Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * NET3 Protocol independent device support routines. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public License | |
6 | * as published by the Free Software Foundation; either version | |
7 | * 2 of the License, or (at your option) any later version. | |
8 | * | |
9 | * Derived from the non IP parts of dev.c 1.0.19 | |
02c30a84 | 10 | * Authors: Ross Biro |
1da177e4 LT |
11 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
12 | * Mark Evans, <evansmp@uhura.aston.ac.uk> | |
13 | * | |
14 | * Additional Authors: | |
15 | * Florian la Roche <rzsfl@rz.uni-sb.de> | |
16 | * Alan Cox <gw4pts@gw4pts.ampr.org> | |
17 | * David Hinds <dahinds@users.sourceforge.net> | |
18 | * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> | |
19 | * Adam Sulmicki <adam@cfar.umd.edu> | |
20 | * Pekka Riikonen <priikone@poesidon.pspt.fi> | |
21 | * | |
22 | * Changes: | |
23 | * D.J. Barrow : Fixed bug where dev->refcnt gets set | |
24 | * to 2 if register_netdev gets called | |
25 | * before net_dev_init & also removed a | |
26 | * few lines of code in the process. | |
27 | * Alan Cox : device private ioctl copies fields back. | |
28 | * Alan Cox : Transmit queue code does relevant | |
29 | * stunts to keep the queue safe. | |
30 | * Alan Cox : Fixed double lock. | |
31 | * Alan Cox : Fixed promisc NULL pointer trap | |
32 | * ???????? : Support the full private ioctl range | |
33 | * Alan Cox : Moved ioctl permission check into | |
34 | * drivers | |
35 | * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI | |
36 | * Alan Cox : 100 backlog just doesn't cut it when | |
37 | * you start doing multicast video 8) | |
38 | * Alan Cox : Rewrote net_bh and list manager. | |
39 | * Alan Cox : Fix ETH_P_ALL echoback lengths. | |
40 | * Alan Cox : Took out transmit every packet pass | |
41 | * Saved a few bytes in the ioctl handler | |
42 | * Alan Cox : Network driver sets packet type before | |
43 | * calling netif_rx. Saves a function | |
44 | * call a packet. | |
45 | * Alan Cox : Hashed net_bh() | |
46 | * Richard Kooijman: Timestamp fixes. | |
47 | * Alan Cox : Wrong field in SIOCGIFDSTADDR | |
48 | * Alan Cox : Device lock protection. | |
49 | * Alan Cox : Fixed nasty side effect of device close | |
50 | * changes. | |
51 | * Rudi Cilibrasi : Pass the right thing to | |
52 | * set_mac_address() | |
53 | * Dave Miller : 32bit quantity for the device lock to | |
54 | * make it work out on a Sparc. | |
55 | * Bjorn Ekwall : Added KERNELD hack. | |
56 | * Alan Cox : Cleaned up the backlog initialise. | |
57 | * Craig Metz : SIOCGIFCONF fix if space for under | |
58 | * 1 device. | |
59 | * Thomas Bogendoerfer : Return ENODEV for dev_open, if there | |
60 | * is no device open function. | |
61 | * Andi Kleen : Fix error reporting for SIOCGIFCONF | |
62 | * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF | |
63 | * Cyrus Durgin : Cleaned for KMOD | |
64 | * Adam Sulmicki : Bug Fix : Network Device Unload | |
65 | * A network device unload needs to purge | |
66 | * the backlog queue. | |
67 | * Paul Rusty Russell : SIOCSIFNAME | |
68 | * Pekka Riikonen : Netdev boot-time settings code | |
69 | * Andrew Morton : Make unregister_netdevice wait | |
70 | * indefinitely on dev->refcnt | |
71 | * J Hadi Salim : - Backlog queue sampling | |
72 | * - netif_rx() feedback | |
73 | */ | |
74 | ||
75 | #include <asm/uaccess.h> | |
76 | #include <asm/system.h> | |
77 | #include <linux/bitops.h> | |
4fc268d2 | 78 | #include <linux/capability.h> |
1da177e4 LT |
79 | #include <linux/cpu.h> |
80 | #include <linux/types.h> | |
81 | #include <linux/kernel.h> | |
82 | #include <linux/sched.h> | |
4a3e2f71 | 83 | #include <linux/mutex.h> |
1da177e4 LT |
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/notifier.h> | |
94 | #include <linux/skbuff.h> | |
457c4cbc | 95 | #include <net/net_namespace.h> |
1da177e4 LT |
96 | #include <net/sock.h> |
97 | #include <linux/rtnetlink.h> | |
98 | #include <linux/proc_fs.h> | |
99 | #include <linux/seq_file.h> | |
100 | #include <linux/stat.h> | |
101 | #include <linux/if_bridge.h> | |
b863ceb7 | 102 | #include <linux/if_macvlan.h> |
1da177e4 LT |
103 | #include <net/dst.h> |
104 | #include <net/pkt_sched.h> | |
105 | #include <net/checksum.h> | |
106 | #include <linux/highmem.h> | |
107 | #include <linux/init.h> | |
108 | #include <linux/kmod.h> | |
109 | #include <linux/module.h> | |
110 | #include <linux/kallsyms.h> | |
111 | #include <linux/netpoll.h> | |
112 | #include <linux/rcupdate.h> | |
113 | #include <linux/delay.h> | |
295f4a1f | 114 | #include <net/wext.h> |
1da177e4 | 115 | #include <net/iw_handler.h> |
1da177e4 | 116 | #include <asm/current.h> |
5bdb9886 | 117 | #include <linux/audit.h> |
db217334 | 118 | #include <linux/dmaengine.h> |
f6a78bfc | 119 | #include <linux/err.h> |
c7fa9d18 | 120 | #include <linux/ctype.h> |
723e98b7 | 121 | #include <linux/if_arp.h> |
1da177e4 | 122 | |
342709ef PE |
123 | #include "net-sysfs.h" |
124 | ||
1da177e4 LT |
125 | /* |
126 | * The list of packet types we will receive (as opposed to discard) | |
127 | * and the routines to invoke. | |
128 | * | |
129 | * Why 16. Because with 16 the only overlap we get on a hash of the | |
130 | * low nibble of the protocol value is RARP/SNAP/X.25. | |
131 | * | |
132 | * NOTE: That is no longer true with the addition of VLAN tags. Not | |
133 | * sure which should go first, but I bet it won't make much | |
134 | * difference if we are running VLANs. The good news is that | |
135 | * this protocol won't be in the list unless compiled in, so | |
3041a069 | 136 | * the average user (w/out VLANs) will not be adversely affected. |
1da177e4 LT |
137 | * --BLG |
138 | * | |
139 | * 0800 IP | |
140 | * 8100 802.1Q VLAN | |
141 | * 0001 802.3 | |
142 | * 0002 AX.25 | |
143 | * 0004 802.2 | |
144 | * 8035 RARP | |
145 | * 0005 SNAP | |
146 | * 0805 X.25 | |
147 | * 0806 ARP | |
148 | * 8137 IPX | |
149 | * 0009 Localtalk | |
150 | * 86DD IPv6 | |
151 | */ | |
152 | ||
153 | static DEFINE_SPINLOCK(ptype_lock); | |
6b2bedc3 SH |
154 | static struct list_head ptype_base[16] __read_mostly; /* 16 way hashed list */ |
155 | static struct list_head ptype_all __read_mostly; /* Taps */ | |
1da177e4 | 156 | |
db217334 | 157 | #ifdef CONFIG_NET_DMA |
d379b01e DW |
158 | struct net_dma { |
159 | struct dma_client client; | |
160 | spinlock_t lock; | |
161 | cpumask_t channel_mask; | |
162 | struct dma_chan *channels[NR_CPUS]; | |
163 | }; | |
164 | ||
165 | static enum dma_state_client | |
166 | netdev_dma_event(struct dma_client *client, struct dma_chan *chan, | |
167 | enum dma_state state); | |
168 | ||
169 | static struct net_dma net_dma = { | |
170 | .client = { | |
171 | .event_callback = netdev_dma_event, | |
172 | }, | |
173 | }; | |
db217334 CL |
174 | #endif |
175 | ||
1da177e4 | 176 | /* |
7562f876 | 177 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
178 | * semaphore. |
179 | * | |
180 | * Pure readers hold dev_base_lock for reading. | |
181 | * | |
182 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 183 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
184 | * actual updates. This allows pure readers to access the list even |
185 | * while a writer is preparing to update it. | |
186 | * | |
187 | * To put it another way, dev_base_lock is held for writing only to | |
188 | * protect against pure readers; the rtnl semaphore provides the | |
189 | * protection against other writers. | |
190 | * | |
191 | * See, for example usages, register_netdevice() and | |
192 | * unregister_netdevice(), which must be called with the rtnl | |
193 | * semaphore held. | |
194 | */ | |
1da177e4 LT |
195 | DEFINE_RWLOCK(dev_base_lock); |
196 | ||
1da177e4 LT |
197 | EXPORT_SYMBOL(dev_base_lock); |
198 | ||
199 | #define NETDEV_HASHBITS 8 | |
881d966b | 200 | #define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS) |
1da177e4 | 201 | |
881d966b | 202 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 LT |
203 | { |
204 | unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ)); | |
881d966b | 205 | return &net->dev_name_head[hash & ((1 << NETDEV_HASHBITS) - 1)]; |
1da177e4 LT |
206 | } |
207 | ||
881d966b | 208 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 209 | { |
881d966b | 210 | return &net->dev_index_head[ifindex & ((1 << NETDEV_HASHBITS) - 1)]; |
1da177e4 LT |
211 | } |
212 | ||
ce286d32 EB |
213 | /* Device list insertion */ |
214 | static int list_netdevice(struct net_device *dev) | |
215 | { | |
216 | struct net *net = dev->nd_net; | |
217 | ||
218 | ASSERT_RTNL(); | |
219 | ||
220 | write_lock_bh(&dev_base_lock); | |
221 | list_add_tail(&dev->dev_list, &net->dev_base_head); | |
222 | hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
223 | hlist_add_head(&dev->index_hlist, dev_index_hash(net, dev->ifindex)); | |
224 | write_unlock_bh(&dev_base_lock); | |
225 | return 0; | |
226 | } | |
227 | ||
228 | /* Device list removal */ | |
229 | static void unlist_netdevice(struct net_device *dev) | |
230 | { | |
231 | ASSERT_RTNL(); | |
232 | ||
233 | /* Unlink dev from the device chain */ | |
234 | write_lock_bh(&dev_base_lock); | |
235 | list_del(&dev->dev_list); | |
236 | hlist_del(&dev->name_hlist); | |
237 | hlist_del(&dev->index_hlist); | |
238 | write_unlock_bh(&dev_base_lock); | |
239 | } | |
240 | ||
1da177e4 LT |
241 | /* |
242 | * Our notifier list | |
243 | */ | |
244 | ||
f07d5b94 | 245 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
246 | |
247 | /* | |
248 | * Device drivers call our routines to queue packets here. We empty the | |
249 | * queue in the local softnet handler. | |
250 | */ | |
bea3348e SH |
251 | |
252 | DEFINE_PER_CPU(struct softnet_data, softnet_data); | |
1da177e4 | 253 | |
723e98b7 JP |
254 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
255 | /* | |
256 | * register_netdevice() inits dev->_xmit_lock and sets lockdep class | |
257 | * according to dev->type | |
258 | */ | |
259 | static const unsigned short netdev_lock_type[] = | |
260 | {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
261 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, | |
262 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
263 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
264 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
265 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
266 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
267 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
268 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
269 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
270 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
271 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
272 | ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211, | |
273 | ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_VOID, | |
274 | ARPHRD_NONE}; | |
275 | ||
276 | static const char *netdev_lock_name[] = | |
277 | {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", | |
278 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
279 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
280 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
281 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
282 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
283 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
284 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
285 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
286 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
287 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
288 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
289 | "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211", | |
290 | "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_VOID", | |
291 | "_xmit_NONE"}; | |
292 | ||
293 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
294 | ||
295 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
296 | { | |
297 | int i; | |
298 | ||
299 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
300 | if (netdev_lock_type[i] == dev_type) | |
301 | return i; | |
302 | /* the last key is used by default */ | |
303 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
304 | } | |
305 | ||
306 | static inline void netdev_set_lockdep_class(spinlock_t *lock, | |
307 | unsigned short dev_type) | |
308 | { | |
309 | int i; | |
310 | ||
311 | i = netdev_lock_pos(dev_type); | |
312 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
313 | netdev_lock_name[i]); | |
314 | } | |
315 | #else | |
316 | static inline void netdev_set_lockdep_class(spinlock_t *lock, | |
317 | unsigned short dev_type) | |
318 | { | |
319 | } | |
320 | #endif | |
1da177e4 LT |
321 | |
322 | /******************************************************************************* | |
323 | ||
324 | Protocol management and registration routines | |
325 | ||
326 | *******************************************************************************/ | |
327 | ||
1da177e4 LT |
328 | /* |
329 | * Add a protocol ID to the list. Now that the input handler is | |
330 | * smarter we can dispense with all the messy stuff that used to be | |
331 | * here. | |
332 | * | |
333 | * BEWARE!!! Protocol handlers, mangling input packets, | |
334 | * MUST BE last in hash buckets and checking protocol handlers | |
335 | * MUST start from promiscuous ptype_all chain in net_bh. | |
336 | * It is true now, do not change it. | |
337 | * Explanation follows: if protocol handler, mangling packet, will | |
338 | * be the first on list, it is not able to sense, that packet | |
339 | * is cloned and should be copied-on-write, so that it will | |
340 | * change it and subsequent readers will get broken packet. | |
341 | * --ANK (980803) | |
342 | */ | |
343 | ||
344 | /** | |
345 | * dev_add_pack - add packet handler | |
346 | * @pt: packet type declaration | |
347 | * | |
348 | * Add a protocol handler to the networking stack. The passed &packet_type | |
349 | * is linked into kernel lists and may not be freed until it has been | |
350 | * removed from the kernel lists. | |
351 | * | |
4ec93edb | 352 | * This call does not sleep therefore it can not |
1da177e4 LT |
353 | * guarantee all CPU's that are in middle of receiving packets |
354 | * will see the new packet type (until the next received packet). | |
355 | */ | |
356 | ||
357 | void dev_add_pack(struct packet_type *pt) | |
358 | { | |
359 | int hash; | |
360 | ||
361 | spin_lock_bh(&ptype_lock); | |
9be9a6b9 | 362 | if (pt->type == htons(ETH_P_ALL)) |
1da177e4 | 363 | list_add_rcu(&pt->list, &ptype_all); |
9be9a6b9 | 364 | else { |
1da177e4 LT |
365 | hash = ntohs(pt->type) & 15; |
366 | list_add_rcu(&pt->list, &ptype_base[hash]); | |
367 | } | |
368 | spin_unlock_bh(&ptype_lock); | |
369 | } | |
370 | ||
1da177e4 LT |
371 | /** |
372 | * __dev_remove_pack - remove packet handler | |
373 | * @pt: packet type declaration | |
374 | * | |
375 | * Remove a protocol handler that was previously added to the kernel | |
376 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
377 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 378 | * returns. |
1da177e4 LT |
379 | * |
380 | * The packet type might still be in use by receivers | |
381 | * and must not be freed until after all the CPU's have gone | |
382 | * through a quiescent state. | |
383 | */ | |
384 | void __dev_remove_pack(struct packet_type *pt) | |
385 | { | |
386 | struct list_head *head; | |
387 | struct packet_type *pt1; | |
388 | ||
389 | spin_lock_bh(&ptype_lock); | |
390 | ||
9be9a6b9 | 391 | if (pt->type == htons(ETH_P_ALL)) |
1da177e4 | 392 | head = &ptype_all; |
9be9a6b9 | 393 | else |
1da177e4 LT |
394 | head = &ptype_base[ntohs(pt->type) & 15]; |
395 | ||
396 | list_for_each_entry(pt1, head, list) { | |
397 | if (pt == pt1) { | |
398 | list_del_rcu(&pt->list); | |
399 | goto out; | |
400 | } | |
401 | } | |
402 | ||
403 | printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt); | |
404 | out: | |
405 | spin_unlock_bh(&ptype_lock); | |
406 | } | |
407 | /** | |
408 | * dev_remove_pack - remove packet handler | |
409 | * @pt: packet type declaration | |
410 | * | |
411 | * Remove a protocol handler that was previously added to the kernel | |
412 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
413 | * from the kernel lists and can be freed or reused once this function | |
414 | * returns. | |
415 | * | |
416 | * This call sleeps to guarantee that no CPU is looking at the packet | |
417 | * type after return. | |
418 | */ | |
419 | void dev_remove_pack(struct packet_type *pt) | |
420 | { | |
421 | __dev_remove_pack(pt); | |
4ec93edb | 422 | |
1da177e4 LT |
423 | synchronize_net(); |
424 | } | |
425 | ||
426 | /****************************************************************************** | |
427 | ||
428 | Device Boot-time Settings Routines | |
429 | ||
430 | *******************************************************************************/ | |
431 | ||
432 | /* Boot time configuration table */ | |
433 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
434 | ||
435 | /** | |
436 | * netdev_boot_setup_add - add new setup entry | |
437 | * @name: name of the device | |
438 | * @map: configured settings for the device | |
439 | * | |
440 | * Adds new setup entry to the dev_boot_setup list. The function | |
441 | * returns 0 on error and 1 on success. This is a generic routine to | |
442 | * all netdevices. | |
443 | */ | |
444 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
445 | { | |
446 | struct netdev_boot_setup *s; | |
447 | int i; | |
448 | ||
449 | s = dev_boot_setup; | |
450 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
451 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
452 | memset(s[i].name, 0, sizeof(s[i].name)); | |
453 | strcpy(s[i].name, name); | |
454 | memcpy(&s[i].map, map, sizeof(s[i].map)); | |
455 | break; | |
456 | } | |
457 | } | |
458 | ||
459 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
460 | } | |
461 | ||
462 | /** | |
463 | * netdev_boot_setup_check - check boot time settings | |
464 | * @dev: the netdevice | |
465 | * | |
466 | * Check boot time settings for the device. | |
467 | * The found settings are set for the device to be used | |
468 | * later in the device probing. | |
469 | * Returns 0 if no settings found, 1 if they are. | |
470 | */ | |
471 | int netdev_boot_setup_check(struct net_device *dev) | |
472 | { | |
473 | struct netdev_boot_setup *s = dev_boot_setup; | |
474 | int i; | |
475 | ||
476 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
477 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
478 | !strncmp(dev->name, s[i].name, strlen(s[i].name))) { | |
479 | dev->irq = s[i].map.irq; | |
480 | dev->base_addr = s[i].map.base_addr; | |
481 | dev->mem_start = s[i].map.mem_start; | |
482 | dev->mem_end = s[i].map.mem_end; | |
483 | return 1; | |
484 | } | |
485 | } | |
486 | return 0; | |
487 | } | |
488 | ||
489 | ||
490 | /** | |
491 | * netdev_boot_base - get address from boot time settings | |
492 | * @prefix: prefix for network device | |
493 | * @unit: id for network device | |
494 | * | |
495 | * Check boot time settings for the base address of device. | |
496 | * The found settings are set for the device to be used | |
497 | * later in the device probing. | |
498 | * Returns 0 if no settings found. | |
499 | */ | |
500 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
501 | { | |
502 | const struct netdev_boot_setup *s = dev_boot_setup; | |
503 | char name[IFNAMSIZ]; | |
504 | int i; | |
505 | ||
506 | sprintf(name, "%s%d", prefix, unit); | |
507 | ||
508 | /* | |
509 | * If device already registered then return base of 1 | |
510 | * to indicate not to probe for this interface | |
511 | */ | |
881d966b | 512 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
513 | return 1; |
514 | ||
515 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
516 | if (!strcmp(name, s[i].name)) | |
517 | return s[i].map.base_addr; | |
518 | return 0; | |
519 | } | |
520 | ||
521 | /* | |
522 | * Saves at boot time configured settings for any netdevice. | |
523 | */ | |
524 | int __init netdev_boot_setup(char *str) | |
525 | { | |
526 | int ints[5]; | |
527 | struct ifmap map; | |
528 | ||
529 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
530 | if (!str || !*str) | |
531 | return 0; | |
532 | ||
533 | /* Save settings */ | |
534 | memset(&map, 0, sizeof(map)); | |
535 | if (ints[0] > 0) | |
536 | map.irq = ints[1]; | |
537 | if (ints[0] > 1) | |
538 | map.base_addr = ints[2]; | |
539 | if (ints[0] > 2) | |
540 | map.mem_start = ints[3]; | |
541 | if (ints[0] > 3) | |
542 | map.mem_end = ints[4]; | |
543 | ||
544 | /* Add new entry to the list */ | |
545 | return netdev_boot_setup_add(str, &map); | |
546 | } | |
547 | ||
548 | __setup("netdev=", netdev_boot_setup); | |
549 | ||
550 | /******************************************************************************* | |
551 | ||
552 | Device Interface Subroutines | |
553 | ||
554 | *******************************************************************************/ | |
555 | ||
556 | /** | |
557 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 558 | * @net: the applicable net namespace |
1da177e4 LT |
559 | * @name: name to find |
560 | * | |
561 | * Find an interface by name. Must be called under RTNL semaphore | |
562 | * or @dev_base_lock. If the name is found a pointer to the device | |
563 | * is returned. If the name is not found then %NULL is returned. The | |
564 | * reference counters are not incremented so the caller must be | |
565 | * careful with locks. | |
566 | */ | |
567 | ||
881d966b | 568 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
569 | { |
570 | struct hlist_node *p; | |
571 | ||
881d966b | 572 | hlist_for_each(p, dev_name_hash(net, name)) { |
1da177e4 LT |
573 | struct net_device *dev |
574 | = hlist_entry(p, struct net_device, name_hlist); | |
575 | if (!strncmp(dev->name, name, IFNAMSIZ)) | |
576 | return dev; | |
577 | } | |
578 | return NULL; | |
579 | } | |
580 | ||
581 | /** | |
582 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 583 | * @net: the applicable net namespace |
1da177e4 LT |
584 | * @name: name to find |
585 | * | |
586 | * Find an interface by name. This can be called from any | |
587 | * context and does its own locking. The returned handle has | |
588 | * the usage count incremented and the caller must use dev_put() to | |
589 | * release it when it is no longer needed. %NULL is returned if no | |
590 | * matching device is found. | |
591 | */ | |
592 | ||
881d966b | 593 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
594 | { |
595 | struct net_device *dev; | |
596 | ||
597 | read_lock(&dev_base_lock); | |
881d966b | 598 | dev = __dev_get_by_name(net, name); |
1da177e4 LT |
599 | if (dev) |
600 | dev_hold(dev); | |
601 | read_unlock(&dev_base_lock); | |
602 | return dev; | |
603 | } | |
604 | ||
605 | /** | |
606 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 607 | * @net: the applicable net namespace |
1da177e4 LT |
608 | * @ifindex: index of device |
609 | * | |
610 | * Search for an interface by index. Returns %NULL if the device | |
611 | * is not found or a pointer to the device. The device has not | |
612 | * had its reference counter increased so the caller must be careful | |
613 | * about locking. The caller must hold either the RTNL semaphore | |
614 | * or @dev_base_lock. | |
615 | */ | |
616 | ||
881d966b | 617 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
618 | { |
619 | struct hlist_node *p; | |
620 | ||
881d966b | 621 | hlist_for_each(p, dev_index_hash(net, ifindex)) { |
1da177e4 LT |
622 | struct net_device *dev |
623 | = hlist_entry(p, struct net_device, index_hlist); | |
624 | if (dev->ifindex == ifindex) | |
625 | return dev; | |
626 | } | |
627 | return NULL; | |
628 | } | |
629 | ||
630 | ||
631 | /** | |
632 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 633 | * @net: the applicable net namespace |
1da177e4 LT |
634 | * @ifindex: index of device |
635 | * | |
636 | * Search for an interface by index. Returns NULL if the device | |
637 | * is not found or a pointer to the device. The device returned has | |
638 | * had a reference added and the pointer is safe until the user calls | |
639 | * dev_put to indicate they have finished with it. | |
640 | */ | |
641 | ||
881d966b | 642 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
643 | { |
644 | struct net_device *dev; | |
645 | ||
646 | read_lock(&dev_base_lock); | |
881d966b | 647 | dev = __dev_get_by_index(net, ifindex); |
1da177e4 LT |
648 | if (dev) |
649 | dev_hold(dev); | |
650 | read_unlock(&dev_base_lock); | |
651 | return dev; | |
652 | } | |
653 | ||
654 | /** | |
655 | * dev_getbyhwaddr - find a device by its hardware address | |
c4ea43c5 | 656 | * @net: the applicable net namespace |
1da177e4 LT |
657 | * @type: media type of device |
658 | * @ha: hardware address | |
659 | * | |
660 | * Search for an interface by MAC address. Returns NULL if the device | |
661 | * is not found or a pointer to the device. The caller must hold the | |
662 | * rtnl semaphore. The returned device has not had its ref count increased | |
663 | * and the caller must therefore be careful about locking | |
664 | * | |
665 | * BUGS: | |
666 | * If the API was consistent this would be __dev_get_by_hwaddr | |
667 | */ | |
668 | ||
881d966b | 669 | struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha) |
1da177e4 LT |
670 | { |
671 | struct net_device *dev; | |
672 | ||
673 | ASSERT_RTNL(); | |
674 | ||
881d966b | 675 | for_each_netdev(&init_net, dev) |
1da177e4 LT |
676 | if (dev->type == type && |
677 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
678 | return dev; |
679 | ||
680 | return NULL; | |
1da177e4 LT |
681 | } |
682 | ||
cf309e3f JF |
683 | EXPORT_SYMBOL(dev_getbyhwaddr); |
684 | ||
881d966b | 685 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
686 | { |
687 | struct net_device *dev; | |
688 | ||
4e9cac2b | 689 | ASSERT_RTNL(); |
881d966b | 690 | for_each_netdev(net, dev) |
4e9cac2b | 691 | if (dev->type == type) |
7562f876 PE |
692 | return dev; |
693 | ||
694 | return NULL; | |
4e9cac2b PM |
695 | } |
696 | ||
697 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); | |
698 | ||
881d966b | 699 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b PM |
700 | { |
701 | struct net_device *dev; | |
702 | ||
703 | rtnl_lock(); | |
881d966b | 704 | dev = __dev_getfirstbyhwtype(net, type); |
4e9cac2b PM |
705 | if (dev) |
706 | dev_hold(dev); | |
1da177e4 LT |
707 | rtnl_unlock(); |
708 | return dev; | |
709 | } | |
710 | ||
711 | EXPORT_SYMBOL(dev_getfirstbyhwtype); | |
712 | ||
713 | /** | |
714 | * dev_get_by_flags - find any device with given flags | |
c4ea43c5 | 715 | * @net: the applicable net namespace |
1da177e4 LT |
716 | * @if_flags: IFF_* values |
717 | * @mask: bitmask of bits in if_flags to check | |
718 | * | |
719 | * Search for any interface with the given flags. Returns NULL if a device | |
4ec93edb | 720 | * is not found or a pointer to the device. The device returned has |
1da177e4 LT |
721 | * had a reference added and the pointer is safe until the user calls |
722 | * dev_put to indicate they have finished with it. | |
723 | */ | |
724 | ||
881d966b | 725 | struct net_device * dev_get_by_flags(struct net *net, unsigned short if_flags, unsigned short mask) |
1da177e4 | 726 | { |
7562f876 | 727 | struct net_device *dev, *ret; |
1da177e4 | 728 | |
7562f876 | 729 | ret = NULL; |
1da177e4 | 730 | read_lock(&dev_base_lock); |
881d966b | 731 | for_each_netdev(net, dev) { |
1da177e4 LT |
732 | if (((dev->flags ^ if_flags) & mask) == 0) { |
733 | dev_hold(dev); | |
7562f876 | 734 | ret = dev; |
1da177e4 LT |
735 | break; |
736 | } | |
737 | } | |
738 | read_unlock(&dev_base_lock); | |
7562f876 | 739 | return ret; |
1da177e4 LT |
740 | } |
741 | ||
742 | /** | |
743 | * dev_valid_name - check if name is okay for network device | |
744 | * @name: name string | |
745 | * | |
746 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
747 | * to allow sysfs to work. We also disallow any kind of |
748 | * whitespace. | |
1da177e4 | 749 | */ |
c2373ee9 | 750 | int dev_valid_name(const char *name) |
1da177e4 | 751 | { |
c7fa9d18 DM |
752 | if (*name == '\0') |
753 | return 0; | |
b6fe17d6 SH |
754 | if (strlen(name) >= IFNAMSIZ) |
755 | return 0; | |
c7fa9d18 DM |
756 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
757 | return 0; | |
758 | ||
759 | while (*name) { | |
760 | if (*name == '/' || isspace(*name)) | |
761 | return 0; | |
762 | name++; | |
763 | } | |
764 | return 1; | |
1da177e4 LT |
765 | } |
766 | ||
767 | /** | |
b267b179 EB |
768 | * __dev_alloc_name - allocate a name for a device |
769 | * @net: network namespace to allocate the device name in | |
1da177e4 | 770 | * @name: name format string |
b267b179 | 771 | * @buf: scratch buffer and result name string |
1da177e4 LT |
772 | * |
773 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
774 | * id. It scans list of devices to build up a free map, then chooses |
775 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
776 | * while allocating the name and adding the device in order to avoid | |
777 | * duplicates. | |
778 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
779 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
780 | */ |
781 | ||
b267b179 | 782 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
783 | { |
784 | int i = 0; | |
1da177e4 LT |
785 | const char *p; |
786 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 787 | unsigned long *inuse; |
1da177e4 LT |
788 | struct net_device *d; |
789 | ||
790 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
791 | if (p) { | |
792 | /* | |
793 | * Verify the string as this thing may have come from | |
794 | * the user. There must be either one "%d" and no other "%" | |
795 | * characters. | |
796 | */ | |
797 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
798 | return -EINVAL; | |
799 | ||
800 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 801 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
802 | if (!inuse) |
803 | return -ENOMEM; | |
804 | ||
881d966b | 805 | for_each_netdev(net, d) { |
1da177e4 LT |
806 | if (!sscanf(d->name, name, &i)) |
807 | continue; | |
808 | if (i < 0 || i >= max_netdevices) | |
809 | continue; | |
810 | ||
811 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 812 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
813 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
814 | set_bit(i, inuse); | |
815 | } | |
816 | ||
817 | i = find_first_zero_bit(inuse, max_netdevices); | |
818 | free_page((unsigned long) inuse); | |
819 | } | |
820 | ||
b267b179 EB |
821 | snprintf(buf, IFNAMSIZ, name, i); |
822 | if (!__dev_get_by_name(net, buf)) | |
1da177e4 | 823 | return i; |
1da177e4 LT |
824 | |
825 | /* It is possible to run out of possible slots | |
826 | * when the name is long and there isn't enough space left | |
827 | * for the digits, or if all bits are used. | |
828 | */ | |
829 | return -ENFILE; | |
830 | } | |
831 | ||
b267b179 EB |
832 | /** |
833 | * dev_alloc_name - allocate a name for a device | |
834 | * @dev: device | |
835 | * @name: name format string | |
836 | * | |
837 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
838 | * id. It scans list of devices to build up a free map, then chooses | |
839 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
840 | * while allocating the name and adding the device in order to avoid | |
841 | * duplicates. | |
842 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
843 | * Returns the number of the unit assigned or a negative errno code. | |
844 | */ | |
845 | ||
846 | int dev_alloc_name(struct net_device *dev, const char *name) | |
847 | { | |
848 | char buf[IFNAMSIZ]; | |
849 | struct net *net; | |
850 | int ret; | |
851 | ||
852 | BUG_ON(!dev->nd_net); | |
853 | net = dev->nd_net; | |
854 | ret = __dev_alloc_name(net, name, buf); | |
855 | if (ret >= 0) | |
856 | strlcpy(dev->name, buf, IFNAMSIZ); | |
857 | return ret; | |
858 | } | |
859 | ||
1da177e4 LT |
860 | |
861 | /** | |
862 | * dev_change_name - change name of a device | |
863 | * @dev: device | |
864 | * @newname: name (or format string) must be at least IFNAMSIZ | |
865 | * | |
866 | * Change name of a device, can pass format strings "eth%d". | |
867 | * for wildcarding. | |
868 | */ | |
869 | int dev_change_name(struct net_device *dev, char *newname) | |
870 | { | |
fcc5a03a | 871 | char oldname[IFNAMSIZ]; |
1da177e4 | 872 | int err = 0; |
fcc5a03a | 873 | int ret; |
881d966b | 874 | struct net *net; |
1da177e4 LT |
875 | |
876 | ASSERT_RTNL(); | |
881d966b | 877 | BUG_ON(!dev->nd_net); |
1da177e4 | 878 | |
881d966b | 879 | net = dev->nd_net; |
1da177e4 LT |
880 | if (dev->flags & IFF_UP) |
881 | return -EBUSY; | |
882 | ||
883 | if (!dev_valid_name(newname)) | |
884 | return -EINVAL; | |
885 | ||
c8d90dca SH |
886 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) |
887 | return 0; | |
888 | ||
fcc5a03a HX |
889 | memcpy(oldname, dev->name, IFNAMSIZ); |
890 | ||
1da177e4 LT |
891 | if (strchr(newname, '%')) { |
892 | err = dev_alloc_name(dev, newname); | |
893 | if (err < 0) | |
894 | return err; | |
895 | strcpy(newname, dev->name); | |
896 | } | |
881d966b | 897 | else if (__dev_get_by_name(net, newname)) |
1da177e4 LT |
898 | return -EEXIST; |
899 | else | |
900 | strlcpy(dev->name, newname, IFNAMSIZ); | |
901 | ||
fcc5a03a | 902 | rollback: |
92749821 | 903 | device_rename(&dev->dev, dev->name); |
7f988eab HX |
904 | |
905 | write_lock_bh(&dev_base_lock); | |
92749821 | 906 | hlist_del(&dev->name_hlist); |
881d966b | 907 | hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name)); |
7f988eab HX |
908 | write_unlock_bh(&dev_base_lock); |
909 | ||
056925ab | 910 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
911 | ret = notifier_to_errno(ret); |
912 | ||
913 | if (ret) { | |
914 | if (err) { | |
915 | printk(KERN_ERR | |
916 | "%s: name change rollback failed: %d.\n", | |
917 | dev->name, ret); | |
918 | } else { | |
919 | err = ret; | |
920 | memcpy(dev->name, oldname, IFNAMSIZ); | |
921 | goto rollback; | |
922 | } | |
923 | } | |
1da177e4 LT |
924 | |
925 | return err; | |
926 | } | |
927 | ||
d8a33ac4 | 928 | /** |
3041a069 | 929 | * netdev_features_change - device changes features |
d8a33ac4 SH |
930 | * @dev: device to cause notification |
931 | * | |
932 | * Called to indicate a device has changed features. | |
933 | */ | |
934 | void netdev_features_change(struct net_device *dev) | |
935 | { | |
056925ab | 936 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
937 | } |
938 | EXPORT_SYMBOL(netdev_features_change); | |
939 | ||
1da177e4 LT |
940 | /** |
941 | * netdev_state_change - device changes state | |
942 | * @dev: device to cause notification | |
943 | * | |
944 | * Called to indicate a device has changed state. This function calls | |
945 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
946 | * to the routing socket. | |
947 | */ | |
948 | void netdev_state_change(struct net_device *dev) | |
949 | { | |
950 | if (dev->flags & IFF_UP) { | |
056925ab | 951 | call_netdevice_notifiers(NETDEV_CHANGE, dev); |
1da177e4 LT |
952 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0); |
953 | } | |
954 | } | |
955 | ||
956 | /** | |
957 | * dev_load - load a network module | |
c4ea43c5 | 958 | * @net: the applicable net namespace |
1da177e4 LT |
959 | * @name: name of interface |
960 | * | |
961 | * If a network interface is not present and the process has suitable | |
962 | * privileges this function loads the module. If module loading is not | |
963 | * available in this kernel then it becomes a nop. | |
964 | */ | |
965 | ||
881d966b | 966 | void dev_load(struct net *net, const char *name) |
1da177e4 | 967 | { |
4ec93edb | 968 | struct net_device *dev; |
1da177e4 LT |
969 | |
970 | read_lock(&dev_base_lock); | |
881d966b | 971 | dev = __dev_get_by_name(net, name); |
1da177e4 LT |
972 | read_unlock(&dev_base_lock); |
973 | ||
974 | if (!dev && capable(CAP_SYS_MODULE)) | |
975 | request_module("%s", name); | |
976 | } | |
977 | ||
1da177e4 LT |
978 | /** |
979 | * dev_open - prepare an interface for use. | |
980 | * @dev: device to open | |
981 | * | |
982 | * Takes a device from down to up state. The device's private open | |
983 | * function is invoked and then the multicast lists are loaded. Finally | |
984 | * the device is moved into the up state and a %NETDEV_UP message is | |
985 | * sent to the netdev notifier chain. | |
986 | * | |
987 | * Calling this function on an active interface is a nop. On a failure | |
988 | * a negative errno code is returned. | |
989 | */ | |
990 | int dev_open(struct net_device *dev) | |
991 | { | |
992 | int ret = 0; | |
993 | ||
994 | /* | |
995 | * Is it already up? | |
996 | */ | |
997 | ||
998 | if (dev->flags & IFF_UP) | |
999 | return 0; | |
1000 | ||
1001 | /* | |
1002 | * Is it even present? | |
1003 | */ | |
1004 | if (!netif_device_present(dev)) | |
1005 | return -ENODEV; | |
1006 | ||
1007 | /* | |
1008 | * Call device private open method | |
1009 | */ | |
1010 | set_bit(__LINK_STATE_START, &dev->state); | |
bada339b JG |
1011 | |
1012 | if (dev->validate_addr) | |
1013 | ret = dev->validate_addr(dev); | |
1014 | ||
1015 | if (!ret && dev->open) | |
1da177e4 | 1016 | ret = dev->open(dev); |
1da177e4 | 1017 | |
4ec93edb | 1018 | /* |
1da177e4 LT |
1019 | * If it went open OK then: |
1020 | */ | |
1021 | ||
bada339b JG |
1022 | if (ret) |
1023 | clear_bit(__LINK_STATE_START, &dev->state); | |
1024 | else { | |
1da177e4 LT |
1025 | /* |
1026 | * Set the flags. | |
1027 | */ | |
1028 | dev->flags |= IFF_UP; | |
1029 | ||
1030 | /* | |
1031 | * Initialize multicasting status | |
1032 | */ | |
4417da66 | 1033 | dev_set_rx_mode(dev); |
1da177e4 LT |
1034 | |
1035 | /* | |
1036 | * Wakeup transmit queue engine | |
1037 | */ | |
1038 | dev_activate(dev); | |
1039 | ||
1040 | /* | |
1041 | * ... and announce new interface. | |
1042 | */ | |
056925ab | 1043 | call_netdevice_notifiers(NETDEV_UP, dev); |
1da177e4 | 1044 | } |
bada339b | 1045 | |
1da177e4 LT |
1046 | return ret; |
1047 | } | |
1048 | ||
1049 | /** | |
1050 | * dev_close - shutdown an interface. | |
1051 | * @dev: device to shutdown | |
1052 | * | |
1053 | * This function moves an active device into down state. A | |
1054 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1055 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1056 | * chain. | |
1057 | */ | |
1058 | int dev_close(struct net_device *dev) | |
1059 | { | |
9d5010db DM |
1060 | might_sleep(); |
1061 | ||
1da177e4 LT |
1062 | if (!(dev->flags & IFF_UP)) |
1063 | return 0; | |
1064 | ||
1065 | /* | |
1066 | * Tell people we are going down, so that they can | |
1067 | * prepare to death, when device is still operating. | |
1068 | */ | |
056925ab | 1069 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 LT |
1070 | |
1071 | dev_deactivate(dev); | |
1072 | ||
1073 | clear_bit(__LINK_STATE_START, &dev->state); | |
1074 | ||
1075 | /* Synchronize to scheduled poll. We cannot touch poll list, | |
bea3348e SH |
1076 | * it can be even on different cpu. So just clear netif_running(). |
1077 | * | |
1078 | * dev->stop() will invoke napi_disable() on all of it's | |
1079 | * napi_struct instances on this device. | |
1080 | */ | |
1da177e4 | 1081 | smp_mb__after_clear_bit(); /* Commit netif_running(). */ |
1da177e4 LT |
1082 | |
1083 | /* | |
1084 | * Call the device specific close. This cannot fail. | |
1085 | * Only if device is UP | |
1086 | * | |
1087 | * We allow it to be called even after a DETACH hot-plug | |
1088 | * event. | |
1089 | */ | |
1090 | if (dev->stop) | |
1091 | dev->stop(dev); | |
1092 | ||
1093 | /* | |
1094 | * Device is now down. | |
1095 | */ | |
1096 | ||
1097 | dev->flags &= ~IFF_UP; | |
1098 | ||
1099 | /* | |
1100 | * Tell people we are down | |
1101 | */ | |
056925ab | 1102 | call_netdevice_notifiers(NETDEV_DOWN, dev); |
1da177e4 LT |
1103 | |
1104 | return 0; | |
1105 | } | |
1106 | ||
1107 | ||
881d966b EB |
1108 | static int dev_boot_phase = 1; |
1109 | ||
1da177e4 LT |
1110 | /* |
1111 | * Device change register/unregister. These are not inline or static | |
1112 | * as we export them to the world. | |
1113 | */ | |
1114 | ||
1115 | /** | |
1116 | * register_netdevice_notifier - register a network notifier block | |
1117 | * @nb: notifier | |
1118 | * | |
1119 | * Register a notifier to be called when network device events occur. | |
1120 | * The notifier passed is linked into the kernel structures and must | |
1121 | * not be reused until it has been unregistered. A negative errno code | |
1122 | * is returned on a failure. | |
1123 | * | |
1124 | * When registered all registration and up events are replayed | |
4ec93edb | 1125 | * to the new notifier to allow device to have a race free |
1da177e4 LT |
1126 | * view of the network device list. |
1127 | */ | |
1128 | ||
1129 | int register_netdevice_notifier(struct notifier_block *nb) | |
1130 | { | |
1131 | struct net_device *dev; | |
fcc5a03a | 1132 | struct net_device *last; |
881d966b | 1133 | struct net *net; |
1da177e4 LT |
1134 | int err; |
1135 | ||
1136 | rtnl_lock(); | |
f07d5b94 | 1137 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1138 | if (err) |
1139 | goto unlock; | |
881d966b EB |
1140 | if (dev_boot_phase) |
1141 | goto unlock; | |
1142 | for_each_net(net) { | |
1143 | for_each_netdev(net, dev) { | |
1144 | err = nb->notifier_call(nb, NETDEV_REGISTER, dev); | |
1145 | err = notifier_to_errno(err); | |
1146 | if (err) | |
1147 | goto rollback; | |
1148 | ||
1149 | if (!(dev->flags & IFF_UP)) | |
1150 | continue; | |
1da177e4 | 1151 | |
881d966b EB |
1152 | nb->notifier_call(nb, NETDEV_UP, dev); |
1153 | } | |
1da177e4 | 1154 | } |
fcc5a03a HX |
1155 | |
1156 | unlock: | |
1da177e4 LT |
1157 | rtnl_unlock(); |
1158 | return err; | |
fcc5a03a HX |
1159 | |
1160 | rollback: | |
1161 | last = dev; | |
881d966b EB |
1162 | for_each_net(net) { |
1163 | for_each_netdev(net, dev) { | |
1164 | if (dev == last) | |
1165 | break; | |
fcc5a03a | 1166 | |
881d966b EB |
1167 | if (dev->flags & IFF_UP) { |
1168 | nb->notifier_call(nb, NETDEV_GOING_DOWN, dev); | |
1169 | nb->notifier_call(nb, NETDEV_DOWN, dev); | |
1170 | } | |
1171 | nb->notifier_call(nb, NETDEV_UNREGISTER, dev); | |
fcc5a03a | 1172 | } |
fcc5a03a HX |
1173 | } |
1174 | goto unlock; | |
1da177e4 LT |
1175 | } |
1176 | ||
1177 | /** | |
1178 | * unregister_netdevice_notifier - unregister a network notifier block | |
1179 | * @nb: notifier | |
1180 | * | |
1181 | * Unregister a notifier previously registered by | |
1182 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1183 | * kernel structures and may then be reused. A negative errno code | |
1184 | * is returned on a failure. | |
1185 | */ | |
1186 | ||
1187 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1188 | { | |
9f514950 HX |
1189 | int err; |
1190 | ||
1191 | rtnl_lock(); | |
f07d5b94 | 1192 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
9f514950 HX |
1193 | rtnl_unlock(); |
1194 | return err; | |
1da177e4 LT |
1195 | } |
1196 | ||
1197 | /** | |
1198 | * call_netdevice_notifiers - call all network notifier blocks | |
1199 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1200 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1201 | * |
1202 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1203 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1204 | */ |
1205 | ||
ad7379d4 | 1206 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1207 | { |
ad7379d4 | 1208 | return raw_notifier_call_chain(&netdev_chain, val, dev); |
1da177e4 LT |
1209 | } |
1210 | ||
1211 | /* When > 0 there are consumers of rx skb time stamps */ | |
1212 | static atomic_t netstamp_needed = ATOMIC_INIT(0); | |
1213 | ||
1214 | void net_enable_timestamp(void) | |
1215 | { | |
1216 | atomic_inc(&netstamp_needed); | |
1217 | } | |
1218 | ||
1219 | void net_disable_timestamp(void) | |
1220 | { | |
1221 | atomic_dec(&netstamp_needed); | |
1222 | } | |
1223 | ||
a61bbcf2 | 1224 | static inline void net_timestamp(struct sk_buff *skb) |
1da177e4 LT |
1225 | { |
1226 | if (atomic_read(&netstamp_needed)) | |
a61bbcf2 | 1227 | __net_timestamp(skb); |
b7aa0bf7 ED |
1228 | else |
1229 | skb->tstamp.tv64 = 0; | |
1da177e4 LT |
1230 | } |
1231 | ||
1232 | /* | |
1233 | * Support routine. Sends outgoing frames to any network | |
1234 | * taps currently in use. | |
1235 | */ | |
1236 | ||
f6a78bfc | 1237 | static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
1238 | { |
1239 | struct packet_type *ptype; | |
a61bbcf2 PM |
1240 | |
1241 | net_timestamp(skb); | |
1da177e4 LT |
1242 | |
1243 | rcu_read_lock(); | |
1244 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
1245 | /* Never send packets back to the socket | |
1246 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1247 | */ | |
1248 | if ((ptype->dev == dev || !ptype->dev) && | |
1249 | (ptype->af_packet_priv == NULL || | |
1250 | (struct sock *)ptype->af_packet_priv != skb->sk)) { | |
1251 | struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC); | |
1252 | if (!skb2) | |
1253 | break; | |
1254 | ||
1255 | /* skb->nh should be correctly | |
1256 | set by sender, so that the second statement is | |
1257 | just protection against buggy protocols. | |
1258 | */ | |
459a98ed | 1259 | skb_reset_mac_header(skb2); |
1da177e4 | 1260 | |
d56f90a7 | 1261 | if (skb_network_header(skb2) < skb2->data || |
27a884dc | 1262 | skb2->network_header > skb2->tail) { |
1da177e4 LT |
1263 | if (net_ratelimit()) |
1264 | printk(KERN_CRIT "protocol %04x is " | |
1265 | "buggy, dev %s\n", | |
1266 | skb2->protocol, dev->name); | |
c1d2bbe1 | 1267 | skb_reset_network_header(skb2); |
1da177e4 LT |
1268 | } |
1269 | ||
b0e380b1 | 1270 | skb2->transport_header = skb2->network_header; |
1da177e4 | 1271 | skb2->pkt_type = PACKET_OUTGOING; |
f2ccd8fa | 1272 | ptype->func(skb2, skb->dev, ptype, skb->dev); |
1da177e4 LT |
1273 | } |
1274 | } | |
1275 | rcu_read_unlock(); | |
1276 | } | |
1277 | ||
56079431 DV |
1278 | |
1279 | void __netif_schedule(struct net_device *dev) | |
1280 | { | |
1281 | if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) { | |
1282 | unsigned long flags; | |
1283 | struct softnet_data *sd; | |
1284 | ||
1285 | local_irq_save(flags); | |
1286 | sd = &__get_cpu_var(softnet_data); | |
1287 | dev->next_sched = sd->output_queue; | |
1288 | sd->output_queue = dev; | |
1289 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
1290 | local_irq_restore(flags); | |
1291 | } | |
1292 | } | |
1293 | EXPORT_SYMBOL(__netif_schedule); | |
1294 | ||
bea3348e | 1295 | void dev_kfree_skb_irq(struct sk_buff *skb) |
56079431 | 1296 | { |
bea3348e SH |
1297 | if (atomic_dec_and_test(&skb->users)) { |
1298 | struct softnet_data *sd; | |
1299 | unsigned long flags; | |
56079431 | 1300 | |
bea3348e SH |
1301 | local_irq_save(flags); |
1302 | sd = &__get_cpu_var(softnet_data); | |
1303 | skb->next = sd->completion_queue; | |
1304 | sd->completion_queue = skb; | |
1305 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
1306 | local_irq_restore(flags); | |
1307 | } | |
56079431 | 1308 | } |
bea3348e | 1309 | EXPORT_SYMBOL(dev_kfree_skb_irq); |
56079431 DV |
1310 | |
1311 | void dev_kfree_skb_any(struct sk_buff *skb) | |
1312 | { | |
1313 | if (in_irq() || irqs_disabled()) | |
1314 | dev_kfree_skb_irq(skb); | |
1315 | else | |
1316 | dev_kfree_skb(skb); | |
1317 | } | |
1318 | EXPORT_SYMBOL(dev_kfree_skb_any); | |
1319 | ||
1320 | ||
bea3348e SH |
1321 | /** |
1322 | * netif_device_detach - mark device as removed | |
1323 | * @dev: network device | |
1324 | * | |
1325 | * Mark device as removed from system and therefore no longer available. | |
1326 | */ | |
56079431 DV |
1327 | void netif_device_detach(struct net_device *dev) |
1328 | { | |
1329 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
1330 | netif_running(dev)) { | |
1331 | netif_stop_queue(dev); | |
1332 | } | |
1333 | } | |
1334 | EXPORT_SYMBOL(netif_device_detach); | |
1335 | ||
bea3348e SH |
1336 | /** |
1337 | * netif_device_attach - mark device as attached | |
1338 | * @dev: network device | |
1339 | * | |
1340 | * Mark device as attached from system and restart if needed. | |
1341 | */ | |
56079431 DV |
1342 | void netif_device_attach(struct net_device *dev) |
1343 | { | |
1344 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
1345 | netif_running(dev)) { | |
1346 | netif_wake_queue(dev); | |
4ec93edb | 1347 | __netdev_watchdog_up(dev); |
56079431 DV |
1348 | } |
1349 | } | |
1350 | EXPORT_SYMBOL(netif_device_attach); | |
1351 | ||
1352 | ||
1da177e4 LT |
1353 | /* |
1354 | * Invalidate hardware checksum when packet is to be mangled, and | |
1355 | * complete checksum manually on outgoing path. | |
1356 | */ | |
84fa7933 | 1357 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 1358 | { |
d3bc23e7 | 1359 | __wsum csum; |
663ead3b | 1360 | int ret = 0, offset; |
1da177e4 | 1361 | |
84fa7933 | 1362 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
1363 | goto out_set_summed; |
1364 | ||
1365 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
a430a43d HX |
1366 | /* Let GSO fix up the checksum. */ |
1367 | goto out_set_summed; | |
1da177e4 LT |
1368 | } |
1369 | ||
a030847e HX |
1370 | offset = skb->csum_start - skb_headroom(skb); |
1371 | BUG_ON(offset >= skb_headlen(skb)); | |
1372 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
1373 | ||
1374 | offset += skb->csum_offset; | |
1375 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
1376 | ||
1377 | if (skb_cloned(skb) && | |
1378 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1da177e4 LT |
1379 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
1380 | if (ret) | |
1381 | goto out; | |
1382 | } | |
1383 | ||
a030847e | 1384 | *(__sum16 *)(skb->data + offset) = csum_fold(csum); |
a430a43d | 1385 | out_set_summed: |
1da177e4 | 1386 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 1387 | out: |
1da177e4 LT |
1388 | return ret; |
1389 | } | |
1390 | ||
f6a78bfc HX |
1391 | /** |
1392 | * skb_gso_segment - Perform segmentation on skb. | |
1393 | * @skb: buffer to segment | |
576a30eb | 1394 | * @features: features for the output path (see dev->features) |
f6a78bfc HX |
1395 | * |
1396 | * This function segments the given skb and returns a list of segments. | |
576a30eb HX |
1397 | * |
1398 | * It may return NULL if the skb requires no segmentation. This is | |
1399 | * only possible when GSO is used for verifying header integrity. | |
f6a78bfc | 1400 | */ |
576a30eb | 1401 | struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features) |
f6a78bfc HX |
1402 | { |
1403 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
1404 | struct packet_type *ptype; | |
252e3346 | 1405 | __be16 type = skb->protocol; |
a430a43d | 1406 | int err; |
f6a78bfc HX |
1407 | |
1408 | BUG_ON(skb_shinfo(skb)->frag_list); | |
f6a78bfc | 1409 | |
459a98ed | 1410 | skb_reset_mac_header(skb); |
b0e380b1 | 1411 | skb->mac_len = skb->network_header - skb->mac_header; |
f6a78bfc HX |
1412 | __skb_pull(skb, skb->mac_len); |
1413 | ||
f9d106a6 | 1414 | if (WARN_ON(skb->ip_summed != CHECKSUM_PARTIAL)) { |
a430a43d HX |
1415 | if (skb_header_cloned(skb) && |
1416 | (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))) | |
1417 | return ERR_PTR(err); | |
1418 | } | |
1419 | ||
f6a78bfc HX |
1420 | rcu_read_lock(); |
1421 | list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type) & 15], list) { | |
1422 | if (ptype->type == type && !ptype->dev && ptype->gso_segment) { | |
84fa7933 | 1423 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { |
a430a43d HX |
1424 | err = ptype->gso_send_check(skb); |
1425 | segs = ERR_PTR(err); | |
1426 | if (err || skb_gso_ok(skb, features)) | |
1427 | break; | |
d56f90a7 ACM |
1428 | __skb_push(skb, (skb->data - |
1429 | skb_network_header(skb))); | |
a430a43d | 1430 | } |
576a30eb | 1431 | segs = ptype->gso_segment(skb, features); |
f6a78bfc HX |
1432 | break; |
1433 | } | |
1434 | } | |
1435 | rcu_read_unlock(); | |
1436 | ||
98e399f8 | 1437 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 1438 | |
f6a78bfc HX |
1439 | return segs; |
1440 | } | |
1441 | ||
1442 | EXPORT_SYMBOL(skb_gso_segment); | |
1443 | ||
fb286bb2 HX |
1444 | /* Take action when hardware reception checksum errors are detected. */ |
1445 | #ifdef CONFIG_BUG | |
1446 | void netdev_rx_csum_fault(struct net_device *dev) | |
1447 | { | |
1448 | if (net_ratelimit()) { | |
4ec93edb | 1449 | printk(KERN_ERR "%s: hw csum failure.\n", |
246a4212 | 1450 | dev ? dev->name : "<unknown>"); |
fb286bb2 HX |
1451 | dump_stack(); |
1452 | } | |
1453 | } | |
1454 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
1455 | #endif | |
1456 | ||
1da177e4 LT |
1457 | /* Actually, we should eliminate this check as soon as we know, that: |
1458 | * 1. IOMMU is present and allows to map all the memory. | |
1459 | * 2. No high memory really exists on this machine. | |
1460 | */ | |
1461 | ||
1462 | static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb) | |
1463 | { | |
3d3a8533 | 1464 | #ifdef CONFIG_HIGHMEM |
1da177e4 LT |
1465 | int i; |
1466 | ||
1467 | if (dev->features & NETIF_F_HIGHDMA) | |
1468 | return 0; | |
1469 | ||
1470 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) | |
1471 | if (PageHighMem(skb_shinfo(skb)->frags[i].page)) | |
1472 | return 1; | |
1473 | ||
3d3a8533 | 1474 | #endif |
1da177e4 LT |
1475 | return 0; |
1476 | } | |
1da177e4 | 1477 | |
f6a78bfc HX |
1478 | struct dev_gso_cb { |
1479 | void (*destructor)(struct sk_buff *skb); | |
1480 | }; | |
1481 | ||
1482 | #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb) | |
1483 | ||
1484 | static void dev_gso_skb_destructor(struct sk_buff *skb) | |
1485 | { | |
1486 | struct dev_gso_cb *cb; | |
1487 | ||
1488 | do { | |
1489 | struct sk_buff *nskb = skb->next; | |
1490 | ||
1491 | skb->next = nskb->next; | |
1492 | nskb->next = NULL; | |
1493 | kfree_skb(nskb); | |
1494 | } while (skb->next); | |
1495 | ||
1496 | cb = DEV_GSO_CB(skb); | |
1497 | if (cb->destructor) | |
1498 | cb->destructor(skb); | |
1499 | } | |
1500 | ||
1501 | /** | |
1502 | * dev_gso_segment - Perform emulated hardware segmentation on skb. | |
1503 | * @skb: buffer to segment | |
1504 | * | |
1505 | * This function segments the given skb and stores the list of segments | |
1506 | * in skb->next. | |
1507 | */ | |
1508 | static int dev_gso_segment(struct sk_buff *skb) | |
1509 | { | |
1510 | struct net_device *dev = skb->dev; | |
1511 | struct sk_buff *segs; | |
576a30eb HX |
1512 | int features = dev->features & ~(illegal_highdma(dev, skb) ? |
1513 | NETIF_F_SG : 0); | |
1514 | ||
1515 | segs = skb_gso_segment(skb, features); | |
1516 | ||
1517 | /* Verifying header integrity only. */ | |
1518 | if (!segs) | |
1519 | return 0; | |
f6a78bfc | 1520 | |
f6a78bfc HX |
1521 | if (unlikely(IS_ERR(segs))) |
1522 | return PTR_ERR(segs); | |
1523 | ||
1524 | skb->next = segs; | |
1525 | DEV_GSO_CB(skb)->destructor = skb->destructor; | |
1526 | skb->destructor = dev_gso_skb_destructor; | |
1527 | ||
1528 | return 0; | |
1529 | } | |
1530 | ||
1531 | int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev) | |
1532 | { | |
1533 | if (likely(!skb->next)) { | |
9be9a6b9 | 1534 | if (!list_empty(&ptype_all)) |
f6a78bfc HX |
1535 | dev_queue_xmit_nit(skb, dev); |
1536 | ||
576a30eb HX |
1537 | if (netif_needs_gso(dev, skb)) { |
1538 | if (unlikely(dev_gso_segment(skb))) | |
1539 | goto out_kfree_skb; | |
1540 | if (skb->next) | |
1541 | goto gso; | |
1542 | } | |
f6a78bfc | 1543 | |
576a30eb | 1544 | return dev->hard_start_xmit(skb, dev); |
f6a78bfc HX |
1545 | } |
1546 | ||
576a30eb | 1547 | gso: |
f6a78bfc HX |
1548 | do { |
1549 | struct sk_buff *nskb = skb->next; | |
1550 | int rc; | |
1551 | ||
1552 | skb->next = nskb->next; | |
1553 | nskb->next = NULL; | |
1554 | rc = dev->hard_start_xmit(nskb, dev); | |
1555 | if (unlikely(rc)) { | |
f54d9e8d | 1556 | nskb->next = skb->next; |
f6a78bfc HX |
1557 | skb->next = nskb; |
1558 | return rc; | |
1559 | } | |
f25f4e44 | 1560 | if (unlikely((netif_queue_stopped(dev) || |
668f895a | 1561 | netif_subqueue_stopped(dev, skb)) && |
f25f4e44 | 1562 | skb->next)) |
f54d9e8d | 1563 | return NETDEV_TX_BUSY; |
f6a78bfc | 1564 | } while (skb->next); |
4ec93edb | 1565 | |
f6a78bfc HX |
1566 | skb->destructor = DEV_GSO_CB(skb)->destructor; |
1567 | ||
1568 | out_kfree_skb: | |
1569 | kfree_skb(skb); | |
1570 | return 0; | |
1571 | } | |
1572 | ||
1da177e4 LT |
1573 | /** |
1574 | * dev_queue_xmit - transmit a buffer | |
1575 | * @skb: buffer to transmit | |
1576 | * | |
1577 | * Queue a buffer for transmission to a network device. The caller must | |
1578 | * have set the device and priority and built the buffer before calling | |
1579 | * this function. The function can be called from an interrupt. | |
1580 | * | |
1581 | * A negative errno code is returned on a failure. A success does not | |
1582 | * guarantee the frame will be transmitted as it may be dropped due | |
1583 | * to congestion or traffic shaping. | |
af191367 BG |
1584 | * |
1585 | * ----------------------------------------------------------------------------------- | |
1586 | * I notice this method can also return errors from the queue disciplines, | |
1587 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
1588 | * be positive. | |
1589 | * | |
1590 | * Regardless of the return value, the skb is consumed, so it is currently | |
1591 | * difficult to retry a send to this method. (You can bump the ref count | |
1592 | * before sending to hold a reference for retry if you are careful.) | |
1593 | * | |
1594 | * When calling this method, interrupts MUST be enabled. This is because | |
1595 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
1596 | * --BLG | |
1da177e4 LT |
1597 | */ |
1598 | ||
1599 | int dev_queue_xmit(struct sk_buff *skb) | |
1600 | { | |
1601 | struct net_device *dev = skb->dev; | |
1602 | struct Qdisc *q; | |
1603 | int rc = -ENOMEM; | |
1604 | ||
f6a78bfc HX |
1605 | /* GSO will handle the following emulations directly. */ |
1606 | if (netif_needs_gso(dev, skb)) | |
1607 | goto gso; | |
1608 | ||
1da177e4 LT |
1609 | if (skb_shinfo(skb)->frag_list && |
1610 | !(dev->features & NETIF_F_FRAGLIST) && | |
364c6bad | 1611 | __skb_linearize(skb)) |
1da177e4 LT |
1612 | goto out_kfree_skb; |
1613 | ||
1614 | /* Fragmented skb is linearized if device does not support SG, | |
1615 | * or if at least one of fragments is in highmem and device | |
1616 | * does not support DMA from it. | |
1617 | */ | |
1618 | if (skb_shinfo(skb)->nr_frags && | |
1619 | (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) && | |
364c6bad | 1620 | __skb_linearize(skb)) |
1da177e4 LT |
1621 | goto out_kfree_skb; |
1622 | ||
1623 | /* If packet is not checksummed and device does not support | |
1624 | * checksumming for this protocol, complete checksumming here. | |
1625 | */ | |
663ead3b HX |
1626 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
1627 | skb_set_transport_header(skb, skb->csum_start - | |
1628 | skb_headroom(skb)); | |
1629 | ||
a298830c HX |
1630 | if (!(dev->features & NETIF_F_GEN_CSUM) && |
1631 | !((dev->features & NETIF_F_IP_CSUM) && | |
1632 | skb->protocol == htons(ETH_P_IP)) && | |
1633 | !((dev->features & NETIF_F_IPV6_CSUM) && | |
1634 | skb->protocol == htons(ETH_P_IPV6))) | |
663ead3b HX |
1635 | if (skb_checksum_help(skb)) |
1636 | goto out_kfree_skb; | |
1637 | } | |
1da177e4 | 1638 | |
f6a78bfc | 1639 | gso: |
2d7ceece ED |
1640 | spin_lock_prefetch(&dev->queue_lock); |
1641 | ||
4ec93edb YH |
1642 | /* Disable soft irqs for various locks below. Also |
1643 | * stops preemption for RCU. | |
1da177e4 | 1644 | */ |
4ec93edb | 1645 | rcu_read_lock_bh(); |
1da177e4 | 1646 | |
4ec93edb YH |
1647 | /* Updates of qdisc are serialized by queue_lock. |
1648 | * The struct Qdisc which is pointed to by qdisc is now a | |
1649 | * rcu structure - it may be accessed without acquiring | |
1da177e4 | 1650 | * a lock (but the structure may be stale.) The freeing of the |
4ec93edb | 1651 | * qdisc will be deferred until it's known that there are no |
1da177e4 | 1652 | * more references to it. |
4ec93edb YH |
1653 | * |
1654 | * If the qdisc has an enqueue function, we still need to | |
1da177e4 LT |
1655 | * hold the queue_lock before calling it, since queue_lock |
1656 | * also serializes access to the device queue. | |
1657 | */ | |
1658 | ||
1659 | q = rcu_dereference(dev->qdisc); | |
1660 | #ifdef CONFIG_NET_CLS_ACT | |
1661 | skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS); | |
1662 | #endif | |
1663 | if (q->enqueue) { | |
1664 | /* Grab device queue */ | |
1665 | spin_lock(&dev->queue_lock); | |
85670cc1 PM |
1666 | q = dev->qdisc; |
1667 | if (q->enqueue) { | |
f25f4e44 | 1668 | /* reset queue_mapping to zero */ |
dfa40911 | 1669 | skb_set_queue_mapping(skb, 0); |
85670cc1 PM |
1670 | rc = q->enqueue(skb, q); |
1671 | qdisc_run(dev); | |
1672 | spin_unlock(&dev->queue_lock); | |
1da177e4 | 1673 | |
85670cc1 PM |
1674 | rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc; |
1675 | goto out; | |
1676 | } | |
1da177e4 | 1677 | spin_unlock(&dev->queue_lock); |
1da177e4 LT |
1678 | } |
1679 | ||
1680 | /* The device has no queue. Common case for software devices: | |
1681 | loopback, all the sorts of tunnels... | |
1682 | ||
932ff279 HX |
1683 | Really, it is unlikely that netif_tx_lock protection is necessary |
1684 | here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
1da177e4 LT |
1685 | counters.) |
1686 | However, it is possible, that they rely on protection | |
1687 | made by us here. | |
1688 | ||
1689 | Check this and shot the lock. It is not prone from deadlocks. | |
1690 | Either shot noqueue qdisc, it is even simpler 8) | |
1691 | */ | |
1692 | if (dev->flags & IFF_UP) { | |
1693 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
1694 | ||
1695 | if (dev->xmit_lock_owner != cpu) { | |
1696 | ||
1697 | HARD_TX_LOCK(dev, cpu); | |
1698 | ||
f25f4e44 | 1699 | if (!netif_queue_stopped(dev) && |
668f895a | 1700 | !netif_subqueue_stopped(dev, skb)) { |
1da177e4 | 1701 | rc = 0; |
f6a78bfc | 1702 | if (!dev_hard_start_xmit(skb, dev)) { |
1da177e4 LT |
1703 | HARD_TX_UNLOCK(dev); |
1704 | goto out; | |
1705 | } | |
1706 | } | |
1707 | HARD_TX_UNLOCK(dev); | |
1708 | if (net_ratelimit()) | |
1709 | printk(KERN_CRIT "Virtual device %s asks to " | |
1710 | "queue packet!\n", dev->name); | |
1711 | } else { | |
1712 | /* Recursion is detected! It is possible, | |
1713 | * unfortunately */ | |
1714 | if (net_ratelimit()) | |
1715 | printk(KERN_CRIT "Dead loop on virtual device " | |
1716 | "%s, fix it urgently!\n", dev->name); | |
1717 | } | |
1718 | } | |
1719 | ||
1720 | rc = -ENETDOWN; | |
d4828d85 | 1721 | rcu_read_unlock_bh(); |
1da177e4 LT |
1722 | |
1723 | out_kfree_skb: | |
1724 | kfree_skb(skb); | |
1725 | return rc; | |
1726 | out: | |
d4828d85 | 1727 | rcu_read_unlock_bh(); |
1da177e4 LT |
1728 | return rc; |
1729 | } | |
1730 | ||
1731 | ||
1732 | /*======================================================================= | |
1733 | Receiver routines | |
1734 | =======================================================================*/ | |
1735 | ||
6b2bedc3 SH |
1736 | int netdev_max_backlog __read_mostly = 1000; |
1737 | int netdev_budget __read_mostly = 300; | |
1738 | int weight_p __read_mostly = 64; /* old backlog weight */ | |
1da177e4 LT |
1739 | |
1740 | DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, }; | |
1741 | ||
1742 | ||
1da177e4 LT |
1743 | /** |
1744 | * netif_rx - post buffer to the network code | |
1745 | * @skb: buffer to post | |
1746 | * | |
1747 | * This function receives a packet from a device driver and queues it for | |
1748 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
1749 | * may be dropped during processing for congestion control or by the | |
1750 | * protocol layers. | |
1751 | * | |
1752 | * return values: | |
1753 | * NET_RX_SUCCESS (no congestion) | |
1da177e4 LT |
1754 | * NET_RX_DROP (packet was dropped) |
1755 | * | |
1756 | */ | |
1757 | ||
1758 | int netif_rx(struct sk_buff *skb) | |
1759 | { | |
1da177e4 LT |
1760 | struct softnet_data *queue; |
1761 | unsigned long flags; | |
1762 | ||
1763 | /* if netpoll wants it, pretend we never saw it */ | |
1764 | if (netpoll_rx(skb)) | |
1765 | return NET_RX_DROP; | |
1766 | ||
b7aa0bf7 | 1767 | if (!skb->tstamp.tv64) |
a61bbcf2 | 1768 | net_timestamp(skb); |
1da177e4 LT |
1769 | |
1770 | /* | |
1771 | * The code is rearranged so that the path is the most | |
1772 | * short when CPU is congested, but is still operating. | |
1773 | */ | |
1774 | local_irq_save(flags); | |
1da177e4 LT |
1775 | queue = &__get_cpu_var(softnet_data); |
1776 | ||
1777 | __get_cpu_var(netdev_rx_stat).total++; | |
1778 | if (queue->input_pkt_queue.qlen <= netdev_max_backlog) { | |
1779 | if (queue->input_pkt_queue.qlen) { | |
1da177e4 LT |
1780 | enqueue: |
1781 | dev_hold(skb->dev); | |
1782 | __skb_queue_tail(&queue->input_pkt_queue, skb); | |
1da177e4 | 1783 | local_irq_restore(flags); |
34008d8c | 1784 | return NET_RX_SUCCESS; |
1da177e4 LT |
1785 | } |
1786 | ||
bea3348e | 1787 | napi_schedule(&queue->backlog); |
1da177e4 LT |
1788 | goto enqueue; |
1789 | } | |
1790 | ||
1da177e4 LT |
1791 | __get_cpu_var(netdev_rx_stat).dropped++; |
1792 | local_irq_restore(flags); | |
1793 | ||
1794 | kfree_skb(skb); | |
1795 | return NET_RX_DROP; | |
1796 | } | |
1797 | ||
1798 | int netif_rx_ni(struct sk_buff *skb) | |
1799 | { | |
1800 | int err; | |
1801 | ||
1802 | preempt_disable(); | |
1803 | err = netif_rx(skb); | |
1804 | if (local_softirq_pending()) | |
1805 | do_softirq(); | |
1806 | preempt_enable(); | |
1807 | ||
1808 | return err; | |
1809 | } | |
1810 | ||
1811 | EXPORT_SYMBOL(netif_rx_ni); | |
1812 | ||
f2ccd8fa | 1813 | static inline struct net_device *skb_bond(struct sk_buff *skb) |
1da177e4 LT |
1814 | { |
1815 | struct net_device *dev = skb->dev; | |
1816 | ||
8f903c70 | 1817 | if (dev->master) { |
7ea49ed7 | 1818 | if (skb_bond_should_drop(skb)) { |
8f903c70 JV |
1819 | kfree_skb(skb); |
1820 | return NULL; | |
1821 | } | |
1da177e4 | 1822 | skb->dev = dev->master; |
8f903c70 | 1823 | } |
f2ccd8fa DM |
1824 | |
1825 | return dev; | |
1da177e4 LT |
1826 | } |
1827 | ||
bea3348e | 1828 | |
1da177e4 LT |
1829 | static void net_tx_action(struct softirq_action *h) |
1830 | { | |
1831 | struct softnet_data *sd = &__get_cpu_var(softnet_data); | |
1832 | ||
1833 | if (sd->completion_queue) { | |
1834 | struct sk_buff *clist; | |
1835 | ||
1836 | local_irq_disable(); | |
1837 | clist = sd->completion_queue; | |
1838 | sd->completion_queue = NULL; | |
1839 | local_irq_enable(); | |
1840 | ||
1841 | while (clist) { | |
1842 | struct sk_buff *skb = clist; | |
1843 | clist = clist->next; | |
1844 | ||
1845 | BUG_TRAP(!atomic_read(&skb->users)); | |
1846 | __kfree_skb(skb); | |
1847 | } | |
1848 | } | |
1849 | ||
1850 | if (sd->output_queue) { | |
1851 | struct net_device *head; | |
1852 | ||
1853 | local_irq_disable(); | |
1854 | head = sd->output_queue; | |
1855 | sd->output_queue = NULL; | |
1856 | local_irq_enable(); | |
1857 | ||
1858 | while (head) { | |
1859 | struct net_device *dev = head; | |
1860 | head = head->next_sched; | |
1861 | ||
1862 | smp_mb__before_clear_bit(); | |
1863 | clear_bit(__LINK_STATE_SCHED, &dev->state); | |
1864 | ||
1865 | if (spin_trylock(&dev->queue_lock)) { | |
1866 | qdisc_run(dev); | |
1867 | spin_unlock(&dev->queue_lock); | |
1868 | } else { | |
1869 | netif_schedule(dev); | |
1870 | } | |
1871 | } | |
1872 | } | |
1873 | } | |
1874 | ||
6f05f629 SH |
1875 | static inline int deliver_skb(struct sk_buff *skb, |
1876 | struct packet_type *pt_prev, | |
1877 | struct net_device *orig_dev) | |
1da177e4 LT |
1878 | { |
1879 | atomic_inc(&skb->users); | |
f2ccd8fa | 1880 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
1da177e4 LT |
1881 | } |
1882 | ||
1883 | #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE) | |
6229e362 | 1884 | /* These hooks defined here for ATM */ |
1da177e4 LT |
1885 | struct net_bridge; |
1886 | struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br, | |
1887 | unsigned char *addr); | |
6229e362 | 1888 | void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent) __read_mostly; |
1da177e4 | 1889 | |
6229e362 SH |
1890 | /* |
1891 | * If bridge module is loaded call bridging hook. | |
1892 | * returns NULL if packet was consumed. | |
1893 | */ | |
1894 | struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p, | |
1895 | struct sk_buff *skb) __read_mostly; | |
1896 | static inline struct sk_buff *handle_bridge(struct sk_buff *skb, | |
1897 | struct packet_type **pt_prev, int *ret, | |
1898 | struct net_device *orig_dev) | |
1da177e4 LT |
1899 | { |
1900 | struct net_bridge_port *port; | |
1901 | ||
6229e362 SH |
1902 | if (skb->pkt_type == PACKET_LOOPBACK || |
1903 | (port = rcu_dereference(skb->dev->br_port)) == NULL) | |
1904 | return skb; | |
1da177e4 LT |
1905 | |
1906 | if (*pt_prev) { | |
6229e362 | 1907 | *ret = deliver_skb(skb, *pt_prev, orig_dev); |
1da177e4 | 1908 | *pt_prev = NULL; |
4ec93edb YH |
1909 | } |
1910 | ||
6229e362 | 1911 | return br_handle_frame_hook(port, skb); |
1da177e4 LT |
1912 | } |
1913 | #else | |
6229e362 | 1914 | #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb) |
1da177e4 LT |
1915 | #endif |
1916 | ||
b863ceb7 PM |
1917 | #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE) |
1918 | struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly; | |
1919 | EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook); | |
1920 | ||
1921 | static inline struct sk_buff *handle_macvlan(struct sk_buff *skb, | |
1922 | struct packet_type **pt_prev, | |
1923 | int *ret, | |
1924 | struct net_device *orig_dev) | |
1925 | { | |
1926 | if (skb->dev->macvlan_port == NULL) | |
1927 | return skb; | |
1928 | ||
1929 | if (*pt_prev) { | |
1930 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
1931 | *pt_prev = NULL; | |
1932 | } | |
1933 | return macvlan_handle_frame_hook(skb); | |
1934 | } | |
1935 | #else | |
1936 | #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb) | |
1937 | #endif | |
1938 | ||
1da177e4 LT |
1939 | #ifdef CONFIG_NET_CLS_ACT |
1940 | /* TODO: Maybe we should just force sch_ingress to be compiled in | |
1941 | * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions | |
1942 | * a compare and 2 stores extra right now if we dont have it on | |
1943 | * but have CONFIG_NET_CLS_ACT | |
4ec93edb | 1944 | * NOTE: This doesnt stop any functionality; if you dont have |
1da177e4 LT |
1945 | * the ingress scheduler, you just cant add policies on ingress. |
1946 | * | |
1947 | */ | |
4ec93edb | 1948 | static int ing_filter(struct sk_buff *skb) |
1da177e4 LT |
1949 | { |
1950 | struct Qdisc *q; | |
1951 | struct net_device *dev = skb->dev; | |
1952 | int result = TC_ACT_OK; | |
f697c3e8 | 1953 | u32 ttl = G_TC_RTTL(skb->tc_verd); |
4ec93edb | 1954 | |
f697c3e8 HX |
1955 | if (MAX_RED_LOOP < ttl++) { |
1956 | printk(KERN_WARNING | |
1957 | "Redir loop detected Dropping packet (%d->%d)\n", | |
1958 | skb->iif, dev->ifindex); | |
1959 | return TC_ACT_SHOT; | |
1960 | } | |
1da177e4 | 1961 | |
f697c3e8 HX |
1962 | skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl); |
1963 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); | |
1da177e4 | 1964 | |
f697c3e8 HX |
1965 | spin_lock(&dev->ingress_lock); |
1966 | if ((q = dev->qdisc_ingress) != NULL) | |
1967 | result = q->enqueue(skb, q); | |
1968 | spin_unlock(&dev->ingress_lock); | |
1969 | ||
1970 | return result; | |
1971 | } | |
86e65da9 | 1972 | |
f697c3e8 HX |
1973 | static inline struct sk_buff *handle_ing(struct sk_buff *skb, |
1974 | struct packet_type **pt_prev, | |
1975 | int *ret, struct net_device *orig_dev) | |
1976 | { | |
1977 | if (!skb->dev->qdisc_ingress) | |
1978 | goto out; | |
1da177e4 | 1979 | |
f697c3e8 HX |
1980 | if (*pt_prev) { |
1981 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
1982 | *pt_prev = NULL; | |
1983 | } else { | |
1984 | /* Huh? Why does turning on AF_PACKET affect this? */ | |
1985 | skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd); | |
1da177e4 LT |
1986 | } |
1987 | ||
f697c3e8 HX |
1988 | switch (ing_filter(skb)) { |
1989 | case TC_ACT_SHOT: | |
1990 | case TC_ACT_STOLEN: | |
1991 | kfree_skb(skb); | |
1992 | return NULL; | |
1993 | } | |
1994 | ||
1995 | out: | |
1996 | skb->tc_verd = 0; | |
1997 | return skb; | |
1da177e4 LT |
1998 | } |
1999 | #endif | |
2000 | ||
3b582cc1 SH |
2001 | /** |
2002 | * netif_receive_skb - process receive buffer from network | |
2003 | * @skb: buffer to process | |
2004 | * | |
2005 | * netif_receive_skb() is the main receive data processing function. | |
2006 | * It always succeeds. The buffer may be dropped during processing | |
2007 | * for congestion control or by the protocol layers. | |
2008 | * | |
2009 | * This function may only be called from softirq context and interrupts | |
2010 | * should be enabled. | |
2011 | * | |
2012 | * Return values (usually ignored): | |
2013 | * NET_RX_SUCCESS: no congestion | |
2014 | * NET_RX_DROP: packet was dropped | |
2015 | */ | |
1da177e4 LT |
2016 | int netif_receive_skb(struct sk_buff *skb) |
2017 | { | |
2018 | struct packet_type *ptype, *pt_prev; | |
f2ccd8fa | 2019 | struct net_device *orig_dev; |
1da177e4 | 2020 | int ret = NET_RX_DROP; |
252e3346 | 2021 | __be16 type; |
1da177e4 LT |
2022 | |
2023 | /* if we've gotten here through NAPI, check netpoll */ | |
bea3348e | 2024 | if (netpoll_receive_skb(skb)) |
1da177e4 LT |
2025 | return NET_RX_DROP; |
2026 | ||
b7aa0bf7 | 2027 | if (!skb->tstamp.tv64) |
a61bbcf2 | 2028 | net_timestamp(skb); |
1da177e4 | 2029 | |
c01003c2 PM |
2030 | if (!skb->iif) |
2031 | skb->iif = skb->dev->ifindex; | |
86e65da9 | 2032 | |
f2ccd8fa | 2033 | orig_dev = skb_bond(skb); |
1da177e4 | 2034 | |
8f903c70 JV |
2035 | if (!orig_dev) |
2036 | return NET_RX_DROP; | |
2037 | ||
1da177e4 LT |
2038 | __get_cpu_var(netdev_rx_stat).total++; |
2039 | ||
c1d2bbe1 | 2040 | skb_reset_network_header(skb); |
badff6d0 | 2041 | skb_reset_transport_header(skb); |
b0e380b1 | 2042 | skb->mac_len = skb->network_header - skb->mac_header; |
1da177e4 LT |
2043 | |
2044 | pt_prev = NULL; | |
2045 | ||
2046 | rcu_read_lock(); | |
2047 | ||
2048 | #ifdef CONFIG_NET_CLS_ACT | |
2049 | if (skb->tc_verd & TC_NCLS) { | |
2050 | skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); | |
2051 | goto ncls; | |
2052 | } | |
2053 | #endif | |
2054 | ||
2055 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
2056 | if (!ptype->dev || ptype->dev == skb->dev) { | |
4ec93edb | 2057 | if (pt_prev) |
f2ccd8fa | 2058 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
2059 | pt_prev = ptype; |
2060 | } | |
2061 | } | |
2062 | ||
2063 | #ifdef CONFIG_NET_CLS_ACT | |
f697c3e8 HX |
2064 | skb = handle_ing(skb, &pt_prev, &ret, orig_dev); |
2065 | if (!skb) | |
1da177e4 | 2066 | goto out; |
1da177e4 LT |
2067 | ncls: |
2068 | #endif | |
2069 | ||
6229e362 | 2070 | skb = handle_bridge(skb, &pt_prev, &ret, orig_dev); |
b863ceb7 PM |
2071 | if (!skb) |
2072 | goto out; | |
2073 | skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev); | |
6229e362 | 2074 | if (!skb) |
1da177e4 LT |
2075 | goto out; |
2076 | ||
2077 | type = skb->protocol; | |
2078 | list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) { | |
2079 | if (ptype->type == type && | |
2080 | (!ptype->dev || ptype->dev == skb->dev)) { | |
4ec93edb | 2081 | if (pt_prev) |
f2ccd8fa | 2082 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
2083 | pt_prev = ptype; |
2084 | } | |
2085 | } | |
2086 | ||
2087 | if (pt_prev) { | |
f2ccd8fa | 2088 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
1da177e4 LT |
2089 | } else { |
2090 | kfree_skb(skb); | |
2091 | /* Jamal, now you will not able to escape explaining | |
2092 | * me how you were going to use this. :-) | |
2093 | */ | |
2094 | ret = NET_RX_DROP; | |
2095 | } | |
2096 | ||
2097 | out: | |
2098 | rcu_read_unlock(); | |
2099 | return ret; | |
2100 | } | |
2101 | ||
bea3348e | 2102 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 LT |
2103 | { |
2104 | int work = 0; | |
1da177e4 LT |
2105 | struct softnet_data *queue = &__get_cpu_var(softnet_data); |
2106 | unsigned long start_time = jiffies; | |
2107 | ||
bea3348e SH |
2108 | napi->weight = weight_p; |
2109 | do { | |
1da177e4 LT |
2110 | struct sk_buff *skb; |
2111 | struct net_device *dev; | |
2112 | ||
2113 | local_irq_disable(); | |
2114 | skb = __skb_dequeue(&queue->input_pkt_queue); | |
bea3348e SH |
2115 | if (!skb) { |
2116 | __napi_complete(napi); | |
2117 | local_irq_enable(); | |
2118 | break; | |
2119 | } | |
2120 | ||
1da177e4 LT |
2121 | local_irq_enable(); |
2122 | ||
2123 | dev = skb->dev; | |
2124 | ||
2125 | netif_receive_skb(skb); | |
2126 | ||
2127 | dev_put(dev); | |
bea3348e | 2128 | } while (++work < quota && jiffies == start_time); |
1da177e4 | 2129 | |
bea3348e SH |
2130 | return work; |
2131 | } | |
1da177e4 | 2132 | |
bea3348e SH |
2133 | /** |
2134 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 2135 | * @n: entry to schedule |
bea3348e SH |
2136 | * |
2137 | * The entry's receive function will be scheduled to run | |
2138 | */ | |
2139 | void fastcall __napi_schedule(struct napi_struct *n) | |
2140 | { | |
2141 | unsigned long flags; | |
1da177e4 | 2142 | |
bea3348e SH |
2143 | local_irq_save(flags); |
2144 | list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list); | |
2145 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2146 | local_irq_restore(flags); | |
1da177e4 | 2147 | } |
bea3348e SH |
2148 | EXPORT_SYMBOL(__napi_schedule); |
2149 | ||
1da177e4 LT |
2150 | |
2151 | static void net_rx_action(struct softirq_action *h) | |
2152 | { | |
bea3348e | 2153 | struct list_head *list = &__get_cpu_var(softnet_data).poll_list; |
1da177e4 | 2154 | unsigned long start_time = jiffies; |
51b0bded | 2155 | int budget = netdev_budget; |
53fb95d3 MM |
2156 | void *have; |
2157 | ||
1da177e4 LT |
2158 | local_irq_disable(); |
2159 | ||
bea3348e SH |
2160 | while (!list_empty(list)) { |
2161 | struct napi_struct *n; | |
2162 | int work, weight; | |
1da177e4 | 2163 | |
bea3348e SH |
2164 | /* If softirq window is exhuasted then punt. |
2165 | * | |
2166 | * Note that this is a slight policy change from the | |
2167 | * previous NAPI code, which would allow up to 2 | |
2168 | * jiffies to pass before breaking out. The test | |
2169 | * used to be "jiffies - start_time > 1". | |
2170 | */ | |
2171 | if (unlikely(budget <= 0 || jiffies != start_time)) | |
1da177e4 LT |
2172 | goto softnet_break; |
2173 | ||
2174 | local_irq_enable(); | |
2175 | ||
bea3348e SH |
2176 | /* Even though interrupts have been re-enabled, this |
2177 | * access is safe because interrupts can only add new | |
2178 | * entries to the tail of this list, and only ->poll() | |
2179 | * calls can remove this head entry from the list. | |
2180 | */ | |
2181 | n = list_entry(list->next, struct napi_struct, poll_list); | |
1da177e4 | 2182 | |
bea3348e SH |
2183 | have = netpoll_poll_lock(n); |
2184 | ||
2185 | weight = n->weight; | |
2186 | ||
0a7606c1 DM |
2187 | /* This NAPI_STATE_SCHED test is for avoiding a race |
2188 | * with netpoll's poll_napi(). Only the entity which | |
2189 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
2190 | * actually make the ->poll() call. Therefore we avoid | |
2191 | * accidently calling ->poll() when NAPI is not scheduled. | |
2192 | */ | |
2193 | work = 0; | |
2194 | if (test_bit(NAPI_STATE_SCHED, &n->state)) | |
2195 | work = n->poll(n, weight); | |
bea3348e SH |
2196 | |
2197 | WARN_ON_ONCE(work > weight); | |
2198 | ||
2199 | budget -= work; | |
2200 | ||
2201 | local_irq_disable(); | |
2202 | ||
2203 | /* Drivers must not modify the NAPI state if they | |
2204 | * consume the entire weight. In such cases this code | |
2205 | * still "owns" the NAPI instance and therefore can | |
2206 | * move the instance around on the list at-will. | |
2207 | */ | |
2208 | if (unlikely(work == weight)) | |
2209 | list_move_tail(&n->poll_list, list); | |
2210 | ||
2211 | netpoll_poll_unlock(have); | |
1da177e4 LT |
2212 | } |
2213 | out: | |
515e06c4 | 2214 | local_irq_enable(); |
bea3348e | 2215 | |
db217334 CL |
2216 | #ifdef CONFIG_NET_DMA |
2217 | /* | |
2218 | * There may not be any more sk_buffs coming right now, so push | |
2219 | * any pending DMA copies to hardware | |
2220 | */ | |
d379b01e DW |
2221 | if (!cpus_empty(net_dma.channel_mask)) { |
2222 | int chan_idx; | |
2223 | for_each_cpu_mask(chan_idx, net_dma.channel_mask) { | |
2224 | struct dma_chan *chan = net_dma.channels[chan_idx]; | |
2225 | if (chan) | |
2226 | dma_async_memcpy_issue_pending(chan); | |
2227 | } | |
db217334 CL |
2228 | } |
2229 | #endif | |
bea3348e | 2230 | |
1da177e4 LT |
2231 | return; |
2232 | ||
2233 | softnet_break: | |
2234 | __get_cpu_var(netdev_rx_stat).time_squeeze++; | |
2235 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2236 | goto out; | |
2237 | } | |
2238 | ||
2239 | static gifconf_func_t * gifconf_list [NPROTO]; | |
2240 | ||
2241 | /** | |
2242 | * register_gifconf - register a SIOCGIF handler | |
2243 | * @family: Address family | |
2244 | * @gifconf: Function handler | |
2245 | * | |
2246 | * Register protocol dependent address dumping routines. The handler | |
2247 | * that is passed must not be freed or reused until it has been replaced | |
2248 | * by another handler. | |
2249 | */ | |
2250 | int register_gifconf(unsigned int family, gifconf_func_t * gifconf) | |
2251 | { | |
2252 | if (family >= NPROTO) | |
2253 | return -EINVAL; | |
2254 | gifconf_list[family] = gifconf; | |
2255 | return 0; | |
2256 | } | |
2257 | ||
2258 | ||
2259 | /* | |
2260 | * Map an interface index to its name (SIOCGIFNAME) | |
2261 | */ | |
2262 | ||
2263 | /* | |
2264 | * We need this ioctl for efficient implementation of the | |
2265 | * if_indextoname() function required by the IPv6 API. Without | |
2266 | * it, we would have to search all the interfaces to find a | |
2267 | * match. --pb | |
2268 | */ | |
2269 | ||
881d966b | 2270 | static int dev_ifname(struct net *net, struct ifreq __user *arg) |
1da177e4 LT |
2271 | { |
2272 | struct net_device *dev; | |
2273 | struct ifreq ifr; | |
2274 | ||
2275 | /* | |
2276 | * Fetch the caller's info block. | |
2277 | */ | |
2278 | ||
2279 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
2280 | return -EFAULT; | |
2281 | ||
2282 | read_lock(&dev_base_lock); | |
881d966b | 2283 | dev = __dev_get_by_index(net, ifr.ifr_ifindex); |
1da177e4 LT |
2284 | if (!dev) { |
2285 | read_unlock(&dev_base_lock); | |
2286 | return -ENODEV; | |
2287 | } | |
2288 | ||
2289 | strcpy(ifr.ifr_name, dev->name); | |
2290 | read_unlock(&dev_base_lock); | |
2291 | ||
2292 | if (copy_to_user(arg, &ifr, sizeof(struct ifreq))) | |
2293 | return -EFAULT; | |
2294 | return 0; | |
2295 | } | |
2296 | ||
2297 | /* | |
2298 | * Perform a SIOCGIFCONF call. This structure will change | |
2299 | * size eventually, and there is nothing I can do about it. | |
2300 | * Thus we will need a 'compatibility mode'. | |
2301 | */ | |
2302 | ||
881d966b | 2303 | static int dev_ifconf(struct net *net, char __user *arg) |
1da177e4 LT |
2304 | { |
2305 | struct ifconf ifc; | |
2306 | struct net_device *dev; | |
2307 | char __user *pos; | |
2308 | int len; | |
2309 | int total; | |
2310 | int i; | |
2311 | ||
2312 | /* | |
2313 | * Fetch the caller's info block. | |
2314 | */ | |
2315 | ||
2316 | if (copy_from_user(&ifc, arg, sizeof(struct ifconf))) | |
2317 | return -EFAULT; | |
2318 | ||
2319 | pos = ifc.ifc_buf; | |
2320 | len = ifc.ifc_len; | |
2321 | ||
2322 | /* | |
2323 | * Loop over the interfaces, and write an info block for each. | |
2324 | */ | |
2325 | ||
2326 | total = 0; | |
881d966b | 2327 | for_each_netdev(net, dev) { |
1da177e4 LT |
2328 | for (i = 0; i < NPROTO; i++) { |
2329 | if (gifconf_list[i]) { | |
2330 | int done; | |
2331 | if (!pos) | |
2332 | done = gifconf_list[i](dev, NULL, 0); | |
2333 | else | |
2334 | done = gifconf_list[i](dev, pos + total, | |
2335 | len - total); | |
2336 | if (done < 0) | |
2337 | return -EFAULT; | |
2338 | total += done; | |
2339 | } | |
2340 | } | |
4ec93edb | 2341 | } |
1da177e4 LT |
2342 | |
2343 | /* | |
2344 | * All done. Write the updated control block back to the caller. | |
2345 | */ | |
2346 | ifc.ifc_len = total; | |
2347 | ||
2348 | /* | |
2349 | * Both BSD and Solaris return 0 here, so we do too. | |
2350 | */ | |
2351 | return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0; | |
2352 | } | |
2353 | ||
2354 | #ifdef CONFIG_PROC_FS | |
2355 | /* | |
2356 | * This is invoked by the /proc filesystem handler to display a device | |
2357 | * in detail. | |
2358 | */ | |
7562f876 | 2359 | void *dev_seq_start(struct seq_file *seq, loff_t *pos) |
1da177e4 | 2360 | { |
881d966b | 2361 | struct net *net = seq->private; |
7562f876 | 2362 | loff_t off; |
1da177e4 | 2363 | struct net_device *dev; |
1da177e4 | 2364 | |
7562f876 PE |
2365 | read_lock(&dev_base_lock); |
2366 | if (!*pos) | |
2367 | return SEQ_START_TOKEN; | |
1da177e4 | 2368 | |
7562f876 | 2369 | off = 1; |
881d966b | 2370 | for_each_netdev(net, dev) |
7562f876 PE |
2371 | if (off++ == *pos) |
2372 | return dev; | |
1da177e4 | 2373 | |
7562f876 | 2374 | return NULL; |
1da177e4 LT |
2375 | } |
2376 | ||
2377 | void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
2378 | { | |
881d966b | 2379 | struct net *net = seq->private; |
1da177e4 | 2380 | ++*pos; |
7562f876 | 2381 | return v == SEQ_START_TOKEN ? |
881d966b | 2382 | first_net_device(net) : next_net_device((struct net_device *)v); |
1da177e4 LT |
2383 | } |
2384 | ||
2385 | void dev_seq_stop(struct seq_file *seq, void *v) | |
2386 | { | |
2387 | read_unlock(&dev_base_lock); | |
2388 | } | |
2389 | ||
2390 | static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev) | |
2391 | { | |
c45d286e | 2392 | struct net_device_stats *stats = dev->get_stats(dev); |
1da177e4 | 2393 | |
5a1b5898 RR |
2394 | seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu " |
2395 | "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n", | |
2396 | dev->name, stats->rx_bytes, stats->rx_packets, | |
2397 | stats->rx_errors, | |
2398 | stats->rx_dropped + stats->rx_missed_errors, | |
2399 | stats->rx_fifo_errors, | |
2400 | stats->rx_length_errors + stats->rx_over_errors + | |
2401 | stats->rx_crc_errors + stats->rx_frame_errors, | |
2402 | stats->rx_compressed, stats->multicast, | |
2403 | stats->tx_bytes, stats->tx_packets, | |
2404 | stats->tx_errors, stats->tx_dropped, | |
2405 | stats->tx_fifo_errors, stats->collisions, | |
2406 | stats->tx_carrier_errors + | |
2407 | stats->tx_aborted_errors + | |
2408 | stats->tx_window_errors + | |
2409 | stats->tx_heartbeat_errors, | |
2410 | stats->tx_compressed); | |
1da177e4 LT |
2411 | } |
2412 | ||
2413 | /* | |
2414 | * Called from the PROCfs module. This now uses the new arbitrary sized | |
2415 | * /proc/net interface to create /proc/net/dev | |
2416 | */ | |
2417 | static int dev_seq_show(struct seq_file *seq, void *v) | |
2418 | { | |
2419 | if (v == SEQ_START_TOKEN) | |
2420 | seq_puts(seq, "Inter-| Receive " | |
2421 | " | Transmit\n" | |
2422 | " face |bytes packets errs drop fifo frame " | |
2423 | "compressed multicast|bytes packets errs " | |
2424 | "drop fifo colls carrier compressed\n"); | |
2425 | else | |
2426 | dev_seq_printf_stats(seq, v); | |
2427 | return 0; | |
2428 | } | |
2429 | ||
2430 | static struct netif_rx_stats *softnet_get_online(loff_t *pos) | |
2431 | { | |
2432 | struct netif_rx_stats *rc = NULL; | |
2433 | ||
2434 | while (*pos < NR_CPUS) | |
4ec93edb | 2435 | if (cpu_online(*pos)) { |
1da177e4 LT |
2436 | rc = &per_cpu(netdev_rx_stat, *pos); |
2437 | break; | |
2438 | } else | |
2439 | ++*pos; | |
2440 | return rc; | |
2441 | } | |
2442 | ||
2443 | static void *softnet_seq_start(struct seq_file *seq, loff_t *pos) | |
2444 | { | |
2445 | return softnet_get_online(pos); | |
2446 | } | |
2447 | ||
2448 | static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
2449 | { | |
2450 | ++*pos; | |
2451 | return softnet_get_online(pos); | |
2452 | } | |
2453 | ||
2454 | static void softnet_seq_stop(struct seq_file *seq, void *v) | |
2455 | { | |
2456 | } | |
2457 | ||
2458 | static int softnet_seq_show(struct seq_file *seq, void *v) | |
2459 | { | |
2460 | struct netif_rx_stats *s = v; | |
2461 | ||
2462 | seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n", | |
31aa02c5 | 2463 | s->total, s->dropped, s->time_squeeze, 0, |
c1ebcdb8 SH |
2464 | 0, 0, 0, 0, /* was fastroute */ |
2465 | s->cpu_collision ); | |
1da177e4 LT |
2466 | return 0; |
2467 | } | |
2468 | ||
f690808e | 2469 | static const struct seq_operations dev_seq_ops = { |
1da177e4 LT |
2470 | .start = dev_seq_start, |
2471 | .next = dev_seq_next, | |
2472 | .stop = dev_seq_stop, | |
2473 | .show = dev_seq_show, | |
2474 | }; | |
2475 | ||
2476 | static int dev_seq_open(struct inode *inode, struct file *file) | |
2477 | { | |
881d966b EB |
2478 | struct seq_file *seq; |
2479 | int res; | |
2480 | res = seq_open(file, &dev_seq_ops); | |
2481 | if (!res) { | |
2482 | seq = file->private_data; | |
077130c0 EB |
2483 | seq->private = get_proc_net(inode); |
2484 | if (!seq->private) { | |
2485 | seq_release(inode, file); | |
2486 | res = -ENXIO; | |
2487 | } | |
881d966b EB |
2488 | } |
2489 | return res; | |
2490 | } | |
2491 | ||
2492 | static int dev_seq_release(struct inode *inode, struct file *file) | |
2493 | { | |
2494 | struct seq_file *seq = file->private_data; | |
2495 | struct net *net = seq->private; | |
2496 | put_net(net); | |
2497 | return seq_release(inode, file); | |
1da177e4 LT |
2498 | } |
2499 | ||
9a32144e | 2500 | static const struct file_operations dev_seq_fops = { |
1da177e4 LT |
2501 | .owner = THIS_MODULE, |
2502 | .open = dev_seq_open, | |
2503 | .read = seq_read, | |
2504 | .llseek = seq_lseek, | |
881d966b | 2505 | .release = dev_seq_release, |
1da177e4 LT |
2506 | }; |
2507 | ||
f690808e | 2508 | static const struct seq_operations softnet_seq_ops = { |
1da177e4 LT |
2509 | .start = softnet_seq_start, |
2510 | .next = softnet_seq_next, | |
2511 | .stop = softnet_seq_stop, | |
2512 | .show = softnet_seq_show, | |
2513 | }; | |
2514 | ||
2515 | static int softnet_seq_open(struct inode *inode, struct file *file) | |
2516 | { | |
2517 | return seq_open(file, &softnet_seq_ops); | |
2518 | } | |
2519 | ||
9a32144e | 2520 | static const struct file_operations softnet_seq_fops = { |
1da177e4 LT |
2521 | .owner = THIS_MODULE, |
2522 | .open = softnet_seq_open, | |
2523 | .read = seq_read, | |
2524 | .llseek = seq_lseek, | |
2525 | .release = seq_release, | |
2526 | }; | |
2527 | ||
0e1256ff SH |
2528 | static void *ptype_get_idx(loff_t pos) |
2529 | { | |
2530 | struct packet_type *pt = NULL; | |
2531 | loff_t i = 0; | |
2532 | int t; | |
2533 | ||
2534 | list_for_each_entry_rcu(pt, &ptype_all, list) { | |
2535 | if (i == pos) | |
2536 | return pt; | |
2537 | ++i; | |
2538 | } | |
2539 | ||
2540 | for (t = 0; t < 16; t++) { | |
2541 | list_for_each_entry_rcu(pt, &ptype_base[t], list) { | |
2542 | if (i == pos) | |
2543 | return pt; | |
2544 | ++i; | |
2545 | } | |
2546 | } | |
2547 | return NULL; | |
2548 | } | |
2549 | ||
2550 | static void *ptype_seq_start(struct seq_file *seq, loff_t *pos) | |
2551 | { | |
2552 | rcu_read_lock(); | |
2553 | return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN; | |
2554 | } | |
2555 | ||
2556 | static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
2557 | { | |
2558 | struct packet_type *pt; | |
2559 | struct list_head *nxt; | |
2560 | int hash; | |
2561 | ||
2562 | ++*pos; | |
2563 | if (v == SEQ_START_TOKEN) | |
2564 | return ptype_get_idx(0); | |
2565 | ||
2566 | pt = v; | |
2567 | nxt = pt->list.next; | |
2568 | if (pt->type == htons(ETH_P_ALL)) { | |
2569 | if (nxt != &ptype_all) | |
2570 | goto found; | |
2571 | hash = 0; | |
2572 | nxt = ptype_base[0].next; | |
2573 | } else | |
2574 | hash = ntohs(pt->type) & 15; | |
2575 | ||
2576 | while (nxt == &ptype_base[hash]) { | |
2577 | if (++hash >= 16) | |
2578 | return NULL; | |
2579 | nxt = ptype_base[hash].next; | |
2580 | } | |
2581 | found: | |
2582 | return list_entry(nxt, struct packet_type, list); | |
2583 | } | |
2584 | ||
2585 | static void ptype_seq_stop(struct seq_file *seq, void *v) | |
2586 | { | |
2587 | rcu_read_unlock(); | |
2588 | } | |
2589 | ||
2590 | static void ptype_seq_decode(struct seq_file *seq, void *sym) | |
2591 | { | |
2592 | #ifdef CONFIG_KALLSYMS | |
2593 | unsigned long offset = 0, symsize; | |
2594 | const char *symname; | |
2595 | char *modname; | |
2596 | char namebuf[128]; | |
2597 | ||
2598 | symname = kallsyms_lookup((unsigned long)sym, &symsize, &offset, | |
2599 | &modname, namebuf); | |
2600 | ||
2601 | if (symname) { | |
2602 | char *delim = ":"; | |
2603 | ||
2604 | if (!modname) | |
2605 | modname = delim = ""; | |
2606 | seq_printf(seq, "%s%s%s%s+0x%lx", delim, modname, delim, | |
2607 | symname, offset); | |
2608 | return; | |
2609 | } | |
2610 | #endif | |
2611 | ||
2612 | seq_printf(seq, "[%p]", sym); | |
2613 | } | |
2614 | ||
2615 | static int ptype_seq_show(struct seq_file *seq, void *v) | |
2616 | { | |
2617 | struct packet_type *pt = v; | |
2618 | ||
2619 | if (v == SEQ_START_TOKEN) | |
2620 | seq_puts(seq, "Type Device Function\n"); | |
2621 | else { | |
2622 | if (pt->type == htons(ETH_P_ALL)) | |
2623 | seq_puts(seq, "ALL "); | |
2624 | else | |
2625 | seq_printf(seq, "%04x", ntohs(pt->type)); | |
2626 | ||
2627 | seq_printf(seq, " %-8s ", | |
2628 | pt->dev ? pt->dev->name : ""); | |
2629 | ptype_seq_decode(seq, pt->func); | |
2630 | seq_putc(seq, '\n'); | |
2631 | } | |
2632 | ||
2633 | return 0; | |
2634 | } | |
2635 | ||
2636 | static const struct seq_operations ptype_seq_ops = { | |
2637 | .start = ptype_seq_start, | |
2638 | .next = ptype_seq_next, | |
2639 | .stop = ptype_seq_stop, | |
2640 | .show = ptype_seq_show, | |
2641 | }; | |
2642 | ||
2643 | static int ptype_seq_open(struct inode *inode, struct file *file) | |
2644 | { | |
2645 | return seq_open(file, &ptype_seq_ops); | |
2646 | } | |
2647 | ||
2648 | static const struct file_operations ptype_seq_fops = { | |
2649 | .owner = THIS_MODULE, | |
2650 | .open = ptype_seq_open, | |
2651 | .read = seq_read, | |
2652 | .llseek = seq_lseek, | |
2653 | .release = seq_release, | |
2654 | }; | |
2655 | ||
2656 | ||
4665079c | 2657 | static int __net_init dev_proc_net_init(struct net *net) |
1da177e4 LT |
2658 | { |
2659 | int rc = -ENOMEM; | |
2660 | ||
881d966b | 2661 | if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops)) |
1da177e4 | 2662 | goto out; |
881d966b | 2663 | if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops)) |
1da177e4 | 2664 | goto out_dev; |
881d966b | 2665 | if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops)) |
457c4cbc | 2666 | goto out_softnet; |
0e1256ff | 2667 | |
881d966b | 2668 | if (wext_proc_init(net)) |
457c4cbc | 2669 | goto out_ptype; |
1da177e4 LT |
2670 | rc = 0; |
2671 | out: | |
2672 | return rc; | |
457c4cbc | 2673 | out_ptype: |
881d966b | 2674 | proc_net_remove(net, "ptype"); |
1da177e4 | 2675 | out_softnet: |
881d966b | 2676 | proc_net_remove(net, "softnet_stat"); |
1da177e4 | 2677 | out_dev: |
881d966b | 2678 | proc_net_remove(net, "dev"); |
1da177e4 LT |
2679 | goto out; |
2680 | } | |
881d966b | 2681 | |
4665079c | 2682 | static void __net_exit dev_proc_net_exit(struct net *net) |
881d966b EB |
2683 | { |
2684 | wext_proc_exit(net); | |
2685 | ||
2686 | proc_net_remove(net, "ptype"); | |
2687 | proc_net_remove(net, "softnet_stat"); | |
2688 | proc_net_remove(net, "dev"); | |
2689 | } | |
2690 | ||
022cbae6 | 2691 | static struct pernet_operations __net_initdata dev_proc_ops = { |
881d966b EB |
2692 | .init = dev_proc_net_init, |
2693 | .exit = dev_proc_net_exit, | |
2694 | }; | |
2695 | ||
2696 | static int __init dev_proc_init(void) | |
2697 | { | |
2698 | return register_pernet_subsys(&dev_proc_ops); | |
2699 | } | |
1da177e4 LT |
2700 | #else |
2701 | #define dev_proc_init() 0 | |
2702 | #endif /* CONFIG_PROC_FS */ | |
2703 | ||
2704 | ||
2705 | /** | |
2706 | * netdev_set_master - set up master/slave pair | |
2707 | * @slave: slave device | |
2708 | * @master: new master device | |
2709 | * | |
2710 | * Changes the master device of the slave. Pass %NULL to break the | |
2711 | * bonding. The caller must hold the RTNL semaphore. On a failure | |
2712 | * a negative errno code is returned. On success the reference counts | |
2713 | * are adjusted, %RTM_NEWLINK is sent to the routing socket and the | |
2714 | * function returns zero. | |
2715 | */ | |
2716 | int netdev_set_master(struct net_device *slave, struct net_device *master) | |
2717 | { | |
2718 | struct net_device *old = slave->master; | |
2719 | ||
2720 | ASSERT_RTNL(); | |
2721 | ||
2722 | if (master) { | |
2723 | if (old) | |
2724 | return -EBUSY; | |
2725 | dev_hold(master); | |
2726 | } | |
2727 | ||
2728 | slave->master = master; | |
4ec93edb | 2729 | |
1da177e4 LT |
2730 | synchronize_net(); |
2731 | ||
2732 | if (old) | |
2733 | dev_put(old); | |
2734 | ||
2735 | if (master) | |
2736 | slave->flags |= IFF_SLAVE; | |
2737 | else | |
2738 | slave->flags &= ~IFF_SLAVE; | |
2739 | ||
2740 | rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE); | |
2741 | return 0; | |
2742 | } | |
2743 | ||
4417da66 | 2744 | static void __dev_set_promiscuity(struct net_device *dev, int inc) |
1da177e4 LT |
2745 | { |
2746 | unsigned short old_flags = dev->flags; | |
2747 | ||
24023451 PM |
2748 | ASSERT_RTNL(); |
2749 | ||
1da177e4 LT |
2750 | if ((dev->promiscuity += inc) == 0) |
2751 | dev->flags &= ~IFF_PROMISC; | |
52609c0b DC |
2752 | else |
2753 | dev->flags |= IFF_PROMISC; | |
2754 | if (dev->flags != old_flags) { | |
1da177e4 LT |
2755 | printk(KERN_INFO "device %s %s promiscuous mode\n", |
2756 | dev->name, (dev->flags & IFF_PROMISC) ? "entered" : | |
4ec93edb | 2757 | "left"); |
5bdb9886 SG |
2758 | audit_log(current->audit_context, GFP_ATOMIC, |
2759 | AUDIT_ANOM_PROMISCUOUS, | |
2760 | "dev=%s prom=%d old_prom=%d auid=%u", | |
2761 | dev->name, (dev->flags & IFF_PROMISC), | |
2762 | (old_flags & IFF_PROMISC), | |
4ec93edb | 2763 | audit_get_loginuid(current->audit_context)); |
24023451 PM |
2764 | |
2765 | if (dev->change_rx_flags) | |
2766 | dev->change_rx_flags(dev, IFF_PROMISC); | |
1da177e4 LT |
2767 | } |
2768 | } | |
2769 | ||
4417da66 PM |
2770 | /** |
2771 | * dev_set_promiscuity - update promiscuity count on a device | |
2772 | * @dev: device | |
2773 | * @inc: modifier | |
2774 | * | |
2775 | * Add or remove promiscuity from a device. While the count in the device | |
2776 | * remains above zero the interface remains promiscuous. Once it hits zero | |
2777 | * the device reverts back to normal filtering operation. A negative inc | |
2778 | * value is used to drop promiscuity on the device. | |
2779 | */ | |
2780 | void dev_set_promiscuity(struct net_device *dev, int inc) | |
2781 | { | |
2782 | unsigned short old_flags = dev->flags; | |
2783 | ||
2784 | __dev_set_promiscuity(dev, inc); | |
2785 | if (dev->flags != old_flags) | |
2786 | dev_set_rx_mode(dev); | |
2787 | } | |
2788 | ||
1da177e4 LT |
2789 | /** |
2790 | * dev_set_allmulti - update allmulti count on a device | |
2791 | * @dev: device | |
2792 | * @inc: modifier | |
2793 | * | |
2794 | * Add or remove reception of all multicast frames to a device. While the | |
2795 | * count in the device remains above zero the interface remains listening | |
2796 | * to all interfaces. Once it hits zero the device reverts back to normal | |
2797 | * filtering operation. A negative @inc value is used to drop the counter | |
2798 | * when releasing a resource needing all multicasts. | |
2799 | */ | |
2800 | ||
2801 | void dev_set_allmulti(struct net_device *dev, int inc) | |
2802 | { | |
2803 | unsigned short old_flags = dev->flags; | |
2804 | ||
24023451 PM |
2805 | ASSERT_RTNL(); |
2806 | ||
1da177e4 LT |
2807 | dev->flags |= IFF_ALLMULTI; |
2808 | if ((dev->allmulti += inc) == 0) | |
2809 | dev->flags &= ~IFF_ALLMULTI; | |
24023451 PM |
2810 | if (dev->flags ^ old_flags) { |
2811 | if (dev->change_rx_flags) | |
2812 | dev->change_rx_flags(dev, IFF_ALLMULTI); | |
4417da66 | 2813 | dev_set_rx_mode(dev); |
24023451 | 2814 | } |
4417da66 PM |
2815 | } |
2816 | ||
2817 | /* | |
2818 | * Upload unicast and multicast address lists to device and | |
2819 | * configure RX filtering. When the device doesn't support unicast | |
2820 | * filtering it is put in promiscous mode while unicast addresses | |
2821 | * are present. | |
2822 | */ | |
2823 | void __dev_set_rx_mode(struct net_device *dev) | |
2824 | { | |
2825 | /* dev_open will call this function so the list will stay sane. */ | |
2826 | if (!(dev->flags&IFF_UP)) | |
2827 | return; | |
2828 | ||
2829 | if (!netif_device_present(dev)) | |
40b77c94 | 2830 | return; |
4417da66 PM |
2831 | |
2832 | if (dev->set_rx_mode) | |
2833 | dev->set_rx_mode(dev); | |
2834 | else { | |
2835 | /* Unicast addresses changes may only happen under the rtnl, | |
2836 | * therefore calling __dev_set_promiscuity here is safe. | |
2837 | */ | |
2838 | if (dev->uc_count > 0 && !dev->uc_promisc) { | |
2839 | __dev_set_promiscuity(dev, 1); | |
2840 | dev->uc_promisc = 1; | |
2841 | } else if (dev->uc_count == 0 && dev->uc_promisc) { | |
2842 | __dev_set_promiscuity(dev, -1); | |
2843 | dev->uc_promisc = 0; | |
2844 | } | |
2845 | ||
2846 | if (dev->set_multicast_list) | |
2847 | dev->set_multicast_list(dev); | |
2848 | } | |
2849 | } | |
2850 | ||
2851 | void dev_set_rx_mode(struct net_device *dev) | |
2852 | { | |
2853 | netif_tx_lock_bh(dev); | |
2854 | __dev_set_rx_mode(dev); | |
2855 | netif_tx_unlock_bh(dev); | |
1da177e4 LT |
2856 | } |
2857 | ||
61cbc2fc PM |
2858 | int __dev_addr_delete(struct dev_addr_list **list, int *count, |
2859 | void *addr, int alen, int glbl) | |
bf742482 PM |
2860 | { |
2861 | struct dev_addr_list *da; | |
2862 | ||
2863 | for (; (da = *list) != NULL; list = &da->next) { | |
2864 | if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 && | |
2865 | alen == da->da_addrlen) { | |
2866 | if (glbl) { | |
2867 | int old_glbl = da->da_gusers; | |
2868 | da->da_gusers = 0; | |
2869 | if (old_glbl == 0) | |
2870 | break; | |
2871 | } | |
2872 | if (--da->da_users) | |
2873 | return 0; | |
2874 | ||
2875 | *list = da->next; | |
2876 | kfree(da); | |
61cbc2fc | 2877 | (*count)--; |
bf742482 PM |
2878 | return 0; |
2879 | } | |
2880 | } | |
2881 | return -ENOENT; | |
2882 | } | |
2883 | ||
61cbc2fc PM |
2884 | int __dev_addr_add(struct dev_addr_list **list, int *count, |
2885 | void *addr, int alen, int glbl) | |
bf742482 PM |
2886 | { |
2887 | struct dev_addr_list *da; | |
2888 | ||
2889 | for (da = *list; da != NULL; da = da->next) { | |
2890 | if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 && | |
2891 | da->da_addrlen == alen) { | |
2892 | if (glbl) { | |
2893 | int old_glbl = da->da_gusers; | |
2894 | da->da_gusers = 1; | |
2895 | if (old_glbl) | |
2896 | return 0; | |
2897 | } | |
2898 | da->da_users++; | |
2899 | return 0; | |
2900 | } | |
2901 | } | |
2902 | ||
2903 | da = kmalloc(sizeof(*da), GFP_ATOMIC); | |
2904 | if (da == NULL) | |
2905 | return -ENOMEM; | |
2906 | memcpy(da->da_addr, addr, alen); | |
2907 | da->da_addrlen = alen; | |
2908 | da->da_users = 1; | |
2909 | da->da_gusers = glbl ? 1 : 0; | |
2910 | da->next = *list; | |
2911 | *list = da; | |
61cbc2fc | 2912 | (*count)++; |
bf742482 PM |
2913 | return 0; |
2914 | } | |
2915 | ||
4417da66 PM |
2916 | /** |
2917 | * dev_unicast_delete - Release secondary unicast address. | |
2918 | * @dev: device | |
0ed72ec4 RD |
2919 | * @addr: address to delete |
2920 | * @alen: length of @addr | |
4417da66 PM |
2921 | * |
2922 | * Release reference to a secondary unicast address and remove it | |
0ed72ec4 | 2923 | * from the device if the reference count drops to zero. |
4417da66 PM |
2924 | * |
2925 | * The caller must hold the rtnl_mutex. | |
2926 | */ | |
2927 | int dev_unicast_delete(struct net_device *dev, void *addr, int alen) | |
2928 | { | |
2929 | int err; | |
2930 | ||
2931 | ASSERT_RTNL(); | |
2932 | ||
2933 | netif_tx_lock_bh(dev); | |
61cbc2fc PM |
2934 | err = __dev_addr_delete(&dev->uc_list, &dev->uc_count, addr, alen, 0); |
2935 | if (!err) | |
4417da66 | 2936 | __dev_set_rx_mode(dev); |
4417da66 PM |
2937 | netif_tx_unlock_bh(dev); |
2938 | return err; | |
2939 | } | |
2940 | EXPORT_SYMBOL(dev_unicast_delete); | |
2941 | ||
2942 | /** | |
2943 | * dev_unicast_add - add a secondary unicast address | |
2944 | * @dev: device | |
0ed72ec4 RD |
2945 | * @addr: address to delete |
2946 | * @alen: length of @addr | |
4417da66 PM |
2947 | * |
2948 | * Add a secondary unicast address to the device or increase | |
2949 | * the reference count if it already exists. | |
2950 | * | |
2951 | * The caller must hold the rtnl_mutex. | |
2952 | */ | |
2953 | int dev_unicast_add(struct net_device *dev, void *addr, int alen) | |
2954 | { | |
2955 | int err; | |
2956 | ||
2957 | ASSERT_RTNL(); | |
2958 | ||
2959 | netif_tx_lock_bh(dev); | |
61cbc2fc PM |
2960 | err = __dev_addr_add(&dev->uc_list, &dev->uc_count, addr, alen, 0); |
2961 | if (!err) | |
4417da66 | 2962 | __dev_set_rx_mode(dev); |
4417da66 PM |
2963 | netif_tx_unlock_bh(dev); |
2964 | return err; | |
2965 | } | |
2966 | EXPORT_SYMBOL(dev_unicast_add); | |
2967 | ||
12972621 DC |
2968 | static void __dev_addr_discard(struct dev_addr_list **list) |
2969 | { | |
2970 | struct dev_addr_list *tmp; | |
2971 | ||
2972 | while (*list != NULL) { | |
2973 | tmp = *list; | |
2974 | *list = tmp->next; | |
2975 | if (tmp->da_users > tmp->da_gusers) | |
2976 | printk("__dev_addr_discard: address leakage! " | |
2977 | "da_users=%d\n", tmp->da_users); | |
2978 | kfree(tmp); | |
2979 | } | |
2980 | } | |
2981 | ||
26cc2522 | 2982 | static void dev_addr_discard(struct net_device *dev) |
4417da66 PM |
2983 | { |
2984 | netif_tx_lock_bh(dev); | |
26cc2522 | 2985 | |
4417da66 PM |
2986 | __dev_addr_discard(&dev->uc_list); |
2987 | dev->uc_count = 0; | |
4417da66 | 2988 | |
456ad75c DC |
2989 | __dev_addr_discard(&dev->mc_list); |
2990 | dev->mc_count = 0; | |
26cc2522 | 2991 | |
456ad75c DC |
2992 | netif_tx_unlock_bh(dev); |
2993 | } | |
2994 | ||
1da177e4 LT |
2995 | unsigned dev_get_flags(const struct net_device *dev) |
2996 | { | |
2997 | unsigned flags; | |
2998 | ||
2999 | flags = (dev->flags & ~(IFF_PROMISC | | |
3000 | IFF_ALLMULTI | | |
b00055aa SR |
3001 | IFF_RUNNING | |
3002 | IFF_LOWER_UP | | |
3003 | IFF_DORMANT)) | | |
1da177e4 LT |
3004 | (dev->gflags & (IFF_PROMISC | |
3005 | IFF_ALLMULTI)); | |
3006 | ||
b00055aa SR |
3007 | if (netif_running(dev)) { |
3008 | if (netif_oper_up(dev)) | |
3009 | flags |= IFF_RUNNING; | |
3010 | if (netif_carrier_ok(dev)) | |
3011 | flags |= IFF_LOWER_UP; | |
3012 | if (netif_dormant(dev)) | |
3013 | flags |= IFF_DORMANT; | |
3014 | } | |
1da177e4 LT |
3015 | |
3016 | return flags; | |
3017 | } | |
3018 | ||
3019 | int dev_change_flags(struct net_device *dev, unsigned flags) | |
3020 | { | |
7c355f53 | 3021 | int ret, changes; |
1da177e4 LT |
3022 | int old_flags = dev->flags; |
3023 | ||
24023451 PM |
3024 | ASSERT_RTNL(); |
3025 | ||
1da177e4 LT |
3026 | /* |
3027 | * Set the flags on our device. | |
3028 | */ | |
3029 | ||
3030 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
3031 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
3032 | IFF_AUTOMEDIA)) | | |
3033 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
3034 | IFF_ALLMULTI)); | |
3035 | ||
3036 | /* | |
3037 | * Load in the correct multicast list now the flags have changed. | |
3038 | */ | |
3039 | ||
24023451 PM |
3040 | if (dev->change_rx_flags && (dev->flags ^ flags) & IFF_MULTICAST) |
3041 | dev->change_rx_flags(dev, IFF_MULTICAST); | |
3042 | ||
4417da66 | 3043 | dev_set_rx_mode(dev); |
1da177e4 LT |
3044 | |
3045 | /* | |
3046 | * Have we downed the interface. We handle IFF_UP ourselves | |
3047 | * according to user attempts to set it, rather than blindly | |
3048 | * setting it. | |
3049 | */ | |
3050 | ||
3051 | ret = 0; | |
3052 | if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */ | |
3053 | ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev); | |
3054 | ||
3055 | if (!ret) | |
4417da66 | 3056 | dev_set_rx_mode(dev); |
1da177e4 LT |
3057 | } |
3058 | ||
3059 | if (dev->flags & IFF_UP && | |
3060 | ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI | | |
3061 | IFF_VOLATILE))) | |
056925ab | 3062 | call_netdevice_notifiers(NETDEV_CHANGE, dev); |
1da177e4 LT |
3063 | |
3064 | if ((flags ^ dev->gflags) & IFF_PROMISC) { | |
3065 | int inc = (flags & IFF_PROMISC) ? +1 : -1; | |
3066 | dev->gflags ^= IFF_PROMISC; | |
3067 | dev_set_promiscuity(dev, inc); | |
3068 | } | |
3069 | ||
3070 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
3071 | is important. Some (broken) drivers set IFF_PROMISC, when | |
3072 | IFF_ALLMULTI is requested not asking us and not reporting. | |
3073 | */ | |
3074 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
3075 | int inc = (flags & IFF_ALLMULTI) ? +1 : -1; | |
3076 | dev->gflags ^= IFF_ALLMULTI; | |
3077 | dev_set_allmulti(dev, inc); | |
3078 | } | |
3079 | ||
7c355f53 TG |
3080 | /* Exclude state transition flags, already notified */ |
3081 | changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING); | |
3082 | if (changes) | |
3083 | rtmsg_ifinfo(RTM_NEWLINK, dev, changes); | |
1da177e4 LT |
3084 | |
3085 | return ret; | |
3086 | } | |
3087 | ||
3088 | int dev_set_mtu(struct net_device *dev, int new_mtu) | |
3089 | { | |
3090 | int err; | |
3091 | ||
3092 | if (new_mtu == dev->mtu) | |
3093 | return 0; | |
3094 | ||
3095 | /* MTU must be positive. */ | |
3096 | if (new_mtu < 0) | |
3097 | return -EINVAL; | |
3098 | ||
3099 | if (!netif_device_present(dev)) | |
3100 | return -ENODEV; | |
3101 | ||
3102 | err = 0; | |
3103 | if (dev->change_mtu) | |
3104 | err = dev->change_mtu(dev, new_mtu); | |
3105 | else | |
3106 | dev->mtu = new_mtu; | |
3107 | if (!err && dev->flags & IFF_UP) | |
056925ab | 3108 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); |
1da177e4 LT |
3109 | return err; |
3110 | } | |
3111 | ||
3112 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) | |
3113 | { | |
3114 | int err; | |
3115 | ||
3116 | if (!dev->set_mac_address) | |
3117 | return -EOPNOTSUPP; | |
3118 | if (sa->sa_family != dev->type) | |
3119 | return -EINVAL; | |
3120 | if (!netif_device_present(dev)) | |
3121 | return -ENODEV; | |
3122 | err = dev->set_mac_address(dev, sa); | |
3123 | if (!err) | |
056925ab | 3124 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
1da177e4 LT |
3125 | return err; |
3126 | } | |
3127 | ||
3128 | /* | |
14e3e079 | 3129 | * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock) |
1da177e4 | 3130 | */ |
14e3e079 | 3131 | static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd) |
1da177e4 LT |
3132 | { |
3133 | int err; | |
881d966b | 3134 | struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name); |
1da177e4 LT |
3135 | |
3136 | if (!dev) | |
3137 | return -ENODEV; | |
3138 | ||
3139 | switch (cmd) { | |
3140 | case SIOCGIFFLAGS: /* Get interface flags */ | |
3141 | ifr->ifr_flags = dev_get_flags(dev); | |
3142 | return 0; | |
3143 | ||
1da177e4 LT |
3144 | case SIOCGIFMETRIC: /* Get the metric on the interface |
3145 | (currently unused) */ | |
3146 | ifr->ifr_metric = 0; | |
3147 | return 0; | |
3148 | ||
1da177e4 LT |
3149 | case SIOCGIFMTU: /* Get the MTU of a device */ |
3150 | ifr->ifr_mtu = dev->mtu; | |
3151 | return 0; | |
3152 | ||
1da177e4 LT |
3153 | case SIOCGIFHWADDR: |
3154 | if (!dev->addr_len) | |
3155 | memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data); | |
3156 | else | |
3157 | memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr, | |
3158 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
3159 | ifr->ifr_hwaddr.sa_family = dev->type; | |
3160 | return 0; | |
3161 | ||
14e3e079 JG |
3162 | case SIOCGIFSLAVE: |
3163 | err = -EINVAL; | |
3164 | break; | |
3165 | ||
3166 | case SIOCGIFMAP: | |
3167 | ifr->ifr_map.mem_start = dev->mem_start; | |
3168 | ifr->ifr_map.mem_end = dev->mem_end; | |
3169 | ifr->ifr_map.base_addr = dev->base_addr; | |
3170 | ifr->ifr_map.irq = dev->irq; | |
3171 | ifr->ifr_map.dma = dev->dma; | |
3172 | ifr->ifr_map.port = dev->if_port; | |
3173 | return 0; | |
3174 | ||
3175 | case SIOCGIFINDEX: | |
3176 | ifr->ifr_ifindex = dev->ifindex; | |
3177 | return 0; | |
3178 | ||
3179 | case SIOCGIFTXQLEN: | |
3180 | ifr->ifr_qlen = dev->tx_queue_len; | |
3181 | return 0; | |
3182 | ||
3183 | default: | |
3184 | /* dev_ioctl() should ensure this case | |
3185 | * is never reached | |
3186 | */ | |
3187 | WARN_ON(1); | |
3188 | err = -EINVAL; | |
3189 | break; | |
3190 | ||
3191 | } | |
3192 | return err; | |
3193 | } | |
3194 | ||
3195 | /* | |
3196 | * Perform the SIOCxIFxxx calls, inside rtnl_lock() | |
3197 | */ | |
3198 | static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd) | |
3199 | { | |
3200 | int err; | |
3201 | struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name); | |
3202 | ||
3203 | if (!dev) | |
3204 | return -ENODEV; | |
3205 | ||
3206 | switch (cmd) { | |
3207 | case SIOCSIFFLAGS: /* Set interface flags */ | |
3208 | return dev_change_flags(dev, ifr->ifr_flags); | |
3209 | ||
3210 | case SIOCSIFMETRIC: /* Set the metric on the interface | |
3211 | (currently unused) */ | |
3212 | return -EOPNOTSUPP; | |
3213 | ||
3214 | case SIOCSIFMTU: /* Set the MTU of a device */ | |
3215 | return dev_set_mtu(dev, ifr->ifr_mtu); | |
3216 | ||
1da177e4 LT |
3217 | case SIOCSIFHWADDR: |
3218 | return dev_set_mac_address(dev, &ifr->ifr_hwaddr); | |
3219 | ||
3220 | case SIOCSIFHWBROADCAST: | |
3221 | if (ifr->ifr_hwaddr.sa_family != dev->type) | |
3222 | return -EINVAL; | |
3223 | memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data, | |
3224 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
056925ab | 3225 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
1da177e4 LT |
3226 | return 0; |
3227 | ||
1da177e4 LT |
3228 | case SIOCSIFMAP: |
3229 | if (dev->set_config) { | |
3230 | if (!netif_device_present(dev)) | |
3231 | return -ENODEV; | |
3232 | return dev->set_config(dev, &ifr->ifr_map); | |
3233 | } | |
3234 | return -EOPNOTSUPP; | |
3235 | ||
3236 | case SIOCADDMULTI: | |
3237 | if (!dev->set_multicast_list || | |
3238 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) | |
3239 | return -EINVAL; | |
3240 | if (!netif_device_present(dev)) | |
3241 | return -ENODEV; | |
3242 | return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data, | |
3243 | dev->addr_len, 1); | |
3244 | ||
3245 | case SIOCDELMULTI: | |
3246 | if (!dev->set_multicast_list || | |
3247 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) | |
3248 | return -EINVAL; | |
3249 | if (!netif_device_present(dev)) | |
3250 | return -ENODEV; | |
3251 | return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data, | |
3252 | dev->addr_len, 1); | |
3253 | ||
1da177e4 LT |
3254 | case SIOCSIFTXQLEN: |
3255 | if (ifr->ifr_qlen < 0) | |
3256 | return -EINVAL; | |
3257 | dev->tx_queue_len = ifr->ifr_qlen; | |
3258 | return 0; | |
3259 | ||
3260 | case SIOCSIFNAME: | |
3261 | ifr->ifr_newname[IFNAMSIZ-1] = '\0'; | |
3262 | return dev_change_name(dev, ifr->ifr_newname); | |
3263 | ||
3264 | /* | |
3265 | * Unknown or private ioctl | |
3266 | */ | |
3267 | ||
3268 | default: | |
3269 | if ((cmd >= SIOCDEVPRIVATE && | |
3270 | cmd <= SIOCDEVPRIVATE + 15) || | |
3271 | cmd == SIOCBONDENSLAVE || | |
3272 | cmd == SIOCBONDRELEASE || | |
3273 | cmd == SIOCBONDSETHWADDR || | |
3274 | cmd == SIOCBONDSLAVEINFOQUERY || | |
3275 | cmd == SIOCBONDINFOQUERY || | |
3276 | cmd == SIOCBONDCHANGEACTIVE || | |
3277 | cmd == SIOCGMIIPHY || | |
3278 | cmd == SIOCGMIIREG || | |
3279 | cmd == SIOCSMIIREG || | |
3280 | cmd == SIOCBRADDIF || | |
3281 | cmd == SIOCBRDELIF || | |
3282 | cmd == SIOCWANDEV) { | |
3283 | err = -EOPNOTSUPP; | |
3284 | if (dev->do_ioctl) { | |
3285 | if (netif_device_present(dev)) | |
3286 | err = dev->do_ioctl(dev, ifr, | |
3287 | cmd); | |
3288 | else | |
3289 | err = -ENODEV; | |
3290 | } | |
3291 | } else | |
3292 | err = -EINVAL; | |
3293 | ||
3294 | } | |
3295 | return err; | |
3296 | } | |
3297 | ||
3298 | /* | |
3299 | * This function handles all "interface"-type I/O control requests. The actual | |
3300 | * 'doing' part of this is dev_ifsioc above. | |
3301 | */ | |
3302 | ||
3303 | /** | |
3304 | * dev_ioctl - network device ioctl | |
c4ea43c5 | 3305 | * @net: the applicable net namespace |
1da177e4 LT |
3306 | * @cmd: command to issue |
3307 | * @arg: pointer to a struct ifreq in user space | |
3308 | * | |
3309 | * Issue ioctl functions to devices. This is normally called by the | |
3310 | * user space syscall interfaces but can sometimes be useful for | |
3311 | * other purposes. The return value is the return from the syscall if | |
3312 | * positive or a negative errno code on error. | |
3313 | */ | |
3314 | ||
881d966b | 3315 | int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg) |
1da177e4 LT |
3316 | { |
3317 | struct ifreq ifr; | |
3318 | int ret; | |
3319 | char *colon; | |
3320 | ||
3321 | /* One special case: SIOCGIFCONF takes ifconf argument | |
3322 | and requires shared lock, because it sleeps writing | |
3323 | to user space. | |
3324 | */ | |
3325 | ||
3326 | if (cmd == SIOCGIFCONF) { | |
6756ae4b | 3327 | rtnl_lock(); |
881d966b | 3328 | ret = dev_ifconf(net, (char __user *) arg); |
6756ae4b | 3329 | rtnl_unlock(); |
1da177e4 LT |
3330 | return ret; |
3331 | } | |
3332 | if (cmd == SIOCGIFNAME) | |
881d966b | 3333 | return dev_ifname(net, (struct ifreq __user *)arg); |
1da177e4 LT |
3334 | |
3335 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
3336 | return -EFAULT; | |
3337 | ||
3338 | ifr.ifr_name[IFNAMSIZ-1] = 0; | |
3339 | ||
3340 | colon = strchr(ifr.ifr_name, ':'); | |
3341 | if (colon) | |
3342 | *colon = 0; | |
3343 | ||
3344 | /* | |
3345 | * See which interface the caller is talking about. | |
3346 | */ | |
3347 | ||
3348 | switch (cmd) { | |
3349 | /* | |
3350 | * These ioctl calls: | |
3351 | * - can be done by all. | |
3352 | * - atomic and do not require locking. | |
3353 | * - return a value | |
3354 | */ | |
3355 | case SIOCGIFFLAGS: | |
3356 | case SIOCGIFMETRIC: | |
3357 | case SIOCGIFMTU: | |
3358 | case SIOCGIFHWADDR: | |
3359 | case SIOCGIFSLAVE: | |
3360 | case SIOCGIFMAP: | |
3361 | case SIOCGIFINDEX: | |
3362 | case SIOCGIFTXQLEN: | |
881d966b | 3363 | dev_load(net, ifr.ifr_name); |
1da177e4 | 3364 | read_lock(&dev_base_lock); |
14e3e079 | 3365 | ret = dev_ifsioc_locked(net, &ifr, cmd); |
1da177e4 LT |
3366 | read_unlock(&dev_base_lock); |
3367 | if (!ret) { | |
3368 | if (colon) | |
3369 | *colon = ':'; | |
3370 | if (copy_to_user(arg, &ifr, | |
3371 | sizeof(struct ifreq))) | |
3372 | ret = -EFAULT; | |
3373 | } | |
3374 | return ret; | |
3375 | ||
3376 | case SIOCETHTOOL: | |
881d966b | 3377 | dev_load(net, ifr.ifr_name); |
1da177e4 | 3378 | rtnl_lock(); |
881d966b | 3379 | ret = dev_ethtool(net, &ifr); |
1da177e4 LT |
3380 | rtnl_unlock(); |
3381 | if (!ret) { | |
3382 | if (colon) | |
3383 | *colon = ':'; | |
3384 | if (copy_to_user(arg, &ifr, | |
3385 | sizeof(struct ifreq))) | |
3386 | ret = -EFAULT; | |
3387 | } | |
3388 | return ret; | |
3389 | ||
3390 | /* | |
3391 | * These ioctl calls: | |
3392 | * - require superuser power. | |
3393 | * - require strict serialization. | |
3394 | * - return a value | |
3395 | */ | |
3396 | case SIOCGMIIPHY: | |
3397 | case SIOCGMIIREG: | |
3398 | case SIOCSIFNAME: | |
3399 | if (!capable(CAP_NET_ADMIN)) | |
3400 | return -EPERM; | |
881d966b | 3401 | dev_load(net, ifr.ifr_name); |
1da177e4 | 3402 | rtnl_lock(); |
881d966b | 3403 | ret = dev_ifsioc(net, &ifr, cmd); |
1da177e4 LT |
3404 | rtnl_unlock(); |
3405 | if (!ret) { | |
3406 | if (colon) | |
3407 | *colon = ':'; | |
3408 | if (copy_to_user(arg, &ifr, | |
3409 | sizeof(struct ifreq))) | |
3410 | ret = -EFAULT; | |
3411 | } | |
3412 | return ret; | |
3413 | ||
3414 | /* | |
3415 | * These ioctl calls: | |
3416 | * - require superuser power. | |
3417 | * - require strict serialization. | |
3418 | * - do not return a value | |
3419 | */ | |
3420 | case SIOCSIFFLAGS: | |
3421 | case SIOCSIFMETRIC: | |
3422 | case SIOCSIFMTU: | |
3423 | case SIOCSIFMAP: | |
3424 | case SIOCSIFHWADDR: | |
3425 | case SIOCSIFSLAVE: | |
3426 | case SIOCADDMULTI: | |
3427 | case SIOCDELMULTI: | |
3428 | case SIOCSIFHWBROADCAST: | |
3429 | case SIOCSIFTXQLEN: | |
3430 | case SIOCSMIIREG: | |
3431 | case SIOCBONDENSLAVE: | |
3432 | case SIOCBONDRELEASE: | |
3433 | case SIOCBONDSETHWADDR: | |
1da177e4 LT |
3434 | case SIOCBONDCHANGEACTIVE: |
3435 | case SIOCBRADDIF: | |
3436 | case SIOCBRDELIF: | |
3437 | if (!capable(CAP_NET_ADMIN)) | |
3438 | return -EPERM; | |
cabcac0b TG |
3439 | /* fall through */ |
3440 | case SIOCBONDSLAVEINFOQUERY: | |
3441 | case SIOCBONDINFOQUERY: | |
881d966b | 3442 | dev_load(net, ifr.ifr_name); |
1da177e4 | 3443 | rtnl_lock(); |
881d966b | 3444 | ret = dev_ifsioc(net, &ifr, cmd); |
1da177e4 LT |
3445 | rtnl_unlock(); |
3446 | return ret; | |
3447 | ||
3448 | case SIOCGIFMEM: | |
3449 | /* Get the per device memory space. We can add this but | |
3450 | * currently do not support it */ | |
3451 | case SIOCSIFMEM: | |
3452 | /* Set the per device memory buffer space. | |
3453 | * Not applicable in our case */ | |
3454 | case SIOCSIFLINK: | |
3455 | return -EINVAL; | |
3456 | ||
3457 | /* | |
3458 | * Unknown or private ioctl. | |
3459 | */ | |
3460 | default: | |
3461 | if (cmd == SIOCWANDEV || | |
3462 | (cmd >= SIOCDEVPRIVATE && | |
3463 | cmd <= SIOCDEVPRIVATE + 15)) { | |
881d966b | 3464 | dev_load(net, ifr.ifr_name); |
1da177e4 | 3465 | rtnl_lock(); |
881d966b | 3466 | ret = dev_ifsioc(net, &ifr, cmd); |
1da177e4 LT |
3467 | rtnl_unlock(); |
3468 | if (!ret && copy_to_user(arg, &ifr, | |
3469 | sizeof(struct ifreq))) | |
3470 | ret = -EFAULT; | |
3471 | return ret; | |
3472 | } | |
1da177e4 | 3473 | /* Take care of Wireless Extensions */ |
295f4a1f | 3474 | if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) |
881d966b | 3475 | return wext_handle_ioctl(net, &ifr, cmd, arg); |
1da177e4 LT |
3476 | return -EINVAL; |
3477 | } | |
3478 | } | |
3479 | ||
3480 | ||
3481 | /** | |
3482 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 3483 | * @net: the applicable net namespace |
1da177e4 LT |
3484 | * |
3485 | * Returns a suitable unique value for a new device interface | |
3486 | * number. The caller must hold the rtnl semaphore or the | |
3487 | * dev_base_lock to be sure it remains unique. | |
3488 | */ | |
881d966b | 3489 | static int dev_new_index(struct net *net) |
1da177e4 LT |
3490 | { |
3491 | static int ifindex; | |
3492 | for (;;) { | |
3493 | if (++ifindex <= 0) | |
3494 | ifindex = 1; | |
881d966b | 3495 | if (!__dev_get_by_index(net, ifindex)) |
1da177e4 LT |
3496 | return ifindex; |
3497 | } | |
3498 | } | |
3499 | ||
1da177e4 LT |
3500 | /* Delayed registration/unregisteration */ |
3501 | static DEFINE_SPINLOCK(net_todo_list_lock); | |
3502 | static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list); | |
3503 | ||
6f05f629 | 3504 | static void net_set_todo(struct net_device *dev) |
1da177e4 LT |
3505 | { |
3506 | spin_lock(&net_todo_list_lock); | |
3507 | list_add_tail(&dev->todo_list, &net_todo_list); | |
3508 | spin_unlock(&net_todo_list_lock); | |
3509 | } | |
3510 | ||
93ee31f1 DL |
3511 | static void rollback_registered(struct net_device *dev) |
3512 | { | |
3513 | BUG_ON(dev_boot_phase); | |
3514 | ASSERT_RTNL(); | |
3515 | ||
3516 | /* Some devices call without registering for initialization unwind. */ | |
3517 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
3518 | printk(KERN_DEBUG "unregister_netdevice: device %s/%p never " | |
3519 | "was registered\n", dev->name, dev); | |
3520 | ||
3521 | WARN_ON(1); | |
3522 | return; | |
3523 | } | |
3524 | ||
3525 | BUG_ON(dev->reg_state != NETREG_REGISTERED); | |
3526 | ||
3527 | /* If device is running, close it first. */ | |
3528 | dev_close(dev); | |
3529 | ||
3530 | /* And unlink it from device chain. */ | |
3531 | unlist_netdevice(dev); | |
3532 | ||
3533 | dev->reg_state = NETREG_UNREGISTERING; | |
3534 | ||
3535 | synchronize_net(); | |
3536 | ||
3537 | /* Shutdown queueing discipline. */ | |
3538 | dev_shutdown(dev); | |
3539 | ||
3540 | ||
3541 | /* Notify protocols, that we are about to destroy | |
3542 | this device. They should clean all the things. | |
3543 | */ | |
3544 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
3545 | ||
3546 | /* | |
3547 | * Flush the unicast and multicast chains | |
3548 | */ | |
3549 | dev_addr_discard(dev); | |
3550 | ||
3551 | if (dev->uninit) | |
3552 | dev->uninit(dev); | |
3553 | ||
3554 | /* Notifier chain MUST detach us from master device. */ | |
3555 | BUG_TRAP(!dev->master); | |
3556 | ||
3557 | /* Remove entries from kobject tree */ | |
3558 | netdev_unregister_kobject(dev); | |
3559 | ||
3560 | synchronize_net(); | |
3561 | ||
3562 | dev_put(dev); | |
3563 | } | |
3564 | ||
1da177e4 LT |
3565 | /** |
3566 | * register_netdevice - register a network device | |
3567 | * @dev: device to register | |
3568 | * | |
3569 | * Take a completed network device structure and add it to the kernel | |
3570 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
3571 | * chain. 0 is returned on success. A negative errno code is returned | |
3572 | * on a failure to set up the device, or if the name is a duplicate. | |
3573 | * | |
3574 | * Callers must hold the rtnl semaphore. You may want | |
3575 | * register_netdev() instead of this. | |
3576 | * | |
3577 | * BUGS: | |
3578 | * The locking appears insufficient to guarantee two parallel registers | |
3579 | * will not get the same name. | |
3580 | */ | |
3581 | ||
3582 | int register_netdevice(struct net_device *dev) | |
3583 | { | |
3584 | struct hlist_head *head; | |
3585 | struct hlist_node *p; | |
3586 | int ret; | |
881d966b | 3587 | struct net *net; |
1da177e4 LT |
3588 | |
3589 | BUG_ON(dev_boot_phase); | |
3590 | ASSERT_RTNL(); | |
3591 | ||
b17a7c17 SH |
3592 | might_sleep(); |
3593 | ||
1da177e4 LT |
3594 | /* When net_device's are persistent, this will be fatal. */ |
3595 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
881d966b EB |
3596 | BUG_ON(!dev->nd_net); |
3597 | net = dev->nd_net; | |
1da177e4 LT |
3598 | |
3599 | spin_lock_init(&dev->queue_lock); | |
932ff279 | 3600 | spin_lock_init(&dev->_xmit_lock); |
723e98b7 | 3601 | netdev_set_lockdep_class(&dev->_xmit_lock, dev->type); |
1da177e4 | 3602 | dev->xmit_lock_owner = -1; |
1da177e4 | 3603 | spin_lock_init(&dev->ingress_lock); |
1da177e4 | 3604 | |
1da177e4 LT |
3605 | dev->iflink = -1; |
3606 | ||
3607 | /* Init, if this function is available */ | |
3608 | if (dev->init) { | |
3609 | ret = dev->init(dev); | |
3610 | if (ret) { | |
3611 | if (ret > 0) | |
3612 | ret = -EIO; | |
90833aa4 | 3613 | goto out; |
1da177e4 LT |
3614 | } |
3615 | } | |
4ec93edb | 3616 | |
1da177e4 LT |
3617 | if (!dev_valid_name(dev->name)) { |
3618 | ret = -EINVAL; | |
7ce1b0ed | 3619 | goto err_uninit; |
1da177e4 LT |
3620 | } |
3621 | ||
881d966b | 3622 | dev->ifindex = dev_new_index(net); |
1da177e4 LT |
3623 | if (dev->iflink == -1) |
3624 | dev->iflink = dev->ifindex; | |
3625 | ||
3626 | /* Check for existence of name */ | |
881d966b | 3627 | head = dev_name_hash(net, dev->name); |
1da177e4 LT |
3628 | hlist_for_each(p, head) { |
3629 | struct net_device *d | |
3630 | = hlist_entry(p, struct net_device, name_hlist); | |
3631 | if (!strncmp(d->name, dev->name, IFNAMSIZ)) { | |
3632 | ret = -EEXIST; | |
7ce1b0ed | 3633 | goto err_uninit; |
1da177e4 | 3634 | } |
4ec93edb | 3635 | } |
1da177e4 | 3636 | |
d212f87b SH |
3637 | /* Fix illegal checksum combinations */ |
3638 | if ((dev->features & NETIF_F_HW_CSUM) && | |
3639 | (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
3640 | printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n", | |
3641 | dev->name); | |
3642 | dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); | |
3643 | } | |
3644 | ||
3645 | if ((dev->features & NETIF_F_NO_CSUM) && | |
3646 | (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
3647 | printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n", | |
3648 | dev->name); | |
3649 | dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM); | |
3650 | } | |
3651 | ||
3652 | ||
1da177e4 LT |
3653 | /* Fix illegal SG+CSUM combinations. */ |
3654 | if ((dev->features & NETIF_F_SG) && | |
8648b305 | 3655 | !(dev->features & NETIF_F_ALL_CSUM)) { |
5a8da02b | 3656 | printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no checksum feature.\n", |
1da177e4 LT |
3657 | dev->name); |
3658 | dev->features &= ~NETIF_F_SG; | |
3659 | } | |
3660 | ||
3661 | /* TSO requires that SG is present as well. */ | |
3662 | if ((dev->features & NETIF_F_TSO) && | |
3663 | !(dev->features & NETIF_F_SG)) { | |
5a8da02b | 3664 | printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no SG feature.\n", |
1da177e4 LT |
3665 | dev->name); |
3666 | dev->features &= ~NETIF_F_TSO; | |
3667 | } | |
e89e9cf5 AR |
3668 | if (dev->features & NETIF_F_UFO) { |
3669 | if (!(dev->features & NETIF_F_HW_CSUM)) { | |
3670 | printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no " | |
3671 | "NETIF_F_HW_CSUM feature.\n", | |
3672 | dev->name); | |
3673 | dev->features &= ~NETIF_F_UFO; | |
3674 | } | |
3675 | if (!(dev->features & NETIF_F_SG)) { | |
3676 | printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no " | |
3677 | "NETIF_F_SG feature.\n", | |
3678 | dev->name); | |
3679 | dev->features &= ~NETIF_F_UFO; | |
3680 | } | |
3681 | } | |
1da177e4 | 3682 | |
8b41d188 | 3683 | ret = netdev_register_kobject(dev); |
b17a7c17 | 3684 | if (ret) |
7ce1b0ed | 3685 | goto err_uninit; |
b17a7c17 SH |
3686 | dev->reg_state = NETREG_REGISTERED; |
3687 | ||
1da177e4 LT |
3688 | /* |
3689 | * Default initial state at registry is that the | |
3690 | * device is present. | |
3691 | */ | |
3692 | ||
3693 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
3694 | ||
1da177e4 | 3695 | dev_init_scheduler(dev); |
1da177e4 | 3696 | dev_hold(dev); |
ce286d32 | 3697 | list_netdevice(dev); |
1da177e4 LT |
3698 | |
3699 | /* Notify protocols, that a new device appeared. */ | |
056925ab | 3700 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 3701 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
3702 | if (ret) { |
3703 | rollback_registered(dev); | |
3704 | dev->reg_state = NETREG_UNREGISTERED; | |
3705 | } | |
1da177e4 LT |
3706 | |
3707 | out: | |
3708 | return ret; | |
7ce1b0ed HX |
3709 | |
3710 | err_uninit: | |
3711 | if (dev->uninit) | |
3712 | dev->uninit(dev); | |
3713 | goto out; | |
1da177e4 LT |
3714 | } |
3715 | ||
3716 | /** | |
3717 | * register_netdev - register a network device | |
3718 | * @dev: device to register | |
3719 | * | |
3720 | * Take a completed network device structure and add it to the kernel | |
3721 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
3722 | * chain. 0 is returned on success. A negative errno code is returned | |
3723 | * on a failure to set up the device, or if the name is a duplicate. | |
3724 | * | |
38b4da38 | 3725 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
3726 | * and expands the device name if you passed a format string to |
3727 | * alloc_netdev. | |
3728 | */ | |
3729 | int register_netdev(struct net_device *dev) | |
3730 | { | |
3731 | int err; | |
3732 | ||
3733 | rtnl_lock(); | |
3734 | ||
3735 | /* | |
3736 | * If the name is a format string the caller wants us to do a | |
3737 | * name allocation. | |
3738 | */ | |
3739 | if (strchr(dev->name, '%')) { | |
3740 | err = dev_alloc_name(dev, dev->name); | |
3741 | if (err < 0) | |
3742 | goto out; | |
3743 | } | |
4ec93edb | 3744 | |
1da177e4 LT |
3745 | err = register_netdevice(dev); |
3746 | out: | |
3747 | rtnl_unlock(); | |
3748 | return err; | |
3749 | } | |
3750 | EXPORT_SYMBOL(register_netdev); | |
3751 | ||
3752 | /* | |
3753 | * netdev_wait_allrefs - wait until all references are gone. | |
3754 | * | |
3755 | * This is called when unregistering network devices. | |
3756 | * | |
3757 | * Any protocol or device that holds a reference should register | |
3758 | * for netdevice notification, and cleanup and put back the | |
3759 | * reference if they receive an UNREGISTER event. | |
3760 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 3761 | * call dev_put. |
1da177e4 LT |
3762 | */ |
3763 | static void netdev_wait_allrefs(struct net_device *dev) | |
3764 | { | |
3765 | unsigned long rebroadcast_time, warning_time; | |
3766 | ||
3767 | rebroadcast_time = warning_time = jiffies; | |
3768 | while (atomic_read(&dev->refcnt) != 0) { | |
3769 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { | |
6756ae4b | 3770 | rtnl_lock(); |
1da177e4 LT |
3771 | |
3772 | /* Rebroadcast unregister notification */ | |
056925ab | 3773 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
1da177e4 LT |
3774 | |
3775 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, | |
3776 | &dev->state)) { | |
3777 | /* We must not have linkwatch events | |
3778 | * pending on unregister. If this | |
3779 | * happens, we simply run the queue | |
3780 | * unscheduled, resulting in a noop | |
3781 | * for this device. | |
3782 | */ | |
3783 | linkwatch_run_queue(); | |
3784 | } | |
3785 | ||
6756ae4b | 3786 | __rtnl_unlock(); |
1da177e4 LT |
3787 | |
3788 | rebroadcast_time = jiffies; | |
3789 | } | |
3790 | ||
3791 | msleep(250); | |
3792 | ||
3793 | if (time_after(jiffies, warning_time + 10 * HZ)) { | |
3794 | printk(KERN_EMERG "unregister_netdevice: " | |
3795 | "waiting for %s to become free. Usage " | |
3796 | "count = %d\n", | |
3797 | dev->name, atomic_read(&dev->refcnt)); | |
3798 | warning_time = jiffies; | |
3799 | } | |
3800 | } | |
3801 | } | |
3802 | ||
3803 | /* The sequence is: | |
3804 | * | |
3805 | * rtnl_lock(); | |
3806 | * ... | |
3807 | * register_netdevice(x1); | |
3808 | * register_netdevice(x2); | |
3809 | * ... | |
3810 | * unregister_netdevice(y1); | |
3811 | * unregister_netdevice(y2); | |
3812 | * ... | |
3813 | * rtnl_unlock(); | |
3814 | * free_netdev(y1); | |
3815 | * free_netdev(y2); | |
3816 | * | |
3817 | * We are invoked by rtnl_unlock() after it drops the semaphore. | |
3818 | * This allows us to deal with problems: | |
b17a7c17 | 3819 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
3820 | * without deadlocking with linkwatch via keventd. |
3821 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
3822 | * safely in order to wait for the netdev refcnt to drop to zero. | |
3823 | */ | |
4a3e2f71 | 3824 | static DEFINE_MUTEX(net_todo_run_mutex); |
1da177e4 LT |
3825 | void netdev_run_todo(void) |
3826 | { | |
626ab0e6 | 3827 | struct list_head list; |
1da177e4 LT |
3828 | |
3829 | /* Need to guard against multiple cpu's getting out of order. */ | |
4a3e2f71 | 3830 | mutex_lock(&net_todo_run_mutex); |
1da177e4 LT |
3831 | |
3832 | /* Not safe to do outside the semaphore. We must not return | |
3833 | * until all unregister events invoked by the local processor | |
3834 | * have been completed (either by this todo run, or one on | |
3835 | * another cpu). | |
3836 | */ | |
3837 | if (list_empty(&net_todo_list)) | |
3838 | goto out; | |
3839 | ||
3840 | /* Snapshot list, allow later requests */ | |
3841 | spin_lock(&net_todo_list_lock); | |
626ab0e6 | 3842 | list_replace_init(&net_todo_list, &list); |
1da177e4 | 3843 | spin_unlock(&net_todo_list_lock); |
626ab0e6 | 3844 | |
1da177e4 LT |
3845 | while (!list_empty(&list)) { |
3846 | struct net_device *dev | |
3847 | = list_entry(list.next, struct net_device, todo_list); | |
3848 | list_del(&dev->todo_list); | |
3849 | ||
b17a7c17 SH |
3850 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
3851 | printk(KERN_ERR "network todo '%s' but state %d\n", | |
3852 | dev->name, dev->reg_state); | |
3853 | dump_stack(); | |
3854 | continue; | |
3855 | } | |
1da177e4 | 3856 | |
b17a7c17 | 3857 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 3858 | |
b17a7c17 | 3859 | netdev_wait_allrefs(dev); |
1da177e4 | 3860 | |
b17a7c17 SH |
3861 | /* paranoia */ |
3862 | BUG_ON(atomic_read(&dev->refcnt)); | |
3863 | BUG_TRAP(!dev->ip_ptr); | |
3864 | BUG_TRAP(!dev->ip6_ptr); | |
3865 | BUG_TRAP(!dev->dn_ptr); | |
1da177e4 | 3866 | |
b17a7c17 SH |
3867 | if (dev->destructor) |
3868 | dev->destructor(dev); | |
9093bbb2 SH |
3869 | |
3870 | /* Free network device */ | |
3871 | kobject_put(&dev->dev.kobj); | |
1da177e4 LT |
3872 | } |
3873 | ||
3874 | out: | |
4a3e2f71 | 3875 | mutex_unlock(&net_todo_run_mutex); |
1da177e4 LT |
3876 | } |
3877 | ||
5a1b5898 | 3878 | static struct net_device_stats *internal_stats(struct net_device *dev) |
c45d286e | 3879 | { |
5a1b5898 | 3880 | return &dev->stats; |
c45d286e RR |
3881 | } |
3882 | ||
1da177e4 | 3883 | /** |
f25f4e44 | 3884 | * alloc_netdev_mq - allocate network device |
1da177e4 LT |
3885 | * @sizeof_priv: size of private data to allocate space for |
3886 | * @name: device name format string | |
3887 | * @setup: callback to initialize device | |
f25f4e44 | 3888 | * @queue_count: the number of subqueues to allocate |
1da177e4 LT |
3889 | * |
3890 | * Allocates a struct net_device with private data area for driver use | |
f25f4e44 PWJ |
3891 | * and performs basic initialization. Also allocates subquue structs |
3892 | * for each queue on the device at the end of the netdevice. | |
1da177e4 | 3893 | */ |
f25f4e44 PWJ |
3894 | struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name, |
3895 | void (*setup)(struct net_device *), unsigned int queue_count) | |
1da177e4 LT |
3896 | { |
3897 | void *p; | |
3898 | struct net_device *dev; | |
3899 | int alloc_size; | |
3900 | ||
b6fe17d6 SH |
3901 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
3902 | ||
1da177e4 | 3903 | /* ensure 32-byte alignment of both the device and private area */ |
f25f4e44 | 3904 | alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST + |
31ce72a6 | 3905 | (sizeof(struct net_device_subqueue) * (queue_count - 1))) & |
f25f4e44 | 3906 | ~NETDEV_ALIGN_CONST; |
1da177e4 LT |
3907 | alloc_size += sizeof_priv + NETDEV_ALIGN_CONST; |
3908 | ||
31380de9 | 3909 | p = kzalloc(alloc_size, GFP_KERNEL); |
1da177e4 | 3910 | if (!p) { |
b6fe17d6 | 3911 | printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n"); |
1da177e4 LT |
3912 | return NULL; |
3913 | } | |
1da177e4 LT |
3914 | |
3915 | dev = (struct net_device *) | |
3916 | (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST); | |
3917 | dev->padded = (char *)dev - (char *)p; | |
6d34b1c2 | 3918 | dev->nd_net = &init_net; |
1da177e4 | 3919 | |
f25f4e44 PWJ |
3920 | if (sizeof_priv) { |
3921 | dev->priv = ((char *)dev + | |
3922 | ((sizeof(struct net_device) + | |
3923 | (sizeof(struct net_device_subqueue) * | |
31ce72a6 | 3924 | (queue_count - 1)) + NETDEV_ALIGN_CONST) |
f25f4e44 PWJ |
3925 | & ~NETDEV_ALIGN_CONST)); |
3926 | } | |
3927 | ||
3928 | dev->egress_subqueue_count = queue_count; | |
1da177e4 | 3929 | |
5a1b5898 | 3930 | dev->get_stats = internal_stats; |
bea3348e | 3931 | netpoll_netdev_init(dev); |
1da177e4 LT |
3932 | setup(dev); |
3933 | strcpy(dev->name, name); | |
3934 | return dev; | |
3935 | } | |
f25f4e44 | 3936 | EXPORT_SYMBOL(alloc_netdev_mq); |
1da177e4 LT |
3937 | |
3938 | /** | |
3939 | * free_netdev - free network device | |
3940 | * @dev: device | |
3941 | * | |
4ec93edb YH |
3942 | * This function does the last stage of destroying an allocated device |
3943 | * interface. The reference to the device object is released. | |
1da177e4 LT |
3944 | * If this is the last reference then it will be freed. |
3945 | */ | |
3946 | void free_netdev(struct net_device *dev) | |
3947 | { | |
3041a069 | 3948 | /* Compatibility with error handling in drivers */ |
1da177e4 LT |
3949 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
3950 | kfree((char *)dev - dev->padded); | |
3951 | return; | |
3952 | } | |
3953 | ||
3954 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
3955 | dev->reg_state = NETREG_RELEASED; | |
3956 | ||
43cb76d9 GKH |
3957 | /* will free via device release */ |
3958 | put_device(&dev->dev); | |
1da177e4 | 3959 | } |
4ec93edb | 3960 | |
1da177e4 | 3961 | /* Synchronize with packet receive processing. */ |
4ec93edb | 3962 | void synchronize_net(void) |
1da177e4 LT |
3963 | { |
3964 | might_sleep(); | |
fbd568a3 | 3965 | synchronize_rcu(); |
1da177e4 LT |
3966 | } |
3967 | ||
3968 | /** | |
3969 | * unregister_netdevice - remove device from the kernel | |
3970 | * @dev: device | |
3971 | * | |
3972 | * This function shuts down a device interface and removes it | |
3973 | * from the kernel tables. On success 0 is returned, on a failure | |
3974 | * a negative errno code is returned. | |
3975 | * | |
3976 | * Callers must hold the rtnl semaphore. You may want | |
3977 | * unregister_netdev() instead of this. | |
3978 | */ | |
3979 | ||
22f8cde5 | 3980 | void unregister_netdevice(struct net_device *dev) |
1da177e4 | 3981 | { |
93ee31f1 | 3982 | rollback_registered(dev); |
1da177e4 LT |
3983 | /* Finish processing unregister after unlock */ |
3984 | net_set_todo(dev); | |
1da177e4 LT |
3985 | } |
3986 | ||
3987 | /** | |
3988 | * unregister_netdev - remove device from the kernel | |
3989 | * @dev: device | |
3990 | * | |
3991 | * This function shuts down a device interface and removes it | |
3992 | * from the kernel tables. On success 0 is returned, on a failure | |
3993 | * a negative errno code is returned. | |
3994 | * | |
3995 | * This is just a wrapper for unregister_netdevice that takes | |
3996 | * the rtnl semaphore. In general you want to use this and not | |
3997 | * unregister_netdevice. | |
3998 | */ | |
3999 | void unregister_netdev(struct net_device *dev) | |
4000 | { | |
4001 | rtnl_lock(); | |
4002 | unregister_netdevice(dev); | |
4003 | rtnl_unlock(); | |
4004 | } | |
4005 | ||
4006 | EXPORT_SYMBOL(unregister_netdev); | |
4007 | ||
ce286d32 EB |
4008 | /** |
4009 | * dev_change_net_namespace - move device to different nethost namespace | |
4010 | * @dev: device | |
4011 | * @net: network namespace | |
4012 | * @pat: If not NULL name pattern to try if the current device name | |
4013 | * is already taken in the destination network namespace. | |
4014 | * | |
4015 | * This function shuts down a device interface and moves it | |
4016 | * to a new network namespace. On success 0 is returned, on | |
4017 | * a failure a netagive errno code is returned. | |
4018 | * | |
4019 | * Callers must hold the rtnl semaphore. | |
4020 | */ | |
4021 | ||
4022 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
4023 | { | |
4024 | char buf[IFNAMSIZ]; | |
4025 | const char *destname; | |
4026 | int err; | |
4027 | ||
4028 | ASSERT_RTNL(); | |
4029 | ||
4030 | /* Don't allow namespace local devices to be moved. */ | |
4031 | err = -EINVAL; | |
4032 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
4033 | goto out; | |
4034 | ||
4035 | /* Ensure the device has been registrered */ | |
4036 | err = -EINVAL; | |
4037 | if (dev->reg_state != NETREG_REGISTERED) | |
4038 | goto out; | |
4039 | ||
4040 | /* Get out if there is nothing todo */ | |
4041 | err = 0; | |
4042 | if (dev->nd_net == net) | |
4043 | goto out; | |
4044 | ||
4045 | /* Pick the destination device name, and ensure | |
4046 | * we can use it in the destination network namespace. | |
4047 | */ | |
4048 | err = -EEXIST; | |
4049 | destname = dev->name; | |
4050 | if (__dev_get_by_name(net, destname)) { | |
4051 | /* We get here if we can't use the current device name */ | |
4052 | if (!pat) | |
4053 | goto out; | |
4054 | if (!dev_valid_name(pat)) | |
4055 | goto out; | |
4056 | if (strchr(pat, '%')) { | |
4057 | if (__dev_alloc_name(net, pat, buf) < 0) | |
4058 | goto out; | |
4059 | destname = buf; | |
4060 | } else | |
4061 | destname = pat; | |
4062 | if (__dev_get_by_name(net, destname)) | |
4063 | goto out; | |
4064 | } | |
4065 | ||
4066 | /* | |
4067 | * And now a mini version of register_netdevice unregister_netdevice. | |
4068 | */ | |
4069 | ||
4070 | /* If device is running close it first. */ | |
9b772652 | 4071 | dev_close(dev); |
ce286d32 EB |
4072 | |
4073 | /* And unlink it from device chain */ | |
4074 | err = -ENODEV; | |
4075 | unlist_netdevice(dev); | |
4076 | ||
4077 | synchronize_net(); | |
4078 | ||
4079 | /* Shutdown queueing discipline. */ | |
4080 | dev_shutdown(dev); | |
4081 | ||
4082 | /* Notify protocols, that we are about to destroy | |
4083 | this device. They should clean all the things. | |
4084 | */ | |
4085 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
4086 | ||
4087 | /* | |
4088 | * Flush the unicast and multicast chains | |
4089 | */ | |
4090 | dev_addr_discard(dev); | |
4091 | ||
4092 | /* Actually switch the network namespace */ | |
4093 | dev->nd_net = net; | |
4094 | ||
4095 | /* Assign the new device name */ | |
4096 | if (destname != dev->name) | |
4097 | strcpy(dev->name, destname); | |
4098 | ||
4099 | /* If there is an ifindex conflict assign a new one */ | |
4100 | if (__dev_get_by_index(net, dev->ifindex)) { | |
4101 | int iflink = (dev->iflink == dev->ifindex); | |
4102 | dev->ifindex = dev_new_index(net); | |
4103 | if (iflink) | |
4104 | dev->iflink = dev->ifindex; | |
4105 | } | |
4106 | ||
8b41d188 | 4107 | /* Fixup kobjects */ |
ce286d32 | 4108 | err = device_rename(&dev->dev, dev->name); |
8b41d188 | 4109 | WARN_ON(err); |
ce286d32 EB |
4110 | |
4111 | /* Add the device back in the hashes */ | |
4112 | list_netdevice(dev); | |
4113 | ||
4114 | /* Notify protocols, that a new device appeared. */ | |
4115 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
4116 | ||
4117 | synchronize_net(); | |
4118 | err = 0; | |
4119 | out: | |
4120 | return err; | |
4121 | } | |
4122 | ||
1da177e4 LT |
4123 | static int dev_cpu_callback(struct notifier_block *nfb, |
4124 | unsigned long action, | |
4125 | void *ocpu) | |
4126 | { | |
4127 | struct sk_buff **list_skb; | |
4128 | struct net_device **list_net; | |
4129 | struct sk_buff *skb; | |
4130 | unsigned int cpu, oldcpu = (unsigned long)ocpu; | |
4131 | struct softnet_data *sd, *oldsd; | |
4132 | ||
8bb78442 | 4133 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
1da177e4 LT |
4134 | return NOTIFY_OK; |
4135 | ||
4136 | local_irq_disable(); | |
4137 | cpu = smp_processor_id(); | |
4138 | sd = &per_cpu(softnet_data, cpu); | |
4139 | oldsd = &per_cpu(softnet_data, oldcpu); | |
4140 | ||
4141 | /* Find end of our completion_queue. */ | |
4142 | list_skb = &sd->completion_queue; | |
4143 | while (*list_skb) | |
4144 | list_skb = &(*list_skb)->next; | |
4145 | /* Append completion queue from offline CPU. */ | |
4146 | *list_skb = oldsd->completion_queue; | |
4147 | oldsd->completion_queue = NULL; | |
4148 | ||
4149 | /* Find end of our output_queue. */ | |
4150 | list_net = &sd->output_queue; | |
4151 | while (*list_net) | |
4152 | list_net = &(*list_net)->next_sched; | |
4153 | /* Append output queue from offline CPU. */ | |
4154 | *list_net = oldsd->output_queue; | |
4155 | oldsd->output_queue = NULL; | |
4156 | ||
4157 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
4158 | local_irq_enable(); | |
4159 | ||
4160 | /* Process offline CPU's input_pkt_queue */ | |
4161 | while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) | |
4162 | netif_rx(skb); | |
4163 | ||
4164 | return NOTIFY_OK; | |
4165 | } | |
1da177e4 | 4166 | |
db217334 CL |
4167 | #ifdef CONFIG_NET_DMA |
4168 | /** | |
0ed72ec4 RD |
4169 | * net_dma_rebalance - try to maintain one DMA channel per CPU |
4170 | * @net_dma: DMA client and associated data (lock, channels, channel_mask) | |
4171 | * | |
4172 | * This is called when the number of channels allocated to the net_dma client | |
4173 | * changes. The net_dma client tries to have one DMA channel per CPU. | |
db217334 | 4174 | */ |
d379b01e DW |
4175 | |
4176 | static void net_dma_rebalance(struct net_dma *net_dma) | |
db217334 | 4177 | { |
d379b01e | 4178 | unsigned int cpu, i, n, chan_idx; |
db217334 CL |
4179 | struct dma_chan *chan; |
4180 | ||
d379b01e | 4181 | if (cpus_empty(net_dma->channel_mask)) { |
db217334 | 4182 | for_each_online_cpu(cpu) |
29bbd72d | 4183 | rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL); |
db217334 CL |
4184 | return; |
4185 | } | |
4186 | ||
4187 | i = 0; | |
4188 | cpu = first_cpu(cpu_online_map); | |
4189 | ||
d379b01e DW |
4190 | for_each_cpu_mask(chan_idx, net_dma->channel_mask) { |
4191 | chan = net_dma->channels[chan_idx]; | |
4192 | ||
4193 | n = ((num_online_cpus() / cpus_weight(net_dma->channel_mask)) | |
4194 | + (i < (num_online_cpus() % | |
4195 | cpus_weight(net_dma->channel_mask)) ? 1 : 0)); | |
db217334 CL |
4196 | |
4197 | while(n) { | |
29bbd72d | 4198 | per_cpu(softnet_data, cpu).net_dma = chan; |
db217334 CL |
4199 | cpu = next_cpu(cpu, cpu_online_map); |
4200 | n--; | |
4201 | } | |
4202 | i++; | |
4203 | } | |
db217334 CL |
4204 | } |
4205 | ||
4206 | /** | |
4207 | * netdev_dma_event - event callback for the net_dma_client | |
4208 | * @client: should always be net_dma_client | |
f4b8ea78 | 4209 | * @chan: DMA channel for the event |
0ed72ec4 | 4210 | * @state: DMA state to be handled |
db217334 | 4211 | */ |
d379b01e DW |
4212 | static enum dma_state_client |
4213 | netdev_dma_event(struct dma_client *client, struct dma_chan *chan, | |
4214 | enum dma_state state) | |
4215 | { | |
4216 | int i, found = 0, pos = -1; | |
4217 | struct net_dma *net_dma = | |
4218 | container_of(client, struct net_dma, client); | |
4219 | enum dma_state_client ack = DMA_DUP; /* default: take no action */ | |
4220 | ||
4221 | spin_lock(&net_dma->lock); | |
4222 | switch (state) { | |
4223 | case DMA_RESOURCE_AVAILABLE: | |
4224 | for (i = 0; i < NR_CPUS; i++) | |
4225 | if (net_dma->channels[i] == chan) { | |
4226 | found = 1; | |
4227 | break; | |
4228 | } else if (net_dma->channels[i] == NULL && pos < 0) | |
4229 | pos = i; | |
4230 | ||
4231 | if (!found && pos >= 0) { | |
4232 | ack = DMA_ACK; | |
4233 | net_dma->channels[pos] = chan; | |
4234 | cpu_set(pos, net_dma->channel_mask); | |
4235 | net_dma_rebalance(net_dma); | |
4236 | } | |
db217334 CL |
4237 | break; |
4238 | case DMA_RESOURCE_REMOVED: | |
d379b01e DW |
4239 | for (i = 0; i < NR_CPUS; i++) |
4240 | if (net_dma->channels[i] == chan) { | |
4241 | found = 1; | |
4242 | pos = i; | |
4243 | break; | |
4244 | } | |
4245 | ||
4246 | if (found) { | |
4247 | ack = DMA_ACK; | |
4248 | cpu_clear(pos, net_dma->channel_mask); | |
4249 | net_dma->channels[i] = NULL; | |
4250 | net_dma_rebalance(net_dma); | |
4251 | } | |
db217334 CL |
4252 | break; |
4253 | default: | |
4254 | break; | |
4255 | } | |
d379b01e DW |
4256 | spin_unlock(&net_dma->lock); |
4257 | ||
4258 | return ack; | |
db217334 CL |
4259 | } |
4260 | ||
4261 | /** | |
4262 | * netdev_dma_regiser - register the networking subsystem as a DMA client | |
4263 | */ | |
4264 | static int __init netdev_dma_register(void) | |
4265 | { | |
d379b01e DW |
4266 | spin_lock_init(&net_dma.lock); |
4267 | dma_cap_set(DMA_MEMCPY, net_dma.client.cap_mask); | |
4268 | dma_async_client_register(&net_dma.client); | |
4269 | dma_async_client_chan_request(&net_dma.client); | |
db217334 CL |
4270 | return 0; |
4271 | } | |
4272 | ||
4273 | #else | |
4274 | static int __init netdev_dma_register(void) { return -ENODEV; } | |
4275 | #endif /* CONFIG_NET_DMA */ | |
1da177e4 | 4276 | |
7f353bf2 HX |
4277 | /** |
4278 | * netdev_compute_feature - compute conjunction of two feature sets | |
4279 | * @all: first feature set | |
4280 | * @one: second feature set | |
4281 | * | |
4282 | * Computes a new feature set after adding a device with feature set | |
4283 | * @one to the master device with current feature set @all. Returns | |
4284 | * the new feature set. | |
4285 | */ | |
4286 | int netdev_compute_features(unsigned long all, unsigned long one) | |
4287 | { | |
4288 | /* if device needs checksumming, downgrade to hw checksumming */ | |
4289 | if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM)) | |
4290 | all ^= NETIF_F_NO_CSUM | NETIF_F_HW_CSUM; | |
4291 | ||
4292 | /* if device can't do all checksum, downgrade to ipv4/ipv6 */ | |
4293 | if (all & NETIF_F_HW_CSUM && !(one & NETIF_F_HW_CSUM)) | |
4294 | all ^= NETIF_F_HW_CSUM | |
4295 | | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM; | |
4296 | ||
4297 | if (one & NETIF_F_GSO) | |
4298 | one |= NETIF_F_GSO_SOFTWARE; | |
4299 | one |= NETIF_F_GSO; | |
4300 | ||
4301 | /* If even one device supports robust GSO, enable it for all. */ | |
4302 | if (one & NETIF_F_GSO_ROBUST) | |
4303 | all |= NETIF_F_GSO_ROBUST; | |
4304 | ||
4305 | all &= one | NETIF_F_LLTX; | |
4306 | ||
4307 | if (!(all & NETIF_F_ALL_CSUM)) | |
4308 | all &= ~NETIF_F_SG; | |
4309 | if (!(all & NETIF_F_SG)) | |
4310 | all &= ~NETIF_F_GSO_MASK; | |
4311 | ||
4312 | return all; | |
4313 | } | |
4314 | EXPORT_SYMBOL(netdev_compute_features); | |
4315 | ||
30d97d35 PE |
4316 | static struct hlist_head *netdev_create_hash(void) |
4317 | { | |
4318 | int i; | |
4319 | struct hlist_head *hash; | |
4320 | ||
4321 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
4322 | if (hash != NULL) | |
4323 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
4324 | INIT_HLIST_HEAD(&hash[i]); | |
4325 | ||
4326 | return hash; | |
4327 | } | |
4328 | ||
881d966b | 4329 | /* Initialize per network namespace state */ |
4665079c | 4330 | static int __net_init netdev_init(struct net *net) |
881d966b | 4331 | { |
881d966b | 4332 | INIT_LIST_HEAD(&net->dev_base_head); |
881d966b | 4333 | |
30d97d35 PE |
4334 | net->dev_name_head = netdev_create_hash(); |
4335 | if (net->dev_name_head == NULL) | |
4336 | goto err_name; | |
881d966b | 4337 | |
30d97d35 PE |
4338 | net->dev_index_head = netdev_create_hash(); |
4339 | if (net->dev_index_head == NULL) | |
4340 | goto err_idx; | |
881d966b EB |
4341 | |
4342 | return 0; | |
30d97d35 PE |
4343 | |
4344 | err_idx: | |
4345 | kfree(net->dev_name_head); | |
4346 | err_name: | |
4347 | return -ENOMEM; | |
881d966b EB |
4348 | } |
4349 | ||
4665079c | 4350 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
4351 | { |
4352 | kfree(net->dev_name_head); | |
4353 | kfree(net->dev_index_head); | |
4354 | } | |
4355 | ||
022cbae6 | 4356 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
4357 | .init = netdev_init, |
4358 | .exit = netdev_exit, | |
4359 | }; | |
4360 | ||
4665079c | 4361 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 EB |
4362 | { |
4363 | struct net_device *dev, *next; | |
4364 | /* | |
4365 | * Push all migratable of the network devices back to the | |
4366 | * initial network namespace | |
4367 | */ | |
4368 | rtnl_lock(); | |
4369 | for_each_netdev_safe(net, dev, next) { | |
4370 | int err; | |
4371 | ||
4372 | /* Ignore unmoveable devices (i.e. loopback) */ | |
4373 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
4374 | continue; | |
4375 | ||
4376 | /* Push remaing network devices to init_net */ | |
4377 | err = dev_change_net_namespace(dev, &init_net, "dev%d"); | |
4378 | if (err) { | |
4379 | printk(KERN_WARNING "%s: failed to move %s to init_net: %d\n", | |
4380 | __func__, dev->name, err); | |
4381 | unregister_netdevice(dev); | |
4382 | } | |
4383 | } | |
4384 | rtnl_unlock(); | |
4385 | } | |
4386 | ||
022cbae6 | 4387 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 EB |
4388 | .exit = default_device_exit, |
4389 | }; | |
4390 | ||
1da177e4 LT |
4391 | /* |
4392 | * Initialize the DEV module. At boot time this walks the device list and | |
4393 | * unhooks any devices that fail to initialise (normally hardware not | |
4394 | * present) and leaves us with a valid list of present and active devices. | |
4395 | * | |
4396 | */ | |
4397 | ||
4398 | /* | |
4399 | * This is called single threaded during boot, so no need | |
4400 | * to take the rtnl semaphore. | |
4401 | */ | |
4402 | static int __init net_dev_init(void) | |
4403 | { | |
4404 | int i, rc = -ENOMEM; | |
4405 | ||
4406 | BUG_ON(!dev_boot_phase); | |
4407 | ||
1da177e4 LT |
4408 | if (dev_proc_init()) |
4409 | goto out; | |
4410 | ||
8b41d188 | 4411 | if (netdev_kobject_init()) |
1da177e4 LT |
4412 | goto out; |
4413 | ||
4414 | INIT_LIST_HEAD(&ptype_all); | |
4ec93edb | 4415 | for (i = 0; i < 16; i++) |
1da177e4 LT |
4416 | INIT_LIST_HEAD(&ptype_base[i]); |
4417 | ||
881d966b EB |
4418 | if (register_pernet_subsys(&netdev_net_ops)) |
4419 | goto out; | |
1da177e4 | 4420 | |
ce286d32 EB |
4421 | if (register_pernet_device(&default_device_ops)) |
4422 | goto out; | |
4423 | ||
1da177e4 LT |
4424 | /* |
4425 | * Initialise the packet receive queues. | |
4426 | */ | |
4427 | ||
6f912042 | 4428 | for_each_possible_cpu(i) { |
1da177e4 LT |
4429 | struct softnet_data *queue; |
4430 | ||
4431 | queue = &per_cpu(softnet_data, i); | |
4432 | skb_queue_head_init(&queue->input_pkt_queue); | |
1da177e4 LT |
4433 | queue->completion_queue = NULL; |
4434 | INIT_LIST_HEAD(&queue->poll_list); | |
bea3348e SH |
4435 | |
4436 | queue->backlog.poll = process_backlog; | |
4437 | queue->backlog.weight = weight_p; | |
1da177e4 LT |
4438 | } |
4439 | ||
db217334 CL |
4440 | netdev_dma_register(); |
4441 | ||
1da177e4 LT |
4442 | dev_boot_phase = 0; |
4443 | ||
4444 | open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL); | |
4445 | open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL); | |
4446 | ||
4447 | hotcpu_notifier(dev_cpu_callback, 0); | |
4448 | dst_init(); | |
4449 | dev_mcast_init(); | |
4450 | rc = 0; | |
4451 | out: | |
4452 | return rc; | |
4453 | } | |
4454 | ||
4455 | subsys_initcall(net_dev_init); | |
4456 | ||
4457 | EXPORT_SYMBOL(__dev_get_by_index); | |
4458 | EXPORT_SYMBOL(__dev_get_by_name); | |
4459 | EXPORT_SYMBOL(__dev_remove_pack); | |
c2373ee9 | 4460 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
4461 | EXPORT_SYMBOL(dev_add_pack); |
4462 | EXPORT_SYMBOL(dev_alloc_name); | |
4463 | EXPORT_SYMBOL(dev_close); | |
4464 | EXPORT_SYMBOL(dev_get_by_flags); | |
4465 | EXPORT_SYMBOL(dev_get_by_index); | |
4466 | EXPORT_SYMBOL(dev_get_by_name); | |
1da177e4 LT |
4467 | EXPORT_SYMBOL(dev_open); |
4468 | EXPORT_SYMBOL(dev_queue_xmit); | |
4469 | EXPORT_SYMBOL(dev_remove_pack); | |
4470 | EXPORT_SYMBOL(dev_set_allmulti); | |
4471 | EXPORT_SYMBOL(dev_set_promiscuity); | |
4472 | EXPORT_SYMBOL(dev_change_flags); | |
4473 | EXPORT_SYMBOL(dev_set_mtu); | |
4474 | EXPORT_SYMBOL(dev_set_mac_address); | |
4475 | EXPORT_SYMBOL(free_netdev); | |
4476 | EXPORT_SYMBOL(netdev_boot_setup_check); | |
4477 | EXPORT_SYMBOL(netdev_set_master); | |
4478 | EXPORT_SYMBOL(netdev_state_change); | |
4479 | EXPORT_SYMBOL(netif_receive_skb); | |
4480 | EXPORT_SYMBOL(netif_rx); | |
4481 | EXPORT_SYMBOL(register_gifconf); | |
4482 | EXPORT_SYMBOL(register_netdevice); | |
4483 | EXPORT_SYMBOL(register_netdevice_notifier); | |
4484 | EXPORT_SYMBOL(skb_checksum_help); | |
4485 | EXPORT_SYMBOL(synchronize_net); | |
4486 | EXPORT_SYMBOL(unregister_netdevice); | |
4487 | EXPORT_SYMBOL(unregister_netdevice_notifier); | |
4488 | EXPORT_SYMBOL(net_enable_timestamp); | |
4489 | EXPORT_SYMBOL(net_disable_timestamp); | |
4490 | EXPORT_SYMBOL(dev_get_flags); | |
4491 | ||
4492 | #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE) | |
4493 | EXPORT_SYMBOL(br_handle_frame_hook); | |
4494 | EXPORT_SYMBOL(br_fdb_get_hook); | |
4495 | EXPORT_SYMBOL(br_fdb_put_hook); | |
4496 | #endif | |
4497 | ||
4498 | #ifdef CONFIG_KMOD | |
4499 | EXPORT_SYMBOL(dev_load); | |
4500 | #endif | |
4501 | ||
4502 | EXPORT_PER_CPU_SYMBOL(softnet_data); |