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>
141 #include <linux/sctp.h>
143 #include "net-sysfs.h"
145 /* Instead of increasing this, you should create a hash table. */
146 #define MAX_GRO_SKBS 8
148 /* This should be increased if a protocol with a bigger head is added. */
149 #define GRO_MAX_HEAD (MAX_HEADER + 128)
151 static DEFINE_SPINLOCK(ptype_lock
);
152 static DEFINE_SPINLOCK(offload_lock
);
153 struct list_head ptype_base
[PTYPE_HASH_SIZE
] __read_mostly
;
154 struct list_head ptype_all __read_mostly
; /* Taps */
155 static struct list_head offload_base __read_mostly
;
157 static int netif_rx_internal(struct sk_buff
*skb
);
158 static int call_netdevice_notifiers_info(unsigned long val
,
159 struct net_device
*dev
,
160 struct netdev_notifier_info
*info
);
163 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
166 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
168 * Writers must hold the rtnl semaphore while they loop through the
169 * dev_base_head list, and hold dev_base_lock for writing when they do the
170 * actual updates. This allows pure readers to access the list even
171 * while a writer is preparing to update it.
173 * To put it another way, dev_base_lock is held for writing only to
174 * protect against pure readers; the rtnl semaphore provides the
175 * protection against other writers.
177 * See, for example usages, register_netdevice() and
178 * unregister_netdevice(), which must be called with the rtnl
181 DEFINE_RWLOCK(dev_base_lock
);
182 EXPORT_SYMBOL(dev_base_lock
);
184 /* protects napi_hash addition/deletion and napi_gen_id */
185 static DEFINE_SPINLOCK(napi_hash_lock
);
187 static unsigned int napi_gen_id
= NR_CPUS
;
188 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash
, 8);
190 static seqcount_t devnet_rename_seq
;
192 static inline void dev_base_seq_inc(struct net
*net
)
194 while (++net
->dev_base_seq
== 0);
197 static inline struct hlist_head
*dev_name_hash(struct net
*net
, const char *name
)
199 unsigned int hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
201 return &net
->dev_name_head
[hash_32(hash
, NETDEV_HASHBITS
)];
204 static inline struct hlist_head
*dev_index_hash(struct net
*net
, int ifindex
)
206 return &net
->dev_index_head
[ifindex
& (NETDEV_HASHENTRIES
- 1)];
209 static inline void rps_lock(struct softnet_data
*sd
)
212 spin_lock(&sd
->input_pkt_queue
.lock
);
216 static inline void rps_unlock(struct softnet_data
*sd
)
219 spin_unlock(&sd
->input_pkt_queue
.lock
);
223 /* Device list insertion */
224 static void list_netdevice(struct net_device
*dev
)
226 struct net
*net
= dev_net(dev
);
230 write_lock_bh(&dev_base_lock
);
231 list_add_tail_rcu(&dev
->dev_list
, &net
->dev_base_head
);
232 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
233 hlist_add_head_rcu(&dev
->index_hlist
,
234 dev_index_hash(net
, dev
->ifindex
));
235 write_unlock_bh(&dev_base_lock
);
237 dev_base_seq_inc(net
);
240 /* Device list removal
241 * caller must respect a RCU grace period before freeing/reusing dev
243 static void unlist_netdevice(struct net_device
*dev
)
247 /* Unlink dev from the device chain */
248 write_lock_bh(&dev_base_lock
);
249 list_del_rcu(&dev
->dev_list
);
250 hlist_del_rcu(&dev
->name_hlist
);
251 hlist_del_rcu(&dev
->index_hlist
);
252 write_unlock_bh(&dev_base_lock
);
254 dev_base_seq_inc(dev_net(dev
));
261 static RAW_NOTIFIER_HEAD(netdev_chain
);
264 * Device drivers call our routines to queue packets here. We empty the
265 * queue in the local softnet handler.
268 DEFINE_PER_CPU_ALIGNED(struct softnet_data
, softnet_data
);
269 EXPORT_PER_CPU_SYMBOL(softnet_data
);
271 #ifdef CONFIG_LOCKDEP
273 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
274 * according to dev->type
276 static const unsigned short netdev_lock_type
[] =
277 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
278 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
279 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
280 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
281 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
282 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
283 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
284 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
285 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
286 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
287 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
288 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
289 ARPHRD_FCFABRIC
, ARPHRD_IEEE80211
, ARPHRD_IEEE80211_PRISM
,
290 ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_PHONET
, ARPHRD_PHONET_PIPE
,
291 ARPHRD_IEEE802154
, ARPHRD_VOID
, ARPHRD_NONE
};
293 static const char *const netdev_lock_name
[] =
294 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
295 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
296 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
297 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
298 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
299 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
300 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
301 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
302 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
303 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
304 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
305 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
306 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
307 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
308 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
310 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
311 static struct lock_class_key netdev_addr_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
313 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
317 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
318 if (netdev_lock_type
[i
] == dev_type
)
320 /* the last key is used by default */
321 return ARRAY_SIZE(netdev_lock_type
) - 1;
324 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
325 unsigned short dev_type
)
329 i
= netdev_lock_pos(dev_type
);
330 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
331 netdev_lock_name
[i
]);
334 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
338 i
= netdev_lock_pos(dev
->type
);
339 lockdep_set_class_and_name(&dev
->addr_list_lock
,
340 &netdev_addr_lock_key
[i
],
341 netdev_lock_name
[i
]);
344 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
345 unsigned short dev_type
)
348 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
353 /*******************************************************************************
355 Protocol management and registration routines
357 *******************************************************************************/
360 * Add a protocol ID to the list. Now that the input handler is
361 * smarter we can dispense with all the messy stuff that used to be
364 * BEWARE!!! Protocol handlers, mangling input packets,
365 * MUST BE last in hash buckets and checking protocol handlers
366 * MUST start from promiscuous ptype_all chain in net_bh.
367 * It is true now, do not change it.
368 * Explanation follows: if protocol handler, mangling packet, will
369 * be the first on list, it is not able to sense, that packet
370 * is cloned and should be copied-on-write, so that it will
371 * change it and subsequent readers will get broken packet.
375 static inline struct list_head
*ptype_head(const struct packet_type
*pt
)
377 if (pt
->type
== htons(ETH_P_ALL
))
378 return pt
->dev
? &pt
->dev
->ptype_all
: &ptype_all
;
380 return pt
->dev
? &pt
->dev
->ptype_specific
:
381 &ptype_base
[ntohs(pt
->type
) & PTYPE_HASH_MASK
];
385 * dev_add_pack - add packet handler
386 * @pt: packet type declaration
388 * Add a protocol handler to the networking stack. The passed &packet_type
389 * is linked into kernel lists and may not be freed until it has been
390 * removed from the kernel lists.
392 * This call does not sleep therefore it can not
393 * guarantee all CPU's that are in middle of receiving packets
394 * will see the new packet type (until the next received packet).
397 void dev_add_pack(struct packet_type
*pt
)
399 struct list_head
*head
= ptype_head(pt
);
401 spin_lock(&ptype_lock
);
402 list_add_rcu(&pt
->list
, head
);
403 spin_unlock(&ptype_lock
);
405 EXPORT_SYMBOL(dev_add_pack
);
408 * __dev_remove_pack - remove packet handler
409 * @pt: packet type declaration
411 * Remove a protocol handler that was previously added to the kernel
412 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
413 * from the kernel lists and can be freed or reused once this function
416 * The packet type might still be in use by receivers
417 * and must not be freed until after all the CPU's have gone
418 * through a quiescent state.
420 void __dev_remove_pack(struct packet_type
*pt
)
422 struct list_head
*head
= ptype_head(pt
);
423 struct packet_type
*pt1
;
425 spin_lock(&ptype_lock
);
427 list_for_each_entry(pt1
, head
, list
) {
429 list_del_rcu(&pt
->list
);
434 pr_warn("dev_remove_pack: %p not found\n", pt
);
436 spin_unlock(&ptype_lock
);
438 EXPORT_SYMBOL(__dev_remove_pack
);
441 * dev_remove_pack - remove packet handler
442 * @pt: packet type declaration
444 * Remove a protocol handler that was previously added to the kernel
445 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
446 * from the kernel lists and can be freed or reused once this function
449 * This call sleeps to guarantee that no CPU is looking at the packet
452 void dev_remove_pack(struct packet_type
*pt
)
454 __dev_remove_pack(pt
);
458 EXPORT_SYMBOL(dev_remove_pack
);
462 * dev_add_offload - register offload handlers
463 * @po: protocol offload declaration
465 * Add protocol offload handlers to the networking stack. The passed
466 * &proto_offload is linked into kernel lists and may not be freed until
467 * it has been removed from the kernel lists.
469 * This call does not sleep therefore it can not
470 * guarantee all CPU's that are in middle of receiving packets
471 * will see the new offload handlers (until the next received packet).
473 void dev_add_offload(struct packet_offload
*po
)
475 struct packet_offload
*elem
;
477 spin_lock(&offload_lock
);
478 list_for_each_entry(elem
, &offload_base
, list
) {
479 if (po
->priority
< elem
->priority
)
482 list_add_rcu(&po
->list
, elem
->list
.prev
);
483 spin_unlock(&offload_lock
);
485 EXPORT_SYMBOL(dev_add_offload
);
488 * __dev_remove_offload - remove offload handler
489 * @po: packet offload declaration
491 * Remove a protocol offload handler that was previously added to the
492 * kernel offload handlers by dev_add_offload(). The passed &offload_type
493 * is removed from the kernel lists and can be freed or reused once this
496 * The packet type might still be in use by receivers
497 * and must not be freed until after all the CPU's have gone
498 * through a quiescent state.
500 static void __dev_remove_offload(struct packet_offload
*po
)
502 struct list_head
*head
= &offload_base
;
503 struct packet_offload
*po1
;
505 spin_lock(&offload_lock
);
507 list_for_each_entry(po1
, head
, list
) {
509 list_del_rcu(&po
->list
);
514 pr_warn("dev_remove_offload: %p not found\n", po
);
516 spin_unlock(&offload_lock
);
520 * dev_remove_offload - remove packet offload handler
521 * @po: packet offload declaration
523 * Remove a packet offload handler that was previously added to the kernel
524 * offload handlers by dev_add_offload(). The passed &offload_type is
525 * removed from the kernel lists and can be freed or reused once this
528 * This call sleeps to guarantee that no CPU is looking at the packet
531 void dev_remove_offload(struct packet_offload
*po
)
533 __dev_remove_offload(po
);
537 EXPORT_SYMBOL(dev_remove_offload
);
539 /******************************************************************************
541 Device Boot-time Settings Routines
543 *******************************************************************************/
545 /* Boot time configuration table */
546 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
549 * netdev_boot_setup_add - add new setup entry
550 * @name: name of the device
551 * @map: configured settings for the device
553 * Adds new setup entry to the dev_boot_setup list. The function
554 * returns 0 on error and 1 on success. This is a generic routine to
557 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
559 struct netdev_boot_setup
*s
;
563 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
564 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
565 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
566 strlcpy(s
[i
].name
, name
, IFNAMSIZ
);
567 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
572 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
576 * netdev_boot_setup_check - check boot time settings
577 * @dev: the netdevice
579 * Check boot time settings for the device.
580 * The found settings are set for the device to be used
581 * later in the device probing.
582 * Returns 0 if no settings found, 1 if they are.
584 int netdev_boot_setup_check(struct net_device
*dev
)
586 struct netdev_boot_setup
*s
= dev_boot_setup
;
589 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
590 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
591 !strcmp(dev
->name
, s
[i
].name
)) {
592 dev
->irq
= s
[i
].map
.irq
;
593 dev
->base_addr
= s
[i
].map
.base_addr
;
594 dev
->mem_start
= s
[i
].map
.mem_start
;
595 dev
->mem_end
= s
[i
].map
.mem_end
;
601 EXPORT_SYMBOL(netdev_boot_setup_check
);
605 * netdev_boot_base - get address from boot time settings
606 * @prefix: prefix for network device
607 * @unit: id for network device
609 * Check boot time settings for the base address of device.
610 * The found settings are set for the device to be used
611 * later in the device probing.
612 * Returns 0 if no settings found.
614 unsigned long netdev_boot_base(const char *prefix
, int unit
)
616 const struct netdev_boot_setup
*s
= dev_boot_setup
;
620 sprintf(name
, "%s%d", prefix
, unit
);
623 * If device already registered then return base of 1
624 * to indicate not to probe for this interface
626 if (__dev_get_by_name(&init_net
, name
))
629 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
630 if (!strcmp(name
, s
[i
].name
))
631 return s
[i
].map
.base_addr
;
636 * Saves at boot time configured settings for any netdevice.
638 int __init
netdev_boot_setup(char *str
)
643 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
648 memset(&map
, 0, sizeof(map
));
652 map
.base_addr
= ints
[2];
654 map
.mem_start
= ints
[3];
656 map
.mem_end
= ints
[4];
658 /* Add new entry to the list */
659 return netdev_boot_setup_add(str
, &map
);
662 __setup("netdev=", netdev_boot_setup
);
664 /*******************************************************************************
666 Device Interface Subroutines
668 *******************************************************************************/
671 * dev_get_iflink - get 'iflink' value of a interface
672 * @dev: targeted interface
674 * Indicates the ifindex the interface is linked to.
675 * Physical interfaces have the same 'ifindex' and 'iflink' values.
678 int dev_get_iflink(const struct net_device
*dev
)
680 if (dev
->netdev_ops
&& dev
->netdev_ops
->ndo_get_iflink
)
681 return dev
->netdev_ops
->ndo_get_iflink(dev
);
685 EXPORT_SYMBOL(dev_get_iflink
);
688 * dev_fill_metadata_dst - Retrieve tunnel egress information.
689 * @dev: targeted interface
692 * For better visibility of tunnel traffic OVS needs to retrieve
693 * egress tunnel information for a packet. Following API allows
694 * user to get this info.
696 int dev_fill_metadata_dst(struct net_device
*dev
, struct sk_buff
*skb
)
698 struct ip_tunnel_info
*info
;
700 if (!dev
->netdev_ops
|| !dev
->netdev_ops
->ndo_fill_metadata_dst
)
703 info
= skb_tunnel_info_unclone(skb
);
706 if (unlikely(!(info
->mode
& IP_TUNNEL_INFO_TX
)))
709 return dev
->netdev_ops
->ndo_fill_metadata_dst(dev
, skb
);
711 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst
);
714 * __dev_get_by_name - find a device by its name
715 * @net: the applicable net namespace
716 * @name: name to find
718 * Find an interface by name. Must be called under RTNL semaphore
719 * or @dev_base_lock. If the name is found a pointer to the device
720 * is returned. If the name is not found then %NULL is returned. The
721 * reference counters are not incremented so the caller must be
722 * careful with locks.
725 struct net_device
*__dev_get_by_name(struct net
*net
, const char *name
)
727 struct net_device
*dev
;
728 struct hlist_head
*head
= dev_name_hash(net
, name
);
730 hlist_for_each_entry(dev
, head
, name_hlist
)
731 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
736 EXPORT_SYMBOL(__dev_get_by_name
);
739 * dev_get_by_name_rcu - find a device by its name
740 * @net: the applicable net namespace
741 * @name: name to find
743 * Find an interface by name.
744 * If the name is found a pointer to the device is returned.
745 * If the name is not found then %NULL is returned.
746 * The reference counters are not incremented so the caller must be
747 * careful with locks. The caller must hold RCU lock.
750 struct net_device
*dev_get_by_name_rcu(struct net
*net
, const char *name
)
752 struct net_device
*dev
;
753 struct hlist_head
*head
= dev_name_hash(net
, name
);
755 hlist_for_each_entry_rcu(dev
, head
, name_hlist
)
756 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
761 EXPORT_SYMBOL(dev_get_by_name_rcu
);
764 * dev_get_by_name - find a device by its name
765 * @net: the applicable net namespace
766 * @name: name to find
768 * Find an interface by name. This can be called from any
769 * context and does its own locking. The returned handle has
770 * the usage count incremented and the caller must use dev_put() to
771 * release it when it is no longer needed. %NULL is returned if no
772 * matching device is found.
775 struct net_device
*dev_get_by_name(struct net
*net
, const char *name
)
777 struct net_device
*dev
;
780 dev
= dev_get_by_name_rcu(net
, name
);
786 EXPORT_SYMBOL(dev_get_by_name
);
789 * __dev_get_by_index - find a device by its ifindex
790 * @net: the applicable net namespace
791 * @ifindex: index of device
793 * Search for an interface by index. Returns %NULL if the device
794 * is not found or a pointer to the device. The device has not
795 * had its reference counter increased so the caller must be careful
796 * about locking. The caller must hold either the RTNL semaphore
800 struct net_device
*__dev_get_by_index(struct net
*net
, int ifindex
)
802 struct net_device
*dev
;
803 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
805 hlist_for_each_entry(dev
, head
, index_hlist
)
806 if (dev
->ifindex
== ifindex
)
811 EXPORT_SYMBOL(__dev_get_by_index
);
814 * dev_get_by_index_rcu - find a device by its ifindex
815 * @net: the applicable net namespace
816 * @ifindex: index of device
818 * Search for an interface by index. Returns %NULL if the device
819 * is not found or a pointer to the device. The device has not
820 * had its reference counter increased so the caller must be careful
821 * about locking. The caller must hold RCU lock.
824 struct net_device
*dev_get_by_index_rcu(struct net
*net
, int ifindex
)
826 struct net_device
*dev
;
827 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
829 hlist_for_each_entry_rcu(dev
, head
, index_hlist
)
830 if (dev
->ifindex
== ifindex
)
835 EXPORT_SYMBOL(dev_get_by_index_rcu
);
839 * dev_get_by_index - find a device by its ifindex
840 * @net: the applicable net namespace
841 * @ifindex: index of device
843 * Search for an interface by index. Returns NULL if the device
844 * is not found or a pointer to the device. The device returned has
845 * had a reference added and the pointer is safe until the user calls
846 * dev_put to indicate they have finished with it.
849 struct net_device
*dev_get_by_index(struct net
*net
, int ifindex
)
851 struct net_device
*dev
;
854 dev
= dev_get_by_index_rcu(net
, ifindex
);
860 EXPORT_SYMBOL(dev_get_by_index
);
863 * netdev_get_name - get a netdevice name, knowing its ifindex.
864 * @net: network namespace
865 * @name: a pointer to the buffer where the name will be stored.
866 * @ifindex: the ifindex of the interface to get the name from.
868 * The use of raw_seqcount_begin() and cond_resched() before
869 * retrying is required as we want to give the writers a chance
870 * to complete when CONFIG_PREEMPT is not set.
872 int netdev_get_name(struct net
*net
, char *name
, int ifindex
)
874 struct net_device
*dev
;
878 seq
= raw_seqcount_begin(&devnet_rename_seq
);
880 dev
= dev_get_by_index_rcu(net
, ifindex
);
886 strcpy(name
, dev
->name
);
888 if (read_seqcount_retry(&devnet_rename_seq
, seq
)) {
897 * dev_getbyhwaddr_rcu - find a device by its hardware address
898 * @net: the applicable net namespace
899 * @type: media type of device
900 * @ha: hardware address
902 * Search for an interface by MAC address. Returns NULL if the device
903 * is not found or a pointer to the device.
904 * The caller must hold RCU or RTNL.
905 * The returned device has not had its ref count increased
906 * and the caller must therefore be careful about locking
910 struct net_device
*dev_getbyhwaddr_rcu(struct net
*net
, unsigned short type
,
913 struct net_device
*dev
;
915 for_each_netdev_rcu(net
, dev
)
916 if (dev
->type
== type
&&
917 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
922 EXPORT_SYMBOL(dev_getbyhwaddr_rcu
);
924 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
926 struct net_device
*dev
;
929 for_each_netdev(net
, dev
)
930 if (dev
->type
== type
)
935 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
937 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
939 struct net_device
*dev
, *ret
= NULL
;
942 for_each_netdev_rcu(net
, dev
)
943 if (dev
->type
== type
) {
951 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
954 * __dev_get_by_flags - find any device with given flags
955 * @net: the applicable net namespace
956 * @if_flags: IFF_* values
957 * @mask: bitmask of bits in if_flags to check
959 * Search for any interface with the given flags. Returns NULL if a device
960 * is not found or a pointer to the device. Must be called inside
961 * rtnl_lock(), and result refcount is unchanged.
964 struct net_device
*__dev_get_by_flags(struct net
*net
, unsigned short if_flags
,
967 struct net_device
*dev
, *ret
;
972 for_each_netdev(net
, dev
) {
973 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
980 EXPORT_SYMBOL(__dev_get_by_flags
);
983 * dev_valid_name - check if name is okay for network device
986 * Network device names need to be valid file names to
987 * to allow sysfs to work. We also disallow any kind of
990 bool dev_valid_name(const char *name
)
994 if (strlen(name
) >= IFNAMSIZ
)
996 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
1000 if (*name
== '/' || *name
== ':' || isspace(*name
))
1006 EXPORT_SYMBOL(dev_valid_name
);
1009 * __dev_alloc_name - allocate a name for a device
1010 * @net: network namespace to allocate the device name in
1011 * @name: name format string
1012 * @buf: scratch buffer and result name string
1014 * Passed a format string - eg "lt%d" it will try and find a suitable
1015 * id. It scans list of devices to build up a free map, then chooses
1016 * the first empty slot. The caller must hold the dev_base or rtnl lock
1017 * while allocating the name and adding the device in order to avoid
1019 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1020 * Returns the number of the unit assigned or a negative errno code.
1023 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
1027 const int max_netdevices
= 8*PAGE_SIZE
;
1028 unsigned long *inuse
;
1029 struct net_device
*d
;
1031 p
= strnchr(name
, IFNAMSIZ
-1, '%');
1034 * Verify the string as this thing may have come from
1035 * the user. There must be either one "%d" and no other "%"
1038 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
1041 /* Use one page as a bit array of possible slots */
1042 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
1046 for_each_netdev(net
, d
) {
1047 if (!sscanf(d
->name
, name
, &i
))
1049 if (i
< 0 || i
>= max_netdevices
)
1052 /* avoid cases where sscanf is not exact inverse of printf */
1053 snprintf(buf
, IFNAMSIZ
, name
, i
);
1054 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
1058 i
= find_first_zero_bit(inuse
, max_netdevices
);
1059 free_page((unsigned long) inuse
);
1063 snprintf(buf
, IFNAMSIZ
, name
, i
);
1064 if (!__dev_get_by_name(net
, buf
))
1067 /* It is possible to run out of possible slots
1068 * when the name is long and there isn't enough space left
1069 * for the digits, or if all bits are used.
1075 * dev_alloc_name - allocate a name for a device
1077 * @name: name format string
1079 * Passed a format string - eg "lt%d" it will try and find a suitable
1080 * id. It scans list of devices to build up a free map, then chooses
1081 * the first empty slot. The caller must hold the dev_base or rtnl lock
1082 * while allocating the name and adding the device in order to avoid
1084 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1085 * Returns the number of the unit assigned or a negative errno code.
1088 int dev_alloc_name(struct net_device
*dev
, const char *name
)
1094 BUG_ON(!dev_net(dev
));
1096 ret
= __dev_alloc_name(net
, name
, buf
);
1098 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
1101 EXPORT_SYMBOL(dev_alloc_name
);
1103 static int dev_alloc_name_ns(struct net
*net
,
1104 struct net_device
*dev
,
1110 ret
= __dev_alloc_name(net
, name
, buf
);
1112 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
1116 static int dev_get_valid_name(struct net
*net
,
1117 struct net_device
*dev
,
1122 if (!dev_valid_name(name
))
1125 if (strchr(name
, '%'))
1126 return dev_alloc_name_ns(net
, dev
, name
);
1127 else if (__dev_get_by_name(net
, name
))
1129 else if (dev
->name
!= name
)
1130 strlcpy(dev
->name
, name
, IFNAMSIZ
);
1136 * dev_change_name - change name of a device
1138 * @newname: name (or format string) must be at least IFNAMSIZ
1140 * Change name of a device, can pass format strings "eth%d".
1143 int dev_change_name(struct net_device
*dev
, const char *newname
)
1145 unsigned char old_assign_type
;
1146 char oldname
[IFNAMSIZ
];
1152 BUG_ON(!dev_net(dev
));
1155 if (dev
->flags
& IFF_UP
)
1158 write_seqcount_begin(&devnet_rename_seq
);
1160 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0) {
1161 write_seqcount_end(&devnet_rename_seq
);
1165 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
1167 err
= dev_get_valid_name(net
, dev
, newname
);
1169 write_seqcount_end(&devnet_rename_seq
);
1173 if (oldname
[0] && !strchr(oldname
, '%'))
1174 netdev_info(dev
, "renamed from %s\n", oldname
);
1176 old_assign_type
= dev
->name_assign_type
;
1177 dev
->name_assign_type
= NET_NAME_RENAMED
;
1180 ret
= device_rename(&dev
->dev
, dev
->name
);
1182 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1183 dev
->name_assign_type
= old_assign_type
;
1184 write_seqcount_end(&devnet_rename_seq
);
1188 write_seqcount_end(&devnet_rename_seq
);
1190 netdev_adjacent_rename_links(dev
, oldname
);
1192 write_lock_bh(&dev_base_lock
);
1193 hlist_del_rcu(&dev
->name_hlist
);
1194 write_unlock_bh(&dev_base_lock
);
1198 write_lock_bh(&dev_base_lock
);
1199 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
1200 write_unlock_bh(&dev_base_lock
);
1202 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
1203 ret
= notifier_to_errno(ret
);
1206 /* err >= 0 after dev_alloc_name() or stores the first errno */
1209 write_seqcount_begin(&devnet_rename_seq
);
1210 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1211 memcpy(oldname
, newname
, IFNAMSIZ
);
1212 dev
->name_assign_type
= old_assign_type
;
1213 old_assign_type
= NET_NAME_RENAMED
;
1216 pr_err("%s: name change rollback failed: %d\n",
1225 * dev_set_alias - change ifalias of a device
1227 * @alias: name up to IFALIASZ
1228 * @len: limit of bytes to copy from info
1230 * Set ifalias for a device,
1232 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
1238 if (len
>= IFALIASZ
)
1242 kfree(dev
->ifalias
);
1243 dev
->ifalias
= NULL
;
1247 new_ifalias
= krealloc(dev
->ifalias
, len
+ 1, GFP_KERNEL
);
1250 dev
->ifalias
= new_ifalias
;
1252 strlcpy(dev
->ifalias
, alias
, len
+1);
1258 * netdev_features_change - device changes features
1259 * @dev: device to cause notification
1261 * Called to indicate a device has changed features.
1263 void netdev_features_change(struct net_device
*dev
)
1265 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1267 EXPORT_SYMBOL(netdev_features_change
);
1270 * netdev_state_change - device changes state
1271 * @dev: device to cause notification
1273 * Called to indicate a device has changed state. This function calls
1274 * the notifier chains for netdev_chain and sends a NEWLINK message
1275 * to the routing socket.
1277 void netdev_state_change(struct net_device
*dev
)
1279 if (dev
->flags
& IFF_UP
) {
1280 struct netdev_notifier_change_info change_info
;
1282 change_info
.flags_changed
= 0;
1283 call_netdevice_notifiers_info(NETDEV_CHANGE
, dev
,
1285 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0, GFP_KERNEL
);
1288 EXPORT_SYMBOL(netdev_state_change
);
1291 * netdev_notify_peers - notify network peers about existence of @dev
1292 * @dev: network device
1294 * Generate traffic such that interested network peers are aware of
1295 * @dev, such as by generating a gratuitous ARP. This may be used when
1296 * a device wants to inform the rest of the network about some sort of
1297 * reconfiguration such as a failover event or virtual machine
1300 void netdev_notify_peers(struct net_device
*dev
)
1303 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
, dev
);
1306 EXPORT_SYMBOL(netdev_notify_peers
);
1308 static int __dev_open(struct net_device
*dev
)
1310 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1315 if (!netif_device_present(dev
))
1318 /* Block netpoll from trying to do any rx path servicing.
1319 * If we don't do this there is a chance ndo_poll_controller
1320 * or ndo_poll may be running while we open the device
1322 netpoll_poll_disable(dev
);
1324 ret
= call_netdevice_notifiers(NETDEV_PRE_UP
, dev
);
1325 ret
= notifier_to_errno(ret
);
1329 set_bit(__LINK_STATE_START
, &dev
->state
);
1331 if (ops
->ndo_validate_addr
)
1332 ret
= ops
->ndo_validate_addr(dev
);
1334 if (!ret
&& ops
->ndo_open
)
1335 ret
= ops
->ndo_open(dev
);
1337 netpoll_poll_enable(dev
);
1340 clear_bit(__LINK_STATE_START
, &dev
->state
);
1342 dev
->flags
|= IFF_UP
;
1343 dev_set_rx_mode(dev
);
1345 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
1352 * dev_open - prepare an interface for use.
1353 * @dev: device to open
1355 * Takes a device from down to up state. The device's private open
1356 * function is invoked and then the multicast lists are loaded. Finally
1357 * the device is moved into the up state and a %NETDEV_UP message is
1358 * sent to the netdev notifier chain.
1360 * Calling this function on an active interface is a nop. On a failure
1361 * a negative errno code is returned.
1363 int dev_open(struct net_device
*dev
)
1367 if (dev
->flags
& IFF_UP
)
1370 ret
= __dev_open(dev
);
1374 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
, GFP_KERNEL
);
1375 call_netdevice_notifiers(NETDEV_UP
, dev
);
1379 EXPORT_SYMBOL(dev_open
);
1381 static int __dev_close_many(struct list_head
*head
)
1383 struct net_device
*dev
;
1388 list_for_each_entry(dev
, head
, close_list
) {
1389 /* Temporarily disable netpoll until the interface is down */
1390 netpoll_poll_disable(dev
);
1392 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1394 clear_bit(__LINK_STATE_START
, &dev
->state
);
1396 /* Synchronize to scheduled poll. We cannot touch poll list, it
1397 * can be even on different cpu. So just clear netif_running().
1399 * dev->stop() will invoke napi_disable() on all of it's
1400 * napi_struct instances on this device.
1402 smp_mb__after_atomic(); /* Commit netif_running(). */
1405 dev_deactivate_many(head
);
1407 list_for_each_entry(dev
, head
, close_list
) {
1408 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1411 * Call the device specific close. This cannot fail.
1412 * Only if device is UP
1414 * We allow it to be called even after a DETACH hot-plug
1420 dev
->flags
&= ~IFF_UP
;
1421 netpoll_poll_enable(dev
);
1427 static int __dev_close(struct net_device
*dev
)
1432 list_add(&dev
->close_list
, &single
);
1433 retval
= __dev_close_many(&single
);
1439 int dev_close_many(struct list_head
*head
, bool unlink
)
1441 struct net_device
*dev
, *tmp
;
1443 /* Remove the devices that don't need to be closed */
1444 list_for_each_entry_safe(dev
, tmp
, head
, close_list
)
1445 if (!(dev
->flags
& IFF_UP
))
1446 list_del_init(&dev
->close_list
);
1448 __dev_close_many(head
);
1450 list_for_each_entry_safe(dev
, tmp
, head
, close_list
) {
1451 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
, GFP_KERNEL
);
1452 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1454 list_del_init(&dev
->close_list
);
1459 EXPORT_SYMBOL(dev_close_many
);
1462 * dev_close - shutdown an interface.
1463 * @dev: device to shutdown
1465 * This function moves an active device into down state. A
1466 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1467 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1470 int dev_close(struct net_device
*dev
)
1472 if (dev
->flags
& IFF_UP
) {
1475 list_add(&dev
->close_list
, &single
);
1476 dev_close_many(&single
, true);
1481 EXPORT_SYMBOL(dev_close
);
1485 * dev_disable_lro - disable Large Receive Offload on a device
1488 * Disable Large Receive Offload (LRO) on a net device. Must be
1489 * called under RTNL. This is needed if received packets may be
1490 * forwarded to another interface.
1492 void dev_disable_lro(struct net_device
*dev
)
1494 struct net_device
*lower_dev
;
1495 struct list_head
*iter
;
1497 dev
->wanted_features
&= ~NETIF_F_LRO
;
1498 netdev_update_features(dev
);
1500 if (unlikely(dev
->features
& NETIF_F_LRO
))
1501 netdev_WARN(dev
, "failed to disable LRO!\n");
1503 netdev_for_each_lower_dev(dev
, lower_dev
, iter
)
1504 dev_disable_lro(lower_dev
);
1506 EXPORT_SYMBOL(dev_disable_lro
);
1508 static int call_netdevice_notifier(struct notifier_block
*nb
, unsigned long val
,
1509 struct net_device
*dev
)
1511 struct netdev_notifier_info info
;
1513 netdev_notifier_info_init(&info
, dev
);
1514 return nb
->notifier_call(nb
, val
, &info
);
1517 static int dev_boot_phase
= 1;
1520 * register_netdevice_notifier - register a network notifier block
1523 * Register a notifier to be called when network device events occur.
1524 * The notifier passed is linked into the kernel structures and must
1525 * not be reused until it has been unregistered. A negative errno code
1526 * is returned on a failure.
1528 * When registered all registration and up events are replayed
1529 * to the new notifier to allow device to have a race free
1530 * view of the network device list.
1533 int register_netdevice_notifier(struct notifier_block
*nb
)
1535 struct net_device
*dev
;
1536 struct net_device
*last
;
1541 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1547 for_each_netdev(net
, dev
) {
1548 err
= call_netdevice_notifier(nb
, NETDEV_REGISTER
, dev
);
1549 err
= notifier_to_errno(err
);
1553 if (!(dev
->flags
& IFF_UP
))
1556 call_netdevice_notifier(nb
, NETDEV_UP
, dev
);
1567 for_each_netdev(net
, dev
) {
1571 if (dev
->flags
& IFF_UP
) {
1572 call_netdevice_notifier(nb
, NETDEV_GOING_DOWN
,
1574 call_netdevice_notifier(nb
, NETDEV_DOWN
, dev
);
1576 call_netdevice_notifier(nb
, NETDEV_UNREGISTER
, dev
);
1581 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1584 EXPORT_SYMBOL(register_netdevice_notifier
);
1587 * unregister_netdevice_notifier - unregister a network notifier block
1590 * Unregister a notifier previously registered by
1591 * register_netdevice_notifier(). The notifier is unlinked into the
1592 * kernel structures and may then be reused. A negative errno code
1593 * is returned on a failure.
1595 * After unregistering unregister and down device events are synthesized
1596 * for all devices on the device list to the removed notifier to remove
1597 * the need for special case cleanup code.
1600 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1602 struct net_device
*dev
;
1607 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1612 for_each_netdev(net
, dev
) {
1613 if (dev
->flags
& IFF_UP
) {
1614 call_netdevice_notifier(nb
, NETDEV_GOING_DOWN
,
1616 call_netdevice_notifier(nb
, NETDEV_DOWN
, dev
);
1618 call_netdevice_notifier(nb
, NETDEV_UNREGISTER
, dev
);
1625 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1628 * call_netdevice_notifiers_info - call all network notifier blocks
1629 * @val: value passed unmodified to notifier function
1630 * @dev: net_device pointer passed unmodified to notifier function
1631 * @info: notifier information data
1633 * Call all network notifier blocks. Parameters and return value
1634 * are as for raw_notifier_call_chain().
1637 static int call_netdevice_notifiers_info(unsigned long val
,
1638 struct net_device
*dev
,
1639 struct netdev_notifier_info
*info
)
1642 netdev_notifier_info_init(info
, dev
);
1643 return raw_notifier_call_chain(&netdev_chain
, val
, info
);
1647 * call_netdevice_notifiers - call all network notifier blocks
1648 * @val: value passed unmodified to notifier function
1649 * @dev: net_device pointer passed unmodified to notifier function
1651 * Call all network notifier blocks. Parameters and return value
1652 * are as for raw_notifier_call_chain().
1655 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1657 struct netdev_notifier_info info
;
1659 return call_netdevice_notifiers_info(val
, dev
, &info
);
1661 EXPORT_SYMBOL(call_netdevice_notifiers
);
1663 #ifdef CONFIG_NET_INGRESS
1664 static struct static_key ingress_needed __read_mostly
;
1666 void net_inc_ingress_queue(void)
1668 static_key_slow_inc(&ingress_needed
);
1670 EXPORT_SYMBOL_GPL(net_inc_ingress_queue
);
1672 void net_dec_ingress_queue(void)
1674 static_key_slow_dec(&ingress_needed
);
1676 EXPORT_SYMBOL_GPL(net_dec_ingress_queue
);
1679 #ifdef CONFIG_NET_EGRESS
1680 static struct static_key egress_needed __read_mostly
;
1682 void net_inc_egress_queue(void)
1684 static_key_slow_inc(&egress_needed
);
1686 EXPORT_SYMBOL_GPL(net_inc_egress_queue
);
1688 void net_dec_egress_queue(void)
1690 static_key_slow_dec(&egress_needed
);
1692 EXPORT_SYMBOL_GPL(net_dec_egress_queue
);
1695 static struct static_key netstamp_needed __read_mostly
;
1696 #ifdef HAVE_JUMP_LABEL
1697 /* We are not allowed to call static_key_slow_dec() from irq context
1698 * If net_disable_timestamp() is called from irq context, defer the
1699 * static_key_slow_dec() calls.
1701 static atomic_t netstamp_needed_deferred
;
1704 void net_enable_timestamp(void)
1706 #ifdef HAVE_JUMP_LABEL
1707 int deferred
= atomic_xchg(&netstamp_needed_deferred
, 0);
1711 static_key_slow_dec(&netstamp_needed
);
1715 static_key_slow_inc(&netstamp_needed
);
1717 EXPORT_SYMBOL(net_enable_timestamp
);
1719 void net_disable_timestamp(void)
1721 #ifdef HAVE_JUMP_LABEL
1722 if (in_interrupt()) {
1723 atomic_inc(&netstamp_needed_deferred
);
1727 static_key_slow_dec(&netstamp_needed
);
1729 EXPORT_SYMBOL(net_disable_timestamp
);
1731 static inline void net_timestamp_set(struct sk_buff
*skb
)
1733 skb
->tstamp
.tv64
= 0;
1734 if (static_key_false(&netstamp_needed
))
1735 __net_timestamp(skb
);
1738 #define net_timestamp_check(COND, SKB) \
1739 if (static_key_false(&netstamp_needed)) { \
1740 if ((COND) && !(SKB)->tstamp.tv64) \
1741 __net_timestamp(SKB); \
1744 bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb)
1748 if (!(dev
->flags
& IFF_UP
))
1751 len
= dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
;
1752 if (skb
->len
<= len
)
1755 /* if TSO is enabled, we don't care about the length as the packet
1756 * could be forwarded without being segmented before
1758 if (skb_is_gso(skb
))
1763 EXPORT_SYMBOL_GPL(is_skb_forwardable
);
1765 int __dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1767 if (skb_orphan_frags(skb
, GFP_ATOMIC
) ||
1768 unlikely(!is_skb_forwardable(dev
, skb
))) {
1769 atomic_long_inc(&dev
->rx_dropped
);
1774 skb_scrub_packet(skb
, true);
1776 skb
->protocol
= eth_type_trans(skb
, dev
);
1777 skb_postpull_rcsum(skb
, eth_hdr(skb
), ETH_HLEN
);
1781 EXPORT_SYMBOL_GPL(__dev_forward_skb
);
1784 * dev_forward_skb - loopback an skb to another netif
1786 * @dev: destination network device
1787 * @skb: buffer to forward
1790 * NET_RX_SUCCESS (no congestion)
1791 * NET_RX_DROP (packet was dropped, but freed)
1793 * dev_forward_skb can be used for injecting an skb from the
1794 * start_xmit function of one device into the receive queue
1795 * of another device.
1797 * The receiving device may be in another namespace, so
1798 * we have to clear all information in the skb that could
1799 * impact namespace isolation.
1801 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1803 return __dev_forward_skb(dev
, skb
) ?: netif_rx_internal(skb
);
1805 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1807 static inline int deliver_skb(struct sk_buff
*skb
,
1808 struct packet_type
*pt_prev
,
1809 struct net_device
*orig_dev
)
1811 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
1813 atomic_inc(&skb
->users
);
1814 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1817 static inline void deliver_ptype_list_skb(struct sk_buff
*skb
,
1818 struct packet_type
**pt
,
1819 struct net_device
*orig_dev
,
1821 struct list_head
*ptype_list
)
1823 struct packet_type
*ptype
, *pt_prev
= *pt
;
1825 list_for_each_entry_rcu(ptype
, ptype_list
, list
) {
1826 if (ptype
->type
!= type
)
1829 deliver_skb(skb
, pt_prev
, orig_dev
);
1835 static inline bool skb_loop_sk(struct packet_type
*ptype
, struct sk_buff
*skb
)
1837 if (!ptype
->af_packet_priv
|| !skb
->sk
)
1840 if (ptype
->id_match
)
1841 return ptype
->id_match(ptype
, skb
->sk
);
1842 else if ((struct sock
*)ptype
->af_packet_priv
== skb
->sk
)
1849 * Support routine. Sends outgoing frames to any network
1850 * taps currently in use.
1853 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1855 struct packet_type
*ptype
;
1856 struct sk_buff
*skb2
= NULL
;
1857 struct packet_type
*pt_prev
= NULL
;
1858 struct list_head
*ptype_list
= &ptype_all
;
1862 list_for_each_entry_rcu(ptype
, ptype_list
, list
) {
1863 /* Never send packets back to the socket
1864 * they originated from - MvS (miquels@drinkel.ow.org)
1866 if (skb_loop_sk(ptype
, skb
))
1870 deliver_skb(skb2
, pt_prev
, skb
->dev
);
1875 /* need to clone skb, done only once */
1876 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1880 net_timestamp_set(skb2
);
1882 /* skb->nh should be correctly
1883 * set by sender, so that the second statement is
1884 * just protection against buggy protocols.
1886 skb_reset_mac_header(skb2
);
1888 if (skb_network_header(skb2
) < skb2
->data
||
1889 skb_network_header(skb2
) > skb_tail_pointer(skb2
)) {
1890 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1891 ntohs(skb2
->protocol
),
1893 skb_reset_network_header(skb2
);
1896 skb2
->transport_header
= skb2
->network_header
;
1897 skb2
->pkt_type
= PACKET_OUTGOING
;
1901 if (ptype_list
== &ptype_all
) {
1902 ptype_list
= &dev
->ptype_all
;
1907 pt_prev
->func(skb2
, skb
->dev
, pt_prev
, skb
->dev
);
1912 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1913 * @dev: Network device
1914 * @txq: number of queues available
1916 * If real_num_tx_queues is changed the tc mappings may no longer be
1917 * valid. To resolve this verify the tc mapping remains valid and if
1918 * not NULL the mapping. With no priorities mapping to this
1919 * offset/count pair it will no longer be used. In the worst case TC0
1920 * is invalid nothing can be done so disable priority mappings. If is
1921 * expected that drivers will fix this mapping if they can before
1922 * calling netif_set_real_num_tx_queues.
1924 static void netif_setup_tc(struct net_device
*dev
, unsigned int txq
)
1927 struct netdev_tc_txq
*tc
= &dev
->tc_to_txq
[0];
1929 /* If TC0 is invalidated disable TC mapping */
1930 if (tc
->offset
+ tc
->count
> txq
) {
1931 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1936 /* Invalidated prio to tc mappings set to TC0 */
1937 for (i
= 1; i
< TC_BITMASK
+ 1; i
++) {
1938 int q
= netdev_get_prio_tc_map(dev
, i
);
1940 tc
= &dev
->tc_to_txq
[q
];
1941 if (tc
->offset
+ tc
->count
> txq
) {
1942 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1944 netdev_set_prio_tc_map(dev
, i
, 0);
1950 static DEFINE_MUTEX(xps_map_mutex
);
1951 #define xmap_dereference(P) \
1952 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1954 static struct xps_map
*remove_xps_queue(struct xps_dev_maps
*dev_maps
,
1957 struct xps_map
*map
= NULL
;
1961 map
= xmap_dereference(dev_maps
->cpu_map
[cpu
]);
1963 for (pos
= 0; map
&& pos
< map
->len
; pos
++) {
1964 if (map
->queues
[pos
] == index
) {
1966 map
->queues
[pos
] = map
->queues
[--map
->len
];
1968 RCU_INIT_POINTER(dev_maps
->cpu_map
[cpu
], NULL
);
1969 kfree_rcu(map
, rcu
);
1979 static void netif_reset_xps_queues_gt(struct net_device
*dev
, u16 index
)
1981 struct xps_dev_maps
*dev_maps
;
1983 bool active
= false;
1985 mutex_lock(&xps_map_mutex
);
1986 dev_maps
= xmap_dereference(dev
->xps_maps
);
1991 for_each_possible_cpu(cpu
) {
1992 for (i
= index
; i
< dev
->num_tx_queues
; i
++) {
1993 if (!remove_xps_queue(dev_maps
, cpu
, i
))
1996 if (i
== dev
->num_tx_queues
)
2001 RCU_INIT_POINTER(dev
->xps_maps
, NULL
);
2002 kfree_rcu(dev_maps
, rcu
);
2005 for (i
= index
; i
< dev
->num_tx_queues
; i
++)
2006 netdev_queue_numa_node_write(netdev_get_tx_queue(dev
, i
),
2010 mutex_unlock(&xps_map_mutex
);
2013 static struct xps_map
*expand_xps_map(struct xps_map
*map
,
2016 struct xps_map
*new_map
;
2017 int alloc_len
= XPS_MIN_MAP_ALLOC
;
2020 for (pos
= 0; map
&& pos
< map
->len
; pos
++) {
2021 if (map
->queues
[pos
] != index
)
2026 /* Need to add queue to this CPU's existing map */
2028 if (pos
< map
->alloc_len
)
2031 alloc_len
= map
->alloc_len
* 2;
2034 /* Need to allocate new map to store queue on this CPU's map */
2035 new_map
= kzalloc_node(XPS_MAP_SIZE(alloc_len
), GFP_KERNEL
,
2040 for (i
= 0; i
< pos
; i
++)
2041 new_map
->queues
[i
] = map
->queues
[i
];
2042 new_map
->alloc_len
= alloc_len
;
2048 int netif_set_xps_queue(struct net_device
*dev
, const struct cpumask
*mask
,
2051 struct xps_dev_maps
*dev_maps
, *new_dev_maps
= NULL
;
2052 struct xps_map
*map
, *new_map
;
2053 int maps_sz
= max_t(unsigned int, XPS_DEV_MAPS_SIZE
, L1_CACHE_BYTES
);
2054 int cpu
, numa_node_id
= -2;
2055 bool active
= false;
2057 mutex_lock(&xps_map_mutex
);
2059 dev_maps
= xmap_dereference(dev
->xps_maps
);
2061 /* allocate memory for queue storage */
2062 for_each_online_cpu(cpu
) {
2063 if (!cpumask_test_cpu(cpu
, mask
))
2067 new_dev_maps
= kzalloc(maps_sz
, GFP_KERNEL
);
2068 if (!new_dev_maps
) {
2069 mutex_unlock(&xps_map_mutex
);
2073 map
= dev_maps
? xmap_dereference(dev_maps
->cpu_map
[cpu
]) :
2076 map
= expand_xps_map(map
, cpu
, index
);
2080 RCU_INIT_POINTER(new_dev_maps
->cpu_map
[cpu
], map
);
2084 goto out_no_new_maps
;
2086 for_each_possible_cpu(cpu
) {
2087 if (cpumask_test_cpu(cpu
, mask
) && cpu_online(cpu
)) {
2088 /* add queue to CPU maps */
2091 map
= xmap_dereference(new_dev_maps
->cpu_map
[cpu
]);
2092 while ((pos
< map
->len
) && (map
->queues
[pos
] != index
))
2095 if (pos
== map
->len
)
2096 map
->queues
[map
->len
++] = index
;
2098 if (numa_node_id
== -2)
2099 numa_node_id
= cpu_to_node(cpu
);
2100 else if (numa_node_id
!= cpu_to_node(cpu
))
2103 } else if (dev_maps
) {
2104 /* fill in the new device map from the old device map */
2105 map
= xmap_dereference(dev_maps
->cpu_map
[cpu
]);
2106 RCU_INIT_POINTER(new_dev_maps
->cpu_map
[cpu
], map
);
2111 rcu_assign_pointer(dev
->xps_maps
, new_dev_maps
);
2113 /* Cleanup old maps */
2115 for_each_possible_cpu(cpu
) {
2116 new_map
= xmap_dereference(new_dev_maps
->cpu_map
[cpu
]);
2117 map
= xmap_dereference(dev_maps
->cpu_map
[cpu
]);
2118 if (map
&& map
!= new_map
)
2119 kfree_rcu(map
, rcu
);
2122 kfree_rcu(dev_maps
, rcu
);
2125 dev_maps
= new_dev_maps
;
2129 /* update Tx queue numa node */
2130 netdev_queue_numa_node_write(netdev_get_tx_queue(dev
, index
),
2131 (numa_node_id
>= 0) ? numa_node_id
:
2137 /* removes queue from unused CPUs */
2138 for_each_possible_cpu(cpu
) {
2139 if (cpumask_test_cpu(cpu
, mask
) && cpu_online(cpu
))
2142 if (remove_xps_queue(dev_maps
, cpu
, index
))
2146 /* free map if not active */
2148 RCU_INIT_POINTER(dev
->xps_maps
, NULL
);
2149 kfree_rcu(dev_maps
, rcu
);
2153 mutex_unlock(&xps_map_mutex
);
2157 /* remove any maps that we added */
2158 for_each_possible_cpu(cpu
) {
2159 new_map
= xmap_dereference(new_dev_maps
->cpu_map
[cpu
]);
2160 map
= dev_maps
? xmap_dereference(dev_maps
->cpu_map
[cpu
]) :
2162 if (new_map
&& new_map
!= map
)
2166 mutex_unlock(&xps_map_mutex
);
2168 kfree(new_dev_maps
);
2171 EXPORT_SYMBOL(netif_set_xps_queue
);
2175 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2176 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2178 int netif_set_real_num_tx_queues(struct net_device
*dev
, unsigned int txq
)
2182 if (txq
< 1 || txq
> dev
->num_tx_queues
)
2185 if (dev
->reg_state
== NETREG_REGISTERED
||
2186 dev
->reg_state
== NETREG_UNREGISTERING
) {
2189 rc
= netdev_queue_update_kobjects(dev
, dev
->real_num_tx_queues
,
2195 netif_setup_tc(dev
, txq
);
2197 if (txq
< dev
->real_num_tx_queues
) {
2198 qdisc_reset_all_tx_gt(dev
, txq
);
2200 netif_reset_xps_queues_gt(dev
, txq
);
2205 dev
->real_num_tx_queues
= txq
;
2208 EXPORT_SYMBOL(netif_set_real_num_tx_queues
);
2212 * netif_set_real_num_rx_queues - set actual number of RX queues used
2213 * @dev: Network device
2214 * @rxq: Actual number of RX queues
2216 * This must be called either with the rtnl_lock held or before
2217 * registration of the net device. Returns 0 on success, or a
2218 * negative error code. If called before registration, it always
2221 int netif_set_real_num_rx_queues(struct net_device
*dev
, unsigned int rxq
)
2225 if (rxq
< 1 || rxq
> dev
->num_rx_queues
)
2228 if (dev
->reg_state
== NETREG_REGISTERED
) {
2231 rc
= net_rx_queue_update_kobjects(dev
, dev
->real_num_rx_queues
,
2237 dev
->real_num_rx_queues
= rxq
;
2240 EXPORT_SYMBOL(netif_set_real_num_rx_queues
);
2244 * netif_get_num_default_rss_queues - default number of RSS queues
2246 * This routine should set an upper limit on the number of RSS queues
2247 * used by default by multiqueue devices.
2249 int netif_get_num_default_rss_queues(void)
2251 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES
, num_online_cpus());
2253 EXPORT_SYMBOL(netif_get_num_default_rss_queues
);
2255 static inline void __netif_reschedule(struct Qdisc
*q
)
2257 struct softnet_data
*sd
;
2258 unsigned long flags
;
2260 local_irq_save(flags
);
2261 sd
= this_cpu_ptr(&softnet_data
);
2262 q
->next_sched
= NULL
;
2263 *sd
->output_queue_tailp
= q
;
2264 sd
->output_queue_tailp
= &q
->next_sched
;
2265 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
2266 local_irq_restore(flags
);
2269 void __netif_schedule(struct Qdisc
*q
)
2271 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
2272 __netif_reschedule(q
);
2274 EXPORT_SYMBOL(__netif_schedule
);
2276 struct dev_kfree_skb_cb
{
2277 enum skb_free_reason reason
;
2280 static struct dev_kfree_skb_cb
*get_kfree_skb_cb(const struct sk_buff
*skb
)
2282 return (struct dev_kfree_skb_cb
*)skb
->cb
;
2285 void netif_schedule_queue(struct netdev_queue
*txq
)
2288 if (!(txq
->state
& QUEUE_STATE_ANY_XOFF
)) {
2289 struct Qdisc
*q
= rcu_dereference(txq
->qdisc
);
2291 __netif_schedule(q
);
2295 EXPORT_SYMBOL(netif_schedule_queue
);
2298 * netif_wake_subqueue - allow sending packets on subqueue
2299 * @dev: network device
2300 * @queue_index: sub queue index
2302 * Resume individual transmit queue of a device with multiple transmit queues.
2304 void netif_wake_subqueue(struct net_device
*dev
, u16 queue_index
)
2306 struct netdev_queue
*txq
= netdev_get_tx_queue(dev
, queue_index
);
2308 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF
, &txq
->state
)) {
2312 q
= rcu_dereference(txq
->qdisc
);
2313 __netif_schedule(q
);
2317 EXPORT_SYMBOL(netif_wake_subqueue
);
2319 void netif_tx_wake_queue(struct netdev_queue
*dev_queue
)
2321 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF
, &dev_queue
->state
)) {
2325 q
= rcu_dereference(dev_queue
->qdisc
);
2326 __netif_schedule(q
);
2330 EXPORT_SYMBOL(netif_tx_wake_queue
);
2332 void __dev_kfree_skb_irq(struct sk_buff
*skb
, enum skb_free_reason reason
)
2334 unsigned long flags
;
2336 if (likely(atomic_read(&skb
->users
) == 1)) {
2338 atomic_set(&skb
->users
, 0);
2339 } else if (likely(!atomic_dec_and_test(&skb
->users
))) {
2342 get_kfree_skb_cb(skb
)->reason
= reason
;
2343 local_irq_save(flags
);
2344 skb
->next
= __this_cpu_read(softnet_data
.completion_queue
);
2345 __this_cpu_write(softnet_data
.completion_queue
, skb
);
2346 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
2347 local_irq_restore(flags
);
2349 EXPORT_SYMBOL(__dev_kfree_skb_irq
);
2351 void __dev_kfree_skb_any(struct sk_buff
*skb
, enum skb_free_reason reason
)
2353 if (in_irq() || irqs_disabled())
2354 __dev_kfree_skb_irq(skb
, reason
);
2358 EXPORT_SYMBOL(__dev_kfree_skb_any
);
2362 * netif_device_detach - mark device as removed
2363 * @dev: network device
2365 * Mark device as removed from system and therefore no longer available.
2367 void netif_device_detach(struct net_device
*dev
)
2369 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
2370 netif_running(dev
)) {
2371 netif_tx_stop_all_queues(dev
);
2374 EXPORT_SYMBOL(netif_device_detach
);
2377 * netif_device_attach - mark device as attached
2378 * @dev: network device
2380 * Mark device as attached from system and restart if needed.
2382 void netif_device_attach(struct net_device
*dev
)
2384 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
2385 netif_running(dev
)) {
2386 netif_tx_wake_all_queues(dev
);
2387 __netdev_watchdog_up(dev
);
2390 EXPORT_SYMBOL(netif_device_attach
);
2393 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2394 * to be used as a distribution range.
2396 u16
__skb_tx_hash(const struct net_device
*dev
, struct sk_buff
*skb
,
2397 unsigned int num_tx_queues
)
2401 u16 qcount
= num_tx_queues
;
2403 if (skb_rx_queue_recorded(skb
)) {
2404 hash
= skb_get_rx_queue(skb
);
2405 while (unlikely(hash
>= num_tx_queues
))
2406 hash
-= num_tx_queues
;
2411 u8 tc
= netdev_get_prio_tc_map(dev
, skb
->priority
);
2412 qoffset
= dev
->tc_to_txq
[tc
].offset
;
2413 qcount
= dev
->tc_to_txq
[tc
].count
;
2416 return (u16
) reciprocal_scale(skb_get_hash(skb
), qcount
) + qoffset
;
2418 EXPORT_SYMBOL(__skb_tx_hash
);
2420 static void skb_warn_bad_offload(const struct sk_buff
*skb
)
2422 static const netdev_features_t null_features
= 0;
2423 struct net_device
*dev
= skb
->dev
;
2424 const char *name
= "";
2426 if (!net_ratelimit())
2430 if (dev
->dev
.parent
)
2431 name
= dev_driver_string(dev
->dev
.parent
);
2433 name
= netdev_name(dev
);
2435 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2436 "gso_type=%d ip_summed=%d\n",
2437 name
, dev
? &dev
->features
: &null_features
,
2438 skb
->sk
? &skb
->sk
->sk_route_caps
: &null_features
,
2439 skb
->len
, skb
->data_len
, skb_shinfo(skb
)->gso_size
,
2440 skb_shinfo(skb
)->gso_type
, skb
->ip_summed
);
2444 * Invalidate hardware checksum when packet is to be mangled, and
2445 * complete checksum manually on outgoing path.
2447 int skb_checksum_help(struct sk_buff
*skb
)
2450 int ret
= 0, offset
;
2452 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
2453 goto out_set_summed
;
2455 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
2456 skb_warn_bad_offload(skb
);
2460 /* Before computing a checksum, we should make sure no frag could
2461 * be modified by an external entity : checksum could be wrong.
2463 if (skb_has_shared_frag(skb
)) {
2464 ret
= __skb_linearize(skb
);
2469 offset
= skb_checksum_start_offset(skb
);
2470 BUG_ON(offset
>= skb_headlen(skb
));
2471 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
2473 offset
+= skb
->csum_offset
;
2474 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
2476 if (skb_cloned(skb
) &&
2477 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
2478 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2483 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
2485 skb
->ip_summed
= CHECKSUM_NONE
;
2489 EXPORT_SYMBOL(skb_checksum_help
);
2491 /* skb_csum_offload_check - Driver helper function to determine if a device
2492 * with limited checksum offload capabilities is able to offload the checksum
2493 * for a given packet.
2496 * skb - sk_buff for the packet in question
2497 * spec - contains the description of what device can offload
2498 * csum_encapped - returns true if the checksum being offloaded is
2499 * encpasulated. That is it is checksum for the transport header
2500 * in the inner headers.
2501 * checksum_help - when set indicates that helper function should
2502 * call skb_checksum_help if offload checks fail
2505 * true: Packet has passed the checksum checks and should be offloadable to
2506 * the device (a driver may still need to check for additional
2507 * restrictions of its device)
2508 * false: Checksum is not offloadable. If checksum_help was set then
2509 * skb_checksum_help was called to resolve checksum for non-GSO
2510 * packets and when IP protocol is not SCTP
2512 bool __skb_csum_offload_chk(struct sk_buff
*skb
,
2513 const struct skb_csum_offl_spec
*spec
,
2514 bool *csum_encapped
,
2518 struct ipv6hdr
*ipv6
;
2523 if (skb
->protocol
== htons(ETH_P_8021Q
) ||
2524 skb
->protocol
== htons(ETH_P_8021AD
)) {
2525 if (!spec
->vlan_okay
)
2529 /* We check whether the checksum refers to a transport layer checksum in
2530 * the outermost header or an encapsulated transport layer checksum that
2531 * corresponds to the inner headers of the skb. If the checksum is for
2532 * something else in the packet we need help.
2534 if (skb_checksum_start_offset(skb
) == skb_transport_offset(skb
)) {
2535 /* Non-encapsulated checksum */
2536 protocol
= eproto_to_ipproto(vlan_get_protocol(skb
));
2537 nhdr
= skb_network_header(skb
);
2538 *csum_encapped
= false;
2539 if (spec
->no_not_encapped
)
2541 } else if (skb
->encapsulation
&& spec
->encap_okay
&&
2542 skb_checksum_start_offset(skb
) ==
2543 skb_inner_transport_offset(skb
)) {
2544 /* Encapsulated checksum */
2545 *csum_encapped
= true;
2546 switch (skb
->inner_protocol_type
) {
2547 case ENCAP_TYPE_ETHER
:
2548 protocol
= eproto_to_ipproto(skb
->inner_protocol
);
2550 case ENCAP_TYPE_IPPROTO
:
2551 protocol
= skb
->inner_protocol
;
2554 nhdr
= skb_inner_network_header(skb
);
2561 if (!spec
->ipv4_okay
)
2564 ip_proto
= iph
->protocol
;
2565 if (iph
->ihl
!= 5 && !spec
->ip_options_okay
)
2569 if (!spec
->ipv6_okay
)
2571 if (spec
->no_encapped_ipv6
&& *csum_encapped
)
2574 nhdr
+= sizeof(*ipv6
);
2575 ip_proto
= ipv6
->nexthdr
;
2584 if (!spec
->tcp_okay
||
2585 skb
->csum_offset
!= offsetof(struct tcphdr
, check
))
2589 if (!spec
->udp_okay
||
2590 skb
->csum_offset
!= offsetof(struct udphdr
, check
))
2594 if (!spec
->sctp_okay
||
2595 skb
->csum_offset
!= offsetof(struct sctphdr
, checksum
))
2599 case NEXTHDR_ROUTING
:
2600 case NEXTHDR_DEST
: {
2603 if (protocol
!= IPPROTO_IPV6
|| !spec
->ext_hdrs_okay
)
2606 ip_proto
= opthdr
[0];
2607 nhdr
+= (opthdr
[1] + 1) << 3;
2609 goto ip_proto_again
;
2615 /* Passed the tests for offloading checksum */
2619 if (csum_help
&& !skb_shinfo(skb
)->gso_size
)
2620 skb_checksum_help(skb
);
2624 EXPORT_SYMBOL(__skb_csum_offload_chk
);
2626 __be16
skb_network_protocol(struct sk_buff
*skb
, int *depth
)
2628 __be16 type
= skb
->protocol
;
2630 /* Tunnel gso handlers can set protocol to ethernet. */
2631 if (type
== htons(ETH_P_TEB
)) {
2634 if (unlikely(!pskb_may_pull(skb
, sizeof(struct ethhdr
))))
2637 eth
= (struct ethhdr
*)skb_mac_header(skb
);
2638 type
= eth
->h_proto
;
2641 return __vlan_get_protocol(skb
, type
, depth
);
2645 * skb_mac_gso_segment - mac layer segmentation handler.
2646 * @skb: buffer to segment
2647 * @features: features for the output path (see dev->features)
2649 struct sk_buff
*skb_mac_gso_segment(struct sk_buff
*skb
,
2650 netdev_features_t features
)
2652 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
2653 struct packet_offload
*ptype
;
2654 int vlan_depth
= skb
->mac_len
;
2655 __be16 type
= skb_network_protocol(skb
, &vlan_depth
);
2657 if (unlikely(!type
))
2658 return ERR_PTR(-EINVAL
);
2660 __skb_pull(skb
, vlan_depth
);
2663 list_for_each_entry_rcu(ptype
, &offload_base
, list
) {
2664 if (ptype
->type
== type
&& ptype
->callbacks
.gso_segment
) {
2665 segs
= ptype
->callbacks
.gso_segment(skb
, features
);
2671 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
2675 EXPORT_SYMBOL(skb_mac_gso_segment
);
2678 /* openvswitch calls this on rx path, so we need a different check.
2680 static inline bool skb_needs_check(struct sk_buff
*skb
, bool tx_path
)
2683 return skb
->ip_summed
!= CHECKSUM_PARTIAL
;
2685 return skb
->ip_summed
== CHECKSUM_NONE
;
2689 * __skb_gso_segment - Perform segmentation on skb.
2690 * @skb: buffer to segment
2691 * @features: features for the output path (see dev->features)
2692 * @tx_path: whether it is called in TX path
2694 * This function segments the given skb and returns a list of segments.
2696 * It may return NULL if the skb requires no segmentation. This is
2697 * only possible when GSO is used for verifying header integrity.
2699 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
2701 struct sk_buff
*__skb_gso_segment(struct sk_buff
*skb
,
2702 netdev_features_t features
, bool tx_path
)
2704 if (unlikely(skb_needs_check(skb
, tx_path
))) {
2707 skb_warn_bad_offload(skb
);
2709 err
= skb_cow_head(skb
, 0);
2711 return ERR_PTR(err
);
2714 BUILD_BUG_ON(SKB_SGO_CB_OFFSET
+
2715 sizeof(*SKB_GSO_CB(skb
)) > sizeof(skb
->cb
));
2717 SKB_GSO_CB(skb
)->mac_offset
= skb_headroom(skb
);
2718 SKB_GSO_CB(skb
)->encap_level
= 0;
2720 skb_reset_mac_header(skb
);
2721 skb_reset_mac_len(skb
);
2723 return skb_mac_gso_segment(skb
, features
);
2725 EXPORT_SYMBOL(__skb_gso_segment
);
2727 /* Take action when hardware reception checksum errors are detected. */
2729 void netdev_rx_csum_fault(struct net_device
*dev
)
2731 if (net_ratelimit()) {
2732 pr_err("%s: hw csum failure\n", dev
? dev
->name
: "<unknown>");
2736 EXPORT_SYMBOL(netdev_rx_csum_fault
);
2739 /* Actually, we should eliminate this check as soon as we know, that:
2740 * 1. IOMMU is present and allows to map all the memory.
2741 * 2. No high memory really exists on this machine.
2744 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
2746 #ifdef CONFIG_HIGHMEM
2748 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
2749 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2750 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2751 if (PageHighMem(skb_frag_page(frag
)))
2756 if (PCI_DMA_BUS_IS_PHYS
) {
2757 struct device
*pdev
= dev
->dev
.parent
;
2761 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2762 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2763 dma_addr_t addr
= page_to_phys(skb_frag_page(frag
));
2764 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
2772 /* If MPLS offload request, verify we are testing hardware MPLS features
2773 * instead of standard features for the netdev.
2775 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
2776 static netdev_features_t
net_mpls_features(struct sk_buff
*skb
,
2777 netdev_features_t features
,
2780 if (eth_p_mpls(type
))
2781 features
&= skb
->dev
->mpls_features
;
2786 static netdev_features_t
net_mpls_features(struct sk_buff
*skb
,
2787 netdev_features_t features
,
2794 static netdev_features_t
harmonize_features(struct sk_buff
*skb
,
2795 netdev_features_t features
)
2800 type
= skb_network_protocol(skb
, &tmp
);
2801 features
= net_mpls_features(skb
, features
, type
);
2803 if (skb
->ip_summed
!= CHECKSUM_NONE
&&
2804 !can_checksum_protocol(features
, type
)) {
2805 features
&= ~NETIF_F_CSUM_MASK
;
2806 } else if (illegal_highdma(skb
->dev
, skb
)) {
2807 features
&= ~NETIF_F_SG
;
2813 netdev_features_t
passthru_features_check(struct sk_buff
*skb
,
2814 struct net_device
*dev
,
2815 netdev_features_t features
)
2819 EXPORT_SYMBOL(passthru_features_check
);
2821 static netdev_features_t
dflt_features_check(const struct sk_buff
*skb
,
2822 struct net_device
*dev
,
2823 netdev_features_t features
)
2825 return vlan_features_check(skb
, features
);
2828 netdev_features_t
netif_skb_features(struct sk_buff
*skb
)
2830 struct net_device
*dev
= skb
->dev
;
2831 netdev_features_t features
= dev
->features
;
2832 u16 gso_segs
= skb_shinfo(skb
)->gso_segs
;
2834 if (gso_segs
> dev
->gso_max_segs
|| gso_segs
< dev
->gso_min_segs
)
2835 features
&= ~NETIF_F_GSO_MASK
;
2837 /* If encapsulation offload request, verify we are testing
2838 * hardware encapsulation features instead of standard
2839 * features for the netdev
2841 if (skb
->encapsulation
)
2842 features
&= dev
->hw_enc_features
;
2844 if (skb_vlan_tagged(skb
))
2845 features
= netdev_intersect_features(features
,
2846 dev
->vlan_features
|
2847 NETIF_F_HW_VLAN_CTAG_TX
|
2848 NETIF_F_HW_VLAN_STAG_TX
);
2850 if (dev
->netdev_ops
->ndo_features_check
)
2851 features
&= dev
->netdev_ops
->ndo_features_check(skb
, dev
,
2854 features
&= dflt_features_check(skb
, dev
, features
);
2856 return harmonize_features(skb
, features
);
2858 EXPORT_SYMBOL(netif_skb_features
);
2860 static int xmit_one(struct sk_buff
*skb
, struct net_device
*dev
,
2861 struct netdev_queue
*txq
, bool more
)
2866 if (!list_empty(&ptype_all
) || !list_empty(&dev
->ptype_all
))
2867 dev_queue_xmit_nit(skb
, dev
);
2870 trace_net_dev_start_xmit(skb
, dev
);
2871 rc
= netdev_start_xmit(skb
, dev
, txq
, more
);
2872 trace_net_dev_xmit(skb
, rc
, dev
, len
);
2877 struct sk_buff
*dev_hard_start_xmit(struct sk_buff
*first
, struct net_device
*dev
,
2878 struct netdev_queue
*txq
, int *ret
)
2880 struct sk_buff
*skb
= first
;
2881 int rc
= NETDEV_TX_OK
;
2884 struct sk_buff
*next
= skb
->next
;
2887 rc
= xmit_one(skb
, dev
, txq
, next
!= NULL
);
2888 if (unlikely(!dev_xmit_complete(rc
))) {
2894 if (netif_xmit_stopped(txq
) && skb
) {
2895 rc
= NETDEV_TX_BUSY
;
2905 static struct sk_buff
*validate_xmit_vlan(struct sk_buff
*skb
,
2906 netdev_features_t features
)
2908 if (skb_vlan_tag_present(skb
) &&
2909 !vlan_hw_offload_capable(features
, skb
->vlan_proto
))
2910 skb
= __vlan_hwaccel_push_inside(skb
);
2914 static struct sk_buff
*validate_xmit_skb(struct sk_buff
*skb
, struct net_device
*dev
)
2916 netdev_features_t features
;
2921 features
= netif_skb_features(skb
);
2922 skb
= validate_xmit_vlan(skb
, features
);
2926 if (netif_needs_gso(skb
, features
)) {
2927 struct sk_buff
*segs
;
2929 segs
= skb_gso_segment(skb
, features
);
2937 if (skb_needs_linearize(skb
, features
) &&
2938 __skb_linearize(skb
))
2941 /* If packet is not checksummed and device does not
2942 * support checksumming for this protocol, complete
2943 * checksumming here.
2945 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2946 if (skb
->encapsulation
)
2947 skb_set_inner_transport_header(skb
,
2948 skb_checksum_start_offset(skb
));
2950 skb_set_transport_header(skb
,
2951 skb_checksum_start_offset(skb
));
2952 if (!(features
& NETIF_F_CSUM_MASK
) &&
2953 skb_checksum_help(skb
))
2966 struct sk_buff
*validate_xmit_skb_list(struct sk_buff
*skb
, struct net_device
*dev
)
2968 struct sk_buff
*next
, *head
= NULL
, *tail
;
2970 for (; skb
!= NULL
; skb
= next
) {
2974 /* in case skb wont be segmented, point to itself */
2977 skb
= validate_xmit_skb(skb
, dev
);
2985 /* If skb was segmented, skb->prev points to
2986 * the last segment. If not, it still contains skb.
2993 static void qdisc_pkt_len_init(struct sk_buff
*skb
)
2995 const struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
2997 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
2999 /* To get more precise estimation of bytes sent on wire,
3000 * we add to pkt_len the headers size of all segments
3002 if (shinfo
->gso_size
) {
3003 unsigned int hdr_len
;
3004 u16 gso_segs
= shinfo
->gso_segs
;
3006 /* mac layer + network layer */
3007 hdr_len
= skb_transport_header(skb
) - skb_mac_header(skb
);
3009 /* + transport layer */
3010 if (likely(shinfo
->gso_type
& (SKB_GSO_TCPV4
| SKB_GSO_TCPV6
)))
3011 hdr_len
+= tcp_hdrlen(skb
);
3013 hdr_len
+= sizeof(struct udphdr
);
3015 if (shinfo
->gso_type
& SKB_GSO_DODGY
)
3016 gso_segs
= DIV_ROUND_UP(skb
->len
- hdr_len
,
3019 qdisc_skb_cb(skb
)->pkt_len
+= (gso_segs
- 1) * hdr_len
;
3023 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
3024 struct net_device
*dev
,
3025 struct netdev_queue
*txq
)
3027 spinlock_t
*root_lock
= qdisc_lock(q
);
3031 qdisc_calculate_pkt_len(skb
, q
);
3033 * Heuristic to force contended enqueues to serialize on a
3034 * separate lock before trying to get qdisc main lock.
3035 * This permits __QDISC___STATE_RUNNING owner to get the lock more
3036 * often and dequeue packets faster.
3038 contended
= qdisc_is_running(q
);
3039 if (unlikely(contended
))
3040 spin_lock(&q
->busylock
);
3042 spin_lock(root_lock
);
3043 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
3046 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
3047 qdisc_run_begin(q
)) {
3049 * This is a work-conserving queue; there are no old skbs
3050 * waiting to be sent out; and the qdisc is not running -
3051 * xmit the skb directly.
3054 qdisc_bstats_update(q
, skb
);
3056 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
, true)) {
3057 if (unlikely(contended
)) {
3058 spin_unlock(&q
->busylock
);
3065 rc
= NET_XMIT_SUCCESS
;
3067 rc
= q
->enqueue(skb
, q
) & NET_XMIT_MASK
;
3068 if (qdisc_run_begin(q
)) {
3069 if (unlikely(contended
)) {
3070 spin_unlock(&q
->busylock
);
3076 spin_unlock(root_lock
);
3077 if (unlikely(contended
))
3078 spin_unlock(&q
->busylock
);
3082 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3083 static void skb_update_prio(struct sk_buff
*skb
)
3085 struct netprio_map
*map
= rcu_dereference_bh(skb
->dev
->priomap
);
3087 if (!skb
->priority
&& skb
->sk
&& map
) {
3088 unsigned int prioidx
=
3089 sock_cgroup_prioidx(&skb
->sk
->sk_cgrp_data
);
3091 if (prioidx
< map
->priomap_len
)
3092 skb
->priority
= map
->priomap
[prioidx
];
3096 #define skb_update_prio(skb)
3099 DEFINE_PER_CPU(int, xmit_recursion
);
3100 EXPORT_SYMBOL(xmit_recursion
);
3102 #define RECURSION_LIMIT 10
3105 * dev_loopback_xmit - loop back @skb
3106 * @net: network namespace this loopback is happening in
3107 * @sk: sk needed to be a netfilter okfn
3108 * @skb: buffer to transmit
3110 int dev_loopback_xmit(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
3112 skb_reset_mac_header(skb
);
3113 __skb_pull(skb
, skb_network_offset(skb
));
3114 skb
->pkt_type
= PACKET_LOOPBACK
;
3115 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
3116 WARN_ON(!skb_dst(skb
));
3121 EXPORT_SYMBOL(dev_loopback_xmit
);
3123 #ifdef CONFIG_NET_EGRESS
3124 static struct sk_buff
*
3125 sch_handle_egress(struct sk_buff
*skb
, int *ret
, struct net_device
*dev
)
3127 struct tcf_proto
*cl
= rcu_dereference_bh(dev
->egress_cl_list
);
3128 struct tcf_result cl_res
;
3133 /* skb->tc_verd and qdisc_skb_cb(skb)->pkt_len were already set
3134 * earlier by the caller.
3136 qdisc_bstats_cpu_update(cl
->q
, skb
);
3138 switch (tc_classify(skb
, cl
, &cl_res
, false)) {
3140 case TC_ACT_RECLASSIFY
:
3141 skb
->tc_index
= TC_H_MIN(cl_res
.classid
);
3144 qdisc_qstats_cpu_drop(cl
->q
);
3145 *ret
= NET_XMIT_DROP
;
3149 *ret
= NET_XMIT_SUCCESS
;
3153 case TC_ACT_REDIRECT
:
3154 /* No need to push/pop skb's mac_header here on egress! */
3155 skb_do_redirect(skb
);
3156 *ret
= NET_XMIT_SUCCESS
;
3164 #endif /* CONFIG_NET_EGRESS */
3166 static inline int get_xps_queue(struct net_device
*dev
, struct sk_buff
*skb
)
3169 struct xps_dev_maps
*dev_maps
;
3170 struct xps_map
*map
;
3171 int queue_index
= -1;
3174 dev_maps
= rcu_dereference(dev
->xps_maps
);
3176 map
= rcu_dereference(
3177 dev_maps
->cpu_map
[skb
->sender_cpu
- 1]);
3180 queue_index
= map
->queues
[0];
3182 queue_index
= map
->queues
[reciprocal_scale(skb_get_hash(skb
),
3184 if (unlikely(queue_index
>= dev
->real_num_tx_queues
))
3196 static u16
__netdev_pick_tx(struct net_device
*dev
, struct sk_buff
*skb
)
3198 struct sock
*sk
= skb
->sk
;
3199 int queue_index
= sk_tx_queue_get(sk
);
3201 if (queue_index
< 0 || skb
->ooo_okay
||
3202 queue_index
>= dev
->real_num_tx_queues
) {
3203 int new_index
= get_xps_queue(dev
, skb
);
3205 new_index
= skb_tx_hash(dev
, skb
);
3207 if (queue_index
!= new_index
&& sk
&&
3209 rcu_access_pointer(sk
->sk_dst_cache
))
3210 sk_tx_queue_set(sk
, new_index
);
3212 queue_index
= new_index
;
3218 struct netdev_queue
*netdev_pick_tx(struct net_device
*dev
,
3219 struct sk_buff
*skb
,
3222 int queue_index
= 0;
3225 u32 sender_cpu
= skb
->sender_cpu
- 1;
3227 if (sender_cpu
>= (u32
)NR_CPUS
)
3228 skb
->sender_cpu
= raw_smp_processor_id() + 1;
3231 if (dev
->real_num_tx_queues
!= 1) {
3232 const struct net_device_ops
*ops
= dev
->netdev_ops
;
3233 if (ops
->ndo_select_queue
)
3234 queue_index
= ops
->ndo_select_queue(dev
, skb
, accel_priv
,
3237 queue_index
= __netdev_pick_tx(dev
, skb
);
3240 queue_index
= netdev_cap_txqueue(dev
, queue_index
);
3243 skb_set_queue_mapping(skb
, queue_index
);
3244 return netdev_get_tx_queue(dev
, queue_index
);
3248 * __dev_queue_xmit - transmit a buffer
3249 * @skb: buffer to transmit
3250 * @accel_priv: private data used for L2 forwarding offload
3252 * Queue a buffer for transmission to a network device. The caller must
3253 * have set the device and priority and built the buffer before calling
3254 * this function. The function can be called from an interrupt.
3256 * A negative errno code is returned on a failure. A success does not
3257 * guarantee the frame will be transmitted as it may be dropped due
3258 * to congestion or traffic shaping.
3260 * -----------------------------------------------------------------------------------
3261 * I notice this method can also return errors from the queue disciplines,
3262 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3265 * Regardless of the return value, the skb is consumed, so it is currently
3266 * difficult to retry a send to this method. (You can bump the ref count
3267 * before sending to hold a reference for retry if you are careful.)
3269 * When calling this method, interrupts MUST be enabled. This is because
3270 * the BH enable code must have IRQs enabled so that it will not deadlock.
3273 static int __dev_queue_xmit(struct sk_buff
*skb
, void *accel_priv
)
3275 struct net_device
*dev
= skb
->dev
;
3276 struct netdev_queue
*txq
;
3280 skb_reset_mac_header(skb
);
3282 if (unlikely(skb_shinfo(skb
)->tx_flags
& SKBTX_SCHED_TSTAMP
))
3283 __skb_tstamp_tx(skb
, NULL
, skb
->sk
, SCM_TSTAMP_SCHED
);
3285 /* Disable soft irqs for various locks below. Also
3286 * stops preemption for RCU.
3290 skb_update_prio(skb
);
3292 qdisc_pkt_len_init(skb
);
3293 #ifdef CONFIG_NET_CLS_ACT
3294 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
3295 # ifdef CONFIG_NET_EGRESS
3296 if (static_key_false(&egress_needed
)) {
3297 skb
= sch_handle_egress(skb
, &rc
, dev
);
3303 /* If device/qdisc don't need skb->dst, release it right now while
3304 * its hot in this cpu cache.
3306 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
3311 #ifdef CONFIG_NET_SWITCHDEV
3312 /* Don't forward if offload device already forwarded */
3313 if (skb
->offload_fwd_mark
&&
3314 skb
->offload_fwd_mark
== dev
->offload_fwd_mark
) {
3316 rc
= NET_XMIT_SUCCESS
;
3321 txq
= netdev_pick_tx(dev
, skb
, accel_priv
);
3322 q
= rcu_dereference_bh(txq
->qdisc
);
3324 trace_net_dev_queue(skb
);
3326 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
3330 /* The device has no queue. Common case for software devices:
3331 loopback, all the sorts of tunnels...
3333 Really, it is unlikely that netif_tx_lock protection is necessary
3334 here. (f.e. loopback and IP tunnels are clean ignoring statistics
3336 However, it is possible, that they rely on protection
3339 Check this and shot the lock. It is not prone from deadlocks.
3340 Either shot noqueue qdisc, it is even simpler 8)
3342 if (dev
->flags
& IFF_UP
) {
3343 int cpu
= smp_processor_id(); /* ok because BHs are off */
3345 if (txq
->xmit_lock_owner
!= cpu
) {
3347 if (__this_cpu_read(xmit_recursion
) > RECURSION_LIMIT
)
3348 goto recursion_alert
;
3350 skb
= validate_xmit_skb(skb
, dev
);
3354 HARD_TX_LOCK(dev
, txq
, cpu
);
3356 if (!netif_xmit_stopped(txq
)) {
3357 __this_cpu_inc(xmit_recursion
);
3358 skb
= dev_hard_start_xmit(skb
, dev
, txq
, &rc
);
3359 __this_cpu_dec(xmit_recursion
);
3360 if (dev_xmit_complete(rc
)) {
3361 HARD_TX_UNLOCK(dev
, txq
);
3365 HARD_TX_UNLOCK(dev
, txq
);
3366 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3369 /* Recursion is detected! It is possible,
3373 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3380 rcu_read_unlock_bh();
3382 atomic_long_inc(&dev
->tx_dropped
);
3383 kfree_skb_list(skb
);
3386 rcu_read_unlock_bh();
3390 int dev_queue_xmit(struct sk_buff
*skb
)
3392 return __dev_queue_xmit(skb
, NULL
);
3394 EXPORT_SYMBOL(dev_queue_xmit
);
3396 int dev_queue_xmit_accel(struct sk_buff
*skb
, void *accel_priv
)
3398 return __dev_queue_xmit(skb
, accel_priv
);
3400 EXPORT_SYMBOL(dev_queue_xmit_accel
);
3403 /*=======================================================================
3405 =======================================================================*/
3407 int netdev_max_backlog __read_mostly
= 1000;
3408 EXPORT_SYMBOL(netdev_max_backlog
);
3410 int netdev_tstamp_prequeue __read_mostly
= 1;
3411 int netdev_budget __read_mostly
= 300;
3412 int weight_p __read_mostly
= 64; /* old backlog weight */
3414 /* Called with irq disabled */
3415 static inline void ____napi_schedule(struct softnet_data
*sd
,
3416 struct napi_struct
*napi
)
3418 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
3419 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3424 /* One global table that all flow-based protocols share. */
3425 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
3426 EXPORT_SYMBOL(rps_sock_flow_table
);
3427 u32 rps_cpu_mask __read_mostly
;
3428 EXPORT_SYMBOL(rps_cpu_mask
);
3430 struct static_key rps_needed __read_mostly
;
3432 static struct rps_dev_flow
*
3433 set_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
3434 struct rps_dev_flow
*rflow
, u16 next_cpu
)
3436 if (next_cpu
< nr_cpu_ids
) {
3437 #ifdef CONFIG_RFS_ACCEL
3438 struct netdev_rx_queue
*rxqueue
;
3439 struct rps_dev_flow_table
*flow_table
;
3440 struct rps_dev_flow
*old_rflow
;
3445 /* Should we steer this flow to a different hardware queue? */
3446 if (!skb_rx_queue_recorded(skb
) || !dev
->rx_cpu_rmap
||
3447 !(dev
->features
& NETIF_F_NTUPLE
))
3449 rxq_index
= cpu_rmap_lookup_index(dev
->rx_cpu_rmap
, next_cpu
);
3450 if (rxq_index
== skb_get_rx_queue(skb
))
3453 rxqueue
= dev
->_rx
+ rxq_index
;
3454 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3457 flow_id
= skb_get_hash(skb
) & flow_table
->mask
;
3458 rc
= dev
->netdev_ops
->ndo_rx_flow_steer(dev
, skb
,
3459 rxq_index
, flow_id
);
3463 rflow
= &flow_table
->flows
[flow_id
];
3465 if (old_rflow
->filter
== rflow
->filter
)
3466 old_rflow
->filter
= RPS_NO_FILTER
;
3470 per_cpu(softnet_data
, next_cpu
).input_queue_head
;
3473 rflow
->cpu
= next_cpu
;
3478 * get_rps_cpu is called from netif_receive_skb and returns the target
3479 * CPU from the RPS map of the receiving queue for a given skb.
3480 * rcu_read_lock must be held on entry.
3482 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
3483 struct rps_dev_flow
**rflowp
)
3485 const struct rps_sock_flow_table
*sock_flow_table
;
3486 struct netdev_rx_queue
*rxqueue
= dev
->_rx
;
3487 struct rps_dev_flow_table
*flow_table
;
3488 struct rps_map
*map
;
3493 if (skb_rx_queue_recorded(skb
)) {
3494 u16 index
= skb_get_rx_queue(skb
);
3496 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
3497 WARN_ONCE(dev
->real_num_rx_queues
> 1,
3498 "%s received packet on queue %u, but number "
3499 "of RX queues is %u\n",
3500 dev
->name
, index
, dev
->real_num_rx_queues
);
3506 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3508 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3509 map
= rcu_dereference(rxqueue
->rps_map
);
3510 if (!flow_table
&& !map
)
3513 skb_reset_network_header(skb
);
3514 hash
= skb_get_hash(skb
);
3518 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
3519 if (flow_table
&& sock_flow_table
) {
3520 struct rps_dev_flow
*rflow
;
3524 /* First check into global flow table if there is a match */
3525 ident
= sock_flow_table
->ents
[hash
& sock_flow_table
->mask
];
3526 if ((ident
^ hash
) & ~rps_cpu_mask
)
3529 next_cpu
= ident
& rps_cpu_mask
;
3531 /* OK, now we know there is a match,
3532 * we can look at the local (per receive queue) flow table
3534 rflow
= &flow_table
->flows
[hash
& flow_table
->mask
];
3538 * If the desired CPU (where last recvmsg was done) is
3539 * different from current CPU (one in the rx-queue flow
3540 * table entry), switch if one of the following holds:
3541 * - Current CPU is unset (>= nr_cpu_ids).
3542 * - Current CPU is offline.
3543 * - The current CPU's queue tail has advanced beyond the
3544 * last packet that was enqueued using this table entry.
3545 * This guarantees that all previous packets for the flow
3546 * have been dequeued, thus preserving in order delivery.
3548 if (unlikely(tcpu
!= next_cpu
) &&
3549 (tcpu
>= nr_cpu_ids
|| !cpu_online(tcpu
) ||
3550 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
3551 rflow
->last_qtail
)) >= 0)) {
3553 rflow
= set_rps_cpu(dev
, skb
, rflow
, next_cpu
);
3556 if (tcpu
< nr_cpu_ids
&& cpu_online(tcpu
)) {
3566 tcpu
= map
->cpus
[reciprocal_scale(hash
, map
->len
)];
3567 if (cpu_online(tcpu
)) {
3577 #ifdef CONFIG_RFS_ACCEL
3580 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3581 * @dev: Device on which the filter was set
3582 * @rxq_index: RX queue index
3583 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3584 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3586 * Drivers that implement ndo_rx_flow_steer() should periodically call
3587 * this function for each installed filter and remove the filters for
3588 * which it returns %true.
3590 bool rps_may_expire_flow(struct net_device
*dev
, u16 rxq_index
,
3591 u32 flow_id
, u16 filter_id
)
3593 struct netdev_rx_queue
*rxqueue
= dev
->_rx
+ rxq_index
;
3594 struct rps_dev_flow_table
*flow_table
;
3595 struct rps_dev_flow
*rflow
;
3600 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3601 if (flow_table
&& flow_id
<= flow_table
->mask
) {
3602 rflow
= &flow_table
->flows
[flow_id
];
3603 cpu
= ACCESS_ONCE(rflow
->cpu
);
3604 if (rflow
->filter
== filter_id
&& cpu
< nr_cpu_ids
&&
3605 ((int)(per_cpu(softnet_data
, cpu
).input_queue_head
-
3606 rflow
->last_qtail
) <
3607 (int)(10 * flow_table
->mask
)))
3613 EXPORT_SYMBOL(rps_may_expire_flow
);
3615 #endif /* CONFIG_RFS_ACCEL */
3617 /* Called from hardirq (IPI) context */
3618 static void rps_trigger_softirq(void *data
)
3620 struct softnet_data
*sd
= data
;
3622 ____napi_schedule(sd
, &sd
->backlog
);
3626 #endif /* CONFIG_RPS */
3629 * Check if this softnet_data structure is another cpu one
3630 * If yes, queue it to our IPI list and return 1
3633 static int rps_ipi_queued(struct softnet_data
*sd
)
3636 struct softnet_data
*mysd
= this_cpu_ptr(&softnet_data
);
3639 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
3640 mysd
->rps_ipi_list
= sd
;
3642 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3645 #endif /* CONFIG_RPS */
3649 #ifdef CONFIG_NET_FLOW_LIMIT
3650 int netdev_flow_limit_table_len __read_mostly
= (1 << 12);
3653 static bool skb_flow_limit(struct sk_buff
*skb
, unsigned int qlen
)
3655 #ifdef CONFIG_NET_FLOW_LIMIT
3656 struct sd_flow_limit
*fl
;
3657 struct softnet_data
*sd
;
3658 unsigned int old_flow
, new_flow
;
3660 if (qlen
< (netdev_max_backlog
>> 1))
3663 sd
= this_cpu_ptr(&softnet_data
);
3666 fl
= rcu_dereference(sd
->flow_limit
);
3668 new_flow
= skb_get_hash(skb
) & (fl
->num_buckets
- 1);
3669 old_flow
= fl
->history
[fl
->history_head
];
3670 fl
->history
[fl
->history_head
] = new_flow
;
3673 fl
->history_head
&= FLOW_LIMIT_HISTORY
- 1;
3675 if (likely(fl
->buckets
[old_flow
]))
3676 fl
->buckets
[old_flow
]--;
3678 if (++fl
->buckets
[new_flow
] > (FLOW_LIMIT_HISTORY
>> 1)) {
3690 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3691 * queue (may be a remote CPU queue).
3693 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
3694 unsigned int *qtail
)
3696 struct softnet_data
*sd
;
3697 unsigned long flags
;
3700 sd
= &per_cpu(softnet_data
, cpu
);
3702 local_irq_save(flags
);
3705 if (!netif_running(skb
->dev
))
3707 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3708 if (qlen
<= netdev_max_backlog
&& !skb_flow_limit(skb
, qlen
)) {
3711 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
3712 input_queue_tail_incr_save(sd
, qtail
);
3714 local_irq_restore(flags
);
3715 return NET_RX_SUCCESS
;
3718 /* Schedule NAPI for backlog device
3719 * We can use non atomic operation since we own the queue lock
3721 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
3722 if (!rps_ipi_queued(sd
))
3723 ____napi_schedule(sd
, &sd
->backlog
);
3732 local_irq_restore(flags
);
3734 atomic_long_inc(&skb
->dev
->rx_dropped
);
3739 static int netif_rx_internal(struct sk_buff
*skb
)
3743 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3745 trace_netif_rx(skb
);
3747 if (static_key_false(&rps_needed
)) {
3748 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3754 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3756 cpu
= smp_processor_id();
3758 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3766 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
3773 * netif_rx - post buffer to the network code
3774 * @skb: buffer to post
3776 * This function receives a packet from a device driver and queues it for
3777 * the upper (protocol) levels to process. It always succeeds. The buffer
3778 * may be dropped during processing for congestion control or by the
3782 * NET_RX_SUCCESS (no congestion)
3783 * NET_RX_DROP (packet was dropped)
3787 int netif_rx(struct sk_buff
*skb
)
3789 trace_netif_rx_entry(skb
);
3791 return netif_rx_internal(skb
);
3793 EXPORT_SYMBOL(netif_rx
);
3795 int netif_rx_ni(struct sk_buff
*skb
)
3799 trace_netif_rx_ni_entry(skb
);
3802 err
= netif_rx_internal(skb
);
3803 if (local_softirq_pending())
3809 EXPORT_SYMBOL(netif_rx_ni
);
3811 static void net_tx_action(struct softirq_action
*h
)
3813 struct softnet_data
*sd
= this_cpu_ptr(&softnet_data
);
3815 if (sd
->completion_queue
) {
3816 struct sk_buff
*clist
;
3818 local_irq_disable();
3819 clist
= sd
->completion_queue
;
3820 sd
->completion_queue
= NULL
;
3824 struct sk_buff
*skb
= clist
;
3825 clist
= clist
->next
;
3827 WARN_ON(atomic_read(&skb
->users
));
3828 if (likely(get_kfree_skb_cb(skb
)->reason
== SKB_REASON_CONSUMED
))
3829 trace_consume_skb(skb
);
3831 trace_kfree_skb(skb
, net_tx_action
);
3833 if (skb
->fclone
!= SKB_FCLONE_UNAVAILABLE
)
3836 __kfree_skb_defer(skb
);
3839 __kfree_skb_flush();
3842 if (sd
->output_queue
) {
3845 local_irq_disable();
3846 head
= sd
->output_queue
;
3847 sd
->output_queue
= NULL
;
3848 sd
->output_queue_tailp
= &sd
->output_queue
;
3852 struct Qdisc
*q
= head
;
3853 spinlock_t
*root_lock
;
3855 head
= head
->next_sched
;
3857 root_lock
= qdisc_lock(q
);
3858 if (spin_trylock(root_lock
)) {
3859 smp_mb__before_atomic();
3860 clear_bit(__QDISC_STATE_SCHED
,
3863 spin_unlock(root_lock
);
3865 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
3867 __netif_reschedule(q
);
3869 smp_mb__before_atomic();
3870 clear_bit(__QDISC_STATE_SCHED
,
3878 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3879 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3880 /* This hook is defined here for ATM LANE */
3881 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
3882 unsigned char *addr
) __read_mostly
;
3883 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
3886 static inline struct sk_buff
*
3887 sch_handle_ingress(struct sk_buff
*skb
, struct packet_type
**pt_prev
, int *ret
,
3888 struct net_device
*orig_dev
)
3890 #ifdef CONFIG_NET_CLS_ACT
3891 struct tcf_proto
*cl
= rcu_dereference_bh(skb
->dev
->ingress_cl_list
);
3892 struct tcf_result cl_res
;
3894 /* If there's at least one ingress present somewhere (so
3895 * we get here via enabled static key), remaining devices
3896 * that are not configured with an ingress qdisc will bail
3902 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
3906 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
3907 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
3908 qdisc_bstats_cpu_update(cl
->q
, skb
);
3910 switch (tc_classify(skb
, cl
, &cl_res
, false)) {
3912 case TC_ACT_RECLASSIFY
:
3913 skb
->tc_index
= TC_H_MIN(cl_res
.classid
);
3916 qdisc_qstats_cpu_drop(cl
->q
);
3921 case TC_ACT_REDIRECT
:
3922 /* skb_mac_header check was done by cls/act_bpf, so
3923 * we can safely push the L2 header back before
3924 * redirecting to another netdev
3926 __skb_push(skb
, skb
->mac_len
);
3927 skb_do_redirect(skb
);
3932 #endif /* CONFIG_NET_CLS_ACT */
3937 * netdev_rx_handler_register - register receive handler
3938 * @dev: device to register a handler for
3939 * @rx_handler: receive handler to register
3940 * @rx_handler_data: data pointer that is used by rx handler
3942 * Register a receive handler for a device. This handler will then be
3943 * called from __netif_receive_skb. A negative errno code is returned
3946 * The caller must hold the rtnl_mutex.
3948 * For a general description of rx_handler, see enum rx_handler_result.
3950 int netdev_rx_handler_register(struct net_device
*dev
,
3951 rx_handler_func_t
*rx_handler
,
3952 void *rx_handler_data
)
3956 if (dev
->rx_handler
)
3959 /* Note: rx_handler_data must be set before rx_handler */
3960 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
3961 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
3965 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
3968 * netdev_rx_handler_unregister - unregister receive handler
3969 * @dev: device to unregister a handler from
3971 * Unregister a receive handler from a device.
3973 * The caller must hold the rtnl_mutex.
3975 void netdev_rx_handler_unregister(struct net_device
*dev
)
3979 RCU_INIT_POINTER(dev
->rx_handler
, NULL
);
3980 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3981 * section has a guarantee to see a non NULL rx_handler_data
3985 RCU_INIT_POINTER(dev
->rx_handler_data
, NULL
);
3987 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
3990 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3991 * the special handling of PFMEMALLOC skbs.
3993 static bool skb_pfmemalloc_protocol(struct sk_buff
*skb
)
3995 switch (skb
->protocol
) {
3996 case htons(ETH_P_ARP
):
3997 case htons(ETH_P_IP
):
3998 case htons(ETH_P_IPV6
):
3999 case htons(ETH_P_8021Q
):
4000 case htons(ETH_P_8021AD
):
4007 static inline int nf_ingress(struct sk_buff
*skb
, struct packet_type
**pt_prev
,
4008 int *ret
, struct net_device
*orig_dev
)
4010 #ifdef CONFIG_NETFILTER_INGRESS
4011 if (nf_hook_ingress_active(skb
)) {
4013 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
4017 return nf_hook_ingress(skb
);
4019 #endif /* CONFIG_NETFILTER_INGRESS */
4023 static int __netif_receive_skb_core(struct sk_buff
*skb
, bool pfmemalloc
)
4025 struct packet_type
*ptype
, *pt_prev
;
4026 rx_handler_func_t
*rx_handler
;
4027 struct net_device
*orig_dev
;
4028 bool deliver_exact
= false;
4029 int ret
= NET_RX_DROP
;
4032 net_timestamp_check(!netdev_tstamp_prequeue
, skb
);
4034 trace_netif_receive_skb(skb
);
4036 orig_dev
= skb
->dev
;
4038 skb_reset_network_header(skb
);
4039 if (!skb_transport_header_was_set(skb
))
4040 skb_reset_transport_header(skb
);
4041 skb_reset_mac_len(skb
);
4046 skb
->skb_iif
= skb
->dev
->ifindex
;
4048 __this_cpu_inc(softnet_data
.processed
);
4050 if (skb
->protocol
== cpu_to_be16(ETH_P_8021Q
) ||
4051 skb
->protocol
== cpu_to_be16(ETH_P_8021AD
)) {
4052 skb
= skb_vlan_untag(skb
);
4057 #ifdef CONFIG_NET_CLS_ACT
4058 if (skb
->tc_verd
& TC_NCLS
) {
4059 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
4067 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
4069 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
4073 list_for_each_entry_rcu(ptype
, &skb
->dev
->ptype_all
, list
) {
4075 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
4080 #ifdef CONFIG_NET_INGRESS
4081 if (static_key_false(&ingress_needed
)) {
4082 skb
= sch_handle_ingress(skb
, &pt_prev
, &ret
, orig_dev
);
4086 if (nf_ingress(skb
, &pt_prev
, &ret
, orig_dev
) < 0)
4090 #ifdef CONFIG_NET_CLS_ACT
4094 if (pfmemalloc
&& !skb_pfmemalloc_protocol(skb
))
4097 if (skb_vlan_tag_present(skb
)) {
4099 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
4102 if (vlan_do_receive(&skb
))
4104 else if (unlikely(!skb
))
4108 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
4111 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
4114 switch (rx_handler(&skb
)) {
4115 case RX_HANDLER_CONSUMED
:
4116 ret
= NET_RX_SUCCESS
;
4118 case RX_HANDLER_ANOTHER
:
4120 case RX_HANDLER_EXACT
:
4121 deliver_exact
= true;
4122 case RX_HANDLER_PASS
:
4129 if (unlikely(skb_vlan_tag_present(skb
))) {
4130 if (skb_vlan_tag_get_id(skb
))
4131 skb
->pkt_type
= PACKET_OTHERHOST
;
4132 /* Note: we might in the future use prio bits
4133 * and set skb->priority like in vlan_do_receive()
4134 * For the time being, just ignore Priority Code Point
4139 type
= skb
->protocol
;
4141 /* deliver only exact match when indicated */
4142 if (likely(!deliver_exact
)) {
4143 deliver_ptype_list_skb(skb
, &pt_prev
, orig_dev
, type
,
4144 &ptype_base
[ntohs(type
) &
4148 deliver_ptype_list_skb(skb
, &pt_prev
, orig_dev
, type
,
4149 &orig_dev
->ptype_specific
);
4151 if (unlikely(skb
->dev
!= orig_dev
)) {
4152 deliver_ptype_list_skb(skb
, &pt_prev
, orig_dev
, type
,
4153 &skb
->dev
->ptype_specific
);
4157 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
4160 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
4164 atomic_long_inc(&skb
->dev
->rx_dropped
);
4166 atomic_long_inc(&skb
->dev
->rx_nohandler
);
4168 /* Jamal, now you will not able to escape explaining
4169 * me how you were going to use this. :-)
4178 static int __netif_receive_skb(struct sk_buff
*skb
)
4182 if (sk_memalloc_socks() && skb_pfmemalloc(skb
)) {
4183 unsigned long pflags
= current
->flags
;
4186 * PFMEMALLOC skbs are special, they should
4187 * - be delivered to SOCK_MEMALLOC sockets only
4188 * - stay away from userspace
4189 * - have bounded memory usage
4191 * Use PF_MEMALLOC as this saves us from propagating the allocation
4192 * context down to all allocation sites.
4194 current
->flags
|= PF_MEMALLOC
;
4195 ret
= __netif_receive_skb_core(skb
, true);
4196 tsk_restore_flags(current
, pflags
, PF_MEMALLOC
);
4198 ret
= __netif_receive_skb_core(skb
, false);
4203 static int netif_receive_skb_internal(struct sk_buff
*skb
)
4207 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
4209 if (skb_defer_rx_timestamp(skb
))
4210 return NET_RX_SUCCESS
;
4215 if (static_key_false(&rps_needed
)) {
4216 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
4217 int cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
4220 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
4226 ret
= __netif_receive_skb(skb
);
4232 * netif_receive_skb - process receive buffer from network
4233 * @skb: buffer to process
4235 * netif_receive_skb() is the main receive data processing function.
4236 * It always succeeds. The buffer may be dropped during processing
4237 * for congestion control or by the protocol layers.
4239 * This function may only be called from softirq context and interrupts
4240 * should be enabled.
4242 * Return values (usually ignored):
4243 * NET_RX_SUCCESS: no congestion
4244 * NET_RX_DROP: packet was dropped
4246 int netif_receive_skb(struct sk_buff
*skb
)
4248 trace_netif_receive_skb_entry(skb
);
4250 return netif_receive_skb_internal(skb
);
4252 EXPORT_SYMBOL(netif_receive_skb
);
4254 /* Network device is going away, flush any packets still pending
4255 * Called with irqs disabled.
4257 static void flush_backlog(void *arg
)
4259 struct net_device
*dev
= arg
;
4260 struct softnet_data
*sd
= this_cpu_ptr(&softnet_data
);
4261 struct sk_buff
*skb
, *tmp
;
4264 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
4265 if (skb
->dev
== dev
) {
4266 __skb_unlink(skb
, &sd
->input_pkt_queue
);
4268 input_queue_head_incr(sd
);
4273 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
4274 if (skb
->dev
== dev
) {
4275 __skb_unlink(skb
, &sd
->process_queue
);
4277 input_queue_head_incr(sd
);
4282 static int napi_gro_complete(struct sk_buff
*skb
)
4284 struct packet_offload
*ptype
;
4285 __be16 type
= skb
->protocol
;
4286 struct list_head
*head
= &offload_base
;
4289 BUILD_BUG_ON(sizeof(struct napi_gro_cb
) > sizeof(skb
->cb
));
4291 if (NAPI_GRO_CB(skb
)->count
== 1) {
4292 skb_shinfo(skb
)->gso_size
= 0;
4297 list_for_each_entry_rcu(ptype
, head
, list
) {
4298 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_complete
)
4301 err
= ptype
->callbacks
.gro_complete(skb
, 0);
4307 WARN_ON(&ptype
->list
== head
);
4309 return NET_RX_SUCCESS
;
4313 return netif_receive_skb_internal(skb
);
4316 /* napi->gro_list contains packets ordered by age.
4317 * youngest packets at the head of it.
4318 * Complete skbs in reverse order to reduce latencies.
4320 void napi_gro_flush(struct napi_struct
*napi
, bool flush_old
)
4322 struct sk_buff
*skb
, *prev
= NULL
;
4324 /* scan list and build reverse chain */
4325 for (skb
= napi
->gro_list
; skb
!= NULL
; skb
= skb
->next
) {
4330 for (skb
= prev
; skb
; skb
= prev
) {
4333 if (flush_old
&& NAPI_GRO_CB(skb
)->age
== jiffies
)
4337 napi_gro_complete(skb
);
4341 napi
->gro_list
= NULL
;
4343 EXPORT_SYMBOL(napi_gro_flush
);
4345 static void gro_list_prepare(struct napi_struct
*napi
, struct sk_buff
*skb
)
4348 unsigned int maclen
= skb
->dev
->hard_header_len
;
4349 u32 hash
= skb_get_hash_raw(skb
);
4351 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
4352 unsigned long diffs
;
4354 NAPI_GRO_CB(p
)->flush
= 0;
4356 if (hash
!= skb_get_hash_raw(p
)) {
4357 NAPI_GRO_CB(p
)->same_flow
= 0;
4361 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
4362 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
4363 diffs
|= skb_metadata_dst_cmp(p
, skb
);
4364 if (maclen
== ETH_HLEN
)
4365 diffs
|= compare_ether_header(skb_mac_header(p
),
4366 skb_mac_header(skb
));
4368 diffs
= memcmp(skb_mac_header(p
),
4369 skb_mac_header(skb
),
4371 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
4375 static void skb_gro_reset_offset(struct sk_buff
*skb
)
4377 const struct skb_shared_info
*pinfo
= skb_shinfo(skb
);
4378 const skb_frag_t
*frag0
= &pinfo
->frags
[0];
4380 NAPI_GRO_CB(skb
)->data_offset
= 0;
4381 NAPI_GRO_CB(skb
)->frag0
= NULL
;
4382 NAPI_GRO_CB(skb
)->frag0_len
= 0;
4384 if (skb_mac_header(skb
) == skb_tail_pointer(skb
) &&
4386 !PageHighMem(skb_frag_page(frag0
))) {
4387 NAPI_GRO_CB(skb
)->frag0
= skb_frag_address(frag0
);
4388 NAPI_GRO_CB(skb
)->frag0_len
= skb_frag_size(frag0
);
4392 static void gro_pull_from_frag0(struct sk_buff
*skb
, int grow
)
4394 struct skb_shared_info
*pinfo
= skb_shinfo(skb
);
4396 BUG_ON(skb
->end
- skb
->tail
< grow
);
4398 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
4400 skb
->data_len
-= grow
;
4403 pinfo
->frags
[0].page_offset
+= grow
;
4404 skb_frag_size_sub(&pinfo
->frags
[0], grow
);
4406 if (unlikely(!skb_frag_size(&pinfo
->frags
[0]))) {
4407 skb_frag_unref(skb
, 0);
4408 memmove(pinfo
->frags
, pinfo
->frags
+ 1,
4409 --pinfo
->nr_frags
* sizeof(pinfo
->frags
[0]));
4413 static enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
4415 struct sk_buff
**pp
= NULL
;
4416 struct packet_offload
*ptype
;
4417 __be16 type
= skb
->protocol
;
4418 struct list_head
*head
= &offload_base
;
4420 enum gro_result ret
;
4423 if (!(skb
->dev
->features
& NETIF_F_GRO
))
4426 if (skb_is_gso(skb
) || skb_has_frag_list(skb
) || skb
->csum_bad
)
4429 gro_list_prepare(napi
, skb
);
4432 list_for_each_entry_rcu(ptype
, head
, list
) {
4433 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_receive
)
4436 skb_set_network_header(skb
, skb_gro_offset(skb
));
4437 skb_reset_mac_len(skb
);
4438 NAPI_GRO_CB(skb
)->same_flow
= 0;
4439 NAPI_GRO_CB(skb
)->flush
= 0;
4440 NAPI_GRO_CB(skb
)->free
= 0;
4441 NAPI_GRO_CB(skb
)->encap_mark
= 0;
4442 NAPI_GRO_CB(skb
)->is_fou
= 0;
4443 NAPI_GRO_CB(skb
)->gro_remcsum_start
= 0;
4445 /* Setup for GRO checksum validation */
4446 switch (skb
->ip_summed
) {
4447 case CHECKSUM_COMPLETE
:
4448 NAPI_GRO_CB(skb
)->csum
= skb
->csum
;
4449 NAPI_GRO_CB(skb
)->csum_valid
= 1;
4450 NAPI_GRO_CB(skb
)->csum_cnt
= 0;
4452 case CHECKSUM_UNNECESSARY
:
4453 NAPI_GRO_CB(skb
)->csum_cnt
= skb
->csum_level
+ 1;
4454 NAPI_GRO_CB(skb
)->csum_valid
= 0;
4457 NAPI_GRO_CB(skb
)->csum_cnt
= 0;
4458 NAPI_GRO_CB(skb
)->csum_valid
= 0;
4461 pp
= ptype
->callbacks
.gro_receive(&napi
->gro_list
, skb
);
4466 if (&ptype
->list
== head
)
4469 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
4470 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
4473 struct sk_buff
*nskb
= *pp
;
4477 napi_gro_complete(nskb
);
4484 if (NAPI_GRO_CB(skb
)->flush
)
4487 if (unlikely(napi
->gro_count
>= MAX_GRO_SKBS
)) {
4488 struct sk_buff
*nskb
= napi
->gro_list
;
4490 /* locate the end of the list to select the 'oldest' flow */
4491 while (nskb
->next
) {
4497 napi_gro_complete(nskb
);
4501 NAPI_GRO_CB(skb
)->count
= 1;
4502 NAPI_GRO_CB(skb
)->age
= jiffies
;
4503 NAPI_GRO_CB(skb
)->last
= skb
;
4504 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
4505 skb
->next
= napi
->gro_list
;
4506 napi
->gro_list
= skb
;
4510 grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
4512 gro_pull_from_frag0(skb
, grow
);
4521 struct packet_offload
*gro_find_receive_by_type(__be16 type
)
4523 struct list_head
*offload_head
= &offload_base
;
4524 struct packet_offload
*ptype
;
4526 list_for_each_entry_rcu(ptype
, offload_head
, list
) {
4527 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_receive
)
4533 EXPORT_SYMBOL(gro_find_receive_by_type
);
4535 struct packet_offload
*gro_find_complete_by_type(__be16 type
)
4537 struct list_head
*offload_head
= &offload_base
;
4538 struct packet_offload
*ptype
;
4540 list_for_each_entry_rcu(ptype
, offload_head
, list
) {
4541 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_complete
)
4547 EXPORT_SYMBOL(gro_find_complete_by_type
);
4549 static gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
4553 if (netif_receive_skb_internal(skb
))
4561 case GRO_MERGED_FREE
:
4562 if (NAPI_GRO_CB(skb
)->free
== NAPI_GRO_FREE_STOLEN_HEAD
) {
4564 kmem_cache_free(skbuff_head_cache
, skb
);
4578 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
4580 skb_mark_napi_id(skb
, napi
);
4581 trace_napi_gro_receive_entry(skb
);
4583 skb_gro_reset_offset(skb
);
4585 return napi_skb_finish(dev_gro_receive(napi
, skb
), skb
);
4587 EXPORT_SYMBOL(napi_gro_receive
);
4589 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
4591 if (unlikely(skb
->pfmemalloc
)) {
4595 __skb_pull(skb
, skb_headlen(skb
));
4596 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4597 skb_reserve(skb
, NET_SKB_PAD
+ NET_IP_ALIGN
- skb_headroom(skb
));
4599 skb
->dev
= napi
->dev
;
4601 skb
->encapsulation
= 0;
4602 skb_shinfo(skb
)->gso_type
= 0;
4603 skb
->truesize
= SKB_TRUESIZE(skb_end_offset(skb
));
4608 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
4610 struct sk_buff
*skb
= napi
->skb
;
4613 skb
= napi_alloc_skb(napi
, GRO_MAX_HEAD
);
4616 skb_mark_napi_id(skb
, napi
);
4621 EXPORT_SYMBOL(napi_get_frags
);
4623 static gro_result_t
napi_frags_finish(struct napi_struct
*napi
,
4624 struct sk_buff
*skb
,
4630 __skb_push(skb
, ETH_HLEN
);
4631 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
4632 if (ret
== GRO_NORMAL
&& netif_receive_skb_internal(skb
))
4637 case GRO_MERGED_FREE
:
4638 napi_reuse_skb(napi
, skb
);
4648 /* Upper GRO stack assumes network header starts at gro_offset=0
4649 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4650 * We copy ethernet header into skb->data to have a common layout.
4652 static struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
4654 struct sk_buff
*skb
= napi
->skb
;
4655 const struct ethhdr
*eth
;
4656 unsigned int hlen
= sizeof(*eth
);
4660 skb_reset_mac_header(skb
);
4661 skb_gro_reset_offset(skb
);
4663 eth
= skb_gro_header_fast(skb
, 0);
4664 if (unlikely(skb_gro_header_hard(skb
, hlen
))) {
4665 eth
= skb_gro_header_slow(skb
, hlen
, 0);
4666 if (unlikely(!eth
)) {
4667 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
4668 __func__
, napi
->dev
->name
);
4669 napi_reuse_skb(napi
, skb
);
4673 gro_pull_from_frag0(skb
, hlen
);
4674 NAPI_GRO_CB(skb
)->frag0
+= hlen
;
4675 NAPI_GRO_CB(skb
)->frag0_len
-= hlen
;
4677 __skb_pull(skb
, hlen
);
4680 * This works because the only protocols we care about don't require
4682 * We'll fix it up properly in napi_frags_finish()
4684 skb
->protocol
= eth
->h_proto
;
4689 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
4691 struct sk_buff
*skb
= napi_frags_skb(napi
);
4696 trace_napi_gro_frags_entry(skb
);
4698 return napi_frags_finish(napi
, skb
, dev_gro_receive(napi
, skb
));
4700 EXPORT_SYMBOL(napi_gro_frags
);
4702 /* Compute the checksum from gro_offset and return the folded value
4703 * after adding in any pseudo checksum.
4705 __sum16
__skb_gro_checksum_complete(struct sk_buff
*skb
)
4710 wsum
= skb_checksum(skb
, skb_gro_offset(skb
), skb_gro_len(skb
), 0);
4712 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4713 sum
= csum_fold(csum_add(NAPI_GRO_CB(skb
)->csum
, wsum
));
4715 if (unlikely(skb
->ip_summed
== CHECKSUM_COMPLETE
) &&
4716 !skb
->csum_complete_sw
)
4717 netdev_rx_csum_fault(skb
->dev
);
4720 NAPI_GRO_CB(skb
)->csum
= wsum
;
4721 NAPI_GRO_CB(skb
)->csum_valid
= 1;
4725 EXPORT_SYMBOL(__skb_gro_checksum_complete
);
4728 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4729 * Note: called with local irq disabled, but exits with local irq enabled.
4731 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
4734 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
4737 sd
->rps_ipi_list
= NULL
;
4741 /* Send pending IPI's to kick RPS processing on remote cpus. */
4743 struct softnet_data
*next
= remsd
->rps_ipi_next
;
4745 if (cpu_online(remsd
->cpu
))
4746 smp_call_function_single_async(remsd
->cpu
,
4755 static bool sd_has_rps_ipi_waiting(struct softnet_data
*sd
)
4758 return sd
->rps_ipi_list
!= NULL
;
4764 static int process_backlog(struct napi_struct
*napi
, int quota
)
4767 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
4769 /* Check if we have pending ipi, its better to send them now,
4770 * not waiting net_rx_action() end.
4772 if (sd_has_rps_ipi_waiting(sd
)) {
4773 local_irq_disable();
4774 net_rps_action_and_irq_enable(sd
);
4777 napi
->weight
= weight_p
;
4778 local_irq_disable();
4780 struct sk_buff
*skb
;
4782 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
4785 __netif_receive_skb(skb
);
4787 local_irq_disable();
4788 input_queue_head_incr(sd
);
4789 if (++work
>= quota
) {
4796 if (skb_queue_empty(&sd
->input_pkt_queue
)) {
4798 * Inline a custom version of __napi_complete().
4799 * only current cpu owns and manipulates this napi,
4800 * and NAPI_STATE_SCHED is the only possible flag set
4802 * We can use a plain write instead of clear_bit(),
4803 * and we dont need an smp_mb() memory barrier.
4811 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
4812 &sd
->process_queue
);
4821 * __napi_schedule - schedule for receive
4822 * @n: entry to schedule
4824 * The entry's receive function will be scheduled to run.
4825 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
4827 void __napi_schedule(struct napi_struct
*n
)
4829 unsigned long flags
;
4831 local_irq_save(flags
);
4832 ____napi_schedule(this_cpu_ptr(&softnet_data
), n
);
4833 local_irq_restore(flags
);
4835 EXPORT_SYMBOL(__napi_schedule
);
4838 * __napi_schedule_irqoff - schedule for receive
4839 * @n: entry to schedule
4841 * Variant of __napi_schedule() assuming hard irqs are masked
4843 void __napi_schedule_irqoff(struct napi_struct
*n
)
4845 ____napi_schedule(this_cpu_ptr(&softnet_data
), n
);
4847 EXPORT_SYMBOL(__napi_schedule_irqoff
);
4849 void __napi_complete(struct napi_struct
*n
)
4851 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
4853 list_del_init(&n
->poll_list
);
4854 smp_mb__before_atomic();
4855 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
4857 EXPORT_SYMBOL(__napi_complete
);
4859 void napi_complete_done(struct napi_struct
*n
, int work_done
)
4861 unsigned long flags
;
4864 * don't let napi dequeue from the cpu poll list
4865 * just in case its running on a different cpu
4867 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
4871 unsigned long timeout
= 0;
4874 timeout
= n
->dev
->gro_flush_timeout
;
4877 hrtimer_start(&n
->timer
, ns_to_ktime(timeout
),
4878 HRTIMER_MODE_REL_PINNED
);
4880 napi_gro_flush(n
, false);
4882 if (likely(list_empty(&n
->poll_list
))) {
4883 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED
, &n
->state
));
4885 /* If n->poll_list is not empty, we need to mask irqs */
4886 local_irq_save(flags
);
4888 local_irq_restore(flags
);
4891 EXPORT_SYMBOL(napi_complete_done
);
4893 /* must be called under rcu_read_lock(), as we dont take a reference */
4894 static struct napi_struct
*napi_by_id(unsigned int napi_id
)
4896 unsigned int hash
= napi_id
% HASH_SIZE(napi_hash
);
4897 struct napi_struct
*napi
;
4899 hlist_for_each_entry_rcu(napi
, &napi_hash
[hash
], napi_hash_node
)
4900 if (napi
->napi_id
== napi_id
)
4906 #if defined(CONFIG_NET_RX_BUSY_POLL)
4907 #define BUSY_POLL_BUDGET 8
4908 bool sk_busy_loop(struct sock
*sk
, int nonblock
)
4910 unsigned long end_time
= !nonblock
? sk_busy_loop_end_time(sk
) : 0;
4911 int (*busy_poll
)(struct napi_struct
*dev
);
4912 struct napi_struct
*napi
;
4917 napi
= napi_by_id(sk
->sk_napi_id
);
4921 /* Note: ndo_busy_poll method is optional in linux-4.5 */
4922 busy_poll
= napi
->dev
->netdev_ops
->ndo_busy_poll
;
4928 rc
= busy_poll(napi
);
4929 } else if (napi_schedule_prep(napi
)) {
4930 void *have
= netpoll_poll_lock(napi
);
4932 if (test_bit(NAPI_STATE_SCHED
, &napi
->state
)) {
4933 rc
= napi
->poll(napi
, BUSY_POLL_BUDGET
);
4934 trace_napi_poll(napi
);
4935 if (rc
== BUSY_POLL_BUDGET
) {
4936 napi_complete_done(napi
, rc
);
4937 napi_schedule(napi
);
4940 netpoll_poll_unlock(have
);
4943 NET_ADD_STATS_BH(sock_net(sk
),
4944 LINUX_MIB_BUSYPOLLRXPACKETS
, rc
);
4947 if (rc
== LL_FLUSH_FAILED
)
4948 break; /* permanent failure */
4951 } while (!nonblock
&& skb_queue_empty(&sk
->sk_receive_queue
) &&
4952 !need_resched() && !busy_loop_timeout(end_time
));
4954 rc
= !skb_queue_empty(&sk
->sk_receive_queue
);
4959 EXPORT_SYMBOL(sk_busy_loop
);
4961 #endif /* CONFIG_NET_RX_BUSY_POLL */
4963 void napi_hash_add(struct napi_struct
*napi
)
4965 if (test_bit(NAPI_STATE_NO_BUSY_POLL
, &napi
->state
) ||
4966 test_and_set_bit(NAPI_STATE_HASHED
, &napi
->state
))
4969 spin_lock(&napi_hash_lock
);
4971 /* 0..NR_CPUS+1 range is reserved for sender_cpu use */
4973 if (unlikely(++napi_gen_id
< NR_CPUS
+ 1))
4974 napi_gen_id
= NR_CPUS
+ 1;
4975 } while (napi_by_id(napi_gen_id
));
4976 napi
->napi_id
= napi_gen_id
;
4978 hlist_add_head_rcu(&napi
->napi_hash_node
,
4979 &napi_hash
[napi
->napi_id
% HASH_SIZE(napi_hash
)]);
4981 spin_unlock(&napi_hash_lock
);
4983 EXPORT_SYMBOL_GPL(napi_hash_add
);
4985 /* Warning : caller is responsible to make sure rcu grace period
4986 * is respected before freeing memory containing @napi
4988 bool napi_hash_del(struct napi_struct
*napi
)
4990 bool rcu_sync_needed
= false;
4992 spin_lock(&napi_hash_lock
);
4994 if (test_and_clear_bit(NAPI_STATE_HASHED
, &napi
->state
)) {
4995 rcu_sync_needed
= true;
4996 hlist_del_rcu(&napi
->napi_hash_node
);
4998 spin_unlock(&napi_hash_lock
);
4999 return rcu_sync_needed
;
5001 EXPORT_SYMBOL_GPL(napi_hash_del
);
5003 static enum hrtimer_restart
napi_watchdog(struct hrtimer
*timer
)
5005 struct napi_struct
*napi
;
5007 napi
= container_of(timer
, struct napi_struct
, timer
);
5009 napi_schedule(napi
);
5011 return HRTIMER_NORESTART
;
5014 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
5015 int (*poll
)(struct napi_struct
*, int), int weight
)
5017 INIT_LIST_HEAD(&napi
->poll_list
);
5018 hrtimer_init(&napi
->timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_REL_PINNED
);
5019 napi
->timer
.function
= napi_watchdog
;
5020 napi
->gro_count
= 0;
5021 napi
->gro_list
= NULL
;
5024 if (weight
> NAPI_POLL_WEIGHT
)
5025 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
5027 napi
->weight
= weight
;
5028 list_add(&napi
->dev_list
, &dev
->napi_list
);
5030 #ifdef CONFIG_NETPOLL
5031 spin_lock_init(&napi
->poll_lock
);
5032 napi
->poll_owner
= -1;
5034 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
5035 napi_hash_add(napi
);
5037 EXPORT_SYMBOL(netif_napi_add
);
5039 void napi_disable(struct napi_struct
*n
)
5042 set_bit(NAPI_STATE_DISABLE
, &n
->state
);
5044 while (test_and_set_bit(NAPI_STATE_SCHED
, &n
->state
))
5046 while (test_and_set_bit(NAPI_STATE_NPSVC
, &n
->state
))
5049 hrtimer_cancel(&n
->timer
);
5051 clear_bit(NAPI_STATE_DISABLE
, &n
->state
);
5053 EXPORT_SYMBOL(napi_disable
);
5055 /* Must be called in process context */
5056 void netif_napi_del(struct napi_struct
*napi
)
5059 if (napi_hash_del(napi
))
5061 list_del_init(&napi
->dev_list
);
5062 napi_free_frags(napi
);
5064 kfree_skb_list(napi
->gro_list
);
5065 napi
->gro_list
= NULL
;
5066 napi
->gro_count
= 0;
5068 EXPORT_SYMBOL(netif_napi_del
);
5070 static int napi_poll(struct napi_struct
*n
, struct list_head
*repoll
)
5075 list_del_init(&n
->poll_list
);
5077 have
= netpoll_poll_lock(n
);
5081 /* This NAPI_STATE_SCHED test is for avoiding a race
5082 * with netpoll's poll_napi(). Only the entity which
5083 * obtains the lock and sees NAPI_STATE_SCHED set will
5084 * actually make the ->poll() call. Therefore we avoid
5085 * accidentally calling ->poll() when NAPI is not scheduled.
5088 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
5089 work
= n
->poll(n
, weight
);
5093 WARN_ON_ONCE(work
> weight
);
5095 if (likely(work
< weight
))
5098 /* Drivers must not modify the NAPI state if they
5099 * consume the entire weight. In such cases this code
5100 * still "owns" the NAPI instance and therefore can
5101 * move the instance around on the list at-will.
5103 if (unlikely(napi_disable_pending(n
))) {
5109 /* flush too old packets
5110 * If HZ < 1000, flush all packets.
5112 napi_gro_flush(n
, HZ
>= 1000);
5115 /* Some drivers may have called napi_schedule
5116 * prior to exhausting their budget.
5118 if (unlikely(!list_empty(&n
->poll_list
))) {
5119 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
5120 n
->dev
? n
->dev
->name
: "backlog");
5124 list_add_tail(&n
->poll_list
, repoll
);
5127 netpoll_poll_unlock(have
);
5132 static void net_rx_action(struct softirq_action
*h
)
5134 struct softnet_data
*sd
= this_cpu_ptr(&softnet_data
);
5135 unsigned long time_limit
= jiffies
+ 2;
5136 int budget
= netdev_budget
;
5140 local_irq_disable();
5141 list_splice_init(&sd
->poll_list
, &list
);
5145 struct napi_struct
*n
;
5147 if (list_empty(&list
)) {
5148 if (!sd_has_rps_ipi_waiting(sd
) && list_empty(&repoll
))
5153 n
= list_first_entry(&list
, struct napi_struct
, poll_list
);
5154 budget
-= napi_poll(n
, &repoll
);
5156 /* If softirq window is exhausted then punt.
5157 * Allow this to run for 2 jiffies since which will allow
5158 * an average latency of 1.5/HZ.
5160 if (unlikely(budget
<= 0 ||
5161 time_after_eq(jiffies
, time_limit
))) {
5167 __kfree_skb_flush();
5168 local_irq_disable();
5170 list_splice_tail_init(&sd
->poll_list
, &list
);
5171 list_splice_tail(&repoll
, &list
);
5172 list_splice(&list
, &sd
->poll_list
);
5173 if (!list_empty(&sd
->poll_list
))
5174 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
5176 net_rps_action_and_irq_enable(sd
);
5179 struct netdev_adjacent
{
5180 struct net_device
*dev
;
5182 /* upper master flag, there can only be one master device per list */
5185 /* counter for the number of times this device was added to us */
5188 /* private field for the users */
5191 struct list_head list
;
5192 struct rcu_head rcu
;
5195 static struct netdev_adjacent
*__netdev_find_adj(struct net_device
*adj_dev
,
5196 struct list_head
*adj_list
)
5198 struct netdev_adjacent
*adj
;
5200 list_for_each_entry(adj
, adj_list
, list
) {
5201 if (adj
->dev
== adj_dev
)
5208 * netdev_has_upper_dev - Check if device is linked to an upper device
5210 * @upper_dev: upper device to check
5212 * Find out if a device is linked to specified upper device and return true
5213 * in case it is. Note that this checks only immediate upper device,
5214 * not through a complete stack of devices. The caller must hold the RTNL lock.
5216 bool netdev_has_upper_dev(struct net_device
*dev
,
5217 struct net_device
*upper_dev
)
5221 return __netdev_find_adj(upper_dev
, &dev
->all_adj_list
.upper
);
5223 EXPORT_SYMBOL(netdev_has_upper_dev
);
5226 * netdev_has_any_upper_dev - Check if device is linked to some device
5229 * Find out if a device is linked to an upper device and return true in case
5230 * it is. The caller must hold the RTNL lock.
5232 static bool netdev_has_any_upper_dev(struct net_device
*dev
)
5236 return !list_empty(&dev
->all_adj_list
.upper
);
5240 * netdev_master_upper_dev_get - Get master upper device
5243 * Find a master upper device and return pointer to it or NULL in case
5244 * it's not there. The caller must hold the RTNL lock.
5246 struct net_device
*netdev_master_upper_dev_get(struct net_device
*dev
)
5248 struct netdev_adjacent
*upper
;
5252 if (list_empty(&dev
->adj_list
.upper
))
5255 upper
= list_first_entry(&dev
->adj_list
.upper
,
5256 struct netdev_adjacent
, list
);
5257 if (likely(upper
->master
))
5261 EXPORT_SYMBOL(netdev_master_upper_dev_get
);
5263 void *netdev_adjacent_get_private(struct list_head
*adj_list
)
5265 struct netdev_adjacent
*adj
;
5267 adj
= list_entry(adj_list
, struct netdev_adjacent
, list
);
5269 return adj
->private;
5271 EXPORT_SYMBOL(netdev_adjacent_get_private
);
5274 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
5276 * @iter: list_head ** of the current position
5278 * Gets the next device from the dev's upper list, starting from iter
5279 * position. The caller must hold RCU read lock.
5281 struct net_device
*netdev_upper_get_next_dev_rcu(struct net_device
*dev
,
5282 struct list_head
**iter
)
5284 struct netdev_adjacent
*upper
;
5286 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5288 upper
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
5290 if (&upper
->list
== &dev
->adj_list
.upper
)
5293 *iter
= &upper
->list
;
5297 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu
);
5300 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
5302 * @iter: list_head ** of the current position
5304 * Gets the next device from the dev's upper list, starting from iter
5305 * position. The caller must hold RCU read lock.
5307 struct net_device
*netdev_all_upper_get_next_dev_rcu(struct net_device
*dev
,
5308 struct list_head
**iter
)
5310 struct netdev_adjacent
*upper
;
5312 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5314 upper
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
5316 if (&upper
->list
== &dev
->all_adj_list
.upper
)
5319 *iter
= &upper
->list
;
5323 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu
);
5326 * netdev_lower_get_next_private - Get the next ->private from the
5327 * lower neighbour list
5329 * @iter: list_head ** of the current position
5331 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5332 * list, starting from iter position. The caller must hold either hold the
5333 * RTNL lock or its own locking that guarantees that the neighbour lower
5334 * list will remain unchanged.
5336 void *netdev_lower_get_next_private(struct net_device
*dev
,
5337 struct list_head
**iter
)
5339 struct netdev_adjacent
*lower
;
5341 lower
= list_entry(*iter
, struct netdev_adjacent
, list
);
5343 if (&lower
->list
== &dev
->adj_list
.lower
)
5346 *iter
= lower
->list
.next
;
5348 return lower
->private;
5350 EXPORT_SYMBOL(netdev_lower_get_next_private
);
5353 * netdev_lower_get_next_private_rcu - Get the next ->private from the
5354 * lower neighbour list, RCU
5357 * @iter: list_head ** of the current position
5359 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5360 * list, starting from iter position. The caller must hold RCU read lock.
5362 void *netdev_lower_get_next_private_rcu(struct net_device
*dev
,
5363 struct list_head
**iter
)
5365 struct netdev_adjacent
*lower
;
5367 WARN_ON_ONCE(!rcu_read_lock_held());
5369 lower
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
5371 if (&lower
->list
== &dev
->adj_list
.lower
)
5374 *iter
= &lower
->list
;
5376 return lower
->private;
5378 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu
);
5381 * netdev_lower_get_next - Get the next device from the lower neighbour
5384 * @iter: list_head ** of the current position
5386 * Gets the next netdev_adjacent from the dev's lower neighbour
5387 * list, starting from iter position. The caller must hold RTNL lock or
5388 * its own locking that guarantees that the neighbour lower
5389 * list will remain unchanged.
5391 void *netdev_lower_get_next(struct net_device
*dev
, struct list_head
**iter
)
5393 struct netdev_adjacent
*lower
;
5395 lower
= list_entry(*iter
, struct netdev_adjacent
, list
);
5397 if (&lower
->list
== &dev
->adj_list
.lower
)
5400 *iter
= lower
->list
.next
;
5404 EXPORT_SYMBOL(netdev_lower_get_next
);
5407 * netdev_lower_get_first_private_rcu - Get the first ->private from the
5408 * lower neighbour list, RCU
5412 * Gets the first netdev_adjacent->private from the dev's lower neighbour
5413 * list. The caller must hold RCU read lock.
5415 void *netdev_lower_get_first_private_rcu(struct net_device
*dev
)
5417 struct netdev_adjacent
*lower
;
5419 lower
= list_first_or_null_rcu(&dev
->adj_list
.lower
,
5420 struct netdev_adjacent
, list
);
5422 return lower
->private;
5425 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu
);
5428 * netdev_master_upper_dev_get_rcu - Get master upper device
5431 * Find a master upper device and return pointer to it or NULL in case
5432 * it's not there. The caller must hold the RCU read lock.
5434 struct net_device
*netdev_master_upper_dev_get_rcu(struct net_device
*dev
)
5436 struct netdev_adjacent
*upper
;
5438 upper
= list_first_or_null_rcu(&dev
->adj_list
.upper
,
5439 struct netdev_adjacent
, list
);
5440 if (upper
&& likely(upper
->master
))
5444 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu
);
5446 static int netdev_adjacent_sysfs_add(struct net_device
*dev
,
5447 struct net_device
*adj_dev
,
5448 struct list_head
*dev_list
)
5450 char linkname
[IFNAMSIZ
+7];
5451 sprintf(linkname
, dev_list
== &dev
->adj_list
.upper
?
5452 "upper_%s" : "lower_%s", adj_dev
->name
);
5453 return sysfs_create_link(&(dev
->dev
.kobj
), &(adj_dev
->dev
.kobj
),
5456 static void netdev_adjacent_sysfs_del(struct net_device
*dev
,
5458 struct list_head
*dev_list
)
5460 char linkname
[IFNAMSIZ
+7];
5461 sprintf(linkname
, dev_list
== &dev
->adj_list
.upper
?
5462 "upper_%s" : "lower_%s", name
);
5463 sysfs_remove_link(&(dev
->dev
.kobj
), linkname
);
5466 static inline bool netdev_adjacent_is_neigh_list(struct net_device
*dev
,
5467 struct net_device
*adj_dev
,
5468 struct list_head
*dev_list
)
5470 return (dev_list
== &dev
->adj_list
.upper
||
5471 dev_list
== &dev
->adj_list
.lower
) &&
5472 net_eq(dev_net(dev
), dev_net(adj_dev
));
5475 static int __netdev_adjacent_dev_insert(struct net_device
*dev
,
5476 struct net_device
*adj_dev
,
5477 struct list_head
*dev_list
,
5478 void *private, bool master
)
5480 struct netdev_adjacent
*adj
;
5483 adj
= __netdev_find_adj(adj_dev
, dev_list
);
5490 adj
= kmalloc(sizeof(*adj
), GFP_KERNEL
);
5495 adj
->master
= master
;
5497 adj
->private = private;
5500 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
5501 adj_dev
->name
, dev
->name
, adj_dev
->name
);
5503 if (netdev_adjacent_is_neigh_list(dev
, adj_dev
, dev_list
)) {
5504 ret
= netdev_adjacent_sysfs_add(dev
, adj_dev
, dev_list
);
5509 /* Ensure that master link is always the first item in list. */
5511 ret
= sysfs_create_link(&(dev
->dev
.kobj
),
5512 &(adj_dev
->dev
.kobj
), "master");
5514 goto remove_symlinks
;
5516 list_add_rcu(&adj
->list
, dev_list
);
5518 list_add_tail_rcu(&adj
->list
, dev_list
);
5524 if (netdev_adjacent_is_neigh_list(dev
, adj_dev
, dev_list
))
5525 netdev_adjacent_sysfs_del(dev
, adj_dev
->name
, dev_list
);
5533 static void __netdev_adjacent_dev_remove(struct net_device
*dev
,
5534 struct net_device
*adj_dev
,
5535 struct list_head
*dev_list
)
5537 struct netdev_adjacent
*adj
;
5539 adj
= __netdev_find_adj(adj_dev
, dev_list
);
5542 pr_err("tried to remove device %s from %s\n",
5543 dev
->name
, adj_dev
->name
);
5547 if (adj
->ref_nr
> 1) {
5548 pr_debug("%s to %s ref_nr-- = %d\n", dev
->name
, adj_dev
->name
,
5555 sysfs_remove_link(&(dev
->dev
.kobj
), "master");
5557 if (netdev_adjacent_is_neigh_list(dev
, adj_dev
, dev_list
))
5558 netdev_adjacent_sysfs_del(dev
, adj_dev
->name
, dev_list
);
5560 list_del_rcu(&adj
->list
);
5561 pr_debug("dev_put for %s, because link removed from %s to %s\n",
5562 adj_dev
->name
, dev
->name
, adj_dev
->name
);
5564 kfree_rcu(adj
, rcu
);
5567 static int __netdev_adjacent_dev_link_lists(struct net_device
*dev
,
5568 struct net_device
*upper_dev
,
5569 struct list_head
*up_list
,
5570 struct list_head
*down_list
,
5571 void *private, bool master
)
5575 ret
= __netdev_adjacent_dev_insert(dev
, upper_dev
, up_list
, private,
5580 ret
= __netdev_adjacent_dev_insert(upper_dev
, dev
, down_list
, private,
5583 __netdev_adjacent_dev_remove(dev
, upper_dev
, up_list
);
5590 static int __netdev_adjacent_dev_link(struct net_device
*dev
,
5591 struct net_device
*upper_dev
)
5593 return __netdev_adjacent_dev_link_lists(dev
, upper_dev
,
5594 &dev
->all_adj_list
.upper
,
5595 &upper_dev
->all_adj_list
.lower
,
5599 static void __netdev_adjacent_dev_unlink_lists(struct net_device
*dev
,
5600 struct net_device
*upper_dev
,
5601 struct list_head
*up_list
,
5602 struct list_head
*down_list
)
5604 __netdev_adjacent_dev_remove(dev
, upper_dev
, up_list
);
5605 __netdev_adjacent_dev_remove(upper_dev
, dev
, down_list
);
5608 static void __netdev_adjacent_dev_unlink(struct net_device
*dev
,
5609 struct net_device
*upper_dev
)
5611 __netdev_adjacent_dev_unlink_lists(dev
, upper_dev
,
5612 &dev
->all_adj_list
.upper
,
5613 &upper_dev
->all_adj_list
.lower
);
5616 static int __netdev_adjacent_dev_link_neighbour(struct net_device
*dev
,
5617 struct net_device
*upper_dev
,
5618 void *private, bool master
)
5620 int ret
= __netdev_adjacent_dev_link(dev
, upper_dev
);
5625 ret
= __netdev_adjacent_dev_link_lists(dev
, upper_dev
,
5626 &dev
->adj_list
.upper
,
5627 &upper_dev
->adj_list
.lower
,
5630 __netdev_adjacent_dev_unlink(dev
, upper_dev
);
5637 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device
*dev
,
5638 struct net_device
*upper_dev
)
5640 __netdev_adjacent_dev_unlink(dev
, upper_dev
);
5641 __netdev_adjacent_dev_unlink_lists(dev
, upper_dev
,
5642 &dev
->adj_list
.upper
,
5643 &upper_dev
->adj_list
.lower
);
5646 static int __netdev_upper_dev_link(struct net_device
*dev
,
5647 struct net_device
*upper_dev
, bool master
,
5648 void *upper_priv
, void *upper_info
)
5650 struct netdev_notifier_changeupper_info changeupper_info
;
5651 struct netdev_adjacent
*i
, *j
, *to_i
, *to_j
;
5656 if (dev
== upper_dev
)
5659 /* To prevent loops, check if dev is not upper device to upper_dev. */
5660 if (__netdev_find_adj(dev
, &upper_dev
->all_adj_list
.upper
))
5663 if (__netdev_find_adj(upper_dev
, &dev
->adj_list
.upper
))
5666 if (master
&& netdev_master_upper_dev_get(dev
))
5669 changeupper_info
.upper_dev
= upper_dev
;
5670 changeupper_info
.master
= master
;
5671 changeupper_info
.linking
= true;
5672 changeupper_info
.upper_info
= upper_info
;
5674 ret
= call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER
, dev
,
5675 &changeupper_info
.info
);
5676 ret
= notifier_to_errno(ret
);
5680 ret
= __netdev_adjacent_dev_link_neighbour(dev
, upper_dev
, upper_priv
,
5685 /* Now that we linked these devs, make all the upper_dev's
5686 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5687 * versa, and don't forget the devices itself. All of these
5688 * links are non-neighbours.
5690 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5691 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
) {
5692 pr_debug("Interlinking %s with %s, non-neighbour\n",
5693 i
->dev
->name
, j
->dev
->name
);
5694 ret
= __netdev_adjacent_dev_link(i
->dev
, j
->dev
);
5700 /* add dev to every upper_dev's upper device */
5701 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
) {
5702 pr_debug("linking %s's upper device %s with %s\n",
5703 upper_dev
->name
, i
->dev
->name
, dev
->name
);
5704 ret
= __netdev_adjacent_dev_link(dev
, i
->dev
);
5706 goto rollback_upper_mesh
;
5709 /* add upper_dev to every dev's lower device */
5710 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5711 pr_debug("linking %s's lower device %s with %s\n", dev
->name
,
5712 i
->dev
->name
, upper_dev
->name
);
5713 ret
= __netdev_adjacent_dev_link(i
->dev
, upper_dev
);
5715 goto rollback_lower_mesh
;
5718 ret
= call_netdevice_notifiers_info(NETDEV_CHANGEUPPER
, dev
,
5719 &changeupper_info
.info
);
5720 ret
= notifier_to_errno(ret
);
5722 goto rollback_lower_mesh
;
5726 rollback_lower_mesh
:
5728 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5731 __netdev_adjacent_dev_unlink(i
->dev
, upper_dev
);
5736 rollback_upper_mesh
:
5738 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
) {
5741 __netdev_adjacent_dev_unlink(dev
, i
->dev
);
5749 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5750 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
) {
5751 if (i
== to_i
&& j
== to_j
)
5753 __netdev_adjacent_dev_unlink(i
->dev
, j
->dev
);
5759 __netdev_adjacent_dev_unlink_neighbour(dev
, upper_dev
);
5765 * netdev_upper_dev_link - Add a link to the upper device
5767 * @upper_dev: new upper device
5769 * Adds a link to device which is upper to this one. The caller must hold
5770 * the RTNL lock. On a failure a negative errno code is returned.
5771 * On success the reference counts are adjusted and the function
5774 int netdev_upper_dev_link(struct net_device
*dev
,
5775 struct net_device
*upper_dev
)
5777 return __netdev_upper_dev_link(dev
, upper_dev
, false, NULL
, NULL
);
5779 EXPORT_SYMBOL(netdev_upper_dev_link
);
5782 * netdev_master_upper_dev_link - Add a master link to the upper device
5784 * @upper_dev: new upper device
5785 * @upper_priv: upper device private
5786 * @upper_info: upper info to be passed down via notifier
5788 * Adds a link to device which is upper to this one. In this case, only
5789 * one master upper device can be linked, although other non-master devices
5790 * might be linked as well. The caller must hold the RTNL lock.
5791 * On a failure a negative errno code is returned. On success the reference
5792 * counts are adjusted and the function returns zero.
5794 int netdev_master_upper_dev_link(struct net_device
*dev
,
5795 struct net_device
*upper_dev
,
5796 void *upper_priv
, void *upper_info
)
5798 return __netdev_upper_dev_link(dev
, upper_dev
, true,
5799 upper_priv
, upper_info
);
5801 EXPORT_SYMBOL(netdev_master_upper_dev_link
);
5804 * netdev_upper_dev_unlink - Removes a link to upper device
5806 * @upper_dev: new upper device
5808 * Removes a link to device which is upper to this one. The caller must hold
5811 void netdev_upper_dev_unlink(struct net_device
*dev
,
5812 struct net_device
*upper_dev
)
5814 struct netdev_notifier_changeupper_info changeupper_info
;
5815 struct netdev_adjacent
*i
, *j
;
5818 changeupper_info
.upper_dev
= upper_dev
;
5819 changeupper_info
.master
= netdev_master_upper_dev_get(dev
) == upper_dev
;
5820 changeupper_info
.linking
= false;
5822 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER
, dev
,
5823 &changeupper_info
.info
);
5825 __netdev_adjacent_dev_unlink_neighbour(dev
, upper_dev
);
5827 /* Here is the tricky part. We must remove all dev's lower
5828 * devices from all upper_dev's upper devices and vice
5829 * versa, to maintain the graph relationship.
5831 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
)
5832 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
)
5833 __netdev_adjacent_dev_unlink(i
->dev
, j
->dev
);
5835 /* remove also the devices itself from lower/upper device
5838 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
)
5839 __netdev_adjacent_dev_unlink(i
->dev
, upper_dev
);
5841 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
)
5842 __netdev_adjacent_dev_unlink(dev
, i
->dev
);
5844 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER
, dev
,
5845 &changeupper_info
.info
);
5847 EXPORT_SYMBOL(netdev_upper_dev_unlink
);
5850 * netdev_bonding_info_change - Dispatch event about slave change
5852 * @bonding_info: info to dispatch
5854 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
5855 * The caller must hold the RTNL lock.
5857 void netdev_bonding_info_change(struct net_device
*dev
,
5858 struct netdev_bonding_info
*bonding_info
)
5860 struct netdev_notifier_bonding_info info
;
5862 memcpy(&info
.bonding_info
, bonding_info
,
5863 sizeof(struct netdev_bonding_info
));
5864 call_netdevice_notifiers_info(NETDEV_BONDING_INFO
, dev
,
5867 EXPORT_SYMBOL(netdev_bonding_info_change
);
5869 static void netdev_adjacent_add_links(struct net_device
*dev
)
5871 struct netdev_adjacent
*iter
;
5873 struct net
*net
= dev_net(dev
);
5875 list_for_each_entry(iter
, &dev
->adj_list
.upper
, list
) {
5876 if (!net_eq(net
,dev_net(iter
->dev
)))
5878 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
5879 &iter
->dev
->adj_list
.lower
);
5880 netdev_adjacent_sysfs_add(dev
, iter
->dev
,
5881 &dev
->adj_list
.upper
);
5884 list_for_each_entry(iter
, &dev
->adj_list
.lower
, list
) {
5885 if (!net_eq(net
,dev_net(iter
->dev
)))
5887 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
5888 &iter
->dev
->adj_list
.upper
);
5889 netdev_adjacent_sysfs_add(dev
, iter
->dev
,
5890 &dev
->adj_list
.lower
);
5894 static void netdev_adjacent_del_links(struct net_device
*dev
)
5896 struct netdev_adjacent
*iter
;
5898 struct net
*net
= dev_net(dev
);
5900 list_for_each_entry(iter
, &dev
->adj_list
.upper
, list
) {
5901 if (!net_eq(net
,dev_net(iter
->dev
)))
5903 netdev_adjacent_sysfs_del(iter
->dev
, dev
->name
,
5904 &iter
->dev
->adj_list
.lower
);
5905 netdev_adjacent_sysfs_del(dev
, iter
->dev
->name
,
5906 &dev
->adj_list
.upper
);
5909 list_for_each_entry(iter
, &dev
->adj_list
.lower
, list
) {
5910 if (!net_eq(net
,dev_net(iter
->dev
)))
5912 netdev_adjacent_sysfs_del(iter
->dev
, dev
->name
,
5913 &iter
->dev
->adj_list
.upper
);
5914 netdev_adjacent_sysfs_del(dev
, iter
->dev
->name
,
5915 &dev
->adj_list
.lower
);
5919 void netdev_adjacent_rename_links(struct net_device
*dev
, char *oldname
)
5921 struct netdev_adjacent
*iter
;
5923 struct net
*net
= dev_net(dev
);
5925 list_for_each_entry(iter
, &dev
->adj_list
.upper
, list
) {
5926 if (!net_eq(net
,dev_net(iter
->dev
)))
5928 netdev_adjacent_sysfs_del(iter
->dev
, oldname
,
5929 &iter
->dev
->adj_list
.lower
);
5930 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
5931 &iter
->dev
->adj_list
.lower
);
5934 list_for_each_entry(iter
, &dev
->adj_list
.lower
, list
) {
5935 if (!net_eq(net
,dev_net(iter
->dev
)))
5937 netdev_adjacent_sysfs_del(iter
->dev
, oldname
,
5938 &iter
->dev
->adj_list
.upper
);
5939 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
5940 &iter
->dev
->adj_list
.upper
);
5944 void *netdev_lower_dev_get_private(struct net_device
*dev
,
5945 struct net_device
*lower_dev
)
5947 struct netdev_adjacent
*lower
;
5951 lower
= __netdev_find_adj(lower_dev
, &dev
->adj_list
.lower
);
5955 return lower
->private;
5957 EXPORT_SYMBOL(netdev_lower_dev_get_private
);
5960 int dev_get_nest_level(struct net_device
*dev
,
5961 bool (*type_check
)(const struct net_device
*dev
))
5963 struct net_device
*lower
= NULL
;
5964 struct list_head
*iter
;
5970 netdev_for_each_lower_dev(dev
, lower
, iter
) {
5971 nest
= dev_get_nest_level(lower
, type_check
);
5972 if (max_nest
< nest
)
5976 if (type_check(dev
))
5981 EXPORT_SYMBOL(dev_get_nest_level
);
5984 * netdev_lower_change - Dispatch event about lower device state change
5985 * @lower_dev: device
5986 * @lower_state_info: state to dispatch
5988 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
5989 * The caller must hold the RTNL lock.
5991 void netdev_lower_state_changed(struct net_device
*lower_dev
,
5992 void *lower_state_info
)
5994 struct netdev_notifier_changelowerstate_info changelowerstate_info
;
5997 changelowerstate_info
.lower_state_info
= lower_state_info
;
5998 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE
, lower_dev
,
5999 &changelowerstate_info
.info
);
6001 EXPORT_SYMBOL(netdev_lower_state_changed
);
6003 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
6005 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6007 if (ops
->ndo_change_rx_flags
)
6008 ops
->ndo_change_rx_flags(dev
, flags
);
6011 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
, bool notify
)
6013 unsigned int old_flags
= dev
->flags
;
6019 dev
->flags
|= IFF_PROMISC
;
6020 dev
->promiscuity
+= inc
;
6021 if (dev
->promiscuity
== 0) {
6024 * If inc causes overflow, untouch promisc and return error.
6027 dev
->flags
&= ~IFF_PROMISC
;
6029 dev
->promiscuity
-= inc
;
6030 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
6035 if (dev
->flags
!= old_flags
) {
6036 pr_info("device %s %s promiscuous mode\n",
6038 dev
->flags
& IFF_PROMISC
? "entered" : "left");
6039 if (audit_enabled
) {
6040 current_uid_gid(&uid
, &gid
);
6041 audit_log(current
->audit_context
, GFP_ATOMIC
,
6042 AUDIT_ANOM_PROMISCUOUS
,
6043 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
6044 dev
->name
, (dev
->flags
& IFF_PROMISC
),
6045 (old_flags
& IFF_PROMISC
),
6046 from_kuid(&init_user_ns
, audit_get_loginuid(current
)),
6047 from_kuid(&init_user_ns
, uid
),
6048 from_kgid(&init_user_ns
, gid
),
6049 audit_get_sessionid(current
));
6052 dev_change_rx_flags(dev
, IFF_PROMISC
);
6055 __dev_notify_flags(dev
, old_flags
, IFF_PROMISC
);
6060 * dev_set_promiscuity - update promiscuity count on a device
6064 * Add or remove promiscuity from a device. While the count in the device
6065 * remains above zero the interface remains promiscuous. Once it hits zero
6066 * the device reverts back to normal filtering operation. A negative inc
6067 * value is used to drop promiscuity on the device.
6068 * Return 0 if successful or a negative errno code on error.
6070 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
6072 unsigned int old_flags
= dev
->flags
;
6075 err
= __dev_set_promiscuity(dev
, inc
, true);
6078 if (dev
->flags
!= old_flags
)
6079 dev_set_rx_mode(dev
);
6082 EXPORT_SYMBOL(dev_set_promiscuity
);
6084 static int __dev_set_allmulti(struct net_device
*dev
, int inc
, bool notify
)
6086 unsigned int old_flags
= dev
->flags
, old_gflags
= dev
->gflags
;
6090 dev
->flags
|= IFF_ALLMULTI
;
6091 dev
->allmulti
+= inc
;
6092 if (dev
->allmulti
== 0) {
6095 * If inc causes overflow, untouch allmulti and return error.
6098 dev
->flags
&= ~IFF_ALLMULTI
;
6100 dev
->allmulti
-= inc
;
6101 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
6106 if (dev
->flags
^ old_flags
) {
6107 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
6108 dev_set_rx_mode(dev
);
6110 __dev_notify_flags(dev
, old_flags
,
6111 dev
->gflags
^ old_gflags
);
6117 * dev_set_allmulti - update allmulti count on a device
6121 * Add or remove reception of all multicast frames to a device. While the
6122 * count in the device remains above zero the interface remains listening
6123 * to all interfaces. Once it hits zero the device reverts back to normal
6124 * filtering operation. A negative @inc value is used to drop the counter
6125 * when releasing a resource needing all multicasts.
6126 * Return 0 if successful or a negative errno code on error.
6129 int dev_set_allmulti(struct net_device
*dev
, int inc
)
6131 return __dev_set_allmulti(dev
, inc
, true);
6133 EXPORT_SYMBOL(dev_set_allmulti
);
6136 * Upload unicast and multicast address lists to device and
6137 * configure RX filtering. When the device doesn't support unicast
6138 * filtering it is put in promiscuous mode while unicast addresses
6141 void __dev_set_rx_mode(struct net_device
*dev
)
6143 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6145 /* dev_open will call this function so the list will stay sane. */
6146 if (!(dev
->flags
&IFF_UP
))
6149 if (!netif_device_present(dev
))
6152 if (!(dev
->priv_flags
& IFF_UNICAST_FLT
)) {
6153 /* Unicast addresses changes may only happen under the rtnl,
6154 * therefore calling __dev_set_promiscuity here is safe.
6156 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
6157 __dev_set_promiscuity(dev
, 1, false);
6158 dev
->uc_promisc
= true;
6159 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
6160 __dev_set_promiscuity(dev
, -1, false);
6161 dev
->uc_promisc
= false;
6165 if (ops
->ndo_set_rx_mode
)
6166 ops
->ndo_set_rx_mode(dev
);
6169 void dev_set_rx_mode(struct net_device
*dev
)
6171 netif_addr_lock_bh(dev
);
6172 __dev_set_rx_mode(dev
);
6173 netif_addr_unlock_bh(dev
);
6177 * dev_get_flags - get flags reported to userspace
6180 * Get the combination of flag bits exported through APIs to userspace.
6182 unsigned int dev_get_flags(const struct net_device
*dev
)
6186 flags
= (dev
->flags
& ~(IFF_PROMISC
|
6191 (dev
->gflags
& (IFF_PROMISC
|
6194 if (netif_running(dev
)) {
6195 if (netif_oper_up(dev
))
6196 flags
|= IFF_RUNNING
;
6197 if (netif_carrier_ok(dev
))
6198 flags
|= IFF_LOWER_UP
;
6199 if (netif_dormant(dev
))
6200 flags
|= IFF_DORMANT
;
6205 EXPORT_SYMBOL(dev_get_flags
);
6207 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
6209 unsigned int old_flags
= dev
->flags
;
6215 * Set the flags on our device.
6218 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
6219 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
6221 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
6225 * Load in the correct multicast list now the flags have changed.
6228 if ((old_flags
^ flags
) & IFF_MULTICAST
)
6229 dev_change_rx_flags(dev
, IFF_MULTICAST
);
6231 dev_set_rx_mode(dev
);
6234 * Have we downed the interface. We handle IFF_UP ourselves
6235 * according to user attempts to set it, rather than blindly
6240 if ((old_flags
^ flags
) & IFF_UP
)
6241 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
6243 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
6244 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
6245 unsigned int old_flags
= dev
->flags
;
6247 dev
->gflags
^= IFF_PROMISC
;
6249 if (__dev_set_promiscuity(dev
, inc
, false) >= 0)
6250 if (dev
->flags
!= old_flags
)
6251 dev_set_rx_mode(dev
);
6254 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
6255 is important. Some (broken) drivers set IFF_PROMISC, when
6256 IFF_ALLMULTI is requested not asking us and not reporting.
6258 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
6259 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
6261 dev
->gflags
^= IFF_ALLMULTI
;
6262 __dev_set_allmulti(dev
, inc
, false);
6268 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
,
6269 unsigned int gchanges
)
6271 unsigned int changes
= dev
->flags
^ old_flags
;
6274 rtmsg_ifinfo(RTM_NEWLINK
, dev
, gchanges
, GFP_ATOMIC
);
6276 if (changes
& IFF_UP
) {
6277 if (dev
->flags
& IFF_UP
)
6278 call_netdevice_notifiers(NETDEV_UP
, dev
);
6280 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
6283 if (dev
->flags
& IFF_UP
&&
6284 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
))) {
6285 struct netdev_notifier_change_info change_info
;
6287 change_info
.flags_changed
= changes
;
6288 call_netdevice_notifiers_info(NETDEV_CHANGE
, dev
,
6294 * dev_change_flags - change device settings
6296 * @flags: device state flags
6298 * Change settings on device based state flags. The flags are
6299 * in the userspace exported format.
6301 int dev_change_flags(struct net_device
*dev
, unsigned int flags
)
6304 unsigned int changes
, old_flags
= dev
->flags
, old_gflags
= dev
->gflags
;
6306 ret
= __dev_change_flags(dev
, flags
);
6310 changes
= (old_flags
^ dev
->flags
) | (old_gflags
^ dev
->gflags
);
6311 __dev_notify_flags(dev
, old_flags
, changes
);
6314 EXPORT_SYMBOL(dev_change_flags
);
6316 static int __dev_set_mtu(struct net_device
*dev
, int new_mtu
)
6318 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6320 if (ops
->ndo_change_mtu
)
6321 return ops
->ndo_change_mtu(dev
, new_mtu
);
6328 * dev_set_mtu - Change maximum transfer unit
6330 * @new_mtu: new transfer unit
6332 * Change the maximum transfer size of the network device.
6334 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
6338 if (new_mtu
== dev
->mtu
)
6341 /* MTU must be positive. */
6345 if (!netif_device_present(dev
))
6348 err
= call_netdevice_notifiers(NETDEV_PRECHANGEMTU
, dev
);
6349 err
= notifier_to_errno(err
);
6353 orig_mtu
= dev
->mtu
;
6354 err
= __dev_set_mtu(dev
, new_mtu
);
6357 err
= call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
6358 err
= notifier_to_errno(err
);
6360 /* setting mtu back and notifying everyone again,
6361 * so that they have a chance to revert changes.
6363 __dev_set_mtu(dev
, orig_mtu
);
6364 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
6369 EXPORT_SYMBOL(dev_set_mtu
);
6372 * dev_set_group - Change group this device belongs to
6374 * @new_group: group this device should belong to
6376 void dev_set_group(struct net_device
*dev
, int new_group
)
6378 dev
->group
= new_group
;
6380 EXPORT_SYMBOL(dev_set_group
);
6383 * dev_set_mac_address - Change Media Access Control Address
6387 * Change the hardware (MAC) address of the device
6389 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
6391 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6394 if (!ops
->ndo_set_mac_address
)
6396 if (sa
->sa_family
!= dev
->type
)
6398 if (!netif_device_present(dev
))
6400 err
= ops
->ndo_set_mac_address(dev
, sa
);
6403 dev
->addr_assign_type
= NET_ADDR_SET
;
6404 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
6405 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
6408 EXPORT_SYMBOL(dev_set_mac_address
);
6411 * dev_change_carrier - Change device carrier
6413 * @new_carrier: new value
6415 * Change device carrier
6417 int dev_change_carrier(struct net_device
*dev
, bool new_carrier
)
6419 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6421 if (!ops
->ndo_change_carrier
)
6423 if (!netif_device_present(dev
))
6425 return ops
->ndo_change_carrier(dev
, new_carrier
);
6427 EXPORT_SYMBOL(dev_change_carrier
);
6430 * dev_get_phys_port_id - Get device physical port ID
6434 * Get device physical port ID
6436 int dev_get_phys_port_id(struct net_device
*dev
,
6437 struct netdev_phys_item_id
*ppid
)
6439 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6441 if (!ops
->ndo_get_phys_port_id
)
6443 return ops
->ndo_get_phys_port_id(dev
, ppid
);
6445 EXPORT_SYMBOL(dev_get_phys_port_id
);
6448 * dev_get_phys_port_name - Get device physical port name
6451 * @len: limit of bytes to copy to name
6453 * Get device physical port name
6455 int dev_get_phys_port_name(struct net_device
*dev
,
6456 char *name
, size_t len
)
6458 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6460 if (!ops
->ndo_get_phys_port_name
)
6462 return ops
->ndo_get_phys_port_name(dev
, name
, len
);
6464 EXPORT_SYMBOL(dev_get_phys_port_name
);
6467 * dev_change_proto_down - update protocol port state information
6469 * @proto_down: new value
6471 * This info can be used by switch drivers to set the phys state of the
6474 int dev_change_proto_down(struct net_device
*dev
, bool proto_down
)
6476 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6478 if (!ops
->ndo_change_proto_down
)
6480 if (!netif_device_present(dev
))
6482 return ops
->ndo_change_proto_down(dev
, proto_down
);
6484 EXPORT_SYMBOL(dev_change_proto_down
);
6487 * dev_new_index - allocate an ifindex
6488 * @net: the applicable net namespace
6490 * Returns a suitable unique value for a new device interface
6491 * number. The caller must hold the rtnl semaphore or the
6492 * dev_base_lock to be sure it remains unique.
6494 static int dev_new_index(struct net
*net
)
6496 int ifindex
= net
->ifindex
;
6500 if (!__dev_get_by_index(net
, ifindex
))
6501 return net
->ifindex
= ifindex
;
6505 /* Delayed registration/unregisteration */
6506 static LIST_HEAD(net_todo_list
);
6507 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq
);
6509 static void net_set_todo(struct net_device
*dev
)
6511 list_add_tail(&dev
->todo_list
, &net_todo_list
);
6512 dev_net(dev
)->dev_unreg_count
++;
6515 static void rollback_registered_many(struct list_head
*head
)
6517 struct net_device
*dev
, *tmp
;
6518 LIST_HEAD(close_head
);
6520 BUG_ON(dev_boot_phase
);
6523 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
6524 /* Some devices call without registering
6525 * for initialization unwind. Remove those
6526 * devices and proceed with the remaining.
6528 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
6529 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
6533 list_del(&dev
->unreg_list
);
6536 dev
->dismantle
= true;
6537 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
6540 /* If device is running, close it first. */
6541 list_for_each_entry(dev
, head
, unreg_list
)
6542 list_add_tail(&dev
->close_list
, &close_head
);
6543 dev_close_many(&close_head
, true);
6545 list_for_each_entry(dev
, head
, unreg_list
) {
6546 /* And unlink it from device chain. */
6547 unlist_netdevice(dev
);
6549 dev
->reg_state
= NETREG_UNREGISTERING
;
6550 on_each_cpu(flush_backlog
, dev
, 1);
6555 list_for_each_entry(dev
, head
, unreg_list
) {
6556 struct sk_buff
*skb
= NULL
;
6558 /* Shutdown queueing discipline. */
6562 /* Notify protocols, that we are about to destroy
6563 this device. They should clean all the things.
6565 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6567 if (!dev
->rtnl_link_ops
||
6568 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
6569 skb
= rtmsg_ifinfo_build_skb(RTM_DELLINK
, dev
, ~0U,
6573 * Flush the unicast and multicast chains
6578 if (dev
->netdev_ops
->ndo_uninit
)
6579 dev
->netdev_ops
->ndo_uninit(dev
);
6582 rtmsg_ifinfo_send(skb
, dev
, GFP_KERNEL
);
6584 /* Notifier chain MUST detach us all upper devices. */
6585 WARN_ON(netdev_has_any_upper_dev(dev
));
6587 /* Remove entries from kobject tree */
6588 netdev_unregister_kobject(dev
);
6590 /* Remove XPS queueing entries */
6591 netif_reset_xps_queues_gt(dev
, 0);
6597 list_for_each_entry(dev
, head
, unreg_list
)
6601 static void rollback_registered(struct net_device
*dev
)
6605 list_add(&dev
->unreg_list
, &single
);
6606 rollback_registered_many(&single
);
6610 static netdev_features_t
netdev_sync_upper_features(struct net_device
*lower
,
6611 struct net_device
*upper
, netdev_features_t features
)
6613 netdev_features_t upper_disables
= NETIF_F_UPPER_DISABLES
;
6614 netdev_features_t feature
;
6617 for_each_netdev_feature(&upper_disables
, feature_bit
) {
6618 feature
= __NETIF_F_BIT(feature_bit
);
6619 if (!(upper
->wanted_features
& feature
)
6620 && (features
& feature
)) {
6621 netdev_dbg(lower
, "Dropping feature %pNF, upper dev %s has it off.\n",
6622 &feature
, upper
->name
);
6623 features
&= ~feature
;
6630 static void netdev_sync_lower_features(struct net_device
*upper
,
6631 struct net_device
*lower
, netdev_features_t features
)
6633 netdev_features_t upper_disables
= NETIF_F_UPPER_DISABLES
;
6634 netdev_features_t feature
;
6637 for_each_netdev_feature(&upper_disables
, feature_bit
) {
6638 feature
= __NETIF_F_BIT(feature_bit
);
6639 if (!(features
& feature
) && (lower
->features
& feature
)) {
6640 netdev_dbg(upper
, "Disabling feature %pNF on lower dev %s.\n",
6641 &feature
, lower
->name
);
6642 lower
->wanted_features
&= ~feature
;
6643 netdev_update_features(lower
);
6645 if (unlikely(lower
->features
& feature
))
6646 netdev_WARN(upper
, "failed to disable %pNF on %s!\n",
6647 &feature
, lower
->name
);
6652 static netdev_features_t
netdev_fix_features(struct net_device
*dev
,
6653 netdev_features_t features
)
6655 /* Fix illegal checksum combinations */
6656 if ((features
& NETIF_F_HW_CSUM
) &&
6657 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
6658 netdev_warn(dev
, "mixed HW and IP checksum settings.\n");
6659 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
6662 /* TSO requires that SG is present as well. */
6663 if ((features
& NETIF_F_ALL_TSO
) && !(features
& NETIF_F_SG
)) {
6664 netdev_dbg(dev
, "Dropping TSO features since no SG feature.\n");
6665 features
&= ~NETIF_F_ALL_TSO
;
6668 if ((features
& NETIF_F_TSO
) && !(features
& NETIF_F_HW_CSUM
) &&
6669 !(features
& NETIF_F_IP_CSUM
)) {
6670 netdev_dbg(dev
, "Dropping TSO features since no CSUM feature.\n");
6671 features
&= ~NETIF_F_TSO
;
6672 features
&= ~NETIF_F_TSO_ECN
;
6675 if ((features
& NETIF_F_TSO6
) && !(features
& NETIF_F_HW_CSUM
) &&
6676 !(features
& NETIF_F_IPV6_CSUM
)) {
6677 netdev_dbg(dev
, "Dropping TSO6 features since no CSUM feature.\n");
6678 features
&= ~NETIF_F_TSO6
;
6681 /* TSO ECN requires that TSO is present as well. */
6682 if ((features
& NETIF_F_ALL_TSO
) == NETIF_F_TSO_ECN
)
6683 features
&= ~NETIF_F_TSO_ECN
;
6685 /* Software GSO depends on SG. */
6686 if ((features
& NETIF_F_GSO
) && !(features
& NETIF_F_SG
)) {
6687 netdev_dbg(dev
, "Dropping NETIF_F_GSO since no SG feature.\n");
6688 features
&= ~NETIF_F_GSO
;
6691 /* UFO needs SG and checksumming */
6692 if (features
& NETIF_F_UFO
) {
6693 /* maybe split UFO into V4 and V6? */
6694 if (!(features
& NETIF_F_HW_CSUM
) &&
6695 ((features
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
)) !=
6696 (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
))) {
6698 "Dropping NETIF_F_UFO since no checksum offload features.\n");
6699 features
&= ~NETIF_F_UFO
;
6702 if (!(features
& NETIF_F_SG
)) {
6704 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
6705 features
&= ~NETIF_F_UFO
;
6709 #ifdef CONFIG_NET_RX_BUSY_POLL
6710 if (dev
->netdev_ops
->ndo_busy_poll
)
6711 features
|= NETIF_F_BUSY_POLL
;
6714 features
&= ~NETIF_F_BUSY_POLL
;
6719 int __netdev_update_features(struct net_device
*dev
)
6721 struct net_device
*upper
, *lower
;
6722 netdev_features_t features
;
6723 struct list_head
*iter
;
6728 features
= netdev_get_wanted_features(dev
);
6730 if (dev
->netdev_ops
->ndo_fix_features
)
6731 features
= dev
->netdev_ops
->ndo_fix_features(dev
, features
);
6733 /* driver might be less strict about feature dependencies */
6734 features
= netdev_fix_features(dev
, features
);
6736 /* some features can't be enabled if they're off an an upper device */
6737 netdev_for_each_upper_dev_rcu(dev
, upper
, iter
)
6738 features
= netdev_sync_upper_features(dev
, upper
, features
);
6740 if (dev
->features
== features
)
6743 netdev_dbg(dev
, "Features changed: %pNF -> %pNF\n",
6744 &dev
->features
, &features
);
6746 if (dev
->netdev_ops
->ndo_set_features
)
6747 err
= dev
->netdev_ops
->ndo_set_features(dev
, features
);
6751 if (unlikely(err
< 0)) {
6753 "set_features() failed (%d); wanted %pNF, left %pNF\n",
6754 err
, &features
, &dev
->features
);
6755 /* return non-0 since some features might have changed and
6756 * it's better to fire a spurious notification than miss it
6762 /* some features must be disabled on lower devices when disabled
6763 * on an upper device (think: bonding master or bridge)
6765 netdev_for_each_lower_dev(dev
, lower
, iter
)
6766 netdev_sync_lower_features(dev
, lower
, features
);
6769 dev
->features
= features
;
6771 return err
< 0 ? 0 : 1;
6775 * netdev_update_features - recalculate device features
6776 * @dev: the device to check
6778 * Recalculate dev->features set and send notifications if it
6779 * has changed. Should be called after driver or hardware dependent
6780 * conditions might have changed that influence the features.
6782 void netdev_update_features(struct net_device
*dev
)
6784 if (__netdev_update_features(dev
))
6785 netdev_features_change(dev
);
6787 EXPORT_SYMBOL(netdev_update_features
);
6790 * netdev_change_features - recalculate device features
6791 * @dev: the device to check
6793 * Recalculate dev->features set and send notifications even
6794 * if they have not changed. Should be called instead of
6795 * netdev_update_features() if also dev->vlan_features might
6796 * have changed to allow the changes to be propagated to stacked
6799 void netdev_change_features(struct net_device
*dev
)
6801 __netdev_update_features(dev
);
6802 netdev_features_change(dev
);
6804 EXPORT_SYMBOL(netdev_change_features
);
6807 * netif_stacked_transfer_operstate - transfer operstate
6808 * @rootdev: the root or lower level device to transfer state from
6809 * @dev: the device to transfer operstate to
6811 * Transfer operational state from root to device. This is normally
6812 * called when a stacking relationship exists between the root
6813 * device and the device(a leaf device).
6815 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
6816 struct net_device
*dev
)
6818 if (rootdev
->operstate
== IF_OPER_DORMANT
)
6819 netif_dormant_on(dev
);
6821 netif_dormant_off(dev
);
6823 if (netif_carrier_ok(rootdev
)) {
6824 if (!netif_carrier_ok(dev
))
6825 netif_carrier_on(dev
);
6827 if (netif_carrier_ok(dev
))
6828 netif_carrier_off(dev
);
6831 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
6834 static int netif_alloc_rx_queues(struct net_device
*dev
)
6836 unsigned int i
, count
= dev
->num_rx_queues
;
6837 struct netdev_rx_queue
*rx
;
6838 size_t sz
= count
* sizeof(*rx
);
6842 rx
= kzalloc(sz
, GFP_KERNEL
| __GFP_NOWARN
| __GFP_REPEAT
);
6850 for (i
= 0; i
< count
; i
++)
6856 static void netdev_init_one_queue(struct net_device
*dev
,
6857 struct netdev_queue
*queue
, void *_unused
)
6859 /* Initialize queue lock */
6860 spin_lock_init(&queue
->_xmit_lock
);
6861 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
6862 queue
->xmit_lock_owner
= -1;
6863 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
6866 dql_init(&queue
->dql
, HZ
);
6870 static void netif_free_tx_queues(struct net_device
*dev
)
6875 static int netif_alloc_netdev_queues(struct net_device
*dev
)
6877 unsigned int count
= dev
->num_tx_queues
;
6878 struct netdev_queue
*tx
;
6879 size_t sz
= count
* sizeof(*tx
);
6881 if (count
< 1 || count
> 0xffff)
6884 tx
= kzalloc(sz
, GFP_KERNEL
| __GFP_NOWARN
| __GFP_REPEAT
);
6892 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
6893 spin_lock_init(&dev
->tx_global_lock
);
6898 void netif_tx_stop_all_queues(struct net_device
*dev
)
6902 for (i
= 0; i
< dev
->num_tx_queues
; i
++) {
6903 struct netdev_queue
*txq
= netdev_get_tx_queue(dev
, i
);
6904 netif_tx_stop_queue(txq
);
6907 EXPORT_SYMBOL(netif_tx_stop_all_queues
);
6910 * register_netdevice - register a network device
6911 * @dev: device to register
6913 * Take a completed network device structure and add it to the kernel
6914 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6915 * chain. 0 is returned on success. A negative errno code is returned
6916 * on a failure to set up the device, or if the name is a duplicate.
6918 * Callers must hold the rtnl semaphore. You may want
6919 * register_netdev() instead of this.
6922 * The locking appears insufficient to guarantee two parallel registers
6923 * will not get the same name.
6926 int register_netdevice(struct net_device
*dev
)
6929 struct net
*net
= dev_net(dev
);
6931 BUG_ON(dev_boot_phase
);
6936 /* When net_device's are persistent, this will be fatal. */
6937 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
6940 spin_lock_init(&dev
->addr_list_lock
);
6941 netdev_set_addr_lockdep_class(dev
);
6943 ret
= dev_get_valid_name(net
, dev
, dev
->name
);
6947 /* Init, if this function is available */
6948 if (dev
->netdev_ops
->ndo_init
) {
6949 ret
= dev
->netdev_ops
->ndo_init(dev
);
6957 if (((dev
->hw_features
| dev
->features
) &
6958 NETIF_F_HW_VLAN_CTAG_FILTER
) &&
6959 (!dev
->netdev_ops
->ndo_vlan_rx_add_vid
||
6960 !dev
->netdev_ops
->ndo_vlan_rx_kill_vid
)) {
6961 netdev_WARN(dev
, "Buggy VLAN acceleration in driver!\n");
6968 dev
->ifindex
= dev_new_index(net
);
6969 else if (__dev_get_by_index(net
, dev
->ifindex
))
6972 /* Transfer changeable features to wanted_features and enable
6973 * software offloads (GSO and GRO).
6975 dev
->hw_features
|= NETIF_F_SOFT_FEATURES
;
6976 dev
->features
|= NETIF_F_SOFT_FEATURES
;
6977 dev
->wanted_features
= dev
->features
& dev
->hw_features
;
6979 if (!(dev
->flags
& IFF_LOOPBACK
)) {
6980 dev
->hw_features
|= NETIF_F_NOCACHE_COPY
;
6983 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6985 dev
->vlan_features
|= NETIF_F_HIGHDMA
;
6987 /* Make NETIF_F_SG inheritable to tunnel devices.
6989 dev
->hw_enc_features
|= NETIF_F_SG
;
6991 /* Make NETIF_F_SG inheritable to MPLS.
6993 dev
->mpls_features
|= NETIF_F_SG
;
6995 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
6996 ret
= notifier_to_errno(ret
);
7000 ret
= netdev_register_kobject(dev
);
7003 dev
->reg_state
= NETREG_REGISTERED
;
7005 __netdev_update_features(dev
);
7008 * Default initial state at registry is that the
7009 * device is present.
7012 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
7014 linkwatch_init_dev(dev
);
7016 dev_init_scheduler(dev
);
7018 list_netdevice(dev
);
7019 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
7021 /* If the device has permanent device address, driver should
7022 * set dev_addr and also addr_assign_type should be set to
7023 * NET_ADDR_PERM (default value).
7025 if (dev
->addr_assign_type
== NET_ADDR_PERM
)
7026 memcpy(dev
->perm_addr
, dev
->dev_addr
, dev
->addr_len
);
7028 /* Notify protocols, that a new device appeared. */
7029 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
7030 ret
= notifier_to_errno(ret
);
7032 rollback_registered(dev
);
7033 dev
->reg_state
= NETREG_UNREGISTERED
;
7036 * Prevent userspace races by waiting until the network
7037 * device is fully setup before sending notifications.
7039 if (!dev
->rtnl_link_ops
||
7040 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
7041 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U, GFP_KERNEL
);
7047 if (dev
->netdev_ops
->ndo_uninit
)
7048 dev
->netdev_ops
->ndo_uninit(dev
);
7051 EXPORT_SYMBOL(register_netdevice
);
7054 * init_dummy_netdev - init a dummy network device for NAPI
7055 * @dev: device to init
7057 * This takes a network device structure and initialize the minimum
7058 * amount of fields so it can be used to schedule NAPI polls without
7059 * registering a full blown interface. This is to be used by drivers
7060 * that need to tie several hardware interfaces to a single NAPI
7061 * poll scheduler due to HW limitations.
7063 int init_dummy_netdev(struct net_device
*dev
)
7065 /* Clear everything. Note we don't initialize spinlocks
7066 * are they aren't supposed to be taken by any of the
7067 * NAPI code and this dummy netdev is supposed to be
7068 * only ever used for NAPI polls
7070 memset(dev
, 0, sizeof(struct net_device
));
7072 /* make sure we BUG if trying to hit standard
7073 * register/unregister code path
7075 dev
->reg_state
= NETREG_DUMMY
;
7077 /* NAPI wants this */
7078 INIT_LIST_HEAD(&dev
->napi_list
);
7080 /* a dummy interface is started by default */
7081 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
7082 set_bit(__LINK_STATE_START
, &dev
->state
);
7084 /* Note : We dont allocate pcpu_refcnt for dummy devices,
7085 * because users of this 'device' dont need to change
7091 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
7095 * register_netdev - register a network device
7096 * @dev: device to register
7098 * Take a completed network device structure and add it to the kernel
7099 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
7100 * chain. 0 is returned on success. A negative errno code is returned
7101 * on a failure to set up the device, or if the name is a duplicate.
7103 * This is a wrapper around register_netdevice that takes the rtnl semaphore
7104 * and expands the device name if you passed a format string to
7107 int register_netdev(struct net_device
*dev
)
7112 err
= register_netdevice(dev
);
7116 EXPORT_SYMBOL(register_netdev
);
7118 int netdev_refcnt_read(const struct net_device
*dev
)
7122 for_each_possible_cpu(i
)
7123 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
7126 EXPORT_SYMBOL(netdev_refcnt_read
);
7129 * netdev_wait_allrefs - wait until all references are gone.
7130 * @dev: target net_device
7132 * This is called when unregistering network devices.
7134 * Any protocol or device that holds a reference should register
7135 * for netdevice notification, and cleanup and put back the
7136 * reference if they receive an UNREGISTER event.
7137 * We can get stuck here if buggy protocols don't correctly
7140 static void netdev_wait_allrefs(struct net_device
*dev
)
7142 unsigned long rebroadcast_time
, warning_time
;
7145 linkwatch_forget_dev(dev
);
7147 rebroadcast_time
= warning_time
= jiffies
;
7148 refcnt
= netdev_refcnt_read(dev
);
7150 while (refcnt
!= 0) {
7151 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
7154 /* Rebroadcast unregister notification */
7155 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
7161 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
7162 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
7164 /* We must not have linkwatch events
7165 * pending on unregister. If this
7166 * happens, we simply run the queue
7167 * unscheduled, resulting in a noop
7170 linkwatch_run_queue();
7175 rebroadcast_time
= jiffies
;
7180 refcnt
= netdev_refcnt_read(dev
);
7182 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
7183 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
7185 warning_time
= jiffies
;
7194 * register_netdevice(x1);
7195 * register_netdevice(x2);
7197 * unregister_netdevice(y1);
7198 * unregister_netdevice(y2);
7204 * We are invoked by rtnl_unlock().
7205 * This allows us to deal with problems:
7206 * 1) We can delete sysfs objects which invoke hotplug
7207 * without deadlocking with linkwatch via keventd.
7208 * 2) Since we run with the RTNL semaphore not held, we can sleep
7209 * safely in order to wait for the netdev refcnt to drop to zero.
7211 * We must not return until all unregister events added during
7212 * the interval the lock was held have been completed.
7214 void netdev_run_todo(void)
7216 struct list_head list
;
7218 /* Snapshot list, allow later requests */
7219 list_replace_init(&net_todo_list
, &list
);
7224 /* Wait for rcu callbacks to finish before next phase */
7225 if (!list_empty(&list
))
7228 while (!list_empty(&list
)) {
7229 struct net_device
*dev
7230 = list_first_entry(&list
, struct net_device
, todo_list
);
7231 list_del(&dev
->todo_list
);
7234 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
7237 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
7238 pr_err("network todo '%s' but state %d\n",
7239 dev
->name
, dev
->reg_state
);
7244 dev
->reg_state
= NETREG_UNREGISTERED
;
7246 netdev_wait_allrefs(dev
);
7249 BUG_ON(netdev_refcnt_read(dev
));
7250 BUG_ON(!list_empty(&dev
->ptype_all
));
7251 BUG_ON(!list_empty(&dev
->ptype_specific
));
7252 WARN_ON(rcu_access_pointer(dev
->ip_ptr
));
7253 WARN_ON(rcu_access_pointer(dev
->ip6_ptr
));
7254 WARN_ON(dev
->dn_ptr
);
7256 if (dev
->destructor
)
7257 dev
->destructor(dev
);
7259 /* Report a network device has been unregistered */
7261 dev_net(dev
)->dev_unreg_count
--;
7263 wake_up(&netdev_unregistering_wq
);
7265 /* Free network device */
7266 kobject_put(&dev
->dev
.kobj
);
7270 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
7271 * all the same fields in the same order as net_device_stats, with only
7272 * the type differing, but rtnl_link_stats64 may have additional fields
7273 * at the end for newer counters.
7275 void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
7276 const struct net_device_stats
*netdev_stats
)
7278 #if BITS_PER_LONG == 64
7279 BUILD_BUG_ON(sizeof(*stats64
) < sizeof(*netdev_stats
));
7280 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
7281 /* zero out counters that only exist in rtnl_link_stats64 */
7282 memset((char *)stats64
+ sizeof(*netdev_stats
), 0,
7283 sizeof(*stats64
) - sizeof(*netdev_stats
));
7285 size_t i
, n
= sizeof(*netdev_stats
) / sizeof(unsigned long);
7286 const unsigned long *src
= (const unsigned long *)netdev_stats
;
7287 u64
*dst
= (u64
*)stats64
;
7289 BUILD_BUG_ON(n
> sizeof(*stats64
) / sizeof(u64
));
7290 for (i
= 0; i
< n
; i
++)
7292 /* zero out counters that only exist in rtnl_link_stats64 */
7293 memset((char *)stats64
+ n
* sizeof(u64
), 0,
7294 sizeof(*stats64
) - n
* sizeof(u64
));
7297 EXPORT_SYMBOL(netdev_stats_to_stats64
);
7300 * dev_get_stats - get network device statistics
7301 * @dev: device to get statistics from
7302 * @storage: place to store stats
7304 * Get network statistics from device. Return @storage.
7305 * The device driver may provide its own method by setting
7306 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
7307 * otherwise the internal statistics structure is used.
7309 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
7310 struct rtnl_link_stats64
*storage
)
7312 const struct net_device_ops
*ops
= dev
->netdev_ops
;
7314 if (ops
->ndo_get_stats64
) {
7315 memset(storage
, 0, sizeof(*storage
));
7316 ops
->ndo_get_stats64(dev
, storage
);
7317 } else if (ops
->ndo_get_stats
) {
7318 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
7320 netdev_stats_to_stats64(storage
, &dev
->stats
);
7322 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
7323 storage
->tx_dropped
+= atomic_long_read(&dev
->tx_dropped
);
7324 storage
->rx_nohandler
+= atomic_long_read(&dev
->rx_nohandler
);
7327 EXPORT_SYMBOL(dev_get_stats
);
7329 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
7331 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
7333 #ifdef CONFIG_NET_CLS_ACT
7336 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
7339 netdev_init_one_queue(dev
, queue
, NULL
);
7340 RCU_INIT_POINTER(queue
->qdisc
, &noop_qdisc
);
7341 queue
->qdisc_sleeping
= &noop_qdisc
;
7342 rcu_assign_pointer(dev
->ingress_queue
, queue
);
7347 static const struct ethtool_ops default_ethtool_ops
;
7349 void netdev_set_default_ethtool_ops(struct net_device
*dev
,
7350 const struct ethtool_ops
*ops
)
7352 if (dev
->ethtool_ops
== &default_ethtool_ops
)
7353 dev
->ethtool_ops
= ops
;
7355 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops
);
7357 void netdev_freemem(struct net_device
*dev
)
7359 char *addr
= (char *)dev
- dev
->padded
;
7365 * alloc_netdev_mqs - allocate network device
7366 * @sizeof_priv: size of private data to allocate space for
7367 * @name: device name format string
7368 * @name_assign_type: origin of device name
7369 * @setup: callback to initialize device
7370 * @txqs: the number of TX subqueues to allocate
7371 * @rxqs: the number of RX subqueues to allocate
7373 * Allocates a struct net_device with private data area for driver use
7374 * and performs basic initialization. Also allocates subqueue structs
7375 * for each queue on the device.
7377 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
7378 unsigned char name_assign_type
,
7379 void (*setup
)(struct net_device
*),
7380 unsigned int txqs
, unsigned int rxqs
)
7382 struct net_device
*dev
;
7384 struct net_device
*p
;
7386 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
7389 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
7395 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
7400 alloc_size
= sizeof(struct net_device
);
7402 /* ensure 32-byte alignment of private area */
7403 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
7404 alloc_size
+= sizeof_priv
;
7406 /* ensure 32-byte alignment of whole construct */
7407 alloc_size
+= NETDEV_ALIGN
- 1;
7409 p
= kzalloc(alloc_size
, GFP_KERNEL
| __GFP_NOWARN
| __GFP_REPEAT
);
7411 p
= vzalloc(alloc_size
);
7415 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
7416 dev
->padded
= (char *)dev
- (char *)p
;
7418 dev
->pcpu_refcnt
= alloc_percpu(int);
7419 if (!dev
->pcpu_refcnt
)
7422 if (dev_addr_init(dev
))
7428 dev_net_set(dev
, &init_net
);
7430 dev
->gso_max_size
= GSO_MAX_SIZE
;
7431 dev
->gso_max_segs
= GSO_MAX_SEGS
;
7432 dev
->gso_min_segs
= 0;
7434 INIT_LIST_HEAD(&dev
->napi_list
);
7435 INIT_LIST_HEAD(&dev
->unreg_list
);
7436 INIT_LIST_HEAD(&dev
->close_list
);
7437 INIT_LIST_HEAD(&dev
->link_watch_list
);
7438 INIT_LIST_HEAD(&dev
->adj_list
.upper
);
7439 INIT_LIST_HEAD(&dev
->adj_list
.lower
);
7440 INIT_LIST_HEAD(&dev
->all_adj_list
.upper
);
7441 INIT_LIST_HEAD(&dev
->all_adj_list
.lower
);
7442 INIT_LIST_HEAD(&dev
->ptype_all
);
7443 INIT_LIST_HEAD(&dev
->ptype_specific
);
7444 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
| IFF_XMIT_DST_RELEASE_PERM
;
7447 if (!dev
->tx_queue_len
) {
7448 dev
->priv_flags
|= IFF_NO_QUEUE
;
7449 dev
->tx_queue_len
= 1;
7452 dev
->num_tx_queues
= txqs
;
7453 dev
->real_num_tx_queues
= txqs
;
7454 if (netif_alloc_netdev_queues(dev
))
7458 dev
->num_rx_queues
= rxqs
;
7459 dev
->real_num_rx_queues
= rxqs
;
7460 if (netif_alloc_rx_queues(dev
))
7464 strcpy(dev
->name
, name
);
7465 dev
->name_assign_type
= name_assign_type
;
7466 dev
->group
= INIT_NETDEV_GROUP
;
7467 if (!dev
->ethtool_ops
)
7468 dev
->ethtool_ops
= &default_ethtool_ops
;
7470 nf_hook_ingress_init(dev
);
7479 free_percpu(dev
->pcpu_refcnt
);
7481 netdev_freemem(dev
);
7484 EXPORT_SYMBOL(alloc_netdev_mqs
);
7487 * free_netdev - free network device
7490 * This function does the last stage of destroying an allocated device
7491 * interface. The reference to the device object is released.
7492 * If this is the last reference then it will be freed.
7493 * Must be called in process context.
7495 void free_netdev(struct net_device
*dev
)
7497 struct napi_struct
*p
, *n
;
7500 netif_free_tx_queues(dev
);
7505 kfree(rcu_dereference_protected(dev
->ingress_queue
, 1));
7507 /* Flush device addresses */
7508 dev_addr_flush(dev
);
7510 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
7513 free_percpu(dev
->pcpu_refcnt
);
7514 dev
->pcpu_refcnt
= NULL
;
7516 /* Compatibility with error handling in drivers */
7517 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
7518 netdev_freemem(dev
);
7522 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
7523 dev
->reg_state
= NETREG_RELEASED
;
7525 /* will free via device release */
7526 put_device(&dev
->dev
);
7528 EXPORT_SYMBOL(free_netdev
);
7531 * synchronize_net - Synchronize with packet receive processing
7533 * Wait for packets currently being received to be done.
7534 * Does not block later packets from starting.
7536 void synchronize_net(void)
7539 if (rtnl_is_locked())
7540 synchronize_rcu_expedited();
7544 EXPORT_SYMBOL(synchronize_net
);
7547 * unregister_netdevice_queue - remove device from the kernel
7551 * This function shuts down a device interface and removes it
7552 * from the kernel tables.
7553 * If head not NULL, device is queued to be unregistered later.
7555 * Callers must hold the rtnl semaphore. You may want
7556 * unregister_netdev() instead of this.
7559 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
7564 list_move_tail(&dev
->unreg_list
, head
);
7566 rollback_registered(dev
);
7567 /* Finish processing unregister after unlock */
7571 EXPORT_SYMBOL(unregister_netdevice_queue
);
7574 * unregister_netdevice_many - unregister many devices
7575 * @head: list of devices
7577 * Note: As most callers use a stack allocated list_head,
7578 * we force a list_del() to make sure stack wont be corrupted later.
7580 void unregister_netdevice_many(struct list_head
*head
)
7582 struct net_device
*dev
;
7584 if (!list_empty(head
)) {
7585 rollback_registered_many(head
);
7586 list_for_each_entry(dev
, head
, unreg_list
)
7591 EXPORT_SYMBOL(unregister_netdevice_many
);
7594 * unregister_netdev - remove device from the kernel
7597 * This function shuts down a device interface and removes it
7598 * from the kernel tables.
7600 * This is just a wrapper for unregister_netdevice that takes
7601 * the rtnl semaphore. In general you want to use this and not
7602 * unregister_netdevice.
7604 void unregister_netdev(struct net_device
*dev
)
7607 unregister_netdevice(dev
);
7610 EXPORT_SYMBOL(unregister_netdev
);
7613 * dev_change_net_namespace - move device to different nethost namespace
7615 * @net: network namespace
7616 * @pat: If not NULL name pattern to try if the current device name
7617 * is already taken in the destination network namespace.
7619 * This function shuts down a device interface and moves it
7620 * to a new network namespace. On success 0 is returned, on
7621 * a failure a netagive errno code is returned.
7623 * Callers must hold the rtnl semaphore.
7626 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
7632 /* Don't allow namespace local devices to be moved. */
7634 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
7637 /* Ensure the device has been registrered */
7638 if (dev
->reg_state
!= NETREG_REGISTERED
)
7641 /* Get out if there is nothing todo */
7643 if (net_eq(dev_net(dev
), net
))
7646 /* Pick the destination device name, and ensure
7647 * we can use it in the destination network namespace.
7650 if (__dev_get_by_name(net
, dev
->name
)) {
7651 /* We get here if we can't use the current device name */
7654 if (dev_get_valid_name(net
, dev
, pat
) < 0)
7659 * And now a mini version of register_netdevice unregister_netdevice.
7662 /* If device is running close it first. */
7665 /* And unlink it from device chain */
7667 unlist_netdevice(dev
);
7671 /* Shutdown queueing discipline. */
7674 /* Notify protocols, that we are about to destroy
7675 this device. They should clean all the things.
7677 Note that dev->reg_state stays at NETREG_REGISTERED.
7678 This is wanted because this way 8021q and macvlan know
7679 the device is just moving and can keep their slaves up.
7681 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
7683 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
7684 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U, GFP_KERNEL
);
7687 * Flush the unicast and multicast chains
7692 /* Send a netdev-removed uevent to the old namespace */
7693 kobject_uevent(&dev
->dev
.kobj
, KOBJ_REMOVE
);
7694 netdev_adjacent_del_links(dev
);
7696 /* Actually switch the network namespace */
7697 dev_net_set(dev
, net
);
7699 /* If there is an ifindex conflict assign a new one */
7700 if (__dev_get_by_index(net
, dev
->ifindex
))
7701 dev
->ifindex
= dev_new_index(net
);
7703 /* Send a netdev-add uevent to the new namespace */
7704 kobject_uevent(&dev
->dev
.kobj
, KOBJ_ADD
);
7705 netdev_adjacent_add_links(dev
);
7707 /* Fixup kobjects */
7708 err
= device_rename(&dev
->dev
, dev
->name
);
7711 /* Add the device back in the hashes */
7712 list_netdevice(dev
);
7714 /* Notify protocols, that a new device appeared. */
7715 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
7718 * Prevent userspace races by waiting until the network
7719 * device is fully setup before sending notifications.
7721 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U, GFP_KERNEL
);
7728 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
7730 static int dev_cpu_callback(struct notifier_block
*nfb
,
7731 unsigned long action
,
7734 struct sk_buff
**list_skb
;
7735 struct sk_buff
*skb
;
7736 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
7737 struct softnet_data
*sd
, *oldsd
;
7739 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
7742 local_irq_disable();
7743 cpu
= smp_processor_id();
7744 sd
= &per_cpu(softnet_data
, cpu
);
7745 oldsd
= &per_cpu(softnet_data
, oldcpu
);
7747 /* Find end of our completion_queue. */
7748 list_skb
= &sd
->completion_queue
;
7750 list_skb
= &(*list_skb
)->next
;
7751 /* Append completion queue from offline CPU. */
7752 *list_skb
= oldsd
->completion_queue
;
7753 oldsd
->completion_queue
= NULL
;
7755 /* Append output queue from offline CPU. */
7756 if (oldsd
->output_queue
) {
7757 *sd
->output_queue_tailp
= oldsd
->output_queue
;
7758 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
7759 oldsd
->output_queue
= NULL
;
7760 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
7762 /* Append NAPI poll list from offline CPU, with one exception :
7763 * process_backlog() must be called by cpu owning percpu backlog.
7764 * We properly handle process_queue & input_pkt_queue later.
7766 while (!list_empty(&oldsd
->poll_list
)) {
7767 struct napi_struct
*napi
= list_first_entry(&oldsd
->poll_list
,
7771 list_del_init(&napi
->poll_list
);
7772 if (napi
->poll
== process_backlog
)
7775 ____napi_schedule(sd
, napi
);
7778 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
7781 /* Process offline CPU's input_pkt_queue */
7782 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
7784 input_queue_head_incr(oldsd
);
7786 while ((skb
= skb_dequeue(&oldsd
->input_pkt_queue
))) {
7788 input_queue_head_incr(oldsd
);
7796 * netdev_increment_features - increment feature set by one
7797 * @all: current feature set
7798 * @one: new feature set
7799 * @mask: mask feature set
7801 * Computes a new feature set after adding a device with feature set
7802 * @one to the master device with current feature set @all. Will not
7803 * enable anything that is off in @mask. Returns the new feature set.
7805 netdev_features_t
netdev_increment_features(netdev_features_t all
,
7806 netdev_features_t one
, netdev_features_t mask
)
7808 if (mask
& NETIF_F_HW_CSUM
)
7809 mask
|= NETIF_F_CSUM_MASK
;
7810 mask
|= NETIF_F_VLAN_CHALLENGED
;
7812 all
|= one
& (NETIF_F_ONE_FOR_ALL
| NETIF_F_CSUM_MASK
) & mask
;
7813 all
&= one
| ~NETIF_F_ALL_FOR_ALL
;
7815 /* If one device supports hw checksumming, set for all. */
7816 if (all
& NETIF_F_HW_CSUM
)
7817 all
&= ~(NETIF_F_CSUM_MASK
& ~NETIF_F_HW_CSUM
);
7821 EXPORT_SYMBOL(netdev_increment_features
);
7823 static struct hlist_head
* __net_init
netdev_create_hash(void)
7826 struct hlist_head
*hash
;
7828 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
7830 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
7831 INIT_HLIST_HEAD(&hash
[i
]);
7836 /* Initialize per network namespace state */
7837 static int __net_init
netdev_init(struct net
*net
)
7839 if (net
!= &init_net
)
7840 INIT_LIST_HEAD(&net
->dev_base_head
);
7842 net
->dev_name_head
= netdev_create_hash();
7843 if (net
->dev_name_head
== NULL
)
7846 net
->dev_index_head
= netdev_create_hash();
7847 if (net
->dev_index_head
== NULL
)
7853 kfree(net
->dev_name_head
);
7859 * netdev_drivername - network driver for the device
7860 * @dev: network device
7862 * Determine network driver for device.
7864 const char *netdev_drivername(const struct net_device
*dev
)
7866 const struct device_driver
*driver
;
7867 const struct device
*parent
;
7868 const char *empty
= "";
7870 parent
= dev
->dev
.parent
;
7874 driver
= parent
->driver
;
7875 if (driver
&& driver
->name
)
7876 return driver
->name
;
7880 static void __netdev_printk(const char *level
, const struct net_device
*dev
,
7881 struct va_format
*vaf
)
7883 if (dev
&& dev
->dev
.parent
) {
7884 dev_printk_emit(level
[1] - '0',
7887 dev_driver_string(dev
->dev
.parent
),
7888 dev_name(dev
->dev
.parent
),
7889 netdev_name(dev
), netdev_reg_state(dev
),
7892 printk("%s%s%s: %pV",
7893 level
, netdev_name(dev
), netdev_reg_state(dev
), vaf
);
7895 printk("%s(NULL net_device): %pV", level
, vaf
);
7899 void netdev_printk(const char *level
, const struct net_device
*dev
,
7900 const char *format
, ...)
7902 struct va_format vaf
;
7905 va_start(args
, format
);
7910 __netdev_printk(level
, dev
, &vaf
);
7914 EXPORT_SYMBOL(netdev_printk
);
7916 #define define_netdev_printk_level(func, level) \
7917 void func(const struct net_device *dev, const char *fmt, ...) \
7919 struct va_format vaf; \
7922 va_start(args, fmt); \
7927 __netdev_printk(level, dev, &vaf); \
7931 EXPORT_SYMBOL(func);
7933 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
7934 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
7935 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
7936 define_netdev_printk_level(netdev_err
, KERN_ERR
);
7937 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
7938 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
7939 define_netdev_printk_level(netdev_info
, KERN_INFO
);
7941 static void __net_exit
netdev_exit(struct net
*net
)
7943 kfree(net
->dev_name_head
);
7944 kfree(net
->dev_index_head
);
7947 static struct pernet_operations __net_initdata netdev_net_ops
= {
7948 .init
= netdev_init
,
7949 .exit
= netdev_exit
,
7952 static void __net_exit
default_device_exit(struct net
*net
)
7954 struct net_device
*dev
, *aux
;
7956 * Push all migratable network devices back to the
7957 * initial network namespace
7960 for_each_netdev_safe(net
, dev
, aux
) {
7962 char fb_name
[IFNAMSIZ
];
7964 /* Ignore unmoveable devices (i.e. loopback) */
7965 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
7968 /* Leave virtual devices for the generic cleanup */
7969 if (dev
->rtnl_link_ops
)
7972 /* Push remaining network devices to init_net */
7973 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
7974 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
7976 pr_emerg("%s: failed to move %s to init_net: %d\n",
7977 __func__
, dev
->name
, err
);
7984 static void __net_exit
rtnl_lock_unregistering(struct list_head
*net_list
)
7986 /* Return with the rtnl_lock held when there are no network
7987 * devices unregistering in any network namespace in net_list.
7991 DEFINE_WAIT_FUNC(wait
, woken_wake_function
);
7993 add_wait_queue(&netdev_unregistering_wq
, &wait
);
7995 unregistering
= false;
7997 list_for_each_entry(net
, net_list
, exit_list
) {
7998 if (net
->dev_unreg_count
> 0) {
7999 unregistering
= true;
8007 wait_woken(&wait
, TASK_UNINTERRUPTIBLE
, MAX_SCHEDULE_TIMEOUT
);
8009 remove_wait_queue(&netdev_unregistering_wq
, &wait
);
8012 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
8014 /* At exit all network devices most be removed from a network
8015 * namespace. Do this in the reverse order of registration.
8016 * Do this across as many network namespaces as possible to
8017 * improve batching efficiency.
8019 struct net_device
*dev
;
8021 LIST_HEAD(dev_kill_list
);
8023 /* To prevent network device cleanup code from dereferencing
8024 * loopback devices or network devices that have been freed
8025 * wait here for all pending unregistrations to complete,
8026 * before unregistring the loopback device and allowing the
8027 * network namespace be freed.
8029 * The netdev todo list containing all network devices
8030 * unregistrations that happen in default_device_exit_batch
8031 * will run in the rtnl_unlock() at the end of
8032 * default_device_exit_batch.
8034 rtnl_lock_unregistering(net_list
);
8035 list_for_each_entry(net
, net_list
, exit_list
) {
8036 for_each_netdev_reverse(net
, dev
) {
8037 if (dev
->rtnl_link_ops
&& dev
->rtnl_link_ops
->dellink
)
8038 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
8040 unregister_netdevice_queue(dev
, &dev_kill_list
);
8043 unregister_netdevice_many(&dev_kill_list
);
8047 static struct pernet_operations __net_initdata default_device_ops
= {
8048 .exit
= default_device_exit
,
8049 .exit_batch
= default_device_exit_batch
,
8053 * Initialize the DEV module. At boot time this walks the device list and
8054 * unhooks any devices that fail to initialise (normally hardware not
8055 * present) and leaves us with a valid list of present and active devices.
8060 * This is called single threaded during boot, so no need
8061 * to take the rtnl semaphore.
8063 static int __init
net_dev_init(void)
8065 int i
, rc
= -ENOMEM
;
8067 BUG_ON(!dev_boot_phase
);
8069 if (dev_proc_init())
8072 if (netdev_kobject_init())
8075 INIT_LIST_HEAD(&ptype_all
);
8076 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
8077 INIT_LIST_HEAD(&ptype_base
[i
]);
8079 INIT_LIST_HEAD(&offload_base
);
8081 if (register_pernet_subsys(&netdev_net_ops
))
8085 * Initialise the packet receive queues.
8088 for_each_possible_cpu(i
) {
8089 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
8091 skb_queue_head_init(&sd
->input_pkt_queue
);
8092 skb_queue_head_init(&sd
->process_queue
);
8093 INIT_LIST_HEAD(&sd
->poll_list
);
8094 sd
->output_queue_tailp
= &sd
->output_queue
;
8096 sd
->csd
.func
= rps_trigger_softirq
;
8101 sd
->backlog
.poll
= process_backlog
;
8102 sd
->backlog
.weight
= weight_p
;
8107 /* The loopback device is special if any other network devices
8108 * is present in a network namespace the loopback device must
8109 * be present. Since we now dynamically allocate and free the
8110 * loopback device ensure this invariant is maintained by
8111 * keeping the loopback device as the first device on the
8112 * list of network devices. Ensuring the loopback devices
8113 * is the first device that appears and the last network device
8116 if (register_pernet_device(&loopback_net_ops
))
8119 if (register_pernet_device(&default_device_ops
))
8122 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
8123 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
8125 hotcpu_notifier(dev_cpu_callback
, 0);
8132 subsys_initcall(net_dev_init
);