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 <net/busy_poll.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/stat.h>
103 #include <net/dst_metadata.h>
104 #include <net/pkt_sched.h>
105 #include <net/checksum.h>
106 #include <net/xfrm.h>
107 #include <linux/highmem.h>
108 #include <linux/init.h>
109 #include <linux/module.h>
110 #include <linux/netpoll.h>
111 #include <linux/rcupdate.h>
112 #include <linux/delay.h>
113 #include <net/iw_handler.h>
114 #include <asm/current.h>
115 #include <linux/audit.h>
116 #include <linux/dmaengine.h>
117 #include <linux/err.h>
118 #include <linux/ctype.h>
119 #include <linux/if_arp.h>
120 #include <linux/if_vlan.h>
121 #include <linux/ip.h>
123 #include <net/mpls.h>
124 #include <linux/ipv6.h>
125 #include <linux/in.h>
126 #include <linux/jhash.h>
127 #include <linux/random.h>
128 #include <trace/events/napi.h>
129 #include <trace/events/net.h>
130 #include <trace/events/skb.h>
131 #include <linux/pci.h>
132 #include <linux/inetdevice.h>
133 #include <linux/cpu_rmap.h>
134 #include <linux/static_key.h>
135 #include <linux/hashtable.h>
136 #include <linux/vmalloc.h>
137 #include <linux/if_macvlan.h>
138 #include <linux/errqueue.h>
139 #include <linux/hrtimer.h>
140 #include <linux/netfilter_ingress.h>
142 #include "net-sysfs.h"
144 /* Instead of increasing this, you should create a hash table. */
145 #define MAX_GRO_SKBS 8
147 /* This should be increased if a protocol with a bigger head is added. */
148 #define GRO_MAX_HEAD (MAX_HEADER + 128)
150 static DEFINE_SPINLOCK(ptype_lock
);
151 static DEFINE_SPINLOCK(offload_lock
);
152 struct list_head ptype_base
[PTYPE_HASH_SIZE
] __read_mostly
;
153 struct list_head ptype_all __read_mostly
; /* Taps */
154 static struct list_head offload_base __read_mostly
;
156 static int netif_rx_internal(struct sk_buff
*skb
);
157 static int call_netdevice_notifiers_info(unsigned long val
,
158 struct net_device
*dev
,
159 struct netdev_notifier_info
*info
);
162 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
165 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
167 * Writers must hold the rtnl semaphore while they loop through the
168 * dev_base_head list, and hold dev_base_lock for writing when they do the
169 * actual updates. This allows pure readers to access the list even
170 * while a writer is preparing to update it.
172 * To put it another way, dev_base_lock is held for writing only to
173 * protect against pure readers; the rtnl semaphore provides the
174 * protection against other writers.
176 * See, for example usages, register_netdevice() and
177 * unregister_netdevice(), which must be called with the rtnl
180 DEFINE_RWLOCK(dev_base_lock
);
181 EXPORT_SYMBOL(dev_base_lock
);
183 /* protects napi_hash addition/deletion and napi_gen_id */
184 static DEFINE_SPINLOCK(napi_hash_lock
);
186 static unsigned int napi_gen_id
= NR_CPUS
;
187 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash
, 8);
189 static seqcount_t devnet_rename_seq
;
191 static inline void dev_base_seq_inc(struct net
*net
)
193 while (++net
->dev_base_seq
== 0);
196 static inline struct hlist_head
*dev_name_hash(struct net
*net
, const char *name
)
198 unsigned int hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
200 return &net
->dev_name_head
[hash_32(hash
, NETDEV_HASHBITS
)];
203 static inline struct hlist_head
*dev_index_hash(struct net
*net
, int ifindex
)
205 return &net
->dev_index_head
[ifindex
& (NETDEV_HASHENTRIES
- 1)];
208 static inline void rps_lock(struct softnet_data
*sd
)
211 spin_lock(&sd
->input_pkt_queue
.lock
);
215 static inline void rps_unlock(struct softnet_data
*sd
)
218 spin_unlock(&sd
->input_pkt_queue
.lock
);
222 /* Device list insertion */
223 static void list_netdevice(struct net_device
*dev
)
225 struct net
*net
= dev_net(dev
);
229 write_lock_bh(&dev_base_lock
);
230 list_add_tail_rcu(&dev
->dev_list
, &net
->dev_base_head
);
231 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
232 hlist_add_head_rcu(&dev
->index_hlist
,
233 dev_index_hash(net
, dev
->ifindex
));
234 write_unlock_bh(&dev_base_lock
);
236 dev_base_seq_inc(net
);
239 /* Device list removal
240 * caller must respect a RCU grace period before freeing/reusing dev
242 static void unlist_netdevice(struct net_device
*dev
)
246 /* Unlink dev from the device chain */
247 write_lock_bh(&dev_base_lock
);
248 list_del_rcu(&dev
->dev_list
);
249 hlist_del_rcu(&dev
->name_hlist
);
250 hlist_del_rcu(&dev
->index_hlist
);
251 write_unlock_bh(&dev_base_lock
);
253 dev_base_seq_inc(dev_net(dev
));
260 static RAW_NOTIFIER_HEAD(netdev_chain
);
263 * Device drivers call our routines to queue packets here. We empty the
264 * queue in the local softnet handler.
267 DEFINE_PER_CPU_ALIGNED(struct softnet_data
, softnet_data
);
268 EXPORT_PER_CPU_SYMBOL(softnet_data
);
270 #ifdef CONFIG_LOCKDEP
272 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
273 * according to dev->type
275 static const unsigned short netdev_lock_type
[] =
276 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
277 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
278 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
279 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
280 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
281 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
282 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
283 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
284 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
285 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
286 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
287 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
288 ARPHRD_FCFABRIC
, ARPHRD_IEEE80211
, ARPHRD_IEEE80211_PRISM
,
289 ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_PHONET
, ARPHRD_PHONET_PIPE
,
290 ARPHRD_IEEE802154
, ARPHRD_VOID
, ARPHRD_NONE
};
292 static const char *const netdev_lock_name
[] =
293 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
294 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
295 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
296 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
297 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
298 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
299 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
300 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
301 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
302 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
303 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
304 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
305 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
306 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
307 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
309 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
310 static struct lock_class_key netdev_addr_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
312 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
316 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
317 if (netdev_lock_type
[i
] == dev_type
)
319 /* the last key is used by default */
320 return ARRAY_SIZE(netdev_lock_type
) - 1;
323 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
324 unsigned short dev_type
)
328 i
= netdev_lock_pos(dev_type
);
329 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
330 netdev_lock_name
[i
]);
333 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
337 i
= netdev_lock_pos(dev
->type
);
338 lockdep_set_class_and_name(&dev
->addr_list_lock
,
339 &netdev_addr_lock_key
[i
],
340 netdev_lock_name
[i
]);
343 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
344 unsigned short dev_type
)
347 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
352 /*******************************************************************************
354 Protocol management and registration routines
356 *******************************************************************************/
359 * Add a protocol ID to the list. Now that the input handler is
360 * smarter we can dispense with all the messy stuff that used to be
363 * BEWARE!!! Protocol handlers, mangling input packets,
364 * MUST BE last in hash buckets and checking protocol handlers
365 * MUST start from promiscuous ptype_all chain in net_bh.
366 * It is true now, do not change it.
367 * Explanation follows: if protocol handler, mangling packet, will
368 * be the first on list, it is not able to sense, that packet
369 * is cloned and should be copied-on-write, so that it will
370 * change it and subsequent readers will get broken packet.
374 static inline struct list_head
*ptype_head(const struct packet_type
*pt
)
376 if (pt
->type
== htons(ETH_P_ALL
))
377 return pt
->dev
? &pt
->dev
->ptype_all
: &ptype_all
;
379 return pt
->dev
? &pt
->dev
->ptype_specific
:
380 &ptype_base
[ntohs(pt
->type
) & PTYPE_HASH_MASK
];
384 * dev_add_pack - add packet handler
385 * @pt: packet type declaration
387 * Add a protocol handler to the networking stack. The passed &packet_type
388 * is linked into kernel lists and may not be freed until it has been
389 * removed from the kernel lists.
391 * This call does not sleep therefore it can not
392 * guarantee all CPU's that are in middle of receiving packets
393 * will see the new packet type (until the next received packet).
396 void dev_add_pack(struct packet_type
*pt
)
398 struct list_head
*head
= ptype_head(pt
);
400 spin_lock(&ptype_lock
);
401 list_add_rcu(&pt
->list
, head
);
402 spin_unlock(&ptype_lock
);
404 EXPORT_SYMBOL(dev_add_pack
);
407 * __dev_remove_pack - remove packet handler
408 * @pt: packet type declaration
410 * Remove a protocol handler that was previously added to the kernel
411 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
412 * from the kernel lists and can be freed or reused once this function
415 * The packet type might still be in use by receivers
416 * and must not be freed until after all the CPU's have gone
417 * through a quiescent state.
419 void __dev_remove_pack(struct packet_type
*pt
)
421 struct list_head
*head
= ptype_head(pt
);
422 struct packet_type
*pt1
;
424 spin_lock(&ptype_lock
);
426 list_for_each_entry(pt1
, head
, list
) {
428 list_del_rcu(&pt
->list
);
433 pr_warn("dev_remove_pack: %p not found\n", pt
);
435 spin_unlock(&ptype_lock
);
437 EXPORT_SYMBOL(__dev_remove_pack
);
440 * dev_remove_pack - remove packet handler
441 * @pt: packet type declaration
443 * Remove a protocol handler that was previously added to the kernel
444 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
445 * from the kernel lists and can be freed or reused once this function
448 * This call sleeps to guarantee that no CPU is looking at the packet
451 void dev_remove_pack(struct packet_type
*pt
)
453 __dev_remove_pack(pt
);
457 EXPORT_SYMBOL(dev_remove_pack
);
461 * dev_add_offload - register offload handlers
462 * @po: protocol offload declaration
464 * Add protocol offload handlers to the networking stack. The passed
465 * &proto_offload is linked into kernel lists and may not be freed until
466 * it has been removed from the kernel lists.
468 * This call does not sleep therefore it can not
469 * guarantee all CPU's that are in middle of receiving packets
470 * will see the new offload handlers (until the next received packet).
472 void dev_add_offload(struct packet_offload
*po
)
474 struct packet_offload
*elem
;
476 spin_lock(&offload_lock
);
477 list_for_each_entry(elem
, &offload_base
, list
) {
478 if (po
->priority
< elem
->priority
)
481 list_add_rcu(&po
->list
, elem
->list
.prev
);
482 spin_unlock(&offload_lock
);
484 EXPORT_SYMBOL(dev_add_offload
);
487 * __dev_remove_offload - remove offload handler
488 * @po: packet offload declaration
490 * Remove a protocol offload handler that was previously added to the
491 * kernel offload handlers by dev_add_offload(). The passed &offload_type
492 * is removed from the kernel lists and can be freed or reused once this
495 * The packet type might still be in use by receivers
496 * and must not be freed until after all the CPU's have gone
497 * through a quiescent state.
499 static void __dev_remove_offload(struct packet_offload
*po
)
501 struct list_head
*head
= &offload_base
;
502 struct packet_offload
*po1
;
504 spin_lock(&offload_lock
);
506 list_for_each_entry(po1
, head
, list
) {
508 list_del_rcu(&po
->list
);
513 pr_warn("dev_remove_offload: %p not found\n", po
);
515 spin_unlock(&offload_lock
);
519 * dev_remove_offload - remove packet offload handler
520 * @po: packet offload declaration
522 * Remove a packet offload handler that was previously added to the kernel
523 * offload handlers by dev_add_offload(). The passed &offload_type is
524 * removed from the kernel lists and can be freed or reused once this
527 * This call sleeps to guarantee that no CPU is looking at the packet
530 void dev_remove_offload(struct packet_offload
*po
)
532 __dev_remove_offload(po
);
536 EXPORT_SYMBOL(dev_remove_offload
);
538 /******************************************************************************
540 Device Boot-time Settings Routines
542 *******************************************************************************/
544 /* Boot time configuration table */
545 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
548 * netdev_boot_setup_add - add new setup entry
549 * @name: name of the device
550 * @map: configured settings for the device
552 * Adds new setup entry to the dev_boot_setup list. The function
553 * returns 0 on error and 1 on success. This is a generic routine to
556 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
558 struct netdev_boot_setup
*s
;
562 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
563 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
564 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
565 strlcpy(s
[i
].name
, name
, IFNAMSIZ
);
566 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
571 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
575 * netdev_boot_setup_check - check boot time settings
576 * @dev: the netdevice
578 * Check boot time settings for the device.
579 * The found settings are set for the device to be used
580 * later in the device probing.
581 * Returns 0 if no settings found, 1 if they are.
583 int netdev_boot_setup_check(struct net_device
*dev
)
585 struct netdev_boot_setup
*s
= dev_boot_setup
;
588 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
589 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
590 !strcmp(dev
->name
, s
[i
].name
)) {
591 dev
->irq
= s
[i
].map
.irq
;
592 dev
->base_addr
= s
[i
].map
.base_addr
;
593 dev
->mem_start
= s
[i
].map
.mem_start
;
594 dev
->mem_end
= s
[i
].map
.mem_end
;
600 EXPORT_SYMBOL(netdev_boot_setup_check
);
604 * netdev_boot_base - get address from boot time settings
605 * @prefix: prefix for network device
606 * @unit: id for network device
608 * Check boot time settings for the base address of device.
609 * The found settings are set for the device to be used
610 * later in the device probing.
611 * Returns 0 if no settings found.
613 unsigned long netdev_boot_base(const char *prefix
, int unit
)
615 const struct netdev_boot_setup
*s
= dev_boot_setup
;
619 sprintf(name
, "%s%d", prefix
, unit
);
622 * If device already registered then return base of 1
623 * to indicate not to probe for this interface
625 if (__dev_get_by_name(&init_net
, name
))
628 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
629 if (!strcmp(name
, s
[i
].name
))
630 return s
[i
].map
.base_addr
;
635 * Saves at boot time configured settings for any netdevice.
637 int __init
netdev_boot_setup(char *str
)
642 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
647 memset(&map
, 0, sizeof(map
));
651 map
.base_addr
= ints
[2];
653 map
.mem_start
= ints
[3];
655 map
.mem_end
= ints
[4];
657 /* Add new entry to the list */
658 return netdev_boot_setup_add(str
, &map
);
661 __setup("netdev=", netdev_boot_setup
);
663 /*******************************************************************************
665 Device Interface Subroutines
667 *******************************************************************************/
670 * dev_get_iflink - get 'iflink' value of a interface
671 * @dev: targeted interface
673 * Indicates the ifindex the interface is linked to.
674 * Physical interfaces have the same 'ifindex' and 'iflink' values.
677 int dev_get_iflink(const struct net_device
*dev
)
679 if (dev
->netdev_ops
&& dev
->netdev_ops
->ndo_get_iflink
)
680 return dev
->netdev_ops
->ndo_get_iflink(dev
);
684 EXPORT_SYMBOL(dev_get_iflink
);
687 * dev_fill_metadata_dst - Retrieve tunnel egress information.
688 * @dev: targeted interface
691 * For better visibility of tunnel traffic OVS needs to retrieve
692 * egress tunnel information for a packet. Following API allows
693 * user to get this info.
695 int dev_fill_metadata_dst(struct net_device
*dev
, struct sk_buff
*skb
)
697 struct ip_tunnel_info
*info
;
699 if (!dev
->netdev_ops
|| !dev
->netdev_ops
->ndo_fill_metadata_dst
)
702 info
= skb_tunnel_info_unclone(skb
);
705 if (unlikely(!(info
->mode
& IP_TUNNEL_INFO_TX
)))
708 return dev
->netdev_ops
->ndo_fill_metadata_dst(dev
, skb
);
710 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst
);
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. Must be called under RTNL semaphore
718 * or @dev_base_lock. If the name is found a pointer to the device
719 * is returned. If the name is not found then %NULL is returned. The
720 * reference counters are not incremented so the caller must be
721 * careful with locks.
724 struct net_device
*__dev_get_by_name(struct net
*net
, const char *name
)
726 struct net_device
*dev
;
727 struct hlist_head
*head
= dev_name_hash(net
, name
);
729 hlist_for_each_entry(dev
, head
, name_hlist
)
730 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
735 EXPORT_SYMBOL(__dev_get_by_name
);
738 * dev_get_by_name_rcu - find a device by its name
739 * @net: the applicable net namespace
740 * @name: name to find
742 * Find an interface by name.
743 * If the name is found a pointer to the device is returned.
744 * If the name is not found then %NULL is returned.
745 * The reference counters are not incremented so the caller must be
746 * careful with locks. The caller must hold RCU lock.
749 struct net_device
*dev_get_by_name_rcu(struct net
*net
, const char *name
)
751 struct net_device
*dev
;
752 struct hlist_head
*head
= dev_name_hash(net
, name
);
754 hlist_for_each_entry_rcu(dev
, head
, name_hlist
)
755 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
760 EXPORT_SYMBOL(dev_get_by_name_rcu
);
763 * dev_get_by_name - find a device by its name
764 * @net: the applicable net namespace
765 * @name: name to find
767 * Find an interface by name. This can be called from any
768 * context and does its own locking. The returned handle has
769 * the usage count incremented and the caller must use dev_put() to
770 * release it when it is no longer needed. %NULL is returned if no
771 * matching device is found.
774 struct net_device
*dev_get_by_name(struct net
*net
, const char *name
)
776 struct net_device
*dev
;
779 dev
= dev_get_by_name_rcu(net
, name
);
785 EXPORT_SYMBOL(dev_get_by_name
);
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 has not
794 * had its reference counter increased so the caller must be careful
795 * about locking. The caller must hold either the RTNL semaphore
799 struct net_device
*__dev_get_by_index(struct net
*net
, int ifindex
)
801 struct net_device
*dev
;
802 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
804 hlist_for_each_entry(dev
, head
, index_hlist
)
805 if (dev
->ifindex
== ifindex
)
810 EXPORT_SYMBOL(__dev_get_by_index
);
813 * dev_get_by_index_rcu - find a device by its ifindex
814 * @net: the applicable net namespace
815 * @ifindex: index of device
817 * Search for an interface by index. Returns %NULL if the device
818 * is not found or a pointer to the device. The device has not
819 * had its reference counter increased so the caller must be careful
820 * about locking. The caller must hold RCU lock.
823 struct net_device
*dev_get_by_index_rcu(struct net
*net
, int ifindex
)
825 struct net_device
*dev
;
826 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
828 hlist_for_each_entry_rcu(dev
, head
, index_hlist
)
829 if (dev
->ifindex
== ifindex
)
834 EXPORT_SYMBOL(dev_get_by_index_rcu
);
838 * dev_get_by_index - find a device by its ifindex
839 * @net: the applicable net namespace
840 * @ifindex: index of device
842 * Search for an interface by index. Returns NULL if the device
843 * is not found or a pointer to the device. The device returned has
844 * had a reference added and the pointer is safe until the user calls
845 * dev_put to indicate they have finished with it.
848 struct net_device
*dev_get_by_index(struct net
*net
, int ifindex
)
850 struct net_device
*dev
;
853 dev
= dev_get_by_index_rcu(net
, ifindex
);
859 EXPORT_SYMBOL(dev_get_by_index
);
862 * netdev_get_name - get a netdevice name, knowing its ifindex.
863 * @net: network namespace
864 * @name: a pointer to the buffer where the name will be stored.
865 * @ifindex: the ifindex of the interface to get the name from.
867 * The use of raw_seqcount_begin() and cond_resched() before
868 * retrying is required as we want to give the writers a chance
869 * to complete when CONFIG_PREEMPT is not set.
871 int netdev_get_name(struct net
*net
, char *name
, int ifindex
)
873 struct net_device
*dev
;
877 seq
= raw_seqcount_begin(&devnet_rename_seq
);
879 dev
= dev_get_by_index_rcu(net
, ifindex
);
885 strcpy(name
, dev
->name
);
887 if (read_seqcount_retry(&devnet_rename_seq
, seq
)) {
896 * dev_getbyhwaddr_rcu - find a device by its hardware address
897 * @net: the applicable net namespace
898 * @type: media type of device
899 * @ha: hardware address
901 * Search for an interface by MAC address. Returns NULL if the device
902 * is not found or a pointer to the device.
903 * The caller must hold RCU or RTNL.
904 * The returned device has not had its ref count increased
905 * and the caller must therefore be careful about locking
909 struct net_device
*dev_getbyhwaddr_rcu(struct net
*net
, unsigned short type
,
912 struct net_device
*dev
;
914 for_each_netdev_rcu(net
, dev
)
915 if (dev
->type
== type
&&
916 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
921 EXPORT_SYMBOL(dev_getbyhwaddr_rcu
);
923 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
925 struct net_device
*dev
;
928 for_each_netdev(net
, dev
)
929 if (dev
->type
== type
)
934 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
936 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
938 struct net_device
*dev
, *ret
= NULL
;
941 for_each_netdev_rcu(net
, dev
)
942 if (dev
->type
== type
) {
950 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
953 * __dev_get_by_flags - find any device with given flags
954 * @net: the applicable net namespace
955 * @if_flags: IFF_* values
956 * @mask: bitmask of bits in if_flags to check
958 * Search for any interface with the given flags. Returns NULL if a device
959 * is not found or a pointer to the device. Must be called inside
960 * rtnl_lock(), and result refcount is unchanged.
963 struct net_device
*__dev_get_by_flags(struct net
*net
, unsigned short if_flags
,
966 struct net_device
*dev
, *ret
;
971 for_each_netdev(net
, dev
) {
972 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
979 EXPORT_SYMBOL(__dev_get_by_flags
);
982 * dev_valid_name - check if name is okay for network device
985 * Network device names need to be valid file names to
986 * to allow sysfs to work. We also disallow any kind of
989 bool dev_valid_name(const char *name
)
993 if (strlen(name
) >= IFNAMSIZ
)
995 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
999 if (*name
== '/' || *name
== ':' || isspace(*name
))
1005 EXPORT_SYMBOL(dev_valid_name
);
1008 * __dev_alloc_name - allocate a name for a device
1009 * @net: network namespace to allocate the device name in
1010 * @name: name format string
1011 * @buf: scratch buffer and result name string
1013 * Passed a format string - eg "lt%d" it will try and find a suitable
1014 * id. It scans list of devices to build up a free map, then chooses
1015 * the first empty slot. The caller must hold the dev_base or rtnl lock
1016 * while allocating the name and adding the device in order to avoid
1018 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1019 * Returns the number of the unit assigned or a negative errno code.
1022 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
1026 const int max_netdevices
= 8*PAGE_SIZE
;
1027 unsigned long *inuse
;
1028 struct net_device
*d
;
1030 p
= strnchr(name
, IFNAMSIZ
-1, '%');
1033 * Verify the string as this thing may have come from
1034 * the user. There must be either one "%d" and no other "%"
1037 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
1040 /* Use one page as a bit array of possible slots */
1041 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
1045 for_each_netdev(net
, d
) {
1046 if (!sscanf(d
->name
, name
, &i
))
1048 if (i
< 0 || i
>= max_netdevices
)
1051 /* avoid cases where sscanf is not exact inverse of printf */
1052 snprintf(buf
, IFNAMSIZ
, name
, i
);
1053 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
1057 i
= find_first_zero_bit(inuse
, max_netdevices
);
1058 free_page((unsigned long) inuse
);
1062 snprintf(buf
, IFNAMSIZ
, name
, i
);
1063 if (!__dev_get_by_name(net
, buf
))
1066 /* It is possible to run out of possible slots
1067 * when the name is long and there isn't enough space left
1068 * for the digits, or if all bits are used.
1074 * dev_alloc_name - allocate a name for a device
1076 * @name: name format string
1078 * Passed a format string - eg "lt%d" it will try and find a suitable
1079 * id. It scans list of devices to build up a free map, then chooses
1080 * the first empty slot. The caller must hold the dev_base or rtnl lock
1081 * while allocating the name and adding the device in order to avoid
1083 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1084 * Returns the number of the unit assigned or a negative errno code.
1087 int dev_alloc_name(struct net_device
*dev
, const char *name
)
1093 BUG_ON(!dev_net(dev
));
1095 ret
= __dev_alloc_name(net
, name
, buf
);
1097 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
1100 EXPORT_SYMBOL(dev_alloc_name
);
1102 static int dev_alloc_name_ns(struct net
*net
,
1103 struct net_device
*dev
,
1109 ret
= __dev_alloc_name(net
, name
, buf
);
1111 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
1115 static int dev_get_valid_name(struct net
*net
,
1116 struct net_device
*dev
,
1121 if (!dev_valid_name(name
))
1124 if (strchr(name
, '%'))
1125 return dev_alloc_name_ns(net
, dev
, name
);
1126 else if (__dev_get_by_name(net
, name
))
1128 else if (dev
->name
!= name
)
1129 strlcpy(dev
->name
, name
, IFNAMSIZ
);
1135 * dev_change_name - change name of a device
1137 * @newname: name (or format string) must be at least IFNAMSIZ
1139 * Change name of a device, can pass format strings "eth%d".
1142 int dev_change_name(struct net_device
*dev
, const char *newname
)
1144 unsigned char old_assign_type
;
1145 char oldname
[IFNAMSIZ
];
1151 BUG_ON(!dev_net(dev
));
1154 if (dev
->flags
& IFF_UP
)
1157 write_seqcount_begin(&devnet_rename_seq
);
1159 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0) {
1160 write_seqcount_end(&devnet_rename_seq
);
1164 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
1166 err
= dev_get_valid_name(net
, dev
, newname
);
1168 write_seqcount_end(&devnet_rename_seq
);
1172 if (oldname
[0] && !strchr(oldname
, '%'))
1173 netdev_info(dev
, "renamed from %s\n", oldname
);
1175 old_assign_type
= dev
->name_assign_type
;
1176 dev
->name_assign_type
= NET_NAME_RENAMED
;
1179 ret
= device_rename(&dev
->dev
, dev
->name
);
1181 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1182 dev
->name_assign_type
= old_assign_type
;
1183 write_seqcount_end(&devnet_rename_seq
);
1187 write_seqcount_end(&devnet_rename_seq
);
1189 netdev_adjacent_rename_links(dev
, oldname
);
1191 write_lock_bh(&dev_base_lock
);
1192 hlist_del_rcu(&dev
->name_hlist
);
1193 write_unlock_bh(&dev_base_lock
);
1197 write_lock_bh(&dev_base_lock
);
1198 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
1199 write_unlock_bh(&dev_base_lock
);
1201 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
1202 ret
= notifier_to_errno(ret
);
1205 /* err >= 0 after dev_alloc_name() or stores the first errno */
1208 write_seqcount_begin(&devnet_rename_seq
);
1209 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1210 memcpy(oldname
, newname
, IFNAMSIZ
);
1211 dev
->name_assign_type
= old_assign_type
;
1212 old_assign_type
= NET_NAME_RENAMED
;
1215 pr_err("%s: name change rollback failed: %d\n",
1224 * dev_set_alias - change ifalias of a device
1226 * @alias: name up to IFALIASZ
1227 * @len: limit of bytes to copy from info
1229 * Set ifalias for a device,
1231 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
1237 if (len
>= IFALIASZ
)
1241 kfree(dev
->ifalias
);
1242 dev
->ifalias
= NULL
;
1246 new_ifalias
= krealloc(dev
->ifalias
, len
+ 1, GFP_KERNEL
);
1249 dev
->ifalias
= new_ifalias
;
1251 strlcpy(dev
->ifalias
, alias
, len
+1);
1257 * netdev_features_change - device changes features
1258 * @dev: device to cause notification
1260 * Called to indicate a device has changed features.
1262 void netdev_features_change(struct net_device
*dev
)
1264 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1266 EXPORT_SYMBOL(netdev_features_change
);
1269 * netdev_state_change - device changes state
1270 * @dev: device to cause notification
1272 * Called to indicate a device has changed state. This function calls
1273 * the notifier chains for netdev_chain and sends a NEWLINK message
1274 * to the routing socket.
1276 void netdev_state_change(struct net_device
*dev
)
1278 if (dev
->flags
& IFF_UP
) {
1279 struct netdev_notifier_change_info change_info
;
1281 change_info
.flags_changed
= 0;
1282 call_netdevice_notifiers_info(NETDEV_CHANGE
, dev
,
1284 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0, GFP_KERNEL
);
1287 EXPORT_SYMBOL(netdev_state_change
);
1290 * netdev_notify_peers - notify network peers about existence of @dev
1291 * @dev: network device
1293 * Generate traffic such that interested network peers are aware of
1294 * @dev, such as by generating a gratuitous ARP. This may be used when
1295 * a device wants to inform the rest of the network about some sort of
1296 * reconfiguration such as a failover event or virtual machine
1299 void netdev_notify_peers(struct net_device
*dev
)
1302 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
, dev
);
1305 EXPORT_SYMBOL(netdev_notify_peers
);
1307 static int __dev_open(struct net_device
*dev
)
1309 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1314 if (!netif_device_present(dev
))
1317 /* Block netpoll from trying to do any rx path servicing.
1318 * If we don't do this there is a chance ndo_poll_controller
1319 * or ndo_poll may be running while we open the device
1321 netpoll_poll_disable(dev
);
1323 ret
= call_netdevice_notifiers(NETDEV_PRE_UP
, dev
);
1324 ret
= notifier_to_errno(ret
);
1328 set_bit(__LINK_STATE_START
, &dev
->state
);
1330 if (ops
->ndo_validate_addr
)
1331 ret
= ops
->ndo_validate_addr(dev
);
1333 if (!ret
&& ops
->ndo_open
)
1334 ret
= ops
->ndo_open(dev
);
1336 netpoll_poll_enable(dev
);
1339 clear_bit(__LINK_STATE_START
, &dev
->state
);
1341 dev
->flags
|= IFF_UP
;
1342 dev_set_rx_mode(dev
);
1344 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
1351 * dev_open - prepare an interface for use.
1352 * @dev: device to open
1354 * Takes a device from down to up state. The device's private open
1355 * function is invoked and then the multicast lists are loaded. Finally
1356 * the device is moved into the up state and a %NETDEV_UP message is
1357 * sent to the netdev notifier chain.
1359 * Calling this function on an active interface is a nop. On a failure
1360 * a negative errno code is returned.
1362 int dev_open(struct net_device
*dev
)
1366 if (dev
->flags
& IFF_UP
)
1369 ret
= __dev_open(dev
);
1373 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
, GFP_KERNEL
);
1374 call_netdevice_notifiers(NETDEV_UP
, dev
);
1378 EXPORT_SYMBOL(dev_open
);
1380 static int __dev_close_many(struct list_head
*head
)
1382 struct net_device
*dev
;
1387 list_for_each_entry(dev
, head
, close_list
) {
1388 /* Temporarily disable netpoll until the interface is down */
1389 netpoll_poll_disable(dev
);
1391 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1393 clear_bit(__LINK_STATE_START
, &dev
->state
);
1395 /* Synchronize to scheduled poll. We cannot touch poll list, it
1396 * can be even on different cpu. So just clear netif_running().
1398 * dev->stop() will invoke napi_disable() on all of it's
1399 * napi_struct instances on this device.
1401 smp_mb__after_atomic(); /* Commit netif_running(). */
1404 dev_deactivate_many(head
);
1406 list_for_each_entry(dev
, head
, close_list
) {
1407 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1410 * Call the device specific close. This cannot fail.
1411 * Only if device is UP
1413 * We allow it to be called even after a DETACH hot-plug
1419 dev
->flags
&= ~IFF_UP
;
1420 netpoll_poll_enable(dev
);
1426 static int __dev_close(struct net_device
*dev
)
1431 list_add(&dev
->close_list
, &single
);
1432 retval
= __dev_close_many(&single
);
1438 int dev_close_many(struct list_head
*head
, bool unlink
)
1440 struct net_device
*dev
, *tmp
;
1442 /* Remove the devices that don't need to be closed */
1443 list_for_each_entry_safe(dev
, tmp
, head
, close_list
)
1444 if (!(dev
->flags
& IFF_UP
))
1445 list_del_init(&dev
->close_list
);
1447 __dev_close_many(head
);
1449 list_for_each_entry_safe(dev
, tmp
, head
, close_list
) {
1450 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
, GFP_KERNEL
);
1451 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1453 list_del_init(&dev
->close_list
);
1458 EXPORT_SYMBOL(dev_close_many
);
1461 * dev_close - shutdown an interface.
1462 * @dev: device to shutdown
1464 * This function moves an active device into down state. A
1465 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1466 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1469 int dev_close(struct net_device
*dev
)
1471 if (dev
->flags
& IFF_UP
) {
1474 list_add(&dev
->close_list
, &single
);
1475 dev_close_many(&single
, true);
1480 EXPORT_SYMBOL(dev_close
);
1484 * dev_disable_lro - disable Large Receive Offload on a device
1487 * Disable Large Receive Offload (LRO) on a net device. Must be
1488 * called under RTNL. This is needed if received packets may be
1489 * forwarded to another interface.
1491 void dev_disable_lro(struct net_device
*dev
)
1493 struct net_device
*lower_dev
;
1494 struct list_head
*iter
;
1496 dev
->wanted_features
&= ~NETIF_F_LRO
;
1497 netdev_update_features(dev
);
1499 if (unlikely(dev
->features
& NETIF_F_LRO
))
1500 netdev_WARN(dev
, "failed to disable LRO!\n");
1502 netdev_for_each_lower_dev(dev
, lower_dev
, iter
)
1503 dev_disable_lro(lower_dev
);
1505 EXPORT_SYMBOL(dev_disable_lro
);
1507 static int call_netdevice_notifier(struct notifier_block
*nb
, unsigned long val
,
1508 struct net_device
*dev
)
1510 struct netdev_notifier_info info
;
1512 netdev_notifier_info_init(&info
, dev
);
1513 return nb
->notifier_call(nb
, val
, &info
);
1516 static int dev_boot_phase
= 1;
1519 * register_netdevice_notifier - register a network notifier block
1522 * Register a notifier to be called when network device events occur.
1523 * The notifier passed is linked into the kernel structures and must
1524 * not be reused until it has been unregistered. A negative errno code
1525 * is returned on a failure.
1527 * When registered all registration and up events are replayed
1528 * to the new notifier to allow device to have a race free
1529 * view of the network device list.
1532 int register_netdevice_notifier(struct notifier_block
*nb
)
1534 struct net_device
*dev
;
1535 struct net_device
*last
;
1540 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1546 for_each_netdev(net
, dev
) {
1547 err
= call_netdevice_notifier(nb
, NETDEV_REGISTER
, dev
);
1548 err
= notifier_to_errno(err
);
1552 if (!(dev
->flags
& IFF_UP
))
1555 call_netdevice_notifier(nb
, NETDEV_UP
, dev
);
1566 for_each_netdev(net
, dev
) {
1570 if (dev
->flags
& IFF_UP
) {
1571 call_netdevice_notifier(nb
, NETDEV_GOING_DOWN
,
1573 call_netdevice_notifier(nb
, NETDEV_DOWN
, dev
);
1575 call_netdevice_notifier(nb
, NETDEV_UNREGISTER
, dev
);
1580 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1583 EXPORT_SYMBOL(register_netdevice_notifier
);
1586 * unregister_netdevice_notifier - unregister a network notifier block
1589 * Unregister a notifier previously registered by
1590 * register_netdevice_notifier(). The notifier is unlinked into the
1591 * kernel structures and may then be reused. A negative errno code
1592 * is returned on a failure.
1594 * After unregistering unregister and down device events are synthesized
1595 * for all devices on the device list to the removed notifier to remove
1596 * the need for special case cleanup code.
1599 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1601 struct net_device
*dev
;
1606 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1611 for_each_netdev(net
, dev
) {
1612 if (dev
->flags
& IFF_UP
) {
1613 call_netdevice_notifier(nb
, NETDEV_GOING_DOWN
,
1615 call_netdevice_notifier(nb
, NETDEV_DOWN
, dev
);
1617 call_netdevice_notifier(nb
, NETDEV_UNREGISTER
, dev
);
1624 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1627 * call_netdevice_notifiers_info - call all network notifier blocks
1628 * @val: value passed unmodified to notifier function
1629 * @dev: net_device pointer passed unmodified to notifier function
1630 * @info: notifier information data
1632 * Call all network notifier blocks. Parameters and return value
1633 * are as for raw_notifier_call_chain().
1636 static int call_netdevice_notifiers_info(unsigned long val
,
1637 struct net_device
*dev
,
1638 struct netdev_notifier_info
*info
)
1641 netdev_notifier_info_init(info
, dev
);
1642 return raw_notifier_call_chain(&netdev_chain
, val
, info
);
1646 * call_netdevice_notifiers - call all network notifier blocks
1647 * @val: value passed unmodified to notifier function
1648 * @dev: net_device pointer passed unmodified to notifier function
1650 * Call all network notifier blocks. Parameters and return value
1651 * are as for raw_notifier_call_chain().
1654 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1656 struct netdev_notifier_info info
;
1658 return call_netdevice_notifiers_info(val
, dev
, &info
);
1660 EXPORT_SYMBOL(call_netdevice_notifiers
);
1662 #ifdef CONFIG_NET_INGRESS
1663 static struct static_key ingress_needed __read_mostly
;
1665 void net_inc_ingress_queue(void)
1667 static_key_slow_inc(&ingress_needed
);
1669 EXPORT_SYMBOL_GPL(net_inc_ingress_queue
);
1671 void net_dec_ingress_queue(void)
1673 static_key_slow_dec(&ingress_needed
);
1675 EXPORT_SYMBOL_GPL(net_dec_ingress_queue
);
1678 static struct static_key netstamp_needed __read_mostly
;
1679 #ifdef HAVE_JUMP_LABEL
1680 /* We are not allowed to call static_key_slow_dec() from irq context
1681 * If net_disable_timestamp() is called from irq context, defer the
1682 * static_key_slow_dec() calls.
1684 static atomic_t netstamp_needed_deferred
;
1687 void net_enable_timestamp(void)
1689 #ifdef HAVE_JUMP_LABEL
1690 int deferred
= atomic_xchg(&netstamp_needed_deferred
, 0);
1694 static_key_slow_dec(&netstamp_needed
);
1698 static_key_slow_inc(&netstamp_needed
);
1700 EXPORT_SYMBOL(net_enable_timestamp
);
1702 void net_disable_timestamp(void)
1704 #ifdef HAVE_JUMP_LABEL
1705 if (in_interrupt()) {
1706 atomic_inc(&netstamp_needed_deferred
);
1710 static_key_slow_dec(&netstamp_needed
);
1712 EXPORT_SYMBOL(net_disable_timestamp
);
1714 static inline void net_timestamp_set(struct sk_buff
*skb
)
1716 skb
->tstamp
.tv64
= 0;
1717 if (static_key_false(&netstamp_needed
))
1718 __net_timestamp(skb
);
1721 #define net_timestamp_check(COND, SKB) \
1722 if (static_key_false(&netstamp_needed)) { \
1723 if ((COND) && !(SKB)->tstamp.tv64) \
1724 __net_timestamp(SKB); \
1727 bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb)
1731 if (!(dev
->flags
& IFF_UP
))
1734 len
= dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
;
1735 if (skb
->len
<= len
)
1738 /* if TSO is enabled, we don't care about the length as the packet
1739 * could be forwarded without being segmented before
1741 if (skb_is_gso(skb
))
1746 EXPORT_SYMBOL_GPL(is_skb_forwardable
);
1748 int __dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1750 if (skb_orphan_frags(skb
, GFP_ATOMIC
) ||
1751 unlikely(!is_skb_forwardable(dev
, skb
))) {
1752 atomic_long_inc(&dev
->rx_dropped
);
1757 skb_scrub_packet(skb
, true);
1759 skb
->protocol
= eth_type_trans(skb
, dev
);
1760 skb_postpull_rcsum(skb
, eth_hdr(skb
), ETH_HLEN
);
1764 EXPORT_SYMBOL_GPL(__dev_forward_skb
);
1767 * dev_forward_skb - loopback an skb to another netif
1769 * @dev: destination network device
1770 * @skb: buffer to forward
1773 * NET_RX_SUCCESS (no congestion)
1774 * NET_RX_DROP (packet was dropped, but freed)
1776 * dev_forward_skb can be used for injecting an skb from the
1777 * start_xmit function of one device into the receive queue
1778 * of another device.
1780 * The receiving device may be in another namespace, so
1781 * we have to clear all information in the skb that could
1782 * impact namespace isolation.
1784 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1786 return __dev_forward_skb(dev
, skb
) ?: netif_rx_internal(skb
);
1788 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1790 static inline int deliver_skb(struct sk_buff
*skb
,
1791 struct packet_type
*pt_prev
,
1792 struct net_device
*orig_dev
)
1794 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
1796 atomic_inc(&skb
->users
);
1797 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1800 static inline void deliver_ptype_list_skb(struct sk_buff
*skb
,
1801 struct packet_type
**pt
,
1802 struct net_device
*orig_dev
,
1804 struct list_head
*ptype_list
)
1806 struct packet_type
*ptype
, *pt_prev
= *pt
;
1808 list_for_each_entry_rcu(ptype
, ptype_list
, list
) {
1809 if (ptype
->type
!= type
)
1812 deliver_skb(skb
, pt_prev
, orig_dev
);
1818 static inline bool skb_loop_sk(struct packet_type
*ptype
, struct sk_buff
*skb
)
1820 if (!ptype
->af_packet_priv
|| !skb
->sk
)
1823 if (ptype
->id_match
)
1824 return ptype
->id_match(ptype
, skb
->sk
);
1825 else if ((struct sock
*)ptype
->af_packet_priv
== skb
->sk
)
1832 * Support routine. Sends outgoing frames to any network
1833 * taps currently in use.
1836 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1838 struct packet_type
*ptype
;
1839 struct sk_buff
*skb2
= NULL
;
1840 struct packet_type
*pt_prev
= NULL
;
1841 struct list_head
*ptype_list
= &ptype_all
;
1845 list_for_each_entry_rcu(ptype
, ptype_list
, list
) {
1846 /* Never send packets back to the socket
1847 * they originated from - MvS (miquels@drinkel.ow.org)
1849 if (skb_loop_sk(ptype
, skb
))
1853 deliver_skb(skb2
, pt_prev
, skb
->dev
);
1858 /* need to clone skb, done only once */
1859 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1863 net_timestamp_set(skb2
);
1865 /* skb->nh should be correctly
1866 * set by sender, so that the second statement is
1867 * just protection against buggy protocols.
1869 skb_reset_mac_header(skb2
);
1871 if (skb_network_header(skb2
) < skb2
->data
||
1872 skb_network_header(skb2
) > skb_tail_pointer(skb2
)) {
1873 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1874 ntohs(skb2
->protocol
),
1876 skb_reset_network_header(skb2
);
1879 skb2
->transport_header
= skb2
->network_header
;
1880 skb2
->pkt_type
= PACKET_OUTGOING
;
1884 if (ptype_list
== &ptype_all
) {
1885 ptype_list
= &dev
->ptype_all
;
1890 pt_prev
->func(skb2
, skb
->dev
, pt_prev
, skb
->dev
);
1895 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1896 * @dev: Network device
1897 * @txq: number of queues available
1899 * If real_num_tx_queues is changed the tc mappings may no longer be
1900 * valid. To resolve this verify the tc mapping remains valid and if
1901 * not NULL the mapping. With no priorities mapping to this
1902 * offset/count pair it will no longer be used. In the worst case TC0
1903 * is invalid nothing can be done so disable priority mappings. If is
1904 * expected that drivers will fix this mapping if they can before
1905 * calling netif_set_real_num_tx_queues.
1907 static void netif_setup_tc(struct net_device
*dev
, unsigned int txq
)
1910 struct netdev_tc_txq
*tc
= &dev
->tc_to_txq
[0];
1912 /* If TC0 is invalidated disable TC mapping */
1913 if (tc
->offset
+ tc
->count
> txq
) {
1914 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1919 /* Invalidated prio to tc mappings set to TC0 */
1920 for (i
= 1; i
< TC_BITMASK
+ 1; i
++) {
1921 int q
= netdev_get_prio_tc_map(dev
, i
);
1923 tc
= &dev
->tc_to_txq
[q
];
1924 if (tc
->offset
+ tc
->count
> txq
) {
1925 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1927 netdev_set_prio_tc_map(dev
, i
, 0);
1933 static DEFINE_MUTEX(xps_map_mutex
);
1934 #define xmap_dereference(P) \
1935 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1937 static struct xps_map
*remove_xps_queue(struct xps_dev_maps
*dev_maps
,
1940 struct xps_map
*map
= NULL
;
1944 map
= xmap_dereference(dev_maps
->cpu_map
[cpu
]);
1946 for (pos
= 0; map
&& pos
< map
->len
; pos
++) {
1947 if (map
->queues
[pos
] == index
) {
1949 map
->queues
[pos
] = map
->queues
[--map
->len
];
1951 RCU_INIT_POINTER(dev_maps
->cpu_map
[cpu
], NULL
);
1952 kfree_rcu(map
, rcu
);
1962 static void netif_reset_xps_queues_gt(struct net_device
*dev
, u16 index
)
1964 struct xps_dev_maps
*dev_maps
;
1966 bool active
= false;
1968 mutex_lock(&xps_map_mutex
);
1969 dev_maps
= xmap_dereference(dev
->xps_maps
);
1974 for_each_possible_cpu(cpu
) {
1975 for (i
= index
; i
< dev
->num_tx_queues
; i
++) {
1976 if (!remove_xps_queue(dev_maps
, cpu
, i
))
1979 if (i
== dev
->num_tx_queues
)
1984 RCU_INIT_POINTER(dev
->xps_maps
, NULL
);
1985 kfree_rcu(dev_maps
, rcu
);
1988 for (i
= index
; i
< dev
->num_tx_queues
; i
++)
1989 netdev_queue_numa_node_write(netdev_get_tx_queue(dev
, i
),
1993 mutex_unlock(&xps_map_mutex
);
1996 static struct xps_map
*expand_xps_map(struct xps_map
*map
,
1999 struct xps_map
*new_map
;
2000 int alloc_len
= XPS_MIN_MAP_ALLOC
;
2003 for (pos
= 0; map
&& pos
< map
->len
; pos
++) {
2004 if (map
->queues
[pos
] != index
)
2009 /* Need to add queue to this CPU's existing map */
2011 if (pos
< map
->alloc_len
)
2014 alloc_len
= map
->alloc_len
* 2;
2017 /* Need to allocate new map to store queue on this CPU's map */
2018 new_map
= kzalloc_node(XPS_MAP_SIZE(alloc_len
), GFP_KERNEL
,
2023 for (i
= 0; i
< pos
; i
++)
2024 new_map
->queues
[i
] = map
->queues
[i
];
2025 new_map
->alloc_len
= alloc_len
;
2031 int netif_set_xps_queue(struct net_device
*dev
, const struct cpumask
*mask
,
2034 struct xps_dev_maps
*dev_maps
, *new_dev_maps
= NULL
;
2035 struct xps_map
*map
, *new_map
;
2036 int maps_sz
= max_t(unsigned int, XPS_DEV_MAPS_SIZE
, L1_CACHE_BYTES
);
2037 int cpu
, numa_node_id
= -2;
2038 bool active
= false;
2040 mutex_lock(&xps_map_mutex
);
2042 dev_maps
= xmap_dereference(dev
->xps_maps
);
2044 /* allocate memory for queue storage */
2045 for_each_online_cpu(cpu
) {
2046 if (!cpumask_test_cpu(cpu
, mask
))
2050 new_dev_maps
= kzalloc(maps_sz
, GFP_KERNEL
);
2051 if (!new_dev_maps
) {
2052 mutex_unlock(&xps_map_mutex
);
2056 map
= dev_maps
? xmap_dereference(dev_maps
->cpu_map
[cpu
]) :
2059 map
= expand_xps_map(map
, cpu
, index
);
2063 RCU_INIT_POINTER(new_dev_maps
->cpu_map
[cpu
], map
);
2067 goto out_no_new_maps
;
2069 for_each_possible_cpu(cpu
) {
2070 if (cpumask_test_cpu(cpu
, mask
) && cpu_online(cpu
)) {
2071 /* add queue to CPU maps */
2074 map
= xmap_dereference(new_dev_maps
->cpu_map
[cpu
]);
2075 while ((pos
< map
->len
) && (map
->queues
[pos
] != index
))
2078 if (pos
== map
->len
)
2079 map
->queues
[map
->len
++] = index
;
2081 if (numa_node_id
== -2)
2082 numa_node_id
= cpu_to_node(cpu
);
2083 else if (numa_node_id
!= cpu_to_node(cpu
))
2086 } else if (dev_maps
) {
2087 /* fill in the new device map from the old device map */
2088 map
= xmap_dereference(dev_maps
->cpu_map
[cpu
]);
2089 RCU_INIT_POINTER(new_dev_maps
->cpu_map
[cpu
], map
);
2094 rcu_assign_pointer(dev
->xps_maps
, new_dev_maps
);
2096 /* Cleanup old maps */
2098 for_each_possible_cpu(cpu
) {
2099 new_map
= xmap_dereference(new_dev_maps
->cpu_map
[cpu
]);
2100 map
= xmap_dereference(dev_maps
->cpu_map
[cpu
]);
2101 if (map
&& map
!= new_map
)
2102 kfree_rcu(map
, rcu
);
2105 kfree_rcu(dev_maps
, rcu
);
2108 dev_maps
= new_dev_maps
;
2112 /* update Tx queue numa node */
2113 netdev_queue_numa_node_write(netdev_get_tx_queue(dev
, index
),
2114 (numa_node_id
>= 0) ? numa_node_id
:
2120 /* removes queue from unused CPUs */
2121 for_each_possible_cpu(cpu
) {
2122 if (cpumask_test_cpu(cpu
, mask
) && cpu_online(cpu
))
2125 if (remove_xps_queue(dev_maps
, cpu
, index
))
2129 /* free map if not active */
2131 RCU_INIT_POINTER(dev
->xps_maps
, NULL
);
2132 kfree_rcu(dev_maps
, rcu
);
2136 mutex_unlock(&xps_map_mutex
);
2140 /* remove any maps that we added */
2141 for_each_possible_cpu(cpu
) {
2142 new_map
= xmap_dereference(new_dev_maps
->cpu_map
[cpu
]);
2143 map
= dev_maps
? xmap_dereference(dev_maps
->cpu_map
[cpu
]) :
2145 if (new_map
&& new_map
!= map
)
2149 mutex_unlock(&xps_map_mutex
);
2151 kfree(new_dev_maps
);
2154 EXPORT_SYMBOL(netif_set_xps_queue
);
2158 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2159 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2161 int netif_set_real_num_tx_queues(struct net_device
*dev
, unsigned int txq
)
2165 if (txq
< 1 || txq
> dev
->num_tx_queues
)
2168 if (dev
->reg_state
== NETREG_REGISTERED
||
2169 dev
->reg_state
== NETREG_UNREGISTERING
) {
2172 rc
= netdev_queue_update_kobjects(dev
, dev
->real_num_tx_queues
,
2178 netif_setup_tc(dev
, txq
);
2180 if (txq
< dev
->real_num_tx_queues
) {
2181 qdisc_reset_all_tx_gt(dev
, txq
);
2183 netif_reset_xps_queues_gt(dev
, txq
);
2188 dev
->real_num_tx_queues
= txq
;
2191 EXPORT_SYMBOL(netif_set_real_num_tx_queues
);
2195 * netif_set_real_num_rx_queues - set actual number of RX queues used
2196 * @dev: Network device
2197 * @rxq: Actual number of RX queues
2199 * This must be called either with the rtnl_lock held or before
2200 * registration of the net device. Returns 0 on success, or a
2201 * negative error code. If called before registration, it always
2204 int netif_set_real_num_rx_queues(struct net_device
*dev
, unsigned int rxq
)
2208 if (rxq
< 1 || rxq
> dev
->num_rx_queues
)
2211 if (dev
->reg_state
== NETREG_REGISTERED
) {
2214 rc
= net_rx_queue_update_kobjects(dev
, dev
->real_num_rx_queues
,
2220 dev
->real_num_rx_queues
= rxq
;
2223 EXPORT_SYMBOL(netif_set_real_num_rx_queues
);
2227 * netif_get_num_default_rss_queues - default number of RSS queues
2229 * This routine should set an upper limit on the number of RSS queues
2230 * used by default by multiqueue devices.
2232 int netif_get_num_default_rss_queues(void)
2234 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES
, num_online_cpus());
2236 EXPORT_SYMBOL(netif_get_num_default_rss_queues
);
2238 static inline void __netif_reschedule(struct Qdisc
*q
)
2240 struct softnet_data
*sd
;
2241 unsigned long flags
;
2243 local_irq_save(flags
);
2244 sd
= this_cpu_ptr(&softnet_data
);
2245 q
->next_sched
= NULL
;
2246 *sd
->output_queue_tailp
= q
;
2247 sd
->output_queue_tailp
= &q
->next_sched
;
2248 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
2249 local_irq_restore(flags
);
2252 void __netif_schedule(struct Qdisc
*q
)
2254 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
2255 __netif_reschedule(q
);
2257 EXPORT_SYMBOL(__netif_schedule
);
2259 struct dev_kfree_skb_cb
{
2260 enum skb_free_reason reason
;
2263 static struct dev_kfree_skb_cb
*get_kfree_skb_cb(const struct sk_buff
*skb
)
2265 return (struct dev_kfree_skb_cb
*)skb
->cb
;
2268 void netif_schedule_queue(struct netdev_queue
*txq
)
2271 if (!(txq
->state
& QUEUE_STATE_ANY_XOFF
)) {
2272 struct Qdisc
*q
= rcu_dereference(txq
->qdisc
);
2274 __netif_schedule(q
);
2278 EXPORT_SYMBOL(netif_schedule_queue
);
2281 * netif_wake_subqueue - allow sending packets on subqueue
2282 * @dev: network device
2283 * @queue_index: sub queue index
2285 * Resume individual transmit queue of a device with multiple transmit queues.
2287 void netif_wake_subqueue(struct net_device
*dev
, u16 queue_index
)
2289 struct netdev_queue
*txq
= netdev_get_tx_queue(dev
, queue_index
);
2291 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF
, &txq
->state
)) {
2295 q
= rcu_dereference(txq
->qdisc
);
2296 __netif_schedule(q
);
2300 EXPORT_SYMBOL(netif_wake_subqueue
);
2302 void netif_tx_wake_queue(struct netdev_queue
*dev_queue
)
2304 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF
, &dev_queue
->state
)) {
2308 q
= rcu_dereference(dev_queue
->qdisc
);
2309 __netif_schedule(q
);
2313 EXPORT_SYMBOL(netif_tx_wake_queue
);
2315 void __dev_kfree_skb_irq(struct sk_buff
*skb
, enum skb_free_reason reason
)
2317 unsigned long flags
;
2319 if (likely(atomic_read(&skb
->users
) == 1)) {
2321 atomic_set(&skb
->users
, 0);
2322 } else if (likely(!atomic_dec_and_test(&skb
->users
))) {
2325 get_kfree_skb_cb(skb
)->reason
= reason
;
2326 local_irq_save(flags
);
2327 skb
->next
= __this_cpu_read(softnet_data
.completion_queue
);
2328 __this_cpu_write(softnet_data
.completion_queue
, skb
);
2329 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
2330 local_irq_restore(flags
);
2332 EXPORT_SYMBOL(__dev_kfree_skb_irq
);
2334 void __dev_kfree_skb_any(struct sk_buff
*skb
, enum skb_free_reason reason
)
2336 if (in_irq() || irqs_disabled())
2337 __dev_kfree_skb_irq(skb
, reason
);
2341 EXPORT_SYMBOL(__dev_kfree_skb_any
);
2345 * netif_device_detach - mark device as removed
2346 * @dev: network device
2348 * Mark device as removed from system and therefore no longer available.
2350 void netif_device_detach(struct net_device
*dev
)
2352 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
2353 netif_running(dev
)) {
2354 netif_tx_stop_all_queues(dev
);
2357 EXPORT_SYMBOL(netif_device_detach
);
2360 * netif_device_attach - mark device as attached
2361 * @dev: network device
2363 * Mark device as attached from system and restart if needed.
2365 void netif_device_attach(struct net_device
*dev
)
2367 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
2368 netif_running(dev
)) {
2369 netif_tx_wake_all_queues(dev
);
2370 __netdev_watchdog_up(dev
);
2373 EXPORT_SYMBOL(netif_device_attach
);
2376 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2377 * to be used as a distribution range.
2379 u16
__skb_tx_hash(const struct net_device
*dev
, struct sk_buff
*skb
,
2380 unsigned int num_tx_queues
)
2384 u16 qcount
= num_tx_queues
;
2386 if (skb_rx_queue_recorded(skb
)) {
2387 hash
= skb_get_rx_queue(skb
);
2388 while (unlikely(hash
>= num_tx_queues
))
2389 hash
-= num_tx_queues
;
2394 u8 tc
= netdev_get_prio_tc_map(dev
, skb
->priority
);
2395 qoffset
= dev
->tc_to_txq
[tc
].offset
;
2396 qcount
= dev
->tc_to_txq
[tc
].count
;
2399 return (u16
) reciprocal_scale(skb_get_hash(skb
), qcount
) + qoffset
;
2401 EXPORT_SYMBOL(__skb_tx_hash
);
2403 static void skb_warn_bad_offload(const struct sk_buff
*skb
)
2405 static const netdev_features_t null_features
= 0;
2406 struct net_device
*dev
= skb
->dev
;
2407 const char *name
= "";
2409 if (!net_ratelimit())
2413 if (dev
->dev
.parent
)
2414 name
= dev_driver_string(dev
->dev
.parent
);
2416 name
= netdev_name(dev
);
2418 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2419 "gso_type=%d ip_summed=%d\n",
2420 name
, dev
? &dev
->features
: &null_features
,
2421 skb
->sk
? &skb
->sk
->sk_route_caps
: &null_features
,
2422 skb
->len
, skb
->data_len
, skb_shinfo(skb
)->gso_size
,
2423 skb_shinfo(skb
)->gso_type
, skb
->ip_summed
);
2427 * Invalidate hardware checksum when packet is to be mangled, and
2428 * complete checksum manually on outgoing path.
2430 int skb_checksum_help(struct sk_buff
*skb
)
2433 int ret
= 0, offset
;
2435 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
2436 goto out_set_summed
;
2438 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
2439 skb_warn_bad_offload(skb
);
2443 /* Before computing a checksum, we should make sure no frag could
2444 * be modified by an external entity : checksum could be wrong.
2446 if (skb_has_shared_frag(skb
)) {
2447 ret
= __skb_linearize(skb
);
2452 offset
= skb_checksum_start_offset(skb
);
2453 BUG_ON(offset
>= skb_headlen(skb
));
2454 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
2456 offset
+= skb
->csum_offset
;
2457 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
2459 if (skb_cloned(skb
) &&
2460 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
2461 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2466 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
2468 skb
->ip_summed
= CHECKSUM_NONE
;
2472 EXPORT_SYMBOL(skb_checksum_help
);
2474 __be16
skb_network_protocol(struct sk_buff
*skb
, int *depth
)
2476 __be16 type
= skb
->protocol
;
2478 /* Tunnel gso handlers can set protocol to ethernet. */
2479 if (type
== htons(ETH_P_TEB
)) {
2482 if (unlikely(!pskb_may_pull(skb
, sizeof(struct ethhdr
))))
2485 eth
= (struct ethhdr
*)skb_mac_header(skb
);
2486 type
= eth
->h_proto
;
2489 return __vlan_get_protocol(skb
, type
, depth
);
2493 * skb_mac_gso_segment - mac layer segmentation handler.
2494 * @skb: buffer to segment
2495 * @features: features for the output path (see dev->features)
2497 struct sk_buff
*skb_mac_gso_segment(struct sk_buff
*skb
,
2498 netdev_features_t features
)
2500 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
2501 struct packet_offload
*ptype
;
2502 int vlan_depth
= skb
->mac_len
;
2503 __be16 type
= skb_network_protocol(skb
, &vlan_depth
);
2505 if (unlikely(!type
))
2506 return ERR_PTR(-EINVAL
);
2508 __skb_pull(skb
, vlan_depth
);
2511 list_for_each_entry_rcu(ptype
, &offload_base
, list
) {
2512 if (ptype
->type
== type
&& ptype
->callbacks
.gso_segment
) {
2513 segs
= ptype
->callbacks
.gso_segment(skb
, features
);
2519 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
2523 EXPORT_SYMBOL(skb_mac_gso_segment
);
2526 /* openvswitch calls this on rx path, so we need a different check.
2528 static inline bool skb_needs_check(struct sk_buff
*skb
, bool tx_path
)
2531 return skb
->ip_summed
!= CHECKSUM_PARTIAL
;
2533 return skb
->ip_summed
== CHECKSUM_NONE
;
2537 * __skb_gso_segment - Perform segmentation on skb.
2538 * @skb: buffer to segment
2539 * @features: features for the output path (see dev->features)
2540 * @tx_path: whether it is called in TX path
2542 * This function segments the given skb and returns a list of segments.
2544 * It may return NULL if the skb requires no segmentation. This is
2545 * only possible when GSO is used for verifying header integrity.
2547 struct sk_buff
*__skb_gso_segment(struct sk_buff
*skb
,
2548 netdev_features_t features
, bool tx_path
)
2550 if (unlikely(skb_needs_check(skb
, tx_path
))) {
2553 skb_warn_bad_offload(skb
);
2555 err
= skb_cow_head(skb
, 0);
2557 return ERR_PTR(err
);
2560 SKB_GSO_CB(skb
)->mac_offset
= skb_headroom(skb
);
2561 SKB_GSO_CB(skb
)->encap_level
= 0;
2563 skb_reset_mac_header(skb
);
2564 skb_reset_mac_len(skb
);
2566 return skb_mac_gso_segment(skb
, features
);
2568 EXPORT_SYMBOL(__skb_gso_segment
);
2570 /* Take action when hardware reception checksum errors are detected. */
2572 void netdev_rx_csum_fault(struct net_device
*dev
)
2574 if (net_ratelimit()) {
2575 pr_err("%s: hw csum failure\n", dev
? dev
->name
: "<unknown>");
2579 EXPORT_SYMBOL(netdev_rx_csum_fault
);
2582 /* Actually, we should eliminate this check as soon as we know, that:
2583 * 1. IOMMU is present and allows to map all the memory.
2584 * 2. No high memory really exists on this machine.
2587 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
2589 #ifdef CONFIG_HIGHMEM
2591 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
2592 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2593 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2594 if (PageHighMem(skb_frag_page(frag
)))
2599 if (PCI_DMA_BUS_IS_PHYS
) {
2600 struct device
*pdev
= dev
->dev
.parent
;
2604 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2605 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2606 dma_addr_t addr
= page_to_phys(skb_frag_page(frag
));
2607 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
2615 /* If MPLS offload request, verify we are testing hardware MPLS features
2616 * instead of standard features for the netdev.
2618 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
2619 static netdev_features_t
net_mpls_features(struct sk_buff
*skb
,
2620 netdev_features_t features
,
2623 if (eth_p_mpls(type
))
2624 features
&= skb
->dev
->mpls_features
;
2629 static netdev_features_t
net_mpls_features(struct sk_buff
*skb
,
2630 netdev_features_t features
,
2637 static netdev_features_t
harmonize_features(struct sk_buff
*skb
,
2638 netdev_features_t features
)
2643 type
= skb_network_protocol(skb
, &tmp
);
2644 features
= net_mpls_features(skb
, features
, type
);
2646 if (skb
->ip_summed
!= CHECKSUM_NONE
&&
2647 !can_checksum_protocol(features
, type
)) {
2648 features
&= ~NETIF_F_ALL_CSUM
;
2649 } else if (illegal_highdma(skb
->dev
, skb
)) {
2650 features
&= ~NETIF_F_SG
;
2656 netdev_features_t
passthru_features_check(struct sk_buff
*skb
,
2657 struct net_device
*dev
,
2658 netdev_features_t features
)
2662 EXPORT_SYMBOL(passthru_features_check
);
2664 static netdev_features_t
dflt_features_check(const struct sk_buff
*skb
,
2665 struct net_device
*dev
,
2666 netdev_features_t features
)
2668 return vlan_features_check(skb
, features
);
2671 netdev_features_t
netif_skb_features(struct sk_buff
*skb
)
2673 struct net_device
*dev
= skb
->dev
;
2674 netdev_features_t features
= dev
->features
;
2675 u16 gso_segs
= skb_shinfo(skb
)->gso_segs
;
2677 if (gso_segs
> dev
->gso_max_segs
|| gso_segs
< dev
->gso_min_segs
)
2678 features
&= ~NETIF_F_GSO_MASK
;
2680 /* If encapsulation offload request, verify we are testing
2681 * hardware encapsulation features instead of standard
2682 * features for the netdev
2684 if (skb
->encapsulation
)
2685 features
&= dev
->hw_enc_features
;
2687 if (skb_vlan_tagged(skb
))
2688 features
= netdev_intersect_features(features
,
2689 dev
->vlan_features
|
2690 NETIF_F_HW_VLAN_CTAG_TX
|
2691 NETIF_F_HW_VLAN_STAG_TX
);
2693 if (dev
->netdev_ops
->ndo_features_check
)
2694 features
&= dev
->netdev_ops
->ndo_features_check(skb
, dev
,
2697 features
&= dflt_features_check(skb
, dev
, features
);
2699 return harmonize_features(skb
, features
);
2701 EXPORT_SYMBOL(netif_skb_features
);
2703 static int xmit_one(struct sk_buff
*skb
, struct net_device
*dev
,
2704 struct netdev_queue
*txq
, bool more
)
2709 if (!list_empty(&ptype_all
) || !list_empty(&dev
->ptype_all
))
2710 dev_queue_xmit_nit(skb
, dev
);
2713 trace_net_dev_start_xmit(skb
, dev
);
2714 rc
= netdev_start_xmit(skb
, dev
, txq
, more
);
2715 trace_net_dev_xmit(skb
, rc
, dev
, len
);
2720 struct sk_buff
*dev_hard_start_xmit(struct sk_buff
*first
, struct net_device
*dev
,
2721 struct netdev_queue
*txq
, int *ret
)
2723 struct sk_buff
*skb
= first
;
2724 int rc
= NETDEV_TX_OK
;
2727 struct sk_buff
*next
= skb
->next
;
2730 rc
= xmit_one(skb
, dev
, txq
, next
!= NULL
);
2731 if (unlikely(!dev_xmit_complete(rc
))) {
2737 if (netif_xmit_stopped(txq
) && skb
) {
2738 rc
= NETDEV_TX_BUSY
;
2748 static struct sk_buff
*validate_xmit_vlan(struct sk_buff
*skb
,
2749 netdev_features_t features
)
2751 if (skb_vlan_tag_present(skb
) &&
2752 !vlan_hw_offload_capable(features
, skb
->vlan_proto
))
2753 skb
= __vlan_hwaccel_push_inside(skb
);
2757 static struct sk_buff
*validate_xmit_skb(struct sk_buff
*skb
, struct net_device
*dev
)
2759 netdev_features_t features
;
2764 features
= netif_skb_features(skb
);
2765 skb
= validate_xmit_vlan(skb
, features
);
2769 if (netif_needs_gso(skb
, features
)) {
2770 struct sk_buff
*segs
;
2772 segs
= skb_gso_segment(skb
, features
);
2780 if (skb_needs_linearize(skb
, features
) &&
2781 __skb_linearize(skb
))
2784 /* If packet is not checksummed and device does not
2785 * support checksumming for this protocol, complete
2786 * checksumming here.
2788 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2789 if (skb
->encapsulation
)
2790 skb_set_inner_transport_header(skb
,
2791 skb_checksum_start_offset(skb
));
2793 skb_set_transport_header(skb
,
2794 skb_checksum_start_offset(skb
));
2795 if (!(features
& NETIF_F_ALL_CSUM
) &&
2796 skb_checksum_help(skb
))
2809 struct sk_buff
*validate_xmit_skb_list(struct sk_buff
*skb
, struct net_device
*dev
)
2811 struct sk_buff
*next
, *head
= NULL
, *tail
;
2813 for (; skb
!= NULL
; skb
= next
) {
2817 /* in case skb wont be segmented, point to itself */
2820 skb
= validate_xmit_skb(skb
, dev
);
2828 /* If skb was segmented, skb->prev points to
2829 * the last segment. If not, it still contains skb.
2836 static void qdisc_pkt_len_init(struct sk_buff
*skb
)
2838 const struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
2840 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
2842 /* To get more precise estimation of bytes sent on wire,
2843 * we add to pkt_len the headers size of all segments
2845 if (shinfo
->gso_size
) {
2846 unsigned int hdr_len
;
2847 u16 gso_segs
= shinfo
->gso_segs
;
2849 /* mac layer + network layer */
2850 hdr_len
= skb_transport_header(skb
) - skb_mac_header(skb
);
2852 /* + transport layer */
2853 if (likely(shinfo
->gso_type
& (SKB_GSO_TCPV4
| SKB_GSO_TCPV6
)))
2854 hdr_len
+= tcp_hdrlen(skb
);
2856 hdr_len
+= sizeof(struct udphdr
);
2858 if (shinfo
->gso_type
& SKB_GSO_DODGY
)
2859 gso_segs
= DIV_ROUND_UP(skb
->len
- hdr_len
,
2862 qdisc_skb_cb(skb
)->pkt_len
+= (gso_segs
- 1) * hdr_len
;
2866 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2867 struct net_device
*dev
,
2868 struct netdev_queue
*txq
)
2870 spinlock_t
*root_lock
= qdisc_lock(q
);
2874 qdisc_pkt_len_init(skb
);
2875 qdisc_calculate_pkt_len(skb
, q
);
2877 * Heuristic to force contended enqueues to serialize on a
2878 * separate lock before trying to get qdisc main lock.
2879 * This permits __QDISC___STATE_RUNNING owner to get the lock more
2880 * often and dequeue packets faster.
2882 contended
= qdisc_is_running(q
);
2883 if (unlikely(contended
))
2884 spin_lock(&q
->busylock
);
2886 spin_lock(root_lock
);
2887 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2890 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2891 qdisc_run_begin(q
)) {
2893 * This is a work-conserving queue; there are no old skbs
2894 * waiting to be sent out; and the qdisc is not running -
2895 * xmit the skb directly.
2898 qdisc_bstats_update(q
, skb
);
2900 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
, true)) {
2901 if (unlikely(contended
)) {
2902 spin_unlock(&q
->busylock
);
2909 rc
= NET_XMIT_SUCCESS
;
2911 rc
= q
->enqueue(skb
, q
) & NET_XMIT_MASK
;
2912 if (qdisc_run_begin(q
)) {
2913 if (unlikely(contended
)) {
2914 spin_unlock(&q
->busylock
);
2920 spin_unlock(root_lock
);
2921 if (unlikely(contended
))
2922 spin_unlock(&q
->busylock
);
2926 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2927 static void skb_update_prio(struct sk_buff
*skb
)
2929 struct netprio_map
*map
= rcu_dereference_bh(skb
->dev
->priomap
);
2931 if (!skb
->priority
&& skb
->sk
&& map
) {
2932 unsigned int prioidx
= skb
->sk
->sk_cgrp_prioidx
;
2934 if (prioidx
< map
->priomap_len
)
2935 skb
->priority
= map
->priomap
[prioidx
];
2939 #define skb_update_prio(skb)
2942 DEFINE_PER_CPU(int, xmit_recursion
);
2943 EXPORT_SYMBOL(xmit_recursion
);
2945 #define RECURSION_LIMIT 10
2948 * dev_loopback_xmit - loop back @skb
2949 * @net: network namespace this loopback is happening in
2950 * @sk: sk needed to be a netfilter okfn
2951 * @skb: buffer to transmit
2953 int dev_loopback_xmit(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
2955 skb_reset_mac_header(skb
);
2956 __skb_pull(skb
, skb_network_offset(skb
));
2957 skb
->pkt_type
= PACKET_LOOPBACK
;
2958 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
2959 WARN_ON(!skb_dst(skb
));
2964 EXPORT_SYMBOL(dev_loopback_xmit
);
2966 static inline int get_xps_queue(struct net_device
*dev
, struct sk_buff
*skb
)
2969 struct xps_dev_maps
*dev_maps
;
2970 struct xps_map
*map
;
2971 int queue_index
= -1;
2974 dev_maps
= rcu_dereference(dev
->xps_maps
);
2976 map
= rcu_dereference(
2977 dev_maps
->cpu_map
[skb
->sender_cpu
- 1]);
2980 queue_index
= map
->queues
[0];
2982 queue_index
= map
->queues
[reciprocal_scale(skb_get_hash(skb
),
2984 if (unlikely(queue_index
>= dev
->real_num_tx_queues
))
2996 static u16
__netdev_pick_tx(struct net_device
*dev
, struct sk_buff
*skb
)
2998 struct sock
*sk
= skb
->sk
;
2999 int queue_index
= sk_tx_queue_get(sk
);
3001 if (queue_index
< 0 || skb
->ooo_okay
||
3002 queue_index
>= dev
->real_num_tx_queues
) {
3003 int new_index
= get_xps_queue(dev
, skb
);
3005 new_index
= skb_tx_hash(dev
, skb
);
3007 if (queue_index
!= new_index
&& sk
&&
3009 rcu_access_pointer(sk
->sk_dst_cache
))
3010 sk_tx_queue_set(sk
, new_index
);
3012 queue_index
= new_index
;
3018 struct netdev_queue
*netdev_pick_tx(struct net_device
*dev
,
3019 struct sk_buff
*skb
,
3022 int queue_index
= 0;
3025 u32 sender_cpu
= skb
->sender_cpu
- 1;
3027 if (sender_cpu
>= (u32
)NR_CPUS
)
3028 skb
->sender_cpu
= raw_smp_processor_id() + 1;
3031 if (dev
->real_num_tx_queues
!= 1) {
3032 const struct net_device_ops
*ops
= dev
->netdev_ops
;
3033 if (ops
->ndo_select_queue
)
3034 queue_index
= ops
->ndo_select_queue(dev
, skb
, accel_priv
,
3037 queue_index
= __netdev_pick_tx(dev
, skb
);
3040 queue_index
= netdev_cap_txqueue(dev
, queue_index
);
3043 skb_set_queue_mapping(skb
, queue_index
);
3044 return netdev_get_tx_queue(dev
, queue_index
);
3048 * __dev_queue_xmit - transmit a buffer
3049 * @skb: buffer to transmit
3050 * @accel_priv: private data used for L2 forwarding offload
3052 * Queue a buffer for transmission to a network device. The caller must
3053 * have set the device and priority and built the buffer before calling
3054 * this function. The function can be called from an interrupt.
3056 * A negative errno code is returned on a failure. A success does not
3057 * guarantee the frame will be transmitted as it may be dropped due
3058 * to congestion or traffic shaping.
3060 * -----------------------------------------------------------------------------------
3061 * I notice this method can also return errors from the queue disciplines,
3062 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3065 * Regardless of the return value, the skb is consumed, so it is currently
3066 * difficult to retry a send to this method. (You can bump the ref count
3067 * before sending to hold a reference for retry if you are careful.)
3069 * When calling this method, interrupts MUST be enabled. This is because
3070 * the BH enable code must have IRQs enabled so that it will not deadlock.
3073 static int __dev_queue_xmit(struct sk_buff
*skb
, void *accel_priv
)
3075 struct net_device
*dev
= skb
->dev
;
3076 struct netdev_queue
*txq
;
3080 skb_reset_mac_header(skb
);
3082 if (unlikely(skb_shinfo(skb
)->tx_flags
& SKBTX_SCHED_TSTAMP
))
3083 __skb_tstamp_tx(skb
, NULL
, skb
->sk
, SCM_TSTAMP_SCHED
);
3085 /* Disable soft irqs for various locks below. Also
3086 * stops preemption for RCU.
3090 skb_update_prio(skb
);
3092 /* If device/qdisc don't need skb->dst, release it right now while
3093 * its hot in this cpu cache.
3095 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
3100 #ifdef CONFIG_NET_SWITCHDEV
3101 /* Don't forward if offload device already forwarded */
3102 if (skb
->offload_fwd_mark
&&
3103 skb
->offload_fwd_mark
== dev
->offload_fwd_mark
) {
3105 rc
= NET_XMIT_SUCCESS
;
3110 txq
= netdev_pick_tx(dev
, skb
, accel_priv
);
3111 q
= rcu_dereference_bh(txq
->qdisc
);
3113 #ifdef CONFIG_NET_CLS_ACT
3114 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
3116 trace_net_dev_queue(skb
);
3118 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
3122 /* The device has no queue. Common case for software devices:
3123 loopback, all the sorts of tunnels...
3125 Really, it is unlikely that netif_tx_lock protection is necessary
3126 here. (f.e. loopback and IP tunnels are clean ignoring statistics
3128 However, it is possible, that they rely on protection
3131 Check this and shot the lock. It is not prone from deadlocks.
3132 Either shot noqueue qdisc, it is even simpler 8)
3134 if (dev
->flags
& IFF_UP
) {
3135 int cpu
= smp_processor_id(); /* ok because BHs are off */
3137 if (txq
->xmit_lock_owner
!= cpu
) {
3139 if (__this_cpu_read(xmit_recursion
) > RECURSION_LIMIT
)
3140 goto recursion_alert
;
3142 skb
= validate_xmit_skb(skb
, dev
);
3146 HARD_TX_LOCK(dev
, txq
, cpu
);
3148 if (!netif_xmit_stopped(txq
)) {
3149 __this_cpu_inc(xmit_recursion
);
3150 skb
= dev_hard_start_xmit(skb
, dev
, txq
, &rc
);
3151 __this_cpu_dec(xmit_recursion
);
3152 if (dev_xmit_complete(rc
)) {
3153 HARD_TX_UNLOCK(dev
, txq
);
3157 HARD_TX_UNLOCK(dev
, txq
);
3158 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3161 /* Recursion is detected! It is possible,
3165 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3172 rcu_read_unlock_bh();
3174 atomic_long_inc(&dev
->tx_dropped
);
3175 kfree_skb_list(skb
);
3178 rcu_read_unlock_bh();
3182 int dev_queue_xmit(struct sk_buff
*skb
)
3184 return __dev_queue_xmit(skb
, NULL
);
3186 EXPORT_SYMBOL(dev_queue_xmit
);
3188 int dev_queue_xmit_accel(struct sk_buff
*skb
, void *accel_priv
)
3190 return __dev_queue_xmit(skb
, accel_priv
);
3192 EXPORT_SYMBOL(dev_queue_xmit_accel
);
3195 /*=======================================================================
3197 =======================================================================*/
3199 int netdev_max_backlog __read_mostly
= 1000;
3200 EXPORT_SYMBOL(netdev_max_backlog
);
3202 int netdev_tstamp_prequeue __read_mostly
= 1;
3203 int netdev_budget __read_mostly
= 300;
3204 int weight_p __read_mostly
= 64; /* old backlog weight */
3206 /* Called with irq disabled */
3207 static inline void ____napi_schedule(struct softnet_data
*sd
,
3208 struct napi_struct
*napi
)
3210 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
3211 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3216 /* One global table that all flow-based protocols share. */
3217 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
3218 EXPORT_SYMBOL(rps_sock_flow_table
);
3219 u32 rps_cpu_mask __read_mostly
;
3220 EXPORT_SYMBOL(rps_cpu_mask
);
3222 struct static_key rps_needed __read_mostly
;
3224 static struct rps_dev_flow
*
3225 set_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
3226 struct rps_dev_flow
*rflow
, u16 next_cpu
)
3228 if (next_cpu
< nr_cpu_ids
) {
3229 #ifdef CONFIG_RFS_ACCEL
3230 struct netdev_rx_queue
*rxqueue
;
3231 struct rps_dev_flow_table
*flow_table
;
3232 struct rps_dev_flow
*old_rflow
;
3237 /* Should we steer this flow to a different hardware queue? */
3238 if (!skb_rx_queue_recorded(skb
) || !dev
->rx_cpu_rmap
||
3239 !(dev
->features
& NETIF_F_NTUPLE
))
3241 rxq_index
= cpu_rmap_lookup_index(dev
->rx_cpu_rmap
, next_cpu
);
3242 if (rxq_index
== skb_get_rx_queue(skb
))
3245 rxqueue
= dev
->_rx
+ rxq_index
;
3246 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3249 flow_id
= skb_get_hash(skb
) & flow_table
->mask
;
3250 rc
= dev
->netdev_ops
->ndo_rx_flow_steer(dev
, skb
,
3251 rxq_index
, flow_id
);
3255 rflow
= &flow_table
->flows
[flow_id
];
3257 if (old_rflow
->filter
== rflow
->filter
)
3258 old_rflow
->filter
= RPS_NO_FILTER
;
3262 per_cpu(softnet_data
, next_cpu
).input_queue_head
;
3265 rflow
->cpu
= next_cpu
;
3270 * get_rps_cpu is called from netif_receive_skb and returns the target
3271 * CPU from the RPS map of the receiving queue for a given skb.
3272 * rcu_read_lock must be held on entry.
3274 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
3275 struct rps_dev_flow
**rflowp
)
3277 const struct rps_sock_flow_table
*sock_flow_table
;
3278 struct netdev_rx_queue
*rxqueue
= dev
->_rx
;
3279 struct rps_dev_flow_table
*flow_table
;
3280 struct rps_map
*map
;
3285 if (skb_rx_queue_recorded(skb
)) {
3286 u16 index
= skb_get_rx_queue(skb
);
3288 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
3289 WARN_ONCE(dev
->real_num_rx_queues
> 1,
3290 "%s received packet on queue %u, but number "
3291 "of RX queues is %u\n",
3292 dev
->name
, index
, dev
->real_num_rx_queues
);
3298 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3300 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3301 map
= rcu_dereference(rxqueue
->rps_map
);
3302 if (!flow_table
&& !map
)
3305 skb_reset_network_header(skb
);
3306 hash
= skb_get_hash(skb
);
3310 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
3311 if (flow_table
&& sock_flow_table
) {
3312 struct rps_dev_flow
*rflow
;
3316 /* First check into global flow table if there is a match */
3317 ident
= sock_flow_table
->ents
[hash
& sock_flow_table
->mask
];
3318 if ((ident
^ hash
) & ~rps_cpu_mask
)
3321 next_cpu
= ident
& rps_cpu_mask
;
3323 /* OK, now we know there is a match,
3324 * we can look at the local (per receive queue) flow table
3326 rflow
= &flow_table
->flows
[hash
& flow_table
->mask
];
3330 * If the desired CPU (where last recvmsg was done) is
3331 * different from current CPU (one in the rx-queue flow
3332 * table entry), switch if one of the following holds:
3333 * - Current CPU is unset (>= nr_cpu_ids).
3334 * - Current CPU is offline.
3335 * - The current CPU's queue tail has advanced beyond the
3336 * last packet that was enqueued using this table entry.
3337 * This guarantees that all previous packets for the flow
3338 * have been dequeued, thus preserving in order delivery.
3340 if (unlikely(tcpu
!= next_cpu
) &&
3341 (tcpu
>= nr_cpu_ids
|| !cpu_online(tcpu
) ||
3342 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
3343 rflow
->last_qtail
)) >= 0)) {
3345 rflow
= set_rps_cpu(dev
, skb
, rflow
, next_cpu
);
3348 if (tcpu
< nr_cpu_ids
&& cpu_online(tcpu
)) {
3358 tcpu
= map
->cpus
[reciprocal_scale(hash
, map
->len
)];
3359 if (cpu_online(tcpu
)) {
3369 #ifdef CONFIG_RFS_ACCEL
3372 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3373 * @dev: Device on which the filter was set
3374 * @rxq_index: RX queue index
3375 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3376 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3378 * Drivers that implement ndo_rx_flow_steer() should periodically call
3379 * this function for each installed filter and remove the filters for
3380 * which it returns %true.
3382 bool rps_may_expire_flow(struct net_device
*dev
, u16 rxq_index
,
3383 u32 flow_id
, u16 filter_id
)
3385 struct netdev_rx_queue
*rxqueue
= dev
->_rx
+ rxq_index
;
3386 struct rps_dev_flow_table
*flow_table
;
3387 struct rps_dev_flow
*rflow
;
3392 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3393 if (flow_table
&& flow_id
<= flow_table
->mask
) {
3394 rflow
= &flow_table
->flows
[flow_id
];
3395 cpu
= ACCESS_ONCE(rflow
->cpu
);
3396 if (rflow
->filter
== filter_id
&& cpu
< nr_cpu_ids
&&
3397 ((int)(per_cpu(softnet_data
, cpu
).input_queue_head
-
3398 rflow
->last_qtail
) <
3399 (int)(10 * flow_table
->mask
)))
3405 EXPORT_SYMBOL(rps_may_expire_flow
);
3407 #endif /* CONFIG_RFS_ACCEL */
3409 /* Called from hardirq (IPI) context */
3410 static void rps_trigger_softirq(void *data
)
3412 struct softnet_data
*sd
= data
;
3414 ____napi_schedule(sd
, &sd
->backlog
);
3418 #endif /* CONFIG_RPS */
3421 * Check if this softnet_data structure is another cpu one
3422 * If yes, queue it to our IPI list and return 1
3425 static int rps_ipi_queued(struct softnet_data
*sd
)
3428 struct softnet_data
*mysd
= this_cpu_ptr(&softnet_data
);
3431 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
3432 mysd
->rps_ipi_list
= sd
;
3434 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3437 #endif /* CONFIG_RPS */
3441 #ifdef CONFIG_NET_FLOW_LIMIT
3442 int netdev_flow_limit_table_len __read_mostly
= (1 << 12);
3445 static bool skb_flow_limit(struct sk_buff
*skb
, unsigned int qlen
)
3447 #ifdef CONFIG_NET_FLOW_LIMIT
3448 struct sd_flow_limit
*fl
;
3449 struct softnet_data
*sd
;
3450 unsigned int old_flow
, new_flow
;
3452 if (qlen
< (netdev_max_backlog
>> 1))
3455 sd
= this_cpu_ptr(&softnet_data
);
3458 fl
= rcu_dereference(sd
->flow_limit
);
3460 new_flow
= skb_get_hash(skb
) & (fl
->num_buckets
- 1);
3461 old_flow
= fl
->history
[fl
->history_head
];
3462 fl
->history
[fl
->history_head
] = new_flow
;
3465 fl
->history_head
&= FLOW_LIMIT_HISTORY
- 1;
3467 if (likely(fl
->buckets
[old_flow
]))
3468 fl
->buckets
[old_flow
]--;
3470 if (++fl
->buckets
[new_flow
] > (FLOW_LIMIT_HISTORY
>> 1)) {
3482 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3483 * queue (may be a remote CPU queue).
3485 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
3486 unsigned int *qtail
)
3488 struct softnet_data
*sd
;
3489 unsigned long flags
;
3492 sd
= &per_cpu(softnet_data
, cpu
);
3494 local_irq_save(flags
);
3497 if (!netif_running(skb
->dev
))
3499 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3500 if (qlen
<= netdev_max_backlog
&& !skb_flow_limit(skb
, qlen
)) {
3503 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
3504 input_queue_tail_incr_save(sd
, qtail
);
3506 local_irq_restore(flags
);
3507 return NET_RX_SUCCESS
;
3510 /* Schedule NAPI for backlog device
3511 * We can use non atomic operation since we own the queue lock
3513 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
3514 if (!rps_ipi_queued(sd
))
3515 ____napi_schedule(sd
, &sd
->backlog
);
3524 local_irq_restore(flags
);
3526 atomic_long_inc(&skb
->dev
->rx_dropped
);
3531 static int netif_rx_internal(struct sk_buff
*skb
)
3535 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3537 trace_netif_rx(skb
);
3539 if (static_key_false(&rps_needed
)) {
3540 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3546 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3548 cpu
= smp_processor_id();
3550 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3558 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
3565 * netif_rx - post buffer to the network code
3566 * @skb: buffer to post
3568 * This function receives a packet from a device driver and queues it for
3569 * the upper (protocol) levels to process. It always succeeds. The buffer
3570 * may be dropped during processing for congestion control or by the
3574 * NET_RX_SUCCESS (no congestion)
3575 * NET_RX_DROP (packet was dropped)
3579 int netif_rx(struct sk_buff
*skb
)
3581 trace_netif_rx_entry(skb
);
3583 return netif_rx_internal(skb
);
3585 EXPORT_SYMBOL(netif_rx
);
3587 int netif_rx_ni(struct sk_buff
*skb
)
3591 trace_netif_rx_ni_entry(skb
);
3594 err
= netif_rx_internal(skb
);
3595 if (local_softirq_pending())
3601 EXPORT_SYMBOL(netif_rx_ni
);
3603 static void net_tx_action(struct softirq_action
*h
)
3605 struct softnet_data
*sd
= this_cpu_ptr(&softnet_data
);
3607 if (sd
->completion_queue
) {
3608 struct sk_buff
*clist
;
3610 local_irq_disable();
3611 clist
= sd
->completion_queue
;
3612 sd
->completion_queue
= NULL
;
3616 struct sk_buff
*skb
= clist
;
3617 clist
= clist
->next
;
3619 WARN_ON(atomic_read(&skb
->users
));
3620 if (likely(get_kfree_skb_cb(skb
)->reason
== SKB_REASON_CONSUMED
))
3621 trace_consume_skb(skb
);
3623 trace_kfree_skb(skb
, net_tx_action
);
3628 if (sd
->output_queue
) {
3631 local_irq_disable();
3632 head
= sd
->output_queue
;
3633 sd
->output_queue
= NULL
;
3634 sd
->output_queue_tailp
= &sd
->output_queue
;
3638 struct Qdisc
*q
= head
;
3639 spinlock_t
*root_lock
;
3641 head
= head
->next_sched
;
3643 root_lock
= qdisc_lock(q
);
3644 if (spin_trylock(root_lock
)) {
3645 smp_mb__before_atomic();
3646 clear_bit(__QDISC_STATE_SCHED
,
3649 spin_unlock(root_lock
);
3651 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
3653 __netif_reschedule(q
);
3655 smp_mb__before_atomic();
3656 clear_bit(__QDISC_STATE_SCHED
,
3664 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3665 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3666 /* This hook is defined here for ATM LANE */
3667 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
3668 unsigned char *addr
) __read_mostly
;
3669 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
3672 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
3673 struct packet_type
**pt_prev
,
3674 int *ret
, struct net_device
*orig_dev
)
3676 #ifdef CONFIG_NET_CLS_ACT
3677 struct tcf_proto
*cl
= rcu_dereference_bh(skb
->dev
->ingress_cl_list
);
3678 struct tcf_result cl_res
;
3680 /* If there's at least one ingress present somewhere (so
3681 * we get here via enabled static key), remaining devices
3682 * that are not configured with an ingress qdisc will bail
3688 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
3692 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
3693 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
3694 qdisc_bstats_cpu_update(cl
->q
, skb
);
3696 switch (tc_classify(skb
, cl
, &cl_res
, false)) {
3698 case TC_ACT_RECLASSIFY
:
3699 skb
->tc_index
= TC_H_MIN(cl_res
.classid
);
3702 qdisc_qstats_cpu_drop(cl
->q
);
3707 case TC_ACT_REDIRECT
:
3708 /* skb_mac_header check was done by cls/act_bpf, so
3709 * we can safely push the L2 header back before
3710 * redirecting to another netdev
3712 __skb_push(skb
, skb
->mac_len
);
3713 skb_do_redirect(skb
);
3718 #endif /* CONFIG_NET_CLS_ACT */
3723 * netdev_rx_handler_register - register receive handler
3724 * @dev: device to register a handler for
3725 * @rx_handler: receive handler to register
3726 * @rx_handler_data: data pointer that is used by rx handler
3728 * Register a receive handler for a device. This handler will then be
3729 * called from __netif_receive_skb. A negative errno code is returned
3732 * The caller must hold the rtnl_mutex.
3734 * For a general description of rx_handler, see enum rx_handler_result.
3736 int netdev_rx_handler_register(struct net_device
*dev
,
3737 rx_handler_func_t
*rx_handler
,
3738 void *rx_handler_data
)
3742 if (dev
->rx_handler
)
3745 /* Note: rx_handler_data must be set before rx_handler */
3746 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
3747 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
3751 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
3754 * netdev_rx_handler_unregister - unregister receive handler
3755 * @dev: device to unregister a handler from
3757 * Unregister a receive handler from a device.
3759 * The caller must hold the rtnl_mutex.
3761 void netdev_rx_handler_unregister(struct net_device
*dev
)
3765 RCU_INIT_POINTER(dev
->rx_handler
, NULL
);
3766 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3767 * section has a guarantee to see a non NULL rx_handler_data
3771 RCU_INIT_POINTER(dev
->rx_handler_data
, NULL
);
3773 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
3776 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3777 * the special handling of PFMEMALLOC skbs.
3779 static bool skb_pfmemalloc_protocol(struct sk_buff
*skb
)
3781 switch (skb
->protocol
) {
3782 case htons(ETH_P_ARP
):
3783 case htons(ETH_P_IP
):
3784 case htons(ETH_P_IPV6
):
3785 case htons(ETH_P_8021Q
):
3786 case htons(ETH_P_8021AD
):
3793 static inline int nf_ingress(struct sk_buff
*skb
, struct packet_type
**pt_prev
,
3794 int *ret
, struct net_device
*orig_dev
)
3796 #ifdef CONFIG_NETFILTER_INGRESS
3797 if (nf_hook_ingress_active(skb
)) {
3799 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
3803 return nf_hook_ingress(skb
);
3805 #endif /* CONFIG_NETFILTER_INGRESS */
3809 static int __netif_receive_skb_core(struct sk_buff
*skb
, bool pfmemalloc
)
3811 struct packet_type
*ptype
, *pt_prev
;
3812 rx_handler_func_t
*rx_handler
;
3813 struct net_device
*orig_dev
;
3814 bool deliver_exact
= false;
3815 int ret
= NET_RX_DROP
;
3818 net_timestamp_check(!netdev_tstamp_prequeue
, skb
);
3820 trace_netif_receive_skb(skb
);
3822 orig_dev
= skb
->dev
;
3824 skb_reset_network_header(skb
);
3825 if (!skb_transport_header_was_set(skb
))
3826 skb_reset_transport_header(skb
);
3827 skb_reset_mac_len(skb
);
3832 skb
->skb_iif
= skb
->dev
->ifindex
;
3834 __this_cpu_inc(softnet_data
.processed
);
3836 if (skb
->protocol
== cpu_to_be16(ETH_P_8021Q
) ||
3837 skb
->protocol
== cpu_to_be16(ETH_P_8021AD
)) {
3838 skb
= skb_vlan_untag(skb
);
3843 #ifdef CONFIG_NET_CLS_ACT
3844 if (skb
->tc_verd
& TC_NCLS
) {
3845 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
3853 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
3855 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3859 list_for_each_entry_rcu(ptype
, &skb
->dev
->ptype_all
, list
) {
3861 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3866 #ifdef CONFIG_NET_INGRESS
3867 if (static_key_false(&ingress_needed
)) {
3868 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
3872 if (nf_ingress(skb
, &pt_prev
, &ret
, orig_dev
) < 0)
3876 #ifdef CONFIG_NET_CLS_ACT
3880 if (pfmemalloc
&& !skb_pfmemalloc_protocol(skb
))
3883 if (skb_vlan_tag_present(skb
)) {
3885 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3888 if (vlan_do_receive(&skb
))
3890 else if (unlikely(!skb
))
3894 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
3897 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3900 switch (rx_handler(&skb
)) {
3901 case RX_HANDLER_CONSUMED
:
3902 ret
= NET_RX_SUCCESS
;
3904 case RX_HANDLER_ANOTHER
:
3906 case RX_HANDLER_EXACT
:
3907 deliver_exact
= true;
3908 case RX_HANDLER_PASS
:
3915 if (unlikely(skb_vlan_tag_present(skb
))) {
3916 if (skb_vlan_tag_get_id(skb
))
3917 skb
->pkt_type
= PACKET_OTHERHOST
;
3918 /* Note: we might in the future use prio bits
3919 * and set skb->priority like in vlan_do_receive()
3920 * For the time being, just ignore Priority Code Point
3925 type
= skb
->protocol
;
3927 /* deliver only exact match when indicated */
3928 if (likely(!deliver_exact
)) {
3929 deliver_ptype_list_skb(skb
, &pt_prev
, orig_dev
, type
,
3930 &ptype_base
[ntohs(type
) &
3934 deliver_ptype_list_skb(skb
, &pt_prev
, orig_dev
, type
,
3935 &orig_dev
->ptype_specific
);
3937 if (unlikely(skb
->dev
!= orig_dev
)) {
3938 deliver_ptype_list_skb(skb
, &pt_prev
, orig_dev
, type
,
3939 &skb
->dev
->ptype_specific
);
3943 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
3946 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
3949 atomic_long_inc(&skb
->dev
->rx_dropped
);
3951 /* Jamal, now you will not able to escape explaining
3952 * me how you were going to use this. :-)
3961 static int __netif_receive_skb(struct sk_buff
*skb
)
3965 if (sk_memalloc_socks() && skb_pfmemalloc(skb
)) {
3966 unsigned long pflags
= current
->flags
;
3969 * PFMEMALLOC skbs are special, they should
3970 * - be delivered to SOCK_MEMALLOC sockets only
3971 * - stay away from userspace
3972 * - have bounded memory usage
3974 * Use PF_MEMALLOC as this saves us from propagating the allocation
3975 * context down to all allocation sites.
3977 current
->flags
|= PF_MEMALLOC
;
3978 ret
= __netif_receive_skb_core(skb
, true);
3979 tsk_restore_flags(current
, pflags
, PF_MEMALLOC
);
3981 ret
= __netif_receive_skb_core(skb
, false);
3986 static int netif_receive_skb_internal(struct sk_buff
*skb
)
3990 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3992 if (skb_defer_rx_timestamp(skb
))
3993 return NET_RX_SUCCESS
;
3998 if (static_key_false(&rps_needed
)) {
3999 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
4000 int cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
4003 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
4009 ret
= __netif_receive_skb(skb
);
4015 * netif_receive_skb - process receive buffer from network
4016 * @skb: buffer to process
4018 * netif_receive_skb() is the main receive data processing function.
4019 * It always succeeds. The buffer may be dropped during processing
4020 * for congestion control or by the protocol layers.
4022 * This function may only be called from softirq context and interrupts
4023 * should be enabled.
4025 * Return values (usually ignored):
4026 * NET_RX_SUCCESS: no congestion
4027 * NET_RX_DROP: packet was dropped
4029 int netif_receive_skb(struct sk_buff
*skb
)
4031 trace_netif_receive_skb_entry(skb
);
4033 return netif_receive_skb_internal(skb
);
4035 EXPORT_SYMBOL(netif_receive_skb
);
4037 /* Network device is going away, flush any packets still pending
4038 * Called with irqs disabled.
4040 static void flush_backlog(void *arg
)
4042 struct net_device
*dev
= arg
;
4043 struct softnet_data
*sd
= this_cpu_ptr(&softnet_data
);
4044 struct sk_buff
*skb
, *tmp
;
4047 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
4048 if (skb
->dev
== dev
) {
4049 __skb_unlink(skb
, &sd
->input_pkt_queue
);
4051 input_queue_head_incr(sd
);
4056 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
4057 if (skb
->dev
== dev
) {
4058 __skb_unlink(skb
, &sd
->process_queue
);
4060 input_queue_head_incr(sd
);
4065 static int napi_gro_complete(struct sk_buff
*skb
)
4067 struct packet_offload
*ptype
;
4068 __be16 type
= skb
->protocol
;
4069 struct list_head
*head
= &offload_base
;
4072 BUILD_BUG_ON(sizeof(struct napi_gro_cb
) > sizeof(skb
->cb
));
4074 if (NAPI_GRO_CB(skb
)->count
== 1) {
4075 skb_shinfo(skb
)->gso_size
= 0;
4080 list_for_each_entry_rcu(ptype
, head
, list
) {
4081 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_complete
)
4084 err
= ptype
->callbacks
.gro_complete(skb
, 0);
4090 WARN_ON(&ptype
->list
== head
);
4092 return NET_RX_SUCCESS
;
4096 return netif_receive_skb_internal(skb
);
4099 /* napi->gro_list contains packets ordered by age.
4100 * youngest packets at the head of it.
4101 * Complete skbs in reverse order to reduce latencies.
4103 void napi_gro_flush(struct napi_struct
*napi
, bool flush_old
)
4105 struct sk_buff
*skb
, *prev
= NULL
;
4107 /* scan list and build reverse chain */
4108 for (skb
= napi
->gro_list
; skb
!= NULL
; skb
= skb
->next
) {
4113 for (skb
= prev
; skb
; skb
= prev
) {
4116 if (flush_old
&& NAPI_GRO_CB(skb
)->age
== jiffies
)
4120 napi_gro_complete(skb
);
4124 napi
->gro_list
= NULL
;
4126 EXPORT_SYMBOL(napi_gro_flush
);
4128 static void gro_list_prepare(struct napi_struct
*napi
, struct sk_buff
*skb
)
4131 unsigned int maclen
= skb
->dev
->hard_header_len
;
4132 u32 hash
= skb_get_hash_raw(skb
);
4134 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
4135 unsigned long diffs
;
4137 NAPI_GRO_CB(p
)->flush
= 0;
4139 if (hash
!= skb_get_hash_raw(p
)) {
4140 NAPI_GRO_CB(p
)->same_flow
= 0;
4144 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
4145 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
4146 if (maclen
== ETH_HLEN
)
4147 diffs
|= compare_ether_header(skb_mac_header(p
),
4148 skb_mac_header(skb
));
4150 diffs
= memcmp(skb_mac_header(p
),
4151 skb_mac_header(skb
),
4153 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
4157 static void skb_gro_reset_offset(struct sk_buff
*skb
)
4159 const struct skb_shared_info
*pinfo
= skb_shinfo(skb
);
4160 const skb_frag_t
*frag0
= &pinfo
->frags
[0];
4162 NAPI_GRO_CB(skb
)->data_offset
= 0;
4163 NAPI_GRO_CB(skb
)->frag0
= NULL
;
4164 NAPI_GRO_CB(skb
)->frag0_len
= 0;
4166 if (skb_mac_header(skb
) == skb_tail_pointer(skb
) &&
4168 !PageHighMem(skb_frag_page(frag0
))) {
4169 NAPI_GRO_CB(skb
)->frag0
= skb_frag_address(frag0
);
4170 NAPI_GRO_CB(skb
)->frag0_len
= skb_frag_size(frag0
);
4174 static void gro_pull_from_frag0(struct sk_buff
*skb
, int grow
)
4176 struct skb_shared_info
*pinfo
= skb_shinfo(skb
);
4178 BUG_ON(skb
->end
- skb
->tail
< grow
);
4180 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
4182 skb
->data_len
-= grow
;
4185 pinfo
->frags
[0].page_offset
+= grow
;
4186 skb_frag_size_sub(&pinfo
->frags
[0], grow
);
4188 if (unlikely(!skb_frag_size(&pinfo
->frags
[0]))) {
4189 skb_frag_unref(skb
, 0);
4190 memmove(pinfo
->frags
, pinfo
->frags
+ 1,
4191 --pinfo
->nr_frags
* sizeof(pinfo
->frags
[0]));
4195 static enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
4197 struct sk_buff
**pp
= NULL
;
4198 struct packet_offload
*ptype
;
4199 __be16 type
= skb
->protocol
;
4200 struct list_head
*head
= &offload_base
;
4202 enum gro_result ret
;
4205 if (!(skb
->dev
->features
& NETIF_F_GRO
))
4208 if (skb_is_gso(skb
) || skb_has_frag_list(skb
) || skb
->csum_bad
)
4211 gro_list_prepare(napi
, skb
);
4214 list_for_each_entry_rcu(ptype
, head
, list
) {
4215 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_receive
)
4218 skb_set_network_header(skb
, skb_gro_offset(skb
));
4219 skb_reset_mac_len(skb
);
4220 NAPI_GRO_CB(skb
)->same_flow
= 0;
4221 NAPI_GRO_CB(skb
)->flush
= 0;
4222 NAPI_GRO_CB(skb
)->free
= 0;
4223 NAPI_GRO_CB(skb
)->udp_mark
= 0;
4224 NAPI_GRO_CB(skb
)->gro_remcsum_start
= 0;
4226 /* Setup for GRO checksum validation */
4227 switch (skb
->ip_summed
) {
4228 case CHECKSUM_COMPLETE
:
4229 NAPI_GRO_CB(skb
)->csum
= skb
->csum
;
4230 NAPI_GRO_CB(skb
)->csum_valid
= 1;
4231 NAPI_GRO_CB(skb
)->csum_cnt
= 0;
4233 case CHECKSUM_UNNECESSARY
:
4234 NAPI_GRO_CB(skb
)->csum_cnt
= skb
->csum_level
+ 1;
4235 NAPI_GRO_CB(skb
)->csum_valid
= 0;
4238 NAPI_GRO_CB(skb
)->csum_cnt
= 0;
4239 NAPI_GRO_CB(skb
)->csum_valid
= 0;
4242 pp
= ptype
->callbacks
.gro_receive(&napi
->gro_list
, skb
);
4247 if (&ptype
->list
== head
)
4250 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
4251 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
4254 struct sk_buff
*nskb
= *pp
;
4258 napi_gro_complete(nskb
);
4265 if (NAPI_GRO_CB(skb
)->flush
)
4268 if (unlikely(napi
->gro_count
>= MAX_GRO_SKBS
)) {
4269 struct sk_buff
*nskb
= napi
->gro_list
;
4271 /* locate the end of the list to select the 'oldest' flow */
4272 while (nskb
->next
) {
4278 napi_gro_complete(nskb
);
4282 NAPI_GRO_CB(skb
)->count
= 1;
4283 NAPI_GRO_CB(skb
)->age
= jiffies
;
4284 NAPI_GRO_CB(skb
)->last
= skb
;
4285 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
4286 skb
->next
= napi
->gro_list
;
4287 napi
->gro_list
= skb
;
4291 grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
4293 gro_pull_from_frag0(skb
, grow
);
4302 struct packet_offload
*gro_find_receive_by_type(__be16 type
)
4304 struct list_head
*offload_head
= &offload_base
;
4305 struct packet_offload
*ptype
;
4307 list_for_each_entry_rcu(ptype
, offload_head
, list
) {
4308 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_receive
)
4314 EXPORT_SYMBOL(gro_find_receive_by_type
);
4316 struct packet_offload
*gro_find_complete_by_type(__be16 type
)
4318 struct list_head
*offload_head
= &offload_base
;
4319 struct packet_offload
*ptype
;
4321 list_for_each_entry_rcu(ptype
, offload_head
, list
) {
4322 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_complete
)
4328 EXPORT_SYMBOL(gro_find_complete_by_type
);
4330 static gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
4334 if (netif_receive_skb_internal(skb
))
4342 case GRO_MERGED_FREE
:
4343 if (NAPI_GRO_CB(skb
)->free
== NAPI_GRO_FREE_STOLEN_HEAD
)
4344 kmem_cache_free(skbuff_head_cache
, skb
);
4357 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
4359 skb_mark_napi_id(skb
, napi
);
4360 trace_napi_gro_receive_entry(skb
);
4362 skb_gro_reset_offset(skb
);
4364 return napi_skb_finish(dev_gro_receive(napi
, skb
), skb
);
4366 EXPORT_SYMBOL(napi_gro_receive
);
4368 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
4370 if (unlikely(skb
->pfmemalloc
)) {
4374 __skb_pull(skb
, skb_headlen(skb
));
4375 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4376 skb_reserve(skb
, NET_SKB_PAD
+ NET_IP_ALIGN
- skb_headroom(skb
));
4378 skb
->dev
= napi
->dev
;
4380 skb
->encapsulation
= 0;
4381 skb_shinfo(skb
)->gso_type
= 0;
4382 skb
->truesize
= SKB_TRUESIZE(skb_end_offset(skb
));
4387 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
4389 struct sk_buff
*skb
= napi
->skb
;
4392 skb
= napi_alloc_skb(napi
, GRO_MAX_HEAD
);
4395 skb_mark_napi_id(skb
, napi
);
4400 EXPORT_SYMBOL(napi_get_frags
);
4402 static gro_result_t
napi_frags_finish(struct napi_struct
*napi
,
4403 struct sk_buff
*skb
,
4409 __skb_push(skb
, ETH_HLEN
);
4410 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
4411 if (ret
== GRO_NORMAL
&& netif_receive_skb_internal(skb
))
4416 case GRO_MERGED_FREE
:
4417 napi_reuse_skb(napi
, skb
);
4427 /* Upper GRO stack assumes network header starts at gro_offset=0
4428 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4429 * We copy ethernet header into skb->data to have a common layout.
4431 static struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
4433 struct sk_buff
*skb
= napi
->skb
;
4434 const struct ethhdr
*eth
;
4435 unsigned int hlen
= sizeof(*eth
);
4439 skb_reset_mac_header(skb
);
4440 skb_gro_reset_offset(skb
);
4442 eth
= skb_gro_header_fast(skb
, 0);
4443 if (unlikely(skb_gro_header_hard(skb
, hlen
))) {
4444 eth
= skb_gro_header_slow(skb
, hlen
, 0);
4445 if (unlikely(!eth
)) {
4446 napi_reuse_skb(napi
, skb
);
4450 gro_pull_from_frag0(skb
, hlen
);
4451 NAPI_GRO_CB(skb
)->frag0
+= hlen
;
4452 NAPI_GRO_CB(skb
)->frag0_len
-= hlen
;
4454 __skb_pull(skb
, hlen
);
4457 * This works because the only protocols we care about don't require
4459 * We'll fix it up properly in napi_frags_finish()
4461 skb
->protocol
= eth
->h_proto
;
4466 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
4468 struct sk_buff
*skb
= napi_frags_skb(napi
);
4473 trace_napi_gro_frags_entry(skb
);
4475 return napi_frags_finish(napi
, skb
, dev_gro_receive(napi
, skb
));
4477 EXPORT_SYMBOL(napi_gro_frags
);
4479 /* Compute the checksum from gro_offset and return the folded value
4480 * after adding in any pseudo checksum.
4482 __sum16
__skb_gro_checksum_complete(struct sk_buff
*skb
)
4487 wsum
= skb_checksum(skb
, skb_gro_offset(skb
), skb_gro_len(skb
), 0);
4489 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4490 sum
= csum_fold(csum_add(NAPI_GRO_CB(skb
)->csum
, wsum
));
4492 if (unlikely(skb
->ip_summed
== CHECKSUM_COMPLETE
) &&
4493 !skb
->csum_complete_sw
)
4494 netdev_rx_csum_fault(skb
->dev
);
4497 NAPI_GRO_CB(skb
)->csum
= wsum
;
4498 NAPI_GRO_CB(skb
)->csum_valid
= 1;
4502 EXPORT_SYMBOL(__skb_gro_checksum_complete
);
4505 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4506 * Note: called with local irq disabled, but exits with local irq enabled.
4508 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
4511 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
4514 sd
->rps_ipi_list
= NULL
;
4518 /* Send pending IPI's to kick RPS processing on remote cpus. */
4520 struct softnet_data
*next
= remsd
->rps_ipi_next
;
4522 if (cpu_online(remsd
->cpu
))
4523 smp_call_function_single_async(remsd
->cpu
,
4532 static bool sd_has_rps_ipi_waiting(struct softnet_data
*sd
)
4535 return sd
->rps_ipi_list
!= NULL
;
4541 static int process_backlog(struct napi_struct
*napi
, int quota
)
4544 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
4546 /* Check if we have pending ipi, its better to send them now,
4547 * not waiting net_rx_action() end.
4549 if (sd_has_rps_ipi_waiting(sd
)) {
4550 local_irq_disable();
4551 net_rps_action_and_irq_enable(sd
);
4554 napi
->weight
= weight_p
;
4555 local_irq_disable();
4557 struct sk_buff
*skb
;
4559 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
4562 __netif_receive_skb(skb
);
4564 local_irq_disable();
4565 input_queue_head_incr(sd
);
4566 if (++work
>= quota
) {
4573 if (skb_queue_empty(&sd
->input_pkt_queue
)) {
4575 * Inline a custom version of __napi_complete().
4576 * only current cpu owns and manipulates this napi,
4577 * and NAPI_STATE_SCHED is the only possible flag set
4579 * We can use a plain write instead of clear_bit(),
4580 * and we dont need an smp_mb() memory barrier.
4588 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
4589 &sd
->process_queue
);
4598 * __napi_schedule - schedule for receive
4599 * @n: entry to schedule
4601 * The entry's receive function will be scheduled to run.
4602 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
4604 void __napi_schedule(struct napi_struct
*n
)
4606 unsigned long flags
;
4608 local_irq_save(flags
);
4609 ____napi_schedule(this_cpu_ptr(&softnet_data
), n
);
4610 local_irq_restore(flags
);
4612 EXPORT_SYMBOL(__napi_schedule
);
4615 * __napi_schedule_irqoff - schedule for receive
4616 * @n: entry to schedule
4618 * Variant of __napi_schedule() assuming hard irqs are masked
4620 void __napi_schedule_irqoff(struct napi_struct
*n
)
4622 ____napi_schedule(this_cpu_ptr(&softnet_data
), n
);
4624 EXPORT_SYMBOL(__napi_schedule_irqoff
);
4626 void __napi_complete(struct napi_struct
*n
)
4628 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
4630 list_del_init(&n
->poll_list
);
4631 smp_mb__before_atomic();
4632 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
4634 EXPORT_SYMBOL(__napi_complete
);
4636 void napi_complete_done(struct napi_struct
*n
, int work_done
)
4638 unsigned long flags
;
4641 * don't let napi dequeue from the cpu poll list
4642 * just in case its running on a different cpu
4644 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
4648 unsigned long timeout
= 0;
4651 timeout
= n
->dev
->gro_flush_timeout
;
4654 hrtimer_start(&n
->timer
, ns_to_ktime(timeout
),
4655 HRTIMER_MODE_REL_PINNED
);
4657 napi_gro_flush(n
, false);
4659 if (likely(list_empty(&n
->poll_list
))) {
4660 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED
, &n
->state
));
4662 /* If n->poll_list is not empty, we need to mask irqs */
4663 local_irq_save(flags
);
4665 local_irq_restore(flags
);
4668 EXPORT_SYMBOL(napi_complete_done
);
4670 /* must be called under rcu_read_lock(), as we dont take a reference */
4671 static struct napi_struct
*napi_by_id(unsigned int napi_id
)
4673 unsigned int hash
= napi_id
% HASH_SIZE(napi_hash
);
4674 struct napi_struct
*napi
;
4676 hlist_for_each_entry_rcu(napi
, &napi_hash
[hash
], napi_hash_node
)
4677 if (napi
->napi_id
== napi_id
)
4683 #if defined(CONFIG_NET_RX_BUSY_POLL)
4684 #define BUSY_POLL_BUDGET 8
4685 bool sk_busy_loop(struct sock
*sk
, int nonblock
)
4687 unsigned long end_time
= !nonblock
? sk_busy_loop_end_time(sk
) : 0;
4688 int (*busy_poll
)(struct napi_struct
*dev
);
4689 struct napi_struct
*napi
;
4694 napi
= napi_by_id(sk
->sk_napi_id
);
4698 /* Note: ndo_busy_poll method is optional in linux-4.5 */
4699 busy_poll
= napi
->dev
->netdev_ops
->ndo_busy_poll
;
4705 rc
= busy_poll(napi
);
4706 } else if (napi_schedule_prep(napi
)) {
4707 void *have
= netpoll_poll_lock(napi
);
4709 if (test_bit(NAPI_STATE_SCHED
, &napi
->state
)) {
4710 rc
= napi
->poll(napi
, BUSY_POLL_BUDGET
);
4711 trace_napi_poll(napi
);
4712 if (rc
== BUSY_POLL_BUDGET
) {
4713 napi_complete_done(napi
, rc
);
4714 napi_schedule(napi
);
4717 netpoll_poll_unlock(have
);
4720 NET_ADD_STATS_BH(sock_net(sk
),
4721 LINUX_MIB_BUSYPOLLRXPACKETS
, rc
);
4724 if (rc
== LL_FLUSH_FAILED
)
4725 break; /* permanent failure */
4728 } while (!nonblock
&& skb_queue_empty(&sk
->sk_receive_queue
) &&
4729 !need_resched() && !busy_loop_timeout(end_time
));
4731 rc
= !skb_queue_empty(&sk
->sk_receive_queue
);
4736 EXPORT_SYMBOL(sk_busy_loop
);
4738 #endif /* CONFIG_NET_RX_BUSY_POLL */
4740 void napi_hash_add(struct napi_struct
*napi
)
4742 if (test_bit(NAPI_STATE_NO_BUSY_POLL
, &napi
->state
) ||
4743 test_and_set_bit(NAPI_STATE_HASHED
, &napi
->state
))
4746 spin_lock(&napi_hash_lock
);
4748 /* 0..NR_CPUS+1 range is reserved for sender_cpu use */
4750 if (unlikely(++napi_gen_id
< NR_CPUS
+ 1))
4751 napi_gen_id
= NR_CPUS
+ 1;
4752 } while (napi_by_id(napi_gen_id
));
4753 napi
->napi_id
= napi_gen_id
;
4755 hlist_add_head_rcu(&napi
->napi_hash_node
,
4756 &napi_hash
[napi
->napi_id
% HASH_SIZE(napi_hash
)]);
4758 spin_unlock(&napi_hash_lock
);
4760 EXPORT_SYMBOL_GPL(napi_hash_add
);
4762 /* Warning : caller is responsible to make sure rcu grace period
4763 * is respected before freeing memory containing @napi
4765 bool napi_hash_del(struct napi_struct
*napi
)
4767 bool rcu_sync_needed
= false;
4769 spin_lock(&napi_hash_lock
);
4771 if (test_and_clear_bit(NAPI_STATE_HASHED
, &napi
->state
)) {
4772 rcu_sync_needed
= true;
4773 hlist_del_rcu(&napi
->napi_hash_node
);
4775 spin_unlock(&napi_hash_lock
);
4776 return rcu_sync_needed
;
4778 EXPORT_SYMBOL_GPL(napi_hash_del
);
4780 static enum hrtimer_restart
napi_watchdog(struct hrtimer
*timer
)
4782 struct napi_struct
*napi
;
4784 napi
= container_of(timer
, struct napi_struct
, timer
);
4786 napi_schedule(napi
);
4788 return HRTIMER_NORESTART
;
4791 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
4792 int (*poll
)(struct napi_struct
*, int), int weight
)
4794 INIT_LIST_HEAD(&napi
->poll_list
);
4795 hrtimer_init(&napi
->timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_REL_PINNED
);
4796 napi
->timer
.function
= napi_watchdog
;
4797 napi
->gro_count
= 0;
4798 napi
->gro_list
= NULL
;
4801 if (weight
> NAPI_POLL_WEIGHT
)
4802 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4804 napi
->weight
= weight
;
4805 list_add(&napi
->dev_list
, &dev
->napi_list
);
4807 #ifdef CONFIG_NETPOLL
4808 spin_lock_init(&napi
->poll_lock
);
4809 napi
->poll_owner
= -1;
4811 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
4812 napi_hash_add(napi
);
4814 EXPORT_SYMBOL(netif_napi_add
);
4816 void napi_disable(struct napi_struct
*n
)
4819 set_bit(NAPI_STATE_DISABLE
, &n
->state
);
4821 while (test_and_set_bit(NAPI_STATE_SCHED
, &n
->state
))
4823 while (test_and_set_bit(NAPI_STATE_NPSVC
, &n
->state
))
4826 hrtimer_cancel(&n
->timer
);
4828 clear_bit(NAPI_STATE_DISABLE
, &n
->state
);
4830 EXPORT_SYMBOL(napi_disable
);
4832 /* Must be called in process context */
4833 void netif_napi_del(struct napi_struct
*napi
)
4836 if (napi_hash_del(napi
))
4838 list_del_init(&napi
->dev_list
);
4839 napi_free_frags(napi
);
4841 kfree_skb_list(napi
->gro_list
);
4842 napi
->gro_list
= NULL
;
4843 napi
->gro_count
= 0;
4845 EXPORT_SYMBOL(netif_napi_del
);
4847 static int napi_poll(struct napi_struct
*n
, struct list_head
*repoll
)
4852 list_del_init(&n
->poll_list
);
4854 have
= netpoll_poll_lock(n
);
4858 /* This NAPI_STATE_SCHED test is for avoiding a race
4859 * with netpoll's poll_napi(). Only the entity which
4860 * obtains the lock and sees NAPI_STATE_SCHED set will
4861 * actually make the ->poll() call. Therefore we avoid
4862 * accidentally calling ->poll() when NAPI is not scheduled.
4865 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
4866 work
= n
->poll(n
, weight
);
4870 WARN_ON_ONCE(work
> weight
);
4872 if (likely(work
< weight
))
4875 /* Drivers must not modify the NAPI state if they
4876 * consume the entire weight. In such cases this code
4877 * still "owns" the NAPI instance and therefore can
4878 * move the instance around on the list at-will.
4880 if (unlikely(napi_disable_pending(n
))) {
4886 /* flush too old packets
4887 * If HZ < 1000, flush all packets.
4889 napi_gro_flush(n
, HZ
>= 1000);
4892 /* Some drivers may have called napi_schedule
4893 * prior to exhausting their budget.
4895 if (unlikely(!list_empty(&n
->poll_list
))) {
4896 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
4897 n
->dev
? n
->dev
->name
: "backlog");
4901 list_add_tail(&n
->poll_list
, repoll
);
4904 netpoll_poll_unlock(have
);
4909 static void net_rx_action(struct softirq_action
*h
)
4911 struct softnet_data
*sd
= this_cpu_ptr(&softnet_data
);
4912 unsigned long time_limit
= jiffies
+ 2;
4913 int budget
= netdev_budget
;
4917 local_irq_disable();
4918 list_splice_init(&sd
->poll_list
, &list
);
4922 struct napi_struct
*n
;
4924 if (list_empty(&list
)) {
4925 if (!sd_has_rps_ipi_waiting(sd
) && list_empty(&repoll
))
4930 n
= list_first_entry(&list
, struct napi_struct
, poll_list
);
4931 budget
-= napi_poll(n
, &repoll
);
4933 /* If softirq window is exhausted then punt.
4934 * Allow this to run for 2 jiffies since which will allow
4935 * an average latency of 1.5/HZ.
4937 if (unlikely(budget
<= 0 ||
4938 time_after_eq(jiffies
, time_limit
))) {
4944 local_irq_disable();
4946 list_splice_tail_init(&sd
->poll_list
, &list
);
4947 list_splice_tail(&repoll
, &list
);
4948 list_splice(&list
, &sd
->poll_list
);
4949 if (!list_empty(&sd
->poll_list
))
4950 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
4952 net_rps_action_and_irq_enable(sd
);
4955 struct netdev_adjacent
{
4956 struct net_device
*dev
;
4958 /* upper master flag, there can only be one master device per list */
4961 /* counter for the number of times this device was added to us */
4964 /* private field for the users */
4967 struct list_head list
;
4968 struct rcu_head rcu
;
4971 static struct netdev_adjacent
*__netdev_find_adj(struct net_device
*adj_dev
,
4972 struct list_head
*adj_list
)
4974 struct netdev_adjacent
*adj
;
4976 list_for_each_entry(adj
, adj_list
, list
) {
4977 if (adj
->dev
== adj_dev
)
4984 * netdev_has_upper_dev - Check if device is linked to an upper device
4986 * @upper_dev: upper device to check
4988 * Find out if a device is linked to specified upper device and return true
4989 * in case it is. Note that this checks only immediate upper device,
4990 * not through a complete stack of devices. The caller must hold the RTNL lock.
4992 bool netdev_has_upper_dev(struct net_device
*dev
,
4993 struct net_device
*upper_dev
)
4997 return __netdev_find_adj(upper_dev
, &dev
->all_adj_list
.upper
);
4999 EXPORT_SYMBOL(netdev_has_upper_dev
);
5002 * netdev_has_any_upper_dev - Check if device is linked to some device
5005 * Find out if a device is linked to an upper device and return true in case
5006 * it is. The caller must hold the RTNL lock.
5008 static bool netdev_has_any_upper_dev(struct net_device
*dev
)
5012 return !list_empty(&dev
->all_adj_list
.upper
);
5016 * netdev_master_upper_dev_get - Get master upper device
5019 * Find a master upper device and return pointer to it or NULL in case
5020 * it's not there. The caller must hold the RTNL lock.
5022 struct net_device
*netdev_master_upper_dev_get(struct net_device
*dev
)
5024 struct netdev_adjacent
*upper
;
5028 if (list_empty(&dev
->adj_list
.upper
))
5031 upper
= list_first_entry(&dev
->adj_list
.upper
,
5032 struct netdev_adjacent
, list
);
5033 if (likely(upper
->master
))
5037 EXPORT_SYMBOL(netdev_master_upper_dev_get
);
5039 void *netdev_adjacent_get_private(struct list_head
*adj_list
)
5041 struct netdev_adjacent
*adj
;
5043 adj
= list_entry(adj_list
, struct netdev_adjacent
, list
);
5045 return adj
->private;
5047 EXPORT_SYMBOL(netdev_adjacent_get_private
);
5050 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
5052 * @iter: list_head ** of the current position
5054 * Gets the next device from the dev's upper list, starting from iter
5055 * position. The caller must hold RCU read lock.
5057 struct net_device
*netdev_upper_get_next_dev_rcu(struct net_device
*dev
,
5058 struct list_head
**iter
)
5060 struct netdev_adjacent
*upper
;
5062 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5064 upper
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
5066 if (&upper
->list
== &dev
->adj_list
.upper
)
5069 *iter
= &upper
->list
;
5073 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu
);
5076 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
5078 * @iter: list_head ** of the current position
5080 * Gets the next device from the dev's upper list, starting from iter
5081 * position. The caller must hold RCU read lock.
5083 struct net_device
*netdev_all_upper_get_next_dev_rcu(struct net_device
*dev
,
5084 struct list_head
**iter
)
5086 struct netdev_adjacent
*upper
;
5088 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5090 upper
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
5092 if (&upper
->list
== &dev
->all_adj_list
.upper
)
5095 *iter
= &upper
->list
;
5099 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu
);
5102 * netdev_lower_get_next_private - Get the next ->private from the
5103 * lower neighbour list
5105 * @iter: list_head ** of the current position
5107 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5108 * list, starting from iter position. The caller must hold either hold the
5109 * RTNL lock or its own locking that guarantees that the neighbour lower
5110 * list will remain unchanged.
5112 void *netdev_lower_get_next_private(struct net_device
*dev
,
5113 struct list_head
**iter
)
5115 struct netdev_adjacent
*lower
;
5117 lower
= list_entry(*iter
, struct netdev_adjacent
, list
);
5119 if (&lower
->list
== &dev
->adj_list
.lower
)
5122 *iter
= lower
->list
.next
;
5124 return lower
->private;
5126 EXPORT_SYMBOL(netdev_lower_get_next_private
);
5129 * netdev_lower_get_next_private_rcu - Get the next ->private from the
5130 * lower neighbour list, RCU
5133 * @iter: list_head ** of the current position
5135 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5136 * list, starting from iter position. The caller must hold RCU read lock.
5138 void *netdev_lower_get_next_private_rcu(struct net_device
*dev
,
5139 struct list_head
**iter
)
5141 struct netdev_adjacent
*lower
;
5143 WARN_ON_ONCE(!rcu_read_lock_held());
5145 lower
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
5147 if (&lower
->list
== &dev
->adj_list
.lower
)
5150 *iter
= &lower
->list
;
5152 return lower
->private;
5154 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu
);
5157 * netdev_lower_get_next - Get the next device from the lower neighbour
5160 * @iter: list_head ** of the current position
5162 * Gets the next netdev_adjacent from the dev's lower neighbour
5163 * list, starting from iter position. The caller must hold RTNL lock or
5164 * its own locking that guarantees that the neighbour lower
5165 * list will remain unchanged.
5167 void *netdev_lower_get_next(struct net_device
*dev
, struct list_head
**iter
)
5169 struct netdev_adjacent
*lower
;
5171 lower
= list_entry((*iter
)->next
, struct netdev_adjacent
, list
);
5173 if (&lower
->list
== &dev
->adj_list
.lower
)
5176 *iter
= &lower
->list
;
5180 EXPORT_SYMBOL(netdev_lower_get_next
);
5183 * netdev_lower_get_first_private_rcu - Get the first ->private from the
5184 * lower neighbour list, RCU
5188 * Gets the first netdev_adjacent->private from the dev's lower neighbour
5189 * list. The caller must hold RCU read lock.
5191 void *netdev_lower_get_first_private_rcu(struct net_device
*dev
)
5193 struct netdev_adjacent
*lower
;
5195 lower
= list_first_or_null_rcu(&dev
->adj_list
.lower
,
5196 struct netdev_adjacent
, list
);
5198 return lower
->private;
5201 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu
);
5204 * netdev_master_upper_dev_get_rcu - Get master upper device
5207 * Find a master upper device and return pointer to it or NULL in case
5208 * it's not there. The caller must hold the RCU read lock.
5210 struct net_device
*netdev_master_upper_dev_get_rcu(struct net_device
*dev
)
5212 struct netdev_adjacent
*upper
;
5214 upper
= list_first_or_null_rcu(&dev
->adj_list
.upper
,
5215 struct netdev_adjacent
, list
);
5216 if (upper
&& likely(upper
->master
))
5220 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu
);
5222 static int netdev_adjacent_sysfs_add(struct net_device
*dev
,
5223 struct net_device
*adj_dev
,
5224 struct list_head
*dev_list
)
5226 char linkname
[IFNAMSIZ
+7];
5227 sprintf(linkname
, dev_list
== &dev
->adj_list
.upper
?
5228 "upper_%s" : "lower_%s", adj_dev
->name
);
5229 return sysfs_create_link(&(dev
->dev
.kobj
), &(adj_dev
->dev
.kobj
),
5232 static void netdev_adjacent_sysfs_del(struct net_device
*dev
,
5234 struct list_head
*dev_list
)
5236 char linkname
[IFNAMSIZ
+7];
5237 sprintf(linkname
, dev_list
== &dev
->adj_list
.upper
?
5238 "upper_%s" : "lower_%s", name
);
5239 sysfs_remove_link(&(dev
->dev
.kobj
), linkname
);
5242 static inline bool netdev_adjacent_is_neigh_list(struct net_device
*dev
,
5243 struct net_device
*adj_dev
,
5244 struct list_head
*dev_list
)
5246 return (dev_list
== &dev
->adj_list
.upper
||
5247 dev_list
== &dev
->adj_list
.lower
) &&
5248 net_eq(dev_net(dev
), dev_net(adj_dev
));
5251 static int __netdev_adjacent_dev_insert(struct net_device
*dev
,
5252 struct net_device
*adj_dev
,
5253 struct list_head
*dev_list
,
5254 void *private, bool master
)
5256 struct netdev_adjacent
*adj
;
5259 adj
= __netdev_find_adj(adj_dev
, dev_list
);
5266 adj
= kmalloc(sizeof(*adj
), GFP_KERNEL
);
5271 adj
->master
= master
;
5273 adj
->private = private;
5276 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
5277 adj_dev
->name
, dev
->name
, adj_dev
->name
);
5279 if (netdev_adjacent_is_neigh_list(dev
, adj_dev
, dev_list
)) {
5280 ret
= netdev_adjacent_sysfs_add(dev
, adj_dev
, dev_list
);
5285 /* Ensure that master link is always the first item in list. */
5287 ret
= sysfs_create_link(&(dev
->dev
.kobj
),
5288 &(adj_dev
->dev
.kobj
), "master");
5290 goto remove_symlinks
;
5292 list_add_rcu(&adj
->list
, dev_list
);
5294 list_add_tail_rcu(&adj
->list
, dev_list
);
5300 if (netdev_adjacent_is_neigh_list(dev
, adj_dev
, dev_list
))
5301 netdev_adjacent_sysfs_del(dev
, adj_dev
->name
, dev_list
);
5309 static void __netdev_adjacent_dev_remove(struct net_device
*dev
,
5310 struct net_device
*adj_dev
,
5311 struct list_head
*dev_list
)
5313 struct netdev_adjacent
*adj
;
5315 adj
= __netdev_find_adj(adj_dev
, dev_list
);
5318 pr_err("tried to remove device %s from %s\n",
5319 dev
->name
, adj_dev
->name
);
5323 if (adj
->ref_nr
> 1) {
5324 pr_debug("%s to %s ref_nr-- = %d\n", dev
->name
, adj_dev
->name
,
5331 sysfs_remove_link(&(dev
->dev
.kobj
), "master");
5333 if (netdev_adjacent_is_neigh_list(dev
, adj_dev
, dev_list
))
5334 netdev_adjacent_sysfs_del(dev
, adj_dev
->name
, dev_list
);
5336 list_del_rcu(&adj
->list
);
5337 pr_debug("dev_put for %s, because link removed from %s to %s\n",
5338 adj_dev
->name
, dev
->name
, adj_dev
->name
);
5340 kfree_rcu(adj
, rcu
);
5343 static int __netdev_adjacent_dev_link_lists(struct net_device
*dev
,
5344 struct net_device
*upper_dev
,
5345 struct list_head
*up_list
,
5346 struct list_head
*down_list
,
5347 void *private, bool master
)
5351 ret
= __netdev_adjacent_dev_insert(dev
, upper_dev
, up_list
, private,
5356 ret
= __netdev_adjacent_dev_insert(upper_dev
, dev
, down_list
, private,
5359 __netdev_adjacent_dev_remove(dev
, upper_dev
, up_list
);
5366 static int __netdev_adjacent_dev_link(struct net_device
*dev
,
5367 struct net_device
*upper_dev
)
5369 return __netdev_adjacent_dev_link_lists(dev
, upper_dev
,
5370 &dev
->all_adj_list
.upper
,
5371 &upper_dev
->all_adj_list
.lower
,
5375 static void __netdev_adjacent_dev_unlink_lists(struct net_device
*dev
,
5376 struct net_device
*upper_dev
,
5377 struct list_head
*up_list
,
5378 struct list_head
*down_list
)
5380 __netdev_adjacent_dev_remove(dev
, upper_dev
, up_list
);
5381 __netdev_adjacent_dev_remove(upper_dev
, dev
, down_list
);
5384 static void __netdev_adjacent_dev_unlink(struct net_device
*dev
,
5385 struct net_device
*upper_dev
)
5387 __netdev_adjacent_dev_unlink_lists(dev
, upper_dev
,
5388 &dev
->all_adj_list
.upper
,
5389 &upper_dev
->all_adj_list
.lower
);
5392 static int __netdev_adjacent_dev_link_neighbour(struct net_device
*dev
,
5393 struct net_device
*upper_dev
,
5394 void *private, bool master
)
5396 int ret
= __netdev_adjacent_dev_link(dev
, upper_dev
);
5401 ret
= __netdev_adjacent_dev_link_lists(dev
, upper_dev
,
5402 &dev
->adj_list
.upper
,
5403 &upper_dev
->adj_list
.lower
,
5406 __netdev_adjacent_dev_unlink(dev
, upper_dev
);
5413 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device
*dev
,
5414 struct net_device
*upper_dev
)
5416 __netdev_adjacent_dev_unlink(dev
, upper_dev
);
5417 __netdev_adjacent_dev_unlink_lists(dev
, upper_dev
,
5418 &dev
->adj_list
.upper
,
5419 &upper_dev
->adj_list
.lower
);
5422 static int __netdev_upper_dev_link(struct net_device
*dev
,
5423 struct net_device
*upper_dev
, bool master
,
5424 void *upper_priv
, void *upper_info
)
5426 struct netdev_notifier_changeupper_info changeupper_info
;
5427 struct netdev_adjacent
*i
, *j
, *to_i
, *to_j
;
5432 if (dev
== upper_dev
)
5435 /* To prevent loops, check if dev is not upper device to upper_dev. */
5436 if (__netdev_find_adj(dev
, &upper_dev
->all_adj_list
.upper
))
5439 if (__netdev_find_adj(upper_dev
, &dev
->adj_list
.upper
))
5442 if (master
&& netdev_master_upper_dev_get(dev
))
5445 changeupper_info
.upper_dev
= upper_dev
;
5446 changeupper_info
.master
= master
;
5447 changeupper_info
.linking
= true;
5448 changeupper_info
.upper_info
= upper_info
;
5450 ret
= call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER
, dev
,
5451 &changeupper_info
.info
);
5452 ret
= notifier_to_errno(ret
);
5456 ret
= __netdev_adjacent_dev_link_neighbour(dev
, upper_dev
, upper_priv
,
5461 /* Now that we linked these devs, make all the upper_dev's
5462 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5463 * versa, and don't forget the devices itself. All of these
5464 * links are non-neighbours.
5466 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5467 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
) {
5468 pr_debug("Interlinking %s with %s, non-neighbour\n",
5469 i
->dev
->name
, j
->dev
->name
);
5470 ret
= __netdev_adjacent_dev_link(i
->dev
, j
->dev
);
5476 /* add dev to every upper_dev's upper device */
5477 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
) {
5478 pr_debug("linking %s's upper device %s with %s\n",
5479 upper_dev
->name
, i
->dev
->name
, dev
->name
);
5480 ret
= __netdev_adjacent_dev_link(dev
, i
->dev
);
5482 goto rollback_upper_mesh
;
5485 /* add upper_dev to every dev's lower device */
5486 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5487 pr_debug("linking %s's lower device %s with %s\n", dev
->name
,
5488 i
->dev
->name
, upper_dev
->name
);
5489 ret
= __netdev_adjacent_dev_link(i
->dev
, upper_dev
);
5491 goto rollback_lower_mesh
;
5494 ret
= call_netdevice_notifiers_info(NETDEV_CHANGEUPPER
, dev
,
5495 &changeupper_info
.info
);
5496 ret
= notifier_to_errno(ret
);
5498 goto rollback_lower_mesh
;
5502 rollback_lower_mesh
:
5504 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5507 __netdev_adjacent_dev_unlink(i
->dev
, upper_dev
);
5512 rollback_upper_mesh
:
5514 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
) {
5517 __netdev_adjacent_dev_unlink(dev
, i
->dev
);
5525 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5526 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
) {
5527 if (i
== to_i
&& j
== to_j
)
5529 __netdev_adjacent_dev_unlink(i
->dev
, j
->dev
);
5535 __netdev_adjacent_dev_unlink_neighbour(dev
, upper_dev
);
5541 * netdev_upper_dev_link - Add a link to the upper device
5543 * @upper_dev: new upper device
5545 * Adds a link to device which is upper to this one. The caller must hold
5546 * the RTNL lock. On a failure a negative errno code is returned.
5547 * On success the reference counts are adjusted and the function
5550 int netdev_upper_dev_link(struct net_device
*dev
,
5551 struct net_device
*upper_dev
)
5553 return __netdev_upper_dev_link(dev
, upper_dev
, false, NULL
, NULL
);
5555 EXPORT_SYMBOL(netdev_upper_dev_link
);
5558 * netdev_master_upper_dev_link - Add a master link to the upper device
5560 * @upper_dev: new upper device
5561 * @upper_priv: upper device private
5562 * @upper_info: upper info to be passed down via notifier
5564 * Adds a link to device which is upper to this one. In this case, only
5565 * one master upper device can be linked, although other non-master devices
5566 * might be linked as well. The caller must hold the RTNL lock.
5567 * On a failure a negative errno code is returned. On success the reference
5568 * counts are adjusted and the function returns zero.
5570 int netdev_master_upper_dev_link(struct net_device
*dev
,
5571 struct net_device
*upper_dev
,
5572 void *upper_priv
, void *upper_info
)
5574 return __netdev_upper_dev_link(dev
, upper_dev
, true,
5575 upper_priv
, upper_info
);
5577 EXPORT_SYMBOL(netdev_master_upper_dev_link
);
5580 * netdev_upper_dev_unlink - Removes a link to upper device
5582 * @upper_dev: new upper device
5584 * Removes a link to device which is upper to this one. The caller must hold
5587 void netdev_upper_dev_unlink(struct net_device
*dev
,
5588 struct net_device
*upper_dev
)
5590 struct netdev_notifier_changeupper_info changeupper_info
;
5591 struct netdev_adjacent
*i
, *j
;
5594 changeupper_info
.upper_dev
= upper_dev
;
5595 changeupper_info
.master
= netdev_master_upper_dev_get(dev
) == upper_dev
;
5596 changeupper_info
.linking
= false;
5598 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER
, dev
,
5599 &changeupper_info
.info
);
5601 __netdev_adjacent_dev_unlink_neighbour(dev
, upper_dev
);
5603 /* Here is the tricky part. We must remove all dev's lower
5604 * devices from all upper_dev's upper devices and vice
5605 * versa, to maintain the graph relationship.
5607 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
)
5608 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
)
5609 __netdev_adjacent_dev_unlink(i
->dev
, j
->dev
);
5611 /* remove also the devices itself from lower/upper device
5614 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
)
5615 __netdev_adjacent_dev_unlink(i
->dev
, upper_dev
);
5617 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
)
5618 __netdev_adjacent_dev_unlink(dev
, i
->dev
);
5620 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER
, dev
,
5621 &changeupper_info
.info
);
5623 EXPORT_SYMBOL(netdev_upper_dev_unlink
);
5626 * netdev_bonding_info_change - Dispatch event about slave change
5628 * @bonding_info: info to dispatch
5630 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
5631 * The caller must hold the RTNL lock.
5633 void netdev_bonding_info_change(struct net_device
*dev
,
5634 struct netdev_bonding_info
*bonding_info
)
5636 struct netdev_notifier_bonding_info info
;
5638 memcpy(&info
.bonding_info
, bonding_info
,
5639 sizeof(struct netdev_bonding_info
));
5640 call_netdevice_notifiers_info(NETDEV_BONDING_INFO
, dev
,
5643 EXPORT_SYMBOL(netdev_bonding_info_change
);
5645 static void netdev_adjacent_add_links(struct net_device
*dev
)
5647 struct netdev_adjacent
*iter
;
5649 struct net
*net
= dev_net(dev
);
5651 list_for_each_entry(iter
, &dev
->adj_list
.upper
, list
) {
5652 if (!net_eq(net
,dev_net(iter
->dev
)))
5654 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
5655 &iter
->dev
->adj_list
.lower
);
5656 netdev_adjacent_sysfs_add(dev
, iter
->dev
,
5657 &dev
->adj_list
.upper
);
5660 list_for_each_entry(iter
, &dev
->adj_list
.lower
, list
) {
5661 if (!net_eq(net
,dev_net(iter
->dev
)))
5663 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
5664 &iter
->dev
->adj_list
.upper
);
5665 netdev_adjacent_sysfs_add(dev
, iter
->dev
,
5666 &dev
->adj_list
.lower
);
5670 static void netdev_adjacent_del_links(struct net_device
*dev
)
5672 struct netdev_adjacent
*iter
;
5674 struct net
*net
= dev_net(dev
);
5676 list_for_each_entry(iter
, &dev
->adj_list
.upper
, list
) {
5677 if (!net_eq(net
,dev_net(iter
->dev
)))
5679 netdev_adjacent_sysfs_del(iter
->dev
, dev
->name
,
5680 &iter
->dev
->adj_list
.lower
);
5681 netdev_adjacent_sysfs_del(dev
, iter
->dev
->name
,
5682 &dev
->adj_list
.upper
);
5685 list_for_each_entry(iter
, &dev
->adj_list
.lower
, list
) {
5686 if (!net_eq(net
,dev_net(iter
->dev
)))
5688 netdev_adjacent_sysfs_del(iter
->dev
, dev
->name
,
5689 &iter
->dev
->adj_list
.upper
);
5690 netdev_adjacent_sysfs_del(dev
, iter
->dev
->name
,
5691 &dev
->adj_list
.lower
);
5695 void netdev_adjacent_rename_links(struct net_device
*dev
, char *oldname
)
5697 struct netdev_adjacent
*iter
;
5699 struct net
*net
= dev_net(dev
);
5701 list_for_each_entry(iter
, &dev
->adj_list
.upper
, list
) {
5702 if (!net_eq(net
,dev_net(iter
->dev
)))
5704 netdev_adjacent_sysfs_del(iter
->dev
, oldname
,
5705 &iter
->dev
->adj_list
.lower
);
5706 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
5707 &iter
->dev
->adj_list
.lower
);
5710 list_for_each_entry(iter
, &dev
->adj_list
.lower
, list
) {
5711 if (!net_eq(net
,dev_net(iter
->dev
)))
5713 netdev_adjacent_sysfs_del(iter
->dev
, oldname
,
5714 &iter
->dev
->adj_list
.upper
);
5715 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
5716 &iter
->dev
->adj_list
.upper
);
5720 void *netdev_lower_dev_get_private(struct net_device
*dev
,
5721 struct net_device
*lower_dev
)
5723 struct netdev_adjacent
*lower
;
5727 lower
= __netdev_find_adj(lower_dev
, &dev
->adj_list
.lower
);
5731 return lower
->private;
5733 EXPORT_SYMBOL(netdev_lower_dev_get_private
);
5736 int dev_get_nest_level(struct net_device
*dev
,
5737 bool (*type_check
)(struct net_device
*dev
))
5739 struct net_device
*lower
= NULL
;
5740 struct list_head
*iter
;
5746 netdev_for_each_lower_dev(dev
, lower
, iter
) {
5747 nest
= dev_get_nest_level(lower
, type_check
);
5748 if (max_nest
< nest
)
5752 if (type_check(dev
))
5757 EXPORT_SYMBOL(dev_get_nest_level
);
5759 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
5761 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5763 if (ops
->ndo_change_rx_flags
)
5764 ops
->ndo_change_rx_flags(dev
, flags
);
5767 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
, bool notify
)
5769 unsigned int old_flags
= dev
->flags
;
5775 dev
->flags
|= IFF_PROMISC
;
5776 dev
->promiscuity
+= inc
;
5777 if (dev
->promiscuity
== 0) {
5780 * If inc causes overflow, untouch promisc and return error.
5783 dev
->flags
&= ~IFF_PROMISC
;
5785 dev
->promiscuity
-= inc
;
5786 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5791 if (dev
->flags
!= old_flags
) {
5792 pr_info("device %s %s promiscuous mode\n",
5794 dev
->flags
& IFF_PROMISC
? "entered" : "left");
5795 if (audit_enabled
) {
5796 current_uid_gid(&uid
, &gid
);
5797 audit_log(current
->audit_context
, GFP_ATOMIC
,
5798 AUDIT_ANOM_PROMISCUOUS
,
5799 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5800 dev
->name
, (dev
->flags
& IFF_PROMISC
),
5801 (old_flags
& IFF_PROMISC
),
5802 from_kuid(&init_user_ns
, audit_get_loginuid(current
)),
5803 from_kuid(&init_user_ns
, uid
),
5804 from_kgid(&init_user_ns
, gid
),
5805 audit_get_sessionid(current
));
5808 dev_change_rx_flags(dev
, IFF_PROMISC
);
5811 __dev_notify_flags(dev
, old_flags
, IFF_PROMISC
);
5816 * dev_set_promiscuity - update promiscuity count on a device
5820 * Add or remove promiscuity from a device. While the count in the device
5821 * remains above zero the interface remains promiscuous. Once it hits zero
5822 * the device reverts back to normal filtering operation. A negative inc
5823 * value is used to drop promiscuity on the device.
5824 * Return 0 if successful or a negative errno code on error.
5826 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
5828 unsigned int old_flags
= dev
->flags
;
5831 err
= __dev_set_promiscuity(dev
, inc
, true);
5834 if (dev
->flags
!= old_flags
)
5835 dev_set_rx_mode(dev
);
5838 EXPORT_SYMBOL(dev_set_promiscuity
);
5840 static int __dev_set_allmulti(struct net_device
*dev
, int inc
, bool notify
)
5842 unsigned int old_flags
= dev
->flags
, old_gflags
= dev
->gflags
;
5846 dev
->flags
|= IFF_ALLMULTI
;
5847 dev
->allmulti
+= inc
;
5848 if (dev
->allmulti
== 0) {
5851 * If inc causes overflow, untouch allmulti and return error.
5854 dev
->flags
&= ~IFF_ALLMULTI
;
5856 dev
->allmulti
-= inc
;
5857 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5862 if (dev
->flags
^ old_flags
) {
5863 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
5864 dev_set_rx_mode(dev
);
5866 __dev_notify_flags(dev
, old_flags
,
5867 dev
->gflags
^ old_gflags
);
5873 * dev_set_allmulti - update allmulti count on a device
5877 * Add or remove reception of all multicast frames to a device. While the
5878 * count in the device remains above zero the interface remains listening
5879 * to all interfaces. Once it hits zero the device reverts back to normal
5880 * filtering operation. A negative @inc value is used to drop the counter
5881 * when releasing a resource needing all multicasts.
5882 * Return 0 if successful or a negative errno code on error.
5885 int dev_set_allmulti(struct net_device
*dev
, int inc
)
5887 return __dev_set_allmulti(dev
, inc
, true);
5889 EXPORT_SYMBOL(dev_set_allmulti
);
5892 * Upload unicast and multicast address lists to device and
5893 * configure RX filtering. When the device doesn't support unicast
5894 * filtering it is put in promiscuous mode while unicast addresses
5897 void __dev_set_rx_mode(struct net_device
*dev
)
5899 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5901 /* dev_open will call this function so the list will stay sane. */
5902 if (!(dev
->flags
&IFF_UP
))
5905 if (!netif_device_present(dev
))
5908 if (!(dev
->priv_flags
& IFF_UNICAST_FLT
)) {
5909 /* Unicast addresses changes may only happen under the rtnl,
5910 * therefore calling __dev_set_promiscuity here is safe.
5912 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
5913 __dev_set_promiscuity(dev
, 1, false);
5914 dev
->uc_promisc
= true;
5915 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
5916 __dev_set_promiscuity(dev
, -1, false);
5917 dev
->uc_promisc
= false;
5921 if (ops
->ndo_set_rx_mode
)
5922 ops
->ndo_set_rx_mode(dev
);
5925 void dev_set_rx_mode(struct net_device
*dev
)
5927 netif_addr_lock_bh(dev
);
5928 __dev_set_rx_mode(dev
);
5929 netif_addr_unlock_bh(dev
);
5933 * dev_get_flags - get flags reported to userspace
5936 * Get the combination of flag bits exported through APIs to userspace.
5938 unsigned int dev_get_flags(const struct net_device
*dev
)
5942 flags
= (dev
->flags
& ~(IFF_PROMISC
|
5947 (dev
->gflags
& (IFF_PROMISC
|
5950 if (netif_running(dev
)) {
5951 if (netif_oper_up(dev
))
5952 flags
|= IFF_RUNNING
;
5953 if (netif_carrier_ok(dev
))
5954 flags
|= IFF_LOWER_UP
;
5955 if (netif_dormant(dev
))
5956 flags
|= IFF_DORMANT
;
5961 EXPORT_SYMBOL(dev_get_flags
);
5963 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
5965 unsigned int old_flags
= dev
->flags
;
5971 * Set the flags on our device.
5974 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
5975 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
5977 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
5981 * Load in the correct multicast list now the flags have changed.
5984 if ((old_flags
^ flags
) & IFF_MULTICAST
)
5985 dev_change_rx_flags(dev
, IFF_MULTICAST
);
5987 dev_set_rx_mode(dev
);
5990 * Have we downed the interface. We handle IFF_UP ourselves
5991 * according to user attempts to set it, rather than blindly
5996 if ((old_flags
^ flags
) & IFF_UP
)
5997 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
5999 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
6000 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
6001 unsigned int old_flags
= dev
->flags
;
6003 dev
->gflags
^= IFF_PROMISC
;
6005 if (__dev_set_promiscuity(dev
, inc
, false) >= 0)
6006 if (dev
->flags
!= old_flags
)
6007 dev_set_rx_mode(dev
);
6010 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
6011 is important. Some (broken) drivers set IFF_PROMISC, when
6012 IFF_ALLMULTI is requested not asking us and not reporting.
6014 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
6015 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
6017 dev
->gflags
^= IFF_ALLMULTI
;
6018 __dev_set_allmulti(dev
, inc
, false);
6024 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
,
6025 unsigned int gchanges
)
6027 unsigned int changes
= dev
->flags
^ old_flags
;
6030 rtmsg_ifinfo(RTM_NEWLINK
, dev
, gchanges
, GFP_ATOMIC
);
6032 if (changes
& IFF_UP
) {
6033 if (dev
->flags
& IFF_UP
)
6034 call_netdevice_notifiers(NETDEV_UP
, dev
);
6036 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
6039 if (dev
->flags
& IFF_UP
&&
6040 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
))) {
6041 struct netdev_notifier_change_info change_info
;
6043 change_info
.flags_changed
= changes
;
6044 call_netdevice_notifiers_info(NETDEV_CHANGE
, dev
,
6050 * dev_change_flags - change device settings
6052 * @flags: device state flags
6054 * Change settings on device based state flags. The flags are
6055 * in the userspace exported format.
6057 int dev_change_flags(struct net_device
*dev
, unsigned int flags
)
6060 unsigned int changes
, old_flags
= dev
->flags
, old_gflags
= dev
->gflags
;
6062 ret
= __dev_change_flags(dev
, flags
);
6066 changes
= (old_flags
^ dev
->flags
) | (old_gflags
^ dev
->gflags
);
6067 __dev_notify_flags(dev
, old_flags
, changes
);
6070 EXPORT_SYMBOL(dev_change_flags
);
6072 static int __dev_set_mtu(struct net_device
*dev
, int new_mtu
)
6074 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6076 if (ops
->ndo_change_mtu
)
6077 return ops
->ndo_change_mtu(dev
, new_mtu
);
6084 * dev_set_mtu - Change maximum transfer unit
6086 * @new_mtu: new transfer unit
6088 * Change the maximum transfer size of the network device.
6090 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
6094 if (new_mtu
== dev
->mtu
)
6097 /* MTU must be positive. */
6101 if (!netif_device_present(dev
))
6104 err
= call_netdevice_notifiers(NETDEV_PRECHANGEMTU
, dev
);
6105 err
= notifier_to_errno(err
);
6109 orig_mtu
= dev
->mtu
;
6110 err
= __dev_set_mtu(dev
, new_mtu
);
6113 err
= call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
6114 err
= notifier_to_errno(err
);
6116 /* setting mtu back and notifying everyone again,
6117 * so that they have a chance to revert changes.
6119 __dev_set_mtu(dev
, orig_mtu
);
6120 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
6125 EXPORT_SYMBOL(dev_set_mtu
);
6128 * dev_set_group - Change group this device belongs to
6130 * @new_group: group this device should belong to
6132 void dev_set_group(struct net_device
*dev
, int new_group
)
6134 dev
->group
= new_group
;
6136 EXPORT_SYMBOL(dev_set_group
);
6139 * dev_set_mac_address - Change Media Access Control Address
6143 * Change the hardware (MAC) address of the device
6145 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
6147 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6150 if (!ops
->ndo_set_mac_address
)
6152 if (sa
->sa_family
!= dev
->type
)
6154 if (!netif_device_present(dev
))
6156 err
= ops
->ndo_set_mac_address(dev
, sa
);
6159 dev
->addr_assign_type
= NET_ADDR_SET
;
6160 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
6161 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
6164 EXPORT_SYMBOL(dev_set_mac_address
);
6167 * dev_change_carrier - Change device carrier
6169 * @new_carrier: new value
6171 * Change device carrier
6173 int dev_change_carrier(struct net_device
*dev
, bool new_carrier
)
6175 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6177 if (!ops
->ndo_change_carrier
)
6179 if (!netif_device_present(dev
))
6181 return ops
->ndo_change_carrier(dev
, new_carrier
);
6183 EXPORT_SYMBOL(dev_change_carrier
);
6186 * dev_get_phys_port_id - Get device physical port ID
6190 * Get device physical port ID
6192 int dev_get_phys_port_id(struct net_device
*dev
,
6193 struct netdev_phys_item_id
*ppid
)
6195 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6197 if (!ops
->ndo_get_phys_port_id
)
6199 return ops
->ndo_get_phys_port_id(dev
, ppid
);
6201 EXPORT_SYMBOL(dev_get_phys_port_id
);
6204 * dev_get_phys_port_name - Get device physical port name
6208 * Get device physical port name
6210 int dev_get_phys_port_name(struct net_device
*dev
,
6211 char *name
, size_t len
)
6213 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6215 if (!ops
->ndo_get_phys_port_name
)
6217 return ops
->ndo_get_phys_port_name(dev
, name
, len
);
6219 EXPORT_SYMBOL(dev_get_phys_port_name
);
6222 * dev_change_proto_down - update protocol port state information
6224 * @proto_down: new value
6226 * This info can be used by switch drivers to set the phys state of the
6229 int dev_change_proto_down(struct net_device
*dev
, bool proto_down
)
6231 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6233 if (!ops
->ndo_change_proto_down
)
6235 if (!netif_device_present(dev
))
6237 return ops
->ndo_change_proto_down(dev
, proto_down
);
6239 EXPORT_SYMBOL(dev_change_proto_down
);
6242 * dev_new_index - allocate an ifindex
6243 * @net: the applicable net namespace
6245 * Returns a suitable unique value for a new device interface
6246 * number. The caller must hold the rtnl semaphore or the
6247 * dev_base_lock to be sure it remains unique.
6249 static int dev_new_index(struct net
*net
)
6251 int ifindex
= net
->ifindex
;
6255 if (!__dev_get_by_index(net
, ifindex
))
6256 return net
->ifindex
= ifindex
;
6260 /* Delayed registration/unregisteration */
6261 static LIST_HEAD(net_todo_list
);
6262 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq
);
6264 static void net_set_todo(struct net_device
*dev
)
6266 list_add_tail(&dev
->todo_list
, &net_todo_list
);
6267 dev_net(dev
)->dev_unreg_count
++;
6270 static void rollback_registered_many(struct list_head
*head
)
6272 struct net_device
*dev
, *tmp
;
6273 LIST_HEAD(close_head
);
6275 BUG_ON(dev_boot_phase
);
6278 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
6279 /* Some devices call without registering
6280 * for initialization unwind. Remove those
6281 * devices and proceed with the remaining.
6283 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
6284 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
6288 list_del(&dev
->unreg_list
);
6291 dev
->dismantle
= true;
6292 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
6295 /* If device is running, close it first. */
6296 list_for_each_entry(dev
, head
, unreg_list
)
6297 list_add_tail(&dev
->close_list
, &close_head
);
6298 dev_close_many(&close_head
, true);
6300 list_for_each_entry(dev
, head
, unreg_list
) {
6301 /* And unlink it from device chain. */
6302 unlist_netdevice(dev
);
6304 dev
->reg_state
= NETREG_UNREGISTERING
;
6305 on_each_cpu(flush_backlog
, dev
, 1);
6310 list_for_each_entry(dev
, head
, unreg_list
) {
6311 struct sk_buff
*skb
= NULL
;
6313 /* Shutdown queueing discipline. */
6317 /* Notify protocols, that we are about to destroy
6318 this device. They should clean all the things.
6320 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6322 if (!dev
->rtnl_link_ops
||
6323 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
6324 skb
= rtmsg_ifinfo_build_skb(RTM_DELLINK
, dev
, ~0U,
6328 * Flush the unicast and multicast chains
6333 if (dev
->netdev_ops
->ndo_uninit
)
6334 dev
->netdev_ops
->ndo_uninit(dev
);
6337 rtmsg_ifinfo_send(skb
, dev
, GFP_KERNEL
);
6339 /* Notifier chain MUST detach us all upper devices. */
6340 WARN_ON(netdev_has_any_upper_dev(dev
));
6342 /* Remove entries from kobject tree */
6343 netdev_unregister_kobject(dev
);
6345 /* Remove XPS queueing entries */
6346 netif_reset_xps_queues_gt(dev
, 0);
6352 list_for_each_entry(dev
, head
, unreg_list
)
6356 static void rollback_registered(struct net_device
*dev
)
6360 list_add(&dev
->unreg_list
, &single
);
6361 rollback_registered_many(&single
);
6365 static netdev_features_t
netdev_sync_upper_features(struct net_device
*lower
,
6366 struct net_device
*upper
, netdev_features_t features
)
6368 netdev_features_t upper_disables
= NETIF_F_UPPER_DISABLES
;
6369 netdev_features_t feature
;
6372 for_each_netdev_feature(&upper_disables
, feature_bit
) {
6373 feature
= __NETIF_F_BIT(feature_bit
);
6374 if (!(upper
->wanted_features
& feature
)
6375 && (features
& feature
)) {
6376 netdev_dbg(lower
, "Dropping feature %pNF, upper dev %s has it off.\n",
6377 &feature
, upper
->name
);
6378 features
&= ~feature
;
6385 static void netdev_sync_lower_features(struct net_device
*upper
,
6386 struct net_device
*lower
, netdev_features_t features
)
6388 netdev_features_t upper_disables
= NETIF_F_UPPER_DISABLES
;
6389 netdev_features_t feature
;
6392 for_each_netdev_feature(&upper_disables
, feature_bit
) {
6393 feature
= __NETIF_F_BIT(feature_bit
);
6394 if (!(features
& feature
) && (lower
->features
& feature
)) {
6395 netdev_dbg(upper
, "Disabling feature %pNF on lower dev %s.\n",
6396 &feature
, lower
->name
);
6397 lower
->wanted_features
&= ~feature
;
6398 netdev_update_features(lower
);
6400 if (unlikely(lower
->features
& feature
))
6401 netdev_WARN(upper
, "failed to disable %pNF on %s!\n",
6402 &feature
, lower
->name
);
6407 static netdev_features_t
netdev_fix_features(struct net_device
*dev
,
6408 netdev_features_t features
)
6410 /* Fix illegal checksum combinations */
6411 if ((features
& NETIF_F_HW_CSUM
) &&
6412 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
6413 netdev_warn(dev
, "mixed HW and IP checksum settings.\n");
6414 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
6417 /* TSO requires that SG is present as well. */
6418 if ((features
& NETIF_F_ALL_TSO
) && !(features
& NETIF_F_SG
)) {
6419 netdev_dbg(dev
, "Dropping TSO features since no SG feature.\n");
6420 features
&= ~NETIF_F_ALL_TSO
;
6423 if ((features
& NETIF_F_TSO
) && !(features
& NETIF_F_HW_CSUM
) &&
6424 !(features
& NETIF_F_IP_CSUM
)) {
6425 netdev_dbg(dev
, "Dropping TSO features since no CSUM feature.\n");
6426 features
&= ~NETIF_F_TSO
;
6427 features
&= ~NETIF_F_TSO_ECN
;
6430 if ((features
& NETIF_F_TSO6
) && !(features
& NETIF_F_HW_CSUM
) &&
6431 !(features
& NETIF_F_IPV6_CSUM
)) {
6432 netdev_dbg(dev
, "Dropping TSO6 features since no CSUM feature.\n");
6433 features
&= ~NETIF_F_TSO6
;
6436 /* TSO ECN requires that TSO is present as well. */
6437 if ((features
& NETIF_F_ALL_TSO
) == NETIF_F_TSO_ECN
)
6438 features
&= ~NETIF_F_TSO_ECN
;
6440 /* Software GSO depends on SG. */
6441 if ((features
& NETIF_F_GSO
) && !(features
& NETIF_F_SG
)) {
6442 netdev_dbg(dev
, "Dropping NETIF_F_GSO since no SG feature.\n");
6443 features
&= ~NETIF_F_GSO
;
6446 /* UFO needs SG and checksumming */
6447 if (features
& NETIF_F_UFO
) {
6448 /* maybe split UFO into V4 and V6? */
6449 if (!((features
& NETIF_F_GEN_CSUM
) ||
6450 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))
6451 == (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
6453 "Dropping NETIF_F_UFO since no checksum offload features.\n");
6454 features
&= ~NETIF_F_UFO
;
6457 if (!(features
& NETIF_F_SG
)) {
6459 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
6460 features
&= ~NETIF_F_UFO
;
6464 #ifdef CONFIG_NET_RX_BUSY_POLL
6465 if (dev
->netdev_ops
->ndo_busy_poll
)
6466 features
|= NETIF_F_BUSY_POLL
;
6469 features
&= ~NETIF_F_BUSY_POLL
;
6474 int __netdev_update_features(struct net_device
*dev
)
6476 struct net_device
*upper
, *lower
;
6477 netdev_features_t features
;
6478 struct list_head
*iter
;
6483 features
= netdev_get_wanted_features(dev
);
6485 if (dev
->netdev_ops
->ndo_fix_features
)
6486 features
= dev
->netdev_ops
->ndo_fix_features(dev
, features
);
6488 /* driver might be less strict about feature dependencies */
6489 features
= netdev_fix_features(dev
, features
);
6491 /* some features can't be enabled if they're off an an upper device */
6492 netdev_for_each_upper_dev_rcu(dev
, upper
, iter
)
6493 features
= netdev_sync_upper_features(dev
, upper
, features
);
6495 if (dev
->features
== features
)
6498 netdev_dbg(dev
, "Features changed: %pNF -> %pNF\n",
6499 &dev
->features
, &features
);
6501 if (dev
->netdev_ops
->ndo_set_features
)
6502 err
= dev
->netdev_ops
->ndo_set_features(dev
, features
);
6506 if (unlikely(err
< 0)) {
6508 "set_features() failed (%d); wanted %pNF, left %pNF\n",
6509 err
, &features
, &dev
->features
);
6510 /* return non-0 since some features might have changed and
6511 * it's better to fire a spurious notification than miss it
6517 /* some features must be disabled on lower devices when disabled
6518 * on an upper device (think: bonding master or bridge)
6520 netdev_for_each_lower_dev(dev
, lower
, iter
)
6521 netdev_sync_lower_features(dev
, lower
, features
);
6524 dev
->features
= features
;
6526 return err
< 0 ? 0 : 1;
6530 * netdev_update_features - recalculate device features
6531 * @dev: the device to check
6533 * Recalculate dev->features set and send notifications if it
6534 * has changed. Should be called after driver or hardware dependent
6535 * conditions might have changed that influence the features.
6537 void netdev_update_features(struct net_device
*dev
)
6539 if (__netdev_update_features(dev
))
6540 netdev_features_change(dev
);
6542 EXPORT_SYMBOL(netdev_update_features
);
6545 * netdev_change_features - recalculate device features
6546 * @dev: the device to check
6548 * Recalculate dev->features set and send notifications even
6549 * if they have not changed. Should be called instead of
6550 * netdev_update_features() if also dev->vlan_features might
6551 * have changed to allow the changes to be propagated to stacked
6554 void netdev_change_features(struct net_device
*dev
)
6556 __netdev_update_features(dev
);
6557 netdev_features_change(dev
);
6559 EXPORT_SYMBOL(netdev_change_features
);
6562 * netif_stacked_transfer_operstate - transfer operstate
6563 * @rootdev: the root or lower level device to transfer state from
6564 * @dev: the device to transfer operstate to
6566 * Transfer operational state from root to device. This is normally
6567 * called when a stacking relationship exists between the root
6568 * device and the device(a leaf device).
6570 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
6571 struct net_device
*dev
)
6573 if (rootdev
->operstate
== IF_OPER_DORMANT
)
6574 netif_dormant_on(dev
);
6576 netif_dormant_off(dev
);
6578 if (netif_carrier_ok(rootdev
)) {
6579 if (!netif_carrier_ok(dev
))
6580 netif_carrier_on(dev
);
6582 if (netif_carrier_ok(dev
))
6583 netif_carrier_off(dev
);
6586 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
6589 static int netif_alloc_rx_queues(struct net_device
*dev
)
6591 unsigned int i
, count
= dev
->num_rx_queues
;
6592 struct netdev_rx_queue
*rx
;
6593 size_t sz
= count
* sizeof(*rx
);
6597 rx
= kzalloc(sz
, GFP_KERNEL
| __GFP_NOWARN
| __GFP_REPEAT
);
6605 for (i
= 0; i
< count
; i
++)
6611 static void netdev_init_one_queue(struct net_device
*dev
,
6612 struct netdev_queue
*queue
, void *_unused
)
6614 /* Initialize queue lock */
6615 spin_lock_init(&queue
->_xmit_lock
);
6616 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
6617 queue
->xmit_lock_owner
= -1;
6618 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
6621 dql_init(&queue
->dql
, HZ
);
6625 static void netif_free_tx_queues(struct net_device
*dev
)
6630 static int netif_alloc_netdev_queues(struct net_device
*dev
)
6632 unsigned int count
= dev
->num_tx_queues
;
6633 struct netdev_queue
*tx
;
6634 size_t sz
= count
* sizeof(*tx
);
6636 if (count
< 1 || count
> 0xffff)
6639 tx
= kzalloc(sz
, GFP_KERNEL
| __GFP_NOWARN
| __GFP_REPEAT
);
6647 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
6648 spin_lock_init(&dev
->tx_global_lock
);
6653 void netif_tx_stop_all_queues(struct net_device
*dev
)
6657 for (i
= 0; i
< dev
->num_tx_queues
; i
++) {
6658 struct netdev_queue
*txq
= netdev_get_tx_queue(dev
, i
);
6659 netif_tx_stop_queue(txq
);
6662 EXPORT_SYMBOL(netif_tx_stop_all_queues
);
6665 * register_netdevice - register a network device
6666 * @dev: device to register
6668 * Take a completed network device structure and add it to the kernel
6669 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6670 * chain. 0 is returned on success. A negative errno code is returned
6671 * on a failure to set up the device, or if the name is a duplicate.
6673 * Callers must hold the rtnl semaphore. You may want
6674 * register_netdev() instead of this.
6677 * The locking appears insufficient to guarantee two parallel registers
6678 * will not get the same name.
6681 int register_netdevice(struct net_device
*dev
)
6684 struct net
*net
= dev_net(dev
);
6686 BUG_ON(dev_boot_phase
);
6691 /* When net_device's are persistent, this will be fatal. */
6692 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
6695 spin_lock_init(&dev
->addr_list_lock
);
6696 netdev_set_addr_lockdep_class(dev
);
6698 ret
= dev_get_valid_name(net
, dev
, dev
->name
);
6702 /* Init, if this function is available */
6703 if (dev
->netdev_ops
->ndo_init
) {
6704 ret
= dev
->netdev_ops
->ndo_init(dev
);
6712 if (((dev
->hw_features
| dev
->features
) &
6713 NETIF_F_HW_VLAN_CTAG_FILTER
) &&
6714 (!dev
->netdev_ops
->ndo_vlan_rx_add_vid
||
6715 !dev
->netdev_ops
->ndo_vlan_rx_kill_vid
)) {
6716 netdev_WARN(dev
, "Buggy VLAN acceleration in driver!\n");
6723 dev
->ifindex
= dev_new_index(net
);
6724 else if (__dev_get_by_index(net
, dev
->ifindex
))
6727 /* Transfer changeable features to wanted_features and enable
6728 * software offloads (GSO and GRO).
6730 dev
->hw_features
|= NETIF_F_SOFT_FEATURES
;
6731 dev
->features
|= NETIF_F_SOFT_FEATURES
;
6732 dev
->wanted_features
= dev
->features
& dev
->hw_features
;
6734 if (!(dev
->flags
& IFF_LOOPBACK
)) {
6735 dev
->hw_features
|= NETIF_F_NOCACHE_COPY
;
6738 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6740 dev
->vlan_features
|= NETIF_F_HIGHDMA
;
6742 /* Make NETIF_F_SG inheritable to tunnel devices.
6744 dev
->hw_enc_features
|= NETIF_F_SG
;
6746 /* Make NETIF_F_SG inheritable to MPLS.
6748 dev
->mpls_features
|= NETIF_F_SG
;
6750 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
6751 ret
= notifier_to_errno(ret
);
6755 ret
= netdev_register_kobject(dev
);
6758 dev
->reg_state
= NETREG_REGISTERED
;
6760 __netdev_update_features(dev
);
6763 * Default initial state at registry is that the
6764 * device is present.
6767 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
6769 linkwatch_init_dev(dev
);
6771 dev_init_scheduler(dev
);
6773 list_netdevice(dev
);
6774 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
6776 /* If the device has permanent device address, driver should
6777 * set dev_addr and also addr_assign_type should be set to
6778 * NET_ADDR_PERM (default value).
6780 if (dev
->addr_assign_type
== NET_ADDR_PERM
)
6781 memcpy(dev
->perm_addr
, dev
->dev_addr
, dev
->addr_len
);
6783 /* Notify protocols, that a new device appeared. */
6784 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
6785 ret
= notifier_to_errno(ret
);
6787 rollback_registered(dev
);
6788 dev
->reg_state
= NETREG_UNREGISTERED
;
6791 * Prevent userspace races by waiting until the network
6792 * device is fully setup before sending notifications.
6794 if (!dev
->rtnl_link_ops
||
6795 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
6796 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U, GFP_KERNEL
);
6802 if (dev
->netdev_ops
->ndo_uninit
)
6803 dev
->netdev_ops
->ndo_uninit(dev
);
6806 EXPORT_SYMBOL(register_netdevice
);
6809 * init_dummy_netdev - init a dummy network device for NAPI
6810 * @dev: device to init
6812 * This takes a network device structure and initialize the minimum
6813 * amount of fields so it can be used to schedule NAPI polls without
6814 * registering a full blown interface. This is to be used by drivers
6815 * that need to tie several hardware interfaces to a single NAPI
6816 * poll scheduler due to HW limitations.
6818 int init_dummy_netdev(struct net_device
*dev
)
6820 /* Clear everything. Note we don't initialize spinlocks
6821 * are they aren't supposed to be taken by any of the
6822 * NAPI code and this dummy netdev is supposed to be
6823 * only ever used for NAPI polls
6825 memset(dev
, 0, sizeof(struct net_device
));
6827 /* make sure we BUG if trying to hit standard
6828 * register/unregister code path
6830 dev
->reg_state
= NETREG_DUMMY
;
6832 /* NAPI wants this */
6833 INIT_LIST_HEAD(&dev
->napi_list
);
6835 /* a dummy interface is started by default */
6836 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
6837 set_bit(__LINK_STATE_START
, &dev
->state
);
6839 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6840 * because users of this 'device' dont need to change
6846 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
6850 * register_netdev - register a network device
6851 * @dev: device to register
6853 * Take a completed network device structure and add it to the kernel
6854 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6855 * chain. 0 is returned on success. A negative errno code is returned
6856 * on a failure to set up the device, or if the name is a duplicate.
6858 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6859 * and expands the device name if you passed a format string to
6862 int register_netdev(struct net_device
*dev
)
6867 err
= register_netdevice(dev
);
6871 EXPORT_SYMBOL(register_netdev
);
6873 int netdev_refcnt_read(const struct net_device
*dev
)
6877 for_each_possible_cpu(i
)
6878 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
6881 EXPORT_SYMBOL(netdev_refcnt_read
);
6884 * netdev_wait_allrefs - wait until all references are gone.
6885 * @dev: target net_device
6887 * This is called when unregistering network devices.
6889 * Any protocol or device that holds a reference should register
6890 * for netdevice notification, and cleanup and put back the
6891 * reference if they receive an UNREGISTER event.
6892 * We can get stuck here if buggy protocols don't correctly
6895 static void netdev_wait_allrefs(struct net_device
*dev
)
6897 unsigned long rebroadcast_time
, warning_time
;
6900 linkwatch_forget_dev(dev
);
6902 rebroadcast_time
= warning_time
= jiffies
;
6903 refcnt
= netdev_refcnt_read(dev
);
6905 while (refcnt
!= 0) {
6906 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
6909 /* Rebroadcast unregister notification */
6910 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6916 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
6917 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
6919 /* We must not have linkwatch events
6920 * pending on unregister. If this
6921 * happens, we simply run the queue
6922 * unscheduled, resulting in a noop
6925 linkwatch_run_queue();
6930 rebroadcast_time
= jiffies
;
6935 refcnt
= netdev_refcnt_read(dev
);
6937 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
6938 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6940 warning_time
= jiffies
;
6949 * register_netdevice(x1);
6950 * register_netdevice(x2);
6952 * unregister_netdevice(y1);
6953 * unregister_netdevice(y2);
6959 * We are invoked by rtnl_unlock().
6960 * This allows us to deal with problems:
6961 * 1) We can delete sysfs objects which invoke hotplug
6962 * without deadlocking with linkwatch via keventd.
6963 * 2) Since we run with the RTNL semaphore not held, we can sleep
6964 * safely in order to wait for the netdev refcnt to drop to zero.
6966 * We must not return until all unregister events added during
6967 * the interval the lock was held have been completed.
6969 void netdev_run_todo(void)
6971 struct list_head list
;
6973 /* Snapshot list, allow later requests */
6974 list_replace_init(&net_todo_list
, &list
);
6979 /* Wait for rcu callbacks to finish before next phase */
6980 if (!list_empty(&list
))
6983 while (!list_empty(&list
)) {
6984 struct net_device
*dev
6985 = list_first_entry(&list
, struct net_device
, todo_list
);
6986 list_del(&dev
->todo_list
);
6989 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
6992 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
6993 pr_err("network todo '%s' but state %d\n",
6994 dev
->name
, dev
->reg_state
);
6999 dev
->reg_state
= NETREG_UNREGISTERED
;
7001 netdev_wait_allrefs(dev
);
7004 BUG_ON(netdev_refcnt_read(dev
));
7005 BUG_ON(!list_empty(&dev
->ptype_all
));
7006 BUG_ON(!list_empty(&dev
->ptype_specific
));
7007 WARN_ON(rcu_access_pointer(dev
->ip_ptr
));
7008 WARN_ON(rcu_access_pointer(dev
->ip6_ptr
));
7009 WARN_ON(dev
->dn_ptr
);
7011 if (dev
->destructor
)
7012 dev
->destructor(dev
);
7014 /* Report a network device has been unregistered */
7016 dev_net(dev
)->dev_unreg_count
--;
7018 wake_up(&netdev_unregistering_wq
);
7020 /* Free network device */
7021 kobject_put(&dev
->dev
.kobj
);
7025 /* Convert net_device_stats to rtnl_link_stats64. They have the same
7026 * fields in the same order, with only the type differing.
7028 void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
7029 const struct net_device_stats
*netdev_stats
)
7031 #if BITS_PER_LONG == 64
7032 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
7033 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
7035 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
7036 const unsigned long *src
= (const unsigned long *)netdev_stats
;
7037 u64
*dst
= (u64
*)stats64
;
7039 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
7040 sizeof(*stats64
) / sizeof(u64
));
7041 for (i
= 0; i
< n
; i
++)
7045 EXPORT_SYMBOL(netdev_stats_to_stats64
);
7048 * dev_get_stats - get network device statistics
7049 * @dev: device to get statistics from
7050 * @storage: place to store stats
7052 * Get network statistics from device. Return @storage.
7053 * The device driver may provide its own method by setting
7054 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
7055 * otherwise the internal statistics structure is used.
7057 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
7058 struct rtnl_link_stats64
*storage
)
7060 const struct net_device_ops
*ops
= dev
->netdev_ops
;
7062 if (ops
->ndo_get_stats64
) {
7063 memset(storage
, 0, sizeof(*storage
));
7064 ops
->ndo_get_stats64(dev
, storage
);
7065 } else if (ops
->ndo_get_stats
) {
7066 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
7068 netdev_stats_to_stats64(storage
, &dev
->stats
);
7070 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
7071 storage
->tx_dropped
+= atomic_long_read(&dev
->tx_dropped
);
7074 EXPORT_SYMBOL(dev_get_stats
);
7076 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
7078 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
7080 #ifdef CONFIG_NET_CLS_ACT
7083 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
7086 netdev_init_one_queue(dev
, queue
, NULL
);
7087 RCU_INIT_POINTER(queue
->qdisc
, &noop_qdisc
);
7088 queue
->qdisc_sleeping
= &noop_qdisc
;
7089 rcu_assign_pointer(dev
->ingress_queue
, queue
);
7094 static const struct ethtool_ops default_ethtool_ops
;
7096 void netdev_set_default_ethtool_ops(struct net_device
*dev
,
7097 const struct ethtool_ops
*ops
)
7099 if (dev
->ethtool_ops
== &default_ethtool_ops
)
7100 dev
->ethtool_ops
= ops
;
7102 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops
);
7104 void netdev_freemem(struct net_device
*dev
)
7106 char *addr
= (char *)dev
- dev
->padded
;
7112 * alloc_netdev_mqs - allocate network device
7113 * @sizeof_priv: size of private data to allocate space for
7114 * @name: device name format string
7115 * @name_assign_type: origin of device name
7116 * @setup: callback to initialize device
7117 * @txqs: the number of TX subqueues to allocate
7118 * @rxqs: the number of RX subqueues to allocate
7120 * Allocates a struct net_device with private data area for driver use
7121 * and performs basic initialization. Also allocates subqueue structs
7122 * for each queue on the device.
7124 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
7125 unsigned char name_assign_type
,
7126 void (*setup
)(struct net_device
*),
7127 unsigned int txqs
, unsigned int rxqs
)
7129 struct net_device
*dev
;
7131 struct net_device
*p
;
7133 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
7136 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
7142 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
7147 alloc_size
= sizeof(struct net_device
);
7149 /* ensure 32-byte alignment of private area */
7150 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
7151 alloc_size
+= sizeof_priv
;
7153 /* ensure 32-byte alignment of whole construct */
7154 alloc_size
+= NETDEV_ALIGN
- 1;
7156 p
= kzalloc(alloc_size
, GFP_KERNEL
| __GFP_NOWARN
| __GFP_REPEAT
);
7158 p
= vzalloc(alloc_size
);
7162 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
7163 dev
->padded
= (char *)dev
- (char *)p
;
7165 dev
->pcpu_refcnt
= alloc_percpu(int);
7166 if (!dev
->pcpu_refcnt
)
7169 if (dev_addr_init(dev
))
7175 dev_net_set(dev
, &init_net
);
7177 dev
->gso_max_size
= GSO_MAX_SIZE
;
7178 dev
->gso_max_segs
= GSO_MAX_SEGS
;
7179 dev
->gso_min_segs
= 0;
7181 INIT_LIST_HEAD(&dev
->napi_list
);
7182 INIT_LIST_HEAD(&dev
->unreg_list
);
7183 INIT_LIST_HEAD(&dev
->close_list
);
7184 INIT_LIST_HEAD(&dev
->link_watch_list
);
7185 INIT_LIST_HEAD(&dev
->adj_list
.upper
);
7186 INIT_LIST_HEAD(&dev
->adj_list
.lower
);
7187 INIT_LIST_HEAD(&dev
->all_adj_list
.upper
);
7188 INIT_LIST_HEAD(&dev
->all_adj_list
.lower
);
7189 INIT_LIST_HEAD(&dev
->ptype_all
);
7190 INIT_LIST_HEAD(&dev
->ptype_specific
);
7191 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
| IFF_XMIT_DST_RELEASE_PERM
;
7194 if (!dev
->tx_queue_len
)
7195 dev
->priv_flags
|= IFF_NO_QUEUE
;
7197 dev
->num_tx_queues
= txqs
;
7198 dev
->real_num_tx_queues
= txqs
;
7199 if (netif_alloc_netdev_queues(dev
))
7203 dev
->num_rx_queues
= rxqs
;
7204 dev
->real_num_rx_queues
= rxqs
;
7205 if (netif_alloc_rx_queues(dev
))
7209 strcpy(dev
->name
, name
);
7210 dev
->name_assign_type
= name_assign_type
;
7211 dev
->group
= INIT_NETDEV_GROUP
;
7212 if (!dev
->ethtool_ops
)
7213 dev
->ethtool_ops
= &default_ethtool_ops
;
7215 nf_hook_ingress_init(dev
);
7224 free_percpu(dev
->pcpu_refcnt
);
7226 netdev_freemem(dev
);
7229 EXPORT_SYMBOL(alloc_netdev_mqs
);
7232 * free_netdev - free network device
7235 * This function does the last stage of destroying an allocated device
7236 * interface. The reference to the device object is released.
7237 * If this is the last reference then it will be freed.
7238 * Must be called in process context.
7240 void free_netdev(struct net_device
*dev
)
7242 struct napi_struct
*p
, *n
;
7245 netif_free_tx_queues(dev
);
7250 kfree(rcu_dereference_protected(dev
->ingress_queue
, 1));
7252 /* Flush device addresses */
7253 dev_addr_flush(dev
);
7255 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
7258 free_percpu(dev
->pcpu_refcnt
);
7259 dev
->pcpu_refcnt
= NULL
;
7261 /* Compatibility with error handling in drivers */
7262 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
7263 netdev_freemem(dev
);
7267 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
7268 dev
->reg_state
= NETREG_RELEASED
;
7270 /* will free via device release */
7271 put_device(&dev
->dev
);
7273 EXPORT_SYMBOL(free_netdev
);
7276 * synchronize_net - Synchronize with packet receive processing
7278 * Wait for packets currently being received to be done.
7279 * Does not block later packets from starting.
7281 void synchronize_net(void)
7284 if (rtnl_is_locked())
7285 synchronize_rcu_expedited();
7289 EXPORT_SYMBOL(synchronize_net
);
7292 * unregister_netdevice_queue - remove device from the kernel
7296 * This function shuts down a device interface and removes it
7297 * from the kernel tables.
7298 * If head not NULL, device is queued to be unregistered later.
7300 * Callers must hold the rtnl semaphore. You may want
7301 * unregister_netdev() instead of this.
7304 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
7309 list_move_tail(&dev
->unreg_list
, head
);
7311 rollback_registered(dev
);
7312 /* Finish processing unregister after unlock */
7316 EXPORT_SYMBOL(unregister_netdevice_queue
);
7319 * unregister_netdevice_many - unregister many devices
7320 * @head: list of devices
7322 * Note: As most callers use a stack allocated list_head,
7323 * we force a list_del() to make sure stack wont be corrupted later.
7325 void unregister_netdevice_many(struct list_head
*head
)
7327 struct net_device
*dev
;
7329 if (!list_empty(head
)) {
7330 rollback_registered_many(head
);
7331 list_for_each_entry(dev
, head
, unreg_list
)
7336 EXPORT_SYMBOL(unregister_netdevice_many
);
7339 * unregister_netdev - remove device from the kernel
7342 * This function shuts down a device interface and removes it
7343 * from the kernel tables.
7345 * This is just a wrapper for unregister_netdevice that takes
7346 * the rtnl semaphore. In general you want to use this and not
7347 * unregister_netdevice.
7349 void unregister_netdev(struct net_device
*dev
)
7352 unregister_netdevice(dev
);
7355 EXPORT_SYMBOL(unregister_netdev
);
7358 * dev_change_net_namespace - move device to different nethost namespace
7360 * @net: network namespace
7361 * @pat: If not NULL name pattern to try if the current device name
7362 * is already taken in the destination network namespace.
7364 * This function shuts down a device interface and moves it
7365 * to a new network namespace. On success 0 is returned, on
7366 * a failure a netagive errno code is returned.
7368 * Callers must hold the rtnl semaphore.
7371 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
7377 /* Don't allow namespace local devices to be moved. */
7379 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
7382 /* Ensure the device has been registrered */
7383 if (dev
->reg_state
!= NETREG_REGISTERED
)
7386 /* Get out if there is nothing todo */
7388 if (net_eq(dev_net(dev
), net
))
7391 /* Pick the destination device name, and ensure
7392 * we can use it in the destination network namespace.
7395 if (__dev_get_by_name(net
, dev
->name
)) {
7396 /* We get here if we can't use the current device name */
7399 if (dev_get_valid_name(net
, dev
, pat
) < 0)
7404 * And now a mini version of register_netdevice unregister_netdevice.
7407 /* If device is running close it first. */
7410 /* And unlink it from device chain */
7412 unlist_netdevice(dev
);
7416 /* Shutdown queueing discipline. */
7419 /* Notify protocols, that we are about to destroy
7420 this device. They should clean all the things.
7422 Note that dev->reg_state stays at NETREG_REGISTERED.
7423 This is wanted because this way 8021q and macvlan know
7424 the device is just moving and can keep their slaves up.
7426 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
7428 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
7429 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U, GFP_KERNEL
);
7432 * Flush the unicast and multicast chains
7437 /* Send a netdev-removed uevent to the old namespace */
7438 kobject_uevent(&dev
->dev
.kobj
, KOBJ_REMOVE
);
7439 netdev_adjacent_del_links(dev
);
7441 /* Actually switch the network namespace */
7442 dev_net_set(dev
, net
);
7444 /* If there is an ifindex conflict assign a new one */
7445 if (__dev_get_by_index(net
, dev
->ifindex
))
7446 dev
->ifindex
= dev_new_index(net
);
7448 /* Send a netdev-add uevent to the new namespace */
7449 kobject_uevent(&dev
->dev
.kobj
, KOBJ_ADD
);
7450 netdev_adjacent_add_links(dev
);
7452 /* Fixup kobjects */
7453 err
= device_rename(&dev
->dev
, dev
->name
);
7456 /* Add the device back in the hashes */
7457 list_netdevice(dev
);
7459 /* Notify protocols, that a new device appeared. */
7460 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
7463 * Prevent userspace races by waiting until the network
7464 * device is fully setup before sending notifications.
7466 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U, GFP_KERNEL
);
7473 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
7475 static int dev_cpu_callback(struct notifier_block
*nfb
,
7476 unsigned long action
,
7479 struct sk_buff
**list_skb
;
7480 struct sk_buff
*skb
;
7481 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
7482 struct softnet_data
*sd
, *oldsd
;
7484 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
7487 local_irq_disable();
7488 cpu
= smp_processor_id();
7489 sd
= &per_cpu(softnet_data
, cpu
);
7490 oldsd
= &per_cpu(softnet_data
, oldcpu
);
7492 /* Find end of our completion_queue. */
7493 list_skb
= &sd
->completion_queue
;
7495 list_skb
= &(*list_skb
)->next
;
7496 /* Append completion queue from offline CPU. */
7497 *list_skb
= oldsd
->completion_queue
;
7498 oldsd
->completion_queue
= NULL
;
7500 /* Append output queue from offline CPU. */
7501 if (oldsd
->output_queue
) {
7502 *sd
->output_queue_tailp
= oldsd
->output_queue
;
7503 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
7504 oldsd
->output_queue
= NULL
;
7505 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
7507 /* Append NAPI poll list from offline CPU, with one exception :
7508 * process_backlog() must be called by cpu owning percpu backlog.
7509 * We properly handle process_queue & input_pkt_queue later.
7511 while (!list_empty(&oldsd
->poll_list
)) {
7512 struct napi_struct
*napi
= list_first_entry(&oldsd
->poll_list
,
7516 list_del_init(&napi
->poll_list
);
7517 if (napi
->poll
== process_backlog
)
7520 ____napi_schedule(sd
, napi
);
7523 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
7526 /* Process offline CPU's input_pkt_queue */
7527 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
7529 input_queue_head_incr(oldsd
);
7531 while ((skb
= skb_dequeue(&oldsd
->input_pkt_queue
))) {
7533 input_queue_head_incr(oldsd
);
7541 * netdev_increment_features - increment feature set by one
7542 * @all: current feature set
7543 * @one: new feature set
7544 * @mask: mask feature set
7546 * Computes a new feature set after adding a device with feature set
7547 * @one to the master device with current feature set @all. Will not
7548 * enable anything that is off in @mask. Returns the new feature set.
7550 netdev_features_t
netdev_increment_features(netdev_features_t all
,
7551 netdev_features_t one
, netdev_features_t mask
)
7553 if (mask
& NETIF_F_GEN_CSUM
)
7554 mask
|= NETIF_F_ALL_CSUM
;
7555 mask
|= NETIF_F_VLAN_CHALLENGED
;
7557 all
|= one
& (NETIF_F_ONE_FOR_ALL
|NETIF_F_ALL_CSUM
) & mask
;
7558 all
&= one
| ~NETIF_F_ALL_FOR_ALL
;
7560 /* If one device supports hw checksumming, set for all. */
7561 if (all
& NETIF_F_GEN_CSUM
)
7562 all
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_GEN_CSUM
);
7566 EXPORT_SYMBOL(netdev_increment_features
);
7568 static struct hlist_head
* __net_init
netdev_create_hash(void)
7571 struct hlist_head
*hash
;
7573 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
7575 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
7576 INIT_HLIST_HEAD(&hash
[i
]);
7581 /* Initialize per network namespace state */
7582 static int __net_init
netdev_init(struct net
*net
)
7584 if (net
!= &init_net
)
7585 INIT_LIST_HEAD(&net
->dev_base_head
);
7587 net
->dev_name_head
= netdev_create_hash();
7588 if (net
->dev_name_head
== NULL
)
7591 net
->dev_index_head
= netdev_create_hash();
7592 if (net
->dev_index_head
== NULL
)
7598 kfree(net
->dev_name_head
);
7604 * netdev_drivername - network driver for the device
7605 * @dev: network device
7607 * Determine network driver for device.
7609 const char *netdev_drivername(const struct net_device
*dev
)
7611 const struct device_driver
*driver
;
7612 const struct device
*parent
;
7613 const char *empty
= "";
7615 parent
= dev
->dev
.parent
;
7619 driver
= parent
->driver
;
7620 if (driver
&& driver
->name
)
7621 return driver
->name
;
7625 static void __netdev_printk(const char *level
, const struct net_device
*dev
,
7626 struct va_format
*vaf
)
7628 if (dev
&& dev
->dev
.parent
) {
7629 dev_printk_emit(level
[1] - '0',
7632 dev_driver_string(dev
->dev
.parent
),
7633 dev_name(dev
->dev
.parent
),
7634 netdev_name(dev
), netdev_reg_state(dev
),
7637 printk("%s%s%s: %pV",
7638 level
, netdev_name(dev
), netdev_reg_state(dev
), vaf
);
7640 printk("%s(NULL net_device): %pV", level
, vaf
);
7644 void netdev_printk(const char *level
, const struct net_device
*dev
,
7645 const char *format
, ...)
7647 struct va_format vaf
;
7650 va_start(args
, format
);
7655 __netdev_printk(level
, dev
, &vaf
);
7659 EXPORT_SYMBOL(netdev_printk
);
7661 #define define_netdev_printk_level(func, level) \
7662 void func(const struct net_device *dev, const char *fmt, ...) \
7664 struct va_format vaf; \
7667 va_start(args, fmt); \
7672 __netdev_printk(level, dev, &vaf); \
7676 EXPORT_SYMBOL(func);
7678 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
7679 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
7680 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
7681 define_netdev_printk_level(netdev_err
, KERN_ERR
);
7682 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
7683 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
7684 define_netdev_printk_level(netdev_info
, KERN_INFO
);
7686 static void __net_exit
netdev_exit(struct net
*net
)
7688 kfree(net
->dev_name_head
);
7689 kfree(net
->dev_index_head
);
7692 static struct pernet_operations __net_initdata netdev_net_ops
= {
7693 .init
= netdev_init
,
7694 .exit
= netdev_exit
,
7697 static void __net_exit
default_device_exit(struct net
*net
)
7699 struct net_device
*dev
, *aux
;
7701 * Push all migratable network devices back to the
7702 * initial network namespace
7705 for_each_netdev_safe(net
, dev
, aux
) {
7707 char fb_name
[IFNAMSIZ
];
7709 /* Ignore unmoveable devices (i.e. loopback) */
7710 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
7713 /* Leave virtual devices for the generic cleanup */
7714 if (dev
->rtnl_link_ops
)
7717 /* Push remaining network devices to init_net */
7718 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
7719 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
7721 pr_emerg("%s: failed to move %s to init_net: %d\n",
7722 __func__
, dev
->name
, err
);
7729 static void __net_exit
rtnl_lock_unregistering(struct list_head
*net_list
)
7731 /* Return with the rtnl_lock held when there are no network
7732 * devices unregistering in any network namespace in net_list.
7736 DEFINE_WAIT_FUNC(wait
, woken_wake_function
);
7738 add_wait_queue(&netdev_unregistering_wq
, &wait
);
7740 unregistering
= false;
7742 list_for_each_entry(net
, net_list
, exit_list
) {
7743 if (net
->dev_unreg_count
> 0) {
7744 unregistering
= true;
7752 wait_woken(&wait
, TASK_UNINTERRUPTIBLE
, MAX_SCHEDULE_TIMEOUT
);
7754 remove_wait_queue(&netdev_unregistering_wq
, &wait
);
7757 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
7759 /* At exit all network devices most be removed from a network
7760 * namespace. Do this in the reverse order of registration.
7761 * Do this across as many network namespaces as possible to
7762 * improve batching efficiency.
7764 struct net_device
*dev
;
7766 LIST_HEAD(dev_kill_list
);
7768 /* To prevent network device cleanup code from dereferencing
7769 * loopback devices or network devices that have been freed
7770 * wait here for all pending unregistrations to complete,
7771 * before unregistring the loopback device and allowing the
7772 * network namespace be freed.
7774 * The netdev todo list containing all network devices
7775 * unregistrations that happen in default_device_exit_batch
7776 * will run in the rtnl_unlock() at the end of
7777 * default_device_exit_batch.
7779 rtnl_lock_unregistering(net_list
);
7780 list_for_each_entry(net
, net_list
, exit_list
) {
7781 for_each_netdev_reverse(net
, dev
) {
7782 if (dev
->rtnl_link_ops
&& dev
->rtnl_link_ops
->dellink
)
7783 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
7785 unregister_netdevice_queue(dev
, &dev_kill_list
);
7788 unregister_netdevice_many(&dev_kill_list
);
7792 static struct pernet_operations __net_initdata default_device_ops
= {
7793 .exit
= default_device_exit
,
7794 .exit_batch
= default_device_exit_batch
,
7798 * Initialize the DEV module. At boot time this walks the device list and
7799 * unhooks any devices that fail to initialise (normally hardware not
7800 * present) and leaves us with a valid list of present and active devices.
7805 * This is called single threaded during boot, so no need
7806 * to take the rtnl semaphore.
7808 static int __init
net_dev_init(void)
7810 int i
, rc
= -ENOMEM
;
7812 BUG_ON(!dev_boot_phase
);
7814 if (dev_proc_init())
7817 if (netdev_kobject_init())
7820 INIT_LIST_HEAD(&ptype_all
);
7821 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
7822 INIT_LIST_HEAD(&ptype_base
[i
]);
7824 INIT_LIST_HEAD(&offload_base
);
7826 if (register_pernet_subsys(&netdev_net_ops
))
7830 * Initialise the packet receive queues.
7833 for_each_possible_cpu(i
) {
7834 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
7836 skb_queue_head_init(&sd
->input_pkt_queue
);
7837 skb_queue_head_init(&sd
->process_queue
);
7838 INIT_LIST_HEAD(&sd
->poll_list
);
7839 sd
->output_queue_tailp
= &sd
->output_queue
;
7841 sd
->csd
.func
= rps_trigger_softirq
;
7846 sd
->backlog
.poll
= process_backlog
;
7847 sd
->backlog
.weight
= weight_p
;
7852 /* The loopback device is special if any other network devices
7853 * is present in a network namespace the loopback device must
7854 * be present. Since we now dynamically allocate and free the
7855 * loopback device ensure this invariant is maintained by
7856 * keeping the loopback device as the first device on the
7857 * list of network devices. Ensuring the loopback devices
7858 * is the first device that appears and the last network device
7861 if (register_pernet_device(&loopback_net_ops
))
7864 if (register_pernet_device(&default_device_ops
))
7867 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
7868 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
7870 hotcpu_notifier(dev_cpu_callback
, 0);
7877 subsys_initcall(net_dev_init
);