]>
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/sched.h> | |
84 | #include <linux/mutex.h> | |
85 | #include <linux/string.h> | |
86 | #include <linux/mm.h> | |
87 | #include <linux/socket.h> | |
88 | #include <linux/sockios.h> | |
89 | #include <linux/errno.h> | |
90 | #include <linux/interrupt.h> | |
91 | #include <linux/if_ether.h> | |
92 | #include <linux/netdevice.h> | |
93 | #include <linux/etherdevice.h> | |
94 | #include <linux/ethtool.h> | |
95 | #include <linux/notifier.h> | |
96 | #include <linux/skbuff.h> | |
97 | #include <net/net_namespace.h> | |
98 | #include <net/sock.h> | |
99 | #include <linux/rtnetlink.h> | |
100 | #include <linux/proc_fs.h> | |
101 | #include <linux/seq_file.h> | |
102 | #include <linux/stat.h> | |
103 | #include <linux/if_bridge.h> | |
104 | #include <linux/if_macvlan.h> | |
105 | #include <net/dst.h> | |
106 | #include <net/pkt_sched.h> | |
107 | #include <net/checksum.h> | |
108 | #include <net/xfrm.h> | |
109 | #include <linux/highmem.h> | |
110 | #include <linux/init.h> | |
111 | #include <linux/kmod.h> | |
112 | #include <linux/module.h> | |
113 | #include <linux/netpoll.h> | |
114 | #include <linux/rcupdate.h> | |
115 | #include <linux/delay.h> | |
116 | #include <net/wext.h> | |
117 | #include <net/iw_handler.h> | |
118 | #include <asm/current.h> | |
119 | #include <linux/audit.h> | |
120 | #include <linux/dmaengine.h> | |
121 | #include <linux/err.h> | |
122 | #include <linux/ctype.h> | |
123 | #include <linux/if_arp.h> | |
124 | #include <linux/if_vlan.h> | |
125 | #include <linux/ip.h> | |
126 | #include <net/ip.h> | |
127 | #include <linux/ipv6.h> | |
128 | #include <linux/in.h> | |
129 | #include <linux/jhash.h> | |
130 | #include <linux/random.h> | |
131 | #include <trace/events/napi.h> | |
132 | ||
133 | #include "net-sysfs.h" | |
134 | ||
135 | /* Instead of increasing this, you should create a hash table. */ | |
136 | #define MAX_GRO_SKBS 8 | |
137 | ||
138 | /* This should be increased if a protocol with a bigger head is added. */ | |
139 | #define GRO_MAX_HEAD (MAX_HEADER + 128) | |
140 | ||
141 | /* | |
142 | * The list of packet types we will receive (as opposed to discard) | |
143 | * and the routines to invoke. | |
144 | * | |
145 | * Why 16. Because with 16 the only overlap we get on a hash of the | |
146 | * low nibble of the protocol value is RARP/SNAP/X.25. | |
147 | * | |
148 | * NOTE: That is no longer true with the addition of VLAN tags. Not | |
149 | * sure which should go first, but I bet it won't make much | |
150 | * difference if we are running VLANs. The good news is that | |
151 | * this protocol won't be in the list unless compiled in, so | |
152 | * the average user (w/out VLANs) will not be adversely affected. | |
153 | * --BLG | |
154 | * | |
155 | * 0800 IP | |
156 | * 8100 802.1Q VLAN | |
157 | * 0001 802.3 | |
158 | * 0002 AX.25 | |
159 | * 0004 802.2 | |
160 | * 8035 RARP | |
161 | * 0005 SNAP | |
162 | * 0805 X.25 | |
163 | * 0806 ARP | |
164 | * 8137 IPX | |
165 | * 0009 Localtalk | |
166 | * 86DD IPv6 | |
167 | */ | |
168 | ||
169 | #define PTYPE_HASH_SIZE (16) | |
170 | #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1) | |
171 | ||
172 | static DEFINE_SPINLOCK(ptype_lock); | |
173 | static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; | |
174 | static struct list_head ptype_all __read_mostly; /* Taps */ | |
175 | ||
176 | /* | |
177 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl | |
178 | * semaphore. | |
179 | * | |
180 | * Pure readers hold dev_base_lock for reading, or rcu_read_lock() | |
181 | * | |
182 | * Writers must hold the rtnl semaphore while they loop through the | |
183 | * dev_base_head list, and hold dev_base_lock for writing when they do the | |
184 | * actual updates. This allows pure readers to access the list even | |
185 | * while a writer is preparing to update it. | |
186 | * | |
187 | * To put it another way, dev_base_lock is held for writing only to | |
188 | * protect against pure readers; the rtnl semaphore provides the | |
189 | * protection against other writers. | |
190 | * | |
191 | * See, for example usages, register_netdevice() and | |
192 | * unregister_netdevice(), which must be called with the rtnl | |
193 | * semaphore held. | |
194 | */ | |
195 | DEFINE_RWLOCK(dev_base_lock); | |
196 | EXPORT_SYMBOL(dev_base_lock); | |
197 | ||
198 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) | |
199 | { | |
200 | unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ)); | |
201 | return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; | |
202 | } | |
203 | ||
204 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) | |
205 | { | |
206 | return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; | |
207 | } | |
208 | ||
209 | /* Device list insertion */ | |
210 | static int list_netdevice(struct net_device *dev) | |
211 | { | |
212 | struct net *net = dev_net(dev); | |
213 | ||
214 | ASSERT_RTNL(); | |
215 | ||
216 | write_lock_bh(&dev_base_lock); | |
217 | list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); | |
218 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
219 | hlist_add_head_rcu(&dev->index_hlist, | |
220 | dev_index_hash(net, dev->ifindex)); | |
221 | write_unlock_bh(&dev_base_lock); | |
222 | return 0; | |
223 | } | |
224 | ||
225 | /* Device list removal | |
226 | * caller must respect a RCU grace period before freeing/reusing dev | |
227 | */ | |
228 | static void unlist_netdevice(struct net_device *dev) | |
229 | { | |
230 | ASSERT_RTNL(); | |
231 | ||
232 | /* Unlink dev from the device chain */ | |
233 | write_lock_bh(&dev_base_lock); | |
234 | list_del_rcu(&dev->dev_list); | |
235 | hlist_del_rcu(&dev->name_hlist); | |
236 | hlist_del_rcu(&dev->index_hlist); | |
237 | write_unlock_bh(&dev_base_lock); | |
238 | } | |
239 | ||
240 | /* | |
241 | * Our notifier list | |
242 | */ | |
243 | ||
244 | static RAW_NOTIFIER_HEAD(netdev_chain); | |
245 | ||
246 | /* | |
247 | * Device drivers call our routines to queue packets here. We empty the | |
248 | * queue in the local softnet handler. | |
249 | */ | |
250 | ||
251 | DEFINE_PER_CPU(struct softnet_data, softnet_data); | |
252 | EXPORT_PER_CPU_SYMBOL(softnet_data); | |
253 | ||
254 | #ifdef CONFIG_LOCKDEP | |
255 | /* | |
256 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class | |
257 | * according to dev->type | |
258 | */ | |
259 | static const unsigned short netdev_lock_type[] = | |
260 | {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
261 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, | |
262 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
263 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
264 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
265 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
266 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
267 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
268 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
269 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
270 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
271 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
272 | ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211, | |
273 | ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, | |
274 | ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154, | |
275 | ARPHRD_VOID, ARPHRD_NONE}; | |
276 | ||
277 | static const char *const netdev_lock_name[] = | |
278 | {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", | |
279 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
280 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
281 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
282 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
283 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
284 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
285 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
286 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
287 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
288 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
289 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
290 | "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211", | |
291 | "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", | |
292 | "_xmit_PHONET_PIPE", "_xmit_IEEE802154", | |
293 | "_xmit_VOID", "_xmit_NONE"}; | |
294 | ||
295 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
296 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
297 | ||
298 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
299 | { | |
300 | int i; | |
301 | ||
302 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
303 | if (netdev_lock_type[i] == dev_type) | |
304 | return i; | |
305 | /* the last key is used by default */ | |
306 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
307 | } | |
308 | ||
309 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, | |
310 | unsigned short dev_type) | |
311 | { | |
312 | int i; | |
313 | ||
314 | i = netdev_lock_pos(dev_type); | |
315 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
316 | netdev_lock_name[i]); | |
317 | } | |
318 | ||
319 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
320 | { | |
321 | int i; | |
322 | ||
323 | i = netdev_lock_pos(dev->type); | |
324 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
325 | &netdev_addr_lock_key[i], | |
326 | netdev_lock_name[i]); | |
327 | } | |
328 | #else | |
329 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, | |
330 | unsigned short dev_type) | |
331 | { | |
332 | } | |
333 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
334 | { | |
335 | } | |
336 | #endif | |
337 | ||
338 | /******************************************************************************* | |
339 | ||
340 | Protocol management and registration routines | |
341 | ||
342 | *******************************************************************************/ | |
343 | ||
344 | /* | |
345 | * Add a protocol ID to the list. Now that the input handler is | |
346 | * smarter we can dispense with all the messy stuff that used to be | |
347 | * here. | |
348 | * | |
349 | * BEWARE!!! Protocol handlers, mangling input packets, | |
350 | * MUST BE last in hash buckets and checking protocol handlers | |
351 | * MUST start from promiscuous ptype_all chain in net_bh. | |
352 | * It is true now, do not change it. | |
353 | * Explanation follows: if protocol handler, mangling packet, will | |
354 | * be the first on list, it is not able to sense, that packet | |
355 | * is cloned and should be copied-on-write, so that it will | |
356 | * change it and subsequent readers will get broken packet. | |
357 | * --ANK (980803) | |
358 | */ | |
359 | ||
360 | /** | |
361 | * dev_add_pack - add packet handler | |
362 | * @pt: packet type declaration | |
363 | * | |
364 | * Add a protocol handler to the networking stack. The passed &packet_type | |
365 | * is linked into kernel lists and may not be freed until it has been | |
366 | * removed from the kernel lists. | |
367 | * | |
368 | * This call does not sleep therefore it can not | |
369 | * guarantee all CPU's that are in middle of receiving packets | |
370 | * will see the new packet type (until the next received packet). | |
371 | */ | |
372 | ||
373 | void dev_add_pack(struct packet_type *pt) | |
374 | { | |
375 | int hash; | |
376 | ||
377 | spin_lock_bh(&ptype_lock); | |
378 | if (pt->type == htons(ETH_P_ALL)) | |
379 | list_add_rcu(&pt->list, &ptype_all); | |
380 | else { | |
381 | hash = ntohs(pt->type) & PTYPE_HASH_MASK; | |
382 | list_add_rcu(&pt->list, &ptype_base[hash]); | |
383 | } | |
384 | spin_unlock_bh(&ptype_lock); | |
385 | } | |
386 | EXPORT_SYMBOL(dev_add_pack); | |
387 | ||
388 | /** | |
389 | * __dev_remove_pack - remove packet handler | |
390 | * @pt: packet type declaration | |
391 | * | |
392 | * Remove a protocol handler that was previously added to the kernel | |
393 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
394 | * from the kernel lists and can be freed or reused once this function | |
395 | * returns. | |
396 | * | |
397 | * The packet type might still be in use by receivers | |
398 | * and must not be freed until after all the CPU's have gone | |
399 | * through a quiescent state. | |
400 | */ | |
401 | void __dev_remove_pack(struct packet_type *pt) | |
402 | { | |
403 | struct list_head *head; | |
404 | struct packet_type *pt1; | |
405 | ||
406 | spin_lock_bh(&ptype_lock); | |
407 | ||
408 | if (pt->type == htons(ETH_P_ALL)) | |
409 | head = &ptype_all; | |
410 | else | |
411 | head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
412 | ||
413 | list_for_each_entry(pt1, head, list) { | |
414 | if (pt == pt1) { | |
415 | list_del_rcu(&pt->list); | |
416 | goto out; | |
417 | } | |
418 | } | |
419 | ||
420 | printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt); | |
421 | out: | |
422 | spin_unlock_bh(&ptype_lock); | |
423 | } | |
424 | EXPORT_SYMBOL(__dev_remove_pack); | |
425 | ||
426 | /** | |
427 | * dev_remove_pack - remove packet handler | |
428 | * @pt: packet type declaration | |
429 | * | |
430 | * Remove a protocol handler that was previously added to the kernel | |
431 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
432 | * from the kernel lists and can be freed or reused once this function | |
433 | * returns. | |
434 | * | |
435 | * This call sleeps to guarantee that no CPU is looking at the packet | |
436 | * type after return. | |
437 | */ | |
438 | void dev_remove_pack(struct packet_type *pt) | |
439 | { | |
440 | __dev_remove_pack(pt); | |
441 | ||
442 | synchronize_net(); | |
443 | } | |
444 | EXPORT_SYMBOL(dev_remove_pack); | |
445 | ||
446 | /****************************************************************************** | |
447 | ||
448 | Device Boot-time Settings Routines | |
449 | ||
450 | *******************************************************************************/ | |
451 | ||
452 | /* Boot time configuration table */ | |
453 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
454 | ||
455 | /** | |
456 | * netdev_boot_setup_add - add new setup entry | |
457 | * @name: name of the device | |
458 | * @map: configured settings for the device | |
459 | * | |
460 | * Adds new setup entry to the dev_boot_setup list. The function | |
461 | * returns 0 on error and 1 on success. This is a generic routine to | |
462 | * all netdevices. | |
463 | */ | |
464 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
465 | { | |
466 | struct netdev_boot_setup *s; | |
467 | int i; | |
468 | ||
469 | s = dev_boot_setup; | |
470 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
471 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
472 | memset(s[i].name, 0, sizeof(s[i].name)); | |
473 | strlcpy(s[i].name, name, IFNAMSIZ); | |
474 | memcpy(&s[i].map, map, sizeof(s[i].map)); | |
475 | break; | |
476 | } | |
477 | } | |
478 | ||
479 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
480 | } | |
481 | ||
482 | /** | |
483 | * netdev_boot_setup_check - check boot time settings | |
484 | * @dev: the netdevice | |
485 | * | |
486 | * Check boot time settings for the device. | |
487 | * The found settings are set for the device to be used | |
488 | * later in the device probing. | |
489 | * Returns 0 if no settings found, 1 if they are. | |
490 | */ | |
491 | int netdev_boot_setup_check(struct net_device *dev) | |
492 | { | |
493 | struct netdev_boot_setup *s = dev_boot_setup; | |
494 | int i; | |
495 | ||
496 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
497 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
498 | !strcmp(dev->name, s[i].name)) { | |
499 | dev->irq = s[i].map.irq; | |
500 | dev->base_addr = s[i].map.base_addr; | |
501 | dev->mem_start = s[i].map.mem_start; | |
502 | dev->mem_end = s[i].map.mem_end; | |
503 | return 1; | |
504 | } | |
505 | } | |
506 | return 0; | |
507 | } | |
508 | EXPORT_SYMBOL(netdev_boot_setup_check); | |
509 | ||
510 | ||
511 | /** | |
512 | * netdev_boot_base - get address from boot time settings | |
513 | * @prefix: prefix for network device | |
514 | * @unit: id for network device | |
515 | * | |
516 | * Check boot time settings for the base address of device. | |
517 | * The found settings are set for the device to be used | |
518 | * later in the device probing. | |
519 | * Returns 0 if no settings found. | |
520 | */ | |
521 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
522 | { | |
523 | const struct netdev_boot_setup *s = dev_boot_setup; | |
524 | char name[IFNAMSIZ]; | |
525 | int i; | |
526 | ||
527 | sprintf(name, "%s%d", prefix, unit); | |
528 | ||
529 | /* | |
530 | * If device already registered then return base of 1 | |
531 | * to indicate not to probe for this interface | |
532 | */ | |
533 | if (__dev_get_by_name(&init_net, name)) | |
534 | return 1; | |
535 | ||
536 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
537 | if (!strcmp(name, s[i].name)) | |
538 | return s[i].map.base_addr; | |
539 | return 0; | |
540 | } | |
541 | ||
542 | /* | |
543 | * Saves at boot time configured settings for any netdevice. | |
544 | */ | |
545 | int __init netdev_boot_setup(char *str) | |
546 | { | |
547 | int ints[5]; | |
548 | struct ifmap map; | |
549 | ||
550 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
551 | if (!str || !*str) | |
552 | return 0; | |
553 | ||
554 | /* Save settings */ | |
555 | memset(&map, 0, sizeof(map)); | |
556 | if (ints[0] > 0) | |
557 | map.irq = ints[1]; | |
558 | if (ints[0] > 1) | |
559 | map.base_addr = ints[2]; | |
560 | if (ints[0] > 2) | |
561 | map.mem_start = ints[3]; | |
562 | if (ints[0] > 3) | |
563 | map.mem_end = ints[4]; | |
564 | ||
565 | /* Add new entry to the list */ | |
566 | return netdev_boot_setup_add(str, &map); | |
567 | } | |
568 | ||
569 | __setup("netdev=", netdev_boot_setup); | |
570 | ||
571 | /******************************************************************************* | |
572 | ||
573 | Device Interface Subroutines | |
574 | ||
575 | *******************************************************************************/ | |
576 | ||
577 | /** | |
578 | * __dev_get_by_name - find a device by its name | |
579 | * @net: the applicable net namespace | |
580 | * @name: name to find | |
581 | * | |
582 | * Find an interface by name. Must be called under RTNL semaphore | |
583 | * or @dev_base_lock. If the name is found a pointer to the device | |
584 | * is returned. If the name is not found then %NULL is returned. The | |
585 | * reference counters are not incremented so the caller must be | |
586 | * careful with locks. | |
587 | */ | |
588 | ||
589 | struct net_device *__dev_get_by_name(struct net *net, const char *name) | |
590 | { | |
591 | struct hlist_node *p; | |
592 | struct net_device *dev; | |
593 | struct hlist_head *head = dev_name_hash(net, name); | |
594 | ||
595 | hlist_for_each_entry(dev, p, head, name_hlist) | |
596 | if (!strncmp(dev->name, name, IFNAMSIZ)) | |
597 | return dev; | |
598 | ||
599 | return NULL; | |
600 | } | |
601 | EXPORT_SYMBOL(__dev_get_by_name); | |
602 | ||
603 | /** | |
604 | * dev_get_by_name_rcu - find a device by its name | |
605 | * @net: the applicable net namespace | |
606 | * @name: name to find | |
607 | * | |
608 | * Find an interface by name. | |
609 | * If the name is found a pointer to the device is returned. | |
610 | * If the name is not found then %NULL is returned. | |
611 | * The reference counters are not incremented so the caller must be | |
612 | * careful with locks. The caller must hold RCU lock. | |
613 | */ | |
614 | ||
615 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
616 | { | |
617 | struct hlist_node *p; | |
618 | struct net_device *dev; | |
619 | struct hlist_head *head = dev_name_hash(net, name); | |
620 | ||
621 | hlist_for_each_entry_rcu(dev, p, head, name_hlist) | |
622 | if (!strncmp(dev->name, name, IFNAMSIZ)) | |
623 | return dev; | |
624 | ||
625 | return NULL; | |
626 | } | |
627 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
628 | ||
629 | /** | |
630 | * dev_get_by_name - find a device by its name | |
631 | * @net: the applicable net namespace | |
632 | * @name: name to find | |
633 | * | |
634 | * Find an interface by name. This can be called from any | |
635 | * context and does its own locking. The returned handle has | |
636 | * the usage count incremented and the caller must use dev_put() to | |
637 | * release it when it is no longer needed. %NULL is returned if no | |
638 | * matching device is found. | |
639 | */ | |
640 | ||
641 | struct net_device *dev_get_by_name(struct net *net, const char *name) | |
642 | { | |
643 | struct net_device *dev; | |
644 | ||
645 | rcu_read_lock(); | |
646 | dev = dev_get_by_name_rcu(net, name); | |
647 | if (dev) | |
648 | dev_hold(dev); | |
649 | rcu_read_unlock(); | |
650 | return dev; | |
651 | } | |
652 | EXPORT_SYMBOL(dev_get_by_name); | |
653 | ||
654 | /** | |
655 | * __dev_get_by_index - find a device by its ifindex | |
656 | * @net: the applicable net namespace | |
657 | * @ifindex: index of device | |
658 | * | |
659 | * Search for an interface by index. Returns %NULL if the device | |
660 | * is not found or a pointer to the device. The device has not | |
661 | * had its reference counter increased so the caller must be careful | |
662 | * about locking. The caller must hold either the RTNL semaphore | |
663 | * or @dev_base_lock. | |
664 | */ | |
665 | ||
666 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) | |
667 | { | |
668 | struct hlist_node *p; | |
669 | struct net_device *dev; | |
670 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
671 | ||
672 | hlist_for_each_entry(dev, p, head, index_hlist) | |
673 | if (dev->ifindex == ifindex) | |
674 | return dev; | |
675 | ||
676 | return NULL; | |
677 | } | |
678 | EXPORT_SYMBOL(__dev_get_by_index); | |
679 | ||
680 | /** | |
681 | * dev_get_by_index_rcu - find a device by its ifindex | |
682 | * @net: the applicable net namespace | |
683 | * @ifindex: index of device | |
684 | * | |
685 | * Search for an interface by index. Returns %NULL if the device | |
686 | * is not found or a pointer to the device. The device has not | |
687 | * had its reference counter increased so the caller must be careful | |
688 | * about locking. The caller must hold RCU lock. | |
689 | */ | |
690 | ||
691 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
692 | { | |
693 | struct hlist_node *p; | |
694 | struct net_device *dev; | |
695 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
696 | ||
697 | hlist_for_each_entry_rcu(dev, p, head, index_hlist) | |
698 | if (dev->ifindex == ifindex) | |
699 | return dev; | |
700 | ||
701 | return NULL; | |
702 | } | |
703 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
704 | ||
705 | ||
706 | /** | |
707 | * dev_get_by_index - find a device by its ifindex | |
708 | * @net: the applicable net namespace | |
709 | * @ifindex: index of device | |
710 | * | |
711 | * Search for an interface by index. Returns NULL if the device | |
712 | * is not found or a pointer to the device. The device returned has | |
713 | * had a reference added and the pointer is safe until the user calls | |
714 | * dev_put to indicate they have finished with it. | |
715 | */ | |
716 | ||
717 | struct net_device *dev_get_by_index(struct net *net, int ifindex) | |
718 | { | |
719 | struct net_device *dev; | |
720 | ||
721 | rcu_read_lock(); | |
722 | dev = dev_get_by_index_rcu(net, ifindex); | |
723 | if (dev) | |
724 | dev_hold(dev); | |
725 | rcu_read_unlock(); | |
726 | return dev; | |
727 | } | |
728 | EXPORT_SYMBOL(dev_get_by_index); | |
729 | ||
730 | /** | |
731 | * dev_getbyhwaddr - find a device by its hardware address | |
732 | * @net: the applicable net namespace | |
733 | * @type: media type of device | |
734 | * @ha: hardware address | |
735 | * | |
736 | * Search for an interface by MAC address. Returns NULL if the device | |
737 | * is not found or a pointer to the device. The caller must hold the | |
738 | * rtnl semaphore. The returned device has not had its ref count increased | |
739 | * and the caller must therefore be careful about locking | |
740 | * | |
741 | * BUGS: | |
742 | * If the API was consistent this would be __dev_get_by_hwaddr | |
743 | */ | |
744 | ||
745 | struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha) | |
746 | { | |
747 | struct net_device *dev; | |
748 | ||
749 | ASSERT_RTNL(); | |
750 | ||
751 | for_each_netdev(net, dev) | |
752 | if (dev->type == type && | |
753 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
754 | return dev; | |
755 | ||
756 | return NULL; | |
757 | } | |
758 | EXPORT_SYMBOL(dev_getbyhwaddr); | |
759 | ||
760 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) | |
761 | { | |
762 | struct net_device *dev; | |
763 | ||
764 | ASSERT_RTNL(); | |
765 | for_each_netdev(net, dev) | |
766 | if (dev->type == type) | |
767 | return dev; | |
768 | ||
769 | return NULL; | |
770 | } | |
771 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); | |
772 | ||
773 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) | |
774 | { | |
775 | struct net_device *dev; | |
776 | ||
777 | rtnl_lock(); | |
778 | dev = __dev_getfirstbyhwtype(net, type); | |
779 | if (dev) | |
780 | dev_hold(dev); | |
781 | rtnl_unlock(); | |
782 | return dev; | |
783 | } | |
784 | EXPORT_SYMBOL(dev_getfirstbyhwtype); | |
785 | ||
786 | /** | |
787 | * dev_get_by_flags - find any device with given flags | |
788 | * @net: the applicable net namespace | |
789 | * @if_flags: IFF_* values | |
790 | * @mask: bitmask of bits in if_flags to check | |
791 | * | |
792 | * Search for any interface with the given flags. Returns NULL if a device | |
793 | * is not found or a pointer to the device. The device returned has | |
794 | * had a reference added and the pointer is safe until the user calls | |
795 | * dev_put to indicate they have finished with it. | |
796 | */ | |
797 | ||
798 | struct net_device *dev_get_by_flags(struct net *net, unsigned short if_flags, | |
799 | unsigned short mask) | |
800 | { | |
801 | struct net_device *dev, *ret; | |
802 | ||
803 | ret = NULL; | |
804 | rcu_read_lock(); | |
805 | for_each_netdev_rcu(net, dev) { | |
806 | if (((dev->flags ^ if_flags) & mask) == 0) { | |
807 | dev_hold(dev); | |
808 | ret = dev; | |
809 | break; | |
810 | } | |
811 | } | |
812 | rcu_read_unlock(); | |
813 | return ret; | |
814 | } | |
815 | EXPORT_SYMBOL(dev_get_by_flags); | |
816 | ||
817 | /** | |
818 | * dev_valid_name - check if name is okay for network device | |
819 | * @name: name string | |
820 | * | |
821 | * Network device names need to be valid file names to | |
822 | * to allow sysfs to work. We also disallow any kind of | |
823 | * whitespace. | |
824 | */ | |
825 | int dev_valid_name(const char *name) | |
826 | { | |
827 | if (*name == '\0') | |
828 | return 0; | |
829 | if (strlen(name) >= IFNAMSIZ) | |
830 | return 0; | |
831 | if (!strcmp(name, ".") || !strcmp(name, "..")) | |
832 | return 0; | |
833 | ||
834 | while (*name) { | |
835 | if (*name == '/' || isspace(*name)) | |
836 | return 0; | |
837 | name++; | |
838 | } | |
839 | return 1; | |
840 | } | |
841 | EXPORT_SYMBOL(dev_valid_name); | |
842 | ||
843 | /** | |
844 | * __dev_alloc_name - allocate a name for a device | |
845 | * @net: network namespace to allocate the device name in | |
846 | * @name: name format string | |
847 | * @buf: scratch buffer and result name string | |
848 | * | |
849 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
850 | * id. It scans list of devices to build up a free map, then chooses | |
851 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
852 | * while allocating the name and adding the device in order to avoid | |
853 | * duplicates. | |
854 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
855 | * Returns the number of the unit assigned or a negative errno code. | |
856 | */ | |
857 | ||
858 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) | |
859 | { | |
860 | int i = 0; | |
861 | const char *p; | |
862 | const int max_netdevices = 8*PAGE_SIZE; | |
863 | unsigned long *inuse; | |
864 | struct net_device *d; | |
865 | ||
866 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
867 | if (p) { | |
868 | /* | |
869 | * Verify the string as this thing may have come from | |
870 | * the user. There must be either one "%d" and no other "%" | |
871 | * characters. | |
872 | */ | |
873 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
874 | return -EINVAL; | |
875 | ||
876 | /* Use one page as a bit array of possible slots */ | |
877 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); | |
878 | if (!inuse) | |
879 | return -ENOMEM; | |
880 | ||
881 | for_each_netdev(net, d) { | |
882 | if (!sscanf(d->name, name, &i)) | |
883 | continue; | |
884 | if (i < 0 || i >= max_netdevices) | |
885 | continue; | |
886 | ||
887 | /* avoid cases where sscanf is not exact inverse of printf */ | |
888 | snprintf(buf, IFNAMSIZ, name, i); | |
889 | if (!strncmp(buf, d->name, IFNAMSIZ)) | |
890 | set_bit(i, inuse); | |
891 | } | |
892 | ||
893 | i = find_first_zero_bit(inuse, max_netdevices); | |
894 | free_page((unsigned long) inuse); | |
895 | } | |
896 | ||
897 | if (buf != name) | |
898 | snprintf(buf, IFNAMSIZ, name, i); | |
899 | if (!__dev_get_by_name(net, buf)) | |
900 | return i; | |
901 | ||
902 | /* It is possible to run out of possible slots | |
903 | * when the name is long and there isn't enough space left | |
904 | * for the digits, or if all bits are used. | |
905 | */ | |
906 | return -ENFILE; | |
907 | } | |
908 | ||
909 | /** | |
910 | * dev_alloc_name - allocate a name for a device | |
911 | * @dev: device | |
912 | * @name: name format string | |
913 | * | |
914 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
915 | * id. It scans list of devices to build up a free map, then chooses | |
916 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
917 | * while allocating the name and adding the device in order to avoid | |
918 | * duplicates. | |
919 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
920 | * Returns the number of the unit assigned or a negative errno code. | |
921 | */ | |
922 | ||
923 | int dev_alloc_name(struct net_device *dev, const char *name) | |
924 | { | |
925 | char buf[IFNAMSIZ]; | |
926 | struct net *net; | |
927 | int ret; | |
928 | ||
929 | BUG_ON(!dev_net(dev)); | |
930 | net = dev_net(dev); | |
931 | ret = __dev_alloc_name(net, name, buf); | |
932 | if (ret >= 0) | |
933 | strlcpy(dev->name, buf, IFNAMSIZ); | |
934 | return ret; | |
935 | } | |
936 | EXPORT_SYMBOL(dev_alloc_name); | |
937 | ||
938 | static int dev_get_valid_name(struct net *net, const char *name, char *buf, | |
939 | bool fmt) | |
940 | { | |
941 | if (!dev_valid_name(name)) | |
942 | return -EINVAL; | |
943 | ||
944 | if (fmt && strchr(name, '%')) | |
945 | return __dev_alloc_name(net, name, buf); | |
946 | else if (__dev_get_by_name(net, name)) | |
947 | return -EEXIST; | |
948 | else if (buf != name) | |
949 | strlcpy(buf, name, IFNAMSIZ); | |
950 | ||
951 | return 0; | |
952 | } | |
953 | ||
954 | /** | |
955 | * dev_change_name - change name of a device | |
956 | * @dev: device | |
957 | * @newname: name (or format string) must be at least IFNAMSIZ | |
958 | * | |
959 | * Change name of a device, can pass format strings "eth%d". | |
960 | * for wildcarding. | |
961 | */ | |
962 | int dev_change_name(struct net_device *dev, const char *newname) | |
963 | { | |
964 | char oldname[IFNAMSIZ]; | |
965 | int err = 0; | |
966 | int ret; | |
967 | struct net *net; | |
968 | ||
969 | ASSERT_RTNL(); | |
970 | BUG_ON(!dev_net(dev)); | |
971 | ||
972 | net = dev_net(dev); | |
973 | if (dev->flags & IFF_UP) | |
974 | return -EBUSY; | |
975 | ||
976 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) | |
977 | return 0; | |
978 | ||
979 | memcpy(oldname, dev->name, IFNAMSIZ); | |
980 | ||
981 | err = dev_get_valid_name(net, newname, dev->name, 1); | |
982 | if (err < 0) | |
983 | return err; | |
984 | ||
985 | rollback: | |
986 | /* For now only devices in the initial network namespace | |
987 | * are in sysfs. | |
988 | */ | |
989 | if (net_eq(net, &init_net)) { | |
990 | ret = device_rename(&dev->dev, dev->name); | |
991 | if (ret) { | |
992 | memcpy(dev->name, oldname, IFNAMSIZ); | |
993 | return ret; | |
994 | } | |
995 | } | |
996 | ||
997 | write_lock_bh(&dev_base_lock); | |
998 | hlist_del(&dev->name_hlist); | |
999 | write_unlock_bh(&dev_base_lock); | |
1000 | ||
1001 | synchronize_rcu(); | |
1002 | ||
1003 | write_lock_bh(&dev_base_lock); | |
1004 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
1005 | write_unlock_bh(&dev_base_lock); | |
1006 | ||
1007 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); | |
1008 | ret = notifier_to_errno(ret); | |
1009 | ||
1010 | if (ret) { | |
1011 | /* err >= 0 after dev_alloc_name() or stores the first errno */ | |
1012 | if (err >= 0) { | |
1013 | err = ret; | |
1014 | memcpy(dev->name, oldname, IFNAMSIZ); | |
1015 | goto rollback; | |
1016 | } else { | |
1017 | printk(KERN_ERR | |
1018 | "%s: name change rollback failed: %d.\n", | |
1019 | dev->name, ret); | |
1020 | } | |
1021 | } | |
1022 | ||
1023 | return err; | |
1024 | } | |
1025 | ||
1026 | /** | |
1027 | * dev_set_alias - change ifalias of a device | |
1028 | * @dev: device | |
1029 | * @alias: name up to IFALIASZ | |
1030 | * @len: limit of bytes to copy from info | |
1031 | * | |
1032 | * Set ifalias for a device, | |
1033 | */ | |
1034 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1035 | { | |
1036 | ASSERT_RTNL(); | |
1037 | ||
1038 | if (len >= IFALIASZ) | |
1039 | return -EINVAL; | |
1040 | ||
1041 | if (!len) { | |
1042 | if (dev->ifalias) { | |
1043 | kfree(dev->ifalias); | |
1044 | dev->ifalias = NULL; | |
1045 | } | |
1046 | return 0; | |
1047 | } | |
1048 | ||
1049 | dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL); | |
1050 | if (!dev->ifalias) | |
1051 | return -ENOMEM; | |
1052 | ||
1053 | strlcpy(dev->ifalias, alias, len+1); | |
1054 | return len; | |
1055 | } | |
1056 | ||
1057 | ||
1058 | /** | |
1059 | * netdev_features_change - device changes features | |
1060 | * @dev: device to cause notification | |
1061 | * | |
1062 | * Called to indicate a device has changed features. | |
1063 | */ | |
1064 | void netdev_features_change(struct net_device *dev) | |
1065 | { | |
1066 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); | |
1067 | } | |
1068 | EXPORT_SYMBOL(netdev_features_change); | |
1069 | ||
1070 | /** | |
1071 | * netdev_state_change - device changes state | |
1072 | * @dev: device to cause notification | |
1073 | * | |
1074 | * Called to indicate a device has changed state. This function calls | |
1075 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1076 | * to the routing socket. | |
1077 | */ | |
1078 | void netdev_state_change(struct net_device *dev) | |
1079 | { | |
1080 | if (dev->flags & IFF_UP) { | |
1081 | call_netdevice_notifiers(NETDEV_CHANGE, dev); | |
1082 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0); | |
1083 | } | |
1084 | } | |
1085 | EXPORT_SYMBOL(netdev_state_change); | |
1086 | ||
1087 | void netdev_bonding_change(struct net_device *dev, unsigned long event) | |
1088 | { | |
1089 | call_netdevice_notifiers(event, dev); | |
1090 | } | |
1091 | EXPORT_SYMBOL(netdev_bonding_change); | |
1092 | ||
1093 | /** | |
1094 | * dev_load - load a network module | |
1095 | * @net: the applicable net namespace | |
1096 | * @name: name of interface | |
1097 | * | |
1098 | * If a network interface is not present and the process has suitable | |
1099 | * privileges this function loads the module. If module loading is not | |
1100 | * available in this kernel then it becomes a nop. | |
1101 | */ | |
1102 | ||
1103 | void dev_load(struct net *net, const char *name) | |
1104 | { | |
1105 | struct net_device *dev; | |
1106 | ||
1107 | rcu_read_lock(); | |
1108 | dev = dev_get_by_name_rcu(net, name); | |
1109 | rcu_read_unlock(); | |
1110 | ||
1111 | if (!dev && capable(CAP_NET_ADMIN)) | |
1112 | request_module("%s", name); | |
1113 | } | |
1114 | EXPORT_SYMBOL(dev_load); | |
1115 | ||
1116 | static int __dev_open(struct net_device *dev) | |
1117 | { | |
1118 | const struct net_device_ops *ops = dev->netdev_ops; | |
1119 | int ret; | |
1120 | ||
1121 | ASSERT_RTNL(); | |
1122 | ||
1123 | /* | |
1124 | * Is it even present? | |
1125 | */ | |
1126 | if (!netif_device_present(dev)) | |
1127 | return -ENODEV; | |
1128 | ||
1129 | ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); | |
1130 | ret = notifier_to_errno(ret); | |
1131 | if (ret) | |
1132 | return ret; | |
1133 | ||
1134 | /* | |
1135 | * Call device private open method | |
1136 | */ | |
1137 | set_bit(__LINK_STATE_START, &dev->state); | |
1138 | ||
1139 | if (ops->ndo_validate_addr) | |
1140 | ret = ops->ndo_validate_addr(dev); | |
1141 | ||
1142 | if (!ret && ops->ndo_open) | |
1143 | ret = ops->ndo_open(dev); | |
1144 | ||
1145 | /* | |
1146 | * If it went open OK then: | |
1147 | */ | |
1148 | ||
1149 | if (ret) | |
1150 | clear_bit(__LINK_STATE_START, &dev->state); | |
1151 | else { | |
1152 | /* | |
1153 | * Set the flags. | |
1154 | */ | |
1155 | dev->flags |= IFF_UP; | |
1156 | ||
1157 | /* | |
1158 | * Enable NET_DMA | |
1159 | */ | |
1160 | net_dmaengine_get(); | |
1161 | ||
1162 | /* | |
1163 | * Initialize multicasting status | |
1164 | */ | |
1165 | dev_set_rx_mode(dev); | |
1166 | ||
1167 | /* | |
1168 | * Wakeup transmit queue engine | |
1169 | */ | |
1170 | dev_activate(dev); | |
1171 | } | |
1172 | ||
1173 | return ret; | |
1174 | } | |
1175 | ||
1176 | /** | |
1177 | * dev_open - prepare an interface for use. | |
1178 | * @dev: device to open | |
1179 | * | |
1180 | * Takes a device from down to up state. The device's private open | |
1181 | * function is invoked and then the multicast lists are loaded. Finally | |
1182 | * the device is moved into the up state and a %NETDEV_UP message is | |
1183 | * sent to the netdev notifier chain. | |
1184 | * | |
1185 | * Calling this function on an active interface is a nop. On a failure | |
1186 | * a negative errno code is returned. | |
1187 | */ | |
1188 | int dev_open(struct net_device *dev) | |
1189 | { | |
1190 | int ret; | |
1191 | ||
1192 | /* | |
1193 | * Is it already up? | |
1194 | */ | |
1195 | if (dev->flags & IFF_UP) | |
1196 | return 0; | |
1197 | ||
1198 | /* | |
1199 | * Open device | |
1200 | */ | |
1201 | ret = __dev_open(dev); | |
1202 | if (ret < 0) | |
1203 | return ret; | |
1204 | ||
1205 | /* | |
1206 | * ... and announce new interface. | |
1207 | */ | |
1208 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING); | |
1209 | call_netdevice_notifiers(NETDEV_UP, dev); | |
1210 | ||
1211 | return ret; | |
1212 | } | |
1213 | EXPORT_SYMBOL(dev_open); | |
1214 | ||
1215 | static int __dev_close(struct net_device *dev) | |
1216 | { | |
1217 | const struct net_device_ops *ops = dev->netdev_ops; | |
1218 | ||
1219 | ASSERT_RTNL(); | |
1220 | might_sleep(); | |
1221 | ||
1222 | /* | |
1223 | * Tell people we are going down, so that they can | |
1224 | * prepare to death, when device is still operating. | |
1225 | */ | |
1226 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); | |
1227 | ||
1228 | clear_bit(__LINK_STATE_START, &dev->state); | |
1229 | ||
1230 | /* Synchronize to scheduled poll. We cannot touch poll list, | |
1231 | * it can be even on different cpu. So just clear netif_running(). | |
1232 | * | |
1233 | * dev->stop() will invoke napi_disable() on all of it's | |
1234 | * napi_struct instances on this device. | |
1235 | */ | |
1236 | smp_mb__after_clear_bit(); /* Commit netif_running(). */ | |
1237 | ||
1238 | dev_deactivate(dev); | |
1239 | ||
1240 | /* | |
1241 | * Call the device specific close. This cannot fail. | |
1242 | * Only if device is UP | |
1243 | * | |
1244 | * We allow it to be called even after a DETACH hot-plug | |
1245 | * event. | |
1246 | */ | |
1247 | if (ops->ndo_stop) | |
1248 | ops->ndo_stop(dev); | |
1249 | ||
1250 | /* | |
1251 | * Device is now down. | |
1252 | */ | |
1253 | ||
1254 | dev->flags &= ~IFF_UP; | |
1255 | ||
1256 | /* | |
1257 | * Shutdown NET_DMA | |
1258 | */ | |
1259 | net_dmaengine_put(); | |
1260 | ||
1261 | return 0; | |
1262 | } | |
1263 | ||
1264 | /** | |
1265 | * dev_close - shutdown an interface. | |
1266 | * @dev: device to shutdown | |
1267 | * | |
1268 | * This function moves an active device into down state. A | |
1269 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1270 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1271 | * chain. | |
1272 | */ | |
1273 | int dev_close(struct net_device *dev) | |
1274 | { | |
1275 | if (!(dev->flags & IFF_UP)) | |
1276 | return 0; | |
1277 | ||
1278 | __dev_close(dev); | |
1279 | ||
1280 | /* | |
1281 | * Tell people we are down | |
1282 | */ | |
1283 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING); | |
1284 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
1285 | ||
1286 | return 0; | |
1287 | } | |
1288 | EXPORT_SYMBOL(dev_close); | |
1289 | ||
1290 | ||
1291 | /** | |
1292 | * dev_disable_lro - disable Large Receive Offload on a device | |
1293 | * @dev: device | |
1294 | * | |
1295 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1296 | * called under RTNL. This is needed if received packets may be | |
1297 | * forwarded to another interface. | |
1298 | */ | |
1299 | void dev_disable_lro(struct net_device *dev) | |
1300 | { | |
1301 | if (dev->ethtool_ops && dev->ethtool_ops->get_flags && | |
1302 | dev->ethtool_ops->set_flags) { | |
1303 | u32 flags = dev->ethtool_ops->get_flags(dev); | |
1304 | if (flags & ETH_FLAG_LRO) { | |
1305 | flags &= ~ETH_FLAG_LRO; | |
1306 | dev->ethtool_ops->set_flags(dev, flags); | |
1307 | } | |
1308 | } | |
1309 | WARN_ON(dev->features & NETIF_F_LRO); | |
1310 | } | |
1311 | EXPORT_SYMBOL(dev_disable_lro); | |
1312 | ||
1313 | ||
1314 | static int dev_boot_phase = 1; | |
1315 | ||
1316 | /* | |
1317 | * Device change register/unregister. These are not inline or static | |
1318 | * as we export them to the world. | |
1319 | */ | |
1320 | ||
1321 | /** | |
1322 | * register_netdevice_notifier - register a network notifier block | |
1323 | * @nb: notifier | |
1324 | * | |
1325 | * Register a notifier to be called when network device events occur. | |
1326 | * The notifier passed is linked into the kernel structures and must | |
1327 | * not be reused until it has been unregistered. A negative errno code | |
1328 | * is returned on a failure. | |
1329 | * | |
1330 | * When registered all registration and up events are replayed | |
1331 | * to the new notifier to allow device to have a race free | |
1332 | * view of the network device list. | |
1333 | */ | |
1334 | ||
1335 | int register_netdevice_notifier(struct notifier_block *nb) | |
1336 | { | |
1337 | struct net_device *dev; | |
1338 | struct net_device *last; | |
1339 | struct net *net; | |
1340 | int err; | |
1341 | ||
1342 | rtnl_lock(); | |
1343 | err = raw_notifier_chain_register(&netdev_chain, nb); | |
1344 | if (err) | |
1345 | goto unlock; | |
1346 | if (dev_boot_phase) | |
1347 | goto unlock; | |
1348 | for_each_net(net) { | |
1349 | for_each_netdev(net, dev) { | |
1350 | err = nb->notifier_call(nb, NETDEV_REGISTER, dev); | |
1351 | err = notifier_to_errno(err); | |
1352 | if (err) | |
1353 | goto rollback; | |
1354 | ||
1355 | if (!(dev->flags & IFF_UP)) | |
1356 | continue; | |
1357 | ||
1358 | nb->notifier_call(nb, NETDEV_UP, dev); | |
1359 | } | |
1360 | } | |
1361 | ||
1362 | unlock: | |
1363 | rtnl_unlock(); | |
1364 | return err; | |
1365 | ||
1366 | rollback: | |
1367 | last = dev; | |
1368 | for_each_net(net) { | |
1369 | for_each_netdev(net, dev) { | |
1370 | if (dev == last) | |
1371 | break; | |
1372 | ||
1373 | if (dev->flags & IFF_UP) { | |
1374 | nb->notifier_call(nb, NETDEV_GOING_DOWN, dev); | |
1375 | nb->notifier_call(nb, NETDEV_DOWN, dev); | |
1376 | } | |
1377 | nb->notifier_call(nb, NETDEV_UNREGISTER, dev); | |
1378 | nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev); | |
1379 | } | |
1380 | } | |
1381 | ||
1382 | raw_notifier_chain_unregister(&netdev_chain, nb); | |
1383 | goto unlock; | |
1384 | } | |
1385 | EXPORT_SYMBOL(register_netdevice_notifier); | |
1386 | ||
1387 | /** | |
1388 | * unregister_netdevice_notifier - unregister a network notifier block | |
1389 | * @nb: notifier | |
1390 | * | |
1391 | * Unregister a notifier previously registered by | |
1392 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1393 | * kernel structures and may then be reused. A negative errno code | |
1394 | * is returned on a failure. | |
1395 | */ | |
1396 | ||
1397 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1398 | { | |
1399 | int err; | |
1400 | ||
1401 | rtnl_lock(); | |
1402 | err = raw_notifier_chain_unregister(&netdev_chain, nb); | |
1403 | rtnl_unlock(); | |
1404 | return err; | |
1405 | } | |
1406 | EXPORT_SYMBOL(unregister_netdevice_notifier); | |
1407 | ||
1408 | /** | |
1409 | * call_netdevice_notifiers - call all network notifier blocks | |
1410 | * @val: value passed unmodified to notifier function | |
1411 | * @dev: net_device pointer passed unmodified to notifier function | |
1412 | * | |
1413 | * Call all network notifier blocks. Parameters and return value | |
1414 | * are as for raw_notifier_call_chain(). | |
1415 | */ | |
1416 | ||
1417 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) | |
1418 | { | |
1419 | return raw_notifier_call_chain(&netdev_chain, val, dev); | |
1420 | } | |
1421 | ||
1422 | /* When > 0 there are consumers of rx skb time stamps */ | |
1423 | static atomic_t netstamp_needed = ATOMIC_INIT(0); | |
1424 | ||
1425 | void net_enable_timestamp(void) | |
1426 | { | |
1427 | atomic_inc(&netstamp_needed); | |
1428 | } | |
1429 | EXPORT_SYMBOL(net_enable_timestamp); | |
1430 | ||
1431 | void net_disable_timestamp(void) | |
1432 | { | |
1433 | atomic_dec(&netstamp_needed); | |
1434 | } | |
1435 | EXPORT_SYMBOL(net_disable_timestamp); | |
1436 | ||
1437 | static inline void net_timestamp(struct sk_buff *skb) | |
1438 | { | |
1439 | if (atomic_read(&netstamp_needed)) | |
1440 | __net_timestamp(skb); | |
1441 | else | |
1442 | skb->tstamp.tv64 = 0; | |
1443 | } | |
1444 | ||
1445 | /** | |
1446 | * dev_forward_skb - loopback an skb to another netif | |
1447 | * | |
1448 | * @dev: destination network device | |
1449 | * @skb: buffer to forward | |
1450 | * | |
1451 | * return values: | |
1452 | * NET_RX_SUCCESS (no congestion) | |
1453 | * NET_RX_DROP (packet was dropped) | |
1454 | * | |
1455 | * dev_forward_skb can be used for injecting an skb from the | |
1456 | * start_xmit function of one device into the receive queue | |
1457 | * of another device. | |
1458 | * | |
1459 | * The receiving device may be in another namespace, so | |
1460 | * we have to clear all information in the skb that could | |
1461 | * impact namespace isolation. | |
1462 | */ | |
1463 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
1464 | { | |
1465 | skb_orphan(skb); | |
1466 | ||
1467 | if (!(dev->flags & IFF_UP)) | |
1468 | return NET_RX_DROP; | |
1469 | ||
1470 | if (skb->len > (dev->mtu + dev->hard_header_len)) | |
1471 | return NET_RX_DROP; | |
1472 | ||
1473 | skb_set_dev(skb, dev); | |
1474 | skb->tstamp.tv64 = 0; | |
1475 | skb->pkt_type = PACKET_HOST; | |
1476 | skb->protocol = eth_type_trans(skb, dev); | |
1477 | return netif_rx(skb); | |
1478 | } | |
1479 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
1480 | ||
1481 | /* | |
1482 | * Support routine. Sends outgoing frames to any network | |
1483 | * taps currently in use. | |
1484 | */ | |
1485 | ||
1486 | static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) | |
1487 | { | |
1488 | struct packet_type *ptype; | |
1489 | ||
1490 | #ifdef CONFIG_NET_CLS_ACT | |
1491 | if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS))) | |
1492 | net_timestamp(skb); | |
1493 | #else | |
1494 | net_timestamp(skb); | |
1495 | #endif | |
1496 | ||
1497 | rcu_read_lock(); | |
1498 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
1499 | /* Never send packets back to the socket | |
1500 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1501 | */ | |
1502 | if ((ptype->dev == dev || !ptype->dev) && | |
1503 | (ptype->af_packet_priv == NULL || | |
1504 | (struct sock *)ptype->af_packet_priv != skb->sk)) { | |
1505 | struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC); | |
1506 | if (!skb2) | |
1507 | break; | |
1508 | ||
1509 | /* skb->nh should be correctly | |
1510 | set by sender, so that the second statement is | |
1511 | just protection against buggy protocols. | |
1512 | */ | |
1513 | skb_reset_mac_header(skb2); | |
1514 | ||
1515 | if (skb_network_header(skb2) < skb2->data || | |
1516 | skb2->network_header > skb2->tail) { | |
1517 | if (net_ratelimit()) | |
1518 | printk(KERN_CRIT "protocol %04x is " | |
1519 | "buggy, dev %s\n", | |
1520 | skb2->protocol, dev->name); | |
1521 | skb_reset_network_header(skb2); | |
1522 | } | |
1523 | ||
1524 | skb2->transport_header = skb2->network_header; | |
1525 | skb2->pkt_type = PACKET_OUTGOING; | |
1526 | ptype->func(skb2, skb->dev, ptype, skb->dev); | |
1527 | } | |
1528 | } | |
1529 | rcu_read_unlock(); | |
1530 | } | |
1531 | ||
1532 | ||
1533 | static inline void __netif_reschedule(struct Qdisc *q) | |
1534 | { | |
1535 | struct softnet_data *sd; | |
1536 | unsigned long flags; | |
1537 | ||
1538 | local_irq_save(flags); | |
1539 | sd = &__get_cpu_var(softnet_data); | |
1540 | q->next_sched = sd->output_queue; | |
1541 | sd->output_queue = q; | |
1542 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
1543 | local_irq_restore(flags); | |
1544 | } | |
1545 | ||
1546 | void __netif_schedule(struct Qdisc *q) | |
1547 | { | |
1548 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
1549 | __netif_reschedule(q); | |
1550 | } | |
1551 | EXPORT_SYMBOL(__netif_schedule); | |
1552 | ||
1553 | void dev_kfree_skb_irq(struct sk_buff *skb) | |
1554 | { | |
1555 | if (atomic_dec_and_test(&skb->users)) { | |
1556 | struct softnet_data *sd; | |
1557 | unsigned long flags; | |
1558 | ||
1559 | local_irq_save(flags); | |
1560 | sd = &__get_cpu_var(softnet_data); | |
1561 | skb->next = sd->completion_queue; | |
1562 | sd->completion_queue = skb; | |
1563 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
1564 | local_irq_restore(flags); | |
1565 | } | |
1566 | } | |
1567 | EXPORT_SYMBOL(dev_kfree_skb_irq); | |
1568 | ||
1569 | void dev_kfree_skb_any(struct sk_buff *skb) | |
1570 | { | |
1571 | if (in_irq() || irqs_disabled()) | |
1572 | dev_kfree_skb_irq(skb); | |
1573 | else | |
1574 | dev_kfree_skb(skb); | |
1575 | } | |
1576 | EXPORT_SYMBOL(dev_kfree_skb_any); | |
1577 | ||
1578 | ||
1579 | /** | |
1580 | * netif_device_detach - mark device as removed | |
1581 | * @dev: network device | |
1582 | * | |
1583 | * Mark device as removed from system and therefore no longer available. | |
1584 | */ | |
1585 | void netif_device_detach(struct net_device *dev) | |
1586 | { | |
1587 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
1588 | netif_running(dev)) { | |
1589 | netif_tx_stop_all_queues(dev); | |
1590 | } | |
1591 | } | |
1592 | EXPORT_SYMBOL(netif_device_detach); | |
1593 | ||
1594 | /** | |
1595 | * netif_device_attach - mark device as attached | |
1596 | * @dev: network device | |
1597 | * | |
1598 | * Mark device as attached from system and restart if needed. | |
1599 | */ | |
1600 | void netif_device_attach(struct net_device *dev) | |
1601 | { | |
1602 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
1603 | netif_running(dev)) { | |
1604 | netif_tx_wake_all_queues(dev); | |
1605 | __netdev_watchdog_up(dev); | |
1606 | } | |
1607 | } | |
1608 | EXPORT_SYMBOL(netif_device_attach); | |
1609 | ||
1610 | static bool can_checksum_protocol(unsigned long features, __be16 protocol) | |
1611 | { | |
1612 | return ((features & NETIF_F_GEN_CSUM) || | |
1613 | ((features & NETIF_F_IP_CSUM) && | |
1614 | protocol == htons(ETH_P_IP)) || | |
1615 | ((features & NETIF_F_IPV6_CSUM) && | |
1616 | protocol == htons(ETH_P_IPV6)) || | |
1617 | ((features & NETIF_F_FCOE_CRC) && | |
1618 | protocol == htons(ETH_P_FCOE))); | |
1619 | } | |
1620 | ||
1621 | static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb) | |
1622 | { | |
1623 | if (can_checksum_protocol(dev->features, skb->protocol)) | |
1624 | return true; | |
1625 | ||
1626 | if (skb->protocol == htons(ETH_P_8021Q)) { | |
1627 | struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data; | |
1628 | if (can_checksum_protocol(dev->features & dev->vlan_features, | |
1629 | veh->h_vlan_encapsulated_proto)) | |
1630 | return true; | |
1631 | } | |
1632 | ||
1633 | return false; | |
1634 | } | |
1635 | ||
1636 | /** | |
1637 | * skb_dev_set -- assign a new device to a buffer | |
1638 | * @skb: buffer for the new device | |
1639 | * @dev: network device | |
1640 | * | |
1641 | * If an skb is owned by a device already, we have to reset | |
1642 | * all data private to the namespace a device belongs to | |
1643 | * before assigning it a new device. | |
1644 | */ | |
1645 | #ifdef CONFIG_NET_NS | |
1646 | void skb_set_dev(struct sk_buff *skb, struct net_device *dev) | |
1647 | { | |
1648 | skb_dst_drop(skb); | |
1649 | if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) { | |
1650 | secpath_reset(skb); | |
1651 | nf_reset(skb); | |
1652 | skb_init_secmark(skb); | |
1653 | skb->mark = 0; | |
1654 | skb->priority = 0; | |
1655 | skb->nf_trace = 0; | |
1656 | skb->ipvs_property = 0; | |
1657 | #ifdef CONFIG_NET_SCHED | |
1658 | skb->tc_index = 0; | |
1659 | #endif | |
1660 | } | |
1661 | skb->dev = dev; | |
1662 | } | |
1663 | EXPORT_SYMBOL(skb_set_dev); | |
1664 | #endif /* CONFIG_NET_NS */ | |
1665 | ||
1666 | /* | |
1667 | * Invalidate hardware checksum when packet is to be mangled, and | |
1668 | * complete checksum manually on outgoing path. | |
1669 | */ | |
1670 | int skb_checksum_help(struct sk_buff *skb) | |
1671 | { | |
1672 | __wsum csum; | |
1673 | int ret = 0, offset; | |
1674 | ||
1675 | if (skb->ip_summed == CHECKSUM_COMPLETE) | |
1676 | goto out_set_summed; | |
1677 | ||
1678 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
1679 | /* Let GSO fix up the checksum. */ | |
1680 | goto out_set_summed; | |
1681 | } | |
1682 | ||
1683 | offset = skb->csum_start - skb_headroom(skb); | |
1684 | BUG_ON(offset >= skb_headlen(skb)); | |
1685 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
1686 | ||
1687 | offset += skb->csum_offset; | |
1688 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
1689 | ||
1690 | if (skb_cloned(skb) && | |
1691 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1692 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); | |
1693 | if (ret) | |
1694 | goto out; | |
1695 | } | |
1696 | ||
1697 | *(__sum16 *)(skb->data + offset) = csum_fold(csum); | |
1698 | out_set_summed: | |
1699 | skb->ip_summed = CHECKSUM_NONE; | |
1700 | out: | |
1701 | return ret; | |
1702 | } | |
1703 | EXPORT_SYMBOL(skb_checksum_help); | |
1704 | ||
1705 | /** | |
1706 | * skb_gso_segment - Perform segmentation on skb. | |
1707 | * @skb: buffer to segment | |
1708 | * @features: features for the output path (see dev->features) | |
1709 | * | |
1710 | * This function segments the given skb and returns a list of segments. | |
1711 | * | |
1712 | * It may return NULL if the skb requires no segmentation. This is | |
1713 | * only possible when GSO is used for verifying header integrity. | |
1714 | */ | |
1715 | struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features) | |
1716 | { | |
1717 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
1718 | struct packet_type *ptype; | |
1719 | __be16 type = skb->protocol; | |
1720 | int err; | |
1721 | ||
1722 | skb_reset_mac_header(skb); | |
1723 | skb->mac_len = skb->network_header - skb->mac_header; | |
1724 | __skb_pull(skb, skb->mac_len); | |
1725 | ||
1726 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { | |
1727 | struct net_device *dev = skb->dev; | |
1728 | struct ethtool_drvinfo info = {}; | |
1729 | ||
1730 | if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo) | |
1731 | dev->ethtool_ops->get_drvinfo(dev, &info); | |
1732 | ||
1733 | WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d " | |
1734 | "ip_summed=%d", | |
1735 | info.driver, dev ? dev->features : 0L, | |
1736 | skb->sk ? skb->sk->sk_route_caps : 0L, | |
1737 | skb->len, skb->data_len, skb->ip_summed); | |
1738 | ||
1739 | if (skb_header_cloned(skb) && | |
1740 | (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))) | |
1741 | return ERR_PTR(err); | |
1742 | } | |
1743 | ||
1744 | rcu_read_lock(); | |
1745 | list_for_each_entry_rcu(ptype, | |
1746 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
1747 | if (ptype->type == type && !ptype->dev && ptype->gso_segment) { | |
1748 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { | |
1749 | err = ptype->gso_send_check(skb); | |
1750 | segs = ERR_PTR(err); | |
1751 | if (err || skb_gso_ok(skb, features)) | |
1752 | break; | |
1753 | __skb_push(skb, (skb->data - | |
1754 | skb_network_header(skb))); | |
1755 | } | |
1756 | segs = ptype->gso_segment(skb, features); | |
1757 | break; | |
1758 | } | |
1759 | } | |
1760 | rcu_read_unlock(); | |
1761 | ||
1762 | __skb_push(skb, skb->data - skb_mac_header(skb)); | |
1763 | ||
1764 | return segs; | |
1765 | } | |
1766 | EXPORT_SYMBOL(skb_gso_segment); | |
1767 | ||
1768 | /* Take action when hardware reception checksum errors are detected. */ | |
1769 | #ifdef CONFIG_BUG | |
1770 | void netdev_rx_csum_fault(struct net_device *dev) | |
1771 | { | |
1772 | if (net_ratelimit()) { | |
1773 | printk(KERN_ERR "%s: hw csum failure.\n", | |
1774 | dev ? dev->name : "<unknown>"); | |
1775 | dump_stack(); | |
1776 | } | |
1777 | } | |
1778 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
1779 | #endif | |
1780 | ||
1781 | /* Actually, we should eliminate this check as soon as we know, that: | |
1782 | * 1. IOMMU is present and allows to map all the memory. | |
1783 | * 2. No high memory really exists on this machine. | |
1784 | */ | |
1785 | ||
1786 | static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb) | |
1787 | { | |
1788 | #ifdef CONFIG_HIGHMEM | |
1789 | int i; | |
1790 | ||
1791 | if (dev->features & NETIF_F_HIGHDMA) | |
1792 | return 0; | |
1793 | ||
1794 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) | |
1795 | if (PageHighMem(skb_shinfo(skb)->frags[i].page)) | |
1796 | return 1; | |
1797 | ||
1798 | #endif | |
1799 | return 0; | |
1800 | } | |
1801 | ||
1802 | struct dev_gso_cb { | |
1803 | void (*destructor)(struct sk_buff *skb); | |
1804 | }; | |
1805 | ||
1806 | #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb) | |
1807 | ||
1808 | static void dev_gso_skb_destructor(struct sk_buff *skb) | |
1809 | { | |
1810 | struct dev_gso_cb *cb; | |
1811 | ||
1812 | do { | |
1813 | struct sk_buff *nskb = skb->next; | |
1814 | ||
1815 | skb->next = nskb->next; | |
1816 | nskb->next = NULL; | |
1817 | kfree_skb(nskb); | |
1818 | } while (skb->next); | |
1819 | ||
1820 | cb = DEV_GSO_CB(skb); | |
1821 | if (cb->destructor) | |
1822 | cb->destructor(skb); | |
1823 | } | |
1824 | ||
1825 | /** | |
1826 | * dev_gso_segment - Perform emulated hardware segmentation on skb. | |
1827 | * @skb: buffer to segment | |
1828 | * | |
1829 | * This function segments the given skb and stores the list of segments | |
1830 | * in skb->next. | |
1831 | */ | |
1832 | static int dev_gso_segment(struct sk_buff *skb) | |
1833 | { | |
1834 | struct net_device *dev = skb->dev; | |
1835 | struct sk_buff *segs; | |
1836 | int features = dev->features & ~(illegal_highdma(dev, skb) ? | |
1837 | NETIF_F_SG : 0); | |
1838 | ||
1839 | segs = skb_gso_segment(skb, features); | |
1840 | ||
1841 | /* Verifying header integrity only. */ | |
1842 | if (!segs) | |
1843 | return 0; | |
1844 | ||
1845 | if (IS_ERR(segs)) | |
1846 | return PTR_ERR(segs); | |
1847 | ||
1848 | skb->next = segs; | |
1849 | DEV_GSO_CB(skb)->destructor = skb->destructor; | |
1850 | skb->destructor = dev_gso_skb_destructor; | |
1851 | ||
1852 | return 0; | |
1853 | } | |
1854 | ||
1855 | int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev, | |
1856 | struct netdev_queue *txq) | |
1857 | { | |
1858 | const struct net_device_ops *ops = dev->netdev_ops; | |
1859 | int rc = NETDEV_TX_OK; | |
1860 | ||
1861 | if (likely(!skb->next)) { | |
1862 | if (!list_empty(&ptype_all)) | |
1863 | dev_queue_xmit_nit(skb, dev); | |
1864 | ||
1865 | if (netif_needs_gso(dev, skb)) { | |
1866 | if (unlikely(dev_gso_segment(skb))) | |
1867 | goto out_kfree_skb; | |
1868 | if (skb->next) | |
1869 | goto gso; | |
1870 | } | |
1871 | ||
1872 | /* | |
1873 | * If device doesnt need skb->dst, release it right now while | |
1874 | * its hot in this cpu cache | |
1875 | */ | |
1876 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
1877 | skb_dst_drop(skb); | |
1878 | ||
1879 | rc = ops->ndo_start_xmit(skb, dev); | |
1880 | if (rc == NETDEV_TX_OK) | |
1881 | txq_trans_update(txq); | |
1882 | /* | |
1883 | * TODO: if skb_orphan() was called by | |
1884 | * dev->hard_start_xmit() (for example, the unmodified | |
1885 | * igb driver does that; bnx2 doesn't), then | |
1886 | * skb_tx_software_timestamp() will be unable to send | |
1887 | * back the time stamp. | |
1888 | * | |
1889 | * How can this be prevented? Always create another | |
1890 | * reference to the socket before calling | |
1891 | * dev->hard_start_xmit()? Prevent that skb_orphan() | |
1892 | * does anything in dev->hard_start_xmit() by clearing | |
1893 | * the skb destructor before the call and restoring it | |
1894 | * afterwards, then doing the skb_orphan() ourselves? | |
1895 | */ | |
1896 | return rc; | |
1897 | } | |
1898 | ||
1899 | gso: | |
1900 | do { | |
1901 | struct sk_buff *nskb = skb->next; | |
1902 | ||
1903 | skb->next = nskb->next; | |
1904 | nskb->next = NULL; | |
1905 | ||
1906 | /* | |
1907 | * If device doesnt need nskb->dst, release it right now while | |
1908 | * its hot in this cpu cache | |
1909 | */ | |
1910 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
1911 | skb_dst_drop(nskb); | |
1912 | ||
1913 | rc = ops->ndo_start_xmit(nskb, dev); | |
1914 | if (unlikely(rc != NETDEV_TX_OK)) { | |
1915 | if (rc & ~NETDEV_TX_MASK) | |
1916 | goto out_kfree_gso_skb; | |
1917 | nskb->next = skb->next; | |
1918 | skb->next = nskb; | |
1919 | return rc; | |
1920 | } | |
1921 | txq_trans_update(txq); | |
1922 | if (unlikely(netif_tx_queue_stopped(txq) && skb->next)) | |
1923 | return NETDEV_TX_BUSY; | |
1924 | } while (skb->next); | |
1925 | ||
1926 | out_kfree_gso_skb: | |
1927 | if (likely(skb->next == NULL)) | |
1928 | skb->destructor = DEV_GSO_CB(skb)->destructor; | |
1929 | out_kfree_skb: | |
1930 | kfree_skb(skb); | |
1931 | return rc; | |
1932 | } | |
1933 | ||
1934 | static u32 hashrnd __read_mostly; | |
1935 | ||
1936 | u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb) | |
1937 | { | |
1938 | u32 hash; | |
1939 | ||
1940 | if (skb_rx_queue_recorded(skb)) { | |
1941 | hash = skb_get_rx_queue(skb); | |
1942 | while (unlikely(hash >= dev->real_num_tx_queues)) | |
1943 | hash -= dev->real_num_tx_queues; | |
1944 | return hash; | |
1945 | } | |
1946 | ||
1947 | if (skb->sk && skb->sk->sk_hash) | |
1948 | hash = skb->sk->sk_hash; | |
1949 | else | |
1950 | hash = skb->protocol; | |
1951 | ||
1952 | hash = jhash_1word(hash, hashrnd); | |
1953 | ||
1954 | return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32); | |
1955 | } | |
1956 | EXPORT_SYMBOL(skb_tx_hash); | |
1957 | ||
1958 | static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index) | |
1959 | { | |
1960 | if (unlikely(queue_index >= dev->real_num_tx_queues)) { | |
1961 | if (net_ratelimit()) { | |
1962 | netdev_warn(dev, "selects TX queue %d, but " | |
1963 | "real number of TX queues is %d\n", | |
1964 | queue_index, dev->real_num_tx_queues); | |
1965 | } | |
1966 | return 0; | |
1967 | } | |
1968 | return queue_index; | |
1969 | } | |
1970 | ||
1971 | static struct netdev_queue *dev_pick_tx(struct net_device *dev, | |
1972 | struct sk_buff *skb) | |
1973 | { | |
1974 | u16 queue_index; | |
1975 | struct sock *sk = skb->sk; | |
1976 | ||
1977 | if (sk_tx_queue_recorded(sk)) { | |
1978 | queue_index = sk_tx_queue_get(sk); | |
1979 | } else { | |
1980 | const struct net_device_ops *ops = dev->netdev_ops; | |
1981 | ||
1982 | if (ops->ndo_select_queue) { | |
1983 | queue_index = ops->ndo_select_queue(dev, skb); | |
1984 | queue_index = dev_cap_txqueue(dev, queue_index); | |
1985 | } else { | |
1986 | queue_index = 0; | |
1987 | if (dev->real_num_tx_queues > 1) | |
1988 | queue_index = skb_tx_hash(dev, skb); | |
1989 | ||
1990 | if (sk && sk->sk_dst_cache) | |
1991 | sk_tx_queue_set(sk, queue_index); | |
1992 | } | |
1993 | } | |
1994 | ||
1995 | skb_set_queue_mapping(skb, queue_index); | |
1996 | return netdev_get_tx_queue(dev, queue_index); | |
1997 | } | |
1998 | ||
1999 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, | |
2000 | struct net_device *dev, | |
2001 | struct netdev_queue *txq) | |
2002 | { | |
2003 | spinlock_t *root_lock = qdisc_lock(q); | |
2004 | int rc; | |
2005 | ||
2006 | spin_lock(root_lock); | |
2007 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
2008 | kfree_skb(skb); | |
2009 | rc = NET_XMIT_DROP; | |
2010 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
2011 | !test_and_set_bit(__QDISC_STATE_RUNNING, &q->state)) { | |
2012 | /* | |
2013 | * This is a work-conserving queue; there are no old skbs | |
2014 | * waiting to be sent out; and the qdisc is not running - | |
2015 | * xmit the skb directly. | |
2016 | */ | |
2017 | __qdisc_update_bstats(q, skb->len); | |
2018 | if (sch_direct_xmit(skb, q, dev, txq, root_lock)) | |
2019 | __qdisc_run(q); | |
2020 | else | |
2021 | clear_bit(__QDISC_STATE_RUNNING, &q->state); | |
2022 | ||
2023 | rc = NET_XMIT_SUCCESS; | |
2024 | } else { | |
2025 | rc = qdisc_enqueue_root(skb, q); | |
2026 | qdisc_run(q); | |
2027 | } | |
2028 | spin_unlock(root_lock); | |
2029 | ||
2030 | return rc; | |
2031 | } | |
2032 | ||
2033 | /* | |
2034 | * Returns true if either: | |
2035 | * 1. skb has frag_list and the device doesn't support FRAGLIST, or | |
2036 | * 2. skb is fragmented and the device does not support SG, or if | |
2037 | * at least one of fragments is in highmem and device does not | |
2038 | * support DMA from it. | |
2039 | */ | |
2040 | static inline int skb_needs_linearize(struct sk_buff *skb, | |
2041 | struct net_device *dev) | |
2042 | { | |
2043 | return (skb_has_frags(skb) && !(dev->features & NETIF_F_FRAGLIST)) || | |
2044 | (skb_shinfo(skb)->nr_frags && (!(dev->features & NETIF_F_SG) || | |
2045 | illegal_highdma(dev, skb))); | |
2046 | } | |
2047 | ||
2048 | /** | |
2049 | * dev_queue_xmit - transmit a buffer | |
2050 | * @skb: buffer to transmit | |
2051 | * | |
2052 | * Queue a buffer for transmission to a network device. The caller must | |
2053 | * have set the device and priority and built the buffer before calling | |
2054 | * this function. The function can be called from an interrupt. | |
2055 | * | |
2056 | * A negative errno code is returned on a failure. A success does not | |
2057 | * guarantee the frame will be transmitted as it may be dropped due | |
2058 | * to congestion or traffic shaping. | |
2059 | * | |
2060 | * ----------------------------------------------------------------------------------- | |
2061 | * I notice this method can also return errors from the queue disciplines, | |
2062 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
2063 | * be positive. | |
2064 | * | |
2065 | * Regardless of the return value, the skb is consumed, so it is currently | |
2066 | * difficult to retry a send to this method. (You can bump the ref count | |
2067 | * before sending to hold a reference for retry if you are careful.) | |
2068 | * | |
2069 | * When calling this method, interrupts MUST be enabled. This is because | |
2070 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
2071 | * --BLG | |
2072 | */ | |
2073 | int dev_queue_xmit(struct sk_buff *skb) | |
2074 | { | |
2075 | struct net_device *dev = skb->dev; | |
2076 | struct netdev_queue *txq; | |
2077 | struct Qdisc *q; | |
2078 | int rc = -ENOMEM; | |
2079 | ||
2080 | /* GSO will handle the following emulations directly. */ | |
2081 | if (netif_needs_gso(dev, skb)) | |
2082 | goto gso; | |
2083 | ||
2084 | /* Convert a paged skb to linear, if required */ | |
2085 | if (skb_needs_linearize(skb, dev) && __skb_linearize(skb)) | |
2086 | goto out_kfree_skb; | |
2087 | ||
2088 | /* If packet is not checksummed and device does not support | |
2089 | * checksumming for this protocol, complete checksumming here. | |
2090 | */ | |
2091 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
2092 | skb_set_transport_header(skb, skb->csum_start - | |
2093 | skb_headroom(skb)); | |
2094 | if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb)) | |
2095 | goto out_kfree_skb; | |
2096 | } | |
2097 | ||
2098 | gso: | |
2099 | /* Disable soft irqs for various locks below. Also | |
2100 | * stops preemption for RCU. | |
2101 | */ | |
2102 | rcu_read_lock_bh(); | |
2103 | ||
2104 | txq = dev_pick_tx(dev, skb); | |
2105 | q = rcu_dereference_bh(txq->qdisc); | |
2106 | ||
2107 | #ifdef CONFIG_NET_CLS_ACT | |
2108 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS); | |
2109 | #endif | |
2110 | if (q->enqueue) { | |
2111 | rc = __dev_xmit_skb(skb, q, dev, txq); | |
2112 | goto out; | |
2113 | } | |
2114 | ||
2115 | /* The device has no queue. Common case for software devices: | |
2116 | loopback, all the sorts of tunnels... | |
2117 | ||
2118 | Really, it is unlikely that netif_tx_lock protection is necessary | |
2119 | here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
2120 | counters.) | |
2121 | However, it is possible, that they rely on protection | |
2122 | made by us here. | |
2123 | ||
2124 | Check this and shot the lock. It is not prone from deadlocks. | |
2125 | Either shot noqueue qdisc, it is even simpler 8) | |
2126 | */ | |
2127 | if (dev->flags & IFF_UP) { | |
2128 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
2129 | ||
2130 | if (txq->xmit_lock_owner != cpu) { | |
2131 | ||
2132 | HARD_TX_LOCK(dev, txq, cpu); | |
2133 | ||
2134 | if (!netif_tx_queue_stopped(txq)) { | |
2135 | rc = dev_hard_start_xmit(skb, dev, txq); | |
2136 | if (dev_xmit_complete(rc)) { | |
2137 | HARD_TX_UNLOCK(dev, txq); | |
2138 | goto out; | |
2139 | } | |
2140 | } | |
2141 | HARD_TX_UNLOCK(dev, txq); | |
2142 | if (net_ratelimit()) | |
2143 | printk(KERN_CRIT "Virtual device %s asks to " | |
2144 | "queue packet!\n", dev->name); | |
2145 | } else { | |
2146 | /* Recursion is detected! It is possible, | |
2147 | * unfortunately */ | |
2148 | if (net_ratelimit()) | |
2149 | printk(KERN_CRIT "Dead loop on virtual device " | |
2150 | "%s, fix it urgently!\n", dev->name); | |
2151 | } | |
2152 | } | |
2153 | ||
2154 | rc = -ENETDOWN; | |
2155 | rcu_read_unlock_bh(); | |
2156 | ||
2157 | out_kfree_skb: | |
2158 | kfree_skb(skb); | |
2159 | return rc; | |
2160 | out: | |
2161 | rcu_read_unlock_bh(); | |
2162 | return rc; | |
2163 | } | |
2164 | EXPORT_SYMBOL(dev_queue_xmit); | |
2165 | ||
2166 | ||
2167 | /*======================================================================= | |
2168 | Receiver routines | |
2169 | =======================================================================*/ | |
2170 | ||
2171 | int netdev_max_backlog __read_mostly = 1000; | |
2172 | int netdev_budget __read_mostly = 300; | |
2173 | int weight_p __read_mostly = 64; /* old backlog weight */ | |
2174 | ||
2175 | DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, }; | |
2176 | ||
2177 | /* | |
2178 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
2179 | * CPU from the RPS map of the receiving queue for a given skb. | |
2180 | */ | |
2181 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb) | |
2182 | { | |
2183 | struct ipv6hdr *ip6; | |
2184 | struct iphdr *ip; | |
2185 | struct netdev_rx_queue *rxqueue; | |
2186 | struct rps_map *map; | |
2187 | int cpu = -1; | |
2188 | u8 ip_proto; | |
2189 | u32 addr1, addr2, ports, ihl; | |
2190 | ||
2191 | rcu_read_lock(); | |
2192 | ||
2193 | if (skb_rx_queue_recorded(skb)) { | |
2194 | u16 index = skb_get_rx_queue(skb); | |
2195 | if (unlikely(index >= dev->num_rx_queues)) { | |
2196 | if (net_ratelimit()) { | |
2197 | netdev_warn(dev, "received packet on queue " | |
2198 | "%u, but number of RX queues is %u\n", | |
2199 | index, dev->num_rx_queues); | |
2200 | } | |
2201 | goto done; | |
2202 | } | |
2203 | rxqueue = dev->_rx + index; | |
2204 | } else | |
2205 | rxqueue = dev->_rx; | |
2206 | ||
2207 | if (!rxqueue->rps_map) | |
2208 | goto done; | |
2209 | ||
2210 | if (skb->rxhash) | |
2211 | goto got_hash; /* Skip hash computation on packet header */ | |
2212 | ||
2213 | switch (skb->protocol) { | |
2214 | case __constant_htons(ETH_P_IP): | |
2215 | if (!pskb_may_pull(skb, sizeof(*ip))) | |
2216 | goto done; | |
2217 | ||
2218 | ip = (struct iphdr *) skb->data; | |
2219 | ip_proto = ip->protocol; | |
2220 | addr1 = ip->saddr; | |
2221 | addr2 = ip->daddr; | |
2222 | ihl = ip->ihl; | |
2223 | break; | |
2224 | case __constant_htons(ETH_P_IPV6): | |
2225 | if (!pskb_may_pull(skb, sizeof(*ip6))) | |
2226 | goto done; | |
2227 | ||
2228 | ip6 = (struct ipv6hdr *) skb->data; | |
2229 | ip_proto = ip6->nexthdr; | |
2230 | addr1 = ip6->saddr.s6_addr32[3]; | |
2231 | addr2 = ip6->daddr.s6_addr32[3]; | |
2232 | ihl = (40 >> 2); | |
2233 | break; | |
2234 | default: | |
2235 | goto done; | |
2236 | } | |
2237 | ports = 0; | |
2238 | switch (ip_proto) { | |
2239 | case IPPROTO_TCP: | |
2240 | case IPPROTO_UDP: | |
2241 | case IPPROTO_DCCP: | |
2242 | case IPPROTO_ESP: | |
2243 | case IPPROTO_AH: | |
2244 | case IPPROTO_SCTP: | |
2245 | case IPPROTO_UDPLITE: | |
2246 | if (pskb_may_pull(skb, (ihl * 4) + 4)) | |
2247 | ports = *((u32 *) (skb->data + (ihl * 4))); | |
2248 | break; | |
2249 | ||
2250 | default: | |
2251 | break; | |
2252 | } | |
2253 | ||
2254 | skb->rxhash = jhash_3words(addr1, addr2, ports, hashrnd); | |
2255 | if (!skb->rxhash) | |
2256 | skb->rxhash = 1; | |
2257 | ||
2258 | got_hash: | |
2259 | map = rcu_dereference(rxqueue->rps_map); | |
2260 | if (map) { | |
2261 | u16 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32]; | |
2262 | ||
2263 | if (cpu_online(tcpu)) { | |
2264 | cpu = tcpu; | |
2265 | goto done; | |
2266 | } | |
2267 | } | |
2268 | ||
2269 | done: | |
2270 | rcu_read_unlock(); | |
2271 | return cpu; | |
2272 | } | |
2273 | ||
2274 | /* | |
2275 | * This structure holds the per-CPU mask of CPUs for which IPIs are scheduled | |
2276 | * to be sent to kick remote softirq processing. There are two masks since | |
2277 | * the sending of IPIs must be done with interrupts enabled. The select field | |
2278 | * indicates the current mask that enqueue_backlog uses to schedule IPIs. | |
2279 | * select is flipped before net_rps_action is called while still under lock, | |
2280 | * net_rps_action then uses the non-selected mask to send the IPIs and clears | |
2281 | * it without conflicting with enqueue_backlog operation. | |
2282 | */ | |
2283 | struct rps_remote_softirq_cpus { | |
2284 | cpumask_t mask[2]; | |
2285 | int select; | |
2286 | }; | |
2287 | static DEFINE_PER_CPU(struct rps_remote_softirq_cpus, rps_remote_softirq_cpus); | |
2288 | ||
2289 | /* Called from hardirq (IPI) context */ | |
2290 | static void trigger_softirq(void *data) | |
2291 | { | |
2292 | struct softnet_data *queue = data; | |
2293 | __napi_schedule(&queue->backlog); | |
2294 | __get_cpu_var(netdev_rx_stat).received_rps++; | |
2295 | } | |
2296 | ||
2297 | /* | |
2298 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
2299 | * queue (may be a remote CPU queue). | |
2300 | */ | |
2301 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu) | |
2302 | { | |
2303 | struct softnet_data *queue; | |
2304 | unsigned long flags; | |
2305 | ||
2306 | queue = &per_cpu(softnet_data, cpu); | |
2307 | ||
2308 | local_irq_save(flags); | |
2309 | __get_cpu_var(netdev_rx_stat).total++; | |
2310 | ||
2311 | spin_lock(&queue->input_pkt_queue.lock); | |
2312 | if (queue->input_pkt_queue.qlen <= netdev_max_backlog) { | |
2313 | if (queue->input_pkt_queue.qlen) { | |
2314 | enqueue: | |
2315 | __skb_queue_tail(&queue->input_pkt_queue, skb); | |
2316 | spin_unlock_irqrestore(&queue->input_pkt_queue.lock, | |
2317 | flags); | |
2318 | return NET_RX_SUCCESS; | |
2319 | } | |
2320 | ||
2321 | /* Schedule NAPI for backlog device */ | |
2322 | if (napi_schedule_prep(&queue->backlog)) { | |
2323 | if (cpu != smp_processor_id()) { | |
2324 | struct rps_remote_softirq_cpus *rcpus = | |
2325 | &__get_cpu_var(rps_remote_softirq_cpus); | |
2326 | ||
2327 | cpu_set(cpu, rcpus->mask[rcpus->select]); | |
2328 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2329 | } else | |
2330 | __napi_schedule(&queue->backlog); | |
2331 | } | |
2332 | goto enqueue; | |
2333 | } | |
2334 | ||
2335 | spin_unlock(&queue->input_pkt_queue.lock); | |
2336 | ||
2337 | __get_cpu_var(netdev_rx_stat).dropped++; | |
2338 | local_irq_restore(flags); | |
2339 | ||
2340 | kfree_skb(skb); | |
2341 | return NET_RX_DROP; | |
2342 | } | |
2343 | ||
2344 | /** | |
2345 | * netif_rx - post buffer to the network code | |
2346 | * @skb: buffer to post | |
2347 | * | |
2348 | * This function receives a packet from a device driver and queues it for | |
2349 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
2350 | * may be dropped during processing for congestion control or by the | |
2351 | * protocol layers. | |
2352 | * | |
2353 | * return values: | |
2354 | * NET_RX_SUCCESS (no congestion) | |
2355 | * NET_RX_DROP (packet was dropped) | |
2356 | * | |
2357 | */ | |
2358 | ||
2359 | int netif_rx(struct sk_buff *skb) | |
2360 | { | |
2361 | int cpu; | |
2362 | ||
2363 | /* if netpoll wants it, pretend we never saw it */ | |
2364 | if (netpoll_rx(skb)) | |
2365 | return NET_RX_DROP; | |
2366 | ||
2367 | if (!skb->tstamp.tv64) | |
2368 | net_timestamp(skb); | |
2369 | ||
2370 | cpu = get_rps_cpu(skb->dev, skb); | |
2371 | if (cpu < 0) | |
2372 | cpu = smp_processor_id(); | |
2373 | ||
2374 | return enqueue_to_backlog(skb, cpu); | |
2375 | } | |
2376 | EXPORT_SYMBOL(netif_rx); | |
2377 | ||
2378 | int netif_rx_ni(struct sk_buff *skb) | |
2379 | { | |
2380 | int err; | |
2381 | ||
2382 | preempt_disable(); | |
2383 | err = netif_rx(skb); | |
2384 | if (local_softirq_pending()) | |
2385 | do_softirq(); | |
2386 | preempt_enable(); | |
2387 | ||
2388 | return err; | |
2389 | } | |
2390 | EXPORT_SYMBOL(netif_rx_ni); | |
2391 | ||
2392 | static void net_tx_action(struct softirq_action *h) | |
2393 | { | |
2394 | struct softnet_data *sd = &__get_cpu_var(softnet_data); | |
2395 | ||
2396 | if (sd->completion_queue) { | |
2397 | struct sk_buff *clist; | |
2398 | ||
2399 | local_irq_disable(); | |
2400 | clist = sd->completion_queue; | |
2401 | sd->completion_queue = NULL; | |
2402 | local_irq_enable(); | |
2403 | ||
2404 | while (clist) { | |
2405 | struct sk_buff *skb = clist; | |
2406 | clist = clist->next; | |
2407 | ||
2408 | WARN_ON(atomic_read(&skb->users)); | |
2409 | __kfree_skb(skb); | |
2410 | } | |
2411 | } | |
2412 | ||
2413 | if (sd->output_queue) { | |
2414 | struct Qdisc *head; | |
2415 | ||
2416 | local_irq_disable(); | |
2417 | head = sd->output_queue; | |
2418 | sd->output_queue = NULL; | |
2419 | local_irq_enable(); | |
2420 | ||
2421 | while (head) { | |
2422 | struct Qdisc *q = head; | |
2423 | spinlock_t *root_lock; | |
2424 | ||
2425 | head = head->next_sched; | |
2426 | ||
2427 | root_lock = qdisc_lock(q); | |
2428 | if (spin_trylock(root_lock)) { | |
2429 | smp_mb__before_clear_bit(); | |
2430 | clear_bit(__QDISC_STATE_SCHED, | |
2431 | &q->state); | |
2432 | qdisc_run(q); | |
2433 | spin_unlock(root_lock); | |
2434 | } else { | |
2435 | if (!test_bit(__QDISC_STATE_DEACTIVATED, | |
2436 | &q->state)) { | |
2437 | __netif_reschedule(q); | |
2438 | } else { | |
2439 | smp_mb__before_clear_bit(); | |
2440 | clear_bit(__QDISC_STATE_SCHED, | |
2441 | &q->state); | |
2442 | } | |
2443 | } | |
2444 | } | |
2445 | } | |
2446 | } | |
2447 | ||
2448 | static inline int deliver_skb(struct sk_buff *skb, | |
2449 | struct packet_type *pt_prev, | |
2450 | struct net_device *orig_dev) | |
2451 | { | |
2452 | atomic_inc(&skb->users); | |
2453 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
2454 | } | |
2455 | ||
2456 | #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE) | |
2457 | ||
2458 | #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE) | |
2459 | /* This hook is defined here for ATM LANE */ | |
2460 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
2461 | unsigned char *addr) __read_mostly; | |
2462 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); | |
2463 | #endif | |
2464 | ||
2465 | /* | |
2466 | * If bridge module is loaded call bridging hook. | |
2467 | * returns NULL if packet was consumed. | |
2468 | */ | |
2469 | struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p, | |
2470 | struct sk_buff *skb) __read_mostly; | |
2471 | EXPORT_SYMBOL_GPL(br_handle_frame_hook); | |
2472 | ||
2473 | static inline struct sk_buff *handle_bridge(struct sk_buff *skb, | |
2474 | struct packet_type **pt_prev, int *ret, | |
2475 | struct net_device *orig_dev) | |
2476 | { | |
2477 | struct net_bridge_port *port; | |
2478 | ||
2479 | if (skb->pkt_type == PACKET_LOOPBACK || | |
2480 | (port = rcu_dereference(skb->dev->br_port)) == NULL) | |
2481 | return skb; | |
2482 | ||
2483 | if (*pt_prev) { | |
2484 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
2485 | *pt_prev = NULL; | |
2486 | } | |
2487 | ||
2488 | return br_handle_frame_hook(port, skb); | |
2489 | } | |
2490 | #else | |
2491 | #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb) | |
2492 | #endif | |
2493 | ||
2494 | #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE) | |
2495 | struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly; | |
2496 | EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook); | |
2497 | ||
2498 | static inline struct sk_buff *handle_macvlan(struct sk_buff *skb, | |
2499 | struct packet_type **pt_prev, | |
2500 | int *ret, | |
2501 | struct net_device *orig_dev) | |
2502 | { | |
2503 | if (skb->dev->macvlan_port == NULL) | |
2504 | return skb; | |
2505 | ||
2506 | if (*pt_prev) { | |
2507 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
2508 | *pt_prev = NULL; | |
2509 | } | |
2510 | return macvlan_handle_frame_hook(skb); | |
2511 | } | |
2512 | #else | |
2513 | #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb) | |
2514 | #endif | |
2515 | ||
2516 | #ifdef CONFIG_NET_CLS_ACT | |
2517 | /* TODO: Maybe we should just force sch_ingress to be compiled in | |
2518 | * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions | |
2519 | * a compare and 2 stores extra right now if we dont have it on | |
2520 | * but have CONFIG_NET_CLS_ACT | |
2521 | * NOTE: This doesnt stop any functionality; if you dont have | |
2522 | * the ingress scheduler, you just cant add policies on ingress. | |
2523 | * | |
2524 | */ | |
2525 | static int ing_filter(struct sk_buff *skb) | |
2526 | { | |
2527 | struct net_device *dev = skb->dev; | |
2528 | u32 ttl = G_TC_RTTL(skb->tc_verd); | |
2529 | struct netdev_queue *rxq; | |
2530 | int result = TC_ACT_OK; | |
2531 | struct Qdisc *q; | |
2532 | ||
2533 | if (MAX_RED_LOOP < ttl++) { | |
2534 | printk(KERN_WARNING | |
2535 | "Redir loop detected Dropping packet (%d->%d)\n", | |
2536 | skb->skb_iif, dev->ifindex); | |
2537 | return TC_ACT_SHOT; | |
2538 | } | |
2539 | ||
2540 | skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl); | |
2541 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); | |
2542 | ||
2543 | rxq = &dev->rx_queue; | |
2544 | ||
2545 | q = rxq->qdisc; | |
2546 | if (q != &noop_qdisc) { | |
2547 | spin_lock(qdisc_lock(q)); | |
2548 | if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) | |
2549 | result = qdisc_enqueue_root(skb, q); | |
2550 | spin_unlock(qdisc_lock(q)); | |
2551 | } | |
2552 | ||
2553 | return result; | |
2554 | } | |
2555 | ||
2556 | static inline struct sk_buff *handle_ing(struct sk_buff *skb, | |
2557 | struct packet_type **pt_prev, | |
2558 | int *ret, struct net_device *orig_dev) | |
2559 | { | |
2560 | if (skb->dev->rx_queue.qdisc == &noop_qdisc) | |
2561 | goto out; | |
2562 | ||
2563 | if (*pt_prev) { | |
2564 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
2565 | *pt_prev = NULL; | |
2566 | } else { | |
2567 | /* Huh? Why does turning on AF_PACKET affect this? */ | |
2568 | skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd); | |
2569 | } | |
2570 | ||
2571 | switch (ing_filter(skb)) { | |
2572 | case TC_ACT_SHOT: | |
2573 | case TC_ACT_STOLEN: | |
2574 | kfree_skb(skb); | |
2575 | return NULL; | |
2576 | } | |
2577 | ||
2578 | out: | |
2579 | skb->tc_verd = 0; | |
2580 | return skb; | |
2581 | } | |
2582 | #endif | |
2583 | ||
2584 | /* | |
2585 | * netif_nit_deliver - deliver received packets to network taps | |
2586 | * @skb: buffer | |
2587 | * | |
2588 | * This function is used to deliver incoming packets to network | |
2589 | * taps. It should be used when the normal netif_receive_skb path | |
2590 | * is bypassed, for example because of VLAN acceleration. | |
2591 | */ | |
2592 | void netif_nit_deliver(struct sk_buff *skb) | |
2593 | { | |
2594 | struct packet_type *ptype; | |
2595 | ||
2596 | if (list_empty(&ptype_all)) | |
2597 | return; | |
2598 | ||
2599 | skb_reset_network_header(skb); | |
2600 | skb_reset_transport_header(skb); | |
2601 | skb->mac_len = skb->network_header - skb->mac_header; | |
2602 | ||
2603 | rcu_read_lock(); | |
2604 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
2605 | if (!ptype->dev || ptype->dev == skb->dev) | |
2606 | deliver_skb(skb, ptype, skb->dev); | |
2607 | } | |
2608 | rcu_read_unlock(); | |
2609 | } | |
2610 | ||
2611 | int __netif_receive_skb(struct sk_buff *skb) | |
2612 | { | |
2613 | struct packet_type *ptype, *pt_prev; | |
2614 | struct net_device *orig_dev; | |
2615 | struct net_device *null_or_orig; | |
2616 | struct net_device *null_or_bond; | |
2617 | int ret = NET_RX_DROP; | |
2618 | __be16 type; | |
2619 | ||
2620 | if (!skb->tstamp.tv64) | |
2621 | net_timestamp(skb); | |
2622 | ||
2623 | if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb)) | |
2624 | return NET_RX_SUCCESS; | |
2625 | ||
2626 | /* if we've gotten here through NAPI, check netpoll */ | |
2627 | if (netpoll_receive_skb(skb)) | |
2628 | return NET_RX_DROP; | |
2629 | ||
2630 | if (!skb->skb_iif) | |
2631 | skb->skb_iif = skb->dev->ifindex; | |
2632 | ||
2633 | null_or_orig = NULL; | |
2634 | orig_dev = skb->dev; | |
2635 | if (orig_dev->master) { | |
2636 | if (skb_bond_should_drop(skb)) | |
2637 | null_or_orig = orig_dev; /* deliver only exact match */ | |
2638 | else | |
2639 | skb->dev = orig_dev->master; | |
2640 | } | |
2641 | ||
2642 | __get_cpu_var(netdev_rx_stat).total++; | |
2643 | ||
2644 | skb_reset_network_header(skb); | |
2645 | skb_reset_transport_header(skb); | |
2646 | skb->mac_len = skb->network_header - skb->mac_header; | |
2647 | ||
2648 | pt_prev = NULL; | |
2649 | ||
2650 | rcu_read_lock(); | |
2651 | ||
2652 | #ifdef CONFIG_NET_CLS_ACT | |
2653 | if (skb->tc_verd & TC_NCLS) { | |
2654 | skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); | |
2655 | goto ncls; | |
2656 | } | |
2657 | #endif | |
2658 | ||
2659 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
2660 | if (ptype->dev == null_or_orig || ptype->dev == skb->dev || | |
2661 | ptype->dev == orig_dev) { | |
2662 | if (pt_prev) | |
2663 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
2664 | pt_prev = ptype; | |
2665 | } | |
2666 | } | |
2667 | ||
2668 | #ifdef CONFIG_NET_CLS_ACT | |
2669 | skb = handle_ing(skb, &pt_prev, &ret, orig_dev); | |
2670 | if (!skb) | |
2671 | goto out; | |
2672 | ncls: | |
2673 | #endif | |
2674 | ||
2675 | skb = handle_bridge(skb, &pt_prev, &ret, orig_dev); | |
2676 | if (!skb) | |
2677 | goto out; | |
2678 | skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev); | |
2679 | if (!skb) | |
2680 | goto out; | |
2681 | ||
2682 | /* | |
2683 | * Make sure frames received on VLAN interfaces stacked on | |
2684 | * bonding interfaces still make their way to any base bonding | |
2685 | * device that may have registered for a specific ptype. The | |
2686 | * handler may have to adjust skb->dev and orig_dev. | |
2687 | */ | |
2688 | null_or_bond = NULL; | |
2689 | if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) && | |
2690 | (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) { | |
2691 | null_or_bond = vlan_dev_real_dev(skb->dev); | |
2692 | } | |
2693 | ||
2694 | type = skb->protocol; | |
2695 | list_for_each_entry_rcu(ptype, | |
2696 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
2697 | if (ptype->type == type && (ptype->dev == null_or_orig || | |
2698 | ptype->dev == skb->dev || ptype->dev == orig_dev || | |
2699 | ptype->dev == null_or_bond)) { | |
2700 | if (pt_prev) | |
2701 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
2702 | pt_prev = ptype; | |
2703 | } | |
2704 | } | |
2705 | ||
2706 | if (pt_prev) { | |
2707 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
2708 | } else { | |
2709 | kfree_skb(skb); | |
2710 | /* Jamal, now you will not able to escape explaining | |
2711 | * me how you were going to use this. :-) | |
2712 | */ | |
2713 | ret = NET_RX_DROP; | |
2714 | } | |
2715 | ||
2716 | out: | |
2717 | rcu_read_unlock(); | |
2718 | return ret; | |
2719 | } | |
2720 | ||
2721 | /** | |
2722 | * netif_receive_skb - process receive buffer from network | |
2723 | * @skb: buffer to process | |
2724 | * | |
2725 | * netif_receive_skb() is the main receive data processing function. | |
2726 | * It always succeeds. The buffer may be dropped during processing | |
2727 | * for congestion control or by the protocol layers. | |
2728 | * | |
2729 | * This function may only be called from softirq context and interrupts | |
2730 | * should be enabled. | |
2731 | * | |
2732 | * Return values (usually ignored): | |
2733 | * NET_RX_SUCCESS: no congestion | |
2734 | * NET_RX_DROP: packet was dropped | |
2735 | */ | |
2736 | int netif_receive_skb(struct sk_buff *skb) | |
2737 | { | |
2738 | int cpu; | |
2739 | ||
2740 | cpu = get_rps_cpu(skb->dev, skb); | |
2741 | ||
2742 | if (cpu < 0) | |
2743 | return __netif_receive_skb(skb); | |
2744 | else | |
2745 | return enqueue_to_backlog(skb, cpu); | |
2746 | } | |
2747 | EXPORT_SYMBOL(netif_receive_skb); | |
2748 | ||
2749 | /* Network device is going away, flush any packets still pending */ | |
2750 | static void flush_backlog(void *arg) | |
2751 | { | |
2752 | struct net_device *dev = arg; | |
2753 | struct softnet_data *queue = &__get_cpu_var(softnet_data); | |
2754 | struct sk_buff *skb, *tmp; | |
2755 | ||
2756 | skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp) | |
2757 | if (skb->dev == dev) { | |
2758 | __skb_unlink(skb, &queue->input_pkt_queue); | |
2759 | kfree_skb(skb); | |
2760 | } | |
2761 | } | |
2762 | ||
2763 | static int napi_gro_complete(struct sk_buff *skb) | |
2764 | { | |
2765 | struct packet_type *ptype; | |
2766 | __be16 type = skb->protocol; | |
2767 | struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK]; | |
2768 | int err = -ENOENT; | |
2769 | ||
2770 | if (NAPI_GRO_CB(skb)->count == 1) { | |
2771 | skb_shinfo(skb)->gso_size = 0; | |
2772 | goto out; | |
2773 | } | |
2774 | ||
2775 | rcu_read_lock(); | |
2776 | list_for_each_entry_rcu(ptype, head, list) { | |
2777 | if (ptype->type != type || ptype->dev || !ptype->gro_complete) | |
2778 | continue; | |
2779 | ||
2780 | err = ptype->gro_complete(skb); | |
2781 | break; | |
2782 | } | |
2783 | rcu_read_unlock(); | |
2784 | ||
2785 | if (err) { | |
2786 | WARN_ON(&ptype->list == head); | |
2787 | kfree_skb(skb); | |
2788 | return NET_RX_SUCCESS; | |
2789 | } | |
2790 | ||
2791 | out: | |
2792 | return netif_receive_skb(skb); | |
2793 | } | |
2794 | ||
2795 | static void napi_gro_flush(struct napi_struct *napi) | |
2796 | { | |
2797 | struct sk_buff *skb, *next; | |
2798 | ||
2799 | for (skb = napi->gro_list; skb; skb = next) { | |
2800 | next = skb->next; | |
2801 | skb->next = NULL; | |
2802 | napi_gro_complete(skb); | |
2803 | } | |
2804 | ||
2805 | napi->gro_count = 0; | |
2806 | napi->gro_list = NULL; | |
2807 | } | |
2808 | ||
2809 | enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) | |
2810 | { | |
2811 | struct sk_buff **pp = NULL; | |
2812 | struct packet_type *ptype; | |
2813 | __be16 type = skb->protocol; | |
2814 | struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK]; | |
2815 | int same_flow; | |
2816 | int mac_len; | |
2817 | enum gro_result ret; | |
2818 | ||
2819 | if (!(skb->dev->features & NETIF_F_GRO)) | |
2820 | goto normal; | |
2821 | ||
2822 | if (skb_is_gso(skb) || skb_has_frags(skb)) | |
2823 | goto normal; | |
2824 | ||
2825 | rcu_read_lock(); | |
2826 | list_for_each_entry_rcu(ptype, head, list) { | |
2827 | if (ptype->type != type || ptype->dev || !ptype->gro_receive) | |
2828 | continue; | |
2829 | ||
2830 | skb_set_network_header(skb, skb_gro_offset(skb)); | |
2831 | mac_len = skb->network_header - skb->mac_header; | |
2832 | skb->mac_len = mac_len; | |
2833 | NAPI_GRO_CB(skb)->same_flow = 0; | |
2834 | NAPI_GRO_CB(skb)->flush = 0; | |
2835 | NAPI_GRO_CB(skb)->free = 0; | |
2836 | ||
2837 | pp = ptype->gro_receive(&napi->gro_list, skb); | |
2838 | break; | |
2839 | } | |
2840 | rcu_read_unlock(); | |
2841 | ||
2842 | if (&ptype->list == head) | |
2843 | goto normal; | |
2844 | ||
2845 | same_flow = NAPI_GRO_CB(skb)->same_flow; | |
2846 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; | |
2847 | ||
2848 | if (pp) { | |
2849 | struct sk_buff *nskb = *pp; | |
2850 | ||
2851 | *pp = nskb->next; | |
2852 | nskb->next = NULL; | |
2853 | napi_gro_complete(nskb); | |
2854 | napi->gro_count--; | |
2855 | } | |
2856 | ||
2857 | if (same_flow) | |
2858 | goto ok; | |
2859 | ||
2860 | if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS) | |
2861 | goto normal; | |
2862 | ||
2863 | napi->gro_count++; | |
2864 | NAPI_GRO_CB(skb)->count = 1; | |
2865 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); | |
2866 | skb->next = napi->gro_list; | |
2867 | napi->gro_list = skb; | |
2868 | ret = GRO_HELD; | |
2869 | ||
2870 | pull: | |
2871 | if (skb_headlen(skb) < skb_gro_offset(skb)) { | |
2872 | int grow = skb_gro_offset(skb) - skb_headlen(skb); | |
2873 | ||
2874 | BUG_ON(skb->end - skb->tail < grow); | |
2875 | ||
2876 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
2877 | ||
2878 | skb->tail += grow; | |
2879 | skb->data_len -= grow; | |
2880 | ||
2881 | skb_shinfo(skb)->frags[0].page_offset += grow; | |
2882 | skb_shinfo(skb)->frags[0].size -= grow; | |
2883 | ||
2884 | if (unlikely(!skb_shinfo(skb)->frags[0].size)) { | |
2885 | put_page(skb_shinfo(skb)->frags[0].page); | |
2886 | memmove(skb_shinfo(skb)->frags, | |
2887 | skb_shinfo(skb)->frags + 1, | |
2888 | --skb_shinfo(skb)->nr_frags); | |
2889 | } | |
2890 | } | |
2891 | ||
2892 | ok: | |
2893 | return ret; | |
2894 | ||
2895 | normal: | |
2896 | ret = GRO_NORMAL; | |
2897 | goto pull; | |
2898 | } | |
2899 | EXPORT_SYMBOL(dev_gro_receive); | |
2900 | ||
2901 | static gro_result_t | |
2902 | __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) | |
2903 | { | |
2904 | struct sk_buff *p; | |
2905 | ||
2906 | if (netpoll_rx_on(skb)) | |
2907 | return GRO_NORMAL; | |
2908 | ||
2909 | for (p = napi->gro_list; p; p = p->next) { | |
2910 | NAPI_GRO_CB(p)->same_flow = | |
2911 | (p->dev == skb->dev) && | |
2912 | !compare_ether_header(skb_mac_header(p), | |
2913 | skb_gro_mac_header(skb)); | |
2914 | NAPI_GRO_CB(p)->flush = 0; | |
2915 | } | |
2916 | ||
2917 | return dev_gro_receive(napi, skb); | |
2918 | } | |
2919 | ||
2920 | gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) | |
2921 | { | |
2922 | switch (ret) { | |
2923 | case GRO_NORMAL: | |
2924 | if (netif_receive_skb(skb)) | |
2925 | ret = GRO_DROP; | |
2926 | break; | |
2927 | ||
2928 | case GRO_DROP: | |
2929 | case GRO_MERGED_FREE: | |
2930 | kfree_skb(skb); | |
2931 | break; | |
2932 | ||
2933 | case GRO_HELD: | |
2934 | case GRO_MERGED: | |
2935 | break; | |
2936 | } | |
2937 | ||
2938 | return ret; | |
2939 | } | |
2940 | EXPORT_SYMBOL(napi_skb_finish); | |
2941 | ||
2942 | void skb_gro_reset_offset(struct sk_buff *skb) | |
2943 | { | |
2944 | NAPI_GRO_CB(skb)->data_offset = 0; | |
2945 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
2946 | NAPI_GRO_CB(skb)->frag0_len = 0; | |
2947 | ||
2948 | if (skb->mac_header == skb->tail && | |
2949 | !PageHighMem(skb_shinfo(skb)->frags[0].page)) { | |
2950 | NAPI_GRO_CB(skb)->frag0 = | |
2951 | page_address(skb_shinfo(skb)->frags[0].page) + | |
2952 | skb_shinfo(skb)->frags[0].page_offset; | |
2953 | NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size; | |
2954 | } | |
2955 | } | |
2956 | EXPORT_SYMBOL(skb_gro_reset_offset); | |
2957 | ||
2958 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) | |
2959 | { | |
2960 | skb_gro_reset_offset(skb); | |
2961 | ||
2962 | return napi_skb_finish(__napi_gro_receive(napi, skb), skb); | |
2963 | } | |
2964 | EXPORT_SYMBOL(napi_gro_receive); | |
2965 | ||
2966 | void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) | |
2967 | { | |
2968 | __skb_pull(skb, skb_headlen(skb)); | |
2969 | skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb)); | |
2970 | ||
2971 | napi->skb = skb; | |
2972 | } | |
2973 | EXPORT_SYMBOL(napi_reuse_skb); | |
2974 | ||
2975 | struct sk_buff *napi_get_frags(struct napi_struct *napi) | |
2976 | { | |
2977 | struct sk_buff *skb = napi->skb; | |
2978 | ||
2979 | if (!skb) { | |
2980 | skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD); | |
2981 | if (skb) | |
2982 | napi->skb = skb; | |
2983 | } | |
2984 | return skb; | |
2985 | } | |
2986 | EXPORT_SYMBOL(napi_get_frags); | |
2987 | ||
2988 | gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb, | |
2989 | gro_result_t ret) | |
2990 | { | |
2991 | switch (ret) { | |
2992 | case GRO_NORMAL: | |
2993 | case GRO_HELD: | |
2994 | skb->protocol = eth_type_trans(skb, skb->dev); | |
2995 | ||
2996 | if (ret == GRO_HELD) | |
2997 | skb_gro_pull(skb, -ETH_HLEN); | |
2998 | else if (netif_receive_skb(skb)) | |
2999 | ret = GRO_DROP; | |
3000 | break; | |
3001 | ||
3002 | case GRO_DROP: | |
3003 | case GRO_MERGED_FREE: | |
3004 | napi_reuse_skb(napi, skb); | |
3005 | break; | |
3006 | ||
3007 | case GRO_MERGED: | |
3008 | break; | |
3009 | } | |
3010 | ||
3011 | return ret; | |
3012 | } | |
3013 | EXPORT_SYMBOL(napi_frags_finish); | |
3014 | ||
3015 | struct sk_buff *napi_frags_skb(struct napi_struct *napi) | |
3016 | { | |
3017 | struct sk_buff *skb = napi->skb; | |
3018 | struct ethhdr *eth; | |
3019 | unsigned int hlen; | |
3020 | unsigned int off; | |
3021 | ||
3022 | napi->skb = NULL; | |
3023 | ||
3024 | skb_reset_mac_header(skb); | |
3025 | skb_gro_reset_offset(skb); | |
3026 | ||
3027 | off = skb_gro_offset(skb); | |
3028 | hlen = off + sizeof(*eth); | |
3029 | eth = skb_gro_header_fast(skb, off); | |
3030 | if (skb_gro_header_hard(skb, hlen)) { | |
3031 | eth = skb_gro_header_slow(skb, hlen, off); | |
3032 | if (unlikely(!eth)) { | |
3033 | napi_reuse_skb(napi, skb); | |
3034 | skb = NULL; | |
3035 | goto out; | |
3036 | } | |
3037 | } | |
3038 | ||
3039 | skb_gro_pull(skb, sizeof(*eth)); | |
3040 | ||
3041 | /* | |
3042 | * This works because the only protocols we care about don't require | |
3043 | * special handling. We'll fix it up properly at the end. | |
3044 | */ | |
3045 | skb->protocol = eth->h_proto; | |
3046 | ||
3047 | out: | |
3048 | return skb; | |
3049 | } | |
3050 | EXPORT_SYMBOL(napi_frags_skb); | |
3051 | ||
3052 | gro_result_t napi_gro_frags(struct napi_struct *napi) | |
3053 | { | |
3054 | struct sk_buff *skb = napi_frags_skb(napi); | |
3055 | ||
3056 | if (!skb) | |
3057 | return GRO_DROP; | |
3058 | ||
3059 | return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb)); | |
3060 | } | |
3061 | EXPORT_SYMBOL(napi_gro_frags); | |
3062 | ||
3063 | static int process_backlog(struct napi_struct *napi, int quota) | |
3064 | { | |
3065 | int work = 0; | |
3066 | struct softnet_data *queue = &__get_cpu_var(softnet_data); | |
3067 | unsigned long start_time = jiffies; | |
3068 | ||
3069 | napi->weight = weight_p; | |
3070 | do { | |
3071 | struct sk_buff *skb; | |
3072 | ||
3073 | spin_lock_irq(&queue->input_pkt_queue.lock); | |
3074 | skb = __skb_dequeue(&queue->input_pkt_queue); | |
3075 | if (!skb) { | |
3076 | __napi_complete(napi); | |
3077 | spin_unlock_irq(&queue->input_pkt_queue.lock); | |
3078 | break; | |
3079 | } | |
3080 | spin_unlock_irq(&queue->input_pkt_queue.lock); | |
3081 | ||
3082 | __netif_receive_skb(skb); | |
3083 | } while (++work < quota && jiffies == start_time); | |
3084 | ||
3085 | return work; | |
3086 | } | |
3087 | ||
3088 | /** | |
3089 | * __napi_schedule - schedule for receive | |
3090 | * @n: entry to schedule | |
3091 | * | |
3092 | * The entry's receive function will be scheduled to run | |
3093 | */ | |
3094 | void __napi_schedule(struct napi_struct *n) | |
3095 | { | |
3096 | unsigned long flags; | |
3097 | ||
3098 | local_irq_save(flags); | |
3099 | list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list); | |
3100 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3101 | local_irq_restore(flags); | |
3102 | } | |
3103 | EXPORT_SYMBOL(__napi_schedule); | |
3104 | ||
3105 | void __napi_complete(struct napi_struct *n) | |
3106 | { | |
3107 | BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); | |
3108 | BUG_ON(n->gro_list); | |
3109 | ||
3110 | list_del(&n->poll_list); | |
3111 | smp_mb__before_clear_bit(); | |
3112 | clear_bit(NAPI_STATE_SCHED, &n->state); | |
3113 | } | |
3114 | EXPORT_SYMBOL(__napi_complete); | |
3115 | ||
3116 | void napi_complete(struct napi_struct *n) | |
3117 | { | |
3118 | unsigned long flags; | |
3119 | ||
3120 | /* | |
3121 | * don't let napi dequeue from the cpu poll list | |
3122 | * just in case its running on a different cpu | |
3123 | */ | |
3124 | if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state))) | |
3125 | return; | |
3126 | ||
3127 | napi_gro_flush(n); | |
3128 | local_irq_save(flags); | |
3129 | __napi_complete(n); | |
3130 | local_irq_restore(flags); | |
3131 | } | |
3132 | EXPORT_SYMBOL(napi_complete); | |
3133 | ||
3134 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, | |
3135 | int (*poll)(struct napi_struct *, int), int weight) | |
3136 | { | |
3137 | INIT_LIST_HEAD(&napi->poll_list); | |
3138 | napi->gro_count = 0; | |
3139 | napi->gro_list = NULL; | |
3140 | napi->skb = NULL; | |
3141 | napi->poll = poll; | |
3142 | napi->weight = weight; | |
3143 | list_add(&napi->dev_list, &dev->napi_list); | |
3144 | napi->dev = dev; | |
3145 | #ifdef CONFIG_NETPOLL | |
3146 | spin_lock_init(&napi->poll_lock); | |
3147 | napi->poll_owner = -1; | |
3148 | #endif | |
3149 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
3150 | } | |
3151 | EXPORT_SYMBOL(netif_napi_add); | |
3152 | ||
3153 | void netif_napi_del(struct napi_struct *napi) | |
3154 | { | |
3155 | struct sk_buff *skb, *next; | |
3156 | ||
3157 | list_del_init(&napi->dev_list); | |
3158 | napi_free_frags(napi); | |
3159 | ||
3160 | for (skb = napi->gro_list; skb; skb = next) { | |
3161 | next = skb->next; | |
3162 | skb->next = NULL; | |
3163 | kfree_skb(skb); | |
3164 | } | |
3165 | ||
3166 | napi->gro_list = NULL; | |
3167 | napi->gro_count = 0; | |
3168 | } | |
3169 | EXPORT_SYMBOL(netif_napi_del); | |
3170 | ||
3171 | /* | |
3172 | * net_rps_action sends any pending IPI's for rps. This is only called from | |
3173 | * softirq and interrupts must be enabled. | |
3174 | */ | |
3175 | static void net_rps_action(cpumask_t *mask) | |
3176 | { | |
3177 | int cpu; | |
3178 | ||
3179 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
3180 | for_each_cpu_mask_nr(cpu, *mask) { | |
3181 | struct softnet_data *queue = &per_cpu(softnet_data, cpu); | |
3182 | if (cpu_online(cpu)) | |
3183 | __smp_call_function_single(cpu, &queue->csd, 0); | |
3184 | } | |
3185 | cpus_clear(*mask); | |
3186 | } | |
3187 | ||
3188 | static void net_rx_action(struct softirq_action *h) | |
3189 | { | |
3190 | struct list_head *list = &__get_cpu_var(softnet_data).poll_list; | |
3191 | unsigned long time_limit = jiffies + 2; | |
3192 | int budget = netdev_budget; | |
3193 | void *have; | |
3194 | int select; | |
3195 | struct rps_remote_softirq_cpus *rcpus; | |
3196 | ||
3197 | local_irq_disable(); | |
3198 | ||
3199 | while (!list_empty(list)) { | |
3200 | struct napi_struct *n; | |
3201 | int work, weight; | |
3202 | ||
3203 | /* If softirq window is exhuasted then punt. | |
3204 | * Allow this to run for 2 jiffies since which will allow | |
3205 | * an average latency of 1.5/HZ. | |
3206 | */ | |
3207 | if (unlikely(budget <= 0 || time_after(jiffies, time_limit))) | |
3208 | goto softnet_break; | |
3209 | ||
3210 | local_irq_enable(); | |
3211 | ||
3212 | /* Even though interrupts have been re-enabled, this | |
3213 | * access is safe because interrupts can only add new | |
3214 | * entries to the tail of this list, and only ->poll() | |
3215 | * calls can remove this head entry from the list. | |
3216 | */ | |
3217 | n = list_first_entry(list, struct napi_struct, poll_list); | |
3218 | ||
3219 | have = netpoll_poll_lock(n); | |
3220 | ||
3221 | weight = n->weight; | |
3222 | ||
3223 | /* This NAPI_STATE_SCHED test is for avoiding a race | |
3224 | * with netpoll's poll_napi(). Only the entity which | |
3225 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
3226 | * actually make the ->poll() call. Therefore we avoid | |
3227 | * accidently calling ->poll() when NAPI is not scheduled. | |
3228 | */ | |
3229 | work = 0; | |
3230 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { | |
3231 | work = n->poll(n, weight); | |
3232 | trace_napi_poll(n); | |
3233 | } | |
3234 | ||
3235 | WARN_ON_ONCE(work > weight); | |
3236 | ||
3237 | budget -= work; | |
3238 | ||
3239 | local_irq_disable(); | |
3240 | ||
3241 | /* Drivers must not modify the NAPI state if they | |
3242 | * consume the entire weight. In such cases this code | |
3243 | * still "owns" the NAPI instance and therefore can | |
3244 | * move the instance around on the list at-will. | |
3245 | */ | |
3246 | if (unlikely(work == weight)) { | |
3247 | if (unlikely(napi_disable_pending(n))) { | |
3248 | local_irq_enable(); | |
3249 | napi_complete(n); | |
3250 | local_irq_disable(); | |
3251 | } else | |
3252 | list_move_tail(&n->poll_list, list); | |
3253 | } | |
3254 | ||
3255 | netpoll_poll_unlock(have); | |
3256 | } | |
3257 | out: | |
3258 | rcpus = &__get_cpu_var(rps_remote_softirq_cpus); | |
3259 | select = rcpus->select; | |
3260 | rcpus->select ^= 1; | |
3261 | ||
3262 | local_irq_enable(); | |
3263 | ||
3264 | net_rps_action(&rcpus->mask[select]); | |
3265 | ||
3266 | #ifdef CONFIG_NET_DMA | |
3267 | /* | |
3268 | * There may not be any more sk_buffs coming right now, so push | |
3269 | * any pending DMA copies to hardware | |
3270 | */ | |
3271 | dma_issue_pending_all(); | |
3272 | #endif | |
3273 | ||
3274 | return; | |
3275 | ||
3276 | softnet_break: | |
3277 | __get_cpu_var(netdev_rx_stat).time_squeeze++; | |
3278 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3279 | goto out; | |
3280 | } | |
3281 | ||
3282 | static gifconf_func_t *gifconf_list[NPROTO]; | |
3283 | ||
3284 | /** | |
3285 | * register_gifconf - register a SIOCGIF handler | |
3286 | * @family: Address family | |
3287 | * @gifconf: Function handler | |
3288 | * | |
3289 | * Register protocol dependent address dumping routines. The handler | |
3290 | * that is passed must not be freed or reused until it has been replaced | |
3291 | * by another handler. | |
3292 | */ | |
3293 | int register_gifconf(unsigned int family, gifconf_func_t *gifconf) | |
3294 | { | |
3295 | if (family >= NPROTO) | |
3296 | return -EINVAL; | |
3297 | gifconf_list[family] = gifconf; | |
3298 | return 0; | |
3299 | } | |
3300 | EXPORT_SYMBOL(register_gifconf); | |
3301 | ||
3302 | ||
3303 | /* | |
3304 | * Map an interface index to its name (SIOCGIFNAME) | |
3305 | */ | |
3306 | ||
3307 | /* | |
3308 | * We need this ioctl for efficient implementation of the | |
3309 | * if_indextoname() function required by the IPv6 API. Without | |
3310 | * it, we would have to search all the interfaces to find a | |
3311 | * match. --pb | |
3312 | */ | |
3313 | ||
3314 | static int dev_ifname(struct net *net, struct ifreq __user *arg) | |
3315 | { | |
3316 | struct net_device *dev; | |
3317 | struct ifreq ifr; | |
3318 | ||
3319 | /* | |
3320 | * Fetch the caller's info block. | |
3321 | */ | |
3322 | ||
3323 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
3324 | return -EFAULT; | |
3325 | ||
3326 | rcu_read_lock(); | |
3327 | dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex); | |
3328 | if (!dev) { | |
3329 | rcu_read_unlock(); | |
3330 | return -ENODEV; | |
3331 | } | |
3332 | ||
3333 | strcpy(ifr.ifr_name, dev->name); | |
3334 | rcu_read_unlock(); | |
3335 | ||
3336 | if (copy_to_user(arg, &ifr, sizeof(struct ifreq))) | |
3337 | return -EFAULT; | |
3338 | return 0; | |
3339 | } | |
3340 | ||
3341 | /* | |
3342 | * Perform a SIOCGIFCONF call. This structure will change | |
3343 | * size eventually, and there is nothing I can do about it. | |
3344 | * Thus we will need a 'compatibility mode'. | |
3345 | */ | |
3346 | ||
3347 | static int dev_ifconf(struct net *net, char __user *arg) | |
3348 | { | |
3349 | struct ifconf ifc; | |
3350 | struct net_device *dev; | |
3351 | char __user *pos; | |
3352 | int len; | |
3353 | int total; | |
3354 | int i; | |
3355 | ||
3356 | /* | |
3357 | * Fetch the caller's info block. | |
3358 | */ | |
3359 | ||
3360 | if (copy_from_user(&ifc, arg, sizeof(struct ifconf))) | |
3361 | return -EFAULT; | |
3362 | ||
3363 | pos = ifc.ifc_buf; | |
3364 | len = ifc.ifc_len; | |
3365 | ||
3366 | /* | |
3367 | * Loop over the interfaces, and write an info block for each. | |
3368 | */ | |
3369 | ||
3370 | total = 0; | |
3371 | for_each_netdev(net, dev) { | |
3372 | for (i = 0; i < NPROTO; i++) { | |
3373 | if (gifconf_list[i]) { | |
3374 | int done; | |
3375 | if (!pos) | |
3376 | done = gifconf_list[i](dev, NULL, 0); | |
3377 | else | |
3378 | done = gifconf_list[i](dev, pos + total, | |
3379 | len - total); | |
3380 | if (done < 0) | |
3381 | return -EFAULT; | |
3382 | total += done; | |
3383 | } | |
3384 | } | |
3385 | } | |
3386 | ||
3387 | /* | |
3388 | * All done. Write the updated control block back to the caller. | |
3389 | */ | |
3390 | ifc.ifc_len = total; | |
3391 | ||
3392 | /* | |
3393 | * Both BSD and Solaris return 0 here, so we do too. | |
3394 | */ | |
3395 | return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0; | |
3396 | } | |
3397 | ||
3398 | #ifdef CONFIG_PROC_FS | |
3399 | /* | |
3400 | * This is invoked by the /proc filesystem handler to display a device | |
3401 | * in detail. | |
3402 | */ | |
3403 | void *dev_seq_start(struct seq_file *seq, loff_t *pos) | |
3404 | __acquires(RCU) | |
3405 | { | |
3406 | struct net *net = seq_file_net(seq); | |
3407 | loff_t off; | |
3408 | struct net_device *dev; | |
3409 | ||
3410 | rcu_read_lock(); | |
3411 | if (!*pos) | |
3412 | return SEQ_START_TOKEN; | |
3413 | ||
3414 | off = 1; | |
3415 | for_each_netdev_rcu(net, dev) | |
3416 | if (off++ == *pos) | |
3417 | return dev; | |
3418 | ||
3419 | return NULL; | |
3420 | } | |
3421 | ||
3422 | void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
3423 | { | |
3424 | struct net_device *dev = (v == SEQ_START_TOKEN) ? | |
3425 | first_net_device(seq_file_net(seq)) : | |
3426 | next_net_device((struct net_device *)v); | |
3427 | ||
3428 | ++*pos; | |
3429 | return rcu_dereference(dev); | |
3430 | } | |
3431 | ||
3432 | void dev_seq_stop(struct seq_file *seq, void *v) | |
3433 | __releases(RCU) | |
3434 | { | |
3435 | rcu_read_unlock(); | |
3436 | } | |
3437 | ||
3438 | static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev) | |
3439 | { | |
3440 | const struct net_device_stats *stats = dev_get_stats(dev); | |
3441 | ||
3442 | seq_printf(seq, "%6s: %7lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu " | |
3443 | "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n", | |
3444 | dev->name, stats->rx_bytes, stats->rx_packets, | |
3445 | stats->rx_errors, | |
3446 | stats->rx_dropped + stats->rx_missed_errors, | |
3447 | stats->rx_fifo_errors, | |
3448 | stats->rx_length_errors + stats->rx_over_errors + | |
3449 | stats->rx_crc_errors + stats->rx_frame_errors, | |
3450 | stats->rx_compressed, stats->multicast, | |
3451 | stats->tx_bytes, stats->tx_packets, | |
3452 | stats->tx_errors, stats->tx_dropped, | |
3453 | stats->tx_fifo_errors, stats->collisions, | |
3454 | stats->tx_carrier_errors + | |
3455 | stats->tx_aborted_errors + | |
3456 | stats->tx_window_errors + | |
3457 | stats->tx_heartbeat_errors, | |
3458 | stats->tx_compressed); | |
3459 | } | |
3460 | ||
3461 | /* | |
3462 | * Called from the PROCfs module. This now uses the new arbitrary sized | |
3463 | * /proc/net interface to create /proc/net/dev | |
3464 | */ | |
3465 | static int dev_seq_show(struct seq_file *seq, void *v) | |
3466 | { | |
3467 | if (v == SEQ_START_TOKEN) | |
3468 | seq_puts(seq, "Inter-| Receive " | |
3469 | " | Transmit\n" | |
3470 | " face |bytes packets errs drop fifo frame " | |
3471 | "compressed multicast|bytes packets errs " | |
3472 | "drop fifo colls carrier compressed\n"); | |
3473 | else | |
3474 | dev_seq_printf_stats(seq, v); | |
3475 | return 0; | |
3476 | } | |
3477 | ||
3478 | static struct netif_rx_stats *softnet_get_online(loff_t *pos) | |
3479 | { | |
3480 | struct netif_rx_stats *rc = NULL; | |
3481 | ||
3482 | while (*pos < nr_cpu_ids) | |
3483 | if (cpu_online(*pos)) { | |
3484 | rc = &per_cpu(netdev_rx_stat, *pos); | |
3485 | break; | |
3486 | } else | |
3487 | ++*pos; | |
3488 | return rc; | |
3489 | } | |
3490 | ||
3491 | static void *softnet_seq_start(struct seq_file *seq, loff_t *pos) | |
3492 | { | |
3493 | return softnet_get_online(pos); | |
3494 | } | |
3495 | ||
3496 | static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
3497 | { | |
3498 | ++*pos; | |
3499 | return softnet_get_online(pos); | |
3500 | } | |
3501 | ||
3502 | static void softnet_seq_stop(struct seq_file *seq, void *v) | |
3503 | { | |
3504 | } | |
3505 | ||
3506 | static int softnet_seq_show(struct seq_file *seq, void *v) | |
3507 | { | |
3508 | struct netif_rx_stats *s = v; | |
3509 | ||
3510 | seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n", | |
3511 | s->total, s->dropped, s->time_squeeze, 0, | |
3512 | 0, 0, 0, 0, /* was fastroute */ | |
3513 | s->cpu_collision, s->received_rps); | |
3514 | return 0; | |
3515 | } | |
3516 | ||
3517 | static const struct seq_operations dev_seq_ops = { | |
3518 | .start = dev_seq_start, | |
3519 | .next = dev_seq_next, | |
3520 | .stop = dev_seq_stop, | |
3521 | .show = dev_seq_show, | |
3522 | }; | |
3523 | ||
3524 | static int dev_seq_open(struct inode *inode, struct file *file) | |
3525 | { | |
3526 | return seq_open_net(inode, file, &dev_seq_ops, | |
3527 | sizeof(struct seq_net_private)); | |
3528 | } | |
3529 | ||
3530 | static const struct file_operations dev_seq_fops = { | |
3531 | .owner = THIS_MODULE, | |
3532 | .open = dev_seq_open, | |
3533 | .read = seq_read, | |
3534 | .llseek = seq_lseek, | |
3535 | .release = seq_release_net, | |
3536 | }; | |
3537 | ||
3538 | static const struct seq_operations softnet_seq_ops = { | |
3539 | .start = softnet_seq_start, | |
3540 | .next = softnet_seq_next, | |
3541 | .stop = softnet_seq_stop, | |
3542 | .show = softnet_seq_show, | |
3543 | }; | |
3544 | ||
3545 | static int softnet_seq_open(struct inode *inode, struct file *file) | |
3546 | { | |
3547 | return seq_open(file, &softnet_seq_ops); | |
3548 | } | |
3549 | ||
3550 | static const struct file_operations softnet_seq_fops = { | |
3551 | .owner = THIS_MODULE, | |
3552 | .open = softnet_seq_open, | |
3553 | .read = seq_read, | |
3554 | .llseek = seq_lseek, | |
3555 | .release = seq_release, | |
3556 | }; | |
3557 | ||
3558 | static void *ptype_get_idx(loff_t pos) | |
3559 | { | |
3560 | struct packet_type *pt = NULL; | |
3561 | loff_t i = 0; | |
3562 | int t; | |
3563 | ||
3564 | list_for_each_entry_rcu(pt, &ptype_all, list) { | |
3565 | if (i == pos) | |
3566 | return pt; | |
3567 | ++i; | |
3568 | } | |
3569 | ||
3570 | for (t = 0; t < PTYPE_HASH_SIZE; t++) { | |
3571 | list_for_each_entry_rcu(pt, &ptype_base[t], list) { | |
3572 | if (i == pos) | |
3573 | return pt; | |
3574 | ++i; | |
3575 | } | |
3576 | } | |
3577 | return NULL; | |
3578 | } | |
3579 | ||
3580 | static void *ptype_seq_start(struct seq_file *seq, loff_t *pos) | |
3581 | __acquires(RCU) | |
3582 | { | |
3583 | rcu_read_lock(); | |
3584 | return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN; | |
3585 | } | |
3586 | ||
3587 | static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
3588 | { | |
3589 | struct packet_type *pt; | |
3590 | struct list_head *nxt; | |
3591 | int hash; | |
3592 | ||
3593 | ++*pos; | |
3594 | if (v == SEQ_START_TOKEN) | |
3595 | return ptype_get_idx(0); | |
3596 | ||
3597 | pt = v; | |
3598 | nxt = pt->list.next; | |
3599 | if (pt->type == htons(ETH_P_ALL)) { | |
3600 | if (nxt != &ptype_all) | |
3601 | goto found; | |
3602 | hash = 0; | |
3603 | nxt = ptype_base[0].next; | |
3604 | } else | |
3605 | hash = ntohs(pt->type) & PTYPE_HASH_MASK; | |
3606 | ||
3607 | while (nxt == &ptype_base[hash]) { | |
3608 | if (++hash >= PTYPE_HASH_SIZE) | |
3609 | return NULL; | |
3610 | nxt = ptype_base[hash].next; | |
3611 | } | |
3612 | found: | |
3613 | return list_entry(nxt, struct packet_type, list); | |
3614 | } | |
3615 | ||
3616 | static void ptype_seq_stop(struct seq_file *seq, void *v) | |
3617 | __releases(RCU) | |
3618 | { | |
3619 | rcu_read_unlock(); | |
3620 | } | |
3621 | ||
3622 | static int ptype_seq_show(struct seq_file *seq, void *v) | |
3623 | { | |
3624 | struct packet_type *pt = v; | |
3625 | ||
3626 | if (v == SEQ_START_TOKEN) | |
3627 | seq_puts(seq, "Type Device Function\n"); | |
3628 | else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) { | |
3629 | if (pt->type == htons(ETH_P_ALL)) | |
3630 | seq_puts(seq, "ALL "); | |
3631 | else | |
3632 | seq_printf(seq, "%04x", ntohs(pt->type)); | |
3633 | ||
3634 | seq_printf(seq, " %-8s %pF\n", | |
3635 | pt->dev ? pt->dev->name : "", pt->func); | |
3636 | } | |
3637 | ||
3638 | return 0; | |
3639 | } | |
3640 | ||
3641 | static const struct seq_operations ptype_seq_ops = { | |
3642 | .start = ptype_seq_start, | |
3643 | .next = ptype_seq_next, | |
3644 | .stop = ptype_seq_stop, | |
3645 | .show = ptype_seq_show, | |
3646 | }; | |
3647 | ||
3648 | static int ptype_seq_open(struct inode *inode, struct file *file) | |
3649 | { | |
3650 | return seq_open_net(inode, file, &ptype_seq_ops, | |
3651 | sizeof(struct seq_net_private)); | |
3652 | } | |
3653 | ||
3654 | static const struct file_operations ptype_seq_fops = { | |
3655 | .owner = THIS_MODULE, | |
3656 | .open = ptype_seq_open, | |
3657 | .read = seq_read, | |
3658 | .llseek = seq_lseek, | |
3659 | .release = seq_release_net, | |
3660 | }; | |
3661 | ||
3662 | ||
3663 | static int __net_init dev_proc_net_init(struct net *net) | |
3664 | { | |
3665 | int rc = -ENOMEM; | |
3666 | ||
3667 | if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops)) | |
3668 | goto out; | |
3669 | if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops)) | |
3670 | goto out_dev; | |
3671 | if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops)) | |
3672 | goto out_softnet; | |
3673 | ||
3674 | if (wext_proc_init(net)) | |
3675 | goto out_ptype; | |
3676 | rc = 0; | |
3677 | out: | |
3678 | return rc; | |
3679 | out_ptype: | |
3680 | proc_net_remove(net, "ptype"); | |
3681 | out_softnet: | |
3682 | proc_net_remove(net, "softnet_stat"); | |
3683 | out_dev: | |
3684 | proc_net_remove(net, "dev"); | |
3685 | goto out; | |
3686 | } | |
3687 | ||
3688 | static void __net_exit dev_proc_net_exit(struct net *net) | |
3689 | { | |
3690 | wext_proc_exit(net); | |
3691 | ||
3692 | proc_net_remove(net, "ptype"); | |
3693 | proc_net_remove(net, "softnet_stat"); | |
3694 | proc_net_remove(net, "dev"); | |
3695 | } | |
3696 | ||
3697 | static struct pernet_operations __net_initdata dev_proc_ops = { | |
3698 | .init = dev_proc_net_init, | |
3699 | .exit = dev_proc_net_exit, | |
3700 | }; | |
3701 | ||
3702 | static int __init dev_proc_init(void) | |
3703 | { | |
3704 | return register_pernet_subsys(&dev_proc_ops); | |
3705 | } | |
3706 | #else | |
3707 | #define dev_proc_init() 0 | |
3708 | #endif /* CONFIG_PROC_FS */ | |
3709 | ||
3710 | ||
3711 | /** | |
3712 | * netdev_set_master - set up master/slave pair | |
3713 | * @slave: slave device | |
3714 | * @master: new master device | |
3715 | * | |
3716 | * Changes the master device of the slave. Pass %NULL to break the | |
3717 | * bonding. The caller must hold the RTNL semaphore. On a failure | |
3718 | * a negative errno code is returned. On success the reference counts | |
3719 | * are adjusted, %RTM_NEWLINK is sent to the routing socket and the | |
3720 | * function returns zero. | |
3721 | */ | |
3722 | int netdev_set_master(struct net_device *slave, struct net_device *master) | |
3723 | { | |
3724 | struct net_device *old = slave->master; | |
3725 | ||
3726 | ASSERT_RTNL(); | |
3727 | ||
3728 | if (master) { | |
3729 | if (old) | |
3730 | return -EBUSY; | |
3731 | dev_hold(master); | |
3732 | } | |
3733 | ||
3734 | slave->master = master; | |
3735 | ||
3736 | synchronize_net(); | |
3737 | ||
3738 | if (old) | |
3739 | dev_put(old); | |
3740 | ||
3741 | if (master) | |
3742 | slave->flags |= IFF_SLAVE; | |
3743 | else | |
3744 | slave->flags &= ~IFF_SLAVE; | |
3745 | ||
3746 | rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE); | |
3747 | return 0; | |
3748 | } | |
3749 | EXPORT_SYMBOL(netdev_set_master); | |
3750 | ||
3751 | static void dev_change_rx_flags(struct net_device *dev, int flags) | |
3752 | { | |
3753 | const struct net_device_ops *ops = dev->netdev_ops; | |
3754 | ||
3755 | if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags) | |
3756 | ops->ndo_change_rx_flags(dev, flags); | |
3757 | } | |
3758 | ||
3759 | static int __dev_set_promiscuity(struct net_device *dev, int inc) | |
3760 | { | |
3761 | unsigned short old_flags = dev->flags; | |
3762 | uid_t uid; | |
3763 | gid_t gid; | |
3764 | ||
3765 | ASSERT_RTNL(); | |
3766 | ||
3767 | dev->flags |= IFF_PROMISC; | |
3768 | dev->promiscuity += inc; | |
3769 | if (dev->promiscuity == 0) { | |
3770 | /* | |
3771 | * Avoid overflow. | |
3772 | * If inc causes overflow, untouch promisc and return error. | |
3773 | */ | |
3774 | if (inc < 0) | |
3775 | dev->flags &= ~IFF_PROMISC; | |
3776 | else { | |
3777 | dev->promiscuity -= inc; | |
3778 | printk(KERN_WARNING "%s: promiscuity touches roof, " | |
3779 | "set promiscuity failed, promiscuity feature " | |
3780 | "of device might be broken.\n", dev->name); | |
3781 | return -EOVERFLOW; | |
3782 | } | |
3783 | } | |
3784 | if (dev->flags != old_flags) { | |
3785 | printk(KERN_INFO "device %s %s promiscuous mode\n", | |
3786 | dev->name, (dev->flags & IFF_PROMISC) ? "entered" : | |
3787 | "left"); | |
3788 | if (audit_enabled) { | |
3789 | current_uid_gid(&uid, &gid); | |
3790 | audit_log(current->audit_context, GFP_ATOMIC, | |
3791 | AUDIT_ANOM_PROMISCUOUS, | |
3792 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
3793 | dev->name, (dev->flags & IFF_PROMISC), | |
3794 | (old_flags & IFF_PROMISC), | |
3795 | audit_get_loginuid(current), | |
3796 | uid, gid, | |
3797 | audit_get_sessionid(current)); | |
3798 | } | |
3799 | ||
3800 | dev_change_rx_flags(dev, IFF_PROMISC); | |
3801 | } | |
3802 | return 0; | |
3803 | } | |
3804 | ||
3805 | /** | |
3806 | * dev_set_promiscuity - update promiscuity count on a device | |
3807 | * @dev: device | |
3808 | * @inc: modifier | |
3809 | * | |
3810 | * Add or remove promiscuity from a device. While the count in the device | |
3811 | * remains above zero the interface remains promiscuous. Once it hits zero | |
3812 | * the device reverts back to normal filtering operation. A negative inc | |
3813 | * value is used to drop promiscuity on the device. | |
3814 | * Return 0 if successful or a negative errno code on error. | |
3815 | */ | |
3816 | int dev_set_promiscuity(struct net_device *dev, int inc) | |
3817 | { | |
3818 | unsigned short old_flags = dev->flags; | |
3819 | int err; | |
3820 | ||
3821 | err = __dev_set_promiscuity(dev, inc); | |
3822 | if (err < 0) | |
3823 | return err; | |
3824 | if (dev->flags != old_flags) | |
3825 | dev_set_rx_mode(dev); | |
3826 | return err; | |
3827 | } | |
3828 | EXPORT_SYMBOL(dev_set_promiscuity); | |
3829 | ||
3830 | /** | |
3831 | * dev_set_allmulti - update allmulti count on a device | |
3832 | * @dev: device | |
3833 | * @inc: modifier | |
3834 | * | |
3835 | * Add or remove reception of all multicast frames to a device. While the | |
3836 | * count in the device remains above zero the interface remains listening | |
3837 | * to all interfaces. Once it hits zero the device reverts back to normal | |
3838 | * filtering operation. A negative @inc value is used to drop the counter | |
3839 | * when releasing a resource needing all multicasts. | |
3840 | * Return 0 if successful or a negative errno code on error. | |
3841 | */ | |
3842 | ||
3843 | int dev_set_allmulti(struct net_device *dev, int inc) | |
3844 | { | |
3845 | unsigned short old_flags = dev->flags; | |
3846 | ||
3847 | ASSERT_RTNL(); | |
3848 | ||
3849 | dev->flags |= IFF_ALLMULTI; | |
3850 | dev->allmulti += inc; | |
3851 | if (dev->allmulti == 0) { | |
3852 | /* | |
3853 | * Avoid overflow. | |
3854 | * If inc causes overflow, untouch allmulti and return error. | |
3855 | */ | |
3856 | if (inc < 0) | |
3857 | dev->flags &= ~IFF_ALLMULTI; | |
3858 | else { | |
3859 | dev->allmulti -= inc; | |
3860 | printk(KERN_WARNING "%s: allmulti touches roof, " | |
3861 | "set allmulti failed, allmulti feature of " | |
3862 | "device might be broken.\n", dev->name); | |
3863 | return -EOVERFLOW; | |
3864 | } | |
3865 | } | |
3866 | if (dev->flags ^ old_flags) { | |
3867 | dev_change_rx_flags(dev, IFF_ALLMULTI); | |
3868 | dev_set_rx_mode(dev); | |
3869 | } | |
3870 | return 0; | |
3871 | } | |
3872 | EXPORT_SYMBOL(dev_set_allmulti); | |
3873 | ||
3874 | /* | |
3875 | * Upload unicast and multicast address lists to device and | |
3876 | * configure RX filtering. When the device doesn't support unicast | |
3877 | * filtering it is put in promiscuous mode while unicast addresses | |
3878 | * are present. | |
3879 | */ | |
3880 | void __dev_set_rx_mode(struct net_device *dev) | |
3881 | { | |
3882 | const struct net_device_ops *ops = dev->netdev_ops; | |
3883 | ||
3884 | /* dev_open will call this function so the list will stay sane. */ | |
3885 | if (!(dev->flags&IFF_UP)) | |
3886 | return; | |
3887 | ||
3888 | if (!netif_device_present(dev)) | |
3889 | return; | |
3890 | ||
3891 | if (ops->ndo_set_rx_mode) | |
3892 | ops->ndo_set_rx_mode(dev); | |
3893 | else { | |
3894 | /* Unicast addresses changes may only happen under the rtnl, | |
3895 | * therefore calling __dev_set_promiscuity here is safe. | |
3896 | */ | |
3897 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { | |
3898 | __dev_set_promiscuity(dev, 1); | |
3899 | dev->uc_promisc = 1; | |
3900 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { | |
3901 | __dev_set_promiscuity(dev, -1); | |
3902 | dev->uc_promisc = 0; | |
3903 | } | |
3904 | ||
3905 | if (ops->ndo_set_multicast_list) | |
3906 | ops->ndo_set_multicast_list(dev); | |
3907 | } | |
3908 | } | |
3909 | ||
3910 | void dev_set_rx_mode(struct net_device *dev) | |
3911 | { | |
3912 | netif_addr_lock_bh(dev); | |
3913 | __dev_set_rx_mode(dev); | |
3914 | netif_addr_unlock_bh(dev); | |
3915 | } | |
3916 | ||
3917 | /* hw addresses list handling functions */ | |
3918 | ||
3919 | static int __hw_addr_add(struct netdev_hw_addr_list *list, unsigned char *addr, | |
3920 | int addr_len, unsigned char addr_type) | |
3921 | { | |
3922 | struct netdev_hw_addr *ha; | |
3923 | int alloc_size; | |
3924 | ||
3925 | if (addr_len > MAX_ADDR_LEN) | |
3926 | return -EINVAL; | |
3927 | ||
3928 | list_for_each_entry(ha, &list->list, list) { | |
3929 | if (!memcmp(ha->addr, addr, addr_len) && | |
3930 | ha->type == addr_type) { | |
3931 | ha->refcount++; | |
3932 | return 0; | |
3933 | } | |
3934 | } | |
3935 | ||
3936 | ||
3937 | alloc_size = sizeof(*ha); | |
3938 | if (alloc_size < L1_CACHE_BYTES) | |
3939 | alloc_size = L1_CACHE_BYTES; | |
3940 | ha = kmalloc(alloc_size, GFP_ATOMIC); | |
3941 | if (!ha) | |
3942 | return -ENOMEM; | |
3943 | memcpy(ha->addr, addr, addr_len); | |
3944 | ha->type = addr_type; | |
3945 | ha->refcount = 1; | |
3946 | ha->synced = false; | |
3947 | list_add_tail_rcu(&ha->list, &list->list); | |
3948 | list->count++; | |
3949 | return 0; | |
3950 | } | |
3951 | ||
3952 | static void ha_rcu_free(struct rcu_head *head) | |
3953 | { | |
3954 | struct netdev_hw_addr *ha; | |
3955 | ||
3956 | ha = container_of(head, struct netdev_hw_addr, rcu_head); | |
3957 | kfree(ha); | |
3958 | } | |
3959 | ||
3960 | static int __hw_addr_del(struct netdev_hw_addr_list *list, unsigned char *addr, | |
3961 | int addr_len, unsigned char addr_type) | |
3962 | { | |
3963 | struct netdev_hw_addr *ha; | |
3964 | ||
3965 | list_for_each_entry(ha, &list->list, list) { | |
3966 | if (!memcmp(ha->addr, addr, addr_len) && | |
3967 | (ha->type == addr_type || !addr_type)) { | |
3968 | if (--ha->refcount) | |
3969 | return 0; | |
3970 | list_del_rcu(&ha->list); | |
3971 | call_rcu(&ha->rcu_head, ha_rcu_free); | |
3972 | list->count--; | |
3973 | return 0; | |
3974 | } | |
3975 | } | |
3976 | return -ENOENT; | |
3977 | } | |
3978 | ||
3979 | static int __hw_addr_add_multiple(struct netdev_hw_addr_list *to_list, | |
3980 | struct netdev_hw_addr_list *from_list, | |
3981 | int addr_len, | |
3982 | unsigned char addr_type) | |
3983 | { | |
3984 | int err; | |
3985 | struct netdev_hw_addr *ha, *ha2; | |
3986 | unsigned char type; | |
3987 | ||
3988 | list_for_each_entry(ha, &from_list->list, list) { | |
3989 | type = addr_type ? addr_type : ha->type; | |
3990 | err = __hw_addr_add(to_list, ha->addr, addr_len, type); | |
3991 | if (err) | |
3992 | goto unroll; | |
3993 | } | |
3994 | return 0; | |
3995 | ||
3996 | unroll: | |
3997 | list_for_each_entry(ha2, &from_list->list, list) { | |
3998 | if (ha2 == ha) | |
3999 | break; | |
4000 | type = addr_type ? addr_type : ha2->type; | |
4001 | __hw_addr_del(to_list, ha2->addr, addr_len, type); | |
4002 | } | |
4003 | return err; | |
4004 | } | |
4005 | ||
4006 | static void __hw_addr_del_multiple(struct netdev_hw_addr_list *to_list, | |
4007 | struct netdev_hw_addr_list *from_list, | |
4008 | int addr_len, | |
4009 | unsigned char addr_type) | |
4010 | { | |
4011 | struct netdev_hw_addr *ha; | |
4012 | unsigned char type; | |
4013 | ||
4014 | list_for_each_entry(ha, &from_list->list, list) { | |
4015 | type = addr_type ? addr_type : ha->type; | |
4016 | __hw_addr_del(to_list, ha->addr, addr_len, addr_type); | |
4017 | } | |
4018 | } | |
4019 | ||
4020 | static int __hw_addr_sync(struct netdev_hw_addr_list *to_list, | |
4021 | struct netdev_hw_addr_list *from_list, | |
4022 | int addr_len) | |
4023 | { | |
4024 | int err = 0; | |
4025 | struct netdev_hw_addr *ha, *tmp; | |
4026 | ||
4027 | list_for_each_entry_safe(ha, tmp, &from_list->list, list) { | |
4028 | if (!ha->synced) { | |
4029 | err = __hw_addr_add(to_list, ha->addr, | |
4030 | addr_len, ha->type); | |
4031 | if (err) | |
4032 | break; | |
4033 | ha->synced = true; | |
4034 | ha->refcount++; | |
4035 | } else if (ha->refcount == 1) { | |
4036 | __hw_addr_del(to_list, ha->addr, addr_len, ha->type); | |
4037 | __hw_addr_del(from_list, ha->addr, addr_len, ha->type); | |
4038 | } | |
4039 | } | |
4040 | return err; | |
4041 | } | |
4042 | ||
4043 | static void __hw_addr_unsync(struct netdev_hw_addr_list *to_list, | |
4044 | struct netdev_hw_addr_list *from_list, | |
4045 | int addr_len) | |
4046 | { | |
4047 | struct netdev_hw_addr *ha, *tmp; | |
4048 | ||
4049 | list_for_each_entry_safe(ha, tmp, &from_list->list, list) { | |
4050 | if (ha->synced) { | |
4051 | __hw_addr_del(to_list, ha->addr, | |
4052 | addr_len, ha->type); | |
4053 | ha->synced = false; | |
4054 | __hw_addr_del(from_list, ha->addr, | |
4055 | addr_len, ha->type); | |
4056 | } | |
4057 | } | |
4058 | } | |
4059 | ||
4060 | static void __hw_addr_flush(struct netdev_hw_addr_list *list) | |
4061 | { | |
4062 | struct netdev_hw_addr *ha, *tmp; | |
4063 | ||
4064 | list_for_each_entry_safe(ha, tmp, &list->list, list) { | |
4065 | list_del_rcu(&ha->list); | |
4066 | call_rcu(&ha->rcu_head, ha_rcu_free); | |
4067 | } | |
4068 | list->count = 0; | |
4069 | } | |
4070 | ||
4071 | static void __hw_addr_init(struct netdev_hw_addr_list *list) | |
4072 | { | |
4073 | INIT_LIST_HEAD(&list->list); | |
4074 | list->count = 0; | |
4075 | } | |
4076 | ||
4077 | /* Device addresses handling functions */ | |
4078 | ||
4079 | static void dev_addr_flush(struct net_device *dev) | |
4080 | { | |
4081 | /* rtnl_mutex must be held here */ | |
4082 | ||
4083 | __hw_addr_flush(&dev->dev_addrs); | |
4084 | dev->dev_addr = NULL; | |
4085 | } | |
4086 | ||
4087 | static int dev_addr_init(struct net_device *dev) | |
4088 | { | |
4089 | unsigned char addr[MAX_ADDR_LEN]; | |
4090 | struct netdev_hw_addr *ha; | |
4091 | int err; | |
4092 | ||
4093 | /* rtnl_mutex must be held here */ | |
4094 | ||
4095 | __hw_addr_init(&dev->dev_addrs); | |
4096 | memset(addr, 0, sizeof(addr)); | |
4097 | err = __hw_addr_add(&dev->dev_addrs, addr, sizeof(addr), | |
4098 | NETDEV_HW_ADDR_T_LAN); | |
4099 | if (!err) { | |
4100 | /* | |
4101 | * Get the first (previously created) address from the list | |
4102 | * and set dev_addr pointer to this location. | |
4103 | */ | |
4104 | ha = list_first_entry(&dev->dev_addrs.list, | |
4105 | struct netdev_hw_addr, list); | |
4106 | dev->dev_addr = ha->addr; | |
4107 | } | |
4108 | return err; | |
4109 | } | |
4110 | ||
4111 | /** | |
4112 | * dev_addr_add - Add a device address | |
4113 | * @dev: device | |
4114 | * @addr: address to add | |
4115 | * @addr_type: address type | |
4116 | * | |
4117 | * Add a device address to the device or increase the reference count if | |
4118 | * it already exists. | |
4119 | * | |
4120 | * The caller must hold the rtnl_mutex. | |
4121 | */ | |
4122 | int dev_addr_add(struct net_device *dev, unsigned char *addr, | |
4123 | unsigned char addr_type) | |
4124 | { | |
4125 | int err; | |
4126 | ||
4127 | ASSERT_RTNL(); | |
4128 | ||
4129 | err = __hw_addr_add(&dev->dev_addrs, addr, dev->addr_len, addr_type); | |
4130 | if (!err) | |
4131 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); | |
4132 | return err; | |
4133 | } | |
4134 | EXPORT_SYMBOL(dev_addr_add); | |
4135 | ||
4136 | /** | |
4137 | * dev_addr_del - Release a device address. | |
4138 | * @dev: device | |
4139 | * @addr: address to delete | |
4140 | * @addr_type: address type | |
4141 | * | |
4142 | * Release reference to a device address and remove it from the device | |
4143 | * if the reference count drops to zero. | |
4144 | * | |
4145 | * The caller must hold the rtnl_mutex. | |
4146 | */ | |
4147 | int dev_addr_del(struct net_device *dev, unsigned char *addr, | |
4148 | unsigned char addr_type) | |
4149 | { | |
4150 | int err; | |
4151 | struct netdev_hw_addr *ha; | |
4152 | ||
4153 | ASSERT_RTNL(); | |
4154 | ||
4155 | /* | |
4156 | * We can not remove the first address from the list because | |
4157 | * dev->dev_addr points to that. | |
4158 | */ | |
4159 | ha = list_first_entry(&dev->dev_addrs.list, | |
4160 | struct netdev_hw_addr, list); | |
4161 | if (ha->addr == dev->dev_addr && ha->refcount == 1) | |
4162 | return -ENOENT; | |
4163 | ||
4164 | err = __hw_addr_del(&dev->dev_addrs, addr, dev->addr_len, | |
4165 | addr_type); | |
4166 | if (!err) | |
4167 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); | |
4168 | return err; | |
4169 | } | |
4170 | EXPORT_SYMBOL(dev_addr_del); | |
4171 | ||
4172 | /** | |
4173 | * dev_addr_add_multiple - Add device addresses from another device | |
4174 | * @to_dev: device to which addresses will be added | |
4175 | * @from_dev: device from which addresses will be added | |
4176 | * @addr_type: address type - 0 means type will be used from from_dev | |
4177 | * | |
4178 | * Add device addresses of the one device to another. | |
4179 | ** | |
4180 | * The caller must hold the rtnl_mutex. | |
4181 | */ | |
4182 | int dev_addr_add_multiple(struct net_device *to_dev, | |
4183 | struct net_device *from_dev, | |
4184 | unsigned char addr_type) | |
4185 | { | |
4186 | int err; | |
4187 | ||
4188 | ASSERT_RTNL(); | |
4189 | ||
4190 | if (from_dev->addr_len != to_dev->addr_len) | |
4191 | return -EINVAL; | |
4192 | err = __hw_addr_add_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs, | |
4193 | to_dev->addr_len, addr_type); | |
4194 | if (!err) | |
4195 | call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev); | |
4196 | return err; | |
4197 | } | |
4198 | EXPORT_SYMBOL(dev_addr_add_multiple); | |
4199 | ||
4200 | /** | |
4201 | * dev_addr_del_multiple - Delete device addresses by another device | |
4202 | * @to_dev: device where the addresses will be deleted | |
4203 | * @from_dev: device by which addresses the addresses will be deleted | |
4204 | * @addr_type: address type - 0 means type will used from from_dev | |
4205 | * | |
4206 | * Deletes addresses in to device by the list of addresses in from device. | |
4207 | * | |
4208 | * The caller must hold the rtnl_mutex. | |
4209 | */ | |
4210 | int dev_addr_del_multiple(struct net_device *to_dev, | |
4211 | struct net_device *from_dev, | |
4212 | unsigned char addr_type) | |
4213 | { | |
4214 | ASSERT_RTNL(); | |
4215 | ||
4216 | if (from_dev->addr_len != to_dev->addr_len) | |
4217 | return -EINVAL; | |
4218 | __hw_addr_del_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs, | |
4219 | to_dev->addr_len, addr_type); | |
4220 | call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev); | |
4221 | return 0; | |
4222 | } | |
4223 | EXPORT_SYMBOL(dev_addr_del_multiple); | |
4224 | ||
4225 | /* multicast addresses handling functions */ | |
4226 | ||
4227 | int __dev_addr_delete(struct dev_addr_list **list, int *count, | |
4228 | void *addr, int alen, int glbl) | |
4229 | { | |
4230 | struct dev_addr_list *da; | |
4231 | ||
4232 | for (; (da = *list) != NULL; list = &da->next) { | |
4233 | if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 && | |
4234 | alen == da->da_addrlen) { | |
4235 | if (glbl) { | |
4236 | int old_glbl = da->da_gusers; | |
4237 | da->da_gusers = 0; | |
4238 | if (old_glbl == 0) | |
4239 | break; | |
4240 | } | |
4241 | if (--da->da_users) | |
4242 | return 0; | |
4243 | ||
4244 | *list = da->next; | |
4245 | kfree(da); | |
4246 | (*count)--; | |
4247 | return 0; | |
4248 | } | |
4249 | } | |
4250 | return -ENOENT; | |
4251 | } | |
4252 | ||
4253 | int __dev_addr_add(struct dev_addr_list **list, int *count, | |
4254 | void *addr, int alen, int glbl) | |
4255 | { | |
4256 | struct dev_addr_list *da; | |
4257 | ||
4258 | for (da = *list; da != NULL; da = da->next) { | |
4259 | if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 && | |
4260 | da->da_addrlen == alen) { | |
4261 | if (glbl) { | |
4262 | int old_glbl = da->da_gusers; | |
4263 | da->da_gusers = 1; | |
4264 | if (old_glbl) | |
4265 | return 0; | |
4266 | } | |
4267 | da->da_users++; | |
4268 | return 0; | |
4269 | } | |
4270 | } | |
4271 | ||
4272 | da = kzalloc(sizeof(*da), GFP_ATOMIC); | |
4273 | if (da == NULL) | |
4274 | return -ENOMEM; | |
4275 | memcpy(da->da_addr, addr, alen); | |
4276 | da->da_addrlen = alen; | |
4277 | da->da_users = 1; | |
4278 | da->da_gusers = glbl ? 1 : 0; | |
4279 | da->next = *list; | |
4280 | *list = da; | |
4281 | (*count)++; | |
4282 | return 0; | |
4283 | } | |
4284 | ||
4285 | /** | |
4286 | * dev_unicast_delete - Release secondary unicast address. | |
4287 | * @dev: device | |
4288 | * @addr: address to delete | |
4289 | * | |
4290 | * Release reference to a secondary unicast address and remove it | |
4291 | * from the device if the reference count drops to zero. | |
4292 | * | |
4293 | * The caller must hold the rtnl_mutex. | |
4294 | */ | |
4295 | int dev_unicast_delete(struct net_device *dev, void *addr) | |
4296 | { | |
4297 | int err; | |
4298 | ||
4299 | ASSERT_RTNL(); | |
4300 | ||
4301 | netif_addr_lock_bh(dev); | |
4302 | err = __hw_addr_del(&dev->uc, addr, dev->addr_len, | |
4303 | NETDEV_HW_ADDR_T_UNICAST); | |
4304 | if (!err) | |
4305 | __dev_set_rx_mode(dev); | |
4306 | netif_addr_unlock_bh(dev); | |
4307 | return err; | |
4308 | } | |
4309 | EXPORT_SYMBOL(dev_unicast_delete); | |
4310 | ||
4311 | /** | |
4312 | * dev_unicast_add - add a secondary unicast address | |
4313 | * @dev: device | |
4314 | * @addr: address to add | |
4315 | * | |
4316 | * Add a secondary unicast address to the device or increase | |
4317 | * the reference count if it already exists. | |
4318 | * | |
4319 | * The caller must hold the rtnl_mutex. | |
4320 | */ | |
4321 | int dev_unicast_add(struct net_device *dev, void *addr) | |
4322 | { | |
4323 | int err; | |
4324 | ||
4325 | ASSERT_RTNL(); | |
4326 | ||
4327 | netif_addr_lock_bh(dev); | |
4328 | err = __hw_addr_add(&dev->uc, addr, dev->addr_len, | |
4329 | NETDEV_HW_ADDR_T_UNICAST); | |
4330 | if (!err) | |
4331 | __dev_set_rx_mode(dev); | |
4332 | netif_addr_unlock_bh(dev); | |
4333 | return err; | |
4334 | } | |
4335 | EXPORT_SYMBOL(dev_unicast_add); | |
4336 | ||
4337 | int __dev_addr_sync(struct dev_addr_list **to, int *to_count, | |
4338 | struct dev_addr_list **from, int *from_count) | |
4339 | { | |
4340 | struct dev_addr_list *da, *next; | |
4341 | int err = 0; | |
4342 | ||
4343 | da = *from; | |
4344 | while (da != NULL) { | |
4345 | next = da->next; | |
4346 | if (!da->da_synced) { | |
4347 | err = __dev_addr_add(to, to_count, | |
4348 | da->da_addr, da->da_addrlen, 0); | |
4349 | if (err < 0) | |
4350 | break; | |
4351 | da->da_synced = 1; | |
4352 | da->da_users++; | |
4353 | } else if (da->da_users == 1) { | |
4354 | __dev_addr_delete(to, to_count, | |
4355 | da->da_addr, da->da_addrlen, 0); | |
4356 | __dev_addr_delete(from, from_count, | |
4357 | da->da_addr, da->da_addrlen, 0); | |
4358 | } | |
4359 | da = next; | |
4360 | } | |
4361 | return err; | |
4362 | } | |
4363 | EXPORT_SYMBOL_GPL(__dev_addr_sync); | |
4364 | ||
4365 | void __dev_addr_unsync(struct dev_addr_list **to, int *to_count, | |
4366 | struct dev_addr_list **from, int *from_count) | |
4367 | { | |
4368 | struct dev_addr_list *da, *next; | |
4369 | ||
4370 | da = *from; | |
4371 | while (da != NULL) { | |
4372 | next = da->next; | |
4373 | if (da->da_synced) { | |
4374 | __dev_addr_delete(to, to_count, | |
4375 | da->da_addr, da->da_addrlen, 0); | |
4376 | da->da_synced = 0; | |
4377 | __dev_addr_delete(from, from_count, | |
4378 | da->da_addr, da->da_addrlen, 0); | |
4379 | } | |
4380 | da = next; | |
4381 | } | |
4382 | } | |
4383 | EXPORT_SYMBOL_GPL(__dev_addr_unsync); | |
4384 | ||
4385 | /** | |
4386 | * dev_unicast_sync - Synchronize device's unicast list to another device | |
4387 | * @to: destination device | |
4388 | * @from: source device | |
4389 | * | |
4390 | * Add newly added addresses to the destination device and release | |
4391 | * addresses that have no users left. The source device must be | |
4392 | * locked by netif_tx_lock_bh. | |
4393 | * | |
4394 | * This function is intended to be called from the dev->set_rx_mode | |
4395 | * function of layered software devices. | |
4396 | */ | |
4397 | int dev_unicast_sync(struct net_device *to, struct net_device *from) | |
4398 | { | |
4399 | int err = 0; | |
4400 | ||
4401 | if (to->addr_len != from->addr_len) | |
4402 | return -EINVAL; | |
4403 | ||
4404 | netif_addr_lock_bh(to); | |
4405 | err = __hw_addr_sync(&to->uc, &from->uc, to->addr_len); | |
4406 | if (!err) | |
4407 | __dev_set_rx_mode(to); | |
4408 | netif_addr_unlock_bh(to); | |
4409 | return err; | |
4410 | } | |
4411 | EXPORT_SYMBOL(dev_unicast_sync); | |
4412 | ||
4413 | /** | |
4414 | * dev_unicast_unsync - Remove synchronized addresses from the destination device | |
4415 | * @to: destination device | |
4416 | * @from: source device | |
4417 | * | |
4418 | * Remove all addresses that were added to the destination device by | |
4419 | * dev_unicast_sync(). This function is intended to be called from the | |
4420 | * dev->stop function of layered software devices. | |
4421 | */ | |
4422 | void dev_unicast_unsync(struct net_device *to, struct net_device *from) | |
4423 | { | |
4424 | if (to->addr_len != from->addr_len) | |
4425 | return; | |
4426 | ||
4427 | netif_addr_lock_bh(from); | |
4428 | netif_addr_lock(to); | |
4429 | __hw_addr_unsync(&to->uc, &from->uc, to->addr_len); | |
4430 | __dev_set_rx_mode(to); | |
4431 | netif_addr_unlock(to); | |
4432 | netif_addr_unlock_bh(from); | |
4433 | } | |
4434 | EXPORT_SYMBOL(dev_unicast_unsync); | |
4435 | ||
4436 | static void dev_unicast_flush(struct net_device *dev) | |
4437 | { | |
4438 | netif_addr_lock_bh(dev); | |
4439 | __hw_addr_flush(&dev->uc); | |
4440 | netif_addr_unlock_bh(dev); | |
4441 | } | |
4442 | ||
4443 | static void dev_unicast_init(struct net_device *dev) | |
4444 | { | |
4445 | __hw_addr_init(&dev->uc); | |
4446 | } | |
4447 | ||
4448 | ||
4449 | static void __dev_addr_discard(struct dev_addr_list **list) | |
4450 | { | |
4451 | struct dev_addr_list *tmp; | |
4452 | ||
4453 | while (*list != NULL) { | |
4454 | tmp = *list; | |
4455 | *list = tmp->next; | |
4456 | if (tmp->da_users > tmp->da_gusers) | |
4457 | printk("__dev_addr_discard: address leakage! " | |
4458 | "da_users=%d\n", tmp->da_users); | |
4459 | kfree(tmp); | |
4460 | } | |
4461 | } | |
4462 | ||
4463 | static void dev_addr_discard(struct net_device *dev) | |
4464 | { | |
4465 | netif_addr_lock_bh(dev); | |
4466 | ||
4467 | __dev_addr_discard(&dev->mc_list); | |
4468 | netdev_mc_count(dev) = 0; | |
4469 | ||
4470 | netif_addr_unlock_bh(dev); | |
4471 | } | |
4472 | ||
4473 | /** | |
4474 | * dev_get_flags - get flags reported to userspace | |
4475 | * @dev: device | |
4476 | * | |
4477 | * Get the combination of flag bits exported through APIs to userspace. | |
4478 | */ | |
4479 | unsigned dev_get_flags(const struct net_device *dev) | |
4480 | { | |
4481 | unsigned flags; | |
4482 | ||
4483 | flags = (dev->flags & ~(IFF_PROMISC | | |
4484 | IFF_ALLMULTI | | |
4485 | IFF_RUNNING | | |
4486 | IFF_LOWER_UP | | |
4487 | IFF_DORMANT)) | | |
4488 | (dev->gflags & (IFF_PROMISC | | |
4489 | IFF_ALLMULTI)); | |
4490 | ||
4491 | if (netif_running(dev)) { | |
4492 | if (netif_oper_up(dev)) | |
4493 | flags |= IFF_RUNNING; | |
4494 | if (netif_carrier_ok(dev)) | |
4495 | flags |= IFF_LOWER_UP; | |
4496 | if (netif_dormant(dev)) | |
4497 | flags |= IFF_DORMANT; | |
4498 | } | |
4499 | ||
4500 | return flags; | |
4501 | } | |
4502 | EXPORT_SYMBOL(dev_get_flags); | |
4503 | ||
4504 | int __dev_change_flags(struct net_device *dev, unsigned int flags) | |
4505 | { | |
4506 | int old_flags = dev->flags; | |
4507 | int ret; | |
4508 | ||
4509 | ASSERT_RTNL(); | |
4510 | ||
4511 | /* | |
4512 | * Set the flags on our device. | |
4513 | */ | |
4514 | ||
4515 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
4516 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
4517 | IFF_AUTOMEDIA)) | | |
4518 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
4519 | IFF_ALLMULTI)); | |
4520 | ||
4521 | /* | |
4522 | * Load in the correct multicast list now the flags have changed. | |
4523 | */ | |
4524 | ||
4525 | if ((old_flags ^ flags) & IFF_MULTICAST) | |
4526 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
4527 | ||
4528 | dev_set_rx_mode(dev); | |
4529 | ||
4530 | /* | |
4531 | * Have we downed the interface. We handle IFF_UP ourselves | |
4532 | * according to user attempts to set it, rather than blindly | |
4533 | * setting it. | |
4534 | */ | |
4535 | ||
4536 | ret = 0; | |
4537 | if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */ | |
4538 | ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev); | |
4539 | ||
4540 | if (!ret) | |
4541 | dev_set_rx_mode(dev); | |
4542 | } | |
4543 | ||
4544 | if ((flags ^ dev->gflags) & IFF_PROMISC) { | |
4545 | int inc = (flags & IFF_PROMISC) ? 1 : -1; | |
4546 | ||
4547 | dev->gflags ^= IFF_PROMISC; | |
4548 | dev_set_promiscuity(dev, inc); | |
4549 | } | |
4550 | ||
4551 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
4552 | is important. Some (broken) drivers set IFF_PROMISC, when | |
4553 | IFF_ALLMULTI is requested not asking us and not reporting. | |
4554 | */ | |
4555 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
4556 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; | |
4557 | ||
4558 | dev->gflags ^= IFF_ALLMULTI; | |
4559 | dev_set_allmulti(dev, inc); | |
4560 | } | |
4561 | ||
4562 | return ret; | |
4563 | } | |
4564 | ||
4565 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags) | |
4566 | { | |
4567 | unsigned int changes = dev->flags ^ old_flags; | |
4568 | ||
4569 | if (changes & IFF_UP) { | |
4570 | if (dev->flags & IFF_UP) | |
4571 | call_netdevice_notifiers(NETDEV_UP, dev); | |
4572 | else | |
4573 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
4574 | } | |
4575 | ||
4576 | if (dev->flags & IFF_UP && | |
4577 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) | |
4578 | call_netdevice_notifiers(NETDEV_CHANGE, dev); | |
4579 | } | |
4580 | ||
4581 | /** | |
4582 | * dev_change_flags - change device settings | |
4583 | * @dev: device | |
4584 | * @flags: device state flags | |
4585 | * | |
4586 | * Change settings on device based state flags. The flags are | |
4587 | * in the userspace exported format. | |
4588 | */ | |
4589 | int dev_change_flags(struct net_device *dev, unsigned flags) | |
4590 | { | |
4591 | int ret, changes; | |
4592 | int old_flags = dev->flags; | |
4593 | ||
4594 | ret = __dev_change_flags(dev, flags); | |
4595 | if (ret < 0) | |
4596 | return ret; | |
4597 | ||
4598 | changes = old_flags ^ dev->flags; | |
4599 | if (changes) | |
4600 | rtmsg_ifinfo(RTM_NEWLINK, dev, changes); | |
4601 | ||
4602 | __dev_notify_flags(dev, old_flags); | |
4603 | return ret; | |
4604 | } | |
4605 | EXPORT_SYMBOL(dev_change_flags); | |
4606 | ||
4607 | /** | |
4608 | * dev_set_mtu - Change maximum transfer unit | |
4609 | * @dev: device | |
4610 | * @new_mtu: new transfer unit | |
4611 | * | |
4612 | * Change the maximum transfer size of the network device. | |
4613 | */ | |
4614 | int dev_set_mtu(struct net_device *dev, int new_mtu) | |
4615 | { | |
4616 | const struct net_device_ops *ops = dev->netdev_ops; | |
4617 | int err; | |
4618 | ||
4619 | if (new_mtu == dev->mtu) | |
4620 | return 0; | |
4621 | ||
4622 | /* MTU must be positive. */ | |
4623 | if (new_mtu < 0) | |
4624 | return -EINVAL; | |
4625 | ||
4626 | if (!netif_device_present(dev)) | |
4627 | return -ENODEV; | |
4628 | ||
4629 | err = 0; | |
4630 | if (ops->ndo_change_mtu) | |
4631 | err = ops->ndo_change_mtu(dev, new_mtu); | |
4632 | else | |
4633 | dev->mtu = new_mtu; | |
4634 | ||
4635 | if (!err && dev->flags & IFF_UP) | |
4636 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); | |
4637 | return err; | |
4638 | } | |
4639 | EXPORT_SYMBOL(dev_set_mtu); | |
4640 | ||
4641 | /** | |
4642 | * dev_set_mac_address - Change Media Access Control Address | |
4643 | * @dev: device | |
4644 | * @sa: new address | |
4645 | * | |
4646 | * Change the hardware (MAC) address of the device | |
4647 | */ | |
4648 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) | |
4649 | { | |
4650 | const struct net_device_ops *ops = dev->netdev_ops; | |
4651 | int err; | |
4652 | ||
4653 | if (!ops->ndo_set_mac_address) | |
4654 | return -EOPNOTSUPP; | |
4655 | if (sa->sa_family != dev->type) | |
4656 | return -EINVAL; | |
4657 | if (!netif_device_present(dev)) | |
4658 | return -ENODEV; | |
4659 | err = ops->ndo_set_mac_address(dev, sa); | |
4660 | if (!err) | |
4661 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); | |
4662 | return err; | |
4663 | } | |
4664 | EXPORT_SYMBOL(dev_set_mac_address); | |
4665 | ||
4666 | /* | |
4667 | * Perform the SIOCxIFxxx calls, inside rcu_read_lock() | |
4668 | */ | |
4669 | static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd) | |
4670 | { | |
4671 | int err; | |
4672 | struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name); | |
4673 | ||
4674 | if (!dev) | |
4675 | return -ENODEV; | |
4676 | ||
4677 | switch (cmd) { | |
4678 | case SIOCGIFFLAGS: /* Get interface flags */ | |
4679 | ifr->ifr_flags = (short) dev_get_flags(dev); | |
4680 | return 0; | |
4681 | ||
4682 | case SIOCGIFMETRIC: /* Get the metric on the interface | |
4683 | (currently unused) */ | |
4684 | ifr->ifr_metric = 0; | |
4685 | return 0; | |
4686 | ||
4687 | case SIOCGIFMTU: /* Get the MTU of a device */ | |
4688 | ifr->ifr_mtu = dev->mtu; | |
4689 | return 0; | |
4690 | ||
4691 | case SIOCGIFHWADDR: | |
4692 | if (!dev->addr_len) | |
4693 | memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data); | |
4694 | else | |
4695 | memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr, | |
4696 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
4697 | ifr->ifr_hwaddr.sa_family = dev->type; | |
4698 | return 0; | |
4699 | ||
4700 | case SIOCGIFSLAVE: | |
4701 | err = -EINVAL; | |
4702 | break; | |
4703 | ||
4704 | case SIOCGIFMAP: | |
4705 | ifr->ifr_map.mem_start = dev->mem_start; | |
4706 | ifr->ifr_map.mem_end = dev->mem_end; | |
4707 | ifr->ifr_map.base_addr = dev->base_addr; | |
4708 | ifr->ifr_map.irq = dev->irq; | |
4709 | ifr->ifr_map.dma = dev->dma; | |
4710 | ifr->ifr_map.port = dev->if_port; | |
4711 | return 0; | |
4712 | ||
4713 | case SIOCGIFINDEX: | |
4714 | ifr->ifr_ifindex = dev->ifindex; | |
4715 | return 0; | |
4716 | ||
4717 | case SIOCGIFTXQLEN: | |
4718 | ifr->ifr_qlen = dev->tx_queue_len; | |
4719 | return 0; | |
4720 | ||
4721 | default: | |
4722 | /* dev_ioctl() should ensure this case | |
4723 | * is never reached | |
4724 | */ | |
4725 | WARN_ON(1); | |
4726 | err = -EINVAL; | |
4727 | break; | |
4728 | ||
4729 | } | |
4730 | return err; | |
4731 | } | |
4732 | ||
4733 | /* | |
4734 | * Perform the SIOCxIFxxx calls, inside rtnl_lock() | |
4735 | */ | |
4736 | static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd) | |
4737 | { | |
4738 | int err; | |
4739 | struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name); | |
4740 | const struct net_device_ops *ops; | |
4741 | ||
4742 | if (!dev) | |
4743 | return -ENODEV; | |
4744 | ||
4745 | ops = dev->netdev_ops; | |
4746 | ||
4747 | switch (cmd) { | |
4748 | case SIOCSIFFLAGS: /* Set interface flags */ | |
4749 | return dev_change_flags(dev, ifr->ifr_flags); | |
4750 | ||
4751 | case SIOCSIFMETRIC: /* Set the metric on the interface | |
4752 | (currently unused) */ | |
4753 | return -EOPNOTSUPP; | |
4754 | ||
4755 | case SIOCSIFMTU: /* Set the MTU of a device */ | |
4756 | return dev_set_mtu(dev, ifr->ifr_mtu); | |
4757 | ||
4758 | case SIOCSIFHWADDR: | |
4759 | return dev_set_mac_address(dev, &ifr->ifr_hwaddr); | |
4760 | ||
4761 | case SIOCSIFHWBROADCAST: | |
4762 | if (ifr->ifr_hwaddr.sa_family != dev->type) | |
4763 | return -EINVAL; | |
4764 | memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data, | |
4765 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
4766 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); | |
4767 | return 0; | |
4768 | ||
4769 | case SIOCSIFMAP: | |
4770 | if (ops->ndo_set_config) { | |
4771 | if (!netif_device_present(dev)) | |
4772 | return -ENODEV; | |
4773 | return ops->ndo_set_config(dev, &ifr->ifr_map); | |
4774 | } | |
4775 | return -EOPNOTSUPP; | |
4776 | ||
4777 | case SIOCADDMULTI: | |
4778 | if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) || | |
4779 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) | |
4780 | return -EINVAL; | |
4781 | if (!netif_device_present(dev)) | |
4782 | return -ENODEV; | |
4783 | return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data, | |
4784 | dev->addr_len, 1); | |
4785 | ||
4786 | case SIOCDELMULTI: | |
4787 | if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) || | |
4788 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) | |
4789 | return -EINVAL; | |
4790 | if (!netif_device_present(dev)) | |
4791 | return -ENODEV; | |
4792 | return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data, | |
4793 | dev->addr_len, 1); | |
4794 | ||
4795 | case SIOCSIFTXQLEN: | |
4796 | if (ifr->ifr_qlen < 0) | |
4797 | return -EINVAL; | |
4798 | dev->tx_queue_len = ifr->ifr_qlen; | |
4799 | return 0; | |
4800 | ||
4801 | case SIOCSIFNAME: | |
4802 | ifr->ifr_newname[IFNAMSIZ-1] = '\0'; | |
4803 | return dev_change_name(dev, ifr->ifr_newname); | |
4804 | ||
4805 | /* | |
4806 | * Unknown or private ioctl | |
4807 | */ | |
4808 | default: | |
4809 | if ((cmd >= SIOCDEVPRIVATE && | |
4810 | cmd <= SIOCDEVPRIVATE + 15) || | |
4811 | cmd == SIOCBONDENSLAVE || | |
4812 | cmd == SIOCBONDRELEASE || | |
4813 | cmd == SIOCBONDSETHWADDR || | |
4814 | cmd == SIOCBONDSLAVEINFOQUERY || | |
4815 | cmd == SIOCBONDINFOQUERY || | |
4816 | cmd == SIOCBONDCHANGEACTIVE || | |
4817 | cmd == SIOCGMIIPHY || | |
4818 | cmd == SIOCGMIIREG || | |
4819 | cmd == SIOCSMIIREG || | |
4820 | cmd == SIOCBRADDIF || | |
4821 | cmd == SIOCBRDELIF || | |
4822 | cmd == SIOCSHWTSTAMP || | |
4823 | cmd == SIOCWANDEV) { | |
4824 | err = -EOPNOTSUPP; | |
4825 | if (ops->ndo_do_ioctl) { | |
4826 | if (netif_device_present(dev)) | |
4827 | err = ops->ndo_do_ioctl(dev, ifr, cmd); | |
4828 | else | |
4829 | err = -ENODEV; | |
4830 | } | |
4831 | } else | |
4832 | err = -EINVAL; | |
4833 | ||
4834 | } | |
4835 | return err; | |
4836 | } | |
4837 | ||
4838 | /* | |
4839 | * This function handles all "interface"-type I/O control requests. The actual | |
4840 | * 'doing' part of this is dev_ifsioc above. | |
4841 | */ | |
4842 | ||
4843 | /** | |
4844 | * dev_ioctl - network device ioctl | |
4845 | * @net: the applicable net namespace | |
4846 | * @cmd: command to issue | |
4847 | * @arg: pointer to a struct ifreq in user space | |
4848 | * | |
4849 | * Issue ioctl functions to devices. This is normally called by the | |
4850 | * user space syscall interfaces but can sometimes be useful for | |
4851 | * other purposes. The return value is the return from the syscall if | |
4852 | * positive or a negative errno code on error. | |
4853 | */ | |
4854 | ||
4855 | int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg) | |
4856 | { | |
4857 | struct ifreq ifr; | |
4858 | int ret; | |
4859 | char *colon; | |
4860 | ||
4861 | /* One special case: SIOCGIFCONF takes ifconf argument | |
4862 | and requires shared lock, because it sleeps writing | |
4863 | to user space. | |
4864 | */ | |
4865 | ||
4866 | if (cmd == SIOCGIFCONF) { | |
4867 | rtnl_lock(); | |
4868 | ret = dev_ifconf(net, (char __user *) arg); | |
4869 | rtnl_unlock(); | |
4870 | return ret; | |
4871 | } | |
4872 | if (cmd == SIOCGIFNAME) | |
4873 | return dev_ifname(net, (struct ifreq __user *)arg); | |
4874 | ||
4875 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
4876 | return -EFAULT; | |
4877 | ||
4878 | ifr.ifr_name[IFNAMSIZ-1] = 0; | |
4879 | ||
4880 | colon = strchr(ifr.ifr_name, ':'); | |
4881 | if (colon) | |
4882 | *colon = 0; | |
4883 | ||
4884 | /* | |
4885 | * See which interface the caller is talking about. | |
4886 | */ | |
4887 | ||
4888 | switch (cmd) { | |
4889 | /* | |
4890 | * These ioctl calls: | |
4891 | * - can be done by all. | |
4892 | * - atomic and do not require locking. | |
4893 | * - return a value | |
4894 | */ | |
4895 | case SIOCGIFFLAGS: | |
4896 | case SIOCGIFMETRIC: | |
4897 | case SIOCGIFMTU: | |
4898 | case SIOCGIFHWADDR: | |
4899 | case SIOCGIFSLAVE: | |
4900 | case SIOCGIFMAP: | |
4901 | case SIOCGIFINDEX: | |
4902 | case SIOCGIFTXQLEN: | |
4903 | dev_load(net, ifr.ifr_name); | |
4904 | rcu_read_lock(); | |
4905 | ret = dev_ifsioc_locked(net, &ifr, cmd); | |
4906 | rcu_read_unlock(); | |
4907 | if (!ret) { | |
4908 | if (colon) | |
4909 | *colon = ':'; | |
4910 | if (copy_to_user(arg, &ifr, | |
4911 | sizeof(struct ifreq))) | |
4912 | ret = -EFAULT; | |
4913 | } | |
4914 | return ret; | |
4915 | ||
4916 | case SIOCETHTOOL: | |
4917 | dev_load(net, ifr.ifr_name); | |
4918 | rtnl_lock(); | |
4919 | ret = dev_ethtool(net, &ifr); | |
4920 | rtnl_unlock(); | |
4921 | if (!ret) { | |
4922 | if (colon) | |
4923 | *colon = ':'; | |
4924 | if (copy_to_user(arg, &ifr, | |
4925 | sizeof(struct ifreq))) | |
4926 | ret = -EFAULT; | |
4927 | } | |
4928 | return ret; | |
4929 | ||
4930 | /* | |
4931 | * These ioctl calls: | |
4932 | * - require superuser power. | |
4933 | * - require strict serialization. | |
4934 | * - return a value | |
4935 | */ | |
4936 | case SIOCGMIIPHY: | |
4937 | case SIOCGMIIREG: | |
4938 | case SIOCSIFNAME: | |
4939 | if (!capable(CAP_NET_ADMIN)) | |
4940 | return -EPERM; | |
4941 | dev_load(net, ifr.ifr_name); | |
4942 | rtnl_lock(); | |
4943 | ret = dev_ifsioc(net, &ifr, cmd); | |
4944 | rtnl_unlock(); | |
4945 | if (!ret) { | |
4946 | if (colon) | |
4947 | *colon = ':'; | |
4948 | if (copy_to_user(arg, &ifr, | |
4949 | sizeof(struct ifreq))) | |
4950 | ret = -EFAULT; | |
4951 | } | |
4952 | return ret; | |
4953 | ||
4954 | /* | |
4955 | * These ioctl calls: | |
4956 | * - require superuser power. | |
4957 | * - require strict serialization. | |
4958 | * - do not return a value | |
4959 | */ | |
4960 | case SIOCSIFFLAGS: | |
4961 | case SIOCSIFMETRIC: | |
4962 | case SIOCSIFMTU: | |
4963 | case SIOCSIFMAP: | |
4964 | case SIOCSIFHWADDR: | |
4965 | case SIOCSIFSLAVE: | |
4966 | case SIOCADDMULTI: | |
4967 | case SIOCDELMULTI: | |
4968 | case SIOCSIFHWBROADCAST: | |
4969 | case SIOCSIFTXQLEN: | |
4970 | case SIOCSMIIREG: | |
4971 | case SIOCBONDENSLAVE: | |
4972 | case SIOCBONDRELEASE: | |
4973 | case SIOCBONDSETHWADDR: | |
4974 | case SIOCBONDCHANGEACTIVE: | |
4975 | case SIOCBRADDIF: | |
4976 | case SIOCBRDELIF: | |
4977 | case SIOCSHWTSTAMP: | |
4978 | if (!capable(CAP_NET_ADMIN)) | |
4979 | return -EPERM; | |
4980 | /* fall through */ | |
4981 | case SIOCBONDSLAVEINFOQUERY: | |
4982 | case SIOCBONDINFOQUERY: | |
4983 | dev_load(net, ifr.ifr_name); | |
4984 | rtnl_lock(); | |
4985 | ret = dev_ifsioc(net, &ifr, cmd); | |
4986 | rtnl_unlock(); | |
4987 | return ret; | |
4988 | ||
4989 | case SIOCGIFMEM: | |
4990 | /* Get the per device memory space. We can add this but | |
4991 | * currently do not support it */ | |
4992 | case SIOCSIFMEM: | |
4993 | /* Set the per device memory buffer space. | |
4994 | * Not applicable in our case */ | |
4995 | case SIOCSIFLINK: | |
4996 | return -EINVAL; | |
4997 | ||
4998 | /* | |
4999 | * Unknown or private ioctl. | |
5000 | */ | |
5001 | default: | |
5002 | if (cmd == SIOCWANDEV || | |
5003 | (cmd >= SIOCDEVPRIVATE && | |
5004 | cmd <= SIOCDEVPRIVATE + 15)) { | |
5005 | dev_load(net, ifr.ifr_name); | |
5006 | rtnl_lock(); | |
5007 | ret = dev_ifsioc(net, &ifr, cmd); | |
5008 | rtnl_unlock(); | |
5009 | if (!ret && copy_to_user(arg, &ifr, | |
5010 | sizeof(struct ifreq))) | |
5011 | ret = -EFAULT; | |
5012 | return ret; | |
5013 | } | |
5014 | /* Take care of Wireless Extensions */ | |
5015 | if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) | |
5016 | return wext_handle_ioctl(net, &ifr, cmd, arg); | |
5017 | return -EINVAL; | |
5018 | } | |
5019 | } | |
5020 | ||
5021 | ||
5022 | /** | |
5023 | * dev_new_index - allocate an ifindex | |
5024 | * @net: the applicable net namespace | |
5025 | * | |
5026 | * Returns a suitable unique value for a new device interface | |
5027 | * number. The caller must hold the rtnl semaphore or the | |
5028 | * dev_base_lock to be sure it remains unique. | |
5029 | */ | |
5030 | static int dev_new_index(struct net *net) | |
5031 | { | |
5032 | static int ifindex; | |
5033 | for (;;) { | |
5034 | if (++ifindex <= 0) | |
5035 | ifindex = 1; | |
5036 | if (!__dev_get_by_index(net, ifindex)) | |
5037 | return ifindex; | |
5038 | } | |
5039 | } | |
5040 | ||
5041 | /* Delayed registration/unregisteration */ | |
5042 | static LIST_HEAD(net_todo_list); | |
5043 | ||
5044 | static void net_set_todo(struct net_device *dev) | |
5045 | { | |
5046 | list_add_tail(&dev->todo_list, &net_todo_list); | |
5047 | } | |
5048 | ||
5049 | static void rollback_registered_many(struct list_head *head) | |
5050 | { | |
5051 | struct net_device *dev, *tmp; | |
5052 | ||
5053 | BUG_ON(dev_boot_phase); | |
5054 | ASSERT_RTNL(); | |
5055 | ||
5056 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { | |
5057 | /* Some devices call without registering | |
5058 | * for initialization unwind. Remove those | |
5059 | * devices and proceed with the remaining. | |
5060 | */ | |
5061 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
5062 | pr_debug("unregister_netdevice: device %s/%p never " | |
5063 | "was registered\n", dev->name, dev); | |
5064 | ||
5065 | WARN_ON(1); | |
5066 | list_del(&dev->unreg_list); | |
5067 | continue; | |
5068 | } | |
5069 | ||
5070 | BUG_ON(dev->reg_state != NETREG_REGISTERED); | |
5071 | ||
5072 | /* If device is running, close it first. */ | |
5073 | dev_close(dev); | |
5074 | ||
5075 | /* And unlink it from device chain. */ | |
5076 | unlist_netdevice(dev); | |
5077 | ||
5078 | dev->reg_state = NETREG_UNREGISTERING; | |
5079 | } | |
5080 | ||
5081 | synchronize_net(); | |
5082 | ||
5083 | list_for_each_entry(dev, head, unreg_list) { | |
5084 | /* Shutdown queueing discipline. */ | |
5085 | dev_shutdown(dev); | |
5086 | ||
5087 | ||
5088 | /* Notify protocols, that we are about to destroy | |
5089 | this device. They should clean all the things. | |
5090 | */ | |
5091 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
5092 | ||
5093 | if (!dev->rtnl_link_ops || | |
5094 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
5095 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U); | |
5096 | ||
5097 | /* | |
5098 | * Flush the unicast and multicast chains | |
5099 | */ | |
5100 | dev_unicast_flush(dev); | |
5101 | dev_addr_discard(dev); | |
5102 | ||
5103 | if (dev->netdev_ops->ndo_uninit) | |
5104 | dev->netdev_ops->ndo_uninit(dev); | |
5105 | ||
5106 | /* Notifier chain MUST detach us from master device. */ | |
5107 | WARN_ON(dev->master); | |
5108 | ||
5109 | /* Remove entries from kobject tree */ | |
5110 | netdev_unregister_kobject(dev); | |
5111 | } | |
5112 | ||
5113 | /* Process any work delayed until the end of the batch */ | |
5114 | dev = list_first_entry(head, struct net_device, unreg_list); | |
5115 | call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev); | |
5116 | ||
5117 | synchronize_net(); | |
5118 | ||
5119 | list_for_each_entry(dev, head, unreg_list) | |
5120 | dev_put(dev); | |
5121 | } | |
5122 | ||
5123 | static void rollback_registered(struct net_device *dev) | |
5124 | { | |
5125 | LIST_HEAD(single); | |
5126 | ||
5127 | list_add(&dev->unreg_list, &single); | |
5128 | rollback_registered_many(&single); | |
5129 | } | |
5130 | ||
5131 | static void __netdev_init_queue_locks_one(struct net_device *dev, | |
5132 | struct netdev_queue *dev_queue, | |
5133 | void *_unused) | |
5134 | { | |
5135 | spin_lock_init(&dev_queue->_xmit_lock); | |
5136 | netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type); | |
5137 | dev_queue->xmit_lock_owner = -1; | |
5138 | } | |
5139 | ||
5140 | static void netdev_init_queue_locks(struct net_device *dev) | |
5141 | { | |
5142 | netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL); | |
5143 | __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL); | |
5144 | } | |
5145 | ||
5146 | unsigned long netdev_fix_features(unsigned long features, const char *name) | |
5147 | { | |
5148 | /* Fix illegal SG+CSUM combinations. */ | |
5149 | if ((features & NETIF_F_SG) && | |
5150 | !(features & NETIF_F_ALL_CSUM)) { | |
5151 | if (name) | |
5152 | printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no " | |
5153 | "checksum feature.\n", name); | |
5154 | features &= ~NETIF_F_SG; | |
5155 | } | |
5156 | ||
5157 | /* TSO requires that SG is present as well. */ | |
5158 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) { | |
5159 | if (name) | |
5160 | printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no " | |
5161 | "SG feature.\n", name); | |
5162 | features &= ~NETIF_F_TSO; | |
5163 | } | |
5164 | ||
5165 | if (features & NETIF_F_UFO) { | |
5166 | if (!(features & NETIF_F_GEN_CSUM)) { | |
5167 | if (name) | |
5168 | printk(KERN_ERR "%s: Dropping NETIF_F_UFO " | |
5169 | "since no NETIF_F_HW_CSUM feature.\n", | |
5170 | name); | |
5171 | features &= ~NETIF_F_UFO; | |
5172 | } | |
5173 | ||
5174 | if (!(features & NETIF_F_SG)) { | |
5175 | if (name) | |
5176 | printk(KERN_ERR "%s: Dropping NETIF_F_UFO " | |
5177 | "since no NETIF_F_SG feature.\n", name); | |
5178 | features &= ~NETIF_F_UFO; | |
5179 | } | |
5180 | } | |
5181 | ||
5182 | return features; | |
5183 | } | |
5184 | EXPORT_SYMBOL(netdev_fix_features); | |
5185 | ||
5186 | /** | |
5187 | * netif_stacked_transfer_operstate - transfer operstate | |
5188 | * @rootdev: the root or lower level device to transfer state from | |
5189 | * @dev: the device to transfer operstate to | |
5190 | * | |
5191 | * Transfer operational state from root to device. This is normally | |
5192 | * called when a stacking relationship exists between the root | |
5193 | * device and the device(a leaf device). | |
5194 | */ | |
5195 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
5196 | struct net_device *dev) | |
5197 | { | |
5198 | if (rootdev->operstate == IF_OPER_DORMANT) | |
5199 | netif_dormant_on(dev); | |
5200 | else | |
5201 | netif_dormant_off(dev); | |
5202 | ||
5203 | if (netif_carrier_ok(rootdev)) { | |
5204 | if (!netif_carrier_ok(dev)) | |
5205 | netif_carrier_on(dev); | |
5206 | } else { | |
5207 | if (netif_carrier_ok(dev)) | |
5208 | netif_carrier_off(dev); | |
5209 | } | |
5210 | } | |
5211 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
5212 | ||
5213 | /** | |
5214 | * register_netdevice - register a network device | |
5215 | * @dev: device to register | |
5216 | * | |
5217 | * Take a completed network device structure and add it to the kernel | |
5218 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
5219 | * chain. 0 is returned on success. A negative errno code is returned | |
5220 | * on a failure to set up the device, or if the name is a duplicate. | |
5221 | * | |
5222 | * Callers must hold the rtnl semaphore. You may want | |
5223 | * register_netdev() instead of this. | |
5224 | * | |
5225 | * BUGS: | |
5226 | * The locking appears insufficient to guarantee two parallel registers | |
5227 | * will not get the same name. | |
5228 | */ | |
5229 | ||
5230 | int register_netdevice(struct net_device *dev) | |
5231 | { | |
5232 | int ret; | |
5233 | struct net *net = dev_net(dev); | |
5234 | ||
5235 | BUG_ON(dev_boot_phase); | |
5236 | ASSERT_RTNL(); | |
5237 | ||
5238 | might_sleep(); | |
5239 | ||
5240 | /* When net_device's are persistent, this will be fatal. */ | |
5241 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
5242 | BUG_ON(!net); | |
5243 | ||
5244 | spin_lock_init(&dev->addr_list_lock); | |
5245 | netdev_set_addr_lockdep_class(dev); | |
5246 | netdev_init_queue_locks(dev); | |
5247 | ||
5248 | dev->iflink = -1; | |
5249 | ||
5250 | if (!dev->num_rx_queues) { | |
5251 | /* | |
5252 | * Allocate a single RX queue if driver never called | |
5253 | * alloc_netdev_mq | |
5254 | */ | |
5255 | ||
5256 | dev->_rx = kzalloc(sizeof(struct netdev_rx_queue), GFP_KERNEL); | |
5257 | if (!dev->_rx) { | |
5258 | ret = -ENOMEM; | |
5259 | goto out; | |
5260 | } | |
5261 | ||
5262 | dev->_rx->first = dev->_rx; | |
5263 | atomic_set(&dev->_rx->count, 1); | |
5264 | dev->num_rx_queues = 1; | |
5265 | } | |
5266 | ||
5267 | /* Init, if this function is available */ | |
5268 | if (dev->netdev_ops->ndo_init) { | |
5269 | ret = dev->netdev_ops->ndo_init(dev); | |
5270 | if (ret) { | |
5271 | if (ret > 0) | |
5272 | ret = -EIO; | |
5273 | goto out; | |
5274 | } | |
5275 | } | |
5276 | ||
5277 | ret = dev_get_valid_name(net, dev->name, dev->name, 0); | |
5278 | if (ret) | |
5279 | goto err_uninit; | |
5280 | ||
5281 | dev->ifindex = dev_new_index(net); | |
5282 | if (dev->iflink == -1) | |
5283 | dev->iflink = dev->ifindex; | |
5284 | ||
5285 | /* Fix illegal checksum combinations */ | |
5286 | if ((dev->features & NETIF_F_HW_CSUM) && | |
5287 | (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
5288 | printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n", | |
5289 | dev->name); | |
5290 | dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); | |
5291 | } | |
5292 | ||
5293 | if ((dev->features & NETIF_F_NO_CSUM) && | |
5294 | (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
5295 | printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n", | |
5296 | dev->name); | |
5297 | dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM); | |
5298 | } | |
5299 | ||
5300 | dev->features = netdev_fix_features(dev->features, dev->name); | |
5301 | ||
5302 | /* Enable software GSO if SG is supported. */ | |
5303 | if (dev->features & NETIF_F_SG) | |
5304 | dev->features |= NETIF_F_GSO; | |
5305 | ||
5306 | netdev_initialize_kobject(dev); | |
5307 | ||
5308 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); | |
5309 | ret = notifier_to_errno(ret); | |
5310 | if (ret) | |
5311 | goto err_uninit; | |
5312 | ||
5313 | ret = netdev_register_kobject(dev); | |
5314 | if (ret) | |
5315 | goto err_uninit; | |
5316 | dev->reg_state = NETREG_REGISTERED; | |
5317 | ||
5318 | /* | |
5319 | * Default initial state at registry is that the | |
5320 | * device is present. | |
5321 | */ | |
5322 | ||
5323 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
5324 | ||
5325 | dev_init_scheduler(dev); | |
5326 | dev_hold(dev); | |
5327 | list_netdevice(dev); | |
5328 | ||
5329 | /* Notify protocols, that a new device appeared. */ | |
5330 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
5331 | ret = notifier_to_errno(ret); | |
5332 | if (ret) { | |
5333 | rollback_registered(dev); | |
5334 | dev->reg_state = NETREG_UNREGISTERED; | |
5335 | } | |
5336 | /* | |
5337 | * Prevent userspace races by waiting until the network | |
5338 | * device is fully setup before sending notifications. | |
5339 | */ | |
5340 | if (!dev->rtnl_link_ops || | |
5341 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
5342 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U); | |
5343 | ||
5344 | out: | |
5345 | return ret; | |
5346 | ||
5347 | err_uninit: | |
5348 | if (dev->netdev_ops->ndo_uninit) | |
5349 | dev->netdev_ops->ndo_uninit(dev); | |
5350 | goto out; | |
5351 | } | |
5352 | EXPORT_SYMBOL(register_netdevice); | |
5353 | ||
5354 | /** | |
5355 | * init_dummy_netdev - init a dummy network device for NAPI | |
5356 | * @dev: device to init | |
5357 | * | |
5358 | * This takes a network device structure and initialize the minimum | |
5359 | * amount of fields so it can be used to schedule NAPI polls without | |
5360 | * registering a full blown interface. This is to be used by drivers | |
5361 | * that need to tie several hardware interfaces to a single NAPI | |
5362 | * poll scheduler due to HW limitations. | |
5363 | */ | |
5364 | int init_dummy_netdev(struct net_device *dev) | |
5365 | { | |
5366 | /* Clear everything. Note we don't initialize spinlocks | |
5367 | * are they aren't supposed to be taken by any of the | |
5368 | * NAPI code and this dummy netdev is supposed to be | |
5369 | * only ever used for NAPI polls | |
5370 | */ | |
5371 | memset(dev, 0, sizeof(struct net_device)); | |
5372 | ||
5373 | /* make sure we BUG if trying to hit standard | |
5374 | * register/unregister code path | |
5375 | */ | |
5376 | dev->reg_state = NETREG_DUMMY; | |
5377 | ||
5378 | /* initialize the ref count */ | |
5379 | atomic_set(&dev->refcnt, 1); | |
5380 | ||
5381 | /* NAPI wants this */ | |
5382 | INIT_LIST_HEAD(&dev->napi_list); | |
5383 | ||
5384 | /* a dummy interface is started by default */ | |
5385 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
5386 | set_bit(__LINK_STATE_START, &dev->state); | |
5387 | ||
5388 | return 0; | |
5389 | } | |
5390 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
5391 | ||
5392 | ||
5393 | /** | |
5394 | * register_netdev - register a network device | |
5395 | * @dev: device to register | |
5396 | * | |
5397 | * Take a completed network device structure and add it to the kernel | |
5398 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
5399 | * chain. 0 is returned on success. A negative errno code is returned | |
5400 | * on a failure to set up the device, or if the name is a duplicate. | |
5401 | * | |
5402 | * This is a wrapper around register_netdevice that takes the rtnl semaphore | |
5403 | * and expands the device name if you passed a format string to | |
5404 | * alloc_netdev. | |
5405 | */ | |
5406 | int register_netdev(struct net_device *dev) | |
5407 | { | |
5408 | int err; | |
5409 | ||
5410 | rtnl_lock(); | |
5411 | ||
5412 | /* | |
5413 | * If the name is a format string the caller wants us to do a | |
5414 | * name allocation. | |
5415 | */ | |
5416 | if (strchr(dev->name, '%')) { | |
5417 | err = dev_alloc_name(dev, dev->name); | |
5418 | if (err < 0) | |
5419 | goto out; | |
5420 | } | |
5421 | ||
5422 | err = register_netdevice(dev); | |
5423 | out: | |
5424 | rtnl_unlock(); | |
5425 | return err; | |
5426 | } | |
5427 | EXPORT_SYMBOL(register_netdev); | |
5428 | ||
5429 | /* | |
5430 | * netdev_wait_allrefs - wait until all references are gone. | |
5431 | * | |
5432 | * This is called when unregistering network devices. | |
5433 | * | |
5434 | * Any protocol or device that holds a reference should register | |
5435 | * for netdevice notification, and cleanup and put back the | |
5436 | * reference if they receive an UNREGISTER event. | |
5437 | * We can get stuck here if buggy protocols don't correctly | |
5438 | * call dev_put. | |
5439 | */ | |
5440 | static void netdev_wait_allrefs(struct net_device *dev) | |
5441 | { | |
5442 | unsigned long rebroadcast_time, warning_time; | |
5443 | ||
5444 | linkwatch_forget_dev(dev); | |
5445 | ||
5446 | rebroadcast_time = warning_time = jiffies; | |
5447 | while (atomic_read(&dev->refcnt) != 0) { | |
5448 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { | |
5449 | rtnl_lock(); | |
5450 | ||
5451 | /* Rebroadcast unregister notification */ | |
5452 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
5453 | /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users | |
5454 | * should have already handle it the first time */ | |
5455 | ||
5456 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, | |
5457 | &dev->state)) { | |
5458 | /* We must not have linkwatch events | |
5459 | * pending on unregister. If this | |
5460 | * happens, we simply run the queue | |
5461 | * unscheduled, resulting in a noop | |
5462 | * for this device. | |
5463 | */ | |
5464 | linkwatch_run_queue(); | |
5465 | } | |
5466 | ||
5467 | __rtnl_unlock(); | |
5468 | ||
5469 | rebroadcast_time = jiffies; | |
5470 | } | |
5471 | ||
5472 | msleep(250); | |
5473 | ||
5474 | if (time_after(jiffies, warning_time + 10 * HZ)) { | |
5475 | printk(KERN_EMERG "unregister_netdevice: " | |
5476 | "waiting for %s to become free. Usage " | |
5477 | "count = %d\n", | |
5478 | dev->name, atomic_read(&dev->refcnt)); | |
5479 | warning_time = jiffies; | |
5480 | } | |
5481 | } | |
5482 | } | |
5483 | ||
5484 | /* The sequence is: | |
5485 | * | |
5486 | * rtnl_lock(); | |
5487 | * ... | |
5488 | * register_netdevice(x1); | |
5489 | * register_netdevice(x2); | |
5490 | * ... | |
5491 | * unregister_netdevice(y1); | |
5492 | * unregister_netdevice(y2); | |
5493 | * ... | |
5494 | * rtnl_unlock(); | |
5495 | * free_netdev(y1); | |
5496 | * free_netdev(y2); | |
5497 | * | |
5498 | * We are invoked by rtnl_unlock(). | |
5499 | * This allows us to deal with problems: | |
5500 | * 1) We can delete sysfs objects which invoke hotplug | |
5501 | * without deadlocking with linkwatch via keventd. | |
5502 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
5503 | * safely in order to wait for the netdev refcnt to drop to zero. | |
5504 | * | |
5505 | * We must not return until all unregister events added during | |
5506 | * the interval the lock was held have been completed. | |
5507 | */ | |
5508 | void netdev_run_todo(void) | |
5509 | { | |
5510 | struct list_head list; | |
5511 | ||
5512 | /* Snapshot list, allow later requests */ | |
5513 | list_replace_init(&net_todo_list, &list); | |
5514 | ||
5515 | __rtnl_unlock(); | |
5516 | ||
5517 | while (!list_empty(&list)) { | |
5518 | struct net_device *dev | |
5519 | = list_first_entry(&list, struct net_device, todo_list); | |
5520 | list_del(&dev->todo_list); | |
5521 | ||
5522 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { | |
5523 | printk(KERN_ERR "network todo '%s' but state %d\n", | |
5524 | dev->name, dev->reg_state); | |
5525 | dump_stack(); | |
5526 | continue; | |
5527 | } | |
5528 | ||
5529 | dev->reg_state = NETREG_UNREGISTERED; | |
5530 | ||
5531 | on_each_cpu(flush_backlog, dev, 1); | |
5532 | ||
5533 | netdev_wait_allrefs(dev); | |
5534 | ||
5535 | /* paranoia */ | |
5536 | BUG_ON(atomic_read(&dev->refcnt)); | |
5537 | WARN_ON(dev->ip_ptr); | |
5538 | WARN_ON(dev->ip6_ptr); | |
5539 | WARN_ON(dev->dn_ptr); | |
5540 | ||
5541 | if (dev->destructor) | |
5542 | dev->destructor(dev); | |
5543 | ||
5544 | /* Free network device */ | |
5545 | kobject_put(&dev->dev.kobj); | |
5546 | } | |
5547 | } | |
5548 | ||
5549 | /** | |
5550 | * dev_txq_stats_fold - fold tx_queues stats | |
5551 | * @dev: device to get statistics from | |
5552 | * @stats: struct net_device_stats to hold results | |
5553 | */ | |
5554 | void dev_txq_stats_fold(const struct net_device *dev, | |
5555 | struct net_device_stats *stats) | |
5556 | { | |
5557 | unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0; | |
5558 | unsigned int i; | |
5559 | struct netdev_queue *txq; | |
5560 | ||
5561 | for (i = 0; i < dev->num_tx_queues; i++) { | |
5562 | txq = netdev_get_tx_queue(dev, i); | |
5563 | tx_bytes += txq->tx_bytes; | |
5564 | tx_packets += txq->tx_packets; | |
5565 | tx_dropped += txq->tx_dropped; | |
5566 | } | |
5567 | if (tx_bytes || tx_packets || tx_dropped) { | |
5568 | stats->tx_bytes = tx_bytes; | |
5569 | stats->tx_packets = tx_packets; | |
5570 | stats->tx_dropped = tx_dropped; | |
5571 | } | |
5572 | } | |
5573 | EXPORT_SYMBOL(dev_txq_stats_fold); | |
5574 | ||
5575 | /** | |
5576 | * dev_get_stats - get network device statistics | |
5577 | * @dev: device to get statistics from | |
5578 | * | |
5579 | * Get network statistics from device. The device driver may provide | |
5580 | * its own method by setting dev->netdev_ops->get_stats; otherwise | |
5581 | * the internal statistics structure is used. | |
5582 | */ | |
5583 | const struct net_device_stats *dev_get_stats(struct net_device *dev) | |
5584 | { | |
5585 | const struct net_device_ops *ops = dev->netdev_ops; | |
5586 | ||
5587 | if (ops->ndo_get_stats) | |
5588 | return ops->ndo_get_stats(dev); | |
5589 | ||
5590 | dev_txq_stats_fold(dev, &dev->stats); | |
5591 | return &dev->stats; | |
5592 | } | |
5593 | EXPORT_SYMBOL(dev_get_stats); | |
5594 | ||
5595 | static void netdev_init_one_queue(struct net_device *dev, | |
5596 | struct netdev_queue *queue, | |
5597 | void *_unused) | |
5598 | { | |
5599 | queue->dev = dev; | |
5600 | } | |
5601 | ||
5602 | static void netdev_init_queues(struct net_device *dev) | |
5603 | { | |
5604 | netdev_init_one_queue(dev, &dev->rx_queue, NULL); | |
5605 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); | |
5606 | spin_lock_init(&dev->tx_global_lock); | |
5607 | } | |
5608 | ||
5609 | /** | |
5610 | * alloc_netdev_mq - allocate network device | |
5611 | * @sizeof_priv: size of private data to allocate space for | |
5612 | * @name: device name format string | |
5613 | * @setup: callback to initialize device | |
5614 | * @queue_count: the number of subqueues to allocate | |
5615 | * | |
5616 | * Allocates a struct net_device with private data area for driver use | |
5617 | * and performs basic initialization. Also allocates subquue structs | |
5618 | * for each queue on the device at the end of the netdevice. | |
5619 | */ | |
5620 | struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name, | |
5621 | void (*setup)(struct net_device *), unsigned int queue_count) | |
5622 | { | |
5623 | struct netdev_queue *tx; | |
5624 | struct netdev_rx_queue *rx; | |
5625 | struct net_device *dev; | |
5626 | size_t alloc_size; | |
5627 | struct net_device *p; | |
5628 | int i; | |
5629 | ||
5630 | BUG_ON(strlen(name) >= sizeof(dev->name)); | |
5631 | ||
5632 | alloc_size = sizeof(struct net_device); | |
5633 | if (sizeof_priv) { | |
5634 | /* ensure 32-byte alignment of private area */ | |
5635 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); | |
5636 | alloc_size += sizeof_priv; | |
5637 | } | |
5638 | /* ensure 32-byte alignment of whole construct */ | |
5639 | alloc_size += NETDEV_ALIGN - 1; | |
5640 | ||
5641 | p = kzalloc(alloc_size, GFP_KERNEL); | |
5642 | if (!p) { | |
5643 | printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n"); | |
5644 | return NULL; | |
5645 | } | |
5646 | ||
5647 | tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL); | |
5648 | if (!tx) { | |
5649 | printk(KERN_ERR "alloc_netdev: Unable to allocate " | |
5650 | "tx qdiscs.\n"); | |
5651 | goto free_p; | |
5652 | } | |
5653 | ||
5654 | rx = kcalloc(queue_count, sizeof(struct netdev_rx_queue), GFP_KERNEL); | |
5655 | if (!rx) { | |
5656 | printk(KERN_ERR "alloc_netdev: Unable to allocate " | |
5657 | "rx queues.\n"); | |
5658 | goto free_tx; | |
5659 | } | |
5660 | ||
5661 | atomic_set(&rx->count, queue_count); | |
5662 | ||
5663 | /* | |
5664 | * Set a pointer to first element in the array which holds the | |
5665 | * reference count. | |
5666 | */ | |
5667 | for (i = 0; i < queue_count; i++) | |
5668 | rx[i].first = rx; | |
5669 | ||
5670 | dev = PTR_ALIGN(p, NETDEV_ALIGN); | |
5671 | dev->padded = (char *)dev - (char *)p; | |
5672 | ||
5673 | if (dev_addr_init(dev)) | |
5674 | goto free_rx; | |
5675 | ||
5676 | dev_unicast_init(dev); | |
5677 | ||
5678 | dev_net_set(dev, &init_net); | |
5679 | ||
5680 | dev->_tx = tx; | |
5681 | dev->num_tx_queues = queue_count; | |
5682 | dev->real_num_tx_queues = queue_count; | |
5683 | ||
5684 | dev->_rx = rx; | |
5685 | dev->num_rx_queues = queue_count; | |
5686 | ||
5687 | dev->gso_max_size = GSO_MAX_SIZE; | |
5688 | ||
5689 | netdev_init_queues(dev); | |
5690 | ||
5691 | INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list); | |
5692 | dev->ethtool_ntuple_list.count = 0; | |
5693 | INIT_LIST_HEAD(&dev->napi_list); | |
5694 | INIT_LIST_HEAD(&dev->unreg_list); | |
5695 | INIT_LIST_HEAD(&dev->link_watch_list); | |
5696 | dev->priv_flags = IFF_XMIT_DST_RELEASE; | |
5697 | setup(dev); | |
5698 | strcpy(dev->name, name); | |
5699 | return dev; | |
5700 | ||
5701 | free_rx: | |
5702 | kfree(rx); | |
5703 | free_tx: | |
5704 | kfree(tx); | |
5705 | free_p: | |
5706 | kfree(p); | |
5707 | return NULL; | |
5708 | } | |
5709 | EXPORT_SYMBOL(alloc_netdev_mq); | |
5710 | ||
5711 | /** | |
5712 | * free_netdev - free network device | |
5713 | * @dev: device | |
5714 | * | |
5715 | * This function does the last stage of destroying an allocated device | |
5716 | * interface. The reference to the device object is released. | |
5717 | * If this is the last reference then it will be freed. | |
5718 | */ | |
5719 | void free_netdev(struct net_device *dev) | |
5720 | { | |
5721 | struct napi_struct *p, *n; | |
5722 | ||
5723 | release_net(dev_net(dev)); | |
5724 | ||
5725 | kfree(dev->_tx); | |
5726 | ||
5727 | /* Flush device addresses */ | |
5728 | dev_addr_flush(dev); | |
5729 | ||
5730 | /* Clear ethtool n-tuple list */ | |
5731 | ethtool_ntuple_flush(dev); | |
5732 | ||
5733 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) | |
5734 | netif_napi_del(p); | |
5735 | ||
5736 | /* Compatibility with error handling in drivers */ | |
5737 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
5738 | kfree((char *)dev - dev->padded); | |
5739 | return; | |
5740 | } | |
5741 | ||
5742 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
5743 | dev->reg_state = NETREG_RELEASED; | |
5744 | ||
5745 | /* will free via device release */ | |
5746 | put_device(&dev->dev); | |
5747 | } | |
5748 | EXPORT_SYMBOL(free_netdev); | |
5749 | ||
5750 | /** | |
5751 | * synchronize_net - Synchronize with packet receive processing | |
5752 | * | |
5753 | * Wait for packets currently being received to be done. | |
5754 | * Does not block later packets from starting. | |
5755 | */ | |
5756 | void synchronize_net(void) | |
5757 | { | |
5758 | might_sleep(); | |
5759 | synchronize_rcu(); | |
5760 | } | |
5761 | EXPORT_SYMBOL(synchronize_net); | |
5762 | ||
5763 | /** | |
5764 | * unregister_netdevice_queue - remove device from the kernel | |
5765 | * @dev: device | |
5766 | * @head: list | |
5767 | * | |
5768 | * This function shuts down a device interface and removes it | |
5769 | * from the kernel tables. | |
5770 | * If head not NULL, device is queued to be unregistered later. | |
5771 | * | |
5772 | * Callers must hold the rtnl semaphore. You may want | |
5773 | * unregister_netdev() instead of this. | |
5774 | */ | |
5775 | ||
5776 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) | |
5777 | { | |
5778 | ASSERT_RTNL(); | |
5779 | ||
5780 | if (head) { | |
5781 | list_move_tail(&dev->unreg_list, head); | |
5782 | } else { | |
5783 | rollback_registered(dev); | |
5784 | /* Finish processing unregister after unlock */ | |
5785 | net_set_todo(dev); | |
5786 | } | |
5787 | } | |
5788 | EXPORT_SYMBOL(unregister_netdevice_queue); | |
5789 | ||
5790 | /** | |
5791 | * unregister_netdevice_many - unregister many devices | |
5792 | * @head: list of devices | |
5793 | */ | |
5794 | void unregister_netdevice_many(struct list_head *head) | |
5795 | { | |
5796 | struct net_device *dev; | |
5797 | ||
5798 | if (!list_empty(head)) { | |
5799 | rollback_registered_many(head); | |
5800 | list_for_each_entry(dev, head, unreg_list) | |
5801 | net_set_todo(dev); | |
5802 | } | |
5803 | } | |
5804 | EXPORT_SYMBOL(unregister_netdevice_many); | |
5805 | ||
5806 | /** | |
5807 | * unregister_netdev - remove device from the kernel | |
5808 | * @dev: device | |
5809 | * | |
5810 | * This function shuts down a device interface and removes it | |
5811 | * from the kernel tables. | |
5812 | * | |
5813 | * This is just a wrapper for unregister_netdevice that takes | |
5814 | * the rtnl semaphore. In general you want to use this and not | |
5815 | * unregister_netdevice. | |
5816 | */ | |
5817 | void unregister_netdev(struct net_device *dev) | |
5818 | { | |
5819 | rtnl_lock(); | |
5820 | unregister_netdevice(dev); | |
5821 | rtnl_unlock(); | |
5822 | } | |
5823 | EXPORT_SYMBOL(unregister_netdev); | |
5824 | ||
5825 | /** | |
5826 | * dev_change_net_namespace - move device to different nethost namespace | |
5827 | * @dev: device | |
5828 | * @net: network namespace | |
5829 | * @pat: If not NULL name pattern to try if the current device name | |
5830 | * is already taken in the destination network namespace. | |
5831 | * | |
5832 | * This function shuts down a device interface and moves it | |
5833 | * to a new network namespace. On success 0 is returned, on | |
5834 | * a failure a netagive errno code is returned. | |
5835 | * | |
5836 | * Callers must hold the rtnl semaphore. | |
5837 | */ | |
5838 | ||
5839 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
5840 | { | |
5841 | int err; | |
5842 | ||
5843 | ASSERT_RTNL(); | |
5844 | ||
5845 | /* Don't allow namespace local devices to be moved. */ | |
5846 | err = -EINVAL; | |
5847 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
5848 | goto out; | |
5849 | ||
5850 | #ifdef CONFIG_SYSFS | |
5851 | /* Don't allow real devices to be moved when sysfs | |
5852 | * is enabled. | |
5853 | */ | |
5854 | err = -EINVAL; | |
5855 | if (dev->dev.parent) | |
5856 | goto out; | |
5857 | #endif | |
5858 | ||
5859 | /* Ensure the device has been registrered */ | |
5860 | err = -EINVAL; | |
5861 | if (dev->reg_state != NETREG_REGISTERED) | |
5862 | goto out; | |
5863 | ||
5864 | /* Get out if there is nothing todo */ | |
5865 | err = 0; | |
5866 | if (net_eq(dev_net(dev), net)) | |
5867 | goto out; | |
5868 | ||
5869 | /* Pick the destination device name, and ensure | |
5870 | * we can use it in the destination network namespace. | |
5871 | */ | |
5872 | err = -EEXIST; | |
5873 | if (__dev_get_by_name(net, dev->name)) { | |
5874 | /* We get here if we can't use the current device name */ | |
5875 | if (!pat) | |
5876 | goto out; | |
5877 | if (dev_get_valid_name(net, pat, dev->name, 1)) | |
5878 | goto out; | |
5879 | } | |
5880 | ||
5881 | /* | |
5882 | * And now a mini version of register_netdevice unregister_netdevice. | |
5883 | */ | |
5884 | ||
5885 | /* If device is running close it first. */ | |
5886 | dev_close(dev); | |
5887 | ||
5888 | /* And unlink it from device chain */ | |
5889 | err = -ENODEV; | |
5890 | unlist_netdevice(dev); | |
5891 | ||
5892 | synchronize_net(); | |
5893 | ||
5894 | /* Shutdown queueing discipline. */ | |
5895 | dev_shutdown(dev); | |
5896 | ||
5897 | /* Notify protocols, that we are about to destroy | |
5898 | this device. They should clean all the things. | |
5899 | */ | |
5900 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
5901 | call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev); | |
5902 | ||
5903 | /* | |
5904 | * Flush the unicast and multicast chains | |
5905 | */ | |
5906 | dev_unicast_flush(dev); | |
5907 | dev_addr_discard(dev); | |
5908 | ||
5909 | netdev_unregister_kobject(dev); | |
5910 | ||
5911 | /* Actually switch the network namespace */ | |
5912 | dev_net_set(dev, net); | |
5913 | ||
5914 | /* If there is an ifindex conflict assign a new one */ | |
5915 | if (__dev_get_by_index(net, dev->ifindex)) { | |
5916 | int iflink = (dev->iflink == dev->ifindex); | |
5917 | dev->ifindex = dev_new_index(net); | |
5918 | if (iflink) | |
5919 | dev->iflink = dev->ifindex; | |
5920 | } | |
5921 | ||
5922 | /* Fixup kobjects */ | |
5923 | err = netdev_register_kobject(dev); | |
5924 | WARN_ON(err); | |
5925 | ||
5926 | /* Add the device back in the hashes */ | |
5927 | list_netdevice(dev); | |
5928 | ||
5929 | /* Notify protocols, that a new device appeared. */ | |
5930 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
5931 | ||
5932 | /* | |
5933 | * Prevent userspace races by waiting until the network | |
5934 | * device is fully setup before sending notifications. | |
5935 | */ | |
5936 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U); | |
5937 | ||
5938 | synchronize_net(); | |
5939 | err = 0; | |
5940 | out: | |
5941 | return err; | |
5942 | } | |
5943 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); | |
5944 | ||
5945 | static int dev_cpu_callback(struct notifier_block *nfb, | |
5946 | unsigned long action, | |
5947 | void *ocpu) | |
5948 | { | |
5949 | struct sk_buff **list_skb; | |
5950 | struct Qdisc **list_net; | |
5951 | struct sk_buff *skb; | |
5952 | unsigned int cpu, oldcpu = (unsigned long)ocpu; | |
5953 | struct softnet_data *sd, *oldsd; | |
5954 | ||
5955 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) | |
5956 | return NOTIFY_OK; | |
5957 | ||
5958 | local_irq_disable(); | |
5959 | cpu = smp_processor_id(); | |
5960 | sd = &per_cpu(softnet_data, cpu); | |
5961 | oldsd = &per_cpu(softnet_data, oldcpu); | |
5962 | ||
5963 | /* Find end of our completion_queue. */ | |
5964 | list_skb = &sd->completion_queue; | |
5965 | while (*list_skb) | |
5966 | list_skb = &(*list_skb)->next; | |
5967 | /* Append completion queue from offline CPU. */ | |
5968 | *list_skb = oldsd->completion_queue; | |
5969 | oldsd->completion_queue = NULL; | |
5970 | ||
5971 | /* Find end of our output_queue. */ | |
5972 | list_net = &sd->output_queue; | |
5973 | while (*list_net) | |
5974 | list_net = &(*list_net)->next_sched; | |
5975 | /* Append output queue from offline CPU. */ | |
5976 | *list_net = oldsd->output_queue; | |
5977 | oldsd->output_queue = NULL; | |
5978 | ||
5979 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
5980 | local_irq_enable(); | |
5981 | ||
5982 | /* Process offline CPU's input_pkt_queue */ | |
5983 | while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) | |
5984 | netif_rx(skb); | |
5985 | ||
5986 | return NOTIFY_OK; | |
5987 | } | |
5988 | ||
5989 | ||
5990 | /** | |
5991 | * netdev_increment_features - increment feature set by one | |
5992 | * @all: current feature set | |
5993 | * @one: new feature set | |
5994 | * @mask: mask feature set | |
5995 | * | |
5996 | * Computes a new feature set after adding a device with feature set | |
5997 | * @one to the master device with current feature set @all. Will not | |
5998 | * enable anything that is off in @mask. Returns the new feature set. | |
5999 | */ | |
6000 | unsigned long netdev_increment_features(unsigned long all, unsigned long one, | |
6001 | unsigned long mask) | |
6002 | { | |
6003 | /* If device needs checksumming, downgrade to it. */ | |
6004 | if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM)) | |
6005 | all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM); | |
6006 | else if (mask & NETIF_F_ALL_CSUM) { | |
6007 | /* If one device supports v4/v6 checksumming, set for all. */ | |
6008 | if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) && | |
6009 | !(all & NETIF_F_GEN_CSUM)) { | |
6010 | all &= ~NETIF_F_ALL_CSUM; | |
6011 | all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); | |
6012 | } | |
6013 | ||
6014 | /* If one device supports hw checksumming, set for all. */ | |
6015 | if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) { | |
6016 | all &= ~NETIF_F_ALL_CSUM; | |
6017 | all |= NETIF_F_HW_CSUM; | |
6018 | } | |
6019 | } | |
6020 | ||
6021 | one |= NETIF_F_ALL_CSUM; | |
6022 | ||
6023 | one |= all & NETIF_F_ONE_FOR_ALL; | |
6024 | all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO; | |
6025 | all |= one & mask & NETIF_F_ONE_FOR_ALL; | |
6026 | ||
6027 | return all; | |
6028 | } | |
6029 | EXPORT_SYMBOL(netdev_increment_features); | |
6030 | ||
6031 | static struct hlist_head *netdev_create_hash(void) | |
6032 | { | |
6033 | int i; | |
6034 | struct hlist_head *hash; | |
6035 | ||
6036 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
6037 | if (hash != NULL) | |
6038 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
6039 | INIT_HLIST_HEAD(&hash[i]); | |
6040 | ||
6041 | return hash; | |
6042 | } | |
6043 | ||
6044 | /* Initialize per network namespace state */ | |
6045 | static int __net_init netdev_init(struct net *net) | |
6046 | { | |
6047 | INIT_LIST_HEAD(&net->dev_base_head); | |
6048 | ||
6049 | net->dev_name_head = netdev_create_hash(); | |
6050 | if (net->dev_name_head == NULL) | |
6051 | goto err_name; | |
6052 | ||
6053 | net->dev_index_head = netdev_create_hash(); | |
6054 | if (net->dev_index_head == NULL) | |
6055 | goto err_idx; | |
6056 | ||
6057 | return 0; | |
6058 | ||
6059 | err_idx: | |
6060 | kfree(net->dev_name_head); | |
6061 | err_name: | |
6062 | return -ENOMEM; | |
6063 | } | |
6064 | ||
6065 | /** | |
6066 | * netdev_drivername - network driver for the device | |
6067 | * @dev: network device | |
6068 | * @buffer: buffer for resulting name | |
6069 | * @len: size of buffer | |
6070 | * | |
6071 | * Determine network driver for device. | |
6072 | */ | |
6073 | char *netdev_drivername(const struct net_device *dev, char *buffer, int len) | |
6074 | { | |
6075 | const struct device_driver *driver; | |
6076 | const struct device *parent; | |
6077 | ||
6078 | if (len <= 0 || !buffer) | |
6079 | return buffer; | |
6080 | buffer[0] = 0; | |
6081 | ||
6082 | parent = dev->dev.parent; | |
6083 | ||
6084 | if (!parent) | |
6085 | return buffer; | |
6086 | ||
6087 | driver = parent->driver; | |
6088 | if (driver && driver->name) | |
6089 | strlcpy(buffer, driver->name, len); | |
6090 | return buffer; | |
6091 | } | |
6092 | ||
6093 | static void __net_exit netdev_exit(struct net *net) | |
6094 | { | |
6095 | kfree(net->dev_name_head); | |
6096 | kfree(net->dev_index_head); | |
6097 | } | |
6098 | ||
6099 | static struct pernet_operations __net_initdata netdev_net_ops = { | |
6100 | .init = netdev_init, | |
6101 | .exit = netdev_exit, | |
6102 | }; | |
6103 | ||
6104 | static void __net_exit default_device_exit(struct net *net) | |
6105 | { | |
6106 | struct net_device *dev, *aux; | |
6107 | /* | |
6108 | * Push all migratable network devices back to the | |
6109 | * initial network namespace | |
6110 | */ | |
6111 | rtnl_lock(); | |
6112 | for_each_netdev_safe(net, dev, aux) { | |
6113 | int err; | |
6114 | char fb_name[IFNAMSIZ]; | |
6115 | ||
6116 | /* Ignore unmoveable devices (i.e. loopback) */ | |
6117 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
6118 | continue; | |
6119 | ||
6120 | /* Leave virtual devices for the generic cleanup */ | |
6121 | if (dev->rtnl_link_ops) | |
6122 | continue; | |
6123 | ||
6124 | /* Push remaing network devices to init_net */ | |
6125 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); | |
6126 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
6127 | if (err) { | |
6128 | printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n", | |
6129 | __func__, dev->name, err); | |
6130 | BUG(); | |
6131 | } | |
6132 | } | |
6133 | rtnl_unlock(); | |
6134 | } | |
6135 | ||
6136 | static void __net_exit default_device_exit_batch(struct list_head *net_list) | |
6137 | { | |
6138 | /* At exit all network devices most be removed from a network | |
6139 | * namespace. Do this in the reverse order of registeration. | |
6140 | * Do this across as many network namespaces as possible to | |
6141 | * improve batching efficiency. | |
6142 | */ | |
6143 | struct net_device *dev; | |
6144 | struct net *net; | |
6145 | LIST_HEAD(dev_kill_list); | |
6146 | ||
6147 | rtnl_lock(); | |
6148 | list_for_each_entry(net, net_list, exit_list) { | |
6149 | for_each_netdev_reverse(net, dev) { | |
6150 | if (dev->rtnl_link_ops) | |
6151 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); | |
6152 | else | |
6153 | unregister_netdevice_queue(dev, &dev_kill_list); | |
6154 | } | |
6155 | } | |
6156 | unregister_netdevice_many(&dev_kill_list); | |
6157 | rtnl_unlock(); | |
6158 | } | |
6159 | ||
6160 | static struct pernet_operations __net_initdata default_device_ops = { | |
6161 | .exit = default_device_exit, | |
6162 | .exit_batch = default_device_exit_batch, | |
6163 | }; | |
6164 | ||
6165 | /* | |
6166 | * Initialize the DEV module. At boot time this walks the device list and | |
6167 | * unhooks any devices that fail to initialise (normally hardware not | |
6168 | * present) and leaves us with a valid list of present and active devices. | |
6169 | * | |
6170 | */ | |
6171 | ||
6172 | /* | |
6173 | * This is called single threaded during boot, so no need | |
6174 | * to take the rtnl semaphore. | |
6175 | */ | |
6176 | static int __init net_dev_init(void) | |
6177 | { | |
6178 | int i, rc = -ENOMEM; | |
6179 | ||
6180 | BUG_ON(!dev_boot_phase); | |
6181 | ||
6182 | if (dev_proc_init()) | |
6183 | goto out; | |
6184 | ||
6185 | if (netdev_kobject_init()) | |
6186 | goto out; | |
6187 | ||
6188 | INIT_LIST_HEAD(&ptype_all); | |
6189 | for (i = 0; i < PTYPE_HASH_SIZE; i++) | |
6190 | INIT_LIST_HEAD(&ptype_base[i]); | |
6191 | ||
6192 | if (register_pernet_subsys(&netdev_net_ops)) | |
6193 | goto out; | |
6194 | ||
6195 | /* | |
6196 | * Initialise the packet receive queues. | |
6197 | */ | |
6198 | ||
6199 | for_each_possible_cpu(i) { | |
6200 | struct softnet_data *queue; | |
6201 | ||
6202 | queue = &per_cpu(softnet_data, i); | |
6203 | skb_queue_head_init(&queue->input_pkt_queue); | |
6204 | queue->completion_queue = NULL; | |
6205 | INIT_LIST_HEAD(&queue->poll_list); | |
6206 | ||
6207 | queue->csd.func = trigger_softirq; | |
6208 | queue->csd.info = queue; | |
6209 | queue->csd.flags = 0; | |
6210 | ||
6211 | queue->backlog.poll = process_backlog; | |
6212 | queue->backlog.weight = weight_p; | |
6213 | queue->backlog.gro_list = NULL; | |
6214 | queue->backlog.gro_count = 0; | |
6215 | } | |
6216 | ||
6217 | dev_boot_phase = 0; | |
6218 | ||
6219 | /* The loopback device is special if any other network devices | |
6220 | * is present in a network namespace the loopback device must | |
6221 | * be present. Since we now dynamically allocate and free the | |
6222 | * loopback device ensure this invariant is maintained by | |
6223 | * keeping the loopback device as the first device on the | |
6224 | * list of network devices. Ensuring the loopback devices | |
6225 | * is the first device that appears and the last network device | |
6226 | * that disappears. | |
6227 | */ | |
6228 | if (register_pernet_device(&loopback_net_ops)) | |
6229 | goto out; | |
6230 | ||
6231 | if (register_pernet_device(&default_device_ops)) | |
6232 | goto out; | |
6233 | ||
6234 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); | |
6235 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
6236 | ||
6237 | hotcpu_notifier(dev_cpu_callback, 0); | |
6238 | dst_init(); | |
6239 | dev_mcast_init(); | |
6240 | rc = 0; | |
6241 | out: | |
6242 | return rc; | |
6243 | } | |
6244 | ||
6245 | subsys_initcall(net_dev_init); | |
6246 | ||
6247 | static int __init initialize_hashrnd(void) | |
6248 | { | |
6249 | get_random_bytes(&hashrnd, sizeof(hashrnd)); | |
6250 | return 0; | |
6251 | } | |
6252 | ||
6253 | late_initcall_sync(initialize_hashrnd); | |
6254 |