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 <linux/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 <linux/bpf.h>
98 #include <net/net_namespace.h>
100 #include <net/busy_poll.h>
101 #include <linux/rtnetlink.h>
102 #include <linux/stat.h>
104 #include <net/dst_metadata.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include <linux/highmem.h>
109 #include <linux/init.h>
110 #include <linux/module.h>
111 #include <linux/netpoll.h>
112 #include <linux/rcupdate.h>
113 #include <linux/delay.h>
114 #include <net/iw_handler.h>
115 #include <asm/current.h>
116 #include <linux/audit.h>
117 #include <linux/dmaengine.h>
118 #include <linux/err.h>
119 #include <linux/ctype.h>
120 #include <linux/if_arp.h>
121 #include <linux/if_vlan.h>
122 #include <linux/ip.h>
124 #include <net/mpls.h>
125 #include <linux/ipv6.h>
126 #include <linux/in.h>
127 #include <linux/jhash.h>
128 #include <linux/random.h>
129 #include <trace/events/napi.h>
130 #include <trace/events/net.h>
131 #include <trace/events/skb.h>
132 #include <linux/pci.h>
133 #include <linux/inetdevice.h>
134 #include <linux/cpu_rmap.h>
135 #include <linux/static_key.h>
136 #include <linux/hashtable.h>
137 #include <linux/vmalloc.h>
138 #include <linux/if_macvlan.h>
139 #include <linux/errqueue.h>
140 #include <linux/hrtimer.h>
141 #include <linux/netfilter_ingress.h>
142 #include <linux/crash_dump.h>
144 #include "net-sysfs.h"
146 /* Instead of increasing this, you should create a hash table. */
147 #define MAX_GRO_SKBS 8
149 /* This should be increased if a protocol with a bigger head is added. */
150 #define GRO_MAX_HEAD (MAX_HEADER + 128)
152 static DEFINE_SPINLOCK(ptype_lock
);
153 static DEFINE_SPINLOCK(offload_lock
);
154 struct list_head ptype_base
[PTYPE_HASH_SIZE
] __read_mostly
;
155 struct list_head ptype_all __read_mostly
; /* Taps */
156 static struct list_head offload_base __read_mostly
;
158 static int netif_rx_internal(struct sk_buff
*skb
);
159 static int call_netdevice_notifiers_info(unsigned long val
,
160 struct net_device
*dev
,
161 struct netdev_notifier_info
*info
);
164 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
167 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
169 * Writers must hold the rtnl semaphore while they loop through the
170 * dev_base_head list, and hold dev_base_lock for writing when they do the
171 * actual updates. This allows pure readers to access the list even
172 * while a writer is preparing to update it.
174 * To put it another way, dev_base_lock is held for writing only to
175 * protect against pure readers; the rtnl semaphore provides the
176 * protection against other writers.
178 * See, for example usages, register_netdevice() and
179 * unregister_netdevice(), which must be called with the rtnl
182 DEFINE_RWLOCK(dev_base_lock
);
183 EXPORT_SYMBOL(dev_base_lock
);
185 /* protects napi_hash addition/deletion and napi_gen_id */
186 static DEFINE_SPINLOCK(napi_hash_lock
);
188 static unsigned int napi_gen_id
= NR_CPUS
;
189 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash
, 8);
191 static seqcount_t devnet_rename_seq
;
193 static inline void dev_base_seq_inc(struct net
*net
)
195 while (++net
->dev_base_seq
== 0)
199 static inline struct hlist_head
*dev_name_hash(struct net
*net
, const char *name
)
201 unsigned int hash
= full_name_hash(net
, name
, strnlen(name
, IFNAMSIZ
));
203 return &net
->dev_name_head
[hash_32(hash
, NETDEV_HASHBITS
)];
206 static inline struct hlist_head
*dev_index_hash(struct net
*net
, int ifindex
)
208 return &net
->dev_index_head
[ifindex
& (NETDEV_HASHENTRIES
- 1)];
211 static inline void rps_lock(struct softnet_data
*sd
)
214 spin_lock(&sd
->input_pkt_queue
.lock
);
218 static inline void rps_unlock(struct softnet_data
*sd
)
221 spin_unlock(&sd
->input_pkt_queue
.lock
);
225 /* Device list insertion */
226 static void list_netdevice(struct net_device
*dev
)
228 struct net
*net
= dev_net(dev
);
232 write_lock_bh(&dev_base_lock
);
233 list_add_tail_rcu(&dev
->dev_list
, &net
->dev_base_head
);
234 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
235 hlist_add_head_rcu(&dev
->index_hlist
,
236 dev_index_hash(net
, dev
->ifindex
));
237 write_unlock_bh(&dev_base_lock
);
239 dev_base_seq_inc(net
);
242 /* Device list removal
243 * caller must respect a RCU grace period before freeing/reusing dev
245 static void unlist_netdevice(struct net_device
*dev
)
249 /* Unlink dev from the device chain */
250 write_lock_bh(&dev_base_lock
);
251 list_del_rcu(&dev
->dev_list
);
252 hlist_del_rcu(&dev
->name_hlist
);
253 hlist_del_rcu(&dev
->index_hlist
);
254 write_unlock_bh(&dev_base_lock
);
256 dev_base_seq_inc(dev_net(dev
));
263 static RAW_NOTIFIER_HEAD(netdev_chain
);
266 * Device drivers call our routines to queue packets here. We empty the
267 * queue in the local softnet handler.
270 DEFINE_PER_CPU_ALIGNED(struct softnet_data
, softnet_data
);
271 EXPORT_PER_CPU_SYMBOL(softnet_data
);
273 #ifdef CONFIG_LOCKDEP
275 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
276 * according to dev->type
278 static const unsigned short netdev_lock_type
[] = {
279 ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
280 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
281 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
282 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
283 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
284 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
285 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
286 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
287 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
288 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
289 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
290 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
291 ARPHRD_FCFABRIC
, ARPHRD_IEEE80211
, ARPHRD_IEEE80211_PRISM
,
292 ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_PHONET
, ARPHRD_PHONET_PIPE
,
293 ARPHRD_IEEE802154
, ARPHRD_VOID
, ARPHRD_NONE
};
295 static const char *const netdev_lock_name
[] = {
296 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
297 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
298 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
299 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
300 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
301 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
302 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
303 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
304 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
305 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
306 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
307 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
308 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
309 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
310 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
312 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
313 static struct lock_class_key netdev_addr_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
315 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
319 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
320 if (netdev_lock_type
[i
] == dev_type
)
322 /* the last key is used by default */
323 return ARRAY_SIZE(netdev_lock_type
) - 1;
326 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
327 unsigned short dev_type
)
331 i
= netdev_lock_pos(dev_type
);
332 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
333 netdev_lock_name
[i
]);
336 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
340 i
= netdev_lock_pos(dev
->type
);
341 lockdep_set_class_and_name(&dev
->addr_list_lock
,
342 &netdev_addr_lock_key
[i
],
343 netdev_lock_name
[i
]);
346 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
347 unsigned short dev_type
)
350 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
355 /*******************************************************************************
357 * Protocol management and registration routines
359 *******************************************************************************/
363 * Add a protocol ID to the list. Now that the input handler is
364 * smarter we can dispense with all the messy stuff that used to be
367 * BEWARE!!! Protocol handlers, mangling input packets,
368 * MUST BE last in hash buckets and checking protocol handlers
369 * MUST start from promiscuous ptype_all chain in net_bh.
370 * It is true now, do not change it.
371 * Explanation follows: if protocol handler, mangling packet, will
372 * be the first on list, it is not able to sense, that packet
373 * is cloned and should be copied-on-write, so that it will
374 * change it and subsequent readers will get broken packet.
378 static inline struct list_head
*ptype_head(const struct packet_type
*pt
)
380 if (pt
->type
== htons(ETH_P_ALL
))
381 return pt
->dev
? &pt
->dev
->ptype_all
: &ptype_all
;
383 return pt
->dev
? &pt
->dev
->ptype_specific
:
384 &ptype_base
[ntohs(pt
->type
) & PTYPE_HASH_MASK
];
388 * dev_add_pack - add packet handler
389 * @pt: packet type declaration
391 * Add a protocol handler to the networking stack. The passed &packet_type
392 * is linked into kernel lists and may not be freed until it has been
393 * removed from the kernel lists.
395 * This call does not sleep therefore it can not
396 * guarantee all CPU's that are in middle of receiving packets
397 * will see the new packet type (until the next received packet).
400 void dev_add_pack(struct packet_type
*pt
)
402 struct list_head
*head
= ptype_head(pt
);
404 spin_lock(&ptype_lock
);
405 list_add_rcu(&pt
->list
, head
);
406 spin_unlock(&ptype_lock
);
408 EXPORT_SYMBOL(dev_add_pack
);
411 * __dev_remove_pack - remove packet handler
412 * @pt: packet type declaration
414 * Remove a protocol handler that was previously added to the kernel
415 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
416 * from the kernel lists and can be freed or reused once this function
419 * The packet type might still be in use by receivers
420 * and must not be freed until after all the CPU's have gone
421 * through a quiescent state.
423 void __dev_remove_pack(struct packet_type
*pt
)
425 struct list_head
*head
= ptype_head(pt
);
426 struct packet_type
*pt1
;
428 spin_lock(&ptype_lock
);
430 list_for_each_entry(pt1
, head
, list
) {
432 list_del_rcu(&pt
->list
);
437 pr_warn("dev_remove_pack: %p not found\n", pt
);
439 spin_unlock(&ptype_lock
);
441 EXPORT_SYMBOL(__dev_remove_pack
);
444 * dev_remove_pack - remove packet handler
445 * @pt: packet type declaration
447 * Remove a protocol handler that was previously added to the kernel
448 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
449 * from the kernel lists and can be freed or reused once this function
452 * This call sleeps to guarantee that no CPU is looking at the packet
455 void dev_remove_pack(struct packet_type
*pt
)
457 __dev_remove_pack(pt
);
461 EXPORT_SYMBOL(dev_remove_pack
);
465 * dev_add_offload - register offload handlers
466 * @po: protocol offload declaration
468 * Add protocol offload handlers to the networking stack. The passed
469 * &proto_offload is linked into kernel lists and may not be freed until
470 * it has been removed from the kernel lists.
472 * This call does not sleep therefore it can not
473 * guarantee all CPU's that are in middle of receiving packets
474 * will see the new offload handlers (until the next received packet).
476 void dev_add_offload(struct packet_offload
*po
)
478 struct packet_offload
*elem
;
480 spin_lock(&offload_lock
);
481 list_for_each_entry(elem
, &offload_base
, list
) {
482 if (po
->priority
< elem
->priority
)
485 list_add_rcu(&po
->list
, elem
->list
.prev
);
486 spin_unlock(&offload_lock
);
488 EXPORT_SYMBOL(dev_add_offload
);
491 * __dev_remove_offload - remove offload handler
492 * @po: packet offload declaration
494 * Remove a protocol offload handler that was previously added to the
495 * kernel offload handlers by dev_add_offload(). The passed &offload_type
496 * is removed from the kernel lists and can be freed or reused once this
499 * The packet type might still be in use by receivers
500 * and must not be freed until after all the CPU's have gone
501 * through a quiescent state.
503 static void __dev_remove_offload(struct packet_offload
*po
)
505 struct list_head
*head
= &offload_base
;
506 struct packet_offload
*po1
;
508 spin_lock(&offload_lock
);
510 list_for_each_entry(po1
, head
, list
) {
512 list_del_rcu(&po
->list
);
517 pr_warn("dev_remove_offload: %p not found\n", po
);
519 spin_unlock(&offload_lock
);
523 * dev_remove_offload - remove packet offload handler
524 * @po: packet offload declaration
526 * Remove a packet offload handler that was previously added to the kernel
527 * offload handlers by dev_add_offload(). The passed &offload_type is
528 * removed from the kernel lists and can be freed or reused once this
531 * This call sleeps to guarantee that no CPU is looking at the packet
534 void dev_remove_offload(struct packet_offload
*po
)
536 __dev_remove_offload(po
);
540 EXPORT_SYMBOL(dev_remove_offload
);
542 /******************************************************************************
544 * Device Boot-time Settings Routines
546 ******************************************************************************/
548 /* Boot time configuration table */
549 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
552 * netdev_boot_setup_add - add new setup entry
553 * @name: name of the device
554 * @map: configured settings for the device
556 * Adds new setup entry to the dev_boot_setup list. The function
557 * returns 0 on error and 1 on success. This is a generic routine to
560 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
562 struct netdev_boot_setup
*s
;
566 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
567 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
568 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
569 strlcpy(s
[i
].name
, name
, IFNAMSIZ
);
570 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
575 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
579 * netdev_boot_setup_check - check boot time settings
580 * @dev: the netdevice
582 * Check boot time settings for the device.
583 * The found settings are set for the device to be used
584 * later in the device probing.
585 * Returns 0 if no settings found, 1 if they are.
587 int netdev_boot_setup_check(struct net_device
*dev
)
589 struct netdev_boot_setup
*s
= dev_boot_setup
;
592 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
593 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
594 !strcmp(dev
->name
, s
[i
].name
)) {
595 dev
->irq
= s
[i
].map
.irq
;
596 dev
->base_addr
= s
[i
].map
.base_addr
;
597 dev
->mem_start
= s
[i
].map
.mem_start
;
598 dev
->mem_end
= s
[i
].map
.mem_end
;
604 EXPORT_SYMBOL(netdev_boot_setup_check
);
608 * netdev_boot_base - get address from boot time settings
609 * @prefix: prefix for network device
610 * @unit: id for network device
612 * Check boot time settings for the base address of device.
613 * The found settings are set for the device to be used
614 * later in the device probing.
615 * Returns 0 if no settings found.
617 unsigned long netdev_boot_base(const char *prefix
, int unit
)
619 const struct netdev_boot_setup
*s
= dev_boot_setup
;
623 sprintf(name
, "%s%d", prefix
, unit
);
626 * If device already registered then return base of 1
627 * to indicate not to probe for this interface
629 if (__dev_get_by_name(&init_net
, name
))
632 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
633 if (!strcmp(name
, s
[i
].name
))
634 return s
[i
].map
.base_addr
;
639 * Saves at boot time configured settings for any netdevice.
641 int __init
netdev_boot_setup(char *str
)
646 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
651 memset(&map
, 0, sizeof(map
));
655 map
.base_addr
= ints
[2];
657 map
.mem_start
= ints
[3];
659 map
.mem_end
= ints
[4];
661 /* Add new entry to the list */
662 return netdev_boot_setup_add(str
, &map
);
665 __setup("netdev=", netdev_boot_setup
);
667 /*******************************************************************************
669 * Device Interface Subroutines
671 *******************************************************************************/
674 * dev_get_iflink - get 'iflink' value of a interface
675 * @dev: targeted interface
677 * Indicates the ifindex the interface is linked to.
678 * Physical interfaces have the same 'ifindex' and 'iflink' values.
681 int dev_get_iflink(const struct net_device
*dev
)
683 if (dev
->netdev_ops
&& dev
->netdev_ops
->ndo_get_iflink
)
684 return dev
->netdev_ops
->ndo_get_iflink(dev
);
688 EXPORT_SYMBOL(dev_get_iflink
);
691 * dev_fill_metadata_dst - Retrieve tunnel egress information.
692 * @dev: targeted interface
695 * For better visibility of tunnel traffic OVS needs to retrieve
696 * egress tunnel information for a packet. Following API allows
697 * user to get this info.
699 int dev_fill_metadata_dst(struct net_device
*dev
, struct sk_buff
*skb
)
701 struct ip_tunnel_info
*info
;
703 if (!dev
->netdev_ops
|| !dev
->netdev_ops
->ndo_fill_metadata_dst
)
706 info
= skb_tunnel_info_unclone(skb
);
709 if (unlikely(!(info
->mode
& IP_TUNNEL_INFO_TX
)))
712 return dev
->netdev_ops
->ndo_fill_metadata_dst(dev
, skb
);
714 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst
);
717 * __dev_get_by_name - find a device by its name
718 * @net: the applicable net namespace
719 * @name: name to find
721 * Find an interface by name. Must be called under RTNL semaphore
722 * or @dev_base_lock. If the name is found a pointer to the device
723 * is returned. If the name is not found then %NULL is returned. The
724 * reference counters are not incremented so the caller must be
725 * careful with locks.
728 struct net_device
*__dev_get_by_name(struct net
*net
, const char *name
)
730 struct net_device
*dev
;
731 struct hlist_head
*head
= dev_name_hash(net
, name
);
733 hlist_for_each_entry(dev
, head
, name_hlist
)
734 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
739 EXPORT_SYMBOL(__dev_get_by_name
);
742 * dev_get_by_name_rcu - find a device by its name
743 * @net: the applicable net namespace
744 * @name: name to find
746 * Find an interface by name.
747 * If the name is found a pointer to the device is returned.
748 * If the name is not found then %NULL is returned.
749 * The reference counters are not incremented so the caller must be
750 * careful with locks. The caller must hold RCU lock.
753 struct net_device
*dev_get_by_name_rcu(struct net
*net
, const char *name
)
755 struct net_device
*dev
;
756 struct hlist_head
*head
= dev_name_hash(net
, name
);
758 hlist_for_each_entry_rcu(dev
, head
, name_hlist
)
759 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
764 EXPORT_SYMBOL(dev_get_by_name_rcu
);
767 * dev_get_by_name - find a device by its name
768 * @net: the applicable net namespace
769 * @name: name to find
771 * Find an interface by name. This can be called from any
772 * context and does its own locking. The returned handle has
773 * the usage count incremented and the caller must use dev_put() to
774 * release it when it is no longer needed. %NULL is returned if no
775 * matching device is found.
778 struct net_device
*dev_get_by_name(struct net
*net
, const char *name
)
780 struct net_device
*dev
;
783 dev
= dev_get_by_name_rcu(net
, name
);
789 EXPORT_SYMBOL(dev_get_by_name
);
792 * __dev_get_by_index - find a device by its ifindex
793 * @net: the applicable net namespace
794 * @ifindex: index of device
796 * Search for an interface by index. Returns %NULL if the device
797 * is not found or a pointer to the device. The device has not
798 * had its reference counter increased so the caller must be careful
799 * about locking. The caller must hold either the RTNL semaphore
803 struct net_device
*__dev_get_by_index(struct net
*net
, int ifindex
)
805 struct net_device
*dev
;
806 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
808 hlist_for_each_entry(dev
, head
, index_hlist
)
809 if (dev
->ifindex
== ifindex
)
814 EXPORT_SYMBOL(__dev_get_by_index
);
817 * dev_get_by_index_rcu - find a device by its ifindex
818 * @net: the applicable net namespace
819 * @ifindex: index of device
821 * Search for an interface by index. Returns %NULL if the device
822 * is not found or a pointer to the device. The device has not
823 * had its reference counter increased so the caller must be careful
824 * about locking. The caller must hold RCU lock.
827 struct net_device
*dev_get_by_index_rcu(struct net
*net
, int ifindex
)
829 struct net_device
*dev
;
830 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
832 hlist_for_each_entry_rcu(dev
, head
, index_hlist
)
833 if (dev
->ifindex
== ifindex
)
838 EXPORT_SYMBOL(dev_get_by_index_rcu
);
842 * dev_get_by_index - find a device by its ifindex
843 * @net: the applicable net namespace
844 * @ifindex: index of device
846 * Search for an interface by index. Returns NULL if the device
847 * is not found or a pointer to the device. The device returned has
848 * had a reference added and the pointer is safe until the user calls
849 * dev_put to indicate they have finished with it.
852 struct net_device
*dev_get_by_index(struct net
*net
, int ifindex
)
854 struct net_device
*dev
;
857 dev
= dev_get_by_index_rcu(net
, ifindex
);
863 EXPORT_SYMBOL(dev_get_by_index
);
866 * netdev_get_name - get a netdevice name, knowing its ifindex.
867 * @net: network namespace
868 * @name: a pointer to the buffer where the name will be stored.
869 * @ifindex: the ifindex of the interface to get the name from.
871 * The use of raw_seqcount_begin() and cond_resched() before
872 * retrying is required as we want to give the writers a chance
873 * to complete when CONFIG_PREEMPT is not set.
875 int netdev_get_name(struct net
*net
, char *name
, int ifindex
)
877 struct net_device
*dev
;
881 seq
= raw_seqcount_begin(&devnet_rename_seq
);
883 dev
= dev_get_by_index_rcu(net
, ifindex
);
889 strcpy(name
, dev
->name
);
891 if (read_seqcount_retry(&devnet_rename_seq
, seq
)) {
900 * dev_getbyhwaddr_rcu - find a device by its hardware address
901 * @net: the applicable net namespace
902 * @type: media type of device
903 * @ha: hardware address
905 * Search for an interface by MAC address. Returns NULL if the device
906 * is not found or a pointer to the device.
907 * The caller must hold RCU or RTNL.
908 * The returned device has not had its ref count increased
909 * and the caller must therefore be careful about locking
913 struct net_device
*dev_getbyhwaddr_rcu(struct net
*net
, unsigned short type
,
916 struct net_device
*dev
;
918 for_each_netdev_rcu(net
, dev
)
919 if (dev
->type
== type
&&
920 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
925 EXPORT_SYMBOL(dev_getbyhwaddr_rcu
);
927 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
929 struct net_device
*dev
;
932 for_each_netdev(net
, dev
)
933 if (dev
->type
== type
)
938 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
940 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
942 struct net_device
*dev
, *ret
= NULL
;
945 for_each_netdev_rcu(net
, dev
)
946 if (dev
->type
== type
) {
954 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
957 * __dev_get_by_flags - find any device with given flags
958 * @net: the applicable net namespace
959 * @if_flags: IFF_* values
960 * @mask: bitmask of bits in if_flags to check
962 * Search for any interface with the given flags. Returns NULL if a device
963 * is not found or a pointer to the device. Must be called inside
964 * rtnl_lock(), and result refcount is unchanged.
967 struct net_device
*__dev_get_by_flags(struct net
*net
, unsigned short if_flags
,
970 struct net_device
*dev
, *ret
;
975 for_each_netdev(net
, dev
) {
976 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
983 EXPORT_SYMBOL(__dev_get_by_flags
);
986 * dev_valid_name - check if name is okay for network device
989 * Network device names need to be valid file names to
990 * to allow sysfs to work. We also disallow any kind of
993 bool dev_valid_name(const char *name
)
997 if (strlen(name
) >= IFNAMSIZ
)
999 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
1003 if (*name
== '/' || *name
== ':' || isspace(*name
))
1009 EXPORT_SYMBOL(dev_valid_name
);
1012 * __dev_alloc_name - allocate a name for a device
1013 * @net: network namespace to allocate the device name in
1014 * @name: name format string
1015 * @buf: scratch buffer and result name string
1017 * Passed a format string - eg "lt%d" it will try and find a suitable
1018 * id. It scans list of devices to build up a free map, then chooses
1019 * the first empty slot. The caller must hold the dev_base or rtnl lock
1020 * while allocating the name and adding the device in order to avoid
1022 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1023 * Returns the number of the unit assigned or a negative errno code.
1026 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
1030 const int max_netdevices
= 8*PAGE_SIZE
;
1031 unsigned long *inuse
;
1032 struct net_device
*d
;
1034 p
= strnchr(name
, IFNAMSIZ
-1, '%');
1037 * Verify the string as this thing may have come from
1038 * the user. There must be either one "%d" and no other "%"
1041 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
1044 /* Use one page as a bit array of possible slots */
1045 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
1049 for_each_netdev(net
, d
) {
1050 if (!sscanf(d
->name
, name
, &i
))
1052 if (i
< 0 || i
>= max_netdevices
)
1055 /* avoid cases where sscanf is not exact inverse of printf */
1056 snprintf(buf
, IFNAMSIZ
, name
, i
);
1057 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
1061 i
= find_first_zero_bit(inuse
, max_netdevices
);
1062 free_page((unsigned long) inuse
);
1066 snprintf(buf
, IFNAMSIZ
, name
, i
);
1067 if (!__dev_get_by_name(net
, buf
))
1070 /* It is possible to run out of possible slots
1071 * when the name is long and there isn't enough space left
1072 * for the digits, or if all bits are used.
1078 * dev_alloc_name - allocate a name for a device
1080 * @name: name format string
1082 * Passed a format string - eg "lt%d" it will try and find a suitable
1083 * id. It scans list of devices to build up a free map, then chooses
1084 * the first empty slot. The caller must hold the dev_base or rtnl lock
1085 * while allocating the name and adding the device in order to avoid
1087 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1088 * Returns the number of the unit assigned or a negative errno code.
1091 int dev_alloc_name(struct net_device
*dev
, const char *name
)
1097 BUG_ON(!dev_net(dev
));
1099 ret
= __dev_alloc_name(net
, name
, buf
);
1101 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
1104 EXPORT_SYMBOL(dev_alloc_name
);
1106 static int dev_alloc_name_ns(struct net
*net
,
1107 struct net_device
*dev
,
1113 ret
= __dev_alloc_name(net
, name
, buf
);
1115 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
1119 static int dev_get_valid_name(struct net
*net
,
1120 struct net_device
*dev
,
1125 if (!dev_valid_name(name
))
1128 if (strchr(name
, '%'))
1129 return dev_alloc_name_ns(net
, dev
, name
);
1130 else if (__dev_get_by_name(net
, name
))
1132 else if (dev
->name
!= name
)
1133 strlcpy(dev
->name
, name
, IFNAMSIZ
);
1139 * dev_change_name - change name of a device
1141 * @newname: name (or format string) must be at least IFNAMSIZ
1143 * Change name of a device, can pass format strings "eth%d".
1146 int dev_change_name(struct net_device
*dev
, const char *newname
)
1148 unsigned char old_assign_type
;
1149 char oldname
[IFNAMSIZ
];
1155 BUG_ON(!dev_net(dev
));
1158 if (dev
->flags
& IFF_UP
)
1161 write_seqcount_begin(&devnet_rename_seq
);
1163 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0) {
1164 write_seqcount_end(&devnet_rename_seq
);
1168 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
1170 err
= dev_get_valid_name(net
, dev
, newname
);
1172 write_seqcount_end(&devnet_rename_seq
);
1176 if (oldname
[0] && !strchr(oldname
, '%'))
1177 netdev_info(dev
, "renamed from %s\n", oldname
);
1179 old_assign_type
= dev
->name_assign_type
;
1180 dev
->name_assign_type
= NET_NAME_RENAMED
;
1183 ret
= device_rename(&dev
->dev
, dev
->name
);
1185 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1186 dev
->name_assign_type
= old_assign_type
;
1187 write_seqcount_end(&devnet_rename_seq
);
1191 write_seqcount_end(&devnet_rename_seq
);
1193 netdev_adjacent_rename_links(dev
, oldname
);
1195 write_lock_bh(&dev_base_lock
);
1196 hlist_del_rcu(&dev
->name_hlist
);
1197 write_unlock_bh(&dev_base_lock
);
1201 write_lock_bh(&dev_base_lock
);
1202 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
1203 write_unlock_bh(&dev_base_lock
);
1205 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
1206 ret
= notifier_to_errno(ret
);
1209 /* err >= 0 after dev_alloc_name() or stores the first errno */
1212 write_seqcount_begin(&devnet_rename_seq
);
1213 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1214 memcpy(oldname
, newname
, IFNAMSIZ
);
1215 dev
->name_assign_type
= old_assign_type
;
1216 old_assign_type
= NET_NAME_RENAMED
;
1219 pr_err("%s: name change rollback failed: %d\n",
1228 * dev_set_alias - change ifalias of a device
1230 * @alias: name up to IFALIASZ
1231 * @len: limit of bytes to copy from info
1233 * Set ifalias for a device,
1235 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
1241 if (len
>= IFALIASZ
)
1245 kfree(dev
->ifalias
);
1246 dev
->ifalias
= NULL
;
1250 new_ifalias
= krealloc(dev
->ifalias
, len
+ 1, GFP_KERNEL
);
1253 dev
->ifalias
= new_ifalias
;
1255 strlcpy(dev
->ifalias
, alias
, len
+1);
1261 * netdev_features_change - device changes features
1262 * @dev: device to cause notification
1264 * Called to indicate a device has changed features.
1266 void netdev_features_change(struct net_device
*dev
)
1268 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1270 EXPORT_SYMBOL(netdev_features_change
);
1273 * netdev_state_change - device changes state
1274 * @dev: device to cause notification
1276 * Called to indicate a device has changed state. This function calls
1277 * the notifier chains for netdev_chain and sends a NEWLINK message
1278 * to the routing socket.
1280 void netdev_state_change(struct net_device
*dev
)
1282 if (dev
->flags
& IFF_UP
) {
1283 struct netdev_notifier_change_info change_info
;
1285 change_info
.flags_changed
= 0;
1286 call_netdevice_notifiers_info(NETDEV_CHANGE
, dev
,
1288 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0, GFP_KERNEL
);
1291 EXPORT_SYMBOL(netdev_state_change
);
1294 * netdev_notify_peers - notify network peers about existence of @dev
1295 * @dev: network device
1297 * Generate traffic such that interested network peers are aware of
1298 * @dev, such as by generating a gratuitous ARP. This may be used when
1299 * a device wants to inform the rest of the network about some sort of
1300 * reconfiguration such as a failover event or virtual machine
1303 void netdev_notify_peers(struct net_device
*dev
)
1306 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
, dev
);
1307 call_netdevice_notifiers(NETDEV_RESEND_IGMP
, dev
);
1310 EXPORT_SYMBOL(netdev_notify_peers
);
1312 static int __dev_open(struct net_device
*dev
)
1314 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1319 if (!netif_device_present(dev
))
1322 /* Block netpoll from trying to do any rx path servicing.
1323 * If we don't do this there is a chance ndo_poll_controller
1324 * or ndo_poll may be running while we open the device
1326 netpoll_poll_disable(dev
);
1328 ret
= call_netdevice_notifiers(NETDEV_PRE_UP
, dev
);
1329 ret
= notifier_to_errno(ret
);
1333 set_bit(__LINK_STATE_START
, &dev
->state
);
1335 if (ops
->ndo_validate_addr
)
1336 ret
= ops
->ndo_validate_addr(dev
);
1338 if (!ret
&& ops
->ndo_open
)
1339 ret
= ops
->ndo_open(dev
);
1341 netpoll_poll_enable(dev
);
1344 clear_bit(__LINK_STATE_START
, &dev
->state
);
1346 dev
->flags
|= IFF_UP
;
1347 dev_set_rx_mode(dev
);
1349 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
1356 * dev_open - prepare an interface for use.
1357 * @dev: device to open
1359 * Takes a device from down to up state. The device's private open
1360 * function is invoked and then the multicast lists are loaded. Finally
1361 * the device is moved into the up state and a %NETDEV_UP message is
1362 * sent to the netdev notifier chain.
1364 * Calling this function on an active interface is a nop. On a failure
1365 * a negative errno code is returned.
1367 int dev_open(struct net_device
*dev
)
1371 if (dev
->flags
& IFF_UP
)
1374 ret
= __dev_open(dev
);
1378 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
, GFP_KERNEL
);
1379 call_netdevice_notifiers(NETDEV_UP
, dev
);
1383 EXPORT_SYMBOL(dev_open
);
1385 static int __dev_close_many(struct list_head
*head
)
1387 struct net_device
*dev
;
1392 list_for_each_entry(dev
, head
, close_list
) {
1393 /* Temporarily disable netpoll until the interface is down */
1394 netpoll_poll_disable(dev
);
1396 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1398 clear_bit(__LINK_STATE_START
, &dev
->state
);
1400 /* Synchronize to scheduled poll. We cannot touch poll list, it
1401 * can be even on different cpu. So just clear netif_running().
1403 * dev->stop() will invoke napi_disable() on all of it's
1404 * napi_struct instances on this device.
1406 smp_mb__after_atomic(); /* Commit netif_running(). */
1409 dev_deactivate_many(head
);
1411 list_for_each_entry(dev
, head
, close_list
) {
1412 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1415 * Call the device specific close. This cannot fail.
1416 * Only if device is UP
1418 * We allow it to be called even after a DETACH hot-plug
1424 dev
->flags
&= ~IFF_UP
;
1425 netpoll_poll_enable(dev
);
1431 static int __dev_close(struct net_device
*dev
)
1436 list_add(&dev
->close_list
, &single
);
1437 retval
= __dev_close_many(&single
);
1443 int dev_close_many(struct list_head
*head
, bool unlink
)
1445 struct net_device
*dev
, *tmp
;
1447 /* Remove the devices that don't need to be closed */
1448 list_for_each_entry_safe(dev
, tmp
, head
, close_list
)
1449 if (!(dev
->flags
& IFF_UP
))
1450 list_del_init(&dev
->close_list
);
1452 __dev_close_many(head
);
1454 list_for_each_entry_safe(dev
, tmp
, head
, close_list
) {
1455 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
, GFP_KERNEL
);
1456 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1458 list_del_init(&dev
->close_list
);
1463 EXPORT_SYMBOL(dev_close_many
);
1466 * dev_close - shutdown an interface.
1467 * @dev: device to shutdown
1469 * This function moves an active device into down state. A
1470 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1471 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1474 int dev_close(struct net_device
*dev
)
1476 if (dev
->flags
& IFF_UP
) {
1479 list_add(&dev
->close_list
, &single
);
1480 dev_close_many(&single
, true);
1485 EXPORT_SYMBOL(dev_close
);
1489 * dev_disable_lro - disable Large Receive Offload on a device
1492 * Disable Large Receive Offload (LRO) on a net device. Must be
1493 * called under RTNL. This is needed if received packets may be
1494 * forwarded to another interface.
1496 void dev_disable_lro(struct net_device
*dev
)
1498 struct net_device
*lower_dev
;
1499 struct list_head
*iter
;
1501 dev
->wanted_features
&= ~NETIF_F_LRO
;
1502 netdev_update_features(dev
);
1504 if (unlikely(dev
->features
& NETIF_F_LRO
))
1505 netdev_WARN(dev
, "failed to disable LRO!\n");
1507 netdev_for_each_lower_dev(dev
, lower_dev
, iter
)
1508 dev_disable_lro(lower_dev
);
1510 EXPORT_SYMBOL(dev_disable_lro
);
1512 static int call_netdevice_notifier(struct notifier_block
*nb
, unsigned long val
,
1513 struct net_device
*dev
)
1515 struct netdev_notifier_info info
;
1517 netdev_notifier_info_init(&info
, dev
);
1518 return nb
->notifier_call(nb
, val
, &info
);
1521 static int dev_boot_phase
= 1;
1524 * register_netdevice_notifier - register a network notifier block
1527 * Register a notifier to be called when network device events occur.
1528 * The notifier passed is linked into the kernel structures and must
1529 * not be reused until it has been unregistered. A negative errno code
1530 * is returned on a failure.
1532 * When registered all registration and up events are replayed
1533 * to the new notifier to allow device to have a race free
1534 * view of the network device list.
1537 int register_netdevice_notifier(struct notifier_block
*nb
)
1539 struct net_device
*dev
;
1540 struct net_device
*last
;
1545 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1551 for_each_netdev(net
, dev
) {
1552 err
= call_netdevice_notifier(nb
, NETDEV_REGISTER
, dev
);
1553 err
= notifier_to_errno(err
);
1557 if (!(dev
->flags
& IFF_UP
))
1560 call_netdevice_notifier(nb
, NETDEV_UP
, dev
);
1571 for_each_netdev(net
, dev
) {
1575 if (dev
->flags
& IFF_UP
) {
1576 call_netdevice_notifier(nb
, NETDEV_GOING_DOWN
,
1578 call_netdevice_notifier(nb
, NETDEV_DOWN
, dev
);
1580 call_netdevice_notifier(nb
, NETDEV_UNREGISTER
, dev
);
1585 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1588 EXPORT_SYMBOL(register_netdevice_notifier
);
1591 * unregister_netdevice_notifier - unregister a network notifier block
1594 * Unregister a notifier previously registered by
1595 * register_netdevice_notifier(). The notifier is unlinked into the
1596 * kernel structures and may then be reused. A negative errno code
1597 * is returned on a failure.
1599 * After unregistering unregister and down device events are synthesized
1600 * for all devices on the device list to the removed notifier to remove
1601 * the need for special case cleanup code.
1604 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1606 struct net_device
*dev
;
1611 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1616 for_each_netdev(net
, dev
) {
1617 if (dev
->flags
& IFF_UP
) {
1618 call_netdevice_notifier(nb
, NETDEV_GOING_DOWN
,
1620 call_netdevice_notifier(nb
, NETDEV_DOWN
, dev
);
1622 call_netdevice_notifier(nb
, NETDEV_UNREGISTER
, dev
);
1629 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1632 * call_netdevice_notifiers_info - call all network notifier blocks
1633 * @val: value passed unmodified to notifier function
1634 * @dev: net_device pointer passed unmodified to notifier function
1635 * @info: notifier information data
1637 * Call all network notifier blocks. Parameters and return value
1638 * are as for raw_notifier_call_chain().
1641 static int call_netdevice_notifiers_info(unsigned long val
,
1642 struct net_device
*dev
,
1643 struct netdev_notifier_info
*info
)
1646 netdev_notifier_info_init(info
, dev
);
1647 return raw_notifier_call_chain(&netdev_chain
, val
, info
);
1651 * call_netdevice_notifiers - call all network notifier blocks
1652 * @val: value passed unmodified to notifier function
1653 * @dev: net_device pointer passed unmodified to notifier function
1655 * Call all network notifier blocks. Parameters and return value
1656 * are as for raw_notifier_call_chain().
1659 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1661 struct netdev_notifier_info info
;
1663 return call_netdevice_notifiers_info(val
, dev
, &info
);
1665 EXPORT_SYMBOL(call_netdevice_notifiers
);
1667 #ifdef CONFIG_NET_INGRESS
1668 static struct static_key ingress_needed __read_mostly
;
1670 void net_inc_ingress_queue(void)
1672 static_key_slow_inc(&ingress_needed
);
1674 EXPORT_SYMBOL_GPL(net_inc_ingress_queue
);
1676 void net_dec_ingress_queue(void)
1678 static_key_slow_dec(&ingress_needed
);
1680 EXPORT_SYMBOL_GPL(net_dec_ingress_queue
);
1683 #ifdef CONFIG_NET_EGRESS
1684 static struct static_key egress_needed __read_mostly
;
1686 void net_inc_egress_queue(void)
1688 static_key_slow_inc(&egress_needed
);
1690 EXPORT_SYMBOL_GPL(net_inc_egress_queue
);
1692 void net_dec_egress_queue(void)
1694 static_key_slow_dec(&egress_needed
);
1696 EXPORT_SYMBOL_GPL(net_dec_egress_queue
);
1699 static struct static_key netstamp_needed __read_mostly
;
1700 #ifdef HAVE_JUMP_LABEL
1701 static atomic_t netstamp_needed_deferred
;
1702 static atomic_t netstamp_wanted
;
1703 static void netstamp_clear(struct work_struct
*work
)
1705 int deferred
= atomic_xchg(&netstamp_needed_deferred
, 0);
1708 wanted
= atomic_add_return(deferred
, &netstamp_wanted
);
1710 static_key_enable(&netstamp_needed
);
1712 static_key_disable(&netstamp_needed
);
1714 static DECLARE_WORK(netstamp_work
, netstamp_clear
);
1717 void net_enable_timestamp(void)
1719 #ifdef HAVE_JUMP_LABEL
1723 wanted
= atomic_read(&netstamp_wanted
);
1726 if (atomic_cmpxchg(&netstamp_wanted
, wanted
, wanted
+ 1) == wanted
)
1729 atomic_inc(&netstamp_needed_deferred
);
1730 schedule_work(&netstamp_work
);
1732 static_key_slow_inc(&netstamp_needed
);
1735 EXPORT_SYMBOL(net_enable_timestamp
);
1737 void net_disable_timestamp(void)
1739 #ifdef HAVE_JUMP_LABEL
1743 wanted
= atomic_read(&netstamp_wanted
);
1746 if (atomic_cmpxchg(&netstamp_wanted
, wanted
, wanted
- 1) == wanted
)
1749 atomic_dec(&netstamp_needed_deferred
);
1750 schedule_work(&netstamp_work
);
1752 static_key_slow_dec(&netstamp_needed
);
1755 EXPORT_SYMBOL(net_disable_timestamp
);
1757 static inline void net_timestamp_set(struct sk_buff
*skb
)
1760 if (static_key_false(&netstamp_needed
))
1761 __net_timestamp(skb
);
1764 #define net_timestamp_check(COND, SKB) \
1765 if (static_key_false(&netstamp_needed)) { \
1766 if ((COND) && !(SKB)->tstamp) \
1767 __net_timestamp(SKB); \
1770 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
1774 if (!(dev
->flags
& IFF_UP
))
1777 len
= dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
;
1778 if (skb
->len
<= len
)
1781 /* if TSO is enabled, we don't care about the length as the packet
1782 * could be forwarded without being segmented before
1784 if (skb_is_gso(skb
))
1789 EXPORT_SYMBOL_GPL(is_skb_forwardable
);
1791 int __dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1793 int ret
= ____dev_forward_skb(dev
, skb
);
1796 skb
->protocol
= eth_type_trans(skb
, dev
);
1797 skb_postpull_rcsum(skb
, eth_hdr(skb
), ETH_HLEN
);
1802 EXPORT_SYMBOL_GPL(__dev_forward_skb
);
1805 * dev_forward_skb - loopback an skb to another netif
1807 * @dev: destination network device
1808 * @skb: buffer to forward
1811 * NET_RX_SUCCESS (no congestion)
1812 * NET_RX_DROP (packet was dropped, but freed)
1814 * dev_forward_skb can be used for injecting an skb from the
1815 * start_xmit function of one device into the receive queue
1816 * of another device.
1818 * The receiving device may be in another namespace, so
1819 * we have to clear all information in the skb that could
1820 * impact namespace isolation.
1822 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1824 return __dev_forward_skb(dev
, skb
) ?: netif_rx_internal(skb
);
1826 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1828 static inline int deliver_skb(struct sk_buff
*skb
,
1829 struct packet_type
*pt_prev
,
1830 struct net_device
*orig_dev
)
1832 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
1834 atomic_inc(&skb
->users
);
1835 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1838 static inline void deliver_ptype_list_skb(struct sk_buff
*skb
,
1839 struct packet_type
**pt
,
1840 struct net_device
*orig_dev
,
1842 struct list_head
*ptype_list
)
1844 struct packet_type
*ptype
, *pt_prev
= *pt
;
1846 list_for_each_entry_rcu(ptype
, ptype_list
, list
) {
1847 if (ptype
->type
!= type
)
1850 deliver_skb(skb
, pt_prev
, orig_dev
);
1856 static inline bool skb_loop_sk(struct packet_type
*ptype
, struct sk_buff
*skb
)
1858 if (!ptype
->af_packet_priv
|| !skb
->sk
)
1861 if (ptype
->id_match
)
1862 return ptype
->id_match(ptype
, skb
->sk
);
1863 else if ((struct sock
*)ptype
->af_packet_priv
== skb
->sk
)
1870 * Support routine. Sends outgoing frames to any network
1871 * taps currently in use.
1874 void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1876 struct packet_type
*ptype
;
1877 struct sk_buff
*skb2
= NULL
;
1878 struct packet_type
*pt_prev
= NULL
;
1879 struct list_head
*ptype_list
= &ptype_all
;
1883 list_for_each_entry_rcu(ptype
, ptype_list
, list
) {
1884 /* Never send packets back to the socket
1885 * they originated from - MvS (miquels@drinkel.ow.org)
1887 if (skb_loop_sk(ptype
, skb
))
1891 deliver_skb(skb2
, pt_prev
, skb
->dev
);
1896 /* need to clone skb, done only once */
1897 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1901 net_timestamp_set(skb2
);
1903 /* skb->nh should be correctly
1904 * set by sender, so that the second statement is
1905 * just protection against buggy protocols.
1907 skb_reset_mac_header(skb2
);
1909 if (skb_network_header(skb2
) < skb2
->data
||
1910 skb_network_header(skb2
) > skb_tail_pointer(skb2
)) {
1911 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1912 ntohs(skb2
->protocol
),
1914 skb_reset_network_header(skb2
);
1917 skb2
->transport_header
= skb2
->network_header
;
1918 skb2
->pkt_type
= PACKET_OUTGOING
;
1922 if (ptype_list
== &ptype_all
) {
1923 ptype_list
= &dev
->ptype_all
;
1928 pt_prev
->func(skb2
, skb
->dev
, pt_prev
, skb
->dev
);
1931 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit
);
1934 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1935 * @dev: Network device
1936 * @txq: number of queues available
1938 * If real_num_tx_queues is changed the tc mappings may no longer be
1939 * valid. To resolve this verify the tc mapping remains valid and if
1940 * not NULL the mapping. With no priorities mapping to this
1941 * offset/count pair it will no longer be used. In the worst case TC0
1942 * is invalid nothing can be done so disable priority mappings. If is
1943 * expected that drivers will fix this mapping if they can before
1944 * calling netif_set_real_num_tx_queues.
1946 static void netif_setup_tc(struct net_device
*dev
, unsigned int txq
)
1949 struct netdev_tc_txq
*tc
= &dev
->tc_to_txq
[0];
1951 /* If TC0 is invalidated disable TC mapping */
1952 if (tc
->offset
+ tc
->count
> txq
) {
1953 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1958 /* Invalidated prio to tc mappings set to TC0 */
1959 for (i
= 1; i
< TC_BITMASK
+ 1; i
++) {
1960 int q
= netdev_get_prio_tc_map(dev
, i
);
1962 tc
= &dev
->tc_to_txq
[q
];
1963 if (tc
->offset
+ tc
->count
> txq
) {
1964 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1966 netdev_set_prio_tc_map(dev
, i
, 0);
1971 int netdev_txq_to_tc(struct net_device
*dev
, unsigned int txq
)
1974 struct netdev_tc_txq
*tc
= &dev
->tc_to_txq
[0];
1977 for (i
= 0; i
< TC_MAX_QUEUE
; i
++, tc
++) {
1978 if ((txq
- tc
->offset
) < tc
->count
)
1989 static DEFINE_MUTEX(xps_map_mutex
);
1990 #define xmap_dereference(P) \
1991 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1993 static bool remove_xps_queue(struct xps_dev_maps
*dev_maps
,
1996 struct xps_map
*map
= NULL
;
2000 map
= xmap_dereference(dev_maps
->cpu_map
[tci
]);
2004 for (pos
= map
->len
; pos
--;) {
2005 if (map
->queues
[pos
] != index
)
2009 map
->queues
[pos
] = map
->queues
[--map
->len
];
2013 RCU_INIT_POINTER(dev_maps
->cpu_map
[tci
], NULL
);
2014 kfree_rcu(map
, rcu
);
2021 static bool remove_xps_queue_cpu(struct net_device
*dev
,
2022 struct xps_dev_maps
*dev_maps
,
2023 int cpu
, u16 offset
, u16 count
)
2025 int num_tc
= dev
->num_tc
? : 1;
2026 bool active
= false;
2029 for (tci
= cpu
* num_tc
; num_tc
--; tci
++) {
2032 for (i
= count
, j
= offset
; i
--; j
++) {
2033 if (!remove_xps_queue(dev_maps
, cpu
, j
))
2043 static void netif_reset_xps_queues(struct net_device
*dev
, u16 offset
,
2046 struct xps_dev_maps
*dev_maps
;
2048 bool active
= false;
2050 mutex_lock(&xps_map_mutex
);
2051 dev_maps
= xmap_dereference(dev
->xps_maps
);
2056 for_each_possible_cpu(cpu
)
2057 active
|= remove_xps_queue_cpu(dev
, dev_maps
, cpu
,
2061 RCU_INIT_POINTER(dev
->xps_maps
, NULL
);
2062 kfree_rcu(dev_maps
, rcu
);
2065 for (i
= offset
+ (count
- 1); count
--; i
--)
2066 netdev_queue_numa_node_write(netdev_get_tx_queue(dev
, i
),
2070 mutex_unlock(&xps_map_mutex
);
2073 static void netif_reset_xps_queues_gt(struct net_device
*dev
, u16 index
)
2075 netif_reset_xps_queues(dev
, index
, dev
->num_tx_queues
- index
);
2078 static struct xps_map
*expand_xps_map(struct xps_map
*map
,
2081 struct xps_map
*new_map
;
2082 int alloc_len
= XPS_MIN_MAP_ALLOC
;
2085 for (pos
= 0; map
&& pos
< map
->len
; pos
++) {
2086 if (map
->queues
[pos
] != index
)
2091 /* Need to add queue to this CPU's existing map */
2093 if (pos
< map
->alloc_len
)
2096 alloc_len
= map
->alloc_len
* 2;
2099 /* Need to allocate new map to store queue on this CPU's map */
2100 new_map
= kzalloc_node(XPS_MAP_SIZE(alloc_len
), GFP_KERNEL
,
2105 for (i
= 0; i
< pos
; i
++)
2106 new_map
->queues
[i
] = map
->queues
[i
];
2107 new_map
->alloc_len
= alloc_len
;
2113 int netif_set_xps_queue(struct net_device
*dev
, const struct cpumask
*mask
,
2116 struct xps_dev_maps
*dev_maps
, *new_dev_maps
= NULL
;
2117 int i
, cpu
, tci
, numa_node_id
= -2;
2118 int maps_sz
, num_tc
= 1, tc
= 0;
2119 struct xps_map
*map
, *new_map
;
2120 bool active
= false;
2123 num_tc
= dev
->num_tc
;
2124 tc
= netdev_txq_to_tc(dev
, index
);
2129 maps_sz
= XPS_DEV_MAPS_SIZE(num_tc
);
2130 if (maps_sz
< L1_CACHE_BYTES
)
2131 maps_sz
= L1_CACHE_BYTES
;
2133 mutex_lock(&xps_map_mutex
);
2135 dev_maps
= xmap_dereference(dev
->xps_maps
);
2137 /* allocate memory for queue storage */
2138 for_each_cpu_and(cpu
, cpu_online_mask
, mask
) {
2140 new_dev_maps
= kzalloc(maps_sz
, GFP_KERNEL
);
2141 if (!new_dev_maps
) {
2142 mutex_unlock(&xps_map_mutex
);
2146 tci
= cpu
* num_tc
+ tc
;
2147 map
= dev_maps
? xmap_dereference(dev_maps
->cpu_map
[tci
]) :
2150 map
= expand_xps_map(map
, cpu
, index
);
2154 RCU_INIT_POINTER(new_dev_maps
->cpu_map
[tci
], map
);
2158 goto out_no_new_maps
;
2160 for_each_possible_cpu(cpu
) {
2161 /* copy maps belonging to foreign traffic classes */
2162 for (i
= tc
, tci
= cpu
* num_tc
; dev_maps
&& i
--; tci
++) {
2163 /* fill in the new device map from the old device map */
2164 map
= xmap_dereference(dev_maps
->cpu_map
[tci
]);
2165 RCU_INIT_POINTER(new_dev_maps
->cpu_map
[tci
], map
);
2168 /* We need to explicitly update tci as prevous loop
2169 * could break out early if dev_maps is NULL.
2171 tci
= cpu
* num_tc
+ tc
;
2173 if (cpumask_test_cpu(cpu
, mask
) && cpu_online(cpu
)) {
2174 /* add queue to CPU maps */
2177 map
= xmap_dereference(new_dev_maps
->cpu_map
[tci
]);
2178 while ((pos
< map
->len
) && (map
->queues
[pos
] != index
))
2181 if (pos
== map
->len
)
2182 map
->queues
[map
->len
++] = index
;
2184 if (numa_node_id
== -2)
2185 numa_node_id
= cpu_to_node(cpu
);
2186 else if (numa_node_id
!= cpu_to_node(cpu
))
2189 } else if (dev_maps
) {
2190 /* fill in the new device map from the old device map */
2191 map
= xmap_dereference(dev_maps
->cpu_map
[tci
]);
2192 RCU_INIT_POINTER(new_dev_maps
->cpu_map
[tci
], map
);
2195 /* copy maps belonging to foreign traffic classes */
2196 for (i
= num_tc
- tc
, tci
++; dev_maps
&& --i
; tci
++) {
2197 /* fill in the new device map from the old device map */
2198 map
= xmap_dereference(dev_maps
->cpu_map
[tci
]);
2199 RCU_INIT_POINTER(new_dev_maps
->cpu_map
[tci
], map
);
2203 rcu_assign_pointer(dev
->xps_maps
, new_dev_maps
);
2205 /* Cleanup old maps */
2207 goto out_no_old_maps
;
2209 for_each_possible_cpu(cpu
) {
2210 for (i
= num_tc
, tci
= cpu
* num_tc
; i
--; tci
++) {
2211 new_map
= xmap_dereference(new_dev_maps
->cpu_map
[tci
]);
2212 map
= xmap_dereference(dev_maps
->cpu_map
[tci
]);
2213 if (map
&& map
!= new_map
)
2214 kfree_rcu(map
, rcu
);
2218 kfree_rcu(dev_maps
, rcu
);
2221 dev_maps
= new_dev_maps
;
2225 /* update Tx queue numa node */
2226 netdev_queue_numa_node_write(netdev_get_tx_queue(dev
, index
),
2227 (numa_node_id
>= 0) ? numa_node_id
:
2233 /* removes queue from unused CPUs */
2234 for_each_possible_cpu(cpu
) {
2235 for (i
= tc
, tci
= cpu
* num_tc
; i
--; tci
++)
2236 active
|= remove_xps_queue(dev_maps
, tci
, index
);
2237 if (!cpumask_test_cpu(cpu
, mask
) || !cpu_online(cpu
))
2238 active
|= remove_xps_queue(dev_maps
, tci
, index
);
2239 for (i
= num_tc
- tc
, tci
++; --i
; tci
++)
2240 active
|= remove_xps_queue(dev_maps
, tci
, index
);
2243 /* free map if not active */
2245 RCU_INIT_POINTER(dev
->xps_maps
, NULL
);
2246 kfree_rcu(dev_maps
, rcu
);
2250 mutex_unlock(&xps_map_mutex
);
2254 /* remove any maps that we added */
2255 for_each_possible_cpu(cpu
) {
2256 for (i
= num_tc
, tci
= cpu
* num_tc
; i
--; tci
++) {
2257 new_map
= xmap_dereference(new_dev_maps
->cpu_map
[tci
]);
2259 xmap_dereference(dev_maps
->cpu_map
[tci
]) :
2261 if (new_map
&& new_map
!= map
)
2266 mutex_unlock(&xps_map_mutex
);
2268 kfree(new_dev_maps
);
2271 EXPORT_SYMBOL(netif_set_xps_queue
);
2274 void netdev_reset_tc(struct net_device
*dev
)
2277 netif_reset_xps_queues_gt(dev
, 0);
2280 memset(dev
->tc_to_txq
, 0, sizeof(dev
->tc_to_txq
));
2281 memset(dev
->prio_tc_map
, 0, sizeof(dev
->prio_tc_map
));
2283 EXPORT_SYMBOL(netdev_reset_tc
);
2285 int netdev_set_tc_queue(struct net_device
*dev
, u8 tc
, u16 count
, u16 offset
)
2287 if (tc
>= dev
->num_tc
)
2291 netif_reset_xps_queues(dev
, offset
, count
);
2293 dev
->tc_to_txq
[tc
].count
= count
;
2294 dev
->tc_to_txq
[tc
].offset
= offset
;
2297 EXPORT_SYMBOL(netdev_set_tc_queue
);
2299 int netdev_set_num_tc(struct net_device
*dev
, u8 num_tc
)
2301 if (num_tc
> TC_MAX_QUEUE
)
2305 netif_reset_xps_queues_gt(dev
, 0);
2307 dev
->num_tc
= num_tc
;
2310 EXPORT_SYMBOL(netdev_set_num_tc
);
2313 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2314 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2316 int netif_set_real_num_tx_queues(struct net_device
*dev
, unsigned int txq
)
2320 if (txq
< 1 || txq
> dev
->num_tx_queues
)
2323 if (dev
->reg_state
== NETREG_REGISTERED
||
2324 dev
->reg_state
== NETREG_UNREGISTERING
) {
2327 rc
= netdev_queue_update_kobjects(dev
, dev
->real_num_tx_queues
,
2333 netif_setup_tc(dev
, txq
);
2335 if (txq
< dev
->real_num_tx_queues
) {
2336 qdisc_reset_all_tx_gt(dev
, txq
);
2338 netif_reset_xps_queues_gt(dev
, txq
);
2343 dev
->real_num_tx_queues
= txq
;
2346 EXPORT_SYMBOL(netif_set_real_num_tx_queues
);
2350 * netif_set_real_num_rx_queues - set actual number of RX queues used
2351 * @dev: Network device
2352 * @rxq: Actual number of RX queues
2354 * This must be called either with the rtnl_lock held or before
2355 * registration of the net device. Returns 0 on success, or a
2356 * negative error code. If called before registration, it always
2359 int netif_set_real_num_rx_queues(struct net_device
*dev
, unsigned int rxq
)
2363 if (rxq
< 1 || rxq
> dev
->num_rx_queues
)
2366 if (dev
->reg_state
== NETREG_REGISTERED
) {
2369 rc
= net_rx_queue_update_kobjects(dev
, dev
->real_num_rx_queues
,
2375 dev
->real_num_rx_queues
= rxq
;
2378 EXPORT_SYMBOL(netif_set_real_num_rx_queues
);
2382 * netif_get_num_default_rss_queues - default number of RSS queues
2384 * This routine should set an upper limit on the number of RSS queues
2385 * used by default by multiqueue devices.
2387 int netif_get_num_default_rss_queues(void)
2389 return is_kdump_kernel() ?
2390 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES
, num_online_cpus());
2392 EXPORT_SYMBOL(netif_get_num_default_rss_queues
);
2394 static void __netif_reschedule(struct Qdisc
*q
)
2396 struct softnet_data
*sd
;
2397 unsigned long flags
;
2399 local_irq_save(flags
);
2400 sd
= this_cpu_ptr(&softnet_data
);
2401 q
->next_sched
= NULL
;
2402 *sd
->output_queue_tailp
= q
;
2403 sd
->output_queue_tailp
= &q
->next_sched
;
2404 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
2405 local_irq_restore(flags
);
2408 void __netif_schedule(struct Qdisc
*q
)
2410 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
2411 __netif_reschedule(q
);
2413 EXPORT_SYMBOL(__netif_schedule
);
2415 struct dev_kfree_skb_cb
{
2416 enum skb_free_reason reason
;
2419 static struct dev_kfree_skb_cb
*get_kfree_skb_cb(const struct sk_buff
*skb
)
2421 return (struct dev_kfree_skb_cb
*)skb
->cb
;
2424 void netif_schedule_queue(struct netdev_queue
*txq
)
2427 if (!(txq
->state
& QUEUE_STATE_ANY_XOFF
)) {
2428 struct Qdisc
*q
= rcu_dereference(txq
->qdisc
);
2430 __netif_schedule(q
);
2434 EXPORT_SYMBOL(netif_schedule_queue
);
2436 void netif_tx_wake_queue(struct netdev_queue
*dev_queue
)
2438 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF
, &dev_queue
->state
)) {
2442 q
= rcu_dereference(dev_queue
->qdisc
);
2443 __netif_schedule(q
);
2447 EXPORT_SYMBOL(netif_tx_wake_queue
);
2449 void __dev_kfree_skb_irq(struct sk_buff
*skb
, enum skb_free_reason reason
)
2451 unsigned long flags
;
2453 if (likely(atomic_read(&skb
->users
) == 1)) {
2455 atomic_set(&skb
->users
, 0);
2456 } else if (likely(!atomic_dec_and_test(&skb
->users
))) {
2459 get_kfree_skb_cb(skb
)->reason
= reason
;
2460 local_irq_save(flags
);
2461 skb
->next
= __this_cpu_read(softnet_data
.completion_queue
);
2462 __this_cpu_write(softnet_data
.completion_queue
, skb
);
2463 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
2464 local_irq_restore(flags
);
2466 EXPORT_SYMBOL(__dev_kfree_skb_irq
);
2468 void __dev_kfree_skb_any(struct sk_buff
*skb
, enum skb_free_reason reason
)
2470 if (in_irq() || irqs_disabled())
2471 __dev_kfree_skb_irq(skb
, reason
);
2475 EXPORT_SYMBOL(__dev_kfree_skb_any
);
2479 * netif_device_detach - mark device as removed
2480 * @dev: network device
2482 * Mark device as removed from system and therefore no longer available.
2484 void netif_device_detach(struct net_device
*dev
)
2486 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
2487 netif_running(dev
)) {
2488 netif_tx_stop_all_queues(dev
);
2491 EXPORT_SYMBOL(netif_device_detach
);
2494 * netif_device_attach - mark device as attached
2495 * @dev: network device
2497 * Mark device as attached from system and restart if needed.
2499 void netif_device_attach(struct net_device
*dev
)
2501 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
2502 netif_running(dev
)) {
2503 netif_tx_wake_all_queues(dev
);
2504 __netdev_watchdog_up(dev
);
2507 EXPORT_SYMBOL(netif_device_attach
);
2510 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2511 * to be used as a distribution range.
2513 u16
__skb_tx_hash(const struct net_device
*dev
, struct sk_buff
*skb
,
2514 unsigned int num_tx_queues
)
2518 u16 qcount
= num_tx_queues
;
2520 if (skb_rx_queue_recorded(skb
)) {
2521 hash
= skb_get_rx_queue(skb
);
2522 while (unlikely(hash
>= num_tx_queues
))
2523 hash
-= num_tx_queues
;
2528 u8 tc
= netdev_get_prio_tc_map(dev
, skb
->priority
);
2530 qoffset
= dev
->tc_to_txq
[tc
].offset
;
2531 qcount
= dev
->tc_to_txq
[tc
].count
;
2534 return (u16
) reciprocal_scale(skb_get_hash(skb
), qcount
) + qoffset
;
2536 EXPORT_SYMBOL(__skb_tx_hash
);
2538 static void skb_warn_bad_offload(const struct sk_buff
*skb
)
2540 static const netdev_features_t null_features
;
2541 struct net_device
*dev
= skb
->dev
;
2542 const char *name
= "";
2544 if (!net_ratelimit())
2548 if (dev
->dev
.parent
)
2549 name
= dev_driver_string(dev
->dev
.parent
);
2551 name
= netdev_name(dev
);
2553 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2554 "gso_type=%d ip_summed=%d\n",
2555 name
, dev
? &dev
->features
: &null_features
,
2556 skb
->sk
? &skb
->sk
->sk_route_caps
: &null_features
,
2557 skb
->len
, skb
->data_len
, skb_shinfo(skb
)->gso_size
,
2558 skb_shinfo(skb
)->gso_type
, skb
->ip_summed
);
2562 * Invalidate hardware checksum when packet is to be mangled, and
2563 * complete checksum manually on outgoing path.
2565 int skb_checksum_help(struct sk_buff
*skb
)
2568 int ret
= 0, offset
;
2570 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
2571 goto out_set_summed
;
2573 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
2574 skb_warn_bad_offload(skb
);
2578 /* Before computing a checksum, we should make sure no frag could
2579 * be modified by an external entity : checksum could be wrong.
2581 if (skb_has_shared_frag(skb
)) {
2582 ret
= __skb_linearize(skb
);
2587 offset
= skb_checksum_start_offset(skb
);
2588 BUG_ON(offset
>= skb_headlen(skb
));
2589 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
2591 offset
+= skb
->csum_offset
;
2592 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
2594 if (skb_cloned(skb
) &&
2595 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
2596 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2601 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
) ?: CSUM_MANGLED_0
;
2603 skb
->ip_summed
= CHECKSUM_NONE
;
2607 EXPORT_SYMBOL(skb_checksum_help
);
2609 __be16
skb_network_protocol(struct sk_buff
*skb
, int *depth
)
2611 __be16 type
= skb
->protocol
;
2613 /* Tunnel gso handlers can set protocol to ethernet. */
2614 if (type
== htons(ETH_P_TEB
)) {
2617 if (unlikely(!pskb_may_pull(skb
, sizeof(struct ethhdr
))))
2620 eth
= (struct ethhdr
*)skb_mac_header(skb
);
2621 type
= eth
->h_proto
;
2624 return __vlan_get_protocol(skb
, type
, depth
);
2628 * skb_mac_gso_segment - mac layer segmentation handler.
2629 * @skb: buffer to segment
2630 * @features: features for the output path (see dev->features)
2632 struct sk_buff
*skb_mac_gso_segment(struct sk_buff
*skb
,
2633 netdev_features_t features
)
2635 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
2636 struct packet_offload
*ptype
;
2637 int vlan_depth
= skb
->mac_len
;
2638 __be16 type
= skb_network_protocol(skb
, &vlan_depth
);
2640 if (unlikely(!type
))
2641 return ERR_PTR(-EINVAL
);
2643 __skb_pull(skb
, vlan_depth
);
2646 list_for_each_entry_rcu(ptype
, &offload_base
, list
) {
2647 if (ptype
->type
== type
&& ptype
->callbacks
.gso_segment
) {
2648 segs
= ptype
->callbacks
.gso_segment(skb
, features
);
2654 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
2658 EXPORT_SYMBOL(skb_mac_gso_segment
);
2661 /* openvswitch calls this on rx path, so we need a different check.
2663 static inline bool skb_needs_check(struct sk_buff
*skb
, bool tx_path
)
2666 return skb
->ip_summed
!= CHECKSUM_PARTIAL
&&
2667 skb
->ip_summed
!= CHECKSUM_NONE
;
2669 return skb
->ip_summed
== CHECKSUM_NONE
;
2673 * __skb_gso_segment - Perform segmentation on skb.
2674 * @skb: buffer to segment
2675 * @features: features for the output path (see dev->features)
2676 * @tx_path: whether it is called in TX path
2678 * This function segments the given skb and returns a list of segments.
2680 * It may return NULL if the skb requires no segmentation. This is
2681 * only possible when GSO is used for verifying header integrity.
2683 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
2685 struct sk_buff
*__skb_gso_segment(struct sk_buff
*skb
,
2686 netdev_features_t features
, bool tx_path
)
2688 struct sk_buff
*segs
;
2690 if (unlikely(skb_needs_check(skb
, tx_path
))) {
2693 /* We're going to init ->check field in TCP or UDP header */
2694 err
= skb_cow_head(skb
, 0);
2696 return ERR_PTR(err
);
2699 /* Only report GSO partial support if it will enable us to
2700 * support segmentation on this frame without needing additional
2703 if (features
& NETIF_F_GSO_PARTIAL
) {
2704 netdev_features_t partial_features
= NETIF_F_GSO_ROBUST
;
2705 struct net_device
*dev
= skb
->dev
;
2707 partial_features
|= dev
->features
& dev
->gso_partial_features
;
2708 if (!skb_gso_ok(skb
, features
| partial_features
))
2709 features
&= ~NETIF_F_GSO_PARTIAL
;
2712 BUILD_BUG_ON(SKB_SGO_CB_OFFSET
+
2713 sizeof(*SKB_GSO_CB(skb
)) > sizeof(skb
->cb
));
2715 SKB_GSO_CB(skb
)->mac_offset
= skb_headroom(skb
);
2716 SKB_GSO_CB(skb
)->encap_level
= 0;
2718 skb_reset_mac_header(skb
);
2719 skb_reset_mac_len(skb
);
2721 segs
= skb_mac_gso_segment(skb
, features
);
2723 if (unlikely(skb_needs_check(skb
, tx_path
)))
2724 skb_warn_bad_offload(skb
);
2728 EXPORT_SYMBOL(__skb_gso_segment
);
2730 /* Take action when hardware reception checksum errors are detected. */
2732 void netdev_rx_csum_fault(struct net_device
*dev
)
2734 if (net_ratelimit()) {
2735 pr_err("%s: hw csum failure\n", dev
? dev
->name
: "<unknown>");
2739 EXPORT_SYMBOL(netdev_rx_csum_fault
);
2742 /* Actually, we should eliminate this check as soon as we know, that:
2743 * 1. IOMMU is present and allows to map all the memory.
2744 * 2. No high memory really exists on this machine.
2747 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
2749 #ifdef CONFIG_HIGHMEM
2752 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
2753 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2754 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2756 if (PageHighMem(skb_frag_page(frag
)))
2761 if (PCI_DMA_BUS_IS_PHYS
) {
2762 struct device
*pdev
= dev
->dev
.parent
;
2766 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2767 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2768 dma_addr_t addr
= page_to_phys(skb_frag_page(frag
));
2770 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
2778 /* If MPLS offload request, verify we are testing hardware MPLS features
2779 * instead of standard features for the netdev.
2781 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
2782 static netdev_features_t
net_mpls_features(struct sk_buff
*skb
,
2783 netdev_features_t features
,
2786 if (eth_p_mpls(type
))
2787 features
&= skb
->dev
->mpls_features
;
2792 static netdev_features_t
net_mpls_features(struct sk_buff
*skb
,
2793 netdev_features_t features
,
2800 static netdev_features_t
harmonize_features(struct sk_buff
*skb
,
2801 netdev_features_t features
)
2806 type
= skb_network_protocol(skb
, &tmp
);
2807 features
= net_mpls_features(skb
, features
, type
);
2809 if (skb
->ip_summed
!= CHECKSUM_NONE
&&
2810 !can_checksum_protocol(features
, type
)) {
2811 features
&= ~(NETIF_F_CSUM_MASK
| NETIF_F_GSO_MASK
);
2813 if (illegal_highdma(skb
->dev
, skb
))
2814 features
&= ~NETIF_F_SG
;
2819 netdev_features_t
passthru_features_check(struct sk_buff
*skb
,
2820 struct net_device
*dev
,
2821 netdev_features_t features
)
2825 EXPORT_SYMBOL(passthru_features_check
);
2827 static netdev_features_t
dflt_features_check(const struct sk_buff
*skb
,
2828 struct net_device
*dev
,
2829 netdev_features_t features
)
2831 return vlan_features_check(skb
, features
);
2834 static netdev_features_t
gso_features_check(const struct sk_buff
*skb
,
2835 struct net_device
*dev
,
2836 netdev_features_t features
)
2838 u16 gso_segs
= skb_shinfo(skb
)->gso_segs
;
2840 if (gso_segs
> dev
->gso_max_segs
)
2841 return features
& ~NETIF_F_GSO_MASK
;
2843 /* Support for GSO partial features requires software
2844 * intervention before we can actually process the packets
2845 * so we need to strip support for any partial features now
2846 * and we can pull them back in after we have partially
2847 * segmented the frame.
2849 if (!(skb_shinfo(skb
)->gso_type
& SKB_GSO_PARTIAL
))
2850 features
&= ~dev
->gso_partial_features
;
2852 /* Make sure to clear the IPv4 ID mangling feature if the
2853 * IPv4 header has the potential to be fragmented.
2855 if (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCPV4
) {
2856 struct iphdr
*iph
= skb
->encapsulation
?
2857 inner_ip_hdr(skb
) : ip_hdr(skb
);
2859 if (!(iph
->frag_off
& htons(IP_DF
)))
2860 features
&= ~NETIF_F_TSO_MANGLEID
;
2866 netdev_features_t
netif_skb_features(struct sk_buff
*skb
)
2868 struct net_device
*dev
= skb
->dev
;
2869 netdev_features_t features
= dev
->features
;
2871 if (skb_is_gso(skb
))
2872 features
= gso_features_check(skb
, dev
, features
);
2874 /* If encapsulation offload request, verify we are testing
2875 * hardware encapsulation features instead of standard
2876 * features for the netdev
2878 if (skb
->encapsulation
)
2879 features
&= dev
->hw_enc_features
;
2881 if (skb_vlan_tagged(skb
))
2882 features
= netdev_intersect_features(features
,
2883 dev
->vlan_features
|
2884 NETIF_F_HW_VLAN_CTAG_TX
|
2885 NETIF_F_HW_VLAN_STAG_TX
);
2887 if (dev
->netdev_ops
->ndo_features_check
)
2888 features
&= dev
->netdev_ops
->ndo_features_check(skb
, dev
,
2891 features
&= dflt_features_check(skb
, dev
, features
);
2893 return harmonize_features(skb
, features
);
2895 EXPORT_SYMBOL(netif_skb_features
);
2897 static int xmit_one(struct sk_buff
*skb
, struct net_device
*dev
,
2898 struct netdev_queue
*txq
, bool more
)
2903 if (!list_empty(&ptype_all
) || !list_empty(&dev
->ptype_all
))
2904 dev_queue_xmit_nit(skb
, dev
);
2907 trace_net_dev_start_xmit(skb
, dev
);
2908 rc
= netdev_start_xmit(skb
, dev
, txq
, more
);
2909 trace_net_dev_xmit(skb
, rc
, dev
, len
);
2914 struct sk_buff
*dev_hard_start_xmit(struct sk_buff
*first
, struct net_device
*dev
,
2915 struct netdev_queue
*txq
, int *ret
)
2917 struct sk_buff
*skb
= first
;
2918 int rc
= NETDEV_TX_OK
;
2921 struct sk_buff
*next
= skb
->next
;
2924 rc
= xmit_one(skb
, dev
, txq
, next
!= NULL
);
2925 if (unlikely(!dev_xmit_complete(rc
))) {
2931 if (netif_xmit_stopped(txq
) && skb
) {
2932 rc
= NETDEV_TX_BUSY
;
2942 static struct sk_buff
*validate_xmit_vlan(struct sk_buff
*skb
,
2943 netdev_features_t features
)
2945 if (skb_vlan_tag_present(skb
) &&
2946 !vlan_hw_offload_capable(features
, skb
->vlan_proto
))
2947 skb
= __vlan_hwaccel_push_inside(skb
);
2951 static struct sk_buff
*validate_xmit_skb(struct sk_buff
*skb
, struct net_device
*dev
)
2953 netdev_features_t features
;
2955 features
= netif_skb_features(skb
);
2956 skb
= validate_xmit_vlan(skb
, features
);
2960 if (netif_needs_gso(skb
, features
)) {
2961 struct sk_buff
*segs
;
2963 segs
= skb_gso_segment(skb
, features
);
2971 if (skb_needs_linearize(skb
, features
) &&
2972 __skb_linearize(skb
))
2975 /* If packet is not checksummed and device does not
2976 * support checksumming for this protocol, complete
2977 * checksumming here.
2979 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2980 if (skb
->encapsulation
)
2981 skb_set_inner_transport_header(skb
,
2982 skb_checksum_start_offset(skb
));
2984 skb_set_transport_header(skb
,
2985 skb_checksum_start_offset(skb
));
2986 if (!(features
& NETIF_F_CSUM_MASK
) &&
2987 skb_checksum_help(skb
))
2997 atomic_long_inc(&dev
->tx_dropped
);
3001 struct sk_buff
*validate_xmit_skb_list(struct sk_buff
*skb
, struct net_device
*dev
)
3003 struct sk_buff
*next
, *head
= NULL
, *tail
;
3005 for (; skb
!= NULL
; skb
= next
) {
3009 /* in case skb wont be segmented, point to itself */
3012 skb
= validate_xmit_skb(skb
, dev
);
3020 /* If skb was segmented, skb->prev points to
3021 * the last segment. If not, it still contains skb.
3027 EXPORT_SYMBOL_GPL(validate_xmit_skb_list
);
3029 static void qdisc_pkt_len_init(struct sk_buff
*skb
)
3031 const struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
3033 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
3035 /* To get more precise estimation of bytes sent on wire,
3036 * we add to pkt_len the headers size of all segments
3038 if (shinfo
->gso_size
) {
3039 unsigned int hdr_len
;
3040 u16 gso_segs
= shinfo
->gso_segs
;
3042 /* mac layer + network layer */
3043 hdr_len
= skb_transport_header(skb
) - skb_mac_header(skb
);
3045 /* + transport layer */
3046 if (likely(shinfo
->gso_type
& (SKB_GSO_TCPV4
| SKB_GSO_TCPV6
)))
3047 hdr_len
+= tcp_hdrlen(skb
);
3049 hdr_len
+= sizeof(struct udphdr
);
3051 if (shinfo
->gso_type
& SKB_GSO_DODGY
)
3052 gso_segs
= DIV_ROUND_UP(skb
->len
- hdr_len
,
3055 qdisc_skb_cb(skb
)->pkt_len
+= (gso_segs
- 1) * hdr_len
;
3059 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
3060 struct net_device
*dev
,
3061 struct netdev_queue
*txq
)
3063 spinlock_t
*root_lock
= qdisc_lock(q
);
3064 struct sk_buff
*to_free
= NULL
;
3068 qdisc_calculate_pkt_len(skb
, q
);
3070 * Heuristic to force contended enqueues to serialize on a
3071 * separate lock before trying to get qdisc main lock.
3072 * This permits qdisc->running owner to get the lock more
3073 * often and dequeue packets faster.
3075 contended
= qdisc_is_running(q
);
3076 if (unlikely(contended
))
3077 spin_lock(&q
->busylock
);
3079 spin_lock(root_lock
);
3080 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
3081 __qdisc_drop(skb
, &to_free
);
3083 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
3084 qdisc_run_begin(q
)) {
3086 * This is a work-conserving queue; there are no old skbs
3087 * waiting to be sent out; and the qdisc is not running -
3088 * xmit the skb directly.
3091 qdisc_bstats_update(q
, skb
);
3093 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
, true)) {
3094 if (unlikely(contended
)) {
3095 spin_unlock(&q
->busylock
);
3102 rc
= NET_XMIT_SUCCESS
;
3104 rc
= q
->enqueue(skb
, q
, &to_free
) & NET_XMIT_MASK
;
3105 if (qdisc_run_begin(q
)) {
3106 if (unlikely(contended
)) {
3107 spin_unlock(&q
->busylock
);
3113 spin_unlock(root_lock
);
3114 if (unlikely(to_free
))
3115 kfree_skb_list(to_free
);
3116 if (unlikely(contended
))
3117 spin_unlock(&q
->busylock
);
3121 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3122 static void skb_update_prio(struct sk_buff
*skb
)
3124 struct netprio_map
*map
= rcu_dereference_bh(skb
->dev
->priomap
);
3126 if (!skb
->priority
&& skb
->sk
&& map
) {
3127 unsigned int prioidx
=
3128 sock_cgroup_prioidx(&skb
->sk
->sk_cgrp_data
);
3130 if (prioidx
< map
->priomap_len
)
3131 skb
->priority
= map
->priomap
[prioidx
];
3135 #define skb_update_prio(skb)
3138 DEFINE_PER_CPU(int, xmit_recursion
);
3139 EXPORT_SYMBOL(xmit_recursion
);
3142 * dev_loopback_xmit - loop back @skb
3143 * @net: network namespace this loopback is happening in
3144 * @sk: sk needed to be a netfilter okfn
3145 * @skb: buffer to transmit
3147 int dev_loopback_xmit(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
3149 skb_reset_mac_header(skb
);
3150 __skb_pull(skb
, skb_network_offset(skb
));
3151 skb
->pkt_type
= PACKET_LOOPBACK
;
3152 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
3153 WARN_ON(!skb_dst(skb
));
3158 EXPORT_SYMBOL(dev_loopback_xmit
);
3160 #ifdef CONFIG_NET_EGRESS
3161 static struct sk_buff
*
3162 sch_handle_egress(struct sk_buff
*skb
, int *ret
, struct net_device
*dev
)
3164 struct tcf_proto
*cl
= rcu_dereference_bh(dev
->egress_cl_list
);
3165 struct tcf_result cl_res
;
3170 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3171 qdisc_bstats_cpu_update(cl
->q
, skb
);
3173 switch (tc_classify(skb
, cl
, &cl_res
, false)) {
3175 case TC_ACT_RECLASSIFY
:
3176 skb
->tc_index
= TC_H_MIN(cl_res
.classid
);
3179 qdisc_qstats_cpu_drop(cl
->q
);
3180 *ret
= NET_XMIT_DROP
;
3185 *ret
= NET_XMIT_SUCCESS
;
3188 case TC_ACT_REDIRECT
:
3189 /* No need to push/pop skb's mac_header here on egress! */
3190 skb_do_redirect(skb
);
3191 *ret
= NET_XMIT_SUCCESS
;
3199 #endif /* CONFIG_NET_EGRESS */
3201 static inline int get_xps_queue(struct net_device
*dev
, struct sk_buff
*skb
)
3204 struct xps_dev_maps
*dev_maps
;
3205 struct xps_map
*map
;
3206 int queue_index
= -1;
3209 dev_maps
= rcu_dereference(dev
->xps_maps
);
3211 unsigned int tci
= skb
->sender_cpu
- 1;
3215 tci
+= netdev_get_prio_tc_map(dev
, skb
->priority
);
3218 map
= rcu_dereference(dev_maps
->cpu_map
[tci
]);
3221 queue_index
= map
->queues
[0];
3223 queue_index
= map
->queues
[reciprocal_scale(skb_get_hash(skb
),
3225 if (unlikely(queue_index
>= dev
->real_num_tx_queues
))
3237 static u16
__netdev_pick_tx(struct net_device
*dev
, struct sk_buff
*skb
)
3239 struct sock
*sk
= skb
->sk
;
3240 int queue_index
= sk_tx_queue_get(sk
);
3242 if (queue_index
< 0 || skb
->ooo_okay
||
3243 queue_index
>= dev
->real_num_tx_queues
) {
3244 int new_index
= get_xps_queue(dev
, skb
);
3247 new_index
= skb_tx_hash(dev
, skb
);
3249 if (queue_index
!= new_index
&& sk
&&
3251 rcu_access_pointer(sk
->sk_dst_cache
))
3252 sk_tx_queue_set(sk
, new_index
);
3254 queue_index
= new_index
;
3260 struct netdev_queue
*netdev_pick_tx(struct net_device
*dev
,
3261 struct sk_buff
*skb
,
3264 int queue_index
= 0;
3267 u32 sender_cpu
= skb
->sender_cpu
- 1;
3269 if (sender_cpu
>= (u32
)NR_CPUS
)
3270 skb
->sender_cpu
= raw_smp_processor_id() + 1;
3273 if (dev
->real_num_tx_queues
!= 1) {
3274 const struct net_device_ops
*ops
= dev
->netdev_ops
;
3276 if (ops
->ndo_select_queue
)
3277 queue_index
= ops
->ndo_select_queue(dev
, skb
, accel_priv
,
3280 queue_index
= __netdev_pick_tx(dev
, skb
);
3283 queue_index
= netdev_cap_txqueue(dev
, queue_index
);
3286 skb_set_queue_mapping(skb
, queue_index
);
3287 return netdev_get_tx_queue(dev
, queue_index
);
3291 * __dev_queue_xmit - transmit a buffer
3292 * @skb: buffer to transmit
3293 * @accel_priv: private data used for L2 forwarding offload
3295 * Queue a buffer for transmission to a network device. The caller must
3296 * have set the device and priority and built the buffer before calling
3297 * this function. The function can be called from an interrupt.
3299 * A negative errno code is returned on a failure. A success does not
3300 * guarantee the frame will be transmitted as it may be dropped due
3301 * to congestion or traffic shaping.
3303 * -----------------------------------------------------------------------------------
3304 * I notice this method can also return errors from the queue disciplines,
3305 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3308 * Regardless of the return value, the skb is consumed, so it is currently
3309 * difficult to retry a send to this method. (You can bump the ref count
3310 * before sending to hold a reference for retry if you are careful.)
3312 * When calling this method, interrupts MUST be enabled. This is because
3313 * the BH enable code must have IRQs enabled so that it will not deadlock.
3316 static int __dev_queue_xmit(struct sk_buff
*skb
, void *accel_priv
)
3318 struct net_device
*dev
= skb
->dev
;
3319 struct netdev_queue
*txq
;
3323 skb_reset_mac_header(skb
);
3325 if (unlikely(skb_shinfo(skb
)->tx_flags
& SKBTX_SCHED_TSTAMP
))
3326 __skb_tstamp_tx(skb
, NULL
, skb
->sk
, SCM_TSTAMP_SCHED
);
3328 /* Disable soft irqs for various locks below. Also
3329 * stops preemption for RCU.
3333 skb_update_prio(skb
);
3335 qdisc_pkt_len_init(skb
);
3336 #ifdef CONFIG_NET_CLS_ACT
3337 skb
->tc_at_ingress
= 0;
3338 # ifdef CONFIG_NET_EGRESS
3339 if (static_key_false(&egress_needed
)) {
3340 skb
= sch_handle_egress(skb
, &rc
, dev
);
3346 /* If device/qdisc don't need skb->dst, release it right now while
3347 * its hot in this cpu cache.
3349 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
3354 txq
= netdev_pick_tx(dev
, skb
, accel_priv
);
3355 q
= rcu_dereference_bh(txq
->qdisc
);
3357 trace_net_dev_queue(skb
);
3359 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
3363 /* The device has no queue. Common case for software devices:
3364 * loopback, all the sorts of tunnels...
3366 * Really, it is unlikely that netif_tx_lock protection is necessary
3367 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
3369 * However, it is possible, that they rely on protection
3372 * Check this and shot the lock. It is not prone from deadlocks.
3373 *Either shot noqueue qdisc, it is even simpler 8)
3375 if (dev
->flags
& IFF_UP
) {
3376 int cpu
= smp_processor_id(); /* ok because BHs are off */
3378 if (txq
->xmit_lock_owner
!= cpu
) {
3379 if (unlikely(__this_cpu_read(xmit_recursion
) >
3380 XMIT_RECURSION_LIMIT
))
3381 goto recursion_alert
;
3383 skb
= validate_xmit_skb(skb
, dev
);
3387 HARD_TX_LOCK(dev
, txq
, cpu
);
3389 if (!netif_xmit_stopped(txq
)) {
3390 __this_cpu_inc(xmit_recursion
);
3391 skb
= dev_hard_start_xmit(skb
, dev
, txq
, &rc
);
3392 __this_cpu_dec(xmit_recursion
);
3393 if (dev_xmit_complete(rc
)) {
3394 HARD_TX_UNLOCK(dev
, txq
);
3398 HARD_TX_UNLOCK(dev
, txq
);
3399 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3402 /* Recursion is detected! It is possible,
3406 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3412 rcu_read_unlock_bh();
3414 atomic_long_inc(&dev
->tx_dropped
);
3415 kfree_skb_list(skb
);
3418 rcu_read_unlock_bh();
3422 int dev_queue_xmit(struct sk_buff
*skb
)
3424 return __dev_queue_xmit(skb
, NULL
);
3426 EXPORT_SYMBOL(dev_queue_xmit
);
3428 int dev_queue_xmit_accel(struct sk_buff
*skb
, void *accel_priv
)
3430 return __dev_queue_xmit(skb
, accel_priv
);
3432 EXPORT_SYMBOL(dev_queue_xmit_accel
);
3435 /*************************************************************************
3437 *************************************************************************/
3439 int netdev_max_backlog __read_mostly
= 1000;
3440 EXPORT_SYMBOL(netdev_max_backlog
);
3442 int netdev_tstamp_prequeue __read_mostly
= 1;
3443 int netdev_budget __read_mostly
= 300;
3444 int weight_p __read_mostly
= 64; /* old backlog weight */
3445 int dev_weight_rx_bias __read_mostly
= 1; /* bias for backlog weight */
3446 int dev_weight_tx_bias __read_mostly
= 1; /* bias for output_queue quota */
3447 int dev_rx_weight __read_mostly
= 64;
3448 int dev_tx_weight __read_mostly
= 64;
3450 /* Called with irq disabled */
3451 static inline void ____napi_schedule(struct softnet_data
*sd
,
3452 struct napi_struct
*napi
)
3454 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
3455 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3460 /* One global table that all flow-based protocols share. */
3461 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
3462 EXPORT_SYMBOL(rps_sock_flow_table
);
3463 u32 rps_cpu_mask __read_mostly
;
3464 EXPORT_SYMBOL(rps_cpu_mask
);
3466 struct static_key rps_needed __read_mostly
;
3467 EXPORT_SYMBOL(rps_needed
);
3468 struct static_key rfs_needed __read_mostly
;
3469 EXPORT_SYMBOL(rfs_needed
);
3471 static struct rps_dev_flow
*
3472 set_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
3473 struct rps_dev_flow
*rflow
, u16 next_cpu
)
3475 if (next_cpu
< nr_cpu_ids
) {
3476 #ifdef CONFIG_RFS_ACCEL
3477 struct netdev_rx_queue
*rxqueue
;
3478 struct rps_dev_flow_table
*flow_table
;
3479 struct rps_dev_flow
*old_rflow
;
3484 /* Should we steer this flow to a different hardware queue? */
3485 if (!skb_rx_queue_recorded(skb
) || !dev
->rx_cpu_rmap
||
3486 !(dev
->features
& NETIF_F_NTUPLE
))
3488 rxq_index
= cpu_rmap_lookup_index(dev
->rx_cpu_rmap
, next_cpu
);
3489 if (rxq_index
== skb_get_rx_queue(skb
))
3492 rxqueue
= dev
->_rx
+ rxq_index
;
3493 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3496 flow_id
= skb_get_hash(skb
) & flow_table
->mask
;
3497 rc
= dev
->netdev_ops
->ndo_rx_flow_steer(dev
, skb
,
3498 rxq_index
, flow_id
);
3502 rflow
= &flow_table
->flows
[flow_id
];
3504 if (old_rflow
->filter
== rflow
->filter
)
3505 old_rflow
->filter
= RPS_NO_FILTER
;
3509 per_cpu(softnet_data
, next_cpu
).input_queue_head
;
3512 rflow
->cpu
= next_cpu
;
3517 * get_rps_cpu is called from netif_receive_skb and returns the target
3518 * CPU from the RPS map of the receiving queue for a given skb.
3519 * rcu_read_lock must be held on entry.
3521 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
3522 struct rps_dev_flow
**rflowp
)
3524 const struct rps_sock_flow_table
*sock_flow_table
;
3525 struct netdev_rx_queue
*rxqueue
= dev
->_rx
;
3526 struct rps_dev_flow_table
*flow_table
;
3527 struct rps_map
*map
;
3532 if (skb_rx_queue_recorded(skb
)) {
3533 u16 index
= skb_get_rx_queue(skb
);
3535 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
3536 WARN_ONCE(dev
->real_num_rx_queues
> 1,
3537 "%s received packet on queue %u, but number "
3538 "of RX queues is %u\n",
3539 dev
->name
, index
, dev
->real_num_rx_queues
);
3545 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3547 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3548 map
= rcu_dereference(rxqueue
->rps_map
);
3549 if (!flow_table
&& !map
)
3552 skb_reset_network_header(skb
);
3553 hash
= skb_get_hash(skb
);
3557 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
3558 if (flow_table
&& sock_flow_table
) {
3559 struct rps_dev_flow
*rflow
;
3563 /* First check into global flow table if there is a match */
3564 ident
= sock_flow_table
->ents
[hash
& sock_flow_table
->mask
];
3565 if ((ident
^ hash
) & ~rps_cpu_mask
)
3568 next_cpu
= ident
& rps_cpu_mask
;
3570 /* OK, now we know there is a match,
3571 * we can look at the local (per receive queue) flow table
3573 rflow
= &flow_table
->flows
[hash
& flow_table
->mask
];
3577 * If the desired CPU (where last recvmsg was done) is
3578 * different from current CPU (one in the rx-queue flow
3579 * table entry), switch if one of the following holds:
3580 * - Current CPU is unset (>= nr_cpu_ids).
3581 * - Current CPU is offline.
3582 * - The current CPU's queue tail has advanced beyond the
3583 * last packet that was enqueued using this table entry.
3584 * This guarantees that all previous packets for the flow
3585 * have been dequeued, thus preserving in order delivery.
3587 if (unlikely(tcpu
!= next_cpu
) &&
3588 (tcpu
>= nr_cpu_ids
|| !cpu_online(tcpu
) ||
3589 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
3590 rflow
->last_qtail
)) >= 0)) {
3592 rflow
= set_rps_cpu(dev
, skb
, rflow
, next_cpu
);
3595 if (tcpu
< nr_cpu_ids
&& cpu_online(tcpu
)) {
3605 tcpu
= map
->cpus
[reciprocal_scale(hash
, map
->len
)];
3606 if (cpu_online(tcpu
)) {
3616 #ifdef CONFIG_RFS_ACCEL
3619 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3620 * @dev: Device on which the filter was set
3621 * @rxq_index: RX queue index
3622 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3623 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3625 * Drivers that implement ndo_rx_flow_steer() should periodically call
3626 * this function for each installed filter and remove the filters for
3627 * which it returns %true.
3629 bool rps_may_expire_flow(struct net_device
*dev
, u16 rxq_index
,
3630 u32 flow_id
, u16 filter_id
)
3632 struct netdev_rx_queue
*rxqueue
= dev
->_rx
+ rxq_index
;
3633 struct rps_dev_flow_table
*flow_table
;
3634 struct rps_dev_flow
*rflow
;
3639 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3640 if (flow_table
&& flow_id
<= flow_table
->mask
) {
3641 rflow
= &flow_table
->flows
[flow_id
];
3642 cpu
= ACCESS_ONCE(rflow
->cpu
);
3643 if (rflow
->filter
== filter_id
&& cpu
< nr_cpu_ids
&&
3644 ((int)(per_cpu(softnet_data
, cpu
).input_queue_head
-
3645 rflow
->last_qtail
) <
3646 (int)(10 * flow_table
->mask
)))
3652 EXPORT_SYMBOL(rps_may_expire_flow
);
3654 #endif /* CONFIG_RFS_ACCEL */
3656 /* Called from hardirq (IPI) context */
3657 static void rps_trigger_softirq(void *data
)
3659 struct softnet_data
*sd
= data
;
3661 ____napi_schedule(sd
, &sd
->backlog
);
3665 #endif /* CONFIG_RPS */
3668 * Check if this softnet_data structure is another cpu one
3669 * If yes, queue it to our IPI list and return 1
3672 static int rps_ipi_queued(struct softnet_data
*sd
)
3675 struct softnet_data
*mysd
= this_cpu_ptr(&softnet_data
);
3678 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
3679 mysd
->rps_ipi_list
= sd
;
3681 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3684 #endif /* CONFIG_RPS */
3688 #ifdef CONFIG_NET_FLOW_LIMIT
3689 int netdev_flow_limit_table_len __read_mostly
= (1 << 12);
3692 static bool skb_flow_limit(struct sk_buff
*skb
, unsigned int qlen
)
3694 #ifdef CONFIG_NET_FLOW_LIMIT
3695 struct sd_flow_limit
*fl
;
3696 struct softnet_data
*sd
;
3697 unsigned int old_flow
, new_flow
;
3699 if (qlen
< (netdev_max_backlog
>> 1))
3702 sd
= this_cpu_ptr(&softnet_data
);
3705 fl
= rcu_dereference(sd
->flow_limit
);
3707 new_flow
= skb_get_hash(skb
) & (fl
->num_buckets
- 1);
3708 old_flow
= fl
->history
[fl
->history_head
];
3709 fl
->history
[fl
->history_head
] = new_flow
;
3712 fl
->history_head
&= FLOW_LIMIT_HISTORY
- 1;
3714 if (likely(fl
->buckets
[old_flow
]))
3715 fl
->buckets
[old_flow
]--;
3717 if (++fl
->buckets
[new_flow
] > (FLOW_LIMIT_HISTORY
>> 1)) {
3729 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3730 * queue (may be a remote CPU queue).
3732 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
3733 unsigned int *qtail
)
3735 struct softnet_data
*sd
;
3736 unsigned long flags
;
3739 sd
= &per_cpu(softnet_data
, cpu
);
3741 local_irq_save(flags
);
3744 if (!netif_running(skb
->dev
))
3746 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3747 if (qlen
<= netdev_max_backlog
&& !skb_flow_limit(skb
, qlen
)) {
3750 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
3751 input_queue_tail_incr_save(sd
, qtail
);
3753 local_irq_restore(flags
);
3754 return NET_RX_SUCCESS
;
3757 /* Schedule NAPI for backlog device
3758 * We can use non atomic operation since we own the queue lock
3760 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
3761 if (!rps_ipi_queued(sd
))
3762 ____napi_schedule(sd
, &sd
->backlog
);
3771 local_irq_restore(flags
);
3773 atomic_long_inc(&skb
->dev
->rx_dropped
);
3778 static int netif_rx_internal(struct sk_buff
*skb
)
3782 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3784 trace_netif_rx(skb
);
3786 if (static_key_false(&rps_needed
)) {
3787 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3793 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3795 cpu
= smp_processor_id();
3797 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3806 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
3813 * netif_rx - post buffer to the network code
3814 * @skb: buffer to post
3816 * This function receives a packet from a device driver and queues it for
3817 * the upper (protocol) levels to process. It always succeeds. The buffer
3818 * may be dropped during processing for congestion control or by the
3822 * NET_RX_SUCCESS (no congestion)
3823 * NET_RX_DROP (packet was dropped)
3827 int netif_rx(struct sk_buff
*skb
)
3829 trace_netif_rx_entry(skb
);
3831 return netif_rx_internal(skb
);
3833 EXPORT_SYMBOL(netif_rx
);
3835 int netif_rx_ni(struct sk_buff
*skb
)
3839 trace_netif_rx_ni_entry(skb
);
3842 err
= netif_rx_internal(skb
);
3843 if (local_softirq_pending())
3849 EXPORT_SYMBOL(netif_rx_ni
);
3851 static __latent_entropy
void net_tx_action(struct softirq_action
*h
)
3853 struct softnet_data
*sd
= this_cpu_ptr(&softnet_data
);
3855 if (sd
->completion_queue
) {
3856 struct sk_buff
*clist
;
3858 local_irq_disable();
3859 clist
= sd
->completion_queue
;
3860 sd
->completion_queue
= NULL
;
3864 struct sk_buff
*skb
= clist
;
3866 clist
= clist
->next
;
3868 WARN_ON(atomic_read(&skb
->users
));
3869 if (likely(get_kfree_skb_cb(skb
)->reason
== SKB_REASON_CONSUMED
))
3870 trace_consume_skb(skb
);
3872 trace_kfree_skb(skb
, net_tx_action
);
3874 if (skb
->fclone
!= SKB_FCLONE_UNAVAILABLE
)
3877 __kfree_skb_defer(skb
);
3880 __kfree_skb_flush();
3883 if (sd
->output_queue
) {
3886 local_irq_disable();
3887 head
= sd
->output_queue
;
3888 sd
->output_queue
= NULL
;
3889 sd
->output_queue_tailp
= &sd
->output_queue
;
3893 struct Qdisc
*q
= head
;
3894 spinlock_t
*root_lock
;
3896 head
= head
->next_sched
;
3898 root_lock
= qdisc_lock(q
);
3899 spin_lock(root_lock
);
3900 /* We need to make sure head->next_sched is read
3901 * before clearing __QDISC_STATE_SCHED
3903 smp_mb__before_atomic();
3904 clear_bit(__QDISC_STATE_SCHED
, &q
->state
);
3906 spin_unlock(root_lock
);
3911 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
3912 /* This hook is defined here for ATM LANE */
3913 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
3914 unsigned char *addr
) __read_mostly
;
3915 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
3918 static inline struct sk_buff
*
3919 sch_handle_ingress(struct sk_buff
*skb
, struct packet_type
**pt_prev
, int *ret
,
3920 struct net_device
*orig_dev
)
3922 #ifdef CONFIG_NET_CLS_ACT
3923 struct tcf_proto
*cl
= rcu_dereference_bh(skb
->dev
->ingress_cl_list
);
3924 struct tcf_result cl_res
;
3926 /* If there's at least one ingress present somewhere (so
3927 * we get here via enabled static key), remaining devices
3928 * that are not configured with an ingress qdisc will bail
3934 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
3938 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
3939 skb
->tc_at_ingress
= 1;
3940 qdisc_bstats_cpu_update(cl
->q
, skb
);
3942 switch (tc_classify(skb
, cl
, &cl_res
, false)) {
3944 case TC_ACT_RECLASSIFY
:
3945 skb
->tc_index
= TC_H_MIN(cl_res
.classid
);
3948 qdisc_qstats_cpu_drop(cl
->q
);
3955 case TC_ACT_REDIRECT
:
3956 /* skb_mac_header check was done by cls/act_bpf, so
3957 * we can safely push the L2 header back before
3958 * redirecting to another netdev
3960 __skb_push(skb
, skb
->mac_len
);
3961 skb_do_redirect(skb
);
3966 #endif /* CONFIG_NET_CLS_ACT */
3971 * netdev_is_rx_handler_busy - check if receive handler is registered
3972 * @dev: device to check
3974 * Check if a receive handler is already registered for a given device.
3975 * Return true if there one.
3977 * The caller must hold the rtnl_mutex.
3979 bool netdev_is_rx_handler_busy(struct net_device
*dev
)
3982 return dev
&& rtnl_dereference(dev
->rx_handler
);
3984 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy
);
3987 * netdev_rx_handler_register - register receive handler
3988 * @dev: device to register a handler for
3989 * @rx_handler: receive handler to register
3990 * @rx_handler_data: data pointer that is used by rx handler
3992 * Register a receive handler for a device. This handler will then be
3993 * called from __netif_receive_skb. A negative errno code is returned
3996 * The caller must hold the rtnl_mutex.
3998 * For a general description of rx_handler, see enum rx_handler_result.
4000 int netdev_rx_handler_register(struct net_device
*dev
,
4001 rx_handler_func_t
*rx_handler
,
4002 void *rx_handler_data
)
4004 if (netdev_is_rx_handler_busy(dev
))
4007 /* Note: rx_handler_data must be set before rx_handler */
4008 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
4009 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
4013 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
4016 * netdev_rx_handler_unregister - unregister receive handler
4017 * @dev: device to unregister a handler from
4019 * Unregister a receive handler from a device.
4021 * The caller must hold the rtnl_mutex.
4023 void netdev_rx_handler_unregister(struct net_device
*dev
)
4027 RCU_INIT_POINTER(dev
->rx_handler
, NULL
);
4028 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
4029 * section has a guarantee to see a non NULL rx_handler_data
4033 RCU_INIT_POINTER(dev
->rx_handler_data
, NULL
);
4035 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
4038 * Limit the use of PFMEMALLOC reserves to those protocols that implement
4039 * the special handling of PFMEMALLOC skbs.
4041 static bool skb_pfmemalloc_protocol(struct sk_buff
*skb
)
4043 switch (skb
->protocol
) {
4044 case htons(ETH_P_ARP
):
4045 case htons(ETH_P_IP
):
4046 case htons(ETH_P_IPV6
):
4047 case htons(ETH_P_8021Q
):
4048 case htons(ETH_P_8021AD
):
4055 static inline int nf_ingress(struct sk_buff
*skb
, struct packet_type
**pt_prev
,
4056 int *ret
, struct net_device
*orig_dev
)
4058 #ifdef CONFIG_NETFILTER_INGRESS
4059 if (nf_hook_ingress_active(skb
)) {
4063 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
4068 ingress_retval
= nf_hook_ingress(skb
);
4070 return ingress_retval
;
4072 #endif /* CONFIG_NETFILTER_INGRESS */
4076 static int __netif_receive_skb_core(struct sk_buff
*skb
, bool pfmemalloc
)
4078 struct packet_type
*ptype
, *pt_prev
;
4079 rx_handler_func_t
*rx_handler
;
4080 struct net_device
*orig_dev
;
4081 bool deliver_exact
= false;
4082 int ret
= NET_RX_DROP
;
4085 net_timestamp_check(!netdev_tstamp_prequeue
, skb
);
4087 trace_netif_receive_skb(skb
);
4089 orig_dev
= skb
->dev
;
4091 skb_reset_network_header(skb
);
4092 if (!skb_transport_header_was_set(skb
))
4093 skb_reset_transport_header(skb
);
4094 skb_reset_mac_len(skb
);
4099 skb
->skb_iif
= skb
->dev
->ifindex
;
4101 __this_cpu_inc(softnet_data
.processed
);
4103 if (skb
->protocol
== cpu_to_be16(ETH_P_8021Q
) ||
4104 skb
->protocol
== cpu_to_be16(ETH_P_8021AD
)) {
4105 skb
= skb_vlan_untag(skb
);
4110 if (skb_skip_tc_classify(skb
))
4116 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
4118 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
4122 list_for_each_entry_rcu(ptype
, &skb
->dev
->ptype_all
, list
) {
4124 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
4129 #ifdef CONFIG_NET_INGRESS
4130 if (static_key_false(&ingress_needed
)) {
4131 skb
= sch_handle_ingress(skb
, &pt_prev
, &ret
, orig_dev
);
4135 if (nf_ingress(skb
, &pt_prev
, &ret
, orig_dev
) < 0)
4141 if (pfmemalloc
&& !skb_pfmemalloc_protocol(skb
))
4144 if (skb_vlan_tag_present(skb
)) {
4146 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
4149 if (vlan_do_receive(&skb
))
4151 else if (unlikely(!skb
))
4155 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
4158 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
4161 switch (rx_handler(&skb
)) {
4162 case RX_HANDLER_CONSUMED
:
4163 ret
= NET_RX_SUCCESS
;
4165 case RX_HANDLER_ANOTHER
:
4167 case RX_HANDLER_EXACT
:
4168 deliver_exact
= true;
4169 case RX_HANDLER_PASS
:
4176 if (unlikely(skb_vlan_tag_present(skb
))) {
4177 if (skb_vlan_tag_get_id(skb
))
4178 skb
->pkt_type
= PACKET_OTHERHOST
;
4179 /* Note: we might in the future use prio bits
4180 * and set skb->priority like in vlan_do_receive()
4181 * For the time being, just ignore Priority Code Point
4186 type
= skb
->protocol
;
4188 /* deliver only exact match when indicated */
4189 if (likely(!deliver_exact
)) {
4190 deliver_ptype_list_skb(skb
, &pt_prev
, orig_dev
, type
,
4191 &ptype_base
[ntohs(type
) &
4195 deliver_ptype_list_skb(skb
, &pt_prev
, orig_dev
, type
,
4196 &orig_dev
->ptype_specific
);
4198 if (unlikely(skb
->dev
!= orig_dev
)) {
4199 deliver_ptype_list_skb(skb
, &pt_prev
, orig_dev
, type
,
4200 &skb
->dev
->ptype_specific
);
4204 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
4207 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
4211 atomic_long_inc(&skb
->dev
->rx_dropped
);
4213 atomic_long_inc(&skb
->dev
->rx_nohandler
);
4215 /* Jamal, now you will not able to escape explaining
4216 * me how you were going to use this. :-)
4225 static int __netif_receive_skb(struct sk_buff
*skb
)
4229 if (sk_memalloc_socks() && skb_pfmemalloc(skb
)) {
4230 unsigned long pflags
= current
->flags
;
4233 * PFMEMALLOC skbs are special, they should
4234 * - be delivered to SOCK_MEMALLOC sockets only
4235 * - stay away from userspace
4236 * - have bounded memory usage
4238 * Use PF_MEMALLOC as this saves us from propagating the allocation
4239 * context down to all allocation sites.
4241 current
->flags
|= PF_MEMALLOC
;
4242 ret
= __netif_receive_skb_core(skb
, true);
4243 tsk_restore_flags(current
, pflags
, PF_MEMALLOC
);
4245 ret
= __netif_receive_skb_core(skb
, false);
4250 static int netif_receive_skb_internal(struct sk_buff
*skb
)
4254 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
4256 if (skb_defer_rx_timestamp(skb
))
4257 return NET_RX_SUCCESS
;
4262 if (static_key_false(&rps_needed
)) {
4263 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
4264 int cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
4267 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
4273 ret
= __netif_receive_skb(skb
);
4279 * netif_receive_skb - process receive buffer from network
4280 * @skb: buffer to process
4282 * netif_receive_skb() is the main receive data processing function.
4283 * It always succeeds. The buffer may be dropped during processing
4284 * for congestion control or by the protocol layers.
4286 * This function may only be called from softirq context and interrupts
4287 * should be enabled.
4289 * Return values (usually ignored):
4290 * NET_RX_SUCCESS: no congestion
4291 * NET_RX_DROP: packet was dropped
4293 int netif_receive_skb(struct sk_buff
*skb
)
4295 trace_netif_receive_skb_entry(skb
);
4297 return netif_receive_skb_internal(skb
);
4299 EXPORT_SYMBOL(netif_receive_skb
);
4301 DEFINE_PER_CPU(struct work_struct
, flush_works
);
4303 /* Network device is going away, flush any packets still pending */
4304 static void flush_backlog(struct work_struct
*work
)
4306 struct sk_buff
*skb
, *tmp
;
4307 struct softnet_data
*sd
;
4310 sd
= this_cpu_ptr(&softnet_data
);
4312 local_irq_disable();
4314 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
4315 if (skb
->dev
->reg_state
== NETREG_UNREGISTERING
) {
4316 __skb_unlink(skb
, &sd
->input_pkt_queue
);
4318 input_queue_head_incr(sd
);
4324 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
4325 if (skb
->dev
->reg_state
== NETREG_UNREGISTERING
) {
4326 __skb_unlink(skb
, &sd
->process_queue
);
4328 input_queue_head_incr(sd
);
4334 static void flush_all_backlogs(void)
4340 for_each_online_cpu(cpu
)
4341 queue_work_on(cpu
, system_highpri_wq
,
4342 per_cpu_ptr(&flush_works
, cpu
));
4344 for_each_online_cpu(cpu
)
4345 flush_work(per_cpu_ptr(&flush_works
, cpu
));
4350 static int napi_gro_complete(struct sk_buff
*skb
)
4352 struct packet_offload
*ptype
;
4353 __be16 type
= skb
->protocol
;
4354 struct list_head
*head
= &offload_base
;
4357 BUILD_BUG_ON(sizeof(struct napi_gro_cb
) > sizeof(skb
->cb
));
4359 if (NAPI_GRO_CB(skb
)->count
== 1) {
4360 skb_shinfo(skb
)->gso_size
= 0;
4365 list_for_each_entry_rcu(ptype
, head
, list
) {
4366 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_complete
)
4369 err
= ptype
->callbacks
.gro_complete(skb
, 0);
4375 WARN_ON(&ptype
->list
== head
);
4377 return NET_RX_SUCCESS
;
4381 return netif_receive_skb_internal(skb
);
4384 /* napi->gro_list contains packets ordered by age.
4385 * youngest packets at the head of it.
4386 * Complete skbs in reverse order to reduce latencies.
4388 void napi_gro_flush(struct napi_struct
*napi
, bool flush_old
)
4390 struct sk_buff
*skb
, *prev
= NULL
;
4392 /* scan list and build reverse chain */
4393 for (skb
= napi
->gro_list
; skb
!= NULL
; skb
= skb
->next
) {
4398 for (skb
= prev
; skb
; skb
= prev
) {
4401 if (flush_old
&& NAPI_GRO_CB(skb
)->age
== jiffies
)
4405 napi_gro_complete(skb
);
4409 napi
->gro_list
= NULL
;
4411 EXPORT_SYMBOL(napi_gro_flush
);
4413 static void gro_list_prepare(struct napi_struct
*napi
, struct sk_buff
*skb
)
4416 unsigned int maclen
= skb
->dev
->hard_header_len
;
4417 u32 hash
= skb_get_hash_raw(skb
);
4419 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
4420 unsigned long diffs
;
4422 NAPI_GRO_CB(p
)->flush
= 0;
4424 if (hash
!= skb_get_hash_raw(p
)) {
4425 NAPI_GRO_CB(p
)->same_flow
= 0;
4429 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
4430 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
4431 diffs
|= skb_metadata_dst_cmp(p
, skb
);
4432 if (maclen
== ETH_HLEN
)
4433 diffs
|= compare_ether_header(skb_mac_header(p
),
4434 skb_mac_header(skb
));
4436 diffs
= memcmp(skb_mac_header(p
),
4437 skb_mac_header(skb
),
4439 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
4443 static void skb_gro_reset_offset(struct sk_buff
*skb
)
4445 const struct skb_shared_info
*pinfo
= skb_shinfo(skb
);
4446 const skb_frag_t
*frag0
= &pinfo
->frags
[0];
4448 NAPI_GRO_CB(skb
)->data_offset
= 0;
4449 NAPI_GRO_CB(skb
)->frag0
= NULL
;
4450 NAPI_GRO_CB(skb
)->frag0_len
= 0;
4452 if (skb_mac_header(skb
) == skb_tail_pointer(skb
) &&
4454 !PageHighMem(skb_frag_page(frag0
))) {
4455 NAPI_GRO_CB(skb
)->frag0
= skb_frag_address(frag0
);
4456 NAPI_GRO_CB(skb
)->frag0_len
= min_t(unsigned int,
4457 skb_frag_size(frag0
),
4458 skb
->end
- skb
->tail
);
4462 static void gro_pull_from_frag0(struct sk_buff
*skb
, int grow
)
4464 struct skb_shared_info
*pinfo
= skb_shinfo(skb
);
4466 BUG_ON(skb
->end
- skb
->tail
< grow
);
4468 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
4470 skb
->data_len
-= grow
;
4473 pinfo
->frags
[0].page_offset
+= grow
;
4474 skb_frag_size_sub(&pinfo
->frags
[0], grow
);
4476 if (unlikely(!skb_frag_size(&pinfo
->frags
[0]))) {
4477 skb_frag_unref(skb
, 0);
4478 memmove(pinfo
->frags
, pinfo
->frags
+ 1,
4479 --pinfo
->nr_frags
* sizeof(pinfo
->frags
[0]));
4483 static enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
4485 struct sk_buff
**pp
= NULL
;
4486 struct packet_offload
*ptype
;
4487 __be16 type
= skb
->protocol
;
4488 struct list_head
*head
= &offload_base
;
4490 enum gro_result ret
;
4493 if (!(skb
->dev
->features
& NETIF_F_GRO
))
4499 gro_list_prepare(napi
, skb
);
4502 list_for_each_entry_rcu(ptype
, head
, list
) {
4503 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_receive
)
4506 skb_set_network_header(skb
, skb_gro_offset(skb
));
4507 skb_reset_mac_len(skb
);
4508 NAPI_GRO_CB(skb
)->same_flow
= 0;
4509 NAPI_GRO_CB(skb
)->flush
= skb_is_gso(skb
) || skb_has_frag_list(skb
);
4510 NAPI_GRO_CB(skb
)->free
= 0;
4511 NAPI_GRO_CB(skb
)->encap_mark
= 0;
4512 NAPI_GRO_CB(skb
)->recursion_counter
= 0;
4513 NAPI_GRO_CB(skb
)->is_fou
= 0;
4514 NAPI_GRO_CB(skb
)->is_atomic
= 1;
4515 NAPI_GRO_CB(skb
)->gro_remcsum_start
= 0;
4517 /* Setup for GRO checksum validation */
4518 switch (skb
->ip_summed
) {
4519 case CHECKSUM_COMPLETE
:
4520 NAPI_GRO_CB(skb
)->csum
= skb
->csum
;
4521 NAPI_GRO_CB(skb
)->csum_valid
= 1;
4522 NAPI_GRO_CB(skb
)->csum_cnt
= 0;
4524 case CHECKSUM_UNNECESSARY
:
4525 NAPI_GRO_CB(skb
)->csum_cnt
= skb
->csum_level
+ 1;
4526 NAPI_GRO_CB(skb
)->csum_valid
= 0;
4529 NAPI_GRO_CB(skb
)->csum_cnt
= 0;
4530 NAPI_GRO_CB(skb
)->csum_valid
= 0;
4533 pp
= ptype
->callbacks
.gro_receive(&napi
->gro_list
, skb
);
4538 if (&ptype
->list
== head
)
4541 if (IS_ERR(pp
) && PTR_ERR(pp
) == -EINPROGRESS
) {
4546 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
4547 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
4550 struct sk_buff
*nskb
= *pp
;
4554 napi_gro_complete(nskb
);
4561 if (NAPI_GRO_CB(skb
)->flush
)
4564 if (unlikely(napi
->gro_count
>= MAX_GRO_SKBS
)) {
4565 struct sk_buff
*nskb
= napi
->gro_list
;
4567 /* locate the end of the list to select the 'oldest' flow */
4568 while (nskb
->next
) {
4574 napi_gro_complete(nskb
);
4578 NAPI_GRO_CB(skb
)->count
= 1;
4579 NAPI_GRO_CB(skb
)->age
= jiffies
;
4580 NAPI_GRO_CB(skb
)->last
= skb
;
4581 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
4582 skb
->next
= napi
->gro_list
;
4583 napi
->gro_list
= skb
;
4587 grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
4589 gro_pull_from_frag0(skb
, grow
);
4598 struct packet_offload
*gro_find_receive_by_type(__be16 type
)
4600 struct list_head
*offload_head
= &offload_base
;
4601 struct packet_offload
*ptype
;
4603 list_for_each_entry_rcu(ptype
, offload_head
, list
) {
4604 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_receive
)
4610 EXPORT_SYMBOL(gro_find_receive_by_type
);
4612 struct packet_offload
*gro_find_complete_by_type(__be16 type
)
4614 struct list_head
*offload_head
= &offload_base
;
4615 struct packet_offload
*ptype
;
4617 list_for_each_entry_rcu(ptype
, offload_head
, list
) {
4618 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_complete
)
4624 EXPORT_SYMBOL(gro_find_complete_by_type
);
4626 static gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
4630 if (netif_receive_skb_internal(skb
))
4638 case GRO_MERGED_FREE
:
4639 if (NAPI_GRO_CB(skb
)->free
== NAPI_GRO_FREE_STOLEN_HEAD
) {
4642 kmem_cache_free(skbuff_head_cache
, skb
);
4657 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
4659 skb_mark_napi_id(skb
, napi
);
4660 trace_napi_gro_receive_entry(skb
);
4662 skb_gro_reset_offset(skb
);
4664 return napi_skb_finish(dev_gro_receive(napi
, skb
), skb
);
4666 EXPORT_SYMBOL(napi_gro_receive
);
4668 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
4670 if (unlikely(skb
->pfmemalloc
)) {
4674 __skb_pull(skb
, skb_headlen(skb
));
4675 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4676 skb_reserve(skb
, NET_SKB_PAD
+ NET_IP_ALIGN
- skb_headroom(skb
));
4678 skb
->dev
= napi
->dev
;
4680 skb
->encapsulation
= 0;
4681 skb_shinfo(skb
)->gso_type
= 0;
4682 skb
->truesize
= SKB_TRUESIZE(skb_end_offset(skb
));
4688 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
4690 struct sk_buff
*skb
= napi
->skb
;
4693 skb
= napi_alloc_skb(napi
, GRO_MAX_HEAD
);
4696 skb_mark_napi_id(skb
, napi
);
4701 EXPORT_SYMBOL(napi_get_frags
);
4703 static gro_result_t
napi_frags_finish(struct napi_struct
*napi
,
4704 struct sk_buff
*skb
,
4710 __skb_push(skb
, ETH_HLEN
);
4711 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
4712 if (ret
== GRO_NORMAL
&& netif_receive_skb_internal(skb
))
4717 case GRO_MERGED_FREE
:
4718 napi_reuse_skb(napi
, skb
);
4729 /* Upper GRO stack assumes network header starts at gro_offset=0
4730 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4731 * We copy ethernet header into skb->data to have a common layout.
4733 static struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
4735 struct sk_buff
*skb
= napi
->skb
;
4736 const struct ethhdr
*eth
;
4737 unsigned int hlen
= sizeof(*eth
);
4741 skb_reset_mac_header(skb
);
4742 skb_gro_reset_offset(skb
);
4744 eth
= skb_gro_header_fast(skb
, 0);
4745 if (unlikely(skb_gro_header_hard(skb
, hlen
))) {
4746 eth
= skb_gro_header_slow(skb
, hlen
, 0);
4747 if (unlikely(!eth
)) {
4748 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
4749 __func__
, napi
->dev
->name
);
4750 napi_reuse_skb(napi
, skb
);
4754 gro_pull_from_frag0(skb
, hlen
);
4755 NAPI_GRO_CB(skb
)->frag0
+= hlen
;
4756 NAPI_GRO_CB(skb
)->frag0_len
-= hlen
;
4758 __skb_pull(skb
, hlen
);
4761 * This works because the only protocols we care about don't require
4763 * We'll fix it up properly in napi_frags_finish()
4765 skb
->protocol
= eth
->h_proto
;
4770 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
4772 struct sk_buff
*skb
= napi_frags_skb(napi
);
4777 trace_napi_gro_frags_entry(skb
);
4779 return napi_frags_finish(napi
, skb
, dev_gro_receive(napi
, skb
));
4781 EXPORT_SYMBOL(napi_gro_frags
);
4783 /* Compute the checksum from gro_offset and return the folded value
4784 * after adding in any pseudo checksum.
4786 __sum16
__skb_gro_checksum_complete(struct sk_buff
*skb
)
4791 wsum
= skb_checksum(skb
, skb_gro_offset(skb
), skb_gro_len(skb
), 0);
4793 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4794 sum
= csum_fold(csum_add(NAPI_GRO_CB(skb
)->csum
, wsum
));
4796 if (unlikely(skb
->ip_summed
== CHECKSUM_COMPLETE
) &&
4797 !skb
->csum_complete_sw
)
4798 netdev_rx_csum_fault(skb
->dev
);
4801 NAPI_GRO_CB(skb
)->csum
= wsum
;
4802 NAPI_GRO_CB(skb
)->csum_valid
= 1;
4806 EXPORT_SYMBOL(__skb_gro_checksum_complete
);
4809 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4810 * Note: called with local irq disabled, but exits with local irq enabled.
4812 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
4815 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
4818 sd
->rps_ipi_list
= NULL
;
4822 /* Send pending IPI's to kick RPS processing on remote cpus. */
4824 struct softnet_data
*next
= remsd
->rps_ipi_next
;
4826 if (cpu_online(remsd
->cpu
))
4827 smp_call_function_single_async(remsd
->cpu
,
4836 static bool sd_has_rps_ipi_waiting(struct softnet_data
*sd
)
4839 return sd
->rps_ipi_list
!= NULL
;
4845 static int process_backlog(struct napi_struct
*napi
, int quota
)
4847 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
4851 /* Check if we have pending ipi, its better to send them now,
4852 * not waiting net_rx_action() end.
4854 if (sd_has_rps_ipi_waiting(sd
)) {
4855 local_irq_disable();
4856 net_rps_action_and_irq_enable(sd
);
4859 napi
->weight
= dev_rx_weight
;
4861 struct sk_buff
*skb
;
4863 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
4865 __netif_receive_skb(skb
);
4867 input_queue_head_incr(sd
);
4868 if (++work
>= quota
)
4873 local_irq_disable();
4875 if (skb_queue_empty(&sd
->input_pkt_queue
)) {
4877 * Inline a custom version of __napi_complete().
4878 * only current cpu owns and manipulates this napi,
4879 * and NAPI_STATE_SCHED is the only possible flag set
4881 * We can use a plain write instead of clear_bit(),
4882 * and we dont need an smp_mb() memory barrier.
4887 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
4888 &sd
->process_queue
);
4898 * __napi_schedule - schedule for receive
4899 * @n: entry to schedule
4901 * The entry's receive function will be scheduled to run.
4902 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
4904 void __napi_schedule(struct napi_struct
*n
)
4906 unsigned long flags
;
4908 local_irq_save(flags
);
4909 ____napi_schedule(this_cpu_ptr(&softnet_data
), n
);
4910 local_irq_restore(flags
);
4912 EXPORT_SYMBOL(__napi_schedule
);
4915 * napi_schedule_prep - check if napi can be scheduled
4918 * Test if NAPI routine is already running, and if not mark
4919 * it as running. This is used as a condition variable
4920 * insure only one NAPI poll instance runs. We also make
4921 * sure there is no pending NAPI disable.
4923 bool napi_schedule_prep(struct napi_struct
*n
)
4925 unsigned long val
, new;
4928 val
= READ_ONCE(n
->state
);
4929 if (unlikely(val
& NAPIF_STATE_DISABLE
))
4931 new = val
| NAPIF_STATE_SCHED
;
4933 /* Sets STATE_MISSED bit if STATE_SCHED was already set
4934 * This was suggested by Alexander Duyck, as compiler
4935 * emits better code than :
4936 * if (val & NAPIF_STATE_SCHED)
4937 * new |= NAPIF_STATE_MISSED;
4939 new |= (val
& NAPIF_STATE_SCHED
) / NAPIF_STATE_SCHED
*
4941 } while (cmpxchg(&n
->state
, val
, new) != val
);
4943 return !(val
& NAPIF_STATE_SCHED
);
4945 EXPORT_SYMBOL(napi_schedule_prep
);
4948 * __napi_schedule_irqoff - schedule for receive
4949 * @n: entry to schedule
4951 * Variant of __napi_schedule() assuming hard irqs are masked
4953 void __napi_schedule_irqoff(struct napi_struct
*n
)
4955 ____napi_schedule(this_cpu_ptr(&softnet_data
), n
);
4957 EXPORT_SYMBOL(__napi_schedule_irqoff
);
4959 bool napi_complete_done(struct napi_struct
*n
, int work_done
)
4961 unsigned long flags
, val
, new;
4964 * 1) Don't let napi dequeue from the cpu poll list
4965 * just in case its running on a different cpu.
4966 * 2) If we are busy polling, do nothing here, we have
4967 * the guarantee we will be called later.
4969 if (unlikely(n
->state
& (NAPIF_STATE_NPSVC
|
4970 NAPIF_STATE_IN_BUSY_POLL
)))
4974 unsigned long timeout
= 0;
4977 timeout
= n
->dev
->gro_flush_timeout
;
4980 hrtimer_start(&n
->timer
, ns_to_ktime(timeout
),
4981 HRTIMER_MODE_REL_PINNED
);
4983 napi_gro_flush(n
, false);
4985 if (unlikely(!list_empty(&n
->poll_list
))) {
4986 /* If n->poll_list is not empty, we need to mask irqs */
4987 local_irq_save(flags
);
4988 list_del_init(&n
->poll_list
);
4989 local_irq_restore(flags
);
4993 val
= READ_ONCE(n
->state
);
4995 WARN_ON_ONCE(!(val
& NAPIF_STATE_SCHED
));
4997 new = val
& ~(NAPIF_STATE_MISSED
| NAPIF_STATE_SCHED
);
4999 /* If STATE_MISSED was set, leave STATE_SCHED set,
5000 * because we will call napi->poll() one more time.
5001 * This C code was suggested by Alexander Duyck to help gcc.
5003 new |= (val
& NAPIF_STATE_MISSED
) / NAPIF_STATE_MISSED
*
5005 } while (cmpxchg(&n
->state
, val
, new) != val
);
5007 if (unlikely(val
& NAPIF_STATE_MISSED
)) {
5014 EXPORT_SYMBOL(napi_complete_done
);
5016 /* must be called under rcu_read_lock(), as we dont take a reference */
5017 static struct napi_struct
*napi_by_id(unsigned int napi_id
)
5019 unsigned int hash
= napi_id
% HASH_SIZE(napi_hash
);
5020 struct napi_struct
*napi
;
5022 hlist_for_each_entry_rcu(napi
, &napi_hash
[hash
], napi_hash_node
)
5023 if (napi
->napi_id
== napi_id
)
5029 #if defined(CONFIG_NET_RX_BUSY_POLL)
5031 #define BUSY_POLL_BUDGET 8
5033 static void busy_poll_stop(struct napi_struct
*napi
, void *have_poll_lock
)
5037 /* Busy polling means there is a high chance device driver hard irq
5038 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
5039 * set in napi_schedule_prep().
5040 * Since we are about to call napi->poll() once more, we can safely
5041 * clear NAPI_STATE_MISSED.
5043 * Note: x86 could use a single "lock and ..." instruction
5044 * to perform these two clear_bit()
5046 clear_bit(NAPI_STATE_MISSED
, &napi
->state
);
5047 clear_bit(NAPI_STATE_IN_BUSY_POLL
, &napi
->state
);
5051 /* All we really want here is to re-enable device interrupts.
5052 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
5054 rc
= napi
->poll(napi
, BUSY_POLL_BUDGET
);
5055 netpoll_poll_unlock(have_poll_lock
);
5056 if (rc
== BUSY_POLL_BUDGET
)
5057 __napi_schedule(napi
);
5059 if (local_softirq_pending())
5063 void sk_busy_loop(struct sock
*sk
, int nonblock
)
5065 unsigned long start_time
= nonblock
? 0 : busy_loop_current_time();
5066 int (*napi_poll
)(struct napi_struct
*napi
, int budget
);
5067 void *have_poll_lock
= NULL
;
5068 struct napi_struct
*napi
;
5069 unsigned int napi_id
;
5072 napi_id
= READ_ONCE(sk
->sk_napi_id
);
5073 if (napi_id
< MIN_NAPI_ID
)
5080 napi
= napi_by_id(napi_id
);
5090 unsigned long val
= READ_ONCE(napi
->state
);
5092 /* If multiple threads are competing for this napi,
5093 * we avoid dirtying napi->state as much as we can.
5095 if (val
& (NAPIF_STATE_DISABLE
| NAPIF_STATE_SCHED
|
5096 NAPIF_STATE_IN_BUSY_POLL
))
5098 if (cmpxchg(&napi
->state
, val
,
5099 val
| NAPIF_STATE_IN_BUSY_POLL
|
5100 NAPIF_STATE_SCHED
) != val
)
5102 have_poll_lock
= netpoll_poll_lock(napi
);
5103 napi_poll
= napi
->poll
;
5105 work
= napi_poll(napi
, BUSY_POLL_BUDGET
);
5106 trace_napi_poll(napi
, work
, BUSY_POLL_BUDGET
);
5109 __NET_ADD_STATS(sock_net(sk
),
5110 LINUX_MIB_BUSYPOLLRXPACKETS
, work
);
5113 if (nonblock
|| !skb_queue_empty(&sk
->sk_receive_queue
) ||
5114 sk_busy_loop_timeout(sk
, start_time
))
5117 if (unlikely(need_resched())) {
5119 busy_poll_stop(napi
, have_poll_lock
);
5123 if (!skb_queue_empty(&sk
->sk_receive_queue
) ||
5124 sk_busy_loop_timeout(sk
, start_time
))
5131 busy_poll_stop(napi
, have_poll_lock
);
5136 EXPORT_SYMBOL(sk_busy_loop
);
5138 #endif /* CONFIG_NET_RX_BUSY_POLL */
5140 static void napi_hash_add(struct napi_struct
*napi
)
5142 if (test_bit(NAPI_STATE_NO_BUSY_POLL
, &napi
->state
) ||
5143 test_and_set_bit(NAPI_STATE_HASHED
, &napi
->state
))
5146 spin_lock(&napi_hash_lock
);
5148 /* 0..NR_CPUS range is reserved for sender_cpu use */
5150 if (unlikely(++napi_gen_id
< MIN_NAPI_ID
))
5151 napi_gen_id
= MIN_NAPI_ID
;
5152 } while (napi_by_id(napi_gen_id
));
5153 napi
->napi_id
= napi_gen_id
;
5155 hlist_add_head_rcu(&napi
->napi_hash_node
,
5156 &napi_hash
[napi
->napi_id
% HASH_SIZE(napi_hash
)]);
5158 spin_unlock(&napi_hash_lock
);
5161 /* Warning : caller is responsible to make sure rcu grace period
5162 * is respected before freeing memory containing @napi
5164 bool napi_hash_del(struct napi_struct
*napi
)
5166 bool rcu_sync_needed
= false;
5168 spin_lock(&napi_hash_lock
);
5170 if (test_and_clear_bit(NAPI_STATE_HASHED
, &napi
->state
)) {
5171 rcu_sync_needed
= true;
5172 hlist_del_rcu(&napi
->napi_hash_node
);
5174 spin_unlock(&napi_hash_lock
);
5175 return rcu_sync_needed
;
5177 EXPORT_SYMBOL_GPL(napi_hash_del
);
5179 static enum hrtimer_restart
napi_watchdog(struct hrtimer
*timer
)
5181 struct napi_struct
*napi
;
5183 napi
= container_of(timer
, struct napi_struct
, timer
);
5185 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
5186 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
5188 if (napi
->gro_list
&& !napi_disable_pending(napi
) &&
5189 !test_and_set_bit(NAPI_STATE_SCHED
, &napi
->state
))
5190 __napi_schedule_irqoff(napi
);
5192 return HRTIMER_NORESTART
;
5195 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
5196 int (*poll
)(struct napi_struct
*, int), int weight
)
5198 INIT_LIST_HEAD(&napi
->poll_list
);
5199 hrtimer_init(&napi
->timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_REL_PINNED
);
5200 napi
->timer
.function
= napi_watchdog
;
5201 napi
->gro_count
= 0;
5202 napi
->gro_list
= NULL
;
5205 if (weight
> NAPI_POLL_WEIGHT
)
5206 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
5208 napi
->weight
= weight
;
5209 list_add(&napi
->dev_list
, &dev
->napi_list
);
5211 #ifdef CONFIG_NETPOLL
5212 napi
->poll_owner
= -1;
5214 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
5215 napi_hash_add(napi
);
5217 EXPORT_SYMBOL(netif_napi_add
);
5219 void napi_disable(struct napi_struct
*n
)
5222 set_bit(NAPI_STATE_DISABLE
, &n
->state
);
5224 while (test_and_set_bit(NAPI_STATE_SCHED
, &n
->state
))
5226 while (test_and_set_bit(NAPI_STATE_NPSVC
, &n
->state
))
5229 hrtimer_cancel(&n
->timer
);
5231 clear_bit(NAPI_STATE_DISABLE
, &n
->state
);
5233 EXPORT_SYMBOL(napi_disable
);
5235 /* Must be called in process context */
5236 void netif_napi_del(struct napi_struct
*napi
)
5239 if (napi_hash_del(napi
))
5241 list_del_init(&napi
->dev_list
);
5242 napi_free_frags(napi
);
5244 kfree_skb_list(napi
->gro_list
);
5245 napi
->gro_list
= NULL
;
5246 napi
->gro_count
= 0;
5248 EXPORT_SYMBOL(netif_napi_del
);
5250 static int napi_poll(struct napi_struct
*n
, struct list_head
*repoll
)
5255 list_del_init(&n
->poll_list
);
5257 have
= netpoll_poll_lock(n
);
5261 /* This NAPI_STATE_SCHED test is for avoiding a race
5262 * with netpoll's poll_napi(). Only the entity which
5263 * obtains the lock and sees NAPI_STATE_SCHED set will
5264 * actually make the ->poll() call. Therefore we avoid
5265 * accidentally calling ->poll() when NAPI is not scheduled.
5268 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
5269 work
= n
->poll(n
, weight
);
5270 trace_napi_poll(n
, work
, weight
);
5273 WARN_ON_ONCE(work
> weight
);
5275 if (likely(work
< weight
))
5278 /* Drivers must not modify the NAPI state if they
5279 * consume the entire weight. In such cases this code
5280 * still "owns" the NAPI instance and therefore can
5281 * move the instance around on the list at-will.
5283 if (unlikely(napi_disable_pending(n
))) {
5289 /* flush too old packets
5290 * If HZ < 1000, flush all packets.
5292 napi_gro_flush(n
, HZ
>= 1000);
5295 /* Some drivers may have called napi_schedule
5296 * prior to exhausting their budget.
5298 if (unlikely(!list_empty(&n
->poll_list
))) {
5299 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
5300 n
->dev
? n
->dev
->name
: "backlog");
5304 list_add_tail(&n
->poll_list
, repoll
);
5307 netpoll_poll_unlock(have
);
5312 static __latent_entropy
void net_rx_action(struct softirq_action
*h
)
5314 struct softnet_data
*sd
= this_cpu_ptr(&softnet_data
);
5315 unsigned long time_limit
= jiffies
+ 2;
5316 int budget
= netdev_budget
;
5320 local_irq_disable();
5321 list_splice_init(&sd
->poll_list
, &list
);
5325 struct napi_struct
*n
;
5327 if (list_empty(&list
)) {
5328 if (!sd_has_rps_ipi_waiting(sd
) && list_empty(&repoll
))
5333 n
= list_first_entry(&list
, struct napi_struct
, poll_list
);
5334 budget
-= napi_poll(n
, &repoll
);
5336 /* If softirq window is exhausted then punt.
5337 * Allow this to run for 2 jiffies since which will allow
5338 * an average latency of 1.5/HZ.
5340 if (unlikely(budget
<= 0 ||
5341 time_after_eq(jiffies
, time_limit
))) {
5347 local_irq_disable();
5349 list_splice_tail_init(&sd
->poll_list
, &list
);
5350 list_splice_tail(&repoll
, &list
);
5351 list_splice(&list
, &sd
->poll_list
);
5352 if (!list_empty(&sd
->poll_list
))
5353 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
5355 net_rps_action_and_irq_enable(sd
);
5357 __kfree_skb_flush();
5360 struct netdev_adjacent
{
5361 struct net_device
*dev
;
5363 /* upper master flag, there can only be one master device per list */
5366 /* counter for the number of times this device was added to us */
5369 /* private field for the users */
5372 struct list_head list
;
5373 struct rcu_head rcu
;
5376 static struct netdev_adjacent
*__netdev_find_adj(struct net_device
*adj_dev
,
5377 struct list_head
*adj_list
)
5379 struct netdev_adjacent
*adj
;
5381 list_for_each_entry(adj
, adj_list
, list
) {
5382 if (adj
->dev
== adj_dev
)
5388 static int __netdev_has_upper_dev(struct net_device
*upper_dev
, void *data
)
5390 struct net_device
*dev
= data
;
5392 return upper_dev
== dev
;
5396 * netdev_has_upper_dev - Check if device is linked to an upper device
5398 * @upper_dev: upper device to check
5400 * Find out if a device is linked to specified upper device and return true
5401 * in case it is. Note that this checks only immediate upper device,
5402 * not through a complete stack of devices. The caller must hold the RTNL lock.
5404 bool netdev_has_upper_dev(struct net_device
*dev
,
5405 struct net_device
*upper_dev
)
5409 return netdev_walk_all_upper_dev_rcu(dev
, __netdev_has_upper_dev
,
5412 EXPORT_SYMBOL(netdev_has_upper_dev
);
5415 * netdev_has_upper_dev_all - Check if device is linked to an upper device
5417 * @upper_dev: upper device to check
5419 * Find out if a device is linked to specified upper device and return true
5420 * in case it is. Note that this checks the entire upper device chain.
5421 * The caller must hold rcu lock.
5424 bool netdev_has_upper_dev_all_rcu(struct net_device
*dev
,
5425 struct net_device
*upper_dev
)
5427 return !!netdev_walk_all_upper_dev_rcu(dev
, __netdev_has_upper_dev
,
5430 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu
);
5433 * netdev_has_any_upper_dev - Check if device is linked to some device
5436 * Find out if a device is linked to an upper device and return true in case
5437 * it is. The caller must hold the RTNL lock.
5439 static bool netdev_has_any_upper_dev(struct net_device
*dev
)
5443 return !list_empty(&dev
->adj_list
.upper
);
5447 * netdev_master_upper_dev_get - Get master upper device
5450 * Find a master upper device and return pointer to it or NULL in case
5451 * it's not there. The caller must hold the RTNL lock.
5453 struct net_device
*netdev_master_upper_dev_get(struct net_device
*dev
)
5455 struct netdev_adjacent
*upper
;
5459 if (list_empty(&dev
->adj_list
.upper
))
5462 upper
= list_first_entry(&dev
->adj_list
.upper
,
5463 struct netdev_adjacent
, list
);
5464 if (likely(upper
->master
))
5468 EXPORT_SYMBOL(netdev_master_upper_dev_get
);
5471 * netdev_has_any_lower_dev - Check if device is linked to some device
5474 * Find out if a device is linked to a lower device and return true in case
5475 * it is. The caller must hold the RTNL lock.
5477 static bool netdev_has_any_lower_dev(struct net_device
*dev
)
5481 return !list_empty(&dev
->adj_list
.lower
);
5484 void *netdev_adjacent_get_private(struct list_head
*adj_list
)
5486 struct netdev_adjacent
*adj
;
5488 adj
= list_entry(adj_list
, struct netdev_adjacent
, list
);
5490 return adj
->private;
5492 EXPORT_SYMBOL(netdev_adjacent_get_private
);
5495 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
5497 * @iter: list_head ** of the current position
5499 * Gets the next device from the dev's upper list, starting from iter
5500 * position. The caller must hold RCU read lock.
5502 struct net_device
*netdev_upper_get_next_dev_rcu(struct net_device
*dev
,
5503 struct list_head
**iter
)
5505 struct netdev_adjacent
*upper
;
5507 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5509 upper
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
5511 if (&upper
->list
== &dev
->adj_list
.upper
)
5514 *iter
= &upper
->list
;
5518 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu
);
5520 static struct net_device
*netdev_next_upper_dev_rcu(struct net_device
*dev
,
5521 struct list_head
**iter
)
5523 struct netdev_adjacent
*upper
;
5525 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5527 upper
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
5529 if (&upper
->list
== &dev
->adj_list
.upper
)
5532 *iter
= &upper
->list
;
5537 int netdev_walk_all_upper_dev_rcu(struct net_device
*dev
,
5538 int (*fn
)(struct net_device
*dev
,
5542 struct net_device
*udev
;
5543 struct list_head
*iter
;
5546 for (iter
= &dev
->adj_list
.upper
,
5547 udev
= netdev_next_upper_dev_rcu(dev
, &iter
);
5549 udev
= netdev_next_upper_dev_rcu(dev
, &iter
)) {
5550 /* first is the upper device itself */
5551 ret
= fn(udev
, data
);
5555 /* then look at all of its upper devices */
5556 ret
= netdev_walk_all_upper_dev_rcu(udev
, fn
, data
);
5563 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu
);
5566 * netdev_lower_get_next_private - Get the next ->private from the
5567 * lower neighbour list
5569 * @iter: list_head ** of the current position
5571 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5572 * list, starting from iter position. The caller must hold either hold the
5573 * RTNL lock or its own locking that guarantees that the neighbour lower
5574 * list will remain unchanged.
5576 void *netdev_lower_get_next_private(struct net_device
*dev
,
5577 struct list_head
**iter
)
5579 struct netdev_adjacent
*lower
;
5581 lower
= list_entry(*iter
, struct netdev_adjacent
, list
);
5583 if (&lower
->list
== &dev
->adj_list
.lower
)
5586 *iter
= lower
->list
.next
;
5588 return lower
->private;
5590 EXPORT_SYMBOL(netdev_lower_get_next_private
);
5593 * netdev_lower_get_next_private_rcu - Get the next ->private from the
5594 * lower neighbour list, RCU
5597 * @iter: list_head ** of the current position
5599 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5600 * list, starting from iter position. The caller must hold RCU read lock.
5602 void *netdev_lower_get_next_private_rcu(struct net_device
*dev
,
5603 struct list_head
**iter
)
5605 struct netdev_adjacent
*lower
;
5607 WARN_ON_ONCE(!rcu_read_lock_held());
5609 lower
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
5611 if (&lower
->list
== &dev
->adj_list
.lower
)
5614 *iter
= &lower
->list
;
5616 return lower
->private;
5618 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu
);
5621 * netdev_lower_get_next - Get the next device from the lower neighbour
5624 * @iter: list_head ** of the current position
5626 * Gets the next netdev_adjacent from the dev's lower neighbour
5627 * list, starting from iter position. The caller must hold RTNL lock or
5628 * its own locking that guarantees that the neighbour lower
5629 * list will remain unchanged.
5631 void *netdev_lower_get_next(struct net_device
*dev
, struct list_head
**iter
)
5633 struct netdev_adjacent
*lower
;
5635 lower
= list_entry(*iter
, struct netdev_adjacent
, list
);
5637 if (&lower
->list
== &dev
->adj_list
.lower
)
5640 *iter
= lower
->list
.next
;
5644 EXPORT_SYMBOL(netdev_lower_get_next
);
5646 static struct net_device
*netdev_next_lower_dev(struct net_device
*dev
,
5647 struct list_head
**iter
)
5649 struct netdev_adjacent
*lower
;
5651 lower
= list_entry((*iter
)->next
, struct netdev_adjacent
, list
);
5653 if (&lower
->list
== &dev
->adj_list
.lower
)
5656 *iter
= &lower
->list
;
5661 int netdev_walk_all_lower_dev(struct net_device
*dev
,
5662 int (*fn
)(struct net_device
*dev
,
5666 struct net_device
*ldev
;
5667 struct list_head
*iter
;
5670 for (iter
= &dev
->adj_list
.lower
,
5671 ldev
= netdev_next_lower_dev(dev
, &iter
);
5673 ldev
= netdev_next_lower_dev(dev
, &iter
)) {
5674 /* first is the lower device itself */
5675 ret
= fn(ldev
, data
);
5679 /* then look at all of its lower devices */
5680 ret
= netdev_walk_all_lower_dev(ldev
, fn
, data
);
5687 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev
);
5689 static struct net_device
*netdev_next_lower_dev_rcu(struct net_device
*dev
,
5690 struct list_head
**iter
)
5692 struct netdev_adjacent
*lower
;
5694 lower
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
5695 if (&lower
->list
== &dev
->adj_list
.lower
)
5698 *iter
= &lower
->list
;
5703 int netdev_walk_all_lower_dev_rcu(struct net_device
*dev
,
5704 int (*fn
)(struct net_device
*dev
,
5708 struct net_device
*ldev
;
5709 struct list_head
*iter
;
5712 for (iter
= &dev
->adj_list
.lower
,
5713 ldev
= netdev_next_lower_dev_rcu(dev
, &iter
);
5715 ldev
= netdev_next_lower_dev_rcu(dev
, &iter
)) {
5716 /* first is the lower device itself */
5717 ret
= fn(ldev
, data
);
5721 /* then look at all of its lower devices */
5722 ret
= netdev_walk_all_lower_dev_rcu(ldev
, fn
, data
);
5729 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu
);
5732 * netdev_lower_get_first_private_rcu - Get the first ->private from the
5733 * lower neighbour list, RCU
5737 * Gets the first netdev_adjacent->private from the dev's lower neighbour
5738 * list. The caller must hold RCU read lock.
5740 void *netdev_lower_get_first_private_rcu(struct net_device
*dev
)
5742 struct netdev_adjacent
*lower
;
5744 lower
= list_first_or_null_rcu(&dev
->adj_list
.lower
,
5745 struct netdev_adjacent
, list
);
5747 return lower
->private;
5750 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu
);
5753 * netdev_master_upper_dev_get_rcu - Get master upper device
5756 * Find a master upper device and return pointer to it or NULL in case
5757 * it's not there. The caller must hold the RCU read lock.
5759 struct net_device
*netdev_master_upper_dev_get_rcu(struct net_device
*dev
)
5761 struct netdev_adjacent
*upper
;
5763 upper
= list_first_or_null_rcu(&dev
->adj_list
.upper
,
5764 struct netdev_adjacent
, list
);
5765 if (upper
&& likely(upper
->master
))
5769 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu
);
5771 static int netdev_adjacent_sysfs_add(struct net_device
*dev
,
5772 struct net_device
*adj_dev
,
5773 struct list_head
*dev_list
)
5775 char linkname
[IFNAMSIZ
+7];
5777 sprintf(linkname
, dev_list
== &dev
->adj_list
.upper
?
5778 "upper_%s" : "lower_%s", adj_dev
->name
);
5779 return sysfs_create_link(&(dev
->dev
.kobj
), &(adj_dev
->dev
.kobj
),
5782 static void netdev_adjacent_sysfs_del(struct net_device
*dev
,
5784 struct list_head
*dev_list
)
5786 char linkname
[IFNAMSIZ
+7];
5788 sprintf(linkname
, dev_list
== &dev
->adj_list
.upper
?
5789 "upper_%s" : "lower_%s", name
);
5790 sysfs_remove_link(&(dev
->dev
.kobj
), linkname
);
5793 static inline bool netdev_adjacent_is_neigh_list(struct net_device
*dev
,
5794 struct net_device
*adj_dev
,
5795 struct list_head
*dev_list
)
5797 return (dev_list
== &dev
->adj_list
.upper
||
5798 dev_list
== &dev
->adj_list
.lower
) &&
5799 net_eq(dev_net(dev
), dev_net(adj_dev
));
5802 static int __netdev_adjacent_dev_insert(struct net_device
*dev
,
5803 struct net_device
*adj_dev
,
5804 struct list_head
*dev_list
,
5805 void *private, bool master
)
5807 struct netdev_adjacent
*adj
;
5810 adj
= __netdev_find_adj(adj_dev
, dev_list
);
5814 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
5815 dev
->name
, adj_dev
->name
, adj
->ref_nr
);
5820 adj
= kmalloc(sizeof(*adj
), GFP_KERNEL
);
5825 adj
->master
= master
;
5827 adj
->private = private;
5830 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
5831 dev
->name
, adj_dev
->name
, adj
->ref_nr
, adj_dev
->name
);
5833 if (netdev_adjacent_is_neigh_list(dev
, adj_dev
, dev_list
)) {
5834 ret
= netdev_adjacent_sysfs_add(dev
, adj_dev
, dev_list
);
5839 /* Ensure that master link is always the first item in list. */
5841 ret
= sysfs_create_link(&(dev
->dev
.kobj
),
5842 &(adj_dev
->dev
.kobj
), "master");
5844 goto remove_symlinks
;
5846 list_add_rcu(&adj
->list
, dev_list
);
5848 list_add_tail_rcu(&adj
->list
, dev_list
);
5854 if (netdev_adjacent_is_neigh_list(dev
, adj_dev
, dev_list
))
5855 netdev_adjacent_sysfs_del(dev
, adj_dev
->name
, dev_list
);
5863 static void __netdev_adjacent_dev_remove(struct net_device
*dev
,
5864 struct net_device
*adj_dev
,
5866 struct list_head
*dev_list
)
5868 struct netdev_adjacent
*adj
;
5870 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
5871 dev
->name
, adj_dev
->name
, ref_nr
);
5873 adj
= __netdev_find_adj(adj_dev
, dev_list
);
5876 pr_err("Adjacency does not exist for device %s from %s\n",
5877 dev
->name
, adj_dev
->name
);
5882 if (adj
->ref_nr
> ref_nr
) {
5883 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
5884 dev
->name
, adj_dev
->name
, ref_nr
,
5885 adj
->ref_nr
- ref_nr
);
5886 adj
->ref_nr
-= ref_nr
;
5891 sysfs_remove_link(&(dev
->dev
.kobj
), "master");
5893 if (netdev_adjacent_is_neigh_list(dev
, adj_dev
, dev_list
))
5894 netdev_adjacent_sysfs_del(dev
, adj_dev
->name
, dev_list
);
5896 list_del_rcu(&adj
->list
);
5897 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
5898 adj_dev
->name
, dev
->name
, adj_dev
->name
);
5900 kfree_rcu(adj
, rcu
);
5903 static int __netdev_adjacent_dev_link_lists(struct net_device
*dev
,
5904 struct net_device
*upper_dev
,
5905 struct list_head
*up_list
,
5906 struct list_head
*down_list
,
5907 void *private, bool master
)
5911 ret
= __netdev_adjacent_dev_insert(dev
, upper_dev
, up_list
,
5916 ret
= __netdev_adjacent_dev_insert(upper_dev
, dev
, down_list
,
5919 __netdev_adjacent_dev_remove(dev
, upper_dev
, 1, up_list
);
5926 static void __netdev_adjacent_dev_unlink_lists(struct net_device
*dev
,
5927 struct net_device
*upper_dev
,
5929 struct list_head
*up_list
,
5930 struct list_head
*down_list
)
5932 __netdev_adjacent_dev_remove(dev
, upper_dev
, ref_nr
, up_list
);
5933 __netdev_adjacent_dev_remove(upper_dev
, dev
, ref_nr
, down_list
);
5936 static int __netdev_adjacent_dev_link_neighbour(struct net_device
*dev
,
5937 struct net_device
*upper_dev
,
5938 void *private, bool master
)
5940 return __netdev_adjacent_dev_link_lists(dev
, upper_dev
,
5941 &dev
->adj_list
.upper
,
5942 &upper_dev
->adj_list
.lower
,
5946 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device
*dev
,
5947 struct net_device
*upper_dev
)
5949 __netdev_adjacent_dev_unlink_lists(dev
, upper_dev
, 1,
5950 &dev
->adj_list
.upper
,
5951 &upper_dev
->adj_list
.lower
);
5954 static int __netdev_upper_dev_link(struct net_device
*dev
,
5955 struct net_device
*upper_dev
, bool master
,
5956 void *upper_priv
, void *upper_info
)
5958 struct netdev_notifier_changeupper_info changeupper_info
;
5963 if (dev
== upper_dev
)
5966 /* To prevent loops, check if dev is not upper device to upper_dev. */
5967 if (netdev_has_upper_dev(upper_dev
, dev
))
5970 if (netdev_has_upper_dev(dev
, upper_dev
))
5973 if (master
&& netdev_master_upper_dev_get(dev
))
5976 changeupper_info
.upper_dev
= upper_dev
;
5977 changeupper_info
.master
= master
;
5978 changeupper_info
.linking
= true;
5979 changeupper_info
.upper_info
= upper_info
;
5981 ret
= call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER
, dev
,
5982 &changeupper_info
.info
);
5983 ret
= notifier_to_errno(ret
);
5987 ret
= __netdev_adjacent_dev_link_neighbour(dev
, upper_dev
, upper_priv
,
5992 ret
= call_netdevice_notifiers_info(NETDEV_CHANGEUPPER
, dev
,
5993 &changeupper_info
.info
);
5994 ret
= notifier_to_errno(ret
);
6001 __netdev_adjacent_dev_unlink_neighbour(dev
, upper_dev
);
6007 * netdev_upper_dev_link - Add a link to the upper device
6009 * @upper_dev: new upper device
6011 * Adds a link to device which is upper to this one. The caller must hold
6012 * the RTNL lock. On a failure a negative errno code is returned.
6013 * On success the reference counts are adjusted and the function
6016 int netdev_upper_dev_link(struct net_device
*dev
,
6017 struct net_device
*upper_dev
)
6019 return __netdev_upper_dev_link(dev
, upper_dev
, false, NULL
, NULL
);
6021 EXPORT_SYMBOL(netdev_upper_dev_link
);
6024 * netdev_master_upper_dev_link - Add a master link to the upper device
6026 * @upper_dev: new upper device
6027 * @upper_priv: upper device private
6028 * @upper_info: upper info to be passed down via notifier
6030 * Adds a link to device which is upper to this one. In this case, only
6031 * one master upper device can be linked, although other non-master devices
6032 * might be linked as well. The caller must hold the RTNL lock.
6033 * On a failure a negative errno code is returned. On success the reference
6034 * counts are adjusted and the function returns zero.
6036 int netdev_master_upper_dev_link(struct net_device
*dev
,
6037 struct net_device
*upper_dev
,
6038 void *upper_priv
, void *upper_info
)
6040 return __netdev_upper_dev_link(dev
, upper_dev
, true,
6041 upper_priv
, upper_info
);
6043 EXPORT_SYMBOL(netdev_master_upper_dev_link
);
6046 * netdev_upper_dev_unlink - Removes a link to upper device
6048 * @upper_dev: new upper device
6050 * Removes a link to device which is upper to this one. The caller must hold
6053 void netdev_upper_dev_unlink(struct net_device
*dev
,
6054 struct net_device
*upper_dev
)
6056 struct netdev_notifier_changeupper_info changeupper_info
;
6060 changeupper_info
.upper_dev
= upper_dev
;
6061 changeupper_info
.master
= netdev_master_upper_dev_get(dev
) == upper_dev
;
6062 changeupper_info
.linking
= false;
6064 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER
, dev
,
6065 &changeupper_info
.info
);
6067 __netdev_adjacent_dev_unlink_neighbour(dev
, upper_dev
);
6069 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER
, dev
,
6070 &changeupper_info
.info
);
6072 EXPORT_SYMBOL(netdev_upper_dev_unlink
);
6075 * netdev_bonding_info_change - Dispatch event about slave change
6077 * @bonding_info: info to dispatch
6079 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
6080 * The caller must hold the RTNL lock.
6082 void netdev_bonding_info_change(struct net_device
*dev
,
6083 struct netdev_bonding_info
*bonding_info
)
6085 struct netdev_notifier_bonding_info info
;
6087 memcpy(&info
.bonding_info
, bonding_info
,
6088 sizeof(struct netdev_bonding_info
));
6089 call_netdevice_notifiers_info(NETDEV_BONDING_INFO
, dev
,
6092 EXPORT_SYMBOL(netdev_bonding_info_change
);
6094 static void netdev_adjacent_add_links(struct net_device
*dev
)
6096 struct netdev_adjacent
*iter
;
6098 struct net
*net
= dev_net(dev
);
6100 list_for_each_entry(iter
, &dev
->adj_list
.upper
, list
) {
6101 if (!net_eq(net
, dev_net(iter
->dev
)))
6103 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
6104 &iter
->dev
->adj_list
.lower
);
6105 netdev_adjacent_sysfs_add(dev
, iter
->dev
,
6106 &dev
->adj_list
.upper
);
6109 list_for_each_entry(iter
, &dev
->adj_list
.lower
, list
) {
6110 if (!net_eq(net
, dev_net(iter
->dev
)))
6112 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
6113 &iter
->dev
->adj_list
.upper
);
6114 netdev_adjacent_sysfs_add(dev
, iter
->dev
,
6115 &dev
->adj_list
.lower
);
6119 static void netdev_adjacent_del_links(struct net_device
*dev
)
6121 struct netdev_adjacent
*iter
;
6123 struct net
*net
= dev_net(dev
);
6125 list_for_each_entry(iter
, &dev
->adj_list
.upper
, list
) {
6126 if (!net_eq(net
, dev_net(iter
->dev
)))
6128 netdev_adjacent_sysfs_del(iter
->dev
, dev
->name
,
6129 &iter
->dev
->adj_list
.lower
);
6130 netdev_adjacent_sysfs_del(dev
, iter
->dev
->name
,
6131 &dev
->adj_list
.upper
);
6134 list_for_each_entry(iter
, &dev
->adj_list
.lower
, list
) {
6135 if (!net_eq(net
, dev_net(iter
->dev
)))
6137 netdev_adjacent_sysfs_del(iter
->dev
, dev
->name
,
6138 &iter
->dev
->adj_list
.upper
);
6139 netdev_adjacent_sysfs_del(dev
, iter
->dev
->name
,
6140 &dev
->adj_list
.lower
);
6144 void netdev_adjacent_rename_links(struct net_device
*dev
, char *oldname
)
6146 struct netdev_adjacent
*iter
;
6148 struct net
*net
= dev_net(dev
);
6150 list_for_each_entry(iter
, &dev
->adj_list
.upper
, list
) {
6151 if (!net_eq(net
, dev_net(iter
->dev
)))
6153 netdev_adjacent_sysfs_del(iter
->dev
, oldname
,
6154 &iter
->dev
->adj_list
.lower
);
6155 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
6156 &iter
->dev
->adj_list
.lower
);
6159 list_for_each_entry(iter
, &dev
->adj_list
.lower
, list
) {
6160 if (!net_eq(net
, dev_net(iter
->dev
)))
6162 netdev_adjacent_sysfs_del(iter
->dev
, oldname
,
6163 &iter
->dev
->adj_list
.upper
);
6164 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
6165 &iter
->dev
->adj_list
.upper
);
6169 void *netdev_lower_dev_get_private(struct net_device
*dev
,
6170 struct net_device
*lower_dev
)
6172 struct netdev_adjacent
*lower
;
6176 lower
= __netdev_find_adj(lower_dev
, &dev
->adj_list
.lower
);
6180 return lower
->private;
6182 EXPORT_SYMBOL(netdev_lower_dev_get_private
);
6185 int dev_get_nest_level(struct net_device
*dev
)
6187 struct net_device
*lower
= NULL
;
6188 struct list_head
*iter
;
6194 netdev_for_each_lower_dev(dev
, lower
, iter
) {
6195 nest
= dev_get_nest_level(lower
);
6196 if (max_nest
< nest
)
6200 return max_nest
+ 1;
6202 EXPORT_SYMBOL(dev_get_nest_level
);
6205 * netdev_lower_change - Dispatch event about lower device state change
6206 * @lower_dev: device
6207 * @lower_state_info: state to dispatch
6209 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
6210 * The caller must hold the RTNL lock.
6212 void netdev_lower_state_changed(struct net_device
*lower_dev
,
6213 void *lower_state_info
)
6215 struct netdev_notifier_changelowerstate_info changelowerstate_info
;
6218 changelowerstate_info
.lower_state_info
= lower_state_info
;
6219 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE
, lower_dev
,
6220 &changelowerstate_info
.info
);
6222 EXPORT_SYMBOL(netdev_lower_state_changed
);
6224 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
6226 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6228 if (ops
->ndo_change_rx_flags
)
6229 ops
->ndo_change_rx_flags(dev
, flags
);
6232 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
, bool notify
)
6234 unsigned int old_flags
= dev
->flags
;
6240 dev
->flags
|= IFF_PROMISC
;
6241 dev
->promiscuity
+= inc
;
6242 if (dev
->promiscuity
== 0) {
6245 * If inc causes overflow, untouch promisc and return error.
6248 dev
->flags
&= ~IFF_PROMISC
;
6250 dev
->promiscuity
-= inc
;
6251 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
6256 if (dev
->flags
!= old_flags
) {
6257 pr_info("device %s %s promiscuous mode\n",
6259 dev
->flags
& IFF_PROMISC
? "entered" : "left");
6260 if (audit_enabled
) {
6261 current_uid_gid(&uid
, &gid
);
6262 audit_log(current
->audit_context
, GFP_ATOMIC
,
6263 AUDIT_ANOM_PROMISCUOUS
,
6264 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
6265 dev
->name
, (dev
->flags
& IFF_PROMISC
),
6266 (old_flags
& IFF_PROMISC
),
6267 from_kuid(&init_user_ns
, audit_get_loginuid(current
)),
6268 from_kuid(&init_user_ns
, uid
),
6269 from_kgid(&init_user_ns
, gid
),
6270 audit_get_sessionid(current
));
6273 dev_change_rx_flags(dev
, IFF_PROMISC
);
6276 __dev_notify_flags(dev
, old_flags
, IFF_PROMISC
);
6281 * dev_set_promiscuity - update promiscuity count on a device
6285 * Add or remove promiscuity from a device. While the count in the device
6286 * remains above zero the interface remains promiscuous. Once it hits zero
6287 * the device reverts back to normal filtering operation. A negative inc
6288 * value is used to drop promiscuity on the device.
6289 * Return 0 if successful or a negative errno code on error.
6291 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
6293 unsigned int old_flags
= dev
->flags
;
6296 err
= __dev_set_promiscuity(dev
, inc
, true);
6299 if (dev
->flags
!= old_flags
)
6300 dev_set_rx_mode(dev
);
6303 EXPORT_SYMBOL(dev_set_promiscuity
);
6305 static int __dev_set_allmulti(struct net_device
*dev
, int inc
, bool notify
)
6307 unsigned int old_flags
= dev
->flags
, old_gflags
= dev
->gflags
;
6311 dev
->flags
|= IFF_ALLMULTI
;
6312 dev
->allmulti
+= inc
;
6313 if (dev
->allmulti
== 0) {
6316 * If inc causes overflow, untouch allmulti and return error.
6319 dev
->flags
&= ~IFF_ALLMULTI
;
6321 dev
->allmulti
-= inc
;
6322 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
6327 if (dev
->flags
^ old_flags
) {
6328 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
6329 dev_set_rx_mode(dev
);
6331 __dev_notify_flags(dev
, old_flags
,
6332 dev
->gflags
^ old_gflags
);
6338 * dev_set_allmulti - update allmulti count on a device
6342 * Add or remove reception of all multicast frames to a device. While the
6343 * count in the device remains above zero the interface remains listening
6344 * to all interfaces. Once it hits zero the device reverts back to normal
6345 * filtering operation. A negative @inc value is used to drop the counter
6346 * when releasing a resource needing all multicasts.
6347 * Return 0 if successful or a negative errno code on error.
6350 int dev_set_allmulti(struct net_device
*dev
, int inc
)
6352 return __dev_set_allmulti(dev
, inc
, true);
6354 EXPORT_SYMBOL(dev_set_allmulti
);
6357 * Upload unicast and multicast address lists to device and
6358 * configure RX filtering. When the device doesn't support unicast
6359 * filtering it is put in promiscuous mode while unicast addresses
6362 void __dev_set_rx_mode(struct net_device
*dev
)
6364 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6366 /* dev_open will call this function so the list will stay sane. */
6367 if (!(dev
->flags
&IFF_UP
))
6370 if (!netif_device_present(dev
))
6373 if (!(dev
->priv_flags
& IFF_UNICAST_FLT
)) {
6374 /* Unicast addresses changes may only happen under the rtnl,
6375 * therefore calling __dev_set_promiscuity here is safe.
6377 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
6378 __dev_set_promiscuity(dev
, 1, false);
6379 dev
->uc_promisc
= true;
6380 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
6381 __dev_set_promiscuity(dev
, -1, false);
6382 dev
->uc_promisc
= false;
6386 if (ops
->ndo_set_rx_mode
)
6387 ops
->ndo_set_rx_mode(dev
);
6390 void dev_set_rx_mode(struct net_device
*dev
)
6392 netif_addr_lock_bh(dev
);
6393 __dev_set_rx_mode(dev
);
6394 netif_addr_unlock_bh(dev
);
6398 * dev_get_flags - get flags reported to userspace
6401 * Get the combination of flag bits exported through APIs to userspace.
6403 unsigned int dev_get_flags(const struct net_device
*dev
)
6407 flags
= (dev
->flags
& ~(IFF_PROMISC
|
6412 (dev
->gflags
& (IFF_PROMISC
|
6415 if (netif_running(dev
)) {
6416 if (netif_oper_up(dev
))
6417 flags
|= IFF_RUNNING
;
6418 if (netif_carrier_ok(dev
))
6419 flags
|= IFF_LOWER_UP
;
6420 if (netif_dormant(dev
))
6421 flags
|= IFF_DORMANT
;
6426 EXPORT_SYMBOL(dev_get_flags
);
6428 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
6430 unsigned int old_flags
= dev
->flags
;
6436 * Set the flags on our device.
6439 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
6440 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
6442 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
6446 * Load in the correct multicast list now the flags have changed.
6449 if ((old_flags
^ flags
) & IFF_MULTICAST
)
6450 dev_change_rx_flags(dev
, IFF_MULTICAST
);
6452 dev_set_rx_mode(dev
);
6455 * Have we downed the interface. We handle IFF_UP ourselves
6456 * according to user attempts to set it, rather than blindly
6461 if ((old_flags
^ flags
) & IFF_UP
)
6462 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
6464 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
6465 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
6466 unsigned int old_flags
= dev
->flags
;
6468 dev
->gflags
^= IFF_PROMISC
;
6470 if (__dev_set_promiscuity(dev
, inc
, false) >= 0)
6471 if (dev
->flags
!= old_flags
)
6472 dev_set_rx_mode(dev
);
6475 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
6476 * is important. Some (broken) drivers set IFF_PROMISC, when
6477 * IFF_ALLMULTI is requested not asking us and not reporting.
6479 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
6480 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
6482 dev
->gflags
^= IFF_ALLMULTI
;
6483 __dev_set_allmulti(dev
, inc
, false);
6489 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
,
6490 unsigned int gchanges
)
6492 unsigned int changes
= dev
->flags
^ old_flags
;
6495 rtmsg_ifinfo(RTM_NEWLINK
, dev
, gchanges
, GFP_ATOMIC
);
6497 if (changes
& IFF_UP
) {
6498 if (dev
->flags
& IFF_UP
)
6499 call_netdevice_notifiers(NETDEV_UP
, dev
);
6501 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
6504 if (dev
->flags
& IFF_UP
&&
6505 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
))) {
6506 struct netdev_notifier_change_info change_info
;
6508 change_info
.flags_changed
= changes
;
6509 call_netdevice_notifiers_info(NETDEV_CHANGE
, dev
,
6515 * dev_change_flags - change device settings
6517 * @flags: device state flags
6519 * Change settings on device based state flags. The flags are
6520 * in the userspace exported format.
6522 int dev_change_flags(struct net_device
*dev
, unsigned int flags
)
6525 unsigned int changes
, old_flags
= dev
->flags
, old_gflags
= dev
->gflags
;
6527 ret
= __dev_change_flags(dev
, flags
);
6531 changes
= (old_flags
^ dev
->flags
) | (old_gflags
^ dev
->gflags
);
6532 __dev_notify_flags(dev
, old_flags
, changes
);
6535 EXPORT_SYMBOL(dev_change_flags
);
6537 static int __dev_set_mtu(struct net_device
*dev
, int new_mtu
)
6539 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6541 if (ops
->ndo_change_mtu
)
6542 return ops
->ndo_change_mtu(dev
, new_mtu
);
6549 * dev_set_mtu - Change maximum transfer unit
6551 * @new_mtu: new transfer unit
6553 * Change the maximum transfer size of the network device.
6555 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
6559 if (new_mtu
== dev
->mtu
)
6562 /* MTU must be positive, and in range */
6563 if (new_mtu
< 0 || new_mtu
< dev
->min_mtu
) {
6564 net_err_ratelimited("%s: Invalid MTU %d requested, hw min %d\n",
6565 dev
->name
, new_mtu
, dev
->min_mtu
);
6569 if (dev
->max_mtu
> 0 && new_mtu
> dev
->max_mtu
) {
6570 net_err_ratelimited("%s: Invalid MTU %d requested, hw max %d\n",
6571 dev
->name
, new_mtu
, dev
->max_mtu
);
6575 if (!netif_device_present(dev
))
6578 err
= call_netdevice_notifiers(NETDEV_PRECHANGEMTU
, dev
);
6579 err
= notifier_to_errno(err
);
6583 orig_mtu
= dev
->mtu
;
6584 err
= __dev_set_mtu(dev
, new_mtu
);
6587 err
= call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
6588 err
= notifier_to_errno(err
);
6590 /* setting mtu back and notifying everyone again,
6591 * so that they have a chance to revert changes.
6593 __dev_set_mtu(dev
, orig_mtu
);
6594 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
6599 EXPORT_SYMBOL(dev_set_mtu
);
6602 * dev_set_group - Change group this device belongs to
6604 * @new_group: group this device should belong to
6606 void dev_set_group(struct net_device
*dev
, int new_group
)
6608 dev
->group
= new_group
;
6610 EXPORT_SYMBOL(dev_set_group
);
6613 * dev_set_mac_address - Change Media Access Control Address
6617 * Change the hardware (MAC) address of the device
6619 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
6621 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6624 if (!ops
->ndo_set_mac_address
)
6626 if (sa
->sa_family
!= dev
->type
)
6628 if (!netif_device_present(dev
))
6630 err
= ops
->ndo_set_mac_address(dev
, sa
);
6633 dev
->addr_assign_type
= NET_ADDR_SET
;
6634 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
6635 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
6638 EXPORT_SYMBOL(dev_set_mac_address
);
6641 * dev_change_carrier - Change device carrier
6643 * @new_carrier: new value
6645 * Change device carrier
6647 int dev_change_carrier(struct net_device
*dev
, bool new_carrier
)
6649 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6651 if (!ops
->ndo_change_carrier
)
6653 if (!netif_device_present(dev
))
6655 return ops
->ndo_change_carrier(dev
, new_carrier
);
6657 EXPORT_SYMBOL(dev_change_carrier
);
6660 * dev_get_phys_port_id - Get device physical port ID
6664 * Get device physical port ID
6666 int dev_get_phys_port_id(struct net_device
*dev
,
6667 struct netdev_phys_item_id
*ppid
)
6669 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6671 if (!ops
->ndo_get_phys_port_id
)
6673 return ops
->ndo_get_phys_port_id(dev
, ppid
);
6675 EXPORT_SYMBOL(dev_get_phys_port_id
);
6678 * dev_get_phys_port_name - Get device physical port name
6681 * @len: limit of bytes to copy to name
6683 * Get device physical port name
6685 int dev_get_phys_port_name(struct net_device
*dev
,
6686 char *name
, size_t len
)
6688 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6690 if (!ops
->ndo_get_phys_port_name
)
6692 return ops
->ndo_get_phys_port_name(dev
, name
, len
);
6694 EXPORT_SYMBOL(dev_get_phys_port_name
);
6697 * dev_change_proto_down - update protocol port state information
6699 * @proto_down: new value
6701 * This info can be used by switch drivers to set the phys state of the
6704 int dev_change_proto_down(struct net_device
*dev
, bool proto_down
)
6706 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6708 if (!ops
->ndo_change_proto_down
)
6710 if (!netif_device_present(dev
))
6712 return ops
->ndo_change_proto_down(dev
, proto_down
);
6714 EXPORT_SYMBOL(dev_change_proto_down
);
6717 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
6719 * @fd: new program fd or negative value to clear
6720 * @flags: xdp-related flags
6722 * Set or clear a bpf program for a device
6724 int dev_change_xdp_fd(struct net_device
*dev
, int fd
, u32 flags
)
6726 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6727 struct bpf_prog
*prog
= NULL
;
6728 struct netdev_xdp xdp
;
6736 if (flags
& XDP_FLAGS_UPDATE_IF_NOEXIST
) {
6737 memset(&xdp
, 0, sizeof(xdp
));
6738 xdp
.command
= XDP_QUERY_PROG
;
6740 err
= ops
->ndo_xdp(dev
, &xdp
);
6743 if (xdp
.prog_attached
)
6747 prog
= bpf_prog_get_type(fd
, BPF_PROG_TYPE_XDP
);
6749 return PTR_ERR(prog
);
6752 memset(&xdp
, 0, sizeof(xdp
));
6753 xdp
.command
= XDP_SETUP_PROG
;
6756 err
= ops
->ndo_xdp(dev
, &xdp
);
6757 if (err
< 0 && prog
)
6762 EXPORT_SYMBOL(dev_change_xdp_fd
);
6765 * dev_new_index - allocate an ifindex
6766 * @net: the applicable net namespace
6768 * Returns a suitable unique value for a new device interface
6769 * number. The caller must hold the rtnl semaphore or the
6770 * dev_base_lock to be sure it remains unique.
6772 static int dev_new_index(struct net
*net
)
6774 int ifindex
= net
->ifindex
;
6779 if (!__dev_get_by_index(net
, ifindex
))
6780 return net
->ifindex
= ifindex
;
6784 /* Delayed registration/unregisteration */
6785 static LIST_HEAD(net_todo_list
);
6786 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq
);
6788 static void net_set_todo(struct net_device
*dev
)
6790 list_add_tail(&dev
->todo_list
, &net_todo_list
);
6791 dev_net(dev
)->dev_unreg_count
++;
6794 static void rollback_registered_many(struct list_head
*head
)
6796 struct net_device
*dev
, *tmp
;
6797 LIST_HEAD(close_head
);
6799 BUG_ON(dev_boot_phase
);
6802 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
6803 /* Some devices call without registering
6804 * for initialization unwind. Remove those
6805 * devices and proceed with the remaining.
6807 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
6808 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
6812 list_del(&dev
->unreg_list
);
6815 dev
->dismantle
= true;
6816 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
6819 /* If device is running, close it first. */
6820 list_for_each_entry(dev
, head
, unreg_list
)
6821 list_add_tail(&dev
->close_list
, &close_head
);
6822 dev_close_many(&close_head
, true);
6824 list_for_each_entry(dev
, head
, unreg_list
) {
6825 /* And unlink it from device chain. */
6826 unlist_netdevice(dev
);
6828 dev
->reg_state
= NETREG_UNREGISTERING
;
6830 flush_all_backlogs();
6834 list_for_each_entry(dev
, head
, unreg_list
) {
6835 struct sk_buff
*skb
= NULL
;
6837 /* Shutdown queueing discipline. */
6841 /* Notify protocols, that we are about to destroy
6842 * this device. They should clean all the things.
6844 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6846 if (!dev
->rtnl_link_ops
||
6847 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
6848 skb
= rtmsg_ifinfo_build_skb(RTM_DELLINK
, dev
, ~0U,
6852 * Flush the unicast and multicast chains
6857 if (dev
->netdev_ops
->ndo_uninit
)
6858 dev
->netdev_ops
->ndo_uninit(dev
);
6861 rtmsg_ifinfo_send(skb
, dev
, GFP_KERNEL
);
6863 /* Notifier chain MUST detach us all upper devices. */
6864 WARN_ON(netdev_has_any_upper_dev(dev
));
6865 WARN_ON(netdev_has_any_lower_dev(dev
));
6867 /* Remove entries from kobject tree */
6868 netdev_unregister_kobject(dev
);
6870 /* Remove XPS queueing entries */
6871 netif_reset_xps_queues_gt(dev
, 0);
6877 list_for_each_entry(dev
, head
, unreg_list
)
6881 static void rollback_registered(struct net_device
*dev
)
6885 list_add(&dev
->unreg_list
, &single
);
6886 rollback_registered_many(&single
);
6890 static netdev_features_t
netdev_sync_upper_features(struct net_device
*lower
,
6891 struct net_device
*upper
, netdev_features_t features
)
6893 netdev_features_t upper_disables
= NETIF_F_UPPER_DISABLES
;
6894 netdev_features_t feature
;
6897 for_each_netdev_feature(&upper_disables
, feature_bit
) {
6898 feature
= __NETIF_F_BIT(feature_bit
);
6899 if (!(upper
->wanted_features
& feature
)
6900 && (features
& feature
)) {
6901 netdev_dbg(lower
, "Dropping feature %pNF, upper dev %s has it off.\n",
6902 &feature
, upper
->name
);
6903 features
&= ~feature
;
6910 static void netdev_sync_lower_features(struct net_device
*upper
,
6911 struct net_device
*lower
, netdev_features_t features
)
6913 netdev_features_t upper_disables
= NETIF_F_UPPER_DISABLES
;
6914 netdev_features_t feature
;
6917 for_each_netdev_feature(&upper_disables
, feature_bit
) {
6918 feature
= __NETIF_F_BIT(feature_bit
);
6919 if (!(features
& feature
) && (lower
->features
& feature
)) {
6920 netdev_dbg(upper
, "Disabling feature %pNF on lower dev %s.\n",
6921 &feature
, lower
->name
);
6922 lower
->wanted_features
&= ~feature
;
6923 netdev_update_features(lower
);
6925 if (unlikely(lower
->features
& feature
))
6926 netdev_WARN(upper
, "failed to disable %pNF on %s!\n",
6927 &feature
, lower
->name
);
6932 static netdev_features_t
netdev_fix_features(struct net_device
*dev
,
6933 netdev_features_t features
)
6935 /* Fix illegal checksum combinations */
6936 if ((features
& NETIF_F_HW_CSUM
) &&
6937 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
6938 netdev_warn(dev
, "mixed HW and IP checksum settings.\n");
6939 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
6942 /* TSO requires that SG is present as well. */
6943 if ((features
& NETIF_F_ALL_TSO
) && !(features
& NETIF_F_SG
)) {
6944 netdev_dbg(dev
, "Dropping TSO features since no SG feature.\n");
6945 features
&= ~NETIF_F_ALL_TSO
;
6948 if ((features
& NETIF_F_TSO
) && !(features
& NETIF_F_HW_CSUM
) &&
6949 !(features
& NETIF_F_IP_CSUM
)) {
6950 netdev_dbg(dev
, "Dropping TSO features since no CSUM feature.\n");
6951 features
&= ~NETIF_F_TSO
;
6952 features
&= ~NETIF_F_TSO_ECN
;
6955 if ((features
& NETIF_F_TSO6
) && !(features
& NETIF_F_HW_CSUM
) &&
6956 !(features
& NETIF_F_IPV6_CSUM
)) {
6957 netdev_dbg(dev
, "Dropping TSO6 features since no CSUM feature.\n");
6958 features
&= ~NETIF_F_TSO6
;
6961 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
6962 if ((features
& NETIF_F_TSO_MANGLEID
) && !(features
& NETIF_F_TSO
))
6963 features
&= ~NETIF_F_TSO_MANGLEID
;
6965 /* TSO ECN requires that TSO is present as well. */
6966 if ((features
& NETIF_F_ALL_TSO
) == NETIF_F_TSO_ECN
)
6967 features
&= ~NETIF_F_TSO_ECN
;
6969 /* Software GSO depends on SG. */
6970 if ((features
& NETIF_F_GSO
) && !(features
& NETIF_F_SG
)) {
6971 netdev_dbg(dev
, "Dropping NETIF_F_GSO since no SG feature.\n");
6972 features
&= ~NETIF_F_GSO
;
6975 /* UFO needs SG and checksumming */
6976 if (features
& NETIF_F_UFO
) {
6977 /* maybe split UFO into V4 and V6? */
6978 if (!(features
& NETIF_F_HW_CSUM
) &&
6979 ((features
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
)) !=
6980 (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
))) {
6982 "Dropping NETIF_F_UFO since no checksum offload features.\n");
6983 features
&= ~NETIF_F_UFO
;
6986 if (!(features
& NETIF_F_SG
)) {
6988 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
6989 features
&= ~NETIF_F_UFO
;
6993 /* GSO partial features require GSO partial be set */
6994 if ((features
& dev
->gso_partial_features
) &&
6995 !(features
& NETIF_F_GSO_PARTIAL
)) {
6997 "Dropping partially supported GSO features since no GSO partial.\n");
6998 features
&= ~dev
->gso_partial_features
;
7004 int __netdev_update_features(struct net_device
*dev
)
7006 struct net_device
*upper
, *lower
;
7007 netdev_features_t features
;
7008 struct list_head
*iter
;
7013 features
= netdev_get_wanted_features(dev
);
7015 if (dev
->netdev_ops
->ndo_fix_features
)
7016 features
= dev
->netdev_ops
->ndo_fix_features(dev
, features
);
7018 /* driver might be less strict about feature dependencies */
7019 features
= netdev_fix_features(dev
, features
);
7021 /* some features can't be enabled if they're off an an upper device */
7022 netdev_for_each_upper_dev_rcu(dev
, upper
, iter
)
7023 features
= netdev_sync_upper_features(dev
, upper
, features
);
7025 if (dev
->features
== features
)
7028 netdev_dbg(dev
, "Features changed: %pNF -> %pNF\n",
7029 &dev
->features
, &features
);
7031 if (dev
->netdev_ops
->ndo_set_features
)
7032 err
= dev
->netdev_ops
->ndo_set_features(dev
, features
);
7036 if (unlikely(err
< 0)) {
7038 "set_features() failed (%d); wanted %pNF, left %pNF\n",
7039 err
, &features
, &dev
->features
);
7040 /* return non-0 since some features might have changed and
7041 * it's better to fire a spurious notification than miss it
7047 /* some features must be disabled on lower devices when disabled
7048 * on an upper device (think: bonding master or bridge)
7050 netdev_for_each_lower_dev(dev
, lower
, iter
)
7051 netdev_sync_lower_features(dev
, lower
, features
);
7054 dev
->features
= features
;
7056 return err
< 0 ? 0 : 1;
7060 * netdev_update_features - recalculate device features
7061 * @dev: the device to check
7063 * Recalculate dev->features set and send notifications if it
7064 * has changed. Should be called after driver or hardware dependent
7065 * conditions might have changed that influence the features.
7067 void netdev_update_features(struct net_device
*dev
)
7069 if (__netdev_update_features(dev
))
7070 netdev_features_change(dev
);
7072 EXPORT_SYMBOL(netdev_update_features
);
7075 * netdev_change_features - recalculate device features
7076 * @dev: the device to check
7078 * Recalculate dev->features set and send notifications even
7079 * if they have not changed. Should be called instead of
7080 * netdev_update_features() if also dev->vlan_features might
7081 * have changed to allow the changes to be propagated to stacked
7084 void netdev_change_features(struct net_device
*dev
)
7086 __netdev_update_features(dev
);
7087 netdev_features_change(dev
);
7089 EXPORT_SYMBOL(netdev_change_features
);
7092 * netif_stacked_transfer_operstate - transfer operstate
7093 * @rootdev: the root or lower level device to transfer state from
7094 * @dev: the device to transfer operstate to
7096 * Transfer operational state from root to device. This is normally
7097 * called when a stacking relationship exists between the root
7098 * device and the device(a leaf device).
7100 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
7101 struct net_device
*dev
)
7103 if (rootdev
->operstate
== IF_OPER_DORMANT
)
7104 netif_dormant_on(dev
);
7106 netif_dormant_off(dev
);
7108 if (netif_carrier_ok(rootdev
)) {
7109 if (!netif_carrier_ok(dev
))
7110 netif_carrier_on(dev
);
7112 if (netif_carrier_ok(dev
))
7113 netif_carrier_off(dev
);
7116 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
7119 static int netif_alloc_rx_queues(struct net_device
*dev
)
7121 unsigned int i
, count
= dev
->num_rx_queues
;
7122 struct netdev_rx_queue
*rx
;
7123 size_t sz
= count
* sizeof(*rx
);
7127 rx
= kzalloc(sz
, GFP_KERNEL
| __GFP_NOWARN
| __GFP_REPEAT
);
7135 for (i
= 0; i
< count
; i
++)
7141 static void netdev_init_one_queue(struct net_device
*dev
,
7142 struct netdev_queue
*queue
, void *_unused
)
7144 /* Initialize queue lock */
7145 spin_lock_init(&queue
->_xmit_lock
);
7146 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
7147 queue
->xmit_lock_owner
= -1;
7148 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
7151 dql_init(&queue
->dql
, HZ
);
7155 static void netif_free_tx_queues(struct net_device
*dev
)
7160 static int netif_alloc_netdev_queues(struct net_device
*dev
)
7162 unsigned int count
= dev
->num_tx_queues
;
7163 struct netdev_queue
*tx
;
7164 size_t sz
= count
* sizeof(*tx
);
7166 if (count
< 1 || count
> 0xffff)
7169 tx
= kzalloc(sz
, GFP_KERNEL
| __GFP_NOWARN
| __GFP_REPEAT
);
7177 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
7178 spin_lock_init(&dev
->tx_global_lock
);
7183 void netif_tx_stop_all_queues(struct net_device
*dev
)
7187 for (i
= 0; i
< dev
->num_tx_queues
; i
++) {
7188 struct netdev_queue
*txq
= netdev_get_tx_queue(dev
, i
);
7190 netif_tx_stop_queue(txq
);
7193 EXPORT_SYMBOL(netif_tx_stop_all_queues
);
7196 * register_netdevice - register a network device
7197 * @dev: device to register
7199 * Take a completed network device structure and add it to the kernel
7200 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
7201 * chain. 0 is returned on success. A negative errno code is returned
7202 * on a failure to set up the device, or if the name is a duplicate.
7204 * Callers must hold the rtnl semaphore. You may want
7205 * register_netdev() instead of this.
7208 * The locking appears insufficient to guarantee two parallel registers
7209 * will not get the same name.
7212 int register_netdevice(struct net_device
*dev
)
7215 struct net
*net
= dev_net(dev
);
7217 BUG_ON(dev_boot_phase
);
7222 /* When net_device's are persistent, this will be fatal. */
7223 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
7226 spin_lock_init(&dev
->addr_list_lock
);
7227 netdev_set_addr_lockdep_class(dev
);
7229 ret
= dev_get_valid_name(net
, dev
, dev
->name
);
7233 /* Init, if this function is available */
7234 if (dev
->netdev_ops
->ndo_init
) {
7235 ret
= dev
->netdev_ops
->ndo_init(dev
);
7243 if (((dev
->hw_features
| dev
->features
) &
7244 NETIF_F_HW_VLAN_CTAG_FILTER
) &&
7245 (!dev
->netdev_ops
->ndo_vlan_rx_add_vid
||
7246 !dev
->netdev_ops
->ndo_vlan_rx_kill_vid
)) {
7247 netdev_WARN(dev
, "Buggy VLAN acceleration in driver!\n");
7254 dev
->ifindex
= dev_new_index(net
);
7255 else if (__dev_get_by_index(net
, dev
->ifindex
))
7258 /* Transfer changeable features to wanted_features and enable
7259 * software offloads (GSO and GRO).
7261 dev
->hw_features
|= NETIF_F_SOFT_FEATURES
;
7262 dev
->features
|= NETIF_F_SOFT_FEATURES
;
7263 dev
->wanted_features
= dev
->features
& dev
->hw_features
;
7265 if (!(dev
->flags
& IFF_LOOPBACK
))
7266 dev
->hw_features
|= NETIF_F_NOCACHE_COPY
;
7268 /* If IPv4 TCP segmentation offload is supported we should also
7269 * allow the device to enable segmenting the frame with the option
7270 * of ignoring a static IP ID value. This doesn't enable the
7271 * feature itself but allows the user to enable it later.
7273 if (dev
->hw_features
& NETIF_F_TSO
)
7274 dev
->hw_features
|= NETIF_F_TSO_MANGLEID
;
7275 if (dev
->vlan_features
& NETIF_F_TSO
)
7276 dev
->vlan_features
|= NETIF_F_TSO_MANGLEID
;
7277 if (dev
->mpls_features
& NETIF_F_TSO
)
7278 dev
->mpls_features
|= NETIF_F_TSO_MANGLEID
;
7279 if (dev
->hw_enc_features
& NETIF_F_TSO
)
7280 dev
->hw_enc_features
|= NETIF_F_TSO_MANGLEID
;
7282 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
7284 dev
->vlan_features
|= NETIF_F_HIGHDMA
;
7286 /* Make NETIF_F_SG inheritable to tunnel devices.
7288 dev
->hw_enc_features
|= NETIF_F_SG
| NETIF_F_GSO_PARTIAL
;
7290 /* Make NETIF_F_SG inheritable to MPLS.
7292 dev
->mpls_features
|= NETIF_F_SG
;
7294 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
7295 ret
= notifier_to_errno(ret
);
7299 ret
= netdev_register_kobject(dev
);
7302 dev
->reg_state
= NETREG_REGISTERED
;
7304 __netdev_update_features(dev
);
7307 * Default initial state at registry is that the
7308 * device is present.
7311 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
7313 linkwatch_init_dev(dev
);
7315 dev_init_scheduler(dev
);
7317 list_netdevice(dev
);
7318 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
7320 /* If the device has permanent device address, driver should
7321 * set dev_addr and also addr_assign_type should be set to
7322 * NET_ADDR_PERM (default value).
7324 if (dev
->addr_assign_type
== NET_ADDR_PERM
)
7325 memcpy(dev
->perm_addr
, dev
->dev_addr
, dev
->addr_len
);
7327 /* Notify protocols, that a new device appeared. */
7328 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
7329 ret
= notifier_to_errno(ret
);
7331 rollback_registered(dev
);
7332 dev
->reg_state
= NETREG_UNREGISTERED
;
7335 * Prevent userspace races by waiting until the network
7336 * device is fully setup before sending notifications.
7338 if (!dev
->rtnl_link_ops
||
7339 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
7340 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U, GFP_KERNEL
);
7346 if (dev
->netdev_ops
->ndo_uninit
)
7347 dev
->netdev_ops
->ndo_uninit(dev
);
7350 EXPORT_SYMBOL(register_netdevice
);
7353 * init_dummy_netdev - init a dummy network device for NAPI
7354 * @dev: device to init
7356 * This takes a network device structure and initialize the minimum
7357 * amount of fields so it can be used to schedule NAPI polls without
7358 * registering a full blown interface. This is to be used by drivers
7359 * that need to tie several hardware interfaces to a single NAPI
7360 * poll scheduler due to HW limitations.
7362 int init_dummy_netdev(struct net_device
*dev
)
7364 /* Clear everything. Note we don't initialize spinlocks
7365 * are they aren't supposed to be taken by any of the
7366 * NAPI code and this dummy netdev is supposed to be
7367 * only ever used for NAPI polls
7369 memset(dev
, 0, sizeof(struct net_device
));
7371 /* make sure we BUG if trying to hit standard
7372 * register/unregister code path
7374 dev
->reg_state
= NETREG_DUMMY
;
7376 /* NAPI wants this */
7377 INIT_LIST_HEAD(&dev
->napi_list
);
7379 /* a dummy interface is started by default */
7380 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
7381 set_bit(__LINK_STATE_START
, &dev
->state
);
7383 /* Note : We dont allocate pcpu_refcnt for dummy devices,
7384 * because users of this 'device' dont need to change
7390 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
7394 * register_netdev - register a network device
7395 * @dev: device to register
7397 * Take a completed network device structure and add it to the kernel
7398 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
7399 * chain. 0 is returned on success. A negative errno code is returned
7400 * on a failure to set up the device, or if the name is a duplicate.
7402 * This is a wrapper around register_netdevice that takes the rtnl semaphore
7403 * and expands the device name if you passed a format string to
7406 int register_netdev(struct net_device
*dev
)
7411 err
= register_netdevice(dev
);
7415 EXPORT_SYMBOL(register_netdev
);
7417 int netdev_refcnt_read(const struct net_device
*dev
)
7421 for_each_possible_cpu(i
)
7422 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
7425 EXPORT_SYMBOL(netdev_refcnt_read
);
7428 * netdev_wait_allrefs - wait until all references are gone.
7429 * @dev: target net_device
7431 * This is called when unregistering network devices.
7433 * Any protocol or device that holds a reference should register
7434 * for netdevice notification, and cleanup and put back the
7435 * reference if they receive an UNREGISTER event.
7436 * We can get stuck here if buggy protocols don't correctly
7439 static void netdev_wait_allrefs(struct net_device
*dev
)
7441 unsigned long rebroadcast_time
, warning_time
;
7444 linkwatch_forget_dev(dev
);
7446 rebroadcast_time
= warning_time
= jiffies
;
7447 refcnt
= netdev_refcnt_read(dev
);
7449 while (refcnt
!= 0) {
7450 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
7453 /* Rebroadcast unregister notification */
7454 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
7460 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
7461 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
7463 /* We must not have linkwatch events
7464 * pending on unregister. If this
7465 * happens, we simply run the queue
7466 * unscheduled, resulting in a noop
7469 linkwatch_run_queue();
7474 rebroadcast_time
= jiffies
;
7479 refcnt
= netdev_refcnt_read(dev
);
7481 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
7482 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
7484 warning_time
= jiffies
;
7493 * register_netdevice(x1);
7494 * register_netdevice(x2);
7496 * unregister_netdevice(y1);
7497 * unregister_netdevice(y2);
7503 * We are invoked by rtnl_unlock().
7504 * This allows us to deal with problems:
7505 * 1) We can delete sysfs objects which invoke hotplug
7506 * without deadlocking with linkwatch via keventd.
7507 * 2) Since we run with the RTNL semaphore not held, we can sleep
7508 * safely in order to wait for the netdev refcnt to drop to zero.
7510 * We must not return until all unregister events added during
7511 * the interval the lock was held have been completed.
7513 void netdev_run_todo(void)
7515 struct list_head list
;
7517 /* Snapshot list, allow later requests */
7518 list_replace_init(&net_todo_list
, &list
);
7523 /* Wait for rcu callbacks to finish before next phase */
7524 if (!list_empty(&list
))
7527 while (!list_empty(&list
)) {
7528 struct net_device
*dev
7529 = list_first_entry(&list
, struct net_device
, todo_list
);
7530 list_del(&dev
->todo_list
);
7533 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
7536 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
7537 pr_err("network todo '%s' but state %d\n",
7538 dev
->name
, dev
->reg_state
);
7543 dev
->reg_state
= NETREG_UNREGISTERED
;
7545 netdev_wait_allrefs(dev
);
7548 BUG_ON(netdev_refcnt_read(dev
));
7549 BUG_ON(!list_empty(&dev
->ptype_all
));
7550 BUG_ON(!list_empty(&dev
->ptype_specific
));
7551 WARN_ON(rcu_access_pointer(dev
->ip_ptr
));
7552 WARN_ON(rcu_access_pointer(dev
->ip6_ptr
));
7553 WARN_ON(dev
->dn_ptr
);
7555 if (dev
->destructor
)
7556 dev
->destructor(dev
);
7558 /* Report a network device has been unregistered */
7560 dev_net(dev
)->dev_unreg_count
--;
7562 wake_up(&netdev_unregistering_wq
);
7564 /* Free network device */
7565 kobject_put(&dev
->dev
.kobj
);
7569 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
7570 * all the same fields in the same order as net_device_stats, with only
7571 * the type differing, but rtnl_link_stats64 may have additional fields
7572 * at the end for newer counters.
7574 void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
7575 const struct net_device_stats
*netdev_stats
)
7577 #if BITS_PER_LONG == 64
7578 BUILD_BUG_ON(sizeof(*stats64
) < sizeof(*netdev_stats
));
7579 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
7580 /* zero out counters that only exist in rtnl_link_stats64 */
7581 memset((char *)stats64
+ sizeof(*netdev_stats
), 0,
7582 sizeof(*stats64
) - sizeof(*netdev_stats
));
7584 size_t i
, n
= sizeof(*netdev_stats
) / sizeof(unsigned long);
7585 const unsigned long *src
= (const unsigned long *)netdev_stats
;
7586 u64
*dst
= (u64
*)stats64
;
7588 BUILD_BUG_ON(n
> sizeof(*stats64
) / sizeof(u64
));
7589 for (i
= 0; i
< n
; i
++)
7591 /* zero out counters that only exist in rtnl_link_stats64 */
7592 memset((char *)stats64
+ n
* sizeof(u64
), 0,
7593 sizeof(*stats64
) - n
* sizeof(u64
));
7596 EXPORT_SYMBOL(netdev_stats_to_stats64
);
7599 * dev_get_stats - get network device statistics
7600 * @dev: device to get statistics from
7601 * @storage: place to store stats
7603 * Get network statistics from device. Return @storage.
7604 * The device driver may provide its own method by setting
7605 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
7606 * otherwise the internal statistics structure is used.
7608 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
7609 struct rtnl_link_stats64
*storage
)
7611 const struct net_device_ops
*ops
= dev
->netdev_ops
;
7613 if (ops
->ndo_get_stats64
) {
7614 memset(storage
, 0, sizeof(*storage
));
7615 ops
->ndo_get_stats64(dev
, storage
);
7616 } else if (ops
->ndo_get_stats
) {
7617 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
7619 netdev_stats_to_stats64(storage
, &dev
->stats
);
7621 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
7622 storage
->tx_dropped
+= atomic_long_read(&dev
->tx_dropped
);
7623 storage
->rx_nohandler
+= atomic_long_read(&dev
->rx_nohandler
);
7626 EXPORT_SYMBOL(dev_get_stats
);
7628 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
7630 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
7632 #ifdef CONFIG_NET_CLS_ACT
7635 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
7638 netdev_init_one_queue(dev
, queue
, NULL
);
7639 RCU_INIT_POINTER(queue
->qdisc
, &noop_qdisc
);
7640 queue
->qdisc_sleeping
= &noop_qdisc
;
7641 rcu_assign_pointer(dev
->ingress_queue
, queue
);
7646 static const struct ethtool_ops default_ethtool_ops
;
7648 void netdev_set_default_ethtool_ops(struct net_device
*dev
,
7649 const struct ethtool_ops
*ops
)
7651 if (dev
->ethtool_ops
== &default_ethtool_ops
)
7652 dev
->ethtool_ops
= ops
;
7654 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops
);
7656 void netdev_freemem(struct net_device
*dev
)
7658 char *addr
= (char *)dev
- dev
->padded
;
7664 * alloc_netdev_mqs - allocate network device
7665 * @sizeof_priv: size of private data to allocate space for
7666 * @name: device name format string
7667 * @name_assign_type: origin of device name
7668 * @setup: callback to initialize device
7669 * @txqs: the number of TX subqueues to allocate
7670 * @rxqs: the number of RX subqueues to allocate
7672 * Allocates a struct net_device with private data area for driver use
7673 * and performs basic initialization. Also allocates subqueue structs
7674 * for each queue on the device.
7676 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
7677 unsigned char name_assign_type
,
7678 void (*setup
)(struct net_device
*),
7679 unsigned int txqs
, unsigned int rxqs
)
7681 struct net_device
*dev
;
7683 struct net_device
*p
;
7685 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
7688 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
7694 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
7699 alloc_size
= sizeof(struct net_device
);
7701 /* ensure 32-byte alignment of private area */
7702 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
7703 alloc_size
+= sizeof_priv
;
7705 /* ensure 32-byte alignment of whole construct */
7706 alloc_size
+= NETDEV_ALIGN
- 1;
7708 p
= kzalloc(alloc_size
, GFP_KERNEL
| __GFP_NOWARN
| __GFP_REPEAT
);
7710 p
= vzalloc(alloc_size
);
7714 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
7715 dev
->padded
= (char *)dev
- (char *)p
;
7717 dev
->pcpu_refcnt
= alloc_percpu(int);
7718 if (!dev
->pcpu_refcnt
)
7721 if (dev_addr_init(dev
))
7727 dev_net_set(dev
, &init_net
);
7729 dev
->gso_max_size
= GSO_MAX_SIZE
;
7730 dev
->gso_max_segs
= GSO_MAX_SEGS
;
7732 INIT_LIST_HEAD(&dev
->napi_list
);
7733 INIT_LIST_HEAD(&dev
->unreg_list
);
7734 INIT_LIST_HEAD(&dev
->close_list
);
7735 INIT_LIST_HEAD(&dev
->link_watch_list
);
7736 INIT_LIST_HEAD(&dev
->adj_list
.upper
);
7737 INIT_LIST_HEAD(&dev
->adj_list
.lower
);
7738 INIT_LIST_HEAD(&dev
->ptype_all
);
7739 INIT_LIST_HEAD(&dev
->ptype_specific
);
7740 #ifdef CONFIG_NET_SCHED
7741 hash_init(dev
->qdisc_hash
);
7743 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
| IFF_XMIT_DST_RELEASE_PERM
;
7746 if (!dev
->tx_queue_len
) {
7747 dev
->priv_flags
|= IFF_NO_QUEUE
;
7748 dev
->tx_queue_len
= DEFAULT_TX_QUEUE_LEN
;
7751 dev
->num_tx_queues
= txqs
;
7752 dev
->real_num_tx_queues
= txqs
;
7753 if (netif_alloc_netdev_queues(dev
))
7757 dev
->num_rx_queues
= rxqs
;
7758 dev
->real_num_rx_queues
= rxqs
;
7759 if (netif_alloc_rx_queues(dev
))
7763 strcpy(dev
->name
, name
);
7764 dev
->name_assign_type
= name_assign_type
;
7765 dev
->group
= INIT_NETDEV_GROUP
;
7766 if (!dev
->ethtool_ops
)
7767 dev
->ethtool_ops
= &default_ethtool_ops
;
7769 nf_hook_ingress_init(dev
);
7778 free_percpu(dev
->pcpu_refcnt
);
7780 netdev_freemem(dev
);
7783 EXPORT_SYMBOL(alloc_netdev_mqs
);
7786 * free_netdev - free network device
7789 * This function does the last stage of destroying an allocated device
7790 * interface. The reference to the device object is released. If this
7791 * is the last reference then it will be freed.Must be called in process
7794 void free_netdev(struct net_device
*dev
)
7796 struct napi_struct
*p
, *n
;
7799 netif_free_tx_queues(dev
);
7804 kfree(rcu_dereference_protected(dev
->ingress_queue
, 1));
7806 /* Flush device addresses */
7807 dev_addr_flush(dev
);
7809 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
7812 free_percpu(dev
->pcpu_refcnt
);
7813 dev
->pcpu_refcnt
= NULL
;
7815 /* Compatibility with error handling in drivers */
7816 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
7817 netdev_freemem(dev
);
7821 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
7822 dev
->reg_state
= NETREG_RELEASED
;
7824 /* will free via device release */
7825 put_device(&dev
->dev
);
7827 EXPORT_SYMBOL(free_netdev
);
7830 * synchronize_net - Synchronize with packet receive processing
7832 * Wait for packets currently being received to be done.
7833 * Does not block later packets from starting.
7835 void synchronize_net(void)
7838 if (rtnl_is_locked())
7839 synchronize_rcu_expedited();
7843 EXPORT_SYMBOL(synchronize_net
);
7846 * unregister_netdevice_queue - remove device from the kernel
7850 * This function shuts down a device interface and removes it
7851 * from the kernel tables.
7852 * If head not NULL, device is queued to be unregistered later.
7854 * Callers must hold the rtnl semaphore. You may want
7855 * unregister_netdev() instead of this.
7858 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
7863 list_move_tail(&dev
->unreg_list
, head
);
7865 rollback_registered(dev
);
7866 /* Finish processing unregister after unlock */
7870 EXPORT_SYMBOL(unregister_netdevice_queue
);
7873 * unregister_netdevice_many - unregister many devices
7874 * @head: list of devices
7876 * Note: As most callers use a stack allocated list_head,
7877 * we force a list_del() to make sure stack wont be corrupted later.
7879 void unregister_netdevice_many(struct list_head
*head
)
7881 struct net_device
*dev
;
7883 if (!list_empty(head
)) {
7884 rollback_registered_many(head
);
7885 list_for_each_entry(dev
, head
, unreg_list
)
7890 EXPORT_SYMBOL(unregister_netdevice_many
);
7893 * unregister_netdev - remove device from the kernel
7896 * This function shuts down a device interface and removes it
7897 * from the kernel tables.
7899 * This is just a wrapper for unregister_netdevice that takes
7900 * the rtnl semaphore. In general you want to use this and not
7901 * unregister_netdevice.
7903 void unregister_netdev(struct net_device
*dev
)
7906 unregister_netdevice(dev
);
7909 EXPORT_SYMBOL(unregister_netdev
);
7912 * dev_change_net_namespace - move device to different nethost namespace
7914 * @net: network namespace
7915 * @pat: If not NULL name pattern to try if the current device name
7916 * is already taken in the destination network namespace.
7918 * This function shuts down a device interface and moves it
7919 * to a new network namespace. On success 0 is returned, on
7920 * a failure a netagive errno code is returned.
7922 * Callers must hold the rtnl semaphore.
7925 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
7931 /* Don't allow namespace local devices to be moved. */
7933 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
7936 /* Ensure the device has been registrered */
7937 if (dev
->reg_state
!= NETREG_REGISTERED
)
7940 /* Get out if there is nothing todo */
7942 if (net_eq(dev_net(dev
), net
))
7945 /* Pick the destination device name, and ensure
7946 * we can use it in the destination network namespace.
7949 if (__dev_get_by_name(net
, dev
->name
)) {
7950 /* We get here if we can't use the current device name */
7953 if (dev_get_valid_name(net
, dev
, pat
) < 0)
7958 * And now a mini version of register_netdevice unregister_netdevice.
7961 /* If device is running close it first. */
7964 /* And unlink it from device chain */
7966 unlist_netdevice(dev
);
7970 /* Shutdown queueing discipline. */
7973 /* Notify protocols, that we are about to destroy
7974 * this device. They should clean all the things.
7976 * Note that dev->reg_state stays at NETREG_REGISTERED.
7977 * This is wanted because this way 8021q and macvlan know
7978 * the device is just moving and can keep their slaves up.
7980 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
7982 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
7983 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U, GFP_KERNEL
);
7986 * Flush the unicast and multicast chains
7991 /* Send a netdev-removed uevent to the old namespace */
7992 kobject_uevent(&dev
->dev
.kobj
, KOBJ_REMOVE
);
7993 netdev_adjacent_del_links(dev
);
7995 /* Actually switch the network namespace */
7996 dev_net_set(dev
, net
);
7998 /* If there is an ifindex conflict assign a new one */
7999 if (__dev_get_by_index(net
, dev
->ifindex
))
8000 dev
->ifindex
= dev_new_index(net
);
8002 /* Send a netdev-add uevent to the new namespace */
8003 kobject_uevent(&dev
->dev
.kobj
, KOBJ_ADD
);
8004 netdev_adjacent_add_links(dev
);
8006 /* Fixup kobjects */
8007 err
= device_rename(&dev
->dev
, dev
->name
);
8010 /* Add the device back in the hashes */
8011 list_netdevice(dev
);
8013 /* Notify protocols, that a new device appeared. */
8014 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
8017 * Prevent userspace races by waiting until the network
8018 * device is fully setup before sending notifications.
8020 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U, GFP_KERNEL
);
8027 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
8029 static int dev_cpu_dead(unsigned int oldcpu
)
8031 struct sk_buff
**list_skb
;
8032 struct sk_buff
*skb
;
8034 struct softnet_data
*sd
, *oldsd
;
8036 local_irq_disable();
8037 cpu
= smp_processor_id();
8038 sd
= &per_cpu(softnet_data
, cpu
);
8039 oldsd
= &per_cpu(softnet_data
, oldcpu
);
8041 /* Find end of our completion_queue. */
8042 list_skb
= &sd
->completion_queue
;
8044 list_skb
= &(*list_skb
)->next
;
8045 /* Append completion queue from offline CPU. */
8046 *list_skb
= oldsd
->completion_queue
;
8047 oldsd
->completion_queue
= NULL
;
8049 /* Append output queue from offline CPU. */
8050 if (oldsd
->output_queue
) {
8051 *sd
->output_queue_tailp
= oldsd
->output_queue
;
8052 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
8053 oldsd
->output_queue
= NULL
;
8054 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
8056 /* Append NAPI poll list from offline CPU, with one exception :
8057 * process_backlog() must be called by cpu owning percpu backlog.
8058 * We properly handle process_queue & input_pkt_queue later.
8060 while (!list_empty(&oldsd
->poll_list
)) {
8061 struct napi_struct
*napi
= list_first_entry(&oldsd
->poll_list
,
8065 list_del_init(&napi
->poll_list
);
8066 if (napi
->poll
== process_backlog
)
8069 ____napi_schedule(sd
, napi
);
8072 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
8075 /* Process offline CPU's input_pkt_queue */
8076 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
8078 input_queue_head_incr(oldsd
);
8080 while ((skb
= skb_dequeue(&oldsd
->input_pkt_queue
))) {
8082 input_queue_head_incr(oldsd
);
8089 * netdev_increment_features - increment feature set by one
8090 * @all: current feature set
8091 * @one: new feature set
8092 * @mask: mask feature set
8094 * Computes a new feature set after adding a device with feature set
8095 * @one to the master device with current feature set @all. Will not
8096 * enable anything that is off in @mask. Returns the new feature set.
8098 netdev_features_t
netdev_increment_features(netdev_features_t all
,
8099 netdev_features_t one
, netdev_features_t mask
)
8101 if (mask
& NETIF_F_HW_CSUM
)
8102 mask
|= NETIF_F_CSUM_MASK
;
8103 mask
|= NETIF_F_VLAN_CHALLENGED
;
8105 all
|= one
& (NETIF_F_ONE_FOR_ALL
| NETIF_F_CSUM_MASK
) & mask
;
8106 all
&= one
| ~NETIF_F_ALL_FOR_ALL
;
8108 /* If one device supports hw checksumming, set for all. */
8109 if (all
& NETIF_F_HW_CSUM
)
8110 all
&= ~(NETIF_F_CSUM_MASK
& ~NETIF_F_HW_CSUM
);
8114 EXPORT_SYMBOL(netdev_increment_features
);
8116 static struct hlist_head
* __net_init
netdev_create_hash(void)
8119 struct hlist_head
*hash
;
8121 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
8123 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
8124 INIT_HLIST_HEAD(&hash
[i
]);
8129 /* Initialize per network namespace state */
8130 static int __net_init
netdev_init(struct net
*net
)
8132 if (net
!= &init_net
)
8133 INIT_LIST_HEAD(&net
->dev_base_head
);
8135 net
->dev_name_head
= netdev_create_hash();
8136 if (net
->dev_name_head
== NULL
)
8139 net
->dev_index_head
= netdev_create_hash();
8140 if (net
->dev_index_head
== NULL
)
8146 kfree(net
->dev_name_head
);
8152 * netdev_drivername - network driver for the device
8153 * @dev: network device
8155 * Determine network driver for device.
8157 const char *netdev_drivername(const struct net_device
*dev
)
8159 const struct device_driver
*driver
;
8160 const struct device
*parent
;
8161 const char *empty
= "";
8163 parent
= dev
->dev
.parent
;
8167 driver
= parent
->driver
;
8168 if (driver
&& driver
->name
)
8169 return driver
->name
;
8173 static void __netdev_printk(const char *level
, const struct net_device
*dev
,
8174 struct va_format
*vaf
)
8176 if (dev
&& dev
->dev
.parent
) {
8177 dev_printk_emit(level
[1] - '0',
8180 dev_driver_string(dev
->dev
.parent
),
8181 dev_name(dev
->dev
.parent
),
8182 netdev_name(dev
), netdev_reg_state(dev
),
8185 printk("%s%s%s: %pV",
8186 level
, netdev_name(dev
), netdev_reg_state(dev
), vaf
);
8188 printk("%s(NULL net_device): %pV", level
, vaf
);
8192 void netdev_printk(const char *level
, const struct net_device
*dev
,
8193 const char *format
, ...)
8195 struct va_format vaf
;
8198 va_start(args
, format
);
8203 __netdev_printk(level
, dev
, &vaf
);
8207 EXPORT_SYMBOL(netdev_printk
);
8209 #define define_netdev_printk_level(func, level) \
8210 void func(const struct net_device *dev, const char *fmt, ...) \
8212 struct va_format vaf; \
8215 va_start(args, fmt); \
8220 __netdev_printk(level, dev, &vaf); \
8224 EXPORT_SYMBOL(func);
8226 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
8227 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
8228 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
8229 define_netdev_printk_level(netdev_err
, KERN_ERR
);
8230 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
8231 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
8232 define_netdev_printk_level(netdev_info
, KERN_INFO
);
8234 static void __net_exit
netdev_exit(struct net
*net
)
8236 kfree(net
->dev_name_head
);
8237 kfree(net
->dev_index_head
);
8240 static struct pernet_operations __net_initdata netdev_net_ops
= {
8241 .init
= netdev_init
,
8242 .exit
= netdev_exit
,
8245 static void __net_exit
default_device_exit(struct net
*net
)
8247 struct net_device
*dev
, *aux
;
8249 * Push all migratable network devices back to the
8250 * initial network namespace
8253 for_each_netdev_safe(net
, dev
, aux
) {
8255 char fb_name
[IFNAMSIZ
];
8257 /* Ignore unmoveable devices (i.e. loopback) */
8258 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
8261 /* Leave virtual devices for the generic cleanup */
8262 if (dev
->rtnl_link_ops
)
8265 /* Push remaining network devices to init_net */
8266 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
8267 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
8269 pr_emerg("%s: failed to move %s to init_net: %d\n",
8270 __func__
, dev
->name
, err
);
8277 static void __net_exit
rtnl_lock_unregistering(struct list_head
*net_list
)
8279 /* Return with the rtnl_lock held when there are no network
8280 * devices unregistering in any network namespace in net_list.
8284 DEFINE_WAIT_FUNC(wait
, woken_wake_function
);
8286 add_wait_queue(&netdev_unregistering_wq
, &wait
);
8288 unregistering
= false;
8290 list_for_each_entry(net
, net_list
, exit_list
) {
8291 if (net
->dev_unreg_count
> 0) {
8292 unregistering
= true;
8300 wait_woken(&wait
, TASK_UNINTERRUPTIBLE
, MAX_SCHEDULE_TIMEOUT
);
8302 remove_wait_queue(&netdev_unregistering_wq
, &wait
);
8305 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
8307 /* At exit all network devices most be removed from a network
8308 * namespace. Do this in the reverse order of registration.
8309 * Do this across as many network namespaces as possible to
8310 * improve batching efficiency.
8312 struct net_device
*dev
;
8314 LIST_HEAD(dev_kill_list
);
8316 /* To prevent network device cleanup code from dereferencing
8317 * loopback devices or network devices that have been freed
8318 * wait here for all pending unregistrations to complete,
8319 * before unregistring the loopback device and allowing the
8320 * network namespace be freed.
8322 * The netdev todo list containing all network devices
8323 * unregistrations that happen in default_device_exit_batch
8324 * will run in the rtnl_unlock() at the end of
8325 * default_device_exit_batch.
8327 rtnl_lock_unregistering(net_list
);
8328 list_for_each_entry(net
, net_list
, exit_list
) {
8329 for_each_netdev_reverse(net
, dev
) {
8330 if (dev
->rtnl_link_ops
&& dev
->rtnl_link_ops
->dellink
)
8331 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
8333 unregister_netdevice_queue(dev
, &dev_kill_list
);
8336 unregister_netdevice_many(&dev_kill_list
);
8340 static struct pernet_operations __net_initdata default_device_ops
= {
8341 .exit
= default_device_exit
,
8342 .exit_batch
= default_device_exit_batch
,
8346 * Initialize the DEV module. At boot time this walks the device list and
8347 * unhooks any devices that fail to initialise (normally hardware not
8348 * present) and leaves us with a valid list of present and active devices.
8353 * This is called single threaded during boot, so no need
8354 * to take the rtnl semaphore.
8356 static int __init
net_dev_init(void)
8358 int i
, rc
= -ENOMEM
;
8360 BUG_ON(!dev_boot_phase
);
8362 if (dev_proc_init())
8365 if (netdev_kobject_init())
8368 INIT_LIST_HEAD(&ptype_all
);
8369 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
8370 INIT_LIST_HEAD(&ptype_base
[i
]);
8372 INIT_LIST_HEAD(&offload_base
);
8374 if (register_pernet_subsys(&netdev_net_ops
))
8378 * Initialise the packet receive queues.
8381 for_each_possible_cpu(i
) {
8382 struct work_struct
*flush
= per_cpu_ptr(&flush_works
, i
);
8383 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
8385 INIT_WORK(flush
, flush_backlog
);
8387 skb_queue_head_init(&sd
->input_pkt_queue
);
8388 skb_queue_head_init(&sd
->process_queue
);
8389 INIT_LIST_HEAD(&sd
->poll_list
);
8390 sd
->output_queue_tailp
= &sd
->output_queue
;
8392 sd
->csd
.func
= rps_trigger_softirq
;
8397 sd
->backlog
.poll
= process_backlog
;
8398 sd
->backlog
.weight
= weight_p
;
8403 /* The loopback device is special if any other network devices
8404 * is present in a network namespace the loopback device must
8405 * be present. Since we now dynamically allocate and free the
8406 * loopback device ensure this invariant is maintained by
8407 * keeping the loopback device as the first device on the
8408 * list of network devices. Ensuring the loopback devices
8409 * is the first device that appears and the last network device
8412 if (register_pernet_device(&loopback_net_ops
))
8415 if (register_pernet_device(&default_device_ops
))
8418 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
8419 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
8421 rc
= cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD
, "net/dev:dead",
8422 NULL
, dev_cpu_dead
);
8430 subsys_initcall(net_dev_init
);