2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <net/net_namespace.h>
99 #include <linux/rtnetlink.h>
100 #include <linux/proc_fs.h>
101 #include <linux/seq_file.h>
102 #include <linux/stat.h>
104 #include <net/pkt_sched.h>
105 #include <net/checksum.h>
106 #include <net/xfrm.h>
107 #include <linux/highmem.h>
108 #include <linux/init.h>
109 #include <linux/kmod.h>
110 #include <linux/module.h>
111 #include <linux/netpoll.h>
112 #include <linux/rcupdate.h>
113 #include <linux/delay.h>
114 #include <net/wext.h>
115 #include <net/iw_handler.h>
116 #include <asm/current.h>
117 #include <linux/audit.h>
118 #include <linux/dmaengine.h>
119 #include <linux/err.h>
120 #include <linux/ctype.h>
121 #include <linux/if_arp.h>
122 #include <linux/if_vlan.h>
123 #include <linux/ip.h>
125 #include <linux/ipv6.h>
126 #include <linux/in.h>
127 #include <linux/jhash.h>
128 #include <linux/random.h>
129 #include <trace/events/napi.h>
130 #include <trace/events/net.h>
131 #include <trace/events/skb.h>
132 #include <linux/pci.h>
133 #include <linux/inetdevice.h>
134 #include <linux/cpu_rmap.h>
135 #include <linux/net_tstamp.h>
136 #include <linux/static_key.h>
137 #include <net/flow_keys.h>
139 #include "net-sysfs.h"
141 /* Instead of increasing this, you should create a hash table. */
142 #define MAX_GRO_SKBS 8
144 /* This should be increased if a protocol with a bigger head is added. */
145 #define GRO_MAX_HEAD (MAX_HEADER + 128)
148 * The list of packet types we will receive (as opposed to discard)
149 * and the routines to invoke.
151 * Why 16. Because with 16 the only overlap we get on a hash of the
152 * low nibble of the protocol value is RARP/SNAP/X.25.
154 * NOTE: That is no longer true with the addition of VLAN tags. Not
155 * sure which should go first, but I bet it won't make much
156 * difference if we are running VLANs. The good news is that
157 * this protocol won't be in the list unless compiled in, so
158 * the average user (w/out VLANs) will not be adversely affected.
175 #define PTYPE_HASH_SIZE (16)
176 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
178 static DEFINE_SPINLOCK(ptype_lock
);
179 static struct list_head ptype_base
[PTYPE_HASH_SIZE
] __read_mostly
;
180 static struct list_head ptype_all __read_mostly
; /* Taps */
183 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
186 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
188 * Writers must hold the rtnl semaphore while they loop through the
189 * dev_base_head list, and hold dev_base_lock for writing when they do the
190 * actual updates. This allows pure readers to access the list even
191 * while a writer is preparing to update it.
193 * To put it another way, dev_base_lock is held for writing only to
194 * protect against pure readers; the rtnl semaphore provides the
195 * protection against other writers.
197 * See, for example usages, register_netdevice() and
198 * unregister_netdevice(), which must be called with the rtnl
201 DEFINE_RWLOCK(dev_base_lock
);
202 EXPORT_SYMBOL(dev_base_lock
);
204 static inline void dev_base_seq_inc(struct net
*net
)
206 while (++net
->dev_base_seq
== 0);
209 static inline struct hlist_head
*dev_name_hash(struct net
*net
, const char *name
)
211 unsigned int hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
213 return &net
->dev_name_head
[hash_32(hash
, NETDEV_HASHBITS
)];
216 static inline struct hlist_head
*dev_index_hash(struct net
*net
, int ifindex
)
218 return &net
->dev_index_head
[ifindex
& (NETDEV_HASHENTRIES
- 1)];
221 static inline void rps_lock(struct softnet_data
*sd
)
224 spin_lock(&sd
->input_pkt_queue
.lock
);
228 static inline void rps_unlock(struct softnet_data
*sd
)
231 spin_unlock(&sd
->input_pkt_queue
.lock
);
235 /* Device list insertion */
236 static int list_netdevice(struct net_device
*dev
)
238 struct net
*net
= dev_net(dev
);
242 write_lock_bh(&dev_base_lock
);
243 list_add_tail_rcu(&dev
->dev_list
, &net
->dev_base_head
);
244 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
245 hlist_add_head_rcu(&dev
->index_hlist
,
246 dev_index_hash(net
, dev
->ifindex
));
247 write_unlock_bh(&dev_base_lock
);
249 dev_base_seq_inc(net
);
254 /* Device list removal
255 * caller must respect a RCU grace period before freeing/reusing dev
257 static void unlist_netdevice(struct net_device
*dev
)
261 /* Unlink dev from the device chain */
262 write_lock_bh(&dev_base_lock
);
263 list_del_rcu(&dev
->dev_list
);
264 hlist_del_rcu(&dev
->name_hlist
);
265 hlist_del_rcu(&dev
->index_hlist
);
266 write_unlock_bh(&dev_base_lock
);
268 dev_base_seq_inc(dev_net(dev
));
275 static RAW_NOTIFIER_HEAD(netdev_chain
);
278 * Device drivers call our routines to queue packets here. We empty the
279 * queue in the local softnet handler.
282 DEFINE_PER_CPU_ALIGNED(struct softnet_data
, softnet_data
);
283 EXPORT_PER_CPU_SYMBOL(softnet_data
);
285 #ifdef CONFIG_LOCKDEP
287 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
288 * according to dev->type
290 static const unsigned short netdev_lock_type
[] =
291 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
292 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
293 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
294 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
295 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
296 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
297 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
298 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
299 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
300 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
301 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
302 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
303 ARPHRD_FCFABRIC
, ARPHRD_IEEE80211
, ARPHRD_IEEE80211_PRISM
,
304 ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_PHONET
, ARPHRD_PHONET_PIPE
,
305 ARPHRD_IEEE802154
, ARPHRD_VOID
, ARPHRD_NONE
};
307 static const char *const netdev_lock_name
[] =
308 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
309 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
310 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
311 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
312 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
313 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
314 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
315 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
316 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
317 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
318 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
319 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
320 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
321 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
322 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
324 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
325 static struct lock_class_key netdev_addr_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
327 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
331 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
332 if (netdev_lock_type
[i
] == dev_type
)
334 /* the last key is used by default */
335 return ARRAY_SIZE(netdev_lock_type
) - 1;
338 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
339 unsigned short dev_type
)
343 i
= netdev_lock_pos(dev_type
);
344 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
345 netdev_lock_name
[i
]);
348 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
352 i
= netdev_lock_pos(dev
->type
);
353 lockdep_set_class_and_name(&dev
->addr_list_lock
,
354 &netdev_addr_lock_key
[i
],
355 netdev_lock_name
[i
]);
358 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
359 unsigned short dev_type
)
362 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
367 /*******************************************************************************
369 Protocol management and registration routines
371 *******************************************************************************/
374 * Add a protocol ID to the list. Now that the input handler is
375 * smarter we can dispense with all the messy stuff that used to be
378 * BEWARE!!! Protocol handlers, mangling input packets,
379 * MUST BE last in hash buckets and checking protocol handlers
380 * MUST start from promiscuous ptype_all chain in net_bh.
381 * It is true now, do not change it.
382 * Explanation follows: if protocol handler, mangling packet, will
383 * be the first on list, it is not able to sense, that packet
384 * is cloned and should be copied-on-write, so that it will
385 * change it and subsequent readers will get broken packet.
389 static inline struct list_head
*ptype_head(const struct packet_type
*pt
)
391 if (pt
->type
== htons(ETH_P_ALL
))
394 return &ptype_base
[ntohs(pt
->type
) & PTYPE_HASH_MASK
];
398 * dev_add_pack - add packet handler
399 * @pt: packet type declaration
401 * Add a protocol handler to the networking stack. The passed &packet_type
402 * is linked into kernel lists and may not be freed until it has been
403 * removed from the kernel lists.
405 * This call does not sleep therefore it can not
406 * guarantee all CPU's that are in middle of receiving packets
407 * will see the new packet type (until the next received packet).
410 void dev_add_pack(struct packet_type
*pt
)
412 struct list_head
*head
= ptype_head(pt
);
414 spin_lock(&ptype_lock
);
415 list_add_rcu(&pt
->list
, head
);
416 spin_unlock(&ptype_lock
);
418 EXPORT_SYMBOL(dev_add_pack
);
421 * __dev_remove_pack - remove packet handler
422 * @pt: packet type declaration
424 * Remove a protocol handler that was previously added to the kernel
425 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
426 * from the kernel lists and can be freed or reused once this function
429 * The packet type might still be in use by receivers
430 * and must not be freed until after all the CPU's have gone
431 * through a quiescent state.
433 void __dev_remove_pack(struct packet_type
*pt
)
435 struct list_head
*head
= ptype_head(pt
);
436 struct packet_type
*pt1
;
438 spin_lock(&ptype_lock
);
440 list_for_each_entry(pt1
, head
, list
) {
442 list_del_rcu(&pt
->list
);
447 pr_warn("dev_remove_pack: %p not found\n", pt
);
449 spin_unlock(&ptype_lock
);
451 EXPORT_SYMBOL(__dev_remove_pack
);
454 * dev_remove_pack - remove packet handler
455 * @pt: packet type declaration
457 * Remove a protocol handler that was previously added to the kernel
458 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
459 * from the kernel lists and can be freed or reused once this function
462 * This call sleeps to guarantee that no CPU is looking at the packet
465 void dev_remove_pack(struct packet_type
*pt
)
467 __dev_remove_pack(pt
);
471 EXPORT_SYMBOL(dev_remove_pack
);
473 /******************************************************************************
475 Device Boot-time Settings Routines
477 *******************************************************************************/
479 /* Boot time configuration table */
480 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
483 * netdev_boot_setup_add - add new setup entry
484 * @name: name of the device
485 * @map: configured settings for the device
487 * Adds new setup entry to the dev_boot_setup list. The function
488 * returns 0 on error and 1 on success. This is a generic routine to
491 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
493 struct netdev_boot_setup
*s
;
497 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
498 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
499 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
500 strlcpy(s
[i
].name
, name
, IFNAMSIZ
);
501 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
506 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
510 * netdev_boot_setup_check - check boot time settings
511 * @dev: the netdevice
513 * Check boot time settings for the device.
514 * The found settings are set for the device to be used
515 * later in the device probing.
516 * Returns 0 if no settings found, 1 if they are.
518 int netdev_boot_setup_check(struct net_device
*dev
)
520 struct netdev_boot_setup
*s
= dev_boot_setup
;
523 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
524 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
525 !strcmp(dev
->name
, s
[i
].name
)) {
526 dev
->irq
= s
[i
].map
.irq
;
527 dev
->base_addr
= s
[i
].map
.base_addr
;
528 dev
->mem_start
= s
[i
].map
.mem_start
;
529 dev
->mem_end
= s
[i
].map
.mem_end
;
535 EXPORT_SYMBOL(netdev_boot_setup_check
);
539 * netdev_boot_base - get address from boot time settings
540 * @prefix: prefix for network device
541 * @unit: id for network device
543 * Check boot time settings for the base address of device.
544 * The found settings are set for the device to be used
545 * later in the device probing.
546 * Returns 0 if no settings found.
548 unsigned long netdev_boot_base(const char *prefix
, int unit
)
550 const struct netdev_boot_setup
*s
= dev_boot_setup
;
554 sprintf(name
, "%s%d", prefix
, unit
);
557 * If device already registered then return base of 1
558 * to indicate not to probe for this interface
560 if (__dev_get_by_name(&init_net
, name
))
563 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
564 if (!strcmp(name
, s
[i
].name
))
565 return s
[i
].map
.base_addr
;
570 * Saves at boot time configured settings for any netdevice.
572 int __init
netdev_boot_setup(char *str
)
577 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
582 memset(&map
, 0, sizeof(map
));
586 map
.base_addr
= ints
[2];
588 map
.mem_start
= ints
[3];
590 map
.mem_end
= ints
[4];
592 /* Add new entry to the list */
593 return netdev_boot_setup_add(str
, &map
);
596 __setup("netdev=", netdev_boot_setup
);
598 /*******************************************************************************
600 Device Interface Subroutines
602 *******************************************************************************/
605 * __dev_get_by_name - find a device by its name
606 * @net: the applicable net namespace
607 * @name: name to find
609 * Find an interface by name. Must be called under RTNL semaphore
610 * or @dev_base_lock. If the name is found a pointer to the device
611 * is returned. If the name is not found then %NULL is returned. The
612 * reference counters are not incremented so the caller must be
613 * careful with locks.
616 struct net_device
*__dev_get_by_name(struct net
*net
, const char *name
)
618 struct hlist_node
*p
;
619 struct net_device
*dev
;
620 struct hlist_head
*head
= dev_name_hash(net
, name
);
622 hlist_for_each_entry(dev
, p
, head
, name_hlist
)
623 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
628 EXPORT_SYMBOL(__dev_get_by_name
);
631 * dev_get_by_name_rcu - find a device by its name
632 * @net: the applicable net namespace
633 * @name: name to find
635 * Find an interface by name.
636 * If the name is found a pointer to the device is returned.
637 * If the name is not found then %NULL is returned.
638 * The reference counters are not incremented so the caller must be
639 * careful with locks. The caller must hold RCU lock.
642 struct net_device
*dev_get_by_name_rcu(struct net
*net
, const char *name
)
644 struct hlist_node
*p
;
645 struct net_device
*dev
;
646 struct hlist_head
*head
= dev_name_hash(net
, name
);
648 hlist_for_each_entry_rcu(dev
, p
, head
, name_hlist
)
649 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
654 EXPORT_SYMBOL(dev_get_by_name_rcu
);
657 * dev_get_by_name - find a device by its name
658 * @net: the applicable net namespace
659 * @name: name to find
661 * Find an interface by name. This can be called from any
662 * context and does its own locking. The returned handle has
663 * the usage count incremented and the caller must use dev_put() to
664 * release it when it is no longer needed. %NULL is returned if no
665 * matching device is found.
668 struct net_device
*dev_get_by_name(struct net
*net
, const char *name
)
670 struct net_device
*dev
;
673 dev
= dev_get_by_name_rcu(net
, name
);
679 EXPORT_SYMBOL(dev_get_by_name
);
682 * __dev_get_by_index - find a device by its ifindex
683 * @net: the applicable net namespace
684 * @ifindex: index of device
686 * Search for an interface by index. Returns %NULL if the device
687 * is not found or a pointer to the device. The device has not
688 * had its reference counter increased so the caller must be careful
689 * about locking. The caller must hold either the RTNL semaphore
693 struct net_device
*__dev_get_by_index(struct net
*net
, int ifindex
)
695 struct hlist_node
*p
;
696 struct net_device
*dev
;
697 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
699 hlist_for_each_entry(dev
, p
, head
, index_hlist
)
700 if (dev
->ifindex
== ifindex
)
705 EXPORT_SYMBOL(__dev_get_by_index
);
708 * dev_get_by_index_rcu - find a device by its ifindex
709 * @net: the applicable net namespace
710 * @ifindex: index of device
712 * Search for an interface by index. Returns %NULL if the device
713 * is not found or a pointer to the device. The device has not
714 * had its reference counter increased so the caller must be careful
715 * about locking. The caller must hold RCU lock.
718 struct net_device
*dev_get_by_index_rcu(struct net
*net
, int ifindex
)
720 struct hlist_node
*p
;
721 struct net_device
*dev
;
722 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
724 hlist_for_each_entry_rcu(dev
, p
, head
, index_hlist
)
725 if (dev
->ifindex
== ifindex
)
730 EXPORT_SYMBOL(dev_get_by_index_rcu
);
734 * dev_get_by_index - find a device by its ifindex
735 * @net: the applicable net namespace
736 * @ifindex: index of device
738 * Search for an interface by index. Returns NULL if the device
739 * is not found or a pointer to the device. The device returned has
740 * had a reference added and the pointer is safe until the user calls
741 * dev_put to indicate they have finished with it.
744 struct net_device
*dev_get_by_index(struct net
*net
, int ifindex
)
746 struct net_device
*dev
;
749 dev
= dev_get_by_index_rcu(net
, ifindex
);
755 EXPORT_SYMBOL(dev_get_by_index
);
758 * dev_getbyhwaddr_rcu - find a device by its hardware address
759 * @net: the applicable net namespace
760 * @type: media type of device
761 * @ha: hardware address
763 * Search for an interface by MAC address. Returns NULL if the device
764 * is not found or a pointer to the device.
765 * The caller must hold RCU or RTNL.
766 * The returned device has not had its ref count increased
767 * and the caller must therefore be careful about locking
771 struct net_device
*dev_getbyhwaddr_rcu(struct net
*net
, unsigned short type
,
774 struct net_device
*dev
;
776 for_each_netdev_rcu(net
, dev
)
777 if (dev
->type
== type
&&
778 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
783 EXPORT_SYMBOL(dev_getbyhwaddr_rcu
);
785 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
787 struct net_device
*dev
;
790 for_each_netdev(net
, dev
)
791 if (dev
->type
== type
)
796 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
798 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
800 struct net_device
*dev
, *ret
= NULL
;
803 for_each_netdev_rcu(net
, dev
)
804 if (dev
->type
== type
) {
812 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
815 * dev_get_by_flags_rcu - find any device with given flags
816 * @net: the applicable net namespace
817 * @if_flags: IFF_* values
818 * @mask: bitmask of bits in if_flags to check
820 * Search for any interface with the given flags. Returns NULL if a device
821 * is not found or a pointer to the device. Must be called inside
822 * rcu_read_lock(), and result refcount is unchanged.
825 struct net_device
*dev_get_by_flags_rcu(struct net
*net
, unsigned short if_flags
,
828 struct net_device
*dev
, *ret
;
831 for_each_netdev_rcu(net
, dev
) {
832 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
839 EXPORT_SYMBOL(dev_get_by_flags_rcu
);
842 * dev_valid_name - check if name is okay for network device
845 * Network device names need to be valid file names to
846 * to allow sysfs to work. We also disallow any kind of
849 bool dev_valid_name(const char *name
)
853 if (strlen(name
) >= IFNAMSIZ
)
855 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
859 if (*name
== '/' || isspace(*name
))
865 EXPORT_SYMBOL(dev_valid_name
);
868 * __dev_alloc_name - allocate a name for a device
869 * @net: network namespace to allocate the device name in
870 * @name: name format string
871 * @buf: scratch buffer and result name string
873 * Passed a format string - eg "lt%d" it will try and find a suitable
874 * id. It scans list of devices to build up a free map, then chooses
875 * the first empty slot. The caller must hold the dev_base or rtnl lock
876 * while allocating the name and adding the device in order to avoid
878 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
879 * Returns the number of the unit assigned or a negative errno code.
882 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
886 const int max_netdevices
= 8*PAGE_SIZE
;
887 unsigned long *inuse
;
888 struct net_device
*d
;
890 p
= strnchr(name
, IFNAMSIZ
-1, '%');
893 * Verify the string as this thing may have come from
894 * the user. There must be either one "%d" and no other "%"
897 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
900 /* Use one page as a bit array of possible slots */
901 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
905 for_each_netdev(net
, d
) {
906 if (!sscanf(d
->name
, name
, &i
))
908 if (i
< 0 || i
>= max_netdevices
)
911 /* avoid cases where sscanf is not exact inverse of printf */
912 snprintf(buf
, IFNAMSIZ
, name
, i
);
913 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
917 i
= find_first_zero_bit(inuse
, max_netdevices
);
918 free_page((unsigned long) inuse
);
922 snprintf(buf
, IFNAMSIZ
, name
, i
);
923 if (!__dev_get_by_name(net
, buf
))
926 /* It is possible to run out of possible slots
927 * when the name is long and there isn't enough space left
928 * for the digits, or if all bits are used.
934 * dev_alloc_name - allocate a name for a device
936 * @name: name format string
938 * Passed a format string - eg "lt%d" it will try and find a suitable
939 * id. It scans list of devices to build up a free map, then chooses
940 * the first empty slot. The caller must hold the dev_base or rtnl lock
941 * while allocating the name and adding the device in order to avoid
943 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
944 * Returns the number of the unit assigned or a negative errno code.
947 int dev_alloc_name(struct net_device
*dev
, const char *name
)
953 BUG_ON(!dev_net(dev
));
955 ret
= __dev_alloc_name(net
, name
, buf
);
957 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
960 EXPORT_SYMBOL(dev_alloc_name
);
962 static int dev_get_valid_name(struct net_device
*dev
, const char *name
)
966 BUG_ON(!dev_net(dev
));
969 if (!dev_valid_name(name
))
972 if (strchr(name
, '%'))
973 return dev_alloc_name(dev
, name
);
974 else if (__dev_get_by_name(net
, name
))
976 else if (dev
->name
!= name
)
977 strlcpy(dev
->name
, name
, IFNAMSIZ
);
983 * dev_change_name - change name of a device
985 * @newname: name (or format string) must be at least IFNAMSIZ
987 * Change name of a device, can pass format strings "eth%d".
990 int dev_change_name(struct net_device
*dev
, const char *newname
)
992 char oldname
[IFNAMSIZ
];
998 BUG_ON(!dev_net(dev
));
1001 if (dev
->flags
& IFF_UP
)
1004 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0)
1007 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
1009 err
= dev_get_valid_name(dev
, newname
);
1014 ret
= device_rename(&dev
->dev
, dev
->name
);
1016 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1020 write_lock_bh(&dev_base_lock
);
1021 hlist_del_rcu(&dev
->name_hlist
);
1022 write_unlock_bh(&dev_base_lock
);
1026 write_lock_bh(&dev_base_lock
);
1027 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
1028 write_unlock_bh(&dev_base_lock
);
1030 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
1031 ret
= notifier_to_errno(ret
);
1034 /* err >= 0 after dev_alloc_name() or stores the first errno */
1037 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1040 pr_err("%s: name change rollback failed: %d\n",
1049 * dev_set_alias - change ifalias of a device
1051 * @alias: name up to IFALIASZ
1052 * @len: limit of bytes to copy from info
1054 * Set ifalias for a device,
1056 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
1060 if (len
>= IFALIASZ
)
1065 kfree(dev
->ifalias
);
1066 dev
->ifalias
= NULL
;
1071 dev
->ifalias
= krealloc(dev
->ifalias
, len
+ 1, GFP_KERNEL
);
1075 strlcpy(dev
->ifalias
, alias
, len
+1);
1081 * netdev_features_change - device changes features
1082 * @dev: device to cause notification
1084 * Called to indicate a device has changed features.
1086 void netdev_features_change(struct net_device
*dev
)
1088 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1090 EXPORT_SYMBOL(netdev_features_change
);
1093 * netdev_state_change - device changes state
1094 * @dev: device to cause notification
1096 * Called to indicate a device has changed state. This function calls
1097 * the notifier chains for netdev_chain and sends a NEWLINK message
1098 * to the routing socket.
1100 void netdev_state_change(struct net_device
*dev
)
1102 if (dev
->flags
& IFF_UP
) {
1103 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
1104 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
1107 EXPORT_SYMBOL(netdev_state_change
);
1109 int netdev_bonding_change(struct net_device
*dev
, unsigned long event
)
1111 return call_netdevice_notifiers(event
, dev
);
1113 EXPORT_SYMBOL(netdev_bonding_change
);
1116 * dev_load - load a network module
1117 * @net: the applicable net namespace
1118 * @name: name of interface
1120 * If a network interface is not present and the process has suitable
1121 * privileges this function loads the module. If module loading is not
1122 * available in this kernel then it becomes a nop.
1125 void dev_load(struct net
*net
, const char *name
)
1127 struct net_device
*dev
;
1131 dev
= dev_get_by_name_rcu(net
, name
);
1135 if (no_module
&& capable(CAP_NET_ADMIN
))
1136 no_module
= request_module("netdev-%s", name
);
1137 if (no_module
&& capable(CAP_SYS_MODULE
)) {
1138 if (!request_module("%s", name
))
1139 pr_err("Loading kernel module for a network device with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s instead.\n",
1143 EXPORT_SYMBOL(dev_load
);
1145 static int __dev_open(struct net_device
*dev
)
1147 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1152 if (!netif_device_present(dev
))
1155 ret
= call_netdevice_notifiers(NETDEV_PRE_UP
, dev
);
1156 ret
= notifier_to_errno(ret
);
1160 set_bit(__LINK_STATE_START
, &dev
->state
);
1162 if (ops
->ndo_validate_addr
)
1163 ret
= ops
->ndo_validate_addr(dev
);
1165 if (!ret
&& ops
->ndo_open
)
1166 ret
= ops
->ndo_open(dev
);
1169 clear_bit(__LINK_STATE_START
, &dev
->state
);
1171 dev
->flags
|= IFF_UP
;
1172 net_dmaengine_get();
1173 dev_set_rx_mode(dev
);
1181 * dev_open - prepare an interface for use.
1182 * @dev: device to open
1184 * Takes a device from down to up state. The device's private open
1185 * function is invoked and then the multicast lists are loaded. Finally
1186 * the device is moved into the up state and a %NETDEV_UP message is
1187 * sent to the netdev notifier chain.
1189 * Calling this function on an active interface is a nop. On a failure
1190 * a negative errno code is returned.
1192 int dev_open(struct net_device
*dev
)
1196 if (dev
->flags
& IFF_UP
)
1199 ret
= __dev_open(dev
);
1203 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1204 call_netdevice_notifiers(NETDEV_UP
, dev
);
1208 EXPORT_SYMBOL(dev_open
);
1210 static int __dev_close_many(struct list_head
*head
)
1212 struct net_device
*dev
;
1217 list_for_each_entry(dev
, head
, unreg_list
) {
1218 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1220 clear_bit(__LINK_STATE_START
, &dev
->state
);
1222 /* Synchronize to scheduled poll. We cannot touch poll list, it
1223 * can be even on different cpu. So just clear netif_running().
1225 * dev->stop() will invoke napi_disable() on all of it's
1226 * napi_struct instances on this device.
1228 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1231 dev_deactivate_many(head
);
1233 list_for_each_entry(dev
, head
, unreg_list
) {
1234 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1237 * Call the device specific close. This cannot fail.
1238 * Only if device is UP
1240 * We allow it to be called even after a DETACH hot-plug
1246 dev
->flags
&= ~IFF_UP
;
1247 net_dmaengine_put();
1253 static int __dev_close(struct net_device
*dev
)
1258 list_add(&dev
->unreg_list
, &single
);
1259 retval
= __dev_close_many(&single
);
1264 static int dev_close_many(struct list_head
*head
)
1266 struct net_device
*dev
, *tmp
;
1267 LIST_HEAD(tmp_list
);
1269 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
)
1270 if (!(dev
->flags
& IFF_UP
))
1271 list_move(&dev
->unreg_list
, &tmp_list
);
1273 __dev_close_many(head
);
1275 list_for_each_entry(dev
, head
, unreg_list
) {
1276 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1277 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1280 /* rollback_registered_many needs the complete original list */
1281 list_splice(&tmp_list
, head
);
1286 * dev_close - shutdown an interface.
1287 * @dev: device to shutdown
1289 * This function moves an active device into down state. A
1290 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1291 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1294 int dev_close(struct net_device
*dev
)
1296 if (dev
->flags
& IFF_UP
) {
1299 list_add(&dev
->unreg_list
, &single
);
1300 dev_close_many(&single
);
1305 EXPORT_SYMBOL(dev_close
);
1309 * dev_disable_lro - disable Large Receive Offload on a device
1312 * Disable Large Receive Offload (LRO) on a net device. Must be
1313 * called under RTNL. This is needed if received packets may be
1314 * forwarded to another interface.
1316 void dev_disable_lro(struct net_device
*dev
)
1319 * If we're trying to disable lro on a vlan device
1320 * use the underlying physical device instead
1322 if (is_vlan_dev(dev
))
1323 dev
= vlan_dev_real_dev(dev
);
1325 dev
->wanted_features
&= ~NETIF_F_LRO
;
1326 netdev_update_features(dev
);
1328 if (unlikely(dev
->features
& NETIF_F_LRO
))
1329 netdev_WARN(dev
, "failed to disable LRO!\n");
1331 EXPORT_SYMBOL(dev_disable_lro
);
1334 static int dev_boot_phase
= 1;
1337 * register_netdevice_notifier - register a network notifier block
1340 * Register a notifier to be called when network device events occur.
1341 * The notifier passed is linked into the kernel structures and must
1342 * not be reused until it has been unregistered. A negative errno code
1343 * is returned on a failure.
1345 * When registered all registration and up events are replayed
1346 * to the new notifier to allow device to have a race free
1347 * view of the network device list.
1350 int register_netdevice_notifier(struct notifier_block
*nb
)
1352 struct net_device
*dev
;
1353 struct net_device
*last
;
1358 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1364 for_each_netdev(net
, dev
) {
1365 err
= nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1366 err
= notifier_to_errno(err
);
1370 if (!(dev
->flags
& IFF_UP
))
1373 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1384 for_each_netdev(net
, dev
) {
1388 if (dev
->flags
& IFF_UP
) {
1389 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1390 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1392 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1393 nb
->notifier_call(nb
, NETDEV_UNREGISTER_BATCH
, dev
);
1398 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1401 EXPORT_SYMBOL(register_netdevice_notifier
);
1404 * unregister_netdevice_notifier - unregister a network notifier block
1407 * Unregister a notifier previously registered by
1408 * register_netdevice_notifier(). The notifier is unlinked into the
1409 * kernel structures and may then be reused. A negative errno code
1410 * is returned on a failure.
1412 * After unregistering unregister and down device events are synthesized
1413 * for all devices on the device list to the removed notifier to remove
1414 * the need for special case cleanup code.
1417 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1419 struct net_device
*dev
;
1424 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1429 for_each_netdev(net
, dev
) {
1430 if (dev
->flags
& IFF_UP
) {
1431 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1432 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1434 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1435 nb
->notifier_call(nb
, NETDEV_UNREGISTER_BATCH
, dev
);
1442 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1445 * call_netdevice_notifiers - call all network notifier blocks
1446 * @val: value passed unmodified to notifier function
1447 * @dev: net_device pointer passed unmodified to notifier function
1449 * Call all network notifier blocks. Parameters and return value
1450 * are as for raw_notifier_call_chain().
1453 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1456 return raw_notifier_call_chain(&netdev_chain
, val
, dev
);
1458 EXPORT_SYMBOL(call_netdevice_notifiers
);
1460 static struct static_key netstamp_needed __read_mostly
;
1461 #ifdef HAVE_JUMP_LABEL
1462 /* We are not allowed to call static_key_slow_dec() from irq context
1463 * If net_disable_timestamp() is called from irq context, defer the
1464 * static_key_slow_dec() calls.
1466 static atomic_t netstamp_needed_deferred
;
1469 void net_enable_timestamp(void)
1471 #ifdef HAVE_JUMP_LABEL
1472 int deferred
= atomic_xchg(&netstamp_needed_deferred
, 0);
1476 static_key_slow_dec(&netstamp_needed
);
1480 WARN_ON(in_interrupt());
1481 static_key_slow_inc(&netstamp_needed
);
1483 EXPORT_SYMBOL(net_enable_timestamp
);
1485 void net_disable_timestamp(void)
1487 #ifdef HAVE_JUMP_LABEL
1488 if (in_interrupt()) {
1489 atomic_inc(&netstamp_needed_deferred
);
1493 static_key_slow_dec(&netstamp_needed
);
1495 EXPORT_SYMBOL(net_disable_timestamp
);
1497 static inline void net_timestamp_set(struct sk_buff
*skb
)
1499 skb
->tstamp
.tv64
= 0;
1500 if (static_key_false(&netstamp_needed
))
1501 __net_timestamp(skb
);
1504 #define net_timestamp_check(COND, SKB) \
1505 if (static_key_false(&netstamp_needed)) { \
1506 if ((COND) && !(SKB)->tstamp.tv64) \
1507 __net_timestamp(SKB); \
1510 static int net_hwtstamp_validate(struct ifreq *ifr)
1512 struct hwtstamp_config cfg
;
1513 enum hwtstamp_tx_types tx_type
;
1514 enum hwtstamp_rx_filters rx_filter
;
1515 int tx_type_valid
= 0;
1516 int rx_filter_valid
= 0;
1518 if (copy_from_user(&cfg
, ifr
->ifr_data
, sizeof(cfg
)))
1521 if (cfg
.flags
) /* reserved for future extensions */
1524 tx_type
= cfg
.tx_type
;
1525 rx_filter
= cfg
.rx_filter
;
1528 case HWTSTAMP_TX_OFF
:
1529 case HWTSTAMP_TX_ON
:
1530 case HWTSTAMP_TX_ONESTEP_SYNC
:
1535 switch (rx_filter
) {
1536 case HWTSTAMP_FILTER_NONE
:
1537 case HWTSTAMP_FILTER_ALL
:
1538 case HWTSTAMP_FILTER_SOME
:
1539 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT
:
1540 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC
:
1541 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ
:
1542 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT
:
1543 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC
:
1544 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ
:
1545 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT
:
1546 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC
:
1547 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ
:
1548 case HWTSTAMP_FILTER_PTP_V2_EVENT
:
1549 case HWTSTAMP_FILTER_PTP_V2_SYNC
:
1550 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ
:
1551 rx_filter_valid
= 1;
1555 if (!tx_type_valid
|| !rx_filter_valid
)
1561 static inline bool is_skb_forwardable(struct net_device
*dev
,
1562 struct sk_buff
*skb
)
1566 if (!(dev
->flags
& IFF_UP
))
1569 len
= dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
;
1570 if (skb
->len
<= len
)
1573 /* if TSO is enabled, we don't care about the length as the packet
1574 * could be forwarded without being segmented before
1576 if (skb_is_gso(skb
))
1583 * dev_forward_skb - loopback an skb to another netif
1585 * @dev: destination network device
1586 * @skb: buffer to forward
1589 * NET_RX_SUCCESS (no congestion)
1590 * NET_RX_DROP (packet was dropped, but freed)
1592 * dev_forward_skb can be used for injecting an skb from the
1593 * start_xmit function of one device into the receive queue
1594 * of another device.
1596 * The receiving device may be in another namespace, so
1597 * we have to clear all information in the skb that could
1598 * impact namespace isolation.
1600 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1602 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
1603 if (skb_copy_ubufs(skb
, GFP_ATOMIC
)) {
1604 atomic_long_inc(&dev
->rx_dropped
);
1613 if (unlikely(!is_skb_forwardable(dev
, skb
))) {
1614 atomic_long_inc(&dev
->rx_dropped
);
1621 skb
->tstamp
.tv64
= 0;
1622 skb
->pkt_type
= PACKET_HOST
;
1623 skb
->protocol
= eth_type_trans(skb
, dev
);
1627 return netif_rx(skb
);
1629 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1631 static inline int deliver_skb(struct sk_buff
*skb
,
1632 struct packet_type
*pt_prev
,
1633 struct net_device
*orig_dev
)
1635 atomic_inc(&skb
->users
);
1636 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1640 * Support routine. Sends outgoing frames to any network
1641 * taps currently in use.
1644 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1646 struct packet_type
*ptype
;
1647 struct sk_buff
*skb2
= NULL
;
1648 struct packet_type
*pt_prev
= NULL
;
1651 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1652 /* Never send packets back to the socket
1653 * they originated from - MvS (miquels@drinkel.ow.org)
1655 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1656 (ptype
->af_packet_priv
== NULL
||
1657 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1659 deliver_skb(skb2
, pt_prev
, skb
->dev
);
1664 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1668 net_timestamp_set(skb2
);
1670 /* skb->nh should be correctly
1671 set by sender, so that the second statement is
1672 just protection against buggy protocols.
1674 skb_reset_mac_header(skb2
);
1676 if (skb_network_header(skb2
) < skb2
->data
||
1677 skb2
->network_header
> skb2
->tail
) {
1678 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1679 ntohs(skb2
->protocol
),
1681 skb_reset_network_header(skb2
);
1684 skb2
->transport_header
= skb2
->network_header
;
1685 skb2
->pkt_type
= PACKET_OUTGOING
;
1690 pt_prev
->func(skb2
, skb
->dev
, pt_prev
, skb
->dev
);
1694 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1695 * @dev: Network device
1696 * @txq: number of queues available
1698 * If real_num_tx_queues is changed the tc mappings may no longer be
1699 * valid. To resolve this verify the tc mapping remains valid and if
1700 * not NULL the mapping. With no priorities mapping to this
1701 * offset/count pair it will no longer be used. In the worst case TC0
1702 * is invalid nothing can be done so disable priority mappings. If is
1703 * expected that drivers will fix this mapping if they can before
1704 * calling netif_set_real_num_tx_queues.
1706 static void netif_setup_tc(struct net_device
*dev
, unsigned int txq
)
1709 struct netdev_tc_txq
*tc
= &dev
->tc_to_txq
[0];
1711 /* If TC0 is invalidated disable TC mapping */
1712 if (tc
->offset
+ tc
->count
> txq
) {
1713 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1718 /* Invalidated prio to tc mappings set to TC0 */
1719 for (i
= 1; i
< TC_BITMASK
+ 1; i
++) {
1720 int q
= netdev_get_prio_tc_map(dev
, i
);
1722 tc
= &dev
->tc_to_txq
[q
];
1723 if (tc
->offset
+ tc
->count
> txq
) {
1724 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1726 netdev_set_prio_tc_map(dev
, i
, 0);
1732 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1733 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1735 int netif_set_real_num_tx_queues(struct net_device
*dev
, unsigned int txq
)
1739 if (txq
< 1 || txq
> dev
->num_tx_queues
)
1742 if (dev
->reg_state
== NETREG_REGISTERED
||
1743 dev
->reg_state
== NETREG_UNREGISTERING
) {
1746 rc
= netdev_queue_update_kobjects(dev
, dev
->real_num_tx_queues
,
1752 netif_setup_tc(dev
, txq
);
1754 if (txq
< dev
->real_num_tx_queues
)
1755 qdisc_reset_all_tx_gt(dev
, txq
);
1758 dev
->real_num_tx_queues
= txq
;
1761 EXPORT_SYMBOL(netif_set_real_num_tx_queues
);
1765 * netif_set_real_num_rx_queues - set actual number of RX queues used
1766 * @dev: Network device
1767 * @rxq: Actual number of RX queues
1769 * This must be called either with the rtnl_lock held or before
1770 * registration of the net device. Returns 0 on success, or a
1771 * negative error code. If called before registration, it always
1774 int netif_set_real_num_rx_queues(struct net_device
*dev
, unsigned int rxq
)
1778 if (rxq
< 1 || rxq
> dev
->num_rx_queues
)
1781 if (dev
->reg_state
== NETREG_REGISTERED
) {
1784 rc
= net_rx_queue_update_kobjects(dev
, dev
->real_num_rx_queues
,
1790 dev
->real_num_rx_queues
= rxq
;
1793 EXPORT_SYMBOL(netif_set_real_num_rx_queues
);
1796 static inline void __netif_reschedule(struct Qdisc
*q
)
1798 struct softnet_data
*sd
;
1799 unsigned long flags
;
1801 local_irq_save(flags
);
1802 sd
= &__get_cpu_var(softnet_data
);
1803 q
->next_sched
= NULL
;
1804 *sd
->output_queue_tailp
= q
;
1805 sd
->output_queue_tailp
= &q
->next_sched
;
1806 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1807 local_irq_restore(flags
);
1810 void __netif_schedule(struct Qdisc
*q
)
1812 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
1813 __netif_reschedule(q
);
1815 EXPORT_SYMBOL(__netif_schedule
);
1817 void dev_kfree_skb_irq(struct sk_buff
*skb
)
1819 if (atomic_dec_and_test(&skb
->users
)) {
1820 struct softnet_data
*sd
;
1821 unsigned long flags
;
1823 local_irq_save(flags
);
1824 sd
= &__get_cpu_var(softnet_data
);
1825 skb
->next
= sd
->completion_queue
;
1826 sd
->completion_queue
= skb
;
1827 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1828 local_irq_restore(flags
);
1831 EXPORT_SYMBOL(dev_kfree_skb_irq
);
1833 void dev_kfree_skb_any(struct sk_buff
*skb
)
1835 if (in_irq() || irqs_disabled())
1836 dev_kfree_skb_irq(skb
);
1840 EXPORT_SYMBOL(dev_kfree_skb_any
);
1844 * netif_device_detach - mark device as removed
1845 * @dev: network device
1847 * Mark device as removed from system and therefore no longer available.
1849 void netif_device_detach(struct net_device
*dev
)
1851 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1852 netif_running(dev
)) {
1853 netif_tx_stop_all_queues(dev
);
1856 EXPORT_SYMBOL(netif_device_detach
);
1859 * netif_device_attach - mark device as attached
1860 * @dev: network device
1862 * Mark device as attached from system and restart if needed.
1864 void netif_device_attach(struct net_device
*dev
)
1866 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1867 netif_running(dev
)) {
1868 netif_tx_wake_all_queues(dev
);
1869 __netdev_watchdog_up(dev
);
1872 EXPORT_SYMBOL(netif_device_attach
);
1874 static void skb_warn_bad_offload(const struct sk_buff
*skb
)
1876 static const netdev_features_t null_features
= 0;
1877 struct net_device
*dev
= skb
->dev
;
1878 const char *driver
= "";
1880 if (dev
&& dev
->dev
.parent
)
1881 driver
= dev_driver_string(dev
->dev
.parent
);
1883 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
1884 "gso_type=%d ip_summed=%d\n",
1885 driver
, dev
? &dev
->features
: &null_features
,
1886 skb
->sk
? &skb
->sk
->sk_route_caps
: &null_features
,
1887 skb
->len
, skb
->data_len
, skb_shinfo(skb
)->gso_size
,
1888 skb_shinfo(skb
)->gso_type
, skb
->ip_summed
);
1892 * Invalidate hardware checksum when packet is to be mangled, and
1893 * complete checksum manually on outgoing path.
1895 int skb_checksum_help(struct sk_buff
*skb
)
1898 int ret
= 0, offset
;
1900 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1901 goto out_set_summed
;
1903 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1904 skb_warn_bad_offload(skb
);
1908 offset
= skb_checksum_start_offset(skb
);
1909 BUG_ON(offset
>= skb_headlen(skb
));
1910 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1912 offset
+= skb
->csum_offset
;
1913 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
1915 if (skb_cloned(skb
) &&
1916 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
1917 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1922 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1924 skb
->ip_summed
= CHECKSUM_NONE
;
1928 EXPORT_SYMBOL(skb_checksum_help
);
1931 * skb_gso_segment - Perform segmentation on skb.
1932 * @skb: buffer to segment
1933 * @features: features for the output path (see dev->features)
1935 * This function segments the given skb and returns a list of segments.
1937 * It may return NULL if the skb requires no segmentation. This is
1938 * only possible when GSO is used for verifying header integrity.
1940 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
,
1941 netdev_features_t features
)
1943 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1944 struct packet_type
*ptype
;
1945 __be16 type
= skb
->protocol
;
1946 int vlan_depth
= ETH_HLEN
;
1949 while (type
== htons(ETH_P_8021Q
)) {
1950 struct vlan_hdr
*vh
;
1952 if (unlikely(!pskb_may_pull(skb
, vlan_depth
+ VLAN_HLEN
)))
1953 return ERR_PTR(-EINVAL
);
1955 vh
= (struct vlan_hdr
*)(skb
->data
+ vlan_depth
);
1956 type
= vh
->h_vlan_encapsulated_proto
;
1957 vlan_depth
+= VLAN_HLEN
;
1960 skb_reset_mac_header(skb
);
1961 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1962 __skb_pull(skb
, skb
->mac_len
);
1964 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1965 skb_warn_bad_offload(skb
);
1967 if (skb_header_cloned(skb
) &&
1968 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
1969 return ERR_PTR(err
);
1973 list_for_each_entry_rcu(ptype
,
1974 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
1975 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
1976 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1977 err
= ptype
->gso_send_check(skb
);
1978 segs
= ERR_PTR(err
);
1979 if (err
|| skb_gso_ok(skb
, features
))
1981 __skb_push(skb
, (skb
->data
-
1982 skb_network_header(skb
)));
1984 segs
= ptype
->gso_segment(skb
, features
);
1990 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1994 EXPORT_SYMBOL(skb_gso_segment
);
1996 /* Take action when hardware reception checksum errors are detected. */
1998 void netdev_rx_csum_fault(struct net_device
*dev
)
2000 if (net_ratelimit()) {
2001 pr_err("%s: hw csum failure\n", dev
? dev
->name
: "<unknown>");
2005 EXPORT_SYMBOL(netdev_rx_csum_fault
);
2008 /* Actually, we should eliminate this check as soon as we know, that:
2009 * 1. IOMMU is present and allows to map all the memory.
2010 * 2. No high memory really exists on this machine.
2013 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
2015 #ifdef CONFIG_HIGHMEM
2017 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
2018 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2019 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2020 if (PageHighMem(skb_frag_page(frag
)))
2025 if (PCI_DMA_BUS_IS_PHYS
) {
2026 struct device
*pdev
= dev
->dev
.parent
;
2030 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2031 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2032 dma_addr_t addr
= page_to_phys(skb_frag_page(frag
));
2033 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
2042 void (*destructor
)(struct sk_buff
*skb
);
2045 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2047 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
2049 struct dev_gso_cb
*cb
;
2052 struct sk_buff
*nskb
= skb
->next
;
2054 skb
->next
= nskb
->next
;
2057 } while (skb
->next
);
2059 cb
= DEV_GSO_CB(skb
);
2061 cb
->destructor(skb
);
2065 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2066 * @skb: buffer to segment
2067 * @features: device features as applicable to this skb
2069 * This function segments the given skb and stores the list of segments
2072 static int dev_gso_segment(struct sk_buff
*skb
, netdev_features_t features
)
2074 struct sk_buff
*segs
;
2076 segs
= skb_gso_segment(skb
, features
);
2078 /* Verifying header integrity only. */
2083 return PTR_ERR(segs
);
2086 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
2087 skb
->destructor
= dev_gso_skb_destructor
;
2092 static bool can_checksum_protocol(netdev_features_t features
, __be16 protocol
)
2094 return ((features
& NETIF_F_GEN_CSUM
) ||
2095 ((features
& NETIF_F_V4_CSUM
) &&
2096 protocol
== htons(ETH_P_IP
)) ||
2097 ((features
& NETIF_F_V6_CSUM
) &&
2098 protocol
== htons(ETH_P_IPV6
)) ||
2099 ((features
& NETIF_F_FCOE_CRC
) &&
2100 protocol
== htons(ETH_P_FCOE
)));
2103 static netdev_features_t
harmonize_features(struct sk_buff
*skb
,
2104 __be16 protocol
, netdev_features_t features
)
2106 if (!can_checksum_protocol(features
, protocol
)) {
2107 features
&= ~NETIF_F_ALL_CSUM
;
2108 features
&= ~NETIF_F_SG
;
2109 } else if (illegal_highdma(skb
->dev
, skb
)) {
2110 features
&= ~NETIF_F_SG
;
2116 netdev_features_t
netif_skb_features(struct sk_buff
*skb
)
2118 __be16 protocol
= skb
->protocol
;
2119 netdev_features_t features
= skb
->dev
->features
;
2121 if (protocol
== htons(ETH_P_8021Q
)) {
2122 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
2123 protocol
= veh
->h_vlan_encapsulated_proto
;
2124 } else if (!vlan_tx_tag_present(skb
)) {
2125 return harmonize_features(skb
, protocol
, features
);
2128 features
&= (skb
->dev
->vlan_features
| NETIF_F_HW_VLAN_TX
);
2130 if (protocol
!= htons(ETH_P_8021Q
)) {
2131 return harmonize_features(skb
, protocol
, features
);
2133 features
&= NETIF_F_SG
| NETIF_F_HIGHDMA
| NETIF_F_FRAGLIST
|
2134 NETIF_F_GEN_CSUM
| NETIF_F_HW_VLAN_TX
;
2135 return harmonize_features(skb
, protocol
, features
);
2138 EXPORT_SYMBOL(netif_skb_features
);
2141 * Returns true if either:
2142 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2143 * 2. skb is fragmented and the device does not support SG, or if
2144 * at least one of fragments is in highmem and device does not
2145 * support DMA from it.
2147 static inline int skb_needs_linearize(struct sk_buff
*skb
,
2150 return skb_is_nonlinear(skb
) &&
2151 ((skb_has_frag_list(skb
) &&
2152 !(features
& NETIF_F_FRAGLIST
)) ||
2153 (skb_shinfo(skb
)->nr_frags
&&
2154 !(features
& NETIF_F_SG
)));
2157 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
2158 struct netdev_queue
*txq
)
2160 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2161 int rc
= NETDEV_TX_OK
;
2162 unsigned int skb_len
;
2164 if (likely(!skb
->next
)) {
2165 netdev_features_t features
;
2168 * If device doesn't need skb->dst, release it right now while
2169 * its hot in this cpu cache
2171 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2174 if (!list_empty(&ptype_all
))
2175 dev_queue_xmit_nit(skb
, dev
);
2177 features
= netif_skb_features(skb
);
2179 if (vlan_tx_tag_present(skb
) &&
2180 !(features
& NETIF_F_HW_VLAN_TX
)) {
2181 skb
= __vlan_put_tag(skb
, vlan_tx_tag_get(skb
));
2188 if (netif_needs_gso(skb
, features
)) {
2189 if (unlikely(dev_gso_segment(skb
, features
)))
2194 if (skb_needs_linearize(skb
, features
) &&
2195 __skb_linearize(skb
))
2198 /* If packet is not checksummed and device does not
2199 * support checksumming for this protocol, complete
2200 * checksumming here.
2202 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2203 skb_set_transport_header(skb
,
2204 skb_checksum_start_offset(skb
));
2205 if (!(features
& NETIF_F_ALL_CSUM
) &&
2206 skb_checksum_help(skb
))
2212 rc
= ops
->ndo_start_xmit(skb
, dev
);
2213 trace_net_dev_xmit(skb
, rc
, dev
, skb_len
);
2214 if (rc
== NETDEV_TX_OK
)
2215 txq_trans_update(txq
);
2221 struct sk_buff
*nskb
= skb
->next
;
2223 skb
->next
= nskb
->next
;
2227 * If device doesn't need nskb->dst, release it right now while
2228 * its hot in this cpu cache
2230 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2233 skb_len
= nskb
->len
;
2234 rc
= ops
->ndo_start_xmit(nskb
, dev
);
2235 trace_net_dev_xmit(nskb
, rc
, dev
, skb_len
);
2236 if (unlikely(rc
!= NETDEV_TX_OK
)) {
2237 if (rc
& ~NETDEV_TX_MASK
)
2238 goto out_kfree_gso_skb
;
2239 nskb
->next
= skb
->next
;
2243 txq_trans_update(txq
);
2244 if (unlikely(netif_xmit_stopped(txq
) && skb
->next
))
2245 return NETDEV_TX_BUSY
;
2246 } while (skb
->next
);
2249 if (likely(skb
->next
== NULL
))
2250 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
2257 static u32 hashrnd __read_mostly
;
2260 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2261 * to be used as a distribution range.
2263 u16
__skb_tx_hash(const struct net_device
*dev
, const struct sk_buff
*skb
,
2264 unsigned int num_tx_queues
)
2268 u16 qcount
= num_tx_queues
;
2270 if (skb_rx_queue_recorded(skb
)) {
2271 hash
= skb_get_rx_queue(skb
);
2272 while (unlikely(hash
>= num_tx_queues
))
2273 hash
-= num_tx_queues
;
2278 u8 tc
= netdev_get_prio_tc_map(dev
, skb
->priority
);
2279 qoffset
= dev
->tc_to_txq
[tc
].offset
;
2280 qcount
= dev
->tc_to_txq
[tc
].count
;
2283 if (skb
->sk
&& skb
->sk
->sk_hash
)
2284 hash
= skb
->sk
->sk_hash
;
2286 hash
= (__force u16
) skb
->protocol
;
2287 hash
= jhash_1word(hash
, hashrnd
);
2289 return (u16
) (((u64
) hash
* qcount
) >> 32) + qoffset
;
2291 EXPORT_SYMBOL(__skb_tx_hash
);
2293 static inline u16
dev_cap_txqueue(struct net_device
*dev
, u16 queue_index
)
2295 if (unlikely(queue_index
>= dev
->real_num_tx_queues
)) {
2296 net_warn_ratelimited("%s selects TX queue %d, but real number of TX queues is %d\n",
2297 dev
->name
, queue_index
,
2298 dev
->real_num_tx_queues
);
2304 static inline int get_xps_queue(struct net_device
*dev
, struct sk_buff
*skb
)
2307 struct xps_dev_maps
*dev_maps
;
2308 struct xps_map
*map
;
2309 int queue_index
= -1;
2312 dev_maps
= rcu_dereference(dev
->xps_maps
);
2314 map
= rcu_dereference(
2315 dev_maps
->cpu_map
[raw_smp_processor_id()]);
2318 queue_index
= map
->queues
[0];
2321 if (skb
->sk
&& skb
->sk
->sk_hash
)
2322 hash
= skb
->sk
->sk_hash
;
2324 hash
= (__force u16
) skb
->protocol
^
2326 hash
= jhash_1word(hash
, hashrnd
);
2327 queue_index
= map
->queues
[
2328 ((u64
)hash
* map
->len
) >> 32];
2330 if (unlikely(queue_index
>= dev
->real_num_tx_queues
))
2342 static struct netdev_queue
*dev_pick_tx(struct net_device
*dev
,
2343 struct sk_buff
*skb
)
2346 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2348 if (dev
->real_num_tx_queues
== 1)
2350 else if (ops
->ndo_select_queue
) {
2351 queue_index
= ops
->ndo_select_queue(dev
, skb
);
2352 queue_index
= dev_cap_txqueue(dev
, queue_index
);
2354 struct sock
*sk
= skb
->sk
;
2355 queue_index
= sk_tx_queue_get(sk
);
2357 if (queue_index
< 0 || skb
->ooo_okay
||
2358 queue_index
>= dev
->real_num_tx_queues
) {
2359 int old_index
= queue_index
;
2361 queue_index
= get_xps_queue(dev
, skb
);
2362 if (queue_index
< 0)
2363 queue_index
= skb_tx_hash(dev
, skb
);
2365 if (queue_index
!= old_index
&& sk
) {
2366 struct dst_entry
*dst
=
2367 rcu_dereference_check(sk
->sk_dst_cache
, 1);
2369 if (dst
&& skb_dst(skb
) == dst
)
2370 sk_tx_queue_set(sk
, queue_index
);
2375 skb_set_queue_mapping(skb
, queue_index
);
2376 return netdev_get_tx_queue(dev
, queue_index
);
2379 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2380 struct net_device
*dev
,
2381 struct netdev_queue
*txq
)
2383 spinlock_t
*root_lock
= qdisc_lock(q
);
2387 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
2388 qdisc_calculate_pkt_len(skb
, q
);
2390 * Heuristic to force contended enqueues to serialize on a
2391 * separate lock before trying to get qdisc main lock.
2392 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2393 * and dequeue packets faster.
2395 contended
= qdisc_is_running(q
);
2396 if (unlikely(contended
))
2397 spin_lock(&q
->busylock
);
2399 spin_lock(root_lock
);
2400 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2403 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2404 qdisc_run_begin(q
)) {
2406 * This is a work-conserving queue; there are no old skbs
2407 * waiting to be sent out; and the qdisc is not running -
2408 * xmit the skb directly.
2410 if (!(dev
->priv_flags
& IFF_XMIT_DST_RELEASE
))
2413 qdisc_bstats_update(q
, skb
);
2415 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
)) {
2416 if (unlikely(contended
)) {
2417 spin_unlock(&q
->busylock
);
2424 rc
= NET_XMIT_SUCCESS
;
2427 rc
= q
->enqueue(skb
, q
) & NET_XMIT_MASK
;
2428 if (qdisc_run_begin(q
)) {
2429 if (unlikely(contended
)) {
2430 spin_unlock(&q
->busylock
);
2436 spin_unlock(root_lock
);
2437 if (unlikely(contended
))
2438 spin_unlock(&q
->busylock
);
2442 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2443 static void skb_update_prio(struct sk_buff
*skb
)
2445 struct netprio_map
*map
= rcu_dereference_bh(skb
->dev
->priomap
);
2447 if ((!skb
->priority
) && (skb
->sk
) && map
)
2448 skb
->priority
= map
->priomap
[skb
->sk
->sk_cgrp_prioidx
];
2451 #define skb_update_prio(skb)
2454 static DEFINE_PER_CPU(int, xmit_recursion
);
2455 #define RECURSION_LIMIT 10
2458 * dev_loopback_xmit - loop back @skb
2459 * @skb: buffer to transmit
2461 int dev_loopback_xmit(struct sk_buff
*skb
)
2463 skb_reset_mac_header(skb
);
2464 __skb_pull(skb
, skb_network_offset(skb
));
2465 skb
->pkt_type
= PACKET_LOOPBACK
;
2466 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
2467 WARN_ON(!skb_dst(skb
));
2472 EXPORT_SYMBOL(dev_loopback_xmit
);
2475 * dev_queue_xmit - transmit a buffer
2476 * @skb: buffer to transmit
2478 * Queue a buffer for transmission to a network device. The caller must
2479 * have set the device and priority and built the buffer before calling
2480 * this function. The function can be called from an interrupt.
2482 * A negative errno code is returned on a failure. A success does not
2483 * guarantee the frame will be transmitted as it may be dropped due
2484 * to congestion or traffic shaping.
2486 * -----------------------------------------------------------------------------------
2487 * I notice this method can also return errors from the queue disciplines,
2488 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2491 * Regardless of the return value, the skb is consumed, so it is currently
2492 * difficult to retry a send to this method. (You can bump the ref count
2493 * before sending to hold a reference for retry if you are careful.)
2495 * When calling this method, interrupts MUST be enabled. This is because
2496 * the BH enable code must have IRQs enabled so that it will not deadlock.
2499 int dev_queue_xmit(struct sk_buff
*skb
)
2501 struct net_device
*dev
= skb
->dev
;
2502 struct netdev_queue
*txq
;
2506 /* Disable soft irqs for various locks below. Also
2507 * stops preemption for RCU.
2511 skb_update_prio(skb
);
2513 txq
= dev_pick_tx(dev
, skb
);
2514 q
= rcu_dereference_bh(txq
->qdisc
);
2516 #ifdef CONFIG_NET_CLS_ACT
2517 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2519 trace_net_dev_queue(skb
);
2521 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2525 /* The device has no queue. Common case for software devices:
2526 loopback, all the sorts of tunnels...
2528 Really, it is unlikely that netif_tx_lock protection is necessary
2529 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2531 However, it is possible, that they rely on protection
2534 Check this and shot the lock. It is not prone from deadlocks.
2535 Either shot noqueue qdisc, it is even simpler 8)
2537 if (dev
->flags
& IFF_UP
) {
2538 int cpu
= smp_processor_id(); /* ok because BHs are off */
2540 if (txq
->xmit_lock_owner
!= cpu
) {
2542 if (__this_cpu_read(xmit_recursion
) > RECURSION_LIMIT
)
2543 goto recursion_alert
;
2545 HARD_TX_LOCK(dev
, txq
, cpu
);
2547 if (!netif_xmit_stopped(txq
)) {
2548 __this_cpu_inc(xmit_recursion
);
2549 rc
= dev_hard_start_xmit(skb
, dev
, txq
);
2550 __this_cpu_dec(xmit_recursion
);
2551 if (dev_xmit_complete(rc
)) {
2552 HARD_TX_UNLOCK(dev
, txq
);
2556 HARD_TX_UNLOCK(dev
, txq
);
2557 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2560 /* Recursion is detected! It is possible,
2564 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2570 rcu_read_unlock_bh();
2575 rcu_read_unlock_bh();
2578 EXPORT_SYMBOL(dev_queue_xmit
);
2581 /*=======================================================================
2583 =======================================================================*/
2585 int netdev_max_backlog __read_mostly
= 1000;
2586 int netdev_tstamp_prequeue __read_mostly
= 1;
2587 int netdev_budget __read_mostly
= 300;
2588 int weight_p __read_mostly
= 64; /* old backlog weight */
2590 /* Called with irq disabled */
2591 static inline void ____napi_schedule(struct softnet_data
*sd
,
2592 struct napi_struct
*napi
)
2594 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
2595 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2599 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2600 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2601 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2602 * if hash is a canonical 4-tuple hash over transport ports.
2604 void __skb_get_rxhash(struct sk_buff
*skb
)
2606 struct flow_keys keys
;
2609 if (!skb_flow_dissect(skb
, &keys
))
2613 if ((__force u16
)keys
.port16
[1] < (__force u16
)keys
.port16
[0])
2614 swap(keys
.port16
[0], keys
.port16
[1]);
2618 /* get a consistent hash (same value on both flow directions) */
2619 if ((__force u32
)keys
.dst
< (__force u32
)keys
.src
)
2620 swap(keys
.dst
, keys
.src
);
2622 hash
= jhash_3words((__force u32
)keys
.dst
,
2623 (__force u32
)keys
.src
,
2624 (__force u32
)keys
.ports
, hashrnd
);
2630 EXPORT_SYMBOL(__skb_get_rxhash
);
2634 /* One global table that all flow-based protocols share. */
2635 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
2636 EXPORT_SYMBOL(rps_sock_flow_table
);
2638 struct static_key rps_needed __read_mostly
;
2640 static struct rps_dev_flow
*
2641 set_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2642 struct rps_dev_flow
*rflow
, u16 next_cpu
)
2644 if (next_cpu
!= RPS_NO_CPU
) {
2645 #ifdef CONFIG_RFS_ACCEL
2646 struct netdev_rx_queue
*rxqueue
;
2647 struct rps_dev_flow_table
*flow_table
;
2648 struct rps_dev_flow
*old_rflow
;
2653 /* Should we steer this flow to a different hardware queue? */
2654 if (!skb_rx_queue_recorded(skb
) || !dev
->rx_cpu_rmap
||
2655 !(dev
->features
& NETIF_F_NTUPLE
))
2657 rxq_index
= cpu_rmap_lookup_index(dev
->rx_cpu_rmap
, next_cpu
);
2658 if (rxq_index
== skb_get_rx_queue(skb
))
2661 rxqueue
= dev
->_rx
+ rxq_index
;
2662 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2665 flow_id
= skb
->rxhash
& flow_table
->mask
;
2666 rc
= dev
->netdev_ops
->ndo_rx_flow_steer(dev
, skb
,
2667 rxq_index
, flow_id
);
2671 rflow
= &flow_table
->flows
[flow_id
];
2673 if (old_rflow
->filter
== rflow
->filter
)
2674 old_rflow
->filter
= RPS_NO_FILTER
;
2678 per_cpu(softnet_data
, next_cpu
).input_queue_head
;
2681 rflow
->cpu
= next_cpu
;
2686 * get_rps_cpu is called from netif_receive_skb and returns the target
2687 * CPU from the RPS map of the receiving queue for a given skb.
2688 * rcu_read_lock must be held on entry.
2690 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2691 struct rps_dev_flow
**rflowp
)
2693 struct netdev_rx_queue
*rxqueue
;
2694 struct rps_map
*map
;
2695 struct rps_dev_flow_table
*flow_table
;
2696 struct rps_sock_flow_table
*sock_flow_table
;
2700 if (skb_rx_queue_recorded(skb
)) {
2701 u16 index
= skb_get_rx_queue(skb
);
2702 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
2703 WARN_ONCE(dev
->real_num_rx_queues
> 1,
2704 "%s received packet on queue %u, but number "
2705 "of RX queues is %u\n",
2706 dev
->name
, index
, dev
->real_num_rx_queues
);
2709 rxqueue
= dev
->_rx
+ index
;
2713 map
= rcu_dereference(rxqueue
->rps_map
);
2715 if (map
->len
== 1 &&
2716 !rcu_access_pointer(rxqueue
->rps_flow_table
)) {
2717 tcpu
= map
->cpus
[0];
2718 if (cpu_online(tcpu
))
2722 } else if (!rcu_access_pointer(rxqueue
->rps_flow_table
)) {
2726 skb_reset_network_header(skb
);
2727 if (!skb_get_rxhash(skb
))
2730 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2731 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
2732 if (flow_table
&& sock_flow_table
) {
2734 struct rps_dev_flow
*rflow
;
2736 rflow
= &flow_table
->flows
[skb
->rxhash
& flow_table
->mask
];
2739 next_cpu
= sock_flow_table
->ents
[skb
->rxhash
&
2740 sock_flow_table
->mask
];
2743 * If the desired CPU (where last recvmsg was done) is
2744 * different from current CPU (one in the rx-queue flow
2745 * table entry), switch if one of the following holds:
2746 * - Current CPU is unset (equal to RPS_NO_CPU).
2747 * - Current CPU is offline.
2748 * - The current CPU's queue tail has advanced beyond the
2749 * last packet that was enqueued using this table entry.
2750 * This guarantees that all previous packets for the flow
2751 * have been dequeued, thus preserving in order delivery.
2753 if (unlikely(tcpu
!= next_cpu
) &&
2754 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
2755 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
2756 rflow
->last_qtail
)) >= 0))
2757 rflow
= set_rps_cpu(dev
, skb
, rflow
, next_cpu
);
2759 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
2767 tcpu
= map
->cpus
[((u64
) skb
->rxhash
* map
->len
) >> 32];
2769 if (cpu_online(tcpu
)) {
2779 #ifdef CONFIG_RFS_ACCEL
2782 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2783 * @dev: Device on which the filter was set
2784 * @rxq_index: RX queue index
2785 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2786 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2788 * Drivers that implement ndo_rx_flow_steer() should periodically call
2789 * this function for each installed filter and remove the filters for
2790 * which it returns %true.
2792 bool rps_may_expire_flow(struct net_device
*dev
, u16 rxq_index
,
2793 u32 flow_id
, u16 filter_id
)
2795 struct netdev_rx_queue
*rxqueue
= dev
->_rx
+ rxq_index
;
2796 struct rps_dev_flow_table
*flow_table
;
2797 struct rps_dev_flow
*rflow
;
2802 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2803 if (flow_table
&& flow_id
<= flow_table
->mask
) {
2804 rflow
= &flow_table
->flows
[flow_id
];
2805 cpu
= ACCESS_ONCE(rflow
->cpu
);
2806 if (rflow
->filter
== filter_id
&& cpu
!= RPS_NO_CPU
&&
2807 ((int)(per_cpu(softnet_data
, cpu
).input_queue_head
-
2808 rflow
->last_qtail
) <
2809 (int)(10 * flow_table
->mask
)))
2815 EXPORT_SYMBOL(rps_may_expire_flow
);
2817 #endif /* CONFIG_RFS_ACCEL */
2819 /* Called from hardirq (IPI) context */
2820 static void rps_trigger_softirq(void *data
)
2822 struct softnet_data
*sd
= data
;
2824 ____napi_schedule(sd
, &sd
->backlog
);
2828 #endif /* CONFIG_RPS */
2831 * Check if this softnet_data structure is another cpu one
2832 * If yes, queue it to our IPI list and return 1
2835 static int rps_ipi_queued(struct softnet_data
*sd
)
2838 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
2841 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
2842 mysd
->rps_ipi_list
= sd
;
2844 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2847 #endif /* CONFIG_RPS */
2852 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2853 * queue (may be a remote CPU queue).
2855 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
2856 unsigned int *qtail
)
2858 struct softnet_data
*sd
;
2859 unsigned long flags
;
2861 sd
= &per_cpu(softnet_data
, cpu
);
2863 local_irq_save(flags
);
2866 if (skb_queue_len(&sd
->input_pkt_queue
) <= netdev_max_backlog
) {
2867 if (skb_queue_len(&sd
->input_pkt_queue
)) {
2869 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
2870 input_queue_tail_incr_save(sd
, qtail
);
2872 local_irq_restore(flags
);
2873 return NET_RX_SUCCESS
;
2876 /* Schedule NAPI for backlog device
2877 * We can use non atomic operation since we own the queue lock
2879 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
2880 if (!rps_ipi_queued(sd
))
2881 ____napi_schedule(sd
, &sd
->backlog
);
2889 local_irq_restore(flags
);
2891 atomic_long_inc(&skb
->dev
->rx_dropped
);
2897 * netif_rx - post buffer to the network code
2898 * @skb: buffer to post
2900 * This function receives a packet from a device driver and queues it for
2901 * the upper (protocol) levels to process. It always succeeds. The buffer
2902 * may be dropped during processing for congestion control or by the
2906 * NET_RX_SUCCESS (no congestion)
2907 * NET_RX_DROP (packet was dropped)
2911 int netif_rx(struct sk_buff
*skb
)
2915 /* if netpoll wants it, pretend we never saw it */
2916 if (netpoll_rx(skb
))
2919 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
2921 trace_netif_rx(skb
);
2923 if (static_key_false(&rps_needed
)) {
2924 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
2930 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
2932 cpu
= smp_processor_id();
2934 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
2942 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
2947 EXPORT_SYMBOL(netif_rx
);
2949 int netif_rx_ni(struct sk_buff
*skb
)
2954 err
= netif_rx(skb
);
2955 if (local_softirq_pending())
2961 EXPORT_SYMBOL(netif_rx_ni
);
2963 static void net_tx_action(struct softirq_action
*h
)
2965 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
2967 if (sd
->completion_queue
) {
2968 struct sk_buff
*clist
;
2970 local_irq_disable();
2971 clist
= sd
->completion_queue
;
2972 sd
->completion_queue
= NULL
;
2976 struct sk_buff
*skb
= clist
;
2977 clist
= clist
->next
;
2979 WARN_ON(atomic_read(&skb
->users
));
2980 trace_kfree_skb(skb
, net_tx_action
);
2985 if (sd
->output_queue
) {
2988 local_irq_disable();
2989 head
= sd
->output_queue
;
2990 sd
->output_queue
= NULL
;
2991 sd
->output_queue_tailp
= &sd
->output_queue
;
2995 struct Qdisc
*q
= head
;
2996 spinlock_t
*root_lock
;
2998 head
= head
->next_sched
;
3000 root_lock
= qdisc_lock(q
);
3001 if (spin_trylock(root_lock
)) {
3002 smp_mb__before_clear_bit();
3003 clear_bit(__QDISC_STATE_SCHED
,
3006 spin_unlock(root_lock
);
3008 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
3010 __netif_reschedule(q
);
3012 smp_mb__before_clear_bit();
3013 clear_bit(__QDISC_STATE_SCHED
,
3021 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3022 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3023 /* This hook is defined here for ATM LANE */
3024 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
3025 unsigned char *addr
) __read_mostly
;
3026 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
3029 #ifdef CONFIG_NET_CLS_ACT
3030 /* TODO: Maybe we should just force sch_ingress to be compiled in
3031 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3032 * a compare and 2 stores extra right now if we dont have it on
3033 * but have CONFIG_NET_CLS_ACT
3034 * NOTE: This doesn't stop any functionality; if you dont have
3035 * the ingress scheduler, you just can't add policies on ingress.
3038 static int ing_filter(struct sk_buff
*skb
, struct netdev_queue
*rxq
)
3040 struct net_device
*dev
= skb
->dev
;
3041 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
3042 int result
= TC_ACT_OK
;
3045 if (unlikely(MAX_RED_LOOP
< ttl
++)) {
3046 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3047 skb
->skb_iif
, dev
->ifindex
);
3051 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
3052 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
3055 if (q
!= &noop_qdisc
) {
3056 spin_lock(qdisc_lock(q
));
3057 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
3058 result
= qdisc_enqueue_root(skb
, q
);
3059 spin_unlock(qdisc_lock(q
));
3065 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
3066 struct packet_type
**pt_prev
,
3067 int *ret
, struct net_device
*orig_dev
)
3069 struct netdev_queue
*rxq
= rcu_dereference(skb
->dev
->ingress_queue
);
3071 if (!rxq
|| rxq
->qdisc
== &noop_qdisc
)
3075 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
3079 switch (ing_filter(skb
, rxq
)) {
3093 * netdev_rx_handler_register - register receive handler
3094 * @dev: device to register a handler for
3095 * @rx_handler: receive handler to register
3096 * @rx_handler_data: data pointer that is used by rx handler
3098 * Register a receive hander for a device. This handler will then be
3099 * called from __netif_receive_skb. A negative errno code is returned
3102 * The caller must hold the rtnl_mutex.
3104 * For a general description of rx_handler, see enum rx_handler_result.
3106 int netdev_rx_handler_register(struct net_device
*dev
,
3107 rx_handler_func_t
*rx_handler
,
3108 void *rx_handler_data
)
3112 if (dev
->rx_handler
)
3115 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
3116 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
3120 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
3123 * netdev_rx_handler_unregister - unregister receive handler
3124 * @dev: device to unregister a handler from
3126 * Unregister a receive hander from a device.
3128 * The caller must hold the rtnl_mutex.
3130 void netdev_rx_handler_unregister(struct net_device
*dev
)
3134 RCU_INIT_POINTER(dev
->rx_handler
, NULL
);
3135 RCU_INIT_POINTER(dev
->rx_handler_data
, NULL
);
3137 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
3139 static int __netif_receive_skb(struct sk_buff
*skb
)
3141 struct packet_type
*ptype
, *pt_prev
;
3142 rx_handler_func_t
*rx_handler
;
3143 struct net_device
*orig_dev
;
3144 struct net_device
*null_or_dev
;
3145 bool deliver_exact
= false;
3146 int ret
= NET_RX_DROP
;
3149 net_timestamp_check(!netdev_tstamp_prequeue
, skb
);
3151 trace_netif_receive_skb(skb
);
3153 /* if we've gotten here through NAPI, check netpoll */
3154 if (netpoll_receive_skb(skb
))
3158 skb
->skb_iif
= skb
->dev
->ifindex
;
3159 orig_dev
= skb
->dev
;
3161 skb_reset_network_header(skb
);
3162 skb_reset_transport_header(skb
);
3163 skb_reset_mac_len(skb
);
3171 __this_cpu_inc(softnet_data
.processed
);
3173 if (skb
->protocol
== cpu_to_be16(ETH_P_8021Q
)) {
3174 skb
= vlan_untag(skb
);
3179 #ifdef CONFIG_NET_CLS_ACT
3180 if (skb
->tc_verd
& TC_NCLS
) {
3181 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
3186 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
3187 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
3189 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3194 #ifdef CONFIG_NET_CLS_ACT
3195 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
3201 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
3202 if (vlan_tx_tag_present(skb
)) {
3204 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3207 if (vlan_do_receive(&skb
, !rx_handler
))
3209 else if (unlikely(!skb
))
3215 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3218 switch (rx_handler(&skb
)) {
3219 case RX_HANDLER_CONSUMED
:
3221 case RX_HANDLER_ANOTHER
:
3223 case RX_HANDLER_EXACT
:
3224 deliver_exact
= true;
3225 case RX_HANDLER_PASS
:
3232 /* deliver only exact match when indicated */
3233 null_or_dev
= deliver_exact
? skb
->dev
: NULL
;
3235 type
= skb
->protocol
;
3236 list_for_each_entry_rcu(ptype
,
3237 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
3238 if (ptype
->type
== type
&&
3239 (ptype
->dev
== null_or_dev
|| ptype
->dev
== skb
->dev
||
3240 ptype
->dev
== orig_dev
)) {
3242 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3248 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
3250 atomic_long_inc(&skb
->dev
->rx_dropped
);
3252 /* Jamal, now you will not able to escape explaining
3253 * me how you were going to use this. :-)
3264 * netif_receive_skb - process receive buffer from network
3265 * @skb: buffer to process
3267 * netif_receive_skb() is the main receive data processing function.
3268 * It always succeeds. The buffer may be dropped during processing
3269 * for congestion control or by the protocol layers.
3271 * This function may only be called from softirq context and interrupts
3272 * should be enabled.
3274 * Return values (usually ignored):
3275 * NET_RX_SUCCESS: no congestion
3276 * NET_RX_DROP: packet was dropped
3278 int netif_receive_skb(struct sk_buff
*skb
)
3280 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3282 if (skb_defer_rx_timestamp(skb
))
3283 return NET_RX_SUCCESS
;
3286 if (static_key_false(&rps_needed
)) {
3287 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3292 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3295 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3302 return __netif_receive_skb(skb
);
3304 EXPORT_SYMBOL(netif_receive_skb
);
3306 /* Network device is going away, flush any packets still pending
3307 * Called with irqs disabled.
3309 static void flush_backlog(void *arg
)
3311 struct net_device
*dev
= arg
;
3312 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3313 struct sk_buff
*skb
, *tmp
;
3316 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3317 if (skb
->dev
== dev
) {
3318 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3320 input_queue_head_incr(sd
);
3325 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3326 if (skb
->dev
== dev
) {
3327 __skb_unlink(skb
, &sd
->process_queue
);
3329 input_queue_head_incr(sd
);
3334 static int napi_gro_complete(struct sk_buff
*skb
)
3336 struct packet_type
*ptype
;
3337 __be16 type
= skb
->protocol
;
3338 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3341 if (NAPI_GRO_CB(skb
)->count
== 1) {
3342 skb_shinfo(skb
)->gso_size
= 0;
3347 list_for_each_entry_rcu(ptype
, head
, list
) {
3348 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_complete
)
3351 err
= ptype
->gro_complete(skb
);
3357 WARN_ON(&ptype
->list
== head
);
3359 return NET_RX_SUCCESS
;
3363 return netif_receive_skb(skb
);
3366 inline void napi_gro_flush(struct napi_struct
*napi
)
3368 struct sk_buff
*skb
, *next
;
3370 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3373 napi_gro_complete(skb
);
3376 napi
->gro_count
= 0;
3377 napi
->gro_list
= NULL
;
3379 EXPORT_SYMBOL(napi_gro_flush
);
3381 enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3383 struct sk_buff
**pp
= NULL
;
3384 struct packet_type
*ptype
;
3385 __be16 type
= skb
->protocol
;
3386 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3389 enum gro_result ret
;
3391 if (!(skb
->dev
->features
& NETIF_F_GRO
) || netpoll_rx_on(skb
))
3394 if (skb_is_gso(skb
) || skb_has_frag_list(skb
))
3398 list_for_each_entry_rcu(ptype
, head
, list
) {
3399 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_receive
)
3402 skb_set_network_header(skb
, skb_gro_offset(skb
));
3403 mac_len
= skb
->network_header
- skb
->mac_header
;
3404 skb
->mac_len
= mac_len
;
3405 NAPI_GRO_CB(skb
)->same_flow
= 0;
3406 NAPI_GRO_CB(skb
)->flush
= 0;
3407 NAPI_GRO_CB(skb
)->free
= 0;
3409 pp
= ptype
->gro_receive(&napi
->gro_list
, skb
);
3414 if (&ptype
->list
== head
)
3417 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3418 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3421 struct sk_buff
*nskb
= *pp
;
3425 napi_gro_complete(nskb
);
3432 if (NAPI_GRO_CB(skb
)->flush
|| napi
->gro_count
>= MAX_GRO_SKBS
)
3436 NAPI_GRO_CB(skb
)->count
= 1;
3437 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
3438 skb
->next
= napi
->gro_list
;
3439 napi
->gro_list
= skb
;
3443 if (skb_headlen(skb
) < skb_gro_offset(skb
)) {
3444 int grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
3446 BUG_ON(skb
->end
- skb
->tail
< grow
);
3448 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3451 skb
->data_len
-= grow
;
3453 skb_shinfo(skb
)->frags
[0].page_offset
+= grow
;
3454 skb_frag_size_sub(&skb_shinfo(skb
)->frags
[0], grow
);
3456 if (unlikely(!skb_frag_size(&skb_shinfo(skb
)->frags
[0]))) {
3457 skb_frag_unref(skb
, 0);
3458 memmove(skb_shinfo(skb
)->frags
,
3459 skb_shinfo(skb
)->frags
+ 1,
3460 --skb_shinfo(skb
)->nr_frags
* sizeof(skb_frag_t
));
3471 EXPORT_SYMBOL(dev_gro_receive
);
3473 static inline gro_result_t
3474 __napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3477 unsigned int maclen
= skb
->dev
->hard_header_len
;
3479 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3480 unsigned long diffs
;
3482 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
3483 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
3484 if (maclen
== ETH_HLEN
)
3485 diffs
|= compare_ether_header(skb_mac_header(p
),
3486 skb_gro_mac_header(skb
));
3488 diffs
= memcmp(skb_mac_header(p
),
3489 skb_gro_mac_header(skb
),
3491 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
3492 NAPI_GRO_CB(p
)->flush
= 0;
3495 return dev_gro_receive(napi
, skb
);
3498 gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
3502 if (netif_receive_skb(skb
))
3510 case GRO_MERGED_FREE
:
3511 if (NAPI_GRO_CB(skb
)->free
== NAPI_GRO_FREE_STOLEN_HEAD
)
3512 kmem_cache_free(skbuff_head_cache
, skb
);
3524 EXPORT_SYMBOL(napi_skb_finish
);
3526 void skb_gro_reset_offset(struct sk_buff
*skb
)
3528 NAPI_GRO_CB(skb
)->data_offset
= 0;
3529 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3530 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3532 if (skb
->mac_header
== skb
->tail
&&
3533 !PageHighMem(skb_frag_page(&skb_shinfo(skb
)->frags
[0]))) {
3534 NAPI_GRO_CB(skb
)->frag0
=
3535 skb_frag_address(&skb_shinfo(skb
)->frags
[0]);
3536 NAPI_GRO_CB(skb
)->frag0_len
= skb_frag_size(&skb_shinfo(skb
)->frags
[0]);
3539 EXPORT_SYMBOL(skb_gro_reset_offset
);
3541 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3543 skb_gro_reset_offset(skb
);
3545 return napi_skb_finish(__napi_gro_receive(napi
, skb
), skb
);
3547 EXPORT_SYMBOL(napi_gro_receive
);
3549 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
3551 __skb_pull(skb
, skb_headlen(skb
));
3552 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3553 skb_reserve(skb
, NET_SKB_PAD
+ NET_IP_ALIGN
- skb_headroom(skb
));
3555 skb
->dev
= napi
->dev
;
3561 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
3563 struct sk_buff
*skb
= napi
->skb
;
3566 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
3572 EXPORT_SYMBOL(napi_get_frags
);
3574 gro_result_t
napi_frags_finish(struct napi_struct
*napi
, struct sk_buff
*skb
,
3580 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
3582 if (ret
== GRO_HELD
)
3583 skb_gro_pull(skb
, -ETH_HLEN
);
3584 else if (netif_receive_skb(skb
))
3589 case GRO_MERGED_FREE
:
3590 napi_reuse_skb(napi
, skb
);
3599 EXPORT_SYMBOL(napi_frags_finish
);
3601 static struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
3603 struct sk_buff
*skb
= napi
->skb
;
3610 skb_reset_mac_header(skb
);
3611 skb_gro_reset_offset(skb
);
3613 off
= skb_gro_offset(skb
);
3614 hlen
= off
+ sizeof(*eth
);
3615 eth
= skb_gro_header_fast(skb
, off
);
3616 if (skb_gro_header_hard(skb
, hlen
)) {
3617 eth
= skb_gro_header_slow(skb
, hlen
, off
);
3618 if (unlikely(!eth
)) {
3619 napi_reuse_skb(napi
, skb
);
3625 skb_gro_pull(skb
, sizeof(*eth
));
3628 * This works because the only protocols we care about don't require
3629 * special handling. We'll fix it up properly at the end.
3631 skb
->protocol
= eth
->h_proto
;
3637 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
3639 struct sk_buff
*skb
= napi_frags_skb(napi
);
3644 return napi_frags_finish(napi
, skb
, __napi_gro_receive(napi
, skb
));
3646 EXPORT_SYMBOL(napi_gro_frags
);
3649 * net_rps_action sends any pending IPI's for rps.
3650 * Note: called with local irq disabled, but exits with local irq enabled.
3652 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
3655 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
3658 sd
->rps_ipi_list
= NULL
;
3662 /* Send pending IPI's to kick RPS processing on remote cpus. */
3664 struct softnet_data
*next
= remsd
->rps_ipi_next
;
3666 if (cpu_online(remsd
->cpu
))
3667 __smp_call_function_single(remsd
->cpu
,
3676 static int process_backlog(struct napi_struct
*napi
, int quota
)
3679 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
3682 /* Check if we have pending ipi, its better to send them now,
3683 * not waiting net_rx_action() end.
3685 if (sd
->rps_ipi_list
) {
3686 local_irq_disable();
3687 net_rps_action_and_irq_enable(sd
);
3690 napi
->weight
= weight_p
;
3691 local_irq_disable();
3692 while (work
< quota
) {
3693 struct sk_buff
*skb
;
3696 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
3698 __netif_receive_skb(skb
);
3699 local_irq_disable();
3700 input_queue_head_incr(sd
);
3701 if (++work
>= quota
) {
3708 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3710 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
3711 &sd
->process_queue
);
3713 if (qlen
< quota
- work
) {
3715 * Inline a custom version of __napi_complete().
3716 * only current cpu owns and manipulates this napi,
3717 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3718 * we can use a plain write instead of clear_bit(),
3719 * and we dont need an smp_mb() memory barrier.
3721 list_del(&napi
->poll_list
);
3724 quota
= work
+ qlen
;
3734 * __napi_schedule - schedule for receive
3735 * @n: entry to schedule
3737 * The entry's receive function will be scheduled to run
3739 void __napi_schedule(struct napi_struct
*n
)
3741 unsigned long flags
;
3743 local_irq_save(flags
);
3744 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
3745 local_irq_restore(flags
);
3747 EXPORT_SYMBOL(__napi_schedule
);
3749 void __napi_complete(struct napi_struct
*n
)
3751 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
3752 BUG_ON(n
->gro_list
);
3754 list_del(&n
->poll_list
);
3755 smp_mb__before_clear_bit();
3756 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
3758 EXPORT_SYMBOL(__napi_complete
);
3760 void napi_complete(struct napi_struct
*n
)
3762 unsigned long flags
;
3765 * don't let napi dequeue from the cpu poll list
3766 * just in case its running on a different cpu
3768 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
3772 local_irq_save(flags
);
3774 local_irq_restore(flags
);
3776 EXPORT_SYMBOL(napi_complete
);
3778 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
3779 int (*poll
)(struct napi_struct
*, int), int weight
)
3781 INIT_LIST_HEAD(&napi
->poll_list
);
3782 napi
->gro_count
= 0;
3783 napi
->gro_list
= NULL
;
3786 napi
->weight
= weight
;
3787 list_add(&napi
->dev_list
, &dev
->napi_list
);
3789 #ifdef CONFIG_NETPOLL
3790 spin_lock_init(&napi
->poll_lock
);
3791 napi
->poll_owner
= -1;
3793 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
3795 EXPORT_SYMBOL(netif_napi_add
);
3797 void netif_napi_del(struct napi_struct
*napi
)
3799 struct sk_buff
*skb
, *next
;
3801 list_del_init(&napi
->dev_list
);
3802 napi_free_frags(napi
);
3804 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3810 napi
->gro_list
= NULL
;
3811 napi
->gro_count
= 0;
3813 EXPORT_SYMBOL(netif_napi_del
);
3815 static void net_rx_action(struct softirq_action
*h
)
3817 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3818 unsigned long time_limit
= jiffies
+ 2;
3819 int budget
= netdev_budget
;
3822 local_irq_disable();
3824 while (!list_empty(&sd
->poll_list
)) {
3825 struct napi_struct
*n
;
3828 /* If softirq window is exhuasted then punt.
3829 * Allow this to run for 2 jiffies since which will allow
3830 * an average latency of 1.5/HZ.
3832 if (unlikely(budget
<= 0 || time_after(jiffies
, time_limit
)))
3837 /* Even though interrupts have been re-enabled, this
3838 * access is safe because interrupts can only add new
3839 * entries to the tail of this list, and only ->poll()
3840 * calls can remove this head entry from the list.
3842 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
3844 have
= netpoll_poll_lock(n
);
3848 /* This NAPI_STATE_SCHED test is for avoiding a race
3849 * with netpoll's poll_napi(). Only the entity which
3850 * obtains the lock and sees NAPI_STATE_SCHED set will
3851 * actually make the ->poll() call. Therefore we avoid
3852 * accidentally calling ->poll() when NAPI is not scheduled.
3855 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
3856 work
= n
->poll(n
, weight
);
3860 WARN_ON_ONCE(work
> weight
);
3864 local_irq_disable();
3866 /* Drivers must not modify the NAPI state if they
3867 * consume the entire weight. In such cases this code
3868 * still "owns" the NAPI instance and therefore can
3869 * move the instance around on the list at-will.
3871 if (unlikely(work
== weight
)) {
3872 if (unlikely(napi_disable_pending(n
))) {
3875 local_irq_disable();
3877 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
3880 netpoll_poll_unlock(have
);
3883 net_rps_action_and_irq_enable(sd
);
3885 #ifdef CONFIG_NET_DMA
3887 * There may not be any more sk_buffs coming right now, so push
3888 * any pending DMA copies to hardware
3890 dma_issue_pending_all();
3897 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3901 static gifconf_func_t
*gifconf_list
[NPROTO
];
3904 * register_gifconf - register a SIOCGIF handler
3905 * @family: Address family
3906 * @gifconf: Function handler
3908 * Register protocol dependent address dumping routines. The handler
3909 * that is passed must not be freed or reused until it has been replaced
3910 * by another handler.
3912 int register_gifconf(unsigned int family
, gifconf_func_t
*gifconf
)
3914 if (family
>= NPROTO
)
3916 gifconf_list
[family
] = gifconf
;
3919 EXPORT_SYMBOL(register_gifconf
);
3923 * Map an interface index to its name (SIOCGIFNAME)
3927 * We need this ioctl for efficient implementation of the
3928 * if_indextoname() function required by the IPv6 API. Without
3929 * it, we would have to search all the interfaces to find a
3933 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
3935 struct net_device
*dev
;
3939 * Fetch the caller's info block.
3942 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
3946 dev
= dev_get_by_index_rcu(net
, ifr
.ifr_ifindex
);
3952 strcpy(ifr
.ifr_name
, dev
->name
);
3955 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
3961 * Perform a SIOCGIFCONF call. This structure will change
3962 * size eventually, and there is nothing I can do about it.
3963 * Thus we will need a 'compatibility mode'.
3966 static int dev_ifconf(struct net
*net
, char __user
*arg
)
3969 struct net_device
*dev
;
3976 * Fetch the caller's info block.
3979 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
3986 * Loop over the interfaces, and write an info block for each.
3990 for_each_netdev(net
, dev
) {
3991 for (i
= 0; i
< NPROTO
; i
++) {
3992 if (gifconf_list
[i
]) {
3995 done
= gifconf_list
[i
](dev
, NULL
, 0);
3997 done
= gifconf_list
[i
](dev
, pos
+ total
,
4007 * All done. Write the updated control block back to the caller.
4009 ifc
.ifc_len
= total
;
4012 * Both BSD and Solaris return 0 here, so we do too.
4014 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
4017 #ifdef CONFIG_PROC_FS
4019 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4021 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4022 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4023 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4025 static inline struct net_device
*dev_from_same_bucket(struct seq_file
*seq
, loff_t
*pos
)
4027 struct net
*net
= seq_file_net(seq
);
4028 struct net_device
*dev
;
4029 struct hlist_node
*p
;
4030 struct hlist_head
*h
;
4031 unsigned int count
= 0, offset
= get_offset(*pos
);
4033 h
= &net
->dev_name_head
[get_bucket(*pos
)];
4034 hlist_for_each_entry_rcu(dev
, p
, h
, name_hlist
) {
4035 if (++count
== offset
)
4042 static inline struct net_device
*dev_from_bucket(struct seq_file
*seq
, loff_t
*pos
)
4044 struct net_device
*dev
;
4045 unsigned int bucket
;
4048 dev
= dev_from_same_bucket(seq
, pos
);
4052 bucket
= get_bucket(*pos
) + 1;
4053 *pos
= set_bucket_offset(bucket
, 1);
4054 } while (bucket
< NETDEV_HASHENTRIES
);
4060 * This is invoked by the /proc filesystem handler to display a device
4063 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4068 return SEQ_START_TOKEN
;
4070 if (get_bucket(*pos
) >= NETDEV_HASHENTRIES
)
4073 return dev_from_bucket(seq
, pos
);
4076 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4079 return dev_from_bucket(seq
, pos
);
4082 void dev_seq_stop(struct seq_file
*seq
, void *v
)
4088 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
4090 struct rtnl_link_stats64 temp
;
4091 const struct rtnl_link_stats64
*stats
= dev_get_stats(dev
, &temp
);
4093 seq_printf(seq
, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4094 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4095 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
4097 stats
->rx_dropped
+ stats
->rx_missed_errors
,
4098 stats
->rx_fifo_errors
,
4099 stats
->rx_length_errors
+ stats
->rx_over_errors
+
4100 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
4101 stats
->rx_compressed
, stats
->multicast
,
4102 stats
->tx_bytes
, stats
->tx_packets
,
4103 stats
->tx_errors
, stats
->tx_dropped
,
4104 stats
->tx_fifo_errors
, stats
->collisions
,
4105 stats
->tx_carrier_errors
+
4106 stats
->tx_aborted_errors
+
4107 stats
->tx_window_errors
+
4108 stats
->tx_heartbeat_errors
,
4109 stats
->tx_compressed
);
4113 * Called from the PROCfs module. This now uses the new arbitrary sized
4114 * /proc/net interface to create /proc/net/dev
4116 static int dev_seq_show(struct seq_file
*seq
, void *v
)
4118 if (v
== SEQ_START_TOKEN
)
4119 seq_puts(seq
, "Inter-| Receive "
4121 " face |bytes packets errs drop fifo frame "
4122 "compressed multicast|bytes packets errs "
4123 "drop fifo colls carrier compressed\n");
4125 dev_seq_printf_stats(seq
, v
);
4129 static struct softnet_data
*softnet_get_online(loff_t
*pos
)
4131 struct softnet_data
*sd
= NULL
;
4133 while (*pos
< nr_cpu_ids
)
4134 if (cpu_online(*pos
)) {
4135 sd
= &per_cpu(softnet_data
, *pos
);
4142 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4144 return softnet_get_online(pos
);
4147 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4150 return softnet_get_online(pos
);
4153 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
4157 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
4159 struct softnet_data
*sd
= v
;
4161 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4162 sd
->processed
, sd
->dropped
, sd
->time_squeeze
, 0,
4163 0, 0, 0, 0, /* was fastroute */
4164 sd
->cpu_collision
, sd
->received_rps
);
4168 static const struct seq_operations dev_seq_ops
= {
4169 .start
= dev_seq_start
,
4170 .next
= dev_seq_next
,
4171 .stop
= dev_seq_stop
,
4172 .show
= dev_seq_show
,
4175 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
4177 return seq_open_net(inode
, file
, &dev_seq_ops
,
4178 sizeof(struct seq_net_private
));
4181 static const struct file_operations dev_seq_fops
= {
4182 .owner
= THIS_MODULE
,
4183 .open
= dev_seq_open
,
4185 .llseek
= seq_lseek
,
4186 .release
= seq_release_net
,
4189 static const struct seq_operations softnet_seq_ops
= {
4190 .start
= softnet_seq_start
,
4191 .next
= softnet_seq_next
,
4192 .stop
= softnet_seq_stop
,
4193 .show
= softnet_seq_show
,
4196 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
4198 return seq_open(file
, &softnet_seq_ops
);
4201 static const struct file_operations softnet_seq_fops
= {
4202 .owner
= THIS_MODULE
,
4203 .open
= softnet_seq_open
,
4205 .llseek
= seq_lseek
,
4206 .release
= seq_release
,
4209 static void *ptype_get_idx(loff_t pos
)
4211 struct packet_type
*pt
= NULL
;
4215 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
4221 for (t
= 0; t
< PTYPE_HASH_SIZE
; t
++) {
4222 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
4231 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4235 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
4238 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4240 struct packet_type
*pt
;
4241 struct list_head
*nxt
;
4245 if (v
== SEQ_START_TOKEN
)
4246 return ptype_get_idx(0);
4249 nxt
= pt
->list
.next
;
4250 if (pt
->type
== htons(ETH_P_ALL
)) {
4251 if (nxt
!= &ptype_all
)
4254 nxt
= ptype_base
[0].next
;
4256 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
4258 while (nxt
== &ptype_base
[hash
]) {
4259 if (++hash
>= PTYPE_HASH_SIZE
)
4261 nxt
= ptype_base
[hash
].next
;
4264 return list_entry(nxt
, struct packet_type
, list
);
4267 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
4273 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
4275 struct packet_type
*pt
= v
;
4277 if (v
== SEQ_START_TOKEN
)
4278 seq_puts(seq
, "Type Device Function\n");
4279 else if (pt
->dev
== NULL
|| dev_net(pt
->dev
) == seq_file_net(seq
)) {
4280 if (pt
->type
== htons(ETH_P_ALL
))
4281 seq_puts(seq
, "ALL ");
4283 seq_printf(seq
, "%04x", ntohs(pt
->type
));
4285 seq_printf(seq
, " %-8s %pF\n",
4286 pt
->dev
? pt
->dev
->name
: "", pt
->func
);
4292 static const struct seq_operations ptype_seq_ops
= {
4293 .start
= ptype_seq_start
,
4294 .next
= ptype_seq_next
,
4295 .stop
= ptype_seq_stop
,
4296 .show
= ptype_seq_show
,
4299 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
4301 return seq_open_net(inode
, file
, &ptype_seq_ops
,
4302 sizeof(struct seq_net_private
));
4305 static const struct file_operations ptype_seq_fops
= {
4306 .owner
= THIS_MODULE
,
4307 .open
= ptype_seq_open
,
4309 .llseek
= seq_lseek
,
4310 .release
= seq_release_net
,
4314 static int __net_init
dev_proc_net_init(struct net
*net
)
4318 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
4320 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
4322 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
4325 if (wext_proc_init(net
))
4331 proc_net_remove(net
, "ptype");
4333 proc_net_remove(net
, "softnet_stat");
4335 proc_net_remove(net
, "dev");
4339 static void __net_exit
dev_proc_net_exit(struct net
*net
)
4341 wext_proc_exit(net
);
4343 proc_net_remove(net
, "ptype");
4344 proc_net_remove(net
, "softnet_stat");
4345 proc_net_remove(net
, "dev");
4348 static struct pernet_operations __net_initdata dev_proc_ops
= {
4349 .init
= dev_proc_net_init
,
4350 .exit
= dev_proc_net_exit
,
4353 static int __init
dev_proc_init(void)
4355 return register_pernet_subsys(&dev_proc_ops
);
4358 #define dev_proc_init() 0
4359 #endif /* CONFIG_PROC_FS */
4363 * netdev_set_master - set up master pointer
4364 * @slave: slave device
4365 * @master: new master device
4367 * Changes the master device of the slave. Pass %NULL to break the
4368 * bonding. The caller must hold the RTNL semaphore. On a failure
4369 * a negative errno code is returned. On success the reference counts
4370 * are adjusted and the function returns zero.
4372 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
4374 struct net_device
*old
= slave
->master
;
4384 slave
->master
= master
;
4390 EXPORT_SYMBOL(netdev_set_master
);
4393 * netdev_set_bond_master - set up bonding master/slave pair
4394 * @slave: slave device
4395 * @master: new master device
4397 * Changes the master device of the slave. Pass %NULL to break the
4398 * bonding. The caller must hold the RTNL semaphore. On a failure
4399 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4400 * to the routing socket and the function returns zero.
4402 int netdev_set_bond_master(struct net_device
*slave
, struct net_device
*master
)
4408 err
= netdev_set_master(slave
, master
);
4412 slave
->flags
|= IFF_SLAVE
;
4414 slave
->flags
&= ~IFF_SLAVE
;
4416 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
4419 EXPORT_SYMBOL(netdev_set_bond_master
);
4421 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
4423 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4425 if ((dev
->flags
& IFF_UP
) && ops
->ndo_change_rx_flags
)
4426 ops
->ndo_change_rx_flags(dev
, flags
);
4429 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
4431 unsigned int old_flags
= dev
->flags
;
4437 dev
->flags
|= IFF_PROMISC
;
4438 dev
->promiscuity
+= inc
;
4439 if (dev
->promiscuity
== 0) {
4442 * If inc causes overflow, untouch promisc and return error.
4445 dev
->flags
&= ~IFF_PROMISC
;
4447 dev
->promiscuity
-= inc
;
4448 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4453 if (dev
->flags
!= old_flags
) {
4454 pr_info("device %s %s promiscuous mode\n",
4456 dev
->flags
& IFF_PROMISC
? "entered" : "left");
4457 if (audit_enabled
) {
4458 current_uid_gid(&uid
, &gid
);
4459 audit_log(current
->audit_context
, GFP_ATOMIC
,
4460 AUDIT_ANOM_PROMISCUOUS
,
4461 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4462 dev
->name
, (dev
->flags
& IFF_PROMISC
),
4463 (old_flags
& IFF_PROMISC
),
4464 audit_get_loginuid(current
),
4466 audit_get_sessionid(current
));
4469 dev_change_rx_flags(dev
, IFF_PROMISC
);
4475 * dev_set_promiscuity - update promiscuity count on a device
4479 * Add or remove promiscuity from a device. While the count in the device
4480 * remains above zero the interface remains promiscuous. Once it hits zero
4481 * the device reverts back to normal filtering operation. A negative inc
4482 * value is used to drop promiscuity on the device.
4483 * Return 0 if successful or a negative errno code on error.
4485 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
4487 unsigned int old_flags
= dev
->flags
;
4490 err
= __dev_set_promiscuity(dev
, inc
);
4493 if (dev
->flags
!= old_flags
)
4494 dev_set_rx_mode(dev
);
4497 EXPORT_SYMBOL(dev_set_promiscuity
);
4500 * dev_set_allmulti - update allmulti count on a device
4504 * Add or remove reception of all multicast frames to a device. While the
4505 * count in the device remains above zero the interface remains listening
4506 * to all interfaces. Once it hits zero the device reverts back to normal
4507 * filtering operation. A negative @inc value is used to drop the counter
4508 * when releasing a resource needing all multicasts.
4509 * Return 0 if successful or a negative errno code on error.
4512 int dev_set_allmulti(struct net_device
*dev
, int inc
)
4514 unsigned int old_flags
= dev
->flags
;
4518 dev
->flags
|= IFF_ALLMULTI
;
4519 dev
->allmulti
+= inc
;
4520 if (dev
->allmulti
== 0) {
4523 * If inc causes overflow, untouch allmulti and return error.
4526 dev
->flags
&= ~IFF_ALLMULTI
;
4528 dev
->allmulti
-= inc
;
4529 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4534 if (dev
->flags
^ old_flags
) {
4535 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
4536 dev_set_rx_mode(dev
);
4540 EXPORT_SYMBOL(dev_set_allmulti
);
4543 * Upload unicast and multicast address lists to device and
4544 * configure RX filtering. When the device doesn't support unicast
4545 * filtering it is put in promiscuous mode while unicast addresses
4548 void __dev_set_rx_mode(struct net_device
*dev
)
4550 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4552 /* dev_open will call this function so the list will stay sane. */
4553 if (!(dev
->flags
&IFF_UP
))
4556 if (!netif_device_present(dev
))
4559 if (!(dev
->priv_flags
& IFF_UNICAST_FLT
)) {
4560 /* Unicast addresses changes may only happen under the rtnl,
4561 * therefore calling __dev_set_promiscuity here is safe.
4563 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
4564 __dev_set_promiscuity(dev
, 1);
4565 dev
->uc_promisc
= true;
4566 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
4567 __dev_set_promiscuity(dev
, -1);
4568 dev
->uc_promisc
= false;
4572 if (ops
->ndo_set_rx_mode
)
4573 ops
->ndo_set_rx_mode(dev
);
4576 void dev_set_rx_mode(struct net_device
*dev
)
4578 netif_addr_lock_bh(dev
);
4579 __dev_set_rx_mode(dev
);
4580 netif_addr_unlock_bh(dev
);
4584 * dev_get_flags - get flags reported to userspace
4587 * Get the combination of flag bits exported through APIs to userspace.
4589 unsigned int dev_get_flags(const struct net_device
*dev
)
4593 flags
= (dev
->flags
& ~(IFF_PROMISC
|
4598 (dev
->gflags
& (IFF_PROMISC
|
4601 if (netif_running(dev
)) {
4602 if (netif_oper_up(dev
))
4603 flags
|= IFF_RUNNING
;
4604 if (netif_carrier_ok(dev
))
4605 flags
|= IFF_LOWER_UP
;
4606 if (netif_dormant(dev
))
4607 flags
|= IFF_DORMANT
;
4612 EXPORT_SYMBOL(dev_get_flags
);
4614 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4616 unsigned int old_flags
= dev
->flags
;
4622 * Set the flags on our device.
4625 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
4626 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
4628 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
4632 * Load in the correct multicast list now the flags have changed.
4635 if ((old_flags
^ flags
) & IFF_MULTICAST
)
4636 dev_change_rx_flags(dev
, IFF_MULTICAST
);
4638 dev_set_rx_mode(dev
);
4641 * Have we downed the interface. We handle IFF_UP ourselves
4642 * according to user attempts to set it, rather than blindly
4647 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
4648 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
4651 dev_set_rx_mode(dev
);
4654 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
4655 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
4657 dev
->gflags
^= IFF_PROMISC
;
4658 dev_set_promiscuity(dev
, inc
);
4661 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4662 is important. Some (broken) drivers set IFF_PROMISC, when
4663 IFF_ALLMULTI is requested not asking us and not reporting.
4665 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
4666 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
4668 dev
->gflags
^= IFF_ALLMULTI
;
4669 dev_set_allmulti(dev
, inc
);
4675 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
)
4677 unsigned int changes
= dev
->flags
^ old_flags
;
4679 if (changes
& IFF_UP
) {
4680 if (dev
->flags
& IFF_UP
)
4681 call_netdevice_notifiers(NETDEV_UP
, dev
);
4683 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
4686 if (dev
->flags
& IFF_UP
&&
4687 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
)))
4688 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
4692 * dev_change_flags - change device settings
4694 * @flags: device state flags
4696 * Change settings on device based state flags. The flags are
4697 * in the userspace exported format.
4699 int dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4702 unsigned int changes
, old_flags
= dev
->flags
;
4704 ret
= __dev_change_flags(dev
, flags
);
4708 changes
= old_flags
^ dev
->flags
;
4710 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
4712 __dev_notify_flags(dev
, old_flags
);
4715 EXPORT_SYMBOL(dev_change_flags
);
4718 * dev_set_mtu - Change maximum transfer unit
4720 * @new_mtu: new transfer unit
4722 * Change the maximum transfer size of the network device.
4724 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
4726 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4729 if (new_mtu
== dev
->mtu
)
4732 /* MTU must be positive. */
4736 if (!netif_device_present(dev
))
4740 if (ops
->ndo_change_mtu
)
4741 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
4745 if (!err
&& dev
->flags
& IFF_UP
)
4746 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
4749 EXPORT_SYMBOL(dev_set_mtu
);
4752 * dev_set_group - Change group this device belongs to
4754 * @new_group: group this device should belong to
4756 void dev_set_group(struct net_device
*dev
, int new_group
)
4758 dev
->group
= new_group
;
4760 EXPORT_SYMBOL(dev_set_group
);
4763 * dev_set_mac_address - Change Media Access Control Address
4767 * Change the hardware (MAC) address of the device
4769 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
4771 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4774 if (!ops
->ndo_set_mac_address
)
4776 if (sa
->sa_family
!= dev
->type
)
4778 if (!netif_device_present(dev
))
4780 err
= ops
->ndo_set_mac_address(dev
, sa
);
4782 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4785 EXPORT_SYMBOL(dev_set_mac_address
);
4788 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4790 static int dev_ifsioc_locked(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4793 struct net_device
*dev
= dev_get_by_name_rcu(net
, ifr
->ifr_name
);
4799 case SIOCGIFFLAGS
: /* Get interface flags */
4800 ifr
->ifr_flags
= (short) dev_get_flags(dev
);
4803 case SIOCGIFMETRIC
: /* Get the metric on the interface
4804 (currently unused) */
4805 ifr
->ifr_metric
= 0;
4808 case SIOCGIFMTU
: /* Get the MTU of a device */
4809 ifr
->ifr_mtu
= dev
->mtu
;
4814 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
4816 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
4817 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4818 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
4826 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
4827 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
4828 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
4829 ifr
->ifr_map
.irq
= dev
->irq
;
4830 ifr
->ifr_map
.dma
= dev
->dma
;
4831 ifr
->ifr_map
.port
= dev
->if_port
;
4835 ifr
->ifr_ifindex
= dev
->ifindex
;
4839 ifr
->ifr_qlen
= dev
->tx_queue_len
;
4843 /* dev_ioctl() should ensure this case
4855 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4857 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4860 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
4861 const struct net_device_ops
*ops
;
4866 ops
= dev
->netdev_ops
;
4869 case SIOCSIFFLAGS
: /* Set interface flags */
4870 return dev_change_flags(dev
, ifr
->ifr_flags
);
4872 case SIOCSIFMETRIC
: /* Set the metric on the interface
4873 (currently unused) */
4876 case SIOCSIFMTU
: /* Set the MTU of a device */
4877 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
4880 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
4882 case SIOCSIFHWBROADCAST
:
4883 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
4885 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
4886 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4887 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4891 if (ops
->ndo_set_config
) {
4892 if (!netif_device_present(dev
))
4894 return ops
->ndo_set_config(dev
, &ifr
->ifr_map
);
4899 if (!ops
->ndo_set_rx_mode
||
4900 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4902 if (!netif_device_present(dev
))
4904 return dev_mc_add_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4907 if (!ops
->ndo_set_rx_mode
||
4908 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4910 if (!netif_device_present(dev
))
4912 return dev_mc_del_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4915 if (ifr
->ifr_qlen
< 0)
4917 dev
->tx_queue_len
= ifr
->ifr_qlen
;
4921 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
4922 return dev_change_name(dev
, ifr
->ifr_newname
);
4925 err
= net_hwtstamp_validate(ifr
);
4931 * Unknown or private ioctl
4934 if ((cmd
>= SIOCDEVPRIVATE
&&
4935 cmd
<= SIOCDEVPRIVATE
+ 15) ||
4936 cmd
== SIOCBONDENSLAVE
||
4937 cmd
== SIOCBONDRELEASE
||
4938 cmd
== SIOCBONDSETHWADDR
||
4939 cmd
== SIOCBONDSLAVEINFOQUERY
||
4940 cmd
== SIOCBONDINFOQUERY
||
4941 cmd
== SIOCBONDCHANGEACTIVE
||
4942 cmd
== SIOCGMIIPHY
||
4943 cmd
== SIOCGMIIREG
||
4944 cmd
== SIOCSMIIREG
||
4945 cmd
== SIOCBRADDIF
||
4946 cmd
== SIOCBRDELIF
||
4947 cmd
== SIOCSHWTSTAMP
||
4948 cmd
== SIOCWANDEV
) {
4950 if (ops
->ndo_do_ioctl
) {
4951 if (netif_device_present(dev
))
4952 err
= ops
->ndo_do_ioctl(dev
, ifr
, cmd
);
4964 * This function handles all "interface"-type I/O control requests. The actual
4965 * 'doing' part of this is dev_ifsioc above.
4969 * dev_ioctl - network device ioctl
4970 * @net: the applicable net namespace
4971 * @cmd: command to issue
4972 * @arg: pointer to a struct ifreq in user space
4974 * Issue ioctl functions to devices. This is normally called by the
4975 * user space syscall interfaces but can sometimes be useful for
4976 * other purposes. The return value is the return from the syscall if
4977 * positive or a negative errno code on error.
4980 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
4986 /* One special case: SIOCGIFCONF takes ifconf argument
4987 and requires shared lock, because it sleeps writing
4991 if (cmd
== SIOCGIFCONF
) {
4993 ret
= dev_ifconf(net
, (char __user
*) arg
);
4997 if (cmd
== SIOCGIFNAME
)
4998 return dev_ifname(net
, (struct ifreq __user
*)arg
);
5000 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
5003 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
5005 colon
= strchr(ifr
.ifr_name
, ':');
5010 * See which interface the caller is talking about.
5015 * These ioctl calls:
5016 * - can be done by all.
5017 * - atomic and do not require locking.
5028 dev_load(net
, ifr
.ifr_name
);
5030 ret
= dev_ifsioc_locked(net
, &ifr
, cmd
);
5035 if (copy_to_user(arg
, &ifr
,
5036 sizeof(struct ifreq
)))
5042 dev_load(net
, ifr
.ifr_name
);
5044 ret
= dev_ethtool(net
, &ifr
);
5049 if (copy_to_user(arg
, &ifr
,
5050 sizeof(struct ifreq
)))
5056 * These ioctl calls:
5057 * - require superuser power.
5058 * - require strict serialization.
5064 if (!capable(CAP_NET_ADMIN
))
5066 dev_load(net
, ifr
.ifr_name
);
5068 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5073 if (copy_to_user(arg
, &ifr
,
5074 sizeof(struct ifreq
)))
5080 * These ioctl calls:
5081 * - require superuser power.
5082 * - require strict serialization.
5083 * - do not return a value
5093 case SIOCSIFHWBROADCAST
:
5096 case SIOCBONDENSLAVE
:
5097 case SIOCBONDRELEASE
:
5098 case SIOCBONDSETHWADDR
:
5099 case SIOCBONDCHANGEACTIVE
:
5103 if (!capable(CAP_NET_ADMIN
))
5106 case SIOCBONDSLAVEINFOQUERY
:
5107 case SIOCBONDINFOQUERY
:
5108 dev_load(net
, ifr
.ifr_name
);
5110 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5115 /* Get the per device memory space. We can add this but
5116 * currently do not support it */
5118 /* Set the per device memory buffer space.
5119 * Not applicable in our case */
5124 * Unknown or private ioctl.
5127 if (cmd
== SIOCWANDEV
||
5128 (cmd
>= SIOCDEVPRIVATE
&&
5129 cmd
<= SIOCDEVPRIVATE
+ 15)) {
5130 dev_load(net
, ifr
.ifr_name
);
5132 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5134 if (!ret
&& copy_to_user(arg
, &ifr
,
5135 sizeof(struct ifreq
)))
5139 /* Take care of Wireless Extensions */
5140 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
5141 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
5148 * dev_new_index - allocate an ifindex
5149 * @net: the applicable net namespace
5151 * Returns a suitable unique value for a new device interface
5152 * number. The caller must hold the rtnl semaphore or the
5153 * dev_base_lock to be sure it remains unique.
5155 static int dev_new_index(struct net
*net
)
5161 if (!__dev_get_by_index(net
, ifindex
))
5166 /* Delayed registration/unregisteration */
5167 static LIST_HEAD(net_todo_list
);
5169 static void net_set_todo(struct net_device
*dev
)
5171 list_add_tail(&dev
->todo_list
, &net_todo_list
);
5174 static void rollback_registered_many(struct list_head
*head
)
5176 struct net_device
*dev
, *tmp
;
5178 BUG_ON(dev_boot_phase
);
5181 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
5182 /* Some devices call without registering
5183 * for initialization unwind. Remove those
5184 * devices and proceed with the remaining.
5186 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5187 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5191 list_del(&dev
->unreg_list
);
5194 dev
->dismantle
= true;
5195 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
5198 /* If device is running, close it first. */
5199 dev_close_many(head
);
5201 list_for_each_entry(dev
, head
, unreg_list
) {
5202 /* And unlink it from device chain. */
5203 unlist_netdevice(dev
);
5205 dev
->reg_state
= NETREG_UNREGISTERING
;
5210 list_for_each_entry(dev
, head
, unreg_list
) {
5211 /* Shutdown queueing discipline. */
5215 /* Notify protocols, that we are about to destroy
5216 this device. They should clean all the things.
5218 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5220 if (!dev
->rtnl_link_ops
||
5221 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5222 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
5225 * Flush the unicast and multicast chains
5230 if (dev
->netdev_ops
->ndo_uninit
)
5231 dev
->netdev_ops
->ndo_uninit(dev
);
5233 /* Notifier chain MUST detach us from master device. */
5234 WARN_ON(dev
->master
);
5236 /* Remove entries from kobject tree */
5237 netdev_unregister_kobject(dev
);
5240 /* Process any work delayed until the end of the batch */
5241 dev
= list_first_entry(head
, struct net_device
, unreg_list
);
5242 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
5246 list_for_each_entry(dev
, head
, unreg_list
)
5250 static void rollback_registered(struct net_device
*dev
)
5254 list_add(&dev
->unreg_list
, &single
);
5255 rollback_registered_many(&single
);
5259 static netdev_features_t
netdev_fix_features(struct net_device
*dev
,
5260 netdev_features_t features
)
5262 /* Fix illegal checksum combinations */
5263 if ((features
& NETIF_F_HW_CSUM
) &&
5264 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5265 netdev_warn(dev
, "mixed HW and IP checksum settings.\n");
5266 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5269 /* Fix illegal SG+CSUM combinations. */
5270 if ((features
& NETIF_F_SG
) &&
5271 !(features
& NETIF_F_ALL_CSUM
)) {
5273 "Dropping NETIF_F_SG since no checksum feature.\n");
5274 features
&= ~NETIF_F_SG
;
5277 /* TSO requires that SG is present as well. */
5278 if ((features
& NETIF_F_ALL_TSO
) && !(features
& NETIF_F_SG
)) {
5279 netdev_dbg(dev
, "Dropping TSO features since no SG feature.\n");
5280 features
&= ~NETIF_F_ALL_TSO
;
5283 /* TSO ECN requires that TSO is present as well. */
5284 if ((features
& NETIF_F_ALL_TSO
) == NETIF_F_TSO_ECN
)
5285 features
&= ~NETIF_F_TSO_ECN
;
5287 /* Software GSO depends on SG. */
5288 if ((features
& NETIF_F_GSO
) && !(features
& NETIF_F_SG
)) {
5289 netdev_dbg(dev
, "Dropping NETIF_F_GSO since no SG feature.\n");
5290 features
&= ~NETIF_F_GSO
;
5293 /* UFO needs SG and checksumming */
5294 if (features
& NETIF_F_UFO
) {
5295 /* maybe split UFO into V4 and V6? */
5296 if (!((features
& NETIF_F_GEN_CSUM
) ||
5297 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))
5298 == (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5300 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5301 features
&= ~NETIF_F_UFO
;
5304 if (!(features
& NETIF_F_SG
)) {
5306 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5307 features
&= ~NETIF_F_UFO
;
5314 int __netdev_update_features(struct net_device
*dev
)
5316 netdev_features_t features
;
5321 features
= netdev_get_wanted_features(dev
);
5323 if (dev
->netdev_ops
->ndo_fix_features
)
5324 features
= dev
->netdev_ops
->ndo_fix_features(dev
, features
);
5326 /* driver might be less strict about feature dependencies */
5327 features
= netdev_fix_features(dev
, features
);
5329 if (dev
->features
== features
)
5332 netdev_dbg(dev
, "Features changed: %pNF -> %pNF\n",
5333 &dev
->features
, &features
);
5335 if (dev
->netdev_ops
->ndo_set_features
)
5336 err
= dev
->netdev_ops
->ndo_set_features(dev
, features
);
5338 if (unlikely(err
< 0)) {
5340 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5341 err
, &features
, &dev
->features
);
5346 dev
->features
= features
;
5352 * netdev_update_features - recalculate device features
5353 * @dev: the device to check
5355 * Recalculate dev->features set and send notifications if it
5356 * has changed. Should be called after driver or hardware dependent
5357 * conditions might have changed that influence the features.
5359 void netdev_update_features(struct net_device
*dev
)
5361 if (__netdev_update_features(dev
))
5362 netdev_features_change(dev
);
5364 EXPORT_SYMBOL(netdev_update_features
);
5367 * netdev_change_features - recalculate device features
5368 * @dev: the device to check
5370 * Recalculate dev->features set and send notifications even
5371 * if they have not changed. Should be called instead of
5372 * netdev_update_features() if also dev->vlan_features might
5373 * have changed to allow the changes to be propagated to stacked
5376 void netdev_change_features(struct net_device
*dev
)
5378 __netdev_update_features(dev
);
5379 netdev_features_change(dev
);
5381 EXPORT_SYMBOL(netdev_change_features
);
5384 * netif_stacked_transfer_operstate - transfer operstate
5385 * @rootdev: the root or lower level device to transfer state from
5386 * @dev: the device to transfer operstate to
5388 * Transfer operational state from root to device. This is normally
5389 * called when a stacking relationship exists between the root
5390 * device and the device(a leaf device).
5392 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
5393 struct net_device
*dev
)
5395 if (rootdev
->operstate
== IF_OPER_DORMANT
)
5396 netif_dormant_on(dev
);
5398 netif_dormant_off(dev
);
5400 if (netif_carrier_ok(rootdev
)) {
5401 if (!netif_carrier_ok(dev
))
5402 netif_carrier_on(dev
);
5404 if (netif_carrier_ok(dev
))
5405 netif_carrier_off(dev
);
5408 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
5411 static int netif_alloc_rx_queues(struct net_device
*dev
)
5413 unsigned int i
, count
= dev
->num_rx_queues
;
5414 struct netdev_rx_queue
*rx
;
5418 rx
= kcalloc(count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5420 pr_err("netdev: Unable to allocate %u rx queues\n", count
);
5425 for (i
= 0; i
< count
; i
++)
5431 static void netdev_init_one_queue(struct net_device
*dev
,
5432 struct netdev_queue
*queue
, void *_unused
)
5434 /* Initialize queue lock */
5435 spin_lock_init(&queue
->_xmit_lock
);
5436 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
5437 queue
->xmit_lock_owner
= -1;
5438 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
5441 dql_init(&queue
->dql
, HZ
);
5445 static int netif_alloc_netdev_queues(struct net_device
*dev
)
5447 unsigned int count
= dev
->num_tx_queues
;
5448 struct netdev_queue
*tx
;
5452 tx
= kcalloc(count
, sizeof(struct netdev_queue
), GFP_KERNEL
);
5454 pr_err("netdev: Unable to allocate %u tx queues\n", count
);
5459 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5460 spin_lock_init(&dev
->tx_global_lock
);
5466 * register_netdevice - register a network device
5467 * @dev: device to register
5469 * Take a completed network device structure and add it to the kernel
5470 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5471 * chain. 0 is returned on success. A negative errno code is returned
5472 * on a failure to set up the device, or if the name is a duplicate.
5474 * Callers must hold the rtnl semaphore. You may want
5475 * register_netdev() instead of this.
5478 * The locking appears insufficient to guarantee two parallel registers
5479 * will not get the same name.
5482 int register_netdevice(struct net_device
*dev
)
5485 struct net
*net
= dev_net(dev
);
5487 BUG_ON(dev_boot_phase
);
5492 /* When net_device's are persistent, this will be fatal. */
5493 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
5496 spin_lock_init(&dev
->addr_list_lock
);
5497 netdev_set_addr_lockdep_class(dev
);
5501 ret
= dev_get_valid_name(dev
, dev
->name
);
5505 /* Init, if this function is available */
5506 if (dev
->netdev_ops
->ndo_init
) {
5507 ret
= dev
->netdev_ops
->ndo_init(dev
);
5515 dev
->ifindex
= dev_new_index(net
);
5516 if (dev
->iflink
== -1)
5517 dev
->iflink
= dev
->ifindex
;
5519 /* Transfer changeable features to wanted_features and enable
5520 * software offloads (GSO and GRO).
5522 dev
->hw_features
|= NETIF_F_SOFT_FEATURES
;
5523 dev
->features
|= NETIF_F_SOFT_FEATURES
;
5524 dev
->wanted_features
= dev
->features
& dev
->hw_features
;
5526 /* Turn on no cache copy if HW is doing checksum */
5527 if (!(dev
->flags
& IFF_LOOPBACK
)) {
5528 dev
->hw_features
|= NETIF_F_NOCACHE_COPY
;
5529 if (dev
->features
& NETIF_F_ALL_CSUM
) {
5530 dev
->wanted_features
|= NETIF_F_NOCACHE_COPY
;
5531 dev
->features
|= NETIF_F_NOCACHE_COPY
;
5535 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5537 dev
->vlan_features
|= NETIF_F_HIGHDMA
;
5539 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
5540 ret
= notifier_to_errno(ret
);
5544 ret
= netdev_register_kobject(dev
);
5547 dev
->reg_state
= NETREG_REGISTERED
;
5549 __netdev_update_features(dev
);
5552 * Default initial state at registry is that the
5553 * device is present.
5556 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5558 dev_init_scheduler(dev
);
5560 list_netdevice(dev
);
5562 /* Notify protocols, that a new device appeared. */
5563 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5564 ret
= notifier_to_errno(ret
);
5566 rollback_registered(dev
);
5567 dev
->reg_state
= NETREG_UNREGISTERED
;
5570 * Prevent userspace races by waiting until the network
5571 * device is fully setup before sending notifications.
5573 if (!dev
->rtnl_link_ops
||
5574 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5575 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5581 if (dev
->netdev_ops
->ndo_uninit
)
5582 dev
->netdev_ops
->ndo_uninit(dev
);
5585 EXPORT_SYMBOL(register_netdevice
);
5588 * init_dummy_netdev - init a dummy network device for NAPI
5589 * @dev: device to init
5591 * This takes a network device structure and initialize the minimum
5592 * amount of fields so it can be used to schedule NAPI polls without
5593 * registering a full blown interface. This is to be used by drivers
5594 * that need to tie several hardware interfaces to a single NAPI
5595 * poll scheduler due to HW limitations.
5597 int init_dummy_netdev(struct net_device
*dev
)
5599 /* Clear everything. Note we don't initialize spinlocks
5600 * are they aren't supposed to be taken by any of the
5601 * NAPI code and this dummy netdev is supposed to be
5602 * only ever used for NAPI polls
5604 memset(dev
, 0, sizeof(struct net_device
));
5606 /* make sure we BUG if trying to hit standard
5607 * register/unregister code path
5609 dev
->reg_state
= NETREG_DUMMY
;
5611 /* NAPI wants this */
5612 INIT_LIST_HEAD(&dev
->napi_list
);
5614 /* a dummy interface is started by default */
5615 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5616 set_bit(__LINK_STATE_START
, &dev
->state
);
5618 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5619 * because users of this 'device' dont need to change
5625 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
5629 * register_netdev - register a network device
5630 * @dev: device to register
5632 * Take a completed network device structure and add it to the kernel
5633 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5634 * chain. 0 is returned on success. A negative errno code is returned
5635 * on a failure to set up the device, or if the name is a duplicate.
5637 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5638 * and expands the device name if you passed a format string to
5641 int register_netdev(struct net_device
*dev
)
5646 err
= register_netdevice(dev
);
5650 EXPORT_SYMBOL(register_netdev
);
5652 int netdev_refcnt_read(const struct net_device
*dev
)
5656 for_each_possible_cpu(i
)
5657 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
5660 EXPORT_SYMBOL(netdev_refcnt_read
);
5663 * netdev_wait_allrefs - wait until all references are gone.
5665 * This is called when unregistering network devices.
5667 * Any protocol or device that holds a reference should register
5668 * for netdevice notification, and cleanup and put back the
5669 * reference if they receive an UNREGISTER event.
5670 * We can get stuck here if buggy protocols don't correctly
5673 static void netdev_wait_allrefs(struct net_device
*dev
)
5675 unsigned long rebroadcast_time
, warning_time
;
5678 linkwatch_forget_dev(dev
);
5680 rebroadcast_time
= warning_time
= jiffies
;
5681 refcnt
= netdev_refcnt_read(dev
);
5683 while (refcnt
!= 0) {
5684 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
5687 /* Rebroadcast unregister notification */
5688 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5689 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5690 * should have already handle it the first time */
5692 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
5694 /* We must not have linkwatch events
5695 * pending on unregister. If this
5696 * happens, we simply run the queue
5697 * unscheduled, resulting in a noop
5700 linkwatch_run_queue();
5705 rebroadcast_time
= jiffies
;
5710 refcnt
= netdev_refcnt_read(dev
);
5712 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
5713 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5715 warning_time
= jiffies
;
5724 * register_netdevice(x1);
5725 * register_netdevice(x2);
5727 * unregister_netdevice(y1);
5728 * unregister_netdevice(y2);
5734 * We are invoked by rtnl_unlock().
5735 * This allows us to deal with problems:
5736 * 1) We can delete sysfs objects which invoke hotplug
5737 * without deadlocking with linkwatch via keventd.
5738 * 2) Since we run with the RTNL semaphore not held, we can sleep
5739 * safely in order to wait for the netdev refcnt to drop to zero.
5741 * We must not return until all unregister events added during
5742 * the interval the lock was held have been completed.
5744 void netdev_run_todo(void)
5746 struct list_head list
;
5748 /* Snapshot list, allow later requests */
5749 list_replace_init(&net_todo_list
, &list
);
5753 /* Wait for rcu callbacks to finish before attempting to drain
5754 * the device list. This usually avoids a 250ms wait.
5756 if (!list_empty(&list
))
5759 while (!list_empty(&list
)) {
5760 struct net_device
*dev
5761 = list_first_entry(&list
, struct net_device
, todo_list
);
5762 list_del(&dev
->todo_list
);
5764 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
5765 pr_err("network todo '%s' but state %d\n",
5766 dev
->name
, dev
->reg_state
);
5771 dev
->reg_state
= NETREG_UNREGISTERED
;
5773 on_each_cpu(flush_backlog
, dev
, 1);
5775 netdev_wait_allrefs(dev
);
5778 BUG_ON(netdev_refcnt_read(dev
));
5779 WARN_ON(rcu_access_pointer(dev
->ip_ptr
));
5780 WARN_ON(rcu_access_pointer(dev
->ip6_ptr
));
5781 WARN_ON(dev
->dn_ptr
);
5783 if (dev
->destructor
)
5784 dev
->destructor(dev
);
5786 /* Free network device */
5787 kobject_put(&dev
->dev
.kobj
);
5791 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5792 * fields in the same order, with only the type differing.
5794 void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
5795 const struct net_device_stats
*netdev_stats
)
5797 #if BITS_PER_LONG == 64
5798 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
5799 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
5801 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
5802 const unsigned long *src
= (const unsigned long *)netdev_stats
;
5803 u64
*dst
= (u64
*)stats64
;
5805 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
5806 sizeof(*stats64
) / sizeof(u64
));
5807 for (i
= 0; i
< n
; i
++)
5811 EXPORT_SYMBOL(netdev_stats_to_stats64
);
5814 * dev_get_stats - get network device statistics
5815 * @dev: device to get statistics from
5816 * @storage: place to store stats
5818 * Get network statistics from device. Return @storage.
5819 * The device driver may provide its own method by setting
5820 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5821 * otherwise the internal statistics structure is used.
5823 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
5824 struct rtnl_link_stats64
*storage
)
5826 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5828 if (ops
->ndo_get_stats64
) {
5829 memset(storage
, 0, sizeof(*storage
));
5830 ops
->ndo_get_stats64(dev
, storage
);
5831 } else if (ops
->ndo_get_stats
) {
5832 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
5834 netdev_stats_to_stats64(storage
, &dev
->stats
);
5836 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
5839 EXPORT_SYMBOL(dev_get_stats
);
5841 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
5843 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
5845 #ifdef CONFIG_NET_CLS_ACT
5848 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
5851 netdev_init_one_queue(dev
, queue
, NULL
);
5852 queue
->qdisc
= &noop_qdisc
;
5853 queue
->qdisc_sleeping
= &noop_qdisc
;
5854 rcu_assign_pointer(dev
->ingress_queue
, queue
);
5860 * alloc_netdev_mqs - allocate network device
5861 * @sizeof_priv: size of private data to allocate space for
5862 * @name: device name format string
5863 * @setup: callback to initialize device
5864 * @txqs: the number of TX subqueues to allocate
5865 * @rxqs: the number of RX subqueues to allocate
5867 * Allocates a struct net_device with private data area for driver use
5868 * and performs basic initialization. Also allocates subquue structs
5869 * for each queue on the device.
5871 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
5872 void (*setup
)(struct net_device
*),
5873 unsigned int txqs
, unsigned int rxqs
)
5875 struct net_device
*dev
;
5877 struct net_device
*p
;
5879 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
5882 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
5888 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
5893 alloc_size
= sizeof(struct net_device
);
5895 /* ensure 32-byte alignment of private area */
5896 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
5897 alloc_size
+= sizeof_priv
;
5899 /* ensure 32-byte alignment of whole construct */
5900 alloc_size
+= NETDEV_ALIGN
- 1;
5902 p
= kzalloc(alloc_size
, GFP_KERNEL
);
5904 pr_err("alloc_netdev: Unable to allocate device\n");
5908 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
5909 dev
->padded
= (char *)dev
- (char *)p
;
5911 dev
->pcpu_refcnt
= alloc_percpu(int);
5912 if (!dev
->pcpu_refcnt
)
5915 if (dev_addr_init(dev
))
5921 dev_net_set(dev
, &init_net
);
5923 dev
->gso_max_size
= GSO_MAX_SIZE
;
5925 INIT_LIST_HEAD(&dev
->napi_list
);
5926 INIT_LIST_HEAD(&dev
->unreg_list
);
5927 INIT_LIST_HEAD(&dev
->link_watch_list
);
5928 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
5931 dev
->num_tx_queues
= txqs
;
5932 dev
->real_num_tx_queues
= txqs
;
5933 if (netif_alloc_netdev_queues(dev
))
5937 dev
->num_rx_queues
= rxqs
;
5938 dev
->real_num_rx_queues
= rxqs
;
5939 if (netif_alloc_rx_queues(dev
))
5943 strcpy(dev
->name
, name
);
5944 dev
->group
= INIT_NETDEV_GROUP
;
5952 free_percpu(dev
->pcpu_refcnt
);
5962 EXPORT_SYMBOL(alloc_netdev_mqs
);
5965 * free_netdev - free network device
5968 * This function does the last stage of destroying an allocated device
5969 * interface. The reference to the device object is released.
5970 * If this is the last reference then it will be freed.
5972 void free_netdev(struct net_device
*dev
)
5974 struct napi_struct
*p
, *n
;
5976 release_net(dev_net(dev
));
5983 kfree(rcu_dereference_protected(dev
->ingress_queue
, 1));
5985 /* Flush device addresses */
5986 dev_addr_flush(dev
);
5988 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
5991 free_percpu(dev
->pcpu_refcnt
);
5992 dev
->pcpu_refcnt
= NULL
;
5994 /* Compatibility with error handling in drivers */
5995 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5996 kfree((char *)dev
- dev
->padded
);
6000 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
6001 dev
->reg_state
= NETREG_RELEASED
;
6003 /* will free via device release */
6004 put_device(&dev
->dev
);
6006 EXPORT_SYMBOL(free_netdev
);
6009 * synchronize_net - Synchronize with packet receive processing
6011 * Wait for packets currently being received to be done.
6012 * Does not block later packets from starting.
6014 void synchronize_net(void)
6017 if (rtnl_is_locked())
6018 synchronize_rcu_expedited();
6022 EXPORT_SYMBOL(synchronize_net
);
6025 * unregister_netdevice_queue - remove device from the kernel
6029 * This function shuts down a device interface and removes it
6030 * from the kernel tables.
6031 * If head not NULL, device is queued to be unregistered later.
6033 * Callers must hold the rtnl semaphore. You may want
6034 * unregister_netdev() instead of this.
6037 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
6042 list_move_tail(&dev
->unreg_list
, head
);
6044 rollback_registered(dev
);
6045 /* Finish processing unregister after unlock */
6049 EXPORT_SYMBOL(unregister_netdevice_queue
);
6052 * unregister_netdevice_many - unregister many devices
6053 * @head: list of devices
6055 void unregister_netdevice_many(struct list_head
*head
)
6057 struct net_device
*dev
;
6059 if (!list_empty(head
)) {
6060 rollback_registered_many(head
);
6061 list_for_each_entry(dev
, head
, unreg_list
)
6065 EXPORT_SYMBOL(unregister_netdevice_many
);
6068 * unregister_netdev - remove device from the kernel
6071 * This function shuts down a device interface and removes it
6072 * from the kernel tables.
6074 * This is just a wrapper for unregister_netdevice that takes
6075 * the rtnl semaphore. In general you want to use this and not
6076 * unregister_netdevice.
6078 void unregister_netdev(struct net_device
*dev
)
6081 unregister_netdevice(dev
);
6084 EXPORT_SYMBOL(unregister_netdev
);
6087 * dev_change_net_namespace - move device to different nethost namespace
6089 * @net: network namespace
6090 * @pat: If not NULL name pattern to try if the current device name
6091 * is already taken in the destination network namespace.
6093 * This function shuts down a device interface and moves it
6094 * to a new network namespace. On success 0 is returned, on
6095 * a failure a netagive errno code is returned.
6097 * Callers must hold the rtnl semaphore.
6100 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
6106 /* Don't allow namespace local devices to be moved. */
6108 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6111 /* Ensure the device has been registrered */
6113 if (dev
->reg_state
!= NETREG_REGISTERED
)
6116 /* Get out if there is nothing todo */
6118 if (net_eq(dev_net(dev
), net
))
6121 /* Pick the destination device name, and ensure
6122 * we can use it in the destination network namespace.
6125 if (__dev_get_by_name(net
, dev
->name
)) {
6126 /* We get here if we can't use the current device name */
6129 if (dev_get_valid_name(dev
, pat
) < 0)
6134 * And now a mini version of register_netdevice unregister_netdevice.
6137 /* If device is running close it first. */
6140 /* And unlink it from device chain */
6142 unlist_netdevice(dev
);
6146 /* Shutdown queueing discipline. */
6149 /* Notify protocols, that we are about to destroy
6150 this device. They should clean all the things.
6152 Note that dev->reg_state stays at NETREG_REGISTERED.
6153 This is wanted because this way 8021q and macvlan know
6154 the device is just moving and can keep their slaves up.
6156 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6157 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
6158 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
6161 * Flush the unicast and multicast chains
6166 /* Actually switch the network namespace */
6167 dev_net_set(dev
, net
);
6169 /* If there is an ifindex conflict assign a new one */
6170 if (__dev_get_by_index(net
, dev
->ifindex
)) {
6171 int iflink
= (dev
->iflink
== dev
->ifindex
);
6172 dev
->ifindex
= dev_new_index(net
);
6174 dev
->iflink
= dev
->ifindex
;
6177 /* Fixup kobjects */
6178 err
= device_rename(&dev
->dev
, dev
->name
);
6181 /* Add the device back in the hashes */
6182 list_netdevice(dev
);
6184 /* Notify protocols, that a new device appeared. */
6185 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
6188 * Prevent userspace races by waiting until the network
6189 * device is fully setup before sending notifications.
6191 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
6198 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
6200 static int dev_cpu_callback(struct notifier_block
*nfb
,
6201 unsigned long action
,
6204 struct sk_buff
**list_skb
;
6205 struct sk_buff
*skb
;
6206 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
6207 struct softnet_data
*sd
, *oldsd
;
6209 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
6212 local_irq_disable();
6213 cpu
= smp_processor_id();
6214 sd
= &per_cpu(softnet_data
, cpu
);
6215 oldsd
= &per_cpu(softnet_data
, oldcpu
);
6217 /* Find end of our completion_queue. */
6218 list_skb
= &sd
->completion_queue
;
6220 list_skb
= &(*list_skb
)->next
;
6221 /* Append completion queue from offline CPU. */
6222 *list_skb
= oldsd
->completion_queue
;
6223 oldsd
->completion_queue
= NULL
;
6225 /* Append output queue from offline CPU. */
6226 if (oldsd
->output_queue
) {
6227 *sd
->output_queue_tailp
= oldsd
->output_queue
;
6228 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
6229 oldsd
->output_queue
= NULL
;
6230 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
6232 /* Append NAPI poll list from offline CPU. */
6233 if (!list_empty(&oldsd
->poll_list
)) {
6234 list_splice_init(&oldsd
->poll_list
, &sd
->poll_list
);
6235 raise_softirq_irqoff(NET_RX_SOFTIRQ
);
6238 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
6241 /* Process offline CPU's input_pkt_queue */
6242 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
6244 input_queue_head_incr(oldsd
);
6246 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
6248 input_queue_head_incr(oldsd
);
6256 * netdev_increment_features - increment feature set by one
6257 * @all: current feature set
6258 * @one: new feature set
6259 * @mask: mask feature set
6261 * Computes a new feature set after adding a device with feature set
6262 * @one to the master device with current feature set @all. Will not
6263 * enable anything that is off in @mask. Returns the new feature set.
6265 netdev_features_t
netdev_increment_features(netdev_features_t all
,
6266 netdev_features_t one
, netdev_features_t mask
)
6268 if (mask
& NETIF_F_GEN_CSUM
)
6269 mask
|= NETIF_F_ALL_CSUM
;
6270 mask
|= NETIF_F_VLAN_CHALLENGED
;
6272 all
|= one
& (NETIF_F_ONE_FOR_ALL
|NETIF_F_ALL_CSUM
) & mask
;
6273 all
&= one
| ~NETIF_F_ALL_FOR_ALL
;
6275 /* If one device supports hw checksumming, set for all. */
6276 if (all
& NETIF_F_GEN_CSUM
)
6277 all
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_GEN_CSUM
);
6281 EXPORT_SYMBOL(netdev_increment_features
);
6283 static struct hlist_head
*netdev_create_hash(void)
6286 struct hlist_head
*hash
;
6288 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
6290 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
6291 INIT_HLIST_HEAD(&hash
[i
]);
6296 /* Initialize per network namespace state */
6297 static int __net_init
netdev_init(struct net
*net
)
6299 INIT_LIST_HEAD(&net
->dev_base_head
);
6301 net
->dev_name_head
= netdev_create_hash();
6302 if (net
->dev_name_head
== NULL
)
6305 net
->dev_index_head
= netdev_create_hash();
6306 if (net
->dev_index_head
== NULL
)
6312 kfree(net
->dev_name_head
);
6318 * netdev_drivername - network driver for the device
6319 * @dev: network device
6321 * Determine network driver for device.
6323 const char *netdev_drivername(const struct net_device
*dev
)
6325 const struct device_driver
*driver
;
6326 const struct device
*parent
;
6327 const char *empty
= "";
6329 parent
= dev
->dev
.parent
;
6333 driver
= parent
->driver
;
6334 if (driver
&& driver
->name
)
6335 return driver
->name
;
6339 int __netdev_printk(const char *level
, const struct net_device
*dev
,
6340 struct va_format
*vaf
)
6344 if (dev
&& dev
->dev
.parent
)
6345 r
= dev_printk(level
, dev
->dev
.parent
, "%s: %pV",
6346 netdev_name(dev
), vaf
);
6348 r
= printk("%s%s: %pV", level
, netdev_name(dev
), vaf
);
6350 r
= printk("%s(NULL net_device): %pV", level
, vaf
);
6354 EXPORT_SYMBOL(__netdev_printk
);
6356 int netdev_printk(const char *level
, const struct net_device
*dev
,
6357 const char *format
, ...)
6359 struct va_format vaf
;
6363 va_start(args
, format
);
6368 r
= __netdev_printk(level
, dev
, &vaf
);
6373 EXPORT_SYMBOL(netdev_printk
);
6375 #define define_netdev_printk_level(func, level) \
6376 int func(const struct net_device *dev, const char *fmt, ...) \
6379 struct va_format vaf; \
6382 va_start(args, fmt); \
6387 r = __netdev_printk(level, dev, &vaf); \
6392 EXPORT_SYMBOL(func);
6394 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
6395 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
6396 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
6397 define_netdev_printk_level(netdev_err
, KERN_ERR
);
6398 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
6399 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
6400 define_netdev_printk_level(netdev_info
, KERN_INFO
);
6402 static void __net_exit
netdev_exit(struct net
*net
)
6404 kfree(net
->dev_name_head
);
6405 kfree(net
->dev_index_head
);
6408 static struct pernet_operations __net_initdata netdev_net_ops
= {
6409 .init
= netdev_init
,
6410 .exit
= netdev_exit
,
6413 static void __net_exit
default_device_exit(struct net
*net
)
6415 struct net_device
*dev
, *aux
;
6417 * Push all migratable network devices back to the
6418 * initial network namespace
6421 for_each_netdev_safe(net
, dev
, aux
) {
6423 char fb_name
[IFNAMSIZ
];
6425 /* Ignore unmoveable devices (i.e. loopback) */
6426 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6429 /* Leave virtual devices for the generic cleanup */
6430 if (dev
->rtnl_link_ops
)
6433 /* Push remaining network devices to init_net */
6434 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
6435 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
6437 pr_emerg("%s: failed to move %s to init_net: %d\n",
6438 __func__
, dev
->name
, err
);
6445 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
6447 /* At exit all network devices most be removed from a network
6448 * namespace. Do this in the reverse order of registration.
6449 * Do this across as many network namespaces as possible to
6450 * improve batching efficiency.
6452 struct net_device
*dev
;
6454 LIST_HEAD(dev_kill_list
);
6457 list_for_each_entry(net
, net_list
, exit_list
) {
6458 for_each_netdev_reverse(net
, dev
) {
6459 if (dev
->rtnl_link_ops
)
6460 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
6462 unregister_netdevice_queue(dev
, &dev_kill_list
);
6465 unregister_netdevice_many(&dev_kill_list
);
6466 list_del(&dev_kill_list
);
6470 static struct pernet_operations __net_initdata default_device_ops
= {
6471 .exit
= default_device_exit
,
6472 .exit_batch
= default_device_exit_batch
,
6476 * Initialize the DEV module. At boot time this walks the device list and
6477 * unhooks any devices that fail to initialise (normally hardware not
6478 * present) and leaves us with a valid list of present and active devices.
6483 * This is called single threaded during boot, so no need
6484 * to take the rtnl semaphore.
6486 static int __init
net_dev_init(void)
6488 int i
, rc
= -ENOMEM
;
6490 BUG_ON(!dev_boot_phase
);
6492 if (dev_proc_init())
6495 if (netdev_kobject_init())
6498 INIT_LIST_HEAD(&ptype_all
);
6499 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
6500 INIT_LIST_HEAD(&ptype_base
[i
]);
6502 if (register_pernet_subsys(&netdev_net_ops
))
6506 * Initialise the packet receive queues.
6509 for_each_possible_cpu(i
) {
6510 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
6512 memset(sd
, 0, sizeof(*sd
));
6513 skb_queue_head_init(&sd
->input_pkt_queue
);
6514 skb_queue_head_init(&sd
->process_queue
);
6515 sd
->completion_queue
= NULL
;
6516 INIT_LIST_HEAD(&sd
->poll_list
);
6517 sd
->output_queue
= NULL
;
6518 sd
->output_queue_tailp
= &sd
->output_queue
;
6520 sd
->csd
.func
= rps_trigger_softirq
;
6526 sd
->backlog
.poll
= process_backlog
;
6527 sd
->backlog
.weight
= weight_p
;
6528 sd
->backlog
.gro_list
= NULL
;
6529 sd
->backlog
.gro_count
= 0;
6534 /* The loopback device is special if any other network devices
6535 * is present in a network namespace the loopback device must
6536 * be present. Since we now dynamically allocate and free the
6537 * loopback device ensure this invariant is maintained by
6538 * keeping the loopback device as the first device on the
6539 * list of network devices. Ensuring the loopback devices
6540 * is the first device that appears and the last network device
6543 if (register_pernet_device(&loopback_net_ops
))
6546 if (register_pernet_device(&default_device_ops
))
6549 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
6550 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
6552 hotcpu_notifier(dev_cpu_callback
, 0);
6560 subsys_initcall(net_dev_init
);
6562 static int __init
initialize_hashrnd(void)
6564 get_random_bytes(&hashrnd
, sizeof(hashrnd
));
6568 late_initcall_sync(initialize_hashrnd
);