]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * NET3 Protocol independent device support routines. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public License | |
6 | * as published by the Free Software Foundation; either version | |
7 | * 2 of the License, or (at your option) any later version. | |
8 | * | |
9 | * Derived from the non IP parts of dev.c 1.0.19 | |
02c30a84 | 10 | * Authors: Ross Biro |
1da177e4 LT |
11 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
12 | * Mark Evans, <evansmp@uhura.aston.ac.uk> | |
13 | * | |
14 | * Additional Authors: | |
15 | * Florian la Roche <rzsfl@rz.uni-sb.de> | |
16 | * Alan Cox <gw4pts@gw4pts.ampr.org> | |
17 | * David Hinds <dahinds@users.sourceforge.net> | |
18 | * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> | |
19 | * Adam Sulmicki <adam@cfar.umd.edu> | |
20 | * Pekka Riikonen <priikone@poesidon.pspt.fi> | |
21 | * | |
22 | * Changes: | |
23 | * D.J. Barrow : Fixed bug where dev->refcnt gets set | |
24 | * to 2 if register_netdev gets called | |
25 | * before net_dev_init & also removed a | |
26 | * few lines of code in the process. | |
27 | * Alan Cox : device private ioctl copies fields back. | |
28 | * Alan Cox : Transmit queue code does relevant | |
29 | * stunts to keep the queue safe. | |
30 | * Alan Cox : Fixed double lock. | |
31 | * Alan Cox : Fixed promisc NULL pointer trap | |
32 | * ???????? : Support the full private ioctl range | |
33 | * Alan Cox : Moved ioctl permission check into | |
34 | * drivers | |
35 | * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI | |
36 | * Alan Cox : 100 backlog just doesn't cut it when | |
37 | * you start doing multicast video 8) | |
38 | * Alan Cox : Rewrote net_bh and list manager. | |
39 | * Alan Cox : Fix ETH_P_ALL echoback lengths. | |
40 | * Alan Cox : Took out transmit every packet pass | |
41 | * Saved a few bytes in the ioctl handler | |
42 | * Alan Cox : Network driver sets packet type before | |
43 | * calling netif_rx. Saves a function | |
44 | * call a packet. | |
45 | * Alan Cox : Hashed net_bh() | |
46 | * Richard Kooijman: Timestamp fixes. | |
47 | * Alan Cox : Wrong field in SIOCGIFDSTADDR | |
48 | * Alan Cox : Device lock protection. | |
49 | * Alan Cox : Fixed nasty side effect of device close | |
50 | * changes. | |
51 | * Rudi Cilibrasi : Pass the right thing to | |
52 | * set_mac_address() | |
53 | * Dave Miller : 32bit quantity for the device lock to | |
54 | * make it work out on a Sparc. | |
55 | * Bjorn Ekwall : Added KERNELD hack. | |
56 | * Alan Cox : Cleaned up the backlog initialise. | |
57 | * Craig Metz : SIOCGIFCONF fix if space for under | |
58 | * 1 device. | |
59 | * Thomas Bogendoerfer : Return ENODEV for dev_open, if there | |
60 | * is no device open function. | |
61 | * Andi Kleen : Fix error reporting for SIOCGIFCONF | |
62 | * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF | |
63 | * Cyrus Durgin : Cleaned for KMOD | |
64 | * Adam Sulmicki : Bug Fix : Network Device Unload | |
65 | * A network device unload needs to purge | |
66 | * the backlog queue. | |
67 | * Paul Rusty Russell : SIOCSIFNAME | |
68 | * Pekka Riikonen : Netdev boot-time settings code | |
69 | * Andrew Morton : Make unregister_netdevice wait | |
70 | * indefinitely on dev->refcnt | |
71 | * J Hadi Salim : - Backlog queue sampling | |
72 | * - netif_rx() feedback | |
73 | */ | |
74 | ||
75 | #include <asm/uaccess.h> | |
1da177e4 | 76 | #include <linux/bitops.h> |
4fc268d2 | 77 | #include <linux/capability.h> |
1da177e4 LT |
78 | #include <linux/cpu.h> |
79 | #include <linux/types.h> | |
80 | #include <linux/kernel.h> | |
08e9897d | 81 | #include <linux/hash.h> |
5a0e3ad6 | 82 | #include <linux/slab.h> |
1da177e4 | 83 | #include <linux/sched.h> |
4a3e2f71 | 84 | #include <linux/mutex.h> |
1da177e4 LT |
85 | #include <linux/string.h> |
86 | #include <linux/mm.h> | |
87 | #include <linux/socket.h> | |
88 | #include <linux/sockios.h> | |
89 | #include <linux/errno.h> | |
90 | #include <linux/interrupt.h> | |
91 | #include <linux/if_ether.h> | |
92 | #include <linux/netdevice.h> | |
93 | #include <linux/etherdevice.h> | |
0187bdfb | 94 | #include <linux/ethtool.h> |
1da177e4 LT |
95 | #include <linux/notifier.h> |
96 | #include <linux/skbuff.h> | |
457c4cbc | 97 | #include <net/net_namespace.h> |
1da177e4 LT |
98 | #include <net/sock.h> |
99 | #include <linux/rtnetlink.h> | |
100 | #include <linux/proc_fs.h> | |
101 | #include <linux/seq_file.h> | |
102 | #include <linux/stat.h> | |
1da177e4 LT |
103 | #include <net/dst.h> |
104 | #include <net/pkt_sched.h> | |
105 | #include <net/checksum.h> | |
44540960 | 106 | #include <net/xfrm.h> |
1da177e4 LT |
107 | #include <linux/highmem.h> |
108 | #include <linux/init.h> | |
109 | #include <linux/kmod.h> | |
110 | #include <linux/module.h> | |
1da177e4 LT |
111 | #include <linux/netpoll.h> |
112 | #include <linux/rcupdate.h> | |
113 | #include <linux/delay.h> | |
295f4a1f | 114 | #include <net/wext.h> |
1da177e4 | 115 | #include <net/iw_handler.h> |
1da177e4 | 116 | #include <asm/current.h> |
5bdb9886 | 117 | #include <linux/audit.h> |
db217334 | 118 | #include <linux/dmaengine.h> |
f6a78bfc | 119 | #include <linux/err.h> |
c7fa9d18 | 120 | #include <linux/ctype.h> |
723e98b7 | 121 | #include <linux/if_arp.h> |
6de329e2 | 122 | #include <linux/if_vlan.h> |
8f0f2223 | 123 | #include <linux/ip.h> |
ad55dcaf | 124 | #include <net/ip.h> |
8f0f2223 DM |
125 | #include <linux/ipv6.h> |
126 | #include <linux/in.h> | |
b6b2fed1 DM |
127 | #include <linux/jhash.h> |
128 | #include <linux/random.h> | |
9cbc1cb8 | 129 | #include <trace/events/napi.h> |
cf66ba58 | 130 | #include <trace/events/net.h> |
07dc22e7 | 131 | #include <trace/events/skb.h> |
5acbbd42 | 132 | #include <linux/pci.h> |
caeda9b9 | 133 | #include <linux/inetdevice.h> |
c445477d | 134 | #include <linux/cpu_rmap.h> |
4dc360c5 | 135 | #include <linux/net_tstamp.h> |
c5905afb | 136 | #include <linux/static_key.h> |
4504b861 | 137 | #include <net/flow_keys.h> |
1da177e4 | 138 | |
342709ef PE |
139 | #include "net-sysfs.h" |
140 | ||
d565b0a1 HX |
141 | /* Instead of increasing this, you should create a hash table. */ |
142 | #define MAX_GRO_SKBS 8 | |
143 | ||
5d38a079 HX |
144 | /* This should be increased if a protocol with a bigger head is added. */ |
145 | #define GRO_MAX_HEAD (MAX_HEADER + 128) | |
146 | ||
1da177e4 LT |
147 | /* |
148 | * The list of packet types we will receive (as opposed to discard) | |
149 | * and the routines to invoke. | |
150 | * | |
151 | * Why 16. Because with 16 the only overlap we get on a hash of the | |
152 | * low nibble of the protocol value is RARP/SNAP/X.25. | |
153 | * | |
154 | * NOTE: That is no longer true with the addition of VLAN tags. Not | |
155 | * sure which should go first, but I bet it won't make much | |
156 | * difference if we are running VLANs. The good news is that | |
157 | * this protocol won't be in the list unless compiled in, so | |
3041a069 | 158 | * the average user (w/out VLANs) will not be adversely affected. |
1da177e4 LT |
159 | * --BLG |
160 | * | |
161 | * 0800 IP | |
162 | * 8100 802.1Q VLAN | |
163 | * 0001 802.3 | |
164 | * 0002 AX.25 | |
165 | * 0004 802.2 | |
166 | * 8035 RARP | |
167 | * 0005 SNAP | |
168 | * 0805 X.25 | |
169 | * 0806 ARP | |
170 | * 8137 IPX | |
171 | * 0009 Localtalk | |
172 | * 86DD IPv6 | |
173 | */ | |
174 | ||
82d8a867 PE |
175 | #define PTYPE_HASH_SIZE (16) |
176 | #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1) | |
177 | ||
1da177e4 | 178 | static DEFINE_SPINLOCK(ptype_lock); |
82d8a867 | 179 | static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
6b2bedc3 | 180 | static struct list_head ptype_all __read_mostly; /* Taps */ |
1da177e4 | 181 | |
1da177e4 | 182 | /* |
7562f876 | 183 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
184 | * semaphore. |
185 | * | |
c6d14c84 | 186 | * Pure readers hold dev_base_lock for reading, or rcu_read_lock() |
1da177e4 LT |
187 | * |
188 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 189 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
190 | * actual updates. This allows pure readers to access the list even |
191 | * while a writer is preparing to update it. | |
192 | * | |
193 | * To put it another way, dev_base_lock is held for writing only to | |
194 | * protect against pure readers; the rtnl semaphore provides the | |
195 | * protection against other writers. | |
196 | * | |
197 | * See, for example usages, register_netdevice() and | |
198 | * unregister_netdevice(), which must be called with the rtnl | |
199 | * semaphore held. | |
200 | */ | |
1da177e4 | 201 | DEFINE_RWLOCK(dev_base_lock); |
1da177e4 LT |
202 | EXPORT_SYMBOL(dev_base_lock); |
203 | ||
4e985ada TG |
204 | static inline void dev_base_seq_inc(struct net *net) |
205 | { | |
206 | while (++net->dev_base_seq == 0); | |
207 | } | |
208 | ||
881d966b | 209 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 | 210 | { |
95c96174 ED |
211 | unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ)); |
212 | ||
08e9897d | 213 | return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; |
1da177e4 LT |
214 | } |
215 | ||
881d966b | 216 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 217 | { |
7c28bd0b | 218 | return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; |
1da177e4 LT |
219 | } |
220 | ||
e36fa2f7 | 221 | static inline void rps_lock(struct softnet_data *sd) |
152102c7 CG |
222 | { |
223 | #ifdef CONFIG_RPS | |
e36fa2f7 | 224 | spin_lock(&sd->input_pkt_queue.lock); |
152102c7 CG |
225 | #endif |
226 | } | |
227 | ||
e36fa2f7 | 228 | static inline void rps_unlock(struct softnet_data *sd) |
152102c7 CG |
229 | { |
230 | #ifdef CONFIG_RPS | |
e36fa2f7 | 231 | spin_unlock(&sd->input_pkt_queue.lock); |
152102c7 CG |
232 | #endif |
233 | } | |
234 | ||
ce286d32 EB |
235 | /* Device list insertion */ |
236 | static int list_netdevice(struct net_device *dev) | |
237 | { | |
c346dca1 | 238 | struct net *net = dev_net(dev); |
ce286d32 EB |
239 | |
240 | ASSERT_RTNL(); | |
241 | ||
242 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 243 | list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); |
72c9528b | 244 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); |
fb699dfd ED |
245 | hlist_add_head_rcu(&dev->index_hlist, |
246 | dev_index_hash(net, dev->ifindex)); | |
ce286d32 | 247 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
248 | |
249 | dev_base_seq_inc(net); | |
250 | ||
ce286d32 EB |
251 | return 0; |
252 | } | |
253 | ||
fb699dfd ED |
254 | /* Device list removal |
255 | * caller must respect a RCU grace period before freeing/reusing dev | |
256 | */ | |
ce286d32 EB |
257 | static void unlist_netdevice(struct net_device *dev) |
258 | { | |
259 | ASSERT_RTNL(); | |
260 | ||
261 | /* Unlink dev from the device chain */ | |
262 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 263 | list_del_rcu(&dev->dev_list); |
72c9528b | 264 | hlist_del_rcu(&dev->name_hlist); |
fb699dfd | 265 | hlist_del_rcu(&dev->index_hlist); |
ce286d32 | 266 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
267 | |
268 | dev_base_seq_inc(dev_net(dev)); | |
ce286d32 EB |
269 | } |
270 | ||
1da177e4 LT |
271 | /* |
272 | * Our notifier list | |
273 | */ | |
274 | ||
f07d5b94 | 275 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
276 | |
277 | /* | |
278 | * Device drivers call our routines to queue packets here. We empty the | |
279 | * queue in the local softnet handler. | |
280 | */ | |
bea3348e | 281 | |
9958da05 | 282 | DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); |
d1b19dff | 283 | EXPORT_PER_CPU_SYMBOL(softnet_data); |
1da177e4 | 284 | |
cf508b12 | 285 | #ifdef CONFIG_LOCKDEP |
723e98b7 | 286 | /* |
c773e847 | 287 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class |
723e98b7 JP |
288 | * according to dev->type |
289 | */ | |
290 | static const unsigned short netdev_lock_type[] = | |
291 | {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
292 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, | |
293 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
294 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
295 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
296 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
297 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
298 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
299 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
300 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
301 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
302 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
211ed865 PG |
303 | ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, |
304 | ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, | |
305 | ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; | |
723e98b7 | 306 | |
36cbd3dc | 307 | static const char *const netdev_lock_name[] = |
723e98b7 JP |
308 | {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", |
309 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
310 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
311 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
312 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
313 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
314 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
315 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
316 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
317 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
318 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
319 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
211ed865 PG |
320 | "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", |
321 | "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", | |
322 | "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; | |
723e98b7 JP |
323 | |
324 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
cf508b12 | 325 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; |
723e98b7 JP |
326 | |
327 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
328 | { | |
329 | int i; | |
330 | ||
331 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
332 | if (netdev_lock_type[i] == dev_type) | |
333 | return i; | |
334 | /* the last key is used by default */ | |
335 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
336 | } | |
337 | ||
cf508b12 DM |
338 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
339 | unsigned short dev_type) | |
723e98b7 JP |
340 | { |
341 | int i; | |
342 | ||
343 | i = netdev_lock_pos(dev_type); | |
344 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
345 | netdev_lock_name[i]); | |
346 | } | |
cf508b12 DM |
347 | |
348 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
349 | { | |
350 | int i; | |
351 | ||
352 | i = netdev_lock_pos(dev->type); | |
353 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
354 | &netdev_addr_lock_key[i], | |
355 | netdev_lock_name[i]); | |
356 | } | |
723e98b7 | 357 | #else |
cf508b12 DM |
358 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
359 | unsigned short dev_type) | |
360 | { | |
361 | } | |
362 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
723e98b7 JP |
363 | { |
364 | } | |
365 | #endif | |
1da177e4 LT |
366 | |
367 | /******************************************************************************* | |
368 | ||
369 | Protocol management and registration routines | |
370 | ||
371 | *******************************************************************************/ | |
372 | ||
1da177e4 LT |
373 | /* |
374 | * Add a protocol ID to the list. Now that the input handler is | |
375 | * smarter we can dispense with all the messy stuff that used to be | |
376 | * here. | |
377 | * | |
378 | * BEWARE!!! Protocol handlers, mangling input packets, | |
379 | * MUST BE last in hash buckets and checking protocol handlers | |
380 | * MUST start from promiscuous ptype_all chain in net_bh. | |
381 | * It is true now, do not change it. | |
382 | * Explanation follows: if protocol handler, mangling packet, will | |
383 | * be the first on list, it is not able to sense, that packet | |
384 | * is cloned and should be copied-on-write, so that it will | |
385 | * change it and subsequent readers will get broken packet. | |
386 | * --ANK (980803) | |
387 | */ | |
388 | ||
c07b68e8 ED |
389 | static inline struct list_head *ptype_head(const struct packet_type *pt) |
390 | { | |
391 | if (pt->type == htons(ETH_P_ALL)) | |
392 | return &ptype_all; | |
393 | else | |
394 | return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
395 | } | |
396 | ||
1da177e4 LT |
397 | /** |
398 | * dev_add_pack - add packet handler | |
399 | * @pt: packet type declaration | |
400 | * | |
401 | * Add a protocol handler to the networking stack. The passed &packet_type | |
402 | * is linked into kernel lists and may not be freed until it has been | |
403 | * removed from the kernel lists. | |
404 | * | |
4ec93edb | 405 | * This call does not sleep therefore it can not |
1da177e4 LT |
406 | * guarantee all CPU's that are in middle of receiving packets |
407 | * will see the new packet type (until the next received packet). | |
408 | */ | |
409 | ||
410 | void dev_add_pack(struct packet_type *pt) | |
411 | { | |
c07b68e8 | 412 | struct list_head *head = ptype_head(pt); |
1da177e4 | 413 | |
c07b68e8 ED |
414 | spin_lock(&ptype_lock); |
415 | list_add_rcu(&pt->list, head); | |
416 | spin_unlock(&ptype_lock); | |
1da177e4 | 417 | } |
d1b19dff | 418 | EXPORT_SYMBOL(dev_add_pack); |
1da177e4 | 419 | |
1da177e4 LT |
420 | /** |
421 | * __dev_remove_pack - remove packet handler | |
422 | * @pt: packet type declaration | |
423 | * | |
424 | * Remove a protocol handler that was previously added to the kernel | |
425 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
426 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 427 | * returns. |
1da177e4 LT |
428 | * |
429 | * The packet type might still be in use by receivers | |
430 | * and must not be freed until after all the CPU's have gone | |
431 | * through a quiescent state. | |
432 | */ | |
433 | void __dev_remove_pack(struct packet_type *pt) | |
434 | { | |
c07b68e8 | 435 | struct list_head *head = ptype_head(pt); |
1da177e4 LT |
436 | struct packet_type *pt1; |
437 | ||
c07b68e8 | 438 | spin_lock(&ptype_lock); |
1da177e4 LT |
439 | |
440 | list_for_each_entry(pt1, head, list) { | |
441 | if (pt == pt1) { | |
442 | list_del_rcu(&pt->list); | |
443 | goto out; | |
444 | } | |
445 | } | |
446 | ||
7b6cd1ce | 447 | pr_warn("dev_remove_pack: %p not found\n", pt); |
1da177e4 | 448 | out: |
c07b68e8 | 449 | spin_unlock(&ptype_lock); |
1da177e4 | 450 | } |
d1b19dff ED |
451 | EXPORT_SYMBOL(__dev_remove_pack); |
452 | ||
1da177e4 LT |
453 | /** |
454 | * dev_remove_pack - remove packet handler | |
455 | * @pt: packet type declaration | |
456 | * | |
457 | * Remove a protocol handler that was previously added to the kernel | |
458 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
459 | * from the kernel lists and can be freed or reused once this function | |
460 | * returns. | |
461 | * | |
462 | * This call sleeps to guarantee that no CPU is looking at the packet | |
463 | * type after return. | |
464 | */ | |
465 | void dev_remove_pack(struct packet_type *pt) | |
466 | { | |
467 | __dev_remove_pack(pt); | |
4ec93edb | 468 | |
1da177e4 LT |
469 | synchronize_net(); |
470 | } | |
d1b19dff | 471 | EXPORT_SYMBOL(dev_remove_pack); |
1da177e4 LT |
472 | |
473 | /****************************************************************************** | |
474 | ||
475 | Device Boot-time Settings Routines | |
476 | ||
477 | *******************************************************************************/ | |
478 | ||
479 | /* Boot time configuration table */ | |
480 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
481 | ||
482 | /** | |
483 | * netdev_boot_setup_add - add new setup entry | |
484 | * @name: name of the device | |
485 | * @map: configured settings for the device | |
486 | * | |
487 | * Adds new setup entry to the dev_boot_setup list. The function | |
488 | * returns 0 on error and 1 on success. This is a generic routine to | |
489 | * all netdevices. | |
490 | */ | |
491 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
492 | { | |
493 | struct netdev_boot_setup *s; | |
494 | int i; | |
495 | ||
496 | s = dev_boot_setup; | |
497 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
498 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
499 | memset(s[i].name, 0, sizeof(s[i].name)); | |
93b3cff9 | 500 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
501 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
502 | break; | |
503 | } | |
504 | } | |
505 | ||
506 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
507 | } | |
508 | ||
509 | /** | |
510 | * netdev_boot_setup_check - check boot time settings | |
511 | * @dev: the netdevice | |
512 | * | |
513 | * Check boot time settings for the device. | |
514 | * The found settings are set for the device to be used | |
515 | * later in the device probing. | |
516 | * Returns 0 if no settings found, 1 if they are. | |
517 | */ | |
518 | int netdev_boot_setup_check(struct net_device *dev) | |
519 | { | |
520 | struct netdev_boot_setup *s = dev_boot_setup; | |
521 | int i; | |
522 | ||
523 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
524 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
93b3cff9 | 525 | !strcmp(dev->name, s[i].name)) { |
1da177e4 LT |
526 | dev->irq = s[i].map.irq; |
527 | dev->base_addr = s[i].map.base_addr; | |
528 | dev->mem_start = s[i].map.mem_start; | |
529 | dev->mem_end = s[i].map.mem_end; | |
530 | return 1; | |
531 | } | |
532 | } | |
533 | return 0; | |
534 | } | |
d1b19dff | 535 | EXPORT_SYMBOL(netdev_boot_setup_check); |
1da177e4 LT |
536 | |
537 | ||
538 | /** | |
539 | * netdev_boot_base - get address from boot time settings | |
540 | * @prefix: prefix for network device | |
541 | * @unit: id for network device | |
542 | * | |
543 | * Check boot time settings for the base address of device. | |
544 | * The found settings are set for the device to be used | |
545 | * later in the device probing. | |
546 | * Returns 0 if no settings found. | |
547 | */ | |
548 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
549 | { | |
550 | const struct netdev_boot_setup *s = dev_boot_setup; | |
551 | char name[IFNAMSIZ]; | |
552 | int i; | |
553 | ||
554 | sprintf(name, "%s%d", prefix, unit); | |
555 | ||
556 | /* | |
557 | * If device already registered then return base of 1 | |
558 | * to indicate not to probe for this interface | |
559 | */ | |
881d966b | 560 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
561 | return 1; |
562 | ||
563 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
564 | if (!strcmp(name, s[i].name)) | |
565 | return s[i].map.base_addr; | |
566 | return 0; | |
567 | } | |
568 | ||
569 | /* | |
570 | * Saves at boot time configured settings for any netdevice. | |
571 | */ | |
572 | int __init netdev_boot_setup(char *str) | |
573 | { | |
574 | int ints[5]; | |
575 | struct ifmap map; | |
576 | ||
577 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
578 | if (!str || !*str) | |
579 | return 0; | |
580 | ||
581 | /* Save settings */ | |
582 | memset(&map, 0, sizeof(map)); | |
583 | if (ints[0] > 0) | |
584 | map.irq = ints[1]; | |
585 | if (ints[0] > 1) | |
586 | map.base_addr = ints[2]; | |
587 | if (ints[0] > 2) | |
588 | map.mem_start = ints[3]; | |
589 | if (ints[0] > 3) | |
590 | map.mem_end = ints[4]; | |
591 | ||
592 | /* Add new entry to the list */ | |
593 | return netdev_boot_setup_add(str, &map); | |
594 | } | |
595 | ||
596 | __setup("netdev=", netdev_boot_setup); | |
597 | ||
598 | /******************************************************************************* | |
599 | ||
600 | Device Interface Subroutines | |
601 | ||
602 | *******************************************************************************/ | |
603 | ||
604 | /** | |
605 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 606 | * @net: the applicable net namespace |
1da177e4 LT |
607 | * @name: name to find |
608 | * | |
609 | * Find an interface by name. Must be called under RTNL semaphore | |
610 | * or @dev_base_lock. If the name is found a pointer to the device | |
611 | * is returned. If the name is not found then %NULL is returned. The | |
612 | * reference counters are not incremented so the caller must be | |
613 | * careful with locks. | |
614 | */ | |
615 | ||
881d966b | 616 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
617 | { |
618 | struct hlist_node *p; | |
0bd8d536 ED |
619 | struct net_device *dev; |
620 | struct hlist_head *head = dev_name_hash(net, name); | |
1da177e4 | 621 | |
0bd8d536 | 622 | hlist_for_each_entry(dev, p, head, name_hlist) |
1da177e4 LT |
623 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
624 | return dev; | |
0bd8d536 | 625 | |
1da177e4 LT |
626 | return NULL; |
627 | } | |
d1b19dff | 628 | EXPORT_SYMBOL(__dev_get_by_name); |
1da177e4 | 629 | |
72c9528b ED |
630 | /** |
631 | * dev_get_by_name_rcu - find a device by its name | |
632 | * @net: the applicable net namespace | |
633 | * @name: name to find | |
634 | * | |
635 | * Find an interface by name. | |
636 | * If the name is found a pointer to the device is returned. | |
637 | * If the name is not found then %NULL is returned. | |
638 | * The reference counters are not incremented so the caller must be | |
639 | * careful with locks. The caller must hold RCU lock. | |
640 | */ | |
641 | ||
642 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
643 | { | |
644 | struct hlist_node *p; | |
645 | struct net_device *dev; | |
646 | struct hlist_head *head = dev_name_hash(net, name); | |
647 | ||
648 | hlist_for_each_entry_rcu(dev, p, head, name_hlist) | |
649 | if (!strncmp(dev->name, name, IFNAMSIZ)) | |
650 | return dev; | |
651 | ||
652 | return NULL; | |
653 | } | |
654 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
655 | ||
1da177e4 LT |
656 | /** |
657 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 658 | * @net: the applicable net namespace |
1da177e4 LT |
659 | * @name: name to find |
660 | * | |
661 | * Find an interface by name. This can be called from any | |
662 | * context and does its own locking. The returned handle has | |
663 | * the usage count incremented and the caller must use dev_put() to | |
664 | * release it when it is no longer needed. %NULL is returned if no | |
665 | * matching device is found. | |
666 | */ | |
667 | ||
881d966b | 668 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
669 | { |
670 | struct net_device *dev; | |
671 | ||
72c9528b ED |
672 | rcu_read_lock(); |
673 | dev = dev_get_by_name_rcu(net, name); | |
1da177e4 LT |
674 | if (dev) |
675 | dev_hold(dev); | |
72c9528b | 676 | rcu_read_unlock(); |
1da177e4 LT |
677 | return dev; |
678 | } | |
d1b19dff | 679 | EXPORT_SYMBOL(dev_get_by_name); |
1da177e4 LT |
680 | |
681 | /** | |
682 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 683 | * @net: the applicable net namespace |
1da177e4 LT |
684 | * @ifindex: index of device |
685 | * | |
686 | * Search for an interface by index. Returns %NULL if the device | |
687 | * is not found or a pointer to the device. The device has not | |
688 | * had its reference counter increased so the caller must be careful | |
689 | * about locking. The caller must hold either the RTNL semaphore | |
690 | * or @dev_base_lock. | |
691 | */ | |
692 | ||
881d966b | 693 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
694 | { |
695 | struct hlist_node *p; | |
0bd8d536 ED |
696 | struct net_device *dev; |
697 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
1da177e4 | 698 | |
0bd8d536 | 699 | hlist_for_each_entry(dev, p, head, index_hlist) |
1da177e4 LT |
700 | if (dev->ifindex == ifindex) |
701 | return dev; | |
0bd8d536 | 702 | |
1da177e4 LT |
703 | return NULL; |
704 | } | |
d1b19dff | 705 | EXPORT_SYMBOL(__dev_get_by_index); |
1da177e4 | 706 | |
fb699dfd ED |
707 | /** |
708 | * dev_get_by_index_rcu - find a device by its ifindex | |
709 | * @net: the applicable net namespace | |
710 | * @ifindex: index of device | |
711 | * | |
712 | * Search for an interface by index. Returns %NULL if the device | |
713 | * is not found or a pointer to the device. The device has not | |
714 | * had its reference counter increased so the caller must be careful | |
715 | * about locking. The caller must hold RCU lock. | |
716 | */ | |
717 | ||
718 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
719 | { | |
720 | struct hlist_node *p; | |
721 | struct net_device *dev; | |
722 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
723 | ||
724 | hlist_for_each_entry_rcu(dev, p, head, index_hlist) | |
725 | if (dev->ifindex == ifindex) | |
726 | return dev; | |
727 | ||
728 | return NULL; | |
729 | } | |
730 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
731 | ||
1da177e4 LT |
732 | |
733 | /** | |
734 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 735 | * @net: the applicable net namespace |
1da177e4 LT |
736 | * @ifindex: index of device |
737 | * | |
738 | * Search for an interface by index. Returns NULL if the device | |
739 | * is not found or a pointer to the device. The device returned has | |
740 | * had a reference added and the pointer is safe until the user calls | |
741 | * dev_put to indicate they have finished with it. | |
742 | */ | |
743 | ||
881d966b | 744 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
745 | { |
746 | struct net_device *dev; | |
747 | ||
fb699dfd ED |
748 | rcu_read_lock(); |
749 | dev = dev_get_by_index_rcu(net, ifindex); | |
1da177e4 LT |
750 | if (dev) |
751 | dev_hold(dev); | |
fb699dfd | 752 | rcu_read_unlock(); |
1da177e4 LT |
753 | return dev; |
754 | } | |
d1b19dff | 755 | EXPORT_SYMBOL(dev_get_by_index); |
1da177e4 LT |
756 | |
757 | /** | |
941666c2 | 758 | * dev_getbyhwaddr_rcu - find a device by its hardware address |
c4ea43c5 | 759 | * @net: the applicable net namespace |
1da177e4 LT |
760 | * @type: media type of device |
761 | * @ha: hardware address | |
762 | * | |
763 | * Search for an interface by MAC address. Returns NULL if the device | |
c506653d ED |
764 | * is not found or a pointer to the device. |
765 | * The caller must hold RCU or RTNL. | |
941666c2 | 766 | * The returned device has not had its ref count increased |
1da177e4 LT |
767 | * and the caller must therefore be careful about locking |
768 | * | |
1da177e4 LT |
769 | */ |
770 | ||
941666c2 ED |
771 | struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, |
772 | const char *ha) | |
1da177e4 LT |
773 | { |
774 | struct net_device *dev; | |
775 | ||
941666c2 | 776 | for_each_netdev_rcu(net, dev) |
1da177e4 LT |
777 | if (dev->type == type && |
778 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
779 | return dev; |
780 | ||
781 | return NULL; | |
1da177e4 | 782 | } |
941666c2 | 783 | EXPORT_SYMBOL(dev_getbyhwaddr_rcu); |
cf309e3f | 784 | |
881d966b | 785 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
786 | { |
787 | struct net_device *dev; | |
788 | ||
4e9cac2b | 789 | ASSERT_RTNL(); |
881d966b | 790 | for_each_netdev(net, dev) |
4e9cac2b | 791 | if (dev->type == type) |
7562f876 PE |
792 | return dev; |
793 | ||
794 | return NULL; | |
4e9cac2b | 795 | } |
4e9cac2b PM |
796 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); |
797 | ||
881d966b | 798 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b | 799 | { |
99fe3c39 | 800 | struct net_device *dev, *ret = NULL; |
4e9cac2b | 801 | |
99fe3c39 ED |
802 | rcu_read_lock(); |
803 | for_each_netdev_rcu(net, dev) | |
804 | if (dev->type == type) { | |
805 | dev_hold(dev); | |
806 | ret = dev; | |
807 | break; | |
808 | } | |
809 | rcu_read_unlock(); | |
810 | return ret; | |
1da177e4 | 811 | } |
1da177e4 LT |
812 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
813 | ||
814 | /** | |
bb69ae04 | 815 | * dev_get_by_flags_rcu - find any device with given flags |
c4ea43c5 | 816 | * @net: the applicable net namespace |
1da177e4 LT |
817 | * @if_flags: IFF_* values |
818 | * @mask: bitmask of bits in if_flags to check | |
819 | * | |
820 | * Search for any interface with the given flags. Returns NULL if a device | |
bb69ae04 ED |
821 | * is not found or a pointer to the device. Must be called inside |
822 | * rcu_read_lock(), and result refcount is unchanged. | |
1da177e4 LT |
823 | */ |
824 | ||
bb69ae04 | 825 | struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags, |
d1b19dff | 826 | unsigned short mask) |
1da177e4 | 827 | { |
7562f876 | 828 | struct net_device *dev, *ret; |
1da177e4 | 829 | |
7562f876 | 830 | ret = NULL; |
c6d14c84 | 831 | for_each_netdev_rcu(net, dev) { |
1da177e4 | 832 | if (((dev->flags ^ if_flags) & mask) == 0) { |
7562f876 | 833 | ret = dev; |
1da177e4 LT |
834 | break; |
835 | } | |
836 | } | |
7562f876 | 837 | return ret; |
1da177e4 | 838 | } |
bb69ae04 | 839 | EXPORT_SYMBOL(dev_get_by_flags_rcu); |
1da177e4 LT |
840 | |
841 | /** | |
842 | * dev_valid_name - check if name is okay for network device | |
843 | * @name: name string | |
844 | * | |
845 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
846 | * to allow sysfs to work. We also disallow any kind of |
847 | * whitespace. | |
1da177e4 | 848 | */ |
95f050bf | 849 | bool dev_valid_name(const char *name) |
1da177e4 | 850 | { |
c7fa9d18 | 851 | if (*name == '\0') |
95f050bf | 852 | return false; |
b6fe17d6 | 853 | if (strlen(name) >= IFNAMSIZ) |
95f050bf | 854 | return false; |
c7fa9d18 | 855 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
95f050bf | 856 | return false; |
c7fa9d18 DM |
857 | |
858 | while (*name) { | |
859 | if (*name == '/' || isspace(*name)) | |
95f050bf | 860 | return false; |
c7fa9d18 DM |
861 | name++; |
862 | } | |
95f050bf | 863 | return true; |
1da177e4 | 864 | } |
d1b19dff | 865 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
866 | |
867 | /** | |
b267b179 EB |
868 | * __dev_alloc_name - allocate a name for a device |
869 | * @net: network namespace to allocate the device name in | |
1da177e4 | 870 | * @name: name format string |
b267b179 | 871 | * @buf: scratch buffer and result name string |
1da177e4 LT |
872 | * |
873 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
874 | * id. It scans list of devices to build up a free map, then chooses |
875 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
876 | * while allocating the name and adding the device in order to avoid | |
877 | * duplicates. | |
878 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
879 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
880 | */ |
881 | ||
b267b179 | 882 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
883 | { |
884 | int i = 0; | |
1da177e4 LT |
885 | const char *p; |
886 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 887 | unsigned long *inuse; |
1da177e4 LT |
888 | struct net_device *d; |
889 | ||
890 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
891 | if (p) { | |
892 | /* | |
893 | * Verify the string as this thing may have come from | |
894 | * the user. There must be either one "%d" and no other "%" | |
895 | * characters. | |
896 | */ | |
897 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
898 | return -EINVAL; | |
899 | ||
900 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 901 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
902 | if (!inuse) |
903 | return -ENOMEM; | |
904 | ||
881d966b | 905 | for_each_netdev(net, d) { |
1da177e4 LT |
906 | if (!sscanf(d->name, name, &i)) |
907 | continue; | |
908 | if (i < 0 || i >= max_netdevices) | |
909 | continue; | |
910 | ||
911 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 912 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
913 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
914 | set_bit(i, inuse); | |
915 | } | |
916 | ||
917 | i = find_first_zero_bit(inuse, max_netdevices); | |
918 | free_page((unsigned long) inuse); | |
919 | } | |
920 | ||
d9031024 OP |
921 | if (buf != name) |
922 | snprintf(buf, IFNAMSIZ, name, i); | |
b267b179 | 923 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 924 | return i; |
1da177e4 LT |
925 | |
926 | /* It is possible to run out of possible slots | |
927 | * when the name is long and there isn't enough space left | |
928 | * for the digits, or if all bits are used. | |
929 | */ | |
930 | return -ENFILE; | |
931 | } | |
932 | ||
b267b179 EB |
933 | /** |
934 | * dev_alloc_name - allocate a name for a device | |
935 | * @dev: device | |
936 | * @name: name format string | |
937 | * | |
938 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
939 | * id. It scans list of devices to build up a free map, then chooses | |
940 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
941 | * while allocating the name and adding the device in order to avoid | |
942 | * duplicates. | |
943 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
944 | * Returns the number of the unit assigned or a negative errno code. | |
945 | */ | |
946 | ||
947 | int dev_alloc_name(struct net_device *dev, const char *name) | |
948 | { | |
949 | char buf[IFNAMSIZ]; | |
950 | struct net *net; | |
951 | int ret; | |
952 | ||
c346dca1 YH |
953 | BUG_ON(!dev_net(dev)); |
954 | net = dev_net(dev); | |
b267b179 EB |
955 | ret = __dev_alloc_name(net, name, buf); |
956 | if (ret >= 0) | |
957 | strlcpy(dev->name, buf, IFNAMSIZ); | |
958 | return ret; | |
959 | } | |
d1b19dff | 960 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 961 | |
828de4f6 G |
962 | static int dev_alloc_name_ns(struct net *net, |
963 | struct net_device *dev, | |
964 | const char *name) | |
d9031024 | 965 | { |
828de4f6 G |
966 | char buf[IFNAMSIZ]; |
967 | int ret; | |
8ce6cebc | 968 | |
828de4f6 G |
969 | ret = __dev_alloc_name(net, name, buf); |
970 | if (ret >= 0) | |
971 | strlcpy(dev->name, buf, IFNAMSIZ); | |
972 | return ret; | |
973 | } | |
974 | ||
975 | static int dev_get_valid_name(struct net *net, | |
976 | struct net_device *dev, | |
977 | const char *name) | |
978 | { | |
979 | BUG_ON(!net); | |
8ce6cebc | 980 | |
d9031024 OP |
981 | if (!dev_valid_name(name)) |
982 | return -EINVAL; | |
983 | ||
1c5cae81 | 984 | if (strchr(name, '%')) |
828de4f6 | 985 | return dev_alloc_name_ns(net, dev, name); |
d9031024 OP |
986 | else if (__dev_get_by_name(net, name)) |
987 | return -EEXIST; | |
8ce6cebc DL |
988 | else if (dev->name != name) |
989 | strlcpy(dev->name, name, IFNAMSIZ); | |
d9031024 OP |
990 | |
991 | return 0; | |
992 | } | |
1da177e4 LT |
993 | |
994 | /** | |
995 | * dev_change_name - change name of a device | |
996 | * @dev: device | |
997 | * @newname: name (or format string) must be at least IFNAMSIZ | |
998 | * | |
999 | * Change name of a device, can pass format strings "eth%d". | |
1000 | * for wildcarding. | |
1001 | */ | |
cf04a4c7 | 1002 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 1003 | { |
fcc5a03a | 1004 | char oldname[IFNAMSIZ]; |
1da177e4 | 1005 | int err = 0; |
fcc5a03a | 1006 | int ret; |
881d966b | 1007 | struct net *net; |
1da177e4 LT |
1008 | |
1009 | ASSERT_RTNL(); | |
c346dca1 | 1010 | BUG_ON(!dev_net(dev)); |
1da177e4 | 1011 | |
c346dca1 | 1012 | net = dev_net(dev); |
1da177e4 LT |
1013 | if (dev->flags & IFF_UP) |
1014 | return -EBUSY; | |
1015 | ||
c8d90dca SH |
1016 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) |
1017 | return 0; | |
1018 | ||
fcc5a03a HX |
1019 | memcpy(oldname, dev->name, IFNAMSIZ); |
1020 | ||
828de4f6 | 1021 | err = dev_get_valid_name(net, dev, newname); |
d9031024 OP |
1022 | if (err < 0) |
1023 | return err; | |
1da177e4 | 1024 | |
fcc5a03a | 1025 | rollback: |
a1b3f594 EB |
1026 | ret = device_rename(&dev->dev, dev->name); |
1027 | if (ret) { | |
1028 | memcpy(dev->name, oldname, IFNAMSIZ); | |
1029 | return ret; | |
dcc99773 | 1030 | } |
7f988eab HX |
1031 | |
1032 | write_lock_bh(&dev_base_lock); | |
372b2312 | 1033 | hlist_del_rcu(&dev->name_hlist); |
72c9528b ED |
1034 | write_unlock_bh(&dev_base_lock); |
1035 | ||
1036 | synchronize_rcu(); | |
1037 | ||
1038 | write_lock_bh(&dev_base_lock); | |
1039 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
7f988eab HX |
1040 | write_unlock_bh(&dev_base_lock); |
1041 | ||
056925ab | 1042 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1043 | ret = notifier_to_errno(ret); |
1044 | ||
1045 | if (ret) { | |
91e9c07b ED |
1046 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1047 | if (err >= 0) { | |
fcc5a03a HX |
1048 | err = ret; |
1049 | memcpy(dev->name, oldname, IFNAMSIZ); | |
1050 | goto rollback; | |
91e9c07b | 1051 | } else { |
7b6cd1ce | 1052 | pr_err("%s: name change rollback failed: %d\n", |
91e9c07b | 1053 | dev->name, ret); |
fcc5a03a HX |
1054 | } |
1055 | } | |
1da177e4 LT |
1056 | |
1057 | return err; | |
1058 | } | |
1059 | ||
0b815a1a SH |
1060 | /** |
1061 | * dev_set_alias - change ifalias of a device | |
1062 | * @dev: device | |
1063 | * @alias: name up to IFALIASZ | |
f0db275a | 1064 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1065 | * |
1066 | * Set ifalias for a device, | |
1067 | */ | |
1068 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1069 | { | |
7364e445 AK |
1070 | char *new_ifalias; |
1071 | ||
0b815a1a SH |
1072 | ASSERT_RTNL(); |
1073 | ||
1074 | if (len >= IFALIASZ) | |
1075 | return -EINVAL; | |
1076 | ||
96ca4a2c OH |
1077 | if (!len) { |
1078 | if (dev->ifalias) { | |
1079 | kfree(dev->ifalias); | |
1080 | dev->ifalias = NULL; | |
1081 | } | |
1082 | return 0; | |
1083 | } | |
1084 | ||
7364e445 AK |
1085 | new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL); |
1086 | if (!new_ifalias) | |
0b815a1a | 1087 | return -ENOMEM; |
7364e445 | 1088 | dev->ifalias = new_ifalias; |
0b815a1a SH |
1089 | |
1090 | strlcpy(dev->ifalias, alias, len+1); | |
1091 | return len; | |
1092 | } | |
1093 | ||
1094 | ||
d8a33ac4 | 1095 | /** |
3041a069 | 1096 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1097 | * @dev: device to cause notification |
1098 | * | |
1099 | * Called to indicate a device has changed features. | |
1100 | */ | |
1101 | void netdev_features_change(struct net_device *dev) | |
1102 | { | |
056925ab | 1103 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1104 | } |
1105 | EXPORT_SYMBOL(netdev_features_change); | |
1106 | ||
1da177e4 LT |
1107 | /** |
1108 | * netdev_state_change - device changes state | |
1109 | * @dev: device to cause notification | |
1110 | * | |
1111 | * Called to indicate a device has changed state. This function calls | |
1112 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1113 | * to the routing socket. | |
1114 | */ | |
1115 | void netdev_state_change(struct net_device *dev) | |
1116 | { | |
1117 | if (dev->flags & IFF_UP) { | |
056925ab | 1118 | call_netdevice_notifiers(NETDEV_CHANGE, dev); |
1da177e4 LT |
1119 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0); |
1120 | } | |
1121 | } | |
d1b19dff | 1122 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1123 | |
ee89bab1 AW |
1124 | /** |
1125 | * netdev_notify_peers - notify network peers about existence of @dev | |
1126 | * @dev: network device | |
1127 | * | |
1128 | * Generate traffic such that interested network peers are aware of | |
1129 | * @dev, such as by generating a gratuitous ARP. This may be used when | |
1130 | * a device wants to inform the rest of the network about some sort of | |
1131 | * reconfiguration such as a failover event or virtual machine | |
1132 | * migration. | |
1133 | */ | |
1134 | void netdev_notify_peers(struct net_device *dev) | |
c1da4ac7 | 1135 | { |
ee89bab1 AW |
1136 | rtnl_lock(); |
1137 | call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); | |
1138 | rtnl_unlock(); | |
c1da4ac7 | 1139 | } |
ee89bab1 | 1140 | EXPORT_SYMBOL(netdev_notify_peers); |
c1da4ac7 | 1141 | |
1da177e4 LT |
1142 | /** |
1143 | * dev_load - load a network module | |
c4ea43c5 | 1144 | * @net: the applicable net namespace |
1da177e4 LT |
1145 | * @name: name of interface |
1146 | * | |
1147 | * If a network interface is not present and the process has suitable | |
1148 | * privileges this function loads the module. If module loading is not | |
1149 | * available in this kernel then it becomes a nop. | |
1150 | */ | |
1151 | ||
881d966b | 1152 | void dev_load(struct net *net, const char *name) |
1da177e4 | 1153 | { |
4ec93edb | 1154 | struct net_device *dev; |
8909c9ad | 1155 | int no_module; |
1da177e4 | 1156 | |
72c9528b ED |
1157 | rcu_read_lock(); |
1158 | dev = dev_get_by_name_rcu(net, name); | |
1159 | rcu_read_unlock(); | |
1da177e4 | 1160 | |
8909c9ad VK |
1161 | no_module = !dev; |
1162 | if (no_module && capable(CAP_NET_ADMIN)) | |
1163 | no_module = request_module("netdev-%s", name); | |
1164 | if (no_module && capable(CAP_SYS_MODULE)) { | |
1165 | if (!request_module("%s", name)) | |
7cecb523 VL |
1166 | pr_warn("Loading kernel module for a network device with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s instead.\n", |
1167 | name); | |
8909c9ad | 1168 | } |
1da177e4 | 1169 | } |
d1b19dff | 1170 | EXPORT_SYMBOL(dev_load); |
1da177e4 | 1171 | |
bd380811 | 1172 | static int __dev_open(struct net_device *dev) |
1da177e4 | 1173 | { |
d314774c | 1174 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1175 | int ret; |
1da177e4 | 1176 | |
e46b66bc BH |
1177 | ASSERT_RTNL(); |
1178 | ||
1da177e4 LT |
1179 | if (!netif_device_present(dev)) |
1180 | return -ENODEV; | |
1181 | ||
3b8bcfd5 JB |
1182 | ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); |
1183 | ret = notifier_to_errno(ret); | |
1184 | if (ret) | |
1185 | return ret; | |
1186 | ||
1da177e4 | 1187 | set_bit(__LINK_STATE_START, &dev->state); |
bada339b | 1188 | |
d314774c SH |
1189 | if (ops->ndo_validate_addr) |
1190 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1191 | |
d314774c SH |
1192 | if (!ret && ops->ndo_open) |
1193 | ret = ops->ndo_open(dev); | |
1da177e4 | 1194 | |
bada339b JG |
1195 | if (ret) |
1196 | clear_bit(__LINK_STATE_START, &dev->state); | |
1197 | else { | |
1da177e4 | 1198 | dev->flags |= IFF_UP; |
b4bd07c2 | 1199 | net_dmaengine_get(); |
4417da66 | 1200 | dev_set_rx_mode(dev); |
1da177e4 | 1201 | dev_activate(dev); |
7bf23575 | 1202 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 1203 | } |
bada339b | 1204 | |
1da177e4 LT |
1205 | return ret; |
1206 | } | |
1207 | ||
1208 | /** | |
bd380811 PM |
1209 | * dev_open - prepare an interface for use. |
1210 | * @dev: device to open | |
1da177e4 | 1211 | * |
bd380811 PM |
1212 | * Takes a device from down to up state. The device's private open |
1213 | * function is invoked and then the multicast lists are loaded. Finally | |
1214 | * the device is moved into the up state and a %NETDEV_UP message is | |
1215 | * sent to the netdev notifier chain. | |
1216 | * | |
1217 | * Calling this function on an active interface is a nop. On a failure | |
1218 | * a negative errno code is returned. | |
1da177e4 | 1219 | */ |
bd380811 PM |
1220 | int dev_open(struct net_device *dev) |
1221 | { | |
1222 | int ret; | |
1223 | ||
bd380811 PM |
1224 | if (dev->flags & IFF_UP) |
1225 | return 0; | |
1226 | ||
bd380811 PM |
1227 | ret = __dev_open(dev); |
1228 | if (ret < 0) | |
1229 | return ret; | |
1230 | ||
bd380811 PM |
1231 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING); |
1232 | call_netdevice_notifiers(NETDEV_UP, dev); | |
1233 | ||
1234 | return ret; | |
1235 | } | |
1236 | EXPORT_SYMBOL(dev_open); | |
1237 | ||
44345724 | 1238 | static int __dev_close_many(struct list_head *head) |
1da177e4 | 1239 | { |
44345724 | 1240 | struct net_device *dev; |
e46b66bc | 1241 | |
bd380811 | 1242 | ASSERT_RTNL(); |
9d5010db DM |
1243 | might_sleep(); |
1244 | ||
44345724 | 1245 | list_for_each_entry(dev, head, unreg_list) { |
44345724 | 1246 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1247 | |
44345724 | 1248 | clear_bit(__LINK_STATE_START, &dev->state); |
1da177e4 | 1249 | |
44345724 OP |
1250 | /* Synchronize to scheduled poll. We cannot touch poll list, it |
1251 | * can be even on different cpu. So just clear netif_running(). | |
1252 | * | |
1253 | * dev->stop() will invoke napi_disable() on all of it's | |
1254 | * napi_struct instances on this device. | |
1255 | */ | |
1256 | smp_mb__after_clear_bit(); /* Commit netif_running(). */ | |
1257 | } | |
1da177e4 | 1258 | |
44345724 | 1259 | dev_deactivate_many(head); |
d8b2a4d2 | 1260 | |
44345724 OP |
1261 | list_for_each_entry(dev, head, unreg_list) { |
1262 | const struct net_device_ops *ops = dev->netdev_ops; | |
1da177e4 | 1263 | |
44345724 OP |
1264 | /* |
1265 | * Call the device specific close. This cannot fail. | |
1266 | * Only if device is UP | |
1267 | * | |
1268 | * We allow it to be called even after a DETACH hot-plug | |
1269 | * event. | |
1270 | */ | |
1271 | if (ops->ndo_stop) | |
1272 | ops->ndo_stop(dev); | |
1273 | ||
44345724 | 1274 | dev->flags &= ~IFF_UP; |
44345724 OP |
1275 | net_dmaengine_put(); |
1276 | } | |
1277 | ||
1278 | return 0; | |
1279 | } | |
1280 | ||
1281 | static int __dev_close(struct net_device *dev) | |
1282 | { | |
f87e6f47 | 1283 | int retval; |
44345724 OP |
1284 | LIST_HEAD(single); |
1285 | ||
1286 | list_add(&dev->unreg_list, &single); | |
f87e6f47 LT |
1287 | retval = __dev_close_many(&single); |
1288 | list_del(&single); | |
1289 | return retval; | |
44345724 OP |
1290 | } |
1291 | ||
3fbd8758 | 1292 | static int dev_close_many(struct list_head *head) |
44345724 OP |
1293 | { |
1294 | struct net_device *dev, *tmp; | |
1295 | LIST_HEAD(tmp_list); | |
1da177e4 | 1296 | |
44345724 OP |
1297 | list_for_each_entry_safe(dev, tmp, head, unreg_list) |
1298 | if (!(dev->flags & IFF_UP)) | |
1299 | list_move(&dev->unreg_list, &tmp_list); | |
1300 | ||
1301 | __dev_close_many(head); | |
1da177e4 | 1302 | |
44345724 OP |
1303 | list_for_each_entry(dev, head, unreg_list) { |
1304 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING); | |
1305 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
1306 | } | |
bd380811 | 1307 | |
44345724 OP |
1308 | /* rollback_registered_many needs the complete original list */ |
1309 | list_splice(&tmp_list, head); | |
bd380811 PM |
1310 | return 0; |
1311 | } | |
1312 | ||
1313 | /** | |
1314 | * dev_close - shutdown an interface. | |
1315 | * @dev: device to shutdown | |
1316 | * | |
1317 | * This function moves an active device into down state. A | |
1318 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1319 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1320 | * chain. | |
1321 | */ | |
1322 | int dev_close(struct net_device *dev) | |
1323 | { | |
e14a5993 ED |
1324 | if (dev->flags & IFF_UP) { |
1325 | LIST_HEAD(single); | |
1da177e4 | 1326 | |
e14a5993 ED |
1327 | list_add(&dev->unreg_list, &single); |
1328 | dev_close_many(&single); | |
1329 | list_del(&single); | |
1330 | } | |
1da177e4 LT |
1331 | return 0; |
1332 | } | |
d1b19dff | 1333 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1334 | |
1335 | ||
0187bdfb BH |
1336 | /** |
1337 | * dev_disable_lro - disable Large Receive Offload on a device | |
1338 | * @dev: device | |
1339 | * | |
1340 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1341 | * called under RTNL. This is needed if received packets may be | |
1342 | * forwarded to another interface. | |
1343 | */ | |
1344 | void dev_disable_lro(struct net_device *dev) | |
1345 | { | |
f11970e3 NH |
1346 | /* |
1347 | * If we're trying to disable lro on a vlan device | |
1348 | * use the underlying physical device instead | |
1349 | */ | |
1350 | if (is_vlan_dev(dev)) | |
1351 | dev = vlan_dev_real_dev(dev); | |
1352 | ||
bc5787c6 MM |
1353 | dev->wanted_features &= ~NETIF_F_LRO; |
1354 | netdev_update_features(dev); | |
27660515 | 1355 | |
22d5969f MM |
1356 | if (unlikely(dev->features & NETIF_F_LRO)) |
1357 | netdev_WARN(dev, "failed to disable LRO!\n"); | |
0187bdfb BH |
1358 | } |
1359 | EXPORT_SYMBOL(dev_disable_lro); | |
1360 | ||
1361 | ||
881d966b EB |
1362 | static int dev_boot_phase = 1; |
1363 | ||
1da177e4 LT |
1364 | /** |
1365 | * register_netdevice_notifier - register a network notifier block | |
1366 | * @nb: notifier | |
1367 | * | |
1368 | * Register a notifier to be called when network device events occur. | |
1369 | * The notifier passed is linked into the kernel structures and must | |
1370 | * not be reused until it has been unregistered. A negative errno code | |
1371 | * is returned on a failure. | |
1372 | * | |
1373 | * When registered all registration and up events are replayed | |
4ec93edb | 1374 | * to the new notifier to allow device to have a race free |
1da177e4 LT |
1375 | * view of the network device list. |
1376 | */ | |
1377 | ||
1378 | int register_netdevice_notifier(struct notifier_block *nb) | |
1379 | { | |
1380 | struct net_device *dev; | |
fcc5a03a | 1381 | struct net_device *last; |
881d966b | 1382 | struct net *net; |
1da177e4 LT |
1383 | int err; |
1384 | ||
1385 | rtnl_lock(); | |
f07d5b94 | 1386 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1387 | if (err) |
1388 | goto unlock; | |
881d966b EB |
1389 | if (dev_boot_phase) |
1390 | goto unlock; | |
1391 | for_each_net(net) { | |
1392 | for_each_netdev(net, dev) { | |
1393 | err = nb->notifier_call(nb, NETDEV_REGISTER, dev); | |
1394 | err = notifier_to_errno(err); | |
1395 | if (err) | |
1396 | goto rollback; | |
1397 | ||
1398 | if (!(dev->flags & IFF_UP)) | |
1399 | continue; | |
1da177e4 | 1400 | |
881d966b EB |
1401 | nb->notifier_call(nb, NETDEV_UP, dev); |
1402 | } | |
1da177e4 | 1403 | } |
fcc5a03a HX |
1404 | |
1405 | unlock: | |
1da177e4 LT |
1406 | rtnl_unlock(); |
1407 | return err; | |
fcc5a03a HX |
1408 | |
1409 | rollback: | |
1410 | last = dev; | |
881d966b EB |
1411 | for_each_net(net) { |
1412 | for_each_netdev(net, dev) { | |
1413 | if (dev == last) | |
8f891489 | 1414 | goto outroll; |
fcc5a03a | 1415 | |
881d966b EB |
1416 | if (dev->flags & IFF_UP) { |
1417 | nb->notifier_call(nb, NETDEV_GOING_DOWN, dev); | |
1418 | nb->notifier_call(nb, NETDEV_DOWN, dev); | |
1419 | } | |
1420 | nb->notifier_call(nb, NETDEV_UNREGISTER, dev); | |
fcc5a03a | 1421 | } |
fcc5a03a | 1422 | } |
c67625a1 | 1423 | |
8f891489 | 1424 | outroll: |
c67625a1 | 1425 | raw_notifier_chain_unregister(&netdev_chain, nb); |
fcc5a03a | 1426 | goto unlock; |
1da177e4 | 1427 | } |
d1b19dff | 1428 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1429 | |
1430 | /** | |
1431 | * unregister_netdevice_notifier - unregister a network notifier block | |
1432 | * @nb: notifier | |
1433 | * | |
1434 | * Unregister a notifier previously registered by | |
1435 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1436 | * kernel structures and may then be reused. A negative errno code | |
1437 | * is returned on a failure. | |
7d3d43da EB |
1438 | * |
1439 | * After unregistering unregister and down device events are synthesized | |
1440 | * for all devices on the device list to the removed notifier to remove | |
1441 | * the need for special case cleanup code. | |
1da177e4 LT |
1442 | */ |
1443 | ||
1444 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1445 | { | |
7d3d43da EB |
1446 | struct net_device *dev; |
1447 | struct net *net; | |
9f514950 HX |
1448 | int err; |
1449 | ||
1450 | rtnl_lock(); | |
f07d5b94 | 1451 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
7d3d43da EB |
1452 | if (err) |
1453 | goto unlock; | |
1454 | ||
1455 | for_each_net(net) { | |
1456 | for_each_netdev(net, dev) { | |
1457 | if (dev->flags & IFF_UP) { | |
1458 | nb->notifier_call(nb, NETDEV_GOING_DOWN, dev); | |
1459 | nb->notifier_call(nb, NETDEV_DOWN, dev); | |
1460 | } | |
1461 | nb->notifier_call(nb, NETDEV_UNREGISTER, dev); | |
7d3d43da EB |
1462 | } |
1463 | } | |
1464 | unlock: | |
9f514950 HX |
1465 | rtnl_unlock(); |
1466 | return err; | |
1da177e4 | 1467 | } |
d1b19dff | 1468 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 LT |
1469 | |
1470 | /** | |
1471 | * call_netdevice_notifiers - call all network notifier blocks | |
1472 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1473 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1474 | * |
1475 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1476 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1477 | */ |
1478 | ||
ad7379d4 | 1479 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1480 | { |
ab930471 | 1481 | ASSERT_RTNL(); |
ad7379d4 | 1482 | return raw_notifier_call_chain(&netdev_chain, val, dev); |
1da177e4 | 1483 | } |
edf947f1 | 1484 | EXPORT_SYMBOL(call_netdevice_notifiers); |
1da177e4 | 1485 | |
c5905afb | 1486 | static struct static_key netstamp_needed __read_mostly; |
b90e5794 | 1487 | #ifdef HAVE_JUMP_LABEL |
c5905afb | 1488 | /* We are not allowed to call static_key_slow_dec() from irq context |
b90e5794 | 1489 | * If net_disable_timestamp() is called from irq context, defer the |
c5905afb | 1490 | * static_key_slow_dec() calls. |
b90e5794 ED |
1491 | */ |
1492 | static atomic_t netstamp_needed_deferred; | |
1493 | #endif | |
1da177e4 LT |
1494 | |
1495 | void net_enable_timestamp(void) | |
1496 | { | |
b90e5794 ED |
1497 | #ifdef HAVE_JUMP_LABEL |
1498 | int deferred = atomic_xchg(&netstamp_needed_deferred, 0); | |
1499 | ||
1500 | if (deferred) { | |
1501 | while (--deferred) | |
c5905afb | 1502 | static_key_slow_dec(&netstamp_needed); |
b90e5794 ED |
1503 | return; |
1504 | } | |
1505 | #endif | |
1506 | WARN_ON(in_interrupt()); | |
c5905afb | 1507 | static_key_slow_inc(&netstamp_needed); |
1da177e4 | 1508 | } |
d1b19dff | 1509 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
1510 | |
1511 | void net_disable_timestamp(void) | |
1512 | { | |
b90e5794 ED |
1513 | #ifdef HAVE_JUMP_LABEL |
1514 | if (in_interrupt()) { | |
1515 | atomic_inc(&netstamp_needed_deferred); | |
1516 | return; | |
1517 | } | |
1518 | #endif | |
c5905afb | 1519 | static_key_slow_dec(&netstamp_needed); |
1da177e4 | 1520 | } |
d1b19dff | 1521 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 1522 | |
3b098e2d | 1523 | static inline void net_timestamp_set(struct sk_buff *skb) |
1da177e4 | 1524 | { |
588f0330 | 1525 | skb->tstamp.tv64 = 0; |
c5905afb | 1526 | if (static_key_false(&netstamp_needed)) |
a61bbcf2 | 1527 | __net_timestamp(skb); |
1da177e4 LT |
1528 | } |
1529 | ||
588f0330 | 1530 | #define net_timestamp_check(COND, SKB) \ |
c5905afb | 1531 | if (static_key_false(&netstamp_needed)) { \ |
588f0330 ED |
1532 | if ((COND) && !(SKB)->tstamp.tv64) \ |
1533 | __net_timestamp(SKB); \ | |
1534 | } \ | |
3b098e2d | 1535 | |
4dc360c5 RC |
1536 | static int net_hwtstamp_validate(struct ifreq *ifr) |
1537 | { | |
1538 | struct hwtstamp_config cfg; | |
1539 | enum hwtstamp_tx_types tx_type; | |
1540 | enum hwtstamp_rx_filters rx_filter; | |
1541 | int tx_type_valid = 0; | |
1542 | int rx_filter_valid = 0; | |
1543 | ||
1544 | if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg))) | |
1545 | return -EFAULT; | |
1546 | ||
1547 | if (cfg.flags) /* reserved for future extensions */ | |
1548 | return -EINVAL; | |
1549 | ||
1550 | tx_type = cfg.tx_type; | |
1551 | rx_filter = cfg.rx_filter; | |
1552 | ||
1553 | switch (tx_type) { | |
1554 | case HWTSTAMP_TX_OFF: | |
1555 | case HWTSTAMP_TX_ON: | |
1556 | case HWTSTAMP_TX_ONESTEP_SYNC: | |
1557 | tx_type_valid = 1; | |
1558 | break; | |
1559 | } | |
1560 | ||
1561 | switch (rx_filter) { | |
1562 | case HWTSTAMP_FILTER_NONE: | |
1563 | case HWTSTAMP_FILTER_ALL: | |
1564 | case HWTSTAMP_FILTER_SOME: | |
1565 | case HWTSTAMP_FILTER_PTP_V1_L4_EVENT: | |
1566 | case HWTSTAMP_FILTER_PTP_V1_L4_SYNC: | |
1567 | case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ: | |
1568 | case HWTSTAMP_FILTER_PTP_V2_L4_EVENT: | |
1569 | case HWTSTAMP_FILTER_PTP_V2_L4_SYNC: | |
1570 | case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ: | |
1571 | case HWTSTAMP_FILTER_PTP_V2_L2_EVENT: | |
1572 | case HWTSTAMP_FILTER_PTP_V2_L2_SYNC: | |
1573 | case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ: | |
1574 | case HWTSTAMP_FILTER_PTP_V2_EVENT: | |
1575 | case HWTSTAMP_FILTER_PTP_V2_SYNC: | |
1576 | case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ: | |
1577 | rx_filter_valid = 1; | |
1578 | break; | |
1579 | } | |
1580 | ||
1581 | if (!tx_type_valid || !rx_filter_valid) | |
1582 | return -ERANGE; | |
1583 | ||
1584 | return 0; | |
1585 | } | |
1586 | ||
79b569f0 DL |
1587 | static inline bool is_skb_forwardable(struct net_device *dev, |
1588 | struct sk_buff *skb) | |
1589 | { | |
1590 | unsigned int len; | |
1591 | ||
1592 | if (!(dev->flags & IFF_UP)) | |
1593 | return false; | |
1594 | ||
1595 | len = dev->mtu + dev->hard_header_len + VLAN_HLEN; | |
1596 | if (skb->len <= len) | |
1597 | return true; | |
1598 | ||
1599 | /* if TSO is enabled, we don't care about the length as the packet | |
1600 | * could be forwarded without being segmented before | |
1601 | */ | |
1602 | if (skb_is_gso(skb)) | |
1603 | return true; | |
1604 | ||
1605 | return false; | |
1606 | } | |
1607 | ||
44540960 AB |
1608 | /** |
1609 | * dev_forward_skb - loopback an skb to another netif | |
1610 | * | |
1611 | * @dev: destination network device | |
1612 | * @skb: buffer to forward | |
1613 | * | |
1614 | * return values: | |
1615 | * NET_RX_SUCCESS (no congestion) | |
6ec82562 | 1616 | * NET_RX_DROP (packet was dropped, but freed) |
44540960 AB |
1617 | * |
1618 | * dev_forward_skb can be used for injecting an skb from the | |
1619 | * start_xmit function of one device into the receive queue | |
1620 | * of another device. | |
1621 | * | |
1622 | * The receiving device may be in another namespace, so | |
1623 | * we have to clear all information in the skb that could | |
1624 | * impact namespace isolation. | |
1625 | */ | |
1626 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
1627 | { | |
48c83012 MT |
1628 | if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) { |
1629 | if (skb_copy_ubufs(skb, GFP_ATOMIC)) { | |
1630 | atomic_long_inc(&dev->rx_dropped); | |
1631 | kfree_skb(skb); | |
1632 | return NET_RX_DROP; | |
1633 | } | |
1634 | } | |
1635 | ||
44540960 | 1636 | skb_orphan(skb); |
c736eefa | 1637 | nf_reset(skb); |
44540960 | 1638 | |
79b569f0 | 1639 | if (unlikely(!is_skb_forwardable(dev, skb))) { |
caf586e5 | 1640 | atomic_long_inc(&dev->rx_dropped); |
6ec82562 | 1641 | kfree_skb(skb); |
44540960 | 1642 | return NET_RX_DROP; |
6ec82562 | 1643 | } |
3b9785c6 | 1644 | skb->skb_iif = 0; |
59b9997b DM |
1645 | skb->dev = dev; |
1646 | skb_dst_drop(skb); | |
44540960 AB |
1647 | skb->tstamp.tv64 = 0; |
1648 | skb->pkt_type = PACKET_HOST; | |
1649 | skb->protocol = eth_type_trans(skb, dev); | |
59b9997b DM |
1650 | skb->mark = 0; |
1651 | secpath_reset(skb); | |
1652 | nf_reset(skb); | |
44540960 AB |
1653 | return netif_rx(skb); |
1654 | } | |
1655 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
1656 | ||
71d9dec2 CG |
1657 | static inline int deliver_skb(struct sk_buff *skb, |
1658 | struct packet_type *pt_prev, | |
1659 | struct net_device *orig_dev) | |
1660 | { | |
1080e512 MT |
1661 | if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) |
1662 | return -ENOMEM; | |
71d9dec2 CG |
1663 | atomic_inc(&skb->users); |
1664 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1665 | } | |
1666 | ||
c0de08d0 EL |
1667 | static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) |
1668 | { | |
1669 | if (ptype->af_packet_priv == NULL) | |
1670 | return false; | |
1671 | ||
1672 | if (ptype->id_match) | |
1673 | return ptype->id_match(ptype, skb->sk); | |
1674 | else if ((struct sock *)ptype->af_packet_priv == skb->sk) | |
1675 | return true; | |
1676 | ||
1677 | return false; | |
1678 | } | |
1679 | ||
1da177e4 LT |
1680 | /* |
1681 | * Support routine. Sends outgoing frames to any network | |
1682 | * taps currently in use. | |
1683 | */ | |
1684 | ||
f6a78bfc | 1685 | static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
1686 | { |
1687 | struct packet_type *ptype; | |
71d9dec2 CG |
1688 | struct sk_buff *skb2 = NULL; |
1689 | struct packet_type *pt_prev = NULL; | |
a61bbcf2 | 1690 | |
1da177e4 LT |
1691 | rcu_read_lock(); |
1692 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
1693 | /* Never send packets back to the socket | |
1694 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1695 | */ | |
1696 | if ((ptype->dev == dev || !ptype->dev) && | |
c0de08d0 | 1697 | (!skb_loop_sk(ptype, skb))) { |
71d9dec2 CG |
1698 | if (pt_prev) { |
1699 | deliver_skb(skb2, pt_prev, skb->dev); | |
1700 | pt_prev = ptype; | |
1701 | continue; | |
1702 | } | |
1703 | ||
1704 | skb2 = skb_clone(skb, GFP_ATOMIC); | |
1da177e4 LT |
1705 | if (!skb2) |
1706 | break; | |
1707 | ||
70978182 ED |
1708 | net_timestamp_set(skb2); |
1709 | ||
1da177e4 LT |
1710 | /* skb->nh should be correctly |
1711 | set by sender, so that the second statement is | |
1712 | just protection against buggy protocols. | |
1713 | */ | |
459a98ed | 1714 | skb_reset_mac_header(skb2); |
1da177e4 | 1715 | |
d56f90a7 | 1716 | if (skb_network_header(skb2) < skb2->data || |
27a884dc | 1717 | skb2->network_header > skb2->tail) { |
e87cc472 JP |
1718 | net_crit_ratelimited("protocol %04x is buggy, dev %s\n", |
1719 | ntohs(skb2->protocol), | |
1720 | dev->name); | |
c1d2bbe1 | 1721 | skb_reset_network_header(skb2); |
1da177e4 LT |
1722 | } |
1723 | ||
b0e380b1 | 1724 | skb2->transport_header = skb2->network_header; |
1da177e4 | 1725 | skb2->pkt_type = PACKET_OUTGOING; |
71d9dec2 | 1726 | pt_prev = ptype; |
1da177e4 LT |
1727 | } |
1728 | } | |
71d9dec2 CG |
1729 | if (pt_prev) |
1730 | pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); | |
1da177e4 LT |
1731 | rcu_read_unlock(); |
1732 | } | |
1733 | ||
2c53040f BH |
1734 | /** |
1735 | * netif_setup_tc - Handle tc mappings on real_num_tx_queues change | |
4f57c087 JF |
1736 | * @dev: Network device |
1737 | * @txq: number of queues available | |
1738 | * | |
1739 | * If real_num_tx_queues is changed the tc mappings may no longer be | |
1740 | * valid. To resolve this verify the tc mapping remains valid and if | |
1741 | * not NULL the mapping. With no priorities mapping to this | |
1742 | * offset/count pair it will no longer be used. In the worst case TC0 | |
1743 | * is invalid nothing can be done so disable priority mappings. If is | |
1744 | * expected that drivers will fix this mapping if they can before | |
1745 | * calling netif_set_real_num_tx_queues. | |
1746 | */ | |
bb134d22 | 1747 | static void netif_setup_tc(struct net_device *dev, unsigned int txq) |
4f57c087 JF |
1748 | { |
1749 | int i; | |
1750 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
1751 | ||
1752 | /* If TC0 is invalidated disable TC mapping */ | |
1753 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce | 1754 | pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); |
4f57c087 JF |
1755 | dev->num_tc = 0; |
1756 | return; | |
1757 | } | |
1758 | ||
1759 | /* Invalidated prio to tc mappings set to TC0 */ | |
1760 | for (i = 1; i < TC_BITMASK + 1; i++) { | |
1761 | int q = netdev_get_prio_tc_map(dev, i); | |
1762 | ||
1763 | tc = &dev->tc_to_txq[q]; | |
1764 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce JP |
1765 | pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", |
1766 | i, q); | |
4f57c087 JF |
1767 | netdev_set_prio_tc_map(dev, i, 0); |
1768 | } | |
1769 | } | |
1770 | } | |
1771 | ||
f0796d5c JF |
1772 | /* |
1773 | * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues | |
1774 | * greater then real_num_tx_queues stale skbs on the qdisc must be flushed. | |
1775 | */ | |
e6484930 | 1776 | int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) |
f0796d5c | 1777 | { |
1d24eb48 TH |
1778 | int rc; |
1779 | ||
e6484930 TH |
1780 | if (txq < 1 || txq > dev->num_tx_queues) |
1781 | return -EINVAL; | |
f0796d5c | 1782 | |
5c56580b BH |
1783 | if (dev->reg_state == NETREG_REGISTERED || |
1784 | dev->reg_state == NETREG_UNREGISTERING) { | |
e6484930 TH |
1785 | ASSERT_RTNL(); |
1786 | ||
1d24eb48 TH |
1787 | rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, |
1788 | txq); | |
bf264145 TH |
1789 | if (rc) |
1790 | return rc; | |
1791 | ||
4f57c087 JF |
1792 | if (dev->num_tc) |
1793 | netif_setup_tc(dev, txq); | |
1794 | ||
e6484930 TH |
1795 | if (txq < dev->real_num_tx_queues) |
1796 | qdisc_reset_all_tx_gt(dev, txq); | |
f0796d5c | 1797 | } |
e6484930 TH |
1798 | |
1799 | dev->real_num_tx_queues = txq; | |
1800 | return 0; | |
f0796d5c JF |
1801 | } |
1802 | EXPORT_SYMBOL(netif_set_real_num_tx_queues); | |
56079431 | 1803 | |
62fe0b40 BH |
1804 | #ifdef CONFIG_RPS |
1805 | /** | |
1806 | * netif_set_real_num_rx_queues - set actual number of RX queues used | |
1807 | * @dev: Network device | |
1808 | * @rxq: Actual number of RX queues | |
1809 | * | |
1810 | * This must be called either with the rtnl_lock held or before | |
1811 | * registration of the net device. Returns 0 on success, or a | |
4e7f7951 BH |
1812 | * negative error code. If called before registration, it always |
1813 | * succeeds. | |
62fe0b40 BH |
1814 | */ |
1815 | int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) | |
1816 | { | |
1817 | int rc; | |
1818 | ||
bd25fa7b TH |
1819 | if (rxq < 1 || rxq > dev->num_rx_queues) |
1820 | return -EINVAL; | |
1821 | ||
62fe0b40 BH |
1822 | if (dev->reg_state == NETREG_REGISTERED) { |
1823 | ASSERT_RTNL(); | |
1824 | ||
62fe0b40 BH |
1825 | rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, |
1826 | rxq); | |
1827 | if (rc) | |
1828 | return rc; | |
62fe0b40 BH |
1829 | } |
1830 | ||
1831 | dev->real_num_rx_queues = rxq; | |
1832 | return 0; | |
1833 | } | |
1834 | EXPORT_SYMBOL(netif_set_real_num_rx_queues); | |
1835 | #endif | |
1836 | ||
2c53040f BH |
1837 | /** |
1838 | * netif_get_num_default_rss_queues - default number of RSS queues | |
16917b87 YM |
1839 | * |
1840 | * This routine should set an upper limit on the number of RSS queues | |
1841 | * used by default by multiqueue devices. | |
1842 | */ | |
a55b138b | 1843 | int netif_get_num_default_rss_queues(void) |
16917b87 YM |
1844 | { |
1845 | return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); | |
1846 | } | |
1847 | EXPORT_SYMBOL(netif_get_num_default_rss_queues); | |
1848 | ||
def82a1d | 1849 | static inline void __netif_reschedule(struct Qdisc *q) |
56079431 | 1850 | { |
def82a1d JP |
1851 | struct softnet_data *sd; |
1852 | unsigned long flags; | |
56079431 | 1853 | |
def82a1d JP |
1854 | local_irq_save(flags); |
1855 | sd = &__get_cpu_var(softnet_data); | |
a9cbd588 CG |
1856 | q->next_sched = NULL; |
1857 | *sd->output_queue_tailp = q; | |
1858 | sd->output_queue_tailp = &q->next_sched; | |
def82a1d JP |
1859 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
1860 | local_irq_restore(flags); | |
1861 | } | |
1862 | ||
1863 | void __netif_schedule(struct Qdisc *q) | |
1864 | { | |
1865 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
1866 | __netif_reschedule(q); | |
56079431 DV |
1867 | } |
1868 | EXPORT_SYMBOL(__netif_schedule); | |
1869 | ||
bea3348e | 1870 | void dev_kfree_skb_irq(struct sk_buff *skb) |
56079431 | 1871 | { |
3578b0c8 | 1872 | if (atomic_dec_and_test(&skb->users)) { |
bea3348e SH |
1873 | struct softnet_data *sd; |
1874 | unsigned long flags; | |
56079431 | 1875 | |
bea3348e SH |
1876 | local_irq_save(flags); |
1877 | sd = &__get_cpu_var(softnet_data); | |
1878 | skb->next = sd->completion_queue; | |
1879 | sd->completion_queue = skb; | |
1880 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
1881 | local_irq_restore(flags); | |
1882 | } | |
56079431 | 1883 | } |
bea3348e | 1884 | EXPORT_SYMBOL(dev_kfree_skb_irq); |
56079431 DV |
1885 | |
1886 | void dev_kfree_skb_any(struct sk_buff *skb) | |
1887 | { | |
1888 | if (in_irq() || irqs_disabled()) | |
1889 | dev_kfree_skb_irq(skb); | |
1890 | else | |
1891 | dev_kfree_skb(skb); | |
1892 | } | |
1893 | EXPORT_SYMBOL(dev_kfree_skb_any); | |
1894 | ||
1895 | ||
bea3348e SH |
1896 | /** |
1897 | * netif_device_detach - mark device as removed | |
1898 | * @dev: network device | |
1899 | * | |
1900 | * Mark device as removed from system and therefore no longer available. | |
1901 | */ | |
56079431 DV |
1902 | void netif_device_detach(struct net_device *dev) |
1903 | { | |
1904 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
1905 | netif_running(dev)) { | |
d543103a | 1906 | netif_tx_stop_all_queues(dev); |
56079431 DV |
1907 | } |
1908 | } | |
1909 | EXPORT_SYMBOL(netif_device_detach); | |
1910 | ||
bea3348e SH |
1911 | /** |
1912 | * netif_device_attach - mark device as attached | |
1913 | * @dev: network device | |
1914 | * | |
1915 | * Mark device as attached from system and restart if needed. | |
1916 | */ | |
56079431 DV |
1917 | void netif_device_attach(struct net_device *dev) |
1918 | { | |
1919 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
1920 | netif_running(dev)) { | |
d543103a | 1921 | netif_tx_wake_all_queues(dev); |
4ec93edb | 1922 | __netdev_watchdog_up(dev); |
56079431 DV |
1923 | } |
1924 | } | |
1925 | EXPORT_SYMBOL(netif_device_attach); | |
1926 | ||
36c92474 BH |
1927 | static void skb_warn_bad_offload(const struct sk_buff *skb) |
1928 | { | |
65e9d2fa | 1929 | static const netdev_features_t null_features = 0; |
36c92474 BH |
1930 | struct net_device *dev = skb->dev; |
1931 | const char *driver = ""; | |
1932 | ||
1933 | if (dev && dev->dev.parent) | |
1934 | driver = dev_driver_string(dev->dev.parent); | |
1935 | ||
1936 | WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d " | |
1937 | "gso_type=%d ip_summed=%d\n", | |
65e9d2fa MM |
1938 | driver, dev ? &dev->features : &null_features, |
1939 | skb->sk ? &skb->sk->sk_route_caps : &null_features, | |
36c92474 BH |
1940 | skb->len, skb->data_len, skb_shinfo(skb)->gso_size, |
1941 | skb_shinfo(skb)->gso_type, skb->ip_summed); | |
1942 | } | |
1943 | ||
1da177e4 LT |
1944 | /* |
1945 | * Invalidate hardware checksum when packet is to be mangled, and | |
1946 | * complete checksum manually on outgoing path. | |
1947 | */ | |
84fa7933 | 1948 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 1949 | { |
d3bc23e7 | 1950 | __wsum csum; |
663ead3b | 1951 | int ret = 0, offset; |
1da177e4 | 1952 | |
84fa7933 | 1953 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
1954 | goto out_set_summed; |
1955 | ||
1956 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
36c92474 BH |
1957 | skb_warn_bad_offload(skb); |
1958 | return -EINVAL; | |
1da177e4 LT |
1959 | } |
1960 | ||
55508d60 | 1961 | offset = skb_checksum_start_offset(skb); |
a030847e HX |
1962 | BUG_ON(offset >= skb_headlen(skb)); |
1963 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
1964 | ||
1965 | offset += skb->csum_offset; | |
1966 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
1967 | ||
1968 | if (skb_cloned(skb) && | |
1969 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1da177e4 LT |
1970 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
1971 | if (ret) | |
1972 | goto out; | |
1973 | } | |
1974 | ||
a030847e | 1975 | *(__sum16 *)(skb->data + offset) = csum_fold(csum); |
a430a43d | 1976 | out_set_summed: |
1da177e4 | 1977 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 1978 | out: |
1da177e4 LT |
1979 | return ret; |
1980 | } | |
d1b19dff | 1981 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 1982 | |
f6a78bfc HX |
1983 | /** |
1984 | * skb_gso_segment - Perform segmentation on skb. | |
1985 | * @skb: buffer to segment | |
576a30eb | 1986 | * @features: features for the output path (see dev->features) |
f6a78bfc HX |
1987 | * |
1988 | * This function segments the given skb and returns a list of segments. | |
576a30eb HX |
1989 | * |
1990 | * It may return NULL if the skb requires no segmentation. This is | |
1991 | * only possible when GSO is used for verifying header integrity. | |
f6a78bfc | 1992 | */ |
c8f44aff MM |
1993 | struct sk_buff *skb_gso_segment(struct sk_buff *skb, |
1994 | netdev_features_t features) | |
f6a78bfc HX |
1995 | { |
1996 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
1997 | struct packet_type *ptype; | |
252e3346 | 1998 | __be16 type = skb->protocol; |
c8d5bcd1 | 1999 | int vlan_depth = ETH_HLEN; |
a430a43d | 2000 | int err; |
f6a78bfc | 2001 | |
c8d5bcd1 JG |
2002 | while (type == htons(ETH_P_8021Q)) { |
2003 | struct vlan_hdr *vh; | |
7b9c6090 | 2004 | |
c8d5bcd1 | 2005 | if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN))) |
7b9c6090 JG |
2006 | return ERR_PTR(-EINVAL); |
2007 | ||
c8d5bcd1 JG |
2008 | vh = (struct vlan_hdr *)(skb->data + vlan_depth); |
2009 | type = vh->h_vlan_encapsulated_proto; | |
2010 | vlan_depth += VLAN_HLEN; | |
7b9c6090 JG |
2011 | } |
2012 | ||
459a98ed | 2013 | skb_reset_mac_header(skb); |
b0e380b1 | 2014 | skb->mac_len = skb->network_header - skb->mac_header; |
f6a78bfc HX |
2015 | __skb_pull(skb, skb->mac_len); |
2016 | ||
67fd1a73 | 2017 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { |
36c92474 | 2018 | skb_warn_bad_offload(skb); |
67fd1a73 | 2019 | |
a430a43d HX |
2020 | if (skb_header_cloned(skb) && |
2021 | (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))) | |
2022 | return ERR_PTR(err); | |
2023 | } | |
2024 | ||
f6a78bfc | 2025 | rcu_read_lock(); |
82d8a867 PE |
2026 | list_for_each_entry_rcu(ptype, |
2027 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
f6a78bfc | 2028 | if (ptype->type == type && !ptype->dev && ptype->gso_segment) { |
84fa7933 | 2029 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { |
a430a43d HX |
2030 | err = ptype->gso_send_check(skb); |
2031 | segs = ERR_PTR(err); | |
2032 | if (err || skb_gso_ok(skb, features)) | |
2033 | break; | |
d56f90a7 ACM |
2034 | __skb_push(skb, (skb->data - |
2035 | skb_network_header(skb))); | |
a430a43d | 2036 | } |
576a30eb | 2037 | segs = ptype->gso_segment(skb, features); |
f6a78bfc HX |
2038 | break; |
2039 | } | |
2040 | } | |
2041 | rcu_read_unlock(); | |
2042 | ||
98e399f8 | 2043 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 2044 | |
f6a78bfc HX |
2045 | return segs; |
2046 | } | |
f6a78bfc HX |
2047 | EXPORT_SYMBOL(skb_gso_segment); |
2048 | ||
fb286bb2 HX |
2049 | /* Take action when hardware reception checksum errors are detected. */ |
2050 | #ifdef CONFIG_BUG | |
2051 | void netdev_rx_csum_fault(struct net_device *dev) | |
2052 | { | |
2053 | if (net_ratelimit()) { | |
7b6cd1ce | 2054 | pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); |
fb286bb2 HX |
2055 | dump_stack(); |
2056 | } | |
2057 | } | |
2058 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
2059 | #endif | |
2060 | ||
1da177e4 LT |
2061 | /* Actually, we should eliminate this check as soon as we know, that: |
2062 | * 1. IOMMU is present and allows to map all the memory. | |
2063 | * 2. No high memory really exists on this machine. | |
2064 | */ | |
2065 | ||
9092c658 | 2066 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
1da177e4 | 2067 | { |
3d3a8533 | 2068 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 2069 | int i; |
5acbbd42 | 2070 | if (!(dev->features & NETIF_F_HIGHDMA)) { |
ea2ab693 IC |
2071 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
2072 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
2073 | if (PageHighMem(skb_frag_page(frag))) | |
5acbbd42 | 2074 | return 1; |
ea2ab693 | 2075 | } |
5acbbd42 | 2076 | } |
1da177e4 | 2077 | |
5acbbd42 FT |
2078 | if (PCI_DMA_BUS_IS_PHYS) { |
2079 | struct device *pdev = dev->dev.parent; | |
1da177e4 | 2080 | |
9092c658 ED |
2081 | if (!pdev) |
2082 | return 0; | |
5acbbd42 | 2083 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
ea2ab693 IC |
2084 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
2085 | dma_addr_t addr = page_to_phys(skb_frag_page(frag)); | |
5acbbd42 FT |
2086 | if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask) |
2087 | return 1; | |
2088 | } | |
2089 | } | |
3d3a8533 | 2090 | #endif |
1da177e4 LT |
2091 | return 0; |
2092 | } | |
1da177e4 | 2093 | |
f6a78bfc HX |
2094 | struct dev_gso_cb { |
2095 | void (*destructor)(struct sk_buff *skb); | |
2096 | }; | |
2097 | ||
2098 | #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb) | |
2099 | ||
2100 | static void dev_gso_skb_destructor(struct sk_buff *skb) | |
2101 | { | |
2102 | struct dev_gso_cb *cb; | |
2103 | ||
2104 | do { | |
2105 | struct sk_buff *nskb = skb->next; | |
2106 | ||
2107 | skb->next = nskb->next; | |
2108 | nskb->next = NULL; | |
2109 | kfree_skb(nskb); | |
2110 | } while (skb->next); | |
2111 | ||
2112 | cb = DEV_GSO_CB(skb); | |
2113 | if (cb->destructor) | |
2114 | cb->destructor(skb); | |
2115 | } | |
2116 | ||
2117 | /** | |
2118 | * dev_gso_segment - Perform emulated hardware segmentation on skb. | |
2119 | * @skb: buffer to segment | |
91ecb63c | 2120 | * @features: device features as applicable to this skb |
f6a78bfc HX |
2121 | * |
2122 | * This function segments the given skb and stores the list of segments | |
2123 | * in skb->next. | |
2124 | */ | |
c8f44aff | 2125 | static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features) |
f6a78bfc | 2126 | { |
f6a78bfc | 2127 | struct sk_buff *segs; |
576a30eb HX |
2128 | |
2129 | segs = skb_gso_segment(skb, features); | |
2130 | ||
2131 | /* Verifying header integrity only. */ | |
2132 | if (!segs) | |
2133 | return 0; | |
f6a78bfc | 2134 | |
801678c5 | 2135 | if (IS_ERR(segs)) |
f6a78bfc HX |
2136 | return PTR_ERR(segs); |
2137 | ||
2138 | skb->next = segs; | |
2139 | DEV_GSO_CB(skb)->destructor = skb->destructor; | |
2140 | skb->destructor = dev_gso_skb_destructor; | |
2141 | ||
2142 | return 0; | |
2143 | } | |
2144 | ||
c8f44aff | 2145 | static bool can_checksum_protocol(netdev_features_t features, __be16 protocol) |
03634668 JG |
2146 | { |
2147 | return ((features & NETIF_F_GEN_CSUM) || | |
2148 | ((features & NETIF_F_V4_CSUM) && | |
2149 | protocol == htons(ETH_P_IP)) || | |
2150 | ((features & NETIF_F_V6_CSUM) && | |
2151 | protocol == htons(ETH_P_IPV6)) || | |
2152 | ((features & NETIF_F_FCOE_CRC) && | |
2153 | protocol == htons(ETH_P_FCOE))); | |
2154 | } | |
2155 | ||
c8f44aff MM |
2156 | static netdev_features_t harmonize_features(struct sk_buff *skb, |
2157 | __be16 protocol, netdev_features_t features) | |
f01a5236 | 2158 | { |
c0d680e5 EC |
2159 | if (skb->ip_summed != CHECKSUM_NONE && |
2160 | !can_checksum_protocol(features, protocol)) { | |
f01a5236 JG |
2161 | features &= ~NETIF_F_ALL_CSUM; |
2162 | features &= ~NETIF_F_SG; | |
2163 | } else if (illegal_highdma(skb->dev, skb)) { | |
2164 | features &= ~NETIF_F_SG; | |
2165 | } | |
2166 | ||
2167 | return features; | |
2168 | } | |
2169 | ||
c8f44aff | 2170 | netdev_features_t netif_skb_features(struct sk_buff *skb) |
58e998c6 JG |
2171 | { |
2172 | __be16 protocol = skb->protocol; | |
c8f44aff | 2173 | netdev_features_t features = skb->dev->features; |
58e998c6 | 2174 | |
30b678d8 BH |
2175 | if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs) |
2176 | features &= ~NETIF_F_GSO_MASK; | |
2177 | ||
58e998c6 JG |
2178 | if (protocol == htons(ETH_P_8021Q)) { |
2179 | struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data; | |
2180 | protocol = veh->h_vlan_encapsulated_proto; | |
f01a5236 JG |
2181 | } else if (!vlan_tx_tag_present(skb)) { |
2182 | return harmonize_features(skb, protocol, features); | |
2183 | } | |
58e998c6 | 2184 | |
6ee400aa | 2185 | features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX); |
f01a5236 JG |
2186 | |
2187 | if (protocol != htons(ETH_P_8021Q)) { | |
2188 | return harmonize_features(skb, protocol, features); | |
2189 | } else { | |
2190 | features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | | |
6ee400aa | 2191 | NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX; |
f01a5236 JG |
2192 | return harmonize_features(skb, protocol, features); |
2193 | } | |
58e998c6 | 2194 | } |
f01a5236 | 2195 | EXPORT_SYMBOL(netif_skb_features); |
58e998c6 | 2196 | |
6afff0ca JF |
2197 | /* |
2198 | * Returns true if either: | |
2199 | * 1. skb has frag_list and the device doesn't support FRAGLIST, or | |
d1a53dfd | 2200 | * 2. skb is fragmented and the device does not support SG. |
6afff0ca JF |
2201 | */ |
2202 | static inline int skb_needs_linearize(struct sk_buff *skb, | |
02932ce9 | 2203 | int features) |
6afff0ca | 2204 | { |
02932ce9 JG |
2205 | return skb_is_nonlinear(skb) && |
2206 | ((skb_has_frag_list(skb) && | |
2207 | !(features & NETIF_F_FRAGLIST)) || | |
e1e78db6 | 2208 | (skb_shinfo(skb)->nr_frags && |
02932ce9 | 2209 | !(features & NETIF_F_SG))); |
6afff0ca JF |
2210 | } |
2211 | ||
fd2ea0a7 DM |
2212 | int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev, |
2213 | struct netdev_queue *txq) | |
f6a78bfc | 2214 | { |
00829823 | 2215 | const struct net_device_ops *ops = dev->netdev_ops; |
572a9d7b | 2216 | int rc = NETDEV_TX_OK; |
ec764bf0 | 2217 | unsigned int skb_len; |
00829823 | 2218 | |
f6a78bfc | 2219 | if (likely(!skb->next)) { |
c8f44aff | 2220 | netdev_features_t features; |
fc741216 | 2221 | |
93f154b5 | 2222 | /* |
25985edc | 2223 | * If device doesn't need skb->dst, release it right now while |
93f154b5 ED |
2224 | * its hot in this cpu cache |
2225 | */ | |
adf30907 ED |
2226 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) |
2227 | skb_dst_drop(skb); | |
2228 | ||
fc741216 JG |
2229 | features = netif_skb_features(skb); |
2230 | ||
7b9c6090 | 2231 | if (vlan_tx_tag_present(skb) && |
fc741216 | 2232 | !(features & NETIF_F_HW_VLAN_TX)) { |
7b9c6090 JG |
2233 | skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb)); |
2234 | if (unlikely(!skb)) | |
2235 | goto out; | |
2236 | ||
2237 | skb->vlan_tci = 0; | |
2238 | } | |
2239 | ||
fc741216 | 2240 | if (netif_needs_gso(skb, features)) { |
91ecb63c | 2241 | if (unlikely(dev_gso_segment(skb, features))) |
9ccb8975 DM |
2242 | goto out_kfree_skb; |
2243 | if (skb->next) | |
2244 | goto gso; | |
6afff0ca | 2245 | } else { |
02932ce9 | 2246 | if (skb_needs_linearize(skb, features) && |
6afff0ca JF |
2247 | __skb_linearize(skb)) |
2248 | goto out_kfree_skb; | |
2249 | ||
2250 | /* If packet is not checksummed and device does not | |
2251 | * support checksumming for this protocol, complete | |
2252 | * checksumming here. | |
2253 | */ | |
2254 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
55508d60 MM |
2255 | skb_set_transport_header(skb, |
2256 | skb_checksum_start_offset(skb)); | |
03634668 | 2257 | if (!(features & NETIF_F_ALL_CSUM) && |
6afff0ca JF |
2258 | skb_checksum_help(skb)) |
2259 | goto out_kfree_skb; | |
2260 | } | |
9ccb8975 DM |
2261 | } |
2262 | ||
b40863c6 ED |
2263 | if (!list_empty(&ptype_all)) |
2264 | dev_queue_xmit_nit(skb, dev); | |
2265 | ||
ec764bf0 | 2266 | skb_len = skb->len; |
ac45f602 | 2267 | rc = ops->ndo_start_xmit(skb, dev); |
ec764bf0 | 2268 | trace_net_dev_xmit(skb, rc, dev, skb_len); |
ec634fe3 | 2269 | if (rc == NETDEV_TX_OK) |
08baf561 | 2270 | txq_trans_update(txq); |
ac45f602 | 2271 | return rc; |
f6a78bfc HX |
2272 | } |
2273 | ||
576a30eb | 2274 | gso: |
f6a78bfc HX |
2275 | do { |
2276 | struct sk_buff *nskb = skb->next; | |
f6a78bfc HX |
2277 | |
2278 | skb->next = nskb->next; | |
2279 | nskb->next = NULL; | |
068a2de5 KK |
2280 | |
2281 | /* | |
25985edc | 2282 | * If device doesn't need nskb->dst, release it right now while |
068a2de5 KK |
2283 | * its hot in this cpu cache |
2284 | */ | |
2285 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
2286 | skb_dst_drop(nskb); | |
2287 | ||
b40863c6 ED |
2288 | if (!list_empty(&ptype_all)) |
2289 | dev_queue_xmit_nit(nskb, dev); | |
2290 | ||
ec764bf0 | 2291 | skb_len = nskb->len; |
00829823 | 2292 | rc = ops->ndo_start_xmit(nskb, dev); |
ec764bf0 | 2293 | trace_net_dev_xmit(nskb, rc, dev, skb_len); |
ec634fe3 | 2294 | if (unlikely(rc != NETDEV_TX_OK)) { |
572a9d7b PM |
2295 | if (rc & ~NETDEV_TX_MASK) |
2296 | goto out_kfree_gso_skb; | |
f54d9e8d | 2297 | nskb->next = skb->next; |
f6a78bfc HX |
2298 | skb->next = nskb; |
2299 | return rc; | |
2300 | } | |
08baf561 | 2301 | txq_trans_update(txq); |
73466498 | 2302 | if (unlikely(netif_xmit_stopped(txq) && skb->next)) |
f54d9e8d | 2303 | return NETDEV_TX_BUSY; |
f6a78bfc | 2304 | } while (skb->next); |
4ec93edb | 2305 | |
572a9d7b PM |
2306 | out_kfree_gso_skb: |
2307 | if (likely(skb->next == NULL)) | |
2308 | skb->destructor = DEV_GSO_CB(skb)->destructor; | |
f6a78bfc HX |
2309 | out_kfree_skb: |
2310 | kfree_skb(skb); | |
7b9c6090 | 2311 | out: |
572a9d7b | 2312 | return rc; |
f6a78bfc HX |
2313 | } |
2314 | ||
0a9627f2 | 2315 | static u32 hashrnd __read_mostly; |
b6b2fed1 | 2316 | |
a3d22a68 VZ |
2317 | /* |
2318 | * Returns a Tx hash based on the given packet descriptor a Tx queues' number | |
2319 | * to be used as a distribution range. | |
2320 | */ | |
2321 | u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb, | |
2322 | unsigned int num_tx_queues) | |
8f0f2223 | 2323 | { |
7019298a | 2324 | u32 hash; |
4f57c087 JF |
2325 | u16 qoffset = 0; |
2326 | u16 qcount = num_tx_queues; | |
b6b2fed1 | 2327 | |
513de11b DM |
2328 | if (skb_rx_queue_recorded(skb)) { |
2329 | hash = skb_get_rx_queue(skb); | |
a3d22a68 VZ |
2330 | while (unlikely(hash >= num_tx_queues)) |
2331 | hash -= num_tx_queues; | |
513de11b DM |
2332 | return hash; |
2333 | } | |
ec581f6a | 2334 | |
4f57c087 JF |
2335 | if (dev->num_tc) { |
2336 | u8 tc = netdev_get_prio_tc_map(dev, skb->priority); | |
2337 | qoffset = dev->tc_to_txq[tc].offset; | |
2338 | qcount = dev->tc_to_txq[tc].count; | |
2339 | } | |
2340 | ||
ec581f6a | 2341 | if (skb->sk && skb->sk->sk_hash) |
7019298a | 2342 | hash = skb->sk->sk_hash; |
ec581f6a | 2343 | else |
62b1a8ab | 2344 | hash = (__force u16) skb->protocol; |
0a9627f2 | 2345 | hash = jhash_1word(hash, hashrnd); |
b6b2fed1 | 2346 | |
4f57c087 | 2347 | return (u16) (((u64) hash * qcount) >> 32) + qoffset; |
8f0f2223 | 2348 | } |
a3d22a68 | 2349 | EXPORT_SYMBOL(__skb_tx_hash); |
8f0f2223 | 2350 | |
ed04642f ED |
2351 | static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index) |
2352 | { | |
2353 | if (unlikely(queue_index >= dev->real_num_tx_queues)) { | |
e87cc472 JP |
2354 | net_warn_ratelimited("%s selects TX queue %d, but real number of TX queues is %d\n", |
2355 | dev->name, queue_index, | |
2356 | dev->real_num_tx_queues); | |
ed04642f ED |
2357 | return 0; |
2358 | } | |
2359 | return queue_index; | |
2360 | } | |
2361 | ||
1d24eb48 TH |
2362 | static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb) |
2363 | { | |
bf264145 | 2364 | #ifdef CONFIG_XPS |
1d24eb48 TH |
2365 | struct xps_dev_maps *dev_maps; |
2366 | struct xps_map *map; | |
2367 | int queue_index = -1; | |
2368 | ||
2369 | rcu_read_lock(); | |
2370 | dev_maps = rcu_dereference(dev->xps_maps); | |
2371 | if (dev_maps) { | |
2372 | map = rcu_dereference( | |
2373 | dev_maps->cpu_map[raw_smp_processor_id()]); | |
2374 | if (map) { | |
2375 | if (map->len == 1) | |
2376 | queue_index = map->queues[0]; | |
2377 | else { | |
2378 | u32 hash; | |
2379 | if (skb->sk && skb->sk->sk_hash) | |
2380 | hash = skb->sk->sk_hash; | |
2381 | else | |
2382 | hash = (__force u16) skb->protocol ^ | |
2383 | skb->rxhash; | |
2384 | hash = jhash_1word(hash, hashrnd); | |
2385 | queue_index = map->queues[ | |
2386 | ((u64)hash * map->len) >> 32]; | |
2387 | } | |
2388 | if (unlikely(queue_index >= dev->real_num_tx_queues)) | |
2389 | queue_index = -1; | |
2390 | } | |
2391 | } | |
2392 | rcu_read_unlock(); | |
2393 | ||
2394 | return queue_index; | |
2395 | #else | |
2396 | return -1; | |
2397 | #endif | |
2398 | } | |
2399 | ||
8c4c49df AW |
2400 | struct netdev_queue *netdev_pick_tx(struct net_device *dev, |
2401 | struct sk_buff *skb) | |
e8a0464c | 2402 | { |
b0f77d0e | 2403 | int queue_index; |
deabc772 | 2404 | const struct net_device_ops *ops = dev->netdev_ops; |
a4ee3ce3 | 2405 | |
3853b584 TH |
2406 | if (dev->real_num_tx_queues == 1) |
2407 | queue_index = 0; | |
2408 | else if (ops->ndo_select_queue) { | |
deabc772 HS |
2409 | queue_index = ops->ndo_select_queue(dev, skb); |
2410 | queue_index = dev_cap_txqueue(dev, queue_index); | |
2411 | } else { | |
2412 | struct sock *sk = skb->sk; | |
2413 | queue_index = sk_tx_queue_get(sk); | |
a4ee3ce3 | 2414 | |
3853b584 TH |
2415 | if (queue_index < 0 || skb->ooo_okay || |
2416 | queue_index >= dev->real_num_tx_queues) { | |
2417 | int old_index = queue_index; | |
fd2ea0a7 | 2418 | |
1d24eb48 TH |
2419 | queue_index = get_xps_queue(dev, skb); |
2420 | if (queue_index < 0) | |
2421 | queue_index = skb_tx_hash(dev, skb); | |
3853b584 TH |
2422 | |
2423 | if (queue_index != old_index && sk) { | |
2424 | struct dst_entry *dst = | |
2425 | rcu_dereference_check(sk->sk_dst_cache, 1); | |
8728c544 ED |
2426 | |
2427 | if (dst && skb_dst(skb) == dst) | |
2428 | sk_tx_queue_set(sk, queue_index); | |
2429 | } | |
a4ee3ce3 KK |
2430 | } |
2431 | } | |
eae792b7 | 2432 | |
fd2ea0a7 DM |
2433 | skb_set_queue_mapping(skb, queue_index); |
2434 | return netdev_get_tx_queue(dev, queue_index); | |
e8a0464c DM |
2435 | } |
2436 | ||
bbd8a0d3 KK |
2437 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
2438 | struct net_device *dev, | |
2439 | struct netdev_queue *txq) | |
2440 | { | |
2441 | spinlock_t *root_lock = qdisc_lock(q); | |
a2da570d | 2442 | bool contended; |
bbd8a0d3 KK |
2443 | int rc; |
2444 | ||
a2da570d ED |
2445 | qdisc_skb_cb(skb)->pkt_len = skb->len; |
2446 | qdisc_calculate_pkt_len(skb, q); | |
79640a4c ED |
2447 | /* |
2448 | * Heuristic to force contended enqueues to serialize on a | |
2449 | * separate lock before trying to get qdisc main lock. | |
2450 | * This permits __QDISC_STATE_RUNNING owner to get the lock more often | |
2451 | * and dequeue packets faster. | |
2452 | */ | |
a2da570d | 2453 | contended = qdisc_is_running(q); |
79640a4c ED |
2454 | if (unlikely(contended)) |
2455 | spin_lock(&q->busylock); | |
2456 | ||
bbd8a0d3 KK |
2457 | spin_lock(root_lock); |
2458 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
2459 | kfree_skb(skb); | |
2460 | rc = NET_XMIT_DROP; | |
2461 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
bc135b23 | 2462 | qdisc_run_begin(q)) { |
bbd8a0d3 KK |
2463 | /* |
2464 | * This is a work-conserving queue; there are no old skbs | |
2465 | * waiting to be sent out; and the qdisc is not running - | |
2466 | * xmit the skb directly. | |
2467 | */ | |
7fee226a ED |
2468 | if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE)) |
2469 | skb_dst_force(skb); | |
bfe0d029 | 2470 | |
bfe0d029 ED |
2471 | qdisc_bstats_update(q, skb); |
2472 | ||
79640a4c ED |
2473 | if (sch_direct_xmit(skb, q, dev, txq, root_lock)) { |
2474 | if (unlikely(contended)) { | |
2475 | spin_unlock(&q->busylock); | |
2476 | contended = false; | |
2477 | } | |
bbd8a0d3 | 2478 | __qdisc_run(q); |
79640a4c | 2479 | } else |
bc135b23 | 2480 | qdisc_run_end(q); |
bbd8a0d3 KK |
2481 | |
2482 | rc = NET_XMIT_SUCCESS; | |
2483 | } else { | |
7fee226a | 2484 | skb_dst_force(skb); |
a2da570d | 2485 | rc = q->enqueue(skb, q) & NET_XMIT_MASK; |
79640a4c ED |
2486 | if (qdisc_run_begin(q)) { |
2487 | if (unlikely(contended)) { | |
2488 | spin_unlock(&q->busylock); | |
2489 | contended = false; | |
2490 | } | |
2491 | __qdisc_run(q); | |
2492 | } | |
bbd8a0d3 KK |
2493 | } |
2494 | spin_unlock(root_lock); | |
79640a4c ED |
2495 | if (unlikely(contended)) |
2496 | spin_unlock(&q->busylock); | |
bbd8a0d3 KK |
2497 | return rc; |
2498 | } | |
2499 | ||
5bc1421e NH |
2500 | #if IS_ENABLED(CONFIG_NETPRIO_CGROUP) |
2501 | static void skb_update_prio(struct sk_buff *skb) | |
2502 | { | |
6977a79d | 2503 | struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap); |
5bc1421e | 2504 | |
91c68ce2 ED |
2505 | if (!skb->priority && skb->sk && map) { |
2506 | unsigned int prioidx = skb->sk->sk_cgrp_prioidx; | |
2507 | ||
2508 | if (prioidx < map->priomap_len) | |
2509 | skb->priority = map->priomap[prioidx]; | |
2510 | } | |
5bc1421e NH |
2511 | } |
2512 | #else | |
2513 | #define skb_update_prio(skb) | |
2514 | #endif | |
2515 | ||
745e20f1 | 2516 | static DEFINE_PER_CPU(int, xmit_recursion); |
11a766ce | 2517 | #define RECURSION_LIMIT 10 |
745e20f1 | 2518 | |
95603e22 MM |
2519 | /** |
2520 | * dev_loopback_xmit - loop back @skb | |
2521 | * @skb: buffer to transmit | |
2522 | */ | |
2523 | int dev_loopback_xmit(struct sk_buff *skb) | |
2524 | { | |
2525 | skb_reset_mac_header(skb); | |
2526 | __skb_pull(skb, skb_network_offset(skb)); | |
2527 | skb->pkt_type = PACKET_LOOPBACK; | |
2528 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
2529 | WARN_ON(!skb_dst(skb)); | |
2530 | skb_dst_force(skb); | |
2531 | netif_rx_ni(skb); | |
2532 | return 0; | |
2533 | } | |
2534 | EXPORT_SYMBOL(dev_loopback_xmit); | |
2535 | ||
d29f749e DJ |
2536 | /** |
2537 | * dev_queue_xmit - transmit a buffer | |
2538 | * @skb: buffer to transmit | |
2539 | * | |
2540 | * Queue a buffer for transmission to a network device. The caller must | |
2541 | * have set the device and priority and built the buffer before calling | |
2542 | * this function. The function can be called from an interrupt. | |
2543 | * | |
2544 | * A negative errno code is returned on a failure. A success does not | |
2545 | * guarantee the frame will be transmitted as it may be dropped due | |
2546 | * to congestion or traffic shaping. | |
2547 | * | |
2548 | * ----------------------------------------------------------------------------------- | |
2549 | * I notice this method can also return errors from the queue disciplines, | |
2550 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
2551 | * be positive. | |
2552 | * | |
2553 | * Regardless of the return value, the skb is consumed, so it is currently | |
2554 | * difficult to retry a send to this method. (You can bump the ref count | |
2555 | * before sending to hold a reference for retry if you are careful.) | |
2556 | * | |
2557 | * When calling this method, interrupts MUST be enabled. This is because | |
2558 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
2559 | * --BLG | |
2560 | */ | |
1da177e4 LT |
2561 | int dev_queue_xmit(struct sk_buff *skb) |
2562 | { | |
2563 | struct net_device *dev = skb->dev; | |
dc2b4847 | 2564 | struct netdev_queue *txq; |
1da177e4 LT |
2565 | struct Qdisc *q; |
2566 | int rc = -ENOMEM; | |
2567 | ||
4ec93edb YH |
2568 | /* Disable soft irqs for various locks below. Also |
2569 | * stops preemption for RCU. | |
1da177e4 | 2570 | */ |
4ec93edb | 2571 | rcu_read_lock_bh(); |
1da177e4 | 2572 | |
5bc1421e NH |
2573 | skb_update_prio(skb); |
2574 | ||
8c4c49df | 2575 | txq = netdev_pick_tx(dev, skb); |
a898def2 | 2576 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 2577 | |
1da177e4 | 2578 | #ifdef CONFIG_NET_CLS_ACT |
d1b19dff | 2579 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS); |
1da177e4 | 2580 | #endif |
cf66ba58 | 2581 | trace_net_dev_queue(skb); |
1da177e4 | 2582 | if (q->enqueue) { |
bbd8a0d3 | 2583 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 2584 | goto out; |
1da177e4 LT |
2585 | } |
2586 | ||
2587 | /* The device has no queue. Common case for software devices: | |
2588 | loopback, all the sorts of tunnels... | |
2589 | ||
932ff279 HX |
2590 | Really, it is unlikely that netif_tx_lock protection is necessary |
2591 | here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
1da177e4 LT |
2592 | counters.) |
2593 | However, it is possible, that they rely on protection | |
2594 | made by us here. | |
2595 | ||
2596 | Check this and shot the lock. It is not prone from deadlocks. | |
2597 | Either shot noqueue qdisc, it is even simpler 8) | |
2598 | */ | |
2599 | if (dev->flags & IFF_UP) { | |
2600 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
2601 | ||
c773e847 | 2602 | if (txq->xmit_lock_owner != cpu) { |
1da177e4 | 2603 | |
745e20f1 ED |
2604 | if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT) |
2605 | goto recursion_alert; | |
2606 | ||
c773e847 | 2607 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 2608 | |
73466498 | 2609 | if (!netif_xmit_stopped(txq)) { |
745e20f1 | 2610 | __this_cpu_inc(xmit_recursion); |
572a9d7b | 2611 | rc = dev_hard_start_xmit(skb, dev, txq); |
745e20f1 | 2612 | __this_cpu_dec(xmit_recursion); |
572a9d7b | 2613 | if (dev_xmit_complete(rc)) { |
c773e847 | 2614 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
2615 | goto out; |
2616 | } | |
2617 | } | |
c773e847 | 2618 | HARD_TX_UNLOCK(dev, txq); |
e87cc472 JP |
2619 | net_crit_ratelimited("Virtual device %s asks to queue packet!\n", |
2620 | dev->name); | |
1da177e4 LT |
2621 | } else { |
2622 | /* Recursion is detected! It is possible, | |
745e20f1 ED |
2623 | * unfortunately |
2624 | */ | |
2625 | recursion_alert: | |
e87cc472 JP |
2626 | net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", |
2627 | dev->name); | |
1da177e4 LT |
2628 | } |
2629 | } | |
2630 | ||
2631 | rc = -ENETDOWN; | |
d4828d85 | 2632 | rcu_read_unlock_bh(); |
1da177e4 | 2633 | |
1da177e4 LT |
2634 | kfree_skb(skb); |
2635 | return rc; | |
2636 | out: | |
d4828d85 | 2637 | rcu_read_unlock_bh(); |
1da177e4 LT |
2638 | return rc; |
2639 | } | |
d1b19dff | 2640 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 LT |
2641 | |
2642 | ||
2643 | /*======================================================================= | |
2644 | Receiver routines | |
2645 | =======================================================================*/ | |
2646 | ||
6b2bedc3 | 2647 | int netdev_max_backlog __read_mostly = 1000; |
c9e6bc64 ED |
2648 | EXPORT_SYMBOL(netdev_max_backlog); |
2649 | ||
3b098e2d | 2650 | int netdev_tstamp_prequeue __read_mostly = 1; |
6b2bedc3 SH |
2651 | int netdev_budget __read_mostly = 300; |
2652 | int weight_p __read_mostly = 64; /* old backlog weight */ | |
1da177e4 | 2653 | |
eecfd7c4 ED |
2654 | /* Called with irq disabled */ |
2655 | static inline void ____napi_schedule(struct softnet_data *sd, | |
2656 | struct napi_struct *napi) | |
2657 | { | |
2658 | list_add_tail(&napi->poll_list, &sd->poll_list); | |
2659 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2660 | } | |
2661 | ||
0a9627f2 | 2662 | /* |
bfb564e7 | 2663 | * __skb_get_rxhash: calculate a flow hash based on src/dst addresses |
bdeab991 TH |
2664 | * and src/dst port numbers. Sets rxhash in skb to non-zero hash value |
2665 | * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb | |
2666 | * if hash is a canonical 4-tuple hash over transport ports. | |
0a9627f2 | 2667 | */ |
bdeab991 | 2668 | void __skb_get_rxhash(struct sk_buff *skb) |
0a9627f2 | 2669 | { |
4504b861 ED |
2670 | struct flow_keys keys; |
2671 | u32 hash; | |
c6865cb3 | 2672 | |
4504b861 ED |
2673 | if (!skb_flow_dissect(skb, &keys)) |
2674 | return; | |
e971b722 | 2675 | |
68622342 | 2676 | if (keys.ports) |
4504b861 | 2677 | skb->l4_rxhash = 1; |
0a9627f2 | 2678 | |
b249dcb8 | 2679 | /* get a consistent hash (same value on both flow directions) */ |
68622342 CG |
2680 | if (((__force u32)keys.dst < (__force u32)keys.src) || |
2681 | (((__force u32)keys.dst == (__force u32)keys.src) && | |
2682 | ((__force u16)keys.port16[1] < (__force u16)keys.port16[0]))) { | |
4504b861 | 2683 | swap(keys.dst, keys.src); |
68622342 CG |
2684 | swap(keys.port16[0], keys.port16[1]); |
2685 | } | |
0a9627f2 | 2686 | |
4504b861 ED |
2687 | hash = jhash_3words((__force u32)keys.dst, |
2688 | (__force u32)keys.src, | |
2689 | (__force u32)keys.ports, hashrnd); | |
bfb564e7 KK |
2690 | if (!hash) |
2691 | hash = 1; | |
2692 | ||
bdeab991 | 2693 | skb->rxhash = hash; |
bfb564e7 KK |
2694 | } |
2695 | EXPORT_SYMBOL(__skb_get_rxhash); | |
2696 | ||
2697 | #ifdef CONFIG_RPS | |
2698 | ||
2699 | /* One global table that all flow-based protocols share. */ | |
6e3f7faf | 2700 | struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; |
bfb564e7 KK |
2701 | EXPORT_SYMBOL(rps_sock_flow_table); |
2702 | ||
c5905afb | 2703 | struct static_key rps_needed __read_mostly; |
adc9300e | 2704 | |
c445477d BH |
2705 | static struct rps_dev_flow * |
2706 | set_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
2707 | struct rps_dev_flow *rflow, u16 next_cpu) | |
2708 | { | |
09994d1b | 2709 | if (next_cpu != RPS_NO_CPU) { |
c445477d BH |
2710 | #ifdef CONFIG_RFS_ACCEL |
2711 | struct netdev_rx_queue *rxqueue; | |
2712 | struct rps_dev_flow_table *flow_table; | |
2713 | struct rps_dev_flow *old_rflow; | |
2714 | u32 flow_id; | |
2715 | u16 rxq_index; | |
2716 | int rc; | |
2717 | ||
2718 | /* Should we steer this flow to a different hardware queue? */ | |
69a19ee6 BH |
2719 | if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || |
2720 | !(dev->features & NETIF_F_NTUPLE)) | |
c445477d BH |
2721 | goto out; |
2722 | rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); | |
2723 | if (rxq_index == skb_get_rx_queue(skb)) | |
2724 | goto out; | |
2725 | ||
2726 | rxqueue = dev->_rx + rxq_index; | |
2727 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
2728 | if (!flow_table) | |
2729 | goto out; | |
2730 | flow_id = skb->rxhash & flow_table->mask; | |
2731 | rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, | |
2732 | rxq_index, flow_id); | |
2733 | if (rc < 0) | |
2734 | goto out; | |
2735 | old_rflow = rflow; | |
2736 | rflow = &flow_table->flows[flow_id]; | |
c445477d BH |
2737 | rflow->filter = rc; |
2738 | if (old_rflow->filter == rflow->filter) | |
2739 | old_rflow->filter = RPS_NO_FILTER; | |
2740 | out: | |
2741 | #endif | |
2742 | rflow->last_qtail = | |
09994d1b | 2743 | per_cpu(softnet_data, next_cpu).input_queue_head; |
c445477d BH |
2744 | } |
2745 | ||
09994d1b | 2746 | rflow->cpu = next_cpu; |
c445477d BH |
2747 | return rflow; |
2748 | } | |
2749 | ||
bfb564e7 KK |
2750 | /* |
2751 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
2752 | * CPU from the RPS map of the receiving queue for a given skb. | |
2753 | * rcu_read_lock must be held on entry. | |
2754 | */ | |
2755 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
2756 | struct rps_dev_flow **rflowp) | |
2757 | { | |
2758 | struct netdev_rx_queue *rxqueue; | |
6e3f7faf | 2759 | struct rps_map *map; |
bfb564e7 KK |
2760 | struct rps_dev_flow_table *flow_table; |
2761 | struct rps_sock_flow_table *sock_flow_table; | |
2762 | int cpu = -1; | |
2763 | u16 tcpu; | |
2764 | ||
2765 | if (skb_rx_queue_recorded(skb)) { | |
2766 | u16 index = skb_get_rx_queue(skb); | |
62fe0b40 BH |
2767 | if (unlikely(index >= dev->real_num_rx_queues)) { |
2768 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
2769 | "%s received packet on queue %u, but number " | |
2770 | "of RX queues is %u\n", | |
2771 | dev->name, index, dev->real_num_rx_queues); | |
bfb564e7 KK |
2772 | goto done; |
2773 | } | |
2774 | rxqueue = dev->_rx + index; | |
2775 | } else | |
2776 | rxqueue = dev->_rx; | |
2777 | ||
6e3f7faf ED |
2778 | map = rcu_dereference(rxqueue->rps_map); |
2779 | if (map) { | |
85875236 | 2780 | if (map->len == 1 && |
33d480ce | 2781 | !rcu_access_pointer(rxqueue->rps_flow_table)) { |
6febfca9 CG |
2782 | tcpu = map->cpus[0]; |
2783 | if (cpu_online(tcpu)) | |
2784 | cpu = tcpu; | |
2785 | goto done; | |
2786 | } | |
33d480ce | 2787 | } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) { |
bfb564e7 | 2788 | goto done; |
6febfca9 | 2789 | } |
bfb564e7 | 2790 | |
2d47b459 | 2791 | skb_reset_network_header(skb); |
bfb564e7 KK |
2792 | if (!skb_get_rxhash(skb)) |
2793 | goto done; | |
2794 | ||
fec5e652 TH |
2795 | flow_table = rcu_dereference(rxqueue->rps_flow_table); |
2796 | sock_flow_table = rcu_dereference(rps_sock_flow_table); | |
2797 | if (flow_table && sock_flow_table) { | |
2798 | u16 next_cpu; | |
2799 | struct rps_dev_flow *rflow; | |
2800 | ||
2801 | rflow = &flow_table->flows[skb->rxhash & flow_table->mask]; | |
2802 | tcpu = rflow->cpu; | |
2803 | ||
2804 | next_cpu = sock_flow_table->ents[skb->rxhash & | |
2805 | sock_flow_table->mask]; | |
2806 | ||
2807 | /* | |
2808 | * If the desired CPU (where last recvmsg was done) is | |
2809 | * different from current CPU (one in the rx-queue flow | |
2810 | * table entry), switch if one of the following holds: | |
2811 | * - Current CPU is unset (equal to RPS_NO_CPU). | |
2812 | * - Current CPU is offline. | |
2813 | * - The current CPU's queue tail has advanced beyond the | |
2814 | * last packet that was enqueued using this table entry. | |
2815 | * This guarantees that all previous packets for the flow | |
2816 | * have been dequeued, thus preserving in order delivery. | |
2817 | */ | |
2818 | if (unlikely(tcpu != next_cpu) && | |
2819 | (tcpu == RPS_NO_CPU || !cpu_online(tcpu) || | |
2820 | ((int)(per_cpu(softnet_data, tcpu).input_queue_head - | |
c445477d BH |
2821 | rflow->last_qtail)) >= 0)) |
2822 | rflow = set_rps_cpu(dev, skb, rflow, next_cpu); | |
2823 | ||
fec5e652 TH |
2824 | if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) { |
2825 | *rflowp = rflow; | |
2826 | cpu = tcpu; | |
2827 | goto done; | |
2828 | } | |
2829 | } | |
2830 | ||
0a9627f2 | 2831 | if (map) { |
fec5e652 | 2832 | tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32]; |
0a9627f2 TH |
2833 | |
2834 | if (cpu_online(tcpu)) { | |
2835 | cpu = tcpu; | |
2836 | goto done; | |
2837 | } | |
2838 | } | |
2839 | ||
2840 | done: | |
0a9627f2 TH |
2841 | return cpu; |
2842 | } | |
2843 | ||
c445477d BH |
2844 | #ifdef CONFIG_RFS_ACCEL |
2845 | ||
2846 | /** | |
2847 | * rps_may_expire_flow - check whether an RFS hardware filter may be removed | |
2848 | * @dev: Device on which the filter was set | |
2849 | * @rxq_index: RX queue index | |
2850 | * @flow_id: Flow ID passed to ndo_rx_flow_steer() | |
2851 | * @filter_id: Filter ID returned by ndo_rx_flow_steer() | |
2852 | * | |
2853 | * Drivers that implement ndo_rx_flow_steer() should periodically call | |
2854 | * this function for each installed filter and remove the filters for | |
2855 | * which it returns %true. | |
2856 | */ | |
2857 | bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, | |
2858 | u32 flow_id, u16 filter_id) | |
2859 | { | |
2860 | struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; | |
2861 | struct rps_dev_flow_table *flow_table; | |
2862 | struct rps_dev_flow *rflow; | |
2863 | bool expire = true; | |
2864 | int cpu; | |
2865 | ||
2866 | rcu_read_lock(); | |
2867 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
2868 | if (flow_table && flow_id <= flow_table->mask) { | |
2869 | rflow = &flow_table->flows[flow_id]; | |
2870 | cpu = ACCESS_ONCE(rflow->cpu); | |
2871 | if (rflow->filter == filter_id && cpu != RPS_NO_CPU && | |
2872 | ((int)(per_cpu(softnet_data, cpu).input_queue_head - | |
2873 | rflow->last_qtail) < | |
2874 | (int)(10 * flow_table->mask))) | |
2875 | expire = false; | |
2876 | } | |
2877 | rcu_read_unlock(); | |
2878 | return expire; | |
2879 | } | |
2880 | EXPORT_SYMBOL(rps_may_expire_flow); | |
2881 | ||
2882 | #endif /* CONFIG_RFS_ACCEL */ | |
2883 | ||
0a9627f2 | 2884 | /* Called from hardirq (IPI) context */ |
e36fa2f7 | 2885 | static void rps_trigger_softirq(void *data) |
0a9627f2 | 2886 | { |
e36fa2f7 ED |
2887 | struct softnet_data *sd = data; |
2888 | ||
eecfd7c4 | 2889 | ____napi_schedule(sd, &sd->backlog); |
dee42870 | 2890 | sd->received_rps++; |
0a9627f2 | 2891 | } |
e36fa2f7 | 2892 | |
fec5e652 | 2893 | #endif /* CONFIG_RPS */ |
0a9627f2 | 2894 | |
e36fa2f7 ED |
2895 | /* |
2896 | * Check if this softnet_data structure is another cpu one | |
2897 | * If yes, queue it to our IPI list and return 1 | |
2898 | * If no, return 0 | |
2899 | */ | |
2900 | static int rps_ipi_queued(struct softnet_data *sd) | |
2901 | { | |
2902 | #ifdef CONFIG_RPS | |
2903 | struct softnet_data *mysd = &__get_cpu_var(softnet_data); | |
2904 | ||
2905 | if (sd != mysd) { | |
2906 | sd->rps_ipi_next = mysd->rps_ipi_list; | |
2907 | mysd->rps_ipi_list = sd; | |
2908 | ||
2909 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2910 | return 1; | |
2911 | } | |
2912 | #endif /* CONFIG_RPS */ | |
2913 | return 0; | |
2914 | } | |
2915 | ||
0a9627f2 TH |
2916 | /* |
2917 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
2918 | * queue (may be a remote CPU queue). | |
2919 | */ | |
fec5e652 TH |
2920 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, |
2921 | unsigned int *qtail) | |
0a9627f2 | 2922 | { |
e36fa2f7 | 2923 | struct softnet_data *sd; |
0a9627f2 TH |
2924 | unsigned long flags; |
2925 | ||
e36fa2f7 | 2926 | sd = &per_cpu(softnet_data, cpu); |
0a9627f2 TH |
2927 | |
2928 | local_irq_save(flags); | |
0a9627f2 | 2929 | |
e36fa2f7 | 2930 | rps_lock(sd); |
6e7676c1 CG |
2931 | if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) { |
2932 | if (skb_queue_len(&sd->input_pkt_queue)) { | |
0a9627f2 | 2933 | enqueue: |
e36fa2f7 | 2934 | __skb_queue_tail(&sd->input_pkt_queue, skb); |
76cc8b13 | 2935 | input_queue_tail_incr_save(sd, qtail); |
e36fa2f7 | 2936 | rps_unlock(sd); |
152102c7 | 2937 | local_irq_restore(flags); |
0a9627f2 TH |
2938 | return NET_RX_SUCCESS; |
2939 | } | |
2940 | ||
ebda37c2 ED |
2941 | /* Schedule NAPI for backlog device |
2942 | * We can use non atomic operation since we own the queue lock | |
2943 | */ | |
2944 | if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { | |
e36fa2f7 | 2945 | if (!rps_ipi_queued(sd)) |
eecfd7c4 | 2946 | ____napi_schedule(sd, &sd->backlog); |
0a9627f2 TH |
2947 | } |
2948 | goto enqueue; | |
2949 | } | |
2950 | ||
dee42870 | 2951 | sd->dropped++; |
e36fa2f7 | 2952 | rps_unlock(sd); |
0a9627f2 | 2953 | |
0a9627f2 TH |
2954 | local_irq_restore(flags); |
2955 | ||
caf586e5 | 2956 | atomic_long_inc(&skb->dev->rx_dropped); |
0a9627f2 TH |
2957 | kfree_skb(skb); |
2958 | return NET_RX_DROP; | |
2959 | } | |
1da177e4 | 2960 | |
1da177e4 LT |
2961 | /** |
2962 | * netif_rx - post buffer to the network code | |
2963 | * @skb: buffer to post | |
2964 | * | |
2965 | * This function receives a packet from a device driver and queues it for | |
2966 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
2967 | * may be dropped during processing for congestion control or by the | |
2968 | * protocol layers. | |
2969 | * | |
2970 | * return values: | |
2971 | * NET_RX_SUCCESS (no congestion) | |
1da177e4 LT |
2972 | * NET_RX_DROP (packet was dropped) |
2973 | * | |
2974 | */ | |
2975 | ||
2976 | int netif_rx(struct sk_buff *skb) | |
2977 | { | |
b0e28f1e | 2978 | int ret; |
1da177e4 LT |
2979 | |
2980 | /* if netpoll wants it, pretend we never saw it */ | |
2981 | if (netpoll_rx(skb)) | |
2982 | return NET_RX_DROP; | |
2983 | ||
588f0330 | 2984 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
1da177e4 | 2985 | |
cf66ba58 | 2986 | trace_netif_rx(skb); |
df334545 | 2987 | #ifdef CONFIG_RPS |
c5905afb | 2988 | if (static_key_false(&rps_needed)) { |
fec5e652 | 2989 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
b0e28f1e ED |
2990 | int cpu; |
2991 | ||
cece1945 | 2992 | preempt_disable(); |
b0e28f1e | 2993 | rcu_read_lock(); |
fec5e652 TH |
2994 | |
2995 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
b0e28f1e ED |
2996 | if (cpu < 0) |
2997 | cpu = smp_processor_id(); | |
fec5e652 TH |
2998 | |
2999 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
3000 | ||
b0e28f1e | 3001 | rcu_read_unlock(); |
cece1945 | 3002 | preempt_enable(); |
adc9300e ED |
3003 | } else |
3004 | #endif | |
fec5e652 TH |
3005 | { |
3006 | unsigned int qtail; | |
3007 | ret = enqueue_to_backlog(skb, get_cpu(), &qtail); | |
3008 | put_cpu(); | |
3009 | } | |
b0e28f1e | 3010 | return ret; |
1da177e4 | 3011 | } |
d1b19dff | 3012 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
3013 | |
3014 | int netif_rx_ni(struct sk_buff *skb) | |
3015 | { | |
3016 | int err; | |
3017 | ||
3018 | preempt_disable(); | |
3019 | err = netif_rx(skb); | |
3020 | if (local_softirq_pending()) | |
3021 | do_softirq(); | |
3022 | preempt_enable(); | |
3023 | ||
3024 | return err; | |
3025 | } | |
1da177e4 LT |
3026 | EXPORT_SYMBOL(netif_rx_ni); |
3027 | ||
1da177e4 LT |
3028 | static void net_tx_action(struct softirq_action *h) |
3029 | { | |
3030 | struct softnet_data *sd = &__get_cpu_var(softnet_data); | |
3031 | ||
3032 | if (sd->completion_queue) { | |
3033 | struct sk_buff *clist; | |
3034 | ||
3035 | local_irq_disable(); | |
3036 | clist = sd->completion_queue; | |
3037 | sd->completion_queue = NULL; | |
3038 | local_irq_enable(); | |
3039 | ||
3040 | while (clist) { | |
3041 | struct sk_buff *skb = clist; | |
3042 | clist = clist->next; | |
3043 | ||
547b792c | 3044 | WARN_ON(atomic_read(&skb->users)); |
07dc22e7 | 3045 | trace_kfree_skb(skb, net_tx_action); |
1da177e4 LT |
3046 | __kfree_skb(skb); |
3047 | } | |
3048 | } | |
3049 | ||
3050 | if (sd->output_queue) { | |
37437bb2 | 3051 | struct Qdisc *head; |
1da177e4 LT |
3052 | |
3053 | local_irq_disable(); | |
3054 | head = sd->output_queue; | |
3055 | sd->output_queue = NULL; | |
a9cbd588 | 3056 | sd->output_queue_tailp = &sd->output_queue; |
1da177e4 LT |
3057 | local_irq_enable(); |
3058 | ||
3059 | while (head) { | |
37437bb2 DM |
3060 | struct Qdisc *q = head; |
3061 | spinlock_t *root_lock; | |
3062 | ||
1da177e4 LT |
3063 | head = head->next_sched; |
3064 | ||
5fb66229 | 3065 | root_lock = qdisc_lock(q); |
37437bb2 | 3066 | if (spin_trylock(root_lock)) { |
def82a1d JP |
3067 | smp_mb__before_clear_bit(); |
3068 | clear_bit(__QDISC_STATE_SCHED, | |
3069 | &q->state); | |
37437bb2 DM |
3070 | qdisc_run(q); |
3071 | spin_unlock(root_lock); | |
1da177e4 | 3072 | } else { |
195648bb | 3073 | if (!test_bit(__QDISC_STATE_DEACTIVATED, |
e8a83e10 | 3074 | &q->state)) { |
195648bb | 3075 | __netif_reschedule(q); |
e8a83e10 JP |
3076 | } else { |
3077 | smp_mb__before_clear_bit(); | |
3078 | clear_bit(__QDISC_STATE_SCHED, | |
3079 | &q->state); | |
3080 | } | |
1da177e4 LT |
3081 | } |
3082 | } | |
3083 | } | |
3084 | } | |
3085 | ||
ab95bfe0 JP |
3086 | #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \ |
3087 | (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)) | |
da678292 MM |
3088 | /* This hook is defined here for ATM LANE */ |
3089 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
3090 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 3091 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 3092 | #endif |
1da177e4 | 3093 | |
1da177e4 LT |
3094 | #ifdef CONFIG_NET_CLS_ACT |
3095 | /* TODO: Maybe we should just force sch_ingress to be compiled in | |
3096 | * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions | |
3097 | * a compare and 2 stores extra right now if we dont have it on | |
3098 | * but have CONFIG_NET_CLS_ACT | |
25985edc LDM |
3099 | * NOTE: This doesn't stop any functionality; if you dont have |
3100 | * the ingress scheduler, you just can't add policies on ingress. | |
1da177e4 LT |
3101 | * |
3102 | */ | |
24824a09 | 3103 | static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq) |
1da177e4 | 3104 | { |
1da177e4 | 3105 | struct net_device *dev = skb->dev; |
f697c3e8 | 3106 | u32 ttl = G_TC_RTTL(skb->tc_verd); |
555353cf DM |
3107 | int result = TC_ACT_OK; |
3108 | struct Qdisc *q; | |
4ec93edb | 3109 | |
de384830 | 3110 | if (unlikely(MAX_RED_LOOP < ttl++)) { |
e87cc472 JP |
3111 | net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n", |
3112 | skb->skb_iif, dev->ifindex); | |
f697c3e8 HX |
3113 | return TC_ACT_SHOT; |
3114 | } | |
1da177e4 | 3115 | |
f697c3e8 HX |
3116 | skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl); |
3117 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); | |
1da177e4 | 3118 | |
83874000 | 3119 | q = rxq->qdisc; |
8d50b53d | 3120 | if (q != &noop_qdisc) { |
83874000 | 3121 | spin_lock(qdisc_lock(q)); |
a9312ae8 DM |
3122 | if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) |
3123 | result = qdisc_enqueue_root(skb, q); | |
83874000 DM |
3124 | spin_unlock(qdisc_lock(q)); |
3125 | } | |
f697c3e8 HX |
3126 | |
3127 | return result; | |
3128 | } | |
86e65da9 | 3129 | |
f697c3e8 HX |
3130 | static inline struct sk_buff *handle_ing(struct sk_buff *skb, |
3131 | struct packet_type **pt_prev, | |
3132 | int *ret, struct net_device *orig_dev) | |
3133 | { | |
24824a09 ED |
3134 | struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue); |
3135 | ||
3136 | if (!rxq || rxq->qdisc == &noop_qdisc) | |
f697c3e8 | 3137 | goto out; |
1da177e4 | 3138 | |
f697c3e8 HX |
3139 | if (*pt_prev) { |
3140 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
3141 | *pt_prev = NULL; | |
1da177e4 LT |
3142 | } |
3143 | ||
24824a09 | 3144 | switch (ing_filter(skb, rxq)) { |
f697c3e8 HX |
3145 | case TC_ACT_SHOT: |
3146 | case TC_ACT_STOLEN: | |
3147 | kfree_skb(skb); | |
3148 | return NULL; | |
3149 | } | |
3150 | ||
3151 | out: | |
3152 | skb->tc_verd = 0; | |
3153 | return skb; | |
1da177e4 LT |
3154 | } |
3155 | #endif | |
3156 | ||
ab95bfe0 JP |
3157 | /** |
3158 | * netdev_rx_handler_register - register receive handler | |
3159 | * @dev: device to register a handler for | |
3160 | * @rx_handler: receive handler to register | |
93e2c32b | 3161 | * @rx_handler_data: data pointer that is used by rx handler |
ab95bfe0 JP |
3162 | * |
3163 | * Register a receive hander for a device. This handler will then be | |
3164 | * called from __netif_receive_skb. A negative errno code is returned | |
3165 | * on a failure. | |
3166 | * | |
3167 | * The caller must hold the rtnl_mutex. | |
8a4eb573 JP |
3168 | * |
3169 | * For a general description of rx_handler, see enum rx_handler_result. | |
ab95bfe0 JP |
3170 | */ |
3171 | int netdev_rx_handler_register(struct net_device *dev, | |
93e2c32b JP |
3172 | rx_handler_func_t *rx_handler, |
3173 | void *rx_handler_data) | |
ab95bfe0 JP |
3174 | { |
3175 | ASSERT_RTNL(); | |
3176 | ||
3177 | if (dev->rx_handler) | |
3178 | return -EBUSY; | |
3179 | ||
93e2c32b | 3180 | rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); |
ab95bfe0 JP |
3181 | rcu_assign_pointer(dev->rx_handler, rx_handler); |
3182 | ||
3183 | return 0; | |
3184 | } | |
3185 | EXPORT_SYMBOL_GPL(netdev_rx_handler_register); | |
3186 | ||
3187 | /** | |
3188 | * netdev_rx_handler_unregister - unregister receive handler | |
3189 | * @dev: device to unregister a handler from | |
3190 | * | |
3191 | * Unregister a receive hander from a device. | |
3192 | * | |
3193 | * The caller must hold the rtnl_mutex. | |
3194 | */ | |
3195 | void netdev_rx_handler_unregister(struct net_device *dev) | |
3196 | { | |
3197 | ||
3198 | ASSERT_RTNL(); | |
a9b3cd7f SH |
3199 | RCU_INIT_POINTER(dev->rx_handler, NULL); |
3200 | RCU_INIT_POINTER(dev->rx_handler_data, NULL); | |
ab95bfe0 JP |
3201 | } |
3202 | EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); | |
3203 | ||
b4b9e355 MG |
3204 | /* |
3205 | * Limit the use of PFMEMALLOC reserves to those protocols that implement | |
3206 | * the special handling of PFMEMALLOC skbs. | |
3207 | */ | |
3208 | static bool skb_pfmemalloc_protocol(struct sk_buff *skb) | |
3209 | { | |
3210 | switch (skb->protocol) { | |
3211 | case __constant_htons(ETH_P_ARP): | |
3212 | case __constant_htons(ETH_P_IP): | |
3213 | case __constant_htons(ETH_P_IPV6): | |
3214 | case __constant_htons(ETH_P_8021Q): | |
3215 | return true; | |
3216 | default: | |
3217 | return false; | |
3218 | } | |
3219 | } | |
3220 | ||
10f744d2 | 3221 | static int __netif_receive_skb(struct sk_buff *skb) |
1da177e4 LT |
3222 | { |
3223 | struct packet_type *ptype, *pt_prev; | |
ab95bfe0 | 3224 | rx_handler_func_t *rx_handler; |
f2ccd8fa | 3225 | struct net_device *orig_dev; |
63d8ea7f | 3226 | struct net_device *null_or_dev; |
8a4eb573 | 3227 | bool deliver_exact = false; |
1da177e4 | 3228 | int ret = NET_RX_DROP; |
252e3346 | 3229 | __be16 type; |
b4b9e355 | 3230 | unsigned long pflags = current->flags; |
1da177e4 | 3231 | |
588f0330 | 3232 | net_timestamp_check(!netdev_tstamp_prequeue, skb); |
81bbb3d4 | 3233 | |
cf66ba58 | 3234 | trace_netif_receive_skb(skb); |
9b22ea56 | 3235 | |
b4b9e355 MG |
3236 | /* |
3237 | * PFMEMALLOC skbs are special, they should | |
3238 | * - be delivered to SOCK_MEMALLOC sockets only | |
3239 | * - stay away from userspace | |
3240 | * - have bounded memory usage | |
3241 | * | |
3242 | * Use PF_MEMALLOC as this saves us from propagating the allocation | |
3243 | * context down to all allocation sites. | |
3244 | */ | |
3245 | if (sk_memalloc_socks() && skb_pfmemalloc(skb)) | |
3246 | current->flags |= PF_MEMALLOC; | |
3247 | ||
1da177e4 | 3248 | /* if we've gotten here through NAPI, check netpoll */ |
bea3348e | 3249 | if (netpoll_receive_skb(skb)) |
b4b9e355 | 3250 | goto out; |
1da177e4 | 3251 | |
cc9bd5ce | 3252 | orig_dev = skb->dev; |
8f903c70 | 3253 | |
c1d2bbe1 | 3254 | skb_reset_network_header(skb); |
badff6d0 | 3255 | skb_reset_transport_header(skb); |
0b5c9db1 | 3256 | skb_reset_mac_len(skb); |
1da177e4 LT |
3257 | |
3258 | pt_prev = NULL; | |
3259 | ||
3260 | rcu_read_lock(); | |
3261 | ||
63d8ea7f | 3262 | another_round: |
b6858177 | 3263 | skb->skb_iif = skb->dev->ifindex; |
63d8ea7f DM |
3264 | |
3265 | __this_cpu_inc(softnet_data.processed); | |
3266 | ||
bcc6d479 JP |
3267 | if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) { |
3268 | skb = vlan_untag(skb); | |
3269 | if (unlikely(!skb)) | |
b4b9e355 | 3270 | goto unlock; |
bcc6d479 JP |
3271 | } |
3272 | ||
1da177e4 LT |
3273 | #ifdef CONFIG_NET_CLS_ACT |
3274 | if (skb->tc_verd & TC_NCLS) { | |
3275 | skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); | |
3276 | goto ncls; | |
3277 | } | |
3278 | #endif | |
3279 | ||
b4b9e355 MG |
3280 | if (sk_memalloc_socks() && skb_pfmemalloc(skb)) |
3281 | goto skip_taps; | |
3282 | ||
1da177e4 | 3283 | list_for_each_entry_rcu(ptype, &ptype_all, list) { |
63d8ea7f | 3284 | if (!ptype->dev || ptype->dev == skb->dev) { |
4ec93edb | 3285 | if (pt_prev) |
f2ccd8fa | 3286 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
3287 | pt_prev = ptype; |
3288 | } | |
3289 | } | |
3290 | ||
b4b9e355 | 3291 | skip_taps: |
1da177e4 | 3292 | #ifdef CONFIG_NET_CLS_ACT |
f697c3e8 HX |
3293 | skb = handle_ing(skb, &pt_prev, &ret, orig_dev); |
3294 | if (!skb) | |
b4b9e355 | 3295 | goto unlock; |
1da177e4 LT |
3296 | ncls: |
3297 | #endif | |
3298 | ||
b4b9e355 MG |
3299 | if (sk_memalloc_socks() && skb_pfmemalloc(skb) |
3300 | && !skb_pfmemalloc_protocol(skb)) | |
3301 | goto drop; | |
3302 | ||
6a32e4f9 | 3303 | rx_handler = rcu_dereference(skb->dev->rx_handler); |
2425717b JF |
3304 | if (vlan_tx_tag_present(skb)) { |
3305 | if (pt_prev) { | |
3306 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3307 | pt_prev = NULL; | |
3308 | } | |
6a32e4f9 | 3309 | if (vlan_do_receive(&skb, !rx_handler)) |
2425717b JF |
3310 | goto another_round; |
3311 | else if (unlikely(!skb)) | |
b4b9e355 | 3312 | goto unlock; |
2425717b JF |
3313 | } |
3314 | ||
ab95bfe0 JP |
3315 | if (rx_handler) { |
3316 | if (pt_prev) { | |
3317 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3318 | pt_prev = NULL; | |
3319 | } | |
8a4eb573 JP |
3320 | switch (rx_handler(&skb)) { |
3321 | case RX_HANDLER_CONSUMED: | |
b4b9e355 | 3322 | goto unlock; |
8a4eb573 | 3323 | case RX_HANDLER_ANOTHER: |
63d8ea7f | 3324 | goto another_round; |
8a4eb573 JP |
3325 | case RX_HANDLER_EXACT: |
3326 | deliver_exact = true; | |
3327 | case RX_HANDLER_PASS: | |
3328 | break; | |
3329 | default: | |
3330 | BUG(); | |
3331 | } | |
ab95bfe0 | 3332 | } |
1da177e4 | 3333 | |
63d8ea7f | 3334 | /* deliver only exact match when indicated */ |
8a4eb573 | 3335 | null_or_dev = deliver_exact ? skb->dev : NULL; |
1f3c8804 | 3336 | |
1da177e4 | 3337 | type = skb->protocol; |
82d8a867 PE |
3338 | list_for_each_entry_rcu(ptype, |
3339 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
63d8ea7f | 3340 | if (ptype->type == type && |
e3f48d37 JP |
3341 | (ptype->dev == null_or_dev || ptype->dev == skb->dev || |
3342 | ptype->dev == orig_dev)) { | |
4ec93edb | 3343 | if (pt_prev) |
f2ccd8fa | 3344 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
3345 | pt_prev = ptype; |
3346 | } | |
3347 | } | |
3348 | ||
3349 | if (pt_prev) { | |
1080e512 | 3350 | if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) |
0e698bf6 | 3351 | goto drop; |
1080e512 MT |
3352 | else |
3353 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1da177e4 | 3354 | } else { |
b4b9e355 | 3355 | drop: |
caf586e5 | 3356 | atomic_long_inc(&skb->dev->rx_dropped); |
1da177e4 LT |
3357 | kfree_skb(skb); |
3358 | /* Jamal, now you will not able to escape explaining | |
3359 | * me how you were going to use this. :-) | |
3360 | */ | |
3361 | ret = NET_RX_DROP; | |
3362 | } | |
3363 | ||
b4b9e355 | 3364 | unlock: |
1da177e4 | 3365 | rcu_read_unlock(); |
b4b9e355 MG |
3366 | out: |
3367 | tsk_restore_flags(current, pflags, PF_MEMALLOC); | |
1da177e4 LT |
3368 | return ret; |
3369 | } | |
0a9627f2 TH |
3370 | |
3371 | /** | |
3372 | * netif_receive_skb - process receive buffer from network | |
3373 | * @skb: buffer to process | |
3374 | * | |
3375 | * netif_receive_skb() is the main receive data processing function. | |
3376 | * It always succeeds. The buffer may be dropped during processing | |
3377 | * for congestion control or by the protocol layers. | |
3378 | * | |
3379 | * This function may only be called from softirq context and interrupts | |
3380 | * should be enabled. | |
3381 | * | |
3382 | * Return values (usually ignored): | |
3383 | * NET_RX_SUCCESS: no congestion | |
3384 | * NET_RX_DROP: packet was dropped | |
3385 | */ | |
3386 | int netif_receive_skb(struct sk_buff *skb) | |
3387 | { | |
588f0330 | 3388 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
3b098e2d | 3389 | |
c1f19b51 RC |
3390 | if (skb_defer_rx_timestamp(skb)) |
3391 | return NET_RX_SUCCESS; | |
3392 | ||
df334545 | 3393 | #ifdef CONFIG_RPS |
c5905afb | 3394 | if (static_key_false(&rps_needed)) { |
3b098e2d ED |
3395 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
3396 | int cpu, ret; | |
fec5e652 | 3397 | |
3b098e2d ED |
3398 | rcu_read_lock(); |
3399 | ||
3400 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
0a9627f2 | 3401 | |
3b098e2d ED |
3402 | if (cpu >= 0) { |
3403 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
3404 | rcu_read_unlock(); | |
adc9300e | 3405 | return ret; |
3b098e2d | 3406 | } |
adc9300e | 3407 | rcu_read_unlock(); |
fec5e652 | 3408 | } |
1e94d72f | 3409 | #endif |
adc9300e | 3410 | return __netif_receive_skb(skb); |
0a9627f2 | 3411 | } |
d1b19dff | 3412 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 3413 | |
88751275 ED |
3414 | /* Network device is going away, flush any packets still pending |
3415 | * Called with irqs disabled. | |
3416 | */ | |
152102c7 | 3417 | static void flush_backlog(void *arg) |
6e583ce5 | 3418 | { |
152102c7 | 3419 | struct net_device *dev = arg; |
e36fa2f7 | 3420 | struct softnet_data *sd = &__get_cpu_var(softnet_data); |
6e583ce5 SH |
3421 | struct sk_buff *skb, *tmp; |
3422 | ||
e36fa2f7 | 3423 | rps_lock(sd); |
6e7676c1 | 3424 | skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { |
6e583ce5 | 3425 | if (skb->dev == dev) { |
e36fa2f7 | 3426 | __skb_unlink(skb, &sd->input_pkt_queue); |
6e583ce5 | 3427 | kfree_skb(skb); |
76cc8b13 | 3428 | input_queue_head_incr(sd); |
6e583ce5 | 3429 | } |
6e7676c1 | 3430 | } |
e36fa2f7 | 3431 | rps_unlock(sd); |
6e7676c1 CG |
3432 | |
3433 | skb_queue_walk_safe(&sd->process_queue, skb, tmp) { | |
3434 | if (skb->dev == dev) { | |
3435 | __skb_unlink(skb, &sd->process_queue); | |
3436 | kfree_skb(skb); | |
76cc8b13 | 3437 | input_queue_head_incr(sd); |
6e7676c1 CG |
3438 | } |
3439 | } | |
6e583ce5 SH |
3440 | } |
3441 | ||
d565b0a1 HX |
3442 | static int napi_gro_complete(struct sk_buff *skb) |
3443 | { | |
3444 | struct packet_type *ptype; | |
3445 | __be16 type = skb->protocol; | |
3446 | struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK]; | |
3447 | int err = -ENOENT; | |
3448 | ||
fc59f9a3 HX |
3449 | if (NAPI_GRO_CB(skb)->count == 1) { |
3450 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 3451 | goto out; |
fc59f9a3 | 3452 | } |
d565b0a1 HX |
3453 | |
3454 | rcu_read_lock(); | |
3455 | list_for_each_entry_rcu(ptype, head, list) { | |
3456 | if (ptype->type != type || ptype->dev || !ptype->gro_complete) | |
3457 | continue; | |
3458 | ||
3459 | err = ptype->gro_complete(skb); | |
3460 | break; | |
3461 | } | |
3462 | rcu_read_unlock(); | |
3463 | ||
3464 | if (err) { | |
3465 | WARN_ON(&ptype->list == head); | |
3466 | kfree_skb(skb); | |
3467 | return NET_RX_SUCCESS; | |
3468 | } | |
3469 | ||
3470 | out: | |
d565b0a1 HX |
3471 | return netif_receive_skb(skb); |
3472 | } | |
3473 | ||
2e71a6f8 ED |
3474 | /* napi->gro_list contains packets ordered by age. |
3475 | * youngest packets at the head of it. | |
3476 | * Complete skbs in reverse order to reduce latencies. | |
3477 | */ | |
3478 | void napi_gro_flush(struct napi_struct *napi, bool flush_old) | |
d565b0a1 | 3479 | { |
2e71a6f8 | 3480 | struct sk_buff *skb, *prev = NULL; |
d565b0a1 | 3481 | |
2e71a6f8 ED |
3482 | /* scan list and build reverse chain */ |
3483 | for (skb = napi->gro_list; skb != NULL; skb = skb->next) { | |
3484 | skb->prev = prev; | |
3485 | prev = skb; | |
3486 | } | |
3487 | ||
3488 | for (skb = prev; skb; skb = prev) { | |
d565b0a1 | 3489 | skb->next = NULL; |
2e71a6f8 ED |
3490 | |
3491 | if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) | |
3492 | return; | |
3493 | ||
3494 | prev = skb->prev; | |
d565b0a1 | 3495 | napi_gro_complete(skb); |
2e71a6f8 | 3496 | napi->gro_count--; |
d565b0a1 HX |
3497 | } |
3498 | ||
3499 | napi->gro_list = NULL; | |
3500 | } | |
86cac58b | 3501 | EXPORT_SYMBOL(napi_gro_flush); |
d565b0a1 | 3502 | |
5b252f0c | 3503 | enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 HX |
3504 | { |
3505 | struct sk_buff **pp = NULL; | |
3506 | struct packet_type *ptype; | |
3507 | __be16 type = skb->protocol; | |
3508 | struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK]; | |
0da2afd5 | 3509 | int same_flow; |
d565b0a1 | 3510 | int mac_len; |
5b252f0c | 3511 | enum gro_result ret; |
d565b0a1 | 3512 | |
ce9e76c8 | 3513 | if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb)) |
d565b0a1 HX |
3514 | goto normal; |
3515 | ||
21dc3301 | 3516 | if (skb_is_gso(skb) || skb_has_frag_list(skb)) |
f17f5c91 HX |
3517 | goto normal; |
3518 | ||
d565b0a1 HX |
3519 | rcu_read_lock(); |
3520 | list_for_each_entry_rcu(ptype, head, list) { | |
d565b0a1 HX |
3521 | if (ptype->type != type || ptype->dev || !ptype->gro_receive) |
3522 | continue; | |
3523 | ||
86911732 | 3524 | skb_set_network_header(skb, skb_gro_offset(skb)); |
d565b0a1 HX |
3525 | mac_len = skb->network_header - skb->mac_header; |
3526 | skb->mac_len = mac_len; | |
3527 | NAPI_GRO_CB(skb)->same_flow = 0; | |
3528 | NAPI_GRO_CB(skb)->flush = 0; | |
5d38a079 | 3529 | NAPI_GRO_CB(skb)->free = 0; |
d565b0a1 | 3530 | |
d565b0a1 HX |
3531 | pp = ptype->gro_receive(&napi->gro_list, skb); |
3532 | break; | |
3533 | } | |
3534 | rcu_read_unlock(); | |
3535 | ||
3536 | if (&ptype->list == head) | |
3537 | goto normal; | |
3538 | ||
0da2afd5 | 3539 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 3540 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 3541 | |
d565b0a1 HX |
3542 | if (pp) { |
3543 | struct sk_buff *nskb = *pp; | |
3544 | ||
3545 | *pp = nskb->next; | |
3546 | nskb->next = NULL; | |
3547 | napi_gro_complete(nskb); | |
4ae5544f | 3548 | napi->gro_count--; |
d565b0a1 HX |
3549 | } |
3550 | ||
0da2afd5 | 3551 | if (same_flow) |
d565b0a1 HX |
3552 | goto ok; |
3553 | ||
4ae5544f | 3554 | if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS) |
d565b0a1 | 3555 | goto normal; |
d565b0a1 | 3556 | |
4ae5544f | 3557 | napi->gro_count++; |
d565b0a1 | 3558 | NAPI_GRO_CB(skb)->count = 1; |
2e71a6f8 | 3559 | NAPI_GRO_CB(skb)->age = jiffies; |
86911732 | 3560 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
d565b0a1 HX |
3561 | skb->next = napi->gro_list; |
3562 | napi->gro_list = skb; | |
5d0d9be8 | 3563 | ret = GRO_HELD; |
d565b0a1 | 3564 | |
ad0f9904 | 3565 | pull: |
cb18978c HX |
3566 | if (skb_headlen(skb) < skb_gro_offset(skb)) { |
3567 | int grow = skb_gro_offset(skb) - skb_headlen(skb); | |
3568 | ||
3569 | BUG_ON(skb->end - skb->tail < grow); | |
3570 | ||
3571 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
3572 | ||
3573 | skb->tail += grow; | |
3574 | skb->data_len -= grow; | |
3575 | ||
3576 | skb_shinfo(skb)->frags[0].page_offset += grow; | |
9e903e08 | 3577 | skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow); |
cb18978c | 3578 | |
9e903e08 | 3579 | if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) { |
ea2ab693 | 3580 | skb_frag_unref(skb, 0); |
cb18978c HX |
3581 | memmove(skb_shinfo(skb)->frags, |
3582 | skb_shinfo(skb)->frags + 1, | |
e5093aec | 3583 | --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t)); |
cb18978c | 3584 | } |
ad0f9904 HX |
3585 | } |
3586 | ||
d565b0a1 | 3587 | ok: |
5d0d9be8 | 3588 | return ret; |
d565b0a1 HX |
3589 | |
3590 | normal: | |
ad0f9904 HX |
3591 | ret = GRO_NORMAL; |
3592 | goto pull; | |
5d38a079 | 3593 | } |
96e93eab HX |
3594 | EXPORT_SYMBOL(dev_gro_receive); |
3595 | ||
40d0802b | 3596 | static inline gro_result_t |
5b252f0c | 3597 | __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab HX |
3598 | { |
3599 | struct sk_buff *p; | |
5ca3b72c | 3600 | unsigned int maclen = skb->dev->hard_header_len; |
96e93eab HX |
3601 | |
3602 | for (p = napi->gro_list; p; p = p->next) { | |
40d0802b ED |
3603 | unsigned long diffs; |
3604 | ||
3605 | diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; | |
3701e513 | 3606 | diffs |= p->vlan_tci ^ skb->vlan_tci; |
5ca3b72c ED |
3607 | if (maclen == ETH_HLEN) |
3608 | diffs |= compare_ether_header(skb_mac_header(p), | |
3609 | skb_gro_mac_header(skb)); | |
3610 | else if (!diffs) | |
3611 | diffs = memcmp(skb_mac_header(p), | |
3612 | skb_gro_mac_header(skb), | |
3613 | maclen); | |
40d0802b | 3614 | NAPI_GRO_CB(p)->same_flow = !diffs; |
96e93eab HX |
3615 | NAPI_GRO_CB(p)->flush = 0; |
3616 | } | |
3617 | ||
3618 | return dev_gro_receive(napi, skb); | |
3619 | } | |
5d38a079 | 3620 | |
c7c4b3b6 | 3621 | gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) |
5d38a079 | 3622 | { |
5d0d9be8 HX |
3623 | switch (ret) { |
3624 | case GRO_NORMAL: | |
c7c4b3b6 BH |
3625 | if (netif_receive_skb(skb)) |
3626 | ret = GRO_DROP; | |
3627 | break; | |
5d38a079 | 3628 | |
5d0d9be8 | 3629 | case GRO_DROP: |
5d38a079 HX |
3630 | kfree_skb(skb); |
3631 | break; | |
5b252f0c | 3632 | |
daa86548 | 3633 | case GRO_MERGED_FREE: |
d7e8883c ED |
3634 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) |
3635 | kmem_cache_free(skbuff_head_cache, skb); | |
3636 | else | |
3637 | __kfree_skb(skb); | |
daa86548 ED |
3638 | break; |
3639 | ||
5b252f0c BH |
3640 | case GRO_HELD: |
3641 | case GRO_MERGED: | |
3642 | break; | |
5d38a079 HX |
3643 | } |
3644 | ||
c7c4b3b6 | 3645 | return ret; |
5d0d9be8 HX |
3646 | } |
3647 | EXPORT_SYMBOL(napi_skb_finish); | |
3648 | ||
ca07e43e | 3649 | static void skb_gro_reset_offset(struct sk_buff *skb) |
78a478d0 | 3650 | { |
ca07e43e ED |
3651 | const struct skb_shared_info *pinfo = skb_shinfo(skb); |
3652 | const skb_frag_t *frag0 = &pinfo->frags[0]; | |
3653 | ||
78a478d0 HX |
3654 | NAPI_GRO_CB(skb)->data_offset = 0; |
3655 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
7489594c | 3656 | NAPI_GRO_CB(skb)->frag0_len = 0; |
78a478d0 | 3657 | |
78d3fd0b | 3658 | if (skb->mac_header == skb->tail && |
ca07e43e ED |
3659 | pinfo->nr_frags && |
3660 | !PageHighMem(skb_frag_page(frag0))) { | |
3661 | NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); | |
3662 | NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0); | |
7489594c | 3663 | } |
78a478d0 | 3664 | } |
78a478d0 | 3665 | |
c7c4b3b6 | 3666 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 3667 | { |
86911732 HX |
3668 | skb_gro_reset_offset(skb); |
3669 | ||
5d0d9be8 | 3670 | return napi_skb_finish(__napi_gro_receive(napi, skb), skb); |
d565b0a1 HX |
3671 | } |
3672 | EXPORT_SYMBOL(napi_gro_receive); | |
3673 | ||
d0c2b0d2 | 3674 | static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab | 3675 | { |
96e93eab | 3676 | __skb_pull(skb, skb_headlen(skb)); |
2a2a459e ED |
3677 | /* restore the reserve we had after netdev_alloc_skb_ip_align() */ |
3678 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); | |
3701e513 | 3679 | skb->vlan_tci = 0; |
66c46d74 | 3680 | skb->dev = napi->dev; |
6d152e23 | 3681 | skb->skb_iif = 0; |
96e93eab HX |
3682 | |
3683 | napi->skb = skb; | |
3684 | } | |
96e93eab | 3685 | |
76620aaf | 3686 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 3687 | { |
5d38a079 | 3688 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
3689 | |
3690 | if (!skb) { | |
89d71a66 ED |
3691 | skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD); |
3692 | if (skb) | |
3693 | napi->skb = skb; | |
80595d59 | 3694 | } |
96e93eab HX |
3695 | return skb; |
3696 | } | |
76620aaf | 3697 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 3698 | |
c7c4b3b6 BH |
3699 | gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb, |
3700 | gro_result_t ret) | |
96e93eab | 3701 | { |
5d0d9be8 HX |
3702 | switch (ret) { |
3703 | case GRO_NORMAL: | |
86911732 | 3704 | case GRO_HELD: |
e76b69cc | 3705 | skb->protocol = eth_type_trans(skb, skb->dev); |
86911732 | 3706 | |
c7c4b3b6 BH |
3707 | if (ret == GRO_HELD) |
3708 | skb_gro_pull(skb, -ETH_HLEN); | |
3709 | else if (netif_receive_skb(skb)) | |
3710 | ret = GRO_DROP; | |
86911732 | 3711 | break; |
5d38a079 | 3712 | |
5d0d9be8 | 3713 | case GRO_DROP: |
5d0d9be8 HX |
3714 | case GRO_MERGED_FREE: |
3715 | napi_reuse_skb(napi, skb); | |
3716 | break; | |
5b252f0c BH |
3717 | |
3718 | case GRO_MERGED: | |
3719 | break; | |
5d0d9be8 | 3720 | } |
5d38a079 | 3721 | |
c7c4b3b6 | 3722 | return ret; |
5d38a079 | 3723 | } |
5d0d9be8 HX |
3724 | EXPORT_SYMBOL(napi_frags_finish); |
3725 | ||
4adb9c4a | 3726 | static struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
76620aaf HX |
3727 | { |
3728 | struct sk_buff *skb = napi->skb; | |
3729 | struct ethhdr *eth; | |
a5b1cf28 HX |
3730 | unsigned int hlen; |
3731 | unsigned int off; | |
76620aaf HX |
3732 | |
3733 | napi->skb = NULL; | |
3734 | ||
3735 | skb_reset_mac_header(skb); | |
3736 | skb_gro_reset_offset(skb); | |
3737 | ||
a5b1cf28 HX |
3738 | off = skb_gro_offset(skb); |
3739 | hlen = off + sizeof(*eth); | |
3740 | eth = skb_gro_header_fast(skb, off); | |
3741 | if (skb_gro_header_hard(skb, hlen)) { | |
3742 | eth = skb_gro_header_slow(skb, hlen, off); | |
3743 | if (unlikely(!eth)) { | |
3744 | napi_reuse_skb(napi, skb); | |
3745 | skb = NULL; | |
3746 | goto out; | |
3747 | } | |
76620aaf HX |
3748 | } |
3749 | ||
3750 | skb_gro_pull(skb, sizeof(*eth)); | |
3751 | ||
3752 | /* | |
3753 | * This works because the only protocols we care about don't require | |
3754 | * special handling. We'll fix it up properly at the end. | |
3755 | */ | |
3756 | skb->protocol = eth->h_proto; | |
3757 | ||
3758 | out: | |
3759 | return skb; | |
3760 | } | |
76620aaf | 3761 | |
c7c4b3b6 | 3762 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 3763 | { |
76620aaf | 3764 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 HX |
3765 | |
3766 | if (!skb) | |
c7c4b3b6 | 3767 | return GRO_DROP; |
5d0d9be8 HX |
3768 | |
3769 | return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb)); | |
3770 | } | |
5d38a079 HX |
3771 | EXPORT_SYMBOL(napi_gro_frags); |
3772 | ||
e326bed2 ED |
3773 | /* |
3774 | * net_rps_action sends any pending IPI's for rps. | |
3775 | * Note: called with local irq disabled, but exits with local irq enabled. | |
3776 | */ | |
3777 | static void net_rps_action_and_irq_enable(struct softnet_data *sd) | |
3778 | { | |
3779 | #ifdef CONFIG_RPS | |
3780 | struct softnet_data *remsd = sd->rps_ipi_list; | |
3781 | ||
3782 | if (remsd) { | |
3783 | sd->rps_ipi_list = NULL; | |
3784 | ||
3785 | local_irq_enable(); | |
3786 | ||
3787 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
3788 | while (remsd) { | |
3789 | struct softnet_data *next = remsd->rps_ipi_next; | |
3790 | ||
3791 | if (cpu_online(remsd->cpu)) | |
3792 | __smp_call_function_single(remsd->cpu, | |
3793 | &remsd->csd, 0); | |
3794 | remsd = next; | |
3795 | } | |
3796 | } else | |
3797 | #endif | |
3798 | local_irq_enable(); | |
3799 | } | |
3800 | ||
bea3348e | 3801 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 LT |
3802 | { |
3803 | int work = 0; | |
eecfd7c4 | 3804 | struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); |
1da177e4 | 3805 | |
e326bed2 ED |
3806 | #ifdef CONFIG_RPS |
3807 | /* Check if we have pending ipi, its better to send them now, | |
3808 | * not waiting net_rx_action() end. | |
3809 | */ | |
3810 | if (sd->rps_ipi_list) { | |
3811 | local_irq_disable(); | |
3812 | net_rps_action_and_irq_enable(sd); | |
3813 | } | |
3814 | #endif | |
bea3348e | 3815 | napi->weight = weight_p; |
6e7676c1 CG |
3816 | local_irq_disable(); |
3817 | while (work < quota) { | |
1da177e4 | 3818 | struct sk_buff *skb; |
6e7676c1 CG |
3819 | unsigned int qlen; |
3820 | ||
3821 | while ((skb = __skb_dequeue(&sd->process_queue))) { | |
3822 | local_irq_enable(); | |
3823 | __netif_receive_skb(skb); | |
6e7676c1 | 3824 | local_irq_disable(); |
76cc8b13 TH |
3825 | input_queue_head_incr(sd); |
3826 | if (++work >= quota) { | |
3827 | local_irq_enable(); | |
3828 | return work; | |
3829 | } | |
6e7676c1 | 3830 | } |
1da177e4 | 3831 | |
e36fa2f7 | 3832 | rps_lock(sd); |
6e7676c1 | 3833 | qlen = skb_queue_len(&sd->input_pkt_queue); |
76cc8b13 | 3834 | if (qlen) |
6e7676c1 CG |
3835 | skb_queue_splice_tail_init(&sd->input_pkt_queue, |
3836 | &sd->process_queue); | |
76cc8b13 | 3837 | |
6e7676c1 | 3838 | if (qlen < quota - work) { |
eecfd7c4 ED |
3839 | /* |
3840 | * Inline a custom version of __napi_complete(). | |
3841 | * only current cpu owns and manipulates this napi, | |
3842 | * and NAPI_STATE_SCHED is the only possible flag set on backlog. | |
3843 | * we can use a plain write instead of clear_bit(), | |
3844 | * and we dont need an smp_mb() memory barrier. | |
3845 | */ | |
3846 | list_del(&napi->poll_list); | |
3847 | napi->state = 0; | |
3848 | ||
6e7676c1 | 3849 | quota = work + qlen; |
bea3348e | 3850 | } |
e36fa2f7 | 3851 | rps_unlock(sd); |
6e7676c1 CG |
3852 | } |
3853 | local_irq_enable(); | |
1da177e4 | 3854 | |
bea3348e SH |
3855 | return work; |
3856 | } | |
1da177e4 | 3857 | |
bea3348e SH |
3858 | /** |
3859 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 3860 | * @n: entry to schedule |
bea3348e SH |
3861 | * |
3862 | * The entry's receive function will be scheduled to run | |
3863 | */ | |
b5606c2d | 3864 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
3865 | { |
3866 | unsigned long flags; | |
1da177e4 | 3867 | |
bea3348e | 3868 | local_irq_save(flags); |
eecfd7c4 | 3869 | ____napi_schedule(&__get_cpu_var(softnet_data), n); |
bea3348e | 3870 | local_irq_restore(flags); |
1da177e4 | 3871 | } |
bea3348e SH |
3872 | EXPORT_SYMBOL(__napi_schedule); |
3873 | ||
d565b0a1 HX |
3874 | void __napi_complete(struct napi_struct *n) |
3875 | { | |
3876 | BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); | |
3877 | BUG_ON(n->gro_list); | |
3878 | ||
3879 | list_del(&n->poll_list); | |
3880 | smp_mb__before_clear_bit(); | |
3881 | clear_bit(NAPI_STATE_SCHED, &n->state); | |
3882 | } | |
3883 | EXPORT_SYMBOL(__napi_complete); | |
3884 | ||
3885 | void napi_complete(struct napi_struct *n) | |
3886 | { | |
3887 | unsigned long flags; | |
3888 | ||
3889 | /* | |
3890 | * don't let napi dequeue from the cpu poll list | |
3891 | * just in case its running on a different cpu | |
3892 | */ | |
3893 | if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state))) | |
3894 | return; | |
3895 | ||
2e71a6f8 | 3896 | napi_gro_flush(n, false); |
d565b0a1 HX |
3897 | local_irq_save(flags); |
3898 | __napi_complete(n); | |
3899 | local_irq_restore(flags); | |
3900 | } | |
3901 | EXPORT_SYMBOL(napi_complete); | |
3902 | ||
3903 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, | |
3904 | int (*poll)(struct napi_struct *, int), int weight) | |
3905 | { | |
3906 | INIT_LIST_HEAD(&napi->poll_list); | |
4ae5544f | 3907 | napi->gro_count = 0; |
d565b0a1 | 3908 | napi->gro_list = NULL; |
5d38a079 | 3909 | napi->skb = NULL; |
d565b0a1 HX |
3910 | napi->poll = poll; |
3911 | napi->weight = weight; | |
3912 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 3913 | napi->dev = dev; |
5d38a079 | 3914 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
3915 | spin_lock_init(&napi->poll_lock); |
3916 | napi->poll_owner = -1; | |
3917 | #endif | |
3918 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
3919 | } | |
3920 | EXPORT_SYMBOL(netif_napi_add); | |
3921 | ||
3922 | void netif_napi_del(struct napi_struct *napi) | |
3923 | { | |
3924 | struct sk_buff *skb, *next; | |
3925 | ||
d7b06636 | 3926 | list_del_init(&napi->dev_list); |
76620aaf | 3927 | napi_free_frags(napi); |
d565b0a1 HX |
3928 | |
3929 | for (skb = napi->gro_list; skb; skb = next) { | |
3930 | next = skb->next; | |
3931 | skb->next = NULL; | |
3932 | kfree_skb(skb); | |
3933 | } | |
3934 | ||
3935 | napi->gro_list = NULL; | |
4ae5544f | 3936 | napi->gro_count = 0; |
d565b0a1 HX |
3937 | } |
3938 | EXPORT_SYMBOL(netif_napi_del); | |
3939 | ||
1da177e4 LT |
3940 | static void net_rx_action(struct softirq_action *h) |
3941 | { | |
e326bed2 | 3942 | struct softnet_data *sd = &__get_cpu_var(softnet_data); |
24f8b238 | 3943 | unsigned long time_limit = jiffies + 2; |
51b0bded | 3944 | int budget = netdev_budget; |
53fb95d3 MM |
3945 | void *have; |
3946 | ||
1da177e4 LT |
3947 | local_irq_disable(); |
3948 | ||
e326bed2 | 3949 | while (!list_empty(&sd->poll_list)) { |
bea3348e SH |
3950 | struct napi_struct *n; |
3951 | int work, weight; | |
1da177e4 | 3952 | |
bea3348e | 3953 | /* If softirq window is exhuasted then punt. |
24f8b238 SH |
3954 | * Allow this to run for 2 jiffies since which will allow |
3955 | * an average latency of 1.5/HZ. | |
bea3348e | 3956 | */ |
24f8b238 | 3957 | if (unlikely(budget <= 0 || time_after(jiffies, time_limit))) |
1da177e4 LT |
3958 | goto softnet_break; |
3959 | ||
3960 | local_irq_enable(); | |
3961 | ||
bea3348e SH |
3962 | /* Even though interrupts have been re-enabled, this |
3963 | * access is safe because interrupts can only add new | |
3964 | * entries to the tail of this list, and only ->poll() | |
3965 | * calls can remove this head entry from the list. | |
3966 | */ | |
e326bed2 | 3967 | n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list); |
1da177e4 | 3968 | |
bea3348e SH |
3969 | have = netpoll_poll_lock(n); |
3970 | ||
3971 | weight = n->weight; | |
3972 | ||
0a7606c1 DM |
3973 | /* This NAPI_STATE_SCHED test is for avoiding a race |
3974 | * with netpoll's poll_napi(). Only the entity which | |
3975 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
3976 | * actually make the ->poll() call. Therefore we avoid | |
25985edc | 3977 | * accidentally calling ->poll() when NAPI is not scheduled. |
0a7606c1 DM |
3978 | */ |
3979 | work = 0; | |
4ea7e386 | 3980 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { |
0a7606c1 | 3981 | work = n->poll(n, weight); |
4ea7e386 NH |
3982 | trace_napi_poll(n); |
3983 | } | |
bea3348e SH |
3984 | |
3985 | WARN_ON_ONCE(work > weight); | |
3986 | ||
3987 | budget -= work; | |
3988 | ||
3989 | local_irq_disable(); | |
3990 | ||
3991 | /* Drivers must not modify the NAPI state if they | |
3992 | * consume the entire weight. In such cases this code | |
3993 | * still "owns" the NAPI instance and therefore can | |
3994 | * move the instance around on the list at-will. | |
3995 | */ | |
fed17f30 | 3996 | if (unlikely(work == weight)) { |
ff780cd8 HX |
3997 | if (unlikely(napi_disable_pending(n))) { |
3998 | local_irq_enable(); | |
3999 | napi_complete(n); | |
4000 | local_irq_disable(); | |
2e71a6f8 ED |
4001 | } else { |
4002 | if (n->gro_list) { | |
4003 | /* flush too old packets | |
4004 | * If HZ < 1000, flush all packets. | |
4005 | */ | |
4006 | local_irq_enable(); | |
4007 | napi_gro_flush(n, HZ >= 1000); | |
4008 | local_irq_disable(); | |
4009 | } | |
e326bed2 | 4010 | list_move_tail(&n->poll_list, &sd->poll_list); |
2e71a6f8 | 4011 | } |
fed17f30 | 4012 | } |
bea3348e SH |
4013 | |
4014 | netpoll_poll_unlock(have); | |
1da177e4 LT |
4015 | } |
4016 | out: | |
e326bed2 | 4017 | net_rps_action_and_irq_enable(sd); |
0a9627f2 | 4018 | |
db217334 CL |
4019 | #ifdef CONFIG_NET_DMA |
4020 | /* | |
4021 | * There may not be any more sk_buffs coming right now, so push | |
4022 | * any pending DMA copies to hardware | |
4023 | */ | |
2ba05622 | 4024 | dma_issue_pending_all(); |
db217334 | 4025 | #endif |
bea3348e | 4026 | |
1da177e4 LT |
4027 | return; |
4028 | ||
4029 | softnet_break: | |
dee42870 | 4030 | sd->time_squeeze++; |
1da177e4 LT |
4031 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); |
4032 | goto out; | |
4033 | } | |
4034 | ||
d1b19dff | 4035 | static gifconf_func_t *gifconf_list[NPROTO]; |
1da177e4 LT |
4036 | |
4037 | /** | |
4038 | * register_gifconf - register a SIOCGIF handler | |
4039 | * @family: Address family | |
4040 | * @gifconf: Function handler | |
4041 | * | |
4042 | * Register protocol dependent address dumping routines. The handler | |
4043 | * that is passed must not be freed or reused until it has been replaced | |
4044 | * by another handler. | |
4045 | */ | |
d1b19dff | 4046 | int register_gifconf(unsigned int family, gifconf_func_t *gifconf) |
1da177e4 LT |
4047 | { |
4048 | if (family >= NPROTO) | |
4049 | return -EINVAL; | |
4050 | gifconf_list[family] = gifconf; | |
4051 | return 0; | |
4052 | } | |
d1b19dff | 4053 | EXPORT_SYMBOL(register_gifconf); |
1da177e4 LT |
4054 | |
4055 | ||
4056 | /* | |
4057 | * Map an interface index to its name (SIOCGIFNAME) | |
4058 | */ | |
4059 | ||
4060 | /* | |
4061 | * We need this ioctl for efficient implementation of the | |
4062 | * if_indextoname() function required by the IPv6 API. Without | |
4063 | * it, we would have to search all the interfaces to find a | |
4064 | * match. --pb | |
4065 | */ | |
4066 | ||
881d966b | 4067 | static int dev_ifname(struct net *net, struct ifreq __user *arg) |
1da177e4 LT |
4068 | { |
4069 | struct net_device *dev; | |
4070 | struct ifreq ifr; | |
4071 | ||
4072 | /* | |
4073 | * Fetch the caller's info block. | |
4074 | */ | |
4075 | ||
4076 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
4077 | return -EFAULT; | |
4078 | ||
fb699dfd ED |
4079 | rcu_read_lock(); |
4080 | dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex); | |
1da177e4 | 4081 | if (!dev) { |
fb699dfd | 4082 | rcu_read_unlock(); |
1da177e4 LT |
4083 | return -ENODEV; |
4084 | } | |
4085 | ||
4086 | strcpy(ifr.ifr_name, dev->name); | |
fb699dfd | 4087 | rcu_read_unlock(); |
1da177e4 LT |
4088 | |
4089 | if (copy_to_user(arg, &ifr, sizeof(struct ifreq))) | |
4090 | return -EFAULT; | |
4091 | return 0; | |
4092 | } | |
4093 | ||
4094 | /* | |
4095 | * Perform a SIOCGIFCONF call. This structure will change | |
4096 | * size eventually, and there is nothing I can do about it. | |
4097 | * Thus we will need a 'compatibility mode'. | |
4098 | */ | |
4099 | ||
881d966b | 4100 | static int dev_ifconf(struct net *net, char __user *arg) |
1da177e4 LT |
4101 | { |
4102 | struct ifconf ifc; | |
4103 | struct net_device *dev; | |
4104 | char __user *pos; | |
4105 | int len; | |
4106 | int total; | |
4107 | int i; | |
4108 | ||
4109 | /* | |
4110 | * Fetch the caller's info block. | |
4111 | */ | |
4112 | ||
4113 | if (copy_from_user(&ifc, arg, sizeof(struct ifconf))) | |
4114 | return -EFAULT; | |
4115 | ||
4116 | pos = ifc.ifc_buf; | |
4117 | len = ifc.ifc_len; | |
4118 | ||
4119 | /* | |
4120 | * Loop over the interfaces, and write an info block for each. | |
4121 | */ | |
4122 | ||
4123 | total = 0; | |
881d966b | 4124 | for_each_netdev(net, dev) { |
1da177e4 LT |
4125 | for (i = 0; i < NPROTO; i++) { |
4126 | if (gifconf_list[i]) { | |
4127 | int done; | |
4128 | if (!pos) | |
4129 | done = gifconf_list[i](dev, NULL, 0); | |
4130 | else | |
4131 | done = gifconf_list[i](dev, pos + total, | |
4132 | len - total); | |
4133 | if (done < 0) | |
4134 | return -EFAULT; | |
4135 | total += done; | |
4136 | } | |
4137 | } | |
4ec93edb | 4138 | } |
1da177e4 LT |
4139 | |
4140 | /* | |
4141 | * All done. Write the updated control block back to the caller. | |
4142 | */ | |
4143 | ifc.ifc_len = total; | |
4144 | ||
4145 | /* | |
4146 | * Both BSD and Solaris return 0 here, so we do too. | |
4147 | */ | |
4148 | return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0; | |
4149 | } | |
4150 | ||
4151 | #ifdef CONFIG_PROC_FS | |
f04565dd | 4152 | |
2def16ae | 4153 | #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1) |
f04565dd MM |
4154 | |
4155 | #define get_bucket(x) ((x) >> BUCKET_SPACE) | |
4156 | #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1)) | |
4157 | #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o)) | |
4158 | ||
2def16ae | 4159 | static inline struct net_device *dev_from_same_bucket(struct seq_file *seq, loff_t *pos) |
f04565dd | 4160 | { |
f04565dd MM |
4161 | struct net *net = seq_file_net(seq); |
4162 | struct net_device *dev; | |
4163 | struct hlist_node *p; | |
4164 | struct hlist_head *h; | |
2def16ae | 4165 | unsigned int count = 0, offset = get_offset(*pos); |
f04565dd | 4166 | |
2def16ae | 4167 | h = &net->dev_name_head[get_bucket(*pos)]; |
f04565dd | 4168 | hlist_for_each_entry_rcu(dev, p, h, name_hlist) { |
2def16ae | 4169 | if (++count == offset) |
f04565dd | 4170 | return dev; |
f04565dd MM |
4171 | } |
4172 | ||
4173 | return NULL; | |
4174 | } | |
4175 | ||
2def16ae | 4176 | static inline struct net_device *dev_from_bucket(struct seq_file *seq, loff_t *pos) |
f04565dd | 4177 | { |
f04565dd MM |
4178 | struct net_device *dev; |
4179 | unsigned int bucket; | |
4180 | ||
f04565dd | 4181 | do { |
2def16ae | 4182 | dev = dev_from_same_bucket(seq, pos); |
f04565dd MM |
4183 | if (dev) |
4184 | return dev; | |
4185 | ||
2def16ae ED |
4186 | bucket = get_bucket(*pos) + 1; |
4187 | *pos = set_bucket_offset(bucket, 1); | |
f04565dd MM |
4188 | } while (bucket < NETDEV_HASHENTRIES); |
4189 | ||
4190 | return NULL; | |
4191 | } | |
4192 | ||
1da177e4 LT |
4193 | /* |
4194 | * This is invoked by the /proc filesystem handler to display a device | |
4195 | * in detail. | |
4196 | */ | |
7562f876 | 4197 | void *dev_seq_start(struct seq_file *seq, loff_t *pos) |
c6d14c84 | 4198 | __acquires(RCU) |
1da177e4 | 4199 | { |
c6d14c84 | 4200 | rcu_read_lock(); |
7562f876 PE |
4201 | if (!*pos) |
4202 | return SEQ_START_TOKEN; | |
1da177e4 | 4203 | |
2def16ae | 4204 | if (get_bucket(*pos) >= NETDEV_HASHENTRIES) |
f04565dd | 4205 | return NULL; |
1da177e4 | 4206 | |
2def16ae | 4207 | return dev_from_bucket(seq, pos); |
1da177e4 LT |
4208 | } |
4209 | ||
4210 | void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
4211 | { | |
f04565dd | 4212 | ++*pos; |
2def16ae | 4213 | return dev_from_bucket(seq, pos); |
1da177e4 LT |
4214 | } |
4215 | ||
4216 | void dev_seq_stop(struct seq_file *seq, void *v) | |
c6d14c84 | 4217 | __releases(RCU) |
1da177e4 | 4218 | { |
c6d14c84 | 4219 | rcu_read_unlock(); |
1da177e4 LT |
4220 | } |
4221 | ||
4222 | static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev) | |
4223 | { | |
28172739 ED |
4224 | struct rtnl_link_stats64 temp; |
4225 | const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp); | |
1da177e4 | 4226 | |
be1f3c2c BH |
4227 | seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu " |
4228 | "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n", | |
5a1b5898 RR |
4229 | dev->name, stats->rx_bytes, stats->rx_packets, |
4230 | stats->rx_errors, | |
4231 | stats->rx_dropped + stats->rx_missed_errors, | |
4232 | stats->rx_fifo_errors, | |
4233 | stats->rx_length_errors + stats->rx_over_errors + | |
4234 | stats->rx_crc_errors + stats->rx_frame_errors, | |
4235 | stats->rx_compressed, stats->multicast, | |
4236 | stats->tx_bytes, stats->tx_packets, | |
4237 | stats->tx_errors, stats->tx_dropped, | |
4238 | stats->tx_fifo_errors, stats->collisions, | |
4239 | stats->tx_carrier_errors + | |
4240 | stats->tx_aborted_errors + | |
4241 | stats->tx_window_errors + | |
4242 | stats->tx_heartbeat_errors, | |
4243 | stats->tx_compressed); | |
1da177e4 LT |
4244 | } |
4245 | ||
4246 | /* | |
4247 | * Called from the PROCfs module. This now uses the new arbitrary sized | |
4248 | * /proc/net interface to create /proc/net/dev | |
4249 | */ | |
4250 | static int dev_seq_show(struct seq_file *seq, void *v) | |
4251 | { | |
4252 | if (v == SEQ_START_TOKEN) | |
4253 | seq_puts(seq, "Inter-| Receive " | |
4254 | " | Transmit\n" | |
4255 | " face |bytes packets errs drop fifo frame " | |
4256 | "compressed multicast|bytes packets errs " | |
4257 | "drop fifo colls carrier compressed\n"); | |
4258 | else | |
4259 | dev_seq_printf_stats(seq, v); | |
4260 | return 0; | |
4261 | } | |
4262 | ||
dee42870 | 4263 | static struct softnet_data *softnet_get_online(loff_t *pos) |
1da177e4 | 4264 | { |
dee42870 | 4265 | struct softnet_data *sd = NULL; |
1da177e4 | 4266 | |
0c0b0aca | 4267 | while (*pos < nr_cpu_ids) |
4ec93edb | 4268 | if (cpu_online(*pos)) { |
dee42870 | 4269 | sd = &per_cpu(softnet_data, *pos); |
1da177e4 LT |
4270 | break; |
4271 | } else | |
4272 | ++*pos; | |
dee42870 | 4273 | return sd; |
1da177e4 LT |
4274 | } |
4275 | ||
4276 | static void *softnet_seq_start(struct seq_file *seq, loff_t *pos) | |
4277 | { | |
4278 | return softnet_get_online(pos); | |
4279 | } | |
4280 | ||
4281 | static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
4282 | { | |
4283 | ++*pos; | |
4284 | return softnet_get_online(pos); | |
4285 | } | |
4286 | ||
4287 | static void softnet_seq_stop(struct seq_file *seq, void *v) | |
4288 | { | |
4289 | } | |
4290 | ||
4291 | static int softnet_seq_show(struct seq_file *seq, void *v) | |
4292 | { | |
dee42870 | 4293 | struct softnet_data *sd = v; |
1da177e4 | 4294 | |
0a9627f2 | 4295 | seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n", |
dee42870 | 4296 | sd->processed, sd->dropped, sd->time_squeeze, 0, |
c1ebcdb8 | 4297 | 0, 0, 0, 0, /* was fastroute */ |
dee42870 | 4298 | sd->cpu_collision, sd->received_rps); |
1da177e4 LT |
4299 | return 0; |
4300 | } | |
4301 | ||
f690808e | 4302 | static const struct seq_operations dev_seq_ops = { |
1da177e4 LT |
4303 | .start = dev_seq_start, |
4304 | .next = dev_seq_next, | |
4305 | .stop = dev_seq_stop, | |
4306 | .show = dev_seq_show, | |
4307 | }; | |
4308 | ||
4309 | static int dev_seq_open(struct inode *inode, struct file *file) | |
4310 | { | |
e372c414 | 4311 | return seq_open_net(inode, file, &dev_seq_ops, |
2def16ae | 4312 | sizeof(struct seq_net_private)); |
5cac98dd AB |
4313 | } |
4314 | ||
9a32144e | 4315 | static const struct file_operations dev_seq_fops = { |
1da177e4 LT |
4316 | .owner = THIS_MODULE, |
4317 | .open = dev_seq_open, | |
4318 | .read = seq_read, | |
4319 | .llseek = seq_lseek, | |
e372c414 | 4320 | .release = seq_release_net, |
1da177e4 LT |
4321 | }; |
4322 | ||
f690808e | 4323 | static const struct seq_operations softnet_seq_ops = { |
1da177e4 LT |
4324 | .start = softnet_seq_start, |
4325 | .next = softnet_seq_next, | |
4326 | .stop = softnet_seq_stop, | |
4327 | .show = softnet_seq_show, | |
4328 | }; | |
4329 | ||
4330 | static int softnet_seq_open(struct inode *inode, struct file *file) | |
4331 | { | |
4332 | return seq_open(file, &softnet_seq_ops); | |
4333 | } | |
4334 | ||
9a32144e | 4335 | static const struct file_operations softnet_seq_fops = { |
1da177e4 LT |
4336 | .owner = THIS_MODULE, |
4337 | .open = softnet_seq_open, | |
4338 | .read = seq_read, | |
4339 | .llseek = seq_lseek, | |
4340 | .release = seq_release, | |
4341 | }; | |
4342 | ||
0e1256ff SH |
4343 | static void *ptype_get_idx(loff_t pos) |
4344 | { | |
4345 | struct packet_type *pt = NULL; | |
4346 | loff_t i = 0; | |
4347 | int t; | |
4348 | ||
4349 | list_for_each_entry_rcu(pt, &ptype_all, list) { | |
4350 | if (i == pos) | |
4351 | return pt; | |
4352 | ++i; | |
4353 | } | |
4354 | ||
82d8a867 | 4355 | for (t = 0; t < PTYPE_HASH_SIZE; t++) { |
0e1256ff SH |
4356 | list_for_each_entry_rcu(pt, &ptype_base[t], list) { |
4357 | if (i == pos) | |
4358 | return pt; | |
4359 | ++i; | |
4360 | } | |
4361 | } | |
4362 | return NULL; | |
4363 | } | |
4364 | ||
4365 | static void *ptype_seq_start(struct seq_file *seq, loff_t *pos) | |
72348a42 | 4366 | __acquires(RCU) |
0e1256ff SH |
4367 | { |
4368 | rcu_read_lock(); | |
4369 | return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN; | |
4370 | } | |
4371 | ||
4372 | static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
4373 | { | |
4374 | struct packet_type *pt; | |
4375 | struct list_head *nxt; | |
4376 | int hash; | |
4377 | ||
4378 | ++*pos; | |
4379 | if (v == SEQ_START_TOKEN) | |
4380 | return ptype_get_idx(0); | |
4381 | ||
4382 | pt = v; | |
4383 | nxt = pt->list.next; | |
4384 | if (pt->type == htons(ETH_P_ALL)) { | |
4385 | if (nxt != &ptype_all) | |
4386 | goto found; | |
4387 | hash = 0; | |
4388 | nxt = ptype_base[0].next; | |
4389 | } else | |
82d8a867 | 4390 | hash = ntohs(pt->type) & PTYPE_HASH_MASK; |
0e1256ff SH |
4391 | |
4392 | while (nxt == &ptype_base[hash]) { | |
82d8a867 | 4393 | if (++hash >= PTYPE_HASH_SIZE) |
0e1256ff SH |
4394 | return NULL; |
4395 | nxt = ptype_base[hash].next; | |
4396 | } | |
4397 | found: | |
4398 | return list_entry(nxt, struct packet_type, list); | |
4399 | } | |
4400 | ||
4401 | static void ptype_seq_stop(struct seq_file *seq, void *v) | |
72348a42 | 4402 | __releases(RCU) |
0e1256ff SH |
4403 | { |
4404 | rcu_read_unlock(); | |
4405 | } | |
4406 | ||
0e1256ff SH |
4407 | static int ptype_seq_show(struct seq_file *seq, void *v) |
4408 | { | |
4409 | struct packet_type *pt = v; | |
4410 | ||
4411 | if (v == SEQ_START_TOKEN) | |
4412 | seq_puts(seq, "Type Device Function\n"); | |
c346dca1 | 4413 | else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) { |
0e1256ff SH |
4414 | if (pt->type == htons(ETH_P_ALL)) |
4415 | seq_puts(seq, "ALL "); | |
4416 | else | |
4417 | seq_printf(seq, "%04x", ntohs(pt->type)); | |
4418 | ||
908cd2da AD |
4419 | seq_printf(seq, " %-8s %pF\n", |
4420 | pt->dev ? pt->dev->name : "", pt->func); | |
0e1256ff SH |
4421 | } |
4422 | ||
4423 | return 0; | |
4424 | } | |
4425 | ||
4426 | static const struct seq_operations ptype_seq_ops = { | |
4427 | .start = ptype_seq_start, | |
4428 | .next = ptype_seq_next, | |
4429 | .stop = ptype_seq_stop, | |
4430 | .show = ptype_seq_show, | |
4431 | }; | |
4432 | ||
4433 | static int ptype_seq_open(struct inode *inode, struct file *file) | |
4434 | { | |
2feb27db PE |
4435 | return seq_open_net(inode, file, &ptype_seq_ops, |
4436 | sizeof(struct seq_net_private)); | |
0e1256ff SH |
4437 | } |
4438 | ||
4439 | static const struct file_operations ptype_seq_fops = { | |
4440 | .owner = THIS_MODULE, | |
4441 | .open = ptype_seq_open, | |
4442 | .read = seq_read, | |
4443 | .llseek = seq_lseek, | |
2feb27db | 4444 | .release = seq_release_net, |
0e1256ff SH |
4445 | }; |
4446 | ||
4447 | ||
4665079c | 4448 | static int __net_init dev_proc_net_init(struct net *net) |
1da177e4 LT |
4449 | { |
4450 | int rc = -ENOMEM; | |
4451 | ||
881d966b | 4452 | if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops)) |
1da177e4 | 4453 | goto out; |
881d966b | 4454 | if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops)) |
1da177e4 | 4455 | goto out_dev; |
881d966b | 4456 | if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops)) |
457c4cbc | 4457 | goto out_softnet; |
0e1256ff | 4458 | |
881d966b | 4459 | if (wext_proc_init(net)) |
457c4cbc | 4460 | goto out_ptype; |
1da177e4 LT |
4461 | rc = 0; |
4462 | out: | |
4463 | return rc; | |
457c4cbc | 4464 | out_ptype: |
881d966b | 4465 | proc_net_remove(net, "ptype"); |
1da177e4 | 4466 | out_softnet: |
881d966b | 4467 | proc_net_remove(net, "softnet_stat"); |
1da177e4 | 4468 | out_dev: |
881d966b | 4469 | proc_net_remove(net, "dev"); |
1da177e4 LT |
4470 | goto out; |
4471 | } | |
881d966b | 4472 | |
4665079c | 4473 | static void __net_exit dev_proc_net_exit(struct net *net) |
881d966b EB |
4474 | { |
4475 | wext_proc_exit(net); | |
4476 | ||
4477 | proc_net_remove(net, "ptype"); | |
4478 | proc_net_remove(net, "softnet_stat"); | |
4479 | proc_net_remove(net, "dev"); | |
4480 | } | |
4481 | ||
022cbae6 | 4482 | static struct pernet_operations __net_initdata dev_proc_ops = { |
881d966b EB |
4483 | .init = dev_proc_net_init, |
4484 | .exit = dev_proc_net_exit, | |
4485 | }; | |
4486 | ||
4487 | static int __init dev_proc_init(void) | |
4488 | { | |
4489 | return register_pernet_subsys(&dev_proc_ops); | |
4490 | } | |
1da177e4 LT |
4491 | #else |
4492 | #define dev_proc_init() 0 | |
4493 | #endif /* CONFIG_PROC_FS */ | |
4494 | ||
4495 | ||
4496 | /** | |
1765a575 | 4497 | * netdev_set_master - set up master pointer |
1da177e4 LT |
4498 | * @slave: slave device |
4499 | * @master: new master device | |
4500 | * | |
4501 | * Changes the master device of the slave. Pass %NULL to break the | |
4502 | * bonding. The caller must hold the RTNL semaphore. On a failure | |
4503 | * a negative errno code is returned. On success the reference counts | |
1765a575 | 4504 | * are adjusted and the function returns zero. |
1da177e4 LT |
4505 | */ |
4506 | int netdev_set_master(struct net_device *slave, struct net_device *master) | |
4507 | { | |
4508 | struct net_device *old = slave->master; | |
4509 | ||
4510 | ASSERT_RTNL(); | |
4511 | ||
4512 | if (master) { | |
4513 | if (old) | |
4514 | return -EBUSY; | |
4515 | dev_hold(master); | |
4516 | } | |
4517 | ||
4518 | slave->master = master; | |
4ec93edb | 4519 | |
6df427fe | 4520 | if (old) |
1da177e4 | 4521 | dev_put(old); |
1765a575 JP |
4522 | return 0; |
4523 | } | |
4524 | EXPORT_SYMBOL(netdev_set_master); | |
4525 | ||
4526 | /** | |
4527 | * netdev_set_bond_master - set up bonding master/slave pair | |
4528 | * @slave: slave device | |
4529 | * @master: new master device | |
4530 | * | |
4531 | * Changes the master device of the slave. Pass %NULL to break the | |
4532 | * bonding. The caller must hold the RTNL semaphore. On a failure | |
4533 | * a negative errno code is returned. On success %RTM_NEWLINK is sent | |
4534 | * to the routing socket and the function returns zero. | |
4535 | */ | |
4536 | int netdev_set_bond_master(struct net_device *slave, struct net_device *master) | |
4537 | { | |
4538 | int err; | |
4539 | ||
4540 | ASSERT_RTNL(); | |
4541 | ||
4542 | err = netdev_set_master(slave, master); | |
4543 | if (err) | |
4544 | return err; | |
1da177e4 LT |
4545 | if (master) |
4546 | slave->flags |= IFF_SLAVE; | |
4547 | else | |
4548 | slave->flags &= ~IFF_SLAVE; | |
4549 | ||
4550 | rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE); | |
4551 | return 0; | |
4552 | } | |
1765a575 | 4553 | EXPORT_SYMBOL(netdev_set_bond_master); |
1da177e4 | 4554 | |
b6c40d68 PM |
4555 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
4556 | { | |
d314774c SH |
4557 | const struct net_device_ops *ops = dev->netdev_ops; |
4558 | ||
4559 | if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags) | |
4560 | ops->ndo_change_rx_flags(dev, flags); | |
b6c40d68 PM |
4561 | } |
4562 | ||
dad9b335 | 4563 | static int __dev_set_promiscuity(struct net_device *dev, int inc) |
1da177e4 | 4564 | { |
b536db93 | 4565 | unsigned int old_flags = dev->flags; |
d04a48b0 EB |
4566 | kuid_t uid; |
4567 | kgid_t gid; | |
1da177e4 | 4568 | |
24023451 PM |
4569 | ASSERT_RTNL(); |
4570 | ||
dad9b335 WC |
4571 | dev->flags |= IFF_PROMISC; |
4572 | dev->promiscuity += inc; | |
4573 | if (dev->promiscuity == 0) { | |
4574 | /* | |
4575 | * Avoid overflow. | |
4576 | * If inc causes overflow, untouch promisc and return error. | |
4577 | */ | |
4578 | if (inc < 0) | |
4579 | dev->flags &= ~IFF_PROMISC; | |
4580 | else { | |
4581 | dev->promiscuity -= inc; | |
7b6cd1ce JP |
4582 | pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", |
4583 | dev->name); | |
dad9b335 WC |
4584 | return -EOVERFLOW; |
4585 | } | |
4586 | } | |
52609c0b | 4587 | if (dev->flags != old_flags) { |
7b6cd1ce JP |
4588 | pr_info("device %s %s promiscuous mode\n", |
4589 | dev->name, | |
4590 | dev->flags & IFF_PROMISC ? "entered" : "left"); | |
8192b0c4 DH |
4591 | if (audit_enabled) { |
4592 | current_uid_gid(&uid, &gid); | |
7759db82 KHK |
4593 | audit_log(current->audit_context, GFP_ATOMIC, |
4594 | AUDIT_ANOM_PROMISCUOUS, | |
4595 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
4596 | dev->name, (dev->flags & IFF_PROMISC), | |
4597 | (old_flags & IFF_PROMISC), | |
e1760bd5 | 4598 | from_kuid(&init_user_ns, audit_get_loginuid(current)), |
d04a48b0 EB |
4599 | from_kuid(&init_user_ns, uid), |
4600 | from_kgid(&init_user_ns, gid), | |
7759db82 | 4601 | audit_get_sessionid(current)); |
8192b0c4 | 4602 | } |
24023451 | 4603 | |
b6c40d68 | 4604 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 4605 | } |
dad9b335 | 4606 | return 0; |
1da177e4 LT |
4607 | } |
4608 | ||
4417da66 PM |
4609 | /** |
4610 | * dev_set_promiscuity - update promiscuity count on a device | |
4611 | * @dev: device | |
4612 | * @inc: modifier | |
4613 | * | |
4614 | * Add or remove promiscuity from a device. While the count in the device | |
4615 | * remains above zero the interface remains promiscuous. Once it hits zero | |
4616 | * the device reverts back to normal filtering operation. A negative inc | |
4617 | * value is used to drop promiscuity on the device. | |
dad9b335 | 4618 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 4619 | */ |
dad9b335 | 4620 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 | 4621 | { |
b536db93 | 4622 | unsigned int old_flags = dev->flags; |
dad9b335 | 4623 | int err; |
4417da66 | 4624 | |
dad9b335 | 4625 | err = __dev_set_promiscuity(dev, inc); |
4b5a698e | 4626 | if (err < 0) |
dad9b335 | 4627 | return err; |
4417da66 PM |
4628 | if (dev->flags != old_flags) |
4629 | dev_set_rx_mode(dev); | |
dad9b335 | 4630 | return err; |
4417da66 | 4631 | } |
d1b19dff | 4632 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 4633 | |
1da177e4 LT |
4634 | /** |
4635 | * dev_set_allmulti - update allmulti count on a device | |
4636 | * @dev: device | |
4637 | * @inc: modifier | |
4638 | * | |
4639 | * Add or remove reception of all multicast frames to a device. While the | |
4640 | * count in the device remains above zero the interface remains listening | |
4641 | * to all interfaces. Once it hits zero the device reverts back to normal | |
4642 | * filtering operation. A negative @inc value is used to drop the counter | |
4643 | * when releasing a resource needing all multicasts. | |
dad9b335 | 4644 | * Return 0 if successful or a negative errno code on error. |
1da177e4 LT |
4645 | */ |
4646 | ||
dad9b335 | 4647 | int dev_set_allmulti(struct net_device *dev, int inc) |
1da177e4 | 4648 | { |
b536db93 | 4649 | unsigned int old_flags = dev->flags; |
1da177e4 | 4650 | |
24023451 PM |
4651 | ASSERT_RTNL(); |
4652 | ||
1da177e4 | 4653 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
4654 | dev->allmulti += inc; |
4655 | if (dev->allmulti == 0) { | |
4656 | /* | |
4657 | * Avoid overflow. | |
4658 | * If inc causes overflow, untouch allmulti and return error. | |
4659 | */ | |
4660 | if (inc < 0) | |
4661 | dev->flags &= ~IFF_ALLMULTI; | |
4662 | else { | |
4663 | dev->allmulti -= inc; | |
7b6cd1ce JP |
4664 | pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", |
4665 | dev->name); | |
dad9b335 WC |
4666 | return -EOVERFLOW; |
4667 | } | |
4668 | } | |
24023451 | 4669 | if (dev->flags ^ old_flags) { |
b6c40d68 | 4670 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 4671 | dev_set_rx_mode(dev); |
24023451 | 4672 | } |
dad9b335 | 4673 | return 0; |
4417da66 | 4674 | } |
d1b19dff | 4675 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
4676 | |
4677 | /* | |
4678 | * Upload unicast and multicast address lists to device and | |
4679 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 4680 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
4681 | * are present. |
4682 | */ | |
4683 | void __dev_set_rx_mode(struct net_device *dev) | |
4684 | { | |
d314774c SH |
4685 | const struct net_device_ops *ops = dev->netdev_ops; |
4686 | ||
4417da66 PM |
4687 | /* dev_open will call this function so the list will stay sane. */ |
4688 | if (!(dev->flags&IFF_UP)) | |
4689 | return; | |
4690 | ||
4691 | if (!netif_device_present(dev)) | |
40b77c94 | 4692 | return; |
4417da66 | 4693 | |
01789349 | 4694 | if (!(dev->priv_flags & IFF_UNICAST_FLT)) { |
4417da66 PM |
4695 | /* Unicast addresses changes may only happen under the rtnl, |
4696 | * therefore calling __dev_set_promiscuity here is safe. | |
4697 | */ | |
32e7bfc4 | 4698 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
4417da66 | 4699 | __dev_set_promiscuity(dev, 1); |
2d348d1f | 4700 | dev->uc_promisc = true; |
32e7bfc4 | 4701 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
4417da66 | 4702 | __dev_set_promiscuity(dev, -1); |
2d348d1f | 4703 | dev->uc_promisc = false; |
4417da66 | 4704 | } |
4417da66 | 4705 | } |
01789349 JP |
4706 | |
4707 | if (ops->ndo_set_rx_mode) | |
4708 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
4709 | } |
4710 | ||
4711 | void dev_set_rx_mode(struct net_device *dev) | |
4712 | { | |
b9e40857 | 4713 | netif_addr_lock_bh(dev); |
4417da66 | 4714 | __dev_set_rx_mode(dev); |
b9e40857 | 4715 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
4716 | } |
4717 | ||
f0db275a SH |
4718 | /** |
4719 | * dev_get_flags - get flags reported to userspace | |
4720 | * @dev: device | |
4721 | * | |
4722 | * Get the combination of flag bits exported through APIs to userspace. | |
4723 | */ | |
95c96174 | 4724 | unsigned int dev_get_flags(const struct net_device *dev) |
1da177e4 | 4725 | { |
95c96174 | 4726 | unsigned int flags; |
1da177e4 LT |
4727 | |
4728 | flags = (dev->flags & ~(IFF_PROMISC | | |
4729 | IFF_ALLMULTI | | |
b00055aa SR |
4730 | IFF_RUNNING | |
4731 | IFF_LOWER_UP | | |
4732 | IFF_DORMANT)) | | |
1da177e4 LT |
4733 | (dev->gflags & (IFF_PROMISC | |
4734 | IFF_ALLMULTI)); | |
4735 | ||
b00055aa SR |
4736 | if (netif_running(dev)) { |
4737 | if (netif_oper_up(dev)) | |
4738 | flags |= IFF_RUNNING; | |
4739 | if (netif_carrier_ok(dev)) | |
4740 | flags |= IFF_LOWER_UP; | |
4741 | if (netif_dormant(dev)) | |
4742 | flags |= IFF_DORMANT; | |
4743 | } | |
1da177e4 LT |
4744 | |
4745 | return flags; | |
4746 | } | |
d1b19dff | 4747 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 4748 | |
bd380811 | 4749 | int __dev_change_flags(struct net_device *dev, unsigned int flags) |
1da177e4 | 4750 | { |
b536db93 | 4751 | unsigned int old_flags = dev->flags; |
bd380811 | 4752 | int ret; |
1da177e4 | 4753 | |
24023451 PM |
4754 | ASSERT_RTNL(); |
4755 | ||
1da177e4 LT |
4756 | /* |
4757 | * Set the flags on our device. | |
4758 | */ | |
4759 | ||
4760 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
4761 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
4762 | IFF_AUTOMEDIA)) | | |
4763 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
4764 | IFF_ALLMULTI)); | |
4765 | ||
4766 | /* | |
4767 | * Load in the correct multicast list now the flags have changed. | |
4768 | */ | |
4769 | ||
b6c40d68 PM |
4770 | if ((old_flags ^ flags) & IFF_MULTICAST) |
4771 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 4772 | |
4417da66 | 4773 | dev_set_rx_mode(dev); |
1da177e4 LT |
4774 | |
4775 | /* | |
4776 | * Have we downed the interface. We handle IFF_UP ourselves | |
4777 | * according to user attempts to set it, rather than blindly | |
4778 | * setting it. | |
4779 | */ | |
4780 | ||
4781 | ret = 0; | |
4782 | if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */ | |
bd380811 | 4783 | ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev); |
1da177e4 LT |
4784 | |
4785 | if (!ret) | |
4417da66 | 4786 | dev_set_rx_mode(dev); |
1da177e4 LT |
4787 | } |
4788 | ||
1da177e4 | 4789 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff ED |
4790 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
4791 | ||
1da177e4 LT |
4792 | dev->gflags ^= IFF_PROMISC; |
4793 | dev_set_promiscuity(dev, inc); | |
4794 | } | |
4795 | ||
4796 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
4797 | is important. Some (broken) drivers set IFF_PROMISC, when | |
4798 | IFF_ALLMULTI is requested not asking us and not reporting. | |
4799 | */ | |
4800 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
4801 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
4802 | ||
1da177e4 LT |
4803 | dev->gflags ^= IFF_ALLMULTI; |
4804 | dev_set_allmulti(dev, inc); | |
4805 | } | |
4806 | ||
bd380811 PM |
4807 | return ret; |
4808 | } | |
4809 | ||
4810 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags) | |
4811 | { | |
4812 | unsigned int changes = dev->flags ^ old_flags; | |
4813 | ||
4814 | if (changes & IFF_UP) { | |
4815 | if (dev->flags & IFF_UP) | |
4816 | call_netdevice_notifiers(NETDEV_UP, dev); | |
4817 | else | |
4818 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
4819 | } | |
4820 | ||
4821 | if (dev->flags & IFF_UP && | |
4822 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) | |
4823 | call_netdevice_notifiers(NETDEV_CHANGE, dev); | |
4824 | } | |
4825 | ||
4826 | /** | |
4827 | * dev_change_flags - change device settings | |
4828 | * @dev: device | |
4829 | * @flags: device state flags | |
4830 | * | |
4831 | * Change settings on device based state flags. The flags are | |
4832 | * in the userspace exported format. | |
4833 | */ | |
b536db93 | 4834 | int dev_change_flags(struct net_device *dev, unsigned int flags) |
bd380811 | 4835 | { |
b536db93 ED |
4836 | int ret; |
4837 | unsigned int changes, old_flags = dev->flags; | |
bd380811 PM |
4838 | |
4839 | ret = __dev_change_flags(dev, flags); | |
4840 | if (ret < 0) | |
4841 | return ret; | |
4842 | ||
4843 | changes = old_flags ^ dev->flags; | |
7c355f53 TG |
4844 | if (changes) |
4845 | rtmsg_ifinfo(RTM_NEWLINK, dev, changes); | |
1da177e4 | 4846 | |
bd380811 | 4847 | __dev_notify_flags(dev, old_flags); |
1da177e4 LT |
4848 | return ret; |
4849 | } | |
d1b19dff | 4850 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 4851 | |
f0db275a SH |
4852 | /** |
4853 | * dev_set_mtu - Change maximum transfer unit | |
4854 | * @dev: device | |
4855 | * @new_mtu: new transfer unit | |
4856 | * | |
4857 | * Change the maximum transfer size of the network device. | |
4858 | */ | |
1da177e4 LT |
4859 | int dev_set_mtu(struct net_device *dev, int new_mtu) |
4860 | { | |
d314774c | 4861 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
4862 | int err; |
4863 | ||
4864 | if (new_mtu == dev->mtu) | |
4865 | return 0; | |
4866 | ||
4867 | /* MTU must be positive. */ | |
4868 | if (new_mtu < 0) | |
4869 | return -EINVAL; | |
4870 | ||
4871 | if (!netif_device_present(dev)) | |
4872 | return -ENODEV; | |
4873 | ||
4874 | err = 0; | |
d314774c SH |
4875 | if (ops->ndo_change_mtu) |
4876 | err = ops->ndo_change_mtu(dev, new_mtu); | |
1da177e4 LT |
4877 | else |
4878 | dev->mtu = new_mtu; | |
d314774c | 4879 | |
1da177e4 | 4880 | if (!err && dev->flags & IFF_UP) |
056925ab | 4881 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); |
1da177e4 LT |
4882 | return err; |
4883 | } | |
d1b19dff | 4884 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 4885 | |
cbda10fa VD |
4886 | /** |
4887 | * dev_set_group - Change group this device belongs to | |
4888 | * @dev: device | |
4889 | * @new_group: group this device should belong to | |
4890 | */ | |
4891 | void dev_set_group(struct net_device *dev, int new_group) | |
4892 | { | |
4893 | dev->group = new_group; | |
4894 | } | |
4895 | EXPORT_SYMBOL(dev_set_group); | |
4896 | ||
f0db275a SH |
4897 | /** |
4898 | * dev_set_mac_address - Change Media Access Control Address | |
4899 | * @dev: device | |
4900 | * @sa: new address | |
4901 | * | |
4902 | * Change the hardware (MAC) address of the device | |
4903 | */ | |
1da177e4 LT |
4904 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) |
4905 | { | |
d314774c | 4906 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
4907 | int err; |
4908 | ||
d314774c | 4909 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
4910 | return -EOPNOTSUPP; |
4911 | if (sa->sa_family != dev->type) | |
4912 | return -EINVAL; | |
4913 | if (!netif_device_present(dev)) | |
4914 | return -ENODEV; | |
d314774c | 4915 | err = ops->ndo_set_mac_address(dev, sa); |
1da177e4 | 4916 | if (!err) |
056925ab | 4917 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
7bf23575 | 4918 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 LT |
4919 | return err; |
4920 | } | |
d1b19dff | 4921 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 LT |
4922 | |
4923 | /* | |
3710becf | 4924 | * Perform the SIOCxIFxxx calls, inside rcu_read_lock() |
1da177e4 | 4925 | */ |
14e3e079 | 4926 | static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd) |
1da177e4 LT |
4927 | { |
4928 | int err; | |
3710becf | 4929 | struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name); |
1da177e4 LT |
4930 | |
4931 | if (!dev) | |
4932 | return -ENODEV; | |
4933 | ||
4934 | switch (cmd) { | |
d1b19dff ED |
4935 | case SIOCGIFFLAGS: /* Get interface flags */ |
4936 | ifr->ifr_flags = (short) dev_get_flags(dev); | |
4937 | return 0; | |
1da177e4 | 4938 | |
d1b19dff ED |
4939 | case SIOCGIFMETRIC: /* Get the metric on the interface |
4940 | (currently unused) */ | |
4941 | ifr->ifr_metric = 0; | |
4942 | return 0; | |
1da177e4 | 4943 | |
d1b19dff ED |
4944 | case SIOCGIFMTU: /* Get the MTU of a device */ |
4945 | ifr->ifr_mtu = dev->mtu; | |
4946 | return 0; | |
1da177e4 | 4947 | |
d1b19dff ED |
4948 | case SIOCGIFHWADDR: |
4949 | if (!dev->addr_len) | |
4950 | memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data); | |
4951 | else | |
4952 | memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr, | |
4953 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
4954 | ifr->ifr_hwaddr.sa_family = dev->type; | |
4955 | return 0; | |
1da177e4 | 4956 | |
d1b19dff ED |
4957 | case SIOCGIFSLAVE: |
4958 | err = -EINVAL; | |
4959 | break; | |
14e3e079 | 4960 | |
d1b19dff ED |
4961 | case SIOCGIFMAP: |
4962 | ifr->ifr_map.mem_start = dev->mem_start; | |
4963 | ifr->ifr_map.mem_end = dev->mem_end; | |
4964 | ifr->ifr_map.base_addr = dev->base_addr; | |
4965 | ifr->ifr_map.irq = dev->irq; | |
4966 | ifr->ifr_map.dma = dev->dma; | |
4967 | ifr->ifr_map.port = dev->if_port; | |
4968 | return 0; | |
14e3e079 | 4969 | |
d1b19dff ED |
4970 | case SIOCGIFINDEX: |
4971 | ifr->ifr_ifindex = dev->ifindex; | |
4972 | return 0; | |
14e3e079 | 4973 | |
d1b19dff ED |
4974 | case SIOCGIFTXQLEN: |
4975 | ifr->ifr_qlen = dev->tx_queue_len; | |
4976 | return 0; | |
14e3e079 | 4977 | |
d1b19dff ED |
4978 | default: |
4979 | /* dev_ioctl() should ensure this case | |
4980 | * is never reached | |
4981 | */ | |
4982 | WARN_ON(1); | |
41c31f31 | 4983 | err = -ENOTTY; |
d1b19dff | 4984 | break; |
14e3e079 JG |
4985 | |
4986 | } | |
4987 | return err; | |
4988 | } | |
4989 | ||
4990 | /* | |
4991 | * Perform the SIOCxIFxxx calls, inside rtnl_lock() | |
4992 | */ | |
4993 | static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd) | |
4994 | { | |
4995 | int err; | |
4996 | struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name); | |
5f2f6da7 | 4997 | const struct net_device_ops *ops; |
14e3e079 JG |
4998 | |
4999 | if (!dev) | |
5000 | return -ENODEV; | |
5001 | ||
5f2f6da7 JP |
5002 | ops = dev->netdev_ops; |
5003 | ||
14e3e079 | 5004 | switch (cmd) { |
d1b19dff ED |
5005 | case SIOCSIFFLAGS: /* Set interface flags */ |
5006 | return dev_change_flags(dev, ifr->ifr_flags); | |
14e3e079 | 5007 | |
d1b19dff ED |
5008 | case SIOCSIFMETRIC: /* Set the metric on the interface |
5009 | (currently unused) */ | |
5010 | return -EOPNOTSUPP; | |
14e3e079 | 5011 | |
d1b19dff ED |
5012 | case SIOCSIFMTU: /* Set the MTU of a device */ |
5013 | return dev_set_mtu(dev, ifr->ifr_mtu); | |
1da177e4 | 5014 | |
d1b19dff ED |
5015 | case SIOCSIFHWADDR: |
5016 | return dev_set_mac_address(dev, &ifr->ifr_hwaddr); | |
1da177e4 | 5017 | |
d1b19dff ED |
5018 | case SIOCSIFHWBROADCAST: |
5019 | if (ifr->ifr_hwaddr.sa_family != dev->type) | |
5020 | return -EINVAL; | |
5021 | memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data, | |
5022 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
5023 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); | |
5024 | return 0; | |
1da177e4 | 5025 | |
d1b19dff ED |
5026 | case SIOCSIFMAP: |
5027 | if (ops->ndo_set_config) { | |
1da177e4 LT |
5028 | if (!netif_device_present(dev)) |
5029 | return -ENODEV; | |
d1b19dff ED |
5030 | return ops->ndo_set_config(dev, &ifr->ifr_map); |
5031 | } | |
5032 | return -EOPNOTSUPP; | |
1da177e4 | 5033 | |
d1b19dff | 5034 | case SIOCADDMULTI: |
b81693d9 | 5035 | if (!ops->ndo_set_rx_mode || |
d1b19dff ED |
5036 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) |
5037 | return -EINVAL; | |
5038 | if (!netif_device_present(dev)) | |
5039 | return -ENODEV; | |
22bedad3 | 5040 | return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data); |
d1b19dff ED |
5041 | |
5042 | case SIOCDELMULTI: | |
b81693d9 | 5043 | if (!ops->ndo_set_rx_mode || |
d1b19dff ED |
5044 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) |
5045 | return -EINVAL; | |
5046 | if (!netif_device_present(dev)) | |
5047 | return -ENODEV; | |
22bedad3 | 5048 | return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data); |
1da177e4 | 5049 | |
d1b19dff ED |
5050 | case SIOCSIFTXQLEN: |
5051 | if (ifr->ifr_qlen < 0) | |
5052 | return -EINVAL; | |
5053 | dev->tx_queue_len = ifr->ifr_qlen; | |
5054 | return 0; | |
1da177e4 | 5055 | |
d1b19dff ED |
5056 | case SIOCSIFNAME: |
5057 | ifr->ifr_newname[IFNAMSIZ-1] = '\0'; | |
5058 | return dev_change_name(dev, ifr->ifr_newname); | |
1da177e4 | 5059 | |
4dc360c5 RC |
5060 | case SIOCSHWTSTAMP: |
5061 | err = net_hwtstamp_validate(ifr); | |
5062 | if (err) | |
5063 | return err; | |
5064 | /* fall through */ | |
5065 | ||
d1b19dff ED |
5066 | /* |
5067 | * Unknown or private ioctl | |
5068 | */ | |
5069 | default: | |
5070 | if ((cmd >= SIOCDEVPRIVATE && | |
5071 | cmd <= SIOCDEVPRIVATE + 15) || | |
5072 | cmd == SIOCBONDENSLAVE || | |
5073 | cmd == SIOCBONDRELEASE || | |
5074 | cmd == SIOCBONDSETHWADDR || | |
5075 | cmd == SIOCBONDSLAVEINFOQUERY || | |
5076 | cmd == SIOCBONDINFOQUERY || | |
5077 | cmd == SIOCBONDCHANGEACTIVE || | |
5078 | cmd == SIOCGMIIPHY || | |
5079 | cmd == SIOCGMIIREG || | |
5080 | cmd == SIOCSMIIREG || | |
5081 | cmd == SIOCBRADDIF || | |
5082 | cmd == SIOCBRDELIF || | |
5083 | cmd == SIOCSHWTSTAMP || | |
5084 | cmd == SIOCWANDEV) { | |
5085 | err = -EOPNOTSUPP; | |
5086 | if (ops->ndo_do_ioctl) { | |
5087 | if (netif_device_present(dev)) | |
5088 | err = ops->ndo_do_ioctl(dev, ifr, cmd); | |
5089 | else | |
5090 | err = -ENODEV; | |
5091 | } | |
5092 | } else | |
5093 | err = -EINVAL; | |
1da177e4 LT |
5094 | |
5095 | } | |
5096 | return err; | |
5097 | } | |
5098 | ||
5099 | /* | |
5100 | * This function handles all "interface"-type I/O control requests. The actual | |
5101 | * 'doing' part of this is dev_ifsioc above. | |
5102 | */ | |
5103 | ||
5104 | /** | |
5105 | * dev_ioctl - network device ioctl | |
c4ea43c5 | 5106 | * @net: the applicable net namespace |
1da177e4 LT |
5107 | * @cmd: command to issue |
5108 | * @arg: pointer to a struct ifreq in user space | |
5109 | * | |
5110 | * Issue ioctl functions to devices. This is normally called by the | |
5111 | * user space syscall interfaces but can sometimes be useful for | |
5112 | * other purposes. The return value is the return from the syscall if | |
5113 | * positive or a negative errno code on error. | |
5114 | */ | |
5115 | ||
881d966b | 5116 | int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg) |
1da177e4 LT |
5117 | { |
5118 | struct ifreq ifr; | |
5119 | int ret; | |
5120 | char *colon; | |
5121 | ||
5122 | /* One special case: SIOCGIFCONF takes ifconf argument | |
5123 | and requires shared lock, because it sleeps writing | |
5124 | to user space. | |
5125 | */ | |
5126 | ||
5127 | if (cmd == SIOCGIFCONF) { | |
6756ae4b | 5128 | rtnl_lock(); |
881d966b | 5129 | ret = dev_ifconf(net, (char __user *) arg); |
6756ae4b | 5130 | rtnl_unlock(); |
1da177e4 LT |
5131 | return ret; |
5132 | } | |
5133 | if (cmd == SIOCGIFNAME) | |
881d966b | 5134 | return dev_ifname(net, (struct ifreq __user *)arg); |
1da177e4 LT |
5135 | |
5136 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
5137 | return -EFAULT; | |
5138 | ||
5139 | ifr.ifr_name[IFNAMSIZ-1] = 0; | |
5140 | ||
5141 | colon = strchr(ifr.ifr_name, ':'); | |
5142 | if (colon) | |
5143 | *colon = 0; | |
5144 | ||
5145 | /* | |
5146 | * See which interface the caller is talking about. | |
5147 | */ | |
5148 | ||
5149 | switch (cmd) { | |
d1b19dff ED |
5150 | /* |
5151 | * These ioctl calls: | |
5152 | * - can be done by all. | |
5153 | * - atomic and do not require locking. | |
5154 | * - return a value | |
5155 | */ | |
5156 | case SIOCGIFFLAGS: | |
5157 | case SIOCGIFMETRIC: | |
5158 | case SIOCGIFMTU: | |
5159 | case SIOCGIFHWADDR: | |
5160 | case SIOCGIFSLAVE: | |
5161 | case SIOCGIFMAP: | |
5162 | case SIOCGIFINDEX: | |
5163 | case SIOCGIFTXQLEN: | |
5164 | dev_load(net, ifr.ifr_name); | |
3710becf | 5165 | rcu_read_lock(); |
d1b19dff | 5166 | ret = dev_ifsioc_locked(net, &ifr, cmd); |
3710becf | 5167 | rcu_read_unlock(); |
d1b19dff ED |
5168 | if (!ret) { |
5169 | if (colon) | |
5170 | *colon = ':'; | |
5171 | if (copy_to_user(arg, &ifr, | |
5172 | sizeof(struct ifreq))) | |
5173 | ret = -EFAULT; | |
5174 | } | |
5175 | return ret; | |
1da177e4 | 5176 | |
d1b19dff ED |
5177 | case SIOCETHTOOL: |
5178 | dev_load(net, ifr.ifr_name); | |
5179 | rtnl_lock(); | |
5180 | ret = dev_ethtool(net, &ifr); | |
5181 | rtnl_unlock(); | |
5182 | if (!ret) { | |
5183 | if (colon) | |
5184 | *colon = ':'; | |
5185 | if (copy_to_user(arg, &ifr, | |
5186 | sizeof(struct ifreq))) | |
5187 | ret = -EFAULT; | |
5188 | } | |
5189 | return ret; | |
1da177e4 | 5190 | |
d1b19dff ED |
5191 | /* |
5192 | * These ioctl calls: | |
5193 | * - require superuser power. | |
5194 | * - require strict serialization. | |
5195 | * - return a value | |
5196 | */ | |
5197 | case SIOCGMIIPHY: | |
5198 | case SIOCGMIIREG: | |
5199 | case SIOCSIFNAME: | |
5200 | if (!capable(CAP_NET_ADMIN)) | |
5201 | return -EPERM; | |
5202 | dev_load(net, ifr.ifr_name); | |
5203 | rtnl_lock(); | |
5204 | ret = dev_ifsioc(net, &ifr, cmd); | |
5205 | rtnl_unlock(); | |
5206 | if (!ret) { | |
5207 | if (colon) | |
5208 | *colon = ':'; | |
5209 | if (copy_to_user(arg, &ifr, | |
5210 | sizeof(struct ifreq))) | |
5211 | ret = -EFAULT; | |
5212 | } | |
5213 | return ret; | |
1da177e4 | 5214 | |
d1b19dff ED |
5215 | /* |
5216 | * These ioctl calls: | |
5217 | * - require superuser power. | |
5218 | * - require strict serialization. | |
5219 | * - do not return a value | |
5220 | */ | |
5221 | case SIOCSIFFLAGS: | |
5222 | case SIOCSIFMETRIC: | |
5223 | case SIOCSIFMTU: | |
5224 | case SIOCSIFMAP: | |
5225 | case SIOCSIFHWADDR: | |
5226 | case SIOCSIFSLAVE: | |
5227 | case SIOCADDMULTI: | |
5228 | case SIOCDELMULTI: | |
5229 | case SIOCSIFHWBROADCAST: | |
5230 | case SIOCSIFTXQLEN: | |
5231 | case SIOCSMIIREG: | |
5232 | case SIOCBONDENSLAVE: | |
5233 | case SIOCBONDRELEASE: | |
5234 | case SIOCBONDSETHWADDR: | |
5235 | case SIOCBONDCHANGEACTIVE: | |
5236 | case SIOCBRADDIF: | |
5237 | case SIOCBRDELIF: | |
5238 | case SIOCSHWTSTAMP: | |
5239 | if (!capable(CAP_NET_ADMIN)) | |
5240 | return -EPERM; | |
5241 | /* fall through */ | |
5242 | case SIOCBONDSLAVEINFOQUERY: | |
5243 | case SIOCBONDINFOQUERY: | |
5244 | dev_load(net, ifr.ifr_name); | |
5245 | rtnl_lock(); | |
5246 | ret = dev_ifsioc(net, &ifr, cmd); | |
5247 | rtnl_unlock(); | |
5248 | return ret; | |
5249 | ||
5250 | case SIOCGIFMEM: | |
5251 | /* Get the per device memory space. We can add this but | |
5252 | * currently do not support it */ | |
5253 | case SIOCSIFMEM: | |
5254 | /* Set the per device memory buffer space. | |
5255 | * Not applicable in our case */ | |
5256 | case SIOCSIFLINK: | |
41c31f31 | 5257 | return -ENOTTY; |
d1b19dff ED |
5258 | |
5259 | /* | |
5260 | * Unknown or private ioctl. | |
5261 | */ | |
5262 | default: | |
5263 | if (cmd == SIOCWANDEV || | |
5264 | (cmd >= SIOCDEVPRIVATE && | |
5265 | cmd <= SIOCDEVPRIVATE + 15)) { | |
881d966b | 5266 | dev_load(net, ifr.ifr_name); |
1da177e4 | 5267 | rtnl_lock(); |
881d966b | 5268 | ret = dev_ifsioc(net, &ifr, cmd); |
1da177e4 | 5269 | rtnl_unlock(); |
d1b19dff ED |
5270 | if (!ret && copy_to_user(arg, &ifr, |
5271 | sizeof(struct ifreq))) | |
5272 | ret = -EFAULT; | |
1da177e4 | 5273 | return ret; |
d1b19dff ED |
5274 | } |
5275 | /* Take care of Wireless Extensions */ | |
5276 | if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) | |
5277 | return wext_handle_ioctl(net, &ifr, cmd, arg); | |
41c31f31 | 5278 | return -ENOTTY; |
1da177e4 LT |
5279 | } |
5280 | } | |
5281 | ||
5282 | ||
5283 | /** | |
5284 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 5285 | * @net: the applicable net namespace |
1da177e4 LT |
5286 | * |
5287 | * Returns a suitable unique value for a new device interface | |
5288 | * number. The caller must hold the rtnl semaphore or the | |
5289 | * dev_base_lock to be sure it remains unique. | |
5290 | */ | |
881d966b | 5291 | static int dev_new_index(struct net *net) |
1da177e4 | 5292 | { |
aa79e66e | 5293 | int ifindex = net->ifindex; |
1da177e4 LT |
5294 | for (;;) { |
5295 | if (++ifindex <= 0) | |
5296 | ifindex = 1; | |
881d966b | 5297 | if (!__dev_get_by_index(net, ifindex)) |
aa79e66e | 5298 | return net->ifindex = ifindex; |
1da177e4 LT |
5299 | } |
5300 | } | |
5301 | ||
1da177e4 | 5302 | /* Delayed registration/unregisteration */ |
3b5b34fd | 5303 | static LIST_HEAD(net_todo_list); |
1da177e4 | 5304 | |
6f05f629 | 5305 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 5306 | { |
1da177e4 | 5307 | list_add_tail(&dev->todo_list, &net_todo_list); |
1da177e4 LT |
5308 | } |
5309 | ||
9b5e383c | 5310 | static void rollback_registered_many(struct list_head *head) |
93ee31f1 | 5311 | { |
e93737b0 | 5312 | struct net_device *dev, *tmp; |
9b5e383c | 5313 | |
93ee31f1 DL |
5314 | BUG_ON(dev_boot_phase); |
5315 | ASSERT_RTNL(); | |
5316 | ||
e93737b0 | 5317 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
9b5e383c | 5318 | /* Some devices call without registering |
e93737b0 KK |
5319 | * for initialization unwind. Remove those |
5320 | * devices and proceed with the remaining. | |
9b5e383c ED |
5321 | */ |
5322 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
7b6cd1ce JP |
5323 | pr_debug("unregister_netdevice: device %s/%p never was registered\n", |
5324 | dev->name, dev); | |
93ee31f1 | 5325 | |
9b5e383c | 5326 | WARN_ON(1); |
e93737b0 KK |
5327 | list_del(&dev->unreg_list); |
5328 | continue; | |
9b5e383c | 5329 | } |
449f4544 | 5330 | dev->dismantle = true; |
9b5e383c | 5331 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
44345724 | 5332 | } |
93ee31f1 | 5333 | |
44345724 OP |
5334 | /* If device is running, close it first. */ |
5335 | dev_close_many(head); | |
93ee31f1 | 5336 | |
44345724 | 5337 | list_for_each_entry(dev, head, unreg_list) { |
9b5e383c ED |
5338 | /* And unlink it from device chain. */ |
5339 | unlist_netdevice(dev); | |
93ee31f1 | 5340 | |
9b5e383c ED |
5341 | dev->reg_state = NETREG_UNREGISTERING; |
5342 | } | |
93ee31f1 DL |
5343 | |
5344 | synchronize_net(); | |
5345 | ||
9b5e383c ED |
5346 | list_for_each_entry(dev, head, unreg_list) { |
5347 | /* Shutdown queueing discipline. */ | |
5348 | dev_shutdown(dev); | |
93ee31f1 DL |
5349 | |
5350 | ||
9b5e383c ED |
5351 | /* Notify protocols, that we are about to destroy |
5352 | this device. They should clean all the things. | |
5353 | */ | |
5354 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
93ee31f1 | 5355 | |
a2835763 PM |
5356 | if (!dev->rtnl_link_ops || |
5357 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
5358 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U); | |
5359 | ||
9b5e383c ED |
5360 | /* |
5361 | * Flush the unicast and multicast chains | |
5362 | */ | |
a748ee24 | 5363 | dev_uc_flush(dev); |
22bedad3 | 5364 | dev_mc_flush(dev); |
93ee31f1 | 5365 | |
9b5e383c ED |
5366 | if (dev->netdev_ops->ndo_uninit) |
5367 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 | 5368 | |
9b5e383c ED |
5369 | /* Notifier chain MUST detach us from master device. */ |
5370 | WARN_ON(dev->master); | |
93ee31f1 | 5371 | |
9b5e383c ED |
5372 | /* Remove entries from kobject tree */ |
5373 | netdev_unregister_kobject(dev); | |
5374 | } | |
93ee31f1 | 5375 | |
850a545b | 5376 | synchronize_net(); |
395264d5 | 5377 | |
a5ee1551 | 5378 | list_for_each_entry(dev, head, unreg_list) |
9b5e383c ED |
5379 | dev_put(dev); |
5380 | } | |
5381 | ||
5382 | static void rollback_registered(struct net_device *dev) | |
5383 | { | |
5384 | LIST_HEAD(single); | |
5385 | ||
5386 | list_add(&dev->unreg_list, &single); | |
5387 | rollback_registered_many(&single); | |
ceaaec98 | 5388 | list_del(&single); |
93ee31f1 DL |
5389 | } |
5390 | ||
c8f44aff MM |
5391 | static netdev_features_t netdev_fix_features(struct net_device *dev, |
5392 | netdev_features_t features) | |
b63365a2 | 5393 | { |
57422dc5 MM |
5394 | /* Fix illegal checksum combinations */ |
5395 | if ((features & NETIF_F_HW_CSUM) && | |
5396 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 5397 | netdev_warn(dev, "mixed HW and IP checksum settings.\n"); |
57422dc5 MM |
5398 | features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); |
5399 | } | |
5400 | ||
b63365a2 HX |
5401 | /* Fix illegal SG+CSUM combinations. */ |
5402 | if ((features & NETIF_F_SG) && | |
5403 | !(features & NETIF_F_ALL_CSUM)) { | |
6f404e44 MM |
5404 | netdev_dbg(dev, |
5405 | "Dropping NETIF_F_SG since no checksum feature.\n"); | |
b63365a2 HX |
5406 | features &= ~NETIF_F_SG; |
5407 | } | |
5408 | ||
5409 | /* TSO requires that SG is present as well. */ | |
ea2d3688 | 5410 | if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { |
6f404e44 | 5411 | netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); |
ea2d3688 | 5412 | features &= ~NETIF_F_ALL_TSO; |
b63365a2 HX |
5413 | } |
5414 | ||
31d8b9e0 BH |
5415 | /* TSO ECN requires that TSO is present as well. */ |
5416 | if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) | |
5417 | features &= ~NETIF_F_TSO_ECN; | |
5418 | ||
212b573f MM |
5419 | /* Software GSO depends on SG. */ |
5420 | if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { | |
6f404e44 | 5421 | netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); |
212b573f MM |
5422 | features &= ~NETIF_F_GSO; |
5423 | } | |
5424 | ||
acd1130e | 5425 | /* UFO needs SG and checksumming */ |
b63365a2 | 5426 | if (features & NETIF_F_UFO) { |
79032644 MM |
5427 | /* maybe split UFO into V4 and V6? */ |
5428 | if (!((features & NETIF_F_GEN_CSUM) || | |
5429 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM)) | |
5430 | == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 5431 | netdev_dbg(dev, |
acd1130e | 5432 | "Dropping NETIF_F_UFO since no checksum offload features.\n"); |
b63365a2 HX |
5433 | features &= ~NETIF_F_UFO; |
5434 | } | |
5435 | ||
5436 | if (!(features & NETIF_F_SG)) { | |
6f404e44 | 5437 | netdev_dbg(dev, |
acd1130e | 5438 | "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n"); |
b63365a2 HX |
5439 | features &= ~NETIF_F_UFO; |
5440 | } | |
5441 | } | |
5442 | ||
5443 | return features; | |
5444 | } | |
b63365a2 | 5445 | |
6cb6a27c | 5446 | int __netdev_update_features(struct net_device *dev) |
5455c699 | 5447 | { |
c8f44aff | 5448 | netdev_features_t features; |
5455c699 MM |
5449 | int err = 0; |
5450 | ||
87267485 MM |
5451 | ASSERT_RTNL(); |
5452 | ||
5455c699 MM |
5453 | features = netdev_get_wanted_features(dev); |
5454 | ||
5455 | if (dev->netdev_ops->ndo_fix_features) | |
5456 | features = dev->netdev_ops->ndo_fix_features(dev, features); | |
5457 | ||
5458 | /* driver might be less strict about feature dependencies */ | |
5459 | features = netdev_fix_features(dev, features); | |
5460 | ||
5461 | if (dev->features == features) | |
6cb6a27c | 5462 | return 0; |
5455c699 | 5463 | |
c8f44aff MM |
5464 | netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", |
5465 | &dev->features, &features); | |
5455c699 MM |
5466 | |
5467 | if (dev->netdev_ops->ndo_set_features) | |
5468 | err = dev->netdev_ops->ndo_set_features(dev, features); | |
5469 | ||
6cb6a27c | 5470 | if (unlikely(err < 0)) { |
5455c699 | 5471 | netdev_err(dev, |
c8f44aff MM |
5472 | "set_features() failed (%d); wanted %pNF, left %pNF\n", |
5473 | err, &features, &dev->features); | |
6cb6a27c MM |
5474 | return -1; |
5475 | } | |
5476 | ||
5477 | if (!err) | |
5478 | dev->features = features; | |
5479 | ||
5480 | return 1; | |
5481 | } | |
5482 | ||
afe12cc8 MM |
5483 | /** |
5484 | * netdev_update_features - recalculate device features | |
5485 | * @dev: the device to check | |
5486 | * | |
5487 | * Recalculate dev->features set and send notifications if it | |
5488 | * has changed. Should be called after driver or hardware dependent | |
5489 | * conditions might have changed that influence the features. | |
5490 | */ | |
6cb6a27c MM |
5491 | void netdev_update_features(struct net_device *dev) |
5492 | { | |
5493 | if (__netdev_update_features(dev)) | |
5494 | netdev_features_change(dev); | |
5455c699 MM |
5495 | } |
5496 | EXPORT_SYMBOL(netdev_update_features); | |
5497 | ||
afe12cc8 MM |
5498 | /** |
5499 | * netdev_change_features - recalculate device features | |
5500 | * @dev: the device to check | |
5501 | * | |
5502 | * Recalculate dev->features set and send notifications even | |
5503 | * if they have not changed. Should be called instead of | |
5504 | * netdev_update_features() if also dev->vlan_features might | |
5505 | * have changed to allow the changes to be propagated to stacked | |
5506 | * VLAN devices. | |
5507 | */ | |
5508 | void netdev_change_features(struct net_device *dev) | |
5509 | { | |
5510 | __netdev_update_features(dev); | |
5511 | netdev_features_change(dev); | |
5512 | } | |
5513 | EXPORT_SYMBOL(netdev_change_features); | |
5514 | ||
fc4a7489 PM |
5515 | /** |
5516 | * netif_stacked_transfer_operstate - transfer operstate | |
5517 | * @rootdev: the root or lower level device to transfer state from | |
5518 | * @dev: the device to transfer operstate to | |
5519 | * | |
5520 | * Transfer operational state from root to device. This is normally | |
5521 | * called when a stacking relationship exists between the root | |
5522 | * device and the device(a leaf device). | |
5523 | */ | |
5524 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
5525 | struct net_device *dev) | |
5526 | { | |
5527 | if (rootdev->operstate == IF_OPER_DORMANT) | |
5528 | netif_dormant_on(dev); | |
5529 | else | |
5530 | netif_dormant_off(dev); | |
5531 | ||
5532 | if (netif_carrier_ok(rootdev)) { | |
5533 | if (!netif_carrier_ok(dev)) | |
5534 | netif_carrier_on(dev); | |
5535 | } else { | |
5536 | if (netif_carrier_ok(dev)) | |
5537 | netif_carrier_off(dev); | |
5538 | } | |
5539 | } | |
5540 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
5541 | ||
bf264145 | 5542 | #ifdef CONFIG_RPS |
1b4bf461 ED |
5543 | static int netif_alloc_rx_queues(struct net_device *dev) |
5544 | { | |
1b4bf461 | 5545 | unsigned int i, count = dev->num_rx_queues; |
bd25fa7b | 5546 | struct netdev_rx_queue *rx; |
1b4bf461 | 5547 | |
bd25fa7b | 5548 | BUG_ON(count < 1); |
1b4bf461 | 5549 | |
bd25fa7b TH |
5550 | rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL); |
5551 | if (!rx) { | |
7b6cd1ce | 5552 | pr_err("netdev: Unable to allocate %u rx queues\n", count); |
bd25fa7b | 5553 | return -ENOMEM; |
1b4bf461 | 5554 | } |
bd25fa7b TH |
5555 | dev->_rx = rx; |
5556 | ||
bd25fa7b | 5557 | for (i = 0; i < count; i++) |
fe822240 | 5558 | rx[i].dev = dev; |
1b4bf461 ED |
5559 | return 0; |
5560 | } | |
bf264145 | 5561 | #endif |
1b4bf461 | 5562 | |
aa942104 CG |
5563 | static void netdev_init_one_queue(struct net_device *dev, |
5564 | struct netdev_queue *queue, void *_unused) | |
5565 | { | |
5566 | /* Initialize queue lock */ | |
5567 | spin_lock_init(&queue->_xmit_lock); | |
5568 | netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); | |
5569 | queue->xmit_lock_owner = -1; | |
b236da69 | 5570 | netdev_queue_numa_node_write(queue, NUMA_NO_NODE); |
aa942104 | 5571 | queue->dev = dev; |
114cf580 TH |
5572 | #ifdef CONFIG_BQL |
5573 | dql_init(&queue->dql, HZ); | |
5574 | #endif | |
aa942104 CG |
5575 | } |
5576 | ||
e6484930 TH |
5577 | static int netif_alloc_netdev_queues(struct net_device *dev) |
5578 | { | |
5579 | unsigned int count = dev->num_tx_queues; | |
5580 | struct netdev_queue *tx; | |
5581 | ||
5582 | BUG_ON(count < 1); | |
5583 | ||
5584 | tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL); | |
5585 | if (!tx) { | |
7b6cd1ce | 5586 | pr_err("netdev: Unable to allocate %u tx queues\n", count); |
e6484930 TH |
5587 | return -ENOMEM; |
5588 | } | |
5589 | dev->_tx = tx; | |
1d24eb48 | 5590 | |
e6484930 TH |
5591 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); |
5592 | spin_lock_init(&dev->tx_global_lock); | |
aa942104 CG |
5593 | |
5594 | return 0; | |
e6484930 TH |
5595 | } |
5596 | ||
1da177e4 LT |
5597 | /** |
5598 | * register_netdevice - register a network device | |
5599 | * @dev: device to register | |
5600 | * | |
5601 | * Take a completed network device structure and add it to the kernel | |
5602 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
5603 | * chain. 0 is returned on success. A negative errno code is returned | |
5604 | * on a failure to set up the device, or if the name is a duplicate. | |
5605 | * | |
5606 | * Callers must hold the rtnl semaphore. You may want | |
5607 | * register_netdev() instead of this. | |
5608 | * | |
5609 | * BUGS: | |
5610 | * The locking appears insufficient to guarantee two parallel registers | |
5611 | * will not get the same name. | |
5612 | */ | |
5613 | ||
5614 | int register_netdevice(struct net_device *dev) | |
5615 | { | |
1da177e4 | 5616 | int ret; |
d314774c | 5617 | struct net *net = dev_net(dev); |
1da177e4 LT |
5618 | |
5619 | BUG_ON(dev_boot_phase); | |
5620 | ASSERT_RTNL(); | |
5621 | ||
b17a7c17 SH |
5622 | might_sleep(); |
5623 | ||
1da177e4 LT |
5624 | /* When net_device's are persistent, this will be fatal. */ |
5625 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 5626 | BUG_ON(!net); |
1da177e4 | 5627 | |
f1f28aa3 | 5628 | spin_lock_init(&dev->addr_list_lock); |
cf508b12 | 5629 | netdev_set_addr_lockdep_class(dev); |
1da177e4 | 5630 | |
1da177e4 LT |
5631 | dev->iflink = -1; |
5632 | ||
828de4f6 | 5633 | ret = dev_get_valid_name(net, dev, dev->name); |
0696c3a8 PP |
5634 | if (ret < 0) |
5635 | goto out; | |
5636 | ||
1da177e4 | 5637 | /* Init, if this function is available */ |
d314774c SH |
5638 | if (dev->netdev_ops->ndo_init) { |
5639 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
5640 | if (ret) { |
5641 | if (ret > 0) | |
5642 | ret = -EIO; | |
90833aa4 | 5643 | goto out; |
1da177e4 LT |
5644 | } |
5645 | } | |
4ec93edb | 5646 | |
9c7dafbf PE |
5647 | ret = -EBUSY; |
5648 | if (!dev->ifindex) | |
5649 | dev->ifindex = dev_new_index(net); | |
5650 | else if (__dev_get_by_index(net, dev->ifindex)) | |
5651 | goto err_uninit; | |
5652 | ||
1da177e4 LT |
5653 | if (dev->iflink == -1) |
5654 | dev->iflink = dev->ifindex; | |
5655 | ||
5455c699 MM |
5656 | /* Transfer changeable features to wanted_features and enable |
5657 | * software offloads (GSO and GRO). | |
5658 | */ | |
5659 | dev->hw_features |= NETIF_F_SOFT_FEATURES; | |
14d1232f MM |
5660 | dev->features |= NETIF_F_SOFT_FEATURES; |
5661 | dev->wanted_features = dev->features & dev->hw_features; | |
1da177e4 | 5662 | |
c6e1a0d1 | 5663 | /* Turn on no cache copy if HW is doing checksum */ |
34324dc2 MM |
5664 | if (!(dev->flags & IFF_LOOPBACK)) { |
5665 | dev->hw_features |= NETIF_F_NOCACHE_COPY; | |
5666 | if (dev->features & NETIF_F_ALL_CSUM) { | |
5667 | dev->wanted_features |= NETIF_F_NOCACHE_COPY; | |
5668 | dev->features |= NETIF_F_NOCACHE_COPY; | |
5669 | } | |
c6e1a0d1 TH |
5670 | } |
5671 | ||
1180e7d6 | 5672 | /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. |
16c3ea78 | 5673 | */ |
1180e7d6 | 5674 | dev->vlan_features |= NETIF_F_HIGHDMA; |
16c3ea78 | 5675 | |
7ffbe3fd JB |
5676 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); |
5677 | ret = notifier_to_errno(ret); | |
5678 | if (ret) | |
5679 | goto err_uninit; | |
5680 | ||
8b41d188 | 5681 | ret = netdev_register_kobject(dev); |
b17a7c17 | 5682 | if (ret) |
7ce1b0ed | 5683 | goto err_uninit; |
b17a7c17 SH |
5684 | dev->reg_state = NETREG_REGISTERED; |
5685 | ||
6cb6a27c | 5686 | __netdev_update_features(dev); |
8e9b59b2 | 5687 | |
1da177e4 LT |
5688 | /* |
5689 | * Default initial state at registry is that the | |
5690 | * device is present. | |
5691 | */ | |
5692 | ||
5693 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
5694 | ||
8f4cccbb BH |
5695 | linkwatch_init_dev(dev); |
5696 | ||
1da177e4 | 5697 | dev_init_scheduler(dev); |
1da177e4 | 5698 | dev_hold(dev); |
ce286d32 | 5699 | list_netdevice(dev); |
7bf23575 | 5700 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 LT |
5701 | |
5702 | /* Notify protocols, that a new device appeared. */ | |
056925ab | 5703 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 5704 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
5705 | if (ret) { |
5706 | rollback_registered(dev); | |
5707 | dev->reg_state = NETREG_UNREGISTERED; | |
5708 | } | |
d90a909e EB |
5709 | /* |
5710 | * Prevent userspace races by waiting until the network | |
5711 | * device is fully setup before sending notifications. | |
5712 | */ | |
a2835763 PM |
5713 | if (!dev->rtnl_link_ops || |
5714 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
5715 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U); | |
1da177e4 LT |
5716 | |
5717 | out: | |
5718 | return ret; | |
7ce1b0ed HX |
5719 | |
5720 | err_uninit: | |
d314774c SH |
5721 | if (dev->netdev_ops->ndo_uninit) |
5722 | dev->netdev_ops->ndo_uninit(dev); | |
7ce1b0ed | 5723 | goto out; |
1da177e4 | 5724 | } |
d1b19dff | 5725 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 5726 | |
937f1ba5 BH |
5727 | /** |
5728 | * init_dummy_netdev - init a dummy network device for NAPI | |
5729 | * @dev: device to init | |
5730 | * | |
5731 | * This takes a network device structure and initialize the minimum | |
5732 | * amount of fields so it can be used to schedule NAPI polls without | |
5733 | * registering a full blown interface. This is to be used by drivers | |
5734 | * that need to tie several hardware interfaces to a single NAPI | |
5735 | * poll scheduler due to HW limitations. | |
5736 | */ | |
5737 | int init_dummy_netdev(struct net_device *dev) | |
5738 | { | |
5739 | /* Clear everything. Note we don't initialize spinlocks | |
5740 | * are they aren't supposed to be taken by any of the | |
5741 | * NAPI code and this dummy netdev is supposed to be | |
5742 | * only ever used for NAPI polls | |
5743 | */ | |
5744 | memset(dev, 0, sizeof(struct net_device)); | |
5745 | ||
5746 | /* make sure we BUG if trying to hit standard | |
5747 | * register/unregister code path | |
5748 | */ | |
5749 | dev->reg_state = NETREG_DUMMY; | |
5750 | ||
937f1ba5 BH |
5751 | /* NAPI wants this */ |
5752 | INIT_LIST_HEAD(&dev->napi_list); | |
5753 | ||
5754 | /* a dummy interface is started by default */ | |
5755 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
5756 | set_bit(__LINK_STATE_START, &dev->state); | |
5757 | ||
29b4433d ED |
5758 | /* Note : We dont allocate pcpu_refcnt for dummy devices, |
5759 | * because users of this 'device' dont need to change | |
5760 | * its refcount. | |
5761 | */ | |
5762 | ||
937f1ba5 BH |
5763 | return 0; |
5764 | } | |
5765 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
5766 | ||
5767 | ||
1da177e4 LT |
5768 | /** |
5769 | * register_netdev - register a network device | |
5770 | * @dev: device to register | |
5771 | * | |
5772 | * Take a completed network device structure and add it to the kernel | |
5773 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
5774 | * chain. 0 is returned on success. A negative errno code is returned | |
5775 | * on a failure to set up the device, or if the name is a duplicate. | |
5776 | * | |
38b4da38 | 5777 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
5778 | * and expands the device name if you passed a format string to |
5779 | * alloc_netdev. | |
5780 | */ | |
5781 | int register_netdev(struct net_device *dev) | |
5782 | { | |
5783 | int err; | |
5784 | ||
5785 | rtnl_lock(); | |
1da177e4 | 5786 | err = register_netdevice(dev); |
1da177e4 LT |
5787 | rtnl_unlock(); |
5788 | return err; | |
5789 | } | |
5790 | EXPORT_SYMBOL(register_netdev); | |
5791 | ||
29b4433d ED |
5792 | int netdev_refcnt_read(const struct net_device *dev) |
5793 | { | |
5794 | int i, refcnt = 0; | |
5795 | ||
5796 | for_each_possible_cpu(i) | |
5797 | refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); | |
5798 | return refcnt; | |
5799 | } | |
5800 | EXPORT_SYMBOL(netdev_refcnt_read); | |
5801 | ||
2c53040f | 5802 | /** |
1da177e4 | 5803 | * netdev_wait_allrefs - wait until all references are gone. |
3de7a37b | 5804 | * @dev: target net_device |
1da177e4 LT |
5805 | * |
5806 | * This is called when unregistering network devices. | |
5807 | * | |
5808 | * Any protocol or device that holds a reference should register | |
5809 | * for netdevice notification, and cleanup and put back the | |
5810 | * reference if they receive an UNREGISTER event. | |
5811 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 5812 | * call dev_put. |
1da177e4 LT |
5813 | */ |
5814 | static void netdev_wait_allrefs(struct net_device *dev) | |
5815 | { | |
5816 | unsigned long rebroadcast_time, warning_time; | |
29b4433d | 5817 | int refcnt; |
1da177e4 | 5818 | |
e014debe ED |
5819 | linkwatch_forget_dev(dev); |
5820 | ||
1da177e4 | 5821 | rebroadcast_time = warning_time = jiffies; |
29b4433d ED |
5822 | refcnt = netdev_refcnt_read(dev); |
5823 | ||
5824 | while (refcnt != 0) { | |
1da177e4 | 5825 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { |
6756ae4b | 5826 | rtnl_lock(); |
1da177e4 LT |
5827 | |
5828 | /* Rebroadcast unregister notification */ | |
056925ab | 5829 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
1da177e4 | 5830 | |
748e2d93 | 5831 | __rtnl_unlock(); |
0115e8e3 | 5832 | rcu_barrier(); |
748e2d93 ED |
5833 | rtnl_lock(); |
5834 | ||
0115e8e3 | 5835 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
1da177e4 LT |
5836 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, |
5837 | &dev->state)) { | |
5838 | /* We must not have linkwatch events | |
5839 | * pending on unregister. If this | |
5840 | * happens, we simply run the queue | |
5841 | * unscheduled, resulting in a noop | |
5842 | * for this device. | |
5843 | */ | |
5844 | linkwatch_run_queue(); | |
5845 | } | |
5846 | ||
6756ae4b | 5847 | __rtnl_unlock(); |
1da177e4 LT |
5848 | |
5849 | rebroadcast_time = jiffies; | |
5850 | } | |
5851 | ||
5852 | msleep(250); | |
5853 | ||
29b4433d ED |
5854 | refcnt = netdev_refcnt_read(dev); |
5855 | ||
1da177e4 | 5856 | if (time_after(jiffies, warning_time + 10 * HZ)) { |
7b6cd1ce JP |
5857 | pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", |
5858 | dev->name, refcnt); | |
1da177e4 LT |
5859 | warning_time = jiffies; |
5860 | } | |
5861 | } | |
5862 | } | |
5863 | ||
5864 | /* The sequence is: | |
5865 | * | |
5866 | * rtnl_lock(); | |
5867 | * ... | |
5868 | * register_netdevice(x1); | |
5869 | * register_netdevice(x2); | |
5870 | * ... | |
5871 | * unregister_netdevice(y1); | |
5872 | * unregister_netdevice(y2); | |
5873 | * ... | |
5874 | * rtnl_unlock(); | |
5875 | * free_netdev(y1); | |
5876 | * free_netdev(y2); | |
5877 | * | |
58ec3b4d | 5878 | * We are invoked by rtnl_unlock(). |
1da177e4 | 5879 | * This allows us to deal with problems: |
b17a7c17 | 5880 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
5881 | * without deadlocking with linkwatch via keventd. |
5882 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
5883 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
5884 | * |
5885 | * We must not return until all unregister events added during | |
5886 | * the interval the lock was held have been completed. | |
1da177e4 | 5887 | */ |
1da177e4 LT |
5888 | void netdev_run_todo(void) |
5889 | { | |
626ab0e6 | 5890 | struct list_head list; |
1da177e4 | 5891 | |
1da177e4 | 5892 | /* Snapshot list, allow later requests */ |
626ab0e6 | 5893 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
5894 | |
5895 | __rtnl_unlock(); | |
626ab0e6 | 5896 | |
0115e8e3 ED |
5897 | |
5898 | /* Wait for rcu callbacks to finish before next phase */ | |
850a545b EB |
5899 | if (!list_empty(&list)) |
5900 | rcu_barrier(); | |
5901 | ||
1da177e4 LT |
5902 | while (!list_empty(&list)) { |
5903 | struct net_device *dev | |
e5e26d75 | 5904 | = list_first_entry(&list, struct net_device, todo_list); |
1da177e4 LT |
5905 | list_del(&dev->todo_list); |
5906 | ||
748e2d93 | 5907 | rtnl_lock(); |
0115e8e3 | 5908 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
748e2d93 | 5909 | __rtnl_unlock(); |
0115e8e3 | 5910 | |
b17a7c17 | 5911 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
7b6cd1ce | 5912 | pr_err("network todo '%s' but state %d\n", |
b17a7c17 SH |
5913 | dev->name, dev->reg_state); |
5914 | dump_stack(); | |
5915 | continue; | |
5916 | } | |
1da177e4 | 5917 | |
b17a7c17 | 5918 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 5919 | |
152102c7 | 5920 | on_each_cpu(flush_backlog, dev, 1); |
6e583ce5 | 5921 | |
b17a7c17 | 5922 | netdev_wait_allrefs(dev); |
1da177e4 | 5923 | |
b17a7c17 | 5924 | /* paranoia */ |
29b4433d | 5925 | BUG_ON(netdev_refcnt_read(dev)); |
33d480ce ED |
5926 | WARN_ON(rcu_access_pointer(dev->ip_ptr)); |
5927 | WARN_ON(rcu_access_pointer(dev->ip6_ptr)); | |
547b792c | 5928 | WARN_ON(dev->dn_ptr); |
1da177e4 | 5929 | |
b17a7c17 SH |
5930 | if (dev->destructor) |
5931 | dev->destructor(dev); | |
9093bbb2 SH |
5932 | |
5933 | /* Free network device */ | |
5934 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 5935 | } |
1da177e4 LT |
5936 | } |
5937 | ||
3cfde79c BH |
5938 | /* Convert net_device_stats to rtnl_link_stats64. They have the same |
5939 | * fields in the same order, with only the type differing. | |
5940 | */ | |
77a1abf5 ED |
5941 | void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, |
5942 | const struct net_device_stats *netdev_stats) | |
3cfde79c BH |
5943 | { |
5944 | #if BITS_PER_LONG == 64 | |
77a1abf5 ED |
5945 | BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats)); |
5946 | memcpy(stats64, netdev_stats, sizeof(*stats64)); | |
3cfde79c BH |
5947 | #else |
5948 | size_t i, n = sizeof(*stats64) / sizeof(u64); | |
5949 | const unsigned long *src = (const unsigned long *)netdev_stats; | |
5950 | u64 *dst = (u64 *)stats64; | |
5951 | ||
5952 | BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) != | |
5953 | sizeof(*stats64) / sizeof(u64)); | |
5954 | for (i = 0; i < n; i++) | |
5955 | dst[i] = src[i]; | |
5956 | #endif | |
5957 | } | |
77a1abf5 | 5958 | EXPORT_SYMBOL(netdev_stats_to_stats64); |
3cfde79c | 5959 | |
eeda3fd6 SH |
5960 | /** |
5961 | * dev_get_stats - get network device statistics | |
5962 | * @dev: device to get statistics from | |
28172739 | 5963 | * @storage: place to store stats |
eeda3fd6 | 5964 | * |
d7753516 BH |
5965 | * Get network statistics from device. Return @storage. |
5966 | * The device driver may provide its own method by setting | |
5967 | * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; | |
5968 | * otherwise the internal statistics structure is used. | |
eeda3fd6 | 5969 | */ |
d7753516 BH |
5970 | struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, |
5971 | struct rtnl_link_stats64 *storage) | |
7004bf25 | 5972 | { |
eeda3fd6 SH |
5973 | const struct net_device_ops *ops = dev->netdev_ops; |
5974 | ||
28172739 ED |
5975 | if (ops->ndo_get_stats64) { |
5976 | memset(storage, 0, sizeof(*storage)); | |
caf586e5 ED |
5977 | ops->ndo_get_stats64(dev, storage); |
5978 | } else if (ops->ndo_get_stats) { | |
3cfde79c | 5979 | netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); |
caf586e5 ED |
5980 | } else { |
5981 | netdev_stats_to_stats64(storage, &dev->stats); | |
28172739 | 5982 | } |
caf586e5 | 5983 | storage->rx_dropped += atomic_long_read(&dev->rx_dropped); |
28172739 | 5984 | return storage; |
c45d286e | 5985 | } |
eeda3fd6 | 5986 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 5987 | |
24824a09 | 5988 | struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) |
dc2b4847 | 5989 | { |
24824a09 | 5990 | struct netdev_queue *queue = dev_ingress_queue(dev); |
dc2b4847 | 5991 | |
24824a09 ED |
5992 | #ifdef CONFIG_NET_CLS_ACT |
5993 | if (queue) | |
5994 | return queue; | |
5995 | queue = kzalloc(sizeof(*queue), GFP_KERNEL); | |
5996 | if (!queue) | |
5997 | return NULL; | |
5998 | netdev_init_one_queue(dev, queue, NULL); | |
24824a09 ED |
5999 | queue->qdisc = &noop_qdisc; |
6000 | queue->qdisc_sleeping = &noop_qdisc; | |
6001 | rcu_assign_pointer(dev->ingress_queue, queue); | |
6002 | #endif | |
6003 | return queue; | |
bb949fbd DM |
6004 | } |
6005 | ||
2c60db03 ED |
6006 | static const struct ethtool_ops default_ethtool_ops; |
6007 | ||
1da177e4 | 6008 | /** |
36909ea4 | 6009 | * alloc_netdev_mqs - allocate network device |
1da177e4 LT |
6010 | * @sizeof_priv: size of private data to allocate space for |
6011 | * @name: device name format string | |
6012 | * @setup: callback to initialize device | |
36909ea4 TH |
6013 | * @txqs: the number of TX subqueues to allocate |
6014 | * @rxqs: the number of RX subqueues to allocate | |
1da177e4 LT |
6015 | * |
6016 | * Allocates a struct net_device with private data area for driver use | |
f25f4e44 | 6017 | * and performs basic initialization. Also allocates subquue structs |
36909ea4 | 6018 | * for each queue on the device. |
1da177e4 | 6019 | */ |
36909ea4 TH |
6020 | struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, |
6021 | void (*setup)(struct net_device *), | |
6022 | unsigned int txqs, unsigned int rxqs) | |
1da177e4 | 6023 | { |
1da177e4 | 6024 | struct net_device *dev; |
7943986c | 6025 | size_t alloc_size; |
1ce8e7b5 | 6026 | struct net_device *p; |
1da177e4 | 6027 | |
b6fe17d6 SH |
6028 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
6029 | ||
36909ea4 | 6030 | if (txqs < 1) { |
7b6cd1ce | 6031 | pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); |
55513fb4 TH |
6032 | return NULL; |
6033 | } | |
6034 | ||
36909ea4 TH |
6035 | #ifdef CONFIG_RPS |
6036 | if (rxqs < 1) { | |
7b6cd1ce | 6037 | pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); |
36909ea4 TH |
6038 | return NULL; |
6039 | } | |
6040 | #endif | |
6041 | ||
fd2ea0a7 | 6042 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
6043 | if (sizeof_priv) { |
6044 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 6045 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
6046 | alloc_size += sizeof_priv; |
6047 | } | |
6048 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 6049 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 6050 | |
31380de9 | 6051 | p = kzalloc(alloc_size, GFP_KERNEL); |
1da177e4 | 6052 | if (!p) { |
7b6cd1ce | 6053 | pr_err("alloc_netdev: Unable to allocate device\n"); |
1da177e4 LT |
6054 | return NULL; |
6055 | } | |
1da177e4 | 6056 | |
1ce8e7b5 | 6057 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 6058 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc | 6059 | |
29b4433d ED |
6060 | dev->pcpu_refcnt = alloc_percpu(int); |
6061 | if (!dev->pcpu_refcnt) | |
e6484930 | 6062 | goto free_p; |
ab9c73cc | 6063 | |
ab9c73cc | 6064 | if (dev_addr_init(dev)) |
29b4433d | 6065 | goto free_pcpu; |
ab9c73cc | 6066 | |
22bedad3 | 6067 | dev_mc_init(dev); |
a748ee24 | 6068 | dev_uc_init(dev); |
ccffad25 | 6069 | |
c346dca1 | 6070 | dev_net_set(dev, &init_net); |
1da177e4 | 6071 | |
8d3bdbd5 | 6072 | dev->gso_max_size = GSO_MAX_SIZE; |
30b678d8 | 6073 | dev->gso_max_segs = GSO_MAX_SEGS; |
8d3bdbd5 | 6074 | |
8d3bdbd5 DM |
6075 | INIT_LIST_HEAD(&dev->napi_list); |
6076 | INIT_LIST_HEAD(&dev->unreg_list); | |
6077 | INIT_LIST_HEAD(&dev->link_watch_list); | |
6078 | dev->priv_flags = IFF_XMIT_DST_RELEASE; | |
6079 | setup(dev); | |
6080 | ||
36909ea4 TH |
6081 | dev->num_tx_queues = txqs; |
6082 | dev->real_num_tx_queues = txqs; | |
ed9af2e8 | 6083 | if (netif_alloc_netdev_queues(dev)) |
8d3bdbd5 | 6084 | goto free_all; |
e8a0464c | 6085 | |
df334545 | 6086 | #ifdef CONFIG_RPS |
36909ea4 TH |
6087 | dev->num_rx_queues = rxqs; |
6088 | dev->real_num_rx_queues = rxqs; | |
fe822240 | 6089 | if (netif_alloc_rx_queues(dev)) |
8d3bdbd5 | 6090 | goto free_all; |
df334545 | 6091 | #endif |
0a9627f2 | 6092 | |
1da177e4 | 6093 | strcpy(dev->name, name); |
cbda10fa | 6094 | dev->group = INIT_NETDEV_GROUP; |
2c60db03 ED |
6095 | if (!dev->ethtool_ops) |
6096 | dev->ethtool_ops = &default_ethtool_ops; | |
1da177e4 | 6097 | return dev; |
ab9c73cc | 6098 | |
8d3bdbd5 DM |
6099 | free_all: |
6100 | free_netdev(dev); | |
6101 | return NULL; | |
6102 | ||
29b4433d ED |
6103 | free_pcpu: |
6104 | free_percpu(dev->pcpu_refcnt); | |
ed9af2e8 | 6105 | kfree(dev->_tx); |
fe822240 TH |
6106 | #ifdef CONFIG_RPS |
6107 | kfree(dev->_rx); | |
6108 | #endif | |
6109 | ||
ab9c73cc JP |
6110 | free_p: |
6111 | kfree(p); | |
6112 | return NULL; | |
1da177e4 | 6113 | } |
36909ea4 | 6114 | EXPORT_SYMBOL(alloc_netdev_mqs); |
1da177e4 LT |
6115 | |
6116 | /** | |
6117 | * free_netdev - free network device | |
6118 | * @dev: device | |
6119 | * | |
4ec93edb YH |
6120 | * This function does the last stage of destroying an allocated device |
6121 | * interface. The reference to the device object is released. | |
1da177e4 LT |
6122 | * If this is the last reference then it will be freed. |
6123 | */ | |
6124 | void free_netdev(struct net_device *dev) | |
6125 | { | |
d565b0a1 HX |
6126 | struct napi_struct *p, *n; |
6127 | ||
f3005d7f DL |
6128 | release_net(dev_net(dev)); |
6129 | ||
e8a0464c | 6130 | kfree(dev->_tx); |
fe822240 TH |
6131 | #ifdef CONFIG_RPS |
6132 | kfree(dev->_rx); | |
6133 | #endif | |
e8a0464c | 6134 | |
33d480ce | 6135 | kfree(rcu_dereference_protected(dev->ingress_queue, 1)); |
24824a09 | 6136 | |
f001fde5 JP |
6137 | /* Flush device addresses */ |
6138 | dev_addr_flush(dev); | |
6139 | ||
d565b0a1 HX |
6140 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
6141 | netif_napi_del(p); | |
6142 | ||
29b4433d ED |
6143 | free_percpu(dev->pcpu_refcnt); |
6144 | dev->pcpu_refcnt = NULL; | |
6145 | ||
3041a069 | 6146 | /* Compatibility with error handling in drivers */ |
1da177e4 LT |
6147 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
6148 | kfree((char *)dev - dev->padded); | |
6149 | return; | |
6150 | } | |
6151 | ||
6152 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
6153 | dev->reg_state = NETREG_RELEASED; | |
6154 | ||
43cb76d9 GKH |
6155 | /* will free via device release */ |
6156 | put_device(&dev->dev); | |
1da177e4 | 6157 | } |
d1b19dff | 6158 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 6159 | |
f0db275a SH |
6160 | /** |
6161 | * synchronize_net - Synchronize with packet receive processing | |
6162 | * | |
6163 | * Wait for packets currently being received to be done. | |
6164 | * Does not block later packets from starting. | |
6165 | */ | |
4ec93edb | 6166 | void synchronize_net(void) |
1da177e4 LT |
6167 | { |
6168 | might_sleep(); | |
be3fc413 ED |
6169 | if (rtnl_is_locked()) |
6170 | synchronize_rcu_expedited(); | |
6171 | else | |
6172 | synchronize_rcu(); | |
1da177e4 | 6173 | } |
d1b19dff | 6174 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
6175 | |
6176 | /** | |
44a0873d | 6177 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 6178 | * @dev: device |
44a0873d | 6179 | * @head: list |
6ebfbc06 | 6180 | * |
1da177e4 | 6181 | * This function shuts down a device interface and removes it |
d59b54b1 | 6182 | * from the kernel tables. |
44a0873d | 6183 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
6184 | * |
6185 | * Callers must hold the rtnl semaphore. You may want | |
6186 | * unregister_netdev() instead of this. | |
6187 | */ | |
6188 | ||
44a0873d | 6189 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 6190 | { |
a6620712 HX |
6191 | ASSERT_RTNL(); |
6192 | ||
44a0873d | 6193 | if (head) { |
9fdce099 | 6194 | list_move_tail(&dev->unreg_list, head); |
44a0873d ED |
6195 | } else { |
6196 | rollback_registered(dev); | |
6197 | /* Finish processing unregister after unlock */ | |
6198 | net_set_todo(dev); | |
6199 | } | |
1da177e4 | 6200 | } |
44a0873d | 6201 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 6202 | |
9b5e383c ED |
6203 | /** |
6204 | * unregister_netdevice_many - unregister many devices | |
6205 | * @head: list of devices | |
9b5e383c ED |
6206 | */ |
6207 | void unregister_netdevice_many(struct list_head *head) | |
6208 | { | |
6209 | struct net_device *dev; | |
6210 | ||
6211 | if (!list_empty(head)) { | |
6212 | rollback_registered_many(head); | |
6213 | list_for_each_entry(dev, head, unreg_list) | |
6214 | net_set_todo(dev); | |
6215 | } | |
6216 | } | |
63c8099d | 6217 | EXPORT_SYMBOL(unregister_netdevice_many); |
9b5e383c | 6218 | |
1da177e4 LT |
6219 | /** |
6220 | * unregister_netdev - remove device from the kernel | |
6221 | * @dev: device | |
6222 | * | |
6223 | * This function shuts down a device interface and removes it | |
d59b54b1 | 6224 | * from the kernel tables. |
1da177e4 LT |
6225 | * |
6226 | * This is just a wrapper for unregister_netdevice that takes | |
6227 | * the rtnl semaphore. In general you want to use this and not | |
6228 | * unregister_netdevice. | |
6229 | */ | |
6230 | void unregister_netdev(struct net_device *dev) | |
6231 | { | |
6232 | rtnl_lock(); | |
6233 | unregister_netdevice(dev); | |
6234 | rtnl_unlock(); | |
6235 | } | |
1da177e4 LT |
6236 | EXPORT_SYMBOL(unregister_netdev); |
6237 | ||
ce286d32 EB |
6238 | /** |
6239 | * dev_change_net_namespace - move device to different nethost namespace | |
6240 | * @dev: device | |
6241 | * @net: network namespace | |
6242 | * @pat: If not NULL name pattern to try if the current device name | |
6243 | * is already taken in the destination network namespace. | |
6244 | * | |
6245 | * This function shuts down a device interface and moves it | |
6246 | * to a new network namespace. On success 0 is returned, on | |
6247 | * a failure a netagive errno code is returned. | |
6248 | * | |
6249 | * Callers must hold the rtnl semaphore. | |
6250 | */ | |
6251 | ||
6252 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
6253 | { | |
ce286d32 EB |
6254 | int err; |
6255 | ||
6256 | ASSERT_RTNL(); | |
6257 | ||
6258 | /* Don't allow namespace local devices to be moved. */ | |
6259 | err = -EINVAL; | |
6260 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
6261 | goto out; | |
6262 | ||
6263 | /* Ensure the device has been registrered */ | |
6264 | err = -EINVAL; | |
6265 | if (dev->reg_state != NETREG_REGISTERED) | |
6266 | goto out; | |
6267 | ||
6268 | /* Get out if there is nothing todo */ | |
6269 | err = 0; | |
878628fb | 6270 | if (net_eq(dev_net(dev), net)) |
ce286d32 EB |
6271 | goto out; |
6272 | ||
6273 | /* Pick the destination device name, and ensure | |
6274 | * we can use it in the destination network namespace. | |
6275 | */ | |
6276 | err = -EEXIST; | |
d9031024 | 6277 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
6278 | /* We get here if we can't use the current device name */ |
6279 | if (!pat) | |
6280 | goto out; | |
828de4f6 | 6281 | if (dev_get_valid_name(net, dev, pat) < 0) |
ce286d32 EB |
6282 | goto out; |
6283 | } | |
6284 | ||
6285 | /* | |
6286 | * And now a mini version of register_netdevice unregister_netdevice. | |
6287 | */ | |
6288 | ||
6289 | /* If device is running close it first. */ | |
9b772652 | 6290 | dev_close(dev); |
ce286d32 EB |
6291 | |
6292 | /* And unlink it from device chain */ | |
6293 | err = -ENODEV; | |
6294 | unlist_netdevice(dev); | |
6295 | ||
6296 | synchronize_net(); | |
6297 | ||
6298 | /* Shutdown queueing discipline. */ | |
6299 | dev_shutdown(dev); | |
6300 | ||
6301 | /* Notify protocols, that we are about to destroy | |
6302 | this device. They should clean all the things. | |
3b27e105 DL |
6303 | |
6304 | Note that dev->reg_state stays at NETREG_REGISTERED. | |
6305 | This is wanted because this way 8021q and macvlan know | |
6306 | the device is just moving and can keep their slaves up. | |
ce286d32 EB |
6307 | */ |
6308 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
6549dd43 G |
6309 | rcu_barrier(); |
6310 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); | |
d2237d35 | 6311 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U); |
ce286d32 EB |
6312 | |
6313 | /* | |
6314 | * Flush the unicast and multicast chains | |
6315 | */ | |
a748ee24 | 6316 | dev_uc_flush(dev); |
22bedad3 | 6317 | dev_mc_flush(dev); |
ce286d32 EB |
6318 | |
6319 | /* Actually switch the network namespace */ | |
c346dca1 | 6320 | dev_net_set(dev, net); |
ce286d32 | 6321 | |
ce286d32 EB |
6322 | /* If there is an ifindex conflict assign a new one */ |
6323 | if (__dev_get_by_index(net, dev->ifindex)) { | |
6324 | int iflink = (dev->iflink == dev->ifindex); | |
6325 | dev->ifindex = dev_new_index(net); | |
6326 | if (iflink) | |
6327 | dev->iflink = dev->ifindex; | |
6328 | } | |
6329 | ||
8b41d188 | 6330 | /* Fixup kobjects */ |
a1b3f594 | 6331 | err = device_rename(&dev->dev, dev->name); |
8b41d188 | 6332 | WARN_ON(err); |
ce286d32 EB |
6333 | |
6334 | /* Add the device back in the hashes */ | |
6335 | list_netdevice(dev); | |
6336 | ||
6337 | /* Notify protocols, that a new device appeared. */ | |
6338 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
6339 | ||
d90a909e EB |
6340 | /* |
6341 | * Prevent userspace races by waiting until the network | |
6342 | * device is fully setup before sending notifications. | |
6343 | */ | |
6344 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U); | |
6345 | ||
ce286d32 EB |
6346 | synchronize_net(); |
6347 | err = 0; | |
6348 | out: | |
6349 | return err; | |
6350 | } | |
463d0183 | 6351 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 6352 | |
1da177e4 LT |
6353 | static int dev_cpu_callback(struct notifier_block *nfb, |
6354 | unsigned long action, | |
6355 | void *ocpu) | |
6356 | { | |
6357 | struct sk_buff **list_skb; | |
1da177e4 LT |
6358 | struct sk_buff *skb; |
6359 | unsigned int cpu, oldcpu = (unsigned long)ocpu; | |
6360 | struct softnet_data *sd, *oldsd; | |
6361 | ||
8bb78442 | 6362 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
1da177e4 LT |
6363 | return NOTIFY_OK; |
6364 | ||
6365 | local_irq_disable(); | |
6366 | cpu = smp_processor_id(); | |
6367 | sd = &per_cpu(softnet_data, cpu); | |
6368 | oldsd = &per_cpu(softnet_data, oldcpu); | |
6369 | ||
6370 | /* Find end of our completion_queue. */ | |
6371 | list_skb = &sd->completion_queue; | |
6372 | while (*list_skb) | |
6373 | list_skb = &(*list_skb)->next; | |
6374 | /* Append completion queue from offline CPU. */ | |
6375 | *list_skb = oldsd->completion_queue; | |
6376 | oldsd->completion_queue = NULL; | |
6377 | ||
1da177e4 | 6378 | /* Append output queue from offline CPU. */ |
a9cbd588 CG |
6379 | if (oldsd->output_queue) { |
6380 | *sd->output_queue_tailp = oldsd->output_queue; | |
6381 | sd->output_queue_tailp = oldsd->output_queue_tailp; | |
6382 | oldsd->output_queue = NULL; | |
6383 | oldsd->output_queue_tailp = &oldsd->output_queue; | |
6384 | } | |
264524d5 HC |
6385 | /* Append NAPI poll list from offline CPU. */ |
6386 | if (!list_empty(&oldsd->poll_list)) { | |
6387 | list_splice_init(&oldsd->poll_list, &sd->poll_list); | |
6388 | raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
6389 | } | |
1da177e4 LT |
6390 | |
6391 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
6392 | local_irq_enable(); | |
6393 | ||
6394 | /* Process offline CPU's input_pkt_queue */ | |
76cc8b13 | 6395 | while ((skb = __skb_dequeue(&oldsd->process_queue))) { |
1da177e4 | 6396 | netif_rx(skb); |
76cc8b13 | 6397 | input_queue_head_incr(oldsd); |
fec5e652 | 6398 | } |
76cc8b13 | 6399 | while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) { |
6e7676c1 | 6400 | netif_rx(skb); |
76cc8b13 TH |
6401 | input_queue_head_incr(oldsd); |
6402 | } | |
1da177e4 LT |
6403 | |
6404 | return NOTIFY_OK; | |
6405 | } | |
1da177e4 LT |
6406 | |
6407 | ||
7f353bf2 | 6408 | /** |
b63365a2 HX |
6409 | * netdev_increment_features - increment feature set by one |
6410 | * @all: current feature set | |
6411 | * @one: new feature set | |
6412 | * @mask: mask feature set | |
7f353bf2 HX |
6413 | * |
6414 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
6415 | * @one to the master device with current feature set @all. Will not |
6416 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 6417 | */ |
c8f44aff MM |
6418 | netdev_features_t netdev_increment_features(netdev_features_t all, |
6419 | netdev_features_t one, netdev_features_t mask) | |
b63365a2 | 6420 | { |
1742f183 MM |
6421 | if (mask & NETIF_F_GEN_CSUM) |
6422 | mask |= NETIF_F_ALL_CSUM; | |
6423 | mask |= NETIF_F_VLAN_CHALLENGED; | |
7f353bf2 | 6424 | |
1742f183 MM |
6425 | all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask; |
6426 | all &= one | ~NETIF_F_ALL_FOR_ALL; | |
c6e1a0d1 | 6427 | |
1742f183 MM |
6428 | /* If one device supports hw checksumming, set for all. */ |
6429 | if (all & NETIF_F_GEN_CSUM) | |
6430 | all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM); | |
7f353bf2 HX |
6431 | |
6432 | return all; | |
6433 | } | |
b63365a2 | 6434 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 6435 | |
30d97d35 PE |
6436 | static struct hlist_head *netdev_create_hash(void) |
6437 | { | |
6438 | int i; | |
6439 | struct hlist_head *hash; | |
6440 | ||
6441 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
6442 | if (hash != NULL) | |
6443 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
6444 | INIT_HLIST_HEAD(&hash[i]); | |
6445 | ||
6446 | return hash; | |
6447 | } | |
6448 | ||
881d966b | 6449 | /* Initialize per network namespace state */ |
4665079c | 6450 | static int __net_init netdev_init(struct net *net) |
881d966b | 6451 | { |
734b6541 RM |
6452 | if (net != &init_net) |
6453 | INIT_LIST_HEAD(&net->dev_base_head); | |
881d966b | 6454 | |
30d97d35 PE |
6455 | net->dev_name_head = netdev_create_hash(); |
6456 | if (net->dev_name_head == NULL) | |
6457 | goto err_name; | |
881d966b | 6458 | |
30d97d35 PE |
6459 | net->dev_index_head = netdev_create_hash(); |
6460 | if (net->dev_index_head == NULL) | |
6461 | goto err_idx; | |
881d966b EB |
6462 | |
6463 | return 0; | |
30d97d35 PE |
6464 | |
6465 | err_idx: | |
6466 | kfree(net->dev_name_head); | |
6467 | err_name: | |
6468 | return -ENOMEM; | |
881d966b EB |
6469 | } |
6470 | ||
f0db275a SH |
6471 | /** |
6472 | * netdev_drivername - network driver for the device | |
6473 | * @dev: network device | |
f0db275a SH |
6474 | * |
6475 | * Determine network driver for device. | |
6476 | */ | |
3019de12 | 6477 | const char *netdev_drivername(const struct net_device *dev) |
6579e57b | 6478 | { |
cf04a4c7 SH |
6479 | const struct device_driver *driver; |
6480 | const struct device *parent; | |
3019de12 | 6481 | const char *empty = ""; |
6579e57b AV |
6482 | |
6483 | parent = dev->dev.parent; | |
6579e57b | 6484 | if (!parent) |
3019de12 | 6485 | return empty; |
6579e57b AV |
6486 | |
6487 | driver = parent->driver; | |
6488 | if (driver && driver->name) | |
3019de12 DM |
6489 | return driver->name; |
6490 | return empty; | |
6579e57b AV |
6491 | } |
6492 | ||
b004ff49 | 6493 | static int __netdev_printk(const char *level, const struct net_device *dev, |
256df2f3 JP |
6494 | struct va_format *vaf) |
6495 | { | |
6496 | int r; | |
6497 | ||
b004ff49 | 6498 | if (dev && dev->dev.parent) { |
666f355f JP |
6499 | r = dev_printk_emit(level[1] - '0', |
6500 | dev->dev.parent, | |
6501 | "%s %s %s: %pV", | |
6502 | dev_driver_string(dev->dev.parent), | |
6503 | dev_name(dev->dev.parent), | |
6504 | netdev_name(dev), vaf); | |
b004ff49 | 6505 | } else if (dev) { |
256df2f3 | 6506 | r = printk("%s%s: %pV", level, netdev_name(dev), vaf); |
b004ff49 | 6507 | } else { |
256df2f3 | 6508 | r = printk("%s(NULL net_device): %pV", level, vaf); |
b004ff49 | 6509 | } |
256df2f3 JP |
6510 | |
6511 | return r; | |
6512 | } | |
6513 | ||
6514 | int netdev_printk(const char *level, const struct net_device *dev, | |
6515 | const char *format, ...) | |
6516 | { | |
6517 | struct va_format vaf; | |
6518 | va_list args; | |
6519 | int r; | |
6520 | ||
6521 | va_start(args, format); | |
6522 | ||
6523 | vaf.fmt = format; | |
6524 | vaf.va = &args; | |
6525 | ||
6526 | r = __netdev_printk(level, dev, &vaf); | |
b004ff49 | 6527 | |
256df2f3 JP |
6528 | va_end(args); |
6529 | ||
6530 | return r; | |
6531 | } | |
6532 | EXPORT_SYMBOL(netdev_printk); | |
6533 | ||
6534 | #define define_netdev_printk_level(func, level) \ | |
6535 | int func(const struct net_device *dev, const char *fmt, ...) \ | |
6536 | { \ | |
6537 | int r; \ | |
6538 | struct va_format vaf; \ | |
6539 | va_list args; \ | |
6540 | \ | |
6541 | va_start(args, fmt); \ | |
6542 | \ | |
6543 | vaf.fmt = fmt; \ | |
6544 | vaf.va = &args; \ | |
6545 | \ | |
6546 | r = __netdev_printk(level, dev, &vaf); \ | |
b004ff49 | 6547 | \ |
256df2f3 JP |
6548 | va_end(args); \ |
6549 | \ | |
6550 | return r; \ | |
6551 | } \ | |
6552 | EXPORT_SYMBOL(func); | |
6553 | ||
6554 | define_netdev_printk_level(netdev_emerg, KERN_EMERG); | |
6555 | define_netdev_printk_level(netdev_alert, KERN_ALERT); | |
6556 | define_netdev_printk_level(netdev_crit, KERN_CRIT); | |
6557 | define_netdev_printk_level(netdev_err, KERN_ERR); | |
6558 | define_netdev_printk_level(netdev_warn, KERN_WARNING); | |
6559 | define_netdev_printk_level(netdev_notice, KERN_NOTICE); | |
6560 | define_netdev_printk_level(netdev_info, KERN_INFO); | |
6561 | ||
4665079c | 6562 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
6563 | { |
6564 | kfree(net->dev_name_head); | |
6565 | kfree(net->dev_index_head); | |
6566 | } | |
6567 | ||
022cbae6 | 6568 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
6569 | .init = netdev_init, |
6570 | .exit = netdev_exit, | |
6571 | }; | |
6572 | ||
4665079c | 6573 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 6574 | { |
e008b5fc | 6575 | struct net_device *dev, *aux; |
ce286d32 | 6576 | /* |
e008b5fc | 6577 | * Push all migratable network devices back to the |
ce286d32 EB |
6578 | * initial network namespace |
6579 | */ | |
6580 | rtnl_lock(); | |
e008b5fc | 6581 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 6582 | int err; |
aca51397 | 6583 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
6584 | |
6585 | /* Ignore unmoveable devices (i.e. loopback) */ | |
6586 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
6587 | continue; | |
6588 | ||
e008b5fc EB |
6589 | /* Leave virtual devices for the generic cleanup */ |
6590 | if (dev->rtnl_link_ops) | |
6591 | continue; | |
d0c082ce | 6592 | |
25985edc | 6593 | /* Push remaining network devices to init_net */ |
aca51397 PE |
6594 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
6595 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
ce286d32 | 6596 | if (err) { |
7b6cd1ce JP |
6597 | pr_emerg("%s: failed to move %s to init_net: %d\n", |
6598 | __func__, dev->name, err); | |
aca51397 | 6599 | BUG(); |
ce286d32 EB |
6600 | } |
6601 | } | |
6602 | rtnl_unlock(); | |
6603 | } | |
6604 | ||
04dc7f6b EB |
6605 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
6606 | { | |
6607 | /* At exit all network devices most be removed from a network | |
b595076a | 6608 | * namespace. Do this in the reverse order of registration. |
04dc7f6b EB |
6609 | * Do this across as many network namespaces as possible to |
6610 | * improve batching efficiency. | |
6611 | */ | |
6612 | struct net_device *dev; | |
6613 | struct net *net; | |
6614 | LIST_HEAD(dev_kill_list); | |
6615 | ||
6616 | rtnl_lock(); | |
6617 | list_for_each_entry(net, net_list, exit_list) { | |
6618 | for_each_netdev_reverse(net, dev) { | |
6619 | if (dev->rtnl_link_ops) | |
6620 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); | |
6621 | else | |
6622 | unregister_netdevice_queue(dev, &dev_kill_list); | |
6623 | } | |
6624 | } | |
6625 | unregister_netdevice_many(&dev_kill_list); | |
ceaaec98 | 6626 | list_del(&dev_kill_list); |
04dc7f6b EB |
6627 | rtnl_unlock(); |
6628 | } | |
6629 | ||
022cbae6 | 6630 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 6631 | .exit = default_device_exit, |
04dc7f6b | 6632 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
6633 | }; |
6634 | ||
1da177e4 LT |
6635 | /* |
6636 | * Initialize the DEV module. At boot time this walks the device list and | |
6637 | * unhooks any devices that fail to initialise (normally hardware not | |
6638 | * present) and leaves us with a valid list of present and active devices. | |
6639 | * | |
6640 | */ | |
6641 | ||
6642 | /* | |
6643 | * This is called single threaded during boot, so no need | |
6644 | * to take the rtnl semaphore. | |
6645 | */ | |
6646 | static int __init net_dev_init(void) | |
6647 | { | |
6648 | int i, rc = -ENOMEM; | |
6649 | ||
6650 | BUG_ON(!dev_boot_phase); | |
6651 | ||
1da177e4 LT |
6652 | if (dev_proc_init()) |
6653 | goto out; | |
6654 | ||
8b41d188 | 6655 | if (netdev_kobject_init()) |
1da177e4 LT |
6656 | goto out; |
6657 | ||
6658 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 6659 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
6660 | INIT_LIST_HEAD(&ptype_base[i]); |
6661 | ||
881d966b EB |
6662 | if (register_pernet_subsys(&netdev_net_ops)) |
6663 | goto out; | |
1da177e4 LT |
6664 | |
6665 | /* | |
6666 | * Initialise the packet receive queues. | |
6667 | */ | |
6668 | ||
6f912042 | 6669 | for_each_possible_cpu(i) { |
e36fa2f7 | 6670 | struct softnet_data *sd = &per_cpu(softnet_data, i); |
1da177e4 | 6671 | |
dee42870 | 6672 | memset(sd, 0, sizeof(*sd)); |
e36fa2f7 | 6673 | skb_queue_head_init(&sd->input_pkt_queue); |
6e7676c1 | 6674 | skb_queue_head_init(&sd->process_queue); |
e36fa2f7 ED |
6675 | sd->completion_queue = NULL; |
6676 | INIT_LIST_HEAD(&sd->poll_list); | |
a9cbd588 CG |
6677 | sd->output_queue = NULL; |
6678 | sd->output_queue_tailp = &sd->output_queue; | |
df334545 | 6679 | #ifdef CONFIG_RPS |
e36fa2f7 ED |
6680 | sd->csd.func = rps_trigger_softirq; |
6681 | sd->csd.info = sd; | |
6682 | sd->csd.flags = 0; | |
6683 | sd->cpu = i; | |
1e94d72f | 6684 | #endif |
0a9627f2 | 6685 | |
e36fa2f7 ED |
6686 | sd->backlog.poll = process_backlog; |
6687 | sd->backlog.weight = weight_p; | |
6688 | sd->backlog.gro_list = NULL; | |
6689 | sd->backlog.gro_count = 0; | |
1da177e4 LT |
6690 | } |
6691 | ||
1da177e4 LT |
6692 | dev_boot_phase = 0; |
6693 | ||
505d4f73 EB |
6694 | /* The loopback device is special if any other network devices |
6695 | * is present in a network namespace the loopback device must | |
6696 | * be present. Since we now dynamically allocate and free the | |
6697 | * loopback device ensure this invariant is maintained by | |
6698 | * keeping the loopback device as the first device on the | |
6699 | * list of network devices. Ensuring the loopback devices | |
6700 | * is the first device that appears and the last network device | |
6701 | * that disappears. | |
6702 | */ | |
6703 | if (register_pernet_device(&loopback_net_ops)) | |
6704 | goto out; | |
6705 | ||
6706 | if (register_pernet_device(&default_device_ops)) | |
6707 | goto out; | |
6708 | ||
962cf36c CM |
6709 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
6710 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 LT |
6711 | |
6712 | hotcpu_notifier(dev_cpu_callback, 0); | |
6713 | dst_init(); | |
6714 | dev_mcast_init(); | |
6715 | rc = 0; | |
6716 | out: | |
6717 | return rc; | |
6718 | } | |
6719 | ||
6720 | subsys_initcall(net_dev_init); | |
6721 | ||
e88721f8 KK |
6722 | static int __init initialize_hashrnd(void) |
6723 | { | |
0a9627f2 | 6724 | get_random_bytes(&hashrnd, sizeof(hashrnd)); |
e88721f8 KK |
6725 | return 0; |
6726 | } | |
6727 | ||
6728 | late_initcall_sync(initialize_hashrnd); | |
6729 |