2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/hash.h>
83 #include <linux/slab.h>
84 #include <linux/sched.h>
85 #include <linux/mutex.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/notifier.h>
97 #include <linux/skbuff.h>
98 #include <net/net_namespace.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/stat.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include <linux/highmem.h>
109 #include <linux/init.h>
110 #include <linux/kmod.h>
111 #include <linux/module.h>
112 #include <linux/netpoll.h>
113 #include <linux/rcupdate.h>
114 #include <linux/delay.h>
115 #include <net/wext.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
126 #include <linux/ipv6.h>
127 #include <linux/in.h>
128 #include <linux/jhash.h>
129 #include <linux/random.h>
130 #include <trace/events/napi.h>
131 #include <trace/events/net.h>
132 #include <trace/events/skb.h>
133 #include <linux/pci.h>
134 #include <linux/inetdevice.h>
135 #include <linux/cpu_rmap.h>
136 #include <linux/net_tstamp.h>
137 #include <linux/static_key.h>
138 #include <net/flow_keys.h>
140 #include "net-sysfs.h"
142 /* Instead of increasing this, you should create a hash table. */
143 #define MAX_GRO_SKBS 8
145 /* This should be increased if a protocol with a bigger head is added. */
146 #define GRO_MAX_HEAD (MAX_HEADER + 128)
149 * The list of packet types we will receive (as opposed to discard)
150 * and the routines to invoke.
152 * Why 16. Because with 16 the only overlap we get on a hash of the
153 * low nibble of the protocol value is RARP/SNAP/X.25.
155 * NOTE: That is no longer true with the addition of VLAN tags. Not
156 * sure which should go first, but I bet it won't make much
157 * difference if we are running VLANs. The good news is that
158 * this protocol won't be in the list unless compiled in, so
159 * the average user (w/out VLANs) will not be adversely affected.
176 #define PTYPE_HASH_SIZE (16)
177 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
179 static DEFINE_SPINLOCK(ptype_lock
);
180 static struct list_head ptype_base
[PTYPE_HASH_SIZE
] __read_mostly
;
181 static struct list_head ptype_all __read_mostly
; /* Taps */
184 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
187 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
189 * Writers must hold the rtnl semaphore while they loop through the
190 * dev_base_head list, and hold dev_base_lock for writing when they do the
191 * actual updates. This allows pure readers to access the list even
192 * while a writer is preparing to update it.
194 * To put it another way, dev_base_lock is held for writing only to
195 * protect against pure readers; the rtnl semaphore provides the
196 * protection against other writers.
198 * See, for example usages, register_netdevice() and
199 * unregister_netdevice(), which must be called with the rtnl
202 DEFINE_RWLOCK(dev_base_lock
);
203 EXPORT_SYMBOL(dev_base_lock
);
205 static inline void dev_base_seq_inc(struct net
*net
)
207 while (++net
->dev_base_seq
== 0);
210 static inline struct hlist_head
*dev_name_hash(struct net
*net
, const char *name
)
212 unsigned 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.
1415 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1420 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1424 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1427 * call_netdevice_notifiers - call all network notifier blocks
1428 * @val: value passed unmodified to notifier function
1429 * @dev: net_device pointer passed unmodified to notifier function
1431 * Call all network notifier blocks. Parameters and return value
1432 * are as for raw_notifier_call_chain().
1435 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1438 return raw_notifier_call_chain(&netdev_chain
, val
, dev
);
1440 EXPORT_SYMBOL(call_netdevice_notifiers
);
1442 static struct static_key netstamp_needed __read_mostly
;
1443 #ifdef HAVE_JUMP_LABEL
1444 /* We are not allowed to call static_key_slow_dec() from irq context
1445 * If net_disable_timestamp() is called from irq context, defer the
1446 * static_key_slow_dec() calls.
1448 static atomic_t netstamp_needed_deferred
;
1451 void net_enable_timestamp(void)
1453 #ifdef HAVE_JUMP_LABEL
1454 int deferred
= atomic_xchg(&netstamp_needed_deferred
, 0);
1458 static_key_slow_dec(&netstamp_needed
);
1462 WARN_ON(in_interrupt());
1463 static_key_slow_inc(&netstamp_needed
);
1465 EXPORT_SYMBOL(net_enable_timestamp
);
1467 void net_disable_timestamp(void)
1469 #ifdef HAVE_JUMP_LABEL
1470 if (in_interrupt()) {
1471 atomic_inc(&netstamp_needed_deferred
);
1475 static_key_slow_dec(&netstamp_needed
);
1477 EXPORT_SYMBOL(net_disable_timestamp
);
1479 static inline void net_timestamp_set(struct sk_buff
*skb
)
1481 skb
->tstamp
.tv64
= 0;
1482 if (static_key_false(&netstamp_needed
))
1483 __net_timestamp(skb
);
1486 #define net_timestamp_check(COND, SKB) \
1487 if (static_key_false(&netstamp_needed)) { \
1488 if ((COND) && !(SKB)->tstamp.tv64) \
1489 __net_timestamp(SKB); \
1492 static int net_hwtstamp_validate(struct ifreq *ifr)
1494 struct hwtstamp_config cfg
;
1495 enum hwtstamp_tx_types tx_type
;
1496 enum hwtstamp_rx_filters rx_filter
;
1497 int tx_type_valid
= 0;
1498 int rx_filter_valid
= 0;
1500 if (copy_from_user(&cfg
, ifr
->ifr_data
, sizeof(cfg
)))
1503 if (cfg
.flags
) /* reserved for future extensions */
1506 tx_type
= cfg
.tx_type
;
1507 rx_filter
= cfg
.rx_filter
;
1510 case HWTSTAMP_TX_OFF
:
1511 case HWTSTAMP_TX_ON
:
1512 case HWTSTAMP_TX_ONESTEP_SYNC
:
1517 switch (rx_filter
) {
1518 case HWTSTAMP_FILTER_NONE
:
1519 case HWTSTAMP_FILTER_ALL
:
1520 case HWTSTAMP_FILTER_SOME
:
1521 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT
:
1522 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC
:
1523 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ
:
1524 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT
:
1525 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC
:
1526 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ
:
1527 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT
:
1528 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC
:
1529 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ
:
1530 case HWTSTAMP_FILTER_PTP_V2_EVENT
:
1531 case HWTSTAMP_FILTER_PTP_V2_SYNC
:
1532 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ
:
1533 rx_filter_valid
= 1;
1537 if (!tx_type_valid
|| !rx_filter_valid
)
1543 static inline bool is_skb_forwardable(struct net_device
*dev
,
1544 struct sk_buff
*skb
)
1548 if (!(dev
->flags
& IFF_UP
))
1551 len
= dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
;
1552 if (skb
->len
<= len
)
1555 /* if TSO is enabled, we don't care about the length as the packet
1556 * could be forwarded without being segmented before
1558 if (skb_is_gso(skb
))
1565 * dev_forward_skb - loopback an skb to another netif
1567 * @dev: destination network device
1568 * @skb: buffer to forward
1571 * NET_RX_SUCCESS (no congestion)
1572 * NET_RX_DROP (packet was dropped, but freed)
1574 * dev_forward_skb can be used for injecting an skb from the
1575 * start_xmit function of one device into the receive queue
1576 * of another device.
1578 * The receiving device may be in another namespace, so
1579 * we have to clear all information in the skb that could
1580 * impact namespace isolation.
1582 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1584 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
1585 if (skb_copy_ubufs(skb
, GFP_ATOMIC
)) {
1586 atomic_long_inc(&dev
->rx_dropped
);
1595 if (unlikely(!is_skb_forwardable(dev
, skb
))) {
1596 atomic_long_inc(&dev
->rx_dropped
);
1601 skb_set_dev(skb
, dev
);
1602 skb
->tstamp
.tv64
= 0;
1603 skb
->pkt_type
= PACKET_HOST
;
1604 skb
->protocol
= eth_type_trans(skb
, dev
);
1605 return netif_rx(skb
);
1607 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1609 static inline int deliver_skb(struct sk_buff
*skb
,
1610 struct packet_type
*pt_prev
,
1611 struct net_device
*orig_dev
)
1613 atomic_inc(&skb
->users
);
1614 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1618 * Support routine. Sends outgoing frames to any network
1619 * taps currently in use.
1622 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1624 struct packet_type
*ptype
;
1625 struct sk_buff
*skb2
= NULL
;
1626 struct packet_type
*pt_prev
= NULL
;
1629 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1630 /* Never send packets back to the socket
1631 * they originated from - MvS (miquels@drinkel.ow.org)
1633 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1634 (ptype
->af_packet_priv
== NULL
||
1635 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1637 deliver_skb(skb2
, pt_prev
, skb
->dev
);
1642 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1646 net_timestamp_set(skb2
);
1648 /* skb->nh should be correctly
1649 set by sender, so that the second statement is
1650 just protection against buggy protocols.
1652 skb_reset_mac_header(skb2
);
1654 if (skb_network_header(skb2
) < skb2
->data
||
1655 skb2
->network_header
> skb2
->tail
) {
1656 if (net_ratelimit())
1657 pr_crit("protocol %04x is buggy, dev %s\n",
1658 ntohs(skb2
->protocol
),
1660 skb_reset_network_header(skb2
);
1663 skb2
->transport_header
= skb2
->network_header
;
1664 skb2
->pkt_type
= PACKET_OUTGOING
;
1669 pt_prev
->func(skb2
, skb
->dev
, pt_prev
, skb
->dev
);
1673 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1674 * @dev: Network device
1675 * @txq: number of queues available
1677 * If real_num_tx_queues is changed the tc mappings may no longer be
1678 * valid. To resolve this verify the tc mapping remains valid and if
1679 * not NULL the mapping. With no priorities mapping to this
1680 * offset/count pair it will no longer be used. In the worst case TC0
1681 * is invalid nothing can be done so disable priority mappings. If is
1682 * expected that drivers will fix this mapping if they can before
1683 * calling netif_set_real_num_tx_queues.
1685 static void netif_setup_tc(struct net_device
*dev
, unsigned int txq
)
1688 struct netdev_tc_txq
*tc
= &dev
->tc_to_txq
[0];
1690 /* If TC0 is invalidated disable TC mapping */
1691 if (tc
->offset
+ tc
->count
> txq
) {
1692 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1697 /* Invalidated prio to tc mappings set to TC0 */
1698 for (i
= 1; i
< TC_BITMASK
+ 1; i
++) {
1699 int q
= netdev_get_prio_tc_map(dev
, i
);
1701 tc
= &dev
->tc_to_txq
[q
];
1702 if (tc
->offset
+ tc
->count
> txq
) {
1703 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1705 netdev_set_prio_tc_map(dev
, i
, 0);
1711 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1712 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1714 int netif_set_real_num_tx_queues(struct net_device
*dev
, unsigned int txq
)
1718 if (txq
< 1 || txq
> dev
->num_tx_queues
)
1721 if (dev
->reg_state
== NETREG_REGISTERED
||
1722 dev
->reg_state
== NETREG_UNREGISTERING
) {
1725 rc
= netdev_queue_update_kobjects(dev
, dev
->real_num_tx_queues
,
1731 netif_setup_tc(dev
, txq
);
1733 if (txq
< dev
->real_num_tx_queues
)
1734 qdisc_reset_all_tx_gt(dev
, txq
);
1737 dev
->real_num_tx_queues
= txq
;
1740 EXPORT_SYMBOL(netif_set_real_num_tx_queues
);
1744 * netif_set_real_num_rx_queues - set actual number of RX queues used
1745 * @dev: Network device
1746 * @rxq: Actual number of RX queues
1748 * This must be called either with the rtnl_lock held or before
1749 * registration of the net device. Returns 0 on success, or a
1750 * negative error code. If called before registration, it always
1753 int netif_set_real_num_rx_queues(struct net_device
*dev
, unsigned int rxq
)
1757 if (rxq
< 1 || rxq
> dev
->num_rx_queues
)
1760 if (dev
->reg_state
== NETREG_REGISTERED
) {
1763 rc
= net_rx_queue_update_kobjects(dev
, dev
->real_num_rx_queues
,
1769 dev
->real_num_rx_queues
= rxq
;
1772 EXPORT_SYMBOL(netif_set_real_num_rx_queues
);
1775 static inline void __netif_reschedule(struct Qdisc
*q
)
1777 struct softnet_data
*sd
;
1778 unsigned long flags
;
1780 local_irq_save(flags
);
1781 sd
= &__get_cpu_var(softnet_data
);
1782 q
->next_sched
= NULL
;
1783 *sd
->output_queue_tailp
= q
;
1784 sd
->output_queue_tailp
= &q
->next_sched
;
1785 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1786 local_irq_restore(flags
);
1789 void __netif_schedule(struct Qdisc
*q
)
1791 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
1792 __netif_reschedule(q
);
1794 EXPORT_SYMBOL(__netif_schedule
);
1796 void dev_kfree_skb_irq(struct sk_buff
*skb
)
1798 if (atomic_dec_and_test(&skb
->users
)) {
1799 struct softnet_data
*sd
;
1800 unsigned long flags
;
1802 local_irq_save(flags
);
1803 sd
= &__get_cpu_var(softnet_data
);
1804 skb
->next
= sd
->completion_queue
;
1805 sd
->completion_queue
= skb
;
1806 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1807 local_irq_restore(flags
);
1810 EXPORT_SYMBOL(dev_kfree_skb_irq
);
1812 void dev_kfree_skb_any(struct sk_buff
*skb
)
1814 if (in_irq() || irqs_disabled())
1815 dev_kfree_skb_irq(skb
);
1819 EXPORT_SYMBOL(dev_kfree_skb_any
);
1823 * netif_device_detach - mark device as removed
1824 * @dev: network device
1826 * Mark device as removed from system and therefore no longer available.
1828 void netif_device_detach(struct net_device
*dev
)
1830 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1831 netif_running(dev
)) {
1832 netif_tx_stop_all_queues(dev
);
1835 EXPORT_SYMBOL(netif_device_detach
);
1838 * netif_device_attach - mark device as attached
1839 * @dev: network device
1841 * Mark device as attached from system and restart if needed.
1843 void netif_device_attach(struct net_device
*dev
)
1845 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1846 netif_running(dev
)) {
1847 netif_tx_wake_all_queues(dev
);
1848 __netdev_watchdog_up(dev
);
1851 EXPORT_SYMBOL(netif_device_attach
);
1854 * skb_dev_set -- assign a new device to a buffer
1855 * @skb: buffer for the new device
1856 * @dev: network device
1858 * If an skb is owned by a device already, we have to reset
1859 * all data private to the namespace a device belongs to
1860 * before assigning it a new device.
1862 #ifdef CONFIG_NET_NS
1863 void skb_set_dev(struct sk_buff
*skb
, struct net_device
*dev
)
1866 if (skb
->dev
&& !net_eq(dev_net(skb
->dev
), dev_net(dev
))) {
1869 skb_init_secmark(skb
);
1873 skb
->ipvs_property
= 0;
1874 #ifdef CONFIG_NET_SCHED
1880 EXPORT_SYMBOL(skb_set_dev
);
1881 #endif /* CONFIG_NET_NS */
1883 static void skb_warn_bad_offload(const struct sk_buff
*skb
)
1885 static const netdev_features_t null_features
= 0;
1886 struct net_device
*dev
= skb
->dev
;
1887 const char *driver
= "";
1889 if (dev
&& dev
->dev
.parent
)
1890 driver
= dev_driver_string(dev
->dev
.parent
);
1892 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
1893 "gso_type=%d ip_summed=%d\n",
1894 driver
, dev
? &dev
->features
: &null_features
,
1895 skb
->sk
? &skb
->sk
->sk_route_caps
: &null_features
,
1896 skb
->len
, skb
->data_len
, skb_shinfo(skb
)->gso_size
,
1897 skb_shinfo(skb
)->gso_type
, skb
->ip_summed
);
1901 * Invalidate hardware checksum when packet is to be mangled, and
1902 * complete checksum manually on outgoing path.
1904 int skb_checksum_help(struct sk_buff
*skb
)
1907 int ret
= 0, offset
;
1909 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1910 goto out_set_summed
;
1912 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1913 skb_warn_bad_offload(skb
);
1917 offset
= skb_checksum_start_offset(skb
);
1918 BUG_ON(offset
>= skb_headlen(skb
));
1919 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1921 offset
+= skb
->csum_offset
;
1922 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
1924 if (skb_cloned(skb
) &&
1925 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
1926 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1931 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1933 skb
->ip_summed
= CHECKSUM_NONE
;
1937 EXPORT_SYMBOL(skb_checksum_help
);
1940 * skb_gso_segment - Perform segmentation on skb.
1941 * @skb: buffer to segment
1942 * @features: features for the output path (see dev->features)
1944 * This function segments the given skb and returns a list of segments.
1946 * It may return NULL if the skb requires no segmentation. This is
1947 * only possible when GSO is used for verifying header integrity.
1949 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
,
1950 netdev_features_t features
)
1952 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1953 struct packet_type
*ptype
;
1954 __be16 type
= skb
->protocol
;
1955 int vlan_depth
= ETH_HLEN
;
1958 while (type
== htons(ETH_P_8021Q
)) {
1959 struct vlan_hdr
*vh
;
1961 if (unlikely(!pskb_may_pull(skb
, vlan_depth
+ VLAN_HLEN
)))
1962 return ERR_PTR(-EINVAL
);
1964 vh
= (struct vlan_hdr
*)(skb
->data
+ vlan_depth
);
1965 type
= vh
->h_vlan_encapsulated_proto
;
1966 vlan_depth
+= VLAN_HLEN
;
1969 skb_reset_mac_header(skb
);
1970 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1971 __skb_pull(skb
, skb
->mac_len
);
1973 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1974 skb_warn_bad_offload(skb
);
1976 if (skb_header_cloned(skb
) &&
1977 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
1978 return ERR_PTR(err
);
1982 list_for_each_entry_rcu(ptype
,
1983 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
1984 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
1985 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1986 err
= ptype
->gso_send_check(skb
);
1987 segs
= ERR_PTR(err
);
1988 if (err
|| skb_gso_ok(skb
, features
))
1990 __skb_push(skb
, (skb
->data
-
1991 skb_network_header(skb
)));
1993 segs
= ptype
->gso_segment(skb
, features
);
1999 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
2003 EXPORT_SYMBOL(skb_gso_segment
);
2005 /* Take action when hardware reception checksum errors are detected. */
2007 void netdev_rx_csum_fault(struct net_device
*dev
)
2009 if (net_ratelimit()) {
2010 pr_err("%s: hw csum failure\n", dev
? dev
->name
: "<unknown>");
2014 EXPORT_SYMBOL(netdev_rx_csum_fault
);
2017 /* Actually, we should eliminate this check as soon as we know, that:
2018 * 1. IOMMU is present and allows to map all the memory.
2019 * 2. No high memory really exists on this machine.
2022 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
2024 #ifdef CONFIG_HIGHMEM
2026 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
2027 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2028 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2029 if (PageHighMem(skb_frag_page(frag
)))
2034 if (PCI_DMA_BUS_IS_PHYS
) {
2035 struct device
*pdev
= dev
->dev
.parent
;
2039 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2040 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2041 dma_addr_t addr
= page_to_phys(skb_frag_page(frag
));
2042 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
2051 void (*destructor
)(struct sk_buff
*skb
);
2054 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2056 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
2058 struct dev_gso_cb
*cb
;
2061 struct sk_buff
*nskb
= skb
->next
;
2063 skb
->next
= nskb
->next
;
2066 } while (skb
->next
);
2068 cb
= DEV_GSO_CB(skb
);
2070 cb
->destructor(skb
);
2074 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2075 * @skb: buffer to segment
2076 * @features: device features as applicable to this skb
2078 * This function segments the given skb and stores the list of segments
2081 static int dev_gso_segment(struct sk_buff
*skb
, netdev_features_t features
)
2083 struct sk_buff
*segs
;
2085 segs
= skb_gso_segment(skb
, features
);
2087 /* Verifying header integrity only. */
2092 return PTR_ERR(segs
);
2095 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
2096 skb
->destructor
= dev_gso_skb_destructor
;
2102 * Try to orphan skb early, right before transmission by the device.
2103 * We cannot orphan skb if tx timestamp is requested or the sk-reference
2104 * is needed on driver level for other reasons, e.g. see net/can/raw.c
2106 static inline void skb_orphan_try(struct sk_buff
*skb
)
2108 struct sock
*sk
= skb
->sk
;
2110 if (sk
&& !skb_shinfo(skb
)->tx_flags
) {
2111 /* skb_tx_hash() wont be able to get sk.
2112 * We copy sk_hash into skb->rxhash
2115 skb
->rxhash
= sk
->sk_hash
;
2120 static bool can_checksum_protocol(netdev_features_t features
, __be16 protocol
)
2122 return ((features
& NETIF_F_GEN_CSUM
) ||
2123 ((features
& NETIF_F_V4_CSUM
) &&
2124 protocol
== htons(ETH_P_IP
)) ||
2125 ((features
& NETIF_F_V6_CSUM
) &&
2126 protocol
== htons(ETH_P_IPV6
)) ||
2127 ((features
& NETIF_F_FCOE_CRC
) &&
2128 protocol
== htons(ETH_P_FCOE
)));
2131 static netdev_features_t
harmonize_features(struct sk_buff
*skb
,
2132 __be16 protocol
, netdev_features_t features
)
2134 if (!can_checksum_protocol(features
, protocol
)) {
2135 features
&= ~NETIF_F_ALL_CSUM
;
2136 features
&= ~NETIF_F_SG
;
2137 } else if (illegal_highdma(skb
->dev
, skb
)) {
2138 features
&= ~NETIF_F_SG
;
2144 netdev_features_t
netif_skb_features(struct sk_buff
*skb
)
2146 __be16 protocol
= skb
->protocol
;
2147 netdev_features_t features
= skb
->dev
->features
;
2149 if (protocol
== htons(ETH_P_8021Q
)) {
2150 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
2151 protocol
= veh
->h_vlan_encapsulated_proto
;
2152 } else if (!vlan_tx_tag_present(skb
)) {
2153 return harmonize_features(skb
, protocol
, features
);
2156 features
&= (skb
->dev
->vlan_features
| NETIF_F_HW_VLAN_TX
);
2158 if (protocol
!= htons(ETH_P_8021Q
)) {
2159 return harmonize_features(skb
, protocol
, features
);
2161 features
&= NETIF_F_SG
| NETIF_F_HIGHDMA
| NETIF_F_FRAGLIST
|
2162 NETIF_F_GEN_CSUM
| NETIF_F_HW_VLAN_TX
;
2163 return harmonize_features(skb
, protocol
, features
);
2166 EXPORT_SYMBOL(netif_skb_features
);
2169 * Returns true if either:
2170 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2171 * 2. skb is fragmented and the device does not support SG, or if
2172 * at least one of fragments is in highmem and device does not
2173 * support DMA from it.
2175 static inline int skb_needs_linearize(struct sk_buff
*skb
,
2178 return skb_is_nonlinear(skb
) &&
2179 ((skb_has_frag_list(skb
) &&
2180 !(features
& NETIF_F_FRAGLIST
)) ||
2181 (skb_shinfo(skb
)->nr_frags
&&
2182 !(features
& NETIF_F_SG
)));
2185 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
2186 struct netdev_queue
*txq
)
2188 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2189 int rc
= NETDEV_TX_OK
;
2190 unsigned int skb_len
;
2192 if (likely(!skb
->next
)) {
2193 netdev_features_t features
;
2196 * If device doesn't need skb->dst, release it right now while
2197 * its hot in this cpu cache
2199 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2202 if (!list_empty(&ptype_all
))
2203 dev_queue_xmit_nit(skb
, dev
);
2205 skb_orphan_try(skb
);
2207 features
= netif_skb_features(skb
);
2209 if (vlan_tx_tag_present(skb
) &&
2210 !(features
& NETIF_F_HW_VLAN_TX
)) {
2211 skb
= __vlan_put_tag(skb
, vlan_tx_tag_get(skb
));
2218 if (netif_needs_gso(skb
, features
)) {
2219 if (unlikely(dev_gso_segment(skb
, features
)))
2224 if (skb_needs_linearize(skb
, features
) &&
2225 __skb_linearize(skb
))
2228 /* If packet is not checksummed and device does not
2229 * support checksumming for this protocol, complete
2230 * checksumming here.
2232 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2233 skb_set_transport_header(skb
,
2234 skb_checksum_start_offset(skb
));
2235 if (!(features
& NETIF_F_ALL_CSUM
) &&
2236 skb_checksum_help(skb
))
2242 rc
= ops
->ndo_start_xmit(skb
, dev
);
2243 trace_net_dev_xmit(skb
, rc
, dev
, skb_len
);
2244 if (rc
== NETDEV_TX_OK
)
2245 txq_trans_update(txq
);
2251 struct sk_buff
*nskb
= skb
->next
;
2253 skb
->next
= nskb
->next
;
2257 * If device doesn't need nskb->dst, release it right now while
2258 * its hot in this cpu cache
2260 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2263 skb_len
= nskb
->len
;
2264 rc
= ops
->ndo_start_xmit(nskb
, dev
);
2265 trace_net_dev_xmit(nskb
, rc
, dev
, skb_len
);
2266 if (unlikely(rc
!= NETDEV_TX_OK
)) {
2267 if (rc
& ~NETDEV_TX_MASK
)
2268 goto out_kfree_gso_skb
;
2269 nskb
->next
= skb
->next
;
2273 txq_trans_update(txq
);
2274 if (unlikely(netif_xmit_stopped(txq
) && skb
->next
))
2275 return NETDEV_TX_BUSY
;
2276 } while (skb
->next
);
2279 if (likely(skb
->next
== NULL
))
2280 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
2287 static u32 hashrnd __read_mostly
;
2290 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2291 * to be used as a distribution range.
2293 u16
__skb_tx_hash(const struct net_device
*dev
, const struct sk_buff
*skb
,
2294 unsigned int num_tx_queues
)
2298 u16 qcount
= num_tx_queues
;
2300 if (skb_rx_queue_recorded(skb
)) {
2301 hash
= skb_get_rx_queue(skb
);
2302 while (unlikely(hash
>= num_tx_queues
))
2303 hash
-= num_tx_queues
;
2308 u8 tc
= netdev_get_prio_tc_map(dev
, skb
->priority
);
2309 qoffset
= dev
->tc_to_txq
[tc
].offset
;
2310 qcount
= dev
->tc_to_txq
[tc
].count
;
2313 if (skb
->sk
&& skb
->sk
->sk_hash
)
2314 hash
= skb
->sk
->sk_hash
;
2316 hash
= (__force u16
) skb
->protocol
^ skb
->rxhash
;
2317 hash
= jhash_1word(hash
, hashrnd
);
2319 return (u16
) (((u64
) hash
* qcount
) >> 32) + qoffset
;
2321 EXPORT_SYMBOL(__skb_tx_hash
);
2323 static inline u16
dev_cap_txqueue(struct net_device
*dev
, u16 queue_index
)
2325 if (unlikely(queue_index
>= dev
->real_num_tx_queues
)) {
2326 if (net_ratelimit()) {
2327 pr_warn("%s selects TX queue %d, but real number of TX queues is %d\n",
2328 dev
->name
, queue_index
,
2329 dev
->real_num_tx_queues
);
2336 static inline int get_xps_queue(struct net_device
*dev
, struct sk_buff
*skb
)
2339 struct xps_dev_maps
*dev_maps
;
2340 struct xps_map
*map
;
2341 int queue_index
= -1;
2344 dev_maps
= rcu_dereference(dev
->xps_maps
);
2346 map
= rcu_dereference(
2347 dev_maps
->cpu_map
[raw_smp_processor_id()]);
2350 queue_index
= map
->queues
[0];
2353 if (skb
->sk
&& skb
->sk
->sk_hash
)
2354 hash
= skb
->sk
->sk_hash
;
2356 hash
= (__force u16
) skb
->protocol
^
2358 hash
= jhash_1word(hash
, hashrnd
);
2359 queue_index
= map
->queues
[
2360 ((u64
)hash
* map
->len
) >> 32];
2362 if (unlikely(queue_index
>= dev
->real_num_tx_queues
))
2374 static struct netdev_queue
*dev_pick_tx(struct net_device
*dev
,
2375 struct sk_buff
*skb
)
2378 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2380 if (dev
->real_num_tx_queues
== 1)
2382 else if (ops
->ndo_select_queue
) {
2383 queue_index
= ops
->ndo_select_queue(dev
, skb
);
2384 queue_index
= dev_cap_txqueue(dev
, queue_index
);
2386 struct sock
*sk
= skb
->sk
;
2387 queue_index
= sk_tx_queue_get(sk
);
2389 if (queue_index
< 0 || skb
->ooo_okay
||
2390 queue_index
>= dev
->real_num_tx_queues
) {
2391 int old_index
= queue_index
;
2393 queue_index
= get_xps_queue(dev
, skb
);
2394 if (queue_index
< 0)
2395 queue_index
= skb_tx_hash(dev
, skb
);
2397 if (queue_index
!= old_index
&& sk
) {
2398 struct dst_entry
*dst
=
2399 rcu_dereference_check(sk
->sk_dst_cache
, 1);
2401 if (dst
&& skb_dst(skb
) == dst
)
2402 sk_tx_queue_set(sk
, queue_index
);
2407 skb_set_queue_mapping(skb
, queue_index
);
2408 return netdev_get_tx_queue(dev
, queue_index
);
2411 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2412 struct net_device
*dev
,
2413 struct netdev_queue
*txq
)
2415 spinlock_t
*root_lock
= qdisc_lock(q
);
2419 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
2420 qdisc_calculate_pkt_len(skb
, q
);
2422 * Heuristic to force contended enqueues to serialize on a
2423 * separate lock before trying to get qdisc main lock.
2424 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2425 * and dequeue packets faster.
2427 contended
= qdisc_is_running(q
);
2428 if (unlikely(contended
))
2429 spin_lock(&q
->busylock
);
2431 spin_lock(root_lock
);
2432 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2435 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2436 qdisc_run_begin(q
)) {
2438 * This is a work-conserving queue; there are no old skbs
2439 * waiting to be sent out; and the qdisc is not running -
2440 * xmit the skb directly.
2442 if (!(dev
->priv_flags
& IFF_XMIT_DST_RELEASE
))
2445 qdisc_bstats_update(q
, skb
);
2447 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
)) {
2448 if (unlikely(contended
)) {
2449 spin_unlock(&q
->busylock
);
2456 rc
= NET_XMIT_SUCCESS
;
2459 rc
= q
->enqueue(skb
, q
) & NET_XMIT_MASK
;
2460 if (qdisc_run_begin(q
)) {
2461 if (unlikely(contended
)) {
2462 spin_unlock(&q
->busylock
);
2468 spin_unlock(root_lock
);
2469 if (unlikely(contended
))
2470 spin_unlock(&q
->busylock
);
2474 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2475 static void skb_update_prio(struct sk_buff
*skb
)
2477 struct netprio_map
*map
= rcu_dereference_bh(skb
->dev
->priomap
);
2479 if ((!skb
->priority
) && (skb
->sk
) && map
)
2480 skb
->priority
= map
->priomap
[skb
->sk
->sk_cgrp_prioidx
];
2483 #define skb_update_prio(skb)
2486 static DEFINE_PER_CPU(int, xmit_recursion
);
2487 #define RECURSION_LIMIT 10
2490 * dev_queue_xmit - transmit a buffer
2491 * @skb: buffer to transmit
2493 * Queue a buffer for transmission to a network device. The caller must
2494 * have set the device and priority and built the buffer before calling
2495 * this function. The function can be called from an interrupt.
2497 * A negative errno code is returned on a failure. A success does not
2498 * guarantee the frame will be transmitted as it may be dropped due
2499 * to congestion or traffic shaping.
2501 * -----------------------------------------------------------------------------------
2502 * I notice this method can also return errors from the queue disciplines,
2503 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2506 * Regardless of the return value, the skb is consumed, so it is currently
2507 * difficult to retry a send to this method. (You can bump the ref count
2508 * before sending to hold a reference for retry if you are careful.)
2510 * When calling this method, interrupts MUST be enabled. This is because
2511 * the BH enable code must have IRQs enabled so that it will not deadlock.
2514 int dev_queue_xmit(struct sk_buff
*skb
)
2516 struct net_device
*dev
= skb
->dev
;
2517 struct netdev_queue
*txq
;
2521 /* Disable soft irqs for various locks below. Also
2522 * stops preemption for RCU.
2526 skb_update_prio(skb
);
2528 txq
= dev_pick_tx(dev
, skb
);
2529 q
= rcu_dereference_bh(txq
->qdisc
);
2531 #ifdef CONFIG_NET_CLS_ACT
2532 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2534 trace_net_dev_queue(skb
);
2536 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2540 /* The device has no queue. Common case for software devices:
2541 loopback, all the sorts of tunnels...
2543 Really, it is unlikely that netif_tx_lock protection is necessary
2544 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2546 However, it is possible, that they rely on protection
2549 Check this and shot the lock. It is not prone from deadlocks.
2550 Either shot noqueue qdisc, it is even simpler 8)
2552 if (dev
->flags
& IFF_UP
) {
2553 int cpu
= smp_processor_id(); /* ok because BHs are off */
2555 if (txq
->xmit_lock_owner
!= cpu
) {
2557 if (__this_cpu_read(xmit_recursion
) > RECURSION_LIMIT
)
2558 goto recursion_alert
;
2560 HARD_TX_LOCK(dev
, txq
, cpu
);
2562 if (!netif_xmit_stopped(txq
)) {
2563 __this_cpu_inc(xmit_recursion
);
2564 rc
= dev_hard_start_xmit(skb
, dev
, txq
);
2565 __this_cpu_dec(xmit_recursion
);
2566 if (dev_xmit_complete(rc
)) {
2567 HARD_TX_UNLOCK(dev
, txq
);
2571 HARD_TX_UNLOCK(dev
, txq
);
2572 if (net_ratelimit())
2573 pr_crit("Virtual device %s asks to queue packet!\n",
2576 /* Recursion is detected! It is possible,
2580 if (net_ratelimit())
2581 pr_crit("Dead loop on virtual device %s, fix it urgently!\n",
2587 rcu_read_unlock_bh();
2592 rcu_read_unlock_bh();
2595 EXPORT_SYMBOL(dev_queue_xmit
);
2598 /*=======================================================================
2600 =======================================================================*/
2602 int netdev_max_backlog __read_mostly
= 1000;
2603 int netdev_tstamp_prequeue __read_mostly
= 1;
2604 int netdev_budget __read_mostly
= 300;
2605 int weight_p __read_mostly
= 64; /* old backlog weight */
2607 /* Called with irq disabled */
2608 static inline void ____napi_schedule(struct softnet_data
*sd
,
2609 struct napi_struct
*napi
)
2611 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
2612 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2616 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2617 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2618 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2619 * if hash is a canonical 4-tuple hash over transport ports.
2621 void __skb_get_rxhash(struct sk_buff
*skb
)
2623 struct flow_keys keys
;
2626 if (!skb_flow_dissect(skb
, &keys
))
2630 if ((__force u16
)keys
.port16
[1] < (__force u16
)keys
.port16
[0])
2631 swap(keys
.port16
[0], keys
.port16
[1]);
2635 /* get a consistent hash (same value on both flow directions) */
2636 if ((__force u32
)keys
.dst
< (__force u32
)keys
.src
)
2637 swap(keys
.dst
, keys
.src
);
2639 hash
= jhash_3words((__force u32
)keys
.dst
,
2640 (__force u32
)keys
.src
,
2641 (__force u32
)keys
.ports
, hashrnd
);
2647 EXPORT_SYMBOL(__skb_get_rxhash
);
2651 /* One global table that all flow-based protocols share. */
2652 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
2653 EXPORT_SYMBOL(rps_sock_flow_table
);
2655 struct static_key rps_needed __read_mostly
;
2657 static struct rps_dev_flow
*
2658 set_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2659 struct rps_dev_flow
*rflow
, u16 next_cpu
)
2661 if (next_cpu
!= RPS_NO_CPU
) {
2662 #ifdef CONFIG_RFS_ACCEL
2663 struct netdev_rx_queue
*rxqueue
;
2664 struct rps_dev_flow_table
*flow_table
;
2665 struct rps_dev_flow
*old_rflow
;
2670 /* Should we steer this flow to a different hardware queue? */
2671 if (!skb_rx_queue_recorded(skb
) || !dev
->rx_cpu_rmap
||
2672 !(dev
->features
& NETIF_F_NTUPLE
))
2674 rxq_index
= cpu_rmap_lookup_index(dev
->rx_cpu_rmap
, next_cpu
);
2675 if (rxq_index
== skb_get_rx_queue(skb
))
2678 rxqueue
= dev
->_rx
+ rxq_index
;
2679 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2682 flow_id
= skb
->rxhash
& flow_table
->mask
;
2683 rc
= dev
->netdev_ops
->ndo_rx_flow_steer(dev
, skb
,
2684 rxq_index
, flow_id
);
2688 rflow
= &flow_table
->flows
[flow_id
];
2690 if (old_rflow
->filter
== rflow
->filter
)
2691 old_rflow
->filter
= RPS_NO_FILTER
;
2695 per_cpu(softnet_data
, next_cpu
).input_queue_head
;
2698 rflow
->cpu
= next_cpu
;
2703 * get_rps_cpu is called from netif_receive_skb and returns the target
2704 * CPU from the RPS map of the receiving queue for a given skb.
2705 * rcu_read_lock must be held on entry.
2707 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2708 struct rps_dev_flow
**rflowp
)
2710 struct netdev_rx_queue
*rxqueue
;
2711 struct rps_map
*map
;
2712 struct rps_dev_flow_table
*flow_table
;
2713 struct rps_sock_flow_table
*sock_flow_table
;
2717 if (skb_rx_queue_recorded(skb
)) {
2718 u16 index
= skb_get_rx_queue(skb
);
2719 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
2720 WARN_ONCE(dev
->real_num_rx_queues
> 1,
2721 "%s received packet on queue %u, but number "
2722 "of RX queues is %u\n",
2723 dev
->name
, index
, dev
->real_num_rx_queues
);
2726 rxqueue
= dev
->_rx
+ index
;
2730 map
= rcu_dereference(rxqueue
->rps_map
);
2732 if (map
->len
== 1 &&
2733 !rcu_access_pointer(rxqueue
->rps_flow_table
)) {
2734 tcpu
= map
->cpus
[0];
2735 if (cpu_online(tcpu
))
2739 } else if (!rcu_access_pointer(rxqueue
->rps_flow_table
)) {
2743 skb_reset_network_header(skb
);
2744 if (!skb_get_rxhash(skb
))
2747 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2748 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
2749 if (flow_table
&& sock_flow_table
) {
2751 struct rps_dev_flow
*rflow
;
2753 rflow
= &flow_table
->flows
[skb
->rxhash
& flow_table
->mask
];
2756 next_cpu
= sock_flow_table
->ents
[skb
->rxhash
&
2757 sock_flow_table
->mask
];
2760 * If the desired CPU (where last recvmsg was done) is
2761 * different from current CPU (one in the rx-queue flow
2762 * table entry), switch if one of the following holds:
2763 * - Current CPU is unset (equal to RPS_NO_CPU).
2764 * - Current CPU is offline.
2765 * - The current CPU's queue tail has advanced beyond the
2766 * last packet that was enqueued using this table entry.
2767 * This guarantees that all previous packets for the flow
2768 * have been dequeued, thus preserving in order delivery.
2770 if (unlikely(tcpu
!= next_cpu
) &&
2771 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
2772 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
2773 rflow
->last_qtail
)) >= 0))
2774 rflow
= set_rps_cpu(dev
, skb
, rflow
, next_cpu
);
2776 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
2784 tcpu
= map
->cpus
[((u64
) skb
->rxhash
* map
->len
) >> 32];
2786 if (cpu_online(tcpu
)) {
2796 #ifdef CONFIG_RFS_ACCEL
2799 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2800 * @dev: Device on which the filter was set
2801 * @rxq_index: RX queue index
2802 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2803 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2805 * Drivers that implement ndo_rx_flow_steer() should periodically call
2806 * this function for each installed filter and remove the filters for
2807 * which it returns %true.
2809 bool rps_may_expire_flow(struct net_device
*dev
, u16 rxq_index
,
2810 u32 flow_id
, u16 filter_id
)
2812 struct netdev_rx_queue
*rxqueue
= dev
->_rx
+ rxq_index
;
2813 struct rps_dev_flow_table
*flow_table
;
2814 struct rps_dev_flow
*rflow
;
2819 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2820 if (flow_table
&& flow_id
<= flow_table
->mask
) {
2821 rflow
= &flow_table
->flows
[flow_id
];
2822 cpu
= ACCESS_ONCE(rflow
->cpu
);
2823 if (rflow
->filter
== filter_id
&& cpu
!= RPS_NO_CPU
&&
2824 ((int)(per_cpu(softnet_data
, cpu
).input_queue_head
-
2825 rflow
->last_qtail
) <
2826 (int)(10 * flow_table
->mask
)))
2832 EXPORT_SYMBOL(rps_may_expire_flow
);
2834 #endif /* CONFIG_RFS_ACCEL */
2836 /* Called from hardirq (IPI) context */
2837 static void rps_trigger_softirq(void *data
)
2839 struct softnet_data
*sd
= data
;
2841 ____napi_schedule(sd
, &sd
->backlog
);
2845 #endif /* CONFIG_RPS */
2848 * Check if this softnet_data structure is another cpu one
2849 * If yes, queue it to our IPI list and return 1
2852 static int rps_ipi_queued(struct softnet_data
*sd
)
2855 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
2858 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
2859 mysd
->rps_ipi_list
= sd
;
2861 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2864 #endif /* CONFIG_RPS */
2869 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2870 * queue (may be a remote CPU queue).
2872 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
2873 unsigned int *qtail
)
2875 struct softnet_data
*sd
;
2876 unsigned long flags
;
2878 sd
= &per_cpu(softnet_data
, cpu
);
2880 local_irq_save(flags
);
2883 if (skb_queue_len(&sd
->input_pkt_queue
) <= netdev_max_backlog
) {
2884 if (skb_queue_len(&sd
->input_pkt_queue
)) {
2886 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
2887 input_queue_tail_incr_save(sd
, qtail
);
2889 local_irq_restore(flags
);
2890 return NET_RX_SUCCESS
;
2893 /* Schedule NAPI for backlog device
2894 * We can use non atomic operation since we own the queue lock
2896 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
2897 if (!rps_ipi_queued(sd
))
2898 ____napi_schedule(sd
, &sd
->backlog
);
2906 local_irq_restore(flags
);
2908 atomic_long_inc(&skb
->dev
->rx_dropped
);
2914 * netif_rx - post buffer to the network code
2915 * @skb: buffer to post
2917 * This function receives a packet from a device driver and queues it for
2918 * the upper (protocol) levels to process. It always succeeds. The buffer
2919 * may be dropped during processing for congestion control or by the
2923 * NET_RX_SUCCESS (no congestion)
2924 * NET_RX_DROP (packet was dropped)
2928 int netif_rx(struct sk_buff
*skb
)
2932 /* if netpoll wants it, pretend we never saw it */
2933 if (netpoll_rx(skb
))
2936 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
2938 trace_netif_rx(skb
);
2940 if (static_key_false(&rps_needed
)) {
2941 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
2947 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
2949 cpu
= smp_processor_id();
2951 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
2959 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
2964 EXPORT_SYMBOL(netif_rx
);
2966 int netif_rx_ni(struct sk_buff
*skb
)
2971 err
= netif_rx(skb
);
2972 if (local_softirq_pending())
2978 EXPORT_SYMBOL(netif_rx_ni
);
2980 static void net_tx_action(struct softirq_action
*h
)
2982 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
2984 if (sd
->completion_queue
) {
2985 struct sk_buff
*clist
;
2987 local_irq_disable();
2988 clist
= sd
->completion_queue
;
2989 sd
->completion_queue
= NULL
;
2993 struct sk_buff
*skb
= clist
;
2994 clist
= clist
->next
;
2996 WARN_ON(atomic_read(&skb
->users
));
2997 trace_kfree_skb(skb
, net_tx_action
);
3002 if (sd
->output_queue
) {
3005 local_irq_disable();
3006 head
= sd
->output_queue
;
3007 sd
->output_queue
= NULL
;
3008 sd
->output_queue_tailp
= &sd
->output_queue
;
3012 struct Qdisc
*q
= head
;
3013 spinlock_t
*root_lock
;
3015 head
= head
->next_sched
;
3017 root_lock
= qdisc_lock(q
);
3018 if (spin_trylock(root_lock
)) {
3019 smp_mb__before_clear_bit();
3020 clear_bit(__QDISC_STATE_SCHED
,
3023 spin_unlock(root_lock
);
3025 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
3027 __netif_reschedule(q
);
3029 smp_mb__before_clear_bit();
3030 clear_bit(__QDISC_STATE_SCHED
,
3038 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3039 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3040 /* This hook is defined here for ATM LANE */
3041 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
3042 unsigned char *addr
) __read_mostly
;
3043 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
3046 #ifdef CONFIG_NET_CLS_ACT
3047 /* TODO: Maybe we should just force sch_ingress to be compiled in
3048 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3049 * a compare and 2 stores extra right now if we dont have it on
3050 * but have CONFIG_NET_CLS_ACT
3051 * NOTE: This doesn't stop any functionality; if you dont have
3052 * the ingress scheduler, you just can't add policies on ingress.
3055 static int ing_filter(struct sk_buff
*skb
, struct netdev_queue
*rxq
)
3057 struct net_device
*dev
= skb
->dev
;
3058 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
3059 int result
= TC_ACT_OK
;
3062 if (unlikely(MAX_RED_LOOP
< ttl
++)) {
3063 if (net_ratelimit())
3064 pr_warn("Redir loop detected Dropping packet (%d->%d)\n",
3065 skb
->skb_iif
, dev
->ifindex
);
3069 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
3070 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
3073 if (q
!= &noop_qdisc
) {
3074 spin_lock(qdisc_lock(q
));
3075 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
3076 result
= qdisc_enqueue_root(skb
, q
);
3077 spin_unlock(qdisc_lock(q
));
3083 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
3084 struct packet_type
**pt_prev
,
3085 int *ret
, struct net_device
*orig_dev
)
3087 struct netdev_queue
*rxq
= rcu_dereference(skb
->dev
->ingress_queue
);
3089 if (!rxq
|| rxq
->qdisc
== &noop_qdisc
)
3093 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
3097 switch (ing_filter(skb
, rxq
)) {
3111 * netdev_rx_handler_register - register receive handler
3112 * @dev: device to register a handler for
3113 * @rx_handler: receive handler to register
3114 * @rx_handler_data: data pointer that is used by rx handler
3116 * Register a receive hander for a device. This handler will then be
3117 * called from __netif_receive_skb. A negative errno code is returned
3120 * The caller must hold the rtnl_mutex.
3122 * For a general description of rx_handler, see enum rx_handler_result.
3124 int netdev_rx_handler_register(struct net_device
*dev
,
3125 rx_handler_func_t
*rx_handler
,
3126 void *rx_handler_data
)
3130 if (dev
->rx_handler
)
3133 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
3134 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
3138 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
3141 * netdev_rx_handler_unregister - unregister receive handler
3142 * @dev: device to unregister a handler from
3144 * Unregister a receive hander from a device.
3146 * The caller must hold the rtnl_mutex.
3148 void netdev_rx_handler_unregister(struct net_device
*dev
)
3152 RCU_INIT_POINTER(dev
->rx_handler
, NULL
);
3153 RCU_INIT_POINTER(dev
->rx_handler_data
, NULL
);
3155 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
3157 static int __netif_receive_skb(struct sk_buff
*skb
)
3159 struct packet_type
*ptype
, *pt_prev
;
3160 rx_handler_func_t
*rx_handler
;
3161 struct net_device
*orig_dev
;
3162 struct net_device
*null_or_dev
;
3163 bool deliver_exact
= false;
3164 int ret
= NET_RX_DROP
;
3167 net_timestamp_check(!netdev_tstamp_prequeue
, skb
);
3169 trace_netif_receive_skb(skb
);
3171 /* if we've gotten here through NAPI, check netpoll */
3172 if (netpoll_receive_skb(skb
))
3176 skb
->skb_iif
= skb
->dev
->ifindex
;
3177 orig_dev
= skb
->dev
;
3179 skb_reset_network_header(skb
);
3180 skb_reset_transport_header(skb
);
3181 skb_reset_mac_len(skb
);
3189 __this_cpu_inc(softnet_data
.processed
);
3191 if (skb
->protocol
== cpu_to_be16(ETH_P_8021Q
)) {
3192 skb
= vlan_untag(skb
);
3197 #ifdef CONFIG_NET_CLS_ACT
3198 if (skb
->tc_verd
& TC_NCLS
) {
3199 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
3204 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
3205 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
3207 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3212 #ifdef CONFIG_NET_CLS_ACT
3213 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
3219 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
3220 if (vlan_tx_tag_present(skb
)) {
3222 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3225 if (vlan_do_receive(&skb
, !rx_handler
))
3227 else if (unlikely(!skb
))
3233 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3236 switch (rx_handler(&skb
)) {
3237 case RX_HANDLER_CONSUMED
:
3239 case RX_HANDLER_ANOTHER
:
3241 case RX_HANDLER_EXACT
:
3242 deliver_exact
= true;
3243 case RX_HANDLER_PASS
:
3250 /* deliver only exact match when indicated */
3251 null_or_dev
= deliver_exact
? skb
->dev
: NULL
;
3253 type
= skb
->protocol
;
3254 list_for_each_entry_rcu(ptype
,
3255 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
3256 if (ptype
->type
== type
&&
3257 (ptype
->dev
== null_or_dev
|| ptype
->dev
== skb
->dev
||
3258 ptype
->dev
== orig_dev
)) {
3260 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3266 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
3268 atomic_long_inc(&skb
->dev
->rx_dropped
);
3270 /* Jamal, now you will not able to escape explaining
3271 * me how you were going to use this. :-)
3282 * netif_receive_skb - process receive buffer from network
3283 * @skb: buffer to process
3285 * netif_receive_skb() is the main receive data processing function.
3286 * It always succeeds. The buffer may be dropped during processing
3287 * for congestion control or by the protocol layers.
3289 * This function may only be called from softirq context and interrupts
3290 * should be enabled.
3292 * Return values (usually ignored):
3293 * NET_RX_SUCCESS: no congestion
3294 * NET_RX_DROP: packet was dropped
3296 int netif_receive_skb(struct sk_buff
*skb
)
3298 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3300 if (skb_defer_rx_timestamp(skb
))
3301 return NET_RX_SUCCESS
;
3304 if (static_key_false(&rps_needed
)) {
3305 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3310 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3313 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3320 return __netif_receive_skb(skb
);
3322 EXPORT_SYMBOL(netif_receive_skb
);
3324 /* Network device is going away, flush any packets still pending
3325 * Called with irqs disabled.
3327 static void flush_backlog(void *arg
)
3329 struct net_device
*dev
= arg
;
3330 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3331 struct sk_buff
*skb
, *tmp
;
3334 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3335 if (skb
->dev
== dev
) {
3336 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3338 input_queue_head_incr(sd
);
3343 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3344 if (skb
->dev
== dev
) {
3345 __skb_unlink(skb
, &sd
->process_queue
);
3347 input_queue_head_incr(sd
);
3352 static int napi_gro_complete(struct sk_buff
*skb
)
3354 struct packet_type
*ptype
;
3355 __be16 type
= skb
->protocol
;
3356 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3359 if (NAPI_GRO_CB(skb
)->count
== 1) {
3360 skb_shinfo(skb
)->gso_size
= 0;
3365 list_for_each_entry_rcu(ptype
, head
, list
) {
3366 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_complete
)
3369 err
= ptype
->gro_complete(skb
);
3375 WARN_ON(&ptype
->list
== head
);
3377 return NET_RX_SUCCESS
;
3381 return netif_receive_skb(skb
);
3384 inline void napi_gro_flush(struct napi_struct
*napi
)
3386 struct sk_buff
*skb
, *next
;
3388 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3391 napi_gro_complete(skb
);
3394 napi
->gro_count
= 0;
3395 napi
->gro_list
= NULL
;
3397 EXPORT_SYMBOL(napi_gro_flush
);
3399 enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3401 struct sk_buff
**pp
= NULL
;
3402 struct packet_type
*ptype
;
3403 __be16 type
= skb
->protocol
;
3404 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3407 enum gro_result ret
;
3409 if (!(skb
->dev
->features
& NETIF_F_GRO
) || netpoll_rx_on(skb
))
3412 if (skb_is_gso(skb
) || skb_has_frag_list(skb
))
3416 list_for_each_entry_rcu(ptype
, head
, list
) {
3417 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_receive
)
3420 skb_set_network_header(skb
, skb_gro_offset(skb
));
3421 mac_len
= skb
->network_header
- skb
->mac_header
;
3422 skb
->mac_len
= mac_len
;
3423 NAPI_GRO_CB(skb
)->same_flow
= 0;
3424 NAPI_GRO_CB(skb
)->flush
= 0;
3425 NAPI_GRO_CB(skb
)->free
= 0;
3427 pp
= ptype
->gro_receive(&napi
->gro_list
, skb
);
3432 if (&ptype
->list
== head
)
3435 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3436 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3439 struct sk_buff
*nskb
= *pp
;
3443 napi_gro_complete(nskb
);
3450 if (NAPI_GRO_CB(skb
)->flush
|| napi
->gro_count
>= MAX_GRO_SKBS
)
3454 NAPI_GRO_CB(skb
)->count
= 1;
3455 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
3456 skb
->next
= napi
->gro_list
;
3457 napi
->gro_list
= skb
;
3461 if (skb_headlen(skb
) < skb_gro_offset(skb
)) {
3462 int grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
3464 BUG_ON(skb
->end
- skb
->tail
< grow
);
3466 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3469 skb
->data_len
-= grow
;
3471 skb_shinfo(skb
)->frags
[0].page_offset
+= grow
;
3472 skb_frag_size_sub(&skb_shinfo(skb
)->frags
[0], grow
);
3474 if (unlikely(!skb_frag_size(&skb_shinfo(skb
)->frags
[0]))) {
3475 skb_frag_unref(skb
, 0);
3476 memmove(skb_shinfo(skb
)->frags
,
3477 skb_shinfo(skb
)->frags
+ 1,
3478 --skb_shinfo(skb
)->nr_frags
* sizeof(skb_frag_t
));
3489 EXPORT_SYMBOL(dev_gro_receive
);
3491 static inline gro_result_t
3492 __napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3495 unsigned int maclen
= skb
->dev
->hard_header_len
;
3497 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3498 unsigned long diffs
;
3500 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
3501 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
3502 if (maclen
== ETH_HLEN
)
3503 diffs
|= compare_ether_header(skb_mac_header(p
),
3504 skb_gro_mac_header(skb
));
3506 diffs
= memcmp(skb_mac_header(p
),
3507 skb_gro_mac_header(skb
),
3509 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
3510 NAPI_GRO_CB(p
)->flush
= 0;
3513 return dev_gro_receive(napi
, skb
);
3516 gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
3520 if (netif_receive_skb(skb
))
3525 case GRO_MERGED_FREE
:
3536 EXPORT_SYMBOL(napi_skb_finish
);
3538 void skb_gro_reset_offset(struct sk_buff
*skb
)
3540 NAPI_GRO_CB(skb
)->data_offset
= 0;
3541 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3542 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3544 if (skb
->mac_header
== skb
->tail
&&
3545 !PageHighMem(skb_frag_page(&skb_shinfo(skb
)->frags
[0]))) {
3546 NAPI_GRO_CB(skb
)->frag0
=
3547 skb_frag_address(&skb_shinfo(skb
)->frags
[0]);
3548 NAPI_GRO_CB(skb
)->frag0_len
= skb_frag_size(&skb_shinfo(skb
)->frags
[0]);
3551 EXPORT_SYMBOL(skb_gro_reset_offset
);
3553 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3555 skb_gro_reset_offset(skb
);
3557 return napi_skb_finish(__napi_gro_receive(napi
, skb
), skb
);
3559 EXPORT_SYMBOL(napi_gro_receive
);
3561 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
3563 __skb_pull(skb
, skb_headlen(skb
));
3564 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3565 skb_reserve(skb
, NET_SKB_PAD
+ NET_IP_ALIGN
- skb_headroom(skb
));
3567 skb
->dev
= napi
->dev
;
3573 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
3575 struct sk_buff
*skb
= napi
->skb
;
3578 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
3584 EXPORT_SYMBOL(napi_get_frags
);
3586 gro_result_t
napi_frags_finish(struct napi_struct
*napi
, struct sk_buff
*skb
,
3592 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
3594 if (ret
== GRO_HELD
)
3595 skb_gro_pull(skb
, -ETH_HLEN
);
3596 else if (netif_receive_skb(skb
))
3601 case GRO_MERGED_FREE
:
3602 napi_reuse_skb(napi
, skb
);
3611 EXPORT_SYMBOL(napi_frags_finish
);
3613 struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
3615 struct sk_buff
*skb
= napi
->skb
;
3622 skb_reset_mac_header(skb
);
3623 skb_gro_reset_offset(skb
);
3625 off
= skb_gro_offset(skb
);
3626 hlen
= off
+ sizeof(*eth
);
3627 eth
= skb_gro_header_fast(skb
, off
);
3628 if (skb_gro_header_hard(skb
, hlen
)) {
3629 eth
= skb_gro_header_slow(skb
, hlen
, off
);
3630 if (unlikely(!eth
)) {
3631 napi_reuse_skb(napi
, skb
);
3637 skb_gro_pull(skb
, sizeof(*eth
));
3640 * This works because the only protocols we care about don't require
3641 * special handling. We'll fix it up properly at the end.
3643 skb
->protocol
= eth
->h_proto
;
3648 EXPORT_SYMBOL(napi_frags_skb
);
3650 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
3652 struct sk_buff
*skb
= napi_frags_skb(napi
);
3657 return napi_frags_finish(napi
, skb
, __napi_gro_receive(napi
, skb
));
3659 EXPORT_SYMBOL(napi_gro_frags
);
3662 * net_rps_action sends any pending IPI's for rps.
3663 * Note: called with local irq disabled, but exits with local irq enabled.
3665 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
3668 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
3671 sd
->rps_ipi_list
= NULL
;
3675 /* Send pending IPI's to kick RPS processing on remote cpus. */
3677 struct softnet_data
*next
= remsd
->rps_ipi_next
;
3679 if (cpu_online(remsd
->cpu
))
3680 __smp_call_function_single(remsd
->cpu
,
3689 static int process_backlog(struct napi_struct
*napi
, int quota
)
3692 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
3695 /* Check if we have pending ipi, its better to send them now,
3696 * not waiting net_rx_action() end.
3698 if (sd
->rps_ipi_list
) {
3699 local_irq_disable();
3700 net_rps_action_and_irq_enable(sd
);
3703 napi
->weight
= weight_p
;
3704 local_irq_disable();
3705 while (work
< quota
) {
3706 struct sk_buff
*skb
;
3709 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
3711 __netif_receive_skb(skb
);
3712 local_irq_disable();
3713 input_queue_head_incr(sd
);
3714 if (++work
>= quota
) {
3721 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3723 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
3724 &sd
->process_queue
);
3726 if (qlen
< quota
- work
) {
3728 * Inline a custom version of __napi_complete().
3729 * only current cpu owns and manipulates this napi,
3730 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3731 * we can use a plain write instead of clear_bit(),
3732 * and we dont need an smp_mb() memory barrier.
3734 list_del(&napi
->poll_list
);
3737 quota
= work
+ qlen
;
3747 * __napi_schedule - schedule for receive
3748 * @n: entry to schedule
3750 * The entry's receive function will be scheduled to run
3752 void __napi_schedule(struct napi_struct
*n
)
3754 unsigned long flags
;
3756 local_irq_save(flags
);
3757 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
3758 local_irq_restore(flags
);
3760 EXPORT_SYMBOL(__napi_schedule
);
3762 void __napi_complete(struct napi_struct
*n
)
3764 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
3765 BUG_ON(n
->gro_list
);
3767 list_del(&n
->poll_list
);
3768 smp_mb__before_clear_bit();
3769 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
3771 EXPORT_SYMBOL(__napi_complete
);
3773 void napi_complete(struct napi_struct
*n
)
3775 unsigned long flags
;
3778 * don't let napi dequeue from the cpu poll list
3779 * just in case its running on a different cpu
3781 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
3785 local_irq_save(flags
);
3787 local_irq_restore(flags
);
3789 EXPORT_SYMBOL(napi_complete
);
3791 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
3792 int (*poll
)(struct napi_struct
*, int), int weight
)
3794 INIT_LIST_HEAD(&napi
->poll_list
);
3795 napi
->gro_count
= 0;
3796 napi
->gro_list
= NULL
;
3799 napi
->weight
= weight
;
3800 list_add(&napi
->dev_list
, &dev
->napi_list
);
3802 #ifdef CONFIG_NETPOLL
3803 spin_lock_init(&napi
->poll_lock
);
3804 napi
->poll_owner
= -1;
3806 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
3808 EXPORT_SYMBOL(netif_napi_add
);
3810 void netif_napi_del(struct napi_struct
*napi
)
3812 struct sk_buff
*skb
, *next
;
3814 list_del_init(&napi
->dev_list
);
3815 napi_free_frags(napi
);
3817 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3823 napi
->gro_list
= NULL
;
3824 napi
->gro_count
= 0;
3826 EXPORT_SYMBOL(netif_napi_del
);
3828 static void net_rx_action(struct softirq_action
*h
)
3830 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3831 unsigned long time_limit
= jiffies
+ 2;
3832 int budget
= netdev_budget
;
3835 local_irq_disable();
3837 while (!list_empty(&sd
->poll_list
)) {
3838 struct napi_struct
*n
;
3841 /* If softirq window is exhuasted then punt.
3842 * Allow this to run for 2 jiffies since which will allow
3843 * an average latency of 1.5/HZ.
3845 if (unlikely(budget
<= 0 || time_after(jiffies
, time_limit
)))
3850 /* Even though interrupts have been re-enabled, this
3851 * access is safe because interrupts can only add new
3852 * entries to the tail of this list, and only ->poll()
3853 * calls can remove this head entry from the list.
3855 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
3857 have
= netpoll_poll_lock(n
);
3861 /* This NAPI_STATE_SCHED test is for avoiding a race
3862 * with netpoll's poll_napi(). Only the entity which
3863 * obtains the lock and sees NAPI_STATE_SCHED set will
3864 * actually make the ->poll() call. Therefore we avoid
3865 * accidentally calling ->poll() when NAPI is not scheduled.
3868 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
3869 work
= n
->poll(n
, weight
);
3873 WARN_ON_ONCE(work
> weight
);
3877 local_irq_disable();
3879 /* Drivers must not modify the NAPI state if they
3880 * consume the entire weight. In such cases this code
3881 * still "owns" the NAPI instance and therefore can
3882 * move the instance around on the list at-will.
3884 if (unlikely(work
== weight
)) {
3885 if (unlikely(napi_disable_pending(n
))) {
3888 local_irq_disable();
3890 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
3893 netpoll_poll_unlock(have
);
3896 net_rps_action_and_irq_enable(sd
);
3898 #ifdef CONFIG_NET_DMA
3900 * There may not be any more sk_buffs coming right now, so push
3901 * any pending DMA copies to hardware
3903 dma_issue_pending_all();
3910 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3914 static gifconf_func_t
*gifconf_list
[NPROTO
];
3917 * register_gifconf - register a SIOCGIF handler
3918 * @family: Address family
3919 * @gifconf: Function handler
3921 * Register protocol dependent address dumping routines. The handler
3922 * that is passed must not be freed or reused until it has been replaced
3923 * by another handler.
3925 int register_gifconf(unsigned int family
, gifconf_func_t
*gifconf
)
3927 if (family
>= NPROTO
)
3929 gifconf_list
[family
] = gifconf
;
3932 EXPORT_SYMBOL(register_gifconf
);
3936 * Map an interface index to its name (SIOCGIFNAME)
3940 * We need this ioctl for efficient implementation of the
3941 * if_indextoname() function required by the IPv6 API. Without
3942 * it, we would have to search all the interfaces to find a
3946 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
3948 struct net_device
*dev
;
3952 * Fetch the caller's info block.
3955 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
3959 dev
= dev_get_by_index_rcu(net
, ifr
.ifr_ifindex
);
3965 strcpy(ifr
.ifr_name
, dev
->name
);
3968 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
3974 * Perform a SIOCGIFCONF call. This structure will change
3975 * size eventually, and there is nothing I can do about it.
3976 * Thus we will need a 'compatibility mode'.
3979 static int dev_ifconf(struct net
*net
, char __user
*arg
)
3982 struct net_device
*dev
;
3989 * Fetch the caller's info block.
3992 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
3999 * Loop over the interfaces, and write an info block for each.
4003 for_each_netdev(net
, dev
) {
4004 for (i
= 0; i
< NPROTO
; i
++) {
4005 if (gifconf_list
[i
]) {
4008 done
= gifconf_list
[i
](dev
, NULL
, 0);
4010 done
= gifconf_list
[i
](dev
, pos
+ total
,
4020 * All done. Write the updated control block back to the caller.
4022 ifc
.ifc_len
= total
;
4025 * Both BSD and Solaris return 0 here, so we do too.
4027 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
4030 #ifdef CONFIG_PROC_FS
4032 #define BUCKET_SPACE (32 - NETDEV_HASHBITS)
4034 struct dev_iter_state
{
4035 struct seq_net_private p
;
4036 unsigned int pos
; /* bucket << BUCKET_SPACE + offset */
4039 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4040 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4041 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4043 static inline struct net_device
*dev_from_same_bucket(struct seq_file
*seq
)
4045 struct dev_iter_state
*state
= seq
->private;
4046 struct net
*net
= seq_file_net(seq
);
4047 struct net_device
*dev
;
4048 struct hlist_node
*p
;
4049 struct hlist_head
*h
;
4050 unsigned int count
, bucket
, offset
;
4052 bucket
= get_bucket(state
->pos
);
4053 offset
= get_offset(state
->pos
);
4054 h
= &net
->dev_name_head
[bucket
];
4056 hlist_for_each_entry_rcu(dev
, p
, h
, name_hlist
) {
4057 if (count
++ == offset
) {
4058 state
->pos
= set_bucket_offset(bucket
, count
);
4066 static inline struct net_device
*dev_from_new_bucket(struct seq_file
*seq
)
4068 struct dev_iter_state
*state
= seq
->private;
4069 struct net_device
*dev
;
4070 unsigned int bucket
;
4072 bucket
= get_bucket(state
->pos
);
4074 dev
= dev_from_same_bucket(seq
);
4079 state
->pos
= set_bucket_offset(bucket
, 0);
4080 } while (bucket
< NETDEV_HASHENTRIES
);
4086 * This is invoked by the /proc filesystem handler to display a device
4089 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4092 struct dev_iter_state
*state
= seq
->private;
4096 return SEQ_START_TOKEN
;
4098 /* check for end of the hash */
4099 if (state
->pos
== 0 && *pos
> 1)
4102 return dev_from_new_bucket(seq
);
4105 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4107 struct net_device
*dev
;
4111 if (v
== SEQ_START_TOKEN
)
4112 return dev_from_new_bucket(seq
);
4114 dev
= dev_from_same_bucket(seq
);
4118 return dev_from_new_bucket(seq
);
4121 void dev_seq_stop(struct seq_file
*seq
, void *v
)
4127 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
4129 struct rtnl_link_stats64 temp
;
4130 const struct rtnl_link_stats64
*stats
= dev_get_stats(dev
, &temp
);
4132 seq_printf(seq
, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4133 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4134 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
4136 stats
->rx_dropped
+ stats
->rx_missed_errors
,
4137 stats
->rx_fifo_errors
,
4138 stats
->rx_length_errors
+ stats
->rx_over_errors
+
4139 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
4140 stats
->rx_compressed
, stats
->multicast
,
4141 stats
->tx_bytes
, stats
->tx_packets
,
4142 stats
->tx_errors
, stats
->tx_dropped
,
4143 stats
->tx_fifo_errors
, stats
->collisions
,
4144 stats
->tx_carrier_errors
+
4145 stats
->tx_aborted_errors
+
4146 stats
->tx_window_errors
+
4147 stats
->tx_heartbeat_errors
,
4148 stats
->tx_compressed
);
4152 * Called from the PROCfs module. This now uses the new arbitrary sized
4153 * /proc/net interface to create /proc/net/dev
4155 static int dev_seq_show(struct seq_file
*seq
, void *v
)
4157 if (v
== SEQ_START_TOKEN
)
4158 seq_puts(seq
, "Inter-| Receive "
4160 " face |bytes packets errs drop fifo frame "
4161 "compressed multicast|bytes packets errs "
4162 "drop fifo colls carrier compressed\n");
4164 dev_seq_printf_stats(seq
, v
);
4168 static struct softnet_data
*softnet_get_online(loff_t
*pos
)
4170 struct softnet_data
*sd
= NULL
;
4172 while (*pos
< nr_cpu_ids
)
4173 if (cpu_online(*pos
)) {
4174 sd
= &per_cpu(softnet_data
, *pos
);
4181 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4183 return softnet_get_online(pos
);
4186 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4189 return softnet_get_online(pos
);
4192 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
4196 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
4198 struct softnet_data
*sd
= v
;
4200 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4201 sd
->processed
, sd
->dropped
, sd
->time_squeeze
, 0,
4202 0, 0, 0, 0, /* was fastroute */
4203 sd
->cpu_collision
, sd
->received_rps
);
4207 static const struct seq_operations dev_seq_ops
= {
4208 .start
= dev_seq_start
,
4209 .next
= dev_seq_next
,
4210 .stop
= dev_seq_stop
,
4211 .show
= dev_seq_show
,
4214 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
4216 return seq_open_net(inode
, file
, &dev_seq_ops
,
4217 sizeof(struct dev_iter_state
));
4220 int dev_seq_open_ops(struct inode
*inode
, struct file
*file
,
4221 const struct seq_operations
*ops
)
4223 return seq_open_net(inode
, file
, ops
, sizeof(struct dev_iter_state
));
4226 static const struct file_operations dev_seq_fops
= {
4227 .owner
= THIS_MODULE
,
4228 .open
= dev_seq_open
,
4230 .llseek
= seq_lseek
,
4231 .release
= seq_release_net
,
4234 static const struct seq_operations softnet_seq_ops
= {
4235 .start
= softnet_seq_start
,
4236 .next
= softnet_seq_next
,
4237 .stop
= softnet_seq_stop
,
4238 .show
= softnet_seq_show
,
4241 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
4243 return seq_open(file
, &softnet_seq_ops
);
4246 static const struct file_operations softnet_seq_fops
= {
4247 .owner
= THIS_MODULE
,
4248 .open
= softnet_seq_open
,
4250 .llseek
= seq_lseek
,
4251 .release
= seq_release
,
4254 static void *ptype_get_idx(loff_t pos
)
4256 struct packet_type
*pt
= NULL
;
4260 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
4266 for (t
= 0; t
< PTYPE_HASH_SIZE
; t
++) {
4267 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
4276 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4280 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
4283 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4285 struct packet_type
*pt
;
4286 struct list_head
*nxt
;
4290 if (v
== SEQ_START_TOKEN
)
4291 return ptype_get_idx(0);
4294 nxt
= pt
->list
.next
;
4295 if (pt
->type
== htons(ETH_P_ALL
)) {
4296 if (nxt
!= &ptype_all
)
4299 nxt
= ptype_base
[0].next
;
4301 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
4303 while (nxt
== &ptype_base
[hash
]) {
4304 if (++hash
>= PTYPE_HASH_SIZE
)
4306 nxt
= ptype_base
[hash
].next
;
4309 return list_entry(nxt
, struct packet_type
, list
);
4312 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
4318 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
4320 struct packet_type
*pt
= v
;
4322 if (v
== SEQ_START_TOKEN
)
4323 seq_puts(seq
, "Type Device Function\n");
4324 else if (pt
->dev
== NULL
|| dev_net(pt
->dev
) == seq_file_net(seq
)) {
4325 if (pt
->type
== htons(ETH_P_ALL
))
4326 seq_puts(seq
, "ALL ");
4328 seq_printf(seq
, "%04x", ntohs(pt
->type
));
4330 seq_printf(seq
, " %-8s %pF\n",
4331 pt
->dev
? pt
->dev
->name
: "", pt
->func
);
4337 static const struct seq_operations ptype_seq_ops
= {
4338 .start
= ptype_seq_start
,
4339 .next
= ptype_seq_next
,
4340 .stop
= ptype_seq_stop
,
4341 .show
= ptype_seq_show
,
4344 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
4346 return seq_open_net(inode
, file
, &ptype_seq_ops
,
4347 sizeof(struct seq_net_private
));
4350 static const struct file_operations ptype_seq_fops
= {
4351 .owner
= THIS_MODULE
,
4352 .open
= ptype_seq_open
,
4354 .llseek
= seq_lseek
,
4355 .release
= seq_release_net
,
4359 static int __net_init
dev_proc_net_init(struct net
*net
)
4363 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
4365 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
4367 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
4370 if (wext_proc_init(net
))
4376 proc_net_remove(net
, "ptype");
4378 proc_net_remove(net
, "softnet_stat");
4380 proc_net_remove(net
, "dev");
4384 static void __net_exit
dev_proc_net_exit(struct net
*net
)
4386 wext_proc_exit(net
);
4388 proc_net_remove(net
, "ptype");
4389 proc_net_remove(net
, "softnet_stat");
4390 proc_net_remove(net
, "dev");
4393 static struct pernet_operations __net_initdata dev_proc_ops
= {
4394 .init
= dev_proc_net_init
,
4395 .exit
= dev_proc_net_exit
,
4398 static int __init
dev_proc_init(void)
4400 return register_pernet_subsys(&dev_proc_ops
);
4403 #define dev_proc_init() 0
4404 #endif /* CONFIG_PROC_FS */
4408 * netdev_set_master - set up master pointer
4409 * @slave: slave device
4410 * @master: new master device
4412 * Changes the master device of the slave. Pass %NULL to break the
4413 * bonding. The caller must hold the RTNL semaphore. On a failure
4414 * a negative errno code is returned. On success the reference counts
4415 * are adjusted and the function returns zero.
4417 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
4419 struct net_device
*old
= slave
->master
;
4429 slave
->master
= master
;
4435 EXPORT_SYMBOL(netdev_set_master
);
4438 * netdev_set_bond_master - set up bonding master/slave pair
4439 * @slave: slave device
4440 * @master: new master device
4442 * Changes the master device of the slave. Pass %NULL to break the
4443 * bonding. The caller must hold the RTNL semaphore. On a failure
4444 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4445 * to the routing socket and the function returns zero.
4447 int netdev_set_bond_master(struct net_device
*slave
, struct net_device
*master
)
4453 err
= netdev_set_master(slave
, master
);
4457 slave
->flags
|= IFF_SLAVE
;
4459 slave
->flags
&= ~IFF_SLAVE
;
4461 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
4464 EXPORT_SYMBOL(netdev_set_bond_master
);
4466 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
4468 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4470 if ((dev
->flags
& IFF_UP
) && ops
->ndo_change_rx_flags
)
4471 ops
->ndo_change_rx_flags(dev
, flags
);
4474 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
4476 unsigned int old_flags
= dev
->flags
;
4482 dev
->flags
|= IFF_PROMISC
;
4483 dev
->promiscuity
+= inc
;
4484 if (dev
->promiscuity
== 0) {
4487 * If inc causes overflow, untouch promisc and return error.
4490 dev
->flags
&= ~IFF_PROMISC
;
4492 dev
->promiscuity
-= inc
;
4493 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4498 if (dev
->flags
!= old_flags
) {
4499 pr_info("device %s %s promiscuous mode\n",
4501 dev
->flags
& IFF_PROMISC
? "entered" : "left");
4502 if (audit_enabled
) {
4503 current_uid_gid(&uid
, &gid
);
4504 audit_log(current
->audit_context
, GFP_ATOMIC
,
4505 AUDIT_ANOM_PROMISCUOUS
,
4506 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4507 dev
->name
, (dev
->flags
& IFF_PROMISC
),
4508 (old_flags
& IFF_PROMISC
),
4509 audit_get_loginuid(current
),
4511 audit_get_sessionid(current
));
4514 dev_change_rx_flags(dev
, IFF_PROMISC
);
4520 * dev_set_promiscuity - update promiscuity count on a device
4524 * Add or remove promiscuity from a device. While the count in the device
4525 * remains above zero the interface remains promiscuous. Once it hits zero
4526 * the device reverts back to normal filtering operation. A negative inc
4527 * value is used to drop promiscuity on the device.
4528 * Return 0 if successful or a negative errno code on error.
4530 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
4532 unsigned int old_flags
= dev
->flags
;
4535 err
= __dev_set_promiscuity(dev
, inc
);
4538 if (dev
->flags
!= old_flags
)
4539 dev_set_rx_mode(dev
);
4542 EXPORT_SYMBOL(dev_set_promiscuity
);
4545 * dev_set_allmulti - update allmulti count on a device
4549 * Add or remove reception of all multicast frames to a device. While the
4550 * count in the device remains above zero the interface remains listening
4551 * to all interfaces. Once it hits zero the device reverts back to normal
4552 * filtering operation. A negative @inc value is used to drop the counter
4553 * when releasing a resource needing all multicasts.
4554 * Return 0 if successful or a negative errno code on error.
4557 int dev_set_allmulti(struct net_device
*dev
, int inc
)
4559 unsigned int old_flags
= dev
->flags
;
4563 dev
->flags
|= IFF_ALLMULTI
;
4564 dev
->allmulti
+= inc
;
4565 if (dev
->allmulti
== 0) {
4568 * If inc causes overflow, untouch allmulti and return error.
4571 dev
->flags
&= ~IFF_ALLMULTI
;
4573 dev
->allmulti
-= inc
;
4574 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4579 if (dev
->flags
^ old_flags
) {
4580 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
4581 dev_set_rx_mode(dev
);
4585 EXPORT_SYMBOL(dev_set_allmulti
);
4588 * Upload unicast and multicast address lists to device and
4589 * configure RX filtering. When the device doesn't support unicast
4590 * filtering it is put in promiscuous mode while unicast addresses
4593 void __dev_set_rx_mode(struct net_device
*dev
)
4595 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4597 /* dev_open will call this function so the list will stay sane. */
4598 if (!(dev
->flags
&IFF_UP
))
4601 if (!netif_device_present(dev
))
4604 if (!(dev
->priv_flags
& IFF_UNICAST_FLT
)) {
4605 /* Unicast addresses changes may only happen under the rtnl,
4606 * therefore calling __dev_set_promiscuity here is safe.
4608 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
4609 __dev_set_promiscuity(dev
, 1);
4610 dev
->uc_promisc
= true;
4611 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
4612 __dev_set_promiscuity(dev
, -1);
4613 dev
->uc_promisc
= false;
4617 if (ops
->ndo_set_rx_mode
)
4618 ops
->ndo_set_rx_mode(dev
);
4621 void dev_set_rx_mode(struct net_device
*dev
)
4623 netif_addr_lock_bh(dev
);
4624 __dev_set_rx_mode(dev
);
4625 netif_addr_unlock_bh(dev
);
4629 * dev_get_flags - get flags reported to userspace
4632 * Get the combination of flag bits exported through APIs to userspace.
4634 unsigned dev_get_flags(const struct net_device
*dev
)
4638 flags
= (dev
->flags
& ~(IFF_PROMISC
|
4643 (dev
->gflags
& (IFF_PROMISC
|
4646 if (netif_running(dev
)) {
4647 if (netif_oper_up(dev
))
4648 flags
|= IFF_RUNNING
;
4649 if (netif_carrier_ok(dev
))
4650 flags
|= IFF_LOWER_UP
;
4651 if (netif_dormant(dev
))
4652 flags
|= IFF_DORMANT
;
4657 EXPORT_SYMBOL(dev_get_flags
);
4659 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4661 unsigned int old_flags
= dev
->flags
;
4667 * Set the flags on our device.
4670 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
4671 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
4673 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
4677 * Load in the correct multicast list now the flags have changed.
4680 if ((old_flags
^ flags
) & IFF_MULTICAST
)
4681 dev_change_rx_flags(dev
, IFF_MULTICAST
);
4683 dev_set_rx_mode(dev
);
4686 * Have we downed the interface. We handle IFF_UP ourselves
4687 * according to user attempts to set it, rather than blindly
4692 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
4693 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
4696 dev_set_rx_mode(dev
);
4699 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
4700 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
4702 dev
->gflags
^= IFF_PROMISC
;
4703 dev_set_promiscuity(dev
, inc
);
4706 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4707 is important. Some (broken) drivers set IFF_PROMISC, when
4708 IFF_ALLMULTI is requested not asking us and not reporting.
4710 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
4711 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
4713 dev
->gflags
^= IFF_ALLMULTI
;
4714 dev_set_allmulti(dev
, inc
);
4720 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
)
4722 unsigned int changes
= dev
->flags
^ old_flags
;
4724 if (changes
& IFF_UP
) {
4725 if (dev
->flags
& IFF_UP
)
4726 call_netdevice_notifiers(NETDEV_UP
, dev
);
4728 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
4731 if (dev
->flags
& IFF_UP
&&
4732 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
)))
4733 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
4737 * dev_change_flags - change device settings
4739 * @flags: device state flags
4741 * Change settings on device based state flags. The flags are
4742 * in the userspace exported format.
4744 int dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4747 unsigned int changes
, old_flags
= dev
->flags
;
4749 ret
= __dev_change_flags(dev
, flags
);
4753 changes
= old_flags
^ dev
->flags
;
4755 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
4757 __dev_notify_flags(dev
, old_flags
);
4760 EXPORT_SYMBOL(dev_change_flags
);
4763 * dev_set_mtu - Change maximum transfer unit
4765 * @new_mtu: new transfer unit
4767 * Change the maximum transfer size of the network device.
4769 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
4771 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4774 if (new_mtu
== dev
->mtu
)
4777 /* MTU must be positive. */
4781 if (!netif_device_present(dev
))
4785 if (ops
->ndo_change_mtu
)
4786 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
4790 if (!err
&& dev
->flags
& IFF_UP
)
4791 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
4794 EXPORT_SYMBOL(dev_set_mtu
);
4797 * dev_set_group - Change group this device belongs to
4799 * @new_group: group this device should belong to
4801 void dev_set_group(struct net_device
*dev
, int new_group
)
4803 dev
->group
= new_group
;
4805 EXPORT_SYMBOL(dev_set_group
);
4808 * dev_set_mac_address - Change Media Access Control Address
4812 * Change the hardware (MAC) address of the device
4814 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
4816 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4819 if (!ops
->ndo_set_mac_address
)
4821 if (sa
->sa_family
!= dev
->type
)
4823 if (!netif_device_present(dev
))
4825 err
= ops
->ndo_set_mac_address(dev
, sa
);
4827 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4830 EXPORT_SYMBOL(dev_set_mac_address
);
4833 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4835 static int dev_ifsioc_locked(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4838 struct net_device
*dev
= dev_get_by_name_rcu(net
, ifr
->ifr_name
);
4844 case SIOCGIFFLAGS
: /* Get interface flags */
4845 ifr
->ifr_flags
= (short) dev_get_flags(dev
);
4848 case SIOCGIFMETRIC
: /* Get the metric on the interface
4849 (currently unused) */
4850 ifr
->ifr_metric
= 0;
4853 case SIOCGIFMTU
: /* Get the MTU of a device */
4854 ifr
->ifr_mtu
= dev
->mtu
;
4859 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
4861 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
4862 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4863 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
4871 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
4872 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
4873 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
4874 ifr
->ifr_map
.irq
= dev
->irq
;
4875 ifr
->ifr_map
.dma
= dev
->dma
;
4876 ifr
->ifr_map
.port
= dev
->if_port
;
4880 ifr
->ifr_ifindex
= dev
->ifindex
;
4884 ifr
->ifr_qlen
= dev
->tx_queue_len
;
4888 /* dev_ioctl() should ensure this case
4900 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4902 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4905 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
4906 const struct net_device_ops
*ops
;
4911 ops
= dev
->netdev_ops
;
4914 case SIOCSIFFLAGS
: /* Set interface flags */
4915 return dev_change_flags(dev
, ifr
->ifr_flags
);
4917 case SIOCSIFMETRIC
: /* Set the metric on the interface
4918 (currently unused) */
4921 case SIOCSIFMTU
: /* Set the MTU of a device */
4922 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
4925 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
4927 case SIOCSIFHWBROADCAST
:
4928 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
4930 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
4931 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4932 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4936 if (ops
->ndo_set_config
) {
4937 if (!netif_device_present(dev
))
4939 return ops
->ndo_set_config(dev
, &ifr
->ifr_map
);
4944 if (!ops
->ndo_set_rx_mode
||
4945 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4947 if (!netif_device_present(dev
))
4949 return dev_mc_add_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4952 if (!ops
->ndo_set_rx_mode
||
4953 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4955 if (!netif_device_present(dev
))
4957 return dev_mc_del_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4960 if (ifr
->ifr_qlen
< 0)
4962 dev
->tx_queue_len
= ifr
->ifr_qlen
;
4966 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
4967 return dev_change_name(dev
, ifr
->ifr_newname
);
4970 err
= net_hwtstamp_validate(ifr
);
4976 * Unknown or private ioctl
4979 if ((cmd
>= SIOCDEVPRIVATE
&&
4980 cmd
<= SIOCDEVPRIVATE
+ 15) ||
4981 cmd
== SIOCBONDENSLAVE
||
4982 cmd
== SIOCBONDRELEASE
||
4983 cmd
== SIOCBONDSETHWADDR
||
4984 cmd
== SIOCBONDSLAVEINFOQUERY
||
4985 cmd
== SIOCBONDINFOQUERY
||
4986 cmd
== SIOCBONDCHANGEACTIVE
||
4987 cmd
== SIOCGMIIPHY
||
4988 cmd
== SIOCGMIIREG
||
4989 cmd
== SIOCSMIIREG
||
4990 cmd
== SIOCBRADDIF
||
4991 cmd
== SIOCBRDELIF
||
4992 cmd
== SIOCSHWTSTAMP
||
4993 cmd
== SIOCWANDEV
) {
4995 if (ops
->ndo_do_ioctl
) {
4996 if (netif_device_present(dev
))
4997 err
= ops
->ndo_do_ioctl(dev
, ifr
, cmd
);
5009 * This function handles all "interface"-type I/O control requests. The actual
5010 * 'doing' part of this is dev_ifsioc above.
5014 * dev_ioctl - network device ioctl
5015 * @net: the applicable net namespace
5016 * @cmd: command to issue
5017 * @arg: pointer to a struct ifreq in user space
5019 * Issue ioctl functions to devices. This is normally called by the
5020 * user space syscall interfaces but can sometimes be useful for
5021 * other purposes. The return value is the return from the syscall if
5022 * positive or a negative errno code on error.
5025 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
5031 /* One special case: SIOCGIFCONF takes ifconf argument
5032 and requires shared lock, because it sleeps writing
5036 if (cmd
== SIOCGIFCONF
) {
5038 ret
= dev_ifconf(net
, (char __user
*) arg
);
5042 if (cmd
== SIOCGIFNAME
)
5043 return dev_ifname(net
, (struct ifreq __user
*)arg
);
5045 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
5048 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
5050 colon
= strchr(ifr
.ifr_name
, ':');
5055 * See which interface the caller is talking about.
5060 * These ioctl calls:
5061 * - can be done by all.
5062 * - atomic and do not require locking.
5073 dev_load(net
, ifr
.ifr_name
);
5075 ret
= dev_ifsioc_locked(net
, &ifr
, cmd
);
5080 if (copy_to_user(arg
, &ifr
,
5081 sizeof(struct ifreq
)))
5087 dev_load(net
, ifr
.ifr_name
);
5089 ret
= dev_ethtool(net
, &ifr
);
5094 if (copy_to_user(arg
, &ifr
,
5095 sizeof(struct ifreq
)))
5101 * These ioctl calls:
5102 * - require superuser power.
5103 * - require strict serialization.
5109 if (!capable(CAP_NET_ADMIN
))
5111 dev_load(net
, ifr
.ifr_name
);
5113 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5118 if (copy_to_user(arg
, &ifr
,
5119 sizeof(struct ifreq
)))
5125 * These ioctl calls:
5126 * - require superuser power.
5127 * - require strict serialization.
5128 * - do not return a value
5138 case SIOCSIFHWBROADCAST
:
5141 case SIOCBONDENSLAVE
:
5142 case SIOCBONDRELEASE
:
5143 case SIOCBONDSETHWADDR
:
5144 case SIOCBONDCHANGEACTIVE
:
5148 if (!capable(CAP_NET_ADMIN
))
5151 case SIOCBONDSLAVEINFOQUERY
:
5152 case SIOCBONDINFOQUERY
:
5153 dev_load(net
, ifr
.ifr_name
);
5155 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5160 /* Get the per device memory space. We can add this but
5161 * currently do not support it */
5163 /* Set the per device memory buffer space.
5164 * Not applicable in our case */
5169 * Unknown or private ioctl.
5172 if (cmd
== SIOCWANDEV
||
5173 (cmd
>= SIOCDEVPRIVATE
&&
5174 cmd
<= SIOCDEVPRIVATE
+ 15)) {
5175 dev_load(net
, ifr
.ifr_name
);
5177 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5179 if (!ret
&& copy_to_user(arg
, &ifr
,
5180 sizeof(struct ifreq
)))
5184 /* Take care of Wireless Extensions */
5185 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
5186 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
5193 * dev_new_index - allocate an ifindex
5194 * @net: the applicable net namespace
5196 * Returns a suitable unique value for a new device interface
5197 * number. The caller must hold the rtnl semaphore or the
5198 * dev_base_lock to be sure it remains unique.
5200 static int dev_new_index(struct net
*net
)
5206 if (!__dev_get_by_index(net
, ifindex
))
5211 /* Delayed registration/unregisteration */
5212 static LIST_HEAD(net_todo_list
);
5214 static void net_set_todo(struct net_device
*dev
)
5216 list_add_tail(&dev
->todo_list
, &net_todo_list
);
5219 static void rollback_registered_many(struct list_head
*head
)
5221 struct net_device
*dev
, *tmp
;
5223 BUG_ON(dev_boot_phase
);
5226 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
5227 /* Some devices call without registering
5228 * for initialization unwind. Remove those
5229 * devices and proceed with the remaining.
5231 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5232 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5236 list_del(&dev
->unreg_list
);
5239 dev
->dismantle
= true;
5240 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
5243 /* If device is running, close it first. */
5244 dev_close_many(head
);
5246 list_for_each_entry(dev
, head
, unreg_list
) {
5247 /* And unlink it from device chain. */
5248 unlist_netdevice(dev
);
5250 dev
->reg_state
= NETREG_UNREGISTERING
;
5255 list_for_each_entry(dev
, head
, unreg_list
) {
5256 /* Shutdown queueing discipline. */
5260 /* Notify protocols, that we are about to destroy
5261 this device. They should clean all the things.
5263 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5265 if (!dev
->rtnl_link_ops
||
5266 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5267 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
5270 * Flush the unicast and multicast chains
5275 if (dev
->netdev_ops
->ndo_uninit
)
5276 dev
->netdev_ops
->ndo_uninit(dev
);
5278 /* Notifier chain MUST detach us from master device. */
5279 WARN_ON(dev
->master
);
5281 /* Remove entries from kobject tree */
5282 netdev_unregister_kobject(dev
);
5285 /* Process any work delayed until the end of the batch */
5286 dev
= list_first_entry(head
, struct net_device
, unreg_list
);
5287 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
5291 list_for_each_entry(dev
, head
, unreg_list
)
5295 static void rollback_registered(struct net_device
*dev
)
5299 list_add(&dev
->unreg_list
, &single
);
5300 rollback_registered_many(&single
);
5304 static netdev_features_t
netdev_fix_features(struct net_device
*dev
,
5305 netdev_features_t features
)
5307 /* Fix illegal checksum combinations */
5308 if ((features
& NETIF_F_HW_CSUM
) &&
5309 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5310 netdev_warn(dev
, "mixed HW and IP checksum settings.\n");
5311 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5314 /* Fix illegal SG+CSUM combinations. */
5315 if ((features
& NETIF_F_SG
) &&
5316 !(features
& NETIF_F_ALL_CSUM
)) {
5318 "Dropping NETIF_F_SG since no checksum feature.\n");
5319 features
&= ~NETIF_F_SG
;
5322 /* TSO requires that SG is present as well. */
5323 if ((features
& NETIF_F_ALL_TSO
) && !(features
& NETIF_F_SG
)) {
5324 netdev_dbg(dev
, "Dropping TSO features since no SG feature.\n");
5325 features
&= ~NETIF_F_ALL_TSO
;
5328 /* TSO ECN requires that TSO is present as well. */
5329 if ((features
& NETIF_F_ALL_TSO
) == NETIF_F_TSO_ECN
)
5330 features
&= ~NETIF_F_TSO_ECN
;
5332 /* Software GSO depends on SG. */
5333 if ((features
& NETIF_F_GSO
) && !(features
& NETIF_F_SG
)) {
5334 netdev_dbg(dev
, "Dropping NETIF_F_GSO since no SG feature.\n");
5335 features
&= ~NETIF_F_GSO
;
5338 /* UFO needs SG and checksumming */
5339 if (features
& NETIF_F_UFO
) {
5340 /* maybe split UFO into V4 and V6? */
5341 if (!((features
& NETIF_F_GEN_CSUM
) ||
5342 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))
5343 == (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5345 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5346 features
&= ~NETIF_F_UFO
;
5349 if (!(features
& NETIF_F_SG
)) {
5351 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5352 features
&= ~NETIF_F_UFO
;
5359 int __netdev_update_features(struct net_device
*dev
)
5361 netdev_features_t features
;
5366 features
= netdev_get_wanted_features(dev
);
5368 if (dev
->netdev_ops
->ndo_fix_features
)
5369 features
= dev
->netdev_ops
->ndo_fix_features(dev
, features
);
5371 /* driver might be less strict about feature dependencies */
5372 features
= netdev_fix_features(dev
, features
);
5374 if (dev
->features
== features
)
5377 netdev_dbg(dev
, "Features changed: %pNF -> %pNF\n",
5378 &dev
->features
, &features
);
5380 if (dev
->netdev_ops
->ndo_set_features
)
5381 err
= dev
->netdev_ops
->ndo_set_features(dev
, features
);
5383 if (unlikely(err
< 0)) {
5385 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5386 err
, &features
, &dev
->features
);
5391 dev
->features
= features
;
5397 * netdev_update_features - recalculate device features
5398 * @dev: the device to check
5400 * Recalculate dev->features set and send notifications if it
5401 * has changed. Should be called after driver or hardware dependent
5402 * conditions might have changed that influence the features.
5404 void netdev_update_features(struct net_device
*dev
)
5406 if (__netdev_update_features(dev
))
5407 netdev_features_change(dev
);
5409 EXPORT_SYMBOL(netdev_update_features
);
5412 * netdev_change_features - recalculate device features
5413 * @dev: the device to check
5415 * Recalculate dev->features set and send notifications even
5416 * if they have not changed. Should be called instead of
5417 * netdev_update_features() if also dev->vlan_features might
5418 * have changed to allow the changes to be propagated to stacked
5421 void netdev_change_features(struct net_device
*dev
)
5423 __netdev_update_features(dev
);
5424 netdev_features_change(dev
);
5426 EXPORT_SYMBOL(netdev_change_features
);
5429 * netif_stacked_transfer_operstate - transfer operstate
5430 * @rootdev: the root or lower level device to transfer state from
5431 * @dev: the device to transfer operstate to
5433 * Transfer operational state from root to device. This is normally
5434 * called when a stacking relationship exists between the root
5435 * device and the device(a leaf device).
5437 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
5438 struct net_device
*dev
)
5440 if (rootdev
->operstate
== IF_OPER_DORMANT
)
5441 netif_dormant_on(dev
);
5443 netif_dormant_off(dev
);
5445 if (netif_carrier_ok(rootdev
)) {
5446 if (!netif_carrier_ok(dev
))
5447 netif_carrier_on(dev
);
5449 if (netif_carrier_ok(dev
))
5450 netif_carrier_off(dev
);
5453 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
5456 static int netif_alloc_rx_queues(struct net_device
*dev
)
5458 unsigned int i
, count
= dev
->num_rx_queues
;
5459 struct netdev_rx_queue
*rx
;
5463 rx
= kcalloc(count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5465 pr_err("netdev: Unable to allocate %u rx queues\n", count
);
5470 for (i
= 0; i
< count
; i
++)
5476 static void netdev_init_one_queue(struct net_device
*dev
,
5477 struct netdev_queue
*queue
, void *_unused
)
5479 /* Initialize queue lock */
5480 spin_lock_init(&queue
->_xmit_lock
);
5481 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
5482 queue
->xmit_lock_owner
= -1;
5483 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
5486 dql_init(&queue
->dql
, HZ
);
5490 static int netif_alloc_netdev_queues(struct net_device
*dev
)
5492 unsigned int count
= dev
->num_tx_queues
;
5493 struct netdev_queue
*tx
;
5497 tx
= kcalloc(count
, sizeof(struct netdev_queue
), GFP_KERNEL
);
5499 pr_err("netdev: Unable to allocate %u tx queues\n", count
);
5504 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5505 spin_lock_init(&dev
->tx_global_lock
);
5511 * register_netdevice - register a network device
5512 * @dev: device to register
5514 * Take a completed network device structure and add it to the kernel
5515 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5516 * chain. 0 is returned on success. A negative errno code is returned
5517 * on a failure to set up the device, or if the name is a duplicate.
5519 * Callers must hold the rtnl semaphore. You may want
5520 * register_netdev() instead of this.
5523 * The locking appears insufficient to guarantee two parallel registers
5524 * will not get the same name.
5527 int register_netdevice(struct net_device
*dev
)
5530 struct net
*net
= dev_net(dev
);
5532 BUG_ON(dev_boot_phase
);
5537 /* When net_device's are persistent, this will be fatal. */
5538 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
5541 spin_lock_init(&dev
->addr_list_lock
);
5542 netdev_set_addr_lockdep_class(dev
);
5546 ret
= dev_get_valid_name(dev
, dev
->name
);
5550 /* Init, if this function is available */
5551 if (dev
->netdev_ops
->ndo_init
) {
5552 ret
= dev
->netdev_ops
->ndo_init(dev
);
5560 dev
->ifindex
= dev_new_index(net
);
5561 if (dev
->iflink
== -1)
5562 dev
->iflink
= dev
->ifindex
;
5564 /* Transfer changeable features to wanted_features and enable
5565 * software offloads (GSO and GRO).
5567 dev
->hw_features
|= NETIF_F_SOFT_FEATURES
;
5568 dev
->features
|= NETIF_F_SOFT_FEATURES
;
5569 dev
->wanted_features
= dev
->features
& dev
->hw_features
;
5571 /* Turn on no cache copy if HW is doing checksum */
5572 if (!(dev
->flags
& IFF_LOOPBACK
)) {
5573 dev
->hw_features
|= NETIF_F_NOCACHE_COPY
;
5574 if (dev
->features
& NETIF_F_ALL_CSUM
) {
5575 dev
->wanted_features
|= NETIF_F_NOCACHE_COPY
;
5576 dev
->features
|= NETIF_F_NOCACHE_COPY
;
5580 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5582 dev
->vlan_features
|= NETIF_F_HIGHDMA
;
5584 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
5585 ret
= notifier_to_errno(ret
);
5589 ret
= netdev_register_kobject(dev
);
5592 dev
->reg_state
= NETREG_REGISTERED
;
5594 __netdev_update_features(dev
);
5597 * Default initial state at registry is that the
5598 * device is present.
5601 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5603 dev_init_scheduler(dev
);
5605 list_netdevice(dev
);
5607 /* Notify protocols, that a new device appeared. */
5608 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5609 ret
= notifier_to_errno(ret
);
5611 rollback_registered(dev
);
5612 dev
->reg_state
= NETREG_UNREGISTERED
;
5615 * Prevent userspace races by waiting until the network
5616 * device is fully setup before sending notifications.
5618 if (!dev
->rtnl_link_ops
||
5619 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5620 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5626 if (dev
->netdev_ops
->ndo_uninit
)
5627 dev
->netdev_ops
->ndo_uninit(dev
);
5630 EXPORT_SYMBOL(register_netdevice
);
5633 * init_dummy_netdev - init a dummy network device for NAPI
5634 * @dev: device to init
5636 * This takes a network device structure and initialize the minimum
5637 * amount of fields so it can be used to schedule NAPI polls without
5638 * registering a full blown interface. This is to be used by drivers
5639 * that need to tie several hardware interfaces to a single NAPI
5640 * poll scheduler due to HW limitations.
5642 int init_dummy_netdev(struct net_device
*dev
)
5644 /* Clear everything. Note we don't initialize spinlocks
5645 * are they aren't supposed to be taken by any of the
5646 * NAPI code and this dummy netdev is supposed to be
5647 * only ever used for NAPI polls
5649 memset(dev
, 0, sizeof(struct net_device
));
5651 /* make sure we BUG if trying to hit standard
5652 * register/unregister code path
5654 dev
->reg_state
= NETREG_DUMMY
;
5656 /* NAPI wants this */
5657 INIT_LIST_HEAD(&dev
->napi_list
);
5659 /* a dummy interface is started by default */
5660 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5661 set_bit(__LINK_STATE_START
, &dev
->state
);
5663 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5664 * because users of this 'device' dont need to change
5670 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
5674 * register_netdev - register a network device
5675 * @dev: device to register
5677 * Take a completed network device structure and add it to the kernel
5678 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5679 * chain. 0 is returned on success. A negative errno code is returned
5680 * on a failure to set up the device, or if the name is a duplicate.
5682 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5683 * and expands the device name if you passed a format string to
5686 int register_netdev(struct net_device
*dev
)
5691 err
= register_netdevice(dev
);
5695 EXPORT_SYMBOL(register_netdev
);
5697 int netdev_refcnt_read(const struct net_device
*dev
)
5701 for_each_possible_cpu(i
)
5702 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
5705 EXPORT_SYMBOL(netdev_refcnt_read
);
5708 * netdev_wait_allrefs - wait until all references are gone.
5710 * This is called when unregistering network devices.
5712 * Any protocol or device that holds a reference should register
5713 * for netdevice notification, and cleanup and put back the
5714 * reference if they receive an UNREGISTER event.
5715 * We can get stuck here if buggy protocols don't correctly
5718 static void netdev_wait_allrefs(struct net_device
*dev
)
5720 unsigned long rebroadcast_time
, warning_time
;
5723 linkwatch_forget_dev(dev
);
5725 rebroadcast_time
= warning_time
= jiffies
;
5726 refcnt
= netdev_refcnt_read(dev
);
5728 while (refcnt
!= 0) {
5729 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
5732 /* Rebroadcast unregister notification */
5733 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5734 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5735 * should have already handle it the first time */
5737 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
5739 /* We must not have linkwatch events
5740 * pending on unregister. If this
5741 * happens, we simply run the queue
5742 * unscheduled, resulting in a noop
5745 linkwatch_run_queue();
5750 rebroadcast_time
= jiffies
;
5755 refcnt
= netdev_refcnt_read(dev
);
5757 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
5758 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5760 warning_time
= jiffies
;
5769 * register_netdevice(x1);
5770 * register_netdevice(x2);
5772 * unregister_netdevice(y1);
5773 * unregister_netdevice(y2);
5779 * We are invoked by rtnl_unlock().
5780 * This allows us to deal with problems:
5781 * 1) We can delete sysfs objects which invoke hotplug
5782 * without deadlocking with linkwatch via keventd.
5783 * 2) Since we run with the RTNL semaphore not held, we can sleep
5784 * safely in order to wait for the netdev refcnt to drop to zero.
5786 * We must not return until all unregister events added during
5787 * the interval the lock was held have been completed.
5789 void netdev_run_todo(void)
5791 struct list_head list
;
5793 /* Snapshot list, allow later requests */
5794 list_replace_init(&net_todo_list
, &list
);
5798 /* Wait for rcu callbacks to finish before attempting to drain
5799 * the device list. This usually avoids a 250ms wait.
5801 if (!list_empty(&list
))
5804 while (!list_empty(&list
)) {
5805 struct net_device
*dev
5806 = list_first_entry(&list
, struct net_device
, todo_list
);
5807 list_del(&dev
->todo_list
);
5809 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
5810 pr_err("network todo '%s' but state %d\n",
5811 dev
->name
, dev
->reg_state
);
5816 dev
->reg_state
= NETREG_UNREGISTERED
;
5818 on_each_cpu(flush_backlog
, dev
, 1);
5820 netdev_wait_allrefs(dev
);
5823 BUG_ON(netdev_refcnt_read(dev
));
5824 WARN_ON(rcu_access_pointer(dev
->ip_ptr
));
5825 WARN_ON(rcu_access_pointer(dev
->ip6_ptr
));
5826 WARN_ON(dev
->dn_ptr
);
5828 if (dev
->destructor
)
5829 dev
->destructor(dev
);
5831 /* Free network device */
5832 kobject_put(&dev
->dev
.kobj
);
5836 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5837 * fields in the same order, with only the type differing.
5839 void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
5840 const struct net_device_stats
*netdev_stats
)
5842 #if BITS_PER_LONG == 64
5843 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
5844 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
5846 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
5847 const unsigned long *src
= (const unsigned long *)netdev_stats
;
5848 u64
*dst
= (u64
*)stats64
;
5850 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
5851 sizeof(*stats64
) / sizeof(u64
));
5852 for (i
= 0; i
< n
; i
++)
5856 EXPORT_SYMBOL(netdev_stats_to_stats64
);
5859 * dev_get_stats - get network device statistics
5860 * @dev: device to get statistics from
5861 * @storage: place to store stats
5863 * Get network statistics from device. Return @storage.
5864 * The device driver may provide its own method by setting
5865 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5866 * otherwise the internal statistics structure is used.
5868 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
5869 struct rtnl_link_stats64
*storage
)
5871 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5873 if (ops
->ndo_get_stats64
) {
5874 memset(storage
, 0, sizeof(*storage
));
5875 ops
->ndo_get_stats64(dev
, storage
);
5876 } else if (ops
->ndo_get_stats
) {
5877 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
5879 netdev_stats_to_stats64(storage
, &dev
->stats
);
5881 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
5884 EXPORT_SYMBOL(dev_get_stats
);
5886 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
5888 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
5890 #ifdef CONFIG_NET_CLS_ACT
5893 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
5896 netdev_init_one_queue(dev
, queue
, NULL
);
5897 queue
->qdisc
= &noop_qdisc
;
5898 queue
->qdisc_sleeping
= &noop_qdisc
;
5899 rcu_assign_pointer(dev
->ingress_queue
, queue
);
5905 * alloc_netdev_mqs - allocate network device
5906 * @sizeof_priv: size of private data to allocate space for
5907 * @name: device name format string
5908 * @setup: callback to initialize device
5909 * @txqs: the number of TX subqueues to allocate
5910 * @rxqs: the number of RX subqueues to allocate
5912 * Allocates a struct net_device with private data area for driver use
5913 * and performs basic initialization. Also allocates subquue structs
5914 * for each queue on the device.
5916 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
5917 void (*setup
)(struct net_device
*),
5918 unsigned int txqs
, unsigned int rxqs
)
5920 struct net_device
*dev
;
5922 struct net_device
*p
;
5924 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
5927 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
5933 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
5938 alloc_size
= sizeof(struct net_device
);
5940 /* ensure 32-byte alignment of private area */
5941 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
5942 alloc_size
+= sizeof_priv
;
5944 /* ensure 32-byte alignment of whole construct */
5945 alloc_size
+= NETDEV_ALIGN
- 1;
5947 p
= kzalloc(alloc_size
, GFP_KERNEL
);
5949 pr_err("alloc_netdev: Unable to allocate device\n");
5953 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
5954 dev
->padded
= (char *)dev
- (char *)p
;
5956 dev
->pcpu_refcnt
= alloc_percpu(int);
5957 if (!dev
->pcpu_refcnt
)
5960 if (dev_addr_init(dev
))
5966 dev_net_set(dev
, &init_net
);
5968 dev
->gso_max_size
= GSO_MAX_SIZE
;
5970 INIT_LIST_HEAD(&dev
->napi_list
);
5971 INIT_LIST_HEAD(&dev
->unreg_list
);
5972 INIT_LIST_HEAD(&dev
->link_watch_list
);
5973 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
5976 dev
->num_tx_queues
= txqs
;
5977 dev
->real_num_tx_queues
= txqs
;
5978 if (netif_alloc_netdev_queues(dev
))
5982 dev
->num_rx_queues
= rxqs
;
5983 dev
->real_num_rx_queues
= rxqs
;
5984 if (netif_alloc_rx_queues(dev
))
5988 strcpy(dev
->name
, name
);
5989 dev
->group
= INIT_NETDEV_GROUP
;
5997 free_percpu(dev
->pcpu_refcnt
);
6007 EXPORT_SYMBOL(alloc_netdev_mqs
);
6010 * free_netdev - free network device
6013 * This function does the last stage of destroying an allocated device
6014 * interface. The reference to the device object is released.
6015 * If this is the last reference then it will be freed.
6017 void free_netdev(struct net_device
*dev
)
6019 struct napi_struct
*p
, *n
;
6021 release_net(dev_net(dev
));
6028 kfree(rcu_dereference_protected(dev
->ingress_queue
, 1));
6030 /* Flush device addresses */
6031 dev_addr_flush(dev
);
6033 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
6036 free_percpu(dev
->pcpu_refcnt
);
6037 dev
->pcpu_refcnt
= NULL
;
6039 /* Compatibility with error handling in drivers */
6040 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
6041 kfree((char *)dev
- dev
->padded
);
6045 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
6046 dev
->reg_state
= NETREG_RELEASED
;
6048 /* will free via device release */
6049 put_device(&dev
->dev
);
6051 EXPORT_SYMBOL(free_netdev
);
6054 * synchronize_net - Synchronize with packet receive processing
6056 * Wait for packets currently being received to be done.
6057 * Does not block later packets from starting.
6059 void synchronize_net(void)
6062 if (rtnl_is_locked())
6063 synchronize_rcu_expedited();
6067 EXPORT_SYMBOL(synchronize_net
);
6070 * unregister_netdevice_queue - remove device from the kernel
6074 * This function shuts down a device interface and removes it
6075 * from the kernel tables.
6076 * If head not NULL, device is queued to be unregistered later.
6078 * Callers must hold the rtnl semaphore. You may want
6079 * unregister_netdev() instead of this.
6082 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
6087 list_move_tail(&dev
->unreg_list
, head
);
6089 rollback_registered(dev
);
6090 /* Finish processing unregister after unlock */
6094 EXPORT_SYMBOL(unregister_netdevice_queue
);
6097 * unregister_netdevice_many - unregister many devices
6098 * @head: list of devices
6100 void unregister_netdevice_many(struct list_head
*head
)
6102 struct net_device
*dev
;
6104 if (!list_empty(head
)) {
6105 rollback_registered_many(head
);
6106 list_for_each_entry(dev
, head
, unreg_list
)
6110 EXPORT_SYMBOL(unregister_netdevice_many
);
6113 * unregister_netdev - remove device from the kernel
6116 * This function shuts down a device interface and removes it
6117 * from the kernel tables.
6119 * This is just a wrapper for unregister_netdevice that takes
6120 * the rtnl semaphore. In general you want to use this and not
6121 * unregister_netdevice.
6123 void unregister_netdev(struct net_device
*dev
)
6126 unregister_netdevice(dev
);
6129 EXPORT_SYMBOL(unregister_netdev
);
6132 * dev_change_net_namespace - move device to different nethost namespace
6134 * @net: network namespace
6135 * @pat: If not NULL name pattern to try if the current device name
6136 * is already taken in the destination network namespace.
6138 * This function shuts down a device interface and moves it
6139 * to a new network namespace. On success 0 is returned, on
6140 * a failure a netagive errno code is returned.
6142 * Callers must hold the rtnl semaphore.
6145 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
6151 /* Don't allow namespace local devices to be moved. */
6153 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6156 /* Ensure the device has been registrered */
6158 if (dev
->reg_state
!= NETREG_REGISTERED
)
6161 /* Get out if there is nothing todo */
6163 if (net_eq(dev_net(dev
), net
))
6166 /* Pick the destination device name, and ensure
6167 * we can use it in the destination network namespace.
6170 if (__dev_get_by_name(net
, dev
->name
)) {
6171 /* We get here if we can't use the current device name */
6174 if (dev_get_valid_name(dev
, pat
) < 0)
6179 * And now a mini version of register_netdevice unregister_netdevice.
6182 /* If device is running close it first. */
6185 /* And unlink it from device chain */
6187 unlist_netdevice(dev
);
6191 /* Shutdown queueing discipline. */
6194 /* Notify protocols, that we are about to destroy
6195 this device. They should clean all the things.
6197 Note that dev->reg_state stays at NETREG_REGISTERED.
6198 This is wanted because this way 8021q and macvlan know
6199 the device is just moving and can keep their slaves up.
6201 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6202 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
6203 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
6206 * Flush the unicast and multicast chains
6211 /* Actually switch the network namespace */
6212 dev_net_set(dev
, net
);
6214 /* If there is an ifindex conflict assign a new one */
6215 if (__dev_get_by_index(net
, dev
->ifindex
)) {
6216 int iflink
= (dev
->iflink
== dev
->ifindex
);
6217 dev
->ifindex
= dev_new_index(net
);
6219 dev
->iflink
= dev
->ifindex
;
6222 /* Fixup kobjects */
6223 err
= device_rename(&dev
->dev
, dev
->name
);
6226 /* Add the device back in the hashes */
6227 list_netdevice(dev
);
6229 /* Notify protocols, that a new device appeared. */
6230 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
6233 * Prevent userspace races by waiting until the network
6234 * device is fully setup before sending notifications.
6236 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
6243 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
6245 static int dev_cpu_callback(struct notifier_block
*nfb
,
6246 unsigned long action
,
6249 struct sk_buff
**list_skb
;
6250 struct sk_buff
*skb
;
6251 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
6252 struct softnet_data
*sd
, *oldsd
;
6254 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
6257 local_irq_disable();
6258 cpu
= smp_processor_id();
6259 sd
= &per_cpu(softnet_data
, cpu
);
6260 oldsd
= &per_cpu(softnet_data
, oldcpu
);
6262 /* Find end of our completion_queue. */
6263 list_skb
= &sd
->completion_queue
;
6265 list_skb
= &(*list_skb
)->next
;
6266 /* Append completion queue from offline CPU. */
6267 *list_skb
= oldsd
->completion_queue
;
6268 oldsd
->completion_queue
= NULL
;
6270 /* Append output queue from offline CPU. */
6271 if (oldsd
->output_queue
) {
6272 *sd
->output_queue_tailp
= oldsd
->output_queue
;
6273 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
6274 oldsd
->output_queue
= NULL
;
6275 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
6277 /* Append NAPI poll list from offline CPU. */
6278 if (!list_empty(&oldsd
->poll_list
)) {
6279 list_splice_init(&oldsd
->poll_list
, &sd
->poll_list
);
6280 raise_softirq_irqoff(NET_RX_SOFTIRQ
);
6283 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
6286 /* Process offline CPU's input_pkt_queue */
6287 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
6289 input_queue_head_incr(oldsd
);
6291 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
6293 input_queue_head_incr(oldsd
);
6301 * netdev_increment_features - increment feature set by one
6302 * @all: current feature set
6303 * @one: new feature set
6304 * @mask: mask feature set
6306 * Computes a new feature set after adding a device with feature set
6307 * @one to the master device with current feature set @all. Will not
6308 * enable anything that is off in @mask. Returns the new feature set.
6310 netdev_features_t
netdev_increment_features(netdev_features_t all
,
6311 netdev_features_t one
, netdev_features_t mask
)
6313 if (mask
& NETIF_F_GEN_CSUM
)
6314 mask
|= NETIF_F_ALL_CSUM
;
6315 mask
|= NETIF_F_VLAN_CHALLENGED
;
6317 all
|= one
& (NETIF_F_ONE_FOR_ALL
|NETIF_F_ALL_CSUM
) & mask
;
6318 all
&= one
| ~NETIF_F_ALL_FOR_ALL
;
6320 /* If one device supports hw checksumming, set for all. */
6321 if (all
& NETIF_F_GEN_CSUM
)
6322 all
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_GEN_CSUM
);
6326 EXPORT_SYMBOL(netdev_increment_features
);
6328 static struct hlist_head
*netdev_create_hash(void)
6331 struct hlist_head
*hash
;
6333 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
6335 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
6336 INIT_HLIST_HEAD(&hash
[i
]);
6341 /* Initialize per network namespace state */
6342 static int __net_init
netdev_init(struct net
*net
)
6344 INIT_LIST_HEAD(&net
->dev_base_head
);
6346 net
->dev_name_head
= netdev_create_hash();
6347 if (net
->dev_name_head
== NULL
)
6350 net
->dev_index_head
= netdev_create_hash();
6351 if (net
->dev_index_head
== NULL
)
6357 kfree(net
->dev_name_head
);
6363 * netdev_drivername - network driver for the device
6364 * @dev: network device
6366 * Determine network driver for device.
6368 const char *netdev_drivername(const struct net_device
*dev
)
6370 const struct device_driver
*driver
;
6371 const struct device
*parent
;
6372 const char *empty
= "";
6374 parent
= dev
->dev
.parent
;
6378 driver
= parent
->driver
;
6379 if (driver
&& driver
->name
)
6380 return driver
->name
;
6384 int __netdev_printk(const char *level
, const struct net_device
*dev
,
6385 struct va_format
*vaf
)
6389 if (dev
&& dev
->dev
.parent
)
6390 r
= dev_printk(level
, dev
->dev
.parent
, "%s: %pV",
6391 netdev_name(dev
), vaf
);
6393 r
= printk("%s%s: %pV", level
, netdev_name(dev
), vaf
);
6395 r
= printk("%s(NULL net_device): %pV", level
, vaf
);
6399 EXPORT_SYMBOL(__netdev_printk
);
6401 int netdev_printk(const char *level
, const struct net_device
*dev
,
6402 const char *format
, ...)
6404 struct va_format vaf
;
6408 va_start(args
, format
);
6413 r
= __netdev_printk(level
, dev
, &vaf
);
6418 EXPORT_SYMBOL(netdev_printk
);
6420 #define define_netdev_printk_level(func, level) \
6421 int func(const struct net_device *dev, const char *fmt, ...) \
6424 struct va_format vaf; \
6427 va_start(args, fmt); \
6432 r = __netdev_printk(level, dev, &vaf); \
6437 EXPORT_SYMBOL(func);
6439 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
6440 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
6441 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
6442 define_netdev_printk_level(netdev_err
, KERN_ERR
);
6443 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
6444 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
6445 define_netdev_printk_level(netdev_info
, KERN_INFO
);
6447 static void __net_exit
netdev_exit(struct net
*net
)
6449 kfree(net
->dev_name_head
);
6450 kfree(net
->dev_index_head
);
6453 static struct pernet_operations __net_initdata netdev_net_ops
= {
6454 .init
= netdev_init
,
6455 .exit
= netdev_exit
,
6458 static void __net_exit
default_device_exit(struct net
*net
)
6460 struct net_device
*dev
, *aux
;
6462 * Push all migratable network devices back to the
6463 * initial network namespace
6466 for_each_netdev_safe(net
, dev
, aux
) {
6468 char fb_name
[IFNAMSIZ
];
6470 /* Ignore unmoveable devices (i.e. loopback) */
6471 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6474 /* Leave virtual devices for the generic cleanup */
6475 if (dev
->rtnl_link_ops
)
6478 /* Push remaining network devices to init_net */
6479 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
6480 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
6482 pr_emerg("%s: failed to move %s to init_net: %d\n",
6483 __func__
, dev
->name
, err
);
6490 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
6492 /* At exit all network devices most be removed from a network
6493 * namespace. Do this in the reverse order of registration.
6494 * Do this across as many network namespaces as possible to
6495 * improve batching efficiency.
6497 struct net_device
*dev
;
6499 LIST_HEAD(dev_kill_list
);
6502 list_for_each_entry(net
, net_list
, exit_list
) {
6503 for_each_netdev_reverse(net
, dev
) {
6504 if (dev
->rtnl_link_ops
)
6505 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
6507 unregister_netdevice_queue(dev
, &dev_kill_list
);
6510 unregister_netdevice_many(&dev_kill_list
);
6511 list_del(&dev_kill_list
);
6515 static struct pernet_operations __net_initdata default_device_ops
= {
6516 .exit
= default_device_exit
,
6517 .exit_batch
= default_device_exit_batch
,
6521 * Initialize the DEV module. At boot time this walks the device list and
6522 * unhooks any devices that fail to initialise (normally hardware not
6523 * present) and leaves us with a valid list of present and active devices.
6528 * This is called single threaded during boot, so no need
6529 * to take the rtnl semaphore.
6531 static int __init
net_dev_init(void)
6533 int i
, rc
= -ENOMEM
;
6535 BUG_ON(!dev_boot_phase
);
6537 if (dev_proc_init())
6540 if (netdev_kobject_init())
6543 INIT_LIST_HEAD(&ptype_all
);
6544 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
6545 INIT_LIST_HEAD(&ptype_base
[i
]);
6547 if (register_pernet_subsys(&netdev_net_ops
))
6551 * Initialise the packet receive queues.
6554 for_each_possible_cpu(i
) {
6555 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
6557 memset(sd
, 0, sizeof(*sd
));
6558 skb_queue_head_init(&sd
->input_pkt_queue
);
6559 skb_queue_head_init(&sd
->process_queue
);
6560 sd
->completion_queue
= NULL
;
6561 INIT_LIST_HEAD(&sd
->poll_list
);
6562 sd
->output_queue
= NULL
;
6563 sd
->output_queue_tailp
= &sd
->output_queue
;
6565 sd
->csd
.func
= rps_trigger_softirq
;
6571 sd
->backlog
.poll
= process_backlog
;
6572 sd
->backlog
.weight
= weight_p
;
6573 sd
->backlog
.gro_list
= NULL
;
6574 sd
->backlog
.gro_count
= 0;
6579 /* The loopback device is special if any other network devices
6580 * is present in a network namespace the loopback device must
6581 * be present. Since we now dynamically allocate and free the
6582 * loopback device ensure this invariant is maintained by
6583 * keeping the loopback device as the first device on the
6584 * list of network devices. Ensuring the loopback devices
6585 * is the first device that appears and the last network device
6588 if (register_pernet_device(&loopback_net_ops
))
6591 if (register_pernet_device(&default_device_ops
))
6594 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
6595 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
6597 hotcpu_notifier(dev_cpu_callback
, 0);
6605 subsys_initcall(net_dev_init
);
6607 static int __init
initialize_hashrnd(void)
6609 get_random_bytes(&hashrnd
, sizeof(hashrnd
));
6613 late_initcall_sync(initialize_hashrnd
);