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_alloc_name_ns(struct net
*net
,
963 struct net_device
*dev
,
969 ret
= __dev_alloc_name(net
, name
, buf
);
971 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
975 static int dev_get_valid_name(struct net
*net
,
976 struct net_device
*dev
,
981 if (!dev_valid_name(name
))
984 if (strchr(name
, '%'))
985 return dev_alloc_name_ns(net
, dev
, name
);
986 else if (__dev_get_by_name(net
, name
))
988 else if (dev
->name
!= name
)
989 strlcpy(dev
->name
, name
, IFNAMSIZ
);
995 * dev_change_name - change name of a device
997 * @newname: name (or format string) must be at least IFNAMSIZ
999 * Change name of a device, can pass format strings "eth%d".
1002 int dev_change_name(struct net_device
*dev
, const char *newname
)
1004 char oldname
[IFNAMSIZ
];
1010 BUG_ON(!dev_net(dev
));
1013 if (dev
->flags
& IFF_UP
)
1016 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0)
1019 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
1021 err
= dev_get_valid_name(net
, dev
, newname
);
1026 ret
= device_rename(&dev
->dev
, dev
->name
);
1028 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1032 write_lock_bh(&dev_base_lock
);
1033 hlist_del_rcu(&dev
->name_hlist
);
1034 write_unlock_bh(&dev_base_lock
);
1038 write_lock_bh(&dev_base_lock
);
1039 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
1040 write_unlock_bh(&dev_base_lock
);
1042 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
1043 ret
= notifier_to_errno(ret
);
1046 /* err >= 0 after dev_alloc_name() or stores the first errno */
1049 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1052 pr_err("%s: name change rollback failed: %d\n",
1061 * dev_set_alias - change ifalias of a device
1063 * @alias: name up to IFALIASZ
1064 * @len: limit of bytes to copy from info
1066 * Set ifalias for a device,
1068 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
1074 if (len
>= IFALIASZ
)
1079 kfree(dev
->ifalias
);
1080 dev
->ifalias
= NULL
;
1085 new_ifalias
= krealloc(dev
->ifalias
, len
+ 1, GFP_KERNEL
);
1088 dev
->ifalias
= new_ifalias
;
1090 strlcpy(dev
->ifalias
, alias
, len
+1);
1096 * netdev_features_change - device changes features
1097 * @dev: device to cause notification
1099 * Called to indicate a device has changed features.
1101 void netdev_features_change(struct net_device
*dev
)
1103 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1105 EXPORT_SYMBOL(netdev_features_change
);
1108 * netdev_state_change - device changes state
1109 * @dev: device to cause notification
1111 * Called to indicate a device has changed state. This function calls
1112 * the notifier chains for netdev_chain and sends a NEWLINK message
1113 * to the routing socket.
1115 void netdev_state_change(struct net_device
*dev
)
1117 if (dev
->flags
& IFF_UP
) {
1118 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
1119 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
1122 EXPORT_SYMBOL(netdev_state_change
);
1125 * netdev_notify_peers - notify network peers about existence of @dev
1126 * @dev: network device
1128 * Generate traffic such that interested network peers are aware of
1129 * @dev, such as by generating a gratuitous ARP. This may be used when
1130 * a device wants to inform the rest of the network about some sort of
1131 * reconfiguration such as a failover event or virtual machine
1134 void netdev_notify_peers(struct net_device
*dev
)
1137 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
, dev
);
1140 EXPORT_SYMBOL(netdev_notify_peers
);
1143 * dev_load - load a network module
1144 * @net: the applicable net namespace
1145 * @name: name of interface
1147 * If a network interface is not present and the process has suitable
1148 * privileges this function loads the module. If module loading is not
1149 * available in this kernel then it becomes a nop.
1152 void dev_load(struct net
*net
, const char *name
)
1154 struct net_device
*dev
;
1158 dev
= dev_get_by_name_rcu(net
, name
);
1162 if (no_module
&& capable(CAP_NET_ADMIN
))
1163 no_module
= request_module("netdev-%s", name
);
1164 if (no_module
&& capable(CAP_SYS_MODULE
)) {
1165 if (!request_module("%s", name
))
1166 pr_warn("Loading kernel module for a network device with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s instead.\n",
1170 EXPORT_SYMBOL(dev_load
);
1172 static int __dev_open(struct net_device
*dev
)
1174 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1179 if (!netif_device_present(dev
))
1182 ret
= call_netdevice_notifiers(NETDEV_PRE_UP
, dev
);
1183 ret
= notifier_to_errno(ret
);
1187 set_bit(__LINK_STATE_START
, &dev
->state
);
1189 if (ops
->ndo_validate_addr
)
1190 ret
= ops
->ndo_validate_addr(dev
);
1192 if (!ret
&& ops
->ndo_open
)
1193 ret
= ops
->ndo_open(dev
);
1196 clear_bit(__LINK_STATE_START
, &dev
->state
);
1198 dev
->flags
|= IFF_UP
;
1199 net_dmaengine_get();
1200 dev_set_rx_mode(dev
);
1202 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
1209 * dev_open - prepare an interface for use.
1210 * @dev: device to open
1212 * Takes a device from down to up state. The device's private open
1213 * function is invoked and then the multicast lists are loaded. Finally
1214 * the device is moved into the up state and a %NETDEV_UP message is
1215 * sent to the netdev notifier chain.
1217 * Calling this function on an active interface is a nop. On a failure
1218 * a negative errno code is returned.
1220 int dev_open(struct net_device
*dev
)
1224 if (dev
->flags
& IFF_UP
)
1227 ret
= __dev_open(dev
);
1231 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1232 call_netdevice_notifiers(NETDEV_UP
, dev
);
1236 EXPORT_SYMBOL(dev_open
);
1238 static int __dev_close_many(struct list_head
*head
)
1240 struct net_device
*dev
;
1245 list_for_each_entry(dev
, head
, unreg_list
) {
1246 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1248 clear_bit(__LINK_STATE_START
, &dev
->state
);
1250 /* Synchronize to scheduled poll. We cannot touch poll list, it
1251 * can be even on different cpu. So just clear netif_running().
1253 * dev->stop() will invoke napi_disable() on all of it's
1254 * napi_struct instances on this device.
1256 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1259 dev_deactivate_many(head
);
1261 list_for_each_entry(dev
, head
, unreg_list
) {
1262 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1265 * Call the device specific close. This cannot fail.
1266 * Only if device is UP
1268 * We allow it to be called even after a DETACH hot-plug
1274 dev
->flags
&= ~IFF_UP
;
1275 net_dmaengine_put();
1281 static int __dev_close(struct net_device
*dev
)
1286 list_add(&dev
->unreg_list
, &single
);
1287 retval
= __dev_close_many(&single
);
1292 static int dev_close_many(struct list_head
*head
)
1294 struct net_device
*dev
, *tmp
;
1295 LIST_HEAD(tmp_list
);
1297 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
)
1298 if (!(dev
->flags
& IFF_UP
))
1299 list_move(&dev
->unreg_list
, &tmp_list
);
1301 __dev_close_many(head
);
1303 list_for_each_entry(dev
, head
, unreg_list
) {
1304 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1305 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1308 /* rollback_registered_many needs the complete original list */
1309 list_splice(&tmp_list
, head
);
1314 * dev_close - shutdown an interface.
1315 * @dev: device to shutdown
1317 * This function moves an active device into down state. A
1318 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1319 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1322 int dev_close(struct net_device
*dev
)
1324 if (dev
->flags
& IFF_UP
) {
1327 list_add(&dev
->unreg_list
, &single
);
1328 dev_close_many(&single
);
1333 EXPORT_SYMBOL(dev_close
);
1337 * dev_disable_lro - disable Large Receive Offload on a device
1340 * Disable Large Receive Offload (LRO) on a net device. Must be
1341 * called under RTNL. This is needed if received packets may be
1342 * forwarded to another interface.
1344 void dev_disable_lro(struct net_device
*dev
)
1347 * If we're trying to disable lro on a vlan device
1348 * use the underlying physical device instead
1350 if (is_vlan_dev(dev
))
1351 dev
= vlan_dev_real_dev(dev
);
1353 dev
->wanted_features
&= ~NETIF_F_LRO
;
1354 netdev_update_features(dev
);
1356 if (unlikely(dev
->features
& NETIF_F_LRO
))
1357 netdev_WARN(dev
, "failed to disable LRO!\n");
1359 EXPORT_SYMBOL(dev_disable_lro
);
1362 static int dev_boot_phase
= 1;
1365 * register_netdevice_notifier - register a network notifier block
1368 * Register a notifier to be called when network device events occur.
1369 * The notifier passed is linked into the kernel structures and must
1370 * not be reused until it has been unregistered. A negative errno code
1371 * is returned on a failure.
1373 * When registered all registration and up events are replayed
1374 * to the new notifier to allow device to have a race free
1375 * view of the network device list.
1378 int register_netdevice_notifier(struct notifier_block
*nb
)
1380 struct net_device
*dev
;
1381 struct net_device
*last
;
1386 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1392 for_each_netdev(net
, dev
) {
1393 err
= nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1394 err
= notifier_to_errno(err
);
1398 if (!(dev
->flags
& IFF_UP
))
1401 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1412 for_each_netdev(net
, dev
) {
1416 if (dev
->flags
& IFF_UP
) {
1417 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1418 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1420 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1425 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1428 EXPORT_SYMBOL(register_netdevice_notifier
);
1431 * unregister_netdevice_notifier - unregister a network notifier block
1434 * Unregister a notifier previously registered by
1435 * register_netdevice_notifier(). The notifier is unlinked into the
1436 * kernel structures and may then be reused. A negative errno code
1437 * is returned on a failure.
1439 * After unregistering unregister and down device events are synthesized
1440 * for all devices on the device list to the removed notifier to remove
1441 * the need for special case cleanup code.
1444 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1446 struct net_device
*dev
;
1451 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1456 for_each_netdev(net
, dev
) {
1457 if (dev
->flags
& IFF_UP
) {
1458 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1459 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1461 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1468 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1471 * call_netdevice_notifiers - call all network notifier blocks
1472 * @val: value passed unmodified to notifier function
1473 * @dev: net_device pointer passed unmodified to notifier function
1475 * Call all network notifier blocks. Parameters and return value
1476 * are as for raw_notifier_call_chain().
1479 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1482 return raw_notifier_call_chain(&netdev_chain
, val
, dev
);
1484 EXPORT_SYMBOL(call_netdevice_notifiers
);
1486 static struct static_key netstamp_needed __read_mostly
;
1487 #ifdef HAVE_JUMP_LABEL
1488 /* We are not allowed to call static_key_slow_dec() from irq context
1489 * If net_disable_timestamp() is called from irq context, defer the
1490 * static_key_slow_dec() calls.
1492 static atomic_t netstamp_needed_deferred
;
1495 void net_enable_timestamp(void)
1497 #ifdef HAVE_JUMP_LABEL
1498 int deferred
= atomic_xchg(&netstamp_needed_deferred
, 0);
1502 static_key_slow_dec(&netstamp_needed
);
1506 WARN_ON(in_interrupt());
1507 static_key_slow_inc(&netstamp_needed
);
1509 EXPORT_SYMBOL(net_enable_timestamp
);
1511 void net_disable_timestamp(void)
1513 #ifdef HAVE_JUMP_LABEL
1514 if (in_interrupt()) {
1515 atomic_inc(&netstamp_needed_deferred
);
1519 static_key_slow_dec(&netstamp_needed
);
1521 EXPORT_SYMBOL(net_disable_timestamp
);
1523 static inline void net_timestamp_set(struct sk_buff
*skb
)
1525 skb
->tstamp
.tv64
= 0;
1526 if (static_key_false(&netstamp_needed
))
1527 __net_timestamp(skb
);
1530 #define net_timestamp_check(COND, SKB) \
1531 if (static_key_false(&netstamp_needed)) { \
1532 if ((COND) && !(SKB)->tstamp.tv64) \
1533 __net_timestamp(SKB); \
1536 static int net_hwtstamp_validate(struct ifreq *ifr)
1538 struct hwtstamp_config cfg
;
1539 enum hwtstamp_tx_types tx_type
;
1540 enum hwtstamp_rx_filters rx_filter
;
1541 int tx_type_valid
= 0;
1542 int rx_filter_valid
= 0;
1544 if (copy_from_user(&cfg
, ifr
->ifr_data
, sizeof(cfg
)))
1547 if (cfg
.flags
) /* reserved for future extensions */
1550 tx_type
= cfg
.tx_type
;
1551 rx_filter
= cfg
.rx_filter
;
1554 case HWTSTAMP_TX_OFF
:
1555 case HWTSTAMP_TX_ON
:
1556 case HWTSTAMP_TX_ONESTEP_SYNC
:
1561 switch (rx_filter
) {
1562 case HWTSTAMP_FILTER_NONE
:
1563 case HWTSTAMP_FILTER_ALL
:
1564 case HWTSTAMP_FILTER_SOME
:
1565 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT
:
1566 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC
:
1567 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ
:
1568 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT
:
1569 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC
:
1570 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ
:
1571 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT
:
1572 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC
:
1573 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ
:
1574 case HWTSTAMP_FILTER_PTP_V2_EVENT
:
1575 case HWTSTAMP_FILTER_PTP_V2_SYNC
:
1576 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ
:
1577 rx_filter_valid
= 1;
1581 if (!tx_type_valid
|| !rx_filter_valid
)
1587 static inline bool is_skb_forwardable(struct net_device
*dev
,
1588 struct sk_buff
*skb
)
1592 if (!(dev
->flags
& IFF_UP
))
1595 len
= dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
;
1596 if (skb
->len
<= len
)
1599 /* if TSO is enabled, we don't care about the length as the packet
1600 * could be forwarded without being segmented before
1602 if (skb_is_gso(skb
))
1609 * dev_forward_skb - loopback an skb to another netif
1611 * @dev: destination network device
1612 * @skb: buffer to forward
1615 * NET_RX_SUCCESS (no congestion)
1616 * NET_RX_DROP (packet was dropped, but freed)
1618 * dev_forward_skb can be used for injecting an skb from the
1619 * start_xmit function of one device into the receive queue
1620 * of another device.
1622 * The receiving device may be in another namespace, so
1623 * we have to clear all information in the skb that could
1624 * impact namespace isolation.
1626 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1628 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
1629 if (skb_copy_ubufs(skb
, GFP_ATOMIC
)) {
1630 atomic_long_inc(&dev
->rx_dropped
);
1639 if (unlikely(!is_skb_forwardable(dev
, skb
))) {
1640 atomic_long_inc(&dev
->rx_dropped
);
1647 skb
->tstamp
.tv64
= 0;
1648 skb
->pkt_type
= PACKET_HOST
;
1649 skb
->protocol
= eth_type_trans(skb
, dev
);
1653 return netif_rx(skb
);
1655 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1657 static inline int deliver_skb(struct sk_buff
*skb
,
1658 struct packet_type
*pt_prev
,
1659 struct net_device
*orig_dev
)
1661 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
1663 atomic_inc(&skb
->users
);
1664 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1667 static inline bool skb_loop_sk(struct packet_type
*ptype
, struct sk_buff
*skb
)
1669 if (ptype
->af_packet_priv
== NULL
)
1672 if (ptype
->id_match
)
1673 return ptype
->id_match(ptype
, skb
->sk
);
1674 else if ((struct sock
*)ptype
->af_packet_priv
== skb
->sk
)
1681 * Support routine. Sends outgoing frames to any network
1682 * taps currently in use.
1685 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1687 struct packet_type
*ptype
;
1688 struct sk_buff
*skb2
= NULL
;
1689 struct packet_type
*pt_prev
= NULL
;
1692 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1693 /* Never send packets back to the socket
1694 * they originated from - MvS (miquels@drinkel.ow.org)
1696 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1697 (!skb_loop_sk(ptype
, skb
))) {
1699 deliver_skb(skb2
, pt_prev
, skb
->dev
);
1704 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1708 net_timestamp_set(skb2
);
1710 /* skb->nh should be correctly
1711 set by sender, so that the second statement is
1712 just protection against buggy protocols.
1714 skb_reset_mac_header(skb2
);
1716 if (skb_network_header(skb2
) < skb2
->data
||
1717 skb2
->network_header
> skb2
->tail
) {
1718 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1719 ntohs(skb2
->protocol
),
1721 skb_reset_network_header(skb2
);
1724 skb2
->transport_header
= skb2
->network_header
;
1725 skb2
->pkt_type
= PACKET_OUTGOING
;
1730 pt_prev
->func(skb2
, skb
->dev
, pt_prev
, skb
->dev
);
1735 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1736 * @dev: Network device
1737 * @txq: number of queues available
1739 * If real_num_tx_queues is changed the tc mappings may no longer be
1740 * valid. To resolve this verify the tc mapping remains valid and if
1741 * not NULL the mapping. With no priorities mapping to this
1742 * offset/count pair it will no longer be used. In the worst case TC0
1743 * is invalid nothing can be done so disable priority mappings. If is
1744 * expected that drivers will fix this mapping if they can before
1745 * calling netif_set_real_num_tx_queues.
1747 static void netif_setup_tc(struct net_device
*dev
, unsigned int txq
)
1750 struct netdev_tc_txq
*tc
= &dev
->tc_to_txq
[0];
1752 /* If TC0 is invalidated disable TC mapping */
1753 if (tc
->offset
+ tc
->count
> txq
) {
1754 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1759 /* Invalidated prio to tc mappings set to TC0 */
1760 for (i
= 1; i
< TC_BITMASK
+ 1; i
++) {
1761 int q
= netdev_get_prio_tc_map(dev
, i
);
1763 tc
= &dev
->tc_to_txq
[q
];
1764 if (tc
->offset
+ tc
->count
> txq
) {
1765 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1767 netdev_set_prio_tc_map(dev
, i
, 0);
1773 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1774 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1776 int netif_set_real_num_tx_queues(struct net_device
*dev
, unsigned int txq
)
1780 if (txq
< 1 || txq
> dev
->num_tx_queues
)
1783 if (dev
->reg_state
== NETREG_REGISTERED
||
1784 dev
->reg_state
== NETREG_UNREGISTERING
) {
1787 rc
= netdev_queue_update_kobjects(dev
, dev
->real_num_tx_queues
,
1793 netif_setup_tc(dev
, txq
);
1795 if (txq
< dev
->real_num_tx_queues
)
1796 qdisc_reset_all_tx_gt(dev
, txq
);
1799 dev
->real_num_tx_queues
= txq
;
1802 EXPORT_SYMBOL(netif_set_real_num_tx_queues
);
1806 * netif_set_real_num_rx_queues - set actual number of RX queues used
1807 * @dev: Network device
1808 * @rxq: Actual number of RX queues
1810 * This must be called either with the rtnl_lock held or before
1811 * registration of the net device. Returns 0 on success, or a
1812 * negative error code. If called before registration, it always
1815 int netif_set_real_num_rx_queues(struct net_device
*dev
, unsigned int rxq
)
1819 if (rxq
< 1 || rxq
> dev
->num_rx_queues
)
1822 if (dev
->reg_state
== NETREG_REGISTERED
) {
1825 rc
= net_rx_queue_update_kobjects(dev
, dev
->real_num_rx_queues
,
1831 dev
->real_num_rx_queues
= rxq
;
1834 EXPORT_SYMBOL(netif_set_real_num_rx_queues
);
1838 * netif_get_num_default_rss_queues - default number of RSS queues
1840 * This routine should set an upper limit on the number of RSS queues
1841 * used by default by multiqueue devices.
1843 int netif_get_num_default_rss_queues(void)
1845 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES
, num_online_cpus());
1847 EXPORT_SYMBOL(netif_get_num_default_rss_queues
);
1849 static inline void __netif_reschedule(struct Qdisc
*q
)
1851 struct softnet_data
*sd
;
1852 unsigned long flags
;
1854 local_irq_save(flags
);
1855 sd
= &__get_cpu_var(softnet_data
);
1856 q
->next_sched
= NULL
;
1857 *sd
->output_queue_tailp
= q
;
1858 sd
->output_queue_tailp
= &q
->next_sched
;
1859 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1860 local_irq_restore(flags
);
1863 void __netif_schedule(struct Qdisc
*q
)
1865 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
1866 __netif_reschedule(q
);
1868 EXPORT_SYMBOL(__netif_schedule
);
1870 void dev_kfree_skb_irq(struct sk_buff
*skb
)
1872 if (atomic_dec_and_test(&skb
->users
)) {
1873 struct softnet_data
*sd
;
1874 unsigned long flags
;
1876 local_irq_save(flags
);
1877 sd
= &__get_cpu_var(softnet_data
);
1878 skb
->next
= sd
->completion_queue
;
1879 sd
->completion_queue
= skb
;
1880 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1881 local_irq_restore(flags
);
1884 EXPORT_SYMBOL(dev_kfree_skb_irq
);
1886 void dev_kfree_skb_any(struct sk_buff
*skb
)
1888 if (in_irq() || irqs_disabled())
1889 dev_kfree_skb_irq(skb
);
1893 EXPORT_SYMBOL(dev_kfree_skb_any
);
1897 * netif_device_detach - mark device as removed
1898 * @dev: network device
1900 * Mark device as removed from system and therefore no longer available.
1902 void netif_device_detach(struct net_device
*dev
)
1904 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1905 netif_running(dev
)) {
1906 netif_tx_stop_all_queues(dev
);
1909 EXPORT_SYMBOL(netif_device_detach
);
1912 * netif_device_attach - mark device as attached
1913 * @dev: network device
1915 * Mark device as attached from system and restart if needed.
1917 void netif_device_attach(struct net_device
*dev
)
1919 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1920 netif_running(dev
)) {
1921 netif_tx_wake_all_queues(dev
);
1922 __netdev_watchdog_up(dev
);
1925 EXPORT_SYMBOL(netif_device_attach
);
1927 static void skb_warn_bad_offload(const struct sk_buff
*skb
)
1929 static const netdev_features_t null_features
= 0;
1930 struct net_device
*dev
= skb
->dev
;
1931 const char *driver
= "";
1933 if (dev
&& dev
->dev
.parent
)
1934 driver
= dev_driver_string(dev
->dev
.parent
);
1936 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
1937 "gso_type=%d ip_summed=%d\n",
1938 driver
, dev
? &dev
->features
: &null_features
,
1939 skb
->sk
? &skb
->sk
->sk_route_caps
: &null_features
,
1940 skb
->len
, skb
->data_len
, skb_shinfo(skb
)->gso_size
,
1941 skb_shinfo(skb
)->gso_type
, skb
->ip_summed
);
1945 * Invalidate hardware checksum when packet is to be mangled, and
1946 * complete checksum manually on outgoing path.
1948 int skb_checksum_help(struct sk_buff
*skb
)
1951 int ret
= 0, offset
;
1953 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1954 goto out_set_summed
;
1956 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1957 skb_warn_bad_offload(skb
);
1961 offset
= skb_checksum_start_offset(skb
);
1962 BUG_ON(offset
>= skb_headlen(skb
));
1963 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1965 offset
+= skb
->csum_offset
;
1966 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
1968 if (skb_cloned(skb
) &&
1969 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
1970 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1975 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1977 skb
->ip_summed
= CHECKSUM_NONE
;
1981 EXPORT_SYMBOL(skb_checksum_help
);
1984 * skb_gso_segment - Perform segmentation on skb.
1985 * @skb: buffer to segment
1986 * @features: features for the output path (see dev->features)
1988 * This function segments the given skb and returns a list of segments.
1990 * It may return NULL if the skb requires no segmentation. This is
1991 * only possible when GSO is used for verifying header integrity.
1993 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
,
1994 netdev_features_t features
)
1996 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1997 struct packet_type
*ptype
;
1998 __be16 type
= skb
->protocol
;
1999 int vlan_depth
= ETH_HLEN
;
2002 while (type
== htons(ETH_P_8021Q
)) {
2003 struct vlan_hdr
*vh
;
2005 if (unlikely(!pskb_may_pull(skb
, vlan_depth
+ VLAN_HLEN
)))
2006 return ERR_PTR(-EINVAL
);
2008 vh
= (struct vlan_hdr
*)(skb
->data
+ vlan_depth
);
2009 type
= vh
->h_vlan_encapsulated_proto
;
2010 vlan_depth
+= VLAN_HLEN
;
2013 skb_reset_mac_header(skb
);
2014 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
2015 __skb_pull(skb
, skb
->mac_len
);
2017 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
2018 skb_warn_bad_offload(skb
);
2020 if (skb_header_cloned(skb
) &&
2021 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
2022 return ERR_PTR(err
);
2026 list_for_each_entry_rcu(ptype
,
2027 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
2028 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
2029 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
2030 err
= ptype
->gso_send_check(skb
);
2031 segs
= ERR_PTR(err
);
2032 if (err
|| skb_gso_ok(skb
, features
))
2034 __skb_push(skb
, (skb
->data
-
2035 skb_network_header(skb
)));
2037 segs
= ptype
->gso_segment(skb
, features
);
2043 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
2047 EXPORT_SYMBOL(skb_gso_segment
);
2049 /* Take action when hardware reception checksum errors are detected. */
2051 void netdev_rx_csum_fault(struct net_device
*dev
)
2053 if (net_ratelimit()) {
2054 pr_err("%s: hw csum failure\n", dev
? dev
->name
: "<unknown>");
2058 EXPORT_SYMBOL(netdev_rx_csum_fault
);
2061 /* Actually, we should eliminate this check as soon as we know, that:
2062 * 1. IOMMU is present and allows to map all the memory.
2063 * 2. No high memory really exists on this machine.
2066 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
2068 #ifdef CONFIG_HIGHMEM
2070 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
2071 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2072 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2073 if (PageHighMem(skb_frag_page(frag
)))
2078 if (PCI_DMA_BUS_IS_PHYS
) {
2079 struct device
*pdev
= dev
->dev
.parent
;
2083 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2084 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2085 dma_addr_t addr
= page_to_phys(skb_frag_page(frag
));
2086 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
2095 void (*destructor
)(struct sk_buff
*skb
);
2098 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2100 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
2102 struct dev_gso_cb
*cb
;
2105 struct sk_buff
*nskb
= skb
->next
;
2107 skb
->next
= nskb
->next
;
2110 } while (skb
->next
);
2112 cb
= DEV_GSO_CB(skb
);
2114 cb
->destructor(skb
);
2118 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2119 * @skb: buffer to segment
2120 * @features: device features as applicable to this skb
2122 * This function segments the given skb and stores the list of segments
2125 static int dev_gso_segment(struct sk_buff
*skb
, netdev_features_t features
)
2127 struct sk_buff
*segs
;
2129 segs
= skb_gso_segment(skb
, features
);
2131 /* Verifying header integrity only. */
2136 return PTR_ERR(segs
);
2139 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
2140 skb
->destructor
= dev_gso_skb_destructor
;
2145 static bool can_checksum_protocol(netdev_features_t features
, __be16 protocol
)
2147 return ((features
& NETIF_F_GEN_CSUM
) ||
2148 ((features
& NETIF_F_V4_CSUM
) &&
2149 protocol
== htons(ETH_P_IP
)) ||
2150 ((features
& NETIF_F_V6_CSUM
) &&
2151 protocol
== htons(ETH_P_IPV6
)) ||
2152 ((features
& NETIF_F_FCOE_CRC
) &&
2153 protocol
== htons(ETH_P_FCOE
)));
2156 static netdev_features_t
harmonize_features(struct sk_buff
*skb
,
2157 __be16 protocol
, netdev_features_t features
)
2159 if (!can_checksum_protocol(features
, protocol
)) {
2160 features
&= ~NETIF_F_ALL_CSUM
;
2161 features
&= ~NETIF_F_SG
;
2162 } else if (illegal_highdma(skb
->dev
, skb
)) {
2163 features
&= ~NETIF_F_SG
;
2169 netdev_features_t
netif_skb_features(struct sk_buff
*skb
)
2171 __be16 protocol
= skb
->protocol
;
2172 netdev_features_t features
= skb
->dev
->features
;
2174 if (skb_shinfo(skb
)->gso_segs
> skb
->dev
->gso_max_segs
)
2175 features
&= ~NETIF_F_GSO_MASK
;
2177 if (protocol
== htons(ETH_P_8021Q
)) {
2178 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
2179 protocol
= veh
->h_vlan_encapsulated_proto
;
2180 } else if (!vlan_tx_tag_present(skb
)) {
2181 return harmonize_features(skb
, protocol
, features
);
2184 features
&= (skb
->dev
->vlan_features
| NETIF_F_HW_VLAN_TX
);
2186 if (protocol
!= htons(ETH_P_8021Q
)) {
2187 return harmonize_features(skb
, protocol
, features
);
2189 features
&= NETIF_F_SG
| NETIF_F_HIGHDMA
| NETIF_F_FRAGLIST
|
2190 NETIF_F_GEN_CSUM
| NETIF_F_HW_VLAN_TX
;
2191 return harmonize_features(skb
, protocol
, features
);
2194 EXPORT_SYMBOL(netif_skb_features
);
2197 * Returns true if either:
2198 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2199 * 2. skb is fragmented and the device does not support SG.
2201 static inline int skb_needs_linearize(struct sk_buff
*skb
,
2204 return skb_is_nonlinear(skb
) &&
2205 ((skb_has_frag_list(skb
) &&
2206 !(features
& NETIF_F_FRAGLIST
)) ||
2207 (skb_shinfo(skb
)->nr_frags
&&
2208 !(features
& NETIF_F_SG
)));
2211 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
2212 struct netdev_queue
*txq
)
2214 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2215 int rc
= NETDEV_TX_OK
;
2216 unsigned int skb_len
;
2218 if (likely(!skb
->next
)) {
2219 netdev_features_t features
;
2222 * If device doesn't need skb->dst, release it right now while
2223 * its hot in this cpu cache
2225 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2228 features
= netif_skb_features(skb
);
2230 if (vlan_tx_tag_present(skb
) &&
2231 !(features
& NETIF_F_HW_VLAN_TX
)) {
2232 skb
= __vlan_put_tag(skb
, vlan_tx_tag_get(skb
));
2239 if (netif_needs_gso(skb
, features
)) {
2240 if (unlikely(dev_gso_segment(skb
, features
)))
2245 if (skb_needs_linearize(skb
, features
) &&
2246 __skb_linearize(skb
))
2249 /* If packet is not checksummed and device does not
2250 * support checksumming for this protocol, complete
2251 * checksumming here.
2253 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2254 skb_set_transport_header(skb
,
2255 skb_checksum_start_offset(skb
));
2256 if (!(features
& NETIF_F_ALL_CSUM
) &&
2257 skb_checksum_help(skb
))
2262 if (!list_empty(&ptype_all
))
2263 dev_queue_xmit_nit(skb
, dev
);
2266 rc
= ops
->ndo_start_xmit(skb
, dev
);
2267 trace_net_dev_xmit(skb
, rc
, dev
, skb_len
);
2268 if (rc
== NETDEV_TX_OK
)
2269 txq_trans_update(txq
);
2275 struct sk_buff
*nskb
= skb
->next
;
2277 skb
->next
= nskb
->next
;
2281 * If device doesn't need nskb->dst, release it right now while
2282 * its hot in this cpu cache
2284 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2287 if (!list_empty(&ptype_all
))
2288 dev_queue_xmit_nit(nskb
, dev
);
2290 skb_len
= nskb
->len
;
2291 rc
= ops
->ndo_start_xmit(nskb
, dev
);
2292 trace_net_dev_xmit(nskb
, rc
, dev
, skb_len
);
2293 if (unlikely(rc
!= NETDEV_TX_OK
)) {
2294 if (rc
& ~NETDEV_TX_MASK
)
2295 goto out_kfree_gso_skb
;
2296 nskb
->next
= skb
->next
;
2300 txq_trans_update(txq
);
2301 if (unlikely(netif_xmit_stopped(txq
) && skb
->next
))
2302 return NETDEV_TX_BUSY
;
2303 } while (skb
->next
);
2306 if (likely(skb
->next
== NULL
))
2307 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
2314 static u32 hashrnd __read_mostly
;
2317 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2318 * to be used as a distribution range.
2320 u16
__skb_tx_hash(const struct net_device
*dev
, const struct sk_buff
*skb
,
2321 unsigned int num_tx_queues
)
2325 u16 qcount
= num_tx_queues
;
2327 if (skb_rx_queue_recorded(skb
)) {
2328 hash
= skb_get_rx_queue(skb
);
2329 while (unlikely(hash
>= num_tx_queues
))
2330 hash
-= num_tx_queues
;
2335 u8 tc
= netdev_get_prio_tc_map(dev
, skb
->priority
);
2336 qoffset
= dev
->tc_to_txq
[tc
].offset
;
2337 qcount
= dev
->tc_to_txq
[tc
].count
;
2340 if (skb
->sk
&& skb
->sk
->sk_hash
)
2341 hash
= skb
->sk
->sk_hash
;
2343 hash
= (__force u16
) skb
->protocol
;
2344 hash
= jhash_1word(hash
, hashrnd
);
2346 return (u16
) (((u64
) hash
* qcount
) >> 32) + qoffset
;
2348 EXPORT_SYMBOL(__skb_tx_hash
);
2350 static inline u16
dev_cap_txqueue(struct net_device
*dev
, u16 queue_index
)
2352 if (unlikely(queue_index
>= dev
->real_num_tx_queues
)) {
2353 net_warn_ratelimited("%s selects TX queue %d, but real number of TX queues is %d\n",
2354 dev
->name
, queue_index
,
2355 dev
->real_num_tx_queues
);
2361 static inline int get_xps_queue(struct net_device
*dev
, struct sk_buff
*skb
)
2364 struct xps_dev_maps
*dev_maps
;
2365 struct xps_map
*map
;
2366 int queue_index
= -1;
2369 dev_maps
= rcu_dereference(dev
->xps_maps
);
2371 map
= rcu_dereference(
2372 dev_maps
->cpu_map
[raw_smp_processor_id()]);
2375 queue_index
= map
->queues
[0];
2378 if (skb
->sk
&& skb
->sk
->sk_hash
)
2379 hash
= skb
->sk
->sk_hash
;
2381 hash
= (__force u16
) skb
->protocol
^
2383 hash
= jhash_1word(hash
, hashrnd
);
2384 queue_index
= map
->queues
[
2385 ((u64
)hash
* map
->len
) >> 32];
2387 if (unlikely(queue_index
>= dev
->real_num_tx_queues
))
2399 static struct netdev_queue
*dev_pick_tx(struct net_device
*dev
,
2400 struct sk_buff
*skb
)
2403 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2405 if (dev
->real_num_tx_queues
== 1)
2407 else if (ops
->ndo_select_queue
) {
2408 queue_index
= ops
->ndo_select_queue(dev
, skb
);
2409 queue_index
= dev_cap_txqueue(dev
, queue_index
);
2411 struct sock
*sk
= skb
->sk
;
2412 queue_index
= sk_tx_queue_get(sk
);
2414 if (queue_index
< 0 || skb
->ooo_okay
||
2415 queue_index
>= dev
->real_num_tx_queues
) {
2416 int old_index
= queue_index
;
2418 queue_index
= get_xps_queue(dev
, skb
);
2419 if (queue_index
< 0)
2420 queue_index
= skb_tx_hash(dev
, skb
);
2422 if (queue_index
!= old_index
&& sk
) {
2423 struct dst_entry
*dst
=
2424 rcu_dereference_check(sk
->sk_dst_cache
, 1);
2426 if (dst
&& skb_dst(skb
) == dst
)
2427 sk_tx_queue_set(sk
, queue_index
);
2432 skb_set_queue_mapping(skb
, queue_index
);
2433 return netdev_get_tx_queue(dev
, queue_index
);
2436 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2437 struct net_device
*dev
,
2438 struct netdev_queue
*txq
)
2440 spinlock_t
*root_lock
= qdisc_lock(q
);
2444 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
2445 qdisc_calculate_pkt_len(skb
, q
);
2447 * Heuristic to force contended enqueues to serialize on a
2448 * separate lock before trying to get qdisc main lock.
2449 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2450 * and dequeue packets faster.
2452 contended
= qdisc_is_running(q
);
2453 if (unlikely(contended
))
2454 spin_lock(&q
->busylock
);
2456 spin_lock(root_lock
);
2457 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2460 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2461 qdisc_run_begin(q
)) {
2463 * This is a work-conserving queue; there are no old skbs
2464 * waiting to be sent out; and the qdisc is not running -
2465 * xmit the skb directly.
2467 if (!(dev
->priv_flags
& IFF_XMIT_DST_RELEASE
))
2470 qdisc_bstats_update(q
, skb
);
2472 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
)) {
2473 if (unlikely(contended
)) {
2474 spin_unlock(&q
->busylock
);
2481 rc
= NET_XMIT_SUCCESS
;
2484 rc
= q
->enqueue(skb
, q
) & NET_XMIT_MASK
;
2485 if (qdisc_run_begin(q
)) {
2486 if (unlikely(contended
)) {
2487 spin_unlock(&q
->busylock
);
2493 spin_unlock(root_lock
);
2494 if (unlikely(contended
))
2495 spin_unlock(&q
->busylock
);
2499 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2500 static void skb_update_prio(struct sk_buff
*skb
)
2502 struct netprio_map
*map
= rcu_dereference_bh(skb
->dev
->priomap
);
2504 if (!skb
->priority
&& skb
->sk
&& map
) {
2505 unsigned int prioidx
= skb
->sk
->sk_cgrp_prioidx
;
2507 if (prioidx
< map
->priomap_len
)
2508 skb
->priority
= map
->priomap
[prioidx
];
2512 #define skb_update_prio(skb)
2515 static DEFINE_PER_CPU(int, xmit_recursion
);
2516 #define RECURSION_LIMIT 10
2519 * dev_loopback_xmit - loop back @skb
2520 * @skb: buffer to transmit
2522 int dev_loopback_xmit(struct sk_buff
*skb
)
2524 skb_reset_mac_header(skb
);
2525 __skb_pull(skb
, skb_network_offset(skb
));
2526 skb
->pkt_type
= PACKET_LOOPBACK
;
2527 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
2528 WARN_ON(!skb_dst(skb
));
2533 EXPORT_SYMBOL(dev_loopback_xmit
);
2536 * dev_queue_xmit - transmit a buffer
2537 * @skb: buffer to transmit
2539 * Queue a buffer for transmission to a network device. The caller must
2540 * have set the device and priority and built the buffer before calling
2541 * this function. The function can be called from an interrupt.
2543 * A negative errno code is returned on a failure. A success does not
2544 * guarantee the frame will be transmitted as it may be dropped due
2545 * to congestion or traffic shaping.
2547 * -----------------------------------------------------------------------------------
2548 * I notice this method can also return errors from the queue disciplines,
2549 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2552 * Regardless of the return value, the skb is consumed, so it is currently
2553 * difficult to retry a send to this method. (You can bump the ref count
2554 * before sending to hold a reference for retry if you are careful.)
2556 * When calling this method, interrupts MUST be enabled. This is because
2557 * the BH enable code must have IRQs enabled so that it will not deadlock.
2560 int dev_queue_xmit(struct sk_buff
*skb
)
2562 struct net_device
*dev
= skb
->dev
;
2563 struct netdev_queue
*txq
;
2567 /* Disable soft irqs for various locks below. Also
2568 * stops preemption for RCU.
2572 skb_update_prio(skb
);
2574 txq
= dev_pick_tx(dev
, skb
);
2575 q
= rcu_dereference_bh(txq
->qdisc
);
2577 #ifdef CONFIG_NET_CLS_ACT
2578 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2580 trace_net_dev_queue(skb
);
2582 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2586 /* The device has no queue. Common case for software devices:
2587 loopback, all the sorts of tunnels...
2589 Really, it is unlikely that netif_tx_lock protection is necessary
2590 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2592 However, it is possible, that they rely on protection
2595 Check this and shot the lock. It is not prone from deadlocks.
2596 Either shot noqueue qdisc, it is even simpler 8)
2598 if (dev
->flags
& IFF_UP
) {
2599 int cpu
= smp_processor_id(); /* ok because BHs are off */
2601 if (txq
->xmit_lock_owner
!= cpu
) {
2603 if (__this_cpu_read(xmit_recursion
) > RECURSION_LIMIT
)
2604 goto recursion_alert
;
2606 HARD_TX_LOCK(dev
, txq
, cpu
);
2608 if (!netif_xmit_stopped(txq
)) {
2609 __this_cpu_inc(xmit_recursion
);
2610 rc
= dev_hard_start_xmit(skb
, dev
, txq
);
2611 __this_cpu_dec(xmit_recursion
);
2612 if (dev_xmit_complete(rc
)) {
2613 HARD_TX_UNLOCK(dev
, txq
);
2617 HARD_TX_UNLOCK(dev
, txq
);
2618 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2621 /* Recursion is detected! It is possible,
2625 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2631 rcu_read_unlock_bh();
2636 rcu_read_unlock_bh();
2639 EXPORT_SYMBOL(dev_queue_xmit
);
2642 /*=======================================================================
2644 =======================================================================*/
2646 int netdev_max_backlog __read_mostly
= 1000;
2647 int netdev_tstamp_prequeue __read_mostly
= 1;
2648 int netdev_budget __read_mostly
= 300;
2649 int weight_p __read_mostly
= 64; /* old backlog weight */
2651 /* Called with irq disabled */
2652 static inline void ____napi_schedule(struct softnet_data
*sd
,
2653 struct napi_struct
*napi
)
2655 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
2656 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2660 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2661 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2662 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2663 * if hash is a canonical 4-tuple hash over transport ports.
2665 void __skb_get_rxhash(struct sk_buff
*skb
)
2667 struct flow_keys keys
;
2670 if (!skb_flow_dissect(skb
, &keys
))
2676 /* get a consistent hash (same value on both flow directions) */
2677 if (((__force u32
)keys
.dst
< (__force u32
)keys
.src
) ||
2678 (((__force u32
)keys
.dst
== (__force u32
)keys
.src
) &&
2679 ((__force u16
)keys
.port16
[1] < (__force u16
)keys
.port16
[0]))) {
2680 swap(keys
.dst
, keys
.src
);
2681 swap(keys
.port16
[0], keys
.port16
[1]);
2684 hash
= jhash_3words((__force u32
)keys
.dst
,
2685 (__force u32
)keys
.src
,
2686 (__force u32
)keys
.ports
, hashrnd
);
2692 EXPORT_SYMBOL(__skb_get_rxhash
);
2696 /* One global table that all flow-based protocols share. */
2697 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
2698 EXPORT_SYMBOL(rps_sock_flow_table
);
2700 struct static_key rps_needed __read_mostly
;
2702 static struct rps_dev_flow
*
2703 set_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2704 struct rps_dev_flow
*rflow
, u16 next_cpu
)
2706 if (next_cpu
!= RPS_NO_CPU
) {
2707 #ifdef CONFIG_RFS_ACCEL
2708 struct netdev_rx_queue
*rxqueue
;
2709 struct rps_dev_flow_table
*flow_table
;
2710 struct rps_dev_flow
*old_rflow
;
2715 /* Should we steer this flow to a different hardware queue? */
2716 if (!skb_rx_queue_recorded(skb
) || !dev
->rx_cpu_rmap
||
2717 !(dev
->features
& NETIF_F_NTUPLE
))
2719 rxq_index
= cpu_rmap_lookup_index(dev
->rx_cpu_rmap
, next_cpu
);
2720 if (rxq_index
== skb_get_rx_queue(skb
))
2723 rxqueue
= dev
->_rx
+ rxq_index
;
2724 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2727 flow_id
= skb
->rxhash
& flow_table
->mask
;
2728 rc
= dev
->netdev_ops
->ndo_rx_flow_steer(dev
, skb
,
2729 rxq_index
, flow_id
);
2733 rflow
= &flow_table
->flows
[flow_id
];
2735 if (old_rflow
->filter
== rflow
->filter
)
2736 old_rflow
->filter
= RPS_NO_FILTER
;
2740 per_cpu(softnet_data
, next_cpu
).input_queue_head
;
2743 rflow
->cpu
= next_cpu
;
2748 * get_rps_cpu is called from netif_receive_skb and returns the target
2749 * CPU from the RPS map of the receiving queue for a given skb.
2750 * rcu_read_lock must be held on entry.
2752 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2753 struct rps_dev_flow
**rflowp
)
2755 struct netdev_rx_queue
*rxqueue
;
2756 struct rps_map
*map
;
2757 struct rps_dev_flow_table
*flow_table
;
2758 struct rps_sock_flow_table
*sock_flow_table
;
2762 if (skb_rx_queue_recorded(skb
)) {
2763 u16 index
= skb_get_rx_queue(skb
);
2764 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
2765 WARN_ONCE(dev
->real_num_rx_queues
> 1,
2766 "%s received packet on queue %u, but number "
2767 "of RX queues is %u\n",
2768 dev
->name
, index
, dev
->real_num_rx_queues
);
2771 rxqueue
= dev
->_rx
+ index
;
2775 map
= rcu_dereference(rxqueue
->rps_map
);
2777 if (map
->len
== 1 &&
2778 !rcu_access_pointer(rxqueue
->rps_flow_table
)) {
2779 tcpu
= map
->cpus
[0];
2780 if (cpu_online(tcpu
))
2784 } else if (!rcu_access_pointer(rxqueue
->rps_flow_table
)) {
2788 skb_reset_network_header(skb
);
2789 if (!skb_get_rxhash(skb
))
2792 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2793 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
2794 if (flow_table
&& sock_flow_table
) {
2796 struct rps_dev_flow
*rflow
;
2798 rflow
= &flow_table
->flows
[skb
->rxhash
& flow_table
->mask
];
2801 next_cpu
= sock_flow_table
->ents
[skb
->rxhash
&
2802 sock_flow_table
->mask
];
2805 * If the desired CPU (where last recvmsg was done) is
2806 * different from current CPU (one in the rx-queue flow
2807 * table entry), switch if one of the following holds:
2808 * - Current CPU is unset (equal to RPS_NO_CPU).
2809 * - Current CPU is offline.
2810 * - The current CPU's queue tail has advanced beyond the
2811 * last packet that was enqueued using this table entry.
2812 * This guarantees that all previous packets for the flow
2813 * have been dequeued, thus preserving in order delivery.
2815 if (unlikely(tcpu
!= next_cpu
) &&
2816 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
2817 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
2818 rflow
->last_qtail
)) >= 0))
2819 rflow
= set_rps_cpu(dev
, skb
, rflow
, next_cpu
);
2821 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
2829 tcpu
= map
->cpus
[((u64
) skb
->rxhash
* map
->len
) >> 32];
2831 if (cpu_online(tcpu
)) {
2841 #ifdef CONFIG_RFS_ACCEL
2844 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2845 * @dev: Device on which the filter was set
2846 * @rxq_index: RX queue index
2847 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2848 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2850 * Drivers that implement ndo_rx_flow_steer() should periodically call
2851 * this function for each installed filter and remove the filters for
2852 * which it returns %true.
2854 bool rps_may_expire_flow(struct net_device
*dev
, u16 rxq_index
,
2855 u32 flow_id
, u16 filter_id
)
2857 struct netdev_rx_queue
*rxqueue
= dev
->_rx
+ rxq_index
;
2858 struct rps_dev_flow_table
*flow_table
;
2859 struct rps_dev_flow
*rflow
;
2864 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2865 if (flow_table
&& flow_id
<= flow_table
->mask
) {
2866 rflow
= &flow_table
->flows
[flow_id
];
2867 cpu
= ACCESS_ONCE(rflow
->cpu
);
2868 if (rflow
->filter
== filter_id
&& cpu
!= RPS_NO_CPU
&&
2869 ((int)(per_cpu(softnet_data
, cpu
).input_queue_head
-
2870 rflow
->last_qtail
) <
2871 (int)(10 * flow_table
->mask
)))
2877 EXPORT_SYMBOL(rps_may_expire_flow
);
2879 #endif /* CONFIG_RFS_ACCEL */
2881 /* Called from hardirq (IPI) context */
2882 static void rps_trigger_softirq(void *data
)
2884 struct softnet_data
*sd
= data
;
2886 ____napi_schedule(sd
, &sd
->backlog
);
2890 #endif /* CONFIG_RPS */
2893 * Check if this softnet_data structure is another cpu one
2894 * If yes, queue it to our IPI list and return 1
2897 static int rps_ipi_queued(struct softnet_data
*sd
)
2900 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
2903 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
2904 mysd
->rps_ipi_list
= sd
;
2906 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2909 #endif /* CONFIG_RPS */
2914 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2915 * queue (may be a remote CPU queue).
2917 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
2918 unsigned int *qtail
)
2920 struct softnet_data
*sd
;
2921 unsigned long flags
;
2923 sd
= &per_cpu(softnet_data
, cpu
);
2925 local_irq_save(flags
);
2928 if (skb_queue_len(&sd
->input_pkt_queue
) <= netdev_max_backlog
) {
2929 if (skb_queue_len(&sd
->input_pkt_queue
)) {
2931 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
2932 input_queue_tail_incr_save(sd
, qtail
);
2934 local_irq_restore(flags
);
2935 return NET_RX_SUCCESS
;
2938 /* Schedule NAPI for backlog device
2939 * We can use non atomic operation since we own the queue lock
2941 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
2942 if (!rps_ipi_queued(sd
))
2943 ____napi_schedule(sd
, &sd
->backlog
);
2951 local_irq_restore(flags
);
2953 atomic_long_inc(&skb
->dev
->rx_dropped
);
2959 * netif_rx - post buffer to the network code
2960 * @skb: buffer to post
2962 * This function receives a packet from a device driver and queues it for
2963 * the upper (protocol) levels to process. It always succeeds. The buffer
2964 * may be dropped during processing for congestion control or by the
2968 * NET_RX_SUCCESS (no congestion)
2969 * NET_RX_DROP (packet was dropped)
2973 int netif_rx(struct sk_buff
*skb
)
2977 /* if netpoll wants it, pretend we never saw it */
2978 if (netpoll_rx(skb
))
2981 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
2983 trace_netif_rx(skb
);
2985 if (static_key_false(&rps_needed
)) {
2986 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
2992 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
2994 cpu
= smp_processor_id();
2996 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3004 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
3009 EXPORT_SYMBOL(netif_rx
);
3011 int netif_rx_ni(struct sk_buff
*skb
)
3016 err
= netif_rx(skb
);
3017 if (local_softirq_pending())
3023 EXPORT_SYMBOL(netif_rx_ni
);
3025 static void net_tx_action(struct softirq_action
*h
)
3027 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3029 if (sd
->completion_queue
) {
3030 struct sk_buff
*clist
;
3032 local_irq_disable();
3033 clist
= sd
->completion_queue
;
3034 sd
->completion_queue
= NULL
;
3038 struct sk_buff
*skb
= clist
;
3039 clist
= clist
->next
;
3041 WARN_ON(atomic_read(&skb
->users
));
3042 trace_kfree_skb(skb
, net_tx_action
);
3047 if (sd
->output_queue
) {
3050 local_irq_disable();
3051 head
= sd
->output_queue
;
3052 sd
->output_queue
= NULL
;
3053 sd
->output_queue_tailp
= &sd
->output_queue
;
3057 struct Qdisc
*q
= head
;
3058 spinlock_t
*root_lock
;
3060 head
= head
->next_sched
;
3062 root_lock
= qdisc_lock(q
);
3063 if (spin_trylock(root_lock
)) {
3064 smp_mb__before_clear_bit();
3065 clear_bit(__QDISC_STATE_SCHED
,
3068 spin_unlock(root_lock
);
3070 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
3072 __netif_reschedule(q
);
3074 smp_mb__before_clear_bit();
3075 clear_bit(__QDISC_STATE_SCHED
,
3083 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3084 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3085 /* This hook is defined here for ATM LANE */
3086 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
3087 unsigned char *addr
) __read_mostly
;
3088 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
3091 #ifdef CONFIG_NET_CLS_ACT
3092 /* TODO: Maybe we should just force sch_ingress to be compiled in
3093 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3094 * a compare and 2 stores extra right now if we dont have it on
3095 * but have CONFIG_NET_CLS_ACT
3096 * NOTE: This doesn't stop any functionality; if you dont have
3097 * the ingress scheduler, you just can't add policies on ingress.
3100 static int ing_filter(struct sk_buff
*skb
, struct netdev_queue
*rxq
)
3102 struct net_device
*dev
= skb
->dev
;
3103 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
3104 int result
= TC_ACT_OK
;
3107 if (unlikely(MAX_RED_LOOP
< ttl
++)) {
3108 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3109 skb
->skb_iif
, dev
->ifindex
);
3113 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
3114 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
3117 if (q
!= &noop_qdisc
) {
3118 spin_lock(qdisc_lock(q
));
3119 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
3120 result
= qdisc_enqueue_root(skb
, q
);
3121 spin_unlock(qdisc_lock(q
));
3127 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
3128 struct packet_type
**pt_prev
,
3129 int *ret
, struct net_device
*orig_dev
)
3131 struct netdev_queue
*rxq
= rcu_dereference(skb
->dev
->ingress_queue
);
3133 if (!rxq
|| rxq
->qdisc
== &noop_qdisc
)
3137 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
3141 switch (ing_filter(skb
, rxq
)) {
3155 * netdev_rx_handler_register - register receive handler
3156 * @dev: device to register a handler for
3157 * @rx_handler: receive handler to register
3158 * @rx_handler_data: data pointer that is used by rx handler
3160 * Register a receive hander for a device. This handler will then be
3161 * called from __netif_receive_skb. A negative errno code is returned
3164 * The caller must hold the rtnl_mutex.
3166 * For a general description of rx_handler, see enum rx_handler_result.
3168 int netdev_rx_handler_register(struct net_device
*dev
,
3169 rx_handler_func_t
*rx_handler
,
3170 void *rx_handler_data
)
3174 if (dev
->rx_handler
)
3177 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
3178 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
3182 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
3185 * netdev_rx_handler_unregister - unregister receive handler
3186 * @dev: device to unregister a handler from
3188 * Unregister a receive hander from a device.
3190 * The caller must hold the rtnl_mutex.
3192 void netdev_rx_handler_unregister(struct net_device
*dev
)
3196 RCU_INIT_POINTER(dev
->rx_handler
, NULL
);
3197 RCU_INIT_POINTER(dev
->rx_handler_data
, NULL
);
3199 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
3202 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3203 * the special handling of PFMEMALLOC skbs.
3205 static bool skb_pfmemalloc_protocol(struct sk_buff
*skb
)
3207 switch (skb
->protocol
) {
3208 case __constant_htons(ETH_P_ARP
):
3209 case __constant_htons(ETH_P_IP
):
3210 case __constant_htons(ETH_P_IPV6
):
3211 case __constant_htons(ETH_P_8021Q
):
3218 static int __netif_receive_skb(struct sk_buff
*skb
)
3220 struct packet_type
*ptype
, *pt_prev
;
3221 rx_handler_func_t
*rx_handler
;
3222 struct net_device
*orig_dev
;
3223 struct net_device
*null_or_dev
;
3224 bool deliver_exact
= false;
3225 int ret
= NET_RX_DROP
;
3227 unsigned long pflags
= current
->flags
;
3229 net_timestamp_check(!netdev_tstamp_prequeue
, skb
);
3231 trace_netif_receive_skb(skb
);
3234 * PFMEMALLOC skbs are special, they should
3235 * - be delivered to SOCK_MEMALLOC sockets only
3236 * - stay away from userspace
3237 * - have bounded memory usage
3239 * Use PF_MEMALLOC as this saves us from propagating the allocation
3240 * context down to all allocation sites.
3242 if (sk_memalloc_socks() && skb_pfmemalloc(skb
))
3243 current
->flags
|= PF_MEMALLOC
;
3245 /* if we've gotten here through NAPI, check netpoll */
3246 if (netpoll_receive_skb(skb
))
3249 orig_dev
= skb
->dev
;
3251 skb_reset_network_header(skb
);
3252 skb_reset_transport_header(skb
);
3253 skb_reset_mac_len(skb
);
3260 skb
->skb_iif
= skb
->dev
->ifindex
;
3262 __this_cpu_inc(softnet_data
.processed
);
3264 if (skb
->protocol
== cpu_to_be16(ETH_P_8021Q
)) {
3265 skb
= vlan_untag(skb
);
3270 #ifdef CONFIG_NET_CLS_ACT
3271 if (skb
->tc_verd
& TC_NCLS
) {
3272 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
3277 if (sk_memalloc_socks() && skb_pfmemalloc(skb
))
3280 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
3281 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
3283 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3289 #ifdef CONFIG_NET_CLS_ACT
3290 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
3296 if (sk_memalloc_socks() && skb_pfmemalloc(skb
)
3297 && !skb_pfmemalloc_protocol(skb
))
3300 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
3301 if (vlan_tx_tag_present(skb
)) {
3303 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3306 if (vlan_do_receive(&skb
, !rx_handler
))
3308 else if (unlikely(!skb
))
3314 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3317 switch (rx_handler(&skb
)) {
3318 case RX_HANDLER_CONSUMED
:
3320 case RX_HANDLER_ANOTHER
:
3322 case RX_HANDLER_EXACT
:
3323 deliver_exact
= true;
3324 case RX_HANDLER_PASS
:
3331 /* deliver only exact match when indicated */
3332 null_or_dev
= deliver_exact
? skb
->dev
: NULL
;
3334 type
= skb
->protocol
;
3335 list_for_each_entry_rcu(ptype
,
3336 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
3337 if (ptype
->type
== type
&&
3338 (ptype
->dev
== null_or_dev
|| ptype
->dev
== skb
->dev
||
3339 ptype
->dev
== orig_dev
)) {
3341 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3347 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
3350 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
3353 atomic_long_inc(&skb
->dev
->rx_dropped
);
3355 /* Jamal, now you will not able to escape explaining
3356 * me how you were going to use this. :-)
3364 tsk_restore_flags(current
, pflags
, PF_MEMALLOC
);
3369 * netif_receive_skb - process receive buffer from network
3370 * @skb: buffer to process
3372 * netif_receive_skb() is the main receive data processing function.
3373 * It always succeeds. The buffer may be dropped during processing
3374 * for congestion control or by the protocol layers.
3376 * This function may only be called from softirq context and interrupts
3377 * should be enabled.
3379 * Return values (usually ignored):
3380 * NET_RX_SUCCESS: no congestion
3381 * NET_RX_DROP: packet was dropped
3383 int netif_receive_skb(struct sk_buff
*skb
)
3385 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3387 if (skb_defer_rx_timestamp(skb
))
3388 return NET_RX_SUCCESS
;
3391 if (static_key_false(&rps_needed
)) {
3392 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3397 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3400 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3407 return __netif_receive_skb(skb
);
3409 EXPORT_SYMBOL(netif_receive_skb
);
3411 /* Network device is going away, flush any packets still pending
3412 * Called with irqs disabled.
3414 static void flush_backlog(void *arg
)
3416 struct net_device
*dev
= arg
;
3417 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3418 struct sk_buff
*skb
, *tmp
;
3421 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3422 if (skb
->dev
== dev
) {
3423 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3425 input_queue_head_incr(sd
);
3430 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3431 if (skb
->dev
== dev
) {
3432 __skb_unlink(skb
, &sd
->process_queue
);
3434 input_queue_head_incr(sd
);
3439 static int napi_gro_complete(struct sk_buff
*skb
)
3441 struct packet_type
*ptype
;
3442 __be16 type
= skb
->protocol
;
3443 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3446 if (NAPI_GRO_CB(skb
)->count
== 1) {
3447 skb_shinfo(skb
)->gso_size
= 0;
3452 list_for_each_entry_rcu(ptype
, head
, list
) {
3453 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_complete
)
3456 err
= ptype
->gro_complete(skb
);
3462 WARN_ON(&ptype
->list
== head
);
3464 return NET_RX_SUCCESS
;
3468 return netif_receive_skb(skb
);
3471 inline void napi_gro_flush(struct napi_struct
*napi
)
3473 struct sk_buff
*skb
, *next
;
3475 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3478 napi_gro_complete(skb
);
3481 napi
->gro_count
= 0;
3482 napi
->gro_list
= NULL
;
3484 EXPORT_SYMBOL(napi_gro_flush
);
3486 enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3488 struct sk_buff
**pp
= NULL
;
3489 struct packet_type
*ptype
;
3490 __be16 type
= skb
->protocol
;
3491 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3494 enum gro_result ret
;
3496 if (!(skb
->dev
->features
& NETIF_F_GRO
) || netpoll_rx_on(skb
))
3499 if (skb_is_gso(skb
) || skb_has_frag_list(skb
))
3503 list_for_each_entry_rcu(ptype
, head
, list
) {
3504 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_receive
)
3507 skb_set_network_header(skb
, skb_gro_offset(skb
));
3508 mac_len
= skb
->network_header
- skb
->mac_header
;
3509 skb
->mac_len
= mac_len
;
3510 NAPI_GRO_CB(skb
)->same_flow
= 0;
3511 NAPI_GRO_CB(skb
)->flush
= 0;
3512 NAPI_GRO_CB(skb
)->free
= 0;
3514 pp
= ptype
->gro_receive(&napi
->gro_list
, skb
);
3519 if (&ptype
->list
== head
)
3522 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3523 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3526 struct sk_buff
*nskb
= *pp
;
3530 napi_gro_complete(nskb
);
3537 if (NAPI_GRO_CB(skb
)->flush
|| napi
->gro_count
>= MAX_GRO_SKBS
)
3541 NAPI_GRO_CB(skb
)->count
= 1;
3542 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
3543 skb
->next
= napi
->gro_list
;
3544 napi
->gro_list
= skb
;
3548 if (skb_headlen(skb
) < skb_gro_offset(skb
)) {
3549 int grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
3551 BUG_ON(skb
->end
- skb
->tail
< grow
);
3553 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3556 skb
->data_len
-= grow
;
3558 skb_shinfo(skb
)->frags
[0].page_offset
+= grow
;
3559 skb_frag_size_sub(&skb_shinfo(skb
)->frags
[0], grow
);
3561 if (unlikely(!skb_frag_size(&skb_shinfo(skb
)->frags
[0]))) {
3562 skb_frag_unref(skb
, 0);
3563 memmove(skb_shinfo(skb
)->frags
,
3564 skb_shinfo(skb
)->frags
+ 1,
3565 --skb_shinfo(skb
)->nr_frags
* sizeof(skb_frag_t
));
3576 EXPORT_SYMBOL(dev_gro_receive
);
3578 static inline gro_result_t
3579 __napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3582 unsigned int maclen
= skb
->dev
->hard_header_len
;
3584 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3585 unsigned long diffs
;
3587 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
3588 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
3589 if (maclen
== ETH_HLEN
)
3590 diffs
|= compare_ether_header(skb_mac_header(p
),
3591 skb_gro_mac_header(skb
));
3593 diffs
= memcmp(skb_mac_header(p
),
3594 skb_gro_mac_header(skb
),
3596 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
3597 NAPI_GRO_CB(p
)->flush
= 0;
3600 return dev_gro_receive(napi
, skb
);
3603 gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
3607 if (netif_receive_skb(skb
))
3615 case GRO_MERGED_FREE
:
3616 if (NAPI_GRO_CB(skb
)->free
== NAPI_GRO_FREE_STOLEN_HEAD
)
3617 kmem_cache_free(skbuff_head_cache
, skb
);
3629 EXPORT_SYMBOL(napi_skb_finish
);
3631 void skb_gro_reset_offset(struct sk_buff
*skb
)
3633 NAPI_GRO_CB(skb
)->data_offset
= 0;
3634 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3635 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3637 if (skb
->mac_header
== skb
->tail
&&
3638 !PageHighMem(skb_frag_page(&skb_shinfo(skb
)->frags
[0]))) {
3639 NAPI_GRO_CB(skb
)->frag0
=
3640 skb_frag_address(&skb_shinfo(skb
)->frags
[0]);
3641 NAPI_GRO_CB(skb
)->frag0_len
= skb_frag_size(&skb_shinfo(skb
)->frags
[0]);
3644 EXPORT_SYMBOL(skb_gro_reset_offset
);
3646 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3648 skb_gro_reset_offset(skb
);
3650 return napi_skb_finish(__napi_gro_receive(napi
, skb
), skb
);
3652 EXPORT_SYMBOL(napi_gro_receive
);
3654 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
3656 __skb_pull(skb
, skb_headlen(skb
));
3657 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3658 skb_reserve(skb
, NET_SKB_PAD
+ NET_IP_ALIGN
- skb_headroom(skb
));
3660 skb
->dev
= napi
->dev
;
3666 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
3668 struct sk_buff
*skb
= napi
->skb
;
3671 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
3677 EXPORT_SYMBOL(napi_get_frags
);
3679 gro_result_t
napi_frags_finish(struct napi_struct
*napi
, struct sk_buff
*skb
,
3685 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
3687 if (ret
== GRO_HELD
)
3688 skb_gro_pull(skb
, -ETH_HLEN
);
3689 else if (netif_receive_skb(skb
))
3694 case GRO_MERGED_FREE
:
3695 napi_reuse_skb(napi
, skb
);
3704 EXPORT_SYMBOL(napi_frags_finish
);
3706 static struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
3708 struct sk_buff
*skb
= napi
->skb
;
3715 skb_reset_mac_header(skb
);
3716 skb_gro_reset_offset(skb
);
3718 off
= skb_gro_offset(skb
);
3719 hlen
= off
+ sizeof(*eth
);
3720 eth
= skb_gro_header_fast(skb
, off
);
3721 if (skb_gro_header_hard(skb
, hlen
)) {
3722 eth
= skb_gro_header_slow(skb
, hlen
, off
);
3723 if (unlikely(!eth
)) {
3724 napi_reuse_skb(napi
, skb
);
3730 skb_gro_pull(skb
, sizeof(*eth
));
3733 * This works because the only protocols we care about don't require
3734 * special handling. We'll fix it up properly at the end.
3736 skb
->protocol
= eth
->h_proto
;
3742 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
3744 struct sk_buff
*skb
= napi_frags_skb(napi
);
3749 return napi_frags_finish(napi
, skb
, __napi_gro_receive(napi
, skb
));
3751 EXPORT_SYMBOL(napi_gro_frags
);
3754 * net_rps_action sends any pending IPI's for rps.
3755 * Note: called with local irq disabled, but exits with local irq enabled.
3757 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
3760 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
3763 sd
->rps_ipi_list
= NULL
;
3767 /* Send pending IPI's to kick RPS processing on remote cpus. */
3769 struct softnet_data
*next
= remsd
->rps_ipi_next
;
3771 if (cpu_online(remsd
->cpu
))
3772 __smp_call_function_single(remsd
->cpu
,
3781 static int process_backlog(struct napi_struct
*napi
, int quota
)
3784 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
3787 /* Check if we have pending ipi, its better to send them now,
3788 * not waiting net_rx_action() end.
3790 if (sd
->rps_ipi_list
) {
3791 local_irq_disable();
3792 net_rps_action_and_irq_enable(sd
);
3795 napi
->weight
= weight_p
;
3796 local_irq_disable();
3797 while (work
< quota
) {
3798 struct sk_buff
*skb
;
3801 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
3803 __netif_receive_skb(skb
);
3804 local_irq_disable();
3805 input_queue_head_incr(sd
);
3806 if (++work
>= quota
) {
3813 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3815 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
3816 &sd
->process_queue
);
3818 if (qlen
< quota
- work
) {
3820 * Inline a custom version of __napi_complete().
3821 * only current cpu owns and manipulates this napi,
3822 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3823 * we can use a plain write instead of clear_bit(),
3824 * and we dont need an smp_mb() memory barrier.
3826 list_del(&napi
->poll_list
);
3829 quota
= work
+ qlen
;
3839 * __napi_schedule - schedule for receive
3840 * @n: entry to schedule
3842 * The entry's receive function will be scheduled to run
3844 void __napi_schedule(struct napi_struct
*n
)
3846 unsigned long flags
;
3848 local_irq_save(flags
);
3849 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
3850 local_irq_restore(flags
);
3852 EXPORT_SYMBOL(__napi_schedule
);
3854 void __napi_complete(struct napi_struct
*n
)
3856 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
3857 BUG_ON(n
->gro_list
);
3859 list_del(&n
->poll_list
);
3860 smp_mb__before_clear_bit();
3861 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
3863 EXPORT_SYMBOL(__napi_complete
);
3865 void napi_complete(struct napi_struct
*n
)
3867 unsigned long flags
;
3870 * don't let napi dequeue from the cpu poll list
3871 * just in case its running on a different cpu
3873 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
3877 local_irq_save(flags
);
3879 local_irq_restore(flags
);
3881 EXPORT_SYMBOL(napi_complete
);
3883 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
3884 int (*poll
)(struct napi_struct
*, int), int weight
)
3886 INIT_LIST_HEAD(&napi
->poll_list
);
3887 napi
->gro_count
= 0;
3888 napi
->gro_list
= NULL
;
3891 napi
->weight
= weight
;
3892 list_add(&napi
->dev_list
, &dev
->napi_list
);
3894 #ifdef CONFIG_NETPOLL
3895 spin_lock_init(&napi
->poll_lock
);
3896 napi
->poll_owner
= -1;
3898 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
3900 EXPORT_SYMBOL(netif_napi_add
);
3902 void netif_napi_del(struct napi_struct
*napi
)
3904 struct sk_buff
*skb
, *next
;
3906 list_del_init(&napi
->dev_list
);
3907 napi_free_frags(napi
);
3909 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3915 napi
->gro_list
= NULL
;
3916 napi
->gro_count
= 0;
3918 EXPORT_SYMBOL(netif_napi_del
);
3920 static void net_rx_action(struct softirq_action
*h
)
3922 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3923 unsigned long time_limit
= jiffies
+ 2;
3924 int budget
= netdev_budget
;
3927 local_irq_disable();
3929 while (!list_empty(&sd
->poll_list
)) {
3930 struct napi_struct
*n
;
3933 /* If softirq window is exhuasted then punt.
3934 * Allow this to run for 2 jiffies since which will allow
3935 * an average latency of 1.5/HZ.
3937 if (unlikely(budget
<= 0 || time_after(jiffies
, time_limit
)))
3942 /* Even though interrupts have been re-enabled, this
3943 * access is safe because interrupts can only add new
3944 * entries to the tail of this list, and only ->poll()
3945 * calls can remove this head entry from the list.
3947 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
3949 have
= netpoll_poll_lock(n
);
3953 /* This NAPI_STATE_SCHED test is for avoiding a race
3954 * with netpoll's poll_napi(). Only the entity which
3955 * obtains the lock and sees NAPI_STATE_SCHED set will
3956 * actually make the ->poll() call. Therefore we avoid
3957 * accidentally calling ->poll() when NAPI is not scheduled.
3960 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
3961 work
= n
->poll(n
, weight
);
3965 WARN_ON_ONCE(work
> weight
);
3969 local_irq_disable();
3971 /* Drivers must not modify the NAPI state if they
3972 * consume the entire weight. In such cases this code
3973 * still "owns" the NAPI instance and therefore can
3974 * move the instance around on the list at-will.
3976 if (unlikely(work
== weight
)) {
3977 if (unlikely(napi_disable_pending(n
))) {
3980 local_irq_disable();
3982 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
3985 netpoll_poll_unlock(have
);
3988 net_rps_action_and_irq_enable(sd
);
3990 #ifdef CONFIG_NET_DMA
3992 * There may not be any more sk_buffs coming right now, so push
3993 * any pending DMA copies to hardware
3995 dma_issue_pending_all();
4002 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
4006 static gifconf_func_t
*gifconf_list
[NPROTO
];
4009 * register_gifconf - register a SIOCGIF handler
4010 * @family: Address family
4011 * @gifconf: Function handler
4013 * Register protocol dependent address dumping routines. The handler
4014 * that is passed must not be freed or reused until it has been replaced
4015 * by another handler.
4017 int register_gifconf(unsigned int family
, gifconf_func_t
*gifconf
)
4019 if (family
>= NPROTO
)
4021 gifconf_list
[family
] = gifconf
;
4024 EXPORT_SYMBOL(register_gifconf
);
4028 * Map an interface index to its name (SIOCGIFNAME)
4032 * We need this ioctl for efficient implementation of the
4033 * if_indextoname() function required by the IPv6 API. Without
4034 * it, we would have to search all the interfaces to find a
4038 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
4040 struct net_device
*dev
;
4044 * Fetch the caller's info block.
4047 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
4051 dev
= dev_get_by_index_rcu(net
, ifr
.ifr_ifindex
);
4057 strcpy(ifr
.ifr_name
, dev
->name
);
4060 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
4066 * Perform a SIOCGIFCONF call. This structure will change
4067 * size eventually, and there is nothing I can do about it.
4068 * Thus we will need a 'compatibility mode'.
4071 static int dev_ifconf(struct net
*net
, char __user
*arg
)
4074 struct net_device
*dev
;
4081 * Fetch the caller's info block.
4084 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
4091 * Loop over the interfaces, and write an info block for each.
4095 for_each_netdev(net
, dev
) {
4096 for (i
= 0; i
< NPROTO
; i
++) {
4097 if (gifconf_list
[i
]) {
4100 done
= gifconf_list
[i
](dev
, NULL
, 0);
4102 done
= gifconf_list
[i
](dev
, pos
+ total
,
4112 * All done. Write the updated control block back to the caller.
4114 ifc
.ifc_len
= total
;
4117 * Both BSD and Solaris return 0 here, so we do too.
4119 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
4122 #ifdef CONFIG_PROC_FS
4124 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4126 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4127 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4128 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4130 static inline struct net_device
*dev_from_same_bucket(struct seq_file
*seq
, loff_t
*pos
)
4132 struct net
*net
= seq_file_net(seq
);
4133 struct net_device
*dev
;
4134 struct hlist_node
*p
;
4135 struct hlist_head
*h
;
4136 unsigned int count
= 0, offset
= get_offset(*pos
);
4138 h
= &net
->dev_name_head
[get_bucket(*pos
)];
4139 hlist_for_each_entry_rcu(dev
, p
, h
, name_hlist
) {
4140 if (++count
== offset
)
4147 static inline struct net_device
*dev_from_bucket(struct seq_file
*seq
, loff_t
*pos
)
4149 struct net_device
*dev
;
4150 unsigned int bucket
;
4153 dev
= dev_from_same_bucket(seq
, pos
);
4157 bucket
= get_bucket(*pos
) + 1;
4158 *pos
= set_bucket_offset(bucket
, 1);
4159 } while (bucket
< NETDEV_HASHENTRIES
);
4165 * This is invoked by the /proc filesystem handler to display a device
4168 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4173 return SEQ_START_TOKEN
;
4175 if (get_bucket(*pos
) >= NETDEV_HASHENTRIES
)
4178 return dev_from_bucket(seq
, pos
);
4181 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4184 return dev_from_bucket(seq
, pos
);
4187 void dev_seq_stop(struct seq_file
*seq
, void *v
)
4193 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
4195 struct rtnl_link_stats64 temp
;
4196 const struct rtnl_link_stats64
*stats
= dev_get_stats(dev
, &temp
);
4198 seq_printf(seq
, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4199 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4200 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
4202 stats
->rx_dropped
+ stats
->rx_missed_errors
,
4203 stats
->rx_fifo_errors
,
4204 stats
->rx_length_errors
+ stats
->rx_over_errors
+
4205 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
4206 stats
->rx_compressed
, stats
->multicast
,
4207 stats
->tx_bytes
, stats
->tx_packets
,
4208 stats
->tx_errors
, stats
->tx_dropped
,
4209 stats
->tx_fifo_errors
, stats
->collisions
,
4210 stats
->tx_carrier_errors
+
4211 stats
->tx_aborted_errors
+
4212 stats
->tx_window_errors
+
4213 stats
->tx_heartbeat_errors
,
4214 stats
->tx_compressed
);
4218 * Called from the PROCfs module. This now uses the new arbitrary sized
4219 * /proc/net interface to create /proc/net/dev
4221 static int dev_seq_show(struct seq_file
*seq
, void *v
)
4223 if (v
== SEQ_START_TOKEN
)
4224 seq_puts(seq
, "Inter-| Receive "
4226 " face |bytes packets errs drop fifo frame "
4227 "compressed multicast|bytes packets errs "
4228 "drop fifo colls carrier compressed\n");
4230 dev_seq_printf_stats(seq
, v
);
4234 static struct softnet_data
*softnet_get_online(loff_t
*pos
)
4236 struct softnet_data
*sd
= NULL
;
4238 while (*pos
< nr_cpu_ids
)
4239 if (cpu_online(*pos
)) {
4240 sd
= &per_cpu(softnet_data
, *pos
);
4247 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4249 return softnet_get_online(pos
);
4252 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4255 return softnet_get_online(pos
);
4258 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
4262 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
4264 struct softnet_data
*sd
= v
;
4266 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4267 sd
->processed
, sd
->dropped
, sd
->time_squeeze
, 0,
4268 0, 0, 0, 0, /* was fastroute */
4269 sd
->cpu_collision
, sd
->received_rps
);
4273 static const struct seq_operations dev_seq_ops
= {
4274 .start
= dev_seq_start
,
4275 .next
= dev_seq_next
,
4276 .stop
= dev_seq_stop
,
4277 .show
= dev_seq_show
,
4280 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
4282 return seq_open_net(inode
, file
, &dev_seq_ops
,
4283 sizeof(struct seq_net_private
));
4286 static const struct file_operations dev_seq_fops
= {
4287 .owner
= THIS_MODULE
,
4288 .open
= dev_seq_open
,
4290 .llseek
= seq_lseek
,
4291 .release
= seq_release_net
,
4294 static const struct seq_operations softnet_seq_ops
= {
4295 .start
= softnet_seq_start
,
4296 .next
= softnet_seq_next
,
4297 .stop
= softnet_seq_stop
,
4298 .show
= softnet_seq_show
,
4301 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
4303 return seq_open(file
, &softnet_seq_ops
);
4306 static const struct file_operations softnet_seq_fops
= {
4307 .owner
= THIS_MODULE
,
4308 .open
= softnet_seq_open
,
4310 .llseek
= seq_lseek
,
4311 .release
= seq_release
,
4314 static void *ptype_get_idx(loff_t pos
)
4316 struct packet_type
*pt
= NULL
;
4320 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
4326 for (t
= 0; t
< PTYPE_HASH_SIZE
; t
++) {
4327 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
4336 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4340 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
4343 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4345 struct packet_type
*pt
;
4346 struct list_head
*nxt
;
4350 if (v
== SEQ_START_TOKEN
)
4351 return ptype_get_idx(0);
4354 nxt
= pt
->list
.next
;
4355 if (pt
->type
== htons(ETH_P_ALL
)) {
4356 if (nxt
!= &ptype_all
)
4359 nxt
= ptype_base
[0].next
;
4361 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
4363 while (nxt
== &ptype_base
[hash
]) {
4364 if (++hash
>= PTYPE_HASH_SIZE
)
4366 nxt
= ptype_base
[hash
].next
;
4369 return list_entry(nxt
, struct packet_type
, list
);
4372 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
4378 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
4380 struct packet_type
*pt
= v
;
4382 if (v
== SEQ_START_TOKEN
)
4383 seq_puts(seq
, "Type Device Function\n");
4384 else if (pt
->dev
== NULL
|| dev_net(pt
->dev
) == seq_file_net(seq
)) {
4385 if (pt
->type
== htons(ETH_P_ALL
))
4386 seq_puts(seq
, "ALL ");
4388 seq_printf(seq
, "%04x", ntohs(pt
->type
));
4390 seq_printf(seq
, " %-8s %pF\n",
4391 pt
->dev
? pt
->dev
->name
: "", pt
->func
);
4397 static const struct seq_operations ptype_seq_ops
= {
4398 .start
= ptype_seq_start
,
4399 .next
= ptype_seq_next
,
4400 .stop
= ptype_seq_stop
,
4401 .show
= ptype_seq_show
,
4404 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
4406 return seq_open_net(inode
, file
, &ptype_seq_ops
,
4407 sizeof(struct seq_net_private
));
4410 static const struct file_operations ptype_seq_fops
= {
4411 .owner
= THIS_MODULE
,
4412 .open
= ptype_seq_open
,
4414 .llseek
= seq_lseek
,
4415 .release
= seq_release_net
,
4419 static int __net_init
dev_proc_net_init(struct net
*net
)
4423 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
4425 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
4427 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
4430 if (wext_proc_init(net
))
4436 proc_net_remove(net
, "ptype");
4438 proc_net_remove(net
, "softnet_stat");
4440 proc_net_remove(net
, "dev");
4444 static void __net_exit
dev_proc_net_exit(struct net
*net
)
4446 wext_proc_exit(net
);
4448 proc_net_remove(net
, "ptype");
4449 proc_net_remove(net
, "softnet_stat");
4450 proc_net_remove(net
, "dev");
4453 static struct pernet_operations __net_initdata dev_proc_ops
= {
4454 .init
= dev_proc_net_init
,
4455 .exit
= dev_proc_net_exit
,
4458 static int __init
dev_proc_init(void)
4460 return register_pernet_subsys(&dev_proc_ops
);
4463 #define dev_proc_init() 0
4464 #endif /* CONFIG_PROC_FS */
4468 * netdev_set_master - set up master pointer
4469 * @slave: slave device
4470 * @master: new master device
4472 * Changes the master device of the slave. Pass %NULL to break the
4473 * bonding. The caller must hold the RTNL semaphore. On a failure
4474 * a negative errno code is returned. On success the reference counts
4475 * are adjusted and the function returns zero.
4477 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
4479 struct net_device
*old
= slave
->master
;
4489 slave
->master
= master
;
4495 EXPORT_SYMBOL(netdev_set_master
);
4498 * netdev_set_bond_master - set up bonding master/slave pair
4499 * @slave: slave device
4500 * @master: new master device
4502 * Changes the master device of the slave. Pass %NULL to break the
4503 * bonding. The caller must hold the RTNL semaphore. On a failure
4504 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4505 * to the routing socket and the function returns zero.
4507 int netdev_set_bond_master(struct net_device
*slave
, struct net_device
*master
)
4513 err
= netdev_set_master(slave
, master
);
4517 slave
->flags
|= IFF_SLAVE
;
4519 slave
->flags
&= ~IFF_SLAVE
;
4521 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
4524 EXPORT_SYMBOL(netdev_set_bond_master
);
4526 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
4528 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4530 if ((dev
->flags
& IFF_UP
) && ops
->ndo_change_rx_flags
)
4531 ops
->ndo_change_rx_flags(dev
, flags
);
4534 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
4536 unsigned int old_flags
= dev
->flags
;
4542 dev
->flags
|= IFF_PROMISC
;
4543 dev
->promiscuity
+= inc
;
4544 if (dev
->promiscuity
== 0) {
4547 * If inc causes overflow, untouch promisc and return error.
4550 dev
->flags
&= ~IFF_PROMISC
;
4552 dev
->promiscuity
-= inc
;
4553 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4558 if (dev
->flags
!= old_flags
) {
4559 pr_info("device %s %s promiscuous mode\n",
4561 dev
->flags
& IFF_PROMISC
? "entered" : "left");
4562 if (audit_enabled
) {
4563 current_uid_gid(&uid
, &gid
);
4564 audit_log(current
->audit_context
, GFP_ATOMIC
,
4565 AUDIT_ANOM_PROMISCUOUS
,
4566 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4567 dev
->name
, (dev
->flags
& IFF_PROMISC
),
4568 (old_flags
& IFF_PROMISC
),
4569 audit_get_loginuid(current
),
4570 from_kuid(&init_user_ns
, uid
),
4571 from_kgid(&init_user_ns
, gid
),
4572 audit_get_sessionid(current
));
4575 dev_change_rx_flags(dev
, IFF_PROMISC
);
4581 * dev_set_promiscuity - update promiscuity count on a device
4585 * Add or remove promiscuity from a device. While the count in the device
4586 * remains above zero the interface remains promiscuous. Once it hits zero
4587 * the device reverts back to normal filtering operation. A negative inc
4588 * value is used to drop promiscuity on the device.
4589 * Return 0 if successful or a negative errno code on error.
4591 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
4593 unsigned int old_flags
= dev
->flags
;
4596 err
= __dev_set_promiscuity(dev
, inc
);
4599 if (dev
->flags
!= old_flags
)
4600 dev_set_rx_mode(dev
);
4603 EXPORT_SYMBOL(dev_set_promiscuity
);
4606 * dev_set_allmulti - update allmulti count on a device
4610 * Add or remove reception of all multicast frames to a device. While the
4611 * count in the device remains above zero the interface remains listening
4612 * to all interfaces. Once it hits zero the device reverts back to normal
4613 * filtering operation. A negative @inc value is used to drop the counter
4614 * when releasing a resource needing all multicasts.
4615 * Return 0 if successful or a negative errno code on error.
4618 int dev_set_allmulti(struct net_device
*dev
, int inc
)
4620 unsigned int old_flags
= dev
->flags
;
4624 dev
->flags
|= IFF_ALLMULTI
;
4625 dev
->allmulti
+= inc
;
4626 if (dev
->allmulti
== 0) {
4629 * If inc causes overflow, untouch allmulti and return error.
4632 dev
->flags
&= ~IFF_ALLMULTI
;
4634 dev
->allmulti
-= inc
;
4635 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4640 if (dev
->flags
^ old_flags
) {
4641 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
4642 dev_set_rx_mode(dev
);
4646 EXPORT_SYMBOL(dev_set_allmulti
);
4649 * Upload unicast and multicast address lists to device and
4650 * configure RX filtering. When the device doesn't support unicast
4651 * filtering it is put in promiscuous mode while unicast addresses
4654 void __dev_set_rx_mode(struct net_device
*dev
)
4656 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4658 /* dev_open will call this function so the list will stay sane. */
4659 if (!(dev
->flags
&IFF_UP
))
4662 if (!netif_device_present(dev
))
4665 if (!(dev
->priv_flags
& IFF_UNICAST_FLT
)) {
4666 /* Unicast addresses changes may only happen under the rtnl,
4667 * therefore calling __dev_set_promiscuity here is safe.
4669 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
4670 __dev_set_promiscuity(dev
, 1);
4671 dev
->uc_promisc
= true;
4672 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
4673 __dev_set_promiscuity(dev
, -1);
4674 dev
->uc_promisc
= false;
4678 if (ops
->ndo_set_rx_mode
)
4679 ops
->ndo_set_rx_mode(dev
);
4682 void dev_set_rx_mode(struct net_device
*dev
)
4684 netif_addr_lock_bh(dev
);
4685 __dev_set_rx_mode(dev
);
4686 netif_addr_unlock_bh(dev
);
4690 * dev_get_flags - get flags reported to userspace
4693 * Get the combination of flag bits exported through APIs to userspace.
4695 unsigned int dev_get_flags(const struct net_device
*dev
)
4699 flags
= (dev
->flags
& ~(IFF_PROMISC
|
4704 (dev
->gflags
& (IFF_PROMISC
|
4707 if (netif_running(dev
)) {
4708 if (netif_oper_up(dev
))
4709 flags
|= IFF_RUNNING
;
4710 if (netif_carrier_ok(dev
))
4711 flags
|= IFF_LOWER_UP
;
4712 if (netif_dormant(dev
))
4713 flags
|= IFF_DORMANT
;
4718 EXPORT_SYMBOL(dev_get_flags
);
4720 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4722 unsigned int old_flags
= dev
->flags
;
4728 * Set the flags on our device.
4731 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
4732 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
4734 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
4738 * Load in the correct multicast list now the flags have changed.
4741 if ((old_flags
^ flags
) & IFF_MULTICAST
)
4742 dev_change_rx_flags(dev
, IFF_MULTICAST
);
4744 dev_set_rx_mode(dev
);
4747 * Have we downed the interface. We handle IFF_UP ourselves
4748 * according to user attempts to set it, rather than blindly
4753 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
4754 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
4757 dev_set_rx_mode(dev
);
4760 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
4761 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
4763 dev
->gflags
^= IFF_PROMISC
;
4764 dev_set_promiscuity(dev
, inc
);
4767 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4768 is important. Some (broken) drivers set IFF_PROMISC, when
4769 IFF_ALLMULTI is requested not asking us and not reporting.
4771 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
4772 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
4774 dev
->gflags
^= IFF_ALLMULTI
;
4775 dev_set_allmulti(dev
, inc
);
4781 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
)
4783 unsigned int changes
= dev
->flags
^ old_flags
;
4785 if (changes
& IFF_UP
) {
4786 if (dev
->flags
& IFF_UP
)
4787 call_netdevice_notifiers(NETDEV_UP
, dev
);
4789 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
4792 if (dev
->flags
& IFF_UP
&&
4793 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
)))
4794 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
4798 * dev_change_flags - change device settings
4800 * @flags: device state flags
4802 * Change settings on device based state flags. The flags are
4803 * in the userspace exported format.
4805 int dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4808 unsigned int changes
, old_flags
= dev
->flags
;
4810 ret
= __dev_change_flags(dev
, flags
);
4814 changes
= old_flags
^ dev
->flags
;
4816 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
4818 __dev_notify_flags(dev
, old_flags
);
4821 EXPORT_SYMBOL(dev_change_flags
);
4824 * dev_set_mtu - Change maximum transfer unit
4826 * @new_mtu: new transfer unit
4828 * Change the maximum transfer size of the network device.
4830 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
4832 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4835 if (new_mtu
== dev
->mtu
)
4838 /* MTU must be positive. */
4842 if (!netif_device_present(dev
))
4846 if (ops
->ndo_change_mtu
)
4847 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
4851 if (!err
&& dev
->flags
& IFF_UP
)
4852 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
4855 EXPORT_SYMBOL(dev_set_mtu
);
4858 * dev_set_group - Change group this device belongs to
4860 * @new_group: group this device should belong to
4862 void dev_set_group(struct net_device
*dev
, int new_group
)
4864 dev
->group
= new_group
;
4866 EXPORT_SYMBOL(dev_set_group
);
4869 * dev_set_mac_address - Change Media Access Control Address
4873 * Change the hardware (MAC) address of the device
4875 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
4877 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4880 if (!ops
->ndo_set_mac_address
)
4882 if (sa
->sa_family
!= dev
->type
)
4884 if (!netif_device_present(dev
))
4886 err
= ops
->ndo_set_mac_address(dev
, sa
);
4888 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4889 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
4892 EXPORT_SYMBOL(dev_set_mac_address
);
4895 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4897 static int dev_ifsioc_locked(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4900 struct net_device
*dev
= dev_get_by_name_rcu(net
, ifr
->ifr_name
);
4906 case SIOCGIFFLAGS
: /* Get interface flags */
4907 ifr
->ifr_flags
= (short) dev_get_flags(dev
);
4910 case SIOCGIFMETRIC
: /* Get the metric on the interface
4911 (currently unused) */
4912 ifr
->ifr_metric
= 0;
4915 case SIOCGIFMTU
: /* Get the MTU of a device */
4916 ifr
->ifr_mtu
= dev
->mtu
;
4921 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
4923 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
4924 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4925 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
4933 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
4934 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
4935 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
4936 ifr
->ifr_map
.irq
= dev
->irq
;
4937 ifr
->ifr_map
.dma
= dev
->dma
;
4938 ifr
->ifr_map
.port
= dev
->if_port
;
4942 ifr
->ifr_ifindex
= dev
->ifindex
;
4946 ifr
->ifr_qlen
= dev
->tx_queue_len
;
4950 /* dev_ioctl() should ensure this case
4962 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4964 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4967 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
4968 const struct net_device_ops
*ops
;
4973 ops
= dev
->netdev_ops
;
4976 case SIOCSIFFLAGS
: /* Set interface flags */
4977 return dev_change_flags(dev
, ifr
->ifr_flags
);
4979 case SIOCSIFMETRIC
: /* Set the metric on the interface
4980 (currently unused) */
4983 case SIOCSIFMTU
: /* Set the MTU of a device */
4984 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
4987 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
4989 case SIOCSIFHWBROADCAST
:
4990 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
4992 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
4993 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4994 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4998 if (ops
->ndo_set_config
) {
4999 if (!netif_device_present(dev
))
5001 return ops
->ndo_set_config(dev
, &ifr
->ifr_map
);
5006 if (!ops
->ndo_set_rx_mode
||
5007 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
5009 if (!netif_device_present(dev
))
5011 return dev_mc_add_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
5014 if (!ops
->ndo_set_rx_mode
||
5015 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
5017 if (!netif_device_present(dev
))
5019 return dev_mc_del_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
5022 if (ifr
->ifr_qlen
< 0)
5024 dev
->tx_queue_len
= ifr
->ifr_qlen
;
5028 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
5029 return dev_change_name(dev
, ifr
->ifr_newname
);
5032 err
= net_hwtstamp_validate(ifr
);
5038 * Unknown or private ioctl
5041 if ((cmd
>= SIOCDEVPRIVATE
&&
5042 cmd
<= SIOCDEVPRIVATE
+ 15) ||
5043 cmd
== SIOCBONDENSLAVE
||
5044 cmd
== SIOCBONDRELEASE
||
5045 cmd
== SIOCBONDSETHWADDR
||
5046 cmd
== SIOCBONDSLAVEINFOQUERY
||
5047 cmd
== SIOCBONDINFOQUERY
||
5048 cmd
== SIOCBONDCHANGEACTIVE
||
5049 cmd
== SIOCGMIIPHY
||
5050 cmd
== SIOCGMIIREG
||
5051 cmd
== SIOCSMIIREG
||
5052 cmd
== SIOCBRADDIF
||
5053 cmd
== SIOCBRDELIF
||
5054 cmd
== SIOCSHWTSTAMP
||
5055 cmd
== SIOCWANDEV
) {
5057 if (ops
->ndo_do_ioctl
) {
5058 if (netif_device_present(dev
))
5059 err
= ops
->ndo_do_ioctl(dev
, ifr
, cmd
);
5071 * This function handles all "interface"-type I/O control requests. The actual
5072 * 'doing' part of this is dev_ifsioc above.
5076 * dev_ioctl - network device ioctl
5077 * @net: the applicable net namespace
5078 * @cmd: command to issue
5079 * @arg: pointer to a struct ifreq in user space
5081 * Issue ioctl functions to devices. This is normally called by the
5082 * user space syscall interfaces but can sometimes be useful for
5083 * other purposes. The return value is the return from the syscall if
5084 * positive or a negative errno code on error.
5087 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
5093 /* One special case: SIOCGIFCONF takes ifconf argument
5094 and requires shared lock, because it sleeps writing
5098 if (cmd
== SIOCGIFCONF
) {
5100 ret
= dev_ifconf(net
, (char __user
*) arg
);
5104 if (cmd
== SIOCGIFNAME
)
5105 return dev_ifname(net
, (struct ifreq __user
*)arg
);
5107 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
5110 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
5112 colon
= strchr(ifr
.ifr_name
, ':');
5117 * See which interface the caller is talking about.
5122 * These ioctl calls:
5123 * - can be done by all.
5124 * - atomic and do not require locking.
5135 dev_load(net
, ifr
.ifr_name
);
5137 ret
= dev_ifsioc_locked(net
, &ifr
, cmd
);
5142 if (copy_to_user(arg
, &ifr
,
5143 sizeof(struct ifreq
)))
5149 dev_load(net
, ifr
.ifr_name
);
5151 ret
= dev_ethtool(net
, &ifr
);
5156 if (copy_to_user(arg
, &ifr
,
5157 sizeof(struct ifreq
)))
5163 * These ioctl calls:
5164 * - require superuser power.
5165 * - require strict serialization.
5171 if (!capable(CAP_NET_ADMIN
))
5173 dev_load(net
, ifr
.ifr_name
);
5175 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5180 if (copy_to_user(arg
, &ifr
,
5181 sizeof(struct ifreq
)))
5187 * These ioctl calls:
5188 * - require superuser power.
5189 * - require strict serialization.
5190 * - do not return a value
5200 case SIOCSIFHWBROADCAST
:
5203 case SIOCBONDENSLAVE
:
5204 case SIOCBONDRELEASE
:
5205 case SIOCBONDSETHWADDR
:
5206 case SIOCBONDCHANGEACTIVE
:
5210 if (!capable(CAP_NET_ADMIN
))
5213 case SIOCBONDSLAVEINFOQUERY
:
5214 case SIOCBONDINFOQUERY
:
5215 dev_load(net
, ifr
.ifr_name
);
5217 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5222 /* Get the per device memory space. We can add this but
5223 * currently do not support it */
5225 /* Set the per device memory buffer space.
5226 * Not applicable in our case */
5231 * Unknown or private ioctl.
5234 if (cmd
== SIOCWANDEV
||
5235 (cmd
>= SIOCDEVPRIVATE
&&
5236 cmd
<= SIOCDEVPRIVATE
+ 15)) {
5237 dev_load(net
, ifr
.ifr_name
);
5239 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5241 if (!ret
&& copy_to_user(arg
, &ifr
,
5242 sizeof(struct ifreq
)))
5246 /* Take care of Wireless Extensions */
5247 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
5248 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
5255 * dev_new_index - allocate an ifindex
5256 * @net: the applicable net namespace
5258 * Returns a suitable unique value for a new device interface
5259 * number. The caller must hold the rtnl semaphore or the
5260 * dev_base_lock to be sure it remains unique.
5262 static int dev_new_index(struct net
*net
)
5264 int ifindex
= net
->ifindex
;
5268 if (!__dev_get_by_index(net
, ifindex
))
5269 return net
->ifindex
= ifindex
;
5273 /* Delayed registration/unregisteration */
5274 static LIST_HEAD(net_todo_list
);
5276 static void net_set_todo(struct net_device
*dev
)
5278 list_add_tail(&dev
->todo_list
, &net_todo_list
);
5281 static void rollback_registered_many(struct list_head
*head
)
5283 struct net_device
*dev
, *tmp
;
5285 BUG_ON(dev_boot_phase
);
5288 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
5289 /* Some devices call without registering
5290 * for initialization unwind. Remove those
5291 * devices and proceed with the remaining.
5293 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5294 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5298 list_del(&dev
->unreg_list
);
5301 dev
->dismantle
= true;
5302 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
5305 /* If device is running, close it first. */
5306 dev_close_many(head
);
5308 list_for_each_entry(dev
, head
, unreg_list
) {
5309 /* And unlink it from device chain. */
5310 unlist_netdevice(dev
);
5312 dev
->reg_state
= NETREG_UNREGISTERING
;
5317 list_for_each_entry(dev
, head
, unreg_list
) {
5318 /* Shutdown queueing discipline. */
5322 /* Notify protocols, that we are about to destroy
5323 this device. They should clean all the things.
5325 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5327 if (!dev
->rtnl_link_ops
||
5328 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5329 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
5332 * Flush the unicast and multicast chains
5337 if (dev
->netdev_ops
->ndo_uninit
)
5338 dev
->netdev_ops
->ndo_uninit(dev
);
5340 /* Notifier chain MUST detach us from master device. */
5341 WARN_ON(dev
->master
);
5343 /* Remove entries from kobject tree */
5344 netdev_unregister_kobject(dev
);
5349 list_for_each_entry(dev
, head
, unreg_list
)
5353 static void rollback_registered(struct net_device
*dev
)
5357 list_add(&dev
->unreg_list
, &single
);
5358 rollback_registered_many(&single
);
5362 static netdev_features_t
netdev_fix_features(struct net_device
*dev
,
5363 netdev_features_t features
)
5365 /* Fix illegal checksum combinations */
5366 if ((features
& NETIF_F_HW_CSUM
) &&
5367 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5368 netdev_warn(dev
, "mixed HW and IP checksum settings.\n");
5369 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5372 /* Fix illegal SG+CSUM combinations. */
5373 if ((features
& NETIF_F_SG
) &&
5374 !(features
& NETIF_F_ALL_CSUM
)) {
5376 "Dropping NETIF_F_SG since no checksum feature.\n");
5377 features
&= ~NETIF_F_SG
;
5380 /* TSO requires that SG is present as well. */
5381 if ((features
& NETIF_F_ALL_TSO
) && !(features
& NETIF_F_SG
)) {
5382 netdev_dbg(dev
, "Dropping TSO features since no SG feature.\n");
5383 features
&= ~NETIF_F_ALL_TSO
;
5386 /* TSO ECN requires that TSO is present as well. */
5387 if ((features
& NETIF_F_ALL_TSO
) == NETIF_F_TSO_ECN
)
5388 features
&= ~NETIF_F_TSO_ECN
;
5390 /* Software GSO depends on SG. */
5391 if ((features
& NETIF_F_GSO
) && !(features
& NETIF_F_SG
)) {
5392 netdev_dbg(dev
, "Dropping NETIF_F_GSO since no SG feature.\n");
5393 features
&= ~NETIF_F_GSO
;
5396 /* UFO needs SG and checksumming */
5397 if (features
& NETIF_F_UFO
) {
5398 /* maybe split UFO into V4 and V6? */
5399 if (!((features
& NETIF_F_GEN_CSUM
) ||
5400 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))
5401 == (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5403 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5404 features
&= ~NETIF_F_UFO
;
5407 if (!(features
& NETIF_F_SG
)) {
5409 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5410 features
&= ~NETIF_F_UFO
;
5417 int __netdev_update_features(struct net_device
*dev
)
5419 netdev_features_t features
;
5424 features
= netdev_get_wanted_features(dev
);
5426 if (dev
->netdev_ops
->ndo_fix_features
)
5427 features
= dev
->netdev_ops
->ndo_fix_features(dev
, features
);
5429 /* driver might be less strict about feature dependencies */
5430 features
= netdev_fix_features(dev
, features
);
5432 if (dev
->features
== features
)
5435 netdev_dbg(dev
, "Features changed: %pNF -> %pNF\n",
5436 &dev
->features
, &features
);
5438 if (dev
->netdev_ops
->ndo_set_features
)
5439 err
= dev
->netdev_ops
->ndo_set_features(dev
, features
);
5441 if (unlikely(err
< 0)) {
5443 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5444 err
, &features
, &dev
->features
);
5449 dev
->features
= features
;
5455 * netdev_update_features - recalculate device features
5456 * @dev: the device to check
5458 * Recalculate dev->features set and send notifications if it
5459 * has changed. Should be called after driver or hardware dependent
5460 * conditions might have changed that influence the features.
5462 void netdev_update_features(struct net_device
*dev
)
5464 if (__netdev_update_features(dev
))
5465 netdev_features_change(dev
);
5467 EXPORT_SYMBOL(netdev_update_features
);
5470 * netdev_change_features - recalculate device features
5471 * @dev: the device to check
5473 * Recalculate dev->features set and send notifications even
5474 * if they have not changed. Should be called instead of
5475 * netdev_update_features() if also dev->vlan_features might
5476 * have changed to allow the changes to be propagated to stacked
5479 void netdev_change_features(struct net_device
*dev
)
5481 __netdev_update_features(dev
);
5482 netdev_features_change(dev
);
5484 EXPORT_SYMBOL(netdev_change_features
);
5487 * netif_stacked_transfer_operstate - transfer operstate
5488 * @rootdev: the root or lower level device to transfer state from
5489 * @dev: the device to transfer operstate to
5491 * Transfer operational state from root to device. This is normally
5492 * called when a stacking relationship exists between the root
5493 * device and the device(a leaf device).
5495 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
5496 struct net_device
*dev
)
5498 if (rootdev
->operstate
== IF_OPER_DORMANT
)
5499 netif_dormant_on(dev
);
5501 netif_dormant_off(dev
);
5503 if (netif_carrier_ok(rootdev
)) {
5504 if (!netif_carrier_ok(dev
))
5505 netif_carrier_on(dev
);
5507 if (netif_carrier_ok(dev
))
5508 netif_carrier_off(dev
);
5511 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
5514 static int netif_alloc_rx_queues(struct net_device
*dev
)
5516 unsigned int i
, count
= dev
->num_rx_queues
;
5517 struct netdev_rx_queue
*rx
;
5521 rx
= kcalloc(count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5523 pr_err("netdev: Unable to allocate %u rx queues\n", count
);
5528 for (i
= 0; i
< count
; i
++)
5534 static void netdev_init_one_queue(struct net_device
*dev
,
5535 struct netdev_queue
*queue
, void *_unused
)
5537 /* Initialize queue lock */
5538 spin_lock_init(&queue
->_xmit_lock
);
5539 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
5540 queue
->xmit_lock_owner
= -1;
5541 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
5544 dql_init(&queue
->dql
, HZ
);
5548 static int netif_alloc_netdev_queues(struct net_device
*dev
)
5550 unsigned int count
= dev
->num_tx_queues
;
5551 struct netdev_queue
*tx
;
5555 tx
= kcalloc(count
, sizeof(struct netdev_queue
), GFP_KERNEL
);
5557 pr_err("netdev: Unable to allocate %u tx queues\n", count
);
5562 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5563 spin_lock_init(&dev
->tx_global_lock
);
5569 * register_netdevice - register a network device
5570 * @dev: device to register
5572 * Take a completed network device structure and add it to the kernel
5573 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5574 * chain. 0 is returned on success. A negative errno code is returned
5575 * on a failure to set up the device, or if the name is a duplicate.
5577 * Callers must hold the rtnl semaphore. You may want
5578 * register_netdev() instead of this.
5581 * The locking appears insufficient to guarantee two parallel registers
5582 * will not get the same name.
5585 int register_netdevice(struct net_device
*dev
)
5588 struct net
*net
= dev_net(dev
);
5590 BUG_ON(dev_boot_phase
);
5595 /* When net_device's are persistent, this will be fatal. */
5596 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
5599 spin_lock_init(&dev
->addr_list_lock
);
5600 netdev_set_addr_lockdep_class(dev
);
5604 ret
= dev_get_valid_name(net
, dev
, dev
->name
);
5608 /* Init, if this function is available */
5609 if (dev
->netdev_ops
->ndo_init
) {
5610 ret
= dev
->netdev_ops
->ndo_init(dev
);
5620 dev
->ifindex
= dev_new_index(net
);
5621 else if (__dev_get_by_index(net
, dev
->ifindex
))
5624 if (dev
->iflink
== -1)
5625 dev
->iflink
= dev
->ifindex
;
5627 /* Transfer changeable features to wanted_features and enable
5628 * software offloads (GSO and GRO).
5630 dev
->hw_features
|= NETIF_F_SOFT_FEATURES
;
5631 dev
->features
|= NETIF_F_SOFT_FEATURES
;
5632 dev
->wanted_features
= dev
->features
& dev
->hw_features
;
5634 /* Turn on no cache copy if HW is doing checksum */
5635 if (!(dev
->flags
& IFF_LOOPBACK
)) {
5636 dev
->hw_features
|= NETIF_F_NOCACHE_COPY
;
5637 if (dev
->features
& NETIF_F_ALL_CSUM
) {
5638 dev
->wanted_features
|= NETIF_F_NOCACHE_COPY
;
5639 dev
->features
|= NETIF_F_NOCACHE_COPY
;
5643 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5645 dev
->vlan_features
|= NETIF_F_HIGHDMA
;
5647 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
5648 ret
= notifier_to_errno(ret
);
5652 ret
= netdev_register_kobject(dev
);
5655 dev
->reg_state
= NETREG_REGISTERED
;
5657 __netdev_update_features(dev
);
5660 * Default initial state at registry is that the
5661 * device is present.
5664 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5666 linkwatch_init_dev(dev
);
5668 dev_init_scheduler(dev
);
5670 list_netdevice(dev
);
5671 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
5673 /* Notify protocols, that a new device appeared. */
5674 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5675 ret
= notifier_to_errno(ret
);
5677 rollback_registered(dev
);
5678 dev
->reg_state
= NETREG_UNREGISTERED
;
5681 * Prevent userspace races by waiting until the network
5682 * device is fully setup before sending notifications.
5684 if (!dev
->rtnl_link_ops
||
5685 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5686 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5692 if (dev
->netdev_ops
->ndo_uninit
)
5693 dev
->netdev_ops
->ndo_uninit(dev
);
5696 EXPORT_SYMBOL(register_netdevice
);
5699 * init_dummy_netdev - init a dummy network device for NAPI
5700 * @dev: device to init
5702 * This takes a network device structure and initialize the minimum
5703 * amount of fields so it can be used to schedule NAPI polls without
5704 * registering a full blown interface. This is to be used by drivers
5705 * that need to tie several hardware interfaces to a single NAPI
5706 * poll scheduler due to HW limitations.
5708 int init_dummy_netdev(struct net_device
*dev
)
5710 /* Clear everything. Note we don't initialize spinlocks
5711 * are they aren't supposed to be taken by any of the
5712 * NAPI code and this dummy netdev is supposed to be
5713 * only ever used for NAPI polls
5715 memset(dev
, 0, sizeof(struct net_device
));
5717 /* make sure we BUG if trying to hit standard
5718 * register/unregister code path
5720 dev
->reg_state
= NETREG_DUMMY
;
5722 /* NAPI wants this */
5723 INIT_LIST_HEAD(&dev
->napi_list
);
5725 /* a dummy interface is started by default */
5726 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5727 set_bit(__LINK_STATE_START
, &dev
->state
);
5729 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5730 * because users of this 'device' dont need to change
5736 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
5740 * register_netdev - register a network device
5741 * @dev: device to register
5743 * Take a completed network device structure and add it to the kernel
5744 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5745 * chain. 0 is returned on success. A negative errno code is returned
5746 * on a failure to set up the device, or if the name is a duplicate.
5748 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5749 * and expands the device name if you passed a format string to
5752 int register_netdev(struct net_device
*dev
)
5757 err
= register_netdevice(dev
);
5761 EXPORT_SYMBOL(register_netdev
);
5763 int netdev_refcnt_read(const struct net_device
*dev
)
5767 for_each_possible_cpu(i
)
5768 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
5771 EXPORT_SYMBOL(netdev_refcnt_read
);
5774 * netdev_wait_allrefs - wait until all references are gone.
5775 * @dev: target net_device
5777 * This is called when unregistering network devices.
5779 * Any protocol or device that holds a reference should register
5780 * for netdevice notification, and cleanup and put back the
5781 * reference if they receive an UNREGISTER event.
5782 * We can get stuck here if buggy protocols don't correctly
5785 static void netdev_wait_allrefs(struct net_device
*dev
)
5787 unsigned long rebroadcast_time
, warning_time
;
5790 linkwatch_forget_dev(dev
);
5792 rebroadcast_time
= warning_time
= jiffies
;
5793 refcnt
= netdev_refcnt_read(dev
);
5795 while (refcnt
!= 0) {
5796 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
5799 /* Rebroadcast unregister notification */
5800 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5806 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
5807 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
5809 /* We must not have linkwatch events
5810 * pending on unregister. If this
5811 * happens, we simply run the queue
5812 * unscheduled, resulting in a noop
5815 linkwatch_run_queue();
5820 rebroadcast_time
= jiffies
;
5825 refcnt
= netdev_refcnt_read(dev
);
5827 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
5828 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5830 warning_time
= jiffies
;
5839 * register_netdevice(x1);
5840 * register_netdevice(x2);
5842 * unregister_netdevice(y1);
5843 * unregister_netdevice(y2);
5849 * We are invoked by rtnl_unlock().
5850 * This allows us to deal with problems:
5851 * 1) We can delete sysfs objects which invoke hotplug
5852 * without deadlocking with linkwatch via keventd.
5853 * 2) Since we run with the RTNL semaphore not held, we can sleep
5854 * safely in order to wait for the netdev refcnt to drop to zero.
5856 * We must not return until all unregister events added during
5857 * the interval the lock was held have been completed.
5859 void netdev_run_todo(void)
5861 struct list_head list
;
5863 /* Snapshot list, allow later requests */
5864 list_replace_init(&net_todo_list
, &list
);
5869 /* Wait for rcu callbacks to finish before next phase */
5870 if (!list_empty(&list
))
5873 while (!list_empty(&list
)) {
5874 struct net_device
*dev
5875 = list_first_entry(&list
, struct net_device
, todo_list
);
5876 list_del(&dev
->todo_list
);
5879 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
5882 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
5883 pr_err("network todo '%s' but state %d\n",
5884 dev
->name
, dev
->reg_state
);
5889 dev
->reg_state
= NETREG_UNREGISTERED
;
5891 on_each_cpu(flush_backlog
, dev
, 1);
5893 netdev_wait_allrefs(dev
);
5896 BUG_ON(netdev_refcnt_read(dev
));
5897 WARN_ON(rcu_access_pointer(dev
->ip_ptr
));
5898 WARN_ON(rcu_access_pointer(dev
->ip6_ptr
));
5899 WARN_ON(dev
->dn_ptr
);
5901 if (dev
->destructor
)
5902 dev
->destructor(dev
);
5904 /* Free network device */
5905 kobject_put(&dev
->dev
.kobj
);
5909 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5910 * fields in the same order, with only the type differing.
5912 void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
5913 const struct net_device_stats
*netdev_stats
)
5915 #if BITS_PER_LONG == 64
5916 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
5917 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
5919 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
5920 const unsigned long *src
= (const unsigned long *)netdev_stats
;
5921 u64
*dst
= (u64
*)stats64
;
5923 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
5924 sizeof(*stats64
) / sizeof(u64
));
5925 for (i
= 0; i
< n
; i
++)
5929 EXPORT_SYMBOL(netdev_stats_to_stats64
);
5932 * dev_get_stats - get network device statistics
5933 * @dev: device to get statistics from
5934 * @storage: place to store stats
5936 * Get network statistics from device. Return @storage.
5937 * The device driver may provide its own method by setting
5938 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5939 * otherwise the internal statistics structure is used.
5941 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
5942 struct rtnl_link_stats64
*storage
)
5944 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5946 if (ops
->ndo_get_stats64
) {
5947 memset(storage
, 0, sizeof(*storage
));
5948 ops
->ndo_get_stats64(dev
, storage
);
5949 } else if (ops
->ndo_get_stats
) {
5950 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
5952 netdev_stats_to_stats64(storage
, &dev
->stats
);
5954 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
5957 EXPORT_SYMBOL(dev_get_stats
);
5959 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
5961 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
5963 #ifdef CONFIG_NET_CLS_ACT
5966 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
5969 netdev_init_one_queue(dev
, queue
, NULL
);
5970 queue
->qdisc
= &noop_qdisc
;
5971 queue
->qdisc_sleeping
= &noop_qdisc
;
5972 rcu_assign_pointer(dev
->ingress_queue
, queue
);
5977 static const struct ethtool_ops default_ethtool_ops
;
5980 * alloc_netdev_mqs - allocate network device
5981 * @sizeof_priv: size of private data to allocate space for
5982 * @name: device name format string
5983 * @setup: callback to initialize device
5984 * @txqs: the number of TX subqueues to allocate
5985 * @rxqs: the number of RX subqueues to allocate
5987 * Allocates a struct net_device with private data area for driver use
5988 * and performs basic initialization. Also allocates subquue structs
5989 * for each queue on the device.
5991 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
5992 void (*setup
)(struct net_device
*),
5993 unsigned int txqs
, unsigned int rxqs
)
5995 struct net_device
*dev
;
5997 struct net_device
*p
;
5999 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
6002 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6008 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6013 alloc_size
= sizeof(struct net_device
);
6015 /* ensure 32-byte alignment of private area */
6016 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
6017 alloc_size
+= sizeof_priv
;
6019 /* ensure 32-byte alignment of whole construct */
6020 alloc_size
+= NETDEV_ALIGN
- 1;
6022 p
= kzalloc(alloc_size
, GFP_KERNEL
);
6024 pr_err("alloc_netdev: Unable to allocate device\n");
6028 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
6029 dev
->padded
= (char *)dev
- (char *)p
;
6031 dev
->pcpu_refcnt
= alloc_percpu(int);
6032 if (!dev
->pcpu_refcnt
)
6035 if (dev_addr_init(dev
))
6041 dev_net_set(dev
, &init_net
);
6043 dev
->gso_max_size
= GSO_MAX_SIZE
;
6044 dev
->gso_max_segs
= GSO_MAX_SEGS
;
6046 INIT_LIST_HEAD(&dev
->napi_list
);
6047 INIT_LIST_HEAD(&dev
->unreg_list
);
6048 INIT_LIST_HEAD(&dev
->link_watch_list
);
6049 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
6052 dev
->num_tx_queues
= txqs
;
6053 dev
->real_num_tx_queues
= txqs
;
6054 if (netif_alloc_netdev_queues(dev
))
6058 dev
->num_rx_queues
= rxqs
;
6059 dev
->real_num_rx_queues
= rxqs
;
6060 if (netif_alloc_rx_queues(dev
))
6064 strcpy(dev
->name
, name
);
6065 dev
->group
= INIT_NETDEV_GROUP
;
6066 if (!dev
->ethtool_ops
)
6067 dev
->ethtool_ops
= &default_ethtool_ops
;
6075 free_percpu(dev
->pcpu_refcnt
);
6085 EXPORT_SYMBOL(alloc_netdev_mqs
);
6088 * free_netdev - free network device
6091 * This function does the last stage of destroying an allocated device
6092 * interface. The reference to the device object is released.
6093 * If this is the last reference then it will be freed.
6095 void free_netdev(struct net_device
*dev
)
6097 struct napi_struct
*p
, *n
;
6099 release_net(dev_net(dev
));
6106 kfree(rcu_dereference_protected(dev
->ingress_queue
, 1));
6108 /* Flush device addresses */
6109 dev_addr_flush(dev
);
6111 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
6114 free_percpu(dev
->pcpu_refcnt
);
6115 dev
->pcpu_refcnt
= NULL
;
6117 /* Compatibility with error handling in drivers */
6118 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
6119 kfree((char *)dev
- dev
->padded
);
6123 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
6124 dev
->reg_state
= NETREG_RELEASED
;
6126 /* will free via device release */
6127 put_device(&dev
->dev
);
6129 EXPORT_SYMBOL(free_netdev
);
6132 * synchronize_net - Synchronize with packet receive processing
6134 * Wait for packets currently being received to be done.
6135 * Does not block later packets from starting.
6137 void synchronize_net(void)
6140 if (rtnl_is_locked())
6141 synchronize_rcu_expedited();
6145 EXPORT_SYMBOL(synchronize_net
);
6148 * unregister_netdevice_queue - remove device from the kernel
6152 * This function shuts down a device interface and removes it
6153 * from the kernel tables.
6154 * If head not NULL, device is queued to be unregistered later.
6156 * Callers must hold the rtnl semaphore. You may want
6157 * unregister_netdev() instead of this.
6160 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
6165 list_move_tail(&dev
->unreg_list
, head
);
6167 rollback_registered(dev
);
6168 /* Finish processing unregister after unlock */
6172 EXPORT_SYMBOL(unregister_netdevice_queue
);
6175 * unregister_netdevice_many - unregister many devices
6176 * @head: list of devices
6178 void unregister_netdevice_many(struct list_head
*head
)
6180 struct net_device
*dev
;
6182 if (!list_empty(head
)) {
6183 rollback_registered_many(head
);
6184 list_for_each_entry(dev
, head
, unreg_list
)
6188 EXPORT_SYMBOL(unregister_netdevice_many
);
6191 * unregister_netdev - remove device from the kernel
6194 * This function shuts down a device interface and removes it
6195 * from the kernel tables.
6197 * This is just a wrapper for unregister_netdevice that takes
6198 * the rtnl semaphore. In general you want to use this and not
6199 * unregister_netdevice.
6201 void unregister_netdev(struct net_device
*dev
)
6204 unregister_netdevice(dev
);
6207 EXPORT_SYMBOL(unregister_netdev
);
6210 * dev_change_net_namespace - move device to different nethost namespace
6212 * @net: network namespace
6213 * @pat: If not NULL name pattern to try if the current device name
6214 * is already taken in the destination network namespace.
6216 * This function shuts down a device interface and moves it
6217 * to a new network namespace. On success 0 is returned, on
6218 * a failure a netagive errno code is returned.
6220 * Callers must hold the rtnl semaphore.
6223 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
6229 /* Don't allow namespace local devices to be moved. */
6231 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6234 /* Ensure the device has been registrered */
6236 if (dev
->reg_state
!= NETREG_REGISTERED
)
6239 /* Get out if there is nothing todo */
6241 if (net_eq(dev_net(dev
), net
))
6244 /* Pick the destination device name, and ensure
6245 * we can use it in the destination network namespace.
6248 if (__dev_get_by_name(net
, dev
->name
)) {
6249 /* We get here if we can't use the current device name */
6252 if (dev_get_valid_name(net
, dev
, pat
) < 0)
6257 * And now a mini version of register_netdevice unregister_netdevice.
6260 /* If device is running close it first. */
6263 /* And unlink it from device chain */
6265 unlist_netdevice(dev
);
6269 /* Shutdown queueing discipline. */
6272 /* Notify protocols, that we are about to destroy
6273 this device. They should clean all the things.
6275 Note that dev->reg_state stays at NETREG_REGISTERED.
6276 This is wanted because this way 8021q and macvlan know
6277 the device is just moving and can keep their slaves up.
6279 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6281 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
6282 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
6285 * Flush the unicast and multicast chains
6290 /* Actually switch the network namespace */
6291 dev_net_set(dev
, net
);
6293 /* If there is an ifindex conflict assign a new one */
6294 if (__dev_get_by_index(net
, dev
->ifindex
)) {
6295 int iflink
= (dev
->iflink
== dev
->ifindex
);
6296 dev
->ifindex
= dev_new_index(net
);
6298 dev
->iflink
= dev
->ifindex
;
6301 /* Fixup kobjects */
6302 err
= device_rename(&dev
->dev
, dev
->name
);
6305 /* Add the device back in the hashes */
6306 list_netdevice(dev
);
6308 /* Notify protocols, that a new device appeared. */
6309 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
6312 * Prevent userspace races by waiting until the network
6313 * device is fully setup before sending notifications.
6315 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
6322 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
6324 static int dev_cpu_callback(struct notifier_block
*nfb
,
6325 unsigned long action
,
6328 struct sk_buff
**list_skb
;
6329 struct sk_buff
*skb
;
6330 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
6331 struct softnet_data
*sd
, *oldsd
;
6333 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
6336 local_irq_disable();
6337 cpu
= smp_processor_id();
6338 sd
= &per_cpu(softnet_data
, cpu
);
6339 oldsd
= &per_cpu(softnet_data
, oldcpu
);
6341 /* Find end of our completion_queue. */
6342 list_skb
= &sd
->completion_queue
;
6344 list_skb
= &(*list_skb
)->next
;
6345 /* Append completion queue from offline CPU. */
6346 *list_skb
= oldsd
->completion_queue
;
6347 oldsd
->completion_queue
= NULL
;
6349 /* Append output queue from offline CPU. */
6350 if (oldsd
->output_queue
) {
6351 *sd
->output_queue_tailp
= oldsd
->output_queue
;
6352 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
6353 oldsd
->output_queue
= NULL
;
6354 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
6356 /* Append NAPI poll list from offline CPU. */
6357 if (!list_empty(&oldsd
->poll_list
)) {
6358 list_splice_init(&oldsd
->poll_list
, &sd
->poll_list
);
6359 raise_softirq_irqoff(NET_RX_SOFTIRQ
);
6362 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
6365 /* Process offline CPU's input_pkt_queue */
6366 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
6368 input_queue_head_incr(oldsd
);
6370 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
6372 input_queue_head_incr(oldsd
);
6380 * netdev_increment_features - increment feature set by one
6381 * @all: current feature set
6382 * @one: new feature set
6383 * @mask: mask feature set
6385 * Computes a new feature set after adding a device with feature set
6386 * @one to the master device with current feature set @all. Will not
6387 * enable anything that is off in @mask. Returns the new feature set.
6389 netdev_features_t
netdev_increment_features(netdev_features_t all
,
6390 netdev_features_t one
, netdev_features_t mask
)
6392 if (mask
& NETIF_F_GEN_CSUM
)
6393 mask
|= NETIF_F_ALL_CSUM
;
6394 mask
|= NETIF_F_VLAN_CHALLENGED
;
6396 all
|= one
& (NETIF_F_ONE_FOR_ALL
|NETIF_F_ALL_CSUM
) & mask
;
6397 all
&= one
| ~NETIF_F_ALL_FOR_ALL
;
6399 /* If one device supports hw checksumming, set for all. */
6400 if (all
& NETIF_F_GEN_CSUM
)
6401 all
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_GEN_CSUM
);
6405 EXPORT_SYMBOL(netdev_increment_features
);
6407 static struct hlist_head
*netdev_create_hash(void)
6410 struct hlist_head
*hash
;
6412 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
6414 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
6415 INIT_HLIST_HEAD(&hash
[i
]);
6420 /* Initialize per network namespace state */
6421 static int __net_init
netdev_init(struct net
*net
)
6423 if (net
!= &init_net
)
6424 INIT_LIST_HEAD(&net
->dev_base_head
);
6426 net
->dev_name_head
= netdev_create_hash();
6427 if (net
->dev_name_head
== NULL
)
6430 net
->dev_index_head
= netdev_create_hash();
6431 if (net
->dev_index_head
== NULL
)
6437 kfree(net
->dev_name_head
);
6443 * netdev_drivername - network driver for the device
6444 * @dev: network device
6446 * Determine network driver for device.
6448 const char *netdev_drivername(const struct net_device
*dev
)
6450 const struct device_driver
*driver
;
6451 const struct device
*parent
;
6452 const char *empty
= "";
6454 parent
= dev
->dev
.parent
;
6458 driver
= parent
->driver
;
6459 if (driver
&& driver
->name
)
6460 return driver
->name
;
6464 int __netdev_printk(const char *level
, const struct net_device
*dev
,
6465 struct va_format
*vaf
)
6469 if (dev
&& dev
->dev
.parent
)
6470 r
= dev_printk(level
, dev
->dev
.parent
, "%s: %pV",
6471 netdev_name(dev
), vaf
);
6473 r
= printk("%s%s: %pV", level
, netdev_name(dev
), vaf
);
6475 r
= printk("%s(NULL net_device): %pV", level
, vaf
);
6479 EXPORT_SYMBOL(__netdev_printk
);
6481 int netdev_printk(const char *level
, const struct net_device
*dev
,
6482 const char *format
, ...)
6484 struct va_format vaf
;
6488 va_start(args
, format
);
6493 r
= __netdev_printk(level
, dev
, &vaf
);
6498 EXPORT_SYMBOL(netdev_printk
);
6500 #define define_netdev_printk_level(func, level) \
6501 int func(const struct net_device *dev, const char *fmt, ...) \
6504 struct va_format vaf; \
6507 va_start(args, fmt); \
6512 r = __netdev_printk(level, dev, &vaf); \
6517 EXPORT_SYMBOL(func);
6519 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
6520 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
6521 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
6522 define_netdev_printk_level(netdev_err
, KERN_ERR
);
6523 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
6524 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
6525 define_netdev_printk_level(netdev_info
, KERN_INFO
);
6527 static void __net_exit
netdev_exit(struct net
*net
)
6529 kfree(net
->dev_name_head
);
6530 kfree(net
->dev_index_head
);
6533 static struct pernet_operations __net_initdata netdev_net_ops
= {
6534 .init
= netdev_init
,
6535 .exit
= netdev_exit
,
6538 static void __net_exit
default_device_exit(struct net
*net
)
6540 struct net_device
*dev
, *aux
;
6542 * Push all migratable network devices back to the
6543 * initial network namespace
6546 for_each_netdev_safe(net
, dev
, aux
) {
6548 char fb_name
[IFNAMSIZ
];
6550 /* Ignore unmoveable devices (i.e. loopback) */
6551 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6554 /* Leave virtual devices for the generic cleanup */
6555 if (dev
->rtnl_link_ops
)
6558 /* Push remaining network devices to init_net */
6559 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
6560 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
6562 pr_emerg("%s: failed to move %s to init_net: %d\n",
6563 __func__
, dev
->name
, err
);
6570 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
6572 /* At exit all network devices most be removed from a network
6573 * namespace. Do this in the reverse order of registration.
6574 * Do this across as many network namespaces as possible to
6575 * improve batching efficiency.
6577 struct net_device
*dev
;
6579 LIST_HEAD(dev_kill_list
);
6582 list_for_each_entry(net
, net_list
, exit_list
) {
6583 for_each_netdev_reverse(net
, dev
) {
6584 if (dev
->rtnl_link_ops
)
6585 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
6587 unregister_netdevice_queue(dev
, &dev_kill_list
);
6590 unregister_netdevice_many(&dev_kill_list
);
6591 list_del(&dev_kill_list
);
6595 static struct pernet_operations __net_initdata default_device_ops
= {
6596 .exit
= default_device_exit
,
6597 .exit_batch
= default_device_exit_batch
,
6601 * Initialize the DEV module. At boot time this walks the device list and
6602 * unhooks any devices that fail to initialise (normally hardware not
6603 * present) and leaves us with a valid list of present and active devices.
6608 * This is called single threaded during boot, so no need
6609 * to take the rtnl semaphore.
6611 static int __init
net_dev_init(void)
6613 int i
, rc
= -ENOMEM
;
6615 BUG_ON(!dev_boot_phase
);
6617 if (dev_proc_init())
6620 if (netdev_kobject_init())
6623 INIT_LIST_HEAD(&ptype_all
);
6624 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
6625 INIT_LIST_HEAD(&ptype_base
[i
]);
6627 if (register_pernet_subsys(&netdev_net_ops
))
6631 * Initialise the packet receive queues.
6634 for_each_possible_cpu(i
) {
6635 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
6637 memset(sd
, 0, sizeof(*sd
));
6638 skb_queue_head_init(&sd
->input_pkt_queue
);
6639 skb_queue_head_init(&sd
->process_queue
);
6640 sd
->completion_queue
= NULL
;
6641 INIT_LIST_HEAD(&sd
->poll_list
);
6642 sd
->output_queue
= NULL
;
6643 sd
->output_queue_tailp
= &sd
->output_queue
;
6645 sd
->csd
.func
= rps_trigger_softirq
;
6651 sd
->backlog
.poll
= process_backlog
;
6652 sd
->backlog
.weight
= weight_p
;
6653 sd
->backlog
.gro_list
= NULL
;
6654 sd
->backlog
.gro_count
= 0;
6659 /* The loopback device is special if any other network devices
6660 * is present in a network namespace the loopback device must
6661 * be present. Since we now dynamically allocate and free the
6662 * loopback device ensure this invariant is maintained by
6663 * keeping the loopback device as the first device on the
6664 * list of network devices. Ensuring the loopback devices
6665 * is the first device that appears and the last network device
6668 if (register_pernet_device(&loopback_net_ops
))
6671 if (register_pernet_device(&default_device_ops
))
6674 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
6675 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
6677 hotcpu_notifier(dev_cpu_callback
, 0);
6685 subsys_initcall(net_dev_init
);
6687 static int __init
initialize_hashrnd(void)
6689 get_random_bytes(&hashrnd
, sizeof(hashrnd
));
6693 late_initcall_sync(initialize_hashrnd
);