da177e4 | 6446 | |
1da177e4 | 6447 | /* Snapshot list, allow later requests */ |
626ab0e6 | 6448 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
6449 | |
6450 | __rtnl_unlock(); | |
626ab0e6 | 6451 | |
0115e8e3 ED |
6452 | |
6453 | /* Wait for rcu callbacks to finish before next phase */ | |
850a545b EB |
6454 | if (!list_empty(&list)) |
6455 | rcu_barrier(); | |
6456 | ||
1da177e4 LT |
6457 | while (!list_empty(&list)) { |
6458 | struct net_device *dev | |
e5e26d75 | 6459 | = list_first_entry(&list, struct net_device, todo_list); |
1da177e4 LT |
6460 | list_del(&dev->todo_list); |
6461 | ||
748e2d93 | 6462 | rtnl_lock(); |
0115e8e3 | 6463 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
748e2d93 | 6464 | __rtnl_unlock(); |
0115e8e3 | 6465 | |
b17a7c17 | 6466 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
7b6cd1ce | 6467 | pr_err("network todo '%s' but state %d\n", |
b17a7c17 SH |
6468 | dev->name, dev->reg_state); |
6469 | dump_stack(); | |
6470 | continue; | |
6471 | } | |
1da177e4 | 6472 | |
b17a7c17 | 6473 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 6474 | |
152102c7 | 6475 | on_each_cpu(flush_backlog, dev, 1); |
6e583ce5 | 6476 | |
b17a7c17 | 6477 | netdev_wait_allrefs(dev); |
1da177e4 | 6478 | |
b17a7c17 | 6479 | /* paranoia */ |
29b4433d | 6480 | BUG_ON(netdev_refcnt_read(dev)); |
33d480ce ED |
6481 | WARN_ON(rcu_access_pointer(dev->ip_ptr)); |
6482 | WARN_ON(rcu_access_pointer(dev->ip6_ptr)); | |
547b792c | 6483 | WARN_ON(dev->dn_ptr); |
1da177e4 | 6484 | |
b17a7c17 SH |
6485 | if (dev->destructor) |
6486 | dev->destructor(dev); | |
9093bbb2 SH |
6487 | |
6488 | /* Free network device */ | |
6489 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 6490 | } |
1da177e4 LT |
6491 | } |
6492 | ||
3cfde79c BH |
6493 | /* Convert net_device_stats to rtnl_link_stats64. They have the same |
6494 | * fields in the same order, with only the type differing. | |
6495 | */ | |
77a1abf5 ED |
6496 | void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, |
6497 | const struct net_device_stats *netdev_stats) | |
3cfde79c BH |
6498 | { |
6499 | #if BITS_PER_LONG == 64 | |
77a1abf5 ED |
6500 | BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats)); |
6501 | memcpy(stats64, netdev_stats, sizeof(*stats64)); | |
3cfde79c BH |
6502 | #else |
6503 | size_t i, n = sizeof(*stats64) / sizeof(u64); | |
6504 | const unsigned long *src = (const unsigned long *)netdev_stats; | |
6505 | u64 *dst = (u64 *)stats64; | |
6506 | ||
6507 | BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) != | |
6508 | sizeof(*stats64) / sizeof(u64)); | |
6509 | for (i = 0; i < n; i++) | |
6510 | dst[i] = src[i]; | |
6511 | #endif | |
6512 | } | |
77a1abf5 | 6513 | EXPORT_SYMBOL(netdev_stats_to_stats64); |
3cfde79c | 6514 | |
eeda3fd6 SH |
6515 | /** |
6516 | * dev_get_stats - get network device statistics | |
6517 | * @dev: device to get statistics from | |
28172739 | 6518 | * @storage: place to store stats |
eeda3fd6 | 6519 | * |
d7753516 BH |
6520 | * Get network statistics from device. Return @storage. |
6521 | * The device driver may provide its own method by setting | |
6522 | * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; | |
6523 | * otherwise the internal statistics structure is used. | |
eeda3fd6 | 6524 | */ |
d7753516 BH |
6525 | struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, |
6526 | struct rtnl_link_stats64 *storage) | |
7004bf25 | 6527 | { |
eeda3fd6 SH |
6528 | const struct net_device_ops *ops = dev->netdev_ops; |
6529 | ||
28172739 ED |
6530 | if (ops->ndo_get_stats64) { |
6531 | memset(storage, 0, sizeof(*storage)); | |
caf586e5 ED |
6532 | ops->ndo_get_stats64(dev, storage); |
6533 | } else if (ops->ndo_get_stats) { | |
3cfde79c | 6534 | netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); |
caf586e5 ED |
6535 | } else { |
6536 | netdev_stats_to_stats64(storage, &dev->stats); | |
28172739 | 6537 | } |
caf586e5 | 6538 | storage->rx_dropped += atomic_long_read(&dev->rx_dropped); |
28172739 | 6539 | return storage; |
c45d286e | 6540 | } |
eeda3fd6 | 6541 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 6542 | |
24824a09 | 6543 | struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) |
dc2b4847 | 6544 | { |
24824a09 | 6545 | struct netdev_queue *queue = dev_ingress_queue(dev); |
dc2b4847 | 6546 | |
24824a09 ED |
6547 | #ifdef CONFIG_NET_CLS_ACT |
6548 | if (queue) | |
6549 | return queue; | |
6550 | queue = kzalloc(sizeof(*queue), GFP_KERNEL); | |
6551 | if (!queue) | |
6552 | return NULL; | |
6553 | netdev_init_one_queue(dev, queue, NULL); | |
24824a09 ED |
6554 | queue->qdisc = &noop_qdisc; |
6555 | queue->qdisc_sleeping = &noop_qdisc; | |
6556 | rcu_assign_pointer(dev->ingress_queue, queue); | |
6557 | #endif | |
6558 | return queue; | |
bb949fbd DM |
6559 | } |
6560 | ||
2c60db03 ED |
6561 | static const struct ethtool_ops default_ethtool_ops; |
6562 | ||
1da177e4 | 6563 | /** |
36909ea4 | 6564 | * alloc_netdev_mqs - allocate network device |
1da177e4 LT |
6565 | * @sizeof_priv: size of private data to allocate space for |
6566 | * @name: device name format string | |
6567 | * @setup: callback to initialize device | |
36909ea4 TH |
6568 | * @txqs: the number of TX subqueues to allocate |
6569 | * @rxqs: the number of RX subqueues to allocate | |
1da177e4 LT |
6570 | * |
6571 | * Allocates a struct net_device with private data area for driver use | |
f25f4e44 | 6572 | * and performs basic initialization. Also allocates subquue structs |
36909ea4 | 6573 | * for each queue on the device. |
1da177e4 | 6574 | */ |
36909ea4 TH |
6575 | struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, |
6576 | void (*setup)(struct net_device *), | |
6577 | unsigned int txqs, unsigned int rxqs) | |
1da177e4 | 6578 | { |
1da177e4 | 6579 | struct net_device *dev; |
7943986c | 6580 | size_t alloc_size; |
1ce8e7b5 | 6581 | struct net_device *p; |
1da177e4 | 6582 | |
b6fe17d6 SH |
6583 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
6584 | ||
36909ea4 | 6585 | if (txqs < 1) { |
7b6cd1ce | 6586 | pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); |
55513fb4 TH |
6587 | return NULL; |
6588 | } | |
6589 | ||
36909ea4 TH |
6590 | #ifdef CONFIG_RPS |
6591 | if (rxqs < 1) { | |
7b6cd1ce | 6592 | pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); |
36909ea4 TH |
6593 | return NULL; |
6594 | } | |
6595 | #endif | |
6596 | ||
fd2ea0a7 | 6597 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
6598 | if (sizeof_priv) { |
6599 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 6600 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
6601 | alloc_size += sizeof_priv; |
6602 | } | |
6603 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 6604 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 6605 | |
31380de9 | 6606 | p = kzalloc(alloc_size, GFP_KERNEL); |
1da177e4 | 6607 | if (!p) { |
7b6cd1ce | 6608 | pr_err("alloc_netdev: Unable to allocate device\n"); |
1da177e4 LT |
6609 | return NULL; |
6610 | } | |
1da177e4 | 6611 | |
1ce8e7b5 | 6612 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 6613 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc | 6614 | |
29b4433d ED |
6615 | dev->pcpu_refcnt = alloc_percpu(int); |
6616 | if (!dev->pcpu_refcnt) | |
e6484930 | 6617 | goto free_p; |
ab9c73cc | 6618 | |
ab9c73cc | 6619 | if (dev_addr_init(dev)) |
29b4433d | 6620 | goto free_pcpu; |
ab9c73cc | 6621 | |
22bedad3 | 6622 | dev_mc_init(dev); |
a748ee24 | 6623 | dev_uc_init(dev); |
ccffad25 | 6624 | |
c346dca1 | 6625 | dev_net_set(dev, &init_net); |
1da177e4 | 6626 | |
8d3bdbd5 | 6627 | dev->gso_max_size = GSO_MAX_SIZE; |
30b678d8 | 6628 | dev->gso_max_segs = GSO_MAX_SEGS; |
8d3bdbd5 | 6629 | |
8d3bdbd5 DM |
6630 | INIT_LIST_HEAD(&dev->napi_list); |
6631 | INIT_LIST_HEAD(&dev->unreg_list); | |
6632 | INIT_LIST_HEAD(&dev->link_watch_list); | |
9ff162a8 | 6633 | INIT_LIST_HEAD(&dev->upper_dev_list); |
8d3bdbd5 DM |
6634 | dev->priv_flags = IFF_XMIT_DST_RELEASE; |
6635 | setup(dev); | |
6636 | ||
36909ea4 TH |
6637 | dev->num_tx_queues = txqs; |
6638 | dev->real_num_tx_queues = txqs; | |
ed9af2e8 | 6639 | if (netif_alloc_netdev_queues(dev)) |
8d3bdbd5 | 6640 | goto free_all; |
e8a0464c | 6641 | |
df334545 | 6642 | #ifdef CONFIG_RPS |
36909ea4 TH |
6643 | dev->num_rx_queues = rxqs; |
6644 | dev->real_num_rx_queues = rxqs; | |
fe822240 | 6645 | if (netif_alloc_rx_queues(dev)) |
8d3bdbd5 | 6646 | goto free_all; |
df334545 | 6647 | #endif |
0a9627f2 | 6648 | |
1da177e4 | 6649 | strcpy(dev->name, name); |
cbda10fa | 6650 | dev->group = INIT_NETDEV_GROUP; |
2c60db03 ED |
6651 | if (!dev->ethtool_ops) |
6652 | dev->ethtool_ops = &default_ethtool_ops; | |
1da177e4 | 6653 | return dev; |
ab9c73cc | 6654 | |
8d3bdbd5 DM |
6655 | free_all: |
6656 | free_netdev(dev); | |
6657 | return NULL; | |
6658 | ||
29b4433d ED |
6659 | free_pcpu: |
6660 | free_percpu(dev->pcpu_refcnt); | |
ed9af2e8 | 6661 | kfree(dev->_tx); |
fe822240 TH |
6662 | #ifdef CONFIG_RPS |
6663 | kfree(dev->_rx); | |
6664 | #endif | |
6665 | ||
ab9c73cc JP |
6666 | free_p: |
6667 | kfree(p); | |
6668 | return NULL; | |
1da177e4 | 6669 | } |
36909ea4 | 6670 | EXPORT_SYMBOL(alloc_netdev_mqs); |
1da177e4 LT |
6671 | |
6672 | /** | |
6673 | * free_netdev - free network device | |
6674 | * @dev: device | |
6675 | * | |
4ec93edb YH |
6676 | * This function does the last stage of destroying an allocated device |
6677 | * interface. The reference to the device object is released. | |
1da177e4 LT |
6678 | * If this is the last reference then it will be freed. |
6679 | */ | |
6680 | void free_netdev(struct net_device *dev) | |
6681 | { | |
d565b0a1 HX |
6682 | struct napi_struct *p, *n; |
6683 | ||
f3005d7f DL |
6684 | release_net(dev_net(dev)); |
6685 | ||
e8a0464c | 6686 | kfree(dev->_tx); |
fe822240 TH |
6687 | #ifdef CONFIG_RPS |
6688 | kfree(dev->_rx); | |
6689 | #endif | |
e8a0464c | 6690 | |
33d480ce | 6691 | kfree(rcu_dereference_protected(dev->ingress_queue, 1)); |
24824a09 | 6692 | |
f001fde5 JP |
6693 | /* Flush device addresses */ |
6694 | dev_addr_flush(dev); | |
6695 | ||
d565b0a1 HX |
6696 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
6697 | netif_napi_del(p); | |
6698 | ||
29b4433d ED |
6699 | free_percpu(dev->pcpu_refcnt); |
6700 | dev->pcpu_refcnt = NULL; | |
6701 | ||
3041a069 | 6702 | /* Compatibility with error handling in drivers */ |
1da177e4 LT |
6703 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
6704 | kfree((char *)dev - dev->padded); | |
6705 | return; | |
6706 | } | |
6707 | ||
6708 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
6709 | dev->reg_state = NETREG_RELEASED; | |
6710 | ||
43cb76d9 GKH |
6711 | /* will free via device release */ |
6712 | put_device(&dev->dev); | |
1da177e4 | 6713 | } |
d1b19dff | 6714 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 6715 | |
f0db275a SH |
6716 | /** |
6717 | * synchronize_net - Synchronize with packet receive processing | |
6718 | * | |
6719 | * Wait for packets currently being received to be done. | |
6720 | * Does not block later packets from starting. | |
6721 | */ | |
4ec93edb | 6722 | void synchronize_net(void) |
1da177e4 LT |
6723 | { |
6724 | might_sleep(); | |
be3fc413 ED |
6725 | if (rtnl_is_locked()) |
6726 | synchronize_rcu_expedited(); | |
6727 | else | |
6728 | synchronize_rcu(); | |
1da177e4 | 6729 | } |
d1b19dff | 6730 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
6731 | |
6732 | /** | |
44a0873d | 6733 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 6734 | * @dev: device |
44a0873d | 6735 | * @head: list |
6ebfbc06 | 6736 | * |
1da177e4 | 6737 | * This function shuts down a device interface and removes it |
d59b54b1 | 6738 | * from the kernel tables. |
44a0873d | 6739 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
6740 | * |
6741 | * Callers must hold the rtnl semaphore. You may want | |
6742 | * unregister_netdev() instead of this. | |
6743 | */ | |
6744 | ||
44a0873d | 6745 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 6746 | { |
a6620712 HX |
6747 | ASSERT_RTNL(); |
6748 | ||
44a0873d | 6749 | if (head) { |
9fdce099 | 6750 | list_move_tail(&dev->unreg_list, head); |
44a0873d ED |
6751 | } else { |
6752 | rollback_registered(dev); | |
6753 | /* Finish processing unregister after unlock */ | |
6754 | net_set_todo(dev); | |
6755 | } | |
1da177e4 | 6756 | } |
44a0873d | 6757 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 6758 | |
9b5e383c ED |
6759 | /** |
6760 | * unregister_netdevice_many - unregister many devices | |
6761 | * @head: list of devices | |
9b5e383c ED |
6762 | */ |
6763 | void unregister_netdevice_many(struct list_head *head) | |
6764 | { | |
6765 | struct net_device *dev; | |
6766 | ||
6767 | if (!list_empty(head)) { | |
6768 | rollback_registered_many(head); | |
6769 | list_for_each_entry(dev, head, unreg_list) | |
6770 | net_set_todo(dev); | |
6771 | } | |
6772 | } | |
63c8099d | 6773 | EXPORT_SYMBOL(unregister_netdevice_many); |
9b5e383c | 6774 | |
1da177e4 LT |
6775 | /** |
6776 | * unregister_netdev - remove device from the kernel | |
6777 | * @dev: device | |
6778 | * | |
6779 | * This function shuts down a device interface and removes it | |
d59b54b1 | 6780 | * from the kernel tables. |
1da177e4 LT |
6781 | * |
6782 | * This is just a wrapper for unregister_netdevice that takes | |
6783 | * the rtnl semaphore. In general you want to use this and not | |
6784 | * unregister_netdevice. | |
6785 | */ | |
6786 | void unregister_netdev(struct net_device *dev) | |
6787 | { | |
6788 | rtnl_lock(); | |
6789 | unregister_netdevice(dev); | |
6790 | rtnl_unlock(); | |
6791 | } | |
1da177e4 LT |
6792 | EXPORT_SYMBOL(unregister_netdev); |
6793 | ||
ce286d32 EB |
6794 | /** |
6795 | * dev_change_net_namespace - move device to different nethost namespace | |
6796 | * @dev: device | |
6797 | * @net: network namespace | |
6798 | * @pat: If not NULL name pattern to try if the current device name | |
6799 | * is already taken in the destination network namespace. | |
6800 | * | |
6801 | * This function shuts down a device interface and moves it | |
6802 | * to a new network namespace. On success 0 is returned, on | |
6803 | * a failure a netagive errno code is returned. | |
6804 | * | |
6805 | * Callers must hold the rtnl semaphore. | |
6806 | */ | |
6807 | ||
6808 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
6809 | { | |
ce286d32 EB |
6810 | int err; |
6811 | ||
6812 | ASSERT_RTNL(); | |
6813 | ||
6814 | /* Don't allow namespace local devices to be moved. */ | |
6815 | err = -EINVAL; | |
6816 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
6817 | goto out; | |
6818 | ||
6819 | /* Ensure the device has been registrered */ | |
ce286d32 EB |
6820 | if (dev->reg_state != NETREG_REGISTERED) |
6821 | goto out; | |
6822 | ||
6823 | /* Get out if there is nothing todo */ | |
6824 | err = 0; | |
878628fb | 6825 | if (net_eq(dev_net(dev), net)) |
ce286d32 EB |
6826 | goto out; |
6827 | ||
6828 | /* Pick the destination device name, and ensure | |
6829 | * we can use it in the destination network namespace. | |
6830 | */ | |
6831 | err = -EEXIST; | |
d9031024 | 6832 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
6833 | /* We get here if we can't use the current device name */ |
6834 | if (!pat) | |
6835 | goto out; | |
828de4f6 | 6836 | if (dev_get_valid_name(net, dev, pat) < 0) |
ce286d32 EB |
6837 | goto out; |
6838 | } | |
6839 | ||
6840 | /* | |
6841 | * And now a mini version of register_netdevice unregister_netdevice. | |
6842 | */ | |
6843 | ||
6844 | /* If device is running close it first. */ | |
9b772652 | 6845 | dev_close(dev); |
ce286d32 EB |
6846 | |
6847 | /* And unlink it from device chain */ | |
6848 | err = -ENODEV; | |
6849 | unlist_netdevice(dev); | |
6850 | ||
6851 | synchronize_net(); | |
6852 | ||
6853 | /* Shutdown queueing discipline. */ | |
6854 | dev_shutdown(dev); | |
6855 | ||
6856 | /* Notify protocols, that we are about to destroy | |
6857 | this device. They should clean all the things. | |
3b27e105 DL |
6858 | |
6859 | Note that dev->reg_state stays at NETREG_REGISTERED. | |
6860 | This is wanted because this way 8021q and macvlan know | |
6861 | the device is just moving and can keep their slaves up. | |
ce286d32 EB |
6862 | */ |
6863 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
6549dd43 G |
6864 | rcu_barrier(); |
6865 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); | |
d2237d35 | 6866 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U); |
ce286d32 EB |
6867 | |
6868 | /* | |
6869 | * Flush the unicast and multicast chains | |
6870 | */ | |
a748ee24 | 6871 | dev_uc_flush(dev); |
22bedad3 | 6872 | dev_mc_flush(dev); |
ce286d32 | 6873 | |
4e66ae2e SH |
6874 | /* Send a netdev-removed uevent to the old namespace */ |
6875 | kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); | |
6876 | ||
ce286d32 | 6877 | /* Actually switch the network namespace */ |
c346dca1 | 6878 | dev_net_set(dev, net); |
ce286d32 | 6879 | |
ce286d32 EB |
6880 | /* If there is an ifindex conflict assign a new one */ |
6881 | if (__dev_get_by_index(net, dev->ifindex)) { | |
6882 | int iflink = (dev->iflink == dev->ifindex); | |
6883 | dev->ifindex = dev_new_index(net); | |
6884 | if (iflink) | |
6885 | dev->iflink = dev->ifindex; | |
6886 | } | |
6887 | ||
4e66ae2e SH |
6888 | /* Send a netdev-add uevent to the new namespace */ |
6889 | kobject_uevent(&dev->dev.kobj, KOBJ_ADD); | |
6890 | ||
8b41d188 | 6891 | /* Fixup kobjects */ |
a1b3f594 | 6892 | err = device_rename(&dev->dev, dev->name); |
8b41d188 | 6893 | WARN_ON(err); |
ce286d32 EB |
6894 | |
6895 | /* Add the device back in the hashes */ | |
6896 | list_netdevice(dev); | |
6897 | ||
6898 | /* Notify protocols, that a new device appeared. */ | |
6899 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
6900 | ||
d90a909e EB |
6901 | /* |
6902 | * Prevent userspace races by waiting until the network | |
6903 | * device is fully setup before sending notifications. | |
6904 | */ | |
6905 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U); | |
6906 | ||
ce286d32 EB |
6907 | synchronize_net(); |
6908 | err = 0; | |
6909 | out: | |
6910 | return err; | |
6911 | } | |
463d0183 | 6912 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 6913 | |
1da177e4 LT |
6914 | static int dev_cpu_callback(struct notifier_block *nfb, |
6915 | unsigned long action, | |
6916 | void *ocpu) | |
6917 | { | |
6918 | struct sk_buff **list_skb; | |
1da177e4 LT |
6919 | struct sk_buff *skb; |
6920 | unsigned int cpu, oldcpu = (unsigned long)ocpu; | |
6921 | struct softnet_data *sd, *oldsd; | |
6922 | ||
8bb78442 | 6923 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
1da177e4 LT |
6924 | return NOTIFY_OK; |
6925 | ||
6926 | local_irq_disable(); | |
6927 | cpu = smp_processor_id(); | |
6928 | sd = &per_cpu(softnet_data, cpu); | |
6929 | oldsd = &per_cpu(softnet_data, oldcpu); | |
6930 | ||
6931 | /* Find end of our completion_queue. */ | |
6932 | list_skb = &sd->completion_queue; | |
6933 | while (*list_skb) | |
6934 | list_skb = &(*list_skb)->next; | |
6935 | /* Append completion queue from offline CPU. */ | |
6936 | *list_skb = oldsd->completion_queue; | |
6937 | oldsd->completion_queue = NULL; | |
6938 | ||
1da177e4 | 6939 | /* Append output queue from offline CPU. */ |
a9cbd588 CG |
6940 | if (oldsd->output_queue) { |
6941 | *sd->output_queue_tailp = oldsd->output_queue; | |
6942 | sd->output_queue_tailp = oldsd->output_queue_tailp; | |
6943 | oldsd->output_queue = NULL; | |
6944 | oldsd->output_queue_tailp = &oldsd->output_queue; | |
6945 | } | |
264524d5 HC |
6946 | /* Append NAPI poll list from offline CPU. */ |
6947 | if (!list_empty(&oldsd->poll_list)) { | |
6948 | list_splice_init(&oldsd->poll_list, &sd->poll_list); | |
6949 | raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
6950 | } | |
1da177e4 LT |
6951 | |
6952 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
6953 | local_irq_enable(); | |
6954 | ||
6955 | /* Process offline CPU's input_pkt_queue */ | |
76cc8b13 | 6956 | while ((skb = __skb_dequeue(&oldsd->process_queue))) { |
1da177e4 | 6957 | netif_rx(skb); |
76cc8b13 | 6958 | input_queue_head_incr(oldsd); |
fec5e652 | 6959 | } |
76cc8b13 | 6960 | while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) { |
6e7676c1 | 6961 | netif_rx(skb); |
76cc8b13 TH |
6962 | input_queue_head_incr(oldsd); |
6963 | } | |
1da177e4 LT |
6964 | |
6965 | return NOTIFY_OK; | |
6966 | } | |
1da177e4 LT |
6967 | |
6968 | ||
7f353bf2 | 6969 | /** |
b63365a2 HX |
6970 | * netdev_increment_features - increment feature set by one |
6971 | * @all: current feature set | |
6972 | * @one: new feature set | |
6973 | * @mask: mask feature set | |
7f353bf2 HX |
6974 | * |
6975 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
6976 | * @one to the master device with current feature set @all. Will not |
6977 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 6978 | */ |
c8f44aff MM |
6979 | netdev_features_t netdev_increment_features(netdev_features_t all, |
6980 | netdev_features_t one, netdev_features_t mask) | |
b63365a2 | 6981 | { |
1742f183 MM |
6982 | if (mask & NETIF_F_GEN_CSUM) |
6983 | mask |= NETIF_F_ALL_CSUM; | |
6984 | mask |= NETIF_F_VLAN_CHALLENGED; | |
7f353bf2 | 6985 | |
1742f183 MM |
6986 | all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask; |
6987 | all &= one | ~NETIF_F_ALL_FOR_ALL; | |
c6e1a0d1 | 6988 | |
1742f183 MM |
6989 | /* If one device supports hw checksumming, set for all. */ |
6990 | if (all & NETIF_F_GEN_CSUM) | |
6991 | all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM); | |
7f353bf2 HX |
6992 | |
6993 | return all; | |
6994 | } | |
b63365a2 | 6995 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 6996 | |
30d97d35 PE |
6997 | static struct hlist_head *netdev_create_hash(void) |
6998 | { | |
6999 | int i; | |
7000 | struct hlist_head *hash; | |
7001 | ||
7002 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
7003 | if (hash != NULL) | |
7004 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
7005 | INIT_HLIST_HEAD(&hash[i]); | |
7006 | ||
7007 | return hash; | |
7008 | } | |
7009 | ||
881d966b | 7010 | /* Initialize per network namespace state */ |
4665079c | 7011 | static int __net_init netdev_init(struct net *net) |
881d966b | 7012 | { |
734b6541 RM |
7013 | if (net != &init_net) |
7014 | INIT_LIST_HEAD(&net->dev_base_head); | |
881d966b | 7015 | |
30d97d35 PE |
7016 | net->dev_name_head = netdev_create_hash(); |
7017 | if (net->dev_name_head == NULL) | |
7018 | goto err_name; | |
881d966b | 7019 | |
30d97d35 PE |
7020 | net->dev_index_head = netdev_create_hash(); |
7021 | if (net->dev_index_head == NULL) | |
7022 | goto err_idx; | |
881d966b EB |
7023 | |
7024 | return 0; | |
30d97d35 PE |
7025 | |
7026 | err_idx: | |
7027 | kfree(net->dev_name_head); | |
7028 | err_name: | |
7029 | return -ENOMEM; | |
881d966b EB |
7030 | } |
7031 | ||
f0db275a SH |
7032 | /** |
7033 | * netdev_drivername - network driver for the device | |
7034 | * @dev: network device | |
f0db275a SH |
7035 | * |
7036 | * Determine network driver for device. | |
7037 | */ | |
3019de12 | 7038 | const char *netdev_drivername(const struct net_device *dev) |
6579e57b | 7039 | { |
cf04a4c7 SH |
7040 | const struct device_driver *driver; |
7041 | const struct device *parent; | |
3019de12 | 7042 | const char *empty = ""; |
6579e57b AV |
7043 | |
7044 | parent = dev->dev.parent; | |
6579e57b | 7045 | if (!parent) |
3019de12 | 7046 | return empty; |
6579e57b AV |
7047 | |
7048 | driver = parent->driver; | |
7049 | if (driver && driver->name) | |
3019de12 DM |
7050 | return driver->name; |
7051 | return empty; | |
6579e57b AV |
7052 | } |
7053 | ||
b004ff49 | 7054 | static int __netdev_printk(const char *level, const struct net_device *dev, |
256df2f3 JP |
7055 | struct va_format *vaf) |
7056 | { | |
7057 | int r; | |
7058 | ||
b004ff49 | 7059 | if (dev && dev->dev.parent) { |
666f355f JP |
7060 | r = dev_printk_emit(level[1] - '0', |
7061 | dev->dev.parent, | |
7062 | "%s %s %s: %pV", | |
7063 | dev_driver_string(dev->dev.parent), | |
7064 | dev_name(dev->dev.parent), | |
7065 | netdev_name(dev), vaf); | |
b004ff49 | 7066 | } else if (dev) { |
256df2f3 | 7067 | r = printk("%s%s: %pV", level, netdev_name(dev), vaf); |
b004ff49 | 7068 | } else { |
256df2f3 | 7069 | r = printk("%s(NULL net_device): %pV", level, vaf); |
b004ff49 | 7070 | } |
256df2f3 JP |
7071 | |
7072 | return r; | |
7073 | } | |
7074 | ||
7075 | int netdev_printk(const char *level, const struct net_device *dev, | |
7076 | const char *format, ...) | |
7077 | { | |
7078 | struct va_format vaf; | |
7079 | va_list args; | |
7080 | int r; | |
7081 | ||
7082 | va_start(args, format); | |
7083 | ||
7084 | vaf.fmt = format; | |
7085 | vaf.va = &args; | |
7086 | ||
7087 | r = __netdev_printk(level, dev, &vaf); | |
b004ff49 | 7088 | |
256df2f3 JP |
7089 | va_end(args); |
7090 | ||
7091 | return r; | |
7092 | } | |
7093 | EXPORT_SYMBOL(netdev_printk); | |
7094 | ||
7095 | #define define_netdev_printk_level(func, level) \ | |
7096 | int func(const struct net_device *dev, const char *fmt, ...) \ | |
7097 | { \ | |
7098 | int r; \ | |
7099 | struct va_format vaf; \ | |
7100 | va_list args; \ | |
7101 | \ | |
7102 | va_start(args, fmt); \ | |
7103 | \ | |
7104 | vaf.fmt = fmt; \ | |
7105 | vaf.va = &args; \ | |
7106 | \ | |
7107 | r = __netdev_printk(level, dev, &vaf); \ | |
b004ff49 | 7108 | \ |
256df2f3 JP |
7109 | va_end(args); \ |
7110 | \ | |
7111 | return r; \ | |
7112 | } \ | |
7113 | EXPORT_SYMBOL(func); | |
7114 | ||
7115 | define_netdev_printk_level(netdev_emerg, KERN_EMERG); | |
7116 | define_netdev_printk_level(netdev_alert, KERN_ALERT); | |
7117 | define_netdev_printk_level(netdev_crit, KERN_CRIT); | |
7118 | define_netdev_printk_level(netdev_err, KERN_ERR); | |
7119 | define_netdev_printk_level(netdev_warn, KERN_WARNING); | |
7120 | define_netdev_printk_level(netdev_notice, KERN_NOTICE); | |
7121 | define_netdev_printk_level(netdev_info, KERN_INFO); | |
7122 | ||
4665079c | 7123 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
7124 | { |
7125 | kfree(net->dev_name_head); | |
7126 | kfree(net->dev_index_head); | |
7127 | } | |
7128 | ||
022cbae6 | 7129 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
7130 | .init = netdev_init, |
7131 | .exit = netdev_exit, | |
7132 | }; | |
7133 | ||
4665079c | 7134 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 7135 | { |
e008b5fc | 7136 | struct net_device *dev, *aux; |
ce286d32 | 7137 | /* |
e008b5fc | 7138 | * Push all migratable network devices back to the |
ce286d32 EB |
7139 | * initial network namespace |
7140 | */ | |
7141 | rtnl_lock(); | |
e008b5fc | 7142 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 7143 | int err; |
aca51397 | 7144 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
7145 | |
7146 | /* Ignore unmoveable devices (i.e. loopback) */ | |
7147 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
7148 | continue; | |
7149 | ||
e008b5fc EB |
7150 | /* Leave virtual devices for the generic cleanup */ |
7151 | if (dev->rtnl_link_ops) | |
7152 | continue; | |
d0c082ce | 7153 | |
25985edc | 7154 | /* Push remaining network devices to init_net */ |
aca51397 PE |
7155 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
7156 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
ce286d32 | 7157 | if (err) { |
7b6cd1ce JP |
7158 | pr_emerg("%s: failed to move %s to init_net: %d\n", |
7159 | __func__, dev->name, err); | |
aca51397 | 7160 | BUG(); |
ce286d32 EB |
7161 | } |
7162 | } | |
7163 | rtnl_unlock(); | |
7164 | } | |
7165 | ||
04dc7f6b EB |
7166 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
7167 | { | |
7168 | /* At exit all network devices most be removed from a network | |
b595076a | 7169 | * namespace. Do this in the reverse order of registration. |
04dc7f6b EB |
7170 | * Do this across as many network namespaces as possible to |
7171 | * improve batching efficiency. | |
7172 | */ | |
7173 | struct net_device *dev; | |
7174 | struct net *net; | |
7175 | LIST_HEAD(dev_kill_list); | |
7176 | ||
7177 | rtnl_lock(); | |
7178 | list_for_each_entry(net, net_list, exit_list) { | |
7179 | for_each_netdev_reverse(net, dev) { | |
7180 | if (dev->rtnl_link_ops) | |
7181 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); | |
7182 | else | |
7183 | unregister_netdevice_queue(dev, &dev_kill_list); | |
7184 | } | |
7185 | } | |
7186 | unregister_netdevice_many(&dev_kill_list); | |
ceaaec98 | 7187 | list_del(&dev_kill_list); |
04dc7f6b EB |
7188 | rtnl_unlock(); |
7189 | } | |
7190 | ||
022cbae6 | 7191 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 7192 | .exit = default_device_exit, |
04dc7f6b | 7193 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
7194 | }; |
7195 | ||
1da177e4 LT |
7196 | /* |
7197 | * Initialize the DEV module. At boot time this walks the device list and | |
7198 | * unhooks any devices that fail to initialise (normally hardware not | |
7199 | * present) and leaves us with a valid list of present and active devices. | |
7200 | * | |
7201 | */ | |
7202 | ||
7203 | /* | |
7204 | * This is called single threaded during boot, so no need | |
7205 | * to take the rtnl semaphore. | |
7206 | */ | |
7207 | static int __init net_dev_init(void) | |
7208 | { | |
7209 | int i, rc = -ENOMEM; | |
7210 | ||
7211 | BUG_ON(!dev_boot_phase); | |
7212 | ||
1da177e4 LT |
7213 | if (dev_proc_init()) |
7214 | goto out; | |
7215 | ||
8b41d188 | 7216 | if (netdev_kobject_init()) |
1da177e4 LT |
7217 | goto out; |
7218 | ||
7219 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 7220 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
7221 | INIT_LIST_HEAD(&ptype_base[i]); |
7222 | ||
62532da9 VY |
7223 | INIT_LIST_HEAD(&offload_base); |
7224 | ||
881d966b EB |
7225 | if (register_pernet_subsys(&netdev_net_ops)) |
7226 | goto out; | |
1da177e4 LT |
7227 | |
7228 | /* | |
7229 | * Initialise the packet receive queues. | |
7230 | */ | |
7231 | ||
6f912042 | 7232 | for_each_possible_cpu(i) { |
e36fa2f7 | 7233 | struct softnet_data *sd = &per_cpu(softnet_data, i); |
1da177e4 | 7234 | |
dee42870 | 7235 | memset(sd, 0, sizeof(*sd)); |
e36fa2f7 | 7236 | skb_queue_head_init(&sd->input_pkt_queue); |
6e7676c1 | 7237 | skb_queue_head_init(&sd->process_queue); |
e36fa2f7 ED |
7238 | sd->completion_queue = NULL; |
7239 | INIT_LIST_HEAD(&sd->poll_list); | |
a9cbd588 CG |
7240 | sd->output_queue = NULL; |
7241 | sd->output_queue_tailp = &sd->output_queue; | |
df334545 | 7242 | #ifdef CONFIG_RPS |
e36fa2f7 ED |
7243 | sd->csd.func = rps_trigger_softirq; |
7244 | sd->csd.info = sd; | |
7245 | sd->csd.flags = 0; | |
7246 | sd->cpu = i; | |
1e94d72f | 7247 | #endif |
0a9627f2 | 7248 | |
e36fa2f7 ED |
7249 | sd->backlog.poll = process_backlog; |
7250 | sd->backlog.weight = weight_p; | |
7251 | sd->backlog.gro_list = NULL; | |
7252 | sd->backlog.gro_count = 0; | |
1da177e4 LT |
7253 | } |
7254 | ||
1da177e4 LT |
7255 | dev_boot_phase = 0; |
7256 | ||
505d4f73 EB |
7257 | /* The loopback device is special if any other network devices |
7258 | * is present in a network namespace the loopback device must | |
7259 | * be present. Since we now dynamically allocate and free the | |
7260 | * loopback device ensure this invariant is maintained by | |
7261 | * keeping the loopback device as the first device on the | |
7262 | * list of network devices. Ensuring the loopback devices | |
7263 | * is the first device that appears and the last network device | |
7264 | * that disappears. | |
7265 | */ | |
7266 | if (register_pernet_device(&loopback_net_ops)) | |
7267 | goto out; | |
7268 | ||
7269 | if (register_pernet_device(&default_device_ops)) | |
7270 | goto out; | |
7271 | ||
962cf36c CM |
7272 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
7273 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 LT |
7274 | |
7275 | hotcpu_notifier(dev_cpu_callback, 0); | |
7276 | dst_init(); | |
7277 | dev_mcast_init(); | |
7278 | rc = 0; | |
7279 | out: | |
7280 | return rc; | |
7281 | } | |
7282 | ||
7283 | subsys_initcall(net_dev_init); | |
7284 | ||
e88721f8 KK |
7285 | static int __init initialize_hashrnd(void) |
7286 | { | |
0a9627f2 | 7287 | get_random_bytes(&hashrnd, sizeof(hashrnd)); |
e88721f8 KK |
7288 | return 0; |
7289 | } | |
7290 | ||
7291 | late_initcall_sync(initialize_hashrnd); | |
7292 |