2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <net/net_namespace.h>
99 #include <linux/rtnetlink.h>
100 #include <linux/stat.h>
102 #include <net/pkt_sched.h>
103 #include <net/checksum.h>
104 #include <net/xfrm.h>
105 #include <linux/highmem.h>
106 #include <linux/init.h>
107 #include <linux/module.h>
108 #include <linux/netpoll.h>
109 #include <linux/rcupdate.h>
110 #include <linux/delay.h>
111 #include <net/iw_handler.h>
112 #include <asm/current.h>
113 #include <linux/audit.h>
114 #include <linux/dmaengine.h>
115 #include <linux/err.h>
116 #include <linux/ctype.h>
117 #include <linux/if_arp.h>
118 #include <linux/if_vlan.h>
119 #include <linux/ip.h>
121 #include <linux/ipv6.h>
122 #include <linux/in.h>
123 #include <linux/jhash.h>
124 #include <linux/random.h>
125 #include <trace/events/napi.h>
126 #include <trace/events/net.h>
127 #include <trace/events/skb.h>
128 #include <linux/pci.h>
129 #include <linux/inetdevice.h>
130 #include <linux/cpu_rmap.h>
131 #include <linux/static_key.h>
132 #include <linux/hashtable.h>
133 #include <linux/vmalloc.h>
134 #include <linux/if_macvlan.h>
136 #include "net-sysfs.h"
138 /* Instead of increasing this, you should create a hash table. */
139 #define MAX_GRO_SKBS 8
141 /* This should be increased if a protocol with a bigger head is added. */
142 #define GRO_MAX_HEAD (MAX_HEADER + 128)
144 static DEFINE_SPINLOCK(ptype_lock
);
145 static DEFINE_SPINLOCK(offload_lock
);
146 struct list_head ptype_base
[PTYPE_HASH_SIZE
] __read_mostly
;
147 struct list_head ptype_all __read_mostly
; /* Taps */
148 static struct list_head offload_base __read_mostly
;
151 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
154 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
156 * Writers must hold the rtnl semaphore while they loop through the
157 * dev_base_head list, and hold dev_base_lock for writing when they do the
158 * actual updates. This allows pure readers to access the list even
159 * while a writer is preparing to update it.
161 * To put it another way, dev_base_lock is held for writing only to
162 * protect against pure readers; the rtnl semaphore provides the
163 * protection against other writers.
165 * See, for example usages, register_netdevice() and
166 * unregister_netdevice(), which must be called with the rtnl
169 DEFINE_RWLOCK(dev_base_lock
);
170 EXPORT_SYMBOL(dev_base_lock
);
172 /* protects napi_hash addition/deletion and napi_gen_id */
173 static DEFINE_SPINLOCK(napi_hash_lock
);
175 static unsigned int napi_gen_id
;
176 static DEFINE_HASHTABLE(napi_hash
, 8);
178 static seqcount_t devnet_rename_seq
;
180 static inline void dev_base_seq_inc(struct net
*net
)
182 while (++net
->dev_base_seq
== 0);
185 static inline struct hlist_head
*dev_name_hash(struct net
*net
, const char *name
)
187 unsigned int hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
189 return &net
->dev_name_head
[hash_32(hash
, NETDEV_HASHBITS
)];
192 static inline struct hlist_head
*dev_index_hash(struct net
*net
, int ifindex
)
194 return &net
->dev_index_head
[ifindex
& (NETDEV_HASHENTRIES
- 1)];
197 static inline void rps_lock(struct softnet_data
*sd
)
200 spin_lock(&sd
->input_pkt_queue
.lock
);
204 static inline void rps_unlock(struct softnet_data
*sd
)
207 spin_unlock(&sd
->input_pkt_queue
.lock
);
211 /* Device list insertion */
212 static void list_netdevice(struct net_device
*dev
)
214 struct net
*net
= dev_net(dev
);
218 write_lock_bh(&dev_base_lock
);
219 list_add_tail_rcu(&dev
->dev_list
, &net
->dev_base_head
);
220 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
221 hlist_add_head_rcu(&dev
->index_hlist
,
222 dev_index_hash(net
, dev
->ifindex
));
223 write_unlock_bh(&dev_base_lock
);
225 dev_base_seq_inc(net
);
228 /* Device list removal
229 * caller must respect a RCU grace period before freeing/reusing dev
231 static void unlist_netdevice(struct net_device
*dev
)
235 /* Unlink dev from the device chain */
236 write_lock_bh(&dev_base_lock
);
237 list_del_rcu(&dev
->dev_list
);
238 hlist_del_rcu(&dev
->name_hlist
);
239 hlist_del_rcu(&dev
->index_hlist
);
240 write_unlock_bh(&dev_base_lock
);
242 dev_base_seq_inc(dev_net(dev
));
249 static RAW_NOTIFIER_HEAD(netdev_chain
);
252 * Device drivers call our routines to queue packets here. We empty the
253 * queue in the local softnet handler.
256 DEFINE_PER_CPU_ALIGNED(struct softnet_data
, softnet_data
);
257 EXPORT_PER_CPU_SYMBOL(softnet_data
);
259 #ifdef CONFIG_LOCKDEP
261 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
262 * according to dev->type
264 static const unsigned short netdev_lock_type
[] =
265 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
266 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
267 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
268 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
269 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
270 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
271 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
272 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
273 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
274 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
275 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
276 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
277 ARPHRD_FCFABRIC
, ARPHRD_IEEE80211
, ARPHRD_IEEE80211_PRISM
,
278 ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_PHONET
, ARPHRD_PHONET_PIPE
,
279 ARPHRD_IEEE802154
, ARPHRD_VOID
, ARPHRD_NONE
};
281 static const char *const netdev_lock_name
[] =
282 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
283 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
284 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
285 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
286 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
287 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
288 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
289 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
290 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
291 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
292 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
293 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
294 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
295 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
296 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
298 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
299 static struct lock_class_key netdev_addr_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
301 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
305 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
306 if (netdev_lock_type
[i
] == dev_type
)
308 /* the last key is used by default */
309 return ARRAY_SIZE(netdev_lock_type
) - 1;
312 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
313 unsigned short dev_type
)
317 i
= netdev_lock_pos(dev_type
);
318 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
319 netdev_lock_name
[i
]);
322 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
326 i
= netdev_lock_pos(dev
->type
);
327 lockdep_set_class_and_name(&dev
->addr_list_lock
,
328 &netdev_addr_lock_key
[i
],
329 netdev_lock_name
[i
]);
332 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
333 unsigned short dev_type
)
336 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
341 /*******************************************************************************
343 Protocol management and registration routines
345 *******************************************************************************/
348 * Add a protocol ID to the list. Now that the input handler is
349 * smarter we can dispense with all the messy stuff that used to be
352 * BEWARE!!! Protocol handlers, mangling input packets,
353 * MUST BE last in hash buckets and checking protocol handlers
354 * MUST start from promiscuous ptype_all chain in net_bh.
355 * It is true now, do not change it.
356 * Explanation follows: if protocol handler, mangling packet, will
357 * be the first on list, it is not able to sense, that packet
358 * is cloned and should be copied-on-write, so that it will
359 * change it and subsequent readers will get broken packet.
363 static inline struct list_head
*ptype_head(const struct packet_type
*pt
)
365 if (pt
->type
== htons(ETH_P_ALL
))
368 return &ptype_base
[ntohs(pt
->type
) & PTYPE_HASH_MASK
];
372 * dev_add_pack - add packet handler
373 * @pt: packet type declaration
375 * Add a protocol handler to the networking stack. The passed &packet_type
376 * is linked into kernel lists and may not be freed until it has been
377 * removed from the kernel lists.
379 * This call does not sleep therefore it can not
380 * guarantee all CPU's that are in middle of receiving packets
381 * will see the new packet type (until the next received packet).
384 void dev_add_pack(struct packet_type
*pt
)
386 struct list_head
*head
= ptype_head(pt
);
388 spin_lock(&ptype_lock
);
389 list_add_rcu(&pt
->list
, head
);
390 spin_unlock(&ptype_lock
);
392 EXPORT_SYMBOL(dev_add_pack
);
395 * __dev_remove_pack - remove packet handler
396 * @pt: packet type declaration
398 * Remove a protocol handler that was previously added to the kernel
399 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
400 * from the kernel lists and can be freed or reused once this function
403 * The packet type might still be in use by receivers
404 * and must not be freed until after all the CPU's have gone
405 * through a quiescent state.
407 void __dev_remove_pack(struct packet_type
*pt
)
409 struct list_head
*head
= ptype_head(pt
);
410 struct packet_type
*pt1
;
412 spin_lock(&ptype_lock
);
414 list_for_each_entry(pt1
, head
, list
) {
416 list_del_rcu(&pt
->list
);
421 pr_warn("dev_remove_pack: %p not found\n", pt
);
423 spin_unlock(&ptype_lock
);
425 EXPORT_SYMBOL(__dev_remove_pack
);
428 * dev_remove_pack - remove packet handler
429 * @pt: packet type declaration
431 * Remove a protocol handler that was previously added to the kernel
432 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
433 * from the kernel lists and can be freed or reused once this function
436 * This call sleeps to guarantee that no CPU is looking at the packet
439 void dev_remove_pack(struct packet_type
*pt
)
441 __dev_remove_pack(pt
);
445 EXPORT_SYMBOL(dev_remove_pack
);
449 * dev_add_offload - register offload handlers
450 * @po: protocol offload declaration
452 * Add protocol offload handlers to the networking stack. The passed
453 * &proto_offload is linked into kernel lists and may not be freed until
454 * it has been removed from the kernel lists.
456 * This call does not sleep therefore it can not
457 * guarantee all CPU's that are in middle of receiving packets
458 * will see the new offload handlers (until the next received packet).
460 void dev_add_offload(struct packet_offload
*po
)
462 struct list_head
*head
= &offload_base
;
464 spin_lock(&offload_lock
);
465 list_add_rcu(&po
->list
, head
);
466 spin_unlock(&offload_lock
);
468 EXPORT_SYMBOL(dev_add_offload
);
471 * __dev_remove_offload - remove offload handler
472 * @po: packet offload declaration
474 * Remove a protocol offload handler that was previously added to the
475 * kernel offload handlers by dev_add_offload(). The passed &offload_type
476 * is removed from the kernel lists and can be freed or reused once this
479 * The packet type might still be in use by receivers
480 * and must not be freed until after all the CPU's have gone
481 * through a quiescent state.
483 void __dev_remove_offload(struct packet_offload
*po
)
485 struct list_head
*head
= &offload_base
;
486 struct packet_offload
*po1
;
488 spin_lock(&offload_lock
);
490 list_for_each_entry(po1
, head
, list
) {
492 list_del_rcu(&po
->list
);
497 pr_warn("dev_remove_offload: %p not found\n", po
);
499 spin_unlock(&offload_lock
);
501 EXPORT_SYMBOL(__dev_remove_offload
);
504 * dev_remove_offload - remove packet offload handler
505 * @po: packet offload declaration
507 * Remove a packet offload handler that was previously added to the kernel
508 * offload handlers by dev_add_offload(). The passed &offload_type is
509 * removed from the kernel lists and can be freed or reused once this
512 * This call sleeps to guarantee that no CPU is looking at the packet
515 void dev_remove_offload(struct packet_offload
*po
)
517 __dev_remove_offload(po
);
521 EXPORT_SYMBOL(dev_remove_offload
);
523 /******************************************************************************
525 Device Boot-time Settings Routines
527 *******************************************************************************/
529 /* Boot time configuration table */
530 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
533 * netdev_boot_setup_add - add new setup entry
534 * @name: name of the device
535 * @map: configured settings for the device
537 * Adds new setup entry to the dev_boot_setup list. The function
538 * returns 0 on error and 1 on success. This is a generic routine to
541 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
543 struct netdev_boot_setup
*s
;
547 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
548 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
549 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
550 strlcpy(s
[i
].name
, name
, IFNAMSIZ
);
551 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
556 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
560 * netdev_boot_setup_check - check boot time settings
561 * @dev: the netdevice
563 * Check boot time settings for the device.
564 * The found settings are set for the device to be used
565 * later in the device probing.
566 * Returns 0 if no settings found, 1 if they are.
568 int netdev_boot_setup_check(struct net_device
*dev
)
570 struct netdev_boot_setup
*s
= dev_boot_setup
;
573 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
574 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
575 !strcmp(dev
->name
, s
[i
].name
)) {
576 dev
->irq
= s
[i
].map
.irq
;
577 dev
->base_addr
= s
[i
].map
.base_addr
;
578 dev
->mem_start
= s
[i
].map
.mem_start
;
579 dev
->mem_end
= s
[i
].map
.mem_end
;
585 EXPORT_SYMBOL(netdev_boot_setup_check
);
589 * netdev_boot_base - get address from boot time settings
590 * @prefix: prefix for network device
591 * @unit: id for network device
593 * Check boot time settings for the base address of device.
594 * The found settings are set for the device to be used
595 * later in the device probing.
596 * Returns 0 if no settings found.
598 unsigned long netdev_boot_base(const char *prefix
, int unit
)
600 const struct netdev_boot_setup
*s
= dev_boot_setup
;
604 sprintf(name
, "%s%d", prefix
, unit
);
607 * If device already registered then return base of 1
608 * to indicate not to probe for this interface
610 if (__dev_get_by_name(&init_net
, name
))
613 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
614 if (!strcmp(name
, s
[i
].name
))
615 return s
[i
].map
.base_addr
;
620 * Saves at boot time configured settings for any netdevice.
622 int __init
netdev_boot_setup(char *str
)
627 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
632 memset(&map
, 0, sizeof(map
));
636 map
.base_addr
= ints
[2];
638 map
.mem_start
= ints
[3];
640 map
.mem_end
= ints
[4];
642 /* Add new entry to the list */
643 return netdev_boot_setup_add(str
, &map
);
646 __setup("netdev=", netdev_boot_setup
);
648 /*******************************************************************************
650 Device Interface Subroutines
652 *******************************************************************************/
655 * __dev_get_by_name - find a device by its name
656 * @net: the applicable net namespace
657 * @name: name to find
659 * Find an interface by name. Must be called under RTNL semaphore
660 * or @dev_base_lock. If the name is found a pointer to the device
661 * is returned. If the name is not found then %NULL is returned. The
662 * reference counters are not incremented so the caller must be
663 * careful with locks.
666 struct net_device
*__dev_get_by_name(struct net
*net
, const char *name
)
668 struct net_device
*dev
;
669 struct hlist_head
*head
= dev_name_hash(net
, name
);
671 hlist_for_each_entry(dev
, head
, name_hlist
)
672 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
677 EXPORT_SYMBOL(__dev_get_by_name
);
680 * dev_get_by_name_rcu - find a device by its name
681 * @net: the applicable net namespace
682 * @name: name to find
684 * Find an interface by name.
685 * If the name is found a pointer to the device is returned.
686 * If the name is not found then %NULL is returned.
687 * The reference counters are not incremented so the caller must be
688 * careful with locks. The caller must hold RCU lock.
691 struct net_device
*dev_get_by_name_rcu(struct net
*net
, const char *name
)
693 struct net_device
*dev
;
694 struct hlist_head
*head
= dev_name_hash(net
, name
);
696 hlist_for_each_entry_rcu(dev
, head
, name_hlist
)
697 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
702 EXPORT_SYMBOL(dev_get_by_name_rcu
);
705 * dev_get_by_name - find a device by its name
706 * @net: the applicable net namespace
707 * @name: name to find
709 * Find an interface by name. This can be called from any
710 * context and does its own locking. The returned handle has
711 * the usage count incremented and the caller must use dev_put() to
712 * release it when it is no longer needed. %NULL is returned if no
713 * matching device is found.
716 struct net_device
*dev_get_by_name(struct net
*net
, const char *name
)
718 struct net_device
*dev
;
721 dev
= dev_get_by_name_rcu(net
, name
);
727 EXPORT_SYMBOL(dev_get_by_name
);
730 * __dev_get_by_index - find a device by its ifindex
731 * @net: the applicable net namespace
732 * @ifindex: index of device
734 * Search for an interface by index. Returns %NULL if the device
735 * is not found or a pointer to the device. The device has not
736 * had its reference counter increased so the caller must be careful
737 * about locking. The caller must hold either the RTNL semaphore
741 struct net_device
*__dev_get_by_index(struct net
*net
, int ifindex
)
743 struct net_device
*dev
;
744 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
746 hlist_for_each_entry(dev
, head
, index_hlist
)
747 if (dev
->ifindex
== ifindex
)
752 EXPORT_SYMBOL(__dev_get_by_index
);
755 * dev_get_by_index_rcu - find a device by its ifindex
756 * @net: the applicable net namespace
757 * @ifindex: index of device
759 * Search for an interface by index. Returns %NULL if the device
760 * is not found or a pointer to the device. The device has not
761 * had its reference counter increased so the caller must be careful
762 * about locking. The caller must hold RCU lock.
765 struct net_device
*dev_get_by_index_rcu(struct net
*net
, int ifindex
)
767 struct net_device
*dev
;
768 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
770 hlist_for_each_entry_rcu(dev
, head
, index_hlist
)
771 if (dev
->ifindex
== ifindex
)
776 EXPORT_SYMBOL(dev_get_by_index_rcu
);
780 * dev_get_by_index - find a device by its ifindex
781 * @net: the applicable net namespace
782 * @ifindex: index of device
784 * Search for an interface by index. Returns NULL if the device
785 * is not found or a pointer to the device. The device returned has
786 * had a reference added and the pointer is safe until the user calls
787 * dev_put to indicate they have finished with it.
790 struct net_device
*dev_get_by_index(struct net
*net
, int ifindex
)
792 struct net_device
*dev
;
795 dev
= dev_get_by_index_rcu(net
, ifindex
);
801 EXPORT_SYMBOL(dev_get_by_index
);
804 * netdev_get_name - get a netdevice name, knowing its ifindex.
805 * @net: network namespace
806 * @name: a pointer to the buffer where the name will be stored.
807 * @ifindex: the ifindex of the interface to get the name from.
809 * The use of raw_seqcount_begin() and cond_resched() before
810 * retrying is required as we want to give the writers a chance
811 * to complete when CONFIG_PREEMPT is not set.
813 int netdev_get_name(struct net
*net
, char *name
, int ifindex
)
815 struct net_device
*dev
;
819 seq
= raw_seqcount_begin(&devnet_rename_seq
);
821 dev
= dev_get_by_index_rcu(net
, ifindex
);
827 strcpy(name
, dev
->name
);
829 if (read_seqcount_retry(&devnet_rename_seq
, seq
)) {
838 * dev_getbyhwaddr_rcu - find a device by its hardware address
839 * @net: the applicable net namespace
840 * @type: media type of device
841 * @ha: hardware address
843 * Search for an interface by MAC address. Returns NULL if the device
844 * is not found or a pointer to the device.
845 * The caller must hold RCU or RTNL.
846 * The returned device has not had its ref count increased
847 * and the caller must therefore be careful about locking
851 struct net_device
*dev_getbyhwaddr_rcu(struct net
*net
, unsigned short type
,
854 struct net_device
*dev
;
856 for_each_netdev_rcu(net
, dev
)
857 if (dev
->type
== type
&&
858 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
863 EXPORT_SYMBOL(dev_getbyhwaddr_rcu
);
865 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
867 struct net_device
*dev
;
870 for_each_netdev(net
, dev
)
871 if (dev
->type
== type
)
876 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
878 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
880 struct net_device
*dev
, *ret
= NULL
;
883 for_each_netdev_rcu(net
, dev
)
884 if (dev
->type
== type
) {
892 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
895 * dev_get_by_flags_rcu - find any device with given flags
896 * @net: the applicable net namespace
897 * @if_flags: IFF_* values
898 * @mask: bitmask of bits in if_flags to check
900 * Search for any interface with the given flags. Returns NULL if a device
901 * is not found or a pointer to the device. Must be called inside
902 * rcu_read_lock(), and result refcount is unchanged.
905 struct net_device
*dev_get_by_flags_rcu(struct net
*net
, unsigned short if_flags
,
908 struct net_device
*dev
, *ret
;
911 for_each_netdev_rcu(net
, dev
) {
912 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
919 EXPORT_SYMBOL(dev_get_by_flags_rcu
);
922 * dev_valid_name - check if name is okay for network device
925 * Network device names need to be valid file names to
926 * to allow sysfs to work. We also disallow any kind of
929 bool dev_valid_name(const char *name
)
933 if (strlen(name
) >= IFNAMSIZ
)
935 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
939 if (*name
== '/' || isspace(*name
))
945 EXPORT_SYMBOL(dev_valid_name
);
948 * __dev_alloc_name - allocate a name for a device
949 * @net: network namespace to allocate the device name in
950 * @name: name format string
951 * @buf: scratch buffer and result name string
953 * Passed a format string - eg "lt%d" it will try and find a suitable
954 * id. It scans list of devices to build up a free map, then chooses
955 * the first empty slot. The caller must hold the dev_base or rtnl lock
956 * while allocating the name and adding the device in order to avoid
958 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
959 * Returns the number of the unit assigned or a negative errno code.
962 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
966 const int max_netdevices
= 8*PAGE_SIZE
;
967 unsigned long *inuse
;
968 struct net_device
*d
;
970 p
= strnchr(name
, IFNAMSIZ
-1, '%');
973 * Verify the string as this thing may have come from
974 * the user. There must be either one "%d" and no other "%"
977 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
980 /* Use one page as a bit array of possible slots */
981 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
985 for_each_netdev(net
, d
) {
986 if (!sscanf(d
->name
, name
, &i
))
988 if (i
< 0 || i
>= max_netdevices
)
991 /* avoid cases where sscanf is not exact inverse of printf */
992 snprintf(buf
, IFNAMSIZ
, name
, i
);
993 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
997 i
= find_first_zero_bit(inuse
, max_netdevices
);
998 free_page((unsigned long) inuse
);
1002 snprintf(buf
, IFNAMSIZ
, name
, i
);
1003 if (!__dev_get_by_name(net
, buf
))
1006 /* It is possible to run out of possible slots
1007 * when the name is long and there isn't enough space left
1008 * for the digits, or if all bits are used.
1014 * dev_alloc_name - allocate a name for a device
1016 * @name: name format string
1018 * Passed a format string - eg "lt%d" it will try and find a suitable
1019 * id. It scans list of devices to build up a free map, then chooses
1020 * the first empty slot. The caller must hold the dev_base or rtnl lock
1021 * while allocating the name and adding the device in order to avoid
1023 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1024 * Returns the number of the unit assigned or a negative errno code.
1027 int dev_alloc_name(struct net_device
*dev
, const char *name
)
1033 BUG_ON(!dev_net(dev
));
1035 ret
= __dev_alloc_name(net
, name
, buf
);
1037 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
1040 EXPORT_SYMBOL(dev_alloc_name
);
1042 static int dev_alloc_name_ns(struct net
*net
,
1043 struct net_device
*dev
,
1049 ret
= __dev_alloc_name(net
, name
, buf
);
1051 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
1055 static int dev_get_valid_name(struct net
*net
,
1056 struct net_device
*dev
,
1061 if (!dev_valid_name(name
))
1064 if (strchr(name
, '%'))
1065 return dev_alloc_name_ns(net
, dev
, name
);
1066 else if (__dev_get_by_name(net
, name
))
1068 else if (dev
->name
!= name
)
1069 strlcpy(dev
->name
, name
, IFNAMSIZ
);
1075 * dev_change_name - change name of a device
1077 * @newname: name (or format string) must be at least IFNAMSIZ
1079 * Change name of a device, can pass format strings "eth%d".
1082 int dev_change_name(struct net_device
*dev
, const char *newname
)
1084 char oldname
[IFNAMSIZ
];
1090 BUG_ON(!dev_net(dev
));
1093 if (dev
->flags
& IFF_UP
)
1096 write_seqcount_begin(&devnet_rename_seq
);
1098 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0) {
1099 write_seqcount_end(&devnet_rename_seq
);
1103 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
1105 err
= dev_get_valid_name(net
, dev
, newname
);
1107 write_seqcount_end(&devnet_rename_seq
);
1112 ret
= device_rename(&dev
->dev
, dev
->name
);
1114 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1115 write_seqcount_end(&devnet_rename_seq
);
1119 write_seqcount_end(&devnet_rename_seq
);
1121 write_lock_bh(&dev_base_lock
);
1122 hlist_del_rcu(&dev
->name_hlist
);
1123 write_unlock_bh(&dev_base_lock
);
1127 write_lock_bh(&dev_base_lock
);
1128 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
1129 write_unlock_bh(&dev_base_lock
);
1131 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
1132 ret
= notifier_to_errno(ret
);
1135 /* err >= 0 after dev_alloc_name() or stores the first errno */
1138 write_seqcount_begin(&devnet_rename_seq
);
1139 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1142 pr_err("%s: name change rollback failed: %d\n",
1151 * dev_set_alias - change ifalias of a device
1153 * @alias: name up to IFALIASZ
1154 * @len: limit of bytes to copy from info
1156 * Set ifalias for a device,
1158 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
1164 if (len
>= IFALIASZ
)
1168 kfree(dev
->ifalias
);
1169 dev
->ifalias
= NULL
;
1173 new_ifalias
= krealloc(dev
->ifalias
, len
+ 1, GFP_KERNEL
);
1176 dev
->ifalias
= new_ifalias
;
1178 strlcpy(dev
->ifalias
, alias
, len
+1);
1184 * netdev_features_change - device changes features
1185 * @dev: device to cause notification
1187 * Called to indicate a device has changed features.
1189 void netdev_features_change(struct net_device
*dev
)
1191 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1193 EXPORT_SYMBOL(netdev_features_change
);
1196 * netdev_state_change - device changes state
1197 * @dev: device to cause notification
1199 * Called to indicate a device has changed state. This function calls
1200 * the notifier chains for netdev_chain and sends a NEWLINK message
1201 * to the routing socket.
1203 void netdev_state_change(struct net_device
*dev
)
1205 if (dev
->flags
& IFF_UP
) {
1206 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
1207 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0, GFP_KERNEL
);
1210 EXPORT_SYMBOL(netdev_state_change
);
1213 * netdev_notify_peers - notify network peers about existence of @dev
1214 * @dev: network device
1216 * Generate traffic such that interested network peers are aware of
1217 * @dev, such as by generating a gratuitous ARP. This may be used when
1218 * a device wants to inform the rest of the network about some sort of
1219 * reconfiguration such as a failover event or virtual machine
1222 void netdev_notify_peers(struct net_device
*dev
)
1225 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
, dev
);
1228 EXPORT_SYMBOL(netdev_notify_peers
);
1230 static int __dev_open(struct net_device
*dev
)
1232 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1237 if (!netif_device_present(dev
))
1240 /* Block netpoll from trying to do any rx path servicing.
1241 * If we don't do this there is a chance ndo_poll_controller
1242 * or ndo_poll may be running while we open the device
1244 netpoll_rx_disable(dev
);
1246 ret
= call_netdevice_notifiers(NETDEV_PRE_UP
, dev
);
1247 ret
= notifier_to_errno(ret
);
1251 set_bit(__LINK_STATE_START
, &dev
->state
);
1253 if (ops
->ndo_validate_addr
)
1254 ret
= ops
->ndo_validate_addr(dev
);
1256 if (!ret
&& ops
->ndo_open
)
1257 ret
= ops
->ndo_open(dev
);
1259 netpoll_rx_enable(dev
);
1262 clear_bit(__LINK_STATE_START
, &dev
->state
);
1264 dev
->flags
|= IFF_UP
;
1265 net_dmaengine_get();
1266 dev_set_rx_mode(dev
);
1268 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
1275 * dev_open - prepare an interface for use.
1276 * @dev: device to open
1278 * Takes a device from down to up state. The device's private open
1279 * function is invoked and then the multicast lists are loaded. Finally
1280 * the device is moved into the up state and a %NETDEV_UP message is
1281 * sent to the netdev notifier chain.
1283 * Calling this function on an active interface is a nop. On a failure
1284 * a negative errno code is returned.
1286 int dev_open(struct net_device
*dev
)
1290 if (dev
->flags
& IFF_UP
)
1293 ret
= __dev_open(dev
);
1297 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
, GFP_KERNEL
);
1298 call_netdevice_notifiers(NETDEV_UP
, dev
);
1302 EXPORT_SYMBOL(dev_open
);
1304 static int __dev_close_many(struct list_head
*head
)
1306 struct net_device
*dev
;
1311 list_for_each_entry(dev
, head
, close_list
) {
1312 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1314 clear_bit(__LINK_STATE_START
, &dev
->state
);
1316 /* Synchronize to scheduled poll. We cannot touch poll list, it
1317 * can be even on different cpu. So just clear netif_running().
1319 * dev->stop() will invoke napi_disable() on all of it's
1320 * napi_struct instances on this device.
1322 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1325 dev_deactivate_many(head
);
1327 list_for_each_entry(dev
, head
, close_list
) {
1328 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1331 * Call the device specific close. This cannot fail.
1332 * Only if device is UP
1334 * We allow it to be called even after a DETACH hot-plug
1340 dev
->flags
&= ~IFF_UP
;
1341 net_dmaengine_put();
1347 static int __dev_close(struct net_device
*dev
)
1352 /* Temporarily disable netpoll until the interface is down */
1353 netpoll_rx_disable(dev
);
1355 list_add(&dev
->close_list
, &single
);
1356 retval
= __dev_close_many(&single
);
1359 netpoll_rx_enable(dev
);
1363 static int dev_close_many(struct list_head
*head
)
1365 struct net_device
*dev
, *tmp
;
1367 /* Remove the devices that don't need to be closed */
1368 list_for_each_entry_safe(dev
, tmp
, head
, close_list
)
1369 if (!(dev
->flags
& IFF_UP
))
1370 list_del_init(&dev
->close_list
);
1372 __dev_close_many(head
);
1374 list_for_each_entry_safe(dev
, tmp
, head
, close_list
) {
1375 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
, GFP_KERNEL
);
1376 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1377 list_del_init(&dev
->close_list
);
1384 * dev_close - shutdown an interface.
1385 * @dev: device to shutdown
1387 * This function moves an active device into down state. A
1388 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1389 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1392 int dev_close(struct net_device
*dev
)
1394 if (dev
->flags
& IFF_UP
) {
1397 /* Block netpoll rx while the interface is going down */
1398 netpoll_rx_disable(dev
);
1400 list_add(&dev
->close_list
, &single
);
1401 dev_close_many(&single
);
1404 netpoll_rx_enable(dev
);
1408 EXPORT_SYMBOL(dev_close
);
1412 * dev_disable_lro - disable Large Receive Offload on a device
1415 * Disable Large Receive Offload (LRO) on a net device. Must be
1416 * called under RTNL. This is needed if received packets may be
1417 * forwarded to another interface.
1419 void dev_disable_lro(struct net_device
*dev
)
1422 * If we're trying to disable lro on a vlan device
1423 * use the underlying physical device instead
1425 if (is_vlan_dev(dev
))
1426 dev
= vlan_dev_real_dev(dev
);
1428 /* the same for macvlan devices */
1429 if (netif_is_macvlan(dev
))
1430 dev
= macvlan_dev_real_dev(dev
);
1432 dev
->wanted_features
&= ~NETIF_F_LRO
;
1433 netdev_update_features(dev
);
1435 if (unlikely(dev
->features
& NETIF_F_LRO
))
1436 netdev_WARN(dev
, "failed to disable LRO!\n");
1438 EXPORT_SYMBOL(dev_disable_lro
);
1440 static int call_netdevice_notifier(struct notifier_block
*nb
, unsigned long val
,
1441 struct net_device
*dev
)
1443 struct netdev_notifier_info info
;
1445 netdev_notifier_info_init(&info
, dev
);
1446 return nb
->notifier_call(nb
, val
, &info
);
1449 static int dev_boot_phase
= 1;
1452 * register_netdevice_notifier - register a network notifier block
1455 * Register a notifier to be called when network device events occur.
1456 * The notifier passed is linked into the kernel structures and must
1457 * not be reused until it has been unregistered. A negative errno code
1458 * is returned on a failure.
1460 * When registered all registration and up events are replayed
1461 * to the new notifier to allow device to have a race free
1462 * view of the network device list.
1465 int register_netdevice_notifier(struct notifier_block
*nb
)
1467 struct net_device
*dev
;
1468 struct net_device
*last
;
1473 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1479 for_each_netdev(net
, dev
) {
1480 err
= call_netdevice_notifier(nb
, NETDEV_REGISTER
, dev
);
1481 err
= notifier_to_errno(err
);
1485 if (!(dev
->flags
& IFF_UP
))
1488 call_netdevice_notifier(nb
, NETDEV_UP
, dev
);
1499 for_each_netdev(net
, dev
) {
1503 if (dev
->flags
& IFF_UP
) {
1504 call_netdevice_notifier(nb
, NETDEV_GOING_DOWN
,
1506 call_netdevice_notifier(nb
, NETDEV_DOWN
, dev
);
1508 call_netdevice_notifier(nb
, NETDEV_UNREGISTER
, dev
);
1513 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1516 EXPORT_SYMBOL(register_netdevice_notifier
);
1519 * unregister_netdevice_notifier - unregister a network notifier block
1522 * Unregister a notifier previously registered by
1523 * register_netdevice_notifier(). The notifier is unlinked into the
1524 * kernel structures and may then be reused. A negative errno code
1525 * is returned on a failure.
1527 * After unregistering unregister and down device events are synthesized
1528 * for all devices on the device list to the removed notifier to remove
1529 * the need for special case cleanup code.
1532 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1534 struct net_device
*dev
;
1539 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1544 for_each_netdev(net
, dev
) {
1545 if (dev
->flags
& IFF_UP
) {
1546 call_netdevice_notifier(nb
, NETDEV_GOING_DOWN
,
1548 call_netdevice_notifier(nb
, NETDEV_DOWN
, dev
);
1550 call_netdevice_notifier(nb
, NETDEV_UNREGISTER
, dev
);
1557 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1560 * call_netdevice_notifiers_info - call all network notifier blocks
1561 * @val: value passed unmodified to notifier function
1562 * @dev: net_device pointer passed unmodified to notifier function
1563 * @info: notifier information data
1565 * Call all network notifier blocks. Parameters and return value
1566 * are as for raw_notifier_call_chain().
1569 int call_netdevice_notifiers_info(unsigned long val
, struct net_device
*dev
,
1570 struct netdev_notifier_info
*info
)
1573 netdev_notifier_info_init(info
, dev
);
1574 return raw_notifier_call_chain(&netdev_chain
, val
, info
);
1576 EXPORT_SYMBOL(call_netdevice_notifiers_info
);
1579 * call_netdevice_notifiers - call all network notifier blocks
1580 * @val: value passed unmodified to notifier function
1581 * @dev: net_device pointer passed unmodified to notifier function
1583 * Call all network notifier blocks. Parameters and return value
1584 * are as for raw_notifier_call_chain().
1587 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1589 struct netdev_notifier_info info
;
1591 return call_netdevice_notifiers_info(val
, dev
, &info
);
1593 EXPORT_SYMBOL(call_netdevice_notifiers
);
1595 static struct static_key netstamp_needed __read_mostly
;
1596 #ifdef HAVE_JUMP_LABEL
1597 /* We are not allowed to call static_key_slow_dec() from irq context
1598 * If net_disable_timestamp() is called from irq context, defer the
1599 * static_key_slow_dec() calls.
1601 static atomic_t netstamp_needed_deferred
;
1604 void net_enable_timestamp(void)
1606 #ifdef HAVE_JUMP_LABEL
1607 int deferred
= atomic_xchg(&netstamp_needed_deferred
, 0);
1611 static_key_slow_dec(&netstamp_needed
);
1615 static_key_slow_inc(&netstamp_needed
);
1617 EXPORT_SYMBOL(net_enable_timestamp
);
1619 void net_disable_timestamp(void)
1621 #ifdef HAVE_JUMP_LABEL
1622 if (in_interrupt()) {
1623 atomic_inc(&netstamp_needed_deferred
);
1627 static_key_slow_dec(&netstamp_needed
);
1629 EXPORT_SYMBOL(net_disable_timestamp
);
1631 static inline void net_timestamp_set(struct sk_buff
*skb
)
1633 skb
->tstamp
.tv64
= 0;
1634 if (static_key_false(&netstamp_needed
))
1635 __net_timestamp(skb
);
1638 #define net_timestamp_check(COND, SKB) \
1639 if (static_key_false(&netstamp_needed)) { \
1640 if ((COND) && !(SKB)->tstamp.tv64) \
1641 __net_timestamp(SKB); \
1644 static inline bool is_skb_forwardable(struct net_device *dev,
1645 struct sk_buff
*skb
)
1649 if (!(dev
->flags
& IFF_UP
))
1652 len
= dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
;
1653 if (skb
->len
<= len
)
1656 /* if TSO is enabled, we don't care about the length as the packet
1657 * could be forwarded without being segmented before
1659 if (skb_is_gso(skb
))
1666 * dev_forward_skb - loopback an skb to another netif
1668 * @dev: destination network device
1669 * @skb: buffer to forward
1672 * NET_RX_SUCCESS (no congestion)
1673 * NET_RX_DROP (packet was dropped, but freed)
1675 * dev_forward_skb can be used for injecting an skb from the
1676 * start_xmit function of one device into the receive queue
1677 * of another device.
1679 * The receiving device may be in another namespace, so
1680 * we have to clear all information in the skb that could
1681 * impact namespace isolation.
1683 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1685 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
1686 if (skb_copy_ubufs(skb
, GFP_ATOMIC
)) {
1687 atomic_long_inc(&dev
->rx_dropped
);
1693 if (unlikely(!is_skb_forwardable(dev
, skb
))) {
1694 atomic_long_inc(&dev
->rx_dropped
);
1699 skb_scrub_packet(skb
, true);
1700 skb
->protocol
= eth_type_trans(skb
, dev
);
1702 return netif_rx(skb
);
1704 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1706 static inline int deliver_skb(struct sk_buff
*skb
,
1707 struct packet_type
*pt_prev
,
1708 struct net_device
*orig_dev
)
1710 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
1712 atomic_inc(&skb
->users
);
1713 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1716 static inline bool skb_loop_sk(struct packet_type
*ptype
, struct sk_buff
*skb
)
1718 if (!ptype
->af_packet_priv
|| !skb
->sk
)
1721 if (ptype
->id_match
)
1722 return ptype
->id_match(ptype
, skb
->sk
);
1723 else if ((struct sock
*)ptype
->af_packet_priv
== skb
->sk
)
1730 * Support routine. Sends outgoing frames to any network
1731 * taps currently in use.
1734 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1736 struct packet_type
*ptype
;
1737 struct sk_buff
*skb2
= NULL
;
1738 struct packet_type
*pt_prev
= NULL
;
1741 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1742 /* Never send packets back to the socket
1743 * they originated from - MvS (miquels@drinkel.ow.org)
1745 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1746 (!skb_loop_sk(ptype
, skb
))) {
1748 deliver_skb(skb2
, pt_prev
, skb
->dev
);
1753 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1757 net_timestamp_set(skb2
);
1759 /* skb->nh should be correctly
1760 set by sender, so that the second statement is
1761 just protection against buggy protocols.
1763 skb_reset_mac_header(skb2
);
1765 if (skb_network_header(skb2
) < skb2
->data
||
1766 skb_network_header(skb2
) > skb_tail_pointer(skb2
)) {
1767 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1768 ntohs(skb2
->protocol
),
1770 skb_reset_network_header(skb2
);
1773 skb2
->transport_header
= skb2
->network_header
;
1774 skb2
->pkt_type
= PACKET_OUTGOING
;
1779 pt_prev
->func(skb2
, skb
->dev
, pt_prev
, skb
->dev
);
1784 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1785 * @dev: Network device
1786 * @txq: number of queues available
1788 * If real_num_tx_queues is changed the tc mappings may no longer be
1789 * valid. To resolve this verify the tc mapping remains valid and if
1790 * not NULL the mapping. With no priorities mapping to this
1791 * offset/count pair it will no longer be used. In the worst case TC0
1792 * is invalid nothing can be done so disable priority mappings. If is
1793 * expected that drivers will fix this mapping if they can before
1794 * calling netif_set_real_num_tx_queues.
1796 static void netif_setup_tc(struct net_device
*dev
, unsigned int txq
)
1799 struct netdev_tc_txq
*tc
= &dev
->tc_to_txq
[0];
1801 /* If TC0 is invalidated disable TC mapping */
1802 if (tc
->offset
+ tc
->count
> txq
) {
1803 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1808 /* Invalidated prio to tc mappings set to TC0 */
1809 for (i
= 1; i
< TC_BITMASK
+ 1; i
++) {
1810 int q
= netdev_get_prio_tc_map(dev
, i
);
1812 tc
= &dev
->tc_to_txq
[q
];
1813 if (tc
->offset
+ tc
->count
> txq
) {
1814 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1816 netdev_set_prio_tc_map(dev
, i
, 0);
1822 static DEFINE_MUTEX(xps_map_mutex
);
1823 #define xmap_dereference(P) \
1824 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1826 static struct xps_map
*remove_xps_queue(struct xps_dev_maps
*dev_maps
,
1829 struct xps_map
*map
= NULL
;
1833 map
= xmap_dereference(dev_maps
->cpu_map
[cpu
]);
1835 for (pos
= 0; map
&& pos
< map
->len
; pos
++) {
1836 if (map
->queues
[pos
] == index
) {
1838 map
->queues
[pos
] = map
->queues
[--map
->len
];
1840 RCU_INIT_POINTER(dev_maps
->cpu_map
[cpu
], NULL
);
1841 kfree_rcu(map
, rcu
);
1851 static void netif_reset_xps_queues_gt(struct net_device
*dev
, u16 index
)
1853 struct xps_dev_maps
*dev_maps
;
1855 bool active
= false;
1857 mutex_lock(&xps_map_mutex
);
1858 dev_maps
= xmap_dereference(dev
->xps_maps
);
1863 for_each_possible_cpu(cpu
) {
1864 for (i
= index
; i
< dev
->num_tx_queues
; i
++) {
1865 if (!remove_xps_queue(dev_maps
, cpu
, i
))
1868 if (i
== dev
->num_tx_queues
)
1873 RCU_INIT_POINTER(dev
->xps_maps
, NULL
);
1874 kfree_rcu(dev_maps
, rcu
);
1877 for (i
= index
; i
< dev
->num_tx_queues
; i
++)
1878 netdev_queue_numa_node_write(netdev_get_tx_queue(dev
, i
),
1882 mutex_unlock(&xps_map_mutex
);
1885 static struct xps_map
*expand_xps_map(struct xps_map
*map
,
1888 struct xps_map
*new_map
;
1889 int alloc_len
= XPS_MIN_MAP_ALLOC
;
1892 for (pos
= 0; map
&& pos
< map
->len
; pos
++) {
1893 if (map
->queues
[pos
] != index
)
1898 /* Need to add queue to this CPU's existing map */
1900 if (pos
< map
->alloc_len
)
1903 alloc_len
= map
->alloc_len
* 2;
1906 /* Need to allocate new map to store queue on this CPU's map */
1907 new_map
= kzalloc_node(XPS_MAP_SIZE(alloc_len
), GFP_KERNEL
,
1912 for (i
= 0; i
< pos
; i
++)
1913 new_map
->queues
[i
] = map
->queues
[i
];
1914 new_map
->alloc_len
= alloc_len
;
1920 int netif_set_xps_queue(struct net_device
*dev
, const struct cpumask
*mask
,
1923 struct xps_dev_maps
*dev_maps
, *new_dev_maps
= NULL
;
1924 struct xps_map
*map
, *new_map
;
1925 int maps_sz
= max_t(unsigned int, XPS_DEV_MAPS_SIZE
, L1_CACHE_BYTES
);
1926 int cpu
, numa_node_id
= -2;
1927 bool active
= false;
1929 mutex_lock(&xps_map_mutex
);
1931 dev_maps
= xmap_dereference(dev
->xps_maps
);
1933 /* allocate memory for queue storage */
1934 for_each_online_cpu(cpu
) {
1935 if (!cpumask_test_cpu(cpu
, mask
))
1939 new_dev_maps
= kzalloc(maps_sz
, GFP_KERNEL
);
1940 if (!new_dev_maps
) {
1941 mutex_unlock(&xps_map_mutex
);
1945 map
= dev_maps
? xmap_dereference(dev_maps
->cpu_map
[cpu
]) :
1948 map
= expand_xps_map(map
, cpu
, index
);
1952 RCU_INIT_POINTER(new_dev_maps
->cpu_map
[cpu
], map
);
1956 goto out_no_new_maps
;
1958 for_each_possible_cpu(cpu
) {
1959 if (cpumask_test_cpu(cpu
, mask
) && cpu_online(cpu
)) {
1960 /* add queue to CPU maps */
1963 map
= xmap_dereference(new_dev_maps
->cpu_map
[cpu
]);
1964 while ((pos
< map
->len
) && (map
->queues
[pos
] != index
))
1967 if (pos
== map
->len
)
1968 map
->queues
[map
->len
++] = index
;
1970 if (numa_node_id
== -2)
1971 numa_node_id
= cpu_to_node(cpu
);
1972 else if (numa_node_id
!= cpu_to_node(cpu
))
1975 } else if (dev_maps
) {
1976 /* fill in the new device map from the old device map */
1977 map
= xmap_dereference(dev_maps
->cpu_map
[cpu
]);
1978 RCU_INIT_POINTER(new_dev_maps
->cpu_map
[cpu
], map
);
1983 rcu_assign_pointer(dev
->xps_maps
, new_dev_maps
);
1985 /* Cleanup old maps */
1987 for_each_possible_cpu(cpu
) {
1988 new_map
= xmap_dereference(new_dev_maps
->cpu_map
[cpu
]);
1989 map
= xmap_dereference(dev_maps
->cpu_map
[cpu
]);
1990 if (map
&& map
!= new_map
)
1991 kfree_rcu(map
, rcu
);
1994 kfree_rcu(dev_maps
, rcu
);
1997 dev_maps
= new_dev_maps
;
2001 /* update Tx queue numa node */
2002 netdev_queue_numa_node_write(netdev_get_tx_queue(dev
, index
),
2003 (numa_node_id
>= 0) ? numa_node_id
:
2009 /* removes queue from unused CPUs */
2010 for_each_possible_cpu(cpu
) {
2011 if (cpumask_test_cpu(cpu
, mask
) && cpu_online(cpu
))
2014 if (remove_xps_queue(dev_maps
, cpu
, index
))
2018 /* free map if not active */
2020 RCU_INIT_POINTER(dev
->xps_maps
, NULL
);
2021 kfree_rcu(dev_maps
, rcu
);
2025 mutex_unlock(&xps_map_mutex
);
2029 /* remove any maps that we added */
2030 for_each_possible_cpu(cpu
) {
2031 new_map
= xmap_dereference(new_dev_maps
->cpu_map
[cpu
]);
2032 map
= dev_maps
? xmap_dereference(dev_maps
->cpu_map
[cpu
]) :
2034 if (new_map
&& new_map
!= map
)
2038 mutex_unlock(&xps_map_mutex
);
2040 kfree(new_dev_maps
);
2043 EXPORT_SYMBOL(netif_set_xps_queue
);
2047 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2048 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2050 int netif_set_real_num_tx_queues(struct net_device
*dev
, unsigned int txq
)
2054 if (txq
< 1 || txq
> dev
->num_tx_queues
)
2057 if (dev
->reg_state
== NETREG_REGISTERED
||
2058 dev
->reg_state
== NETREG_UNREGISTERING
) {
2061 rc
= netdev_queue_update_kobjects(dev
, dev
->real_num_tx_queues
,
2067 netif_setup_tc(dev
, txq
);
2069 if (txq
< dev
->real_num_tx_queues
) {
2070 qdisc_reset_all_tx_gt(dev
, txq
);
2072 netif_reset_xps_queues_gt(dev
, txq
);
2077 dev
->real_num_tx_queues
= txq
;
2080 EXPORT_SYMBOL(netif_set_real_num_tx_queues
);
2084 * netif_set_real_num_rx_queues - set actual number of RX queues used
2085 * @dev: Network device
2086 * @rxq: Actual number of RX queues
2088 * This must be called either with the rtnl_lock held or before
2089 * registration of the net device. Returns 0 on success, or a
2090 * negative error code. If called before registration, it always
2093 int netif_set_real_num_rx_queues(struct net_device
*dev
, unsigned int rxq
)
2097 if (rxq
< 1 || rxq
> dev
->num_rx_queues
)
2100 if (dev
->reg_state
== NETREG_REGISTERED
) {
2103 rc
= net_rx_queue_update_kobjects(dev
, dev
->real_num_rx_queues
,
2109 dev
->real_num_rx_queues
= rxq
;
2112 EXPORT_SYMBOL(netif_set_real_num_rx_queues
);
2116 * netif_get_num_default_rss_queues - default number of RSS queues
2118 * This routine should set an upper limit on the number of RSS queues
2119 * used by default by multiqueue devices.
2121 int netif_get_num_default_rss_queues(void)
2123 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES
, num_online_cpus());
2125 EXPORT_SYMBOL(netif_get_num_default_rss_queues
);
2127 static inline void __netif_reschedule(struct Qdisc
*q
)
2129 struct softnet_data
*sd
;
2130 unsigned long flags
;
2132 local_irq_save(flags
);
2133 sd
= &__get_cpu_var(softnet_data
);
2134 q
->next_sched
= NULL
;
2135 *sd
->output_queue_tailp
= q
;
2136 sd
->output_queue_tailp
= &q
->next_sched
;
2137 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
2138 local_irq_restore(flags
);
2141 void __netif_schedule(struct Qdisc
*q
)
2143 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
2144 __netif_reschedule(q
);
2146 EXPORT_SYMBOL(__netif_schedule
);
2148 void dev_kfree_skb_irq(struct sk_buff
*skb
)
2150 if (atomic_dec_and_test(&skb
->users
)) {
2151 struct softnet_data
*sd
;
2152 unsigned long flags
;
2154 local_irq_save(flags
);
2155 sd
= &__get_cpu_var(softnet_data
);
2156 skb
->next
= sd
->completion_queue
;
2157 sd
->completion_queue
= skb
;
2158 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
2159 local_irq_restore(flags
);
2162 EXPORT_SYMBOL(dev_kfree_skb_irq
);
2164 void dev_kfree_skb_any(struct sk_buff
*skb
)
2166 if (in_irq() || irqs_disabled())
2167 dev_kfree_skb_irq(skb
);
2171 EXPORT_SYMBOL(dev_kfree_skb_any
);
2175 * netif_device_detach - mark device as removed
2176 * @dev: network device
2178 * Mark device as removed from system and therefore no longer available.
2180 void netif_device_detach(struct net_device
*dev
)
2182 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
2183 netif_running(dev
)) {
2184 netif_tx_stop_all_queues(dev
);
2187 EXPORT_SYMBOL(netif_device_detach
);
2190 * netif_device_attach - mark device as attached
2191 * @dev: network device
2193 * Mark device as attached from system and restart if needed.
2195 void netif_device_attach(struct net_device
*dev
)
2197 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
2198 netif_running(dev
)) {
2199 netif_tx_wake_all_queues(dev
);
2200 __netdev_watchdog_up(dev
);
2203 EXPORT_SYMBOL(netif_device_attach
);
2205 static void skb_warn_bad_offload(const struct sk_buff
*skb
)
2207 static const netdev_features_t null_features
= 0;
2208 struct net_device
*dev
= skb
->dev
;
2209 const char *driver
= "";
2211 if (!net_ratelimit())
2214 if (dev
&& dev
->dev
.parent
)
2215 driver
= dev_driver_string(dev
->dev
.parent
);
2217 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2218 "gso_type=%d ip_summed=%d\n",
2219 driver
, dev
? &dev
->features
: &null_features
,
2220 skb
->sk
? &skb
->sk
->sk_route_caps
: &null_features
,
2221 skb
->len
, skb
->data_len
, skb_shinfo(skb
)->gso_size
,
2222 skb_shinfo(skb
)->gso_type
, skb
->ip_summed
);
2226 * Invalidate hardware checksum when packet is to be mangled, and
2227 * complete checksum manually on outgoing path.
2229 int skb_checksum_help(struct sk_buff
*skb
)
2232 int ret
= 0, offset
;
2234 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
2235 goto out_set_summed
;
2237 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
2238 skb_warn_bad_offload(skb
);
2242 /* Before computing a checksum, we should make sure no frag could
2243 * be modified by an external entity : checksum could be wrong.
2245 if (skb_has_shared_frag(skb
)) {
2246 ret
= __skb_linearize(skb
);
2251 offset
= skb_checksum_start_offset(skb
);
2252 BUG_ON(offset
>= skb_headlen(skb
));
2253 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
2255 offset
+= skb
->csum_offset
;
2256 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
2258 if (skb_cloned(skb
) &&
2259 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
2260 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2265 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
2267 skb
->ip_summed
= CHECKSUM_NONE
;
2271 EXPORT_SYMBOL(skb_checksum_help
);
2273 __be16
skb_network_protocol(struct sk_buff
*skb
)
2275 __be16 type
= skb
->protocol
;
2276 int vlan_depth
= ETH_HLEN
;
2278 /* Tunnel gso handlers can set protocol to ethernet. */
2279 if (type
== htons(ETH_P_TEB
)) {
2282 if (unlikely(!pskb_may_pull(skb
, sizeof(struct ethhdr
))))
2285 eth
= (struct ethhdr
*)skb_mac_header(skb
);
2286 type
= eth
->h_proto
;
2289 while (type
== htons(ETH_P_8021Q
) || type
== htons(ETH_P_8021AD
)) {
2290 struct vlan_hdr
*vh
;
2292 if (unlikely(!pskb_may_pull(skb
, vlan_depth
+ VLAN_HLEN
)))
2295 vh
= (struct vlan_hdr
*)(skb
->data
+ vlan_depth
);
2296 type
= vh
->h_vlan_encapsulated_proto
;
2297 vlan_depth
+= VLAN_HLEN
;
2304 * skb_mac_gso_segment - mac layer segmentation handler.
2305 * @skb: buffer to segment
2306 * @features: features for the output path (see dev->features)
2308 struct sk_buff
*skb_mac_gso_segment(struct sk_buff
*skb
,
2309 netdev_features_t features
)
2311 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
2312 struct packet_offload
*ptype
;
2313 __be16 type
= skb_network_protocol(skb
);
2315 if (unlikely(!type
))
2316 return ERR_PTR(-EINVAL
);
2318 __skb_pull(skb
, skb
->mac_len
);
2321 list_for_each_entry_rcu(ptype
, &offload_base
, list
) {
2322 if (ptype
->type
== type
&& ptype
->callbacks
.gso_segment
) {
2323 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
2326 err
= ptype
->callbacks
.gso_send_check(skb
);
2327 segs
= ERR_PTR(err
);
2328 if (err
|| skb_gso_ok(skb
, features
))
2330 __skb_push(skb
, (skb
->data
-
2331 skb_network_header(skb
)));
2333 segs
= ptype
->callbacks
.gso_segment(skb
, features
);
2339 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
2343 EXPORT_SYMBOL(skb_mac_gso_segment
);
2346 /* openvswitch calls this on rx path, so we need a different check.
2348 static inline bool skb_needs_check(struct sk_buff
*skb
, bool tx_path
)
2351 return skb
->ip_summed
!= CHECKSUM_PARTIAL
;
2353 return skb
->ip_summed
== CHECKSUM_NONE
;
2357 * __skb_gso_segment - Perform segmentation on skb.
2358 * @skb: buffer to segment
2359 * @features: features for the output path (see dev->features)
2360 * @tx_path: whether it is called in TX path
2362 * This function segments the given skb and returns a list of segments.
2364 * It may return NULL if the skb requires no segmentation. This is
2365 * only possible when GSO is used for verifying header integrity.
2367 struct sk_buff
*__skb_gso_segment(struct sk_buff
*skb
,
2368 netdev_features_t features
, bool tx_path
)
2370 if (unlikely(skb_needs_check(skb
, tx_path
))) {
2373 skb_warn_bad_offload(skb
);
2375 if (skb_header_cloned(skb
) &&
2376 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
2377 return ERR_PTR(err
);
2380 SKB_GSO_CB(skb
)->mac_offset
= skb_headroom(skb
);
2381 SKB_GSO_CB(skb
)->encap_level
= 0;
2383 skb_reset_mac_header(skb
);
2384 skb_reset_mac_len(skb
);
2386 return skb_mac_gso_segment(skb
, features
);
2388 EXPORT_SYMBOL(__skb_gso_segment
);
2390 /* Take action when hardware reception checksum errors are detected. */
2392 void netdev_rx_csum_fault(struct net_device
*dev
)
2394 if (net_ratelimit()) {
2395 pr_err("%s: hw csum failure\n", dev
? dev
->name
: "<unknown>");
2399 EXPORT_SYMBOL(netdev_rx_csum_fault
);
2402 /* Actually, we should eliminate this check as soon as we know, that:
2403 * 1. IOMMU is present and allows to map all the memory.
2404 * 2. No high memory really exists on this machine.
2407 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
2409 #ifdef CONFIG_HIGHMEM
2411 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
2412 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2413 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2414 if (PageHighMem(skb_frag_page(frag
)))
2419 if (PCI_DMA_BUS_IS_PHYS
) {
2420 struct device
*pdev
= dev
->dev
.parent
;
2424 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2425 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2426 dma_addr_t addr
= page_to_phys(skb_frag_page(frag
));
2427 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
2436 void (*destructor
)(struct sk_buff
*skb
);
2439 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2441 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
2443 struct dev_gso_cb
*cb
;
2446 struct sk_buff
*nskb
= skb
->next
;
2448 skb
->next
= nskb
->next
;
2451 } while (skb
->next
);
2453 cb
= DEV_GSO_CB(skb
);
2455 cb
->destructor(skb
);
2459 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2460 * @skb: buffer to segment
2461 * @features: device features as applicable to this skb
2463 * This function segments the given skb and stores the list of segments
2466 static int dev_gso_segment(struct sk_buff
*skb
, netdev_features_t features
)
2468 struct sk_buff
*segs
;
2470 segs
= skb_gso_segment(skb
, features
);
2472 /* Verifying header integrity only. */
2477 return PTR_ERR(segs
);
2480 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
2481 skb
->destructor
= dev_gso_skb_destructor
;
2486 static netdev_features_t
harmonize_features(struct sk_buff
*skb
,
2487 netdev_features_t features
)
2489 if (skb
->ip_summed
!= CHECKSUM_NONE
&&
2490 !can_checksum_protocol(features
, skb_network_protocol(skb
))) {
2491 features
&= ~NETIF_F_ALL_CSUM
;
2492 } else if (illegal_highdma(skb
->dev
, skb
)) {
2493 features
&= ~NETIF_F_SG
;
2499 netdev_features_t
netif_skb_features(struct sk_buff
*skb
)
2501 __be16 protocol
= skb
->protocol
;
2502 netdev_features_t features
= skb
->dev
->features
;
2504 if (skb_shinfo(skb
)->gso_segs
> skb
->dev
->gso_max_segs
)
2505 features
&= ~NETIF_F_GSO_MASK
;
2507 if (protocol
== htons(ETH_P_8021Q
) || protocol
== htons(ETH_P_8021AD
)) {
2508 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
2509 protocol
= veh
->h_vlan_encapsulated_proto
;
2510 } else if (!vlan_tx_tag_present(skb
)) {
2511 return harmonize_features(skb
, features
);
2514 features
&= (skb
->dev
->vlan_features
| NETIF_F_HW_VLAN_CTAG_TX
|
2515 NETIF_F_HW_VLAN_STAG_TX
);
2517 if (protocol
== htons(ETH_P_8021Q
) || protocol
== htons(ETH_P_8021AD
))
2518 features
&= NETIF_F_SG
| NETIF_F_HIGHDMA
| NETIF_F_FRAGLIST
|
2519 NETIF_F_GEN_CSUM
| NETIF_F_HW_VLAN_CTAG_TX
|
2520 NETIF_F_HW_VLAN_STAG_TX
;
2522 return harmonize_features(skb
, features
);
2524 EXPORT_SYMBOL(netif_skb_features
);
2527 * Returns true if either:
2528 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2529 * 2. skb is fragmented and the device does not support SG.
2531 static inline int skb_needs_linearize(struct sk_buff
*skb
,
2532 netdev_features_t features
)
2534 return skb_is_nonlinear(skb
) &&
2535 ((skb_has_frag_list(skb
) &&
2536 !(features
& NETIF_F_FRAGLIST
)) ||
2537 (skb_shinfo(skb
)->nr_frags
&&
2538 !(features
& NETIF_F_SG
)));
2541 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
2542 struct netdev_queue
*txq
, void *accel_priv
)
2544 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2545 int rc
= NETDEV_TX_OK
;
2546 unsigned int skb_len
;
2548 if (likely(!skb
->next
)) {
2549 netdev_features_t features
;
2552 * If device doesn't need skb->dst, release it right now while
2553 * its hot in this cpu cache
2555 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2558 features
= netif_skb_features(skb
);
2560 if (vlan_tx_tag_present(skb
) &&
2561 !vlan_hw_offload_capable(features
, skb
->vlan_proto
)) {
2562 skb
= __vlan_put_tag(skb
, skb
->vlan_proto
,
2563 vlan_tx_tag_get(skb
));
2570 /* If encapsulation offload request, verify we are testing
2571 * hardware encapsulation features instead of standard
2572 * features for the netdev
2574 if (skb
->encapsulation
)
2575 features
&= dev
->hw_enc_features
;
2577 if (netif_needs_gso(skb
, features
)) {
2578 if (unlikely(dev_gso_segment(skb
, features
)))
2583 if (skb_needs_linearize(skb
, features
) &&
2584 __skb_linearize(skb
))
2587 /* If packet is not checksummed and device does not
2588 * support checksumming for this protocol, complete
2589 * checksumming here.
2591 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2592 if (skb
->encapsulation
)
2593 skb_set_inner_transport_header(skb
,
2594 skb_checksum_start_offset(skb
));
2596 skb_set_transport_header(skb
,
2597 skb_checksum_start_offset(skb
));
2598 if (!(features
& NETIF_F_ALL_CSUM
) &&
2599 skb_checksum_help(skb
))
2604 if (!list_empty(&ptype_all
))
2605 dev_queue_xmit_nit(skb
, dev
);
2609 rc
= ops
->ndo_dfwd_start_xmit(skb
, dev
, accel_priv
);
2611 rc
= ops
->ndo_start_xmit(skb
, dev
);
2613 trace_net_dev_xmit(skb
, rc
, dev
, skb_len
);
2614 if (rc
== NETDEV_TX_OK
&& txq
)
2615 txq_trans_update(txq
);
2621 struct sk_buff
*nskb
= skb
->next
;
2623 skb
->next
= nskb
->next
;
2626 if (!list_empty(&ptype_all
))
2627 dev_queue_xmit_nit(nskb
, dev
);
2629 skb_len
= nskb
->len
;
2631 rc
= ops
->ndo_dfwd_start_xmit(nskb
, dev
, accel_priv
);
2633 rc
= ops
->ndo_start_xmit(nskb
, dev
);
2634 trace_net_dev_xmit(nskb
, rc
, dev
, skb_len
);
2635 if (unlikely(rc
!= NETDEV_TX_OK
)) {
2636 if (rc
& ~NETDEV_TX_MASK
)
2637 goto out_kfree_gso_skb
;
2638 nskb
->next
= skb
->next
;
2642 txq_trans_update(txq
);
2643 if (unlikely(netif_xmit_stopped(txq
) && skb
->next
))
2644 return NETDEV_TX_BUSY
;
2645 } while (skb
->next
);
2648 if (likely(skb
->next
== NULL
)) {
2649 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
2658 EXPORT_SYMBOL_GPL(dev_hard_start_xmit
);
2660 static void qdisc_pkt_len_init(struct sk_buff
*skb
)
2662 const struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
2664 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
2666 /* To get more precise estimation of bytes sent on wire,
2667 * we add to pkt_len the headers size of all segments
2669 if (shinfo
->gso_size
) {
2670 unsigned int hdr_len
;
2671 u16 gso_segs
= shinfo
->gso_segs
;
2673 /* mac layer + network layer */
2674 hdr_len
= skb_transport_header(skb
) - skb_mac_header(skb
);
2676 /* + transport layer */
2677 if (likely(shinfo
->gso_type
& (SKB_GSO_TCPV4
| SKB_GSO_TCPV6
)))
2678 hdr_len
+= tcp_hdrlen(skb
);
2680 hdr_len
+= sizeof(struct udphdr
);
2682 if (shinfo
->gso_type
& SKB_GSO_DODGY
)
2683 gso_segs
= DIV_ROUND_UP(skb
->len
- hdr_len
,
2686 qdisc_skb_cb(skb
)->pkt_len
+= (gso_segs
- 1) * hdr_len
;
2690 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2691 struct net_device
*dev
,
2692 struct netdev_queue
*txq
)
2694 spinlock_t
*root_lock
= qdisc_lock(q
);
2698 qdisc_pkt_len_init(skb
);
2699 qdisc_calculate_pkt_len(skb
, q
);
2701 * Heuristic to force contended enqueues to serialize on a
2702 * separate lock before trying to get qdisc main lock.
2703 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2704 * and dequeue packets faster.
2706 contended
= qdisc_is_running(q
);
2707 if (unlikely(contended
))
2708 spin_lock(&q
->busylock
);
2710 spin_lock(root_lock
);
2711 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2714 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2715 qdisc_run_begin(q
)) {
2717 * This is a work-conserving queue; there are no old skbs
2718 * waiting to be sent out; and the qdisc is not running -
2719 * xmit the skb directly.
2721 if (!(dev
->priv_flags
& IFF_XMIT_DST_RELEASE
))
2724 qdisc_bstats_update(q
, skb
);
2726 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
)) {
2727 if (unlikely(contended
)) {
2728 spin_unlock(&q
->busylock
);
2735 rc
= NET_XMIT_SUCCESS
;
2738 rc
= q
->enqueue(skb
, q
) & NET_XMIT_MASK
;
2739 if (qdisc_run_begin(q
)) {
2740 if (unlikely(contended
)) {
2741 spin_unlock(&q
->busylock
);
2747 spin_unlock(root_lock
);
2748 if (unlikely(contended
))
2749 spin_unlock(&q
->busylock
);
2753 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2754 static void skb_update_prio(struct sk_buff
*skb
)
2756 struct netprio_map
*map
= rcu_dereference_bh(skb
->dev
->priomap
);
2758 if (!skb
->priority
&& skb
->sk
&& map
) {
2759 unsigned int prioidx
= skb
->sk
->sk_cgrp_prioidx
;
2761 if (prioidx
< map
->priomap_len
)
2762 skb
->priority
= map
->priomap
[prioidx
];
2766 #define skb_update_prio(skb)
2769 static DEFINE_PER_CPU(int, xmit_recursion
);
2770 #define RECURSION_LIMIT 10
2773 * dev_loopback_xmit - loop back @skb
2774 * @skb: buffer to transmit
2776 int dev_loopback_xmit(struct sk_buff
*skb
)
2778 skb_reset_mac_header(skb
);
2779 __skb_pull(skb
, skb_network_offset(skb
));
2780 skb
->pkt_type
= PACKET_LOOPBACK
;
2781 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
2782 WARN_ON(!skb_dst(skb
));
2787 EXPORT_SYMBOL(dev_loopback_xmit
);
2790 * dev_queue_xmit - transmit a buffer
2791 * @skb: buffer to transmit
2793 * Queue a buffer for transmission to a network device. The caller must
2794 * have set the device and priority and built the buffer before calling
2795 * this function. The function can be called from an interrupt.
2797 * A negative errno code is returned on a failure. A success does not
2798 * guarantee the frame will be transmitted as it may be dropped due
2799 * to congestion or traffic shaping.
2801 * -----------------------------------------------------------------------------------
2802 * I notice this method can also return errors from the queue disciplines,
2803 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2806 * Regardless of the return value, the skb is consumed, so it is currently
2807 * difficult to retry a send to this method. (You can bump the ref count
2808 * before sending to hold a reference for retry if you are careful.)
2810 * When calling this method, interrupts MUST be enabled. This is because
2811 * the BH enable code must have IRQs enabled so that it will not deadlock.
2814 int dev_queue_xmit(struct sk_buff
*skb
)
2816 struct net_device
*dev
= skb
->dev
;
2817 struct netdev_queue
*txq
;
2821 skb_reset_mac_header(skb
);
2823 /* Disable soft irqs for various locks below. Also
2824 * stops preemption for RCU.
2828 skb_update_prio(skb
);
2830 txq
= netdev_pick_tx(dev
, skb
);
2831 q
= rcu_dereference_bh(txq
->qdisc
);
2833 #ifdef CONFIG_NET_CLS_ACT
2834 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2836 trace_net_dev_queue(skb
);
2838 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2842 /* The device has no queue. Common case for software devices:
2843 loopback, all the sorts of tunnels...
2845 Really, it is unlikely that netif_tx_lock protection is necessary
2846 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2848 However, it is possible, that they rely on protection
2851 Check this and shot the lock. It is not prone from deadlocks.
2852 Either shot noqueue qdisc, it is even simpler 8)
2854 if (dev
->flags
& IFF_UP
) {
2855 int cpu
= smp_processor_id(); /* ok because BHs are off */
2857 if (txq
->xmit_lock_owner
!= cpu
) {
2859 if (__this_cpu_read(xmit_recursion
) > RECURSION_LIMIT
)
2860 goto recursion_alert
;
2862 HARD_TX_LOCK(dev
, txq
, cpu
);
2864 if (!netif_xmit_stopped(txq
)) {
2865 __this_cpu_inc(xmit_recursion
);
2866 rc
= dev_hard_start_xmit(skb
, dev
, txq
, NULL
);
2867 __this_cpu_dec(xmit_recursion
);
2868 if (dev_xmit_complete(rc
)) {
2869 HARD_TX_UNLOCK(dev
, txq
);
2873 HARD_TX_UNLOCK(dev
, txq
);
2874 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2877 /* Recursion is detected! It is possible,
2881 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2887 rcu_read_unlock_bh();
2892 rcu_read_unlock_bh();
2895 EXPORT_SYMBOL(dev_queue_xmit
);
2898 /*=======================================================================
2900 =======================================================================*/
2902 int netdev_max_backlog __read_mostly
= 1000;
2903 EXPORT_SYMBOL(netdev_max_backlog
);
2905 int netdev_tstamp_prequeue __read_mostly
= 1;
2906 int netdev_budget __read_mostly
= 300;
2907 int weight_p __read_mostly
= 64; /* old backlog weight */
2909 /* Called with irq disabled */
2910 static inline void ____napi_schedule(struct softnet_data
*sd
,
2911 struct napi_struct
*napi
)
2913 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
2914 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2919 /* One global table that all flow-based protocols share. */
2920 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
2921 EXPORT_SYMBOL(rps_sock_flow_table
);
2923 struct static_key rps_needed __read_mostly
;
2925 static struct rps_dev_flow
*
2926 set_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2927 struct rps_dev_flow
*rflow
, u16 next_cpu
)
2929 if (next_cpu
!= RPS_NO_CPU
) {
2930 #ifdef CONFIG_RFS_ACCEL
2931 struct netdev_rx_queue
*rxqueue
;
2932 struct rps_dev_flow_table
*flow_table
;
2933 struct rps_dev_flow
*old_rflow
;
2938 /* Should we steer this flow to a different hardware queue? */
2939 if (!skb_rx_queue_recorded(skb
) || !dev
->rx_cpu_rmap
||
2940 !(dev
->features
& NETIF_F_NTUPLE
))
2942 rxq_index
= cpu_rmap_lookup_index(dev
->rx_cpu_rmap
, next_cpu
);
2943 if (rxq_index
== skb_get_rx_queue(skb
))
2946 rxqueue
= dev
->_rx
+ rxq_index
;
2947 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2950 flow_id
= skb
->rxhash
& flow_table
->mask
;
2951 rc
= dev
->netdev_ops
->ndo_rx_flow_steer(dev
, skb
,
2952 rxq_index
, flow_id
);
2956 rflow
= &flow_table
->flows
[flow_id
];
2958 if (old_rflow
->filter
== rflow
->filter
)
2959 old_rflow
->filter
= RPS_NO_FILTER
;
2963 per_cpu(softnet_data
, next_cpu
).input_queue_head
;
2966 rflow
->cpu
= next_cpu
;
2971 * get_rps_cpu is called from netif_receive_skb and returns the target
2972 * CPU from the RPS map of the receiving queue for a given skb.
2973 * rcu_read_lock must be held on entry.
2975 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2976 struct rps_dev_flow
**rflowp
)
2978 struct netdev_rx_queue
*rxqueue
;
2979 struct rps_map
*map
;
2980 struct rps_dev_flow_table
*flow_table
;
2981 struct rps_sock_flow_table
*sock_flow_table
;
2985 if (skb_rx_queue_recorded(skb
)) {
2986 u16 index
= skb_get_rx_queue(skb
);
2987 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
2988 WARN_ONCE(dev
->real_num_rx_queues
> 1,
2989 "%s received packet on queue %u, but number "
2990 "of RX queues is %u\n",
2991 dev
->name
, index
, dev
->real_num_rx_queues
);
2994 rxqueue
= dev
->_rx
+ index
;
2998 map
= rcu_dereference(rxqueue
->rps_map
);
3000 if (map
->len
== 1 &&
3001 !rcu_access_pointer(rxqueue
->rps_flow_table
)) {
3002 tcpu
= map
->cpus
[0];
3003 if (cpu_online(tcpu
))
3007 } else if (!rcu_access_pointer(rxqueue
->rps_flow_table
)) {
3011 skb_reset_network_header(skb
);
3012 if (!skb_get_rxhash(skb
))
3015 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3016 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
3017 if (flow_table
&& sock_flow_table
) {
3019 struct rps_dev_flow
*rflow
;
3021 rflow
= &flow_table
->flows
[skb
->rxhash
& flow_table
->mask
];
3024 next_cpu
= sock_flow_table
->ents
[skb
->rxhash
&
3025 sock_flow_table
->mask
];
3028 * If the desired CPU (where last recvmsg was done) is
3029 * different from current CPU (one in the rx-queue flow
3030 * table entry), switch if one of the following holds:
3031 * - Current CPU is unset (equal to RPS_NO_CPU).
3032 * - Current CPU is offline.
3033 * - The current CPU's queue tail has advanced beyond the
3034 * last packet that was enqueued using this table entry.
3035 * This guarantees that all previous packets for the flow
3036 * have been dequeued, thus preserving in order delivery.
3038 if (unlikely(tcpu
!= next_cpu
) &&
3039 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
3040 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
3041 rflow
->last_qtail
)) >= 0)) {
3043 rflow
= set_rps_cpu(dev
, skb
, rflow
, next_cpu
);
3046 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
3054 tcpu
= map
->cpus
[((u64
) skb
->rxhash
* map
->len
) >> 32];
3056 if (cpu_online(tcpu
)) {
3066 #ifdef CONFIG_RFS_ACCEL
3069 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3070 * @dev: Device on which the filter was set
3071 * @rxq_index: RX queue index
3072 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3073 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3075 * Drivers that implement ndo_rx_flow_steer() should periodically call
3076 * this function for each installed filter and remove the filters for
3077 * which it returns %true.
3079 bool rps_may_expire_flow(struct net_device
*dev
, u16 rxq_index
,
3080 u32 flow_id
, u16 filter_id
)
3082 struct netdev_rx_queue
*rxqueue
= dev
->_rx
+ rxq_index
;
3083 struct rps_dev_flow_table
*flow_table
;
3084 struct rps_dev_flow
*rflow
;
3089 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3090 if (flow_table
&& flow_id
<= flow_table
->mask
) {
3091 rflow
= &flow_table
->flows
[flow_id
];
3092 cpu
= ACCESS_ONCE(rflow
->cpu
);
3093 if (rflow
->filter
== filter_id
&& cpu
!= RPS_NO_CPU
&&
3094 ((int)(per_cpu(softnet_data
, cpu
).input_queue_head
-
3095 rflow
->last_qtail
) <
3096 (int)(10 * flow_table
->mask
)))
3102 EXPORT_SYMBOL(rps_may_expire_flow
);
3104 #endif /* CONFIG_RFS_ACCEL */
3106 /* Called from hardirq (IPI) context */
3107 static void rps_trigger_softirq(void *data
)
3109 struct softnet_data
*sd
= data
;
3111 ____napi_schedule(sd
, &sd
->backlog
);
3115 #endif /* CONFIG_RPS */
3118 * Check if this softnet_data structure is another cpu one
3119 * If yes, queue it to our IPI list and return 1
3122 static int rps_ipi_queued(struct softnet_data
*sd
)
3125 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
3128 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
3129 mysd
->rps_ipi_list
= sd
;
3131 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3134 #endif /* CONFIG_RPS */
3138 #ifdef CONFIG_NET_FLOW_LIMIT
3139 int netdev_flow_limit_table_len __read_mostly
= (1 << 12);
3142 static bool skb_flow_limit(struct sk_buff
*skb
, unsigned int qlen
)
3144 #ifdef CONFIG_NET_FLOW_LIMIT
3145 struct sd_flow_limit
*fl
;
3146 struct softnet_data
*sd
;
3147 unsigned int old_flow
, new_flow
;
3149 if (qlen
< (netdev_max_backlog
>> 1))
3152 sd
= &__get_cpu_var(softnet_data
);
3155 fl
= rcu_dereference(sd
->flow_limit
);
3157 new_flow
= skb_get_rxhash(skb
) & (fl
->num_buckets
- 1);
3158 old_flow
= fl
->history
[fl
->history_head
];
3159 fl
->history
[fl
->history_head
] = new_flow
;
3162 fl
->history_head
&= FLOW_LIMIT_HISTORY
- 1;
3164 if (likely(fl
->buckets
[old_flow
]))
3165 fl
->buckets
[old_flow
]--;
3167 if (++fl
->buckets
[new_flow
] > (FLOW_LIMIT_HISTORY
>> 1)) {
3179 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3180 * queue (may be a remote CPU queue).
3182 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
3183 unsigned int *qtail
)
3185 struct softnet_data
*sd
;
3186 unsigned long flags
;
3189 sd
= &per_cpu(softnet_data
, cpu
);
3191 local_irq_save(flags
);
3194 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3195 if (qlen
<= netdev_max_backlog
&& !skb_flow_limit(skb
, qlen
)) {
3196 if (skb_queue_len(&sd
->input_pkt_queue
)) {
3198 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
3199 input_queue_tail_incr_save(sd
, qtail
);
3201 local_irq_restore(flags
);
3202 return NET_RX_SUCCESS
;
3205 /* Schedule NAPI for backlog device
3206 * We can use non atomic operation since we own the queue lock
3208 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
3209 if (!rps_ipi_queued(sd
))
3210 ____napi_schedule(sd
, &sd
->backlog
);
3218 local_irq_restore(flags
);
3220 atomic_long_inc(&skb
->dev
->rx_dropped
);
3226 * netif_rx - post buffer to the network code
3227 * @skb: buffer to post
3229 * This function receives a packet from a device driver and queues it for
3230 * the upper (protocol) levels to process. It always succeeds. The buffer
3231 * may be dropped during processing for congestion control or by the
3235 * NET_RX_SUCCESS (no congestion)
3236 * NET_RX_DROP (packet was dropped)
3240 int netif_rx(struct sk_buff
*skb
)
3244 /* if netpoll wants it, pretend we never saw it */
3245 if (netpoll_rx(skb
))
3248 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3250 trace_netif_rx(skb
);
3252 if (static_key_false(&rps_needed
)) {
3253 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3259 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3261 cpu
= smp_processor_id();
3263 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3271 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
3276 EXPORT_SYMBOL(netif_rx
);
3278 int netif_rx_ni(struct sk_buff
*skb
)
3283 err
= netif_rx(skb
);
3284 if (local_softirq_pending())
3290 EXPORT_SYMBOL(netif_rx_ni
);
3292 static void net_tx_action(struct softirq_action
*h
)
3294 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3296 if (sd
->completion_queue
) {
3297 struct sk_buff
*clist
;
3299 local_irq_disable();
3300 clist
= sd
->completion_queue
;
3301 sd
->completion_queue
= NULL
;
3305 struct sk_buff
*skb
= clist
;
3306 clist
= clist
->next
;
3308 WARN_ON(atomic_read(&skb
->users
));
3309 trace_kfree_skb(skb
, net_tx_action
);
3314 if (sd
->output_queue
) {
3317 local_irq_disable();
3318 head
= sd
->output_queue
;
3319 sd
->output_queue
= NULL
;
3320 sd
->output_queue_tailp
= &sd
->output_queue
;
3324 struct Qdisc
*q
= head
;
3325 spinlock_t
*root_lock
;
3327 head
= head
->next_sched
;
3329 root_lock
= qdisc_lock(q
);
3330 if (spin_trylock(root_lock
)) {
3331 smp_mb__before_clear_bit();
3332 clear_bit(__QDISC_STATE_SCHED
,
3335 spin_unlock(root_lock
);
3337 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
3339 __netif_reschedule(q
);
3341 smp_mb__before_clear_bit();
3342 clear_bit(__QDISC_STATE_SCHED
,
3350 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3351 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3352 /* This hook is defined here for ATM LANE */
3353 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
3354 unsigned char *addr
) __read_mostly
;
3355 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
3358 #ifdef CONFIG_NET_CLS_ACT
3359 /* TODO: Maybe we should just force sch_ingress to be compiled in
3360 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3361 * a compare and 2 stores extra right now if we dont have it on
3362 * but have CONFIG_NET_CLS_ACT
3363 * NOTE: This doesn't stop any functionality; if you dont have
3364 * the ingress scheduler, you just can't add policies on ingress.
3367 static int ing_filter(struct sk_buff
*skb
, struct netdev_queue
*rxq
)
3369 struct net_device
*dev
= skb
->dev
;
3370 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
3371 int result
= TC_ACT_OK
;
3374 if (unlikely(MAX_RED_LOOP
< ttl
++)) {
3375 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3376 skb
->skb_iif
, dev
->ifindex
);
3380 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
3381 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
3384 if (q
!= &noop_qdisc
) {
3385 spin_lock(qdisc_lock(q
));
3386 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
3387 result
= qdisc_enqueue_root(skb
, q
);
3388 spin_unlock(qdisc_lock(q
));
3394 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
3395 struct packet_type
**pt_prev
,
3396 int *ret
, struct net_device
*orig_dev
)
3398 struct netdev_queue
*rxq
= rcu_dereference(skb
->dev
->ingress_queue
);
3400 if (!rxq
|| rxq
->qdisc
== &noop_qdisc
)
3404 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
3408 switch (ing_filter(skb
, rxq
)) {
3422 * netdev_rx_handler_register - register receive handler
3423 * @dev: device to register a handler for
3424 * @rx_handler: receive handler to register
3425 * @rx_handler_data: data pointer that is used by rx handler
3427 * Register a receive hander for a device. This handler will then be
3428 * called from __netif_receive_skb. A negative errno code is returned
3431 * The caller must hold the rtnl_mutex.
3433 * For a general description of rx_handler, see enum rx_handler_result.
3435 int netdev_rx_handler_register(struct net_device
*dev
,
3436 rx_handler_func_t
*rx_handler
,
3437 void *rx_handler_data
)
3441 if (dev
->rx_handler
)
3444 /* Note: rx_handler_data must be set before rx_handler */
3445 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
3446 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
3450 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
3453 * netdev_rx_handler_unregister - unregister receive handler
3454 * @dev: device to unregister a handler from
3456 * Unregister a receive handler from a device.
3458 * The caller must hold the rtnl_mutex.
3460 void netdev_rx_handler_unregister(struct net_device
*dev
)
3464 RCU_INIT_POINTER(dev
->rx_handler
, NULL
);
3465 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3466 * section has a guarantee to see a non NULL rx_handler_data
3470 RCU_INIT_POINTER(dev
->rx_handler_data
, NULL
);
3472 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
3475 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3476 * the special handling of PFMEMALLOC skbs.
3478 static bool skb_pfmemalloc_protocol(struct sk_buff
*skb
)
3480 switch (skb
->protocol
) {
3481 case __constant_htons(ETH_P_ARP
):
3482 case __constant_htons(ETH_P_IP
):
3483 case __constant_htons(ETH_P_IPV6
):
3484 case __constant_htons(ETH_P_8021Q
):
3485 case __constant_htons(ETH_P_8021AD
):
3492 static int __netif_receive_skb_core(struct sk_buff
*skb
, bool pfmemalloc
)
3494 struct packet_type
*ptype
, *pt_prev
;
3495 rx_handler_func_t
*rx_handler
;
3496 struct net_device
*orig_dev
;
3497 struct net_device
*null_or_dev
;
3498 bool deliver_exact
= false;
3499 int ret
= NET_RX_DROP
;
3502 net_timestamp_check(!netdev_tstamp_prequeue
, skb
);
3504 trace_netif_receive_skb(skb
);
3506 /* if we've gotten here through NAPI, check netpoll */
3507 if (netpoll_receive_skb(skb
))
3510 orig_dev
= skb
->dev
;
3512 skb_reset_network_header(skb
);
3513 if (!skb_transport_header_was_set(skb
))
3514 skb_reset_transport_header(skb
);
3515 skb_reset_mac_len(skb
);
3522 skb
->skb_iif
= skb
->dev
->ifindex
;
3524 __this_cpu_inc(softnet_data
.processed
);
3526 if (skb
->protocol
== cpu_to_be16(ETH_P_8021Q
) ||
3527 skb
->protocol
== cpu_to_be16(ETH_P_8021AD
)) {
3528 skb
= vlan_untag(skb
);
3533 #ifdef CONFIG_NET_CLS_ACT
3534 if (skb
->tc_verd
& TC_NCLS
) {
3535 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
3543 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
3544 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
3546 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3552 #ifdef CONFIG_NET_CLS_ACT
3553 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
3559 if (pfmemalloc
&& !skb_pfmemalloc_protocol(skb
))
3562 if (vlan_tx_tag_present(skb
)) {
3564 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3567 if (vlan_do_receive(&skb
))
3569 else if (unlikely(!skb
))
3573 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
3576 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3579 switch (rx_handler(&skb
)) {
3580 case RX_HANDLER_CONSUMED
:
3581 ret
= NET_RX_SUCCESS
;
3583 case RX_HANDLER_ANOTHER
:
3585 case RX_HANDLER_EXACT
:
3586 deliver_exact
= true;
3587 case RX_HANDLER_PASS
:
3594 if (unlikely(vlan_tx_tag_present(skb
))) {
3595 if (vlan_tx_tag_get_id(skb
))
3596 skb
->pkt_type
= PACKET_OTHERHOST
;
3597 /* Note: we might in the future use prio bits
3598 * and set skb->priority like in vlan_do_receive()
3599 * For the time being, just ignore Priority Code Point
3604 /* deliver only exact match when indicated */
3605 null_or_dev
= deliver_exact
? skb
->dev
: NULL
;
3607 type
= skb
->protocol
;
3608 list_for_each_entry_rcu(ptype
,
3609 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
3610 if (ptype
->type
== type
&&
3611 (ptype
->dev
== null_or_dev
|| ptype
->dev
== skb
->dev
||
3612 ptype
->dev
== orig_dev
)) {
3614 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3620 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
3623 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
3626 atomic_long_inc(&skb
->dev
->rx_dropped
);
3628 /* Jamal, now you will not able to escape explaining
3629 * me how you were going to use this. :-)
3640 static int __netif_receive_skb(struct sk_buff
*skb
)
3644 if (sk_memalloc_socks() && skb_pfmemalloc(skb
)) {
3645 unsigned long pflags
= current
->flags
;
3648 * PFMEMALLOC skbs are special, they should
3649 * - be delivered to SOCK_MEMALLOC sockets only
3650 * - stay away from userspace
3651 * - have bounded memory usage
3653 * Use PF_MEMALLOC as this saves us from propagating the allocation
3654 * context down to all allocation sites.
3656 current
->flags
|= PF_MEMALLOC
;
3657 ret
= __netif_receive_skb_core(skb
, true);
3658 tsk_restore_flags(current
, pflags
, PF_MEMALLOC
);
3660 ret
= __netif_receive_skb_core(skb
, false);
3666 * netif_receive_skb - process receive buffer from network
3667 * @skb: buffer to process
3669 * netif_receive_skb() is the main receive data processing function.
3670 * It always succeeds. The buffer may be dropped during processing
3671 * for congestion control or by the protocol layers.
3673 * This function may only be called from softirq context and interrupts
3674 * should be enabled.
3676 * Return values (usually ignored):
3677 * NET_RX_SUCCESS: no congestion
3678 * NET_RX_DROP: packet was dropped
3680 int netif_receive_skb(struct sk_buff
*skb
)
3682 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3684 if (skb_defer_rx_timestamp(skb
))
3685 return NET_RX_SUCCESS
;
3688 if (static_key_false(&rps_needed
)) {
3689 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3694 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3697 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3704 return __netif_receive_skb(skb
);
3706 EXPORT_SYMBOL(netif_receive_skb
);
3708 /* Network device is going away, flush any packets still pending
3709 * Called with irqs disabled.
3711 static void flush_backlog(void *arg
)
3713 struct net_device
*dev
= arg
;
3714 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3715 struct sk_buff
*skb
, *tmp
;
3718 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3719 if (skb
->dev
== dev
) {
3720 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3722 input_queue_head_incr(sd
);
3727 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3728 if (skb
->dev
== dev
) {
3729 __skb_unlink(skb
, &sd
->process_queue
);
3731 input_queue_head_incr(sd
);
3736 static int napi_gro_complete(struct sk_buff
*skb
)
3738 struct packet_offload
*ptype
;
3739 __be16 type
= skb
->protocol
;
3740 struct list_head
*head
= &offload_base
;
3743 BUILD_BUG_ON(sizeof(struct napi_gro_cb
) > sizeof(skb
->cb
));
3745 if (NAPI_GRO_CB(skb
)->count
== 1) {
3746 skb_shinfo(skb
)->gso_size
= 0;
3751 list_for_each_entry_rcu(ptype
, head
, list
) {
3752 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_complete
)
3755 err
= ptype
->callbacks
.gro_complete(skb
);
3761 WARN_ON(&ptype
->list
== head
);
3763 return NET_RX_SUCCESS
;
3767 return netif_receive_skb(skb
);
3770 /* napi->gro_list contains packets ordered by age.
3771 * youngest packets at the head of it.
3772 * Complete skbs in reverse order to reduce latencies.
3774 void napi_gro_flush(struct napi_struct
*napi
, bool flush_old
)
3776 struct sk_buff
*skb
, *prev
= NULL
;
3778 /* scan list and build reverse chain */
3779 for (skb
= napi
->gro_list
; skb
!= NULL
; skb
= skb
->next
) {
3784 for (skb
= prev
; skb
; skb
= prev
) {
3787 if (flush_old
&& NAPI_GRO_CB(skb
)->age
== jiffies
)
3791 napi_gro_complete(skb
);
3795 napi
->gro_list
= NULL
;
3797 EXPORT_SYMBOL(napi_gro_flush
);
3799 static void gro_list_prepare(struct napi_struct
*napi
, struct sk_buff
*skb
)
3802 unsigned int maclen
= skb
->dev
->hard_header_len
;
3804 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3805 unsigned long diffs
;
3807 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
3808 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
3809 if (maclen
== ETH_HLEN
)
3810 diffs
|= compare_ether_header(skb_mac_header(p
),
3811 skb_gro_mac_header(skb
));
3813 diffs
= memcmp(skb_mac_header(p
),
3814 skb_gro_mac_header(skb
),
3816 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
3817 NAPI_GRO_CB(p
)->flush
= 0;
3821 static enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3823 struct sk_buff
**pp
= NULL
;
3824 struct packet_offload
*ptype
;
3825 __be16 type
= skb
->protocol
;
3826 struct list_head
*head
= &offload_base
;
3828 enum gro_result ret
;
3830 if (!(skb
->dev
->features
& NETIF_F_GRO
) || netpoll_rx_on(skb
))
3833 if (skb_is_gso(skb
) || skb_has_frag_list(skb
))
3836 gro_list_prepare(napi
, skb
);
3839 list_for_each_entry_rcu(ptype
, head
, list
) {
3840 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_receive
)
3843 skb_set_network_header(skb
, skb_gro_offset(skb
));
3844 skb_reset_mac_len(skb
);
3845 NAPI_GRO_CB(skb
)->same_flow
= 0;
3846 NAPI_GRO_CB(skb
)->flush
= 0;
3847 NAPI_GRO_CB(skb
)->free
= 0;
3849 pp
= ptype
->callbacks
.gro_receive(&napi
->gro_list
, skb
);
3854 if (&ptype
->list
== head
)
3857 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3858 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3861 struct sk_buff
*nskb
= *pp
;
3865 napi_gro_complete(nskb
);
3872 if (NAPI_GRO_CB(skb
)->flush
|| napi
->gro_count
>= MAX_GRO_SKBS
)
3876 NAPI_GRO_CB(skb
)->count
= 1;
3877 NAPI_GRO_CB(skb
)->age
= jiffies
;
3878 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
3879 skb
->next
= napi
->gro_list
;
3880 napi
->gro_list
= skb
;
3884 if (skb_headlen(skb
) < skb_gro_offset(skb
)) {
3885 int grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
3887 BUG_ON(skb
->end
- skb
->tail
< grow
);
3889 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3892 skb
->data_len
-= grow
;
3894 skb_shinfo(skb
)->frags
[0].page_offset
+= grow
;
3895 skb_frag_size_sub(&skb_shinfo(skb
)->frags
[0], grow
);
3897 if (unlikely(!skb_frag_size(&skb_shinfo(skb
)->frags
[0]))) {
3898 skb_frag_unref(skb
, 0);
3899 memmove(skb_shinfo(skb
)->frags
,
3900 skb_shinfo(skb
)->frags
+ 1,
3901 --skb_shinfo(skb
)->nr_frags
* sizeof(skb_frag_t
));
3914 static gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
3918 if (netif_receive_skb(skb
))
3926 case GRO_MERGED_FREE
:
3927 if (NAPI_GRO_CB(skb
)->free
== NAPI_GRO_FREE_STOLEN_HEAD
)
3928 kmem_cache_free(skbuff_head_cache
, skb
);
3941 static void skb_gro_reset_offset(struct sk_buff
*skb
)
3943 const struct skb_shared_info
*pinfo
= skb_shinfo(skb
);
3944 const skb_frag_t
*frag0
= &pinfo
->frags
[0];
3946 NAPI_GRO_CB(skb
)->data_offset
= 0;
3947 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3948 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3950 if (skb_mac_header(skb
) == skb_tail_pointer(skb
) &&
3952 !PageHighMem(skb_frag_page(frag0
))) {
3953 NAPI_GRO_CB(skb
)->frag0
= skb_frag_address(frag0
);
3954 NAPI_GRO_CB(skb
)->frag0_len
= skb_frag_size(frag0
);
3958 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3960 skb_gro_reset_offset(skb
);
3962 return napi_skb_finish(dev_gro_receive(napi
, skb
), skb
);
3964 EXPORT_SYMBOL(napi_gro_receive
);
3966 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
3968 __skb_pull(skb
, skb_headlen(skb
));
3969 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3970 skb_reserve(skb
, NET_SKB_PAD
+ NET_IP_ALIGN
- skb_headroom(skb
));
3972 skb
->dev
= napi
->dev
;
3978 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
3980 struct sk_buff
*skb
= napi
->skb
;
3983 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
3988 EXPORT_SYMBOL(napi_get_frags
);
3990 static gro_result_t
napi_frags_finish(struct napi_struct
*napi
, struct sk_buff
*skb
,
3996 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
3998 if (ret
== GRO_HELD
)
3999 skb_gro_pull(skb
, -ETH_HLEN
);
4000 else if (netif_receive_skb(skb
))
4005 case GRO_MERGED_FREE
:
4006 napi_reuse_skb(napi
, skb
);
4016 static struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
4018 struct sk_buff
*skb
= napi
->skb
;
4025 skb_reset_mac_header(skb
);
4026 skb_gro_reset_offset(skb
);
4028 off
= skb_gro_offset(skb
);
4029 hlen
= off
+ sizeof(*eth
);
4030 eth
= skb_gro_header_fast(skb
, off
);
4031 if (skb_gro_header_hard(skb
, hlen
)) {
4032 eth
= skb_gro_header_slow(skb
, hlen
, off
);
4033 if (unlikely(!eth
)) {
4034 napi_reuse_skb(napi
, skb
);
4040 skb_gro_pull(skb
, sizeof(*eth
));
4043 * This works because the only protocols we care about don't require
4044 * special handling. We'll fix it up properly at the end.
4046 skb
->protocol
= eth
->h_proto
;
4052 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
4054 struct sk_buff
*skb
= napi_frags_skb(napi
);
4059 return napi_frags_finish(napi
, skb
, dev_gro_receive(napi
, skb
));
4061 EXPORT_SYMBOL(napi_gro_frags
);
4064 * net_rps_action sends any pending IPI's for rps.
4065 * Note: called with local irq disabled, but exits with local irq enabled.
4067 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
4070 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
4073 sd
->rps_ipi_list
= NULL
;
4077 /* Send pending IPI's to kick RPS processing on remote cpus. */
4079 struct softnet_data
*next
= remsd
->rps_ipi_next
;
4081 if (cpu_online(remsd
->cpu
))
4082 __smp_call_function_single(remsd
->cpu
,
4091 static int process_backlog(struct napi_struct
*napi
, int quota
)
4094 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
4097 /* Check if we have pending ipi, its better to send them now,
4098 * not waiting net_rx_action() end.
4100 if (sd
->rps_ipi_list
) {
4101 local_irq_disable();
4102 net_rps_action_and_irq_enable(sd
);
4105 napi
->weight
= weight_p
;
4106 local_irq_disable();
4107 while (work
< quota
) {
4108 struct sk_buff
*skb
;
4111 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
4113 __netif_receive_skb(skb
);
4114 local_irq_disable();
4115 input_queue_head_incr(sd
);
4116 if (++work
>= quota
) {
4123 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
4125 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
4126 &sd
->process_queue
);
4128 if (qlen
< quota
- work
) {
4130 * Inline a custom version of __napi_complete().
4131 * only current cpu owns and manipulates this napi,
4132 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
4133 * we can use a plain write instead of clear_bit(),
4134 * and we dont need an smp_mb() memory barrier.
4136 list_del(&napi
->poll_list
);
4139 quota
= work
+ qlen
;
4149 * __napi_schedule - schedule for receive
4150 * @n: entry to schedule
4152 * The entry's receive function will be scheduled to run
4154 void __napi_schedule(struct napi_struct
*n
)
4156 unsigned long flags
;
4158 local_irq_save(flags
);
4159 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
4160 local_irq_restore(flags
);
4162 EXPORT_SYMBOL(__napi_schedule
);
4164 void __napi_complete(struct napi_struct
*n
)
4166 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
4167 BUG_ON(n
->gro_list
);
4169 list_del(&n
->poll_list
);
4170 smp_mb__before_clear_bit();
4171 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
4173 EXPORT_SYMBOL(__napi_complete
);
4175 void napi_complete(struct napi_struct
*n
)
4177 unsigned long flags
;
4180 * don't let napi dequeue from the cpu poll list
4181 * just in case its running on a different cpu
4183 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
4186 napi_gro_flush(n
, false);
4187 local_irq_save(flags
);
4189 local_irq_restore(flags
);
4191 EXPORT_SYMBOL(napi_complete
);
4193 /* must be called under rcu_read_lock(), as we dont take a reference */
4194 struct napi_struct
*napi_by_id(unsigned int napi_id
)
4196 unsigned int hash
= napi_id
% HASH_SIZE(napi_hash
);
4197 struct napi_struct
*napi
;
4199 hlist_for_each_entry_rcu(napi
, &napi_hash
[hash
], napi_hash_node
)
4200 if (napi
->napi_id
== napi_id
)
4205 EXPORT_SYMBOL_GPL(napi_by_id
);
4207 void napi_hash_add(struct napi_struct
*napi
)
4209 if (!test_and_set_bit(NAPI_STATE_HASHED
, &napi
->state
)) {
4211 spin_lock(&napi_hash_lock
);
4213 /* 0 is not a valid id, we also skip an id that is taken
4214 * we expect both events to be extremely rare
4217 while (!napi
->napi_id
) {
4218 napi
->napi_id
= ++napi_gen_id
;
4219 if (napi_by_id(napi
->napi_id
))
4223 hlist_add_head_rcu(&napi
->napi_hash_node
,
4224 &napi_hash
[napi
->napi_id
% HASH_SIZE(napi_hash
)]);
4226 spin_unlock(&napi_hash_lock
);
4229 EXPORT_SYMBOL_GPL(napi_hash_add
);
4231 /* Warning : caller is responsible to make sure rcu grace period
4232 * is respected before freeing memory containing @napi
4234 void napi_hash_del(struct napi_struct
*napi
)
4236 spin_lock(&napi_hash_lock
);
4238 if (test_and_clear_bit(NAPI_STATE_HASHED
, &napi
->state
))
4239 hlist_del_rcu(&napi
->napi_hash_node
);
4241 spin_unlock(&napi_hash_lock
);
4243 EXPORT_SYMBOL_GPL(napi_hash_del
);
4245 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
4246 int (*poll
)(struct napi_struct
*, int), int weight
)
4248 INIT_LIST_HEAD(&napi
->poll_list
);
4249 napi
->gro_count
= 0;
4250 napi
->gro_list
= NULL
;
4253 if (weight
> NAPI_POLL_WEIGHT
)
4254 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4256 napi
->weight
= weight
;
4257 list_add(&napi
->dev_list
, &dev
->napi_list
);
4259 #ifdef CONFIG_NETPOLL
4260 spin_lock_init(&napi
->poll_lock
);
4261 napi
->poll_owner
= -1;
4263 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
4265 EXPORT_SYMBOL(netif_napi_add
);
4267 void netif_napi_del(struct napi_struct
*napi
)
4269 struct sk_buff
*skb
, *next
;
4271 list_del_init(&napi
->dev_list
);
4272 napi_free_frags(napi
);
4274 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
4280 napi
->gro_list
= NULL
;
4281 napi
->gro_count
= 0;
4283 EXPORT_SYMBOL(netif_napi_del
);
4285 static void net_rx_action(struct softirq_action
*h
)
4287 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
4288 unsigned long time_limit
= jiffies
+ 2;
4289 int budget
= netdev_budget
;
4292 local_irq_disable();
4294 while (!list_empty(&sd
->poll_list
)) {
4295 struct napi_struct
*n
;
4298 /* If softirq window is exhuasted then punt.
4299 * Allow this to run for 2 jiffies since which will allow
4300 * an average latency of 1.5/HZ.
4302 if (unlikely(budget
<= 0 || time_after_eq(jiffies
, time_limit
)))
4307 /* Even though interrupts have been re-enabled, this
4308 * access is safe because interrupts can only add new
4309 * entries to the tail of this list, and only ->poll()
4310 * calls can remove this head entry from the list.
4312 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
4314 have
= netpoll_poll_lock(n
);
4318 /* This NAPI_STATE_SCHED test is for avoiding a race
4319 * with netpoll's poll_napi(). Only the entity which
4320 * obtains the lock and sees NAPI_STATE_SCHED set will
4321 * actually make the ->poll() call. Therefore we avoid
4322 * accidentally calling ->poll() when NAPI is not scheduled.
4325 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
4326 work
= n
->poll(n
, weight
);
4330 WARN_ON_ONCE(work
> weight
);
4334 local_irq_disable();
4336 /* Drivers must not modify the NAPI state if they
4337 * consume the entire weight. In such cases this code
4338 * still "owns" the NAPI instance and therefore can
4339 * move the instance around on the list at-will.
4341 if (unlikely(work
== weight
)) {
4342 if (unlikely(napi_disable_pending(n
))) {
4345 local_irq_disable();
4348 /* flush too old packets
4349 * If HZ < 1000, flush all packets.
4352 napi_gro_flush(n
, HZ
>= 1000);
4353 local_irq_disable();
4355 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
4359 netpoll_poll_unlock(have
);
4362 net_rps_action_and_irq_enable(sd
);
4364 #ifdef CONFIG_NET_DMA
4366 * There may not be any more sk_buffs coming right now, so push
4367 * any pending DMA copies to hardware
4369 dma_issue_pending_all();
4376 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
4380 struct netdev_adjacent
{
4381 struct net_device
*dev
;
4383 /* upper master flag, there can only be one master device per list */
4386 /* counter for the number of times this device was added to us */
4389 /* private field for the users */
4392 struct list_head list
;
4393 struct rcu_head rcu
;
4396 static struct netdev_adjacent
*__netdev_find_adj_rcu(struct net_device
*dev
,
4397 struct net_device
*adj_dev
,
4398 struct list_head
*adj_list
)
4400 struct netdev_adjacent
*adj
;
4402 list_for_each_entry_rcu(adj
, adj_list
, list
) {
4403 if (adj
->dev
== adj_dev
)
4409 static struct netdev_adjacent
*__netdev_find_adj(struct net_device
*dev
,
4410 struct net_device
*adj_dev
,
4411 struct list_head
*adj_list
)
4413 struct netdev_adjacent
*adj
;
4415 list_for_each_entry(adj
, adj_list
, list
) {
4416 if (adj
->dev
== adj_dev
)
4423 * netdev_has_upper_dev - Check if device is linked to an upper device
4425 * @upper_dev: upper device to check
4427 * Find out if a device is linked to specified upper device and return true
4428 * in case it is. Note that this checks only immediate upper device,
4429 * not through a complete stack of devices. The caller must hold the RTNL lock.
4431 bool netdev_has_upper_dev(struct net_device
*dev
,
4432 struct net_device
*upper_dev
)
4436 return __netdev_find_adj(dev
, upper_dev
, &dev
->all_adj_list
.upper
);
4438 EXPORT_SYMBOL(netdev_has_upper_dev
);
4441 * netdev_has_any_upper_dev - Check if device is linked to some device
4444 * Find out if a device is linked to an upper device and return true in case
4445 * it is. The caller must hold the RTNL lock.
4447 bool netdev_has_any_upper_dev(struct net_device
*dev
)
4451 return !list_empty(&dev
->all_adj_list
.upper
);
4453 EXPORT_SYMBOL(netdev_has_any_upper_dev
);
4456 * netdev_master_upper_dev_get - Get master upper device
4459 * Find a master upper device and return pointer to it or NULL in case
4460 * it's not there. The caller must hold the RTNL lock.
4462 struct net_device
*netdev_master_upper_dev_get(struct net_device
*dev
)
4464 struct netdev_adjacent
*upper
;
4468 if (list_empty(&dev
->adj_list
.upper
))
4471 upper
= list_first_entry(&dev
->adj_list
.upper
,
4472 struct netdev_adjacent
, list
);
4473 if (likely(upper
->master
))
4477 EXPORT_SYMBOL(netdev_master_upper_dev_get
);
4479 void *netdev_adjacent_get_private(struct list_head
*adj_list
)
4481 struct netdev_adjacent
*adj
;
4483 adj
= list_entry(adj_list
, struct netdev_adjacent
, list
);
4485 return adj
->private;
4487 EXPORT_SYMBOL(netdev_adjacent_get_private
);
4490 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4492 * @iter: list_head ** of the current position
4494 * Gets the next device from the dev's upper list, starting from iter
4495 * position. The caller must hold RCU read lock.
4497 struct net_device
*netdev_all_upper_get_next_dev_rcu(struct net_device
*dev
,
4498 struct list_head
**iter
)
4500 struct netdev_adjacent
*upper
;
4502 WARN_ON_ONCE(!rcu_read_lock_held());
4504 upper
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
4506 if (&upper
->list
== &dev
->all_adj_list
.upper
)
4509 *iter
= &upper
->list
;
4513 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu
);
4516 * netdev_lower_get_next_private - Get the next ->private from the
4517 * lower neighbour list
4519 * @iter: list_head ** of the current position
4521 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4522 * list, starting from iter position. The caller must hold either hold the
4523 * RTNL lock or its own locking that guarantees that the neighbour lower
4524 * list will remain unchainged.
4526 void *netdev_lower_get_next_private(struct net_device
*dev
,
4527 struct list_head
**iter
)
4529 struct netdev_adjacent
*lower
;
4531 lower
= list_entry(*iter
, struct netdev_adjacent
, list
);
4533 if (&lower
->list
== &dev
->adj_list
.lower
)
4537 *iter
= lower
->list
.next
;
4539 return lower
->private;
4541 EXPORT_SYMBOL(netdev_lower_get_next_private
);
4544 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4545 * lower neighbour list, RCU
4548 * @iter: list_head ** of the current position
4550 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4551 * list, starting from iter position. The caller must hold RCU read lock.
4553 void *netdev_lower_get_next_private_rcu(struct net_device
*dev
,
4554 struct list_head
**iter
)
4556 struct netdev_adjacent
*lower
;
4558 WARN_ON_ONCE(!rcu_read_lock_held());
4560 lower
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
4562 if (&lower
->list
== &dev
->adj_list
.lower
)
4566 *iter
= &lower
->list
;
4568 return lower
->private;
4570 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu
);
4573 * netdev_master_upper_dev_get_rcu - Get master upper device
4576 * Find a master upper device and return pointer to it or NULL in case
4577 * it's not there. The caller must hold the RCU read lock.
4579 struct net_device
*netdev_master_upper_dev_get_rcu(struct net_device
*dev
)
4581 struct netdev_adjacent
*upper
;
4583 upper
= list_first_or_null_rcu(&dev
->adj_list
.upper
,
4584 struct netdev_adjacent
, list
);
4585 if (upper
&& likely(upper
->master
))
4589 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu
);
4591 static int __netdev_adjacent_dev_insert(struct net_device
*dev
,
4592 struct net_device
*adj_dev
,
4593 struct list_head
*dev_list
,
4594 void *private, bool master
)
4596 struct netdev_adjacent
*adj
;
4597 char linkname
[IFNAMSIZ
+7];
4600 adj
= __netdev_find_adj(dev
, adj_dev
, dev_list
);
4607 adj
= kmalloc(sizeof(*adj
), GFP_KERNEL
);
4612 adj
->master
= master
;
4614 adj
->private = private;
4617 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
4618 adj_dev
->name
, dev
->name
, adj_dev
->name
);
4620 if (dev_list
== &dev
->adj_list
.lower
) {
4621 sprintf(linkname
, "lower_%s", adj_dev
->name
);
4622 ret
= sysfs_create_link(&(dev
->dev
.kobj
),
4623 &(adj_dev
->dev
.kobj
), linkname
);
4626 } else if (dev_list
== &dev
->adj_list
.upper
) {
4627 sprintf(linkname
, "upper_%s", adj_dev
->name
);
4628 ret
= sysfs_create_link(&(dev
->dev
.kobj
),
4629 &(adj_dev
->dev
.kobj
), linkname
);
4634 /* Ensure that master link is always the first item in list. */
4636 ret
= sysfs_create_link(&(dev
->dev
.kobj
),
4637 &(adj_dev
->dev
.kobj
), "master");
4639 goto remove_symlinks
;
4641 list_add_rcu(&adj
->list
, dev_list
);
4643 list_add_tail_rcu(&adj
->list
, dev_list
);
4649 if (dev_list
== &dev
->adj_list
.lower
) {
4650 sprintf(linkname
, "lower_%s", adj_dev
->name
);
4651 sysfs_remove_link(&(dev
->dev
.kobj
), linkname
);
4652 } else if (dev_list
== &dev
->adj_list
.upper
) {
4653 sprintf(linkname
, "upper_%s", adj_dev
->name
);
4654 sysfs_remove_link(&(dev
->dev
.kobj
), linkname
);
4664 void __netdev_adjacent_dev_remove(struct net_device
*dev
,
4665 struct net_device
*adj_dev
,
4666 struct list_head
*dev_list
)
4668 struct netdev_adjacent
*adj
;
4669 char linkname
[IFNAMSIZ
+7];
4671 adj
= __netdev_find_adj(dev
, adj_dev
, dev_list
);
4674 pr_err("tried to remove device %s from %s\n",
4675 dev
->name
, adj_dev
->name
);
4679 if (adj
->ref_nr
> 1) {
4680 pr_debug("%s to %s ref_nr-- = %d\n", dev
->name
, adj_dev
->name
,
4687 sysfs_remove_link(&(dev
->dev
.kobj
), "master");
4689 if (dev_list
== &dev
->adj_list
.lower
) {
4690 sprintf(linkname
, "lower_%s", adj_dev
->name
);
4691 sysfs_remove_link(&(dev
->dev
.kobj
), linkname
);
4692 } else if (dev_list
== &dev
->adj_list
.upper
) {
4693 sprintf(linkname
, "upper_%s", adj_dev
->name
);
4694 sysfs_remove_link(&(dev
->dev
.kobj
), linkname
);
4697 list_del_rcu(&adj
->list
);
4698 pr_debug("dev_put for %s, because link removed from %s to %s\n",
4699 adj_dev
->name
, dev
->name
, adj_dev
->name
);
4701 kfree_rcu(adj
, rcu
);
4704 int __netdev_adjacent_dev_link_lists(struct net_device
*dev
,
4705 struct net_device
*upper_dev
,
4706 struct list_head
*up_list
,
4707 struct list_head
*down_list
,
4708 void *private, bool master
)
4712 ret
= __netdev_adjacent_dev_insert(dev
, upper_dev
, up_list
, private,
4717 ret
= __netdev_adjacent_dev_insert(upper_dev
, dev
, down_list
, private,
4720 __netdev_adjacent_dev_remove(dev
, upper_dev
, up_list
);
4727 int __netdev_adjacent_dev_link(struct net_device
*dev
,
4728 struct net_device
*upper_dev
)
4730 return __netdev_adjacent_dev_link_lists(dev
, upper_dev
,
4731 &dev
->all_adj_list
.upper
,
4732 &upper_dev
->all_adj_list
.lower
,
4736 void __netdev_adjacent_dev_unlink_lists(struct net_device
*dev
,
4737 struct net_device
*upper_dev
,
4738 struct list_head
*up_list
,
4739 struct list_head
*down_list
)
4741 __netdev_adjacent_dev_remove(dev
, upper_dev
, up_list
);
4742 __netdev_adjacent_dev_remove(upper_dev
, dev
, down_list
);
4745 void __netdev_adjacent_dev_unlink(struct net_device
*dev
,
4746 struct net_device
*upper_dev
)
4748 __netdev_adjacent_dev_unlink_lists(dev
, upper_dev
,
4749 &dev
->all_adj_list
.upper
,
4750 &upper_dev
->all_adj_list
.lower
);
4753 int __netdev_adjacent_dev_link_neighbour(struct net_device
*dev
,
4754 struct net_device
*upper_dev
,
4755 void *private, bool master
)
4757 int ret
= __netdev_adjacent_dev_link(dev
, upper_dev
);
4762 ret
= __netdev_adjacent_dev_link_lists(dev
, upper_dev
,
4763 &dev
->adj_list
.upper
,
4764 &upper_dev
->adj_list
.lower
,
4767 __netdev_adjacent_dev_unlink(dev
, upper_dev
);
4774 void __netdev_adjacent_dev_unlink_neighbour(struct net_device
*dev
,
4775 struct net_device
*upper_dev
)
4777 __netdev_adjacent_dev_unlink(dev
, upper_dev
);
4778 __netdev_adjacent_dev_unlink_lists(dev
, upper_dev
,
4779 &dev
->adj_list
.upper
,
4780 &upper_dev
->adj_list
.lower
);
4783 static int __netdev_upper_dev_link(struct net_device
*dev
,
4784 struct net_device
*upper_dev
, bool master
,
4787 struct netdev_adjacent
*i
, *j
, *to_i
, *to_j
;
4792 if (dev
== upper_dev
)
4795 /* To prevent loops, check if dev is not upper device to upper_dev. */
4796 if (__netdev_find_adj(upper_dev
, dev
, &upper_dev
->all_adj_list
.upper
))
4799 if (__netdev_find_adj(dev
, upper_dev
, &dev
->all_adj_list
.upper
))
4802 if (master
&& netdev_master_upper_dev_get(dev
))
4805 ret
= __netdev_adjacent_dev_link_neighbour(dev
, upper_dev
, private,
4810 /* Now that we linked these devs, make all the upper_dev's
4811 * all_adj_list.upper visible to every dev's all_adj_list.lower an
4812 * versa, and don't forget the devices itself. All of these
4813 * links are non-neighbours.
4815 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
4816 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
) {
4817 pr_debug("Interlinking %s with %s, non-neighbour\n",
4818 i
->dev
->name
, j
->dev
->name
);
4819 ret
= __netdev_adjacent_dev_link(i
->dev
, j
->dev
);
4825 /* add dev to every upper_dev's upper device */
4826 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
) {
4827 pr_debug("linking %s's upper device %s with %s\n",
4828 upper_dev
->name
, i
->dev
->name
, dev
->name
);
4829 ret
= __netdev_adjacent_dev_link(dev
, i
->dev
);
4831 goto rollback_upper_mesh
;
4834 /* add upper_dev to every dev's lower device */
4835 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
4836 pr_debug("linking %s's lower device %s with %s\n", dev
->name
,
4837 i
->dev
->name
, upper_dev
->name
);
4838 ret
= __netdev_adjacent_dev_link(i
->dev
, upper_dev
);
4840 goto rollback_lower_mesh
;
4843 call_netdevice_notifiers(NETDEV_CHANGEUPPER
, dev
);
4846 rollback_lower_mesh
:
4848 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
4851 __netdev_adjacent_dev_unlink(i
->dev
, upper_dev
);
4856 rollback_upper_mesh
:
4858 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
) {
4861 __netdev_adjacent_dev_unlink(dev
, i
->dev
);
4869 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
4870 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
) {
4871 if (i
== to_i
&& j
== to_j
)
4873 __netdev_adjacent_dev_unlink(i
->dev
, j
->dev
);
4879 __netdev_adjacent_dev_unlink_neighbour(dev
, upper_dev
);
4885 * netdev_upper_dev_link - Add a link to the upper device
4887 * @upper_dev: new upper device
4889 * Adds a link to device which is upper to this one. The caller must hold
4890 * the RTNL lock. On a failure a negative errno code is returned.
4891 * On success the reference counts are adjusted and the function
4894 int netdev_upper_dev_link(struct net_device
*dev
,
4895 struct net_device
*upper_dev
)
4897 return __netdev_upper_dev_link(dev
, upper_dev
, false, NULL
);
4899 EXPORT_SYMBOL(netdev_upper_dev_link
);
4902 * netdev_master_upper_dev_link - Add a master link to the upper device
4904 * @upper_dev: new upper device
4906 * Adds a link to device which is upper to this one. In this case, only
4907 * one master upper device can be linked, although other non-master devices
4908 * might be linked as well. The caller must hold the RTNL lock.
4909 * On a failure a negative errno code is returned. On success the reference
4910 * counts are adjusted and the function returns zero.
4912 int netdev_master_upper_dev_link(struct net_device
*dev
,
4913 struct net_device
*upper_dev
)
4915 return __netdev_upper_dev_link(dev
, upper_dev
, true, NULL
);
4917 EXPORT_SYMBOL(netdev_master_upper_dev_link
);
4919 int netdev_master_upper_dev_link_private(struct net_device
*dev
,
4920 struct net_device
*upper_dev
,
4923 return __netdev_upper_dev_link(dev
, upper_dev
, true, private);
4925 EXPORT_SYMBOL(netdev_master_upper_dev_link_private
);
4928 * netdev_upper_dev_unlink - Removes a link to upper device
4930 * @upper_dev: new upper device
4932 * Removes a link to device which is upper to this one. The caller must hold
4935 void netdev_upper_dev_unlink(struct net_device
*dev
,
4936 struct net_device
*upper_dev
)
4938 struct netdev_adjacent
*i
, *j
;
4941 __netdev_adjacent_dev_unlink_neighbour(dev
, upper_dev
);
4943 /* Here is the tricky part. We must remove all dev's lower
4944 * devices from all upper_dev's upper devices and vice
4945 * versa, to maintain the graph relationship.
4947 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
)
4948 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
)
4949 __netdev_adjacent_dev_unlink(i
->dev
, j
->dev
);
4951 /* remove also the devices itself from lower/upper device
4954 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
)
4955 __netdev_adjacent_dev_unlink(i
->dev
, upper_dev
);
4957 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
)
4958 __netdev_adjacent_dev_unlink(dev
, i
->dev
);
4960 call_netdevice_notifiers(NETDEV_CHANGEUPPER
, dev
);
4962 EXPORT_SYMBOL(netdev_upper_dev_unlink
);
4964 void *netdev_lower_dev_get_private_rcu(struct net_device
*dev
,
4965 struct net_device
*lower_dev
)
4967 struct netdev_adjacent
*lower
;
4971 lower
= __netdev_find_adj_rcu(dev
, lower_dev
, &dev
->adj_list
.lower
);
4975 return lower
->private;
4977 EXPORT_SYMBOL(netdev_lower_dev_get_private_rcu
);
4979 void *netdev_lower_dev_get_private(struct net_device
*dev
,
4980 struct net_device
*lower_dev
)
4982 struct netdev_adjacent
*lower
;
4986 lower
= __netdev_find_adj(dev
, lower_dev
, &dev
->adj_list
.lower
);
4990 return lower
->private;
4992 EXPORT_SYMBOL(netdev_lower_dev_get_private
);
4994 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
4996 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4998 if (ops
->ndo_change_rx_flags
)
4999 ops
->ndo_change_rx_flags(dev
, flags
);
5002 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
, bool notify
)
5004 unsigned int old_flags
= dev
->flags
;
5010 dev
->flags
|= IFF_PROMISC
;
5011 dev
->promiscuity
+= inc
;
5012 if (dev
->promiscuity
== 0) {
5015 * If inc causes overflow, untouch promisc and return error.
5018 dev
->flags
&= ~IFF_PROMISC
;
5020 dev
->promiscuity
-= inc
;
5021 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5026 if (dev
->flags
!= old_flags
) {
5027 pr_info("device %s %s promiscuous mode\n",
5029 dev
->flags
& IFF_PROMISC
? "entered" : "left");
5030 if (audit_enabled
) {
5031 current_uid_gid(&uid
, &gid
);
5032 audit_log(current
->audit_context
, GFP_ATOMIC
,
5033 AUDIT_ANOM_PROMISCUOUS
,
5034 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5035 dev
->name
, (dev
->flags
& IFF_PROMISC
),
5036 (old_flags
& IFF_PROMISC
),
5037 from_kuid(&init_user_ns
, audit_get_loginuid(current
)),
5038 from_kuid(&init_user_ns
, uid
),
5039 from_kgid(&init_user_ns
, gid
),
5040 audit_get_sessionid(current
));
5043 dev_change_rx_flags(dev
, IFF_PROMISC
);
5046 __dev_notify_flags(dev
, old_flags
, IFF_PROMISC
);
5051 * dev_set_promiscuity - update promiscuity count on a device
5055 * Add or remove promiscuity from a device. While the count in the device
5056 * remains above zero the interface remains promiscuous. Once it hits zero
5057 * the device reverts back to normal filtering operation. A negative inc
5058 * value is used to drop promiscuity on the device.
5059 * Return 0 if successful or a negative errno code on error.
5061 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
5063 unsigned int old_flags
= dev
->flags
;
5066 err
= __dev_set_promiscuity(dev
, inc
, true);
5069 if (dev
->flags
!= old_flags
)
5070 dev_set_rx_mode(dev
);
5073 EXPORT_SYMBOL(dev_set_promiscuity
);
5075 static int __dev_set_allmulti(struct net_device
*dev
, int inc
, bool notify
)
5077 unsigned int old_flags
= dev
->flags
, old_gflags
= dev
->gflags
;
5081 dev
->flags
|= IFF_ALLMULTI
;
5082 dev
->allmulti
+= inc
;
5083 if (dev
->allmulti
== 0) {
5086 * If inc causes overflow, untouch allmulti and return error.
5089 dev
->flags
&= ~IFF_ALLMULTI
;
5091 dev
->allmulti
-= inc
;
5092 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5097 if (dev
->flags
^ old_flags
) {
5098 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
5099 dev_set_rx_mode(dev
);
5101 __dev_notify_flags(dev
, old_flags
,
5102 dev
->gflags
^ old_gflags
);
5108 * dev_set_allmulti - update allmulti count on a device
5112 * Add or remove reception of all multicast frames to a device. While the
5113 * count in the device remains above zero the interface remains listening
5114 * to all interfaces. Once it hits zero the device reverts back to normal
5115 * filtering operation. A negative @inc value is used to drop the counter
5116 * when releasing a resource needing all multicasts.
5117 * Return 0 if successful or a negative errno code on error.
5120 int dev_set_allmulti(struct net_device
*dev
, int inc
)
5122 return __dev_set_allmulti(dev
, inc
, true);
5124 EXPORT_SYMBOL(dev_set_allmulti
);
5127 * Upload unicast and multicast address lists to device and
5128 * configure RX filtering. When the device doesn't support unicast
5129 * filtering it is put in promiscuous mode while unicast addresses
5132 void __dev_set_rx_mode(struct net_device
*dev
)
5134 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5136 /* dev_open will call this function so the list will stay sane. */
5137 if (!(dev
->flags
&IFF_UP
))
5140 if (!netif_device_present(dev
))
5143 if (!(dev
->priv_flags
& IFF_UNICAST_FLT
)) {
5144 /* Unicast addresses changes may only happen under the rtnl,
5145 * therefore calling __dev_set_promiscuity here is safe.
5147 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
5148 __dev_set_promiscuity(dev
, 1, false);
5149 dev
->uc_promisc
= true;
5150 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
5151 __dev_set_promiscuity(dev
, -1, false);
5152 dev
->uc_promisc
= false;
5156 if (ops
->ndo_set_rx_mode
)
5157 ops
->ndo_set_rx_mode(dev
);
5160 void dev_set_rx_mode(struct net_device
*dev
)
5162 netif_addr_lock_bh(dev
);
5163 __dev_set_rx_mode(dev
);
5164 netif_addr_unlock_bh(dev
);
5168 * dev_get_flags - get flags reported to userspace
5171 * Get the combination of flag bits exported through APIs to userspace.
5173 unsigned int dev_get_flags(const struct net_device
*dev
)
5177 flags
= (dev
->flags
& ~(IFF_PROMISC
|
5182 (dev
->gflags
& (IFF_PROMISC
|
5185 if (netif_running(dev
)) {
5186 if (netif_oper_up(dev
))
5187 flags
|= IFF_RUNNING
;
5188 if (netif_carrier_ok(dev
))
5189 flags
|= IFF_LOWER_UP
;
5190 if (netif_dormant(dev
))
5191 flags
|= IFF_DORMANT
;
5196 EXPORT_SYMBOL(dev_get_flags
);
5198 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
5200 unsigned int old_flags
= dev
->flags
;
5206 * Set the flags on our device.
5209 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
5210 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
5212 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
5216 * Load in the correct multicast list now the flags have changed.
5219 if ((old_flags
^ flags
) & IFF_MULTICAST
)
5220 dev_change_rx_flags(dev
, IFF_MULTICAST
);
5222 dev_set_rx_mode(dev
);
5225 * Have we downed the interface. We handle IFF_UP ourselves
5226 * according to user attempts to set it, rather than blindly
5231 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
5232 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
5235 dev_set_rx_mode(dev
);
5238 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
5239 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
5240 unsigned int old_flags
= dev
->flags
;
5242 dev
->gflags
^= IFF_PROMISC
;
5244 if (__dev_set_promiscuity(dev
, inc
, false) >= 0)
5245 if (dev
->flags
!= old_flags
)
5246 dev_set_rx_mode(dev
);
5249 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5250 is important. Some (broken) drivers set IFF_PROMISC, when
5251 IFF_ALLMULTI is requested not asking us and not reporting.
5253 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
5254 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
5256 dev
->gflags
^= IFF_ALLMULTI
;
5257 __dev_set_allmulti(dev
, inc
, false);
5263 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
,
5264 unsigned int gchanges
)
5266 unsigned int changes
= dev
->flags
^ old_flags
;
5269 rtmsg_ifinfo(RTM_NEWLINK
, dev
, gchanges
, GFP_ATOMIC
);
5271 if (changes
& IFF_UP
) {
5272 if (dev
->flags
& IFF_UP
)
5273 call_netdevice_notifiers(NETDEV_UP
, dev
);
5275 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
5278 if (dev
->flags
& IFF_UP
&&
5279 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
))) {
5280 struct netdev_notifier_change_info change_info
;
5282 change_info
.flags_changed
= changes
;
5283 call_netdevice_notifiers_info(NETDEV_CHANGE
, dev
,
5289 * dev_change_flags - change device settings
5291 * @flags: device state flags
5293 * Change settings on device based state flags. The flags are
5294 * in the userspace exported format.
5296 int dev_change_flags(struct net_device
*dev
, unsigned int flags
)
5299 unsigned int changes
, old_flags
= dev
->flags
, old_gflags
= dev
->gflags
;
5301 ret
= __dev_change_flags(dev
, flags
);
5305 changes
= (old_flags
^ dev
->flags
) | (old_gflags
^ dev
->gflags
);
5306 __dev_notify_flags(dev
, old_flags
, changes
);
5309 EXPORT_SYMBOL(dev_change_flags
);
5312 * dev_set_mtu - Change maximum transfer unit
5314 * @new_mtu: new transfer unit
5316 * Change the maximum transfer size of the network device.
5318 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
5320 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5323 if (new_mtu
== dev
->mtu
)
5326 /* MTU must be positive. */
5330 if (!netif_device_present(dev
))
5334 if (ops
->ndo_change_mtu
)
5335 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
5340 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
5343 EXPORT_SYMBOL(dev_set_mtu
);
5346 * dev_set_group - Change group this device belongs to
5348 * @new_group: group this device should belong to
5350 void dev_set_group(struct net_device
*dev
, int new_group
)
5352 dev
->group
= new_group
;
5354 EXPORT_SYMBOL(dev_set_group
);
5357 * dev_set_mac_address - Change Media Access Control Address
5361 * Change the hardware (MAC) address of the device
5363 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
5365 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5368 if (!ops
->ndo_set_mac_address
)
5370 if (sa
->sa_family
!= dev
->type
)
5372 if (!netif_device_present(dev
))
5374 err
= ops
->ndo_set_mac_address(dev
, sa
);
5377 dev
->addr_assign_type
= NET_ADDR_SET
;
5378 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
5379 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
5382 EXPORT_SYMBOL(dev_set_mac_address
);
5385 * dev_change_carrier - Change device carrier
5387 * @new_carrier: new value
5389 * Change device carrier
5391 int dev_change_carrier(struct net_device
*dev
, bool new_carrier
)
5393 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5395 if (!ops
->ndo_change_carrier
)
5397 if (!netif_device_present(dev
))
5399 return ops
->ndo_change_carrier(dev
, new_carrier
);
5401 EXPORT_SYMBOL(dev_change_carrier
);
5404 * dev_get_phys_port_id - Get device physical port ID
5408 * Get device physical port ID
5410 int dev_get_phys_port_id(struct net_device
*dev
,
5411 struct netdev_phys_port_id
*ppid
)
5413 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5415 if (!ops
->ndo_get_phys_port_id
)
5417 return ops
->ndo_get_phys_port_id(dev
, ppid
);
5419 EXPORT_SYMBOL(dev_get_phys_port_id
);
5422 * dev_new_index - allocate an ifindex
5423 * @net: the applicable net namespace
5425 * Returns a suitable unique value for a new device interface
5426 * number. The caller must hold the rtnl semaphore or the
5427 * dev_base_lock to be sure it remains unique.
5429 static int dev_new_index(struct net
*net
)
5431 int ifindex
= net
->ifindex
;
5435 if (!__dev_get_by_index(net
, ifindex
))
5436 return net
->ifindex
= ifindex
;
5440 /* Delayed registration/unregisteration */
5441 static LIST_HEAD(net_todo_list
);
5442 static DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq
);
5444 static void net_set_todo(struct net_device
*dev
)
5446 list_add_tail(&dev
->todo_list
, &net_todo_list
);
5447 dev_net(dev
)->dev_unreg_count
++;
5450 static void rollback_registered_many(struct list_head
*head
)
5452 struct net_device
*dev
, *tmp
;
5453 LIST_HEAD(close_head
);
5455 BUG_ON(dev_boot_phase
);
5458 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
5459 /* Some devices call without registering
5460 * for initialization unwind. Remove those
5461 * devices and proceed with the remaining.
5463 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5464 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5468 list_del(&dev
->unreg_list
);
5471 dev
->dismantle
= true;
5472 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
5475 /* If device is running, close it first. */
5476 list_for_each_entry(dev
, head
, unreg_list
)
5477 list_add_tail(&dev
->close_list
, &close_head
);
5478 dev_close_many(&close_head
);
5480 list_for_each_entry(dev
, head
, unreg_list
) {
5481 /* And unlink it from device chain. */
5482 unlist_netdevice(dev
);
5484 dev
->reg_state
= NETREG_UNREGISTERING
;
5489 list_for_each_entry(dev
, head
, unreg_list
) {
5490 /* Shutdown queueing discipline. */
5494 /* Notify protocols, that we are about to destroy
5495 this device. They should clean all the things.
5497 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5499 if (!dev
->rtnl_link_ops
||
5500 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5501 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U, GFP_KERNEL
);
5504 * Flush the unicast and multicast chains
5509 if (dev
->netdev_ops
->ndo_uninit
)
5510 dev
->netdev_ops
->ndo_uninit(dev
);
5512 /* Notifier chain MUST detach us all upper devices. */
5513 WARN_ON(netdev_has_any_upper_dev(dev
));
5515 /* Remove entries from kobject tree */
5516 netdev_unregister_kobject(dev
);
5518 /* Remove XPS queueing entries */
5519 netif_reset_xps_queues_gt(dev
, 0);
5525 list_for_each_entry(dev
, head
, unreg_list
)
5529 static void rollback_registered(struct net_device
*dev
)
5533 list_add(&dev
->unreg_list
, &single
);
5534 rollback_registered_many(&single
);
5538 static netdev_features_t
netdev_fix_features(struct net_device
*dev
,
5539 netdev_features_t features
)
5541 /* Fix illegal checksum combinations */
5542 if ((features
& NETIF_F_HW_CSUM
) &&
5543 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5544 netdev_warn(dev
, "mixed HW and IP checksum settings.\n");
5545 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5548 /* TSO requires that SG is present as well. */
5549 if ((features
& NETIF_F_ALL_TSO
) && !(features
& NETIF_F_SG
)) {
5550 netdev_dbg(dev
, "Dropping TSO features since no SG feature.\n");
5551 features
&= ~NETIF_F_ALL_TSO
;
5554 if ((features
& NETIF_F_TSO
) && !(features
& NETIF_F_HW_CSUM
) &&
5555 !(features
& NETIF_F_IP_CSUM
)) {
5556 netdev_dbg(dev
, "Dropping TSO features since no CSUM feature.\n");
5557 features
&= ~NETIF_F_TSO
;
5558 features
&= ~NETIF_F_TSO_ECN
;
5561 if ((features
& NETIF_F_TSO6
) && !(features
& NETIF_F_HW_CSUM
) &&
5562 !(features
& NETIF_F_IPV6_CSUM
)) {
5563 netdev_dbg(dev
, "Dropping TSO6 features since no CSUM feature.\n");
5564 features
&= ~NETIF_F_TSO6
;
5567 /* TSO ECN requires that TSO is present as well. */
5568 if ((features
& NETIF_F_ALL_TSO
) == NETIF_F_TSO_ECN
)
5569 features
&= ~NETIF_F_TSO_ECN
;
5571 /* Software GSO depends on SG. */
5572 if ((features
& NETIF_F_GSO
) && !(features
& NETIF_F_SG
)) {
5573 netdev_dbg(dev
, "Dropping NETIF_F_GSO since no SG feature.\n");
5574 features
&= ~NETIF_F_GSO
;
5577 /* UFO needs SG and checksumming */
5578 if (features
& NETIF_F_UFO
) {
5579 /* maybe split UFO into V4 and V6? */
5580 if (!((features
& NETIF_F_GEN_CSUM
) ||
5581 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))
5582 == (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5584 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5585 features
&= ~NETIF_F_UFO
;
5588 if (!(features
& NETIF_F_SG
)) {
5590 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5591 features
&= ~NETIF_F_UFO
;
5598 int __netdev_update_features(struct net_device
*dev
)
5600 netdev_features_t features
;
5605 features
= netdev_get_wanted_features(dev
);
5607 if (dev
->netdev_ops
->ndo_fix_features
)
5608 features
= dev
->netdev_ops
->ndo_fix_features(dev
, features
);
5610 /* driver might be less strict about feature dependencies */
5611 features
= netdev_fix_features(dev
, features
);
5613 if (dev
->features
== features
)
5616 netdev_dbg(dev
, "Features changed: %pNF -> %pNF\n",
5617 &dev
->features
, &features
);
5619 if (dev
->netdev_ops
->ndo_set_features
)
5620 err
= dev
->netdev_ops
->ndo_set_features(dev
, features
);
5622 if (unlikely(err
< 0)) {
5624 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5625 err
, &features
, &dev
->features
);
5630 dev
->features
= features
;
5636 * netdev_update_features - recalculate device features
5637 * @dev: the device to check
5639 * Recalculate dev->features set and send notifications if it
5640 * has changed. Should be called after driver or hardware dependent
5641 * conditions might have changed that influence the features.
5643 void netdev_update_features(struct net_device
*dev
)
5645 if (__netdev_update_features(dev
))
5646 netdev_features_change(dev
);
5648 EXPORT_SYMBOL(netdev_update_features
);
5651 * netdev_change_features - recalculate device features
5652 * @dev: the device to check
5654 * Recalculate dev->features set and send notifications even
5655 * if they have not changed. Should be called instead of
5656 * netdev_update_features() if also dev->vlan_features might
5657 * have changed to allow the changes to be propagated to stacked
5660 void netdev_change_features(struct net_device
*dev
)
5662 __netdev_update_features(dev
);
5663 netdev_features_change(dev
);
5665 EXPORT_SYMBOL(netdev_change_features
);
5668 * netif_stacked_transfer_operstate - transfer operstate
5669 * @rootdev: the root or lower level device to transfer state from
5670 * @dev: the device to transfer operstate to
5672 * Transfer operational state from root to device. This is normally
5673 * called when a stacking relationship exists between the root
5674 * device and the device(a leaf device).
5676 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
5677 struct net_device
*dev
)
5679 if (rootdev
->operstate
== IF_OPER_DORMANT
)
5680 netif_dormant_on(dev
);
5682 netif_dormant_off(dev
);
5684 if (netif_carrier_ok(rootdev
)) {
5685 if (!netif_carrier_ok(dev
))
5686 netif_carrier_on(dev
);
5688 if (netif_carrier_ok(dev
))
5689 netif_carrier_off(dev
);
5692 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
5695 static int netif_alloc_rx_queues(struct net_device
*dev
)
5697 unsigned int i
, count
= dev
->num_rx_queues
;
5698 struct netdev_rx_queue
*rx
;
5702 rx
= kcalloc(count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5708 for (i
= 0; i
< count
; i
++)
5714 static void netdev_init_one_queue(struct net_device
*dev
,
5715 struct netdev_queue
*queue
, void *_unused
)
5717 /* Initialize queue lock */
5718 spin_lock_init(&queue
->_xmit_lock
);
5719 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
5720 queue
->xmit_lock_owner
= -1;
5721 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
5724 dql_init(&queue
->dql
, HZ
);
5728 static void netif_free_tx_queues(struct net_device
*dev
)
5730 if (is_vmalloc_addr(dev
->_tx
))
5736 static int netif_alloc_netdev_queues(struct net_device
*dev
)
5738 unsigned int count
= dev
->num_tx_queues
;
5739 struct netdev_queue
*tx
;
5740 size_t sz
= count
* sizeof(*tx
);
5742 BUG_ON(count
< 1 || count
> 0xffff);
5744 tx
= kzalloc(sz
, GFP_KERNEL
| __GFP_NOWARN
| __GFP_REPEAT
);
5752 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5753 spin_lock_init(&dev
->tx_global_lock
);
5759 * register_netdevice - register a network device
5760 * @dev: device to register
5762 * Take a completed network device structure and add it to the kernel
5763 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5764 * chain. 0 is returned on success. A negative errno code is returned
5765 * on a failure to set up the device, or if the name is a duplicate.
5767 * Callers must hold the rtnl semaphore. You may want
5768 * register_netdev() instead of this.
5771 * The locking appears insufficient to guarantee two parallel registers
5772 * will not get the same name.
5775 int register_netdevice(struct net_device
*dev
)
5778 struct net
*net
= dev_net(dev
);
5780 BUG_ON(dev_boot_phase
);
5785 /* When net_device's are persistent, this will be fatal. */
5786 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
5789 spin_lock_init(&dev
->addr_list_lock
);
5790 netdev_set_addr_lockdep_class(dev
);
5794 ret
= dev_get_valid_name(net
, dev
, dev
->name
);
5798 /* Init, if this function is available */
5799 if (dev
->netdev_ops
->ndo_init
) {
5800 ret
= dev
->netdev_ops
->ndo_init(dev
);
5808 if (((dev
->hw_features
| dev
->features
) &
5809 NETIF_F_HW_VLAN_CTAG_FILTER
) &&
5810 (!dev
->netdev_ops
->ndo_vlan_rx_add_vid
||
5811 !dev
->netdev_ops
->ndo_vlan_rx_kill_vid
)) {
5812 netdev_WARN(dev
, "Buggy VLAN acceleration in driver!\n");
5819 dev
->ifindex
= dev_new_index(net
);
5820 else if (__dev_get_by_index(net
, dev
->ifindex
))
5823 if (dev
->iflink
== -1)
5824 dev
->iflink
= dev
->ifindex
;
5826 /* Transfer changeable features to wanted_features and enable
5827 * software offloads (GSO and GRO).
5829 dev
->hw_features
|= NETIF_F_SOFT_FEATURES
;
5830 dev
->features
|= NETIF_F_SOFT_FEATURES
;
5831 dev
->wanted_features
= dev
->features
& dev
->hw_features
;
5833 /* Turn on no cache copy if HW is doing checksum */
5834 if (!(dev
->flags
& IFF_LOOPBACK
)) {
5835 dev
->hw_features
|= NETIF_F_NOCACHE_COPY
;
5836 if (dev
->features
& NETIF_F_ALL_CSUM
) {
5837 dev
->wanted_features
|= NETIF_F_NOCACHE_COPY
;
5838 dev
->features
|= NETIF_F_NOCACHE_COPY
;
5842 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5844 dev
->vlan_features
|= NETIF_F_HIGHDMA
;
5846 /* Make NETIF_F_SG inheritable to tunnel devices.
5848 dev
->hw_enc_features
|= NETIF_F_SG
;
5850 /* Make NETIF_F_SG inheritable to MPLS.
5852 dev
->mpls_features
|= NETIF_F_SG
;
5854 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
5855 ret
= notifier_to_errno(ret
);
5859 ret
= netdev_register_kobject(dev
);
5862 dev
->reg_state
= NETREG_REGISTERED
;
5864 __netdev_update_features(dev
);
5867 * Default initial state at registry is that the
5868 * device is present.
5871 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5873 linkwatch_init_dev(dev
);
5875 dev_init_scheduler(dev
);
5877 list_netdevice(dev
);
5878 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
5880 /* If the device has permanent device address, driver should
5881 * set dev_addr and also addr_assign_type should be set to
5882 * NET_ADDR_PERM (default value).
5884 if (dev
->addr_assign_type
== NET_ADDR_PERM
)
5885 memcpy(dev
->perm_addr
, dev
->dev_addr
, dev
->addr_len
);
5887 /* Notify protocols, that a new device appeared. */
5888 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5889 ret
= notifier_to_errno(ret
);
5891 rollback_registered(dev
);
5892 dev
->reg_state
= NETREG_UNREGISTERED
;
5895 * Prevent userspace races by waiting until the network
5896 * device is fully setup before sending notifications.
5898 if (!dev
->rtnl_link_ops
||
5899 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5900 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U, GFP_KERNEL
);
5906 if (dev
->netdev_ops
->ndo_uninit
)
5907 dev
->netdev_ops
->ndo_uninit(dev
);
5910 EXPORT_SYMBOL(register_netdevice
);
5913 * init_dummy_netdev - init a dummy network device for NAPI
5914 * @dev: device to init
5916 * This takes a network device structure and initialize the minimum
5917 * amount of fields so it can be used to schedule NAPI polls without
5918 * registering a full blown interface. This is to be used by drivers
5919 * that need to tie several hardware interfaces to a single NAPI
5920 * poll scheduler due to HW limitations.
5922 int init_dummy_netdev(struct net_device
*dev
)
5924 /* Clear everything. Note we don't initialize spinlocks
5925 * are they aren't supposed to be taken by any of the
5926 * NAPI code and this dummy netdev is supposed to be
5927 * only ever used for NAPI polls
5929 memset(dev
, 0, sizeof(struct net_device
));
5931 /* make sure we BUG if trying to hit standard
5932 * register/unregister code path
5934 dev
->reg_state
= NETREG_DUMMY
;
5936 /* NAPI wants this */
5937 INIT_LIST_HEAD(&dev
->napi_list
);
5939 /* a dummy interface is started by default */
5940 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5941 set_bit(__LINK_STATE_START
, &dev
->state
);
5943 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5944 * because users of this 'device' dont need to change
5950 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
5954 * register_netdev - register a network device
5955 * @dev: device to register
5957 * Take a completed network device structure and add it to the kernel
5958 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5959 * chain. 0 is returned on success. A negative errno code is returned
5960 * on a failure to set up the device, or if the name is a duplicate.
5962 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5963 * and expands the device name if you passed a format string to
5966 int register_netdev(struct net_device
*dev
)
5971 err
= register_netdevice(dev
);
5975 EXPORT_SYMBOL(register_netdev
);
5977 int netdev_refcnt_read(const struct net_device
*dev
)
5981 for_each_possible_cpu(i
)
5982 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
5985 EXPORT_SYMBOL(netdev_refcnt_read
);
5988 * netdev_wait_allrefs - wait until all references are gone.
5989 * @dev: target net_device
5991 * This is called when unregistering network devices.
5993 * Any protocol or device that holds a reference should register
5994 * for netdevice notification, and cleanup and put back the
5995 * reference if they receive an UNREGISTER event.
5996 * We can get stuck here if buggy protocols don't correctly
5999 static void netdev_wait_allrefs(struct net_device
*dev
)
6001 unsigned long rebroadcast_time
, warning_time
;
6004 linkwatch_forget_dev(dev
);
6006 rebroadcast_time
= warning_time
= jiffies
;
6007 refcnt
= netdev_refcnt_read(dev
);
6009 while (refcnt
!= 0) {
6010 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
6013 /* Rebroadcast unregister notification */
6014 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6020 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
6021 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
6023 /* We must not have linkwatch events
6024 * pending on unregister. If this
6025 * happens, we simply run the queue
6026 * unscheduled, resulting in a noop
6029 linkwatch_run_queue();
6034 rebroadcast_time
= jiffies
;
6039 refcnt
= netdev_refcnt_read(dev
);
6041 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
6042 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6044 warning_time
= jiffies
;
6053 * register_netdevice(x1);
6054 * register_netdevice(x2);
6056 * unregister_netdevice(y1);
6057 * unregister_netdevice(y2);
6063 * We are invoked by rtnl_unlock().
6064 * This allows us to deal with problems:
6065 * 1) We can delete sysfs objects which invoke hotplug
6066 * without deadlocking with linkwatch via keventd.
6067 * 2) Since we run with the RTNL semaphore not held, we can sleep
6068 * safely in order to wait for the netdev refcnt to drop to zero.
6070 * We must not return until all unregister events added during
6071 * the interval the lock was held have been completed.
6073 void netdev_run_todo(void)
6075 struct list_head list
;
6077 /* Snapshot list, allow later requests */
6078 list_replace_init(&net_todo_list
, &list
);
6083 /* Wait for rcu callbacks to finish before next phase */
6084 if (!list_empty(&list
))
6087 while (!list_empty(&list
)) {
6088 struct net_device
*dev
6089 = list_first_entry(&list
, struct net_device
, todo_list
);
6090 list_del(&dev
->todo_list
);
6093 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
6096 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
6097 pr_err("network todo '%s' but state %d\n",
6098 dev
->name
, dev
->reg_state
);
6103 dev
->reg_state
= NETREG_UNREGISTERED
;
6105 on_each_cpu(flush_backlog
, dev
, 1);
6107 netdev_wait_allrefs(dev
);
6110 BUG_ON(netdev_refcnt_read(dev
));
6111 WARN_ON(rcu_access_pointer(dev
->ip_ptr
));
6112 WARN_ON(rcu_access_pointer(dev
->ip6_ptr
));
6113 WARN_ON(dev
->dn_ptr
);
6115 if (dev
->destructor
)
6116 dev
->destructor(dev
);
6118 /* Report a network device has been unregistered */
6120 dev_net(dev
)->dev_unreg_count
--;
6122 wake_up(&netdev_unregistering_wq
);
6124 /* Free network device */
6125 kobject_put(&dev
->dev
.kobj
);
6129 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6130 * fields in the same order, with only the type differing.
6132 void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
6133 const struct net_device_stats
*netdev_stats
)
6135 #if BITS_PER_LONG == 64
6136 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
6137 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
6139 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
6140 const unsigned long *src
= (const unsigned long *)netdev_stats
;
6141 u64
*dst
= (u64
*)stats64
;
6143 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
6144 sizeof(*stats64
) / sizeof(u64
));
6145 for (i
= 0; i
< n
; i
++)
6149 EXPORT_SYMBOL(netdev_stats_to_stats64
);
6152 * dev_get_stats - get network device statistics
6153 * @dev: device to get statistics from
6154 * @storage: place to store stats
6156 * Get network statistics from device. Return @storage.
6157 * The device driver may provide its own method by setting
6158 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6159 * otherwise the internal statistics structure is used.
6161 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
6162 struct rtnl_link_stats64
*storage
)
6164 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6166 if (ops
->ndo_get_stats64
) {
6167 memset(storage
, 0, sizeof(*storage
));
6168 ops
->ndo_get_stats64(dev
, storage
);
6169 } else if (ops
->ndo_get_stats
) {
6170 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
6172 netdev_stats_to_stats64(storage
, &dev
->stats
);
6174 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
6177 EXPORT_SYMBOL(dev_get_stats
);
6179 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
6181 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
6183 #ifdef CONFIG_NET_CLS_ACT
6186 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
6189 netdev_init_one_queue(dev
, queue
, NULL
);
6190 queue
->qdisc
= &noop_qdisc
;
6191 queue
->qdisc_sleeping
= &noop_qdisc
;
6192 rcu_assign_pointer(dev
->ingress_queue
, queue
);
6197 static const struct ethtool_ops default_ethtool_ops
;
6199 void netdev_set_default_ethtool_ops(struct net_device
*dev
,
6200 const struct ethtool_ops
*ops
)
6202 if (dev
->ethtool_ops
== &default_ethtool_ops
)
6203 dev
->ethtool_ops
= ops
;
6205 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops
);
6207 void netdev_freemem(struct net_device
*dev
)
6209 char *addr
= (char *)dev
- dev
->padded
;
6211 if (is_vmalloc_addr(addr
))
6218 * alloc_netdev_mqs - allocate network device
6219 * @sizeof_priv: size of private data to allocate space for
6220 * @name: device name format string
6221 * @setup: callback to initialize device
6222 * @txqs: the number of TX subqueues to allocate
6223 * @rxqs: the number of RX subqueues to allocate
6225 * Allocates a struct net_device with private data area for driver use
6226 * and performs basic initialization. Also allocates subquue structs
6227 * for each queue on the device.
6229 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
6230 void (*setup
)(struct net_device
*),
6231 unsigned int txqs
, unsigned int rxqs
)
6233 struct net_device
*dev
;
6235 struct net_device
*p
;
6237 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
6240 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6246 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6251 alloc_size
= sizeof(struct net_device
);
6253 /* ensure 32-byte alignment of private area */
6254 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
6255 alloc_size
+= sizeof_priv
;
6257 /* ensure 32-byte alignment of whole construct */
6258 alloc_size
+= NETDEV_ALIGN
- 1;
6260 p
= kzalloc(alloc_size
, GFP_KERNEL
| __GFP_NOWARN
| __GFP_REPEAT
);
6262 p
= vzalloc(alloc_size
);
6266 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
6267 dev
->padded
= (char *)dev
- (char *)p
;
6269 dev
->pcpu_refcnt
= alloc_percpu(int);
6270 if (!dev
->pcpu_refcnt
)
6273 if (dev_addr_init(dev
))
6279 dev_net_set(dev
, &init_net
);
6281 dev
->gso_max_size
= GSO_MAX_SIZE
;
6282 dev
->gso_max_segs
= GSO_MAX_SEGS
;
6284 INIT_LIST_HEAD(&dev
->napi_list
);
6285 INIT_LIST_HEAD(&dev
->unreg_list
);
6286 INIT_LIST_HEAD(&dev
->close_list
);
6287 INIT_LIST_HEAD(&dev
->link_watch_list
);
6288 INIT_LIST_HEAD(&dev
->adj_list
.upper
);
6289 INIT_LIST_HEAD(&dev
->adj_list
.lower
);
6290 INIT_LIST_HEAD(&dev
->all_adj_list
.upper
);
6291 INIT_LIST_HEAD(&dev
->all_adj_list
.lower
);
6292 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
6295 dev
->num_tx_queues
= txqs
;
6296 dev
->real_num_tx_queues
= txqs
;
6297 if (netif_alloc_netdev_queues(dev
))
6301 dev
->num_rx_queues
= rxqs
;
6302 dev
->real_num_rx_queues
= rxqs
;
6303 if (netif_alloc_rx_queues(dev
))
6307 strcpy(dev
->name
, name
);
6308 dev
->group
= INIT_NETDEV_GROUP
;
6309 if (!dev
->ethtool_ops
)
6310 dev
->ethtool_ops
= &default_ethtool_ops
;
6318 free_percpu(dev
->pcpu_refcnt
);
6319 netif_free_tx_queues(dev
);
6325 netdev_freemem(dev
);
6328 EXPORT_SYMBOL(alloc_netdev_mqs
);
6331 * free_netdev - free network device
6334 * This function does the last stage of destroying an allocated device
6335 * interface. The reference to the device object is released.
6336 * If this is the last reference then it will be freed.
6338 void free_netdev(struct net_device
*dev
)
6340 struct napi_struct
*p
, *n
;
6342 release_net(dev_net(dev
));
6344 netif_free_tx_queues(dev
);
6349 kfree(rcu_dereference_protected(dev
->ingress_queue
, 1));
6351 /* Flush device addresses */
6352 dev_addr_flush(dev
);
6354 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
6357 free_percpu(dev
->pcpu_refcnt
);
6358 dev
->pcpu_refcnt
= NULL
;
6360 /* Compatibility with error handling in drivers */
6361 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
6362 netdev_freemem(dev
);
6366 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
6367 dev
->reg_state
= NETREG_RELEASED
;
6369 /* will free via device release */
6370 put_device(&dev
->dev
);
6372 EXPORT_SYMBOL(free_netdev
);
6375 * synchronize_net - Synchronize with packet receive processing
6377 * Wait for packets currently being received to be done.
6378 * Does not block later packets from starting.
6380 void synchronize_net(void)
6383 if (rtnl_is_locked())
6384 synchronize_rcu_expedited();
6388 EXPORT_SYMBOL(synchronize_net
);
6391 * unregister_netdevice_queue - remove device from the kernel
6395 * This function shuts down a device interface and removes it
6396 * from the kernel tables.
6397 * If head not NULL, device is queued to be unregistered later.
6399 * Callers must hold the rtnl semaphore. You may want
6400 * unregister_netdev() instead of this.
6403 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
6408 list_move_tail(&dev
->unreg_list
, head
);
6410 rollback_registered(dev
);
6411 /* Finish processing unregister after unlock */
6415 EXPORT_SYMBOL(unregister_netdevice_queue
);
6418 * unregister_netdevice_many - unregister many devices
6419 * @head: list of devices
6421 void unregister_netdevice_many(struct list_head
*head
)
6423 struct net_device
*dev
;
6425 if (!list_empty(head
)) {
6426 rollback_registered_many(head
);
6427 list_for_each_entry(dev
, head
, unreg_list
)
6431 EXPORT_SYMBOL(unregister_netdevice_many
);
6434 * unregister_netdev - remove device from the kernel
6437 * This function shuts down a device interface and removes it
6438 * from the kernel tables.
6440 * This is just a wrapper for unregister_netdevice that takes
6441 * the rtnl semaphore. In general you want to use this and not
6442 * unregister_netdevice.
6444 void unregister_netdev(struct net_device
*dev
)
6447 unregister_netdevice(dev
);
6450 EXPORT_SYMBOL(unregister_netdev
);
6453 * dev_change_net_namespace - move device to different nethost namespace
6455 * @net: network namespace
6456 * @pat: If not NULL name pattern to try if the current device name
6457 * is already taken in the destination network namespace.
6459 * This function shuts down a device interface and moves it
6460 * to a new network namespace. On success 0 is returned, on
6461 * a failure a netagive errno code is returned.
6463 * Callers must hold the rtnl semaphore.
6466 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
6472 /* Don't allow namespace local devices to be moved. */
6474 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6477 /* Ensure the device has been registrered */
6478 if (dev
->reg_state
!= NETREG_REGISTERED
)
6481 /* Get out if there is nothing todo */
6483 if (net_eq(dev_net(dev
), net
))
6486 /* Pick the destination device name, and ensure
6487 * we can use it in the destination network namespace.
6490 if (__dev_get_by_name(net
, dev
->name
)) {
6491 /* We get here if we can't use the current device name */
6494 if (dev_get_valid_name(net
, dev
, pat
) < 0)
6499 * And now a mini version of register_netdevice unregister_netdevice.
6502 /* If device is running close it first. */
6505 /* And unlink it from device chain */
6507 unlist_netdevice(dev
);
6511 /* Shutdown queueing discipline. */
6514 /* Notify protocols, that we are about to destroy
6515 this device. They should clean all the things.
6517 Note that dev->reg_state stays at NETREG_REGISTERED.
6518 This is wanted because this way 8021q and macvlan know
6519 the device is just moving and can keep their slaves up.
6521 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6523 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
6524 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U, GFP_KERNEL
);
6527 * Flush the unicast and multicast chains
6532 /* Send a netdev-removed uevent to the old namespace */
6533 kobject_uevent(&dev
->dev
.kobj
, KOBJ_REMOVE
);
6535 /* Actually switch the network namespace */
6536 dev_net_set(dev
, net
);
6538 /* If there is an ifindex conflict assign a new one */
6539 if (__dev_get_by_index(net
, dev
->ifindex
)) {
6540 int iflink
= (dev
->iflink
== dev
->ifindex
);
6541 dev
->ifindex
= dev_new_index(net
);
6543 dev
->iflink
= dev
->ifindex
;
6546 /* Send a netdev-add uevent to the new namespace */
6547 kobject_uevent(&dev
->dev
.kobj
, KOBJ_ADD
);
6549 /* Fixup kobjects */
6550 err
= device_rename(&dev
->dev
, dev
->name
);
6553 /* Add the device back in the hashes */
6554 list_netdevice(dev
);
6556 /* Notify protocols, that a new device appeared. */
6557 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
6560 * Prevent userspace races by waiting until the network
6561 * device is fully setup before sending notifications.
6563 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U, GFP_KERNEL
);
6570 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
6572 static int dev_cpu_callback(struct notifier_block
*nfb
,
6573 unsigned long action
,
6576 struct sk_buff
**list_skb
;
6577 struct sk_buff
*skb
;
6578 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
6579 struct softnet_data
*sd
, *oldsd
;
6581 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
6584 local_irq_disable();
6585 cpu
= smp_processor_id();
6586 sd
= &per_cpu(softnet_data
, cpu
);
6587 oldsd
= &per_cpu(softnet_data
, oldcpu
);
6589 /* Find end of our completion_queue. */
6590 list_skb
= &sd
->completion_queue
;
6592 list_skb
= &(*list_skb
)->next
;
6593 /* Append completion queue from offline CPU. */
6594 *list_skb
= oldsd
->completion_queue
;
6595 oldsd
->completion_queue
= NULL
;
6597 /* Append output queue from offline CPU. */
6598 if (oldsd
->output_queue
) {
6599 *sd
->output_queue_tailp
= oldsd
->output_queue
;
6600 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
6601 oldsd
->output_queue
= NULL
;
6602 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
6604 /* Append NAPI poll list from offline CPU. */
6605 if (!list_empty(&oldsd
->poll_list
)) {
6606 list_splice_init(&oldsd
->poll_list
, &sd
->poll_list
);
6607 raise_softirq_irqoff(NET_RX_SOFTIRQ
);
6610 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
6613 /* Process offline CPU's input_pkt_queue */
6614 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
6616 input_queue_head_incr(oldsd
);
6618 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
6620 input_queue_head_incr(oldsd
);
6628 * netdev_increment_features - increment feature set by one
6629 * @all: current feature set
6630 * @one: new feature set
6631 * @mask: mask feature set
6633 * Computes a new feature set after adding a device with feature set
6634 * @one to the master device with current feature set @all. Will not
6635 * enable anything that is off in @mask. Returns the new feature set.
6637 netdev_features_t
netdev_increment_features(netdev_features_t all
,
6638 netdev_features_t one
, netdev_features_t mask
)
6640 if (mask
& NETIF_F_GEN_CSUM
)
6641 mask
|= NETIF_F_ALL_CSUM
;
6642 mask
|= NETIF_F_VLAN_CHALLENGED
;
6644 all
|= one
& (NETIF_F_ONE_FOR_ALL
|NETIF_F_ALL_CSUM
) & mask
;
6645 all
&= one
| ~NETIF_F_ALL_FOR_ALL
;
6647 /* If one device supports hw checksumming, set for all. */
6648 if (all
& NETIF_F_GEN_CSUM
)
6649 all
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_GEN_CSUM
);
6653 EXPORT_SYMBOL(netdev_increment_features
);
6655 static struct hlist_head
* __net_init
netdev_create_hash(void)
6658 struct hlist_head
*hash
;
6660 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
6662 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
6663 INIT_HLIST_HEAD(&hash
[i
]);
6668 /* Initialize per network namespace state */
6669 static int __net_init
netdev_init(struct net
*net
)
6671 if (net
!= &init_net
)
6672 INIT_LIST_HEAD(&net
->dev_base_head
);
6674 net
->dev_name_head
= netdev_create_hash();
6675 if (net
->dev_name_head
== NULL
)
6678 net
->dev_index_head
= netdev_create_hash();
6679 if (net
->dev_index_head
== NULL
)
6685 kfree(net
->dev_name_head
);
6691 * netdev_drivername - network driver for the device
6692 * @dev: network device
6694 * Determine network driver for device.
6696 const char *netdev_drivername(const struct net_device
*dev
)
6698 const struct device_driver
*driver
;
6699 const struct device
*parent
;
6700 const char *empty
= "";
6702 parent
= dev
->dev
.parent
;
6706 driver
= parent
->driver
;
6707 if (driver
&& driver
->name
)
6708 return driver
->name
;
6712 static int __netdev_printk(const char *level
, const struct net_device
*dev
,
6713 struct va_format
*vaf
)
6717 if (dev
&& dev
->dev
.parent
) {
6718 r
= dev_printk_emit(level
[1] - '0',
6721 dev_driver_string(dev
->dev
.parent
),
6722 dev_name(dev
->dev
.parent
),
6723 netdev_name(dev
), vaf
);
6725 r
= printk("%s%s: %pV", level
, netdev_name(dev
), vaf
);
6727 r
= printk("%s(NULL net_device): %pV", level
, vaf
);
6733 int netdev_printk(const char *level
, const struct net_device
*dev
,
6734 const char *format
, ...)
6736 struct va_format vaf
;
6740 va_start(args
, format
);
6745 r
= __netdev_printk(level
, dev
, &vaf
);
6751 EXPORT_SYMBOL(netdev_printk
);
6753 #define define_netdev_printk_level(func, level) \
6754 int func(const struct net_device *dev, const char *fmt, ...) \
6757 struct va_format vaf; \
6760 va_start(args, fmt); \
6765 r = __netdev_printk(level, dev, &vaf); \
6771 EXPORT_SYMBOL(func);
6773 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
6774 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
6775 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
6776 define_netdev_printk_level(netdev_err
, KERN_ERR
);
6777 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
6778 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
6779 define_netdev_printk_level(netdev_info
, KERN_INFO
);
6781 static void __net_exit
netdev_exit(struct net
*net
)
6783 kfree(net
->dev_name_head
);
6784 kfree(net
->dev_index_head
);
6787 static struct pernet_operations __net_initdata netdev_net_ops
= {
6788 .init
= netdev_init
,
6789 .exit
= netdev_exit
,
6792 static void __net_exit
default_device_exit(struct net
*net
)
6794 struct net_device
*dev
, *aux
;
6796 * Push all migratable network devices back to the
6797 * initial network namespace
6800 for_each_netdev_safe(net
, dev
, aux
) {
6802 char fb_name
[IFNAMSIZ
];
6804 /* Ignore unmoveable devices (i.e. loopback) */
6805 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6808 /* Leave virtual devices for the generic cleanup */
6809 if (dev
->rtnl_link_ops
)
6812 /* Push remaining network devices to init_net */
6813 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
6814 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
6816 pr_emerg("%s: failed to move %s to init_net: %d\n",
6817 __func__
, dev
->name
, err
);
6824 static void __net_exit
rtnl_lock_unregistering(struct list_head
*net_list
)
6826 /* Return with the rtnl_lock held when there are no network
6827 * devices unregistering in any network namespace in net_list.
6834 prepare_to_wait(&netdev_unregistering_wq
, &wait
,
6835 TASK_UNINTERRUPTIBLE
);
6836 unregistering
= false;
6838 list_for_each_entry(net
, net_list
, exit_list
) {
6839 if (net
->dev_unreg_count
> 0) {
6840 unregistering
= true;
6849 finish_wait(&netdev_unregistering_wq
, &wait
);
6852 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
6854 /* At exit all network devices most be removed from a network
6855 * namespace. Do this in the reverse order of registration.
6856 * Do this across as many network namespaces as possible to
6857 * improve batching efficiency.
6859 struct net_device
*dev
;
6861 LIST_HEAD(dev_kill_list
);
6863 /* To prevent network device cleanup code from dereferencing
6864 * loopback devices or network devices that have been freed
6865 * wait here for all pending unregistrations to complete,
6866 * before unregistring the loopback device and allowing the
6867 * network namespace be freed.
6869 * The netdev todo list containing all network devices
6870 * unregistrations that happen in default_device_exit_batch
6871 * will run in the rtnl_unlock() at the end of
6872 * default_device_exit_batch.
6874 rtnl_lock_unregistering(net_list
);
6875 list_for_each_entry(net
, net_list
, exit_list
) {
6876 for_each_netdev_reverse(net
, dev
) {
6877 if (dev
->rtnl_link_ops
)
6878 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
6880 unregister_netdevice_queue(dev
, &dev_kill_list
);
6883 unregister_netdevice_many(&dev_kill_list
);
6884 list_del(&dev_kill_list
);
6888 static struct pernet_operations __net_initdata default_device_ops
= {
6889 .exit
= default_device_exit
,
6890 .exit_batch
= default_device_exit_batch
,
6894 * Initialize the DEV module. At boot time this walks the device list and
6895 * unhooks any devices that fail to initialise (normally hardware not
6896 * present) and leaves us with a valid list of present and active devices.
6901 * This is called single threaded during boot, so no need
6902 * to take the rtnl semaphore.
6904 static int __init
net_dev_init(void)
6906 int i
, rc
= -ENOMEM
;
6908 BUG_ON(!dev_boot_phase
);
6910 if (dev_proc_init())
6913 if (netdev_kobject_init())
6916 INIT_LIST_HEAD(&ptype_all
);
6917 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
6918 INIT_LIST_HEAD(&ptype_base
[i
]);
6920 INIT_LIST_HEAD(&offload_base
);
6922 if (register_pernet_subsys(&netdev_net_ops
))
6926 * Initialise the packet receive queues.
6929 for_each_possible_cpu(i
) {
6930 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
6932 memset(sd
, 0, sizeof(*sd
));
6933 skb_queue_head_init(&sd
->input_pkt_queue
);
6934 skb_queue_head_init(&sd
->process_queue
);
6935 sd
->completion_queue
= NULL
;
6936 INIT_LIST_HEAD(&sd
->poll_list
);
6937 sd
->output_queue
= NULL
;
6938 sd
->output_queue_tailp
= &sd
->output_queue
;
6940 sd
->csd
.func
= rps_trigger_softirq
;
6946 sd
->backlog
.poll
= process_backlog
;
6947 sd
->backlog
.weight
= weight_p
;
6948 sd
->backlog
.gro_list
= NULL
;
6949 sd
->backlog
.gro_count
= 0;
6951 #ifdef CONFIG_NET_FLOW_LIMIT
6952 sd
->flow_limit
= NULL
;
6958 /* The loopback device is special if any other network devices
6959 * is present in a network namespace the loopback device must
6960 * be present. Since we now dynamically allocate and free the
6961 * loopback device ensure this invariant is maintained by
6962 * keeping the loopback device as the first device on the
6963 * list of network devices. Ensuring the loopback devices
6964 * is the first device that appears and the last network device
6967 if (register_pernet_device(&loopback_net_ops
))
6970 if (register_pernet_device(&default_device_ops
))
6973 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
6974 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
6976 hotcpu_notifier(dev_cpu_callback
, 0);
6983 subsys_initcall(net_dev_init
);