| 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/ethtool.h>
|
| 94 | #include <linux/notifier.h>
|
| 95 | #include <linux/skbuff.h>
|
| 96 | #include <net/net_namespace.h>
|
| 97 | #include <net/sock.h>
|
| 98 | #include <linux/rtnetlink.h>
|
| 99 | #include <linux/proc_fs.h>
|
| 100 | #include <linux/seq_file.h>
|
| 101 | #include <linux/stat.h>
|
| 102 | #include <linux/if_bridge.h>
|
| 103 | #include <linux/if_macvlan.h>
|
| 104 | #include <net/dst.h>
|
| 105 | #include <net/pkt_sched.h>
|
| 106 | #include <net/checksum.h>
|
| 107 | #include <linux/highmem.h>
|
| 108 | #include <linux/init.h>
|
| 109 | #include <linux/kmod.h>
|
| 110 | #include <linux/module.h>
|
| 111 | #include <linux/netpoll.h>
|
| 112 | #include <linux/rcupdate.h>
|
| 113 | #include <linux/delay.h>
|
| 114 | #include <net/wext.h>
|
| 115 | #include <net/iw_handler.h>
|
| 116 | #include <asm/current.h>
|
| 117 | #include <linux/audit.h>
|
| 118 | #include <linux/dmaengine.h>
|
| 119 | #include <linux/err.h>
|
| 120 | #include <linux/ctype.h>
|
| 121 | #include <linux/if_arp.h>
|
| 122 | #include <linux/if_vlan.h>
|
| 123 | #include <linux/ip.h>
|
| 124 | #include <net/ip.h>
|
| 125 | #include <linux/ipv6.h>
|
| 126 | #include <linux/in.h>
|
| 127 | #include <linux/jhash.h>
|
| 128 | #include <linux/random.h>
|
| 129 | #include <trace/events/napi.h>
|
| 130 |
|
| 131 | #include "net-sysfs.h"
|
| 132 |
|
| 133 |
|
| 134 | #define MAX_GRO_SKBS 8
|
| 135 |
|
| 136 |
|
| 137 | #define GRO_MAX_HEAD (MAX_HEADER + 128)
|
| 138 |
|
| 139 | |
| 140 | * The list of packet types we will receive (as opposed to discard)
|
| 141 | * and the routines to invoke.
|
| 142 | *
|
| 143 | * Why 16. Because with 16 the only overlap we get on a hash of the
|
| 144 | * low nibble of the protocol value is RARP/SNAP/X.25.
|
| 145 | *
|
| 146 | * NOTE: That is no longer true with the addition of VLAN tags. Not
|
| 147 | * sure which should go first, but I bet it won't make much
|
| 148 | * difference if we are running VLANs. The good news is that
|
| 149 | * this protocol won't be in the list unless compiled in, so
|
| 150 | * the average user (w/out VLANs) will not be adversely affected.
|
| 151 | * --BLG
|
| 152 | *
|
| 153 | * 0800 IP
|
| 154 | * 8100 802.1Q VLAN
|
| 155 | * 0001 802.3
|
| 156 | * 0002 AX.25
|
| 157 | * 0004 802.2
|
| 158 | * 8035 RARP
|
| 159 | * 0005 SNAP
|
| 160 | * 0805 X.25
|
| 161 | * 0806 ARP
|
| 162 | * 8137 IPX
|
| 163 | * 0009 Localtalk
|
| 164 | * 86DD IPv6
|
| 165 |
|
| 166 |
|
| 167 | #define PTYPE_HASH_SIZE (16)
|
| 168 | #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
|
| 169 |
|
| 170 | static DEFINE_SPINLOCK(ptype_lock);
|
| 171 | static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
|
| 172 | static struct list_head ptype_all __read_mostly;
|
| 173 |
|
| 174 | |
| 175 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl
|
| 176 | * semaphore.
|
| 177 | *
|
| 178 | * Pure readers hold dev_base_lock for reading.
|
| 179 | *
|
| 180 | * Writers must hold the rtnl semaphore while they loop through the
|
| 181 | * dev_base_head list, and hold dev_base_lock for writing when they do the
|
| 182 | * actual updates. This allows pure readers to access the list even
|
| 183 | * while a writer is preparing to update it.
|
| 184 | *
|
| 185 | * To put it another way, dev_base_lock is held for writing only to
|
| 186 | * protect against pure readers; the rtnl semaphore provides the
|
| 187 | * protection against other writers.
|
| 188 | *
|
| 189 | * See, for example usages, register_netdevice() and
|
| 190 | * unregister_netdevice(), which must be called with the rtnl
|
| 191 | * semaphore held.
|
| 192 |
|
| 193 | DEFINE_RWLOCK(dev_base_lock);
|
| 194 |
|
| 195 | EXPORT_SYMBOL(dev_base_lock);
|
| 196 |
|
| 197 | #define NETDEV_HASHBITS 8
|
| 198 | #define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS)
|
| 199 |
|
| 200 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
|
| 201 | {
|
| 202 | unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
|
| 203 | return &net->dev_name_head[hash & ((1 << NETDEV_HASHBITS) - 1)];
|
| 204 | }
|
| 205 |
|
| 206 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
|
| 207 | {
|
| 208 | return &net->dev_index_head[ifindex & ((1 << NETDEV_HASHBITS) - 1)];
|
| 209 | }
|
| 210 |
|
| 211 |
|
| 212 | static int list_netdevice(struct net_device *dev)
|
| 213 | {
|
| 214 | struct net *net = dev_net(dev);
|
| 215 |
|
| 216 | ASSERT_RTNL();
|
| 217 |
|
| 218 | write_lock_bh(&dev_base_lock);
|
| 219 | list_add_tail(&dev->dev_list, &net->dev_base_head);
|
| 220 | hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
|
| 221 | hlist_add_head(&dev->index_hlist, dev_index_hash(net, dev->ifindex));
|
| 222 | write_unlock_bh(&dev_base_lock);
|
| 223 | return 0;
|
| 224 | }
|
| 225 |
|
| 226 |
|
| 227 | static void unlist_netdevice(struct net_device *dev)
|
| 228 | {
|
| 229 | ASSERT_RTNL();
|
| 230 |
|
| 231 |
|
| 232 | write_lock_bh(&dev_base_lock);
|
| 233 | list_del(&dev->dev_list);
|
| 234 | hlist_del(&dev->name_hlist);
|
| 235 | hlist_del(&dev->index_hlist);
|
| 236 | write_unlock_bh(&dev_base_lock);
|
| 237 | }
|
| 238 |
|
| 239 | |
| 240 | * Our notifier list
|
| 241 |
|
| 242 |
|
| 243 | static RAW_NOTIFIER_HEAD(netdev_chain);
|
| 244 |
|
| 245 | |
| 246 | * Device drivers call our routines to queue packets here. We empty the
|
| 247 | * queue in the local softnet handler.
|
| 248 |
|
| 249 |
|
| 250 | DEFINE_PER_CPU(struct softnet_data, softnet_data);
|
| 251 |
|
| 252 | #ifdef CONFIG_LOCKDEP
|
| 253 | |
| 254 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class
|
| 255 | * according to dev->type
|
| 256 |
|
| 257 | static const unsigned short netdev_lock_type[] =
|
| 258 | {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
|
| 259 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
|
| 260 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
|
| 261 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
|
| 262 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
|
| 263 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
|
| 264 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
|
| 265 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
|
| 266 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
|
| 267 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
|
| 268 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
|
| 269 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
|
| 270 | ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
|
| 271 | ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
|
| 272 | ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154, ARPHRD_IEEE802154_PHY,
|
| 273 | ARPHRD_VOID, ARPHRD_NONE};
|
| 274 |
|
| 275 | static const char *netdev_lock_name[] =
|
| 276 | {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
|
| 277 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
|
| 278 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
|
| 279 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
|
| 280 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
|
| 281 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
|
| 282 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
|
| 283 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
|
| 284 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
|
| 285 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
|
| 286 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
|
| 287 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
|
| 288 | "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
|
| 289 | "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
|
| 290 | "_xmit_PHONET_PIPE", "_xmit_IEEE802154", "_xmit_IEEE802154_PHY",
|
| 291 | "_xmit_VOID", "_xmit_NONE"};
|
| 292 |
|
| 293 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
|
| 294 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
|
| 295 |
|
| 296 | static inline unsigned short netdev_lock_pos(unsigned short dev_type)
|
| 297 | {
|
| 298 | int i;
|
| 299 |
|
| 300 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
|
| 301 | if (netdev_lock_type[i] == dev_type)
|
| 302 | return i;
|
| 303 |
|
| 304 | return ARRAY_SIZE(netdev_lock_type) - 1;
|
| 305 | }
|
| 306 |
|
| 307 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
|
| 308 | unsigned short dev_type)
|
| 309 | {
|
| 310 | int i;
|
| 311 |
|
| 312 | i = netdev_lock_pos(dev_type);
|
| 313 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
|
| 314 | netdev_lock_name[i]);
|
| 315 | }
|
| 316 |
|
| 317 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
|
| 318 | {
|
| 319 | int i;
|
| 320 |
|
| 321 | i = netdev_lock_pos(dev->type);
|
| 322 | lockdep_set_class_and_name(&dev->addr_list_lock,
|
| 323 | &netdev_addr_lock_key[i],
|
| 324 | netdev_lock_name[i]);
|
| 325 | }
|
| 326 | #else
|
| 327 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
|
| 328 | unsigned short dev_type)
|
| 329 | {
|
| 330 | }
|
| 331 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
|
| 332 | {
|
| 333 | }
|
| 334 | #endif
|
| 335 |
|
| 336 | |
| 337 |
|
| 338 | Protocol management and registration routines
|
| 339 |
|
| 340 | ******************************************************************************
|
| 341 |
|
| 342 | |
| 343 | * Add a protocol ID to the list. Now that the input handler is
|
| 344 | * smarter we can dispense with all the messy stuff that used to be
|
| 345 | * here.
|
| 346 | *
|
| 347 | * BEWARE!!! Protocol handlers, mangling input packets,
|
| 348 | * MUST BE last in hash buckets and checking protocol handlers
|
| 349 | * MUST start from promiscuous ptype_all chain in net_bh.
|
| 350 | * It is true now, do not change it.
|
| 351 | * Explanation follows: if protocol handler, mangling packet, will
|
| 352 | * be the first on list, it is not able to sense, that packet
|
| 353 | * is cloned and should be copied-on-write, so that it will
|
| 354 | * change it and subsequent readers will get broken packet.
|
| 355 | * --ANK (980803)
|
| 356 |
|
| 357 |
|
| 358 | |
| 359 | * dev_add_pack - add packet handler
|
| 360 | * @pt: packet type declaration
|
| 361 | *
|
| 362 | * Add a protocol handler to the networking stack. The passed &packet_type
|
| 363 | * is linked into kernel lists and may not be freed until it has been
|
| 364 | * removed from the kernel lists.
|
| 365 | *
|
| 366 | * This call does not sleep therefore it can not
|
| 367 | * guarantee all CPU's that are in middle of receiving packets
|
| 368 | * will see the new packet type (until the next received packet).
|
| 369 |
|
| 370 |
|
| 371 | void dev_add_pack(struct packet_type *pt)
|
| 372 | {
|
| 373 | int hash;
|
| 374 |
|
| 375 | spin_lock_bh(&ptype_lock);
|
| 376 | if (pt->type == htons(ETH_P_ALL))
|
| 377 | list_add_rcu(&pt->list, &ptype_all);
|
| 378 | else {
|
| 379 | hash = ntohs(pt->type) & PTYPE_HASH_MASK;
|
| 380 | list_add_rcu(&pt->list, &ptype_base[hash]);
|
| 381 | }
|
| 382 | spin_unlock_bh(&ptype_lock);
|
| 383 | }
|
| 384 |
|
| 385 | |
| 386 | * __dev_remove_pack - remove packet handler
|
| 387 | * @pt: packet type declaration
|
| 388 | *
|
| 389 | * Remove a protocol handler that was previously added to the kernel
|
| 390 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed
|
| 391 | * from the kernel lists and can be freed or reused once this function
|
| 392 | * returns.
|
| 393 | *
|
| 394 | * The packet type might still be in use by receivers
|
| 395 | * and must not be freed until after all the CPU's have gone
|
| 396 | * through a quiescent state.
|
| 397 |
|
| 398 | void __dev_remove_pack(struct packet_type *pt)
|
| 399 | {
|
| 400 | struct list_head *head;
|
| 401 | struct packet_type *pt1;
|
| 402 |
|
| 403 | spin_lock_bh(&ptype_lock);
|
| 404 |
|
| 405 | if (pt->type == htons(ETH_P_ALL))
|
| 406 | head = &ptype_all;
|
| 407 | else
|
| 408 | head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
|
| 409 |
|
| 410 | list_for_each_entry(pt1, head, list) {
|
| 411 | if (pt == pt1) {
|
| 412 | list_del_rcu(&pt->list);
|
| 413 | goto out;
|
| 414 | }
|
| 415 | }
|
| 416 |
|
| 417 | printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
|
| 418 | out:
|
| 419 | spin_unlock_bh(&ptype_lock);
|
| 420 | }
|
| 421 | |
| 422 | * dev_remove_pack - remove packet handler
|
| 423 | * @pt: packet type declaration
|
| 424 | *
|
| 425 | * Remove a protocol handler that was previously added to the kernel
|
| 426 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed
|
| 427 | * from the kernel lists and can be freed or reused once this function
|
| 428 | * returns.
|
| 429 | *
|
| 430 | * This call sleeps to guarantee that no CPU is looking at the packet
|
| 431 | * type after return.
|
| 432 |
|
| 433 | void dev_remove_pack(struct packet_type *pt)
|
| 434 | {
|
| 435 | __dev_remove_pack(pt);
|
| 436 |
|
| 437 | synchronize_net();
|
| 438 | }
|
| 439 |
|
| 440 | |
| 441 |
|
| 442 | Device Boot-time Settings Routines
|
| 443 |
|
| 444 | ******************************************************************************
|
| 445 |
|
| 446 |
|
| 447 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
|
| 448 |
|
| 449 | |
| 450 | * netdev_boot_setup_add - add new setup entry
|
| 451 | * @name: name of the device
|
| 452 | * @map: configured settings for the device
|
| 453 | *
|
| 454 | * Adds new setup entry to the dev_boot_setup list. The function
|
| 455 | * returns 0 on error and 1 on success. This is a generic routine to
|
| 456 | * all netdevices.
|
| 457 |
|
| 458 | static int netdev_boot_setup_add(char *name, struct ifmap *map)
|
| 459 | {
|
| 460 | struct netdev_boot_setup *s;
|
| 461 | int i;
|
| 462 |
|
| 463 | s = dev_boot_setup;
|
| 464 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
|
| 465 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
|
| 466 | memset(s[i].name, 0, sizeof(s[i].name));
|
| 467 | strlcpy(s[i].name, name, IFNAMSIZ);
|
| 468 | memcpy(&s[i].map, map, sizeof(s[i].map));
|
| 469 | break;
|
| 470 | }
|
| 471 | }
|
| 472 |
|
| 473 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
|
| 474 | }
|
| 475 |
|
| 476 | |
| 477 | * netdev_boot_setup_check - check boot time settings
|
| 478 | * @dev: the netdevice
|
| 479 | *
|
| 480 | * Check boot time settings for the device.
|
| 481 | * The found settings are set for the device to be used
|
| 482 | * later in the device probing.
|
| 483 | * Returns 0 if no settings found, 1 if they are.
|
| 484 |
|
| 485 | int netdev_boot_setup_check(struct net_device *dev)
|
| 486 | {
|
| 487 | struct netdev_boot_setup *s = dev_boot_setup;
|
| 488 | int i;
|
| 489 |
|
| 490 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
|
| 491 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
|
| 492 | !strcmp(dev->name, s[i].name)) {
|
| 493 | dev->irq = s[i].map.irq;
|
| 494 | dev->base_addr = s[i].map.base_addr;
|
| 495 | dev->mem_start = s[i].map.mem_start;
|
| 496 | dev->mem_end = s[i].map.mem_end;
|
| 497 | return 1;
|
| 498 | }
|
| 499 | }
|
| 500 | return 0;
|
| 501 | }
|
| 502 |
|
| 503 |
|
| 504 | |
| 505 | * netdev_boot_base - get address from boot time settings
|
| 506 | * @prefix: prefix for network device
|
| 507 | * @unit: id for network device
|
| 508 | *
|
| 509 | * Check boot time settings for the base address of device.
|
| 510 | * The found settings are set for the device to be used
|
| 511 | * later in the device probing.
|
| 512 | * Returns 0 if no settings found.
|
| 513 |
|
| 514 | unsigned long netdev_boot_base(const char *prefix, int unit)
|
| 515 | {
|
| 516 | const struct netdev_boot_setup *s = dev_boot_setup;
|
| 517 | char name[IFNAMSIZ];
|
| 518 | int i;
|
| 519 |
|
| 520 | sprintf(name, "%s%d", prefix, unit);
|
| 521 |
|
| 522 | |
| 523 | * If device already registered then return base of 1
|
| 524 | * to indicate not to probe for this interface
|
| 525 |
|
| 526 | if (__dev_get_by_name(&init_net, name))
|
| 527 | return 1;
|
| 528 |
|
| 529 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
|
| 530 | if (!strcmp(name, s[i].name))
|
| 531 | return s[i].map.base_addr;
|
| 532 | return 0;
|
| 533 | }
|
| 534 |
|
| 535 | |
| 536 | * Saves at boot time configured settings for any netdevice.
|
| 537 |
|
| 538 | int __init netdev_boot_setup(char *str)
|
| 539 | {
|
| 540 | int ints[5];
|
| 541 | struct ifmap map;
|
| 542 |
|
| 543 | str = get_options(str, ARRAY_SIZE(ints), ints);
|
| 544 | if (!str || !*str)
|
| 545 | return 0;
|
| 546 |
|
| 547 |
|
| 548 | memset(&map, 0, sizeof(map));
|
| 549 | if (ints[0] > 0)
|
| 550 | map.irq = ints[1];
|
| 551 | if (ints[0] > 1)
|
| 552 | map.base_addr = ints[2];
|
| 553 | if (ints[0] > 2)
|
| 554 | map.mem_start = ints[3];
|
| 555 | if (ints[0] > 3)
|
| 556 | map.mem_end = ints[4];
|
| 557 |
|
| 558 |
|
| 559 | return netdev_boot_setup_add(str, &map);
|
| 560 | }
|
| 561 |
|
| 562 | __setup("netdev=", netdev_boot_setup);
|
| 563 |
|
| 564 | |
| 565 |
|
| 566 | Device Interface Subroutines
|
| 567 |
|
| 568 | ******************************************************************************
|
| 569 |
|
| 570 | |
| 571 | * __dev_get_by_name - find a device by its name
|
| 572 | * @net: the applicable net namespace
|
| 573 | * @name: name to find
|
| 574 | *
|
| 575 | * Find an interface by name. Must be called under RTNL semaphore
|
| 576 | * or @dev_base_lock. If the name is found a pointer to the device
|
| 577 | * is returned. If the name is not found then %NULL is returned. The
|
| 578 | * reference counters are not incremented so the caller must be
|
| 579 | * careful with locks.
|
| 580 |
|
| 581 |
|
| 582 | struct net_device *__dev_get_by_name(struct net *net, const char *name)
|
| 583 | {
|
| 584 | struct hlist_node *p;
|
| 585 |
|
| 586 | hlist_for_each(p, dev_name_hash(net, name)) {
|
| 587 | struct net_device *dev
|
| 588 | = hlist_entry(p, struct net_device, name_hlist);
|
| 589 | if (!strncmp(dev->name, name, IFNAMSIZ))
|
| 590 | return dev;
|
| 591 | }
|
| 592 | return NULL;
|
| 593 | }
|
| 594 |
|
| 595 | |
| 596 | * dev_get_by_name - find a device by its name
|
| 597 | * @net: the applicable net namespace
|
| 598 | * @name: name to find
|
| 599 | *
|
| 600 | * Find an interface by name. This can be called from any
|
| 601 | * context and does its own locking. The returned handle has
|
| 602 | * the usage count incremented and the caller must use dev_put() to
|
| 603 | * release it when it is no longer needed. %NULL is returned if no
|
| 604 | * matching device is found.
|
| 605 |
|
| 606 |
|
| 607 | struct net_device *dev_get_by_name(struct net *net, const char *name)
|
| 608 | {
|
| 609 | struct net_device *dev;
|
| 610 |
|
| 611 | read_lock(&dev_base_lock);
|
| 612 | dev = __dev_get_by_name(net, name);
|
| 613 | if (dev)
|
| 614 | dev_hold(dev);
|
| 615 | read_unlock(&dev_base_lock);
|
| 616 | return dev;
|
| 617 | }
|
| 618 |
|
| 619 | |
| 620 | * __dev_get_by_index - find a device by its ifindex
|
| 621 | * @net: the applicable net namespace
|
| 622 | * @ifindex: index of device
|
| 623 | *
|
| 624 | * Search for an interface by index. Returns %NULL if the device
|
| 625 | * is not found or a pointer to the device. The device has not
|
| 626 | * had its reference counter increased so the caller must be careful
|
| 627 | * about locking. The caller must hold either the RTNL semaphore
|
| 628 | * or @dev_base_lock.
|
| 629 |
|
| 630 |
|
| 631 | struct net_device *__dev_get_by_index(struct net *net, int ifindex)
|
| 632 | {
|
| 633 | struct hlist_node *p;
|
| 634 |
|
| 635 | hlist_for_each(p, dev_index_hash(net, ifindex)) {
|
| 636 | struct net_device *dev
|
| 637 | = hlist_entry(p, struct net_device, index_hlist);
|
| 638 | if (dev->ifindex == ifindex)
|
| 639 | return dev;
|
| 640 | }
|
| 641 | return NULL;
|
| 642 | }
|
| 643 |
|
| 644 |
|
| 645 | |
| 646 | * dev_get_by_index - find a device by its ifindex
|
| 647 | * @net: the applicable net namespace
|
| 648 | * @ifindex: index of device
|
| 649 | *
|
| 650 | * Search for an interface by index. Returns NULL if the device
|
| 651 | * is not found or a pointer to the device. The device returned has
|
| 652 | * had a reference added and the pointer is safe until the user calls
|
| 653 | * dev_put to indicate they have finished with it.
|
| 654 |
|
| 655 |
|
| 656 | struct net_device *dev_get_by_index(struct net *net, int ifindex)
|
| 657 | {
|
| 658 | struct net_device *dev;
|
| 659 |
|
| 660 | read_lock(&dev_base_lock);
|
| 661 | dev = __dev_get_by_index(net, ifindex);
|
| 662 | if (dev)
|
| 663 | dev_hold(dev);
|
| 664 | read_unlock(&dev_base_lock);
|
| 665 | return dev;
|
| 666 | }
|
| 667 |
|
| 668 | |
| 669 | * dev_getbyhwaddr - find a device by its hardware address
|
| 670 | * @net: the applicable net namespace
|
| 671 | * @type: media type of device
|
| 672 | * @ha: hardware address
|
| 673 | *
|
| 674 | * Search for an interface by MAC address. Returns NULL if the device
|
| 675 | * is not found or a pointer to the device. The caller must hold the
|
| 676 | * rtnl semaphore. The returned device has not had its ref count increased
|
| 677 | * and the caller must therefore be careful about locking
|
| 678 | *
|
| 679 | * BUGS:
|
| 680 | * If the API was consistent this would be __dev_get_by_hwaddr
|
| 681 |
|
| 682 |
|
| 683 | struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
|
| 684 | {
|
| 685 | struct net_device *dev;
|
| 686 |
|
| 687 | ASSERT_RTNL();
|
| 688 |
|
| 689 | for_each_netdev(net, dev)
|
| 690 | if (dev->type == type &&
|
| 691 | !memcmp(dev->dev_addr, ha, dev->addr_len))
|
| 692 | return dev;
|
| 693 |
|
| 694 | return NULL;
|
| 695 | }
|
| 696 |
|
| 697 | EXPORT_SYMBOL(dev_getbyhwaddr);
|
| 698 |
|
| 699 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
|
| 700 | {
|
| 701 | struct net_device *dev;
|
| 702 |
|
| 703 | ASSERT_RTNL();
|
| 704 | for_each_netdev(net, dev)
|
| 705 | if (dev->type == type)
|
| 706 | return dev;
|
| 707 |
|
| 708 | return NULL;
|
| 709 | }
|
| 710 |
|
| 711 | EXPORT_SYMBOL(__dev_getfirstbyhwtype);
|
| 712 |
|
| 713 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
|
| 714 | {
|
| 715 | struct net_device *dev;
|
| 716 |
|
| 717 | rtnl_lock();
|
| 718 | dev = __dev_getfirstbyhwtype(net, type);
|
| 719 | if (dev)
|
| 720 | dev_hold(dev);
|
| 721 | rtnl_unlock();
|
| 722 | return dev;
|
| 723 | }
|
| 724 |
|
| 725 | EXPORT_SYMBOL(dev_getfirstbyhwtype);
|
| 726 |
|
| 727 | |
| 728 | * dev_get_by_flags - find any device with given flags
|
| 729 | * @net: the applicable net namespace
|
| 730 | * @if_flags: IFF_* values
|
| 731 | * @mask: bitmask of bits in if_flags to check
|
| 732 | *
|
| 733 | * Search for any interface with the given flags. Returns NULL if a device
|
| 734 | * is not found or a pointer to the device. The device returned has
|
| 735 | * had a reference added and the pointer is safe until the user calls
|
| 736 | * dev_put to indicate they have finished with it.
|
| 737 |
|
| 738 |
|
| 739 | struct net_device * dev_get_by_flags(struct net *net, unsigned short if_flags, unsigned short mask)
|
| 740 | {
|
| 741 | struct net_device *dev, *ret;
|
| 742 |
|
| 743 | ret = NULL;
|
| 744 | read_lock(&dev_base_lock);
|
| 745 | for_each_netdev(net, dev) {
|
| 746 | if (((dev->flags ^ if_flags) & mask) == 0) {
|
| 747 | dev_hold(dev);
|
| 748 | ret = dev;
|
| 749 | break;
|
| 750 | }
|
| 751 | }
|
| 752 | read_unlock(&dev_base_lock);
|
| 753 | return ret;
|
| 754 | }
|
| 755 |
|
| 756 | |
| 757 | * dev_valid_name - check if name is okay for network device
|
| 758 | * @name: name string
|
| 759 | *
|
| 760 | * Network device names need to be valid file names to
|
| 761 | * to allow sysfs to work. We also disallow any kind of
|
| 762 | * whitespace.
|
| 763 |
|
| 764 | int dev_valid_name(const char *name)
|
| 765 | {
|
| 766 | if (*name == '\0')
|
| 767 | return 0;
|
| 768 | if (strlen(name) >= IFNAMSIZ)
|
| 769 | return 0;
|
| 770 | if (!strcmp(name, ".") || !strcmp(name, ".."))
|
| 771 | return 0;
|
| 772 |
|
| 773 | while (*name) {
|
| 774 | if (*name == '/' || isspace(*name))
|
| 775 | return 0;
|
| 776 | name++;
|
| 777 | }
|
| 778 | return 1;
|
| 779 | }
|
| 780 |
|
| 781 | |
| 782 | * __dev_alloc_name - allocate a name for a device
|
| 783 | * @net: network namespace to allocate the device name in
|
| 784 | * @name: name format string
|
| 785 | * @buf: scratch buffer and result name string
|
| 786 | *
|
| 787 | * Passed a format string - eg "lt%d" it will try and find a suitable
|
| 788 | * id. It scans list of devices to build up a free map, then chooses
|
| 789 | * the first empty slot. The caller must hold the dev_base or rtnl lock
|
| 790 | * while allocating the name and adding the device in order to avoid
|
| 791 | * duplicates.
|
| 792 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
|
| 793 | * Returns the number of the unit assigned or a negative errno code.
|
| 794 |
|
| 795 |
|
| 796 | static int __dev_alloc_name(struct net *net, const char *name, char *buf)
|
| 797 | {
|
| 798 | int i = 0;
|
| 799 | const char *p;
|
| 800 | const int max_netdevices = 8*PAGE_SIZE;
|
| 801 | unsigned long *inuse;
|
| 802 | struct net_device *d;
|
| 803 |
|
| 804 | p = strnchr(name, IFNAMSIZ-1, '%');
|
| 805 | if (p) {
|
| 806 | |
| 807 | * Verify the string as this thing may have come from
|
| 808 | * the user. There must be either one "%d" and no other "%"
|
| 809 | * characters.
|
| 810 |
|
| 811 | if (p[1] != 'd' || strchr(p + 2, '%'))
|
| 812 | return -EINVAL;
|
| 813 |
|
| 814 |
|
| 815 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
|
| 816 | if (!inuse)
|
| 817 | return -ENOMEM;
|
| 818 |
|
| 819 | for_each_netdev(net, d) {
|
| 820 | if (!sscanf(d->name, name, &i))
|
| 821 | continue;
|
| 822 | if (i < 0 || i >= max_netdevices)
|
| 823 | continue;
|
| 824 |
|
| 825 |
|
| 826 | snprintf(buf, IFNAMSIZ, name, i);
|
| 827 | if (!strncmp(buf, d->name, IFNAMSIZ))
|
| 828 | set_bit(i, inuse);
|
| 829 | }
|
| 830 |
|
| 831 | i = find_first_zero_bit(inuse, max_netdevices);
|
| 832 | free_page((unsigned long) inuse);
|
| 833 | }
|
| 834 |
|
| 835 | snprintf(buf, IFNAMSIZ, name, i);
|
| 836 | if (!__dev_get_by_name(net, buf))
|
| 837 | return i;
|
| 838 |
|
| 839 | |
| 840 | * when the name is long and there isn't enough space left
|
| 841 | * for the digits, or if all bits are used.
|
| 842 |
|
| 843 | return -ENFILE;
|
| 844 | }
|
| 845 |
|
| 846 | |
| 847 | * dev_alloc_name - allocate a name for a device
|
| 848 | * @dev: device
|
| 849 | * @name: name format string
|
| 850 | *
|
| 851 | * Passed a format string - eg "lt%d" it will try and find a suitable
|
| 852 | * id. It scans list of devices to build up a free map, then chooses
|
| 853 | * the first empty slot. The caller must hold the dev_base or rtnl lock
|
| 854 | * while allocating the name and adding the device in order to avoid
|
| 855 | * duplicates.
|
| 856 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
|
| 857 | * Returns the number of the unit assigned or a negative errno code.
|
| 858 |
|
| 859 |
|
| 860 | int dev_alloc_name(struct net_device *dev, const char *name)
|
| 861 | {
|
| 862 | char buf[IFNAMSIZ];
|
| 863 | struct net *net;
|
| 864 | int ret;
|
| 865 |
|
| 866 | BUG_ON(!dev_net(dev));
|
| 867 | net = dev_net(dev);
|
| 868 | ret = __dev_alloc_name(net, name, buf);
|
| 869 | if (ret >= 0)
|
| 870 | strlcpy(dev->name, buf, IFNAMSIZ);
|
| 871 | return ret;
|
| 872 | }
|
| 873 |
|
| 874 |
|
| 875 | |
| 876 | * dev_change_name - change name of a device
|
| 877 | * @dev: device
|
| 878 | * @newname: name (or format string) must be at least IFNAMSIZ
|
| 879 | *
|
| 880 | * Change name of a device, can pass format strings "eth%d".
|
| 881 | * for wildcarding.
|
| 882 |
|
| 883 | int dev_change_name(struct net_device *dev, const char *newname)
|
| 884 | {
|
| 885 | char oldname[IFNAMSIZ];
|
| 886 | int err = 0;
|
| 887 | int ret;
|
| 888 | struct net *net;
|
| 889 |
|
| 890 | ASSERT_RTNL();
|
| 891 | BUG_ON(!dev_net(dev));
|
| 892 |
|
| 893 | net = dev_net(dev);
|
| 894 | if (dev->flags & IFF_UP)
|
| 895 | return -EBUSY;
|
| 896 |
|
| 897 | if (!dev_valid_name(newname))
|
| 898 | return -EINVAL;
|
| 899 |
|
| 900 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
|
| 901 | return 0;
|
| 902 |
|
| 903 | memcpy(oldname, dev->name, IFNAMSIZ);
|
| 904 |
|
| 905 | if (strchr(newname, '%')) {
|
| 906 | err = dev_alloc_name(dev, newname);
|
| 907 | if (err < 0)
|
| 908 | return err;
|
| 909 | }
|
| 910 | else if (__dev_get_by_name(net, newname))
|
| 911 | return -EEXIST;
|
| 912 | else
|
| 913 | strlcpy(dev->name, newname, IFNAMSIZ);
|
| 914 |
|
| 915 | rollback:
|
| 916 | |
| 917 | * are in sysfs.
|
| 918 |
|
| 919 | if (net == &init_net) {
|
| 920 | ret = device_rename(&dev->dev, dev->name);
|
| 921 | if (ret) {
|
| 922 | memcpy(dev->name, oldname, IFNAMSIZ);
|
| 923 | return ret;
|
| 924 | }
|
| 925 | }
|
| 926 |
|
| 927 | write_lock_bh(&dev_base_lock);
|
| 928 | hlist_del(&dev->name_hlist);
|
| 929 | hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
|
| 930 | write_unlock_bh(&dev_base_lock);
|
| 931 |
|
| 932 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
|
| 933 | ret = notifier_to_errno(ret);
|
| 934 |
|
| 935 | if (ret) {
|
| 936 | if (err) {
|
| 937 | printk(KERN_ERR
|
| 938 | "%s: name change rollback failed: %d.\n",
|
| 939 | dev->name, ret);
|
| 940 | } else {
|
| 941 | err = ret;
|
| 942 | memcpy(dev->name, oldname, IFNAMSIZ);
|
| 943 | goto rollback;
|
| 944 | }
|
| 945 | }
|
| 946 |
|
| 947 | return err;
|
| 948 | }
|
| 949 |
|
| 950 | |
| 951 | * dev_set_alias - change ifalias of a device
|
| 952 | * @dev: device
|
| 953 | * @alias: name up to IFALIASZ
|
| 954 | * @len: limit of bytes to copy from info
|
| 955 | *
|
| 956 | * Set ifalias for a device,
|
| 957 |
|
| 958 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
|
| 959 | {
|
| 960 | ASSERT_RTNL();
|
| 961 |
|
| 962 | if (len >= IFALIASZ)
|
| 963 | return -EINVAL;
|
| 964 |
|
| 965 | if (!len) {
|
| 966 | if (dev->ifalias) {
|
| 967 | kfree(dev->ifalias);
|
| 968 | dev->ifalias = NULL;
|
| 969 | }
|
| 970 | return 0;
|
| 971 | }
|
| 972 |
|
| 973 | dev->ifalias = krealloc(dev->ifalias, len+1, GFP_KERNEL);
|
| 974 | if (!dev->ifalias)
|
| 975 | return -ENOMEM;
|
| 976 |
|
| 977 | strlcpy(dev->ifalias, alias, len+1);
|
| 978 | return len;
|
| 979 | }
|
| 980 |
|
| 981 |
|
| 982 | |
| 983 | * netdev_features_change - device changes features
|
| 984 | * @dev: device to cause notification
|
| 985 | *
|
| 986 | * Called to indicate a device has changed features.
|
| 987 |
|
| 988 | void netdev_features_change(struct net_device *dev)
|
| 989 | {
|
| 990 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
|
| 991 | }
|
| 992 | EXPORT_SYMBOL(netdev_features_change);
|
| 993 |
|
| 994 | |
| 995 | * netdev_state_change - device changes state
|
| 996 | * @dev: device to cause notification
|
| 997 | *
|
| 998 | * Called to indicate a device has changed state. This function calls
|
| 999 | * the notifier chains for netdev_chain and sends a NEWLINK message
|
| 1000 | * to the routing socket.
|
| 1001 |
|
| 1002 | void netdev_state_change(struct net_device *dev)
|
| 1003 | {
|
| 1004 | if (dev->flags & IFF_UP) {
|
| 1005 | call_netdevice_notifiers(NETDEV_CHANGE, dev);
|
| 1006 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
|
| 1007 | }
|
| 1008 | }
|
| 1009 |
|
| 1010 | void netdev_bonding_change(struct net_device *dev)
|
| 1011 | {
|
| 1012 | call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, dev);
|
| 1013 | }
|
| 1014 | EXPORT_SYMBOL(netdev_bonding_change);
|
| 1015 |
|
| 1016 | |
| 1017 | * dev_load - load a network module
|
| 1018 | * @net: the applicable net namespace
|
| 1019 | * @name: name of interface
|
| 1020 | *
|
| 1021 | * If a network interface is not present and the process has suitable
|
| 1022 | * privileges this function loads the module. If module loading is not
|
| 1023 | * available in this kernel then it becomes a nop.
|
| 1024 |
|
| 1025 |
|
| 1026 | void dev_load(struct net *net, const char *name)
|
| 1027 | {
|
| 1028 | struct net_device *dev;
|
| 1029 |
|
| 1030 | read_lock(&dev_base_lock);
|
| 1031 | dev = __dev_get_by_name(net, name);
|
| 1032 | read_unlock(&dev_base_lock);
|
| 1033 |
|
| 1034 | if (!dev && capable(CAP_SYS_MODULE))
|
| 1035 | request_module("%s", name);
|
| 1036 | }
|
| 1037 |
|
| 1038 | |
| 1039 | * dev_open - prepare an interface for use.
|
| 1040 | * @dev: device to open
|
| 1041 | *
|
| 1042 | * Takes a device from down to up state. The device's private open
|
| 1043 | * function is invoked and then the multicast lists are loaded. Finally
|
| 1044 | * the device is moved into the up state and a %NETDEV_UP message is
|
| 1045 | * sent to the netdev notifier chain.
|
| 1046 | *
|
| 1047 | * Calling this function on an active interface is a nop. On a failure
|
| 1048 | * a negative errno code is returned.
|
| 1049 |
|
| 1050 | int dev_open(struct net_device *dev)
|
| 1051 | {
|
| 1052 | const struct net_device_ops *ops = dev->netdev_ops;
|
| 1053 | int ret;
|
| 1054 |
|
| 1055 | ASSERT_RTNL();
|
| 1056 |
|
| 1057 | |
| 1058 | * Is it already up?
|
| 1059 |
|
| 1060 |
|
| 1061 | if (dev->flags & IFF_UP)
|
| 1062 | return 0;
|
| 1063 |
|
| 1064 | |
| 1065 | * Is it even present?
|
| 1066 |
|
| 1067 | if (!netif_device_present(dev))
|
| 1068 | return -ENODEV;
|
| 1069 |
|
| 1070 | ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
|
| 1071 | ret = notifier_to_errno(ret);
|
| 1072 | if (ret)
|
| 1073 | return ret;
|
| 1074 |
|
| 1075 | |
| 1076 | * Call device private open method
|
| 1077 |
|
| 1078 | set_bit(__LINK_STATE_START, &dev->state);
|
| 1079 |
|
| 1080 | if (ops->ndo_validate_addr)
|
| 1081 | ret = ops->ndo_validate_addr(dev);
|
| 1082 |
|
| 1083 | if (!ret && ops->ndo_open)
|
| 1084 | ret = ops->ndo_open(dev);
|
| 1085 |
|
| 1086 | |
| 1087 | * If it went open OK then:
|
| 1088 |
|
| 1089 |
|
| 1090 | if (ret)
|
| 1091 | clear_bit(__LINK_STATE_START, &dev->state);
|
| 1092 | else {
|
| 1093 | |
| 1094 | * Set the flags.
|
| 1095 |
|
| 1096 | dev->flags |= IFF_UP;
|
| 1097 |
|
| 1098 | |
| 1099 | * Enable NET_DMA
|
| 1100 |
|
| 1101 | net_dmaengine_get();
|
| 1102 |
|
| 1103 | |
| 1104 | * Initialize multicasting status
|
| 1105 |
|
| 1106 | dev_set_rx_mode(dev);
|
| 1107 |
|
| 1108 | |
| 1109 | * Wakeup transmit queue engine
|
| 1110 |
|
| 1111 | dev_activate(dev);
|
| 1112 |
|
| 1113 | |
| 1114 | * ... and announce new interface.
|
| 1115 |
|
| 1116 | call_netdevice_notifiers(NETDEV_UP, dev);
|
| 1117 | }
|
| 1118 |
|
| 1119 | return ret;
|
| 1120 | }
|
| 1121 |
|
| 1122 | |
| 1123 | * dev_close - shutdown an interface.
|
| 1124 | * @dev: device to shutdown
|
| 1125 | *
|
| 1126 | * This function moves an active device into down state. A
|
| 1127 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
|
| 1128 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
|
| 1129 | * chain.
|
| 1130 |
|
| 1131 | int dev_close(struct net_device *dev)
|
| 1132 | {
|
| 1133 | const struct net_device_ops *ops = dev->netdev_ops;
|
| 1134 | ASSERT_RTNL();
|
| 1135 |
|
| 1136 | might_sleep();
|
| 1137 |
|
| 1138 | if (!(dev->flags & IFF_UP))
|
| 1139 | return 0;
|
| 1140 |
|
| 1141 | |
| 1142 | * Tell people we are going down, so that they can
|
| 1143 | * prepare to death, when device is still operating.
|
| 1144 |
|
| 1145 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
|
| 1146 |
|
| 1147 | clear_bit(__LINK_STATE_START, &dev->state);
|
| 1148 |
|
| 1149 | |
| 1150 | * it can be even on different cpu. So just clear netif_running().
|
| 1151 | *
|
| 1152 | * dev->stop() will invoke napi_disable() on all of it's
|
| 1153 | * napi_struct instances on this device.
|
| 1154 |
|
| 1155 | smp_mb__after_clear_bit();
|
| 1156 |
|
| 1157 | dev_deactivate(dev);
|
| 1158 |
|
| 1159 | |
| 1160 | * Call the device specific close. This cannot fail.
|
| 1161 | * Only if device is UP
|
| 1162 | *
|
| 1163 | * We allow it to be called even after a DETACH hot-plug
|
| 1164 | * event.
|
| 1165 |
|
| 1166 | if (ops->ndo_stop)
|
| 1167 | ops->ndo_stop(dev);
|
| 1168 |
|
| 1169 | |
| 1170 | * Device is now down.
|
| 1171 |
|
| 1172 |
|
| 1173 | dev->flags &= ~IFF_UP;
|
| 1174 |
|
| 1175 | |
| 1176 | * Tell people we are down
|
| 1177 |
|
| 1178 | call_netdevice_notifiers(NETDEV_DOWN, dev);
|
| 1179 |
|
| 1180 | |
| 1181 | * Shutdown NET_DMA
|
| 1182 |
|
| 1183 | net_dmaengine_put();
|
| 1184 |
|
| 1185 | return 0;
|
| 1186 | }
|
| 1187 |
|
| 1188 |
|
| 1189 | |
| 1190 | * dev_disable_lro - disable Large Receive Offload on a device
|
| 1191 | * @dev: device
|
| 1192 | *
|
| 1193 | * Disable Large Receive Offload (LRO) on a net device. Must be
|
| 1194 | * called under RTNL. This is needed if received packets may be
|
| 1195 | * forwarded to another interface.
|
| 1196 |
|
| 1197 | void dev_disable_lro(struct net_device *dev)
|
| 1198 | {
|
| 1199 | if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
|
| 1200 | dev->ethtool_ops->set_flags) {
|
| 1201 | u32 flags = dev->ethtool_ops->get_flags(dev);
|
| 1202 | if (flags & ETH_FLAG_LRO) {
|
| 1203 | flags &= ~ETH_FLAG_LRO;
|
| 1204 | dev->ethtool_ops->set_flags(dev, flags);
|
| 1205 | }
|
| 1206 | }
|
| 1207 | WARN_ON(dev->features & NETIF_F_LRO);
|
| 1208 | }
|
| 1209 | EXPORT_SYMBOL(dev_disable_lro);
|
| 1210 |
|
| 1211 |
|
| 1212 | static int dev_boot_phase = 1;
|
| 1213 |
|
| 1214 | |
| 1215 | * Device change register/unregister. These are not inline or static
|
| 1216 | * as we export them to the world.
|
| 1217 |
|
| 1218 |
|
| 1219 | |
| 1220 | * register_netdevice_notifier - register a network notifier block
|
| 1221 | * @nb: notifier
|
| 1222 | *
|
| 1223 | * Register a notifier to be called when network device events occur.
|
| 1224 | * The notifier passed is linked into the kernel structures and must
|
| 1225 | * not be reused until it has been unregistered. A negative errno code
|
| 1226 | * is returned on a failure.
|
| 1227 | *
|
| 1228 | * When registered all registration and up events are replayed
|
| 1229 | * to the new notifier to allow device to have a race free
|
| 1230 | * view of the network device list.
|
| 1231 |
|
| 1232 |
|
| 1233 | int register_netdevice_notifier(struct notifier_block *nb)
|
| 1234 | {
|
| 1235 | struct net_device *dev;
|
| 1236 | struct net_device *last;
|
| 1237 | struct net *net;
|
| 1238 | int err;
|
| 1239 |
|
| 1240 | rtnl_lock();
|
| 1241 | err = raw_notifier_chain_register(&netdev_chain, nb);
|
| 1242 | if (err)
|
| 1243 | goto unlock;
|
| 1244 | if (dev_boot_phase)
|
| 1245 | goto unlock;
|
| 1246 | for_each_net(net) {
|
| 1247 | for_each_netdev(net, dev) {
|
| 1248 | err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
|
| 1249 | err = notifier_to_errno(err);
|
| 1250 | if (err)
|
| 1251 | goto rollback;
|
| 1252 |
|
| 1253 | if (!(dev->flags & IFF_UP))
|
| 1254 | continue;
|
| 1255 |
|
| 1256 | nb->notifier_call(nb, NETDEV_UP, dev);
|
| 1257 | }
|
| 1258 | }
|
| 1259 |
|
| 1260 | unlock:
|
| 1261 | rtnl_unlock();
|
| 1262 | return err;
|
| 1263 |
|
| 1264 | rollback:
|
| 1265 | last = dev;
|
| 1266 | for_each_net(net) {
|
| 1267 | for_each_netdev(net, dev) {
|
| 1268 | if (dev == last)
|
| 1269 | break;
|
| 1270 |
|
| 1271 | if (dev->flags & IFF_UP) {
|
| 1272 | nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
|
| 1273 | nb->notifier_call(nb, NETDEV_DOWN, dev);
|
| 1274 | }
|
| 1275 | nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
|
| 1276 | }
|
| 1277 | }
|
| 1278 |
|
| 1279 | raw_notifier_chain_unregister(&netdev_chain, nb);
|
| 1280 | goto unlock;
|
| 1281 | }
|
| 1282 |
|
| 1283 | |
| 1284 | * unregister_netdevice_notifier - unregister a network notifier block
|
| 1285 | * @nb: notifier
|
| 1286 | *
|
| 1287 | * Unregister a notifier previously registered by
|
| 1288 | * register_netdevice_notifier(). The notifier is unlinked into the
|
| 1289 | * kernel structures and may then be reused. A negative errno code
|
| 1290 | * is returned on a failure.
|
| 1291 |
|
| 1292 |
|
| 1293 | int unregister_netdevice_notifier(struct notifier_block *nb)
|
| 1294 | {
|
| 1295 | int err;
|
| 1296 |
|
| 1297 | rtnl_lock();
|
| 1298 | err = raw_notifier_chain_unregister(&netdev_chain, nb);
|
| 1299 | rtnl_unlock();
|
| 1300 | return err;
|
| 1301 | }
|
| 1302 |
|
| 1303 | |
| 1304 | * call_netdevice_notifiers - call all network notifier blocks
|
| 1305 | * @val: value passed unmodified to notifier function
|
| 1306 | * @dev: net_device pointer passed unmodified to notifier function
|
| 1307 | *
|
| 1308 | * Call all network notifier blocks. Parameters and return value
|
| 1309 | * are as for raw_notifier_call_chain().
|
| 1310 |
|
| 1311 |
|
| 1312 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
|
| 1313 | {
|
| 1314 | return raw_notifier_call_chain(&netdev_chain, val, dev);
|
| 1315 | }
|
| 1316 |
|
| 1317 |
|
| 1318 | static atomic_t netstamp_needed = ATOMIC_INIT(0);
|
| 1319 |
|
| 1320 | void net_enable_timestamp(void)
|
| 1321 | {
|
| 1322 | atomic_inc(&netstamp_needed);
|
| 1323 | }
|
| 1324 |
|
| 1325 | void net_disable_timestamp(void)
|
| 1326 | {
|
| 1327 | atomic_dec(&netstamp_needed);
|
| 1328 | }
|
| 1329 |
|
| 1330 | static inline void net_timestamp(struct sk_buff *skb)
|
| 1331 | {
|
| 1332 | if (atomic_read(&netstamp_needed))
|
| 1333 | __net_timestamp(skb);
|
| 1334 | else
|
| 1335 | skb->tstamp.tv64 = 0;
|
| 1336 | }
|
| 1337 |
|
| 1338 | |
| 1339 | * Support routine. Sends outgoing frames to any network
|
| 1340 | * taps currently in use.
|
| 1341 |
|
| 1342 |
|
| 1343 | static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
|
| 1344 | {
|
| 1345 | struct packet_type *ptype;
|
| 1346 |
|
| 1347 | #ifdef CONFIG_NET_CLS_ACT
|
| 1348 | if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
|
| 1349 | net_timestamp(skb);
|
| 1350 | #else
|
| 1351 | net_timestamp(skb);
|
| 1352 | #endif
|
| 1353 |
|
| 1354 | rcu_read_lock();
|
| 1355 | list_for_each_entry_rcu(ptype, &ptype_all, list) {
|
| 1356 | |
| 1357 | * they originated from - MvS ([email protected])
|
| 1358 |
|
| 1359 | if ((ptype->dev == dev || !ptype->dev) &&
|
| 1360 | (ptype->af_packet_priv == NULL ||
|
| 1361 | (struct sock *)ptype->af_packet_priv != skb->sk)) {
|
| 1362 | struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
|
| 1363 | if (!skb2)
|
| 1364 | break;
|
| 1365 |
|
| 1366 | |
| 1367 | set by sender, so that the second statement is
|
| 1368 | just protection against buggy protocols.
|
| 1369 |
|
| 1370 | skb_reset_mac_header(skb2);
|
| 1371 |
|
| 1372 | if (skb_network_header(skb2) < skb2->data ||
|
| 1373 | skb2->network_header > skb2->tail) {
|
| 1374 | if (net_ratelimit())
|
| 1375 | printk(KERN_CRIT "protocol %04x is "
|
| 1376 | "buggy, dev %s\n",
|
| 1377 | skb2->protocol, dev->name);
|
| 1378 | skb_reset_network_header(skb2);
|
| 1379 | }
|
| 1380 |
|
| 1381 | skb2->transport_header = skb2->network_header;
|
| 1382 | skb2->pkt_type = PACKET_OUTGOING;
|
| 1383 | ptype->func(skb2, skb->dev, ptype, skb->dev);
|
| 1384 | }
|
| 1385 | }
|
| 1386 | rcu_read_unlock();
|
| 1387 | }
|
| 1388 |
|
| 1389 |
|
| 1390 | static inline void __netif_reschedule(struct Qdisc *q)
|
| 1391 | {
|
| 1392 | struct softnet_data *sd;
|
| 1393 | unsigned long flags;
|
| 1394 |
|
| 1395 | local_irq_save(flags);
|
| 1396 | sd = &__get_cpu_var(softnet_data);
|
| 1397 | q->next_sched = sd->output_queue;
|
| 1398 | sd->output_queue = q;
|
| 1399 | raise_softirq_irqoff(NET_TX_SOFTIRQ);
|
| 1400 | local_irq_restore(flags);
|
| 1401 | }
|
| 1402 |
|
| 1403 | void __netif_schedule(struct Qdisc *q)
|
| 1404 | {
|
| 1405 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
|
| 1406 | __netif_reschedule(q);
|
| 1407 | }
|
| 1408 | EXPORT_SYMBOL(__netif_schedule);
|
| 1409 |
|
| 1410 | void dev_kfree_skb_irq(struct sk_buff *skb)
|
| 1411 | {
|
| 1412 | if (atomic_dec_and_test(&skb->users)) {
|
| 1413 | struct softnet_data *sd;
|
| 1414 | unsigned long flags;
|
| 1415 |
|
| 1416 | local_irq_save(flags);
|
| 1417 | sd = &__get_cpu_var(softnet_data);
|
| 1418 | skb->next = sd->completion_queue;
|
| 1419 | sd->completion_queue = skb;
|
| 1420 | raise_softirq_irqoff(NET_TX_SOFTIRQ);
|
| 1421 | local_irq_restore(flags);
|
| 1422 | }
|
| 1423 | }
|
| 1424 | EXPORT_SYMBOL(dev_kfree_skb_irq);
|
| 1425 |
|
| 1426 | void dev_kfree_skb_any(struct sk_buff *skb)
|
| 1427 | {
|
| 1428 | if (in_irq() || irqs_disabled())
|
| 1429 | dev_kfree_skb_irq(skb);
|
| 1430 | else
|
| 1431 | dev_kfree_skb(skb);
|
| 1432 | }
|
| 1433 | EXPORT_SYMBOL(dev_kfree_skb_any);
|
| 1434 |
|
| 1435 |
|
| 1436 | |
| 1437 | * netif_device_detach - mark device as removed
|
| 1438 | * @dev: network device
|
| 1439 | *
|
| 1440 | * Mark device as removed from system and therefore no longer available.
|
| 1441 |
|
| 1442 | void netif_device_detach(struct net_device *dev)
|
| 1443 | {
|
| 1444 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
|
| 1445 | netif_running(dev)) {
|
| 1446 | netif_tx_stop_all_queues(dev);
|
| 1447 | }
|
| 1448 | }
|
| 1449 | EXPORT_SYMBOL(netif_device_detach);
|
| 1450 |
|
| 1451 | |
| 1452 | * netif_device_attach - mark device as attached
|
| 1453 | * @dev: network device
|
| 1454 | *
|
| 1455 | * Mark device as attached from system and restart if needed.
|
| 1456 |
|
| 1457 | void netif_device_attach(struct net_device *dev)
|
| 1458 | {
|
| 1459 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
|
| 1460 | netif_running(dev)) {
|
| 1461 | netif_tx_wake_all_queues(dev);
|
| 1462 | __netdev_watchdog_up(dev);
|
| 1463 | }
|
| 1464 | }
|
| 1465 | EXPORT_SYMBOL(netif_device_attach);
|
| 1466 |
|
| 1467 | static bool can_checksum_protocol(unsigned long features, __be16 protocol)
|
| 1468 | {
|
| 1469 | return ((features & NETIF_F_GEN_CSUM) ||
|
| 1470 | ((features & NETIF_F_IP_CSUM) &&
|
| 1471 | protocol == htons(ETH_P_IP)) ||
|
| 1472 | ((features & NETIF_F_IPV6_CSUM) &&
|
| 1473 | protocol == htons(ETH_P_IPV6)) ||
|
| 1474 | ((features & NETIF_F_FCOE_CRC) &&
|
| 1475 | protocol == htons(ETH_P_FCOE)));
|
| 1476 | }
|
| 1477 |
|
| 1478 | static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
|
| 1479 | {
|
| 1480 | if (can_checksum_protocol(dev->features, skb->protocol))
|
| 1481 | return true;
|
| 1482 |
|
| 1483 | if (skb->protocol == htons(ETH_P_8021Q)) {
|
| 1484 | struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
|
| 1485 | if (can_checksum_protocol(dev->features & dev->vlan_features,
|
| 1486 | veh->h_vlan_encapsulated_proto))
|
| 1487 | return true;
|
| 1488 | }
|
| 1489 |
|
| 1490 | return false;
|
| 1491 | }
|
| 1492 |
|
| 1493 | |
| 1494 | * Invalidate hardware checksum when packet is to be mangled, and
|
| 1495 | * complete checksum manually on outgoing path.
|
| 1496 |
|
| 1497 | int skb_checksum_help(struct sk_buff *skb)
|
| 1498 | {
|
| 1499 | __wsum csum;
|
| 1500 | int ret = 0, offset;
|
| 1501 |
|
| 1502 | if (skb->ip_summed == CHECKSUM_COMPLETE)
|
| 1503 | goto out_set_summed;
|
| 1504 |
|
| 1505 | if (unlikely(skb_shinfo(skb)->gso_size)) {
|
| 1506 |
|
| 1507 | goto out_set_summed;
|
| 1508 | }
|
| 1509 |
|
| 1510 | offset = skb->csum_start - skb_headroom(skb);
|
| 1511 | BUG_ON(offset >= skb_headlen(skb));
|
| 1512 | csum = skb_checksum(skb, offset, skb->len - offset, 0);
|
| 1513 |
|
| 1514 | offset += skb->csum_offset;
|
| 1515 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
|
| 1516 |
|
| 1517 | if (skb_cloned(skb) &&
|
| 1518 | !skb_clone_writable(skb, offset + sizeof(__sum16))) {
|
| 1519 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
|
| 1520 | if (ret)
|
| 1521 | goto out;
|
| 1522 | }
|
| 1523 |
|
| 1524 | *(__sum16 *)(skb->data + offset) = csum_fold(csum);
|
| 1525 | out_set_summed:
|
| 1526 | skb->ip_summed = CHECKSUM_NONE;
|
| 1527 | out:
|
| 1528 | return ret;
|
| 1529 | }
|
| 1530 |
|
| 1531 | |
| 1532 | * skb_gso_segment - Perform segmentation on skb.
|
| 1533 | * @skb: buffer to segment
|
| 1534 | * @features: features for the output path (see dev->features)
|
| 1535 | *
|
| 1536 | * This function segments the given skb and returns a list of segments.
|
| 1537 | *
|
| 1538 | * It may return NULL if the skb requires no segmentation. This is
|
| 1539 | * only possible when GSO is used for verifying header integrity.
|
| 1540 |
|
| 1541 | struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
|
| 1542 | {
|
| 1543 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
|
| 1544 | struct packet_type *ptype;
|
| 1545 | __be16 type = skb->protocol;
|
| 1546 | int err;
|
| 1547 |
|
| 1548 | skb_reset_mac_header(skb);
|
| 1549 | skb->mac_len = skb->network_header - skb->mac_header;
|
| 1550 | __skb_pull(skb, skb->mac_len);
|
| 1551 |
|
| 1552 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
|
| 1553 | struct net_device *dev = skb->dev;
|
| 1554 | struct ethtool_drvinfo info = {};
|
| 1555 |
|
| 1556 | if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
|
| 1557 | dev->ethtool_ops->get_drvinfo(dev, &info);
|
| 1558 |
|
| 1559 | WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
|
| 1560 | "ip_summed=%d",
|
| 1561 | info.driver, dev ? dev->features : 0L,
|
| 1562 | skb->sk ? skb->sk->sk_route_caps : 0L,
|
| 1563 | skb->len, skb->data_len, skb->ip_summed);
|
| 1564 |
|
| 1565 | if (skb_header_cloned(skb) &&
|
| 1566 | (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
|
| 1567 | return ERR_PTR(err);
|
| 1568 | }
|
| 1569 |
|
| 1570 | rcu_read_lock();
|
| 1571 | list_for_each_entry_rcu(ptype,
|
| 1572 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
|
| 1573 | if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
|
| 1574 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
|
| 1575 | err = ptype->gso_send_check(skb);
|
| 1576 | segs = ERR_PTR(err);
|
| 1577 | if (err || skb_gso_ok(skb, features))
|
| 1578 | break;
|
| 1579 | __skb_push(skb, (skb->data -
|
| 1580 | skb_network_header(skb)));
|
| 1581 | }
|
| 1582 | segs = ptype->gso_segment(skb, features);
|
| 1583 | break;
|
| 1584 | }
|
| 1585 | }
|
| 1586 | rcu_read_unlock();
|
| 1587 |
|
| 1588 | __skb_push(skb, skb->data - skb_mac_header(skb));
|
| 1589 |
|
| 1590 | return segs;
|
| 1591 | }
|
| 1592 |
|
| 1593 | EXPORT_SYMBOL(skb_gso_segment);
|
| 1594 |
|
| 1595 |
|
| 1596 | #ifdef CONFIG_BUG
|
| 1597 | void netdev_rx_csum_fault(struct net_device *dev)
|
| 1598 | {
|
| 1599 | if (net_ratelimit()) {
|
| 1600 | printk(KERN_ERR "%s: hw csum failure.\n",
|
| 1601 | dev ? dev->name : "<unknown>");
|
| 1602 | dump_stack();
|
| 1603 | }
|
| 1604 | }
|
| 1605 | EXPORT_SYMBOL(netdev_rx_csum_fault);
|
| 1606 | #endif
|
| 1607 |
|
| 1608 | |
| 1609 | * 1. IOMMU is present and allows to map all the memory.
|
| 1610 | * 2. No high memory really exists on this machine.
|
| 1611 |
|
| 1612 |
|
| 1613 | static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
|
| 1614 | {
|
| 1615 | #ifdef CONFIG_HIGHMEM
|
| 1616 | int i;
|
| 1617 |
|
| 1618 | if (dev->features & NETIF_F_HIGHDMA)
|
| 1619 | return 0;
|
| 1620 |
|
| 1621 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
|
| 1622 | if (PageHighMem(skb_shinfo(skb)->frags[i].page))
|
| 1623 | return 1;
|
| 1624 |
|
| 1625 | #endif
|
| 1626 | return 0;
|
| 1627 | }
|
| 1628 |
|
| 1629 | struct dev_gso_cb {
|
| 1630 | void (*destructor)(struct sk_buff *skb);
|
| 1631 | };
|
| 1632 |
|
| 1633 | #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
|
| 1634 |
|
| 1635 | static void dev_gso_skb_destructor(struct sk_buff *skb)
|
| 1636 | {
|
| 1637 | struct dev_gso_cb *cb;
|
| 1638 |
|
| 1639 | do {
|
| 1640 | struct sk_buff *nskb = skb->next;
|
| 1641 |
|
| 1642 | skb->next = nskb->next;
|
| 1643 | nskb->next = NULL;
|
| 1644 | kfree_skb(nskb);
|
| 1645 | } while (skb->next);
|
| 1646 |
|
| 1647 | cb = DEV_GSO_CB(skb);
|
| 1648 | if (cb->destructor)
|
| 1649 | cb->destructor(skb);
|
| 1650 | }
|
| 1651 |
|
| 1652 | |
| 1653 | * dev_gso_segment - Perform emulated hardware segmentation on skb.
|
| 1654 | * @skb: buffer to segment
|
| 1655 | *
|
| 1656 | * This function segments the given skb and stores the list of segments
|
| 1657 | * in skb->next.
|
| 1658 |
|
| 1659 | static int dev_gso_segment(struct sk_buff *skb)
|
| 1660 | {
|
| 1661 | struct net_device *dev = skb->dev;
|
| 1662 | struct sk_buff *segs;
|
| 1663 | int features = dev->features & ~(illegal_highdma(dev, skb) ?
|
| 1664 | NETIF_F_SG : 0);
|
| 1665 |
|
| 1666 | segs = skb_gso_segment(skb, features);
|
| 1667 |
|
| 1668 |
|
| 1669 | if (!segs)
|
| 1670 | return 0;
|
| 1671 |
|
| 1672 | if (IS_ERR(segs))
|
| 1673 | return PTR_ERR(segs);
|
| 1674 |
|
| 1675 | skb->next = segs;
|
| 1676 | DEV_GSO_CB(skb)->destructor = skb->destructor;
|
| 1677 | skb->destructor = dev_gso_skb_destructor;
|
| 1678 |
|
| 1679 | return 0;
|
| 1680 | }
|
| 1681 |
|
| 1682 | int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
|
| 1683 | struct netdev_queue *txq)
|
| 1684 | {
|
| 1685 | const struct net_device_ops *ops = dev->netdev_ops;
|
| 1686 | int rc;
|
| 1687 |
|
| 1688 | if (likely(!skb->next)) {
|
| 1689 | if (!list_empty(&ptype_all))
|
| 1690 | dev_queue_xmit_nit(skb, dev);
|
| 1691 |
|
| 1692 | if (netif_needs_gso(dev, skb)) {
|
| 1693 | if (unlikely(dev_gso_segment(skb)))
|
| 1694 | goto out_kfree_skb;
|
| 1695 | if (skb->next)
|
| 1696 | goto gso;
|
| 1697 | }
|
| 1698 |
|
| 1699 | |
| 1700 | * If device doesnt need skb->dst, release it right now while
|
| 1701 | * its hot in this cpu cache
|
| 1702 |
|
| 1703 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
|
| 1704 | skb_dst_drop(skb);
|
| 1705 |
|
| 1706 | rc = ops->ndo_start_xmit(skb, dev);
|
| 1707 | if (rc == 0)
|
| 1708 | txq_trans_update(txq);
|
| 1709 | |
| 1710 | * TODO: if skb_orphan() was called by
|
| 1711 | * dev->hard_start_xmit() (for example, the unmodified
|
| 1712 | * igb driver does that; bnx2 doesn't), then
|
| 1713 | * skb_tx_software_timestamp() will be unable to send
|
| 1714 | * back the time stamp.
|
| 1715 | *
|
| 1716 | * How can this be prevented? Always create another
|
| 1717 | * reference to the socket before calling
|
| 1718 | * dev->hard_start_xmit()? Prevent that skb_orphan()
|
| 1719 | * does anything in dev->hard_start_xmit() by clearing
|
| 1720 | * the skb destructor before the call and restoring it
|
| 1721 | * afterwards, then doing the skb_orphan() ourselves?
|
| 1722 |
|
| 1723 | return rc;
|
| 1724 | }
|
| 1725 |
|
| 1726 | gso:
|
| 1727 | do {
|
| 1728 | struct sk_buff *nskb = skb->next;
|
| 1729 |
|
| 1730 | skb->next = nskb->next;
|
| 1731 | nskb->next = NULL;
|
| 1732 | rc = ops->ndo_start_xmit(nskb, dev);
|
| 1733 | if (unlikely(rc)) {
|
| 1734 | nskb->next = skb->next;
|
| 1735 | skb->next = nskb;
|
| 1736 | return rc;
|
| 1737 | }
|
| 1738 | txq_trans_update(txq);
|
| 1739 | if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
|
| 1740 | return NETDEV_TX_BUSY;
|
| 1741 | } while (skb->next);
|
| 1742 |
|
| 1743 | skb->destructor = DEV_GSO_CB(skb)->destructor;
|
| 1744 |
|
| 1745 | out_kfree_skb:
|
| 1746 | kfree_skb(skb);
|
| 1747 | return 0;
|
| 1748 | }
|
| 1749 |
|
| 1750 | static u32 skb_tx_hashrnd;
|
| 1751 |
|
| 1752 | u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
|
| 1753 | {
|
| 1754 | u32 hash;
|
| 1755 |
|
| 1756 | if (skb_rx_queue_recorded(skb)) {
|
| 1757 | hash = skb_get_rx_queue(skb);
|
| 1758 | while (unlikely (hash >= dev->real_num_tx_queues))
|
| 1759 | hash -= dev->real_num_tx_queues;
|
| 1760 | return hash;
|
| 1761 | }
|
| 1762 |
|
| 1763 | if (skb->sk && skb->sk->sk_hash)
|
| 1764 | hash = skb->sk->sk_hash;
|
| 1765 | else
|
| 1766 | hash = skb->protocol;
|
| 1767 |
|
| 1768 | hash = jhash_1word(hash, skb_tx_hashrnd);
|
| 1769 |
|
| 1770 | return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
|
| 1771 | }
|
| 1772 | EXPORT_SYMBOL(skb_tx_hash);
|
| 1773 |
|
| 1774 | static struct netdev_queue *dev_pick_tx(struct net_device *dev,
|
| 1775 | struct sk_buff *skb)
|
| 1776 | {
|
| 1777 | const struct net_device_ops *ops = dev->netdev_ops;
|
| 1778 | u16 queue_index = 0;
|
| 1779 |
|
| 1780 | if (ops->ndo_select_queue)
|
| 1781 | queue_index = ops->ndo_select_queue(dev, skb);
|
| 1782 | else if (dev->real_num_tx_queues > 1)
|
| 1783 | queue_index = skb_tx_hash(dev, skb);
|
| 1784 |
|
| 1785 | skb_set_queue_mapping(skb, queue_index);
|
| 1786 | return netdev_get_tx_queue(dev, queue_index);
|
| 1787 | }
|
| 1788 |
|
| 1789 | |
| 1790 | * dev_queue_xmit - transmit a buffer
|
| 1791 | * @skb: buffer to transmit
|
| 1792 | *
|
| 1793 | * Queue a buffer for transmission to a network device. The caller must
|
| 1794 | * have set the device and priority and built the buffer before calling
|
| 1795 | * this function. The function can be called from an interrupt.
|
| 1796 | *
|
| 1797 | * A negative errno code is returned on a failure. A success does not
|
| 1798 | * guarantee the frame will be transmitted as it may be dropped due
|
| 1799 | * to congestion or traffic shaping.
|
| 1800 | *
|
| 1801 | * -----------------------------------------------------------------------------------
|
| 1802 | * I notice this method can also return errors from the queue disciplines,
|
| 1803 | * including NET_XMIT_DROP, which is a positive value. So, errors can also
|
| 1804 | * be positive.
|
| 1805 | *
|
| 1806 | * Regardless of the return value, the skb is consumed, so it is currently
|
| 1807 | * difficult to retry a send to this method. (You can bump the ref count
|
| 1808 | * before sending to hold a reference for retry if you are careful.)
|
| 1809 | *
|
| 1810 | * When calling this method, interrupts MUST be enabled. This is because
|
| 1811 | * the BH enable code must have IRQs enabled so that it will not deadlock.
|
| 1812 | * --BLG
|
| 1813 |
|
| 1814 | int dev_queue_xmit(struct sk_buff *skb)
|
| 1815 | {
|
| 1816 | struct net_device *dev = skb->dev;
|
| 1817 | struct netdev_queue *txq;
|
| 1818 | struct Qdisc *q;
|
| 1819 | int rc = -ENOMEM;
|
| 1820 |
|
| 1821 |
|
| 1822 | if (netif_needs_gso(dev, skb))
|
| 1823 | goto gso;
|
| 1824 |
|
| 1825 | if (skb_has_frags(skb) &&
|
| 1826 | !(dev->features & NETIF_F_FRAGLIST) &&
|
| 1827 | __skb_linearize(skb))
|
| 1828 | goto out_kfree_skb;
|
| 1829 |
|
| 1830 | |
| 1831 | * or if at least one of fragments is in highmem and device
|
| 1832 | * does not support DMA from it.
|
| 1833 |
|
| 1834 | if (skb_shinfo(skb)->nr_frags &&
|
| 1835 | (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
|
| 1836 | __skb_linearize(skb))
|
| 1837 | goto out_kfree_skb;
|
| 1838 |
|
| 1839 | |
| 1840 | * checksumming for this protocol, complete checksumming here.
|
| 1841 |
|
| 1842 | if (skb->ip_summed == CHECKSUM_PARTIAL) {
|
| 1843 | skb_set_transport_header(skb, skb->csum_start -
|
| 1844 | skb_headroom(skb));
|
| 1845 | if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
|
| 1846 | goto out_kfree_skb;
|
| 1847 | }
|
| 1848 |
|
| 1849 | gso:
|
| 1850 | |
| 1851 | * stops preemption for RCU.
|
| 1852 |
|
| 1853 | rcu_read_lock_bh();
|
| 1854 |
|
| 1855 | txq = dev_pick_tx(dev, skb);
|
| 1856 | q = rcu_dereference(txq->qdisc);
|
| 1857 |
|
| 1858 | #ifdef CONFIG_NET_CLS_ACT
|
| 1859 | skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
|
| 1860 | #endif
|
| 1861 | if (q->enqueue) {
|
| 1862 | spinlock_t *root_lock = qdisc_lock(q);
|
| 1863 |
|
| 1864 | spin_lock(root_lock);
|
| 1865 |
|
| 1866 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
|
| 1867 | kfree_skb(skb);
|
| 1868 | rc = NET_XMIT_DROP;
|
| 1869 | } else {
|
| 1870 | rc = qdisc_enqueue_root(skb, q);
|
| 1871 | qdisc_run(q);
|
| 1872 | }
|
| 1873 | spin_unlock(root_lock);
|
| 1874 |
|
| 1875 | goto out;
|
| 1876 | }
|
| 1877 |
|
| 1878 | |
| 1879 | loopback, all the sorts of tunnels...
|
| 1880 |
|
| 1881 | Really, it is unlikely that netif_tx_lock protection is necessary
|
| 1882 | here. (f.e. loopback and IP tunnels are clean ignoring statistics
|
| 1883 | counters.)
|
| 1884 | However, it is possible, that they rely on protection
|
| 1885 | made by us here.
|
| 1886 |
|
| 1887 | Check this and shot the lock. It is not prone from deadlocks.
|
| 1888 | Either shot noqueue qdisc, it is even simpler 8)
|
| 1889 |
|
| 1890 | if (dev->flags & IFF_UP) {
|
| 1891 | int cpu = smp_processor_id();
|
| 1892 |
|
| 1893 | if (txq->xmit_lock_owner != cpu) {
|
| 1894 |
|
| 1895 | HARD_TX_LOCK(dev, txq, cpu);
|
| 1896 |
|
| 1897 | if (!netif_tx_queue_stopped(txq)) {
|
| 1898 | rc = 0;
|
| 1899 | if (!dev_hard_start_xmit(skb, dev, txq)) {
|
| 1900 | HARD_TX_UNLOCK(dev, txq);
|
| 1901 | goto out;
|
| 1902 | }
|
| 1903 | }
|
| 1904 | HARD_TX_UNLOCK(dev, txq);
|
| 1905 | if (net_ratelimit())
|
| 1906 | printk(KERN_CRIT "Virtual device %s asks to "
|
| 1907 | "queue packet!\n", dev->name);
|
| 1908 | } else {
|
| 1909 | |
| 1910 | * unfortunately
|
| 1911 | if (net_ratelimit())
|
| 1912 | printk(KERN_CRIT "Dead loop on virtual device "
|
| 1913 | "%s, fix it urgently!\n", dev->name);
|
| 1914 | }
|
| 1915 | }
|
| 1916 |
|
| 1917 | rc = -ENETDOWN;
|
| 1918 | rcu_read_unlock_bh();
|
| 1919 |
|
| 1920 | out_kfree_skb:
|
| 1921 | kfree_skb(skb);
|
| 1922 | return rc;
|
| 1923 | out:
|
| 1924 | rcu_read_unlock_bh();
|
| 1925 | return rc;
|
| 1926 | }
|
| 1927 |
|
| 1928 |
|
| 1929 | |
| 1930 | Receiver routines
|
| 1931 | =======================================================================
|
| 1932 |
|
| 1933 | int netdev_max_backlog __read_mostly = 1000;
|
| 1934 | int netdev_budget __read_mostly = 300;
|
| 1935 | int weight_p __read_mostly = 64;
|
| 1936 |
|
| 1937 | DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
|
| 1938 |
|
| 1939 |
|
| 1940 | |
| 1941 | * netif_rx - post buffer to the network code
|
| 1942 | * @skb: buffer to post
|
| 1943 | *
|
| 1944 | * This function receives a packet from a device driver and queues it for
|
| 1945 | * the upper (protocol) levels to process. It always succeeds. The buffer
|
| 1946 | * may be dropped during processing for congestion control or by the
|
| 1947 | * protocol layers.
|
| 1948 | *
|
| 1949 | * return values:
|
| 1950 | * NET_RX_SUCCESS (no congestion)
|
| 1951 | * NET_RX_DROP (packet was dropped)
|
| 1952 | *
|
| 1953 |
|
| 1954 |
|
| 1955 | int netif_rx(struct sk_buff *skb)
|
| 1956 | {
|
| 1957 | struct softnet_data *queue;
|
| 1958 | unsigned long flags;
|
| 1959 |
|
| 1960 |
|
| 1961 | if (netpoll_rx(skb))
|
| 1962 | return NET_RX_DROP;
|
| 1963 |
|
| 1964 | if (!skb->tstamp.tv64)
|
| 1965 | net_timestamp(skb);
|
| 1966 |
|
| 1967 | |
| 1968 | * The code is rearranged so that the path is the most
|
| 1969 | * short when CPU is congested, but is still operating.
|
| 1970 |
|
| 1971 | local_irq_save(flags);
|
| 1972 | queue = &__get_cpu_var(softnet_data);
|
| 1973 |
|
| 1974 | __get_cpu_var(netdev_rx_stat).total++;
|
| 1975 | if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
|
| 1976 | if (queue->input_pkt_queue.qlen) {
|
| 1977 | enqueue:
|
| 1978 | __skb_queue_tail(&queue->input_pkt_queue, skb);
|
| 1979 | local_irq_restore(flags);
|
| 1980 | return NET_RX_SUCCESS;
|
| 1981 | }
|
| 1982 |
|
| 1983 | napi_schedule(&queue->backlog);
|
| 1984 | goto enqueue;
|
| 1985 | }
|
| 1986 |
|
| 1987 | __get_cpu_var(netdev_rx_stat).dropped++;
|
| 1988 | local_irq_restore(flags);
|
| 1989 |
|
| 1990 | kfree_skb(skb);
|
| 1991 | return NET_RX_DROP;
|
| 1992 | }
|
| 1993 |
|
| 1994 | int netif_rx_ni(struct sk_buff *skb)
|
| 1995 | {
|
| 1996 | int err;
|
| 1997 |
|
| 1998 | preempt_disable();
|
| 1999 | err = netif_rx(skb);
|
| 2000 | if (local_softirq_pending())
|
| 2001 | do_softirq();
|
| 2002 | preempt_enable();
|
| 2003 |
|
| 2004 | return err;
|
| 2005 | }
|
| 2006 |
|
| 2007 | EXPORT_SYMBOL(netif_rx_ni);
|
| 2008 |
|
| 2009 | static void net_tx_action(struct softirq_action *h)
|
| 2010 | {
|
| 2011 | struct softnet_data *sd = &__get_cpu_var(softnet_data);
|
| 2012 |
|
| 2013 | if (sd->completion_queue) {
|
| 2014 | struct sk_buff *clist;
|
| 2015 |
|
| 2016 | local_irq_disable();
|
| 2017 | clist = sd->completion_queue;
|
| 2018 | sd->completion_queue = NULL;
|
| 2019 | local_irq_enable();
|
| 2020 |
|
| 2021 | while (clist) {
|
| 2022 | struct sk_buff *skb = clist;
|
| 2023 | clist = clist->next;
|
| 2024 |
|
| 2025 | WARN_ON(atomic_read(&skb->users));
|
| 2026 | __kfree_skb(skb);
|
| 2027 | }
|
| 2028 | }
|
| 2029 |
|
| 2030 | if (sd->output_queue) {
|
| 2031 | struct Qdisc *head;
|
| 2032 |
|
| 2033 | local_irq_disable();
|
| 2034 | head = sd->output_queue;
|
| 2035 | sd->output_queue = NULL;
|
| 2036 | local_irq_enable();
|
| 2037 |
|
| 2038 | while (head) {
|
| 2039 | struct Qdisc *q = head;
|
| 2040 | spinlock_t *root_lock;
|
| 2041 |
|
| 2042 | head = head->next_sched;
|
| 2043 |
|
| 2044 | root_lock = qdisc_lock(q);
|
| 2045 | if (spin_trylock(root_lock)) {
|
| 2046 | smp_mb__before_clear_bit();
|
| 2047 | clear_bit(__QDISC_STATE_SCHED,
|
| 2048 | &q->state);
|
| 2049 | qdisc_run(q);
|
| 2050 | spin_unlock(root_lock);
|
| 2051 | } else {
|
| 2052 | if (!test_bit(__QDISC_STATE_DEACTIVATED,
|
| 2053 | &q->state)) {
|
| 2054 | __netif_reschedule(q);
|
| 2055 | } else {
|
| 2056 | smp_mb__before_clear_bit();
|
| 2057 | clear_bit(__QDISC_STATE_SCHED,
|
| 2058 | &q->state);
|
| 2059 | }
|
| 2060 | }
|
| 2061 | }
|
| 2062 | }
|
| 2063 | }
|
| 2064 |
|
| 2065 | static inline int deliver_skb(struct sk_buff *skb,
|
| 2066 | struct packet_type *pt_prev,
|
| 2067 | struct net_device *orig_dev)
|
| 2068 | {
|
| 2069 | atomic_inc(&skb->users);
|
| 2070 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
|
| 2071 | }
|
| 2072 |
|
| 2073 | #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
|
| 2074 |
|
| 2075 | #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)
|
| 2076 |
|
| 2077 | int (*br_fdb_test_addr_hook)(struct net_device *dev,
|
| 2078 | unsigned char *addr) __read_mostly;
|
| 2079 | EXPORT_SYMBOL(br_fdb_test_addr_hook);
|
| 2080 | #endif
|
| 2081 |
|
| 2082 | |
| 2083 | * If bridge module is loaded call bridging hook.
|
| 2084 | * returns NULL if packet was consumed.
|
| 2085 |
|
| 2086 | struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
|
| 2087 | struct sk_buff *skb) __read_mostly;
|
| 2088 | EXPORT_SYMBOL(br_handle_frame_hook);
|
| 2089 |
|
| 2090 | static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
|
| 2091 | struct packet_type **pt_prev, int *ret,
|
| 2092 | struct net_device *orig_dev)
|
| 2093 | {
|
| 2094 | struct net_bridge_port *port;
|
| 2095 |
|
| 2096 | if (skb->pkt_type == PACKET_LOOPBACK ||
|
| 2097 | (port = rcu_dereference(skb->dev->br_port)) == NULL)
|
| 2098 | return skb;
|
| 2099 |
|
| 2100 | if (*pt_prev) {
|
| 2101 | *ret = deliver_skb(skb, *pt_prev, orig_dev);
|
| 2102 | *pt_prev = NULL;
|
| 2103 | }
|
| 2104 |
|
| 2105 | return br_handle_frame_hook(port, skb);
|
| 2106 | }
|
| 2107 | #else
|
| 2108 | #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
|
| 2109 | #endif
|
| 2110 |
|
| 2111 | #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
|
| 2112 | struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
|
| 2113 | EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
|
| 2114 |
|
| 2115 | static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
|
| 2116 | struct packet_type **pt_prev,
|
| 2117 | int *ret,
|
| 2118 | struct net_device *orig_dev)
|
| 2119 | {
|
| 2120 | if (skb->dev->macvlan_port == NULL)
|
| 2121 | return skb;
|
| 2122 |
|
| 2123 | if (*pt_prev) {
|
| 2124 | *ret = deliver_skb(skb, *pt_prev, orig_dev);
|
| 2125 | *pt_prev = NULL;
|
| 2126 | }
|
| 2127 | return macvlan_handle_frame_hook(skb);
|
| 2128 | }
|
| 2129 | #else
|
| 2130 | #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
|
| 2131 | #endif
|
| 2132 |
|
| 2133 | #ifdef CONFIG_NET_CLS_ACT
|
| 2134 | |
| 2135 | * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
|
| 2136 | * a compare and 2 stores extra right now if we dont have it on
|
| 2137 | * but have CONFIG_NET_CLS_ACT
|
| 2138 | * NOTE: This doesnt stop any functionality; if you dont have
|
| 2139 | * the ingress scheduler, you just cant add policies on ingress.
|
| 2140 | *
|
| 2141 |
|
| 2142 | static int ing_filter(struct sk_buff *skb)
|
| 2143 | {
|
| 2144 | struct net_device *dev = skb->dev;
|
| 2145 | u32 ttl = G_TC_RTTL(skb->tc_verd);
|
| 2146 | struct netdev_queue *rxq;
|
| 2147 | int result = TC_ACT_OK;
|
| 2148 | struct Qdisc *q;
|
| 2149 |
|
| 2150 | if (MAX_RED_LOOP < ttl++) {
|
| 2151 | printk(KERN_WARNING
|
| 2152 | "Redir loop detected Dropping packet (%d->%d)\n",
|
| 2153 | skb->iif, dev->ifindex);
|
| 2154 | return TC_ACT_SHOT;
|
| 2155 | }
|
| 2156 |
|
| 2157 | skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
|
| 2158 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
|
| 2159 |
|
| 2160 | rxq = &dev->rx_queue;
|
| 2161 |
|
| 2162 | q = rxq->qdisc;
|
| 2163 | if (q != &noop_qdisc) {
|
| 2164 | spin_lock(qdisc_lock(q));
|
| 2165 | if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
|
| 2166 | result = qdisc_enqueue_root(skb, q);
|
| 2167 | spin_unlock(qdisc_lock(q));
|
| 2168 | }
|
| 2169 |
|
| 2170 | return result;
|
| 2171 | }
|
| 2172 |
|
| 2173 | static inline struct sk_buff *handle_ing(struct sk_buff *skb,
|
| 2174 | struct packet_type **pt_prev,
|
| 2175 | int *ret, struct net_device *orig_dev)
|
| 2176 | {
|
| 2177 | if (skb->dev->rx_queue.qdisc == &noop_qdisc)
|
| 2178 | goto out;
|
| 2179 |
|
| 2180 | if (*pt_prev) {
|
| 2181 | *ret = deliver_skb(skb, *pt_prev, orig_dev);
|
| 2182 | *pt_prev = NULL;
|
| 2183 | } else {
|
| 2184 |
|
| 2185 | skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
|
| 2186 | }
|
| 2187 |
|
| 2188 | switch (ing_filter(skb)) {
|
| 2189 | case TC_ACT_SHOT:
|
| 2190 | case TC_ACT_STOLEN:
|
| 2191 | kfree_skb(skb);
|
| 2192 | return NULL;
|
| 2193 | }
|
| 2194 |
|
| 2195 | out:
|
| 2196 | skb->tc_verd = 0;
|
| 2197 | return skb;
|
| 2198 | }
|
| 2199 | #endif
|
| 2200 |
|
| 2201 | |
| 2202 | * netif_nit_deliver - deliver received packets to network taps
|
| 2203 | * @skb: buffer
|
| 2204 | *
|
| 2205 | * This function is used to deliver incoming packets to network
|
| 2206 | * taps. It should be used when the normal netif_receive_skb path
|
| 2207 | * is bypassed, for example because of VLAN acceleration.
|
| 2208 |
|
| 2209 | void netif_nit_deliver(struct sk_buff *skb)
|
| 2210 | {
|
| 2211 | struct packet_type *ptype;
|
| 2212 |
|
| 2213 | if (list_empty(&ptype_all))
|
| 2214 | return;
|
| 2215 |
|
| 2216 | skb_reset_network_header(skb);
|
| 2217 | skb_reset_transport_header(skb);
|
| 2218 | skb->mac_len = skb->network_header - skb->mac_header;
|
| 2219 |
|
| 2220 | rcu_read_lock();
|
| 2221 | list_for_each_entry_rcu(ptype, &ptype_all, list) {
|
| 2222 | if (!ptype->dev || ptype->dev == skb->dev)
|
| 2223 | deliver_skb(skb, ptype, skb->dev);
|
| 2224 | }
|
| 2225 | rcu_read_unlock();
|
| 2226 | }
|
| 2227 |
|
| 2228 | |
| 2229 | * netif_receive_skb - process receive buffer from network
|
| 2230 | * @skb: buffer to process
|
| 2231 | *
|
| 2232 | * netif_receive_skb() is the main receive data processing function.
|
| 2233 | * It always succeeds. The buffer may be dropped during processing
|
| 2234 | * for congestion control or by the protocol layers.
|
| 2235 | *
|
| 2236 | * This function may only be called from softirq context and interrupts
|
| 2237 | * should be enabled.
|
| 2238 | *
|
| 2239 | * Return values (usually ignored):
|
| 2240 | * NET_RX_SUCCESS: no congestion
|
| 2241 | * NET_RX_DROP: packet was dropped
|
| 2242 |
|
| 2243 | int netif_receive_skb(struct sk_buff *skb)
|
| 2244 | {
|
| 2245 | struct packet_type *ptype, *pt_prev;
|
| 2246 | struct net_device *orig_dev;
|
| 2247 | struct net_device *null_or_orig;
|
| 2248 | int ret = NET_RX_DROP;
|
| 2249 | __be16 type;
|
| 2250 |
|
| 2251 | if (skb->vlan_tci && vlan_hwaccel_do_receive(skb))
|
| 2252 | return NET_RX_SUCCESS;
|
| 2253 |
|
| 2254 |
|
| 2255 | if (netpoll_receive_skb(skb))
|
| 2256 | return NET_RX_DROP;
|
| 2257 |
|
| 2258 | if (!skb->tstamp.tv64)
|
| 2259 | net_timestamp(skb);
|
| 2260 |
|
| 2261 | if (!skb->iif)
|
| 2262 | skb->iif = skb->dev->ifindex;
|
| 2263 |
|
| 2264 | null_or_orig = NULL;
|
| 2265 | orig_dev = skb->dev;
|
| 2266 | if (orig_dev->master) {
|
| 2267 | if (skb_bond_should_drop(skb))
|
| 2268 | null_or_orig = orig_dev;
|
| 2269 | else
|
| 2270 | skb->dev = orig_dev->master;
|
| 2271 | }
|
| 2272 |
|
| 2273 | __get_cpu_var(netdev_rx_stat).total++;
|
| 2274 |
|
| 2275 | skb_reset_network_header(skb);
|
| 2276 | skb_reset_transport_header(skb);
|
| 2277 | skb->mac_len = skb->network_header - skb->mac_header;
|
| 2278 |
|
| 2279 | pt_prev = NULL;
|
| 2280 |
|
| 2281 | rcu_read_lock();
|
| 2282 |
|
| 2283 | #ifdef CONFIG_NET_CLS_ACT
|
| 2284 | if (skb->tc_verd & TC_NCLS) {
|
| 2285 | skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
|
| 2286 | goto ncls;
|
| 2287 | }
|
| 2288 | #endif
|
| 2289 |
|
| 2290 | list_for_each_entry_rcu(ptype, &ptype_all, list) {
|
| 2291 | if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
|
| 2292 | ptype->dev == orig_dev) {
|
| 2293 | if (pt_prev)
|
| 2294 | ret = deliver_skb(skb, pt_prev, orig_dev);
|
| 2295 | pt_prev = ptype;
|
| 2296 | }
|
| 2297 | }
|
| 2298 |
|
| 2299 | #ifdef CONFIG_NET_CLS_ACT
|
| 2300 | skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
|
| 2301 | if (!skb)
|
| 2302 | goto out;
|
| 2303 | ncls:
|
| 2304 | #endif
|
| 2305 |
|
| 2306 | skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
|
| 2307 | if (!skb)
|
| 2308 | goto out;
|
| 2309 | skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
|
| 2310 | if (!skb)
|
| 2311 | goto out;
|
| 2312 |
|
| 2313 | skb_orphan(skb);
|
| 2314 |
|
| 2315 | type = skb->protocol;
|
| 2316 | list_for_each_entry_rcu(ptype,
|
| 2317 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
|
| 2318 | if (ptype->type == type &&
|
| 2319 | (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
|
| 2320 | ptype->dev == orig_dev)) {
|
| 2321 | if (pt_prev)
|
| 2322 | ret = deliver_skb(skb, pt_prev, orig_dev);
|
| 2323 | pt_prev = ptype;
|
| 2324 | }
|
| 2325 | }
|
| 2326 |
|
| 2327 | if (pt_prev) {
|
| 2328 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
|
| 2329 | } else {
|
| 2330 | kfree_skb(skb);
|
| 2331 | |
| 2332 | * me how you were going to use this. :-)
|
| 2333 |
|
| 2334 | ret = NET_RX_DROP;
|
| 2335 | }
|
| 2336 |
|
| 2337 | out:
|
| 2338 | rcu_read_unlock();
|
| 2339 | return ret;
|
| 2340 | }
|
| 2341 |
|
| 2342 |
|
| 2343 | static void flush_backlog(void *arg)
|
| 2344 | {
|
| 2345 | struct net_device *dev = arg;
|
| 2346 | struct softnet_data *queue = &__get_cpu_var(softnet_data);
|
| 2347 | struct sk_buff *skb, *tmp;
|
| 2348 |
|
| 2349 | skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
|
| 2350 | if (skb->dev == dev) {
|
| 2351 | __skb_unlink(skb, &queue->input_pkt_queue);
|
| 2352 | kfree_skb(skb);
|
| 2353 | }
|
| 2354 | }
|
| 2355 |
|
| 2356 | static int napi_gro_complete(struct sk_buff *skb)
|
| 2357 | {
|
| 2358 | struct packet_type *ptype;
|
| 2359 | __be16 type = skb->protocol;
|
| 2360 | struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
|
| 2361 | int err = -ENOENT;
|
| 2362 |
|
| 2363 | if (NAPI_GRO_CB(skb)->count == 1) {
|
| 2364 | skb_shinfo(skb)->gso_size = 0;
|
| 2365 | goto out;
|
| 2366 | }
|
| 2367 |
|
| 2368 | rcu_read_lock();
|
| 2369 | list_for_each_entry_rcu(ptype, head, list) {
|
| 2370 | if (ptype->type != type || ptype->dev || !ptype->gro_complete)
|
| 2371 | continue;
|
| 2372 |
|
| 2373 | err = ptype->gro_complete(skb);
|
| 2374 | break;
|
| 2375 | }
|
| 2376 | rcu_read_unlock();
|
| 2377 |
|
| 2378 | if (err) {
|
| 2379 | WARN_ON(&ptype->list == head);
|
| 2380 | kfree_skb(skb);
|
| 2381 | return NET_RX_SUCCESS;
|
| 2382 | }
|
| 2383 |
|
| 2384 | out:
|
| 2385 | return netif_receive_skb(skb);
|
| 2386 | }
|
| 2387 |
|
| 2388 | void napi_gro_flush(struct napi_struct *napi)
|
| 2389 | {
|
| 2390 | struct sk_buff *skb, *next;
|
| 2391 |
|
| 2392 | for (skb = napi->gro_list; skb; skb = next) {
|
| 2393 | next = skb->next;
|
| 2394 | skb->next = NULL;
|
| 2395 | napi_gro_complete(skb);
|
| 2396 | }
|
| 2397 |
|
| 2398 | napi->gro_count = 0;
|
| 2399 | napi->gro_list = NULL;
|
| 2400 | }
|
| 2401 | EXPORT_SYMBOL(napi_gro_flush);
|
| 2402 |
|
| 2403 | int dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
|
| 2404 | {
|
| 2405 | struct sk_buff **pp = NULL;
|
| 2406 | struct packet_type *ptype;
|
| 2407 | __be16 type = skb->protocol;
|
| 2408 | struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
|
| 2409 | int same_flow;
|
| 2410 | int mac_len;
|
| 2411 | int ret;
|
| 2412 |
|
| 2413 | if (!(skb->dev->features & NETIF_F_GRO))
|
| 2414 | goto normal;
|
| 2415 |
|
| 2416 | if (skb_is_gso(skb) || skb_has_frags(skb))
|
| 2417 | goto normal;
|
| 2418 |
|
| 2419 | rcu_read_lock();
|
| 2420 | list_for_each_entry_rcu(ptype, head, list) {
|
| 2421 | if (ptype->type != type || ptype->dev || !ptype->gro_receive)
|
| 2422 | continue;
|
| 2423 |
|
| 2424 | skb_set_network_header(skb, skb_gro_offset(skb));
|
| 2425 | mac_len = skb->network_header - skb->mac_header;
|
| 2426 | skb->mac_len = mac_len;
|
| 2427 | NAPI_GRO_CB(skb)->same_flow = 0;
|
| 2428 | NAPI_GRO_CB(skb)->flush = 0;
|
| 2429 | NAPI_GRO_CB(skb)->free = 0;
|
| 2430 |
|
| 2431 | pp = ptype->gro_receive(&napi->gro_list, skb);
|
| 2432 | break;
|
| 2433 | }
|
| 2434 | rcu_read_unlock();
|
| 2435 |
|
| 2436 | if (&ptype->list == head)
|
| 2437 | goto normal;
|
| 2438 |
|
| 2439 | same_flow = NAPI_GRO_CB(skb)->same_flow;
|
| 2440 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
|
| 2441 |
|
| 2442 | if (pp) {
|
| 2443 | struct sk_buff *nskb = *pp;
|
| 2444 |
|
| 2445 | *pp = nskb->next;
|
| 2446 | nskb->next = NULL;
|
| 2447 | napi_gro_complete(nskb);
|
| 2448 | napi->gro_count--;
|
| 2449 | }
|
| 2450 |
|
| 2451 | if (same_flow)
|
| 2452 | goto ok;
|
| 2453 |
|
| 2454 | if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
|
| 2455 | goto normal;
|
| 2456 |
|
| 2457 | napi->gro_count++;
|
| 2458 | NAPI_GRO_CB(skb)->count = 1;
|
| 2459 | skb_shinfo(skb)->gso_size = skb_gro_len(skb);
|
| 2460 | skb->next = napi->gro_list;
|
| 2461 | napi->gro_list = skb;
|
| 2462 | ret = GRO_HELD;
|
| 2463 |
|
| 2464 | pull:
|
| 2465 | if (skb_headlen(skb) < skb_gro_offset(skb)) {
|
| 2466 | int grow = skb_gro_offset(skb) - skb_headlen(skb);
|
| 2467 |
|
| 2468 | BUG_ON(skb->end - skb->tail < grow);
|
| 2469 |
|
| 2470 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
|
| 2471 |
|
| 2472 | skb->tail += grow;
|
| 2473 | skb->data_len -= grow;
|
| 2474 |
|
| 2475 | skb_shinfo(skb)->frags[0].page_offset += grow;
|
| 2476 | skb_shinfo(skb)->frags[0].size -= grow;
|
| 2477 |
|
| 2478 | if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
|
| 2479 | put_page(skb_shinfo(skb)->frags[0].page);
|
| 2480 | memmove(skb_shinfo(skb)->frags,
|
| 2481 | skb_shinfo(skb)->frags + 1,
|
| 2482 | --skb_shinfo(skb)->nr_frags);
|
| 2483 | }
|
| 2484 | }
|
| 2485 |
|
| 2486 | ok:
|
| 2487 | return ret;
|
| 2488 |
|
| 2489 | normal:
|
| 2490 | ret = GRO_NORMAL;
|
| 2491 | goto pull;
|
| 2492 | }
|
| 2493 | EXPORT_SYMBOL(dev_gro_receive);
|
| 2494 |
|
| 2495 | static int __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
|
| 2496 | {
|
| 2497 | struct sk_buff *p;
|
| 2498 |
|
| 2499 | if (netpoll_rx_on(skb))
|
| 2500 | return GRO_NORMAL;
|
| 2501 |
|
| 2502 | for (p = napi->gro_list; p; p = p->next) {
|
| 2503 | NAPI_GRO_CB(p)->same_flow = (p->dev == skb->dev)
|
| 2504 | && !compare_ether_header(skb_mac_header(p),
|
| 2505 | skb_gro_mac_header(skb));
|
| 2506 | NAPI_GRO_CB(p)->flush = 0;
|
| 2507 | }
|
| 2508 |
|
| 2509 | return dev_gro_receive(napi, skb);
|
| 2510 | }
|
| 2511 |
|
| 2512 | int napi_skb_finish(int ret, struct sk_buff *skb)
|
| 2513 | {
|
| 2514 | int err = NET_RX_SUCCESS;
|
| 2515 |
|
| 2516 | switch (ret) {
|
| 2517 | case GRO_NORMAL:
|
| 2518 | return netif_receive_skb(skb);
|
| 2519 |
|
| 2520 | case GRO_DROP:
|
| 2521 | err = NET_RX_DROP;
|
| 2522 |
|
| 2523 |
|
| 2524 | case GRO_MERGED_FREE:
|
| 2525 | kfree_skb(skb);
|
| 2526 | break;
|
| 2527 | }
|
| 2528 |
|
| 2529 | return err;
|
| 2530 | }
|
| 2531 | EXPORT_SYMBOL(napi_skb_finish);
|
| 2532 |
|
| 2533 | void skb_gro_reset_offset(struct sk_buff *skb)
|
| 2534 | {
|
| 2535 | NAPI_GRO_CB(skb)->data_offset = 0;
|
| 2536 | NAPI_GRO_CB(skb)->frag0 = NULL;
|
| 2537 | NAPI_GRO_CB(skb)->frag0_len = 0;
|
| 2538 |
|
| 2539 | if (skb->mac_header == skb->tail &&
|
| 2540 | !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
|
| 2541 | NAPI_GRO_CB(skb)->frag0 =
|
| 2542 | page_address(skb_shinfo(skb)->frags[0].page) +
|
| 2543 | skb_shinfo(skb)->frags[0].page_offset;
|
| 2544 | NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
|
| 2545 | }
|
| 2546 | }
|
| 2547 | EXPORT_SYMBOL(skb_gro_reset_offset);
|
| 2548 |
|
| 2549 | int napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
|
| 2550 | {
|
| 2551 | skb_gro_reset_offset(skb);
|
| 2552 |
|
| 2553 | return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
|
| 2554 | }
|
| 2555 | EXPORT_SYMBOL(napi_gro_receive);
|
| 2556 |
|
| 2557 | void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
|
| 2558 | {
|
| 2559 | __skb_pull(skb, skb_headlen(skb));
|
| 2560 | skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
|
| 2561 |
|
| 2562 | napi->skb = skb;
|
| 2563 | }
|
| 2564 | EXPORT_SYMBOL(napi_reuse_skb);
|
| 2565 |
|
| 2566 | struct sk_buff *napi_get_frags(struct napi_struct *napi)
|
| 2567 | {
|
| 2568 | struct net_device *dev = napi->dev;
|
| 2569 | struct sk_buff *skb = napi->skb;
|
| 2570 |
|
| 2571 | if (!skb) {
|
| 2572 | skb = netdev_alloc_skb(dev, GRO_MAX_HEAD + NET_IP_ALIGN);
|
| 2573 | if (!skb)
|
| 2574 | goto out;
|
| 2575 |
|
| 2576 | skb_reserve(skb, NET_IP_ALIGN);
|
| 2577 |
|
| 2578 | napi->skb = skb;
|
| 2579 | }
|
| 2580 |
|
| 2581 | out:
|
| 2582 | return skb;
|
| 2583 | }
|
| 2584 | EXPORT_SYMBOL(napi_get_frags);
|
| 2585 |
|
| 2586 | int napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb, int ret)
|
| 2587 | {
|
| 2588 | int err = NET_RX_SUCCESS;
|
| 2589 |
|
| 2590 | switch (ret) {
|
| 2591 | case GRO_NORMAL:
|
| 2592 | case GRO_HELD:
|
| 2593 | skb->protocol = eth_type_trans(skb, napi->dev);
|
| 2594 |
|
| 2595 | if (ret == GRO_NORMAL)
|
| 2596 | return netif_receive_skb(skb);
|
| 2597 |
|
| 2598 | skb_gro_pull(skb, -ETH_HLEN);
|
| 2599 | break;
|
| 2600 |
|
| 2601 | case GRO_DROP:
|
| 2602 | err = NET_RX_DROP;
|
| 2603 |
|
| 2604 |
|
| 2605 | case GRO_MERGED_FREE:
|
| 2606 | napi_reuse_skb(napi, skb);
|
| 2607 | break;
|
| 2608 | }
|
| 2609 |
|
| 2610 | return err;
|
| 2611 | }
|
| 2612 | EXPORT_SYMBOL(napi_frags_finish);
|
| 2613 |
|
| 2614 | struct sk_buff *napi_frags_skb(struct napi_struct *napi)
|
| 2615 | {
|
| 2616 | struct sk_buff *skb = napi->skb;
|
| 2617 | struct ethhdr *eth;
|
| 2618 | unsigned int hlen;
|
| 2619 | unsigned int off;
|
| 2620 |
|
| 2621 | napi->skb = NULL;
|
| 2622 |
|
| 2623 | skb_reset_mac_header(skb);
|
| 2624 | skb_gro_reset_offset(skb);
|
| 2625 |
|
| 2626 | off = skb_gro_offset(skb);
|
| 2627 | hlen = off + sizeof(*eth);
|
| 2628 | eth = skb_gro_header_fast(skb, off);
|
| 2629 | if (skb_gro_header_hard(skb, hlen)) {
|
| 2630 | eth = skb_gro_header_slow(skb, hlen, off);
|
| 2631 | if (unlikely(!eth)) {
|
| 2632 | napi_reuse_skb(napi, skb);
|
| 2633 | skb = NULL;
|
| 2634 | goto out;
|
| 2635 | }
|
| 2636 | }
|
| 2637 |
|
| 2638 | skb_gro_pull(skb, sizeof(*eth));
|
| 2639 |
|
| 2640 | |
| 2641 | * This works because the only protocols we care about don't require
|
| 2642 | * special handling. We'll fix it up properly at the end.
|
| 2643 |
|
| 2644 | skb->protocol = eth->h_proto;
|
| 2645 |
|
| 2646 | out:
|
| 2647 | return skb;
|
| 2648 | }
|
| 2649 | EXPORT_SYMBOL(napi_frags_skb);
|
| 2650 |
|
| 2651 | int napi_gro_frags(struct napi_struct *napi)
|
| 2652 | {
|
| 2653 | struct sk_buff *skb = napi_frags_skb(napi);
|
| 2654 |
|
| 2655 | if (!skb)
|
| 2656 | return NET_RX_DROP;
|
| 2657 |
|
| 2658 | return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
|
| 2659 | }
|
| 2660 | EXPORT_SYMBOL(napi_gro_frags);
|
| 2661 |
|
| 2662 | static int process_backlog(struct napi_struct *napi, int quota)
|
| 2663 | {
|
| 2664 | int work = 0;
|
| 2665 | struct softnet_data *queue = &__get_cpu_var(softnet_data);
|
| 2666 | unsigned long start_time = jiffies;
|
| 2667 |
|
| 2668 | napi->weight = weight_p;
|
| 2669 | do {
|
| 2670 | struct sk_buff *skb;
|
| 2671 |
|
| 2672 | local_irq_disable();
|
| 2673 | skb = __skb_dequeue(&queue->input_pkt_queue);
|
| 2674 | if (!skb) {
|
| 2675 | __napi_complete(napi);
|
| 2676 | local_irq_enable();
|
| 2677 | break;
|
| 2678 | }
|
| 2679 | local_irq_enable();
|
| 2680 |
|
| 2681 | netif_receive_skb(skb);
|
| 2682 | } while (++work < quota && jiffies == start_time);
|
| 2683 |
|
| 2684 | return work;
|
| 2685 | }
|
| 2686 |
|
| 2687 | |
| 2688 | * __napi_schedule - schedule for receive
|
| 2689 | * @n: entry to schedule
|
| 2690 | *
|
| 2691 | * The entry's receive function will be scheduled to run
|
| 2692 |
|
| 2693 | void __napi_schedule(struct napi_struct *n)
|
| 2694 | {
|
| 2695 | unsigned long flags;
|
| 2696 |
|
| 2697 | local_irq_save(flags);
|
| 2698 | list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
|
| 2699 | __raise_softirq_irqoff(NET_RX_SOFTIRQ);
|
| 2700 | local_irq_restore(flags);
|
| 2701 | }
|
| 2702 | EXPORT_SYMBOL(__napi_schedule);
|
| 2703 |
|
| 2704 | void __napi_complete(struct napi_struct *n)
|
| 2705 | {
|
| 2706 | BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
|
| 2707 | BUG_ON(n->gro_list);
|
| 2708 |
|
| 2709 | list_del(&n->poll_list);
|
| 2710 | smp_mb__before_clear_bit();
|
| 2711 | clear_bit(NAPI_STATE_SCHED, &n->state);
|
| 2712 | }
|
| 2713 | EXPORT_SYMBOL(__napi_complete);
|
| 2714 |
|
| 2715 | void napi_complete(struct napi_struct *n)
|
| 2716 | {
|
| 2717 | unsigned long flags;
|
| 2718 |
|
| 2719 | |
| 2720 | * don't let napi dequeue from the cpu poll list
|
| 2721 | * just in case its running on a different cpu
|
| 2722 |
|
| 2723 | if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
|
| 2724 | return;
|
| 2725 |
|
| 2726 | napi_gro_flush(n);
|
| 2727 | local_irq_save(flags);
|
| 2728 | __napi_complete(n);
|
| 2729 | local_irq_restore(flags);
|
| 2730 | }
|
| 2731 | EXPORT_SYMBOL(napi_complete);
|
| 2732 |
|
| 2733 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
|
| 2734 | int (*poll)(struct napi_struct *, int), int weight)
|
| 2735 | {
|
| 2736 | INIT_LIST_HEAD(&napi->poll_list);
|
| 2737 | napi->gro_count = 0;
|
| 2738 | napi->gro_list = NULL;
|
| 2739 | napi->skb = NULL;
|
| 2740 | napi->poll = poll;
|
| 2741 | napi->weight = weight;
|
| 2742 | list_add(&napi->dev_list, &dev->napi_list);
|
| 2743 | napi->dev = dev;
|
| 2744 | #ifdef CONFIG_NETPOLL
|
| 2745 | spin_lock_init(&napi->poll_lock);
|
| 2746 | napi->poll_owner = -1;
|
| 2747 | #endif
|
| 2748 | set_bit(NAPI_STATE_SCHED, &napi->state);
|
| 2749 | }
|
| 2750 | EXPORT_SYMBOL(netif_napi_add);
|
| 2751 |
|
| 2752 | void netif_napi_del(struct napi_struct *napi)
|
| 2753 | {
|
| 2754 | struct sk_buff *skb, *next;
|
| 2755 |
|
| 2756 | list_del_init(&napi->dev_list);
|
| 2757 | napi_free_frags(napi);
|
| 2758 |
|
| 2759 | for (skb = napi->gro_list; skb; skb = next) {
|
| 2760 | next = skb->next;
|
| 2761 | skb->next = NULL;
|
| 2762 | kfree_skb(skb);
|
| 2763 | }
|
| 2764 |
|
| 2765 | napi->gro_list = NULL;
|
| 2766 | napi->gro_count = 0;
|
| 2767 | }
|
| 2768 | EXPORT_SYMBOL(netif_napi_del);
|
| 2769 |
|
| 2770 |
|
| 2771 | static void net_rx_action(struct softirq_action *h)
|
| 2772 | {
|
| 2773 | struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
|
| 2774 | unsigned long time_limit = jiffies + 2;
|
| 2775 | int budget = netdev_budget;
|
| 2776 | void *have;
|
| 2777 |
|
| 2778 | local_irq_disable();
|
| 2779 |
|
| 2780 | while (!list_empty(list)) {
|
| 2781 | struct napi_struct *n;
|
| 2782 | int work, weight;
|
| 2783 |
|
| 2784 | |
| 2785 | * Allow this to run for 2 jiffies since which will allow
|
| 2786 | * an average latency of 1.5/HZ.
|
| 2787 |
|
| 2788 | if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
|
| 2789 | goto softnet_break;
|
| 2790 |
|
| 2791 | local_irq_enable();
|
| 2792 |
|
| 2793 | |
| 2794 | * access is safe because interrupts can only add new
|
| 2795 | * entries to the tail of this list, and only ->poll()
|
| 2796 | * calls can remove this head entry from the list.
|
| 2797 |
|
| 2798 | n = list_entry(list->next, struct napi_struct, poll_list);
|
| 2799 |
|
| 2800 | have = netpoll_poll_lock(n);
|
| 2801 |
|
| 2802 | weight = n->weight;
|
| 2803 |
|
| 2804 | |
| 2805 | * with netpoll's poll_napi(). Only the entity which
|
| 2806 | * obtains the lock and sees NAPI_STATE_SCHED set will
|
| 2807 | * actually make the ->poll() call. Therefore we avoid
|
| 2808 | * accidently calling ->poll() when NAPI is not scheduled.
|
| 2809 |
|
| 2810 | work = 0;
|
| 2811 | if (test_bit(NAPI_STATE_SCHED, &n->state)) {
|
| 2812 | work = n->poll(n, weight);
|
| 2813 | trace_napi_poll(n);
|
| 2814 | }
|
| 2815 |
|
| 2816 | WARN_ON_ONCE(work > weight);
|
| 2817 |
|
| 2818 | budget -= work;
|
| 2819 |
|
| 2820 | local_irq_disable();
|
| 2821 |
|
| 2822 | |
| 2823 | * consume the entire weight. In such cases this code
|
| 2824 | * still "owns" the NAPI instance and therefore can
|
| 2825 | * move the instance around on the list at-will.
|
| 2826 |
|
| 2827 | if (unlikely(work == weight)) {
|
| 2828 | if (unlikely(napi_disable_pending(n)))
|
| 2829 | __napi_complete(n);
|
| 2830 | else
|
| 2831 | list_move_tail(&n->poll_list, list);
|
| 2832 | }
|
| 2833 |
|
| 2834 | netpoll_poll_unlock(have);
|
| 2835 | }
|
| 2836 | out:
|
| 2837 | local_irq_enable();
|
| 2838 |
|
| 2839 | #ifdef CONFIG_NET_DMA
|
| 2840 | |
| 2841 | * There may not be any more sk_buffs coming right now, so push
|
| 2842 | * any pending DMA copies to hardware
|
| 2843 |
|
| 2844 | dma_issue_pending_all();
|
| 2845 | #endif
|
| 2846 |
|
| 2847 | return;
|
| 2848 |
|
| 2849 | softnet_break:
|
| 2850 | __get_cpu_var(netdev_rx_stat).time_squeeze++;
|
| 2851 | __raise_softirq_irqoff(NET_RX_SOFTIRQ);
|
| 2852 | goto out;
|
| 2853 | }
|
| 2854 |
|
| 2855 | static gifconf_func_t * gifconf_list [NPROTO];
|
| 2856 |
|
| 2857 | |
| 2858 | * register_gifconf - register a SIOCGIF handler
|
| 2859 | * @family: Address family
|
| 2860 | * @gifconf: Function handler
|
| 2861 | *
|
| 2862 | * Register protocol dependent address dumping routines. The handler
|
| 2863 | * that is passed must not be freed or reused until it has been replaced
|
| 2864 | * by another handler.
|
| 2865 |
|
| 2866 | int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
|
| 2867 | {
|
| 2868 | if (family >= NPROTO)
|
| 2869 | return -EINVAL;
|
| 2870 | gifconf_list[family] = gifconf;
|
| 2871 | return 0;
|
| 2872 | }
|
| 2873 |
|
| 2874 |
|
| 2875 | |
| 2876 | * Map an interface index to its name (SIOCGIFNAME)
|
| 2877 |
|
| 2878 |
|
| 2879 | |
| 2880 | * We need this ioctl for efficient implementation of the
|
| 2881 | * if_indextoname() function required by the IPv6 API. Without
|
| 2882 | * it, we would have to search all the interfaces to find a
|
| 2883 | * match. --pb
|
| 2884 |
|
| 2885 |
|
| 2886 | static int dev_ifname(struct net *net, struct ifreq __user *arg)
|
| 2887 | {
|
| 2888 | struct net_device *dev;
|
| 2889 | struct ifreq ifr;
|
| 2890 |
|
| 2891 | |
| 2892 | * Fetch the caller's info block.
|
| 2893 |
|
| 2894 |
|
| 2895 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
|
| 2896 | return -EFAULT;
|
| 2897 |
|
| 2898 | read_lock(&dev_base_lock);
|
| 2899 | dev = __dev_get_by_index(net, ifr.ifr_ifindex);
|
| 2900 | if (!dev) {
|
| 2901 | read_unlock(&dev_base_lock);
|
| 2902 | return -ENODEV;
|
| 2903 | }
|
| 2904 |
|
| 2905 | strcpy(ifr.ifr_name, dev->name);
|
| 2906 | read_unlock(&dev_base_lock);
|
| 2907 |
|
| 2908 | if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
|
| 2909 | return -EFAULT;
|
| 2910 | return 0;
|
| 2911 | }
|
| 2912 |
|
| 2913 | |
| 2914 | * Perform a SIOCGIFCONF call. This structure will change
|
| 2915 | * size eventually, and there is nothing I can do about it.
|
| 2916 | * Thus we will need a 'compatibility mode'.
|
| 2917 |
|
| 2918 |
|
| 2919 | static int dev_ifconf(struct net *net, char __user *arg)
|
| 2920 | {
|
| 2921 | struct ifconf ifc;
|
| 2922 | struct net_device *dev;
|
| 2923 | char __user *pos;
|
| 2924 | int len;
|
| 2925 | int total;
|
| 2926 | int i;
|
| 2927 |
|
| 2928 | |
| 2929 | * Fetch the caller's info block.
|
| 2930 |
|
| 2931 |
|
| 2932 | if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
|
| 2933 | return -EFAULT;
|
| 2934 |
|
| 2935 | pos = ifc.ifc_buf;
|
| 2936 | len = ifc.ifc_len;
|
| 2937 |
|
| 2938 | |
| 2939 | * Loop over the interfaces, and write an info block for each.
|
| 2940 |
|
| 2941 |
|
| 2942 | total = 0;
|
| 2943 | for_each_netdev(net, dev) {
|
| 2944 | for (i = 0; i < NPROTO; i++) {
|
| 2945 | if (gifconf_list[i]) {
|
| 2946 | int done;
|
| 2947 | if (!pos)
|
| 2948 | done = gifconf_list[i](dev, NULL, 0);
|
| 2949 | else
|
| 2950 | done = gifconf_list[i](dev, pos + total,
|
| 2951 | len - total);
|
| 2952 | if (done < 0)
|
| 2953 | return -EFAULT;
|
| 2954 | total += done;
|
| 2955 | }
|
| 2956 | }
|
| 2957 | }
|
| 2958 |
|
| 2959 | |
| 2960 | * All done. Write the updated control block back to the caller.
|
| 2961 |
|
| 2962 | ifc.ifc_len = total;
|
| 2963 |
|
| 2964 | |
| 2965 | * Both BSD and Solaris return 0 here, so we do too.
|
| 2966 |
|
| 2967 | return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
|
| 2968 | }
|
| 2969 |
|
| 2970 | #ifdef CONFIG_PROC_FS
|
| 2971 | |
| 2972 | * This is invoked by the /proc filesystem handler to display a device
|
| 2973 | * in detail.
|
| 2974 |
|
| 2975 | void *dev_seq_start(struct seq_file *seq, loff_t *pos)
|
| 2976 | __acquires(dev_base_lock)
|
| 2977 | {
|
| 2978 | struct net *net = seq_file_net(seq);
|
| 2979 | loff_t off;
|
| 2980 | struct net_device *dev;
|
| 2981 |
|
| 2982 | read_lock(&dev_base_lock);
|
| 2983 | if (!*pos)
|
| 2984 | return SEQ_START_TOKEN;
|
| 2985 |
|
| 2986 | off = 1;
|
| 2987 | for_each_netdev(net, dev)
|
| 2988 | if (off++ == *pos)
|
| 2989 | return dev;
|
| 2990 |
|
| 2991 | return NULL;
|
| 2992 | }
|
| 2993 |
|
| 2994 | void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
|
| 2995 | {
|
| 2996 | struct net *net = seq_file_net(seq);
|
| 2997 | ++*pos;
|
| 2998 | return v == SEQ_START_TOKEN ?
|
| 2999 | first_net_device(net) : next_net_device((struct net_device *)v);
|
| 3000 | }
|
| 3001 |
|
| 3002 | void dev_seq_stop(struct seq_file *seq, void *v)
|
| 3003 | __releases(dev_base_lock)
|
| 3004 | {
|
| 3005 | read_unlock(&dev_base_lock);
|
| 3006 | }
|
| 3007 |
|
| 3008 | static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
|
| 3009 | {
|
| 3010 | const struct net_device_stats *stats = dev_get_stats(dev);
|
| 3011 |
|
| 3012 | seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
|
| 3013 | "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
|
| 3014 | dev->name, stats->rx_bytes, stats->rx_packets,
|
| 3015 | stats->rx_errors,
|
| 3016 | stats->rx_dropped + stats->rx_missed_errors,
|
| 3017 | stats->rx_fifo_errors,
|
| 3018 | stats->rx_length_errors + stats->rx_over_errors +
|
| 3019 | stats->rx_crc_errors + stats->rx_frame_errors,
|
| 3020 | stats->rx_compressed, stats->multicast,
|
| 3021 | stats->tx_bytes, stats->tx_packets,
|
| 3022 | stats->tx_errors, stats->tx_dropped,
|
| 3023 | stats->tx_fifo_errors, stats->collisions,
|
| 3024 | stats->tx_carrier_errors +
|
| 3025 | stats->tx_aborted_errors +
|
| 3026 | stats->tx_window_errors +
|
| 3027 | stats->tx_heartbeat_errors,
|
| 3028 | stats->tx_compressed);
|
| 3029 | }
|
| 3030 |
|
| 3031 | |
| 3032 | * Called from the PROCfs module. This now uses the new arbitrary sized
|
| 3033 | * /proc/net interface to create /proc/net/dev
|
| 3034 |
|
| 3035 | static int dev_seq_show(struct seq_file *seq, void *v)
|
| 3036 | {
|
| 3037 | if (v == SEQ_START_TOKEN)
|
| 3038 | seq_puts(seq, "Inter-| Receive "
|
| 3039 | " | Transmit\n"
|
| 3040 | " face |bytes packets errs drop fifo frame "
|
| 3041 | "compressed multicast|bytes packets errs "
|
| 3042 | "drop fifo colls carrier compressed\n");
|
| 3043 | else
|
| 3044 | dev_seq_printf_stats(seq, v);
|
| 3045 | return 0;
|
| 3046 | }
|
| 3047 |
|
| 3048 | static struct netif_rx_stats *softnet_get_online(loff_t *pos)
|
| 3049 | {
|
| 3050 | struct netif_rx_stats *rc = NULL;
|
| 3051 |
|
| 3052 | while (*pos < nr_cpu_ids)
|
| 3053 | if (cpu_online(*pos)) {
|
| 3054 | rc = &per_cpu(netdev_rx_stat, *pos);
|
| 3055 | break;
|
| 3056 | } else
|
| 3057 | ++*pos;
|
| 3058 | return rc;
|
| 3059 | }
|
| 3060 |
|
| 3061 | static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
|
| 3062 | {
|
| 3063 | return softnet_get_online(pos);
|
| 3064 | }
|
| 3065 |
|
| 3066 | static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
|
| 3067 | {
|
| 3068 | ++*pos;
|
| 3069 | return softnet_get_online(pos);
|
| 3070 | }
|
| 3071 |
|
| 3072 | static void softnet_seq_stop(struct seq_file *seq, void *v)
|
| 3073 | {
|
| 3074 | }
|
| 3075 |
|
| 3076 | static int softnet_seq_show(struct seq_file *seq, void *v)
|
| 3077 | {
|
| 3078 | struct netif_rx_stats *s = v;
|
| 3079 |
|
| 3080 | seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
|
| 3081 | s->total, s->dropped, s->time_squeeze, 0,
|
| 3082 | 0, 0, 0, 0,
|
| 3083 | s->cpu_collision );
|
| 3084 | return 0;
|
| 3085 | }
|
| 3086 |
|
| 3087 | static const struct seq_operations dev_seq_ops = {
|
| 3088 | .start = dev_seq_start,
|
| 3089 | .next = dev_seq_next,
|
| 3090 | .stop = dev_seq_stop,
|
| 3091 | .show = dev_seq_show,
|
| 3092 | };
|
| 3093 |
|
| 3094 | static int dev_seq_open(struct inode *inode, struct file *file)
|
| 3095 | {
|
| 3096 | return seq_open_net(inode, file, &dev_seq_ops,
|
| 3097 | sizeof(struct seq_net_private));
|
| 3098 | }
|
| 3099 |
|
| 3100 | static const struct file_operations dev_seq_fops = {
|
| 3101 | .owner = THIS_MODULE,
|
| 3102 | .open = dev_seq_open,
|
| 3103 | .read = seq_read,
|
| 3104 | .llseek = seq_lseek,
|
| 3105 | .release = seq_release_net,
|
| 3106 | };
|
| 3107 |
|
| 3108 | static const struct seq_operations softnet_seq_ops = {
|
| 3109 | .start = softnet_seq_start,
|
| 3110 | .next = softnet_seq_next,
|
| 3111 | .stop = softnet_seq_stop,
|
| 3112 | .show = softnet_seq_show,
|
| 3113 | };
|
| 3114 |
|
| 3115 | static int softnet_seq_open(struct inode *inode, struct file *file)
|
| 3116 | {
|
| 3117 | return seq_open(file, &softnet_seq_ops);
|
| 3118 | }
|
| 3119 |
|
| 3120 | static const struct file_operations softnet_seq_fops = {
|
| 3121 | .owner = THIS_MODULE,
|
| 3122 | .open = softnet_seq_open,
|
| 3123 | .read = seq_read,
|
| 3124 | .llseek = seq_lseek,
|
| 3125 | .release = seq_release,
|
| 3126 | };
|
| 3127 |
|
| 3128 | static void *ptype_get_idx(loff_t pos)
|
| 3129 | {
|
| 3130 | struct packet_type *pt = NULL;
|
| 3131 | loff_t i = 0;
|
| 3132 | int t;
|
| 3133 |
|
| 3134 | list_for_each_entry_rcu(pt, &ptype_all, list) {
|
| 3135 | if (i == pos)
|
| 3136 | return pt;
|
| 3137 | ++i;
|
| 3138 | }
|
| 3139 |
|
| 3140 | for (t = 0; t < PTYPE_HASH_SIZE; t++) {
|
| 3141 | list_for_each_entry_rcu(pt, &ptype_base[t], list) {
|
| 3142 | if (i == pos)
|
| 3143 | return pt;
|
| 3144 | ++i;
|
| 3145 | }
|
| 3146 | }
|
| 3147 | return NULL;
|
| 3148 | }
|
| 3149 |
|
| 3150 | static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
|
| 3151 | __acquires(RCU)
|
| 3152 | {
|
| 3153 | rcu_read_lock();
|
| 3154 | return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
|
| 3155 | }
|
| 3156 |
|
| 3157 | static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
|
| 3158 | {
|
| 3159 | struct packet_type *pt;
|
| 3160 | struct list_head *nxt;
|
| 3161 | int hash;
|
| 3162 |
|
| 3163 | ++*pos;
|
| 3164 | if (v == SEQ_START_TOKEN)
|
| 3165 | return ptype_get_idx(0);
|
| 3166 |
|
| 3167 | pt = v;
|
| 3168 | nxt = pt->list.next;
|
| 3169 | if (pt->type == htons(ETH_P_ALL)) {
|
| 3170 | if (nxt != &ptype_all)
|
| 3171 | goto found;
|
| 3172 | hash = 0;
|
| 3173 | nxt = ptype_base[0].next;
|
| 3174 | } else
|
| 3175 | hash = ntohs(pt->type) & PTYPE_HASH_MASK;
|
| 3176 |
|
| 3177 | while (nxt == &ptype_base[hash]) {
|
| 3178 | if (++hash >= PTYPE_HASH_SIZE)
|
| 3179 | return NULL;
|
| 3180 | nxt = ptype_base[hash].next;
|
| 3181 | }
|
| 3182 | found:
|
| 3183 | return list_entry(nxt, struct packet_type, list);
|
| 3184 | }
|
| 3185 |
|
| 3186 | static void ptype_seq_stop(struct seq_file *seq, void *v)
|
| 3187 | __releases(RCU)
|
| 3188 | {
|
| 3189 | rcu_read_unlock();
|
| 3190 | }
|
| 3191 |
|
| 3192 | static int ptype_seq_show(struct seq_file *seq, void *v)
|
| 3193 | {
|
| 3194 | struct packet_type *pt = v;
|
| 3195 |
|
| 3196 | if (v == SEQ_START_TOKEN)
|
| 3197 | seq_puts(seq, "Type Device Function\n");
|
| 3198 | else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
|
| 3199 | if (pt->type == htons(ETH_P_ALL))
|
| 3200 | seq_puts(seq, "ALL ");
|
| 3201 | else
|
| 3202 | seq_printf(seq, "%04x", ntohs(pt->type));
|
| 3203 |
|
| 3204 | seq_printf(seq, " %-8s %pF\n",
|
| 3205 | pt->dev ? pt->dev->name : "", pt->func);
|
| 3206 | }
|
| 3207 |
|
| 3208 | return 0;
|
| 3209 | }
|
| 3210 |
|
| 3211 | static const struct seq_operations ptype_seq_ops = {
|
| 3212 | .start = ptype_seq_start,
|
| 3213 | .next = ptype_seq_next,
|
| 3214 | .stop = ptype_seq_stop,
|
| 3215 | .show = ptype_seq_show,
|
| 3216 | };
|
| 3217 |
|
| 3218 | static int ptype_seq_open(struct inode *inode, struct file *file)
|
| 3219 | {
|
| 3220 | return seq_open_net(inode, file, &ptype_seq_ops,
|
| 3221 | sizeof(struct seq_net_private));
|
| 3222 | }
|
| 3223 |
|
| 3224 | static const struct file_operations ptype_seq_fops = {
|
| 3225 | .owner = THIS_MODULE,
|
| 3226 | .open = ptype_seq_open,
|
| 3227 | .read = seq_read,
|
| 3228 | .llseek = seq_lseek,
|
| 3229 | .release = seq_release_net,
|
| 3230 | };
|
| 3231 |
|
| 3232 |
|
| 3233 | static int __net_init dev_proc_net_init(struct net *net)
|
| 3234 | {
|
| 3235 | int rc = -ENOMEM;
|
| 3236 |
|
| 3237 | if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
|
| 3238 | goto out;
|
| 3239 | if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
|
| 3240 | goto out_dev;
|
| 3241 | if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
|
| 3242 | goto out_softnet;
|
| 3243 |
|
| 3244 | if (wext_proc_init(net))
|
| 3245 | goto out_ptype;
|
| 3246 | rc = 0;
|
| 3247 | out:
|
| 3248 | return rc;
|
| 3249 | out_ptype:
|
| 3250 | proc_net_remove(net, "ptype");
|
| 3251 | out_softnet:
|
| 3252 | proc_net_remove(net, "softnet_stat");
|
| 3253 | out_dev:
|
| 3254 | proc_net_remove(net, "dev");
|
| 3255 | goto out;
|
| 3256 | }
|
| 3257 |
|
| 3258 | static void __net_exit dev_proc_net_exit(struct net *net)
|
| 3259 | {
|
| 3260 | wext_proc_exit(net);
|
| 3261 |
|
| 3262 | proc_net_remove(net, "ptype");
|
| 3263 | proc_net_remove(net, "softnet_stat");
|
| 3264 | proc_net_remove(net, "dev");
|
| 3265 | }
|
| 3266 |
|
| 3267 | static struct pernet_operations __net_initdata dev_proc_ops = {
|
| 3268 | .init = dev_proc_net_init,
|
| 3269 | .exit = dev_proc_net_exit,
|
| 3270 | };
|
| 3271 |
|
| 3272 | static int __init dev_proc_init(void)
|
| 3273 | {
|
| 3274 | return register_pernet_subsys(&dev_proc_ops);
|
| 3275 | }
|
| 3276 | #else
|
| 3277 | #define dev_proc_init() 0
|
| 3278 | #endif
|
| 3279 |
|
| 3280 |
|
| 3281 | |
| 3282 | * netdev_set_master - set up master/slave pair
|
| 3283 | * @slave: slave device
|
| 3284 | * @master: new master device
|
| 3285 | *
|
| 3286 | * Changes the master device of the slave. Pass %NULL to break the
|
| 3287 | * bonding. The caller must hold the RTNL semaphore. On a failure
|
| 3288 | * a negative errno code is returned. On success the reference counts
|
| 3289 | * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
|
| 3290 | * function returns zero.
|
| 3291 |
|
| 3292 | int netdev_set_master(struct net_device *slave, struct net_device *master)
|
| 3293 | {
|
| 3294 | struct net_device *old = slave->master;
|
| 3295 |
|
| 3296 | ASSERT_RTNL();
|
| 3297 |
|
| 3298 | if (master) {
|
| 3299 | if (old)
|
| 3300 | return -EBUSY;
|
| 3301 | dev_hold(master);
|
| 3302 | }
|
| 3303 |
|
| 3304 | slave->master = master;
|
| 3305 |
|
| 3306 | synchronize_net();
|
| 3307 |
|
| 3308 | if (old)
|
| 3309 | dev_put(old);
|
| 3310 |
|
| 3311 | if (master)
|
| 3312 | slave->flags |= IFF_SLAVE;
|
| 3313 | else
|
| 3314 | slave->flags &= ~IFF_SLAVE;
|
| 3315 |
|
| 3316 | rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
|
| 3317 | return 0;
|
| 3318 | }
|
| 3319 |
|
| 3320 | static void dev_change_rx_flags(struct net_device *dev, int flags)
|
| 3321 | {
|
| 3322 | const struct net_device_ops *ops = dev->netdev_ops;
|
| 3323 |
|
| 3324 | if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
|
| 3325 | ops->ndo_change_rx_flags(dev, flags);
|
| 3326 | }
|
| 3327 |
|
| 3328 | static int __dev_set_promiscuity(struct net_device *dev, int inc)
|
| 3329 | {
|
| 3330 | unsigned short old_flags = dev->flags;
|
| 3331 | uid_t uid;
|
| 3332 | gid_t gid;
|
| 3333 |
|
| 3334 | ASSERT_RTNL();
|
| 3335 |
|
| 3336 | dev->flags |= IFF_PROMISC;
|
| 3337 | dev->promiscuity += inc;
|
| 3338 | if (dev->promiscuity == 0) {
|
| 3339 | |
| 3340 | * Avoid overflow.
|
| 3341 | * If inc causes overflow, untouch promisc and return error.
|
| 3342 |
|
| 3343 | if (inc < 0)
|
| 3344 | dev->flags &= ~IFF_PROMISC;
|
| 3345 | else {
|
| 3346 | dev->promiscuity -= inc;
|
| 3347 | printk(KERN_WARNING "%s: promiscuity touches roof, "
|
| 3348 | "set promiscuity failed, promiscuity feature "
|
| 3349 | "of device might be broken.\n", dev->name);
|
| 3350 | return -EOVERFLOW;
|
| 3351 | }
|
| 3352 | }
|
| 3353 | if (dev->flags != old_flags) {
|
| 3354 | printk(KERN_INFO "device %s %s promiscuous mode\n",
|
| 3355 | dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
|
| 3356 | "left");
|
| 3357 | if (audit_enabled) {
|
| 3358 | current_uid_gid(&uid, &gid);
|
| 3359 | audit_log(current->audit_context, GFP_ATOMIC,
|
| 3360 | AUDIT_ANOM_PROMISCUOUS,
|
| 3361 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
|
| 3362 | dev->name, (dev->flags & IFF_PROMISC),
|
| 3363 | (old_flags & IFF_PROMISC),
|
| 3364 | audit_get_loginuid(current),
|
| 3365 | uid, gid,
|
| 3366 | audit_get_sessionid(current));
|
| 3367 | }
|
| 3368 |
|
| 3369 | dev_change_rx_flags(dev, IFF_PROMISC);
|
| 3370 | }
|
| 3371 | return 0;
|
| 3372 | }
|
| 3373 |
|
| 3374 | |
| 3375 | * dev_set_promiscuity - update promiscuity count on a device
|
| 3376 | * @dev: device
|
| 3377 | * @inc: modifier
|
| 3378 | *
|
| 3379 | * Add or remove promiscuity from a device. While the count in the device
|
| 3380 | * remains above zero the interface remains promiscuous. Once it hits zero
|
| 3381 | * the device reverts back to normal filtering operation. A negative inc
|
| 3382 | * value is used to drop promiscuity on the device.
|
| 3383 | * Return 0 if successful or a negative errno code on error.
|
| 3384 |
|
| 3385 | int dev_set_promiscuity(struct net_device *dev, int inc)
|
| 3386 | {
|
| 3387 | unsigned short old_flags = dev->flags;
|
| 3388 | int err;
|
| 3389 |
|
| 3390 | err = __dev_set_promiscuity(dev, inc);
|
| 3391 | if (err < 0)
|
| 3392 | return err;
|
| 3393 | if (dev->flags != old_flags)
|
| 3394 | dev_set_rx_mode(dev);
|
| 3395 | return err;
|
| 3396 | }
|
| 3397 |
|
| 3398 | |
| 3399 | * dev_set_allmulti - update allmulti count on a device
|
| 3400 | * @dev: device
|
| 3401 | * @inc: modifier
|
| 3402 | *
|
| 3403 | * Add or remove reception of all multicast frames to a device. While the
|
| 3404 | * count in the device remains above zero the interface remains listening
|
| 3405 | * to all interfaces. Once it hits zero the device reverts back to normal
|
| 3406 | * filtering operation. A negative @inc value is used to drop the counter
|
| 3407 | * when releasing a resource needing all multicasts.
|
| 3408 | * Return 0 if successful or a negative errno code on error.
|
| 3409 |
|
| 3410 |
|
| 3411 | int dev_set_allmulti(struct net_device *dev, int inc)
|
| 3412 | {
|
| 3413 | unsigned short old_flags = dev->flags;
|
| 3414 |
|
| 3415 | ASSERT_RTNL();
|
| 3416 |
|
| 3417 | dev->flags |= IFF_ALLMULTI;
|
| 3418 | dev->allmulti += inc;
|
| 3419 | if (dev->allmulti == 0) {
|
| 3420 | |
| 3421 | * Avoid overflow.
|
| 3422 | * If inc causes overflow, untouch allmulti and return error.
|
| 3423 |
|
| 3424 | if (inc < 0)
|
| 3425 | dev->flags &= ~IFF_ALLMULTI;
|
| 3426 | else {
|
| 3427 | dev->allmulti -= inc;
|
| 3428 | printk(KERN_WARNING "%s: allmulti touches roof, "
|
| 3429 | "set allmulti failed, allmulti feature of "
|
| 3430 | "device might be broken.\n", dev->name);
|
| 3431 | return -EOVERFLOW;
|
| 3432 | }
|
| 3433 | }
|
| 3434 | if (dev->flags ^ old_flags) {
|
| 3435 | dev_change_rx_flags(dev, IFF_ALLMULTI);
|
| 3436 | dev_set_rx_mode(dev);
|
| 3437 | }
|
| 3438 | return 0;
|
| 3439 | }
|
| 3440 |
|
| 3441 | |
| 3442 | * Upload unicast and multicast address lists to device and
|
| 3443 | * configure RX filtering. When the device doesn't support unicast
|
| 3444 | * filtering it is put in promiscuous mode while unicast addresses
|
| 3445 | * are present.
|
| 3446 |
|
| 3447 | void __dev_set_rx_mode(struct net_device *dev)
|
| 3448 | {
|
| 3449 | const struct net_device_ops *ops = dev->netdev_ops;
|
| 3450 |
|
| 3451 |
|
| 3452 | if (!(dev->flags&IFF_UP))
|
| 3453 | return;
|
| 3454 |
|
| 3455 | if (!netif_device_present(dev))
|
| 3456 | return;
|
| 3457 |
|
| 3458 | if (ops->ndo_set_rx_mode)
|
| 3459 | ops->ndo_set_rx_mode(dev);
|
| 3460 | else {
|
| 3461 | |
| 3462 | * therefore calling __dev_set_promiscuity here is safe.
|
| 3463 |
|
| 3464 | if (dev->uc.count > 0 && !dev->uc_promisc) {
|
| 3465 | __dev_set_promiscuity(dev, 1);
|
| 3466 | dev->uc_promisc = 1;
|
| 3467 | } else if (dev->uc.count == 0 && dev->uc_promisc) {
|
| 3468 | __dev_set_promiscuity(dev, -1);
|
| 3469 | dev->uc_promisc = 0;
|
| 3470 | }
|
| 3471 |
|
| 3472 | if (ops->ndo_set_multicast_list)
|
| 3473 | ops->ndo_set_multicast_list(dev);
|
| 3474 | }
|
| 3475 | }
|
| 3476 |
|
| 3477 | void dev_set_rx_mode(struct net_device *dev)
|
| 3478 | {
|
| 3479 | netif_addr_lock_bh(dev);
|
| 3480 | __dev_set_rx_mode(dev);
|
| 3481 | netif_addr_unlock_bh(dev);
|
| 3482 | }
|
| 3483 |
|
| 3484 |
|
| 3485 |
|
| 3486 | static int __hw_addr_add(struct netdev_hw_addr_list *list, unsigned char *addr,
|
| 3487 | int addr_len, unsigned char addr_type)
|
| 3488 | {
|
| 3489 | struct netdev_hw_addr *ha;
|
| 3490 | int alloc_size;
|
| 3491 |
|
| 3492 | if (addr_len > MAX_ADDR_LEN)
|
| 3493 | return -EINVAL;
|
| 3494 |
|
| 3495 | list_for_each_entry(ha, &list->list, list) {
|
| 3496 | if (!memcmp(ha->addr, addr, addr_len) &&
|
| 3497 | ha->type == addr_type) {
|
| 3498 | ha->refcount++;
|
| 3499 | return 0;
|
| 3500 | }
|
| 3501 | }
|
| 3502 |
|
| 3503 |
|
| 3504 | alloc_size = sizeof(*ha);
|
| 3505 | if (alloc_size < L1_CACHE_BYTES)
|
| 3506 | alloc_size = L1_CACHE_BYTES;
|
| 3507 | ha = kmalloc(alloc_size, GFP_ATOMIC);
|
| 3508 | if (!ha)
|
| 3509 | return -ENOMEM;
|
| 3510 | memcpy(ha->addr, addr, addr_len);
|
| 3511 | ha->type = addr_type;
|
| 3512 | ha->refcount = 1;
|
| 3513 | ha->synced = false;
|
| 3514 | list_add_tail_rcu(&ha->list, &list->list);
|
| 3515 | list->count++;
|
| 3516 | return 0;
|
| 3517 | }
|
| 3518 |
|
| 3519 | static void ha_rcu_free(struct rcu_head *head)
|
| 3520 | {
|
| 3521 | struct netdev_hw_addr *ha;
|
| 3522 |
|
| 3523 | ha = container_of(head, struct netdev_hw_addr, rcu_head);
|
| 3524 | kfree(ha);
|
| 3525 | }
|
| 3526 |
|
| 3527 | static int __hw_addr_del(struct netdev_hw_addr_list *list, unsigned char *addr,
|
| 3528 | int addr_len, unsigned char addr_type)
|
| 3529 | {
|
| 3530 | struct netdev_hw_addr *ha;
|
| 3531 |
|
| 3532 | list_for_each_entry(ha, &list->list, list) {
|
| 3533 | if (!memcmp(ha->addr, addr, addr_len) &&
|
| 3534 | (ha->type == addr_type || !addr_type)) {
|
| 3535 | if (--ha->refcount)
|
| 3536 | return 0;
|
| 3537 | list_del_rcu(&ha->list);
|
| 3538 | call_rcu(&ha->rcu_head, ha_rcu_free);
|
| 3539 | list->count--;
|
| 3540 | return 0;
|
| 3541 | }
|
| 3542 | }
|
| 3543 | return -ENOENT;
|
| 3544 | }
|
| 3545 |
|
| 3546 | static int __hw_addr_add_multiple(struct netdev_hw_addr_list *to_list,
|
| 3547 | struct netdev_hw_addr_list *from_list,
|
| 3548 | int addr_len,
|
| 3549 | unsigned char addr_type)
|
| 3550 | {
|
| 3551 | int err;
|
| 3552 | struct netdev_hw_addr *ha, *ha2;
|
| 3553 | unsigned char type;
|
| 3554 |
|
| 3555 | list_for_each_entry(ha, &from_list->list, list) {
|
| 3556 | type = addr_type ? addr_type : ha->type;
|
| 3557 | err = __hw_addr_add(to_list, ha->addr, addr_len, type);
|
| 3558 | if (err)
|
| 3559 | goto unroll;
|
| 3560 | }
|
| 3561 | return 0;
|
| 3562 |
|
| 3563 | unroll:
|
| 3564 | list_for_each_entry(ha2, &from_list->list, list) {
|
| 3565 | if (ha2 == ha)
|
| 3566 | break;
|
| 3567 | type = addr_type ? addr_type : ha2->type;
|
| 3568 | __hw_addr_del(to_list, ha2->addr, addr_len, type);
|
| 3569 | }
|
| 3570 | return err;
|
| 3571 | }
|
| 3572 |
|
| 3573 | static void __hw_addr_del_multiple(struct netdev_hw_addr_list *to_list,
|
| 3574 | struct netdev_hw_addr_list *from_list,
|
| 3575 | int addr_len,
|
| 3576 | unsigned char addr_type)
|
| 3577 | {
|
| 3578 | struct netdev_hw_addr *ha;
|
| 3579 | unsigned char type;
|
| 3580 |
|
| 3581 | list_for_each_entry(ha, &from_list->list, list) {
|
| 3582 | type = addr_type ? addr_type : ha->type;
|
| 3583 | __hw_addr_del(to_list, ha->addr, addr_len, addr_type);
|
| 3584 | }
|
| 3585 | }
|
| 3586 |
|
| 3587 | static int __hw_addr_sync(struct netdev_hw_addr_list *to_list,
|
| 3588 | struct netdev_hw_addr_list *from_list,
|
| 3589 | int addr_len)
|
| 3590 | {
|
| 3591 | int err = 0;
|
| 3592 | struct netdev_hw_addr *ha, *tmp;
|
| 3593 |
|
| 3594 | list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
|
| 3595 | if (!ha->synced) {
|
| 3596 | err = __hw_addr_add(to_list, ha->addr,
|
| 3597 | addr_len, ha->type);
|
| 3598 | if (err)
|
| 3599 | break;
|
| 3600 | ha->synced = true;
|
| 3601 | ha->refcount++;
|
| 3602 | } else if (ha->refcount == 1) {
|
| 3603 | __hw_addr_del(to_list, ha->addr, addr_len, ha->type);
|
| 3604 | __hw_addr_del(from_list, ha->addr, addr_len, ha->type);
|
| 3605 | }
|
| 3606 | }
|
| 3607 | return err;
|
| 3608 | }
|
| 3609 |
|
| 3610 | static void __hw_addr_unsync(struct netdev_hw_addr_list *to_list,
|
| 3611 | struct netdev_hw_addr_list *from_list,
|
| 3612 | int addr_len)
|
| 3613 | {
|
| 3614 | struct netdev_hw_addr *ha, *tmp;
|
| 3615 |
|
| 3616 | list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
|
| 3617 | if (ha->synced) {
|
| 3618 | __hw_addr_del(to_list, ha->addr,
|
| 3619 | addr_len, ha->type);
|
| 3620 | ha->synced = false;
|
| 3621 | __hw_addr_del(from_list, ha->addr,
|
| 3622 | addr_len, ha->type);
|
| 3623 | }
|
| 3624 | }
|
| 3625 | }
|
| 3626 |
|
| 3627 | static void __hw_addr_flush(struct netdev_hw_addr_list *list)
|
| 3628 | {
|
| 3629 | struct netdev_hw_addr *ha, *tmp;
|
| 3630 |
|
| 3631 | list_for_each_entry_safe(ha, tmp, &list->list, list) {
|
| 3632 | list_del_rcu(&ha->list);
|
| 3633 | call_rcu(&ha->rcu_head, ha_rcu_free);
|
| 3634 | }
|
| 3635 | list->count = 0;
|
| 3636 | }
|
| 3637 |
|
| 3638 | static void __hw_addr_init(struct netdev_hw_addr_list *list)
|
| 3639 | {
|
| 3640 | INIT_LIST_HEAD(&list->list);
|
| 3641 | list->count = 0;
|
| 3642 | }
|
| 3643 |
|
| 3644 |
|
| 3645 |
|
| 3646 | static void dev_addr_flush(struct net_device *dev)
|
| 3647 | {
|
| 3648 |
|
| 3649 |
|
| 3650 | __hw_addr_flush(&dev->dev_addrs);
|
| 3651 | dev->dev_addr = NULL;
|
| 3652 | }
|
| 3653 |
|
| 3654 | static int dev_addr_init(struct net_device *dev)
|
| 3655 | {
|
| 3656 | unsigned char addr[MAX_ADDR_LEN];
|
| 3657 | struct netdev_hw_addr *ha;
|
| 3658 | int err;
|
| 3659 |
|
| 3660 |
|
| 3661 |
|
| 3662 | __hw_addr_init(&dev->dev_addrs);
|
| 3663 | memset(addr, 0, sizeof(addr));
|
| 3664 | err = __hw_addr_add(&dev->dev_addrs, addr, sizeof(addr),
|
| 3665 | NETDEV_HW_ADDR_T_LAN);
|
| 3666 | if (!err) {
|
| 3667 | |
| 3668 | * Get the first (previously created) address from the list
|
| 3669 | * and set dev_addr pointer to this location.
|
| 3670 |
|
| 3671 | ha = list_first_entry(&dev->dev_addrs.list,
|
| 3672 | struct netdev_hw_addr, list);
|
| 3673 | dev->dev_addr = ha->addr;
|
| 3674 | }
|
| 3675 | return err;
|
| 3676 | }
|
| 3677 |
|
| 3678 | |
| 3679 | * dev_addr_add - Add a device address
|
| 3680 | * @dev: device
|
| 3681 | * @addr: address to add
|
| 3682 | * @addr_type: address type
|
| 3683 | *
|
| 3684 | * Add a device address to the device or increase the reference count if
|
| 3685 | * it already exists.
|
| 3686 | *
|
| 3687 | * The caller must hold the rtnl_mutex.
|
| 3688 |
|
| 3689 | int dev_addr_add(struct net_device *dev, unsigned char *addr,
|
| 3690 | unsigned char addr_type)
|
| 3691 | {
|
| 3692 | int err;
|
| 3693 |
|
| 3694 | ASSERT_RTNL();
|
| 3695 |
|
| 3696 | err = __hw_addr_add(&dev->dev_addrs, addr, dev->addr_len, addr_type);
|
| 3697 | if (!err)
|
| 3698 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
|
| 3699 | return err;
|
| 3700 | }
|
| 3701 | EXPORT_SYMBOL(dev_addr_add);
|
| 3702 |
|
| 3703 | |
| 3704 | * dev_addr_del - Release a device address.
|
| 3705 | * @dev: device
|
| 3706 | * @addr: address to delete
|
| 3707 | * @addr_type: address type
|
| 3708 | *
|
| 3709 | * Release reference to a device address and remove it from the device
|
| 3710 | * if the reference count drops to zero.
|
| 3711 | *
|
| 3712 | * The caller must hold the rtnl_mutex.
|
| 3713 |
|
| 3714 | int dev_addr_del(struct net_device *dev, unsigned char *addr,
|
| 3715 | unsigned char addr_type)
|
| 3716 | {
|
| 3717 | int err;
|
| 3718 | struct netdev_hw_addr *ha;
|
| 3719 |
|
| 3720 | ASSERT_RTNL();
|
| 3721 |
|
| 3722 | |
| 3723 | * We can not remove the first address from the list because
|
| 3724 | * dev->dev_addr points to that.
|
| 3725 |
|
| 3726 | ha = list_first_entry(&dev->dev_addrs.list,
|
| 3727 | struct netdev_hw_addr, list);
|
| 3728 | if (ha->addr == dev->dev_addr && ha->refcount == 1)
|
| 3729 | return -ENOENT;
|
| 3730 |
|
| 3731 | err = __hw_addr_del(&dev->dev_addrs, addr, dev->addr_len,
|
| 3732 | addr_type);
|
| 3733 | if (!err)
|
| 3734 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
|
| 3735 | return err;
|
| 3736 | }
|
| 3737 | EXPORT_SYMBOL(dev_addr_del);
|
| 3738 |
|
| 3739 | |
| 3740 | * dev_addr_add_multiple - Add device addresses from another device
|
| 3741 | * @to_dev: device to which addresses will be added
|
| 3742 | * @from_dev: device from which addresses will be added
|
| 3743 | * @addr_type: address type - 0 means type will be used from from_dev
|
| 3744 | *
|
| 3745 | * Add device addresses of the one device to another.
|
| 3746 | **
|
| 3747 | * The caller must hold the rtnl_mutex.
|
| 3748 |
|
| 3749 | int dev_addr_add_multiple(struct net_device *to_dev,
|
| 3750 | struct net_device *from_dev,
|
| 3751 | unsigned char addr_type)
|
| 3752 | {
|
| 3753 | int err;
|
| 3754 |
|
| 3755 | ASSERT_RTNL();
|
| 3756 |
|
| 3757 | if (from_dev->addr_len != to_dev->addr_len)
|
| 3758 | return -EINVAL;
|
| 3759 | err = __hw_addr_add_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
|
| 3760 | to_dev->addr_len, addr_type);
|
| 3761 | if (!err)
|
| 3762 | call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
|
| 3763 | return err;
|
| 3764 | }
|
| 3765 | EXPORT_SYMBOL(dev_addr_add_multiple);
|
| 3766 |
|
| 3767 | |
| 3768 | * dev_addr_del_multiple - Delete device addresses by another device
|
| 3769 | * @to_dev: device where the addresses will be deleted
|
| 3770 | * @from_dev: device by which addresses the addresses will be deleted
|
| 3771 | * @addr_type: address type - 0 means type will used from from_dev
|
| 3772 | *
|
| 3773 | * Deletes addresses in to device by the list of addresses in from device.
|
| 3774 | *
|
| 3775 | * The caller must hold the rtnl_mutex.
|
| 3776 |
|
| 3777 | int dev_addr_del_multiple(struct net_device *to_dev,
|
| 3778 | struct net_device *from_dev,
|
| 3779 | unsigned char addr_type)
|
| 3780 | {
|
| 3781 | ASSERT_RTNL();
|
| 3782 |
|
| 3783 | if (from_dev->addr_len != to_dev->addr_len)
|
| 3784 | return -EINVAL;
|
| 3785 | __hw_addr_del_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
|
| 3786 | to_dev->addr_len, addr_type);
|
| 3787 | call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
|
| 3788 | return 0;
|
| 3789 | }
|
| 3790 | EXPORT_SYMBOL(dev_addr_del_multiple);
|
| 3791 |
|
| 3792 |
|
| 3793 |
|
| 3794 | int __dev_addr_delete(struct dev_addr_list **list, int *count,
|
| 3795 | void *addr, int alen, int glbl)
|
| 3796 | {
|
| 3797 | struct dev_addr_list *da;
|
| 3798 |
|
| 3799 | for (; (da = *list) != NULL; list = &da->next) {
|
| 3800 | if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
|
| 3801 | alen == da->da_addrlen) {
|
| 3802 | if (glbl) {
|
| 3803 | int old_glbl = da->da_gusers;
|
| 3804 | da->da_gusers = 0;
|
| 3805 | if (old_glbl == 0)
|
| 3806 | break;
|
| 3807 | }
|
| 3808 | if (--da->da_users)
|
| 3809 | return 0;
|
| 3810 |
|
| 3811 | *list = da->next;
|
| 3812 | kfree(da);
|
| 3813 | (*count)--;
|
| 3814 | return 0;
|
| 3815 | }
|
| 3816 | }
|
| 3817 | return -ENOENT;
|
| 3818 | }
|
| 3819 |
|
| 3820 | int __dev_addr_add(struct dev_addr_list **list, int *count,
|
| 3821 | void *addr, int alen, int glbl)
|
| 3822 | {
|
| 3823 | struct dev_addr_list *da;
|
| 3824 |
|
| 3825 | for (da = *list; da != NULL; da = da->next) {
|
| 3826 | if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
|
| 3827 | da->da_addrlen == alen) {
|
| 3828 | if (glbl) {
|
| 3829 | int old_glbl = da->da_gusers;
|
| 3830 | da->da_gusers = 1;
|
| 3831 | if (old_glbl)
|
| 3832 | return 0;
|
| 3833 | }
|
| 3834 | da->da_users++;
|
| 3835 | return 0;
|
| 3836 | }
|
| 3837 | }
|
| 3838 |
|
| 3839 | da = kzalloc(sizeof(*da), GFP_ATOMIC);
|
| 3840 | if (da == NULL)
|
| 3841 | return -ENOMEM;
|
| 3842 | memcpy(da->da_addr, addr, alen);
|
| 3843 | da->da_addrlen = alen;
|
| 3844 | da->da_users = 1;
|
| 3845 | da->da_gusers = glbl ? 1 : 0;
|
| 3846 | da->next = *list;
|
| 3847 | *list = da;
|
| 3848 | (*count)++;
|
| 3849 | return 0;
|
| 3850 | }
|
| 3851 |
|
| 3852 | |
| 3853 | * dev_unicast_delete - Release secondary unicast address.
|
| 3854 | * @dev: device
|
| 3855 | * @addr: address to delete
|
| 3856 | *
|
| 3857 | * Release reference to a secondary unicast address and remove it
|
| 3858 | * from the device if the reference count drops to zero.
|
| 3859 | *
|
| 3860 | * The caller must hold the rtnl_mutex.
|
| 3861 |
|
| 3862 | int dev_unicast_delete(struct net_device *dev, void *addr)
|
| 3863 | {
|
| 3864 | int err;
|
| 3865 |
|
| 3866 | ASSERT_RTNL();
|
| 3867 |
|
| 3868 | err = __hw_addr_del(&dev->uc, addr, dev->addr_len,
|
| 3869 | NETDEV_HW_ADDR_T_UNICAST);
|
| 3870 | if (!err)
|
| 3871 | __dev_set_rx_mode(dev);
|
| 3872 | return err;
|
| 3873 | }
|
| 3874 | EXPORT_SYMBOL(dev_unicast_delete);
|
| 3875 |
|
| 3876 | |
| 3877 | * dev_unicast_add - add a secondary unicast address
|
| 3878 | * @dev: device
|
| 3879 | * @addr: address to add
|
| 3880 | *
|
| 3881 | * Add a secondary unicast address to the device or increase
|
| 3882 | * the reference count if it already exists.
|
| 3883 | *
|
| 3884 | * The caller must hold the rtnl_mutex.
|
| 3885 |
|
| 3886 | int dev_unicast_add(struct net_device *dev, void *addr)
|
| 3887 | {
|
| 3888 | int err;
|
| 3889 |
|
| 3890 | ASSERT_RTNL();
|
| 3891 |
|
| 3892 | err = __hw_addr_add(&dev->uc, addr, dev->addr_len,
|
| 3893 | NETDEV_HW_ADDR_T_UNICAST);
|
| 3894 | if (!err)
|
| 3895 | __dev_set_rx_mode(dev);
|
| 3896 | return err;
|
| 3897 | }
|
| 3898 | EXPORT_SYMBOL(dev_unicast_add);
|
| 3899 |
|
| 3900 | int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
|
| 3901 | struct dev_addr_list **from, int *from_count)
|
| 3902 | {
|
| 3903 | struct dev_addr_list *da, *next;
|
| 3904 | int err = 0;
|
| 3905 |
|
| 3906 | da = *from;
|
| 3907 | while (da != NULL) {
|
| 3908 | next = da->next;
|
| 3909 | if (!da->da_synced) {
|
| 3910 | err = __dev_addr_add(to, to_count,
|
| 3911 | da->da_addr, da->da_addrlen, 0);
|
| 3912 | if (err < 0)
|
| 3913 | break;
|
| 3914 | da->da_synced = 1;
|
| 3915 | da->da_users++;
|
| 3916 | } else if (da->da_users == 1) {
|
| 3917 | __dev_addr_delete(to, to_count,
|
| 3918 | da->da_addr, da->da_addrlen, 0);
|
| 3919 | __dev_addr_delete(from, from_count,
|
| 3920 | da->da_addr, da->da_addrlen, 0);
|
| 3921 | }
|
| 3922 | da = next;
|
| 3923 | }
|
| 3924 | return err;
|
| 3925 | }
|
| 3926 |
|
| 3927 | void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
|
| 3928 | struct dev_addr_list **from, int *from_count)
|
| 3929 | {
|
| 3930 | struct dev_addr_list *da, *next;
|
| 3931 |
|
| 3932 | da = *from;
|
| 3933 | while (da != NULL) {
|
| 3934 | next = da->next;
|
| 3935 | if (da->da_synced) {
|
| 3936 | __dev_addr_delete(to, to_count,
|
| 3937 | da->da_addr, da->da_addrlen, 0);
|
| 3938 | da->da_synced = 0;
|
| 3939 | __dev_addr_delete(from, from_count,
|
| 3940 | da->da_addr, da->da_addrlen, 0);
|
| 3941 | }
|
| 3942 | da = next;
|
| 3943 | }
|
| 3944 | }
|
| 3945 |
|
| 3946 | |
| 3947 | * dev_unicast_sync - Synchronize device's unicast list to another device
|
| 3948 | * @to: destination device
|
| 3949 | * @from: source device
|
| 3950 | *
|
| 3951 | * Add newly added addresses to the destination device and release
|
| 3952 | * addresses that have no users left.
|
| 3953 | *
|
| 3954 | * This function is intended to be called from the dev->set_rx_mode
|
| 3955 | * function of layered software devices.
|
| 3956 |
|
| 3957 | int dev_unicast_sync(struct net_device *to, struct net_device *from)
|
| 3958 | {
|
| 3959 | int err = 0;
|
| 3960 |
|
| 3961 | ASSERT_RTNL();
|
| 3962 |
|
| 3963 | if (to->addr_len != from->addr_len)
|
| 3964 | return -EINVAL;
|
| 3965 |
|
| 3966 | err = __hw_addr_sync(&to->uc, &from->uc, to->addr_len);
|
| 3967 | if (!err)
|
| 3968 | __dev_set_rx_mode(to);
|
| 3969 | return err;
|
| 3970 | }
|
| 3971 | EXPORT_SYMBOL(dev_unicast_sync);
|
| 3972 |
|
| 3973 | |
| 3974 | * dev_unicast_unsync - Remove synchronized addresses from the destination device
|
| 3975 | * @to: destination device
|
| 3976 | * @from: source device
|
| 3977 | *
|
| 3978 | * Remove all addresses that were added to the destination device by
|
| 3979 | * dev_unicast_sync(). This function is intended to be called from the
|
| 3980 | * dev->stop function of layered software devices.
|
| 3981 |
|
| 3982 | void dev_unicast_unsync(struct net_device *to, struct net_device *from)
|
| 3983 | {
|
| 3984 | ASSERT_RTNL();
|
| 3985 |
|
| 3986 | if (to->addr_len != from->addr_len)
|
| 3987 | return;
|
| 3988 |
|
| 3989 | __hw_addr_unsync(&to->uc, &from->uc, to->addr_len);
|
| 3990 | __dev_set_rx_mode(to);
|
| 3991 | }
|
| 3992 | EXPORT_SYMBOL(dev_unicast_unsync);
|
| 3993 |
|
| 3994 | static void dev_unicast_flush(struct net_device *dev)
|
| 3995 | {
|
| 3996 |
|
| 3997 |
|
| 3998 | __hw_addr_flush(&dev->uc);
|
| 3999 | }
|
| 4000 |
|
| 4001 | static void dev_unicast_init(struct net_device *dev)
|
| 4002 | {
|
| 4003 |
|
| 4004 |
|
| 4005 | __hw_addr_init(&dev->uc);
|
| 4006 | }
|
| 4007 |
|
| 4008 |
|
| 4009 | static void __dev_addr_discard(struct dev_addr_list **list)
|
| 4010 | {
|
| 4011 | struct dev_addr_list *tmp;
|
| 4012 |
|
| 4013 | while (*list != NULL) {
|
| 4014 | tmp = *list;
|
| 4015 | *list = tmp->next;
|
| 4016 | if (tmp->da_users > tmp->da_gusers)
|
| 4017 | printk("__dev_addr_discard: address leakage! "
|
| 4018 | "da_users=%d\n", tmp->da_users);
|
| 4019 | kfree(tmp);
|
| 4020 | }
|
| 4021 | }
|
| 4022 |
|
| 4023 | static void dev_addr_discard(struct net_device *dev)
|
| 4024 | {
|
| 4025 | netif_addr_lock_bh(dev);
|
| 4026 |
|
| 4027 | __dev_addr_discard(&dev->mc_list);
|
| 4028 | dev->mc_count = 0;
|
| 4029 |
|
| 4030 | netif_addr_unlock_bh(dev);
|
| 4031 | }
|
| 4032 |
|
| 4033 | |
| 4034 | * dev_get_flags - get flags reported to userspace
|
| 4035 | * @dev: device
|
| 4036 | *
|
| 4037 | * Get the combination of flag bits exported through APIs to userspace.
|
| 4038 |
|
| 4039 | unsigned dev_get_flags(const struct net_device *dev)
|
| 4040 | {
|
| 4041 | unsigned flags;
|
| 4042 |
|
| 4043 | flags = (dev->flags & ~(IFF_PROMISC |
|
| 4044 | IFF_ALLMULTI |
|
| 4045 | IFF_RUNNING |
|
| 4046 | IFF_LOWER_UP |
|
| 4047 | IFF_DORMANT)) |
|
| 4048 | (dev->gflags & (IFF_PROMISC |
|
| 4049 | IFF_ALLMULTI));
|
| 4050 |
|
| 4051 | if (netif_running(dev)) {
|
| 4052 | if (netif_oper_up(dev))
|
| 4053 | flags |= IFF_RUNNING;
|
| 4054 | if (netif_carrier_ok(dev))
|
| 4055 | flags |= IFF_LOWER_UP;
|
| 4056 | if (netif_dormant(dev))
|
| 4057 | flags |= IFF_DORMANT;
|
| 4058 | }
|
| 4059 |
|
| 4060 | return flags;
|
| 4061 | }
|
| 4062 |
|
| 4063 | |
| 4064 | * dev_change_flags - change device settings
|
| 4065 | * @dev: device
|
| 4066 | * @flags: device state flags
|
| 4067 | *
|
| 4068 | * Change settings on device based state flags. The flags are
|
| 4069 | * in the userspace exported format.
|
| 4070 |
|
| 4071 | int dev_change_flags(struct net_device *dev, unsigned flags)
|
| 4072 | {
|
| 4073 | int ret, changes;
|
| 4074 | int old_flags = dev->flags;
|
| 4075 |
|
| 4076 | ASSERT_RTNL();
|
| 4077 |
|
| 4078 | |
| 4079 | * Set the flags on our device.
|
| 4080 |
|
| 4081 |
|
| 4082 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
|
| 4083 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
|
| 4084 | IFF_AUTOMEDIA)) |
|
| 4085 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
|
| 4086 | IFF_ALLMULTI));
|
| 4087 |
|
| 4088 | |
| 4089 | * Load in the correct multicast list now the flags have changed.
|
| 4090 |
|
| 4091 |
|
| 4092 | if ((old_flags ^ flags) & IFF_MULTICAST)
|
| 4093 | dev_change_rx_flags(dev, IFF_MULTICAST);
|
| 4094 |
|
| 4095 | dev_set_rx_mode(dev);
|
| 4096 |
|
| 4097 | |
| 4098 | * Have we downed the interface. We handle IFF_UP ourselves
|
| 4099 | * according to user attempts to set it, rather than blindly
|
| 4100 | * setting it.
|
| 4101 |
|
| 4102 |
|
| 4103 | ret = 0;
|
| 4104 | if ((old_flags ^ flags) & IFF_UP) {
|
| 4105 | ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
|
| 4106 |
|
| 4107 | if (!ret)
|
| 4108 | dev_set_rx_mode(dev);
|
| 4109 | }
|
| 4110 |
|
| 4111 | if (dev->flags & IFF_UP &&
|
| 4112 | ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
|
| 4113 | IFF_VOLATILE)))
|
| 4114 | call_netdevice_notifiers(NETDEV_CHANGE, dev);
|
| 4115 |
|
| 4116 | if ((flags ^ dev->gflags) & IFF_PROMISC) {
|
| 4117 | int inc = (flags & IFF_PROMISC) ? +1 : -1;
|
| 4118 | dev->gflags ^= IFF_PROMISC;
|
| 4119 | dev_set_promiscuity(dev, inc);
|
| 4120 | }
|
| 4121 |
|
| 4122 | |
| 4123 | is important. Some (broken) drivers set IFF_PROMISC, when
|
| 4124 | IFF_ALLMULTI is requested not asking us and not reporting.
|
| 4125 |
|
| 4126 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
|
| 4127 | int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
|
| 4128 | dev->gflags ^= IFF_ALLMULTI;
|
| 4129 | dev_set_allmulti(dev, inc);
|
| 4130 | }
|
| 4131 |
|
| 4132 |
|
| 4133 | changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
|
| 4134 | if (changes)
|
| 4135 | rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
|
| 4136 |
|
| 4137 | return ret;
|
| 4138 | }
|
| 4139 |
|
| 4140 | |
| 4141 | * dev_set_mtu - Change maximum transfer unit
|
| 4142 | * @dev: device
|
| 4143 | * @new_mtu: new transfer unit
|
| 4144 | *
|
| 4145 | * Change the maximum transfer size of the network device.
|
| 4146 |
|
| 4147 | int dev_set_mtu(struct net_device *dev, int new_mtu)
|
| 4148 | {
|
| 4149 | const struct net_device_ops *ops = dev->netdev_ops;
|
| 4150 | int err;
|
| 4151 |
|
| 4152 | if (new_mtu == dev->mtu)
|
| 4153 | return 0;
|
| 4154 |
|
| 4155 |
|
| 4156 | if (new_mtu < 0)
|
| 4157 | return -EINVAL;
|
| 4158 |
|
| 4159 | if (!netif_device_present(dev))
|
| 4160 | return -ENODEV;
|
| 4161 |
|
| 4162 | err = 0;
|
| 4163 | if (ops->ndo_change_mtu)
|
| 4164 | err = ops->ndo_change_mtu(dev, new_mtu);
|
| 4165 | else
|
| 4166 | dev->mtu = new_mtu;
|
| 4167 |
|
| 4168 | if (!err && dev->flags & IFF_UP)
|
| 4169 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
|
| 4170 | return err;
|
| 4171 | }
|
| 4172 |
|
| 4173 | |
| 4174 | * dev_set_mac_address - Change Media Access Control Address
|
| 4175 | * @dev: device
|
| 4176 | * @sa: new address
|
| 4177 | *
|
| 4178 | * Change the hardware (MAC) address of the device
|
| 4179 |
|
| 4180 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
|
| 4181 | {
|
| 4182 | const struct net_device_ops *ops = dev->netdev_ops;
|
| 4183 | int err;
|
| 4184 |
|
| 4185 | if (!ops->ndo_set_mac_address)
|
| 4186 | return -EOPNOTSUPP;
|
| 4187 | if (sa->sa_family != dev->type)
|
| 4188 | return -EINVAL;
|
| 4189 | if (!netif_device_present(dev))
|
| 4190 | return -ENODEV;
|
| 4191 | err = ops->ndo_set_mac_address(dev, sa);
|
| 4192 | if (!err)
|
| 4193 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
|
| 4194 | return err;
|
| 4195 | }
|
| 4196 |
|
| 4197 | |
| 4198 | * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
|
| 4199 |
|
| 4200 | static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
|
| 4201 | {
|
| 4202 | int err;
|
| 4203 | struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
|
| 4204 |
|
| 4205 | if (!dev)
|
| 4206 | return -ENODEV;
|
| 4207 |
|
| 4208 | switch (cmd) {
|
| 4209 | case SIOCGIFFLAGS:
|
| 4210 | ifr->ifr_flags = (short) dev_get_flags(dev);
|
| 4211 | return 0;
|
| 4212 |
|
| 4213 | case SIOCGIFMETRIC: |
| 4214 | (currently unused)
|
| 4215 | ifr->ifr_metric = 0;
|
| 4216 | return 0;
|
| 4217 |
|
| 4218 | case SIOCGIFMTU:
|
| 4219 | ifr->ifr_mtu = dev->mtu;
|
| 4220 | return 0;
|
| 4221 |
|
| 4222 | case SIOCGIFHWADDR:
|
| 4223 | if (!dev->addr_len)
|
| 4224 | memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
|
| 4225 | else
|
| 4226 | memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
|
| 4227 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
|
| 4228 | ifr->ifr_hwaddr.sa_family = dev->type;
|
| 4229 | return 0;
|
| 4230 |
|
| 4231 | case SIOCGIFSLAVE:
|
| 4232 | err = -EINVAL;
|
| 4233 | break;
|
| 4234 |
|
| 4235 | case SIOCGIFMAP:
|
| 4236 | ifr->ifr_map.mem_start = dev->mem_start;
|
| 4237 | ifr->ifr_map.mem_end = dev->mem_end;
|
| 4238 | ifr->ifr_map.base_addr = dev->base_addr;
|
| 4239 | ifr->ifr_map.irq = dev->irq;
|
| 4240 | ifr->ifr_map.dma = dev->dma;
|
| 4241 | ifr->ifr_map.port = dev->if_port;
|
| 4242 | return 0;
|
| 4243 |
|
| 4244 | case SIOCGIFINDEX:
|
| 4245 | ifr->ifr_ifindex = dev->ifindex;
|
| 4246 | return 0;
|
| 4247 |
|
| 4248 | case SIOCGIFTXQLEN:
|
| 4249 | ifr->ifr_qlen = dev->tx_queue_len;
|
| 4250 | return 0;
|
| 4251 |
|
| 4252 | default:
|
| 4253 | |
| 4254 | * is never reached
|
| 4255 |
|
| 4256 | WARN_ON(1);
|
| 4257 | err = -EINVAL;
|
| 4258 | break;
|
| 4259 |
|
| 4260 | }
|
| 4261 | return err;
|
| 4262 | }
|
| 4263 |
|
| 4264 | |
| 4265 | * Perform the SIOCxIFxxx calls, inside rtnl_lock()
|
| 4266 |
|
| 4267 | static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
|
| 4268 | {
|
| 4269 | int err;
|
| 4270 | struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
|
| 4271 | const struct net_device_ops *ops;
|
| 4272 |
|
| 4273 | if (!dev)
|
| 4274 | return -ENODEV;
|
| 4275 |
|
| 4276 | ops = dev->netdev_ops;
|
| 4277 |
|
| 4278 | switch (cmd) {
|
| 4279 | case SIOCSIFFLAGS:
|
| 4280 | return dev_change_flags(dev, ifr->ifr_flags);
|
| 4281 |
|
| 4282 | case SIOCSIFMETRIC: |
| 4283 | (currently unused)
|
| 4284 | return -EOPNOTSUPP;
|
| 4285 |
|
| 4286 | case SIOCSIFMTU:
|
| 4287 | return dev_set_mtu(dev, ifr->ifr_mtu);
|
| 4288 |
|
| 4289 | case SIOCSIFHWADDR:
|
| 4290 | return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
|
| 4291 |
|
| 4292 | case SIOCSIFHWBROADCAST:
|
| 4293 | if (ifr->ifr_hwaddr.sa_family != dev->type)
|
| 4294 | return -EINVAL;
|
| 4295 | memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
|
| 4296 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
|
| 4297 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
|
| 4298 | return 0;
|
| 4299 |
|
| 4300 | case SIOCSIFMAP:
|
| 4301 | if (ops->ndo_set_config) {
|
| 4302 | if (!netif_device_present(dev))
|
| 4303 | return -ENODEV;
|
| 4304 | return ops->ndo_set_config(dev, &ifr->ifr_map);
|
| 4305 | }
|
| 4306 | return -EOPNOTSUPP;
|
| 4307 |
|
| 4308 | case SIOCADDMULTI:
|
| 4309 | if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
|
| 4310 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
|
| 4311 | return -EINVAL;
|
| 4312 | if (!netif_device_present(dev))
|
| 4313 | return -ENODEV;
|
| 4314 | return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
|
| 4315 | dev->addr_len, 1);
|
| 4316 |
|
| 4317 | case SIOCDELMULTI:
|
| 4318 | if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
|
| 4319 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
|
| 4320 | return -EINVAL;
|
| 4321 | if (!netif_device_present(dev))
|
| 4322 | return -ENODEV;
|
| 4323 | return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
|
| 4324 | dev->addr_len, 1);
|
| 4325 |
|
| 4326 | case SIOCSIFTXQLEN:
|
| 4327 | if (ifr->ifr_qlen < 0)
|
| 4328 | return -EINVAL;
|
| 4329 | dev->tx_queue_len = ifr->ifr_qlen;
|
| 4330 | return 0;
|
| 4331 |
|
| 4332 | case SIOCSIFNAME:
|
| 4333 | ifr->ifr_newname[IFNAMSIZ-1] = '\0';
|
| 4334 | return dev_change_name(dev, ifr->ifr_newname);
|
| 4335 |
|
| 4336 | |
| 4337 | * Unknown or private ioctl
|
| 4338 |
|
| 4339 |
|
| 4340 | default:
|
| 4341 | if ((cmd >= SIOCDEVPRIVATE &&
|
| 4342 | cmd <= SIOCDEVPRIVATE + 15) ||
|
| 4343 | cmd == SIOCBONDENSLAVE ||
|
| 4344 | cmd == SIOCBONDRELEASE ||
|
| 4345 | cmd == SIOCBONDSETHWADDR ||
|
| 4346 | cmd == SIOCBONDSLAVEINFOQUERY ||
|
| 4347 | cmd == SIOCBONDINFOQUERY ||
|
| 4348 | cmd == SIOCBONDCHANGEACTIVE ||
|
| 4349 | cmd == SIOCGMIIPHY ||
|
| 4350 | cmd == SIOCGMIIREG ||
|
| 4351 | cmd == SIOCSMIIREG ||
|
| 4352 | cmd == SIOCBRADDIF ||
|
| 4353 | cmd == SIOCBRDELIF ||
|
| 4354 | cmd == SIOCSHWTSTAMP ||
|
| 4355 | cmd == SIOCWANDEV) {
|
| 4356 | err = -EOPNOTSUPP;
|
| 4357 | if (ops->ndo_do_ioctl) {
|
| 4358 | if (netif_device_present(dev))
|
| 4359 | err = ops->ndo_do_ioctl(dev, ifr, cmd);
|
| 4360 | else
|
| 4361 | err = -ENODEV;
|
| 4362 | }
|
| 4363 | } else
|
| 4364 | err = -EINVAL;
|
| 4365 |
|
| 4366 | }
|
| 4367 | return err;
|
| 4368 | }
|
| 4369 |
|
| 4370 | |
| 4371 | * This function handles all "interface"-type I/O control requests. The actual
|
| 4372 | * 'doing' part of this is dev_ifsioc above.
|
| 4373 |
|
| 4374 |
|
| 4375 | |
| 4376 | * dev_ioctl - network device ioctl
|
| 4377 | * @net: the applicable net namespace
|
| 4378 | * @cmd: command to issue
|
| 4379 | * @arg: pointer to a struct ifreq in user space
|
| 4380 | *
|
| 4381 | * Issue ioctl functions to devices. This is normally called by the
|
| 4382 | * user space syscall interfaces but can sometimes be useful for
|
| 4383 | * other purposes. The return value is the return from the syscall if
|
| 4384 | * positive or a negative errno code on error.
|
| 4385 |
|
| 4386 |
|
| 4387 | int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
|
| 4388 | {
|
| 4389 | struct ifreq ifr;
|
| 4390 | int ret;
|
| 4391 | char *colon;
|
| 4392 |
|
| 4393 | |
| 4394 | and requires shared lock, because it sleeps writing
|
| 4395 | to user space.
|
| 4396 |
|
| 4397 |
|
| 4398 | if (cmd == SIOCGIFCONF) {
|
| 4399 | rtnl_lock();
|
| 4400 | ret = dev_ifconf(net, (char __user *) arg);
|
| 4401 | rtnl_unlock();
|
| 4402 | return ret;
|
| 4403 | }
|
| 4404 | if (cmd == SIOCGIFNAME)
|
| 4405 | return dev_ifname(net, (struct ifreq __user *)arg);
|
| 4406 |
|
| 4407 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
|
| 4408 | return -EFAULT;
|
| 4409 |
|
| 4410 | ifr.ifr_name[IFNAMSIZ-1] = 0;
|
| 4411 |
|
| 4412 | colon = strchr(ifr.ifr_name, ':');
|
| 4413 | if (colon)
|
| 4414 | *colon = 0;
|
| 4415 |
|
| 4416 | |
| 4417 | * See which interface the caller is talking about.
|
| 4418 |
|
| 4419 |
|
| 4420 | switch (cmd) {
|
| 4421 | |
| 4422 | * These ioctl calls:
|
| 4423 | * - can be done by all.
|
| 4424 | * - atomic and do not require locking.
|
| 4425 | * - return a value
|
| 4426 |
|
| 4427 | case SIOCGIFFLAGS:
|
| 4428 | case SIOCGIFMETRIC:
|
| 4429 | case SIOCGIFMTU:
|
| 4430 | case SIOCGIFHWADDR:
|
| 4431 | case SIOCGIFSLAVE:
|
| 4432 | case SIOCGIFMAP:
|
| 4433 | case SIOCGIFINDEX:
|
| 4434 | case SIOCGIFTXQLEN:
|
| 4435 | dev_load(net, ifr.ifr_name);
|
| 4436 | read_lock(&dev_base_lock);
|
| 4437 | ret = dev_ifsioc_locked(net, &ifr, cmd);
|
| 4438 | read_unlock(&dev_base_lock);
|
| 4439 | if (!ret) {
|
| 4440 | if (colon)
|
| 4441 | *colon = ':';
|
| 4442 | if (copy_to_user(arg, &ifr,
|
| 4443 | sizeof(struct ifreq)))
|
| 4444 | ret = -EFAULT;
|
| 4445 | }
|
| 4446 | return ret;
|
| 4447 |
|
| 4448 | case SIOCETHTOOL:
|
| 4449 | dev_load(net, ifr.ifr_name);
|
| 4450 | rtnl_lock();
|
| 4451 | ret = dev_ethtool(net, &ifr);
|
| 4452 | rtnl_unlock();
|
| 4453 | if (!ret) {
|
| 4454 | if (colon)
|
| 4455 | *colon = ':';
|
| 4456 | if (copy_to_user(arg, &ifr,
|
| 4457 | sizeof(struct ifreq)))
|
| 4458 | ret = -EFAULT;
|
| 4459 | }
|
| 4460 | return ret;
|
| 4461 |
|
| 4462 | |
| 4463 | * These ioctl calls:
|
| 4464 | * - require superuser power.
|
| 4465 | * - require strict serialization.
|
| 4466 | * - return a value
|
| 4467 |
|
| 4468 | case SIOCGMIIPHY:
|
| 4469 | case SIOCGMIIREG:
|
| 4470 | case SIOCSIFNAME:
|
| 4471 | if (!capable(CAP_NET_ADMIN))
|
| 4472 | return -EPERM;
|
| 4473 | dev_load(net, ifr.ifr_name);
|
| 4474 | rtnl_lock();
|
| 4475 | ret = dev_ifsioc(net, &ifr, cmd);
|
| 4476 | rtnl_unlock();
|
| 4477 | if (!ret) {
|
| 4478 | if (colon)
|
| 4479 | *colon = ':';
|
| 4480 | if (copy_to_user(arg, &ifr,
|
| 4481 | sizeof(struct ifreq)))
|
| 4482 | ret = -EFAULT;
|
| 4483 | }
|
| 4484 | return ret;
|
| 4485 |
|
| 4486 | |
| 4487 | * These ioctl calls:
|
| 4488 | * - require superuser power.
|
| 4489 | * - require strict serialization.
|
| 4490 | * - do not return a value
|
| 4491 |
|
| 4492 | case SIOCSIFFLAGS:
|
| 4493 | case SIOCSIFMETRIC:
|
| 4494 | case SIOCSIFMTU:
|
| 4495 | case SIOCSIFMAP:
|
| 4496 | case SIOCSIFHWADDR:
|
| 4497 | case SIOCSIFSLAVE:
|
| 4498 | case SIOCADDMULTI:
|
| 4499 | case SIOCDELMULTI:
|
| 4500 | case SIOCSIFHWBROADCAST:
|
| 4501 | case SIOCSIFTXQLEN:
|
| 4502 | case SIOCSMIIREG:
|
| 4503 | case SIOCBONDENSLAVE:
|
| 4504 | case SIOCBONDRELEASE:
|
| 4505 | case SIOCBONDSETHWADDR:
|
| 4506 | case SIOCBONDCHANGEACTIVE:
|
| 4507 | case SIOCBRADDIF:
|
| 4508 | case SIOCBRDELIF:
|
| 4509 | case SIOCSHWTSTAMP:
|
| 4510 | if (!capable(CAP_NET_ADMIN))
|
| 4511 | return -EPERM;
|
| 4512 |
|
| 4513 | case SIOCBONDSLAVEINFOQUERY:
|
| 4514 | case SIOCBONDINFOQUERY:
|
| 4515 | dev_load(net, ifr.ifr_name);
|
| 4516 | rtnl_lock();
|
| 4517 | ret = dev_ifsioc(net, &ifr, cmd);
|
| 4518 | rtnl_unlock();
|
| 4519 | return ret;
|
| 4520 |
|
| 4521 | case SIOCGIFMEM:
|
| 4522 | |
| 4523 | * currently do not support it
|
| 4524 | case SIOCSIFMEM:
|
| 4525 | |
| 4526 | * Not applicable in our case
|
| 4527 | case SIOCSIFLINK:
|
| 4528 | return -EINVAL;
|
| 4529 |
|
| 4530 | |
| 4531 | * Unknown or private ioctl.
|
| 4532 |
|
| 4533 | default:
|
| 4534 | if (cmd == SIOCWANDEV ||
|
| 4535 | (cmd >= SIOCDEVPRIVATE &&
|
| 4536 | cmd <= SIOCDEVPRIVATE + 15)) {
|
| 4537 | dev_load(net, ifr.ifr_name);
|
| 4538 | rtnl_lock();
|
| 4539 | ret = dev_ifsioc(net, &ifr, cmd);
|
| 4540 | rtnl_unlock();
|
| 4541 | if (!ret && copy_to_user(arg, &ifr,
|
| 4542 | sizeof(struct ifreq)))
|
| 4543 | ret = -EFAULT;
|
| 4544 | return ret;
|
| 4545 | }
|
| 4546 |
|
| 4547 | if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
|
| 4548 | return wext_handle_ioctl(net, &ifr, cmd, arg);
|
| 4549 | return -EINVAL;
|
| 4550 | }
|
| 4551 | }
|
| 4552 |
|
| 4553 |
|
| 4554 | |
| 4555 | * dev_new_index - allocate an ifindex
|
| 4556 | * @net: the applicable net namespace
|
| 4557 | *
|
| 4558 | * Returns a suitable unique value for a new device interface
|
| 4559 | * number. The caller must hold the rtnl semaphore or the
|
| 4560 | * dev_base_lock to be sure it remains unique.
|
| 4561 |
|
| 4562 | static int dev_new_index(struct net *net)
|
| 4563 | {
|
| 4564 | static int ifindex;
|
| 4565 | for (;;) {
|
| 4566 | if (++ifindex <= 0)
|
| 4567 | ifindex = 1;
|
| 4568 | if (!__dev_get_by_index(net, ifindex))
|
| 4569 | return ifindex;
|
| 4570 | }
|
| 4571 | }
|
| 4572 |
|
| 4573 |
|
| 4574 | static LIST_HEAD(net_todo_list);
|
| 4575 |
|
| 4576 | static void net_set_todo(struct net_device *dev)
|
| 4577 | {
|
| 4578 | list_add_tail(&dev->todo_list, &net_todo_list);
|
| 4579 | }
|
| 4580 |
|
| 4581 | static void rollback_registered(struct net_device *dev)
|
| 4582 | {
|
| 4583 | BUG_ON(dev_boot_phase);
|
| 4584 | ASSERT_RTNL();
|
| 4585 |
|
| 4586 |
|
| 4587 | if (dev->reg_state == NETREG_UNINITIALIZED) {
|
| 4588 | printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
|
| 4589 | "was registered\n", dev->name, dev);
|
| 4590 |
|
| 4591 | WARN_ON(1);
|
| 4592 | return;
|
| 4593 | }
|
| 4594 |
|
| 4595 | BUG_ON(dev->reg_state != NETREG_REGISTERED);
|
| 4596 |
|
| 4597 |
|
| 4598 | dev_close(dev);
|
| 4599 |
|
| 4600 |
|
| 4601 | unlist_netdevice(dev);
|
| 4602 |
|
| 4603 | dev->reg_state = NETREG_UNREGISTERING;
|
| 4604 |
|
| 4605 | synchronize_net();
|
| 4606 |
|
| 4607 |
|
| 4608 | dev_shutdown(dev);
|
| 4609 |
|
| 4610 |
|
| 4611 | |
| 4612 | this device. They should clean all the things.
|
| 4613 |
|
| 4614 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
|
| 4615 |
|
| 4616 | |
| 4617 | * Flush the unicast and multicast chains
|
| 4618 |
|
| 4619 | dev_unicast_flush(dev);
|
| 4620 | dev_addr_discard(dev);
|
| 4621 |
|
| 4622 | if (dev->netdev_ops->ndo_uninit)
|
| 4623 | dev->netdev_ops->ndo_uninit(dev);
|
| 4624 |
|
| 4625 |
|
| 4626 | WARN_ON(dev->master);
|
| 4627 |
|
| 4628 |
|
| 4629 | netdev_unregister_kobject(dev);
|
| 4630 |
|
| 4631 | synchronize_net();
|
| 4632 |
|
| 4633 | dev_put(dev);
|
| 4634 | }
|
| 4635 |
|
| 4636 | static void __netdev_init_queue_locks_one(struct net_device *dev,
|
| 4637 | struct netdev_queue *dev_queue,
|
| 4638 | void *_unused)
|
| 4639 | {
|
| 4640 | spin_lock_init(&dev_queue->_xmit_lock);
|
| 4641 | netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
|
| 4642 | dev_queue->xmit_lock_owner = -1;
|
| 4643 | }
|
| 4644 |
|
| 4645 | static void netdev_init_queue_locks(struct net_device *dev)
|
| 4646 | {
|
| 4647 | netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
|
| 4648 | __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
|
| 4649 | }
|
| 4650 |
|
| 4651 | unsigned long netdev_fix_features(unsigned long features, const char *name)
|
| 4652 | {
|
| 4653 |
|
| 4654 | if ((features & NETIF_F_SG) &&
|
| 4655 | !(features & NETIF_F_ALL_CSUM)) {
|
| 4656 | if (name)
|
| 4657 | printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
|
| 4658 | "checksum feature.\n", name);
|
| 4659 | features &= ~NETIF_F_SG;
|
| 4660 | }
|
| 4661 |
|
| 4662 |
|
| 4663 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
|
| 4664 | if (name)
|
| 4665 | printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
|
| 4666 | "SG feature.\n", name);
|
| 4667 | features &= ~NETIF_F_TSO;
|
| 4668 | }
|
| 4669 |
|
| 4670 | if (features & NETIF_F_UFO) {
|
| 4671 | if (!(features & NETIF_F_GEN_CSUM)) {
|
| 4672 | if (name)
|
| 4673 | printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
|
| 4674 | "since no NETIF_F_HW_CSUM feature.\n",
|
| 4675 | name);
|
| 4676 | features &= ~NETIF_F_UFO;
|
| 4677 | }
|
| 4678 |
|
| 4679 | if (!(features & NETIF_F_SG)) {
|
| 4680 | if (name)
|
| 4681 | printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
|
| 4682 | "since no NETIF_F_SG feature.\n", name);
|
| 4683 | features &= ~NETIF_F_UFO;
|
| 4684 | }
|
| 4685 | }
|
| 4686 |
|
| 4687 | return features;
|
| 4688 | }
|
| 4689 | EXPORT_SYMBOL(netdev_fix_features);
|
| 4690 |
|
| 4691 | |
| 4692 | * register_netdevice - register a network device
|
| 4693 | * @dev: device to register
|
| 4694 | *
|
| 4695 | * Take a completed network device structure and add it to the kernel
|
| 4696 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
|
| 4697 | * chain. 0 is returned on success. A negative errno code is returned
|
| 4698 | * on a failure to set up the device, or if the name is a duplicate.
|
| 4699 | *
|
| 4700 | * Callers must hold the rtnl semaphore. You may want
|
| 4701 | * register_netdev() instead of this.
|
| 4702 | *
|
| 4703 | * BUGS:
|
| 4704 | * The locking appears insufficient to guarantee two parallel registers
|
| 4705 | * will not get the same name.
|
| 4706 |
|
| 4707 |
|
| 4708 | int register_netdevice(struct net_device *dev)
|
| 4709 | {
|
| 4710 | struct hlist_head *head;
|
| 4711 | struct hlist_node *p;
|
| 4712 | int ret;
|
| 4713 | struct net *net = dev_net(dev);
|
| 4714 |
|
| 4715 | BUG_ON(dev_boot_phase);
|
| 4716 | ASSERT_RTNL();
|
| 4717 |
|
| 4718 | might_sleep();
|
| 4719 |
|
| 4720 |
|
| 4721 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
|
| 4722 | BUG_ON(!net);
|
| 4723 |
|
| 4724 | spin_lock_init(&dev->addr_list_lock);
|
| 4725 | netdev_set_addr_lockdep_class(dev);
|
| 4726 | netdev_init_queue_locks(dev);
|
| 4727 |
|
| 4728 | dev->iflink = -1;
|
| 4729 |
|
| 4730 |
|
| 4731 | if (dev->netdev_ops->ndo_init) {
|
| 4732 | ret = dev->netdev_ops->ndo_init(dev);
|
| 4733 | if (ret) {
|
| 4734 | if (ret > 0)
|
| 4735 | ret = -EIO;
|
| 4736 | goto out;
|
| 4737 | }
|
| 4738 | }
|
| 4739 |
|
| 4740 | if (!dev_valid_name(dev->name)) {
|
| 4741 | ret = -EINVAL;
|
| 4742 | goto err_uninit;
|
| 4743 | }
|
| 4744 |
|
| 4745 | dev->ifindex = dev_new_index(net);
|
| 4746 | if (dev->iflink == -1)
|
| 4747 | dev->iflink = dev->ifindex;
|
| 4748 |
|
| 4749 |
|
| 4750 | head = dev_name_hash(net, dev->name);
|
| 4751 | hlist_for_each(p, head) {
|
| 4752 | struct net_device *d
|
| 4753 | = hlist_entry(p, struct net_device, name_hlist);
|
| 4754 | if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
|
| 4755 | ret = -EEXIST;
|
| 4756 | goto err_uninit;
|
| 4757 | }
|
| 4758 | }
|
| 4759 |
|
| 4760 |
|
| 4761 | if ((dev->features & NETIF_F_HW_CSUM) &&
|
| 4762 | (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
|
| 4763 | printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
|
| 4764 | dev->name);
|
| 4765 | dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
|
| 4766 | }
|
| 4767 |
|
| 4768 | if ((dev->features & NETIF_F_NO_CSUM) &&
|
| 4769 | (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
|
| 4770 | printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
|
| 4771 | dev->name);
|
| 4772 | dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
|
| 4773 | }
|
| 4774 |
|
| 4775 | dev->features = netdev_fix_features(dev->features, dev->name);
|
| 4776 |
|
| 4777 |
|
| 4778 | if (dev->features & NETIF_F_SG)
|
| 4779 | dev->features |= NETIF_F_GSO;
|
| 4780 |
|
| 4781 | netdev_initialize_kobject(dev);
|
| 4782 | ret = netdev_register_kobject(dev);
|
| 4783 | if (ret)
|
| 4784 | goto err_uninit;
|
| 4785 | dev->reg_state = NETREG_REGISTERED;
|
| 4786 |
|
| 4787 | |
| 4788 | * Default initial state at registry is that the
|
| 4789 | * device is present.
|
| 4790 |
|
| 4791 |
|
| 4792 | set_bit(__LINK_STATE_PRESENT, &dev->state);
|
| 4793 |
|
| 4794 | dev_init_scheduler(dev);
|
| 4795 | dev_hold(dev);
|
| 4796 | list_netdevice(dev);
|
| 4797 |
|
| 4798 |
|
| 4799 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
|
| 4800 | ret = notifier_to_errno(ret);
|
| 4801 | if (ret) {
|
| 4802 | rollback_registered(dev);
|
| 4803 | dev->reg_state = NETREG_UNREGISTERED;
|
| 4804 | }
|
| 4805 |
|
| 4806 | out:
|
| 4807 | return ret;
|
| 4808 |
|
| 4809 | err_uninit:
|
| 4810 | if (dev->netdev_ops->ndo_uninit)
|
| 4811 | dev->netdev_ops->ndo_uninit(dev);
|
| 4812 | goto out;
|
| 4813 | }
|
| 4814 |
|
| 4815 | |
| 4816 | * init_dummy_netdev - init a dummy network device for NAPI
|
| 4817 | * @dev: device to init
|
| 4818 | *
|
| 4819 | * This takes a network device structure and initialize the minimum
|
| 4820 | * amount of fields so it can be used to schedule NAPI polls without
|
| 4821 | * registering a full blown interface. This is to be used by drivers
|
| 4822 | * that need to tie several hardware interfaces to a single NAPI
|
| 4823 | * poll scheduler due to HW limitations.
|
| 4824 |
|
| 4825 | int init_dummy_netdev(struct net_device *dev)
|
| 4826 | {
|
| 4827 | |
| 4828 | * are they aren't supposed to be taken by any of the
|
| 4829 | * NAPI code and this dummy netdev is supposed to be
|
| 4830 | * only ever used for NAPI polls
|
| 4831 |
|
| 4832 | memset(dev, 0, sizeof(struct net_device));
|
| 4833 |
|
| 4834 | |
| 4835 | * register/unregister code path
|
| 4836 |
|
| 4837 | dev->reg_state = NETREG_DUMMY;
|
| 4838 |
|
| 4839 |
|
| 4840 | atomic_set(&dev->refcnt, 1);
|
| 4841 |
|
| 4842 |
|
| 4843 | INIT_LIST_HEAD(&dev->napi_list);
|
| 4844 |
|
| 4845 |
|
| 4846 | set_bit(__LINK_STATE_PRESENT, &dev->state);
|
| 4847 | set_bit(__LINK_STATE_START, &dev->state);
|
| 4848 |
|
| 4849 | return 0;
|
| 4850 | }
|
| 4851 | EXPORT_SYMBOL_GPL(init_dummy_netdev);
|
| 4852 |
|
| 4853 |
|
| 4854 | |
| 4855 | * register_netdev - register a network device
|
| 4856 | * @dev: device to register
|
| 4857 | *
|
| 4858 | * Take a completed network device structure and add it to the kernel
|
| 4859 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
|
| 4860 | * chain. 0 is returned on success. A negative errno code is returned
|
| 4861 | * on a failure to set up the device, or if the name is a duplicate.
|
| 4862 | *
|
| 4863 | * This is a wrapper around register_netdevice that takes the rtnl semaphore
|
| 4864 | * and expands the device name if you passed a format string to
|
| 4865 | * alloc_netdev.
|
| 4866 |
|
| 4867 | int register_netdev(struct net_device *dev)
|
| 4868 | {
|
| 4869 | int err;
|
| 4870 |
|
| 4871 | rtnl_lock();
|
| 4872 |
|
| 4873 | |
| 4874 | * If the name is a format string the caller wants us to do a
|
| 4875 | * name allocation.
|
| 4876 |
|
| 4877 | if (strchr(dev->name, '%')) {
|
| 4878 | err = dev_alloc_name(dev, dev->name);
|
| 4879 | if (err < 0)
|
| 4880 | goto out;
|
| 4881 | }
|
| 4882 |
|
| 4883 | err = register_netdevice(dev);
|
| 4884 | out:
|
| 4885 | rtnl_unlock();
|
| 4886 | return err;
|
| 4887 | }
|
| 4888 | EXPORT_SYMBOL(register_netdev);
|
| 4889 |
|
| 4890 | |
| 4891 | * netdev_wait_allrefs - wait until all references are gone.
|
| 4892 | *
|
| 4893 | * This is called when unregistering network devices.
|
| 4894 | *
|
| 4895 | * Any protocol or device that holds a reference should register
|
| 4896 | * for netdevice notification, and cleanup and put back the
|
| 4897 | * reference if they receive an UNREGISTER event.
|
| 4898 | * We can get stuck here if buggy protocols don't correctly
|
| 4899 | * call dev_put.
|
| 4900 |
|
| 4901 | static void netdev_wait_allrefs(struct net_device *dev)
|
| 4902 | {
|
| 4903 | unsigned long rebroadcast_time, warning_time;
|
| 4904 |
|
| 4905 | rebroadcast_time = warning_time = jiffies;
|
| 4906 | while (atomic_read(&dev->refcnt) != 0) {
|
| 4907 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
|
| 4908 | rtnl_lock();
|
| 4909 |
|
| 4910 |
|
| 4911 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
|
| 4912 |
|
| 4913 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
|
| 4914 | &dev->state)) {
|
| 4915 | |
| 4916 | * pending on unregister. If this
|
| 4917 | * happens, we simply run the queue
|
| 4918 | * unscheduled, resulting in a noop
|
| 4919 | * for this device.
|
| 4920 |
|
| 4921 | linkwatch_run_queue();
|
| 4922 | }
|
| 4923 |
|
| 4924 | __rtnl_unlock();
|
| 4925 |
|
| 4926 | rebroadcast_time = jiffies;
|
| 4927 | }
|
| 4928 |
|
| 4929 | msleep(250);
|
| 4930 |
|
| 4931 | if (time_after(jiffies, warning_time + 10 * HZ)) {
|
| 4932 | printk(KERN_EMERG "unregister_netdevice: "
|
| 4933 | "waiting for %s to become free. Usage "
|
| 4934 | "count = %d\n",
|
| 4935 | dev->name, atomic_read(&dev->refcnt));
|
| 4936 | warning_time = jiffies;
|
| 4937 | }
|
| 4938 | }
|
| 4939 | }
|
| 4940 |
|
| 4941 | |
| 4942 | *
|
| 4943 | * rtnl_lock();
|
| 4944 | * ...
|
| 4945 | * register_netdevice(x1);
|
| 4946 | * register_netdevice(x2);
|
| 4947 | * ...
|
| 4948 | * unregister_netdevice(y1);
|
| 4949 | * unregister_netdevice(y2);
|
| 4950 | * ...
|
| 4951 | * rtnl_unlock();
|
| 4952 | * free_netdev(y1);
|
| 4953 | * free_netdev(y2);
|
| 4954 | *
|
| 4955 | * We are invoked by rtnl_unlock().
|
| 4956 | * This allows us to deal with problems:
|
| 4957 | * 1) We can delete sysfs objects which invoke hotplug
|
| 4958 | * without deadlocking with linkwatch via keventd.
|
| 4959 | * 2) Since we run with the RTNL semaphore not held, we can sleep
|
| 4960 | * safely in order to wait for the netdev refcnt to drop to zero.
|
| 4961 | *
|
| 4962 | * We must not return until all unregister events added during
|
| 4963 | * the interval the lock was held have been completed.
|
| 4964 |
|
| 4965 | void netdev_run_todo(void)
|
| 4966 | {
|
| 4967 | struct list_head list;
|
| 4968 |
|
| 4969 |
|
| 4970 | list_replace_init(&net_todo_list, &list);
|
| 4971 |
|
| 4972 | __rtnl_unlock();
|
| 4973 |
|
| 4974 | while (!list_empty(&list)) {
|
| 4975 | struct net_device *dev
|
| 4976 | = list_entry(list.next, struct net_device, todo_list);
|
| 4977 | list_del(&dev->todo_list);
|
| 4978 |
|
| 4979 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
|
| 4980 | printk(KERN_ERR "network todo '%s' but state %d\n",
|
| 4981 | dev->name, dev->reg_state);
|
| 4982 | dump_stack();
|
| 4983 | continue;
|
| 4984 | }
|
| 4985 |
|
| 4986 | dev->reg_state = NETREG_UNREGISTERED;
|
| 4987 |
|
| 4988 | on_each_cpu(flush_backlog, dev, 1);
|
| 4989 |
|
| 4990 | netdev_wait_allrefs(dev);
|
| 4991 |
|
| 4992 |
|
| 4993 | BUG_ON(atomic_read(&dev->refcnt));
|
| 4994 | WARN_ON(dev->ip_ptr);
|
| 4995 | WARN_ON(dev->ip6_ptr);
|
| 4996 | WARN_ON(dev->dn_ptr);
|
| 4997 |
|
| 4998 | if (dev->destructor)
|
| 4999 | dev->destructor(dev);
|
| 5000 |
|
| 5001 |
|
| 5002 | kobject_put(&dev->dev.kobj);
|
| 5003 | }
|
| 5004 | }
|
| 5005 |
|
| 5006 | |
| 5007 | * dev_get_stats - get network device statistics
|
| 5008 | * @dev: device to get statistics from
|
| 5009 | *
|
| 5010 | * Get network statistics from device. The device driver may provide
|
| 5011 | * its own method by setting dev->netdev_ops->get_stats; otherwise
|
| 5012 | * the internal statistics structure is used.
|
| 5013 |
|
| 5014 | const struct net_device_stats *dev_get_stats(struct net_device *dev)
|
| 5015 | {
|
| 5016 | const struct net_device_ops *ops = dev->netdev_ops;
|
| 5017 |
|
| 5018 | if (ops->ndo_get_stats)
|
| 5019 | return ops->ndo_get_stats(dev);
|
| 5020 | else {
|
| 5021 | unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
|
| 5022 | struct net_device_stats *stats = &dev->stats;
|
| 5023 | unsigned int i;
|
| 5024 | struct netdev_queue *txq;
|
| 5025 |
|
| 5026 | for (i = 0; i < dev->num_tx_queues; i++) {
|
| 5027 | txq = netdev_get_tx_queue(dev, i);
|
| 5028 | tx_bytes += txq->tx_bytes;
|
| 5029 | tx_packets += txq->tx_packets;
|
| 5030 | tx_dropped += txq->tx_dropped;
|
| 5031 | }
|
| 5032 | if (tx_bytes || tx_packets || tx_dropped) {
|
| 5033 | stats->tx_bytes = tx_bytes;
|
| 5034 | stats->tx_packets = tx_packets;
|
| 5035 | stats->tx_dropped = tx_dropped;
|
| 5036 | }
|
| 5037 | return stats;
|
| 5038 | }
|
| 5039 | }
|
| 5040 | EXPORT_SYMBOL(dev_get_stats);
|
| 5041 |
|
| 5042 | static void netdev_init_one_queue(struct net_device *dev,
|
| 5043 | struct netdev_queue *queue,
|
| 5044 | void *_unused)
|
| 5045 | {
|
| 5046 | queue->dev = dev;
|
| 5047 | }
|
| 5048 |
|
| 5049 | static void netdev_init_queues(struct net_device *dev)
|
| 5050 | {
|
| 5051 | netdev_init_one_queue(dev, &dev->rx_queue, NULL);
|
| 5052 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
|
| 5053 | spin_lock_init(&dev->tx_global_lock);
|
| 5054 | }
|
| 5055 |
|
| 5056 | |
| 5057 | * alloc_netdev_mq - allocate network device
|
| 5058 | * @sizeof_priv: size of private data to allocate space for
|
| 5059 | * @name: device name format string
|
| 5060 | * @setup: callback to initialize device
|
| 5061 | * @queue_count: the number of subqueues to allocate
|
| 5062 | *
|
| 5063 | * Allocates a struct net_device with private data area for driver use
|
| 5064 | * and performs basic initialization. Also allocates subquue structs
|
| 5065 | * for each queue on the device at the end of the netdevice.
|
| 5066 |
|
| 5067 | struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
|
| 5068 | void (*setup)(struct net_device *), unsigned int queue_count)
|
| 5069 | {
|
| 5070 | struct netdev_queue *tx;
|
| 5071 | struct net_device *dev;
|
| 5072 | size_t alloc_size;
|
| 5073 | struct net_device *p;
|
| 5074 |
|
| 5075 | BUG_ON(strlen(name) >= sizeof(dev->name));
|
| 5076 |
|
| 5077 | alloc_size = sizeof(struct net_device);
|
| 5078 | if (sizeof_priv) {
|
| 5079 |
|
| 5080 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
|
| 5081 | alloc_size += sizeof_priv;
|
| 5082 | }
|
| 5083 |
|
| 5084 | alloc_size += NETDEV_ALIGN - 1;
|
| 5085 |
|
| 5086 | p = kzalloc(alloc_size, GFP_KERNEL);
|
| 5087 | if (!p) {
|
| 5088 | printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
|
| 5089 | return NULL;
|
| 5090 | }
|
| 5091 |
|
| 5092 | tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
|
| 5093 | if (!tx) {
|
| 5094 | printk(KERN_ERR "alloc_netdev: Unable to allocate "
|
| 5095 | "tx qdiscs.\n");
|
| 5096 | goto free_p;
|
| 5097 | }
|
| 5098 |
|
| 5099 | dev = PTR_ALIGN(p, NETDEV_ALIGN);
|
| 5100 | dev->padded = (char *)dev - (char *)p;
|
| 5101 |
|
| 5102 | if (dev_addr_init(dev))
|
| 5103 | goto free_tx;
|
| 5104 |
|
| 5105 | dev_unicast_init(dev);
|
| 5106 |
|
| 5107 | dev_net_set(dev, &init_net);
|
| 5108 |
|
| 5109 | dev->_tx = tx;
|
| 5110 | dev->num_tx_queues = queue_count;
|
| 5111 | dev->real_num_tx_queues = queue_count;
|
| 5112 |
|
| 5113 | dev->gso_max_size = GSO_MAX_SIZE;
|
| 5114 |
|
| 5115 | netdev_init_queues(dev);
|
| 5116 |
|
| 5117 | INIT_LIST_HEAD(&dev->napi_list);
|
| 5118 | dev->priv_flags = IFF_XMIT_DST_RELEASE;
|
| 5119 | setup(dev);
|
| 5120 | strcpy(dev->name, name);
|
| 5121 | return dev;
|
| 5122 |
|
| 5123 | free_tx:
|
| 5124 | kfree(tx);
|
| 5125 |
|
| 5126 | free_p:
|
| 5127 | kfree(p);
|
| 5128 | return NULL;
|
| 5129 | }
|
| 5130 | EXPORT_SYMBOL(alloc_netdev_mq);
|
| 5131 |
|
| 5132 | |
| 5133 | * free_netdev - free network device
|
| 5134 | * @dev: device
|
| 5135 | *
|
| 5136 | * This function does the last stage of destroying an allocated device
|
| 5137 | * interface. The reference to the device object is released.
|
| 5138 | * If this is the last reference then it will be freed.
|
| 5139 |
|
| 5140 | void free_netdev(struct net_device *dev)
|
| 5141 | {
|
| 5142 | struct napi_struct *p, *n;
|
| 5143 |
|
| 5144 | release_net(dev_net(dev));
|
| 5145 |
|
| 5146 | kfree(dev->_tx);
|
| 5147 |
|
| 5148 |
|
| 5149 | dev_addr_flush(dev);
|
| 5150 |
|
| 5151 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
|
| 5152 | netif_napi_del(p);
|
| 5153 |
|
| 5154 |
|
| 5155 | if (dev->reg_state == NETREG_UNINITIALIZED) {
|
| 5156 | kfree((char *)dev - dev->padded);
|
| 5157 | return;
|
| 5158 | }
|
| 5159 |
|
| 5160 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
|
| 5161 | dev->reg_state = NETREG_RELEASED;
|
| 5162 |
|
| 5163 |
|
| 5164 | put_device(&dev->dev);
|
| 5165 | }
|
| 5166 |
|
| 5167 | |
| 5168 | * synchronize_net - Synchronize with packet receive processing
|
| 5169 | *
|
| 5170 | * Wait for packets currently being received to be done.
|
| 5171 | * Does not block later packets from starting.
|
| 5172 |
|
| 5173 | void synchronize_net(void)
|
| 5174 | {
|
| 5175 | might_sleep();
|
| 5176 | synchronize_rcu();
|
| 5177 | }
|
| 5178 |
|
| 5179 | |
| 5180 | * unregister_netdevice - remove device from the kernel
|
| 5181 | * @dev: device
|
| 5182 | *
|
| 5183 | * This function shuts down a device interface and removes it
|
| 5184 | * from the kernel tables.
|
| 5185 | *
|
| 5186 | * Callers must hold the rtnl semaphore. You may want
|
| 5187 | * unregister_netdev() instead of this.
|
| 5188 |
|
| 5189 |
|
| 5190 | void unregister_netdevice(struct net_device *dev)
|
| 5191 | {
|
| 5192 | ASSERT_RTNL();
|
| 5193 |
|
| 5194 | rollback_registered(dev);
|
| 5195 |
|
| 5196 | net_set_todo(dev);
|
| 5197 | }
|
| 5198 |
|
| 5199 | |
| 5200 | * unregister_netdev - remove device from the kernel
|
| 5201 | * @dev: device
|
| 5202 | *
|
| 5203 | * This function shuts down a device interface and removes it
|
| 5204 | * from the kernel tables.
|
| 5205 | *
|
| 5206 | * This is just a wrapper for unregister_netdevice that takes
|
| 5207 | * the rtnl semaphore. In general you want to use this and not
|
| 5208 | * unregister_netdevice.
|
| 5209 |
|
| 5210 | void unregister_netdev(struct net_device *dev)
|
| 5211 | {
|
| 5212 | rtnl_lock();
|
| 5213 | unregister_netdevice(dev);
|
| 5214 | rtnl_unlock();
|
| 5215 | }
|
| 5216 |
|
| 5217 | EXPORT_SYMBOL(unregister_netdev);
|
| 5218 |
|
| 5219 | |
| 5220 | * dev_change_net_namespace - move device to different nethost namespace
|
| 5221 | * @dev: device
|
| 5222 | * @net: network namespace
|
| 5223 | * @pat: If not NULL name pattern to try if the current device name
|
| 5224 | * is already taken in the destination network namespace.
|
| 5225 | *
|
| 5226 | * This function shuts down a device interface and moves it
|
| 5227 | * to a new network namespace. On success 0 is returned, on
|
| 5228 | * a failure a netagive errno code is returned.
|
| 5229 | *
|
| 5230 | * Callers must hold the rtnl semaphore.
|
| 5231 |
|
| 5232 |
|
| 5233 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
|
| 5234 | {
|
| 5235 | char buf[IFNAMSIZ];
|
| 5236 | const char *destname;
|
| 5237 | int err;
|
| 5238 |
|
| 5239 | ASSERT_RTNL();
|
| 5240 |
|
| 5241 |
|
| 5242 | err = -EINVAL;
|
| 5243 | if (dev->features & NETIF_F_NETNS_LOCAL)
|
| 5244 | goto out;
|
| 5245 |
|
| 5246 | #ifdef CONFIG_SYSFS
|
| 5247 | |
| 5248 | * is enabled.
|
| 5249 |
|
| 5250 | err = -EINVAL;
|
| 5251 | if (dev->dev.parent)
|
| 5252 | goto out;
|
| 5253 | #endif
|
| 5254 |
|
| 5255 |
|
| 5256 | err = -EINVAL;
|
| 5257 | if (dev->reg_state != NETREG_REGISTERED)
|
| 5258 | goto out;
|
| 5259 |
|
| 5260 |
|
| 5261 | err = 0;
|
| 5262 | if (net_eq(dev_net(dev), net))
|
| 5263 | goto out;
|
| 5264 |
|
| 5265 | |
| 5266 | * we can use it in the destination network namespace.
|
| 5267 |
|
| 5268 | err = -EEXIST;
|
| 5269 | destname = dev->name;
|
| 5270 | if (__dev_get_by_name(net, destname)) {
|
| 5271 |
|
| 5272 | if (!pat)
|
| 5273 | goto out;
|
| 5274 | if (!dev_valid_name(pat))
|
| 5275 | goto out;
|
| 5276 | if (strchr(pat, '%')) {
|
| 5277 | if (__dev_alloc_name(net, pat, buf) < 0)
|
| 5278 | goto out;
|
| 5279 | destname = buf;
|
| 5280 | } else
|
| 5281 | destname = pat;
|
| 5282 | if (__dev_get_by_name(net, destname))
|
| 5283 | goto out;
|
| 5284 | }
|
| 5285 |
|
| 5286 | |
| 5287 | * And now a mini version of register_netdevice unregister_netdevice.
|
| 5288 |
|
| 5289 |
|
| 5290 |
|
| 5291 | dev_close(dev);
|
| 5292 |
|
| 5293 |
|
| 5294 | err = -ENODEV;
|
| 5295 | unlist_netdevice(dev);
|
| 5296 |
|
| 5297 | synchronize_net();
|
| 5298 |
|
| 5299 |
|
| 5300 | dev_shutdown(dev);
|
| 5301 |
|
| 5302 | |
| 5303 | this device. They should clean all the things.
|
| 5304 |
|
| 5305 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
|
| 5306 |
|
| 5307 | |
| 5308 | * Flush the unicast and multicast chains
|
| 5309 |
|
| 5310 | dev_unicast_flush(dev);
|
| 5311 | dev_addr_discard(dev);
|
| 5312 |
|
| 5313 | netdev_unregister_kobject(dev);
|
| 5314 |
|
| 5315 |
|
| 5316 | dev_net_set(dev, net);
|
| 5317 |
|
| 5318 |
|
| 5319 | if (destname != dev->name)
|
| 5320 | strcpy(dev->name, destname);
|
| 5321 |
|
| 5322 |
|
| 5323 | if (__dev_get_by_index(net, dev->ifindex)) {
|
| 5324 | int iflink = (dev->iflink == dev->ifindex);
|
| 5325 | dev->ifindex = dev_new_index(net);
|
| 5326 | if (iflink)
|
| 5327 | dev->iflink = dev->ifindex;
|
| 5328 | }
|
| 5329 |
|
| 5330 |
|
| 5331 | err = netdev_register_kobject(dev);
|
| 5332 | WARN_ON(err);
|
| 5333 |
|
| 5334 |
|
| 5335 | list_netdevice(dev);
|
| 5336 |
|
| 5337 |
|
| 5338 | call_netdevice_notifiers(NETDEV_REGISTER, dev);
|
| 5339 |
|
| 5340 | synchronize_net();
|
| 5341 | err = 0;
|
| 5342 | out:
|
| 5343 | return err;
|
| 5344 | }
|
| 5345 |
|
| 5346 | static int dev_cpu_callback(struct notifier_block *nfb,
|
| 5347 | unsigned long action,
|
| 5348 | void *ocpu)
|
| 5349 | {
|
| 5350 | struct sk_buff **list_skb;
|
| 5351 | struct Qdisc **list_net;
|
| 5352 | struct sk_buff *skb;
|
| 5353 | unsigned int cpu, oldcpu = (unsigned long)ocpu;
|
| 5354 | struct softnet_data *sd, *oldsd;
|
| 5355 |
|
| 5356 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
|
| 5357 | return NOTIFY_OK;
|
| 5358 |
|
| 5359 | local_irq_disable();
|
| 5360 | cpu = smp_processor_id();
|
| 5361 | sd = &per_cpu(softnet_data, cpu);
|
| 5362 | oldsd = &per_cpu(softnet_data, oldcpu);
|
| 5363 |
|
| 5364 |
|
| 5365 | list_skb = &sd->completion_queue;
|
| 5366 | while (*list_skb)
|
| 5367 | list_skb = &(*list_skb)->next;
|
| 5368 |
|
| 5369 | *list_skb = oldsd->completion_queue;
|
| 5370 | oldsd->completion_queue = NULL;
|
| 5371 |
|
| 5372 |
|
| 5373 | list_net = &sd->output_queue;
|
| 5374 | while (*list_net)
|
| 5375 | list_net = &(*list_net)->next_sched;
|
| 5376 |
|
| 5377 | *list_net = oldsd->output_queue;
|
| 5378 | oldsd->output_queue = NULL;
|
| 5379 |
|
| 5380 | raise_softirq_irqoff(NET_TX_SOFTIRQ);
|
| 5381 | local_irq_enable();
|
| 5382 |
|
| 5383 |
|
| 5384 | while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
|
| 5385 | netif_rx(skb);
|
| 5386 |
|
| 5387 | return NOTIFY_OK;
|
| 5388 | }
|
| 5389 |
|
| 5390 |
|
| 5391 | |
| 5392 | * netdev_increment_features - increment feature set by one
|
| 5393 | * @all: current feature set
|
| 5394 | * @one: new feature set
|
| 5395 | * @mask: mask feature set
|
| 5396 | *
|
| 5397 | * Computes a new feature set after adding a device with feature set
|
| 5398 | * @one to the master device with current feature set @all. Will not
|
| 5399 | * enable anything that is off in @mask. Returns the new feature set.
|
| 5400 |
|
| 5401 | unsigned long netdev_increment_features(unsigned long all, unsigned long one,
|
| 5402 | unsigned long mask)
|
| 5403 | {
|
| 5404 |
|
| 5405 | if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
|
| 5406 | all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
|
| 5407 | else if (mask & NETIF_F_ALL_CSUM) {
|
| 5408 |
|
| 5409 | if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
|
| 5410 | !(all & NETIF_F_GEN_CSUM)) {
|
| 5411 | all &= ~NETIF_F_ALL_CSUM;
|
| 5412 | all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
|
| 5413 | }
|
| 5414 |
|
| 5415 |
|
| 5416 | if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
|
| 5417 | all &= ~NETIF_F_ALL_CSUM;
|
| 5418 | all |= NETIF_F_HW_CSUM;
|
| 5419 | }
|
| 5420 | }
|
| 5421 |
|
| 5422 | one |= NETIF_F_ALL_CSUM;
|
| 5423 |
|
| 5424 | one |= all & NETIF_F_ONE_FOR_ALL;
|
| 5425 | all &= one | NETIF_F_LLTX | NETIF_F_GSO;
|
| 5426 | all |= one & mask & NETIF_F_ONE_FOR_ALL;
|
| 5427 |
|
| 5428 | return all;
|
| 5429 | }
|
| 5430 | EXPORT_SYMBOL(netdev_increment_features);
|
| 5431 |
|
| 5432 | static struct hlist_head *netdev_create_hash(void)
|
| 5433 | {
|
| 5434 | int i;
|
| 5435 | struct hlist_head *hash;
|
| 5436 |
|
| 5437 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
|
| 5438 | if (hash != NULL)
|
| 5439 | for (i = 0; i < NETDEV_HASHENTRIES; i++)
|
| 5440 | INIT_HLIST_HEAD(&hash[i]);
|
| 5441 |
|
| 5442 | return hash;
|
| 5443 | }
|
| 5444 |
|
| 5445 |
|
| 5446 | static int __net_init netdev_init(struct net *net)
|
| 5447 | {
|
| 5448 | INIT_LIST_HEAD(&net->dev_base_head);
|
| 5449 |
|
| 5450 | net->dev_name_head = netdev_create_hash();
|
| 5451 | if (net->dev_name_head == NULL)
|
| 5452 | goto err_name;
|
| 5453 |
|
| 5454 | net->dev_index_head = netdev_create_hash();
|
| 5455 | if (net->dev_index_head == NULL)
|
| 5456 | goto err_idx;
|
| 5457 |
|
| 5458 | return 0;
|
| 5459 |
|
| 5460 | err_idx:
|
| 5461 | kfree(net->dev_name_head);
|
| 5462 | err_name:
|
| 5463 | return -ENOMEM;
|
| 5464 | }
|
| 5465 |
|
| 5466 | |
| 5467 | * netdev_drivername - network driver for the device
|
| 5468 | * @dev: network device
|
| 5469 | * @buffer: buffer for resulting name
|
| 5470 | * @len: size of buffer
|
| 5471 | *
|
| 5472 | * Determine network driver for device.
|
| 5473 |
|
| 5474 | char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
|
| 5475 | {
|
| 5476 | const struct device_driver *driver;
|
| 5477 | const struct device *parent;
|
| 5478 |
|
| 5479 | if (len <= 0 || !buffer)
|
| 5480 | return buffer;
|
| 5481 | buffer[0] = 0;
|
| 5482 |
|
| 5483 | parent = dev->dev.parent;
|
| 5484 |
|
| 5485 | if (!parent)
|
| 5486 | return buffer;
|
| 5487 |
|
| 5488 | driver = parent->driver;
|
| 5489 | if (driver && driver->name)
|
| 5490 | strlcpy(buffer, driver->name, len);
|
| 5491 | return buffer;
|
| 5492 | }
|
| 5493 |
|
| 5494 | static void __net_exit netdev_exit(struct net *net)
|
| 5495 | {
|
| 5496 | kfree(net->dev_name_head);
|
| 5497 | kfree(net->dev_index_head);
|
| 5498 | }
|
| 5499 |
|
| 5500 | static struct pernet_operations __net_initdata netdev_net_ops = {
|
| 5501 | .init = netdev_init,
|
| 5502 | .exit = netdev_exit,
|
| 5503 | };
|
| 5504 |
|
| 5505 | static void __net_exit default_device_exit(struct net *net)
|
| 5506 | {
|
| 5507 | struct net_device *dev;
|
| 5508 | |
| 5509 | * Push all migratable of the network devices back to the
|
| 5510 | * initial network namespace
|
| 5511 |
|
| 5512 | rtnl_lock();
|
| 5513 | restart:
|
| 5514 | for_each_netdev(net, dev) {
|
| 5515 | int err;
|
| 5516 | char fb_name[IFNAMSIZ];
|
| 5517 |
|
| 5518 |
|
| 5519 | if (dev->features & NETIF_F_NETNS_LOCAL)
|
| 5520 | continue;
|
| 5521 |
|
| 5522 |
|
| 5523 | if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) {
|
| 5524 | dev->rtnl_link_ops->dellink(dev);
|
| 5525 | goto restart;
|
| 5526 | }
|
| 5527 |
|
| 5528 |
|
| 5529 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
|
| 5530 | err = dev_change_net_namespace(dev, &init_net, fb_name);
|
| 5531 | if (err) {
|
| 5532 | printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
|
| 5533 | __func__, dev->name, err);
|
| 5534 | BUG();
|
| 5535 | }
|
| 5536 | goto restart;
|
| 5537 | }
|
| 5538 | rtnl_unlock();
|
| 5539 | }
|
| 5540 |
|
| 5541 | static struct pernet_operations __net_initdata default_device_ops = {
|
| 5542 | .exit = default_device_exit,
|
| 5543 | };
|
| 5544 |
|
| 5545 | |
| 5546 | * Initialize the DEV module. At boot time this walks the device list and
|
| 5547 | * unhooks any devices that fail to initialise (normally hardware not
|
| 5548 | * present) and leaves us with a valid list of present and active devices.
|
| 5549 | *
|
| 5550 |
|
| 5551 |
|
| 5552 | |
| 5553 | * This is called single threaded during boot, so no need
|
| 5554 | * to take the rtnl semaphore.
|
| 5555 |
|
| 5556 | static int __init net_dev_init(void)
|
| 5557 | {
|
| 5558 | int i, rc = -ENOMEM;
|
| 5559 |
|
| 5560 | BUG_ON(!dev_boot_phase);
|
| 5561 |
|
| 5562 | if (dev_proc_init())
|
| 5563 | goto out;
|
| 5564 |
|
| 5565 | if (netdev_kobject_init())
|
| 5566 | goto out;
|
| 5567 |
|
| 5568 | INIT_LIST_HEAD(&ptype_all);
|
| 5569 | for (i = 0; i < PTYPE_HASH_SIZE; i++)
|
| 5570 | INIT_LIST_HEAD(&ptype_base[i]);
|
| 5571 |
|
| 5572 | if (register_pernet_subsys(&netdev_net_ops))
|
| 5573 | goto out;
|
| 5574 |
|
| 5575 | |
| 5576 | * Initialise the packet receive queues.
|
| 5577 |
|
| 5578 |
|
| 5579 | for_each_possible_cpu(i) {
|
| 5580 | struct softnet_data *queue;
|
| 5581 |
|
| 5582 | queue = &per_cpu(softnet_data, i);
|
| 5583 | skb_queue_head_init(&queue->input_pkt_queue);
|
| 5584 | queue->completion_queue = NULL;
|
| 5585 | INIT_LIST_HEAD(&queue->poll_list);
|
| 5586 |
|
| 5587 | queue->backlog.poll = process_backlog;
|
| 5588 | queue->backlog.weight = weight_p;
|
| 5589 | queue->backlog.gro_list = NULL;
|
| 5590 | queue->backlog.gro_count = 0;
|
| 5591 | }
|
| 5592 |
|
| 5593 | dev_boot_phase = 0;
|
| 5594 |
|
| 5595 | |
| 5596 | * is present in a network namespace the loopback device must
|
| 5597 | * be present. Since we now dynamically allocate and free the
|
| 5598 | * loopback device ensure this invariant is maintained by
|
| 5599 | * keeping the loopback device as the first device on the
|
| 5600 | * list of network devices. Ensuring the loopback devices
|
| 5601 | * is the first device that appears and the last network device
|
| 5602 | * that disappears.
|
| 5603 |
|
| 5604 | if (register_pernet_device(&loopback_net_ops))
|
| 5605 | goto out;
|
| 5606 |
|
| 5607 | if (register_pernet_device(&default_device_ops))
|
| 5608 | goto out;
|
| 5609 |
|
| 5610 | open_softirq(NET_TX_SOFTIRQ, net_tx_action);
|
| 5611 | open_softirq(NET_RX_SOFTIRQ, net_rx_action);
|
| 5612 |
|
| 5613 | hotcpu_notifier(dev_cpu_callback, 0);
|
| 5614 | dst_init();
|
| 5615 | dev_mcast_init();
|
| 5616 | rc = 0;
|
| 5617 | out:
|
| 5618 | return rc;
|
| 5619 | }
|
| 5620 |
|
| 5621 | subsys_initcall(net_dev_init);
|
| 5622 |
|
| 5623 | static int __init initialize_hashrnd(void)
|
| 5624 | {
|
| 5625 | get_random_bytes(&skb_tx_hashrnd, sizeof(skb_tx_hashrnd));
|
| 5626 | return 0;
|
| 5627 | }
|
| 5628 |
|
| 5629 | late_initcall_sync(initialize_hashrnd);
|
| 5630 |
|
| 5631 | EXPORT_SYMBOL(__dev_get_by_index);
|
| 5632 | EXPORT_SYMBOL(__dev_get_by_name);
|
| 5633 | EXPORT_SYMBOL(__dev_remove_pack);
|
| 5634 | EXPORT_SYMBOL(dev_valid_name);
|
| 5635 | EXPORT_SYMBOL(dev_add_pack);
|
| 5636 | EXPORT_SYMBOL(dev_alloc_name);
|
| 5637 | EXPORT_SYMBOL(dev_close);
|
| 5638 | EXPORT_SYMBOL(dev_get_by_flags);
|
| 5639 | EXPORT_SYMBOL(dev_get_by_index);
|
| 5640 | EXPORT_SYMBOL(dev_get_by_name);
|
| 5641 | EXPORT_SYMBOL(dev_open);
|
| 5642 | EXPORT_SYMBOL(dev_queue_xmit);
|
| 5643 | EXPORT_SYMBOL(dev_remove_pack);
|
| 5644 | EXPORT_SYMBOL(dev_set_allmulti);
|
| 5645 | EXPORT_SYMBOL(dev_set_promiscuity);
|
| 5646 | EXPORT_SYMBOL(dev_change_flags);
|
| 5647 | EXPORT_SYMBOL(dev_set_mtu);
|
| 5648 | EXPORT_SYMBOL(dev_set_mac_address);
|
| 5649 | EXPORT_SYMBOL(free_netdev);
|
| 5650 | EXPORT_SYMBOL(netdev_boot_setup_check);
|
| 5651 | EXPORT_SYMBOL(netdev_set_master);
|
| 5652 | EXPORT_SYMBOL(netdev_state_change);
|
| 5653 | EXPORT_SYMBOL(netif_receive_skb);
|
| 5654 | EXPORT_SYMBOL(netif_rx);
|
| 5655 | EXPORT_SYMBOL(register_gifconf);
|
| 5656 | EXPORT_SYMBOL(register_netdevice);
|
| 5657 | EXPORT_SYMBOL(register_netdevice_notifier);
|
| 5658 | EXPORT_SYMBOL(skb_checksum_help);
|
| 5659 | EXPORT_SYMBOL(synchronize_net);
|
| 5660 | EXPORT_SYMBOL(unregister_netdevice);
|
| 5661 | EXPORT_SYMBOL(unregister_netdevice_notifier);
|
| 5662 | EXPORT_SYMBOL(net_enable_timestamp);
|
| 5663 | EXPORT_SYMBOL(net_disable_timestamp);
|
| 5664 | EXPORT_SYMBOL(dev_get_flags);
|
| 5665 |
|
| 5666 | EXPORT_SYMBOL(dev_load);
|
| 5667 |
|
| 5668 | EXPORT_PER_CPU_SYMBOL(softnet_data);
|