| 1 | |
| 2 | * NET3 Protocol independent device support routines.
|
| 3 | *
|
| 4 | * This program is free software; you can redistribute it and/or
|
| 5 | * modify it under the terms of the GNU General Public License
|
| 6 | * as published by the Free Software Foundation; either version
|
| 7 | * 2 of the License, or (at your option) any later version.
|
| 8 | *
|
| 9 | * Derived from the non IP parts of dev.c 1.0.19
|
| 10 | * Authors: Ross Biro
|
| 11 | * Fred N. van Kempen, <[email protected]>
|
| 12 | * Mark Evans, <[email protected]>
|
| 13 | *
|
| 14 | * Additional Authors:
|
| 15 | * Florian la Roche <[email protected]>
|
| 16 | * Alan Cox <[email protected]>
|
| 17 | * David Hinds <[email protected]>
|
| 18 | * Alexey Kuznetsov <[email protected]>
|
| 19 | * Adam Sulmicki <[email protected]>
|
| 20 | * Pekka Riikonen <[email protected]>
|
| 21 | *
|
| 22 | * Changes:
|
| 23 | * D.J. Barrow : Fixed bug where dev->refcnt gets set
|
| 24 | * to 2 if register_netdev gets called
|
| 25 | * before net_dev_init & also removed a
|
| 26 | * few lines of code in the process.
|
| 27 | * Alan Cox : device private ioctl copies fields back.
|
| 28 | * Alan Cox : Transmit queue code does relevant
|
| 29 | * stunts to keep the queue safe.
|
| 30 | * Alan Cox : Fixed double lock.
|
| 31 | * Alan Cox : Fixed promisc NULL pointer trap
|
| 32 | * ???????? : Support the full private ioctl range
|
| 33 | * Alan Cox : Moved ioctl permission check into
|
| 34 | * drivers
|
| 35 | * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
|
| 36 | * Alan Cox : 100 backlog just doesn't cut it when
|
| 37 | * you start doing multicast video 8)
|
| 38 | * Alan Cox : Rewrote net_bh and list manager.
|
| 39 | * Alan Cox : Fix ETH_P_ALL echoback lengths.
|
| 40 | * Alan Cox : Took out transmit every packet pass
|
| 41 | * Saved a few bytes in the ioctl handler
|
| 42 | * Alan Cox : Network driver sets packet type before
|
| 43 | * calling netif_rx. Saves a function
|
| 44 | * call a packet.
|
| 45 | * Alan Cox : Hashed net_bh()
|
| 46 | * Richard Kooijman: Timestamp fixes.
|
| 47 | * Alan Cox : Wrong field in SIOCGIFDSTADDR
|
| 48 | * Alan Cox : Device lock protection.
|
| 49 | * Alan Cox : Fixed nasty side effect of device close
|
| 50 | * changes.
|
| 51 | * Rudi Cilibrasi : Pass the right thing to
|
| 52 | * set_mac_address()
|
| 53 | * Dave Miller : 32bit quantity for the device lock to
|
| 54 | * make it work out on a Sparc.
|
| 55 | * Bjorn Ekwall : Added KERNELD hack.
|
| 56 | * Alan Cox : Cleaned up the backlog initialise.
|
| 57 | * Craig Metz : SIOCGIFCONF fix if space for under
|
| 58 | * 1 device.
|
| 59 | * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
|
| 60 | * is no device open function.
|
| 61 | * Andi Kleen : Fix error reporting for SIOCGIFCONF
|
| 62 | * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
|
| 63 | * Cyrus Durgin : Cleaned for KMOD
|
| 64 | * Adam Sulmicki : Bug Fix : Network Device Unload
|
| 65 | * A network device unload needs to purge
|
| 66 | * the backlog queue.
|
| 67 | * Paul Rusty Russell : SIOCSIFNAME
|
| 68 | * Pekka Riikonen : Netdev boot-time settings code
|
| 69 | * Andrew Morton : Make unregister_netdevice wait
|
| 70 | * indefinitely on dev->refcnt
|
| 71 | * J Hadi Salim : - Backlog queue sampling
|
| 72 | * - netif_rx() feedback
|
| 73 |
|
| 74 |
|
| 75 | #include <asm/uaccess.h>
|
| 76 | #include <asm/system.h>
|
| 77 | #include <linux/bitops.h>
|
| 78 | #include <linux/capability.h>
|
| 79 | #include <linux/cpu.h>
|
| 80 | #include <linux/types.h>
|
| 81 | #include <linux/kernel.h>
|
| 82 | #include <linux/sched.h>
|
| 83 | #include <linux/mutex.h>
|
| 84 | #include <linux/string.h>
|
| 85 | #include <linux/mm.h>
|
| 86 | #include <linux/socket.h>
|
| 87 | #include <linux/sockios.h>
|
| 88 | #include <linux/errno.h>
|
| 89 | #include <linux/interrupt.h>
|
| 90 | #include <linux/if_ether.h>
|
| 91 | #include <linux/netdevice.h>
|
| 92 | #include <linux/etherdevice.h>
|
| 93 | #include <linux/notifier.h>
|
| 94 | #include <linux/skbuff.h>
|
| 95 | #include <net/net_namespace.h>
|
| 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>
|
| 102 | #include <linux/if_macvlan.h>
|
| 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>
|
| 114 | #include <net/wext.h>
|
| 115 | #include <net/iw_handler.h>
|
| 116 | #include <asm/current.h>
|
| 117 | #include <linux/audit.h>
|
| 118 | #include <linux/dmaengine.h>
|
| 119 | #include <linux/err.h>
|
| 120 | #include <linux/ctype.h>
|
| 121 | #include <linux/if_arp.h>
|
| 122 |
|
| 123 | #include "net-sysfs.h"
|
| 124 |
|
| 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
|
| 136 | * the average user (w/out VLANs) will not be adversely affected.
|
| 137 | * --BLG
|
| 138 | *
|
| 139 | * 0800 IP
|
| 140 | * 8100 802.1Q VLAN
|
| 141 | * 0001 802.3
|
| 142 | * 0002 AX.25
|
| 143 | * 0004 802.2
|
| 144 | * 8035 RARP
|
| 145 | * 0005 SNAP
|
| 146 | * 0805 X.25
|
| 147 | * 0806 ARP
|
| 148 | * 8137 IPX
|
| 149 | * 0009 Localtalk
|
| 150 | * 86DD IPv6
|
| 151 |
|
| 152 |
|
| 153 | #define PTYPE_HASH_SIZE (16)
|
| 154 | #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
|
| 155 |
|
| 156 | static DEFINE_SPINLOCK(ptype_lock);
|
| 157 | static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
|
| 158 | static struct list_head ptype_all __read_mostly;
|
| 159 |
|
| 160 | #ifdef CONFIG_NET_DMA
|
| 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 | };
|
| 177 | #endif
|
| 178 |
|
| 179 | |
| 180 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl
|
| 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
|
| 186 | * dev_base_head list, and hold dev_base_lock for writing when they do the
|
| 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 |
|
| 198 | DEFINE_RWLOCK(dev_base_lock);
|
| 199 |
|
| 200 | EXPORT_SYMBOL(dev_base_lock);
|
| 201 |
|
| 202 | #define NETDEV_HASHBITS 8
|
| 203 | #define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS)
|
| 204 |
|
| 205 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
|
| 206 | {
|
| 207 | unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
|
| 208 | return &net->dev_name_head[hash & ((1 << NETDEV_HASHBITS) - 1)];
|
| 209 | }
|
| 210 |
|
| 211 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
|
| 212 | {
|
| 213 | return &net->dev_index_head[ifindex & ((1 << NETDEV_HASHBITS) - 1)];
|
| 214 | }
|
| 215 |
|
| 216 |
|
| 217 | static int list_netdevice(struct net_device *dev)
|
| 218 | {
|
| 219 | struct net *net = dev_net(dev);
|
| 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 |
|
| 232 | static void unlist_netdevice(struct net_device *dev)
|
| 233 | {
|
| 234 | ASSERT_RTNL();
|
| 235 |
|
| 236 |
|
| 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 |
|
| 244 | |
| 245 | * Our notifier list
|
| 246 |
|
| 247 |
|
| 248 | static RAW_NOTIFIER_HEAD(netdev_chain);
|
| 249 |
|
| 250 | |
| 251 | * Device drivers call our routines to queue packets here. We empty the
|
| 252 | * queue in the local softnet handler.
|
| 253 |
|
| 254 |
|
| 255 | DEFINE_PER_CPU(struct softnet_data, softnet_data);
|
| 256 |
|
| 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 |
|
| 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
|
| 324 |
|
| 325 | |
| 326 |
|
| 327 | Protocol management and registration routines
|
| 328 |
|
| 329 | ******************************************************************************
|
| 330 |
|
| 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 | *
|
| 355 | * This call does not sleep therefore it can not
|
| 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);
|
| 365 | if (pt->type == htons(ETH_P_ALL))
|
| 366 | list_add_rcu(&pt->list, &ptype_all);
|
| 367 | else {
|
| 368 | hash = ntohs(pt->type) & PTYPE_HASH_MASK;
|
| 369 | list_add_rcu(&pt->list, &ptype_base[hash]);
|
| 370 | }
|
| 371 | spin_unlock_bh(&ptype_lock);
|
| 372 | }
|
| 373 |
|
| 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
|
| 381 | * returns.
|
| 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 |
|
| 394 | if (pt->type == htons(ETH_P_ALL))
|
| 395 | head = &ptype_all;
|
| 396 | else
|
| 397 | head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
|
| 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);
|
| 425 |
|
| 426 | synchronize_net();
|
| 427 | }
|
| 428 |
|
| 429 | |
| 430 |
|
| 431 | Device Boot-time Settings Routines
|
| 432 |
|
| 433 | ******************************************************************************
|
| 434 |
|
| 435 |
|
| 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 |
|
| 515 | if (__dev_get_by_name(&init_net, name))
|
| 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 |
|
| 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 |
|
| 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
|
| 561 | * @net: the applicable net namespace
|
| 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 |
|
| 571 | struct net_device *__dev_get_by_name(struct net *net, const char *name)
|
| 572 | {
|
| 573 | struct hlist_node *p;
|
| 574 |
|
| 575 | hlist_for_each(p, dev_name_hash(net, name)) {
|
| 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
|
| 586 | * @net: the applicable net namespace
|
| 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 |
|
| 596 | struct net_device *dev_get_by_name(struct net *net, const char *name)
|
| 597 | {
|
| 598 | struct net_device *dev;
|
| 599 |
|
| 600 | read_lock(&dev_base_lock);
|
| 601 | dev = __dev_get_by_name(net, name);
|
| 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
|
| 610 | * @net: the applicable net namespace
|
| 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 |
|
| 620 | struct net_device *__dev_get_by_index(struct net *net, int ifindex)
|
| 621 | {
|
| 622 | struct hlist_node *p;
|
| 623 |
|
| 624 | hlist_for_each(p, dev_index_hash(net, ifindex)) {
|
| 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
|
| 636 | * @net: the applicable net namespace
|
| 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 |
|
| 645 | struct net_device *dev_get_by_index(struct net *net, int ifindex)
|
| 646 | {
|
| 647 | struct net_device *dev;
|
| 648 |
|
| 649 | read_lock(&dev_base_lock);
|
| 650 | dev = __dev_get_by_index(net, ifindex);
|
| 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
|
| 659 | * @net: the applicable net namespace
|
| 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 |
|
| 672 | struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
|
| 673 | {
|
| 674 | struct net_device *dev;
|
| 675 |
|
| 676 | ASSERT_RTNL();
|
| 677 |
|
| 678 | for_each_netdev(net, dev)
|
| 679 | if (dev->type == type &&
|
| 680 | !memcmp(dev->dev_addr, ha, dev->addr_len))
|
| 681 | return dev;
|
| 682 |
|
| 683 | return NULL;
|
| 684 | }
|
| 685 |
|
| 686 | EXPORT_SYMBOL(dev_getbyhwaddr);
|
| 687 |
|
| 688 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
|
| 689 | {
|
| 690 | struct net_device *dev;
|
| 691 |
|
| 692 | ASSERT_RTNL();
|
| 693 | for_each_netdev(net, dev)
|
| 694 | if (dev->type == type)
|
| 695 | return dev;
|
| 696 |
|
| 697 | return NULL;
|
| 698 | }
|
| 699 |
|
| 700 | EXPORT_SYMBOL(__dev_getfirstbyhwtype);
|
| 701 |
|
| 702 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
|
| 703 | {
|
| 704 | struct net_device *dev;
|
| 705 |
|
| 706 | rtnl_lock();
|
| 707 | dev = __dev_getfirstbyhwtype(net, type);
|
| 708 | if (dev)
|
| 709 | dev_hold(dev);
|
| 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
|
| 718 | * @net: the applicable net namespace
|
| 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
|
| 723 | * is not found or a pointer to the device. The device returned has
|
| 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 |
|
| 728 | struct net_device * dev_get_by_flags(struct net *net, unsigned short if_flags, unsigned short mask)
|
| 729 | {
|
| 730 | struct net_device *dev, *ret;
|
| 731 |
|
| 732 | ret = NULL;
|
| 733 | read_lock(&dev_base_lock);
|
| 734 | for_each_netdev(net, dev) {
|
| 735 | if (((dev->flags ^ if_flags) & mask) == 0) {
|
| 736 | dev_hold(dev);
|
| 737 | ret = dev;
|
| 738 | break;
|
| 739 | }
|
| 740 | }
|
| 741 | read_unlock(&dev_base_lock);
|
| 742 | return ret;
|
| 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
|
| 750 | * to allow sysfs to work. We also disallow any kind of
|
| 751 | * whitespace.
|
| 752 |
|
| 753 | int dev_valid_name(const char *name)
|
| 754 | {
|
| 755 | if (*name == '\0')
|
| 756 | return 0;
|
| 757 | if (strlen(name) >= IFNAMSIZ)
|
| 758 | return 0;
|
| 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;
|
| 768 | }
|
| 769 |
|
| 770 | |
| 771 | * __dev_alloc_name - allocate a name for a device
|
| 772 | * @net: network namespace to allocate the device name in
|
| 773 | * @name: name format string
|
| 774 | * @buf: scratch buffer and result name string
|
| 775 | *
|
| 776 | * Passed a format string - eg "lt%d" it will try and find a suitable
|
| 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.
|
| 783 |
|
| 784 |
|
| 785 | static int __dev_alloc_name(struct net *net, const char *name, char *buf)
|
| 786 | {
|
| 787 | int i = 0;
|
| 788 | const char *p;
|
| 789 | const int max_netdevices = 8*PAGE_SIZE;
|
| 790 | unsigned long *inuse;
|
| 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 |
|
| 804 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
|
| 805 | if (!inuse)
|
| 806 | return -ENOMEM;
|
| 807 |
|
| 808 | for_each_netdev(net, d) {
|
| 809 | if (!sscanf(d->name, name, &i))
|
| 810 | continue;
|
| 811 | if (i < 0 || i >= max_netdevices)
|
| 812 | continue;
|
| 813 |
|
| 814 |
|
| 815 | snprintf(buf, IFNAMSIZ, name, i);
|
| 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 |
|
| 824 | snprintf(buf, IFNAMSIZ, name, i);
|
| 825 | if (!__dev_get_by_name(net, buf))
|
| 826 | return i;
|
| 827 |
|
| 828 | |
| 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 |
|
| 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_net(dev));
|
| 856 | net = dev_net(dev);
|
| 857 | ret = __dev_alloc_name(net, name, buf);
|
| 858 | if (ret >= 0)
|
| 859 | strlcpy(dev->name, buf, IFNAMSIZ);
|
| 860 | return ret;
|
| 861 | }
|
| 862 |
|
| 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 | {
|
| 874 | char oldname[IFNAMSIZ];
|
| 875 | int err = 0;
|
| 876 | int ret;
|
| 877 | struct net *net;
|
| 878 |
|
| 879 | ASSERT_RTNL();
|
| 880 | BUG_ON(!dev_net(dev));
|
| 881 |
|
| 882 | net = dev_net(dev);
|
| 883 | if (dev->flags & IFF_UP)
|
| 884 | return -EBUSY;
|
| 885 |
|
| 886 | if (!dev_valid_name(newname))
|
| 887 | return -EINVAL;
|
| 888 |
|
| 889 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
|
| 890 | return 0;
|
| 891 |
|
| 892 | memcpy(oldname, dev->name, IFNAMSIZ);
|
| 893 |
|
| 894 | if (strchr(newname, '%')) {
|
| 895 | err = dev_alloc_name(dev, newname);
|
| 896 | if (err < 0)
|
| 897 | return err;
|
| 898 | strcpy(newname, dev->name);
|
| 899 | }
|
| 900 | else if (__dev_get_by_name(net, newname))
|
| 901 | return -EEXIST;
|
| 902 | else
|
| 903 | strlcpy(dev->name, newname, IFNAMSIZ);
|
| 904 |
|
| 905 | rollback:
|
| 906 | device_rename(&dev->dev, dev->name);
|
| 907 |
|
| 908 | write_lock_bh(&dev_base_lock);
|
| 909 | hlist_del(&dev->name_hlist);
|
| 910 | hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
|
| 911 | write_unlock_bh(&dev_base_lock);
|
| 912 |
|
| 913 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
|
| 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 | }
|
| 927 |
|
| 928 | return err;
|
| 929 | }
|
| 930 |
|
| 931 | |
| 932 | * netdev_features_change - device changes features
|
| 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 | {
|
| 939 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
|
| 940 | }
|
| 941 | EXPORT_SYMBOL(netdev_features_change);
|
| 942 |
|
| 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) {
|
| 954 | call_netdevice_notifiers(NETDEV_CHANGE, dev);
|
| 955 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
|
| 956 | }
|
| 957 | }
|
| 958 |
|
| 959 | |
| 960 | * dev_load - load a network module
|
| 961 | * @net: the applicable net namespace
|
| 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 |
|
| 969 | void dev_load(struct net *net, const char *name)
|
| 970 | {
|
| 971 | struct net_device *dev;
|
| 972 |
|
| 973 | read_lock(&dev_base_lock);
|
| 974 | dev = __dev_get_by_name(net, name);
|
| 975 | read_unlock(&dev_base_lock);
|
| 976 |
|
| 977 | if (!dev && capable(CAP_SYS_MODULE))
|
| 978 | request_module("%s", name);
|
| 979 | }
|
| 980 |
|
| 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);
|
| 1014 |
|
| 1015 | if (dev->validate_addr)
|
| 1016 | ret = dev->validate_addr(dev);
|
| 1017 |
|
| 1018 | if (!ret && dev->open)
|
| 1019 | ret = dev->open(dev);
|
| 1020 |
|
| 1021 | |
| 1022 | * If it went open OK then:
|
| 1023 |
|
| 1024 |
|
| 1025 | if (ret)
|
| 1026 | clear_bit(__LINK_STATE_START, &dev->state);
|
| 1027 | else {
|
| 1028 | |
| 1029 | * Set the flags.
|
| 1030 |
|
| 1031 | dev->flags |= IFF_UP;
|
| 1032 |
|
| 1033 | |
| 1034 | * Initialize multicasting status
|
| 1035 |
|
| 1036 | dev_set_rx_mode(dev);
|
| 1037 |
|
| 1038 | |
| 1039 | * Wakeup transmit queue engine
|
| 1040 |
|
| 1041 | dev_activate(dev);
|
| 1042 |
|
| 1043 | |
| 1044 | * ... and announce new interface.
|
| 1045 |
|
| 1046 | call_netdevice_notifiers(NETDEV_UP, dev);
|
| 1047 | }
|
| 1048 |
|
| 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 | {
|
| 1063 | might_sleep();
|
| 1064 |
|
| 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 |
|
| 1072 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
|
| 1073 |
|
| 1074 | clear_bit(__LINK_STATE_START, &dev->state);
|
| 1075 |
|
| 1076 | |
| 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 |
|
| 1082 | smp_mb__after_clear_bit();
|
| 1083 |
|
| 1084 | dev_deactivate(dev);
|
| 1085 |
|
| 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 |
|
| 1105 | call_netdevice_notifiers(NETDEV_DOWN, dev);
|
| 1106 |
|
| 1107 | return 0;
|
| 1108 | }
|
| 1109 |
|
| 1110 |
|
| 1111 | static int dev_boot_phase = 1;
|
| 1112 |
|
| 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
|
| 1128 | * to the new notifier to allow device to have a race free
|
| 1129 | * view of the network device list.
|
| 1130 |
|
| 1131 |
|
| 1132 | int register_netdevice_notifier(struct notifier_block *nb)
|
| 1133 | {
|
| 1134 | struct net_device *dev;
|
| 1135 | struct net_device *last;
|
| 1136 | struct net *net;
|
| 1137 | int err;
|
| 1138 |
|
| 1139 | rtnl_lock();
|
| 1140 | err = raw_notifier_chain_register(&netdev_chain, nb);
|
| 1141 | if (err)
|
| 1142 | goto unlock;
|
| 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;
|
| 1154 |
|
| 1155 | nb->notifier_call(nb, NETDEV_UP, dev);
|
| 1156 | }
|
| 1157 | }
|
| 1158 |
|
| 1159 | unlock:
|
| 1160 | rtnl_unlock();
|
| 1161 | return err;
|
| 1162 |
|
| 1163 | rollback:
|
| 1164 | last = dev;
|
| 1165 | for_each_net(net) {
|
| 1166 | for_each_netdev(net, dev) {
|
| 1167 | if (dev == last)
|
| 1168 | break;
|
| 1169 |
|
| 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);
|
| 1175 | }
|
| 1176 | }
|
| 1177 |
|
| 1178 | raw_notifier_chain_unregister(&netdev_chain, nb);
|
| 1179 | goto unlock;
|
| 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 | {
|
| 1194 | int err;
|
| 1195 |
|
| 1196 | rtnl_lock();
|
| 1197 | err = raw_notifier_chain_unregister(&netdev_chain, nb);
|
| 1198 | rtnl_unlock();
|
| 1199 | return err;
|
| 1200 | }
|
| 1201 |
|
| 1202 | |
| 1203 | * call_netdevice_notifiers - call all network notifier blocks
|
| 1204 | * @val: value passed unmodified to notifier function
|
| 1205 | * @dev: net_device pointer passed unmodified to notifier function
|
| 1206 | *
|
| 1207 | * Call all network notifier blocks. Parameters and return value
|
| 1208 | * are as for raw_notifier_call_chain().
|
| 1209 |
|
| 1210 |
|
| 1211 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
|
| 1212 | {
|
| 1213 | return raw_notifier_call_chain(&netdev_chain, val, dev);
|
| 1214 | }
|
| 1215 |
|
| 1216 |
|
| 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 |
|
| 1229 | static inline void net_timestamp(struct sk_buff *skb)
|
| 1230 | {
|
| 1231 | if (atomic_read(&netstamp_needed))
|
| 1232 | __net_timestamp(skb);
|
| 1233 | else
|
| 1234 | skb->tstamp.tv64 = 0;
|
| 1235 | }
|
| 1236 |
|
| 1237 | |
| 1238 | * Support routine. Sends outgoing frames to any network
|
| 1239 | * taps currently in use.
|
| 1240 |
|
| 1241 |
|
| 1242 | static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
|
| 1243 | {
|
| 1244 | struct packet_type *ptype;
|
| 1245 |
|
| 1246 | net_timestamp(skb);
|
| 1247 |
|
| 1248 | rcu_read_lock();
|
| 1249 | list_for_each_entry_rcu(ptype, &ptype_all, list) {
|
| 1250 | |
| 1251 | * they originated from - MvS ([email protected])
|
| 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 | |
| 1261 | set by sender, so that the second statement is
|
| 1262 | just protection against buggy protocols.
|
| 1263 |
|
| 1264 | skb_reset_mac_header(skb2);
|
| 1265 |
|
| 1266 | if (skb_network_header(skb2) < skb2->data ||
|
| 1267 | skb2->network_header > skb2->tail) {
|
| 1268 | if (net_ratelimit())
|
| 1269 | printk(KERN_CRIT "protocol %04x is "
|
| 1270 | "buggy, dev %s\n",
|
| 1271 | skb2->protocol, dev->name);
|
| 1272 | skb_reset_network_header(skb2);
|
| 1273 | }
|
| 1274 |
|
| 1275 | skb2->transport_header = skb2->network_header;
|
| 1276 | skb2->pkt_type = PACKET_OUTGOING;
|
| 1277 | ptype->func(skb2, skb->dev, ptype, skb->dev);
|
| 1278 | }
|
| 1279 | }
|
| 1280 | rcu_read_unlock();
|
| 1281 | }
|
| 1282 |
|
| 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 |
|
| 1300 | void dev_kfree_skb_irq(struct sk_buff *skb)
|
| 1301 | {
|
| 1302 | if (atomic_dec_and_test(&skb->users)) {
|
| 1303 | struct softnet_data *sd;
|
| 1304 | unsigned long flags;
|
| 1305 |
|
| 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 | }
|
| 1313 | }
|
| 1314 | EXPORT_SYMBOL(dev_kfree_skb_irq);
|
| 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 |
|
| 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 |
|
| 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 |
|
| 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 |
|
| 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);
|
| 1352 | __netdev_watchdog_up(dev);
|
| 1353 | }
|
| 1354 | }
|
| 1355 | EXPORT_SYMBOL(netif_device_attach);
|
| 1356 |
|
| 1357 |
|
| 1358 | |
| 1359 | * Invalidate hardware checksum when packet is to be mangled, and
|
| 1360 | * complete checksum manually on outgoing path.
|
| 1361 |
|
| 1362 | int skb_checksum_help(struct sk_buff *skb)
|
| 1363 | {
|
| 1364 | __wsum csum;
|
| 1365 | int ret = 0, offset;
|
| 1366 |
|
| 1367 | if (skb->ip_summed == CHECKSUM_COMPLETE)
|
| 1368 | goto out_set_summed;
|
| 1369 |
|
| 1370 | if (unlikely(skb_shinfo(skb)->gso_size)) {
|
| 1371 |
|
| 1372 | goto out_set_summed;
|
| 1373 | }
|
| 1374 |
|
| 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))) {
|
| 1384 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
|
| 1385 | if (ret)
|
| 1386 | goto out;
|
| 1387 | }
|
| 1388 |
|
| 1389 | *(__sum16 *)(skb->data + offset) = csum_fold(csum);
|
| 1390 | out_set_summed:
|
| 1391 | skb->ip_summed = CHECKSUM_NONE;
|
| 1392 | out:
|
| 1393 | return ret;
|
| 1394 | }
|
| 1395 |
|
| 1396 | |
| 1397 | * skb_gso_segment - Perform segmentation on skb.
|
| 1398 | * @skb: buffer to segment
|
| 1399 | * @features: features for the output path (see dev->features)
|
| 1400 | *
|
| 1401 | * This function segments the given skb and returns a list of segments.
|
| 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.
|
| 1405 |
|
| 1406 | struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
|
| 1407 | {
|
| 1408 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
|
| 1409 | struct packet_type *ptype;
|
| 1410 | __be16 type = skb->protocol;
|
| 1411 | int err;
|
| 1412 |
|
| 1413 | BUG_ON(skb_shinfo(skb)->frag_list);
|
| 1414 |
|
| 1415 | skb_reset_mac_header(skb);
|
| 1416 | skb->mac_len = skb->network_header - skb->mac_header;
|
| 1417 | __skb_pull(skb, skb->mac_len);
|
| 1418 |
|
| 1419 | if (WARN_ON(skb->ip_summed != CHECKSUM_PARTIAL)) {
|
| 1420 | if (skb_header_cloned(skb) &&
|
| 1421 | (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
|
| 1422 | return ERR_PTR(err);
|
| 1423 | }
|
| 1424 |
|
| 1425 | rcu_read_lock();
|
| 1426 | list_for_each_entry_rcu(ptype,
|
| 1427 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
|
| 1428 | if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
|
| 1429 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
|
| 1430 | err = ptype->gso_send_check(skb);
|
| 1431 | segs = ERR_PTR(err);
|
| 1432 | if (err || skb_gso_ok(skb, features))
|
| 1433 | break;
|
| 1434 | __skb_push(skb, (skb->data -
|
| 1435 | skb_network_header(skb)));
|
| 1436 | }
|
| 1437 | segs = ptype->gso_segment(skb, features);
|
| 1438 | break;
|
| 1439 | }
|
| 1440 | }
|
| 1441 | rcu_read_unlock();
|
| 1442 |
|
| 1443 | __skb_push(skb, skb->data - skb_mac_header(skb));
|
| 1444 |
|
| 1445 | return segs;
|
| 1446 | }
|
| 1447 |
|
| 1448 | EXPORT_SYMBOL(skb_gso_segment);
|
| 1449 |
|
| 1450 |
|
| 1451 | #ifdef CONFIG_BUG
|
| 1452 | void netdev_rx_csum_fault(struct net_device *dev)
|
| 1453 | {
|
| 1454 | if (net_ratelimit()) {
|
| 1455 | printk(KERN_ERR "%s: hw csum failure.\n",
|
| 1456 | dev ? dev->name : "<unknown>");
|
| 1457 | dump_stack();
|
| 1458 | }
|
| 1459 | }
|
| 1460 | EXPORT_SYMBOL(netdev_rx_csum_fault);
|
| 1461 | #endif
|
| 1462 |
|
| 1463 | |
| 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 | {
|
| 1470 | #ifdef CONFIG_HIGHMEM
|
| 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 |
|
| 1480 | #endif
|
| 1481 | return 0;
|
| 1482 | }
|
| 1483 |
|
| 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;
|
| 1518 | int features = dev->features & ~(illegal_highdma(dev, skb) ?
|
| 1519 | NETIF_F_SG : 0);
|
| 1520 |
|
| 1521 | segs = skb_gso_segment(skb, features);
|
| 1522 |
|
| 1523 |
|
| 1524 | if (!segs)
|
| 1525 | return 0;
|
| 1526 |
|
| 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)) {
|
| 1540 | if (!list_empty(&ptype_all))
|
| 1541 | dev_queue_xmit_nit(skb, dev);
|
| 1542 |
|
| 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 | }
|
| 1549 |
|
| 1550 | return dev->hard_start_xmit(skb, dev);
|
| 1551 | }
|
| 1552 |
|
| 1553 | gso:
|
| 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)) {
|
| 1562 | nskb->next = skb->next;
|
| 1563 | skb->next = nskb;
|
| 1564 | return rc;
|
| 1565 | }
|
| 1566 | if (unlikely((netif_queue_stopped(dev) ||
|
| 1567 | netif_subqueue_stopped(dev, skb)) &&
|
| 1568 | skb->next))
|
| 1569 | return NETDEV_TX_BUSY;
|
| 1570 | } while (skb->next);
|
| 1571 |
|
| 1572 | skb->destructor = DEV_GSO_CB(skb)->destructor;
|
| 1573 |
|
| 1574 | out_kfree_skb:
|
| 1575 | kfree_skb(skb);
|
| 1576 | return 0;
|
| 1577 | }
|
| 1578 |
|
| 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.
|
| 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
|
| 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 |
|
| 1611 |
|
| 1612 | if (netif_needs_gso(dev, skb))
|
| 1613 | goto gso;
|
| 1614 |
|
| 1615 | if (skb_shinfo(skb)->frag_list &&
|
| 1616 | !(dev->features & NETIF_F_FRAGLIST) &&
|
| 1617 | __skb_linearize(skb))
|
| 1618 | goto out_kfree_skb;
|
| 1619 |
|
| 1620 | |
| 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)) &&
|
| 1626 | __skb_linearize(skb))
|
| 1627 | goto out_kfree_skb;
|
| 1628 |
|
| 1629 | |
| 1630 | * checksumming for this protocol, complete checksumming here.
|
| 1631 |
|
| 1632 | if (skb->ip_summed == CHECKSUM_PARTIAL) {
|
| 1633 | skb_set_transport_header(skb, skb->csum_start -
|
| 1634 | skb_headroom(skb));
|
| 1635 |
|
| 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)))
|
| 1641 | if (skb_checksum_help(skb))
|
| 1642 | goto out_kfree_skb;
|
| 1643 | }
|
| 1644 |
|
| 1645 | gso:
|
| 1646 | spin_lock_prefetch(&dev->queue_lock);
|
| 1647 |
|
| 1648 | |
| 1649 | * stops preemption for RCU.
|
| 1650 |
|
| 1651 | rcu_read_lock_bh();
|
| 1652 |
|
| 1653 | |
| 1654 | * The struct Qdisc which is pointed to by qdisc is now a
|
| 1655 | * rcu structure - it may be accessed without acquiring
|
| 1656 | * a lock (but the structure may be stale.) The freeing of the
|
| 1657 | * qdisc will be deferred until it's known that there are no
|
| 1658 | * more references to it.
|
| 1659 | *
|
| 1660 | * If the qdisc has an enqueue function, we still need to
|
| 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 |
|
| 1671 | spin_lock(&dev->queue_lock);
|
| 1672 | q = dev->qdisc;
|
| 1673 | if (q->enqueue) {
|
| 1674 |
|
| 1675 | skb_set_queue_mapping(skb, 0);
|
| 1676 | rc = q->enqueue(skb, q);
|
| 1677 | qdisc_run(dev);
|
| 1678 | spin_unlock(&dev->queue_lock);
|
| 1679 |
|
| 1680 | rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
|
| 1681 | goto out;
|
| 1682 | }
|
| 1683 | spin_unlock(&dev->queue_lock);
|
| 1684 | }
|
| 1685 |
|
| 1686 | |
| 1687 | loopback, all the sorts of tunnels...
|
| 1688 |
|
| 1689 | Really, it is unlikely that netif_tx_lock protection is necessary
|
| 1690 | here. (f.e. loopback and IP tunnels are clean ignoring statistics
|
| 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();
|
| 1700 |
|
| 1701 | if (dev->xmit_lock_owner != cpu) {
|
| 1702 |
|
| 1703 | HARD_TX_LOCK(dev, cpu);
|
| 1704 |
|
| 1705 | if (!netif_queue_stopped(dev) &&
|
| 1706 | !netif_subqueue_stopped(dev, skb)) {
|
| 1707 | rc = 0;
|
| 1708 | if (!dev_hard_start_xmit(skb, dev)) {
|
| 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 | |
| 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;
|
| 1727 | rcu_read_unlock_bh();
|
| 1728 |
|
| 1729 | out_kfree_skb:
|
| 1730 | kfree_skb(skb);
|
| 1731 | return rc;
|
| 1732 | out:
|
| 1733 | rcu_read_unlock_bh();
|
| 1734 | return rc;
|
| 1735 | }
|
| 1736 |
|
| 1737 |
|
| 1738 | |
| 1739 | Receiver routines
|
| 1740 | =======================================================================
|
| 1741 |
|
| 1742 | int netdev_max_backlog __read_mostly = 1000;
|
| 1743 | int netdev_budget __read_mostly = 300;
|
| 1744 | int weight_p __read_mostly = 64;
|
| 1745 |
|
| 1746 | DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
|
| 1747 |
|
| 1748 |
|
| 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)
|
| 1760 | * NET_RX_DROP (packet was dropped)
|
| 1761 | *
|
| 1762 |
|
| 1763 |
|
| 1764 | int netif_rx(struct sk_buff *skb)
|
| 1765 | {
|
| 1766 | struct softnet_data *queue;
|
| 1767 | unsigned long flags;
|
| 1768 |
|
| 1769 |
|
| 1770 | if (netpoll_rx(skb))
|
| 1771 | return NET_RX_DROP;
|
| 1772 |
|
| 1773 | if (!skb->tstamp.tv64)
|
| 1774 | net_timestamp(skb);
|
| 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);
|
| 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) {
|
| 1786 | enqueue:
|
| 1787 | dev_hold(skb->dev);
|
| 1788 | __skb_queue_tail(&queue->input_pkt_queue, skb);
|
| 1789 | local_irq_restore(flags);
|
| 1790 | return NET_RX_SUCCESS;
|
| 1791 | }
|
| 1792 |
|
| 1793 | napi_schedule(&queue->backlog);
|
| 1794 | goto enqueue;
|
| 1795 | }
|
| 1796 |
|
| 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 |
|
| 1819 | static inline struct net_device *skb_bond(struct sk_buff *skb)
|
| 1820 | {
|
| 1821 | struct net_device *dev = skb->dev;
|
| 1822 |
|
| 1823 | if (dev->master) {
|
| 1824 | if (skb_bond_should_drop(skb)) {
|
| 1825 | kfree_skb(skb);
|
| 1826 | return NULL;
|
| 1827 | }
|
| 1828 | skb->dev = dev->master;
|
| 1829 | }
|
| 1830 |
|
| 1831 | return dev;
|
| 1832 | }
|
| 1833 |
|
| 1834 |
|
| 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 |
|
| 1881 | static inline int deliver_skb(struct sk_buff *skb,
|
| 1882 | struct packet_type *pt_prev,
|
| 1883 | struct net_device *orig_dev)
|
| 1884 | {
|
| 1885 | atomic_inc(&skb->users);
|
| 1886 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
|
| 1887 | }
|
| 1888 |
|
| 1889 | #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
|
| 1890 |
|
| 1891 | struct net_bridge;
|
| 1892 | struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
|
| 1893 | unsigned char *addr);
|
| 1894 | void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent) __read_mostly;
|
| 1895 |
|
| 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)
|
| 1905 | {
|
| 1906 | struct net_bridge_port *port;
|
| 1907 |
|
| 1908 | if (skb->pkt_type == PACKET_LOOPBACK ||
|
| 1909 | (port = rcu_dereference(skb->dev->br_port)) == NULL)
|
| 1910 | return skb;
|
| 1911 |
|
| 1912 | if (*pt_prev) {
|
| 1913 | *ret = deliver_skb(skb, *pt_prev, orig_dev);
|
| 1914 | *pt_prev = NULL;
|
| 1915 | }
|
| 1916 |
|
| 1917 | return br_handle_frame_hook(port, skb);
|
| 1918 | }
|
| 1919 | #else
|
| 1920 | #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
|
| 1921 | #endif
|
| 1922 |
|
| 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 |
|
| 1945 | #ifdef CONFIG_NET_CLS_ACT
|
| 1946 | |
| 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
|
| 1950 | * NOTE: This doesnt stop any functionality; if you dont have
|
| 1951 | * the ingress scheduler, you just cant add policies on ingress.
|
| 1952 | *
|
| 1953 |
|
| 1954 | static int ing_filter(struct sk_buff *skb)
|
| 1955 | {
|
| 1956 | struct Qdisc *q;
|
| 1957 | struct net_device *dev = skb->dev;
|
| 1958 | int result = TC_ACT_OK;
|
| 1959 | u32 ttl = G_TC_RTTL(skb->tc_verd);
|
| 1960 |
|
| 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 | }
|
| 1967 |
|
| 1968 | skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
|
| 1969 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
|
| 1970 |
|
| 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 | }
|
| 1978 |
|
| 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;
|
| 1985 |
|
| 1986 | if (*pt_prev) {
|
| 1987 | *ret = deliver_skb(skb, *pt_prev, orig_dev);
|
| 1988 | *pt_prev = NULL;
|
| 1989 | } else {
|
| 1990 |
|
| 1991 | skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
|
| 1992 | }
|
| 1993 |
|
| 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;
|
| 2004 | }
|
| 2005 | #endif
|
| 2006 |
|
| 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 |
|
| 2022 | int netif_receive_skb(struct sk_buff *skb)
|
| 2023 | {
|
| 2024 | struct packet_type *ptype, *pt_prev;
|
| 2025 | struct net_device *orig_dev;
|
| 2026 | int ret = NET_RX_DROP;
|
| 2027 | __be16 type;
|
| 2028 |
|
| 2029 |
|
| 2030 | if (netpoll_receive_skb(skb))
|
| 2031 | return NET_RX_DROP;
|
| 2032 |
|
| 2033 | if (!skb->tstamp.tv64)
|
| 2034 | net_timestamp(skb);
|
| 2035 |
|
| 2036 | if (!skb->iif)
|
| 2037 | skb->iif = skb->dev->ifindex;
|
| 2038 |
|
| 2039 | orig_dev = skb_bond(skb);
|
| 2040 |
|
| 2041 | if (!orig_dev)
|
| 2042 | return NET_RX_DROP;
|
| 2043 |
|
| 2044 | __get_cpu_var(netdev_rx_stat).total++;
|
| 2045 |
|
| 2046 | skb_reset_network_header(skb);
|
| 2047 | skb_reset_transport_header(skb);
|
| 2048 | skb->mac_len = skb->network_header - skb->mac_header;
|
| 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) {
|
| 2063 | if (pt_prev)
|
| 2064 | ret = deliver_skb(skb, pt_prev, orig_dev);
|
| 2065 | pt_prev = ptype;
|
| 2066 | }
|
| 2067 | }
|
| 2068 |
|
| 2069 | #ifdef CONFIG_NET_CLS_ACT
|
| 2070 | skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
|
| 2071 | if (!skb)
|
| 2072 | goto out;
|
| 2073 | ncls:
|
| 2074 | #endif
|
| 2075 |
|
| 2076 | skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
|
| 2077 | if (!skb)
|
| 2078 | goto out;
|
| 2079 | skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
|
| 2080 | if (!skb)
|
| 2081 | goto out;
|
| 2082 |
|
| 2083 | type = skb->protocol;
|
| 2084 | list_for_each_entry_rcu(ptype,
|
| 2085 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
|
| 2086 | if (ptype->type == type &&
|
| 2087 | (!ptype->dev || ptype->dev == skb->dev)) {
|
| 2088 | if (pt_prev)
|
| 2089 | ret = deliver_skb(skb, pt_prev, orig_dev);
|
| 2090 | pt_prev = ptype;
|
| 2091 | }
|
| 2092 | }
|
| 2093 |
|
| 2094 | if (pt_prev) {
|
| 2095 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
|
| 2096 | } else {
|
| 2097 | kfree_skb(skb);
|
| 2098 | |
| 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 |
|
| 2109 | static int process_backlog(struct napi_struct *napi, int quota)
|
| 2110 | {
|
| 2111 | int work = 0;
|
| 2112 | struct softnet_data *queue = &__get_cpu_var(softnet_data);
|
| 2113 | unsigned long start_time = jiffies;
|
| 2114 |
|
| 2115 | napi->weight = weight_p;
|
| 2116 | do {
|
| 2117 | struct sk_buff *skb;
|
| 2118 | struct net_device *dev;
|
| 2119 |
|
| 2120 | local_irq_disable();
|
| 2121 | skb = __skb_dequeue(&queue->input_pkt_queue);
|
| 2122 | if (!skb) {
|
| 2123 | __napi_complete(napi);
|
| 2124 | local_irq_enable();
|
| 2125 | break;
|
| 2126 | }
|
| 2127 |
|
| 2128 | local_irq_enable();
|
| 2129 |
|
| 2130 | dev = skb->dev;
|
| 2131 |
|
| 2132 | netif_receive_skb(skb);
|
| 2133 |
|
| 2134 | dev_put(dev);
|
| 2135 | } while (++work < quota && jiffies == start_time);
|
| 2136 |
|
| 2137 | return work;
|
| 2138 | }
|
| 2139 |
|
| 2140 | |
| 2141 | * __napi_schedule - schedule for receive
|
| 2142 | * @n: entry to schedule
|
| 2143 | *
|
| 2144 | * The entry's receive function will be scheduled to run
|
| 2145 |
|
| 2146 | void __napi_schedule(struct napi_struct *n)
|
| 2147 | {
|
| 2148 | unsigned long flags;
|
| 2149 |
|
| 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);
|
| 2154 | }
|
| 2155 | EXPORT_SYMBOL(__napi_schedule);
|
| 2156 |
|
| 2157 |
|
| 2158 | static void net_rx_action(struct softirq_action *h)
|
| 2159 | {
|
| 2160 | struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
|
| 2161 | unsigned long start_time = jiffies;
|
| 2162 | int budget = netdev_budget;
|
| 2163 | void *have;
|
| 2164 |
|
| 2165 | local_irq_disable();
|
| 2166 |
|
| 2167 | while (!list_empty(list)) {
|
| 2168 | struct napi_struct *n;
|
| 2169 | int work, weight;
|
| 2170 |
|
| 2171 | |
| 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))
|
| 2179 | goto softnet_break;
|
| 2180 |
|
| 2181 | local_irq_enable();
|
| 2182 |
|
| 2183 | |
| 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);
|
| 2189 |
|
| 2190 | have = netpoll_poll_lock(n);
|
| 2191 |
|
| 2192 | weight = n->weight;
|
| 2193 |
|
| 2194 | |
| 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);
|
| 2203 |
|
| 2204 | WARN_ON_ONCE(work > weight);
|
| 2205 |
|
| 2206 | budget -= work;
|
| 2207 |
|
| 2208 | local_irq_disable();
|
| 2209 |
|
| 2210 | |
| 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 |
|
| 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 | }
|
| 2221 |
|
| 2222 | netpoll_poll_unlock(have);
|
| 2223 | }
|
| 2224 | out:
|
| 2225 | local_irq_enable();
|
| 2226 |
|
| 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 |
|
| 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 | }
|
| 2239 | }
|
| 2240 | #endif
|
| 2241 |
|
| 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 |
|
| 2281 | static int dev_ifname(struct net *net, struct ifreq __user *arg)
|
| 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);
|
| 2294 | dev = __dev_get_by_index(net, ifr.ifr_ifindex);
|
| 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 |
|
| 2314 | static int dev_ifconf(struct net *net, char __user *arg)
|
| 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;
|
| 2338 | for_each_netdev(net, dev) {
|
| 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 | }
|
| 2352 | }
|
| 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 |
|
| 2370 | void *dev_seq_start(struct seq_file *seq, loff_t *pos)
|
| 2371 | __acquires(dev_base_lock)
|
| 2372 | {
|
| 2373 | struct net *net = seq_file_net(seq);
|
| 2374 | loff_t off;
|
| 2375 | struct net_device *dev;
|
| 2376 |
|
| 2377 | read_lock(&dev_base_lock);
|
| 2378 | if (!*pos)
|
| 2379 | return SEQ_START_TOKEN;
|
| 2380 |
|
| 2381 | off = 1;
|
| 2382 | for_each_netdev(net, dev)
|
| 2383 | if (off++ == *pos)
|
| 2384 | return dev;
|
| 2385 |
|
| 2386 | return NULL;
|
| 2387 | }
|
| 2388 |
|
| 2389 | void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
|
| 2390 | {
|
| 2391 | struct net *net = seq_file_net(seq);
|
| 2392 | ++*pos;
|
| 2393 | return v == SEQ_START_TOKEN ?
|
| 2394 | first_net_device(net) : next_net_device((struct net_device *)v);
|
| 2395 | }
|
| 2396 |
|
| 2397 | void dev_seq_stop(struct seq_file *seq, void *v)
|
| 2398 | __releases(dev_base_lock)
|
| 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 | {
|
| 2405 | struct net_device_stats *stats = dev->get_stats(dev);
|
| 2406 |
|
| 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);
|
| 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)
|
| 2448 | if (cpu_online(*pos)) {
|
| 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",
|
| 2476 | s->total, s->dropped, s->time_squeeze, 0,
|
| 2477 | 0, 0, 0, 0,
|
| 2478 | s->cpu_collision );
|
| 2479 | return 0;
|
| 2480 | }
|
| 2481 |
|
| 2482 | static const struct seq_operations dev_seq_ops = {
|
| 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 | {
|
| 2491 | return seq_open_net(inode, file, &dev_seq_ops,
|
| 2492 | sizeof(struct seq_net_private));
|
| 2493 | }
|
| 2494 |
|
| 2495 | static const struct file_operations dev_seq_fops = {
|
| 2496 | .owner = THIS_MODULE,
|
| 2497 | .open = dev_seq_open,
|
| 2498 | .read = seq_read,
|
| 2499 | .llseek = seq_lseek,
|
| 2500 | .release = seq_release_net,
|
| 2501 | };
|
| 2502 |
|
| 2503 | static const struct seq_operations softnet_seq_ops = {
|
| 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 |
|
| 2515 | static const struct file_operations softnet_seq_fops = {
|
| 2516 | .owner = THIS_MODULE,
|
| 2517 | .open = softnet_seq_open,
|
| 2518 | .read = seq_read,
|
| 2519 | .llseek = seq_lseek,
|
| 2520 | .release = seq_release,
|
| 2521 | };
|
| 2522 |
|
| 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 |
|
| 2535 | for (t = 0; t < PTYPE_HASH_SIZE; t++) {
|
| 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)
|
| 2546 | __acquires(RCU)
|
| 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
|
| 2570 | hash = ntohs(pt->type) & PTYPE_HASH_MASK;
|
| 2571 |
|
| 2572 | while (nxt == &ptype_base[hash]) {
|
| 2573 | if (++hash >= PTYPE_HASH_SIZE)
|
| 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)
|
| 2582 | __releases(RCU)
|
| 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 if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
|
| 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_net(inode, file, &ptype_seq_ops,
|
| 2643 | sizeof(struct seq_net_private));
|
| 2644 | }
|
| 2645 |
|
| 2646 | static const struct file_operations ptype_seq_fops = {
|
| 2647 | .owner = THIS_MODULE,
|
| 2648 | .open = ptype_seq_open,
|
| 2649 | .read = seq_read,
|
| 2650 | .llseek = seq_lseek,
|
| 2651 | .release = seq_release_net,
|
| 2652 | };
|
| 2653 |
|
| 2654 |
|
| 2655 | static int __net_init dev_proc_net_init(struct net *net)
|
| 2656 | {
|
| 2657 | int rc = -ENOMEM;
|
| 2658 |
|
| 2659 | if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
|
| 2660 | goto out;
|
| 2661 | if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
|
| 2662 | goto out_dev;
|
| 2663 | if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
|
| 2664 | goto out_softnet;
|
| 2665 |
|
| 2666 | if (wext_proc_init(net))
|
| 2667 | goto out_ptype;
|
| 2668 | rc = 0;
|
| 2669 | out:
|
| 2670 | return rc;
|
| 2671 | out_ptype:
|
| 2672 | proc_net_remove(net, "ptype");
|
| 2673 | out_softnet:
|
| 2674 | proc_net_remove(net, "softnet_stat");
|
| 2675 | out_dev:
|
| 2676 | proc_net_remove(net, "dev");
|
| 2677 | goto out;
|
| 2678 | }
|
| 2679 |
|
| 2680 | static void __net_exit dev_proc_net_exit(struct net *net)
|
| 2681 | {
|
| 2682 | wext_proc_exit(net);
|
| 2683 |
|
| 2684 | proc_net_remove(net, "ptype");
|
| 2685 | proc_net_remove(net, "softnet_stat");
|
| 2686 | proc_net_remove(net, "dev");
|
| 2687 | }
|
| 2688 |
|
| 2689 | static struct pernet_operations __net_initdata dev_proc_ops = {
|
| 2690 | .init = dev_proc_net_init,
|
| 2691 | .exit = dev_proc_net_exit,
|
| 2692 | };
|
| 2693 |
|
| 2694 | static int __init dev_proc_init(void)
|
| 2695 | {
|
| 2696 | return register_pernet_subsys(&dev_proc_ops);
|
| 2697 | }
|
| 2698 | #else
|
| 2699 | #define dev_proc_init() 0
|
| 2700 | #endif
|
| 2701 |
|
| 2702 |
|
| 2703 | |
| 2704 | * netdev_set_master - set up master/slave pair
|
| 2705 | * @slave: slave device
|
| 2706 | * @master: new master device
|
| 2707 | *
|
| 2708 | * Changes the master device of the slave. Pass %NULL to break the
|
| 2709 | * bonding. The caller must hold the RTNL semaphore. On a failure
|
| 2710 | * a negative errno code is returned. On success the reference counts
|
| 2711 | * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
|
| 2712 | * function returns zero.
|
| 2713 |
|
| 2714 | int netdev_set_master(struct net_device *slave, struct net_device *master)
|
| 2715 | {
|
| 2716 | struct net_device *old = slave->master;
|
| 2717 |
|
| 2718 | ASSERT_RTNL();
|
| 2719 |
|
| 2720 | if (master) {
|
| 2721 | if (old)
|
| 2722 | return -EBUSY;
|
| 2723 | dev_hold(master);
|
| 2724 | }
|
| 2725 |
|
| 2726 | slave->master = master;
|
| 2727 |
|
| 2728 | synchronize_net();
|
| 2729 |
|
| 2730 | if (old)
|
| 2731 | dev_put(old);
|
| 2732 |
|
| 2733 | if (master)
|
| 2734 | slave->flags |= IFF_SLAVE;
|
| 2735 | else
|
| 2736 | slave->flags &= ~IFF_SLAVE;
|
| 2737 |
|
| 2738 | rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
|
| 2739 | return 0;
|
| 2740 | }
|
| 2741 |
|
| 2742 | static void __dev_set_promiscuity(struct net_device *dev, int inc)
|
| 2743 | {
|
| 2744 | unsigned short old_flags = dev->flags;
|
| 2745 |
|
| 2746 | ASSERT_RTNL();
|
| 2747 |
|
| 2748 | if ((dev->promiscuity += inc) == 0)
|
| 2749 | dev->flags &= ~IFF_PROMISC;
|
| 2750 | else
|
| 2751 | dev->flags |= IFF_PROMISC;
|
| 2752 | if (dev->flags != old_flags) {
|
| 2753 | printk(KERN_INFO "device %s %s promiscuous mode\n",
|
| 2754 | dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
|
| 2755 | "left");
|
| 2756 | if (audit_enabled)
|
| 2757 | audit_log(current->audit_context, GFP_ATOMIC,
|
| 2758 | AUDIT_ANOM_PROMISCUOUS,
|
| 2759 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
|
| 2760 | dev->name, (dev->flags & IFF_PROMISC),
|
| 2761 | (old_flags & IFF_PROMISC),
|
| 2762 | audit_get_loginuid(current),
|
| 2763 | current->uid, current->gid,
|
| 2764 | audit_get_sessionid(current));
|
| 2765 |
|
| 2766 | if (dev->change_rx_flags)
|
| 2767 | dev->change_rx_flags(dev, IFF_PROMISC);
|
| 2768 | }
|
| 2769 | }
|
| 2770 |
|
| 2771 | |
| 2772 | * dev_set_promiscuity - update promiscuity count on a device
|
| 2773 | * @dev: device
|
| 2774 | * @inc: modifier
|
| 2775 | *
|
| 2776 | * Add or remove promiscuity from a device. While the count in the device
|
| 2777 | * remains above zero the interface remains promiscuous. Once it hits zero
|
| 2778 | * the device reverts back to normal filtering operation. A negative inc
|
| 2779 | * value is used to drop promiscuity on the device.
|
| 2780 |
|
| 2781 | void dev_set_promiscuity(struct net_device *dev, int inc)
|
| 2782 | {
|
| 2783 | unsigned short old_flags = dev->flags;
|
| 2784 |
|
| 2785 | __dev_set_promiscuity(dev, inc);
|
| 2786 | if (dev->flags != old_flags)
|
| 2787 | dev_set_rx_mode(dev);
|
| 2788 | }
|
| 2789 |
|
| 2790 | |
| 2791 | * dev_set_allmulti - update allmulti count on a device
|
| 2792 | * @dev: device
|
| 2793 | * @inc: modifier
|
| 2794 | *
|
| 2795 | * Add or remove reception of all multicast frames to a device. While the
|
| 2796 | * count in the device remains above zero the interface remains listening
|
| 2797 | * to all interfaces. Once it hits zero the device reverts back to normal
|
| 2798 | * filtering operation. A negative @inc value is used to drop the counter
|
| 2799 | * when releasing a resource needing all multicasts.
|
| 2800 |
|
| 2801 |
|
| 2802 | void dev_set_allmulti(struct net_device *dev, int inc)
|
| 2803 | {
|
| 2804 | unsigned short old_flags = dev->flags;
|
| 2805 |
|
| 2806 | ASSERT_RTNL();
|
| 2807 |
|
| 2808 | dev->flags |= IFF_ALLMULTI;
|
| 2809 | if ((dev->allmulti += inc) == 0)
|
| 2810 | dev->flags &= ~IFF_ALLMULTI;
|
| 2811 | if (dev->flags ^ old_flags) {
|
| 2812 | if (dev->change_rx_flags)
|
| 2813 | dev->change_rx_flags(dev, IFF_ALLMULTI);
|
| 2814 | dev_set_rx_mode(dev);
|
| 2815 | }
|
| 2816 | }
|
| 2817 |
|
| 2818 | |
| 2819 | * Upload unicast and multicast address lists to device and
|
| 2820 | * configure RX filtering. When the device doesn't support unicast
|
| 2821 | * filtering it is put in promiscuous mode while unicast addresses
|
| 2822 | * are present.
|
| 2823 |
|
| 2824 | void __dev_set_rx_mode(struct net_device *dev)
|
| 2825 | {
|
| 2826 |
|
| 2827 | if (!(dev->flags&IFF_UP))
|
| 2828 | return;
|
| 2829 |
|
| 2830 | if (!netif_device_present(dev))
|
| 2831 | return;
|
| 2832 |
|
| 2833 | if (dev->set_rx_mode)
|
| 2834 | dev->set_rx_mode(dev);
|
| 2835 | else {
|
| 2836 | |
| 2837 | * therefore calling __dev_set_promiscuity here is safe.
|
| 2838 |
|
| 2839 | if (dev->uc_count > 0 && !dev->uc_promisc) {
|
| 2840 | __dev_set_promiscuity(dev, 1);
|
| 2841 | dev->uc_promisc = 1;
|
| 2842 | } else if (dev->uc_count == 0 && dev->uc_promisc) {
|
| 2843 | __dev_set_promiscuity(dev, -1);
|
| 2844 | dev->uc_promisc = 0;
|
| 2845 | }
|
| 2846 |
|
| 2847 | if (dev->set_multicast_list)
|
| 2848 | dev->set_multicast_list(dev);
|
| 2849 | }
|
| 2850 | }
|
| 2851 |
|
| 2852 | void dev_set_rx_mode(struct net_device *dev)
|
| 2853 | {
|
| 2854 | netif_tx_lock_bh(dev);
|
| 2855 | __dev_set_rx_mode(dev);
|
| 2856 | netif_tx_unlock_bh(dev);
|
| 2857 | }
|
| 2858 |
|
| 2859 | int __dev_addr_delete(struct dev_addr_list **list, int *count,
|
| 2860 | void *addr, int alen, int glbl)
|
| 2861 | {
|
| 2862 | struct dev_addr_list *da;
|
| 2863 |
|
| 2864 | for (; (da = *list) != NULL; list = &da->next) {
|
| 2865 | if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
|
| 2866 | alen == da->da_addrlen) {
|
| 2867 | if (glbl) {
|
| 2868 | int old_glbl = da->da_gusers;
|
| 2869 | da->da_gusers = 0;
|
| 2870 | if (old_glbl == 0)
|
| 2871 | break;
|
| 2872 | }
|
| 2873 | if (--da->da_users)
|
| 2874 | return 0;
|
| 2875 |
|
| 2876 | *list = da->next;
|
| 2877 | kfree(da);
|
| 2878 | (*count)--;
|
| 2879 | return 0;
|
| 2880 | }
|
| 2881 | }
|
| 2882 | return -ENOENT;
|
| 2883 | }
|
| 2884 |
|
| 2885 | int __dev_addr_add(struct dev_addr_list **list, int *count,
|
| 2886 | void *addr, int alen, int glbl)
|
| 2887 | {
|
| 2888 | struct dev_addr_list *da;
|
| 2889 |
|
| 2890 | for (da = *list; da != NULL; da = da->next) {
|
| 2891 | if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
|
| 2892 | da->da_addrlen == alen) {
|
| 2893 | if (glbl) {
|
| 2894 | int old_glbl = da->da_gusers;
|
| 2895 | da->da_gusers = 1;
|
| 2896 | if (old_glbl)
|
| 2897 | return 0;
|
| 2898 | }
|
| 2899 | da->da_users++;
|
| 2900 | return 0;
|
| 2901 | }
|
| 2902 | }
|
| 2903 |
|
| 2904 | da = kzalloc(sizeof(*da), GFP_ATOMIC);
|
| 2905 | if (da == NULL)
|
| 2906 | return -ENOMEM;
|
| 2907 | memcpy(da->da_addr, addr, alen);
|
| 2908 | da->da_addrlen = alen;
|
| 2909 | da->da_users = 1;
|
| 2910 | da->da_gusers = glbl ? 1 : 0;
|
| 2911 | da->next = *list;
|
| 2912 | *list = da;
|
| 2913 | (*count)++;
|
| 2914 | return 0;
|
| 2915 | }
|
| 2916 |
|
| 2917 | |
| 2918 | * dev_unicast_delete - Release secondary unicast address.
|
| 2919 | * @dev: device
|
| 2920 | * @addr: address to delete
|
| 2921 | * @alen: length of @addr
|
| 2922 | *
|
| 2923 | * Release reference to a secondary unicast address and remove it
|
| 2924 | * from the device if the reference count drops to zero.
|
| 2925 | *
|
| 2926 | * The caller must hold the rtnl_mutex.
|
| 2927 |
|
| 2928 | int dev_unicast_delete(struct net_device *dev, void *addr, int alen)
|
| 2929 | {
|
| 2930 | int err;
|
| 2931 |
|
| 2932 | ASSERT_RTNL();
|
| 2933 |
|
| 2934 | netif_tx_lock_bh(dev);
|
| 2935 | err = __dev_addr_delete(&dev->uc_list, &dev->uc_count, addr, alen, 0);
|
| 2936 | if (!err)
|
| 2937 | __dev_set_rx_mode(dev);
|
| 2938 | netif_tx_unlock_bh(dev);
|
| 2939 | return err;
|
| 2940 | }
|
| 2941 | EXPORT_SYMBOL(dev_unicast_delete);
|
| 2942 |
|
| 2943 | |
| 2944 | * dev_unicast_add - add a secondary unicast address
|
| 2945 | * @dev: device
|
| 2946 | * @addr: address to delete
|
| 2947 | * @alen: length of @addr
|
| 2948 | *
|
| 2949 | * Add a secondary unicast address to the device or increase
|
| 2950 | * the reference count if it already exists.
|
| 2951 | *
|
| 2952 | * The caller must hold the rtnl_mutex.
|
| 2953 |
|
| 2954 | int dev_unicast_add(struct net_device *dev, void *addr, int alen)
|
| 2955 | {
|
| 2956 | int err;
|
| 2957 |
|
| 2958 | ASSERT_RTNL();
|
| 2959 |
|
| 2960 | netif_tx_lock_bh(dev);
|
| 2961 | err = __dev_addr_add(&dev->uc_list, &dev->uc_count, addr, alen, 0);
|
| 2962 | if (!err)
|
| 2963 | __dev_set_rx_mode(dev);
|
| 2964 | netif_tx_unlock_bh(dev);
|
| 2965 | return err;
|
| 2966 | }
|
| 2967 | EXPORT_SYMBOL(dev_unicast_add);
|
| 2968 |
|
| 2969 | int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
|
| 2970 | struct dev_addr_list **from, int *from_count)
|
| 2971 | {
|
| 2972 | struct dev_addr_list *da, *next;
|
| 2973 | int err = 0;
|
| 2974 |
|
| 2975 | da = *from;
|
| 2976 | while (da != NULL) {
|
| 2977 | next = da->next;
|
| 2978 | if (!da->da_synced) {
|
| 2979 | err = __dev_addr_add(to, to_count,
|
| 2980 | da->da_addr, da->da_addrlen, 0);
|
| 2981 | if (err < 0)
|
| 2982 | break;
|
| 2983 | da->da_synced = 1;
|
| 2984 | da->da_users++;
|
| 2985 | } else if (da->da_users == 1) {
|
| 2986 | __dev_addr_delete(to, to_count,
|
| 2987 | da->da_addr, da->da_addrlen, 0);
|
| 2988 | __dev_addr_delete(from, from_count,
|
| 2989 | da->da_addr, da->da_addrlen, 0);
|
| 2990 | }
|
| 2991 | da = next;
|
| 2992 | }
|
| 2993 | return err;
|
| 2994 | }
|
| 2995 |
|
| 2996 | void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
|
| 2997 | struct dev_addr_list **from, int *from_count)
|
| 2998 | {
|
| 2999 | struct dev_addr_list *da, *next;
|
| 3000 |
|
| 3001 | da = *from;
|
| 3002 | while (da != NULL) {
|
| 3003 | next = da->next;
|
| 3004 | if (da->da_synced) {
|
| 3005 | __dev_addr_delete(to, to_count,
|
| 3006 | da->da_addr, da->da_addrlen, 0);
|
| 3007 | da->da_synced = 0;
|
| 3008 | __dev_addr_delete(from, from_count,
|
| 3009 | da->da_addr, da->da_addrlen, 0);
|
| 3010 | }
|
| 3011 | da = next;
|
| 3012 | }
|
| 3013 | }
|
| 3014 |
|
| 3015 | |
| 3016 | * dev_unicast_sync - Synchronize device's unicast list to another device
|
| 3017 | * @to: destination device
|
| 3018 | * @from: source device
|
| 3019 | *
|
| 3020 | * Add newly added addresses to the destination device and release
|
| 3021 | * addresses that have no users left. The source device must be
|
| 3022 | * locked by netif_tx_lock_bh.
|
| 3023 | *
|
| 3024 | * This function is intended to be called from the dev->set_rx_mode
|
| 3025 | * function of layered software devices.
|
| 3026 |
|
| 3027 | int dev_unicast_sync(struct net_device *to, struct net_device *from)
|
| 3028 | {
|
| 3029 | int err = 0;
|
| 3030 |
|
| 3031 | netif_tx_lock_bh(to);
|
| 3032 | err = __dev_addr_sync(&to->uc_list, &to->uc_count,
|
| 3033 | &from->uc_list, &from->uc_count);
|
| 3034 | if (!err)
|
| 3035 | __dev_set_rx_mode(to);
|
| 3036 | netif_tx_unlock_bh(to);
|
| 3037 | return err;
|
| 3038 | }
|
| 3039 | EXPORT_SYMBOL(dev_unicast_sync);
|
| 3040 |
|
| 3041 | |
| 3042 | * dev_unicast_unsync - Remove synchronized addresses from the destination 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 |
|
| 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 |
|
| 3078 | static void dev_addr_discard(struct net_device *dev)
|
| 3079 | {
|
| 3080 | netif_tx_lock_bh(dev);
|
| 3081 |
|
| 3082 | __dev_addr_discard(&dev->uc_list);
|
| 3083 | dev->uc_count = 0;
|
| 3084 |
|
| 3085 | __dev_addr_discard(&dev->mc_list);
|
| 3086 | dev->mc_count = 0;
|
| 3087 |
|
| 3088 | netif_tx_unlock_bh(dev);
|
| 3089 | }
|
| 3090 |
|
| 3091 | unsigned dev_get_flags(const struct net_device *dev)
|
| 3092 | {
|
| 3093 | unsigned flags;
|
| 3094 |
|
| 3095 | flags = (dev->flags & ~(IFF_PROMISC |
|
| 3096 | IFF_ALLMULTI |
|
| 3097 | IFF_RUNNING |
|
| 3098 | IFF_LOWER_UP |
|
| 3099 | IFF_DORMANT)) |
|
| 3100 | (dev->gflags & (IFF_PROMISC |
|
| 3101 | IFF_ALLMULTI));
|
| 3102 |
|
| 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 | }
|
| 3111 |
|
| 3112 | return flags;
|
| 3113 | }
|
| 3114 |
|
| 3115 | int dev_change_flags(struct net_device *dev, unsigned flags)
|
| 3116 | {
|
| 3117 | int ret, changes;
|
| 3118 | int old_flags = dev->flags;
|
| 3119 |
|
| 3120 | ASSERT_RTNL();
|
| 3121 |
|
| 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 |
|
| 3136 | if (dev->change_rx_flags && (dev->flags ^ flags) & IFF_MULTICAST)
|
| 3137 | dev->change_rx_flags(dev, IFF_MULTICAST);
|
| 3138 |
|
| 3139 | dev_set_rx_mode(dev);
|
| 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) {
|
| 3149 | ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
|
| 3150 |
|
| 3151 | if (!ret)
|
| 3152 | dev_set_rx_mode(dev);
|
| 3153 | }
|
| 3154 |
|
| 3155 | if (dev->flags & IFF_UP &&
|
| 3156 | ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
|
| 3157 | IFF_VOLATILE)))
|
| 3158 | call_netdevice_notifiers(NETDEV_CHANGE, dev);
|
| 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 | |
| 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 |
|
| 3176 |
|
| 3177 | changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
|
| 3178 | if (changes)
|
| 3179 | rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
|
| 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 |
|
| 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)
|
| 3204 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
|
| 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)
|
| 3220 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
|
| 3221 | return err;
|
| 3222 | }
|
| 3223 |
|
| 3224 | |
| 3225 | * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
|
| 3226 |
|
| 3227 | static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
|
| 3228 | {
|
| 3229 | int err;
|
| 3230 | struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
|
| 3231 |
|
| 3232 | if (!dev)
|
| 3233 | return -ENODEV;
|
| 3234 |
|
| 3235 | switch (cmd) {
|
| 3236 | case SIOCGIFFLAGS:
|
| 3237 | ifr->ifr_flags = dev_get_flags(dev);
|
| 3238 | return 0;
|
| 3239 |
|
| 3240 | case SIOCGIFMETRIC: |
| 3241 | (currently unused)
|
| 3242 | ifr->ifr_metric = 0;
|
| 3243 | return 0;
|
| 3244 |
|
| 3245 | case SIOCGIFMTU:
|
| 3246 | ifr->ifr_mtu = dev->mtu;
|
| 3247 | return 0;
|
| 3248 |
|
| 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 |
|
| 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 | |
| 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:
|
| 3304 | return dev_change_flags(dev, ifr->ifr_flags);
|
| 3305 |
|
| 3306 | case SIOCSIFMETRIC: |
| 3307 | (currently unused)
|
| 3308 | return -EOPNOTSUPP;
|
| 3309 |
|
| 3310 | case SIOCSIFMTU:
|
| 3311 | return dev_set_mtu(dev, ifr->ifr_mtu);
|
| 3312 |
|
| 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));
|
| 3321 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
|
| 3322 | return 0;
|
| 3323 |
|
| 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 && !dev->set_rx_mode) ||
|
| 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 && !dev->set_rx_mode) ||
|
| 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 |
|
| 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
|
| 3401 | * @net: the applicable net namespace
|
| 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 |
|
| 3411 | int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
|
| 3412 | {
|
| 3413 | struct ifreq ifr;
|
| 3414 | int ret;
|
| 3415 | char *colon;
|
| 3416 |
|
| 3417 | |
| 3418 | and requires shared lock, because it sleeps writing
|
| 3419 | to user space.
|
| 3420 |
|
| 3421 |
|
| 3422 | if (cmd == SIOCGIFCONF) {
|
| 3423 | rtnl_lock();
|
| 3424 | ret = dev_ifconf(net, (char __user *) arg);
|
| 3425 | rtnl_unlock();
|
| 3426 | return ret;
|
| 3427 | }
|
| 3428 | if (cmd == SIOCGIFNAME)
|
| 3429 | return dev_ifname(net, (struct ifreq __user *)arg);
|
| 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:
|
| 3459 | dev_load(net, ifr.ifr_name);
|
| 3460 | read_lock(&dev_base_lock);
|
| 3461 | ret = dev_ifsioc_locked(net, &ifr, cmd);
|
| 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:
|
| 3473 | dev_load(net, ifr.ifr_name);
|
| 3474 | rtnl_lock();
|
| 3475 | ret = dev_ethtool(net, &ifr);
|
| 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;
|
| 3497 | dev_load(net, ifr.ifr_name);
|
| 3498 | rtnl_lock();
|
| 3499 | ret = dev_ifsioc(net, &ifr, cmd);
|
| 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:
|
| 3530 | case SIOCBONDCHANGEACTIVE:
|
| 3531 | case SIOCBRADDIF:
|
| 3532 | case SIOCBRDELIF:
|
| 3533 | if (!capable(CAP_NET_ADMIN))
|
| 3534 | return -EPERM;
|
| 3535 |
|
| 3536 | case SIOCBONDSLAVEINFOQUERY:
|
| 3537 | case SIOCBONDINFOQUERY:
|
| 3538 | dev_load(net, ifr.ifr_name);
|
| 3539 | rtnl_lock();
|
| 3540 | ret = dev_ifsioc(net, &ifr, cmd);
|
| 3541 | rtnl_unlock();
|
| 3542 | return ret;
|
| 3543 |
|
| 3544 | case SIOCGIFMEM:
|
| 3545 | |
| 3546 | * currently do not support it
|
| 3547 | case SIOCSIFMEM:
|
| 3548 | |
| 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)) {
|
| 3560 | dev_load(net, ifr.ifr_name);
|
| 3561 | rtnl_lock();
|
| 3562 | ret = dev_ifsioc(net, &ifr, cmd);
|
| 3563 | rtnl_unlock();
|
| 3564 | if (!ret && copy_to_user(arg, &ifr,
|
| 3565 | sizeof(struct ifreq)))
|
| 3566 | ret = -EFAULT;
|
| 3567 | return ret;
|
| 3568 | }
|
| 3569 |
|
| 3570 | if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
|
| 3571 | return wext_handle_ioctl(net, &ifr, cmd, arg);
|
| 3572 | return -EINVAL;
|
| 3573 | }
|
| 3574 | }
|
| 3575 |
|
| 3576 |
|
| 3577 | |
| 3578 | * dev_new_index - allocate an ifindex
|
| 3579 | * @net: the applicable net namespace
|
| 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 |
|
| 3585 | static int dev_new_index(struct net *net)
|
| 3586 | {
|
| 3587 | static int ifindex;
|
| 3588 | for (;;) {
|
| 3589 | if (++ifindex <= 0)
|
| 3590 | ifindex = 1;
|
| 3591 | if (!__dev_get_by_index(net, ifindex))
|
| 3592 | return ifindex;
|
| 3593 | }
|
| 3594 | }
|
| 3595 |
|
| 3596 |
|
| 3597 | static DEFINE_SPINLOCK(net_todo_list_lock);
|
| 3598 | static LIST_HEAD(net_todo_list);
|
| 3599 |
|
| 3600 | static void net_set_todo(struct net_device *dev)
|
| 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 |
|
| 3607 | static void rollback_registered(struct net_device *dev)
|
| 3608 | {
|
| 3609 | BUG_ON(dev_boot_phase);
|
| 3610 | ASSERT_RTNL();
|
| 3611 |
|
| 3612 |
|
| 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 |
|
| 3624 | dev_close(dev);
|
| 3625 |
|
| 3626 |
|
| 3627 | unlist_netdevice(dev);
|
| 3628 |
|
| 3629 | dev->reg_state = NETREG_UNREGISTERING;
|
| 3630 |
|
| 3631 | synchronize_net();
|
| 3632 |
|
| 3633 |
|
| 3634 | dev_shutdown(dev);
|
| 3635 |
|
| 3636 |
|
| 3637 | |
| 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 |
|
| 3651 | BUG_TRAP(!dev->master);
|
| 3652 |
|
| 3653 |
|
| 3654 | netdev_unregister_kobject(dev);
|
| 3655 |
|
| 3656 | synchronize_net();
|
| 3657 |
|
| 3658 | dev_put(dev);
|
| 3659 | }
|
| 3660 |
|
| 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;
|
| 3683 | struct net *net;
|
| 3684 |
|
| 3685 | BUG_ON(dev_boot_phase);
|
| 3686 | ASSERT_RTNL();
|
| 3687 |
|
| 3688 | might_sleep();
|
| 3689 |
|
| 3690 |
|
| 3691 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
|
| 3692 | BUG_ON(!dev_net(dev));
|
| 3693 | net = dev_net(dev);
|
| 3694 |
|
| 3695 | spin_lock_init(&dev->queue_lock);
|
| 3696 | spin_lock_init(&dev->_xmit_lock);
|
| 3697 | netdev_set_lockdep_class(&dev->_xmit_lock, dev->type);
|
| 3698 | dev->xmit_lock_owner = -1;
|
| 3699 | spin_lock_init(&dev->ingress_lock);
|
| 3700 |
|
| 3701 | dev->iflink = -1;
|
| 3702 |
|
| 3703 |
|
| 3704 | if (dev->init) {
|
| 3705 | ret = dev->init(dev);
|
| 3706 | if (ret) {
|
| 3707 | if (ret > 0)
|
| 3708 | ret = -EIO;
|
| 3709 | goto out;
|
| 3710 | }
|
| 3711 | }
|
| 3712 |
|
| 3713 | if (!dev_valid_name(dev->name)) {
|
| 3714 | ret = -EINVAL;
|
| 3715 | goto err_uninit;
|
| 3716 | }
|
| 3717 |
|
| 3718 | dev->ifindex = dev_new_index(net);
|
| 3719 | if (dev->iflink == -1)
|
| 3720 | dev->iflink = dev->ifindex;
|
| 3721 |
|
| 3722 |
|
| 3723 | head = dev_name_hash(net, dev->name);
|
| 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;
|
| 3729 | goto err_uninit;
|
| 3730 | }
|
| 3731 | }
|
| 3732 |
|
| 3733 |
|
| 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 |
|
| 3749 |
|
| 3750 | if ((dev->features & NETIF_F_SG) &&
|
| 3751 | !(dev->features & NETIF_F_ALL_CSUM)) {
|
| 3752 | printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no checksum feature.\n",
|
| 3753 | dev->name);
|
| 3754 | dev->features &= ~NETIF_F_SG;
|
| 3755 | }
|
| 3756 |
|
| 3757 |
|
| 3758 | if ((dev->features & NETIF_F_TSO) &&
|
| 3759 | !(dev->features & NETIF_F_SG)) {
|
| 3760 | printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no SG feature.\n",
|
| 3761 | dev->name);
|
| 3762 | dev->features &= ~NETIF_F_TSO;
|
| 3763 | }
|
| 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 | }
|
| 3778 |
|
| 3779 | ret = netdev_register_kobject(dev);
|
| 3780 | if (ret)
|
| 3781 | goto err_uninit;
|
| 3782 | dev->reg_state = NETREG_REGISTERED;
|
| 3783 |
|
| 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 |
|
| 3791 | dev_init_scheduler(dev);
|
| 3792 | dev_hold(dev);
|
| 3793 | list_netdevice(dev);
|
| 3794 |
|
| 3795 |
|
| 3796 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
|
| 3797 | ret = notifier_to_errno(ret);
|
| 3798 | if (ret) {
|
| 3799 | rollback_registered(dev);
|
| 3800 | dev->reg_state = NETREG_UNREGISTERED;
|
| 3801 | }
|
| 3802 |
|
| 3803 | out:
|
| 3804 | return ret;
|
| 3805 |
|
| 3806 | err_uninit:
|
| 3807 | if (dev->uninit)
|
| 3808 | dev->uninit(dev);
|
| 3809 | goto out;
|
| 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 | *
|
| 3821 | * This is a wrapper around register_netdevice that takes the rtnl semaphore
|
| 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 | }
|
| 3840 |
|
| 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
|
| 3857 | * call dev_put.
|
| 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)) {
|
| 3866 | rtnl_lock();
|
| 3867 |
|
| 3868 |
|
| 3869 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
|
| 3870 |
|
| 3871 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
|
| 3872 | &dev->state)) {
|
| 3873 | |
| 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 |
|
| 3882 | __rtnl_unlock();
|
| 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 | |
| 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:
|
| 3915 | * 1) We can delete sysfs objects which invoke hotplug
|
| 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 |
|
| 3920 | static DEFINE_MUTEX(net_todo_run_mutex);
|
| 3921 | void netdev_run_todo(void)
|
| 3922 | {
|
| 3923 | struct list_head list;
|
| 3924 |
|
| 3925 |
|
| 3926 | mutex_lock(&net_todo_run_mutex);
|
| 3927 |
|
| 3928 | |
| 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 |
|
| 3937 | spin_lock(&net_todo_list_lock);
|
| 3938 | list_replace_init(&net_todo_list, &list);
|
| 3939 | spin_unlock(&net_todo_list_lock);
|
| 3940 |
|
| 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 |
|
| 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 | }
|
| 3952 |
|
| 3953 | dev->reg_state = NETREG_UNREGISTERED;
|
| 3954 |
|
| 3955 | netdev_wait_allrefs(dev);
|
| 3956 |
|
| 3957 |
|
| 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);
|
| 3962 |
|
| 3963 | if (dev->destructor)
|
| 3964 | dev->destructor(dev);
|
| 3965 |
|
| 3966 |
|
| 3967 | kobject_put(&dev->dev.kobj);
|
| 3968 | }
|
| 3969 |
|
| 3970 | out:
|
| 3971 | mutex_unlock(&net_todo_run_mutex);
|
| 3972 | }
|
| 3973 |
|
| 3974 | static struct net_device_stats *internal_stats(struct net_device *dev)
|
| 3975 | {
|
| 3976 | return &dev->stats;
|
| 3977 | }
|
| 3978 |
|
| 3979 | |
| 3980 | * alloc_netdev_mq - allocate network device
|
| 3981 | * @sizeof_priv: size of private data to allocate space for
|
| 3982 | * @name: device name format string
|
| 3983 | * @setup: callback to initialize device
|
| 3984 | * @queue_count: the number of subqueues to allocate
|
| 3985 | *
|
| 3986 | * Allocates a struct net_device with private data area for driver use
|
| 3987 | * and performs basic initialization. Also allocates subquue structs
|
| 3988 | * for each queue on the device at the end of the netdevice.
|
| 3989 |
|
| 3990 | struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
|
| 3991 | void (*setup)(struct net_device *), unsigned int queue_count)
|
| 3992 | {
|
| 3993 | void *p;
|
| 3994 | struct net_device *dev;
|
| 3995 | int alloc_size;
|
| 3996 |
|
| 3997 | BUG_ON(strlen(name) >= sizeof(dev->name));
|
| 3998 |
|
| 3999 | alloc_size = sizeof(struct net_device) +
|
| 4000 | sizeof(struct net_device_subqueue) * (queue_count - 1);
|
| 4001 | if (sizeof_priv) {
|
| 4002 |
|
| 4003 | alloc_size = (alloc_size + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
|
| 4004 | alloc_size += sizeof_priv;
|
| 4005 | }
|
| 4006 |
|
| 4007 | alloc_size += NETDEV_ALIGN_CONST;
|
| 4008 |
|
| 4009 | p = kzalloc(alloc_size, GFP_KERNEL);
|
| 4010 | if (!p) {
|
| 4011 | printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
|
| 4012 | return NULL;
|
| 4013 | }
|
| 4014 |
|
| 4015 | dev = (struct net_device *)
|
| 4016 | (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
|
| 4017 | dev->padded = (char *)dev - (char *)p;
|
| 4018 | dev_net_set(dev, &init_net);
|
| 4019 |
|
| 4020 | if (sizeof_priv) {
|
| 4021 | dev->priv = ((char *)dev +
|
| 4022 | ((sizeof(struct net_device) +
|
| 4023 | (sizeof(struct net_device_subqueue) *
|
| 4024 | (queue_count - 1)) + NETDEV_ALIGN_CONST)
|
| 4025 | & ~NETDEV_ALIGN_CONST));
|
| 4026 | }
|
| 4027 |
|
| 4028 | dev->egress_subqueue_count = queue_count;
|
| 4029 | dev->gso_max_size = GSO_MAX_SIZE;
|
| 4030 |
|
| 4031 | dev->get_stats = internal_stats;
|
| 4032 | netpoll_netdev_init(dev);
|
| 4033 | setup(dev);
|
| 4034 | strcpy(dev->name, name);
|
| 4035 | return dev;
|
| 4036 | }
|
| 4037 | EXPORT_SYMBOL(alloc_netdev_mq);
|
| 4038 |
|
| 4039 | |
| 4040 | * free_netdev - free network device
|
| 4041 | * @dev: device
|
| 4042 | *
|
| 4043 | * This function does the last stage of destroying an allocated device
|
| 4044 | * interface. The reference to the device object is released.
|
| 4045 | * If this is the last reference then it will be freed.
|
| 4046 |
|
| 4047 | void free_netdev(struct net_device *dev)
|
| 4048 | {
|
| 4049 | release_net(dev_net(dev));
|
| 4050 |
|
| 4051 |
|
| 4052 | if (dev->reg_state == NETREG_UNINITIALIZED) {
|
| 4053 | kfree((char *)dev - dev->padded);
|
| 4054 | return;
|
| 4055 | }
|
| 4056 |
|
| 4057 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
|
| 4058 | dev->reg_state = NETREG_RELEASED;
|
| 4059 |
|
| 4060 |
|
| 4061 | put_device(&dev->dev);
|
| 4062 | }
|
| 4063 |
|
| 4064 |
|
| 4065 | void synchronize_net(void)
|
| 4066 | {
|
| 4067 | might_sleep();
|
| 4068 | synchronize_rcu();
|
| 4069 | }
|
| 4070 |
|
| 4071 | |
| 4072 | * unregister_netdevice - remove device from the kernel
|
| 4073 | * @dev: device
|
| 4074 | *
|
| 4075 | * This function shuts down a device interface and removes it
|
| 4076 | * from the kernel tables.
|
| 4077 | *
|
| 4078 | * Callers must hold the rtnl semaphore. You may want
|
| 4079 | * unregister_netdev() instead of this.
|
| 4080 |
|
| 4081 |
|
| 4082 | void unregister_netdevice(struct net_device *dev)
|
| 4083 | {
|
| 4084 | ASSERT_RTNL();
|
| 4085 |
|
| 4086 | rollback_registered(dev);
|
| 4087 |
|
| 4088 | net_set_todo(dev);
|
| 4089 | }
|
| 4090 |
|
| 4091 | |
| 4092 | * unregister_netdev - remove device from the kernel
|
| 4093 | * @dev: device
|
| 4094 | *
|
| 4095 | * This function shuts down a device interface and removes it
|
| 4096 | * from the kernel tables.
|
| 4097 | *
|
| 4098 | * This is just a wrapper for unregister_netdevice that takes
|
| 4099 | * the rtnl semaphore. In general you want to use this and not
|
| 4100 | * unregister_netdevice.
|
| 4101 |
|
| 4102 | void unregister_netdev(struct net_device *dev)
|
| 4103 | {
|
| 4104 | rtnl_lock();
|
| 4105 | unregister_netdevice(dev);
|
| 4106 | rtnl_unlock();
|
| 4107 | }
|
| 4108 |
|
| 4109 | EXPORT_SYMBOL(unregister_netdev);
|
| 4110 |
|
| 4111 | |
| 4112 | * dev_change_net_namespace - move device to different nethost namespace
|
| 4113 | * @dev: device
|
| 4114 | * @net: network namespace
|
| 4115 | * @pat: If not NULL name pattern to try if the current device name
|
| 4116 | * is already taken in the destination network namespace.
|
| 4117 | *
|
| 4118 | * This function shuts down a device interface and moves it
|
| 4119 | * to a new network namespace. On success 0 is returned, on
|
| 4120 | * a failure a netagive errno code is returned.
|
| 4121 | *
|
| 4122 | * Callers must hold the rtnl semaphore.
|
| 4123 |
|
| 4124 |
|
| 4125 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
|
| 4126 | {
|
| 4127 | char buf[IFNAMSIZ];
|
| 4128 | const char *destname;
|
| 4129 | int err;
|
| 4130 |
|
| 4131 | ASSERT_RTNL();
|
| 4132 |
|
| 4133 |
|
| 4134 | err = -EINVAL;
|
| 4135 | if (dev->features & NETIF_F_NETNS_LOCAL)
|
| 4136 | goto out;
|
| 4137 |
|
| 4138 |
|
| 4139 | err = -EINVAL;
|
| 4140 | if (dev->reg_state != NETREG_REGISTERED)
|
| 4141 | goto out;
|
| 4142 |
|
| 4143 |
|
| 4144 | err = 0;
|
| 4145 | if (net_eq(dev_net(dev), net))
|
| 4146 | goto out;
|
| 4147 |
|
| 4148 | |
| 4149 | * we can use it in the destination network namespace.
|
| 4150 |
|
| 4151 | err = -EEXIST;
|
| 4152 | destname = dev->name;
|
| 4153 | if (__dev_get_by_name(net, destname)) {
|
| 4154 |
|
| 4155 | if (!pat)
|
| 4156 | goto out;
|
| 4157 | if (!dev_valid_name(pat))
|
| 4158 | goto out;
|
| 4159 | if (strchr(pat, '%')) {
|
| 4160 | if (__dev_alloc_name(net, pat, buf) < 0)
|
| 4161 | goto out;
|
| 4162 | destname = buf;
|
| 4163 | } else
|
| 4164 | destname = pat;
|
| 4165 | if (__dev_get_by_name(net, destname))
|
| 4166 | goto out;
|
| 4167 | }
|
| 4168 |
|
| 4169 | |
| 4170 | * And now a mini version of register_netdevice unregister_netdevice.
|
| 4171 |
|
| 4172 |
|
| 4173 |
|
| 4174 | dev_close(dev);
|
| 4175 |
|
| 4176 |
|
| 4177 | err = -ENODEV;
|
| 4178 | unlist_netdevice(dev);
|
| 4179 |
|
| 4180 | synchronize_net();
|
| 4181 |
|
| 4182 |
|
| 4183 | dev_shutdown(dev);
|
| 4184 |
|
| 4185 | |
| 4186 | this device. They should clean all the things.
|
| 4187 |
|
| 4188 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
|
| 4189 |
|
| 4190 | |
| 4191 | * Flush the unicast and multicast chains
|
| 4192 |
|
| 4193 | dev_addr_discard(dev);
|
| 4194 |
|
| 4195 |
|
| 4196 | dev_net_set(dev, net);
|
| 4197 |
|
| 4198 |
|
| 4199 | if (destname != dev->name)
|
| 4200 | strcpy(dev->name, destname);
|
| 4201 |
|
| 4202 |
|
| 4203 | if (__dev_get_by_index(net, dev->ifindex)) {
|
| 4204 | int iflink = (dev->iflink == dev->ifindex);
|
| 4205 | dev->ifindex = dev_new_index(net);
|
| 4206 | if (iflink)
|
| 4207 | dev->iflink = dev->ifindex;
|
| 4208 | }
|
| 4209 |
|
| 4210 |
|
| 4211 | err = device_rename(&dev->dev, dev->name);
|
| 4212 | WARN_ON(err);
|
| 4213 |
|
| 4214 |
|
| 4215 | list_netdevice(dev);
|
| 4216 |
|
| 4217 |
|
| 4218 | call_netdevice_notifiers(NETDEV_REGISTER, dev);
|
| 4219 |
|
| 4220 | synchronize_net();
|
| 4221 | err = 0;
|
| 4222 | out:
|
| 4223 | return err;
|
| 4224 | }
|
| 4225 |
|
| 4226 | static int dev_cpu_callback(struct notifier_block *nfb,
|
| 4227 | unsigned long action,
|
| 4228 | void *ocpu)
|
| 4229 | {
|
| 4230 | struct sk_buff **list_skb;
|
| 4231 | struct net_device **list_net;
|
| 4232 | struct sk_buff *skb;
|
| 4233 | unsigned int cpu, oldcpu = (unsigned long)ocpu;
|
| 4234 | struct softnet_data *sd, *oldsd;
|
| 4235 |
|
| 4236 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
|
| 4237 | return NOTIFY_OK;
|
| 4238 |
|
| 4239 | local_irq_disable();
|
| 4240 | cpu = smp_processor_id();
|
| 4241 | sd = &per_cpu(softnet_data, cpu);
|
| 4242 | oldsd = &per_cpu(softnet_data, oldcpu);
|
| 4243 |
|
| 4244 |
|
| 4245 | list_skb = &sd->completion_queue;
|
| 4246 | while (*list_skb)
|
| 4247 | list_skb = &(*list_skb)->next;
|
| 4248 |
|
| 4249 | *list_skb = oldsd->completion_queue;
|
| 4250 | oldsd->completion_queue = NULL;
|
| 4251 |
|
| 4252 |
|
| 4253 | list_net = &sd->output_queue;
|
| 4254 | while (*list_net)
|
| 4255 | list_net = &(*list_net)->next_sched;
|
| 4256 |
|
| 4257 | *list_net = oldsd->output_queue;
|
| 4258 | oldsd->output_queue = NULL;
|
| 4259 |
|
| 4260 | raise_softirq_irqoff(NET_TX_SOFTIRQ);
|
| 4261 | local_irq_enable();
|
| 4262 |
|
| 4263 |
|
| 4264 | while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
|
| 4265 | netif_rx(skb);
|
| 4266 |
|
| 4267 | return NOTIFY_OK;
|
| 4268 | }
|
| 4269 |
|
| 4270 | #ifdef CONFIG_NET_DMA
|
| 4271 | |
| 4272 | * net_dma_rebalance - try to maintain one DMA channel per CPU
|
| 4273 | * @net_dma: DMA client and associated data (lock, channels, channel_mask)
|
| 4274 | *
|
| 4275 | * This is called when the number of channels allocated to the net_dma client
|
| 4276 | * changes. The net_dma client tries to have one DMA channel per CPU.
|
| 4277 |
|
| 4278 |
|
| 4279 | static void net_dma_rebalance(struct net_dma *net_dma)
|
| 4280 | {
|
| 4281 | unsigned int cpu, i, n, chan_idx;
|
| 4282 | struct dma_chan *chan;
|
| 4283 |
|
| 4284 | if (cpus_empty(net_dma->channel_mask)) {
|
| 4285 | for_each_online_cpu(cpu)
|
| 4286 | rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL);
|
| 4287 | return;
|
| 4288 | }
|
| 4289 |
|
| 4290 | i = 0;
|
| 4291 | cpu = first_cpu(cpu_online_map);
|
| 4292 |
|
| 4293 | for_each_cpu_mask(chan_idx, net_dma->channel_mask) {
|
| 4294 | chan = net_dma->channels[chan_idx];
|
| 4295 |
|
| 4296 | n = ((num_online_cpus() / cpus_weight(net_dma->channel_mask))
|
| 4297 | + (i < (num_online_cpus() %
|
| 4298 | cpus_weight(net_dma->channel_mask)) ? 1 : 0));
|
| 4299 |
|
| 4300 | while(n) {
|
| 4301 | per_cpu(softnet_data, cpu).net_dma = chan;
|
| 4302 | cpu = next_cpu(cpu, cpu_online_map);
|
| 4303 | n--;
|
| 4304 | }
|
| 4305 | i++;
|
| 4306 | }
|
| 4307 | }
|
| 4308 |
|
| 4309 | |
| 4310 | * netdev_dma_event - event callback for the net_dma_client
|
| 4311 | * @client: should always be net_dma_client
|
| 4312 | * @chan: DMA channel for the event
|
| 4313 | * @state: DMA state to be handled
|
| 4314 |
|
| 4315 | static enum dma_state_client
|
| 4316 | netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
|
| 4317 | enum dma_state state)
|
| 4318 | {
|
| 4319 | int i, found = 0, pos = -1;
|
| 4320 | struct net_dma *net_dma =
|
| 4321 | container_of(client, struct net_dma, client);
|
| 4322 | enum dma_state_client ack = DMA_DUP;
|
| 4323 |
|
| 4324 | spin_lock(&net_dma->lock);
|
| 4325 | switch (state) {
|
| 4326 | case DMA_RESOURCE_AVAILABLE:
|
| 4327 | for (i = 0; i < NR_CPUS; i++)
|
| 4328 | if (net_dma->channels[i] == chan) {
|
| 4329 | found = 1;
|
| 4330 | break;
|
| 4331 | } else if (net_dma->channels[i] == NULL && pos < 0)
|
| 4332 | pos = i;
|
| 4333 |
|
| 4334 | if (!found && pos >= 0) {
|
| 4335 | ack = DMA_ACK;
|
| 4336 | net_dma->channels[pos] = chan;
|
| 4337 | cpu_set(pos, net_dma->channel_mask);
|
| 4338 | net_dma_rebalance(net_dma);
|
| 4339 | }
|
| 4340 | break;
|
| 4341 | case DMA_RESOURCE_REMOVED:
|
| 4342 | for (i = 0; i < NR_CPUS; i++)
|
| 4343 | if (net_dma->channels[i] == chan) {
|
| 4344 | found = 1;
|
| 4345 | pos = i;
|
| 4346 | break;
|
| 4347 | }
|
| 4348 |
|
| 4349 | if (found) {
|
| 4350 | ack = DMA_ACK;
|
| 4351 | cpu_clear(pos, net_dma->channel_mask);
|
| 4352 | net_dma->channels[i] = NULL;
|
| 4353 | net_dma_rebalance(net_dma);
|
| 4354 | }
|
| 4355 | break;
|
| 4356 | default:
|
| 4357 | break;
|
| 4358 | }
|
| 4359 | spin_unlock(&net_dma->lock);
|
| 4360 |
|
| 4361 | return ack;
|
| 4362 | }
|
| 4363 |
|
| 4364 | |
| 4365 | * netdev_dma_regiser - register the networking subsystem as a DMA client
|
| 4366 |
|
| 4367 | static int __init netdev_dma_register(void)
|
| 4368 | {
|
| 4369 | spin_lock_init(&net_dma.lock);
|
| 4370 | dma_cap_set(DMA_MEMCPY, net_dma.client.cap_mask);
|
| 4371 | dma_async_client_register(&net_dma.client);
|
| 4372 | dma_async_client_chan_request(&net_dma.client);
|
| 4373 | return 0;
|
| 4374 | }
|
| 4375 |
|
| 4376 | #else
|
| 4377 | static int __init netdev_dma_register(void) { return -ENODEV; }
|
| 4378 | #endif
|
| 4379 |
|
| 4380 | |
| 4381 | * netdev_compute_feature - compute conjunction of two feature sets
|
| 4382 | * @all: first feature set
|
| 4383 | * @one: second feature set
|
| 4384 | *
|
| 4385 | * Computes a new feature set after adding a device with feature set
|
| 4386 | * @one to the master device with current feature set @all. Returns
|
| 4387 | * the new feature set.
|
| 4388 |
|
| 4389 | int netdev_compute_features(unsigned long all, unsigned long one)
|
| 4390 | {
|
| 4391 |
|
| 4392 | if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
|
| 4393 | all ^= NETIF_F_NO_CSUM | NETIF_F_HW_CSUM;
|
| 4394 |
|
| 4395 |
|
| 4396 | if (all & NETIF_F_HW_CSUM && !(one & NETIF_F_HW_CSUM))
|
| 4397 | all ^= NETIF_F_HW_CSUM
|
| 4398 | | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
|
| 4399 |
|
| 4400 | if (one & NETIF_F_GSO)
|
| 4401 | one |= NETIF_F_GSO_SOFTWARE;
|
| 4402 | one |= NETIF_F_GSO;
|
| 4403 |
|
| 4404 |
|
| 4405 | if (one & NETIF_F_GSO_ROBUST)
|
| 4406 | all |= NETIF_F_GSO_ROBUST;
|
| 4407 |
|
| 4408 | all &= one | NETIF_F_LLTX;
|
| 4409 |
|
| 4410 | if (!(all & NETIF_F_ALL_CSUM))
|
| 4411 | all &= ~NETIF_F_SG;
|
| 4412 | if (!(all & NETIF_F_SG))
|
| 4413 | all &= ~NETIF_F_GSO_MASK;
|
| 4414 |
|
| 4415 | return all;
|
| 4416 | }
|
| 4417 | EXPORT_SYMBOL(netdev_compute_features);
|
| 4418 |
|
| 4419 | static struct hlist_head *netdev_create_hash(void)
|
| 4420 | {
|
| 4421 | int i;
|
| 4422 | struct hlist_head *hash;
|
| 4423 |
|
| 4424 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
|
| 4425 | if (hash != NULL)
|
| 4426 | for (i = 0; i < NETDEV_HASHENTRIES; i++)
|
| 4427 | INIT_HLIST_HEAD(&hash[i]);
|
| 4428 |
|
| 4429 | return hash;
|
| 4430 | }
|
| 4431 |
|
| 4432 |
|
| 4433 | static int __net_init netdev_init(struct net *net)
|
| 4434 | {
|
| 4435 | INIT_LIST_HEAD(&net->dev_base_head);
|
| 4436 |
|
| 4437 | net->dev_name_head = netdev_create_hash();
|
| 4438 | if (net->dev_name_head == NULL)
|
| 4439 | goto err_name;
|
| 4440 |
|
| 4441 | net->dev_index_head = netdev_create_hash();
|
| 4442 | if (net->dev_index_head == NULL)
|
| 4443 | goto err_idx;
|
| 4444 |
|
| 4445 | return 0;
|
| 4446 |
|
| 4447 | err_idx:
|
| 4448 | kfree(net->dev_name_head);
|
| 4449 | err_name:
|
| 4450 | return -ENOMEM;
|
| 4451 | }
|
| 4452 |
|
| 4453 | static void __net_exit netdev_exit(struct net *net)
|
| 4454 | {
|
| 4455 | kfree(net->dev_name_head);
|
| 4456 | kfree(net->dev_index_head);
|
| 4457 | }
|
| 4458 |
|
| 4459 | static struct pernet_operations __net_initdata netdev_net_ops = {
|
| 4460 | .init = netdev_init,
|
| 4461 | .exit = netdev_exit,
|
| 4462 | };
|
| 4463 |
|
| 4464 | static void __net_exit default_device_exit(struct net *net)
|
| 4465 | {
|
| 4466 | struct net_device *dev, *next;
|
| 4467 | |
| 4468 | * Push all migratable of the network devices back to the
|
| 4469 | * initial network namespace
|
| 4470 |
|
| 4471 | rtnl_lock();
|
| 4472 | for_each_netdev_safe(net, dev, next) {
|
| 4473 | int err;
|
| 4474 |
|
| 4475 |
|
| 4476 | if (dev->features & NETIF_F_NETNS_LOCAL)
|
| 4477 | continue;
|
| 4478 |
|
| 4479 |
|
| 4480 | err = dev_change_net_namespace(dev, &init_net, "dev%d");
|
| 4481 | if (err) {
|
| 4482 | printk(KERN_WARNING "%s: failed to move %s to init_net: %d\n",
|
| 4483 | __func__, dev->name, err);
|
| 4484 | unregister_netdevice(dev);
|
| 4485 | }
|
| 4486 | }
|
| 4487 | rtnl_unlock();
|
| 4488 | }
|
| 4489 |
|
| 4490 | static struct pernet_operations __net_initdata default_device_ops = {
|
| 4491 | .exit = default_device_exit,
|
| 4492 | };
|
| 4493 |
|
| 4494 | |
| 4495 | * Initialize the DEV module. At boot time this walks the device list and
|
| 4496 | * unhooks any devices that fail to initialise (normally hardware not
|
| 4497 | * present) and leaves us with a valid list of present and active devices.
|
| 4498 | *
|
| 4499 |
|
| 4500 |
|
| 4501 | |
| 4502 | * This is called single threaded during boot, so no need
|
| 4503 | * to take the rtnl semaphore.
|
| 4504 |
|
| 4505 | static int __init net_dev_init(void)
|
| 4506 | {
|
| 4507 | int i, rc = -ENOMEM;
|
| 4508 |
|
| 4509 | BUG_ON(!dev_boot_phase);
|
| 4510 |
|
| 4511 | if (dev_proc_init())
|
| 4512 | goto out;
|
| 4513 |
|
| 4514 | if (netdev_kobject_init())
|
| 4515 | goto out;
|
| 4516 |
|
| 4517 | INIT_LIST_HEAD(&ptype_all);
|
| 4518 | for (i = 0; i < PTYPE_HASH_SIZE; i++)
|
| 4519 | INIT_LIST_HEAD(&ptype_base[i]);
|
| 4520 |
|
| 4521 | if (register_pernet_subsys(&netdev_net_ops))
|
| 4522 | goto out;
|
| 4523 |
|
| 4524 | if (register_pernet_device(&default_device_ops))
|
| 4525 | goto out;
|
| 4526 |
|
| 4527 | |
| 4528 | * Initialise the packet receive queues.
|
| 4529 |
|
| 4530 |
|
| 4531 | for_each_possible_cpu(i) {
|
| 4532 | struct softnet_data *queue;
|
| 4533 |
|
| 4534 | queue = &per_cpu(softnet_data, i);
|
| 4535 | skb_queue_head_init(&queue->input_pkt_queue);
|
| 4536 | queue->completion_queue = NULL;
|
| 4537 | INIT_LIST_HEAD(&queue->poll_list);
|
| 4538 |
|
| 4539 | queue->backlog.poll = process_backlog;
|
| 4540 | queue->backlog.weight = weight_p;
|
| 4541 | }
|
| 4542 |
|
| 4543 | netdev_dma_register();
|
| 4544 |
|
| 4545 | dev_boot_phase = 0;
|
| 4546 |
|
| 4547 | open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
|
| 4548 | open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
|
| 4549 |
|
| 4550 | hotcpu_notifier(dev_cpu_callback, 0);
|
| 4551 | dst_init();
|
| 4552 | dev_mcast_init();
|
| 4553 | rc = 0;
|
| 4554 | out:
|
| 4555 | return rc;
|
| 4556 | }
|
| 4557 |
|
| 4558 | subsys_initcall(net_dev_init);
|
| 4559 |
|
| 4560 | EXPORT_SYMBOL(__dev_get_by_index);
|
| 4561 | EXPORT_SYMBOL(__dev_get_by_name);
|
| 4562 | EXPORT_SYMBOL(__dev_remove_pack);
|
| 4563 | EXPORT_SYMBOL(dev_valid_name);
|
| 4564 | EXPORT_SYMBOL(dev_add_pack);
|
| 4565 | EXPORT_SYMBOL(dev_alloc_name);
|
| 4566 | EXPORT_SYMBOL(dev_close);
|
| 4567 | EXPORT_SYMBOL(dev_get_by_flags);
|
| 4568 | EXPORT_SYMBOL(dev_get_by_index);
|
| 4569 | EXPORT_SYMBOL(dev_get_by_name);
|
| 4570 | EXPORT_SYMBOL(dev_open);
|
| 4571 | EXPORT_SYMBOL(dev_queue_xmit);
|
| 4572 | EXPORT_SYMBOL(dev_remove_pack);
|
| 4573 | EXPORT_SYMBOL(dev_set_allmulti);
|
| 4574 | EXPORT_SYMBOL(dev_set_promiscuity);
|
| 4575 | EXPORT_SYMBOL(dev_change_flags);
|
| 4576 | EXPORT_SYMBOL(dev_set_mtu);
|
| 4577 | EXPORT_SYMBOL(dev_set_mac_address);
|
| 4578 | EXPORT_SYMBOL(free_netdev);
|
| 4579 | EXPORT_SYMBOL(netdev_boot_setup_check);
|
| 4580 | EXPORT_SYMBOL(netdev_set_master);
|
| 4581 | EXPORT_SYMBOL(netdev_state_change);
|
| 4582 | EXPORT_SYMBOL(netif_receive_skb);
|
| 4583 | EXPORT_SYMBOL(netif_rx);
|
| 4584 | EXPORT_SYMBOL(register_gifconf);
|
| 4585 | EXPORT_SYMBOL(register_netdevice);
|
| 4586 | EXPORT_SYMBOL(register_netdevice_notifier);
|
| 4587 | EXPORT_SYMBOL(skb_checksum_help);
|
| 4588 | EXPORT_SYMBOL(synchronize_net);
|
| 4589 | EXPORT_SYMBOL(unregister_netdevice);
|
| 4590 | EXPORT_SYMBOL(unregister_netdevice_notifier);
|
| 4591 | EXPORT_SYMBOL(net_enable_timestamp);
|
| 4592 | EXPORT_SYMBOL(net_disable_timestamp);
|
| 4593 | EXPORT_SYMBOL(dev_get_flags);
|
| 4594 |
|
| 4595 | #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
|
| 4596 | EXPORT_SYMBOL(br_handle_frame_hook);
|
| 4597 | EXPORT_SYMBOL(br_fdb_get_hook);
|
| 4598 | EXPORT_SYMBOL(br_fdb_put_hook);
|
| 4599 | #endif
|
| 4600 |
|
| 4601 | #ifdef CONFIG_KMOD
|
| 4602 | EXPORT_SYMBOL(dev_load);
|
| 4603 | #endif
|
| 4604 |
|
| 4605 | EXPORT_PER_CPU_SYMBOL(softnet_data);
|