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 <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <net/net_namespace.h>
99 #include <linux/rtnetlink.h>
100 #include <linux/proc_fs.h>
101 #include <linux/seq_file.h>
102 #include <linux/stat.h>
104 #include <net/pkt_sched.h>
105 #include <net/checksum.h>
106 #include <net/xfrm.h>
107 #include <linux/highmem.h>
108 #include <linux/init.h>
109 #include <linux/kmod.h>
110 #include <linux/module.h>
111 #include <linux/netpoll.h>
112 #include <linux/rcupdate.h>
113 #include <linux/delay.h>
114 #include <net/wext.h>
115 #include <net/iw_handler.h>
116 #include <asm/current.h>
117 #include <linux/audit.h>
118 #include <linux/dmaengine.h>
119 #include <linux/err.h>
120 #include <linux/ctype.h>
121 #include <linux/if_arp.h>
122 #include <linux/if_vlan.h>
123 #include <linux/ip.h>
125 #include <linux/ipv6.h>
126 #include <linux/in.h>
127 #include <linux/jhash.h>
128 #include <linux/random.h>
129 #include <trace/events/napi.h>
130 #include <trace/events/net.h>
131 #include <trace/events/skb.h>
132 #include <linux/pci.h>
133 #include <linux/inetdevice.h>
134 #include <linux/cpu_rmap.h>
135 #include <linux/net_tstamp.h>
136 #include <linux/static_key.h>
137 #include <net/flow_keys.h>
139 #include "net-sysfs.h"
141 /* Instead of increasing this, you should create a hash table. */
142 #define MAX_GRO_SKBS 8
144 /* This should be increased if a protocol with a bigger head is added. */
145 #define GRO_MAX_HEAD (MAX_HEADER + 128)
148 * The list of packet types we will receive (as opposed to discard)
149 * and the routines to invoke.
151 * Why 16. Because with 16 the only overlap we get on a hash of the
152 * low nibble of the protocol value is RARP/SNAP/X.25.
154 * NOTE: That is no longer true with the addition of VLAN tags. Not
155 * sure which should go first, but I bet it won't make much
156 * difference if we are running VLANs. The good news is that
157 * this protocol won't be in the list unless compiled in, so
158 * the average user (w/out VLANs) will not be adversely affected.
175 #define PTYPE_HASH_SIZE (16)
176 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
178 static DEFINE_SPINLOCK(ptype_lock
);
179 static struct list_head ptype_base
[PTYPE_HASH_SIZE
] __read_mostly
;
180 static struct list_head ptype_all __read_mostly
; /* Taps */
183 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
186 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
188 * Writers must hold the rtnl semaphore while they loop through the
189 * dev_base_head list, and hold dev_base_lock for writing when they do the
190 * actual updates. This allows pure readers to access the list even
191 * while a writer is preparing to update it.
193 * To put it another way, dev_base_lock is held for writing only to
194 * protect against pure readers; the rtnl semaphore provides the
195 * protection against other writers.
197 * See, for example usages, register_netdevice() and
198 * unregister_netdevice(), which must be called with the rtnl
201 DEFINE_RWLOCK(dev_base_lock
);
202 EXPORT_SYMBOL(dev_base_lock
);
204 static inline void dev_base_seq_inc(struct net
*net
)
206 while (++net
->dev_base_seq
== 0);
209 static inline struct hlist_head
*dev_name_hash(struct net
*net
, const char *name
)
211 unsigned int hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
213 return &net
->dev_name_head
[hash_32(hash
, NETDEV_HASHBITS
)];
216 static inline struct hlist_head
*dev_index_hash(struct net
*net
, int ifindex
)
218 return &net
->dev_index_head
[ifindex
& (NETDEV_HASHENTRIES
- 1)];
221 static inline void rps_lock(struct softnet_data
*sd
)
224 spin_lock(&sd
->input_pkt_queue
.lock
);
228 static inline void rps_unlock(struct softnet_data
*sd
)
231 spin_unlock(&sd
->input_pkt_queue
.lock
);
235 /* Device list insertion */
236 static int list_netdevice(struct net_device
*dev
)
238 struct net
*net
= dev_net(dev
);
242 write_lock_bh(&dev_base_lock
);
243 list_add_tail_rcu(&dev
->dev_list
, &net
->dev_base_head
);
244 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
245 hlist_add_head_rcu(&dev
->index_hlist
,
246 dev_index_hash(net
, dev
->ifindex
));
247 write_unlock_bh(&dev_base_lock
);
249 dev_base_seq_inc(net
);
254 /* Device list removal
255 * caller must respect a RCU grace period before freeing/reusing dev
257 static void unlist_netdevice(struct net_device
*dev
)
261 /* Unlink dev from the device chain */
262 write_lock_bh(&dev_base_lock
);
263 list_del_rcu(&dev
->dev_list
);
264 hlist_del_rcu(&dev
->name_hlist
);
265 hlist_del_rcu(&dev
->index_hlist
);
266 write_unlock_bh(&dev_base_lock
);
268 dev_base_seq_inc(dev_net(dev
));
275 static RAW_NOTIFIER_HEAD(netdev_chain
);
278 * Device drivers call our routines to queue packets here. We empty the
279 * queue in the local softnet handler.
282 DEFINE_PER_CPU_ALIGNED(struct softnet_data
, softnet_data
);
283 EXPORT_PER_CPU_SYMBOL(softnet_data
);
285 #ifdef CONFIG_LOCKDEP
287 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
288 * according to dev->type
290 static const unsigned short netdev_lock_type
[] =
291 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
292 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
293 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
294 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
295 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
296 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
297 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
298 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
299 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
300 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
301 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
302 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
303 ARPHRD_FCFABRIC
, ARPHRD_IEEE802_TR
, ARPHRD_IEEE80211
,
304 ARPHRD_IEEE80211_PRISM
, ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_PHONET
,
305 ARPHRD_PHONET_PIPE
, ARPHRD_IEEE802154
,
306 ARPHRD_VOID
, ARPHRD_NONE
};
308 static const char *const netdev_lock_name
[] =
309 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
310 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
311 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
312 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
313 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
314 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
315 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
316 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
317 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
318 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
319 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
320 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
321 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
322 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
323 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
324 "_xmit_VOID", "_xmit_NONE"};
326 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
327 static struct lock_class_key netdev_addr_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
329 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
333 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
334 if (netdev_lock_type
[i
] == dev_type
)
336 /* the last key is used by default */
337 return ARRAY_SIZE(netdev_lock_type
) - 1;
340 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
341 unsigned short dev_type
)
345 i
= netdev_lock_pos(dev_type
);
346 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
347 netdev_lock_name
[i
]);
350 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
354 i
= netdev_lock_pos(dev
->type
);
355 lockdep_set_class_and_name(&dev
->addr_list_lock
,
356 &netdev_addr_lock_key
[i
],
357 netdev_lock_name
[i
]);
360 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
361 unsigned short dev_type
)
364 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
369 /*******************************************************************************
371 Protocol management and registration routines
373 *******************************************************************************/
376 * Add a protocol ID to the list. Now that the input handler is
377 * smarter we can dispense with all the messy stuff that used to be
380 * BEWARE!!! Protocol handlers, mangling input packets,
381 * MUST BE last in hash buckets and checking protocol handlers
382 * MUST start from promiscuous ptype_all chain in net_bh.
383 * It is true now, do not change it.
384 * Explanation follows: if protocol handler, mangling packet, will
385 * be the first on list, it is not able to sense, that packet
386 * is cloned and should be copied-on-write, so that it will
387 * change it and subsequent readers will get broken packet.
391 static inline struct list_head
*ptype_head(const struct packet_type
*pt
)
393 if (pt
->type
== htons(ETH_P_ALL
))
396 return &ptype_base
[ntohs(pt
->type
) & PTYPE_HASH_MASK
];
400 * dev_add_pack - add packet handler
401 * @pt: packet type declaration
403 * Add a protocol handler to the networking stack. The passed &packet_type
404 * is linked into kernel lists and may not be freed until it has been
405 * removed from the kernel lists.
407 * This call does not sleep therefore it can not
408 * guarantee all CPU's that are in middle of receiving packets
409 * will see the new packet type (until the next received packet).
412 void dev_add_pack(struct packet_type
*pt
)
414 struct list_head
*head
= ptype_head(pt
);
416 spin_lock(&ptype_lock
);
417 list_add_rcu(&pt
->list
, head
);
418 spin_unlock(&ptype_lock
);
420 EXPORT_SYMBOL(dev_add_pack
);
423 * __dev_remove_pack - remove packet handler
424 * @pt: packet type declaration
426 * Remove a protocol handler that was previously added to the kernel
427 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
428 * from the kernel lists and can be freed or reused once this function
431 * The packet type might still be in use by receivers
432 * and must not be freed until after all the CPU's have gone
433 * through a quiescent state.
435 void __dev_remove_pack(struct packet_type
*pt
)
437 struct list_head
*head
= ptype_head(pt
);
438 struct packet_type
*pt1
;
440 spin_lock(&ptype_lock
);
442 list_for_each_entry(pt1
, head
, list
) {
444 list_del_rcu(&pt
->list
);
449 pr_warn("dev_remove_pack: %p not found\n", pt
);
451 spin_unlock(&ptype_lock
);
453 EXPORT_SYMBOL(__dev_remove_pack
);
456 * dev_remove_pack - remove packet handler
457 * @pt: packet type declaration
459 * Remove a protocol handler that was previously added to the kernel
460 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
461 * from the kernel lists and can be freed or reused once this function
464 * This call sleeps to guarantee that no CPU is looking at the packet
467 void dev_remove_pack(struct packet_type
*pt
)
469 __dev_remove_pack(pt
);
473 EXPORT_SYMBOL(dev_remove_pack
);
475 /******************************************************************************
477 Device Boot-time Settings Routines
479 *******************************************************************************/
481 /* Boot time configuration table */
482 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
485 * netdev_boot_setup_add - add new setup entry
486 * @name: name of the device
487 * @map: configured settings for the device
489 * Adds new setup entry to the dev_boot_setup list. The function
490 * returns 0 on error and 1 on success. This is a generic routine to
493 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
495 struct netdev_boot_setup
*s
;
499 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
500 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
501 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
502 strlcpy(s
[i
].name
, name
, IFNAMSIZ
);
503 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
508 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
512 * netdev_boot_setup_check - check boot time settings
513 * @dev: the netdevice
515 * Check boot time settings for the device.
516 * The found settings are set for the device to be used
517 * later in the device probing.
518 * Returns 0 if no settings found, 1 if they are.
520 int netdev_boot_setup_check(struct net_device
*dev
)
522 struct netdev_boot_setup
*s
= dev_boot_setup
;
525 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
526 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
527 !strcmp(dev
->name
, s
[i
].name
)) {
528 dev
->irq
= s
[i
].map
.irq
;
529 dev
->base_addr
= s
[i
].map
.base_addr
;
530 dev
->mem_start
= s
[i
].map
.mem_start
;
531 dev
->mem_end
= s
[i
].map
.mem_end
;
537 EXPORT_SYMBOL(netdev_boot_setup_check
);
541 * netdev_boot_base - get address from boot time settings
542 * @prefix: prefix for network device
543 * @unit: id for network device
545 * Check boot time settings for the base address of device.
546 * The found settings are set for the device to be used
547 * later in the device probing.
548 * Returns 0 if no settings found.
550 unsigned long netdev_boot_base(const char *prefix
, int unit
)
552 const struct netdev_boot_setup
*s
= dev_boot_setup
;
556 sprintf(name
, "%s%d", prefix
, unit
);
559 * If device already registered then return base of 1
560 * to indicate not to probe for this interface
562 if (__dev_get_by_name(&init_net
, name
))
565 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
566 if (!strcmp(name
, s
[i
].name
))
567 return s
[i
].map
.base_addr
;
572 * Saves at boot time configured settings for any netdevice.
574 int __init
netdev_boot_setup(char *str
)
579 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
584 memset(&map
, 0, sizeof(map
));
588 map
.base_addr
= ints
[2];
590 map
.mem_start
= ints
[3];
592 map
.mem_end
= ints
[4];
594 /* Add new entry to the list */
595 return netdev_boot_setup_add(str
, &map
);
598 __setup("netdev=", netdev_boot_setup
);
600 /*******************************************************************************
602 Device Interface Subroutines
604 *******************************************************************************/
607 * __dev_get_by_name - find a device by its name
608 * @net: the applicable net namespace
609 * @name: name to find
611 * Find an interface by name. Must be called under RTNL semaphore
612 * or @dev_base_lock. If the name is found a pointer to the device
613 * is returned. If the name is not found then %NULL is returned. The
614 * reference counters are not incremented so the caller must be
615 * careful with locks.
618 struct net_device
*__dev_get_by_name(struct net
*net
, const char *name
)
620 struct hlist_node
*p
;
621 struct net_device
*dev
;
622 struct hlist_head
*head
= dev_name_hash(net
, name
);
624 hlist_for_each_entry(dev
, p
, head
, name_hlist
)
625 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
630 EXPORT_SYMBOL(__dev_get_by_name
);
633 * dev_get_by_name_rcu - find a device by its name
634 * @net: the applicable net namespace
635 * @name: name to find
637 * Find an interface by name.
638 * If the name is found a pointer to the device is returned.
639 * If the name is not found then %NULL is returned.
640 * The reference counters are not incremented so the caller must be
641 * careful with locks. The caller must hold RCU lock.
644 struct net_device
*dev_get_by_name_rcu(struct net
*net
, const char *name
)
646 struct hlist_node
*p
;
647 struct net_device
*dev
;
648 struct hlist_head
*head
= dev_name_hash(net
, name
);
650 hlist_for_each_entry_rcu(dev
, p
, head
, name_hlist
)
651 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
656 EXPORT_SYMBOL(dev_get_by_name_rcu
);
659 * dev_get_by_name - find a device by its name
660 * @net: the applicable net namespace
661 * @name: name to find
663 * Find an interface by name. This can be called from any
664 * context and does its own locking. The returned handle has
665 * the usage count incremented and the caller must use dev_put() to
666 * release it when it is no longer needed. %NULL is returned if no
667 * matching device is found.
670 struct net_device
*dev_get_by_name(struct net
*net
, const char *name
)
672 struct net_device
*dev
;
675 dev
= dev_get_by_name_rcu(net
, name
);
681 EXPORT_SYMBOL(dev_get_by_name
);
684 * __dev_get_by_index - find a device by its ifindex
685 * @net: the applicable net namespace
686 * @ifindex: index of device
688 * Search for an interface by index. Returns %NULL if the device
689 * is not found or a pointer to the device. The device has not
690 * had its reference counter increased so the caller must be careful
691 * about locking. The caller must hold either the RTNL semaphore
695 struct net_device
*__dev_get_by_index(struct net
*net
, int ifindex
)
697 struct hlist_node
*p
;
698 struct net_device
*dev
;
699 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
701 hlist_for_each_entry(dev
, p
, head
, index_hlist
)
702 if (dev
->ifindex
== ifindex
)
707 EXPORT_SYMBOL(__dev_get_by_index
);
710 * dev_get_by_index_rcu - find a device by its ifindex
711 * @net: the applicable net namespace
712 * @ifindex: index of device
714 * Search for an interface by index. Returns %NULL if the device
715 * is not found or a pointer to the device. The device has not
716 * had its reference counter increased so the caller must be careful
717 * about locking. The caller must hold RCU lock.
720 struct net_device
*dev_get_by_index_rcu(struct net
*net
, int ifindex
)
722 struct hlist_node
*p
;
723 struct net_device
*dev
;
724 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
726 hlist_for_each_entry_rcu(dev
, p
, head
, index_hlist
)
727 if (dev
->ifindex
== ifindex
)
732 EXPORT_SYMBOL(dev_get_by_index_rcu
);
736 * dev_get_by_index - find a device by its ifindex
737 * @net: the applicable net namespace
738 * @ifindex: index of device
740 * Search for an interface by index. Returns NULL if the device
741 * is not found or a pointer to the device. The device returned has
742 * had a reference added and the pointer is safe until the user calls
743 * dev_put to indicate they have finished with it.
746 struct net_device
*dev_get_by_index(struct net
*net
, int ifindex
)
748 struct net_device
*dev
;
751 dev
= dev_get_by_index_rcu(net
, ifindex
);
757 EXPORT_SYMBOL(dev_get_by_index
);
760 * dev_getbyhwaddr_rcu - find a device by its hardware address
761 * @net: the applicable net namespace
762 * @type: media type of device
763 * @ha: hardware address
765 * Search for an interface by MAC address. Returns NULL if the device
766 * is not found or a pointer to the device.
767 * The caller must hold RCU or RTNL.
768 * The returned device has not had its ref count increased
769 * and the caller must therefore be careful about locking
773 struct net_device
*dev_getbyhwaddr_rcu(struct net
*net
, unsigned short type
,
776 struct net_device
*dev
;
778 for_each_netdev_rcu(net
, dev
)
779 if (dev
->type
== type
&&
780 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
785 EXPORT_SYMBOL(dev_getbyhwaddr_rcu
);
787 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
789 struct net_device
*dev
;
792 for_each_netdev(net
, dev
)
793 if (dev
->type
== type
)
798 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
800 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
802 struct net_device
*dev
, *ret
= NULL
;
805 for_each_netdev_rcu(net
, dev
)
806 if (dev
->type
== type
) {
814 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
817 * dev_get_by_flags_rcu - find any device with given flags
818 * @net: the applicable net namespace
819 * @if_flags: IFF_* values
820 * @mask: bitmask of bits in if_flags to check
822 * Search for any interface with the given flags. Returns NULL if a device
823 * is not found or a pointer to the device. Must be called inside
824 * rcu_read_lock(), and result refcount is unchanged.
827 struct net_device
*dev_get_by_flags_rcu(struct net
*net
, unsigned short if_flags
,
830 struct net_device
*dev
, *ret
;
833 for_each_netdev_rcu(net
, dev
) {
834 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
841 EXPORT_SYMBOL(dev_get_by_flags_rcu
);
844 * dev_valid_name - check if name is okay for network device
847 * Network device names need to be valid file names to
848 * to allow sysfs to work. We also disallow any kind of
851 bool dev_valid_name(const char *name
)
855 if (strlen(name
) >= IFNAMSIZ
)
857 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
861 if (*name
== '/' || isspace(*name
))
867 EXPORT_SYMBOL(dev_valid_name
);
870 * __dev_alloc_name - allocate a name for a device
871 * @net: network namespace to allocate the device name in
872 * @name: name format string
873 * @buf: scratch buffer and result name string
875 * Passed a format string - eg "lt%d" it will try and find a suitable
876 * id. It scans list of devices to build up a free map, then chooses
877 * the first empty slot. The caller must hold the dev_base or rtnl lock
878 * while allocating the name and adding the device in order to avoid
880 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
881 * Returns the number of the unit assigned or a negative errno code.
884 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
888 const int max_netdevices
= 8*PAGE_SIZE
;
889 unsigned long *inuse
;
890 struct net_device
*d
;
892 p
= strnchr(name
, IFNAMSIZ
-1, '%');
895 * Verify the string as this thing may have come from
896 * the user. There must be either one "%d" and no other "%"
899 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
902 /* Use one page as a bit array of possible slots */
903 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
907 for_each_netdev(net
, d
) {
908 if (!sscanf(d
->name
, name
, &i
))
910 if (i
< 0 || i
>= max_netdevices
)
913 /* avoid cases where sscanf is not exact inverse of printf */
914 snprintf(buf
, IFNAMSIZ
, name
, i
);
915 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
919 i
= find_first_zero_bit(inuse
, max_netdevices
);
920 free_page((unsigned long) inuse
);
924 snprintf(buf
, IFNAMSIZ
, name
, i
);
925 if (!__dev_get_by_name(net
, buf
))
928 /* It is possible to run out of possible slots
929 * when the name is long and there isn't enough space left
930 * for the digits, or if all bits are used.
936 * dev_alloc_name - allocate a name for a device
938 * @name: name format string
940 * Passed a format string - eg "lt%d" it will try and find a suitable
941 * id. It scans list of devices to build up a free map, then chooses
942 * the first empty slot. The caller must hold the dev_base or rtnl lock
943 * while allocating the name and adding the device in order to avoid
945 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
946 * Returns the number of the unit assigned or a negative errno code.
949 int dev_alloc_name(struct net_device
*dev
, const char *name
)
955 BUG_ON(!dev_net(dev
));
957 ret
= __dev_alloc_name(net
, name
, buf
);
959 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
962 EXPORT_SYMBOL(dev_alloc_name
);
964 static int dev_get_valid_name(struct net_device
*dev
, const char *name
)
968 BUG_ON(!dev_net(dev
));
971 if (!dev_valid_name(name
))
974 if (strchr(name
, '%'))
975 return dev_alloc_name(dev
, name
);
976 else if (__dev_get_by_name(net
, name
))
978 else if (dev
->name
!= name
)
979 strlcpy(dev
->name
, name
, IFNAMSIZ
);
985 * dev_change_name - change name of a device
987 * @newname: name (or format string) must be at least IFNAMSIZ
989 * Change name of a device, can pass format strings "eth%d".
992 int dev_change_name(struct net_device
*dev
, const char *newname
)
994 char oldname
[IFNAMSIZ
];
1000 BUG_ON(!dev_net(dev
));
1003 if (dev
->flags
& IFF_UP
)
1006 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0)
1009 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
1011 err
= dev_get_valid_name(dev
, newname
);
1016 ret
= device_rename(&dev
->dev
, dev
->name
);
1018 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1022 write_lock_bh(&dev_base_lock
);
1023 hlist_del_rcu(&dev
->name_hlist
);
1024 write_unlock_bh(&dev_base_lock
);
1028 write_lock_bh(&dev_base_lock
);
1029 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
1030 write_unlock_bh(&dev_base_lock
);
1032 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
1033 ret
= notifier_to_errno(ret
);
1036 /* err >= 0 after dev_alloc_name() or stores the first errno */
1039 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1042 pr_err("%s: name change rollback failed: %d\n",
1051 * dev_set_alias - change ifalias of a device
1053 * @alias: name up to IFALIASZ
1054 * @len: limit of bytes to copy from info
1056 * Set ifalias for a device,
1058 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
1062 if (len
>= IFALIASZ
)
1067 kfree(dev
->ifalias
);
1068 dev
->ifalias
= NULL
;
1073 dev
->ifalias
= krealloc(dev
->ifalias
, len
+ 1, GFP_KERNEL
);
1077 strlcpy(dev
->ifalias
, alias
, len
+1);
1083 * netdev_features_change - device changes features
1084 * @dev: device to cause notification
1086 * Called to indicate a device has changed features.
1088 void netdev_features_change(struct net_device
*dev
)
1090 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1092 EXPORT_SYMBOL(netdev_features_change
);
1095 * netdev_state_change - device changes state
1096 * @dev: device to cause notification
1098 * Called to indicate a device has changed state. This function calls
1099 * the notifier chains for netdev_chain and sends a NEWLINK message
1100 * to the routing socket.
1102 void netdev_state_change(struct net_device
*dev
)
1104 if (dev
->flags
& IFF_UP
) {
1105 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
1106 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
1109 EXPORT_SYMBOL(netdev_state_change
);
1111 int netdev_bonding_change(struct net_device
*dev
, unsigned long event
)
1113 return call_netdevice_notifiers(event
, dev
);
1115 EXPORT_SYMBOL(netdev_bonding_change
);
1118 * dev_load - load a network module
1119 * @net: the applicable net namespace
1120 * @name: name of interface
1122 * If a network interface is not present and the process has suitable
1123 * privileges this function loads the module. If module loading is not
1124 * available in this kernel then it becomes a nop.
1127 void dev_load(struct net
*net
, const char *name
)
1129 struct net_device
*dev
;
1133 dev
= dev_get_by_name_rcu(net
, name
);
1137 if (no_module
&& capable(CAP_NET_ADMIN
))
1138 no_module
= request_module("netdev-%s", name
);
1139 if (no_module
&& capable(CAP_SYS_MODULE
)) {
1140 if (!request_module("%s", name
))
1141 pr_err("Loading kernel module for a network device with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s instead.\n",
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
)
1321 * If we're trying to disable lro on a vlan device
1322 * use the underlying physical device instead
1324 if (is_vlan_dev(dev
))
1325 dev
= vlan_dev_real_dev(dev
);
1327 dev
->wanted_features
&= ~NETIF_F_LRO
;
1328 netdev_update_features(dev
);
1330 if (unlikely(dev
->features
& NETIF_F_LRO
))
1331 netdev_WARN(dev
, "failed to disable LRO!\n");
1333 EXPORT_SYMBOL(dev_disable_lro
);
1336 static int dev_boot_phase
= 1;
1339 * register_netdevice_notifier - register a network notifier block
1342 * Register a notifier to be called when network device events occur.
1343 * The notifier passed is linked into the kernel structures and must
1344 * not be reused until it has been unregistered. A negative errno code
1345 * is returned on a failure.
1347 * When registered all registration and up events are replayed
1348 * to the new notifier to allow device to have a race free
1349 * view of the network device list.
1352 int register_netdevice_notifier(struct notifier_block
*nb
)
1354 struct net_device
*dev
;
1355 struct net_device
*last
;
1360 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1366 for_each_netdev(net
, dev
) {
1367 err
= nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1368 err
= notifier_to_errno(err
);
1372 if (!(dev
->flags
& IFF_UP
))
1375 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1386 for_each_netdev(net
, dev
) {
1390 if (dev
->flags
& IFF_UP
) {
1391 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1392 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1394 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1395 nb
->notifier_call(nb
, NETDEV_UNREGISTER_BATCH
, dev
);
1400 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1403 EXPORT_SYMBOL(register_netdevice_notifier
);
1406 * unregister_netdevice_notifier - unregister a network notifier block
1409 * Unregister a notifier previously registered by
1410 * register_netdevice_notifier(). The notifier is unlinked into the
1411 * kernel structures and may then be reused. A negative errno code
1412 * is returned on a failure.
1414 * After unregistering unregister and down device events are synthesized
1415 * for all devices on the device list to the removed notifier to remove
1416 * the need for special case cleanup code.
1419 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1421 struct net_device
*dev
;
1426 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1431 for_each_netdev(net
, dev
) {
1432 if (dev
->flags
& IFF_UP
) {
1433 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1434 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1436 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1437 nb
->notifier_call(nb
, NETDEV_UNREGISTER_BATCH
, dev
);
1444 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1447 * call_netdevice_notifiers - call all network notifier blocks
1448 * @val: value passed unmodified to notifier function
1449 * @dev: net_device pointer passed unmodified to notifier function
1451 * Call all network notifier blocks. Parameters and return value
1452 * are as for raw_notifier_call_chain().
1455 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1458 return raw_notifier_call_chain(&netdev_chain
, val
, dev
);
1460 EXPORT_SYMBOL(call_netdevice_notifiers
);
1462 static struct static_key netstamp_needed __read_mostly
;
1463 #ifdef HAVE_JUMP_LABEL
1464 /* We are not allowed to call static_key_slow_dec() from irq context
1465 * If net_disable_timestamp() is called from irq context, defer the
1466 * static_key_slow_dec() calls.
1468 static atomic_t netstamp_needed_deferred
;
1471 void net_enable_timestamp(void)
1473 #ifdef HAVE_JUMP_LABEL
1474 int deferred
= atomic_xchg(&netstamp_needed_deferred
, 0);
1478 static_key_slow_dec(&netstamp_needed
);
1482 WARN_ON(in_interrupt());
1483 static_key_slow_inc(&netstamp_needed
);
1485 EXPORT_SYMBOL(net_enable_timestamp
);
1487 void net_disable_timestamp(void)
1489 #ifdef HAVE_JUMP_LABEL
1490 if (in_interrupt()) {
1491 atomic_inc(&netstamp_needed_deferred
);
1495 static_key_slow_dec(&netstamp_needed
);
1497 EXPORT_SYMBOL(net_disable_timestamp
);
1499 static inline void net_timestamp_set(struct sk_buff
*skb
)
1501 skb
->tstamp
.tv64
= 0;
1502 if (static_key_false(&netstamp_needed
))
1503 __net_timestamp(skb
);
1506 #define net_timestamp_check(COND, SKB) \
1507 if (static_key_false(&netstamp_needed)) { \
1508 if ((COND) && !(SKB)->tstamp.tv64) \
1509 __net_timestamp(SKB); \
1512 static int net_hwtstamp_validate(struct ifreq *ifr)
1514 struct hwtstamp_config cfg
;
1515 enum hwtstamp_tx_types tx_type
;
1516 enum hwtstamp_rx_filters rx_filter
;
1517 int tx_type_valid
= 0;
1518 int rx_filter_valid
= 0;
1520 if (copy_from_user(&cfg
, ifr
->ifr_data
, sizeof(cfg
)))
1523 if (cfg
.flags
) /* reserved for future extensions */
1526 tx_type
= cfg
.tx_type
;
1527 rx_filter
= cfg
.rx_filter
;
1530 case HWTSTAMP_TX_OFF
:
1531 case HWTSTAMP_TX_ON
:
1532 case HWTSTAMP_TX_ONESTEP_SYNC
:
1537 switch (rx_filter
) {
1538 case HWTSTAMP_FILTER_NONE
:
1539 case HWTSTAMP_FILTER_ALL
:
1540 case HWTSTAMP_FILTER_SOME
:
1541 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT
:
1542 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC
:
1543 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ
:
1544 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT
:
1545 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC
:
1546 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ
:
1547 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT
:
1548 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC
:
1549 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ
:
1550 case HWTSTAMP_FILTER_PTP_V2_EVENT
:
1551 case HWTSTAMP_FILTER_PTP_V2_SYNC
:
1552 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ
:
1553 rx_filter_valid
= 1;
1557 if (!tx_type_valid
|| !rx_filter_valid
)
1563 static inline bool is_skb_forwardable(struct net_device
*dev
,
1564 struct sk_buff
*skb
)
1568 if (!(dev
->flags
& IFF_UP
))
1571 len
= dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
;
1572 if (skb
->len
<= len
)
1575 /* if TSO is enabled, we don't care about the length as the packet
1576 * could be forwarded without being segmented before
1578 if (skb_is_gso(skb
))
1585 * dev_forward_skb - loopback an skb to another netif
1587 * @dev: destination network device
1588 * @skb: buffer to forward
1591 * NET_RX_SUCCESS (no congestion)
1592 * NET_RX_DROP (packet was dropped, but freed)
1594 * dev_forward_skb can be used for injecting an skb from the
1595 * start_xmit function of one device into the receive queue
1596 * of another device.
1598 * The receiving device may be in another namespace, so
1599 * we have to clear all information in the skb that could
1600 * impact namespace isolation.
1602 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1604 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
1605 if (skb_copy_ubufs(skb
, GFP_ATOMIC
)) {
1606 atomic_long_inc(&dev
->rx_dropped
);
1615 if (unlikely(!is_skb_forwardable(dev
, skb
))) {
1616 atomic_long_inc(&dev
->rx_dropped
);
1621 skb_set_dev(skb
, dev
);
1622 skb
->tstamp
.tv64
= 0;
1623 skb
->pkt_type
= PACKET_HOST
;
1624 skb
->protocol
= eth_type_trans(skb
, dev
);
1625 return netif_rx(skb
);
1627 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1629 static inline int deliver_skb(struct sk_buff
*skb
,
1630 struct packet_type
*pt_prev
,
1631 struct net_device
*orig_dev
)
1633 atomic_inc(&skb
->users
);
1634 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1638 * Support routine. Sends outgoing frames to any network
1639 * taps currently in use.
1642 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1644 struct packet_type
*ptype
;
1645 struct sk_buff
*skb2
= NULL
;
1646 struct packet_type
*pt_prev
= NULL
;
1649 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1650 /* Never send packets back to the socket
1651 * they originated from - MvS (miquels@drinkel.ow.org)
1653 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1654 (ptype
->af_packet_priv
== NULL
||
1655 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1657 deliver_skb(skb2
, pt_prev
, skb
->dev
);
1662 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1666 net_timestamp_set(skb2
);
1668 /* skb->nh should be correctly
1669 set by sender, so that the second statement is
1670 just protection against buggy protocols.
1672 skb_reset_mac_header(skb2
);
1674 if (skb_network_header(skb2
) < skb2
->data
||
1675 skb2
->network_header
> skb2
->tail
) {
1676 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1677 ntohs(skb2
->protocol
),
1679 skb_reset_network_header(skb2
);
1682 skb2
->transport_header
= skb2
->network_header
;
1683 skb2
->pkt_type
= PACKET_OUTGOING
;
1688 pt_prev
->func(skb2
, skb
->dev
, pt_prev
, skb
->dev
);
1692 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1693 * @dev: Network device
1694 * @txq: number of queues available
1696 * If real_num_tx_queues is changed the tc mappings may no longer be
1697 * valid. To resolve this verify the tc mapping remains valid and if
1698 * not NULL the mapping. With no priorities mapping to this
1699 * offset/count pair it will no longer be used. In the worst case TC0
1700 * is invalid nothing can be done so disable priority mappings. If is
1701 * expected that drivers will fix this mapping if they can before
1702 * calling netif_set_real_num_tx_queues.
1704 static void netif_setup_tc(struct net_device
*dev
, unsigned int txq
)
1707 struct netdev_tc_txq
*tc
= &dev
->tc_to_txq
[0];
1709 /* If TC0 is invalidated disable TC mapping */
1710 if (tc
->offset
+ tc
->count
> txq
) {
1711 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1716 /* Invalidated prio to tc mappings set to TC0 */
1717 for (i
= 1; i
< TC_BITMASK
+ 1; i
++) {
1718 int q
= netdev_get_prio_tc_map(dev
, i
);
1720 tc
= &dev
->tc_to_txq
[q
];
1721 if (tc
->offset
+ tc
->count
> txq
) {
1722 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1724 netdev_set_prio_tc_map(dev
, i
, 0);
1730 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1731 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1733 int netif_set_real_num_tx_queues(struct net_device
*dev
, unsigned int txq
)
1737 if (txq
< 1 || txq
> dev
->num_tx_queues
)
1740 if (dev
->reg_state
== NETREG_REGISTERED
||
1741 dev
->reg_state
== NETREG_UNREGISTERING
) {
1744 rc
= netdev_queue_update_kobjects(dev
, dev
->real_num_tx_queues
,
1750 netif_setup_tc(dev
, txq
);
1752 if (txq
< dev
->real_num_tx_queues
)
1753 qdisc_reset_all_tx_gt(dev
, txq
);
1756 dev
->real_num_tx_queues
= txq
;
1759 EXPORT_SYMBOL(netif_set_real_num_tx_queues
);
1763 * netif_set_real_num_rx_queues - set actual number of RX queues used
1764 * @dev: Network device
1765 * @rxq: Actual number of RX queues
1767 * This must be called either with the rtnl_lock held or before
1768 * registration of the net device. Returns 0 on success, or a
1769 * negative error code. If called before registration, it always
1772 int netif_set_real_num_rx_queues(struct net_device
*dev
, unsigned int rxq
)
1776 if (rxq
< 1 || rxq
> dev
->num_rx_queues
)
1779 if (dev
->reg_state
== NETREG_REGISTERED
) {
1782 rc
= net_rx_queue_update_kobjects(dev
, dev
->real_num_rx_queues
,
1788 dev
->real_num_rx_queues
= rxq
;
1791 EXPORT_SYMBOL(netif_set_real_num_rx_queues
);
1794 static inline void __netif_reschedule(struct Qdisc
*q
)
1796 struct softnet_data
*sd
;
1797 unsigned long flags
;
1799 local_irq_save(flags
);
1800 sd
= &__get_cpu_var(softnet_data
);
1801 q
->next_sched
= NULL
;
1802 *sd
->output_queue_tailp
= q
;
1803 sd
->output_queue_tailp
= &q
->next_sched
;
1804 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1805 local_irq_restore(flags
);
1808 void __netif_schedule(struct Qdisc
*q
)
1810 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
1811 __netif_reschedule(q
);
1813 EXPORT_SYMBOL(__netif_schedule
);
1815 void dev_kfree_skb_irq(struct sk_buff
*skb
)
1817 if (atomic_dec_and_test(&skb
->users
)) {
1818 struct softnet_data
*sd
;
1819 unsigned long flags
;
1821 local_irq_save(flags
);
1822 sd
= &__get_cpu_var(softnet_data
);
1823 skb
->next
= sd
->completion_queue
;
1824 sd
->completion_queue
= skb
;
1825 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1826 local_irq_restore(flags
);
1829 EXPORT_SYMBOL(dev_kfree_skb_irq
);
1831 void dev_kfree_skb_any(struct sk_buff
*skb
)
1833 if (in_irq() || irqs_disabled())
1834 dev_kfree_skb_irq(skb
);
1838 EXPORT_SYMBOL(dev_kfree_skb_any
);
1842 * netif_device_detach - mark device as removed
1843 * @dev: network device
1845 * Mark device as removed from system and therefore no longer available.
1847 void netif_device_detach(struct net_device
*dev
)
1849 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1850 netif_running(dev
)) {
1851 netif_tx_stop_all_queues(dev
);
1854 EXPORT_SYMBOL(netif_device_detach
);
1857 * netif_device_attach - mark device as attached
1858 * @dev: network device
1860 * Mark device as attached from system and restart if needed.
1862 void netif_device_attach(struct net_device
*dev
)
1864 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1865 netif_running(dev
)) {
1866 netif_tx_wake_all_queues(dev
);
1867 __netdev_watchdog_up(dev
);
1870 EXPORT_SYMBOL(netif_device_attach
);
1873 * skb_dev_set -- assign a new device to a buffer
1874 * @skb: buffer for the new device
1875 * @dev: network device
1877 * If an skb is owned by a device already, we have to reset
1878 * all data private to the namespace a device belongs to
1879 * before assigning it a new device.
1881 #ifdef CONFIG_NET_NS
1882 void skb_set_dev(struct sk_buff
*skb
, struct net_device
*dev
)
1885 if (skb
->dev
&& !net_eq(dev_net(skb
->dev
), dev_net(dev
))) {
1888 skb_init_secmark(skb
);
1892 skb
->ipvs_property
= 0;
1893 #ifdef CONFIG_NET_SCHED
1899 EXPORT_SYMBOL(skb_set_dev
);
1900 #endif /* CONFIG_NET_NS */
1902 static void skb_warn_bad_offload(const struct sk_buff
*skb
)
1904 static const netdev_features_t null_features
= 0;
1905 struct net_device
*dev
= skb
->dev
;
1906 const char *driver
= "";
1908 if (dev
&& dev
->dev
.parent
)
1909 driver
= dev_driver_string(dev
->dev
.parent
);
1911 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
1912 "gso_type=%d ip_summed=%d\n",
1913 driver
, dev
? &dev
->features
: &null_features
,
1914 skb
->sk
? &skb
->sk
->sk_route_caps
: &null_features
,
1915 skb
->len
, skb
->data_len
, skb_shinfo(skb
)->gso_size
,
1916 skb_shinfo(skb
)->gso_type
, skb
->ip_summed
);
1920 * Invalidate hardware checksum when packet is to be mangled, and
1921 * complete checksum manually on outgoing path.
1923 int skb_checksum_help(struct sk_buff
*skb
)
1926 int ret
= 0, offset
;
1928 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1929 goto out_set_summed
;
1931 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1932 skb_warn_bad_offload(skb
);
1936 offset
= skb_checksum_start_offset(skb
);
1937 BUG_ON(offset
>= skb_headlen(skb
));
1938 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1940 offset
+= skb
->csum_offset
;
1941 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
1943 if (skb_cloned(skb
) &&
1944 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
1945 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1950 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1952 skb
->ip_summed
= CHECKSUM_NONE
;
1956 EXPORT_SYMBOL(skb_checksum_help
);
1959 * skb_gso_segment - Perform segmentation on skb.
1960 * @skb: buffer to segment
1961 * @features: features for the output path (see dev->features)
1963 * This function segments the given skb and returns a list of segments.
1965 * It may return NULL if the skb requires no segmentation. This is
1966 * only possible when GSO is used for verifying header integrity.
1968 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
,
1969 netdev_features_t features
)
1971 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1972 struct packet_type
*ptype
;
1973 __be16 type
= skb
->protocol
;
1974 int vlan_depth
= ETH_HLEN
;
1977 while (type
== htons(ETH_P_8021Q
)) {
1978 struct vlan_hdr
*vh
;
1980 if (unlikely(!pskb_may_pull(skb
, vlan_depth
+ VLAN_HLEN
)))
1981 return ERR_PTR(-EINVAL
);
1983 vh
= (struct vlan_hdr
*)(skb
->data
+ vlan_depth
);
1984 type
= vh
->h_vlan_encapsulated_proto
;
1985 vlan_depth
+= VLAN_HLEN
;
1988 skb_reset_mac_header(skb
);
1989 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1990 __skb_pull(skb
, skb
->mac_len
);
1992 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1993 skb_warn_bad_offload(skb
);
1995 if (skb_header_cloned(skb
) &&
1996 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
1997 return ERR_PTR(err
);
2001 list_for_each_entry_rcu(ptype
,
2002 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
2003 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
2004 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
2005 err
= ptype
->gso_send_check(skb
);
2006 segs
= ERR_PTR(err
);
2007 if (err
|| skb_gso_ok(skb
, features
))
2009 __skb_push(skb
, (skb
->data
-
2010 skb_network_header(skb
)));
2012 segs
= ptype
->gso_segment(skb
, features
);
2018 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
2022 EXPORT_SYMBOL(skb_gso_segment
);
2024 /* Take action when hardware reception checksum errors are detected. */
2026 void netdev_rx_csum_fault(struct net_device
*dev
)
2028 if (net_ratelimit()) {
2029 pr_err("%s: hw csum failure\n", dev
? dev
->name
: "<unknown>");
2033 EXPORT_SYMBOL(netdev_rx_csum_fault
);
2036 /* Actually, we should eliminate this check as soon as we know, that:
2037 * 1. IOMMU is present and allows to map all the memory.
2038 * 2. No high memory really exists on this machine.
2041 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
2043 #ifdef CONFIG_HIGHMEM
2045 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
2046 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2047 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2048 if (PageHighMem(skb_frag_page(frag
)))
2053 if (PCI_DMA_BUS_IS_PHYS
) {
2054 struct device
*pdev
= dev
->dev
.parent
;
2058 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2059 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2060 dma_addr_t addr
= page_to_phys(skb_frag_page(frag
));
2061 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
2070 void (*destructor
)(struct sk_buff
*skb
);
2073 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2075 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
2077 struct dev_gso_cb
*cb
;
2080 struct sk_buff
*nskb
= skb
->next
;
2082 skb
->next
= nskb
->next
;
2085 } while (skb
->next
);
2087 cb
= DEV_GSO_CB(skb
);
2089 cb
->destructor(skb
);
2093 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2094 * @skb: buffer to segment
2095 * @features: device features as applicable to this skb
2097 * This function segments the given skb and stores the list of segments
2100 static int dev_gso_segment(struct sk_buff
*skb
, netdev_features_t features
)
2102 struct sk_buff
*segs
;
2104 segs
= skb_gso_segment(skb
, features
);
2106 /* Verifying header integrity only. */
2111 return PTR_ERR(segs
);
2114 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
2115 skb
->destructor
= dev_gso_skb_destructor
;
2121 * Try to orphan skb early, right before transmission by the device.
2122 * We cannot orphan skb if tx timestamp is requested or the sk-reference
2123 * is needed on driver level for other reasons, e.g. see net/can/raw.c
2125 static inline void skb_orphan_try(struct sk_buff
*skb
)
2127 struct sock
*sk
= skb
->sk
;
2129 if (sk
&& !skb_shinfo(skb
)->tx_flags
) {
2130 /* skb_tx_hash() wont be able to get sk.
2131 * We copy sk_hash into skb->rxhash
2134 skb
->rxhash
= sk
->sk_hash
;
2139 static bool can_checksum_protocol(netdev_features_t features
, __be16 protocol
)
2141 return ((features
& NETIF_F_GEN_CSUM
) ||
2142 ((features
& NETIF_F_V4_CSUM
) &&
2143 protocol
== htons(ETH_P_IP
)) ||
2144 ((features
& NETIF_F_V6_CSUM
) &&
2145 protocol
== htons(ETH_P_IPV6
)) ||
2146 ((features
& NETIF_F_FCOE_CRC
) &&
2147 protocol
== htons(ETH_P_FCOE
)));
2150 static netdev_features_t
harmonize_features(struct sk_buff
*skb
,
2151 __be16 protocol
, netdev_features_t features
)
2153 if (!can_checksum_protocol(features
, protocol
)) {
2154 features
&= ~NETIF_F_ALL_CSUM
;
2155 features
&= ~NETIF_F_SG
;
2156 } else if (illegal_highdma(skb
->dev
, skb
)) {
2157 features
&= ~NETIF_F_SG
;
2163 netdev_features_t
netif_skb_features(struct sk_buff
*skb
)
2165 __be16 protocol
= skb
->protocol
;
2166 netdev_features_t features
= skb
->dev
->features
;
2168 if (protocol
== htons(ETH_P_8021Q
)) {
2169 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
2170 protocol
= veh
->h_vlan_encapsulated_proto
;
2171 } else if (!vlan_tx_tag_present(skb
)) {
2172 return harmonize_features(skb
, protocol
, features
);
2175 features
&= (skb
->dev
->vlan_features
| NETIF_F_HW_VLAN_TX
);
2177 if (protocol
!= htons(ETH_P_8021Q
)) {
2178 return harmonize_features(skb
, protocol
, features
);
2180 features
&= NETIF_F_SG
| NETIF_F_HIGHDMA
| NETIF_F_FRAGLIST
|
2181 NETIF_F_GEN_CSUM
| NETIF_F_HW_VLAN_TX
;
2182 return harmonize_features(skb
, protocol
, features
);
2185 EXPORT_SYMBOL(netif_skb_features
);
2188 * Returns true if either:
2189 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2190 * 2. skb is fragmented and the device does not support SG, or if
2191 * at least one of fragments is in highmem and device does not
2192 * support DMA from it.
2194 static inline int skb_needs_linearize(struct sk_buff
*skb
,
2197 return skb_is_nonlinear(skb
) &&
2198 ((skb_has_frag_list(skb
) &&
2199 !(features
& NETIF_F_FRAGLIST
)) ||
2200 (skb_shinfo(skb
)->nr_frags
&&
2201 !(features
& NETIF_F_SG
)));
2204 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
2205 struct netdev_queue
*txq
)
2207 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2208 int rc
= NETDEV_TX_OK
;
2209 unsigned int skb_len
;
2211 if (likely(!skb
->next
)) {
2212 netdev_features_t features
;
2215 * If device doesn't need skb->dst, release it right now while
2216 * its hot in this cpu cache
2218 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2221 if (!list_empty(&ptype_all
))
2222 dev_queue_xmit_nit(skb
, dev
);
2224 skb_orphan_try(skb
);
2226 features
= netif_skb_features(skb
);
2228 if (vlan_tx_tag_present(skb
) &&
2229 !(features
& NETIF_F_HW_VLAN_TX
)) {
2230 skb
= __vlan_put_tag(skb
, vlan_tx_tag_get(skb
));
2237 if (netif_needs_gso(skb
, features
)) {
2238 if (unlikely(dev_gso_segment(skb
, features
)))
2243 if (skb_needs_linearize(skb
, features
) &&
2244 __skb_linearize(skb
))
2247 /* If packet is not checksummed and device does not
2248 * support checksumming for this protocol, complete
2249 * checksumming here.
2251 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2252 skb_set_transport_header(skb
,
2253 skb_checksum_start_offset(skb
));
2254 if (!(features
& NETIF_F_ALL_CSUM
) &&
2255 skb_checksum_help(skb
))
2261 rc
= ops
->ndo_start_xmit(skb
, dev
);
2262 trace_net_dev_xmit(skb
, rc
, dev
, skb_len
);
2263 if (rc
== NETDEV_TX_OK
)
2264 txq_trans_update(txq
);
2270 struct sk_buff
*nskb
= skb
->next
;
2272 skb
->next
= nskb
->next
;
2276 * If device doesn't need nskb->dst, release it right now while
2277 * its hot in this cpu cache
2279 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2282 skb_len
= nskb
->len
;
2283 rc
= ops
->ndo_start_xmit(nskb
, dev
);
2284 trace_net_dev_xmit(nskb
, rc
, dev
, skb_len
);
2285 if (unlikely(rc
!= NETDEV_TX_OK
)) {
2286 if (rc
& ~NETDEV_TX_MASK
)
2287 goto out_kfree_gso_skb
;
2288 nskb
->next
= skb
->next
;
2292 txq_trans_update(txq
);
2293 if (unlikely(netif_xmit_stopped(txq
) && skb
->next
))
2294 return NETDEV_TX_BUSY
;
2295 } while (skb
->next
);
2298 if (likely(skb
->next
== NULL
))
2299 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
2306 static u32 hashrnd __read_mostly
;
2309 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2310 * to be used as a distribution range.
2312 u16
__skb_tx_hash(const struct net_device
*dev
, const struct sk_buff
*skb
,
2313 unsigned int num_tx_queues
)
2317 u16 qcount
= num_tx_queues
;
2319 if (skb_rx_queue_recorded(skb
)) {
2320 hash
= skb_get_rx_queue(skb
);
2321 while (unlikely(hash
>= num_tx_queues
))
2322 hash
-= num_tx_queues
;
2327 u8 tc
= netdev_get_prio_tc_map(dev
, skb
->priority
);
2328 qoffset
= dev
->tc_to_txq
[tc
].offset
;
2329 qcount
= dev
->tc_to_txq
[tc
].count
;
2332 if (skb
->sk
&& skb
->sk
->sk_hash
)
2333 hash
= skb
->sk
->sk_hash
;
2335 hash
= (__force u16
) skb
->protocol
^ skb
->rxhash
;
2336 hash
= jhash_1word(hash
, hashrnd
);
2338 return (u16
) (((u64
) hash
* qcount
) >> 32) + qoffset
;
2340 EXPORT_SYMBOL(__skb_tx_hash
);
2342 static inline u16
dev_cap_txqueue(struct net_device
*dev
, u16 queue_index
)
2344 if (unlikely(queue_index
>= dev
->real_num_tx_queues
)) {
2345 net_warn_ratelimited("%s selects TX queue %d, but real number of TX queues is %d\n",
2346 dev
->name
, queue_index
,
2347 dev
->real_num_tx_queues
);
2353 static inline int get_xps_queue(struct net_device
*dev
, struct sk_buff
*skb
)
2356 struct xps_dev_maps
*dev_maps
;
2357 struct xps_map
*map
;
2358 int queue_index
= -1;
2361 dev_maps
= rcu_dereference(dev
->xps_maps
);
2363 map
= rcu_dereference(
2364 dev_maps
->cpu_map
[raw_smp_processor_id()]);
2367 queue_index
= map
->queues
[0];
2370 if (skb
->sk
&& skb
->sk
->sk_hash
)
2371 hash
= skb
->sk
->sk_hash
;
2373 hash
= (__force u16
) skb
->protocol
^
2375 hash
= jhash_1word(hash
, hashrnd
);
2376 queue_index
= map
->queues
[
2377 ((u64
)hash
* map
->len
) >> 32];
2379 if (unlikely(queue_index
>= dev
->real_num_tx_queues
))
2391 static struct netdev_queue
*dev_pick_tx(struct net_device
*dev
,
2392 struct sk_buff
*skb
)
2395 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2397 if (dev
->real_num_tx_queues
== 1)
2399 else if (ops
->ndo_select_queue
) {
2400 queue_index
= ops
->ndo_select_queue(dev
, skb
);
2401 queue_index
= dev_cap_txqueue(dev
, queue_index
);
2403 struct sock
*sk
= skb
->sk
;
2404 queue_index
= sk_tx_queue_get(sk
);
2406 if (queue_index
< 0 || skb
->ooo_okay
||
2407 queue_index
>= dev
->real_num_tx_queues
) {
2408 int old_index
= queue_index
;
2410 queue_index
= get_xps_queue(dev
, skb
);
2411 if (queue_index
< 0)
2412 queue_index
= skb_tx_hash(dev
, skb
);
2414 if (queue_index
!= old_index
&& sk
) {
2415 struct dst_entry
*dst
=
2416 rcu_dereference_check(sk
->sk_dst_cache
, 1);
2418 if (dst
&& skb_dst(skb
) == dst
)
2419 sk_tx_queue_set(sk
, queue_index
);
2424 skb_set_queue_mapping(skb
, queue_index
);
2425 return netdev_get_tx_queue(dev
, queue_index
);
2428 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2429 struct net_device
*dev
,
2430 struct netdev_queue
*txq
)
2432 spinlock_t
*root_lock
= qdisc_lock(q
);
2436 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
2437 qdisc_calculate_pkt_len(skb
, q
);
2439 * Heuristic to force contended enqueues to serialize on a
2440 * separate lock before trying to get qdisc main lock.
2441 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2442 * and dequeue packets faster.
2444 contended
= qdisc_is_running(q
);
2445 if (unlikely(contended
))
2446 spin_lock(&q
->busylock
);
2448 spin_lock(root_lock
);
2449 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2452 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2453 qdisc_run_begin(q
)) {
2455 * This is a work-conserving queue; there are no old skbs
2456 * waiting to be sent out; and the qdisc is not running -
2457 * xmit the skb directly.
2459 if (!(dev
->priv_flags
& IFF_XMIT_DST_RELEASE
))
2462 qdisc_bstats_update(q
, skb
);
2464 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
)) {
2465 if (unlikely(contended
)) {
2466 spin_unlock(&q
->busylock
);
2473 rc
= NET_XMIT_SUCCESS
;
2476 rc
= q
->enqueue(skb
, q
) & NET_XMIT_MASK
;
2477 if (qdisc_run_begin(q
)) {
2478 if (unlikely(contended
)) {
2479 spin_unlock(&q
->busylock
);
2485 spin_unlock(root_lock
);
2486 if (unlikely(contended
))
2487 spin_unlock(&q
->busylock
);
2491 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2492 static void skb_update_prio(struct sk_buff
*skb
)
2494 struct netprio_map
*map
= rcu_dereference_bh(skb
->dev
->priomap
);
2496 if ((!skb
->priority
) && (skb
->sk
) && map
)
2497 skb
->priority
= map
->priomap
[skb
->sk
->sk_cgrp_prioidx
];
2500 #define skb_update_prio(skb)
2503 static DEFINE_PER_CPU(int, xmit_recursion
);
2504 #define RECURSION_LIMIT 10
2507 * dev_queue_xmit - transmit a buffer
2508 * @skb: buffer to transmit
2510 * Queue a buffer for transmission to a network device. The caller must
2511 * have set the device and priority and built the buffer before calling
2512 * this function. The function can be called from an interrupt.
2514 * A negative errno code is returned on a failure. A success does not
2515 * guarantee the frame will be transmitted as it may be dropped due
2516 * to congestion or traffic shaping.
2518 * -----------------------------------------------------------------------------------
2519 * I notice this method can also return errors from the queue disciplines,
2520 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2523 * Regardless of the return value, the skb is consumed, so it is currently
2524 * difficult to retry a send to this method. (You can bump the ref count
2525 * before sending to hold a reference for retry if you are careful.)
2527 * When calling this method, interrupts MUST be enabled. This is because
2528 * the BH enable code must have IRQs enabled so that it will not deadlock.
2531 int dev_queue_xmit(struct sk_buff
*skb
)
2533 struct net_device
*dev
= skb
->dev
;
2534 struct netdev_queue
*txq
;
2538 /* Disable soft irqs for various locks below. Also
2539 * stops preemption for RCU.
2543 skb_update_prio(skb
);
2545 txq
= dev_pick_tx(dev
, skb
);
2546 q
= rcu_dereference_bh(txq
->qdisc
);
2548 #ifdef CONFIG_NET_CLS_ACT
2549 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2551 trace_net_dev_queue(skb
);
2553 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2557 /* The device has no queue. Common case for software devices:
2558 loopback, all the sorts of tunnels...
2560 Really, it is unlikely that netif_tx_lock protection is necessary
2561 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2563 However, it is possible, that they rely on protection
2566 Check this and shot the lock. It is not prone from deadlocks.
2567 Either shot noqueue qdisc, it is even simpler 8)
2569 if (dev
->flags
& IFF_UP
) {
2570 int cpu
= smp_processor_id(); /* ok because BHs are off */
2572 if (txq
->xmit_lock_owner
!= cpu
) {
2574 if (__this_cpu_read(xmit_recursion
) > RECURSION_LIMIT
)
2575 goto recursion_alert
;
2577 HARD_TX_LOCK(dev
, txq
, cpu
);
2579 if (!netif_xmit_stopped(txq
)) {
2580 __this_cpu_inc(xmit_recursion
);
2581 rc
= dev_hard_start_xmit(skb
, dev
, txq
);
2582 __this_cpu_dec(xmit_recursion
);
2583 if (dev_xmit_complete(rc
)) {
2584 HARD_TX_UNLOCK(dev
, txq
);
2588 HARD_TX_UNLOCK(dev
, txq
);
2589 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2592 /* Recursion is detected! It is possible,
2596 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2602 rcu_read_unlock_bh();
2607 rcu_read_unlock_bh();
2610 EXPORT_SYMBOL(dev_queue_xmit
);
2613 /*=======================================================================
2615 =======================================================================*/
2617 int netdev_max_backlog __read_mostly
= 1000;
2618 int netdev_tstamp_prequeue __read_mostly
= 1;
2619 int netdev_budget __read_mostly
= 300;
2620 int weight_p __read_mostly
= 64; /* old backlog weight */
2622 /* Called with irq disabled */
2623 static inline void ____napi_schedule(struct softnet_data
*sd
,
2624 struct napi_struct
*napi
)
2626 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
2627 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2631 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2632 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2633 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2634 * if hash is a canonical 4-tuple hash over transport ports.
2636 void __skb_get_rxhash(struct sk_buff
*skb
)
2638 struct flow_keys keys
;
2641 if (!skb_flow_dissect(skb
, &keys
))
2645 if ((__force u16
)keys
.port16
[1] < (__force u16
)keys
.port16
[0])
2646 swap(keys
.port16
[0], keys
.port16
[1]);
2650 /* get a consistent hash (same value on both flow directions) */
2651 if ((__force u32
)keys
.dst
< (__force u32
)keys
.src
)
2652 swap(keys
.dst
, keys
.src
);
2654 hash
= jhash_3words((__force u32
)keys
.dst
,
2655 (__force u32
)keys
.src
,
2656 (__force u32
)keys
.ports
, hashrnd
);
2662 EXPORT_SYMBOL(__skb_get_rxhash
);
2666 /* One global table that all flow-based protocols share. */
2667 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
2668 EXPORT_SYMBOL(rps_sock_flow_table
);
2670 struct static_key rps_needed __read_mostly
;
2672 static struct rps_dev_flow
*
2673 set_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2674 struct rps_dev_flow
*rflow
, u16 next_cpu
)
2676 if (next_cpu
!= RPS_NO_CPU
) {
2677 #ifdef CONFIG_RFS_ACCEL
2678 struct netdev_rx_queue
*rxqueue
;
2679 struct rps_dev_flow_table
*flow_table
;
2680 struct rps_dev_flow
*old_rflow
;
2685 /* Should we steer this flow to a different hardware queue? */
2686 if (!skb_rx_queue_recorded(skb
) || !dev
->rx_cpu_rmap
||
2687 !(dev
->features
& NETIF_F_NTUPLE
))
2689 rxq_index
= cpu_rmap_lookup_index(dev
->rx_cpu_rmap
, next_cpu
);
2690 if (rxq_index
== skb_get_rx_queue(skb
))
2693 rxqueue
= dev
->_rx
+ rxq_index
;
2694 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2697 flow_id
= skb
->rxhash
& flow_table
->mask
;
2698 rc
= dev
->netdev_ops
->ndo_rx_flow_steer(dev
, skb
,
2699 rxq_index
, flow_id
);
2703 rflow
= &flow_table
->flows
[flow_id
];
2705 if (old_rflow
->filter
== rflow
->filter
)
2706 old_rflow
->filter
= RPS_NO_FILTER
;
2710 per_cpu(softnet_data
, next_cpu
).input_queue_head
;
2713 rflow
->cpu
= next_cpu
;
2718 * get_rps_cpu is called from netif_receive_skb and returns the target
2719 * CPU from the RPS map of the receiving queue for a given skb.
2720 * rcu_read_lock must be held on entry.
2722 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2723 struct rps_dev_flow
**rflowp
)
2725 struct netdev_rx_queue
*rxqueue
;
2726 struct rps_map
*map
;
2727 struct rps_dev_flow_table
*flow_table
;
2728 struct rps_sock_flow_table
*sock_flow_table
;
2732 if (skb_rx_queue_recorded(skb
)) {
2733 u16 index
= skb_get_rx_queue(skb
);
2734 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
2735 WARN_ONCE(dev
->real_num_rx_queues
> 1,
2736 "%s received packet on queue %u, but number "
2737 "of RX queues is %u\n",
2738 dev
->name
, index
, dev
->real_num_rx_queues
);
2741 rxqueue
= dev
->_rx
+ index
;
2745 map
= rcu_dereference(rxqueue
->rps_map
);
2747 if (map
->len
== 1 &&
2748 !rcu_access_pointer(rxqueue
->rps_flow_table
)) {
2749 tcpu
= map
->cpus
[0];
2750 if (cpu_online(tcpu
))
2754 } else if (!rcu_access_pointer(rxqueue
->rps_flow_table
)) {
2758 skb_reset_network_header(skb
);
2759 if (!skb_get_rxhash(skb
))
2762 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2763 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
2764 if (flow_table
&& sock_flow_table
) {
2766 struct rps_dev_flow
*rflow
;
2768 rflow
= &flow_table
->flows
[skb
->rxhash
& flow_table
->mask
];
2771 next_cpu
= sock_flow_table
->ents
[skb
->rxhash
&
2772 sock_flow_table
->mask
];
2775 * If the desired CPU (where last recvmsg was done) is
2776 * different from current CPU (one in the rx-queue flow
2777 * table entry), switch if one of the following holds:
2778 * - Current CPU is unset (equal to RPS_NO_CPU).
2779 * - Current CPU is offline.
2780 * - The current CPU's queue tail has advanced beyond the
2781 * last packet that was enqueued using this table entry.
2782 * This guarantees that all previous packets for the flow
2783 * have been dequeued, thus preserving in order delivery.
2785 if (unlikely(tcpu
!= next_cpu
) &&
2786 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
2787 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
2788 rflow
->last_qtail
)) >= 0))
2789 rflow
= set_rps_cpu(dev
, skb
, rflow
, next_cpu
);
2791 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
2799 tcpu
= map
->cpus
[((u64
) skb
->rxhash
* map
->len
) >> 32];
2801 if (cpu_online(tcpu
)) {
2811 #ifdef CONFIG_RFS_ACCEL
2814 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2815 * @dev: Device on which the filter was set
2816 * @rxq_index: RX queue index
2817 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2818 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2820 * Drivers that implement ndo_rx_flow_steer() should periodically call
2821 * this function for each installed filter and remove the filters for
2822 * which it returns %true.
2824 bool rps_may_expire_flow(struct net_device
*dev
, u16 rxq_index
,
2825 u32 flow_id
, u16 filter_id
)
2827 struct netdev_rx_queue
*rxqueue
= dev
->_rx
+ rxq_index
;
2828 struct rps_dev_flow_table
*flow_table
;
2829 struct rps_dev_flow
*rflow
;
2834 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2835 if (flow_table
&& flow_id
<= flow_table
->mask
) {
2836 rflow
= &flow_table
->flows
[flow_id
];
2837 cpu
= ACCESS_ONCE(rflow
->cpu
);
2838 if (rflow
->filter
== filter_id
&& cpu
!= RPS_NO_CPU
&&
2839 ((int)(per_cpu(softnet_data
, cpu
).input_queue_head
-
2840 rflow
->last_qtail
) <
2841 (int)(10 * flow_table
->mask
)))
2847 EXPORT_SYMBOL(rps_may_expire_flow
);
2849 #endif /* CONFIG_RFS_ACCEL */
2851 /* Called from hardirq (IPI) context */
2852 static void rps_trigger_softirq(void *data
)
2854 struct softnet_data
*sd
= data
;
2856 ____napi_schedule(sd
, &sd
->backlog
);
2860 #endif /* CONFIG_RPS */
2863 * Check if this softnet_data structure is another cpu one
2864 * If yes, queue it to our IPI list and return 1
2867 static int rps_ipi_queued(struct softnet_data
*sd
)
2870 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
2873 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
2874 mysd
->rps_ipi_list
= sd
;
2876 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2879 #endif /* CONFIG_RPS */
2884 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2885 * queue (may be a remote CPU queue).
2887 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
2888 unsigned int *qtail
)
2890 struct softnet_data
*sd
;
2891 unsigned long flags
;
2893 sd
= &per_cpu(softnet_data
, cpu
);
2895 local_irq_save(flags
);
2898 if (skb_queue_len(&sd
->input_pkt_queue
) <= netdev_max_backlog
) {
2899 if (skb_queue_len(&sd
->input_pkt_queue
)) {
2901 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
2902 input_queue_tail_incr_save(sd
, qtail
);
2904 local_irq_restore(flags
);
2905 return NET_RX_SUCCESS
;
2908 /* Schedule NAPI for backlog device
2909 * We can use non atomic operation since we own the queue lock
2911 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
2912 if (!rps_ipi_queued(sd
))
2913 ____napi_schedule(sd
, &sd
->backlog
);
2921 local_irq_restore(flags
);
2923 atomic_long_inc(&skb
->dev
->rx_dropped
);
2929 * netif_rx - post buffer to the network code
2930 * @skb: buffer to post
2932 * This function receives a packet from a device driver and queues it for
2933 * the upper (protocol) levels to process. It always succeeds. The buffer
2934 * may be dropped during processing for congestion control or by the
2938 * NET_RX_SUCCESS (no congestion)
2939 * NET_RX_DROP (packet was dropped)
2943 int netif_rx(struct sk_buff
*skb
)
2947 /* if netpoll wants it, pretend we never saw it */
2948 if (netpoll_rx(skb
))
2951 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
2953 trace_netif_rx(skb
);
2955 if (static_key_false(&rps_needed
)) {
2956 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
2962 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
2964 cpu
= smp_processor_id();
2966 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
2974 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
2979 EXPORT_SYMBOL(netif_rx
);
2981 int netif_rx_ni(struct sk_buff
*skb
)
2986 err
= netif_rx(skb
);
2987 if (local_softirq_pending())
2993 EXPORT_SYMBOL(netif_rx_ni
);
2995 static void net_tx_action(struct softirq_action
*h
)
2997 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
2999 if (sd
->completion_queue
) {
3000 struct sk_buff
*clist
;
3002 local_irq_disable();
3003 clist
= sd
->completion_queue
;
3004 sd
->completion_queue
= NULL
;
3008 struct sk_buff
*skb
= clist
;
3009 clist
= clist
->next
;
3011 WARN_ON(atomic_read(&skb
->users
));
3012 trace_kfree_skb(skb
, net_tx_action
);
3017 if (sd
->output_queue
) {
3020 local_irq_disable();
3021 head
= sd
->output_queue
;
3022 sd
->output_queue
= NULL
;
3023 sd
->output_queue_tailp
= &sd
->output_queue
;
3027 struct Qdisc
*q
= head
;
3028 spinlock_t
*root_lock
;
3030 head
= head
->next_sched
;
3032 root_lock
= qdisc_lock(q
);
3033 if (spin_trylock(root_lock
)) {
3034 smp_mb__before_clear_bit();
3035 clear_bit(__QDISC_STATE_SCHED
,
3038 spin_unlock(root_lock
);
3040 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
3042 __netif_reschedule(q
);
3044 smp_mb__before_clear_bit();
3045 clear_bit(__QDISC_STATE_SCHED
,
3053 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3054 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3055 /* This hook is defined here for ATM LANE */
3056 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
3057 unsigned char *addr
) __read_mostly
;
3058 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
3061 #ifdef CONFIG_NET_CLS_ACT
3062 /* TODO: Maybe we should just force sch_ingress to be compiled in
3063 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3064 * a compare and 2 stores extra right now if we dont have it on
3065 * but have CONFIG_NET_CLS_ACT
3066 * NOTE: This doesn't stop any functionality; if you dont have
3067 * the ingress scheduler, you just can't add policies on ingress.
3070 static int ing_filter(struct sk_buff
*skb
, struct netdev_queue
*rxq
)
3072 struct net_device
*dev
= skb
->dev
;
3073 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
3074 int result
= TC_ACT_OK
;
3077 if (unlikely(MAX_RED_LOOP
< ttl
++)) {
3078 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3079 skb
->skb_iif
, dev
->ifindex
);
3083 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
3084 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
3087 if (q
!= &noop_qdisc
) {
3088 spin_lock(qdisc_lock(q
));
3089 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
3090 result
= qdisc_enqueue_root(skb
, q
);
3091 spin_unlock(qdisc_lock(q
));
3097 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
3098 struct packet_type
**pt_prev
,
3099 int *ret
, struct net_device
*orig_dev
)
3101 struct netdev_queue
*rxq
= rcu_dereference(skb
->dev
->ingress_queue
);
3103 if (!rxq
|| rxq
->qdisc
== &noop_qdisc
)
3107 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
3111 switch (ing_filter(skb
, rxq
)) {
3125 * netdev_rx_handler_register - register receive handler
3126 * @dev: device to register a handler for
3127 * @rx_handler: receive handler to register
3128 * @rx_handler_data: data pointer that is used by rx handler
3130 * Register a receive hander for a device. This handler will then be
3131 * called from __netif_receive_skb. A negative errno code is returned
3134 * The caller must hold the rtnl_mutex.
3136 * For a general description of rx_handler, see enum rx_handler_result.
3138 int netdev_rx_handler_register(struct net_device
*dev
,
3139 rx_handler_func_t
*rx_handler
,
3140 void *rx_handler_data
)
3144 if (dev
->rx_handler
)
3147 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
3148 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
3152 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
3155 * netdev_rx_handler_unregister - unregister receive handler
3156 * @dev: device to unregister a handler from
3158 * Unregister a receive hander from a device.
3160 * The caller must hold the rtnl_mutex.
3162 void netdev_rx_handler_unregister(struct net_device
*dev
)
3166 RCU_INIT_POINTER(dev
->rx_handler
, NULL
);
3167 RCU_INIT_POINTER(dev
->rx_handler_data
, NULL
);
3169 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
3171 static int __netif_receive_skb(struct sk_buff
*skb
)
3173 struct packet_type
*ptype
, *pt_prev
;
3174 rx_handler_func_t
*rx_handler
;
3175 struct net_device
*orig_dev
;
3176 struct net_device
*null_or_dev
;
3177 bool deliver_exact
= false;
3178 int ret
= NET_RX_DROP
;
3181 net_timestamp_check(!netdev_tstamp_prequeue
, skb
);
3183 trace_netif_receive_skb(skb
);
3185 /* if we've gotten here through NAPI, check netpoll */
3186 if (netpoll_receive_skb(skb
))
3190 skb
->skb_iif
= skb
->dev
->ifindex
;
3191 orig_dev
= skb
->dev
;
3193 skb_reset_network_header(skb
);
3194 skb_reset_transport_header(skb
);
3195 skb_reset_mac_len(skb
);
3203 __this_cpu_inc(softnet_data
.processed
);
3205 if (skb
->protocol
== cpu_to_be16(ETH_P_8021Q
)) {
3206 skb
= vlan_untag(skb
);
3211 #ifdef CONFIG_NET_CLS_ACT
3212 if (skb
->tc_verd
& TC_NCLS
) {
3213 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
3218 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
3219 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
3221 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3226 #ifdef CONFIG_NET_CLS_ACT
3227 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
3233 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
3234 if (vlan_tx_tag_present(skb
)) {
3236 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3239 if (vlan_do_receive(&skb
, !rx_handler
))
3241 else if (unlikely(!skb
))
3247 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3250 switch (rx_handler(&skb
)) {
3251 case RX_HANDLER_CONSUMED
:
3253 case RX_HANDLER_ANOTHER
:
3255 case RX_HANDLER_EXACT
:
3256 deliver_exact
= true;
3257 case RX_HANDLER_PASS
:
3264 /* deliver only exact match when indicated */
3265 null_or_dev
= deliver_exact
? skb
->dev
: NULL
;
3267 type
= skb
->protocol
;
3268 list_for_each_entry_rcu(ptype
,
3269 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
3270 if (ptype
->type
== type
&&
3271 (ptype
->dev
== null_or_dev
|| ptype
->dev
== skb
->dev
||
3272 ptype
->dev
== orig_dev
)) {
3274 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3280 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
3282 atomic_long_inc(&skb
->dev
->rx_dropped
);
3284 /* Jamal, now you will not able to escape explaining
3285 * me how you were going to use this. :-)
3296 * netif_receive_skb - process receive buffer from network
3297 * @skb: buffer to process
3299 * netif_receive_skb() is the main receive data processing function.
3300 * It always succeeds. The buffer may be dropped during processing
3301 * for congestion control or by the protocol layers.
3303 * This function may only be called from softirq context and interrupts
3304 * should be enabled.
3306 * Return values (usually ignored):
3307 * NET_RX_SUCCESS: no congestion
3308 * NET_RX_DROP: packet was dropped
3310 int netif_receive_skb(struct sk_buff
*skb
)
3312 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3314 if (skb_defer_rx_timestamp(skb
))
3315 return NET_RX_SUCCESS
;
3318 if (static_key_false(&rps_needed
)) {
3319 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3324 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3327 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3334 return __netif_receive_skb(skb
);
3336 EXPORT_SYMBOL(netif_receive_skb
);
3338 /* Network device is going away, flush any packets still pending
3339 * Called with irqs disabled.
3341 static void flush_backlog(void *arg
)
3343 struct net_device
*dev
= arg
;
3344 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3345 struct sk_buff
*skb
, *tmp
;
3348 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3349 if (skb
->dev
== dev
) {
3350 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3352 input_queue_head_incr(sd
);
3357 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3358 if (skb
->dev
== dev
) {
3359 __skb_unlink(skb
, &sd
->process_queue
);
3361 input_queue_head_incr(sd
);
3366 static int napi_gro_complete(struct sk_buff
*skb
)
3368 struct packet_type
*ptype
;
3369 __be16 type
= skb
->protocol
;
3370 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3373 if (NAPI_GRO_CB(skb
)->count
== 1) {
3374 skb_shinfo(skb
)->gso_size
= 0;
3379 list_for_each_entry_rcu(ptype
, head
, list
) {
3380 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_complete
)
3383 err
= ptype
->gro_complete(skb
);
3389 WARN_ON(&ptype
->list
== head
);
3391 return NET_RX_SUCCESS
;
3395 return netif_receive_skb(skb
);
3398 inline void napi_gro_flush(struct napi_struct
*napi
)
3400 struct sk_buff
*skb
, *next
;
3402 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3405 napi_gro_complete(skb
);
3408 napi
->gro_count
= 0;
3409 napi
->gro_list
= NULL
;
3411 EXPORT_SYMBOL(napi_gro_flush
);
3413 enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3415 struct sk_buff
**pp
= NULL
;
3416 struct packet_type
*ptype
;
3417 __be16 type
= skb
->protocol
;
3418 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3421 enum gro_result ret
;
3423 if (!(skb
->dev
->features
& NETIF_F_GRO
) || netpoll_rx_on(skb
))
3426 if (skb_is_gso(skb
) || skb_has_frag_list(skb
))
3430 list_for_each_entry_rcu(ptype
, head
, list
) {
3431 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_receive
)
3434 skb_set_network_header(skb
, skb_gro_offset(skb
));
3435 mac_len
= skb
->network_header
- skb
->mac_header
;
3436 skb
->mac_len
= mac_len
;
3437 NAPI_GRO_CB(skb
)->same_flow
= 0;
3438 NAPI_GRO_CB(skb
)->flush
= 0;
3439 NAPI_GRO_CB(skb
)->free
= 0;
3441 pp
= ptype
->gro_receive(&napi
->gro_list
, skb
);
3446 if (&ptype
->list
== head
)
3449 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3450 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3453 struct sk_buff
*nskb
= *pp
;
3457 napi_gro_complete(nskb
);
3464 if (NAPI_GRO_CB(skb
)->flush
|| napi
->gro_count
>= MAX_GRO_SKBS
)
3468 NAPI_GRO_CB(skb
)->count
= 1;
3469 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
3470 skb
->next
= napi
->gro_list
;
3471 napi
->gro_list
= skb
;
3475 if (skb_headlen(skb
) < skb_gro_offset(skb
)) {
3476 int grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
3478 BUG_ON(skb
->end
- skb
->tail
< grow
);
3480 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3483 skb
->data_len
-= grow
;
3485 skb_shinfo(skb
)->frags
[0].page_offset
+= grow
;
3486 skb_frag_size_sub(&skb_shinfo(skb
)->frags
[0], grow
);
3488 if (unlikely(!skb_frag_size(&skb_shinfo(skb
)->frags
[0]))) {
3489 skb_frag_unref(skb
, 0);
3490 memmove(skb_shinfo(skb
)->frags
,
3491 skb_shinfo(skb
)->frags
+ 1,
3492 --skb_shinfo(skb
)->nr_frags
* sizeof(skb_frag_t
));
3503 EXPORT_SYMBOL(dev_gro_receive
);
3505 static inline gro_result_t
3506 __napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3509 unsigned int maclen
= skb
->dev
->hard_header_len
;
3511 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3512 unsigned long diffs
;
3514 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
3515 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
3516 if (maclen
== ETH_HLEN
)
3517 diffs
|= compare_ether_header(skb_mac_header(p
),
3518 skb_gro_mac_header(skb
));
3520 diffs
= memcmp(skb_mac_header(p
),
3521 skb_gro_mac_header(skb
),
3523 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
3524 NAPI_GRO_CB(p
)->flush
= 0;
3527 return dev_gro_receive(napi
, skb
);
3530 gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
3534 if (netif_receive_skb(skb
))
3542 case GRO_MERGED_FREE
:
3543 if (NAPI_GRO_CB(skb
)->free
== NAPI_GRO_FREE_STOLEN_HEAD
)
3544 kmem_cache_free(skbuff_head_cache
, skb
);
3556 EXPORT_SYMBOL(napi_skb_finish
);
3558 void skb_gro_reset_offset(struct sk_buff
*skb
)
3560 NAPI_GRO_CB(skb
)->data_offset
= 0;
3561 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3562 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3564 if (skb
->mac_header
== skb
->tail
&&
3565 !PageHighMem(skb_frag_page(&skb_shinfo(skb
)->frags
[0]))) {
3566 NAPI_GRO_CB(skb
)->frag0
=
3567 skb_frag_address(&skb_shinfo(skb
)->frags
[0]);
3568 NAPI_GRO_CB(skb
)->frag0_len
= skb_frag_size(&skb_shinfo(skb
)->frags
[0]);
3571 EXPORT_SYMBOL(skb_gro_reset_offset
);
3573 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3575 skb_gro_reset_offset(skb
);
3577 return napi_skb_finish(__napi_gro_receive(napi
, skb
), skb
);
3579 EXPORT_SYMBOL(napi_gro_receive
);
3581 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
3583 __skb_pull(skb
, skb_headlen(skb
));
3584 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3585 skb_reserve(skb
, NET_SKB_PAD
+ NET_IP_ALIGN
- skb_headroom(skb
));
3587 skb
->dev
= napi
->dev
;
3593 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
3595 struct sk_buff
*skb
= napi
->skb
;
3598 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
3604 EXPORT_SYMBOL(napi_get_frags
);
3606 gro_result_t
napi_frags_finish(struct napi_struct
*napi
, struct sk_buff
*skb
,
3612 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
3614 if (ret
== GRO_HELD
)
3615 skb_gro_pull(skb
, -ETH_HLEN
);
3616 else if (netif_receive_skb(skb
))
3621 case GRO_MERGED_FREE
:
3622 napi_reuse_skb(napi
, skb
);
3631 EXPORT_SYMBOL(napi_frags_finish
);
3633 struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
3635 struct sk_buff
*skb
= napi
->skb
;
3642 skb_reset_mac_header(skb
);
3643 skb_gro_reset_offset(skb
);
3645 off
= skb_gro_offset(skb
);
3646 hlen
= off
+ sizeof(*eth
);
3647 eth
= skb_gro_header_fast(skb
, off
);
3648 if (skb_gro_header_hard(skb
, hlen
)) {
3649 eth
= skb_gro_header_slow(skb
, hlen
, off
);
3650 if (unlikely(!eth
)) {
3651 napi_reuse_skb(napi
, skb
);
3657 skb_gro_pull(skb
, sizeof(*eth
));
3660 * This works because the only protocols we care about don't require
3661 * special handling. We'll fix it up properly at the end.
3663 skb
->protocol
= eth
->h_proto
;
3668 EXPORT_SYMBOL(napi_frags_skb
);
3670 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
3672 struct sk_buff
*skb
= napi_frags_skb(napi
);
3677 return napi_frags_finish(napi
, skb
, __napi_gro_receive(napi
, skb
));
3679 EXPORT_SYMBOL(napi_gro_frags
);
3682 * net_rps_action sends any pending IPI's for rps.
3683 * Note: called with local irq disabled, but exits with local irq enabled.
3685 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
3688 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
3691 sd
->rps_ipi_list
= NULL
;
3695 /* Send pending IPI's to kick RPS processing on remote cpus. */
3697 struct softnet_data
*next
= remsd
->rps_ipi_next
;
3699 if (cpu_online(remsd
->cpu
))
3700 __smp_call_function_single(remsd
->cpu
,
3709 static int process_backlog(struct napi_struct
*napi
, int quota
)
3712 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
3715 /* Check if we have pending ipi, its better to send them now,
3716 * not waiting net_rx_action() end.
3718 if (sd
->rps_ipi_list
) {
3719 local_irq_disable();
3720 net_rps_action_and_irq_enable(sd
);
3723 napi
->weight
= weight_p
;
3724 local_irq_disable();
3725 while (work
< quota
) {
3726 struct sk_buff
*skb
;
3729 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
3731 __netif_receive_skb(skb
);
3732 local_irq_disable();
3733 input_queue_head_incr(sd
);
3734 if (++work
>= quota
) {
3741 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3743 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
3744 &sd
->process_queue
);
3746 if (qlen
< quota
- work
) {
3748 * Inline a custom version of __napi_complete().
3749 * only current cpu owns and manipulates this napi,
3750 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3751 * we can use a plain write instead of clear_bit(),
3752 * and we dont need an smp_mb() memory barrier.
3754 list_del(&napi
->poll_list
);
3757 quota
= work
+ qlen
;
3767 * __napi_schedule - schedule for receive
3768 * @n: entry to schedule
3770 * The entry's receive function will be scheduled to run
3772 void __napi_schedule(struct napi_struct
*n
)
3774 unsigned long flags
;
3776 local_irq_save(flags
);
3777 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
3778 local_irq_restore(flags
);
3780 EXPORT_SYMBOL(__napi_schedule
);
3782 void __napi_complete(struct napi_struct
*n
)
3784 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
3785 BUG_ON(n
->gro_list
);
3787 list_del(&n
->poll_list
);
3788 smp_mb__before_clear_bit();
3789 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
3791 EXPORT_SYMBOL(__napi_complete
);
3793 void napi_complete(struct napi_struct
*n
)
3795 unsigned long flags
;
3798 * don't let napi dequeue from the cpu poll list
3799 * just in case its running on a different cpu
3801 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
3805 local_irq_save(flags
);
3807 local_irq_restore(flags
);
3809 EXPORT_SYMBOL(napi_complete
);
3811 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
3812 int (*poll
)(struct napi_struct
*, int), int weight
)
3814 INIT_LIST_HEAD(&napi
->poll_list
);
3815 napi
->gro_count
= 0;
3816 napi
->gro_list
= NULL
;
3819 napi
->weight
= weight
;
3820 list_add(&napi
->dev_list
, &dev
->napi_list
);
3822 #ifdef CONFIG_NETPOLL
3823 spin_lock_init(&napi
->poll_lock
);
3824 napi
->poll_owner
= -1;
3826 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
3828 EXPORT_SYMBOL(netif_napi_add
);
3830 void netif_napi_del(struct napi_struct
*napi
)
3832 struct sk_buff
*skb
, *next
;
3834 list_del_init(&napi
->dev_list
);
3835 napi_free_frags(napi
);
3837 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3843 napi
->gro_list
= NULL
;
3844 napi
->gro_count
= 0;
3846 EXPORT_SYMBOL(netif_napi_del
);
3848 static void net_rx_action(struct softirq_action
*h
)
3850 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3851 unsigned long time_limit
= jiffies
+ 2;
3852 int budget
= netdev_budget
;
3855 local_irq_disable();
3857 while (!list_empty(&sd
->poll_list
)) {
3858 struct napi_struct
*n
;
3861 /* If softirq window is exhuasted then punt.
3862 * Allow this to run for 2 jiffies since which will allow
3863 * an average latency of 1.5/HZ.
3865 if (unlikely(budget
<= 0 || time_after(jiffies
, time_limit
)))
3870 /* Even though interrupts have been re-enabled, this
3871 * access is safe because interrupts can only add new
3872 * entries to the tail of this list, and only ->poll()
3873 * calls can remove this head entry from the list.
3875 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
3877 have
= netpoll_poll_lock(n
);
3881 /* This NAPI_STATE_SCHED test is for avoiding a race
3882 * with netpoll's poll_napi(). Only the entity which
3883 * obtains the lock and sees NAPI_STATE_SCHED set will
3884 * actually make the ->poll() call. Therefore we avoid
3885 * accidentally calling ->poll() when NAPI is not scheduled.
3888 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
3889 work
= n
->poll(n
, weight
);
3893 WARN_ON_ONCE(work
> weight
);
3897 local_irq_disable();
3899 /* Drivers must not modify the NAPI state if they
3900 * consume the entire weight. In such cases this code
3901 * still "owns" the NAPI instance and therefore can
3902 * move the instance around on the list at-will.
3904 if (unlikely(work
== weight
)) {
3905 if (unlikely(napi_disable_pending(n
))) {
3908 local_irq_disable();
3910 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
3913 netpoll_poll_unlock(have
);
3916 net_rps_action_and_irq_enable(sd
);
3918 #ifdef CONFIG_NET_DMA
3920 * There may not be any more sk_buffs coming right now, so push
3921 * any pending DMA copies to hardware
3923 dma_issue_pending_all();
3930 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3934 static gifconf_func_t
*gifconf_list
[NPROTO
];
3937 * register_gifconf - register a SIOCGIF handler
3938 * @family: Address family
3939 * @gifconf: Function handler
3941 * Register protocol dependent address dumping routines. The handler
3942 * that is passed must not be freed or reused until it has been replaced
3943 * by another handler.
3945 int register_gifconf(unsigned int family
, gifconf_func_t
*gifconf
)
3947 if (family
>= NPROTO
)
3949 gifconf_list
[family
] = gifconf
;
3952 EXPORT_SYMBOL(register_gifconf
);
3956 * Map an interface index to its name (SIOCGIFNAME)
3960 * We need this ioctl for efficient implementation of the
3961 * if_indextoname() function required by the IPv6 API. Without
3962 * it, we would have to search all the interfaces to find a
3966 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
3968 struct net_device
*dev
;
3972 * Fetch the caller's info block.
3975 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
3979 dev
= dev_get_by_index_rcu(net
, ifr
.ifr_ifindex
);
3985 strcpy(ifr
.ifr_name
, dev
->name
);
3988 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
3994 * Perform a SIOCGIFCONF call. This structure will change
3995 * size eventually, and there is nothing I can do about it.
3996 * Thus we will need a 'compatibility mode'.
3999 static int dev_ifconf(struct net
*net
, char __user
*arg
)
4002 struct net_device
*dev
;
4009 * Fetch the caller's info block.
4012 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
4019 * Loop over the interfaces, and write an info block for each.
4023 for_each_netdev(net
, dev
) {
4024 for (i
= 0; i
< NPROTO
; i
++) {
4025 if (gifconf_list
[i
]) {
4028 done
= gifconf_list
[i
](dev
, NULL
, 0);
4030 done
= gifconf_list
[i
](dev
, pos
+ total
,
4040 * All done. Write the updated control block back to the caller.
4042 ifc
.ifc_len
= total
;
4045 * Both BSD and Solaris return 0 here, so we do too.
4047 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
4050 #ifdef CONFIG_PROC_FS
4052 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4054 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4055 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4056 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4058 static inline struct net_device
*dev_from_same_bucket(struct seq_file
*seq
, loff_t
*pos
)
4060 struct net
*net
= seq_file_net(seq
);
4061 struct net_device
*dev
;
4062 struct hlist_node
*p
;
4063 struct hlist_head
*h
;
4064 unsigned int count
= 0, offset
= get_offset(*pos
);
4066 h
= &net
->dev_name_head
[get_bucket(*pos
)];
4067 hlist_for_each_entry_rcu(dev
, p
, h
, name_hlist
) {
4068 if (++count
== offset
)
4075 static inline struct net_device
*dev_from_bucket(struct seq_file
*seq
, loff_t
*pos
)
4077 struct net_device
*dev
;
4078 unsigned int bucket
;
4081 dev
= dev_from_same_bucket(seq
, pos
);
4085 bucket
= get_bucket(*pos
) + 1;
4086 *pos
= set_bucket_offset(bucket
, 1);
4087 } while (bucket
< NETDEV_HASHENTRIES
);
4093 * This is invoked by the /proc filesystem handler to display a device
4096 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4101 return SEQ_START_TOKEN
;
4103 if (get_bucket(*pos
) >= NETDEV_HASHENTRIES
)
4106 return dev_from_bucket(seq
, pos
);
4109 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4112 return dev_from_bucket(seq
, pos
);
4115 void dev_seq_stop(struct seq_file
*seq
, void *v
)
4121 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
4123 struct rtnl_link_stats64 temp
;
4124 const struct rtnl_link_stats64
*stats
= dev_get_stats(dev
, &temp
);
4126 seq_printf(seq
, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4127 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4128 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
4130 stats
->rx_dropped
+ stats
->rx_missed_errors
,
4131 stats
->rx_fifo_errors
,
4132 stats
->rx_length_errors
+ stats
->rx_over_errors
+
4133 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
4134 stats
->rx_compressed
, stats
->multicast
,
4135 stats
->tx_bytes
, stats
->tx_packets
,
4136 stats
->tx_errors
, stats
->tx_dropped
,
4137 stats
->tx_fifo_errors
, stats
->collisions
,
4138 stats
->tx_carrier_errors
+
4139 stats
->tx_aborted_errors
+
4140 stats
->tx_window_errors
+
4141 stats
->tx_heartbeat_errors
,
4142 stats
->tx_compressed
);
4146 * Called from the PROCfs module. This now uses the new arbitrary sized
4147 * /proc/net interface to create /proc/net/dev
4149 static int dev_seq_show(struct seq_file
*seq
, void *v
)
4151 if (v
== SEQ_START_TOKEN
)
4152 seq_puts(seq
, "Inter-| Receive "
4154 " face |bytes packets errs drop fifo frame "
4155 "compressed multicast|bytes packets errs "
4156 "drop fifo colls carrier compressed\n");
4158 dev_seq_printf_stats(seq
, v
);
4162 static struct softnet_data
*softnet_get_online(loff_t
*pos
)
4164 struct softnet_data
*sd
= NULL
;
4166 while (*pos
< nr_cpu_ids
)
4167 if (cpu_online(*pos
)) {
4168 sd
= &per_cpu(softnet_data
, *pos
);
4175 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4177 return softnet_get_online(pos
);
4180 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4183 return softnet_get_online(pos
);
4186 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
4190 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
4192 struct softnet_data
*sd
= v
;
4194 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4195 sd
->processed
, sd
->dropped
, sd
->time_squeeze
, 0,
4196 0, 0, 0, 0, /* was fastroute */
4197 sd
->cpu_collision
, sd
->received_rps
);
4201 static const struct seq_operations dev_seq_ops
= {
4202 .start
= dev_seq_start
,
4203 .next
= dev_seq_next
,
4204 .stop
= dev_seq_stop
,
4205 .show
= dev_seq_show
,
4208 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
4210 return seq_open_net(inode
, file
, &dev_seq_ops
,
4211 sizeof(struct seq_net_private
));
4214 static const struct file_operations dev_seq_fops
= {
4215 .owner
= THIS_MODULE
,
4216 .open
= dev_seq_open
,
4218 .llseek
= seq_lseek
,
4219 .release
= seq_release_net
,
4222 static const struct seq_operations softnet_seq_ops
= {
4223 .start
= softnet_seq_start
,
4224 .next
= softnet_seq_next
,
4225 .stop
= softnet_seq_stop
,
4226 .show
= softnet_seq_show
,
4229 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
4231 return seq_open(file
, &softnet_seq_ops
);
4234 static const struct file_operations softnet_seq_fops
= {
4235 .owner
= THIS_MODULE
,
4236 .open
= softnet_seq_open
,
4238 .llseek
= seq_lseek
,
4239 .release
= seq_release
,
4242 static void *ptype_get_idx(loff_t pos
)
4244 struct packet_type
*pt
= NULL
;
4248 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
4254 for (t
= 0; t
< PTYPE_HASH_SIZE
; t
++) {
4255 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
4264 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4268 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
4271 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4273 struct packet_type
*pt
;
4274 struct list_head
*nxt
;
4278 if (v
== SEQ_START_TOKEN
)
4279 return ptype_get_idx(0);
4282 nxt
= pt
->list
.next
;
4283 if (pt
->type
== htons(ETH_P_ALL
)) {
4284 if (nxt
!= &ptype_all
)
4287 nxt
= ptype_base
[0].next
;
4289 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
4291 while (nxt
== &ptype_base
[hash
]) {
4292 if (++hash
>= PTYPE_HASH_SIZE
)
4294 nxt
= ptype_base
[hash
].next
;
4297 return list_entry(nxt
, struct packet_type
, list
);
4300 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
4306 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
4308 struct packet_type
*pt
= v
;
4310 if (v
== SEQ_START_TOKEN
)
4311 seq_puts(seq
, "Type Device Function\n");
4312 else if (pt
->dev
== NULL
|| dev_net(pt
->dev
) == seq_file_net(seq
)) {
4313 if (pt
->type
== htons(ETH_P_ALL
))
4314 seq_puts(seq
, "ALL ");
4316 seq_printf(seq
, "%04x", ntohs(pt
->type
));
4318 seq_printf(seq
, " %-8s %pF\n",
4319 pt
->dev
? pt
->dev
->name
: "", pt
->func
);
4325 static const struct seq_operations ptype_seq_ops
= {
4326 .start
= ptype_seq_start
,
4327 .next
= ptype_seq_next
,
4328 .stop
= ptype_seq_stop
,
4329 .show
= ptype_seq_show
,
4332 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
4334 return seq_open_net(inode
, file
, &ptype_seq_ops
,
4335 sizeof(struct seq_net_private
));
4338 static const struct file_operations ptype_seq_fops
= {
4339 .owner
= THIS_MODULE
,
4340 .open
= ptype_seq_open
,
4342 .llseek
= seq_lseek
,
4343 .release
= seq_release_net
,
4347 static int __net_init
dev_proc_net_init(struct net
*net
)
4351 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
4353 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
4355 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
4358 if (wext_proc_init(net
))
4364 proc_net_remove(net
, "ptype");
4366 proc_net_remove(net
, "softnet_stat");
4368 proc_net_remove(net
, "dev");
4372 static void __net_exit
dev_proc_net_exit(struct net
*net
)
4374 wext_proc_exit(net
);
4376 proc_net_remove(net
, "ptype");
4377 proc_net_remove(net
, "softnet_stat");
4378 proc_net_remove(net
, "dev");
4381 static struct pernet_operations __net_initdata dev_proc_ops
= {
4382 .init
= dev_proc_net_init
,
4383 .exit
= dev_proc_net_exit
,
4386 static int __init
dev_proc_init(void)
4388 return register_pernet_subsys(&dev_proc_ops
);
4391 #define dev_proc_init() 0
4392 #endif /* CONFIG_PROC_FS */
4396 * netdev_set_master - set up master pointer
4397 * @slave: slave device
4398 * @master: new master device
4400 * Changes the master device of the slave. Pass %NULL to break the
4401 * bonding. The caller must hold the RTNL semaphore. On a failure
4402 * a negative errno code is returned. On success the reference counts
4403 * are adjusted and the function returns zero.
4405 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
4407 struct net_device
*old
= slave
->master
;
4417 slave
->master
= master
;
4423 EXPORT_SYMBOL(netdev_set_master
);
4426 * netdev_set_bond_master - set up bonding master/slave pair
4427 * @slave: slave device
4428 * @master: new master device
4430 * Changes the master device of the slave. Pass %NULL to break the
4431 * bonding. The caller must hold the RTNL semaphore. On a failure
4432 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4433 * to the routing socket and the function returns zero.
4435 int netdev_set_bond_master(struct net_device
*slave
, struct net_device
*master
)
4441 err
= netdev_set_master(slave
, master
);
4445 slave
->flags
|= IFF_SLAVE
;
4447 slave
->flags
&= ~IFF_SLAVE
;
4449 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
4452 EXPORT_SYMBOL(netdev_set_bond_master
);
4454 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
4456 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4458 if ((dev
->flags
& IFF_UP
) && ops
->ndo_change_rx_flags
)
4459 ops
->ndo_change_rx_flags(dev
, flags
);
4462 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
4464 unsigned int old_flags
= dev
->flags
;
4470 dev
->flags
|= IFF_PROMISC
;
4471 dev
->promiscuity
+= inc
;
4472 if (dev
->promiscuity
== 0) {
4475 * If inc causes overflow, untouch promisc and return error.
4478 dev
->flags
&= ~IFF_PROMISC
;
4480 dev
->promiscuity
-= inc
;
4481 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4486 if (dev
->flags
!= old_flags
) {
4487 pr_info("device %s %s promiscuous mode\n",
4489 dev
->flags
& IFF_PROMISC
? "entered" : "left");
4490 if (audit_enabled
) {
4491 current_uid_gid(&uid
, &gid
);
4492 audit_log(current
->audit_context
, GFP_ATOMIC
,
4493 AUDIT_ANOM_PROMISCUOUS
,
4494 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4495 dev
->name
, (dev
->flags
& IFF_PROMISC
),
4496 (old_flags
& IFF_PROMISC
),
4497 audit_get_loginuid(current
),
4499 audit_get_sessionid(current
));
4502 dev_change_rx_flags(dev
, IFF_PROMISC
);
4508 * dev_set_promiscuity - update promiscuity count on a device
4512 * Add or remove promiscuity from a device. While the count in the device
4513 * remains above zero the interface remains promiscuous. Once it hits zero
4514 * the device reverts back to normal filtering operation. A negative inc
4515 * value is used to drop promiscuity on the device.
4516 * Return 0 if successful or a negative errno code on error.
4518 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
4520 unsigned int old_flags
= dev
->flags
;
4523 err
= __dev_set_promiscuity(dev
, inc
);
4526 if (dev
->flags
!= old_flags
)
4527 dev_set_rx_mode(dev
);
4530 EXPORT_SYMBOL(dev_set_promiscuity
);
4533 * dev_set_allmulti - update allmulti count on a device
4537 * Add or remove reception of all multicast frames to a device. While the
4538 * count in the device remains above zero the interface remains listening
4539 * to all interfaces. Once it hits zero the device reverts back to normal
4540 * filtering operation. A negative @inc value is used to drop the counter
4541 * when releasing a resource needing all multicasts.
4542 * Return 0 if successful or a negative errno code on error.
4545 int dev_set_allmulti(struct net_device
*dev
, int inc
)
4547 unsigned int old_flags
= dev
->flags
;
4551 dev
->flags
|= IFF_ALLMULTI
;
4552 dev
->allmulti
+= inc
;
4553 if (dev
->allmulti
== 0) {
4556 * If inc causes overflow, untouch allmulti and return error.
4559 dev
->flags
&= ~IFF_ALLMULTI
;
4561 dev
->allmulti
-= inc
;
4562 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4567 if (dev
->flags
^ old_flags
) {
4568 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
4569 dev_set_rx_mode(dev
);
4573 EXPORT_SYMBOL(dev_set_allmulti
);
4576 * Upload unicast and multicast address lists to device and
4577 * configure RX filtering. When the device doesn't support unicast
4578 * filtering it is put in promiscuous mode while unicast addresses
4581 void __dev_set_rx_mode(struct net_device
*dev
)
4583 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4585 /* dev_open will call this function so the list will stay sane. */
4586 if (!(dev
->flags
&IFF_UP
))
4589 if (!netif_device_present(dev
))
4592 if (!(dev
->priv_flags
& IFF_UNICAST_FLT
)) {
4593 /* Unicast addresses changes may only happen under the rtnl,
4594 * therefore calling __dev_set_promiscuity here is safe.
4596 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
4597 __dev_set_promiscuity(dev
, 1);
4598 dev
->uc_promisc
= true;
4599 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
4600 __dev_set_promiscuity(dev
, -1);
4601 dev
->uc_promisc
= false;
4605 if (ops
->ndo_set_rx_mode
)
4606 ops
->ndo_set_rx_mode(dev
);
4609 void dev_set_rx_mode(struct net_device
*dev
)
4611 netif_addr_lock_bh(dev
);
4612 __dev_set_rx_mode(dev
);
4613 netif_addr_unlock_bh(dev
);
4617 * dev_get_flags - get flags reported to userspace
4620 * Get the combination of flag bits exported through APIs to userspace.
4622 unsigned int dev_get_flags(const struct net_device
*dev
)
4626 flags
= (dev
->flags
& ~(IFF_PROMISC
|
4631 (dev
->gflags
& (IFF_PROMISC
|
4634 if (netif_running(dev
)) {
4635 if (netif_oper_up(dev
))
4636 flags
|= IFF_RUNNING
;
4637 if (netif_carrier_ok(dev
))
4638 flags
|= IFF_LOWER_UP
;
4639 if (netif_dormant(dev
))
4640 flags
|= IFF_DORMANT
;
4645 EXPORT_SYMBOL(dev_get_flags
);
4647 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4649 unsigned int old_flags
= dev
->flags
;
4655 * Set the flags on our device.
4658 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
4659 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
4661 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
4665 * Load in the correct multicast list now the flags have changed.
4668 if ((old_flags
^ flags
) & IFF_MULTICAST
)
4669 dev_change_rx_flags(dev
, IFF_MULTICAST
);
4671 dev_set_rx_mode(dev
);
4674 * Have we downed the interface. We handle IFF_UP ourselves
4675 * according to user attempts to set it, rather than blindly
4680 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
4681 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
4684 dev_set_rx_mode(dev
);
4687 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
4688 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
4690 dev
->gflags
^= IFF_PROMISC
;
4691 dev_set_promiscuity(dev
, inc
);
4694 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4695 is important. Some (broken) drivers set IFF_PROMISC, when
4696 IFF_ALLMULTI is requested not asking us and not reporting.
4698 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
4699 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
4701 dev
->gflags
^= IFF_ALLMULTI
;
4702 dev_set_allmulti(dev
, inc
);
4708 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
)
4710 unsigned int changes
= dev
->flags
^ old_flags
;
4712 if (changes
& IFF_UP
) {
4713 if (dev
->flags
& IFF_UP
)
4714 call_netdevice_notifiers(NETDEV_UP
, dev
);
4716 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
4719 if (dev
->flags
& IFF_UP
&&
4720 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
)))
4721 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
4725 * dev_change_flags - change device settings
4727 * @flags: device state flags
4729 * Change settings on device based state flags. The flags are
4730 * in the userspace exported format.
4732 int dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4735 unsigned int changes
, old_flags
= dev
->flags
;
4737 ret
= __dev_change_flags(dev
, flags
);
4741 changes
= old_flags
^ dev
->flags
;
4743 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
4745 __dev_notify_flags(dev
, old_flags
);
4748 EXPORT_SYMBOL(dev_change_flags
);
4751 * dev_set_mtu - Change maximum transfer unit
4753 * @new_mtu: new transfer unit
4755 * Change the maximum transfer size of the network device.
4757 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
4759 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4762 if (new_mtu
== dev
->mtu
)
4765 /* MTU must be positive. */
4769 if (!netif_device_present(dev
))
4773 if (ops
->ndo_change_mtu
)
4774 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
4778 if (!err
&& dev
->flags
& IFF_UP
)
4779 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
4782 EXPORT_SYMBOL(dev_set_mtu
);
4785 * dev_set_group - Change group this device belongs to
4787 * @new_group: group this device should belong to
4789 void dev_set_group(struct net_device
*dev
, int new_group
)
4791 dev
->group
= new_group
;
4793 EXPORT_SYMBOL(dev_set_group
);
4796 * dev_set_mac_address - Change Media Access Control Address
4800 * Change the hardware (MAC) address of the device
4802 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
4804 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4807 if (!ops
->ndo_set_mac_address
)
4809 if (sa
->sa_family
!= dev
->type
)
4811 if (!netif_device_present(dev
))
4813 err
= ops
->ndo_set_mac_address(dev
, sa
);
4815 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4818 EXPORT_SYMBOL(dev_set_mac_address
);
4821 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4823 static int dev_ifsioc_locked(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4826 struct net_device
*dev
= dev_get_by_name_rcu(net
, ifr
->ifr_name
);
4832 case SIOCGIFFLAGS
: /* Get interface flags */
4833 ifr
->ifr_flags
= (short) dev_get_flags(dev
);
4836 case SIOCGIFMETRIC
: /* Get the metric on the interface
4837 (currently unused) */
4838 ifr
->ifr_metric
= 0;
4841 case SIOCGIFMTU
: /* Get the MTU of a device */
4842 ifr
->ifr_mtu
= dev
->mtu
;
4847 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
4849 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
4850 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4851 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
4859 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
4860 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
4861 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
4862 ifr
->ifr_map
.irq
= dev
->irq
;
4863 ifr
->ifr_map
.dma
= dev
->dma
;
4864 ifr
->ifr_map
.port
= dev
->if_port
;
4868 ifr
->ifr_ifindex
= dev
->ifindex
;
4872 ifr
->ifr_qlen
= dev
->tx_queue_len
;
4876 /* dev_ioctl() should ensure this case
4888 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4890 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4893 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
4894 const struct net_device_ops
*ops
;
4899 ops
= dev
->netdev_ops
;
4902 case SIOCSIFFLAGS
: /* Set interface flags */
4903 return dev_change_flags(dev
, ifr
->ifr_flags
);
4905 case SIOCSIFMETRIC
: /* Set the metric on the interface
4906 (currently unused) */
4909 case SIOCSIFMTU
: /* Set the MTU of a device */
4910 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
4913 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
4915 case SIOCSIFHWBROADCAST
:
4916 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
4918 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
4919 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4920 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4924 if (ops
->ndo_set_config
) {
4925 if (!netif_device_present(dev
))
4927 return ops
->ndo_set_config(dev
, &ifr
->ifr_map
);
4932 if (!ops
->ndo_set_rx_mode
||
4933 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4935 if (!netif_device_present(dev
))
4937 return dev_mc_add_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4940 if (!ops
->ndo_set_rx_mode
||
4941 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4943 if (!netif_device_present(dev
))
4945 return dev_mc_del_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4948 if (ifr
->ifr_qlen
< 0)
4950 dev
->tx_queue_len
= ifr
->ifr_qlen
;
4954 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
4955 return dev_change_name(dev
, ifr
->ifr_newname
);
4958 err
= net_hwtstamp_validate(ifr
);
4964 * Unknown or private ioctl
4967 if ((cmd
>= SIOCDEVPRIVATE
&&
4968 cmd
<= SIOCDEVPRIVATE
+ 15) ||
4969 cmd
== SIOCBONDENSLAVE
||
4970 cmd
== SIOCBONDRELEASE
||
4971 cmd
== SIOCBONDSETHWADDR
||
4972 cmd
== SIOCBONDSLAVEINFOQUERY
||
4973 cmd
== SIOCBONDINFOQUERY
||
4974 cmd
== SIOCBONDCHANGEACTIVE
||
4975 cmd
== SIOCGMIIPHY
||
4976 cmd
== SIOCGMIIREG
||
4977 cmd
== SIOCSMIIREG
||
4978 cmd
== SIOCBRADDIF
||
4979 cmd
== SIOCBRDELIF
||
4980 cmd
== SIOCSHWTSTAMP
||
4981 cmd
== SIOCWANDEV
) {
4983 if (ops
->ndo_do_ioctl
) {
4984 if (netif_device_present(dev
))
4985 err
= ops
->ndo_do_ioctl(dev
, ifr
, cmd
);
4997 * This function handles all "interface"-type I/O control requests. The actual
4998 * 'doing' part of this is dev_ifsioc above.
5002 * dev_ioctl - network device ioctl
5003 * @net: the applicable net namespace
5004 * @cmd: command to issue
5005 * @arg: pointer to a struct ifreq in user space
5007 * Issue ioctl functions to devices. This is normally called by the
5008 * user space syscall interfaces but can sometimes be useful for
5009 * other purposes. The return value is the return from the syscall if
5010 * positive or a negative errno code on error.
5013 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
5019 /* One special case: SIOCGIFCONF takes ifconf argument
5020 and requires shared lock, because it sleeps writing
5024 if (cmd
== SIOCGIFCONF
) {
5026 ret
= dev_ifconf(net
, (char __user
*) arg
);
5030 if (cmd
== SIOCGIFNAME
)
5031 return dev_ifname(net
, (struct ifreq __user
*)arg
);
5033 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
5036 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
5038 colon
= strchr(ifr
.ifr_name
, ':');
5043 * See which interface the caller is talking about.
5048 * These ioctl calls:
5049 * - can be done by all.
5050 * - atomic and do not require locking.
5061 dev_load(net
, ifr
.ifr_name
);
5063 ret
= dev_ifsioc_locked(net
, &ifr
, cmd
);
5068 if (copy_to_user(arg
, &ifr
,
5069 sizeof(struct ifreq
)))
5075 dev_load(net
, ifr
.ifr_name
);
5077 ret
= dev_ethtool(net
, &ifr
);
5082 if (copy_to_user(arg
, &ifr
,
5083 sizeof(struct ifreq
)))
5089 * These ioctl calls:
5090 * - require superuser power.
5091 * - require strict serialization.
5097 if (!capable(CAP_NET_ADMIN
))
5099 dev_load(net
, ifr
.ifr_name
);
5101 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5106 if (copy_to_user(arg
, &ifr
,
5107 sizeof(struct ifreq
)))
5113 * These ioctl calls:
5114 * - require superuser power.
5115 * - require strict serialization.
5116 * - do not return a value
5126 case SIOCSIFHWBROADCAST
:
5129 case SIOCBONDENSLAVE
:
5130 case SIOCBONDRELEASE
:
5131 case SIOCBONDSETHWADDR
:
5132 case SIOCBONDCHANGEACTIVE
:
5136 if (!capable(CAP_NET_ADMIN
))
5139 case SIOCBONDSLAVEINFOQUERY
:
5140 case SIOCBONDINFOQUERY
:
5141 dev_load(net
, ifr
.ifr_name
);
5143 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5148 /* Get the per device memory space. We can add this but
5149 * currently do not support it */
5151 /* Set the per device memory buffer space.
5152 * Not applicable in our case */
5157 * Unknown or private ioctl.
5160 if (cmd
== SIOCWANDEV
||
5161 (cmd
>= SIOCDEVPRIVATE
&&
5162 cmd
<= SIOCDEVPRIVATE
+ 15)) {
5163 dev_load(net
, ifr
.ifr_name
);
5165 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5167 if (!ret
&& copy_to_user(arg
, &ifr
,
5168 sizeof(struct ifreq
)))
5172 /* Take care of Wireless Extensions */
5173 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
5174 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
5181 * dev_new_index - allocate an ifindex
5182 * @net: the applicable net namespace
5184 * Returns a suitable unique value for a new device interface
5185 * number. The caller must hold the rtnl semaphore or the
5186 * dev_base_lock to be sure it remains unique.
5188 static int dev_new_index(struct net
*net
)
5194 if (!__dev_get_by_index(net
, ifindex
))
5199 /* Delayed registration/unregisteration */
5200 static LIST_HEAD(net_todo_list
);
5202 static void net_set_todo(struct net_device
*dev
)
5204 list_add_tail(&dev
->todo_list
, &net_todo_list
);
5207 static void rollback_registered_many(struct list_head
*head
)
5209 struct net_device
*dev
, *tmp
;
5211 BUG_ON(dev_boot_phase
);
5214 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
5215 /* Some devices call without registering
5216 * for initialization unwind. Remove those
5217 * devices and proceed with the remaining.
5219 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5220 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5224 list_del(&dev
->unreg_list
);
5227 dev
->dismantle
= true;
5228 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
5231 /* If device is running, close it first. */
5232 dev_close_many(head
);
5234 list_for_each_entry(dev
, head
, unreg_list
) {
5235 /* And unlink it from device chain. */
5236 unlist_netdevice(dev
);
5238 dev
->reg_state
= NETREG_UNREGISTERING
;
5243 list_for_each_entry(dev
, head
, unreg_list
) {
5244 /* Shutdown queueing discipline. */
5248 /* Notify protocols, that we are about to destroy
5249 this device. They should clean all the things.
5251 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5253 if (!dev
->rtnl_link_ops
||
5254 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5255 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
5258 * Flush the unicast and multicast chains
5263 if (dev
->netdev_ops
->ndo_uninit
)
5264 dev
->netdev_ops
->ndo_uninit(dev
);
5266 /* Notifier chain MUST detach us from master device. */
5267 WARN_ON(dev
->master
);
5269 /* Remove entries from kobject tree */
5270 netdev_unregister_kobject(dev
);
5273 /* Process any work delayed until the end of the batch */
5274 dev
= list_first_entry(head
, struct net_device
, unreg_list
);
5275 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
5279 list_for_each_entry(dev
, head
, unreg_list
)
5283 static void rollback_registered(struct net_device
*dev
)
5287 list_add(&dev
->unreg_list
, &single
);
5288 rollback_registered_many(&single
);
5292 static netdev_features_t
netdev_fix_features(struct net_device
*dev
,
5293 netdev_features_t features
)
5295 /* Fix illegal checksum combinations */
5296 if ((features
& NETIF_F_HW_CSUM
) &&
5297 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5298 netdev_warn(dev
, "mixed HW and IP checksum settings.\n");
5299 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5302 /* Fix illegal SG+CSUM combinations. */
5303 if ((features
& NETIF_F_SG
) &&
5304 !(features
& NETIF_F_ALL_CSUM
)) {
5306 "Dropping NETIF_F_SG since no checksum feature.\n");
5307 features
&= ~NETIF_F_SG
;
5310 /* TSO requires that SG is present as well. */
5311 if ((features
& NETIF_F_ALL_TSO
) && !(features
& NETIF_F_SG
)) {
5312 netdev_dbg(dev
, "Dropping TSO features since no SG feature.\n");
5313 features
&= ~NETIF_F_ALL_TSO
;
5316 /* TSO ECN requires that TSO is present as well. */
5317 if ((features
& NETIF_F_ALL_TSO
) == NETIF_F_TSO_ECN
)
5318 features
&= ~NETIF_F_TSO_ECN
;
5320 /* Software GSO depends on SG. */
5321 if ((features
& NETIF_F_GSO
) && !(features
& NETIF_F_SG
)) {
5322 netdev_dbg(dev
, "Dropping NETIF_F_GSO since no SG feature.\n");
5323 features
&= ~NETIF_F_GSO
;
5326 /* UFO needs SG and checksumming */
5327 if (features
& NETIF_F_UFO
) {
5328 /* maybe split UFO into V4 and V6? */
5329 if (!((features
& NETIF_F_GEN_CSUM
) ||
5330 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))
5331 == (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5333 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5334 features
&= ~NETIF_F_UFO
;
5337 if (!(features
& NETIF_F_SG
)) {
5339 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5340 features
&= ~NETIF_F_UFO
;
5347 int __netdev_update_features(struct net_device
*dev
)
5349 netdev_features_t features
;
5354 features
= netdev_get_wanted_features(dev
);
5356 if (dev
->netdev_ops
->ndo_fix_features
)
5357 features
= dev
->netdev_ops
->ndo_fix_features(dev
, features
);
5359 /* driver might be less strict about feature dependencies */
5360 features
= netdev_fix_features(dev
, features
);
5362 if (dev
->features
== features
)
5365 netdev_dbg(dev
, "Features changed: %pNF -> %pNF\n",
5366 &dev
->features
, &features
);
5368 if (dev
->netdev_ops
->ndo_set_features
)
5369 err
= dev
->netdev_ops
->ndo_set_features(dev
, features
);
5371 if (unlikely(err
< 0)) {
5373 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5374 err
, &features
, &dev
->features
);
5379 dev
->features
= features
;
5385 * netdev_update_features - recalculate device features
5386 * @dev: the device to check
5388 * Recalculate dev->features set and send notifications if it
5389 * has changed. Should be called after driver or hardware dependent
5390 * conditions might have changed that influence the features.
5392 void netdev_update_features(struct net_device
*dev
)
5394 if (__netdev_update_features(dev
))
5395 netdev_features_change(dev
);
5397 EXPORT_SYMBOL(netdev_update_features
);
5400 * netdev_change_features - recalculate device features
5401 * @dev: the device to check
5403 * Recalculate dev->features set and send notifications even
5404 * if they have not changed. Should be called instead of
5405 * netdev_update_features() if also dev->vlan_features might
5406 * have changed to allow the changes to be propagated to stacked
5409 void netdev_change_features(struct net_device
*dev
)
5411 __netdev_update_features(dev
);
5412 netdev_features_change(dev
);
5414 EXPORT_SYMBOL(netdev_change_features
);
5417 * netif_stacked_transfer_operstate - transfer operstate
5418 * @rootdev: the root or lower level device to transfer state from
5419 * @dev: the device to transfer operstate to
5421 * Transfer operational state from root to device. This is normally
5422 * called when a stacking relationship exists between the root
5423 * device and the device(a leaf device).
5425 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
5426 struct net_device
*dev
)
5428 if (rootdev
->operstate
== IF_OPER_DORMANT
)
5429 netif_dormant_on(dev
);
5431 netif_dormant_off(dev
);
5433 if (netif_carrier_ok(rootdev
)) {
5434 if (!netif_carrier_ok(dev
))
5435 netif_carrier_on(dev
);
5437 if (netif_carrier_ok(dev
))
5438 netif_carrier_off(dev
);
5441 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
5444 static int netif_alloc_rx_queues(struct net_device
*dev
)
5446 unsigned int i
, count
= dev
->num_rx_queues
;
5447 struct netdev_rx_queue
*rx
;
5451 rx
= kcalloc(count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5453 pr_err("netdev: Unable to allocate %u rx queues\n", count
);
5458 for (i
= 0; i
< count
; i
++)
5464 static void netdev_init_one_queue(struct net_device
*dev
,
5465 struct netdev_queue
*queue
, void *_unused
)
5467 /* Initialize queue lock */
5468 spin_lock_init(&queue
->_xmit_lock
);
5469 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
5470 queue
->xmit_lock_owner
= -1;
5471 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
5474 dql_init(&queue
->dql
, HZ
);
5478 static int netif_alloc_netdev_queues(struct net_device
*dev
)
5480 unsigned int count
= dev
->num_tx_queues
;
5481 struct netdev_queue
*tx
;
5485 tx
= kcalloc(count
, sizeof(struct netdev_queue
), GFP_KERNEL
);
5487 pr_err("netdev: Unable to allocate %u tx queues\n", count
);
5492 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5493 spin_lock_init(&dev
->tx_global_lock
);
5499 * register_netdevice - register a network device
5500 * @dev: device to register
5502 * Take a completed network device structure and add it to the kernel
5503 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5504 * chain. 0 is returned on success. A negative errno code is returned
5505 * on a failure to set up the device, or if the name is a duplicate.
5507 * Callers must hold the rtnl semaphore. You may want
5508 * register_netdev() instead of this.
5511 * The locking appears insufficient to guarantee two parallel registers
5512 * will not get the same name.
5515 int register_netdevice(struct net_device
*dev
)
5518 struct net
*net
= dev_net(dev
);
5520 BUG_ON(dev_boot_phase
);
5525 /* When net_device's are persistent, this will be fatal. */
5526 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
5529 spin_lock_init(&dev
->addr_list_lock
);
5530 netdev_set_addr_lockdep_class(dev
);
5534 ret
= dev_get_valid_name(dev
, dev
->name
);
5538 /* Init, if this function is available */
5539 if (dev
->netdev_ops
->ndo_init
) {
5540 ret
= dev
->netdev_ops
->ndo_init(dev
);
5548 dev
->ifindex
= dev_new_index(net
);
5549 if (dev
->iflink
== -1)
5550 dev
->iflink
= dev
->ifindex
;
5552 /* Transfer changeable features to wanted_features and enable
5553 * software offloads (GSO and GRO).
5555 dev
->hw_features
|= NETIF_F_SOFT_FEATURES
;
5556 dev
->features
|= NETIF_F_SOFT_FEATURES
;
5557 dev
->wanted_features
= dev
->features
& dev
->hw_features
;
5559 /* Turn on no cache copy if HW is doing checksum */
5560 if (!(dev
->flags
& IFF_LOOPBACK
)) {
5561 dev
->hw_features
|= NETIF_F_NOCACHE_COPY
;
5562 if (dev
->features
& NETIF_F_ALL_CSUM
) {
5563 dev
->wanted_features
|= NETIF_F_NOCACHE_COPY
;
5564 dev
->features
|= NETIF_F_NOCACHE_COPY
;
5568 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5570 dev
->vlan_features
|= NETIF_F_HIGHDMA
;
5572 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
5573 ret
= notifier_to_errno(ret
);
5577 ret
= netdev_register_kobject(dev
);
5580 dev
->reg_state
= NETREG_REGISTERED
;
5582 __netdev_update_features(dev
);
5585 * Default initial state at registry is that the
5586 * device is present.
5589 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5591 dev_init_scheduler(dev
);
5593 list_netdevice(dev
);
5595 /* Notify protocols, that a new device appeared. */
5596 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5597 ret
= notifier_to_errno(ret
);
5599 rollback_registered(dev
);
5600 dev
->reg_state
= NETREG_UNREGISTERED
;
5603 * Prevent userspace races by waiting until the network
5604 * device is fully setup before sending notifications.
5606 if (!dev
->rtnl_link_ops
||
5607 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5608 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5614 if (dev
->netdev_ops
->ndo_uninit
)
5615 dev
->netdev_ops
->ndo_uninit(dev
);
5618 EXPORT_SYMBOL(register_netdevice
);
5621 * init_dummy_netdev - init a dummy network device for NAPI
5622 * @dev: device to init
5624 * This takes a network device structure and initialize the minimum
5625 * amount of fields so it can be used to schedule NAPI polls without
5626 * registering a full blown interface. This is to be used by drivers
5627 * that need to tie several hardware interfaces to a single NAPI
5628 * poll scheduler due to HW limitations.
5630 int init_dummy_netdev(struct net_device
*dev
)
5632 /* Clear everything. Note we don't initialize spinlocks
5633 * are they aren't supposed to be taken by any of the
5634 * NAPI code and this dummy netdev is supposed to be
5635 * only ever used for NAPI polls
5637 memset(dev
, 0, sizeof(struct net_device
));
5639 /* make sure we BUG if trying to hit standard
5640 * register/unregister code path
5642 dev
->reg_state
= NETREG_DUMMY
;
5644 /* NAPI wants this */
5645 INIT_LIST_HEAD(&dev
->napi_list
);
5647 /* a dummy interface is started by default */
5648 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5649 set_bit(__LINK_STATE_START
, &dev
->state
);
5651 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5652 * because users of this 'device' dont need to change
5658 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
5662 * register_netdev - register a network device
5663 * @dev: device to register
5665 * Take a completed network device structure and add it to the kernel
5666 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5667 * chain. 0 is returned on success. A negative errno code is returned
5668 * on a failure to set up the device, or if the name is a duplicate.
5670 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5671 * and expands the device name if you passed a format string to
5674 int register_netdev(struct net_device
*dev
)
5679 err
= register_netdevice(dev
);
5683 EXPORT_SYMBOL(register_netdev
);
5685 int netdev_refcnt_read(const struct net_device
*dev
)
5689 for_each_possible_cpu(i
)
5690 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
5693 EXPORT_SYMBOL(netdev_refcnt_read
);
5696 * netdev_wait_allrefs - wait until all references are gone.
5698 * This is called when unregistering network devices.
5700 * Any protocol or device that holds a reference should register
5701 * for netdevice notification, and cleanup and put back the
5702 * reference if they receive an UNREGISTER event.
5703 * We can get stuck here if buggy protocols don't correctly
5706 static void netdev_wait_allrefs(struct net_device
*dev
)
5708 unsigned long rebroadcast_time
, warning_time
;
5711 linkwatch_forget_dev(dev
);
5713 rebroadcast_time
= warning_time
= jiffies
;
5714 refcnt
= netdev_refcnt_read(dev
);
5716 while (refcnt
!= 0) {
5717 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
5720 /* Rebroadcast unregister notification */
5721 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5722 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5723 * should have already handle it the first time */
5725 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
5727 /* We must not have linkwatch events
5728 * pending on unregister. If this
5729 * happens, we simply run the queue
5730 * unscheduled, resulting in a noop
5733 linkwatch_run_queue();
5738 rebroadcast_time
= jiffies
;
5743 refcnt
= netdev_refcnt_read(dev
);
5745 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
5746 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5748 warning_time
= jiffies
;
5757 * register_netdevice(x1);
5758 * register_netdevice(x2);
5760 * unregister_netdevice(y1);
5761 * unregister_netdevice(y2);
5767 * We are invoked by rtnl_unlock().
5768 * This allows us to deal with problems:
5769 * 1) We can delete sysfs objects which invoke hotplug
5770 * without deadlocking with linkwatch via keventd.
5771 * 2) Since we run with the RTNL semaphore not held, we can sleep
5772 * safely in order to wait for the netdev refcnt to drop to zero.
5774 * We must not return until all unregister events added during
5775 * the interval the lock was held have been completed.
5777 void netdev_run_todo(void)
5779 struct list_head list
;
5781 /* Snapshot list, allow later requests */
5782 list_replace_init(&net_todo_list
, &list
);
5786 /* Wait for rcu callbacks to finish before attempting to drain
5787 * the device list. This usually avoids a 250ms wait.
5789 if (!list_empty(&list
))
5792 while (!list_empty(&list
)) {
5793 struct net_device
*dev
5794 = list_first_entry(&list
, struct net_device
, todo_list
);
5795 list_del(&dev
->todo_list
);
5797 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
5798 pr_err("network todo '%s' but state %d\n",
5799 dev
->name
, dev
->reg_state
);
5804 dev
->reg_state
= NETREG_UNREGISTERED
;
5806 on_each_cpu(flush_backlog
, dev
, 1);
5808 netdev_wait_allrefs(dev
);
5811 BUG_ON(netdev_refcnt_read(dev
));
5812 WARN_ON(rcu_access_pointer(dev
->ip_ptr
));
5813 WARN_ON(rcu_access_pointer(dev
->ip6_ptr
));
5814 WARN_ON(dev
->dn_ptr
);
5816 if (dev
->destructor
)
5817 dev
->destructor(dev
);
5819 /* Free network device */
5820 kobject_put(&dev
->dev
.kobj
);
5824 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5825 * fields in the same order, with only the type differing.
5827 void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
5828 const struct net_device_stats
*netdev_stats
)
5830 #if BITS_PER_LONG == 64
5831 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
5832 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
5834 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
5835 const unsigned long *src
= (const unsigned long *)netdev_stats
;
5836 u64
*dst
= (u64
*)stats64
;
5838 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
5839 sizeof(*stats64
) / sizeof(u64
));
5840 for (i
= 0; i
< n
; i
++)
5844 EXPORT_SYMBOL(netdev_stats_to_stats64
);
5847 * dev_get_stats - get network device statistics
5848 * @dev: device to get statistics from
5849 * @storage: place to store stats
5851 * Get network statistics from device. Return @storage.
5852 * The device driver may provide its own method by setting
5853 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5854 * otherwise the internal statistics structure is used.
5856 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
5857 struct rtnl_link_stats64
*storage
)
5859 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5861 if (ops
->ndo_get_stats64
) {
5862 memset(storage
, 0, sizeof(*storage
));
5863 ops
->ndo_get_stats64(dev
, storage
);
5864 } else if (ops
->ndo_get_stats
) {
5865 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
5867 netdev_stats_to_stats64(storage
, &dev
->stats
);
5869 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
5872 EXPORT_SYMBOL(dev_get_stats
);
5874 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
5876 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
5878 #ifdef CONFIG_NET_CLS_ACT
5881 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
5884 netdev_init_one_queue(dev
, queue
, NULL
);
5885 queue
->qdisc
= &noop_qdisc
;
5886 queue
->qdisc_sleeping
= &noop_qdisc
;
5887 rcu_assign_pointer(dev
->ingress_queue
, queue
);
5893 * alloc_netdev_mqs - allocate network device
5894 * @sizeof_priv: size of private data to allocate space for
5895 * @name: device name format string
5896 * @setup: callback to initialize device
5897 * @txqs: the number of TX subqueues to allocate
5898 * @rxqs: the number of RX subqueues to allocate
5900 * Allocates a struct net_device with private data area for driver use
5901 * and performs basic initialization. Also allocates subquue structs
5902 * for each queue on the device.
5904 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
5905 void (*setup
)(struct net_device
*),
5906 unsigned int txqs
, unsigned int rxqs
)
5908 struct net_device
*dev
;
5910 struct net_device
*p
;
5912 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
5915 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
5921 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
5926 alloc_size
= sizeof(struct net_device
);
5928 /* ensure 32-byte alignment of private area */
5929 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
5930 alloc_size
+= sizeof_priv
;
5932 /* ensure 32-byte alignment of whole construct */
5933 alloc_size
+= NETDEV_ALIGN
- 1;
5935 p
= kzalloc(alloc_size
, GFP_KERNEL
);
5937 pr_err("alloc_netdev: Unable to allocate device\n");
5941 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
5942 dev
->padded
= (char *)dev
- (char *)p
;
5944 dev
->pcpu_refcnt
= alloc_percpu(int);
5945 if (!dev
->pcpu_refcnt
)
5948 if (dev_addr_init(dev
))
5954 dev_net_set(dev
, &init_net
);
5956 dev
->gso_max_size
= GSO_MAX_SIZE
;
5958 INIT_LIST_HEAD(&dev
->napi_list
);
5959 INIT_LIST_HEAD(&dev
->unreg_list
);
5960 INIT_LIST_HEAD(&dev
->link_watch_list
);
5961 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
5964 dev
->num_tx_queues
= txqs
;
5965 dev
->real_num_tx_queues
= txqs
;
5966 if (netif_alloc_netdev_queues(dev
))
5970 dev
->num_rx_queues
= rxqs
;
5971 dev
->real_num_rx_queues
= rxqs
;
5972 if (netif_alloc_rx_queues(dev
))
5976 strcpy(dev
->name
, name
);
5977 dev
->group
= INIT_NETDEV_GROUP
;
5985 free_percpu(dev
->pcpu_refcnt
);
5995 EXPORT_SYMBOL(alloc_netdev_mqs
);
5998 * free_netdev - free network device
6001 * This function does the last stage of destroying an allocated device
6002 * interface. The reference to the device object is released.
6003 * If this is the last reference then it will be freed.
6005 void free_netdev(struct net_device
*dev
)
6007 struct napi_struct
*p
, *n
;
6009 release_net(dev_net(dev
));
6016 kfree(rcu_dereference_protected(dev
->ingress_queue
, 1));
6018 /* Flush device addresses */
6019 dev_addr_flush(dev
);
6021 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
6024 free_percpu(dev
->pcpu_refcnt
);
6025 dev
->pcpu_refcnt
= NULL
;
6027 /* Compatibility with error handling in drivers */
6028 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
6029 kfree((char *)dev
- dev
->padded
);
6033 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
6034 dev
->reg_state
= NETREG_RELEASED
;
6036 /* will free via device release */
6037 put_device(&dev
->dev
);
6039 EXPORT_SYMBOL(free_netdev
);
6042 * synchronize_net - Synchronize with packet receive processing
6044 * Wait for packets currently being received to be done.
6045 * Does not block later packets from starting.
6047 void synchronize_net(void)
6050 if (rtnl_is_locked())
6051 synchronize_rcu_expedited();
6055 EXPORT_SYMBOL(synchronize_net
);
6058 * unregister_netdevice_queue - remove device from the kernel
6062 * This function shuts down a device interface and removes it
6063 * from the kernel tables.
6064 * If head not NULL, device is queued to be unregistered later.
6066 * Callers must hold the rtnl semaphore. You may want
6067 * unregister_netdev() instead of this.
6070 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
6075 list_move_tail(&dev
->unreg_list
, head
);
6077 rollback_registered(dev
);
6078 /* Finish processing unregister after unlock */
6082 EXPORT_SYMBOL(unregister_netdevice_queue
);
6085 * unregister_netdevice_many - unregister many devices
6086 * @head: list of devices
6088 void unregister_netdevice_many(struct list_head
*head
)
6090 struct net_device
*dev
;
6092 if (!list_empty(head
)) {
6093 rollback_registered_many(head
);
6094 list_for_each_entry(dev
, head
, unreg_list
)
6098 EXPORT_SYMBOL(unregister_netdevice_many
);
6101 * unregister_netdev - remove device from the kernel
6104 * This function shuts down a device interface and removes it
6105 * from the kernel tables.
6107 * This is just a wrapper for unregister_netdevice that takes
6108 * the rtnl semaphore. In general you want to use this and not
6109 * unregister_netdevice.
6111 void unregister_netdev(struct net_device
*dev
)
6114 unregister_netdevice(dev
);
6117 EXPORT_SYMBOL(unregister_netdev
);
6120 * dev_change_net_namespace - move device to different nethost namespace
6122 * @net: network namespace
6123 * @pat: If not NULL name pattern to try if the current device name
6124 * is already taken in the destination network namespace.
6126 * This function shuts down a device interface and moves it
6127 * to a new network namespace. On success 0 is returned, on
6128 * a failure a netagive errno code is returned.
6130 * Callers must hold the rtnl semaphore.
6133 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
6139 /* Don't allow namespace local devices to be moved. */
6141 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6144 /* Ensure the device has been registrered */
6146 if (dev
->reg_state
!= NETREG_REGISTERED
)
6149 /* Get out if there is nothing todo */
6151 if (net_eq(dev_net(dev
), net
))
6154 /* Pick the destination device name, and ensure
6155 * we can use it in the destination network namespace.
6158 if (__dev_get_by_name(net
, dev
->name
)) {
6159 /* We get here if we can't use the current device name */
6162 if (dev_get_valid_name(dev
, pat
) < 0)
6167 * And now a mini version of register_netdevice unregister_netdevice.
6170 /* If device is running close it first. */
6173 /* And unlink it from device chain */
6175 unlist_netdevice(dev
);
6179 /* Shutdown queueing discipline. */
6182 /* Notify protocols, that we are about to destroy
6183 this device. They should clean all the things.
6185 Note that dev->reg_state stays at NETREG_REGISTERED.
6186 This is wanted because this way 8021q and macvlan know
6187 the device is just moving and can keep their slaves up.
6189 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6190 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
6191 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
6194 * Flush the unicast and multicast chains
6199 /* Actually switch the network namespace */
6200 dev_net_set(dev
, net
);
6202 /* If there is an ifindex conflict assign a new one */
6203 if (__dev_get_by_index(net
, dev
->ifindex
)) {
6204 int iflink
= (dev
->iflink
== dev
->ifindex
);
6205 dev
->ifindex
= dev_new_index(net
);
6207 dev
->iflink
= dev
->ifindex
;
6210 /* Fixup kobjects */
6211 err
= device_rename(&dev
->dev
, dev
->name
);
6214 /* Add the device back in the hashes */
6215 list_netdevice(dev
);
6217 /* Notify protocols, that a new device appeared. */
6218 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
6221 * Prevent userspace races by waiting until the network
6222 * device is fully setup before sending notifications.
6224 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
6231 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
6233 static int dev_cpu_callback(struct notifier_block
*nfb
,
6234 unsigned long action
,
6237 struct sk_buff
**list_skb
;
6238 struct sk_buff
*skb
;
6239 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
6240 struct softnet_data
*sd
, *oldsd
;
6242 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
6245 local_irq_disable();
6246 cpu
= smp_processor_id();
6247 sd
= &per_cpu(softnet_data
, cpu
);
6248 oldsd
= &per_cpu(softnet_data
, oldcpu
);
6250 /* Find end of our completion_queue. */
6251 list_skb
= &sd
->completion_queue
;
6253 list_skb
= &(*list_skb
)->next
;
6254 /* Append completion queue from offline CPU. */
6255 *list_skb
= oldsd
->completion_queue
;
6256 oldsd
->completion_queue
= NULL
;
6258 /* Append output queue from offline CPU. */
6259 if (oldsd
->output_queue
) {
6260 *sd
->output_queue_tailp
= oldsd
->output_queue
;
6261 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
6262 oldsd
->output_queue
= NULL
;
6263 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
6265 /* Append NAPI poll list from offline CPU. */
6266 if (!list_empty(&oldsd
->poll_list
)) {
6267 list_splice_init(&oldsd
->poll_list
, &sd
->poll_list
);
6268 raise_softirq_irqoff(NET_RX_SOFTIRQ
);
6271 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
6274 /* Process offline CPU's input_pkt_queue */
6275 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
6277 input_queue_head_incr(oldsd
);
6279 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
6281 input_queue_head_incr(oldsd
);
6289 * netdev_increment_features - increment feature set by one
6290 * @all: current feature set
6291 * @one: new feature set
6292 * @mask: mask feature set
6294 * Computes a new feature set after adding a device with feature set
6295 * @one to the master device with current feature set @all. Will not
6296 * enable anything that is off in @mask. Returns the new feature set.
6298 netdev_features_t
netdev_increment_features(netdev_features_t all
,
6299 netdev_features_t one
, netdev_features_t mask
)
6301 if (mask
& NETIF_F_GEN_CSUM
)
6302 mask
|= NETIF_F_ALL_CSUM
;
6303 mask
|= NETIF_F_VLAN_CHALLENGED
;
6305 all
|= one
& (NETIF_F_ONE_FOR_ALL
|NETIF_F_ALL_CSUM
) & mask
;
6306 all
&= one
| ~NETIF_F_ALL_FOR_ALL
;
6308 /* If one device supports hw checksumming, set for all. */
6309 if (all
& NETIF_F_GEN_CSUM
)
6310 all
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_GEN_CSUM
);
6314 EXPORT_SYMBOL(netdev_increment_features
);
6316 static struct hlist_head
*netdev_create_hash(void)
6319 struct hlist_head
*hash
;
6321 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
6323 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
6324 INIT_HLIST_HEAD(&hash
[i
]);
6329 /* Initialize per network namespace state */
6330 static int __net_init
netdev_init(struct net
*net
)
6332 INIT_LIST_HEAD(&net
->dev_base_head
);
6334 net
->dev_name_head
= netdev_create_hash();
6335 if (net
->dev_name_head
== NULL
)
6338 net
->dev_index_head
= netdev_create_hash();
6339 if (net
->dev_index_head
== NULL
)
6345 kfree(net
->dev_name_head
);
6351 * netdev_drivername - network driver for the device
6352 * @dev: network device
6354 * Determine network driver for device.
6356 const char *netdev_drivername(const struct net_device
*dev
)
6358 const struct device_driver
*driver
;
6359 const struct device
*parent
;
6360 const char *empty
= "";
6362 parent
= dev
->dev
.parent
;
6366 driver
= parent
->driver
;
6367 if (driver
&& driver
->name
)
6368 return driver
->name
;
6372 int __netdev_printk(const char *level
, const struct net_device
*dev
,
6373 struct va_format
*vaf
)
6377 if (dev
&& dev
->dev
.parent
)
6378 r
= dev_printk(level
, dev
->dev
.parent
, "%s: %pV",
6379 netdev_name(dev
), vaf
);
6381 r
= printk("%s%s: %pV", level
, netdev_name(dev
), vaf
);
6383 r
= printk("%s(NULL net_device): %pV", level
, vaf
);
6387 EXPORT_SYMBOL(__netdev_printk
);
6389 int netdev_printk(const char *level
, const struct net_device
*dev
,
6390 const char *format
, ...)
6392 struct va_format vaf
;
6396 va_start(args
, format
);
6401 r
= __netdev_printk(level
, dev
, &vaf
);
6406 EXPORT_SYMBOL(netdev_printk
);
6408 #define define_netdev_printk_level(func, level) \
6409 int func(const struct net_device *dev, const char *fmt, ...) \
6412 struct va_format vaf; \
6415 va_start(args, fmt); \
6420 r = __netdev_printk(level, dev, &vaf); \
6425 EXPORT_SYMBOL(func);
6427 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
6428 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
6429 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
6430 define_netdev_printk_level(netdev_err
, KERN_ERR
);
6431 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
6432 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
6433 define_netdev_printk_level(netdev_info
, KERN_INFO
);
6435 static void __net_exit
netdev_exit(struct net
*net
)
6437 kfree(net
->dev_name_head
);
6438 kfree(net
->dev_index_head
);
6441 static struct pernet_operations __net_initdata netdev_net_ops
= {
6442 .init
= netdev_init
,
6443 .exit
= netdev_exit
,
6446 static void __net_exit
default_device_exit(struct net
*net
)
6448 struct net_device
*dev
, *aux
;
6450 * Push all migratable network devices back to the
6451 * initial network namespace
6454 for_each_netdev_safe(net
, dev
, aux
) {
6456 char fb_name
[IFNAMSIZ
];
6458 /* Ignore unmoveable devices (i.e. loopback) */
6459 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6462 /* Leave virtual devices for the generic cleanup */
6463 if (dev
->rtnl_link_ops
)
6466 /* Push remaining network devices to init_net */
6467 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
6468 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
6470 pr_emerg("%s: failed to move %s to init_net: %d\n",
6471 __func__
, dev
->name
, err
);
6478 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
6480 /* At exit all network devices most be removed from a network
6481 * namespace. Do this in the reverse order of registration.
6482 * Do this across as many network namespaces as possible to
6483 * improve batching efficiency.
6485 struct net_device
*dev
;
6487 LIST_HEAD(dev_kill_list
);
6490 list_for_each_entry(net
, net_list
, exit_list
) {
6491 for_each_netdev_reverse(net
, dev
) {
6492 if (dev
->rtnl_link_ops
)
6493 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
6495 unregister_netdevice_queue(dev
, &dev_kill_list
);
6498 unregister_netdevice_many(&dev_kill_list
);
6499 list_del(&dev_kill_list
);
6503 static struct pernet_operations __net_initdata default_device_ops
= {
6504 .exit
= default_device_exit
,
6505 .exit_batch
= default_device_exit_batch
,
6509 * Initialize the DEV module. At boot time this walks the device list and
6510 * unhooks any devices that fail to initialise (normally hardware not
6511 * present) and leaves us with a valid list of present and active devices.
6516 * This is called single threaded during boot, so no need
6517 * to take the rtnl semaphore.
6519 static int __init
net_dev_init(void)
6521 int i
, rc
= -ENOMEM
;
6523 BUG_ON(!dev_boot_phase
);
6525 if (dev_proc_init())
6528 if (netdev_kobject_init())
6531 INIT_LIST_HEAD(&ptype_all
);
6532 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
6533 INIT_LIST_HEAD(&ptype_base
[i
]);
6535 if (register_pernet_subsys(&netdev_net_ops
))
6539 * Initialise the packet receive queues.
6542 for_each_possible_cpu(i
) {
6543 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
6545 memset(sd
, 0, sizeof(*sd
));
6546 skb_queue_head_init(&sd
->input_pkt_queue
);
6547 skb_queue_head_init(&sd
->process_queue
);
6548 sd
->completion_queue
= NULL
;
6549 INIT_LIST_HEAD(&sd
->poll_list
);
6550 sd
->output_queue
= NULL
;
6551 sd
->output_queue_tailp
= &sd
->output_queue
;
6553 sd
->csd
.func
= rps_trigger_softirq
;
6559 sd
->backlog
.poll
= process_backlog
;
6560 sd
->backlog
.weight
= weight_p
;
6561 sd
->backlog
.gro_list
= NULL
;
6562 sd
->backlog
.gro_count
= 0;
6567 /* The loopback device is special if any other network devices
6568 * is present in a network namespace the loopback device must
6569 * be present. Since we now dynamically allocate and free the
6570 * loopback device ensure this invariant is maintained by
6571 * keeping the loopback device as the first device on the
6572 * list of network devices. Ensuring the loopback devices
6573 * is the first device that appears and the last network device
6576 if (register_pernet_device(&loopback_net_ops
))
6579 if (register_pernet_device(&default_device_ops
))
6582 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
6583 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
6585 hotcpu_notifier(dev_cpu_callback
, 0);
6593 subsys_initcall(net_dev_init
);
6595 static int __init
initialize_hashrnd(void)
6597 get_random_bytes(&hashrnd
, sizeof(hashrnd
));
6601 late_initcall_sync(initialize_hashrnd
);