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 (skb
->ip_summed
!= CHECKSUM_NONE
&&
2160 !can_checksum_protocol(features
, protocol
)) {
2161 features
&= ~NETIF_F_ALL_CSUM
;
2162 features
&= ~NETIF_F_SG
;
2163 } else if (illegal_highdma(skb
->dev
, skb
)) {
2164 features
&= ~NETIF_F_SG
;
2170 netdev_features_t
netif_skb_features(struct sk_buff
*skb
)
2172 __be16 protocol
= skb
->protocol
;
2173 netdev_features_t features
= skb
->dev
->features
;
2175 if (skb_shinfo(skb
)->gso_segs
> skb
->dev
->gso_max_segs
)
2176 features
&= ~NETIF_F_GSO_MASK
;
2178 if (protocol
== htons(ETH_P_8021Q
)) {
2179 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
2180 protocol
= veh
->h_vlan_encapsulated_proto
;
2181 } else if (!vlan_tx_tag_present(skb
)) {
2182 return harmonize_features(skb
, protocol
, features
);
2185 features
&= (skb
->dev
->vlan_features
| NETIF_F_HW_VLAN_TX
);
2187 if (protocol
!= htons(ETH_P_8021Q
)) {
2188 return harmonize_features(skb
, protocol
, features
);
2190 features
&= NETIF_F_SG
| NETIF_F_HIGHDMA
| NETIF_F_FRAGLIST
|
2191 NETIF_F_GEN_CSUM
| NETIF_F_HW_VLAN_TX
;
2192 return harmonize_features(skb
, protocol
, features
);
2195 EXPORT_SYMBOL(netif_skb_features
);
2198 * Returns true if either:
2199 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2200 * 2. skb is fragmented and the device does not support SG.
2202 static inline int skb_needs_linearize(struct sk_buff
*skb
,
2205 return skb_is_nonlinear(skb
) &&
2206 ((skb_has_frag_list(skb
) &&
2207 !(features
& NETIF_F_FRAGLIST
)) ||
2208 (skb_shinfo(skb
)->nr_frags
&&
2209 !(features
& NETIF_F_SG
)));
2212 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
2213 struct netdev_queue
*txq
)
2215 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2216 int rc
= NETDEV_TX_OK
;
2217 unsigned int skb_len
;
2219 if (likely(!skb
->next
)) {
2220 netdev_features_t features
;
2223 * If device doesn't need skb->dst, release it right now while
2224 * its hot in this cpu cache
2226 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2229 features
= netif_skb_features(skb
);
2231 if (vlan_tx_tag_present(skb
) &&
2232 !(features
& NETIF_F_HW_VLAN_TX
)) {
2233 skb
= __vlan_put_tag(skb
, vlan_tx_tag_get(skb
));
2240 if (netif_needs_gso(skb
, features
)) {
2241 if (unlikely(dev_gso_segment(skb
, features
)))
2246 if (skb_needs_linearize(skb
, features
) &&
2247 __skb_linearize(skb
))
2250 /* If packet is not checksummed and device does not
2251 * support checksumming for this protocol, complete
2252 * checksumming here.
2254 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2255 skb_set_transport_header(skb
,
2256 skb_checksum_start_offset(skb
));
2257 if (!(features
& NETIF_F_ALL_CSUM
) &&
2258 skb_checksum_help(skb
))
2263 if (!list_empty(&ptype_all
))
2264 dev_queue_xmit_nit(skb
, dev
);
2267 rc
= ops
->ndo_start_xmit(skb
, dev
);
2268 trace_net_dev_xmit(skb
, rc
, dev
, skb_len
);
2269 if (rc
== NETDEV_TX_OK
)
2270 txq_trans_update(txq
);
2276 struct sk_buff
*nskb
= skb
->next
;
2278 skb
->next
= nskb
->next
;
2282 * If device doesn't need nskb->dst, release it right now while
2283 * its hot in this cpu cache
2285 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2288 if (!list_empty(&ptype_all
))
2289 dev_queue_xmit_nit(nskb
, dev
);
2291 skb_len
= nskb
->len
;
2292 rc
= ops
->ndo_start_xmit(nskb
, dev
);
2293 trace_net_dev_xmit(nskb
, rc
, dev
, skb_len
);
2294 if (unlikely(rc
!= NETDEV_TX_OK
)) {
2295 if (rc
& ~NETDEV_TX_MASK
)
2296 goto out_kfree_gso_skb
;
2297 nskb
->next
= skb
->next
;
2301 txq_trans_update(txq
);
2302 if (unlikely(netif_xmit_stopped(txq
) && skb
->next
))
2303 return NETDEV_TX_BUSY
;
2304 } while (skb
->next
);
2307 if (likely(skb
->next
== NULL
))
2308 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
2315 static u32 hashrnd __read_mostly
;
2318 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2319 * to be used as a distribution range.
2321 u16
__skb_tx_hash(const struct net_device
*dev
, const struct sk_buff
*skb
,
2322 unsigned int num_tx_queues
)
2326 u16 qcount
= num_tx_queues
;
2328 if (skb_rx_queue_recorded(skb
)) {
2329 hash
= skb_get_rx_queue(skb
);
2330 while (unlikely(hash
>= num_tx_queues
))
2331 hash
-= num_tx_queues
;
2336 u8 tc
= netdev_get_prio_tc_map(dev
, skb
->priority
);
2337 qoffset
= dev
->tc_to_txq
[tc
].offset
;
2338 qcount
= dev
->tc_to_txq
[tc
].count
;
2341 if (skb
->sk
&& skb
->sk
->sk_hash
)
2342 hash
= skb
->sk
->sk_hash
;
2344 hash
= (__force u16
) skb
->protocol
;
2345 hash
= jhash_1word(hash
, hashrnd
);
2347 return (u16
) (((u64
) hash
* qcount
) >> 32) + qoffset
;
2349 EXPORT_SYMBOL(__skb_tx_hash
);
2351 static inline u16
dev_cap_txqueue(struct net_device
*dev
, u16 queue_index
)
2353 if (unlikely(queue_index
>= dev
->real_num_tx_queues
)) {
2354 net_warn_ratelimited("%s selects TX queue %d, but real number of TX queues is %d\n",
2355 dev
->name
, queue_index
,
2356 dev
->real_num_tx_queues
);
2362 static inline int get_xps_queue(struct net_device
*dev
, struct sk_buff
*skb
)
2365 struct xps_dev_maps
*dev_maps
;
2366 struct xps_map
*map
;
2367 int queue_index
= -1;
2370 dev_maps
= rcu_dereference(dev
->xps_maps
);
2372 map
= rcu_dereference(
2373 dev_maps
->cpu_map
[raw_smp_processor_id()]);
2376 queue_index
= map
->queues
[0];
2379 if (skb
->sk
&& skb
->sk
->sk_hash
)
2380 hash
= skb
->sk
->sk_hash
;
2382 hash
= (__force u16
) skb
->protocol
^
2384 hash
= jhash_1word(hash
, hashrnd
);
2385 queue_index
= map
->queues
[
2386 ((u64
)hash
* map
->len
) >> 32];
2388 if (unlikely(queue_index
>= dev
->real_num_tx_queues
))
2400 struct netdev_queue
*netdev_pick_tx(struct net_device
*dev
,
2401 struct sk_buff
*skb
)
2404 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2406 if (dev
->real_num_tx_queues
== 1)
2408 else if (ops
->ndo_select_queue
) {
2409 queue_index
= ops
->ndo_select_queue(dev
, skb
);
2410 queue_index
= dev_cap_txqueue(dev
, queue_index
);
2412 struct sock
*sk
= skb
->sk
;
2413 queue_index
= sk_tx_queue_get(sk
);
2415 if (queue_index
< 0 || skb
->ooo_okay
||
2416 queue_index
>= dev
->real_num_tx_queues
) {
2417 int old_index
= queue_index
;
2419 queue_index
= get_xps_queue(dev
, skb
);
2420 if (queue_index
< 0)
2421 queue_index
= skb_tx_hash(dev
, skb
);
2423 if (queue_index
!= old_index
&& sk
) {
2424 struct dst_entry
*dst
=
2425 rcu_dereference_check(sk
->sk_dst_cache
, 1);
2427 if (dst
&& skb_dst(skb
) == dst
)
2428 sk_tx_queue_set(sk
, queue_index
);
2433 skb_set_queue_mapping(skb
, queue_index
);
2434 return netdev_get_tx_queue(dev
, queue_index
);
2437 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2438 struct net_device
*dev
,
2439 struct netdev_queue
*txq
)
2441 spinlock_t
*root_lock
= qdisc_lock(q
);
2445 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
2446 qdisc_calculate_pkt_len(skb
, q
);
2448 * Heuristic to force contended enqueues to serialize on a
2449 * separate lock before trying to get qdisc main lock.
2450 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2451 * and dequeue packets faster.
2453 contended
= qdisc_is_running(q
);
2454 if (unlikely(contended
))
2455 spin_lock(&q
->busylock
);
2457 spin_lock(root_lock
);
2458 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2461 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2462 qdisc_run_begin(q
)) {
2464 * This is a work-conserving queue; there are no old skbs
2465 * waiting to be sent out; and the qdisc is not running -
2466 * xmit the skb directly.
2468 if (!(dev
->priv_flags
& IFF_XMIT_DST_RELEASE
))
2471 qdisc_bstats_update(q
, skb
);
2473 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
)) {
2474 if (unlikely(contended
)) {
2475 spin_unlock(&q
->busylock
);
2482 rc
= NET_XMIT_SUCCESS
;
2485 rc
= q
->enqueue(skb
, q
) & NET_XMIT_MASK
;
2486 if (qdisc_run_begin(q
)) {
2487 if (unlikely(contended
)) {
2488 spin_unlock(&q
->busylock
);
2494 spin_unlock(root_lock
);
2495 if (unlikely(contended
))
2496 spin_unlock(&q
->busylock
);
2500 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2501 static void skb_update_prio(struct sk_buff
*skb
)
2503 struct netprio_map
*map
= rcu_dereference_bh(skb
->dev
->priomap
);
2505 if (!skb
->priority
&& skb
->sk
&& map
) {
2506 unsigned int prioidx
= skb
->sk
->sk_cgrp_prioidx
;
2508 if (prioidx
< map
->priomap_len
)
2509 skb
->priority
= map
->priomap
[prioidx
];
2513 #define skb_update_prio(skb)
2516 static DEFINE_PER_CPU(int, xmit_recursion
);
2517 #define RECURSION_LIMIT 10
2520 * dev_loopback_xmit - loop back @skb
2521 * @skb: buffer to transmit
2523 int dev_loopback_xmit(struct sk_buff
*skb
)
2525 skb_reset_mac_header(skb
);
2526 __skb_pull(skb
, skb_network_offset(skb
));
2527 skb
->pkt_type
= PACKET_LOOPBACK
;
2528 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
2529 WARN_ON(!skb_dst(skb
));
2534 EXPORT_SYMBOL(dev_loopback_xmit
);
2537 * dev_queue_xmit - transmit a buffer
2538 * @skb: buffer to transmit
2540 * Queue a buffer for transmission to a network device. The caller must
2541 * have set the device and priority and built the buffer before calling
2542 * this function. The function can be called from an interrupt.
2544 * A negative errno code is returned on a failure. A success does not
2545 * guarantee the frame will be transmitted as it may be dropped due
2546 * to congestion or traffic shaping.
2548 * -----------------------------------------------------------------------------------
2549 * I notice this method can also return errors from the queue disciplines,
2550 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2553 * Regardless of the return value, the skb is consumed, so it is currently
2554 * difficult to retry a send to this method. (You can bump the ref count
2555 * before sending to hold a reference for retry if you are careful.)
2557 * When calling this method, interrupts MUST be enabled. This is because
2558 * the BH enable code must have IRQs enabled so that it will not deadlock.
2561 int dev_queue_xmit(struct sk_buff
*skb
)
2563 struct net_device
*dev
= skb
->dev
;
2564 struct netdev_queue
*txq
;
2568 /* Disable soft irqs for various locks below. Also
2569 * stops preemption for RCU.
2573 skb_update_prio(skb
);
2575 txq
= netdev_pick_tx(dev
, skb
);
2576 q
= rcu_dereference_bh(txq
->qdisc
);
2578 #ifdef CONFIG_NET_CLS_ACT
2579 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2581 trace_net_dev_queue(skb
);
2583 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2587 /* The device has no queue. Common case for software devices:
2588 loopback, all the sorts of tunnels...
2590 Really, it is unlikely that netif_tx_lock protection is necessary
2591 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2593 However, it is possible, that they rely on protection
2596 Check this and shot the lock. It is not prone from deadlocks.
2597 Either shot noqueue qdisc, it is even simpler 8)
2599 if (dev
->flags
& IFF_UP
) {
2600 int cpu
= smp_processor_id(); /* ok because BHs are off */
2602 if (txq
->xmit_lock_owner
!= cpu
) {
2604 if (__this_cpu_read(xmit_recursion
) > RECURSION_LIMIT
)
2605 goto recursion_alert
;
2607 HARD_TX_LOCK(dev
, txq
, cpu
);
2609 if (!netif_xmit_stopped(txq
)) {
2610 __this_cpu_inc(xmit_recursion
);
2611 rc
= dev_hard_start_xmit(skb
, dev
, txq
);
2612 __this_cpu_dec(xmit_recursion
);
2613 if (dev_xmit_complete(rc
)) {
2614 HARD_TX_UNLOCK(dev
, txq
);
2618 HARD_TX_UNLOCK(dev
, txq
);
2619 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2622 /* Recursion is detected! It is possible,
2626 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2632 rcu_read_unlock_bh();
2637 rcu_read_unlock_bh();
2640 EXPORT_SYMBOL(dev_queue_xmit
);
2643 /*=======================================================================
2645 =======================================================================*/
2647 int netdev_max_backlog __read_mostly
= 1000;
2648 int netdev_tstamp_prequeue __read_mostly
= 1;
2649 int netdev_budget __read_mostly
= 300;
2650 int weight_p __read_mostly
= 64; /* old backlog weight */
2652 /* Called with irq disabled */
2653 static inline void ____napi_schedule(struct softnet_data
*sd
,
2654 struct napi_struct
*napi
)
2656 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
2657 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2661 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2662 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2663 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2664 * if hash is a canonical 4-tuple hash over transport ports.
2666 void __skb_get_rxhash(struct sk_buff
*skb
)
2668 struct flow_keys keys
;
2671 if (!skb_flow_dissect(skb
, &keys
))
2677 /* get a consistent hash (same value on both flow directions) */
2678 if (((__force u32
)keys
.dst
< (__force u32
)keys
.src
) ||
2679 (((__force u32
)keys
.dst
== (__force u32
)keys
.src
) &&
2680 ((__force u16
)keys
.port16
[1] < (__force u16
)keys
.port16
[0]))) {
2681 swap(keys
.dst
, keys
.src
);
2682 swap(keys
.port16
[0], keys
.port16
[1]);
2685 hash
= jhash_3words((__force u32
)keys
.dst
,
2686 (__force u32
)keys
.src
,
2687 (__force u32
)keys
.ports
, hashrnd
);
2693 EXPORT_SYMBOL(__skb_get_rxhash
);
2697 /* One global table that all flow-based protocols share. */
2698 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
2699 EXPORT_SYMBOL(rps_sock_flow_table
);
2701 struct static_key rps_needed __read_mostly
;
2703 static struct rps_dev_flow
*
2704 set_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2705 struct rps_dev_flow
*rflow
, u16 next_cpu
)
2707 if (next_cpu
!= RPS_NO_CPU
) {
2708 #ifdef CONFIG_RFS_ACCEL
2709 struct netdev_rx_queue
*rxqueue
;
2710 struct rps_dev_flow_table
*flow_table
;
2711 struct rps_dev_flow
*old_rflow
;
2716 /* Should we steer this flow to a different hardware queue? */
2717 if (!skb_rx_queue_recorded(skb
) || !dev
->rx_cpu_rmap
||
2718 !(dev
->features
& NETIF_F_NTUPLE
))
2720 rxq_index
= cpu_rmap_lookup_index(dev
->rx_cpu_rmap
, next_cpu
);
2721 if (rxq_index
== skb_get_rx_queue(skb
))
2724 rxqueue
= dev
->_rx
+ rxq_index
;
2725 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2728 flow_id
= skb
->rxhash
& flow_table
->mask
;
2729 rc
= dev
->netdev_ops
->ndo_rx_flow_steer(dev
, skb
,
2730 rxq_index
, flow_id
);
2734 rflow
= &flow_table
->flows
[flow_id
];
2736 if (old_rflow
->filter
== rflow
->filter
)
2737 old_rflow
->filter
= RPS_NO_FILTER
;
2741 per_cpu(softnet_data
, next_cpu
).input_queue_head
;
2744 rflow
->cpu
= next_cpu
;
2749 * get_rps_cpu is called from netif_receive_skb and returns the target
2750 * CPU from the RPS map of the receiving queue for a given skb.
2751 * rcu_read_lock must be held on entry.
2753 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2754 struct rps_dev_flow
**rflowp
)
2756 struct netdev_rx_queue
*rxqueue
;
2757 struct rps_map
*map
;
2758 struct rps_dev_flow_table
*flow_table
;
2759 struct rps_sock_flow_table
*sock_flow_table
;
2763 if (skb_rx_queue_recorded(skb
)) {
2764 u16 index
= skb_get_rx_queue(skb
);
2765 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
2766 WARN_ONCE(dev
->real_num_rx_queues
> 1,
2767 "%s received packet on queue %u, but number "
2768 "of RX queues is %u\n",
2769 dev
->name
, index
, dev
->real_num_rx_queues
);
2772 rxqueue
= dev
->_rx
+ index
;
2776 map
= rcu_dereference(rxqueue
->rps_map
);
2778 if (map
->len
== 1 &&
2779 !rcu_access_pointer(rxqueue
->rps_flow_table
)) {
2780 tcpu
= map
->cpus
[0];
2781 if (cpu_online(tcpu
))
2785 } else if (!rcu_access_pointer(rxqueue
->rps_flow_table
)) {
2789 skb_reset_network_header(skb
);
2790 if (!skb_get_rxhash(skb
))
2793 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2794 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
2795 if (flow_table
&& sock_flow_table
) {
2797 struct rps_dev_flow
*rflow
;
2799 rflow
= &flow_table
->flows
[skb
->rxhash
& flow_table
->mask
];
2802 next_cpu
= sock_flow_table
->ents
[skb
->rxhash
&
2803 sock_flow_table
->mask
];
2806 * If the desired CPU (where last recvmsg was done) is
2807 * different from current CPU (one in the rx-queue flow
2808 * table entry), switch if one of the following holds:
2809 * - Current CPU is unset (equal to RPS_NO_CPU).
2810 * - Current CPU is offline.
2811 * - The current CPU's queue tail has advanced beyond the
2812 * last packet that was enqueued using this table entry.
2813 * This guarantees that all previous packets for the flow
2814 * have been dequeued, thus preserving in order delivery.
2816 if (unlikely(tcpu
!= next_cpu
) &&
2817 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
2818 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
2819 rflow
->last_qtail
)) >= 0))
2820 rflow
= set_rps_cpu(dev
, skb
, rflow
, next_cpu
);
2822 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
2830 tcpu
= map
->cpus
[((u64
) skb
->rxhash
* map
->len
) >> 32];
2832 if (cpu_online(tcpu
)) {
2842 #ifdef CONFIG_RFS_ACCEL
2845 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2846 * @dev: Device on which the filter was set
2847 * @rxq_index: RX queue index
2848 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2849 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2851 * Drivers that implement ndo_rx_flow_steer() should periodically call
2852 * this function for each installed filter and remove the filters for
2853 * which it returns %true.
2855 bool rps_may_expire_flow(struct net_device
*dev
, u16 rxq_index
,
2856 u32 flow_id
, u16 filter_id
)
2858 struct netdev_rx_queue
*rxqueue
= dev
->_rx
+ rxq_index
;
2859 struct rps_dev_flow_table
*flow_table
;
2860 struct rps_dev_flow
*rflow
;
2865 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2866 if (flow_table
&& flow_id
<= flow_table
->mask
) {
2867 rflow
= &flow_table
->flows
[flow_id
];
2868 cpu
= ACCESS_ONCE(rflow
->cpu
);
2869 if (rflow
->filter
== filter_id
&& cpu
!= RPS_NO_CPU
&&
2870 ((int)(per_cpu(softnet_data
, cpu
).input_queue_head
-
2871 rflow
->last_qtail
) <
2872 (int)(10 * flow_table
->mask
)))
2878 EXPORT_SYMBOL(rps_may_expire_flow
);
2880 #endif /* CONFIG_RFS_ACCEL */
2882 /* Called from hardirq (IPI) context */
2883 static void rps_trigger_softirq(void *data
)
2885 struct softnet_data
*sd
= data
;
2887 ____napi_schedule(sd
, &sd
->backlog
);
2891 #endif /* CONFIG_RPS */
2894 * Check if this softnet_data structure is another cpu one
2895 * If yes, queue it to our IPI list and return 1
2898 static int rps_ipi_queued(struct softnet_data
*sd
)
2901 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
2904 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
2905 mysd
->rps_ipi_list
= sd
;
2907 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2910 #endif /* CONFIG_RPS */
2915 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2916 * queue (may be a remote CPU queue).
2918 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
2919 unsigned int *qtail
)
2921 struct softnet_data
*sd
;
2922 unsigned long flags
;
2924 sd
= &per_cpu(softnet_data
, cpu
);
2926 local_irq_save(flags
);
2929 if (skb_queue_len(&sd
->input_pkt_queue
) <= netdev_max_backlog
) {
2930 if (skb_queue_len(&sd
->input_pkt_queue
)) {
2932 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
2933 input_queue_tail_incr_save(sd
, qtail
);
2935 local_irq_restore(flags
);
2936 return NET_RX_SUCCESS
;
2939 /* Schedule NAPI for backlog device
2940 * We can use non atomic operation since we own the queue lock
2942 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
2943 if (!rps_ipi_queued(sd
))
2944 ____napi_schedule(sd
, &sd
->backlog
);
2952 local_irq_restore(flags
);
2954 atomic_long_inc(&skb
->dev
->rx_dropped
);
2960 * netif_rx - post buffer to the network code
2961 * @skb: buffer to post
2963 * This function receives a packet from a device driver and queues it for
2964 * the upper (protocol) levels to process. It always succeeds. The buffer
2965 * may be dropped during processing for congestion control or by the
2969 * NET_RX_SUCCESS (no congestion)
2970 * NET_RX_DROP (packet was dropped)
2974 int netif_rx(struct sk_buff
*skb
)
2978 /* if netpoll wants it, pretend we never saw it */
2979 if (netpoll_rx(skb
))
2982 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
2984 trace_netif_rx(skb
);
2986 if (static_key_false(&rps_needed
)) {
2987 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
2993 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
2995 cpu
= smp_processor_id();
2997 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3005 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
3010 EXPORT_SYMBOL(netif_rx
);
3012 int netif_rx_ni(struct sk_buff
*skb
)
3017 err
= netif_rx(skb
);
3018 if (local_softirq_pending())
3024 EXPORT_SYMBOL(netif_rx_ni
);
3026 static void net_tx_action(struct softirq_action
*h
)
3028 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3030 if (sd
->completion_queue
) {
3031 struct sk_buff
*clist
;
3033 local_irq_disable();
3034 clist
= sd
->completion_queue
;
3035 sd
->completion_queue
= NULL
;
3039 struct sk_buff
*skb
= clist
;
3040 clist
= clist
->next
;
3042 WARN_ON(atomic_read(&skb
->users
));
3043 trace_kfree_skb(skb
, net_tx_action
);
3048 if (sd
->output_queue
) {
3051 local_irq_disable();
3052 head
= sd
->output_queue
;
3053 sd
->output_queue
= NULL
;
3054 sd
->output_queue_tailp
= &sd
->output_queue
;
3058 struct Qdisc
*q
= head
;
3059 spinlock_t
*root_lock
;
3061 head
= head
->next_sched
;
3063 root_lock
= qdisc_lock(q
);
3064 if (spin_trylock(root_lock
)) {
3065 smp_mb__before_clear_bit();
3066 clear_bit(__QDISC_STATE_SCHED
,
3069 spin_unlock(root_lock
);
3071 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
3073 __netif_reschedule(q
);
3075 smp_mb__before_clear_bit();
3076 clear_bit(__QDISC_STATE_SCHED
,
3084 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3085 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3086 /* This hook is defined here for ATM LANE */
3087 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
3088 unsigned char *addr
) __read_mostly
;
3089 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
3092 #ifdef CONFIG_NET_CLS_ACT
3093 /* TODO: Maybe we should just force sch_ingress to be compiled in
3094 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3095 * a compare and 2 stores extra right now if we dont have it on
3096 * but have CONFIG_NET_CLS_ACT
3097 * NOTE: This doesn't stop any functionality; if you dont have
3098 * the ingress scheduler, you just can't add policies on ingress.
3101 static int ing_filter(struct sk_buff
*skb
, struct netdev_queue
*rxq
)
3103 struct net_device
*dev
= skb
->dev
;
3104 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
3105 int result
= TC_ACT_OK
;
3108 if (unlikely(MAX_RED_LOOP
< ttl
++)) {
3109 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3110 skb
->skb_iif
, dev
->ifindex
);
3114 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
3115 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
3118 if (q
!= &noop_qdisc
) {
3119 spin_lock(qdisc_lock(q
));
3120 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
3121 result
= qdisc_enqueue_root(skb
, q
);
3122 spin_unlock(qdisc_lock(q
));
3128 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
3129 struct packet_type
**pt_prev
,
3130 int *ret
, struct net_device
*orig_dev
)
3132 struct netdev_queue
*rxq
= rcu_dereference(skb
->dev
->ingress_queue
);
3134 if (!rxq
|| rxq
->qdisc
== &noop_qdisc
)
3138 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
3142 switch (ing_filter(skb
, rxq
)) {
3156 * netdev_rx_handler_register - register receive handler
3157 * @dev: device to register a handler for
3158 * @rx_handler: receive handler to register
3159 * @rx_handler_data: data pointer that is used by rx handler
3161 * Register a receive hander for a device. This handler will then be
3162 * called from __netif_receive_skb. A negative errno code is returned
3165 * The caller must hold the rtnl_mutex.
3167 * For a general description of rx_handler, see enum rx_handler_result.
3169 int netdev_rx_handler_register(struct net_device
*dev
,
3170 rx_handler_func_t
*rx_handler
,
3171 void *rx_handler_data
)
3175 if (dev
->rx_handler
)
3178 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
3179 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
3183 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
3186 * netdev_rx_handler_unregister - unregister receive handler
3187 * @dev: device to unregister a handler from
3189 * Unregister a receive hander from a device.
3191 * The caller must hold the rtnl_mutex.
3193 void netdev_rx_handler_unregister(struct net_device
*dev
)
3197 RCU_INIT_POINTER(dev
->rx_handler
, NULL
);
3198 RCU_INIT_POINTER(dev
->rx_handler_data
, NULL
);
3200 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
3203 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3204 * the special handling of PFMEMALLOC skbs.
3206 static bool skb_pfmemalloc_protocol(struct sk_buff
*skb
)
3208 switch (skb
->protocol
) {
3209 case __constant_htons(ETH_P_ARP
):
3210 case __constant_htons(ETH_P_IP
):
3211 case __constant_htons(ETH_P_IPV6
):
3212 case __constant_htons(ETH_P_8021Q
):
3219 static int __netif_receive_skb(struct sk_buff
*skb
)
3221 struct packet_type
*ptype
, *pt_prev
;
3222 rx_handler_func_t
*rx_handler
;
3223 struct net_device
*orig_dev
;
3224 struct net_device
*null_or_dev
;
3225 bool deliver_exact
= false;
3226 int ret
= NET_RX_DROP
;
3228 unsigned long pflags
= current
->flags
;
3230 net_timestamp_check(!netdev_tstamp_prequeue
, skb
);
3232 trace_netif_receive_skb(skb
);
3235 * PFMEMALLOC skbs are special, they should
3236 * - be delivered to SOCK_MEMALLOC sockets only
3237 * - stay away from userspace
3238 * - have bounded memory usage
3240 * Use PF_MEMALLOC as this saves us from propagating the allocation
3241 * context down to all allocation sites.
3243 if (sk_memalloc_socks() && skb_pfmemalloc(skb
))
3244 current
->flags
|= PF_MEMALLOC
;
3246 /* if we've gotten here through NAPI, check netpoll */
3247 if (netpoll_receive_skb(skb
))
3250 orig_dev
= skb
->dev
;
3252 skb_reset_network_header(skb
);
3253 skb_reset_transport_header(skb
);
3254 skb_reset_mac_len(skb
);
3261 skb
->skb_iif
= skb
->dev
->ifindex
;
3263 __this_cpu_inc(softnet_data
.processed
);
3265 if (skb
->protocol
== cpu_to_be16(ETH_P_8021Q
)) {
3266 skb
= vlan_untag(skb
);
3271 #ifdef CONFIG_NET_CLS_ACT
3272 if (skb
->tc_verd
& TC_NCLS
) {
3273 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
3278 if (sk_memalloc_socks() && skb_pfmemalloc(skb
))
3281 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
3282 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
3284 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3290 #ifdef CONFIG_NET_CLS_ACT
3291 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
3297 if (sk_memalloc_socks() && skb_pfmemalloc(skb
)
3298 && !skb_pfmemalloc_protocol(skb
))
3301 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
3302 if (vlan_tx_tag_present(skb
)) {
3304 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3307 if (vlan_do_receive(&skb
, !rx_handler
))
3309 else if (unlikely(!skb
))
3315 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3318 switch (rx_handler(&skb
)) {
3319 case RX_HANDLER_CONSUMED
:
3321 case RX_HANDLER_ANOTHER
:
3323 case RX_HANDLER_EXACT
:
3324 deliver_exact
= true;
3325 case RX_HANDLER_PASS
:
3332 /* deliver only exact match when indicated */
3333 null_or_dev
= deliver_exact
? skb
->dev
: NULL
;
3335 type
= skb
->protocol
;
3336 list_for_each_entry_rcu(ptype
,
3337 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
3338 if (ptype
->type
== type
&&
3339 (ptype
->dev
== null_or_dev
|| ptype
->dev
== skb
->dev
||
3340 ptype
->dev
== orig_dev
)) {
3342 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3348 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
3351 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
3354 atomic_long_inc(&skb
->dev
->rx_dropped
);
3356 /* Jamal, now you will not able to escape explaining
3357 * me how you were going to use this. :-)
3365 tsk_restore_flags(current
, pflags
, PF_MEMALLOC
);
3370 * netif_receive_skb - process receive buffer from network
3371 * @skb: buffer to process
3373 * netif_receive_skb() is the main receive data processing function.
3374 * It always succeeds. The buffer may be dropped during processing
3375 * for congestion control or by the protocol layers.
3377 * This function may only be called from softirq context and interrupts
3378 * should be enabled.
3380 * Return values (usually ignored):
3381 * NET_RX_SUCCESS: no congestion
3382 * NET_RX_DROP: packet was dropped
3384 int netif_receive_skb(struct sk_buff
*skb
)
3386 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3388 if (skb_defer_rx_timestamp(skb
))
3389 return NET_RX_SUCCESS
;
3392 if (static_key_false(&rps_needed
)) {
3393 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3398 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3401 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3408 return __netif_receive_skb(skb
);
3410 EXPORT_SYMBOL(netif_receive_skb
);
3412 /* Network device is going away, flush any packets still pending
3413 * Called with irqs disabled.
3415 static void flush_backlog(void *arg
)
3417 struct net_device
*dev
= arg
;
3418 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3419 struct sk_buff
*skb
, *tmp
;
3422 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3423 if (skb
->dev
== dev
) {
3424 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3426 input_queue_head_incr(sd
);
3431 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3432 if (skb
->dev
== dev
) {
3433 __skb_unlink(skb
, &sd
->process_queue
);
3435 input_queue_head_incr(sd
);
3440 static int napi_gro_complete(struct sk_buff
*skb
)
3442 struct packet_type
*ptype
;
3443 __be16 type
= skb
->protocol
;
3444 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3447 if (NAPI_GRO_CB(skb
)->count
== 1) {
3448 skb_shinfo(skb
)->gso_size
= 0;
3453 list_for_each_entry_rcu(ptype
, head
, list
) {
3454 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_complete
)
3457 err
= ptype
->gro_complete(skb
);
3463 WARN_ON(&ptype
->list
== head
);
3465 return NET_RX_SUCCESS
;
3469 return netif_receive_skb(skb
);
3472 inline void napi_gro_flush(struct napi_struct
*napi
)
3474 struct sk_buff
*skb
, *next
;
3476 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3479 napi_gro_complete(skb
);
3482 napi
->gro_count
= 0;
3483 napi
->gro_list
= NULL
;
3485 EXPORT_SYMBOL(napi_gro_flush
);
3487 enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3489 struct sk_buff
**pp
= NULL
;
3490 struct packet_type
*ptype
;
3491 __be16 type
= skb
->protocol
;
3492 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3495 enum gro_result ret
;
3497 if (!(skb
->dev
->features
& NETIF_F_GRO
) || netpoll_rx_on(skb
))
3500 if (skb_is_gso(skb
) || skb_has_frag_list(skb
))
3504 list_for_each_entry_rcu(ptype
, head
, list
) {
3505 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_receive
)
3508 skb_set_network_header(skb
, skb_gro_offset(skb
));
3509 mac_len
= skb
->network_header
- skb
->mac_header
;
3510 skb
->mac_len
= mac_len
;
3511 NAPI_GRO_CB(skb
)->same_flow
= 0;
3512 NAPI_GRO_CB(skb
)->flush
= 0;
3513 NAPI_GRO_CB(skb
)->free
= 0;
3515 pp
= ptype
->gro_receive(&napi
->gro_list
, skb
);
3520 if (&ptype
->list
== head
)
3523 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3524 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3527 struct sk_buff
*nskb
= *pp
;
3531 napi_gro_complete(nskb
);
3538 if (NAPI_GRO_CB(skb
)->flush
|| napi
->gro_count
>= MAX_GRO_SKBS
)
3542 NAPI_GRO_CB(skb
)->count
= 1;
3543 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
3544 skb
->next
= napi
->gro_list
;
3545 napi
->gro_list
= skb
;
3549 if (skb_headlen(skb
) < skb_gro_offset(skb
)) {
3550 int grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
3552 BUG_ON(skb
->end
- skb
->tail
< grow
);
3554 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3557 skb
->data_len
-= grow
;
3559 skb_shinfo(skb
)->frags
[0].page_offset
+= grow
;
3560 skb_frag_size_sub(&skb_shinfo(skb
)->frags
[0], grow
);
3562 if (unlikely(!skb_frag_size(&skb_shinfo(skb
)->frags
[0]))) {
3563 skb_frag_unref(skb
, 0);
3564 memmove(skb_shinfo(skb
)->frags
,
3565 skb_shinfo(skb
)->frags
+ 1,
3566 --skb_shinfo(skb
)->nr_frags
* sizeof(skb_frag_t
));
3577 EXPORT_SYMBOL(dev_gro_receive
);
3579 static inline gro_result_t
3580 __napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3583 unsigned int maclen
= skb
->dev
->hard_header_len
;
3585 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3586 unsigned long diffs
;
3588 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
3589 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
3590 if (maclen
== ETH_HLEN
)
3591 diffs
|= compare_ether_header(skb_mac_header(p
),
3592 skb_gro_mac_header(skb
));
3594 diffs
= memcmp(skb_mac_header(p
),
3595 skb_gro_mac_header(skb
),
3597 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
3598 NAPI_GRO_CB(p
)->flush
= 0;
3601 return dev_gro_receive(napi
, skb
);
3604 gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
3608 if (netif_receive_skb(skb
))
3616 case GRO_MERGED_FREE
:
3617 if (NAPI_GRO_CB(skb
)->free
== NAPI_GRO_FREE_STOLEN_HEAD
)
3618 kmem_cache_free(skbuff_head_cache
, skb
);
3630 EXPORT_SYMBOL(napi_skb_finish
);
3632 void skb_gro_reset_offset(struct sk_buff
*skb
)
3634 NAPI_GRO_CB(skb
)->data_offset
= 0;
3635 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3636 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3638 if (skb
->mac_header
== skb
->tail
&&
3639 !PageHighMem(skb_frag_page(&skb_shinfo(skb
)->frags
[0]))) {
3640 NAPI_GRO_CB(skb
)->frag0
=
3641 skb_frag_address(&skb_shinfo(skb
)->frags
[0]);
3642 NAPI_GRO_CB(skb
)->frag0_len
= skb_frag_size(&skb_shinfo(skb
)->frags
[0]);
3645 EXPORT_SYMBOL(skb_gro_reset_offset
);
3647 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3649 skb_gro_reset_offset(skb
);
3651 return napi_skb_finish(__napi_gro_receive(napi
, skb
), skb
);
3653 EXPORT_SYMBOL(napi_gro_receive
);
3655 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
3657 __skb_pull(skb
, skb_headlen(skb
));
3658 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3659 skb_reserve(skb
, NET_SKB_PAD
+ NET_IP_ALIGN
- skb_headroom(skb
));
3661 skb
->dev
= napi
->dev
;
3667 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
3669 struct sk_buff
*skb
= napi
->skb
;
3672 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
3678 EXPORT_SYMBOL(napi_get_frags
);
3680 gro_result_t
napi_frags_finish(struct napi_struct
*napi
, struct sk_buff
*skb
,
3686 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
3688 if (ret
== GRO_HELD
)
3689 skb_gro_pull(skb
, -ETH_HLEN
);
3690 else if (netif_receive_skb(skb
))
3695 case GRO_MERGED_FREE
:
3696 napi_reuse_skb(napi
, skb
);
3705 EXPORT_SYMBOL(napi_frags_finish
);
3707 static struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
3709 struct sk_buff
*skb
= napi
->skb
;
3716 skb_reset_mac_header(skb
);
3717 skb_gro_reset_offset(skb
);
3719 off
= skb_gro_offset(skb
);
3720 hlen
= off
+ sizeof(*eth
);
3721 eth
= skb_gro_header_fast(skb
, off
);
3722 if (skb_gro_header_hard(skb
, hlen
)) {
3723 eth
= skb_gro_header_slow(skb
, hlen
, off
);
3724 if (unlikely(!eth
)) {
3725 napi_reuse_skb(napi
, skb
);
3731 skb_gro_pull(skb
, sizeof(*eth
));
3734 * This works because the only protocols we care about don't require
3735 * special handling. We'll fix it up properly at the end.
3737 skb
->protocol
= eth
->h_proto
;
3743 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
3745 struct sk_buff
*skb
= napi_frags_skb(napi
);
3750 return napi_frags_finish(napi
, skb
, __napi_gro_receive(napi
, skb
));
3752 EXPORT_SYMBOL(napi_gro_frags
);
3755 * net_rps_action sends any pending IPI's for rps.
3756 * Note: called with local irq disabled, but exits with local irq enabled.
3758 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
3761 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
3764 sd
->rps_ipi_list
= NULL
;
3768 /* Send pending IPI's to kick RPS processing on remote cpus. */
3770 struct softnet_data
*next
= remsd
->rps_ipi_next
;
3772 if (cpu_online(remsd
->cpu
))
3773 __smp_call_function_single(remsd
->cpu
,
3782 static int process_backlog(struct napi_struct
*napi
, int quota
)
3785 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
3788 /* Check if we have pending ipi, its better to send them now,
3789 * not waiting net_rx_action() end.
3791 if (sd
->rps_ipi_list
) {
3792 local_irq_disable();
3793 net_rps_action_and_irq_enable(sd
);
3796 napi
->weight
= weight_p
;
3797 local_irq_disable();
3798 while (work
< quota
) {
3799 struct sk_buff
*skb
;
3802 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
3804 __netif_receive_skb(skb
);
3805 local_irq_disable();
3806 input_queue_head_incr(sd
);
3807 if (++work
>= quota
) {
3814 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3816 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
3817 &sd
->process_queue
);
3819 if (qlen
< quota
- work
) {
3821 * Inline a custom version of __napi_complete().
3822 * only current cpu owns and manipulates this napi,
3823 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3824 * we can use a plain write instead of clear_bit(),
3825 * and we dont need an smp_mb() memory barrier.
3827 list_del(&napi
->poll_list
);
3830 quota
= work
+ qlen
;
3840 * __napi_schedule - schedule for receive
3841 * @n: entry to schedule
3843 * The entry's receive function will be scheduled to run
3845 void __napi_schedule(struct napi_struct
*n
)
3847 unsigned long flags
;
3849 local_irq_save(flags
);
3850 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
3851 local_irq_restore(flags
);
3853 EXPORT_SYMBOL(__napi_schedule
);
3855 void __napi_complete(struct napi_struct
*n
)
3857 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
3858 BUG_ON(n
->gro_list
);
3860 list_del(&n
->poll_list
);
3861 smp_mb__before_clear_bit();
3862 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
3864 EXPORT_SYMBOL(__napi_complete
);
3866 void napi_complete(struct napi_struct
*n
)
3868 unsigned long flags
;
3871 * don't let napi dequeue from the cpu poll list
3872 * just in case its running on a different cpu
3874 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
3878 local_irq_save(flags
);
3880 local_irq_restore(flags
);
3882 EXPORT_SYMBOL(napi_complete
);
3884 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
3885 int (*poll
)(struct napi_struct
*, int), int weight
)
3887 INIT_LIST_HEAD(&napi
->poll_list
);
3888 napi
->gro_count
= 0;
3889 napi
->gro_list
= NULL
;
3892 napi
->weight
= weight
;
3893 list_add(&napi
->dev_list
, &dev
->napi_list
);
3895 #ifdef CONFIG_NETPOLL
3896 spin_lock_init(&napi
->poll_lock
);
3897 napi
->poll_owner
= -1;
3899 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
3901 EXPORT_SYMBOL(netif_napi_add
);
3903 void netif_napi_del(struct napi_struct
*napi
)
3905 struct sk_buff
*skb
, *next
;
3907 list_del_init(&napi
->dev_list
);
3908 napi_free_frags(napi
);
3910 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3916 napi
->gro_list
= NULL
;
3917 napi
->gro_count
= 0;
3919 EXPORT_SYMBOL(netif_napi_del
);
3921 static void net_rx_action(struct softirq_action
*h
)
3923 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3924 unsigned long time_limit
= jiffies
+ 2;
3925 int budget
= netdev_budget
;
3928 local_irq_disable();
3930 while (!list_empty(&sd
->poll_list
)) {
3931 struct napi_struct
*n
;
3934 /* If softirq window is exhuasted then punt.
3935 * Allow this to run for 2 jiffies since which will allow
3936 * an average latency of 1.5/HZ.
3938 if (unlikely(budget
<= 0 || time_after(jiffies
, time_limit
)))
3943 /* Even though interrupts have been re-enabled, this
3944 * access is safe because interrupts can only add new
3945 * entries to the tail of this list, and only ->poll()
3946 * calls can remove this head entry from the list.
3948 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
3950 have
= netpoll_poll_lock(n
);
3954 /* This NAPI_STATE_SCHED test is for avoiding a race
3955 * with netpoll's poll_napi(). Only the entity which
3956 * obtains the lock and sees NAPI_STATE_SCHED set will
3957 * actually make the ->poll() call. Therefore we avoid
3958 * accidentally calling ->poll() when NAPI is not scheduled.
3961 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
3962 work
= n
->poll(n
, weight
);
3966 WARN_ON_ONCE(work
> weight
);
3970 local_irq_disable();
3972 /* Drivers must not modify the NAPI state if they
3973 * consume the entire weight. In such cases this code
3974 * still "owns" the NAPI instance and therefore can
3975 * move the instance around on the list at-will.
3977 if (unlikely(work
== weight
)) {
3978 if (unlikely(napi_disable_pending(n
))) {
3981 local_irq_disable();
3983 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
3986 netpoll_poll_unlock(have
);
3989 net_rps_action_and_irq_enable(sd
);
3991 #ifdef CONFIG_NET_DMA
3993 * There may not be any more sk_buffs coming right now, so push
3994 * any pending DMA copies to hardware
3996 dma_issue_pending_all();
4003 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
4007 static gifconf_func_t
*gifconf_list
[NPROTO
];
4010 * register_gifconf - register a SIOCGIF handler
4011 * @family: Address family
4012 * @gifconf: Function handler
4014 * Register protocol dependent address dumping routines. The handler
4015 * that is passed must not be freed or reused until it has been replaced
4016 * by another handler.
4018 int register_gifconf(unsigned int family
, gifconf_func_t
*gifconf
)
4020 if (family
>= NPROTO
)
4022 gifconf_list
[family
] = gifconf
;
4025 EXPORT_SYMBOL(register_gifconf
);
4029 * Map an interface index to its name (SIOCGIFNAME)
4033 * We need this ioctl for efficient implementation of the
4034 * if_indextoname() function required by the IPv6 API. Without
4035 * it, we would have to search all the interfaces to find a
4039 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
4041 struct net_device
*dev
;
4045 * Fetch the caller's info block.
4048 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
4052 dev
= dev_get_by_index_rcu(net
, ifr
.ifr_ifindex
);
4058 strcpy(ifr
.ifr_name
, dev
->name
);
4061 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
4067 * Perform a SIOCGIFCONF call. This structure will change
4068 * size eventually, and there is nothing I can do about it.
4069 * Thus we will need a 'compatibility mode'.
4072 static int dev_ifconf(struct net
*net
, char __user
*arg
)
4075 struct net_device
*dev
;
4082 * Fetch the caller's info block.
4085 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
4092 * Loop over the interfaces, and write an info block for each.
4096 for_each_netdev(net
, dev
) {
4097 for (i
= 0; i
< NPROTO
; i
++) {
4098 if (gifconf_list
[i
]) {
4101 done
= gifconf_list
[i
](dev
, NULL
, 0);
4103 done
= gifconf_list
[i
](dev
, pos
+ total
,
4113 * All done. Write the updated control block back to the caller.
4115 ifc
.ifc_len
= total
;
4118 * Both BSD and Solaris return 0 here, so we do too.
4120 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
4123 #ifdef CONFIG_PROC_FS
4125 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4127 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4128 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4129 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4131 static inline struct net_device
*dev_from_same_bucket(struct seq_file
*seq
, loff_t
*pos
)
4133 struct net
*net
= seq_file_net(seq
);
4134 struct net_device
*dev
;
4135 struct hlist_node
*p
;
4136 struct hlist_head
*h
;
4137 unsigned int count
= 0, offset
= get_offset(*pos
);
4139 h
= &net
->dev_name_head
[get_bucket(*pos
)];
4140 hlist_for_each_entry_rcu(dev
, p
, h
, name_hlist
) {
4141 if (++count
== offset
)
4148 static inline struct net_device
*dev_from_bucket(struct seq_file
*seq
, loff_t
*pos
)
4150 struct net_device
*dev
;
4151 unsigned int bucket
;
4154 dev
= dev_from_same_bucket(seq
, pos
);
4158 bucket
= get_bucket(*pos
) + 1;
4159 *pos
= set_bucket_offset(bucket
, 1);
4160 } while (bucket
< NETDEV_HASHENTRIES
);
4166 * This is invoked by the /proc filesystem handler to display a device
4169 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4174 return SEQ_START_TOKEN
;
4176 if (get_bucket(*pos
) >= NETDEV_HASHENTRIES
)
4179 return dev_from_bucket(seq
, pos
);
4182 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4185 return dev_from_bucket(seq
, pos
);
4188 void dev_seq_stop(struct seq_file
*seq
, void *v
)
4194 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
4196 struct rtnl_link_stats64 temp
;
4197 const struct rtnl_link_stats64
*stats
= dev_get_stats(dev
, &temp
);
4199 seq_printf(seq
, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4200 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4201 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
4203 stats
->rx_dropped
+ stats
->rx_missed_errors
,
4204 stats
->rx_fifo_errors
,
4205 stats
->rx_length_errors
+ stats
->rx_over_errors
+
4206 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
4207 stats
->rx_compressed
, stats
->multicast
,
4208 stats
->tx_bytes
, stats
->tx_packets
,
4209 stats
->tx_errors
, stats
->tx_dropped
,
4210 stats
->tx_fifo_errors
, stats
->collisions
,
4211 stats
->tx_carrier_errors
+
4212 stats
->tx_aborted_errors
+
4213 stats
->tx_window_errors
+
4214 stats
->tx_heartbeat_errors
,
4215 stats
->tx_compressed
);
4219 * Called from the PROCfs module. This now uses the new arbitrary sized
4220 * /proc/net interface to create /proc/net/dev
4222 static int dev_seq_show(struct seq_file
*seq
, void *v
)
4224 if (v
== SEQ_START_TOKEN
)
4225 seq_puts(seq
, "Inter-| Receive "
4227 " face |bytes packets errs drop fifo frame "
4228 "compressed multicast|bytes packets errs "
4229 "drop fifo colls carrier compressed\n");
4231 dev_seq_printf_stats(seq
, v
);
4235 static struct softnet_data
*softnet_get_online(loff_t
*pos
)
4237 struct softnet_data
*sd
= NULL
;
4239 while (*pos
< nr_cpu_ids
)
4240 if (cpu_online(*pos
)) {
4241 sd
= &per_cpu(softnet_data
, *pos
);
4248 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4250 return softnet_get_online(pos
);
4253 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4256 return softnet_get_online(pos
);
4259 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
4263 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
4265 struct softnet_data
*sd
= v
;
4267 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4268 sd
->processed
, sd
->dropped
, sd
->time_squeeze
, 0,
4269 0, 0, 0, 0, /* was fastroute */
4270 sd
->cpu_collision
, sd
->received_rps
);
4274 static const struct seq_operations dev_seq_ops
= {
4275 .start
= dev_seq_start
,
4276 .next
= dev_seq_next
,
4277 .stop
= dev_seq_stop
,
4278 .show
= dev_seq_show
,
4281 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
4283 return seq_open_net(inode
, file
, &dev_seq_ops
,
4284 sizeof(struct seq_net_private
));
4287 static const struct file_operations dev_seq_fops
= {
4288 .owner
= THIS_MODULE
,
4289 .open
= dev_seq_open
,
4291 .llseek
= seq_lseek
,
4292 .release
= seq_release_net
,
4295 static const struct seq_operations softnet_seq_ops
= {
4296 .start
= softnet_seq_start
,
4297 .next
= softnet_seq_next
,
4298 .stop
= softnet_seq_stop
,
4299 .show
= softnet_seq_show
,
4302 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
4304 return seq_open(file
, &softnet_seq_ops
);
4307 static const struct file_operations softnet_seq_fops
= {
4308 .owner
= THIS_MODULE
,
4309 .open
= softnet_seq_open
,
4311 .llseek
= seq_lseek
,
4312 .release
= seq_release
,
4315 static void *ptype_get_idx(loff_t pos
)
4317 struct packet_type
*pt
= NULL
;
4321 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
4327 for (t
= 0; t
< PTYPE_HASH_SIZE
; t
++) {
4328 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
4337 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4341 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
4344 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4346 struct packet_type
*pt
;
4347 struct list_head
*nxt
;
4351 if (v
== SEQ_START_TOKEN
)
4352 return ptype_get_idx(0);
4355 nxt
= pt
->list
.next
;
4356 if (pt
->type
== htons(ETH_P_ALL
)) {
4357 if (nxt
!= &ptype_all
)
4360 nxt
= ptype_base
[0].next
;
4362 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
4364 while (nxt
== &ptype_base
[hash
]) {
4365 if (++hash
>= PTYPE_HASH_SIZE
)
4367 nxt
= ptype_base
[hash
].next
;
4370 return list_entry(nxt
, struct packet_type
, list
);
4373 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
4379 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
4381 struct packet_type
*pt
= v
;
4383 if (v
== SEQ_START_TOKEN
)
4384 seq_puts(seq
, "Type Device Function\n");
4385 else if (pt
->dev
== NULL
|| dev_net(pt
->dev
) == seq_file_net(seq
)) {
4386 if (pt
->type
== htons(ETH_P_ALL
))
4387 seq_puts(seq
, "ALL ");
4389 seq_printf(seq
, "%04x", ntohs(pt
->type
));
4391 seq_printf(seq
, " %-8s %pF\n",
4392 pt
->dev
? pt
->dev
->name
: "", pt
->func
);
4398 static const struct seq_operations ptype_seq_ops
= {
4399 .start
= ptype_seq_start
,
4400 .next
= ptype_seq_next
,
4401 .stop
= ptype_seq_stop
,
4402 .show
= ptype_seq_show
,
4405 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
4407 return seq_open_net(inode
, file
, &ptype_seq_ops
,
4408 sizeof(struct seq_net_private
));
4411 static const struct file_operations ptype_seq_fops
= {
4412 .owner
= THIS_MODULE
,
4413 .open
= ptype_seq_open
,
4415 .llseek
= seq_lseek
,
4416 .release
= seq_release_net
,
4420 static int __net_init
dev_proc_net_init(struct net
*net
)
4424 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
4426 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
4428 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
4431 if (wext_proc_init(net
))
4437 proc_net_remove(net
, "ptype");
4439 proc_net_remove(net
, "softnet_stat");
4441 proc_net_remove(net
, "dev");
4445 static void __net_exit
dev_proc_net_exit(struct net
*net
)
4447 wext_proc_exit(net
);
4449 proc_net_remove(net
, "ptype");
4450 proc_net_remove(net
, "softnet_stat");
4451 proc_net_remove(net
, "dev");
4454 static struct pernet_operations __net_initdata dev_proc_ops
= {
4455 .init
= dev_proc_net_init
,
4456 .exit
= dev_proc_net_exit
,
4459 static int __init
dev_proc_init(void)
4461 return register_pernet_subsys(&dev_proc_ops
);
4464 #define dev_proc_init() 0
4465 #endif /* CONFIG_PROC_FS */
4469 * netdev_set_master - set up master pointer
4470 * @slave: slave device
4471 * @master: new master device
4473 * Changes the master device of the slave. Pass %NULL to break the
4474 * bonding. The caller must hold the RTNL semaphore. On a failure
4475 * a negative errno code is returned. On success the reference counts
4476 * are adjusted and the function returns zero.
4478 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
4480 struct net_device
*old
= slave
->master
;
4490 slave
->master
= master
;
4496 EXPORT_SYMBOL(netdev_set_master
);
4499 * netdev_set_bond_master - set up bonding master/slave pair
4500 * @slave: slave device
4501 * @master: new master device
4503 * Changes the master device of the slave. Pass %NULL to break the
4504 * bonding. The caller must hold the RTNL semaphore. On a failure
4505 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4506 * to the routing socket and the function returns zero.
4508 int netdev_set_bond_master(struct net_device
*slave
, struct net_device
*master
)
4514 err
= netdev_set_master(slave
, master
);
4518 slave
->flags
|= IFF_SLAVE
;
4520 slave
->flags
&= ~IFF_SLAVE
;
4522 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
4525 EXPORT_SYMBOL(netdev_set_bond_master
);
4527 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
4529 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4531 if ((dev
->flags
& IFF_UP
) && ops
->ndo_change_rx_flags
)
4532 ops
->ndo_change_rx_flags(dev
, flags
);
4535 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
4537 unsigned int old_flags
= dev
->flags
;
4543 dev
->flags
|= IFF_PROMISC
;
4544 dev
->promiscuity
+= inc
;
4545 if (dev
->promiscuity
== 0) {
4548 * If inc causes overflow, untouch promisc and return error.
4551 dev
->flags
&= ~IFF_PROMISC
;
4553 dev
->promiscuity
-= inc
;
4554 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4559 if (dev
->flags
!= old_flags
) {
4560 pr_info("device %s %s promiscuous mode\n",
4562 dev
->flags
& IFF_PROMISC
? "entered" : "left");
4563 if (audit_enabled
) {
4564 current_uid_gid(&uid
, &gid
);
4565 audit_log(current
->audit_context
, GFP_ATOMIC
,
4566 AUDIT_ANOM_PROMISCUOUS
,
4567 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4568 dev
->name
, (dev
->flags
& IFF_PROMISC
),
4569 (old_flags
& IFF_PROMISC
),
4570 audit_get_loginuid(current
),
4571 from_kuid(&init_user_ns
, uid
),
4572 from_kgid(&init_user_ns
, gid
),
4573 audit_get_sessionid(current
));
4576 dev_change_rx_flags(dev
, IFF_PROMISC
);
4582 * dev_set_promiscuity - update promiscuity count on a device
4586 * Add or remove promiscuity from a device. While the count in the device
4587 * remains above zero the interface remains promiscuous. Once it hits zero
4588 * the device reverts back to normal filtering operation. A negative inc
4589 * value is used to drop promiscuity on the device.
4590 * Return 0 if successful or a negative errno code on error.
4592 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
4594 unsigned int old_flags
= dev
->flags
;
4597 err
= __dev_set_promiscuity(dev
, inc
);
4600 if (dev
->flags
!= old_flags
)
4601 dev_set_rx_mode(dev
);
4604 EXPORT_SYMBOL(dev_set_promiscuity
);
4607 * dev_set_allmulti - update allmulti count on a device
4611 * Add or remove reception of all multicast frames to a device. While the
4612 * count in the device remains above zero the interface remains listening
4613 * to all interfaces. Once it hits zero the device reverts back to normal
4614 * filtering operation. A negative @inc value is used to drop the counter
4615 * when releasing a resource needing all multicasts.
4616 * Return 0 if successful or a negative errno code on error.
4619 int dev_set_allmulti(struct net_device
*dev
, int inc
)
4621 unsigned int old_flags
= dev
->flags
;
4625 dev
->flags
|= IFF_ALLMULTI
;
4626 dev
->allmulti
+= inc
;
4627 if (dev
->allmulti
== 0) {
4630 * If inc causes overflow, untouch allmulti and return error.
4633 dev
->flags
&= ~IFF_ALLMULTI
;
4635 dev
->allmulti
-= inc
;
4636 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4641 if (dev
->flags
^ old_flags
) {
4642 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
4643 dev_set_rx_mode(dev
);
4647 EXPORT_SYMBOL(dev_set_allmulti
);
4650 * Upload unicast and multicast address lists to device and
4651 * configure RX filtering. When the device doesn't support unicast
4652 * filtering it is put in promiscuous mode while unicast addresses
4655 void __dev_set_rx_mode(struct net_device
*dev
)
4657 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4659 /* dev_open will call this function so the list will stay sane. */
4660 if (!(dev
->flags
&IFF_UP
))
4663 if (!netif_device_present(dev
))
4666 if (!(dev
->priv_flags
& IFF_UNICAST_FLT
)) {
4667 /* Unicast addresses changes may only happen under the rtnl,
4668 * therefore calling __dev_set_promiscuity here is safe.
4670 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
4671 __dev_set_promiscuity(dev
, 1);
4672 dev
->uc_promisc
= true;
4673 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
4674 __dev_set_promiscuity(dev
, -1);
4675 dev
->uc_promisc
= false;
4679 if (ops
->ndo_set_rx_mode
)
4680 ops
->ndo_set_rx_mode(dev
);
4683 void dev_set_rx_mode(struct net_device
*dev
)
4685 netif_addr_lock_bh(dev
);
4686 __dev_set_rx_mode(dev
);
4687 netif_addr_unlock_bh(dev
);
4691 * dev_get_flags - get flags reported to userspace
4694 * Get the combination of flag bits exported through APIs to userspace.
4696 unsigned int dev_get_flags(const struct net_device
*dev
)
4700 flags
= (dev
->flags
& ~(IFF_PROMISC
|
4705 (dev
->gflags
& (IFF_PROMISC
|
4708 if (netif_running(dev
)) {
4709 if (netif_oper_up(dev
))
4710 flags
|= IFF_RUNNING
;
4711 if (netif_carrier_ok(dev
))
4712 flags
|= IFF_LOWER_UP
;
4713 if (netif_dormant(dev
))
4714 flags
|= IFF_DORMANT
;
4719 EXPORT_SYMBOL(dev_get_flags
);
4721 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4723 unsigned int old_flags
= dev
->flags
;
4729 * Set the flags on our device.
4732 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
4733 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
4735 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
4739 * Load in the correct multicast list now the flags have changed.
4742 if ((old_flags
^ flags
) & IFF_MULTICAST
)
4743 dev_change_rx_flags(dev
, IFF_MULTICAST
);
4745 dev_set_rx_mode(dev
);
4748 * Have we downed the interface. We handle IFF_UP ourselves
4749 * according to user attempts to set it, rather than blindly
4754 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
4755 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
4758 dev_set_rx_mode(dev
);
4761 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
4762 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
4764 dev
->gflags
^= IFF_PROMISC
;
4765 dev_set_promiscuity(dev
, inc
);
4768 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4769 is important. Some (broken) drivers set IFF_PROMISC, when
4770 IFF_ALLMULTI is requested not asking us and not reporting.
4772 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
4773 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
4775 dev
->gflags
^= IFF_ALLMULTI
;
4776 dev_set_allmulti(dev
, inc
);
4782 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
)
4784 unsigned int changes
= dev
->flags
^ old_flags
;
4786 if (changes
& IFF_UP
) {
4787 if (dev
->flags
& IFF_UP
)
4788 call_netdevice_notifiers(NETDEV_UP
, dev
);
4790 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
4793 if (dev
->flags
& IFF_UP
&&
4794 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
)))
4795 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
4799 * dev_change_flags - change device settings
4801 * @flags: device state flags
4803 * Change settings on device based state flags. The flags are
4804 * in the userspace exported format.
4806 int dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4809 unsigned int changes
, old_flags
= dev
->flags
;
4811 ret
= __dev_change_flags(dev
, flags
);
4815 changes
= old_flags
^ dev
->flags
;
4817 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
4819 __dev_notify_flags(dev
, old_flags
);
4822 EXPORT_SYMBOL(dev_change_flags
);
4825 * dev_set_mtu - Change maximum transfer unit
4827 * @new_mtu: new transfer unit
4829 * Change the maximum transfer size of the network device.
4831 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
4833 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4836 if (new_mtu
== dev
->mtu
)
4839 /* MTU must be positive. */
4843 if (!netif_device_present(dev
))
4847 if (ops
->ndo_change_mtu
)
4848 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
4852 if (!err
&& dev
->flags
& IFF_UP
)
4853 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
4856 EXPORT_SYMBOL(dev_set_mtu
);
4859 * dev_set_group - Change group this device belongs to
4861 * @new_group: group this device should belong to
4863 void dev_set_group(struct net_device
*dev
, int new_group
)
4865 dev
->group
= new_group
;
4867 EXPORT_SYMBOL(dev_set_group
);
4870 * dev_set_mac_address - Change Media Access Control Address
4874 * Change the hardware (MAC) address of the device
4876 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
4878 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4881 if (!ops
->ndo_set_mac_address
)
4883 if (sa
->sa_family
!= dev
->type
)
4885 if (!netif_device_present(dev
))
4887 err
= ops
->ndo_set_mac_address(dev
, sa
);
4889 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4890 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
4893 EXPORT_SYMBOL(dev_set_mac_address
);
4896 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4898 static int dev_ifsioc_locked(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4901 struct net_device
*dev
= dev_get_by_name_rcu(net
, ifr
->ifr_name
);
4907 case SIOCGIFFLAGS
: /* Get interface flags */
4908 ifr
->ifr_flags
= (short) dev_get_flags(dev
);
4911 case SIOCGIFMETRIC
: /* Get the metric on the interface
4912 (currently unused) */
4913 ifr
->ifr_metric
= 0;
4916 case SIOCGIFMTU
: /* Get the MTU of a device */
4917 ifr
->ifr_mtu
= dev
->mtu
;
4922 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
4924 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
4925 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4926 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
4934 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
4935 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
4936 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
4937 ifr
->ifr_map
.irq
= dev
->irq
;
4938 ifr
->ifr_map
.dma
= dev
->dma
;
4939 ifr
->ifr_map
.port
= dev
->if_port
;
4943 ifr
->ifr_ifindex
= dev
->ifindex
;
4947 ifr
->ifr_qlen
= dev
->tx_queue_len
;
4951 /* dev_ioctl() should ensure this case
4963 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4965 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4968 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
4969 const struct net_device_ops
*ops
;
4974 ops
= dev
->netdev_ops
;
4977 case SIOCSIFFLAGS
: /* Set interface flags */
4978 return dev_change_flags(dev
, ifr
->ifr_flags
);
4980 case SIOCSIFMETRIC
: /* Set the metric on the interface
4981 (currently unused) */
4984 case SIOCSIFMTU
: /* Set the MTU of a device */
4985 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
4988 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
4990 case SIOCSIFHWBROADCAST
:
4991 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
4993 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
4994 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4995 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4999 if (ops
->ndo_set_config
) {
5000 if (!netif_device_present(dev
))
5002 return ops
->ndo_set_config(dev
, &ifr
->ifr_map
);
5007 if (!ops
->ndo_set_rx_mode
||
5008 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
5010 if (!netif_device_present(dev
))
5012 return dev_mc_add_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
5015 if (!ops
->ndo_set_rx_mode
||
5016 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
5018 if (!netif_device_present(dev
))
5020 return dev_mc_del_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
5023 if (ifr
->ifr_qlen
< 0)
5025 dev
->tx_queue_len
= ifr
->ifr_qlen
;
5029 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
5030 return dev_change_name(dev
, ifr
->ifr_newname
);
5033 err
= net_hwtstamp_validate(ifr
);
5039 * Unknown or private ioctl
5042 if ((cmd
>= SIOCDEVPRIVATE
&&
5043 cmd
<= SIOCDEVPRIVATE
+ 15) ||
5044 cmd
== SIOCBONDENSLAVE
||
5045 cmd
== SIOCBONDRELEASE
||
5046 cmd
== SIOCBONDSETHWADDR
||
5047 cmd
== SIOCBONDSLAVEINFOQUERY
||
5048 cmd
== SIOCBONDINFOQUERY
||
5049 cmd
== SIOCBONDCHANGEACTIVE
||
5050 cmd
== SIOCGMIIPHY
||
5051 cmd
== SIOCGMIIREG
||
5052 cmd
== SIOCSMIIREG
||
5053 cmd
== SIOCBRADDIF
||
5054 cmd
== SIOCBRDELIF
||
5055 cmd
== SIOCSHWTSTAMP
||
5056 cmd
== SIOCWANDEV
) {
5058 if (ops
->ndo_do_ioctl
) {
5059 if (netif_device_present(dev
))
5060 err
= ops
->ndo_do_ioctl(dev
, ifr
, cmd
);
5072 * This function handles all "interface"-type I/O control requests. The actual
5073 * 'doing' part of this is dev_ifsioc above.
5077 * dev_ioctl - network device ioctl
5078 * @net: the applicable net namespace
5079 * @cmd: command to issue
5080 * @arg: pointer to a struct ifreq in user space
5082 * Issue ioctl functions to devices. This is normally called by the
5083 * user space syscall interfaces but can sometimes be useful for
5084 * other purposes. The return value is the return from the syscall if
5085 * positive or a negative errno code on error.
5088 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
5094 /* One special case: SIOCGIFCONF takes ifconf argument
5095 and requires shared lock, because it sleeps writing
5099 if (cmd
== SIOCGIFCONF
) {
5101 ret
= dev_ifconf(net
, (char __user
*) arg
);
5105 if (cmd
== SIOCGIFNAME
)
5106 return dev_ifname(net
, (struct ifreq __user
*)arg
);
5108 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
5111 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
5113 colon
= strchr(ifr
.ifr_name
, ':');
5118 * See which interface the caller is talking about.
5123 * These ioctl calls:
5124 * - can be done by all.
5125 * - atomic and do not require locking.
5136 dev_load(net
, ifr
.ifr_name
);
5138 ret
= dev_ifsioc_locked(net
, &ifr
, cmd
);
5143 if (copy_to_user(arg
, &ifr
,
5144 sizeof(struct ifreq
)))
5150 dev_load(net
, ifr
.ifr_name
);
5152 ret
= dev_ethtool(net
, &ifr
);
5157 if (copy_to_user(arg
, &ifr
,
5158 sizeof(struct ifreq
)))
5164 * These ioctl calls:
5165 * - require superuser power.
5166 * - require strict serialization.
5172 if (!capable(CAP_NET_ADMIN
))
5174 dev_load(net
, ifr
.ifr_name
);
5176 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5181 if (copy_to_user(arg
, &ifr
,
5182 sizeof(struct ifreq
)))
5188 * These ioctl calls:
5189 * - require superuser power.
5190 * - require strict serialization.
5191 * - do not return a value
5201 case SIOCSIFHWBROADCAST
:
5204 case SIOCBONDENSLAVE
:
5205 case SIOCBONDRELEASE
:
5206 case SIOCBONDSETHWADDR
:
5207 case SIOCBONDCHANGEACTIVE
:
5211 if (!capable(CAP_NET_ADMIN
))
5214 case SIOCBONDSLAVEINFOQUERY
:
5215 case SIOCBONDINFOQUERY
:
5216 dev_load(net
, ifr
.ifr_name
);
5218 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5223 /* Get the per device memory space. We can add this but
5224 * currently do not support it */
5226 /* Set the per device memory buffer space.
5227 * Not applicable in our case */
5232 * Unknown or private ioctl.
5235 if (cmd
== SIOCWANDEV
||
5236 (cmd
>= SIOCDEVPRIVATE
&&
5237 cmd
<= SIOCDEVPRIVATE
+ 15)) {
5238 dev_load(net
, ifr
.ifr_name
);
5240 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5242 if (!ret
&& copy_to_user(arg
, &ifr
,
5243 sizeof(struct ifreq
)))
5247 /* Take care of Wireless Extensions */
5248 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
5249 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
5256 * dev_new_index - allocate an ifindex
5257 * @net: the applicable net namespace
5259 * Returns a suitable unique value for a new device interface
5260 * number. The caller must hold the rtnl semaphore or the
5261 * dev_base_lock to be sure it remains unique.
5263 static int dev_new_index(struct net
*net
)
5265 int ifindex
= net
->ifindex
;
5269 if (!__dev_get_by_index(net
, ifindex
))
5270 return net
->ifindex
= ifindex
;
5274 /* Delayed registration/unregisteration */
5275 static LIST_HEAD(net_todo_list
);
5277 static void net_set_todo(struct net_device
*dev
)
5279 list_add_tail(&dev
->todo_list
, &net_todo_list
);
5282 static void rollback_registered_many(struct list_head
*head
)
5284 struct net_device
*dev
, *tmp
;
5286 BUG_ON(dev_boot_phase
);
5289 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
5290 /* Some devices call without registering
5291 * for initialization unwind. Remove those
5292 * devices and proceed with the remaining.
5294 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5295 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5299 list_del(&dev
->unreg_list
);
5302 dev
->dismantle
= true;
5303 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
5306 /* If device is running, close it first. */
5307 dev_close_many(head
);
5309 list_for_each_entry(dev
, head
, unreg_list
) {
5310 /* And unlink it from device chain. */
5311 unlist_netdevice(dev
);
5313 dev
->reg_state
= NETREG_UNREGISTERING
;
5318 list_for_each_entry(dev
, head
, unreg_list
) {
5319 /* Shutdown queueing discipline. */
5323 /* Notify protocols, that we are about to destroy
5324 this device. They should clean all the things.
5326 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5328 if (!dev
->rtnl_link_ops
||
5329 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5330 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
5333 * Flush the unicast and multicast chains
5338 if (dev
->netdev_ops
->ndo_uninit
)
5339 dev
->netdev_ops
->ndo_uninit(dev
);
5341 /* Notifier chain MUST detach us from master device. */
5342 WARN_ON(dev
->master
);
5344 /* Remove entries from kobject tree */
5345 netdev_unregister_kobject(dev
);
5350 list_for_each_entry(dev
, head
, unreg_list
)
5354 static void rollback_registered(struct net_device
*dev
)
5358 list_add(&dev
->unreg_list
, &single
);
5359 rollback_registered_many(&single
);
5363 static netdev_features_t
netdev_fix_features(struct net_device
*dev
,
5364 netdev_features_t features
)
5366 /* Fix illegal checksum combinations */
5367 if ((features
& NETIF_F_HW_CSUM
) &&
5368 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5369 netdev_warn(dev
, "mixed HW and IP checksum settings.\n");
5370 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5373 /* Fix illegal SG+CSUM combinations. */
5374 if ((features
& NETIF_F_SG
) &&
5375 !(features
& NETIF_F_ALL_CSUM
)) {
5377 "Dropping NETIF_F_SG since no checksum feature.\n");
5378 features
&= ~NETIF_F_SG
;
5381 /* TSO requires that SG is present as well. */
5382 if ((features
& NETIF_F_ALL_TSO
) && !(features
& NETIF_F_SG
)) {
5383 netdev_dbg(dev
, "Dropping TSO features since no SG feature.\n");
5384 features
&= ~NETIF_F_ALL_TSO
;
5387 /* TSO ECN requires that TSO is present as well. */
5388 if ((features
& NETIF_F_ALL_TSO
) == NETIF_F_TSO_ECN
)
5389 features
&= ~NETIF_F_TSO_ECN
;
5391 /* Software GSO depends on SG. */
5392 if ((features
& NETIF_F_GSO
) && !(features
& NETIF_F_SG
)) {
5393 netdev_dbg(dev
, "Dropping NETIF_F_GSO since no SG feature.\n");
5394 features
&= ~NETIF_F_GSO
;
5397 /* UFO needs SG and checksumming */
5398 if (features
& NETIF_F_UFO
) {
5399 /* maybe split UFO into V4 and V6? */
5400 if (!((features
& NETIF_F_GEN_CSUM
) ||
5401 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))
5402 == (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5404 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5405 features
&= ~NETIF_F_UFO
;
5408 if (!(features
& NETIF_F_SG
)) {
5410 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5411 features
&= ~NETIF_F_UFO
;
5418 int __netdev_update_features(struct net_device
*dev
)
5420 netdev_features_t features
;
5425 features
= netdev_get_wanted_features(dev
);
5427 if (dev
->netdev_ops
->ndo_fix_features
)
5428 features
= dev
->netdev_ops
->ndo_fix_features(dev
, features
);
5430 /* driver might be less strict about feature dependencies */
5431 features
= netdev_fix_features(dev
, features
);
5433 if (dev
->features
== features
)
5436 netdev_dbg(dev
, "Features changed: %pNF -> %pNF\n",
5437 &dev
->features
, &features
);
5439 if (dev
->netdev_ops
->ndo_set_features
)
5440 err
= dev
->netdev_ops
->ndo_set_features(dev
, features
);
5442 if (unlikely(err
< 0)) {
5444 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5445 err
, &features
, &dev
->features
);
5450 dev
->features
= features
;
5456 * netdev_update_features - recalculate device features
5457 * @dev: the device to check
5459 * Recalculate dev->features set and send notifications if it
5460 * has changed. Should be called after driver or hardware dependent
5461 * conditions might have changed that influence the features.
5463 void netdev_update_features(struct net_device
*dev
)
5465 if (__netdev_update_features(dev
))
5466 netdev_features_change(dev
);
5468 EXPORT_SYMBOL(netdev_update_features
);
5471 * netdev_change_features - recalculate device features
5472 * @dev: the device to check
5474 * Recalculate dev->features set and send notifications even
5475 * if they have not changed. Should be called instead of
5476 * netdev_update_features() if also dev->vlan_features might
5477 * have changed to allow the changes to be propagated to stacked
5480 void netdev_change_features(struct net_device
*dev
)
5482 __netdev_update_features(dev
);
5483 netdev_features_change(dev
);
5485 EXPORT_SYMBOL(netdev_change_features
);
5488 * netif_stacked_transfer_operstate - transfer operstate
5489 * @rootdev: the root or lower level device to transfer state from
5490 * @dev: the device to transfer operstate to
5492 * Transfer operational state from root to device. This is normally
5493 * called when a stacking relationship exists between the root
5494 * device and the device(a leaf device).
5496 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
5497 struct net_device
*dev
)
5499 if (rootdev
->operstate
== IF_OPER_DORMANT
)
5500 netif_dormant_on(dev
);
5502 netif_dormant_off(dev
);
5504 if (netif_carrier_ok(rootdev
)) {
5505 if (!netif_carrier_ok(dev
))
5506 netif_carrier_on(dev
);
5508 if (netif_carrier_ok(dev
))
5509 netif_carrier_off(dev
);
5512 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
5515 static int netif_alloc_rx_queues(struct net_device
*dev
)
5517 unsigned int i
, count
= dev
->num_rx_queues
;
5518 struct netdev_rx_queue
*rx
;
5522 rx
= kcalloc(count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5524 pr_err("netdev: Unable to allocate %u rx queues\n", count
);
5529 for (i
= 0; i
< count
; i
++)
5535 static void netdev_init_one_queue(struct net_device
*dev
,
5536 struct netdev_queue
*queue
, void *_unused
)
5538 /* Initialize queue lock */
5539 spin_lock_init(&queue
->_xmit_lock
);
5540 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
5541 queue
->xmit_lock_owner
= -1;
5542 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
5545 dql_init(&queue
->dql
, HZ
);
5549 static int netif_alloc_netdev_queues(struct net_device
*dev
)
5551 unsigned int count
= dev
->num_tx_queues
;
5552 struct netdev_queue
*tx
;
5556 tx
= kcalloc(count
, sizeof(struct netdev_queue
), GFP_KERNEL
);
5558 pr_err("netdev: Unable to allocate %u tx queues\n", count
);
5563 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5564 spin_lock_init(&dev
->tx_global_lock
);
5570 * register_netdevice - register a network device
5571 * @dev: device to register
5573 * Take a completed network device structure and add it to the kernel
5574 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5575 * chain. 0 is returned on success. A negative errno code is returned
5576 * on a failure to set up the device, or if the name is a duplicate.
5578 * Callers must hold the rtnl semaphore. You may want
5579 * register_netdev() instead of this.
5582 * The locking appears insufficient to guarantee two parallel registers
5583 * will not get the same name.
5586 int register_netdevice(struct net_device
*dev
)
5589 struct net
*net
= dev_net(dev
);
5591 BUG_ON(dev_boot_phase
);
5596 /* When net_device's are persistent, this will be fatal. */
5597 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
5600 spin_lock_init(&dev
->addr_list_lock
);
5601 netdev_set_addr_lockdep_class(dev
);
5605 ret
= dev_get_valid_name(net
, dev
, dev
->name
);
5609 /* Init, if this function is available */
5610 if (dev
->netdev_ops
->ndo_init
) {
5611 ret
= dev
->netdev_ops
->ndo_init(dev
);
5621 dev
->ifindex
= dev_new_index(net
);
5622 else if (__dev_get_by_index(net
, dev
->ifindex
))
5625 if (dev
->iflink
== -1)
5626 dev
->iflink
= dev
->ifindex
;
5628 /* Transfer changeable features to wanted_features and enable
5629 * software offloads (GSO and GRO).
5631 dev
->hw_features
|= NETIF_F_SOFT_FEATURES
;
5632 dev
->features
|= NETIF_F_SOFT_FEATURES
;
5633 dev
->wanted_features
= dev
->features
& dev
->hw_features
;
5635 /* Turn on no cache copy if HW is doing checksum */
5636 if (!(dev
->flags
& IFF_LOOPBACK
)) {
5637 dev
->hw_features
|= NETIF_F_NOCACHE_COPY
;
5638 if (dev
->features
& NETIF_F_ALL_CSUM
) {
5639 dev
->wanted_features
|= NETIF_F_NOCACHE_COPY
;
5640 dev
->features
|= NETIF_F_NOCACHE_COPY
;
5644 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5646 dev
->vlan_features
|= NETIF_F_HIGHDMA
;
5648 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
5649 ret
= notifier_to_errno(ret
);
5653 ret
= netdev_register_kobject(dev
);
5656 dev
->reg_state
= NETREG_REGISTERED
;
5658 __netdev_update_features(dev
);
5661 * Default initial state at registry is that the
5662 * device is present.
5665 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5667 linkwatch_init_dev(dev
);
5669 dev_init_scheduler(dev
);
5671 list_netdevice(dev
);
5672 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
5674 /* Notify protocols, that a new device appeared. */
5675 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5676 ret
= notifier_to_errno(ret
);
5678 rollback_registered(dev
);
5679 dev
->reg_state
= NETREG_UNREGISTERED
;
5682 * Prevent userspace races by waiting until the network
5683 * device is fully setup before sending notifications.
5685 if (!dev
->rtnl_link_ops
||
5686 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5687 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5693 if (dev
->netdev_ops
->ndo_uninit
)
5694 dev
->netdev_ops
->ndo_uninit(dev
);
5697 EXPORT_SYMBOL(register_netdevice
);
5700 * init_dummy_netdev - init a dummy network device for NAPI
5701 * @dev: device to init
5703 * This takes a network device structure and initialize the minimum
5704 * amount of fields so it can be used to schedule NAPI polls without
5705 * registering a full blown interface. This is to be used by drivers
5706 * that need to tie several hardware interfaces to a single NAPI
5707 * poll scheduler due to HW limitations.
5709 int init_dummy_netdev(struct net_device
*dev
)
5711 /* Clear everything. Note we don't initialize spinlocks
5712 * are they aren't supposed to be taken by any of the
5713 * NAPI code and this dummy netdev is supposed to be
5714 * only ever used for NAPI polls
5716 memset(dev
, 0, sizeof(struct net_device
));
5718 /* make sure we BUG if trying to hit standard
5719 * register/unregister code path
5721 dev
->reg_state
= NETREG_DUMMY
;
5723 /* NAPI wants this */
5724 INIT_LIST_HEAD(&dev
->napi_list
);
5726 /* a dummy interface is started by default */
5727 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5728 set_bit(__LINK_STATE_START
, &dev
->state
);
5730 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5731 * because users of this 'device' dont need to change
5737 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
5741 * register_netdev - register a network device
5742 * @dev: device to register
5744 * Take a completed network device structure and add it to the kernel
5745 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5746 * chain. 0 is returned on success. A negative errno code is returned
5747 * on a failure to set up the device, or if the name is a duplicate.
5749 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5750 * and expands the device name if you passed a format string to
5753 int register_netdev(struct net_device
*dev
)
5758 err
= register_netdevice(dev
);
5762 EXPORT_SYMBOL(register_netdev
);
5764 int netdev_refcnt_read(const struct net_device
*dev
)
5768 for_each_possible_cpu(i
)
5769 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
5772 EXPORT_SYMBOL(netdev_refcnt_read
);
5775 * netdev_wait_allrefs - wait until all references are gone.
5776 * @dev: target net_device
5778 * This is called when unregistering network devices.
5780 * Any protocol or device that holds a reference should register
5781 * for netdevice notification, and cleanup and put back the
5782 * reference if they receive an UNREGISTER event.
5783 * We can get stuck here if buggy protocols don't correctly
5786 static void netdev_wait_allrefs(struct net_device
*dev
)
5788 unsigned long rebroadcast_time
, warning_time
;
5791 linkwatch_forget_dev(dev
);
5793 rebroadcast_time
= warning_time
= jiffies
;
5794 refcnt
= netdev_refcnt_read(dev
);
5796 while (refcnt
!= 0) {
5797 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
5800 /* Rebroadcast unregister notification */
5801 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5807 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
5808 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
5810 /* We must not have linkwatch events
5811 * pending on unregister. If this
5812 * happens, we simply run the queue
5813 * unscheduled, resulting in a noop
5816 linkwatch_run_queue();
5821 rebroadcast_time
= jiffies
;
5826 refcnt
= netdev_refcnt_read(dev
);
5828 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
5829 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5831 warning_time
= jiffies
;
5840 * register_netdevice(x1);
5841 * register_netdevice(x2);
5843 * unregister_netdevice(y1);
5844 * unregister_netdevice(y2);
5850 * We are invoked by rtnl_unlock().
5851 * This allows us to deal with problems:
5852 * 1) We can delete sysfs objects which invoke hotplug
5853 * without deadlocking with linkwatch via keventd.
5854 * 2) Since we run with the RTNL semaphore not held, we can sleep
5855 * safely in order to wait for the netdev refcnt to drop to zero.
5857 * We must not return until all unregister events added during
5858 * the interval the lock was held have been completed.
5860 void netdev_run_todo(void)
5862 struct list_head list
;
5864 /* Snapshot list, allow later requests */
5865 list_replace_init(&net_todo_list
, &list
);
5870 /* Wait for rcu callbacks to finish before next phase */
5871 if (!list_empty(&list
))
5874 while (!list_empty(&list
)) {
5875 struct net_device
*dev
5876 = list_first_entry(&list
, struct net_device
, todo_list
);
5877 list_del(&dev
->todo_list
);
5880 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
5883 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
5884 pr_err("network todo '%s' but state %d\n",
5885 dev
->name
, dev
->reg_state
);
5890 dev
->reg_state
= NETREG_UNREGISTERED
;
5892 on_each_cpu(flush_backlog
, dev
, 1);
5894 netdev_wait_allrefs(dev
);
5897 BUG_ON(netdev_refcnt_read(dev
));
5898 WARN_ON(rcu_access_pointer(dev
->ip_ptr
));
5899 WARN_ON(rcu_access_pointer(dev
->ip6_ptr
));
5900 WARN_ON(dev
->dn_ptr
);
5902 if (dev
->destructor
)
5903 dev
->destructor(dev
);
5905 /* Free network device */
5906 kobject_put(&dev
->dev
.kobj
);
5910 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5911 * fields in the same order, with only the type differing.
5913 void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
5914 const struct net_device_stats
*netdev_stats
)
5916 #if BITS_PER_LONG == 64
5917 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
5918 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
5920 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
5921 const unsigned long *src
= (const unsigned long *)netdev_stats
;
5922 u64
*dst
= (u64
*)stats64
;
5924 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
5925 sizeof(*stats64
) / sizeof(u64
));
5926 for (i
= 0; i
< n
; i
++)
5930 EXPORT_SYMBOL(netdev_stats_to_stats64
);
5933 * dev_get_stats - get network device statistics
5934 * @dev: device to get statistics from
5935 * @storage: place to store stats
5937 * Get network statistics from device. Return @storage.
5938 * The device driver may provide its own method by setting
5939 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5940 * otherwise the internal statistics structure is used.
5942 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
5943 struct rtnl_link_stats64
*storage
)
5945 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5947 if (ops
->ndo_get_stats64
) {
5948 memset(storage
, 0, sizeof(*storage
));
5949 ops
->ndo_get_stats64(dev
, storage
);
5950 } else if (ops
->ndo_get_stats
) {
5951 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
5953 netdev_stats_to_stats64(storage
, &dev
->stats
);
5955 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
5958 EXPORT_SYMBOL(dev_get_stats
);
5960 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
5962 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
5964 #ifdef CONFIG_NET_CLS_ACT
5967 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
5970 netdev_init_one_queue(dev
, queue
, NULL
);
5971 queue
->qdisc
= &noop_qdisc
;
5972 queue
->qdisc_sleeping
= &noop_qdisc
;
5973 rcu_assign_pointer(dev
->ingress_queue
, queue
);
5978 static const struct ethtool_ops default_ethtool_ops
;
5981 * alloc_netdev_mqs - allocate network device
5982 * @sizeof_priv: size of private data to allocate space for
5983 * @name: device name format string
5984 * @setup: callback to initialize device
5985 * @txqs: the number of TX subqueues to allocate
5986 * @rxqs: the number of RX subqueues to allocate
5988 * Allocates a struct net_device with private data area for driver use
5989 * and performs basic initialization. Also allocates subquue structs
5990 * for each queue on the device.
5992 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
5993 void (*setup
)(struct net_device
*),
5994 unsigned int txqs
, unsigned int rxqs
)
5996 struct net_device
*dev
;
5998 struct net_device
*p
;
6000 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
6003 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6009 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6014 alloc_size
= sizeof(struct net_device
);
6016 /* ensure 32-byte alignment of private area */
6017 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
6018 alloc_size
+= sizeof_priv
;
6020 /* ensure 32-byte alignment of whole construct */
6021 alloc_size
+= NETDEV_ALIGN
- 1;
6023 p
= kzalloc(alloc_size
, GFP_KERNEL
);
6025 pr_err("alloc_netdev: Unable to allocate device\n");
6029 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
6030 dev
->padded
= (char *)dev
- (char *)p
;
6032 dev
->pcpu_refcnt
= alloc_percpu(int);
6033 if (!dev
->pcpu_refcnt
)
6036 if (dev_addr_init(dev
))
6042 dev_net_set(dev
, &init_net
);
6044 dev
->gso_max_size
= GSO_MAX_SIZE
;
6045 dev
->gso_max_segs
= GSO_MAX_SEGS
;
6047 INIT_LIST_HEAD(&dev
->napi_list
);
6048 INIT_LIST_HEAD(&dev
->unreg_list
);
6049 INIT_LIST_HEAD(&dev
->link_watch_list
);
6050 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
6053 dev
->num_tx_queues
= txqs
;
6054 dev
->real_num_tx_queues
= txqs
;
6055 if (netif_alloc_netdev_queues(dev
))
6059 dev
->num_rx_queues
= rxqs
;
6060 dev
->real_num_rx_queues
= rxqs
;
6061 if (netif_alloc_rx_queues(dev
))
6065 strcpy(dev
->name
, name
);
6066 dev
->group
= INIT_NETDEV_GROUP
;
6067 if (!dev
->ethtool_ops
)
6068 dev
->ethtool_ops
= &default_ethtool_ops
;
6076 free_percpu(dev
->pcpu_refcnt
);
6086 EXPORT_SYMBOL(alloc_netdev_mqs
);
6089 * free_netdev - free network device
6092 * This function does the last stage of destroying an allocated device
6093 * interface. The reference to the device object is released.
6094 * If this is the last reference then it will be freed.
6096 void free_netdev(struct net_device
*dev
)
6098 struct napi_struct
*p
, *n
;
6100 release_net(dev_net(dev
));
6107 kfree(rcu_dereference_protected(dev
->ingress_queue
, 1));
6109 /* Flush device addresses */
6110 dev_addr_flush(dev
);
6112 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
6115 free_percpu(dev
->pcpu_refcnt
);
6116 dev
->pcpu_refcnt
= NULL
;
6118 /* Compatibility with error handling in drivers */
6119 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
6120 kfree((char *)dev
- dev
->padded
);
6124 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
6125 dev
->reg_state
= NETREG_RELEASED
;
6127 /* will free via device release */
6128 put_device(&dev
->dev
);
6130 EXPORT_SYMBOL(free_netdev
);
6133 * synchronize_net - Synchronize with packet receive processing
6135 * Wait for packets currently being received to be done.
6136 * Does not block later packets from starting.
6138 void synchronize_net(void)
6141 if (rtnl_is_locked())
6142 synchronize_rcu_expedited();
6146 EXPORT_SYMBOL(synchronize_net
);
6149 * unregister_netdevice_queue - remove device from the kernel
6153 * This function shuts down a device interface and removes it
6154 * from the kernel tables.
6155 * If head not NULL, device is queued to be unregistered later.
6157 * Callers must hold the rtnl semaphore. You may want
6158 * unregister_netdev() instead of this.
6161 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
6166 list_move_tail(&dev
->unreg_list
, head
);
6168 rollback_registered(dev
);
6169 /* Finish processing unregister after unlock */
6173 EXPORT_SYMBOL(unregister_netdevice_queue
);
6176 * unregister_netdevice_many - unregister many devices
6177 * @head: list of devices
6179 void unregister_netdevice_many(struct list_head
*head
)
6181 struct net_device
*dev
;
6183 if (!list_empty(head
)) {
6184 rollback_registered_many(head
);
6185 list_for_each_entry(dev
, head
, unreg_list
)
6189 EXPORT_SYMBOL(unregister_netdevice_many
);
6192 * unregister_netdev - remove device from the kernel
6195 * This function shuts down a device interface and removes it
6196 * from the kernel tables.
6198 * This is just a wrapper for unregister_netdevice that takes
6199 * the rtnl semaphore. In general you want to use this and not
6200 * unregister_netdevice.
6202 void unregister_netdev(struct net_device
*dev
)
6205 unregister_netdevice(dev
);
6208 EXPORT_SYMBOL(unregister_netdev
);
6211 * dev_change_net_namespace - move device to different nethost namespace
6213 * @net: network namespace
6214 * @pat: If not NULL name pattern to try if the current device name
6215 * is already taken in the destination network namespace.
6217 * This function shuts down a device interface and moves it
6218 * to a new network namespace. On success 0 is returned, on
6219 * a failure a netagive errno code is returned.
6221 * Callers must hold the rtnl semaphore.
6224 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
6230 /* Don't allow namespace local devices to be moved. */
6232 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6235 /* Ensure the device has been registrered */
6237 if (dev
->reg_state
!= NETREG_REGISTERED
)
6240 /* Get out if there is nothing todo */
6242 if (net_eq(dev_net(dev
), net
))
6245 /* Pick the destination device name, and ensure
6246 * we can use it in the destination network namespace.
6249 if (__dev_get_by_name(net
, dev
->name
)) {
6250 /* We get here if we can't use the current device name */
6253 if (dev_get_valid_name(net
, dev
, pat
) < 0)
6258 * And now a mini version of register_netdevice unregister_netdevice.
6261 /* If device is running close it first. */
6264 /* And unlink it from device chain */
6266 unlist_netdevice(dev
);
6270 /* Shutdown queueing discipline. */
6273 /* Notify protocols, that we are about to destroy
6274 this device. They should clean all the things.
6276 Note that dev->reg_state stays at NETREG_REGISTERED.
6277 This is wanted because this way 8021q and macvlan know
6278 the device is just moving and can keep their slaves up.
6280 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6282 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
6283 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
6286 * Flush the unicast and multicast chains
6291 /* Actually switch the network namespace */
6292 dev_net_set(dev
, net
);
6294 /* If there is an ifindex conflict assign a new one */
6295 if (__dev_get_by_index(net
, dev
->ifindex
)) {
6296 int iflink
= (dev
->iflink
== dev
->ifindex
);
6297 dev
->ifindex
= dev_new_index(net
);
6299 dev
->iflink
= dev
->ifindex
;
6302 /* Fixup kobjects */
6303 err
= device_rename(&dev
->dev
, dev
->name
);
6306 /* Add the device back in the hashes */
6307 list_netdevice(dev
);
6309 /* Notify protocols, that a new device appeared. */
6310 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
6313 * Prevent userspace races by waiting until the network
6314 * device is fully setup before sending notifications.
6316 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
6323 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
6325 static int dev_cpu_callback(struct notifier_block
*nfb
,
6326 unsigned long action
,
6329 struct sk_buff
**list_skb
;
6330 struct sk_buff
*skb
;
6331 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
6332 struct softnet_data
*sd
, *oldsd
;
6334 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
6337 local_irq_disable();
6338 cpu
= smp_processor_id();
6339 sd
= &per_cpu(softnet_data
, cpu
);
6340 oldsd
= &per_cpu(softnet_data
, oldcpu
);
6342 /* Find end of our completion_queue. */
6343 list_skb
= &sd
->completion_queue
;
6345 list_skb
= &(*list_skb
)->next
;
6346 /* Append completion queue from offline CPU. */
6347 *list_skb
= oldsd
->completion_queue
;
6348 oldsd
->completion_queue
= NULL
;
6350 /* Append output queue from offline CPU. */
6351 if (oldsd
->output_queue
) {
6352 *sd
->output_queue_tailp
= oldsd
->output_queue
;
6353 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
6354 oldsd
->output_queue
= NULL
;
6355 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
6357 /* Append NAPI poll list from offline CPU. */
6358 if (!list_empty(&oldsd
->poll_list
)) {
6359 list_splice_init(&oldsd
->poll_list
, &sd
->poll_list
);
6360 raise_softirq_irqoff(NET_RX_SOFTIRQ
);
6363 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
6366 /* Process offline CPU's input_pkt_queue */
6367 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
6369 input_queue_head_incr(oldsd
);
6371 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
6373 input_queue_head_incr(oldsd
);
6381 * netdev_increment_features - increment feature set by one
6382 * @all: current feature set
6383 * @one: new feature set
6384 * @mask: mask feature set
6386 * Computes a new feature set after adding a device with feature set
6387 * @one to the master device with current feature set @all. Will not
6388 * enable anything that is off in @mask. Returns the new feature set.
6390 netdev_features_t
netdev_increment_features(netdev_features_t all
,
6391 netdev_features_t one
, netdev_features_t mask
)
6393 if (mask
& NETIF_F_GEN_CSUM
)
6394 mask
|= NETIF_F_ALL_CSUM
;
6395 mask
|= NETIF_F_VLAN_CHALLENGED
;
6397 all
|= one
& (NETIF_F_ONE_FOR_ALL
|NETIF_F_ALL_CSUM
) & mask
;
6398 all
&= one
| ~NETIF_F_ALL_FOR_ALL
;
6400 /* If one device supports hw checksumming, set for all. */
6401 if (all
& NETIF_F_GEN_CSUM
)
6402 all
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_GEN_CSUM
);
6406 EXPORT_SYMBOL(netdev_increment_features
);
6408 static struct hlist_head
*netdev_create_hash(void)
6411 struct hlist_head
*hash
;
6413 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
6415 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
6416 INIT_HLIST_HEAD(&hash
[i
]);
6421 /* Initialize per network namespace state */
6422 static int __net_init
netdev_init(struct net
*net
)
6424 if (net
!= &init_net
)
6425 INIT_LIST_HEAD(&net
->dev_base_head
);
6427 net
->dev_name_head
= netdev_create_hash();
6428 if (net
->dev_name_head
== NULL
)
6431 net
->dev_index_head
= netdev_create_hash();
6432 if (net
->dev_index_head
== NULL
)
6438 kfree(net
->dev_name_head
);
6444 * netdev_drivername - network driver for the device
6445 * @dev: network device
6447 * Determine network driver for device.
6449 const char *netdev_drivername(const struct net_device
*dev
)
6451 const struct device_driver
*driver
;
6452 const struct device
*parent
;
6453 const char *empty
= "";
6455 parent
= dev
->dev
.parent
;
6459 driver
= parent
->driver
;
6460 if (driver
&& driver
->name
)
6461 return driver
->name
;
6465 int __netdev_printk(const char *level
, const struct net_device
*dev
,
6466 struct va_format
*vaf
)
6470 if (dev
&& dev
->dev
.parent
)
6471 r
= dev_printk(level
, dev
->dev
.parent
, "%s: %pV",
6472 netdev_name(dev
), vaf
);
6474 r
= printk("%s%s: %pV", level
, netdev_name(dev
), vaf
);
6476 r
= printk("%s(NULL net_device): %pV", level
, vaf
);
6480 EXPORT_SYMBOL(__netdev_printk
);
6482 int netdev_printk(const char *level
, const struct net_device
*dev
,
6483 const char *format
, ...)
6485 struct va_format vaf
;
6489 va_start(args
, format
);
6494 r
= __netdev_printk(level
, dev
, &vaf
);
6499 EXPORT_SYMBOL(netdev_printk
);
6501 #define define_netdev_printk_level(func, level) \
6502 int func(const struct net_device *dev, const char *fmt, ...) \
6505 struct va_format vaf; \
6508 va_start(args, fmt); \
6513 r = __netdev_printk(level, dev, &vaf); \
6518 EXPORT_SYMBOL(func);
6520 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
6521 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
6522 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
6523 define_netdev_printk_level(netdev_err
, KERN_ERR
);
6524 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
6525 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
6526 define_netdev_printk_level(netdev_info
, KERN_INFO
);
6528 static void __net_exit
netdev_exit(struct net
*net
)
6530 kfree(net
->dev_name_head
);
6531 kfree(net
->dev_index_head
);
6534 static struct pernet_operations __net_initdata netdev_net_ops
= {
6535 .init
= netdev_init
,
6536 .exit
= netdev_exit
,
6539 static void __net_exit
default_device_exit(struct net
*net
)
6541 struct net_device
*dev
, *aux
;
6543 * Push all migratable network devices back to the
6544 * initial network namespace
6547 for_each_netdev_safe(net
, dev
, aux
) {
6549 char fb_name
[IFNAMSIZ
];
6551 /* Ignore unmoveable devices (i.e. loopback) */
6552 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6555 /* Leave virtual devices for the generic cleanup */
6556 if (dev
->rtnl_link_ops
)
6559 /* Push remaining network devices to init_net */
6560 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
6561 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
6563 pr_emerg("%s: failed to move %s to init_net: %d\n",
6564 __func__
, dev
->name
, err
);
6571 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
6573 /* At exit all network devices most be removed from a network
6574 * namespace. Do this in the reverse order of registration.
6575 * Do this across as many network namespaces as possible to
6576 * improve batching efficiency.
6578 struct net_device
*dev
;
6580 LIST_HEAD(dev_kill_list
);
6583 list_for_each_entry(net
, net_list
, exit_list
) {
6584 for_each_netdev_reverse(net
, dev
) {
6585 if (dev
->rtnl_link_ops
)
6586 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
6588 unregister_netdevice_queue(dev
, &dev_kill_list
);
6591 unregister_netdevice_many(&dev_kill_list
);
6592 list_del(&dev_kill_list
);
6596 static struct pernet_operations __net_initdata default_device_ops
= {
6597 .exit
= default_device_exit
,
6598 .exit_batch
= default_device_exit_batch
,
6602 * Initialize the DEV module. At boot time this walks the device list and
6603 * unhooks any devices that fail to initialise (normally hardware not
6604 * present) and leaves us with a valid list of present and active devices.
6609 * This is called single threaded during boot, so no need
6610 * to take the rtnl semaphore.
6612 static int __init
net_dev_init(void)
6614 int i
, rc
= -ENOMEM
;
6616 BUG_ON(!dev_boot_phase
);
6618 if (dev_proc_init())
6621 if (netdev_kobject_init())
6624 INIT_LIST_HEAD(&ptype_all
);
6625 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
6626 INIT_LIST_HEAD(&ptype_base
[i
]);
6628 if (register_pernet_subsys(&netdev_net_ops
))
6632 * Initialise the packet receive queues.
6635 for_each_possible_cpu(i
) {
6636 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
6638 memset(sd
, 0, sizeof(*sd
));
6639 skb_queue_head_init(&sd
->input_pkt_queue
);
6640 skb_queue_head_init(&sd
->process_queue
);
6641 sd
->completion_queue
= NULL
;
6642 INIT_LIST_HEAD(&sd
->poll_list
);
6643 sd
->output_queue
= NULL
;
6644 sd
->output_queue_tailp
= &sd
->output_queue
;
6646 sd
->csd
.func
= rps_trigger_softirq
;
6652 sd
->backlog
.poll
= process_backlog
;
6653 sd
->backlog
.weight
= weight_p
;
6654 sd
->backlog
.gro_list
= NULL
;
6655 sd
->backlog
.gro_count
= 0;
6660 /* The loopback device is special if any other network devices
6661 * is present in a network namespace the loopback device must
6662 * be present. Since we now dynamically allocate and free the
6663 * loopback device ensure this invariant is maintained by
6664 * keeping the loopback device as the first device on the
6665 * list of network devices. Ensuring the loopback devices
6666 * is the first device that appears and the last network device
6669 if (register_pernet_device(&loopback_net_ops
))
6672 if (register_pernet_device(&default_device_ops
))
6675 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
6676 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
6678 hotcpu_notifier(dev_cpu_callback
, 0);
6686 subsys_initcall(net_dev_init
);
6688 static int __init
initialize_hashrnd(void)
6690 get_random_bytes(&hashrnd
, sizeof(hashrnd
));
6694 late_initcall_sync(initialize_hashrnd
);