Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * NET3 Protocol independent device support routines. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public License | |
6 | * as published by the Free Software Foundation; either version | |
7 | * 2 of the License, or (at your option) any later version. | |
8 | * | |
9 | * Derived from the non IP parts of dev.c 1.0.19 | |
02c30a84 | 10 | * Authors: Ross Biro |
1da177e4 LT |
11 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
12 | * Mark Evans, <evansmp@uhura.aston.ac.uk> | |
13 | * | |
14 | * Additional Authors: | |
15 | * Florian la Roche <rzsfl@rz.uni-sb.de> | |
16 | * Alan Cox <gw4pts@gw4pts.ampr.org> | |
17 | * David Hinds <dahinds@users.sourceforge.net> | |
18 | * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> | |
19 | * Adam Sulmicki <adam@cfar.umd.edu> | |
20 | * Pekka Riikonen <priikone@poesidon.pspt.fi> | |
21 | * | |
22 | * Changes: | |
23 | * D.J. Barrow : Fixed bug where dev->refcnt gets set | |
24 | * to 2 if register_netdev gets called | |
25 | * before net_dev_init & also removed a | |
26 | * few lines of code in the process. | |
27 | * Alan Cox : device private ioctl copies fields back. | |
28 | * Alan Cox : Transmit queue code does relevant | |
29 | * stunts to keep the queue safe. | |
30 | * Alan Cox : Fixed double lock. | |
31 | * Alan Cox : Fixed promisc NULL pointer trap | |
32 | * ???????? : Support the full private ioctl range | |
33 | * Alan Cox : Moved ioctl permission check into | |
34 | * drivers | |
35 | * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI | |
36 | * Alan Cox : 100 backlog just doesn't cut it when | |
37 | * you start doing multicast video 8) | |
38 | * Alan Cox : Rewrote net_bh and list manager. | |
39 | * Alan Cox : Fix ETH_P_ALL echoback lengths. | |
40 | * Alan Cox : Took out transmit every packet pass | |
41 | * Saved a few bytes in the ioctl handler | |
42 | * Alan Cox : Network driver sets packet type before | |
43 | * calling netif_rx. Saves a function | |
44 | * call a packet. | |
45 | * Alan Cox : Hashed net_bh() | |
46 | * Richard Kooijman: Timestamp fixes. | |
47 | * Alan Cox : Wrong field in SIOCGIFDSTADDR | |
48 | * Alan Cox : Device lock protection. | |
49 | * Alan Cox : Fixed nasty side effect of device close | |
50 | * changes. | |
51 | * Rudi Cilibrasi : Pass the right thing to | |
52 | * set_mac_address() | |
53 | * Dave Miller : 32bit quantity for the device lock to | |
54 | * make it work out on a Sparc. | |
55 | * Bjorn Ekwall : Added KERNELD hack. | |
56 | * Alan Cox : Cleaned up the backlog initialise. | |
57 | * Craig Metz : SIOCGIFCONF fix if space for under | |
58 | * 1 device. | |
59 | * Thomas Bogendoerfer : Return ENODEV for dev_open, if there | |
60 | * is no device open function. | |
61 | * Andi Kleen : Fix error reporting for SIOCGIFCONF | |
62 | * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF | |
63 | * Cyrus Durgin : Cleaned for KMOD | |
64 | * Adam Sulmicki : Bug Fix : Network Device Unload | |
65 | * A network device unload needs to purge | |
66 | * the backlog queue. | |
67 | * Paul Rusty Russell : SIOCSIFNAME | |
68 | * Pekka Riikonen : Netdev boot-time settings code | |
69 | * Andrew Morton : Make unregister_netdevice wait | |
70 | * indefinitely on dev->refcnt | |
71 | * J Hadi Salim : - Backlog queue sampling | |
72 | * - netif_rx() feedback | |
73 | */ | |
74 | ||
75 | #include <asm/uaccess.h> | |
76 | #include <asm/system.h> | |
77 | #include <linux/bitops.h> | |
4fc268d2 | 78 | #include <linux/capability.h> |
1da177e4 LT |
79 | #include <linux/cpu.h> |
80 | #include <linux/types.h> | |
81 | #include <linux/kernel.h> | |
82 | #include <linux/sched.h> | |
4a3e2f71 | 83 | #include <linux/mutex.h> |
1da177e4 LT |
84 | #include <linux/string.h> |
85 | #include <linux/mm.h> | |
86 | #include <linux/socket.h> | |
87 | #include <linux/sockios.h> | |
88 | #include <linux/errno.h> | |
89 | #include <linux/interrupt.h> | |
90 | #include <linux/if_ether.h> | |
91 | #include <linux/netdevice.h> | |
92 | #include <linux/etherdevice.h> | |
93 | #include <linux/notifier.h> | |
94 | #include <linux/skbuff.h> | |
457c4cbc | 95 | #include <net/net_namespace.h> |
1da177e4 LT |
96 | #include <net/sock.h> |
97 | #include <linux/rtnetlink.h> | |
98 | #include <linux/proc_fs.h> | |
99 | #include <linux/seq_file.h> | |
100 | #include <linux/stat.h> | |
101 | #include <linux/if_bridge.h> | |
b863ceb7 | 102 | #include <linux/if_macvlan.h> |
1da177e4 LT |
103 | #include <net/dst.h> |
104 | #include <net/pkt_sched.h> | |
105 | #include <net/checksum.h> | |
106 | #include <linux/highmem.h> | |
107 | #include <linux/init.h> | |
108 | #include <linux/kmod.h> | |
109 | #include <linux/module.h> | |
110 | #include <linux/kallsyms.h> | |
111 | #include <linux/netpoll.h> | |
112 | #include <linux/rcupdate.h> | |
113 | #include <linux/delay.h> | |
295f4a1f | 114 | #include <net/wext.h> |
1da177e4 | 115 | #include <net/iw_handler.h> |
1da177e4 | 116 | #include <asm/current.h> |
5bdb9886 | 117 | #include <linux/audit.h> |
db217334 | 118 | #include <linux/dmaengine.h> |
f6a78bfc | 119 | #include <linux/err.h> |
c7fa9d18 | 120 | #include <linux/ctype.h> |
723e98b7 | 121 | #include <linux/if_arp.h> |
1da177e4 | 122 | |
342709ef PE |
123 | #include "net-sysfs.h" |
124 | ||
1da177e4 LT |
125 | /* |
126 | * The list of packet types we will receive (as opposed to discard) | |
127 | * and the routines to invoke. | |
128 | * | |
129 | * Why 16. Because with 16 the only overlap we get on a hash of the | |
130 | * low nibble of the protocol value is RARP/SNAP/X.25. | |
131 | * | |
132 | * NOTE: That is no longer true with the addition of VLAN tags. Not | |
133 | * sure which should go first, but I bet it won't make much | |
134 | * difference if we are running VLANs. The good news is that | |
135 | * this protocol won't be in the list unless compiled in, so | |
3041a069 | 136 | * the average user (w/out VLANs) will not be adversely affected. |
1da177e4 LT |
137 | * --BLG |
138 | * | |
139 | * 0800 IP | |
140 | * 8100 802.1Q VLAN | |
141 | * 0001 802.3 | |
142 | * 0002 AX.25 | |
143 | * 0004 802.2 | |
144 | * 8035 RARP | |
145 | * 0005 SNAP | |
146 | * 0805 X.25 | |
147 | * 0806 ARP | |
148 | * 8137 IPX | |
149 | * 0009 Localtalk | |
150 | * 86DD IPv6 | |
151 | */ | |
152 | ||
82d8a867 PE |
153 | #define PTYPE_HASH_SIZE (16) |
154 | #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1) | |
155 | ||
1da177e4 | 156 | static DEFINE_SPINLOCK(ptype_lock); |
82d8a867 | 157 | static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
6b2bedc3 | 158 | static struct list_head ptype_all __read_mostly; /* Taps */ |
1da177e4 | 159 | |
db217334 | 160 | #ifdef CONFIG_NET_DMA |
d379b01e DW |
161 | struct net_dma { |
162 | struct dma_client client; | |
163 | spinlock_t lock; | |
164 | cpumask_t channel_mask; | |
165 | struct dma_chan *channels[NR_CPUS]; | |
166 | }; | |
167 | ||
168 | static enum dma_state_client | |
169 | netdev_dma_event(struct dma_client *client, struct dma_chan *chan, | |
170 | enum dma_state state); | |
171 | ||
172 | static struct net_dma net_dma = { | |
173 | .client = { | |
174 | .event_callback = netdev_dma_event, | |
175 | }, | |
176 | }; | |
db217334 CL |
177 | #endif |
178 | ||
1da177e4 | 179 | /* |
7562f876 | 180 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
181 | * semaphore. |
182 | * | |
183 | * Pure readers hold dev_base_lock for reading. | |
184 | * | |
185 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 186 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
187 | * actual updates. This allows pure readers to access the list even |
188 | * while a writer is preparing to update it. | |
189 | * | |
190 | * To put it another way, dev_base_lock is held for writing only to | |
191 | * protect against pure readers; the rtnl semaphore provides the | |
192 | * protection against other writers. | |
193 | * | |
194 | * See, for example usages, register_netdevice() and | |
195 | * unregister_netdevice(), which must be called with the rtnl | |
196 | * semaphore held. | |
197 | */ | |
1da177e4 LT |
198 | DEFINE_RWLOCK(dev_base_lock); |
199 | ||
1da177e4 LT |
200 | EXPORT_SYMBOL(dev_base_lock); |
201 | ||
202 | #define NETDEV_HASHBITS 8 | |
881d966b | 203 | #define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS) |
1da177e4 | 204 | |
881d966b | 205 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 LT |
206 | { |
207 | unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ)); | |
881d966b | 208 | return &net->dev_name_head[hash & ((1 << NETDEV_HASHBITS) - 1)]; |
1da177e4 LT |
209 | } |
210 | ||
881d966b | 211 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 212 | { |
881d966b | 213 | return &net->dev_index_head[ifindex & ((1 << NETDEV_HASHBITS) - 1)]; |
1da177e4 LT |
214 | } |
215 | ||
ce286d32 EB |
216 | /* Device list insertion */ |
217 | static int list_netdevice(struct net_device *dev) | |
218 | { | |
219 | struct net *net = dev->nd_net; | |
220 | ||
221 | ASSERT_RTNL(); | |
222 | ||
223 | write_lock_bh(&dev_base_lock); | |
224 | list_add_tail(&dev->dev_list, &net->dev_base_head); | |
225 | hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
226 | hlist_add_head(&dev->index_hlist, dev_index_hash(net, dev->ifindex)); | |
227 | write_unlock_bh(&dev_base_lock); | |
228 | return 0; | |
229 | } | |
230 | ||
231 | /* Device list removal */ | |
232 | static void unlist_netdevice(struct net_device *dev) | |
233 | { | |
234 | ASSERT_RTNL(); | |
235 | ||
236 | /* Unlink dev from the device chain */ | |
237 | write_lock_bh(&dev_base_lock); | |
238 | list_del(&dev->dev_list); | |
239 | hlist_del(&dev->name_hlist); | |
240 | hlist_del(&dev->index_hlist); | |
241 | write_unlock_bh(&dev_base_lock); | |
242 | } | |
243 | ||
1da177e4 LT |
244 | /* |
245 | * Our notifier list | |
246 | */ | |
247 | ||
f07d5b94 | 248 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
249 | |
250 | /* | |
251 | * Device drivers call our routines to queue packets here. We empty the | |
252 | * queue in the local softnet handler. | |
253 | */ | |
bea3348e SH |
254 | |
255 | DEFINE_PER_CPU(struct softnet_data, softnet_data); | |
1da177e4 | 256 | |
723e98b7 JP |
257 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
258 | /* | |
259 | * register_netdevice() inits dev->_xmit_lock and sets lockdep class | |
260 | * according to dev->type | |
261 | */ | |
262 | static const unsigned short netdev_lock_type[] = | |
263 | {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
264 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, | |
265 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
266 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
267 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
268 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
269 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
270 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
271 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
272 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
273 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
274 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
275 | ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211, | |
276 | ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_VOID, | |
277 | ARPHRD_NONE}; | |
278 | ||
279 | static const char *netdev_lock_name[] = | |
280 | {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", | |
281 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
282 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
283 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
284 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
285 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
286 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
287 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
288 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
289 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
290 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
291 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
292 | "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211", | |
293 | "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_VOID", | |
294 | "_xmit_NONE"}; | |
295 | ||
296 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
297 | ||
298 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
299 | { | |
300 | int i; | |
301 | ||
302 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
303 | if (netdev_lock_type[i] == dev_type) | |
304 | return i; | |
305 | /* the last key is used by default */ | |
306 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
307 | } | |
308 | ||
309 | static inline void netdev_set_lockdep_class(spinlock_t *lock, | |
310 | unsigned short dev_type) | |
311 | { | |
312 | int i; | |
313 | ||
314 | i = netdev_lock_pos(dev_type); | |
315 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
316 | netdev_lock_name[i]); | |
317 | } | |
318 | #else | |
319 | static inline void netdev_set_lockdep_class(spinlock_t *lock, | |
320 | unsigned short dev_type) | |
321 | { | |
322 | } | |
323 | #endif | |
1da177e4 LT |
324 | |
325 | /******************************************************************************* | |
326 | ||
327 | Protocol management and registration routines | |
328 | ||
329 | *******************************************************************************/ | |
330 | ||
1da177e4 LT |
331 | /* |
332 | * Add a protocol ID to the list. Now that the input handler is | |
333 | * smarter we can dispense with all the messy stuff that used to be | |
334 | * here. | |
335 | * | |
336 | * BEWARE!!! Protocol handlers, mangling input packets, | |
337 | * MUST BE last in hash buckets and checking protocol handlers | |
338 | * MUST start from promiscuous ptype_all chain in net_bh. | |
339 | * It is true now, do not change it. | |
340 | * Explanation follows: if protocol handler, mangling packet, will | |
341 | * be the first on list, it is not able to sense, that packet | |
342 | * is cloned and should be copied-on-write, so that it will | |
343 | * change it and subsequent readers will get broken packet. | |
344 | * --ANK (980803) | |
345 | */ | |
346 | ||
347 | /** | |
348 | * dev_add_pack - add packet handler | |
349 | * @pt: packet type declaration | |
350 | * | |
351 | * Add a protocol handler to the networking stack. The passed &packet_type | |
352 | * is linked into kernel lists and may not be freed until it has been | |
353 | * removed from the kernel lists. | |
354 | * | |
4ec93edb | 355 | * This call does not sleep therefore it can not |
1da177e4 LT |
356 | * guarantee all CPU's that are in middle of receiving packets |
357 | * will see the new packet type (until the next received packet). | |
358 | */ | |
359 | ||
360 | void dev_add_pack(struct packet_type *pt) | |
361 | { | |
362 | int hash; | |
363 | ||
364 | spin_lock_bh(&ptype_lock); | |
9be9a6b9 | 365 | if (pt->type == htons(ETH_P_ALL)) |
1da177e4 | 366 | list_add_rcu(&pt->list, &ptype_all); |
9be9a6b9 | 367 | else { |
82d8a867 | 368 | hash = ntohs(pt->type) & PTYPE_HASH_MASK; |
1da177e4 LT |
369 | list_add_rcu(&pt->list, &ptype_base[hash]); |
370 | } | |
371 | spin_unlock_bh(&ptype_lock); | |
372 | } | |
373 | ||
1da177e4 LT |
374 | /** |
375 | * __dev_remove_pack - remove packet handler | |
376 | * @pt: packet type declaration | |
377 | * | |
378 | * Remove a protocol handler that was previously added to the kernel | |
379 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
380 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 381 | * returns. |
1da177e4 LT |
382 | * |
383 | * The packet type might still be in use by receivers | |
384 | * and must not be freed until after all the CPU's have gone | |
385 | * through a quiescent state. | |
386 | */ | |
387 | void __dev_remove_pack(struct packet_type *pt) | |
388 | { | |
389 | struct list_head *head; | |
390 | struct packet_type *pt1; | |
391 | ||
392 | spin_lock_bh(&ptype_lock); | |
393 | ||
9be9a6b9 | 394 | if (pt->type == htons(ETH_P_ALL)) |
1da177e4 | 395 | head = &ptype_all; |
9be9a6b9 | 396 | else |
82d8a867 | 397 | head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; |
1da177e4 LT |
398 | |
399 | list_for_each_entry(pt1, head, list) { | |
400 | if (pt == pt1) { | |
401 | list_del_rcu(&pt->list); | |
402 | goto out; | |
403 | } | |
404 | } | |
405 | ||
406 | printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt); | |
407 | out: | |
408 | spin_unlock_bh(&ptype_lock); | |
409 | } | |
410 | /** | |
411 | * dev_remove_pack - remove packet handler | |
412 | * @pt: packet type declaration | |
413 | * | |
414 | * Remove a protocol handler that was previously added to the kernel | |
415 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
416 | * from the kernel lists and can be freed or reused once this function | |
417 | * returns. | |
418 | * | |
419 | * This call sleeps to guarantee that no CPU is looking at the packet | |
420 | * type after return. | |
421 | */ | |
422 | void dev_remove_pack(struct packet_type *pt) | |
423 | { | |
424 | __dev_remove_pack(pt); | |
4ec93edb | 425 | |
1da177e4 LT |
426 | synchronize_net(); |
427 | } | |
428 | ||
429 | /****************************************************************************** | |
430 | ||
431 | Device Boot-time Settings Routines | |
432 | ||
433 | *******************************************************************************/ | |
434 | ||
435 | /* Boot time configuration table */ | |
436 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
437 | ||
438 | /** | |
439 | * netdev_boot_setup_add - add new setup entry | |
440 | * @name: name of the device | |
441 | * @map: configured settings for the device | |
442 | * | |
443 | * Adds new setup entry to the dev_boot_setup list. The function | |
444 | * returns 0 on error and 1 on success. This is a generic routine to | |
445 | * all netdevices. | |
446 | */ | |
447 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
448 | { | |
449 | struct netdev_boot_setup *s; | |
450 | int i; | |
451 | ||
452 | s = dev_boot_setup; | |
453 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
454 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
455 | memset(s[i].name, 0, sizeof(s[i].name)); | |
456 | strcpy(s[i].name, name); | |
457 | memcpy(&s[i].map, map, sizeof(s[i].map)); | |
458 | break; | |
459 | } | |
460 | } | |
461 | ||
462 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
463 | } | |
464 | ||
465 | /** | |
466 | * netdev_boot_setup_check - check boot time settings | |
467 | * @dev: the netdevice | |
468 | * | |
469 | * Check boot time settings for the device. | |
470 | * The found settings are set for the device to be used | |
471 | * later in the device probing. | |
472 | * Returns 0 if no settings found, 1 if they are. | |
473 | */ | |
474 | int netdev_boot_setup_check(struct net_device *dev) | |
475 | { | |
476 | struct netdev_boot_setup *s = dev_boot_setup; | |
477 | int i; | |
478 | ||
479 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
480 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
481 | !strncmp(dev->name, s[i].name, strlen(s[i].name))) { | |
482 | dev->irq = s[i].map.irq; | |
483 | dev->base_addr = s[i].map.base_addr; | |
484 | dev->mem_start = s[i].map.mem_start; | |
485 | dev->mem_end = s[i].map.mem_end; | |
486 | return 1; | |
487 | } | |
488 | } | |
489 | return 0; | |
490 | } | |
491 | ||
492 | ||
493 | /** | |
494 | * netdev_boot_base - get address from boot time settings | |
495 | * @prefix: prefix for network device | |
496 | * @unit: id for network device | |
497 | * | |
498 | * Check boot time settings for the base address of device. | |
499 | * The found settings are set for the device to be used | |
500 | * later in the device probing. | |
501 | * Returns 0 if no settings found. | |
502 | */ | |
503 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
504 | { | |
505 | const struct netdev_boot_setup *s = dev_boot_setup; | |
506 | char name[IFNAMSIZ]; | |
507 | int i; | |
508 | ||
509 | sprintf(name, "%s%d", prefix, unit); | |
510 | ||
511 | /* | |
512 | * If device already registered then return base of 1 | |
513 | * to indicate not to probe for this interface | |
514 | */ | |
881d966b | 515 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
516 | return 1; |
517 | ||
518 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
519 | if (!strcmp(name, s[i].name)) | |
520 | return s[i].map.base_addr; | |
521 | return 0; | |
522 | } | |
523 | ||
524 | /* | |
525 | * Saves at boot time configured settings for any netdevice. | |
526 | */ | |
527 | int __init netdev_boot_setup(char *str) | |
528 | { | |
529 | int ints[5]; | |
530 | struct ifmap map; | |
531 | ||
532 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
533 | if (!str || !*str) | |
534 | return 0; | |
535 | ||
536 | /* Save settings */ | |
537 | memset(&map, 0, sizeof(map)); | |
538 | if (ints[0] > 0) | |
539 | map.irq = ints[1]; | |
540 | if (ints[0] > 1) | |
541 | map.base_addr = ints[2]; | |
542 | if (ints[0] > 2) | |
543 | map.mem_start = ints[3]; | |
544 | if (ints[0] > 3) | |
545 | map.mem_end = ints[4]; | |
546 | ||
547 | /* Add new entry to the list */ | |
548 | return netdev_boot_setup_add(str, &map); | |
549 | } | |
550 | ||
551 | __setup("netdev=", netdev_boot_setup); | |
552 | ||
553 | /******************************************************************************* | |
554 | ||
555 | Device Interface Subroutines | |
556 | ||
557 | *******************************************************************************/ | |
558 | ||
559 | /** | |
560 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 561 | * @net: the applicable net namespace |
1da177e4 LT |
562 | * @name: name to find |
563 | * | |
564 | * Find an interface by name. Must be called under RTNL semaphore | |
565 | * or @dev_base_lock. If the name is found a pointer to the device | |
566 | * is returned. If the name is not found then %NULL is returned. The | |
567 | * reference counters are not incremented so the caller must be | |
568 | * careful with locks. | |
569 | */ | |
570 | ||
881d966b | 571 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
572 | { |
573 | struct hlist_node *p; | |
574 | ||
881d966b | 575 | hlist_for_each(p, dev_name_hash(net, name)) { |
1da177e4 LT |
576 | struct net_device *dev |
577 | = hlist_entry(p, struct net_device, name_hlist); | |
578 | if (!strncmp(dev->name, name, IFNAMSIZ)) | |
579 | return dev; | |
580 | } | |
581 | return NULL; | |
582 | } | |
583 | ||
584 | /** | |
585 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 586 | * @net: the applicable net namespace |
1da177e4 LT |
587 | * @name: name to find |
588 | * | |
589 | * Find an interface by name. This can be called from any | |
590 | * context and does its own locking. The returned handle has | |
591 | * the usage count incremented and the caller must use dev_put() to | |
592 | * release it when it is no longer needed. %NULL is returned if no | |
593 | * matching device is found. | |
594 | */ | |
595 | ||
881d966b | 596 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
597 | { |
598 | struct net_device *dev; | |
599 | ||
600 | read_lock(&dev_base_lock); | |
881d966b | 601 | dev = __dev_get_by_name(net, name); |
1da177e4 LT |
602 | if (dev) |
603 | dev_hold(dev); | |
604 | read_unlock(&dev_base_lock); | |
605 | return dev; | |
606 | } | |
607 | ||
608 | /** | |
609 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 610 | * @net: the applicable net namespace |
1da177e4 LT |
611 | * @ifindex: index of device |
612 | * | |
613 | * Search for an interface by index. Returns %NULL if the device | |
614 | * is not found or a pointer to the device. The device has not | |
615 | * had its reference counter increased so the caller must be careful | |
616 | * about locking. The caller must hold either the RTNL semaphore | |
617 | * or @dev_base_lock. | |
618 | */ | |
619 | ||
881d966b | 620 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
621 | { |
622 | struct hlist_node *p; | |
623 | ||
881d966b | 624 | hlist_for_each(p, dev_index_hash(net, ifindex)) { |
1da177e4 LT |
625 | struct net_device *dev |
626 | = hlist_entry(p, struct net_device, index_hlist); | |
627 | if (dev->ifindex == ifindex) | |
628 | return dev; | |
629 | } | |
630 | return NULL; | |
631 | } | |
632 | ||
633 | ||
634 | /** | |
635 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 636 | * @net: the applicable net namespace |
1da177e4 LT |
637 | * @ifindex: index of device |
638 | * | |
639 | * Search for an interface by index. Returns NULL if the device | |
640 | * is not found or a pointer to the device. The device returned has | |
641 | * had a reference added and the pointer is safe until the user calls | |
642 | * dev_put to indicate they have finished with it. | |
643 | */ | |
644 | ||
881d966b | 645 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
646 | { |
647 | struct net_device *dev; | |
648 | ||
649 | read_lock(&dev_base_lock); | |
881d966b | 650 | dev = __dev_get_by_index(net, ifindex); |
1da177e4 LT |
651 | if (dev) |
652 | dev_hold(dev); | |
653 | read_unlock(&dev_base_lock); | |
654 | return dev; | |
655 | } | |
656 | ||
657 | /** | |
658 | * dev_getbyhwaddr - find a device by its hardware address | |
c4ea43c5 | 659 | * @net: the applicable net namespace |
1da177e4 LT |
660 | * @type: media type of device |
661 | * @ha: hardware address | |
662 | * | |
663 | * Search for an interface by MAC address. Returns NULL if the device | |
664 | * is not found or a pointer to the device. The caller must hold the | |
665 | * rtnl semaphore. The returned device has not had its ref count increased | |
666 | * and the caller must therefore be careful about locking | |
667 | * | |
668 | * BUGS: | |
669 | * If the API was consistent this would be __dev_get_by_hwaddr | |
670 | */ | |
671 | ||
881d966b | 672 | struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha) |
1da177e4 LT |
673 | { |
674 | struct net_device *dev; | |
675 | ||
676 | ASSERT_RTNL(); | |
677 | ||
81103a52 | 678 | for_each_netdev(net, dev) |
1da177e4 LT |
679 | if (dev->type == type && |
680 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
681 | return dev; |
682 | ||
683 | return NULL; | |
1da177e4 LT |
684 | } |
685 | ||
cf309e3f JF |
686 | EXPORT_SYMBOL(dev_getbyhwaddr); |
687 | ||
881d966b | 688 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
689 | { |
690 | struct net_device *dev; | |
691 | ||
4e9cac2b | 692 | ASSERT_RTNL(); |
881d966b | 693 | for_each_netdev(net, dev) |
4e9cac2b | 694 | if (dev->type == type) |
7562f876 PE |
695 | return dev; |
696 | ||
697 | return NULL; | |
4e9cac2b PM |
698 | } |
699 | ||
700 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); | |
701 | ||
881d966b | 702 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b PM |
703 | { |
704 | struct net_device *dev; | |
705 | ||
706 | rtnl_lock(); | |
881d966b | 707 | dev = __dev_getfirstbyhwtype(net, type); |
4e9cac2b PM |
708 | if (dev) |
709 | dev_hold(dev); | |
1da177e4 LT |
710 | rtnl_unlock(); |
711 | return dev; | |
712 | } | |
713 | ||
714 | EXPORT_SYMBOL(dev_getfirstbyhwtype); | |
715 | ||
716 | /** | |
717 | * dev_get_by_flags - find any device with given flags | |
c4ea43c5 | 718 | * @net: the applicable net namespace |
1da177e4 LT |
719 | * @if_flags: IFF_* values |
720 | * @mask: bitmask of bits in if_flags to check | |
721 | * | |
722 | * Search for any interface with the given flags. Returns NULL if a device | |
4ec93edb | 723 | * is not found or a pointer to the device. The device returned has |
1da177e4 LT |
724 | * had a reference added and the pointer is safe until the user calls |
725 | * dev_put to indicate they have finished with it. | |
726 | */ | |
727 | ||
881d966b | 728 | struct net_device * dev_get_by_flags(struct net *net, unsigned short if_flags, unsigned short mask) |
1da177e4 | 729 | { |
7562f876 | 730 | struct net_device *dev, *ret; |
1da177e4 | 731 | |
7562f876 | 732 | ret = NULL; |
1da177e4 | 733 | read_lock(&dev_base_lock); |
881d966b | 734 | for_each_netdev(net, dev) { |
1da177e4 LT |
735 | if (((dev->flags ^ if_flags) & mask) == 0) { |
736 | dev_hold(dev); | |
7562f876 | 737 | ret = dev; |
1da177e4 LT |
738 | break; |
739 | } | |
740 | } | |
741 | read_unlock(&dev_base_lock); | |
7562f876 | 742 | return ret; |
1da177e4 LT |
743 | } |
744 | ||
745 | /** | |
746 | * dev_valid_name - check if name is okay for network device | |
747 | * @name: name string | |
748 | * | |
749 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
750 | * to allow sysfs to work. We also disallow any kind of |
751 | * whitespace. | |
1da177e4 | 752 | */ |
c2373ee9 | 753 | int dev_valid_name(const char *name) |
1da177e4 | 754 | { |
c7fa9d18 DM |
755 | if (*name == '\0') |
756 | return 0; | |
b6fe17d6 SH |
757 | if (strlen(name) >= IFNAMSIZ) |
758 | return 0; | |
c7fa9d18 DM |
759 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
760 | return 0; | |
761 | ||
762 | while (*name) { | |
763 | if (*name == '/' || isspace(*name)) | |
764 | return 0; | |
765 | name++; | |
766 | } | |
767 | return 1; | |
1da177e4 LT |
768 | } |
769 | ||
770 | /** | |
b267b179 EB |
771 | * __dev_alloc_name - allocate a name for a device |
772 | * @net: network namespace to allocate the device name in | |
1da177e4 | 773 | * @name: name format string |
b267b179 | 774 | * @buf: scratch buffer and result name string |
1da177e4 LT |
775 | * |
776 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
777 | * id. It scans list of devices to build up a free map, then chooses |
778 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
779 | * while allocating the name and adding the device in order to avoid | |
780 | * duplicates. | |
781 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
782 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
783 | */ |
784 | ||
b267b179 | 785 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
786 | { |
787 | int i = 0; | |
1da177e4 LT |
788 | const char *p; |
789 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 790 | unsigned long *inuse; |
1da177e4 LT |
791 | struct net_device *d; |
792 | ||
793 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
794 | if (p) { | |
795 | /* | |
796 | * Verify the string as this thing may have come from | |
797 | * the user. There must be either one "%d" and no other "%" | |
798 | * characters. | |
799 | */ | |
800 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
801 | return -EINVAL; | |
802 | ||
803 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 804 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
805 | if (!inuse) |
806 | return -ENOMEM; | |
807 | ||
881d966b | 808 | for_each_netdev(net, d) { |
1da177e4 LT |
809 | if (!sscanf(d->name, name, &i)) |
810 | continue; | |
811 | if (i < 0 || i >= max_netdevices) | |
812 | continue; | |
813 | ||
814 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 815 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
816 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
817 | set_bit(i, inuse); | |
818 | } | |
819 | ||
820 | i = find_first_zero_bit(inuse, max_netdevices); | |
821 | free_page((unsigned long) inuse); | |
822 | } | |
823 | ||
b267b179 EB |
824 | snprintf(buf, IFNAMSIZ, name, i); |
825 | if (!__dev_get_by_name(net, buf)) | |
1da177e4 | 826 | return i; |
1da177e4 LT |
827 | |
828 | /* It is possible to run out of possible slots | |
829 | * when the name is long and there isn't enough space left | |
830 | * for the digits, or if all bits are used. | |
831 | */ | |
832 | return -ENFILE; | |
833 | } | |
834 | ||
b267b179 EB |
835 | /** |
836 | * dev_alloc_name - allocate a name for a device | |
837 | * @dev: device | |
838 | * @name: name format string | |
839 | * | |
840 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
841 | * id. It scans list of devices to build up a free map, then chooses | |
842 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
843 | * while allocating the name and adding the device in order to avoid | |
844 | * duplicates. | |
845 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
846 | * Returns the number of the unit assigned or a negative errno code. | |
847 | */ | |
848 | ||
849 | int dev_alloc_name(struct net_device *dev, const char *name) | |
850 | { | |
851 | char buf[IFNAMSIZ]; | |
852 | struct net *net; | |
853 | int ret; | |
854 | ||
855 | BUG_ON(!dev->nd_net); | |
856 | net = dev->nd_net; | |
857 | ret = __dev_alloc_name(net, name, buf); | |
858 | if (ret >= 0) | |
859 | strlcpy(dev->name, buf, IFNAMSIZ); | |
860 | return ret; | |
861 | } | |
862 | ||
1da177e4 LT |
863 | |
864 | /** | |
865 | * dev_change_name - change name of a device | |
866 | * @dev: device | |
867 | * @newname: name (or format string) must be at least IFNAMSIZ | |
868 | * | |
869 | * Change name of a device, can pass format strings "eth%d". | |
870 | * for wildcarding. | |
871 | */ | |
872 | int dev_change_name(struct net_device *dev, char *newname) | |
873 | { | |
fcc5a03a | 874 | char oldname[IFNAMSIZ]; |
1da177e4 | 875 | int err = 0; |
fcc5a03a | 876 | int ret; |
881d966b | 877 | struct net *net; |
1da177e4 LT |
878 | |
879 | ASSERT_RTNL(); | |
881d966b | 880 | BUG_ON(!dev->nd_net); |
1da177e4 | 881 | |
881d966b | 882 | net = dev->nd_net; |
1da177e4 LT |
883 | if (dev->flags & IFF_UP) |
884 | return -EBUSY; | |
885 | ||
886 | if (!dev_valid_name(newname)) | |
887 | return -EINVAL; | |
888 | ||
c8d90dca SH |
889 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) |
890 | return 0; | |
891 | ||
fcc5a03a HX |
892 | memcpy(oldname, dev->name, IFNAMSIZ); |
893 | ||
1da177e4 LT |
894 | if (strchr(newname, '%')) { |
895 | err = dev_alloc_name(dev, newname); | |
896 | if (err < 0) | |
897 | return err; | |
898 | strcpy(newname, dev->name); | |
899 | } | |
881d966b | 900 | else if (__dev_get_by_name(net, newname)) |
1da177e4 LT |
901 | return -EEXIST; |
902 | else | |
903 | strlcpy(dev->name, newname, IFNAMSIZ); | |
904 | ||
fcc5a03a | 905 | rollback: |
92749821 | 906 | device_rename(&dev->dev, dev->name); |
7f988eab HX |
907 | |
908 | write_lock_bh(&dev_base_lock); | |
92749821 | 909 | hlist_del(&dev->name_hlist); |
881d966b | 910 | hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name)); |
7f988eab HX |
911 | write_unlock_bh(&dev_base_lock); |
912 | ||
056925ab | 913 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
914 | ret = notifier_to_errno(ret); |
915 | ||
916 | if (ret) { | |
917 | if (err) { | |
918 | printk(KERN_ERR | |
919 | "%s: name change rollback failed: %d.\n", | |
920 | dev->name, ret); | |
921 | } else { | |
922 | err = ret; | |
923 | memcpy(dev->name, oldname, IFNAMSIZ); | |
924 | goto rollback; | |
925 | } | |
926 | } | |
1da177e4 LT |
927 | |
928 | return err; | |
929 | } | |
930 | ||
d8a33ac4 | 931 | /** |
3041a069 | 932 | * netdev_features_change - device changes features |
d8a33ac4 SH |
933 | * @dev: device to cause notification |
934 | * | |
935 | * Called to indicate a device has changed features. | |
936 | */ | |
937 | void netdev_features_change(struct net_device *dev) | |
938 | { | |
056925ab | 939 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
940 | } |
941 | EXPORT_SYMBOL(netdev_features_change); | |
942 | ||
1da177e4 LT |
943 | /** |
944 | * netdev_state_change - device changes state | |
945 | * @dev: device to cause notification | |
946 | * | |
947 | * Called to indicate a device has changed state. This function calls | |
948 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
949 | * to the routing socket. | |
950 | */ | |
951 | void netdev_state_change(struct net_device *dev) | |
952 | { | |
953 | if (dev->flags & IFF_UP) { | |
056925ab | 954 | call_netdevice_notifiers(NETDEV_CHANGE, dev); |
1da177e4 LT |
955 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0); |
956 | } | |
957 | } | |
958 | ||
959 | /** | |
960 | * dev_load - load a network module | |
c4ea43c5 | 961 | * @net: the applicable net namespace |
1da177e4 LT |
962 | * @name: name of interface |
963 | * | |
964 | * If a network interface is not present and the process has suitable | |
965 | * privileges this function loads the module. If module loading is not | |
966 | * available in this kernel then it becomes a nop. | |
967 | */ | |
968 | ||
881d966b | 969 | void dev_load(struct net *net, const char *name) |
1da177e4 | 970 | { |
4ec93edb | 971 | struct net_device *dev; |
1da177e4 LT |
972 | |
973 | read_lock(&dev_base_lock); | |
881d966b | 974 | dev = __dev_get_by_name(net, name); |
1da177e4 LT |
975 | read_unlock(&dev_base_lock); |
976 | ||
977 | if (!dev && capable(CAP_SYS_MODULE)) | |
978 | request_module("%s", name); | |
979 | } | |
980 | ||
1da177e4 LT |
981 | /** |
982 | * dev_open - prepare an interface for use. | |
983 | * @dev: device to open | |
984 | * | |
985 | * Takes a device from down to up state. The device's private open | |
986 | * function is invoked and then the multicast lists are loaded. Finally | |
987 | * the device is moved into the up state and a %NETDEV_UP message is | |
988 | * sent to the netdev notifier chain. | |
989 | * | |
990 | * Calling this function on an active interface is a nop. On a failure | |
991 | * a negative errno code is returned. | |
992 | */ | |
993 | int dev_open(struct net_device *dev) | |
994 | { | |
995 | int ret = 0; | |
996 | ||
997 | /* | |
998 | * Is it already up? | |
999 | */ | |
1000 | ||
1001 | if (dev->flags & IFF_UP) | |
1002 | return 0; | |
1003 | ||
1004 | /* | |
1005 | * Is it even present? | |
1006 | */ | |
1007 | if (!netif_device_present(dev)) | |
1008 | return -ENODEV; | |
1009 | ||
1010 | /* | |
1011 | * Call device private open method | |
1012 | */ | |
1013 | set_bit(__LINK_STATE_START, &dev->state); | |
bada339b JG |
1014 | |
1015 | if (dev->validate_addr) | |
1016 | ret = dev->validate_addr(dev); | |
1017 | ||
1018 | if (!ret && dev->open) | |
1da177e4 | 1019 | ret = dev->open(dev); |
1da177e4 | 1020 | |
4ec93edb | 1021 | /* |
1da177e4 LT |
1022 | * If it went open OK then: |
1023 | */ | |
1024 | ||
bada339b JG |
1025 | if (ret) |
1026 | clear_bit(__LINK_STATE_START, &dev->state); | |
1027 | else { | |
1da177e4 LT |
1028 | /* |
1029 | * Set the flags. | |
1030 | */ | |
1031 | dev->flags |= IFF_UP; | |
1032 | ||
1033 | /* | |
1034 | * Initialize multicasting status | |
1035 | */ | |
4417da66 | 1036 | dev_set_rx_mode(dev); |
1da177e4 LT |
1037 | |
1038 | /* | |
1039 | * Wakeup transmit queue engine | |
1040 | */ | |
1041 | dev_activate(dev); | |
1042 | ||
1043 | /* | |
1044 | * ... and announce new interface. | |
1045 | */ | |
056925ab | 1046 | call_netdevice_notifiers(NETDEV_UP, dev); |
1da177e4 | 1047 | } |
bada339b | 1048 | |
1da177e4 LT |
1049 | return ret; |
1050 | } | |
1051 | ||
1052 | /** | |
1053 | * dev_close - shutdown an interface. | |
1054 | * @dev: device to shutdown | |
1055 | * | |
1056 | * This function moves an active device into down state. A | |
1057 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1058 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1059 | * chain. | |
1060 | */ | |
1061 | int dev_close(struct net_device *dev) | |
1062 | { | |
9d5010db DM |
1063 | might_sleep(); |
1064 | ||
1da177e4 LT |
1065 | if (!(dev->flags & IFF_UP)) |
1066 | return 0; | |
1067 | ||
1068 | /* | |
1069 | * Tell people we are going down, so that they can | |
1070 | * prepare to death, when device is still operating. | |
1071 | */ | |
056925ab | 1072 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1073 | |
1da177e4 LT |
1074 | clear_bit(__LINK_STATE_START, &dev->state); |
1075 | ||
1076 | /* Synchronize to scheduled poll. We cannot touch poll list, | |
bea3348e SH |
1077 | * it can be even on different cpu. So just clear netif_running(). |
1078 | * | |
1079 | * dev->stop() will invoke napi_disable() on all of it's | |
1080 | * napi_struct instances on this device. | |
1081 | */ | |
1da177e4 | 1082 | smp_mb__after_clear_bit(); /* Commit netif_running(). */ |
1da177e4 | 1083 | |
d8b2a4d2 ML |
1084 | dev_deactivate(dev); |
1085 | ||
1da177e4 LT |
1086 | /* |
1087 | * Call the device specific close. This cannot fail. | |
1088 | * Only if device is UP | |
1089 | * | |
1090 | * We allow it to be called even after a DETACH hot-plug | |
1091 | * event. | |
1092 | */ | |
1093 | if (dev->stop) | |
1094 | dev->stop(dev); | |
1095 | ||
1096 | /* | |
1097 | * Device is now down. | |
1098 | */ | |
1099 | ||
1100 | dev->flags &= ~IFF_UP; | |
1101 | ||
1102 | /* | |
1103 | * Tell people we are down | |
1104 | */ | |
056925ab | 1105 | call_netdevice_notifiers(NETDEV_DOWN, dev); |
1da177e4 LT |
1106 | |
1107 | return 0; | |
1108 | } | |
1109 | ||
1110 | ||
881d966b EB |
1111 | static int dev_boot_phase = 1; |
1112 | ||
1da177e4 LT |
1113 | /* |
1114 | * Device change register/unregister. These are not inline or static | |
1115 | * as we export them to the world. | |
1116 | */ | |
1117 | ||
1118 | /** | |
1119 | * register_netdevice_notifier - register a network notifier block | |
1120 | * @nb: notifier | |
1121 | * | |
1122 | * Register a notifier to be called when network device events occur. | |
1123 | * The notifier passed is linked into the kernel structures and must | |
1124 | * not be reused until it has been unregistered. A negative errno code | |
1125 | * is returned on a failure. | |
1126 | * | |
1127 | * When registered all registration and up events are replayed | |
4ec93edb | 1128 | * to the new notifier to allow device to have a race free |
1da177e4 LT |
1129 | * view of the network device list. |
1130 | */ | |
1131 | ||
1132 | int register_netdevice_notifier(struct notifier_block *nb) | |
1133 | { | |
1134 | struct net_device *dev; | |
fcc5a03a | 1135 | struct net_device *last; |
881d966b | 1136 | struct net *net; |
1da177e4 LT |
1137 | int err; |
1138 | ||
1139 | rtnl_lock(); | |
f07d5b94 | 1140 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1141 | if (err) |
1142 | goto unlock; | |
881d966b EB |
1143 | if (dev_boot_phase) |
1144 | goto unlock; | |
1145 | for_each_net(net) { | |
1146 | for_each_netdev(net, dev) { | |
1147 | err = nb->notifier_call(nb, NETDEV_REGISTER, dev); | |
1148 | err = notifier_to_errno(err); | |
1149 | if (err) | |
1150 | goto rollback; | |
1151 | ||
1152 | if (!(dev->flags & IFF_UP)) | |
1153 | continue; | |
1da177e4 | 1154 | |
881d966b EB |
1155 | nb->notifier_call(nb, NETDEV_UP, dev); |
1156 | } | |
1da177e4 | 1157 | } |
fcc5a03a HX |
1158 | |
1159 | unlock: | |
1da177e4 LT |
1160 | rtnl_unlock(); |
1161 | return err; | |
fcc5a03a HX |
1162 | |
1163 | rollback: | |
1164 | last = dev; | |
881d966b EB |
1165 | for_each_net(net) { |
1166 | for_each_netdev(net, dev) { | |
1167 | if (dev == last) | |
1168 | break; | |
fcc5a03a | 1169 | |
881d966b EB |
1170 | if (dev->flags & IFF_UP) { |
1171 | nb->notifier_call(nb, NETDEV_GOING_DOWN, dev); | |
1172 | nb->notifier_call(nb, NETDEV_DOWN, dev); | |
1173 | } | |
1174 | nb->notifier_call(nb, NETDEV_UNREGISTER, dev); | |
fcc5a03a | 1175 | } |
fcc5a03a | 1176 | } |
c67625a1 PE |
1177 | |
1178 | raw_notifier_chain_unregister(&netdev_chain, nb); | |
fcc5a03a | 1179 | goto unlock; |
1da177e4 LT |
1180 | } |
1181 | ||
1182 | /** | |
1183 | * unregister_netdevice_notifier - unregister a network notifier block | |
1184 | * @nb: notifier | |
1185 | * | |
1186 | * Unregister a notifier previously registered by | |
1187 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1188 | * kernel structures and may then be reused. A negative errno code | |
1189 | * is returned on a failure. | |
1190 | */ | |
1191 | ||
1192 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1193 | { | |
9f514950 HX |
1194 | int err; |
1195 | ||
1196 | rtnl_lock(); | |
f07d5b94 | 1197 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
9f514950 HX |
1198 | rtnl_unlock(); |
1199 | return err; | |
1da177e4 LT |
1200 | } |
1201 | ||
1202 | /** | |
1203 | * call_netdevice_notifiers - call all network notifier blocks | |
1204 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1205 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1206 | * |
1207 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1208 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1209 | */ |
1210 | ||
ad7379d4 | 1211 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1212 | { |
ad7379d4 | 1213 | return raw_notifier_call_chain(&netdev_chain, val, dev); |
1da177e4 LT |
1214 | } |
1215 | ||
1216 | /* When > 0 there are consumers of rx skb time stamps */ | |
1217 | static atomic_t netstamp_needed = ATOMIC_INIT(0); | |
1218 | ||
1219 | void net_enable_timestamp(void) | |
1220 | { | |
1221 | atomic_inc(&netstamp_needed); | |
1222 | } | |
1223 | ||
1224 | void net_disable_timestamp(void) | |
1225 | { | |
1226 | atomic_dec(&netstamp_needed); | |
1227 | } | |
1228 | ||
a61bbcf2 | 1229 | static inline void net_timestamp(struct sk_buff *skb) |
1da177e4 LT |
1230 | { |
1231 | if (atomic_read(&netstamp_needed)) | |
a61bbcf2 | 1232 | __net_timestamp(skb); |
b7aa0bf7 ED |
1233 | else |
1234 | skb->tstamp.tv64 = 0; | |
1da177e4 LT |
1235 | } |
1236 | ||
1237 | /* | |
1238 | * Support routine. Sends outgoing frames to any network | |
1239 | * taps currently in use. | |
1240 | */ | |
1241 | ||
f6a78bfc | 1242 | static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
1243 | { |
1244 | struct packet_type *ptype; | |
a61bbcf2 PM |
1245 | |
1246 | net_timestamp(skb); | |
1da177e4 LT |
1247 | |
1248 | rcu_read_lock(); | |
1249 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
1250 | /* Never send packets back to the socket | |
1251 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1252 | */ | |
1253 | if ((ptype->dev == dev || !ptype->dev) && | |
1254 | (ptype->af_packet_priv == NULL || | |
1255 | (struct sock *)ptype->af_packet_priv != skb->sk)) { | |
1256 | struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC); | |
1257 | if (!skb2) | |
1258 | break; | |
1259 | ||
1260 | /* skb->nh should be correctly | |
1261 | set by sender, so that the second statement is | |
1262 | just protection against buggy protocols. | |
1263 | */ | |
459a98ed | 1264 | skb_reset_mac_header(skb2); |
1da177e4 | 1265 | |
d56f90a7 | 1266 | if (skb_network_header(skb2) < skb2->data || |
27a884dc | 1267 | skb2->network_header > skb2->tail) { |
1da177e4 LT |
1268 | if (net_ratelimit()) |
1269 | printk(KERN_CRIT "protocol %04x is " | |
1270 | "buggy, dev %s\n", | |
1271 | skb2->protocol, dev->name); | |
c1d2bbe1 | 1272 | skb_reset_network_header(skb2); |
1da177e4 LT |
1273 | } |
1274 | ||
b0e380b1 | 1275 | skb2->transport_header = skb2->network_header; |
1da177e4 | 1276 | skb2->pkt_type = PACKET_OUTGOING; |
f2ccd8fa | 1277 | ptype->func(skb2, skb->dev, ptype, skb->dev); |
1da177e4 LT |
1278 | } |
1279 | } | |
1280 | rcu_read_unlock(); | |
1281 | } | |
1282 | ||
56079431 DV |
1283 | |
1284 | void __netif_schedule(struct net_device *dev) | |
1285 | { | |
1286 | if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) { | |
1287 | unsigned long flags; | |
1288 | struct softnet_data *sd; | |
1289 | ||
1290 | local_irq_save(flags); | |
1291 | sd = &__get_cpu_var(softnet_data); | |
1292 | dev->next_sched = sd->output_queue; | |
1293 | sd->output_queue = dev; | |
1294 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
1295 | local_irq_restore(flags); | |
1296 | } | |
1297 | } | |
1298 | EXPORT_SYMBOL(__netif_schedule); | |
1299 | ||
bea3348e | 1300 | void dev_kfree_skb_irq(struct sk_buff *skb) |
56079431 | 1301 | { |
bea3348e SH |
1302 | if (atomic_dec_and_test(&skb->users)) { |
1303 | struct softnet_data *sd; | |
1304 | unsigned long flags; | |
56079431 | 1305 | |
bea3348e SH |
1306 | local_irq_save(flags); |
1307 | sd = &__get_cpu_var(softnet_data); | |
1308 | skb->next = sd->completion_queue; | |
1309 | sd->completion_queue = skb; | |
1310 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
1311 | local_irq_restore(flags); | |
1312 | } | |
56079431 | 1313 | } |
bea3348e | 1314 | EXPORT_SYMBOL(dev_kfree_skb_irq); |
56079431 DV |
1315 | |
1316 | void dev_kfree_skb_any(struct sk_buff *skb) | |
1317 | { | |
1318 | if (in_irq() || irqs_disabled()) | |
1319 | dev_kfree_skb_irq(skb); | |
1320 | else | |
1321 | dev_kfree_skb(skb); | |
1322 | } | |
1323 | EXPORT_SYMBOL(dev_kfree_skb_any); | |
1324 | ||
1325 | ||
bea3348e SH |
1326 | /** |
1327 | * netif_device_detach - mark device as removed | |
1328 | * @dev: network device | |
1329 | * | |
1330 | * Mark device as removed from system and therefore no longer available. | |
1331 | */ | |
56079431 DV |
1332 | void netif_device_detach(struct net_device *dev) |
1333 | { | |
1334 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
1335 | netif_running(dev)) { | |
1336 | netif_stop_queue(dev); | |
1337 | } | |
1338 | } | |
1339 | EXPORT_SYMBOL(netif_device_detach); | |
1340 | ||
bea3348e SH |
1341 | /** |
1342 | * netif_device_attach - mark device as attached | |
1343 | * @dev: network device | |
1344 | * | |
1345 | * Mark device as attached from system and restart if needed. | |
1346 | */ | |
56079431 DV |
1347 | void netif_device_attach(struct net_device *dev) |
1348 | { | |
1349 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
1350 | netif_running(dev)) { | |
1351 | netif_wake_queue(dev); | |
4ec93edb | 1352 | __netdev_watchdog_up(dev); |
56079431 DV |
1353 | } |
1354 | } | |
1355 | EXPORT_SYMBOL(netif_device_attach); | |
1356 | ||
1357 | ||
1da177e4 LT |
1358 | /* |
1359 | * Invalidate hardware checksum when packet is to be mangled, and | |
1360 | * complete checksum manually on outgoing path. | |
1361 | */ | |
84fa7933 | 1362 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 1363 | { |
d3bc23e7 | 1364 | __wsum csum; |
663ead3b | 1365 | int ret = 0, offset; |
1da177e4 | 1366 | |
84fa7933 | 1367 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
1368 | goto out_set_summed; |
1369 | ||
1370 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
a430a43d HX |
1371 | /* Let GSO fix up the checksum. */ |
1372 | goto out_set_summed; | |
1da177e4 LT |
1373 | } |
1374 | ||
a030847e HX |
1375 | offset = skb->csum_start - skb_headroom(skb); |
1376 | BUG_ON(offset >= skb_headlen(skb)); | |
1377 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
1378 | ||
1379 | offset += skb->csum_offset; | |
1380 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
1381 | ||
1382 | if (skb_cloned(skb) && | |
1383 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1da177e4 LT |
1384 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
1385 | if (ret) | |
1386 | goto out; | |
1387 | } | |
1388 | ||
a030847e | 1389 | *(__sum16 *)(skb->data + offset) = csum_fold(csum); |
a430a43d | 1390 | out_set_summed: |
1da177e4 | 1391 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 1392 | out: |
1da177e4 LT |
1393 | return ret; |
1394 | } | |
1395 | ||
f6a78bfc HX |
1396 | /** |
1397 | * skb_gso_segment - Perform segmentation on skb. | |
1398 | * @skb: buffer to segment | |
576a30eb | 1399 | * @features: features for the output path (see dev->features) |
f6a78bfc HX |
1400 | * |
1401 | * This function segments the given skb and returns a list of segments. | |
576a30eb HX |
1402 | * |
1403 | * It may return NULL if the skb requires no segmentation. This is | |
1404 | * only possible when GSO is used for verifying header integrity. | |
f6a78bfc | 1405 | */ |
576a30eb | 1406 | struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features) |
f6a78bfc HX |
1407 | { |
1408 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
1409 | struct packet_type *ptype; | |
252e3346 | 1410 | __be16 type = skb->protocol; |
a430a43d | 1411 | int err; |
f6a78bfc HX |
1412 | |
1413 | BUG_ON(skb_shinfo(skb)->frag_list); | |
f6a78bfc | 1414 | |
459a98ed | 1415 | skb_reset_mac_header(skb); |
b0e380b1 | 1416 | skb->mac_len = skb->network_header - skb->mac_header; |
f6a78bfc HX |
1417 | __skb_pull(skb, skb->mac_len); |
1418 | ||
f9d106a6 | 1419 | if (WARN_ON(skb->ip_summed != CHECKSUM_PARTIAL)) { |
a430a43d HX |
1420 | if (skb_header_cloned(skb) && |
1421 | (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))) | |
1422 | return ERR_PTR(err); | |
1423 | } | |
1424 | ||
f6a78bfc | 1425 | rcu_read_lock(); |
82d8a867 PE |
1426 | list_for_each_entry_rcu(ptype, |
1427 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
f6a78bfc | 1428 | if (ptype->type == type && !ptype->dev && ptype->gso_segment) { |
84fa7933 | 1429 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { |
a430a43d HX |
1430 | err = ptype->gso_send_check(skb); |
1431 | segs = ERR_PTR(err); | |
1432 | if (err || skb_gso_ok(skb, features)) | |
1433 | break; | |
d56f90a7 ACM |
1434 | __skb_push(skb, (skb->data - |
1435 | skb_network_header(skb))); | |
a430a43d | 1436 | } |
576a30eb | 1437 | segs = ptype->gso_segment(skb, features); |
f6a78bfc HX |
1438 | break; |
1439 | } | |
1440 | } | |
1441 | rcu_read_unlock(); | |
1442 | ||
98e399f8 | 1443 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 1444 | |
f6a78bfc HX |
1445 | return segs; |
1446 | } | |
1447 | ||
1448 | EXPORT_SYMBOL(skb_gso_segment); | |
1449 | ||
fb286bb2 HX |
1450 | /* Take action when hardware reception checksum errors are detected. */ |
1451 | #ifdef CONFIG_BUG | |
1452 | void netdev_rx_csum_fault(struct net_device *dev) | |
1453 | { | |
1454 | if (net_ratelimit()) { | |
4ec93edb | 1455 | printk(KERN_ERR "%s: hw csum failure.\n", |
246a4212 | 1456 | dev ? dev->name : "<unknown>"); |
fb286bb2 HX |
1457 | dump_stack(); |
1458 | } | |
1459 | } | |
1460 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
1461 | #endif | |
1462 | ||
1da177e4 LT |
1463 | /* Actually, we should eliminate this check as soon as we know, that: |
1464 | * 1. IOMMU is present and allows to map all the memory. | |
1465 | * 2. No high memory really exists on this machine. | |
1466 | */ | |
1467 | ||
1468 | static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb) | |
1469 | { | |
3d3a8533 | 1470 | #ifdef CONFIG_HIGHMEM |
1da177e4 LT |
1471 | int i; |
1472 | ||
1473 | if (dev->features & NETIF_F_HIGHDMA) | |
1474 | return 0; | |
1475 | ||
1476 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) | |
1477 | if (PageHighMem(skb_shinfo(skb)->frags[i].page)) | |
1478 | return 1; | |
1479 | ||
3d3a8533 | 1480 | #endif |
1da177e4 LT |
1481 | return 0; |
1482 | } | |
1da177e4 | 1483 | |
f6a78bfc HX |
1484 | struct dev_gso_cb { |
1485 | void (*destructor)(struct sk_buff *skb); | |
1486 | }; | |
1487 | ||
1488 | #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb) | |
1489 | ||
1490 | static void dev_gso_skb_destructor(struct sk_buff *skb) | |
1491 | { | |
1492 | struct dev_gso_cb *cb; | |
1493 | ||
1494 | do { | |
1495 | struct sk_buff *nskb = skb->next; | |
1496 | ||
1497 | skb->next = nskb->next; | |
1498 | nskb->next = NULL; | |
1499 | kfree_skb(nskb); | |
1500 | } while (skb->next); | |
1501 | ||
1502 | cb = DEV_GSO_CB(skb); | |
1503 | if (cb->destructor) | |
1504 | cb->destructor(skb); | |
1505 | } | |
1506 | ||
1507 | /** | |
1508 | * dev_gso_segment - Perform emulated hardware segmentation on skb. | |
1509 | * @skb: buffer to segment | |
1510 | * | |
1511 | * This function segments the given skb and stores the list of segments | |
1512 | * in skb->next. | |
1513 | */ | |
1514 | static int dev_gso_segment(struct sk_buff *skb) | |
1515 | { | |
1516 | struct net_device *dev = skb->dev; | |
1517 | struct sk_buff *segs; | |
576a30eb HX |
1518 | int features = dev->features & ~(illegal_highdma(dev, skb) ? |
1519 | NETIF_F_SG : 0); | |
1520 | ||
1521 | segs = skb_gso_segment(skb, features); | |
1522 | ||
1523 | /* Verifying header integrity only. */ | |
1524 | if (!segs) | |
1525 | return 0; | |
f6a78bfc | 1526 | |
f6a78bfc HX |
1527 | if (unlikely(IS_ERR(segs))) |
1528 | return PTR_ERR(segs); | |
1529 | ||
1530 | skb->next = segs; | |
1531 | DEV_GSO_CB(skb)->destructor = skb->destructor; | |
1532 | skb->destructor = dev_gso_skb_destructor; | |
1533 | ||
1534 | return 0; | |
1535 | } | |
1536 | ||
1537 | int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev) | |
1538 | { | |
1539 | if (likely(!skb->next)) { | |
9be9a6b9 | 1540 | if (!list_empty(&ptype_all)) |
f6a78bfc HX |
1541 | dev_queue_xmit_nit(skb, dev); |
1542 | ||
576a30eb HX |
1543 | if (netif_needs_gso(dev, skb)) { |
1544 | if (unlikely(dev_gso_segment(skb))) | |
1545 | goto out_kfree_skb; | |
1546 | if (skb->next) | |
1547 | goto gso; | |
1548 | } | |
f6a78bfc | 1549 | |
576a30eb | 1550 | return dev->hard_start_xmit(skb, dev); |
f6a78bfc HX |
1551 | } |
1552 | ||
576a30eb | 1553 | gso: |
f6a78bfc HX |
1554 | do { |
1555 | struct sk_buff *nskb = skb->next; | |
1556 | int rc; | |
1557 | ||
1558 | skb->next = nskb->next; | |
1559 | nskb->next = NULL; | |
1560 | rc = dev->hard_start_xmit(nskb, dev); | |
1561 | if (unlikely(rc)) { | |
f54d9e8d | 1562 | nskb->next = skb->next; |
f6a78bfc HX |
1563 | skb->next = nskb; |
1564 | return rc; | |
1565 | } | |
f25f4e44 | 1566 | if (unlikely((netif_queue_stopped(dev) || |
668f895a | 1567 | netif_subqueue_stopped(dev, skb)) && |
f25f4e44 | 1568 | skb->next)) |
f54d9e8d | 1569 | return NETDEV_TX_BUSY; |
f6a78bfc | 1570 | } while (skb->next); |
4ec93edb | 1571 | |
f6a78bfc HX |
1572 | skb->destructor = DEV_GSO_CB(skb)->destructor; |
1573 | ||
1574 | out_kfree_skb: | |
1575 | kfree_skb(skb); | |
1576 | return 0; | |
1577 | } | |
1578 | ||
1da177e4 LT |
1579 | /** |
1580 | * dev_queue_xmit - transmit a buffer | |
1581 | * @skb: buffer to transmit | |
1582 | * | |
1583 | * Queue a buffer for transmission to a network device. The caller must | |
1584 | * have set the device and priority and built the buffer before calling | |
1585 | * this function. The function can be called from an interrupt. | |
1586 | * | |
1587 | * A negative errno code is returned on a failure. A success does not | |
1588 | * guarantee the frame will be transmitted as it may be dropped due | |
1589 | * to congestion or traffic shaping. | |
af191367 BG |
1590 | * |
1591 | * ----------------------------------------------------------------------------------- | |
1592 | * I notice this method can also return errors from the queue disciplines, | |
1593 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
1594 | * be positive. | |
1595 | * | |
1596 | * Regardless of the return value, the skb is consumed, so it is currently | |
1597 | * difficult to retry a send to this method. (You can bump the ref count | |
1598 | * before sending to hold a reference for retry if you are careful.) | |
1599 | * | |
1600 | * When calling this method, interrupts MUST be enabled. This is because | |
1601 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
1602 | * --BLG | |
1da177e4 LT |
1603 | */ |
1604 | ||
1605 | int dev_queue_xmit(struct sk_buff *skb) | |
1606 | { | |
1607 | struct net_device *dev = skb->dev; | |
1608 | struct Qdisc *q; | |
1609 | int rc = -ENOMEM; | |
1610 | ||
f6a78bfc HX |
1611 | /* GSO will handle the following emulations directly. */ |
1612 | if (netif_needs_gso(dev, skb)) | |
1613 | goto gso; | |
1614 | ||
1da177e4 LT |
1615 | if (skb_shinfo(skb)->frag_list && |
1616 | !(dev->features & NETIF_F_FRAGLIST) && | |
364c6bad | 1617 | __skb_linearize(skb)) |
1da177e4 LT |
1618 | goto out_kfree_skb; |
1619 | ||
1620 | /* Fragmented skb is linearized if device does not support SG, | |
1621 | * or if at least one of fragments is in highmem and device | |
1622 | * does not support DMA from it. | |
1623 | */ | |
1624 | if (skb_shinfo(skb)->nr_frags && | |
1625 | (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) && | |
364c6bad | 1626 | __skb_linearize(skb)) |
1da177e4 LT |
1627 | goto out_kfree_skb; |
1628 | ||
1629 | /* If packet is not checksummed and device does not support | |
1630 | * checksumming for this protocol, complete checksumming here. | |
1631 | */ | |
663ead3b HX |
1632 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
1633 | skb_set_transport_header(skb, skb->csum_start - | |
1634 | skb_headroom(skb)); | |
1635 | ||
a298830c HX |
1636 | if (!(dev->features & NETIF_F_GEN_CSUM) && |
1637 | !((dev->features & NETIF_F_IP_CSUM) && | |
1638 | skb->protocol == htons(ETH_P_IP)) && | |
1639 | !((dev->features & NETIF_F_IPV6_CSUM) && | |
1640 | skb->protocol == htons(ETH_P_IPV6))) | |
663ead3b HX |
1641 | if (skb_checksum_help(skb)) |
1642 | goto out_kfree_skb; | |
1643 | } | |
1da177e4 | 1644 | |
f6a78bfc | 1645 | gso: |
2d7ceece ED |
1646 | spin_lock_prefetch(&dev->queue_lock); |
1647 | ||
4ec93edb YH |
1648 | /* Disable soft irqs for various locks below. Also |
1649 | * stops preemption for RCU. | |
1da177e4 | 1650 | */ |
4ec93edb | 1651 | rcu_read_lock_bh(); |
1da177e4 | 1652 | |
4ec93edb YH |
1653 | /* Updates of qdisc are serialized by queue_lock. |
1654 | * The struct Qdisc which is pointed to by qdisc is now a | |
1655 | * rcu structure - it may be accessed without acquiring | |
1da177e4 | 1656 | * a lock (but the structure may be stale.) The freeing of the |
4ec93edb | 1657 | * qdisc will be deferred until it's known that there are no |
1da177e4 | 1658 | * more references to it. |
4ec93edb YH |
1659 | * |
1660 | * If the qdisc has an enqueue function, we still need to | |
1da177e4 LT |
1661 | * hold the queue_lock before calling it, since queue_lock |
1662 | * also serializes access to the device queue. | |
1663 | */ | |
1664 | ||
1665 | q = rcu_dereference(dev->qdisc); | |
1666 | #ifdef CONFIG_NET_CLS_ACT | |
1667 | skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS); | |
1668 | #endif | |
1669 | if (q->enqueue) { | |
1670 | /* Grab device queue */ | |
1671 | spin_lock(&dev->queue_lock); | |
85670cc1 PM |
1672 | q = dev->qdisc; |
1673 | if (q->enqueue) { | |
f25f4e44 | 1674 | /* reset queue_mapping to zero */ |
dfa40911 | 1675 | skb_set_queue_mapping(skb, 0); |
85670cc1 PM |
1676 | rc = q->enqueue(skb, q); |
1677 | qdisc_run(dev); | |
1678 | spin_unlock(&dev->queue_lock); | |
1da177e4 | 1679 | |
85670cc1 PM |
1680 | rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc; |
1681 | goto out; | |
1682 | } | |
1da177e4 | 1683 | spin_unlock(&dev->queue_lock); |
1da177e4 LT |
1684 | } |
1685 | ||
1686 | /* The device has no queue. Common case for software devices: | |
1687 | loopback, all the sorts of tunnels... | |
1688 | ||
932ff279 HX |
1689 | Really, it is unlikely that netif_tx_lock protection is necessary |
1690 | here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
1da177e4 LT |
1691 | counters.) |
1692 | However, it is possible, that they rely on protection | |
1693 | made by us here. | |
1694 | ||
1695 | Check this and shot the lock. It is not prone from deadlocks. | |
1696 | Either shot noqueue qdisc, it is even simpler 8) | |
1697 | */ | |
1698 | if (dev->flags & IFF_UP) { | |
1699 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
1700 | ||
1701 | if (dev->xmit_lock_owner != cpu) { | |
1702 | ||
1703 | HARD_TX_LOCK(dev, cpu); | |
1704 | ||
f25f4e44 | 1705 | if (!netif_queue_stopped(dev) && |
668f895a | 1706 | !netif_subqueue_stopped(dev, skb)) { |
1da177e4 | 1707 | rc = 0; |
f6a78bfc | 1708 | if (!dev_hard_start_xmit(skb, dev)) { |
1da177e4 LT |
1709 | HARD_TX_UNLOCK(dev); |
1710 | goto out; | |
1711 | } | |
1712 | } | |
1713 | HARD_TX_UNLOCK(dev); | |
1714 | if (net_ratelimit()) | |
1715 | printk(KERN_CRIT "Virtual device %s asks to " | |
1716 | "queue packet!\n", dev->name); | |
1717 | } else { | |
1718 | /* Recursion is detected! It is possible, | |
1719 | * unfortunately */ | |
1720 | if (net_ratelimit()) | |
1721 | printk(KERN_CRIT "Dead loop on virtual device " | |
1722 | "%s, fix it urgently!\n", dev->name); | |
1723 | } | |
1724 | } | |
1725 | ||
1726 | rc = -ENETDOWN; | |
d4828d85 | 1727 | rcu_read_unlock_bh(); |
1da177e4 LT |
1728 | |
1729 | out_kfree_skb: | |
1730 | kfree_skb(skb); | |
1731 | return rc; | |
1732 | out: | |
d4828d85 | 1733 | rcu_read_unlock_bh(); |
1da177e4 LT |
1734 | return rc; |
1735 | } | |
1736 | ||
1737 | ||
1738 | /*======================================================================= | |
1739 | Receiver routines | |
1740 | =======================================================================*/ | |
1741 | ||
6b2bedc3 SH |
1742 | int netdev_max_backlog __read_mostly = 1000; |
1743 | int netdev_budget __read_mostly = 300; | |
1744 | int weight_p __read_mostly = 64; /* old backlog weight */ | |
1da177e4 LT |
1745 | |
1746 | DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, }; | |
1747 | ||
1748 | ||
1da177e4 LT |
1749 | /** |
1750 | * netif_rx - post buffer to the network code | |
1751 | * @skb: buffer to post | |
1752 | * | |
1753 | * This function receives a packet from a device driver and queues it for | |
1754 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
1755 | * may be dropped during processing for congestion control or by the | |
1756 | * protocol layers. | |
1757 | * | |
1758 | * return values: | |
1759 | * NET_RX_SUCCESS (no congestion) | |
1da177e4 LT |
1760 | * NET_RX_DROP (packet was dropped) |
1761 | * | |
1762 | */ | |
1763 | ||
1764 | int netif_rx(struct sk_buff *skb) | |
1765 | { | |
1da177e4 LT |
1766 | struct softnet_data *queue; |
1767 | unsigned long flags; | |
1768 | ||
1769 | /* if netpoll wants it, pretend we never saw it */ | |
1770 | if (netpoll_rx(skb)) | |
1771 | return NET_RX_DROP; | |
1772 | ||
b7aa0bf7 | 1773 | if (!skb->tstamp.tv64) |
a61bbcf2 | 1774 | net_timestamp(skb); |
1da177e4 LT |
1775 | |
1776 | /* | |
1777 | * The code is rearranged so that the path is the most | |
1778 | * short when CPU is congested, but is still operating. | |
1779 | */ | |
1780 | local_irq_save(flags); | |
1da177e4 LT |
1781 | queue = &__get_cpu_var(softnet_data); |
1782 | ||
1783 | __get_cpu_var(netdev_rx_stat).total++; | |
1784 | if (queue->input_pkt_queue.qlen <= netdev_max_backlog) { | |
1785 | if (queue->input_pkt_queue.qlen) { | |
1da177e4 LT |
1786 | enqueue: |
1787 | dev_hold(skb->dev); | |
1788 | __skb_queue_tail(&queue->input_pkt_queue, skb); | |
1da177e4 | 1789 | local_irq_restore(flags); |
34008d8c | 1790 | return NET_RX_SUCCESS; |
1da177e4 LT |
1791 | } |
1792 | ||
bea3348e | 1793 | napi_schedule(&queue->backlog); |
1da177e4 LT |
1794 | goto enqueue; |
1795 | } | |
1796 | ||
1da177e4 LT |
1797 | __get_cpu_var(netdev_rx_stat).dropped++; |
1798 | local_irq_restore(flags); | |
1799 | ||
1800 | kfree_skb(skb); | |
1801 | return NET_RX_DROP; | |
1802 | } | |
1803 | ||
1804 | int netif_rx_ni(struct sk_buff *skb) | |
1805 | { | |
1806 | int err; | |
1807 | ||
1808 | preempt_disable(); | |
1809 | err = netif_rx(skb); | |
1810 | if (local_softirq_pending()) | |
1811 | do_softirq(); | |
1812 | preempt_enable(); | |
1813 | ||
1814 | return err; | |
1815 | } | |
1816 | ||
1817 | EXPORT_SYMBOL(netif_rx_ni); | |
1818 | ||
f2ccd8fa | 1819 | static inline struct net_device *skb_bond(struct sk_buff *skb) |
1da177e4 LT |
1820 | { |
1821 | struct net_device *dev = skb->dev; | |
1822 | ||
8f903c70 | 1823 | if (dev->master) { |
7ea49ed7 | 1824 | if (skb_bond_should_drop(skb)) { |
8f903c70 JV |
1825 | kfree_skb(skb); |
1826 | return NULL; | |
1827 | } | |
1da177e4 | 1828 | skb->dev = dev->master; |
8f903c70 | 1829 | } |
f2ccd8fa DM |
1830 | |
1831 | return dev; | |
1da177e4 LT |
1832 | } |
1833 | ||
bea3348e | 1834 | |
1da177e4 LT |
1835 | static void net_tx_action(struct softirq_action *h) |
1836 | { | |
1837 | struct softnet_data *sd = &__get_cpu_var(softnet_data); | |
1838 | ||
1839 | if (sd->completion_queue) { | |
1840 | struct sk_buff *clist; | |
1841 | ||
1842 | local_irq_disable(); | |
1843 | clist = sd->completion_queue; | |
1844 | sd->completion_queue = NULL; | |
1845 | local_irq_enable(); | |
1846 | ||
1847 | while (clist) { | |
1848 | struct sk_buff *skb = clist; | |
1849 | clist = clist->next; | |
1850 | ||
1851 | BUG_TRAP(!atomic_read(&skb->users)); | |
1852 | __kfree_skb(skb); | |
1853 | } | |
1854 | } | |
1855 | ||
1856 | if (sd->output_queue) { | |
1857 | struct net_device *head; | |
1858 | ||
1859 | local_irq_disable(); | |
1860 | head = sd->output_queue; | |
1861 | sd->output_queue = NULL; | |
1862 | local_irq_enable(); | |
1863 | ||
1864 | while (head) { | |
1865 | struct net_device *dev = head; | |
1866 | head = head->next_sched; | |
1867 | ||
1868 | smp_mb__before_clear_bit(); | |
1869 | clear_bit(__LINK_STATE_SCHED, &dev->state); | |
1870 | ||
1871 | if (spin_trylock(&dev->queue_lock)) { | |
1872 | qdisc_run(dev); | |
1873 | spin_unlock(&dev->queue_lock); | |
1874 | } else { | |
1875 | netif_schedule(dev); | |
1876 | } | |
1877 | } | |
1878 | } | |
1879 | } | |
1880 | ||
6f05f629 SH |
1881 | static inline int deliver_skb(struct sk_buff *skb, |
1882 | struct packet_type *pt_prev, | |
1883 | struct net_device *orig_dev) | |
1da177e4 LT |
1884 | { |
1885 | atomic_inc(&skb->users); | |
f2ccd8fa | 1886 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
1da177e4 LT |
1887 | } |
1888 | ||
1889 | #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE) | |
6229e362 | 1890 | /* These hooks defined here for ATM */ |
1da177e4 LT |
1891 | struct net_bridge; |
1892 | struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br, | |
1893 | unsigned char *addr); | |
6229e362 | 1894 | void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent) __read_mostly; |
1da177e4 | 1895 | |
6229e362 SH |
1896 | /* |
1897 | * If bridge module is loaded call bridging hook. | |
1898 | * returns NULL if packet was consumed. | |
1899 | */ | |
1900 | struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p, | |
1901 | struct sk_buff *skb) __read_mostly; | |
1902 | static inline struct sk_buff *handle_bridge(struct sk_buff *skb, | |
1903 | struct packet_type **pt_prev, int *ret, | |
1904 | struct net_device *orig_dev) | |
1da177e4 LT |
1905 | { |
1906 | struct net_bridge_port *port; | |
1907 | ||
6229e362 SH |
1908 | if (skb->pkt_type == PACKET_LOOPBACK || |
1909 | (port = rcu_dereference(skb->dev->br_port)) == NULL) | |
1910 | return skb; | |
1da177e4 LT |
1911 | |
1912 | if (*pt_prev) { | |
6229e362 | 1913 | *ret = deliver_skb(skb, *pt_prev, orig_dev); |
1da177e4 | 1914 | *pt_prev = NULL; |
4ec93edb YH |
1915 | } |
1916 | ||
6229e362 | 1917 | return br_handle_frame_hook(port, skb); |
1da177e4 LT |
1918 | } |
1919 | #else | |
6229e362 | 1920 | #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb) |
1da177e4 LT |
1921 | #endif |
1922 | ||
b863ceb7 PM |
1923 | #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE) |
1924 | struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly; | |
1925 | EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook); | |
1926 | ||
1927 | static inline struct sk_buff *handle_macvlan(struct sk_buff *skb, | |
1928 | struct packet_type **pt_prev, | |
1929 | int *ret, | |
1930 | struct net_device *orig_dev) | |
1931 | { | |
1932 | if (skb->dev->macvlan_port == NULL) | |
1933 | return skb; | |
1934 | ||
1935 | if (*pt_prev) { | |
1936 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
1937 | *pt_prev = NULL; | |
1938 | } | |
1939 | return macvlan_handle_frame_hook(skb); | |
1940 | } | |
1941 | #else | |
1942 | #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb) | |
1943 | #endif | |
1944 | ||
1da177e4 LT |
1945 | #ifdef CONFIG_NET_CLS_ACT |
1946 | /* TODO: Maybe we should just force sch_ingress to be compiled in | |
1947 | * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions | |
1948 | * a compare and 2 stores extra right now if we dont have it on | |
1949 | * but have CONFIG_NET_CLS_ACT | |
4ec93edb | 1950 | * NOTE: This doesnt stop any functionality; if you dont have |
1da177e4 LT |
1951 | * the ingress scheduler, you just cant add policies on ingress. |
1952 | * | |
1953 | */ | |
4ec93edb | 1954 | static int ing_filter(struct sk_buff *skb) |
1da177e4 LT |
1955 | { |
1956 | struct Qdisc *q; | |
1957 | struct net_device *dev = skb->dev; | |
1958 | int result = TC_ACT_OK; | |
f697c3e8 | 1959 | u32 ttl = G_TC_RTTL(skb->tc_verd); |
4ec93edb | 1960 | |
f697c3e8 HX |
1961 | if (MAX_RED_LOOP < ttl++) { |
1962 | printk(KERN_WARNING | |
1963 | "Redir loop detected Dropping packet (%d->%d)\n", | |
1964 | skb->iif, dev->ifindex); | |
1965 | return TC_ACT_SHOT; | |
1966 | } | |
1da177e4 | 1967 | |
f697c3e8 HX |
1968 | skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl); |
1969 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); | |
1da177e4 | 1970 | |
f697c3e8 HX |
1971 | spin_lock(&dev->ingress_lock); |
1972 | if ((q = dev->qdisc_ingress) != NULL) | |
1973 | result = q->enqueue(skb, q); | |
1974 | spin_unlock(&dev->ingress_lock); | |
1975 | ||
1976 | return result; | |
1977 | } | |
86e65da9 | 1978 | |
f697c3e8 HX |
1979 | static inline struct sk_buff *handle_ing(struct sk_buff *skb, |
1980 | struct packet_type **pt_prev, | |
1981 | int *ret, struct net_device *orig_dev) | |
1982 | { | |
1983 | if (!skb->dev->qdisc_ingress) | |
1984 | goto out; | |
1da177e4 | 1985 | |
f697c3e8 HX |
1986 | if (*pt_prev) { |
1987 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
1988 | *pt_prev = NULL; | |
1989 | } else { | |
1990 | /* Huh? Why does turning on AF_PACKET affect this? */ | |
1991 | skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd); | |
1da177e4 LT |
1992 | } |
1993 | ||
f697c3e8 HX |
1994 | switch (ing_filter(skb)) { |
1995 | case TC_ACT_SHOT: | |
1996 | case TC_ACT_STOLEN: | |
1997 | kfree_skb(skb); | |
1998 | return NULL; | |
1999 | } | |
2000 | ||
2001 | out: | |
2002 | skb->tc_verd = 0; | |
2003 | return skb; | |
1da177e4 LT |
2004 | } |
2005 | #endif | |
2006 | ||
3b582cc1 SH |
2007 | /** |
2008 | * netif_receive_skb - process receive buffer from network | |
2009 | * @skb: buffer to process | |
2010 | * | |
2011 | * netif_receive_skb() is the main receive data processing function. | |
2012 | * It always succeeds. The buffer may be dropped during processing | |
2013 | * for congestion control or by the protocol layers. | |
2014 | * | |
2015 | * This function may only be called from softirq context and interrupts | |
2016 | * should be enabled. | |
2017 | * | |
2018 | * Return values (usually ignored): | |
2019 | * NET_RX_SUCCESS: no congestion | |
2020 | * NET_RX_DROP: packet was dropped | |
2021 | */ | |
1da177e4 LT |
2022 | int netif_receive_skb(struct sk_buff *skb) |
2023 | { | |
2024 | struct packet_type *ptype, *pt_prev; | |
f2ccd8fa | 2025 | struct net_device *orig_dev; |
1da177e4 | 2026 | int ret = NET_RX_DROP; |
252e3346 | 2027 | __be16 type; |
1da177e4 LT |
2028 | |
2029 | /* if we've gotten here through NAPI, check netpoll */ | |
bea3348e | 2030 | if (netpoll_receive_skb(skb)) |
1da177e4 LT |
2031 | return NET_RX_DROP; |
2032 | ||
b7aa0bf7 | 2033 | if (!skb->tstamp.tv64) |
a61bbcf2 | 2034 | net_timestamp(skb); |
1da177e4 | 2035 | |
c01003c2 PM |
2036 | if (!skb->iif) |
2037 | skb->iif = skb->dev->ifindex; | |
86e65da9 | 2038 | |
f2ccd8fa | 2039 | orig_dev = skb_bond(skb); |
1da177e4 | 2040 | |
8f903c70 JV |
2041 | if (!orig_dev) |
2042 | return NET_RX_DROP; | |
2043 | ||
1da177e4 LT |
2044 | __get_cpu_var(netdev_rx_stat).total++; |
2045 | ||
c1d2bbe1 | 2046 | skb_reset_network_header(skb); |
badff6d0 | 2047 | skb_reset_transport_header(skb); |
b0e380b1 | 2048 | skb->mac_len = skb->network_header - skb->mac_header; |
1da177e4 LT |
2049 | |
2050 | pt_prev = NULL; | |
2051 | ||
2052 | rcu_read_lock(); | |
2053 | ||
2054 | #ifdef CONFIG_NET_CLS_ACT | |
2055 | if (skb->tc_verd & TC_NCLS) { | |
2056 | skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); | |
2057 | goto ncls; | |
2058 | } | |
2059 | #endif | |
2060 | ||
2061 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
2062 | if (!ptype->dev || ptype->dev == skb->dev) { | |
4ec93edb | 2063 | if (pt_prev) |
f2ccd8fa | 2064 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
2065 | pt_prev = ptype; |
2066 | } | |
2067 | } | |
2068 | ||
2069 | #ifdef CONFIG_NET_CLS_ACT | |
f697c3e8 HX |
2070 | skb = handle_ing(skb, &pt_prev, &ret, orig_dev); |
2071 | if (!skb) | |
1da177e4 | 2072 | goto out; |
1da177e4 LT |
2073 | ncls: |
2074 | #endif | |
2075 | ||
6229e362 | 2076 | skb = handle_bridge(skb, &pt_prev, &ret, orig_dev); |
b863ceb7 PM |
2077 | if (!skb) |
2078 | goto out; | |
2079 | skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev); | |
6229e362 | 2080 | if (!skb) |
1da177e4 LT |
2081 | goto out; |
2082 | ||
2083 | type = skb->protocol; | |
82d8a867 PE |
2084 | list_for_each_entry_rcu(ptype, |
2085 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
1da177e4 LT |
2086 | if (ptype->type == type && |
2087 | (!ptype->dev || ptype->dev == skb->dev)) { | |
4ec93edb | 2088 | if (pt_prev) |
f2ccd8fa | 2089 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
2090 | pt_prev = ptype; |
2091 | } | |
2092 | } | |
2093 | ||
2094 | if (pt_prev) { | |
f2ccd8fa | 2095 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
1da177e4 LT |
2096 | } else { |
2097 | kfree_skb(skb); | |
2098 | /* Jamal, now you will not able to escape explaining | |
2099 | * me how you were going to use this. :-) | |
2100 | */ | |
2101 | ret = NET_RX_DROP; | |
2102 | } | |
2103 | ||
2104 | out: | |
2105 | rcu_read_unlock(); | |
2106 | return ret; | |
2107 | } | |
2108 | ||
bea3348e | 2109 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 LT |
2110 | { |
2111 | int work = 0; | |
1da177e4 LT |
2112 | struct softnet_data *queue = &__get_cpu_var(softnet_data); |
2113 | unsigned long start_time = jiffies; | |
2114 | ||
bea3348e SH |
2115 | napi->weight = weight_p; |
2116 | do { | |
1da177e4 LT |
2117 | struct sk_buff *skb; |
2118 | struct net_device *dev; | |
2119 | ||
2120 | local_irq_disable(); | |
2121 | skb = __skb_dequeue(&queue->input_pkt_queue); | |
bea3348e SH |
2122 | if (!skb) { |
2123 | __napi_complete(napi); | |
2124 | local_irq_enable(); | |
2125 | break; | |
2126 | } | |
2127 | ||
1da177e4 LT |
2128 | local_irq_enable(); |
2129 | ||
2130 | dev = skb->dev; | |
2131 | ||
2132 | netif_receive_skb(skb); | |
2133 | ||
2134 | dev_put(dev); | |
bea3348e | 2135 | } while (++work < quota && jiffies == start_time); |
1da177e4 | 2136 | |
bea3348e SH |
2137 | return work; |
2138 | } | |
1da177e4 | 2139 | |
bea3348e SH |
2140 | /** |
2141 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 2142 | * @n: entry to schedule |
bea3348e SH |
2143 | * |
2144 | * The entry's receive function will be scheduled to run | |
2145 | */ | |
2146 | void fastcall __napi_schedule(struct napi_struct *n) | |
2147 | { | |
2148 | unsigned long flags; | |
1da177e4 | 2149 | |
bea3348e SH |
2150 | local_irq_save(flags); |
2151 | list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list); | |
2152 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2153 | local_irq_restore(flags); | |
1da177e4 | 2154 | } |
bea3348e SH |
2155 | EXPORT_SYMBOL(__napi_schedule); |
2156 | ||
1da177e4 LT |
2157 | |
2158 | static void net_rx_action(struct softirq_action *h) | |
2159 | { | |
bea3348e | 2160 | struct list_head *list = &__get_cpu_var(softnet_data).poll_list; |
1da177e4 | 2161 | unsigned long start_time = jiffies; |
51b0bded | 2162 | int budget = netdev_budget; |
53fb95d3 MM |
2163 | void *have; |
2164 | ||
1da177e4 LT |
2165 | local_irq_disable(); |
2166 | ||
bea3348e SH |
2167 | while (!list_empty(list)) { |
2168 | struct napi_struct *n; | |
2169 | int work, weight; | |
1da177e4 | 2170 | |
bea3348e SH |
2171 | /* If softirq window is exhuasted then punt. |
2172 | * | |
2173 | * Note that this is a slight policy change from the | |
2174 | * previous NAPI code, which would allow up to 2 | |
2175 | * jiffies to pass before breaking out. The test | |
2176 | * used to be "jiffies - start_time > 1". | |
2177 | */ | |
2178 | if (unlikely(budget <= 0 || jiffies != start_time)) | |
1da177e4 LT |
2179 | goto softnet_break; |
2180 | ||
2181 | local_irq_enable(); | |
2182 | ||
bea3348e SH |
2183 | /* Even though interrupts have been re-enabled, this |
2184 | * access is safe because interrupts can only add new | |
2185 | * entries to the tail of this list, and only ->poll() | |
2186 | * calls can remove this head entry from the list. | |
2187 | */ | |
2188 | n = list_entry(list->next, struct napi_struct, poll_list); | |
1da177e4 | 2189 | |
bea3348e SH |
2190 | have = netpoll_poll_lock(n); |
2191 | ||
2192 | weight = n->weight; | |
2193 | ||
0a7606c1 DM |
2194 | /* This NAPI_STATE_SCHED test is for avoiding a race |
2195 | * with netpoll's poll_napi(). Only the entity which | |
2196 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
2197 | * actually make the ->poll() call. Therefore we avoid | |
2198 | * accidently calling ->poll() when NAPI is not scheduled. | |
2199 | */ | |
2200 | work = 0; | |
2201 | if (test_bit(NAPI_STATE_SCHED, &n->state)) | |
2202 | work = n->poll(n, weight); | |
bea3348e SH |
2203 | |
2204 | WARN_ON_ONCE(work > weight); | |
2205 | ||
2206 | budget -= work; | |
2207 | ||
2208 | local_irq_disable(); | |
2209 | ||
2210 | /* Drivers must not modify the NAPI state if they | |
2211 | * consume the entire weight. In such cases this code | |
2212 | * still "owns" the NAPI instance and therefore can | |
2213 | * move the instance around on the list at-will. | |
2214 | */ | |
fed17f30 DM |
2215 | if (unlikely(work == weight)) { |
2216 | if (unlikely(napi_disable_pending(n))) | |
2217 | __napi_complete(n); | |
2218 | else | |
2219 | list_move_tail(&n->poll_list, list); | |
2220 | } | |
bea3348e SH |
2221 | |
2222 | netpoll_poll_unlock(have); | |
1da177e4 LT |
2223 | } |
2224 | out: | |
515e06c4 | 2225 | local_irq_enable(); |
bea3348e | 2226 | |
db217334 CL |
2227 | #ifdef CONFIG_NET_DMA |
2228 | /* | |
2229 | * There may not be any more sk_buffs coming right now, so push | |
2230 | * any pending DMA copies to hardware | |
2231 | */ | |
d379b01e DW |
2232 | if (!cpus_empty(net_dma.channel_mask)) { |
2233 | int chan_idx; | |
2234 | for_each_cpu_mask(chan_idx, net_dma.channel_mask) { | |
2235 | struct dma_chan *chan = net_dma.channels[chan_idx]; | |
2236 | if (chan) | |
2237 | dma_async_memcpy_issue_pending(chan); | |
2238 | } | |
db217334 CL |
2239 | } |
2240 | #endif | |
bea3348e | 2241 | |
1da177e4 LT |
2242 | return; |
2243 | ||
2244 | softnet_break: | |
2245 | __get_cpu_var(netdev_rx_stat).time_squeeze++; | |
2246 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2247 | goto out; | |
2248 | } | |
2249 | ||
2250 | static gifconf_func_t * gifconf_list [NPROTO]; | |
2251 | ||
2252 | /** | |
2253 | * register_gifconf - register a SIOCGIF handler | |
2254 | * @family: Address family | |
2255 | * @gifconf: Function handler | |
2256 | * | |
2257 | * Register protocol dependent address dumping routines. The handler | |
2258 | * that is passed must not be freed or reused until it has been replaced | |
2259 | * by another handler. | |
2260 | */ | |
2261 | int register_gifconf(unsigned int family, gifconf_func_t * gifconf) | |
2262 | { | |
2263 | if (family >= NPROTO) | |
2264 | return -EINVAL; | |
2265 | gifconf_list[family] = gifconf; | |
2266 | return 0; | |
2267 | } | |
2268 | ||
2269 | ||
2270 | /* | |
2271 | * Map an interface index to its name (SIOCGIFNAME) | |
2272 | */ | |
2273 | ||
2274 | /* | |
2275 | * We need this ioctl for efficient implementation of the | |
2276 | * if_indextoname() function required by the IPv6 API. Without | |
2277 | * it, we would have to search all the interfaces to find a | |
2278 | * match. --pb | |
2279 | */ | |
2280 | ||
881d966b | 2281 | static int dev_ifname(struct net *net, struct ifreq __user *arg) |
1da177e4 LT |
2282 | { |
2283 | struct net_device *dev; | |
2284 | struct ifreq ifr; | |
2285 | ||
2286 | /* | |
2287 | * Fetch the caller's info block. | |
2288 | */ | |
2289 | ||
2290 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
2291 | return -EFAULT; | |
2292 | ||
2293 | read_lock(&dev_base_lock); | |
881d966b | 2294 | dev = __dev_get_by_index(net, ifr.ifr_ifindex); |
1da177e4 LT |
2295 | if (!dev) { |
2296 | read_unlock(&dev_base_lock); | |
2297 | return -ENODEV; | |
2298 | } | |
2299 | ||
2300 | strcpy(ifr.ifr_name, dev->name); | |
2301 | read_unlock(&dev_base_lock); | |
2302 | ||
2303 | if (copy_to_user(arg, &ifr, sizeof(struct ifreq))) | |
2304 | return -EFAULT; | |
2305 | return 0; | |
2306 | } | |
2307 | ||
2308 | /* | |
2309 | * Perform a SIOCGIFCONF call. This structure will change | |
2310 | * size eventually, and there is nothing I can do about it. | |
2311 | * Thus we will need a 'compatibility mode'. | |
2312 | */ | |
2313 | ||
881d966b | 2314 | static int dev_ifconf(struct net *net, char __user *arg) |
1da177e4 LT |
2315 | { |
2316 | struct ifconf ifc; | |
2317 | struct net_device *dev; | |
2318 | char __user *pos; | |
2319 | int len; | |
2320 | int total; | |
2321 | int i; | |
2322 | ||
2323 | /* | |
2324 | * Fetch the caller's info block. | |
2325 | */ | |
2326 | ||
2327 | if (copy_from_user(&ifc, arg, sizeof(struct ifconf))) | |
2328 | return -EFAULT; | |
2329 | ||
2330 | pos = ifc.ifc_buf; | |
2331 | len = ifc.ifc_len; | |
2332 | ||
2333 | /* | |
2334 | * Loop over the interfaces, and write an info block for each. | |
2335 | */ | |
2336 | ||
2337 | total = 0; | |
881d966b | 2338 | for_each_netdev(net, dev) { |
1da177e4 LT |
2339 | for (i = 0; i < NPROTO; i++) { |
2340 | if (gifconf_list[i]) { | |
2341 | int done; | |
2342 | if (!pos) | |
2343 | done = gifconf_list[i](dev, NULL, 0); | |
2344 | else | |
2345 | done = gifconf_list[i](dev, pos + total, | |
2346 | len - total); | |
2347 | if (done < 0) | |
2348 | return -EFAULT; | |
2349 | total += done; | |
2350 | } | |
2351 | } | |
4ec93edb | 2352 | } |
1da177e4 LT |
2353 | |
2354 | /* | |
2355 | * All done. Write the updated control block back to the caller. | |
2356 | */ | |
2357 | ifc.ifc_len = total; | |
2358 | ||
2359 | /* | |
2360 | * Both BSD and Solaris return 0 here, so we do too. | |
2361 | */ | |
2362 | return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0; | |
2363 | } | |
2364 | ||
2365 | #ifdef CONFIG_PROC_FS | |
2366 | /* | |
2367 | * This is invoked by the /proc filesystem handler to display a device | |
2368 | * in detail. | |
2369 | */ | |
7562f876 | 2370 | void *dev_seq_start(struct seq_file *seq, loff_t *pos) |
9a429c49 | 2371 | __acquires(dev_base_lock) |
1da177e4 | 2372 | { |
e372c414 | 2373 | struct net *net = seq_file_net(seq); |
7562f876 | 2374 | loff_t off; |
1da177e4 | 2375 | struct net_device *dev; |
1da177e4 | 2376 | |
7562f876 PE |
2377 | read_lock(&dev_base_lock); |
2378 | if (!*pos) | |
2379 | return SEQ_START_TOKEN; | |
1da177e4 | 2380 | |
7562f876 | 2381 | off = 1; |
881d966b | 2382 | for_each_netdev(net, dev) |
7562f876 PE |
2383 | if (off++ == *pos) |
2384 | return dev; | |
1da177e4 | 2385 | |
7562f876 | 2386 | return NULL; |
1da177e4 LT |
2387 | } |
2388 | ||
2389 | void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
2390 | { | |
e372c414 | 2391 | struct net *net = seq_file_net(seq); |
1da177e4 | 2392 | ++*pos; |
7562f876 | 2393 | return v == SEQ_START_TOKEN ? |
881d966b | 2394 | first_net_device(net) : next_net_device((struct net_device *)v); |
1da177e4 LT |
2395 | } |
2396 | ||
2397 | void dev_seq_stop(struct seq_file *seq, void *v) | |
9a429c49 | 2398 | __releases(dev_base_lock) |
1da177e4 LT |
2399 | { |
2400 | read_unlock(&dev_base_lock); | |
2401 | } | |
2402 | ||
2403 | static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev) | |
2404 | { | |
c45d286e | 2405 | struct net_device_stats *stats = dev->get_stats(dev); |
1da177e4 | 2406 | |
5a1b5898 RR |
2407 | seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu " |
2408 | "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n", | |
2409 | dev->name, stats->rx_bytes, stats->rx_packets, | |
2410 | stats->rx_errors, | |
2411 | stats->rx_dropped + stats->rx_missed_errors, | |
2412 | stats->rx_fifo_errors, | |
2413 | stats->rx_length_errors + stats->rx_over_errors + | |
2414 | stats->rx_crc_errors + stats->rx_frame_errors, | |
2415 | stats->rx_compressed, stats->multicast, | |
2416 | stats->tx_bytes, stats->tx_packets, | |
2417 | stats->tx_errors, stats->tx_dropped, | |
2418 | stats->tx_fifo_errors, stats->collisions, | |
2419 | stats->tx_carrier_errors + | |
2420 | stats->tx_aborted_errors + | |
2421 | stats->tx_window_errors + | |
2422 | stats->tx_heartbeat_errors, | |
2423 | stats->tx_compressed); | |
1da177e4 LT |
2424 | } |
2425 | ||
2426 | /* | |
2427 | * Called from the PROCfs module. This now uses the new arbitrary sized | |
2428 | * /proc/net interface to create /proc/net/dev | |
2429 | */ | |
2430 | static int dev_seq_show(struct seq_file *seq, void *v) | |
2431 | { | |
2432 | if (v == SEQ_START_TOKEN) | |
2433 | seq_puts(seq, "Inter-| Receive " | |
2434 | " | Transmit\n" | |
2435 | " face |bytes packets errs drop fifo frame " | |
2436 | "compressed multicast|bytes packets errs " | |
2437 | "drop fifo colls carrier compressed\n"); | |
2438 | else | |
2439 | dev_seq_printf_stats(seq, v); | |
2440 | return 0; | |
2441 | } | |
2442 | ||
2443 | static struct netif_rx_stats *softnet_get_online(loff_t *pos) | |
2444 | { | |
2445 | struct netif_rx_stats *rc = NULL; | |
2446 | ||
2447 | while (*pos < NR_CPUS) | |
4ec93edb | 2448 | if (cpu_online(*pos)) { |
1da177e4 LT |
2449 | rc = &per_cpu(netdev_rx_stat, *pos); |
2450 | break; | |
2451 | } else | |
2452 | ++*pos; | |
2453 | return rc; | |
2454 | } | |
2455 | ||
2456 | static void *softnet_seq_start(struct seq_file *seq, loff_t *pos) | |
2457 | { | |
2458 | return softnet_get_online(pos); | |
2459 | } | |
2460 | ||
2461 | static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
2462 | { | |
2463 | ++*pos; | |
2464 | return softnet_get_online(pos); | |
2465 | } | |
2466 | ||
2467 | static void softnet_seq_stop(struct seq_file *seq, void *v) | |
2468 | { | |
2469 | } | |
2470 | ||
2471 | static int softnet_seq_show(struct seq_file *seq, void *v) | |
2472 | { | |
2473 | struct netif_rx_stats *s = v; | |
2474 | ||
2475 | seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n", | |
31aa02c5 | 2476 | s->total, s->dropped, s->time_squeeze, 0, |
c1ebcdb8 SH |
2477 | 0, 0, 0, 0, /* was fastroute */ |
2478 | s->cpu_collision ); | |
1da177e4 LT |
2479 | return 0; |
2480 | } | |
2481 | ||
f690808e | 2482 | static const struct seq_operations dev_seq_ops = { |
1da177e4 LT |
2483 | .start = dev_seq_start, |
2484 | .next = dev_seq_next, | |
2485 | .stop = dev_seq_stop, | |
2486 | .show = dev_seq_show, | |
2487 | }; | |
2488 | ||
2489 | static int dev_seq_open(struct inode *inode, struct file *file) | |
2490 | { | |
e372c414 DL |
2491 | return seq_open_net(inode, file, &dev_seq_ops, |
2492 | sizeof(struct seq_net_private)); | |
1da177e4 LT |
2493 | } |
2494 | ||
9a32144e | 2495 | static const struct file_operations dev_seq_fops = { |
1da177e4 LT |
2496 | .owner = THIS_MODULE, |
2497 | .open = dev_seq_open, | |
2498 | .read = seq_read, | |
2499 | .llseek = seq_lseek, | |
e372c414 | 2500 | .release = seq_release_net, |
1da177e4 LT |
2501 | }; |
2502 | ||
f690808e | 2503 | static const struct seq_operations softnet_seq_ops = { |
1da177e4 LT |
2504 | .start = softnet_seq_start, |
2505 | .next = softnet_seq_next, | |
2506 | .stop = softnet_seq_stop, | |
2507 | .show = softnet_seq_show, | |
2508 | }; | |
2509 | ||
2510 | static int softnet_seq_open(struct inode *inode, struct file *file) | |
2511 | { | |
2512 | return seq_open(file, &softnet_seq_ops); | |
2513 | } | |
2514 | ||
9a32144e | 2515 | static const struct file_operations softnet_seq_fops = { |
1da177e4 LT |
2516 | .owner = THIS_MODULE, |
2517 | .open = softnet_seq_open, | |
2518 | .read = seq_read, | |
2519 | .llseek = seq_lseek, | |
2520 | .release = seq_release, | |
2521 | }; | |
2522 | ||
0e1256ff SH |
2523 | static void *ptype_get_idx(loff_t pos) |
2524 | { | |
2525 | struct packet_type *pt = NULL; | |
2526 | loff_t i = 0; | |
2527 | int t; | |
2528 | ||
2529 | list_for_each_entry_rcu(pt, &ptype_all, list) { | |
2530 | if (i == pos) | |
2531 | return pt; | |
2532 | ++i; | |
2533 | } | |
2534 | ||
82d8a867 | 2535 | for (t = 0; t < PTYPE_HASH_SIZE; t++) { |
0e1256ff SH |
2536 | list_for_each_entry_rcu(pt, &ptype_base[t], list) { |
2537 | if (i == pos) | |
2538 | return pt; | |
2539 | ++i; | |
2540 | } | |
2541 | } | |
2542 | return NULL; | |
2543 | } | |
2544 | ||
2545 | static void *ptype_seq_start(struct seq_file *seq, loff_t *pos) | |
72348a42 | 2546 | __acquires(RCU) |
0e1256ff SH |
2547 | { |
2548 | rcu_read_lock(); | |
2549 | return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN; | |
2550 | } | |
2551 | ||
2552 | static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
2553 | { | |
2554 | struct packet_type *pt; | |
2555 | struct list_head *nxt; | |
2556 | int hash; | |
2557 | ||
2558 | ++*pos; | |
2559 | if (v == SEQ_START_TOKEN) | |
2560 | return ptype_get_idx(0); | |
2561 | ||
2562 | pt = v; | |
2563 | nxt = pt->list.next; | |
2564 | if (pt->type == htons(ETH_P_ALL)) { | |
2565 | if (nxt != &ptype_all) | |
2566 | goto found; | |
2567 | hash = 0; | |
2568 | nxt = ptype_base[0].next; | |
2569 | } else | |
82d8a867 | 2570 | hash = ntohs(pt->type) & PTYPE_HASH_MASK; |
0e1256ff SH |
2571 | |
2572 | while (nxt == &ptype_base[hash]) { | |
82d8a867 | 2573 | if (++hash >= PTYPE_HASH_SIZE) |
0e1256ff SH |
2574 | return NULL; |
2575 | nxt = ptype_base[hash].next; | |
2576 | } | |
2577 | found: | |
2578 | return list_entry(nxt, struct packet_type, list); | |
2579 | } | |
2580 | ||
2581 | static void ptype_seq_stop(struct seq_file *seq, void *v) | |
72348a42 | 2582 | __releases(RCU) |
0e1256ff SH |
2583 | { |
2584 | rcu_read_unlock(); | |
2585 | } | |
2586 | ||
2587 | static void ptype_seq_decode(struct seq_file *seq, void *sym) | |
2588 | { | |
2589 | #ifdef CONFIG_KALLSYMS | |
2590 | unsigned long offset = 0, symsize; | |
2591 | const char *symname; | |
2592 | char *modname; | |
2593 | char namebuf[128]; | |
2594 | ||
2595 | symname = kallsyms_lookup((unsigned long)sym, &symsize, &offset, | |
2596 | &modname, namebuf); | |
2597 | ||
2598 | if (symname) { | |
2599 | char *delim = ":"; | |
2600 | ||
2601 | if (!modname) | |
2602 | modname = delim = ""; | |
2603 | seq_printf(seq, "%s%s%s%s+0x%lx", delim, modname, delim, | |
2604 | symname, offset); | |
2605 | return; | |
2606 | } | |
2607 | #endif | |
2608 | ||
2609 | seq_printf(seq, "[%p]", sym); | |
2610 | } | |
2611 | ||
2612 | static int ptype_seq_show(struct seq_file *seq, void *v) | |
2613 | { | |
2614 | struct packet_type *pt = v; | |
2615 | ||
2616 | if (v == SEQ_START_TOKEN) | |
2617 | seq_puts(seq, "Type Device Function\n"); | |
2618 | else { | |
2619 | if (pt->type == htons(ETH_P_ALL)) | |
2620 | seq_puts(seq, "ALL "); | |
2621 | else | |
2622 | seq_printf(seq, "%04x", ntohs(pt->type)); | |
2623 | ||
2624 | seq_printf(seq, " %-8s ", | |
2625 | pt->dev ? pt->dev->name : ""); | |
2626 | ptype_seq_decode(seq, pt->func); | |
2627 | seq_putc(seq, '\n'); | |
2628 | } | |
2629 | ||
2630 | return 0; | |
2631 | } | |
2632 | ||
2633 | static const struct seq_operations ptype_seq_ops = { | |
2634 | .start = ptype_seq_start, | |
2635 | .next = ptype_seq_next, | |
2636 | .stop = ptype_seq_stop, | |
2637 | .show = ptype_seq_show, | |
2638 | }; | |
2639 | ||
2640 | static int ptype_seq_open(struct inode *inode, struct file *file) | |
2641 | { | |
2642 | return seq_open(file, &ptype_seq_ops); | |
2643 | } | |
2644 | ||
2645 | static const struct file_operations ptype_seq_fops = { | |
2646 | .owner = THIS_MODULE, | |
2647 | .open = ptype_seq_open, | |
2648 | .read = seq_read, | |
2649 | .llseek = seq_lseek, | |
2650 | .release = seq_release, | |
2651 | }; | |
2652 | ||
2653 | ||
4665079c | 2654 | static int __net_init dev_proc_net_init(struct net *net) |
1da177e4 LT |
2655 | { |
2656 | int rc = -ENOMEM; | |
2657 | ||
881d966b | 2658 | if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops)) |
1da177e4 | 2659 | goto out; |
881d966b | 2660 | if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops)) |
1da177e4 | 2661 | goto out_dev; |
881d966b | 2662 | if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops)) |
457c4cbc | 2663 | goto out_softnet; |
0e1256ff | 2664 | |
881d966b | 2665 | if (wext_proc_init(net)) |
457c4cbc | 2666 | goto out_ptype; |
1da177e4 LT |
2667 | rc = 0; |
2668 | out: | |
2669 | return rc; | |
457c4cbc | 2670 | out_ptype: |
881d966b | 2671 | proc_net_remove(net, "ptype"); |
1da177e4 | 2672 | out_softnet: |
881d966b | 2673 | proc_net_remove(net, "softnet_stat"); |
1da177e4 | 2674 | out_dev: |
881d966b | 2675 | proc_net_remove(net, "dev"); |
1da177e4 LT |
2676 | goto out; |
2677 | } | |
881d966b | 2678 | |
4665079c | 2679 | static void __net_exit dev_proc_net_exit(struct net *net) |
881d966b EB |
2680 | { |
2681 | wext_proc_exit(net); | |
2682 | ||
2683 | proc_net_remove(net, "ptype"); | |
2684 | proc_net_remove(net, "softnet_stat"); | |
2685 | proc_net_remove(net, "dev"); | |
2686 | } | |
2687 | ||
022cbae6 | 2688 | static struct pernet_operations __net_initdata dev_proc_ops = { |
881d966b EB |
2689 | .init = dev_proc_net_init, |
2690 | .exit = dev_proc_net_exit, | |
2691 | }; | |
2692 | ||
2693 | static int __init dev_proc_init(void) | |
2694 | { | |
2695 | return register_pernet_subsys(&dev_proc_ops); | |
2696 | } | |
1da177e4 LT |
2697 | #else |
2698 | #define dev_proc_init() 0 | |
2699 | #endif /* CONFIG_PROC_FS */ | |
2700 | ||
2701 | ||
2702 | /** | |
2703 | * netdev_set_master - set up master/slave pair | |
2704 | * @slave: slave device | |
2705 | * @master: new master device | |
2706 | * | |
2707 | * Changes the master device of the slave. Pass %NULL to break the | |
2708 | * bonding. The caller must hold the RTNL semaphore. On a failure | |
2709 | * a negative errno code is returned. On success the reference counts | |
2710 | * are adjusted, %RTM_NEWLINK is sent to the routing socket and the | |
2711 | * function returns zero. | |
2712 | */ | |
2713 | int netdev_set_master(struct net_device *slave, struct net_device *master) | |
2714 | { | |
2715 | struct net_device *old = slave->master; | |
2716 | ||
2717 | ASSERT_RTNL(); | |
2718 | ||
2719 | if (master) { | |
2720 | if (old) | |
2721 | return -EBUSY; | |
2722 | dev_hold(master); | |
2723 | } | |
2724 | ||
2725 | slave->master = master; | |
4ec93edb | 2726 | |
1da177e4 LT |
2727 | synchronize_net(); |
2728 | ||
2729 | if (old) | |
2730 | dev_put(old); | |
2731 | ||
2732 | if (master) | |
2733 | slave->flags |= IFF_SLAVE; | |
2734 | else | |
2735 | slave->flags &= ~IFF_SLAVE; | |
2736 | ||
2737 | rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE); | |
2738 | return 0; | |
2739 | } | |
2740 | ||
4417da66 | 2741 | static void __dev_set_promiscuity(struct net_device *dev, int inc) |
1da177e4 LT |
2742 | { |
2743 | unsigned short old_flags = dev->flags; | |
2744 | ||
24023451 PM |
2745 | ASSERT_RTNL(); |
2746 | ||
1da177e4 LT |
2747 | if ((dev->promiscuity += inc) == 0) |
2748 | dev->flags &= ~IFF_PROMISC; | |
52609c0b DC |
2749 | else |
2750 | dev->flags |= IFF_PROMISC; | |
2751 | if (dev->flags != old_flags) { | |
1da177e4 LT |
2752 | printk(KERN_INFO "device %s %s promiscuous mode\n", |
2753 | dev->name, (dev->flags & IFF_PROMISC) ? "entered" : | |
4ec93edb | 2754 | "left"); |
7759db82 KHK |
2755 | if (audit_enabled) |
2756 | audit_log(current->audit_context, GFP_ATOMIC, | |
2757 | AUDIT_ANOM_PROMISCUOUS, | |
2758 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
2759 | dev->name, (dev->flags & IFF_PROMISC), | |
2760 | (old_flags & IFF_PROMISC), | |
2761 | audit_get_loginuid(current), | |
2762 | current->uid, current->gid, | |
2763 | audit_get_sessionid(current)); | |
24023451 PM |
2764 | |
2765 | if (dev->change_rx_flags) | |
2766 | dev->change_rx_flags(dev, IFF_PROMISC); | |
1da177e4 LT |
2767 | } |
2768 | } | |
2769 | ||
4417da66 PM |
2770 | /** |
2771 | * dev_set_promiscuity - update promiscuity count on a device | |
2772 | * @dev: device | |
2773 | * @inc: modifier | |
2774 | * | |
2775 | * Add or remove promiscuity from a device. While the count in the device | |
2776 | * remains above zero the interface remains promiscuous. Once it hits zero | |
2777 | * the device reverts back to normal filtering operation. A negative inc | |
2778 | * value is used to drop promiscuity on the device. | |
2779 | */ | |
2780 | void dev_set_promiscuity(struct net_device *dev, int inc) | |
2781 | { | |
2782 | unsigned short old_flags = dev->flags; | |
2783 | ||
2784 | __dev_set_promiscuity(dev, inc); | |
2785 | if (dev->flags != old_flags) | |
2786 | dev_set_rx_mode(dev); | |
2787 | } | |
2788 | ||
1da177e4 LT |
2789 | /** |
2790 | * dev_set_allmulti - update allmulti count on a device | |
2791 | * @dev: device | |
2792 | * @inc: modifier | |
2793 | * | |
2794 | * Add or remove reception of all multicast frames to a device. While the | |
2795 | * count in the device remains above zero the interface remains listening | |
2796 | * to all interfaces. Once it hits zero the device reverts back to normal | |
2797 | * filtering operation. A negative @inc value is used to drop the counter | |
2798 | * when releasing a resource needing all multicasts. | |
2799 | */ | |
2800 | ||
2801 | void dev_set_allmulti(struct net_device *dev, int inc) | |
2802 | { | |
2803 | unsigned short old_flags = dev->flags; | |
2804 | ||
24023451 PM |
2805 | ASSERT_RTNL(); |
2806 | ||
1da177e4 LT |
2807 | dev->flags |= IFF_ALLMULTI; |
2808 | if ((dev->allmulti += inc) == 0) | |
2809 | dev->flags &= ~IFF_ALLMULTI; | |
24023451 PM |
2810 | if (dev->flags ^ old_flags) { |
2811 | if (dev->change_rx_flags) | |
2812 | dev->change_rx_flags(dev, IFF_ALLMULTI); | |
4417da66 | 2813 | dev_set_rx_mode(dev); |
24023451 | 2814 | } |
4417da66 PM |
2815 | } |
2816 | ||
2817 | /* | |
2818 | * Upload unicast and multicast address lists to device and | |
2819 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 2820 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
2821 | * are present. |
2822 | */ | |
2823 | void __dev_set_rx_mode(struct net_device *dev) | |
2824 | { | |
2825 | /* dev_open will call this function so the list will stay sane. */ | |
2826 | if (!(dev->flags&IFF_UP)) | |
2827 | return; | |
2828 | ||
2829 | if (!netif_device_present(dev)) | |
40b77c94 | 2830 | return; |
4417da66 PM |
2831 | |
2832 | if (dev->set_rx_mode) | |
2833 | dev->set_rx_mode(dev); | |
2834 | else { | |
2835 | /* Unicast addresses changes may only happen under the rtnl, | |
2836 | * therefore calling __dev_set_promiscuity here is safe. | |
2837 | */ | |
2838 | if (dev->uc_count > 0 && !dev->uc_promisc) { | |
2839 | __dev_set_promiscuity(dev, 1); | |
2840 | dev->uc_promisc = 1; | |
2841 | } else if (dev->uc_count == 0 && dev->uc_promisc) { | |
2842 | __dev_set_promiscuity(dev, -1); | |
2843 | dev->uc_promisc = 0; | |
2844 | } | |
2845 | ||
2846 | if (dev->set_multicast_list) | |
2847 | dev->set_multicast_list(dev); | |
2848 | } | |
2849 | } | |
2850 | ||
2851 | void dev_set_rx_mode(struct net_device *dev) | |
2852 | { | |
2853 | netif_tx_lock_bh(dev); | |
2854 | __dev_set_rx_mode(dev); | |
2855 | netif_tx_unlock_bh(dev); | |
1da177e4 LT |
2856 | } |
2857 | ||
61cbc2fc PM |
2858 | int __dev_addr_delete(struct dev_addr_list **list, int *count, |
2859 | void *addr, int alen, int glbl) | |
bf742482 PM |
2860 | { |
2861 | struct dev_addr_list *da; | |
2862 | ||
2863 | for (; (da = *list) != NULL; list = &da->next) { | |
2864 | if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 && | |
2865 | alen == da->da_addrlen) { | |
2866 | if (glbl) { | |
2867 | int old_glbl = da->da_gusers; | |
2868 | da->da_gusers = 0; | |
2869 | if (old_glbl == 0) | |
2870 | break; | |
2871 | } | |
2872 | if (--da->da_users) | |
2873 | return 0; | |
2874 | ||
2875 | *list = da->next; | |
2876 | kfree(da); | |
61cbc2fc | 2877 | (*count)--; |
bf742482 PM |
2878 | return 0; |
2879 | } | |
2880 | } | |
2881 | return -ENOENT; | |
2882 | } | |
2883 | ||
61cbc2fc PM |
2884 | int __dev_addr_add(struct dev_addr_list **list, int *count, |
2885 | void *addr, int alen, int glbl) | |
bf742482 PM |
2886 | { |
2887 | struct dev_addr_list *da; | |
2888 | ||
2889 | for (da = *list; da != NULL; da = da->next) { | |
2890 | if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 && | |
2891 | da->da_addrlen == alen) { | |
2892 | if (glbl) { | |
2893 | int old_glbl = da->da_gusers; | |
2894 | da->da_gusers = 1; | |
2895 | if (old_glbl) | |
2896 | return 0; | |
2897 | } | |
2898 | da->da_users++; | |
2899 | return 0; | |
2900 | } | |
2901 | } | |
2902 | ||
2903 | da = kmalloc(sizeof(*da), GFP_ATOMIC); | |
2904 | if (da == NULL) | |
2905 | return -ENOMEM; | |
2906 | memcpy(da->da_addr, addr, alen); | |
2907 | da->da_addrlen = alen; | |
2908 | da->da_users = 1; | |
2909 | da->da_gusers = glbl ? 1 : 0; | |
2910 | da->next = *list; | |
2911 | *list = da; | |
61cbc2fc | 2912 | (*count)++; |
bf742482 PM |
2913 | return 0; |
2914 | } | |
2915 | ||
4417da66 PM |
2916 | /** |
2917 | * dev_unicast_delete - Release secondary unicast address. | |
2918 | * @dev: device | |
0ed72ec4 RD |
2919 | * @addr: address to delete |
2920 | * @alen: length of @addr | |
4417da66 PM |
2921 | * |
2922 | * Release reference to a secondary unicast address and remove it | |
0ed72ec4 | 2923 | * from the device if the reference count drops to zero. |
4417da66 PM |
2924 | * |
2925 | * The caller must hold the rtnl_mutex. | |
2926 | */ | |
2927 | int dev_unicast_delete(struct net_device *dev, void *addr, int alen) | |
2928 | { | |
2929 | int err; | |
2930 | ||
2931 | ASSERT_RTNL(); | |
2932 | ||
2933 | netif_tx_lock_bh(dev); | |
61cbc2fc PM |
2934 | err = __dev_addr_delete(&dev->uc_list, &dev->uc_count, addr, alen, 0); |
2935 | if (!err) | |
4417da66 | 2936 | __dev_set_rx_mode(dev); |
4417da66 PM |
2937 | netif_tx_unlock_bh(dev); |
2938 | return err; | |
2939 | } | |
2940 | EXPORT_SYMBOL(dev_unicast_delete); | |
2941 | ||
2942 | /** | |
2943 | * dev_unicast_add - add a secondary unicast address | |
2944 | * @dev: device | |
0ed72ec4 RD |
2945 | * @addr: address to delete |
2946 | * @alen: length of @addr | |
4417da66 PM |
2947 | * |
2948 | * Add a secondary unicast address to the device or increase | |
2949 | * the reference count if it already exists. | |
2950 | * | |
2951 | * The caller must hold the rtnl_mutex. | |
2952 | */ | |
2953 | int dev_unicast_add(struct net_device *dev, void *addr, int alen) | |
2954 | { | |
2955 | int err; | |
2956 | ||
2957 | ASSERT_RTNL(); | |
2958 | ||
2959 | netif_tx_lock_bh(dev); | |
61cbc2fc PM |
2960 | err = __dev_addr_add(&dev->uc_list, &dev->uc_count, addr, alen, 0); |
2961 | if (!err) | |
4417da66 | 2962 | __dev_set_rx_mode(dev); |
4417da66 PM |
2963 | netif_tx_unlock_bh(dev); |
2964 | return err; | |
2965 | } | |
2966 | EXPORT_SYMBOL(dev_unicast_add); | |
2967 | ||
e83a2ea8 CL |
2968 | int __dev_addr_sync(struct dev_addr_list **to, int *to_count, |
2969 | struct dev_addr_list **from, int *from_count) | |
2970 | { | |
2971 | struct dev_addr_list *da, *next; | |
2972 | int err = 0; | |
2973 | ||
2974 | da = *from; | |
2975 | while (da != NULL) { | |
2976 | next = da->next; | |
2977 | if (!da->da_synced) { | |
2978 | err = __dev_addr_add(to, to_count, | |
2979 | da->da_addr, da->da_addrlen, 0); | |
2980 | if (err < 0) | |
2981 | break; | |
2982 | da->da_synced = 1; | |
2983 | da->da_users++; | |
2984 | } else if (da->da_users == 1) { | |
2985 | __dev_addr_delete(to, to_count, | |
2986 | da->da_addr, da->da_addrlen, 0); | |
2987 | __dev_addr_delete(from, from_count, | |
2988 | da->da_addr, da->da_addrlen, 0); | |
2989 | } | |
2990 | da = next; | |
2991 | } | |
2992 | return err; | |
2993 | } | |
2994 | ||
2995 | void __dev_addr_unsync(struct dev_addr_list **to, int *to_count, | |
2996 | struct dev_addr_list **from, int *from_count) | |
2997 | { | |
2998 | struct dev_addr_list *da, *next; | |
2999 | ||
3000 | da = *from; | |
3001 | while (da != NULL) { | |
3002 | next = da->next; | |
3003 | if (da->da_synced) { | |
3004 | __dev_addr_delete(to, to_count, | |
3005 | da->da_addr, da->da_addrlen, 0); | |
3006 | da->da_synced = 0; | |
3007 | __dev_addr_delete(from, from_count, | |
3008 | da->da_addr, da->da_addrlen, 0); | |
3009 | } | |
3010 | da = next; | |
3011 | } | |
3012 | } | |
3013 | ||
3014 | /** | |
3015 | * dev_unicast_sync - Synchronize device's unicast list to another device | |
3016 | * @to: destination device | |
3017 | * @from: source device | |
3018 | * | |
3019 | * Add newly added addresses to the destination device and release | |
3020 | * addresses that have no users left. The source device must be | |
3021 | * locked by netif_tx_lock_bh. | |
3022 | * | |
3023 | * This function is intended to be called from the dev->set_rx_mode | |
3024 | * function of layered software devices. | |
3025 | */ | |
3026 | int dev_unicast_sync(struct net_device *to, struct net_device *from) | |
3027 | { | |
3028 | int err = 0; | |
3029 | ||
3030 | netif_tx_lock_bh(to); | |
3031 | err = __dev_addr_sync(&to->uc_list, &to->uc_count, | |
3032 | &from->uc_list, &from->uc_count); | |
3033 | if (!err) | |
3034 | __dev_set_rx_mode(to); | |
3035 | netif_tx_unlock_bh(to); | |
3036 | return err; | |
3037 | } | |
3038 | EXPORT_SYMBOL(dev_unicast_sync); | |
3039 | ||
3040 | /** | |
3041 | * dev_unicast_unsync - Remove synchronized addresses from the destination | |
3042 | * device | |
3043 | * @to: destination device | |
3044 | * @from: source device | |
3045 | * | |
3046 | * Remove all addresses that were added to the destination device by | |
3047 | * dev_unicast_sync(). This function is intended to be called from the | |
3048 | * dev->stop function of layered software devices. | |
3049 | */ | |
3050 | void dev_unicast_unsync(struct net_device *to, struct net_device *from) | |
3051 | { | |
3052 | netif_tx_lock_bh(from); | |
3053 | netif_tx_lock_bh(to); | |
3054 | ||
3055 | __dev_addr_unsync(&to->uc_list, &to->uc_count, | |
3056 | &from->uc_list, &from->uc_count); | |
3057 | __dev_set_rx_mode(to); | |
3058 | ||
3059 | netif_tx_unlock_bh(to); | |
3060 | netif_tx_unlock_bh(from); | |
3061 | } | |
3062 | EXPORT_SYMBOL(dev_unicast_unsync); | |
3063 | ||
12972621 DC |
3064 | static void __dev_addr_discard(struct dev_addr_list **list) |
3065 | { | |
3066 | struct dev_addr_list *tmp; | |
3067 | ||
3068 | while (*list != NULL) { | |
3069 | tmp = *list; | |
3070 | *list = tmp->next; | |
3071 | if (tmp->da_users > tmp->da_gusers) | |
3072 | printk("__dev_addr_discard: address leakage! " | |
3073 | "da_users=%d\n", tmp->da_users); | |
3074 | kfree(tmp); | |
3075 | } | |
3076 | } | |
3077 | ||
26cc2522 | 3078 | static void dev_addr_discard(struct net_device *dev) |
4417da66 PM |
3079 | { |
3080 | netif_tx_lock_bh(dev); | |
26cc2522 | 3081 | |
4417da66 PM |
3082 | __dev_addr_discard(&dev->uc_list); |
3083 | dev->uc_count = 0; | |
4417da66 | 3084 | |
456ad75c DC |
3085 | __dev_addr_discard(&dev->mc_list); |
3086 | dev->mc_count = 0; | |
26cc2522 | 3087 | |
456ad75c DC |
3088 | netif_tx_unlock_bh(dev); |
3089 | } | |
3090 | ||
1da177e4 LT |
3091 | unsigned dev_get_flags(const struct net_device *dev) |
3092 | { | |
3093 | unsigned flags; | |
3094 | ||
3095 | flags = (dev->flags & ~(IFF_PROMISC | | |
3096 | IFF_ALLMULTI | | |
b00055aa SR |
3097 | IFF_RUNNING | |
3098 | IFF_LOWER_UP | | |
3099 | IFF_DORMANT)) | | |
1da177e4 LT |
3100 | (dev->gflags & (IFF_PROMISC | |
3101 | IFF_ALLMULTI)); | |
3102 | ||
b00055aa SR |
3103 | if (netif_running(dev)) { |
3104 | if (netif_oper_up(dev)) | |
3105 | flags |= IFF_RUNNING; | |
3106 | if (netif_carrier_ok(dev)) | |
3107 | flags |= IFF_LOWER_UP; | |
3108 | if (netif_dormant(dev)) | |
3109 | flags |= IFF_DORMANT; | |
3110 | } | |
1da177e4 LT |
3111 | |
3112 | return flags; | |
3113 | } | |
3114 | ||
3115 | int dev_change_flags(struct net_device *dev, unsigned flags) | |
3116 | { | |
7c355f53 | 3117 | int ret, changes; |
1da177e4 LT |
3118 | int old_flags = dev->flags; |
3119 | ||
24023451 PM |
3120 | ASSERT_RTNL(); |
3121 | ||
1da177e4 LT |
3122 | /* |
3123 | * Set the flags on our device. | |
3124 | */ | |
3125 | ||
3126 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
3127 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
3128 | IFF_AUTOMEDIA)) | | |
3129 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
3130 | IFF_ALLMULTI)); | |
3131 | ||
3132 | /* | |
3133 | * Load in the correct multicast list now the flags have changed. | |
3134 | */ | |
3135 | ||
24023451 PM |
3136 | if (dev->change_rx_flags && (dev->flags ^ flags) & IFF_MULTICAST) |
3137 | dev->change_rx_flags(dev, IFF_MULTICAST); | |
3138 | ||
4417da66 | 3139 | dev_set_rx_mode(dev); |
1da177e4 LT |
3140 | |
3141 | /* | |
3142 | * Have we downed the interface. We handle IFF_UP ourselves | |
3143 | * according to user attempts to set it, rather than blindly | |
3144 | * setting it. | |
3145 | */ | |
3146 | ||
3147 | ret = 0; | |
3148 | if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */ | |
3149 | ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev); | |
3150 | ||
3151 | if (!ret) | |
4417da66 | 3152 | dev_set_rx_mode(dev); |
1da177e4 LT |
3153 | } |
3154 | ||
3155 | if (dev->flags & IFF_UP && | |
3156 | ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI | | |
3157 | IFF_VOLATILE))) | |
056925ab | 3158 | call_netdevice_notifiers(NETDEV_CHANGE, dev); |
1da177e4 LT |
3159 | |
3160 | if ((flags ^ dev->gflags) & IFF_PROMISC) { | |
3161 | int inc = (flags & IFF_PROMISC) ? +1 : -1; | |
3162 | dev->gflags ^= IFF_PROMISC; | |
3163 | dev_set_promiscuity(dev, inc); | |
3164 | } | |
3165 | ||
3166 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
3167 | is important. Some (broken) drivers set IFF_PROMISC, when | |
3168 | IFF_ALLMULTI is requested not asking us and not reporting. | |
3169 | */ | |
3170 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
3171 | int inc = (flags & IFF_ALLMULTI) ? +1 : -1; | |
3172 | dev->gflags ^= IFF_ALLMULTI; | |
3173 | dev_set_allmulti(dev, inc); | |
3174 | } | |
3175 | ||
7c355f53 TG |
3176 | /* Exclude state transition flags, already notified */ |
3177 | changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING); | |
3178 | if (changes) | |
3179 | rtmsg_ifinfo(RTM_NEWLINK, dev, changes); | |
1da177e4 LT |
3180 | |
3181 | return ret; | |
3182 | } | |
3183 | ||
3184 | int dev_set_mtu(struct net_device *dev, int new_mtu) | |
3185 | { | |
3186 | int err; | |
3187 | ||
3188 | if (new_mtu == dev->mtu) | |
3189 | return 0; | |
3190 | ||
3191 | /* MTU must be positive. */ | |
3192 | if (new_mtu < 0) | |
3193 | return -EINVAL; | |
3194 | ||
3195 | if (!netif_device_present(dev)) | |
3196 | return -ENODEV; | |
3197 | ||
3198 | err = 0; | |
3199 | if (dev->change_mtu) | |
3200 | err = dev->change_mtu(dev, new_mtu); | |
3201 | else | |
3202 | dev->mtu = new_mtu; | |
3203 | if (!err && dev->flags & IFF_UP) | |
056925ab | 3204 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); |
1da177e4 LT |
3205 | return err; |
3206 | } | |
3207 | ||
3208 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) | |
3209 | { | |
3210 | int err; | |
3211 | ||
3212 | if (!dev->set_mac_address) | |
3213 | return -EOPNOTSUPP; | |
3214 | if (sa->sa_family != dev->type) | |
3215 | return -EINVAL; | |
3216 | if (!netif_device_present(dev)) | |
3217 | return -ENODEV; | |
3218 | err = dev->set_mac_address(dev, sa); | |
3219 | if (!err) | |
056925ab | 3220 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
1da177e4 LT |
3221 | return err; |
3222 | } | |
3223 | ||
3224 | /* | |
14e3e079 | 3225 | * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock) |
1da177e4 | 3226 | */ |
14e3e079 | 3227 | static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd) |
1da177e4 LT |
3228 | { |
3229 | int err; | |
881d966b | 3230 | struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name); |
1da177e4 LT |
3231 | |
3232 | if (!dev) | |
3233 | return -ENODEV; | |
3234 | ||
3235 | switch (cmd) { | |
3236 | case SIOCGIFFLAGS: /* Get interface flags */ | |
3237 | ifr->ifr_flags = dev_get_flags(dev); | |
3238 | return 0; | |
3239 | ||
1da177e4 LT |
3240 | case SIOCGIFMETRIC: /* Get the metric on the interface |
3241 | (currently unused) */ | |
3242 | ifr->ifr_metric = 0; | |
3243 | return 0; | |
3244 | ||
1da177e4 LT |
3245 | case SIOCGIFMTU: /* Get the MTU of a device */ |
3246 | ifr->ifr_mtu = dev->mtu; | |
3247 | return 0; | |
3248 | ||
1da177e4 LT |
3249 | case SIOCGIFHWADDR: |
3250 | if (!dev->addr_len) | |
3251 | memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data); | |
3252 | else | |
3253 | memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr, | |
3254 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
3255 | ifr->ifr_hwaddr.sa_family = dev->type; | |
3256 | return 0; | |
3257 | ||
14e3e079 JG |
3258 | case SIOCGIFSLAVE: |
3259 | err = -EINVAL; | |
3260 | break; | |
3261 | ||
3262 | case SIOCGIFMAP: | |
3263 | ifr->ifr_map.mem_start = dev->mem_start; | |
3264 | ifr->ifr_map.mem_end = dev->mem_end; | |
3265 | ifr->ifr_map.base_addr = dev->base_addr; | |
3266 | ifr->ifr_map.irq = dev->irq; | |
3267 | ifr->ifr_map.dma = dev->dma; | |
3268 | ifr->ifr_map.port = dev->if_port; | |
3269 | return 0; | |
3270 | ||
3271 | case SIOCGIFINDEX: | |
3272 | ifr->ifr_ifindex = dev->ifindex; | |
3273 | return 0; | |
3274 | ||
3275 | case SIOCGIFTXQLEN: | |
3276 | ifr->ifr_qlen = dev->tx_queue_len; | |
3277 | return 0; | |
3278 | ||
3279 | default: | |
3280 | /* dev_ioctl() should ensure this case | |
3281 | * is never reached | |
3282 | */ | |
3283 | WARN_ON(1); | |
3284 | err = -EINVAL; | |
3285 | break; | |
3286 | ||
3287 | } | |
3288 | return err; | |
3289 | } | |
3290 | ||
3291 | /* | |
3292 | * Perform the SIOCxIFxxx calls, inside rtnl_lock() | |
3293 | */ | |
3294 | static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd) | |
3295 | { | |
3296 | int err; | |
3297 | struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name); | |
3298 | ||
3299 | if (!dev) | |
3300 | return -ENODEV; | |
3301 | ||
3302 | switch (cmd) { | |
3303 | case SIOCSIFFLAGS: /* Set interface flags */ | |
3304 | return dev_change_flags(dev, ifr->ifr_flags); | |
3305 | ||
3306 | case SIOCSIFMETRIC: /* Set the metric on the interface | |
3307 | (currently unused) */ | |
3308 | return -EOPNOTSUPP; | |
3309 | ||
3310 | case SIOCSIFMTU: /* Set the MTU of a device */ | |
3311 | return dev_set_mtu(dev, ifr->ifr_mtu); | |
3312 | ||
1da177e4 LT |
3313 | case SIOCSIFHWADDR: |
3314 | return dev_set_mac_address(dev, &ifr->ifr_hwaddr); | |
3315 | ||
3316 | case SIOCSIFHWBROADCAST: | |
3317 | if (ifr->ifr_hwaddr.sa_family != dev->type) | |
3318 | return -EINVAL; | |
3319 | memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data, | |
3320 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
056925ab | 3321 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
1da177e4 LT |
3322 | return 0; |
3323 | ||
1da177e4 LT |
3324 | case SIOCSIFMAP: |
3325 | if (dev->set_config) { | |
3326 | if (!netif_device_present(dev)) | |
3327 | return -ENODEV; | |
3328 | return dev->set_config(dev, &ifr->ifr_map); | |
3329 | } | |
3330 | return -EOPNOTSUPP; | |
3331 | ||
3332 | case SIOCADDMULTI: | |
3333 | if (!dev->set_multicast_list || | |
3334 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) | |
3335 | return -EINVAL; | |
3336 | if (!netif_device_present(dev)) | |
3337 | return -ENODEV; | |
3338 | return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data, | |
3339 | dev->addr_len, 1); | |
3340 | ||
3341 | case SIOCDELMULTI: | |
3342 | if (!dev->set_multicast_list || | |
3343 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) | |
3344 | return -EINVAL; | |
3345 | if (!netif_device_present(dev)) | |
3346 | return -ENODEV; | |
3347 | return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data, | |
3348 | dev->addr_len, 1); | |
3349 | ||
1da177e4 LT |
3350 | case SIOCSIFTXQLEN: |
3351 | if (ifr->ifr_qlen < 0) | |
3352 | return -EINVAL; | |
3353 | dev->tx_queue_len = ifr->ifr_qlen; | |
3354 | return 0; | |
3355 | ||
3356 | case SIOCSIFNAME: | |
3357 | ifr->ifr_newname[IFNAMSIZ-1] = '\0'; | |
3358 | return dev_change_name(dev, ifr->ifr_newname); | |
3359 | ||
3360 | /* | |
3361 | * Unknown or private ioctl | |
3362 | */ | |
3363 | ||
3364 | default: | |
3365 | if ((cmd >= SIOCDEVPRIVATE && | |
3366 | cmd <= SIOCDEVPRIVATE + 15) || | |
3367 | cmd == SIOCBONDENSLAVE || | |
3368 | cmd == SIOCBONDRELEASE || | |
3369 | cmd == SIOCBONDSETHWADDR || | |
3370 | cmd == SIOCBONDSLAVEINFOQUERY || | |
3371 | cmd == SIOCBONDINFOQUERY || | |
3372 | cmd == SIOCBONDCHANGEACTIVE || | |
3373 | cmd == SIOCGMIIPHY || | |
3374 | cmd == SIOCGMIIREG || | |
3375 | cmd == SIOCSMIIREG || | |
3376 | cmd == SIOCBRADDIF || | |
3377 | cmd == SIOCBRDELIF || | |
3378 | cmd == SIOCWANDEV) { | |
3379 | err = -EOPNOTSUPP; | |
3380 | if (dev->do_ioctl) { | |
3381 | if (netif_device_present(dev)) | |
3382 | err = dev->do_ioctl(dev, ifr, | |
3383 | cmd); | |
3384 | else | |
3385 | err = -ENODEV; | |
3386 | } | |
3387 | } else | |
3388 | err = -EINVAL; | |
3389 | ||
3390 | } | |
3391 | return err; | |
3392 | } | |
3393 | ||
3394 | /* | |
3395 | * This function handles all "interface"-type I/O control requests. The actual | |
3396 | * 'doing' part of this is dev_ifsioc above. | |
3397 | */ | |
3398 | ||
3399 | /** | |
3400 | * dev_ioctl - network device ioctl | |
c4ea43c5 | 3401 | * @net: the applicable net namespace |
1da177e4 LT |
3402 | * @cmd: command to issue |
3403 | * @arg: pointer to a struct ifreq in user space | |
3404 | * | |
3405 | * Issue ioctl functions to devices. This is normally called by the | |
3406 | * user space syscall interfaces but can sometimes be useful for | |
3407 | * other purposes. The return value is the return from the syscall if | |
3408 | * positive or a negative errno code on error. | |
3409 | */ | |
3410 | ||
881d966b | 3411 | int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg) |
1da177e4 LT |
3412 | { |
3413 | struct ifreq ifr; | |
3414 | int ret; | |
3415 | char *colon; | |
3416 | ||
3417 | /* One special case: SIOCGIFCONF takes ifconf argument | |
3418 | and requires shared lock, because it sleeps writing | |
3419 | to user space. | |
3420 | */ | |
3421 | ||
3422 | if (cmd == SIOCGIFCONF) { | |
6756ae4b | 3423 | rtnl_lock(); |
881d966b | 3424 | ret = dev_ifconf(net, (char __user *) arg); |
6756ae4b | 3425 | rtnl_unlock(); |
1da177e4 LT |
3426 | return ret; |
3427 | } | |
3428 | if (cmd == SIOCGIFNAME) | |
881d966b | 3429 | return dev_ifname(net, (struct ifreq __user *)arg); |
1da177e4 LT |
3430 | |
3431 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
3432 | return -EFAULT; | |
3433 | ||
3434 | ifr.ifr_name[IFNAMSIZ-1] = 0; | |
3435 | ||
3436 | colon = strchr(ifr.ifr_name, ':'); | |
3437 | if (colon) | |
3438 | *colon = 0; | |
3439 | ||
3440 | /* | |
3441 | * See which interface the caller is talking about. | |
3442 | */ | |
3443 | ||
3444 | switch (cmd) { | |
3445 | /* | |
3446 | * These ioctl calls: | |
3447 | * - can be done by all. | |
3448 | * - atomic and do not require locking. | |
3449 | * - return a value | |
3450 | */ | |
3451 | case SIOCGIFFLAGS: | |
3452 | case SIOCGIFMETRIC: | |
3453 | case SIOCGIFMTU: | |
3454 | case SIOCGIFHWADDR: | |
3455 | case SIOCGIFSLAVE: | |
3456 | case SIOCGIFMAP: | |
3457 | case SIOCGIFINDEX: | |
3458 | case SIOCGIFTXQLEN: | |
881d966b | 3459 | dev_load(net, ifr.ifr_name); |
1da177e4 | 3460 | read_lock(&dev_base_lock); |
14e3e079 | 3461 | ret = dev_ifsioc_locked(net, &ifr, cmd); |
1da177e4 LT |
3462 | read_unlock(&dev_base_lock); |
3463 | if (!ret) { | |
3464 | if (colon) | |
3465 | *colon = ':'; | |
3466 | if (copy_to_user(arg, &ifr, | |
3467 | sizeof(struct ifreq))) | |
3468 | ret = -EFAULT; | |
3469 | } | |
3470 | return ret; | |
3471 | ||
3472 | case SIOCETHTOOL: | |
881d966b | 3473 | dev_load(net, ifr.ifr_name); |
1da177e4 | 3474 | rtnl_lock(); |
881d966b | 3475 | ret = dev_ethtool(net, &ifr); |
1da177e4 LT |
3476 | rtnl_unlock(); |
3477 | if (!ret) { | |
3478 | if (colon) | |
3479 | *colon = ':'; | |
3480 | if (copy_to_user(arg, &ifr, | |
3481 | sizeof(struct ifreq))) | |
3482 | ret = -EFAULT; | |
3483 | } | |
3484 | return ret; | |
3485 | ||
3486 | /* | |
3487 | * These ioctl calls: | |
3488 | * - require superuser power. | |
3489 | * - require strict serialization. | |
3490 | * - return a value | |
3491 | */ | |
3492 | case SIOCGMIIPHY: | |
3493 | case SIOCGMIIREG: | |
3494 | case SIOCSIFNAME: | |
3495 | if (!capable(CAP_NET_ADMIN)) | |
3496 | return -EPERM; | |
881d966b | 3497 | dev_load(net, ifr.ifr_name); |
1da177e4 | 3498 | rtnl_lock(); |
881d966b | 3499 | ret = dev_ifsioc(net, &ifr, cmd); |
1da177e4 LT |
3500 | rtnl_unlock(); |
3501 | if (!ret) { | |
3502 | if (colon) | |
3503 | *colon = ':'; | |
3504 | if (copy_to_user(arg, &ifr, | |
3505 | sizeof(struct ifreq))) | |
3506 | ret = -EFAULT; | |
3507 | } | |
3508 | return ret; | |
3509 | ||
3510 | /* | |
3511 | * These ioctl calls: | |
3512 | * - require superuser power. | |
3513 | * - require strict serialization. | |
3514 | * - do not return a value | |
3515 | */ | |
3516 | case SIOCSIFFLAGS: | |
3517 | case SIOCSIFMETRIC: | |
3518 | case SIOCSIFMTU: | |
3519 | case SIOCSIFMAP: | |
3520 | case SIOCSIFHWADDR: | |
3521 | case SIOCSIFSLAVE: | |
3522 | case SIOCADDMULTI: | |
3523 | case SIOCDELMULTI: | |
3524 | case SIOCSIFHWBROADCAST: | |
3525 | case SIOCSIFTXQLEN: | |
3526 | case SIOCSMIIREG: | |
3527 | case SIOCBONDENSLAVE: | |
3528 | case SIOCBONDRELEASE: | |
3529 | case SIOCBONDSETHWADDR: | |
1da177e4 LT |
3530 | case SIOCBONDCHANGEACTIVE: |
3531 | case SIOCBRADDIF: | |
3532 | case SIOCBRDELIF: | |
3533 | if (!capable(CAP_NET_ADMIN)) | |
3534 | return -EPERM; | |
cabcac0b TG |
3535 | /* fall through */ |
3536 | case SIOCBONDSLAVEINFOQUERY: | |
3537 | case SIOCBONDINFOQUERY: | |
881d966b | 3538 | dev_load(net, ifr.ifr_name); |
1da177e4 | 3539 | rtnl_lock(); |
881d966b | 3540 | ret = dev_ifsioc(net, &ifr, cmd); |
1da177e4 LT |
3541 | rtnl_unlock(); |
3542 | return ret; | |
3543 | ||
3544 | case SIOCGIFMEM: | |
3545 | /* Get the per device memory space. We can add this but | |
3546 | * currently do not support it */ | |
3547 | case SIOCSIFMEM: | |
3548 | /* Set the per device memory buffer space. | |
3549 | * Not applicable in our case */ | |
3550 | case SIOCSIFLINK: | |
3551 | return -EINVAL; | |
3552 | ||
3553 | /* | |
3554 | * Unknown or private ioctl. | |
3555 | */ | |
3556 | default: | |
3557 | if (cmd == SIOCWANDEV || | |
3558 | (cmd >= SIOCDEVPRIVATE && | |
3559 | cmd <= SIOCDEVPRIVATE + 15)) { | |
881d966b | 3560 | dev_load(net, ifr.ifr_name); |
1da177e4 | 3561 | rtnl_lock(); |
881d966b | 3562 | ret = dev_ifsioc(net, &ifr, cmd); |
1da177e4 LT |
3563 | rtnl_unlock(); |
3564 | if (!ret && copy_to_user(arg, &ifr, | |
3565 | sizeof(struct ifreq))) | |
3566 | ret = -EFAULT; | |
3567 | return ret; | |
3568 | } | |
1da177e4 | 3569 | /* Take care of Wireless Extensions */ |
295f4a1f | 3570 | if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) |
881d966b | 3571 | return wext_handle_ioctl(net, &ifr, cmd, arg); |
1da177e4 LT |
3572 | return -EINVAL; |
3573 | } | |
3574 | } | |
3575 | ||
3576 | ||
3577 | /** | |
3578 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 3579 | * @net: the applicable net namespace |
1da177e4 LT |
3580 | * |
3581 | * Returns a suitable unique value for a new device interface | |
3582 | * number. The caller must hold the rtnl semaphore or the | |
3583 | * dev_base_lock to be sure it remains unique. | |
3584 | */ | |
881d966b | 3585 | static int dev_new_index(struct net *net) |
1da177e4 LT |
3586 | { |
3587 | static int ifindex; | |
3588 | for (;;) { | |
3589 | if (++ifindex <= 0) | |
3590 | ifindex = 1; | |
881d966b | 3591 | if (!__dev_get_by_index(net, ifindex)) |
1da177e4 LT |
3592 | return ifindex; |
3593 | } | |
3594 | } | |
3595 | ||
1da177e4 LT |
3596 | /* Delayed registration/unregisteration */ |
3597 | static DEFINE_SPINLOCK(net_todo_list_lock); | |
3b5b34fd | 3598 | static LIST_HEAD(net_todo_list); |
1da177e4 | 3599 | |
6f05f629 | 3600 | static void net_set_todo(struct net_device *dev) |
1da177e4 LT |
3601 | { |
3602 | spin_lock(&net_todo_list_lock); | |
3603 | list_add_tail(&dev->todo_list, &net_todo_list); | |
3604 | spin_unlock(&net_todo_list_lock); | |
3605 | } | |
3606 | ||
93ee31f1 DL |
3607 | static void rollback_registered(struct net_device *dev) |
3608 | { | |
3609 | BUG_ON(dev_boot_phase); | |
3610 | ASSERT_RTNL(); | |
3611 | ||
3612 | /* Some devices call without registering for initialization unwind. */ | |
3613 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
3614 | printk(KERN_DEBUG "unregister_netdevice: device %s/%p never " | |
3615 | "was registered\n", dev->name, dev); | |
3616 | ||
3617 | WARN_ON(1); | |
3618 | return; | |
3619 | } | |
3620 | ||
3621 | BUG_ON(dev->reg_state != NETREG_REGISTERED); | |
3622 | ||
3623 | /* If device is running, close it first. */ | |
3624 | dev_close(dev); | |
3625 | ||
3626 | /* And unlink it from device chain. */ | |
3627 | unlist_netdevice(dev); | |
3628 | ||
3629 | dev->reg_state = NETREG_UNREGISTERING; | |
3630 | ||
3631 | synchronize_net(); | |
3632 | ||
3633 | /* Shutdown queueing discipline. */ | |
3634 | dev_shutdown(dev); | |
3635 | ||
3636 | ||
3637 | /* Notify protocols, that we are about to destroy | |
3638 | this device. They should clean all the things. | |
3639 | */ | |
3640 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
3641 | ||
3642 | /* | |
3643 | * Flush the unicast and multicast chains | |
3644 | */ | |
3645 | dev_addr_discard(dev); | |
3646 | ||
3647 | if (dev->uninit) | |
3648 | dev->uninit(dev); | |
3649 | ||
3650 | /* Notifier chain MUST detach us from master device. */ | |
3651 | BUG_TRAP(!dev->master); | |
3652 | ||
3653 | /* Remove entries from kobject tree */ | |
3654 | netdev_unregister_kobject(dev); | |
3655 | ||
3656 | synchronize_net(); | |
3657 | ||
3658 | dev_put(dev); | |
3659 | } | |
3660 | ||
1da177e4 LT |
3661 | /** |
3662 | * register_netdevice - register a network device | |
3663 | * @dev: device to register | |
3664 | * | |
3665 | * Take a completed network device structure and add it to the kernel | |
3666 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
3667 | * chain. 0 is returned on success. A negative errno code is returned | |
3668 | * on a failure to set up the device, or if the name is a duplicate. | |
3669 | * | |
3670 | * Callers must hold the rtnl semaphore. You may want | |
3671 | * register_netdev() instead of this. | |
3672 | * | |
3673 | * BUGS: | |
3674 | * The locking appears insufficient to guarantee two parallel registers | |
3675 | * will not get the same name. | |
3676 | */ | |
3677 | ||
3678 | int register_netdevice(struct net_device *dev) | |
3679 | { | |
3680 | struct hlist_head *head; | |
3681 | struct hlist_node *p; | |
3682 | int ret; | |
881d966b | 3683 | struct net *net; |
1da177e4 LT |
3684 | |
3685 | BUG_ON(dev_boot_phase); | |
3686 | ASSERT_RTNL(); | |
3687 | ||
b17a7c17 SH |
3688 | might_sleep(); |
3689 | ||
1da177e4 LT |
3690 | /* When net_device's are persistent, this will be fatal. */ |
3691 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
881d966b EB |
3692 | BUG_ON(!dev->nd_net); |
3693 | net = dev->nd_net; | |
1da177e4 LT |
3694 | |
3695 | spin_lock_init(&dev->queue_lock); | |
932ff279 | 3696 | spin_lock_init(&dev->_xmit_lock); |
723e98b7 | 3697 | netdev_set_lockdep_class(&dev->_xmit_lock, dev->type); |
1da177e4 | 3698 | dev->xmit_lock_owner = -1; |
1da177e4 | 3699 | spin_lock_init(&dev->ingress_lock); |
1da177e4 | 3700 | |
1da177e4 LT |
3701 | dev->iflink = -1; |
3702 | ||
3703 | /* Init, if this function is available */ | |
3704 | if (dev->init) { | |
3705 | ret = dev->init(dev); | |
3706 | if (ret) { | |
3707 | if (ret > 0) | |
3708 | ret = -EIO; | |
90833aa4 | 3709 | goto out; |
1da177e4 LT |
3710 | } |
3711 | } | |
4ec93edb | 3712 | |
1da177e4 LT |
3713 | if (!dev_valid_name(dev->name)) { |
3714 | ret = -EINVAL; | |
7ce1b0ed | 3715 | goto err_uninit; |
1da177e4 LT |
3716 | } |
3717 | ||
881d966b | 3718 | dev->ifindex = dev_new_index(net); |
1da177e4 LT |
3719 | if (dev->iflink == -1) |
3720 | dev->iflink = dev->ifindex; | |
3721 | ||
3722 | /* Check for existence of name */ | |
881d966b | 3723 | head = dev_name_hash(net, dev->name); |
1da177e4 LT |
3724 | hlist_for_each(p, head) { |
3725 | struct net_device *d | |
3726 | = hlist_entry(p, struct net_device, name_hlist); | |
3727 | if (!strncmp(d->name, dev->name, IFNAMSIZ)) { | |
3728 | ret = -EEXIST; | |
7ce1b0ed | 3729 | goto err_uninit; |
1da177e4 | 3730 | } |
4ec93edb | 3731 | } |
1da177e4 | 3732 | |
d212f87b SH |
3733 | /* Fix illegal checksum combinations */ |
3734 | if ((dev->features & NETIF_F_HW_CSUM) && | |
3735 | (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
3736 | printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n", | |
3737 | dev->name); | |
3738 | dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); | |
3739 | } | |
3740 | ||
3741 | if ((dev->features & NETIF_F_NO_CSUM) && | |
3742 | (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
3743 | printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n", | |
3744 | dev->name); | |
3745 | dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM); | |
3746 | } | |
3747 | ||
3748 | ||
1da177e4 LT |
3749 | /* Fix illegal SG+CSUM combinations. */ |
3750 | if ((dev->features & NETIF_F_SG) && | |
8648b305 | 3751 | !(dev->features & NETIF_F_ALL_CSUM)) { |
5a8da02b | 3752 | printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no checksum feature.\n", |
1da177e4 LT |
3753 | dev->name); |
3754 | dev->features &= ~NETIF_F_SG; | |
3755 | } | |
3756 | ||
3757 | /* TSO requires that SG is present as well. */ | |
3758 | if ((dev->features & NETIF_F_TSO) && | |
3759 | !(dev->features & NETIF_F_SG)) { | |
5a8da02b | 3760 | printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no SG feature.\n", |
1da177e4 LT |
3761 | dev->name); |
3762 | dev->features &= ~NETIF_F_TSO; | |
3763 | } | |
e89e9cf5 AR |
3764 | if (dev->features & NETIF_F_UFO) { |
3765 | if (!(dev->features & NETIF_F_HW_CSUM)) { | |
3766 | printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no " | |
3767 | "NETIF_F_HW_CSUM feature.\n", | |
3768 | dev->name); | |
3769 | dev->features &= ~NETIF_F_UFO; | |
3770 | } | |
3771 | if (!(dev->features & NETIF_F_SG)) { | |
3772 | printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no " | |
3773 | "NETIF_F_SG feature.\n", | |
3774 | dev->name); | |
3775 | dev->features &= ~NETIF_F_UFO; | |
3776 | } | |
3777 | } | |
1da177e4 | 3778 | |
8b41d188 | 3779 | ret = netdev_register_kobject(dev); |
b17a7c17 | 3780 | if (ret) |
7ce1b0ed | 3781 | goto err_uninit; |
b17a7c17 SH |
3782 | dev->reg_state = NETREG_REGISTERED; |
3783 | ||
1da177e4 LT |
3784 | /* |
3785 | * Default initial state at registry is that the | |
3786 | * device is present. | |
3787 | */ | |
3788 | ||
3789 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
3790 | ||
1da177e4 | 3791 | dev_init_scheduler(dev); |
1da177e4 | 3792 | dev_hold(dev); |
ce286d32 | 3793 | list_netdevice(dev); |
1da177e4 LT |
3794 | |
3795 | /* Notify protocols, that a new device appeared. */ | |
056925ab | 3796 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 3797 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
3798 | if (ret) { |
3799 | rollback_registered(dev); | |
3800 | dev->reg_state = NETREG_UNREGISTERED; | |
3801 | } | |
1da177e4 LT |
3802 | |
3803 | out: | |
3804 | return ret; | |
7ce1b0ed HX |
3805 | |
3806 | err_uninit: | |
3807 | if (dev->uninit) | |
3808 | dev->uninit(dev); | |
3809 | goto out; | |
1da177e4 LT |
3810 | } |
3811 | ||
3812 | /** | |
3813 | * register_netdev - register a network device | |
3814 | * @dev: device to register | |
3815 | * | |
3816 | * Take a completed network device structure and add it to the kernel | |
3817 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
3818 | * chain. 0 is returned on success. A negative errno code is returned | |
3819 | * on a failure to set up the device, or if the name is a duplicate. | |
3820 | * | |
38b4da38 | 3821 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
3822 | * and expands the device name if you passed a format string to |
3823 | * alloc_netdev. | |
3824 | */ | |
3825 | int register_netdev(struct net_device *dev) | |
3826 | { | |
3827 | int err; | |
3828 | ||
3829 | rtnl_lock(); | |
3830 | ||
3831 | /* | |
3832 | * If the name is a format string the caller wants us to do a | |
3833 | * name allocation. | |
3834 | */ | |
3835 | if (strchr(dev->name, '%')) { | |
3836 | err = dev_alloc_name(dev, dev->name); | |
3837 | if (err < 0) | |
3838 | goto out; | |
3839 | } | |
4ec93edb | 3840 | |
1da177e4 LT |
3841 | err = register_netdevice(dev); |
3842 | out: | |
3843 | rtnl_unlock(); | |
3844 | return err; | |
3845 | } | |
3846 | EXPORT_SYMBOL(register_netdev); | |
3847 | ||
3848 | /* | |
3849 | * netdev_wait_allrefs - wait until all references are gone. | |
3850 | * | |
3851 | * This is called when unregistering network devices. | |
3852 | * | |
3853 | * Any protocol or device that holds a reference should register | |
3854 | * for netdevice notification, and cleanup and put back the | |
3855 | * reference if they receive an UNREGISTER event. | |
3856 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 3857 | * call dev_put. |
1da177e4 LT |
3858 | */ |
3859 | static void netdev_wait_allrefs(struct net_device *dev) | |
3860 | { | |
3861 | unsigned long rebroadcast_time, warning_time; | |
3862 | ||
3863 | rebroadcast_time = warning_time = jiffies; | |
3864 | while (atomic_read(&dev->refcnt) != 0) { | |
3865 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { | |
6756ae4b | 3866 | rtnl_lock(); |
1da177e4 LT |
3867 | |
3868 | /* Rebroadcast unregister notification */ | |
056925ab | 3869 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
1da177e4 LT |
3870 | |
3871 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, | |
3872 | &dev->state)) { | |
3873 | /* We must not have linkwatch events | |
3874 | * pending on unregister. If this | |
3875 | * happens, we simply run the queue | |
3876 | * unscheduled, resulting in a noop | |
3877 | * for this device. | |
3878 | */ | |
3879 | linkwatch_run_queue(); | |
3880 | } | |
3881 | ||
6756ae4b | 3882 | __rtnl_unlock(); |
1da177e4 LT |
3883 | |
3884 | rebroadcast_time = jiffies; | |
3885 | } | |
3886 | ||
3887 | msleep(250); | |
3888 | ||
3889 | if (time_after(jiffies, warning_time + 10 * HZ)) { | |
3890 | printk(KERN_EMERG "unregister_netdevice: " | |
3891 | "waiting for %s to become free. Usage " | |
3892 | "count = %d\n", | |
3893 | dev->name, atomic_read(&dev->refcnt)); | |
3894 | warning_time = jiffies; | |
3895 | } | |
3896 | } | |
3897 | } | |
3898 | ||
3899 | /* The sequence is: | |
3900 | * | |
3901 | * rtnl_lock(); | |
3902 | * ... | |
3903 | * register_netdevice(x1); | |
3904 | * register_netdevice(x2); | |
3905 | * ... | |
3906 | * unregister_netdevice(y1); | |
3907 | * unregister_netdevice(y2); | |
3908 | * ... | |
3909 | * rtnl_unlock(); | |
3910 | * free_netdev(y1); | |
3911 | * free_netdev(y2); | |
3912 | * | |
3913 | * We are invoked by rtnl_unlock() after it drops the semaphore. | |
3914 | * This allows us to deal with problems: | |
b17a7c17 | 3915 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
3916 | * without deadlocking with linkwatch via keventd. |
3917 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
3918 | * safely in order to wait for the netdev refcnt to drop to zero. | |
3919 | */ | |
4a3e2f71 | 3920 | static DEFINE_MUTEX(net_todo_run_mutex); |
1da177e4 LT |
3921 | void netdev_run_todo(void) |
3922 | { | |
626ab0e6 | 3923 | struct list_head list; |
1da177e4 LT |
3924 | |
3925 | /* Need to guard against multiple cpu's getting out of order. */ | |
4a3e2f71 | 3926 | mutex_lock(&net_todo_run_mutex); |
1da177e4 LT |
3927 | |
3928 | /* Not safe to do outside the semaphore. We must not return | |
3929 | * until all unregister events invoked by the local processor | |
3930 | * have been completed (either by this todo run, or one on | |
3931 | * another cpu). | |
3932 | */ | |
3933 | if (list_empty(&net_todo_list)) | |
3934 | goto out; | |
3935 | ||
3936 | /* Snapshot list, allow later requests */ | |
3937 | spin_lock(&net_todo_list_lock); | |
626ab0e6 | 3938 | list_replace_init(&net_todo_list, &list); |
1da177e4 | 3939 | spin_unlock(&net_todo_list_lock); |
626ab0e6 | 3940 | |
1da177e4 LT |
3941 | while (!list_empty(&list)) { |
3942 | struct net_device *dev | |
3943 | = list_entry(list.next, struct net_device, todo_list); | |
3944 | list_del(&dev->todo_list); | |
3945 | ||
b17a7c17 SH |
3946 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
3947 | printk(KERN_ERR "network todo '%s' but state %d\n", | |
3948 | dev->name, dev->reg_state); | |
3949 | dump_stack(); | |
3950 | continue; | |
3951 | } | |
1da177e4 | 3952 | |
b17a7c17 | 3953 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 3954 | |
b17a7c17 | 3955 | netdev_wait_allrefs(dev); |
1da177e4 | 3956 | |
b17a7c17 SH |
3957 | /* paranoia */ |
3958 | BUG_ON(atomic_read(&dev->refcnt)); | |
3959 | BUG_TRAP(!dev->ip_ptr); | |
3960 | BUG_TRAP(!dev->ip6_ptr); | |
3961 | BUG_TRAP(!dev->dn_ptr); | |
1da177e4 | 3962 | |
b17a7c17 SH |
3963 | if (dev->destructor) |
3964 | dev->destructor(dev); | |
9093bbb2 SH |
3965 | |
3966 | /* Free network device */ | |
3967 | kobject_put(&dev->dev.kobj); | |
1da177e4 LT |
3968 | } |
3969 | ||
3970 | out: | |
4a3e2f71 | 3971 | mutex_unlock(&net_todo_run_mutex); |
1da177e4 LT |
3972 | } |
3973 | ||
5a1b5898 | 3974 | static struct net_device_stats *internal_stats(struct net_device *dev) |
c45d286e | 3975 | { |
5a1b5898 | 3976 | return &dev->stats; |
c45d286e RR |
3977 | } |
3978 | ||
1da177e4 | 3979 | /** |
f25f4e44 | 3980 | * alloc_netdev_mq - allocate network device |
1da177e4 LT |
3981 | * @sizeof_priv: size of private data to allocate space for |
3982 | * @name: device name format string | |
3983 | * @setup: callback to initialize device | |
f25f4e44 | 3984 | * @queue_count: the number of subqueues to allocate |
1da177e4 LT |
3985 | * |
3986 | * Allocates a struct net_device with private data area for driver use | |
f25f4e44 PWJ |
3987 | * and performs basic initialization. Also allocates subquue structs |
3988 | * for each queue on the device at the end of the netdevice. | |
1da177e4 | 3989 | */ |
f25f4e44 PWJ |
3990 | struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name, |
3991 | void (*setup)(struct net_device *), unsigned int queue_count) | |
1da177e4 LT |
3992 | { |
3993 | void *p; | |
3994 | struct net_device *dev; | |
3995 | int alloc_size; | |
3996 | ||
b6fe17d6 SH |
3997 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
3998 | ||
1da177e4 | 3999 | /* ensure 32-byte alignment of both the device and private area */ |
f25f4e44 | 4000 | alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST + |
31ce72a6 | 4001 | (sizeof(struct net_device_subqueue) * (queue_count - 1))) & |
f25f4e44 | 4002 | ~NETDEV_ALIGN_CONST; |
1da177e4 LT |
4003 | alloc_size += sizeof_priv + NETDEV_ALIGN_CONST; |
4004 | ||
31380de9 | 4005 | p = kzalloc(alloc_size, GFP_KERNEL); |
1da177e4 | 4006 | if (!p) { |
b6fe17d6 | 4007 | printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n"); |
1da177e4 LT |
4008 | return NULL; |
4009 | } | |
1da177e4 LT |
4010 | |
4011 | dev = (struct net_device *) | |
4012 | (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST); | |
4013 | dev->padded = (char *)dev - (char *)p; | |
6d34b1c2 | 4014 | dev->nd_net = &init_net; |
1da177e4 | 4015 | |
f25f4e44 PWJ |
4016 | if (sizeof_priv) { |
4017 | dev->priv = ((char *)dev + | |
4018 | ((sizeof(struct net_device) + | |
4019 | (sizeof(struct net_device_subqueue) * | |
31ce72a6 | 4020 | (queue_count - 1)) + NETDEV_ALIGN_CONST) |
f25f4e44 PWJ |
4021 | & ~NETDEV_ALIGN_CONST)); |
4022 | } | |
4023 | ||
4024 | dev->egress_subqueue_count = queue_count; | |
1da177e4 | 4025 | |
5a1b5898 | 4026 | dev->get_stats = internal_stats; |
bea3348e | 4027 | netpoll_netdev_init(dev); |
1da177e4 LT |
4028 | setup(dev); |
4029 | strcpy(dev->name, name); | |
4030 | return dev; | |
4031 | } | |
f25f4e44 | 4032 | EXPORT_SYMBOL(alloc_netdev_mq); |
1da177e4 LT |
4033 | |
4034 | /** | |
4035 | * free_netdev - free network device | |
4036 | * @dev: device | |
4037 | * | |
4ec93edb YH |
4038 | * This function does the last stage of destroying an allocated device |
4039 | * interface. The reference to the device object is released. | |
1da177e4 LT |
4040 | * If this is the last reference then it will be freed. |
4041 | */ | |
4042 | void free_netdev(struct net_device *dev) | |
4043 | { | |
3041a069 | 4044 | /* Compatibility with error handling in drivers */ |
1da177e4 LT |
4045 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
4046 | kfree((char *)dev - dev->padded); | |
4047 | return; | |
4048 | } | |
4049 | ||
4050 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
4051 | dev->reg_state = NETREG_RELEASED; | |
4052 | ||
43cb76d9 GKH |
4053 | /* will free via device release */ |
4054 | put_device(&dev->dev); | |
1da177e4 | 4055 | } |
4ec93edb | 4056 | |
1da177e4 | 4057 | /* Synchronize with packet receive processing. */ |
4ec93edb | 4058 | void synchronize_net(void) |
1da177e4 LT |
4059 | { |
4060 | might_sleep(); | |
fbd568a3 | 4061 | synchronize_rcu(); |
1da177e4 LT |
4062 | } |
4063 | ||
4064 | /** | |
4065 | * unregister_netdevice - remove device from the kernel | |
4066 | * @dev: device | |
4067 | * | |
4068 | * This function shuts down a device interface and removes it | |
d59b54b1 | 4069 | * from the kernel tables. |
1da177e4 LT |
4070 | * |
4071 | * Callers must hold the rtnl semaphore. You may want | |
4072 | * unregister_netdev() instead of this. | |
4073 | */ | |
4074 | ||
22f8cde5 | 4075 | void unregister_netdevice(struct net_device *dev) |
1da177e4 | 4076 | { |
a6620712 HX |
4077 | ASSERT_RTNL(); |
4078 | ||
93ee31f1 | 4079 | rollback_registered(dev); |
1da177e4 LT |
4080 | /* Finish processing unregister after unlock */ |
4081 | net_set_todo(dev); | |
1da177e4 LT |
4082 | } |
4083 | ||
4084 | /** | |
4085 | * unregister_netdev - remove device from the kernel | |
4086 | * @dev: device | |
4087 | * | |
4088 | * This function shuts down a device interface and removes it | |
d59b54b1 | 4089 | * from the kernel tables. |
1da177e4 LT |
4090 | * |
4091 | * This is just a wrapper for unregister_netdevice that takes | |
4092 | * the rtnl semaphore. In general you want to use this and not | |
4093 | * unregister_netdevice. | |
4094 | */ | |
4095 | void unregister_netdev(struct net_device *dev) | |
4096 | { | |
4097 | rtnl_lock(); | |
4098 | unregister_netdevice(dev); | |
4099 | rtnl_unlock(); | |
4100 | } | |
4101 | ||
4102 | EXPORT_SYMBOL(unregister_netdev); | |
4103 | ||
ce286d32 EB |
4104 | /** |
4105 | * dev_change_net_namespace - move device to different nethost namespace | |
4106 | * @dev: device | |
4107 | * @net: network namespace | |
4108 | * @pat: If not NULL name pattern to try if the current device name | |
4109 | * is already taken in the destination network namespace. | |
4110 | * | |
4111 | * This function shuts down a device interface and moves it | |
4112 | * to a new network namespace. On success 0 is returned, on | |
4113 | * a failure a netagive errno code is returned. | |
4114 | * | |
4115 | * Callers must hold the rtnl semaphore. | |
4116 | */ | |
4117 | ||
4118 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
4119 | { | |
4120 | char buf[IFNAMSIZ]; | |
4121 | const char *destname; | |
4122 | int err; | |
4123 | ||
4124 | ASSERT_RTNL(); | |
4125 | ||
4126 | /* Don't allow namespace local devices to be moved. */ | |
4127 | err = -EINVAL; | |
4128 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
4129 | goto out; | |
4130 | ||
4131 | /* Ensure the device has been registrered */ | |
4132 | err = -EINVAL; | |
4133 | if (dev->reg_state != NETREG_REGISTERED) | |
4134 | goto out; | |
4135 | ||
4136 | /* Get out if there is nothing todo */ | |
4137 | err = 0; | |
4138 | if (dev->nd_net == net) | |
4139 | goto out; | |
4140 | ||
4141 | /* Pick the destination device name, and ensure | |
4142 | * we can use it in the destination network namespace. | |
4143 | */ | |
4144 | err = -EEXIST; | |
4145 | destname = dev->name; | |
4146 | if (__dev_get_by_name(net, destname)) { | |
4147 | /* We get here if we can't use the current device name */ | |
4148 | if (!pat) | |
4149 | goto out; | |
4150 | if (!dev_valid_name(pat)) | |
4151 | goto out; | |
4152 | if (strchr(pat, '%')) { | |
4153 | if (__dev_alloc_name(net, pat, buf) < 0) | |
4154 | goto out; | |
4155 | destname = buf; | |
4156 | } else | |
4157 | destname = pat; | |
4158 | if (__dev_get_by_name(net, destname)) | |
4159 | goto out; | |
4160 | } | |
4161 | ||
4162 | /* | |
4163 | * And now a mini version of register_netdevice unregister_netdevice. | |
4164 | */ | |
4165 | ||
4166 | /* If device is running close it first. */ | |
9b772652 | 4167 | dev_close(dev); |
ce286d32 EB |
4168 | |
4169 | /* And unlink it from device chain */ | |
4170 | err = -ENODEV; | |
4171 | unlist_netdevice(dev); | |
4172 | ||
4173 | synchronize_net(); | |
4174 | ||
4175 | /* Shutdown queueing discipline. */ | |
4176 | dev_shutdown(dev); | |
4177 | ||
4178 | /* Notify protocols, that we are about to destroy | |
4179 | this device. They should clean all the things. | |
4180 | */ | |
4181 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
4182 | ||
4183 | /* | |
4184 | * Flush the unicast and multicast chains | |
4185 | */ | |
4186 | dev_addr_discard(dev); | |
4187 | ||
4188 | /* Actually switch the network namespace */ | |
4189 | dev->nd_net = net; | |
4190 | ||
4191 | /* Assign the new device name */ | |
4192 | if (destname != dev->name) | |
4193 | strcpy(dev->name, destname); | |
4194 | ||
4195 | /* If there is an ifindex conflict assign a new one */ | |
4196 | if (__dev_get_by_index(net, dev->ifindex)) { | |
4197 | int iflink = (dev->iflink == dev->ifindex); | |
4198 | dev->ifindex = dev_new_index(net); | |
4199 | if (iflink) | |
4200 | dev->iflink = dev->ifindex; | |
4201 | } | |
4202 | ||
8b41d188 | 4203 | /* Fixup kobjects */ |
ce286d32 | 4204 | err = device_rename(&dev->dev, dev->name); |
8b41d188 | 4205 | WARN_ON(err); |
ce286d32 EB |
4206 | |
4207 | /* Add the device back in the hashes */ | |
4208 | list_netdevice(dev); | |
4209 | ||
4210 | /* Notify protocols, that a new device appeared. */ | |
4211 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
4212 | ||
4213 | synchronize_net(); | |
4214 | err = 0; | |
4215 | out: | |
4216 | return err; | |
4217 | } | |
4218 | ||
1da177e4 LT |
4219 | static int dev_cpu_callback(struct notifier_block *nfb, |
4220 | unsigned long action, | |
4221 | void *ocpu) | |
4222 | { | |
4223 | struct sk_buff **list_skb; | |
4224 | struct net_device **list_net; | |
4225 | struct sk_buff *skb; | |
4226 | unsigned int cpu, oldcpu = (unsigned long)ocpu; | |
4227 | struct softnet_data *sd, *oldsd; | |
4228 | ||
8bb78442 | 4229 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
1da177e4 LT |
4230 | return NOTIFY_OK; |
4231 | ||
4232 | local_irq_disable(); | |
4233 | cpu = smp_processor_id(); | |
4234 | sd = &per_cpu(softnet_data, cpu); | |
4235 | oldsd = &per_cpu(softnet_data, oldcpu); | |
4236 | ||
4237 | /* Find end of our completion_queue. */ | |
4238 | list_skb = &sd->completion_queue; | |
4239 | while (*list_skb) | |
4240 | list_skb = &(*list_skb)->next; | |
4241 | /* Append completion queue from offline CPU. */ | |
4242 | *list_skb = oldsd->completion_queue; | |
4243 | oldsd->completion_queue = NULL; | |
4244 | ||
4245 | /* Find end of our output_queue. */ | |
4246 | list_net = &sd->output_queue; | |
4247 | while (*list_net) | |
4248 | list_net = &(*list_net)->next_sched; | |
4249 | /* Append output queue from offline CPU. */ | |
4250 | *list_net = oldsd->output_queue; | |
4251 | oldsd->output_queue = NULL; | |
4252 | ||
4253 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
4254 | local_irq_enable(); | |
4255 | ||
4256 | /* Process offline CPU's input_pkt_queue */ | |
4257 | while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) | |
4258 | netif_rx(skb); | |
4259 | ||
4260 | return NOTIFY_OK; | |
4261 | } | |
1da177e4 | 4262 | |
db217334 CL |
4263 | #ifdef CONFIG_NET_DMA |
4264 | /** | |
0ed72ec4 RD |
4265 | * net_dma_rebalance - try to maintain one DMA channel per CPU |
4266 | * @net_dma: DMA client and associated data (lock, channels, channel_mask) | |
4267 | * | |
4268 | * This is called when the number of channels allocated to the net_dma client | |
4269 | * changes. The net_dma client tries to have one DMA channel per CPU. | |
db217334 | 4270 | */ |
d379b01e DW |
4271 | |
4272 | static void net_dma_rebalance(struct net_dma *net_dma) | |
db217334 | 4273 | { |
d379b01e | 4274 | unsigned int cpu, i, n, chan_idx; |
db217334 CL |
4275 | struct dma_chan *chan; |
4276 | ||
d379b01e | 4277 | if (cpus_empty(net_dma->channel_mask)) { |
db217334 | 4278 | for_each_online_cpu(cpu) |
29bbd72d | 4279 | rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL); |
db217334 CL |
4280 | return; |
4281 | } | |
4282 | ||
4283 | i = 0; | |
4284 | cpu = first_cpu(cpu_online_map); | |
4285 | ||
d379b01e DW |
4286 | for_each_cpu_mask(chan_idx, net_dma->channel_mask) { |
4287 | chan = net_dma->channels[chan_idx]; | |
4288 | ||
4289 | n = ((num_online_cpus() / cpus_weight(net_dma->channel_mask)) | |
4290 | + (i < (num_online_cpus() % | |
4291 | cpus_weight(net_dma->channel_mask)) ? 1 : 0)); | |
db217334 CL |
4292 | |
4293 | while(n) { | |
29bbd72d | 4294 | per_cpu(softnet_data, cpu).net_dma = chan; |
db217334 CL |
4295 | cpu = next_cpu(cpu, cpu_online_map); |
4296 | n--; | |
4297 | } | |
4298 | i++; | |
4299 | } | |
db217334 CL |
4300 | } |
4301 | ||
4302 | /** | |
4303 | * netdev_dma_event - event callback for the net_dma_client | |
4304 | * @client: should always be net_dma_client | |
f4b8ea78 | 4305 | * @chan: DMA channel for the event |
0ed72ec4 | 4306 | * @state: DMA state to be handled |
db217334 | 4307 | */ |
d379b01e DW |
4308 | static enum dma_state_client |
4309 | netdev_dma_event(struct dma_client *client, struct dma_chan *chan, | |
4310 | enum dma_state state) | |
4311 | { | |
4312 | int i, found = 0, pos = -1; | |
4313 | struct net_dma *net_dma = | |
4314 | container_of(client, struct net_dma, client); | |
4315 | enum dma_state_client ack = DMA_DUP; /* default: take no action */ | |
4316 | ||
4317 | spin_lock(&net_dma->lock); | |
4318 | switch (state) { | |
4319 | case DMA_RESOURCE_AVAILABLE: | |
4320 | for (i = 0; i < NR_CPUS; i++) | |
4321 | if (net_dma->channels[i] == chan) { | |
4322 | found = 1; | |
4323 | break; | |
4324 | } else if (net_dma->channels[i] == NULL && pos < 0) | |
4325 | pos = i; | |
4326 | ||
4327 | if (!found && pos >= 0) { | |
4328 | ack = DMA_ACK; | |
4329 | net_dma->channels[pos] = chan; | |
4330 | cpu_set(pos, net_dma->channel_mask); | |
4331 | net_dma_rebalance(net_dma); | |
4332 | } | |
db217334 CL |
4333 | break; |
4334 | case DMA_RESOURCE_REMOVED: | |
d379b01e DW |
4335 | for (i = 0; i < NR_CPUS; i++) |
4336 | if (net_dma->channels[i] == chan) { | |
4337 | found = 1; | |
4338 | pos = i; | |
4339 | break; | |
4340 | } | |
4341 | ||
4342 | if (found) { | |
4343 | ack = DMA_ACK; | |
4344 | cpu_clear(pos, net_dma->channel_mask); | |
4345 | net_dma->channels[i] = NULL; | |
4346 | net_dma_rebalance(net_dma); | |
4347 | } | |
db217334 CL |
4348 | break; |
4349 | default: | |
4350 | break; | |
4351 | } | |
d379b01e DW |
4352 | spin_unlock(&net_dma->lock); |
4353 | ||
4354 | return ack; | |
db217334 CL |
4355 | } |
4356 | ||
4357 | /** | |
4358 | * netdev_dma_regiser - register the networking subsystem as a DMA client | |
4359 | */ | |
4360 | static int __init netdev_dma_register(void) | |
4361 | { | |
d379b01e DW |
4362 | spin_lock_init(&net_dma.lock); |
4363 | dma_cap_set(DMA_MEMCPY, net_dma.client.cap_mask); | |
4364 | dma_async_client_register(&net_dma.client); | |
4365 | dma_async_client_chan_request(&net_dma.client); | |
db217334 CL |
4366 | return 0; |
4367 | } | |
4368 | ||
4369 | #else | |
4370 | static int __init netdev_dma_register(void) { return -ENODEV; } | |
4371 | #endif /* CONFIG_NET_DMA */ | |
1da177e4 | 4372 | |
7f353bf2 HX |
4373 | /** |
4374 | * netdev_compute_feature - compute conjunction of two feature sets | |
4375 | * @all: first feature set | |
4376 | * @one: second feature set | |
4377 | * | |
4378 | * Computes a new feature set after adding a device with feature set | |
4379 | * @one to the master device with current feature set @all. Returns | |
4380 | * the new feature set. | |
4381 | */ | |
4382 | int netdev_compute_features(unsigned long all, unsigned long one) | |
4383 | { | |
4384 | /* if device needs checksumming, downgrade to hw checksumming */ | |
4385 | if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM)) | |
4386 | all ^= NETIF_F_NO_CSUM | NETIF_F_HW_CSUM; | |
4387 | ||
4388 | /* if device can't do all checksum, downgrade to ipv4/ipv6 */ | |
4389 | if (all & NETIF_F_HW_CSUM && !(one & NETIF_F_HW_CSUM)) | |
4390 | all ^= NETIF_F_HW_CSUM | |
4391 | | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM; | |
4392 | ||
4393 | if (one & NETIF_F_GSO) | |
4394 | one |= NETIF_F_GSO_SOFTWARE; | |
4395 | one |= NETIF_F_GSO; | |
4396 | ||
4397 | /* If even one device supports robust GSO, enable it for all. */ | |
4398 | if (one & NETIF_F_GSO_ROBUST) | |
4399 | all |= NETIF_F_GSO_ROBUST; | |
4400 | ||
4401 | all &= one | NETIF_F_LLTX; | |
4402 | ||
4403 | if (!(all & NETIF_F_ALL_CSUM)) | |
4404 | all &= ~NETIF_F_SG; | |
4405 | if (!(all & NETIF_F_SG)) | |
4406 | all &= ~NETIF_F_GSO_MASK; | |
4407 | ||
4408 | return all; | |
4409 | } | |
4410 | EXPORT_SYMBOL(netdev_compute_features); | |
4411 | ||
30d97d35 PE |
4412 | static struct hlist_head *netdev_create_hash(void) |
4413 | { | |
4414 | int i; | |
4415 | struct hlist_head *hash; | |
4416 | ||
4417 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
4418 | if (hash != NULL) | |
4419 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
4420 | INIT_HLIST_HEAD(&hash[i]); | |
4421 | ||
4422 | return hash; | |
4423 | } | |
4424 | ||
881d966b | 4425 | /* Initialize per network namespace state */ |
4665079c | 4426 | static int __net_init netdev_init(struct net *net) |
881d966b | 4427 | { |
881d966b | 4428 | INIT_LIST_HEAD(&net->dev_base_head); |
881d966b | 4429 | |
30d97d35 PE |
4430 | net->dev_name_head = netdev_create_hash(); |
4431 | if (net->dev_name_head == NULL) | |
4432 | goto err_name; | |
881d966b | 4433 | |
30d97d35 PE |
4434 | net->dev_index_head = netdev_create_hash(); |
4435 | if (net->dev_index_head == NULL) | |
4436 | goto err_idx; | |
881d966b EB |
4437 | |
4438 | return 0; | |
30d97d35 PE |
4439 | |
4440 | err_idx: | |
4441 | kfree(net->dev_name_head); | |
4442 | err_name: | |
4443 | return -ENOMEM; | |
881d966b EB |
4444 | } |
4445 | ||
4665079c | 4446 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
4447 | { |
4448 | kfree(net->dev_name_head); | |
4449 | kfree(net->dev_index_head); | |
4450 | } | |
4451 | ||
022cbae6 | 4452 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
4453 | .init = netdev_init, |
4454 | .exit = netdev_exit, | |
4455 | }; | |
4456 | ||
4665079c | 4457 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 EB |
4458 | { |
4459 | struct net_device *dev, *next; | |
4460 | /* | |
4461 | * Push all migratable of the network devices back to the | |
4462 | * initial network namespace | |
4463 | */ | |
4464 | rtnl_lock(); | |
4465 | for_each_netdev_safe(net, dev, next) { | |
4466 | int err; | |
4467 | ||
4468 | /* Ignore unmoveable devices (i.e. loopback) */ | |
4469 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
4470 | continue; | |
4471 | ||
4472 | /* Push remaing network devices to init_net */ | |
4473 | err = dev_change_net_namespace(dev, &init_net, "dev%d"); | |
4474 | if (err) { | |
4475 | printk(KERN_WARNING "%s: failed to move %s to init_net: %d\n", | |
4476 | __func__, dev->name, err); | |
4477 | unregister_netdevice(dev); | |
4478 | } | |
4479 | } | |
4480 | rtnl_unlock(); | |
4481 | } | |
4482 | ||
022cbae6 | 4483 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 EB |
4484 | .exit = default_device_exit, |
4485 | }; | |
4486 | ||
1da177e4 LT |
4487 | /* |
4488 | * Initialize the DEV module. At boot time this walks the device list and | |
4489 | * unhooks any devices that fail to initialise (normally hardware not | |
4490 | * present) and leaves us with a valid list of present and active devices. | |
4491 | * | |
4492 | */ | |
4493 | ||
4494 | /* | |
4495 | * This is called single threaded during boot, so no need | |
4496 | * to take the rtnl semaphore. | |
4497 | */ | |
4498 | static int __init net_dev_init(void) | |
4499 | { | |
4500 | int i, rc = -ENOMEM; | |
4501 | ||
4502 | BUG_ON(!dev_boot_phase); | |
4503 | ||
1da177e4 LT |
4504 | if (dev_proc_init()) |
4505 | goto out; | |
4506 | ||
8b41d188 | 4507 | if (netdev_kobject_init()) |
1da177e4 LT |
4508 | goto out; |
4509 | ||
4510 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 4511 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
4512 | INIT_LIST_HEAD(&ptype_base[i]); |
4513 | ||
881d966b EB |
4514 | if (register_pernet_subsys(&netdev_net_ops)) |
4515 | goto out; | |
1da177e4 | 4516 | |
ce286d32 EB |
4517 | if (register_pernet_device(&default_device_ops)) |
4518 | goto out; | |
4519 | ||
1da177e4 LT |
4520 | /* |
4521 | * Initialise the packet receive queues. | |
4522 | */ | |
4523 | ||
6f912042 | 4524 | for_each_possible_cpu(i) { |
1da177e4 LT |
4525 | struct softnet_data *queue; |
4526 | ||
4527 | queue = &per_cpu(softnet_data, i); | |
4528 | skb_queue_head_init(&queue->input_pkt_queue); | |
1da177e4 LT |
4529 | queue->completion_queue = NULL; |
4530 | INIT_LIST_HEAD(&queue->poll_list); | |
bea3348e SH |
4531 | |
4532 | queue->backlog.poll = process_backlog; | |
4533 | queue->backlog.weight = weight_p; | |
1da177e4 LT |
4534 | } |
4535 | ||
db217334 CL |
4536 | netdev_dma_register(); |
4537 | ||
1da177e4 LT |
4538 | dev_boot_phase = 0; |
4539 | ||
4540 | open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL); | |
4541 | open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL); | |
4542 | ||
4543 | hotcpu_notifier(dev_cpu_callback, 0); | |
4544 | dst_init(); | |
4545 | dev_mcast_init(); | |
4546 | rc = 0; | |
4547 | out: | |
4548 | return rc; | |
4549 | } | |
4550 | ||
4551 | subsys_initcall(net_dev_init); | |
4552 | ||
4553 | EXPORT_SYMBOL(__dev_get_by_index); | |
4554 | EXPORT_SYMBOL(__dev_get_by_name); | |
4555 | EXPORT_SYMBOL(__dev_remove_pack); | |
c2373ee9 | 4556 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
4557 | EXPORT_SYMBOL(dev_add_pack); |
4558 | EXPORT_SYMBOL(dev_alloc_name); | |
4559 | EXPORT_SYMBOL(dev_close); | |
4560 | EXPORT_SYMBOL(dev_get_by_flags); | |
4561 | EXPORT_SYMBOL(dev_get_by_index); | |
4562 | EXPORT_SYMBOL(dev_get_by_name); | |
1da177e4 LT |
4563 | EXPORT_SYMBOL(dev_open); |
4564 | EXPORT_SYMBOL(dev_queue_xmit); | |
4565 | EXPORT_SYMBOL(dev_remove_pack); | |
4566 | EXPORT_SYMBOL(dev_set_allmulti); | |
4567 | EXPORT_SYMBOL(dev_set_promiscuity); | |
4568 | EXPORT_SYMBOL(dev_change_flags); | |
4569 | EXPORT_SYMBOL(dev_set_mtu); | |
4570 | EXPORT_SYMBOL(dev_set_mac_address); | |
4571 | EXPORT_SYMBOL(free_netdev); | |
4572 | EXPORT_SYMBOL(netdev_boot_setup_check); | |
4573 | EXPORT_SYMBOL(netdev_set_master); | |
4574 | EXPORT_SYMBOL(netdev_state_change); | |
4575 | EXPORT_SYMBOL(netif_receive_skb); | |
4576 | EXPORT_SYMBOL(netif_rx); | |
4577 | EXPORT_SYMBOL(register_gifconf); | |
4578 | EXPORT_SYMBOL(register_netdevice); | |
4579 | EXPORT_SYMBOL(register_netdevice_notifier); | |
4580 | EXPORT_SYMBOL(skb_checksum_help); | |
4581 | EXPORT_SYMBOL(synchronize_net); | |
4582 | EXPORT_SYMBOL(unregister_netdevice); | |
4583 | EXPORT_SYMBOL(unregister_netdevice_notifier); | |
4584 | EXPORT_SYMBOL(net_enable_timestamp); | |
4585 | EXPORT_SYMBOL(net_disable_timestamp); | |
4586 | EXPORT_SYMBOL(dev_get_flags); | |
4587 | ||
4588 | #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE) | |
4589 | EXPORT_SYMBOL(br_handle_frame_hook); | |
4590 | EXPORT_SYMBOL(br_fdb_get_hook); | |
4591 | EXPORT_SYMBOL(br_fdb_put_hook); | |
4592 | #endif | |
4593 | ||
4594 | #ifdef CONFIG_KMOD | |
4595 | EXPORT_SYMBOL(dev_load); | |
4596 | #endif | |
4597 | ||
4598 | EXPORT_PER_CPU_SYMBOL(softnet_data); |