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