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/stat.h>
102 #include <net/pkt_sched.h>
103 #include <net/checksum.h>
104 #include <net/xfrm.h>
105 #include <linux/highmem.h>
106 #include <linux/init.h>
107 #include <linux/module.h>
108 #include <linux/netpoll.h>
109 #include <linux/rcupdate.h>
110 #include <linux/delay.h>
111 #include <net/iw_handler.h>
112 #include <asm/current.h>
113 #include <linux/audit.h>
114 #include <linux/dmaengine.h>
115 #include <linux/err.h>
116 #include <linux/ctype.h>
117 #include <linux/if_arp.h>
118 #include <linux/if_vlan.h>
119 #include <linux/ip.h>
121 #include <linux/ipv6.h>
122 #include <linux/in.h>
123 #include <linux/jhash.h>
124 #include <linux/random.h>
125 #include <trace/events/napi.h>
126 #include <trace/events/net.h>
127 #include <trace/events/skb.h>
128 #include <linux/pci.h>
129 #include <linux/inetdevice.h>
130 #include <linux/cpu_rmap.h>
131 #include <linux/static_key.h>
132 #include <linux/hashtable.h>
133 #include <linux/vmalloc.h>
135 #include "net-sysfs.h"
137 /* Instead of increasing this, you should create a hash table. */
138 #define MAX_GRO_SKBS 8
140 /* This should be increased if a protocol with a bigger head is added. */
141 #define GRO_MAX_HEAD (MAX_HEADER + 128)
143 static DEFINE_SPINLOCK(ptype_lock
);
144 static DEFINE_SPINLOCK(offload_lock
);
145 struct list_head ptype_base
[PTYPE_HASH_SIZE
] __read_mostly
;
146 struct list_head ptype_all __read_mostly
; /* Taps */
147 static struct list_head offload_base __read_mostly
;
150 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
153 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
155 * Writers must hold the rtnl semaphore while they loop through the
156 * dev_base_head list, and hold dev_base_lock for writing when they do the
157 * actual updates. This allows pure readers to access the list even
158 * while a writer is preparing to update it.
160 * To put it another way, dev_base_lock is held for writing only to
161 * protect against pure readers; the rtnl semaphore provides the
162 * protection against other writers.
164 * See, for example usages, register_netdevice() and
165 * unregister_netdevice(), which must be called with the rtnl
168 DEFINE_RWLOCK(dev_base_lock
);
169 EXPORT_SYMBOL(dev_base_lock
);
171 /* protects napi_hash addition/deletion and napi_gen_id */
172 static DEFINE_SPINLOCK(napi_hash_lock
);
174 static unsigned int napi_gen_id
;
175 static DEFINE_HASHTABLE(napi_hash
, 8);
177 static seqcount_t devnet_rename_seq
;
179 static inline void dev_base_seq_inc(struct net
*net
)
181 while (++net
->dev_base_seq
== 0);
184 static inline struct hlist_head
*dev_name_hash(struct net
*net
, const char *name
)
186 unsigned int hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
188 return &net
->dev_name_head
[hash_32(hash
, NETDEV_HASHBITS
)];
191 static inline struct hlist_head
*dev_index_hash(struct net
*net
, int ifindex
)
193 return &net
->dev_index_head
[ifindex
& (NETDEV_HASHENTRIES
- 1)];
196 static inline void rps_lock(struct softnet_data
*sd
)
199 spin_lock(&sd
->input_pkt_queue
.lock
);
203 static inline void rps_unlock(struct softnet_data
*sd
)
206 spin_unlock(&sd
->input_pkt_queue
.lock
);
210 /* Device list insertion */
211 static void list_netdevice(struct net_device
*dev
)
213 struct net
*net
= dev_net(dev
);
217 write_lock_bh(&dev_base_lock
);
218 list_add_tail_rcu(&dev
->dev_list
, &net
->dev_base_head
);
219 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
220 hlist_add_head_rcu(&dev
->index_hlist
,
221 dev_index_hash(net
, dev
->ifindex
));
222 write_unlock_bh(&dev_base_lock
);
224 dev_base_seq_inc(net
);
227 /* Device list removal
228 * caller must respect a RCU grace period before freeing/reusing dev
230 static void unlist_netdevice(struct net_device
*dev
)
234 /* Unlink dev from the device chain */
235 write_lock_bh(&dev_base_lock
);
236 list_del_rcu(&dev
->dev_list
);
237 hlist_del_rcu(&dev
->name_hlist
);
238 hlist_del_rcu(&dev
->index_hlist
);
239 write_unlock_bh(&dev_base_lock
);
241 dev_base_seq_inc(dev_net(dev
));
248 static RAW_NOTIFIER_HEAD(netdev_chain
);
251 * Device drivers call our routines to queue packets here. We empty the
252 * queue in the local softnet handler.
255 DEFINE_PER_CPU_ALIGNED(struct softnet_data
, softnet_data
);
256 EXPORT_PER_CPU_SYMBOL(softnet_data
);
258 #ifdef CONFIG_LOCKDEP
260 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
261 * according to dev->type
263 static const unsigned short netdev_lock_type
[] =
264 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
265 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
266 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
267 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
268 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
269 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
270 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
271 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
272 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
273 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
274 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
275 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
276 ARPHRD_FCFABRIC
, ARPHRD_IEEE80211
, ARPHRD_IEEE80211_PRISM
,
277 ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_PHONET
, ARPHRD_PHONET_PIPE
,
278 ARPHRD_IEEE802154
, ARPHRD_VOID
, ARPHRD_NONE
};
280 static const char *const netdev_lock_name
[] =
281 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
282 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
283 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
284 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
285 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
286 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
287 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
288 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
289 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
290 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
291 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
292 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
293 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
294 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
295 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
297 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
298 static struct lock_class_key netdev_addr_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
300 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
304 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
305 if (netdev_lock_type
[i
] == dev_type
)
307 /* the last key is used by default */
308 return ARRAY_SIZE(netdev_lock_type
) - 1;
311 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
312 unsigned short dev_type
)
316 i
= netdev_lock_pos(dev_type
);
317 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
318 netdev_lock_name
[i
]);
321 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
325 i
= netdev_lock_pos(dev
->type
);
326 lockdep_set_class_and_name(&dev
->addr_list_lock
,
327 &netdev_addr_lock_key
[i
],
328 netdev_lock_name
[i
]);
331 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
332 unsigned short dev_type
)
335 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
340 /*******************************************************************************
342 Protocol management and registration routines
344 *******************************************************************************/
347 * Add a protocol ID to the list. Now that the input handler is
348 * smarter we can dispense with all the messy stuff that used to be
351 * BEWARE!!! Protocol handlers, mangling input packets,
352 * MUST BE last in hash buckets and checking protocol handlers
353 * MUST start from promiscuous ptype_all chain in net_bh.
354 * It is true now, do not change it.
355 * Explanation follows: if protocol handler, mangling packet, will
356 * be the first on list, it is not able to sense, that packet
357 * is cloned and should be copied-on-write, so that it will
358 * change it and subsequent readers will get broken packet.
362 static inline struct list_head
*ptype_head(const struct packet_type
*pt
)
364 if (pt
->type
== htons(ETH_P_ALL
))
367 return &ptype_base
[ntohs(pt
->type
) & PTYPE_HASH_MASK
];
371 * dev_add_pack - add packet handler
372 * @pt: packet type declaration
374 * Add a protocol handler to the networking stack. The passed &packet_type
375 * is linked into kernel lists and may not be freed until it has been
376 * removed from the kernel lists.
378 * This call does not sleep therefore it can not
379 * guarantee all CPU's that are in middle of receiving packets
380 * will see the new packet type (until the next received packet).
383 void dev_add_pack(struct packet_type
*pt
)
385 struct list_head
*head
= ptype_head(pt
);
387 spin_lock(&ptype_lock
);
388 list_add_rcu(&pt
->list
, head
);
389 spin_unlock(&ptype_lock
);
391 EXPORT_SYMBOL(dev_add_pack
);
394 * __dev_remove_pack - remove packet handler
395 * @pt: packet type declaration
397 * Remove a protocol handler that was previously added to the kernel
398 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
399 * from the kernel lists and can be freed or reused once this function
402 * The packet type might still be in use by receivers
403 * and must not be freed until after all the CPU's have gone
404 * through a quiescent state.
406 void __dev_remove_pack(struct packet_type
*pt
)
408 struct list_head
*head
= ptype_head(pt
);
409 struct packet_type
*pt1
;
411 spin_lock(&ptype_lock
);
413 list_for_each_entry(pt1
, head
, list
) {
415 list_del_rcu(&pt
->list
);
420 pr_warn("dev_remove_pack: %p not found\n", pt
);
422 spin_unlock(&ptype_lock
);
424 EXPORT_SYMBOL(__dev_remove_pack
);
427 * dev_remove_pack - remove packet handler
428 * @pt: packet type declaration
430 * Remove a protocol handler that was previously added to the kernel
431 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
432 * from the kernel lists and can be freed or reused once this function
435 * This call sleeps to guarantee that no CPU is looking at the packet
438 void dev_remove_pack(struct packet_type
*pt
)
440 __dev_remove_pack(pt
);
444 EXPORT_SYMBOL(dev_remove_pack
);
448 * dev_add_offload - register offload handlers
449 * @po: protocol offload declaration
451 * Add protocol offload handlers to the networking stack. The passed
452 * &proto_offload is linked into kernel lists and may not be freed until
453 * it has been removed from the kernel lists.
455 * This call does not sleep therefore it can not
456 * guarantee all CPU's that are in middle of receiving packets
457 * will see the new offload handlers (until the next received packet).
459 void dev_add_offload(struct packet_offload
*po
)
461 struct list_head
*head
= &offload_base
;
463 spin_lock(&offload_lock
);
464 list_add_rcu(&po
->list
, head
);
465 spin_unlock(&offload_lock
);
467 EXPORT_SYMBOL(dev_add_offload
);
470 * __dev_remove_offload - remove offload handler
471 * @po: packet offload declaration
473 * Remove a protocol offload handler that was previously added to the
474 * kernel offload handlers by dev_add_offload(). The passed &offload_type
475 * is removed from the kernel lists and can be freed or reused once this
478 * The packet type might still be in use by receivers
479 * and must not be freed until after all the CPU's have gone
480 * through a quiescent state.
482 void __dev_remove_offload(struct packet_offload
*po
)
484 struct list_head
*head
= &offload_base
;
485 struct packet_offload
*po1
;
487 spin_lock(&offload_lock
);
489 list_for_each_entry(po1
, head
, list
) {
491 list_del_rcu(&po
->list
);
496 pr_warn("dev_remove_offload: %p not found\n", po
);
498 spin_unlock(&offload_lock
);
500 EXPORT_SYMBOL(__dev_remove_offload
);
503 * dev_remove_offload - remove packet offload handler
504 * @po: packet offload declaration
506 * Remove a packet offload handler that was previously added to the kernel
507 * offload handlers by dev_add_offload(). The passed &offload_type is
508 * removed from the kernel lists and can be freed or reused once this
511 * This call sleeps to guarantee that no CPU is looking at the packet
514 void dev_remove_offload(struct packet_offload
*po
)
516 __dev_remove_offload(po
);
520 EXPORT_SYMBOL(dev_remove_offload
);
522 /******************************************************************************
524 Device Boot-time Settings Routines
526 *******************************************************************************/
528 /* Boot time configuration table */
529 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
532 * netdev_boot_setup_add - add new setup entry
533 * @name: name of the device
534 * @map: configured settings for the device
536 * Adds new setup entry to the dev_boot_setup list. The function
537 * returns 0 on error and 1 on success. This is a generic routine to
540 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
542 struct netdev_boot_setup
*s
;
546 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
547 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
548 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
549 strlcpy(s
[i
].name
, name
, IFNAMSIZ
);
550 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
555 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
559 * netdev_boot_setup_check - check boot time settings
560 * @dev: the netdevice
562 * Check boot time settings for the device.
563 * The found settings are set for the device to be used
564 * later in the device probing.
565 * Returns 0 if no settings found, 1 if they are.
567 int netdev_boot_setup_check(struct net_device
*dev
)
569 struct netdev_boot_setup
*s
= dev_boot_setup
;
572 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
573 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
574 !strcmp(dev
->name
, s
[i
].name
)) {
575 dev
->irq
= s
[i
].map
.irq
;
576 dev
->base_addr
= s
[i
].map
.base_addr
;
577 dev
->mem_start
= s
[i
].map
.mem_start
;
578 dev
->mem_end
= s
[i
].map
.mem_end
;
584 EXPORT_SYMBOL(netdev_boot_setup_check
);
588 * netdev_boot_base - get address from boot time settings
589 * @prefix: prefix for network device
590 * @unit: id for network device
592 * Check boot time settings for the base address of device.
593 * The found settings are set for the device to be used
594 * later in the device probing.
595 * Returns 0 if no settings found.
597 unsigned long netdev_boot_base(const char *prefix
, int unit
)
599 const struct netdev_boot_setup
*s
= dev_boot_setup
;
603 sprintf(name
, "%s%d", prefix
, unit
);
606 * If device already registered then return base of 1
607 * to indicate not to probe for this interface
609 if (__dev_get_by_name(&init_net
, name
))
612 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
613 if (!strcmp(name
, s
[i
].name
))
614 return s
[i
].map
.base_addr
;
619 * Saves at boot time configured settings for any netdevice.
621 int __init
netdev_boot_setup(char *str
)
626 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
631 memset(&map
, 0, sizeof(map
));
635 map
.base_addr
= ints
[2];
637 map
.mem_start
= ints
[3];
639 map
.mem_end
= ints
[4];
641 /* Add new entry to the list */
642 return netdev_boot_setup_add(str
, &map
);
645 __setup("netdev=", netdev_boot_setup
);
647 /*******************************************************************************
649 Device Interface Subroutines
651 *******************************************************************************/
654 * __dev_get_by_name - find a device by its name
655 * @net: the applicable net namespace
656 * @name: name to find
658 * Find an interface by name. Must be called under RTNL semaphore
659 * or @dev_base_lock. If the name is found a pointer to the device
660 * is returned. If the name is not found then %NULL is returned. The
661 * reference counters are not incremented so the caller must be
662 * careful with locks.
665 struct net_device
*__dev_get_by_name(struct net
*net
, const char *name
)
667 struct net_device
*dev
;
668 struct hlist_head
*head
= dev_name_hash(net
, name
);
670 hlist_for_each_entry(dev
, head
, name_hlist
)
671 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
676 EXPORT_SYMBOL(__dev_get_by_name
);
679 * dev_get_by_name_rcu - find a device by its name
680 * @net: the applicable net namespace
681 * @name: name to find
683 * Find an interface by name.
684 * If the name is found a pointer to the device is returned.
685 * If the name is not found then %NULL is returned.
686 * The reference counters are not incremented so the caller must be
687 * careful with locks. The caller must hold RCU lock.
690 struct net_device
*dev_get_by_name_rcu(struct net
*net
, const char *name
)
692 struct net_device
*dev
;
693 struct hlist_head
*head
= dev_name_hash(net
, name
);
695 hlist_for_each_entry_rcu(dev
, head
, name_hlist
)
696 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
701 EXPORT_SYMBOL(dev_get_by_name_rcu
);
704 * dev_get_by_name - find a device by its name
705 * @net: the applicable net namespace
706 * @name: name to find
708 * Find an interface by name. This can be called from any
709 * context and does its own locking. The returned handle has
710 * the usage count incremented and the caller must use dev_put() to
711 * release it when it is no longer needed. %NULL is returned if no
712 * matching device is found.
715 struct net_device
*dev_get_by_name(struct net
*net
, const char *name
)
717 struct net_device
*dev
;
720 dev
= dev_get_by_name_rcu(net
, name
);
726 EXPORT_SYMBOL(dev_get_by_name
);
729 * __dev_get_by_index - find a device by its ifindex
730 * @net: the applicable net namespace
731 * @ifindex: index of device
733 * Search for an interface by index. Returns %NULL if the device
734 * is not found or a pointer to the device. The device has not
735 * had its reference counter increased so the caller must be careful
736 * about locking. The caller must hold either the RTNL semaphore
740 struct net_device
*__dev_get_by_index(struct net
*net
, int ifindex
)
742 struct net_device
*dev
;
743 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
745 hlist_for_each_entry(dev
, head
, index_hlist
)
746 if (dev
->ifindex
== ifindex
)
751 EXPORT_SYMBOL(__dev_get_by_index
);
754 * dev_get_by_index_rcu - find a device by its ifindex
755 * @net: the applicable net namespace
756 * @ifindex: index of device
758 * Search for an interface by index. Returns %NULL if the device
759 * is not found or a pointer to the device. The device has not
760 * had its reference counter increased so the caller must be careful
761 * about locking. The caller must hold RCU lock.
764 struct net_device
*dev_get_by_index_rcu(struct net
*net
, int ifindex
)
766 struct net_device
*dev
;
767 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
769 hlist_for_each_entry_rcu(dev
, head
, index_hlist
)
770 if (dev
->ifindex
== ifindex
)
775 EXPORT_SYMBOL(dev_get_by_index_rcu
);
779 * dev_get_by_index - find a device by its ifindex
780 * @net: the applicable net namespace
781 * @ifindex: index of device
783 * Search for an interface by index. Returns NULL if the device
784 * is not found or a pointer to the device. The device returned has
785 * had a reference added and the pointer is safe until the user calls
786 * dev_put to indicate they have finished with it.
789 struct net_device
*dev_get_by_index(struct net
*net
, int ifindex
)
791 struct net_device
*dev
;
794 dev
= dev_get_by_index_rcu(net
, ifindex
);
800 EXPORT_SYMBOL(dev_get_by_index
);
803 * netdev_get_name - get a netdevice name, knowing its ifindex.
804 * @net: network namespace
805 * @name: a pointer to the buffer where the name will be stored.
806 * @ifindex: the ifindex of the interface to get the name from.
808 * The use of raw_seqcount_begin() and cond_resched() before
809 * retrying is required as we want to give the writers a chance
810 * to complete when CONFIG_PREEMPT is not set.
812 int netdev_get_name(struct net
*net
, char *name
, int ifindex
)
814 struct net_device
*dev
;
818 seq
= raw_seqcount_begin(&devnet_rename_seq
);
820 dev
= dev_get_by_index_rcu(net
, ifindex
);
826 strcpy(name
, dev
->name
);
828 if (read_seqcount_retry(&devnet_rename_seq
, seq
)) {
837 * dev_getbyhwaddr_rcu - find a device by its hardware address
838 * @net: the applicable net namespace
839 * @type: media type of device
840 * @ha: hardware address
842 * Search for an interface by MAC address. Returns NULL if the device
843 * is not found or a pointer to the device.
844 * The caller must hold RCU or RTNL.
845 * The returned device has not had its ref count increased
846 * and the caller must therefore be careful about locking
850 struct net_device
*dev_getbyhwaddr_rcu(struct net
*net
, unsigned short type
,
853 struct net_device
*dev
;
855 for_each_netdev_rcu(net
, dev
)
856 if (dev
->type
== type
&&
857 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
862 EXPORT_SYMBOL(dev_getbyhwaddr_rcu
);
864 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
866 struct net_device
*dev
;
869 for_each_netdev(net
, dev
)
870 if (dev
->type
== type
)
875 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
877 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
879 struct net_device
*dev
, *ret
= NULL
;
882 for_each_netdev_rcu(net
, dev
)
883 if (dev
->type
== type
) {
891 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
894 * dev_get_by_flags_rcu - find any device with given flags
895 * @net: the applicable net namespace
896 * @if_flags: IFF_* values
897 * @mask: bitmask of bits in if_flags to check
899 * Search for any interface with the given flags. Returns NULL if a device
900 * is not found or a pointer to the device. Must be called inside
901 * rcu_read_lock(), and result refcount is unchanged.
904 struct net_device
*dev_get_by_flags_rcu(struct net
*net
, unsigned short if_flags
,
907 struct net_device
*dev
, *ret
;
910 for_each_netdev_rcu(net
, dev
) {
911 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
918 EXPORT_SYMBOL(dev_get_by_flags_rcu
);
921 * dev_valid_name - check if name is okay for network device
924 * Network device names need to be valid file names to
925 * to allow sysfs to work. We also disallow any kind of
928 bool dev_valid_name(const char *name
)
932 if (strlen(name
) >= IFNAMSIZ
)
934 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
938 if (*name
== '/' || isspace(*name
))
944 EXPORT_SYMBOL(dev_valid_name
);
947 * __dev_alloc_name - allocate a name for a device
948 * @net: network namespace to allocate the device name in
949 * @name: name format string
950 * @buf: scratch buffer and result name string
952 * Passed a format string - eg "lt%d" it will try and find a suitable
953 * id. It scans list of devices to build up a free map, then chooses
954 * the first empty slot. The caller must hold the dev_base or rtnl lock
955 * while allocating the name and adding the device in order to avoid
957 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
958 * Returns the number of the unit assigned or a negative errno code.
961 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
965 const int max_netdevices
= 8*PAGE_SIZE
;
966 unsigned long *inuse
;
967 struct net_device
*d
;
969 p
= strnchr(name
, IFNAMSIZ
-1, '%');
972 * Verify the string as this thing may have come from
973 * the user. There must be either one "%d" and no other "%"
976 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
979 /* Use one page as a bit array of possible slots */
980 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
984 for_each_netdev(net
, d
) {
985 if (!sscanf(d
->name
, name
, &i
))
987 if (i
< 0 || i
>= max_netdevices
)
990 /* avoid cases where sscanf is not exact inverse of printf */
991 snprintf(buf
, IFNAMSIZ
, name
, i
);
992 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
996 i
= find_first_zero_bit(inuse
, max_netdevices
);
997 free_page((unsigned long) inuse
);
1001 snprintf(buf
, IFNAMSIZ
, name
, i
);
1002 if (!__dev_get_by_name(net
, buf
))
1005 /* It is possible to run out of possible slots
1006 * when the name is long and there isn't enough space left
1007 * for the digits, or if all bits are used.
1013 * dev_alloc_name - allocate a name for a device
1015 * @name: name format string
1017 * Passed a format string - eg "lt%d" it will try and find a suitable
1018 * id. It scans list of devices to build up a free map, then chooses
1019 * the first empty slot. The caller must hold the dev_base or rtnl lock
1020 * while allocating the name and adding the device in order to avoid
1022 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1023 * Returns the number of the unit assigned or a negative errno code.
1026 int dev_alloc_name(struct net_device
*dev
, const char *name
)
1032 BUG_ON(!dev_net(dev
));
1034 ret
= __dev_alloc_name(net
, name
, buf
);
1036 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
1039 EXPORT_SYMBOL(dev_alloc_name
);
1041 static int dev_alloc_name_ns(struct net
*net
,
1042 struct net_device
*dev
,
1048 ret
= __dev_alloc_name(net
, name
, buf
);
1050 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
1054 static int dev_get_valid_name(struct net
*net
,
1055 struct net_device
*dev
,
1060 if (!dev_valid_name(name
))
1063 if (strchr(name
, '%'))
1064 return dev_alloc_name_ns(net
, dev
, name
);
1065 else if (__dev_get_by_name(net
, name
))
1067 else if (dev
->name
!= name
)
1068 strlcpy(dev
->name
, name
, IFNAMSIZ
);
1074 * dev_change_name - change name of a device
1076 * @newname: name (or format string) must be at least IFNAMSIZ
1078 * Change name of a device, can pass format strings "eth%d".
1081 int dev_change_name(struct net_device
*dev
, const char *newname
)
1083 char oldname
[IFNAMSIZ
];
1089 BUG_ON(!dev_net(dev
));
1092 if (dev
->flags
& IFF_UP
)
1095 write_seqcount_begin(&devnet_rename_seq
);
1097 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0) {
1098 write_seqcount_end(&devnet_rename_seq
);
1102 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
1104 err
= dev_get_valid_name(net
, dev
, newname
);
1106 write_seqcount_end(&devnet_rename_seq
);
1111 ret
= device_rename(&dev
->dev
, dev
->name
);
1113 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1114 write_seqcount_end(&devnet_rename_seq
);
1118 write_seqcount_end(&devnet_rename_seq
);
1120 write_lock_bh(&dev_base_lock
);
1121 hlist_del_rcu(&dev
->name_hlist
);
1122 write_unlock_bh(&dev_base_lock
);
1126 write_lock_bh(&dev_base_lock
);
1127 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
1128 write_unlock_bh(&dev_base_lock
);
1130 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
1131 ret
= notifier_to_errno(ret
);
1134 /* err >= 0 after dev_alloc_name() or stores the first errno */
1137 write_seqcount_begin(&devnet_rename_seq
);
1138 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1141 pr_err("%s: name change rollback failed: %d\n",
1150 * dev_set_alias - change ifalias of a device
1152 * @alias: name up to IFALIASZ
1153 * @len: limit of bytes to copy from info
1155 * Set ifalias for a device,
1157 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
1163 if (len
>= IFALIASZ
)
1167 kfree(dev
->ifalias
);
1168 dev
->ifalias
= NULL
;
1172 new_ifalias
= krealloc(dev
->ifalias
, len
+ 1, GFP_KERNEL
);
1175 dev
->ifalias
= new_ifalias
;
1177 strlcpy(dev
->ifalias
, alias
, len
+1);
1183 * netdev_features_change - device changes features
1184 * @dev: device to cause notification
1186 * Called to indicate a device has changed features.
1188 void netdev_features_change(struct net_device
*dev
)
1190 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1192 EXPORT_SYMBOL(netdev_features_change
);
1195 * netdev_state_change - device changes state
1196 * @dev: device to cause notification
1198 * Called to indicate a device has changed state. This function calls
1199 * the notifier chains for netdev_chain and sends a NEWLINK message
1200 * to the routing socket.
1202 void netdev_state_change(struct net_device
*dev
)
1204 if (dev
->flags
& IFF_UP
) {
1205 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
1206 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
1209 EXPORT_SYMBOL(netdev_state_change
);
1212 * netdev_notify_peers - notify network peers about existence of @dev
1213 * @dev: network device
1215 * Generate traffic such that interested network peers are aware of
1216 * @dev, such as by generating a gratuitous ARP. This may be used when
1217 * a device wants to inform the rest of the network about some sort of
1218 * reconfiguration such as a failover event or virtual machine
1221 void netdev_notify_peers(struct net_device
*dev
)
1224 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
, dev
);
1227 EXPORT_SYMBOL(netdev_notify_peers
);
1229 static int __dev_open(struct net_device
*dev
)
1231 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1236 if (!netif_device_present(dev
))
1239 /* Block netpoll from trying to do any rx path servicing.
1240 * If we don't do this there is a chance ndo_poll_controller
1241 * or ndo_poll may be running while we open the device
1243 netpoll_rx_disable(dev
);
1245 ret
= call_netdevice_notifiers(NETDEV_PRE_UP
, dev
);
1246 ret
= notifier_to_errno(ret
);
1250 set_bit(__LINK_STATE_START
, &dev
->state
);
1252 if (ops
->ndo_validate_addr
)
1253 ret
= ops
->ndo_validate_addr(dev
);
1255 if (!ret
&& ops
->ndo_open
)
1256 ret
= ops
->ndo_open(dev
);
1258 netpoll_rx_enable(dev
);
1261 clear_bit(__LINK_STATE_START
, &dev
->state
);
1263 dev
->flags
|= IFF_UP
;
1264 net_dmaengine_get();
1265 dev_set_rx_mode(dev
);
1267 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
1274 * dev_open - prepare an interface for use.
1275 * @dev: device to open
1277 * Takes a device from down to up state. The device's private open
1278 * function is invoked and then the multicast lists are loaded. Finally
1279 * the device is moved into the up state and a %NETDEV_UP message is
1280 * sent to the netdev notifier chain.
1282 * Calling this function on an active interface is a nop. On a failure
1283 * a negative errno code is returned.
1285 int dev_open(struct net_device
*dev
)
1289 if (dev
->flags
& IFF_UP
)
1292 ret
= __dev_open(dev
);
1296 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1297 call_netdevice_notifiers(NETDEV_UP
, dev
);
1301 EXPORT_SYMBOL(dev_open
);
1303 static int __dev_close_many(struct list_head
*head
)
1305 struct net_device
*dev
;
1310 list_for_each_entry(dev
, head
, unreg_list
) {
1311 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1313 clear_bit(__LINK_STATE_START
, &dev
->state
);
1315 /* Synchronize to scheduled poll. We cannot touch poll list, it
1316 * can be even on different cpu. So just clear netif_running().
1318 * dev->stop() will invoke napi_disable() on all of it's
1319 * napi_struct instances on this device.
1321 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1324 dev_deactivate_many(head
);
1326 list_for_each_entry(dev
, head
, unreg_list
) {
1327 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1330 * Call the device specific close. This cannot fail.
1331 * Only if device is UP
1333 * We allow it to be called even after a DETACH hot-plug
1339 dev
->flags
&= ~IFF_UP
;
1340 net_dmaengine_put();
1346 static int __dev_close(struct net_device
*dev
)
1351 /* Temporarily disable netpoll until the interface is down */
1352 netpoll_rx_disable(dev
);
1354 list_add(&dev
->unreg_list
, &single
);
1355 retval
= __dev_close_many(&single
);
1358 netpoll_rx_enable(dev
);
1362 static int dev_close_many(struct list_head
*head
)
1364 struct net_device
*dev
, *tmp
;
1365 LIST_HEAD(tmp_list
);
1367 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
)
1368 if (!(dev
->flags
& IFF_UP
))
1369 list_move(&dev
->unreg_list
, &tmp_list
);
1371 __dev_close_many(head
);
1373 list_for_each_entry(dev
, head
, unreg_list
) {
1374 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1375 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1378 /* rollback_registered_many needs the complete original list */
1379 list_splice(&tmp_list
, head
);
1384 * dev_close - shutdown an interface.
1385 * @dev: device to shutdown
1387 * This function moves an active device into down state. A
1388 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1389 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1392 int dev_close(struct net_device
*dev
)
1394 if (dev
->flags
& IFF_UP
) {
1397 /* Block netpoll rx while the interface is going down */
1398 netpoll_rx_disable(dev
);
1400 list_add(&dev
->unreg_list
, &single
);
1401 dev_close_many(&single
);
1404 netpoll_rx_enable(dev
);
1408 EXPORT_SYMBOL(dev_close
);
1412 * dev_disable_lro - disable Large Receive Offload on a device
1415 * Disable Large Receive Offload (LRO) on a net device. Must be
1416 * called under RTNL. This is needed if received packets may be
1417 * forwarded to another interface.
1419 void dev_disable_lro(struct net_device
*dev
)
1422 * If we're trying to disable lro on a vlan device
1423 * use the underlying physical device instead
1425 if (is_vlan_dev(dev
))
1426 dev
= vlan_dev_real_dev(dev
);
1428 dev
->wanted_features
&= ~NETIF_F_LRO
;
1429 netdev_update_features(dev
);
1431 if (unlikely(dev
->features
& NETIF_F_LRO
))
1432 netdev_WARN(dev
, "failed to disable LRO!\n");
1434 EXPORT_SYMBOL(dev_disable_lro
);
1436 static int call_netdevice_notifier(struct notifier_block
*nb
, unsigned long val
,
1437 struct net_device
*dev
)
1439 struct netdev_notifier_info info
;
1441 netdev_notifier_info_init(&info
, dev
);
1442 return nb
->notifier_call(nb
, val
, &info
);
1445 static int dev_boot_phase
= 1;
1448 * register_netdevice_notifier - register a network notifier block
1451 * Register a notifier to be called when network device events occur.
1452 * The notifier passed is linked into the kernel structures and must
1453 * not be reused until it has been unregistered. A negative errno code
1454 * is returned on a failure.
1456 * When registered all registration and up events are replayed
1457 * to the new notifier to allow device to have a race free
1458 * view of the network device list.
1461 int register_netdevice_notifier(struct notifier_block
*nb
)
1463 struct net_device
*dev
;
1464 struct net_device
*last
;
1469 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1475 for_each_netdev(net
, dev
) {
1476 err
= call_netdevice_notifier(nb
, NETDEV_REGISTER
, dev
);
1477 err
= notifier_to_errno(err
);
1481 if (!(dev
->flags
& IFF_UP
))
1484 call_netdevice_notifier(nb
, NETDEV_UP
, dev
);
1495 for_each_netdev(net
, dev
) {
1499 if (dev
->flags
& IFF_UP
) {
1500 call_netdevice_notifier(nb
, NETDEV_GOING_DOWN
,
1502 call_netdevice_notifier(nb
, NETDEV_DOWN
, dev
);
1504 call_netdevice_notifier(nb
, NETDEV_UNREGISTER
, dev
);
1509 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1512 EXPORT_SYMBOL(register_netdevice_notifier
);
1515 * unregister_netdevice_notifier - unregister a network notifier block
1518 * Unregister a notifier previously registered by
1519 * register_netdevice_notifier(). The notifier is unlinked into the
1520 * kernel structures and may then be reused. A negative errno code
1521 * is returned on a failure.
1523 * After unregistering unregister and down device events are synthesized
1524 * for all devices on the device list to the removed notifier to remove
1525 * the need for special case cleanup code.
1528 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1530 struct net_device
*dev
;
1535 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1540 for_each_netdev(net
, dev
) {
1541 if (dev
->flags
& IFF_UP
) {
1542 call_netdevice_notifier(nb
, NETDEV_GOING_DOWN
,
1544 call_netdevice_notifier(nb
, NETDEV_DOWN
, dev
);
1546 call_netdevice_notifier(nb
, NETDEV_UNREGISTER
, dev
);
1553 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1556 * call_netdevice_notifiers_info - call all network notifier blocks
1557 * @val: value passed unmodified to notifier function
1558 * @dev: net_device pointer passed unmodified to notifier function
1559 * @info: notifier information data
1561 * Call all network notifier blocks. Parameters and return value
1562 * are as for raw_notifier_call_chain().
1565 int call_netdevice_notifiers_info(unsigned long val
, struct net_device
*dev
,
1566 struct netdev_notifier_info
*info
)
1569 netdev_notifier_info_init(info
, dev
);
1570 return raw_notifier_call_chain(&netdev_chain
, val
, info
);
1572 EXPORT_SYMBOL(call_netdevice_notifiers_info
);
1575 * call_netdevice_notifiers - call all network notifier blocks
1576 * @val: value passed unmodified to notifier function
1577 * @dev: net_device pointer passed unmodified to notifier function
1579 * Call all network notifier blocks. Parameters and return value
1580 * are as for raw_notifier_call_chain().
1583 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1585 struct netdev_notifier_info info
;
1587 return call_netdevice_notifiers_info(val
, dev
, &info
);
1589 EXPORT_SYMBOL(call_netdevice_notifiers
);
1591 static struct static_key netstamp_needed __read_mostly
;
1592 #ifdef HAVE_JUMP_LABEL
1593 /* We are not allowed to call static_key_slow_dec() from irq context
1594 * If net_disable_timestamp() is called from irq context, defer the
1595 * static_key_slow_dec() calls.
1597 static atomic_t netstamp_needed_deferred
;
1600 void net_enable_timestamp(void)
1602 #ifdef HAVE_JUMP_LABEL
1603 int deferred
= atomic_xchg(&netstamp_needed_deferred
, 0);
1607 static_key_slow_dec(&netstamp_needed
);
1611 static_key_slow_inc(&netstamp_needed
);
1613 EXPORT_SYMBOL(net_enable_timestamp
);
1615 void net_disable_timestamp(void)
1617 #ifdef HAVE_JUMP_LABEL
1618 if (in_interrupt()) {
1619 atomic_inc(&netstamp_needed_deferred
);
1623 static_key_slow_dec(&netstamp_needed
);
1625 EXPORT_SYMBOL(net_disable_timestamp
);
1627 static inline void net_timestamp_set(struct sk_buff
*skb
)
1629 skb
->tstamp
.tv64
= 0;
1630 if (static_key_false(&netstamp_needed
))
1631 __net_timestamp(skb
);
1634 #define net_timestamp_check(COND, SKB) \
1635 if (static_key_false(&netstamp_needed)) { \
1636 if ((COND) && !(SKB)->tstamp.tv64) \
1637 __net_timestamp(SKB); \
1640 static inline bool is_skb_forwardable(struct net_device *dev,
1641 struct sk_buff
*skb
)
1645 if (!(dev
->flags
& IFF_UP
))
1648 len
= dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
;
1649 if (skb
->len
<= len
)
1652 /* if TSO is enabled, we don't care about the length as the packet
1653 * could be forwarded without being segmented before
1655 if (skb_is_gso(skb
))
1662 * dev_forward_skb - loopback an skb to another netif
1664 * @dev: destination network device
1665 * @skb: buffer to forward
1668 * NET_RX_SUCCESS (no congestion)
1669 * NET_RX_DROP (packet was dropped, but freed)
1671 * dev_forward_skb can be used for injecting an skb from the
1672 * start_xmit function of one device into the receive queue
1673 * of another device.
1675 * The receiving device may be in another namespace, so
1676 * we have to clear all information in the skb that could
1677 * impact namespace isolation.
1679 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1681 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
1682 if (skb_copy_ubufs(skb
, GFP_ATOMIC
)) {
1683 atomic_long_inc(&dev
->rx_dropped
);
1689 if (unlikely(!is_skb_forwardable(dev
, skb
))) {
1690 atomic_long_inc(&dev
->rx_dropped
);
1694 skb
->protocol
= eth_type_trans(skb
, dev
);
1696 /* eth_type_trans() can set pkt_type.
1697 * call skb_scrub_packet() after it to clear pkt_type _after_ calling
1700 skb_scrub_packet(skb
, true);
1702 return netif_rx(skb
);
1704 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1706 static inline int deliver_skb(struct sk_buff
*skb
,
1707 struct packet_type
*pt_prev
,
1708 struct net_device
*orig_dev
)
1710 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
1712 atomic_inc(&skb
->users
);
1713 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1716 static inline bool skb_loop_sk(struct packet_type
*ptype
, struct sk_buff
*skb
)
1718 if (!ptype
->af_packet_priv
|| !skb
->sk
)
1721 if (ptype
->id_match
)
1722 return ptype
->id_match(ptype
, skb
->sk
);
1723 else if ((struct sock
*)ptype
->af_packet_priv
== skb
->sk
)
1730 * Support routine. Sends outgoing frames to any network
1731 * taps currently in use.
1734 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1736 struct packet_type
*ptype
;
1737 struct sk_buff
*skb2
= NULL
;
1738 struct packet_type
*pt_prev
= NULL
;
1741 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1742 /* Never send packets back to the socket
1743 * they originated from - MvS (miquels@drinkel.ow.org)
1745 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1746 (!skb_loop_sk(ptype
, skb
))) {
1748 deliver_skb(skb2
, pt_prev
, skb
->dev
);
1753 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1757 net_timestamp_set(skb2
);
1759 /* skb->nh should be correctly
1760 set by sender, so that the second statement is
1761 just protection against buggy protocols.
1763 skb_reset_mac_header(skb2
);
1765 if (skb_network_header(skb2
) < skb2
->data
||
1766 skb_network_header(skb2
) > skb_tail_pointer(skb2
)) {
1767 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1768 ntohs(skb2
->protocol
),
1770 skb_reset_network_header(skb2
);
1773 skb2
->transport_header
= skb2
->network_header
;
1774 skb2
->pkt_type
= PACKET_OUTGOING
;
1779 pt_prev
->func(skb2
, skb
->dev
, pt_prev
, skb
->dev
);
1784 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1785 * @dev: Network device
1786 * @txq: number of queues available
1788 * If real_num_tx_queues is changed the tc mappings may no longer be
1789 * valid. To resolve this verify the tc mapping remains valid and if
1790 * not NULL the mapping. With no priorities mapping to this
1791 * offset/count pair it will no longer be used. In the worst case TC0
1792 * is invalid nothing can be done so disable priority mappings. If is
1793 * expected that drivers will fix this mapping if they can before
1794 * calling netif_set_real_num_tx_queues.
1796 static void netif_setup_tc(struct net_device
*dev
, unsigned int txq
)
1799 struct netdev_tc_txq
*tc
= &dev
->tc_to_txq
[0];
1801 /* If TC0 is invalidated disable TC mapping */
1802 if (tc
->offset
+ tc
->count
> txq
) {
1803 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1808 /* Invalidated prio to tc mappings set to TC0 */
1809 for (i
= 1; i
< TC_BITMASK
+ 1; i
++) {
1810 int q
= netdev_get_prio_tc_map(dev
, i
);
1812 tc
= &dev
->tc_to_txq
[q
];
1813 if (tc
->offset
+ tc
->count
> txq
) {
1814 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1816 netdev_set_prio_tc_map(dev
, i
, 0);
1822 static DEFINE_MUTEX(xps_map_mutex
);
1823 #define xmap_dereference(P) \
1824 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1826 static struct xps_map
*remove_xps_queue(struct xps_dev_maps
*dev_maps
,
1829 struct xps_map
*map
= NULL
;
1833 map
= xmap_dereference(dev_maps
->cpu_map
[cpu
]);
1835 for (pos
= 0; map
&& pos
< map
->len
; pos
++) {
1836 if (map
->queues
[pos
] == index
) {
1838 map
->queues
[pos
] = map
->queues
[--map
->len
];
1840 RCU_INIT_POINTER(dev_maps
->cpu_map
[cpu
], NULL
);
1841 kfree_rcu(map
, rcu
);
1851 static void netif_reset_xps_queues_gt(struct net_device
*dev
, u16 index
)
1853 struct xps_dev_maps
*dev_maps
;
1855 bool active
= false;
1857 mutex_lock(&xps_map_mutex
);
1858 dev_maps
= xmap_dereference(dev
->xps_maps
);
1863 for_each_possible_cpu(cpu
) {
1864 for (i
= index
; i
< dev
->num_tx_queues
; i
++) {
1865 if (!remove_xps_queue(dev_maps
, cpu
, i
))
1868 if (i
== dev
->num_tx_queues
)
1873 RCU_INIT_POINTER(dev
->xps_maps
, NULL
);
1874 kfree_rcu(dev_maps
, rcu
);
1877 for (i
= index
; i
< dev
->num_tx_queues
; i
++)
1878 netdev_queue_numa_node_write(netdev_get_tx_queue(dev
, i
),
1882 mutex_unlock(&xps_map_mutex
);
1885 static struct xps_map
*expand_xps_map(struct xps_map
*map
,
1888 struct xps_map
*new_map
;
1889 int alloc_len
= XPS_MIN_MAP_ALLOC
;
1892 for (pos
= 0; map
&& pos
< map
->len
; pos
++) {
1893 if (map
->queues
[pos
] != index
)
1898 /* Need to add queue to this CPU's existing map */
1900 if (pos
< map
->alloc_len
)
1903 alloc_len
= map
->alloc_len
* 2;
1906 /* Need to allocate new map to store queue on this CPU's map */
1907 new_map
= kzalloc_node(XPS_MAP_SIZE(alloc_len
), GFP_KERNEL
,
1912 for (i
= 0; i
< pos
; i
++)
1913 new_map
->queues
[i
] = map
->queues
[i
];
1914 new_map
->alloc_len
= alloc_len
;
1920 int netif_set_xps_queue(struct net_device
*dev
, const struct cpumask
*mask
,
1923 struct xps_dev_maps
*dev_maps
, *new_dev_maps
= NULL
;
1924 struct xps_map
*map
, *new_map
;
1925 int maps_sz
= max_t(unsigned int, XPS_DEV_MAPS_SIZE
, L1_CACHE_BYTES
);
1926 int cpu
, numa_node_id
= -2;
1927 bool active
= false;
1929 mutex_lock(&xps_map_mutex
);
1931 dev_maps
= xmap_dereference(dev
->xps_maps
);
1933 /* allocate memory for queue storage */
1934 for_each_online_cpu(cpu
) {
1935 if (!cpumask_test_cpu(cpu
, mask
))
1939 new_dev_maps
= kzalloc(maps_sz
, GFP_KERNEL
);
1940 if (!new_dev_maps
) {
1941 mutex_unlock(&xps_map_mutex
);
1945 map
= dev_maps
? xmap_dereference(dev_maps
->cpu_map
[cpu
]) :
1948 map
= expand_xps_map(map
, cpu
, index
);
1952 RCU_INIT_POINTER(new_dev_maps
->cpu_map
[cpu
], map
);
1956 goto out_no_new_maps
;
1958 for_each_possible_cpu(cpu
) {
1959 if (cpumask_test_cpu(cpu
, mask
) && cpu_online(cpu
)) {
1960 /* add queue to CPU maps */
1963 map
= xmap_dereference(new_dev_maps
->cpu_map
[cpu
]);
1964 while ((pos
< map
->len
) && (map
->queues
[pos
] != index
))
1967 if (pos
== map
->len
)
1968 map
->queues
[map
->len
++] = index
;
1970 if (numa_node_id
== -2)
1971 numa_node_id
= cpu_to_node(cpu
);
1972 else if (numa_node_id
!= cpu_to_node(cpu
))
1975 } else if (dev_maps
) {
1976 /* fill in the new device map from the old device map */
1977 map
= xmap_dereference(dev_maps
->cpu_map
[cpu
]);
1978 RCU_INIT_POINTER(new_dev_maps
->cpu_map
[cpu
], map
);
1983 rcu_assign_pointer(dev
->xps_maps
, new_dev_maps
);
1985 /* Cleanup old maps */
1987 for_each_possible_cpu(cpu
) {
1988 new_map
= xmap_dereference(new_dev_maps
->cpu_map
[cpu
]);
1989 map
= xmap_dereference(dev_maps
->cpu_map
[cpu
]);
1990 if (map
&& map
!= new_map
)
1991 kfree_rcu(map
, rcu
);
1994 kfree_rcu(dev_maps
, rcu
);
1997 dev_maps
= new_dev_maps
;
2001 /* update Tx queue numa node */
2002 netdev_queue_numa_node_write(netdev_get_tx_queue(dev
, index
),
2003 (numa_node_id
>= 0) ? numa_node_id
:
2009 /* removes queue from unused CPUs */
2010 for_each_possible_cpu(cpu
) {
2011 if (cpumask_test_cpu(cpu
, mask
) && cpu_online(cpu
))
2014 if (remove_xps_queue(dev_maps
, cpu
, index
))
2018 /* free map if not active */
2020 RCU_INIT_POINTER(dev
->xps_maps
, NULL
);
2021 kfree_rcu(dev_maps
, rcu
);
2025 mutex_unlock(&xps_map_mutex
);
2029 /* remove any maps that we added */
2030 for_each_possible_cpu(cpu
) {
2031 new_map
= xmap_dereference(new_dev_maps
->cpu_map
[cpu
]);
2032 map
= dev_maps
? xmap_dereference(dev_maps
->cpu_map
[cpu
]) :
2034 if (new_map
&& new_map
!= map
)
2038 mutex_unlock(&xps_map_mutex
);
2040 kfree(new_dev_maps
);
2043 EXPORT_SYMBOL(netif_set_xps_queue
);
2047 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2048 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2050 int netif_set_real_num_tx_queues(struct net_device
*dev
, unsigned int txq
)
2054 if (txq
< 1 || txq
> dev
->num_tx_queues
)
2057 if (dev
->reg_state
== NETREG_REGISTERED
||
2058 dev
->reg_state
== NETREG_UNREGISTERING
) {
2061 rc
= netdev_queue_update_kobjects(dev
, dev
->real_num_tx_queues
,
2067 netif_setup_tc(dev
, txq
);
2069 if (txq
< dev
->real_num_tx_queues
) {
2070 qdisc_reset_all_tx_gt(dev
, txq
);
2072 netif_reset_xps_queues_gt(dev
, txq
);
2077 dev
->real_num_tx_queues
= txq
;
2080 EXPORT_SYMBOL(netif_set_real_num_tx_queues
);
2084 * netif_set_real_num_rx_queues - set actual number of RX queues used
2085 * @dev: Network device
2086 * @rxq: Actual number of RX queues
2088 * This must be called either with the rtnl_lock held or before
2089 * registration of the net device. Returns 0 on success, or a
2090 * negative error code. If called before registration, it always
2093 int netif_set_real_num_rx_queues(struct net_device
*dev
, unsigned int rxq
)
2097 if (rxq
< 1 || rxq
> dev
->num_rx_queues
)
2100 if (dev
->reg_state
== NETREG_REGISTERED
) {
2103 rc
= net_rx_queue_update_kobjects(dev
, dev
->real_num_rx_queues
,
2109 dev
->real_num_rx_queues
= rxq
;
2112 EXPORT_SYMBOL(netif_set_real_num_rx_queues
);
2116 * netif_get_num_default_rss_queues - default number of RSS queues
2118 * This routine should set an upper limit on the number of RSS queues
2119 * used by default by multiqueue devices.
2121 int netif_get_num_default_rss_queues(void)
2123 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES
, num_online_cpus());
2125 EXPORT_SYMBOL(netif_get_num_default_rss_queues
);
2127 static inline void __netif_reschedule(struct Qdisc
*q
)
2129 struct softnet_data
*sd
;
2130 unsigned long flags
;
2132 local_irq_save(flags
);
2133 sd
= &__get_cpu_var(softnet_data
);
2134 q
->next_sched
= NULL
;
2135 *sd
->output_queue_tailp
= q
;
2136 sd
->output_queue_tailp
= &q
->next_sched
;
2137 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
2138 local_irq_restore(flags
);
2141 void __netif_schedule(struct Qdisc
*q
)
2143 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
2144 __netif_reschedule(q
);
2146 EXPORT_SYMBOL(__netif_schedule
);
2148 void dev_kfree_skb_irq(struct sk_buff
*skb
)
2150 if (atomic_dec_and_test(&skb
->users
)) {
2151 struct softnet_data
*sd
;
2152 unsigned long flags
;
2154 local_irq_save(flags
);
2155 sd
= &__get_cpu_var(softnet_data
);
2156 skb
->next
= sd
->completion_queue
;
2157 sd
->completion_queue
= skb
;
2158 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
2159 local_irq_restore(flags
);
2162 EXPORT_SYMBOL(dev_kfree_skb_irq
);
2164 void dev_kfree_skb_any(struct sk_buff
*skb
)
2166 if (in_irq() || irqs_disabled())
2167 dev_kfree_skb_irq(skb
);
2171 EXPORT_SYMBOL(dev_kfree_skb_any
);
2175 * netif_device_detach - mark device as removed
2176 * @dev: network device
2178 * Mark device as removed from system and therefore no longer available.
2180 void netif_device_detach(struct net_device
*dev
)
2182 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
2183 netif_running(dev
)) {
2184 netif_tx_stop_all_queues(dev
);
2187 EXPORT_SYMBOL(netif_device_detach
);
2190 * netif_device_attach - mark device as attached
2191 * @dev: network device
2193 * Mark device as attached from system and restart if needed.
2195 void netif_device_attach(struct net_device
*dev
)
2197 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
2198 netif_running(dev
)) {
2199 netif_tx_wake_all_queues(dev
);
2200 __netdev_watchdog_up(dev
);
2203 EXPORT_SYMBOL(netif_device_attach
);
2205 static void skb_warn_bad_offload(const struct sk_buff
*skb
)
2207 static const netdev_features_t null_features
= 0;
2208 struct net_device
*dev
= skb
->dev
;
2209 const char *driver
= "";
2211 if (!net_ratelimit())
2214 if (dev
&& dev
->dev
.parent
)
2215 driver
= dev_driver_string(dev
->dev
.parent
);
2217 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2218 "gso_type=%d ip_summed=%d\n",
2219 driver
, dev
? &dev
->features
: &null_features
,
2220 skb
->sk
? &skb
->sk
->sk_route_caps
: &null_features
,
2221 skb
->len
, skb
->data_len
, skb_shinfo(skb
)->gso_size
,
2222 skb_shinfo(skb
)->gso_type
, skb
->ip_summed
);
2226 * Invalidate hardware checksum when packet is to be mangled, and
2227 * complete checksum manually on outgoing path.
2229 int skb_checksum_help(struct sk_buff
*skb
)
2232 int ret
= 0, offset
;
2234 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
2235 goto out_set_summed
;
2237 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
2238 skb_warn_bad_offload(skb
);
2242 /* Before computing a checksum, we should make sure no frag could
2243 * be modified by an external entity : checksum could be wrong.
2245 if (skb_has_shared_frag(skb
)) {
2246 ret
= __skb_linearize(skb
);
2251 offset
= skb_checksum_start_offset(skb
);
2252 BUG_ON(offset
>= skb_headlen(skb
));
2253 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
2255 offset
+= skb
->csum_offset
;
2256 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
2258 if (skb_cloned(skb
) &&
2259 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
2260 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2265 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
2267 skb
->ip_summed
= CHECKSUM_NONE
;
2271 EXPORT_SYMBOL(skb_checksum_help
);
2273 __be16
skb_network_protocol(struct sk_buff
*skb
)
2275 __be16 type
= skb
->protocol
;
2276 int vlan_depth
= ETH_HLEN
;
2278 /* Tunnel gso handlers can set protocol to ethernet. */
2279 if (type
== htons(ETH_P_TEB
)) {
2282 if (unlikely(!pskb_may_pull(skb
, sizeof(struct ethhdr
))))
2285 eth
= (struct ethhdr
*)skb_mac_header(skb
);
2286 type
= eth
->h_proto
;
2289 while (type
== htons(ETH_P_8021Q
) || type
== htons(ETH_P_8021AD
)) {
2290 struct vlan_hdr
*vh
;
2292 if (unlikely(!pskb_may_pull(skb
, vlan_depth
+ VLAN_HLEN
)))
2295 vh
= (struct vlan_hdr
*)(skb
->data
+ vlan_depth
);
2296 type
= vh
->h_vlan_encapsulated_proto
;
2297 vlan_depth
+= VLAN_HLEN
;
2304 * skb_mac_gso_segment - mac layer segmentation handler.
2305 * @skb: buffer to segment
2306 * @features: features for the output path (see dev->features)
2308 struct sk_buff
*skb_mac_gso_segment(struct sk_buff
*skb
,
2309 netdev_features_t features
)
2311 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
2312 struct packet_offload
*ptype
;
2313 __be16 type
= skb_network_protocol(skb
);
2315 if (unlikely(!type
))
2316 return ERR_PTR(-EINVAL
);
2318 __skb_pull(skb
, skb
->mac_len
);
2321 list_for_each_entry_rcu(ptype
, &offload_base
, list
) {
2322 if (ptype
->type
== type
&& ptype
->callbacks
.gso_segment
) {
2323 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
2326 err
= ptype
->callbacks
.gso_send_check(skb
);
2327 segs
= ERR_PTR(err
);
2328 if (err
|| skb_gso_ok(skb
, features
))
2330 __skb_push(skb
, (skb
->data
-
2331 skb_network_header(skb
)));
2333 segs
= ptype
->callbacks
.gso_segment(skb
, features
);
2339 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
2343 EXPORT_SYMBOL(skb_mac_gso_segment
);
2346 /* openvswitch calls this on rx path, so we need a different check.
2348 static inline bool skb_needs_check(struct sk_buff
*skb
, bool tx_path
)
2351 return skb
->ip_summed
!= CHECKSUM_PARTIAL
;
2353 return skb
->ip_summed
== CHECKSUM_NONE
;
2357 * __skb_gso_segment - Perform segmentation on skb.
2358 * @skb: buffer to segment
2359 * @features: features for the output path (see dev->features)
2360 * @tx_path: whether it is called in TX path
2362 * This function segments the given skb and returns a list of segments.
2364 * It may return NULL if the skb requires no segmentation. This is
2365 * only possible when GSO is used for verifying header integrity.
2367 struct sk_buff
*__skb_gso_segment(struct sk_buff
*skb
,
2368 netdev_features_t features
, bool tx_path
)
2370 if (unlikely(skb_needs_check(skb
, tx_path
))) {
2373 skb_warn_bad_offload(skb
);
2375 if (skb_header_cloned(skb
) &&
2376 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
2377 return ERR_PTR(err
);
2380 SKB_GSO_CB(skb
)->mac_offset
= skb_headroom(skb
);
2381 skb_reset_mac_header(skb
);
2382 skb_reset_mac_len(skb
);
2384 return skb_mac_gso_segment(skb
, features
);
2386 EXPORT_SYMBOL(__skb_gso_segment
);
2388 /* Take action when hardware reception checksum errors are detected. */
2390 void netdev_rx_csum_fault(struct net_device
*dev
)
2392 if (net_ratelimit()) {
2393 pr_err("%s: hw csum failure\n", dev
? dev
->name
: "<unknown>");
2397 EXPORT_SYMBOL(netdev_rx_csum_fault
);
2400 /* Actually, we should eliminate this check as soon as we know, that:
2401 * 1. IOMMU is present and allows to map all the memory.
2402 * 2. No high memory really exists on this machine.
2405 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
2407 #ifdef CONFIG_HIGHMEM
2409 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
2410 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2411 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2412 if (PageHighMem(skb_frag_page(frag
)))
2417 if (PCI_DMA_BUS_IS_PHYS
) {
2418 struct device
*pdev
= dev
->dev
.parent
;
2422 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2423 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2424 dma_addr_t addr
= page_to_phys(skb_frag_page(frag
));
2425 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
2434 void (*destructor
)(struct sk_buff
*skb
);
2437 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2439 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
2441 struct dev_gso_cb
*cb
;
2444 struct sk_buff
*nskb
= skb
->next
;
2446 skb
->next
= nskb
->next
;
2449 } while (skb
->next
);
2451 cb
= DEV_GSO_CB(skb
);
2453 cb
->destructor(skb
);
2457 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2458 * @skb: buffer to segment
2459 * @features: device features as applicable to this skb
2461 * This function segments the given skb and stores the list of segments
2464 static int dev_gso_segment(struct sk_buff
*skb
, netdev_features_t features
)
2466 struct sk_buff
*segs
;
2468 segs
= skb_gso_segment(skb
, features
);
2470 /* Verifying header integrity only. */
2475 return PTR_ERR(segs
);
2478 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
2479 skb
->destructor
= dev_gso_skb_destructor
;
2484 static netdev_features_t
harmonize_features(struct sk_buff
*skb
,
2485 netdev_features_t features
)
2487 if (skb
->ip_summed
!= CHECKSUM_NONE
&&
2488 !can_checksum_protocol(features
, skb_network_protocol(skb
))) {
2489 features
&= ~NETIF_F_ALL_CSUM
;
2490 } else if (illegal_highdma(skb
->dev
, skb
)) {
2491 features
&= ~NETIF_F_SG
;
2497 netdev_features_t
netif_skb_features(struct sk_buff
*skb
)
2499 __be16 protocol
= skb
->protocol
;
2500 netdev_features_t features
= skb
->dev
->features
;
2502 if (skb_shinfo(skb
)->gso_segs
> skb
->dev
->gso_max_segs
)
2503 features
&= ~NETIF_F_GSO_MASK
;
2505 if (protocol
== htons(ETH_P_8021Q
) || protocol
== htons(ETH_P_8021AD
)) {
2506 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
2507 protocol
= veh
->h_vlan_encapsulated_proto
;
2508 } else if (!vlan_tx_tag_present(skb
)) {
2509 return harmonize_features(skb
, features
);
2512 features
&= (skb
->dev
->vlan_features
| NETIF_F_HW_VLAN_CTAG_TX
|
2513 NETIF_F_HW_VLAN_STAG_TX
);
2515 if (protocol
== htons(ETH_P_8021Q
) || protocol
== htons(ETH_P_8021AD
))
2516 features
&= NETIF_F_SG
| NETIF_F_HIGHDMA
| NETIF_F_FRAGLIST
|
2517 NETIF_F_GEN_CSUM
| NETIF_F_HW_VLAN_CTAG_TX
|
2518 NETIF_F_HW_VLAN_STAG_TX
;
2520 return harmonize_features(skb
, features
);
2522 EXPORT_SYMBOL(netif_skb_features
);
2525 * Returns true if either:
2526 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2527 * 2. skb is fragmented and the device does not support SG.
2529 static inline int skb_needs_linearize(struct sk_buff
*skb
,
2530 netdev_features_t features
)
2532 return skb_is_nonlinear(skb
) &&
2533 ((skb_has_frag_list(skb
) &&
2534 !(features
& NETIF_F_FRAGLIST
)) ||
2535 (skb_shinfo(skb
)->nr_frags
&&
2536 !(features
& NETIF_F_SG
)));
2539 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
2540 struct netdev_queue
*txq
)
2542 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2543 int rc
= NETDEV_TX_OK
;
2544 unsigned int skb_len
;
2546 if (likely(!skb
->next
)) {
2547 netdev_features_t features
;
2550 * If device doesn't need skb->dst, release it right now while
2551 * its hot in this cpu cache
2553 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2556 features
= netif_skb_features(skb
);
2558 if (vlan_tx_tag_present(skb
) &&
2559 !vlan_hw_offload_capable(features
, skb
->vlan_proto
)) {
2560 skb
= __vlan_put_tag(skb
, skb
->vlan_proto
,
2561 vlan_tx_tag_get(skb
));
2568 /* If encapsulation offload request, verify we are testing
2569 * hardware encapsulation features instead of standard
2570 * features for the netdev
2572 if (skb
->encapsulation
)
2573 features
&= dev
->hw_enc_features
;
2575 if (netif_needs_gso(skb
, features
)) {
2576 if (unlikely(dev_gso_segment(skb
, features
)))
2581 if (skb_needs_linearize(skb
, features
) &&
2582 __skb_linearize(skb
))
2585 /* If packet is not checksummed and device does not
2586 * support checksumming for this protocol, complete
2587 * checksumming here.
2589 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2590 if (skb
->encapsulation
)
2591 skb_set_inner_transport_header(skb
,
2592 skb_checksum_start_offset(skb
));
2594 skb_set_transport_header(skb
,
2595 skb_checksum_start_offset(skb
));
2596 if (!(features
& NETIF_F_ALL_CSUM
) &&
2597 skb_checksum_help(skb
))
2602 if (!list_empty(&ptype_all
))
2603 dev_queue_xmit_nit(skb
, dev
);
2606 rc
= ops
->ndo_start_xmit(skb
, dev
);
2607 trace_net_dev_xmit(skb
, rc
, dev
, skb_len
);
2608 if (rc
== NETDEV_TX_OK
)
2609 txq_trans_update(txq
);
2615 struct sk_buff
*nskb
= skb
->next
;
2617 skb
->next
= nskb
->next
;
2620 if (!list_empty(&ptype_all
))
2621 dev_queue_xmit_nit(nskb
, dev
);
2623 skb_len
= nskb
->len
;
2624 rc
= ops
->ndo_start_xmit(nskb
, dev
);
2625 trace_net_dev_xmit(nskb
, rc
, dev
, skb_len
);
2626 if (unlikely(rc
!= NETDEV_TX_OK
)) {
2627 if (rc
& ~NETDEV_TX_MASK
)
2628 goto out_kfree_gso_skb
;
2629 nskb
->next
= skb
->next
;
2633 txq_trans_update(txq
);
2634 if (unlikely(netif_xmit_stopped(txq
) && skb
->next
))
2635 return NETDEV_TX_BUSY
;
2636 } while (skb
->next
);
2639 if (likely(skb
->next
== NULL
)) {
2640 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
2650 static void qdisc_pkt_len_init(struct sk_buff
*skb
)
2652 const struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
2654 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
2656 /* To get more precise estimation of bytes sent on wire,
2657 * we add to pkt_len the headers size of all segments
2659 if (shinfo
->gso_size
) {
2660 unsigned int hdr_len
;
2661 u16 gso_segs
= shinfo
->gso_segs
;
2663 /* mac layer + network layer */
2664 hdr_len
= skb_transport_header(skb
) - skb_mac_header(skb
);
2666 /* + transport layer */
2667 if (likely(shinfo
->gso_type
& (SKB_GSO_TCPV4
| SKB_GSO_TCPV6
)))
2668 hdr_len
+= tcp_hdrlen(skb
);
2670 hdr_len
+= sizeof(struct udphdr
);
2672 if (shinfo
->gso_type
& SKB_GSO_DODGY
)
2673 gso_segs
= DIV_ROUND_UP(skb
->len
- hdr_len
,
2676 qdisc_skb_cb(skb
)->pkt_len
+= (gso_segs
- 1) * hdr_len
;
2680 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2681 struct net_device
*dev
,
2682 struct netdev_queue
*txq
)
2684 spinlock_t
*root_lock
= qdisc_lock(q
);
2688 qdisc_pkt_len_init(skb
);
2689 qdisc_calculate_pkt_len(skb
, q
);
2691 * Heuristic to force contended enqueues to serialize on a
2692 * separate lock before trying to get qdisc main lock.
2693 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2694 * and dequeue packets faster.
2696 contended
= qdisc_is_running(q
);
2697 if (unlikely(contended
))
2698 spin_lock(&q
->busylock
);
2700 spin_lock(root_lock
);
2701 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2704 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2705 qdisc_run_begin(q
)) {
2707 * This is a work-conserving queue; there are no old skbs
2708 * waiting to be sent out; and the qdisc is not running -
2709 * xmit the skb directly.
2711 if (!(dev
->priv_flags
& IFF_XMIT_DST_RELEASE
))
2714 qdisc_bstats_update(q
, skb
);
2716 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
)) {
2717 if (unlikely(contended
)) {
2718 spin_unlock(&q
->busylock
);
2725 rc
= NET_XMIT_SUCCESS
;
2728 rc
= q
->enqueue(skb
, q
) & NET_XMIT_MASK
;
2729 if (qdisc_run_begin(q
)) {
2730 if (unlikely(contended
)) {
2731 spin_unlock(&q
->busylock
);
2737 spin_unlock(root_lock
);
2738 if (unlikely(contended
))
2739 spin_unlock(&q
->busylock
);
2743 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2744 static void skb_update_prio(struct sk_buff
*skb
)
2746 struct netprio_map
*map
= rcu_dereference_bh(skb
->dev
->priomap
);
2748 if (!skb
->priority
&& skb
->sk
&& map
) {
2749 unsigned int prioidx
= skb
->sk
->sk_cgrp_prioidx
;
2751 if (prioidx
< map
->priomap_len
)
2752 skb
->priority
= map
->priomap
[prioidx
];
2756 #define skb_update_prio(skb)
2759 static DEFINE_PER_CPU(int, xmit_recursion
);
2760 #define RECURSION_LIMIT 10
2763 * dev_loopback_xmit - loop back @skb
2764 * @skb: buffer to transmit
2766 int dev_loopback_xmit(struct sk_buff
*skb
)
2768 skb_reset_mac_header(skb
);
2769 __skb_pull(skb
, skb_network_offset(skb
));
2770 skb
->pkt_type
= PACKET_LOOPBACK
;
2771 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
2772 WARN_ON(!skb_dst(skb
));
2777 EXPORT_SYMBOL(dev_loopback_xmit
);
2780 * dev_queue_xmit - transmit a buffer
2781 * @skb: buffer to transmit
2783 * Queue a buffer for transmission to a network device. The caller must
2784 * have set the device and priority and built the buffer before calling
2785 * this function. The function can be called from an interrupt.
2787 * A negative errno code is returned on a failure. A success does not
2788 * guarantee the frame will be transmitted as it may be dropped due
2789 * to congestion or traffic shaping.
2791 * -----------------------------------------------------------------------------------
2792 * I notice this method can also return errors from the queue disciplines,
2793 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2796 * Regardless of the return value, the skb is consumed, so it is currently
2797 * difficult to retry a send to this method. (You can bump the ref count
2798 * before sending to hold a reference for retry if you are careful.)
2800 * When calling this method, interrupts MUST be enabled. This is because
2801 * the BH enable code must have IRQs enabled so that it will not deadlock.
2804 int dev_queue_xmit(struct sk_buff
*skb
)
2806 struct net_device
*dev
= skb
->dev
;
2807 struct netdev_queue
*txq
;
2811 skb_reset_mac_header(skb
);
2813 /* Disable soft irqs for various locks below. Also
2814 * stops preemption for RCU.
2818 skb_update_prio(skb
);
2820 txq
= netdev_pick_tx(dev
, skb
);
2821 q
= rcu_dereference_bh(txq
->qdisc
);
2823 #ifdef CONFIG_NET_CLS_ACT
2824 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2826 trace_net_dev_queue(skb
);
2828 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2832 /* The device has no queue. Common case for software devices:
2833 loopback, all the sorts of tunnels...
2835 Really, it is unlikely that netif_tx_lock protection is necessary
2836 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2838 However, it is possible, that they rely on protection
2841 Check this and shot the lock. It is not prone from deadlocks.
2842 Either shot noqueue qdisc, it is even simpler 8)
2844 if (dev
->flags
& IFF_UP
) {
2845 int cpu
= smp_processor_id(); /* ok because BHs are off */
2847 if (txq
->xmit_lock_owner
!= cpu
) {
2849 if (__this_cpu_read(xmit_recursion
) > RECURSION_LIMIT
)
2850 goto recursion_alert
;
2852 HARD_TX_LOCK(dev
, txq
, cpu
);
2854 if (!netif_xmit_stopped(txq
)) {
2855 __this_cpu_inc(xmit_recursion
);
2856 rc
= dev_hard_start_xmit(skb
, dev
, txq
);
2857 __this_cpu_dec(xmit_recursion
);
2858 if (dev_xmit_complete(rc
)) {
2859 HARD_TX_UNLOCK(dev
, txq
);
2863 HARD_TX_UNLOCK(dev
, txq
);
2864 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2867 /* Recursion is detected! It is possible,
2871 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2877 rcu_read_unlock_bh();
2882 rcu_read_unlock_bh();
2885 EXPORT_SYMBOL(dev_queue_xmit
);
2888 /*=======================================================================
2890 =======================================================================*/
2892 int netdev_max_backlog __read_mostly
= 1000;
2893 EXPORT_SYMBOL(netdev_max_backlog
);
2895 int netdev_tstamp_prequeue __read_mostly
= 1;
2896 int netdev_budget __read_mostly
= 300;
2897 int weight_p __read_mostly
= 64; /* old backlog weight */
2899 /* Called with irq disabled */
2900 static inline void ____napi_schedule(struct softnet_data
*sd
,
2901 struct napi_struct
*napi
)
2903 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
2904 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2909 /* One global table that all flow-based protocols share. */
2910 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
2911 EXPORT_SYMBOL(rps_sock_flow_table
);
2913 struct static_key rps_needed __read_mostly
;
2915 static struct rps_dev_flow
*
2916 set_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2917 struct rps_dev_flow
*rflow
, u16 next_cpu
)
2919 if (next_cpu
!= RPS_NO_CPU
) {
2920 #ifdef CONFIG_RFS_ACCEL
2921 struct netdev_rx_queue
*rxqueue
;
2922 struct rps_dev_flow_table
*flow_table
;
2923 struct rps_dev_flow
*old_rflow
;
2928 /* Should we steer this flow to a different hardware queue? */
2929 if (!skb_rx_queue_recorded(skb
) || !dev
->rx_cpu_rmap
||
2930 !(dev
->features
& NETIF_F_NTUPLE
))
2932 rxq_index
= cpu_rmap_lookup_index(dev
->rx_cpu_rmap
, next_cpu
);
2933 if (rxq_index
== skb_get_rx_queue(skb
))
2936 rxqueue
= dev
->_rx
+ rxq_index
;
2937 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2940 flow_id
= skb
->rxhash
& flow_table
->mask
;
2941 rc
= dev
->netdev_ops
->ndo_rx_flow_steer(dev
, skb
,
2942 rxq_index
, flow_id
);
2946 rflow
= &flow_table
->flows
[flow_id
];
2948 if (old_rflow
->filter
== rflow
->filter
)
2949 old_rflow
->filter
= RPS_NO_FILTER
;
2953 per_cpu(softnet_data
, next_cpu
).input_queue_head
;
2956 rflow
->cpu
= next_cpu
;
2961 * get_rps_cpu is called from netif_receive_skb and returns the target
2962 * CPU from the RPS map of the receiving queue for a given skb.
2963 * rcu_read_lock must be held on entry.
2965 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2966 struct rps_dev_flow
**rflowp
)
2968 struct netdev_rx_queue
*rxqueue
;
2969 struct rps_map
*map
;
2970 struct rps_dev_flow_table
*flow_table
;
2971 struct rps_sock_flow_table
*sock_flow_table
;
2975 if (skb_rx_queue_recorded(skb
)) {
2976 u16 index
= skb_get_rx_queue(skb
);
2977 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
2978 WARN_ONCE(dev
->real_num_rx_queues
> 1,
2979 "%s received packet on queue %u, but number "
2980 "of RX queues is %u\n",
2981 dev
->name
, index
, dev
->real_num_rx_queues
);
2984 rxqueue
= dev
->_rx
+ index
;
2988 map
= rcu_dereference(rxqueue
->rps_map
);
2990 if (map
->len
== 1 &&
2991 !rcu_access_pointer(rxqueue
->rps_flow_table
)) {
2992 tcpu
= map
->cpus
[0];
2993 if (cpu_online(tcpu
))
2997 } else if (!rcu_access_pointer(rxqueue
->rps_flow_table
)) {
3001 skb_reset_network_header(skb
);
3002 if (!skb_get_rxhash(skb
))
3005 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3006 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
3007 if (flow_table
&& sock_flow_table
) {
3009 struct rps_dev_flow
*rflow
;
3011 rflow
= &flow_table
->flows
[skb
->rxhash
& flow_table
->mask
];
3014 next_cpu
= sock_flow_table
->ents
[skb
->rxhash
&
3015 sock_flow_table
->mask
];
3018 * If the desired CPU (where last recvmsg was done) is
3019 * different from current CPU (one in the rx-queue flow
3020 * table entry), switch if one of the following holds:
3021 * - Current CPU is unset (equal to RPS_NO_CPU).
3022 * - Current CPU is offline.
3023 * - The current CPU's queue tail has advanced beyond the
3024 * last packet that was enqueued using this table entry.
3025 * This guarantees that all previous packets for the flow
3026 * have been dequeued, thus preserving in order delivery.
3028 if (unlikely(tcpu
!= next_cpu
) &&
3029 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
3030 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
3031 rflow
->last_qtail
)) >= 0)) {
3033 rflow
= set_rps_cpu(dev
, skb
, rflow
, next_cpu
);
3036 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
3044 tcpu
= map
->cpus
[((u64
) skb
->rxhash
* map
->len
) >> 32];
3046 if (cpu_online(tcpu
)) {
3056 #ifdef CONFIG_RFS_ACCEL
3059 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3060 * @dev: Device on which the filter was set
3061 * @rxq_index: RX queue index
3062 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3063 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3065 * Drivers that implement ndo_rx_flow_steer() should periodically call
3066 * this function for each installed filter and remove the filters for
3067 * which it returns %true.
3069 bool rps_may_expire_flow(struct net_device
*dev
, u16 rxq_index
,
3070 u32 flow_id
, u16 filter_id
)
3072 struct netdev_rx_queue
*rxqueue
= dev
->_rx
+ rxq_index
;
3073 struct rps_dev_flow_table
*flow_table
;
3074 struct rps_dev_flow
*rflow
;
3079 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3080 if (flow_table
&& flow_id
<= flow_table
->mask
) {
3081 rflow
= &flow_table
->flows
[flow_id
];
3082 cpu
= ACCESS_ONCE(rflow
->cpu
);
3083 if (rflow
->filter
== filter_id
&& cpu
!= RPS_NO_CPU
&&
3084 ((int)(per_cpu(softnet_data
, cpu
).input_queue_head
-
3085 rflow
->last_qtail
) <
3086 (int)(10 * flow_table
->mask
)))
3092 EXPORT_SYMBOL(rps_may_expire_flow
);
3094 #endif /* CONFIG_RFS_ACCEL */
3096 /* Called from hardirq (IPI) context */
3097 static void rps_trigger_softirq(void *data
)
3099 struct softnet_data
*sd
= data
;
3101 ____napi_schedule(sd
, &sd
->backlog
);
3105 #endif /* CONFIG_RPS */
3108 * Check if this softnet_data structure is another cpu one
3109 * If yes, queue it to our IPI list and return 1
3112 static int rps_ipi_queued(struct softnet_data
*sd
)
3115 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
3118 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
3119 mysd
->rps_ipi_list
= sd
;
3121 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3124 #endif /* CONFIG_RPS */
3128 #ifdef CONFIG_NET_FLOW_LIMIT
3129 int netdev_flow_limit_table_len __read_mostly
= (1 << 12);
3132 static bool skb_flow_limit(struct sk_buff
*skb
, unsigned int qlen
)
3134 #ifdef CONFIG_NET_FLOW_LIMIT
3135 struct sd_flow_limit
*fl
;
3136 struct softnet_data
*sd
;
3137 unsigned int old_flow
, new_flow
;
3139 if (qlen
< (netdev_max_backlog
>> 1))
3142 sd
= &__get_cpu_var(softnet_data
);
3145 fl
= rcu_dereference(sd
->flow_limit
);
3147 new_flow
= skb_get_rxhash(skb
) & (fl
->num_buckets
- 1);
3148 old_flow
= fl
->history
[fl
->history_head
];
3149 fl
->history
[fl
->history_head
] = new_flow
;
3152 fl
->history_head
&= FLOW_LIMIT_HISTORY
- 1;
3154 if (likely(fl
->buckets
[old_flow
]))
3155 fl
->buckets
[old_flow
]--;
3157 if (++fl
->buckets
[new_flow
] > (FLOW_LIMIT_HISTORY
>> 1)) {
3169 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3170 * queue (may be a remote CPU queue).
3172 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
3173 unsigned int *qtail
)
3175 struct softnet_data
*sd
;
3176 unsigned long flags
;
3179 sd
= &per_cpu(softnet_data
, cpu
);
3181 local_irq_save(flags
);
3184 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3185 if (qlen
<= netdev_max_backlog
&& !skb_flow_limit(skb
, qlen
)) {
3186 if (skb_queue_len(&sd
->input_pkt_queue
)) {
3188 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
3189 input_queue_tail_incr_save(sd
, qtail
);
3191 local_irq_restore(flags
);
3192 return NET_RX_SUCCESS
;
3195 /* Schedule NAPI for backlog device
3196 * We can use non atomic operation since we own the queue lock
3198 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
3199 if (!rps_ipi_queued(sd
))
3200 ____napi_schedule(sd
, &sd
->backlog
);
3208 local_irq_restore(flags
);
3210 atomic_long_inc(&skb
->dev
->rx_dropped
);
3216 * netif_rx - post buffer to the network code
3217 * @skb: buffer to post
3219 * This function receives a packet from a device driver and queues it for
3220 * the upper (protocol) levels to process. It always succeeds. The buffer
3221 * may be dropped during processing for congestion control or by the
3225 * NET_RX_SUCCESS (no congestion)
3226 * NET_RX_DROP (packet was dropped)
3230 int netif_rx(struct sk_buff
*skb
)
3234 /* if netpoll wants it, pretend we never saw it */
3235 if (netpoll_rx(skb
))
3238 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3240 trace_netif_rx(skb
);
3242 if (static_key_false(&rps_needed
)) {
3243 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3249 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3251 cpu
= smp_processor_id();
3253 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3261 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
3266 EXPORT_SYMBOL(netif_rx
);
3268 int netif_rx_ni(struct sk_buff
*skb
)
3273 err
= netif_rx(skb
);
3274 if (local_softirq_pending())
3280 EXPORT_SYMBOL(netif_rx_ni
);
3282 static void net_tx_action(struct softirq_action
*h
)
3284 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3286 if (sd
->completion_queue
) {
3287 struct sk_buff
*clist
;
3289 local_irq_disable();
3290 clist
= sd
->completion_queue
;
3291 sd
->completion_queue
= NULL
;
3295 struct sk_buff
*skb
= clist
;
3296 clist
= clist
->next
;
3298 WARN_ON(atomic_read(&skb
->users
));
3299 trace_kfree_skb(skb
, net_tx_action
);
3304 if (sd
->output_queue
) {
3307 local_irq_disable();
3308 head
= sd
->output_queue
;
3309 sd
->output_queue
= NULL
;
3310 sd
->output_queue_tailp
= &sd
->output_queue
;
3314 struct Qdisc
*q
= head
;
3315 spinlock_t
*root_lock
;
3317 head
= head
->next_sched
;
3319 root_lock
= qdisc_lock(q
);
3320 if (spin_trylock(root_lock
)) {
3321 smp_mb__before_clear_bit();
3322 clear_bit(__QDISC_STATE_SCHED
,
3325 spin_unlock(root_lock
);
3327 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
3329 __netif_reschedule(q
);
3331 smp_mb__before_clear_bit();
3332 clear_bit(__QDISC_STATE_SCHED
,
3340 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3341 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3342 /* This hook is defined here for ATM LANE */
3343 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
3344 unsigned char *addr
) __read_mostly
;
3345 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
3348 #ifdef CONFIG_NET_CLS_ACT
3349 /* TODO: Maybe we should just force sch_ingress to be compiled in
3350 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3351 * a compare and 2 stores extra right now if we dont have it on
3352 * but have CONFIG_NET_CLS_ACT
3353 * NOTE: This doesn't stop any functionality; if you dont have
3354 * the ingress scheduler, you just can't add policies on ingress.
3357 static int ing_filter(struct sk_buff
*skb
, struct netdev_queue
*rxq
)
3359 struct net_device
*dev
= skb
->dev
;
3360 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
3361 int result
= TC_ACT_OK
;
3364 if (unlikely(MAX_RED_LOOP
< ttl
++)) {
3365 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3366 skb
->skb_iif
, dev
->ifindex
);
3370 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
3371 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
3374 if (q
!= &noop_qdisc
) {
3375 spin_lock(qdisc_lock(q
));
3376 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
3377 result
= qdisc_enqueue_root(skb
, q
);
3378 spin_unlock(qdisc_lock(q
));
3384 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
3385 struct packet_type
**pt_prev
,
3386 int *ret
, struct net_device
*orig_dev
)
3388 struct netdev_queue
*rxq
= rcu_dereference(skb
->dev
->ingress_queue
);
3390 if (!rxq
|| rxq
->qdisc
== &noop_qdisc
)
3394 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
3398 switch (ing_filter(skb
, rxq
)) {
3412 * netdev_rx_handler_register - register receive handler
3413 * @dev: device to register a handler for
3414 * @rx_handler: receive handler to register
3415 * @rx_handler_data: data pointer that is used by rx handler
3417 * Register a receive hander for a device. This handler will then be
3418 * called from __netif_receive_skb. A negative errno code is returned
3421 * The caller must hold the rtnl_mutex.
3423 * For a general description of rx_handler, see enum rx_handler_result.
3425 int netdev_rx_handler_register(struct net_device
*dev
,
3426 rx_handler_func_t
*rx_handler
,
3427 void *rx_handler_data
)
3431 if (dev
->rx_handler
)
3434 /* Note: rx_handler_data must be set before rx_handler */
3435 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
3436 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
3440 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
3443 * netdev_rx_handler_unregister - unregister receive handler
3444 * @dev: device to unregister a handler from
3446 * Unregister a receive handler from a device.
3448 * The caller must hold the rtnl_mutex.
3450 void netdev_rx_handler_unregister(struct net_device
*dev
)
3454 RCU_INIT_POINTER(dev
->rx_handler
, NULL
);
3455 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3456 * section has a guarantee to see a non NULL rx_handler_data
3460 RCU_INIT_POINTER(dev
->rx_handler_data
, NULL
);
3462 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
3465 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3466 * the special handling of PFMEMALLOC skbs.
3468 static bool skb_pfmemalloc_protocol(struct sk_buff
*skb
)
3470 switch (skb
->protocol
) {
3471 case __constant_htons(ETH_P_ARP
):
3472 case __constant_htons(ETH_P_IP
):
3473 case __constant_htons(ETH_P_IPV6
):
3474 case __constant_htons(ETH_P_8021Q
):
3475 case __constant_htons(ETH_P_8021AD
):
3482 static int __netif_receive_skb_core(struct sk_buff
*skb
, bool pfmemalloc
)
3484 struct packet_type
*ptype
, *pt_prev
;
3485 rx_handler_func_t
*rx_handler
;
3486 struct net_device
*orig_dev
;
3487 struct net_device
*null_or_dev
;
3488 bool deliver_exact
= false;
3489 int ret
= NET_RX_DROP
;
3492 net_timestamp_check(!netdev_tstamp_prequeue
, skb
);
3494 trace_netif_receive_skb(skb
);
3496 /* if we've gotten here through NAPI, check netpoll */
3497 if (netpoll_receive_skb(skb
))
3500 orig_dev
= skb
->dev
;
3502 skb_reset_network_header(skb
);
3503 if (!skb_transport_header_was_set(skb
))
3504 skb_reset_transport_header(skb
);
3505 skb_reset_mac_len(skb
);
3512 skb
->skb_iif
= skb
->dev
->ifindex
;
3514 __this_cpu_inc(softnet_data
.processed
);
3516 if (skb
->protocol
== cpu_to_be16(ETH_P_8021Q
) ||
3517 skb
->protocol
== cpu_to_be16(ETH_P_8021AD
)) {
3518 skb
= vlan_untag(skb
);
3523 #ifdef CONFIG_NET_CLS_ACT
3524 if (skb
->tc_verd
& TC_NCLS
) {
3525 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
3533 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
3534 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
3536 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3542 #ifdef CONFIG_NET_CLS_ACT
3543 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
3549 if (pfmemalloc
&& !skb_pfmemalloc_protocol(skb
))
3552 if (vlan_tx_tag_present(skb
)) {
3554 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3557 if (vlan_do_receive(&skb
))
3559 else if (unlikely(!skb
))
3563 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
3566 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3569 switch (rx_handler(&skb
)) {
3570 case RX_HANDLER_CONSUMED
:
3571 ret
= NET_RX_SUCCESS
;
3573 case RX_HANDLER_ANOTHER
:
3575 case RX_HANDLER_EXACT
:
3576 deliver_exact
= true;
3577 case RX_HANDLER_PASS
:
3584 if (unlikely(vlan_tx_tag_present(skb
))) {
3585 if (vlan_tx_tag_get_id(skb
))
3586 skb
->pkt_type
= PACKET_OTHERHOST
;
3587 /* Note: we might in the future use prio bits
3588 * and set skb->priority like in vlan_do_receive()
3589 * For the time being, just ignore Priority Code Point
3594 /* deliver only exact match when indicated */
3595 null_or_dev
= deliver_exact
? skb
->dev
: NULL
;
3597 type
= skb
->protocol
;
3598 list_for_each_entry_rcu(ptype
,
3599 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
3600 if (ptype
->type
== type
&&
3601 (ptype
->dev
== null_or_dev
|| ptype
->dev
== skb
->dev
||
3602 ptype
->dev
== orig_dev
)) {
3604 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3610 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
3613 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
3616 atomic_long_inc(&skb
->dev
->rx_dropped
);
3618 /* Jamal, now you will not able to escape explaining
3619 * me how you were going to use this. :-)
3630 static int __netif_receive_skb(struct sk_buff
*skb
)
3634 if (sk_memalloc_socks() && skb_pfmemalloc(skb
)) {
3635 unsigned long pflags
= current
->flags
;
3638 * PFMEMALLOC skbs are special, they should
3639 * - be delivered to SOCK_MEMALLOC sockets only
3640 * - stay away from userspace
3641 * - have bounded memory usage
3643 * Use PF_MEMALLOC as this saves us from propagating the allocation
3644 * context down to all allocation sites.
3646 current
->flags
|= PF_MEMALLOC
;
3647 ret
= __netif_receive_skb_core(skb
, true);
3648 tsk_restore_flags(current
, pflags
, PF_MEMALLOC
);
3650 ret
= __netif_receive_skb_core(skb
, false);
3656 * netif_receive_skb - process receive buffer from network
3657 * @skb: buffer to process
3659 * netif_receive_skb() is the main receive data processing function.
3660 * It always succeeds. The buffer may be dropped during processing
3661 * for congestion control or by the protocol layers.
3663 * This function may only be called from softirq context and interrupts
3664 * should be enabled.
3666 * Return values (usually ignored):
3667 * NET_RX_SUCCESS: no congestion
3668 * NET_RX_DROP: packet was dropped
3670 int netif_receive_skb(struct sk_buff
*skb
)
3672 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3674 if (skb_defer_rx_timestamp(skb
))
3675 return NET_RX_SUCCESS
;
3678 if (static_key_false(&rps_needed
)) {
3679 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3684 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3687 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3694 return __netif_receive_skb(skb
);
3696 EXPORT_SYMBOL(netif_receive_skb
);
3698 /* Network device is going away, flush any packets still pending
3699 * Called with irqs disabled.
3701 static void flush_backlog(void *arg
)
3703 struct net_device
*dev
= arg
;
3704 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3705 struct sk_buff
*skb
, *tmp
;
3708 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3709 if (skb
->dev
== dev
) {
3710 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3712 input_queue_head_incr(sd
);
3717 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3718 if (skb
->dev
== dev
) {
3719 __skb_unlink(skb
, &sd
->process_queue
);
3721 input_queue_head_incr(sd
);
3726 static int napi_gro_complete(struct sk_buff
*skb
)
3728 struct packet_offload
*ptype
;
3729 __be16 type
= skb
->protocol
;
3730 struct list_head
*head
= &offload_base
;
3733 BUILD_BUG_ON(sizeof(struct napi_gro_cb
) > sizeof(skb
->cb
));
3735 if (NAPI_GRO_CB(skb
)->count
== 1) {
3736 skb_shinfo(skb
)->gso_size
= 0;
3741 list_for_each_entry_rcu(ptype
, head
, list
) {
3742 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_complete
)
3745 err
= ptype
->callbacks
.gro_complete(skb
);
3751 WARN_ON(&ptype
->list
== head
);
3753 return NET_RX_SUCCESS
;
3757 return netif_receive_skb(skb
);
3760 /* napi->gro_list contains packets ordered by age.
3761 * youngest packets at the head of it.
3762 * Complete skbs in reverse order to reduce latencies.
3764 void napi_gro_flush(struct napi_struct
*napi
, bool flush_old
)
3766 struct sk_buff
*skb
, *prev
= NULL
;
3768 /* scan list and build reverse chain */
3769 for (skb
= napi
->gro_list
; skb
!= NULL
; skb
= skb
->next
) {
3774 for (skb
= prev
; skb
; skb
= prev
) {
3777 if (flush_old
&& NAPI_GRO_CB(skb
)->age
== jiffies
)
3781 napi_gro_complete(skb
);
3785 napi
->gro_list
= NULL
;
3787 EXPORT_SYMBOL(napi_gro_flush
);
3789 static void gro_list_prepare(struct napi_struct
*napi
, struct sk_buff
*skb
)
3792 unsigned int maclen
= skb
->dev
->hard_header_len
;
3794 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3795 unsigned long diffs
;
3797 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
3798 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
3799 if (maclen
== ETH_HLEN
)
3800 diffs
|= compare_ether_header(skb_mac_header(p
),
3801 skb_gro_mac_header(skb
));
3803 diffs
= memcmp(skb_mac_header(p
),
3804 skb_gro_mac_header(skb
),
3806 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
3807 NAPI_GRO_CB(p
)->flush
= 0;
3811 static enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3813 struct sk_buff
**pp
= NULL
;
3814 struct packet_offload
*ptype
;
3815 __be16 type
= skb
->protocol
;
3816 struct list_head
*head
= &offload_base
;
3818 enum gro_result ret
;
3820 if (!(skb
->dev
->features
& NETIF_F_GRO
) || netpoll_rx_on(skb
))
3823 if (skb_is_gso(skb
) || skb_has_frag_list(skb
))
3826 gro_list_prepare(napi
, skb
);
3829 list_for_each_entry_rcu(ptype
, head
, list
) {
3830 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_receive
)
3833 skb_set_network_header(skb
, skb_gro_offset(skb
));
3834 skb_reset_mac_len(skb
);
3835 NAPI_GRO_CB(skb
)->same_flow
= 0;
3836 NAPI_GRO_CB(skb
)->flush
= 0;
3837 NAPI_GRO_CB(skb
)->free
= 0;
3839 pp
= ptype
->callbacks
.gro_receive(&napi
->gro_list
, skb
);
3844 if (&ptype
->list
== head
)
3847 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3848 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3851 struct sk_buff
*nskb
= *pp
;
3855 napi_gro_complete(nskb
);
3862 if (NAPI_GRO_CB(skb
)->flush
|| napi
->gro_count
>= MAX_GRO_SKBS
)
3866 NAPI_GRO_CB(skb
)->count
= 1;
3867 NAPI_GRO_CB(skb
)->age
= jiffies
;
3868 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
3869 skb
->next
= napi
->gro_list
;
3870 napi
->gro_list
= skb
;
3874 if (skb_headlen(skb
) < skb_gro_offset(skb
)) {
3875 int grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
3877 BUG_ON(skb
->end
- skb
->tail
< grow
);
3879 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3882 skb
->data_len
-= grow
;
3884 skb_shinfo(skb
)->frags
[0].page_offset
+= grow
;
3885 skb_frag_size_sub(&skb_shinfo(skb
)->frags
[0], grow
);
3887 if (unlikely(!skb_frag_size(&skb_shinfo(skb
)->frags
[0]))) {
3888 skb_frag_unref(skb
, 0);
3889 memmove(skb_shinfo(skb
)->frags
,
3890 skb_shinfo(skb
)->frags
+ 1,
3891 --skb_shinfo(skb
)->nr_frags
* sizeof(skb_frag_t
));
3904 static gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
3908 if (netif_receive_skb(skb
))
3916 case GRO_MERGED_FREE
:
3917 if (NAPI_GRO_CB(skb
)->free
== NAPI_GRO_FREE_STOLEN_HEAD
)
3918 kmem_cache_free(skbuff_head_cache
, skb
);
3931 static void skb_gro_reset_offset(struct sk_buff
*skb
)
3933 const struct skb_shared_info
*pinfo
= skb_shinfo(skb
);
3934 const skb_frag_t
*frag0
= &pinfo
->frags
[0];
3936 NAPI_GRO_CB(skb
)->data_offset
= 0;
3937 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3938 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3940 if (skb_mac_header(skb
) == skb_tail_pointer(skb
) &&
3942 !PageHighMem(skb_frag_page(frag0
))) {
3943 NAPI_GRO_CB(skb
)->frag0
= skb_frag_address(frag0
);
3944 NAPI_GRO_CB(skb
)->frag0_len
= skb_frag_size(frag0
);
3948 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3950 skb_gro_reset_offset(skb
);
3952 return napi_skb_finish(dev_gro_receive(napi
, skb
), skb
);
3954 EXPORT_SYMBOL(napi_gro_receive
);
3956 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
3958 __skb_pull(skb
, skb_headlen(skb
));
3959 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3960 skb_reserve(skb
, NET_SKB_PAD
+ NET_IP_ALIGN
- skb_headroom(skb
));
3962 skb
->dev
= napi
->dev
;
3968 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
3970 struct sk_buff
*skb
= napi
->skb
;
3973 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
3979 EXPORT_SYMBOL(napi_get_frags
);
3981 static gro_result_t
napi_frags_finish(struct napi_struct
*napi
, struct sk_buff
*skb
,
3987 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
3989 if (ret
== GRO_HELD
)
3990 skb_gro_pull(skb
, -ETH_HLEN
);
3991 else if (netif_receive_skb(skb
))
3996 case GRO_MERGED_FREE
:
3997 napi_reuse_skb(napi
, skb
);
4007 static struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
4009 struct sk_buff
*skb
= napi
->skb
;
4016 skb_reset_mac_header(skb
);
4017 skb_gro_reset_offset(skb
);
4019 off
= skb_gro_offset(skb
);
4020 hlen
= off
+ sizeof(*eth
);
4021 eth
= skb_gro_header_fast(skb
, off
);
4022 if (skb_gro_header_hard(skb
, hlen
)) {
4023 eth
= skb_gro_header_slow(skb
, hlen
, off
);
4024 if (unlikely(!eth
)) {
4025 napi_reuse_skb(napi
, skb
);
4031 skb_gro_pull(skb
, sizeof(*eth
));
4034 * This works because the only protocols we care about don't require
4035 * special handling. We'll fix it up properly at the end.
4037 skb
->protocol
= eth
->h_proto
;
4043 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
4045 struct sk_buff
*skb
= napi_frags_skb(napi
);
4050 return napi_frags_finish(napi
, skb
, dev_gro_receive(napi
, skb
));
4052 EXPORT_SYMBOL(napi_gro_frags
);
4055 * net_rps_action sends any pending IPI's for rps.
4056 * Note: called with local irq disabled, but exits with local irq enabled.
4058 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
4061 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
4064 sd
->rps_ipi_list
= NULL
;
4068 /* Send pending IPI's to kick RPS processing on remote cpus. */
4070 struct softnet_data
*next
= remsd
->rps_ipi_next
;
4072 if (cpu_online(remsd
->cpu
))
4073 __smp_call_function_single(remsd
->cpu
,
4082 static int process_backlog(struct napi_struct
*napi
, int quota
)
4085 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
4088 /* Check if we have pending ipi, its better to send them now,
4089 * not waiting net_rx_action() end.
4091 if (sd
->rps_ipi_list
) {
4092 local_irq_disable();
4093 net_rps_action_and_irq_enable(sd
);
4096 napi
->weight
= weight_p
;
4097 local_irq_disable();
4098 while (work
< quota
) {
4099 struct sk_buff
*skb
;
4102 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
4104 __netif_receive_skb(skb
);
4105 local_irq_disable();
4106 input_queue_head_incr(sd
);
4107 if (++work
>= quota
) {
4114 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
4116 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
4117 &sd
->process_queue
);
4119 if (qlen
< quota
- work
) {
4121 * Inline a custom version of __napi_complete().
4122 * only current cpu owns and manipulates this napi,
4123 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
4124 * we can use a plain write instead of clear_bit(),
4125 * and we dont need an smp_mb() memory barrier.
4127 list_del(&napi
->poll_list
);
4130 quota
= work
+ qlen
;
4140 * __napi_schedule - schedule for receive
4141 * @n: entry to schedule
4143 * The entry's receive function will be scheduled to run
4145 void __napi_schedule(struct napi_struct
*n
)
4147 unsigned long flags
;
4149 local_irq_save(flags
);
4150 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
4151 local_irq_restore(flags
);
4153 EXPORT_SYMBOL(__napi_schedule
);
4155 void __napi_complete(struct napi_struct
*n
)
4157 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
4158 BUG_ON(n
->gro_list
);
4160 list_del(&n
->poll_list
);
4161 smp_mb__before_clear_bit();
4162 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
4164 EXPORT_SYMBOL(__napi_complete
);
4166 void napi_complete(struct napi_struct
*n
)
4168 unsigned long flags
;
4171 * don't let napi dequeue from the cpu poll list
4172 * just in case its running on a different cpu
4174 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
4177 napi_gro_flush(n
, false);
4178 local_irq_save(flags
);
4180 local_irq_restore(flags
);
4182 EXPORT_SYMBOL(napi_complete
);
4184 /* must be called under rcu_read_lock(), as we dont take a reference */
4185 struct napi_struct
*napi_by_id(unsigned int napi_id
)
4187 unsigned int hash
= napi_id
% HASH_SIZE(napi_hash
);
4188 struct napi_struct
*napi
;
4190 hlist_for_each_entry_rcu(napi
, &napi_hash
[hash
], napi_hash_node
)
4191 if (napi
->napi_id
== napi_id
)
4196 EXPORT_SYMBOL_GPL(napi_by_id
);
4198 void napi_hash_add(struct napi_struct
*napi
)
4200 if (!test_and_set_bit(NAPI_STATE_HASHED
, &napi
->state
)) {
4202 spin_lock(&napi_hash_lock
);
4204 /* 0 is not a valid id, we also skip an id that is taken
4205 * we expect both events to be extremely rare
4208 while (!napi
->napi_id
) {
4209 napi
->napi_id
= ++napi_gen_id
;
4210 if (napi_by_id(napi
->napi_id
))
4214 hlist_add_head_rcu(&napi
->napi_hash_node
,
4215 &napi_hash
[napi
->napi_id
% HASH_SIZE(napi_hash
)]);
4217 spin_unlock(&napi_hash_lock
);
4220 EXPORT_SYMBOL_GPL(napi_hash_add
);
4222 /* Warning : caller is responsible to make sure rcu grace period
4223 * is respected before freeing memory containing @napi
4225 void napi_hash_del(struct napi_struct
*napi
)
4227 spin_lock(&napi_hash_lock
);
4229 if (test_and_clear_bit(NAPI_STATE_HASHED
, &napi
->state
))
4230 hlist_del_rcu(&napi
->napi_hash_node
);
4232 spin_unlock(&napi_hash_lock
);
4234 EXPORT_SYMBOL_GPL(napi_hash_del
);
4236 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
4237 int (*poll
)(struct napi_struct
*, int), int weight
)
4239 INIT_LIST_HEAD(&napi
->poll_list
);
4240 napi
->gro_count
= 0;
4241 napi
->gro_list
= NULL
;
4244 if (weight
> NAPI_POLL_WEIGHT
)
4245 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4247 napi
->weight
= weight
;
4248 list_add(&napi
->dev_list
, &dev
->napi_list
);
4250 #ifdef CONFIG_NETPOLL
4251 spin_lock_init(&napi
->poll_lock
);
4252 napi
->poll_owner
= -1;
4254 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
4256 EXPORT_SYMBOL(netif_napi_add
);
4258 void netif_napi_del(struct napi_struct
*napi
)
4260 struct sk_buff
*skb
, *next
;
4262 list_del_init(&napi
->dev_list
);
4263 napi_free_frags(napi
);
4265 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
4271 napi
->gro_list
= NULL
;
4272 napi
->gro_count
= 0;
4274 EXPORT_SYMBOL(netif_napi_del
);
4276 static void net_rx_action(struct softirq_action
*h
)
4278 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
4279 unsigned long time_limit
= jiffies
+ 2;
4280 int budget
= netdev_budget
;
4283 local_irq_disable();
4285 while (!list_empty(&sd
->poll_list
)) {
4286 struct napi_struct
*n
;
4289 /* If softirq window is exhuasted then punt.
4290 * Allow this to run for 2 jiffies since which will allow
4291 * an average latency of 1.5/HZ.
4293 if (unlikely(budget
<= 0 || time_after_eq(jiffies
, time_limit
)))
4298 /* Even though interrupts have been re-enabled, this
4299 * access is safe because interrupts can only add new
4300 * entries to the tail of this list, and only ->poll()
4301 * calls can remove this head entry from the list.
4303 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
4305 have
= netpoll_poll_lock(n
);
4309 /* This NAPI_STATE_SCHED test is for avoiding a race
4310 * with netpoll's poll_napi(). Only the entity which
4311 * obtains the lock and sees NAPI_STATE_SCHED set will
4312 * actually make the ->poll() call. Therefore we avoid
4313 * accidentally calling ->poll() when NAPI is not scheduled.
4316 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
4317 work
= n
->poll(n
, weight
);
4321 WARN_ON_ONCE(work
> weight
);
4325 local_irq_disable();
4327 /* Drivers must not modify the NAPI state if they
4328 * consume the entire weight. In such cases this code
4329 * still "owns" the NAPI instance and therefore can
4330 * move the instance around on the list at-will.
4332 if (unlikely(work
== weight
)) {
4333 if (unlikely(napi_disable_pending(n
))) {
4336 local_irq_disable();
4339 /* flush too old packets
4340 * If HZ < 1000, flush all packets.
4343 napi_gro_flush(n
, HZ
>= 1000);
4344 local_irq_disable();
4346 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
4350 netpoll_poll_unlock(have
);
4353 net_rps_action_and_irq_enable(sd
);
4355 #ifdef CONFIG_NET_DMA
4357 * There may not be any more sk_buffs coming right now, so push
4358 * any pending DMA copies to hardware
4360 dma_issue_pending_all();
4367 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
4371 struct netdev_adjacent
{
4372 struct net_device
*dev
;
4374 /* upper master flag, there can only be one master device per list */
4377 /* indicates that this dev is our first-level lower/upper device */
4380 /* counter for the number of times this device was added to us */
4383 struct list_head list
;
4384 struct rcu_head rcu
;
4387 static struct netdev_adjacent
*__netdev_find_adj(struct net_device
*dev
,
4388 struct net_device
*adj_dev
,
4391 struct netdev_adjacent
*adj
;
4392 struct list_head
*dev_list
;
4394 dev_list
= upper
? &dev
->upper_dev_list
: &dev
->lower_dev_list
;
4396 list_for_each_entry(adj
, dev_list
, list
) {
4397 if (adj
->dev
== adj_dev
)
4403 static inline struct netdev_adjacent
*__netdev_find_upper(struct net_device
*dev
,
4404 struct net_device
*udev
)
4406 return __netdev_find_adj(dev
, udev
, true);
4409 static inline struct netdev_adjacent
*__netdev_find_lower(struct net_device
*dev
,
4410 struct net_device
*ldev
)
4412 return __netdev_find_adj(dev
, ldev
, false);
4416 * netdev_has_upper_dev - Check if device is linked to an upper device
4418 * @upper_dev: upper device to check
4420 * Find out if a device is linked to specified upper device and return true
4421 * in case it is. Note that this checks only immediate upper device,
4422 * not through a complete stack of devices. The caller must hold the RTNL lock.
4424 bool netdev_has_upper_dev(struct net_device
*dev
,
4425 struct net_device
*upper_dev
)
4429 return __netdev_find_upper(dev
, upper_dev
);
4431 EXPORT_SYMBOL(netdev_has_upper_dev
);
4434 * netdev_has_any_upper_dev - Check if device is linked to some device
4437 * Find out if a device is linked to an upper device and return true in case
4438 * it is. The caller must hold the RTNL lock.
4440 bool netdev_has_any_upper_dev(struct net_device
*dev
)
4444 return !list_empty(&dev
->upper_dev_list
);
4446 EXPORT_SYMBOL(netdev_has_any_upper_dev
);
4449 * netdev_master_upper_dev_get - Get master upper device
4452 * Find a master upper device and return pointer to it or NULL in case
4453 * it's not there. The caller must hold the RTNL lock.
4455 struct net_device
*netdev_master_upper_dev_get(struct net_device
*dev
)
4457 struct netdev_adjacent
*upper
;
4461 if (list_empty(&dev
->upper_dev_list
))
4464 upper
= list_first_entry(&dev
->upper_dev_list
,
4465 struct netdev_adjacent
, list
);
4466 if (likely(upper
->master
))
4470 EXPORT_SYMBOL(netdev_master_upper_dev_get
);
4472 /* netdev_upper_get_next_dev_rcu - Get the next dev from upper list
4474 * @iter: list_head ** of the current position
4476 * Gets the next device from the dev's upper list, starting from iter
4477 * position. The caller must hold RCU read lock.
4479 struct net_device
*netdev_upper_get_next_dev_rcu(struct net_device
*dev
,
4480 struct list_head
**iter
)
4482 struct netdev_adjacent
*upper
;
4484 WARN_ON_ONCE(!rcu_read_lock_held());
4486 upper
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
4488 if (&upper
->list
== &dev
->upper_dev_list
)
4491 *iter
= &upper
->list
;
4495 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu
);
4498 * netdev_master_upper_dev_get_rcu - Get master upper device
4501 * Find a master upper device and return pointer to it or NULL in case
4502 * it's not there. The caller must hold the RCU read lock.
4504 struct net_device
*netdev_master_upper_dev_get_rcu(struct net_device
*dev
)
4506 struct netdev_adjacent
*upper
;
4508 upper
= list_first_or_null_rcu(&dev
->upper_dev_list
,
4509 struct netdev_adjacent
, list
);
4510 if (upper
&& likely(upper
->master
))
4514 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu
);
4516 static int __netdev_adjacent_dev_insert(struct net_device
*dev
,
4517 struct net_device
*adj_dev
,
4518 bool neighbour
, bool master
,
4521 struct netdev_adjacent
*adj
;
4523 adj
= __netdev_find_adj(dev
, adj_dev
, upper
);
4531 adj
= kmalloc(sizeof(*adj
), GFP_KERNEL
);
4536 adj
->master
= master
;
4537 adj
->neighbour
= neighbour
;
4541 pr_debug("dev_hold for %s, because of %s link added from %s to %s\n",
4542 adj_dev
->name
, upper
? "upper" : "lower", dev
->name
,
4546 list_add_tail_rcu(&adj
->list
, &dev
->lower_dev_list
);
4550 /* Ensure that master upper link is always the first item in list. */
4552 list_add_rcu(&adj
->list
, &dev
->upper_dev_list
);
4554 list_add_tail_rcu(&adj
->list
, &dev
->upper_dev_list
);
4559 static inline int __netdev_upper_dev_insert(struct net_device
*dev
,
4560 struct net_device
*udev
,
4561 bool master
, bool neighbour
)
4563 return __netdev_adjacent_dev_insert(dev
, udev
, neighbour
, master
,
4567 static inline int __netdev_lower_dev_insert(struct net_device
*dev
,
4568 struct net_device
*ldev
,
4571 return __netdev_adjacent_dev_insert(dev
, ldev
, neighbour
, false,
4575 void __netdev_adjacent_dev_remove(struct net_device
*dev
,
4576 struct net_device
*adj_dev
, bool upper
)
4578 struct netdev_adjacent
*adj
;
4581 adj
= __netdev_find_upper(dev
, adj_dev
);
4583 adj
= __netdev_find_lower(dev
, adj_dev
);
4588 if (adj
->ref_nr
> 1) {
4593 list_del_rcu(&adj
->list
);
4594 pr_debug("dev_put for %s, because of %s link removed from %s to %s\n",
4595 adj_dev
->name
, upper
? "upper" : "lower", dev
->name
,
4598 kfree_rcu(adj
, rcu
);
4601 static inline void __netdev_upper_dev_remove(struct net_device
*dev
,
4602 struct net_device
*udev
)
4604 return __netdev_adjacent_dev_remove(dev
, udev
, true);
4607 static inline void __netdev_lower_dev_remove(struct net_device
*dev
,
4608 struct net_device
*ldev
)
4610 return __netdev_adjacent_dev_remove(dev
, ldev
, false);
4613 int __netdev_adjacent_dev_insert_link(struct net_device
*dev
,
4614 struct net_device
*upper_dev
,
4615 bool master
, bool neighbour
)
4619 ret
= __netdev_upper_dev_insert(dev
, upper_dev
, master
, neighbour
);
4623 ret
= __netdev_lower_dev_insert(upper_dev
, dev
, neighbour
);
4625 __netdev_upper_dev_remove(dev
, upper_dev
);
4632 static inline int __netdev_adjacent_dev_link(struct net_device
*dev
,
4633 struct net_device
*udev
)
4635 return __netdev_adjacent_dev_insert_link(dev
, udev
, false, false);
4638 static inline int __netdev_adjacent_dev_link_neighbour(struct net_device
*dev
,
4639 struct net_device
*udev
,
4642 return __netdev_adjacent_dev_insert_link(dev
, udev
, master
, true);
4645 void __netdev_adjacent_dev_unlink(struct net_device
*dev
,
4646 struct net_device
*upper_dev
)
4648 __netdev_upper_dev_remove(dev
, upper_dev
);
4649 __netdev_lower_dev_remove(upper_dev
, dev
);
4653 static int __netdev_upper_dev_link(struct net_device
*dev
,
4654 struct net_device
*upper_dev
, bool master
)
4656 struct netdev_adjacent
*i
, *j
, *to_i
, *to_j
;
4661 if (dev
== upper_dev
)
4664 /* To prevent loops, check if dev is not upper device to upper_dev. */
4665 if (__netdev_find_upper(upper_dev
, dev
))
4668 if (__netdev_find_upper(dev
, upper_dev
))
4671 if (master
&& netdev_master_upper_dev_get(dev
))
4674 ret
= __netdev_adjacent_dev_link_neighbour(dev
, upper_dev
, master
);
4678 /* Now that we linked these devs, make all the upper_dev's
4679 * upper_dev_list visible to every dev's lower_dev_list and vice
4680 * versa, and don't forget the devices itself. All of these
4681 * links are non-neighbours.
4683 list_for_each_entry(i
, &dev
->lower_dev_list
, list
) {
4684 list_for_each_entry(j
, &upper_dev
->upper_dev_list
, list
) {
4685 ret
= __netdev_adjacent_dev_link(i
->dev
, j
->dev
);
4691 /* add dev to every upper_dev's upper device */
4692 list_for_each_entry(i
, &upper_dev
->upper_dev_list
, list
) {
4693 ret
= __netdev_adjacent_dev_link(dev
, i
->dev
);
4695 goto rollback_upper_mesh
;
4698 /* add upper_dev to every dev's lower device */
4699 list_for_each_entry(i
, &dev
->lower_dev_list
, list
) {
4700 ret
= __netdev_adjacent_dev_link(i
->dev
, upper_dev
);
4702 goto rollback_lower_mesh
;
4705 call_netdevice_notifiers(NETDEV_CHANGEUPPER
, dev
);
4708 rollback_lower_mesh
:
4710 list_for_each_entry(i
, &dev
->lower_dev_list
, list
) {
4713 __netdev_adjacent_dev_unlink(i
->dev
, upper_dev
);
4718 rollback_upper_mesh
:
4720 list_for_each_entry(i
, &upper_dev
->upper_dev_list
, list
) {
4723 __netdev_adjacent_dev_unlink(dev
, i
->dev
);
4731 list_for_each_entry(i
, &dev
->lower_dev_list
, list
) {
4732 list_for_each_entry(j
, &upper_dev
->upper_dev_list
, list
) {
4733 if (i
== to_i
&& j
== to_j
)
4735 __netdev_adjacent_dev_unlink(i
->dev
, j
->dev
);
4741 __netdev_adjacent_dev_unlink(dev
, upper_dev
);
4747 * netdev_upper_dev_link - Add a link to the upper device
4749 * @upper_dev: new upper device
4751 * Adds a link to device which is upper to this one. The caller must hold
4752 * the RTNL lock. On a failure a negative errno code is returned.
4753 * On success the reference counts are adjusted and the function
4756 int netdev_upper_dev_link(struct net_device
*dev
,
4757 struct net_device
*upper_dev
)
4759 return __netdev_upper_dev_link(dev
, upper_dev
, false);
4761 EXPORT_SYMBOL(netdev_upper_dev_link
);
4764 * netdev_master_upper_dev_link - Add a master link to the upper device
4766 * @upper_dev: new upper device
4768 * Adds a link to device which is upper to this one. In this case, only
4769 * one master upper device can be linked, although other non-master devices
4770 * might be linked as well. The caller must hold the RTNL lock.
4771 * On a failure a negative errno code is returned. On success the reference
4772 * counts are adjusted and the function returns zero.
4774 int netdev_master_upper_dev_link(struct net_device
*dev
,
4775 struct net_device
*upper_dev
)
4777 return __netdev_upper_dev_link(dev
, upper_dev
, true);
4779 EXPORT_SYMBOL(netdev_master_upper_dev_link
);
4782 * netdev_upper_dev_unlink - Removes a link to upper device
4784 * @upper_dev: new upper device
4786 * Removes a link to device which is upper to this one. The caller must hold
4789 void netdev_upper_dev_unlink(struct net_device
*dev
,
4790 struct net_device
*upper_dev
)
4792 struct netdev_adjacent
*i
, *j
;
4795 __netdev_adjacent_dev_unlink(dev
, upper_dev
);
4797 /* Here is the tricky part. We must remove all dev's lower
4798 * devices from all upper_dev's upper devices and vice
4799 * versa, to maintain the graph relationship.
4801 list_for_each_entry(i
, &dev
->lower_dev_list
, list
)
4802 list_for_each_entry(j
, &upper_dev
->upper_dev_list
, list
)
4803 __netdev_adjacent_dev_unlink(i
->dev
, j
->dev
);
4805 /* remove also the devices itself from lower/upper device
4808 list_for_each_entry(i
, &dev
->lower_dev_list
, list
)
4809 __netdev_adjacent_dev_unlink(i
->dev
, upper_dev
);
4811 list_for_each_entry(i
, &upper_dev
->upper_dev_list
, list
)
4812 __netdev_adjacent_dev_unlink(dev
, i
->dev
);
4814 call_netdevice_notifiers(NETDEV_CHANGEUPPER
, dev
);
4816 EXPORT_SYMBOL(netdev_upper_dev_unlink
);
4818 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
4820 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4822 if ((dev
->flags
& IFF_UP
) && ops
->ndo_change_rx_flags
)
4823 ops
->ndo_change_rx_flags(dev
, flags
);
4826 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
4828 unsigned int old_flags
= dev
->flags
;
4834 dev
->flags
|= IFF_PROMISC
;
4835 dev
->promiscuity
+= inc
;
4836 if (dev
->promiscuity
== 0) {
4839 * If inc causes overflow, untouch promisc and return error.
4842 dev
->flags
&= ~IFF_PROMISC
;
4844 dev
->promiscuity
-= inc
;
4845 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4850 if (dev
->flags
!= old_flags
) {
4851 pr_info("device %s %s promiscuous mode\n",
4853 dev
->flags
& IFF_PROMISC
? "entered" : "left");
4854 if (audit_enabled
) {
4855 current_uid_gid(&uid
, &gid
);
4856 audit_log(current
->audit_context
, GFP_ATOMIC
,
4857 AUDIT_ANOM_PROMISCUOUS
,
4858 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4859 dev
->name
, (dev
->flags
& IFF_PROMISC
),
4860 (old_flags
& IFF_PROMISC
),
4861 from_kuid(&init_user_ns
, audit_get_loginuid(current
)),
4862 from_kuid(&init_user_ns
, uid
),
4863 from_kgid(&init_user_ns
, gid
),
4864 audit_get_sessionid(current
));
4867 dev_change_rx_flags(dev
, IFF_PROMISC
);
4873 * dev_set_promiscuity - update promiscuity count on a device
4877 * Add or remove promiscuity from a device. While the count in the device
4878 * remains above zero the interface remains promiscuous. Once it hits zero
4879 * the device reverts back to normal filtering operation. A negative inc
4880 * value is used to drop promiscuity on the device.
4881 * Return 0 if successful or a negative errno code on error.
4883 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
4885 unsigned int old_flags
= dev
->flags
;
4888 err
= __dev_set_promiscuity(dev
, inc
);
4891 if (dev
->flags
!= old_flags
)
4892 dev_set_rx_mode(dev
);
4895 EXPORT_SYMBOL(dev_set_promiscuity
);
4898 * dev_set_allmulti - update allmulti count on a device
4902 * Add or remove reception of all multicast frames to a device. While the
4903 * count in the device remains above zero the interface remains listening
4904 * to all interfaces. Once it hits zero the device reverts back to normal
4905 * filtering operation. A negative @inc value is used to drop the counter
4906 * when releasing a resource needing all multicasts.
4907 * Return 0 if successful or a negative errno code on error.
4910 int dev_set_allmulti(struct net_device
*dev
, int inc
)
4912 unsigned int old_flags
= dev
->flags
;
4916 dev
->flags
|= IFF_ALLMULTI
;
4917 dev
->allmulti
+= inc
;
4918 if (dev
->allmulti
== 0) {
4921 * If inc causes overflow, untouch allmulti and return error.
4924 dev
->flags
&= ~IFF_ALLMULTI
;
4926 dev
->allmulti
-= inc
;
4927 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4932 if (dev
->flags
^ old_flags
) {
4933 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
4934 dev_set_rx_mode(dev
);
4938 EXPORT_SYMBOL(dev_set_allmulti
);
4941 * Upload unicast and multicast address lists to device and
4942 * configure RX filtering. When the device doesn't support unicast
4943 * filtering it is put in promiscuous mode while unicast addresses
4946 void __dev_set_rx_mode(struct net_device
*dev
)
4948 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4950 /* dev_open will call this function so the list will stay sane. */
4951 if (!(dev
->flags
&IFF_UP
))
4954 if (!netif_device_present(dev
))
4957 if (!(dev
->priv_flags
& IFF_UNICAST_FLT
)) {
4958 /* Unicast addresses changes may only happen under the rtnl,
4959 * therefore calling __dev_set_promiscuity here is safe.
4961 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
4962 __dev_set_promiscuity(dev
, 1);
4963 dev
->uc_promisc
= true;
4964 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
4965 __dev_set_promiscuity(dev
, -1);
4966 dev
->uc_promisc
= false;
4970 if (ops
->ndo_set_rx_mode
)
4971 ops
->ndo_set_rx_mode(dev
);
4974 void dev_set_rx_mode(struct net_device
*dev
)
4976 netif_addr_lock_bh(dev
);
4977 __dev_set_rx_mode(dev
);
4978 netif_addr_unlock_bh(dev
);
4982 * dev_get_flags - get flags reported to userspace
4985 * Get the combination of flag bits exported through APIs to userspace.
4987 unsigned int dev_get_flags(const struct net_device
*dev
)
4991 flags
= (dev
->flags
& ~(IFF_PROMISC
|
4996 (dev
->gflags
& (IFF_PROMISC
|
4999 if (netif_running(dev
)) {
5000 if (netif_oper_up(dev
))
5001 flags
|= IFF_RUNNING
;
5002 if (netif_carrier_ok(dev
))
5003 flags
|= IFF_LOWER_UP
;
5004 if (netif_dormant(dev
))
5005 flags
|= IFF_DORMANT
;
5010 EXPORT_SYMBOL(dev_get_flags
);
5012 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
5014 unsigned int old_flags
= dev
->flags
;
5020 * Set the flags on our device.
5023 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
5024 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
5026 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
5030 * Load in the correct multicast list now the flags have changed.
5033 if ((old_flags
^ flags
) & IFF_MULTICAST
)
5034 dev_change_rx_flags(dev
, IFF_MULTICAST
);
5036 dev_set_rx_mode(dev
);
5039 * Have we downed the interface. We handle IFF_UP ourselves
5040 * according to user attempts to set it, rather than blindly
5045 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
5046 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
5049 dev_set_rx_mode(dev
);
5052 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
5053 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
5055 dev
->gflags
^= IFF_PROMISC
;
5056 dev_set_promiscuity(dev
, inc
);
5059 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5060 is important. Some (broken) drivers set IFF_PROMISC, when
5061 IFF_ALLMULTI is requested not asking us and not reporting.
5063 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
5064 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
5066 dev
->gflags
^= IFF_ALLMULTI
;
5067 dev_set_allmulti(dev
, inc
);
5073 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
)
5075 unsigned int changes
= dev
->flags
^ old_flags
;
5077 if (changes
& IFF_UP
) {
5078 if (dev
->flags
& IFF_UP
)
5079 call_netdevice_notifiers(NETDEV_UP
, dev
);
5081 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
5084 if (dev
->flags
& IFF_UP
&&
5085 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
))) {
5086 struct netdev_notifier_change_info change_info
;
5088 change_info
.flags_changed
= changes
;
5089 call_netdevice_notifiers_info(NETDEV_CHANGE
, dev
,
5095 * dev_change_flags - change device settings
5097 * @flags: device state flags
5099 * Change settings on device based state flags. The flags are
5100 * in the userspace exported format.
5102 int dev_change_flags(struct net_device
*dev
, unsigned int flags
)
5105 unsigned int changes
, old_flags
= dev
->flags
;
5107 ret
= __dev_change_flags(dev
, flags
);
5111 changes
= old_flags
^ dev
->flags
;
5113 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
5115 __dev_notify_flags(dev
, old_flags
);
5118 EXPORT_SYMBOL(dev_change_flags
);
5121 * dev_set_mtu - Change maximum transfer unit
5123 * @new_mtu: new transfer unit
5125 * Change the maximum transfer size of the network device.
5127 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
5129 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5132 if (new_mtu
== dev
->mtu
)
5135 /* MTU must be positive. */
5139 if (!netif_device_present(dev
))
5143 if (ops
->ndo_change_mtu
)
5144 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
5149 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
5152 EXPORT_SYMBOL(dev_set_mtu
);
5155 * dev_set_group - Change group this device belongs to
5157 * @new_group: group this device should belong to
5159 void dev_set_group(struct net_device
*dev
, int new_group
)
5161 dev
->group
= new_group
;
5163 EXPORT_SYMBOL(dev_set_group
);
5166 * dev_set_mac_address - Change Media Access Control Address
5170 * Change the hardware (MAC) address of the device
5172 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
5174 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5177 if (!ops
->ndo_set_mac_address
)
5179 if (sa
->sa_family
!= dev
->type
)
5181 if (!netif_device_present(dev
))
5183 err
= ops
->ndo_set_mac_address(dev
, sa
);
5186 dev
->addr_assign_type
= NET_ADDR_SET
;
5187 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
5188 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
5191 EXPORT_SYMBOL(dev_set_mac_address
);
5194 * dev_change_carrier - Change device carrier
5196 * @new_carrier: new value
5198 * Change device carrier
5200 int dev_change_carrier(struct net_device
*dev
, bool new_carrier
)
5202 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5204 if (!ops
->ndo_change_carrier
)
5206 if (!netif_device_present(dev
))
5208 return ops
->ndo_change_carrier(dev
, new_carrier
);
5210 EXPORT_SYMBOL(dev_change_carrier
);
5213 * dev_get_phys_port_id - Get device physical port ID
5217 * Get device physical port ID
5219 int dev_get_phys_port_id(struct net_device
*dev
,
5220 struct netdev_phys_port_id
*ppid
)
5222 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5224 if (!ops
->ndo_get_phys_port_id
)
5226 return ops
->ndo_get_phys_port_id(dev
, ppid
);
5228 EXPORT_SYMBOL(dev_get_phys_port_id
);
5231 * dev_new_index - allocate an ifindex
5232 * @net: the applicable net namespace
5234 * Returns a suitable unique value for a new device interface
5235 * number. The caller must hold the rtnl semaphore or the
5236 * dev_base_lock to be sure it remains unique.
5238 static int dev_new_index(struct net
*net
)
5240 int ifindex
= net
->ifindex
;
5244 if (!__dev_get_by_index(net
, ifindex
))
5245 return net
->ifindex
= ifindex
;
5249 /* Delayed registration/unregisteration */
5250 static LIST_HEAD(net_todo_list
);
5251 static DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq
);
5253 static void net_set_todo(struct net_device
*dev
)
5255 list_add_tail(&dev
->todo_list
, &net_todo_list
);
5256 dev_net(dev
)->dev_unreg_count
++;
5259 static void rollback_registered_many(struct list_head
*head
)
5261 struct net_device
*dev
, *tmp
;
5263 BUG_ON(dev_boot_phase
);
5266 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
5267 /* Some devices call without registering
5268 * for initialization unwind. Remove those
5269 * devices and proceed with the remaining.
5271 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5272 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5276 list_del(&dev
->unreg_list
);
5279 dev
->dismantle
= true;
5280 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
5283 /* If device is running, close it first. */
5284 dev_close_many(head
);
5286 list_for_each_entry(dev
, head
, unreg_list
) {
5287 /* And unlink it from device chain. */
5288 unlist_netdevice(dev
);
5290 dev
->reg_state
= NETREG_UNREGISTERING
;
5295 list_for_each_entry(dev
, head
, unreg_list
) {
5296 /* Shutdown queueing discipline. */
5300 /* Notify protocols, that we are about to destroy
5301 this device. They should clean all the things.
5303 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5305 if (!dev
->rtnl_link_ops
||
5306 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5307 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
5310 * Flush the unicast and multicast chains
5315 if (dev
->netdev_ops
->ndo_uninit
)
5316 dev
->netdev_ops
->ndo_uninit(dev
);
5318 /* Notifier chain MUST detach us all upper devices. */
5319 WARN_ON(netdev_has_any_upper_dev(dev
));
5321 /* Remove entries from kobject tree */
5322 netdev_unregister_kobject(dev
);
5324 /* Remove XPS queueing entries */
5325 netif_reset_xps_queues_gt(dev
, 0);
5331 list_for_each_entry(dev
, head
, unreg_list
)
5335 static void rollback_registered(struct net_device
*dev
)
5339 list_add(&dev
->unreg_list
, &single
);
5340 rollback_registered_many(&single
);
5344 static netdev_features_t
netdev_fix_features(struct net_device
*dev
,
5345 netdev_features_t features
)
5347 /* Fix illegal checksum combinations */
5348 if ((features
& NETIF_F_HW_CSUM
) &&
5349 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5350 netdev_warn(dev
, "mixed HW and IP checksum settings.\n");
5351 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5354 /* TSO requires that SG is present as well. */
5355 if ((features
& NETIF_F_ALL_TSO
) && !(features
& NETIF_F_SG
)) {
5356 netdev_dbg(dev
, "Dropping TSO features since no SG feature.\n");
5357 features
&= ~NETIF_F_ALL_TSO
;
5360 if ((features
& NETIF_F_TSO
) && !(features
& NETIF_F_HW_CSUM
) &&
5361 !(features
& NETIF_F_IP_CSUM
)) {
5362 netdev_dbg(dev
, "Dropping TSO features since no CSUM feature.\n");
5363 features
&= ~NETIF_F_TSO
;
5364 features
&= ~NETIF_F_TSO_ECN
;
5367 if ((features
& NETIF_F_TSO6
) && !(features
& NETIF_F_HW_CSUM
) &&
5368 !(features
& NETIF_F_IPV6_CSUM
)) {
5369 netdev_dbg(dev
, "Dropping TSO6 features since no CSUM feature.\n");
5370 features
&= ~NETIF_F_TSO6
;
5373 /* TSO ECN requires that TSO is present as well. */
5374 if ((features
& NETIF_F_ALL_TSO
) == NETIF_F_TSO_ECN
)
5375 features
&= ~NETIF_F_TSO_ECN
;
5377 /* Software GSO depends on SG. */
5378 if ((features
& NETIF_F_GSO
) && !(features
& NETIF_F_SG
)) {
5379 netdev_dbg(dev
, "Dropping NETIF_F_GSO since no SG feature.\n");
5380 features
&= ~NETIF_F_GSO
;
5383 /* UFO needs SG and checksumming */
5384 if (features
& NETIF_F_UFO
) {
5385 /* maybe split UFO into V4 and V6? */
5386 if (!((features
& NETIF_F_GEN_CSUM
) ||
5387 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))
5388 == (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5390 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5391 features
&= ~NETIF_F_UFO
;
5394 if (!(features
& NETIF_F_SG
)) {
5396 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5397 features
&= ~NETIF_F_UFO
;
5404 int __netdev_update_features(struct net_device
*dev
)
5406 netdev_features_t features
;
5411 features
= netdev_get_wanted_features(dev
);
5413 if (dev
->netdev_ops
->ndo_fix_features
)
5414 features
= dev
->netdev_ops
->ndo_fix_features(dev
, features
);
5416 /* driver might be less strict about feature dependencies */
5417 features
= netdev_fix_features(dev
, features
);
5419 if (dev
->features
== features
)
5422 netdev_dbg(dev
, "Features changed: %pNF -> %pNF\n",
5423 &dev
->features
, &features
);
5425 if (dev
->netdev_ops
->ndo_set_features
)
5426 err
= dev
->netdev_ops
->ndo_set_features(dev
, features
);
5428 if (unlikely(err
< 0)) {
5430 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5431 err
, &features
, &dev
->features
);
5436 dev
->features
= features
;
5442 * netdev_update_features - recalculate device features
5443 * @dev: the device to check
5445 * Recalculate dev->features set and send notifications if it
5446 * has changed. Should be called after driver or hardware dependent
5447 * conditions might have changed that influence the features.
5449 void netdev_update_features(struct net_device
*dev
)
5451 if (__netdev_update_features(dev
))
5452 netdev_features_change(dev
);
5454 EXPORT_SYMBOL(netdev_update_features
);
5457 * netdev_change_features - recalculate device features
5458 * @dev: the device to check
5460 * Recalculate dev->features set and send notifications even
5461 * if they have not changed. Should be called instead of
5462 * netdev_update_features() if also dev->vlan_features might
5463 * have changed to allow the changes to be propagated to stacked
5466 void netdev_change_features(struct net_device
*dev
)
5468 __netdev_update_features(dev
);
5469 netdev_features_change(dev
);
5471 EXPORT_SYMBOL(netdev_change_features
);
5474 * netif_stacked_transfer_operstate - transfer operstate
5475 * @rootdev: the root or lower level device to transfer state from
5476 * @dev: the device to transfer operstate to
5478 * Transfer operational state from root to device. This is normally
5479 * called when a stacking relationship exists between the root
5480 * device and the device(a leaf device).
5482 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
5483 struct net_device
*dev
)
5485 if (rootdev
->operstate
== IF_OPER_DORMANT
)
5486 netif_dormant_on(dev
);
5488 netif_dormant_off(dev
);
5490 if (netif_carrier_ok(rootdev
)) {
5491 if (!netif_carrier_ok(dev
))
5492 netif_carrier_on(dev
);
5494 if (netif_carrier_ok(dev
))
5495 netif_carrier_off(dev
);
5498 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
5501 static int netif_alloc_rx_queues(struct net_device
*dev
)
5503 unsigned int i
, count
= dev
->num_rx_queues
;
5504 struct netdev_rx_queue
*rx
;
5508 rx
= kcalloc(count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5514 for (i
= 0; i
< count
; i
++)
5520 static void netdev_init_one_queue(struct net_device
*dev
,
5521 struct netdev_queue
*queue
, void *_unused
)
5523 /* Initialize queue lock */
5524 spin_lock_init(&queue
->_xmit_lock
);
5525 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
5526 queue
->xmit_lock_owner
= -1;
5527 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
5530 dql_init(&queue
->dql
, HZ
);
5534 static void netif_free_tx_queues(struct net_device
*dev
)
5536 if (is_vmalloc_addr(dev
->_tx
))
5542 static int netif_alloc_netdev_queues(struct net_device
*dev
)
5544 unsigned int count
= dev
->num_tx_queues
;
5545 struct netdev_queue
*tx
;
5546 size_t sz
= count
* sizeof(*tx
);
5548 BUG_ON(count
< 1 || count
> 0xffff);
5550 tx
= kzalloc(sz
, GFP_KERNEL
| __GFP_NOWARN
| __GFP_REPEAT
);
5558 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5559 spin_lock_init(&dev
->tx_global_lock
);
5565 * register_netdevice - register a network device
5566 * @dev: device to register
5568 * Take a completed network device structure and add it to the kernel
5569 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5570 * chain. 0 is returned on success. A negative errno code is returned
5571 * on a failure to set up the device, or if the name is a duplicate.
5573 * Callers must hold the rtnl semaphore. You may want
5574 * register_netdev() instead of this.
5577 * The locking appears insufficient to guarantee two parallel registers
5578 * will not get the same name.
5581 int register_netdevice(struct net_device
*dev
)
5584 struct net
*net
= dev_net(dev
);
5586 BUG_ON(dev_boot_phase
);
5591 /* When net_device's are persistent, this will be fatal. */
5592 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
5595 spin_lock_init(&dev
->addr_list_lock
);
5596 netdev_set_addr_lockdep_class(dev
);
5600 ret
= dev_get_valid_name(net
, dev
, dev
->name
);
5604 /* Init, if this function is available */
5605 if (dev
->netdev_ops
->ndo_init
) {
5606 ret
= dev
->netdev_ops
->ndo_init(dev
);
5614 if (((dev
->hw_features
| dev
->features
) &
5615 NETIF_F_HW_VLAN_CTAG_FILTER
) &&
5616 (!dev
->netdev_ops
->ndo_vlan_rx_add_vid
||
5617 !dev
->netdev_ops
->ndo_vlan_rx_kill_vid
)) {
5618 netdev_WARN(dev
, "Buggy VLAN acceleration in driver!\n");
5625 dev
->ifindex
= dev_new_index(net
);
5626 else if (__dev_get_by_index(net
, dev
->ifindex
))
5629 if (dev
->iflink
== -1)
5630 dev
->iflink
= dev
->ifindex
;
5632 /* Transfer changeable features to wanted_features and enable
5633 * software offloads (GSO and GRO).
5635 dev
->hw_features
|= NETIF_F_SOFT_FEATURES
;
5636 dev
->features
|= NETIF_F_SOFT_FEATURES
;
5637 dev
->wanted_features
= dev
->features
& dev
->hw_features
;
5639 /* Turn on no cache copy if HW is doing checksum */
5640 if (!(dev
->flags
& IFF_LOOPBACK
)) {
5641 dev
->hw_features
|= NETIF_F_NOCACHE_COPY
;
5642 if (dev
->features
& NETIF_F_ALL_CSUM
) {
5643 dev
->wanted_features
|= NETIF_F_NOCACHE_COPY
;
5644 dev
->features
|= NETIF_F_NOCACHE_COPY
;
5648 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5650 dev
->vlan_features
|= NETIF_F_HIGHDMA
;
5652 /* Make NETIF_F_SG inheritable to tunnel devices.
5654 dev
->hw_enc_features
|= NETIF_F_SG
;
5656 /* Make NETIF_F_SG inheritable to MPLS.
5658 dev
->mpls_features
|= NETIF_F_SG
;
5660 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
5661 ret
= notifier_to_errno(ret
);
5665 ret
= netdev_register_kobject(dev
);
5668 dev
->reg_state
= NETREG_REGISTERED
;
5670 __netdev_update_features(dev
);
5673 * Default initial state at registry is that the
5674 * device is present.
5677 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5679 linkwatch_init_dev(dev
);
5681 dev_init_scheduler(dev
);
5683 list_netdevice(dev
);
5684 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
5686 /* If the device has permanent device address, driver should
5687 * set dev_addr and also addr_assign_type should be set to
5688 * NET_ADDR_PERM (default value).
5690 if (dev
->addr_assign_type
== NET_ADDR_PERM
)
5691 memcpy(dev
->perm_addr
, dev
->dev_addr
, dev
->addr_len
);
5693 /* Notify protocols, that a new device appeared. */
5694 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5695 ret
= notifier_to_errno(ret
);
5697 rollback_registered(dev
);
5698 dev
->reg_state
= NETREG_UNREGISTERED
;
5701 * Prevent userspace races by waiting until the network
5702 * device is fully setup before sending notifications.
5704 if (!dev
->rtnl_link_ops
||
5705 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5706 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5712 if (dev
->netdev_ops
->ndo_uninit
)
5713 dev
->netdev_ops
->ndo_uninit(dev
);
5716 EXPORT_SYMBOL(register_netdevice
);
5719 * init_dummy_netdev - init a dummy network device for NAPI
5720 * @dev: device to init
5722 * This takes a network device structure and initialize the minimum
5723 * amount of fields so it can be used to schedule NAPI polls without
5724 * registering a full blown interface. This is to be used by drivers
5725 * that need to tie several hardware interfaces to a single NAPI
5726 * poll scheduler due to HW limitations.
5728 int init_dummy_netdev(struct net_device
*dev
)
5730 /* Clear everything. Note we don't initialize spinlocks
5731 * are they aren't supposed to be taken by any of the
5732 * NAPI code and this dummy netdev is supposed to be
5733 * only ever used for NAPI polls
5735 memset(dev
, 0, sizeof(struct net_device
));
5737 /* make sure we BUG if trying to hit standard
5738 * register/unregister code path
5740 dev
->reg_state
= NETREG_DUMMY
;
5742 /* NAPI wants this */
5743 INIT_LIST_HEAD(&dev
->napi_list
);
5745 /* a dummy interface is started by default */
5746 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5747 set_bit(__LINK_STATE_START
, &dev
->state
);
5749 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5750 * because users of this 'device' dont need to change
5756 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
5760 * register_netdev - register a network device
5761 * @dev: device to register
5763 * Take a completed network device structure and add it to the kernel
5764 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5765 * chain. 0 is returned on success. A negative errno code is returned
5766 * on a failure to set up the device, or if the name is a duplicate.
5768 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5769 * and expands the device name if you passed a format string to
5772 int register_netdev(struct net_device
*dev
)
5777 err
= register_netdevice(dev
);
5781 EXPORT_SYMBOL(register_netdev
);
5783 int netdev_refcnt_read(const struct net_device
*dev
)
5787 for_each_possible_cpu(i
)
5788 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
5791 EXPORT_SYMBOL(netdev_refcnt_read
);
5794 * netdev_wait_allrefs - wait until all references are gone.
5795 * @dev: target net_device
5797 * This is called when unregistering network devices.
5799 * Any protocol or device that holds a reference should register
5800 * for netdevice notification, and cleanup and put back the
5801 * reference if they receive an UNREGISTER event.
5802 * We can get stuck here if buggy protocols don't correctly
5805 static void netdev_wait_allrefs(struct net_device
*dev
)
5807 unsigned long rebroadcast_time
, warning_time
;
5810 linkwatch_forget_dev(dev
);
5812 rebroadcast_time
= warning_time
= jiffies
;
5813 refcnt
= netdev_refcnt_read(dev
);
5815 while (refcnt
!= 0) {
5816 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
5819 /* Rebroadcast unregister notification */
5820 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5826 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
5827 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
5829 /* We must not have linkwatch events
5830 * pending on unregister. If this
5831 * happens, we simply run the queue
5832 * unscheduled, resulting in a noop
5835 linkwatch_run_queue();
5840 rebroadcast_time
= jiffies
;
5845 refcnt
= netdev_refcnt_read(dev
);
5847 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
5848 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5850 warning_time
= jiffies
;
5859 * register_netdevice(x1);
5860 * register_netdevice(x2);
5862 * unregister_netdevice(y1);
5863 * unregister_netdevice(y2);
5869 * We are invoked by rtnl_unlock().
5870 * This allows us to deal with problems:
5871 * 1) We can delete sysfs objects which invoke hotplug
5872 * without deadlocking with linkwatch via keventd.
5873 * 2) Since we run with the RTNL semaphore not held, we can sleep
5874 * safely in order to wait for the netdev refcnt to drop to zero.
5876 * We must not return until all unregister events added during
5877 * the interval the lock was held have been completed.
5879 void netdev_run_todo(void)
5881 struct list_head list
;
5883 /* Snapshot list, allow later requests */
5884 list_replace_init(&net_todo_list
, &list
);
5889 /* Wait for rcu callbacks to finish before next phase */
5890 if (!list_empty(&list
))
5893 while (!list_empty(&list
)) {
5894 struct net_device
*dev
5895 = list_first_entry(&list
, struct net_device
, todo_list
);
5896 list_del(&dev
->todo_list
);
5899 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
5902 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
5903 pr_err("network todo '%s' but state %d\n",
5904 dev
->name
, dev
->reg_state
);
5909 dev
->reg_state
= NETREG_UNREGISTERED
;
5911 on_each_cpu(flush_backlog
, dev
, 1);
5913 netdev_wait_allrefs(dev
);
5916 BUG_ON(netdev_refcnt_read(dev
));
5917 WARN_ON(rcu_access_pointer(dev
->ip_ptr
));
5918 WARN_ON(rcu_access_pointer(dev
->ip6_ptr
));
5919 WARN_ON(dev
->dn_ptr
);
5921 if (dev
->destructor
)
5922 dev
->destructor(dev
);
5924 /* Report a network device has been unregistered */
5926 dev_net(dev
)->dev_unreg_count
--;
5928 wake_up(&netdev_unregistering_wq
);
5930 /* Free network device */
5931 kobject_put(&dev
->dev
.kobj
);
5935 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5936 * fields in the same order, with only the type differing.
5938 void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
5939 const struct net_device_stats
*netdev_stats
)
5941 #if BITS_PER_LONG == 64
5942 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
5943 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
5945 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
5946 const unsigned long *src
= (const unsigned long *)netdev_stats
;
5947 u64
*dst
= (u64
*)stats64
;
5949 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
5950 sizeof(*stats64
) / sizeof(u64
));
5951 for (i
= 0; i
< n
; i
++)
5955 EXPORT_SYMBOL(netdev_stats_to_stats64
);
5958 * dev_get_stats - get network device statistics
5959 * @dev: device to get statistics from
5960 * @storage: place to store stats
5962 * Get network statistics from device. Return @storage.
5963 * The device driver may provide its own method by setting
5964 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5965 * otherwise the internal statistics structure is used.
5967 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
5968 struct rtnl_link_stats64
*storage
)
5970 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5972 if (ops
->ndo_get_stats64
) {
5973 memset(storage
, 0, sizeof(*storage
));
5974 ops
->ndo_get_stats64(dev
, storage
);
5975 } else if (ops
->ndo_get_stats
) {
5976 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
5978 netdev_stats_to_stats64(storage
, &dev
->stats
);
5980 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
5983 EXPORT_SYMBOL(dev_get_stats
);
5985 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
5987 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
5989 #ifdef CONFIG_NET_CLS_ACT
5992 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
5995 netdev_init_one_queue(dev
, queue
, NULL
);
5996 queue
->qdisc
= &noop_qdisc
;
5997 queue
->qdisc_sleeping
= &noop_qdisc
;
5998 rcu_assign_pointer(dev
->ingress_queue
, queue
);
6003 static const struct ethtool_ops default_ethtool_ops
;
6005 void netdev_set_default_ethtool_ops(struct net_device
*dev
,
6006 const struct ethtool_ops
*ops
)
6008 if (dev
->ethtool_ops
== &default_ethtool_ops
)
6009 dev
->ethtool_ops
= ops
;
6011 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops
);
6014 * alloc_netdev_mqs - allocate network device
6015 * @sizeof_priv: size of private data to allocate space for
6016 * @name: device name format string
6017 * @setup: callback to initialize device
6018 * @txqs: the number of TX subqueues to allocate
6019 * @rxqs: the number of RX subqueues to allocate
6021 * Allocates a struct net_device with private data area for driver use
6022 * and performs basic initialization. Also allocates subquue structs
6023 * for each queue on the device.
6025 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
6026 void (*setup
)(struct net_device
*),
6027 unsigned int txqs
, unsigned int rxqs
)
6029 struct net_device
*dev
;
6031 struct net_device
*p
;
6033 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
6036 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6042 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6047 alloc_size
= sizeof(struct net_device
);
6049 /* ensure 32-byte alignment of private area */
6050 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
6051 alloc_size
+= sizeof_priv
;
6053 /* ensure 32-byte alignment of whole construct */
6054 alloc_size
+= NETDEV_ALIGN
- 1;
6056 p
= kzalloc(alloc_size
, GFP_KERNEL
);
6060 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
6061 dev
->padded
= (char *)dev
- (char *)p
;
6063 dev
->pcpu_refcnt
= alloc_percpu(int);
6064 if (!dev
->pcpu_refcnt
)
6067 if (dev_addr_init(dev
))
6073 dev_net_set(dev
, &init_net
);
6075 dev
->gso_max_size
= GSO_MAX_SIZE
;
6076 dev
->gso_max_segs
= GSO_MAX_SEGS
;
6078 INIT_LIST_HEAD(&dev
->napi_list
);
6079 INIT_LIST_HEAD(&dev
->unreg_list
);
6080 INIT_LIST_HEAD(&dev
->link_watch_list
);
6081 INIT_LIST_HEAD(&dev
->upper_dev_list
);
6082 INIT_LIST_HEAD(&dev
->lower_dev_list
);
6083 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
6086 dev
->num_tx_queues
= txqs
;
6087 dev
->real_num_tx_queues
= txqs
;
6088 if (netif_alloc_netdev_queues(dev
))
6092 dev
->num_rx_queues
= rxqs
;
6093 dev
->real_num_rx_queues
= rxqs
;
6094 if (netif_alloc_rx_queues(dev
))
6098 strcpy(dev
->name
, name
);
6099 dev
->group
= INIT_NETDEV_GROUP
;
6100 if (!dev
->ethtool_ops
)
6101 dev
->ethtool_ops
= &default_ethtool_ops
;
6109 free_percpu(dev
->pcpu_refcnt
);
6110 netif_free_tx_queues(dev
);
6119 EXPORT_SYMBOL(alloc_netdev_mqs
);
6122 * free_netdev - free network device
6125 * This function does the last stage of destroying an allocated device
6126 * interface. The reference to the device object is released.
6127 * If this is the last reference then it will be freed.
6129 void free_netdev(struct net_device
*dev
)
6131 struct napi_struct
*p
, *n
;
6133 release_net(dev_net(dev
));
6135 netif_free_tx_queues(dev
);
6140 kfree(rcu_dereference_protected(dev
->ingress_queue
, 1));
6142 /* Flush device addresses */
6143 dev_addr_flush(dev
);
6145 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
6148 free_percpu(dev
->pcpu_refcnt
);
6149 dev
->pcpu_refcnt
= NULL
;
6151 /* Compatibility with error handling in drivers */
6152 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
6153 kfree((char *)dev
- dev
->padded
);
6157 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
6158 dev
->reg_state
= NETREG_RELEASED
;
6160 /* will free via device release */
6161 put_device(&dev
->dev
);
6163 EXPORT_SYMBOL(free_netdev
);
6166 * synchronize_net - Synchronize with packet receive processing
6168 * Wait for packets currently being received to be done.
6169 * Does not block later packets from starting.
6171 void synchronize_net(void)
6174 if (rtnl_is_locked())
6175 synchronize_rcu_expedited();
6179 EXPORT_SYMBOL(synchronize_net
);
6182 * unregister_netdevice_queue - remove device from the kernel
6186 * This function shuts down a device interface and removes it
6187 * from the kernel tables.
6188 * If head not NULL, device is queued to be unregistered later.
6190 * Callers must hold the rtnl semaphore. You may want
6191 * unregister_netdev() instead of this.
6194 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
6199 list_move_tail(&dev
->unreg_list
, head
);
6201 rollback_registered(dev
);
6202 /* Finish processing unregister after unlock */
6206 EXPORT_SYMBOL(unregister_netdevice_queue
);
6209 * unregister_netdevice_many - unregister many devices
6210 * @head: list of devices
6212 void unregister_netdevice_many(struct list_head
*head
)
6214 struct net_device
*dev
;
6216 if (!list_empty(head
)) {
6217 rollback_registered_many(head
);
6218 list_for_each_entry(dev
, head
, unreg_list
)
6222 EXPORT_SYMBOL(unregister_netdevice_many
);
6225 * unregister_netdev - remove device from the kernel
6228 * This function shuts down a device interface and removes it
6229 * from the kernel tables.
6231 * This is just a wrapper for unregister_netdevice that takes
6232 * the rtnl semaphore. In general you want to use this and not
6233 * unregister_netdevice.
6235 void unregister_netdev(struct net_device
*dev
)
6238 unregister_netdevice(dev
);
6241 EXPORT_SYMBOL(unregister_netdev
);
6244 * dev_change_net_namespace - move device to different nethost namespace
6246 * @net: network namespace
6247 * @pat: If not NULL name pattern to try if the current device name
6248 * is already taken in the destination network namespace.
6250 * This function shuts down a device interface and moves it
6251 * to a new network namespace. On success 0 is returned, on
6252 * a failure a netagive errno code is returned.
6254 * Callers must hold the rtnl semaphore.
6257 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
6263 /* Don't allow namespace local devices to be moved. */
6265 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6268 /* Ensure the device has been registrered */
6269 if (dev
->reg_state
!= NETREG_REGISTERED
)
6272 /* Get out if there is nothing todo */
6274 if (net_eq(dev_net(dev
), net
))
6277 /* Pick the destination device name, and ensure
6278 * we can use it in the destination network namespace.
6281 if (__dev_get_by_name(net
, dev
->name
)) {
6282 /* We get here if we can't use the current device name */
6285 if (dev_get_valid_name(net
, dev
, pat
) < 0)
6290 * And now a mini version of register_netdevice unregister_netdevice.
6293 /* If device is running close it first. */
6296 /* And unlink it from device chain */
6298 unlist_netdevice(dev
);
6302 /* Shutdown queueing discipline. */
6305 /* Notify protocols, that we are about to destroy
6306 this device. They should clean all the things.
6308 Note that dev->reg_state stays at NETREG_REGISTERED.
6309 This is wanted because this way 8021q and macvlan know
6310 the device is just moving and can keep their slaves up.
6312 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6314 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
6315 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
6318 * Flush the unicast and multicast chains
6323 /* Send a netdev-removed uevent to the old namespace */
6324 kobject_uevent(&dev
->dev
.kobj
, KOBJ_REMOVE
);
6326 /* Actually switch the network namespace */
6327 dev_net_set(dev
, net
);
6329 /* If there is an ifindex conflict assign a new one */
6330 if (__dev_get_by_index(net
, dev
->ifindex
)) {
6331 int iflink
= (dev
->iflink
== dev
->ifindex
);
6332 dev
->ifindex
= dev_new_index(net
);
6334 dev
->iflink
= dev
->ifindex
;
6337 /* Send a netdev-add uevent to the new namespace */
6338 kobject_uevent(&dev
->dev
.kobj
, KOBJ_ADD
);
6340 /* Fixup kobjects */
6341 err
= device_rename(&dev
->dev
, dev
->name
);
6344 /* Add the device back in the hashes */
6345 list_netdevice(dev
);
6347 /* Notify protocols, that a new device appeared. */
6348 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
6351 * Prevent userspace races by waiting until the network
6352 * device is fully setup before sending notifications.
6354 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
6361 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
6363 static int dev_cpu_callback(struct notifier_block
*nfb
,
6364 unsigned long action
,
6367 struct sk_buff
**list_skb
;
6368 struct sk_buff
*skb
;
6369 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
6370 struct softnet_data
*sd
, *oldsd
;
6372 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
6375 local_irq_disable();
6376 cpu
= smp_processor_id();
6377 sd
= &per_cpu(softnet_data
, cpu
);
6378 oldsd
= &per_cpu(softnet_data
, oldcpu
);
6380 /* Find end of our completion_queue. */
6381 list_skb
= &sd
->completion_queue
;
6383 list_skb
= &(*list_skb
)->next
;
6384 /* Append completion queue from offline CPU. */
6385 *list_skb
= oldsd
->completion_queue
;
6386 oldsd
->completion_queue
= NULL
;
6388 /* Append output queue from offline CPU. */
6389 if (oldsd
->output_queue
) {
6390 *sd
->output_queue_tailp
= oldsd
->output_queue
;
6391 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
6392 oldsd
->output_queue
= NULL
;
6393 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
6395 /* Append NAPI poll list from offline CPU. */
6396 if (!list_empty(&oldsd
->poll_list
)) {
6397 list_splice_init(&oldsd
->poll_list
, &sd
->poll_list
);
6398 raise_softirq_irqoff(NET_RX_SOFTIRQ
);
6401 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
6404 /* Process offline CPU's input_pkt_queue */
6405 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
6407 input_queue_head_incr(oldsd
);
6409 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
6411 input_queue_head_incr(oldsd
);
6419 * netdev_increment_features - increment feature set by one
6420 * @all: current feature set
6421 * @one: new feature set
6422 * @mask: mask feature set
6424 * Computes a new feature set after adding a device with feature set
6425 * @one to the master device with current feature set @all. Will not
6426 * enable anything that is off in @mask. Returns the new feature set.
6428 netdev_features_t
netdev_increment_features(netdev_features_t all
,
6429 netdev_features_t one
, netdev_features_t mask
)
6431 if (mask
& NETIF_F_GEN_CSUM
)
6432 mask
|= NETIF_F_ALL_CSUM
;
6433 mask
|= NETIF_F_VLAN_CHALLENGED
;
6435 all
|= one
& (NETIF_F_ONE_FOR_ALL
|NETIF_F_ALL_CSUM
) & mask
;
6436 all
&= one
| ~NETIF_F_ALL_FOR_ALL
;
6438 /* If one device supports hw checksumming, set for all. */
6439 if (all
& NETIF_F_GEN_CSUM
)
6440 all
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_GEN_CSUM
);
6444 EXPORT_SYMBOL(netdev_increment_features
);
6446 static struct hlist_head
* __net_init
netdev_create_hash(void)
6449 struct hlist_head
*hash
;
6451 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
6453 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
6454 INIT_HLIST_HEAD(&hash
[i
]);
6459 /* Initialize per network namespace state */
6460 static int __net_init
netdev_init(struct net
*net
)
6462 if (net
!= &init_net
)
6463 INIT_LIST_HEAD(&net
->dev_base_head
);
6465 net
->dev_name_head
= netdev_create_hash();
6466 if (net
->dev_name_head
== NULL
)
6469 net
->dev_index_head
= netdev_create_hash();
6470 if (net
->dev_index_head
== NULL
)
6476 kfree(net
->dev_name_head
);
6482 * netdev_drivername - network driver for the device
6483 * @dev: network device
6485 * Determine network driver for device.
6487 const char *netdev_drivername(const struct net_device
*dev
)
6489 const struct device_driver
*driver
;
6490 const struct device
*parent
;
6491 const char *empty
= "";
6493 parent
= dev
->dev
.parent
;
6497 driver
= parent
->driver
;
6498 if (driver
&& driver
->name
)
6499 return driver
->name
;
6503 static int __netdev_printk(const char *level
, const struct net_device
*dev
,
6504 struct va_format
*vaf
)
6508 if (dev
&& dev
->dev
.parent
) {
6509 r
= dev_printk_emit(level
[1] - '0',
6512 dev_driver_string(dev
->dev
.parent
),
6513 dev_name(dev
->dev
.parent
),
6514 netdev_name(dev
), vaf
);
6516 r
= printk("%s%s: %pV", level
, netdev_name(dev
), vaf
);
6518 r
= printk("%s(NULL net_device): %pV", level
, vaf
);
6524 int netdev_printk(const char *level
, const struct net_device
*dev
,
6525 const char *format
, ...)
6527 struct va_format vaf
;
6531 va_start(args
, format
);
6536 r
= __netdev_printk(level
, dev
, &vaf
);
6542 EXPORT_SYMBOL(netdev_printk
);
6544 #define define_netdev_printk_level(func, level) \
6545 int func(const struct net_device *dev, const char *fmt, ...) \
6548 struct va_format vaf; \
6551 va_start(args, fmt); \
6556 r = __netdev_printk(level, dev, &vaf); \
6562 EXPORT_SYMBOL(func);
6564 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
6565 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
6566 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
6567 define_netdev_printk_level(netdev_err
, KERN_ERR
);
6568 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
6569 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
6570 define_netdev_printk_level(netdev_info
, KERN_INFO
);
6572 static void __net_exit
netdev_exit(struct net
*net
)
6574 kfree(net
->dev_name_head
);
6575 kfree(net
->dev_index_head
);
6578 static struct pernet_operations __net_initdata netdev_net_ops
= {
6579 .init
= netdev_init
,
6580 .exit
= netdev_exit
,
6583 static void __net_exit
default_device_exit(struct net
*net
)
6585 struct net_device
*dev
, *aux
;
6587 * Push all migratable network devices back to the
6588 * initial network namespace
6591 for_each_netdev_safe(net
, dev
, aux
) {
6593 char fb_name
[IFNAMSIZ
];
6595 /* Ignore unmoveable devices (i.e. loopback) */
6596 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6599 /* Leave virtual devices for the generic cleanup */
6600 if (dev
->rtnl_link_ops
)
6603 /* Push remaining network devices to init_net */
6604 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
6605 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
6607 pr_emerg("%s: failed to move %s to init_net: %d\n",
6608 __func__
, dev
->name
, err
);
6615 static void __net_exit
rtnl_lock_unregistering(struct list_head
*net_list
)
6617 /* Return with the rtnl_lock held when there are no network
6618 * devices unregistering in any network namespace in net_list.
6625 prepare_to_wait(&netdev_unregistering_wq
, &wait
,
6626 TASK_UNINTERRUPTIBLE
);
6627 unregistering
= false;
6629 list_for_each_entry(net
, net_list
, exit_list
) {
6630 if (net
->dev_unreg_count
> 0) {
6631 unregistering
= true;
6640 finish_wait(&netdev_unregistering_wq
, &wait
);
6643 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
6645 /* At exit all network devices most be removed from a network
6646 * namespace. Do this in the reverse order of registration.
6647 * Do this across as many network namespaces as possible to
6648 * improve batching efficiency.
6650 struct net_device
*dev
;
6652 LIST_HEAD(dev_kill_list
);
6654 /* To prevent network device cleanup code from dereferencing
6655 * loopback devices or network devices that have been freed
6656 * wait here for all pending unregistrations to complete,
6657 * before unregistring the loopback device and allowing the
6658 * network namespace be freed.
6660 * The netdev todo list containing all network devices
6661 * unregistrations that happen in default_device_exit_batch
6662 * will run in the rtnl_unlock() at the end of
6663 * default_device_exit_batch.
6665 rtnl_lock_unregistering(net_list
);
6666 list_for_each_entry(net
, net_list
, exit_list
) {
6667 for_each_netdev_reverse(net
, dev
) {
6668 if (dev
->rtnl_link_ops
)
6669 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
6671 unregister_netdevice_queue(dev
, &dev_kill_list
);
6674 unregister_netdevice_many(&dev_kill_list
);
6675 list_del(&dev_kill_list
);
6679 static struct pernet_operations __net_initdata default_device_ops
= {
6680 .exit
= default_device_exit
,
6681 .exit_batch
= default_device_exit_batch
,
6685 * Initialize the DEV module. At boot time this walks the device list and
6686 * unhooks any devices that fail to initialise (normally hardware not
6687 * present) and leaves us with a valid list of present and active devices.
6692 * This is called single threaded during boot, so no need
6693 * to take the rtnl semaphore.
6695 static int __init
net_dev_init(void)
6697 int i
, rc
= -ENOMEM
;
6699 BUG_ON(!dev_boot_phase
);
6701 if (dev_proc_init())
6704 if (netdev_kobject_init())
6707 INIT_LIST_HEAD(&ptype_all
);
6708 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
6709 INIT_LIST_HEAD(&ptype_base
[i
]);
6711 INIT_LIST_HEAD(&offload_base
);
6713 if (register_pernet_subsys(&netdev_net_ops
))
6717 * Initialise the packet receive queues.
6720 for_each_possible_cpu(i
) {
6721 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
6723 memset(sd
, 0, sizeof(*sd
));
6724 skb_queue_head_init(&sd
->input_pkt_queue
);
6725 skb_queue_head_init(&sd
->process_queue
);
6726 sd
->completion_queue
= NULL
;
6727 INIT_LIST_HEAD(&sd
->poll_list
);
6728 sd
->output_queue
= NULL
;
6729 sd
->output_queue_tailp
= &sd
->output_queue
;
6731 sd
->csd
.func
= rps_trigger_softirq
;
6737 sd
->backlog
.poll
= process_backlog
;
6738 sd
->backlog
.weight
= weight_p
;
6739 sd
->backlog
.gro_list
= NULL
;
6740 sd
->backlog
.gro_count
= 0;
6742 #ifdef CONFIG_NET_FLOW_LIMIT
6743 sd
->flow_limit
= NULL
;
6749 /* The loopback device is special if any other network devices
6750 * is present in a network namespace the loopback device must
6751 * be present. Since we now dynamically allocate and free the
6752 * loopback device ensure this invariant is maintained by
6753 * keeping the loopback device as the first device on the
6754 * list of network devices. Ensuring the loopback devices
6755 * is the first device that appears and the last network device
6758 if (register_pernet_device(&loopback_net_ops
))
6761 if (register_pernet_device(&default_device_ops
))
6764 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
6765 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
6767 hotcpu_notifier(dev_cpu_callback
, 0);
6774 subsys_initcall(net_dev_init
);