]>
Commit | Line | Data |
---|---|---|
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 | |
10 | * Authors: Ross Biro | |
11 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> | |
12 | * Mark Evans, <evansmp@uhura.aston.ac.uk> | |
13 | * | |
14 | * Additional Authors: | |
15 | * Florian la Roche <rzsfl@rz.uni-sb.de> | |
16 | * Alan Cox <gw4pts@gw4pts.ampr.org> | |
17 | * David Hinds <dahinds@users.sourceforge.net> | |
18 | * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> | |
19 | * Adam Sulmicki <adam@cfar.umd.edu> | |
20 | * Pekka Riikonen <priikone@poesidon.pspt.fi> | |
21 | * | |
22 | * Changes: | |
23 | * D.J. Barrow : Fixed bug where dev->refcnt gets set | |
24 | * to 2 if register_netdev gets called | |
25 | * before net_dev_init & also removed a | |
26 | * few lines of code in the process. | |
27 | * Alan Cox : device private ioctl copies fields back. | |
28 | * Alan Cox : Transmit queue code does relevant | |
29 | * stunts to keep the queue safe. | |
30 | * Alan Cox : Fixed double lock. | |
31 | * Alan Cox : Fixed promisc NULL pointer trap | |
32 | * ???????? : Support the full private ioctl range | |
33 | * Alan Cox : Moved ioctl permission check into | |
34 | * drivers | |
35 | * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI | |
36 | * Alan Cox : 100 backlog just doesn't cut it when | |
37 | * you start doing multicast video 8) | |
38 | * Alan Cox : Rewrote net_bh and list manager. | |
39 | * Alan Cox : Fix ETH_P_ALL echoback lengths. | |
40 | * Alan Cox : Took out transmit every packet pass | |
41 | * Saved a few bytes in the ioctl handler | |
42 | * Alan Cox : Network driver sets packet type before | |
43 | * calling netif_rx. Saves a function | |
44 | * call a packet. | |
45 | * Alan Cox : Hashed net_bh() | |
46 | * Richard Kooijman: Timestamp fixes. | |
47 | * Alan Cox : Wrong field in SIOCGIFDSTADDR | |
48 | * Alan Cox : Device lock protection. | |
49 | * Alan Cox : Fixed nasty side effect of device close | |
50 | * changes. | |
51 | * Rudi Cilibrasi : Pass the right thing to | |
52 | * set_mac_address() | |
53 | * Dave Miller : 32bit quantity for the device lock to | |
54 | * make it work out on a Sparc. | |
55 | * Bjorn Ekwall : Added KERNELD hack. | |
56 | * Alan Cox : Cleaned up the backlog initialise. | |
57 | * Craig Metz : SIOCGIFCONF fix if space for under | |
58 | * 1 device. | |
59 | * Thomas Bogendoerfer : Return ENODEV for dev_open, if there | |
60 | * is no device open function. | |
61 | * Andi Kleen : Fix error reporting for SIOCGIFCONF | |
62 | * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF | |
63 | * Cyrus Durgin : Cleaned for KMOD | |
64 | * Adam Sulmicki : Bug Fix : Network Device Unload | |
65 | * A network device unload needs to purge | |
66 | * the backlog queue. | |
67 | * Paul Rusty Russell : SIOCSIFNAME | |
68 | * Pekka Riikonen : Netdev boot-time settings code | |
69 | * Andrew Morton : Make unregister_netdevice wait | |
70 | * indefinitely on dev->refcnt | |
71 | * J Hadi Salim : - Backlog queue sampling | |
72 | * - netif_rx() feedback | |
73 | */ | |
74 | ||
75 | #include <asm/uaccess.h> | |
76 | #include <asm/system.h> | |
77 | #include <linux/bitops.h> | |
78 | #include <linux/capability.h> | |
79 | #include <linux/cpu.h> | |
80 | #include <linux/types.h> | |
81 | #include <linux/kernel.h> | |
82 | #include <linux/hash.h> | |
83 | #include <linux/slab.h> | |
84 | #include <linux/sched.h> | |
85 | #include <linux/mutex.h> | |
86 | #include <linux/string.h> | |
87 | #include <linux/mm.h> | |
88 | #include <linux/socket.h> | |
89 | #include <linux/sockios.h> | |
90 | #include <linux/errno.h> | |
91 | #include <linux/interrupt.h> | |
92 | #include <linux/if_ether.h> | |
93 | #include <linux/netdevice.h> | |
94 | #include <linux/etherdevice.h> | |
95 | #include <linux/ethtool.h> | |
96 | #include <linux/notifier.h> | |
97 | #include <linux/skbuff.h> | |
98 | #include <net/net_namespace.h> | |
99 | #include <net/sock.h> | |
100 | #include <linux/rtnetlink.h> | |
101 | #include <linux/proc_fs.h> | |
102 | #include <linux/seq_file.h> | |
103 | #include <linux/stat.h> | |
104 | #include <net/dst.h> | |
105 | #include <net/pkt_sched.h> | |
106 | #include <net/checksum.h> | |
107 | #include <net/xfrm.h> | |
108 | #include <linux/highmem.h> | |
109 | #include <linux/init.h> | |
110 | #include <linux/kmod.h> | |
111 | #include <linux/module.h> | |
112 | #include <linux/netpoll.h> | |
113 | #include <linux/rcupdate.h> | |
114 | #include <linux/delay.h> | |
115 | #include <net/wext.h> | |
116 | #include <net/iw_handler.h> | |
117 | #include <asm/current.h> | |
118 | #include <linux/audit.h> | |
119 | #include <linux/dmaengine.h> | |
120 | #include <linux/err.h> | |
121 | #include <linux/ctype.h> | |
122 | #include <linux/if_arp.h> | |
123 | #include <linux/if_vlan.h> | |
124 | #include <linux/ip.h> | |
125 | #include <net/ip.h> | |
126 | #include <linux/ipv6.h> | |
127 | #include <linux/in.h> | |
128 | #include <linux/jhash.h> | |
129 | #include <linux/random.h> | |
130 | #include <trace/events/napi.h> | |
131 | #include <trace/events/net.h> | |
132 | #include <trace/events/skb.h> | |
133 | #include <linux/pci.h> | |
134 | #include <linux/inetdevice.h> | |
135 | ||
136 | #include "net-sysfs.h" | |
137 | ||
138 | /* Instead of increasing this, you should create a hash table. */ | |
139 | #define MAX_GRO_SKBS 8 | |
140 | ||
141 | /* This should be increased if a protocol with a bigger head is added. */ | |
142 | #define GRO_MAX_HEAD (MAX_HEADER + 128) | |
143 | ||
144 | /* | |
145 | * The list of packet types we will receive (as opposed to discard) | |
146 | * and the routines to invoke. | |
147 | * | |
148 | * Why 16. Because with 16 the only overlap we get on a hash of the | |
149 | * low nibble of the protocol value is RARP/SNAP/X.25. | |
150 | * | |
151 | * NOTE: That is no longer true with the addition of VLAN tags. Not | |
152 | * sure which should go first, but I bet it won't make much | |
153 | * difference if we are running VLANs. The good news is that | |
154 | * this protocol won't be in the list unless compiled in, so | |
155 | * the average user (w/out VLANs) will not be adversely affected. | |
156 | * --BLG | |
157 | * | |
158 | * 0800 IP | |
159 | * 8100 802.1Q VLAN | |
160 | * 0001 802.3 | |
161 | * 0002 AX.25 | |
162 | * 0004 802.2 | |
163 | * 8035 RARP | |
164 | * 0005 SNAP | |
165 | * 0805 X.25 | |
166 | * 0806 ARP | |
167 | * 8137 IPX | |
168 | * 0009 Localtalk | |
169 | * 86DD IPv6 | |
170 | */ | |
171 | ||
172 | #define PTYPE_HASH_SIZE (16) | |
173 | #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1) | |
174 | ||
175 | static DEFINE_SPINLOCK(ptype_lock); | |
176 | static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; | |
177 | static struct list_head ptype_all __read_mostly; /* Taps */ | |
178 | ||
179 | /* | |
180 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl | |
181 | * semaphore. | |
182 | * | |
183 | * Pure readers hold dev_base_lock for reading, or rcu_read_lock() | |
184 | * | |
185 | * Writers must hold the rtnl semaphore while they loop through the | |
186 | * dev_base_head list, and hold dev_base_lock for writing when they do the | |
187 | * actual updates. This allows pure readers to access the list even | |
188 | * while a writer is preparing to update it. | |
189 | * | |
190 | * To put it another way, dev_base_lock is held for writing only to | |
191 | * protect against pure readers; the rtnl semaphore provides the | |
192 | * protection against other writers. | |
193 | * | |
194 | * See, for example usages, register_netdevice() and | |
195 | * unregister_netdevice(), which must be called with the rtnl | |
196 | * semaphore held. | |
197 | */ | |
198 | DEFINE_RWLOCK(dev_base_lock); | |
199 | EXPORT_SYMBOL(dev_base_lock); | |
200 | ||
201 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) | |
202 | { | |
203 | unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ)); | |
204 | return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; | |
205 | } | |
206 | ||
207 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) | |
208 | { | |
209 | return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; | |
210 | } | |
211 | ||
212 | static inline void rps_lock(struct softnet_data *sd) | |
213 | { | |
214 | #ifdef CONFIG_RPS | |
215 | spin_lock(&sd->input_pkt_queue.lock); | |
216 | #endif | |
217 | } | |
218 | ||
219 | static inline void rps_unlock(struct softnet_data *sd) | |
220 | { | |
221 | #ifdef CONFIG_RPS | |
222 | spin_unlock(&sd->input_pkt_queue.lock); | |
223 | #endif | |
224 | } | |
225 | ||
226 | /* Device list insertion */ | |
227 | static int list_netdevice(struct net_device *dev) | |
228 | { | |
229 | struct net *net = dev_net(dev); | |
230 | ||
231 | ASSERT_RTNL(); | |
232 | ||
233 | write_lock_bh(&dev_base_lock); | |
234 | list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); | |
235 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
236 | hlist_add_head_rcu(&dev->index_hlist, | |
237 | dev_index_hash(net, dev->ifindex)); | |
238 | write_unlock_bh(&dev_base_lock); | |
239 | return 0; | |
240 | } | |
241 | ||
242 | /* Device list removal | |
243 | * caller must respect a RCU grace period before freeing/reusing dev | |
244 | */ | |
245 | static void unlist_netdevice(struct net_device *dev) | |
246 | { | |
247 | ASSERT_RTNL(); | |
248 | ||
249 | /* Unlink dev from the device chain */ | |
250 | write_lock_bh(&dev_base_lock); | |
251 | list_del_rcu(&dev->dev_list); | |
252 | hlist_del_rcu(&dev->name_hlist); | |
253 | hlist_del_rcu(&dev->index_hlist); | |
254 | write_unlock_bh(&dev_base_lock); | |
255 | } | |
256 | ||
257 | /* | |
258 | * Our notifier list | |
259 | */ | |
260 | ||
261 | static RAW_NOTIFIER_HEAD(netdev_chain); | |
262 | ||
263 | /* | |
264 | * Device drivers call our routines to queue packets here. We empty the | |
265 | * queue in the local softnet handler. | |
266 | */ | |
267 | ||
268 | DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); | |
269 | EXPORT_PER_CPU_SYMBOL(softnet_data); | |
270 | ||
271 | #ifdef CONFIG_LOCKDEP | |
272 | /* | |
273 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class | |
274 | * according to dev->type | |
275 | */ | |
276 | static const unsigned short netdev_lock_type[] = | |
277 | {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
278 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, | |
279 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
280 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
281 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
282 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
283 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
284 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
285 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
286 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
287 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
288 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
289 | ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211, | |
290 | ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, | |
291 | ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154, | |
292 | ARPHRD_VOID, ARPHRD_NONE}; | |
293 | ||
294 | static const char *const netdev_lock_name[] = | |
295 | {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", | |
296 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
297 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
298 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
299 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
300 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
301 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
302 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
303 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
304 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
305 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
306 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
307 | "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211", | |
308 | "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", | |
309 | "_xmit_PHONET_PIPE", "_xmit_IEEE802154", | |
310 | "_xmit_VOID", "_xmit_NONE"}; | |
311 | ||
312 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
313 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
314 | ||
315 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
316 | { | |
317 | int i; | |
318 | ||
319 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
320 | if (netdev_lock_type[i] == dev_type) | |
321 | return i; | |
322 | /* the last key is used by default */ | |
323 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
324 | } | |
325 | ||
326 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, | |
327 | unsigned short dev_type) | |
328 | { | |
329 | int i; | |
330 | ||
331 | i = netdev_lock_pos(dev_type); | |
332 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
333 | netdev_lock_name[i]); | |
334 | } | |
335 | ||
336 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
337 | { | |
338 | int i; | |
339 | ||
340 | i = netdev_lock_pos(dev->type); | |
341 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
342 | &netdev_addr_lock_key[i], | |
343 | netdev_lock_name[i]); | |
344 | } | |
345 | #else | |
346 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, | |
347 | unsigned short dev_type) | |
348 | { | |
349 | } | |
350 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
351 | { | |
352 | } | |
353 | #endif | |
354 | ||
355 | /******************************************************************************* | |
356 | ||
357 | Protocol management and registration routines | |
358 | ||
359 | *******************************************************************************/ | |
360 | ||
361 | /* | |
362 | * Add a protocol ID to the list. Now that the input handler is | |
363 | * smarter we can dispense with all the messy stuff that used to be | |
364 | * here. | |
365 | * | |
366 | * BEWARE!!! Protocol handlers, mangling input packets, | |
367 | * MUST BE last in hash buckets and checking protocol handlers | |
368 | * MUST start from promiscuous ptype_all chain in net_bh. | |
369 | * It is true now, do not change it. | |
370 | * Explanation follows: if protocol handler, mangling packet, will | |
371 | * be the first on list, it is not able to sense, that packet | |
372 | * is cloned and should be copied-on-write, so that it will | |
373 | * change it and subsequent readers will get broken packet. | |
374 | * --ANK (980803) | |
375 | */ | |
376 | ||
377 | static inline struct list_head *ptype_head(const struct packet_type *pt) | |
378 | { | |
379 | if (pt->type == htons(ETH_P_ALL)) | |
380 | return &ptype_all; | |
381 | else | |
382 | return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
383 | } | |
384 | ||
385 | /** | |
386 | * dev_add_pack - add packet handler | |
387 | * @pt: packet type declaration | |
388 | * | |
389 | * Add a protocol handler to the networking stack. The passed &packet_type | |
390 | * is linked into kernel lists and may not be freed until it has been | |
391 | * removed from the kernel lists. | |
392 | * | |
393 | * This call does not sleep therefore it can not | |
394 | * guarantee all CPU's that are in middle of receiving packets | |
395 | * will see the new packet type (until the next received packet). | |
396 | */ | |
397 | ||
398 | void dev_add_pack(struct packet_type *pt) | |
399 | { | |
400 | struct list_head *head = ptype_head(pt); | |
401 | ||
402 | spin_lock(&ptype_lock); | |
403 | list_add_rcu(&pt->list, head); | |
404 | spin_unlock(&ptype_lock); | |
405 | } | |
406 | EXPORT_SYMBOL(dev_add_pack); | |
407 | ||
408 | /** | |
409 | * __dev_remove_pack - remove packet handler | |
410 | * @pt: packet type declaration | |
411 | * | |
412 | * Remove a protocol handler that was previously added to the kernel | |
413 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
414 | * from the kernel lists and can be freed or reused once this function | |
415 | * returns. | |
416 | * | |
417 | * The packet type might still be in use by receivers | |
418 | * and must not be freed until after all the CPU's have gone | |
419 | * through a quiescent state. | |
420 | */ | |
421 | void __dev_remove_pack(struct packet_type *pt) | |
422 | { | |
423 | struct list_head *head = ptype_head(pt); | |
424 | struct packet_type *pt1; | |
425 | ||
426 | spin_lock(&ptype_lock); | |
427 | ||
428 | list_for_each_entry(pt1, head, list) { | |
429 | if (pt == pt1) { | |
430 | list_del_rcu(&pt->list); | |
431 | goto out; | |
432 | } | |
433 | } | |
434 | ||
435 | printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt); | |
436 | out: | |
437 | spin_unlock(&ptype_lock); | |
438 | } | |
439 | EXPORT_SYMBOL(__dev_remove_pack); | |
440 | ||
441 | /** | |
442 | * dev_remove_pack - remove packet handler | |
443 | * @pt: packet type declaration | |
444 | * | |
445 | * Remove a protocol handler that was previously added to the kernel | |
446 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
447 | * from the kernel lists and can be freed or reused once this function | |
448 | * returns. | |
449 | * | |
450 | * This call sleeps to guarantee that no CPU is looking at the packet | |
451 | * type after return. | |
452 | */ | |
453 | void dev_remove_pack(struct packet_type *pt) | |
454 | { | |
455 | __dev_remove_pack(pt); | |
456 | ||
457 | synchronize_net(); | |
458 | } | |
459 | EXPORT_SYMBOL(dev_remove_pack); | |
460 | ||
461 | /****************************************************************************** | |
462 | ||
463 | Device Boot-time Settings Routines | |
464 | ||
465 | *******************************************************************************/ | |
466 | ||
467 | /* Boot time configuration table */ | |
468 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
469 | ||
470 | /** | |
471 | * netdev_boot_setup_add - add new setup entry | |
472 | * @name: name of the device | |
473 | * @map: configured settings for the device | |
474 | * | |
475 | * Adds new setup entry to the dev_boot_setup list. The function | |
476 | * returns 0 on error and 1 on success. This is a generic routine to | |
477 | * all netdevices. | |
478 | */ | |
479 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
480 | { | |
481 | struct netdev_boot_setup *s; | |
482 | int i; | |
483 | ||
484 | s = dev_boot_setup; | |
485 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
486 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
487 | memset(s[i].name, 0, sizeof(s[i].name)); | |
488 | strlcpy(s[i].name, name, IFNAMSIZ); | |
489 | memcpy(&s[i].map, map, sizeof(s[i].map)); | |
490 | break; | |
491 | } | |
492 | } | |
493 | ||
494 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
495 | } | |
496 | ||
497 | /** | |
498 | * netdev_boot_setup_check - check boot time settings | |
499 | * @dev: the netdevice | |
500 | * | |
501 | * Check boot time settings for the device. | |
502 | * The found settings are set for the device to be used | |
503 | * later in the device probing. | |
504 | * Returns 0 if no settings found, 1 if they are. | |
505 | */ | |
506 | int netdev_boot_setup_check(struct net_device *dev) | |
507 | { | |
508 | struct netdev_boot_setup *s = dev_boot_setup; | |
509 | int i; | |
510 | ||
511 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
512 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
513 | !strcmp(dev->name, s[i].name)) { | |
514 | dev->irq = s[i].map.irq; | |
515 | dev->base_addr = s[i].map.base_addr; | |
516 | dev->mem_start = s[i].map.mem_start; | |
517 | dev->mem_end = s[i].map.mem_end; | |
518 | return 1; | |
519 | } | |
520 | } | |
521 | return 0; | |
522 | } | |
523 | EXPORT_SYMBOL(netdev_boot_setup_check); | |
524 | ||
525 | ||
526 | /** | |
527 | * netdev_boot_base - get address from boot time settings | |
528 | * @prefix: prefix for network device | |
529 | * @unit: id for network device | |
530 | * | |
531 | * Check boot time settings for the base address of device. | |
532 | * The found settings are set for the device to be used | |
533 | * later in the device probing. | |
534 | * Returns 0 if no settings found. | |
535 | */ | |
536 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
537 | { | |
538 | const struct netdev_boot_setup *s = dev_boot_setup; | |
539 | char name[IFNAMSIZ]; | |
540 | int i; | |
541 | ||
542 | sprintf(name, "%s%d", prefix, unit); | |
543 | ||
544 | /* | |
545 | * If device already registered then return base of 1 | |
546 | * to indicate not to probe for this interface | |
547 | */ | |
548 | if (__dev_get_by_name(&init_net, name)) | |
549 | return 1; | |
550 | ||
551 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
552 | if (!strcmp(name, s[i].name)) | |
553 | return s[i].map.base_addr; | |
554 | return 0; | |
555 | } | |
556 | ||
557 | /* | |
558 | * Saves at boot time configured settings for any netdevice. | |
559 | */ | |
560 | int __init netdev_boot_setup(char *str) | |
561 | { | |
562 | int ints[5]; | |
563 | struct ifmap map; | |
564 | ||
565 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
566 | if (!str || !*str) | |
567 | return 0; | |
568 | ||
569 | /* Save settings */ | |
570 | memset(&map, 0, sizeof(map)); | |
571 | if (ints[0] > 0) | |
572 | map.irq = ints[1]; | |
573 | if (ints[0] > 1) | |
574 | map.base_addr = ints[2]; | |
575 | if (ints[0] > 2) | |
576 | map.mem_start = ints[3]; | |
577 | if (ints[0] > 3) | |
578 | map.mem_end = ints[4]; | |
579 | ||
580 | /* Add new entry to the list */ | |
581 | return netdev_boot_setup_add(str, &map); | |
582 | } | |
583 | ||
584 | __setup("netdev=", netdev_boot_setup); | |
585 | ||
586 | /******************************************************************************* | |
587 | ||
588 | Device Interface Subroutines | |
589 | ||
590 | *******************************************************************************/ | |
591 | ||
592 | /** | |
593 | * __dev_get_by_name - find a device by its name | |
594 | * @net: the applicable net namespace | |
595 | * @name: name to find | |
596 | * | |
597 | * Find an interface by name. Must be called under RTNL semaphore | |
598 | * or @dev_base_lock. If the name is found a pointer to the device | |
599 | * is returned. If the name is not found then %NULL is returned. The | |
600 | * reference counters are not incremented so the caller must be | |
601 | * careful with locks. | |
602 | */ | |
603 | ||
604 | struct net_device *__dev_get_by_name(struct net *net, const char *name) | |
605 | { | |
606 | struct hlist_node *p; | |
607 | struct net_device *dev; | |
608 | struct hlist_head *head = dev_name_hash(net, name); | |
609 | ||
610 | hlist_for_each_entry(dev, p, head, name_hlist) | |
611 | if (!strncmp(dev->name, name, IFNAMSIZ)) | |
612 | return dev; | |
613 | ||
614 | return NULL; | |
615 | } | |
616 | EXPORT_SYMBOL(__dev_get_by_name); | |
617 | ||
618 | /** | |
619 | * dev_get_by_name_rcu - find a device by its name | |
620 | * @net: the applicable net namespace | |
621 | * @name: name to find | |
622 | * | |
623 | * Find an interface by name. | |
624 | * If the name is found a pointer to the device is returned. | |
625 | * If the name is not found then %NULL is returned. | |
626 | * The reference counters are not incremented so the caller must be | |
627 | * careful with locks. The caller must hold RCU lock. | |
628 | */ | |
629 | ||
630 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
631 | { | |
632 | struct hlist_node *p; | |
633 | struct net_device *dev; | |
634 | struct hlist_head *head = dev_name_hash(net, name); | |
635 | ||
636 | hlist_for_each_entry_rcu(dev, p, head, name_hlist) | |
637 | if (!strncmp(dev->name, name, IFNAMSIZ)) | |
638 | return dev; | |
639 | ||
640 | return NULL; | |
641 | } | |
642 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
643 | ||
644 | /** | |
645 | * dev_get_by_name - find a device by its name | |
646 | * @net: the applicable net namespace | |
647 | * @name: name to find | |
648 | * | |
649 | * Find an interface by name. This can be called from any | |
650 | * context and does its own locking. The returned handle has | |
651 | * the usage count incremented and the caller must use dev_put() to | |
652 | * release it when it is no longer needed. %NULL is returned if no | |
653 | * matching device is found. | |
654 | */ | |
655 | ||
656 | struct net_device *dev_get_by_name(struct net *net, const char *name) | |
657 | { | |
658 | struct net_device *dev; | |
659 | ||
660 | rcu_read_lock(); | |
661 | dev = dev_get_by_name_rcu(net, name); | |
662 | if (dev) | |
663 | dev_hold(dev); | |
664 | rcu_read_unlock(); | |
665 | return dev; | |
666 | } | |
667 | EXPORT_SYMBOL(dev_get_by_name); | |
668 | ||
669 | /** | |
670 | * __dev_get_by_index - find a device by its ifindex | |
671 | * @net: the applicable net namespace | |
672 | * @ifindex: index of device | |
673 | * | |
674 | * Search for an interface by index. Returns %NULL if the device | |
675 | * is not found or a pointer to the device. The device has not | |
676 | * had its reference counter increased so the caller must be careful | |
677 | * about locking. The caller must hold either the RTNL semaphore | |
678 | * or @dev_base_lock. | |
679 | */ | |
680 | ||
681 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) | |
682 | { | |
683 | struct hlist_node *p; | |
684 | struct net_device *dev; | |
685 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
686 | ||
687 | hlist_for_each_entry(dev, p, head, index_hlist) | |
688 | if (dev->ifindex == ifindex) | |
689 | return dev; | |
690 | ||
691 | return NULL; | |
692 | } | |
693 | EXPORT_SYMBOL(__dev_get_by_index); | |
694 | ||
695 | /** | |
696 | * dev_get_by_index_rcu - find a device by its ifindex | |
697 | * @net: the applicable net namespace | |
698 | * @ifindex: index of device | |
699 | * | |
700 | * Search for an interface by index. Returns %NULL if the device | |
701 | * is not found or a pointer to the device. The device has not | |
702 | * had its reference counter increased so the caller must be careful | |
703 | * about locking. The caller must hold RCU lock. | |
704 | */ | |
705 | ||
706 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
707 | { | |
708 | struct hlist_node *p; | |
709 | struct net_device *dev; | |
710 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
711 | ||
712 | hlist_for_each_entry_rcu(dev, p, head, index_hlist) | |
713 | if (dev->ifindex == ifindex) | |
714 | return dev; | |
715 | ||
716 | return NULL; | |
717 | } | |
718 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
719 | ||
720 | ||
721 | /** | |
722 | * dev_get_by_index - find a device by its ifindex | |
723 | * @net: the applicable net namespace | |
724 | * @ifindex: index of device | |
725 | * | |
726 | * Search for an interface by index. Returns NULL if the device | |
727 | * is not found or a pointer to the device. The device returned has | |
728 | * had a reference added and the pointer is safe until the user calls | |
729 | * dev_put to indicate they have finished with it. | |
730 | */ | |
731 | ||
732 | struct net_device *dev_get_by_index(struct net *net, int ifindex) | |
733 | { | |
734 | struct net_device *dev; | |
735 | ||
736 | rcu_read_lock(); | |
737 | dev = dev_get_by_index_rcu(net, ifindex); | |
738 | if (dev) | |
739 | dev_hold(dev); | |
740 | rcu_read_unlock(); | |
741 | return dev; | |
742 | } | |
743 | EXPORT_SYMBOL(dev_get_by_index); | |
744 | ||
745 | /** | |
746 | * dev_getbyhwaddr - find a device by its hardware address | |
747 | * @net: the applicable net namespace | |
748 | * @type: media type of device | |
749 | * @ha: hardware address | |
750 | * | |
751 | * Search for an interface by MAC address. Returns NULL if the device | |
752 | * is not found or a pointer to the device. The caller must hold the | |
753 | * rtnl semaphore. The returned device has not had its ref count increased | |
754 | * and the caller must therefore be careful about locking | |
755 | * | |
756 | * BUGS: | |
757 | * If the API was consistent this would be __dev_get_by_hwaddr | |
758 | */ | |
759 | ||
760 | struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha) | |
761 | { | |
762 | struct net_device *dev; | |
763 | ||
764 | ASSERT_RTNL(); | |
765 | ||
766 | for_each_netdev(net, dev) | |
767 | if (dev->type == type && | |
768 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
769 | return dev; | |
770 | ||
771 | return NULL; | |
772 | } | |
773 | EXPORT_SYMBOL(dev_getbyhwaddr); | |
774 | ||
775 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) | |
776 | { | |
777 | struct net_device *dev; | |
778 | ||
779 | ASSERT_RTNL(); | |
780 | for_each_netdev(net, dev) | |
781 | if (dev->type == type) | |
782 | return dev; | |
783 | ||
784 | return NULL; | |
785 | } | |
786 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); | |
787 | ||
788 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) | |
789 | { | |
790 | struct net_device *dev, *ret = NULL; | |
791 | ||
792 | rcu_read_lock(); | |
793 | for_each_netdev_rcu(net, dev) | |
794 | if (dev->type == type) { | |
795 | dev_hold(dev); | |
796 | ret = dev; | |
797 | break; | |
798 | } | |
799 | rcu_read_unlock(); | |
800 | return ret; | |
801 | } | |
802 | EXPORT_SYMBOL(dev_getfirstbyhwtype); | |
803 | ||
804 | /** | |
805 | * dev_get_by_flags_rcu - find any device with given flags | |
806 | * @net: the applicable net namespace | |
807 | * @if_flags: IFF_* values | |
808 | * @mask: bitmask of bits in if_flags to check | |
809 | * | |
810 | * Search for any interface with the given flags. Returns NULL if a device | |
811 | * is not found or a pointer to the device. Must be called inside | |
812 | * rcu_read_lock(), and result refcount is unchanged. | |
813 | */ | |
814 | ||
815 | struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags, | |
816 | unsigned short mask) | |
817 | { | |
818 | struct net_device *dev, *ret; | |
819 | ||
820 | ret = NULL; | |
821 | for_each_netdev_rcu(net, dev) { | |
822 | if (((dev->flags ^ if_flags) & mask) == 0) { | |
823 | ret = dev; | |
824 | break; | |
825 | } | |
826 | } | |
827 | return ret; | |
828 | } | |
829 | EXPORT_SYMBOL(dev_get_by_flags_rcu); | |
830 | ||
831 | /** | |
832 | * dev_valid_name - check if name is okay for network device | |
833 | * @name: name string | |
834 | * | |
835 | * Network device names need to be valid file names to | |
836 | * to allow sysfs to work. We also disallow any kind of | |
837 | * whitespace. | |
838 | */ | |
839 | int dev_valid_name(const char *name) | |
840 | { | |
841 | if (*name == '\0') | |
842 | return 0; | |
843 | if (strlen(name) >= IFNAMSIZ) | |
844 | return 0; | |
845 | if (!strcmp(name, ".") || !strcmp(name, "..")) | |
846 | return 0; | |
847 | ||
848 | while (*name) { | |
849 | if (*name == '/' || isspace(*name)) | |
850 | return 0; | |
851 | name++; | |
852 | } | |
853 | return 1; | |
854 | } | |
855 | EXPORT_SYMBOL(dev_valid_name); | |
856 | ||
857 | /** | |
858 | * __dev_alloc_name - allocate a name for a device | |
859 | * @net: network namespace to allocate the device name in | |
860 | * @name: name format string | |
861 | * @buf: scratch buffer and result name string | |
862 | * | |
863 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
864 | * id. It scans list of devices to build up a free map, then chooses | |
865 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
866 | * while allocating the name and adding the device in order to avoid | |
867 | * duplicates. | |
868 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
869 | * Returns the number of the unit assigned or a negative errno code. | |
870 | */ | |
871 | ||
872 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) | |
873 | { | |
874 | int i = 0; | |
875 | const char *p; | |
876 | const int max_netdevices = 8*PAGE_SIZE; | |
877 | unsigned long *inuse; | |
878 | struct net_device *d; | |
879 | ||
880 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
881 | if (p) { | |
882 | /* | |
883 | * Verify the string as this thing may have come from | |
884 | * the user. There must be either one "%d" and no other "%" | |
885 | * characters. | |
886 | */ | |
887 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
888 | return -EINVAL; | |
889 | ||
890 | /* Use one page as a bit array of possible slots */ | |
891 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); | |
892 | if (!inuse) | |
893 | return -ENOMEM; | |
894 | ||
895 | for_each_netdev(net, d) { | |
896 | if (!sscanf(d->name, name, &i)) | |
897 | continue; | |
898 | if (i < 0 || i >= max_netdevices) | |
899 | continue; | |
900 | ||
901 | /* avoid cases where sscanf is not exact inverse of printf */ | |
902 | snprintf(buf, IFNAMSIZ, name, i); | |
903 | if (!strncmp(buf, d->name, IFNAMSIZ)) | |
904 | set_bit(i, inuse); | |
905 | } | |
906 | ||
907 | i = find_first_zero_bit(inuse, max_netdevices); | |
908 | free_page((unsigned long) inuse); | |
909 | } | |
910 | ||
911 | if (buf != name) | |
912 | snprintf(buf, IFNAMSIZ, name, i); | |
913 | if (!__dev_get_by_name(net, buf)) | |
914 | return i; | |
915 | ||
916 | /* It is possible to run out of possible slots | |
917 | * when the name is long and there isn't enough space left | |
918 | * for the digits, or if all bits are used. | |
919 | */ | |
920 | return -ENFILE; | |
921 | } | |
922 | ||
923 | /** | |
924 | * dev_alloc_name - allocate a name for a device | |
925 | * @dev: device | |
926 | * @name: name format string | |
927 | * | |
928 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
929 | * id. It scans list of devices to build up a free map, then chooses | |
930 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
931 | * while allocating the name and adding the device in order to avoid | |
932 | * duplicates. | |
933 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
934 | * Returns the number of the unit assigned or a negative errno code. | |
935 | */ | |
936 | ||
937 | int dev_alloc_name(struct net_device *dev, const char *name) | |
938 | { | |
939 | char buf[IFNAMSIZ]; | |
940 | struct net *net; | |
941 | int ret; | |
942 | ||
943 | BUG_ON(!dev_net(dev)); | |
944 | net = dev_net(dev); | |
945 | ret = __dev_alloc_name(net, name, buf); | |
946 | if (ret >= 0) | |
947 | strlcpy(dev->name, buf, IFNAMSIZ); | |
948 | return ret; | |
949 | } | |
950 | EXPORT_SYMBOL(dev_alloc_name); | |
951 | ||
952 | static int dev_get_valid_name(struct net_device *dev, const char *name, bool fmt) | |
953 | { | |
954 | struct net *net; | |
955 | ||
956 | BUG_ON(!dev_net(dev)); | |
957 | net = dev_net(dev); | |
958 | ||
959 | if (!dev_valid_name(name)) | |
960 | return -EINVAL; | |
961 | ||
962 | if (fmt && strchr(name, '%')) | |
963 | return dev_alloc_name(dev, name); | |
964 | else if (__dev_get_by_name(net, name)) | |
965 | return -EEXIST; | |
966 | else if (dev->name != name) | |
967 | strlcpy(dev->name, name, IFNAMSIZ); | |
968 | ||
969 | return 0; | |
970 | } | |
971 | ||
972 | /** | |
973 | * dev_change_name - change name of a device | |
974 | * @dev: device | |
975 | * @newname: name (or format string) must be at least IFNAMSIZ | |
976 | * | |
977 | * Change name of a device, can pass format strings "eth%d". | |
978 | * for wildcarding. | |
979 | */ | |
980 | int dev_change_name(struct net_device *dev, const char *newname) | |
981 | { | |
982 | char oldname[IFNAMSIZ]; | |
983 | int err = 0; | |
984 | int ret; | |
985 | struct net *net; | |
986 | ||
987 | ASSERT_RTNL(); | |
988 | BUG_ON(!dev_net(dev)); | |
989 | ||
990 | net = dev_net(dev); | |
991 | if (dev->flags & IFF_UP) | |
992 | return -EBUSY; | |
993 | ||
994 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) | |
995 | return 0; | |
996 | ||
997 | memcpy(oldname, dev->name, IFNAMSIZ); | |
998 | ||
999 | err = dev_get_valid_name(dev, newname, 1); | |
1000 | if (err < 0) | |
1001 | return err; | |
1002 | ||
1003 | rollback: | |
1004 | ret = device_rename(&dev->dev, dev->name); | |
1005 | if (ret) { | |
1006 | memcpy(dev->name, oldname, IFNAMSIZ); | |
1007 | return ret; | |
1008 | } | |
1009 | ||
1010 | write_lock_bh(&dev_base_lock); | |
1011 | hlist_del(&dev->name_hlist); | |
1012 | write_unlock_bh(&dev_base_lock); | |
1013 | ||
1014 | synchronize_rcu(); | |
1015 | ||
1016 | write_lock_bh(&dev_base_lock); | |
1017 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
1018 | write_unlock_bh(&dev_base_lock); | |
1019 | ||
1020 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); | |
1021 | ret = notifier_to_errno(ret); | |
1022 | ||
1023 | if (ret) { | |
1024 | /* err >= 0 after dev_alloc_name() or stores the first errno */ | |
1025 | if (err >= 0) { | |
1026 | err = ret; | |
1027 | memcpy(dev->name, oldname, IFNAMSIZ); | |
1028 | goto rollback; | |
1029 | } else { | |
1030 | printk(KERN_ERR | |
1031 | "%s: name change rollback failed: %d.\n", | |
1032 | dev->name, ret); | |
1033 | } | |
1034 | } | |
1035 | ||
1036 | return err; | |
1037 | } | |
1038 | ||
1039 | /** | |
1040 | * dev_set_alias - change ifalias of a device | |
1041 | * @dev: device | |
1042 | * @alias: name up to IFALIASZ | |
1043 | * @len: limit of bytes to copy from info | |
1044 | * | |
1045 | * Set ifalias for a device, | |
1046 | */ | |
1047 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1048 | { | |
1049 | ASSERT_RTNL(); | |
1050 | ||
1051 | if (len >= IFALIASZ) | |
1052 | return -EINVAL; | |
1053 | ||
1054 | if (!len) { | |
1055 | if (dev->ifalias) { | |
1056 | kfree(dev->ifalias); | |
1057 | dev->ifalias = NULL; | |
1058 | } | |
1059 | return 0; | |
1060 | } | |
1061 | ||
1062 | dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL); | |
1063 | if (!dev->ifalias) | |
1064 | return -ENOMEM; | |
1065 | ||
1066 | strlcpy(dev->ifalias, alias, len+1); | |
1067 | return len; | |
1068 | } | |
1069 | ||
1070 | ||
1071 | /** | |
1072 | * netdev_features_change - device changes features | |
1073 | * @dev: device to cause notification | |
1074 | * | |
1075 | * Called to indicate a device has changed features. | |
1076 | */ | |
1077 | void netdev_features_change(struct net_device *dev) | |
1078 | { | |
1079 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); | |
1080 | } | |
1081 | EXPORT_SYMBOL(netdev_features_change); | |
1082 | ||
1083 | /** | |
1084 | * netdev_state_change - device changes state | |
1085 | * @dev: device to cause notification | |
1086 | * | |
1087 | * Called to indicate a device has changed state. This function calls | |
1088 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1089 | * to the routing socket. | |
1090 | */ | |
1091 | void netdev_state_change(struct net_device *dev) | |
1092 | { | |
1093 | if (dev->flags & IFF_UP) { | |
1094 | call_netdevice_notifiers(NETDEV_CHANGE, dev); | |
1095 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0); | |
1096 | } | |
1097 | } | |
1098 | EXPORT_SYMBOL(netdev_state_change); | |
1099 | ||
1100 | int netdev_bonding_change(struct net_device *dev, unsigned long event) | |
1101 | { | |
1102 | return call_netdevice_notifiers(event, dev); | |
1103 | } | |
1104 | EXPORT_SYMBOL(netdev_bonding_change); | |
1105 | ||
1106 | /** | |
1107 | * dev_load - load a network module | |
1108 | * @net: the applicable net namespace | |
1109 | * @name: name of interface | |
1110 | * | |
1111 | * If a network interface is not present and the process has suitable | |
1112 | * privileges this function loads the module. If module loading is not | |
1113 | * available in this kernel then it becomes a nop. | |
1114 | */ | |
1115 | ||
1116 | void dev_load(struct net *net, const char *name) | |
1117 | { | |
1118 | struct net_device *dev; | |
1119 | ||
1120 | rcu_read_lock(); | |
1121 | dev = dev_get_by_name_rcu(net, name); | |
1122 | rcu_read_unlock(); | |
1123 | ||
1124 | if (!dev && capable(CAP_NET_ADMIN)) | |
1125 | request_module("%s", name); | |
1126 | } | |
1127 | EXPORT_SYMBOL(dev_load); | |
1128 | ||
1129 | static int __dev_open(struct net_device *dev) | |
1130 | { | |
1131 | const struct net_device_ops *ops = dev->netdev_ops; | |
1132 | int ret; | |
1133 | ||
1134 | ASSERT_RTNL(); | |
1135 | ||
1136 | /* | |
1137 | * Is it even present? | |
1138 | */ | |
1139 | if (!netif_device_present(dev)) | |
1140 | return -ENODEV; | |
1141 | ||
1142 | ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); | |
1143 | ret = notifier_to_errno(ret); | |
1144 | if (ret) | |
1145 | return ret; | |
1146 | ||
1147 | /* | |
1148 | * Call device private open method | |
1149 | */ | |
1150 | set_bit(__LINK_STATE_START, &dev->state); | |
1151 | ||
1152 | if (ops->ndo_validate_addr) | |
1153 | ret = ops->ndo_validate_addr(dev); | |
1154 | ||
1155 | if (!ret && ops->ndo_open) | |
1156 | ret = ops->ndo_open(dev); | |
1157 | ||
1158 | /* | |
1159 | * If it went open OK then: | |
1160 | */ | |
1161 | ||
1162 | if (ret) | |
1163 | clear_bit(__LINK_STATE_START, &dev->state); | |
1164 | else { | |
1165 | /* | |
1166 | * Set the flags. | |
1167 | */ | |
1168 | dev->flags |= IFF_UP; | |
1169 | ||
1170 | /* | |
1171 | * Enable NET_DMA | |
1172 | */ | |
1173 | net_dmaengine_get(); | |
1174 | ||
1175 | /* | |
1176 | * Initialize multicasting status | |
1177 | */ | |
1178 | dev_set_rx_mode(dev); | |
1179 | ||
1180 | /* | |
1181 | * Wakeup transmit queue engine | |
1182 | */ | |
1183 | dev_activate(dev); | |
1184 | } | |
1185 | ||
1186 | return ret; | |
1187 | } | |
1188 | ||
1189 | /** | |
1190 | * dev_open - prepare an interface for use. | |
1191 | * @dev: device to open | |
1192 | * | |
1193 | * Takes a device from down to up state. The device's private open | |
1194 | * function is invoked and then the multicast lists are loaded. Finally | |
1195 | * the device is moved into the up state and a %NETDEV_UP message is | |
1196 | * sent to the netdev notifier chain. | |
1197 | * | |
1198 | * Calling this function on an active interface is a nop. On a failure | |
1199 | * a negative errno code is returned. | |
1200 | */ | |
1201 | int dev_open(struct net_device *dev) | |
1202 | { | |
1203 | int ret; | |
1204 | ||
1205 | /* | |
1206 | * Is it already up? | |
1207 | */ | |
1208 | if (dev->flags & IFF_UP) | |
1209 | return 0; | |
1210 | ||
1211 | /* | |
1212 | * Open device | |
1213 | */ | |
1214 | ret = __dev_open(dev); | |
1215 | if (ret < 0) | |
1216 | return ret; | |
1217 | ||
1218 | /* | |
1219 | * ... and announce new interface. | |
1220 | */ | |
1221 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING); | |
1222 | call_netdevice_notifiers(NETDEV_UP, dev); | |
1223 | ||
1224 | return ret; | |
1225 | } | |
1226 | EXPORT_SYMBOL(dev_open); | |
1227 | ||
1228 | static int __dev_close(struct net_device *dev) | |
1229 | { | |
1230 | const struct net_device_ops *ops = dev->netdev_ops; | |
1231 | ||
1232 | ASSERT_RTNL(); | |
1233 | might_sleep(); | |
1234 | ||
1235 | /* | |
1236 | * Tell people we are going down, so that they can | |
1237 | * prepare to death, when device is still operating. | |
1238 | */ | |
1239 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); | |
1240 | ||
1241 | clear_bit(__LINK_STATE_START, &dev->state); | |
1242 | ||
1243 | /* Synchronize to scheduled poll. We cannot touch poll list, | |
1244 | * it can be even on different cpu. So just clear netif_running(). | |
1245 | * | |
1246 | * dev->stop() will invoke napi_disable() on all of it's | |
1247 | * napi_struct instances on this device. | |
1248 | */ | |
1249 | smp_mb__after_clear_bit(); /* Commit netif_running(). */ | |
1250 | ||
1251 | dev_deactivate(dev); | |
1252 | ||
1253 | /* | |
1254 | * Call the device specific close. This cannot fail. | |
1255 | * Only if device is UP | |
1256 | * | |
1257 | * We allow it to be called even after a DETACH hot-plug | |
1258 | * event. | |
1259 | */ | |
1260 | if (ops->ndo_stop) | |
1261 | ops->ndo_stop(dev); | |
1262 | ||
1263 | /* | |
1264 | * Device is now down. | |
1265 | */ | |
1266 | ||
1267 | dev->flags &= ~IFF_UP; | |
1268 | ||
1269 | /* | |
1270 | * Shutdown NET_DMA | |
1271 | */ | |
1272 | net_dmaengine_put(); | |
1273 | ||
1274 | return 0; | |
1275 | } | |
1276 | ||
1277 | /** | |
1278 | * dev_close - shutdown an interface. | |
1279 | * @dev: device to shutdown | |
1280 | * | |
1281 | * This function moves an active device into down state. A | |
1282 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1283 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1284 | * chain. | |
1285 | */ | |
1286 | int dev_close(struct net_device *dev) | |
1287 | { | |
1288 | if (!(dev->flags & IFF_UP)) | |
1289 | return 0; | |
1290 | ||
1291 | __dev_close(dev); | |
1292 | ||
1293 | /* | |
1294 | * Tell people we are down | |
1295 | */ | |
1296 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING); | |
1297 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
1298 | ||
1299 | return 0; | |
1300 | } | |
1301 | EXPORT_SYMBOL(dev_close); | |
1302 | ||
1303 | ||
1304 | /** | |
1305 | * dev_disable_lro - disable Large Receive Offload on a device | |
1306 | * @dev: device | |
1307 | * | |
1308 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1309 | * called under RTNL. This is needed if received packets may be | |
1310 | * forwarded to another interface. | |
1311 | */ | |
1312 | void dev_disable_lro(struct net_device *dev) | |
1313 | { | |
1314 | if (dev->ethtool_ops && dev->ethtool_ops->get_flags && | |
1315 | dev->ethtool_ops->set_flags) { | |
1316 | u32 flags = dev->ethtool_ops->get_flags(dev); | |
1317 | if (flags & ETH_FLAG_LRO) { | |
1318 | flags &= ~ETH_FLAG_LRO; | |
1319 | dev->ethtool_ops->set_flags(dev, flags); | |
1320 | } | |
1321 | } | |
1322 | WARN_ON(dev->features & NETIF_F_LRO); | |
1323 | } | |
1324 | EXPORT_SYMBOL(dev_disable_lro); | |
1325 | ||
1326 | ||
1327 | static int dev_boot_phase = 1; | |
1328 | ||
1329 | /* | |
1330 | * Device change register/unregister. These are not inline or static | |
1331 | * as we export them to the world. | |
1332 | */ | |
1333 | ||
1334 | /** | |
1335 | * register_netdevice_notifier - register a network notifier block | |
1336 | * @nb: notifier | |
1337 | * | |
1338 | * Register a notifier to be called when network device events occur. | |
1339 | * The notifier passed is linked into the kernel structures and must | |
1340 | * not be reused until it has been unregistered. A negative errno code | |
1341 | * is returned on a failure. | |
1342 | * | |
1343 | * When registered all registration and up events are replayed | |
1344 | * to the new notifier to allow device to have a race free | |
1345 | * view of the network device list. | |
1346 | */ | |
1347 | ||
1348 | int register_netdevice_notifier(struct notifier_block *nb) | |
1349 | { | |
1350 | struct net_device *dev; | |
1351 | struct net_device *last; | |
1352 | struct net *net; | |
1353 | int err; | |
1354 | ||
1355 | rtnl_lock(); | |
1356 | err = raw_notifier_chain_register(&netdev_chain, nb); | |
1357 | if (err) | |
1358 | goto unlock; | |
1359 | if (dev_boot_phase) | |
1360 | goto unlock; | |
1361 | for_each_net(net) { | |
1362 | for_each_netdev(net, dev) { | |
1363 | err = nb->notifier_call(nb, NETDEV_REGISTER, dev); | |
1364 | err = notifier_to_errno(err); | |
1365 | if (err) | |
1366 | goto rollback; | |
1367 | ||
1368 | if (!(dev->flags & IFF_UP)) | |
1369 | continue; | |
1370 | ||
1371 | nb->notifier_call(nb, NETDEV_UP, dev); | |
1372 | } | |
1373 | } | |
1374 | ||
1375 | unlock: | |
1376 | rtnl_unlock(); | |
1377 | return err; | |
1378 | ||
1379 | rollback: | |
1380 | last = dev; | |
1381 | for_each_net(net) { | |
1382 | for_each_netdev(net, dev) { | |
1383 | if (dev == last) | |
1384 | break; | |
1385 | ||
1386 | if (dev->flags & IFF_UP) { | |
1387 | nb->notifier_call(nb, NETDEV_GOING_DOWN, dev); | |
1388 | nb->notifier_call(nb, NETDEV_DOWN, dev); | |
1389 | } | |
1390 | nb->notifier_call(nb, NETDEV_UNREGISTER, dev); | |
1391 | nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev); | |
1392 | } | |
1393 | } | |
1394 | ||
1395 | raw_notifier_chain_unregister(&netdev_chain, nb); | |
1396 | goto unlock; | |
1397 | } | |
1398 | EXPORT_SYMBOL(register_netdevice_notifier); | |
1399 | ||
1400 | /** | |
1401 | * unregister_netdevice_notifier - unregister a network notifier block | |
1402 | * @nb: notifier | |
1403 | * | |
1404 | * Unregister a notifier previously registered by | |
1405 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1406 | * kernel structures and may then be reused. A negative errno code | |
1407 | * is returned on a failure. | |
1408 | */ | |
1409 | ||
1410 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1411 | { | |
1412 | int err; | |
1413 | ||
1414 | rtnl_lock(); | |
1415 | err = raw_notifier_chain_unregister(&netdev_chain, nb); | |
1416 | rtnl_unlock(); | |
1417 | return err; | |
1418 | } | |
1419 | EXPORT_SYMBOL(unregister_netdevice_notifier); | |
1420 | ||
1421 | /** | |
1422 | * call_netdevice_notifiers - call all network notifier blocks | |
1423 | * @val: value passed unmodified to notifier function | |
1424 | * @dev: net_device pointer passed unmodified to notifier function | |
1425 | * | |
1426 | * Call all network notifier blocks. Parameters and return value | |
1427 | * are as for raw_notifier_call_chain(). | |
1428 | */ | |
1429 | ||
1430 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) | |
1431 | { | |
1432 | ASSERT_RTNL(); | |
1433 | return raw_notifier_call_chain(&netdev_chain, val, dev); | |
1434 | } | |
1435 | ||
1436 | /* When > 0 there are consumers of rx skb time stamps */ | |
1437 | static atomic_t netstamp_needed = ATOMIC_INIT(0); | |
1438 | ||
1439 | void net_enable_timestamp(void) | |
1440 | { | |
1441 | atomic_inc(&netstamp_needed); | |
1442 | } | |
1443 | EXPORT_SYMBOL(net_enable_timestamp); | |
1444 | ||
1445 | void net_disable_timestamp(void) | |
1446 | { | |
1447 | atomic_dec(&netstamp_needed); | |
1448 | } | |
1449 | EXPORT_SYMBOL(net_disable_timestamp); | |
1450 | ||
1451 | static inline void net_timestamp_set(struct sk_buff *skb) | |
1452 | { | |
1453 | if (atomic_read(&netstamp_needed)) | |
1454 | __net_timestamp(skb); | |
1455 | else | |
1456 | skb->tstamp.tv64 = 0; | |
1457 | } | |
1458 | ||
1459 | static inline void net_timestamp_check(struct sk_buff *skb) | |
1460 | { | |
1461 | if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed)) | |
1462 | __net_timestamp(skb); | |
1463 | } | |
1464 | ||
1465 | /** | |
1466 | * dev_forward_skb - loopback an skb to another netif | |
1467 | * | |
1468 | * @dev: destination network device | |
1469 | * @skb: buffer to forward | |
1470 | * | |
1471 | * return values: | |
1472 | * NET_RX_SUCCESS (no congestion) | |
1473 | * NET_RX_DROP (packet was dropped, but freed) | |
1474 | * | |
1475 | * dev_forward_skb can be used for injecting an skb from the | |
1476 | * start_xmit function of one device into the receive queue | |
1477 | * of another device. | |
1478 | * | |
1479 | * The receiving device may be in another namespace, so | |
1480 | * we have to clear all information in the skb that could | |
1481 | * impact namespace isolation. | |
1482 | */ | |
1483 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
1484 | { | |
1485 | skb_orphan(skb); | |
1486 | nf_reset(skb); | |
1487 | ||
1488 | if (unlikely(!(dev->flags & IFF_UP) || | |
1489 | (skb->len > (dev->mtu + dev->hard_header_len + VLAN_HLEN)))) { | |
1490 | atomic_long_inc(&dev->rx_dropped); | |
1491 | kfree_skb(skb); | |
1492 | return NET_RX_DROP; | |
1493 | } | |
1494 | skb_set_dev(skb, dev); | |
1495 | skb->tstamp.tv64 = 0; | |
1496 | skb->pkt_type = PACKET_HOST; | |
1497 | skb->protocol = eth_type_trans(skb, dev); | |
1498 | return netif_rx(skb); | |
1499 | } | |
1500 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
1501 | ||
1502 | /* | |
1503 | * Support routine. Sends outgoing frames to any network | |
1504 | * taps currently in use. | |
1505 | */ | |
1506 | ||
1507 | static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) | |
1508 | { | |
1509 | struct packet_type *ptype; | |
1510 | ||
1511 | #ifdef CONFIG_NET_CLS_ACT | |
1512 | if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS))) | |
1513 | net_timestamp_set(skb); | |
1514 | #else | |
1515 | net_timestamp_set(skb); | |
1516 | #endif | |
1517 | ||
1518 | rcu_read_lock(); | |
1519 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
1520 | /* Never send packets back to the socket | |
1521 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1522 | */ | |
1523 | if ((ptype->dev == dev || !ptype->dev) && | |
1524 | (ptype->af_packet_priv == NULL || | |
1525 | (struct sock *)ptype->af_packet_priv != skb->sk)) { | |
1526 | struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC); | |
1527 | if (!skb2) | |
1528 | break; | |
1529 | ||
1530 | /* skb->nh should be correctly | |
1531 | set by sender, so that the second statement is | |
1532 | just protection against buggy protocols. | |
1533 | */ | |
1534 | skb_reset_mac_header(skb2); | |
1535 | ||
1536 | if (skb_network_header(skb2) < skb2->data || | |
1537 | skb2->network_header > skb2->tail) { | |
1538 | if (net_ratelimit()) | |
1539 | printk(KERN_CRIT "protocol %04x is " | |
1540 | "buggy, dev %s\n", | |
1541 | ntohs(skb2->protocol), | |
1542 | dev->name); | |
1543 | skb_reset_network_header(skb2); | |
1544 | } | |
1545 | ||
1546 | skb2->transport_header = skb2->network_header; | |
1547 | skb2->pkt_type = PACKET_OUTGOING; | |
1548 | ptype->func(skb2, skb->dev, ptype, skb->dev); | |
1549 | } | |
1550 | } | |
1551 | rcu_read_unlock(); | |
1552 | } | |
1553 | ||
1554 | /* | |
1555 | * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues | |
1556 | * greater then real_num_tx_queues stale skbs on the qdisc must be flushed. | |
1557 | */ | |
1558 | int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) | |
1559 | { | |
1560 | if (txq < 1 || txq > dev->num_tx_queues) | |
1561 | return -EINVAL; | |
1562 | ||
1563 | if (dev->reg_state == NETREG_REGISTERED) { | |
1564 | ASSERT_RTNL(); | |
1565 | ||
1566 | if (txq < dev->real_num_tx_queues) | |
1567 | qdisc_reset_all_tx_gt(dev, txq); | |
1568 | } | |
1569 | ||
1570 | dev->real_num_tx_queues = txq; | |
1571 | return 0; | |
1572 | } | |
1573 | EXPORT_SYMBOL(netif_set_real_num_tx_queues); | |
1574 | ||
1575 | #ifdef CONFIG_RPS | |
1576 | /** | |
1577 | * netif_set_real_num_rx_queues - set actual number of RX queues used | |
1578 | * @dev: Network device | |
1579 | * @rxq: Actual number of RX queues | |
1580 | * | |
1581 | * This must be called either with the rtnl_lock held or before | |
1582 | * registration of the net device. Returns 0 on success, or a | |
1583 | * negative error code. If called before registration, it always | |
1584 | * succeeds. | |
1585 | */ | |
1586 | int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) | |
1587 | { | |
1588 | int rc; | |
1589 | ||
1590 | if (rxq < 1 || rxq > dev->num_rx_queues) | |
1591 | return -EINVAL; | |
1592 | ||
1593 | if (dev->reg_state == NETREG_REGISTERED) { | |
1594 | ASSERT_RTNL(); | |
1595 | ||
1596 | rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, | |
1597 | rxq); | |
1598 | if (rc) | |
1599 | return rc; | |
1600 | } | |
1601 | ||
1602 | dev->real_num_rx_queues = rxq; | |
1603 | return 0; | |
1604 | } | |
1605 | EXPORT_SYMBOL(netif_set_real_num_rx_queues); | |
1606 | #endif | |
1607 | ||
1608 | static inline void __netif_reschedule(struct Qdisc *q) | |
1609 | { | |
1610 | struct softnet_data *sd; | |
1611 | unsigned long flags; | |
1612 | ||
1613 | local_irq_save(flags); | |
1614 | sd = &__get_cpu_var(softnet_data); | |
1615 | q->next_sched = NULL; | |
1616 | *sd->output_queue_tailp = q; | |
1617 | sd->output_queue_tailp = &q->next_sched; | |
1618 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
1619 | local_irq_restore(flags); | |
1620 | } | |
1621 | ||
1622 | void __netif_schedule(struct Qdisc *q) | |
1623 | { | |
1624 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
1625 | __netif_reschedule(q); | |
1626 | } | |
1627 | EXPORT_SYMBOL(__netif_schedule); | |
1628 | ||
1629 | void dev_kfree_skb_irq(struct sk_buff *skb) | |
1630 | { | |
1631 | if (atomic_dec_and_test(&skb->users)) { | |
1632 | struct softnet_data *sd; | |
1633 | unsigned long flags; | |
1634 | ||
1635 | local_irq_save(flags); | |
1636 | sd = &__get_cpu_var(softnet_data); | |
1637 | skb->next = sd->completion_queue; | |
1638 | sd->completion_queue = skb; | |
1639 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
1640 | local_irq_restore(flags); | |
1641 | } | |
1642 | } | |
1643 | EXPORT_SYMBOL(dev_kfree_skb_irq); | |
1644 | ||
1645 | void dev_kfree_skb_any(struct sk_buff *skb) | |
1646 | { | |
1647 | if (in_irq() || irqs_disabled()) | |
1648 | dev_kfree_skb_irq(skb); | |
1649 | else | |
1650 | dev_kfree_skb(skb); | |
1651 | } | |
1652 | EXPORT_SYMBOL(dev_kfree_skb_any); | |
1653 | ||
1654 | ||
1655 | /** | |
1656 | * netif_device_detach - mark device as removed | |
1657 | * @dev: network device | |
1658 | * | |
1659 | * Mark device as removed from system and therefore no longer available. | |
1660 | */ | |
1661 | void netif_device_detach(struct net_device *dev) | |
1662 | { | |
1663 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
1664 | netif_running(dev)) { | |
1665 | netif_tx_stop_all_queues(dev); | |
1666 | } | |
1667 | } | |
1668 | EXPORT_SYMBOL(netif_device_detach); | |
1669 | ||
1670 | /** | |
1671 | * netif_device_attach - mark device as attached | |
1672 | * @dev: network device | |
1673 | * | |
1674 | * Mark device as attached from system and restart if needed. | |
1675 | */ | |
1676 | void netif_device_attach(struct net_device *dev) | |
1677 | { | |
1678 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
1679 | netif_running(dev)) { | |
1680 | netif_tx_wake_all_queues(dev); | |
1681 | __netdev_watchdog_up(dev); | |
1682 | } | |
1683 | } | |
1684 | EXPORT_SYMBOL(netif_device_attach); | |
1685 | ||
1686 | static bool can_checksum_protocol(unsigned long features, __be16 protocol) | |
1687 | { | |
1688 | return ((features & NETIF_F_NO_CSUM) || | |
1689 | ((features & NETIF_F_V4_CSUM) && | |
1690 | protocol == htons(ETH_P_IP)) || | |
1691 | ((features & NETIF_F_V6_CSUM) && | |
1692 | protocol == htons(ETH_P_IPV6)) || | |
1693 | ((features & NETIF_F_FCOE_CRC) && | |
1694 | protocol == htons(ETH_P_FCOE))); | |
1695 | } | |
1696 | ||
1697 | static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb) | |
1698 | { | |
1699 | __be16 protocol = skb->protocol; | |
1700 | int features = dev->features; | |
1701 | ||
1702 | if (vlan_tx_tag_present(skb)) { | |
1703 | features &= dev->vlan_features; | |
1704 | } else if (protocol == htons(ETH_P_8021Q)) { | |
1705 | struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data; | |
1706 | protocol = veh->h_vlan_encapsulated_proto; | |
1707 | features &= dev->vlan_features; | |
1708 | } | |
1709 | ||
1710 | return can_checksum_protocol(features, protocol); | |
1711 | } | |
1712 | ||
1713 | /** | |
1714 | * skb_dev_set -- assign a new device to a buffer | |
1715 | * @skb: buffer for the new device | |
1716 | * @dev: network device | |
1717 | * | |
1718 | * If an skb is owned by a device already, we have to reset | |
1719 | * all data private to the namespace a device belongs to | |
1720 | * before assigning it a new device. | |
1721 | */ | |
1722 | #ifdef CONFIG_NET_NS | |
1723 | void skb_set_dev(struct sk_buff *skb, struct net_device *dev) | |
1724 | { | |
1725 | skb_dst_drop(skb); | |
1726 | if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) { | |
1727 | secpath_reset(skb); | |
1728 | nf_reset(skb); | |
1729 | skb_init_secmark(skb); | |
1730 | skb->mark = 0; | |
1731 | skb->priority = 0; | |
1732 | skb->nf_trace = 0; | |
1733 | skb->ipvs_property = 0; | |
1734 | #ifdef CONFIG_NET_SCHED | |
1735 | skb->tc_index = 0; | |
1736 | #endif | |
1737 | } | |
1738 | skb->dev = dev; | |
1739 | } | |
1740 | EXPORT_SYMBOL(skb_set_dev); | |
1741 | #endif /* CONFIG_NET_NS */ | |
1742 | ||
1743 | /* | |
1744 | * Invalidate hardware checksum when packet is to be mangled, and | |
1745 | * complete checksum manually on outgoing path. | |
1746 | */ | |
1747 | int skb_checksum_help(struct sk_buff *skb) | |
1748 | { | |
1749 | __wsum csum; | |
1750 | int ret = 0, offset; | |
1751 | ||
1752 | if (skb->ip_summed == CHECKSUM_COMPLETE) | |
1753 | goto out_set_summed; | |
1754 | ||
1755 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
1756 | /* Let GSO fix up the checksum. */ | |
1757 | goto out_set_summed; | |
1758 | } | |
1759 | ||
1760 | offset = skb->csum_start - skb_headroom(skb); | |
1761 | BUG_ON(offset >= skb_headlen(skb)); | |
1762 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
1763 | ||
1764 | offset += skb->csum_offset; | |
1765 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
1766 | ||
1767 | if (skb_cloned(skb) && | |
1768 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1769 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); | |
1770 | if (ret) | |
1771 | goto out; | |
1772 | } | |
1773 | ||
1774 | *(__sum16 *)(skb->data + offset) = csum_fold(csum); | |
1775 | out_set_summed: | |
1776 | skb->ip_summed = CHECKSUM_NONE; | |
1777 | out: | |
1778 | return ret; | |
1779 | } | |
1780 | EXPORT_SYMBOL(skb_checksum_help); | |
1781 | ||
1782 | /** | |
1783 | * skb_gso_segment - Perform segmentation on skb. | |
1784 | * @skb: buffer to segment | |
1785 | * @features: features for the output path (see dev->features) | |
1786 | * | |
1787 | * This function segments the given skb and returns a list of segments. | |
1788 | * | |
1789 | * It may return NULL if the skb requires no segmentation. This is | |
1790 | * only possible when GSO is used for verifying header integrity. | |
1791 | */ | |
1792 | struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features) | |
1793 | { | |
1794 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
1795 | struct packet_type *ptype; | |
1796 | __be16 type = skb->protocol; | |
1797 | int err; | |
1798 | ||
1799 | if (type == htons(ETH_P_8021Q)) { | |
1800 | struct vlan_ethhdr *veh; | |
1801 | ||
1802 | if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN))) | |
1803 | return ERR_PTR(-EINVAL); | |
1804 | ||
1805 | veh = (struct vlan_ethhdr *)skb->data; | |
1806 | type = veh->h_vlan_encapsulated_proto; | |
1807 | } | |
1808 | ||
1809 | skb_reset_mac_header(skb); | |
1810 | skb->mac_len = skb->network_header - skb->mac_header; | |
1811 | __skb_pull(skb, skb->mac_len); | |
1812 | ||
1813 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { | |
1814 | struct net_device *dev = skb->dev; | |
1815 | struct ethtool_drvinfo info = {}; | |
1816 | ||
1817 | if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo) | |
1818 | dev->ethtool_ops->get_drvinfo(dev, &info); | |
1819 | ||
1820 | WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d " | |
1821 | "ip_summed=%d", | |
1822 | info.driver, dev ? dev->features : 0L, | |
1823 | skb->sk ? skb->sk->sk_route_caps : 0L, | |
1824 | skb->len, skb->data_len, skb->ip_summed); | |
1825 | ||
1826 | if (skb_header_cloned(skb) && | |
1827 | (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))) | |
1828 | return ERR_PTR(err); | |
1829 | } | |
1830 | ||
1831 | rcu_read_lock(); | |
1832 | list_for_each_entry_rcu(ptype, | |
1833 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
1834 | if (ptype->type == type && !ptype->dev && ptype->gso_segment) { | |
1835 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { | |
1836 | err = ptype->gso_send_check(skb); | |
1837 | segs = ERR_PTR(err); | |
1838 | if (err || skb_gso_ok(skb, features)) | |
1839 | break; | |
1840 | __skb_push(skb, (skb->data - | |
1841 | skb_network_header(skb))); | |
1842 | } | |
1843 | segs = ptype->gso_segment(skb, features); | |
1844 | break; | |
1845 | } | |
1846 | } | |
1847 | rcu_read_unlock(); | |
1848 | ||
1849 | __skb_push(skb, skb->data - skb_mac_header(skb)); | |
1850 | ||
1851 | return segs; | |
1852 | } | |
1853 | EXPORT_SYMBOL(skb_gso_segment); | |
1854 | ||
1855 | /* Take action when hardware reception checksum errors are detected. */ | |
1856 | #ifdef CONFIG_BUG | |
1857 | void netdev_rx_csum_fault(struct net_device *dev) | |
1858 | { | |
1859 | if (net_ratelimit()) { | |
1860 | printk(KERN_ERR "%s: hw csum failure.\n", | |
1861 | dev ? dev->name : "<unknown>"); | |
1862 | dump_stack(); | |
1863 | } | |
1864 | } | |
1865 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
1866 | #endif | |
1867 | ||
1868 | /* Actually, we should eliminate this check as soon as we know, that: | |
1869 | * 1. IOMMU is present and allows to map all the memory. | |
1870 | * 2. No high memory really exists on this machine. | |
1871 | */ | |
1872 | ||
1873 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) | |
1874 | { | |
1875 | #ifdef CONFIG_HIGHMEM | |
1876 | int i; | |
1877 | if (!(dev->features & NETIF_F_HIGHDMA)) { | |
1878 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) | |
1879 | if (PageHighMem(skb_shinfo(skb)->frags[i].page)) | |
1880 | return 1; | |
1881 | } | |
1882 | ||
1883 | if (PCI_DMA_BUS_IS_PHYS) { | |
1884 | struct device *pdev = dev->dev.parent; | |
1885 | ||
1886 | if (!pdev) | |
1887 | return 0; | |
1888 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
1889 | dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page); | |
1890 | if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask) | |
1891 | return 1; | |
1892 | } | |
1893 | } | |
1894 | #endif | |
1895 | return 0; | |
1896 | } | |
1897 | ||
1898 | struct dev_gso_cb { | |
1899 | void (*destructor)(struct sk_buff *skb); | |
1900 | }; | |
1901 | ||
1902 | #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb) | |
1903 | ||
1904 | static void dev_gso_skb_destructor(struct sk_buff *skb) | |
1905 | { | |
1906 | struct dev_gso_cb *cb; | |
1907 | ||
1908 | do { | |
1909 | struct sk_buff *nskb = skb->next; | |
1910 | ||
1911 | skb->next = nskb->next; | |
1912 | nskb->next = NULL; | |
1913 | kfree_skb(nskb); | |
1914 | } while (skb->next); | |
1915 | ||
1916 | cb = DEV_GSO_CB(skb); | |
1917 | if (cb->destructor) | |
1918 | cb->destructor(skb); | |
1919 | } | |
1920 | ||
1921 | /** | |
1922 | * dev_gso_segment - Perform emulated hardware segmentation on skb. | |
1923 | * @skb: buffer to segment | |
1924 | * | |
1925 | * This function segments the given skb and stores the list of segments | |
1926 | * in skb->next. | |
1927 | */ | |
1928 | static int dev_gso_segment(struct sk_buff *skb) | |
1929 | { | |
1930 | struct net_device *dev = skb->dev; | |
1931 | struct sk_buff *segs; | |
1932 | int features = dev->features & ~(illegal_highdma(dev, skb) ? | |
1933 | NETIF_F_SG : 0); | |
1934 | ||
1935 | segs = skb_gso_segment(skb, features); | |
1936 | ||
1937 | /* Verifying header integrity only. */ | |
1938 | if (!segs) | |
1939 | return 0; | |
1940 | ||
1941 | if (IS_ERR(segs)) | |
1942 | return PTR_ERR(segs); | |
1943 | ||
1944 | skb->next = segs; | |
1945 | DEV_GSO_CB(skb)->destructor = skb->destructor; | |
1946 | skb->destructor = dev_gso_skb_destructor; | |
1947 | ||
1948 | return 0; | |
1949 | } | |
1950 | ||
1951 | /* | |
1952 | * Try to orphan skb early, right before transmission by the device. | |
1953 | * We cannot orphan skb if tx timestamp is requested or the sk-reference | |
1954 | * is needed on driver level for other reasons, e.g. see net/can/raw.c | |
1955 | */ | |
1956 | static inline void skb_orphan_try(struct sk_buff *skb) | |
1957 | { | |
1958 | struct sock *sk = skb->sk; | |
1959 | ||
1960 | if (sk && !skb_shinfo(skb)->tx_flags) { | |
1961 | /* skb_tx_hash() wont be able to get sk. | |
1962 | * We copy sk_hash into skb->rxhash | |
1963 | */ | |
1964 | if (!skb->rxhash) | |
1965 | skb->rxhash = sk->sk_hash; | |
1966 | skb_orphan(skb); | |
1967 | } | |
1968 | } | |
1969 | ||
1970 | /* | |
1971 | * Returns true if either: | |
1972 | * 1. skb has frag_list and the device doesn't support FRAGLIST, or | |
1973 | * 2. skb is fragmented and the device does not support SG, or if | |
1974 | * at least one of fragments is in highmem and device does not | |
1975 | * support DMA from it. | |
1976 | */ | |
1977 | static inline int skb_needs_linearize(struct sk_buff *skb, | |
1978 | struct net_device *dev) | |
1979 | { | |
1980 | int features = dev->features; | |
1981 | ||
1982 | if (skb->protocol == htons(ETH_P_8021Q) || vlan_tx_tag_present(skb)) | |
1983 | features &= dev->vlan_features; | |
1984 | ||
1985 | return skb_is_nonlinear(skb) && | |
1986 | ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || | |
1987 | (skb_shinfo(skb)->nr_frags && (!(features & NETIF_F_SG) || | |
1988 | illegal_highdma(dev, skb)))); | |
1989 | } | |
1990 | ||
1991 | int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev, | |
1992 | struct netdev_queue *txq) | |
1993 | { | |
1994 | const struct net_device_ops *ops = dev->netdev_ops; | |
1995 | int rc = NETDEV_TX_OK; | |
1996 | ||
1997 | if (likely(!skb->next)) { | |
1998 | if (!list_empty(&ptype_all)) | |
1999 | dev_queue_xmit_nit(skb, dev); | |
2000 | ||
2001 | /* | |
2002 | * If device doesnt need skb->dst, release it right now while | |
2003 | * its hot in this cpu cache | |
2004 | */ | |
2005 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
2006 | skb_dst_drop(skb); | |
2007 | ||
2008 | skb_orphan_try(skb); | |
2009 | ||
2010 | if (vlan_tx_tag_present(skb) && | |
2011 | !(dev->features & NETIF_F_HW_VLAN_TX)) { | |
2012 | skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb)); | |
2013 | if (unlikely(!skb)) | |
2014 | goto out; | |
2015 | ||
2016 | skb->vlan_tci = 0; | |
2017 | } | |
2018 | ||
2019 | if (netif_needs_gso(dev, skb)) { | |
2020 | if (unlikely(dev_gso_segment(skb))) | |
2021 | goto out_kfree_skb; | |
2022 | if (skb->next) | |
2023 | goto gso; | |
2024 | } else { | |
2025 | if (skb_needs_linearize(skb, dev) && | |
2026 | __skb_linearize(skb)) | |
2027 | goto out_kfree_skb; | |
2028 | ||
2029 | /* If packet is not checksummed and device does not | |
2030 | * support checksumming for this protocol, complete | |
2031 | * checksumming here. | |
2032 | */ | |
2033 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
2034 | skb_set_transport_header(skb, skb->csum_start - | |
2035 | skb_headroom(skb)); | |
2036 | if (!dev_can_checksum(dev, skb) && | |
2037 | skb_checksum_help(skb)) | |
2038 | goto out_kfree_skb; | |
2039 | } | |
2040 | } | |
2041 | ||
2042 | rc = ops->ndo_start_xmit(skb, dev); | |
2043 | trace_net_dev_xmit(skb, rc); | |
2044 | if (rc == NETDEV_TX_OK) | |
2045 | txq_trans_update(txq); | |
2046 | return rc; | |
2047 | } | |
2048 | ||
2049 | gso: | |
2050 | do { | |
2051 | struct sk_buff *nskb = skb->next; | |
2052 | ||
2053 | skb->next = nskb->next; | |
2054 | nskb->next = NULL; | |
2055 | ||
2056 | /* | |
2057 | * If device doesnt need nskb->dst, release it right now while | |
2058 | * its hot in this cpu cache | |
2059 | */ | |
2060 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
2061 | skb_dst_drop(nskb); | |
2062 | ||
2063 | rc = ops->ndo_start_xmit(nskb, dev); | |
2064 | trace_net_dev_xmit(nskb, rc); | |
2065 | if (unlikely(rc != NETDEV_TX_OK)) { | |
2066 | if (rc & ~NETDEV_TX_MASK) | |
2067 | goto out_kfree_gso_skb; | |
2068 | nskb->next = skb->next; | |
2069 | skb->next = nskb; | |
2070 | return rc; | |
2071 | } | |
2072 | txq_trans_update(txq); | |
2073 | if (unlikely(netif_tx_queue_stopped(txq) && skb->next)) | |
2074 | return NETDEV_TX_BUSY; | |
2075 | } while (skb->next); | |
2076 | ||
2077 | out_kfree_gso_skb: | |
2078 | if (likely(skb->next == NULL)) | |
2079 | skb->destructor = DEV_GSO_CB(skb)->destructor; | |
2080 | out_kfree_skb: | |
2081 | kfree_skb(skb); | |
2082 | out: | |
2083 | return rc; | |
2084 | } | |
2085 | ||
2086 | static u32 hashrnd __read_mostly; | |
2087 | ||
2088 | u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb) | |
2089 | { | |
2090 | u32 hash; | |
2091 | ||
2092 | if (skb_rx_queue_recorded(skb)) { | |
2093 | hash = skb_get_rx_queue(skb); | |
2094 | while (unlikely(hash >= dev->real_num_tx_queues)) | |
2095 | hash -= dev->real_num_tx_queues; | |
2096 | return hash; | |
2097 | } | |
2098 | ||
2099 | if (skb->sk && skb->sk->sk_hash) | |
2100 | hash = skb->sk->sk_hash; | |
2101 | else | |
2102 | hash = (__force u16) skb->protocol ^ skb->rxhash; | |
2103 | hash = jhash_1word(hash, hashrnd); | |
2104 | ||
2105 | return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32); | |
2106 | } | |
2107 | EXPORT_SYMBOL(skb_tx_hash); | |
2108 | ||
2109 | static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index) | |
2110 | { | |
2111 | if (unlikely(queue_index >= dev->real_num_tx_queues)) { | |
2112 | if (net_ratelimit()) { | |
2113 | pr_warning("%s selects TX queue %d, but " | |
2114 | "real number of TX queues is %d\n", | |
2115 | dev->name, queue_index, dev->real_num_tx_queues); | |
2116 | } | |
2117 | return 0; | |
2118 | } | |
2119 | return queue_index; | |
2120 | } | |
2121 | ||
2122 | static struct netdev_queue *dev_pick_tx(struct net_device *dev, | |
2123 | struct sk_buff *skb) | |
2124 | { | |
2125 | int queue_index; | |
2126 | const struct net_device_ops *ops = dev->netdev_ops; | |
2127 | ||
2128 | if (ops->ndo_select_queue) { | |
2129 | queue_index = ops->ndo_select_queue(dev, skb); | |
2130 | queue_index = dev_cap_txqueue(dev, queue_index); | |
2131 | } else { | |
2132 | struct sock *sk = skb->sk; | |
2133 | queue_index = sk_tx_queue_get(sk); | |
2134 | if (queue_index < 0) { | |
2135 | ||
2136 | queue_index = 0; | |
2137 | if (dev->real_num_tx_queues > 1) | |
2138 | queue_index = skb_tx_hash(dev, skb); | |
2139 | ||
2140 | if (sk) { | |
2141 | struct dst_entry *dst = rcu_dereference_check(sk->sk_dst_cache, 1); | |
2142 | ||
2143 | if (dst && skb_dst(skb) == dst) | |
2144 | sk_tx_queue_set(sk, queue_index); | |
2145 | } | |
2146 | } | |
2147 | } | |
2148 | ||
2149 | skb_set_queue_mapping(skb, queue_index); | |
2150 | return netdev_get_tx_queue(dev, queue_index); | |
2151 | } | |
2152 | ||
2153 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, | |
2154 | struct net_device *dev, | |
2155 | struct netdev_queue *txq) | |
2156 | { | |
2157 | spinlock_t *root_lock = qdisc_lock(q); | |
2158 | bool contended = qdisc_is_running(q); | |
2159 | int rc; | |
2160 | ||
2161 | /* | |
2162 | * Heuristic to force contended enqueues to serialize on a | |
2163 | * separate lock before trying to get qdisc main lock. | |
2164 | * This permits __QDISC_STATE_RUNNING owner to get the lock more often | |
2165 | * and dequeue packets faster. | |
2166 | */ | |
2167 | if (unlikely(contended)) | |
2168 | spin_lock(&q->busylock); | |
2169 | ||
2170 | spin_lock(root_lock); | |
2171 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
2172 | kfree_skb(skb); | |
2173 | rc = NET_XMIT_DROP; | |
2174 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
2175 | qdisc_run_begin(q)) { | |
2176 | /* | |
2177 | * This is a work-conserving queue; there are no old skbs | |
2178 | * waiting to be sent out; and the qdisc is not running - | |
2179 | * xmit the skb directly. | |
2180 | */ | |
2181 | if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE)) | |
2182 | skb_dst_force(skb); | |
2183 | __qdisc_update_bstats(q, skb->len); | |
2184 | if (sch_direct_xmit(skb, q, dev, txq, root_lock)) { | |
2185 | if (unlikely(contended)) { | |
2186 | spin_unlock(&q->busylock); | |
2187 | contended = false; | |
2188 | } | |
2189 | __qdisc_run(q); | |
2190 | } else | |
2191 | qdisc_run_end(q); | |
2192 | ||
2193 | rc = NET_XMIT_SUCCESS; | |
2194 | } else { | |
2195 | skb_dst_force(skb); | |
2196 | rc = qdisc_enqueue_root(skb, q); | |
2197 | if (qdisc_run_begin(q)) { | |
2198 | if (unlikely(contended)) { | |
2199 | spin_unlock(&q->busylock); | |
2200 | contended = false; | |
2201 | } | |
2202 | __qdisc_run(q); | |
2203 | } | |
2204 | } | |
2205 | spin_unlock(root_lock); | |
2206 | if (unlikely(contended)) | |
2207 | spin_unlock(&q->busylock); | |
2208 | return rc; | |
2209 | } | |
2210 | ||
2211 | static DEFINE_PER_CPU(int, xmit_recursion); | |
2212 | #define RECURSION_LIMIT 10 | |
2213 | ||
2214 | /** | |
2215 | * dev_queue_xmit - transmit a buffer | |
2216 | * @skb: buffer to transmit | |
2217 | * | |
2218 | * Queue a buffer for transmission to a network device. The caller must | |
2219 | * have set the device and priority and built the buffer before calling | |
2220 | * this function. The function can be called from an interrupt. | |
2221 | * | |
2222 | * A negative errno code is returned on a failure. A success does not | |
2223 | * guarantee the frame will be transmitted as it may be dropped due | |
2224 | * to congestion or traffic shaping. | |
2225 | * | |
2226 | * ----------------------------------------------------------------------------------- | |
2227 | * I notice this method can also return errors from the queue disciplines, | |
2228 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
2229 | * be positive. | |
2230 | * | |
2231 | * Regardless of the return value, the skb is consumed, so it is currently | |
2232 | * difficult to retry a send to this method. (You can bump the ref count | |
2233 | * before sending to hold a reference for retry if you are careful.) | |
2234 | * | |
2235 | * When calling this method, interrupts MUST be enabled. This is because | |
2236 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
2237 | * --BLG | |
2238 | */ | |
2239 | int dev_queue_xmit(struct sk_buff *skb) | |
2240 | { | |
2241 | struct net_device *dev = skb->dev; | |
2242 | struct netdev_queue *txq; | |
2243 | struct Qdisc *q; | |
2244 | int rc = -ENOMEM; | |
2245 | ||
2246 | /* Disable soft irqs for various locks below. Also | |
2247 | * stops preemption for RCU. | |
2248 | */ | |
2249 | rcu_read_lock_bh(); | |
2250 | ||
2251 | txq = dev_pick_tx(dev, skb); | |
2252 | q = rcu_dereference_bh(txq->qdisc); | |
2253 | ||
2254 | #ifdef CONFIG_NET_CLS_ACT | |
2255 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS); | |
2256 | #endif | |
2257 | trace_net_dev_queue(skb); | |
2258 | if (q->enqueue) { | |
2259 | rc = __dev_xmit_skb(skb, q, dev, txq); | |
2260 | goto out; | |
2261 | } | |
2262 | ||
2263 | /* The device has no queue. Common case for software devices: | |
2264 | loopback, all the sorts of tunnels... | |
2265 | ||
2266 | Really, it is unlikely that netif_tx_lock protection is necessary | |
2267 | here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
2268 | counters.) | |
2269 | However, it is possible, that they rely on protection | |
2270 | made by us here. | |
2271 | ||
2272 | Check this and shot the lock. It is not prone from deadlocks. | |
2273 | Either shot noqueue qdisc, it is even simpler 8) | |
2274 | */ | |
2275 | if (dev->flags & IFF_UP) { | |
2276 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
2277 | ||
2278 | if (txq->xmit_lock_owner != cpu) { | |
2279 | ||
2280 | if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT) | |
2281 | goto recursion_alert; | |
2282 | ||
2283 | HARD_TX_LOCK(dev, txq, cpu); | |
2284 | ||
2285 | if (!netif_tx_queue_stopped(txq)) { | |
2286 | __this_cpu_inc(xmit_recursion); | |
2287 | rc = dev_hard_start_xmit(skb, dev, txq); | |
2288 | __this_cpu_dec(xmit_recursion); | |
2289 | if (dev_xmit_complete(rc)) { | |
2290 | HARD_TX_UNLOCK(dev, txq); | |
2291 | goto out; | |
2292 | } | |
2293 | } | |
2294 | HARD_TX_UNLOCK(dev, txq); | |
2295 | if (net_ratelimit()) | |
2296 | printk(KERN_CRIT "Virtual device %s asks to " | |
2297 | "queue packet!\n", dev->name); | |
2298 | } else { | |
2299 | /* Recursion is detected! It is possible, | |
2300 | * unfortunately | |
2301 | */ | |
2302 | recursion_alert: | |
2303 | if (net_ratelimit()) | |
2304 | printk(KERN_CRIT "Dead loop on virtual device " | |
2305 | "%s, fix it urgently!\n", dev->name); | |
2306 | } | |
2307 | } | |
2308 | ||
2309 | rc = -ENETDOWN; | |
2310 | rcu_read_unlock_bh(); | |
2311 | ||
2312 | kfree_skb(skb); | |
2313 | return rc; | |
2314 | out: | |
2315 | rcu_read_unlock_bh(); | |
2316 | return rc; | |
2317 | } | |
2318 | EXPORT_SYMBOL(dev_queue_xmit); | |
2319 | ||
2320 | ||
2321 | /*======================================================================= | |
2322 | Receiver routines | |
2323 | =======================================================================*/ | |
2324 | ||
2325 | int netdev_max_backlog __read_mostly = 1000; | |
2326 | int netdev_tstamp_prequeue __read_mostly = 1; | |
2327 | int netdev_budget __read_mostly = 300; | |
2328 | int weight_p __read_mostly = 64; /* old backlog weight */ | |
2329 | ||
2330 | /* Called with irq disabled */ | |
2331 | static inline void ____napi_schedule(struct softnet_data *sd, | |
2332 | struct napi_struct *napi) | |
2333 | { | |
2334 | list_add_tail(&napi->poll_list, &sd->poll_list); | |
2335 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2336 | } | |
2337 | ||
2338 | /* | |
2339 | * __skb_get_rxhash: calculate a flow hash based on src/dst addresses | |
2340 | * and src/dst port numbers. Returns a non-zero hash number on success | |
2341 | * and 0 on failure. | |
2342 | */ | |
2343 | __u32 __skb_get_rxhash(struct sk_buff *skb) | |
2344 | { | |
2345 | int nhoff, hash = 0, poff; | |
2346 | struct ipv6hdr *ip6; | |
2347 | struct iphdr *ip; | |
2348 | u8 ip_proto; | |
2349 | u32 addr1, addr2, ihl; | |
2350 | union { | |
2351 | u32 v32; | |
2352 | u16 v16[2]; | |
2353 | } ports; | |
2354 | ||
2355 | nhoff = skb_network_offset(skb); | |
2356 | ||
2357 | switch (skb->protocol) { | |
2358 | case __constant_htons(ETH_P_IP): | |
2359 | if (!pskb_may_pull(skb, sizeof(*ip) + nhoff)) | |
2360 | goto done; | |
2361 | ||
2362 | ip = (struct iphdr *) (skb->data + nhoff); | |
2363 | if (ip->frag_off & htons(IP_MF | IP_OFFSET)) | |
2364 | ip_proto = 0; | |
2365 | else | |
2366 | ip_proto = ip->protocol; | |
2367 | addr1 = (__force u32) ip->saddr; | |
2368 | addr2 = (__force u32) ip->daddr; | |
2369 | ihl = ip->ihl; | |
2370 | break; | |
2371 | case __constant_htons(ETH_P_IPV6): | |
2372 | if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff)) | |
2373 | goto done; | |
2374 | ||
2375 | ip6 = (struct ipv6hdr *) (skb->data + nhoff); | |
2376 | ip_proto = ip6->nexthdr; | |
2377 | addr1 = (__force u32) ip6->saddr.s6_addr32[3]; | |
2378 | addr2 = (__force u32) ip6->daddr.s6_addr32[3]; | |
2379 | ihl = (40 >> 2); | |
2380 | break; | |
2381 | default: | |
2382 | goto done; | |
2383 | } | |
2384 | ||
2385 | ports.v32 = 0; | |
2386 | poff = proto_ports_offset(ip_proto); | |
2387 | if (poff >= 0) { | |
2388 | nhoff += ihl * 4 + poff; | |
2389 | if (pskb_may_pull(skb, nhoff + 4)) { | |
2390 | ports.v32 = * (__force u32 *) (skb->data + nhoff); | |
2391 | if (ports.v16[1] < ports.v16[0]) | |
2392 | swap(ports.v16[0], ports.v16[1]); | |
2393 | } | |
2394 | } | |
2395 | ||
2396 | /* get a consistent hash (same value on both flow directions) */ | |
2397 | if (addr2 < addr1) | |
2398 | swap(addr1, addr2); | |
2399 | ||
2400 | hash = jhash_3words(addr1, addr2, ports.v32, hashrnd); | |
2401 | if (!hash) | |
2402 | hash = 1; | |
2403 | ||
2404 | done: | |
2405 | return hash; | |
2406 | } | |
2407 | EXPORT_SYMBOL(__skb_get_rxhash); | |
2408 | ||
2409 | #ifdef CONFIG_RPS | |
2410 | ||
2411 | /* One global table that all flow-based protocols share. */ | |
2412 | struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; | |
2413 | EXPORT_SYMBOL(rps_sock_flow_table); | |
2414 | ||
2415 | /* | |
2416 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
2417 | * CPU from the RPS map of the receiving queue for a given skb. | |
2418 | * rcu_read_lock must be held on entry. | |
2419 | */ | |
2420 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
2421 | struct rps_dev_flow **rflowp) | |
2422 | { | |
2423 | struct netdev_rx_queue *rxqueue; | |
2424 | struct rps_map *map; | |
2425 | struct rps_dev_flow_table *flow_table; | |
2426 | struct rps_sock_flow_table *sock_flow_table; | |
2427 | int cpu = -1; | |
2428 | u16 tcpu; | |
2429 | ||
2430 | if (skb_rx_queue_recorded(skb)) { | |
2431 | u16 index = skb_get_rx_queue(skb); | |
2432 | if (unlikely(index >= dev->real_num_rx_queues)) { | |
2433 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
2434 | "%s received packet on queue %u, but number " | |
2435 | "of RX queues is %u\n", | |
2436 | dev->name, index, dev->real_num_rx_queues); | |
2437 | goto done; | |
2438 | } | |
2439 | rxqueue = dev->_rx + index; | |
2440 | } else | |
2441 | rxqueue = dev->_rx; | |
2442 | ||
2443 | map = rcu_dereference(rxqueue->rps_map); | |
2444 | if (map) { | |
2445 | if (map->len == 1) { | |
2446 | tcpu = map->cpus[0]; | |
2447 | if (cpu_online(tcpu)) | |
2448 | cpu = tcpu; | |
2449 | goto done; | |
2450 | } | |
2451 | } else if (!rcu_dereference_raw(rxqueue->rps_flow_table)) { | |
2452 | goto done; | |
2453 | } | |
2454 | ||
2455 | skb_reset_network_header(skb); | |
2456 | if (!skb_get_rxhash(skb)) | |
2457 | goto done; | |
2458 | ||
2459 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
2460 | sock_flow_table = rcu_dereference(rps_sock_flow_table); | |
2461 | if (flow_table && sock_flow_table) { | |
2462 | u16 next_cpu; | |
2463 | struct rps_dev_flow *rflow; | |
2464 | ||
2465 | rflow = &flow_table->flows[skb->rxhash & flow_table->mask]; | |
2466 | tcpu = rflow->cpu; | |
2467 | ||
2468 | next_cpu = sock_flow_table->ents[skb->rxhash & | |
2469 | sock_flow_table->mask]; | |
2470 | ||
2471 | /* | |
2472 | * If the desired CPU (where last recvmsg was done) is | |
2473 | * different from current CPU (one in the rx-queue flow | |
2474 | * table entry), switch if one of the following holds: | |
2475 | * - Current CPU is unset (equal to RPS_NO_CPU). | |
2476 | * - Current CPU is offline. | |
2477 | * - The current CPU's queue tail has advanced beyond the | |
2478 | * last packet that was enqueued using this table entry. | |
2479 | * This guarantees that all previous packets for the flow | |
2480 | * have been dequeued, thus preserving in order delivery. | |
2481 | */ | |
2482 | if (unlikely(tcpu != next_cpu) && | |
2483 | (tcpu == RPS_NO_CPU || !cpu_online(tcpu) || | |
2484 | ((int)(per_cpu(softnet_data, tcpu).input_queue_head - | |
2485 | rflow->last_qtail)) >= 0)) { | |
2486 | tcpu = rflow->cpu = next_cpu; | |
2487 | if (tcpu != RPS_NO_CPU) | |
2488 | rflow->last_qtail = per_cpu(softnet_data, | |
2489 | tcpu).input_queue_head; | |
2490 | } | |
2491 | if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) { | |
2492 | *rflowp = rflow; | |
2493 | cpu = tcpu; | |
2494 | goto done; | |
2495 | } | |
2496 | } | |
2497 | ||
2498 | if (map) { | |
2499 | tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32]; | |
2500 | ||
2501 | if (cpu_online(tcpu)) { | |
2502 | cpu = tcpu; | |
2503 | goto done; | |
2504 | } | |
2505 | } | |
2506 | ||
2507 | done: | |
2508 | return cpu; | |
2509 | } | |
2510 | ||
2511 | /* Called from hardirq (IPI) context */ | |
2512 | static void rps_trigger_softirq(void *data) | |
2513 | { | |
2514 | struct softnet_data *sd = data; | |
2515 | ||
2516 | ____napi_schedule(sd, &sd->backlog); | |
2517 | sd->received_rps++; | |
2518 | } | |
2519 | ||
2520 | #endif /* CONFIG_RPS */ | |
2521 | ||
2522 | /* | |
2523 | * Check if this softnet_data structure is another cpu one | |
2524 | * If yes, queue it to our IPI list and return 1 | |
2525 | * If no, return 0 | |
2526 | */ | |
2527 | static int rps_ipi_queued(struct softnet_data *sd) | |
2528 | { | |
2529 | #ifdef CONFIG_RPS | |
2530 | struct softnet_data *mysd = &__get_cpu_var(softnet_data); | |
2531 | ||
2532 | if (sd != mysd) { | |
2533 | sd->rps_ipi_next = mysd->rps_ipi_list; | |
2534 | mysd->rps_ipi_list = sd; | |
2535 | ||
2536 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2537 | return 1; | |
2538 | } | |
2539 | #endif /* CONFIG_RPS */ | |
2540 | return 0; | |
2541 | } | |
2542 | ||
2543 | /* | |
2544 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
2545 | * queue (may be a remote CPU queue). | |
2546 | */ | |
2547 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, | |
2548 | unsigned int *qtail) | |
2549 | { | |
2550 | struct softnet_data *sd; | |
2551 | unsigned long flags; | |
2552 | ||
2553 | sd = &per_cpu(softnet_data, cpu); | |
2554 | ||
2555 | local_irq_save(flags); | |
2556 | ||
2557 | rps_lock(sd); | |
2558 | if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) { | |
2559 | if (skb_queue_len(&sd->input_pkt_queue)) { | |
2560 | enqueue: | |
2561 | __skb_queue_tail(&sd->input_pkt_queue, skb); | |
2562 | input_queue_tail_incr_save(sd, qtail); | |
2563 | rps_unlock(sd); | |
2564 | local_irq_restore(flags); | |
2565 | return NET_RX_SUCCESS; | |
2566 | } | |
2567 | ||
2568 | /* Schedule NAPI for backlog device | |
2569 | * We can use non atomic operation since we own the queue lock | |
2570 | */ | |
2571 | if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { | |
2572 | if (!rps_ipi_queued(sd)) | |
2573 | ____napi_schedule(sd, &sd->backlog); | |
2574 | } | |
2575 | goto enqueue; | |
2576 | } | |
2577 | ||
2578 | sd->dropped++; | |
2579 | rps_unlock(sd); | |
2580 | ||
2581 | local_irq_restore(flags); | |
2582 | ||
2583 | atomic_long_inc(&skb->dev->rx_dropped); | |
2584 | kfree_skb(skb); | |
2585 | return NET_RX_DROP; | |
2586 | } | |
2587 | ||
2588 | /** | |
2589 | * netif_rx - post buffer to the network code | |
2590 | * @skb: buffer to post | |
2591 | * | |
2592 | * This function receives a packet from a device driver and queues it for | |
2593 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
2594 | * may be dropped during processing for congestion control or by the | |
2595 | * protocol layers. | |
2596 | * | |
2597 | * return values: | |
2598 | * NET_RX_SUCCESS (no congestion) | |
2599 | * NET_RX_DROP (packet was dropped) | |
2600 | * | |
2601 | */ | |
2602 | ||
2603 | int netif_rx(struct sk_buff *skb) | |
2604 | { | |
2605 | int ret; | |
2606 | ||
2607 | /* if netpoll wants it, pretend we never saw it */ | |
2608 | if (netpoll_rx(skb)) | |
2609 | return NET_RX_DROP; | |
2610 | ||
2611 | if (netdev_tstamp_prequeue) | |
2612 | net_timestamp_check(skb); | |
2613 | ||
2614 | trace_netif_rx(skb); | |
2615 | #ifdef CONFIG_RPS | |
2616 | { | |
2617 | struct rps_dev_flow voidflow, *rflow = &voidflow; | |
2618 | int cpu; | |
2619 | ||
2620 | preempt_disable(); | |
2621 | rcu_read_lock(); | |
2622 | ||
2623 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
2624 | if (cpu < 0) | |
2625 | cpu = smp_processor_id(); | |
2626 | ||
2627 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
2628 | ||
2629 | rcu_read_unlock(); | |
2630 | preempt_enable(); | |
2631 | } | |
2632 | #else | |
2633 | { | |
2634 | unsigned int qtail; | |
2635 | ret = enqueue_to_backlog(skb, get_cpu(), &qtail); | |
2636 | put_cpu(); | |
2637 | } | |
2638 | #endif | |
2639 | return ret; | |
2640 | } | |
2641 | EXPORT_SYMBOL(netif_rx); | |
2642 | ||
2643 | int netif_rx_ni(struct sk_buff *skb) | |
2644 | { | |
2645 | int err; | |
2646 | ||
2647 | preempt_disable(); | |
2648 | err = netif_rx(skb); | |
2649 | if (local_softirq_pending()) | |
2650 | do_softirq(); | |
2651 | preempt_enable(); | |
2652 | ||
2653 | return err; | |
2654 | } | |
2655 | EXPORT_SYMBOL(netif_rx_ni); | |
2656 | ||
2657 | static void net_tx_action(struct softirq_action *h) | |
2658 | { | |
2659 | struct softnet_data *sd = &__get_cpu_var(softnet_data); | |
2660 | ||
2661 | if (sd->completion_queue) { | |
2662 | struct sk_buff *clist; | |
2663 | ||
2664 | local_irq_disable(); | |
2665 | clist = sd->completion_queue; | |
2666 | sd->completion_queue = NULL; | |
2667 | local_irq_enable(); | |
2668 | ||
2669 | while (clist) { | |
2670 | struct sk_buff *skb = clist; | |
2671 | clist = clist->next; | |
2672 | ||
2673 | WARN_ON(atomic_read(&skb->users)); | |
2674 | trace_kfree_skb(skb, net_tx_action); | |
2675 | __kfree_skb(skb); | |
2676 | } | |
2677 | } | |
2678 | ||
2679 | if (sd->output_queue) { | |
2680 | struct Qdisc *head; | |
2681 | ||
2682 | local_irq_disable(); | |
2683 | head = sd->output_queue; | |
2684 | sd->output_queue = NULL; | |
2685 | sd->output_queue_tailp = &sd->output_queue; | |
2686 | local_irq_enable(); | |
2687 | ||
2688 | while (head) { | |
2689 | struct Qdisc *q = head; | |
2690 | spinlock_t *root_lock; | |
2691 | ||
2692 | head = head->next_sched; | |
2693 | ||
2694 | root_lock = qdisc_lock(q); | |
2695 | if (spin_trylock(root_lock)) { | |
2696 | smp_mb__before_clear_bit(); | |
2697 | clear_bit(__QDISC_STATE_SCHED, | |
2698 | &q->state); | |
2699 | qdisc_run(q); | |
2700 | spin_unlock(root_lock); | |
2701 | } else { | |
2702 | if (!test_bit(__QDISC_STATE_DEACTIVATED, | |
2703 | &q->state)) { | |
2704 | __netif_reschedule(q); | |
2705 | } else { | |
2706 | smp_mb__before_clear_bit(); | |
2707 | clear_bit(__QDISC_STATE_SCHED, | |
2708 | &q->state); | |
2709 | } | |
2710 | } | |
2711 | } | |
2712 | } | |
2713 | } | |
2714 | ||
2715 | static inline int deliver_skb(struct sk_buff *skb, | |
2716 | struct packet_type *pt_prev, | |
2717 | struct net_device *orig_dev) | |
2718 | { | |
2719 | atomic_inc(&skb->users); | |
2720 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
2721 | } | |
2722 | ||
2723 | #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \ | |
2724 | (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)) | |
2725 | /* This hook is defined here for ATM LANE */ | |
2726 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
2727 | unsigned char *addr) __read_mostly; | |
2728 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); | |
2729 | #endif | |
2730 | ||
2731 | #ifdef CONFIG_NET_CLS_ACT | |
2732 | /* TODO: Maybe we should just force sch_ingress to be compiled in | |
2733 | * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions | |
2734 | * a compare and 2 stores extra right now if we dont have it on | |
2735 | * but have CONFIG_NET_CLS_ACT | |
2736 | * NOTE: This doesnt stop any functionality; if you dont have | |
2737 | * the ingress scheduler, you just cant add policies on ingress. | |
2738 | * | |
2739 | */ | |
2740 | static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq) | |
2741 | { | |
2742 | struct net_device *dev = skb->dev; | |
2743 | u32 ttl = G_TC_RTTL(skb->tc_verd); | |
2744 | int result = TC_ACT_OK; | |
2745 | struct Qdisc *q; | |
2746 | ||
2747 | if (unlikely(MAX_RED_LOOP < ttl++)) { | |
2748 | if (net_ratelimit()) | |
2749 | pr_warning( "Redir loop detected Dropping packet (%d->%d)\n", | |
2750 | skb->skb_iif, dev->ifindex); | |
2751 | return TC_ACT_SHOT; | |
2752 | } | |
2753 | ||
2754 | skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl); | |
2755 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); | |
2756 | ||
2757 | q = rxq->qdisc; | |
2758 | if (q != &noop_qdisc) { | |
2759 | spin_lock(qdisc_lock(q)); | |
2760 | if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) | |
2761 | result = qdisc_enqueue_root(skb, q); | |
2762 | spin_unlock(qdisc_lock(q)); | |
2763 | } | |
2764 | ||
2765 | return result; | |
2766 | } | |
2767 | ||
2768 | static inline struct sk_buff *handle_ing(struct sk_buff *skb, | |
2769 | struct packet_type **pt_prev, | |
2770 | int *ret, struct net_device *orig_dev) | |
2771 | { | |
2772 | struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue); | |
2773 | ||
2774 | if (!rxq || rxq->qdisc == &noop_qdisc) | |
2775 | goto out; | |
2776 | ||
2777 | if (*pt_prev) { | |
2778 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
2779 | *pt_prev = NULL; | |
2780 | } | |
2781 | ||
2782 | switch (ing_filter(skb, rxq)) { | |
2783 | case TC_ACT_SHOT: | |
2784 | case TC_ACT_STOLEN: | |
2785 | kfree_skb(skb); | |
2786 | return NULL; | |
2787 | } | |
2788 | ||
2789 | out: | |
2790 | skb->tc_verd = 0; | |
2791 | return skb; | |
2792 | } | |
2793 | #endif | |
2794 | ||
2795 | /** | |
2796 | * netdev_rx_handler_register - register receive handler | |
2797 | * @dev: device to register a handler for | |
2798 | * @rx_handler: receive handler to register | |
2799 | * @rx_handler_data: data pointer that is used by rx handler | |
2800 | * | |
2801 | * Register a receive hander for a device. This handler will then be | |
2802 | * called from __netif_receive_skb. A negative errno code is returned | |
2803 | * on a failure. | |
2804 | * | |
2805 | * The caller must hold the rtnl_mutex. | |
2806 | */ | |
2807 | int netdev_rx_handler_register(struct net_device *dev, | |
2808 | rx_handler_func_t *rx_handler, | |
2809 | void *rx_handler_data) | |
2810 | { | |
2811 | ASSERT_RTNL(); | |
2812 | ||
2813 | if (dev->rx_handler) | |
2814 | return -EBUSY; | |
2815 | ||
2816 | rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); | |
2817 | rcu_assign_pointer(dev->rx_handler, rx_handler); | |
2818 | ||
2819 | return 0; | |
2820 | } | |
2821 | EXPORT_SYMBOL_GPL(netdev_rx_handler_register); | |
2822 | ||
2823 | /** | |
2824 | * netdev_rx_handler_unregister - unregister receive handler | |
2825 | * @dev: device to unregister a handler from | |
2826 | * | |
2827 | * Unregister a receive hander from a device. | |
2828 | * | |
2829 | * The caller must hold the rtnl_mutex. | |
2830 | */ | |
2831 | void netdev_rx_handler_unregister(struct net_device *dev) | |
2832 | { | |
2833 | ||
2834 | ASSERT_RTNL(); | |
2835 | rcu_assign_pointer(dev->rx_handler, NULL); | |
2836 | rcu_assign_pointer(dev->rx_handler_data, NULL); | |
2837 | } | |
2838 | EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); | |
2839 | ||
2840 | static inline void skb_bond_set_mac_by_master(struct sk_buff *skb, | |
2841 | struct net_device *master) | |
2842 | { | |
2843 | if (skb->pkt_type == PACKET_HOST) { | |
2844 | u16 *dest = (u16 *) eth_hdr(skb)->h_dest; | |
2845 | ||
2846 | memcpy(dest, master->dev_addr, ETH_ALEN); | |
2847 | } | |
2848 | } | |
2849 | ||
2850 | /* On bonding slaves other than the currently active slave, suppress | |
2851 | * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and | |
2852 | * ARP on active-backup slaves with arp_validate enabled. | |
2853 | */ | |
2854 | int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master) | |
2855 | { | |
2856 | struct net_device *dev = skb->dev; | |
2857 | ||
2858 | if (master->priv_flags & IFF_MASTER_ARPMON) | |
2859 | dev->last_rx = jiffies; | |
2860 | ||
2861 | if ((master->priv_flags & IFF_MASTER_ALB) && | |
2862 | (master->priv_flags & IFF_BRIDGE_PORT)) { | |
2863 | /* Do address unmangle. The local destination address | |
2864 | * will be always the one master has. Provides the right | |
2865 | * functionality in a bridge. | |
2866 | */ | |
2867 | skb_bond_set_mac_by_master(skb, master); | |
2868 | } | |
2869 | ||
2870 | if (dev->priv_flags & IFF_SLAVE_INACTIVE) { | |
2871 | if ((dev->priv_flags & IFF_SLAVE_NEEDARP) && | |
2872 | skb->protocol == __cpu_to_be16(ETH_P_ARP)) | |
2873 | return 0; | |
2874 | ||
2875 | if (master->priv_flags & IFF_MASTER_ALB) { | |
2876 | if (skb->pkt_type != PACKET_BROADCAST && | |
2877 | skb->pkt_type != PACKET_MULTICAST) | |
2878 | return 0; | |
2879 | } | |
2880 | if (master->priv_flags & IFF_MASTER_8023AD && | |
2881 | skb->protocol == __cpu_to_be16(ETH_P_SLOW)) | |
2882 | return 0; | |
2883 | ||
2884 | return 1; | |
2885 | } | |
2886 | return 0; | |
2887 | } | |
2888 | EXPORT_SYMBOL(__skb_bond_should_drop); | |
2889 | ||
2890 | static int __netif_receive_skb(struct sk_buff *skb) | |
2891 | { | |
2892 | struct packet_type *ptype, *pt_prev; | |
2893 | rx_handler_func_t *rx_handler; | |
2894 | struct net_device *orig_dev; | |
2895 | struct net_device *master; | |
2896 | struct net_device *null_or_orig; | |
2897 | struct net_device *orig_or_bond; | |
2898 | int ret = NET_RX_DROP; | |
2899 | __be16 type; | |
2900 | ||
2901 | if (!netdev_tstamp_prequeue) | |
2902 | net_timestamp_check(skb); | |
2903 | ||
2904 | trace_netif_receive_skb(skb); | |
2905 | ||
2906 | /* if we've gotten here through NAPI, check netpoll */ | |
2907 | if (netpoll_receive_skb(skb)) | |
2908 | return NET_RX_DROP; | |
2909 | ||
2910 | if (!skb->skb_iif) | |
2911 | skb->skb_iif = skb->dev->ifindex; | |
2912 | ||
2913 | /* | |
2914 | * bonding note: skbs received on inactive slaves should only | |
2915 | * be delivered to pkt handlers that are exact matches. Also | |
2916 | * the deliver_no_wcard flag will be set. If packet handlers | |
2917 | * are sensitive to duplicate packets these skbs will need to | |
2918 | * be dropped at the handler. | |
2919 | */ | |
2920 | null_or_orig = NULL; | |
2921 | orig_dev = skb->dev; | |
2922 | master = ACCESS_ONCE(orig_dev->master); | |
2923 | if (skb->deliver_no_wcard) | |
2924 | null_or_orig = orig_dev; | |
2925 | else if (master) { | |
2926 | if (skb_bond_should_drop(skb, master)) { | |
2927 | skb->deliver_no_wcard = 1; | |
2928 | null_or_orig = orig_dev; /* deliver only exact match */ | |
2929 | } else | |
2930 | skb->dev = master; | |
2931 | } | |
2932 | ||
2933 | __this_cpu_inc(softnet_data.processed); | |
2934 | skb_reset_network_header(skb); | |
2935 | skb_reset_transport_header(skb); | |
2936 | skb->mac_len = skb->network_header - skb->mac_header; | |
2937 | ||
2938 | pt_prev = NULL; | |
2939 | ||
2940 | rcu_read_lock(); | |
2941 | ||
2942 | #ifdef CONFIG_NET_CLS_ACT | |
2943 | if (skb->tc_verd & TC_NCLS) { | |
2944 | skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); | |
2945 | goto ncls; | |
2946 | } | |
2947 | #endif | |
2948 | ||
2949 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
2950 | if (ptype->dev == null_or_orig || ptype->dev == skb->dev || | |
2951 | ptype->dev == orig_dev) { | |
2952 | if (pt_prev) | |
2953 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
2954 | pt_prev = ptype; | |
2955 | } | |
2956 | } | |
2957 | ||
2958 | #ifdef CONFIG_NET_CLS_ACT | |
2959 | skb = handle_ing(skb, &pt_prev, &ret, orig_dev); | |
2960 | if (!skb) | |
2961 | goto out; | |
2962 | ncls: | |
2963 | #endif | |
2964 | ||
2965 | /* Handle special case of bridge or macvlan */ | |
2966 | rx_handler = rcu_dereference(skb->dev->rx_handler); | |
2967 | if (rx_handler) { | |
2968 | if (pt_prev) { | |
2969 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
2970 | pt_prev = NULL; | |
2971 | } | |
2972 | skb = rx_handler(skb); | |
2973 | if (!skb) | |
2974 | goto out; | |
2975 | } | |
2976 | ||
2977 | if (vlan_tx_tag_present(skb)) { | |
2978 | if (pt_prev) { | |
2979 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
2980 | pt_prev = NULL; | |
2981 | } | |
2982 | if (vlan_hwaccel_do_receive(&skb)) { | |
2983 | ret = __netif_receive_skb(skb); | |
2984 | goto out; | |
2985 | } else if (unlikely(!skb)) | |
2986 | goto out; | |
2987 | } | |
2988 | ||
2989 | /* | |
2990 | * Make sure frames received on VLAN interfaces stacked on | |
2991 | * bonding interfaces still make their way to any base bonding | |
2992 | * device that may have registered for a specific ptype. The | |
2993 | * handler may have to adjust skb->dev and orig_dev. | |
2994 | */ | |
2995 | orig_or_bond = orig_dev; | |
2996 | if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) && | |
2997 | (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) { | |
2998 | orig_or_bond = vlan_dev_real_dev(skb->dev); | |
2999 | } | |
3000 | ||
3001 | type = skb->protocol; | |
3002 | list_for_each_entry_rcu(ptype, | |
3003 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
3004 | if (ptype->type == type && (ptype->dev == null_or_orig || | |
3005 | ptype->dev == skb->dev || ptype->dev == orig_dev || | |
3006 | ptype->dev == orig_or_bond)) { | |
3007 | if (pt_prev) | |
3008 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
3009 | pt_prev = ptype; | |
3010 | } | |
3011 | } | |
3012 | ||
3013 | if (pt_prev) { | |
3014 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
3015 | } else { | |
3016 | atomic_long_inc(&skb->dev->rx_dropped); | |
3017 | kfree_skb(skb); | |
3018 | /* Jamal, now you will not able to escape explaining | |
3019 | * me how you were going to use this. :-) | |
3020 | */ | |
3021 | ret = NET_RX_DROP; | |
3022 | } | |
3023 | ||
3024 | out: | |
3025 | rcu_read_unlock(); | |
3026 | return ret; | |
3027 | } | |
3028 | ||
3029 | /** | |
3030 | * netif_receive_skb - process receive buffer from network | |
3031 | * @skb: buffer to process | |
3032 | * | |
3033 | * netif_receive_skb() is the main receive data processing function. | |
3034 | * It always succeeds. The buffer may be dropped during processing | |
3035 | * for congestion control or by the protocol layers. | |
3036 | * | |
3037 | * This function may only be called from softirq context and interrupts | |
3038 | * should be enabled. | |
3039 | * | |
3040 | * Return values (usually ignored): | |
3041 | * NET_RX_SUCCESS: no congestion | |
3042 | * NET_RX_DROP: packet was dropped | |
3043 | */ | |
3044 | int netif_receive_skb(struct sk_buff *skb) | |
3045 | { | |
3046 | if (netdev_tstamp_prequeue) | |
3047 | net_timestamp_check(skb); | |
3048 | ||
3049 | if (skb_defer_rx_timestamp(skb)) | |
3050 | return NET_RX_SUCCESS; | |
3051 | ||
3052 | #ifdef CONFIG_RPS | |
3053 | { | |
3054 | struct rps_dev_flow voidflow, *rflow = &voidflow; | |
3055 | int cpu, ret; | |
3056 | ||
3057 | rcu_read_lock(); | |
3058 | ||
3059 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
3060 | ||
3061 | if (cpu >= 0) { | |
3062 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
3063 | rcu_read_unlock(); | |
3064 | } else { | |
3065 | rcu_read_unlock(); | |
3066 | ret = __netif_receive_skb(skb); | |
3067 | } | |
3068 | ||
3069 | return ret; | |
3070 | } | |
3071 | #else | |
3072 | return __netif_receive_skb(skb); | |
3073 | #endif | |
3074 | } | |
3075 | EXPORT_SYMBOL(netif_receive_skb); | |
3076 | ||
3077 | /* Network device is going away, flush any packets still pending | |
3078 | * Called with irqs disabled. | |
3079 | */ | |
3080 | static void flush_backlog(void *arg) | |
3081 | { | |
3082 | struct net_device *dev = arg; | |
3083 | struct softnet_data *sd = &__get_cpu_var(softnet_data); | |
3084 | struct sk_buff *skb, *tmp; | |
3085 | ||
3086 | rps_lock(sd); | |
3087 | skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { | |
3088 | if (skb->dev == dev) { | |
3089 | __skb_unlink(skb, &sd->input_pkt_queue); | |
3090 | kfree_skb(skb); | |
3091 | input_queue_head_incr(sd); | |
3092 | } | |
3093 | } | |
3094 | rps_unlock(sd); | |
3095 | ||
3096 | skb_queue_walk_safe(&sd->process_queue, skb, tmp) { | |
3097 | if (skb->dev == dev) { | |
3098 | __skb_unlink(skb, &sd->process_queue); | |
3099 | kfree_skb(skb); | |
3100 | input_queue_head_incr(sd); | |
3101 | } | |
3102 | } | |
3103 | } | |
3104 | ||
3105 | static int napi_gro_complete(struct sk_buff *skb) | |
3106 | { | |
3107 | struct packet_type *ptype; | |
3108 | __be16 type = skb->protocol; | |
3109 | struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK]; | |
3110 | int err = -ENOENT; | |
3111 | ||
3112 | if (NAPI_GRO_CB(skb)->count == 1) { | |
3113 | skb_shinfo(skb)->gso_size = 0; | |
3114 | goto out; | |
3115 | } | |
3116 | ||
3117 | rcu_read_lock(); | |
3118 | list_for_each_entry_rcu(ptype, head, list) { | |
3119 | if (ptype->type != type || ptype->dev || !ptype->gro_complete) | |
3120 | continue; | |
3121 | ||
3122 | err = ptype->gro_complete(skb); | |
3123 | break; | |
3124 | } | |
3125 | rcu_read_unlock(); | |
3126 | ||
3127 | if (err) { | |
3128 | WARN_ON(&ptype->list == head); | |
3129 | kfree_skb(skb); | |
3130 | return NET_RX_SUCCESS; | |
3131 | } | |
3132 | ||
3133 | out: | |
3134 | return netif_receive_skb(skb); | |
3135 | } | |
3136 | ||
3137 | inline void napi_gro_flush(struct napi_struct *napi) | |
3138 | { | |
3139 | struct sk_buff *skb, *next; | |
3140 | ||
3141 | for (skb = napi->gro_list; skb; skb = next) { | |
3142 | next = skb->next; | |
3143 | skb->next = NULL; | |
3144 | napi_gro_complete(skb); | |
3145 | } | |
3146 | ||
3147 | napi->gro_count = 0; | |
3148 | napi->gro_list = NULL; | |
3149 | } | |
3150 | EXPORT_SYMBOL(napi_gro_flush); | |
3151 | ||
3152 | enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) | |
3153 | { | |
3154 | struct sk_buff **pp = NULL; | |
3155 | struct packet_type *ptype; | |
3156 | __be16 type = skb->protocol; | |
3157 | struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK]; | |
3158 | int same_flow; | |
3159 | int mac_len; | |
3160 | enum gro_result ret; | |
3161 | ||
3162 | if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb)) | |
3163 | goto normal; | |
3164 | ||
3165 | if (skb_is_gso(skb) || skb_has_frag_list(skb)) | |
3166 | goto normal; | |
3167 | ||
3168 | rcu_read_lock(); | |
3169 | list_for_each_entry_rcu(ptype, head, list) { | |
3170 | if (ptype->type != type || ptype->dev || !ptype->gro_receive) | |
3171 | continue; | |
3172 | ||
3173 | skb_set_network_header(skb, skb_gro_offset(skb)); | |
3174 | mac_len = skb->network_header - skb->mac_header; | |
3175 | skb->mac_len = mac_len; | |
3176 | NAPI_GRO_CB(skb)->same_flow = 0; | |
3177 | NAPI_GRO_CB(skb)->flush = 0; | |
3178 | NAPI_GRO_CB(skb)->free = 0; | |
3179 | ||
3180 | pp = ptype->gro_receive(&napi->gro_list, skb); | |
3181 | break; | |
3182 | } | |
3183 | rcu_read_unlock(); | |
3184 | ||
3185 | if (&ptype->list == head) | |
3186 | goto normal; | |
3187 | ||
3188 | same_flow = NAPI_GRO_CB(skb)->same_flow; | |
3189 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; | |
3190 | ||
3191 | if (pp) { | |
3192 | struct sk_buff *nskb = *pp; | |
3193 | ||
3194 | *pp = nskb->next; | |
3195 | nskb->next = NULL; | |
3196 | napi_gro_complete(nskb); | |
3197 | napi->gro_count--; | |
3198 | } | |
3199 | ||
3200 | if (same_flow) | |
3201 | goto ok; | |
3202 | ||
3203 | if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS) | |
3204 | goto normal; | |
3205 | ||
3206 | napi->gro_count++; | |
3207 | NAPI_GRO_CB(skb)->count = 1; | |
3208 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); | |
3209 | skb->next = napi->gro_list; | |
3210 | napi->gro_list = skb; | |
3211 | ret = GRO_HELD; | |
3212 | ||
3213 | pull: | |
3214 | if (skb_headlen(skb) < skb_gro_offset(skb)) { | |
3215 | int grow = skb_gro_offset(skb) - skb_headlen(skb); | |
3216 | ||
3217 | BUG_ON(skb->end - skb->tail < grow); | |
3218 | ||
3219 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
3220 | ||
3221 | skb->tail += grow; | |
3222 | skb->data_len -= grow; | |
3223 | ||
3224 | skb_shinfo(skb)->frags[0].page_offset += grow; | |
3225 | skb_shinfo(skb)->frags[0].size -= grow; | |
3226 | ||
3227 | if (unlikely(!skb_shinfo(skb)->frags[0].size)) { | |
3228 | put_page(skb_shinfo(skb)->frags[0].page); | |
3229 | memmove(skb_shinfo(skb)->frags, | |
3230 | skb_shinfo(skb)->frags + 1, | |
3231 | --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t)); | |
3232 | } | |
3233 | } | |
3234 | ||
3235 | ok: | |
3236 | return ret; | |
3237 | ||
3238 | normal: | |
3239 | ret = GRO_NORMAL; | |
3240 | goto pull; | |
3241 | } | |
3242 | EXPORT_SYMBOL(dev_gro_receive); | |
3243 | ||
3244 | static inline gro_result_t | |
3245 | __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) | |
3246 | { | |
3247 | struct sk_buff *p; | |
3248 | ||
3249 | for (p = napi->gro_list; p; p = p->next) { | |
3250 | unsigned long diffs; | |
3251 | ||
3252 | diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; | |
3253 | diffs |= p->vlan_tci ^ skb->vlan_tci; | |
3254 | diffs |= compare_ether_header(skb_mac_header(p), | |
3255 | skb_gro_mac_header(skb)); | |
3256 | NAPI_GRO_CB(p)->same_flow = !diffs; | |
3257 | NAPI_GRO_CB(p)->flush = 0; | |
3258 | } | |
3259 | ||
3260 | return dev_gro_receive(napi, skb); | |
3261 | } | |
3262 | ||
3263 | gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) | |
3264 | { | |
3265 | switch (ret) { | |
3266 | case GRO_NORMAL: | |
3267 | if (netif_receive_skb(skb)) | |
3268 | ret = GRO_DROP; | |
3269 | break; | |
3270 | ||
3271 | case GRO_DROP: | |
3272 | case GRO_MERGED_FREE: | |
3273 | kfree_skb(skb); | |
3274 | break; | |
3275 | ||
3276 | case GRO_HELD: | |
3277 | case GRO_MERGED: | |
3278 | break; | |
3279 | } | |
3280 | ||
3281 | return ret; | |
3282 | } | |
3283 | EXPORT_SYMBOL(napi_skb_finish); | |
3284 | ||
3285 | void skb_gro_reset_offset(struct sk_buff *skb) | |
3286 | { | |
3287 | NAPI_GRO_CB(skb)->data_offset = 0; | |
3288 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
3289 | NAPI_GRO_CB(skb)->frag0_len = 0; | |
3290 | ||
3291 | if (skb->mac_header == skb->tail && | |
3292 | !PageHighMem(skb_shinfo(skb)->frags[0].page)) { | |
3293 | NAPI_GRO_CB(skb)->frag0 = | |
3294 | page_address(skb_shinfo(skb)->frags[0].page) + | |
3295 | skb_shinfo(skb)->frags[0].page_offset; | |
3296 | NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size; | |
3297 | } | |
3298 | } | |
3299 | EXPORT_SYMBOL(skb_gro_reset_offset); | |
3300 | ||
3301 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) | |
3302 | { | |
3303 | skb_gro_reset_offset(skb); | |
3304 | ||
3305 | return napi_skb_finish(__napi_gro_receive(napi, skb), skb); | |
3306 | } | |
3307 | EXPORT_SYMBOL(napi_gro_receive); | |
3308 | ||
3309 | static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) | |
3310 | { | |
3311 | __skb_pull(skb, skb_headlen(skb)); | |
3312 | skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb)); | |
3313 | skb->vlan_tci = 0; | |
3314 | ||
3315 | napi->skb = skb; | |
3316 | } | |
3317 | ||
3318 | struct sk_buff *napi_get_frags(struct napi_struct *napi) | |
3319 | { | |
3320 | struct sk_buff *skb = napi->skb; | |
3321 | ||
3322 | if (!skb) { | |
3323 | skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD); | |
3324 | if (skb) | |
3325 | napi->skb = skb; | |
3326 | } | |
3327 | return skb; | |
3328 | } | |
3329 | EXPORT_SYMBOL(napi_get_frags); | |
3330 | ||
3331 | gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb, | |
3332 | gro_result_t ret) | |
3333 | { | |
3334 | switch (ret) { | |
3335 | case GRO_NORMAL: | |
3336 | case GRO_HELD: | |
3337 | skb->protocol = eth_type_trans(skb, skb->dev); | |
3338 | ||
3339 | if (ret == GRO_HELD) | |
3340 | skb_gro_pull(skb, -ETH_HLEN); | |
3341 | else if (netif_receive_skb(skb)) | |
3342 | ret = GRO_DROP; | |
3343 | break; | |
3344 | ||
3345 | case GRO_DROP: | |
3346 | case GRO_MERGED_FREE: | |
3347 | napi_reuse_skb(napi, skb); | |
3348 | break; | |
3349 | ||
3350 | case GRO_MERGED: | |
3351 | break; | |
3352 | } | |
3353 | ||
3354 | return ret; | |
3355 | } | |
3356 | EXPORT_SYMBOL(napi_frags_finish); | |
3357 | ||
3358 | struct sk_buff *napi_frags_skb(struct napi_struct *napi) | |
3359 | { | |
3360 | struct sk_buff *skb = napi->skb; | |
3361 | struct ethhdr *eth; | |
3362 | unsigned int hlen; | |
3363 | unsigned int off; | |
3364 | ||
3365 | napi->skb = NULL; | |
3366 | ||
3367 | skb_reset_mac_header(skb); | |
3368 | skb_gro_reset_offset(skb); | |
3369 | ||
3370 | off = skb_gro_offset(skb); | |
3371 | hlen = off + sizeof(*eth); | |
3372 | eth = skb_gro_header_fast(skb, off); | |
3373 | if (skb_gro_header_hard(skb, hlen)) { | |
3374 | eth = skb_gro_header_slow(skb, hlen, off); | |
3375 | if (unlikely(!eth)) { | |
3376 | napi_reuse_skb(napi, skb); | |
3377 | skb = NULL; | |
3378 | goto out; | |
3379 | } | |
3380 | } | |
3381 | ||
3382 | skb_gro_pull(skb, sizeof(*eth)); | |
3383 | ||
3384 | /* | |
3385 | * This works because the only protocols we care about don't require | |
3386 | * special handling. We'll fix it up properly at the end. | |
3387 | */ | |
3388 | skb->protocol = eth->h_proto; | |
3389 | ||
3390 | out: | |
3391 | return skb; | |
3392 | } | |
3393 | EXPORT_SYMBOL(napi_frags_skb); | |
3394 | ||
3395 | gro_result_t napi_gro_frags(struct napi_struct *napi) | |
3396 | { | |
3397 | struct sk_buff *skb = napi_frags_skb(napi); | |
3398 | ||
3399 | if (!skb) | |
3400 | return GRO_DROP; | |
3401 | ||
3402 | return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb)); | |
3403 | } | |
3404 | EXPORT_SYMBOL(napi_gro_frags); | |
3405 | ||
3406 | /* | |
3407 | * net_rps_action sends any pending IPI's for rps. | |
3408 | * Note: called with local irq disabled, but exits with local irq enabled. | |
3409 | */ | |
3410 | static void net_rps_action_and_irq_enable(struct softnet_data *sd) | |
3411 | { | |
3412 | #ifdef CONFIG_RPS | |
3413 | struct softnet_data *remsd = sd->rps_ipi_list; | |
3414 | ||
3415 | if (remsd) { | |
3416 | sd->rps_ipi_list = NULL; | |
3417 | ||
3418 | local_irq_enable(); | |
3419 | ||
3420 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
3421 | while (remsd) { | |
3422 | struct softnet_data *next = remsd->rps_ipi_next; | |
3423 | ||
3424 | if (cpu_online(remsd->cpu)) | |
3425 | __smp_call_function_single(remsd->cpu, | |
3426 | &remsd->csd, 0); | |
3427 | remsd = next; | |
3428 | } | |
3429 | } else | |
3430 | #endif | |
3431 | local_irq_enable(); | |
3432 | } | |
3433 | ||
3434 | static int process_backlog(struct napi_struct *napi, int quota) | |
3435 | { | |
3436 | int work = 0; | |
3437 | struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); | |
3438 | ||
3439 | #ifdef CONFIG_RPS | |
3440 | /* Check if we have pending ipi, its better to send them now, | |
3441 | * not waiting net_rx_action() end. | |
3442 | */ | |
3443 | if (sd->rps_ipi_list) { | |
3444 | local_irq_disable(); | |
3445 | net_rps_action_and_irq_enable(sd); | |
3446 | } | |
3447 | #endif | |
3448 | napi->weight = weight_p; | |
3449 | local_irq_disable(); | |
3450 | while (work < quota) { | |
3451 | struct sk_buff *skb; | |
3452 | unsigned int qlen; | |
3453 | ||
3454 | while ((skb = __skb_dequeue(&sd->process_queue))) { | |
3455 | local_irq_enable(); | |
3456 | __netif_receive_skb(skb); | |
3457 | local_irq_disable(); | |
3458 | input_queue_head_incr(sd); | |
3459 | if (++work >= quota) { | |
3460 | local_irq_enable(); | |
3461 | return work; | |
3462 | } | |
3463 | } | |
3464 | ||
3465 | rps_lock(sd); | |
3466 | qlen = skb_queue_len(&sd->input_pkt_queue); | |
3467 | if (qlen) | |
3468 | skb_queue_splice_tail_init(&sd->input_pkt_queue, | |
3469 | &sd->process_queue); | |
3470 | ||
3471 | if (qlen < quota - work) { | |
3472 | /* | |
3473 | * Inline a custom version of __napi_complete(). | |
3474 | * only current cpu owns and manipulates this napi, | |
3475 | * and NAPI_STATE_SCHED is the only possible flag set on backlog. | |
3476 | * we can use a plain write instead of clear_bit(), | |
3477 | * and we dont need an smp_mb() memory barrier. | |
3478 | */ | |
3479 | list_del(&napi->poll_list); | |
3480 | napi->state = 0; | |
3481 | ||
3482 | quota = work + qlen; | |
3483 | } | |
3484 | rps_unlock(sd); | |
3485 | } | |
3486 | local_irq_enable(); | |
3487 | ||
3488 | return work; | |
3489 | } | |
3490 | ||
3491 | /** | |
3492 | * __napi_schedule - schedule for receive | |
3493 | * @n: entry to schedule | |
3494 | * | |
3495 | * The entry's receive function will be scheduled to run | |
3496 | */ | |
3497 | void __napi_schedule(struct napi_struct *n) | |
3498 | { | |
3499 | unsigned long flags; | |
3500 | ||
3501 | local_irq_save(flags); | |
3502 | ____napi_schedule(&__get_cpu_var(softnet_data), n); | |
3503 | local_irq_restore(flags); | |
3504 | } | |
3505 | EXPORT_SYMBOL(__napi_schedule); | |
3506 | ||
3507 | void __napi_complete(struct napi_struct *n) | |
3508 | { | |
3509 | BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); | |
3510 | BUG_ON(n->gro_list); | |
3511 | ||
3512 | list_del(&n->poll_list); | |
3513 | smp_mb__before_clear_bit(); | |
3514 | clear_bit(NAPI_STATE_SCHED, &n->state); | |
3515 | } | |
3516 | EXPORT_SYMBOL(__napi_complete); | |
3517 | ||
3518 | void napi_complete(struct napi_struct *n) | |
3519 | { | |
3520 | unsigned long flags; | |
3521 | ||
3522 | /* | |
3523 | * don't let napi dequeue from the cpu poll list | |
3524 | * just in case its running on a different cpu | |
3525 | */ | |
3526 | if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state))) | |
3527 | return; | |
3528 | ||
3529 | napi_gro_flush(n); | |
3530 | local_irq_save(flags); | |
3531 | __napi_complete(n); | |
3532 | local_irq_restore(flags); | |
3533 | } | |
3534 | EXPORT_SYMBOL(napi_complete); | |
3535 | ||
3536 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, | |
3537 | int (*poll)(struct napi_struct *, int), int weight) | |
3538 | { | |
3539 | INIT_LIST_HEAD(&napi->poll_list); | |
3540 | napi->gro_count = 0; | |
3541 | napi->gro_list = NULL; | |
3542 | napi->skb = NULL; | |
3543 | napi->poll = poll; | |
3544 | napi->weight = weight; | |
3545 | list_add(&napi->dev_list, &dev->napi_list); | |
3546 | napi->dev = dev; | |
3547 | #ifdef CONFIG_NETPOLL | |
3548 | spin_lock_init(&napi->poll_lock); | |
3549 | napi->poll_owner = -1; | |
3550 | #endif | |
3551 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
3552 | } | |
3553 | EXPORT_SYMBOL(netif_napi_add); | |
3554 | ||
3555 | void netif_napi_del(struct napi_struct *napi) | |
3556 | { | |
3557 | struct sk_buff *skb, *next; | |
3558 | ||
3559 | list_del_init(&napi->dev_list); | |
3560 | napi_free_frags(napi); | |
3561 | ||
3562 | for (skb = napi->gro_list; skb; skb = next) { | |
3563 | next = skb->next; | |
3564 | skb->next = NULL; | |
3565 | kfree_skb(skb); | |
3566 | } | |
3567 | ||
3568 | napi->gro_list = NULL; | |
3569 | napi->gro_count = 0; | |
3570 | } | |
3571 | EXPORT_SYMBOL(netif_napi_del); | |
3572 | ||
3573 | static void net_rx_action(struct softirq_action *h) | |
3574 | { | |
3575 | struct softnet_data *sd = &__get_cpu_var(softnet_data); | |
3576 | unsigned long time_limit = jiffies + 2; | |
3577 | int budget = netdev_budget; | |
3578 | void *have; | |
3579 | ||
3580 | local_irq_disable(); | |
3581 | ||
3582 | while (!list_empty(&sd->poll_list)) { | |
3583 | struct napi_struct *n; | |
3584 | int work, weight; | |
3585 | ||
3586 | /* If softirq window is exhuasted then punt. | |
3587 | * Allow this to run for 2 jiffies since which will allow | |
3588 | * an average latency of 1.5/HZ. | |
3589 | */ | |
3590 | if (unlikely(budget <= 0 || time_after(jiffies, time_limit))) | |
3591 | goto softnet_break; | |
3592 | ||
3593 | local_irq_enable(); | |
3594 | ||
3595 | /* Even though interrupts have been re-enabled, this | |
3596 | * access is safe because interrupts can only add new | |
3597 | * entries to the tail of this list, and only ->poll() | |
3598 | * calls can remove this head entry from the list. | |
3599 | */ | |
3600 | n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list); | |
3601 | ||
3602 | have = netpoll_poll_lock(n); | |
3603 | ||
3604 | weight = n->weight; | |
3605 | ||
3606 | /* This NAPI_STATE_SCHED test is for avoiding a race | |
3607 | * with netpoll's poll_napi(). Only the entity which | |
3608 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
3609 | * actually make the ->poll() call. Therefore we avoid | |
3610 | * accidently calling ->poll() when NAPI is not scheduled. | |
3611 | */ | |
3612 | work = 0; | |
3613 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { | |
3614 | work = n->poll(n, weight); | |
3615 | trace_napi_poll(n); | |
3616 | } | |
3617 | ||
3618 | WARN_ON_ONCE(work > weight); | |
3619 | ||
3620 | budget -= work; | |
3621 | ||
3622 | local_irq_disable(); | |
3623 | ||
3624 | /* Drivers must not modify the NAPI state if they | |
3625 | * consume the entire weight. In such cases this code | |
3626 | * still "owns" the NAPI instance and therefore can | |
3627 | * move the instance around on the list at-will. | |
3628 | */ | |
3629 | if (unlikely(work == weight)) { | |
3630 | if (unlikely(napi_disable_pending(n))) { | |
3631 | local_irq_enable(); | |
3632 | napi_complete(n); | |
3633 | local_irq_disable(); | |
3634 | } else | |
3635 | list_move_tail(&n->poll_list, &sd->poll_list); | |
3636 | } | |
3637 | ||
3638 | netpoll_poll_unlock(have); | |
3639 | } | |
3640 | out: | |
3641 | net_rps_action_and_irq_enable(sd); | |
3642 | ||
3643 | #ifdef CONFIG_NET_DMA | |
3644 | /* | |
3645 | * There may not be any more sk_buffs coming right now, so push | |
3646 | * any pending DMA copies to hardware | |
3647 | */ | |
3648 | dma_issue_pending_all(); | |
3649 | #endif | |
3650 | ||
3651 | return; | |
3652 | ||
3653 | softnet_break: | |
3654 | sd->time_squeeze++; | |
3655 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3656 | goto out; | |
3657 | } | |
3658 | ||
3659 | static gifconf_func_t *gifconf_list[NPROTO]; | |
3660 | ||
3661 | /** | |
3662 | * register_gifconf - register a SIOCGIF handler | |
3663 | * @family: Address family | |
3664 | * @gifconf: Function handler | |
3665 | * | |
3666 | * Register protocol dependent address dumping routines. The handler | |
3667 | * that is passed must not be freed or reused until it has been replaced | |
3668 | * by another handler. | |
3669 | */ | |
3670 | int register_gifconf(unsigned int family, gifconf_func_t *gifconf) | |
3671 | { | |
3672 | if (family >= NPROTO) | |
3673 | return -EINVAL; | |
3674 | gifconf_list[family] = gifconf; | |
3675 | return 0; | |
3676 | } | |
3677 | EXPORT_SYMBOL(register_gifconf); | |
3678 | ||
3679 | ||
3680 | /* | |
3681 | * Map an interface index to its name (SIOCGIFNAME) | |
3682 | */ | |
3683 | ||
3684 | /* | |
3685 | * We need this ioctl for efficient implementation of the | |
3686 | * if_indextoname() function required by the IPv6 API. Without | |
3687 | * it, we would have to search all the interfaces to find a | |
3688 | * match. --pb | |
3689 | */ | |
3690 | ||
3691 | static int dev_ifname(struct net *net, struct ifreq __user *arg) | |
3692 | { | |
3693 | struct net_device *dev; | |
3694 | struct ifreq ifr; | |
3695 | ||
3696 | /* | |
3697 | * Fetch the caller's info block. | |
3698 | */ | |
3699 | ||
3700 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
3701 | return -EFAULT; | |
3702 | ||
3703 | rcu_read_lock(); | |
3704 | dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex); | |
3705 | if (!dev) { | |
3706 | rcu_read_unlock(); | |
3707 | return -ENODEV; | |
3708 | } | |
3709 | ||
3710 | strcpy(ifr.ifr_name, dev->name); | |
3711 | rcu_read_unlock(); | |
3712 | ||
3713 | if (copy_to_user(arg, &ifr, sizeof(struct ifreq))) | |
3714 | return -EFAULT; | |
3715 | return 0; | |
3716 | } | |
3717 | ||
3718 | /* | |
3719 | * Perform a SIOCGIFCONF call. This structure will change | |
3720 | * size eventually, and there is nothing I can do about it. | |
3721 | * Thus we will need a 'compatibility mode'. | |
3722 | */ | |
3723 | ||
3724 | static int dev_ifconf(struct net *net, char __user *arg) | |
3725 | { | |
3726 | struct ifconf ifc; | |
3727 | struct net_device *dev; | |
3728 | char __user *pos; | |
3729 | int len; | |
3730 | int total; | |
3731 | int i; | |
3732 | ||
3733 | /* | |
3734 | * Fetch the caller's info block. | |
3735 | */ | |
3736 | ||
3737 | if (copy_from_user(&ifc, arg, sizeof(struct ifconf))) | |
3738 | return -EFAULT; | |
3739 | ||
3740 | pos = ifc.ifc_buf; | |
3741 | len = ifc.ifc_len; | |
3742 | ||
3743 | /* | |
3744 | * Loop over the interfaces, and write an info block for each. | |
3745 | */ | |
3746 | ||
3747 | total = 0; | |
3748 | for_each_netdev(net, dev) { | |
3749 | for (i = 0; i < NPROTO; i++) { | |
3750 | if (gifconf_list[i]) { | |
3751 | int done; | |
3752 | if (!pos) | |
3753 | done = gifconf_list[i](dev, NULL, 0); | |
3754 | else | |
3755 | done = gifconf_list[i](dev, pos + total, | |
3756 | len - total); | |
3757 | if (done < 0) | |
3758 | return -EFAULT; | |
3759 | total += done; | |
3760 | } | |
3761 | } | |
3762 | } | |
3763 | ||
3764 | /* | |
3765 | * All done. Write the updated control block back to the caller. | |
3766 | */ | |
3767 | ifc.ifc_len = total; | |
3768 | ||
3769 | /* | |
3770 | * Both BSD and Solaris return 0 here, so we do too. | |
3771 | */ | |
3772 | return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0; | |
3773 | } | |
3774 | ||
3775 | #ifdef CONFIG_PROC_FS | |
3776 | /* | |
3777 | * This is invoked by the /proc filesystem handler to display a device | |
3778 | * in detail. | |
3779 | */ | |
3780 | void *dev_seq_start(struct seq_file *seq, loff_t *pos) | |
3781 | __acquires(RCU) | |
3782 | { | |
3783 | struct net *net = seq_file_net(seq); | |
3784 | loff_t off; | |
3785 | struct net_device *dev; | |
3786 | ||
3787 | rcu_read_lock(); | |
3788 | if (!*pos) | |
3789 | return SEQ_START_TOKEN; | |
3790 | ||
3791 | off = 1; | |
3792 | for_each_netdev_rcu(net, dev) | |
3793 | if (off++ == *pos) | |
3794 | return dev; | |
3795 | ||
3796 | return NULL; | |
3797 | } | |
3798 | ||
3799 | void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
3800 | { | |
3801 | struct net_device *dev = (v == SEQ_START_TOKEN) ? | |
3802 | first_net_device(seq_file_net(seq)) : | |
3803 | next_net_device((struct net_device *)v); | |
3804 | ||
3805 | ++*pos; | |
3806 | return rcu_dereference(dev); | |
3807 | } | |
3808 | ||
3809 | void dev_seq_stop(struct seq_file *seq, void *v) | |
3810 | __releases(RCU) | |
3811 | { | |
3812 | rcu_read_unlock(); | |
3813 | } | |
3814 | ||
3815 | static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev) | |
3816 | { | |
3817 | struct rtnl_link_stats64 temp; | |
3818 | const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp); | |
3819 | ||
3820 | seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu " | |
3821 | "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n", | |
3822 | dev->name, stats->rx_bytes, stats->rx_packets, | |
3823 | stats->rx_errors, | |
3824 | stats->rx_dropped + stats->rx_missed_errors, | |
3825 | stats->rx_fifo_errors, | |
3826 | stats->rx_length_errors + stats->rx_over_errors + | |
3827 | stats->rx_crc_errors + stats->rx_frame_errors, | |
3828 | stats->rx_compressed, stats->multicast, | |
3829 | stats->tx_bytes, stats->tx_packets, | |
3830 | stats->tx_errors, stats->tx_dropped, | |
3831 | stats->tx_fifo_errors, stats->collisions, | |
3832 | stats->tx_carrier_errors + | |
3833 | stats->tx_aborted_errors + | |
3834 | stats->tx_window_errors + | |
3835 | stats->tx_heartbeat_errors, | |
3836 | stats->tx_compressed); | |
3837 | } | |
3838 | ||
3839 | /* | |
3840 | * Called from the PROCfs module. This now uses the new arbitrary sized | |
3841 | * /proc/net interface to create /proc/net/dev | |
3842 | */ | |
3843 | static int dev_seq_show(struct seq_file *seq, void *v) | |
3844 | { | |
3845 | if (v == SEQ_START_TOKEN) | |
3846 | seq_puts(seq, "Inter-| Receive " | |
3847 | " | Transmit\n" | |
3848 | " face |bytes packets errs drop fifo frame " | |
3849 | "compressed multicast|bytes packets errs " | |
3850 | "drop fifo colls carrier compressed\n"); | |
3851 | else | |
3852 | dev_seq_printf_stats(seq, v); | |
3853 | return 0; | |
3854 | } | |
3855 | ||
3856 | static struct softnet_data *softnet_get_online(loff_t *pos) | |
3857 | { | |
3858 | struct softnet_data *sd = NULL; | |
3859 | ||
3860 | while (*pos < nr_cpu_ids) | |
3861 | if (cpu_online(*pos)) { | |
3862 | sd = &per_cpu(softnet_data, *pos); | |
3863 | break; | |
3864 | } else | |
3865 | ++*pos; | |
3866 | return sd; | |
3867 | } | |
3868 | ||
3869 | static void *softnet_seq_start(struct seq_file *seq, loff_t *pos) | |
3870 | { | |
3871 | return softnet_get_online(pos); | |
3872 | } | |
3873 | ||
3874 | static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
3875 | { | |
3876 | ++*pos; | |
3877 | return softnet_get_online(pos); | |
3878 | } | |
3879 | ||
3880 | static void softnet_seq_stop(struct seq_file *seq, void *v) | |
3881 | { | |
3882 | } | |
3883 | ||
3884 | static int softnet_seq_show(struct seq_file *seq, void *v) | |
3885 | { | |
3886 | struct softnet_data *sd = v; | |
3887 | ||
3888 | seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n", | |
3889 | sd->processed, sd->dropped, sd->time_squeeze, 0, | |
3890 | 0, 0, 0, 0, /* was fastroute */ | |
3891 | sd->cpu_collision, sd->received_rps); | |
3892 | return 0; | |
3893 | } | |
3894 | ||
3895 | static const struct seq_operations dev_seq_ops = { | |
3896 | .start = dev_seq_start, | |
3897 | .next = dev_seq_next, | |
3898 | .stop = dev_seq_stop, | |
3899 | .show = dev_seq_show, | |
3900 | }; | |
3901 | ||
3902 | static int dev_seq_open(struct inode *inode, struct file *file) | |
3903 | { | |
3904 | return seq_open_net(inode, file, &dev_seq_ops, | |
3905 | sizeof(struct seq_net_private)); | |
3906 | } | |
3907 | ||
3908 | static const struct file_operations dev_seq_fops = { | |
3909 | .owner = THIS_MODULE, | |
3910 | .open = dev_seq_open, | |
3911 | .read = seq_read, | |
3912 | .llseek = seq_lseek, | |
3913 | .release = seq_release_net, | |
3914 | }; | |
3915 | ||
3916 | static const struct seq_operations softnet_seq_ops = { | |
3917 | .start = softnet_seq_start, | |
3918 | .next = softnet_seq_next, | |
3919 | .stop = softnet_seq_stop, | |
3920 | .show = softnet_seq_show, | |
3921 | }; | |
3922 | ||
3923 | static int softnet_seq_open(struct inode *inode, struct file *file) | |
3924 | { | |
3925 | return seq_open(file, &softnet_seq_ops); | |
3926 | } | |
3927 | ||
3928 | static const struct file_operations softnet_seq_fops = { | |
3929 | .owner = THIS_MODULE, | |
3930 | .open = softnet_seq_open, | |
3931 | .read = seq_read, | |
3932 | .llseek = seq_lseek, | |
3933 | .release = seq_release, | |
3934 | }; | |
3935 | ||
3936 | static void *ptype_get_idx(loff_t pos) | |
3937 | { | |
3938 | struct packet_type *pt = NULL; | |
3939 | loff_t i = 0; | |
3940 | int t; | |
3941 | ||
3942 | list_for_each_entry_rcu(pt, &ptype_all, list) { | |
3943 | if (i == pos) | |
3944 | return pt; | |
3945 | ++i; | |
3946 | } | |
3947 | ||
3948 | for (t = 0; t < PTYPE_HASH_SIZE; t++) { | |
3949 | list_for_each_entry_rcu(pt, &ptype_base[t], list) { | |
3950 | if (i == pos) | |
3951 | return pt; | |
3952 | ++i; | |
3953 | } | |
3954 | } | |
3955 | return NULL; | |
3956 | } | |
3957 | ||
3958 | static void *ptype_seq_start(struct seq_file *seq, loff_t *pos) | |
3959 | __acquires(RCU) | |
3960 | { | |
3961 | rcu_read_lock(); | |
3962 | return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN; | |
3963 | } | |
3964 | ||
3965 | static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
3966 | { | |
3967 | struct packet_type *pt; | |
3968 | struct list_head *nxt; | |
3969 | int hash; | |
3970 | ||
3971 | ++*pos; | |
3972 | if (v == SEQ_START_TOKEN) | |
3973 | return ptype_get_idx(0); | |
3974 | ||
3975 | pt = v; | |
3976 | nxt = pt->list.next; | |
3977 | if (pt->type == htons(ETH_P_ALL)) { | |
3978 | if (nxt != &ptype_all) | |
3979 | goto found; | |
3980 | hash = 0; | |
3981 | nxt = ptype_base[0].next; | |
3982 | } else | |
3983 | hash = ntohs(pt->type) & PTYPE_HASH_MASK; | |
3984 | ||
3985 | while (nxt == &ptype_base[hash]) { | |
3986 | if (++hash >= PTYPE_HASH_SIZE) | |
3987 | return NULL; | |
3988 | nxt = ptype_base[hash].next; | |
3989 | } | |
3990 | found: | |
3991 | return list_entry(nxt, struct packet_type, list); | |
3992 | } | |
3993 | ||
3994 | static void ptype_seq_stop(struct seq_file *seq, void *v) | |
3995 | __releases(RCU) | |
3996 | { | |
3997 | rcu_read_unlock(); | |
3998 | } | |
3999 | ||
4000 | static int ptype_seq_show(struct seq_file *seq, void *v) | |
4001 | { | |
4002 | struct packet_type *pt = v; | |
4003 | ||
4004 | if (v == SEQ_START_TOKEN) | |
4005 | seq_puts(seq, "Type Device Function\n"); | |
4006 | else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) { | |
4007 | if (pt->type == htons(ETH_P_ALL)) | |
4008 | seq_puts(seq, "ALL "); | |
4009 | else | |
4010 | seq_printf(seq, "%04x", ntohs(pt->type)); | |
4011 | ||
4012 | seq_printf(seq, " %-8s %pF\n", | |
4013 | pt->dev ? pt->dev->name : "", pt->func); | |
4014 | } | |
4015 | ||
4016 | return 0; | |
4017 | } | |
4018 | ||
4019 | static const struct seq_operations ptype_seq_ops = { | |
4020 | .start = ptype_seq_start, | |
4021 | .next = ptype_seq_next, | |
4022 | .stop = ptype_seq_stop, | |
4023 | .show = ptype_seq_show, | |
4024 | }; | |
4025 | ||
4026 | static int ptype_seq_open(struct inode *inode, struct file *file) | |
4027 | { | |
4028 | return seq_open_net(inode, file, &ptype_seq_ops, | |
4029 | sizeof(struct seq_net_private)); | |
4030 | } | |
4031 | ||
4032 | static const struct file_operations ptype_seq_fops = { | |
4033 | .owner = THIS_MODULE, | |
4034 | .open = ptype_seq_open, | |
4035 | .read = seq_read, | |
4036 | .llseek = seq_lseek, | |
4037 | .release = seq_release_net, | |
4038 | }; | |
4039 | ||
4040 | ||
4041 | static int __net_init dev_proc_net_init(struct net *net) | |
4042 | { | |
4043 | int rc = -ENOMEM; | |
4044 | ||
4045 | if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops)) | |
4046 | goto out; | |
4047 | if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops)) | |
4048 | goto out_dev; | |
4049 | if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops)) | |
4050 | goto out_softnet; | |
4051 | ||
4052 | if (wext_proc_init(net)) | |
4053 | goto out_ptype; | |
4054 | rc = 0; | |
4055 | out: | |
4056 | return rc; | |
4057 | out_ptype: | |
4058 | proc_net_remove(net, "ptype"); | |
4059 | out_softnet: | |
4060 | proc_net_remove(net, "softnet_stat"); | |
4061 | out_dev: | |
4062 | proc_net_remove(net, "dev"); | |
4063 | goto out; | |
4064 | } | |
4065 | ||
4066 | static void __net_exit dev_proc_net_exit(struct net *net) | |
4067 | { | |
4068 | wext_proc_exit(net); | |
4069 | ||
4070 | proc_net_remove(net, "ptype"); | |
4071 | proc_net_remove(net, "softnet_stat"); | |
4072 | proc_net_remove(net, "dev"); | |
4073 | } | |
4074 | ||
4075 | static struct pernet_operations __net_initdata dev_proc_ops = { | |
4076 | .init = dev_proc_net_init, | |
4077 | .exit = dev_proc_net_exit, | |
4078 | }; | |
4079 | ||
4080 | static int __init dev_proc_init(void) | |
4081 | { | |
4082 | return register_pernet_subsys(&dev_proc_ops); | |
4083 | } | |
4084 | #else | |
4085 | #define dev_proc_init() 0 | |
4086 | #endif /* CONFIG_PROC_FS */ | |
4087 | ||
4088 | ||
4089 | /** | |
4090 | * netdev_set_master - set up master/slave pair | |
4091 | * @slave: slave device | |
4092 | * @master: new master device | |
4093 | * | |
4094 | * Changes the master device of the slave. Pass %NULL to break the | |
4095 | * bonding. The caller must hold the RTNL semaphore. On a failure | |
4096 | * a negative errno code is returned. On success the reference counts | |
4097 | * are adjusted, %RTM_NEWLINK is sent to the routing socket and the | |
4098 | * function returns zero. | |
4099 | */ | |
4100 | int netdev_set_master(struct net_device *slave, struct net_device *master) | |
4101 | { | |
4102 | struct net_device *old = slave->master; | |
4103 | ||
4104 | ASSERT_RTNL(); | |
4105 | ||
4106 | if (master) { | |
4107 | if (old) | |
4108 | return -EBUSY; | |
4109 | dev_hold(master); | |
4110 | } | |
4111 | ||
4112 | slave->master = master; | |
4113 | ||
4114 | if (old) { | |
4115 | synchronize_net(); | |
4116 | dev_put(old); | |
4117 | } | |
4118 | if (master) | |
4119 | slave->flags |= IFF_SLAVE; | |
4120 | else | |
4121 | slave->flags &= ~IFF_SLAVE; | |
4122 | ||
4123 | rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE); | |
4124 | return 0; | |
4125 | } | |
4126 | EXPORT_SYMBOL(netdev_set_master); | |
4127 | ||
4128 | static void dev_change_rx_flags(struct net_device *dev, int flags) | |
4129 | { | |
4130 | const struct net_device_ops *ops = dev->netdev_ops; | |
4131 | ||
4132 | if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags) | |
4133 | ops->ndo_change_rx_flags(dev, flags); | |
4134 | } | |
4135 | ||
4136 | static int __dev_set_promiscuity(struct net_device *dev, int inc) | |
4137 | { | |
4138 | unsigned short old_flags = dev->flags; | |
4139 | uid_t uid; | |
4140 | gid_t gid; | |
4141 | ||
4142 | ASSERT_RTNL(); | |
4143 | ||
4144 | dev->flags |= IFF_PROMISC; | |
4145 | dev->promiscuity += inc; | |
4146 | if (dev->promiscuity == 0) { | |
4147 | /* | |
4148 | * Avoid overflow. | |
4149 | * If inc causes overflow, untouch promisc and return error. | |
4150 | */ | |
4151 | if (inc < 0) | |
4152 | dev->flags &= ~IFF_PROMISC; | |
4153 | else { | |
4154 | dev->promiscuity -= inc; | |
4155 | printk(KERN_WARNING "%s: promiscuity touches roof, " | |
4156 | "set promiscuity failed, promiscuity feature " | |
4157 | "of device might be broken.\n", dev->name); | |
4158 | return -EOVERFLOW; | |
4159 | } | |
4160 | } | |
4161 | if (dev->flags != old_flags) { | |
4162 | printk(KERN_INFO "device %s %s promiscuous mode\n", | |
4163 | dev->name, (dev->flags & IFF_PROMISC) ? "entered" : | |
4164 | "left"); | |
4165 | if (audit_enabled) { | |
4166 | current_uid_gid(&uid, &gid); | |
4167 | audit_log(current->audit_context, GFP_ATOMIC, | |
4168 | AUDIT_ANOM_PROMISCUOUS, | |
4169 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
4170 | dev->name, (dev->flags & IFF_PROMISC), | |
4171 | (old_flags & IFF_PROMISC), | |
4172 | audit_get_loginuid(current), | |
4173 | uid, gid, | |
4174 | audit_get_sessionid(current)); | |
4175 | } | |
4176 | ||
4177 | dev_change_rx_flags(dev, IFF_PROMISC); | |
4178 | } | |
4179 | return 0; | |
4180 | } | |
4181 | ||
4182 | /** | |
4183 | * dev_set_promiscuity - update promiscuity count on a device | |
4184 | * @dev: device | |
4185 | * @inc: modifier | |
4186 | * | |
4187 | * Add or remove promiscuity from a device. While the count in the device | |
4188 | * remains above zero the interface remains promiscuous. Once it hits zero | |
4189 | * the device reverts back to normal filtering operation. A negative inc | |
4190 | * value is used to drop promiscuity on the device. | |
4191 | * Return 0 if successful or a negative errno code on error. | |
4192 | */ | |
4193 | int dev_set_promiscuity(struct net_device *dev, int inc) | |
4194 | { | |
4195 | unsigned short old_flags = dev->flags; | |
4196 | int err; | |
4197 | ||
4198 | err = __dev_set_promiscuity(dev, inc); | |
4199 | if (err < 0) | |
4200 | return err; | |
4201 | if (dev->flags != old_flags) | |
4202 | dev_set_rx_mode(dev); | |
4203 | return err; | |
4204 | } | |
4205 | EXPORT_SYMBOL(dev_set_promiscuity); | |
4206 | ||
4207 | /** | |
4208 | * dev_set_allmulti - update allmulti count on a device | |
4209 | * @dev: device | |
4210 | * @inc: modifier | |
4211 | * | |
4212 | * Add or remove reception of all multicast frames to a device. While the | |
4213 | * count in the device remains above zero the interface remains listening | |
4214 | * to all interfaces. Once it hits zero the device reverts back to normal | |
4215 | * filtering operation. A negative @inc value is used to drop the counter | |
4216 | * when releasing a resource needing all multicasts. | |
4217 | * Return 0 if successful or a negative errno code on error. | |
4218 | */ | |
4219 | ||
4220 | int dev_set_allmulti(struct net_device *dev, int inc) | |
4221 | { | |
4222 | unsigned short old_flags = dev->flags; | |
4223 | ||
4224 | ASSERT_RTNL(); | |
4225 | ||
4226 | dev->flags |= IFF_ALLMULTI; | |
4227 | dev->allmulti += inc; | |
4228 | if (dev->allmulti == 0) { | |
4229 | /* | |
4230 | * Avoid overflow. | |
4231 | * If inc causes overflow, untouch allmulti and return error. | |
4232 | */ | |
4233 | if (inc < 0) | |
4234 | dev->flags &= ~IFF_ALLMULTI; | |
4235 | else { | |
4236 | dev->allmulti -= inc; | |
4237 | printk(KERN_WARNING "%s: allmulti touches roof, " | |
4238 | "set allmulti failed, allmulti feature of " | |
4239 | "device might be broken.\n", dev->name); | |
4240 | return -EOVERFLOW; | |
4241 | } | |
4242 | } | |
4243 | if (dev->flags ^ old_flags) { | |
4244 | dev_change_rx_flags(dev, IFF_ALLMULTI); | |
4245 | dev_set_rx_mode(dev); | |
4246 | } | |
4247 | return 0; | |
4248 | } | |
4249 | EXPORT_SYMBOL(dev_set_allmulti); | |
4250 | ||
4251 | /* | |
4252 | * Upload unicast and multicast address lists to device and | |
4253 | * configure RX filtering. When the device doesn't support unicast | |
4254 | * filtering it is put in promiscuous mode while unicast addresses | |
4255 | * are present. | |
4256 | */ | |
4257 | void __dev_set_rx_mode(struct net_device *dev) | |
4258 | { | |
4259 | const struct net_device_ops *ops = dev->netdev_ops; | |
4260 | ||
4261 | /* dev_open will call this function so the list will stay sane. */ | |
4262 | if (!(dev->flags&IFF_UP)) | |
4263 | return; | |
4264 | ||
4265 | if (!netif_device_present(dev)) | |
4266 | return; | |
4267 | ||
4268 | if (ops->ndo_set_rx_mode) | |
4269 | ops->ndo_set_rx_mode(dev); | |
4270 | else { | |
4271 | /* Unicast addresses changes may only happen under the rtnl, | |
4272 | * therefore calling __dev_set_promiscuity here is safe. | |
4273 | */ | |
4274 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { | |
4275 | __dev_set_promiscuity(dev, 1); | |
4276 | dev->uc_promisc = 1; | |
4277 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { | |
4278 | __dev_set_promiscuity(dev, -1); | |
4279 | dev->uc_promisc = 0; | |
4280 | } | |
4281 | ||
4282 | if (ops->ndo_set_multicast_list) | |
4283 | ops->ndo_set_multicast_list(dev); | |
4284 | } | |
4285 | } | |
4286 | ||
4287 | void dev_set_rx_mode(struct net_device *dev) | |
4288 | { | |
4289 | netif_addr_lock_bh(dev); | |
4290 | __dev_set_rx_mode(dev); | |
4291 | netif_addr_unlock_bh(dev); | |
4292 | } | |
4293 | ||
4294 | /** | |
4295 | * dev_get_flags - get flags reported to userspace | |
4296 | * @dev: device | |
4297 | * | |
4298 | * Get the combination of flag bits exported through APIs to userspace. | |
4299 | */ | |
4300 | unsigned dev_get_flags(const struct net_device *dev) | |
4301 | { | |
4302 | unsigned flags; | |
4303 | ||
4304 | flags = (dev->flags & ~(IFF_PROMISC | | |
4305 | IFF_ALLMULTI | | |
4306 | IFF_RUNNING | | |
4307 | IFF_LOWER_UP | | |
4308 | IFF_DORMANT)) | | |
4309 | (dev->gflags & (IFF_PROMISC | | |
4310 | IFF_ALLMULTI)); | |
4311 | ||
4312 | if (netif_running(dev)) { | |
4313 | if (netif_oper_up(dev)) | |
4314 | flags |= IFF_RUNNING; | |
4315 | if (netif_carrier_ok(dev)) | |
4316 | flags |= IFF_LOWER_UP; | |
4317 | if (netif_dormant(dev)) | |
4318 | flags |= IFF_DORMANT; | |
4319 | } | |
4320 | ||
4321 | return flags; | |
4322 | } | |
4323 | EXPORT_SYMBOL(dev_get_flags); | |
4324 | ||
4325 | int __dev_change_flags(struct net_device *dev, unsigned int flags) | |
4326 | { | |
4327 | int old_flags = dev->flags; | |
4328 | int ret; | |
4329 | ||
4330 | ASSERT_RTNL(); | |
4331 | ||
4332 | /* | |
4333 | * Set the flags on our device. | |
4334 | */ | |
4335 | ||
4336 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
4337 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
4338 | IFF_AUTOMEDIA)) | | |
4339 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
4340 | IFF_ALLMULTI)); | |
4341 | ||
4342 | /* | |
4343 | * Load in the correct multicast list now the flags have changed. | |
4344 | */ | |
4345 | ||
4346 | if ((old_flags ^ flags) & IFF_MULTICAST) | |
4347 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
4348 | ||
4349 | dev_set_rx_mode(dev); | |
4350 | ||
4351 | /* | |
4352 | * Have we downed the interface. We handle IFF_UP ourselves | |
4353 | * according to user attempts to set it, rather than blindly | |
4354 | * setting it. | |
4355 | */ | |
4356 | ||
4357 | ret = 0; | |
4358 | if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */ | |
4359 | ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev); | |
4360 | ||
4361 | if (!ret) | |
4362 | dev_set_rx_mode(dev); | |
4363 | } | |
4364 | ||
4365 | if ((flags ^ dev->gflags) & IFF_PROMISC) { | |
4366 | int inc = (flags & IFF_PROMISC) ? 1 : -1; | |
4367 | ||
4368 | dev->gflags ^= IFF_PROMISC; | |
4369 | dev_set_promiscuity(dev, inc); | |
4370 | } | |
4371 | ||
4372 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
4373 | is important. Some (broken) drivers set IFF_PROMISC, when | |
4374 | IFF_ALLMULTI is requested not asking us and not reporting. | |
4375 | */ | |
4376 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
4377 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; | |
4378 | ||
4379 | dev->gflags ^= IFF_ALLMULTI; | |
4380 | dev_set_allmulti(dev, inc); | |
4381 | } | |
4382 | ||
4383 | return ret; | |
4384 | } | |
4385 | ||
4386 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags) | |
4387 | { | |
4388 | unsigned int changes = dev->flags ^ old_flags; | |
4389 | ||
4390 | if (changes & IFF_UP) { | |
4391 | if (dev->flags & IFF_UP) | |
4392 | call_netdevice_notifiers(NETDEV_UP, dev); | |
4393 | else | |
4394 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
4395 | } | |
4396 | ||
4397 | if (dev->flags & IFF_UP && | |
4398 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) | |
4399 | call_netdevice_notifiers(NETDEV_CHANGE, dev); | |
4400 | } | |
4401 | ||
4402 | /** | |
4403 | * dev_change_flags - change device settings | |
4404 | * @dev: device | |
4405 | * @flags: device state flags | |
4406 | * | |
4407 | * Change settings on device based state flags. The flags are | |
4408 | * in the userspace exported format. | |
4409 | */ | |
4410 | int dev_change_flags(struct net_device *dev, unsigned flags) | |
4411 | { | |
4412 | int ret, changes; | |
4413 | int old_flags = dev->flags; | |
4414 | ||
4415 | ret = __dev_change_flags(dev, flags); | |
4416 | if (ret < 0) | |
4417 | return ret; | |
4418 | ||
4419 | changes = old_flags ^ dev->flags; | |
4420 | if (changes) | |
4421 | rtmsg_ifinfo(RTM_NEWLINK, dev, changes); | |
4422 | ||
4423 | __dev_notify_flags(dev, old_flags); | |
4424 | return ret; | |
4425 | } | |
4426 | EXPORT_SYMBOL(dev_change_flags); | |
4427 | ||
4428 | /** | |
4429 | * dev_set_mtu - Change maximum transfer unit | |
4430 | * @dev: device | |
4431 | * @new_mtu: new transfer unit | |
4432 | * | |
4433 | * Change the maximum transfer size of the network device. | |
4434 | */ | |
4435 | int dev_set_mtu(struct net_device *dev, int new_mtu) | |
4436 | { | |
4437 | const struct net_device_ops *ops = dev->netdev_ops; | |
4438 | int err; | |
4439 | ||
4440 | if (new_mtu == dev->mtu) | |
4441 | return 0; | |
4442 | ||
4443 | /* MTU must be positive. */ | |
4444 | if (new_mtu < 0) | |
4445 | return -EINVAL; | |
4446 | ||
4447 | if (!netif_device_present(dev)) | |
4448 | return -ENODEV; | |
4449 | ||
4450 | err = 0; | |
4451 | if (ops->ndo_change_mtu) | |
4452 | err = ops->ndo_change_mtu(dev, new_mtu); | |
4453 | else | |
4454 | dev->mtu = new_mtu; | |
4455 | ||
4456 | if (!err && dev->flags & IFF_UP) | |
4457 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); | |
4458 | return err; | |
4459 | } | |
4460 | EXPORT_SYMBOL(dev_set_mtu); | |
4461 | ||
4462 | /** | |
4463 | * dev_set_mac_address - Change Media Access Control Address | |
4464 | * @dev: device | |
4465 | * @sa: new address | |
4466 | * | |
4467 | * Change the hardware (MAC) address of the device | |
4468 | */ | |
4469 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) | |
4470 | { | |
4471 | const struct net_device_ops *ops = dev->netdev_ops; | |
4472 | int err; | |
4473 | ||
4474 | if (!ops->ndo_set_mac_address) | |
4475 | return -EOPNOTSUPP; | |
4476 | if (sa->sa_family != dev->type) | |
4477 | return -EINVAL; | |
4478 | if (!netif_device_present(dev)) | |
4479 | return -ENODEV; | |
4480 | err = ops->ndo_set_mac_address(dev, sa); | |
4481 | if (!err) | |
4482 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); | |
4483 | return err; | |
4484 | } | |
4485 | EXPORT_SYMBOL(dev_set_mac_address); | |
4486 | ||
4487 | /* | |
4488 | * Perform the SIOCxIFxxx calls, inside rcu_read_lock() | |
4489 | */ | |
4490 | static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd) | |
4491 | { | |
4492 | int err; | |
4493 | struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name); | |
4494 | ||
4495 | if (!dev) | |
4496 | return -ENODEV; | |
4497 | ||
4498 | switch (cmd) { | |
4499 | case SIOCGIFFLAGS: /* Get interface flags */ | |
4500 | ifr->ifr_flags = (short) dev_get_flags(dev); | |
4501 | return 0; | |
4502 | ||
4503 | case SIOCGIFMETRIC: /* Get the metric on the interface | |
4504 | (currently unused) */ | |
4505 | ifr->ifr_metric = 0; | |
4506 | return 0; | |
4507 | ||
4508 | case SIOCGIFMTU: /* Get the MTU of a device */ | |
4509 | ifr->ifr_mtu = dev->mtu; | |
4510 | return 0; | |
4511 | ||
4512 | case SIOCGIFHWADDR: | |
4513 | if (!dev->addr_len) | |
4514 | memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data); | |
4515 | else | |
4516 | memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr, | |
4517 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
4518 | ifr->ifr_hwaddr.sa_family = dev->type; | |
4519 | return 0; | |
4520 | ||
4521 | case SIOCGIFSLAVE: | |
4522 | err = -EINVAL; | |
4523 | break; | |
4524 | ||
4525 | case SIOCGIFMAP: | |
4526 | ifr->ifr_map.mem_start = dev->mem_start; | |
4527 | ifr->ifr_map.mem_end = dev->mem_end; | |
4528 | ifr->ifr_map.base_addr = dev->base_addr; | |
4529 | ifr->ifr_map.irq = dev->irq; | |
4530 | ifr->ifr_map.dma = dev->dma; | |
4531 | ifr->ifr_map.port = dev->if_port; | |
4532 | return 0; | |
4533 | ||
4534 | case SIOCGIFINDEX: | |
4535 | ifr->ifr_ifindex = dev->ifindex; | |
4536 | return 0; | |
4537 | ||
4538 | case SIOCGIFTXQLEN: | |
4539 | ifr->ifr_qlen = dev->tx_queue_len; | |
4540 | return 0; | |
4541 | ||
4542 | default: | |
4543 | /* dev_ioctl() should ensure this case | |
4544 | * is never reached | |
4545 | */ | |
4546 | WARN_ON(1); | |
4547 | err = -EINVAL; | |
4548 | break; | |
4549 | ||
4550 | } | |
4551 | return err; | |
4552 | } | |
4553 | ||
4554 | /* | |
4555 | * Perform the SIOCxIFxxx calls, inside rtnl_lock() | |
4556 | */ | |
4557 | static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd) | |
4558 | { | |
4559 | int err; | |
4560 | struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name); | |
4561 | const struct net_device_ops *ops; | |
4562 | ||
4563 | if (!dev) | |
4564 | return -ENODEV; | |
4565 | ||
4566 | ops = dev->netdev_ops; | |
4567 | ||
4568 | switch (cmd) { | |
4569 | case SIOCSIFFLAGS: /* Set interface flags */ | |
4570 | return dev_change_flags(dev, ifr->ifr_flags); | |
4571 | ||
4572 | case SIOCSIFMETRIC: /* Set the metric on the interface | |
4573 | (currently unused) */ | |
4574 | return -EOPNOTSUPP; | |
4575 | ||
4576 | case SIOCSIFMTU: /* Set the MTU of a device */ | |
4577 | return dev_set_mtu(dev, ifr->ifr_mtu); | |
4578 | ||
4579 | case SIOCSIFHWADDR: | |
4580 | return dev_set_mac_address(dev, &ifr->ifr_hwaddr); | |
4581 | ||
4582 | case SIOCSIFHWBROADCAST: | |
4583 | if (ifr->ifr_hwaddr.sa_family != dev->type) | |
4584 | return -EINVAL; | |
4585 | memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data, | |
4586 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
4587 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); | |
4588 | return 0; | |
4589 | ||
4590 | case SIOCSIFMAP: | |
4591 | if (ops->ndo_set_config) { | |
4592 | if (!netif_device_present(dev)) | |
4593 | return -ENODEV; | |
4594 | return ops->ndo_set_config(dev, &ifr->ifr_map); | |
4595 | } | |
4596 | return -EOPNOTSUPP; | |
4597 | ||
4598 | case SIOCADDMULTI: | |
4599 | if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) || | |
4600 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) | |
4601 | return -EINVAL; | |
4602 | if (!netif_device_present(dev)) | |
4603 | return -ENODEV; | |
4604 | return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data); | |
4605 | ||
4606 | case SIOCDELMULTI: | |
4607 | if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) || | |
4608 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) | |
4609 | return -EINVAL; | |
4610 | if (!netif_device_present(dev)) | |
4611 | return -ENODEV; | |
4612 | return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data); | |
4613 | ||
4614 | case SIOCSIFTXQLEN: | |
4615 | if (ifr->ifr_qlen < 0) | |
4616 | return -EINVAL; | |
4617 | dev->tx_queue_len = ifr->ifr_qlen; | |
4618 | return 0; | |
4619 | ||
4620 | case SIOCSIFNAME: | |
4621 | ifr->ifr_newname[IFNAMSIZ-1] = '\0'; | |
4622 | return dev_change_name(dev, ifr->ifr_newname); | |
4623 | ||
4624 | /* | |
4625 | * Unknown or private ioctl | |
4626 | */ | |
4627 | default: | |
4628 | if ((cmd >= SIOCDEVPRIVATE && | |
4629 | cmd <= SIOCDEVPRIVATE + 15) || | |
4630 | cmd == SIOCBONDENSLAVE || | |
4631 | cmd == SIOCBONDRELEASE || | |
4632 | cmd == SIOCBONDSETHWADDR || | |
4633 | cmd == SIOCBONDSLAVEINFOQUERY || | |
4634 | cmd == SIOCBONDINFOQUERY || | |
4635 | cmd == SIOCBONDCHANGEACTIVE || | |
4636 | cmd == SIOCGMIIPHY || | |
4637 | cmd == SIOCGMIIREG || | |
4638 | cmd == SIOCSMIIREG || | |
4639 | cmd == SIOCBRADDIF || | |
4640 | cmd == SIOCBRDELIF || | |
4641 | cmd == SIOCSHWTSTAMP || | |
4642 | cmd == SIOCWANDEV) { | |
4643 | err = -EOPNOTSUPP; | |
4644 | if (ops->ndo_do_ioctl) { | |
4645 | if (netif_device_present(dev)) | |
4646 | err = ops->ndo_do_ioctl(dev, ifr, cmd); | |
4647 | else | |
4648 | err = -ENODEV; | |
4649 | } | |
4650 | } else | |
4651 | err = -EINVAL; | |
4652 | ||
4653 | } | |
4654 | return err; | |
4655 | } | |
4656 | ||
4657 | /* | |
4658 | * This function handles all "interface"-type I/O control requests. The actual | |
4659 | * 'doing' part of this is dev_ifsioc above. | |
4660 | */ | |
4661 | ||
4662 | /** | |
4663 | * dev_ioctl - network device ioctl | |
4664 | * @net: the applicable net namespace | |
4665 | * @cmd: command to issue | |
4666 | * @arg: pointer to a struct ifreq in user space | |
4667 | * | |
4668 | * Issue ioctl functions to devices. This is normally called by the | |
4669 | * user space syscall interfaces but can sometimes be useful for | |
4670 | * other purposes. The return value is the return from the syscall if | |
4671 | * positive or a negative errno code on error. | |
4672 | */ | |
4673 | ||
4674 | int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg) | |
4675 | { | |
4676 | struct ifreq ifr; | |
4677 | int ret; | |
4678 | char *colon; | |
4679 | ||
4680 | /* One special case: SIOCGIFCONF takes ifconf argument | |
4681 | and requires shared lock, because it sleeps writing | |
4682 | to user space. | |
4683 | */ | |
4684 | ||
4685 | if (cmd == SIOCGIFCONF) { | |
4686 | rtnl_lock(); | |
4687 | ret = dev_ifconf(net, (char __user *) arg); | |
4688 | rtnl_unlock(); | |
4689 | return ret; | |
4690 | } | |
4691 | if (cmd == SIOCGIFNAME) | |
4692 | return dev_ifname(net, (struct ifreq __user *)arg); | |
4693 | ||
4694 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
4695 | return -EFAULT; | |
4696 | ||
4697 | ifr.ifr_name[IFNAMSIZ-1] = 0; | |
4698 | ||
4699 | colon = strchr(ifr.ifr_name, ':'); | |
4700 | if (colon) | |
4701 | *colon = 0; | |
4702 | ||
4703 | /* | |
4704 | * See which interface the caller is talking about. | |
4705 | */ | |
4706 | ||
4707 | switch (cmd) { | |
4708 | /* | |
4709 | * These ioctl calls: | |
4710 | * - can be done by all. | |
4711 | * - atomic and do not require locking. | |
4712 | * - return a value | |
4713 | */ | |
4714 | case SIOCGIFFLAGS: | |
4715 | case SIOCGIFMETRIC: | |
4716 | case SIOCGIFMTU: | |
4717 | case SIOCGIFHWADDR: | |
4718 | case SIOCGIFSLAVE: | |
4719 | case SIOCGIFMAP: | |
4720 | case SIOCGIFINDEX: | |
4721 | case SIOCGIFTXQLEN: | |
4722 | dev_load(net, ifr.ifr_name); | |
4723 | rcu_read_lock(); | |
4724 | ret = dev_ifsioc_locked(net, &ifr, cmd); | |
4725 | rcu_read_unlock(); | |
4726 | if (!ret) { | |
4727 | if (colon) | |
4728 | *colon = ':'; | |
4729 | if (copy_to_user(arg, &ifr, | |
4730 | sizeof(struct ifreq))) | |
4731 | ret = -EFAULT; | |
4732 | } | |
4733 | return ret; | |
4734 | ||
4735 | case SIOCETHTOOL: | |
4736 | dev_load(net, ifr.ifr_name); | |
4737 | rtnl_lock(); | |
4738 | ret = dev_ethtool(net, &ifr); | |
4739 | rtnl_unlock(); | |
4740 | if (!ret) { | |
4741 | if (colon) | |
4742 | *colon = ':'; | |
4743 | if (copy_to_user(arg, &ifr, | |
4744 | sizeof(struct ifreq))) | |
4745 | ret = -EFAULT; | |
4746 | } | |
4747 | return ret; | |
4748 | ||
4749 | /* | |
4750 | * These ioctl calls: | |
4751 | * - require superuser power. | |
4752 | * - require strict serialization. | |
4753 | * - return a value | |
4754 | */ | |
4755 | case SIOCGMIIPHY: | |
4756 | case SIOCGMIIREG: | |
4757 | case SIOCSIFNAME: | |
4758 | if (!capable(CAP_NET_ADMIN)) | |
4759 | return -EPERM; | |
4760 | dev_load(net, ifr.ifr_name); | |
4761 | rtnl_lock(); | |
4762 | ret = dev_ifsioc(net, &ifr, cmd); | |
4763 | rtnl_unlock(); | |
4764 | if (!ret) { | |
4765 | if (colon) | |
4766 | *colon = ':'; | |
4767 | if (copy_to_user(arg, &ifr, | |
4768 | sizeof(struct ifreq))) | |
4769 | ret = -EFAULT; | |
4770 | } | |
4771 | return ret; | |
4772 | ||
4773 | /* | |
4774 | * These ioctl calls: | |
4775 | * - require superuser power. | |
4776 | * - require strict serialization. | |
4777 | * - do not return a value | |
4778 | */ | |
4779 | case SIOCSIFFLAGS: | |
4780 | case SIOCSIFMETRIC: | |
4781 | case SIOCSIFMTU: | |
4782 | case SIOCSIFMAP: | |
4783 | case SIOCSIFHWADDR: | |
4784 | case SIOCSIFSLAVE: | |
4785 | case SIOCADDMULTI: | |
4786 | case SIOCDELMULTI: | |
4787 | case SIOCSIFHWBROADCAST: | |
4788 | case SIOCSIFTXQLEN: | |
4789 | case SIOCSMIIREG: | |
4790 | case SIOCBONDENSLAVE: | |
4791 | case SIOCBONDRELEASE: | |
4792 | case SIOCBONDSETHWADDR: | |
4793 | case SIOCBONDCHANGEACTIVE: | |
4794 | case SIOCBRADDIF: | |
4795 | case SIOCBRDELIF: | |
4796 | case SIOCSHWTSTAMP: | |
4797 | if (!capable(CAP_NET_ADMIN)) | |
4798 | return -EPERM; | |
4799 | /* fall through */ | |
4800 | case SIOCBONDSLAVEINFOQUERY: | |
4801 | case SIOCBONDINFOQUERY: | |
4802 | dev_load(net, ifr.ifr_name); | |
4803 | rtnl_lock(); | |
4804 | ret = dev_ifsioc(net, &ifr, cmd); | |
4805 | rtnl_unlock(); | |
4806 | return ret; | |
4807 | ||
4808 | case SIOCGIFMEM: | |
4809 | /* Get the per device memory space. We can add this but | |
4810 | * currently do not support it */ | |
4811 | case SIOCSIFMEM: | |
4812 | /* Set the per device memory buffer space. | |
4813 | * Not applicable in our case */ | |
4814 | case SIOCSIFLINK: | |
4815 | return -EINVAL; | |
4816 | ||
4817 | /* | |
4818 | * Unknown or private ioctl. | |
4819 | */ | |
4820 | default: | |
4821 | if (cmd == SIOCWANDEV || | |
4822 | (cmd >= SIOCDEVPRIVATE && | |
4823 | cmd <= SIOCDEVPRIVATE + 15)) { | |
4824 | dev_load(net, ifr.ifr_name); | |
4825 | rtnl_lock(); | |
4826 | ret = dev_ifsioc(net, &ifr, cmd); | |
4827 | rtnl_unlock(); | |
4828 | if (!ret && copy_to_user(arg, &ifr, | |
4829 | sizeof(struct ifreq))) | |
4830 | ret = -EFAULT; | |
4831 | return ret; | |
4832 | } | |
4833 | /* Take care of Wireless Extensions */ | |
4834 | if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) | |
4835 | return wext_handle_ioctl(net, &ifr, cmd, arg); | |
4836 | return -EINVAL; | |
4837 | } | |
4838 | } | |
4839 | ||
4840 | ||
4841 | /** | |
4842 | * dev_new_index - allocate an ifindex | |
4843 | * @net: the applicable net namespace | |
4844 | * | |
4845 | * Returns a suitable unique value for a new device interface | |
4846 | * number. The caller must hold the rtnl semaphore or the | |
4847 | * dev_base_lock to be sure it remains unique. | |
4848 | */ | |
4849 | static int dev_new_index(struct net *net) | |
4850 | { | |
4851 | static int ifindex; | |
4852 | for (;;) { | |
4853 | if (++ifindex <= 0) | |
4854 | ifindex = 1; | |
4855 | if (!__dev_get_by_index(net, ifindex)) | |
4856 | return ifindex; | |
4857 | } | |
4858 | } | |
4859 | ||
4860 | /* Delayed registration/unregisteration */ | |
4861 | static LIST_HEAD(net_todo_list); | |
4862 | ||
4863 | static void net_set_todo(struct net_device *dev) | |
4864 | { | |
4865 | list_add_tail(&dev->todo_list, &net_todo_list); | |
4866 | } | |
4867 | ||
4868 | static void rollback_registered_many(struct list_head *head) | |
4869 | { | |
4870 | struct net_device *dev, *tmp; | |
4871 | ||
4872 | BUG_ON(dev_boot_phase); | |
4873 | ASSERT_RTNL(); | |
4874 | ||
4875 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { | |
4876 | /* Some devices call without registering | |
4877 | * for initialization unwind. Remove those | |
4878 | * devices and proceed with the remaining. | |
4879 | */ | |
4880 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
4881 | pr_debug("unregister_netdevice: device %s/%p never " | |
4882 | "was registered\n", dev->name, dev); | |
4883 | ||
4884 | WARN_ON(1); | |
4885 | list_del(&dev->unreg_list); | |
4886 | continue; | |
4887 | } | |
4888 | ||
4889 | BUG_ON(dev->reg_state != NETREG_REGISTERED); | |
4890 | ||
4891 | /* If device is running, close it first. */ | |
4892 | dev_close(dev); | |
4893 | ||
4894 | /* And unlink it from device chain. */ | |
4895 | unlist_netdevice(dev); | |
4896 | ||
4897 | dev->reg_state = NETREG_UNREGISTERING; | |
4898 | } | |
4899 | ||
4900 | synchronize_net(); | |
4901 | ||
4902 | list_for_each_entry(dev, head, unreg_list) { | |
4903 | /* Shutdown queueing discipline. */ | |
4904 | dev_shutdown(dev); | |
4905 | ||
4906 | ||
4907 | /* Notify protocols, that we are about to destroy | |
4908 | this device. They should clean all the things. | |
4909 | */ | |
4910 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
4911 | ||
4912 | if (!dev->rtnl_link_ops || | |
4913 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
4914 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U); | |
4915 | ||
4916 | /* | |
4917 | * Flush the unicast and multicast chains | |
4918 | */ | |
4919 | dev_uc_flush(dev); | |
4920 | dev_mc_flush(dev); | |
4921 | ||
4922 | if (dev->netdev_ops->ndo_uninit) | |
4923 | dev->netdev_ops->ndo_uninit(dev); | |
4924 | ||
4925 | /* Notifier chain MUST detach us from master device. */ | |
4926 | WARN_ON(dev->master); | |
4927 | ||
4928 | /* Remove entries from kobject tree */ | |
4929 | netdev_unregister_kobject(dev); | |
4930 | } | |
4931 | ||
4932 | /* Process any work delayed until the end of the batch */ | |
4933 | dev = list_first_entry(head, struct net_device, unreg_list); | |
4934 | call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev); | |
4935 | ||
4936 | rcu_barrier(); | |
4937 | ||
4938 | list_for_each_entry(dev, head, unreg_list) | |
4939 | dev_put(dev); | |
4940 | } | |
4941 | ||
4942 | static void rollback_registered(struct net_device *dev) | |
4943 | { | |
4944 | LIST_HEAD(single); | |
4945 | ||
4946 | list_add(&dev->unreg_list, &single); | |
4947 | rollback_registered_many(&single); | |
4948 | } | |
4949 | ||
4950 | unsigned long netdev_fix_features(unsigned long features, const char *name) | |
4951 | { | |
4952 | /* Fix illegal SG+CSUM combinations. */ | |
4953 | if ((features & NETIF_F_SG) && | |
4954 | !(features & NETIF_F_ALL_CSUM)) { | |
4955 | if (name) | |
4956 | printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no " | |
4957 | "checksum feature.\n", name); | |
4958 | features &= ~NETIF_F_SG; | |
4959 | } | |
4960 | ||
4961 | /* TSO requires that SG is present as well. */ | |
4962 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) { | |
4963 | if (name) | |
4964 | printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no " | |
4965 | "SG feature.\n", name); | |
4966 | features &= ~NETIF_F_TSO; | |
4967 | } | |
4968 | ||
4969 | if (features & NETIF_F_UFO) { | |
4970 | if (!(features & NETIF_F_GEN_CSUM)) { | |
4971 | if (name) | |
4972 | printk(KERN_ERR "%s: Dropping NETIF_F_UFO " | |
4973 | "since no NETIF_F_HW_CSUM feature.\n", | |
4974 | name); | |
4975 | features &= ~NETIF_F_UFO; | |
4976 | } | |
4977 | ||
4978 | if (!(features & NETIF_F_SG)) { | |
4979 | if (name) | |
4980 | printk(KERN_ERR "%s: Dropping NETIF_F_UFO " | |
4981 | "since no NETIF_F_SG feature.\n", name); | |
4982 | features &= ~NETIF_F_UFO; | |
4983 | } | |
4984 | } | |
4985 | ||
4986 | return features; | |
4987 | } | |
4988 | EXPORT_SYMBOL(netdev_fix_features); | |
4989 | ||
4990 | /** | |
4991 | * netif_stacked_transfer_operstate - transfer operstate | |
4992 | * @rootdev: the root or lower level device to transfer state from | |
4993 | * @dev: the device to transfer operstate to | |
4994 | * | |
4995 | * Transfer operational state from root to device. This is normally | |
4996 | * called when a stacking relationship exists between the root | |
4997 | * device and the device(a leaf device). | |
4998 | */ | |
4999 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
5000 | struct net_device *dev) | |
5001 | { | |
5002 | if (rootdev->operstate == IF_OPER_DORMANT) | |
5003 | netif_dormant_on(dev); | |
5004 | else | |
5005 | netif_dormant_off(dev); | |
5006 | ||
5007 | if (netif_carrier_ok(rootdev)) { | |
5008 | if (!netif_carrier_ok(dev)) | |
5009 | netif_carrier_on(dev); | |
5010 | } else { | |
5011 | if (netif_carrier_ok(dev)) | |
5012 | netif_carrier_off(dev); | |
5013 | } | |
5014 | } | |
5015 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
5016 | ||
5017 | static int netif_alloc_rx_queues(struct net_device *dev) | |
5018 | { | |
5019 | #ifdef CONFIG_RPS | |
5020 | unsigned int i, count = dev->num_rx_queues; | |
5021 | struct netdev_rx_queue *rx; | |
5022 | ||
5023 | BUG_ON(count < 1); | |
5024 | ||
5025 | rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL); | |
5026 | if (!rx) { | |
5027 | pr_err("netdev: Unable to allocate %u rx queues.\n", count); | |
5028 | return -ENOMEM; | |
5029 | } | |
5030 | dev->_rx = rx; | |
5031 | ||
5032 | /* | |
5033 | * Set a pointer to first element in the array which holds the | |
5034 | * reference count. | |
5035 | */ | |
5036 | for (i = 0; i < count; i++) | |
5037 | rx[i].first = rx; | |
5038 | #endif | |
5039 | return 0; | |
5040 | } | |
5041 | ||
5042 | static int netif_alloc_netdev_queues(struct net_device *dev) | |
5043 | { | |
5044 | unsigned int count = dev->num_tx_queues; | |
5045 | struct netdev_queue *tx; | |
5046 | ||
5047 | BUG_ON(count < 1); | |
5048 | ||
5049 | tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL); | |
5050 | if (!tx) { | |
5051 | pr_err("netdev: Unable to allocate %u tx queues.\n", | |
5052 | count); | |
5053 | return -ENOMEM; | |
5054 | } | |
5055 | dev->_tx = tx; | |
5056 | return 0; | |
5057 | } | |
5058 | ||
5059 | static void netdev_init_one_queue(struct net_device *dev, | |
5060 | struct netdev_queue *queue, | |
5061 | void *_unused) | |
5062 | { | |
5063 | queue->dev = dev; | |
5064 | ||
5065 | /* Initialize queue lock */ | |
5066 | spin_lock_init(&queue->_xmit_lock); | |
5067 | netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); | |
5068 | queue->xmit_lock_owner = -1; | |
5069 | } | |
5070 | ||
5071 | static void netdev_init_queues(struct net_device *dev) | |
5072 | { | |
5073 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); | |
5074 | spin_lock_init(&dev->tx_global_lock); | |
5075 | } | |
5076 | ||
5077 | /** | |
5078 | * register_netdevice - register a network device | |
5079 | * @dev: device to register | |
5080 | * | |
5081 | * Take a completed network device structure and add it to the kernel | |
5082 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
5083 | * chain. 0 is returned on success. A negative errno code is returned | |
5084 | * on a failure to set up the device, or if the name is a duplicate. | |
5085 | * | |
5086 | * Callers must hold the rtnl semaphore. You may want | |
5087 | * register_netdev() instead of this. | |
5088 | * | |
5089 | * BUGS: | |
5090 | * The locking appears insufficient to guarantee two parallel registers | |
5091 | * will not get the same name. | |
5092 | */ | |
5093 | ||
5094 | int register_netdevice(struct net_device *dev) | |
5095 | { | |
5096 | int ret; | |
5097 | struct net *net = dev_net(dev); | |
5098 | ||
5099 | BUG_ON(dev_boot_phase); | |
5100 | ASSERT_RTNL(); | |
5101 | ||
5102 | might_sleep(); | |
5103 | ||
5104 | /* When net_device's are persistent, this will be fatal. */ | |
5105 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
5106 | BUG_ON(!net); | |
5107 | ||
5108 | spin_lock_init(&dev->addr_list_lock); | |
5109 | netdev_set_addr_lockdep_class(dev); | |
5110 | ||
5111 | dev->iflink = -1; | |
5112 | ||
5113 | ret = netif_alloc_rx_queues(dev); | |
5114 | if (ret) | |
5115 | goto out; | |
5116 | ||
5117 | ret = netif_alloc_netdev_queues(dev); | |
5118 | if (ret) | |
5119 | goto out; | |
5120 | ||
5121 | netdev_init_queues(dev); | |
5122 | ||
5123 | /* Init, if this function is available */ | |
5124 | if (dev->netdev_ops->ndo_init) { | |
5125 | ret = dev->netdev_ops->ndo_init(dev); | |
5126 | if (ret) { | |
5127 | if (ret > 0) | |
5128 | ret = -EIO; | |
5129 | goto out; | |
5130 | } | |
5131 | } | |
5132 | ||
5133 | ret = dev_get_valid_name(dev, dev->name, 0); | |
5134 | if (ret) | |
5135 | goto err_uninit; | |
5136 | ||
5137 | dev->ifindex = dev_new_index(net); | |
5138 | if (dev->iflink == -1) | |
5139 | dev->iflink = dev->ifindex; | |
5140 | ||
5141 | /* Fix illegal checksum combinations */ | |
5142 | if ((dev->features & NETIF_F_HW_CSUM) && | |
5143 | (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
5144 | printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n", | |
5145 | dev->name); | |
5146 | dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); | |
5147 | } | |
5148 | ||
5149 | if ((dev->features & NETIF_F_NO_CSUM) && | |
5150 | (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
5151 | printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n", | |
5152 | dev->name); | |
5153 | dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM); | |
5154 | } | |
5155 | ||
5156 | dev->features = netdev_fix_features(dev->features, dev->name); | |
5157 | ||
5158 | /* Enable software GSO if SG is supported. */ | |
5159 | if (dev->features & NETIF_F_SG) | |
5160 | dev->features |= NETIF_F_GSO; | |
5161 | ||
5162 | /* Enable GRO and NETIF_F_HIGHDMA for vlans by default, | |
5163 | * vlan_dev_init() will do the dev->features check, so these features | |
5164 | * are enabled only if supported by underlying device. | |
5165 | */ | |
5166 | dev->vlan_features |= (NETIF_F_GRO | NETIF_F_HIGHDMA); | |
5167 | ||
5168 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); | |
5169 | ret = notifier_to_errno(ret); | |
5170 | if (ret) | |
5171 | goto err_uninit; | |
5172 | ||
5173 | ret = netdev_register_kobject(dev); | |
5174 | if (ret) | |
5175 | goto err_uninit; | |
5176 | dev->reg_state = NETREG_REGISTERED; | |
5177 | ||
5178 | /* | |
5179 | * Default initial state at registry is that the | |
5180 | * device is present. | |
5181 | */ | |
5182 | ||
5183 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
5184 | ||
5185 | dev_init_scheduler(dev); | |
5186 | dev_hold(dev); | |
5187 | list_netdevice(dev); | |
5188 | ||
5189 | /* Notify protocols, that a new device appeared. */ | |
5190 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
5191 | ret = notifier_to_errno(ret); | |
5192 | if (ret) { | |
5193 | rollback_registered(dev); | |
5194 | dev->reg_state = NETREG_UNREGISTERED; | |
5195 | } | |
5196 | /* | |
5197 | * Prevent userspace races by waiting until the network | |
5198 | * device is fully setup before sending notifications. | |
5199 | */ | |
5200 | if (!dev->rtnl_link_ops || | |
5201 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
5202 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U); | |
5203 | ||
5204 | out: | |
5205 | return ret; | |
5206 | ||
5207 | err_uninit: | |
5208 | if (dev->netdev_ops->ndo_uninit) | |
5209 | dev->netdev_ops->ndo_uninit(dev); | |
5210 | goto out; | |
5211 | } | |
5212 | EXPORT_SYMBOL(register_netdevice); | |
5213 | ||
5214 | /** | |
5215 | * init_dummy_netdev - init a dummy network device for NAPI | |
5216 | * @dev: device to init | |
5217 | * | |
5218 | * This takes a network device structure and initialize the minimum | |
5219 | * amount of fields so it can be used to schedule NAPI polls without | |
5220 | * registering a full blown interface. This is to be used by drivers | |
5221 | * that need to tie several hardware interfaces to a single NAPI | |
5222 | * poll scheduler due to HW limitations. | |
5223 | */ | |
5224 | int init_dummy_netdev(struct net_device *dev) | |
5225 | { | |
5226 | /* Clear everything. Note we don't initialize spinlocks | |
5227 | * are they aren't supposed to be taken by any of the | |
5228 | * NAPI code and this dummy netdev is supposed to be | |
5229 | * only ever used for NAPI polls | |
5230 | */ | |
5231 | memset(dev, 0, sizeof(struct net_device)); | |
5232 | ||
5233 | /* make sure we BUG if trying to hit standard | |
5234 | * register/unregister code path | |
5235 | */ | |
5236 | dev->reg_state = NETREG_DUMMY; | |
5237 | ||
5238 | /* NAPI wants this */ | |
5239 | INIT_LIST_HEAD(&dev->napi_list); | |
5240 | ||
5241 | /* a dummy interface is started by default */ | |
5242 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
5243 | set_bit(__LINK_STATE_START, &dev->state); | |
5244 | ||
5245 | /* Note : We dont allocate pcpu_refcnt for dummy devices, | |
5246 | * because users of this 'device' dont need to change | |
5247 | * its refcount. | |
5248 | */ | |
5249 | ||
5250 | return 0; | |
5251 | } | |
5252 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
5253 | ||
5254 | ||
5255 | /** | |
5256 | * register_netdev - register a network device | |
5257 | * @dev: device to register | |
5258 | * | |
5259 | * Take a completed network device structure and add it to the kernel | |
5260 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
5261 | * chain. 0 is returned on success. A negative errno code is returned | |
5262 | * on a failure to set up the device, or if the name is a duplicate. | |
5263 | * | |
5264 | * This is a wrapper around register_netdevice that takes the rtnl semaphore | |
5265 | * and expands the device name if you passed a format string to | |
5266 | * alloc_netdev. | |
5267 | */ | |
5268 | int register_netdev(struct net_device *dev) | |
5269 | { | |
5270 | int err; | |
5271 | ||
5272 | rtnl_lock(); | |
5273 | ||
5274 | /* | |
5275 | * If the name is a format string the caller wants us to do a | |
5276 | * name allocation. | |
5277 | */ | |
5278 | if (strchr(dev->name, '%')) { | |
5279 | err = dev_alloc_name(dev, dev->name); | |
5280 | if (err < 0) | |
5281 | goto out; | |
5282 | } | |
5283 | ||
5284 | err = register_netdevice(dev); | |
5285 | out: | |
5286 | rtnl_unlock(); | |
5287 | return err; | |
5288 | } | |
5289 | EXPORT_SYMBOL(register_netdev); | |
5290 | ||
5291 | int netdev_refcnt_read(const struct net_device *dev) | |
5292 | { | |
5293 | int i, refcnt = 0; | |
5294 | ||
5295 | for_each_possible_cpu(i) | |
5296 | refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); | |
5297 | return refcnt; | |
5298 | } | |
5299 | EXPORT_SYMBOL(netdev_refcnt_read); | |
5300 | ||
5301 | /* | |
5302 | * netdev_wait_allrefs - wait until all references are gone. | |
5303 | * | |
5304 | * This is called when unregistering network devices. | |
5305 | * | |
5306 | * Any protocol or device that holds a reference should register | |
5307 | * for netdevice notification, and cleanup and put back the | |
5308 | * reference if they receive an UNREGISTER event. | |
5309 | * We can get stuck here if buggy protocols don't correctly | |
5310 | * call dev_put. | |
5311 | */ | |
5312 | static void netdev_wait_allrefs(struct net_device *dev) | |
5313 | { | |
5314 | unsigned long rebroadcast_time, warning_time; | |
5315 | int refcnt; | |
5316 | ||
5317 | linkwatch_forget_dev(dev); | |
5318 | ||
5319 | rebroadcast_time = warning_time = jiffies; | |
5320 | refcnt = netdev_refcnt_read(dev); | |
5321 | ||
5322 | while (refcnt != 0) { | |
5323 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { | |
5324 | rtnl_lock(); | |
5325 | ||
5326 | /* Rebroadcast unregister notification */ | |
5327 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
5328 | /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users | |
5329 | * should have already handle it the first time */ | |
5330 | ||
5331 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, | |
5332 | &dev->state)) { | |
5333 | /* We must not have linkwatch events | |
5334 | * pending on unregister. If this | |
5335 | * happens, we simply run the queue | |
5336 | * unscheduled, resulting in a noop | |
5337 | * for this device. | |
5338 | */ | |
5339 | linkwatch_run_queue(); | |
5340 | } | |
5341 | ||
5342 | __rtnl_unlock(); | |
5343 | ||
5344 | rebroadcast_time = jiffies; | |
5345 | } | |
5346 | ||
5347 | msleep(250); | |
5348 | ||
5349 | refcnt = netdev_refcnt_read(dev); | |
5350 | ||
5351 | if (time_after(jiffies, warning_time + 10 * HZ)) { | |
5352 | printk(KERN_EMERG "unregister_netdevice: " | |
5353 | "waiting for %s to become free. Usage " | |
5354 | "count = %d\n", | |
5355 | dev->name, refcnt); | |
5356 | warning_time = jiffies; | |
5357 | } | |
5358 | } | |
5359 | } | |
5360 | ||
5361 | /* The sequence is: | |
5362 | * | |
5363 | * rtnl_lock(); | |
5364 | * ... | |
5365 | * register_netdevice(x1); | |
5366 | * register_netdevice(x2); | |
5367 | * ... | |
5368 | * unregister_netdevice(y1); | |
5369 | * unregister_netdevice(y2); | |
5370 | * ... | |
5371 | * rtnl_unlock(); | |
5372 | * free_netdev(y1); | |
5373 | * free_netdev(y2); | |
5374 | * | |
5375 | * We are invoked by rtnl_unlock(). | |
5376 | * This allows us to deal with problems: | |
5377 | * 1) We can delete sysfs objects which invoke hotplug | |
5378 | * without deadlocking with linkwatch via keventd. | |
5379 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
5380 | * safely in order to wait for the netdev refcnt to drop to zero. | |
5381 | * | |
5382 | * We must not return until all unregister events added during | |
5383 | * the interval the lock was held have been completed. | |
5384 | */ | |
5385 | void netdev_run_todo(void) | |
5386 | { | |
5387 | struct list_head list; | |
5388 | ||
5389 | /* Snapshot list, allow later requests */ | |
5390 | list_replace_init(&net_todo_list, &list); | |
5391 | ||
5392 | __rtnl_unlock(); | |
5393 | ||
5394 | while (!list_empty(&list)) { | |
5395 | struct net_device *dev | |
5396 | = list_first_entry(&list, struct net_device, todo_list); | |
5397 | list_del(&dev->todo_list); | |
5398 | ||
5399 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { | |
5400 | printk(KERN_ERR "network todo '%s' but state %d\n", | |
5401 | dev->name, dev->reg_state); | |
5402 | dump_stack(); | |
5403 | continue; | |
5404 | } | |
5405 | ||
5406 | dev->reg_state = NETREG_UNREGISTERED; | |
5407 | ||
5408 | on_each_cpu(flush_backlog, dev, 1); | |
5409 | ||
5410 | netdev_wait_allrefs(dev); | |
5411 | ||
5412 | /* paranoia */ | |
5413 | BUG_ON(netdev_refcnt_read(dev)); | |
5414 | WARN_ON(rcu_dereference_raw(dev->ip_ptr)); | |
5415 | WARN_ON(rcu_dereference_raw(dev->ip6_ptr)); | |
5416 | WARN_ON(dev->dn_ptr); | |
5417 | ||
5418 | if (dev->destructor) | |
5419 | dev->destructor(dev); | |
5420 | ||
5421 | /* Free network device */ | |
5422 | kobject_put(&dev->dev.kobj); | |
5423 | } | |
5424 | } | |
5425 | ||
5426 | /** | |
5427 | * dev_txq_stats_fold - fold tx_queues stats | |
5428 | * @dev: device to get statistics from | |
5429 | * @stats: struct rtnl_link_stats64 to hold results | |
5430 | */ | |
5431 | void dev_txq_stats_fold(const struct net_device *dev, | |
5432 | struct rtnl_link_stats64 *stats) | |
5433 | { | |
5434 | u64 tx_bytes = 0, tx_packets = 0, tx_dropped = 0; | |
5435 | unsigned int i; | |
5436 | struct netdev_queue *txq; | |
5437 | ||
5438 | for (i = 0; i < dev->num_tx_queues; i++) { | |
5439 | txq = netdev_get_tx_queue(dev, i); | |
5440 | spin_lock_bh(&txq->_xmit_lock); | |
5441 | tx_bytes += txq->tx_bytes; | |
5442 | tx_packets += txq->tx_packets; | |
5443 | tx_dropped += txq->tx_dropped; | |
5444 | spin_unlock_bh(&txq->_xmit_lock); | |
5445 | } | |
5446 | if (tx_bytes || tx_packets || tx_dropped) { | |
5447 | stats->tx_bytes = tx_bytes; | |
5448 | stats->tx_packets = tx_packets; | |
5449 | stats->tx_dropped = tx_dropped; | |
5450 | } | |
5451 | } | |
5452 | EXPORT_SYMBOL(dev_txq_stats_fold); | |
5453 | ||
5454 | /* Convert net_device_stats to rtnl_link_stats64. They have the same | |
5455 | * fields in the same order, with only the type differing. | |
5456 | */ | |
5457 | static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, | |
5458 | const struct net_device_stats *netdev_stats) | |
5459 | { | |
5460 | #if BITS_PER_LONG == 64 | |
5461 | BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats)); | |
5462 | memcpy(stats64, netdev_stats, sizeof(*stats64)); | |
5463 | #else | |
5464 | size_t i, n = sizeof(*stats64) / sizeof(u64); | |
5465 | const unsigned long *src = (const unsigned long *)netdev_stats; | |
5466 | u64 *dst = (u64 *)stats64; | |
5467 | ||
5468 | BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) != | |
5469 | sizeof(*stats64) / sizeof(u64)); | |
5470 | for (i = 0; i < n; i++) | |
5471 | dst[i] = src[i]; | |
5472 | #endif | |
5473 | } | |
5474 | ||
5475 | /** | |
5476 | * dev_get_stats - get network device statistics | |
5477 | * @dev: device to get statistics from | |
5478 | * @storage: place to store stats | |
5479 | * | |
5480 | * Get network statistics from device. Return @storage. | |
5481 | * The device driver may provide its own method by setting | |
5482 | * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; | |
5483 | * otherwise the internal statistics structure is used. | |
5484 | */ | |
5485 | struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, | |
5486 | struct rtnl_link_stats64 *storage) | |
5487 | { | |
5488 | const struct net_device_ops *ops = dev->netdev_ops; | |
5489 | ||
5490 | if (ops->ndo_get_stats64) { | |
5491 | memset(storage, 0, sizeof(*storage)); | |
5492 | ops->ndo_get_stats64(dev, storage); | |
5493 | } else if (ops->ndo_get_stats) { | |
5494 | netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); | |
5495 | } else { | |
5496 | netdev_stats_to_stats64(storage, &dev->stats); | |
5497 | dev_txq_stats_fold(dev, storage); | |
5498 | } | |
5499 | storage->rx_dropped += atomic_long_read(&dev->rx_dropped); | |
5500 | return storage; | |
5501 | } | |
5502 | EXPORT_SYMBOL(dev_get_stats); | |
5503 | ||
5504 | struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) | |
5505 | { | |
5506 | struct netdev_queue *queue = dev_ingress_queue(dev); | |
5507 | ||
5508 | #ifdef CONFIG_NET_CLS_ACT | |
5509 | if (queue) | |
5510 | return queue; | |
5511 | queue = kzalloc(sizeof(*queue), GFP_KERNEL); | |
5512 | if (!queue) | |
5513 | return NULL; | |
5514 | netdev_init_one_queue(dev, queue, NULL); | |
5515 | queue->qdisc = &noop_qdisc; | |
5516 | queue->qdisc_sleeping = &noop_qdisc; | |
5517 | rcu_assign_pointer(dev->ingress_queue, queue); | |
5518 | #endif | |
5519 | return queue; | |
5520 | } | |
5521 | ||
5522 | /** | |
5523 | * alloc_netdev_mq - allocate network device | |
5524 | * @sizeof_priv: size of private data to allocate space for | |
5525 | * @name: device name format string | |
5526 | * @setup: callback to initialize device | |
5527 | * @queue_count: the number of subqueues to allocate | |
5528 | * | |
5529 | * Allocates a struct net_device with private data area for driver use | |
5530 | * and performs basic initialization. Also allocates subquue structs | |
5531 | * for each queue on the device at the end of the netdevice. | |
5532 | */ | |
5533 | struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name, | |
5534 | void (*setup)(struct net_device *), unsigned int queue_count) | |
5535 | { | |
5536 | struct net_device *dev; | |
5537 | size_t alloc_size; | |
5538 | struct net_device *p; | |
5539 | ||
5540 | BUG_ON(strlen(name) >= sizeof(dev->name)); | |
5541 | ||
5542 | if (queue_count < 1) { | |
5543 | pr_err("alloc_netdev: Unable to allocate device " | |
5544 | "with zero queues.\n"); | |
5545 | return NULL; | |
5546 | } | |
5547 | ||
5548 | alloc_size = sizeof(struct net_device); | |
5549 | if (sizeof_priv) { | |
5550 | /* ensure 32-byte alignment of private area */ | |
5551 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); | |
5552 | alloc_size += sizeof_priv; | |
5553 | } | |
5554 | /* ensure 32-byte alignment of whole construct */ | |
5555 | alloc_size += NETDEV_ALIGN - 1; | |
5556 | ||
5557 | p = kzalloc(alloc_size, GFP_KERNEL); | |
5558 | if (!p) { | |
5559 | printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n"); | |
5560 | return NULL; | |
5561 | } | |
5562 | ||
5563 | dev = PTR_ALIGN(p, NETDEV_ALIGN); | |
5564 | dev->padded = (char *)dev - (char *)p; | |
5565 | ||
5566 | dev->pcpu_refcnt = alloc_percpu(int); | |
5567 | if (!dev->pcpu_refcnt) | |
5568 | goto free_p; | |
5569 | ||
5570 | if (dev_addr_init(dev)) | |
5571 | goto free_pcpu; | |
5572 | ||
5573 | dev_mc_init(dev); | |
5574 | dev_uc_init(dev); | |
5575 | ||
5576 | dev_net_set(dev, &init_net); | |
5577 | ||
5578 | dev->num_tx_queues = queue_count; | |
5579 | dev->real_num_tx_queues = queue_count; | |
5580 | ||
5581 | #ifdef CONFIG_RPS | |
5582 | dev->num_rx_queues = queue_count; | |
5583 | dev->real_num_rx_queues = queue_count; | |
5584 | #endif | |
5585 | ||
5586 | dev->gso_max_size = GSO_MAX_SIZE; | |
5587 | ||
5588 | INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list); | |
5589 | dev->ethtool_ntuple_list.count = 0; | |
5590 | INIT_LIST_HEAD(&dev->napi_list); | |
5591 | INIT_LIST_HEAD(&dev->unreg_list); | |
5592 | INIT_LIST_HEAD(&dev->link_watch_list); | |
5593 | dev->priv_flags = IFF_XMIT_DST_RELEASE; | |
5594 | setup(dev); | |
5595 | strcpy(dev->name, name); | |
5596 | return dev; | |
5597 | ||
5598 | free_pcpu: | |
5599 | free_percpu(dev->pcpu_refcnt); | |
5600 | free_p: | |
5601 | kfree(p); | |
5602 | return NULL; | |
5603 | } | |
5604 | EXPORT_SYMBOL(alloc_netdev_mq); | |
5605 | ||
5606 | /** | |
5607 | * free_netdev - free network device | |
5608 | * @dev: device | |
5609 | * | |
5610 | * This function does the last stage of destroying an allocated device | |
5611 | * interface. The reference to the device object is released. | |
5612 | * If this is the last reference then it will be freed. | |
5613 | */ | |
5614 | void free_netdev(struct net_device *dev) | |
5615 | { | |
5616 | struct napi_struct *p, *n; | |
5617 | ||
5618 | release_net(dev_net(dev)); | |
5619 | ||
5620 | kfree(dev->_tx); | |
5621 | ||
5622 | kfree(rcu_dereference_raw(dev->ingress_queue)); | |
5623 | ||
5624 | /* Flush device addresses */ | |
5625 | dev_addr_flush(dev); | |
5626 | ||
5627 | /* Clear ethtool n-tuple list */ | |
5628 | ethtool_ntuple_flush(dev); | |
5629 | ||
5630 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) | |
5631 | netif_napi_del(p); | |
5632 | ||
5633 | free_percpu(dev->pcpu_refcnt); | |
5634 | dev->pcpu_refcnt = NULL; | |
5635 | ||
5636 | /* Compatibility with error handling in drivers */ | |
5637 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
5638 | kfree((char *)dev - dev->padded); | |
5639 | return; | |
5640 | } | |
5641 | ||
5642 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
5643 | dev->reg_state = NETREG_RELEASED; | |
5644 | ||
5645 | /* will free via device release */ | |
5646 | put_device(&dev->dev); | |
5647 | } | |
5648 | EXPORT_SYMBOL(free_netdev); | |
5649 | ||
5650 | /** | |
5651 | * synchronize_net - Synchronize with packet receive processing | |
5652 | * | |
5653 | * Wait for packets currently being received to be done. | |
5654 | * Does not block later packets from starting. | |
5655 | */ | |
5656 | void synchronize_net(void) | |
5657 | { | |
5658 | might_sleep(); | |
5659 | synchronize_rcu(); | |
5660 | } | |
5661 | EXPORT_SYMBOL(synchronize_net); | |
5662 | ||
5663 | /** | |
5664 | * unregister_netdevice_queue - remove device from the kernel | |
5665 | * @dev: device | |
5666 | * @head: list | |
5667 | * | |
5668 | * This function shuts down a device interface and removes it | |
5669 | * from the kernel tables. | |
5670 | * If head not NULL, device is queued to be unregistered later. | |
5671 | * | |
5672 | * Callers must hold the rtnl semaphore. You may want | |
5673 | * unregister_netdev() instead of this. | |
5674 | */ | |
5675 | ||
5676 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) | |
5677 | { | |
5678 | ASSERT_RTNL(); | |
5679 | ||
5680 | if (head) { | |
5681 | list_move_tail(&dev->unreg_list, head); | |
5682 | } else { | |
5683 | rollback_registered(dev); | |
5684 | /* Finish processing unregister after unlock */ | |
5685 | net_set_todo(dev); | |
5686 | } | |
5687 | } | |
5688 | EXPORT_SYMBOL(unregister_netdevice_queue); | |
5689 | ||
5690 | /** | |
5691 | * unregister_netdevice_many - unregister many devices | |
5692 | * @head: list of devices | |
5693 | */ | |
5694 | void unregister_netdevice_many(struct list_head *head) | |
5695 | { | |
5696 | struct net_device *dev; | |
5697 | ||
5698 | if (!list_empty(head)) { | |
5699 | rollback_registered_many(head); | |
5700 | list_for_each_entry(dev, head, unreg_list) | |
5701 | net_set_todo(dev); | |
5702 | } | |
5703 | } | |
5704 | EXPORT_SYMBOL(unregister_netdevice_many); | |
5705 | ||
5706 | /** | |
5707 | * unregister_netdev - remove device from the kernel | |
5708 | * @dev: device | |
5709 | * | |
5710 | * This function shuts down a device interface and removes it | |
5711 | * from the kernel tables. | |
5712 | * | |
5713 | * This is just a wrapper for unregister_netdevice that takes | |
5714 | * the rtnl semaphore. In general you want to use this and not | |
5715 | * unregister_netdevice. | |
5716 | */ | |
5717 | void unregister_netdev(struct net_device *dev) | |
5718 | { | |
5719 | rtnl_lock(); | |
5720 | unregister_netdevice(dev); | |
5721 | rtnl_unlock(); | |
5722 | } | |
5723 | EXPORT_SYMBOL(unregister_netdev); | |
5724 | ||
5725 | /** | |
5726 | * dev_change_net_namespace - move device to different nethost namespace | |
5727 | * @dev: device | |
5728 | * @net: network namespace | |
5729 | * @pat: If not NULL name pattern to try if the current device name | |
5730 | * is already taken in the destination network namespace. | |
5731 | * | |
5732 | * This function shuts down a device interface and moves it | |
5733 | * to a new network namespace. On success 0 is returned, on | |
5734 | * a failure a netagive errno code is returned. | |
5735 | * | |
5736 | * Callers must hold the rtnl semaphore. | |
5737 | */ | |
5738 | ||
5739 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
5740 | { | |
5741 | int err; | |
5742 | ||
5743 | ASSERT_RTNL(); | |
5744 | ||
5745 | /* Don't allow namespace local devices to be moved. */ | |
5746 | err = -EINVAL; | |
5747 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
5748 | goto out; | |
5749 | ||
5750 | /* Ensure the device has been registrered */ | |
5751 | err = -EINVAL; | |
5752 | if (dev->reg_state != NETREG_REGISTERED) | |
5753 | goto out; | |
5754 | ||
5755 | /* Get out if there is nothing todo */ | |
5756 | err = 0; | |
5757 | if (net_eq(dev_net(dev), net)) | |
5758 | goto out; | |
5759 | ||
5760 | /* Pick the destination device name, and ensure | |
5761 | * we can use it in the destination network namespace. | |
5762 | */ | |
5763 | err = -EEXIST; | |
5764 | if (__dev_get_by_name(net, dev->name)) { | |
5765 | /* We get here if we can't use the current device name */ | |
5766 | if (!pat) | |
5767 | goto out; | |
5768 | if (dev_get_valid_name(dev, pat, 1)) | |
5769 | goto out; | |
5770 | } | |
5771 | ||
5772 | /* | |
5773 | * And now a mini version of register_netdevice unregister_netdevice. | |
5774 | */ | |
5775 | ||
5776 | /* If device is running close it first. */ | |
5777 | dev_close(dev); | |
5778 | ||
5779 | /* And unlink it from device chain */ | |
5780 | err = -ENODEV; | |
5781 | unlist_netdevice(dev); | |
5782 | ||
5783 | synchronize_net(); | |
5784 | ||
5785 | /* Shutdown queueing discipline. */ | |
5786 | dev_shutdown(dev); | |
5787 | ||
5788 | /* Notify protocols, that we are about to destroy | |
5789 | this device. They should clean all the things. | |
5790 | ||
5791 | Note that dev->reg_state stays at NETREG_REGISTERED. | |
5792 | This is wanted because this way 8021q and macvlan know | |
5793 | the device is just moving and can keep their slaves up. | |
5794 | */ | |
5795 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
5796 | call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev); | |
5797 | ||
5798 | /* | |
5799 | * Flush the unicast and multicast chains | |
5800 | */ | |
5801 | dev_uc_flush(dev); | |
5802 | dev_mc_flush(dev); | |
5803 | ||
5804 | /* Actually switch the network namespace */ | |
5805 | dev_net_set(dev, net); | |
5806 | ||
5807 | /* If there is an ifindex conflict assign a new one */ | |
5808 | if (__dev_get_by_index(net, dev->ifindex)) { | |
5809 | int iflink = (dev->iflink == dev->ifindex); | |
5810 | dev->ifindex = dev_new_index(net); | |
5811 | if (iflink) | |
5812 | dev->iflink = dev->ifindex; | |
5813 | } | |
5814 | ||
5815 | /* Fixup kobjects */ | |
5816 | err = device_rename(&dev->dev, dev->name); | |
5817 | WARN_ON(err); | |
5818 | ||
5819 | /* Add the device back in the hashes */ | |
5820 | list_netdevice(dev); | |
5821 | ||
5822 | /* Notify protocols, that a new device appeared. */ | |
5823 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
5824 | ||
5825 | /* | |
5826 | * Prevent userspace races by waiting until the network | |
5827 | * device is fully setup before sending notifications. | |
5828 | */ | |
5829 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U); | |
5830 | ||
5831 | synchronize_net(); | |
5832 | err = 0; | |
5833 | out: | |
5834 | return err; | |
5835 | } | |
5836 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); | |
5837 | ||
5838 | static int dev_cpu_callback(struct notifier_block *nfb, | |
5839 | unsigned long action, | |
5840 | void *ocpu) | |
5841 | { | |
5842 | struct sk_buff **list_skb; | |
5843 | struct sk_buff *skb; | |
5844 | unsigned int cpu, oldcpu = (unsigned long)ocpu; | |
5845 | struct softnet_data *sd, *oldsd; | |
5846 | ||
5847 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) | |
5848 | return NOTIFY_OK; | |
5849 | ||
5850 | local_irq_disable(); | |
5851 | cpu = smp_processor_id(); | |
5852 | sd = &per_cpu(softnet_data, cpu); | |
5853 | oldsd = &per_cpu(softnet_data, oldcpu); | |
5854 | ||
5855 | /* Find end of our completion_queue. */ | |
5856 | list_skb = &sd->completion_queue; | |
5857 | while (*list_skb) | |
5858 | list_skb = &(*list_skb)->next; | |
5859 | /* Append completion queue from offline CPU. */ | |
5860 | *list_skb = oldsd->completion_queue; | |
5861 | oldsd->completion_queue = NULL; | |
5862 | ||
5863 | /* Append output queue from offline CPU. */ | |
5864 | if (oldsd->output_queue) { | |
5865 | *sd->output_queue_tailp = oldsd->output_queue; | |
5866 | sd->output_queue_tailp = oldsd->output_queue_tailp; | |
5867 | oldsd->output_queue = NULL; | |
5868 | oldsd->output_queue_tailp = &oldsd->output_queue; | |
5869 | } | |
5870 | ||
5871 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
5872 | local_irq_enable(); | |
5873 | ||
5874 | /* Process offline CPU's input_pkt_queue */ | |
5875 | while ((skb = __skb_dequeue(&oldsd->process_queue))) { | |
5876 | netif_rx(skb); | |
5877 | input_queue_head_incr(oldsd); | |
5878 | } | |
5879 | while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) { | |
5880 | netif_rx(skb); | |
5881 | input_queue_head_incr(oldsd); | |
5882 | } | |
5883 | ||
5884 | return NOTIFY_OK; | |
5885 | } | |
5886 | ||
5887 | ||
5888 | /** | |
5889 | * netdev_increment_features - increment feature set by one | |
5890 | * @all: current feature set | |
5891 | * @one: new feature set | |
5892 | * @mask: mask feature set | |
5893 | * | |
5894 | * Computes a new feature set after adding a device with feature set | |
5895 | * @one to the master device with current feature set @all. Will not | |
5896 | * enable anything that is off in @mask. Returns the new feature set. | |
5897 | */ | |
5898 | unsigned long netdev_increment_features(unsigned long all, unsigned long one, | |
5899 | unsigned long mask) | |
5900 | { | |
5901 | /* If device needs checksumming, downgrade to it. */ | |
5902 | if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM)) | |
5903 | all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM); | |
5904 | else if (mask & NETIF_F_ALL_CSUM) { | |
5905 | /* If one device supports v4/v6 checksumming, set for all. */ | |
5906 | if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) && | |
5907 | !(all & NETIF_F_GEN_CSUM)) { | |
5908 | all &= ~NETIF_F_ALL_CSUM; | |
5909 | all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); | |
5910 | } | |
5911 | ||
5912 | /* If one device supports hw checksumming, set for all. */ | |
5913 | if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) { | |
5914 | all &= ~NETIF_F_ALL_CSUM; | |
5915 | all |= NETIF_F_HW_CSUM; | |
5916 | } | |
5917 | } | |
5918 | ||
5919 | one |= NETIF_F_ALL_CSUM; | |
5920 | ||
5921 | one |= all & NETIF_F_ONE_FOR_ALL; | |
5922 | all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO; | |
5923 | all |= one & mask & NETIF_F_ONE_FOR_ALL; | |
5924 | ||
5925 | return all; | |
5926 | } | |
5927 | EXPORT_SYMBOL(netdev_increment_features); | |
5928 | ||
5929 | static struct hlist_head *netdev_create_hash(void) | |
5930 | { | |
5931 | int i; | |
5932 | struct hlist_head *hash; | |
5933 | ||
5934 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
5935 | if (hash != NULL) | |
5936 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
5937 | INIT_HLIST_HEAD(&hash[i]); | |
5938 | ||
5939 | return hash; | |
5940 | } | |
5941 | ||
5942 | /* Initialize per network namespace state */ | |
5943 | static int __net_init netdev_init(struct net *net) | |
5944 | { | |
5945 | INIT_LIST_HEAD(&net->dev_base_head); | |
5946 | ||
5947 | net->dev_name_head = netdev_create_hash(); | |
5948 | if (net->dev_name_head == NULL) | |
5949 | goto err_name; | |
5950 | ||
5951 | net->dev_index_head = netdev_create_hash(); | |
5952 | if (net->dev_index_head == NULL) | |
5953 | goto err_idx; | |
5954 | ||
5955 | return 0; | |
5956 | ||
5957 | err_idx: | |
5958 | kfree(net->dev_name_head); | |
5959 | err_name: | |
5960 | return -ENOMEM; | |
5961 | } | |
5962 | ||
5963 | /** | |
5964 | * netdev_drivername - network driver for the device | |
5965 | * @dev: network device | |
5966 | * @buffer: buffer for resulting name | |
5967 | * @len: size of buffer | |
5968 | * | |
5969 | * Determine network driver for device. | |
5970 | */ | |
5971 | char *netdev_drivername(const struct net_device *dev, char *buffer, int len) | |
5972 | { | |
5973 | const struct device_driver *driver; | |
5974 | const struct device *parent; | |
5975 | ||
5976 | if (len <= 0 || !buffer) | |
5977 | return buffer; | |
5978 | buffer[0] = 0; | |
5979 | ||
5980 | parent = dev->dev.parent; | |
5981 | ||
5982 | if (!parent) | |
5983 | return buffer; | |
5984 | ||
5985 | driver = parent->driver; | |
5986 | if (driver && driver->name) | |
5987 | strlcpy(buffer, driver->name, len); | |
5988 | return buffer; | |
5989 | } | |
5990 | ||
5991 | static int __netdev_printk(const char *level, const struct net_device *dev, | |
5992 | struct va_format *vaf) | |
5993 | { | |
5994 | int r; | |
5995 | ||
5996 | if (dev && dev->dev.parent) | |
5997 | r = dev_printk(level, dev->dev.parent, "%s: %pV", | |
5998 | netdev_name(dev), vaf); | |
5999 | else if (dev) | |
6000 | r = printk("%s%s: %pV", level, netdev_name(dev), vaf); | |
6001 | else | |
6002 | r = printk("%s(NULL net_device): %pV", level, vaf); | |
6003 | ||
6004 | return r; | |
6005 | } | |
6006 | ||
6007 | int netdev_printk(const char *level, const struct net_device *dev, | |
6008 | const char *format, ...) | |
6009 | { | |
6010 | struct va_format vaf; | |
6011 | va_list args; | |
6012 | int r; | |
6013 | ||
6014 | va_start(args, format); | |
6015 | ||
6016 | vaf.fmt = format; | |
6017 | vaf.va = &args; | |
6018 | ||
6019 | r = __netdev_printk(level, dev, &vaf); | |
6020 | va_end(args); | |
6021 | ||
6022 | return r; | |
6023 | } | |
6024 | EXPORT_SYMBOL(netdev_printk); | |
6025 | ||
6026 | #define define_netdev_printk_level(func, level) \ | |
6027 | int func(const struct net_device *dev, const char *fmt, ...) \ | |
6028 | { \ | |
6029 | int r; \ | |
6030 | struct va_format vaf; \ | |
6031 | va_list args; \ | |
6032 | \ | |
6033 | va_start(args, fmt); \ | |
6034 | \ | |
6035 | vaf.fmt = fmt; \ | |
6036 | vaf.va = &args; \ | |
6037 | \ | |
6038 | r = __netdev_printk(level, dev, &vaf); \ | |
6039 | va_end(args); \ | |
6040 | \ | |
6041 | return r; \ | |
6042 | } \ | |
6043 | EXPORT_SYMBOL(func); | |
6044 | ||
6045 | define_netdev_printk_level(netdev_emerg, KERN_EMERG); | |
6046 | define_netdev_printk_level(netdev_alert, KERN_ALERT); | |
6047 | define_netdev_printk_level(netdev_crit, KERN_CRIT); | |
6048 | define_netdev_printk_level(netdev_err, KERN_ERR); | |
6049 | define_netdev_printk_level(netdev_warn, KERN_WARNING); | |
6050 | define_netdev_printk_level(netdev_notice, KERN_NOTICE); | |
6051 | define_netdev_printk_level(netdev_info, KERN_INFO); | |
6052 | ||
6053 | static void __net_exit netdev_exit(struct net *net) | |
6054 | { | |
6055 | kfree(net->dev_name_head); | |
6056 | kfree(net->dev_index_head); | |
6057 | } | |
6058 | ||
6059 | static struct pernet_operations __net_initdata netdev_net_ops = { | |
6060 | .init = netdev_init, | |
6061 | .exit = netdev_exit, | |
6062 | }; | |
6063 | ||
6064 | static void __net_exit default_device_exit(struct net *net) | |
6065 | { | |
6066 | struct net_device *dev, *aux; | |
6067 | /* | |
6068 | * Push all migratable network devices back to the | |
6069 | * initial network namespace | |
6070 | */ | |
6071 | rtnl_lock(); | |
6072 | for_each_netdev_safe(net, dev, aux) { | |
6073 | int err; | |
6074 | char fb_name[IFNAMSIZ]; | |
6075 | ||
6076 | /* Ignore unmoveable devices (i.e. loopback) */ | |
6077 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
6078 | continue; | |
6079 | ||
6080 | /* Leave virtual devices for the generic cleanup */ | |
6081 | if (dev->rtnl_link_ops) | |
6082 | continue; | |
6083 | ||
6084 | /* Push remaing network devices to init_net */ | |
6085 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); | |
6086 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
6087 | if (err) { | |
6088 | printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n", | |
6089 | __func__, dev->name, err); | |
6090 | BUG(); | |
6091 | } | |
6092 | } | |
6093 | rtnl_unlock(); | |
6094 | } | |
6095 | ||
6096 | static void __net_exit default_device_exit_batch(struct list_head *net_list) | |
6097 | { | |
6098 | /* At exit all network devices most be removed from a network | |
6099 | * namespace. Do this in the reverse order of registeration. | |
6100 | * Do this across as many network namespaces as possible to | |
6101 | * improve batching efficiency. | |
6102 | */ | |
6103 | struct net_device *dev; | |
6104 | struct net *net; | |
6105 | LIST_HEAD(dev_kill_list); | |
6106 | ||
6107 | rtnl_lock(); | |
6108 | list_for_each_entry(net, net_list, exit_list) { | |
6109 | for_each_netdev_reverse(net, dev) { | |
6110 | if (dev->rtnl_link_ops) | |
6111 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); | |
6112 | else | |
6113 | unregister_netdevice_queue(dev, &dev_kill_list); | |
6114 | } | |
6115 | } | |
6116 | unregister_netdevice_many(&dev_kill_list); | |
6117 | rtnl_unlock(); | |
6118 | } | |
6119 | ||
6120 | static struct pernet_operations __net_initdata default_device_ops = { | |
6121 | .exit = default_device_exit, | |
6122 | .exit_batch = default_device_exit_batch, | |
6123 | }; | |
6124 | ||
6125 | /* | |
6126 | * Initialize the DEV module. At boot time this walks the device list and | |
6127 | * unhooks any devices that fail to initialise (normally hardware not | |
6128 | * present) and leaves us with a valid list of present and active devices. | |
6129 | * | |
6130 | */ | |
6131 | ||
6132 | /* | |
6133 | * This is called single threaded during boot, so no need | |
6134 | * to take the rtnl semaphore. | |
6135 | */ | |
6136 | static int __init net_dev_init(void) | |
6137 | { | |
6138 | int i, rc = -ENOMEM; | |
6139 | ||
6140 | BUG_ON(!dev_boot_phase); | |
6141 | ||
6142 | if (dev_proc_init()) | |
6143 | goto out; | |
6144 | ||
6145 | if (netdev_kobject_init()) | |
6146 | goto out; | |
6147 | ||
6148 | INIT_LIST_HEAD(&ptype_all); | |
6149 | for (i = 0; i < PTYPE_HASH_SIZE; i++) | |
6150 | INIT_LIST_HEAD(&ptype_base[i]); | |
6151 | ||
6152 | if (register_pernet_subsys(&netdev_net_ops)) | |
6153 | goto out; | |
6154 | ||
6155 | /* | |
6156 | * Initialise the packet receive queues. | |
6157 | */ | |
6158 | ||
6159 | for_each_possible_cpu(i) { | |
6160 | struct softnet_data *sd = &per_cpu(softnet_data, i); | |
6161 | ||
6162 | memset(sd, 0, sizeof(*sd)); | |
6163 | skb_queue_head_init(&sd->input_pkt_queue); | |
6164 | skb_queue_head_init(&sd->process_queue); | |
6165 | sd->completion_queue = NULL; | |
6166 | INIT_LIST_HEAD(&sd->poll_list); | |
6167 | sd->output_queue = NULL; | |
6168 | sd->output_queue_tailp = &sd->output_queue; | |
6169 | #ifdef CONFIG_RPS | |
6170 | sd->csd.func = rps_trigger_softirq; | |
6171 | sd->csd.info = sd; | |
6172 | sd->csd.flags = 0; | |
6173 | sd->cpu = i; | |
6174 | #endif | |
6175 | ||
6176 | sd->backlog.poll = process_backlog; | |
6177 | sd->backlog.weight = weight_p; | |
6178 | sd->backlog.gro_list = NULL; | |
6179 | sd->backlog.gro_count = 0; | |
6180 | } | |
6181 | ||
6182 | dev_boot_phase = 0; | |
6183 | ||
6184 | /* The loopback device is special if any other network devices | |
6185 | * is present in a network namespace the loopback device must | |
6186 | * be present. Since we now dynamically allocate and free the | |
6187 | * loopback device ensure this invariant is maintained by | |
6188 | * keeping the loopback device as the first device on the | |
6189 | * list of network devices. Ensuring the loopback devices | |
6190 | * is the first device that appears and the last network device | |
6191 | * that disappears. | |
6192 | */ | |
6193 | if (register_pernet_device(&loopback_net_ops)) | |
6194 | goto out; | |
6195 | ||
6196 | if (register_pernet_device(&default_device_ops)) | |
6197 | goto out; | |
6198 | ||
6199 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); | |
6200 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
6201 | ||
6202 | hotcpu_notifier(dev_cpu_callback, 0); | |
6203 | dst_init(); | |
6204 | dev_mcast_init(); | |
6205 | rc = 0; | |
6206 | out: | |
6207 | return rc; | |
6208 | } | |
6209 | ||
6210 | subsys_initcall(net_dev_init); | |
6211 | ||
6212 | static int __init initialize_hashrnd(void) | |
6213 | { | |
6214 | get_random_bytes(&hashrnd, sizeof(hashrnd)); | |
6215 | return 0; | |
6216 | } | |
6217 | ||
6218 | late_initcall_sync(initialize_hashrnd); | |
6219 |