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