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 <net/mpls.h>
122 #include <linux/ipv6.h>
123 #include <linux/in.h>
124 #include <linux/jhash.h>
125 #include <linux/random.h>
126 #include <trace/events/napi.h>
127 #include <trace/events/net.h>
128 #include <trace/events/skb.h>
129 #include <linux/pci.h>
130 #include <linux/inetdevice.h>
131 #include <linux/cpu_rmap.h>
132 #include <linux/static_key.h>
133 #include <linux/hashtable.h>
134 #include <linux/vmalloc.h>
135 #include <linux/if_macvlan.h>
136 #include <linux/errqueue.h>
137 #include <linux/hrtimer.h>
139 #include "net-sysfs.h"
141 /* Instead of increasing this, you should create a hash table. */
142 #define MAX_GRO_SKBS 8
144 /* This should be increased if a protocol with a bigger head is added. */
145 #define GRO_MAX_HEAD (MAX_HEADER + 128)
147 static DEFINE_SPINLOCK(ptype_lock
);
148 static DEFINE_SPINLOCK(offload_lock
);
149 struct list_head ptype_base
[PTYPE_HASH_SIZE
] __read_mostly
;
150 struct list_head ptype_all __read_mostly
; /* Taps */
151 static struct list_head offload_base __read_mostly
;
153 static int netif_rx_internal(struct sk_buff
*skb
);
154 static int call_netdevice_notifiers_info(unsigned long val
,
155 struct net_device
*dev
,
156 struct netdev_notifier_info
*info
);
159 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
162 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
164 * Writers must hold the rtnl semaphore while they loop through the
165 * dev_base_head list, and hold dev_base_lock for writing when they do the
166 * actual updates. This allows pure readers to access the list even
167 * while a writer is preparing to update it.
169 * To put it another way, dev_base_lock is held for writing only to
170 * protect against pure readers; the rtnl semaphore provides the
171 * protection against other writers.
173 * See, for example usages, register_netdevice() and
174 * unregister_netdevice(), which must be called with the rtnl
177 DEFINE_RWLOCK(dev_base_lock
);
178 EXPORT_SYMBOL(dev_base_lock
);
180 /* protects napi_hash addition/deletion and napi_gen_id */
181 static DEFINE_SPINLOCK(napi_hash_lock
);
183 static unsigned int napi_gen_id
;
184 static DEFINE_HASHTABLE(napi_hash
, 8);
186 static seqcount_t devnet_rename_seq
;
188 static inline void dev_base_seq_inc(struct net
*net
)
190 while (++net
->dev_base_seq
== 0);
193 static inline struct hlist_head
*dev_name_hash(struct net
*net
, const char *name
)
195 unsigned int hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
197 return &net
->dev_name_head
[hash_32(hash
, NETDEV_HASHBITS
)];
200 static inline struct hlist_head
*dev_index_hash(struct net
*net
, int ifindex
)
202 return &net
->dev_index_head
[ifindex
& (NETDEV_HASHENTRIES
- 1)];
205 static inline void rps_lock(struct softnet_data
*sd
)
208 spin_lock(&sd
->input_pkt_queue
.lock
);
212 static inline void rps_unlock(struct softnet_data
*sd
)
215 spin_unlock(&sd
->input_pkt_queue
.lock
);
219 /* Device list insertion */
220 static void list_netdevice(struct net_device
*dev
)
222 struct net
*net
= dev_net(dev
);
226 write_lock_bh(&dev_base_lock
);
227 list_add_tail_rcu(&dev
->dev_list
, &net
->dev_base_head
);
228 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
229 hlist_add_head_rcu(&dev
->index_hlist
,
230 dev_index_hash(net
, dev
->ifindex
));
231 write_unlock_bh(&dev_base_lock
);
233 dev_base_seq_inc(net
);
236 /* Device list removal
237 * caller must respect a RCU grace period before freeing/reusing dev
239 static void unlist_netdevice(struct net_device
*dev
)
243 /* Unlink dev from the device chain */
244 write_lock_bh(&dev_base_lock
);
245 list_del_rcu(&dev
->dev_list
);
246 hlist_del_rcu(&dev
->name_hlist
);
247 hlist_del_rcu(&dev
->index_hlist
);
248 write_unlock_bh(&dev_base_lock
);
250 dev_base_seq_inc(dev_net(dev
));
257 static RAW_NOTIFIER_HEAD(netdev_chain
);
260 * Device drivers call our routines to queue packets here. We empty the
261 * queue in the local softnet handler.
264 DEFINE_PER_CPU_ALIGNED(struct softnet_data
, softnet_data
);
265 EXPORT_PER_CPU_SYMBOL(softnet_data
);
267 #ifdef CONFIG_LOCKDEP
269 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
270 * according to dev->type
272 static const unsigned short netdev_lock_type
[] =
273 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
274 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
275 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
276 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
277 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
278 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
279 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
280 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
281 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
282 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
283 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
284 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
285 ARPHRD_FCFABRIC
, ARPHRD_IEEE80211
, ARPHRD_IEEE80211_PRISM
,
286 ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_PHONET
, ARPHRD_PHONET_PIPE
,
287 ARPHRD_IEEE802154
, ARPHRD_VOID
, ARPHRD_NONE
};
289 static const char *const netdev_lock_name
[] =
290 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
291 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
292 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
293 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
294 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
295 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
296 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
297 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
298 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
299 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
300 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
301 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
302 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
303 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
304 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
306 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
307 static struct lock_class_key netdev_addr_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
309 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
313 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
314 if (netdev_lock_type
[i
] == dev_type
)
316 /* the last key is used by default */
317 return ARRAY_SIZE(netdev_lock_type
) - 1;
320 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
321 unsigned short dev_type
)
325 i
= netdev_lock_pos(dev_type
);
326 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
327 netdev_lock_name
[i
]);
330 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
334 i
= netdev_lock_pos(dev
->type
);
335 lockdep_set_class_and_name(&dev
->addr_list_lock
,
336 &netdev_addr_lock_key
[i
],
337 netdev_lock_name
[i
]);
340 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
341 unsigned short dev_type
)
344 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
349 /*******************************************************************************
351 Protocol management and registration routines
353 *******************************************************************************/
356 * Add a protocol ID to the list. Now that the input handler is
357 * smarter we can dispense with all the messy stuff that used to be
360 * BEWARE!!! Protocol handlers, mangling input packets,
361 * MUST BE last in hash buckets and checking protocol handlers
362 * MUST start from promiscuous ptype_all chain in net_bh.
363 * It is true now, do not change it.
364 * Explanation follows: if protocol handler, mangling packet, will
365 * be the first on list, it is not able to sense, that packet
366 * is cloned and should be copied-on-write, so that it will
367 * change it and subsequent readers will get broken packet.
371 static inline struct list_head
*ptype_head(const struct packet_type
*pt
)
373 if (pt
->type
== htons(ETH_P_ALL
))
374 return pt
->dev
? &pt
->dev
->ptype_all
: &ptype_all
;
376 return pt
->dev
? &pt
->dev
->ptype_specific
:
377 &ptype_base
[ntohs(pt
->type
) & PTYPE_HASH_MASK
];
381 * dev_add_pack - add packet handler
382 * @pt: packet type declaration
384 * Add a protocol handler to the networking stack. The passed &packet_type
385 * is linked into kernel lists and may not be freed until it has been
386 * removed from the kernel lists.
388 * This call does not sleep therefore it can not
389 * guarantee all CPU's that are in middle of receiving packets
390 * will see the new packet type (until the next received packet).
393 void dev_add_pack(struct packet_type
*pt
)
395 struct list_head
*head
= ptype_head(pt
);
397 spin_lock(&ptype_lock
);
398 list_add_rcu(&pt
->list
, head
);
399 spin_unlock(&ptype_lock
);
401 EXPORT_SYMBOL(dev_add_pack
);
404 * __dev_remove_pack - remove packet handler
405 * @pt: packet type declaration
407 * Remove a protocol handler that was previously added to the kernel
408 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
409 * from the kernel lists and can be freed or reused once this function
412 * The packet type might still be in use by receivers
413 * and must not be freed until after all the CPU's have gone
414 * through a quiescent state.
416 void __dev_remove_pack(struct packet_type
*pt
)
418 struct list_head
*head
= ptype_head(pt
);
419 struct packet_type
*pt1
;
421 spin_lock(&ptype_lock
);
423 list_for_each_entry(pt1
, head
, list
) {
425 list_del_rcu(&pt
->list
);
430 pr_warn("dev_remove_pack: %p not found\n", pt
);
432 spin_unlock(&ptype_lock
);
434 EXPORT_SYMBOL(__dev_remove_pack
);
437 * dev_remove_pack - remove packet handler
438 * @pt: packet type declaration
440 * Remove a protocol handler that was previously added to the kernel
441 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
442 * from the kernel lists and can be freed or reused once this function
445 * This call sleeps to guarantee that no CPU is looking at the packet
448 void dev_remove_pack(struct packet_type
*pt
)
450 __dev_remove_pack(pt
);
454 EXPORT_SYMBOL(dev_remove_pack
);
458 * dev_add_offload - register offload handlers
459 * @po: protocol offload declaration
461 * Add protocol offload handlers to the networking stack. The passed
462 * &proto_offload is linked into kernel lists and may not be freed until
463 * it has been removed from the kernel lists.
465 * This call does not sleep therefore it can not
466 * guarantee all CPU's that are in middle of receiving packets
467 * will see the new offload handlers (until the next received packet).
469 void dev_add_offload(struct packet_offload
*po
)
471 struct list_head
*head
= &offload_base
;
473 spin_lock(&offload_lock
);
474 list_add_rcu(&po
->list
, head
);
475 spin_unlock(&offload_lock
);
477 EXPORT_SYMBOL(dev_add_offload
);
480 * __dev_remove_offload - remove offload handler
481 * @po: packet offload declaration
483 * Remove a protocol offload handler that was previously added to the
484 * kernel offload handlers by dev_add_offload(). The passed &offload_type
485 * is removed from the kernel lists and can be freed or reused once this
488 * The packet type might still be in use by receivers
489 * and must not be freed until after all the CPU's have gone
490 * through a quiescent state.
492 static void __dev_remove_offload(struct packet_offload
*po
)
494 struct list_head
*head
= &offload_base
;
495 struct packet_offload
*po1
;
497 spin_lock(&offload_lock
);
499 list_for_each_entry(po1
, head
, list
) {
501 list_del_rcu(&po
->list
);
506 pr_warn("dev_remove_offload: %p not found\n", po
);
508 spin_unlock(&offload_lock
);
512 * dev_remove_offload - remove packet offload handler
513 * @po: packet offload declaration
515 * Remove a packet offload handler that was previously added to the kernel
516 * offload handlers by dev_add_offload(). The passed &offload_type is
517 * removed from the kernel lists and can be freed or reused once this
520 * This call sleeps to guarantee that no CPU is looking at the packet
523 void dev_remove_offload(struct packet_offload
*po
)
525 __dev_remove_offload(po
);
529 EXPORT_SYMBOL(dev_remove_offload
);
531 /******************************************************************************
533 Device Boot-time Settings Routines
535 *******************************************************************************/
537 /* Boot time configuration table */
538 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
541 * netdev_boot_setup_add - add new setup entry
542 * @name: name of the device
543 * @map: configured settings for the device
545 * Adds new setup entry to the dev_boot_setup list. The function
546 * returns 0 on error and 1 on success. This is a generic routine to
549 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
551 struct netdev_boot_setup
*s
;
555 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
556 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
557 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
558 strlcpy(s
[i
].name
, name
, IFNAMSIZ
);
559 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
564 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
568 * netdev_boot_setup_check - check boot time settings
569 * @dev: the netdevice
571 * Check boot time settings for the device.
572 * The found settings are set for the device to be used
573 * later in the device probing.
574 * Returns 0 if no settings found, 1 if they are.
576 int netdev_boot_setup_check(struct net_device
*dev
)
578 struct netdev_boot_setup
*s
= dev_boot_setup
;
581 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
582 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
583 !strcmp(dev
->name
, s
[i
].name
)) {
584 dev
->irq
= s
[i
].map
.irq
;
585 dev
->base_addr
= s
[i
].map
.base_addr
;
586 dev
->mem_start
= s
[i
].map
.mem_start
;
587 dev
->mem_end
= s
[i
].map
.mem_end
;
593 EXPORT_SYMBOL(netdev_boot_setup_check
);
597 * netdev_boot_base - get address from boot time settings
598 * @prefix: prefix for network device
599 * @unit: id for network device
601 * Check boot time settings for the base address of device.
602 * The found settings are set for the device to be used
603 * later in the device probing.
604 * Returns 0 if no settings found.
606 unsigned long netdev_boot_base(const char *prefix
, int unit
)
608 const struct netdev_boot_setup
*s
= dev_boot_setup
;
612 sprintf(name
, "%s%d", prefix
, unit
);
615 * If device already registered then return base of 1
616 * to indicate not to probe for this interface
618 if (__dev_get_by_name(&init_net
, name
))
621 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
622 if (!strcmp(name
, s
[i
].name
))
623 return s
[i
].map
.base_addr
;
628 * Saves at boot time configured settings for any netdevice.
630 int __init
netdev_boot_setup(char *str
)
635 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
640 memset(&map
, 0, sizeof(map
));
644 map
.base_addr
= ints
[2];
646 map
.mem_start
= ints
[3];
648 map
.mem_end
= ints
[4];
650 /* Add new entry to the list */
651 return netdev_boot_setup_add(str
, &map
);
654 __setup("netdev=", netdev_boot_setup
);
656 /*******************************************************************************
658 Device Interface Subroutines
660 *******************************************************************************/
663 * __dev_get_by_name - find a device by its name
664 * @net: the applicable net namespace
665 * @name: name to find
667 * Find an interface by name. Must be called under RTNL semaphore
668 * or @dev_base_lock. If the name is found a pointer to the device
669 * is returned. If the name is not found then %NULL is returned. The
670 * reference counters are not incremented so the caller must be
671 * careful with locks.
674 struct net_device
*__dev_get_by_name(struct net
*net
, const char *name
)
676 struct net_device
*dev
;
677 struct hlist_head
*head
= dev_name_hash(net
, name
);
679 hlist_for_each_entry(dev
, head
, name_hlist
)
680 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
685 EXPORT_SYMBOL(__dev_get_by_name
);
688 * dev_get_by_name_rcu - find a device by its name
689 * @net: the applicable net namespace
690 * @name: name to find
692 * Find an interface by name.
693 * If the name is found a pointer to the device is returned.
694 * If the name is not found then %NULL is returned.
695 * The reference counters are not incremented so the caller must be
696 * careful with locks. The caller must hold RCU lock.
699 struct net_device
*dev_get_by_name_rcu(struct net
*net
, const char *name
)
701 struct net_device
*dev
;
702 struct hlist_head
*head
= dev_name_hash(net
, name
);
704 hlist_for_each_entry_rcu(dev
, head
, name_hlist
)
705 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
710 EXPORT_SYMBOL(dev_get_by_name_rcu
);
713 * dev_get_by_name - find a device by its name
714 * @net: the applicable net namespace
715 * @name: name to find
717 * Find an interface by name. This can be called from any
718 * context and does its own locking. The returned handle has
719 * the usage count incremented and the caller must use dev_put() to
720 * release it when it is no longer needed. %NULL is returned if no
721 * matching device is found.
724 struct net_device
*dev_get_by_name(struct net
*net
, const char *name
)
726 struct net_device
*dev
;
729 dev
= dev_get_by_name_rcu(net
, name
);
735 EXPORT_SYMBOL(dev_get_by_name
);
738 * __dev_get_by_index - find a device by its ifindex
739 * @net: the applicable net namespace
740 * @ifindex: index of device
742 * Search for an interface by index. Returns %NULL if the device
743 * is not found or a pointer to the device. The device has not
744 * had its reference counter increased so the caller must be careful
745 * about locking. The caller must hold either the RTNL semaphore
749 struct net_device
*__dev_get_by_index(struct net
*net
, int ifindex
)
751 struct net_device
*dev
;
752 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
754 hlist_for_each_entry(dev
, head
, index_hlist
)
755 if (dev
->ifindex
== ifindex
)
760 EXPORT_SYMBOL(__dev_get_by_index
);
763 * dev_get_by_index_rcu - find a device by its ifindex
764 * @net: the applicable net namespace
765 * @ifindex: index of device
767 * Search for an interface by index. Returns %NULL if the device
768 * is not found or a pointer to the device. The device has not
769 * had its reference counter increased so the caller must be careful
770 * about locking. The caller must hold RCU lock.
773 struct net_device
*dev_get_by_index_rcu(struct net
*net
, int ifindex
)
775 struct net_device
*dev
;
776 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
778 hlist_for_each_entry_rcu(dev
, head
, index_hlist
)
779 if (dev
->ifindex
== ifindex
)
784 EXPORT_SYMBOL(dev_get_by_index_rcu
);
788 * dev_get_by_index - find a device by its ifindex
789 * @net: the applicable net namespace
790 * @ifindex: index of device
792 * Search for an interface by index. Returns NULL if the device
793 * is not found or a pointer to the device. The device returned has
794 * had a reference added and the pointer is safe until the user calls
795 * dev_put to indicate they have finished with it.
798 struct net_device
*dev_get_by_index(struct net
*net
, int ifindex
)
800 struct net_device
*dev
;
803 dev
= dev_get_by_index_rcu(net
, ifindex
);
809 EXPORT_SYMBOL(dev_get_by_index
);
812 * netdev_get_name - get a netdevice name, knowing its ifindex.
813 * @net: network namespace
814 * @name: a pointer to the buffer where the name will be stored.
815 * @ifindex: the ifindex of the interface to get the name from.
817 * The use of raw_seqcount_begin() and cond_resched() before
818 * retrying is required as we want to give the writers a chance
819 * to complete when CONFIG_PREEMPT is not set.
821 int netdev_get_name(struct net
*net
, char *name
, int ifindex
)
823 struct net_device
*dev
;
827 seq
= raw_seqcount_begin(&devnet_rename_seq
);
829 dev
= dev_get_by_index_rcu(net
, ifindex
);
835 strcpy(name
, dev
->name
);
837 if (read_seqcount_retry(&devnet_rename_seq
, seq
)) {
846 * dev_getbyhwaddr_rcu - find a device by its hardware address
847 * @net: the applicable net namespace
848 * @type: media type of device
849 * @ha: hardware address
851 * Search for an interface by MAC address. Returns NULL if the device
852 * is not found or a pointer to the device.
853 * The caller must hold RCU or RTNL.
854 * The returned device has not had its ref count increased
855 * and the caller must therefore be careful about locking
859 struct net_device
*dev_getbyhwaddr_rcu(struct net
*net
, unsigned short type
,
862 struct net_device
*dev
;
864 for_each_netdev_rcu(net
, dev
)
865 if (dev
->type
== type
&&
866 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
871 EXPORT_SYMBOL(dev_getbyhwaddr_rcu
);
873 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
875 struct net_device
*dev
;
878 for_each_netdev(net
, dev
)
879 if (dev
->type
== type
)
884 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
886 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
888 struct net_device
*dev
, *ret
= NULL
;
891 for_each_netdev_rcu(net
, dev
)
892 if (dev
->type
== type
) {
900 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
903 * __dev_get_by_flags - find any device with given flags
904 * @net: the applicable net namespace
905 * @if_flags: IFF_* values
906 * @mask: bitmask of bits in if_flags to check
908 * Search for any interface with the given flags. Returns NULL if a device
909 * is not found or a pointer to the device. Must be called inside
910 * rtnl_lock(), and result refcount is unchanged.
913 struct net_device
*__dev_get_by_flags(struct net
*net
, unsigned short if_flags
,
916 struct net_device
*dev
, *ret
;
921 for_each_netdev(net
, dev
) {
922 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
929 EXPORT_SYMBOL(__dev_get_by_flags
);
932 * dev_valid_name - check if name is okay for network device
935 * Network device names need to be valid file names to
936 * to allow sysfs to work. We also disallow any kind of
939 bool dev_valid_name(const char *name
)
943 if (strlen(name
) >= IFNAMSIZ
)
945 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
949 if (*name
== '/' || *name
== ':' || isspace(*name
))
955 EXPORT_SYMBOL(dev_valid_name
);
958 * __dev_alloc_name - allocate a name for a device
959 * @net: network namespace to allocate the device name in
960 * @name: name format string
961 * @buf: scratch buffer and result name string
963 * Passed a format string - eg "lt%d" it will try and find a suitable
964 * id. It scans list of devices to build up a free map, then chooses
965 * the first empty slot. The caller must hold the dev_base or rtnl lock
966 * while allocating the name and adding the device in order to avoid
968 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
969 * Returns the number of the unit assigned or a negative errno code.
972 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
976 const int max_netdevices
= 8*PAGE_SIZE
;
977 unsigned long *inuse
;
978 struct net_device
*d
;
980 p
= strnchr(name
, IFNAMSIZ
-1, '%');
983 * Verify the string as this thing may have come from
984 * the user. There must be either one "%d" and no other "%"
987 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
990 /* Use one page as a bit array of possible slots */
991 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
995 for_each_netdev(net
, d
) {
996 if (!sscanf(d
->name
, name
, &i
))
998 if (i
< 0 || i
>= max_netdevices
)
1001 /* avoid cases where sscanf is not exact inverse of printf */
1002 snprintf(buf
, IFNAMSIZ
, name
, i
);
1003 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
1007 i
= find_first_zero_bit(inuse
, max_netdevices
);
1008 free_page((unsigned long) inuse
);
1012 snprintf(buf
, IFNAMSIZ
, name
, i
);
1013 if (!__dev_get_by_name(net
, buf
))
1016 /* It is possible to run out of possible slots
1017 * when the name is long and there isn't enough space left
1018 * for the digits, or if all bits are used.
1024 * dev_alloc_name - allocate a name for a device
1026 * @name: name format string
1028 * Passed a format string - eg "lt%d" it will try and find a suitable
1029 * id. It scans list of devices to build up a free map, then chooses
1030 * the first empty slot. The caller must hold the dev_base or rtnl lock
1031 * while allocating the name and adding the device in order to avoid
1033 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1034 * Returns the number of the unit assigned or a negative errno code.
1037 int dev_alloc_name(struct net_device
*dev
, const char *name
)
1043 BUG_ON(!dev_net(dev
));
1045 ret
= __dev_alloc_name(net
, name
, buf
);
1047 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
1050 EXPORT_SYMBOL(dev_alloc_name
);
1052 static int dev_alloc_name_ns(struct net
*net
,
1053 struct net_device
*dev
,
1059 ret
= __dev_alloc_name(net
, name
, buf
);
1061 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
1065 static int dev_get_valid_name(struct net
*net
,
1066 struct net_device
*dev
,
1071 if (!dev_valid_name(name
))
1074 if (strchr(name
, '%'))
1075 return dev_alloc_name_ns(net
, dev
, name
);
1076 else if (__dev_get_by_name(net
, name
))
1078 else if (dev
->name
!= name
)
1079 strlcpy(dev
->name
, name
, IFNAMSIZ
);
1085 * dev_change_name - change name of a device
1087 * @newname: name (or format string) must be at least IFNAMSIZ
1089 * Change name of a device, can pass format strings "eth%d".
1092 int dev_change_name(struct net_device
*dev
, const char *newname
)
1094 unsigned char old_assign_type
;
1095 char oldname
[IFNAMSIZ
];
1101 BUG_ON(!dev_net(dev
));
1104 if (dev
->flags
& IFF_UP
)
1107 write_seqcount_begin(&devnet_rename_seq
);
1109 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0) {
1110 write_seqcount_end(&devnet_rename_seq
);
1114 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
1116 err
= dev_get_valid_name(net
, dev
, newname
);
1118 write_seqcount_end(&devnet_rename_seq
);
1122 if (oldname
[0] && !strchr(oldname
, '%'))
1123 netdev_info(dev
, "renamed from %s\n", oldname
);
1125 old_assign_type
= dev
->name_assign_type
;
1126 dev
->name_assign_type
= NET_NAME_RENAMED
;
1129 ret
= device_rename(&dev
->dev
, dev
->name
);
1131 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1132 dev
->name_assign_type
= old_assign_type
;
1133 write_seqcount_end(&devnet_rename_seq
);
1137 write_seqcount_end(&devnet_rename_seq
);
1139 netdev_adjacent_rename_links(dev
, oldname
);
1141 write_lock_bh(&dev_base_lock
);
1142 hlist_del_rcu(&dev
->name_hlist
);
1143 write_unlock_bh(&dev_base_lock
);
1147 write_lock_bh(&dev_base_lock
);
1148 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
1149 write_unlock_bh(&dev_base_lock
);
1151 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
1152 ret
= notifier_to_errno(ret
);
1155 /* err >= 0 after dev_alloc_name() or stores the first errno */
1158 write_seqcount_begin(&devnet_rename_seq
);
1159 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1160 memcpy(oldname
, newname
, IFNAMSIZ
);
1161 dev
->name_assign_type
= old_assign_type
;
1162 old_assign_type
= NET_NAME_RENAMED
;
1165 pr_err("%s: name change rollback failed: %d\n",
1174 * dev_set_alias - change ifalias of a device
1176 * @alias: name up to IFALIASZ
1177 * @len: limit of bytes to copy from info
1179 * Set ifalias for a device,
1181 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
1187 if (len
>= IFALIASZ
)
1191 kfree(dev
->ifalias
);
1192 dev
->ifalias
= NULL
;
1196 new_ifalias
= krealloc(dev
->ifalias
, len
+ 1, GFP_KERNEL
);
1199 dev
->ifalias
= new_ifalias
;
1201 strlcpy(dev
->ifalias
, alias
, len
+1);
1207 * netdev_features_change - device changes features
1208 * @dev: device to cause notification
1210 * Called to indicate a device has changed features.
1212 void netdev_features_change(struct net_device
*dev
)
1214 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1216 EXPORT_SYMBOL(netdev_features_change
);
1219 * netdev_state_change - device changes state
1220 * @dev: device to cause notification
1222 * Called to indicate a device has changed state. This function calls
1223 * the notifier chains for netdev_chain and sends a NEWLINK message
1224 * to the routing socket.
1226 void netdev_state_change(struct net_device
*dev
)
1228 if (dev
->flags
& IFF_UP
) {
1229 struct netdev_notifier_change_info change_info
;
1231 change_info
.flags_changed
= 0;
1232 call_netdevice_notifiers_info(NETDEV_CHANGE
, dev
,
1234 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0, GFP_KERNEL
);
1237 EXPORT_SYMBOL(netdev_state_change
);
1240 * netdev_notify_peers - notify network peers about existence of @dev
1241 * @dev: network device
1243 * Generate traffic such that interested network peers are aware of
1244 * @dev, such as by generating a gratuitous ARP. This may be used when
1245 * a device wants to inform the rest of the network about some sort of
1246 * reconfiguration such as a failover event or virtual machine
1249 void netdev_notify_peers(struct net_device
*dev
)
1252 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
, dev
);
1255 EXPORT_SYMBOL(netdev_notify_peers
);
1257 static int __dev_open(struct net_device
*dev
)
1259 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1264 if (!netif_device_present(dev
))
1267 /* Block netpoll from trying to do any rx path servicing.
1268 * If we don't do this there is a chance ndo_poll_controller
1269 * or ndo_poll may be running while we open the device
1271 netpoll_poll_disable(dev
);
1273 ret
= call_netdevice_notifiers(NETDEV_PRE_UP
, dev
);
1274 ret
= notifier_to_errno(ret
);
1278 set_bit(__LINK_STATE_START
, &dev
->state
);
1280 if (ops
->ndo_validate_addr
)
1281 ret
= ops
->ndo_validate_addr(dev
);
1283 if (!ret
&& ops
->ndo_open
)
1284 ret
= ops
->ndo_open(dev
);
1286 netpoll_poll_enable(dev
);
1289 clear_bit(__LINK_STATE_START
, &dev
->state
);
1291 dev
->flags
|= IFF_UP
;
1292 dev_set_rx_mode(dev
);
1294 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
1301 * dev_open - prepare an interface for use.
1302 * @dev: device to open
1304 * Takes a device from down to up state. The device's private open
1305 * function is invoked and then the multicast lists are loaded. Finally
1306 * the device is moved into the up state and a %NETDEV_UP message is
1307 * sent to the netdev notifier chain.
1309 * Calling this function on an active interface is a nop. On a failure
1310 * a negative errno code is returned.
1312 int dev_open(struct net_device
*dev
)
1316 if (dev
->flags
& IFF_UP
)
1319 ret
= __dev_open(dev
);
1323 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
, GFP_KERNEL
);
1324 call_netdevice_notifiers(NETDEV_UP
, dev
);
1328 EXPORT_SYMBOL(dev_open
);
1330 static int __dev_close_many(struct list_head
*head
)
1332 struct net_device
*dev
;
1337 list_for_each_entry(dev
, head
, close_list
) {
1338 /* Temporarily disable netpoll until the interface is down */
1339 netpoll_poll_disable(dev
);
1341 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1343 clear_bit(__LINK_STATE_START
, &dev
->state
);
1345 /* Synchronize to scheduled poll. We cannot touch poll list, it
1346 * can be even on different cpu. So just clear netif_running().
1348 * dev->stop() will invoke napi_disable() on all of it's
1349 * napi_struct instances on this device.
1351 smp_mb__after_atomic(); /* Commit netif_running(). */
1354 dev_deactivate_many(head
);
1356 list_for_each_entry(dev
, head
, close_list
) {
1357 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1360 * Call the device specific close. This cannot fail.
1361 * Only if device is UP
1363 * We allow it to be called even after a DETACH hot-plug
1369 dev
->flags
&= ~IFF_UP
;
1370 netpoll_poll_enable(dev
);
1376 static int __dev_close(struct net_device
*dev
)
1381 list_add(&dev
->close_list
, &single
);
1382 retval
= __dev_close_many(&single
);
1388 int dev_close_many(struct list_head
*head
, bool unlink
)
1390 struct net_device
*dev
, *tmp
;
1392 /* Remove the devices that don't need to be closed */
1393 list_for_each_entry_safe(dev
, tmp
, head
, close_list
)
1394 if (!(dev
->flags
& IFF_UP
))
1395 list_del_init(&dev
->close_list
);
1397 __dev_close_many(head
);
1399 list_for_each_entry_safe(dev
, tmp
, head
, close_list
) {
1400 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
, GFP_KERNEL
);
1401 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1403 list_del_init(&dev
->close_list
);
1408 EXPORT_SYMBOL(dev_close_many
);
1411 * dev_close - shutdown an interface.
1412 * @dev: device to shutdown
1414 * This function moves an active device into down state. A
1415 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1416 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1419 int dev_close(struct net_device
*dev
)
1421 if (dev
->flags
& IFF_UP
) {
1424 list_add(&dev
->close_list
, &single
);
1425 dev_close_many(&single
, true);
1430 EXPORT_SYMBOL(dev_close
);
1434 * dev_disable_lro - disable Large Receive Offload on a device
1437 * Disable Large Receive Offload (LRO) on a net device. Must be
1438 * called under RTNL. This is needed if received packets may be
1439 * forwarded to another interface.
1441 void dev_disable_lro(struct net_device
*dev
)
1443 struct net_device
*lower_dev
;
1444 struct list_head
*iter
;
1446 dev
->wanted_features
&= ~NETIF_F_LRO
;
1447 netdev_update_features(dev
);
1449 if (unlikely(dev
->features
& NETIF_F_LRO
))
1450 netdev_WARN(dev
, "failed to disable LRO!\n");
1452 netdev_for_each_lower_dev(dev
, lower_dev
, iter
)
1453 dev_disable_lro(lower_dev
);
1455 EXPORT_SYMBOL(dev_disable_lro
);
1457 static int call_netdevice_notifier(struct notifier_block
*nb
, unsigned long val
,
1458 struct net_device
*dev
)
1460 struct netdev_notifier_info info
;
1462 netdev_notifier_info_init(&info
, dev
);
1463 return nb
->notifier_call(nb
, val
, &info
);
1466 static int dev_boot_phase
= 1;
1469 * register_netdevice_notifier - register a network notifier block
1472 * Register a notifier to be called when network device events occur.
1473 * The notifier passed is linked into the kernel structures and must
1474 * not be reused until it has been unregistered. A negative errno code
1475 * is returned on a failure.
1477 * When registered all registration and up events are replayed
1478 * to the new notifier to allow device to have a race free
1479 * view of the network device list.
1482 int register_netdevice_notifier(struct notifier_block
*nb
)
1484 struct net_device
*dev
;
1485 struct net_device
*last
;
1490 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1496 for_each_netdev(net
, dev
) {
1497 err
= call_netdevice_notifier(nb
, NETDEV_REGISTER
, dev
);
1498 err
= notifier_to_errno(err
);
1502 if (!(dev
->flags
& IFF_UP
))
1505 call_netdevice_notifier(nb
, NETDEV_UP
, dev
);
1516 for_each_netdev(net
, dev
) {
1520 if (dev
->flags
& IFF_UP
) {
1521 call_netdevice_notifier(nb
, NETDEV_GOING_DOWN
,
1523 call_netdevice_notifier(nb
, NETDEV_DOWN
, dev
);
1525 call_netdevice_notifier(nb
, NETDEV_UNREGISTER
, dev
);
1530 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1533 EXPORT_SYMBOL(register_netdevice_notifier
);
1536 * unregister_netdevice_notifier - unregister a network notifier block
1539 * Unregister a notifier previously registered by
1540 * register_netdevice_notifier(). The notifier is unlinked into the
1541 * kernel structures and may then be reused. A negative errno code
1542 * is returned on a failure.
1544 * After unregistering unregister and down device events are synthesized
1545 * for all devices on the device list to the removed notifier to remove
1546 * the need for special case cleanup code.
1549 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1551 struct net_device
*dev
;
1556 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1561 for_each_netdev(net
, dev
) {
1562 if (dev
->flags
& IFF_UP
) {
1563 call_netdevice_notifier(nb
, NETDEV_GOING_DOWN
,
1565 call_netdevice_notifier(nb
, NETDEV_DOWN
, dev
);
1567 call_netdevice_notifier(nb
, NETDEV_UNREGISTER
, dev
);
1574 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1577 * call_netdevice_notifiers_info - call all network notifier blocks
1578 * @val: value passed unmodified to notifier function
1579 * @dev: net_device pointer passed unmodified to notifier function
1580 * @info: notifier information data
1582 * Call all network notifier blocks. Parameters and return value
1583 * are as for raw_notifier_call_chain().
1586 static int call_netdevice_notifiers_info(unsigned long val
,
1587 struct net_device
*dev
,
1588 struct netdev_notifier_info
*info
)
1591 netdev_notifier_info_init(info
, dev
);
1592 return raw_notifier_call_chain(&netdev_chain
, val
, info
);
1596 * call_netdevice_notifiers - call all network notifier blocks
1597 * @val: value passed unmodified to notifier function
1598 * @dev: net_device pointer passed unmodified to notifier function
1600 * Call all network notifier blocks. Parameters and return value
1601 * are as for raw_notifier_call_chain().
1604 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1606 struct netdev_notifier_info info
;
1608 return call_netdevice_notifiers_info(val
, dev
, &info
);
1610 EXPORT_SYMBOL(call_netdevice_notifiers
);
1612 static struct static_key netstamp_needed __read_mostly
;
1613 #ifdef HAVE_JUMP_LABEL
1614 /* We are not allowed to call static_key_slow_dec() from irq context
1615 * If net_disable_timestamp() is called from irq context, defer the
1616 * static_key_slow_dec() calls.
1618 static atomic_t netstamp_needed_deferred
;
1621 void net_enable_timestamp(void)
1623 #ifdef HAVE_JUMP_LABEL
1624 int deferred
= atomic_xchg(&netstamp_needed_deferred
, 0);
1628 static_key_slow_dec(&netstamp_needed
);
1632 static_key_slow_inc(&netstamp_needed
);
1634 EXPORT_SYMBOL(net_enable_timestamp
);
1636 void net_disable_timestamp(void)
1638 #ifdef HAVE_JUMP_LABEL
1639 if (in_interrupt()) {
1640 atomic_inc(&netstamp_needed_deferred
);
1644 static_key_slow_dec(&netstamp_needed
);
1646 EXPORT_SYMBOL(net_disable_timestamp
);
1648 static inline void net_timestamp_set(struct sk_buff
*skb
)
1650 skb
->tstamp
.tv64
= 0;
1651 if (static_key_false(&netstamp_needed
))
1652 __net_timestamp(skb
);
1655 #define net_timestamp_check(COND, SKB) \
1656 if (static_key_false(&netstamp_needed)) { \
1657 if ((COND) && !(SKB)->tstamp.tv64) \
1658 __net_timestamp(SKB); \
1661 bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb)
1665 if (!(dev
->flags
& IFF_UP
))
1668 len
= dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
;
1669 if (skb
->len
<= len
)
1672 /* if TSO is enabled, we don't care about the length as the packet
1673 * could be forwarded without being segmented before
1675 if (skb_is_gso(skb
))
1680 EXPORT_SYMBOL_GPL(is_skb_forwardable
);
1682 int __dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1684 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
1685 if (skb_copy_ubufs(skb
, GFP_ATOMIC
)) {
1686 atomic_long_inc(&dev
->rx_dropped
);
1692 if (unlikely(!is_skb_forwardable(dev
, skb
))) {
1693 atomic_long_inc(&dev
->rx_dropped
);
1698 skb_scrub_packet(skb
, true);
1700 skb
->protocol
= eth_type_trans(skb
, dev
);
1701 skb_postpull_rcsum(skb
, eth_hdr(skb
), ETH_HLEN
);
1705 EXPORT_SYMBOL_GPL(__dev_forward_skb
);
1708 * dev_forward_skb - loopback an skb to another netif
1710 * @dev: destination network device
1711 * @skb: buffer to forward
1714 * NET_RX_SUCCESS (no congestion)
1715 * NET_RX_DROP (packet was dropped, but freed)
1717 * dev_forward_skb can be used for injecting an skb from the
1718 * start_xmit function of one device into the receive queue
1719 * of another device.
1721 * The receiving device may be in another namespace, so
1722 * we have to clear all information in the skb that could
1723 * impact namespace isolation.
1725 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1727 return __dev_forward_skb(dev
, skb
) ?: netif_rx_internal(skb
);
1729 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1731 static inline int deliver_skb(struct sk_buff
*skb
,
1732 struct packet_type
*pt_prev
,
1733 struct net_device
*orig_dev
)
1735 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
1737 atomic_inc(&skb
->users
);
1738 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1741 static inline void deliver_ptype_list_skb(struct sk_buff
*skb
,
1742 struct packet_type
**pt
,
1743 struct net_device
*dev
, __be16 type
,
1744 struct list_head
*ptype_list
)
1746 struct packet_type
*ptype
, *pt_prev
= *pt
;
1748 list_for_each_entry_rcu(ptype
, ptype_list
, list
) {
1749 if (ptype
->type
!= type
)
1752 deliver_skb(skb
, pt_prev
, dev
);
1758 static inline bool skb_loop_sk(struct packet_type
*ptype
, struct sk_buff
*skb
)
1760 if (!ptype
->af_packet_priv
|| !skb
->sk
)
1763 if (ptype
->id_match
)
1764 return ptype
->id_match(ptype
, skb
->sk
);
1765 else if ((struct sock
*)ptype
->af_packet_priv
== skb
->sk
)
1772 * Support routine. Sends outgoing frames to any network
1773 * taps currently in use.
1776 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1778 struct packet_type
*ptype
;
1779 struct sk_buff
*skb2
= NULL
;
1780 struct packet_type
*pt_prev
= NULL
;
1781 struct list_head
*ptype_list
= &ptype_all
;
1785 list_for_each_entry_rcu(ptype
, ptype_list
, list
) {
1786 /* Never send packets back to the socket
1787 * they originated from - MvS (miquels@drinkel.ow.org)
1789 if (skb_loop_sk(ptype
, skb
))
1793 deliver_skb(skb2
, pt_prev
, skb
->dev
);
1798 /* need to clone skb, done only once */
1799 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1803 net_timestamp_set(skb2
);
1805 /* skb->nh should be correctly
1806 * set by sender, so that the second statement is
1807 * just protection against buggy protocols.
1809 skb_reset_mac_header(skb2
);
1811 if (skb_network_header(skb2
) < skb2
->data
||
1812 skb_network_header(skb2
) > skb_tail_pointer(skb2
)) {
1813 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1814 ntohs(skb2
->protocol
),
1816 skb_reset_network_header(skb2
);
1819 skb2
->transport_header
= skb2
->network_header
;
1820 skb2
->pkt_type
= PACKET_OUTGOING
;
1824 if (ptype_list
== &ptype_all
) {
1825 ptype_list
= &dev
->ptype_all
;
1830 pt_prev
->func(skb2
, skb
->dev
, pt_prev
, skb
->dev
);
1835 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1836 * @dev: Network device
1837 * @txq: number of queues available
1839 * If real_num_tx_queues is changed the tc mappings may no longer be
1840 * valid. To resolve this verify the tc mapping remains valid and if
1841 * not NULL the mapping. With no priorities mapping to this
1842 * offset/count pair it will no longer be used. In the worst case TC0
1843 * is invalid nothing can be done so disable priority mappings. If is
1844 * expected that drivers will fix this mapping if they can before
1845 * calling netif_set_real_num_tx_queues.
1847 static void netif_setup_tc(struct net_device
*dev
, unsigned int txq
)
1850 struct netdev_tc_txq
*tc
= &dev
->tc_to_txq
[0];
1852 /* If TC0 is invalidated disable TC mapping */
1853 if (tc
->offset
+ tc
->count
> txq
) {
1854 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1859 /* Invalidated prio to tc mappings set to TC0 */
1860 for (i
= 1; i
< TC_BITMASK
+ 1; i
++) {
1861 int q
= netdev_get_prio_tc_map(dev
, i
);
1863 tc
= &dev
->tc_to_txq
[q
];
1864 if (tc
->offset
+ tc
->count
> txq
) {
1865 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1867 netdev_set_prio_tc_map(dev
, i
, 0);
1873 static DEFINE_MUTEX(xps_map_mutex
);
1874 #define xmap_dereference(P) \
1875 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1877 static struct xps_map
*remove_xps_queue(struct xps_dev_maps
*dev_maps
,
1880 struct xps_map
*map
= NULL
;
1884 map
= xmap_dereference(dev_maps
->cpu_map
[cpu
]);
1886 for (pos
= 0; map
&& pos
< map
->len
; pos
++) {
1887 if (map
->queues
[pos
] == index
) {
1889 map
->queues
[pos
] = map
->queues
[--map
->len
];
1891 RCU_INIT_POINTER(dev_maps
->cpu_map
[cpu
], NULL
);
1892 kfree_rcu(map
, rcu
);
1902 static void netif_reset_xps_queues_gt(struct net_device
*dev
, u16 index
)
1904 struct xps_dev_maps
*dev_maps
;
1906 bool active
= false;
1908 mutex_lock(&xps_map_mutex
);
1909 dev_maps
= xmap_dereference(dev
->xps_maps
);
1914 for_each_possible_cpu(cpu
) {
1915 for (i
= index
; i
< dev
->num_tx_queues
; i
++) {
1916 if (!remove_xps_queue(dev_maps
, cpu
, i
))
1919 if (i
== dev
->num_tx_queues
)
1924 RCU_INIT_POINTER(dev
->xps_maps
, NULL
);
1925 kfree_rcu(dev_maps
, rcu
);
1928 for (i
= index
; i
< dev
->num_tx_queues
; i
++)
1929 netdev_queue_numa_node_write(netdev_get_tx_queue(dev
, i
),
1933 mutex_unlock(&xps_map_mutex
);
1936 static struct xps_map
*expand_xps_map(struct xps_map
*map
,
1939 struct xps_map
*new_map
;
1940 int alloc_len
= XPS_MIN_MAP_ALLOC
;
1943 for (pos
= 0; map
&& pos
< map
->len
; pos
++) {
1944 if (map
->queues
[pos
] != index
)
1949 /* Need to add queue to this CPU's existing map */
1951 if (pos
< map
->alloc_len
)
1954 alloc_len
= map
->alloc_len
* 2;
1957 /* Need to allocate new map to store queue on this CPU's map */
1958 new_map
= kzalloc_node(XPS_MAP_SIZE(alloc_len
), GFP_KERNEL
,
1963 for (i
= 0; i
< pos
; i
++)
1964 new_map
->queues
[i
] = map
->queues
[i
];
1965 new_map
->alloc_len
= alloc_len
;
1971 int netif_set_xps_queue(struct net_device
*dev
, const struct cpumask
*mask
,
1974 struct xps_dev_maps
*dev_maps
, *new_dev_maps
= NULL
;
1975 struct xps_map
*map
, *new_map
;
1976 int maps_sz
= max_t(unsigned int, XPS_DEV_MAPS_SIZE
, L1_CACHE_BYTES
);
1977 int cpu
, numa_node_id
= -2;
1978 bool active
= false;
1980 mutex_lock(&xps_map_mutex
);
1982 dev_maps
= xmap_dereference(dev
->xps_maps
);
1984 /* allocate memory for queue storage */
1985 for_each_online_cpu(cpu
) {
1986 if (!cpumask_test_cpu(cpu
, mask
))
1990 new_dev_maps
= kzalloc(maps_sz
, GFP_KERNEL
);
1991 if (!new_dev_maps
) {
1992 mutex_unlock(&xps_map_mutex
);
1996 map
= dev_maps
? xmap_dereference(dev_maps
->cpu_map
[cpu
]) :
1999 map
= expand_xps_map(map
, cpu
, index
);
2003 RCU_INIT_POINTER(new_dev_maps
->cpu_map
[cpu
], map
);
2007 goto out_no_new_maps
;
2009 for_each_possible_cpu(cpu
) {
2010 if (cpumask_test_cpu(cpu
, mask
) && cpu_online(cpu
)) {
2011 /* add queue to CPU maps */
2014 map
= xmap_dereference(new_dev_maps
->cpu_map
[cpu
]);
2015 while ((pos
< map
->len
) && (map
->queues
[pos
] != index
))
2018 if (pos
== map
->len
)
2019 map
->queues
[map
->len
++] = index
;
2021 if (numa_node_id
== -2)
2022 numa_node_id
= cpu_to_node(cpu
);
2023 else if (numa_node_id
!= cpu_to_node(cpu
))
2026 } else if (dev_maps
) {
2027 /* fill in the new device map from the old device map */
2028 map
= xmap_dereference(dev_maps
->cpu_map
[cpu
]);
2029 RCU_INIT_POINTER(new_dev_maps
->cpu_map
[cpu
], map
);
2034 rcu_assign_pointer(dev
->xps_maps
, new_dev_maps
);
2036 /* Cleanup old maps */
2038 for_each_possible_cpu(cpu
) {
2039 new_map
= xmap_dereference(new_dev_maps
->cpu_map
[cpu
]);
2040 map
= xmap_dereference(dev_maps
->cpu_map
[cpu
]);
2041 if (map
&& map
!= new_map
)
2042 kfree_rcu(map
, rcu
);
2045 kfree_rcu(dev_maps
, rcu
);
2048 dev_maps
= new_dev_maps
;
2052 /* update Tx queue numa node */
2053 netdev_queue_numa_node_write(netdev_get_tx_queue(dev
, index
),
2054 (numa_node_id
>= 0) ? numa_node_id
:
2060 /* removes queue from unused CPUs */
2061 for_each_possible_cpu(cpu
) {
2062 if (cpumask_test_cpu(cpu
, mask
) && cpu_online(cpu
))
2065 if (remove_xps_queue(dev_maps
, cpu
, index
))
2069 /* free map if not active */
2071 RCU_INIT_POINTER(dev
->xps_maps
, NULL
);
2072 kfree_rcu(dev_maps
, rcu
);
2076 mutex_unlock(&xps_map_mutex
);
2080 /* remove any maps that we added */
2081 for_each_possible_cpu(cpu
) {
2082 new_map
= xmap_dereference(new_dev_maps
->cpu_map
[cpu
]);
2083 map
= dev_maps
? xmap_dereference(dev_maps
->cpu_map
[cpu
]) :
2085 if (new_map
&& new_map
!= map
)
2089 mutex_unlock(&xps_map_mutex
);
2091 kfree(new_dev_maps
);
2094 EXPORT_SYMBOL(netif_set_xps_queue
);
2098 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2099 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2101 int netif_set_real_num_tx_queues(struct net_device
*dev
, unsigned int txq
)
2105 if (txq
< 1 || txq
> dev
->num_tx_queues
)
2108 if (dev
->reg_state
== NETREG_REGISTERED
||
2109 dev
->reg_state
== NETREG_UNREGISTERING
) {
2112 rc
= netdev_queue_update_kobjects(dev
, dev
->real_num_tx_queues
,
2118 netif_setup_tc(dev
, txq
);
2120 if (txq
< dev
->real_num_tx_queues
) {
2121 qdisc_reset_all_tx_gt(dev
, txq
);
2123 netif_reset_xps_queues_gt(dev
, txq
);
2128 dev
->real_num_tx_queues
= txq
;
2131 EXPORT_SYMBOL(netif_set_real_num_tx_queues
);
2135 * netif_set_real_num_rx_queues - set actual number of RX queues used
2136 * @dev: Network device
2137 * @rxq: Actual number of RX queues
2139 * This must be called either with the rtnl_lock held or before
2140 * registration of the net device. Returns 0 on success, or a
2141 * negative error code. If called before registration, it always
2144 int netif_set_real_num_rx_queues(struct net_device
*dev
, unsigned int rxq
)
2148 if (rxq
< 1 || rxq
> dev
->num_rx_queues
)
2151 if (dev
->reg_state
== NETREG_REGISTERED
) {
2154 rc
= net_rx_queue_update_kobjects(dev
, dev
->real_num_rx_queues
,
2160 dev
->real_num_rx_queues
= rxq
;
2163 EXPORT_SYMBOL(netif_set_real_num_rx_queues
);
2167 * netif_get_num_default_rss_queues - default number of RSS queues
2169 * This routine should set an upper limit on the number of RSS queues
2170 * used by default by multiqueue devices.
2172 int netif_get_num_default_rss_queues(void)
2174 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES
, num_online_cpus());
2176 EXPORT_SYMBOL(netif_get_num_default_rss_queues
);
2178 static inline void __netif_reschedule(struct Qdisc
*q
)
2180 struct softnet_data
*sd
;
2181 unsigned long flags
;
2183 local_irq_save(flags
);
2184 sd
= this_cpu_ptr(&softnet_data
);
2185 q
->next_sched
= NULL
;
2186 *sd
->output_queue_tailp
= q
;
2187 sd
->output_queue_tailp
= &q
->next_sched
;
2188 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
2189 local_irq_restore(flags
);
2192 void __netif_schedule(struct Qdisc
*q
)
2194 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
2195 __netif_reschedule(q
);
2197 EXPORT_SYMBOL(__netif_schedule
);
2199 struct dev_kfree_skb_cb
{
2200 enum skb_free_reason reason
;
2203 static struct dev_kfree_skb_cb
*get_kfree_skb_cb(const struct sk_buff
*skb
)
2205 return (struct dev_kfree_skb_cb
*)skb
->cb
;
2208 void netif_schedule_queue(struct netdev_queue
*txq
)
2211 if (!(txq
->state
& QUEUE_STATE_ANY_XOFF
)) {
2212 struct Qdisc
*q
= rcu_dereference(txq
->qdisc
);
2214 __netif_schedule(q
);
2218 EXPORT_SYMBOL(netif_schedule_queue
);
2221 * netif_wake_subqueue - allow sending packets on subqueue
2222 * @dev: network device
2223 * @queue_index: sub queue index
2225 * Resume individual transmit queue of a device with multiple transmit queues.
2227 void netif_wake_subqueue(struct net_device
*dev
, u16 queue_index
)
2229 struct netdev_queue
*txq
= netdev_get_tx_queue(dev
, queue_index
);
2231 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF
, &txq
->state
)) {
2235 q
= rcu_dereference(txq
->qdisc
);
2236 __netif_schedule(q
);
2240 EXPORT_SYMBOL(netif_wake_subqueue
);
2242 void netif_tx_wake_queue(struct netdev_queue
*dev_queue
)
2244 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF
, &dev_queue
->state
)) {
2248 q
= rcu_dereference(dev_queue
->qdisc
);
2249 __netif_schedule(q
);
2253 EXPORT_SYMBOL(netif_tx_wake_queue
);
2255 void __dev_kfree_skb_irq(struct sk_buff
*skb
, enum skb_free_reason reason
)
2257 unsigned long flags
;
2259 if (likely(atomic_read(&skb
->users
) == 1)) {
2261 atomic_set(&skb
->users
, 0);
2262 } else if (likely(!atomic_dec_and_test(&skb
->users
))) {
2265 get_kfree_skb_cb(skb
)->reason
= reason
;
2266 local_irq_save(flags
);
2267 skb
->next
= __this_cpu_read(softnet_data
.completion_queue
);
2268 __this_cpu_write(softnet_data
.completion_queue
, skb
);
2269 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
2270 local_irq_restore(flags
);
2272 EXPORT_SYMBOL(__dev_kfree_skb_irq
);
2274 void __dev_kfree_skb_any(struct sk_buff
*skb
, enum skb_free_reason reason
)
2276 if (in_irq() || irqs_disabled())
2277 __dev_kfree_skb_irq(skb
, reason
);
2281 EXPORT_SYMBOL(__dev_kfree_skb_any
);
2285 * netif_device_detach - mark device as removed
2286 * @dev: network device
2288 * Mark device as removed from system and therefore no longer available.
2290 void netif_device_detach(struct net_device
*dev
)
2292 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
2293 netif_running(dev
)) {
2294 netif_tx_stop_all_queues(dev
);
2297 EXPORT_SYMBOL(netif_device_detach
);
2300 * netif_device_attach - mark device as attached
2301 * @dev: network device
2303 * Mark device as attached from system and restart if needed.
2305 void netif_device_attach(struct net_device
*dev
)
2307 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
2308 netif_running(dev
)) {
2309 netif_tx_wake_all_queues(dev
);
2310 __netdev_watchdog_up(dev
);
2313 EXPORT_SYMBOL(netif_device_attach
);
2315 static void skb_warn_bad_offload(const struct sk_buff
*skb
)
2317 static const netdev_features_t null_features
= 0;
2318 struct net_device
*dev
= skb
->dev
;
2319 const char *driver
= "";
2321 if (!net_ratelimit())
2324 if (dev
&& dev
->dev
.parent
)
2325 driver
= dev_driver_string(dev
->dev
.parent
);
2327 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2328 "gso_type=%d ip_summed=%d\n",
2329 driver
, dev
? &dev
->features
: &null_features
,
2330 skb
->sk
? &skb
->sk
->sk_route_caps
: &null_features
,
2331 skb
->len
, skb
->data_len
, skb_shinfo(skb
)->gso_size
,
2332 skb_shinfo(skb
)->gso_type
, skb
->ip_summed
);
2336 * Invalidate hardware checksum when packet is to be mangled, and
2337 * complete checksum manually on outgoing path.
2339 int skb_checksum_help(struct sk_buff
*skb
)
2342 int ret
= 0, offset
;
2344 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
2345 goto out_set_summed
;
2347 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
2348 skb_warn_bad_offload(skb
);
2352 /* Before computing a checksum, we should make sure no frag could
2353 * be modified by an external entity : checksum could be wrong.
2355 if (skb_has_shared_frag(skb
)) {
2356 ret
= __skb_linearize(skb
);
2361 offset
= skb_checksum_start_offset(skb
);
2362 BUG_ON(offset
>= skb_headlen(skb
));
2363 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
2365 offset
+= skb
->csum_offset
;
2366 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
2368 if (skb_cloned(skb
) &&
2369 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
2370 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2375 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
2377 skb
->ip_summed
= CHECKSUM_NONE
;
2381 EXPORT_SYMBOL(skb_checksum_help
);
2383 __be16
skb_network_protocol(struct sk_buff
*skb
, int *depth
)
2385 __be16 type
= skb
->protocol
;
2387 /* Tunnel gso handlers can set protocol to ethernet. */
2388 if (type
== htons(ETH_P_TEB
)) {
2391 if (unlikely(!pskb_may_pull(skb
, sizeof(struct ethhdr
))))
2394 eth
= (struct ethhdr
*)skb_mac_header(skb
);
2395 type
= eth
->h_proto
;
2398 return __vlan_get_protocol(skb
, type
, depth
);
2402 * skb_mac_gso_segment - mac layer segmentation handler.
2403 * @skb: buffer to segment
2404 * @features: features for the output path (see dev->features)
2406 struct sk_buff
*skb_mac_gso_segment(struct sk_buff
*skb
,
2407 netdev_features_t features
)
2409 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
2410 struct packet_offload
*ptype
;
2411 int vlan_depth
= skb
->mac_len
;
2412 __be16 type
= skb_network_protocol(skb
, &vlan_depth
);
2414 if (unlikely(!type
))
2415 return ERR_PTR(-EINVAL
);
2417 __skb_pull(skb
, vlan_depth
);
2420 list_for_each_entry_rcu(ptype
, &offload_base
, list
) {
2421 if (ptype
->type
== type
&& ptype
->callbacks
.gso_segment
) {
2422 segs
= ptype
->callbacks
.gso_segment(skb
, features
);
2428 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
2432 EXPORT_SYMBOL(skb_mac_gso_segment
);
2435 /* openvswitch calls this on rx path, so we need a different check.
2437 static inline bool skb_needs_check(struct sk_buff
*skb
, bool tx_path
)
2440 return skb
->ip_summed
!= CHECKSUM_PARTIAL
;
2442 return skb
->ip_summed
== CHECKSUM_NONE
;
2446 * __skb_gso_segment - Perform segmentation on skb.
2447 * @skb: buffer to segment
2448 * @features: features for the output path (see dev->features)
2449 * @tx_path: whether it is called in TX path
2451 * This function segments the given skb and returns a list of segments.
2453 * It may return NULL if the skb requires no segmentation. This is
2454 * only possible when GSO is used for verifying header integrity.
2456 struct sk_buff
*__skb_gso_segment(struct sk_buff
*skb
,
2457 netdev_features_t features
, bool tx_path
)
2459 if (unlikely(skb_needs_check(skb
, tx_path
))) {
2462 skb_warn_bad_offload(skb
);
2464 err
= skb_cow_head(skb
, 0);
2466 return ERR_PTR(err
);
2469 SKB_GSO_CB(skb
)->mac_offset
= skb_headroom(skb
);
2470 SKB_GSO_CB(skb
)->encap_level
= 0;
2472 skb_reset_mac_header(skb
);
2473 skb_reset_mac_len(skb
);
2475 return skb_mac_gso_segment(skb
, features
);
2477 EXPORT_SYMBOL(__skb_gso_segment
);
2479 /* Take action when hardware reception checksum errors are detected. */
2481 void netdev_rx_csum_fault(struct net_device
*dev
)
2483 if (net_ratelimit()) {
2484 pr_err("%s: hw csum failure\n", dev
? dev
->name
: "<unknown>");
2488 EXPORT_SYMBOL(netdev_rx_csum_fault
);
2491 /* Actually, we should eliminate this check as soon as we know, that:
2492 * 1. IOMMU is present and allows to map all the memory.
2493 * 2. No high memory really exists on this machine.
2496 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
2498 #ifdef CONFIG_HIGHMEM
2500 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
2501 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2502 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2503 if (PageHighMem(skb_frag_page(frag
)))
2508 if (PCI_DMA_BUS_IS_PHYS
) {
2509 struct device
*pdev
= dev
->dev
.parent
;
2513 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2514 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2515 dma_addr_t addr
= page_to_phys(skb_frag_page(frag
));
2516 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
2524 /* If MPLS offload request, verify we are testing hardware MPLS features
2525 * instead of standard features for the netdev.
2527 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
2528 static netdev_features_t
net_mpls_features(struct sk_buff
*skb
,
2529 netdev_features_t features
,
2532 if (eth_p_mpls(type
))
2533 features
&= skb
->dev
->mpls_features
;
2538 static netdev_features_t
net_mpls_features(struct sk_buff
*skb
,
2539 netdev_features_t features
,
2546 static netdev_features_t
harmonize_features(struct sk_buff
*skb
,
2547 netdev_features_t features
)
2552 type
= skb_network_protocol(skb
, &tmp
);
2553 features
= net_mpls_features(skb
, features
, type
);
2555 if (skb
->ip_summed
!= CHECKSUM_NONE
&&
2556 !can_checksum_protocol(features
, type
)) {
2557 features
&= ~NETIF_F_ALL_CSUM
;
2558 } else if (illegal_highdma(skb
->dev
, skb
)) {
2559 features
&= ~NETIF_F_SG
;
2565 netdev_features_t
passthru_features_check(struct sk_buff
*skb
,
2566 struct net_device
*dev
,
2567 netdev_features_t features
)
2571 EXPORT_SYMBOL(passthru_features_check
);
2573 static netdev_features_t
dflt_features_check(const struct sk_buff
*skb
,
2574 struct net_device
*dev
,
2575 netdev_features_t features
)
2577 return vlan_features_check(skb
, features
);
2580 netdev_features_t
netif_skb_features(struct sk_buff
*skb
)
2582 struct net_device
*dev
= skb
->dev
;
2583 netdev_features_t features
= dev
->features
;
2584 u16 gso_segs
= skb_shinfo(skb
)->gso_segs
;
2586 if (gso_segs
> dev
->gso_max_segs
|| gso_segs
< dev
->gso_min_segs
)
2587 features
&= ~NETIF_F_GSO_MASK
;
2589 /* If encapsulation offload request, verify we are testing
2590 * hardware encapsulation features instead of standard
2591 * features for the netdev
2593 if (skb
->encapsulation
)
2594 features
&= dev
->hw_enc_features
;
2596 if (skb_vlan_tagged(skb
))
2597 features
= netdev_intersect_features(features
,
2598 dev
->vlan_features
|
2599 NETIF_F_HW_VLAN_CTAG_TX
|
2600 NETIF_F_HW_VLAN_STAG_TX
);
2602 if (dev
->netdev_ops
->ndo_features_check
)
2603 features
&= dev
->netdev_ops
->ndo_features_check(skb
, dev
,
2606 features
&= dflt_features_check(skb
, dev
, features
);
2608 return harmonize_features(skb
, features
);
2610 EXPORT_SYMBOL(netif_skb_features
);
2612 static int xmit_one(struct sk_buff
*skb
, struct net_device
*dev
,
2613 struct netdev_queue
*txq
, bool more
)
2618 if (!list_empty(&ptype_all
) || !list_empty(&dev
->ptype_all
))
2619 dev_queue_xmit_nit(skb
, dev
);
2622 trace_net_dev_start_xmit(skb
, dev
);
2623 rc
= netdev_start_xmit(skb
, dev
, txq
, more
);
2624 trace_net_dev_xmit(skb
, rc
, dev
, len
);
2629 struct sk_buff
*dev_hard_start_xmit(struct sk_buff
*first
, struct net_device
*dev
,
2630 struct netdev_queue
*txq
, int *ret
)
2632 struct sk_buff
*skb
= first
;
2633 int rc
= NETDEV_TX_OK
;
2636 struct sk_buff
*next
= skb
->next
;
2639 rc
= xmit_one(skb
, dev
, txq
, next
!= NULL
);
2640 if (unlikely(!dev_xmit_complete(rc
))) {
2646 if (netif_xmit_stopped(txq
) && skb
) {
2647 rc
= NETDEV_TX_BUSY
;
2657 static struct sk_buff
*validate_xmit_vlan(struct sk_buff
*skb
,
2658 netdev_features_t features
)
2660 if (skb_vlan_tag_present(skb
) &&
2661 !vlan_hw_offload_capable(features
, skb
->vlan_proto
))
2662 skb
= __vlan_hwaccel_push_inside(skb
);
2666 static struct sk_buff
*validate_xmit_skb(struct sk_buff
*skb
, struct net_device
*dev
)
2668 netdev_features_t features
;
2673 features
= netif_skb_features(skb
);
2674 skb
= validate_xmit_vlan(skb
, features
);
2678 if (netif_needs_gso(dev
, skb
, features
)) {
2679 struct sk_buff
*segs
;
2681 segs
= skb_gso_segment(skb
, features
);
2689 if (skb_needs_linearize(skb
, features
) &&
2690 __skb_linearize(skb
))
2693 /* If packet is not checksummed and device does not
2694 * support checksumming for this protocol, complete
2695 * checksumming here.
2697 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2698 if (skb
->encapsulation
)
2699 skb_set_inner_transport_header(skb
,
2700 skb_checksum_start_offset(skb
));
2702 skb_set_transport_header(skb
,
2703 skb_checksum_start_offset(skb
));
2704 if (!(features
& NETIF_F_ALL_CSUM
) &&
2705 skb_checksum_help(skb
))
2718 struct sk_buff
*validate_xmit_skb_list(struct sk_buff
*skb
, struct net_device
*dev
)
2720 struct sk_buff
*next
, *head
= NULL
, *tail
;
2722 for (; skb
!= NULL
; skb
= next
) {
2726 /* in case skb wont be segmented, point to itself */
2729 skb
= validate_xmit_skb(skb
, dev
);
2737 /* If skb was segmented, skb->prev points to
2738 * the last segment. If not, it still contains skb.
2745 static void qdisc_pkt_len_init(struct sk_buff
*skb
)
2747 const struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
2749 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
2751 /* To get more precise estimation of bytes sent on wire,
2752 * we add to pkt_len the headers size of all segments
2754 if (shinfo
->gso_size
) {
2755 unsigned int hdr_len
;
2756 u16 gso_segs
= shinfo
->gso_segs
;
2758 /* mac layer + network layer */
2759 hdr_len
= skb_transport_header(skb
) - skb_mac_header(skb
);
2761 /* + transport layer */
2762 if (likely(shinfo
->gso_type
& (SKB_GSO_TCPV4
| SKB_GSO_TCPV6
)))
2763 hdr_len
+= tcp_hdrlen(skb
);
2765 hdr_len
+= sizeof(struct udphdr
);
2767 if (shinfo
->gso_type
& SKB_GSO_DODGY
)
2768 gso_segs
= DIV_ROUND_UP(skb
->len
- hdr_len
,
2771 qdisc_skb_cb(skb
)->pkt_len
+= (gso_segs
- 1) * hdr_len
;
2775 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2776 struct net_device
*dev
,
2777 struct netdev_queue
*txq
)
2779 spinlock_t
*root_lock
= qdisc_lock(q
);
2783 qdisc_pkt_len_init(skb
);
2784 qdisc_calculate_pkt_len(skb
, q
);
2786 * Heuristic to force contended enqueues to serialize on a
2787 * separate lock before trying to get qdisc main lock.
2788 * This permits __QDISC___STATE_RUNNING owner to get the lock more
2789 * often and dequeue packets faster.
2791 contended
= qdisc_is_running(q
);
2792 if (unlikely(contended
))
2793 spin_lock(&q
->busylock
);
2795 spin_lock(root_lock
);
2796 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2799 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2800 qdisc_run_begin(q
)) {
2802 * This is a work-conserving queue; there are no old skbs
2803 * waiting to be sent out; and the qdisc is not running -
2804 * xmit the skb directly.
2807 qdisc_bstats_update(q
, skb
);
2809 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
, true)) {
2810 if (unlikely(contended
)) {
2811 spin_unlock(&q
->busylock
);
2818 rc
= NET_XMIT_SUCCESS
;
2820 rc
= q
->enqueue(skb
, q
) & NET_XMIT_MASK
;
2821 if (qdisc_run_begin(q
)) {
2822 if (unlikely(contended
)) {
2823 spin_unlock(&q
->busylock
);
2829 spin_unlock(root_lock
);
2830 if (unlikely(contended
))
2831 spin_unlock(&q
->busylock
);
2835 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2836 static void skb_update_prio(struct sk_buff
*skb
)
2838 struct netprio_map
*map
= rcu_dereference_bh(skb
->dev
->priomap
);
2840 if (!skb
->priority
&& skb
->sk
&& map
) {
2841 unsigned int prioidx
= skb
->sk
->sk_cgrp_prioidx
;
2843 if (prioidx
< map
->priomap_len
)
2844 skb
->priority
= map
->priomap
[prioidx
];
2848 #define skb_update_prio(skb)
2851 static DEFINE_PER_CPU(int, xmit_recursion
);
2852 #define RECURSION_LIMIT 10
2855 * dev_loopback_xmit - loop back @skb
2856 * @skb: buffer to transmit
2858 int dev_loopback_xmit(struct sk_buff
*skb
)
2860 skb_reset_mac_header(skb
);
2861 __skb_pull(skb
, skb_network_offset(skb
));
2862 skb
->pkt_type
= PACKET_LOOPBACK
;
2863 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
2864 WARN_ON(!skb_dst(skb
));
2869 EXPORT_SYMBOL(dev_loopback_xmit
);
2872 * __dev_queue_xmit - transmit a buffer
2873 * @skb: buffer to transmit
2874 * @accel_priv: private data used for L2 forwarding offload
2876 * Queue a buffer for transmission to a network device. The caller must
2877 * have set the device and priority and built the buffer before calling
2878 * this function. The function can be called from an interrupt.
2880 * A negative errno code is returned on a failure. A success does not
2881 * guarantee the frame will be transmitted as it may be dropped due
2882 * to congestion or traffic shaping.
2884 * -----------------------------------------------------------------------------------
2885 * I notice this method can also return errors from the queue disciplines,
2886 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2889 * Regardless of the return value, the skb is consumed, so it is currently
2890 * difficult to retry a send to this method. (You can bump the ref count
2891 * before sending to hold a reference for retry if you are careful.)
2893 * When calling this method, interrupts MUST be enabled. This is because
2894 * the BH enable code must have IRQs enabled so that it will not deadlock.
2897 static int __dev_queue_xmit(struct sk_buff
*skb
, void *accel_priv
)
2899 struct net_device
*dev
= skb
->dev
;
2900 struct netdev_queue
*txq
;
2904 skb_reset_mac_header(skb
);
2906 if (unlikely(skb_shinfo(skb
)->tx_flags
& SKBTX_SCHED_TSTAMP
))
2907 __skb_tstamp_tx(skb
, NULL
, skb
->sk
, SCM_TSTAMP_SCHED
);
2909 /* Disable soft irqs for various locks below. Also
2910 * stops preemption for RCU.
2914 skb_update_prio(skb
);
2916 /* If device/qdisc don't need skb->dst, release it right now while
2917 * its hot in this cpu cache.
2919 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2924 txq
= netdev_pick_tx(dev
, skb
, accel_priv
);
2925 q
= rcu_dereference_bh(txq
->qdisc
);
2927 #ifdef CONFIG_NET_CLS_ACT
2928 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2930 trace_net_dev_queue(skb
);
2932 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2936 /* The device has no queue. Common case for software devices:
2937 loopback, all the sorts of tunnels...
2939 Really, it is unlikely that netif_tx_lock protection is necessary
2940 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2942 However, it is possible, that they rely on protection
2945 Check this and shot the lock. It is not prone from deadlocks.
2946 Either shot noqueue qdisc, it is even simpler 8)
2948 if (dev
->flags
& IFF_UP
) {
2949 int cpu
= smp_processor_id(); /* ok because BHs are off */
2951 if (txq
->xmit_lock_owner
!= cpu
) {
2953 if (__this_cpu_read(xmit_recursion
) > RECURSION_LIMIT
)
2954 goto recursion_alert
;
2956 skb
= validate_xmit_skb(skb
, dev
);
2960 HARD_TX_LOCK(dev
, txq
, cpu
);
2962 if (!netif_xmit_stopped(txq
)) {
2963 __this_cpu_inc(xmit_recursion
);
2964 skb
= dev_hard_start_xmit(skb
, dev
, txq
, &rc
);
2965 __this_cpu_dec(xmit_recursion
);
2966 if (dev_xmit_complete(rc
)) {
2967 HARD_TX_UNLOCK(dev
, txq
);
2971 HARD_TX_UNLOCK(dev
, txq
);
2972 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2975 /* Recursion is detected! It is possible,
2979 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2986 rcu_read_unlock_bh();
2988 atomic_long_inc(&dev
->tx_dropped
);
2989 kfree_skb_list(skb
);
2992 rcu_read_unlock_bh();
2996 int dev_queue_xmit(struct sk_buff
*skb
)
2998 return __dev_queue_xmit(skb
, NULL
);
3000 EXPORT_SYMBOL(dev_queue_xmit
);
3002 int dev_queue_xmit_accel(struct sk_buff
*skb
, void *accel_priv
)
3004 return __dev_queue_xmit(skb
, accel_priv
);
3006 EXPORT_SYMBOL(dev_queue_xmit_accel
);
3009 /*=======================================================================
3011 =======================================================================*/
3013 int netdev_max_backlog __read_mostly
= 1000;
3014 EXPORT_SYMBOL(netdev_max_backlog
);
3016 int netdev_tstamp_prequeue __read_mostly
= 1;
3017 int netdev_budget __read_mostly
= 300;
3018 int weight_p __read_mostly
= 64; /* old backlog weight */
3020 /* Called with irq disabled */
3021 static inline void ____napi_schedule(struct softnet_data
*sd
,
3022 struct napi_struct
*napi
)
3024 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
3025 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3030 /* One global table that all flow-based protocols share. */
3031 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
3032 EXPORT_SYMBOL(rps_sock_flow_table
);
3033 u32 rps_cpu_mask __read_mostly
;
3034 EXPORT_SYMBOL(rps_cpu_mask
);
3036 struct static_key rps_needed __read_mostly
;
3038 static struct rps_dev_flow
*
3039 set_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
3040 struct rps_dev_flow
*rflow
, u16 next_cpu
)
3042 if (next_cpu
!= RPS_NO_CPU
) {
3043 #ifdef CONFIG_RFS_ACCEL
3044 struct netdev_rx_queue
*rxqueue
;
3045 struct rps_dev_flow_table
*flow_table
;
3046 struct rps_dev_flow
*old_rflow
;
3051 /* Should we steer this flow to a different hardware queue? */
3052 if (!skb_rx_queue_recorded(skb
) || !dev
->rx_cpu_rmap
||
3053 !(dev
->features
& NETIF_F_NTUPLE
))
3055 rxq_index
= cpu_rmap_lookup_index(dev
->rx_cpu_rmap
, next_cpu
);
3056 if (rxq_index
== skb_get_rx_queue(skb
))
3059 rxqueue
= dev
->_rx
+ rxq_index
;
3060 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3063 flow_id
= skb_get_hash(skb
) & flow_table
->mask
;
3064 rc
= dev
->netdev_ops
->ndo_rx_flow_steer(dev
, skb
,
3065 rxq_index
, flow_id
);
3069 rflow
= &flow_table
->flows
[flow_id
];
3071 if (old_rflow
->filter
== rflow
->filter
)
3072 old_rflow
->filter
= RPS_NO_FILTER
;
3076 per_cpu(softnet_data
, next_cpu
).input_queue_head
;
3079 rflow
->cpu
= next_cpu
;
3084 * get_rps_cpu is called from netif_receive_skb and returns the target
3085 * CPU from the RPS map of the receiving queue for a given skb.
3086 * rcu_read_lock must be held on entry.
3088 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
3089 struct rps_dev_flow
**rflowp
)
3091 const struct rps_sock_flow_table
*sock_flow_table
;
3092 struct netdev_rx_queue
*rxqueue
= dev
->_rx
;
3093 struct rps_dev_flow_table
*flow_table
;
3094 struct rps_map
*map
;
3099 if (skb_rx_queue_recorded(skb
)) {
3100 u16 index
= skb_get_rx_queue(skb
);
3102 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
3103 WARN_ONCE(dev
->real_num_rx_queues
> 1,
3104 "%s received packet on queue %u, but number "
3105 "of RX queues is %u\n",
3106 dev
->name
, index
, dev
->real_num_rx_queues
);
3112 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3114 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3115 map
= rcu_dereference(rxqueue
->rps_map
);
3116 if (!flow_table
&& !map
)
3119 skb_reset_network_header(skb
);
3120 hash
= skb_get_hash(skb
);
3124 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
3125 if (flow_table
&& sock_flow_table
) {
3126 struct rps_dev_flow
*rflow
;
3130 /* First check into global flow table if there is a match */
3131 ident
= sock_flow_table
->ents
[hash
& sock_flow_table
->mask
];
3132 if ((ident
^ hash
) & ~rps_cpu_mask
)
3135 next_cpu
= ident
& rps_cpu_mask
;
3137 /* OK, now we know there is a match,
3138 * we can look at the local (per receive queue) flow table
3140 rflow
= &flow_table
->flows
[hash
& flow_table
->mask
];
3144 * If the desired CPU (where last recvmsg was done) is
3145 * different from current CPU (one in the rx-queue flow
3146 * table entry), switch if one of the following holds:
3147 * - Current CPU is unset (equal to RPS_NO_CPU).
3148 * - Current CPU is offline.
3149 * - The current CPU's queue tail has advanced beyond the
3150 * last packet that was enqueued using this table entry.
3151 * This guarantees that all previous packets for the flow
3152 * have been dequeued, thus preserving in order delivery.
3154 if (unlikely(tcpu
!= next_cpu
) &&
3155 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
3156 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
3157 rflow
->last_qtail
)) >= 0)) {
3159 rflow
= set_rps_cpu(dev
, skb
, rflow
, next_cpu
);
3162 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
3172 tcpu
= map
->cpus
[reciprocal_scale(hash
, map
->len
)];
3173 if (cpu_online(tcpu
)) {
3183 #ifdef CONFIG_RFS_ACCEL
3186 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3187 * @dev: Device on which the filter was set
3188 * @rxq_index: RX queue index
3189 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3190 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3192 * Drivers that implement ndo_rx_flow_steer() should periodically call
3193 * this function for each installed filter and remove the filters for
3194 * which it returns %true.
3196 bool rps_may_expire_flow(struct net_device
*dev
, u16 rxq_index
,
3197 u32 flow_id
, u16 filter_id
)
3199 struct netdev_rx_queue
*rxqueue
= dev
->_rx
+ rxq_index
;
3200 struct rps_dev_flow_table
*flow_table
;
3201 struct rps_dev_flow
*rflow
;
3206 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3207 if (flow_table
&& flow_id
<= flow_table
->mask
) {
3208 rflow
= &flow_table
->flows
[flow_id
];
3209 cpu
= ACCESS_ONCE(rflow
->cpu
);
3210 if (rflow
->filter
== filter_id
&& cpu
!= RPS_NO_CPU
&&
3211 ((int)(per_cpu(softnet_data
, cpu
).input_queue_head
-
3212 rflow
->last_qtail
) <
3213 (int)(10 * flow_table
->mask
)))
3219 EXPORT_SYMBOL(rps_may_expire_flow
);
3221 #endif /* CONFIG_RFS_ACCEL */
3223 /* Called from hardirq (IPI) context */
3224 static void rps_trigger_softirq(void *data
)
3226 struct softnet_data
*sd
= data
;
3228 ____napi_schedule(sd
, &sd
->backlog
);
3232 #endif /* CONFIG_RPS */
3235 * Check if this softnet_data structure is another cpu one
3236 * If yes, queue it to our IPI list and return 1
3239 static int rps_ipi_queued(struct softnet_data
*sd
)
3242 struct softnet_data
*mysd
= this_cpu_ptr(&softnet_data
);
3245 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
3246 mysd
->rps_ipi_list
= sd
;
3248 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3251 #endif /* CONFIG_RPS */
3255 #ifdef CONFIG_NET_FLOW_LIMIT
3256 int netdev_flow_limit_table_len __read_mostly
= (1 << 12);
3259 static bool skb_flow_limit(struct sk_buff
*skb
, unsigned int qlen
)
3261 #ifdef CONFIG_NET_FLOW_LIMIT
3262 struct sd_flow_limit
*fl
;
3263 struct softnet_data
*sd
;
3264 unsigned int old_flow
, new_flow
;
3266 if (qlen
< (netdev_max_backlog
>> 1))
3269 sd
= this_cpu_ptr(&softnet_data
);
3272 fl
= rcu_dereference(sd
->flow_limit
);
3274 new_flow
= skb_get_hash(skb
) & (fl
->num_buckets
- 1);
3275 old_flow
= fl
->history
[fl
->history_head
];
3276 fl
->history
[fl
->history_head
] = new_flow
;
3279 fl
->history_head
&= FLOW_LIMIT_HISTORY
- 1;
3281 if (likely(fl
->buckets
[old_flow
]))
3282 fl
->buckets
[old_flow
]--;
3284 if (++fl
->buckets
[new_flow
] > (FLOW_LIMIT_HISTORY
>> 1)) {
3296 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3297 * queue (may be a remote CPU queue).
3299 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
3300 unsigned int *qtail
)
3302 struct softnet_data
*sd
;
3303 unsigned long flags
;
3306 sd
= &per_cpu(softnet_data
, cpu
);
3308 local_irq_save(flags
);
3311 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3312 if (qlen
<= netdev_max_backlog
&& !skb_flow_limit(skb
, qlen
)) {
3315 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
3316 input_queue_tail_incr_save(sd
, qtail
);
3318 local_irq_restore(flags
);
3319 return NET_RX_SUCCESS
;
3322 /* Schedule NAPI for backlog device
3323 * We can use non atomic operation since we own the queue lock
3325 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
3326 if (!rps_ipi_queued(sd
))
3327 ____napi_schedule(sd
, &sd
->backlog
);
3335 local_irq_restore(flags
);
3337 atomic_long_inc(&skb
->dev
->rx_dropped
);
3342 static int netif_rx_internal(struct sk_buff
*skb
)
3346 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3348 trace_netif_rx(skb
);
3350 if (static_key_false(&rps_needed
)) {
3351 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3357 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3359 cpu
= smp_processor_id();
3361 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3369 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
3376 * netif_rx - post buffer to the network code
3377 * @skb: buffer to post
3379 * This function receives a packet from a device driver and queues it for
3380 * the upper (protocol) levels to process. It always succeeds. The buffer
3381 * may be dropped during processing for congestion control or by the
3385 * NET_RX_SUCCESS (no congestion)
3386 * NET_RX_DROP (packet was dropped)
3390 int netif_rx(struct sk_buff
*skb
)
3392 trace_netif_rx_entry(skb
);
3394 return netif_rx_internal(skb
);
3396 EXPORT_SYMBOL(netif_rx
);
3398 int netif_rx_ni(struct sk_buff
*skb
)
3402 trace_netif_rx_ni_entry(skb
);
3405 err
= netif_rx_internal(skb
);
3406 if (local_softirq_pending())
3412 EXPORT_SYMBOL(netif_rx_ni
);
3414 static void net_tx_action(struct softirq_action
*h
)
3416 struct softnet_data
*sd
= this_cpu_ptr(&softnet_data
);
3418 if (sd
->completion_queue
) {
3419 struct sk_buff
*clist
;
3421 local_irq_disable();
3422 clist
= sd
->completion_queue
;
3423 sd
->completion_queue
= NULL
;
3427 struct sk_buff
*skb
= clist
;
3428 clist
= clist
->next
;
3430 WARN_ON(atomic_read(&skb
->users
));
3431 if (likely(get_kfree_skb_cb(skb
)->reason
== SKB_REASON_CONSUMED
))
3432 trace_consume_skb(skb
);
3434 trace_kfree_skb(skb
, net_tx_action
);
3439 if (sd
->output_queue
) {
3442 local_irq_disable();
3443 head
= sd
->output_queue
;
3444 sd
->output_queue
= NULL
;
3445 sd
->output_queue_tailp
= &sd
->output_queue
;
3449 struct Qdisc
*q
= head
;
3450 spinlock_t
*root_lock
;
3452 head
= head
->next_sched
;
3454 root_lock
= qdisc_lock(q
);
3455 if (spin_trylock(root_lock
)) {
3456 smp_mb__before_atomic();
3457 clear_bit(__QDISC_STATE_SCHED
,
3460 spin_unlock(root_lock
);
3462 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
3464 __netif_reschedule(q
);
3466 smp_mb__before_atomic();
3467 clear_bit(__QDISC_STATE_SCHED
,
3475 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3476 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3477 /* This hook is defined here for ATM LANE */
3478 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
3479 unsigned char *addr
) __read_mostly
;
3480 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
3483 #ifdef CONFIG_NET_CLS_ACT
3484 /* TODO: Maybe we should just force sch_ingress to be compiled in
3485 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3486 * a compare and 2 stores extra right now if we dont have it on
3487 * but have CONFIG_NET_CLS_ACT
3488 * NOTE: This doesn't stop any functionality; if you dont have
3489 * the ingress scheduler, you just can't add policies on ingress.
3492 static int ing_filter(struct sk_buff
*skb
, struct netdev_queue
*rxq
)
3494 struct net_device
*dev
= skb
->dev
;
3495 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
3496 int result
= TC_ACT_OK
;
3499 if (unlikely(MAX_RED_LOOP
< ttl
++)) {
3500 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3501 skb
->skb_iif
, dev
->ifindex
);
3505 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
3506 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
3508 q
= rcu_dereference(rxq
->qdisc
);
3509 if (q
!= &noop_qdisc
) {
3510 spin_lock(qdisc_lock(q
));
3511 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
3512 result
= qdisc_enqueue_root(skb
, q
);
3513 spin_unlock(qdisc_lock(q
));
3519 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
3520 struct packet_type
**pt_prev
,
3521 int *ret
, struct net_device
*orig_dev
)
3523 struct netdev_queue
*rxq
= rcu_dereference(skb
->dev
->ingress_queue
);
3525 if (!rxq
|| rcu_access_pointer(rxq
->qdisc
) == &noop_qdisc
)
3529 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
3533 switch (ing_filter(skb
, rxq
)) {
3547 * netdev_rx_handler_register - register receive handler
3548 * @dev: device to register a handler for
3549 * @rx_handler: receive handler to register
3550 * @rx_handler_data: data pointer that is used by rx handler
3552 * Register a receive handler for a device. This handler will then be
3553 * called from __netif_receive_skb. A negative errno code is returned
3556 * The caller must hold the rtnl_mutex.
3558 * For a general description of rx_handler, see enum rx_handler_result.
3560 int netdev_rx_handler_register(struct net_device
*dev
,
3561 rx_handler_func_t
*rx_handler
,
3562 void *rx_handler_data
)
3566 if (dev
->rx_handler
)
3569 /* Note: rx_handler_data must be set before rx_handler */
3570 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
3571 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
3575 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
3578 * netdev_rx_handler_unregister - unregister receive handler
3579 * @dev: device to unregister a handler from
3581 * Unregister a receive handler from a device.
3583 * The caller must hold the rtnl_mutex.
3585 void netdev_rx_handler_unregister(struct net_device
*dev
)
3589 RCU_INIT_POINTER(dev
->rx_handler
, NULL
);
3590 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3591 * section has a guarantee to see a non NULL rx_handler_data
3595 RCU_INIT_POINTER(dev
->rx_handler_data
, NULL
);
3597 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
3600 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3601 * the special handling of PFMEMALLOC skbs.
3603 static bool skb_pfmemalloc_protocol(struct sk_buff
*skb
)
3605 switch (skb
->protocol
) {
3606 case htons(ETH_P_ARP
):
3607 case htons(ETH_P_IP
):
3608 case htons(ETH_P_IPV6
):
3609 case htons(ETH_P_8021Q
):
3610 case htons(ETH_P_8021AD
):
3617 static int __netif_receive_skb_core(struct sk_buff
*skb
, bool pfmemalloc
)
3619 struct packet_type
*ptype
, *pt_prev
;
3620 rx_handler_func_t
*rx_handler
;
3621 struct net_device
*orig_dev
;
3622 bool deliver_exact
= false;
3623 int ret
= NET_RX_DROP
;
3626 net_timestamp_check(!netdev_tstamp_prequeue
, skb
);
3628 trace_netif_receive_skb(skb
);
3630 orig_dev
= skb
->dev
;
3632 skb_reset_network_header(skb
);
3633 if (!skb_transport_header_was_set(skb
))
3634 skb_reset_transport_header(skb
);
3635 skb_reset_mac_len(skb
);
3642 skb
->skb_iif
= skb
->dev
->ifindex
;
3644 __this_cpu_inc(softnet_data
.processed
);
3646 if (skb
->protocol
== cpu_to_be16(ETH_P_8021Q
) ||
3647 skb
->protocol
== cpu_to_be16(ETH_P_8021AD
)) {
3648 skb
= skb_vlan_untag(skb
);
3653 #ifdef CONFIG_NET_CLS_ACT
3654 if (skb
->tc_verd
& TC_NCLS
) {
3655 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
3663 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
3665 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3669 list_for_each_entry_rcu(ptype
, &skb
->dev
->ptype_all
, list
) {
3671 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3676 #ifdef CONFIG_NET_CLS_ACT
3677 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
3683 if (pfmemalloc
&& !skb_pfmemalloc_protocol(skb
))
3686 if (skb_vlan_tag_present(skb
)) {
3688 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3691 if (vlan_do_receive(&skb
))
3693 else if (unlikely(!skb
))
3697 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
3700 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3703 switch (rx_handler(&skb
)) {
3704 case RX_HANDLER_CONSUMED
:
3705 ret
= NET_RX_SUCCESS
;
3707 case RX_HANDLER_ANOTHER
:
3709 case RX_HANDLER_EXACT
:
3710 deliver_exact
= true;
3711 case RX_HANDLER_PASS
:
3718 if (unlikely(skb_vlan_tag_present(skb
))) {
3719 if (skb_vlan_tag_get_id(skb
))
3720 skb
->pkt_type
= PACKET_OTHERHOST
;
3721 /* Note: we might in the future use prio bits
3722 * and set skb->priority like in vlan_do_receive()
3723 * For the time being, just ignore Priority Code Point
3728 type
= skb
->protocol
;
3730 /* deliver only exact match when indicated */
3731 if (likely(!deliver_exact
)) {
3732 deliver_ptype_list_skb(skb
, &pt_prev
, orig_dev
, type
,
3733 &ptype_base
[ntohs(type
) &
3737 deliver_ptype_list_skb(skb
, &pt_prev
, orig_dev
, type
,
3738 &orig_dev
->ptype_specific
);
3740 if (unlikely(skb
->dev
!= orig_dev
)) {
3741 deliver_ptype_list_skb(skb
, &pt_prev
, orig_dev
, type
,
3742 &skb
->dev
->ptype_specific
);
3746 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
3749 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
3752 atomic_long_inc(&skb
->dev
->rx_dropped
);
3754 /* Jamal, now you will not able to escape explaining
3755 * me how you were going to use this. :-)
3765 static int __netif_receive_skb(struct sk_buff
*skb
)
3769 if (sk_memalloc_socks() && skb_pfmemalloc(skb
)) {
3770 unsigned long pflags
= current
->flags
;
3773 * PFMEMALLOC skbs are special, they should
3774 * - be delivered to SOCK_MEMALLOC sockets only
3775 * - stay away from userspace
3776 * - have bounded memory usage
3778 * Use PF_MEMALLOC as this saves us from propagating the allocation
3779 * context down to all allocation sites.
3781 current
->flags
|= PF_MEMALLOC
;
3782 ret
= __netif_receive_skb_core(skb
, true);
3783 tsk_restore_flags(current
, pflags
, PF_MEMALLOC
);
3785 ret
= __netif_receive_skb_core(skb
, false);
3790 static int netif_receive_skb_internal(struct sk_buff
*skb
)
3792 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3794 if (skb_defer_rx_timestamp(skb
))
3795 return NET_RX_SUCCESS
;
3798 if (static_key_false(&rps_needed
)) {
3799 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3804 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3807 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3814 return __netif_receive_skb(skb
);
3818 * netif_receive_skb - process receive buffer from network
3819 * @skb: buffer to process
3821 * netif_receive_skb() is the main receive data processing function.
3822 * It always succeeds. The buffer may be dropped during processing
3823 * for congestion control or by the protocol layers.
3825 * This function may only be called from softirq context and interrupts
3826 * should be enabled.
3828 * Return values (usually ignored):
3829 * NET_RX_SUCCESS: no congestion
3830 * NET_RX_DROP: packet was dropped
3832 int netif_receive_skb(struct sk_buff
*skb
)
3834 trace_netif_receive_skb_entry(skb
);
3836 return netif_receive_skb_internal(skb
);
3838 EXPORT_SYMBOL(netif_receive_skb
);
3840 /* Network device is going away, flush any packets still pending
3841 * Called with irqs disabled.
3843 static void flush_backlog(void *arg
)
3845 struct net_device
*dev
= arg
;
3846 struct softnet_data
*sd
= this_cpu_ptr(&softnet_data
);
3847 struct sk_buff
*skb
, *tmp
;
3850 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3851 if (skb
->dev
== dev
) {
3852 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3854 input_queue_head_incr(sd
);
3859 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3860 if (skb
->dev
== dev
) {
3861 __skb_unlink(skb
, &sd
->process_queue
);
3863 input_queue_head_incr(sd
);
3868 static int napi_gro_complete(struct sk_buff
*skb
)
3870 struct packet_offload
*ptype
;
3871 __be16 type
= skb
->protocol
;
3872 struct list_head
*head
= &offload_base
;
3875 BUILD_BUG_ON(sizeof(struct napi_gro_cb
) > sizeof(skb
->cb
));
3877 if (NAPI_GRO_CB(skb
)->count
== 1) {
3878 skb_shinfo(skb
)->gso_size
= 0;
3883 list_for_each_entry_rcu(ptype
, head
, list
) {
3884 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_complete
)
3887 err
= ptype
->callbacks
.gro_complete(skb
, 0);
3893 WARN_ON(&ptype
->list
== head
);
3895 return NET_RX_SUCCESS
;
3899 return netif_receive_skb_internal(skb
);
3902 /* napi->gro_list contains packets ordered by age.
3903 * youngest packets at the head of it.
3904 * Complete skbs in reverse order to reduce latencies.
3906 void napi_gro_flush(struct napi_struct
*napi
, bool flush_old
)
3908 struct sk_buff
*skb
, *prev
= NULL
;
3910 /* scan list and build reverse chain */
3911 for (skb
= napi
->gro_list
; skb
!= NULL
; skb
= skb
->next
) {
3916 for (skb
= prev
; skb
; skb
= prev
) {
3919 if (flush_old
&& NAPI_GRO_CB(skb
)->age
== jiffies
)
3923 napi_gro_complete(skb
);
3927 napi
->gro_list
= NULL
;
3929 EXPORT_SYMBOL(napi_gro_flush
);
3931 static void gro_list_prepare(struct napi_struct
*napi
, struct sk_buff
*skb
)
3934 unsigned int maclen
= skb
->dev
->hard_header_len
;
3935 u32 hash
= skb_get_hash_raw(skb
);
3937 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3938 unsigned long diffs
;
3940 NAPI_GRO_CB(p
)->flush
= 0;
3942 if (hash
!= skb_get_hash_raw(p
)) {
3943 NAPI_GRO_CB(p
)->same_flow
= 0;
3947 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
3948 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
3949 if (maclen
== ETH_HLEN
)
3950 diffs
|= compare_ether_header(skb_mac_header(p
),
3951 skb_mac_header(skb
));
3953 diffs
= memcmp(skb_mac_header(p
),
3954 skb_mac_header(skb
),
3956 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
3960 static void skb_gro_reset_offset(struct sk_buff
*skb
)
3962 const struct skb_shared_info
*pinfo
= skb_shinfo(skb
);
3963 const skb_frag_t
*frag0
= &pinfo
->frags
[0];
3965 NAPI_GRO_CB(skb
)->data_offset
= 0;
3966 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3967 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3969 if (skb_mac_header(skb
) == skb_tail_pointer(skb
) &&
3971 !PageHighMem(skb_frag_page(frag0
))) {
3972 NAPI_GRO_CB(skb
)->frag0
= skb_frag_address(frag0
);
3973 NAPI_GRO_CB(skb
)->frag0_len
= skb_frag_size(frag0
);
3977 static void gro_pull_from_frag0(struct sk_buff
*skb
, int grow
)
3979 struct skb_shared_info
*pinfo
= skb_shinfo(skb
);
3981 BUG_ON(skb
->end
- skb
->tail
< grow
);
3983 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3985 skb
->data_len
-= grow
;
3988 pinfo
->frags
[0].page_offset
+= grow
;
3989 skb_frag_size_sub(&pinfo
->frags
[0], grow
);
3991 if (unlikely(!skb_frag_size(&pinfo
->frags
[0]))) {
3992 skb_frag_unref(skb
, 0);
3993 memmove(pinfo
->frags
, pinfo
->frags
+ 1,
3994 --pinfo
->nr_frags
* sizeof(pinfo
->frags
[0]));
3998 static enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
4000 struct sk_buff
**pp
= NULL
;
4001 struct packet_offload
*ptype
;
4002 __be16 type
= skb
->protocol
;
4003 struct list_head
*head
= &offload_base
;
4005 enum gro_result ret
;
4008 if (!(skb
->dev
->features
& NETIF_F_GRO
))
4011 if (skb_is_gso(skb
) || skb_has_frag_list(skb
) || skb
->csum_bad
)
4014 gro_list_prepare(napi
, skb
);
4017 list_for_each_entry_rcu(ptype
, head
, list
) {
4018 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_receive
)
4021 skb_set_network_header(skb
, skb_gro_offset(skb
));
4022 skb_reset_mac_len(skb
);
4023 NAPI_GRO_CB(skb
)->same_flow
= 0;
4024 NAPI_GRO_CB(skb
)->flush
= 0;
4025 NAPI_GRO_CB(skb
)->free
= 0;
4026 NAPI_GRO_CB(skb
)->udp_mark
= 0;
4027 NAPI_GRO_CB(skb
)->gro_remcsum_start
= 0;
4029 /* Setup for GRO checksum validation */
4030 switch (skb
->ip_summed
) {
4031 case CHECKSUM_COMPLETE
:
4032 NAPI_GRO_CB(skb
)->csum
= skb
->csum
;
4033 NAPI_GRO_CB(skb
)->csum_valid
= 1;
4034 NAPI_GRO_CB(skb
)->csum_cnt
= 0;
4036 case CHECKSUM_UNNECESSARY
:
4037 NAPI_GRO_CB(skb
)->csum_cnt
= skb
->csum_level
+ 1;
4038 NAPI_GRO_CB(skb
)->csum_valid
= 0;
4041 NAPI_GRO_CB(skb
)->csum_cnt
= 0;
4042 NAPI_GRO_CB(skb
)->csum_valid
= 0;
4045 pp
= ptype
->callbacks
.gro_receive(&napi
->gro_list
, skb
);
4050 if (&ptype
->list
== head
)
4053 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
4054 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
4057 struct sk_buff
*nskb
= *pp
;
4061 napi_gro_complete(nskb
);
4068 if (NAPI_GRO_CB(skb
)->flush
)
4071 if (unlikely(napi
->gro_count
>= MAX_GRO_SKBS
)) {
4072 struct sk_buff
*nskb
= napi
->gro_list
;
4074 /* locate the end of the list to select the 'oldest' flow */
4075 while (nskb
->next
) {
4081 napi_gro_complete(nskb
);
4085 NAPI_GRO_CB(skb
)->count
= 1;
4086 NAPI_GRO_CB(skb
)->age
= jiffies
;
4087 NAPI_GRO_CB(skb
)->last
= skb
;
4088 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
4089 skb
->next
= napi
->gro_list
;
4090 napi
->gro_list
= skb
;
4094 grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
4096 gro_pull_from_frag0(skb
, grow
);
4105 struct packet_offload
*gro_find_receive_by_type(__be16 type
)
4107 struct list_head
*offload_head
= &offload_base
;
4108 struct packet_offload
*ptype
;
4110 list_for_each_entry_rcu(ptype
, offload_head
, list
) {
4111 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_receive
)
4117 EXPORT_SYMBOL(gro_find_receive_by_type
);
4119 struct packet_offload
*gro_find_complete_by_type(__be16 type
)
4121 struct list_head
*offload_head
= &offload_base
;
4122 struct packet_offload
*ptype
;
4124 list_for_each_entry_rcu(ptype
, offload_head
, list
) {
4125 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_complete
)
4131 EXPORT_SYMBOL(gro_find_complete_by_type
);
4133 static gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
4137 if (netif_receive_skb_internal(skb
))
4145 case GRO_MERGED_FREE
:
4146 if (NAPI_GRO_CB(skb
)->free
== NAPI_GRO_FREE_STOLEN_HEAD
)
4147 kmem_cache_free(skbuff_head_cache
, skb
);
4160 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
4162 trace_napi_gro_receive_entry(skb
);
4164 skb_gro_reset_offset(skb
);
4166 return napi_skb_finish(dev_gro_receive(napi
, skb
), skb
);
4168 EXPORT_SYMBOL(napi_gro_receive
);
4170 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
4172 if (unlikely(skb
->pfmemalloc
)) {
4176 __skb_pull(skb
, skb_headlen(skb
));
4177 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4178 skb_reserve(skb
, NET_SKB_PAD
+ NET_IP_ALIGN
- skb_headroom(skb
));
4180 skb
->dev
= napi
->dev
;
4182 skb
->encapsulation
= 0;
4183 skb_shinfo(skb
)->gso_type
= 0;
4184 skb
->truesize
= SKB_TRUESIZE(skb_end_offset(skb
));
4189 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
4191 struct sk_buff
*skb
= napi
->skb
;
4194 skb
= napi_alloc_skb(napi
, GRO_MAX_HEAD
);
4199 EXPORT_SYMBOL(napi_get_frags
);
4201 static gro_result_t
napi_frags_finish(struct napi_struct
*napi
,
4202 struct sk_buff
*skb
,
4208 __skb_push(skb
, ETH_HLEN
);
4209 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
4210 if (ret
== GRO_NORMAL
&& netif_receive_skb_internal(skb
))
4215 case GRO_MERGED_FREE
:
4216 napi_reuse_skb(napi
, skb
);
4226 /* Upper GRO stack assumes network header starts at gro_offset=0
4227 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4228 * We copy ethernet header into skb->data to have a common layout.
4230 static struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
4232 struct sk_buff
*skb
= napi
->skb
;
4233 const struct ethhdr
*eth
;
4234 unsigned int hlen
= sizeof(*eth
);
4238 skb_reset_mac_header(skb
);
4239 skb_gro_reset_offset(skb
);
4241 eth
= skb_gro_header_fast(skb
, 0);
4242 if (unlikely(skb_gro_header_hard(skb
, hlen
))) {
4243 eth
= skb_gro_header_slow(skb
, hlen
, 0);
4244 if (unlikely(!eth
)) {
4245 napi_reuse_skb(napi
, skb
);
4249 gro_pull_from_frag0(skb
, hlen
);
4250 NAPI_GRO_CB(skb
)->frag0
+= hlen
;
4251 NAPI_GRO_CB(skb
)->frag0_len
-= hlen
;
4253 __skb_pull(skb
, hlen
);
4256 * This works because the only protocols we care about don't require
4258 * We'll fix it up properly in napi_frags_finish()
4260 skb
->protocol
= eth
->h_proto
;
4265 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
4267 struct sk_buff
*skb
= napi_frags_skb(napi
);
4272 trace_napi_gro_frags_entry(skb
);
4274 return napi_frags_finish(napi
, skb
, dev_gro_receive(napi
, skb
));
4276 EXPORT_SYMBOL(napi_gro_frags
);
4278 /* Compute the checksum from gro_offset and return the folded value
4279 * after adding in any pseudo checksum.
4281 __sum16
__skb_gro_checksum_complete(struct sk_buff
*skb
)
4286 wsum
= skb_checksum(skb
, skb_gro_offset(skb
), skb_gro_len(skb
), 0);
4288 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4289 sum
= csum_fold(csum_add(NAPI_GRO_CB(skb
)->csum
, wsum
));
4291 if (unlikely(skb
->ip_summed
== CHECKSUM_COMPLETE
) &&
4292 !skb
->csum_complete_sw
)
4293 netdev_rx_csum_fault(skb
->dev
);
4296 NAPI_GRO_CB(skb
)->csum
= wsum
;
4297 NAPI_GRO_CB(skb
)->csum_valid
= 1;
4301 EXPORT_SYMBOL(__skb_gro_checksum_complete
);
4304 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4305 * Note: called with local irq disabled, but exits with local irq enabled.
4307 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
4310 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
4313 sd
->rps_ipi_list
= NULL
;
4317 /* Send pending IPI's to kick RPS processing on remote cpus. */
4319 struct softnet_data
*next
= remsd
->rps_ipi_next
;
4321 if (cpu_online(remsd
->cpu
))
4322 smp_call_function_single_async(remsd
->cpu
,
4331 static bool sd_has_rps_ipi_waiting(struct softnet_data
*sd
)
4334 return sd
->rps_ipi_list
!= NULL
;
4340 static int process_backlog(struct napi_struct
*napi
, int quota
)
4343 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
4345 /* Check if we have pending ipi, its better to send them now,
4346 * not waiting net_rx_action() end.
4348 if (sd_has_rps_ipi_waiting(sd
)) {
4349 local_irq_disable();
4350 net_rps_action_and_irq_enable(sd
);
4353 napi
->weight
= weight_p
;
4354 local_irq_disable();
4356 struct sk_buff
*skb
;
4358 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
4360 __netif_receive_skb(skb
);
4361 local_irq_disable();
4362 input_queue_head_incr(sd
);
4363 if (++work
>= quota
) {
4370 if (skb_queue_empty(&sd
->input_pkt_queue
)) {
4372 * Inline a custom version of __napi_complete().
4373 * only current cpu owns and manipulates this napi,
4374 * and NAPI_STATE_SCHED is the only possible flag set
4376 * We can use a plain write instead of clear_bit(),
4377 * and we dont need an smp_mb() memory barrier.
4385 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
4386 &sd
->process_queue
);
4395 * __napi_schedule - schedule for receive
4396 * @n: entry to schedule
4398 * The entry's receive function will be scheduled to run.
4399 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
4401 void __napi_schedule(struct napi_struct
*n
)
4403 unsigned long flags
;
4405 local_irq_save(flags
);
4406 ____napi_schedule(this_cpu_ptr(&softnet_data
), n
);
4407 local_irq_restore(flags
);
4409 EXPORT_SYMBOL(__napi_schedule
);
4412 * __napi_schedule_irqoff - schedule for receive
4413 * @n: entry to schedule
4415 * Variant of __napi_schedule() assuming hard irqs are masked
4417 void __napi_schedule_irqoff(struct napi_struct
*n
)
4419 ____napi_schedule(this_cpu_ptr(&softnet_data
), n
);
4421 EXPORT_SYMBOL(__napi_schedule_irqoff
);
4423 void __napi_complete(struct napi_struct
*n
)
4425 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
4427 list_del_init(&n
->poll_list
);
4428 smp_mb__before_atomic();
4429 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
4431 EXPORT_SYMBOL(__napi_complete
);
4433 void napi_complete_done(struct napi_struct
*n
, int work_done
)
4435 unsigned long flags
;
4438 * don't let napi dequeue from the cpu poll list
4439 * just in case its running on a different cpu
4441 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
4445 unsigned long timeout
= 0;
4448 timeout
= n
->dev
->gro_flush_timeout
;
4451 hrtimer_start(&n
->timer
, ns_to_ktime(timeout
),
4452 HRTIMER_MODE_REL_PINNED
);
4454 napi_gro_flush(n
, false);
4456 if (likely(list_empty(&n
->poll_list
))) {
4457 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED
, &n
->state
));
4459 /* If n->poll_list is not empty, we need to mask irqs */
4460 local_irq_save(flags
);
4462 local_irq_restore(flags
);
4465 EXPORT_SYMBOL(napi_complete_done
);
4467 /* must be called under rcu_read_lock(), as we dont take a reference */
4468 struct napi_struct
*napi_by_id(unsigned int napi_id
)
4470 unsigned int hash
= napi_id
% HASH_SIZE(napi_hash
);
4471 struct napi_struct
*napi
;
4473 hlist_for_each_entry_rcu(napi
, &napi_hash
[hash
], napi_hash_node
)
4474 if (napi
->napi_id
== napi_id
)
4479 EXPORT_SYMBOL_GPL(napi_by_id
);
4481 void napi_hash_add(struct napi_struct
*napi
)
4483 if (!test_and_set_bit(NAPI_STATE_HASHED
, &napi
->state
)) {
4485 spin_lock(&napi_hash_lock
);
4487 /* 0 is not a valid id, we also skip an id that is taken
4488 * we expect both events to be extremely rare
4491 while (!napi
->napi_id
) {
4492 napi
->napi_id
= ++napi_gen_id
;
4493 if (napi_by_id(napi
->napi_id
))
4497 hlist_add_head_rcu(&napi
->napi_hash_node
,
4498 &napi_hash
[napi
->napi_id
% HASH_SIZE(napi_hash
)]);
4500 spin_unlock(&napi_hash_lock
);
4503 EXPORT_SYMBOL_GPL(napi_hash_add
);
4505 /* Warning : caller is responsible to make sure rcu grace period
4506 * is respected before freeing memory containing @napi
4508 void napi_hash_del(struct napi_struct
*napi
)
4510 spin_lock(&napi_hash_lock
);
4512 if (test_and_clear_bit(NAPI_STATE_HASHED
, &napi
->state
))
4513 hlist_del_rcu(&napi
->napi_hash_node
);
4515 spin_unlock(&napi_hash_lock
);
4517 EXPORT_SYMBOL_GPL(napi_hash_del
);
4519 static enum hrtimer_restart
napi_watchdog(struct hrtimer
*timer
)
4521 struct napi_struct
*napi
;
4523 napi
= container_of(timer
, struct napi_struct
, timer
);
4525 napi_schedule(napi
);
4527 return HRTIMER_NORESTART
;
4530 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
4531 int (*poll
)(struct napi_struct
*, int), int weight
)
4533 INIT_LIST_HEAD(&napi
->poll_list
);
4534 hrtimer_init(&napi
->timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_REL_PINNED
);
4535 napi
->timer
.function
= napi_watchdog
;
4536 napi
->gro_count
= 0;
4537 napi
->gro_list
= NULL
;
4540 if (weight
> NAPI_POLL_WEIGHT
)
4541 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4543 napi
->weight
= weight
;
4544 list_add(&napi
->dev_list
, &dev
->napi_list
);
4546 #ifdef CONFIG_NETPOLL
4547 spin_lock_init(&napi
->poll_lock
);
4548 napi
->poll_owner
= -1;
4550 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
4552 EXPORT_SYMBOL(netif_napi_add
);
4554 void napi_disable(struct napi_struct
*n
)
4557 set_bit(NAPI_STATE_DISABLE
, &n
->state
);
4559 while (test_and_set_bit(NAPI_STATE_SCHED
, &n
->state
))
4562 hrtimer_cancel(&n
->timer
);
4564 clear_bit(NAPI_STATE_DISABLE
, &n
->state
);
4566 EXPORT_SYMBOL(napi_disable
);
4568 void netif_napi_del(struct napi_struct
*napi
)
4570 list_del_init(&napi
->dev_list
);
4571 napi_free_frags(napi
);
4573 kfree_skb_list(napi
->gro_list
);
4574 napi
->gro_list
= NULL
;
4575 napi
->gro_count
= 0;
4577 EXPORT_SYMBOL(netif_napi_del
);
4579 static int napi_poll(struct napi_struct
*n
, struct list_head
*repoll
)
4584 list_del_init(&n
->poll_list
);
4586 have
= netpoll_poll_lock(n
);
4590 /* This NAPI_STATE_SCHED test is for avoiding a race
4591 * with netpoll's poll_napi(). Only the entity which
4592 * obtains the lock and sees NAPI_STATE_SCHED set will
4593 * actually make the ->poll() call. Therefore we avoid
4594 * accidentally calling ->poll() when NAPI is not scheduled.
4597 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
4598 work
= n
->poll(n
, weight
);
4602 WARN_ON_ONCE(work
> weight
);
4604 if (likely(work
< weight
))
4607 /* Drivers must not modify the NAPI state if they
4608 * consume the entire weight. In such cases this code
4609 * still "owns" the NAPI instance and therefore can
4610 * move the instance around on the list at-will.
4612 if (unlikely(napi_disable_pending(n
))) {
4618 /* flush too old packets
4619 * If HZ < 1000, flush all packets.
4621 napi_gro_flush(n
, HZ
>= 1000);
4624 /* Some drivers may have called napi_schedule
4625 * prior to exhausting their budget.
4627 if (unlikely(!list_empty(&n
->poll_list
))) {
4628 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
4629 n
->dev
? n
->dev
->name
: "backlog");
4633 list_add_tail(&n
->poll_list
, repoll
);
4636 netpoll_poll_unlock(have
);
4641 static void net_rx_action(struct softirq_action
*h
)
4643 struct softnet_data
*sd
= this_cpu_ptr(&softnet_data
);
4644 unsigned long time_limit
= jiffies
+ 2;
4645 int budget
= netdev_budget
;
4649 local_irq_disable();
4650 list_splice_init(&sd
->poll_list
, &list
);
4654 struct napi_struct
*n
;
4656 if (list_empty(&list
)) {
4657 if (!sd_has_rps_ipi_waiting(sd
) && list_empty(&repoll
))
4662 n
= list_first_entry(&list
, struct napi_struct
, poll_list
);
4663 budget
-= napi_poll(n
, &repoll
);
4665 /* If softirq window is exhausted then punt.
4666 * Allow this to run for 2 jiffies since which will allow
4667 * an average latency of 1.5/HZ.
4669 if (unlikely(budget
<= 0 ||
4670 time_after_eq(jiffies
, time_limit
))) {
4676 local_irq_disable();
4678 list_splice_tail_init(&sd
->poll_list
, &list
);
4679 list_splice_tail(&repoll
, &list
);
4680 list_splice(&list
, &sd
->poll_list
);
4681 if (!list_empty(&sd
->poll_list
))
4682 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
4684 net_rps_action_and_irq_enable(sd
);
4687 struct netdev_adjacent
{
4688 struct net_device
*dev
;
4690 /* upper master flag, there can only be one master device per list */
4693 /* counter for the number of times this device was added to us */
4696 /* private field for the users */
4699 struct list_head list
;
4700 struct rcu_head rcu
;
4703 static struct netdev_adjacent
*__netdev_find_adj(struct net_device
*dev
,
4704 struct net_device
*adj_dev
,
4705 struct list_head
*adj_list
)
4707 struct netdev_adjacent
*adj
;
4709 list_for_each_entry(adj
, adj_list
, list
) {
4710 if (adj
->dev
== adj_dev
)
4717 * netdev_has_upper_dev - Check if device is linked to an upper device
4719 * @upper_dev: upper device to check
4721 * Find out if a device is linked to specified upper device and return true
4722 * in case it is. Note that this checks only immediate upper device,
4723 * not through a complete stack of devices. The caller must hold the RTNL lock.
4725 bool netdev_has_upper_dev(struct net_device
*dev
,
4726 struct net_device
*upper_dev
)
4730 return __netdev_find_adj(dev
, upper_dev
, &dev
->all_adj_list
.upper
);
4732 EXPORT_SYMBOL(netdev_has_upper_dev
);
4735 * netdev_has_any_upper_dev - Check if device is linked to some device
4738 * Find out if a device is linked to an upper device and return true in case
4739 * it is. The caller must hold the RTNL lock.
4741 static bool netdev_has_any_upper_dev(struct net_device
*dev
)
4745 return !list_empty(&dev
->all_adj_list
.upper
);
4749 * netdev_master_upper_dev_get - Get master upper device
4752 * Find a master upper device and return pointer to it or NULL in case
4753 * it's not there. The caller must hold the RTNL lock.
4755 struct net_device
*netdev_master_upper_dev_get(struct net_device
*dev
)
4757 struct netdev_adjacent
*upper
;
4761 if (list_empty(&dev
->adj_list
.upper
))
4764 upper
= list_first_entry(&dev
->adj_list
.upper
,
4765 struct netdev_adjacent
, list
);
4766 if (likely(upper
->master
))
4770 EXPORT_SYMBOL(netdev_master_upper_dev_get
);
4772 void *netdev_adjacent_get_private(struct list_head
*adj_list
)
4774 struct netdev_adjacent
*adj
;
4776 adj
= list_entry(adj_list
, struct netdev_adjacent
, list
);
4778 return adj
->private;
4780 EXPORT_SYMBOL(netdev_adjacent_get_private
);
4783 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
4785 * @iter: list_head ** of the current position
4787 * Gets the next device from the dev's upper list, starting from iter
4788 * position. The caller must hold RCU read lock.
4790 struct net_device
*netdev_upper_get_next_dev_rcu(struct net_device
*dev
,
4791 struct list_head
**iter
)
4793 struct netdev_adjacent
*upper
;
4795 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4797 upper
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
4799 if (&upper
->list
== &dev
->adj_list
.upper
)
4802 *iter
= &upper
->list
;
4806 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu
);
4809 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4811 * @iter: list_head ** of the current position
4813 * Gets the next device from the dev's upper list, starting from iter
4814 * position. The caller must hold RCU read lock.
4816 struct net_device
*netdev_all_upper_get_next_dev_rcu(struct net_device
*dev
,
4817 struct list_head
**iter
)
4819 struct netdev_adjacent
*upper
;
4821 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4823 upper
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
4825 if (&upper
->list
== &dev
->all_adj_list
.upper
)
4828 *iter
= &upper
->list
;
4832 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu
);
4835 * netdev_lower_get_next_private - Get the next ->private from the
4836 * lower neighbour list
4838 * @iter: list_head ** of the current position
4840 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4841 * list, starting from iter position. The caller must hold either hold the
4842 * RTNL lock or its own locking that guarantees that the neighbour lower
4843 * list will remain unchainged.
4845 void *netdev_lower_get_next_private(struct net_device
*dev
,
4846 struct list_head
**iter
)
4848 struct netdev_adjacent
*lower
;
4850 lower
= list_entry(*iter
, struct netdev_adjacent
, list
);
4852 if (&lower
->list
== &dev
->adj_list
.lower
)
4855 *iter
= lower
->list
.next
;
4857 return lower
->private;
4859 EXPORT_SYMBOL(netdev_lower_get_next_private
);
4862 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4863 * lower neighbour list, RCU
4866 * @iter: list_head ** of the current position
4868 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4869 * list, starting from iter position. The caller must hold RCU read lock.
4871 void *netdev_lower_get_next_private_rcu(struct net_device
*dev
,
4872 struct list_head
**iter
)
4874 struct netdev_adjacent
*lower
;
4876 WARN_ON_ONCE(!rcu_read_lock_held());
4878 lower
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
4880 if (&lower
->list
== &dev
->adj_list
.lower
)
4883 *iter
= &lower
->list
;
4885 return lower
->private;
4887 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu
);
4890 * netdev_lower_get_next - Get the next device from the lower neighbour
4893 * @iter: list_head ** of the current position
4895 * Gets the next netdev_adjacent from the dev's lower neighbour
4896 * list, starting from iter position. The caller must hold RTNL lock or
4897 * its own locking that guarantees that the neighbour lower
4898 * list will remain unchainged.
4900 void *netdev_lower_get_next(struct net_device
*dev
, struct list_head
**iter
)
4902 struct netdev_adjacent
*lower
;
4904 lower
= list_entry((*iter
)->next
, struct netdev_adjacent
, list
);
4906 if (&lower
->list
== &dev
->adj_list
.lower
)
4909 *iter
= &lower
->list
;
4913 EXPORT_SYMBOL(netdev_lower_get_next
);
4916 * netdev_lower_get_first_private_rcu - Get the first ->private from the
4917 * lower neighbour list, RCU
4921 * Gets the first netdev_adjacent->private from the dev's lower neighbour
4922 * list. The caller must hold RCU read lock.
4924 void *netdev_lower_get_first_private_rcu(struct net_device
*dev
)
4926 struct netdev_adjacent
*lower
;
4928 lower
= list_first_or_null_rcu(&dev
->adj_list
.lower
,
4929 struct netdev_adjacent
, list
);
4931 return lower
->private;
4934 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu
);
4937 * netdev_master_upper_dev_get_rcu - Get master upper device
4940 * Find a master upper device and return pointer to it or NULL in case
4941 * it's not there. The caller must hold the RCU read lock.
4943 struct net_device
*netdev_master_upper_dev_get_rcu(struct net_device
*dev
)
4945 struct netdev_adjacent
*upper
;
4947 upper
= list_first_or_null_rcu(&dev
->adj_list
.upper
,
4948 struct netdev_adjacent
, list
);
4949 if (upper
&& likely(upper
->master
))
4953 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu
);
4955 static int netdev_adjacent_sysfs_add(struct net_device
*dev
,
4956 struct net_device
*adj_dev
,
4957 struct list_head
*dev_list
)
4959 char linkname
[IFNAMSIZ
+7];
4960 sprintf(linkname
, dev_list
== &dev
->adj_list
.upper
?
4961 "upper_%s" : "lower_%s", adj_dev
->name
);
4962 return sysfs_create_link(&(dev
->dev
.kobj
), &(adj_dev
->dev
.kobj
),
4965 static void netdev_adjacent_sysfs_del(struct net_device
*dev
,
4967 struct list_head
*dev_list
)
4969 char linkname
[IFNAMSIZ
+7];
4970 sprintf(linkname
, dev_list
== &dev
->adj_list
.upper
?
4971 "upper_%s" : "lower_%s", name
);
4972 sysfs_remove_link(&(dev
->dev
.kobj
), linkname
);
4975 static inline bool netdev_adjacent_is_neigh_list(struct net_device
*dev
,
4976 struct net_device
*adj_dev
,
4977 struct list_head
*dev_list
)
4979 return (dev_list
== &dev
->adj_list
.upper
||
4980 dev_list
== &dev
->adj_list
.lower
) &&
4981 net_eq(dev_net(dev
), dev_net(adj_dev
));
4984 static int __netdev_adjacent_dev_insert(struct net_device
*dev
,
4985 struct net_device
*adj_dev
,
4986 struct list_head
*dev_list
,
4987 void *private, bool master
)
4989 struct netdev_adjacent
*adj
;
4992 adj
= __netdev_find_adj(dev
, adj_dev
, dev_list
);
4999 adj
= kmalloc(sizeof(*adj
), GFP_KERNEL
);
5004 adj
->master
= master
;
5006 adj
->private = private;
5009 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
5010 adj_dev
->name
, dev
->name
, adj_dev
->name
);
5012 if (netdev_adjacent_is_neigh_list(dev
, adj_dev
, dev_list
)) {
5013 ret
= netdev_adjacent_sysfs_add(dev
, adj_dev
, dev_list
);
5018 /* Ensure that master link is always the first item in list. */
5020 ret
= sysfs_create_link(&(dev
->dev
.kobj
),
5021 &(adj_dev
->dev
.kobj
), "master");
5023 goto remove_symlinks
;
5025 list_add_rcu(&adj
->list
, dev_list
);
5027 list_add_tail_rcu(&adj
->list
, dev_list
);
5033 if (netdev_adjacent_is_neigh_list(dev
, adj_dev
, dev_list
))
5034 netdev_adjacent_sysfs_del(dev
, adj_dev
->name
, dev_list
);
5042 static void __netdev_adjacent_dev_remove(struct net_device
*dev
,
5043 struct net_device
*adj_dev
,
5044 struct list_head
*dev_list
)
5046 struct netdev_adjacent
*adj
;
5048 adj
= __netdev_find_adj(dev
, adj_dev
, dev_list
);
5051 pr_err("tried to remove device %s from %s\n",
5052 dev
->name
, adj_dev
->name
);
5056 if (adj
->ref_nr
> 1) {
5057 pr_debug("%s to %s ref_nr-- = %d\n", dev
->name
, adj_dev
->name
,
5064 sysfs_remove_link(&(dev
->dev
.kobj
), "master");
5066 if (netdev_adjacent_is_neigh_list(dev
, adj_dev
, dev_list
))
5067 netdev_adjacent_sysfs_del(dev
, adj_dev
->name
, dev_list
);
5069 list_del_rcu(&adj
->list
);
5070 pr_debug("dev_put for %s, because link removed from %s to %s\n",
5071 adj_dev
->name
, dev
->name
, adj_dev
->name
);
5073 kfree_rcu(adj
, rcu
);
5076 static int __netdev_adjacent_dev_link_lists(struct net_device
*dev
,
5077 struct net_device
*upper_dev
,
5078 struct list_head
*up_list
,
5079 struct list_head
*down_list
,
5080 void *private, bool master
)
5084 ret
= __netdev_adjacent_dev_insert(dev
, upper_dev
, up_list
, private,
5089 ret
= __netdev_adjacent_dev_insert(upper_dev
, dev
, down_list
, private,
5092 __netdev_adjacent_dev_remove(dev
, upper_dev
, up_list
);
5099 static int __netdev_adjacent_dev_link(struct net_device
*dev
,
5100 struct net_device
*upper_dev
)
5102 return __netdev_adjacent_dev_link_lists(dev
, upper_dev
,
5103 &dev
->all_adj_list
.upper
,
5104 &upper_dev
->all_adj_list
.lower
,
5108 static void __netdev_adjacent_dev_unlink_lists(struct net_device
*dev
,
5109 struct net_device
*upper_dev
,
5110 struct list_head
*up_list
,
5111 struct list_head
*down_list
)
5113 __netdev_adjacent_dev_remove(dev
, upper_dev
, up_list
);
5114 __netdev_adjacent_dev_remove(upper_dev
, dev
, down_list
);
5117 static void __netdev_adjacent_dev_unlink(struct net_device
*dev
,
5118 struct net_device
*upper_dev
)
5120 __netdev_adjacent_dev_unlink_lists(dev
, upper_dev
,
5121 &dev
->all_adj_list
.upper
,
5122 &upper_dev
->all_adj_list
.lower
);
5125 static int __netdev_adjacent_dev_link_neighbour(struct net_device
*dev
,
5126 struct net_device
*upper_dev
,
5127 void *private, bool master
)
5129 int ret
= __netdev_adjacent_dev_link(dev
, upper_dev
);
5134 ret
= __netdev_adjacent_dev_link_lists(dev
, upper_dev
,
5135 &dev
->adj_list
.upper
,
5136 &upper_dev
->adj_list
.lower
,
5139 __netdev_adjacent_dev_unlink(dev
, upper_dev
);
5146 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device
*dev
,
5147 struct net_device
*upper_dev
)
5149 __netdev_adjacent_dev_unlink(dev
, upper_dev
);
5150 __netdev_adjacent_dev_unlink_lists(dev
, upper_dev
,
5151 &dev
->adj_list
.upper
,
5152 &upper_dev
->adj_list
.lower
);
5155 static int __netdev_upper_dev_link(struct net_device
*dev
,
5156 struct net_device
*upper_dev
, bool master
,
5159 struct netdev_adjacent
*i
, *j
, *to_i
, *to_j
;
5164 if (dev
== upper_dev
)
5167 /* To prevent loops, check if dev is not upper device to upper_dev. */
5168 if (__netdev_find_adj(upper_dev
, dev
, &upper_dev
->all_adj_list
.upper
))
5171 if (__netdev_find_adj(dev
, upper_dev
, &dev
->all_adj_list
.upper
))
5174 if (master
&& netdev_master_upper_dev_get(dev
))
5177 ret
= __netdev_adjacent_dev_link_neighbour(dev
, upper_dev
, private,
5182 /* Now that we linked these devs, make all the upper_dev's
5183 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5184 * versa, and don't forget the devices itself. All of these
5185 * links are non-neighbours.
5187 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5188 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
) {
5189 pr_debug("Interlinking %s with %s, non-neighbour\n",
5190 i
->dev
->name
, j
->dev
->name
);
5191 ret
= __netdev_adjacent_dev_link(i
->dev
, j
->dev
);
5197 /* add dev to every upper_dev's upper device */
5198 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
) {
5199 pr_debug("linking %s's upper device %s with %s\n",
5200 upper_dev
->name
, i
->dev
->name
, dev
->name
);
5201 ret
= __netdev_adjacent_dev_link(dev
, i
->dev
);
5203 goto rollback_upper_mesh
;
5206 /* add upper_dev to every dev's lower device */
5207 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5208 pr_debug("linking %s's lower device %s with %s\n", dev
->name
,
5209 i
->dev
->name
, upper_dev
->name
);
5210 ret
= __netdev_adjacent_dev_link(i
->dev
, upper_dev
);
5212 goto rollback_lower_mesh
;
5215 call_netdevice_notifiers(NETDEV_CHANGEUPPER
, dev
);
5218 rollback_lower_mesh
:
5220 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5223 __netdev_adjacent_dev_unlink(i
->dev
, upper_dev
);
5228 rollback_upper_mesh
:
5230 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
) {
5233 __netdev_adjacent_dev_unlink(dev
, i
->dev
);
5241 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5242 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
) {
5243 if (i
== to_i
&& j
== to_j
)
5245 __netdev_adjacent_dev_unlink(i
->dev
, j
->dev
);
5251 __netdev_adjacent_dev_unlink_neighbour(dev
, upper_dev
);
5257 * netdev_upper_dev_link - Add a link to the upper device
5259 * @upper_dev: new upper device
5261 * Adds a link to device which is upper to this one. The caller must hold
5262 * the RTNL lock. On a failure a negative errno code is returned.
5263 * On success the reference counts are adjusted and the function
5266 int netdev_upper_dev_link(struct net_device
*dev
,
5267 struct net_device
*upper_dev
)
5269 return __netdev_upper_dev_link(dev
, upper_dev
, false, NULL
);
5271 EXPORT_SYMBOL(netdev_upper_dev_link
);
5274 * netdev_master_upper_dev_link - Add a master link to the upper device
5276 * @upper_dev: new upper device
5278 * Adds a link to device which is upper to this one. In this case, only
5279 * one master upper device can be linked, although other non-master devices
5280 * might be linked as well. The caller must hold the RTNL lock.
5281 * On a failure a negative errno code is returned. On success the reference
5282 * counts are adjusted and the function returns zero.
5284 int netdev_master_upper_dev_link(struct net_device
*dev
,
5285 struct net_device
*upper_dev
)
5287 return __netdev_upper_dev_link(dev
, upper_dev
, true, NULL
);
5289 EXPORT_SYMBOL(netdev_master_upper_dev_link
);
5291 int netdev_master_upper_dev_link_private(struct net_device
*dev
,
5292 struct net_device
*upper_dev
,
5295 return __netdev_upper_dev_link(dev
, upper_dev
, true, private);
5297 EXPORT_SYMBOL(netdev_master_upper_dev_link_private
);
5300 * netdev_upper_dev_unlink - Removes a link to upper device
5302 * @upper_dev: new upper device
5304 * Removes a link to device which is upper to this one. The caller must hold
5307 void netdev_upper_dev_unlink(struct net_device
*dev
,
5308 struct net_device
*upper_dev
)
5310 struct netdev_adjacent
*i
, *j
;
5313 __netdev_adjacent_dev_unlink_neighbour(dev
, upper_dev
);
5315 /* Here is the tricky part. We must remove all dev's lower
5316 * devices from all upper_dev's upper devices and vice
5317 * versa, to maintain the graph relationship.
5319 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
)
5320 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
)
5321 __netdev_adjacent_dev_unlink(i
->dev
, j
->dev
);
5323 /* remove also the devices itself from lower/upper device
5326 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
)
5327 __netdev_adjacent_dev_unlink(i
->dev
, upper_dev
);
5329 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
)
5330 __netdev_adjacent_dev_unlink(dev
, i
->dev
);
5332 call_netdevice_notifiers(NETDEV_CHANGEUPPER
, dev
);
5334 EXPORT_SYMBOL(netdev_upper_dev_unlink
);
5337 * netdev_bonding_info_change - Dispatch event about slave change
5339 * @bonding_info: info to dispatch
5341 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
5342 * The caller must hold the RTNL lock.
5344 void netdev_bonding_info_change(struct net_device
*dev
,
5345 struct netdev_bonding_info
*bonding_info
)
5347 struct netdev_notifier_bonding_info info
;
5349 memcpy(&info
.bonding_info
, bonding_info
,
5350 sizeof(struct netdev_bonding_info
));
5351 call_netdevice_notifiers_info(NETDEV_BONDING_INFO
, dev
,
5354 EXPORT_SYMBOL(netdev_bonding_info_change
);
5356 static void netdev_adjacent_add_links(struct net_device
*dev
)
5358 struct netdev_adjacent
*iter
;
5360 struct net
*net
= dev_net(dev
);
5362 list_for_each_entry(iter
, &dev
->adj_list
.upper
, list
) {
5363 if (!net_eq(net
,dev_net(iter
->dev
)))
5365 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
5366 &iter
->dev
->adj_list
.lower
);
5367 netdev_adjacent_sysfs_add(dev
, iter
->dev
,
5368 &dev
->adj_list
.upper
);
5371 list_for_each_entry(iter
, &dev
->adj_list
.lower
, list
) {
5372 if (!net_eq(net
,dev_net(iter
->dev
)))
5374 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
5375 &iter
->dev
->adj_list
.upper
);
5376 netdev_adjacent_sysfs_add(dev
, iter
->dev
,
5377 &dev
->adj_list
.lower
);
5381 static void netdev_adjacent_del_links(struct net_device
*dev
)
5383 struct netdev_adjacent
*iter
;
5385 struct net
*net
= dev_net(dev
);
5387 list_for_each_entry(iter
, &dev
->adj_list
.upper
, list
) {
5388 if (!net_eq(net
,dev_net(iter
->dev
)))
5390 netdev_adjacent_sysfs_del(iter
->dev
, dev
->name
,
5391 &iter
->dev
->adj_list
.lower
);
5392 netdev_adjacent_sysfs_del(dev
, iter
->dev
->name
,
5393 &dev
->adj_list
.upper
);
5396 list_for_each_entry(iter
, &dev
->adj_list
.lower
, list
) {
5397 if (!net_eq(net
,dev_net(iter
->dev
)))
5399 netdev_adjacent_sysfs_del(iter
->dev
, dev
->name
,
5400 &iter
->dev
->adj_list
.upper
);
5401 netdev_adjacent_sysfs_del(dev
, iter
->dev
->name
,
5402 &dev
->adj_list
.lower
);
5406 void netdev_adjacent_rename_links(struct net_device
*dev
, char *oldname
)
5408 struct netdev_adjacent
*iter
;
5410 struct net
*net
= dev_net(dev
);
5412 list_for_each_entry(iter
, &dev
->adj_list
.upper
, list
) {
5413 if (!net_eq(net
,dev_net(iter
->dev
)))
5415 netdev_adjacent_sysfs_del(iter
->dev
, oldname
,
5416 &iter
->dev
->adj_list
.lower
);
5417 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
5418 &iter
->dev
->adj_list
.lower
);
5421 list_for_each_entry(iter
, &dev
->adj_list
.lower
, list
) {
5422 if (!net_eq(net
,dev_net(iter
->dev
)))
5424 netdev_adjacent_sysfs_del(iter
->dev
, oldname
,
5425 &iter
->dev
->adj_list
.upper
);
5426 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
5427 &iter
->dev
->adj_list
.upper
);
5431 void *netdev_lower_dev_get_private(struct net_device
*dev
,
5432 struct net_device
*lower_dev
)
5434 struct netdev_adjacent
*lower
;
5438 lower
= __netdev_find_adj(dev
, lower_dev
, &dev
->adj_list
.lower
);
5442 return lower
->private;
5444 EXPORT_SYMBOL(netdev_lower_dev_get_private
);
5447 int dev_get_nest_level(struct net_device
*dev
,
5448 bool (*type_check
)(struct net_device
*dev
))
5450 struct net_device
*lower
= NULL
;
5451 struct list_head
*iter
;
5457 netdev_for_each_lower_dev(dev
, lower
, iter
) {
5458 nest
= dev_get_nest_level(lower
, type_check
);
5459 if (max_nest
< nest
)
5463 if (type_check(dev
))
5468 EXPORT_SYMBOL(dev_get_nest_level
);
5470 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
5472 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5474 if (ops
->ndo_change_rx_flags
)
5475 ops
->ndo_change_rx_flags(dev
, flags
);
5478 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
, bool notify
)
5480 unsigned int old_flags
= dev
->flags
;
5486 dev
->flags
|= IFF_PROMISC
;
5487 dev
->promiscuity
+= inc
;
5488 if (dev
->promiscuity
== 0) {
5491 * If inc causes overflow, untouch promisc and return error.
5494 dev
->flags
&= ~IFF_PROMISC
;
5496 dev
->promiscuity
-= inc
;
5497 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5502 if (dev
->flags
!= old_flags
) {
5503 pr_info("device %s %s promiscuous mode\n",
5505 dev
->flags
& IFF_PROMISC
? "entered" : "left");
5506 if (audit_enabled
) {
5507 current_uid_gid(&uid
, &gid
);
5508 audit_log(current
->audit_context
, GFP_ATOMIC
,
5509 AUDIT_ANOM_PROMISCUOUS
,
5510 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5511 dev
->name
, (dev
->flags
& IFF_PROMISC
),
5512 (old_flags
& IFF_PROMISC
),
5513 from_kuid(&init_user_ns
, audit_get_loginuid(current
)),
5514 from_kuid(&init_user_ns
, uid
),
5515 from_kgid(&init_user_ns
, gid
),
5516 audit_get_sessionid(current
));
5519 dev_change_rx_flags(dev
, IFF_PROMISC
);
5522 __dev_notify_flags(dev
, old_flags
, IFF_PROMISC
);
5527 * dev_set_promiscuity - update promiscuity count on a device
5531 * Add or remove promiscuity from a device. While the count in the device
5532 * remains above zero the interface remains promiscuous. Once it hits zero
5533 * the device reverts back to normal filtering operation. A negative inc
5534 * value is used to drop promiscuity on the device.
5535 * Return 0 if successful or a negative errno code on error.
5537 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
5539 unsigned int old_flags
= dev
->flags
;
5542 err
= __dev_set_promiscuity(dev
, inc
, true);
5545 if (dev
->flags
!= old_flags
)
5546 dev_set_rx_mode(dev
);
5549 EXPORT_SYMBOL(dev_set_promiscuity
);
5551 static int __dev_set_allmulti(struct net_device
*dev
, int inc
, bool notify
)
5553 unsigned int old_flags
= dev
->flags
, old_gflags
= dev
->gflags
;
5557 dev
->flags
|= IFF_ALLMULTI
;
5558 dev
->allmulti
+= inc
;
5559 if (dev
->allmulti
== 0) {
5562 * If inc causes overflow, untouch allmulti and return error.
5565 dev
->flags
&= ~IFF_ALLMULTI
;
5567 dev
->allmulti
-= inc
;
5568 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5573 if (dev
->flags
^ old_flags
) {
5574 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
5575 dev_set_rx_mode(dev
);
5577 __dev_notify_flags(dev
, old_flags
,
5578 dev
->gflags
^ old_gflags
);
5584 * dev_set_allmulti - update allmulti count on a device
5588 * Add or remove reception of all multicast frames to a device. While the
5589 * count in the device remains above zero the interface remains listening
5590 * to all interfaces. Once it hits zero the device reverts back to normal
5591 * filtering operation. A negative @inc value is used to drop the counter
5592 * when releasing a resource needing all multicasts.
5593 * Return 0 if successful or a negative errno code on error.
5596 int dev_set_allmulti(struct net_device
*dev
, int inc
)
5598 return __dev_set_allmulti(dev
, inc
, true);
5600 EXPORT_SYMBOL(dev_set_allmulti
);
5603 * Upload unicast and multicast address lists to device and
5604 * configure RX filtering. When the device doesn't support unicast
5605 * filtering it is put in promiscuous mode while unicast addresses
5608 void __dev_set_rx_mode(struct net_device
*dev
)
5610 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5612 /* dev_open will call this function so the list will stay sane. */
5613 if (!(dev
->flags
&IFF_UP
))
5616 if (!netif_device_present(dev
))
5619 if (!(dev
->priv_flags
& IFF_UNICAST_FLT
)) {
5620 /* Unicast addresses changes may only happen under the rtnl,
5621 * therefore calling __dev_set_promiscuity here is safe.
5623 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
5624 __dev_set_promiscuity(dev
, 1, false);
5625 dev
->uc_promisc
= true;
5626 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
5627 __dev_set_promiscuity(dev
, -1, false);
5628 dev
->uc_promisc
= false;
5632 if (ops
->ndo_set_rx_mode
)
5633 ops
->ndo_set_rx_mode(dev
);
5636 void dev_set_rx_mode(struct net_device
*dev
)
5638 netif_addr_lock_bh(dev
);
5639 __dev_set_rx_mode(dev
);
5640 netif_addr_unlock_bh(dev
);
5644 * dev_get_flags - get flags reported to userspace
5647 * Get the combination of flag bits exported through APIs to userspace.
5649 unsigned int dev_get_flags(const struct net_device
*dev
)
5653 flags
= (dev
->flags
& ~(IFF_PROMISC
|
5658 (dev
->gflags
& (IFF_PROMISC
|
5661 if (netif_running(dev
)) {
5662 if (netif_oper_up(dev
))
5663 flags
|= IFF_RUNNING
;
5664 if (netif_carrier_ok(dev
))
5665 flags
|= IFF_LOWER_UP
;
5666 if (netif_dormant(dev
))
5667 flags
|= IFF_DORMANT
;
5672 EXPORT_SYMBOL(dev_get_flags
);
5674 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
5676 unsigned int old_flags
= dev
->flags
;
5682 * Set the flags on our device.
5685 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
5686 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
5688 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
5692 * Load in the correct multicast list now the flags have changed.
5695 if ((old_flags
^ flags
) & IFF_MULTICAST
)
5696 dev_change_rx_flags(dev
, IFF_MULTICAST
);
5698 dev_set_rx_mode(dev
);
5701 * Have we downed the interface. We handle IFF_UP ourselves
5702 * according to user attempts to set it, rather than blindly
5707 if ((old_flags
^ flags
) & IFF_UP
)
5708 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
5710 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
5711 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
5712 unsigned int old_flags
= dev
->flags
;
5714 dev
->gflags
^= IFF_PROMISC
;
5716 if (__dev_set_promiscuity(dev
, inc
, false) >= 0)
5717 if (dev
->flags
!= old_flags
)
5718 dev_set_rx_mode(dev
);
5721 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5722 is important. Some (broken) drivers set IFF_PROMISC, when
5723 IFF_ALLMULTI is requested not asking us and not reporting.
5725 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
5726 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
5728 dev
->gflags
^= IFF_ALLMULTI
;
5729 __dev_set_allmulti(dev
, inc
, false);
5735 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
,
5736 unsigned int gchanges
)
5738 unsigned int changes
= dev
->flags
^ old_flags
;
5741 rtmsg_ifinfo(RTM_NEWLINK
, dev
, gchanges
, GFP_ATOMIC
);
5743 if (changes
& IFF_UP
) {
5744 if (dev
->flags
& IFF_UP
)
5745 call_netdevice_notifiers(NETDEV_UP
, dev
);
5747 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
5750 if (dev
->flags
& IFF_UP
&&
5751 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
))) {
5752 struct netdev_notifier_change_info change_info
;
5754 change_info
.flags_changed
= changes
;
5755 call_netdevice_notifiers_info(NETDEV_CHANGE
, dev
,
5761 * dev_change_flags - change device settings
5763 * @flags: device state flags
5765 * Change settings on device based state flags. The flags are
5766 * in the userspace exported format.
5768 int dev_change_flags(struct net_device
*dev
, unsigned int flags
)
5771 unsigned int changes
, old_flags
= dev
->flags
, old_gflags
= dev
->gflags
;
5773 ret
= __dev_change_flags(dev
, flags
);
5777 changes
= (old_flags
^ dev
->flags
) | (old_gflags
^ dev
->gflags
);
5778 __dev_notify_flags(dev
, old_flags
, changes
);
5781 EXPORT_SYMBOL(dev_change_flags
);
5783 static int __dev_set_mtu(struct net_device
*dev
, int new_mtu
)
5785 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5787 if (ops
->ndo_change_mtu
)
5788 return ops
->ndo_change_mtu(dev
, new_mtu
);
5795 * dev_set_mtu - Change maximum transfer unit
5797 * @new_mtu: new transfer unit
5799 * Change the maximum transfer size of the network device.
5801 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
5805 if (new_mtu
== dev
->mtu
)
5808 /* MTU must be positive. */
5812 if (!netif_device_present(dev
))
5815 err
= call_netdevice_notifiers(NETDEV_PRECHANGEMTU
, dev
);
5816 err
= notifier_to_errno(err
);
5820 orig_mtu
= dev
->mtu
;
5821 err
= __dev_set_mtu(dev
, new_mtu
);
5824 err
= call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
5825 err
= notifier_to_errno(err
);
5827 /* setting mtu back and notifying everyone again,
5828 * so that they have a chance to revert changes.
5830 __dev_set_mtu(dev
, orig_mtu
);
5831 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
5836 EXPORT_SYMBOL(dev_set_mtu
);
5839 * dev_set_group - Change group this device belongs to
5841 * @new_group: group this device should belong to
5843 void dev_set_group(struct net_device
*dev
, int new_group
)
5845 dev
->group
= new_group
;
5847 EXPORT_SYMBOL(dev_set_group
);
5850 * dev_set_mac_address - Change Media Access Control Address
5854 * Change the hardware (MAC) address of the device
5856 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
5858 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5861 if (!ops
->ndo_set_mac_address
)
5863 if (sa
->sa_family
!= dev
->type
)
5865 if (!netif_device_present(dev
))
5867 err
= ops
->ndo_set_mac_address(dev
, sa
);
5870 dev
->addr_assign_type
= NET_ADDR_SET
;
5871 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
5872 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
5875 EXPORT_SYMBOL(dev_set_mac_address
);
5878 * dev_change_carrier - Change device carrier
5880 * @new_carrier: new value
5882 * Change device carrier
5884 int dev_change_carrier(struct net_device
*dev
, bool new_carrier
)
5886 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5888 if (!ops
->ndo_change_carrier
)
5890 if (!netif_device_present(dev
))
5892 return ops
->ndo_change_carrier(dev
, new_carrier
);
5894 EXPORT_SYMBOL(dev_change_carrier
);
5897 * dev_get_phys_port_id - Get device physical port ID
5901 * Get device physical port ID
5903 int dev_get_phys_port_id(struct net_device
*dev
,
5904 struct netdev_phys_item_id
*ppid
)
5906 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5908 if (!ops
->ndo_get_phys_port_id
)
5910 return ops
->ndo_get_phys_port_id(dev
, ppid
);
5912 EXPORT_SYMBOL(dev_get_phys_port_id
);
5915 * dev_get_phys_port_name - Get device physical port name
5919 * Get device physical port name
5921 int dev_get_phys_port_name(struct net_device
*dev
,
5922 char *name
, size_t len
)
5924 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5926 if (!ops
->ndo_get_phys_port_name
)
5928 return ops
->ndo_get_phys_port_name(dev
, name
, len
);
5930 EXPORT_SYMBOL(dev_get_phys_port_name
);
5933 * dev_new_index - allocate an ifindex
5934 * @net: the applicable net namespace
5936 * Returns a suitable unique value for a new device interface
5937 * number. The caller must hold the rtnl semaphore or the
5938 * dev_base_lock to be sure it remains unique.
5940 static int dev_new_index(struct net
*net
)
5942 int ifindex
= net
->ifindex
;
5946 if (!__dev_get_by_index(net
, ifindex
))
5947 return net
->ifindex
= ifindex
;
5951 /* Delayed registration/unregisteration */
5952 static LIST_HEAD(net_todo_list
);
5953 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq
);
5955 static void net_set_todo(struct net_device
*dev
)
5957 list_add_tail(&dev
->todo_list
, &net_todo_list
);
5958 dev_net(dev
)->dev_unreg_count
++;
5961 static void rollback_registered_many(struct list_head
*head
)
5963 struct net_device
*dev
, *tmp
;
5964 LIST_HEAD(close_head
);
5966 BUG_ON(dev_boot_phase
);
5969 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
5970 /* Some devices call without registering
5971 * for initialization unwind. Remove those
5972 * devices and proceed with the remaining.
5974 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5975 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5979 list_del(&dev
->unreg_list
);
5982 dev
->dismantle
= true;
5983 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
5986 /* If device is running, close it first. */
5987 list_for_each_entry(dev
, head
, unreg_list
)
5988 list_add_tail(&dev
->close_list
, &close_head
);
5989 dev_close_many(&close_head
, true);
5991 list_for_each_entry(dev
, head
, unreg_list
) {
5992 /* And unlink it from device chain. */
5993 unlist_netdevice(dev
);
5995 dev
->reg_state
= NETREG_UNREGISTERING
;
6000 list_for_each_entry(dev
, head
, unreg_list
) {
6001 struct sk_buff
*skb
= NULL
;
6003 /* Shutdown queueing discipline. */
6007 /* Notify protocols, that we are about to destroy
6008 this device. They should clean all the things.
6010 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6012 if (!dev
->rtnl_link_ops
||
6013 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
6014 skb
= rtmsg_ifinfo_build_skb(RTM_DELLINK
, dev
, ~0U,
6018 * Flush the unicast and multicast chains
6023 if (dev
->netdev_ops
->ndo_uninit
)
6024 dev
->netdev_ops
->ndo_uninit(dev
);
6027 rtmsg_ifinfo_send(skb
, dev
, GFP_KERNEL
);
6029 /* Notifier chain MUST detach us all upper devices. */
6030 WARN_ON(netdev_has_any_upper_dev(dev
));
6032 /* Remove entries from kobject tree */
6033 netdev_unregister_kobject(dev
);
6035 /* Remove XPS queueing entries */
6036 netif_reset_xps_queues_gt(dev
, 0);
6042 list_for_each_entry(dev
, head
, unreg_list
)
6046 static void rollback_registered(struct net_device
*dev
)
6050 list_add(&dev
->unreg_list
, &single
);
6051 rollback_registered_many(&single
);
6055 static netdev_features_t
netdev_fix_features(struct net_device
*dev
,
6056 netdev_features_t features
)
6058 /* Fix illegal checksum combinations */
6059 if ((features
& NETIF_F_HW_CSUM
) &&
6060 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
6061 netdev_warn(dev
, "mixed HW and IP checksum settings.\n");
6062 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
6065 /* TSO requires that SG is present as well. */
6066 if ((features
& NETIF_F_ALL_TSO
) && !(features
& NETIF_F_SG
)) {
6067 netdev_dbg(dev
, "Dropping TSO features since no SG feature.\n");
6068 features
&= ~NETIF_F_ALL_TSO
;
6071 if ((features
& NETIF_F_TSO
) && !(features
& NETIF_F_HW_CSUM
) &&
6072 !(features
& NETIF_F_IP_CSUM
)) {
6073 netdev_dbg(dev
, "Dropping TSO features since no CSUM feature.\n");
6074 features
&= ~NETIF_F_TSO
;
6075 features
&= ~NETIF_F_TSO_ECN
;
6078 if ((features
& NETIF_F_TSO6
) && !(features
& NETIF_F_HW_CSUM
) &&
6079 !(features
& NETIF_F_IPV6_CSUM
)) {
6080 netdev_dbg(dev
, "Dropping TSO6 features since no CSUM feature.\n");
6081 features
&= ~NETIF_F_TSO6
;
6084 /* TSO ECN requires that TSO is present as well. */
6085 if ((features
& NETIF_F_ALL_TSO
) == NETIF_F_TSO_ECN
)
6086 features
&= ~NETIF_F_TSO_ECN
;
6088 /* Software GSO depends on SG. */
6089 if ((features
& NETIF_F_GSO
) && !(features
& NETIF_F_SG
)) {
6090 netdev_dbg(dev
, "Dropping NETIF_F_GSO since no SG feature.\n");
6091 features
&= ~NETIF_F_GSO
;
6094 /* UFO needs SG and checksumming */
6095 if (features
& NETIF_F_UFO
) {
6096 /* maybe split UFO into V4 and V6? */
6097 if (!((features
& NETIF_F_GEN_CSUM
) ||
6098 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))
6099 == (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
6101 "Dropping NETIF_F_UFO since no checksum offload features.\n");
6102 features
&= ~NETIF_F_UFO
;
6105 if (!(features
& NETIF_F_SG
)) {
6107 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
6108 features
&= ~NETIF_F_UFO
;
6112 #ifdef CONFIG_NET_RX_BUSY_POLL
6113 if (dev
->netdev_ops
->ndo_busy_poll
)
6114 features
|= NETIF_F_BUSY_POLL
;
6117 features
&= ~NETIF_F_BUSY_POLL
;
6122 int __netdev_update_features(struct net_device
*dev
)
6124 netdev_features_t features
;
6129 features
= netdev_get_wanted_features(dev
);
6131 if (dev
->netdev_ops
->ndo_fix_features
)
6132 features
= dev
->netdev_ops
->ndo_fix_features(dev
, features
);
6134 /* driver might be less strict about feature dependencies */
6135 features
= netdev_fix_features(dev
, features
);
6137 if (dev
->features
== features
)
6140 netdev_dbg(dev
, "Features changed: %pNF -> %pNF\n",
6141 &dev
->features
, &features
);
6143 if (dev
->netdev_ops
->ndo_set_features
)
6144 err
= dev
->netdev_ops
->ndo_set_features(dev
, features
);
6146 if (unlikely(err
< 0)) {
6148 "set_features() failed (%d); wanted %pNF, left %pNF\n",
6149 err
, &features
, &dev
->features
);
6154 dev
->features
= features
;
6160 * netdev_update_features - recalculate device features
6161 * @dev: the device to check
6163 * Recalculate dev->features set and send notifications if it
6164 * has changed. Should be called after driver or hardware dependent
6165 * conditions might have changed that influence the features.
6167 void netdev_update_features(struct net_device
*dev
)
6169 if (__netdev_update_features(dev
))
6170 netdev_features_change(dev
);
6172 EXPORT_SYMBOL(netdev_update_features
);
6175 * netdev_change_features - recalculate device features
6176 * @dev: the device to check
6178 * Recalculate dev->features set and send notifications even
6179 * if they have not changed. Should be called instead of
6180 * netdev_update_features() if also dev->vlan_features might
6181 * have changed to allow the changes to be propagated to stacked
6184 void netdev_change_features(struct net_device
*dev
)
6186 __netdev_update_features(dev
);
6187 netdev_features_change(dev
);
6189 EXPORT_SYMBOL(netdev_change_features
);
6192 * netif_stacked_transfer_operstate - transfer operstate
6193 * @rootdev: the root or lower level device to transfer state from
6194 * @dev: the device to transfer operstate to
6196 * Transfer operational state from root to device. This is normally
6197 * called when a stacking relationship exists between the root
6198 * device and the device(a leaf device).
6200 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
6201 struct net_device
*dev
)
6203 if (rootdev
->operstate
== IF_OPER_DORMANT
)
6204 netif_dormant_on(dev
);
6206 netif_dormant_off(dev
);
6208 if (netif_carrier_ok(rootdev
)) {
6209 if (!netif_carrier_ok(dev
))
6210 netif_carrier_on(dev
);
6212 if (netif_carrier_ok(dev
))
6213 netif_carrier_off(dev
);
6216 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
6219 static int netif_alloc_rx_queues(struct net_device
*dev
)
6221 unsigned int i
, count
= dev
->num_rx_queues
;
6222 struct netdev_rx_queue
*rx
;
6223 size_t sz
= count
* sizeof(*rx
);
6227 rx
= kzalloc(sz
, GFP_KERNEL
| __GFP_NOWARN
| __GFP_REPEAT
);
6235 for (i
= 0; i
< count
; i
++)
6241 static void netdev_init_one_queue(struct net_device
*dev
,
6242 struct netdev_queue
*queue
, void *_unused
)
6244 /* Initialize queue lock */
6245 spin_lock_init(&queue
->_xmit_lock
);
6246 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
6247 queue
->xmit_lock_owner
= -1;
6248 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
6251 dql_init(&queue
->dql
, HZ
);
6255 static void netif_free_tx_queues(struct net_device
*dev
)
6260 static int netif_alloc_netdev_queues(struct net_device
*dev
)
6262 unsigned int count
= dev
->num_tx_queues
;
6263 struct netdev_queue
*tx
;
6264 size_t sz
= count
* sizeof(*tx
);
6266 BUG_ON(count
< 1 || count
> 0xffff);
6268 tx
= kzalloc(sz
, GFP_KERNEL
| __GFP_NOWARN
| __GFP_REPEAT
);
6276 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
6277 spin_lock_init(&dev
->tx_global_lock
);
6283 * register_netdevice - register a network device
6284 * @dev: device to register
6286 * Take a completed network device structure and add it to the kernel
6287 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6288 * chain. 0 is returned on success. A negative errno code is returned
6289 * on a failure to set up the device, or if the name is a duplicate.
6291 * Callers must hold the rtnl semaphore. You may want
6292 * register_netdev() instead of this.
6295 * The locking appears insufficient to guarantee two parallel registers
6296 * will not get the same name.
6299 int register_netdevice(struct net_device
*dev
)
6302 struct net
*net
= dev_net(dev
);
6304 BUG_ON(dev_boot_phase
);
6309 /* When net_device's are persistent, this will be fatal. */
6310 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
6313 spin_lock_init(&dev
->addr_list_lock
);
6314 netdev_set_addr_lockdep_class(dev
);
6318 ret
= dev_get_valid_name(net
, dev
, dev
->name
);
6322 /* Init, if this function is available */
6323 if (dev
->netdev_ops
->ndo_init
) {
6324 ret
= dev
->netdev_ops
->ndo_init(dev
);
6332 if (((dev
->hw_features
| dev
->features
) &
6333 NETIF_F_HW_VLAN_CTAG_FILTER
) &&
6334 (!dev
->netdev_ops
->ndo_vlan_rx_add_vid
||
6335 !dev
->netdev_ops
->ndo_vlan_rx_kill_vid
)) {
6336 netdev_WARN(dev
, "Buggy VLAN acceleration in driver!\n");
6343 dev
->ifindex
= dev_new_index(net
);
6344 else if (__dev_get_by_index(net
, dev
->ifindex
))
6347 if (dev
->iflink
== -1)
6348 dev
->iflink
= dev
->ifindex
;
6350 /* Transfer changeable features to wanted_features and enable
6351 * software offloads (GSO and GRO).
6353 dev
->hw_features
|= NETIF_F_SOFT_FEATURES
;
6354 dev
->features
|= NETIF_F_SOFT_FEATURES
;
6355 dev
->wanted_features
= dev
->features
& dev
->hw_features
;
6357 if (!(dev
->flags
& IFF_LOOPBACK
)) {
6358 dev
->hw_features
|= NETIF_F_NOCACHE_COPY
;
6361 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6363 dev
->vlan_features
|= NETIF_F_HIGHDMA
;
6365 /* Make NETIF_F_SG inheritable to tunnel devices.
6367 dev
->hw_enc_features
|= NETIF_F_SG
;
6369 /* Make NETIF_F_SG inheritable to MPLS.
6371 dev
->mpls_features
|= NETIF_F_SG
;
6373 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
6374 ret
= notifier_to_errno(ret
);
6378 ret
= netdev_register_kobject(dev
);
6381 dev
->reg_state
= NETREG_REGISTERED
;
6383 __netdev_update_features(dev
);
6386 * Default initial state at registry is that the
6387 * device is present.
6390 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
6392 linkwatch_init_dev(dev
);
6394 dev_init_scheduler(dev
);
6396 list_netdevice(dev
);
6397 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
6399 /* If the device has permanent device address, driver should
6400 * set dev_addr and also addr_assign_type should be set to
6401 * NET_ADDR_PERM (default value).
6403 if (dev
->addr_assign_type
== NET_ADDR_PERM
)
6404 memcpy(dev
->perm_addr
, dev
->dev_addr
, dev
->addr_len
);
6406 /* Notify protocols, that a new device appeared. */
6407 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
6408 ret
= notifier_to_errno(ret
);
6410 rollback_registered(dev
);
6411 dev
->reg_state
= NETREG_UNREGISTERED
;
6414 * Prevent userspace races by waiting until the network
6415 * device is fully setup before sending notifications.
6417 if (!dev
->rtnl_link_ops
||
6418 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
6419 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U, GFP_KERNEL
);
6425 if (dev
->netdev_ops
->ndo_uninit
)
6426 dev
->netdev_ops
->ndo_uninit(dev
);
6429 EXPORT_SYMBOL(register_netdevice
);
6432 * init_dummy_netdev - init a dummy network device for NAPI
6433 * @dev: device to init
6435 * This takes a network device structure and initialize the minimum
6436 * amount of fields so it can be used to schedule NAPI polls without
6437 * registering a full blown interface. This is to be used by drivers
6438 * that need to tie several hardware interfaces to a single NAPI
6439 * poll scheduler due to HW limitations.
6441 int init_dummy_netdev(struct net_device
*dev
)
6443 /* Clear everything. Note we don't initialize spinlocks
6444 * are they aren't supposed to be taken by any of the
6445 * NAPI code and this dummy netdev is supposed to be
6446 * only ever used for NAPI polls
6448 memset(dev
, 0, sizeof(struct net_device
));
6450 /* make sure we BUG if trying to hit standard
6451 * register/unregister code path
6453 dev
->reg_state
= NETREG_DUMMY
;
6455 /* NAPI wants this */
6456 INIT_LIST_HEAD(&dev
->napi_list
);
6458 /* a dummy interface is started by default */
6459 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
6460 set_bit(__LINK_STATE_START
, &dev
->state
);
6462 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6463 * because users of this 'device' dont need to change
6469 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
6473 * register_netdev - register a network device
6474 * @dev: device to register
6476 * Take a completed network device structure and add it to the kernel
6477 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6478 * chain. 0 is returned on success. A negative errno code is returned
6479 * on a failure to set up the device, or if the name is a duplicate.
6481 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6482 * and expands the device name if you passed a format string to
6485 int register_netdev(struct net_device
*dev
)
6490 err
= register_netdevice(dev
);
6494 EXPORT_SYMBOL(register_netdev
);
6496 int netdev_refcnt_read(const struct net_device
*dev
)
6500 for_each_possible_cpu(i
)
6501 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
6504 EXPORT_SYMBOL(netdev_refcnt_read
);
6507 * netdev_wait_allrefs - wait until all references are gone.
6508 * @dev: target net_device
6510 * This is called when unregistering network devices.
6512 * Any protocol or device that holds a reference should register
6513 * for netdevice notification, and cleanup and put back the
6514 * reference if they receive an UNREGISTER event.
6515 * We can get stuck here if buggy protocols don't correctly
6518 static void netdev_wait_allrefs(struct net_device
*dev
)
6520 unsigned long rebroadcast_time
, warning_time
;
6523 linkwatch_forget_dev(dev
);
6525 rebroadcast_time
= warning_time
= jiffies
;
6526 refcnt
= netdev_refcnt_read(dev
);
6528 while (refcnt
!= 0) {
6529 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
6532 /* Rebroadcast unregister notification */
6533 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6539 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
6540 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
6542 /* We must not have linkwatch events
6543 * pending on unregister. If this
6544 * happens, we simply run the queue
6545 * unscheduled, resulting in a noop
6548 linkwatch_run_queue();
6553 rebroadcast_time
= jiffies
;
6558 refcnt
= netdev_refcnt_read(dev
);
6560 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
6561 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6563 warning_time
= jiffies
;
6572 * register_netdevice(x1);
6573 * register_netdevice(x2);
6575 * unregister_netdevice(y1);
6576 * unregister_netdevice(y2);
6582 * We are invoked by rtnl_unlock().
6583 * This allows us to deal with problems:
6584 * 1) We can delete sysfs objects which invoke hotplug
6585 * without deadlocking with linkwatch via keventd.
6586 * 2) Since we run with the RTNL semaphore not held, we can sleep
6587 * safely in order to wait for the netdev refcnt to drop to zero.
6589 * We must not return until all unregister events added during
6590 * the interval the lock was held have been completed.
6592 void netdev_run_todo(void)
6594 struct list_head list
;
6596 /* Snapshot list, allow later requests */
6597 list_replace_init(&net_todo_list
, &list
);
6602 /* Wait for rcu callbacks to finish before next phase */
6603 if (!list_empty(&list
))
6606 while (!list_empty(&list
)) {
6607 struct net_device
*dev
6608 = list_first_entry(&list
, struct net_device
, todo_list
);
6609 list_del(&dev
->todo_list
);
6612 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
6615 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
6616 pr_err("network todo '%s' but state %d\n",
6617 dev
->name
, dev
->reg_state
);
6622 dev
->reg_state
= NETREG_UNREGISTERED
;
6624 on_each_cpu(flush_backlog
, dev
, 1);
6626 netdev_wait_allrefs(dev
);
6629 BUG_ON(netdev_refcnt_read(dev
));
6630 BUG_ON(!list_empty(&dev
->ptype_all
));
6631 BUG_ON(!list_empty(&dev
->ptype_specific
));
6632 WARN_ON(rcu_access_pointer(dev
->ip_ptr
));
6633 WARN_ON(rcu_access_pointer(dev
->ip6_ptr
));
6634 WARN_ON(dev
->dn_ptr
);
6636 if (dev
->destructor
)
6637 dev
->destructor(dev
);
6639 /* Report a network device has been unregistered */
6641 dev_net(dev
)->dev_unreg_count
--;
6643 wake_up(&netdev_unregistering_wq
);
6645 /* Free network device */
6646 kobject_put(&dev
->dev
.kobj
);
6650 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6651 * fields in the same order, with only the type differing.
6653 void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
6654 const struct net_device_stats
*netdev_stats
)
6656 #if BITS_PER_LONG == 64
6657 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
6658 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
6660 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
6661 const unsigned long *src
= (const unsigned long *)netdev_stats
;
6662 u64
*dst
= (u64
*)stats64
;
6664 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
6665 sizeof(*stats64
) / sizeof(u64
));
6666 for (i
= 0; i
< n
; i
++)
6670 EXPORT_SYMBOL(netdev_stats_to_stats64
);
6673 * dev_get_stats - get network device statistics
6674 * @dev: device to get statistics from
6675 * @storage: place to store stats
6677 * Get network statistics from device. Return @storage.
6678 * The device driver may provide its own method by setting
6679 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6680 * otherwise the internal statistics structure is used.
6682 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
6683 struct rtnl_link_stats64
*storage
)
6685 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6687 if (ops
->ndo_get_stats64
) {
6688 memset(storage
, 0, sizeof(*storage
));
6689 ops
->ndo_get_stats64(dev
, storage
);
6690 } else if (ops
->ndo_get_stats
) {
6691 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
6693 netdev_stats_to_stats64(storage
, &dev
->stats
);
6695 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
6696 storage
->tx_dropped
+= atomic_long_read(&dev
->tx_dropped
);
6699 EXPORT_SYMBOL(dev_get_stats
);
6701 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
6703 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
6705 #ifdef CONFIG_NET_CLS_ACT
6708 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
6711 netdev_init_one_queue(dev
, queue
, NULL
);
6712 RCU_INIT_POINTER(queue
->qdisc
, &noop_qdisc
);
6713 queue
->qdisc_sleeping
= &noop_qdisc
;
6714 rcu_assign_pointer(dev
->ingress_queue
, queue
);
6719 static const struct ethtool_ops default_ethtool_ops
;
6721 void netdev_set_default_ethtool_ops(struct net_device
*dev
,
6722 const struct ethtool_ops
*ops
)
6724 if (dev
->ethtool_ops
== &default_ethtool_ops
)
6725 dev
->ethtool_ops
= ops
;
6727 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops
);
6729 void netdev_freemem(struct net_device
*dev
)
6731 char *addr
= (char *)dev
- dev
->padded
;
6737 * alloc_netdev_mqs - allocate network device
6738 * @sizeof_priv: size of private data to allocate space for
6739 * @name: device name format string
6740 * @name_assign_type: origin of device name
6741 * @setup: callback to initialize device
6742 * @txqs: the number of TX subqueues to allocate
6743 * @rxqs: the number of RX subqueues to allocate
6745 * Allocates a struct net_device with private data area for driver use
6746 * and performs basic initialization. Also allocates subqueue structs
6747 * for each queue on the device.
6749 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
6750 unsigned char name_assign_type
,
6751 void (*setup
)(struct net_device
*),
6752 unsigned int txqs
, unsigned int rxqs
)
6754 struct net_device
*dev
;
6756 struct net_device
*p
;
6758 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
6761 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6767 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6772 alloc_size
= sizeof(struct net_device
);
6774 /* ensure 32-byte alignment of private area */
6775 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
6776 alloc_size
+= sizeof_priv
;
6778 /* ensure 32-byte alignment of whole construct */
6779 alloc_size
+= NETDEV_ALIGN
- 1;
6781 p
= kzalloc(alloc_size
, GFP_KERNEL
| __GFP_NOWARN
| __GFP_REPEAT
);
6783 p
= vzalloc(alloc_size
);
6787 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
6788 dev
->padded
= (char *)dev
- (char *)p
;
6790 dev
->pcpu_refcnt
= alloc_percpu(int);
6791 if (!dev
->pcpu_refcnt
)
6794 if (dev_addr_init(dev
))
6800 dev_net_set(dev
, &init_net
);
6802 dev
->gso_max_size
= GSO_MAX_SIZE
;
6803 dev
->gso_max_segs
= GSO_MAX_SEGS
;
6804 dev
->gso_min_segs
= 0;
6806 INIT_LIST_HEAD(&dev
->napi_list
);
6807 INIT_LIST_HEAD(&dev
->unreg_list
);
6808 INIT_LIST_HEAD(&dev
->close_list
);
6809 INIT_LIST_HEAD(&dev
->link_watch_list
);
6810 INIT_LIST_HEAD(&dev
->adj_list
.upper
);
6811 INIT_LIST_HEAD(&dev
->adj_list
.lower
);
6812 INIT_LIST_HEAD(&dev
->all_adj_list
.upper
);
6813 INIT_LIST_HEAD(&dev
->all_adj_list
.lower
);
6814 INIT_LIST_HEAD(&dev
->ptype_all
);
6815 INIT_LIST_HEAD(&dev
->ptype_specific
);
6816 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
| IFF_XMIT_DST_RELEASE_PERM
;
6819 dev
->num_tx_queues
= txqs
;
6820 dev
->real_num_tx_queues
= txqs
;
6821 if (netif_alloc_netdev_queues(dev
))
6825 dev
->num_rx_queues
= rxqs
;
6826 dev
->real_num_rx_queues
= rxqs
;
6827 if (netif_alloc_rx_queues(dev
))
6831 strcpy(dev
->name
, name
);
6832 dev
->name_assign_type
= name_assign_type
;
6833 dev
->group
= INIT_NETDEV_GROUP
;
6834 if (!dev
->ethtool_ops
)
6835 dev
->ethtool_ops
= &default_ethtool_ops
;
6843 free_percpu(dev
->pcpu_refcnt
);
6845 netdev_freemem(dev
);
6848 EXPORT_SYMBOL(alloc_netdev_mqs
);
6851 * free_netdev - free network device
6854 * This function does the last stage of destroying an allocated device
6855 * interface. The reference to the device object is released.
6856 * If this is the last reference then it will be freed.
6858 void free_netdev(struct net_device
*dev
)
6860 struct napi_struct
*p
, *n
;
6862 netif_free_tx_queues(dev
);
6867 kfree(rcu_dereference_protected(dev
->ingress_queue
, 1));
6869 /* Flush device addresses */
6870 dev_addr_flush(dev
);
6872 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
6875 free_percpu(dev
->pcpu_refcnt
);
6876 dev
->pcpu_refcnt
= NULL
;
6878 /* Compatibility with error handling in drivers */
6879 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
6880 netdev_freemem(dev
);
6884 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
6885 dev
->reg_state
= NETREG_RELEASED
;
6887 /* will free via device release */
6888 put_device(&dev
->dev
);
6890 EXPORT_SYMBOL(free_netdev
);
6893 * synchronize_net - Synchronize with packet receive processing
6895 * Wait for packets currently being received to be done.
6896 * Does not block later packets from starting.
6898 void synchronize_net(void)
6901 if (rtnl_is_locked())
6902 synchronize_rcu_expedited();
6906 EXPORT_SYMBOL(synchronize_net
);
6909 * unregister_netdevice_queue - remove device from the kernel
6913 * This function shuts down a device interface and removes it
6914 * from the kernel tables.
6915 * If head not NULL, device is queued to be unregistered later.
6917 * Callers must hold the rtnl semaphore. You may want
6918 * unregister_netdev() instead of this.
6921 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
6926 list_move_tail(&dev
->unreg_list
, head
);
6928 rollback_registered(dev
);
6929 /* Finish processing unregister after unlock */
6933 EXPORT_SYMBOL(unregister_netdevice_queue
);
6936 * unregister_netdevice_many - unregister many devices
6937 * @head: list of devices
6939 * Note: As most callers use a stack allocated list_head,
6940 * we force a list_del() to make sure stack wont be corrupted later.
6942 void unregister_netdevice_many(struct list_head
*head
)
6944 struct net_device
*dev
;
6946 if (!list_empty(head
)) {
6947 rollback_registered_many(head
);
6948 list_for_each_entry(dev
, head
, unreg_list
)
6953 EXPORT_SYMBOL(unregister_netdevice_many
);
6956 * unregister_netdev - remove device from the kernel
6959 * This function shuts down a device interface and removes it
6960 * from the kernel tables.
6962 * This is just a wrapper for unregister_netdevice that takes
6963 * the rtnl semaphore. In general you want to use this and not
6964 * unregister_netdevice.
6966 void unregister_netdev(struct net_device
*dev
)
6969 unregister_netdevice(dev
);
6972 EXPORT_SYMBOL(unregister_netdev
);
6975 * dev_change_net_namespace - move device to different nethost namespace
6977 * @net: network namespace
6978 * @pat: If not NULL name pattern to try if the current device name
6979 * is already taken in the destination network namespace.
6981 * This function shuts down a device interface and moves it
6982 * to a new network namespace. On success 0 is returned, on
6983 * a failure a netagive errno code is returned.
6985 * Callers must hold the rtnl semaphore.
6988 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
6994 /* Don't allow namespace local devices to be moved. */
6996 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6999 /* Ensure the device has been registrered */
7000 if (dev
->reg_state
!= NETREG_REGISTERED
)
7003 /* Get out if there is nothing todo */
7005 if (net_eq(dev_net(dev
), net
))
7008 /* Pick the destination device name, and ensure
7009 * we can use it in the destination network namespace.
7012 if (__dev_get_by_name(net
, dev
->name
)) {
7013 /* We get here if we can't use the current device name */
7016 if (dev_get_valid_name(net
, dev
, pat
) < 0)
7021 * And now a mini version of register_netdevice unregister_netdevice.
7024 /* If device is running close it first. */
7027 /* And unlink it from device chain */
7029 unlist_netdevice(dev
);
7033 /* Shutdown queueing discipline. */
7036 /* Notify protocols, that we are about to destroy
7037 this device. They should clean all the things.
7039 Note that dev->reg_state stays at NETREG_REGISTERED.
7040 This is wanted because this way 8021q and macvlan know
7041 the device is just moving and can keep their slaves up.
7043 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
7045 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
7046 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U, GFP_KERNEL
);
7049 * Flush the unicast and multicast chains
7054 /* Send a netdev-removed uevent to the old namespace */
7055 kobject_uevent(&dev
->dev
.kobj
, KOBJ_REMOVE
);
7056 netdev_adjacent_del_links(dev
);
7058 /* Actually switch the network namespace */
7059 dev_net_set(dev
, net
);
7061 /* If there is an ifindex conflict assign a new one */
7062 if (__dev_get_by_index(net
, dev
->ifindex
)) {
7063 int iflink
= (dev
->iflink
== dev
->ifindex
);
7064 dev
->ifindex
= dev_new_index(net
);
7066 dev
->iflink
= dev
->ifindex
;
7069 /* Send a netdev-add uevent to the new namespace */
7070 kobject_uevent(&dev
->dev
.kobj
, KOBJ_ADD
);
7071 netdev_adjacent_add_links(dev
);
7073 /* Fixup kobjects */
7074 err
= device_rename(&dev
->dev
, dev
->name
);
7077 /* Add the device back in the hashes */
7078 list_netdevice(dev
);
7080 /* Notify protocols, that a new device appeared. */
7081 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
7084 * Prevent userspace races by waiting until the network
7085 * device is fully setup before sending notifications.
7087 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U, GFP_KERNEL
);
7094 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
7096 static int dev_cpu_callback(struct notifier_block
*nfb
,
7097 unsigned long action
,
7100 struct sk_buff
**list_skb
;
7101 struct sk_buff
*skb
;
7102 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
7103 struct softnet_data
*sd
, *oldsd
;
7105 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
7108 local_irq_disable();
7109 cpu
= smp_processor_id();
7110 sd
= &per_cpu(softnet_data
, cpu
);
7111 oldsd
= &per_cpu(softnet_data
, oldcpu
);
7113 /* Find end of our completion_queue. */
7114 list_skb
= &sd
->completion_queue
;
7116 list_skb
= &(*list_skb
)->next
;
7117 /* Append completion queue from offline CPU. */
7118 *list_skb
= oldsd
->completion_queue
;
7119 oldsd
->completion_queue
= NULL
;
7121 /* Append output queue from offline CPU. */
7122 if (oldsd
->output_queue
) {
7123 *sd
->output_queue_tailp
= oldsd
->output_queue
;
7124 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
7125 oldsd
->output_queue
= NULL
;
7126 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
7128 /* Append NAPI poll list from offline CPU, with one exception :
7129 * process_backlog() must be called by cpu owning percpu backlog.
7130 * We properly handle process_queue & input_pkt_queue later.
7132 while (!list_empty(&oldsd
->poll_list
)) {
7133 struct napi_struct
*napi
= list_first_entry(&oldsd
->poll_list
,
7137 list_del_init(&napi
->poll_list
);
7138 if (napi
->poll
== process_backlog
)
7141 ____napi_schedule(sd
, napi
);
7144 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
7147 /* Process offline CPU's input_pkt_queue */
7148 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
7150 input_queue_head_incr(oldsd
);
7152 while ((skb
= skb_dequeue(&oldsd
->input_pkt_queue
))) {
7154 input_queue_head_incr(oldsd
);
7162 * netdev_increment_features - increment feature set by one
7163 * @all: current feature set
7164 * @one: new feature set
7165 * @mask: mask feature set
7167 * Computes a new feature set after adding a device with feature set
7168 * @one to the master device with current feature set @all. Will not
7169 * enable anything that is off in @mask. Returns the new feature set.
7171 netdev_features_t
netdev_increment_features(netdev_features_t all
,
7172 netdev_features_t one
, netdev_features_t mask
)
7174 if (mask
& NETIF_F_GEN_CSUM
)
7175 mask
|= NETIF_F_ALL_CSUM
;
7176 mask
|= NETIF_F_VLAN_CHALLENGED
;
7178 all
|= one
& (NETIF_F_ONE_FOR_ALL
|NETIF_F_ALL_CSUM
) & mask
;
7179 all
&= one
| ~NETIF_F_ALL_FOR_ALL
;
7181 /* If one device supports hw checksumming, set for all. */
7182 if (all
& NETIF_F_GEN_CSUM
)
7183 all
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_GEN_CSUM
);
7187 EXPORT_SYMBOL(netdev_increment_features
);
7189 static struct hlist_head
* __net_init
netdev_create_hash(void)
7192 struct hlist_head
*hash
;
7194 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
7196 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
7197 INIT_HLIST_HEAD(&hash
[i
]);
7202 /* Initialize per network namespace state */
7203 static int __net_init
netdev_init(struct net
*net
)
7205 if (net
!= &init_net
)
7206 INIT_LIST_HEAD(&net
->dev_base_head
);
7208 net
->dev_name_head
= netdev_create_hash();
7209 if (net
->dev_name_head
== NULL
)
7212 net
->dev_index_head
= netdev_create_hash();
7213 if (net
->dev_index_head
== NULL
)
7219 kfree(net
->dev_name_head
);
7225 * netdev_drivername - network driver for the device
7226 * @dev: network device
7228 * Determine network driver for device.
7230 const char *netdev_drivername(const struct net_device
*dev
)
7232 const struct device_driver
*driver
;
7233 const struct device
*parent
;
7234 const char *empty
= "";
7236 parent
= dev
->dev
.parent
;
7240 driver
= parent
->driver
;
7241 if (driver
&& driver
->name
)
7242 return driver
->name
;
7246 static void __netdev_printk(const char *level
, const struct net_device
*dev
,
7247 struct va_format
*vaf
)
7249 if (dev
&& dev
->dev
.parent
) {
7250 dev_printk_emit(level
[1] - '0',
7253 dev_driver_string(dev
->dev
.parent
),
7254 dev_name(dev
->dev
.parent
),
7255 netdev_name(dev
), netdev_reg_state(dev
),
7258 printk("%s%s%s: %pV",
7259 level
, netdev_name(dev
), netdev_reg_state(dev
), vaf
);
7261 printk("%s(NULL net_device): %pV", level
, vaf
);
7265 void netdev_printk(const char *level
, const struct net_device
*dev
,
7266 const char *format
, ...)
7268 struct va_format vaf
;
7271 va_start(args
, format
);
7276 __netdev_printk(level
, dev
, &vaf
);
7280 EXPORT_SYMBOL(netdev_printk
);
7282 #define define_netdev_printk_level(func, level) \
7283 void func(const struct net_device *dev, const char *fmt, ...) \
7285 struct va_format vaf; \
7288 va_start(args, fmt); \
7293 __netdev_printk(level, dev, &vaf); \
7297 EXPORT_SYMBOL(func);
7299 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
7300 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
7301 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
7302 define_netdev_printk_level(netdev_err
, KERN_ERR
);
7303 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
7304 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
7305 define_netdev_printk_level(netdev_info
, KERN_INFO
);
7307 static void __net_exit
netdev_exit(struct net
*net
)
7309 kfree(net
->dev_name_head
);
7310 kfree(net
->dev_index_head
);
7313 static struct pernet_operations __net_initdata netdev_net_ops
= {
7314 .init
= netdev_init
,
7315 .exit
= netdev_exit
,
7318 static void __net_exit
default_device_exit(struct net
*net
)
7320 struct net_device
*dev
, *aux
;
7322 * Push all migratable network devices back to the
7323 * initial network namespace
7326 for_each_netdev_safe(net
, dev
, aux
) {
7328 char fb_name
[IFNAMSIZ
];
7330 /* Ignore unmoveable devices (i.e. loopback) */
7331 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
7334 /* Leave virtual devices for the generic cleanup */
7335 if (dev
->rtnl_link_ops
)
7338 /* Push remaining network devices to init_net */
7339 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
7340 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
7342 pr_emerg("%s: failed to move %s to init_net: %d\n",
7343 __func__
, dev
->name
, err
);
7350 static void __net_exit
rtnl_lock_unregistering(struct list_head
*net_list
)
7352 /* Return with the rtnl_lock held when there are no network
7353 * devices unregistering in any network namespace in net_list.
7357 DEFINE_WAIT_FUNC(wait
, woken_wake_function
);
7359 add_wait_queue(&netdev_unregistering_wq
, &wait
);
7361 unregistering
= false;
7363 list_for_each_entry(net
, net_list
, exit_list
) {
7364 if (net
->dev_unreg_count
> 0) {
7365 unregistering
= true;
7373 wait_woken(&wait
, TASK_UNINTERRUPTIBLE
, MAX_SCHEDULE_TIMEOUT
);
7375 remove_wait_queue(&netdev_unregistering_wq
, &wait
);
7378 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
7380 /* At exit all network devices most be removed from a network
7381 * namespace. Do this in the reverse order of registration.
7382 * Do this across as many network namespaces as possible to
7383 * improve batching efficiency.
7385 struct net_device
*dev
;
7387 LIST_HEAD(dev_kill_list
);
7389 /* To prevent network device cleanup code from dereferencing
7390 * loopback devices or network devices that have been freed
7391 * wait here for all pending unregistrations to complete,
7392 * before unregistring the loopback device and allowing the
7393 * network namespace be freed.
7395 * The netdev todo list containing all network devices
7396 * unregistrations that happen in default_device_exit_batch
7397 * will run in the rtnl_unlock() at the end of
7398 * default_device_exit_batch.
7400 rtnl_lock_unregistering(net_list
);
7401 list_for_each_entry(net
, net_list
, exit_list
) {
7402 for_each_netdev_reverse(net
, dev
) {
7403 if (dev
->rtnl_link_ops
&& dev
->rtnl_link_ops
->dellink
)
7404 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
7406 unregister_netdevice_queue(dev
, &dev_kill_list
);
7409 unregister_netdevice_many(&dev_kill_list
);
7413 static struct pernet_operations __net_initdata default_device_ops
= {
7414 .exit
= default_device_exit
,
7415 .exit_batch
= default_device_exit_batch
,
7419 * Initialize the DEV module. At boot time this walks the device list and
7420 * unhooks any devices that fail to initialise (normally hardware not
7421 * present) and leaves us with a valid list of present and active devices.
7426 * This is called single threaded during boot, so no need
7427 * to take the rtnl semaphore.
7429 static int __init
net_dev_init(void)
7431 int i
, rc
= -ENOMEM
;
7433 BUG_ON(!dev_boot_phase
);
7435 if (dev_proc_init())
7438 if (netdev_kobject_init())
7441 INIT_LIST_HEAD(&ptype_all
);
7442 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
7443 INIT_LIST_HEAD(&ptype_base
[i
]);
7445 INIT_LIST_HEAD(&offload_base
);
7447 if (register_pernet_subsys(&netdev_net_ops
))
7451 * Initialise the packet receive queues.
7454 for_each_possible_cpu(i
) {
7455 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
7457 skb_queue_head_init(&sd
->input_pkt_queue
);
7458 skb_queue_head_init(&sd
->process_queue
);
7459 INIT_LIST_HEAD(&sd
->poll_list
);
7460 sd
->output_queue_tailp
= &sd
->output_queue
;
7462 sd
->csd
.func
= rps_trigger_softirq
;
7467 sd
->backlog
.poll
= process_backlog
;
7468 sd
->backlog
.weight
= weight_p
;
7473 /* The loopback device is special if any other network devices
7474 * is present in a network namespace the loopback device must
7475 * be present. Since we now dynamically allocate and free the
7476 * loopback device ensure this invariant is maintained by
7477 * keeping the loopback device as the first device on the
7478 * list of network devices. Ensuring the loopback devices
7479 * is the first device that appears and the last network device
7482 if (register_pernet_device(&loopback_net_ops
))
7485 if (register_pernet_device(&default_device_ops
))
7488 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
7489 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
7491 hotcpu_notifier(dev_cpu_callback
, 0);
7498 subsys_initcall(net_dev_init
);