2 * NET3 Protocol independent device support routines.
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.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
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
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
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
51 * Rudi Cilibrasi : Pass the right thing to
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
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
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
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/hash.h>
83 #include <linux/slab.h>
84 #include <linux/sched.h>
85 #include <linux/mutex.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/notifier.h>
97 #include <linux/skbuff.h>
98 #include <net/net_namespace.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/stat.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include <linux/highmem.h>
109 #include <linux/init.h>
110 #include <linux/kmod.h>
111 #include <linux/module.h>
112 #include <linux/netpoll.h>
113 #include <linux/rcupdate.h>
114 #include <linux/delay.h>
115 #include <net/wext.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
126 #include <linux/ipv6.h>
127 #include <linux/in.h>
128 #include <linux/jhash.h>
129 #include <linux/random.h>
130 #include <trace/events/napi.h>
131 #include <trace/events/net.h>
132 #include <trace/events/skb.h>
133 #include <linux/pci.h>
134 #include <linux/inetdevice.h>
135 #include <linux/cpu_rmap.h>
136 #include <linux/if_tunnel.h>
138 #include "net-sysfs.h"
140 /* Instead of increasing this, you should create a hash table. */
141 #define MAX_GRO_SKBS 8
143 /* This should be increased if a protocol with a bigger head is added. */
144 #define GRO_MAX_HEAD (MAX_HEADER + 128)
147 * The list of packet types we will receive (as opposed to discard)
148 * and the routines to invoke.
150 * Why 16. Because with 16 the only overlap we get on a hash of the
151 * low nibble of the protocol value is RARP/SNAP/X.25.
153 * NOTE: That is no longer true with the addition of VLAN tags. Not
154 * sure which should go first, but I bet it won't make much
155 * difference if we are running VLANs. The good news is that
156 * this protocol won't be in the list unless compiled in, so
157 * the average user (w/out VLANs) will not be adversely affected.
174 #define PTYPE_HASH_SIZE (16)
175 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
177 static DEFINE_SPINLOCK(ptype_lock
);
178 static struct list_head ptype_base
[PTYPE_HASH_SIZE
] __read_mostly
;
179 static struct list_head ptype_all __read_mostly
; /* Taps */
182 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
185 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
187 * Writers must hold the rtnl semaphore while they loop through the
188 * dev_base_head list, and hold dev_base_lock for writing when they do the
189 * actual updates. This allows pure readers to access the list even
190 * while a writer is preparing to update it.
192 * To put it another way, dev_base_lock is held for writing only to
193 * protect against pure readers; the rtnl semaphore provides the
194 * protection against other writers.
196 * See, for example usages, register_netdevice() and
197 * unregister_netdevice(), which must be called with the rtnl
200 DEFINE_RWLOCK(dev_base_lock
);
201 EXPORT_SYMBOL(dev_base_lock
);
203 static inline void dev_base_seq_inc(struct net
*net
)
205 while (++net
->dev_base_seq
== 0);
208 static inline struct hlist_head
*dev_name_hash(struct net
*net
, const char *name
)
210 unsigned hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
211 return &net
->dev_name_head
[hash_32(hash
, NETDEV_HASHBITS
)];
214 static inline struct hlist_head
*dev_index_hash(struct net
*net
, int ifindex
)
216 return &net
->dev_index_head
[ifindex
& (NETDEV_HASHENTRIES
- 1)];
219 static inline void rps_lock(struct softnet_data
*sd
)
222 spin_lock(&sd
->input_pkt_queue
.lock
);
226 static inline void rps_unlock(struct softnet_data
*sd
)
229 spin_unlock(&sd
->input_pkt_queue
.lock
);
233 /* Device list insertion */
234 static int list_netdevice(struct net_device
*dev
)
236 struct net
*net
= dev_net(dev
);
240 write_lock_bh(&dev_base_lock
);
241 list_add_tail_rcu(&dev
->dev_list
, &net
->dev_base_head
);
242 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
243 hlist_add_head_rcu(&dev
->index_hlist
,
244 dev_index_hash(net
, dev
->ifindex
));
245 write_unlock_bh(&dev_base_lock
);
247 dev_base_seq_inc(net
);
252 /* Device list removal
253 * caller must respect a RCU grace period before freeing/reusing dev
255 static void unlist_netdevice(struct net_device
*dev
)
259 /* Unlink dev from the device chain */
260 write_lock_bh(&dev_base_lock
);
261 list_del_rcu(&dev
->dev_list
);
262 hlist_del_rcu(&dev
->name_hlist
);
263 hlist_del_rcu(&dev
->index_hlist
);
264 write_unlock_bh(&dev_base_lock
);
266 dev_base_seq_inc(dev_net(dev
));
273 static RAW_NOTIFIER_HEAD(netdev_chain
);
276 * Device drivers call our routines to queue packets here. We empty the
277 * queue in the local softnet handler.
280 DEFINE_PER_CPU_ALIGNED(struct softnet_data
, softnet_data
);
281 EXPORT_PER_CPU_SYMBOL(softnet_data
);
283 #ifdef CONFIG_LOCKDEP
285 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
286 * according to dev->type
288 static const unsigned short netdev_lock_type
[] =
289 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
290 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
291 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
292 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
293 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
294 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
295 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
296 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
297 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
298 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
299 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
300 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
301 ARPHRD_FCFABRIC
, ARPHRD_IEEE802_TR
, ARPHRD_IEEE80211
,
302 ARPHRD_IEEE80211_PRISM
, ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_PHONET
,
303 ARPHRD_PHONET_PIPE
, ARPHRD_IEEE802154
,
304 ARPHRD_VOID
, ARPHRD_NONE
};
306 static const char *const netdev_lock_name
[] =
307 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
308 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
309 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
310 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
311 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
312 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
313 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
314 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
315 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
316 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
317 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
318 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
319 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
320 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
321 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
322 "_xmit_VOID", "_xmit_NONE"};
324 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
325 static struct lock_class_key netdev_addr_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
327 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
331 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
332 if (netdev_lock_type
[i
] == dev_type
)
334 /* the last key is used by default */
335 return ARRAY_SIZE(netdev_lock_type
) - 1;
338 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
339 unsigned short dev_type
)
343 i
= netdev_lock_pos(dev_type
);
344 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
345 netdev_lock_name
[i
]);
348 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
352 i
= netdev_lock_pos(dev
->type
);
353 lockdep_set_class_and_name(&dev
->addr_list_lock
,
354 &netdev_addr_lock_key
[i
],
355 netdev_lock_name
[i
]);
358 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
359 unsigned short dev_type
)
362 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
367 /*******************************************************************************
369 Protocol management and registration routines
371 *******************************************************************************/
374 * Add a protocol ID to the list. Now that the input handler is
375 * smarter we can dispense with all the messy stuff that used to be
378 * BEWARE!!! Protocol handlers, mangling input packets,
379 * MUST BE last in hash buckets and checking protocol handlers
380 * MUST start from promiscuous ptype_all chain in net_bh.
381 * It is true now, do not change it.
382 * Explanation follows: if protocol handler, mangling packet, will
383 * be the first on list, it is not able to sense, that packet
384 * is cloned and should be copied-on-write, so that it will
385 * change it and subsequent readers will get broken packet.
389 static inline struct list_head
*ptype_head(const struct packet_type
*pt
)
391 if (pt
->type
== htons(ETH_P_ALL
))
394 return &ptype_base
[ntohs(pt
->type
) & PTYPE_HASH_MASK
];
398 * dev_add_pack - add packet handler
399 * @pt: packet type declaration
401 * Add a protocol handler to the networking stack. The passed &packet_type
402 * is linked into kernel lists and may not be freed until it has been
403 * removed from the kernel lists.
405 * This call does not sleep therefore it can not
406 * guarantee all CPU's that are in middle of receiving packets
407 * will see the new packet type (until the next received packet).
410 void dev_add_pack(struct packet_type
*pt
)
412 struct list_head
*head
= ptype_head(pt
);
414 spin_lock(&ptype_lock
);
415 list_add_rcu(&pt
->list
, head
);
416 spin_unlock(&ptype_lock
);
418 EXPORT_SYMBOL(dev_add_pack
);
421 * __dev_remove_pack - remove packet handler
422 * @pt: packet type declaration
424 * Remove a protocol handler that was previously added to the kernel
425 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
426 * from the kernel lists and can be freed or reused once this function
429 * The packet type might still be in use by receivers
430 * and must not be freed until after all the CPU's have gone
431 * through a quiescent state.
433 void __dev_remove_pack(struct packet_type
*pt
)
435 struct list_head
*head
= ptype_head(pt
);
436 struct packet_type
*pt1
;
438 spin_lock(&ptype_lock
);
440 list_for_each_entry(pt1
, head
, list
) {
442 list_del_rcu(&pt
->list
);
447 printk(KERN_WARNING
"dev_remove_pack: %p not found.\n", pt
);
449 spin_unlock(&ptype_lock
);
451 EXPORT_SYMBOL(__dev_remove_pack
);
454 * dev_remove_pack - remove packet handler
455 * @pt: packet type declaration
457 * Remove a protocol handler that was previously added to the kernel
458 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
459 * from the kernel lists and can be freed or reused once this function
462 * This call sleeps to guarantee that no CPU is looking at the packet
465 void dev_remove_pack(struct packet_type
*pt
)
467 __dev_remove_pack(pt
);
471 EXPORT_SYMBOL(dev_remove_pack
);
473 /******************************************************************************
475 Device Boot-time Settings Routines
477 *******************************************************************************/
479 /* Boot time configuration table */
480 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
483 * netdev_boot_setup_add - add new setup entry
484 * @name: name of the device
485 * @map: configured settings for the device
487 * Adds new setup entry to the dev_boot_setup list. The function
488 * returns 0 on error and 1 on success. This is a generic routine to
491 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
493 struct netdev_boot_setup
*s
;
497 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
498 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
499 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
500 strlcpy(s
[i
].name
, name
, IFNAMSIZ
);
501 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
506 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
510 * netdev_boot_setup_check - check boot time settings
511 * @dev: the netdevice
513 * Check boot time settings for the device.
514 * The found settings are set for the device to be used
515 * later in the device probing.
516 * Returns 0 if no settings found, 1 if they are.
518 int netdev_boot_setup_check(struct net_device
*dev
)
520 struct netdev_boot_setup
*s
= dev_boot_setup
;
523 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
524 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
525 !strcmp(dev
->name
, s
[i
].name
)) {
526 dev
->irq
= s
[i
].map
.irq
;
527 dev
->base_addr
= s
[i
].map
.base_addr
;
528 dev
->mem_start
= s
[i
].map
.mem_start
;
529 dev
->mem_end
= s
[i
].map
.mem_end
;
535 EXPORT_SYMBOL(netdev_boot_setup_check
);
539 * netdev_boot_base - get address from boot time settings
540 * @prefix: prefix for network device
541 * @unit: id for network device
543 * Check boot time settings for the base address of device.
544 * The found settings are set for the device to be used
545 * later in the device probing.
546 * Returns 0 if no settings found.
548 unsigned long netdev_boot_base(const char *prefix
, int unit
)
550 const struct netdev_boot_setup
*s
= dev_boot_setup
;
554 sprintf(name
, "%s%d", prefix
, unit
);
557 * If device already registered then return base of 1
558 * to indicate not to probe for this interface
560 if (__dev_get_by_name(&init_net
, name
))
563 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
564 if (!strcmp(name
, s
[i
].name
))
565 return s
[i
].map
.base_addr
;
570 * Saves at boot time configured settings for any netdevice.
572 int __init
netdev_boot_setup(char *str
)
577 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
582 memset(&map
, 0, sizeof(map
));
586 map
.base_addr
= ints
[2];
588 map
.mem_start
= ints
[3];
590 map
.mem_end
= ints
[4];
592 /* Add new entry to the list */
593 return netdev_boot_setup_add(str
, &map
);
596 __setup("netdev=", netdev_boot_setup
);
598 /*******************************************************************************
600 Device Interface Subroutines
602 *******************************************************************************/
605 * __dev_get_by_name - find a device by its name
606 * @net: the applicable net namespace
607 * @name: name to find
609 * Find an interface by name. Must be called under RTNL semaphore
610 * or @dev_base_lock. If the name is found a pointer to the device
611 * is returned. If the name is not found then %NULL is returned. The
612 * reference counters are not incremented so the caller must be
613 * careful with locks.
616 struct net_device
*__dev_get_by_name(struct net
*net
, const char *name
)
618 struct hlist_node
*p
;
619 struct net_device
*dev
;
620 struct hlist_head
*head
= dev_name_hash(net
, name
);
622 hlist_for_each_entry(dev
, p
, head
, name_hlist
)
623 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
628 EXPORT_SYMBOL(__dev_get_by_name
);
631 * dev_get_by_name_rcu - find a device by its name
632 * @net: the applicable net namespace
633 * @name: name to find
635 * Find an interface by name.
636 * If the name is found a pointer to the device is returned.
637 * If the name is not found then %NULL is returned.
638 * The reference counters are not incremented so the caller must be
639 * careful with locks. The caller must hold RCU lock.
642 struct net_device
*dev_get_by_name_rcu(struct net
*net
, const char *name
)
644 struct hlist_node
*p
;
645 struct net_device
*dev
;
646 struct hlist_head
*head
= dev_name_hash(net
, name
);
648 hlist_for_each_entry_rcu(dev
, p
, head
, name_hlist
)
649 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
654 EXPORT_SYMBOL(dev_get_by_name_rcu
);
657 * dev_get_by_name - find a device by its name
658 * @net: the applicable net namespace
659 * @name: name to find
661 * Find an interface by name. This can be called from any
662 * context and does its own locking. The returned handle has
663 * the usage count incremented and the caller must use dev_put() to
664 * release it when it is no longer needed. %NULL is returned if no
665 * matching device is found.
668 struct net_device
*dev_get_by_name(struct net
*net
, const char *name
)
670 struct net_device
*dev
;
673 dev
= dev_get_by_name_rcu(net
, name
);
679 EXPORT_SYMBOL(dev_get_by_name
);
682 * __dev_get_by_index - find a device by its ifindex
683 * @net: the applicable net namespace
684 * @ifindex: index of device
686 * Search for an interface by index. Returns %NULL if the device
687 * is not found or a pointer to the device. The device has not
688 * had its reference counter increased so the caller must be careful
689 * about locking. The caller must hold either the RTNL semaphore
693 struct net_device
*__dev_get_by_index(struct net
*net
, int ifindex
)
695 struct hlist_node
*p
;
696 struct net_device
*dev
;
697 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
699 hlist_for_each_entry(dev
, p
, head
, index_hlist
)
700 if (dev
->ifindex
== ifindex
)
705 EXPORT_SYMBOL(__dev_get_by_index
);
708 * dev_get_by_index_rcu - find a device by its ifindex
709 * @net: the applicable net namespace
710 * @ifindex: index of device
712 * Search for an interface by index. Returns %NULL if the device
713 * is not found or a pointer to the device. The device has not
714 * had its reference counter increased so the caller must be careful
715 * about locking. The caller must hold RCU lock.
718 struct net_device
*dev_get_by_index_rcu(struct net
*net
, int ifindex
)
720 struct hlist_node
*p
;
721 struct net_device
*dev
;
722 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
724 hlist_for_each_entry_rcu(dev
, p
, head
, index_hlist
)
725 if (dev
->ifindex
== ifindex
)
730 EXPORT_SYMBOL(dev_get_by_index_rcu
);
734 * dev_get_by_index - find a device by its ifindex
735 * @net: the applicable net namespace
736 * @ifindex: index of device
738 * Search for an interface by index. Returns NULL if the device
739 * is not found or a pointer to the device. The device returned has
740 * had a reference added and the pointer is safe until the user calls
741 * dev_put to indicate they have finished with it.
744 struct net_device
*dev_get_by_index(struct net
*net
, int ifindex
)
746 struct net_device
*dev
;
749 dev
= dev_get_by_index_rcu(net
, ifindex
);
755 EXPORT_SYMBOL(dev_get_by_index
);
758 * dev_getbyhwaddr_rcu - find a device by its hardware address
759 * @net: the applicable net namespace
760 * @type: media type of device
761 * @ha: hardware address
763 * Search for an interface by MAC address. Returns NULL if the device
764 * is not found or a pointer to the device.
765 * The caller must hold RCU or RTNL.
766 * The returned device has not had its ref count increased
767 * and the caller must therefore be careful about locking
771 struct net_device
*dev_getbyhwaddr_rcu(struct net
*net
, unsigned short type
,
774 struct net_device
*dev
;
776 for_each_netdev_rcu(net
, dev
)
777 if (dev
->type
== type
&&
778 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
783 EXPORT_SYMBOL(dev_getbyhwaddr_rcu
);
785 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
787 struct net_device
*dev
;
790 for_each_netdev(net
, dev
)
791 if (dev
->type
== type
)
796 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
798 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
800 struct net_device
*dev
, *ret
= NULL
;
803 for_each_netdev_rcu(net
, dev
)
804 if (dev
->type
== type
) {
812 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
815 * dev_get_by_flags_rcu - find any device with given flags
816 * @net: the applicable net namespace
817 * @if_flags: IFF_* values
818 * @mask: bitmask of bits in if_flags to check
820 * Search for any interface with the given flags. Returns NULL if a device
821 * is not found or a pointer to the device. Must be called inside
822 * rcu_read_lock(), and result refcount is unchanged.
825 struct net_device
*dev_get_by_flags_rcu(struct net
*net
, unsigned short if_flags
,
828 struct net_device
*dev
, *ret
;
831 for_each_netdev_rcu(net
, dev
) {
832 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
839 EXPORT_SYMBOL(dev_get_by_flags_rcu
);
842 * dev_valid_name - check if name is okay for network device
845 * Network device names need to be valid file names to
846 * to allow sysfs to work. We also disallow any kind of
849 int dev_valid_name(const char *name
)
853 if (strlen(name
) >= IFNAMSIZ
)
855 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
859 if (*name
== '/' || isspace(*name
))
865 EXPORT_SYMBOL(dev_valid_name
);
868 * __dev_alloc_name - allocate a name for a device
869 * @net: network namespace to allocate the device name in
870 * @name: name format string
871 * @buf: scratch buffer and result name string
873 * Passed a format string - eg "lt%d" it will try and find a suitable
874 * id. It scans list of devices to build up a free map, then chooses
875 * the first empty slot. The caller must hold the dev_base or rtnl lock
876 * while allocating the name and adding the device in order to avoid
878 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
879 * Returns the number of the unit assigned or a negative errno code.
882 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
886 const int max_netdevices
= 8*PAGE_SIZE
;
887 unsigned long *inuse
;
888 struct net_device
*d
;
890 p
= strnchr(name
, IFNAMSIZ
-1, '%');
893 * Verify the string as this thing may have come from
894 * the user. There must be either one "%d" and no other "%"
897 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
900 /* Use one page as a bit array of possible slots */
901 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
905 for_each_netdev(net
, d
) {
906 if (!sscanf(d
->name
, name
, &i
))
908 if (i
< 0 || i
>= max_netdevices
)
911 /* avoid cases where sscanf is not exact inverse of printf */
912 snprintf(buf
, IFNAMSIZ
, name
, i
);
913 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
917 i
= find_first_zero_bit(inuse
, max_netdevices
);
918 free_page((unsigned long) inuse
);
922 snprintf(buf
, IFNAMSIZ
, name
, i
);
923 if (!__dev_get_by_name(net
, buf
))
926 /* It is possible to run out of possible slots
927 * when the name is long and there isn't enough space left
928 * for the digits, or if all bits are used.
934 * dev_alloc_name - allocate a name for a device
936 * @name: name format string
938 * Passed a format string - eg "lt%d" it will try and find a suitable
939 * id. It scans list of devices to build up a free map, then chooses
940 * the first empty slot. The caller must hold the dev_base or rtnl lock
941 * while allocating the name and adding the device in order to avoid
943 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
944 * Returns the number of the unit assigned or a negative errno code.
947 int dev_alloc_name(struct net_device
*dev
, const char *name
)
953 BUG_ON(!dev_net(dev
));
955 ret
= __dev_alloc_name(net
, name
, buf
);
957 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
960 EXPORT_SYMBOL(dev_alloc_name
);
962 static int dev_get_valid_name(struct net_device
*dev
, const char *name
)
966 BUG_ON(!dev_net(dev
));
969 if (!dev_valid_name(name
))
972 if (strchr(name
, '%'))
973 return dev_alloc_name(dev
, name
);
974 else if (__dev_get_by_name(net
, name
))
976 else if (dev
->name
!= name
)
977 strlcpy(dev
->name
, name
, IFNAMSIZ
);
983 * dev_change_name - change name of a device
985 * @newname: name (or format string) must be at least IFNAMSIZ
987 * Change name of a device, can pass format strings "eth%d".
990 int dev_change_name(struct net_device
*dev
, const char *newname
)
992 char oldname
[IFNAMSIZ
];
998 BUG_ON(!dev_net(dev
));
1001 if (dev
->flags
& IFF_UP
)
1004 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0)
1007 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
1009 err
= dev_get_valid_name(dev
, newname
);
1014 ret
= device_rename(&dev
->dev
, dev
->name
);
1016 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1020 write_lock_bh(&dev_base_lock
);
1021 hlist_del_rcu(&dev
->name_hlist
);
1022 write_unlock_bh(&dev_base_lock
);
1026 write_lock_bh(&dev_base_lock
);
1027 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
1028 write_unlock_bh(&dev_base_lock
);
1030 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
1031 ret
= notifier_to_errno(ret
);
1034 /* err >= 0 after dev_alloc_name() or stores the first errno */
1037 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1041 "%s: name change rollback failed: %d.\n",
1050 * dev_set_alias - change ifalias of a device
1052 * @alias: name up to IFALIASZ
1053 * @len: limit of bytes to copy from info
1055 * Set ifalias for a device,
1057 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
1061 if (len
>= IFALIASZ
)
1066 kfree(dev
->ifalias
);
1067 dev
->ifalias
= NULL
;
1072 dev
->ifalias
= krealloc(dev
->ifalias
, len
+ 1, GFP_KERNEL
);
1076 strlcpy(dev
->ifalias
, alias
, len
+1);
1082 * netdev_features_change - device changes features
1083 * @dev: device to cause notification
1085 * Called to indicate a device has changed features.
1087 void netdev_features_change(struct net_device
*dev
)
1089 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1091 EXPORT_SYMBOL(netdev_features_change
);
1094 * netdev_state_change - device changes state
1095 * @dev: device to cause notification
1097 * Called to indicate a device has changed state. This function calls
1098 * the notifier chains for netdev_chain and sends a NEWLINK message
1099 * to the routing socket.
1101 void netdev_state_change(struct net_device
*dev
)
1103 if (dev
->flags
& IFF_UP
) {
1104 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
1105 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
1108 EXPORT_SYMBOL(netdev_state_change
);
1110 int netdev_bonding_change(struct net_device
*dev
, unsigned long event
)
1112 return call_netdevice_notifiers(event
, dev
);
1114 EXPORT_SYMBOL(netdev_bonding_change
);
1117 * dev_load - load a network module
1118 * @net: the applicable net namespace
1119 * @name: name of interface
1121 * If a network interface is not present and the process has suitable
1122 * privileges this function loads the module. If module loading is not
1123 * available in this kernel then it becomes a nop.
1126 void dev_load(struct net
*net
, const char *name
)
1128 struct net_device
*dev
;
1132 dev
= dev_get_by_name_rcu(net
, name
);
1136 if (no_module
&& capable(CAP_NET_ADMIN
))
1137 no_module
= request_module("netdev-%s", name
);
1138 if (no_module
&& capable(CAP_SYS_MODULE
)) {
1139 if (!request_module("%s", name
))
1140 pr_err("Loading kernel module for a network device "
1141 "with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s "
1145 EXPORT_SYMBOL(dev_load
);
1147 static int __dev_open(struct net_device
*dev
)
1149 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1154 if (!netif_device_present(dev
))
1157 ret
= call_netdevice_notifiers(NETDEV_PRE_UP
, dev
);
1158 ret
= notifier_to_errno(ret
);
1162 set_bit(__LINK_STATE_START
, &dev
->state
);
1164 if (ops
->ndo_validate_addr
)
1165 ret
= ops
->ndo_validate_addr(dev
);
1167 if (!ret
&& ops
->ndo_open
)
1168 ret
= ops
->ndo_open(dev
);
1171 clear_bit(__LINK_STATE_START
, &dev
->state
);
1173 dev
->flags
|= IFF_UP
;
1174 net_dmaengine_get();
1175 dev_set_rx_mode(dev
);
1183 * dev_open - prepare an interface for use.
1184 * @dev: device to open
1186 * Takes a device from down to up state. The device's private open
1187 * function is invoked and then the multicast lists are loaded. Finally
1188 * the device is moved into the up state and a %NETDEV_UP message is
1189 * sent to the netdev notifier chain.
1191 * Calling this function on an active interface is a nop. On a failure
1192 * a negative errno code is returned.
1194 int dev_open(struct net_device
*dev
)
1198 if (dev
->flags
& IFF_UP
)
1201 ret
= __dev_open(dev
);
1205 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1206 call_netdevice_notifiers(NETDEV_UP
, dev
);
1210 EXPORT_SYMBOL(dev_open
);
1212 static int __dev_close_many(struct list_head
*head
)
1214 struct net_device
*dev
;
1219 list_for_each_entry(dev
, head
, unreg_list
) {
1220 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1222 clear_bit(__LINK_STATE_START
, &dev
->state
);
1224 /* Synchronize to scheduled poll. We cannot touch poll list, it
1225 * can be even on different cpu. So just clear netif_running().
1227 * dev->stop() will invoke napi_disable() on all of it's
1228 * napi_struct instances on this device.
1230 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1233 dev_deactivate_many(head
);
1235 list_for_each_entry(dev
, head
, unreg_list
) {
1236 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1239 * Call the device specific close. This cannot fail.
1240 * Only if device is UP
1242 * We allow it to be called even after a DETACH hot-plug
1248 dev
->flags
&= ~IFF_UP
;
1249 net_dmaengine_put();
1255 static int __dev_close(struct net_device
*dev
)
1260 list_add(&dev
->unreg_list
, &single
);
1261 retval
= __dev_close_many(&single
);
1266 static int dev_close_many(struct list_head
*head
)
1268 struct net_device
*dev
, *tmp
;
1269 LIST_HEAD(tmp_list
);
1271 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
)
1272 if (!(dev
->flags
& IFF_UP
))
1273 list_move(&dev
->unreg_list
, &tmp_list
);
1275 __dev_close_many(head
);
1277 list_for_each_entry(dev
, head
, unreg_list
) {
1278 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1279 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1282 /* rollback_registered_many needs the complete original list */
1283 list_splice(&tmp_list
, head
);
1288 * dev_close - shutdown an interface.
1289 * @dev: device to shutdown
1291 * This function moves an active device into down state. A
1292 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1293 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1296 int dev_close(struct net_device
*dev
)
1298 if (dev
->flags
& IFF_UP
) {
1301 list_add(&dev
->unreg_list
, &single
);
1302 dev_close_many(&single
);
1307 EXPORT_SYMBOL(dev_close
);
1311 * dev_disable_lro - disable Large Receive Offload on a device
1314 * Disable Large Receive Offload (LRO) on a net device. Must be
1315 * called under RTNL. This is needed if received packets may be
1316 * forwarded to another interface.
1318 void dev_disable_lro(struct net_device
*dev
)
1323 * If we're trying to disable lro on a vlan device
1324 * use the underlying physical device instead
1326 if (is_vlan_dev(dev
))
1327 dev
= vlan_dev_real_dev(dev
);
1329 if (dev
->ethtool_ops
&& dev
->ethtool_ops
->get_flags
)
1330 flags
= dev
->ethtool_ops
->get_flags(dev
);
1332 flags
= ethtool_op_get_flags(dev
);
1334 if (!(flags
& ETH_FLAG_LRO
))
1337 __ethtool_set_flags(dev
, flags
& ~ETH_FLAG_LRO
);
1338 if (unlikely(dev
->features
& NETIF_F_LRO
))
1339 netdev_WARN(dev
, "failed to disable LRO!\n");
1341 EXPORT_SYMBOL(dev_disable_lro
);
1344 static int dev_boot_phase
= 1;
1347 * register_netdevice_notifier - register a network notifier block
1350 * Register a notifier to be called when network device events occur.
1351 * The notifier passed is linked into the kernel structures and must
1352 * not be reused until it has been unregistered. A negative errno code
1353 * is returned on a failure.
1355 * When registered all registration and up events are replayed
1356 * to the new notifier to allow device to have a race free
1357 * view of the network device list.
1360 int register_netdevice_notifier(struct notifier_block
*nb
)
1362 struct net_device
*dev
;
1363 struct net_device
*last
;
1368 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1374 for_each_netdev(net
, dev
) {
1375 err
= nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1376 err
= notifier_to_errno(err
);
1380 if (!(dev
->flags
& IFF_UP
))
1383 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1394 for_each_netdev(net
, dev
) {
1398 if (dev
->flags
& IFF_UP
) {
1399 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1400 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1402 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1403 nb
->notifier_call(nb
, NETDEV_UNREGISTER_BATCH
, dev
);
1407 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1410 EXPORT_SYMBOL(register_netdevice_notifier
);
1413 * unregister_netdevice_notifier - unregister a network notifier block
1416 * Unregister a notifier previously registered by
1417 * register_netdevice_notifier(). The notifier is unlinked into the
1418 * kernel structures and may then be reused. A negative errno code
1419 * is returned on a failure.
1422 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1427 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1431 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1434 * call_netdevice_notifiers - call all network notifier blocks
1435 * @val: value passed unmodified to notifier function
1436 * @dev: net_device pointer passed unmodified to notifier function
1438 * Call all network notifier blocks. Parameters and return value
1439 * are as for raw_notifier_call_chain().
1442 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1445 return raw_notifier_call_chain(&netdev_chain
, val
, dev
);
1447 EXPORT_SYMBOL(call_netdevice_notifiers
);
1449 /* When > 0 there are consumers of rx skb time stamps */
1450 static atomic_t netstamp_needed
= ATOMIC_INIT(0);
1452 void net_enable_timestamp(void)
1454 atomic_inc(&netstamp_needed
);
1456 EXPORT_SYMBOL(net_enable_timestamp
);
1458 void net_disable_timestamp(void)
1460 atomic_dec(&netstamp_needed
);
1462 EXPORT_SYMBOL(net_disable_timestamp
);
1464 static inline void net_timestamp_set(struct sk_buff
*skb
)
1466 if (atomic_read(&netstamp_needed
))
1467 __net_timestamp(skb
);
1469 skb
->tstamp
.tv64
= 0;
1472 static inline void net_timestamp_check(struct sk_buff
*skb
)
1474 if (!skb
->tstamp
.tv64
&& atomic_read(&netstamp_needed
))
1475 __net_timestamp(skb
);
1478 static inline bool is_skb_forwardable(struct net_device
*dev
,
1479 struct sk_buff
*skb
)
1483 if (!(dev
->flags
& IFF_UP
))
1486 len
= dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
;
1487 if (skb
->len
<= len
)
1490 /* if TSO is enabled, we don't care about the length as the packet
1491 * could be forwarded without being segmented before
1493 if (skb_is_gso(skb
))
1500 * dev_forward_skb - loopback an skb to another netif
1502 * @dev: destination network device
1503 * @skb: buffer to forward
1506 * NET_RX_SUCCESS (no congestion)
1507 * NET_RX_DROP (packet was dropped, but freed)
1509 * dev_forward_skb can be used for injecting an skb from the
1510 * start_xmit function of one device into the receive queue
1511 * of another device.
1513 * The receiving device may be in another namespace, so
1514 * we have to clear all information in the skb that could
1515 * impact namespace isolation.
1517 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1522 if (unlikely(!is_skb_forwardable(dev
, skb
))) {
1523 atomic_long_inc(&dev
->rx_dropped
);
1527 skb_set_dev(skb
, dev
);
1528 skb
->tstamp
.tv64
= 0;
1529 skb
->pkt_type
= PACKET_HOST
;
1530 skb
->protocol
= eth_type_trans(skb
, dev
);
1531 return netif_rx(skb
);
1533 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1535 static inline int deliver_skb(struct sk_buff
*skb
,
1536 struct packet_type
*pt_prev
,
1537 struct net_device
*orig_dev
)
1539 atomic_inc(&skb
->users
);
1540 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1544 * Support routine. Sends outgoing frames to any network
1545 * taps currently in use.
1548 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1550 struct packet_type
*ptype
;
1551 struct sk_buff
*skb2
= NULL
;
1552 struct packet_type
*pt_prev
= NULL
;
1555 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1556 /* Never send packets back to the socket
1557 * they originated from - MvS (miquels@drinkel.ow.org)
1559 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1560 (ptype
->af_packet_priv
== NULL
||
1561 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1563 deliver_skb(skb2
, pt_prev
, skb
->dev
);
1568 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1572 net_timestamp_set(skb2
);
1574 /* skb->nh should be correctly
1575 set by sender, so that the second statement is
1576 just protection against buggy protocols.
1578 skb_reset_mac_header(skb2
);
1580 if (skb_network_header(skb2
) < skb2
->data
||
1581 skb2
->network_header
> skb2
->tail
) {
1582 if (net_ratelimit())
1583 printk(KERN_CRIT
"protocol %04x is "
1585 ntohs(skb2
->protocol
),
1587 skb_reset_network_header(skb2
);
1590 skb2
->transport_header
= skb2
->network_header
;
1591 skb2
->pkt_type
= PACKET_OUTGOING
;
1596 pt_prev
->func(skb2
, skb
->dev
, pt_prev
, skb
->dev
);
1600 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1601 * @dev: Network device
1602 * @txq: number of queues available
1604 * If real_num_tx_queues is changed the tc mappings may no longer be
1605 * valid. To resolve this verify the tc mapping remains valid and if
1606 * not NULL the mapping. With no priorities mapping to this
1607 * offset/count pair it will no longer be used. In the worst case TC0
1608 * is invalid nothing can be done so disable priority mappings. If is
1609 * expected that drivers will fix this mapping if they can before
1610 * calling netif_set_real_num_tx_queues.
1612 static void netif_setup_tc(struct net_device
*dev
, unsigned int txq
)
1615 struct netdev_tc_txq
*tc
= &dev
->tc_to_txq
[0];
1617 /* If TC0 is invalidated disable TC mapping */
1618 if (tc
->offset
+ tc
->count
> txq
) {
1619 pr_warning("Number of in use tx queues changed "
1620 "invalidating tc mappings. Priority "
1621 "traffic classification disabled!\n");
1626 /* Invalidated prio to tc mappings set to TC0 */
1627 for (i
= 1; i
< TC_BITMASK
+ 1; i
++) {
1628 int q
= netdev_get_prio_tc_map(dev
, i
);
1630 tc
= &dev
->tc_to_txq
[q
];
1631 if (tc
->offset
+ tc
->count
> txq
) {
1632 pr_warning("Number of in use tx queues "
1633 "changed. Priority %i to tc "
1634 "mapping %i is no longer valid "
1635 "setting map to 0\n",
1637 netdev_set_prio_tc_map(dev
, i
, 0);
1643 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1644 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1646 int netif_set_real_num_tx_queues(struct net_device
*dev
, unsigned int txq
)
1650 if (txq
< 1 || txq
> dev
->num_tx_queues
)
1653 if (dev
->reg_state
== NETREG_REGISTERED
||
1654 dev
->reg_state
== NETREG_UNREGISTERING
) {
1657 rc
= netdev_queue_update_kobjects(dev
, dev
->real_num_tx_queues
,
1663 netif_setup_tc(dev
, txq
);
1665 if (txq
< dev
->real_num_tx_queues
)
1666 qdisc_reset_all_tx_gt(dev
, txq
);
1669 dev
->real_num_tx_queues
= txq
;
1672 EXPORT_SYMBOL(netif_set_real_num_tx_queues
);
1676 * netif_set_real_num_rx_queues - set actual number of RX queues used
1677 * @dev: Network device
1678 * @rxq: Actual number of RX queues
1680 * This must be called either with the rtnl_lock held or before
1681 * registration of the net device. Returns 0 on success, or a
1682 * negative error code. If called before registration, it always
1685 int netif_set_real_num_rx_queues(struct net_device
*dev
, unsigned int rxq
)
1689 if (rxq
< 1 || rxq
> dev
->num_rx_queues
)
1692 if (dev
->reg_state
== NETREG_REGISTERED
) {
1695 rc
= net_rx_queue_update_kobjects(dev
, dev
->real_num_rx_queues
,
1701 dev
->real_num_rx_queues
= rxq
;
1704 EXPORT_SYMBOL(netif_set_real_num_rx_queues
);
1707 static inline void __netif_reschedule(struct Qdisc
*q
)
1709 struct softnet_data
*sd
;
1710 unsigned long flags
;
1712 local_irq_save(flags
);
1713 sd
= &__get_cpu_var(softnet_data
);
1714 q
->next_sched
= NULL
;
1715 *sd
->output_queue_tailp
= q
;
1716 sd
->output_queue_tailp
= &q
->next_sched
;
1717 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1718 local_irq_restore(flags
);
1721 void __netif_schedule(struct Qdisc
*q
)
1723 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
1724 __netif_reschedule(q
);
1726 EXPORT_SYMBOL(__netif_schedule
);
1728 void dev_kfree_skb_irq(struct sk_buff
*skb
)
1730 if (atomic_dec_and_test(&skb
->users
)) {
1731 struct softnet_data
*sd
;
1732 unsigned long flags
;
1734 local_irq_save(flags
);
1735 sd
= &__get_cpu_var(softnet_data
);
1736 skb
->next
= sd
->completion_queue
;
1737 sd
->completion_queue
= skb
;
1738 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1739 local_irq_restore(flags
);
1742 EXPORT_SYMBOL(dev_kfree_skb_irq
);
1744 void dev_kfree_skb_any(struct sk_buff
*skb
)
1746 if (in_irq() || irqs_disabled())
1747 dev_kfree_skb_irq(skb
);
1751 EXPORT_SYMBOL(dev_kfree_skb_any
);
1755 * netif_device_detach - mark device as removed
1756 * @dev: network device
1758 * Mark device as removed from system and therefore no longer available.
1760 void netif_device_detach(struct net_device
*dev
)
1762 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1763 netif_running(dev
)) {
1764 netif_tx_stop_all_queues(dev
);
1767 EXPORT_SYMBOL(netif_device_detach
);
1770 * netif_device_attach - mark device as attached
1771 * @dev: network device
1773 * Mark device as attached from system and restart if needed.
1775 void netif_device_attach(struct net_device
*dev
)
1777 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1778 netif_running(dev
)) {
1779 netif_tx_wake_all_queues(dev
);
1780 __netdev_watchdog_up(dev
);
1783 EXPORT_SYMBOL(netif_device_attach
);
1786 * skb_dev_set -- assign a new device to a buffer
1787 * @skb: buffer for the new device
1788 * @dev: network device
1790 * If an skb is owned by a device already, we have to reset
1791 * all data private to the namespace a device belongs to
1792 * before assigning it a new device.
1794 #ifdef CONFIG_NET_NS
1795 void skb_set_dev(struct sk_buff
*skb
, struct net_device
*dev
)
1798 if (skb
->dev
&& !net_eq(dev_net(skb
->dev
), dev_net(dev
))) {
1801 skb_init_secmark(skb
);
1805 skb
->ipvs_property
= 0;
1806 #ifdef CONFIG_NET_SCHED
1812 EXPORT_SYMBOL(skb_set_dev
);
1813 #endif /* CONFIG_NET_NS */
1816 * Invalidate hardware checksum when packet is to be mangled, and
1817 * complete checksum manually on outgoing path.
1819 int skb_checksum_help(struct sk_buff
*skb
)
1822 int ret
= 0, offset
;
1824 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1825 goto out_set_summed
;
1827 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1828 /* Let GSO fix up the checksum. */
1829 goto out_set_summed
;
1832 offset
= skb_checksum_start_offset(skb
);
1833 BUG_ON(offset
>= skb_headlen(skb
));
1834 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1836 offset
+= skb
->csum_offset
;
1837 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
1839 if (skb_cloned(skb
) &&
1840 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
1841 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1846 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1848 skb
->ip_summed
= CHECKSUM_NONE
;
1852 EXPORT_SYMBOL(skb_checksum_help
);
1855 * skb_gso_segment - Perform segmentation on skb.
1856 * @skb: buffer to segment
1857 * @features: features for the output path (see dev->features)
1859 * This function segments the given skb and returns a list of segments.
1861 * It may return NULL if the skb requires no segmentation. This is
1862 * only possible when GSO is used for verifying header integrity.
1864 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
, u32 features
)
1866 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1867 struct packet_type
*ptype
;
1868 __be16 type
= skb
->protocol
;
1869 int vlan_depth
= ETH_HLEN
;
1872 while (type
== htons(ETH_P_8021Q
)) {
1873 struct vlan_hdr
*vh
;
1875 if (unlikely(!pskb_may_pull(skb
, vlan_depth
+ VLAN_HLEN
)))
1876 return ERR_PTR(-EINVAL
);
1878 vh
= (struct vlan_hdr
*)(skb
->data
+ vlan_depth
);
1879 type
= vh
->h_vlan_encapsulated_proto
;
1880 vlan_depth
+= VLAN_HLEN
;
1883 skb_reset_mac_header(skb
);
1884 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1885 __skb_pull(skb
, skb
->mac_len
);
1887 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1888 struct net_device
*dev
= skb
->dev
;
1889 struct ethtool_drvinfo info
= {};
1891 if (dev
&& dev
->ethtool_ops
&& dev
->ethtool_ops
->get_drvinfo
)
1892 dev
->ethtool_ops
->get_drvinfo(dev
, &info
);
1894 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1895 info
.driver
, dev
? dev
->features
: 0L,
1896 skb
->sk
? skb
->sk
->sk_route_caps
: 0L,
1897 skb
->len
, skb
->data_len
, skb
->ip_summed
);
1899 if (skb_header_cloned(skb
) &&
1900 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
1901 return ERR_PTR(err
);
1905 list_for_each_entry_rcu(ptype
,
1906 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
1907 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
1908 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1909 err
= ptype
->gso_send_check(skb
);
1910 segs
= ERR_PTR(err
);
1911 if (err
|| skb_gso_ok(skb
, features
))
1913 __skb_push(skb
, (skb
->data
-
1914 skb_network_header(skb
)));
1916 segs
= ptype
->gso_segment(skb
, features
);
1922 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1926 EXPORT_SYMBOL(skb_gso_segment
);
1928 /* Take action when hardware reception checksum errors are detected. */
1930 void netdev_rx_csum_fault(struct net_device
*dev
)
1932 if (net_ratelimit()) {
1933 printk(KERN_ERR
"%s: hw csum failure.\n",
1934 dev
? dev
->name
: "<unknown>");
1938 EXPORT_SYMBOL(netdev_rx_csum_fault
);
1941 /* Actually, we should eliminate this check as soon as we know, that:
1942 * 1. IOMMU is present and allows to map all the memory.
1943 * 2. No high memory really exists on this machine.
1946 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
1948 #ifdef CONFIG_HIGHMEM
1950 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
1951 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1952 if (PageHighMem(skb_shinfo(skb
)->frags
[i
].page
))
1956 if (PCI_DMA_BUS_IS_PHYS
) {
1957 struct device
*pdev
= dev
->dev
.parent
;
1961 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1962 dma_addr_t addr
= page_to_phys(skb_shinfo(skb
)->frags
[i
].page
);
1963 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
1972 void (*destructor
)(struct sk_buff
*skb
);
1975 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1977 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
1979 struct dev_gso_cb
*cb
;
1982 struct sk_buff
*nskb
= skb
->next
;
1984 skb
->next
= nskb
->next
;
1987 } while (skb
->next
);
1989 cb
= DEV_GSO_CB(skb
);
1991 cb
->destructor(skb
);
1995 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1996 * @skb: buffer to segment
1997 * @features: device features as applicable to this skb
1999 * This function segments the given skb and stores the list of segments
2002 static int dev_gso_segment(struct sk_buff
*skb
, int features
)
2004 struct sk_buff
*segs
;
2006 segs
= skb_gso_segment(skb
, features
);
2008 /* Verifying header integrity only. */
2013 return PTR_ERR(segs
);
2016 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
2017 skb
->destructor
= dev_gso_skb_destructor
;
2023 * Try to orphan skb early, right before transmission by the device.
2024 * We cannot orphan skb if tx timestamp is requested or the sk-reference
2025 * is needed on driver level for other reasons, e.g. see net/can/raw.c
2027 static inline void skb_orphan_try(struct sk_buff
*skb
)
2029 struct sock
*sk
= skb
->sk
;
2031 if (sk
&& !skb_shinfo(skb
)->tx_flags
) {
2032 /* skb_tx_hash() wont be able to get sk.
2033 * We copy sk_hash into skb->rxhash
2036 skb
->rxhash
= sk
->sk_hash
;
2041 static bool can_checksum_protocol(unsigned long features
, __be16 protocol
)
2043 return ((features
& NETIF_F_GEN_CSUM
) ||
2044 ((features
& NETIF_F_V4_CSUM
) &&
2045 protocol
== htons(ETH_P_IP
)) ||
2046 ((features
& NETIF_F_V6_CSUM
) &&
2047 protocol
== htons(ETH_P_IPV6
)) ||
2048 ((features
& NETIF_F_FCOE_CRC
) &&
2049 protocol
== htons(ETH_P_FCOE
)));
2052 static u32
harmonize_features(struct sk_buff
*skb
, __be16 protocol
, u32 features
)
2054 if (!can_checksum_protocol(features
, protocol
)) {
2055 features
&= ~NETIF_F_ALL_CSUM
;
2056 features
&= ~NETIF_F_SG
;
2057 } else if (illegal_highdma(skb
->dev
, skb
)) {
2058 features
&= ~NETIF_F_SG
;
2064 u32
netif_skb_features(struct sk_buff
*skb
)
2066 __be16 protocol
= skb
->protocol
;
2067 u32 features
= skb
->dev
->features
;
2069 if (protocol
== htons(ETH_P_8021Q
)) {
2070 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
2071 protocol
= veh
->h_vlan_encapsulated_proto
;
2072 } else if (!vlan_tx_tag_present(skb
)) {
2073 return harmonize_features(skb
, protocol
, features
);
2076 features
&= (skb
->dev
->vlan_features
| NETIF_F_HW_VLAN_TX
);
2078 if (protocol
!= htons(ETH_P_8021Q
)) {
2079 return harmonize_features(skb
, protocol
, features
);
2081 features
&= NETIF_F_SG
| NETIF_F_HIGHDMA
| NETIF_F_FRAGLIST
|
2082 NETIF_F_GEN_CSUM
| NETIF_F_HW_VLAN_TX
;
2083 return harmonize_features(skb
, protocol
, features
);
2086 EXPORT_SYMBOL(netif_skb_features
);
2089 * Returns true if either:
2090 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2091 * 2. skb is fragmented and the device does not support SG, or if
2092 * at least one of fragments is in highmem and device does not
2093 * support DMA from it.
2095 static inline int skb_needs_linearize(struct sk_buff
*skb
,
2098 return skb_is_nonlinear(skb
) &&
2099 ((skb_has_frag_list(skb
) &&
2100 !(features
& NETIF_F_FRAGLIST
)) ||
2101 (skb_shinfo(skb
)->nr_frags
&&
2102 !(features
& NETIF_F_SG
)));
2105 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
2106 struct netdev_queue
*txq
)
2108 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2109 int rc
= NETDEV_TX_OK
;
2110 unsigned int skb_len
;
2112 if (likely(!skb
->next
)) {
2116 * If device doesn't need skb->dst, release it right now while
2117 * its hot in this cpu cache
2119 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2122 if (!list_empty(&ptype_all
))
2123 dev_queue_xmit_nit(skb
, dev
);
2125 skb_orphan_try(skb
);
2127 features
= netif_skb_features(skb
);
2129 if (vlan_tx_tag_present(skb
) &&
2130 !(features
& NETIF_F_HW_VLAN_TX
)) {
2131 skb
= __vlan_put_tag(skb
, vlan_tx_tag_get(skb
));
2138 if (netif_needs_gso(skb
, features
)) {
2139 if (unlikely(dev_gso_segment(skb
, features
)))
2144 if (skb_needs_linearize(skb
, features
) &&
2145 __skb_linearize(skb
))
2148 /* If packet is not checksummed and device does not
2149 * support checksumming for this protocol, complete
2150 * checksumming here.
2152 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2153 skb_set_transport_header(skb
,
2154 skb_checksum_start_offset(skb
));
2155 if (!(features
& NETIF_F_ALL_CSUM
) &&
2156 skb_checksum_help(skb
))
2162 rc
= ops
->ndo_start_xmit(skb
, dev
);
2163 trace_net_dev_xmit(skb
, rc
, dev
, skb_len
);
2164 if (rc
== NETDEV_TX_OK
)
2165 txq_trans_update(txq
);
2171 struct sk_buff
*nskb
= skb
->next
;
2173 skb
->next
= nskb
->next
;
2177 * If device doesn't need nskb->dst, release it right now while
2178 * its hot in this cpu cache
2180 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2183 skb_len
= nskb
->len
;
2184 rc
= ops
->ndo_start_xmit(nskb
, dev
);
2185 trace_net_dev_xmit(nskb
, rc
, dev
, skb_len
);
2186 if (unlikely(rc
!= NETDEV_TX_OK
)) {
2187 if (rc
& ~NETDEV_TX_MASK
)
2188 goto out_kfree_gso_skb
;
2189 nskb
->next
= skb
->next
;
2193 txq_trans_update(txq
);
2194 if (unlikely(netif_tx_queue_stopped(txq
) && skb
->next
))
2195 return NETDEV_TX_BUSY
;
2196 } while (skb
->next
);
2199 if (likely(skb
->next
== NULL
))
2200 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
2207 static u32 hashrnd __read_mostly
;
2210 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2211 * to be used as a distribution range.
2213 u16
__skb_tx_hash(const struct net_device
*dev
, const struct sk_buff
*skb
,
2214 unsigned int num_tx_queues
)
2218 u16 qcount
= num_tx_queues
;
2220 if (skb_rx_queue_recorded(skb
)) {
2221 hash
= skb_get_rx_queue(skb
);
2222 while (unlikely(hash
>= num_tx_queues
))
2223 hash
-= num_tx_queues
;
2228 u8 tc
= netdev_get_prio_tc_map(dev
, skb
->priority
);
2229 qoffset
= dev
->tc_to_txq
[tc
].offset
;
2230 qcount
= dev
->tc_to_txq
[tc
].count
;
2233 if (skb
->sk
&& skb
->sk
->sk_hash
)
2234 hash
= skb
->sk
->sk_hash
;
2236 hash
= (__force u16
) skb
->protocol
^ skb
->rxhash
;
2237 hash
= jhash_1word(hash
, hashrnd
);
2239 return (u16
) (((u64
) hash
* qcount
) >> 32) + qoffset
;
2241 EXPORT_SYMBOL(__skb_tx_hash
);
2243 static inline u16
dev_cap_txqueue(struct net_device
*dev
, u16 queue_index
)
2245 if (unlikely(queue_index
>= dev
->real_num_tx_queues
)) {
2246 if (net_ratelimit()) {
2247 pr_warning("%s selects TX queue %d, but "
2248 "real number of TX queues is %d\n",
2249 dev
->name
, queue_index
, dev
->real_num_tx_queues
);
2256 static inline int get_xps_queue(struct net_device
*dev
, struct sk_buff
*skb
)
2259 struct xps_dev_maps
*dev_maps
;
2260 struct xps_map
*map
;
2261 int queue_index
= -1;
2264 dev_maps
= rcu_dereference(dev
->xps_maps
);
2266 map
= rcu_dereference(
2267 dev_maps
->cpu_map
[raw_smp_processor_id()]);
2270 queue_index
= map
->queues
[0];
2273 if (skb
->sk
&& skb
->sk
->sk_hash
)
2274 hash
= skb
->sk
->sk_hash
;
2276 hash
= (__force u16
) skb
->protocol
^
2278 hash
= jhash_1word(hash
, hashrnd
);
2279 queue_index
= map
->queues
[
2280 ((u64
)hash
* map
->len
) >> 32];
2282 if (unlikely(queue_index
>= dev
->real_num_tx_queues
))
2294 static struct netdev_queue
*dev_pick_tx(struct net_device
*dev
,
2295 struct sk_buff
*skb
)
2298 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2300 if (dev
->real_num_tx_queues
== 1)
2302 else if (ops
->ndo_select_queue
) {
2303 queue_index
= ops
->ndo_select_queue(dev
, skb
);
2304 queue_index
= dev_cap_txqueue(dev
, queue_index
);
2306 struct sock
*sk
= skb
->sk
;
2307 queue_index
= sk_tx_queue_get(sk
);
2309 if (queue_index
< 0 || skb
->ooo_okay
||
2310 queue_index
>= dev
->real_num_tx_queues
) {
2311 int old_index
= queue_index
;
2313 queue_index
= get_xps_queue(dev
, skb
);
2314 if (queue_index
< 0)
2315 queue_index
= skb_tx_hash(dev
, skb
);
2317 if (queue_index
!= old_index
&& sk
) {
2318 struct dst_entry
*dst
=
2319 rcu_dereference_check(sk
->sk_dst_cache
, 1);
2321 if (dst
&& skb_dst(skb
) == dst
)
2322 sk_tx_queue_set(sk
, queue_index
);
2327 skb_set_queue_mapping(skb
, queue_index
);
2328 return netdev_get_tx_queue(dev
, queue_index
);
2331 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2332 struct net_device
*dev
,
2333 struct netdev_queue
*txq
)
2335 spinlock_t
*root_lock
= qdisc_lock(q
);
2339 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
2340 qdisc_calculate_pkt_len(skb
, q
);
2342 * Heuristic to force contended enqueues to serialize on a
2343 * separate lock before trying to get qdisc main lock.
2344 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2345 * and dequeue packets faster.
2347 contended
= qdisc_is_running(q
);
2348 if (unlikely(contended
))
2349 spin_lock(&q
->busylock
);
2351 spin_lock(root_lock
);
2352 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2355 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2356 qdisc_run_begin(q
)) {
2358 * This is a work-conserving queue; there are no old skbs
2359 * waiting to be sent out; and the qdisc is not running -
2360 * xmit the skb directly.
2362 if (!(dev
->priv_flags
& IFF_XMIT_DST_RELEASE
))
2365 qdisc_bstats_update(q
, skb
);
2367 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
)) {
2368 if (unlikely(contended
)) {
2369 spin_unlock(&q
->busylock
);
2376 rc
= NET_XMIT_SUCCESS
;
2379 rc
= q
->enqueue(skb
, q
) & NET_XMIT_MASK
;
2380 if (qdisc_run_begin(q
)) {
2381 if (unlikely(contended
)) {
2382 spin_unlock(&q
->busylock
);
2388 spin_unlock(root_lock
);
2389 if (unlikely(contended
))
2390 spin_unlock(&q
->busylock
);
2394 static DEFINE_PER_CPU(int, xmit_recursion
);
2395 #define RECURSION_LIMIT 10
2398 * dev_queue_xmit - transmit a buffer
2399 * @skb: buffer to transmit
2401 * Queue a buffer for transmission to a network device. The caller must
2402 * have set the device and priority and built the buffer before calling
2403 * this function. The function can be called from an interrupt.
2405 * A negative errno code is returned on a failure. A success does not
2406 * guarantee the frame will be transmitted as it may be dropped due
2407 * to congestion or traffic shaping.
2409 * -----------------------------------------------------------------------------------
2410 * I notice this method can also return errors from the queue disciplines,
2411 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2414 * Regardless of the return value, the skb is consumed, so it is currently
2415 * difficult to retry a send to this method. (You can bump the ref count
2416 * before sending to hold a reference for retry if you are careful.)
2418 * When calling this method, interrupts MUST be enabled. This is because
2419 * the BH enable code must have IRQs enabled so that it will not deadlock.
2422 int dev_queue_xmit(struct sk_buff
*skb
)
2424 struct net_device
*dev
= skb
->dev
;
2425 struct netdev_queue
*txq
;
2429 /* Disable soft irqs for various locks below. Also
2430 * stops preemption for RCU.
2434 txq
= dev_pick_tx(dev
, skb
);
2435 q
= rcu_dereference_bh(txq
->qdisc
);
2437 #ifdef CONFIG_NET_CLS_ACT
2438 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2440 trace_net_dev_queue(skb
);
2442 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2446 /* The device has no queue. Common case for software devices:
2447 loopback, all the sorts of tunnels...
2449 Really, it is unlikely that netif_tx_lock protection is necessary
2450 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2452 However, it is possible, that they rely on protection
2455 Check this and shot the lock. It is not prone from deadlocks.
2456 Either shot noqueue qdisc, it is even simpler 8)
2458 if (dev
->flags
& IFF_UP
) {
2459 int cpu
= smp_processor_id(); /* ok because BHs are off */
2461 if (txq
->xmit_lock_owner
!= cpu
) {
2463 if (__this_cpu_read(xmit_recursion
) > RECURSION_LIMIT
)
2464 goto recursion_alert
;
2466 HARD_TX_LOCK(dev
, txq
, cpu
);
2468 if (!netif_tx_queue_stopped(txq
)) {
2469 __this_cpu_inc(xmit_recursion
);
2470 rc
= dev_hard_start_xmit(skb
, dev
, txq
);
2471 __this_cpu_dec(xmit_recursion
);
2472 if (dev_xmit_complete(rc
)) {
2473 HARD_TX_UNLOCK(dev
, txq
);
2477 HARD_TX_UNLOCK(dev
, txq
);
2478 if (net_ratelimit())
2479 printk(KERN_CRIT
"Virtual device %s asks to "
2480 "queue packet!\n", dev
->name
);
2482 /* Recursion is detected! It is possible,
2486 if (net_ratelimit())
2487 printk(KERN_CRIT
"Dead loop on virtual device "
2488 "%s, fix it urgently!\n", dev
->name
);
2493 rcu_read_unlock_bh();
2498 rcu_read_unlock_bh();
2501 EXPORT_SYMBOL(dev_queue_xmit
);
2504 /*=======================================================================
2506 =======================================================================*/
2508 int netdev_max_backlog __read_mostly
= 1000;
2509 int netdev_tstamp_prequeue __read_mostly
= 1;
2510 int netdev_budget __read_mostly
= 300;
2511 int weight_p __read_mostly
= 64; /* old backlog weight */
2513 /* Called with irq disabled */
2514 static inline void ____napi_schedule(struct softnet_data
*sd
,
2515 struct napi_struct
*napi
)
2517 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
2518 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2522 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2523 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2524 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2525 * if hash is a canonical 4-tuple hash over transport ports.
2527 void __skb_get_rxhash(struct sk_buff
*skb
)
2529 int nhoff
, hash
= 0, poff
;
2530 const struct ipv6hdr
*ip6
;
2531 const struct iphdr
*ip
;
2540 nhoff
= skb_network_offset(skb
);
2541 proto
= skb
->protocol
;
2545 case __constant_htons(ETH_P_IP
):
2546 if (!pskb_may_pull(skb
, sizeof(*ip
) + nhoff
))
2549 ip
= (const struct iphdr
*) (skb
->data
+ nhoff
);
2550 if (ip_is_fragment(ip
))
2553 ip_proto
= ip
->protocol
;
2554 addr1
= (__force u32
) ip
->saddr
;
2555 addr2
= (__force u32
) ip
->daddr
;
2556 nhoff
+= ip
->ihl
* 4;
2558 case __constant_htons(ETH_P_IPV6
):
2559 if (!pskb_may_pull(skb
, sizeof(*ip6
) + nhoff
))
2562 ip6
= (const struct ipv6hdr
*) (skb
->data
+ nhoff
);
2563 ip_proto
= ip6
->nexthdr
;
2564 addr1
= (__force u32
) ip6
->saddr
.s6_addr32
[3];
2565 addr2
= (__force u32
) ip6
->daddr
.s6_addr32
[3];
2574 if (pskb_may_pull(skb
, nhoff
+ 16)) {
2575 u8
*h
= skb
->data
+ nhoff
;
2576 __be16 flags
= *(__be16
*)h
;
2579 * Only look inside GRE if version zero and no
2582 if (!(flags
& (GRE_VERSION
|GRE_ROUTING
))) {
2583 proto
= *(__be16
*)(h
+ 2);
2585 if (flags
& GRE_CSUM
)
2587 if (flags
& GRE_KEY
)
2589 if (flags
& GRE_SEQ
)
2600 poff
= proto_ports_offset(ip_proto
);
2603 if (pskb_may_pull(skb
, nhoff
+ 4)) {
2604 ports
.v32
= * (__force u32
*) (skb
->data
+ nhoff
);
2605 if (ports
.v16
[1] < ports
.v16
[0])
2606 swap(ports
.v16
[0], ports
.v16
[1]);
2611 /* get a consistent hash (same value on both flow directions) */
2615 hash
= jhash_3words(addr1
, addr2
, ports
.v32
, hashrnd
);
2622 EXPORT_SYMBOL(__skb_get_rxhash
);
2626 /* One global table that all flow-based protocols share. */
2627 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
2628 EXPORT_SYMBOL(rps_sock_flow_table
);
2630 static struct rps_dev_flow
*
2631 set_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2632 struct rps_dev_flow
*rflow
, u16 next_cpu
)
2636 tcpu
= rflow
->cpu
= next_cpu
;
2637 if (tcpu
!= RPS_NO_CPU
) {
2638 #ifdef CONFIG_RFS_ACCEL
2639 struct netdev_rx_queue
*rxqueue
;
2640 struct rps_dev_flow_table
*flow_table
;
2641 struct rps_dev_flow
*old_rflow
;
2646 /* Should we steer this flow to a different hardware queue? */
2647 if (!skb_rx_queue_recorded(skb
) || !dev
->rx_cpu_rmap
||
2648 !(dev
->features
& NETIF_F_NTUPLE
))
2650 rxq_index
= cpu_rmap_lookup_index(dev
->rx_cpu_rmap
, next_cpu
);
2651 if (rxq_index
== skb_get_rx_queue(skb
))
2654 rxqueue
= dev
->_rx
+ rxq_index
;
2655 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2658 flow_id
= skb
->rxhash
& flow_table
->mask
;
2659 rc
= dev
->netdev_ops
->ndo_rx_flow_steer(dev
, skb
,
2660 rxq_index
, flow_id
);
2664 rflow
= &flow_table
->flows
[flow_id
];
2665 rflow
->cpu
= next_cpu
;
2667 if (old_rflow
->filter
== rflow
->filter
)
2668 old_rflow
->filter
= RPS_NO_FILTER
;
2672 per_cpu(softnet_data
, tcpu
).input_queue_head
;
2679 * get_rps_cpu is called from netif_receive_skb and returns the target
2680 * CPU from the RPS map of the receiving queue for a given skb.
2681 * rcu_read_lock must be held on entry.
2683 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2684 struct rps_dev_flow
**rflowp
)
2686 struct netdev_rx_queue
*rxqueue
;
2687 struct rps_map
*map
;
2688 struct rps_dev_flow_table
*flow_table
;
2689 struct rps_sock_flow_table
*sock_flow_table
;
2693 if (skb_rx_queue_recorded(skb
)) {
2694 u16 index
= skb_get_rx_queue(skb
);
2695 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
2696 WARN_ONCE(dev
->real_num_rx_queues
> 1,
2697 "%s received packet on queue %u, but number "
2698 "of RX queues is %u\n",
2699 dev
->name
, index
, dev
->real_num_rx_queues
);
2702 rxqueue
= dev
->_rx
+ index
;
2706 map
= rcu_dereference(rxqueue
->rps_map
);
2708 if (map
->len
== 1 &&
2709 !rcu_access_pointer(rxqueue
->rps_flow_table
)) {
2710 tcpu
= map
->cpus
[0];
2711 if (cpu_online(tcpu
))
2715 } else if (!rcu_access_pointer(rxqueue
->rps_flow_table
)) {
2719 skb_reset_network_header(skb
);
2720 if (!skb_get_rxhash(skb
))
2723 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2724 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
2725 if (flow_table
&& sock_flow_table
) {
2727 struct rps_dev_flow
*rflow
;
2729 rflow
= &flow_table
->flows
[skb
->rxhash
& flow_table
->mask
];
2732 next_cpu
= sock_flow_table
->ents
[skb
->rxhash
&
2733 sock_flow_table
->mask
];
2736 * If the desired CPU (where last recvmsg was done) is
2737 * different from current CPU (one in the rx-queue flow
2738 * table entry), switch if one of the following holds:
2739 * - Current CPU is unset (equal to RPS_NO_CPU).
2740 * - Current CPU is offline.
2741 * - The current CPU's queue tail has advanced beyond the
2742 * last packet that was enqueued using this table entry.
2743 * This guarantees that all previous packets for the flow
2744 * have been dequeued, thus preserving in order delivery.
2746 if (unlikely(tcpu
!= next_cpu
) &&
2747 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
2748 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
2749 rflow
->last_qtail
)) >= 0))
2750 rflow
= set_rps_cpu(dev
, skb
, rflow
, next_cpu
);
2752 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
2760 tcpu
= map
->cpus
[((u64
) skb
->rxhash
* map
->len
) >> 32];
2762 if (cpu_online(tcpu
)) {
2772 #ifdef CONFIG_RFS_ACCEL
2775 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2776 * @dev: Device on which the filter was set
2777 * @rxq_index: RX queue index
2778 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2779 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2781 * Drivers that implement ndo_rx_flow_steer() should periodically call
2782 * this function for each installed filter and remove the filters for
2783 * which it returns %true.
2785 bool rps_may_expire_flow(struct net_device
*dev
, u16 rxq_index
,
2786 u32 flow_id
, u16 filter_id
)
2788 struct netdev_rx_queue
*rxqueue
= dev
->_rx
+ rxq_index
;
2789 struct rps_dev_flow_table
*flow_table
;
2790 struct rps_dev_flow
*rflow
;
2795 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2796 if (flow_table
&& flow_id
<= flow_table
->mask
) {
2797 rflow
= &flow_table
->flows
[flow_id
];
2798 cpu
= ACCESS_ONCE(rflow
->cpu
);
2799 if (rflow
->filter
== filter_id
&& cpu
!= RPS_NO_CPU
&&
2800 ((int)(per_cpu(softnet_data
, cpu
).input_queue_head
-
2801 rflow
->last_qtail
) <
2802 (int)(10 * flow_table
->mask
)))
2808 EXPORT_SYMBOL(rps_may_expire_flow
);
2810 #endif /* CONFIG_RFS_ACCEL */
2812 /* Called from hardirq (IPI) context */
2813 static void rps_trigger_softirq(void *data
)
2815 struct softnet_data
*sd
= data
;
2817 ____napi_schedule(sd
, &sd
->backlog
);
2821 #endif /* CONFIG_RPS */
2824 * Check if this softnet_data structure is another cpu one
2825 * If yes, queue it to our IPI list and return 1
2828 static int rps_ipi_queued(struct softnet_data
*sd
)
2831 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
2834 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
2835 mysd
->rps_ipi_list
= sd
;
2837 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2840 #endif /* CONFIG_RPS */
2845 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2846 * queue (may be a remote CPU queue).
2848 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
2849 unsigned int *qtail
)
2851 struct softnet_data
*sd
;
2852 unsigned long flags
;
2854 sd
= &per_cpu(softnet_data
, cpu
);
2856 local_irq_save(flags
);
2859 if (skb_queue_len(&sd
->input_pkt_queue
) <= netdev_max_backlog
) {
2860 if (skb_queue_len(&sd
->input_pkt_queue
)) {
2862 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
2863 input_queue_tail_incr_save(sd
, qtail
);
2865 local_irq_restore(flags
);
2866 return NET_RX_SUCCESS
;
2869 /* Schedule NAPI for backlog device
2870 * We can use non atomic operation since we own the queue lock
2872 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
2873 if (!rps_ipi_queued(sd
))
2874 ____napi_schedule(sd
, &sd
->backlog
);
2882 local_irq_restore(flags
);
2884 atomic_long_inc(&skb
->dev
->rx_dropped
);
2890 * netif_rx - post buffer to the network code
2891 * @skb: buffer to post
2893 * This function receives a packet from a device driver and queues it for
2894 * the upper (protocol) levels to process. It always succeeds. The buffer
2895 * may be dropped during processing for congestion control or by the
2899 * NET_RX_SUCCESS (no congestion)
2900 * NET_RX_DROP (packet was dropped)
2904 int netif_rx(struct sk_buff
*skb
)
2908 /* if netpoll wants it, pretend we never saw it */
2909 if (netpoll_rx(skb
))
2912 if (netdev_tstamp_prequeue
)
2913 net_timestamp_check(skb
);
2915 trace_netif_rx(skb
);
2918 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
2924 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
2926 cpu
= smp_processor_id();
2928 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
2936 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
2942 EXPORT_SYMBOL(netif_rx
);
2944 int netif_rx_ni(struct sk_buff
*skb
)
2949 err
= netif_rx(skb
);
2950 if (local_softirq_pending())
2956 EXPORT_SYMBOL(netif_rx_ni
);
2958 static void net_tx_action(struct softirq_action
*h
)
2960 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
2962 if (sd
->completion_queue
) {
2963 struct sk_buff
*clist
;
2965 local_irq_disable();
2966 clist
= sd
->completion_queue
;
2967 sd
->completion_queue
= NULL
;
2971 struct sk_buff
*skb
= clist
;
2972 clist
= clist
->next
;
2974 WARN_ON(atomic_read(&skb
->users
));
2975 trace_kfree_skb(skb
, net_tx_action
);
2980 if (sd
->output_queue
) {
2983 local_irq_disable();
2984 head
= sd
->output_queue
;
2985 sd
->output_queue
= NULL
;
2986 sd
->output_queue_tailp
= &sd
->output_queue
;
2990 struct Qdisc
*q
= head
;
2991 spinlock_t
*root_lock
;
2993 head
= head
->next_sched
;
2995 root_lock
= qdisc_lock(q
);
2996 if (spin_trylock(root_lock
)) {
2997 smp_mb__before_clear_bit();
2998 clear_bit(__QDISC_STATE_SCHED
,
3001 spin_unlock(root_lock
);
3003 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
3005 __netif_reschedule(q
);
3007 smp_mb__before_clear_bit();
3008 clear_bit(__QDISC_STATE_SCHED
,
3016 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3017 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3018 /* This hook is defined here for ATM LANE */
3019 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
3020 unsigned char *addr
) __read_mostly
;
3021 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
3024 #ifdef CONFIG_NET_CLS_ACT
3025 /* TODO: Maybe we should just force sch_ingress to be compiled in
3026 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3027 * a compare and 2 stores extra right now if we dont have it on
3028 * but have CONFIG_NET_CLS_ACT
3029 * NOTE: This doesn't stop any functionality; if you dont have
3030 * the ingress scheduler, you just can't add policies on ingress.
3033 static int ing_filter(struct sk_buff
*skb
, struct netdev_queue
*rxq
)
3035 struct net_device
*dev
= skb
->dev
;
3036 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
3037 int result
= TC_ACT_OK
;
3040 if (unlikely(MAX_RED_LOOP
< ttl
++)) {
3041 if (net_ratelimit())
3042 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
3043 skb
->skb_iif
, dev
->ifindex
);
3047 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
3048 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
3051 if (q
!= &noop_qdisc
) {
3052 spin_lock(qdisc_lock(q
));
3053 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
3054 result
= qdisc_enqueue_root(skb
, q
);
3055 spin_unlock(qdisc_lock(q
));
3061 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
3062 struct packet_type
**pt_prev
,
3063 int *ret
, struct net_device
*orig_dev
)
3065 struct netdev_queue
*rxq
= rcu_dereference(skb
->dev
->ingress_queue
);
3067 if (!rxq
|| rxq
->qdisc
== &noop_qdisc
)
3071 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
3075 switch (ing_filter(skb
, rxq
)) {
3089 * netdev_rx_handler_register - register receive handler
3090 * @dev: device to register a handler for
3091 * @rx_handler: receive handler to register
3092 * @rx_handler_data: data pointer that is used by rx handler
3094 * Register a receive hander for a device. This handler will then be
3095 * called from __netif_receive_skb. A negative errno code is returned
3098 * The caller must hold the rtnl_mutex.
3100 * For a general description of rx_handler, see enum rx_handler_result.
3102 int netdev_rx_handler_register(struct net_device
*dev
,
3103 rx_handler_func_t
*rx_handler
,
3104 void *rx_handler_data
)
3108 if (dev
->rx_handler
)
3111 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
3112 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
3116 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
3119 * netdev_rx_handler_unregister - unregister receive handler
3120 * @dev: device to unregister a handler from
3122 * Unregister a receive hander from a device.
3124 * The caller must hold the rtnl_mutex.
3126 void netdev_rx_handler_unregister(struct net_device
*dev
)
3130 RCU_INIT_POINTER(dev
->rx_handler
, NULL
);
3131 RCU_INIT_POINTER(dev
->rx_handler_data
, NULL
);
3133 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
3135 static int __netif_receive_skb(struct sk_buff
*skb
)
3137 struct packet_type
*ptype
, *pt_prev
;
3138 rx_handler_func_t
*rx_handler
;
3139 struct net_device
*orig_dev
;
3140 struct net_device
*null_or_dev
;
3141 bool deliver_exact
= false;
3142 int ret
= NET_RX_DROP
;
3145 if (!netdev_tstamp_prequeue
)
3146 net_timestamp_check(skb
);
3148 trace_netif_receive_skb(skb
);
3150 /* if we've gotten here through NAPI, check netpoll */
3151 if (netpoll_receive_skb(skb
))
3155 skb
->skb_iif
= skb
->dev
->ifindex
;
3156 orig_dev
= skb
->dev
;
3158 skb_reset_network_header(skb
);
3159 skb_reset_transport_header(skb
);
3160 skb_reset_mac_len(skb
);
3168 __this_cpu_inc(softnet_data
.processed
);
3170 if (skb
->protocol
== cpu_to_be16(ETH_P_8021Q
)) {
3171 skb
= vlan_untag(skb
);
3176 #ifdef CONFIG_NET_CLS_ACT
3177 if (skb
->tc_verd
& TC_NCLS
) {
3178 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
3183 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
3184 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
3186 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3191 #ifdef CONFIG_NET_CLS_ACT
3192 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
3198 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
3201 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3204 switch (rx_handler(&skb
)) {
3205 case RX_HANDLER_CONSUMED
:
3207 case RX_HANDLER_ANOTHER
:
3209 case RX_HANDLER_EXACT
:
3210 deliver_exact
= true;
3211 case RX_HANDLER_PASS
:
3218 if (vlan_tx_tag_present(skb
)) {
3220 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3223 if (vlan_do_receive(&skb
)) {
3224 ret
= __netif_receive_skb(skb
);
3226 } else if (unlikely(!skb
))
3230 /* deliver only exact match when indicated */
3231 null_or_dev
= deliver_exact
? skb
->dev
: NULL
;
3233 type
= skb
->protocol
;
3234 list_for_each_entry_rcu(ptype
,
3235 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
3236 if (ptype
->type
== type
&&
3237 (ptype
->dev
== null_or_dev
|| ptype
->dev
== skb
->dev
||
3238 ptype
->dev
== orig_dev
)) {
3240 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3246 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
3248 atomic_long_inc(&skb
->dev
->rx_dropped
);
3250 /* Jamal, now you will not able to escape explaining
3251 * me how you were going to use this. :-)
3262 * netif_receive_skb - process receive buffer from network
3263 * @skb: buffer to process
3265 * netif_receive_skb() is the main receive data processing function.
3266 * It always succeeds. The buffer may be dropped during processing
3267 * for congestion control or by the protocol layers.
3269 * This function may only be called from softirq context and interrupts
3270 * should be enabled.
3272 * Return values (usually ignored):
3273 * NET_RX_SUCCESS: no congestion
3274 * NET_RX_DROP: packet was dropped
3276 int netif_receive_skb(struct sk_buff
*skb
)
3278 if (netdev_tstamp_prequeue
)
3279 net_timestamp_check(skb
);
3281 if (skb_defer_rx_timestamp(skb
))
3282 return NET_RX_SUCCESS
;
3286 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3291 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3294 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3298 ret
= __netif_receive_skb(skb
);
3304 return __netif_receive_skb(skb
);
3307 EXPORT_SYMBOL(netif_receive_skb
);
3309 /* Network device is going away, flush any packets still pending
3310 * Called with irqs disabled.
3312 static void flush_backlog(void *arg
)
3314 struct net_device
*dev
= arg
;
3315 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3316 struct sk_buff
*skb
, *tmp
;
3319 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3320 if (skb
->dev
== dev
) {
3321 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3323 input_queue_head_incr(sd
);
3328 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3329 if (skb
->dev
== dev
) {
3330 __skb_unlink(skb
, &sd
->process_queue
);
3332 input_queue_head_incr(sd
);
3337 static int napi_gro_complete(struct sk_buff
*skb
)
3339 struct packet_type
*ptype
;
3340 __be16 type
= skb
->protocol
;
3341 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3344 if (NAPI_GRO_CB(skb
)->count
== 1) {
3345 skb_shinfo(skb
)->gso_size
= 0;
3350 list_for_each_entry_rcu(ptype
, head
, list
) {
3351 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_complete
)
3354 err
= ptype
->gro_complete(skb
);
3360 WARN_ON(&ptype
->list
== head
);
3362 return NET_RX_SUCCESS
;
3366 return netif_receive_skb(skb
);
3369 inline void napi_gro_flush(struct napi_struct
*napi
)
3371 struct sk_buff
*skb
, *next
;
3373 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3376 napi_gro_complete(skb
);
3379 napi
->gro_count
= 0;
3380 napi
->gro_list
= NULL
;
3382 EXPORT_SYMBOL(napi_gro_flush
);
3384 enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3386 struct sk_buff
**pp
= NULL
;
3387 struct packet_type
*ptype
;
3388 __be16 type
= skb
->protocol
;
3389 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3392 enum gro_result ret
;
3394 if (!(skb
->dev
->features
& NETIF_F_GRO
) || netpoll_rx_on(skb
))
3397 if (skb_is_gso(skb
) || skb_has_frag_list(skb
))
3401 list_for_each_entry_rcu(ptype
, head
, list
) {
3402 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_receive
)
3405 skb_set_network_header(skb
, skb_gro_offset(skb
));
3406 mac_len
= skb
->network_header
- skb
->mac_header
;
3407 skb
->mac_len
= mac_len
;
3408 NAPI_GRO_CB(skb
)->same_flow
= 0;
3409 NAPI_GRO_CB(skb
)->flush
= 0;
3410 NAPI_GRO_CB(skb
)->free
= 0;
3412 pp
= ptype
->gro_receive(&napi
->gro_list
, skb
);
3417 if (&ptype
->list
== head
)
3420 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3421 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3424 struct sk_buff
*nskb
= *pp
;
3428 napi_gro_complete(nskb
);
3435 if (NAPI_GRO_CB(skb
)->flush
|| napi
->gro_count
>= MAX_GRO_SKBS
)
3439 NAPI_GRO_CB(skb
)->count
= 1;
3440 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
3441 skb
->next
= napi
->gro_list
;
3442 napi
->gro_list
= skb
;
3446 if (skb_headlen(skb
) < skb_gro_offset(skb
)) {
3447 int grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
3449 BUG_ON(skb
->end
- skb
->tail
< grow
);
3451 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3454 skb
->data_len
-= grow
;
3456 skb_shinfo(skb
)->frags
[0].page_offset
+= grow
;
3457 skb_shinfo(skb
)->frags
[0].size
-= grow
;
3459 if (unlikely(!skb_shinfo(skb
)->frags
[0].size
)) {
3460 put_page(skb_shinfo(skb
)->frags
[0].page
);
3461 memmove(skb_shinfo(skb
)->frags
,
3462 skb_shinfo(skb
)->frags
+ 1,
3463 --skb_shinfo(skb
)->nr_frags
* sizeof(skb_frag_t
));
3474 EXPORT_SYMBOL(dev_gro_receive
);
3476 static inline gro_result_t
3477 __napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3481 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3482 unsigned long diffs
;
3484 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
3485 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
3486 diffs
|= compare_ether_header(skb_mac_header(p
),
3487 skb_gro_mac_header(skb
));
3488 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
3489 NAPI_GRO_CB(p
)->flush
= 0;
3492 return dev_gro_receive(napi
, skb
);
3495 gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
3499 if (netif_receive_skb(skb
))
3504 case GRO_MERGED_FREE
:
3515 EXPORT_SYMBOL(napi_skb_finish
);
3517 void skb_gro_reset_offset(struct sk_buff
*skb
)
3519 NAPI_GRO_CB(skb
)->data_offset
= 0;
3520 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3521 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3523 if (skb
->mac_header
== skb
->tail
&&
3524 !PageHighMem(skb_shinfo(skb
)->frags
[0].page
)) {
3525 NAPI_GRO_CB(skb
)->frag0
=
3526 page_address(skb_shinfo(skb
)->frags
[0].page
) +
3527 skb_shinfo(skb
)->frags
[0].page_offset
;
3528 NAPI_GRO_CB(skb
)->frag0_len
= skb_shinfo(skb
)->frags
[0].size
;
3531 EXPORT_SYMBOL(skb_gro_reset_offset
);
3533 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3535 skb_gro_reset_offset(skb
);
3537 return napi_skb_finish(__napi_gro_receive(napi
, skb
), skb
);
3539 EXPORT_SYMBOL(napi_gro_receive
);
3541 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
3543 __skb_pull(skb
, skb_headlen(skb
));
3544 skb_reserve(skb
, NET_IP_ALIGN
- skb_headroom(skb
));
3546 skb
->dev
= napi
->dev
;
3552 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
3554 struct sk_buff
*skb
= napi
->skb
;
3557 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
3563 EXPORT_SYMBOL(napi_get_frags
);
3565 gro_result_t
napi_frags_finish(struct napi_struct
*napi
, struct sk_buff
*skb
,
3571 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
3573 if (ret
== GRO_HELD
)
3574 skb_gro_pull(skb
, -ETH_HLEN
);
3575 else if (netif_receive_skb(skb
))
3580 case GRO_MERGED_FREE
:
3581 napi_reuse_skb(napi
, skb
);
3590 EXPORT_SYMBOL(napi_frags_finish
);
3592 struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
3594 struct sk_buff
*skb
= napi
->skb
;
3601 skb_reset_mac_header(skb
);
3602 skb_gro_reset_offset(skb
);
3604 off
= skb_gro_offset(skb
);
3605 hlen
= off
+ sizeof(*eth
);
3606 eth
= skb_gro_header_fast(skb
, off
);
3607 if (skb_gro_header_hard(skb
, hlen
)) {
3608 eth
= skb_gro_header_slow(skb
, hlen
, off
);
3609 if (unlikely(!eth
)) {
3610 napi_reuse_skb(napi
, skb
);
3616 skb_gro_pull(skb
, sizeof(*eth
));
3619 * This works because the only protocols we care about don't require
3620 * special handling. We'll fix it up properly at the end.
3622 skb
->protocol
= eth
->h_proto
;
3627 EXPORT_SYMBOL(napi_frags_skb
);
3629 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
3631 struct sk_buff
*skb
= napi_frags_skb(napi
);
3636 return napi_frags_finish(napi
, skb
, __napi_gro_receive(napi
, skb
));
3638 EXPORT_SYMBOL(napi_gro_frags
);
3641 * net_rps_action sends any pending IPI's for rps.
3642 * Note: called with local irq disabled, but exits with local irq enabled.
3644 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
3647 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
3650 sd
->rps_ipi_list
= NULL
;
3654 /* Send pending IPI's to kick RPS processing on remote cpus. */
3656 struct softnet_data
*next
= remsd
->rps_ipi_next
;
3658 if (cpu_online(remsd
->cpu
))
3659 __smp_call_function_single(remsd
->cpu
,
3668 static int process_backlog(struct napi_struct
*napi
, int quota
)
3671 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
3674 /* Check if we have pending ipi, its better to send them now,
3675 * not waiting net_rx_action() end.
3677 if (sd
->rps_ipi_list
) {
3678 local_irq_disable();
3679 net_rps_action_and_irq_enable(sd
);
3682 napi
->weight
= weight_p
;
3683 local_irq_disable();
3684 while (work
< quota
) {
3685 struct sk_buff
*skb
;
3688 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
3690 __netif_receive_skb(skb
);
3691 local_irq_disable();
3692 input_queue_head_incr(sd
);
3693 if (++work
>= quota
) {
3700 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3702 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
3703 &sd
->process_queue
);
3705 if (qlen
< quota
- work
) {
3707 * Inline a custom version of __napi_complete().
3708 * only current cpu owns and manipulates this napi,
3709 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3710 * we can use a plain write instead of clear_bit(),
3711 * and we dont need an smp_mb() memory barrier.
3713 list_del(&napi
->poll_list
);
3716 quota
= work
+ qlen
;
3726 * __napi_schedule - schedule for receive
3727 * @n: entry to schedule
3729 * The entry's receive function will be scheduled to run
3731 void __napi_schedule(struct napi_struct
*n
)
3733 unsigned long flags
;
3735 local_irq_save(flags
);
3736 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
3737 local_irq_restore(flags
);
3739 EXPORT_SYMBOL(__napi_schedule
);
3741 void __napi_complete(struct napi_struct
*n
)
3743 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
3744 BUG_ON(n
->gro_list
);
3746 list_del(&n
->poll_list
);
3747 smp_mb__before_clear_bit();
3748 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
3750 EXPORT_SYMBOL(__napi_complete
);
3752 void napi_complete(struct napi_struct
*n
)
3754 unsigned long flags
;
3757 * don't let napi dequeue from the cpu poll list
3758 * just in case its running on a different cpu
3760 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
3764 local_irq_save(flags
);
3766 local_irq_restore(flags
);
3768 EXPORT_SYMBOL(napi_complete
);
3770 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
3771 int (*poll
)(struct napi_struct
*, int), int weight
)
3773 INIT_LIST_HEAD(&napi
->poll_list
);
3774 napi
->gro_count
= 0;
3775 napi
->gro_list
= NULL
;
3778 napi
->weight
= weight
;
3779 list_add(&napi
->dev_list
, &dev
->napi_list
);
3781 #ifdef CONFIG_NETPOLL
3782 spin_lock_init(&napi
->poll_lock
);
3783 napi
->poll_owner
= -1;
3785 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
3787 EXPORT_SYMBOL(netif_napi_add
);
3789 void netif_napi_del(struct napi_struct
*napi
)
3791 struct sk_buff
*skb
, *next
;
3793 list_del_init(&napi
->dev_list
);
3794 napi_free_frags(napi
);
3796 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3802 napi
->gro_list
= NULL
;
3803 napi
->gro_count
= 0;
3805 EXPORT_SYMBOL(netif_napi_del
);
3807 static void net_rx_action(struct softirq_action
*h
)
3809 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3810 unsigned long time_limit
= jiffies
+ 2;
3811 int budget
= netdev_budget
;
3814 local_irq_disable();
3816 while (!list_empty(&sd
->poll_list
)) {
3817 struct napi_struct
*n
;
3820 /* If softirq window is exhuasted then punt.
3821 * Allow this to run for 2 jiffies since which will allow
3822 * an average latency of 1.5/HZ.
3824 if (unlikely(budget
<= 0 || time_after(jiffies
, time_limit
)))
3829 /* Even though interrupts have been re-enabled, this
3830 * access is safe because interrupts can only add new
3831 * entries to the tail of this list, and only ->poll()
3832 * calls can remove this head entry from the list.
3834 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
3836 have
= netpoll_poll_lock(n
);
3840 /* This NAPI_STATE_SCHED test is for avoiding a race
3841 * with netpoll's poll_napi(). Only the entity which
3842 * obtains the lock and sees NAPI_STATE_SCHED set will
3843 * actually make the ->poll() call. Therefore we avoid
3844 * accidentally calling ->poll() when NAPI is not scheduled.
3847 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
3848 work
= n
->poll(n
, weight
);
3852 WARN_ON_ONCE(work
> weight
);
3856 local_irq_disable();
3858 /* Drivers must not modify the NAPI state if they
3859 * consume the entire weight. In such cases this code
3860 * still "owns" the NAPI instance and therefore can
3861 * move the instance around on the list at-will.
3863 if (unlikely(work
== weight
)) {
3864 if (unlikely(napi_disable_pending(n
))) {
3867 local_irq_disable();
3869 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
3872 netpoll_poll_unlock(have
);
3875 net_rps_action_and_irq_enable(sd
);
3877 #ifdef CONFIG_NET_DMA
3879 * There may not be any more sk_buffs coming right now, so push
3880 * any pending DMA copies to hardware
3882 dma_issue_pending_all();
3889 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3893 static gifconf_func_t
*gifconf_list
[NPROTO
];
3896 * register_gifconf - register a SIOCGIF handler
3897 * @family: Address family
3898 * @gifconf: Function handler
3900 * Register protocol dependent address dumping routines. The handler
3901 * that is passed must not be freed or reused until it has been replaced
3902 * by another handler.
3904 int register_gifconf(unsigned int family
, gifconf_func_t
*gifconf
)
3906 if (family
>= NPROTO
)
3908 gifconf_list
[family
] = gifconf
;
3911 EXPORT_SYMBOL(register_gifconf
);
3915 * Map an interface index to its name (SIOCGIFNAME)
3919 * We need this ioctl for efficient implementation of the
3920 * if_indextoname() function required by the IPv6 API. Without
3921 * it, we would have to search all the interfaces to find a
3925 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
3927 struct net_device
*dev
;
3931 * Fetch the caller's info block.
3934 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
3938 dev
= dev_get_by_index_rcu(net
, ifr
.ifr_ifindex
);
3944 strcpy(ifr
.ifr_name
, dev
->name
);
3947 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
3953 * Perform a SIOCGIFCONF call. This structure will change
3954 * size eventually, and there is nothing I can do about it.
3955 * Thus we will need a 'compatibility mode'.
3958 static int dev_ifconf(struct net
*net
, char __user
*arg
)
3961 struct net_device
*dev
;
3968 * Fetch the caller's info block.
3971 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
3978 * Loop over the interfaces, and write an info block for each.
3982 for_each_netdev(net
, dev
) {
3983 for (i
= 0; i
< NPROTO
; i
++) {
3984 if (gifconf_list
[i
]) {
3987 done
= gifconf_list
[i
](dev
, NULL
, 0);
3989 done
= gifconf_list
[i
](dev
, pos
+ total
,
3999 * All done. Write the updated control block back to the caller.
4001 ifc
.ifc_len
= total
;
4004 * Both BSD and Solaris return 0 here, so we do too.
4006 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
4009 #ifdef CONFIG_PROC_FS
4011 * This is invoked by the /proc filesystem handler to display a device
4014 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4017 struct net
*net
= seq_file_net(seq
);
4019 struct net_device
*dev
;
4023 return SEQ_START_TOKEN
;
4026 for_each_netdev_rcu(net
, dev
)
4033 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4035 struct net_device
*dev
= v
;
4037 if (v
== SEQ_START_TOKEN
)
4038 dev
= first_net_device_rcu(seq_file_net(seq
));
4040 dev
= next_net_device_rcu(dev
);
4046 void dev_seq_stop(struct seq_file
*seq
, void *v
)
4052 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
4054 struct rtnl_link_stats64 temp
;
4055 const struct rtnl_link_stats64
*stats
= dev_get_stats(dev
, &temp
);
4057 seq_printf(seq
, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4058 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4059 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
4061 stats
->rx_dropped
+ stats
->rx_missed_errors
,
4062 stats
->rx_fifo_errors
,
4063 stats
->rx_length_errors
+ stats
->rx_over_errors
+
4064 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
4065 stats
->rx_compressed
, stats
->multicast
,
4066 stats
->tx_bytes
, stats
->tx_packets
,
4067 stats
->tx_errors
, stats
->tx_dropped
,
4068 stats
->tx_fifo_errors
, stats
->collisions
,
4069 stats
->tx_carrier_errors
+
4070 stats
->tx_aborted_errors
+
4071 stats
->tx_window_errors
+
4072 stats
->tx_heartbeat_errors
,
4073 stats
->tx_compressed
);
4077 * Called from the PROCfs module. This now uses the new arbitrary sized
4078 * /proc/net interface to create /proc/net/dev
4080 static int dev_seq_show(struct seq_file
*seq
, void *v
)
4082 if (v
== SEQ_START_TOKEN
)
4083 seq_puts(seq
, "Inter-| Receive "
4085 " face |bytes packets errs drop fifo frame "
4086 "compressed multicast|bytes packets errs "
4087 "drop fifo colls carrier compressed\n");
4089 dev_seq_printf_stats(seq
, v
);
4093 static struct softnet_data
*softnet_get_online(loff_t
*pos
)
4095 struct softnet_data
*sd
= NULL
;
4097 while (*pos
< nr_cpu_ids
)
4098 if (cpu_online(*pos
)) {
4099 sd
= &per_cpu(softnet_data
, *pos
);
4106 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4108 return softnet_get_online(pos
);
4111 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4114 return softnet_get_online(pos
);
4117 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
4121 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
4123 struct softnet_data
*sd
= v
;
4125 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4126 sd
->processed
, sd
->dropped
, sd
->time_squeeze
, 0,
4127 0, 0, 0, 0, /* was fastroute */
4128 sd
->cpu_collision
, sd
->received_rps
);
4132 static const struct seq_operations dev_seq_ops
= {
4133 .start
= dev_seq_start
,
4134 .next
= dev_seq_next
,
4135 .stop
= dev_seq_stop
,
4136 .show
= dev_seq_show
,
4139 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
4141 return seq_open_net(inode
, file
, &dev_seq_ops
,
4142 sizeof(struct seq_net_private
));
4145 static const struct file_operations dev_seq_fops
= {
4146 .owner
= THIS_MODULE
,
4147 .open
= dev_seq_open
,
4149 .llseek
= seq_lseek
,
4150 .release
= seq_release_net
,
4153 static const struct seq_operations softnet_seq_ops
= {
4154 .start
= softnet_seq_start
,
4155 .next
= softnet_seq_next
,
4156 .stop
= softnet_seq_stop
,
4157 .show
= softnet_seq_show
,
4160 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
4162 return seq_open(file
, &softnet_seq_ops
);
4165 static const struct file_operations softnet_seq_fops
= {
4166 .owner
= THIS_MODULE
,
4167 .open
= softnet_seq_open
,
4169 .llseek
= seq_lseek
,
4170 .release
= seq_release
,
4173 static void *ptype_get_idx(loff_t pos
)
4175 struct packet_type
*pt
= NULL
;
4179 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
4185 for (t
= 0; t
< PTYPE_HASH_SIZE
; t
++) {
4186 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
4195 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4199 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
4202 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4204 struct packet_type
*pt
;
4205 struct list_head
*nxt
;
4209 if (v
== SEQ_START_TOKEN
)
4210 return ptype_get_idx(0);
4213 nxt
= pt
->list
.next
;
4214 if (pt
->type
== htons(ETH_P_ALL
)) {
4215 if (nxt
!= &ptype_all
)
4218 nxt
= ptype_base
[0].next
;
4220 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
4222 while (nxt
== &ptype_base
[hash
]) {
4223 if (++hash
>= PTYPE_HASH_SIZE
)
4225 nxt
= ptype_base
[hash
].next
;
4228 return list_entry(nxt
, struct packet_type
, list
);
4231 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
4237 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
4239 struct packet_type
*pt
= v
;
4241 if (v
== SEQ_START_TOKEN
)
4242 seq_puts(seq
, "Type Device Function\n");
4243 else if (pt
->dev
== NULL
|| dev_net(pt
->dev
) == seq_file_net(seq
)) {
4244 if (pt
->type
== htons(ETH_P_ALL
))
4245 seq_puts(seq
, "ALL ");
4247 seq_printf(seq
, "%04x", ntohs(pt
->type
));
4249 seq_printf(seq
, " %-8s %pF\n",
4250 pt
->dev
? pt
->dev
->name
: "", pt
->func
);
4256 static const struct seq_operations ptype_seq_ops
= {
4257 .start
= ptype_seq_start
,
4258 .next
= ptype_seq_next
,
4259 .stop
= ptype_seq_stop
,
4260 .show
= ptype_seq_show
,
4263 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
4265 return seq_open_net(inode
, file
, &ptype_seq_ops
,
4266 sizeof(struct seq_net_private
));
4269 static const struct file_operations ptype_seq_fops
= {
4270 .owner
= THIS_MODULE
,
4271 .open
= ptype_seq_open
,
4273 .llseek
= seq_lseek
,
4274 .release
= seq_release_net
,
4278 static int __net_init
dev_proc_net_init(struct net
*net
)
4282 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
4284 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
4286 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
4289 if (wext_proc_init(net
))
4295 proc_net_remove(net
, "ptype");
4297 proc_net_remove(net
, "softnet_stat");
4299 proc_net_remove(net
, "dev");
4303 static void __net_exit
dev_proc_net_exit(struct net
*net
)
4305 wext_proc_exit(net
);
4307 proc_net_remove(net
, "ptype");
4308 proc_net_remove(net
, "softnet_stat");
4309 proc_net_remove(net
, "dev");
4312 static struct pernet_operations __net_initdata dev_proc_ops
= {
4313 .init
= dev_proc_net_init
,
4314 .exit
= dev_proc_net_exit
,
4317 static int __init
dev_proc_init(void)
4319 return register_pernet_subsys(&dev_proc_ops
);
4322 #define dev_proc_init() 0
4323 #endif /* CONFIG_PROC_FS */
4327 * netdev_set_master - set up master pointer
4328 * @slave: slave device
4329 * @master: new master device
4331 * Changes the master device of the slave. Pass %NULL to break the
4332 * bonding. The caller must hold the RTNL semaphore. On a failure
4333 * a negative errno code is returned. On success the reference counts
4334 * are adjusted and the function returns zero.
4336 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
4338 struct net_device
*old
= slave
->master
;
4348 slave
->master
= master
;
4354 EXPORT_SYMBOL(netdev_set_master
);
4357 * netdev_set_bond_master - set up bonding master/slave pair
4358 * @slave: slave device
4359 * @master: new master device
4361 * Changes the master device of the slave. Pass %NULL to break the
4362 * bonding. The caller must hold the RTNL semaphore. On a failure
4363 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4364 * to the routing socket and the function returns zero.
4366 int netdev_set_bond_master(struct net_device
*slave
, struct net_device
*master
)
4372 err
= netdev_set_master(slave
, master
);
4376 slave
->flags
|= IFF_SLAVE
;
4378 slave
->flags
&= ~IFF_SLAVE
;
4380 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
4383 EXPORT_SYMBOL(netdev_set_bond_master
);
4385 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
4387 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4389 if ((dev
->flags
& IFF_UP
) && ops
->ndo_change_rx_flags
)
4390 ops
->ndo_change_rx_flags(dev
, flags
);
4393 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
4395 unsigned short old_flags
= dev
->flags
;
4401 dev
->flags
|= IFF_PROMISC
;
4402 dev
->promiscuity
+= inc
;
4403 if (dev
->promiscuity
== 0) {
4406 * If inc causes overflow, untouch promisc and return error.
4409 dev
->flags
&= ~IFF_PROMISC
;
4411 dev
->promiscuity
-= inc
;
4412 printk(KERN_WARNING
"%s: promiscuity touches roof, "
4413 "set promiscuity failed, promiscuity feature "
4414 "of device might be broken.\n", dev
->name
);
4418 if (dev
->flags
!= old_flags
) {
4419 printk(KERN_INFO
"device %s %s promiscuous mode\n",
4420 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
4422 if (audit_enabled
) {
4423 current_uid_gid(&uid
, &gid
);
4424 audit_log(current
->audit_context
, GFP_ATOMIC
,
4425 AUDIT_ANOM_PROMISCUOUS
,
4426 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4427 dev
->name
, (dev
->flags
& IFF_PROMISC
),
4428 (old_flags
& IFF_PROMISC
),
4429 audit_get_loginuid(current
),
4431 audit_get_sessionid(current
));
4434 dev_change_rx_flags(dev
, IFF_PROMISC
);
4440 * dev_set_promiscuity - update promiscuity count on a device
4444 * Add or remove promiscuity from a device. While the count in the device
4445 * remains above zero the interface remains promiscuous. Once it hits zero
4446 * the device reverts back to normal filtering operation. A negative inc
4447 * value is used to drop promiscuity on the device.
4448 * Return 0 if successful or a negative errno code on error.
4450 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
4452 unsigned short old_flags
= dev
->flags
;
4455 err
= __dev_set_promiscuity(dev
, inc
);
4458 if (dev
->flags
!= old_flags
)
4459 dev_set_rx_mode(dev
);
4462 EXPORT_SYMBOL(dev_set_promiscuity
);
4465 * dev_set_allmulti - update allmulti count on a device
4469 * Add or remove reception of all multicast frames to a device. While the
4470 * count in the device remains above zero the interface remains listening
4471 * to all interfaces. Once it hits zero the device reverts back to normal
4472 * filtering operation. A negative @inc value is used to drop the counter
4473 * when releasing a resource needing all multicasts.
4474 * Return 0 if successful or a negative errno code on error.
4477 int dev_set_allmulti(struct net_device
*dev
, int inc
)
4479 unsigned short old_flags
= dev
->flags
;
4483 dev
->flags
|= IFF_ALLMULTI
;
4484 dev
->allmulti
+= inc
;
4485 if (dev
->allmulti
== 0) {
4488 * If inc causes overflow, untouch allmulti and return error.
4491 dev
->flags
&= ~IFF_ALLMULTI
;
4493 dev
->allmulti
-= inc
;
4494 printk(KERN_WARNING
"%s: allmulti touches roof, "
4495 "set allmulti failed, allmulti feature of "
4496 "device might be broken.\n", dev
->name
);
4500 if (dev
->flags
^ old_flags
) {
4501 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
4502 dev_set_rx_mode(dev
);
4506 EXPORT_SYMBOL(dev_set_allmulti
);
4509 * Upload unicast and multicast address lists to device and
4510 * configure RX filtering. When the device doesn't support unicast
4511 * filtering it is put in promiscuous mode while unicast addresses
4514 void __dev_set_rx_mode(struct net_device
*dev
)
4516 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4518 /* dev_open will call this function so the list will stay sane. */
4519 if (!(dev
->flags
&IFF_UP
))
4522 if (!netif_device_present(dev
))
4525 if (!(dev
->priv_flags
& IFF_UNICAST_FLT
)) {
4526 /* Unicast addresses changes may only happen under the rtnl,
4527 * therefore calling __dev_set_promiscuity here is safe.
4529 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
4530 __dev_set_promiscuity(dev
, 1);
4531 dev
->uc_promisc
= true;
4532 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
4533 __dev_set_promiscuity(dev
, -1);
4534 dev
->uc_promisc
= false;
4538 if (ops
->ndo_set_rx_mode
)
4539 ops
->ndo_set_rx_mode(dev
);
4540 else if (ops
->ndo_set_multicast_list
)
4541 ops
->ndo_set_multicast_list(dev
);
4544 void dev_set_rx_mode(struct net_device
*dev
)
4546 netif_addr_lock_bh(dev
);
4547 __dev_set_rx_mode(dev
);
4548 netif_addr_unlock_bh(dev
);
4552 * dev_ethtool_get_settings - call device's ethtool_ops::get_settings()
4554 * @cmd: memory area for ethtool_ops::get_settings() result
4556 * The cmd arg is initialized properly (cleared and
4557 * ethtool_cmd::cmd field set to ETHTOOL_GSET).
4559 * Return device's ethtool_ops::get_settings() result value or
4560 * -EOPNOTSUPP when device doesn't expose
4561 * ethtool_ops::get_settings() operation.
4563 int dev_ethtool_get_settings(struct net_device
*dev
,
4564 struct ethtool_cmd
*cmd
)
4566 if (!dev
->ethtool_ops
|| !dev
->ethtool_ops
->get_settings
)
4569 memset(cmd
, 0, sizeof(struct ethtool_cmd
));
4570 cmd
->cmd
= ETHTOOL_GSET
;
4571 return dev
->ethtool_ops
->get_settings(dev
, cmd
);
4573 EXPORT_SYMBOL(dev_ethtool_get_settings
);
4576 * dev_get_flags - get flags reported to userspace
4579 * Get the combination of flag bits exported through APIs to userspace.
4581 unsigned dev_get_flags(const struct net_device
*dev
)
4585 flags
= (dev
->flags
& ~(IFF_PROMISC
|
4590 (dev
->gflags
& (IFF_PROMISC
|
4593 if (netif_running(dev
)) {
4594 if (netif_oper_up(dev
))
4595 flags
|= IFF_RUNNING
;
4596 if (netif_carrier_ok(dev
))
4597 flags
|= IFF_LOWER_UP
;
4598 if (netif_dormant(dev
))
4599 flags
|= IFF_DORMANT
;
4604 EXPORT_SYMBOL(dev_get_flags
);
4606 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4608 int old_flags
= dev
->flags
;
4614 * Set the flags on our device.
4617 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
4618 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
4620 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
4624 * Load in the correct multicast list now the flags have changed.
4627 if ((old_flags
^ flags
) & IFF_MULTICAST
)
4628 dev_change_rx_flags(dev
, IFF_MULTICAST
);
4630 dev_set_rx_mode(dev
);
4633 * Have we downed the interface. We handle IFF_UP ourselves
4634 * according to user attempts to set it, rather than blindly
4639 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
4640 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
4643 dev_set_rx_mode(dev
);
4646 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
4647 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
4649 dev
->gflags
^= IFF_PROMISC
;
4650 dev_set_promiscuity(dev
, inc
);
4653 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4654 is important. Some (broken) drivers set IFF_PROMISC, when
4655 IFF_ALLMULTI is requested not asking us and not reporting.
4657 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
4658 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
4660 dev
->gflags
^= IFF_ALLMULTI
;
4661 dev_set_allmulti(dev
, inc
);
4667 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
)
4669 unsigned int changes
= dev
->flags
^ old_flags
;
4671 if (changes
& IFF_UP
) {
4672 if (dev
->flags
& IFF_UP
)
4673 call_netdevice_notifiers(NETDEV_UP
, dev
);
4675 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
4678 if (dev
->flags
& IFF_UP
&&
4679 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
)))
4680 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
4684 * dev_change_flags - change device settings
4686 * @flags: device state flags
4688 * Change settings on device based state flags. The flags are
4689 * in the userspace exported format.
4691 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
4694 int old_flags
= dev
->flags
;
4696 ret
= __dev_change_flags(dev
, flags
);
4700 changes
= old_flags
^ dev
->flags
;
4702 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
4704 __dev_notify_flags(dev
, old_flags
);
4707 EXPORT_SYMBOL(dev_change_flags
);
4710 * dev_set_mtu - Change maximum transfer unit
4712 * @new_mtu: new transfer unit
4714 * Change the maximum transfer size of the network device.
4716 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
4718 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4721 if (new_mtu
== dev
->mtu
)
4724 /* MTU must be positive. */
4728 if (!netif_device_present(dev
))
4732 if (ops
->ndo_change_mtu
)
4733 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
4737 if (!err
&& dev
->flags
& IFF_UP
)
4738 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
4741 EXPORT_SYMBOL(dev_set_mtu
);
4744 * dev_set_group - Change group this device belongs to
4746 * @new_group: group this device should belong to
4748 void dev_set_group(struct net_device
*dev
, int new_group
)
4750 dev
->group
= new_group
;
4752 EXPORT_SYMBOL(dev_set_group
);
4755 * dev_set_mac_address - Change Media Access Control Address
4759 * Change the hardware (MAC) address of the device
4761 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
4763 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4766 if (!ops
->ndo_set_mac_address
)
4768 if (sa
->sa_family
!= dev
->type
)
4770 if (!netif_device_present(dev
))
4772 err
= ops
->ndo_set_mac_address(dev
, sa
);
4774 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4777 EXPORT_SYMBOL(dev_set_mac_address
);
4780 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4782 static int dev_ifsioc_locked(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4785 struct net_device
*dev
= dev_get_by_name_rcu(net
, ifr
->ifr_name
);
4791 case SIOCGIFFLAGS
: /* Get interface flags */
4792 ifr
->ifr_flags
= (short) dev_get_flags(dev
);
4795 case SIOCGIFMETRIC
: /* Get the metric on the interface
4796 (currently unused) */
4797 ifr
->ifr_metric
= 0;
4800 case SIOCGIFMTU
: /* Get the MTU of a device */
4801 ifr
->ifr_mtu
= dev
->mtu
;
4806 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
4808 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
4809 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4810 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
4818 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
4819 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
4820 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
4821 ifr
->ifr_map
.irq
= dev
->irq
;
4822 ifr
->ifr_map
.dma
= dev
->dma
;
4823 ifr
->ifr_map
.port
= dev
->if_port
;
4827 ifr
->ifr_ifindex
= dev
->ifindex
;
4831 ifr
->ifr_qlen
= dev
->tx_queue_len
;
4835 /* dev_ioctl() should ensure this case
4847 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4849 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4852 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
4853 const struct net_device_ops
*ops
;
4858 ops
= dev
->netdev_ops
;
4861 case SIOCSIFFLAGS
: /* Set interface flags */
4862 return dev_change_flags(dev
, ifr
->ifr_flags
);
4864 case SIOCSIFMETRIC
: /* Set the metric on the interface
4865 (currently unused) */
4868 case SIOCSIFMTU
: /* Set the MTU of a device */
4869 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
4872 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
4874 case SIOCSIFHWBROADCAST
:
4875 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
4877 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
4878 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4879 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4883 if (ops
->ndo_set_config
) {
4884 if (!netif_device_present(dev
))
4886 return ops
->ndo_set_config(dev
, &ifr
->ifr_map
);
4891 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
4892 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4894 if (!netif_device_present(dev
))
4896 return dev_mc_add_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4899 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
4900 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4902 if (!netif_device_present(dev
))
4904 return dev_mc_del_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4907 if (ifr
->ifr_qlen
< 0)
4909 dev
->tx_queue_len
= ifr
->ifr_qlen
;
4913 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
4914 return dev_change_name(dev
, ifr
->ifr_newname
);
4917 * Unknown or private ioctl
4920 if ((cmd
>= SIOCDEVPRIVATE
&&
4921 cmd
<= SIOCDEVPRIVATE
+ 15) ||
4922 cmd
== SIOCBONDENSLAVE
||
4923 cmd
== SIOCBONDRELEASE
||
4924 cmd
== SIOCBONDSETHWADDR
||
4925 cmd
== SIOCBONDSLAVEINFOQUERY
||
4926 cmd
== SIOCBONDINFOQUERY
||
4927 cmd
== SIOCBONDCHANGEACTIVE
||
4928 cmd
== SIOCGMIIPHY
||
4929 cmd
== SIOCGMIIREG
||
4930 cmd
== SIOCSMIIREG
||
4931 cmd
== SIOCBRADDIF
||
4932 cmd
== SIOCBRDELIF
||
4933 cmd
== SIOCSHWTSTAMP
||
4934 cmd
== SIOCWANDEV
) {
4936 if (ops
->ndo_do_ioctl
) {
4937 if (netif_device_present(dev
))
4938 err
= ops
->ndo_do_ioctl(dev
, ifr
, cmd
);
4950 * This function handles all "interface"-type I/O control requests. The actual
4951 * 'doing' part of this is dev_ifsioc above.
4955 * dev_ioctl - network device ioctl
4956 * @net: the applicable net namespace
4957 * @cmd: command to issue
4958 * @arg: pointer to a struct ifreq in user space
4960 * Issue ioctl functions to devices. This is normally called by the
4961 * user space syscall interfaces but can sometimes be useful for
4962 * other purposes. The return value is the return from the syscall if
4963 * positive or a negative errno code on error.
4966 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
4972 /* One special case: SIOCGIFCONF takes ifconf argument
4973 and requires shared lock, because it sleeps writing
4977 if (cmd
== SIOCGIFCONF
) {
4979 ret
= dev_ifconf(net
, (char __user
*) arg
);
4983 if (cmd
== SIOCGIFNAME
)
4984 return dev_ifname(net
, (struct ifreq __user
*)arg
);
4986 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
4989 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
4991 colon
= strchr(ifr
.ifr_name
, ':');
4996 * See which interface the caller is talking about.
5001 * These ioctl calls:
5002 * - can be done by all.
5003 * - atomic and do not require locking.
5014 dev_load(net
, ifr
.ifr_name
);
5016 ret
= dev_ifsioc_locked(net
, &ifr
, cmd
);
5021 if (copy_to_user(arg
, &ifr
,
5022 sizeof(struct ifreq
)))
5028 dev_load(net
, ifr
.ifr_name
);
5030 ret
= dev_ethtool(net
, &ifr
);
5035 if (copy_to_user(arg
, &ifr
,
5036 sizeof(struct ifreq
)))
5042 * These ioctl calls:
5043 * - require superuser power.
5044 * - require strict serialization.
5050 if (!capable(CAP_NET_ADMIN
))
5052 dev_load(net
, ifr
.ifr_name
);
5054 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5059 if (copy_to_user(arg
, &ifr
,
5060 sizeof(struct ifreq
)))
5066 * These ioctl calls:
5067 * - require superuser power.
5068 * - require strict serialization.
5069 * - do not return a value
5079 case SIOCSIFHWBROADCAST
:
5082 case SIOCBONDENSLAVE
:
5083 case SIOCBONDRELEASE
:
5084 case SIOCBONDSETHWADDR
:
5085 case SIOCBONDCHANGEACTIVE
:
5089 if (!capable(CAP_NET_ADMIN
))
5092 case SIOCBONDSLAVEINFOQUERY
:
5093 case SIOCBONDINFOQUERY
:
5094 dev_load(net
, ifr
.ifr_name
);
5096 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5101 /* Get the per device memory space. We can add this but
5102 * currently do not support it */
5104 /* Set the per device memory buffer space.
5105 * Not applicable in our case */
5110 * Unknown or private ioctl.
5113 if (cmd
== SIOCWANDEV
||
5114 (cmd
>= SIOCDEVPRIVATE
&&
5115 cmd
<= SIOCDEVPRIVATE
+ 15)) {
5116 dev_load(net
, ifr
.ifr_name
);
5118 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5120 if (!ret
&& copy_to_user(arg
, &ifr
,
5121 sizeof(struct ifreq
)))
5125 /* Take care of Wireless Extensions */
5126 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
5127 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
5134 * dev_new_index - allocate an ifindex
5135 * @net: the applicable net namespace
5137 * Returns a suitable unique value for a new device interface
5138 * number. The caller must hold the rtnl semaphore or the
5139 * dev_base_lock to be sure it remains unique.
5141 static int dev_new_index(struct net
*net
)
5147 if (!__dev_get_by_index(net
, ifindex
))
5152 /* Delayed registration/unregisteration */
5153 static LIST_HEAD(net_todo_list
);
5155 static void net_set_todo(struct net_device
*dev
)
5157 list_add_tail(&dev
->todo_list
, &net_todo_list
);
5160 static void rollback_registered_many(struct list_head
*head
)
5162 struct net_device
*dev
, *tmp
;
5164 BUG_ON(dev_boot_phase
);
5167 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
5168 /* Some devices call without registering
5169 * for initialization unwind. Remove those
5170 * devices and proceed with the remaining.
5172 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5173 pr_debug("unregister_netdevice: device %s/%p never "
5174 "was registered\n", dev
->name
, dev
);
5177 list_del(&dev
->unreg_list
);
5180 dev
->dismantle
= true;
5181 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
5184 /* If device is running, close it first. */
5185 dev_close_many(head
);
5187 list_for_each_entry(dev
, head
, unreg_list
) {
5188 /* And unlink it from device chain. */
5189 unlist_netdevice(dev
);
5191 dev
->reg_state
= NETREG_UNREGISTERING
;
5196 list_for_each_entry(dev
, head
, unreg_list
) {
5197 /* Shutdown queueing discipline. */
5201 /* Notify protocols, that we are about to destroy
5202 this device. They should clean all the things.
5204 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5206 if (!dev
->rtnl_link_ops
||
5207 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5208 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
5211 * Flush the unicast and multicast chains
5216 if (dev
->netdev_ops
->ndo_uninit
)
5217 dev
->netdev_ops
->ndo_uninit(dev
);
5219 /* Notifier chain MUST detach us from master device. */
5220 WARN_ON(dev
->master
);
5222 /* Remove entries from kobject tree */
5223 netdev_unregister_kobject(dev
);
5226 /* Process any work delayed until the end of the batch */
5227 dev
= list_first_entry(head
, struct net_device
, unreg_list
);
5228 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
5232 list_for_each_entry(dev
, head
, unreg_list
)
5236 static void rollback_registered(struct net_device
*dev
)
5240 list_add(&dev
->unreg_list
, &single
);
5241 rollback_registered_many(&single
);
5245 static u32
netdev_fix_features(struct net_device
*dev
, u32 features
)
5247 /* Fix illegal checksum combinations */
5248 if ((features
& NETIF_F_HW_CSUM
) &&
5249 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5250 netdev_warn(dev
, "mixed HW and IP checksum settings.\n");
5251 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5254 if ((features
& NETIF_F_NO_CSUM
) &&
5255 (features
& (NETIF_F_HW_CSUM
|NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5256 netdev_warn(dev
, "mixed no checksumming and other settings.\n");
5257 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
|NETIF_F_HW_CSUM
);
5260 /* Fix illegal SG+CSUM combinations. */
5261 if ((features
& NETIF_F_SG
) &&
5262 !(features
& NETIF_F_ALL_CSUM
)) {
5264 "Dropping NETIF_F_SG since no checksum feature.\n");
5265 features
&= ~NETIF_F_SG
;
5268 /* TSO requires that SG is present as well. */
5269 if ((features
& NETIF_F_ALL_TSO
) && !(features
& NETIF_F_SG
)) {
5270 netdev_dbg(dev
, "Dropping TSO features since no SG feature.\n");
5271 features
&= ~NETIF_F_ALL_TSO
;
5274 /* TSO ECN requires that TSO is present as well. */
5275 if ((features
& NETIF_F_ALL_TSO
) == NETIF_F_TSO_ECN
)
5276 features
&= ~NETIF_F_TSO_ECN
;
5278 /* Software GSO depends on SG. */
5279 if ((features
& NETIF_F_GSO
) && !(features
& NETIF_F_SG
)) {
5280 netdev_dbg(dev
, "Dropping NETIF_F_GSO since no SG feature.\n");
5281 features
&= ~NETIF_F_GSO
;
5284 /* UFO needs SG and checksumming */
5285 if (features
& NETIF_F_UFO
) {
5286 /* maybe split UFO into V4 and V6? */
5287 if (!((features
& NETIF_F_GEN_CSUM
) ||
5288 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))
5289 == (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5291 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5292 features
&= ~NETIF_F_UFO
;
5295 if (!(features
& NETIF_F_SG
)) {
5297 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5298 features
&= ~NETIF_F_UFO
;
5305 int __netdev_update_features(struct net_device
*dev
)
5312 features
= netdev_get_wanted_features(dev
);
5314 if (dev
->netdev_ops
->ndo_fix_features
)
5315 features
= dev
->netdev_ops
->ndo_fix_features(dev
, features
);
5317 /* driver might be less strict about feature dependencies */
5318 features
= netdev_fix_features(dev
, features
);
5320 if (dev
->features
== features
)
5323 netdev_dbg(dev
, "Features changed: 0x%08x -> 0x%08x\n",
5324 dev
->features
, features
);
5326 if (dev
->netdev_ops
->ndo_set_features
)
5327 err
= dev
->netdev_ops
->ndo_set_features(dev
, features
);
5329 if (unlikely(err
< 0)) {
5331 "set_features() failed (%d); wanted 0x%08x, left 0x%08x\n",
5332 err
, features
, dev
->features
);
5337 dev
->features
= features
;
5343 * netdev_update_features - recalculate device features
5344 * @dev: the device to check
5346 * Recalculate dev->features set and send notifications if it
5347 * has changed. Should be called after driver or hardware dependent
5348 * conditions might have changed that influence the features.
5350 void netdev_update_features(struct net_device
*dev
)
5352 if (__netdev_update_features(dev
))
5353 netdev_features_change(dev
);
5355 EXPORT_SYMBOL(netdev_update_features
);
5358 * netdev_change_features - recalculate device features
5359 * @dev: the device to check
5361 * Recalculate dev->features set and send notifications even
5362 * if they have not changed. Should be called instead of
5363 * netdev_update_features() if also dev->vlan_features might
5364 * have changed to allow the changes to be propagated to stacked
5367 void netdev_change_features(struct net_device
*dev
)
5369 __netdev_update_features(dev
);
5370 netdev_features_change(dev
);
5372 EXPORT_SYMBOL(netdev_change_features
);
5375 * netif_stacked_transfer_operstate - transfer operstate
5376 * @rootdev: the root or lower level device to transfer state from
5377 * @dev: the device to transfer operstate to
5379 * Transfer operational state from root to device. This is normally
5380 * called when a stacking relationship exists between the root
5381 * device and the device(a leaf device).
5383 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
5384 struct net_device
*dev
)
5386 if (rootdev
->operstate
== IF_OPER_DORMANT
)
5387 netif_dormant_on(dev
);
5389 netif_dormant_off(dev
);
5391 if (netif_carrier_ok(rootdev
)) {
5392 if (!netif_carrier_ok(dev
))
5393 netif_carrier_on(dev
);
5395 if (netif_carrier_ok(dev
))
5396 netif_carrier_off(dev
);
5399 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
5402 static int netif_alloc_rx_queues(struct net_device
*dev
)
5404 unsigned int i
, count
= dev
->num_rx_queues
;
5405 struct netdev_rx_queue
*rx
;
5409 rx
= kcalloc(count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5411 pr_err("netdev: Unable to allocate %u rx queues.\n", count
);
5416 for (i
= 0; i
< count
; i
++)
5422 static void netdev_init_one_queue(struct net_device
*dev
,
5423 struct netdev_queue
*queue
, void *_unused
)
5425 /* Initialize queue lock */
5426 spin_lock_init(&queue
->_xmit_lock
);
5427 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
5428 queue
->xmit_lock_owner
= -1;
5429 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
5433 static int netif_alloc_netdev_queues(struct net_device
*dev
)
5435 unsigned int count
= dev
->num_tx_queues
;
5436 struct netdev_queue
*tx
;
5440 tx
= kcalloc(count
, sizeof(struct netdev_queue
), GFP_KERNEL
);
5442 pr_err("netdev: Unable to allocate %u tx queues.\n",
5448 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5449 spin_lock_init(&dev
->tx_global_lock
);
5455 * register_netdevice - register a network device
5456 * @dev: device to register
5458 * Take a completed network device structure and add it to the kernel
5459 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5460 * chain. 0 is returned on success. A negative errno code is returned
5461 * on a failure to set up the device, or if the name is a duplicate.
5463 * Callers must hold the rtnl semaphore. You may want
5464 * register_netdev() instead of this.
5467 * The locking appears insufficient to guarantee two parallel registers
5468 * will not get the same name.
5471 int register_netdevice(struct net_device
*dev
)
5474 struct net
*net
= dev_net(dev
);
5476 BUG_ON(dev_boot_phase
);
5481 /* When net_device's are persistent, this will be fatal. */
5482 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
5485 spin_lock_init(&dev
->addr_list_lock
);
5486 netdev_set_addr_lockdep_class(dev
);
5490 ret
= dev_get_valid_name(dev
, dev
->name
);
5494 /* Init, if this function is available */
5495 if (dev
->netdev_ops
->ndo_init
) {
5496 ret
= dev
->netdev_ops
->ndo_init(dev
);
5504 dev
->ifindex
= dev_new_index(net
);
5505 if (dev
->iflink
== -1)
5506 dev
->iflink
= dev
->ifindex
;
5508 /* Transfer changeable features to wanted_features and enable
5509 * software offloads (GSO and GRO).
5511 dev
->hw_features
|= NETIF_F_SOFT_FEATURES
;
5512 dev
->features
|= NETIF_F_SOFT_FEATURES
;
5513 dev
->wanted_features
= dev
->features
& dev
->hw_features
;
5515 /* Turn on no cache copy if HW is doing checksum */
5516 dev
->hw_features
|= NETIF_F_NOCACHE_COPY
;
5517 if ((dev
->features
& NETIF_F_ALL_CSUM
) &&
5518 !(dev
->features
& NETIF_F_NO_CSUM
)) {
5519 dev
->wanted_features
|= NETIF_F_NOCACHE_COPY
;
5520 dev
->features
|= NETIF_F_NOCACHE_COPY
;
5523 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5525 dev
->vlan_features
|= NETIF_F_HIGHDMA
;
5527 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
5528 ret
= notifier_to_errno(ret
);
5532 ret
= netdev_register_kobject(dev
);
5535 dev
->reg_state
= NETREG_REGISTERED
;
5537 __netdev_update_features(dev
);
5540 * Default initial state at registry is that the
5541 * device is present.
5544 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5546 dev_init_scheduler(dev
);
5548 list_netdevice(dev
);
5550 /* Notify protocols, that a new device appeared. */
5551 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5552 ret
= notifier_to_errno(ret
);
5554 rollback_registered(dev
);
5555 dev
->reg_state
= NETREG_UNREGISTERED
;
5558 * Prevent userspace races by waiting until the network
5559 * device is fully setup before sending notifications.
5561 if (!dev
->rtnl_link_ops
||
5562 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5563 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5569 if (dev
->netdev_ops
->ndo_uninit
)
5570 dev
->netdev_ops
->ndo_uninit(dev
);
5573 EXPORT_SYMBOL(register_netdevice
);
5576 * init_dummy_netdev - init a dummy network device for NAPI
5577 * @dev: device to init
5579 * This takes a network device structure and initialize the minimum
5580 * amount of fields so it can be used to schedule NAPI polls without
5581 * registering a full blown interface. This is to be used by drivers
5582 * that need to tie several hardware interfaces to a single NAPI
5583 * poll scheduler due to HW limitations.
5585 int init_dummy_netdev(struct net_device
*dev
)
5587 /* Clear everything. Note we don't initialize spinlocks
5588 * are they aren't supposed to be taken by any of the
5589 * NAPI code and this dummy netdev is supposed to be
5590 * only ever used for NAPI polls
5592 memset(dev
, 0, sizeof(struct net_device
));
5594 /* make sure we BUG if trying to hit standard
5595 * register/unregister code path
5597 dev
->reg_state
= NETREG_DUMMY
;
5599 /* NAPI wants this */
5600 INIT_LIST_HEAD(&dev
->napi_list
);
5602 /* a dummy interface is started by default */
5603 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5604 set_bit(__LINK_STATE_START
, &dev
->state
);
5606 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5607 * because users of this 'device' dont need to change
5613 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
5617 * register_netdev - register a network device
5618 * @dev: device to register
5620 * Take a completed network device structure and add it to the kernel
5621 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5622 * chain. 0 is returned on success. A negative errno code is returned
5623 * on a failure to set up the device, or if the name is a duplicate.
5625 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5626 * and expands the device name if you passed a format string to
5629 int register_netdev(struct net_device
*dev
)
5634 err
= register_netdevice(dev
);
5638 EXPORT_SYMBOL(register_netdev
);
5640 int netdev_refcnt_read(const struct net_device
*dev
)
5644 for_each_possible_cpu(i
)
5645 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
5648 EXPORT_SYMBOL(netdev_refcnt_read
);
5651 * netdev_wait_allrefs - wait until all references are gone.
5653 * This is called when unregistering network devices.
5655 * Any protocol or device that holds a reference should register
5656 * for netdevice notification, and cleanup and put back the
5657 * reference if they receive an UNREGISTER event.
5658 * We can get stuck here if buggy protocols don't correctly
5661 static void netdev_wait_allrefs(struct net_device
*dev
)
5663 unsigned long rebroadcast_time
, warning_time
;
5666 linkwatch_forget_dev(dev
);
5668 rebroadcast_time
= warning_time
= jiffies
;
5669 refcnt
= netdev_refcnt_read(dev
);
5671 while (refcnt
!= 0) {
5672 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
5675 /* Rebroadcast unregister notification */
5676 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5677 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5678 * should have already handle it the first time */
5680 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
5682 /* We must not have linkwatch events
5683 * pending on unregister. If this
5684 * happens, we simply run the queue
5685 * unscheduled, resulting in a noop
5688 linkwatch_run_queue();
5693 rebroadcast_time
= jiffies
;
5698 refcnt
= netdev_refcnt_read(dev
);
5700 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
5701 printk(KERN_EMERG
"unregister_netdevice: "
5702 "waiting for %s to become free. Usage "
5705 warning_time
= jiffies
;
5714 * register_netdevice(x1);
5715 * register_netdevice(x2);
5717 * unregister_netdevice(y1);
5718 * unregister_netdevice(y2);
5724 * We are invoked by rtnl_unlock().
5725 * This allows us to deal with problems:
5726 * 1) We can delete sysfs objects which invoke hotplug
5727 * without deadlocking with linkwatch via keventd.
5728 * 2) Since we run with the RTNL semaphore not held, we can sleep
5729 * safely in order to wait for the netdev refcnt to drop to zero.
5731 * We must not return until all unregister events added during
5732 * the interval the lock was held have been completed.
5734 void netdev_run_todo(void)
5736 struct list_head list
;
5738 /* Snapshot list, allow later requests */
5739 list_replace_init(&net_todo_list
, &list
);
5743 while (!list_empty(&list
)) {
5744 struct net_device
*dev
5745 = list_first_entry(&list
, struct net_device
, todo_list
);
5746 list_del(&dev
->todo_list
);
5748 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
5749 printk(KERN_ERR
"network todo '%s' but state %d\n",
5750 dev
->name
, dev
->reg_state
);
5755 dev
->reg_state
= NETREG_UNREGISTERED
;
5757 on_each_cpu(flush_backlog
, dev
, 1);
5759 netdev_wait_allrefs(dev
);
5762 BUG_ON(netdev_refcnt_read(dev
));
5763 WARN_ON(rcu_access_pointer(dev
->ip_ptr
));
5764 WARN_ON(rcu_access_pointer(dev
->ip6_ptr
));
5765 WARN_ON(dev
->dn_ptr
);
5767 if (dev
->destructor
)
5768 dev
->destructor(dev
);
5770 /* Free network device */
5771 kobject_put(&dev
->dev
.kobj
);
5775 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5776 * fields in the same order, with only the type differing.
5778 static void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
5779 const struct net_device_stats
*netdev_stats
)
5781 #if BITS_PER_LONG == 64
5782 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
5783 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
5785 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
5786 const unsigned long *src
= (const unsigned long *)netdev_stats
;
5787 u64
*dst
= (u64
*)stats64
;
5789 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
5790 sizeof(*stats64
) / sizeof(u64
));
5791 for (i
= 0; i
< n
; i
++)
5797 * dev_get_stats - get network device statistics
5798 * @dev: device to get statistics from
5799 * @storage: place to store stats
5801 * Get network statistics from device. Return @storage.
5802 * The device driver may provide its own method by setting
5803 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5804 * otherwise the internal statistics structure is used.
5806 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
5807 struct rtnl_link_stats64
*storage
)
5809 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5811 if (ops
->ndo_get_stats64
) {
5812 memset(storage
, 0, sizeof(*storage
));
5813 ops
->ndo_get_stats64(dev
, storage
);
5814 } else if (ops
->ndo_get_stats
) {
5815 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
5817 netdev_stats_to_stats64(storage
, &dev
->stats
);
5819 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
5822 EXPORT_SYMBOL(dev_get_stats
);
5824 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
5826 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
5828 #ifdef CONFIG_NET_CLS_ACT
5831 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
5834 netdev_init_one_queue(dev
, queue
, NULL
);
5835 queue
->qdisc
= &noop_qdisc
;
5836 queue
->qdisc_sleeping
= &noop_qdisc
;
5837 rcu_assign_pointer(dev
->ingress_queue
, queue
);
5843 * alloc_netdev_mqs - allocate network device
5844 * @sizeof_priv: size of private data to allocate space for
5845 * @name: device name format string
5846 * @setup: callback to initialize device
5847 * @txqs: the number of TX subqueues to allocate
5848 * @rxqs: the number of RX subqueues to allocate
5850 * Allocates a struct net_device with private data area for driver use
5851 * and performs basic initialization. Also allocates subquue structs
5852 * for each queue on the device.
5854 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
5855 void (*setup
)(struct net_device
*),
5856 unsigned int txqs
, unsigned int rxqs
)
5858 struct net_device
*dev
;
5860 struct net_device
*p
;
5862 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
5865 pr_err("alloc_netdev: Unable to allocate device "
5866 "with zero queues.\n");
5872 pr_err("alloc_netdev: Unable to allocate device "
5873 "with zero RX queues.\n");
5878 alloc_size
= sizeof(struct net_device
);
5880 /* ensure 32-byte alignment of private area */
5881 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
5882 alloc_size
+= sizeof_priv
;
5884 /* ensure 32-byte alignment of whole construct */
5885 alloc_size
+= NETDEV_ALIGN
- 1;
5887 p
= kzalloc(alloc_size
, GFP_KERNEL
);
5889 printk(KERN_ERR
"alloc_netdev: Unable to allocate device.\n");
5893 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
5894 dev
->padded
= (char *)dev
- (char *)p
;
5896 dev
->pcpu_refcnt
= alloc_percpu(int);
5897 if (!dev
->pcpu_refcnt
)
5900 if (dev_addr_init(dev
))
5906 dev_net_set(dev
, &init_net
);
5908 dev
->gso_max_size
= GSO_MAX_SIZE
;
5910 INIT_LIST_HEAD(&dev
->napi_list
);
5911 INIT_LIST_HEAD(&dev
->unreg_list
);
5912 INIT_LIST_HEAD(&dev
->link_watch_list
);
5913 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
5916 dev
->num_tx_queues
= txqs
;
5917 dev
->real_num_tx_queues
= txqs
;
5918 if (netif_alloc_netdev_queues(dev
))
5922 dev
->num_rx_queues
= rxqs
;
5923 dev
->real_num_rx_queues
= rxqs
;
5924 if (netif_alloc_rx_queues(dev
))
5928 strcpy(dev
->name
, name
);
5929 dev
->group
= INIT_NETDEV_GROUP
;
5937 free_percpu(dev
->pcpu_refcnt
);
5947 EXPORT_SYMBOL(alloc_netdev_mqs
);
5950 * free_netdev - free network device
5953 * This function does the last stage of destroying an allocated device
5954 * interface. The reference to the device object is released.
5955 * If this is the last reference then it will be freed.
5957 void free_netdev(struct net_device
*dev
)
5959 struct napi_struct
*p
, *n
;
5961 release_net(dev_net(dev
));
5968 kfree(rcu_dereference_protected(dev
->ingress_queue
, 1));
5970 /* Flush device addresses */
5971 dev_addr_flush(dev
);
5973 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
5976 free_percpu(dev
->pcpu_refcnt
);
5977 dev
->pcpu_refcnt
= NULL
;
5979 /* Compatibility with error handling in drivers */
5980 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5981 kfree((char *)dev
- dev
->padded
);
5985 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
5986 dev
->reg_state
= NETREG_RELEASED
;
5988 /* will free via device release */
5989 put_device(&dev
->dev
);
5991 EXPORT_SYMBOL(free_netdev
);
5994 * synchronize_net - Synchronize with packet receive processing
5996 * Wait for packets currently being received to be done.
5997 * Does not block later packets from starting.
5999 void synchronize_net(void)
6002 if (rtnl_is_locked())
6003 synchronize_rcu_expedited();
6007 EXPORT_SYMBOL(synchronize_net
);
6010 * unregister_netdevice_queue - remove device from the kernel
6014 * This function shuts down a device interface and removes it
6015 * from the kernel tables.
6016 * If head not NULL, device is queued to be unregistered later.
6018 * Callers must hold the rtnl semaphore. You may want
6019 * unregister_netdev() instead of this.
6022 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
6027 list_move_tail(&dev
->unreg_list
, head
);
6029 rollback_registered(dev
);
6030 /* Finish processing unregister after unlock */
6034 EXPORT_SYMBOL(unregister_netdevice_queue
);
6037 * unregister_netdevice_many - unregister many devices
6038 * @head: list of devices
6040 void unregister_netdevice_many(struct list_head
*head
)
6042 struct net_device
*dev
;
6044 if (!list_empty(head
)) {
6045 rollback_registered_many(head
);
6046 list_for_each_entry(dev
, head
, unreg_list
)
6050 EXPORT_SYMBOL(unregister_netdevice_many
);
6053 * unregister_netdev - remove device from the kernel
6056 * This function shuts down a device interface and removes it
6057 * from the kernel tables.
6059 * This is just a wrapper for unregister_netdevice that takes
6060 * the rtnl semaphore. In general you want to use this and not
6061 * unregister_netdevice.
6063 void unregister_netdev(struct net_device
*dev
)
6066 unregister_netdevice(dev
);
6069 EXPORT_SYMBOL(unregister_netdev
);
6072 * dev_change_net_namespace - move device to different nethost namespace
6074 * @net: network namespace
6075 * @pat: If not NULL name pattern to try if the current device name
6076 * is already taken in the destination network namespace.
6078 * This function shuts down a device interface and moves it
6079 * to a new network namespace. On success 0 is returned, on
6080 * a failure a netagive errno code is returned.
6082 * Callers must hold the rtnl semaphore.
6085 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
6091 /* Don't allow namespace local devices to be moved. */
6093 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6096 /* Ensure the device has been registrered */
6098 if (dev
->reg_state
!= NETREG_REGISTERED
)
6101 /* Get out if there is nothing todo */
6103 if (net_eq(dev_net(dev
), net
))
6106 /* Pick the destination device name, and ensure
6107 * we can use it in the destination network namespace.
6110 if (__dev_get_by_name(net
, dev
->name
)) {
6111 /* We get here if we can't use the current device name */
6114 if (dev_get_valid_name(dev
, pat
) < 0)
6119 * And now a mini version of register_netdevice unregister_netdevice.
6122 /* If device is running close it first. */
6125 /* And unlink it from device chain */
6127 unlist_netdevice(dev
);
6131 /* Shutdown queueing discipline. */
6134 /* Notify protocols, that we are about to destroy
6135 this device. They should clean all the things.
6137 Note that dev->reg_state stays at NETREG_REGISTERED.
6138 This is wanted because this way 8021q and macvlan know
6139 the device is just moving and can keep their slaves up.
6141 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6142 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
6145 * Flush the unicast and multicast chains
6150 /* Actually switch the network namespace */
6151 dev_net_set(dev
, net
);
6153 /* If there is an ifindex conflict assign a new one */
6154 if (__dev_get_by_index(net
, dev
->ifindex
)) {
6155 int iflink
= (dev
->iflink
== dev
->ifindex
);
6156 dev
->ifindex
= dev_new_index(net
);
6158 dev
->iflink
= dev
->ifindex
;
6161 /* Fixup kobjects */
6162 err
= device_rename(&dev
->dev
, dev
->name
);
6165 /* Add the device back in the hashes */
6166 list_netdevice(dev
);
6168 /* Notify protocols, that a new device appeared. */
6169 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
6172 * Prevent userspace races by waiting until the network
6173 * device is fully setup before sending notifications.
6175 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
6182 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
6184 static int dev_cpu_callback(struct notifier_block
*nfb
,
6185 unsigned long action
,
6188 struct sk_buff
**list_skb
;
6189 struct sk_buff
*skb
;
6190 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
6191 struct softnet_data
*sd
, *oldsd
;
6193 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
6196 local_irq_disable();
6197 cpu
= smp_processor_id();
6198 sd
= &per_cpu(softnet_data
, cpu
);
6199 oldsd
= &per_cpu(softnet_data
, oldcpu
);
6201 /* Find end of our completion_queue. */
6202 list_skb
= &sd
->completion_queue
;
6204 list_skb
= &(*list_skb
)->next
;
6205 /* Append completion queue from offline CPU. */
6206 *list_skb
= oldsd
->completion_queue
;
6207 oldsd
->completion_queue
= NULL
;
6209 /* Append output queue from offline CPU. */
6210 if (oldsd
->output_queue
) {
6211 *sd
->output_queue_tailp
= oldsd
->output_queue
;
6212 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
6213 oldsd
->output_queue
= NULL
;
6214 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
6216 /* Append NAPI poll list from offline CPU. */
6217 if (!list_empty(&oldsd
->poll_list
)) {
6218 list_splice_init(&oldsd
->poll_list
, &sd
->poll_list
);
6219 raise_softirq_irqoff(NET_RX_SOFTIRQ
);
6222 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
6225 /* Process offline CPU's input_pkt_queue */
6226 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
6228 input_queue_head_incr(oldsd
);
6230 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
6232 input_queue_head_incr(oldsd
);
6240 * netdev_increment_features - increment feature set by one
6241 * @all: current feature set
6242 * @one: new feature set
6243 * @mask: mask feature set
6245 * Computes a new feature set after adding a device with feature set
6246 * @one to the master device with current feature set @all. Will not
6247 * enable anything that is off in @mask. Returns the new feature set.
6249 u32
netdev_increment_features(u32 all
, u32 one
, u32 mask
)
6251 if (mask
& NETIF_F_GEN_CSUM
)
6252 mask
|= NETIF_F_ALL_CSUM
;
6253 mask
|= NETIF_F_VLAN_CHALLENGED
;
6255 all
|= one
& (NETIF_F_ONE_FOR_ALL
|NETIF_F_ALL_CSUM
) & mask
;
6256 all
&= one
| ~NETIF_F_ALL_FOR_ALL
;
6258 /* If device needs checksumming, downgrade to it. */
6259 if (all
& (NETIF_F_ALL_CSUM
& ~NETIF_F_NO_CSUM
))
6260 all
&= ~NETIF_F_NO_CSUM
;
6262 /* If one device supports hw checksumming, set for all. */
6263 if (all
& NETIF_F_GEN_CSUM
)
6264 all
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_GEN_CSUM
);
6268 EXPORT_SYMBOL(netdev_increment_features
);
6270 static struct hlist_head
*netdev_create_hash(void)
6273 struct hlist_head
*hash
;
6275 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
6277 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
6278 INIT_HLIST_HEAD(&hash
[i
]);
6283 /* Initialize per network namespace state */
6284 static int __net_init
netdev_init(struct net
*net
)
6286 INIT_LIST_HEAD(&net
->dev_base_head
);
6288 net
->dev_name_head
= netdev_create_hash();
6289 if (net
->dev_name_head
== NULL
)
6292 net
->dev_index_head
= netdev_create_hash();
6293 if (net
->dev_index_head
== NULL
)
6299 kfree(net
->dev_name_head
);
6305 * netdev_drivername - network driver for the device
6306 * @dev: network device
6308 * Determine network driver for device.
6310 const char *netdev_drivername(const struct net_device
*dev
)
6312 const struct device_driver
*driver
;
6313 const struct device
*parent
;
6314 const char *empty
= "";
6316 parent
= dev
->dev
.parent
;
6320 driver
= parent
->driver
;
6321 if (driver
&& driver
->name
)
6322 return driver
->name
;
6326 static int __netdev_printk(const char *level
, const struct net_device
*dev
,
6327 struct va_format
*vaf
)
6331 if (dev
&& dev
->dev
.parent
)
6332 r
= dev_printk(level
, dev
->dev
.parent
, "%s: %pV",
6333 netdev_name(dev
), vaf
);
6335 r
= printk("%s%s: %pV", level
, netdev_name(dev
), vaf
);
6337 r
= printk("%s(NULL net_device): %pV", level
, vaf
);
6342 int netdev_printk(const char *level
, const struct net_device
*dev
,
6343 const char *format
, ...)
6345 struct va_format vaf
;
6349 va_start(args
, format
);
6354 r
= __netdev_printk(level
, dev
, &vaf
);
6359 EXPORT_SYMBOL(netdev_printk
);
6361 #define define_netdev_printk_level(func, level) \
6362 int func(const struct net_device *dev, const char *fmt, ...) \
6365 struct va_format vaf; \
6368 va_start(args, fmt); \
6373 r = __netdev_printk(level, dev, &vaf); \
6378 EXPORT_SYMBOL(func);
6380 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
6381 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
6382 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
6383 define_netdev_printk_level(netdev_err
, KERN_ERR
);
6384 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
6385 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
6386 define_netdev_printk_level(netdev_info
, KERN_INFO
);
6388 static void __net_exit
netdev_exit(struct net
*net
)
6390 kfree(net
->dev_name_head
);
6391 kfree(net
->dev_index_head
);
6394 static struct pernet_operations __net_initdata netdev_net_ops
= {
6395 .init
= netdev_init
,
6396 .exit
= netdev_exit
,
6399 static void __net_exit
default_device_exit(struct net
*net
)
6401 struct net_device
*dev
, *aux
;
6403 * Push all migratable network devices back to the
6404 * initial network namespace
6407 for_each_netdev_safe(net
, dev
, aux
) {
6409 char fb_name
[IFNAMSIZ
];
6411 /* Ignore unmoveable devices (i.e. loopback) */
6412 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6415 /* Leave virtual devices for the generic cleanup */
6416 if (dev
->rtnl_link_ops
)
6419 /* Push remaining network devices to init_net */
6420 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
6421 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
6423 printk(KERN_EMERG
"%s: failed to move %s to init_net: %d\n",
6424 __func__
, dev
->name
, err
);
6431 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
6433 /* At exit all network devices most be removed from a network
6434 * namespace. Do this in the reverse order of registration.
6435 * Do this across as many network namespaces as possible to
6436 * improve batching efficiency.
6438 struct net_device
*dev
;
6440 LIST_HEAD(dev_kill_list
);
6443 list_for_each_entry(net
, net_list
, exit_list
) {
6444 for_each_netdev_reverse(net
, dev
) {
6445 if (dev
->rtnl_link_ops
)
6446 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
6448 unregister_netdevice_queue(dev
, &dev_kill_list
);
6451 unregister_netdevice_many(&dev_kill_list
);
6452 list_del(&dev_kill_list
);
6456 static struct pernet_operations __net_initdata default_device_ops
= {
6457 .exit
= default_device_exit
,
6458 .exit_batch
= default_device_exit_batch
,
6462 * Initialize the DEV module. At boot time this walks the device list and
6463 * unhooks any devices that fail to initialise (normally hardware not
6464 * present) and leaves us with a valid list of present and active devices.
6469 * This is called single threaded during boot, so no need
6470 * to take the rtnl semaphore.
6472 static int __init
net_dev_init(void)
6474 int i
, rc
= -ENOMEM
;
6476 BUG_ON(!dev_boot_phase
);
6478 if (dev_proc_init())
6481 if (netdev_kobject_init())
6484 INIT_LIST_HEAD(&ptype_all
);
6485 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
6486 INIT_LIST_HEAD(&ptype_base
[i
]);
6488 if (register_pernet_subsys(&netdev_net_ops
))
6492 * Initialise the packet receive queues.
6495 for_each_possible_cpu(i
) {
6496 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
6498 memset(sd
, 0, sizeof(*sd
));
6499 skb_queue_head_init(&sd
->input_pkt_queue
);
6500 skb_queue_head_init(&sd
->process_queue
);
6501 sd
->completion_queue
= NULL
;
6502 INIT_LIST_HEAD(&sd
->poll_list
);
6503 sd
->output_queue
= NULL
;
6504 sd
->output_queue_tailp
= &sd
->output_queue
;
6506 sd
->csd
.func
= rps_trigger_softirq
;
6512 sd
->backlog
.poll
= process_backlog
;
6513 sd
->backlog
.weight
= weight_p
;
6514 sd
->backlog
.gro_list
= NULL
;
6515 sd
->backlog
.gro_count
= 0;
6520 /* The loopback device is special if any other network devices
6521 * is present in a network namespace the loopback device must
6522 * be present. Since we now dynamically allocate and free the
6523 * loopback device ensure this invariant is maintained by
6524 * keeping the loopback device as the first device on the
6525 * list of network devices. Ensuring the loopback devices
6526 * is the first device that appears and the last network device
6529 if (register_pernet_device(&loopback_net_ops
))
6532 if (register_pernet_device(&default_device_ops
))
6535 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
6536 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
6538 hotcpu_notifier(dev_cpu_callback
, 0);
6546 subsys_initcall(net_dev_init
);
6548 static int __init
initialize_hashrnd(void)
6550 get_random_bytes(&hashrnd
, sizeof(hashrnd
));
6554 late_initcall_sync(initialize_hashrnd
);