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 struct sk_buff
*__skb_gso_segment(struct sk_buff
*skb
,
2700 netdev_features_t features
, bool tx_path
)
2702 if (unlikely(skb_needs_check(skb
, tx_path
))) {
2705 skb_warn_bad_offload(skb
);
2707 err
= skb_cow_head(skb
, 0);
2709 return ERR_PTR(err
);
2712 SKB_GSO_CB(skb
)->mac_offset
= skb_headroom(skb
);
2713 SKB_GSO_CB(skb
)->encap_level
= 0;
2715 skb_reset_mac_header(skb
);
2716 skb_reset_mac_len(skb
);
2718 return skb_mac_gso_segment(skb
, features
);
2720 EXPORT_SYMBOL(__skb_gso_segment
);
2722 /* Take action when hardware reception checksum errors are detected. */
2724 void netdev_rx_csum_fault(struct net_device
*dev
)
2726 if (net_ratelimit()) {
2727 pr_err("%s: hw csum failure\n", dev
? dev
->name
: "<unknown>");
2731 EXPORT_SYMBOL(netdev_rx_csum_fault
);
2734 /* Actually, we should eliminate this check as soon as we know, that:
2735 * 1. IOMMU is present and allows to map all the memory.
2736 * 2. No high memory really exists on this machine.
2739 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
2741 #ifdef CONFIG_HIGHMEM
2743 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
2744 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2745 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2746 if (PageHighMem(skb_frag_page(frag
)))
2751 if (PCI_DMA_BUS_IS_PHYS
) {
2752 struct device
*pdev
= dev
->dev
.parent
;
2756 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2757 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2758 dma_addr_t addr
= page_to_phys(skb_frag_page(frag
));
2759 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
2767 /* If MPLS offload request, verify we are testing hardware MPLS features
2768 * instead of standard features for the netdev.
2770 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
2771 static netdev_features_t
net_mpls_features(struct sk_buff
*skb
,
2772 netdev_features_t features
,
2775 if (eth_p_mpls(type
))
2776 features
&= skb
->dev
->mpls_features
;
2781 static netdev_features_t
net_mpls_features(struct sk_buff
*skb
,
2782 netdev_features_t features
,
2789 static netdev_features_t
harmonize_features(struct sk_buff
*skb
,
2790 netdev_features_t features
)
2795 type
= skb_network_protocol(skb
, &tmp
);
2796 features
= net_mpls_features(skb
, features
, type
);
2798 if (skb
->ip_summed
!= CHECKSUM_NONE
&&
2799 !can_checksum_protocol(features
, type
)) {
2800 features
&= ~NETIF_F_CSUM_MASK
;
2801 } else if (illegal_highdma(skb
->dev
, skb
)) {
2802 features
&= ~NETIF_F_SG
;
2808 netdev_features_t
passthru_features_check(struct sk_buff
*skb
,
2809 struct net_device
*dev
,
2810 netdev_features_t features
)
2814 EXPORT_SYMBOL(passthru_features_check
);
2816 static netdev_features_t
dflt_features_check(const struct sk_buff
*skb
,
2817 struct net_device
*dev
,
2818 netdev_features_t features
)
2820 return vlan_features_check(skb
, features
);
2823 netdev_features_t
netif_skb_features(struct sk_buff
*skb
)
2825 struct net_device
*dev
= skb
->dev
;
2826 netdev_features_t features
= dev
->features
;
2827 u16 gso_segs
= skb_shinfo(skb
)->gso_segs
;
2829 if (gso_segs
> dev
->gso_max_segs
|| gso_segs
< dev
->gso_min_segs
)
2830 features
&= ~NETIF_F_GSO_MASK
;
2832 /* If encapsulation offload request, verify we are testing
2833 * hardware encapsulation features instead of standard
2834 * features for the netdev
2836 if (skb
->encapsulation
)
2837 features
&= dev
->hw_enc_features
;
2839 if (skb_vlan_tagged(skb
))
2840 features
= netdev_intersect_features(features
,
2841 dev
->vlan_features
|
2842 NETIF_F_HW_VLAN_CTAG_TX
|
2843 NETIF_F_HW_VLAN_STAG_TX
);
2845 if (dev
->netdev_ops
->ndo_features_check
)
2846 features
&= dev
->netdev_ops
->ndo_features_check(skb
, dev
,
2849 features
&= dflt_features_check(skb
, dev
, features
);
2851 return harmonize_features(skb
, features
);
2853 EXPORT_SYMBOL(netif_skb_features
);
2855 static int xmit_one(struct sk_buff
*skb
, struct net_device
*dev
,
2856 struct netdev_queue
*txq
, bool more
)
2861 if (!list_empty(&ptype_all
) || !list_empty(&dev
->ptype_all
))
2862 dev_queue_xmit_nit(skb
, dev
);
2865 trace_net_dev_start_xmit(skb
, dev
);
2866 rc
= netdev_start_xmit(skb
, dev
, txq
, more
);
2867 trace_net_dev_xmit(skb
, rc
, dev
, len
);
2872 struct sk_buff
*dev_hard_start_xmit(struct sk_buff
*first
, struct net_device
*dev
,
2873 struct netdev_queue
*txq
, int *ret
)
2875 struct sk_buff
*skb
= first
;
2876 int rc
= NETDEV_TX_OK
;
2879 struct sk_buff
*next
= skb
->next
;
2882 rc
= xmit_one(skb
, dev
, txq
, next
!= NULL
);
2883 if (unlikely(!dev_xmit_complete(rc
))) {
2889 if (netif_xmit_stopped(txq
) && skb
) {
2890 rc
= NETDEV_TX_BUSY
;
2900 static struct sk_buff
*validate_xmit_vlan(struct sk_buff
*skb
,
2901 netdev_features_t features
)
2903 if (skb_vlan_tag_present(skb
) &&
2904 !vlan_hw_offload_capable(features
, skb
->vlan_proto
))
2905 skb
= __vlan_hwaccel_push_inside(skb
);
2909 static struct sk_buff
*validate_xmit_skb(struct sk_buff
*skb
, struct net_device
*dev
)
2911 netdev_features_t features
;
2916 features
= netif_skb_features(skb
);
2917 skb
= validate_xmit_vlan(skb
, features
);
2921 if (netif_needs_gso(skb
, features
)) {
2922 struct sk_buff
*segs
;
2924 segs
= skb_gso_segment(skb
, features
);
2932 if (skb_needs_linearize(skb
, features
) &&
2933 __skb_linearize(skb
))
2936 /* If packet is not checksummed and device does not
2937 * support checksumming for this protocol, complete
2938 * checksumming here.
2940 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2941 if (skb
->encapsulation
)
2942 skb_set_inner_transport_header(skb
,
2943 skb_checksum_start_offset(skb
));
2945 skb_set_transport_header(skb
,
2946 skb_checksum_start_offset(skb
));
2947 if (!(features
& NETIF_F_CSUM_MASK
) &&
2948 skb_checksum_help(skb
))
2961 struct sk_buff
*validate_xmit_skb_list(struct sk_buff
*skb
, struct net_device
*dev
)
2963 struct sk_buff
*next
, *head
= NULL
, *tail
;
2965 for (; skb
!= NULL
; skb
= next
) {
2969 /* in case skb wont be segmented, point to itself */
2972 skb
= validate_xmit_skb(skb
, dev
);
2980 /* If skb was segmented, skb->prev points to
2981 * the last segment. If not, it still contains skb.
2988 static void qdisc_pkt_len_init(struct sk_buff
*skb
)
2990 const struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
2992 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
2994 /* To get more precise estimation of bytes sent on wire,
2995 * we add to pkt_len the headers size of all segments
2997 if (shinfo
->gso_size
) {
2998 unsigned int hdr_len
;
2999 u16 gso_segs
= shinfo
->gso_segs
;
3001 /* mac layer + network layer */
3002 hdr_len
= skb_transport_header(skb
) - skb_mac_header(skb
);
3004 /* + transport layer */
3005 if (likely(shinfo
->gso_type
& (SKB_GSO_TCPV4
| SKB_GSO_TCPV6
)))
3006 hdr_len
+= tcp_hdrlen(skb
);
3008 hdr_len
+= sizeof(struct udphdr
);
3010 if (shinfo
->gso_type
& SKB_GSO_DODGY
)
3011 gso_segs
= DIV_ROUND_UP(skb
->len
- hdr_len
,
3014 qdisc_skb_cb(skb
)->pkt_len
+= (gso_segs
- 1) * hdr_len
;
3018 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
3019 struct net_device
*dev
,
3020 struct netdev_queue
*txq
)
3022 spinlock_t
*root_lock
= qdisc_lock(q
);
3026 qdisc_calculate_pkt_len(skb
, q
);
3028 * Heuristic to force contended enqueues to serialize on a
3029 * separate lock before trying to get qdisc main lock.
3030 * This permits __QDISC___STATE_RUNNING owner to get the lock more
3031 * often and dequeue packets faster.
3033 contended
= qdisc_is_running(q
);
3034 if (unlikely(contended
))
3035 spin_lock(&q
->busylock
);
3037 spin_lock(root_lock
);
3038 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
3041 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
3042 qdisc_run_begin(q
)) {
3044 * This is a work-conserving queue; there are no old skbs
3045 * waiting to be sent out; and the qdisc is not running -
3046 * xmit the skb directly.
3049 qdisc_bstats_update(q
, skb
);
3051 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
, true)) {
3052 if (unlikely(contended
)) {
3053 spin_unlock(&q
->busylock
);
3060 rc
= NET_XMIT_SUCCESS
;
3062 rc
= q
->enqueue(skb
, q
) & NET_XMIT_MASK
;
3063 if (qdisc_run_begin(q
)) {
3064 if (unlikely(contended
)) {
3065 spin_unlock(&q
->busylock
);
3071 spin_unlock(root_lock
);
3072 if (unlikely(contended
))
3073 spin_unlock(&q
->busylock
);
3077 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3078 static void skb_update_prio(struct sk_buff
*skb
)
3080 struct netprio_map
*map
= rcu_dereference_bh(skb
->dev
->priomap
);
3082 if (!skb
->priority
&& skb
->sk
&& map
) {
3083 unsigned int prioidx
=
3084 sock_cgroup_prioidx(&skb
->sk
->sk_cgrp_data
);
3086 if (prioidx
< map
->priomap_len
)
3087 skb
->priority
= map
->priomap
[prioidx
];
3091 #define skb_update_prio(skb)
3094 DEFINE_PER_CPU(int, xmit_recursion
);
3095 EXPORT_SYMBOL(xmit_recursion
);
3097 #define RECURSION_LIMIT 10
3100 * dev_loopback_xmit - loop back @skb
3101 * @net: network namespace this loopback is happening in
3102 * @sk: sk needed to be a netfilter okfn
3103 * @skb: buffer to transmit
3105 int dev_loopback_xmit(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
3107 skb_reset_mac_header(skb
);
3108 __skb_pull(skb
, skb_network_offset(skb
));
3109 skb
->pkt_type
= PACKET_LOOPBACK
;
3110 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
3111 WARN_ON(!skb_dst(skb
));
3116 EXPORT_SYMBOL(dev_loopback_xmit
);
3118 #ifdef CONFIG_NET_EGRESS
3119 static struct sk_buff
*
3120 sch_handle_egress(struct sk_buff
*skb
, int *ret
, struct net_device
*dev
)
3122 struct tcf_proto
*cl
= rcu_dereference_bh(dev
->egress_cl_list
);
3123 struct tcf_result cl_res
;
3128 /* skb->tc_verd and qdisc_skb_cb(skb)->pkt_len were already set
3129 * earlier by the caller.
3131 qdisc_bstats_cpu_update(cl
->q
, skb
);
3133 switch (tc_classify(skb
, cl
, &cl_res
, false)) {
3135 case TC_ACT_RECLASSIFY
:
3136 skb
->tc_index
= TC_H_MIN(cl_res
.classid
);
3139 qdisc_qstats_cpu_drop(cl
->q
);
3140 *ret
= NET_XMIT_DROP
;
3144 *ret
= NET_XMIT_SUCCESS
;
3148 case TC_ACT_REDIRECT
:
3149 /* No need to push/pop skb's mac_header here on egress! */
3150 skb_do_redirect(skb
);
3151 *ret
= NET_XMIT_SUCCESS
;
3159 #endif /* CONFIG_NET_EGRESS */
3161 static inline int get_xps_queue(struct net_device
*dev
, struct sk_buff
*skb
)
3164 struct xps_dev_maps
*dev_maps
;
3165 struct xps_map
*map
;
3166 int queue_index
= -1;
3169 dev_maps
= rcu_dereference(dev
->xps_maps
);
3171 map
= rcu_dereference(
3172 dev_maps
->cpu_map
[skb
->sender_cpu
- 1]);
3175 queue_index
= map
->queues
[0];
3177 queue_index
= map
->queues
[reciprocal_scale(skb_get_hash(skb
),
3179 if (unlikely(queue_index
>= dev
->real_num_tx_queues
))
3191 static u16
__netdev_pick_tx(struct net_device
*dev
, struct sk_buff
*skb
)
3193 struct sock
*sk
= skb
->sk
;
3194 int queue_index
= sk_tx_queue_get(sk
);
3196 if (queue_index
< 0 || skb
->ooo_okay
||
3197 queue_index
>= dev
->real_num_tx_queues
) {
3198 int new_index
= get_xps_queue(dev
, skb
);
3200 new_index
= skb_tx_hash(dev
, skb
);
3202 if (queue_index
!= new_index
&& sk
&&
3204 rcu_access_pointer(sk
->sk_dst_cache
))
3205 sk_tx_queue_set(sk
, new_index
);
3207 queue_index
= new_index
;
3213 struct netdev_queue
*netdev_pick_tx(struct net_device
*dev
,
3214 struct sk_buff
*skb
,
3217 int queue_index
= 0;
3220 u32 sender_cpu
= skb
->sender_cpu
- 1;
3222 if (sender_cpu
>= (u32
)NR_CPUS
)
3223 skb
->sender_cpu
= raw_smp_processor_id() + 1;
3226 if (dev
->real_num_tx_queues
!= 1) {
3227 const struct net_device_ops
*ops
= dev
->netdev_ops
;
3228 if (ops
->ndo_select_queue
)
3229 queue_index
= ops
->ndo_select_queue(dev
, skb
, accel_priv
,
3232 queue_index
= __netdev_pick_tx(dev
, skb
);
3235 queue_index
= netdev_cap_txqueue(dev
, queue_index
);
3238 skb_set_queue_mapping(skb
, queue_index
);
3239 return netdev_get_tx_queue(dev
, queue_index
);
3243 * __dev_queue_xmit - transmit a buffer
3244 * @skb: buffer to transmit
3245 * @accel_priv: private data used for L2 forwarding offload
3247 * Queue a buffer for transmission to a network device. The caller must
3248 * have set the device and priority and built the buffer before calling
3249 * this function. The function can be called from an interrupt.
3251 * A negative errno code is returned on a failure. A success does not
3252 * guarantee the frame will be transmitted as it may be dropped due
3253 * to congestion or traffic shaping.
3255 * -----------------------------------------------------------------------------------
3256 * I notice this method can also return errors from the queue disciplines,
3257 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3260 * Regardless of the return value, the skb is consumed, so it is currently
3261 * difficult to retry a send to this method. (You can bump the ref count
3262 * before sending to hold a reference for retry if you are careful.)
3264 * When calling this method, interrupts MUST be enabled. This is because
3265 * the BH enable code must have IRQs enabled so that it will not deadlock.
3268 static int __dev_queue_xmit(struct sk_buff
*skb
, void *accel_priv
)
3270 struct net_device
*dev
= skb
->dev
;
3271 struct netdev_queue
*txq
;
3275 skb_reset_mac_header(skb
);
3277 if (unlikely(skb_shinfo(skb
)->tx_flags
& SKBTX_SCHED_TSTAMP
))
3278 __skb_tstamp_tx(skb
, NULL
, skb
->sk
, SCM_TSTAMP_SCHED
);
3280 /* Disable soft irqs for various locks below. Also
3281 * stops preemption for RCU.
3285 skb_update_prio(skb
);
3287 qdisc_pkt_len_init(skb
);
3288 #ifdef CONFIG_NET_CLS_ACT
3289 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
3290 # ifdef CONFIG_NET_EGRESS
3291 if (static_key_false(&egress_needed
)) {
3292 skb
= sch_handle_egress(skb
, &rc
, dev
);
3298 /* If device/qdisc don't need skb->dst, release it right now while
3299 * its hot in this cpu cache.
3301 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
3306 #ifdef CONFIG_NET_SWITCHDEV
3307 /* Don't forward if offload device already forwarded */
3308 if (skb
->offload_fwd_mark
&&
3309 skb
->offload_fwd_mark
== dev
->offload_fwd_mark
) {
3311 rc
= NET_XMIT_SUCCESS
;
3316 txq
= netdev_pick_tx(dev
, skb
, accel_priv
);
3317 q
= rcu_dereference_bh(txq
->qdisc
);
3319 trace_net_dev_queue(skb
);
3321 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
3325 /* The device has no queue. Common case for software devices:
3326 loopback, all the sorts of tunnels...
3328 Really, it is unlikely that netif_tx_lock protection is necessary
3329 here. (f.e. loopback and IP tunnels are clean ignoring statistics
3331 However, it is possible, that they rely on protection
3334 Check this and shot the lock. It is not prone from deadlocks.
3335 Either shot noqueue qdisc, it is even simpler 8)
3337 if (dev
->flags
& IFF_UP
) {
3338 int cpu
= smp_processor_id(); /* ok because BHs are off */
3340 if (txq
->xmit_lock_owner
!= cpu
) {
3342 if (__this_cpu_read(xmit_recursion
) > RECURSION_LIMIT
)
3343 goto recursion_alert
;
3345 skb
= validate_xmit_skb(skb
, dev
);
3349 HARD_TX_LOCK(dev
, txq
, cpu
);
3351 if (!netif_xmit_stopped(txq
)) {
3352 __this_cpu_inc(xmit_recursion
);
3353 skb
= dev_hard_start_xmit(skb
, dev
, txq
, &rc
);
3354 __this_cpu_dec(xmit_recursion
);
3355 if (dev_xmit_complete(rc
)) {
3356 HARD_TX_UNLOCK(dev
, txq
);
3360 HARD_TX_UNLOCK(dev
, txq
);
3361 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3364 /* Recursion is detected! It is possible,
3368 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3375 rcu_read_unlock_bh();
3377 atomic_long_inc(&dev
->tx_dropped
);
3378 kfree_skb_list(skb
);
3381 rcu_read_unlock_bh();
3385 int dev_queue_xmit(struct sk_buff
*skb
)
3387 return __dev_queue_xmit(skb
, NULL
);
3389 EXPORT_SYMBOL(dev_queue_xmit
);
3391 int dev_queue_xmit_accel(struct sk_buff
*skb
, void *accel_priv
)
3393 return __dev_queue_xmit(skb
, accel_priv
);
3395 EXPORT_SYMBOL(dev_queue_xmit_accel
);
3398 /*=======================================================================
3400 =======================================================================*/
3402 int netdev_max_backlog __read_mostly
= 1000;
3403 EXPORT_SYMBOL(netdev_max_backlog
);
3405 int netdev_tstamp_prequeue __read_mostly
= 1;
3406 int netdev_budget __read_mostly
= 300;
3407 int weight_p __read_mostly
= 64; /* old backlog weight */
3409 /* Called with irq disabled */
3410 static inline void ____napi_schedule(struct softnet_data
*sd
,
3411 struct napi_struct
*napi
)
3413 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
3414 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3419 /* One global table that all flow-based protocols share. */
3420 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
3421 EXPORT_SYMBOL(rps_sock_flow_table
);
3422 u32 rps_cpu_mask __read_mostly
;
3423 EXPORT_SYMBOL(rps_cpu_mask
);
3425 struct static_key rps_needed __read_mostly
;
3427 static struct rps_dev_flow
*
3428 set_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
3429 struct rps_dev_flow
*rflow
, u16 next_cpu
)
3431 if (next_cpu
< nr_cpu_ids
) {
3432 #ifdef CONFIG_RFS_ACCEL
3433 struct netdev_rx_queue
*rxqueue
;
3434 struct rps_dev_flow_table
*flow_table
;
3435 struct rps_dev_flow
*old_rflow
;
3440 /* Should we steer this flow to a different hardware queue? */
3441 if (!skb_rx_queue_recorded(skb
) || !dev
->rx_cpu_rmap
||
3442 !(dev
->features
& NETIF_F_NTUPLE
))
3444 rxq_index
= cpu_rmap_lookup_index(dev
->rx_cpu_rmap
, next_cpu
);
3445 if (rxq_index
== skb_get_rx_queue(skb
))
3448 rxqueue
= dev
->_rx
+ rxq_index
;
3449 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3452 flow_id
= skb_get_hash(skb
) & flow_table
->mask
;
3453 rc
= dev
->netdev_ops
->ndo_rx_flow_steer(dev
, skb
,
3454 rxq_index
, flow_id
);
3458 rflow
= &flow_table
->flows
[flow_id
];
3460 if (old_rflow
->filter
== rflow
->filter
)
3461 old_rflow
->filter
= RPS_NO_FILTER
;
3465 per_cpu(softnet_data
, next_cpu
).input_queue_head
;
3468 rflow
->cpu
= next_cpu
;
3473 * get_rps_cpu is called from netif_receive_skb and returns the target
3474 * CPU from the RPS map of the receiving queue for a given skb.
3475 * rcu_read_lock must be held on entry.
3477 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
3478 struct rps_dev_flow
**rflowp
)
3480 const struct rps_sock_flow_table
*sock_flow_table
;
3481 struct netdev_rx_queue
*rxqueue
= dev
->_rx
;
3482 struct rps_dev_flow_table
*flow_table
;
3483 struct rps_map
*map
;
3488 if (skb_rx_queue_recorded(skb
)) {
3489 u16 index
= skb_get_rx_queue(skb
);
3491 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
3492 WARN_ONCE(dev
->real_num_rx_queues
> 1,
3493 "%s received packet on queue %u, but number "
3494 "of RX queues is %u\n",
3495 dev
->name
, index
, dev
->real_num_rx_queues
);
3501 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3503 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3504 map
= rcu_dereference(rxqueue
->rps_map
);
3505 if (!flow_table
&& !map
)
3508 skb_reset_network_header(skb
);
3509 hash
= skb_get_hash(skb
);
3513 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
3514 if (flow_table
&& sock_flow_table
) {
3515 struct rps_dev_flow
*rflow
;
3519 /* First check into global flow table if there is a match */
3520 ident
= sock_flow_table
->ents
[hash
& sock_flow_table
->mask
];
3521 if ((ident
^ hash
) & ~rps_cpu_mask
)
3524 next_cpu
= ident
& rps_cpu_mask
;
3526 /* OK, now we know there is a match,
3527 * we can look at the local (per receive queue) flow table
3529 rflow
= &flow_table
->flows
[hash
& flow_table
->mask
];
3533 * If the desired CPU (where last recvmsg was done) is
3534 * different from current CPU (one in the rx-queue flow
3535 * table entry), switch if one of the following holds:
3536 * - Current CPU is unset (>= nr_cpu_ids).
3537 * - Current CPU is offline.
3538 * - The current CPU's queue tail has advanced beyond the
3539 * last packet that was enqueued using this table entry.
3540 * This guarantees that all previous packets for the flow
3541 * have been dequeued, thus preserving in order delivery.
3543 if (unlikely(tcpu
!= next_cpu
) &&
3544 (tcpu
>= nr_cpu_ids
|| !cpu_online(tcpu
) ||
3545 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
3546 rflow
->last_qtail
)) >= 0)) {
3548 rflow
= set_rps_cpu(dev
, skb
, rflow
, next_cpu
);
3551 if (tcpu
< nr_cpu_ids
&& cpu_online(tcpu
)) {
3561 tcpu
= map
->cpus
[reciprocal_scale(hash
, map
->len
)];
3562 if (cpu_online(tcpu
)) {
3572 #ifdef CONFIG_RFS_ACCEL
3575 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3576 * @dev: Device on which the filter was set
3577 * @rxq_index: RX queue index
3578 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3579 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3581 * Drivers that implement ndo_rx_flow_steer() should periodically call
3582 * this function for each installed filter and remove the filters for
3583 * which it returns %true.
3585 bool rps_may_expire_flow(struct net_device
*dev
, u16 rxq_index
,
3586 u32 flow_id
, u16 filter_id
)
3588 struct netdev_rx_queue
*rxqueue
= dev
->_rx
+ rxq_index
;
3589 struct rps_dev_flow_table
*flow_table
;
3590 struct rps_dev_flow
*rflow
;
3595 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3596 if (flow_table
&& flow_id
<= flow_table
->mask
) {
3597 rflow
= &flow_table
->flows
[flow_id
];
3598 cpu
= ACCESS_ONCE(rflow
->cpu
);
3599 if (rflow
->filter
== filter_id
&& cpu
< nr_cpu_ids
&&
3600 ((int)(per_cpu(softnet_data
, cpu
).input_queue_head
-
3601 rflow
->last_qtail
) <
3602 (int)(10 * flow_table
->mask
)))
3608 EXPORT_SYMBOL(rps_may_expire_flow
);
3610 #endif /* CONFIG_RFS_ACCEL */
3612 /* Called from hardirq (IPI) context */
3613 static void rps_trigger_softirq(void *data
)
3615 struct softnet_data
*sd
= data
;
3617 ____napi_schedule(sd
, &sd
->backlog
);
3621 #endif /* CONFIG_RPS */
3624 * Check if this softnet_data structure is another cpu one
3625 * If yes, queue it to our IPI list and return 1
3628 static int rps_ipi_queued(struct softnet_data
*sd
)
3631 struct softnet_data
*mysd
= this_cpu_ptr(&softnet_data
);
3634 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
3635 mysd
->rps_ipi_list
= sd
;
3637 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3640 #endif /* CONFIG_RPS */
3644 #ifdef CONFIG_NET_FLOW_LIMIT
3645 int netdev_flow_limit_table_len __read_mostly
= (1 << 12);
3648 static bool skb_flow_limit(struct sk_buff
*skb
, unsigned int qlen
)
3650 #ifdef CONFIG_NET_FLOW_LIMIT
3651 struct sd_flow_limit
*fl
;
3652 struct softnet_data
*sd
;
3653 unsigned int old_flow
, new_flow
;
3655 if (qlen
< (netdev_max_backlog
>> 1))
3658 sd
= this_cpu_ptr(&softnet_data
);
3661 fl
= rcu_dereference(sd
->flow_limit
);
3663 new_flow
= skb_get_hash(skb
) & (fl
->num_buckets
- 1);
3664 old_flow
= fl
->history
[fl
->history_head
];
3665 fl
->history
[fl
->history_head
] = new_flow
;
3668 fl
->history_head
&= FLOW_LIMIT_HISTORY
- 1;
3670 if (likely(fl
->buckets
[old_flow
]))
3671 fl
->buckets
[old_flow
]--;
3673 if (++fl
->buckets
[new_flow
] > (FLOW_LIMIT_HISTORY
>> 1)) {
3685 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3686 * queue (may be a remote CPU queue).
3688 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
3689 unsigned int *qtail
)
3691 struct softnet_data
*sd
;
3692 unsigned long flags
;
3695 sd
= &per_cpu(softnet_data
, cpu
);
3697 local_irq_save(flags
);
3700 if (!netif_running(skb
->dev
))
3702 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3703 if (qlen
<= netdev_max_backlog
&& !skb_flow_limit(skb
, qlen
)) {
3706 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
3707 input_queue_tail_incr_save(sd
, qtail
);
3709 local_irq_restore(flags
);
3710 return NET_RX_SUCCESS
;
3713 /* Schedule NAPI for backlog device
3714 * We can use non atomic operation since we own the queue lock
3716 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
3717 if (!rps_ipi_queued(sd
))
3718 ____napi_schedule(sd
, &sd
->backlog
);
3727 local_irq_restore(flags
);
3729 atomic_long_inc(&skb
->dev
->rx_dropped
);
3734 static int netif_rx_internal(struct sk_buff
*skb
)
3738 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3740 trace_netif_rx(skb
);
3742 if (static_key_false(&rps_needed
)) {
3743 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3749 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3751 cpu
= smp_processor_id();
3753 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3761 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
3768 * netif_rx - post buffer to the network code
3769 * @skb: buffer to post
3771 * This function receives a packet from a device driver and queues it for
3772 * the upper (protocol) levels to process. It always succeeds. The buffer
3773 * may be dropped during processing for congestion control or by the
3777 * NET_RX_SUCCESS (no congestion)
3778 * NET_RX_DROP (packet was dropped)
3782 int netif_rx(struct sk_buff
*skb
)
3784 trace_netif_rx_entry(skb
);
3786 return netif_rx_internal(skb
);
3788 EXPORT_SYMBOL(netif_rx
);
3790 int netif_rx_ni(struct sk_buff
*skb
)
3794 trace_netif_rx_ni_entry(skb
);
3797 err
= netif_rx_internal(skb
);
3798 if (local_softirq_pending())
3804 EXPORT_SYMBOL(netif_rx_ni
);
3806 static void net_tx_action(struct softirq_action
*h
)
3808 struct softnet_data
*sd
= this_cpu_ptr(&softnet_data
);
3810 if (sd
->completion_queue
) {
3811 struct sk_buff
*clist
;
3813 local_irq_disable();
3814 clist
= sd
->completion_queue
;
3815 sd
->completion_queue
= NULL
;
3819 struct sk_buff
*skb
= clist
;
3820 clist
= clist
->next
;
3822 WARN_ON(atomic_read(&skb
->users
));
3823 if (likely(get_kfree_skb_cb(skb
)->reason
== SKB_REASON_CONSUMED
))
3824 trace_consume_skb(skb
);
3826 trace_kfree_skb(skb
, net_tx_action
);
3831 if (sd
->output_queue
) {
3834 local_irq_disable();
3835 head
= sd
->output_queue
;
3836 sd
->output_queue
= NULL
;
3837 sd
->output_queue_tailp
= &sd
->output_queue
;
3841 struct Qdisc
*q
= head
;
3842 spinlock_t
*root_lock
;
3844 head
= head
->next_sched
;
3846 root_lock
= qdisc_lock(q
);
3847 if (spin_trylock(root_lock
)) {
3848 smp_mb__before_atomic();
3849 clear_bit(__QDISC_STATE_SCHED
,
3852 spin_unlock(root_lock
);
3854 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
3856 __netif_reschedule(q
);
3858 smp_mb__before_atomic();
3859 clear_bit(__QDISC_STATE_SCHED
,
3867 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3868 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3869 /* This hook is defined here for ATM LANE */
3870 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
3871 unsigned char *addr
) __read_mostly
;
3872 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
3875 static inline struct sk_buff
*
3876 sch_handle_ingress(struct sk_buff
*skb
, struct packet_type
**pt_prev
, int *ret
,
3877 struct net_device
*orig_dev
)
3879 #ifdef CONFIG_NET_CLS_ACT
3880 struct tcf_proto
*cl
= rcu_dereference_bh(skb
->dev
->ingress_cl_list
);
3881 struct tcf_result cl_res
;
3883 /* If there's at least one ingress present somewhere (so
3884 * we get here via enabled static key), remaining devices
3885 * that are not configured with an ingress qdisc will bail
3891 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
3895 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
3896 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
3897 qdisc_bstats_cpu_update(cl
->q
, skb
);
3899 switch (tc_classify(skb
, cl
, &cl_res
, false)) {
3901 case TC_ACT_RECLASSIFY
:
3902 skb
->tc_index
= TC_H_MIN(cl_res
.classid
);
3905 qdisc_qstats_cpu_drop(cl
->q
);
3910 case TC_ACT_REDIRECT
:
3911 /* skb_mac_header check was done by cls/act_bpf, so
3912 * we can safely push the L2 header back before
3913 * redirecting to another netdev
3915 __skb_push(skb
, skb
->mac_len
);
3916 skb_do_redirect(skb
);
3921 #endif /* CONFIG_NET_CLS_ACT */
3926 * netdev_rx_handler_register - register receive handler
3927 * @dev: device to register a handler for
3928 * @rx_handler: receive handler to register
3929 * @rx_handler_data: data pointer that is used by rx handler
3931 * Register a receive handler for a device. This handler will then be
3932 * called from __netif_receive_skb. A negative errno code is returned
3935 * The caller must hold the rtnl_mutex.
3937 * For a general description of rx_handler, see enum rx_handler_result.
3939 int netdev_rx_handler_register(struct net_device
*dev
,
3940 rx_handler_func_t
*rx_handler
,
3941 void *rx_handler_data
)
3945 if (dev
->rx_handler
)
3948 /* Note: rx_handler_data must be set before rx_handler */
3949 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
3950 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
3954 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
3957 * netdev_rx_handler_unregister - unregister receive handler
3958 * @dev: device to unregister a handler from
3960 * Unregister a receive handler from a device.
3962 * The caller must hold the rtnl_mutex.
3964 void netdev_rx_handler_unregister(struct net_device
*dev
)
3968 RCU_INIT_POINTER(dev
->rx_handler
, NULL
);
3969 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3970 * section has a guarantee to see a non NULL rx_handler_data
3974 RCU_INIT_POINTER(dev
->rx_handler_data
, NULL
);
3976 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
3979 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3980 * the special handling of PFMEMALLOC skbs.
3982 static bool skb_pfmemalloc_protocol(struct sk_buff
*skb
)
3984 switch (skb
->protocol
) {
3985 case htons(ETH_P_ARP
):
3986 case htons(ETH_P_IP
):
3987 case htons(ETH_P_IPV6
):
3988 case htons(ETH_P_8021Q
):
3989 case htons(ETH_P_8021AD
):
3996 static inline int nf_ingress(struct sk_buff
*skb
, struct packet_type
**pt_prev
,
3997 int *ret
, struct net_device
*orig_dev
)
3999 #ifdef CONFIG_NETFILTER_INGRESS
4000 if (nf_hook_ingress_active(skb
)) {
4002 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
4006 return nf_hook_ingress(skb
);
4008 #endif /* CONFIG_NETFILTER_INGRESS */
4012 static int __netif_receive_skb_core(struct sk_buff
*skb
, bool pfmemalloc
)
4014 struct packet_type
*ptype
, *pt_prev
;
4015 rx_handler_func_t
*rx_handler
;
4016 struct net_device
*orig_dev
;
4017 bool deliver_exact
= false;
4018 int ret
= NET_RX_DROP
;
4021 net_timestamp_check(!netdev_tstamp_prequeue
, skb
);
4023 trace_netif_receive_skb(skb
);
4025 orig_dev
= skb
->dev
;
4027 skb_reset_network_header(skb
);
4028 if (!skb_transport_header_was_set(skb
))
4029 skb_reset_transport_header(skb
);
4030 skb_reset_mac_len(skb
);
4035 skb
->skb_iif
= skb
->dev
->ifindex
;
4037 __this_cpu_inc(softnet_data
.processed
);
4039 if (skb
->protocol
== cpu_to_be16(ETH_P_8021Q
) ||
4040 skb
->protocol
== cpu_to_be16(ETH_P_8021AD
)) {
4041 skb
= skb_vlan_untag(skb
);
4046 #ifdef CONFIG_NET_CLS_ACT
4047 if (skb
->tc_verd
& TC_NCLS
) {
4048 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
4056 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
4058 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
4062 list_for_each_entry_rcu(ptype
, &skb
->dev
->ptype_all
, list
) {
4064 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
4069 #ifdef CONFIG_NET_INGRESS
4070 if (static_key_false(&ingress_needed
)) {
4071 skb
= sch_handle_ingress(skb
, &pt_prev
, &ret
, orig_dev
);
4075 if (nf_ingress(skb
, &pt_prev
, &ret
, orig_dev
) < 0)
4079 #ifdef CONFIG_NET_CLS_ACT
4083 if (pfmemalloc
&& !skb_pfmemalloc_protocol(skb
))
4086 if (skb_vlan_tag_present(skb
)) {
4088 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
4091 if (vlan_do_receive(&skb
))
4093 else if (unlikely(!skb
))
4097 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
4100 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
4103 switch (rx_handler(&skb
)) {
4104 case RX_HANDLER_CONSUMED
:
4105 ret
= NET_RX_SUCCESS
;
4107 case RX_HANDLER_ANOTHER
:
4109 case RX_HANDLER_EXACT
:
4110 deliver_exact
= true;
4111 case RX_HANDLER_PASS
:
4118 if (unlikely(skb_vlan_tag_present(skb
))) {
4119 if (skb_vlan_tag_get_id(skb
))
4120 skb
->pkt_type
= PACKET_OTHERHOST
;
4121 /* Note: we might in the future use prio bits
4122 * and set skb->priority like in vlan_do_receive()
4123 * For the time being, just ignore Priority Code Point
4128 type
= skb
->protocol
;
4130 /* deliver only exact match when indicated */
4131 if (likely(!deliver_exact
)) {
4132 deliver_ptype_list_skb(skb
, &pt_prev
, orig_dev
, type
,
4133 &ptype_base
[ntohs(type
) &
4137 deliver_ptype_list_skb(skb
, &pt_prev
, orig_dev
, type
,
4138 &orig_dev
->ptype_specific
);
4140 if (unlikely(skb
->dev
!= orig_dev
)) {
4141 deliver_ptype_list_skb(skb
, &pt_prev
, orig_dev
, type
,
4142 &skb
->dev
->ptype_specific
);
4146 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
4149 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
4152 atomic_long_inc(&skb
->dev
->rx_dropped
);
4154 /* Jamal, now you will not able to escape explaining
4155 * me how you were going to use this. :-)
4164 static int __netif_receive_skb(struct sk_buff
*skb
)
4168 if (sk_memalloc_socks() && skb_pfmemalloc(skb
)) {
4169 unsigned long pflags
= current
->flags
;
4172 * PFMEMALLOC skbs are special, they should
4173 * - be delivered to SOCK_MEMALLOC sockets only
4174 * - stay away from userspace
4175 * - have bounded memory usage
4177 * Use PF_MEMALLOC as this saves us from propagating the allocation
4178 * context down to all allocation sites.
4180 current
->flags
|= PF_MEMALLOC
;
4181 ret
= __netif_receive_skb_core(skb
, true);
4182 tsk_restore_flags(current
, pflags
, PF_MEMALLOC
);
4184 ret
= __netif_receive_skb_core(skb
, false);
4189 static int netif_receive_skb_internal(struct sk_buff
*skb
)
4193 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
4195 if (skb_defer_rx_timestamp(skb
))
4196 return NET_RX_SUCCESS
;
4201 if (static_key_false(&rps_needed
)) {
4202 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
4203 int cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
4206 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
4212 ret
= __netif_receive_skb(skb
);
4218 * netif_receive_skb - process receive buffer from network
4219 * @skb: buffer to process
4221 * netif_receive_skb() is the main receive data processing function.
4222 * It always succeeds. The buffer may be dropped during processing
4223 * for congestion control or by the protocol layers.
4225 * This function may only be called from softirq context and interrupts
4226 * should be enabled.
4228 * Return values (usually ignored):
4229 * NET_RX_SUCCESS: no congestion
4230 * NET_RX_DROP: packet was dropped
4232 int netif_receive_skb(struct sk_buff
*skb
)
4234 trace_netif_receive_skb_entry(skb
);
4236 return netif_receive_skb_internal(skb
);
4238 EXPORT_SYMBOL(netif_receive_skb
);
4240 /* Network device is going away, flush any packets still pending
4241 * Called with irqs disabled.
4243 static void flush_backlog(void *arg
)
4245 struct net_device
*dev
= arg
;
4246 struct softnet_data
*sd
= this_cpu_ptr(&softnet_data
);
4247 struct sk_buff
*skb
, *tmp
;
4250 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
4251 if (skb
->dev
== dev
) {
4252 __skb_unlink(skb
, &sd
->input_pkt_queue
);
4254 input_queue_head_incr(sd
);
4259 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
4260 if (skb
->dev
== dev
) {
4261 __skb_unlink(skb
, &sd
->process_queue
);
4263 input_queue_head_incr(sd
);
4268 static int napi_gro_complete(struct sk_buff
*skb
)
4270 struct packet_offload
*ptype
;
4271 __be16 type
= skb
->protocol
;
4272 struct list_head
*head
= &offload_base
;
4275 BUILD_BUG_ON(sizeof(struct napi_gro_cb
) > sizeof(skb
->cb
));
4277 if (NAPI_GRO_CB(skb
)->count
== 1) {
4278 skb_shinfo(skb
)->gso_size
= 0;
4283 list_for_each_entry_rcu(ptype
, head
, list
) {
4284 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_complete
)
4287 err
= ptype
->callbacks
.gro_complete(skb
, 0);
4293 WARN_ON(&ptype
->list
== head
);
4295 return NET_RX_SUCCESS
;
4299 return netif_receive_skb_internal(skb
);
4302 /* napi->gro_list contains packets ordered by age.
4303 * youngest packets at the head of it.
4304 * Complete skbs in reverse order to reduce latencies.
4306 void napi_gro_flush(struct napi_struct
*napi
, bool flush_old
)
4308 struct sk_buff
*skb
, *prev
= NULL
;
4310 /* scan list and build reverse chain */
4311 for (skb
= napi
->gro_list
; skb
!= NULL
; skb
= skb
->next
) {
4316 for (skb
= prev
; skb
; skb
= prev
) {
4319 if (flush_old
&& NAPI_GRO_CB(skb
)->age
== jiffies
)
4323 napi_gro_complete(skb
);
4327 napi
->gro_list
= NULL
;
4329 EXPORT_SYMBOL(napi_gro_flush
);
4331 static void gro_list_prepare(struct napi_struct
*napi
, struct sk_buff
*skb
)
4334 unsigned int maclen
= skb
->dev
->hard_header_len
;
4335 u32 hash
= skb_get_hash_raw(skb
);
4337 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
4338 unsigned long diffs
;
4340 NAPI_GRO_CB(p
)->flush
= 0;
4342 if (hash
!= skb_get_hash_raw(p
)) {
4343 NAPI_GRO_CB(p
)->same_flow
= 0;
4347 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
4348 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
4349 if (maclen
== ETH_HLEN
)
4350 diffs
|= compare_ether_header(skb_mac_header(p
),
4351 skb_mac_header(skb
));
4353 diffs
= memcmp(skb_mac_header(p
),
4354 skb_mac_header(skb
),
4356 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
4360 static void skb_gro_reset_offset(struct sk_buff
*skb
)
4362 const struct skb_shared_info
*pinfo
= skb_shinfo(skb
);
4363 const skb_frag_t
*frag0
= &pinfo
->frags
[0];
4365 NAPI_GRO_CB(skb
)->data_offset
= 0;
4366 NAPI_GRO_CB(skb
)->frag0
= NULL
;
4367 NAPI_GRO_CB(skb
)->frag0_len
= 0;
4369 if (skb_mac_header(skb
) == skb_tail_pointer(skb
) &&
4371 !PageHighMem(skb_frag_page(frag0
))) {
4372 NAPI_GRO_CB(skb
)->frag0
= skb_frag_address(frag0
);
4373 NAPI_GRO_CB(skb
)->frag0_len
= skb_frag_size(frag0
);
4377 static void gro_pull_from_frag0(struct sk_buff
*skb
, int grow
)
4379 struct skb_shared_info
*pinfo
= skb_shinfo(skb
);
4381 BUG_ON(skb
->end
- skb
->tail
< grow
);
4383 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
4385 skb
->data_len
-= grow
;
4388 pinfo
->frags
[0].page_offset
+= grow
;
4389 skb_frag_size_sub(&pinfo
->frags
[0], grow
);
4391 if (unlikely(!skb_frag_size(&pinfo
->frags
[0]))) {
4392 skb_frag_unref(skb
, 0);
4393 memmove(pinfo
->frags
, pinfo
->frags
+ 1,
4394 --pinfo
->nr_frags
* sizeof(pinfo
->frags
[0]));
4398 static enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
4400 struct sk_buff
**pp
= NULL
;
4401 struct packet_offload
*ptype
;
4402 __be16 type
= skb
->protocol
;
4403 struct list_head
*head
= &offload_base
;
4405 enum gro_result ret
;
4408 if (!(skb
->dev
->features
& NETIF_F_GRO
))
4411 if (skb_is_gso(skb
) || skb_has_frag_list(skb
) || skb
->csum_bad
)
4414 gro_list_prepare(napi
, skb
);
4417 list_for_each_entry_rcu(ptype
, head
, list
) {
4418 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_receive
)
4421 skb_set_network_header(skb
, skb_gro_offset(skb
));
4422 skb_reset_mac_len(skb
);
4423 NAPI_GRO_CB(skb
)->same_flow
= 0;
4424 NAPI_GRO_CB(skb
)->flush
= 0;
4425 NAPI_GRO_CB(skb
)->free
= 0;
4426 NAPI_GRO_CB(skb
)->udp_mark
= 0;
4427 NAPI_GRO_CB(skb
)->gro_remcsum_start
= 0;
4429 /* Setup for GRO checksum validation */
4430 switch (skb
->ip_summed
) {
4431 case CHECKSUM_COMPLETE
:
4432 NAPI_GRO_CB(skb
)->csum
= skb
->csum
;
4433 NAPI_GRO_CB(skb
)->csum_valid
= 1;
4434 NAPI_GRO_CB(skb
)->csum_cnt
= 0;
4436 case CHECKSUM_UNNECESSARY
:
4437 NAPI_GRO_CB(skb
)->csum_cnt
= skb
->csum_level
+ 1;
4438 NAPI_GRO_CB(skb
)->csum_valid
= 0;
4441 NAPI_GRO_CB(skb
)->csum_cnt
= 0;
4442 NAPI_GRO_CB(skb
)->csum_valid
= 0;
4445 pp
= ptype
->callbacks
.gro_receive(&napi
->gro_list
, skb
);
4450 if (&ptype
->list
== head
)
4453 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
4454 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
4457 struct sk_buff
*nskb
= *pp
;
4461 napi_gro_complete(nskb
);
4468 if (NAPI_GRO_CB(skb
)->flush
)
4471 if (unlikely(napi
->gro_count
>= MAX_GRO_SKBS
)) {
4472 struct sk_buff
*nskb
= napi
->gro_list
;
4474 /* locate the end of the list to select the 'oldest' flow */
4475 while (nskb
->next
) {
4481 napi_gro_complete(nskb
);
4485 NAPI_GRO_CB(skb
)->count
= 1;
4486 NAPI_GRO_CB(skb
)->age
= jiffies
;
4487 NAPI_GRO_CB(skb
)->last
= skb
;
4488 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
4489 skb
->next
= napi
->gro_list
;
4490 napi
->gro_list
= skb
;
4494 grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
4496 gro_pull_from_frag0(skb
, grow
);
4505 struct packet_offload
*gro_find_receive_by_type(__be16 type
)
4507 struct list_head
*offload_head
= &offload_base
;
4508 struct packet_offload
*ptype
;
4510 list_for_each_entry_rcu(ptype
, offload_head
, list
) {
4511 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_receive
)
4517 EXPORT_SYMBOL(gro_find_receive_by_type
);
4519 struct packet_offload
*gro_find_complete_by_type(__be16 type
)
4521 struct list_head
*offload_head
= &offload_base
;
4522 struct packet_offload
*ptype
;
4524 list_for_each_entry_rcu(ptype
, offload_head
, list
) {
4525 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_complete
)
4531 EXPORT_SYMBOL(gro_find_complete_by_type
);
4533 static gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
4537 if (netif_receive_skb_internal(skb
))
4545 case GRO_MERGED_FREE
:
4546 if (NAPI_GRO_CB(skb
)->free
== NAPI_GRO_FREE_STOLEN_HEAD
)
4547 kmem_cache_free(skbuff_head_cache
, skb
);
4560 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
4562 skb_mark_napi_id(skb
, napi
);
4563 trace_napi_gro_receive_entry(skb
);
4565 skb_gro_reset_offset(skb
);
4567 return napi_skb_finish(dev_gro_receive(napi
, skb
), skb
);
4569 EXPORT_SYMBOL(napi_gro_receive
);
4571 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
4573 if (unlikely(skb
->pfmemalloc
)) {
4577 __skb_pull(skb
, skb_headlen(skb
));
4578 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4579 skb_reserve(skb
, NET_SKB_PAD
+ NET_IP_ALIGN
- skb_headroom(skb
));
4581 skb
->dev
= napi
->dev
;
4583 skb
->encapsulation
= 0;
4584 skb_shinfo(skb
)->gso_type
= 0;
4585 skb
->truesize
= SKB_TRUESIZE(skb_end_offset(skb
));
4590 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
4592 struct sk_buff
*skb
= napi
->skb
;
4595 skb
= napi_alloc_skb(napi
, GRO_MAX_HEAD
);
4598 skb_mark_napi_id(skb
, napi
);
4603 EXPORT_SYMBOL(napi_get_frags
);
4605 static gro_result_t
napi_frags_finish(struct napi_struct
*napi
,
4606 struct sk_buff
*skb
,
4612 __skb_push(skb
, ETH_HLEN
);
4613 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
4614 if (ret
== GRO_NORMAL
&& netif_receive_skb_internal(skb
))
4619 case GRO_MERGED_FREE
:
4620 napi_reuse_skb(napi
, skb
);
4630 /* Upper GRO stack assumes network header starts at gro_offset=0
4631 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4632 * We copy ethernet header into skb->data to have a common layout.
4634 static struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
4636 struct sk_buff
*skb
= napi
->skb
;
4637 const struct ethhdr
*eth
;
4638 unsigned int hlen
= sizeof(*eth
);
4642 skb_reset_mac_header(skb
);
4643 skb_gro_reset_offset(skb
);
4645 eth
= skb_gro_header_fast(skb
, 0);
4646 if (unlikely(skb_gro_header_hard(skb
, hlen
))) {
4647 eth
= skb_gro_header_slow(skb
, hlen
, 0);
4648 if (unlikely(!eth
)) {
4649 napi_reuse_skb(napi
, skb
);
4653 gro_pull_from_frag0(skb
, hlen
);
4654 NAPI_GRO_CB(skb
)->frag0
+= hlen
;
4655 NAPI_GRO_CB(skb
)->frag0_len
-= hlen
;
4657 __skb_pull(skb
, hlen
);
4660 * This works because the only protocols we care about don't require
4662 * We'll fix it up properly in napi_frags_finish()
4664 skb
->protocol
= eth
->h_proto
;
4669 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
4671 struct sk_buff
*skb
= napi_frags_skb(napi
);
4676 trace_napi_gro_frags_entry(skb
);
4678 return napi_frags_finish(napi
, skb
, dev_gro_receive(napi
, skb
));
4680 EXPORT_SYMBOL(napi_gro_frags
);
4682 /* Compute the checksum from gro_offset and return the folded value
4683 * after adding in any pseudo checksum.
4685 __sum16
__skb_gro_checksum_complete(struct sk_buff
*skb
)
4690 wsum
= skb_checksum(skb
, skb_gro_offset(skb
), skb_gro_len(skb
), 0);
4692 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4693 sum
= csum_fold(csum_add(NAPI_GRO_CB(skb
)->csum
, wsum
));
4695 if (unlikely(skb
->ip_summed
== CHECKSUM_COMPLETE
) &&
4696 !skb
->csum_complete_sw
)
4697 netdev_rx_csum_fault(skb
->dev
);
4700 NAPI_GRO_CB(skb
)->csum
= wsum
;
4701 NAPI_GRO_CB(skb
)->csum_valid
= 1;
4705 EXPORT_SYMBOL(__skb_gro_checksum_complete
);
4708 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4709 * Note: called with local irq disabled, but exits with local irq enabled.
4711 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
4714 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
4717 sd
->rps_ipi_list
= NULL
;
4721 /* Send pending IPI's to kick RPS processing on remote cpus. */
4723 struct softnet_data
*next
= remsd
->rps_ipi_next
;
4725 if (cpu_online(remsd
->cpu
))
4726 smp_call_function_single_async(remsd
->cpu
,
4735 static bool sd_has_rps_ipi_waiting(struct softnet_data
*sd
)
4738 return sd
->rps_ipi_list
!= NULL
;
4744 static int process_backlog(struct napi_struct
*napi
, int quota
)
4747 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
4749 /* Check if we have pending ipi, its better to send them now,
4750 * not waiting net_rx_action() end.
4752 if (sd_has_rps_ipi_waiting(sd
)) {
4753 local_irq_disable();
4754 net_rps_action_and_irq_enable(sd
);
4757 napi
->weight
= weight_p
;
4758 local_irq_disable();
4760 struct sk_buff
*skb
;
4762 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
4765 __netif_receive_skb(skb
);
4767 local_irq_disable();
4768 input_queue_head_incr(sd
);
4769 if (++work
>= quota
) {
4776 if (skb_queue_empty(&sd
->input_pkt_queue
)) {
4778 * Inline a custom version of __napi_complete().
4779 * only current cpu owns and manipulates this napi,
4780 * and NAPI_STATE_SCHED is the only possible flag set
4782 * We can use a plain write instead of clear_bit(),
4783 * and we dont need an smp_mb() memory barrier.
4791 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
4792 &sd
->process_queue
);
4801 * __napi_schedule - schedule for receive
4802 * @n: entry to schedule
4804 * The entry's receive function will be scheduled to run.
4805 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
4807 void __napi_schedule(struct napi_struct
*n
)
4809 unsigned long flags
;
4811 local_irq_save(flags
);
4812 ____napi_schedule(this_cpu_ptr(&softnet_data
), n
);
4813 local_irq_restore(flags
);
4815 EXPORT_SYMBOL(__napi_schedule
);
4818 * __napi_schedule_irqoff - schedule for receive
4819 * @n: entry to schedule
4821 * Variant of __napi_schedule() assuming hard irqs are masked
4823 void __napi_schedule_irqoff(struct napi_struct
*n
)
4825 ____napi_schedule(this_cpu_ptr(&softnet_data
), n
);
4827 EXPORT_SYMBOL(__napi_schedule_irqoff
);
4829 void __napi_complete(struct napi_struct
*n
)
4831 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
4833 list_del_init(&n
->poll_list
);
4834 smp_mb__before_atomic();
4835 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
4837 EXPORT_SYMBOL(__napi_complete
);
4839 void napi_complete_done(struct napi_struct
*n
, int work_done
)
4841 unsigned long flags
;
4844 * don't let napi dequeue from the cpu poll list
4845 * just in case its running on a different cpu
4847 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
4851 unsigned long timeout
= 0;
4854 timeout
= n
->dev
->gro_flush_timeout
;
4857 hrtimer_start(&n
->timer
, ns_to_ktime(timeout
),
4858 HRTIMER_MODE_REL_PINNED
);
4860 napi_gro_flush(n
, false);
4862 if (likely(list_empty(&n
->poll_list
))) {
4863 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED
, &n
->state
));
4865 /* If n->poll_list is not empty, we need to mask irqs */
4866 local_irq_save(flags
);
4868 local_irq_restore(flags
);
4871 EXPORT_SYMBOL(napi_complete_done
);
4873 /* must be called under rcu_read_lock(), as we dont take a reference */
4874 static struct napi_struct
*napi_by_id(unsigned int napi_id
)
4876 unsigned int hash
= napi_id
% HASH_SIZE(napi_hash
);
4877 struct napi_struct
*napi
;
4879 hlist_for_each_entry_rcu(napi
, &napi_hash
[hash
], napi_hash_node
)
4880 if (napi
->napi_id
== napi_id
)
4886 #if defined(CONFIG_NET_RX_BUSY_POLL)
4887 #define BUSY_POLL_BUDGET 8
4888 bool sk_busy_loop(struct sock
*sk
, int nonblock
)
4890 unsigned long end_time
= !nonblock
? sk_busy_loop_end_time(sk
) : 0;
4891 int (*busy_poll
)(struct napi_struct
*dev
);
4892 struct napi_struct
*napi
;
4897 napi
= napi_by_id(sk
->sk_napi_id
);
4901 /* Note: ndo_busy_poll method is optional in linux-4.5 */
4902 busy_poll
= napi
->dev
->netdev_ops
->ndo_busy_poll
;
4908 rc
= busy_poll(napi
);
4909 } else if (napi_schedule_prep(napi
)) {
4910 void *have
= netpoll_poll_lock(napi
);
4912 if (test_bit(NAPI_STATE_SCHED
, &napi
->state
)) {
4913 rc
= napi
->poll(napi
, BUSY_POLL_BUDGET
);
4914 trace_napi_poll(napi
);
4915 if (rc
== BUSY_POLL_BUDGET
) {
4916 napi_complete_done(napi
, rc
);
4917 napi_schedule(napi
);
4920 netpoll_poll_unlock(have
);
4923 NET_ADD_STATS_BH(sock_net(sk
),
4924 LINUX_MIB_BUSYPOLLRXPACKETS
, rc
);
4927 if (rc
== LL_FLUSH_FAILED
)
4928 break; /* permanent failure */
4931 } while (!nonblock
&& skb_queue_empty(&sk
->sk_receive_queue
) &&
4932 !need_resched() && !busy_loop_timeout(end_time
));
4934 rc
= !skb_queue_empty(&sk
->sk_receive_queue
);
4939 EXPORT_SYMBOL(sk_busy_loop
);
4941 #endif /* CONFIG_NET_RX_BUSY_POLL */
4943 void napi_hash_add(struct napi_struct
*napi
)
4945 if (test_bit(NAPI_STATE_NO_BUSY_POLL
, &napi
->state
) ||
4946 test_and_set_bit(NAPI_STATE_HASHED
, &napi
->state
))
4949 spin_lock(&napi_hash_lock
);
4951 /* 0..NR_CPUS+1 range is reserved for sender_cpu use */
4953 if (unlikely(++napi_gen_id
< NR_CPUS
+ 1))
4954 napi_gen_id
= NR_CPUS
+ 1;
4955 } while (napi_by_id(napi_gen_id
));
4956 napi
->napi_id
= napi_gen_id
;
4958 hlist_add_head_rcu(&napi
->napi_hash_node
,
4959 &napi_hash
[napi
->napi_id
% HASH_SIZE(napi_hash
)]);
4961 spin_unlock(&napi_hash_lock
);
4963 EXPORT_SYMBOL_GPL(napi_hash_add
);
4965 /* Warning : caller is responsible to make sure rcu grace period
4966 * is respected before freeing memory containing @napi
4968 bool napi_hash_del(struct napi_struct
*napi
)
4970 bool rcu_sync_needed
= false;
4972 spin_lock(&napi_hash_lock
);
4974 if (test_and_clear_bit(NAPI_STATE_HASHED
, &napi
->state
)) {
4975 rcu_sync_needed
= true;
4976 hlist_del_rcu(&napi
->napi_hash_node
);
4978 spin_unlock(&napi_hash_lock
);
4979 return rcu_sync_needed
;
4981 EXPORT_SYMBOL_GPL(napi_hash_del
);
4983 static enum hrtimer_restart
napi_watchdog(struct hrtimer
*timer
)
4985 struct napi_struct
*napi
;
4987 napi
= container_of(timer
, struct napi_struct
, timer
);
4989 napi_schedule(napi
);
4991 return HRTIMER_NORESTART
;
4994 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
4995 int (*poll
)(struct napi_struct
*, int), int weight
)
4997 INIT_LIST_HEAD(&napi
->poll_list
);
4998 hrtimer_init(&napi
->timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_REL_PINNED
);
4999 napi
->timer
.function
= napi_watchdog
;
5000 napi
->gro_count
= 0;
5001 napi
->gro_list
= NULL
;
5004 if (weight
> NAPI_POLL_WEIGHT
)
5005 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
5007 napi
->weight
= weight
;
5008 list_add(&napi
->dev_list
, &dev
->napi_list
);
5010 #ifdef CONFIG_NETPOLL
5011 spin_lock_init(&napi
->poll_lock
);
5012 napi
->poll_owner
= -1;
5014 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
5015 napi_hash_add(napi
);
5017 EXPORT_SYMBOL(netif_napi_add
);
5019 void napi_disable(struct napi_struct
*n
)
5022 set_bit(NAPI_STATE_DISABLE
, &n
->state
);
5024 while (test_and_set_bit(NAPI_STATE_SCHED
, &n
->state
))
5026 while (test_and_set_bit(NAPI_STATE_NPSVC
, &n
->state
))
5029 hrtimer_cancel(&n
->timer
);
5031 clear_bit(NAPI_STATE_DISABLE
, &n
->state
);
5033 EXPORT_SYMBOL(napi_disable
);
5035 /* Must be called in process context */
5036 void netif_napi_del(struct napi_struct
*napi
)
5039 if (napi_hash_del(napi
))
5041 list_del_init(&napi
->dev_list
);
5042 napi_free_frags(napi
);
5044 kfree_skb_list(napi
->gro_list
);
5045 napi
->gro_list
= NULL
;
5046 napi
->gro_count
= 0;
5048 EXPORT_SYMBOL(netif_napi_del
);
5050 static int napi_poll(struct napi_struct
*n
, struct list_head
*repoll
)
5055 list_del_init(&n
->poll_list
);
5057 have
= netpoll_poll_lock(n
);
5061 /* This NAPI_STATE_SCHED test is for avoiding a race
5062 * with netpoll's poll_napi(). Only the entity which
5063 * obtains the lock and sees NAPI_STATE_SCHED set will
5064 * actually make the ->poll() call. Therefore we avoid
5065 * accidentally calling ->poll() when NAPI is not scheduled.
5068 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
5069 work
= n
->poll(n
, weight
);
5073 WARN_ON_ONCE(work
> weight
);
5075 if (likely(work
< weight
))
5078 /* Drivers must not modify the NAPI state if they
5079 * consume the entire weight. In such cases this code
5080 * still "owns" the NAPI instance and therefore can
5081 * move the instance around on the list at-will.
5083 if (unlikely(napi_disable_pending(n
))) {
5089 /* flush too old packets
5090 * If HZ < 1000, flush all packets.
5092 napi_gro_flush(n
, HZ
>= 1000);
5095 /* Some drivers may have called napi_schedule
5096 * prior to exhausting their budget.
5098 if (unlikely(!list_empty(&n
->poll_list
))) {
5099 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
5100 n
->dev
? n
->dev
->name
: "backlog");
5104 list_add_tail(&n
->poll_list
, repoll
);
5107 netpoll_poll_unlock(have
);
5112 static void net_rx_action(struct softirq_action
*h
)
5114 struct softnet_data
*sd
= this_cpu_ptr(&softnet_data
);
5115 unsigned long time_limit
= jiffies
+ 2;
5116 int budget
= netdev_budget
;
5120 local_irq_disable();
5121 list_splice_init(&sd
->poll_list
, &list
);
5125 struct napi_struct
*n
;
5127 if (list_empty(&list
)) {
5128 if (!sd_has_rps_ipi_waiting(sd
) && list_empty(&repoll
))
5133 n
= list_first_entry(&list
, struct napi_struct
, poll_list
);
5134 budget
-= napi_poll(n
, &repoll
);
5136 /* If softirq window is exhausted then punt.
5137 * Allow this to run for 2 jiffies since which will allow
5138 * an average latency of 1.5/HZ.
5140 if (unlikely(budget
<= 0 ||
5141 time_after_eq(jiffies
, time_limit
))) {
5147 local_irq_disable();
5149 list_splice_tail_init(&sd
->poll_list
, &list
);
5150 list_splice_tail(&repoll
, &list
);
5151 list_splice(&list
, &sd
->poll_list
);
5152 if (!list_empty(&sd
->poll_list
))
5153 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
5155 net_rps_action_and_irq_enable(sd
);
5158 struct netdev_adjacent
{
5159 struct net_device
*dev
;
5161 /* upper master flag, there can only be one master device per list */
5164 /* counter for the number of times this device was added to us */
5167 /* private field for the users */
5170 struct list_head list
;
5171 struct rcu_head rcu
;
5174 static struct netdev_adjacent
*__netdev_find_adj(struct net_device
*adj_dev
,
5175 struct list_head
*adj_list
)
5177 struct netdev_adjacent
*adj
;
5179 list_for_each_entry(adj
, adj_list
, list
) {
5180 if (adj
->dev
== adj_dev
)
5187 * netdev_has_upper_dev - Check if device is linked to an upper device
5189 * @upper_dev: upper device to check
5191 * Find out if a device is linked to specified upper device and return true
5192 * in case it is. Note that this checks only immediate upper device,
5193 * not through a complete stack of devices. The caller must hold the RTNL lock.
5195 bool netdev_has_upper_dev(struct net_device
*dev
,
5196 struct net_device
*upper_dev
)
5200 return __netdev_find_adj(upper_dev
, &dev
->all_adj_list
.upper
);
5202 EXPORT_SYMBOL(netdev_has_upper_dev
);
5205 * netdev_has_any_upper_dev - Check if device is linked to some device
5208 * Find out if a device is linked to an upper device and return true in case
5209 * it is. The caller must hold the RTNL lock.
5211 static bool netdev_has_any_upper_dev(struct net_device
*dev
)
5215 return !list_empty(&dev
->all_adj_list
.upper
);
5219 * netdev_master_upper_dev_get - Get master upper device
5222 * Find a master upper device and return pointer to it or NULL in case
5223 * it's not there. The caller must hold the RTNL lock.
5225 struct net_device
*netdev_master_upper_dev_get(struct net_device
*dev
)
5227 struct netdev_adjacent
*upper
;
5231 if (list_empty(&dev
->adj_list
.upper
))
5234 upper
= list_first_entry(&dev
->adj_list
.upper
,
5235 struct netdev_adjacent
, list
);
5236 if (likely(upper
->master
))
5240 EXPORT_SYMBOL(netdev_master_upper_dev_get
);
5242 void *netdev_adjacent_get_private(struct list_head
*adj_list
)
5244 struct netdev_adjacent
*adj
;
5246 adj
= list_entry(adj_list
, struct netdev_adjacent
, list
);
5248 return adj
->private;
5250 EXPORT_SYMBOL(netdev_adjacent_get_private
);
5253 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
5255 * @iter: list_head ** of the current position
5257 * Gets the next device from the dev's upper list, starting from iter
5258 * position. The caller must hold RCU read lock.
5260 struct net_device
*netdev_upper_get_next_dev_rcu(struct net_device
*dev
,
5261 struct list_head
**iter
)
5263 struct netdev_adjacent
*upper
;
5265 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5267 upper
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
5269 if (&upper
->list
== &dev
->adj_list
.upper
)
5272 *iter
= &upper
->list
;
5276 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu
);
5279 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
5281 * @iter: list_head ** of the current position
5283 * Gets the next device from the dev's upper list, starting from iter
5284 * position. The caller must hold RCU read lock.
5286 struct net_device
*netdev_all_upper_get_next_dev_rcu(struct net_device
*dev
,
5287 struct list_head
**iter
)
5289 struct netdev_adjacent
*upper
;
5291 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5293 upper
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
5295 if (&upper
->list
== &dev
->all_adj_list
.upper
)
5298 *iter
= &upper
->list
;
5302 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu
);
5305 * netdev_lower_get_next_private - Get the next ->private from the
5306 * lower neighbour list
5308 * @iter: list_head ** of the current position
5310 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5311 * list, starting from iter position. The caller must hold either hold the
5312 * RTNL lock or its own locking that guarantees that the neighbour lower
5313 * list will remain unchanged.
5315 void *netdev_lower_get_next_private(struct net_device
*dev
,
5316 struct list_head
**iter
)
5318 struct netdev_adjacent
*lower
;
5320 lower
= list_entry(*iter
, struct netdev_adjacent
, list
);
5322 if (&lower
->list
== &dev
->adj_list
.lower
)
5325 *iter
= lower
->list
.next
;
5327 return lower
->private;
5329 EXPORT_SYMBOL(netdev_lower_get_next_private
);
5332 * netdev_lower_get_next_private_rcu - Get the next ->private from the
5333 * lower neighbour list, RCU
5336 * @iter: list_head ** of the current position
5338 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5339 * list, starting from iter position. The caller must hold RCU read lock.
5341 void *netdev_lower_get_next_private_rcu(struct net_device
*dev
,
5342 struct list_head
**iter
)
5344 struct netdev_adjacent
*lower
;
5346 WARN_ON_ONCE(!rcu_read_lock_held());
5348 lower
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
5350 if (&lower
->list
== &dev
->adj_list
.lower
)
5353 *iter
= &lower
->list
;
5355 return lower
->private;
5357 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu
);
5360 * netdev_lower_get_next - Get the next device from the lower neighbour
5363 * @iter: list_head ** of the current position
5365 * Gets the next netdev_adjacent from the dev's lower neighbour
5366 * list, starting from iter position. The caller must hold RTNL lock or
5367 * its own locking that guarantees that the neighbour lower
5368 * list will remain unchanged.
5370 void *netdev_lower_get_next(struct net_device
*dev
, struct list_head
**iter
)
5372 struct netdev_adjacent
*lower
;
5374 lower
= list_entry((*iter
)->next
, struct netdev_adjacent
, list
);
5376 if (&lower
->list
== &dev
->adj_list
.lower
)
5379 *iter
= &lower
->list
;
5383 EXPORT_SYMBOL(netdev_lower_get_next
);
5386 * netdev_lower_get_first_private_rcu - Get the first ->private from the
5387 * lower neighbour list, RCU
5391 * Gets the first netdev_adjacent->private from the dev's lower neighbour
5392 * list. The caller must hold RCU read lock.
5394 void *netdev_lower_get_first_private_rcu(struct net_device
*dev
)
5396 struct netdev_adjacent
*lower
;
5398 lower
= list_first_or_null_rcu(&dev
->adj_list
.lower
,
5399 struct netdev_adjacent
, list
);
5401 return lower
->private;
5404 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu
);
5407 * netdev_master_upper_dev_get_rcu - Get master upper device
5410 * Find a master upper device and return pointer to it or NULL in case
5411 * it's not there. The caller must hold the RCU read lock.
5413 struct net_device
*netdev_master_upper_dev_get_rcu(struct net_device
*dev
)
5415 struct netdev_adjacent
*upper
;
5417 upper
= list_first_or_null_rcu(&dev
->adj_list
.upper
,
5418 struct netdev_adjacent
, list
);
5419 if (upper
&& likely(upper
->master
))
5423 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu
);
5425 static int netdev_adjacent_sysfs_add(struct net_device
*dev
,
5426 struct net_device
*adj_dev
,
5427 struct list_head
*dev_list
)
5429 char linkname
[IFNAMSIZ
+7];
5430 sprintf(linkname
, dev_list
== &dev
->adj_list
.upper
?
5431 "upper_%s" : "lower_%s", adj_dev
->name
);
5432 return sysfs_create_link(&(dev
->dev
.kobj
), &(adj_dev
->dev
.kobj
),
5435 static void netdev_adjacent_sysfs_del(struct net_device
*dev
,
5437 struct list_head
*dev_list
)
5439 char linkname
[IFNAMSIZ
+7];
5440 sprintf(linkname
, dev_list
== &dev
->adj_list
.upper
?
5441 "upper_%s" : "lower_%s", name
);
5442 sysfs_remove_link(&(dev
->dev
.kobj
), linkname
);
5445 static inline bool netdev_adjacent_is_neigh_list(struct net_device
*dev
,
5446 struct net_device
*adj_dev
,
5447 struct list_head
*dev_list
)
5449 return (dev_list
== &dev
->adj_list
.upper
||
5450 dev_list
== &dev
->adj_list
.lower
) &&
5451 net_eq(dev_net(dev
), dev_net(adj_dev
));
5454 static int __netdev_adjacent_dev_insert(struct net_device
*dev
,
5455 struct net_device
*adj_dev
,
5456 struct list_head
*dev_list
,
5457 void *private, bool master
)
5459 struct netdev_adjacent
*adj
;
5462 adj
= __netdev_find_adj(adj_dev
, dev_list
);
5469 adj
= kmalloc(sizeof(*adj
), GFP_KERNEL
);
5474 adj
->master
= master
;
5476 adj
->private = private;
5479 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
5480 adj_dev
->name
, dev
->name
, adj_dev
->name
);
5482 if (netdev_adjacent_is_neigh_list(dev
, adj_dev
, dev_list
)) {
5483 ret
= netdev_adjacent_sysfs_add(dev
, adj_dev
, dev_list
);
5488 /* Ensure that master link is always the first item in list. */
5490 ret
= sysfs_create_link(&(dev
->dev
.kobj
),
5491 &(adj_dev
->dev
.kobj
), "master");
5493 goto remove_symlinks
;
5495 list_add_rcu(&adj
->list
, dev_list
);
5497 list_add_tail_rcu(&adj
->list
, dev_list
);
5503 if (netdev_adjacent_is_neigh_list(dev
, adj_dev
, dev_list
))
5504 netdev_adjacent_sysfs_del(dev
, adj_dev
->name
, dev_list
);
5512 static void __netdev_adjacent_dev_remove(struct net_device
*dev
,
5513 struct net_device
*adj_dev
,
5514 struct list_head
*dev_list
)
5516 struct netdev_adjacent
*adj
;
5518 adj
= __netdev_find_adj(adj_dev
, dev_list
);
5521 pr_err("tried to remove device %s from %s\n",
5522 dev
->name
, adj_dev
->name
);
5526 if (adj
->ref_nr
> 1) {
5527 pr_debug("%s to %s ref_nr-- = %d\n", dev
->name
, adj_dev
->name
,
5534 sysfs_remove_link(&(dev
->dev
.kobj
), "master");
5536 if (netdev_adjacent_is_neigh_list(dev
, adj_dev
, dev_list
))
5537 netdev_adjacent_sysfs_del(dev
, adj_dev
->name
, dev_list
);
5539 list_del_rcu(&adj
->list
);
5540 pr_debug("dev_put for %s, because link removed from %s to %s\n",
5541 adj_dev
->name
, dev
->name
, adj_dev
->name
);
5543 kfree_rcu(adj
, rcu
);
5546 static int __netdev_adjacent_dev_link_lists(struct net_device
*dev
,
5547 struct net_device
*upper_dev
,
5548 struct list_head
*up_list
,
5549 struct list_head
*down_list
,
5550 void *private, bool master
)
5554 ret
= __netdev_adjacent_dev_insert(dev
, upper_dev
, up_list
, private,
5559 ret
= __netdev_adjacent_dev_insert(upper_dev
, dev
, down_list
, private,
5562 __netdev_adjacent_dev_remove(dev
, upper_dev
, up_list
);
5569 static int __netdev_adjacent_dev_link(struct net_device
*dev
,
5570 struct net_device
*upper_dev
)
5572 return __netdev_adjacent_dev_link_lists(dev
, upper_dev
,
5573 &dev
->all_adj_list
.upper
,
5574 &upper_dev
->all_adj_list
.lower
,
5578 static void __netdev_adjacent_dev_unlink_lists(struct net_device
*dev
,
5579 struct net_device
*upper_dev
,
5580 struct list_head
*up_list
,
5581 struct list_head
*down_list
)
5583 __netdev_adjacent_dev_remove(dev
, upper_dev
, up_list
);
5584 __netdev_adjacent_dev_remove(upper_dev
, dev
, down_list
);
5587 static void __netdev_adjacent_dev_unlink(struct net_device
*dev
,
5588 struct net_device
*upper_dev
)
5590 __netdev_adjacent_dev_unlink_lists(dev
, upper_dev
,
5591 &dev
->all_adj_list
.upper
,
5592 &upper_dev
->all_adj_list
.lower
);
5595 static int __netdev_adjacent_dev_link_neighbour(struct net_device
*dev
,
5596 struct net_device
*upper_dev
,
5597 void *private, bool master
)
5599 int ret
= __netdev_adjacent_dev_link(dev
, upper_dev
);
5604 ret
= __netdev_adjacent_dev_link_lists(dev
, upper_dev
,
5605 &dev
->adj_list
.upper
,
5606 &upper_dev
->adj_list
.lower
,
5609 __netdev_adjacent_dev_unlink(dev
, upper_dev
);
5616 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device
*dev
,
5617 struct net_device
*upper_dev
)
5619 __netdev_adjacent_dev_unlink(dev
, upper_dev
);
5620 __netdev_adjacent_dev_unlink_lists(dev
, upper_dev
,
5621 &dev
->adj_list
.upper
,
5622 &upper_dev
->adj_list
.lower
);
5625 static int __netdev_upper_dev_link(struct net_device
*dev
,
5626 struct net_device
*upper_dev
, bool master
,
5627 void *upper_priv
, void *upper_info
)
5629 struct netdev_notifier_changeupper_info changeupper_info
;
5630 struct netdev_adjacent
*i
, *j
, *to_i
, *to_j
;
5635 if (dev
== upper_dev
)
5638 /* To prevent loops, check if dev is not upper device to upper_dev. */
5639 if (__netdev_find_adj(dev
, &upper_dev
->all_adj_list
.upper
))
5642 if (__netdev_find_adj(upper_dev
, &dev
->adj_list
.upper
))
5645 if (master
&& netdev_master_upper_dev_get(dev
))
5648 changeupper_info
.upper_dev
= upper_dev
;
5649 changeupper_info
.master
= master
;
5650 changeupper_info
.linking
= true;
5651 changeupper_info
.upper_info
= upper_info
;
5653 ret
= call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER
, dev
,
5654 &changeupper_info
.info
);
5655 ret
= notifier_to_errno(ret
);
5659 ret
= __netdev_adjacent_dev_link_neighbour(dev
, upper_dev
, upper_priv
,
5664 /* Now that we linked these devs, make all the upper_dev's
5665 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5666 * versa, and don't forget the devices itself. All of these
5667 * links are non-neighbours.
5669 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5670 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
) {
5671 pr_debug("Interlinking %s with %s, non-neighbour\n",
5672 i
->dev
->name
, j
->dev
->name
);
5673 ret
= __netdev_adjacent_dev_link(i
->dev
, j
->dev
);
5679 /* add dev to every upper_dev's upper device */
5680 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
) {
5681 pr_debug("linking %s's upper device %s with %s\n",
5682 upper_dev
->name
, i
->dev
->name
, dev
->name
);
5683 ret
= __netdev_adjacent_dev_link(dev
, i
->dev
);
5685 goto rollback_upper_mesh
;
5688 /* add upper_dev to every dev's lower device */
5689 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5690 pr_debug("linking %s's lower device %s with %s\n", dev
->name
,
5691 i
->dev
->name
, upper_dev
->name
);
5692 ret
= __netdev_adjacent_dev_link(i
->dev
, upper_dev
);
5694 goto rollback_lower_mesh
;
5697 ret
= call_netdevice_notifiers_info(NETDEV_CHANGEUPPER
, dev
,
5698 &changeupper_info
.info
);
5699 ret
= notifier_to_errno(ret
);
5701 goto rollback_lower_mesh
;
5705 rollback_lower_mesh
:
5707 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5710 __netdev_adjacent_dev_unlink(i
->dev
, upper_dev
);
5715 rollback_upper_mesh
:
5717 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
) {
5720 __netdev_adjacent_dev_unlink(dev
, i
->dev
);
5728 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5729 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
) {
5730 if (i
== to_i
&& j
== to_j
)
5732 __netdev_adjacent_dev_unlink(i
->dev
, j
->dev
);
5738 __netdev_adjacent_dev_unlink_neighbour(dev
, upper_dev
);
5744 * netdev_upper_dev_link - Add a link to the upper device
5746 * @upper_dev: new upper device
5748 * Adds a link to device which is upper to this one. The caller must hold
5749 * the RTNL lock. On a failure a negative errno code is returned.
5750 * On success the reference counts are adjusted and the function
5753 int netdev_upper_dev_link(struct net_device
*dev
,
5754 struct net_device
*upper_dev
)
5756 return __netdev_upper_dev_link(dev
, upper_dev
, false, NULL
, NULL
);
5758 EXPORT_SYMBOL(netdev_upper_dev_link
);
5761 * netdev_master_upper_dev_link - Add a master link to the upper device
5763 * @upper_dev: new upper device
5764 * @upper_priv: upper device private
5765 * @upper_info: upper info to be passed down via notifier
5767 * Adds a link to device which is upper to this one. In this case, only
5768 * one master upper device can be linked, although other non-master devices
5769 * might be linked as well. The caller must hold the RTNL lock.
5770 * On a failure a negative errno code is returned. On success the reference
5771 * counts are adjusted and the function returns zero.
5773 int netdev_master_upper_dev_link(struct net_device
*dev
,
5774 struct net_device
*upper_dev
,
5775 void *upper_priv
, void *upper_info
)
5777 return __netdev_upper_dev_link(dev
, upper_dev
, true,
5778 upper_priv
, upper_info
);
5780 EXPORT_SYMBOL(netdev_master_upper_dev_link
);
5783 * netdev_upper_dev_unlink - Removes a link to upper device
5785 * @upper_dev: new upper device
5787 * Removes a link to device which is upper to this one. The caller must hold
5790 void netdev_upper_dev_unlink(struct net_device
*dev
,
5791 struct net_device
*upper_dev
)
5793 struct netdev_notifier_changeupper_info changeupper_info
;
5794 struct netdev_adjacent
*i
, *j
;
5797 changeupper_info
.upper_dev
= upper_dev
;
5798 changeupper_info
.master
= netdev_master_upper_dev_get(dev
) == upper_dev
;
5799 changeupper_info
.linking
= false;
5801 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER
, dev
,
5802 &changeupper_info
.info
);
5804 __netdev_adjacent_dev_unlink_neighbour(dev
, upper_dev
);
5806 /* Here is the tricky part. We must remove all dev's lower
5807 * devices from all upper_dev's upper devices and vice
5808 * versa, to maintain the graph relationship.
5810 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
)
5811 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
)
5812 __netdev_adjacent_dev_unlink(i
->dev
, j
->dev
);
5814 /* remove also the devices itself from lower/upper device
5817 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
)
5818 __netdev_adjacent_dev_unlink(i
->dev
, upper_dev
);
5820 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
)
5821 __netdev_adjacent_dev_unlink(dev
, i
->dev
);
5823 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER
, dev
,
5824 &changeupper_info
.info
);
5826 EXPORT_SYMBOL(netdev_upper_dev_unlink
);
5829 * netdev_bonding_info_change - Dispatch event about slave change
5831 * @bonding_info: info to dispatch
5833 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
5834 * The caller must hold the RTNL lock.
5836 void netdev_bonding_info_change(struct net_device
*dev
,
5837 struct netdev_bonding_info
*bonding_info
)
5839 struct netdev_notifier_bonding_info info
;
5841 memcpy(&info
.bonding_info
, bonding_info
,
5842 sizeof(struct netdev_bonding_info
));
5843 call_netdevice_notifiers_info(NETDEV_BONDING_INFO
, dev
,
5846 EXPORT_SYMBOL(netdev_bonding_info_change
);
5848 static void netdev_adjacent_add_links(struct net_device
*dev
)
5850 struct netdev_adjacent
*iter
;
5852 struct net
*net
= dev_net(dev
);
5854 list_for_each_entry(iter
, &dev
->adj_list
.upper
, list
) {
5855 if (!net_eq(net
,dev_net(iter
->dev
)))
5857 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
5858 &iter
->dev
->adj_list
.lower
);
5859 netdev_adjacent_sysfs_add(dev
, iter
->dev
,
5860 &dev
->adj_list
.upper
);
5863 list_for_each_entry(iter
, &dev
->adj_list
.lower
, list
) {
5864 if (!net_eq(net
,dev_net(iter
->dev
)))
5866 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
5867 &iter
->dev
->adj_list
.upper
);
5868 netdev_adjacent_sysfs_add(dev
, iter
->dev
,
5869 &dev
->adj_list
.lower
);
5873 static void netdev_adjacent_del_links(struct net_device
*dev
)
5875 struct netdev_adjacent
*iter
;
5877 struct net
*net
= dev_net(dev
);
5879 list_for_each_entry(iter
, &dev
->adj_list
.upper
, list
) {
5880 if (!net_eq(net
,dev_net(iter
->dev
)))
5882 netdev_adjacent_sysfs_del(iter
->dev
, dev
->name
,
5883 &iter
->dev
->adj_list
.lower
);
5884 netdev_adjacent_sysfs_del(dev
, iter
->dev
->name
,
5885 &dev
->adj_list
.upper
);
5888 list_for_each_entry(iter
, &dev
->adj_list
.lower
, list
) {
5889 if (!net_eq(net
,dev_net(iter
->dev
)))
5891 netdev_adjacent_sysfs_del(iter
->dev
, dev
->name
,
5892 &iter
->dev
->adj_list
.upper
);
5893 netdev_adjacent_sysfs_del(dev
, iter
->dev
->name
,
5894 &dev
->adj_list
.lower
);
5898 void netdev_adjacent_rename_links(struct net_device
*dev
, char *oldname
)
5900 struct netdev_adjacent
*iter
;
5902 struct net
*net
= dev_net(dev
);
5904 list_for_each_entry(iter
, &dev
->adj_list
.upper
, list
) {
5905 if (!net_eq(net
,dev_net(iter
->dev
)))
5907 netdev_adjacent_sysfs_del(iter
->dev
, oldname
,
5908 &iter
->dev
->adj_list
.lower
);
5909 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
5910 &iter
->dev
->adj_list
.lower
);
5913 list_for_each_entry(iter
, &dev
->adj_list
.lower
, list
) {
5914 if (!net_eq(net
,dev_net(iter
->dev
)))
5916 netdev_adjacent_sysfs_del(iter
->dev
, oldname
,
5917 &iter
->dev
->adj_list
.upper
);
5918 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
5919 &iter
->dev
->adj_list
.upper
);
5923 void *netdev_lower_dev_get_private(struct net_device
*dev
,
5924 struct net_device
*lower_dev
)
5926 struct netdev_adjacent
*lower
;
5930 lower
= __netdev_find_adj(lower_dev
, &dev
->adj_list
.lower
);
5934 return lower
->private;
5936 EXPORT_SYMBOL(netdev_lower_dev_get_private
);
5939 int dev_get_nest_level(struct net_device
*dev
,
5940 bool (*type_check
)(const struct net_device
*dev
))
5942 struct net_device
*lower
= NULL
;
5943 struct list_head
*iter
;
5949 netdev_for_each_lower_dev(dev
, lower
, iter
) {
5950 nest
= dev_get_nest_level(lower
, type_check
);
5951 if (max_nest
< nest
)
5955 if (type_check(dev
))
5960 EXPORT_SYMBOL(dev_get_nest_level
);
5963 * netdev_lower_change - Dispatch event about lower device state change
5964 * @lower_dev: device
5965 * @lower_state_info: state to dispatch
5967 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
5968 * The caller must hold the RTNL lock.
5970 void netdev_lower_state_changed(struct net_device
*lower_dev
,
5971 void *lower_state_info
)
5973 struct netdev_notifier_changelowerstate_info changelowerstate_info
;
5976 changelowerstate_info
.lower_state_info
= lower_state_info
;
5977 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE
, lower_dev
,
5978 &changelowerstate_info
.info
);
5980 EXPORT_SYMBOL(netdev_lower_state_changed
);
5982 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
5984 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5986 if (ops
->ndo_change_rx_flags
)
5987 ops
->ndo_change_rx_flags(dev
, flags
);
5990 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
, bool notify
)
5992 unsigned int old_flags
= dev
->flags
;
5998 dev
->flags
|= IFF_PROMISC
;
5999 dev
->promiscuity
+= inc
;
6000 if (dev
->promiscuity
== 0) {
6003 * If inc causes overflow, untouch promisc and return error.
6006 dev
->flags
&= ~IFF_PROMISC
;
6008 dev
->promiscuity
-= inc
;
6009 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
6014 if (dev
->flags
!= old_flags
) {
6015 pr_info("device %s %s promiscuous mode\n",
6017 dev
->flags
& IFF_PROMISC
? "entered" : "left");
6018 if (audit_enabled
) {
6019 current_uid_gid(&uid
, &gid
);
6020 audit_log(current
->audit_context
, GFP_ATOMIC
,
6021 AUDIT_ANOM_PROMISCUOUS
,
6022 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
6023 dev
->name
, (dev
->flags
& IFF_PROMISC
),
6024 (old_flags
& IFF_PROMISC
),
6025 from_kuid(&init_user_ns
, audit_get_loginuid(current
)),
6026 from_kuid(&init_user_ns
, uid
),
6027 from_kgid(&init_user_ns
, gid
),
6028 audit_get_sessionid(current
));
6031 dev_change_rx_flags(dev
, IFF_PROMISC
);
6034 __dev_notify_flags(dev
, old_flags
, IFF_PROMISC
);
6039 * dev_set_promiscuity - update promiscuity count on a device
6043 * Add or remove promiscuity from a device. While the count in the device
6044 * remains above zero the interface remains promiscuous. Once it hits zero
6045 * the device reverts back to normal filtering operation. A negative inc
6046 * value is used to drop promiscuity on the device.
6047 * Return 0 if successful or a negative errno code on error.
6049 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
6051 unsigned int old_flags
= dev
->flags
;
6054 err
= __dev_set_promiscuity(dev
, inc
, true);
6057 if (dev
->flags
!= old_flags
)
6058 dev_set_rx_mode(dev
);
6061 EXPORT_SYMBOL(dev_set_promiscuity
);
6063 static int __dev_set_allmulti(struct net_device
*dev
, int inc
, bool notify
)
6065 unsigned int old_flags
= dev
->flags
, old_gflags
= dev
->gflags
;
6069 dev
->flags
|= IFF_ALLMULTI
;
6070 dev
->allmulti
+= inc
;
6071 if (dev
->allmulti
== 0) {
6074 * If inc causes overflow, untouch allmulti and return error.
6077 dev
->flags
&= ~IFF_ALLMULTI
;
6079 dev
->allmulti
-= inc
;
6080 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
6085 if (dev
->flags
^ old_flags
) {
6086 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
6087 dev_set_rx_mode(dev
);
6089 __dev_notify_flags(dev
, old_flags
,
6090 dev
->gflags
^ old_gflags
);
6096 * dev_set_allmulti - update allmulti count on a device
6100 * Add or remove reception of all multicast frames to a device. While the
6101 * count in the device remains above zero the interface remains listening
6102 * to all interfaces. Once it hits zero the device reverts back to normal
6103 * filtering operation. A negative @inc value is used to drop the counter
6104 * when releasing a resource needing all multicasts.
6105 * Return 0 if successful or a negative errno code on error.
6108 int dev_set_allmulti(struct net_device
*dev
, int inc
)
6110 return __dev_set_allmulti(dev
, inc
, true);
6112 EXPORT_SYMBOL(dev_set_allmulti
);
6115 * Upload unicast and multicast address lists to device and
6116 * configure RX filtering. When the device doesn't support unicast
6117 * filtering it is put in promiscuous mode while unicast addresses
6120 void __dev_set_rx_mode(struct net_device
*dev
)
6122 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6124 /* dev_open will call this function so the list will stay sane. */
6125 if (!(dev
->flags
&IFF_UP
))
6128 if (!netif_device_present(dev
))
6131 if (!(dev
->priv_flags
& IFF_UNICAST_FLT
)) {
6132 /* Unicast addresses changes may only happen under the rtnl,
6133 * therefore calling __dev_set_promiscuity here is safe.
6135 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
6136 __dev_set_promiscuity(dev
, 1, false);
6137 dev
->uc_promisc
= true;
6138 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
6139 __dev_set_promiscuity(dev
, -1, false);
6140 dev
->uc_promisc
= false;
6144 if (ops
->ndo_set_rx_mode
)
6145 ops
->ndo_set_rx_mode(dev
);
6148 void dev_set_rx_mode(struct net_device
*dev
)
6150 netif_addr_lock_bh(dev
);
6151 __dev_set_rx_mode(dev
);
6152 netif_addr_unlock_bh(dev
);
6156 * dev_get_flags - get flags reported to userspace
6159 * Get the combination of flag bits exported through APIs to userspace.
6161 unsigned int dev_get_flags(const struct net_device
*dev
)
6165 flags
= (dev
->flags
& ~(IFF_PROMISC
|
6170 (dev
->gflags
& (IFF_PROMISC
|
6173 if (netif_running(dev
)) {
6174 if (netif_oper_up(dev
))
6175 flags
|= IFF_RUNNING
;
6176 if (netif_carrier_ok(dev
))
6177 flags
|= IFF_LOWER_UP
;
6178 if (netif_dormant(dev
))
6179 flags
|= IFF_DORMANT
;
6184 EXPORT_SYMBOL(dev_get_flags
);
6186 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
6188 unsigned int old_flags
= dev
->flags
;
6194 * Set the flags on our device.
6197 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
6198 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
6200 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
6204 * Load in the correct multicast list now the flags have changed.
6207 if ((old_flags
^ flags
) & IFF_MULTICAST
)
6208 dev_change_rx_flags(dev
, IFF_MULTICAST
);
6210 dev_set_rx_mode(dev
);
6213 * Have we downed the interface. We handle IFF_UP ourselves
6214 * according to user attempts to set it, rather than blindly
6219 if ((old_flags
^ flags
) & IFF_UP
)
6220 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
6222 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
6223 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
6224 unsigned int old_flags
= dev
->flags
;
6226 dev
->gflags
^= IFF_PROMISC
;
6228 if (__dev_set_promiscuity(dev
, inc
, false) >= 0)
6229 if (dev
->flags
!= old_flags
)
6230 dev_set_rx_mode(dev
);
6233 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
6234 is important. Some (broken) drivers set IFF_PROMISC, when
6235 IFF_ALLMULTI is requested not asking us and not reporting.
6237 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
6238 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
6240 dev
->gflags
^= IFF_ALLMULTI
;
6241 __dev_set_allmulti(dev
, inc
, false);
6247 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
,
6248 unsigned int gchanges
)
6250 unsigned int changes
= dev
->flags
^ old_flags
;
6253 rtmsg_ifinfo(RTM_NEWLINK
, dev
, gchanges
, GFP_ATOMIC
);
6255 if (changes
& IFF_UP
) {
6256 if (dev
->flags
& IFF_UP
)
6257 call_netdevice_notifiers(NETDEV_UP
, dev
);
6259 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
6262 if (dev
->flags
& IFF_UP
&&
6263 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
))) {
6264 struct netdev_notifier_change_info change_info
;
6266 change_info
.flags_changed
= changes
;
6267 call_netdevice_notifiers_info(NETDEV_CHANGE
, dev
,
6273 * dev_change_flags - change device settings
6275 * @flags: device state flags
6277 * Change settings on device based state flags. The flags are
6278 * in the userspace exported format.
6280 int dev_change_flags(struct net_device
*dev
, unsigned int flags
)
6283 unsigned int changes
, old_flags
= dev
->flags
, old_gflags
= dev
->gflags
;
6285 ret
= __dev_change_flags(dev
, flags
);
6289 changes
= (old_flags
^ dev
->flags
) | (old_gflags
^ dev
->gflags
);
6290 __dev_notify_flags(dev
, old_flags
, changes
);
6293 EXPORT_SYMBOL(dev_change_flags
);
6295 static int __dev_set_mtu(struct net_device
*dev
, int new_mtu
)
6297 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6299 if (ops
->ndo_change_mtu
)
6300 return ops
->ndo_change_mtu(dev
, new_mtu
);
6307 * dev_set_mtu - Change maximum transfer unit
6309 * @new_mtu: new transfer unit
6311 * Change the maximum transfer size of the network device.
6313 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
6317 if (new_mtu
== dev
->mtu
)
6320 /* MTU must be positive. */
6324 if (!netif_device_present(dev
))
6327 err
= call_netdevice_notifiers(NETDEV_PRECHANGEMTU
, dev
);
6328 err
= notifier_to_errno(err
);
6332 orig_mtu
= dev
->mtu
;
6333 err
= __dev_set_mtu(dev
, new_mtu
);
6336 err
= call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
6337 err
= notifier_to_errno(err
);
6339 /* setting mtu back and notifying everyone again,
6340 * so that they have a chance to revert changes.
6342 __dev_set_mtu(dev
, orig_mtu
);
6343 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
6348 EXPORT_SYMBOL(dev_set_mtu
);
6351 * dev_set_group - Change group this device belongs to
6353 * @new_group: group this device should belong to
6355 void dev_set_group(struct net_device
*dev
, int new_group
)
6357 dev
->group
= new_group
;
6359 EXPORT_SYMBOL(dev_set_group
);
6362 * dev_set_mac_address - Change Media Access Control Address
6366 * Change the hardware (MAC) address of the device
6368 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
6370 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6373 if (!ops
->ndo_set_mac_address
)
6375 if (sa
->sa_family
!= dev
->type
)
6377 if (!netif_device_present(dev
))
6379 err
= ops
->ndo_set_mac_address(dev
, sa
);
6382 dev
->addr_assign_type
= NET_ADDR_SET
;
6383 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
6384 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
6387 EXPORT_SYMBOL(dev_set_mac_address
);
6390 * dev_change_carrier - Change device carrier
6392 * @new_carrier: new value
6394 * Change device carrier
6396 int dev_change_carrier(struct net_device
*dev
, bool new_carrier
)
6398 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6400 if (!ops
->ndo_change_carrier
)
6402 if (!netif_device_present(dev
))
6404 return ops
->ndo_change_carrier(dev
, new_carrier
);
6406 EXPORT_SYMBOL(dev_change_carrier
);
6409 * dev_get_phys_port_id - Get device physical port ID
6413 * Get device physical port ID
6415 int dev_get_phys_port_id(struct net_device
*dev
,
6416 struct netdev_phys_item_id
*ppid
)
6418 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6420 if (!ops
->ndo_get_phys_port_id
)
6422 return ops
->ndo_get_phys_port_id(dev
, ppid
);
6424 EXPORT_SYMBOL(dev_get_phys_port_id
);
6427 * dev_get_phys_port_name - Get device physical port name
6431 * Get device physical port name
6433 int dev_get_phys_port_name(struct net_device
*dev
,
6434 char *name
, size_t len
)
6436 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6438 if (!ops
->ndo_get_phys_port_name
)
6440 return ops
->ndo_get_phys_port_name(dev
, name
, len
);
6442 EXPORT_SYMBOL(dev_get_phys_port_name
);
6445 * dev_change_proto_down - update protocol port state information
6447 * @proto_down: new value
6449 * This info can be used by switch drivers to set the phys state of the
6452 int dev_change_proto_down(struct net_device
*dev
, bool proto_down
)
6454 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6456 if (!ops
->ndo_change_proto_down
)
6458 if (!netif_device_present(dev
))
6460 return ops
->ndo_change_proto_down(dev
, proto_down
);
6462 EXPORT_SYMBOL(dev_change_proto_down
);
6465 * dev_new_index - allocate an ifindex
6466 * @net: the applicable net namespace
6468 * Returns a suitable unique value for a new device interface
6469 * number. The caller must hold the rtnl semaphore or the
6470 * dev_base_lock to be sure it remains unique.
6472 static int dev_new_index(struct net
*net
)
6474 int ifindex
= net
->ifindex
;
6478 if (!__dev_get_by_index(net
, ifindex
))
6479 return net
->ifindex
= ifindex
;
6483 /* Delayed registration/unregisteration */
6484 static LIST_HEAD(net_todo_list
);
6485 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq
);
6487 static void net_set_todo(struct net_device
*dev
)
6489 list_add_tail(&dev
->todo_list
, &net_todo_list
);
6490 dev_net(dev
)->dev_unreg_count
++;
6493 static void rollback_registered_many(struct list_head
*head
)
6495 struct net_device
*dev
, *tmp
;
6496 LIST_HEAD(close_head
);
6498 BUG_ON(dev_boot_phase
);
6501 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
6502 /* Some devices call without registering
6503 * for initialization unwind. Remove those
6504 * devices and proceed with the remaining.
6506 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
6507 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
6511 list_del(&dev
->unreg_list
);
6514 dev
->dismantle
= true;
6515 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
6518 /* If device is running, close it first. */
6519 list_for_each_entry(dev
, head
, unreg_list
)
6520 list_add_tail(&dev
->close_list
, &close_head
);
6521 dev_close_many(&close_head
, true);
6523 list_for_each_entry(dev
, head
, unreg_list
) {
6524 /* And unlink it from device chain. */
6525 unlist_netdevice(dev
);
6527 dev
->reg_state
= NETREG_UNREGISTERING
;
6528 on_each_cpu(flush_backlog
, dev
, 1);
6533 list_for_each_entry(dev
, head
, unreg_list
) {
6534 struct sk_buff
*skb
= NULL
;
6536 /* Shutdown queueing discipline. */
6540 /* Notify protocols, that we are about to destroy
6541 this device. They should clean all the things.
6543 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6545 if (!dev
->rtnl_link_ops
||
6546 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
6547 skb
= rtmsg_ifinfo_build_skb(RTM_DELLINK
, dev
, ~0U,
6551 * Flush the unicast and multicast chains
6556 if (dev
->netdev_ops
->ndo_uninit
)
6557 dev
->netdev_ops
->ndo_uninit(dev
);
6560 rtmsg_ifinfo_send(skb
, dev
, GFP_KERNEL
);
6562 /* Notifier chain MUST detach us all upper devices. */
6563 WARN_ON(netdev_has_any_upper_dev(dev
));
6565 /* Remove entries from kobject tree */
6566 netdev_unregister_kobject(dev
);
6568 /* Remove XPS queueing entries */
6569 netif_reset_xps_queues_gt(dev
, 0);
6575 list_for_each_entry(dev
, head
, unreg_list
)
6579 static void rollback_registered(struct net_device
*dev
)
6583 list_add(&dev
->unreg_list
, &single
);
6584 rollback_registered_many(&single
);
6588 static netdev_features_t
netdev_sync_upper_features(struct net_device
*lower
,
6589 struct net_device
*upper
, netdev_features_t features
)
6591 netdev_features_t upper_disables
= NETIF_F_UPPER_DISABLES
;
6592 netdev_features_t feature
;
6595 for_each_netdev_feature(&upper_disables
, feature_bit
) {
6596 feature
= __NETIF_F_BIT(feature_bit
);
6597 if (!(upper
->wanted_features
& feature
)
6598 && (features
& feature
)) {
6599 netdev_dbg(lower
, "Dropping feature %pNF, upper dev %s has it off.\n",
6600 &feature
, upper
->name
);
6601 features
&= ~feature
;
6608 static void netdev_sync_lower_features(struct net_device
*upper
,
6609 struct net_device
*lower
, netdev_features_t features
)
6611 netdev_features_t upper_disables
= NETIF_F_UPPER_DISABLES
;
6612 netdev_features_t feature
;
6615 for_each_netdev_feature(&upper_disables
, feature_bit
) {
6616 feature
= __NETIF_F_BIT(feature_bit
);
6617 if (!(features
& feature
) && (lower
->features
& feature
)) {
6618 netdev_dbg(upper
, "Disabling feature %pNF on lower dev %s.\n",
6619 &feature
, lower
->name
);
6620 lower
->wanted_features
&= ~feature
;
6621 netdev_update_features(lower
);
6623 if (unlikely(lower
->features
& feature
))
6624 netdev_WARN(upper
, "failed to disable %pNF on %s!\n",
6625 &feature
, lower
->name
);
6630 static netdev_features_t
netdev_fix_features(struct net_device
*dev
,
6631 netdev_features_t features
)
6633 /* Fix illegal checksum combinations */
6634 if ((features
& NETIF_F_HW_CSUM
) &&
6635 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
6636 netdev_warn(dev
, "mixed HW and IP checksum settings.\n");
6637 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
6640 /* TSO requires that SG is present as well. */
6641 if ((features
& NETIF_F_ALL_TSO
) && !(features
& NETIF_F_SG
)) {
6642 netdev_dbg(dev
, "Dropping TSO features since no SG feature.\n");
6643 features
&= ~NETIF_F_ALL_TSO
;
6646 if ((features
& NETIF_F_TSO
) && !(features
& NETIF_F_HW_CSUM
) &&
6647 !(features
& NETIF_F_IP_CSUM
)) {
6648 netdev_dbg(dev
, "Dropping TSO features since no CSUM feature.\n");
6649 features
&= ~NETIF_F_TSO
;
6650 features
&= ~NETIF_F_TSO_ECN
;
6653 if ((features
& NETIF_F_TSO6
) && !(features
& NETIF_F_HW_CSUM
) &&
6654 !(features
& NETIF_F_IPV6_CSUM
)) {
6655 netdev_dbg(dev
, "Dropping TSO6 features since no CSUM feature.\n");
6656 features
&= ~NETIF_F_TSO6
;
6659 /* TSO ECN requires that TSO is present as well. */
6660 if ((features
& NETIF_F_ALL_TSO
) == NETIF_F_TSO_ECN
)
6661 features
&= ~NETIF_F_TSO_ECN
;
6663 /* Software GSO depends on SG. */
6664 if ((features
& NETIF_F_GSO
) && !(features
& NETIF_F_SG
)) {
6665 netdev_dbg(dev
, "Dropping NETIF_F_GSO since no SG feature.\n");
6666 features
&= ~NETIF_F_GSO
;
6669 /* UFO needs SG and checksumming */
6670 if (features
& NETIF_F_UFO
) {
6671 /* maybe split UFO into V4 and V6? */
6672 if (!(features
& NETIF_F_HW_CSUM
) &&
6673 ((features
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
)) !=
6674 (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
))) {
6676 "Dropping NETIF_F_UFO since no checksum offload features.\n");
6677 features
&= ~NETIF_F_UFO
;
6680 if (!(features
& NETIF_F_SG
)) {
6682 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
6683 features
&= ~NETIF_F_UFO
;
6687 #ifdef CONFIG_NET_RX_BUSY_POLL
6688 if (dev
->netdev_ops
->ndo_busy_poll
)
6689 features
|= NETIF_F_BUSY_POLL
;
6692 features
&= ~NETIF_F_BUSY_POLL
;
6697 int __netdev_update_features(struct net_device
*dev
)
6699 struct net_device
*upper
, *lower
;
6700 netdev_features_t features
;
6701 struct list_head
*iter
;
6706 features
= netdev_get_wanted_features(dev
);
6708 if (dev
->netdev_ops
->ndo_fix_features
)
6709 features
= dev
->netdev_ops
->ndo_fix_features(dev
, features
);
6711 /* driver might be less strict about feature dependencies */
6712 features
= netdev_fix_features(dev
, features
);
6714 /* some features can't be enabled if they're off an an upper device */
6715 netdev_for_each_upper_dev_rcu(dev
, upper
, iter
)
6716 features
= netdev_sync_upper_features(dev
, upper
, features
);
6718 if (dev
->features
== features
)
6721 netdev_dbg(dev
, "Features changed: %pNF -> %pNF\n",
6722 &dev
->features
, &features
);
6724 if (dev
->netdev_ops
->ndo_set_features
)
6725 err
= dev
->netdev_ops
->ndo_set_features(dev
, features
);
6729 if (unlikely(err
< 0)) {
6731 "set_features() failed (%d); wanted %pNF, left %pNF\n",
6732 err
, &features
, &dev
->features
);
6733 /* return non-0 since some features might have changed and
6734 * it's better to fire a spurious notification than miss it
6740 /* some features must be disabled on lower devices when disabled
6741 * on an upper device (think: bonding master or bridge)
6743 netdev_for_each_lower_dev(dev
, lower
, iter
)
6744 netdev_sync_lower_features(dev
, lower
, features
);
6747 dev
->features
= features
;
6749 return err
< 0 ? 0 : 1;
6753 * netdev_update_features - recalculate device features
6754 * @dev: the device to check
6756 * Recalculate dev->features set and send notifications if it
6757 * has changed. Should be called after driver or hardware dependent
6758 * conditions might have changed that influence the features.
6760 void netdev_update_features(struct net_device
*dev
)
6762 if (__netdev_update_features(dev
))
6763 netdev_features_change(dev
);
6765 EXPORT_SYMBOL(netdev_update_features
);
6768 * netdev_change_features - recalculate device features
6769 * @dev: the device to check
6771 * Recalculate dev->features set and send notifications even
6772 * if they have not changed. Should be called instead of
6773 * netdev_update_features() if also dev->vlan_features might
6774 * have changed to allow the changes to be propagated to stacked
6777 void netdev_change_features(struct net_device
*dev
)
6779 __netdev_update_features(dev
);
6780 netdev_features_change(dev
);
6782 EXPORT_SYMBOL(netdev_change_features
);
6785 * netif_stacked_transfer_operstate - transfer operstate
6786 * @rootdev: the root or lower level device to transfer state from
6787 * @dev: the device to transfer operstate to
6789 * Transfer operational state from root to device. This is normally
6790 * called when a stacking relationship exists between the root
6791 * device and the device(a leaf device).
6793 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
6794 struct net_device
*dev
)
6796 if (rootdev
->operstate
== IF_OPER_DORMANT
)
6797 netif_dormant_on(dev
);
6799 netif_dormant_off(dev
);
6801 if (netif_carrier_ok(rootdev
)) {
6802 if (!netif_carrier_ok(dev
))
6803 netif_carrier_on(dev
);
6805 if (netif_carrier_ok(dev
))
6806 netif_carrier_off(dev
);
6809 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
6812 static int netif_alloc_rx_queues(struct net_device
*dev
)
6814 unsigned int i
, count
= dev
->num_rx_queues
;
6815 struct netdev_rx_queue
*rx
;
6816 size_t sz
= count
* sizeof(*rx
);
6820 rx
= kzalloc(sz
, GFP_KERNEL
| __GFP_NOWARN
| __GFP_REPEAT
);
6828 for (i
= 0; i
< count
; i
++)
6834 static void netdev_init_one_queue(struct net_device
*dev
,
6835 struct netdev_queue
*queue
, void *_unused
)
6837 /* Initialize queue lock */
6838 spin_lock_init(&queue
->_xmit_lock
);
6839 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
6840 queue
->xmit_lock_owner
= -1;
6841 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
6844 dql_init(&queue
->dql
, HZ
);
6848 static void netif_free_tx_queues(struct net_device
*dev
)
6853 static int netif_alloc_netdev_queues(struct net_device
*dev
)
6855 unsigned int count
= dev
->num_tx_queues
;
6856 struct netdev_queue
*tx
;
6857 size_t sz
= count
* sizeof(*tx
);
6859 if (count
< 1 || count
> 0xffff)
6862 tx
= kzalloc(sz
, GFP_KERNEL
| __GFP_NOWARN
| __GFP_REPEAT
);
6870 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
6871 spin_lock_init(&dev
->tx_global_lock
);
6876 void netif_tx_stop_all_queues(struct net_device
*dev
)
6880 for (i
= 0; i
< dev
->num_tx_queues
; i
++) {
6881 struct netdev_queue
*txq
= netdev_get_tx_queue(dev
, i
);
6882 netif_tx_stop_queue(txq
);
6885 EXPORT_SYMBOL(netif_tx_stop_all_queues
);
6888 * register_netdevice - register a network device
6889 * @dev: device to register
6891 * Take a completed network device structure and add it to the kernel
6892 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6893 * chain. 0 is returned on success. A negative errno code is returned
6894 * on a failure to set up the device, or if the name is a duplicate.
6896 * Callers must hold the rtnl semaphore. You may want
6897 * register_netdev() instead of this.
6900 * The locking appears insufficient to guarantee two parallel registers
6901 * will not get the same name.
6904 int register_netdevice(struct net_device
*dev
)
6907 struct net
*net
= dev_net(dev
);
6909 BUG_ON(dev_boot_phase
);
6914 /* When net_device's are persistent, this will be fatal. */
6915 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
6918 spin_lock_init(&dev
->addr_list_lock
);
6919 netdev_set_addr_lockdep_class(dev
);
6921 ret
= dev_get_valid_name(net
, dev
, dev
->name
);
6925 /* Init, if this function is available */
6926 if (dev
->netdev_ops
->ndo_init
) {
6927 ret
= dev
->netdev_ops
->ndo_init(dev
);
6935 if (((dev
->hw_features
| dev
->features
) &
6936 NETIF_F_HW_VLAN_CTAG_FILTER
) &&
6937 (!dev
->netdev_ops
->ndo_vlan_rx_add_vid
||
6938 !dev
->netdev_ops
->ndo_vlan_rx_kill_vid
)) {
6939 netdev_WARN(dev
, "Buggy VLAN acceleration in driver!\n");
6946 dev
->ifindex
= dev_new_index(net
);
6947 else if (__dev_get_by_index(net
, dev
->ifindex
))
6950 /* Transfer changeable features to wanted_features and enable
6951 * software offloads (GSO and GRO).
6953 dev
->hw_features
|= NETIF_F_SOFT_FEATURES
;
6954 dev
->features
|= NETIF_F_SOFT_FEATURES
;
6955 dev
->wanted_features
= dev
->features
& dev
->hw_features
;
6957 if (!(dev
->flags
& IFF_LOOPBACK
)) {
6958 dev
->hw_features
|= NETIF_F_NOCACHE_COPY
;
6961 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6963 dev
->vlan_features
|= NETIF_F_HIGHDMA
;
6965 /* Make NETIF_F_SG inheritable to tunnel devices.
6967 dev
->hw_enc_features
|= NETIF_F_SG
;
6969 /* Make NETIF_F_SG inheritable to MPLS.
6971 dev
->mpls_features
|= NETIF_F_SG
;
6973 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
6974 ret
= notifier_to_errno(ret
);
6978 ret
= netdev_register_kobject(dev
);
6981 dev
->reg_state
= NETREG_REGISTERED
;
6983 __netdev_update_features(dev
);
6986 * Default initial state at registry is that the
6987 * device is present.
6990 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
6992 linkwatch_init_dev(dev
);
6994 dev_init_scheduler(dev
);
6996 list_netdevice(dev
);
6997 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
6999 /* If the device has permanent device address, driver should
7000 * set dev_addr and also addr_assign_type should be set to
7001 * NET_ADDR_PERM (default value).
7003 if (dev
->addr_assign_type
== NET_ADDR_PERM
)
7004 memcpy(dev
->perm_addr
, dev
->dev_addr
, dev
->addr_len
);
7006 /* Notify protocols, that a new device appeared. */
7007 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
7008 ret
= notifier_to_errno(ret
);
7010 rollback_registered(dev
);
7011 dev
->reg_state
= NETREG_UNREGISTERED
;
7014 * Prevent userspace races by waiting until the network
7015 * device is fully setup before sending notifications.
7017 if (!dev
->rtnl_link_ops
||
7018 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
7019 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U, GFP_KERNEL
);
7025 if (dev
->netdev_ops
->ndo_uninit
)
7026 dev
->netdev_ops
->ndo_uninit(dev
);
7029 EXPORT_SYMBOL(register_netdevice
);
7032 * init_dummy_netdev - init a dummy network device for NAPI
7033 * @dev: device to init
7035 * This takes a network device structure and initialize the minimum
7036 * amount of fields so it can be used to schedule NAPI polls without
7037 * registering a full blown interface. This is to be used by drivers
7038 * that need to tie several hardware interfaces to a single NAPI
7039 * poll scheduler due to HW limitations.
7041 int init_dummy_netdev(struct net_device
*dev
)
7043 /* Clear everything. Note we don't initialize spinlocks
7044 * are they aren't supposed to be taken by any of the
7045 * NAPI code and this dummy netdev is supposed to be
7046 * only ever used for NAPI polls
7048 memset(dev
, 0, sizeof(struct net_device
));
7050 /* make sure we BUG if trying to hit standard
7051 * register/unregister code path
7053 dev
->reg_state
= NETREG_DUMMY
;
7055 /* NAPI wants this */
7056 INIT_LIST_HEAD(&dev
->napi_list
);
7058 /* a dummy interface is started by default */
7059 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
7060 set_bit(__LINK_STATE_START
, &dev
->state
);
7062 /* Note : We dont allocate pcpu_refcnt for dummy devices,
7063 * because users of this 'device' dont need to change
7069 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
7073 * register_netdev - register a network device
7074 * @dev: device to register
7076 * Take a completed network device structure and add it to the kernel
7077 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
7078 * chain. 0 is returned on success. A negative errno code is returned
7079 * on a failure to set up the device, or if the name is a duplicate.
7081 * This is a wrapper around register_netdevice that takes the rtnl semaphore
7082 * and expands the device name if you passed a format string to
7085 int register_netdev(struct net_device
*dev
)
7090 err
= register_netdevice(dev
);
7094 EXPORT_SYMBOL(register_netdev
);
7096 int netdev_refcnt_read(const struct net_device
*dev
)
7100 for_each_possible_cpu(i
)
7101 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
7104 EXPORT_SYMBOL(netdev_refcnt_read
);
7107 * netdev_wait_allrefs - wait until all references are gone.
7108 * @dev: target net_device
7110 * This is called when unregistering network devices.
7112 * Any protocol or device that holds a reference should register
7113 * for netdevice notification, and cleanup and put back the
7114 * reference if they receive an UNREGISTER event.
7115 * We can get stuck here if buggy protocols don't correctly
7118 static void netdev_wait_allrefs(struct net_device
*dev
)
7120 unsigned long rebroadcast_time
, warning_time
;
7123 linkwatch_forget_dev(dev
);
7125 rebroadcast_time
= warning_time
= jiffies
;
7126 refcnt
= netdev_refcnt_read(dev
);
7128 while (refcnt
!= 0) {
7129 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
7132 /* Rebroadcast unregister notification */
7133 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
7139 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
7140 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
7142 /* We must not have linkwatch events
7143 * pending on unregister. If this
7144 * happens, we simply run the queue
7145 * unscheduled, resulting in a noop
7148 linkwatch_run_queue();
7153 rebroadcast_time
= jiffies
;
7158 refcnt
= netdev_refcnt_read(dev
);
7160 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
7161 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
7163 warning_time
= jiffies
;
7172 * register_netdevice(x1);
7173 * register_netdevice(x2);
7175 * unregister_netdevice(y1);
7176 * unregister_netdevice(y2);
7182 * We are invoked by rtnl_unlock().
7183 * This allows us to deal with problems:
7184 * 1) We can delete sysfs objects which invoke hotplug
7185 * without deadlocking with linkwatch via keventd.
7186 * 2) Since we run with the RTNL semaphore not held, we can sleep
7187 * safely in order to wait for the netdev refcnt to drop to zero.
7189 * We must not return until all unregister events added during
7190 * the interval the lock was held have been completed.
7192 void netdev_run_todo(void)
7194 struct list_head list
;
7196 /* Snapshot list, allow later requests */
7197 list_replace_init(&net_todo_list
, &list
);
7202 /* Wait for rcu callbacks to finish before next phase */
7203 if (!list_empty(&list
))
7206 while (!list_empty(&list
)) {
7207 struct net_device
*dev
7208 = list_first_entry(&list
, struct net_device
, todo_list
);
7209 list_del(&dev
->todo_list
);
7212 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
7215 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
7216 pr_err("network todo '%s' but state %d\n",
7217 dev
->name
, dev
->reg_state
);
7222 dev
->reg_state
= NETREG_UNREGISTERED
;
7224 netdev_wait_allrefs(dev
);
7227 BUG_ON(netdev_refcnt_read(dev
));
7228 BUG_ON(!list_empty(&dev
->ptype_all
));
7229 BUG_ON(!list_empty(&dev
->ptype_specific
));
7230 WARN_ON(rcu_access_pointer(dev
->ip_ptr
));
7231 WARN_ON(rcu_access_pointer(dev
->ip6_ptr
));
7232 WARN_ON(dev
->dn_ptr
);
7234 if (dev
->destructor
)
7235 dev
->destructor(dev
);
7237 /* Report a network device has been unregistered */
7239 dev_net(dev
)->dev_unreg_count
--;
7241 wake_up(&netdev_unregistering_wq
);
7243 /* Free network device */
7244 kobject_put(&dev
->dev
.kobj
);
7248 /* Convert net_device_stats to rtnl_link_stats64. They have the same
7249 * fields in the same order, with only the type differing.
7251 void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
7252 const struct net_device_stats
*netdev_stats
)
7254 #if BITS_PER_LONG == 64
7255 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
7256 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
7258 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
7259 const unsigned long *src
= (const unsigned long *)netdev_stats
;
7260 u64
*dst
= (u64
*)stats64
;
7262 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
7263 sizeof(*stats64
) / sizeof(u64
));
7264 for (i
= 0; i
< n
; i
++)
7268 EXPORT_SYMBOL(netdev_stats_to_stats64
);
7271 * dev_get_stats - get network device statistics
7272 * @dev: device to get statistics from
7273 * @storage: place to store stats
7275 * Get network statistics from device. Return @storage.
7276 * The device driver may provide its own method by setting
7277 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
7278 * otherwise the internal statistics structure is used.
7280 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
7281 struct rtnl_link_stats64
*storage
)
7283 const struct net_device_ops
*ops
= dev
->netdev_ops
;
7285 if (ops
->ndo_get_stats64
) {
7286 memset(storage
, 0, sizeof(*storage
));
7287 ops
->ndo_get_stats64(dev
, storage
);
7288 } else if (ops
->ndo_get_stats
) {
7289 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
7291 netdev_stats_to_stats64(storage
, &dev
->stats
);
7293 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
7294 storage
->tx_dropped
+= atomic_long_read(&dev
->tx_dropped
);
7297 EXPORT_SYMBOL(dev_get_stats
);
7299 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
7301 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
7303 #ifdef CONFIG_NET_CLS_ACT
7306 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
7309 netdev_init_one_queue(dev
, queue
, NULL
);
7310 RCU_INIT_POINTER(queue
->qdisc
, &noop_qdisc
);
7311 queue
->qdisc_sleeping
= &noop_qdisc
;
7312 rcu_assign_pointer(dev
->ingress_queue
, queue
);
7317 static const struct ethtool_ops default_ethtool_ops
;
7319 void netdev_set_default_ethtool_ops(struct net_device
*dev
,
7320 const struct ethtool_ops
*ops
)
7322 if (dev
->ethtool_ops
== &default_ethtool_ops
)
7323 dev
->ethtool_ops
= ops
;
7325 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops
);
7327 void netdev_freemem(struct net_device
*dev
)
7329 char *addr
= (char *)dev
- dev
->padded
;
7335 * alloc_netdev_mqs - allocate network device
7336 * @sizeof_priv: size of private data to allocate space for
7337 * @name: device name format string
7338 * @name_assign_type: origin of device name
7339 * @setup: callback to initialize device
7340 * @txqs: the number of TX subqueues to allocate
7341 * @rxqs: the number of RX subqueues to allocate
7343 * Allocates a struct net_device with private data area for driver use
7344 * and performs basic initialization. Also allocates subqueue structs
7345 * for each queue on the device.
7347 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
7348 unsigned char name_assign_type
,
7349 void (*setup
)(struct net_device
*),
7350 unsigned int txqs
, unsigned int rxqs
)
7352 struct net_device
*dev
;
7354 struct net_device
*p
;
7356 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
7359 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
7365 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
7370 alloc_size
= sizeof(struct net_device
);
7372 /* ensure 32-byte alignment of private area */
7373 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
7374 alloc_size
+= sizeof_priv
;
7376 /* ensure 32-byte alignment of whole construct */
7377 alloc_size
+= NETDEV_ALIGN
- 1;
7379 p
= kzalloc(alloc_size
, GFP_KERNEL
| __GFP_NOWARN
| __GFP_REPEAT
);
7381 p
= vzalloc(alloc_size
);
7385 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
7386 dev
->padded
= (char *)dev
- (char *)p
;
7388 dev
->pcpu_refcnt
= alloc_percpu(int);
7389 if (!dev
->pcpu_refcnt
)
7392 if (dev_addr_init(dev
))
7398 dev_net_set(dev
, &init_net
);
7400 dev
->gso_max_size
= GSO_MAX_SIZE
;
7401 dev
->gso_max_segs
= GSO_MAX_SEGS
;
7402 dev
->gso_min_segs
= 0;
7404 INIT_LIST_HEAD(&dev
->napi_list
);
7405 INIT_LIST_HEAD(&dev
->unreg_list
);
7406 INIT_LIST_HEAD(&dev
->close_list
);
7407 INIT_LIST_HEAD(&dev
->link_watch_list
);
7408 INIT_LIST_HEAD(&dev
->adj_list
.upper
);
7409 INIT_LIST_HEAD(&dev
->adj_list
.lower
);
7410 INIT_LIST_HEAD(&dev
->all_adj_list
.upper
);
7411 INIT_LIST_HEAD(&dev
->all_adj_list
.lower
);
7412 INIT_LIST_HEAD(&dev
->ptype_all
);
7413 INIT_LIST_HEAD(&dev
->ptype_specific
);
7414 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
| IFF_XMIT_DST_RELEASE_PERM
;
7417 if (!dev
->tx_queue_len
)
7418 dev
->priv_flags
|= IFF_NO_QUEUE
;
7420 dev
->num_tx_queues
= txqs
;
7421 dev
->real_num_tx_queues
= txqs
;
7422 if (netif_alloc_netdev_queues(dev
))
7426 dev
->num_rx_queues
= rxqs
;
7427 dev
->real_num_rx_queues
= rxqs
;
7428 if (netif_alloc_rx_queues(dev
))
7432 strcpy(dev
->name
, name
);
7433 dev
->name_assign_type
= name_assign_type
;
7434 dev
->group
= INIT_NETDEV_GROUP
;
7435 if (!dev
->ethtool_ops
)
7436 dev
->ethtool_ops
= &default_ethtool_ops
;
7438 nf_hook_ingress_init(dev
);
7447 free_percpu(dev
->pcpu_refcnt
);
7449 netdev_freemem(dev
);
7452 EXPORT_SYMBOL(alloc_netdev_mqs
);
7455 * free_netdev - free network device
7458 * This function does the last stage of destroying an allocated device
7459 * interface. The reference to the device object is released.
7460 * If this is the last reference then it will be freed.
7461 * Must be called in process context.
7463 void free_netdev(struct net_device
*dev
)
7465 struct napi_struct
*p
, *n
;
7468 netif_free_tx_queues(dev
);
7473 kfree(rcu_dereference_protected(dev
->ingress_queue
, 1));
7475 /* Flush device addresses */
7476 dev_addr_flush(dev
);
7478 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
7481 free_percpu(dev
->pcpu_refcnt
);
7482 dev
->pcpu_refcnt
= NULL
;
7484 /* Compatibility with error handling in drivers */
7485 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
7486 netdev_freemem(dev
);
7490 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
7491 dev
->reg_state
= NETREG_RELEASED
;
7493 /* will free via device release */
7494 put_device(&dev
->dev
);
7496 EXPORT_SYMBOL(free_netdev
);
7499 * synchronize_net - Synchronize with packet receive processing
7501 * Wait for packets currently being received to be done.
7502 * Does not block later packets from starting.
7504 void synchronize_net(void)
7507 if (rtnl_is_locked())
7508 synchronize_rcu_expedited();
7512 EXPORT_SYMBOL(synchronize_net
);
7515 * unregister_netdevice_queue - remove device from the kernel
7519 * This function shuts down a device interface and removes it
7520 * from the kernel tables.
7521 * If head not NULL, device is queued to be unregistered later.
7523 * Callers must hold the rtnl semaphore. You may want
7524 * unregister_netdev() instead of this.
7527 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
7532 list_move_tail(&dev
->unreg_list
, head
);
7534 rollback_registered(dev
);
7535 /* Finish processing unregister after unlock */
7539 EXPORT_SYMBOL(unregister_netdevice_queue
);
7542 * unregister_netdevice_many - unregister many devices
7543 * @head: list of devices
7545 * Note: As most callers use a stack allocated list_head,
7546 * we force a list_del() to make sure stack wont be corrupted later.
7548 void unregister_netdevice_many(struct list_head
*head
)
7550 struct net_device
*dev
;
7552 if (!list_empty(head
)) {
7553 rollback_registered_many(head
);
7554 list_for_each_entry(dev
, head
, unreg_list
)
7559 EXPORT_SYMBOL(unregister_netdevice_many
);
7562 * unregister_netdev - remove device from the kernel
7565 * This function shuts down a device interface and removes it
7566 * from the kernel tables.
7568 * This is just a wrapper for unregister_netdevice that takes
7569 * the rtnl semaphore. In general you want to use this and not
7570 * unregister_netdevice.
7572 void unregister_netdev(struct net_device
*dev
)
7575 unregister_netdevice(dev
);
7578 EXPORT_SYMBOL(unregister_netdev
);
7581 * dev_change_net_namespace - move device to different nethost namespace
7583 * @net: network namespace
7584 * @pat: If not NULL name pattern to try if the current device name
7585 * is already taken in the destination network namespace.
7587 * This function shuts down a device interface and moves it
7588 * to a new network namespace. On success 0 is returned, on
7589 * a failure a netagive errno code is returned.
7591 * Callers must hold the rtnl semaphore.
7594 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
7600 /* Don't allow namespace local devices to be moved. */
7602 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
7605 /* Ensure the device has been registrered */
7606 if (dev
->reg_state
!= NETREG_REGISTERED
)
7609 /* Get out if there is nothing todo */
7611 if (net_eq(dev_net(dev
), net
))
7614 /* Pick the destination device name, and ensure
7615 * we can use it in the destination network namespace.
7618 if (__dev_get_by_name(net
, dev
->name
)) {
7619 /* We get here if we can't use the current device name */
7622 if (dev_get_valid_name(net
, dev
, pat
) < 0)
7627 * And now a mini version of register_netdevice unregister_netdevice.
7630 /* If device is running close it first. */
7633 /* And unlink it from device chain */
7635 unlist_netdevice(dev
);
7639 /* Shutdown queueing discipline. */
7642 /* Notify protocols, that we are about to destroy
7643 this device. They should clean all the things.
7645 Note that dev->reg_state stays at NETREG_REGISTERED.
7646 This is wanted because this way 8021q and macvlan know
7647 the device is just moving and can keep their slaves up.
7649 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
7651 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
7652 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U, GFP_KERNEL
);
7655 * Flush the unicast and multicast chains
7660 /* Send a netdev-removed uevent to the old namespace */
7661 kobject_uevent(&dev
->dev
.kobj
, KOBJ_REMOVE
);
7662 netdev_adjacent_del_links(dev
);
7664 /* Actually switch the network namespace */
7665 dev_net_set(dev
, net
);
7667 /* If there is an ifindex conflict assign a new one */
7668 if (__dev_get_by_index(net
, dev
->ifindex
))
7669 dev
->ifindex
= dev_new_index(net
);
7671 /* Send a netdev-add uevent to the new namespace */
7672 kobject_uevent(&dev
->dev
.kobj
, KOBJ_ADD
);
7673 netdev_adjacent_add_links(dev
);
7675 /* Fixup kobjects */
7676 err
= device_rename(&dev
->dev
, dev
->name
);
7679 /* Add the device back in the hashes */
7680 list_netdevice(dev
);
7682 /* Notify protocols, that a new device appeared. */
7683 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
7686 * Prevent userspace races by waiting until the network
7687 * device is fully setup before sending notifications.
7689 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U, GFP_KERNEL
);
7696 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
7698 static int dev_cpu_callback(struct notifier_block
*nfb
,
7699 unsigned long action
,
7702 struct sk_buff
**list_skb
;
7703 struct sk_buff
*skb
;
7704 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
7705 struct softnet_data
*sd
, *oldsd
;
7707 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
7710 local_irq_disable();
7711 cpu
= smp_processor_id();
7712 sd
= &per_cpu(softnet_data
, cpu
);
7713 oldsd
= &per_cpu(softnet_data
, oldcpu
);
7715 /* Find end of our completion_queue. */
7716 list_skb
= &sd
->completion_queue
;
7718 list_skb
= &(*list_skb
)->next
;
7719 /* Append completion queue from offline CPU. */
7720 *list_skb
= oldsd
->completion_queue
;
7721 oldsd
->completion_queue
= NULL
;
7723 /* Append output queue from offline CPU. */
7724 if (oldsd
->output_queue
) {
7725 *sd
->output_queue_tailp
= oldsd
->output_queue
;
7726 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
7727 oldsd
->output_queue
= NULL
;
7728 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
7730 /* Append NAPI poll list from offline CPU, with one exception :
7731 * process_backlog() must be called by cpu owning percpu backlog.
7732 * We properly handle process_queue & input_pkt_queue later.
7734 while (!list_empty(&oldsd
->poll_list
)) {
7735 struct napi_struct
*napi
= list_first_entry(&oldsd
->poll_list
,
7739 list_del_init(&napi
->poll_list
);
7740 if (napi
->poll
== process_backlog
)
7743 ____napi_schedule(sd
, napi
);
7746 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
7749 /* Process offline CPU's input_pkt_queue */
7750 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
7752 input_queue_head_incr(oldsd
);
7754 while ((skb
= skb_dequeue(&oldsd
->input_pkt_queue
))) {
7756 input_queue_head_incr(oldsd
);
7764 * netdev_increment_features - increment feature set by one
7765 * @all: current feature set
7766 * @one: new feature set
7767 * @mask: mask feature set
7769 * Computes a new feature set after adding a device with feature set
7770 * @one to the master device with current feature set @all. Will not
7771 * enable anything that is off in @mask. Returns the new feature set.
7773 netdev_features_t
netdev_increment_features(netdev_features_t all
,
7774 netdev_features_t one
, netdev_features_t mask
)
7776 if (mask
& NETIF_F_HW_CSUM
)
7777 mask
|= NETIF_F_CSUM_MASK
;
7778 mask
|= NETIF_F_VLAN_CHALLENGED
;
7780 all
|= one
& (NETIF_F_ONE_FOR_ALL
| NETIF_F_CSUM_MASK
) & mask
;
7781 all
&= one
| ~NETIF_F_ALL_FOR_ALL
;
7783 /* If one device supports hw checksumming, set for all. */
7784 if (all
& NETIF_F_HW_CSUM
)
7785 all
&= ~(NETIF_F_CSUM_MASK
& ~NETIF_F_HW_CSUM
);
7789 EXPORT_SYMBOL(netdev_increment_features
);
7791 static struct hlist_head
* __net_init
netdev_create_hash(void)
7794 struct hlist_head
*hash
;
7796 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
7798 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
7799 INIT_HLIST_HEAD(&hash
[i
]);
7804 /* Initialize per network namespace state */
7805 static int __net_init
netdev_init(struct net
*net
)
7807 if (net
!= &init_net
)
7808 INIT_LIST_HEAD(&net
->dev_base_head
);
7810 net
->dev_name_head
= netdev_create_hash();
7811 if (net
->dev_name_head
== NULL
)
7814 net
->dev_index_head
= netdev_create_hash();
7815 if (net
->dev_index_head
== NULL
)
7821 kfree(net
->dev_name_head
);
7827 * netdev_drivername - network driver for the device
7828 * @dev: network device
7830 * Determine network driver for device.
7832 const char *netdev_drivername(const struct net_device
*dev
)
7834 const struct device_driver
*driver
;
7835 const struct device
*parent
;
7836 const char *empty
= "";
7838 parent
= dev
->dev
.parent
;
7842 driver
= parent
->driver
;
7843 if (driver
&& driver
->name
)
7844 return driver
->name
;
7848 static void __netdev_printk(const char *level
, const struct net_device
*dev
,
7849 struct va_format
*vaf
)
7851 if (dev
&& dev
->dev
.parent
) {
7852 dev_printk_emit(level
[1] - '0',
7855 dev_driver_string(dev
->dev
.parent
),
7856 dev_name(dev
->dev
.parent
),
7857 netdev_name(dev
), netdev_reg_state(dev
),
7860 printk("%s%s%s: %pV",
7861 level
, netdev_name(dev
), netdev_reg_state(dev
), vaf
);
7863 printk("%s(NULL net_device): %pV", level
, vaf
);
7867 void netdev_printk(const char *level
, const struct net_device
*dev
,
7868 const char *format
, ...)
7870 struct va_format vaf
;
7873 va_start(args
, format
);
7878 __netdev_printk(level
, dev
, &vaf
);
7882 EXPORT_SYMBOL(netdev_printk
);
7884 #define define_netdev_printk_level(func, level) \
7885 void func(const struct net_device *dev, const char *fmt, ...) \
7887 struct va_format vaf; \
7890 va_start(args, fmt); \
7895 __netdev_printk(level, dev, &vaf); \
7899 EXPORT_SYMBOL(func);
7901 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
7902 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
7903 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
7904 define_netdev_printk_level(netdev_err
, KERN_ERR
);
7905 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
7906 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
7907 define_netdev_printk_level(netdev_info
, KERN_INFO
);
7909 static void __net_exit
netdev_exit(struct net
*net
)
7911 kfree(net
->dev_name_head
);
7912 kfree(net
->dev_index_head
);
7915 static struct pernet_operations __net_initdata netdev_net_ops
= {
7916 .init
= netdev_init
,
7917 .exit
= netdev_exit
,
7920 static void __net_exit
default_device_exit(struct net
*net
)
7922 struct net_device
*dev
, *aux
;
7924 * Push all migratable network devices back to the
7925 * initial network namespace
7928 for_each_netdev_safe(net
, dev
, aux
) {
7930 char fb_name
[IFNAMSIZ
];
7932 /* Ignore unmoveable devices (i.e. loopback) */
7933 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
7936 /* Leave virtual devices for the generic cleanup */
7937 if (dev
->rtnl_link_ops
)
7940 /* Push remaining network devices to init_net */
7941 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
7942 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
7944 pr_emerg("%s: failed to move %s to init_net: %d\n",
7945 __func__
, dev
->name
, err
);
7952 static void __net_exit
rtnl_lock_unregistering(struct list_head
*net_list
)
7954 /* Return with the rtnl_lock held when there are no network
7955 * devices unregistering in any network namespace in net_list.
7959 DEFINE_WAIT_FUNC(wait
, woken_wake_function
);
7961 add_wait_queue(&netdev_unregistering_wq
, &wait
);
7963 unregistering
= false;
7965 list_for_each_entry(net
, net_list
, exit_list
) {
7966 if (net
->dev_unreg_count
> 0) {
7967 unregistering
= true;
7975 wait_woken(&wait
, TASK_UNINTERRUPTIBLE
, MAX_SCHEDULE_TIMEOUT
);
7977 remove_wait_queue(&netdev_unregistering_wq
, &wait
);
7980 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
7982 /* At exit all network devices most be removed from a network
7983 * namespace. Do this in the reverse order of registration.
7984 * Do this across as many network namespaces as possible to
7985 * improve batching efficiency.
7987 struct net_device
*dev
;
7989 LIST_HEAD(dev_kill_list
);
7991 /* To prevent network device cleanup code from dereferencing
7992 * loopback devices or network devices that have been freed
7993 * wait here for all pending unregistrations to complete,
7994 * before unregistring the loopback device and allowing the
7995 * network namespace be freed.
7997 * The netdev todo list containing all network devices
7998 * unregistrations that happen in default_device_exit_batch
7999 * will run in the rtnl_unlock() at the end of
8000 * default_device_exit_batch.
8002 rtnl_lock_unregistering(net_list
);
8003 list_for_each_entry(net
, net_list
, exit_list
) {
8004 for_each_netdev_reverse(net
, dev
) {
8005 if (dev
->rtnl_link_ops
&& dev
->rtnl_link_ops
->dellink
)
8006 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
8008 unregister_netdevice_queue(dev
, &dev_kill_list
);
8011 unregister_netdevice_many(&dev_kill_list
);
8015 static struct pernet_operations __net_initdata default_device_ops
= {
8016 .exit
= default_device_exit
,
8017 .exit_batch
= default_device_exit_batch
,
8021 * Initialize the DEV module. At boot time this walks the device list and
8022 * unhooks any devices that fail to initialise (normally hardware not
8023 * present) and leaves us with a valid list of present and active devices.
8028 * This is called single threaded during boot, so no need
8029 * to take the rtnl semaphore.
8031 static int __init
net_dev_init(void)
8033 int i
, rc
= -ENOMEM
;
8035 BUG_ON(!dev_boot_phase
);
8037 if (dev_proc_init())
8040 if (netdev_kobject_init())
8043 INIT_LIST_HEAD(&ptype_all
);
8044 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
8045 INIT_LIST_HEAD(&ptype_base
[i
]);
8047 INIT_LIST_HEAD(&offload_base
);
8049 if (register_pernet_subsys(&netdev_net_ops
))
8053 * Initialise the packet receive queues.
8056 for_each_possible_cpu(i
) {
8057 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
8059 skb_queue_head_init(&sd
->input_pkt_queue
);
8060 skb_queue_head_init(&sd
->process_queue
);
8061 INIT_LIST_HEAD(&sd
->poll_list
);
8062 sd
->output_queue_tailp
= &sd
->output_queue
;
8064 sd
->csd
.func
= rps_trigger_softirq
;
8069 sd
->backlog
.poll
= process_backlog
;
8070 sd
->backlog
.weight
= weight_p
;
8075 /* The loopback device is special if any other network devices
8076 * is present in a network namespace the loopback device must
8077 * be present. Since we now dynamically allocate and free the
8078 * loopback device ensure this invariant is maintained by
8079 * keeping the loopback device as the first device on the
8080 * list of network devices. Ensuring the loopback devices
8081 * is the first device that appears and the last network device
8084 if (register_pernet_device(&loopback_net_ops
))
8087 if (register_pernet_device(&default_device_ops
))
8090 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
8091 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
8093 hotcpu_notifier(dev_cpu_callback
, 0);
8100 subsys_initcall(net_dev_init
);