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>
135 #include "net-sysfs.h"
137 /* Instead of increasing this, you should create a hash table. */
138 #define MAX_GRO_SKBS 8
140 /* This should be increased if a protocol with a bigger head is added. */
141 #define GRO_MAX_HEAD (MAX_HEADER + 128)
143 static DEFINE_SPINLOCK(ptype_lock
);
144 static DEFINE_SPINLOCK(offload_lock
);
145 struct list_head ptype_base
[PTYPE_HASH_SIZE
] __read_mostly
;
146 struct list_head ptype_all __read_mostly
; /* Taps */
147 static struct list_head offload_base __read_mostly
;
150 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
153 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
155 * Writers must hold the rtnl semaphore while they loop through the
156 * dev_base_head list, and hold dev_base_lock for writing when they do the
157 * actual updates. This allows pure readers to access the list even
158 * while a writer is preparing to update it.
160 * To put it another way, dev_base_lock is held for writing only to
161 * protect against pure readers; the rtnl semaphore provides the
162 * protection against other writers.
164 * See, for example usages, register_netdevice() and
165 * unregister_netdevice(), which must be called with the rtnl
168 DEFINE_RWLOCK(dev_base_lock
);
169 EXPORT_SYMBOL(dev_base_lock
);
171 /* protects napi_hash addition/deletion and napi_gen_id */
172 static DEFINE_SPINLOCK(napi_hash_lock
);
174 static unsigned int napi_gen_id
;
175 static DEFINE_HASHTABLE(napi_hash
, 8);
177 static seqcount_t devnet_rename_seq
;
179 static inline void dev_base_seq_inc(struct net
*net
)
181 while (++net
->dev_base_seq
== 0);
184 static inline struct hlist_head
*dev_name_hash(struct net
*net
, const char *name
)
186 unsigned int hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
188 return &net
->dev_name_head
[hash_32(hash
, NETDEV_HASHBITS
)];
191 static inline struct hlist_head
*dev_index_hash(struct net
*net
, int ifindex
)
193 return &net
->dev_index_head
[ifindex
& (NETDEV_HASHENTRIES
- 1)];
196 static inline void rps_lock(struct softnet_data
*sd
)
199 spin_lock(&sd
->input_pkt_queue
.lock
);
203 static inline void rps_unlock(struct softnet_data
*sd
)
206 spin_unlock(&sd
->input_pkt_queue
.lock
);
210 /* Device list insertion */
211 static void list_netdevice(struct net_device
*dev
)
213 struct net
*net
= dev_net(dev
);
217 write_lock_bh(&dev_base_lock
);
218 list_add_tail_rcu(&dev
->dev_list
, &net
->dev_base_head
);
219 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
220 hlist_add_head_rcu(&dev
->index_hlist
,
221 dev_index_hash(net
, dev
->ifindex
));
222 write_unlock_bh(&dev_base_lock
);
224 dev_base_seq_inc(net
);
227 /* Device list removal
228 * caller must respect a RCU grace period before freeing/reusing dev
230 static void unlist_netdevice(struct net_device
*dev
)
234 /* Unlink dev from the device chain */
235 write_lock_bh(&dev_base_lock
);
236 list_del_rcu(&dev
->dev_list
);
237 hlist_del_rcu(&dev
->name_hlist
);
238 hlist_del_rcu(&dev
->index_hlist
);
239 write_unlock_bh(&dev_base_lock
);
241 dev_base_seq_inc(dev_net(dev
));
248 static RAW_NOTIFIER_HEAD(netdev_chain
);
251 * Device drivers call our routines to queue packets here. We empty the
252 * queue in the local softnet handler.
255 DEFINE_PER_CPU_ALIGNED(struct softnet_data
, softnet_data
);
256 EXPORT_PER_CPU_SYMBOL(softnet_data
);
258 #ifdef CONFIG_LOCKDEP
260 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
261 * according to dev->type
263 static const unsigned short netdev_lock_type
[] =
264 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
265 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
266 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
267 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
268 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
269 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
270 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
271 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
272 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
273 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
274 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
275 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
276 ARPHRD_FCFABRIC
, ARPHRD_IEEE80211
, ARPHRD_IEEE80211_PRISM
,
277 ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_PHONET
, ARPHRD_PHONET_PIPE
,
278 ARPHRD_IEEE802154
, ARPHRD_VOID
, ARPHRD_NONE
};
280 static const char *const netdev_lock_name
[] =
281 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
282 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
283 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
284 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
285 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
286 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
287 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
288 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
289 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
290 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
291 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
292 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
293 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
294 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
295 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
297 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
298 static struct lock_class_key netdev_addr_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
300 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
304 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
305 if (netdev_lock_type
[i
] == dev_type
)
307 /* the last key is used by default */
308 return ARRAY_SIZE(netdev_lock_type
) - 1;
311 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
312 unsigned short dev_type
)
316 i
= netdev_lock_pos(dev_type
);
317 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
318 netdev_lock_name
[i
]);
321 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
325 i
= netdev_lock_pos(dev
->type
);
326 lockdep_set_class_and_name(&dev
->addr_list_lock
,
327 &netdev_addr_lock_key
[i
],
328 netdev_lock_name
[i
]);
331 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
332 unsigned short dev_type
)
335 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
340 /*******************************************************************************
342 Protocol management and registration routines
344 *******************************************************************************/
347 * Add a protocol ID to the list. Now that the input handler is
348 * smarter we can dispense with all the messy stuff that used to be
351 * BEWARE!!! Protocol handlers, mangling input packets,
352 * MUST BE last in hash buckets and checking protocol handlers
353 * MUST start from promiscuous ptype_all chain in net_bh.
354 * It is true now, do not change it.
355 * Explanation follows: if protocol handler, mangling packet, will
356 * be the first on list, it is not able to sense, that packet
357 * is cloned and should be copied-on-write, so that it will
358 * change it and subsequent readers will get broken packet.
362 static inline struct list_head
*ptype_head(const struct packet_type
*pt
)
364 if (pt
->type
== htons(ETH_P_ALL
))
367 return &ptype_base
[ntohs(pt
->type
) & PTYPE_HASH_MASK
];
371 * dev_add_pack - add packet handler
372 * @pt: packet type declaration
374 * Add a protocol handler to the networking stack. The passed &packet_type
375 * is linked into kernel lists and may not be freed until it has been
376 * removed from the kernel lists.
378 * This call does not sleep therefore it can not
379 * guarantee all CPU's that are in middle of receiving packets
380 * will see the new packet type (until the next received packet).
383 void dev_add_pack(struct packet_type
*pt
)
385 struct list_head
*head
= ptype_head(pt
);
387 spin_lock(&ptype_lock
);
388 list_add_rcu(&pt
->list
, head
);
389 spin_unlock(&ptype_lock
);
391 EXPORT_SYMBOL(dev_add_pack
);
394 * __dev_remove_pack - remove packet handler
395 * @pt: packet type declaration
397 * Remove a protocol handler that was previously added to the kernel
398 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
399 * from the kernel lists and can be freed or reused once this function
402 * The packet type might still be in use by receivers
403 * and must not be freed until after all the CPU's have gone
404 * through a quiescent state.
406 void __dev_remove_pack(struct packet_type
*pt
)
408 struct list_head
*head
= ptype_head(pt
);
409 struct packet_type
*pt1
;
411 spin_lock(&ptype_lock
);
413 list_for_each_entry(pt1
, head
, list
) {
415 list_del_rcu(&pt
->list
);
420 pr_warn("dev_remove_pack: %p not found\n", pt
);
422 spin_unlock(&ptype_lock
);
424 EXPORT_SYMBOL(__dev_remove_pack
);
427 * dev_remove_pack - remove packet handler
428 * @pt: packet type declaration
430 * Remove a protocol handler that was previously added to the kernel
431 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
432 * from the kernel lists and can be freed or reused once this function
435 * This call sleeps to guarantee that no CPU is looking at the packet
438 void dev_remove_pack(struct packet_type
*pt
)
440 __dev_remove_pack(pt
);
444 EXPORT_SYMBOL(dev_remove_pack
);
448 * dev_add_offload - register offload handlers
449 * @po: protocol offload declaration
451 * Add protocol offload handlers to the networking stack. The passed
452 * &proto_offload is linked into kernel lists and may not be freed until
453 * it has been removed from the kernel lists.
455 * This call does not sleep therefore it can not
456 * guarantee all CPU's that are in middle of receiving packets
457 * will see the new offload handlers (until the next received packet).
459 void dev_add_offload(struct packet_offload
*po
)
461 struct list_head
*head
= &offload_base
;
463 spin_lock(&offload_lock
);
464 list_add_rcu(&po
->list
, head
);
465 spin_unlock(&offload_lock
);
467 EXPORT_SYMBOL(dev_add_offload
);
470 * __dev_remove_offload - remove offload handler
471 * @po: packet offload declaration
473 * Remove a protocol offload handler that was previously added to the
474 * kernel offload handlers by dev_add_offload(). The passed &offload_type
475 * is removed from the kernel lists and can be freed or reused once this
478 * The packet type might still be in use by receivers
479 * and must not be freed until after all the CPU's have gone
480 * through a quiescent state.
482 void __dev_remove_offload(struct packet_offload
*po
)
484 struct list_head
*head
= &offload_base
;
485 struct packet_offload
*po1
;
487 spin_lock(&offload_lock
);
489 list_for_each_entry(po1
, head
, list
) {
491 list_del_rcu(&po
->list
);
496 pr_warn("dev_remove_offload: %p not found\n", po
);
498 spin_unlock(&offload_lock
);
500 EXPORT_SYMBOL(__dev_remove_offload
);
503 * dev_remove_offload - remove packet offload handler
504 * @po: packet offload declaration
506 * Remove a packet offload handler that was previously added to the kernel
507 * offload handlers by dev_add_offload(). The passed &offload_type is
508 * removed from the kernel lists and can be freed or reused once this
511 * This call sleeps to guarantee that no CPU is looking at the packet
514 void dev_remove_offload(struct packet_offload
*po
)
516 __dev_remove_offload(po
);
520 EXPORT_SYMBOL(dev_remove_offload
);
522 /******************************************************************************
524 Device Boot-time Settings Routines
526 *******************************************************************************/
528 /* Boot time configuration table */
529 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
532 * netdev_boot_setup_add - add new setup entry
533 * @name: name of the device
534 * @map: configured settings for the device
536 * Adds new setup entry to the dev_boot_setup list. The function
537 * returns 0 on error and 1 on success. This is a generic routine to
540 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
542 struct netdev_boot_setup
*s
;
546 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
547 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
548 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
549 strlcpy(s
[i
].name
, name
, IFNAMSIZ
);
550 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
555 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
559 * netdev_boot_setup_check - check boot time settings
560 * @dev: the netdevice
562 * Check boot time settings for the device.
563 * The found settings are set for the device to be used
564 * later in the device probing.
565 * Returns 0 if no settings found, 1 if they are.
567 int netdev_boot_setup_check(struct net_device
*dev
)
569 struct netdev_boot_setup
*s
= dev_boot_setup
;
572 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
573 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
574 !strcmp(dev
->name
, s
[i
].name
)) {
575 dev
->irq
= s
[i
].map
.irq
;
576 dev
->base_addr
= s
[i
].map
.base_addr
;
577 dev
->mem_start
= s
[i
].map
.mem_start
;
578 dev
->mem_end
= s
[i
].map
.mem_end
;
584 EXPORT_SYMBOL(netdev_boot_setup_check
);
588 * netdev_boot_base - get address from boot time settings
589 * @prefix: prefix for network device
590 * @unit: id for network device
592 * Check boot time settings for the base address of device.
593 * The found settings are set for the device to be used
594 * later in the device probing.
595 * Returns 0 if no settings found.
597 unsigned long netdev_boot_base(const char *prefix
, int unit
)
599 const struct netdev_boot_setup
*s
= dev_boot_setup
;
603 sprintf(name
, "%s%d", prefix
, unit
);
606 * If device already registered then return base of 1
607 * to indicate not to probe for this interface
609 if (__dev_get_by_name(&init_net
, name
))
612 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
613 if (!strcmp(name
, s
[i
].name
))
614 return s
[i
].map
.base_addr
;
619 * Saves at boot time configured settings for any netdevice.
621 int __init
netdev_boot_setup(char *str
)
626 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
631 memset(&map
, 0, sizeof(map
));
635 map
.base_addr
= ints
[2];
637 map
.mem_start
= ints
[3];
639 map
.mem_end
= ints
[4];
641 /* Add new entry to the list */
642 return netdev_boot_setup_add(str
, &map
);
645 __setup("netdev=", netdev_boot_setup
);
647 /*******************************************************************************
649 Device Interface Subroutines
651 *******************************************************************************/
654 * __dev_get_by_name - find a device by its name
655 * @net: the applicable net namespace
656 * @name: name to find
658 * Find an interface by name. Must be called under RTNL semaphore
659 * or @dev_base_lock. If the name is found a pointer to the device
660 * is returned. If the name is not found then %NULL is returned. The
661 * reference counters are not incremented so the caller must be
662 * careful with locks.
665 struct net_device
*__dev_get_by_name(struct net
*net
, const char *name
)
667 struct net_device
*dev
;
668 struct hlist_head
*head
= dev_name_hash(net
, name
);
670 hlist_for_each_entry(dev
, head
, name_hlist
)
671 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
676 EXPORT_SYMBOL(__dev_get_by_name
);
679 * dev_get_by_name_rcu - find a device by its name
680 * @net: the applicable net namespace
681 * @name: name to find
683 * Find an interface by name.
684 * If the name is found a pointer to the device is returned.
685 * If the name is not found then %NULL is returned.
686 * The reference counters are not incremented so the caller must be
687 * careful with locks. The caller must hold RCU lock.
690 struct net_device
*dev_get_by_name_rcu(struct net
*net
, const char *name
)
692 struct net_device
*dev
;
693 struct hlist_head
*head
= dev_name_hash(net
, name
);
695 hlist_for_each_entry_rcu(dev
, head
, name_hlist
)
696 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
701 EXPORT_SYMBOL(dev_get_by_name_rcu
);
704 * dev_get_by_name - find a device by its name
705 * @net: the applicable net namespace
706 * @name: name to find
708 * Find an interface by name. This can be called from any
709 * context and does its own locking. The returned handle has
710 * the usage count incremented and the caller must use dev_put() to
711 * release it when it is no longer needed. %NULL is returned if no
712 * matching device is found.
715 struct net_device
*dev_get_by_name(struct net
*net
, const char *name
)
717 struct net_device
*dev
;
720 dev
= dev_get_by_name_rcu(net
, name
);
726 EXPORT_SYMBOL(dev_get_by_name
);
729 * __dev_get_by_index - find a device by its ifindex
730 * @net: the applicable net namespace
731 * @ifindex: index of device
733 * Search for an interface by index. Returns %NULL if the device
734 * is not found or a pointer to the device. The device has not
735 * had its reference counter increased so the caller must be careful
736 * about locking. The caller must hold either the RTNL semaphore
740 struct net_device
*__dev_get_by_index(struct net
*net
, int ifindex
)
742 struct net_device
*dev
;
743 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
745 hlist_for_each_entry(dev
, head
, index_hlist
)
746 if (dev
->ifindex
== ifindex
)
751 EXPORT_SYMBOL(__dev_get_by_index
);
754 * dev_get_by_index_rcu - find a device by its ifindex
755 * @net: the applicable net namespace
756 * @ifindex: index of device
758 * Search for an interface by index. Returns %NULL if the device
759 * is not found or a pointer to the device. The device has not
760 * had its reference counter increased so the caller must be careful
761 * about locking. The caller must hold RCU lock.
764 struct net_device
*dev_get_by_index_rcu(struct net
*net
, int ifindex
)
766 struct net_device
*dev
;
767 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
769 hlist_for_each_entry_rcu(dev
, head
, index_hlist
)
770 if (dev
->ifindex
== ifindex
)
775 EXPORT_SYMBOL(dev_get_by_index_rcu
);
779 * dev_get_by_index - find a device by its ifindex
780 * @net: the applicable net namespace
781 * @ifindex: index of device
783 * Search for an interface by index. Returns NULL if the device
784 * is not found or a pointer to the device. The device returned has
785 * had a reference added and the pointer is safe until the user calls
786 * dev_put to indicate they have finished with it.
789 struct net_device
*dev_get_by_index(struct net
*net
, int ifindex
)
791 struct net_device
*dev
;
794 dev
= dev_get_by_index_rcu(net
, ifindex
);
800 EXPORT_SYMBOL(dev_get_by_index
);
803 * netdev_get_name - get a netdevice name, knowing its ifindex.
804 * @net: network namespace
805 * @name: a pointer to the buffer where the name will be stored.
806 * @ifindex: the ifindex of the interface to get the name from.
808 * The use of raw_seqcount_begin() and cond_resched() before
809 * retrying is required as we want to give the writers a chance
810 * to complete when CONFIG_PREEMPT is not set.
812 int netdev_get_name(struct net
*net
, char *name
, int ifindex
)
814 struct net_device
*dev
;
818 seq
= raw_seqcount_begin(&devnet_rename_seq
);
820 dev
= dev_get_by_index_rcu(net
, ifindex
);
826 strcpy(name
, dev
->name
);
828 if (read_seqcount_retry(&devnet_rename_seq
, seq
)) {
837 * dev_getbyhwaddr_rcu - find a device by its hardware address
838 * @net: the applicable net namespace
839 * @type: media type of device
840 * @ha: hardware address
842 * Search for an interface by MAC address. Returns NULL if the device
843 * is not found or a pointer to the device.
844 * The caller must hold RCU or RTNL.
845 * The returned device has not had its ref count increased
846 * and the caller must therefore be careful about locking
850 struct net_device
*dev_getbyhwaddr_rcu(struct net
*net
, unsigned short type
,
853 struct net_device
*dev
;
855 for_each_netdev_rcu(net
, dev
)
856 if (dev
->type
== type
&&
857 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
862 EXPORT_SYMBOL(dev_getbyhwaddr_rcu
);
864 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
866 struct net_device
*dev
;
869 for_each_netdev(net
, dev
)
870 if (dev
->type
== type
)
875 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
877 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
879 struct net_device
*dev
, *ret
= NULL
;
882 for_each_netdev_rcu(net
, dev
)
883 if (dev
->type
== type
) {
891 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
894 * dev_get_by_flags_rcu - find any device with given flags
895 * @net: the applicable net namespace
896 * @if_flags: IFF_* values
897 * @mask: bitmask of bits in if_flags to check
899 * Search for any interface with the given flags. Returns NULL if a device
900 * is not found or a pointer to the device. Must be called inside
901 * rcu_read_lock(), and result refcount is unchanged.
904 struct net_device
*dev_get_by_flags_rcu(struct net
*net
, unsigned short if_flags
,
907 struct net_device
*dev
, *ret
;
910 for_each_netdev_rcu(net
, dev
) {
911 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
918 EXPORT_SYMBOL(dev_get_by_flags_rcu
);
921 * dev_valid_name - check if name is okay for network device
924 * Network device names need to be valid file names to
925 * to allow sysfs to work. We also disallow any kind of
928 bool dev_valid_name(const char *name
)
932 if (strlen(name
) >= IFNAMSIZ
)
934 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
938 if (*name
== '/' || isspace(*name
))
944 EXPORT_SYMBOL(dev_valid_name
);
947 * __dev_alloc_name - allocate a name for a device
948 * @net: network namespace to allocate the device name in
949 * @name: name format string
950 * @buf: scratch buffer and result name string
952 * Passed a format string - eg "lt%d" it will try and find a suitable
953 * id. It scans list of devices to build up a free map, then chooses
954 * the first empty slot. The caller must hold the dev_base or rtnl lock
955 * while allocating the name and adding the device in order to avoid
957 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
958 * Returns the number of the unit assigned or a negative errno code.
961 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
965 const int max_netdevices
= 8*PAGE_SIZE
;
966 unsigned long *inuse
;
967 struct net_device
*d
;
969 p
= strnchr(name
, IFNAMSIZ
-1, '%');
972 * Verify the string as this thing may have come from
973 * the user. There must be either one "%d" and no other "%"
976 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
979 /* Use one page as a bit array of possible slots */
980 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
984 for_each_netdev(net
, d
) {
985 if (!sscanf(d
->name
, name
, &i
))
987 if (i
< 0 || i
>= max_netdevices
)
990 /* avoid cases where sscanf is not exact inverse of printf */
991 snprintf(buf
, IFNAMSIZ
, name
, i
);
992 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
996 i
= find_first_zero_bit(inuse
, max_netdevices
);
997 free_page((unsigned long) inuse
);
1001 snprintf(buf
, IFNAMSIZ
, name
, i
);
1002 if (!__dev_get_by_name(net
, buf
))
1005 /* It is possible to run out of possible slots
1006 * when the name is long and there isn't enough space left
1007 * for the digits, or if all bits are used.
1013 * dev_alloc_name - allocate a name for a device
1015 * @name: name format string
1017 * Passed a format string - eg "lt%d" it will try and find a suitable
1018 * id. It scans list of devices to build up a free map, then chooses
1019 * the first empty slot. The caller must hold the dev_base or rtnl lock
1020 * while allocating the name and adding the device in order to avoid
1022 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1023 * Returns the number of the unit assigned or a negative errno code.
1026 int dev_alloc_name(struct net_device
*dev
, const char *name
)
1032 BUG_ON(!dev_net(dev
));
1034 ret
= __dev_alloc_name(net
, name
, buf
);
1036 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
1039 EXPORT_SYMBOL(dev_alloc_name
);
1041 static int dev_alloc_name_ns(struct net
*net
,
1042 struct net_device
*dev
,
1048 ret
= __dev_alloc_name(net
, name
, buf
);
1050 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
1054 static int dev_get_valid_name(struct net
*net
,
1055 struct net_device
*dev
,
1060 if (!dev_valid_name(name
))
1063 if (strchr(name
, '%'))
1064 return dev_alloc_name_ns(net
, dev
, name
);
1065 else if (__dev_get_by_name(net
, name
))
1067 else if (dev
->name
!= name
)
1068 strlcpy(dev
->name
, name
, IFNAMSIZ
);
1074 * dev_change_name - change name of a device
1076 * @newname: name (or format string) must be at least IFNAMSIZ
1078 * Change name of a device, can pass format strings "eth%d".
1081 int dev_change_name(struct net_device
*dev
, const char *newname
)
1083 char oldname
[IFNAMSIZ
];
1089 BUG_ON(!dev_net(dev
));
1092 if (dev
->flags
& IFF_UP
)
1095 write_seqcount_begin(&devnet_rename_seq
);
1097 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0) {
1098 write_seqcount_end(&devnet_rename_seq
);
1102 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
1104 err
= dev_get_valid_name(net
, dev
, newname
);
1106 write_seqcount_end(&devnet_rename_seq
);
1111 ret
= device_rename(&dev
->dev
, dev
->name
);
1113 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1114 write_seqcount_end(&devnet_rename_seq
);
1118 write_seqcount_end(&devnet_rename_seq
);
1120 write_lock_bh(&dev_base_lock
);
1121 hlist_del_rcu(&dev
->name_hlist
);
1122 write_unlock_bh(&dev_base_lock
);
1126 write_lock_bh(&dev_base_lock
);
1127 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
1128 write_unlock_bh(&dev_base_lock
);
1130 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
1131 ret
= notifier_to_errno(ret
);
1134 /* err >= 0 after dev_alloc_name() or stores the first errno */
1137 write_seqcount_begin(&devnet_rename_seq
);
1138 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1141 pr_err("%s: name change rollback failed: %d\n",
1150 * dev_set_alias - change ifalias of a device
1152 * @alias: name up to IFALIASZ
1153 * @len: limit of bytes to copy from info
1155 * Set ifalias for a device,
1157 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
1163 if (len
>= IFALIASZ
)
1167 kfree(dev
->ifalias
);
1168 dev
->ifalias
= NULL
;
1172 new_ifalias
= krealloc(dev
->ifalias
, len
+ 1, GFP_KERNEL
);
1175 dev
->ifalias
= new_ifalias
;
1177 strlcpy(dev
->ifalias
, alias
, len
+1);
1183 * netdev_features_change - device changes features
1184 * @dev: device to cause notification
1186 * Called to indicate a device has changed features.
1188 void netdev_features_change(struct net_device
*dev
)
1190 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1192 EXPORT_SYMBOL(netdev_features_change
);
1195 * netdev_state_change - device changes state
1196 * @dev: device to cause notification
1198 * Called to indicate a device has changed state. This function calls
1199 * the notifier chains for netdev_chain and sends a NEWLINK message
1200 * to the routing socket.
1202 void netdev_state_change(struct net_device
*dev
)
1204 if (dev
->flags
& IFF_UP
) {
1205 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
1206 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0, GFP_KERNEL
);
1209 EXPORT_SYMBOL(netdev_state_change
);
1212 * netdev_notify_peers - notify network peers about existence of @dev
1213 * @dev: network device
1215 * Generate traffic such that interested network peers are aware of
1216 * @dev, such as by generating a gratuitous ARP. This may be used when
1217 * a device wants to inform the rest of the network about some sort of
1218 * reconfiguration such as a failover event or virtual machine
1221 void netdev_notify_peers(struct net_device
*dev
)
1224 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
, dev
);
1227 EXPORT_SYMBOL(netdev_notify_peers
);
1229 static int __dev_open(struct net_device
*dev
)
1231 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1236 if (!netif_device_present(dev
))
1239 /* Block netpoll from trying to do any rx path servicing.
1240 * If we don't do this there is a chance ndo_poll_controller
1241 * or ndo_poll may be running while we open the device
1243 netpoll_rx_disable(dev
);
1245 ret
= call_netdevice_notifiers(NETDEV_PRE_UP
, dev
);
1246 ret
= notifier_to_errno(ret
);
1250 set_bit(__LINK_STATE_START
, &dev
->state
);
1252 if (ops
->ndo_validate_addr
)
1253 ret
= ops
->ndo_validate_addr(dev
);
1255 if (!ret
&& ops
->ndo_open
)
1256 ret
= ops
->ndo_open(dev
);
1258 netpoll_rx_enable(dev
);
1261 clear_bit(__LINK_STATE_START
, &dev
->state
);
1263 dev
->flags
|= IFF_UP
;
1264 net_dmaengine_get();
1265 dev_set_rx_mode(dev
);
1267 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
1274 * dev_open - prepare an interface for use.
1275 * @dev: device to open
1277 * Takes a device from down to up state. The device's private open
1278 * function is invoked and then the multicast lists are loaded. Finally
1279 * the device is moved into the up state and a %NETDEV_UP message is
1280 * sent to the netdev notifier chain.
1282 * Calling this function on an active interface is a nop. On a failure
1283 * a negative errno code is returned.
1285 int dev_open(struct net_device
*dev
)
1289 if (dev
->flags
& IFF_UP
)
1292 ret
= __dev_open(dev
);
1296 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
, GFP_KERNEL
);
1297 call_netdevice_notifiers(NETDEV_UP
, dev
);
1301 EXPORT_SYMBOL(dev_open
);
1303 static int __dev_close_many(struct list_head
*head
)
1305 struct net_device
*dev
;
1310 list_for_each_entry(dev
, head
, close_list
) {
1311 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1313 clear_bit(__LINK_STATE_START
, &dev
->state
);
1315 /* Synchronize to scheduled poll. We cannot touch poll list, it
1316 * can be even on different cpu. So just clear netif_running().
1318 * dev->stop() will invoke napi_disable() on all of it's
1319 * napi_struct instances on this device.
1321 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1324 dev_deactivate_many(head
);
1326 list_for_each_entry(dev
, head
, close_list
) {
1327 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1330 * Call the device specific close. This cannot fail.
1331 * Only if device is UP
1333 * We allow it to be called even after a DETACH hot-plug
1339 dev
->flags
&= ~IFF_UP
;
1340 net_dmaengine_put();
1346 static int __dev_close(struct net_device
*dev
)
1351 /* Temporarily disable netpoll until the interface is down */
1352 netpoll_rx_disable(dev
);
1354 list_add(&dev
->close_list
, &single
);
1355 retval
= __dev_close_many(&single
);
1358 netpoll_rx_enable(dev
);
1362 static int dev_close_many(struct list_head
*head
)
1364 struct net_device
*dev
, *tmp
;
1366 /* Remove the devices that don't need to be closed */
1367 list_for_each_entry_safe(dev
, tmp
, head
, close_list
)
1368 if (!(dev
->flags
& IFF_UP
))
1369 list_del_init(&dev
->close_list
);
1371 __dev_close_many(head
);
1373 list_for_each_entry_safe(dev
, tmp
, head
, close_list
) {
1374 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
, GFP_KERNEL
);
1375 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1376 list_del_init(&dev
->close_list
);
1383 * dev_close - shutdown an interface.
1384 * @dev: device to shutdown
1386 * This function moves an active device into down state. A
1387 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1388 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1391 int dev_close(struct net_device
*dev
)
1393 if (dev
->flags
& IFF_UP
) {
1396 /* Block netpoll rx while the interface is going down */
1397 netpoll_rx_disable(dev
);
1399 list_add(&dev
->close_list
, &single
);
1400 dev_close_many(&single
);
1403 netpoll_rx_enable(dev
);
1407 EXPORT_SYMBOL(dev_close
);
1411 * dev_disable_lro - disable Large Receive Offload on a device
1414 * Disable Large Receive Offload (LRO) on a net device. Must be
1415 * called under RTNL. This is needed if received packets may be
1416 * forwarded to another interface.
1418 void dev_disable_lro(struct net_device
*dev
)
1421 * If we're trying to disable lro on a vlan device
1422 * use the underlying physical device instead
1424 if (is_vlan_dev(dev
))
1425 dev
= vlan_dev_real_dev(dev
);
1427 dev
->wanted_features
&= ~NETIF_F_LRO
;
1428 netdev_update_features(dev
);
1430 if (unlikely(dev
->features
& NETIF_F_LRO
))
1431 netdev_WARN(dev
, "failed to disable LRO!\n");
1433 EXPORT_SYMBOL(dev_disable_lro
);
1435 static int call_netdevice_notifier(struct notifier_block
*nb
, unsigned long val
,
1436 struct net_device
*dev
)
1438 struct netdev_notifier_info info
;
1440 netdev_notifier_info_init(&info
, dev
);
1441 return nb
->notifier_call(nb
, val
, &info
);
1444 static int dev_boot_phase
= 1;
1447 * register_netdevice_notifier - register a network notifier block
1450 * Register a notifier to be called when network device events occur.
1451 * The notifier passed is linked into the kernel structures and must
1452 * not be reused until it has been unregistered. A negative errno code
1453 * is returned on a failure.
1455 * When registered all registration and up events are replayed
1456 * to the new notifier to allow device to have a race free
1457 * view of the network device list.
1460 int register_netdevice_notifier(struct notifier_block
*nb
)
1462 struct net_device
*dev
;
1463 struct net_device
*last
;
1468 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1474 for_each_netdev(net
, dev
) {
1475 err
= call_netdevice_notifier(nb
, NETDEV_REGISTER
, dev
);
1476 err
= notifier_to_errno(err
);
1480 if (!(dev
->flags
& IFF_UP
))
1483 call_netdevice_notifier(nb
, NETDEV_UP
, dev
);
1494 for_each_netdev(net
, dev
) {
1498 if (dev
->flags
& IFF_UP
) {
1499 call_netdevice_notifier(nb
, NETDEV_GOING_DOWN
,
1501 call_netdevice_notifier(nb
, NETDEV_DOWN
, dev
);
1503 call_netdevice_notifier(nb
, NETDEV_UNREGISTER
, dev
);
1508 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1511 EXPORT_SYMBOL(register_netdevice_notifier
);
1514 * unregister_netdevice_notifier - unregister a network notifier block
1517 * Unregister a notifier previously registered by
1518 * register_netdevice_notifier(). The notifier is unlinked into the
1519 * kernel structures and may then be reused. A negative errno code
1520 * is returned on a failure.
1522 * After unregistering unregister and down device events are synthesized
1523 * for all devices on the device list to the removed notifier to remove
1524 * the need for special case cleanup code.
1527 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1529 struct net_device
*dev
;
1534 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1539 for_each_netdev(net
, dev
) {
1540 if (dev
->flags
& IFF_UP
) {
1541 call_netdevice_notifier(nb
, NETDEV_GOING_DOWN
,
1543 call_netdevice_notifier(nb
, NETDEV_DOWN
, dev
);
1545 call_netdevice_notifier(nb
, NETDEV_UNREGISTER
, dev
);
1552 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1555 * call_netdevice_notifiers_info - call all network notifier blocks
1556 * @val: value passed unmodified to notifier function
1557 * @dev: net_device pointer passed unmodified to notifier function
1558 * @info: notifier information data
1560 * Call all network notifier blocks. Parameters and return value
1561 * are as for raw_notifier_call_chain().
1564 int call_netdevice_notifiers_info(unsigned long val
, struct net_device
*dev
,
1565 struct netdev_notifier_info
*info
)
1568 netdev_notifier_info_init(info
, dev
);
1569 return raw_notifier_call_chain(&netdev_chain
, val
, info
);
1571 EXPORT_SYMBOL(call_netdevice_notifiers_info
);
1574 * call_netdevice_notifiers - call all network notifier blocks
1575 * @val: value passed unmodified to notifier function
1576 * @dev: net_device pointer passed unmodified to notifier function
1578 * Call all network notifier blocks. Parameters and return value
1579 * are as for raw_notifier_call_chain().
1582 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1584 struct netdev_notifier_info info
;
1586 return call_netdevice_notifiers_info(val
, dev
, &info
);
1588 EXPORT_SYMBOL(call_netdevice_notifiers
);
1590 static struct static_key netstamp_needed __read_mostly
;
1591 #ifdef HAVE_JUMP_LABEL
1592 /* We are not allowed to call static_key_slow_dec() from irq context
1593 * If net_disable_timestamp() is called from irq context, defer the
1594 * static_key_slow_dec() calls.
1596 static atomic_t netstamp_needed_deferred
;
1599 void net_enable_timestamp(void)
1601 #ifdef HAVE_JUMP_LABEL
1602 int deferred
= atomic_xchg(&netstamp_needed_deferred
, 0);
1606 static_key_slow_dec(&netstamp_needed
);
1610 static_key_slow_inc(&netstamp_needed
);
1612 EXPORT_SYMBOL(net_enable_timestamp
);
1614 void net_disable_timestamp(void)
1616 #ifdef HAVE_JUMP_LABEL
1617 if (in_interrupt()) {
1618 atomic_inc(&netstamp_needed_deferred
);
1622 static_key_slow_dec(&netstamp_needed
);
1624 EXPORT_SYMBOL(net_disable_timestamp
);
1626 static inline void net_timestamp_set(struct sk_buff
*skb
)
1628 skb
->tstamp
.tv64
= 0;
1629 if (static_key_false(&netstamp_needed
))
1630 __net_timestamp(skb
);
1633 #define net_timestamp_check(COND, SKB) \
1634 if (static_key_false(&netstamp_needed)) { \
1635 if ((COND) && !(SKB)->tstamp.tv64) \
1636 __net_timestamp(SKB); \
1639 static inline bool is_skb_forwardable(struct net_device *dev,
1640 struct sk_buff
*skb
)
1644 if (!(dev
->flags
& IFF_UP
))
1647 len
= dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
;
1648 if (skb
->len
<= len
)
1651 /* if TSO is enabled, we don't care about the length as the packet
1652 * could be forwarded without being segmented before
1654 if (skb_is_gso(skb
))
1661 * dev_forward_skb - loopback an skb to another netif
1663 * @dev: destination network device
1664 * @skb: buffer to forward
1667 * NET_RX_SUCCESS (no congestion)
1668 * NET_RX_DROP (packet was dropped, but freed)
1670 * dev_forward_skb can be used for injecting an skb from the
1671 * start_xmit function of one device into the receive queue
1672 * of another device.
1674 * The receiving device may be in another namespace, so
1675 * we have to clear all information in the skb that could
1676 * impact namespace isolation.
1678 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1680 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
1681 if (skb_copy_ubufs(skb
, GFP_ATOMIC
)) {
1682 atomic_long_inc(&dev
->rx_dropped
);
1688 if (unlikely(!is_skb_forwardable(dev
, skb
))) {
1689 atomic_long_inc(&dev
->rx_dropped
);
1694 skb_scrub_packet(skb
, true);
1695 skb
->protocol
= eth_type_trans(skb
, dev
);
1697 return netif_rx(skb
);
1699 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1701 static inline int deliver_skb(struct sk_buff
*skb
,
1702 struct packet_type
*pt_prev
,
1703 struct net_device
*orig_dev
)
1705 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
1707 atomic_inc(&skb
->users
);
1708 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1711 static inline bool skb_loop_sk(struct packet_type
*ptype
, struct sk_buff
*skb
)
1713 if (!ptype
->af_packet_priv
|| !skb
->sk
)
1716 if (ptype
->id_match
)
1717 return ptype
->id_match(ptype
, skb
->sk
);
1718 else if ((struct sock
*)ptype
->af_packet_priv
== skb
->sk
)
1725 * Support routine. Sends outgoing frames to any network
1726 * taps currently in use.
1729 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1731 struct packet_type
*ptype
;
1732 struct sk_buff
*skb2
= NULL
;
1733 struct packet_type
*pt_prev
= NULL
;
1736 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1737 /* Never send packets back to the socket
1738 * they originated from - MvS (miquels@drinkel.ow.org)
1740 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1741 (!skb_loop_sk(ptype
, skb
))) {
1743 deliver_skb(skb2
, pt_prev
, skb
->dev
);
1748 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1752 net_timestamp_set(skb2
);
1754 /* skb->nh should be correctly
1755 set by sender, so that the second statement is
1756 just protection against buggy protocols.
1758 skb_reset_mac_header(skb2
);
1760 if (skb_network_header(skb2
) < skb2
->data
||
1761 skb_network_header(skb2
) > skb_tail_pointer(skb2
)) {
1762 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1763 ntohs(skb2
->protocol
),
1765 skb_reset_network_header(skb2
);
1768 skb2
->transport_header
= skb2
->network_header
;
1769 skb2
->pkt_type
= PACKET_OUTGOING
;
1774 pt_prev
->func(skb2
, skb
->dev
, pt_prev
, skb
->dev
);
1779 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1780 * @dev: Network device
1781 * @txq: number of queues available
1783 * If real_num_tx_queues is changed the tc mappings may no longer be
1784 * valid. To resolve this verify the tc mapping remains valid and if
1785 * not NULL the mapping. With no priorities mapping to this
1786 * offset/count pair it will no longer be used. In the worst case TC0
1787 * is invalid nothing can be done so disable priority mappings. If is
1788 * expected that drivers will fix this mapping if they can before
1789 * calling netif_set_real_num_tx_queues.
1791 static void netif_setup_tc(struct net_device
*dev
, unsigned int txq
)
1794 struct netdev_tc_txq
*tc
= &dev
->tc_to_txq
[0];
1796 /* If TC0 is invalidated disable TC mapping */
1797 if (tc
->offset
+ tc
->count
> txq
) {
1798 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1803 /* Invalidated prio to tc mappings set to TC0 */
1804 for (i
= 1; i
< TC_BITMASK
+ 1; i
++) {
1805 int q
= netdev_get_prio_tc_map(dev
, i
);
1807 tc
= &dev
->tc_to_txq
[q
];
1808 if (tc
->offset
+ tc
->count
> txq
) {
1809 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1811 netdev_set_prio_tc_map(dev
, i
, 0);
1817 static DEFINE_MUTEX(xps_map_mutex
);
1818 #define xmap_dereference(P) \
1819 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1821 static struct xps_map
*remove_xps_queue(struct xps_dev_maps
*dev_maps
,
1824 struct xps_map
*map
= NULL
;
1828 map
= xmap_dereference(dev_maps
->cpu_map
[cpu
]);
1830 for (pos
= 0; map
&& pos
< map
->len
; pos
++) {
1831 if (map
->queues
[pos
] == index
) {
1833 map
->queues
[pos
] = map
->queues
[--map
->len
];
1835 RCU_INIT_POINTER(dev_maps
->cpu_map
[cpu
], NULL
);
1836 kfree_rcu(map
, rcu
);
1846 static void netif_reset_xps_queues_gt(struct net_device
*dev
, u16 index
)
1848 struct xps_dev_maps
*dev_maps
;
1850 bool active
= false;
1852 mutex_lock(&xps_map_mutex
);
1853 dev_maps
= xmap_dereference(dev
->xps_maps
);
1858 for_each_possible_cpu(cpu
) {
1859 for (i
= index
; i
< dev
->num_tx_queues
; i
++) {
1860 if (!remove_xps_queue(dev_maps
, cpu
, i
))
1863 if (i
== dev
->num_tx_queues
)
1868 RCU_INIT_POINTER(dev
->xps_maps
, NULL
);
1869 kfree_rcu(dev_maps
, rcu
);
1872 for (i
= index
; i
< dev
->num_tx_queues
; i
++)
1873 netdev_queue_numa_node_write(netdev_get_tx_queue(dev
, i
),
1877 mutex_unlock(&xps_map_mutex
);
1880 static struct xps_map
*expand_xps_map(struct xps_map
*map
,
1883 struct xps_map
*new_map
;
1884 int alloc_len
= XPS_MIN_MAP_ALLOC
;
1887 for (pos
= 0; map
&& pos
< map
->len
; pos
++) {
1888 if (map
->queues
[pos
] != index
)
1893 /* Need to add queue to this CPU's existing map */
1895 if (pos
< map
->alloc_len
)
1898 alloc_len
= map
->alloc_len
* 2;
1901 /* Need to allocate new map to store queue on this CPU's map */
1902 new_map
= kzalloc_node(XPS_MAP_SIZE(alloc_len
), GFP_KERNEL
,
1907 for (i
= 0; i
< pos
; i
++)
1908 new_map
->queues
[i
] = map
->queues
[i
];
1909 new_map
->alloc_len
= alloc_len
;
1915 int netif_set_xps_queue(struct net_device
*dev
, const struct cpumask
*mask
,
1918 struct xps_dev_maps
*dev_maps
, *new_dev_maps
= NULL
;
1919 struct xps_map
*map
, *new_map
;
1920 int maps_sz
= max_t(unsigned int, XPS_DEV_MAPS_SIZE
, L1_CACHE_BYTES
);
1921 int cpu
, numa_node_id
= -2;
1922 bool active
= false;
1924 mutex_lock(&xps_map_mutex
);
1926 dev_maps
= xmap_dereference(dev
->xps_maps
);
1928 /* allocate memory for queue storage */
1929 for_each_online_cpu(cpu
) {
1930 if (!cpumask_test_cpu(cpu
, mask
))
1934 new_dev_maps
= kzalloc(maps_sz
, GFP_KERNEL
);
1935 if (!new_dev_maps
) {
1936 mutex_unlock(&xps_map_mutex
);
1940 map
= dev_maps
? xmap_dereference(dev_maps
->cpu_map
[cpu
]) :
1943 map
= expand_xps_map(map
, cpu
, index
);
1947 RCU_INIT_POINTER(new_dev_maps
->cpu_map
[cpu
], map
);
1951 goto out_no_new_maps
;
1953 for_each_possible_cpu(cpu
) {
1954 if (cpumask_test_cpu(cpu
, mask
) && cpu_online(cpu
)) {
1955 /* add queue to CPU maps */
1958 map
= xmap_dereference(new_dev_maps
->cpu_map
[cpu
]);
1959 while ((pos
< map
->len
) && (map
->queues
[pos
] != index
))
1962 if (pos
== map
->len
)
1963 map
->queues
[map
->len
++] = index
;
1965 if (numa_node_id
== -2)
1966 numa_node_id
= cpu_to_node(cpu
);
1967 else if (numa_node_id
!= cpu_to_node(cpu
))
1970 } else if (dev_maps
) {
1971 /* fill in the new device map from the old device map */
1972 map
= xmap_dereference(dev_maps
->cpu_map
[cpu
]);
1973 RCU_INIT_POINTER(new_dev_maps
->cpu_map
[cpu
], map
);
1978 rcu_assign_pointer(dev
->xps_maps
, new_dev_maps
);
1980 /* Cleanup old maps */
1982 for_each_possible_cpu(cpu
) {
1983 new_map
= xmap_dereference(new_dev_maps
->cpu_map
[cpu
]);
1984 map
= xmap_dereference(dev_maps
->cpu_map
[cpu
]);
1985 if (map
&& map
!= new_map
)
1986 kfree_rcu(map
, rcu
);
1989 kfree_rcu(dev_maps
, rcu
);
1992 dev_maps
= new_dev_maps
;
1996 /* update Tx queue numa node */
1997 netdev_queue_numa_node_write(netdev_get_tx_queue(dev
, index
),
1998 (numa_node_id
>= 0) ? numa_node_id
:
2004 /* removes queue from unused CPUs */
2005 for_each_possible_cpu(cpu
) {
2006 if (cpumask_test_cpu(cpu
, mask
) && cpu_online(cpu
))
2009 if (remove_xps_queue(dev_maps
, cpu
, index
))
2013 /* free map if not active */
2015 RCU_INIT_POINTER(dev
->xps_maps
, NULL
);
2016 kfree_rcu(dev_maps
, rcu
);
2020 mutex_unlock(&xps_map_mutex
);
2024 /* remove any maps that we added */
2025 for_each_possible_cpu(cpu
) {
2026 new_map
= xmap_dereference(new_dev_maps
->cpu_map
[cpu
]);
2027 map
= dev_maps
? xmap_dereference(dev_maps
->cpu_map
[cpu
]) :
2029 if (new_map
&& new_map
!= map
)
2033 mutex_unlock(&xps_map_mutex
);
2035 kfree(new_dev_maps
);
2038 EXPORT_SYMBOL(netif_set_xps_queue
);
2042 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2043 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2045 int netif_set_real_num_tx_queues(struct net_device
*dev
, unsigned int txq
)
2049 if (txq
< 1 || txq
> dev
->num_tx_queues
)
2052 if (dev
->reg_state
== NETREG_REGISTERED
||
2053 dev
->reg_state
== NETREG_UNREGISTERING
) {
2056 rc
= netdev_queue_update_kobjects(dev
, dev
->real_num_tx_queues
,
2062 netif_setup_tc(dev
, txq
);
2064 if (txq
< dev
->real_num_tx_queues
) {
2065 qdisc_reset_all_tx_gt(dev
, txq
);
2067 netif_reset_xps_queues_gt(dev
, txq
);
2072 dev
->real_num_tx_queues
= txq
;
2075 EXPORT_SYMBOL(netif_set_real_num_tx_queues
);
2079 * netif_set_real_num_rx_queues - set actual number of RX queues used
2080 * @dev: Network device
2081 * @rxq: Actual number of RX queues
2083 * This must be called either with the rtnl_lock held or before
2084 * registration of the net device. Returns 0 on success, or a
2085 * negative error code. If called before registration, it always
2088 int netif_set_real_num_rx_queues(struct net_device
*dev
, unsigned int rxq
)
2092 if (rxq
< 1 || rxq
> dev
->num_rx_queues
)
2095 if (dev
->reg_state
== NETREG_REGISTERED
) {
2098 rc
= net_rx_queue_update_kobjects(dev
, dev
->real_num_rx_queues
,
2104 dev
->real_num_rx_queues
= rxq
;
2107 EXPORT_SYMBOL(netif_set_real_num_rx_queues
);
2111 * netif_get_num_default_rss_queues - default number of RSS queues
2113 * This routine should set an upper limit on the number of RSS queues
2114 * used by default by multiqueue devices.
2116 int netif_get_num_default_rss_queues(void)
2118 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES
, num_online_cpus());
2120 EXPORT_SYMBOL(netif_get_num_default_rss_queues
);
2122 static inline void __netif_reschedule(struct Qdisc
*q
)
2124 struct softnet_data
*sd
;
2125 unsigned long flags
;
2127 local_irq_save(flags
);
2128 sd
= &__get_cpu_var(softnet_data
);
2129 q
->next_sched
= NULL
;
2130 *sd
->output_queue_tailp
= q
;
2131 sd
->output_queue_tailp
= &q
->next_sched
;
2132 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
2133 local_irq_restore(flags
);
2136 void __netif_schedule(struct Qdisc
*q
)
2138 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
2139 __netif_reschedule(q
);
2141 EXPORT_SYMBOL(__netif_schedule
);
2143 void dev_kfree_skb_irq(struct sk_buff
*skb
)
2145 if (atomic_dec_and_test(&skb
->users
)) {
2146 struct softnet_data
*sd
;
2147 unsigned long flags
;
2149 local_irq_save(flags
);
2150 sd
= &__get_cpu_var(softnet_data
);
2151 skb
->next
= sd
->completion_queue
;
2152 sd
->completion_queue
= skb
;
2153 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
2154 local_irq_restore(flags
);
2157 EXPORT_SYMBOL(dev_kfree_skb_irq
);
2159 void dev_kfree_skb_any(struct sk_buff
*skb
)
2161 if (in_irq() || irqs_disabled())
2162 dev_kfree_skb_irq(skb
);
2166 EXPORT_SYMBOL(dev_kfree_skb_any
);
2170 * netif_device_detach - mark device as removed
2171 * @dev: network device
2173 * Mark device as removed from system and therefore no longer available.
2175 void netif_device_detach(struct net_device
*dev
)
2177 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
2178 netif_running(dev
)) {
2179 netif_tx_stop_all_queues(dev
);
2182 EXPORT_SYMBOL(netif_device_detach
);
2185 * netif_device_attach - mark device as attached
2186 * @dev: network device
2188 * Mark device as attached from system and restart if needed.
2190 void netif_device_attach(struct net_device
*dev
)
2192 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
2193 netif_running(dev
)) {
2194 netif_tx_wake_all_queues(dev
);
2195 __netdev_watchdog_up(dev
);
2198 EXPORT_SYMBOL(netif_device_attach
);
2200 static void skb_warn_bad_offload(const struct sk_buff
*skb
)
2202 static const netdev_features_t null_features
= 0;
2203 struct net_device
*dev
= skb
->dev
;
2204 const char *driver
= "";
2206 if (!net_ratelimit())
2209 if (dev
&& dev
->dev
.parent
)
2210 driver
= dev_driver_string(dev
->dev
.parent
);
2212 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2213 "gso_type=%d ip_summed=%d\n",
2214 driver
, dev
? &dev
->features
: &null_features
,
2215 skb
->sk
? &skb
->sk
->sk_route_caps
: &null_features
,
2216 skb
->len
, skb
->data_len
, skb_shinfo(skb
)->gso_size
,
2217 skb_shinfo(skb
)->gso_type
, skb
->ip_summed
);
2221 * Invalidate hardware checksum when packet is to be mangled, and
2222 * complete checksum manually on outgoing path.
2224 int skb_checksum_help(struct sk_buff
*skb
)
2227 int ret
= 0, offset
;
2229 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
2230 goto out_set_summed
;
2232 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
2233 skb_warn_bad_offload(skb
);
2237 /* Before computing a checksum, we should make sure no frag could
2238 * be modified by an external entity : checksum could be wrong.
2240 if (skb_has_shared_frag(skb
)) {
2241 ret
= __skb_linearize(skb
);
2246 offset
= skb_checksum_start_offset(skb
);
2247 BUG_ON(offset
>= skb_headlen(skb
));
2248 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
2250 offset
+= skb
->csum_offset
;
2251 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
2253 if (skb_cloned(skb
) &&
2254 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
2255 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2260 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
2262 skb
->ip_summed
= CHECKSUM_NONE
;
2266 EXPORT_SYMBOL(skb_checksum_help
);
2268 __be16
skb_network_protocol(struct sk_buff
*skb
)
2270 __be16 type
= skb
->protocol
;
2271 int vlan_depth
= ETH_HLEN
;
2273 /* Tunnel gso handlers can set protocol to ethernet. */
2274 if (type
== htons(ETH_P_TEB
)) {
2277 if (unlikely(!pskb_may_pull(skb
, sizeof(struct ethhdr
))))
2280 eth
= (struct ethhdr
*)skb_mac_header(skb
);
2281 type
= eth
->h_proto
;
2284 while (type
== htons(ETH_P_8021Q
) || type
== htons(ETH_P_8021AD
)) {
2285 struct vlan_hdr
*vh
;
2287 if (unlikely(!pskb_may_pull(skb
, vlan_depth
+ VLAN_HLEN
)))
2290 vh
= (struct vlan_hdr
*)(skb
->data
+ vlan_depth
);
2291 type
= vh
->h_vlan_encapsulated_proto
;
2292 vlan_depth
+= VLAN_HLEN
;
2299 * skb_mac_gso_segment - mac layer segmentation handler.
2300 * @skb: buffer to segment
2301 * @features: features for the output path (see dev->features)
2303 struct sk_buff
*skb_mac_gso_segment(struct sk_buff
*skb
,
2304 netdev_features_t features
)
2306 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
2307 struct packet_offload
*ptype
;
2308 __be16 type
= skb_network_protocol(skb
);
2310 if (unlikely(!type
))
2311 return ERR_PTR(-EINVAL
);
2313 __skb_pull(skb
, skb
->mac_len
);
2316 list_for_each_entry_rcu(ptype
, &offload_base
, list
) {
2317 if (ptype
->type
== type
&& ptype
->callbacks
.gso_segment
) {
2318 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
2321 err
= ptype
->callbacks
.gso_send_check(skb
);
2322 segs
= ERR_PTR(err
);
2323 if (err
|| skb_gso_ok(skb
, features
))
2325 __skb_push(skb
, (skb
->data
-
2326 skb_network_header(skb
)));
2328 segs
= ptype
->callbacks
.gso_segment(skb
, features
);
2334 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
2338 EXPORT_SYMBOL(skb_mac_gso_segment
);
2341 /* openvswitch calls this on rx path, so we need a different check.
2343 static inline bool skb_needs_check(struct sk_buff
*skb
, bool tx_path
)
2346 return skb
->ip_summed
!= CHECKSUM_PARTIAL
;
2348 return skb
->ip_summed
== CHECKSUM_NONE
;
2352 * __skb_gso_segment - Perform segmentation on skb.
2353 * @skb: buffer to segment
2354 * @features: features for the output path (see dev->features)
2355 * @tx_path: whether it is called in TX path
2357 * This function segments the given skb and returns a list of segments.
2359 * It may return NULL if the skb requires no segmentation. This is
2360 * only possible when GSO is used for verifying header integrity.
2362 struct sk_buff
*__skb_gso_segment(struct sk_buff
*skb
,
2363 netdev_features_t features
, bool tx_path
)
2365 if (unlikely(skb_needs_check(skb
, tx_path
))) {
2368 skb_warn_bad_offload(skb
);
2370 if (skb_header_cloned(skb
) &&
2371 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
2372 return ERR_PTR(err
);
2375 SKB_GSO_CB(skb
)->mac_offset
= skb_headroom(skb
);
2376 SKB_GSO_CB(skb
)->encap_level
= 0;
2378 skb_reset_mac_header(skb
);
2379 skb_reset_mac_len(skb
);
2381 return skb_mac_gso_segment(skb
, features
);
2383 EXPORT_SYMBOL(__skb_gso_segment
);
2385 /* Take action when hardware reception checksum errors are detected. */
2387 void netdev_rx_csum_fault(struct net_device
*dev
)
2389 if (net_ratelimit()) {
2390 pr_err("%s: hw csum failure\n", dev
? dev
->name
: "<unknown>");
2394 EXPORT_SYMBOL(netdev_rx_csum_fault
);
2397 /* Actually, we should eliminate this check as soon as we know, that:
2398 * 1. IOMMU is present and allows to map all the memory.
2399 * 2. No high memory really exists on this machine.
2402 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
2404 #ifdef CONFIG_HIGHMEM
2406 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
2407 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2408 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2409 if (PageHighMem(skb_frag_page(frag
)))
2414 if (PCI_DMA_BUS_IS_PHYS
) {
2415 struct device
*pdev
= dev
->dev
.parent
;
2419 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2420 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2421 dma_addr_t addr
= page_to_phys(skb_frag_page(frag
));
2422 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
2431 void (*destructor
)(struct sk_buff
*skb
);
2434 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2436 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
2438 struct dev_gso_cb
*cb
;
2441 struct sk_buff
*nskb
= skb
->next
;
2443 skb
->next
= nskb
->next
;
2446 } while (skb
->next
);
2448 cb
= DEV_GSO_CB(skb
);
2450 cb
->destructor(skb
);
2454 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2455 * @skb: buffer to segment
2456 * @features: device features as applicable to this skb
2458 * This function segments the given skb and stores the list of segments
2461 static int dev_gso_segment(struct sk_buff
*skb
, netdev_features_t features
)
2463 struct sk_buff
*segs
;
2465 segs
= skb_gso_segment(skb
, features
);
2467 /* Verifying header integrity only. */
2472 return PTR_ERR(segs
);
2475 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
2476 skb
->destructor
= dev_gso_skb_destructor
;
2481 static netdev_features_t
harmonize_features(struct sk_buff
*skb
,
2482 netdev_features_t features
)
2484 if (skb
->ip_summed
!= CHECKSUM_NONE
&&
2485 !can_checksum_protocol(features
, skb_network_protocol(skb
))) {
2486 features
&= ~NETIF_F_ALL_CSUM
;
2487 } else if (illegal_highdma(skb
->dev
, skb
)) {
2488 features
&= ~NETIF_F_SG
;
2494 netdev_features_t
netif_skb_features(struct sk_buff
*skb
)
2496 __be16 protocol
= skb
->protocol
;
2497 netdev_features_t features
= skb
->dev
->features
;
2499 if (skb_shinfo(skb
)->gso_segs
> skb
->dev
->gso_max_segs
)
2500 features
&= ~NETIF_F_GSO_MASK
;
2502 if (protocol
== htons(ETH_P_8021Q
) || protocol
== htons(ETH_P_8021AD
)) {
2503 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
2504 protocol
= veh
->h_vlan_encapsulated_proto
;
2505 } else if (!vlan_tx_tag_present(skb
)) {
2506 return harmonize_features(skb
, features
);
2509 features
&= (skb
->dev
->vlan_features
| NETIF_F_HW_VLAN_CTAG_TX
|
2510 NETIF_F_HW_VLAN_STAG_TX
);
2512 if (protocol
== htons(ETH_P_8021Q
) || protocol
== htons(ETH_P_8021AD
))
2513 features
&= NETIF_F_SG
| NETIF_F_HIGHDMA
| NETIF_F_FRAGLIST
|
2514 NETIF_F_GEN_CSUM
| NETIF_F_HW_VLAN_CTAG_TX
|
2515 NETIF_F_HW_VLAN_STAG_TX
;
2517 return harmonize_features(skb
, features
);
2519 EXPORT_SYMBOL(netif_skb_features
);
2522 * Returns true if either:
2523 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2524 * 2. skb is fragmented and the device does not support SG.
2526 static inline int skb_needs_linearize(struct sk_buff
*skb
,
2527 netdev_features_t features
)
2529 return skb_is_nonlinear(skb
) &&
2530 ((skb_has_frag_list(skb
) &&
2531 !(features
& NETIF_F_FRAGLIST
)) ||
2532 (skb_shinfo(skb
)->nr_frags
&&
2533 !(features
& NETIF_F_SG
)));
2536 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
2537 struct netdev_queue
*txq
, void *accel_priv
)
2539 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2540 int rc
= NETDEV_TX_OK
;
2541 unsigned int skb_len
;
2543 if (likely(!skb
->next
)) {
2544 netdev_features_t features
;
2547 * If device doesn't need skb->dst, release it right now while
2548 * its hot in this cpu cache
2550 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2553 features
= netif_skb_features(skb
);
2555 if (vlan_tx_tag_present(skb
) &&
2556 !vlan_hw_offload_capable(features
, skb
->vlan_proto
)) {
2557 skb
= __vlan_put_tag(skb
, skb
->vlan_proto
,
2558 vlan_tx_tag_get(skb
));
2565 /* If encapsulation offload request, verify we are testing
2566 * hardware encapsulation features instead of standard
2567 * features for the netdev
2569 if (skb
->encapsulation
)
2570 features
&= dev
->hw_enc_features
;
2572 if (netif_needs_gso(skb
, features
)) {
2573 if (unlikely(dev_gso_segment(skb
, features
)))
2578 if (skb_needs_linearize(skb
, features
) &&
2579 __skb_linearize(skb
))
2582 /* If packet is not checksummed and device does not
2583 * support checksumming for this protocol, complete
2584 * checksumming here.
2586 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2587 if (skb
->encapsulation
)
2588 skb_set_inner_transport_header(skb
,
2589 skb_checksum_start_offset(skb
));
2591 skb_set_transport_header(skb
,
2592 skb_checksum_start_offset(skb
));
2593 if (!(features
& NETIF_F_ALL_CSUM
) &&
2594 skb_checksum_help(skb
))
2599 if (!list_empty(&ptype_all
))
2600 dev_queue_xmit_nit(skb
, dev
);
2604 rc
= ops
->ndo_dfwd_start_xmit(skb
, dev
, accel_priv
);
2606 rc
= ops
->ndo_start_xmit(skb
, dev
);
2608 trace_net_dev_xmit(skb
, rc
, dev
, skb_len
);
2609 if (rc
== NETDEV_TX_OK
&& txq
)
2610 txq_trans_update(txq
);
2616 struct sk_buff
*nskb
= skb
->next
;
2618 skb
->next
= nskb
->next
;
2621 if (!list_empty(&ptype_all
))
2622 dev_queue_xmit_nit(nskb
, dev
);
2624 skb_len
= nskb
->len
;
2626 rc
= ops
->ndo_dfwd_start_xmit(nskb
, dev
, accel_priv
);
2628 rc
= ops
->ndo_start_xmit(nskb
, dev
);
2629 trace_net_dev_xmit(nskb
, rc
, dev
, skb_len
);
2630 if (unlikely(rc
!= NETDEV_TX_OK
)) {
2631 if (rc
& ~NETDEV_TX_MASK
)
2632 goto out_kfree_gso_skb
;
2633 nskb
->next
= skb
->next
;
2637 txq_trans_update(txq
);
2638 if (unlikely(netif_xmit_stopped(txq
) && skb
->next
))
2639 return NETDEV_TX_BUSY
;
2640 } while (skb
->next
);
2643 if (likely(skb
->next
== NULL
)) {
2644 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
2653 EXPORT_SYMBOL_GPL(dev_hard_start_xmit
);
2655 static void qdisc_pkt_len_init(struct sk_buff
*skb
)
2657 const struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
2659 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
2661 /* To get more precise estimation of bytes sent on wire,
2662 * we add to pkt_len the headers size of all segments
2664 if (shinfo
->gso_size
) {
2665 unsigned int hdr_len
;
2666 u16 gso_segs
= shinfo
->gso_segs
;
2668 /* mac layer + network layer */
2669 hdr_len
= skb_transport_header(skb
) - skb_mac_header(skb
);
2671 /* + transport layer */
2672 if (likely(shinfo
->gso_type
& (SKB_GSO_TCPV4
| SKB_GSO_TCPV6
)))
2673 hdr_len
+= tcp_hdrlen(skb
);
2675 hdr_len
+= sizeof(struct udphdr
);
2677 if (shinfo
->gso_type
& SKB_GSO_DODGY
)
2678 gso_segs
= DIV_ROUND_UP(skb
->len
- hdr_len
,
2681 qdisc_skb_cb(skb
)->pkt_len
+= (gso_segs
- 1) * hdr_len
;
2685 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2686 struct net_device
*dev
,
2687 struct netdev_queue
*txq
)
2689 spinlock_t
*root_lock
= qdisc_lock(q
);
2693 qdisc_pkt_len_init(skb
);
2694 qdisc_calculate_pkt_len(skb
, q
);
2696 * Heuristic to force contended enqueues to serialize on a
2697 * separate lock before trying to get qdisc main lock.
2698 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2699 * and dequeue packets faster.
2701 contended
= qdisc_is_running(q
);
2702 if (unlikely(contended
))
2703 spin_lock(&q
->busylock
);
2705 spin_lock(root_lock
);
2706 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2709 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2710 qdisc_run_begin(q
)) {
2712 * This is a work-conserving queue; there are no old skbs
2713 * waiting to be sent out; and the qdisc is not running -
2714 * xmit the skb directly.
2716 if (!(dev
->priv_flags
& IFF_XMIT_DST_RELEASE
))
2719 qdisc_bstats_update(q
, skb
);
2721 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
)) {
2722 if (unlikely(contended
)) {
2723 spin_unlock(&q
->busylock
);
2730 rc
= NET_XMIT_SUCCESS
;
2733 rc
= q
->enqueue(skb
, q
) & NET_XMIT_MASK
;
2734 if (qdisc_run_begin(q
)) {
2735 if (unlikely(contended
)) {
2736 spin_unlock(&q
->busylock
);
2742 spin_unlock(root_lock
);
2743 if (unlikely(contended
))
2744 spin_unlock(&q
->busylock
);
2748 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2749 static void skb_update_prio(struct sk_buff
*skb
)
2751 struct netprio_map
*map
= rcu_dereference_bh(skb
->dev
->priomap
);
2753 if (!skb
->priority
&& skb
->sk
&& map
) {
2754 unsigned int prioidx
= skb
->sk
->sk_cgrp_prioidx
;
2756 if (prioidx
< map
->priomap_len
)
2757 skb
->priority
= map
->priomap
[prioidx
];
2761 #define skb_update_prio(skb)
2764 static DEFINE_PER_CPU(int, xmit_recursion
);
2765 #define RECURSION_LIMIT 10
2768 * dev_loopback_xmit - loop back @skb
2769 * @skb: buffer to transmit
2771 int dev_loopback_xmit(struct sk_buff
*skb
)
2773 skb_reset_mac_header(skb
);
2774 __skb_pull(skb
, skb_network_offset(skb
));
2775 skb
->pkt_type
= PACKET_LOOPBACK
;
2776 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
2777 WARN_ON(!skb_dst(skb
));
2782 EXPORT_SYMBOL(dev_loopback_xmit
);
2785 * dev_queue_xmit - transmit a buffer
2786 * @skb: buffer to transmit
2788 * Queue a buffer for transmission to a network device. The caller must
2789 * have set the device and priority and built the buffer before calling
2790 * this function. The function can be called from an interrupt.
2792 * A negative errno code is returned on a failure. A success does not
2793 * guarantee the frame will be transmitted as it may be dropped due
2794 * to congestion or traffic shaping.
2796 * -----------------------------------------------------------------------------------
2797 * I notice this method can also return errors from the queue disciplines,
2798 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2801 * Regardless of the return value, the skb is consumed, so it is currently
2802 * difficult to retry a send to this method. (You can bump the ref count
2803 * before sending to hold a reference for retry if you are careful.)
2805 * When calling this method, interrupts MUST be enabled. This is because
2806 * the BH enable code must have IRQs enabled so that it will not deadlock.
2809 int dev_queue_xmit(struct sk_buff
*skb
)
2811 struct net_device
*dev
= skb
->dev
;
2812 struct netdev_queue
*txq
;
2816 skb_reset_mac_header(skb
);
2818 /* Disable soft irqs for various locks below. Also
2819 * stops preemption for RCU.
2823 skb_update_prio(skb
);
2825 txq
= netdev_pick_tx(dev
, skb
);
2826 q
= rcu_dereference_bh(txq
->qdisc
);
2828 #ifdef CONFIG_NET_CLS_ACT
2829 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2831 trace_net_dev_queue(skb
);
2833 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2837 /* The device has no queue. Common case for software devices:
2838 loopback, all the sorts of tunnels...
2840 Really, it is unlikely that netif_tx_lock protection is necessary
2841 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2843 However, it is possible, that they rely on protection
2846 Check this and shot the lock. It is not prone from deadlocks.
2847 Either shot noqueue qdisc, it is even simpler 8)
2849 if (dev
->flags
& IFF_UP
) {
2850 int cpu
= smp_processor_id(); /* ok because BHs are off */
2852 if (txq
->xmit_lock_owner
!= cpu
) {
2854 if (__this_cpu_read(xmit_recursion
) > RECURSION_LIMIT
)
2855 goto recursion_alert
;
2857 HARD_TX_LOCK(dev
, txq
, cpu
);
2859 if (!netif_xmit_stopped(txq
)) {
2860 __this_cpu_inc(xmit_recursion
);
2861 rc
= dev_hard_start_xmit(skb
, dev
, txq
, NULL
);
2862 __this_cpu_dec(xmit_recursion
);
2863 if (dev_xmit_complete(rc
)) {
2864 HARD_TX_UNLOCK(dev
, txq
);
2868 HARD_TX_UNLOCK(dev
, txq
);
2869 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2872 /* Recursion is detected! It is possible,
2876 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2882 rcu_read_unlock_bh();
2887 rcu_read_unlock_bh();
2890 EXPORT_SYMBOL(dev_queue_xmit
);
2893 /*=======================================================================
2895 =======================================================================*/
2897 int netdev_max_backlog __read_mostly
= 1000;
2898 EXPORT_SYMBOL(netdev_max_backlog
);
2900 int netdev_tstamp_prequeue __read_mostly
= 1;
2901 int netdev_budget __read_mostly
= 300;
2902 int weight_p __read_mostly
= 64; /* old backlog weight */
2904 /* Called with irq disabled */
2905 static inline void ____napi_schedule(struct softnet_data
*sd
,
2906 struct napi_struct
*napi
)
2908 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
2909 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2914 /* One global table that all flow-based protocols share. */
2915 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
2916 EXPORT_SYMBOL(rps_sock_flow_table
);
2918 struct static_key rps_needed __read_mostly
;
2920 static struct rps_dev_flow
*
2921 set_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2922 struct rps_dev_flow
*rflow
, u16 next_cpu
)
2924 if (next_cpu
!= RPS_NO_CPU
) {
2925 #ifdef CONFIG_RFS_ACCEL
2926 struct netdev_rx_queue
*rxqueue
;
2927 struct rps_dev_flow_table
*flow_table
;
2928 struct rps_dev_flow
*old_rflow
;
2933 /* Should we steer this flow to a different hardware queue? */
2934 if (!skb_rx_queue_recorded(skb
) || !dev
->rx_cpu_rmap
||
2935 !(dev
->features
& NETIF_F_NTUPLE
))
2937 rxq_index
= cpu_rmap_lookup_index(dev
->rx_cpu_rmap
, next_cpu
);
2938 if (rxq_index
== skb_get_rx_queue(skb
))
2941 rxqueue
= dev
->_rx
+ rxq_index
;
2942 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2945 flow_id
= skb
->rxhash
& flow_table
->mask
;
2946 rc
= dev
->netdev_ops
->ndo_rx_flow_steer(dev
, skb
,
2947 rxq_index
, flow_id
);
2951 rflow
= &flow_table
->flows
[flow_id
];
2953 if (old_rflow
->filter
== rflow
->filter
)
2954 old_rflow
->filter
= RPS_NO_FILTER
;
2958 per_cpu(softnet_data
, next_cpu
).input_queue_head
;
2961 rflow
->cpu
= next_cpu
;
2966 * get_rps_cpu is called from netif_receive_skb and returns the target
2967 * CPU from the RPS map of the receiving queue for a given skb.
2968 * rcu_read_lock must be held on entry.
2970 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2971 struct rps_dev_flow
**rflowp
)
2973 struct netdev_rx_queue
*rxqueue
;
2974 struct rps_map
*map
;
2975 struct rps_dev_flow_table
*flow_table
;
2976 struct rps_sock_flow_table
*sock_flow_table
;
2980 if (skb_rx_queue_recorded(skb
)) {
2981 u16 index
= skb_get_rx_queue(skb
);
2982 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
2983 WARN_ONCE(dev
->real_num_rx_queues
> 1,
2984 "%s received packet on queue %u, but number "
2985 "of RX queues is %u\n",
2986 dev
->name
, index
, dev
->real_num_rx_queues
);
2989 rxqueue
= dev
->_rx
+ index
;
2993 map
= rcu_dereference(rxqueue
->rps_map
);
2995 if (map
->len
== 1 &&
2996 !rcu_access_pointer(rxqueue
->rps_flow_table
)) {
2997 tcpu
= map
->cpus
[0];
2998 if (cpu_online(tcpu
))
3002 } else if (!rcu_access_pointer(rxqueue
->rps_flow_table
)) {
3006 skb_reset_network_header(skb
);
3007 if (!skb_get_rxhash(skb
))
3010 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3011 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
3012 if (flow_table
&& sock_flow_table
) {
3014 struct rps_dev_flow
*rflow
;
3016 rflow
= &flow_table
->flows
[skb
->rxhash
& flow_table
->mask
];
3019 next_cpu
= sock_flow_table
->ents
[skb
->rxhash
&
3020 sock_flow_table
->mask
];
3023 * If the desired CPU (where last recvmsg was done) is
3024 * different from current CPU (one in the rx-queue flow
3025 * table entry), switch if one of the following holds:
3026 * - Current CPU is unset (equal to RPS_NO_CPU).
3027 * - Current CPU is offline.
3028 * - The current CPU's queue tail has advanced beyond the
3029 * last packet that was enqueued using this table entry.
3030 * This guarantees that all previous packets for the flow
3031 * have been dequeued, thus preserving in order delivery.
3033 if (unlikely(tcpu
!= next_cpu
) &&
3034 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
3035 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
3036 rflow
->last_qtail
)) >= 0)) {
3038 rflow
= set_rps_cpu(dev
, skb
, rflow
, next_cpu
);
3041 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
3049 tcpu
= map
->cpus
[((u64
) skb
->rxhash
* map
->len
) >> 32];
3051 if (cpu_online(tcpu
)) {
3061 #ifdef CONFIG_RFS_ACCEL
3064 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3065 * @dev: Device on which the filter was set
3066 * @rxq_index: RX queue index
3067 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3068 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3070 * Drivers that implement ndo_rx_flow_steer() should periodically call
3071 * this function for each installed filter and remove the filters for
3072 * which it returns %true.
3074 bool rps_may_expire_flow(struct net_device
*dev
, u16 rxq_index
,
3075 u32 flow_id
, u16 filter_id
)
3077 struct netdev_rx_queue
*rxqueue
= dev
->_rx
+ rxq_index
;
3078 struct rps_dev_flow_table
*flow_table
;
3079 struct rps_dev_flow
*rflow
;
3084 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3085 if (flow_table
&& flow_id
<= flow_table
->mask
) {
3086 rflow
= &flow_table
->flows
[flow_id
];
3087 cpu
= ACCESS_ONCE(rflow
->cpu
);
3088 if (rflow
->filter
== filter_id
&& cpu
!= RPS_NO_CPU
&&
3089 ((int)(per_cpu(softnet_data
, cpu
).input_queue_head
-
3090 rflow
->last_qtail
) <
3091 (int)(10 * flow_table
->mask
)))
3097 EXPORT_SYMBOL(rps_may_expire_flow
);
3099 #endif /* CONFIG_RFS_ACCEL */
3101 /* Called from hardirq (IPI) context */
3102 static void rps_trigger_softirq(void *data
)
3104 struct softnet_data
*sd
= data
;
3106 ____napi_schedule(sd
, &sd
->backlog
);
3110 #endif /* CONFIG_RPS */
3113 * Check if this softnet_data structure is another cpu one
3114 * If yes, queue it to our IPI list and return 1
3117 static int rps_ipi_queued(struct softnet_data
*sd
)
3120 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
3123 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
3124 mysd
->rps_ipi_list
= sd
;
3126 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3129 #endif /* CONFIG_RPS */
3133 #ifdef CONFIG_NET_FLOW_LIMIT
3134 int netdev_flow_limit_table_len __read_mostly
= (1 << 12);
3137 static bool skb_flow_limit(struct sk_buff
*skb
, unsigned int qlen
)
3139 #ifdef CONFIG_NET_FLOW_LIMIT
3140 struct sd_flow_limit
*fl
;
3141 struct softnet_data
*sd
;
3142 unsigned int old_flow
, new_flow
;
3144 if (qlen
< (netdev_max_backlog
>> 1))
3147 sd
= &__get_cpu_var(softnet_data
);
3150 fl
= rcu_dereference(sd
->flow_limit
);
3152 new_flow
= skb_get_rxhash(skb
) & (fl
->num_buckets
- 1);
3153 old_flow
= fl
->history
[fl
->history_head
];
3154 fl
->history
[fl
->history_head
] = new_flow
;
3157 fl
->history_head
&= FLOW_LIMIT_HISTORY
- 1;
3159 if (likely(fl
->buckets
[old_flow
]))
3160 fl
->buckets
[old_flow
]--;
3162 if (++fl
->buckets
[new_flow
] > (FLOW_LIMIT_HISTORY
>> 1)) {
3174 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3175 * queue (may be a remote CPU queue).
3177 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
3178 unsigned int *qtail
)
3180 struct softnet_data
*sd
;
3181 unsigned long flags
;
3184 sd
= &per_cpu(softnet_data
, cpu
);
3186 local_irq_save(flags
);
3189 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3190 if (qlen
<= netdev_max_backlog
&& !skb_flow_limit(skb
, qlen
)) {
3191 if (skb_queue_len(&sd
->input_pkt_queue
)) {
3193 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
3194 input_queue_tail_incr_save(sd
, qtail
);
3196 local_irq_restore(flags
);
3197 return NET_RX_SUCCESS
;
3200 /* Schedule NAPI for backlog device
3201 * We can use non atomic operation since we own the queue lock
3203 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
3204 if (!rps_ipi_queued(sd
))
3205 ____napi_schedule(sd
, &sd
->backlog
);
3213 local_irq_restore(flags
);
3215 atomic_long_inc(&skb
->dev
->rx_dropped
);
3221 * netif_rx - post buffer to the network code
3222 * @skb: buffer to post
3224 * This function receives a packet from a device driver and queues it for
3225 * the upper (protocol) levels to process. It always succeeds. The buffer
3226 * may be dropped during processing for congestion control or by the
3230 * NET_RX_SUCCESS (no congestion)
3231 * NET_RX_DROP (packet was dropped)
3235 int netif_rx(struct sk_buff
*skb
)
3239 /* if netpoll wants it, pretend we never saw it */
3240 if (netpoll_rx(skb
))
3243 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3245 trace_netif_rx(skb
);
3247 if (static_key_false(&rps_needed
)) {
3248 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3254 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3256 cpu
= smp_processor_id();
3258 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3266 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
3271 EXPORT_SYMBOL(netif_rx
);
3273 int netif_rx_ni(struct sk_buff
*skb
)
3278 err
= netif_rx(skb
);
3279 if (local_softirq_pending())
3285 EXPORT_SYMBOL(netif_rx_ni
);
3287 static void net_tx_action(struct softirq_action
*h
)
3289 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3291 if (sd
->completion_queue
) {
3292 struct sk_buff
*clist
;
3294 local_irq_disable();
3295 clist
= sd
->completion_queue
;
3296 sd
->completion_queue
= NULL
;
3300 struct sk_buff
*skb
= clist
;
3301 clist
= clist
->next
;
3303 WARN_ON(atomic_read(&skb
->users
));
3304 trace_kfree_skb(skb
, net_tx_action
);
3309 if (sd
->output_queue
) {
3312 local_irq_disable();
3313 head
= sd
->output_queue
;
3314 sd
->output_queue
= NULL
;
3315 sd
->output_queue_tailp
= &sd
->output_queue
;
3319 struct Qdisc
*q
= head
;
3320 spinlock_t
*root_lock
;
3322 head
= head
->next_sched
;
3324 root_lock
= qdisc_lock(q
);
3325 if (spin_trylock(root_lock
)) {
3326 smp_mb__before_clear_bit();
3327 clear_bit(__QDISC_STATE_SCHED
,
3330 spin_unlock(root_lock
);
3332 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
3334 __netif_reschedule(q
);
3336 smp_mb__before_clear_bit();
3337 clear_bit(__QDISC_STATE_SCHED
,
3345 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3346 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3347 /* This hook is defined here for ATM LANE */
3348 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
3349 unsigned char *addr
) __read_mostly
;
3350 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
3353 #ifdef CONFIG_NET_CLS_ACT
3354 /* TODO: Maybe we should just force sch_ingress to be compiled in
3355 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3356 * a compare and 2 stores extra right now if we dont have it on
3357 * but have CONFIG_NET_CLS_ACT
3358 * NOTE: This doesn't stop any functionality; if you dont have
3359 * the ingress scheduler, you just can't add policies on ingress.
3362 static int ing_filter(struct sk_buff
*skb
, struct netdev_queue
*rxq
)
3364 struct net_device
*dev
= skb
->dev
;
3365 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
3366 int result
= TC_ACT_OK
;
3369 if (unlikely(MAX_RED_LOOP
< ttl
++)) {
3370 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3371 skb
->skb_iif
, dev
->ifindex
);
3375 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
3376 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
3379 if (q
!= &noop_qdisc
) {
3380 spin_lock(qdisc_lock(q
));
3381 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
3382 result
= qdisc_enqueue_root(skb
, q
);
3383 spin_unlock(qdisc_lock(q
));
3389 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
3390 struct packet_type
**pt_prev
,
3391 int *ret
, struct net_device
*orig_dev
)
3393 struct netdev_queue
*rxq
= rcu_dereference(skb
->dev
->ingress_queue
);
3395 if (!rxq
|| rxq
->qdisc
== &noop_qdisc
)
3399 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
3403 switch (ing_filter(skb
, rxq
)) {
3417 * netdev_rx_handler_register - register receive handler
3418 * @dev: device to register a handler for
3419 * @rx_handler: receive handler to register
3420 * @rx_handler_data: data pointer that is used by rx handler
3422 * Register a receive hander for a device. This handler will then be
3423 * called from __netif_receive_skb. A negative errno code is returned
3426 * The caller must hold the rtnl_mutex.
3428 * For a general description of rx_handler, see enum rx_handler_result.
3430 int netdev_rx_handler_register(struct net_device
*dev
,
3431 rx_handler_func_t
*rx_handler
,
3432 void *rx_handler_data
)
3436 if (dev
->rx_handler
)
3439 /* Note: rx_handler_data must be set before rx_handler */
3440 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
3441 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
3445 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
3448 * netdev_rx_handler_unregister - unregister receive handler
3449 * @dev: device to unregister a handler from
3451 * Unregister a receive handler from a device.
3453 * The caller must hold the rtnl_mutex.
3455 void netdev_rx_handler_unregister(struct net_device
*dev
)
3459 RCU_INIT_POINTER(dev
->rx_handler
, NULL
);
3460 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3461 * section has a guarantee to see a non NULL rx_handler_data
3465 RCU_INIT_POINTER(dev
->rx_handler_data
, NULL
);
3467 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
3470 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3471 * the special handling of PFMEMALLOC skbs.
3473 static bool skb_pfmemalloc_protocol(struct sk_buff
*skb
)
3475 switch (skb
->protocol
) {
3476 case __constant_htons(ETH_P_ARP
):
3477 case __constant_htons(ETH_P_IP
):
3478 case __constant_htons(ETH_P_IPV6
):
3479 case __constant_htons(ETH_P_8021Q
):
3480 case __constant_htons(ETH_P_8021AD
):
3487 static int __netif_receive_skb_core(struct sk_buff
*skb
, bool pfmemalloc
)
3489 struct packet_type
*ptype
, *pt_prev
;
3490 rx_handler_func_t
*rx_handler
;
3491 struct net_device
*orig_dev
;
3492 struct net_device
*null_or_dev
;
3493 bool deliver_exact
= false;
3494 int ret
= NET_RX_DROP
;
3497 net_timestamp_check(!netdev_tstamp_prequeue
, skb
);
3499 trace_netif_receive_skb(skb
);
3501 /* if we've gotten here through NAPI, check netpoll */
3502 if (netpoll_receive_skb(skb
))
3505 orig_dev
= skb
->dev
;
3507 skb_reset_network_header(skb
);
3508 if (!skb_transport_header_was_set(skb
))
3509 skb_reset_transport_header(skb
);
3510 skb_reset_mac_len(skb
);
3517 skb
->skb_iif
= skb
->dev
->ifindex
;
3519 __this_cpu_inc(softnet_data
.processed
);
3521 if (skb
->protocol
== cpu_to_be16(ETH_P_8021Q
) ||
3522 skb
->protocol
== cpu_to_be16(ETH_P_8021AD
)) {
3523 skb
= vlan_untag(skb
);
3528 #ifdef CONFIG_NET_CLS_ACT
3529 if (skb
->tc_verd
& TC_NCLS
) {
3530 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
3538 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
3539 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
3541 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3547 #ifdef CONFIG_NET_CLS_ACT
3548 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
3554 if (pfmemalloc
&& !skb_pfmemalloc_protocol(skb
))
3557 if (vlan_tx_tag_present(skb
)) {
3559 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3562 if (vlan_do_receive(&skb
))
3564 else if (unlikely(!skb
))
3568 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
3571 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3574 switch (rx_handler(&skb
)) {
3575 case RX_HANDLER_CONSUMED
:
3576 ret
= NET_RX_SUCCESS
;
3578 case RX_HANDLER_ANOTHER
:
3580 case RX_HANDLER_EXACT
:
3581 deliver_exact
= true;
3582 case RX_HANDLER_PASS
:
3589 if (unlikely(vlan_tx_tag_present(skb
))) {
3590 if (vlan_tx_tag_get_id(skb
))
3591 skb
->pkt_type
= PACKET_OTHERHOST
;
3592 /* Note: we might in the future use prio bits
3593 * and set skb->priority like in vlan_do_receive()
3594 * For the time being, just ignore Priority Code Point
3599 /* deliver only exact match when indicated */
3600 null_or_dev
= deliver_exact
? skb
->dev
: NULL
;
3602 type
= skb
->protocol
;
3603 list_for_each_entry_rcu(ptype
,
3604 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
3605 if (ptype
->type
== type
&&
3606 (ptype
->dev
== null_or_dev
|| ptype
->dev
== skb
->dev
||
3607 ptype
->dev
== orig_dev
)) {
3609 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3615 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
3618 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
3621 atomic_long_inc(&skb
->dev
->rx_dropped
);
3623 /* Jamal, now you will not able to escape explaining
3624 * me how you were going to use this. :-)
3635 static int __netif_receive_skb(struct sk_buff
*skb
)
3639 if (sk_memalloc_socks() && skb_pfmemalloc(skb
)) {
3640 unsigned long pflags
= current
->flags
;
3643 * PFMEMALLOC skbs are special, they should
3644 * - be delivered to SOCK_MEMALLOC sockets only
3645 * - stay away from userspace
3646 * - have bounded memory usage
3648 * Use PF_MEMALLOC as this saves us from propagating the allocation
3649 * context down to all allocation sites.
3651 current
->flags
|= PF_MEMALLOC
;
3652 ret
= __netif_receive_skb_core(skb
, true);
3653 tsk_restore_flags(current
, pflags
, PF_MEMALLOC
);
3655 ret
= __netif_receive_skb_core(skb
, false);
3661 * netif_receive_skb - process receive buffer from network
3662 * @skb: buffer to process
3664 * netif_receive_skb() is the main receive data processing function.
3665 * It always succeeds. The buffer may be dropped during processing
3666 * for congestion control or by the protocol layers.
3668 * This function may only be called from softirq context and interrupts
3669 * should be enabled.
3671 * Return values (usually ignored):
3672 * NET_RX_SUCCESS: no congestion
3673 * NET_RX_DROP: packet was dropped
3675 int netif_receive_skb(struct sk_buff
*skb
)
3677 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3679 if (skb_defer_rx_timestamp(skb
))
3680 return NET_RX_SUCCESS
;
3683 if (static_key_false(&rps_needed
)) {
3684 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3689 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3692 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3699 return __netif_receive_skb(skb
);
3701 EXPORT_SYMBOL(netif_receive_skb
);
3703 /* Network device is going away, flush any packets still pending
3704 * Called with irqs disabled.
3706 static void flush_backlog(void *arg
)
3708 struct net_device
*dev
= arg
;
3709 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3710 struct sk_buff
*skb
, *tmp
;
3713 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3714 if (skb
->dev
== dev
) {
3715 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3717 input_queue_head_incr(sd
);
3722 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3723 if (skb
->dev
== dev
) {
3724 __skb_unlink(skb
, &sd
->process_queue
);
3726 input_queue_head_incr(sd
);
3731 static int napi_gro_complete(struct sk_buff
*skb
)
3733 struct packet_offload
*ptype
;
3734 __be16 type
= skb
->protocol
;
3735 struct list_head
*head
= &offload_base
;
3738 BUILD_BUG_ON(sizeof(struct napi_gro_cb
) > sizeof(skb
->cb
));
3740 if (NAPI_GRO_CB(skb
)->count
== 1) {
3741 skb_shinfo(skb
)->gso_size
= 0;
3746 list_for_each_entry_rcu(ptype
, head
, list
) {
3747 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_complete
)
3750 err
= ptype
->callbacks
.gro_complete(skb
);
3756 WARN_ON(&ptype
->list
== head
);
3758 return NET_RX_SUCCESS
;
3762 return netif_receive_skb(skb
);
3765 /* napi->gro_list contains packets ordered by age.
3766 * youngest packets at the head of it.
3767 * Complete skbs in reverse order to reduce latencies.
3769 void napi_gro_flush(struct napi_struct
*napi
, bool flush_old
)
3771 struct sk_buff
*skb
, *prev
= NULL
;
3773 /* scan list and build reverse chain */
3774 for (skb
= napi
->gro_list
; skb
!= NULL
; skb
= skb
->next
) {
3779 for (skb
= prev
; skb
; skb
= prev
) {
3782 if (flush_old
&& NAPI_GRO_CB(skb
)->age
== jiffies
)
3786 napi_gro_complete(skb
);
3790 napi
->gro_list
= NULL
;
3792 EXPORT_SYMBOL(napi_gro_flush
);
3794 static void gro_list_prepare(struct napi_struct
*napi
, struct sk_buff
*skb
)
3797 unsigned int maclen
= skb
->dev
->hard_header_len
;
3799 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3800 unsigned long diffs
;
3802 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
3803 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
3804 if (maclen
== ETH_HLEN
)
3805 diffs
|= compare_ether_header(skb_mac_header(p
),
3806 skb_gro_mac_header(skb
));
3808 diffs
= memcmp(skb_mac_header(p
),
3809 skb_gro_mac_header(skb
),
3811 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
3812 NAPI_GRO_CB(p
)->flush
= 0;
3816 static enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3818 struct sk_buff
**pp
= NULL
;
3819 struct packet_offload
*ptype
;
3820 __be16 type
= skb
->protocol
;
3821 struct list_head
*head
= &offload_base
;
3823 enum gro_result ret
;
3825 if (!(skb
->dev
->features
& NETIF_F_GRO
) || netpoll_rx_on(skb
))
3828 if (skb_is_gso(skb
) || skb_has_frag_list(skb
))
3831 gro_list_prepare(napi
, skb
);
3834 list_for_each_entry_rcu(ptype
, head
, list
) {
3835 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_receive
)
3838 skb_set_network_header(skb
, skb_gro_offset(skb
));
3839 skb_reset_mac_len(skb
);
3840 NAPI_GRO_CB(skb
)->same_flow
= 0;
3841 NAPI_GRO_CB(skb
)->flush
= 0;
3842 NAPI_GRO_CB(skb
)->free
= 0;
3844 pp
= ptype
->callbacks
.gro_receive(&napi
->gro_list
, skb
);
3849 if (&ptype
->list
== head
)
3852 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3853 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3856 struct sk_buff
*nskb
= *pp
;
3860 napi_gro_complete(nskb
);
3867 if (NAPI_GRO_CB(skb
)->flush
|| napi
->gro_count
>= MAX_GRO_SKBS
)
3871 NAPI_GRO_CB(skb
)->count
= 1;
3872 NAPI_GRO_CB(skb
)->age
= jiffies
;
3873 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
3874 skb
->next
= napi
->gro_list
;
3875 napi
->gro_list
= skb
;
3879 if (skb_headlen(skb
) < skb_gro_offset(skb
)) {
3880 int grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
3882 BUG_ON(skb
->end
- skb
->tail
< grow
);
3884 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3887 skb
->data_len
-= grow
;
3889 skb_shinfo(skb
)->frags
[0].page_offset
+= grow
;
3890 skb_frag_size_sub(&skb_shinfo(skb
)->frags
[0], grow
);
3892 if (unlikely(!skb_frag_size(&skb_shinfo(skb
)->frags
[0]))) {
3893 skb_frag_unref(skb
, 0);
3894 memmove(skb_shinfo(skb
)->frags
,
3895 skb_shinfo(skb
)->frags
+ 1,
3896 --skb_shinfo(skb
)->nr_frags
* sizeof(skb_frag_t
));
3909 static gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
3913 if (netif_receive_skb(skb
))
3921 case GRO_MERGED_FREE
:
3922 if (NAPI_GRO_CB(skb
)->free
== NAPI_GRO_FREE_STOLEN_HEAD
)
3923 kmem_cache_free(skbuff_head_cache
, skb
);
3936 static void skb_gro_reset_offset(struct sk_buff
*skb
)
3938 const struct skb_shared_info
*pinfo
= skb_shinfo(skb
);
3939 const skb_frag_t
*frag0
= &pinfo
->frags
[0];
3941 NAPI_GRO_CB(skb
)->data_offset
= 0;
3942 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3943 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3945 if (skb_mac_header(skb
) == skb_tail_pointer(skb
) &&
3947 !PageHighMem(skb_frag_page(frag0
))) {
3948 NAPI_GRO_CB(skb
)->frag0
= skb_frag_address(frag0
);
3949 NAPI_GRO_CB(skb
)->frag0_len
= skb_frag_size(frag0
);
3953 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3955 skb_gro_reset_offset(skb
);
3957 return napi_skb_finish(dev_gro_receive(napi
, skb
), skb
);
3959 EXPORT_SYMBOL(napi_gro_receive
);
3961 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
3963 __skb_pull(skb
, skb_headlen(skb
));
3964 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3965 skb_reserve(skb
, NET_SKB_PAD
+ NET_IP_ALIGN
- skb_headroom(skb
));
3967 skb
->dev
= napi
->dev
;
3973 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
3975 struct sk_buff
*skb
= napi
->skb
;
3978 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
3984 EXPORT_SYMBOL(napi_get_frags
);
3986 static gro_result_t
napi_frags_finish(struct napi_struct
*napi
, struct sk_buff
*skb
,
3992 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
3994 if (ret
== GRO_HELD
)
3995 skb_gro_pull(skb
, -ETH_HLEN
);
3996 else if (netif_receive_skb(skb
))
4001 case GRO_MERGED_FREE
:
4002 napi_reuse_skb(napi
, skb
);
4012 static struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
4014 struct sk_buff
*skb
= napi
->skb
;
4021 skb_reset_mac_header(skb
);
4022 skb_gro_reset_offset(skb
);
4024 off
= skb_gro_offset(skb
);
4025 hlen
= off
+ sizeof(*eth
);
4026 eth
= skb_gro_header_fast(skb
, off
);
4027 if (skb_gro_header_hard(skb
, hlen
)) {
4028 eth
= skb_gro_header_slow(skb
, hlen
, off
);
4029 if (unlikely(!eth
)) {
4030 napi_reuse_skb(napi
, skb
);
4036 skb_gro_pull(skb
, sizeof(*eth
));
4039 * This works because the only protocols we care about don't require
4040 * special handling. We'll fix it up properly at the end.
4042 skb
->protocol
= eth
->h_proto
;
4048 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
4050 struct sk_buff
*skb
= napi_frags_skb(napi
);
4055 return napi_frags_finish(napi
, skb
, dev_gro_receive(napi
, skb
));
4057 EXPORT_SYMBOL(napi_gro_frags
);
4060 * net_rps_action sends any pending IPI's for rps.
4061 * Note: called with local irq disabled, but exits with local irq enabled.
4063 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
4066 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
4069 sd
->rps_ipi_list
= NULL
;
4073 /* Send pending IPI's to kick RPS processing on remote cpus. */
4075 struct softnet_data
*next
= remsd
->rps_ipi_next
;
4077 if (cpu_online(remsd
->cpu
))
4078 __smp_call_function_single(remsd
->cpu
,
4087 static int process_backlog(struct napi_struct
*napi
, int quota
)
4090 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
4093 /* Check if we have pending ipi, its better to send them now,
4094 * not waiting net_rx_action() end.
4096 if (sd
->rps_ipi_list
) {
4097 local_irq_disable();
4098 net_rps_action_and_irq_enable(sd
);
4101 napi
->weight
= weight_p
;
4102 local_irq_disable();
4103 while (work
< quota
) {
4104 struct sk_buff
*skb
;
4107 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
4109 __netif_receive_skb(skb
);
4110 local_irq_disable();
4111 input_queue_head_incr(sd
);
4112 if (++work
>= quota
) {
4119 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
4121 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
4122 &sd
->process_queue
);
4124 if (qlen
< quota
- work
) {
4126 * Inline a custom version of __napi_complete().
4127 * only current cpu owns and manipulates this napi,
4128 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
4129 * we can use a plain write instead of clear_bit(),
4130 * and we dont need an smp_mb() memory barrier.
4132 list_del(&napi
->poll_list
);
4135 quota
= work
+ qlen
;
4145 * __napi_schedule - schedule for receive
4146 * @n: entry to schedule
4148 * The entry's receive function will be scheduled to run
4150 void __napi_schedule(struct napi_struct
*n
)
4152 unsigned long flags
;
4154 local_irq_save(flags
);
4155 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
4156 local_irq_restore(flags
);
4158 EXPORT_SYMBOL(__napi_schedule
);
4160 void __napi_complete(struct napi_struct
*n
)
4162 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
4163 BUG_ON(n
->gro_list
);
4165 list_del(&n
->poll_list
);
4166 smp_mb__before_clear_bit();
4167 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
4169 EXPORT_SYMBOL(__napi_complete
);
4171 void napi_complete(struct napi_struct
*n
)
4173 unsigned long flags
;
4176 * don't let napi dequeue from the cpu poll list
4177 * just in case its running on a different cpu
4179 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
4182 napi_gro_flush(n
, false);
4183 local_irq_save(flags
);
4185 local_irq_restore(flags
);
4187 EXPORT_SYMBOL(napi_complete
);
4189 /* must be called under rcu_read_lock(), as we dont take a reference */
4190 struct napi_struct
*napi_by_id(unsigned int napi_id
)
4192 unsigned int hash
= napi_id
% HASH_SIZE(napi_hash
);
4193 struct napi_struct
*napi
;
4195 hlist_for_each_entry_rcu(napi
, &napi_hash
[hash
], napi_hash_node
)
4196 if (napi
->napi_id
== napi_id
)
4201 EXPORT_SYMBOL_GPL(napi_by_id
);
4203 void napi_hash_add(struct napi_struct
*napi
)
4205 if (!test_and_set_bit(NAPI_STATE_HASHED
, &napi
->state
)) {
4207 spin_lock(&napi_hash_lock
);
4209 /* 0 is not a valid id, we also skip an id that is taken
4210 * we expect both events to be extremely rare
4213 while (!napi
->napi_id
) {
4214 napi
->napi_id
= ++napi_gen_id
;
4215 if (napi_by_id(napi
->napi_id
))
4219 hlist_add_head_rcu(&napi
->napi_hash_node
,
4220 &napi_hash
[napi
->napi_id
% HASH_SIZE(napi_hash
)]);
4222 spin_unlock(&napi_hash_lock
);
4225 EXPORT_SYMBOL_GPL(napi_hash_add
);
4227 /* Warning : caller is responsible to make sure rcu grace period
4228 * is respected before freeing memory containing @napi
4230 void napi_hash_del(struct napi_struct
*napi
)
4232 spin_lock(&napi_hash_lock
);
4234 if (test_and_clear_bit(NAPI_STATE_HASHED
, &napi
->state
))
4235 hlist_del_rcu(&napi
->napi_hash_node
);
4237 spin_unlock(&napi_hash_lock
);
4239 EXPORT_SYMBOL_GPL(napi_hash_del
);
4241 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
4242 int (*poll
)(struct napi_struct
*, int), int weight
)
4244 INIT_LIST_HEAD(&napi
->poll_list
);
4245 napi
->gro_count
= 0;
4246 napi
->gro_list
= NULL
;
4249 if (weight
> NAPI_POLL_WEIGHT
)
4250 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4252 napi
->weight
= weight
;
4253 list_add(&napi
->dev_list
, &dev
->napi_list
);
4255 #ifdef CONFIG_NETPOLL
4256 spin_lock_init(&napi
->poll_lock
);
4257 napi
->poll_owner
= -1;
4259 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
4261 EXPORT_SYMBOL(netif_napi_add
);
4263 void netif_napi_del(struct napi_struct
*napi
)
4265 struct sk_buff
*skb
, *next
;
4267 list_del_init(&napi
->dev_list
);
4268 napi_free_frags(napi
);
4270 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
4276 napi
->gro_list
= NULL
;
4277 napi
->gro_count
= 0;
4279 EXPORT_SYMBOL(netif_napi_del
);
4281 static void net_rx_action(struct softirq_action
*h
)
4283 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
4284 unsigned long time_limit
= jiffies
+ 2;
4285 int budget
= netdev_budget
;
4288 local_irq_disable();
4290 while (!list_empty(&sd
->poll_list
)) {
4291 struct napi_struct
*n
;
4294 /* If softirq window is exhuasted then punt.
4295 * Allow this to run for 2 jiffies since which will allow
4296 * an average latency of 1.5/HZ.
4298 if (unlikely(budget
<= 0 || time_after_eq(jiffies
, time_limit
)))
4303 /* Even though interrupts have been re-enabled, this
4304 * access is safe because interrupts can only add new
4305 * entries to the tail of this list, and only ->poll()
4306 * calls can remove this head entry from the list.
4308 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
4310 have
= netpoll_poll_lock(n
);
4314 /* This NAPI_STATE_SCHED test is for avoiding a race
4315 * with netpoll's poll_napi(). Only the entity which
4316 * obtains the lock and sees NAPI_STATE_SCHED set will
4317 * actually make the ->poll() call. Therefore we avoid
4318 * accidentally calling ->poll() when NAPI is not scheduled.
4321 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
4322 work
= n
->poll(n
, weight
);
4326 WARN_ON_ONCE(work
> weight
);
4330 local_irq_disable();
4332 /* Drivers must not modify the NAPI state if they
4333 * consume the entire weight. In such cases this code
4334 * still "owns" the NAPI instance and therefore can
4335 * move the instance around on the list at-will.
4337 if (unlikely(work
== weight
)) {
4338 if (unlikely(napi_disable_pending(n
))) {
4341 local_irq_disable();
4344 /* flush too old packets
4345 * If HZ < 1000, flush all packets.
4348 napi_gro_flush(n
, HZ
>= 1000);
4349 local_irq_disable();
4351 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
4355 netpoll_poll_unlock(have
);
4358 net_rps_action_and_irq_enable(sd
);
4360 #ifdef CONFIG_NET_DMA
4362 * There may not be any more sk_buffs coming right now, so push
4363 * any pending DMA copies to hardware
4365 dma_issue_pending_all();
4372 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
4376 struct netdev_adjacent
{
4377 struct net_device
*dev
;
4379 /* upper master flag, there can only be one master device per list */
4382 /* counter for the number of times this device was added to us */
4385 /* private field for the users */
4388 struct list_head list
;
4389 struct rcu_head rcu
;
4392 static struct netdev_adjacent
*__netdev_find_adj_rcu(struct net_device
*dev
,
4393 struct net_device
*adj_dev
,
4394 struct list_head
*adj_list
)
4396 struct netdev_adjacent
*adj
;
4398 list_for_each_entry_rcu(adj
, adj_list
, list
) {
4399 if (adj
->dev
== adj_dev
)
4405 static struct netdev_adjacent
*__netdev_find_adj(struct net_device
*dev
,
4406 struct net_device
*adj_dev
,
4407 struct list_head
*adj_list
)
4409 struct netdev_adjacent
*adj
;
4411 list_for_each_entry(adj
, adj_list
, list
) {
4412 if (adj
->dev
== adj_dev
)
4419 * netdev_has_upper_dev - Check if device is linked to an upper device
4421 * @upper_dev: upper device to check
4423 * Find out if a device is linked to specified upper device and return true
4424 * in case it is. Note that this checks only immediate upper device,
4425 * not through a complete stack of devices. The caller must hold the RTNL lock.
4427 bool netdev_has_upper_dev(struct net_device
*dev
,
4428 struct net_device
*upper_dev
)
4432 return __netdev_find_adj(dev
, upper_dev
, &dev
->all_adj_list
.upper
);
4434 EXPORT_SYMBOL(netdev_has_upper_dev
);
4437 * netdev_has_any_upper_dev - Check if device is linked to some device
4440 * Find out if a device is linked to an upper device and return true in case
4441 * it is. The caller must hold the RTNL lock.
4443 bool netdev_has_any_upper_dev(struct net_device
*dev
)
4447 return !list_empty(&dev
->all_adj_list
.upper
);
4449 EXPORT_SYMBOL(netdev_has_any_upper_dev
);
4452 * netdev_master_upper_dev_get - Get master upper device
4455 * Find a master upper device and return pointer to it or NULL in case
4456 * it's not there. The caller must hold the RTNL lock.
4458 struct net_device
*netdev_master_upper_dev_get(struct net_device
*dev
)
4460 struct netdev_adjacent
*upper
;
4464 if (list_empty(&dev
->adj_list
.upper
))
4467 upper
= list_first_entry(&dev
->adj_list
.upper
,
4468 struct netdev_adjacent
, list
);
4469 if (likely(upper
->master
))
4473 EXPORT_SYMBOL(netdev_master_upper_dev_get
);
4475 void *netdev_adjacent_get_private(struct list_head
*adj_list
)
4477 struct netdev_adjacent
*adj
;
4479 adj
= list_entry(adj_list
, struct netdev_adjacent
, list
);
4481 return adj
->private;
4483 EXPORT_SYMBOL(netdev_adjacent_get_private
);
4486 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4488 * @iter: list_head ** of the current position
4490 * Gets the next device from the dev's upper list, starting from iter
4491 * position. The caller must hold RCU read lock.
4493 struct net_device
*netdev_all_upper_get_next_dev_rcu(struct net_device
*dev
,
4494 struct list_head
**iter
)
4496 struct netdev_adjacent
*upper
;
4498 WARN_ON_ONCE(!rcu_read_lock_held());
4500 upper
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
4502 if (&upper
->list
== &dev
->all_adj_list
.upper
)
4505 *iter
= &upper
->list
;
4509 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu
);
4512 * netdev_lower_get_next_private - Get the next ->private from the
4513 * lower neighbour list
4515 * @iter: list_head ** of the current position
4517 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4518 * list, starting from iter position. The caller must hold either hold the
4519 * RTNL lock or its own locking that guarantees that the neighbour lower
4520 * list will remain unchainged.
4522 void *netdev_lower_get_next_private(struct net_device
*dev
,
4523 struct list_head
**iter
)
4525 struct netdev_adjacent
*lower
;
4527 lower
= list_entry(*iter
, struct netdev_adjacent
, list
);
4529 if (&lower
->list
== &dev
->adj_list
.lower
)
4533 *iter
= lower
->list
.next
;
4535 return lower
->private;
4537 EXPORT_SYMBOL(netdev_lower_get_next_private
);
4540 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4541 * lower neighbour list, RCU
4544 * @iter: list_head ** of the current position
4546 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4547 * list, starting from iter position. The caller must hold RCU read lock.
4549 void *netdev_lower_get_next_private_rcu(struct net_device
*dev
,
4550 struct list_head
**iter
)
4552 struct netdev_adjacent
*lower
;
4554 WARN_ON_ONCE(!rcu_read_lock_held());
4556 lower
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
4558 if (&lower
->list
== &dev
->adj_list
.lower
)
4562 *iter
= &lower
->list
;
4564 return lower
->private;
4566 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu
);
4569 * netdev_master_upper_dev_get_rcu - Get master upper device
4572 * Find a master upper device and return pointer to it or NULL in case
4573 * it's not there. The caller must hold the RCU read lock.
4575 struct net_device
*netdev_master_upper_dev_get_rcu(struct net_device
*dev
)
4577 struct netdev_adjacent
*upper
;
4579 upper
= list_first_or_null_rcu(&dev
->adj_list
.upper
,
4580 struct netdev_adjacent
, list
);
4581 if (upper
&& likely(upper
->master
))
4585 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu
);
4587 static int __netdev_adjacent_dev_insert(struct net_device
*dev
,
4588 struct net_device
*adj_dev
,
4589 struct list_head
*dev_list
,
4590 void *private, bool master
)
4592 struct netdev_adjacent
*adj
;
4593 char linkname
[IFNAMSIZ
+7];
4596 adj
= __netdev_find_adj(dev
, adj_dev
, dev_list
);
4603 adj
= kmalloc(sizeof(*adj
), GFP_KERNEL
);
4608 adj
->master
= master
;
4610 adj
->private = private;
4613 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
4614 adj_dev
->name
, dev
->name
, adj_dev
->name
);
4616 if (dev_list
== &dev
->adj_list
.lower
) {
4617 sprintf(linkname
, "lower_%s", adj_dev
->name
);
4618 ret
= sysfs_create_link(&(dev
->dev
.kobj
),
4619 &(adj_dev
->dev
.kobj
), linkname
);
4622 } else if (dev_list
== &dev
->adj_list
.upper
) {
4623 sprintf(linkname
, "upper_%s", adj_dev
->name
);
4624 ret
= sysfs_create_link(&(dev
->dev
.kobj
),
4625 &(adj_dev
->dev
.kobj
), linkname
);
4630 /* Ensure that master link is always the first item in list. */
4632 ret
= sysfs_create_link(&(dev
->dev
.kobj
),
4633 &(adj_dev
->dev
.kobj
), "master");
4635 goto remove_symlinks
;
4637 list_add_rcu(&adj
->list
, dev_list
);
4639 list_add_tail_rcu(&adj
->list
, dev_list
);
4645 if (dev_list
== &dev
->adj_list
.lower
) {
4646 sprintf(linkname
, "lower_%s", adj_dev
->name
);
4647 sysfs_remove_link(&(dev
->dev
.kobj
), linkname
);
4648 } else if (dev_list
== &dev
->adj_list
.upper
) {
4649 sprintf(linkname
, "upper_%s", adj_dev
->name
);
4650 sysfs_remove_link(&(dev
->dev
.kobj
), linkname
);
4660 void __netdev_adjacent_dev_remove(struct net_device
*dev
,
4661 struct net_device
*adj_dev
,
4662 struct list_head
*dev_list
)
4664 struct netdev_adjacent
*adj
;
4665 char linkname
[IFNAMSIZ
+7];
4667 adj
= __netdev_find_adj(dev
, adj_dev
, dev_list
);
4670 pr_err("tried to remove device %s from %s\n",
4671 dev
->name
, adj_dev
->name
);
4675 if (adj
->ref_nr
> 1) {
4676 pr_debug("%s to %s ref_nr-- = %d\n", dev
->name
, adj_dev
->name
,
4683 sysfs_remove_link(&(dev
->dev
.kobj
), "master");
4685 if (dev_list
== &dev
->adj_list
.lower
) {
4686 sprintf(linkname
, "lower_%s", adj_dev
->name
);
4687 sysfs_remove_link(&(dev
->dev
.kobj
), linkname
);
4688 } else if (dev_list
== &dev
->adj_list
.upper
) {
4689 sprintf(linkname
, "upper_%s", adj_dev
->name
);
4690 sysfs_remove_link(&(dev
->dev
.kobj
), linkname
);
4693 list_del_rcu(&adj
->list
);
4694 pr_debug("dev_put for %s, because link removed from %s to %s\n",
4695 adj_dev
->name
, dev
->name
, adj_dev
->name
);
4697 kfree_rcu(adj
, rcu
);
4700 int __netdev_adjacent_dev_link_lists(struct net_device
*dev
,
4701 struct net_device
*upper_dev
,
4702 struct list_head
*up_list
,
4703 struct list_head
*down_list
,
4704 void *private, bool master
)
4708 ret
= __netdev_adjacent_dev_insert(dev
, upper_dev
, up_list
, private,
4713 ret
= __netdev_adjacent_dev_insert(upper_dev
, dev
, down_list
, private,
4716 __netdev_adjacent_dev_remove(dev
, upper_dev
, up_list
);
4723 int __netdev_adjacent_dev_link(struct net_device
*dev
,
4724 struct net_device
*upper_dev
)
4726 return __netdev_adjacent_dev_link_lists(dev
, upper_dev
,
4727 &dev
->all_adj_list
.upper
,
4728 &upper_dev
->all_adj_list
.lower
,
4732 void __netdev_adjacent_dev_unlink_lists(struct net_device
*dev
,
4733 struct net_device
*upper_dev
,
4734 struct list_head
*up_list
,
4735 struct list_head
*down_list
)
4737 __netdev_adjacent_dev_remove(dev
, upper_dev
, up_list
);
4738 __netdev_adjacent_dev_remove(upper_dev
, dev
, down_list
);
4741 void __netdev_adjacent_dev_unlink(struct net_device
*dev
,
4742 struct net_device
*upper_dev
)
4744 __netdev_adjacent_dev_unlink_lists(dev
, upper_dev
,
4745 &dev
->all_adj_list
.upper
,
4746 &upper_dev
->all_adj_list
.lower
);
4749 int __netdev_adjacent_dev_link_neighbour(struct net_device
*dev
,
4750 struct net_device
*upper_dev
,
4751 void *private, bool master
)
4753 int ret
= __netdev_adjacent_dev_link(dev
, upper_dev
);
4758 ret
= __netdev_adjacent_dev_link_lists(dev
, upper_dev
,
4759 &dev
->adj_list
.upper
,
4760 &upper_dev
->adj_list
.lower
,
4763 __netdev_adjacent_dev_unlink(dev
, upper_dev
);
4770 void __netdev_adjacent_dev_unlink_neighbour(struct net_device
*dev
,
4771 struct net_device
*upper_dev
)
4773 __netdev_adjacent_dev_unlink(dev
, upper_dev
);
4774 __netdev_adjacent_dev_unlink_lists(dev
, upper_dev
,
4775 &dev
->adj_list
.upper
,
4776 &upper_dev
->adj_list
.lower
);
4779 static int __netdev_upper_dev_link(struct net_device
*dev
,
4780 struct net_device
*upper_dev
, bool master
,
4783 struct netdev_adjacent
*i
, *j
, *to_i
, *to_j
;
4788 if (dev
== upper_dev
)
4791 /* To prevent loops, check if dev is not upper device to upper_dev. */
4792 if (__netdev_find_adj(upper_dev
, dev
, &upper_dev
->all_adj_list
.upper
))
4795 if (__netdev_find_adj(dev
, upper_dev
, &dev
->all_adj_list
.upper
))
4798 if (master
&& netdev_master_upper_dev_get(dev
))
4801 ret
= __netdev_adjacent_dev_link_neighbour(dev
, upper_dev
, private,
4806 /* Now that we linked these devs, make all the upper_dev's
4807 * all_adj_list.upper visible to every dev's all_adj_list.lower an
4808 * versa, and don't forget the devices itself. All of these
4809 * links are non-neighbours.
4811 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
4812 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
) {
4813 pr_debug("Interlinking %s with %s, non-neighbour\n",
4814 i
->dev
->name
, j
->dev
->name
);
4815 ret
= __netdev_adjacent_dev_link(i
->dev
, j
->dev
);
4821 /* add dev to every upper_dev's upper device */
4822 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
) {
4823 pr_debug("linking %s's upper device %s with %s\n",
4824 upper_dev
->name
, i
->dev
->name
, dev
->name
);
4825 ret
= __netdev_adjacent_dev_link(dev
, i
->dev
);
4827 goto rollback_upper_mesh
;
4830 /* add upper_dev to every dev's lower device */
4831 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
4832 pr_debug("linking %s's lower device %s with %s\n", dev
->name
,
4833 i
->dev
->name
, upper_dev
->name
);
4834 ret
= __netdev_adjacent_dev_link(i
->dev
, upper_dev
);
4836 goto rollback_lower_mesh
;
4839 call_netdevice_notifiers(NETDEV_CHANGEUPPER
, dev
);
4842 rollback_lower_mesh
:
4844 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
4847 __netdev_adjacent_dev_unlink(i
->dev
, upper_dev
);
4852 rollback_upper_mesh
:
4854 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
) {
4857 __netdev_adjacent_dev_unlink(dev
, i
->dev
);
4865 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
4866 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
) {
4867 if (i
== to_i
&& j
== to_j
)
4869 __netdev_adjacent_dev_unlink(i
->dev
, j
->dev
);
4875 __netdev_adjacent_dev_unlink_neighbour(dev
, upper_dev
);
4881 * netdev_upper_dev_link - Add a link to the upper device
4883 * @upper_dev: new upper device
4885 * Adds a link to device which is upper to this one. The caller must hold
4886 * the RTNL lock. On a failure a negative errno code is returned.
4887 * On success the reference counts are adjusted and the function
4890 int netdev_upper_dev_link(struct net_device
*dev
,
4891 struct net_device
*upper_dev
)
4893 return __netdev_upper_dev_link(dev
, upper_dev
, false, NULL
);
4895 EXPORT_SYMBOL(netdev_upper_dev_link
);
4898 * netdev_master_upper_dev_link - Add a master link to the upper device
4900 * @upper_dev: new upper device
4902 * Adds a link to device which is upper to this one. In this case, only
4903 * one master upper device can be linked, although other non-master devices
4904 * might be linked as well. The caller must hold the RTNL lock.
4905 * On a failure a negative errno code is returned. On success the reference
4906 * counts are adjusted and the function returns zero.
4908 int netdev_master_upper_dev_link(struct net_device
*dev
,
4909 struct net_device
*upper_dev
)
4911 return __netdev_upper_dev_link(dev
, upper_dev
, true, NULL
);
4913 EXPORT_SYMBOL(netdev_master_upper_dev_link
);
4915 int netdev_master_upper_dev_link_private(struct net_device
*dev
,
4916 struct net_device
*upper_dev
,
4919 return __netdev_upper_dev_link(dev
, upper_dev
, true, private);
4921 EXPORT_SYMBOL(netdev_master_upper_dev_link_private
);
4924 * netdev_upper_dev_unlink - Removes a link to upper device
4926 * @upper_dev: new upper device
4928 * Removes a link to device which is upper to this one. The caller must hold
4931 void netdev_upper_dev_unlink(struct net_device
*dev
,
4932 struct net_device
*upper_dev
)
4934 struct netdev_adjacent
*i
, *j
;
4937 __netdev_adjacent_dev_unlink_neighbour(dev
, upper_dev
);
4939 /* Here is the tricky part. We must remove all dev's lower
4940 * devices from all upper_dev's upper devices and vice
4941 * versa, to maintain the graph relationship.
4943 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
)
4944 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
)
4945 __netdev_adjacent_dev_unlink(i
->dev
, j
->dev
);
4947 /* remove also the devices itself from lower/upper device
4950 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
)
4951 __netdev_adjacent_dev_unlink(i
->dev
, upper_dev
);
4953 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
)
4954 __netdev_adjacent_dev_unlink(dev
, i
->dev
);
4956 call_netdevice_notifiers(NETDEV_CHANGEUPPER
, dev
);
4958 EXPORT_SYMBOL(netdev_upper_dev_unlink
);
4960 void *netdev_lower_dev_get_private_rcu(struct net_device
*dev
,
4961 struct net_device
*lower_dev
)
4963 struct netdev_adjacent
*lower
;
4967 lower
= __netdev_find_adj_rcu(dev
, lower_dev
, &dev
->adj_list
.lower
);
4971 return lower
->private;
4973 EXPORT_SYMBOL(netdev_lower_dev_get_private_rcu
);
4975 void *netdev_lower_dev_get_private(struct net_device
*dev
,
4976 struct net_device
*lower_dev
)
4978 struct netdev_adjacent
*lower
;
4982 lower
= __netdev_find_adj(dev
, lower_dev
, &dev
->adj_list
.lower
);
4986 return lower
->private;
4988 EXPORT_SYMBOL(netdev_lower_dev_get_private
);
4990 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
4992 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4994 if ((dev
->flags
& IFF_UP
) && ops
->ndo_change_rx_flags
)
4995 ops
->ndo_change_rx_flags(dev
, flags
);
4998 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
, bool notify
)
5000 unsigned int old_flags
= dev
->flags
;
5006 dev
->flags
|= IFF_PROMISC
;
5007 dev
->promiscuity
+= inc
;
5008 if (dev
->promiscuity
== 0) {
5011 * If inc causes overflow, untouch promisc and return error.
5014 dev
->flags
&= ~IFF_PROMISC
;
5016 dev
->promiscuity
-= inc
;
5017 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5022 if (dev
->flags
!= old_flags
) {
5023 pr_info("device %s %s promiscuous mode\n",
5025 dev
->flags
& IFF_PROMISC
? "entered" : "left");
5026 if (audit_enabled
) {
5027 current_uid_gid(&uid
, &gid
);
5028 audit_log(current
->audit_context
, GFP_ATOMIC
,
5029 AUDIT_ANOM_PROMISCUOUS
,
5030 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5031 dev
->name
, (dev
->flags
& IFF_PROMISC
),
5032 (old_flags
& IFF_PROMISC
),
5033 from_kuid(&init_user_ns
, audit_get_loginuid(current
)),
5034 from_kuid(&init_user_ns
, uid
),
5035 from_kgid(&init_user_ns
, gid
),
5036 audit_get_sessionid(current
));
5039 dev_change_rx_flags(dev
, IFF_PROMISC
);
5042 __dev_notify_flags(dev
, old_flags
, IFF_PROMISC
);
5047 * dev_set_promiscuity - update promiscuity count on a device
5051 * Add or remove promiscuity from a device. While the count in the device
5052 * remains above zero the interface remains promiscuous. Once it hits zero
5053 * the device reverts back to normal filtering operation. A negative inc
5054 * value is used to drop promiscuity on the device.
5055 * Return 0 if successful or a negative errno code on error.
5057 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
5059 unsigned int old_flags
= dev
->flags
;
5062 err
= __dev_set_promiscuity(dev
, inc
, true);
5065 if (dev
->flags
!= old_flags
)
5066 dev_set_rx_mode(dev
);
5069 EXPORT_SYMBOL(dev_set_promiscuity
);
5071 static int __dev_set_allmulti(struct net_device
*dev
, int inc
, bool notify
)
5073 unsigned int old_flags
= dev
->flags
, old_gflags
= dev
->gflags
;
5077 dev
->flags
|= IFF_ALLMULTI
;
5078 dev
->allmulti
+= inc
;
5079 if (dev
->allmulti
== 0) {
5082 * If inc causes overflow, untouch allmulti and return error.
5085 dev
->flags
&= ~IFF_ALLMULTI
;
5087 dev
->allmulti
-= inc
;
5088 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5093 if (dev
->flags
^ old_flags
) {
5094 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
5095 dev_set_rx_mode(dev
);
5097 __dev_notify_flags(dev
, old_flags
,
5098 dev
->gflags
^ old_gflags
);
5104 * dev_set_allmulti - update allmulti count on a device
5108 * Add or remove reception of all multicast frames to a device. While the
5109 * count in the device remains above zero the interface remains listening
5110 * to all interfaces. Once it hits zero the device reverts back to normal
5111 * filtering operation. A negative @inc value is used to drop the counter
5112 * when releasing a resource needing all multicasts.
5113 * Return 0 if successful or a negative errno code on error.
5116 int dev_set_allmulti(struct net_device
*dev
, int inc
)
5118 return __dev_set_allmulti(dev
, inc
, true);
5120 EXPORT_SYMBOL(dev_set_allmulti
);
5123 * Upload unicast and multicast address lists to device and
5124 * configure RX filtering. When the device doesn't support unicast
5125 * filtering it is put in promiscuous mode while unicast addresses
5128 void __dev_set_rx_mode(struct net_device
*dev
)
5130 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5132 /* dev_open will call this function so the list will stay sane. */
5133 if (!(dev
->flags
&IFF_UP
))
5136 if (!netif_device_present(dev
))
5139 if (!(dev
->priv_flags
& IFF_UNICAST_FLT
)) {
5140 /* Unicast addresses changes may only happen under the rtnl,
5141 * therefore calling __dev_set_promiscuity here is safe.
5143 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
5144 __dev_set_promiscuity(dev
, 1, false);
5145 dev
->uc_promisc
= true;
5146 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
5147 __dev_set_promiscuity(dev
, -1, false);
5148 dev
->uc_promisc
= false;
5152 if (ops
->ndo_set_rx_mode
)
5153 ops
->ndo_set_rx_mode(dev
);
5156 void dev_set_rx_mode(struct net_device
*dev
)
5158 netif_addr_lock_bh(dev
);
5159 __dev_set_rx_mode(dev
);
5160 netif_addr_unlock_bh(dev
);
5164 * dev_get_flags - get flags reported to userspace
5167 * Get the combination of flag bits exported through APIs to userspace.
5169 unsigned int dev_get_flags(const struct net_device
*dev
)
5173 flags
= (dev
->flags
& ~(IFF_PROMISC
|
5178 (dev
->gflags
& (IFF_PROMISC
|
5181 if (netif_running(dev
)) {
5182 if (netif_oper_up(dev
))
5183 flags
|= IFF_RUNNING
;
5184 if (netif_carrier_ok(dev
))
5185 flags
|= IFF_LOWER_UP
;
5186 if (netif_dormant(dev
))
5187 flags
|= IFF_DORMANT
;
5192 EXPORT_SYMBOL(dev_get_flags
);
5194 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
5196 unsigned int old_flags
= dev
->flags
;
5202 * Set the flags on our device.
5205 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
5206 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
5208 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
5212 * Load in the correct multicast list now the flags have changed.
5215 if ((old_flags
^ flags
) & IFF_MULTICAST
)
5216 dev_change_rx_flags(dev
, IFF_MULTICAST
);
5218 dev_set_rx_mode(dev
);
5221 * Have we downed the interface. We handle IFF_UP ourselves
5222 * according to user attempts to set it, rather than blindly
5227 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
5228 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
5231 dev_set_rx_mode(dev
);
5234 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
5235 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
5236 unsigned int old_flags
= dev
->flags
;
5238 dev
->gflags
^= IFF_PROMISC
;
5240 if (__dev_set_promiscuity(dev
, inc
, false) >= 0)
5241 if (dev
->flags
!= old_flags
)
5242 dev_set_rx_mode(dev
);
5245 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5246 is important. Some (broken) drivers set IFF_PROMISC, when
5247 IFF_ALLMULTI is requested not asking us and not reporting.
5249 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
5250 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
5252 dev
->gflags
^= IFF_ALLMULTI
;
5253 __dev_set_allmulti(dev
, inc
, false);
5259 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
,
5260 unsigned int gchanges
)
5262 unsigned int changes
= dev
->flags
^ old_flags
;
5265 rtmsg_ifinfo(RTM_NEWLINK
, dev
, gchanges
, GFP_ATOMIC
);
5267 if (changes
& IFF_UP
) {
5268 if (dev
->flags
& IFF_UP
)
5269 call_netdevice_notifiers(NETDEV_UP
, dev
);
5271 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
5274 if (dev
->flags
& IFF_UP
&&
5275 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
))) {
5276 struct netdev_notifier_change_info change_info
;
5278 change_info
.flags_changed
= changes
;
5279 call_netdevice_notifiers_info(NETDEV_CHANGE
, dev
,
5285 * dev_change_flags - change device settings
5287 * @flags: device state flags
5289 * Change settings on device based state flags. The flags are
5290 * in the userspace exported format.
5292 int dev_change_flags(struct net_device
*dev
, unsigned int flags
)
5295 unsigned int changes
, old_flags
= dev
->flags
, old_gflags
= dev
->gflags
;
5297 ret
= __dev_change_flags(dev
, flags
);
5301 changes
= (old_flags
^ dev
->flags
) | (old_gflags
^ dev
->gflags
);
5302 __dev_notify_flags(dev
, old_flags
, changes
);
5305 EXPORT_SYMBOL(dev_change_flags
);
5308 * dev_set_mtu - Change maximum transfer unit
5310 * @new_mtu: new transfer unit
5312 * Change the maximum transfer size of the network device.
5314 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
5316 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5319 if (new_mtu
== dev
->mtu
)
5322 /* MTU must be positive. */
5326 if (!netif_device_present(dev
))
5330 if (ops
->ndo_change_mtu
)
5331 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
5336 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
5339 EXPORT_SYMBOL(dev_set_mtu
);
5342 * dev_set_group - Change group this device belongs to
5344 * @new_group: group this device should belong to
5346 void dev_set_group(struct net_device
*dev
, int new_group
)
5348 dev
->group
= new_group
;
5350 EXPORT_SYMBOL(dev_set_group
);
5353 * dev_set_mac_address - Change Media Access Control Address
5357 * Change the hardware (MAC) address of the device
5359 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
5361 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5364 if (!ops
->ndo_set_mac_address
)
5366 if (sa
->sa_family
!= dev
->type
)
5368 if (!netif_device_present(dev
))
5370 err
= ops
->ndo_set_mac_address(dev
, sa
);
5373 dev
->addr_assign_type
= NET_ADDR_SET
;
5374 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
5375 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
5378 EXPORT_SYMBOL(dev_set_mac_address
);
5381 * dev_change_carrier - Change device carrier
5383 * @new_carrier: new value
5385 * Change device carrier
5387 int dev_change_carrier(struct net_device
*dev
, bool new_carrier
)
5389 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5391 if (!ops
->ndo_change_carrier
)
5393 if (!netif_device_present(dev
))
5395 return ops
->ndo_change_carrier(dev
, new_carrier
);
5397 EXPORT_SYMBOL(dev_change_carrier
);
5400 * dev_get_phys_port_id - Get device physical port ID
5404 * Get device physical port ID
5406 int dev_get_phys_port_id(struct net_device
*dev
,
5407 struct netdev_phys_port_id
*ppid
)
5409 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5411 if (!ops
->ndo_get_phys_port_id
)
5413 return ops
->ndo_get_phys_port_id(dev
, ppid
);
5415 EXPORT_SYMBOL(dev_get_phys_port_id
);
5418 * dev_new_index - allocate an ifindex
5419 * @net: the applicable net namespace
5421 * Returns a suitable unique value for a new device interface
5422 * number. The caller must hold the rtnl semaphore or the
5423 * dev_base_lock to be sure it remains unique.
5425 static int dev_new_index(struct net
*net
)
5427 int ifindex
= net
->ifindex
;
5431 if (!__dev_get_by_index(net
, ifindex
))
5432 return net
->ifindex
= ifindex
;
5436 /* Delayed registration/unregisteration */
5437 static LIST_HEAD(net_todo_list
);
5438 static DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq
);
5440 static void net_set_todo(struct net_device
*dev
)
5442 list_add_tail(&dev
->todo_list
, &net_todo_list
);
5443 dev_net(dev
)->dev_unreg_count
++;
5446 static void rollback_registered_many(struct list_head
*head
)
5448 struct net_device
*dev
, *tmp
;
5449 LIST_HEAD(close_head
);
5451 BUG_ON(dev_boot_phase
);
5454 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
5455 /* Some devices call without registering
5456 * for initialization unwind. Remove those
5457 * devices and proceed with the remaining.
5459 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5460 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5464 list_del(&dev
->unreg_list
);
5467 dev
->dismantle
= true;
5468 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
5471 /* If device is running, close it first. */
5472 list_for_each_entry(dev
, head
, unreg_list
)
5473 list_add_tail(&dev
->close_list
, &close_head
);
5474 dev_close_many(&close_head
);
5476 list_for_each_entry(dev
, head
, unreg_list
) {
5477 /* And unlink it from device chain. */
5478 unlist_netdevice(dev
);
5480 dev
->reg_state
= NETREG_UNREGISTERING
;
5485 list_for_each_entry(dev
, head
, unreg_list
) {
5486 /* Shutdown queueing discipline. */
5490 /* Notify protocols, that we are about to destroy
5491 this device. They should clean all the things.
5493 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5495 if (!dev
->rtnl_link_ops
||
5496 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5497 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U, GFP_KERNEL
);
5500 * Flush the unicast and multicast chains
5505 if (dev
->netdev_ops
->ndo_uninit
)
5506 dev
->netdev_ops
->ndo_uninit(dev
);
5508 /* Notifier chain MUST detach us all upper devices. */
5509 WARN_ON(netdev_has_any_upper_dev(dev
));
5511 /* Remove entries from kobject tree */
5512 netdev_unregister_kobject(dev
);
5514 /* Remove XPS queueing entries */
5515 netif_reset_xps_queues_gt(dev
, 0);
5521 list_for_each_entry(dev
, head
, unreg_list
)
5525 static void rollback_registered(struct net_device
*dev
)
5529 list_add(&dev
->unreg_list
, &single
);
5530 rollback_registered_many(&single
);
5534 static netdev_features_t
netdev_fix_features(struct net_device
*dev
,
5535 netdev_features_t features
)
5537 /* Fix illegal checksum combinations */
5538 if ((features
& NETIF_F_HW_CSUM
) &&
5539 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5540 netdev_warn(dev
, "mixed HW and IP checksum settings.\n");
5541 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5544 /* TSO requires that SG is present as well. */
5545 if ((features
& NETIF_F_ALL_TSO
) && !(features
& NETIF_F_SG
)) {
5546 netdev_dbg(dev
, "Dropping TSO features since no SG feature.\n");
5547 features
&= ~NETIF_F_ALL_TSO
;
5550 if ((features
& NETIF_F_TSO
) && !(features
& NETIF_F_HW_CSUM
) &&
5551 !(features
& NETIF_F_IP_CSUM
)) {
5552 netdev_dbg(dev
, "Dropping TSO features since no CSUM feature.\n");
5553 features
&= ~NETIF_F_TSO
;
5554 features
&= ~NETIF_F_TSO_ECN
;
5557 if ((features
& NETIF_F_TSO6
) && !(features
& NETIF_F_HW_CSUM
) &&
5558 !(features
& NETIF_F_IPV6_CSUM
)) {
5559 netdev_dbg(dev
, "Dropping TSO6 features since no CSUM feature.\n");
5560 features
&= ~NETIF_F_TSO6
;
5563 /* TSO ECN requires that TSO is present as well. */
5564 if ((features
& NETIF_F_ALL_TSO
) == NETIF_F_TSO_ECN
)
5565 features
&= ~NETIF_F_TSO_ECN
;
5567 /* Software GSO depends on SG. */
5568 if ((features
& NETIF_F_GSO
) && !(features
& NETIF_F_SG
)) {
5569 netdev_dbg(dev
, "Dropping NETIF_F_GSO since no SG feature.\n");
5570 features
&= ~NETIF_F_GSO
;
5573 /* UFO needs SG and checksumming */
5574 if (features
& NETIF_F_UFO
) {
5575 /* maybe split UFO into V4 and V6? */
5576 if (!((features
& NETIF_F_GEN_CSUM
) ||
5577 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))
5578 == (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5580 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5581 features
&= ~NETIF_F_UFO
;
5584 if (!(features
& NETIF_F_SG
)) {
5586 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5587 features
&= ~NETIF_F_UFO
;
5594 int __netdev_update_features(struct net_device
*dev
)
5596 netdev_features_t features
;
5601 features
= netdev_get_wanted_features(dev
);
5603 if (dev
->netdev_ops
->ndo_fix_features
)
5604 features
= dev
->netdev_ops
->ndo_fix_features(dev
, features
);
5606 /* driver might be less strict about feature dependencies */
5607 features
= netdev_fix_features(dev
, features
);
5609 if (dev
->features
== features
)
5612 netdev_dbg(dev
, "Features changed: %pNF -> %pNF\n",
5613 &dev
->features
, &features
);
5615 if (dev
->netdev_ops
->ndo_set_features
)
5616 err
= dev
->netdev_ops
->ndo_set_features(dev
, features
);
5618 if (unlikely(err
< 0)) {
5620 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5621 err
, &features
, &dev
->features
);
5626 dev
->features
= features
;
5632 * netdev_update_features - recalculate device features
5633 * @dev: the device to check
5635 * Recalculate dev->features set and send notifications if it
5636 * has changed. Should be called after driver or hardware dependent
5637 * conditions might have changed that influence the features.
5639 void netdev_update_features(struct net_device
*dev
)
5641 if (__netdev_update_features(dev
))
5642 netdev_features_change(dev
);
5644 EXPORT_SYMBOL(netdev_update_features
);
5647 * netdev_change_features - recalculate device features
5648 * @dev: the device to check
5650 * Recalculate dev->features set and send notifications even
5651 * if they have not changed. Should be called instead of
5652 * netdev_update_features() if also dev->vlan_features might
5653 * have changed to allow the changes to be propagated to stacked
5656 void netdev_change_features(struct net_device
*dev
)
5658 __netdev_update_features(dev
);
5659 netdev_features_change(dev
);
5661 EXPORT_SYMBOL(netdev_change_features
);
5664 * netif_stacked_transfer_operstate - transfer operstate
5665 * @rootdev: the root or lower level device to transfer state from
5666 * @dev: the device to transfer operstate to
5668 * Transfer operational state from root to device. This is normally
5669 * called when a stacking relationship exists between the root
5670 * device and the device(a leaf device).
5672 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
5673 struct net_device
*dev
)
5675 if (rootdev
->operstate
== IF_OPER_DORMANT
)
5676 netif_dormant_on(dev
);
5678 netif_dormant_off(dev
);
5680 if (netif_carrier_ok(rootdev
)) {
5681 if (!netif_carrier_ok(dev
))
5682 netif_carrier_on(dev
);
5684 if (netif_carrier_ok(dev
))
5685 netif_carrier_off(dev
);
5688 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
5691 static int netif_alloc_rx_queues(struct net_device
*dev
)
5693 unsigned int i
, count
= dev
->num_rx_queues
;
5694 struct netdev_rx_queue
*rx
;
5698 rx
= kcalloc(count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5704 for (i
= 0; i
< count
; i
++)
5710 static void netdev_init_one_queue(struct net_device
*dev
,
5711 struct netdev_queue
*queue
, void *_unused
)
5713 /* Initialize queue lock */
5714 spin_lock_init(&queue
->_xmit_lock
);
5715 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
5716 queue
->xmit_lock_owner
= -1;
5717 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
5720 dql_init(&queue
->dql
, HZ
);
5724 static void netif_free_tx_queues(struct net_device
*dev
)
5726 if (is_vmalloc_addr(dev
->_tx
))
5732 static int netif_alloc_netdev_queues(struct net_device
*dev
)
5734 unsigned int count
= dev
->num_tx_queues
;
5735 struct netdev_queue
*tx
;
5736 size_t sz
= count
* sizeof(*tx
);
5738 BUG_ON(count
< 1 || count
> 0xffff);
5740 tx
= kzalloc(sz
, GFP_KERNEL
| __GFP_NOWARN
| __GFP_REPEAT
);
5748 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5749 spin_lock_init(&dev
->tx_global_lock
);
5755 * register_netdevice - register a network device
5756 * @dev: device to register
5758 * Take a completed network device structure and add it to the kernel
5759 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5760 * chain. 0 is returned on success. A negative errno code is returned
5761 * on a failure to set up the device, or if the name is a duplicate.
5763 * Callers must hold the rtnl semaphore. You may want
5764 * register_netdev() instead of this.
5767 * The locking appears insufficient to guarantee two parallel registers
5768 * will not get the same name.
5771 int register_netdevice(struct net_device
*dev
)
5774 struct net
*net
= dev_net(dev
);
5776 BUG_ON(dev_boot_phase
);
5781 /* When net_device's are persistent, this will be fatal. */
5782 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
5785 spin_lock_init(&dev
->addr_list_lock
);
5786 netdev_set_addr_lockdep_class(dev
);
5790 ret
= dev_get_valid_name(net
, dev
, dev
->name
);
5794 /* Init, if this function is available */
5795 if (dev
->netdev_ops
->ndo_init
) {
5796 ret
= dev
->netdev_ops
->ndo_init(dev
);
5804 if (((dev
->hw_features
| dev
->features
) &
5805 NETIF_F_HW_VLAN_CTAG_FILTER
) &&
5806 (!dev
->netdev_ops
->ndo_vlan_rx_add_vid
||
5807 !dev
->netdev_ops
->ndo_vlan_rx_kill_vid
)) {
5808 netdev_WARN(dev
, "Buggy VLAN acceleration in driver!\n");
5815 dev
->ifindex
= dev_new_index(net
);
5816 else if (__dev_get_by_index(net
, dev
->ifindex
))
5819 if (dev
->iflink
== -1)
5820 dev
->iflink
= dev
->ifindex
;
5822 /* Transfer changeable features to wanted_features and enable
5823 * software offloads (GSO and GRO).
5825 dev
->hw_features
|= NETIF_F_SOFT_FEATURES
;
5826 dev
->features
|= NETIF_F_SOFT_FEATURES
;
5827 dev
->wanted_features
= dev
->features
& dev
->hw_features
;
5829 /* Turn on no cache copy if HW is doing checksum */
5830 if (!(dev
->flags
& IFF_LOOPBACK
)) {
5831 dev
->hw_features
|= NETIF_F_NOCACHE_COPY
;
5832 if (dev
->features
& NETIF_F_ALL_CSUM
) {
5833 dev
->wanted_features
|= NETIF_F_NOCACHE_COPY
;
5834 dev
->features
|= NETIF_F_NOCACHE_COPY
;
5838 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5840 dev
->vlan_features
|= NETIF_F_HIGHDMA
;
5842 /* Make NETIF_F_SG inheritable to tunnel devices.
5844 dev
->hw_enc_features
|= NETIF_F_SG
;
5846 /* Make NETIF_F_SG inheritable to MPLS.
5848 dev
->mpls_features
|= NETIF_F_SG
;
5850 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
5851 ret
= notifier_to_errno(ret
);
5855 ret
= netdev_register_kobject(dev
);
5858 dev
->reg_state
= NETREG_REGISTERED
;
5860 __netdev_update_features(dev
);
5863 * Default initial state at registry is that the
5864 * device is present.
5867 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5869 linkwatch_init_dev(dev
);
5871 dev_init_scheduler(dev
);
5873 list_netdevice(dev
);
5874 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
5876 /* If the device has permanent device address, driver should
5877 * set dev_addr and also addr_assign_type should be set to
5878 * NET_ADDR_PERM (default value).
5880 if (dev
->addr_assign_type
== NET_ADDR_PERM
)
5881 memcpy(dev
->perm_addr
, dev
->dev_addr
, dev
->addr_len
);
5883 /* Notify protocols, that a new device appeared. */
5884 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5885 ret
= notifier_to_errno(ret
);
5887 rollback_registered(dev
);
5888 dev
->reg_state
= NETREG_UNREGISTERED
;
5891 * Prevent userspace races by waiting until the network
5892 * device is fully setup before sending notifications.
5894 if (!dev
->rtnl_link_ops
||
5895 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5896 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U, GFP_KERNEL
);
5902 if (dev
->netdev_ops
->ndo_uninit
)
5903 dev
->netdev_ops
->ndo_uninit(dev
);
5906 EXPORT_SYMBOL(register_netdevice
);
5909 * init_dummy_netdev - init a dummy network device for NAPI
5910 * @dev: device to init
5912 * This takes a network device structure and initialize the minimum
5913 * amount of fields so it can be used to schedule NAPI polls without
5914 * registering a full blown interface. This is to be used by drivers
5915 * that need to tie several hardware interfaces to a single NAPI
5916 * poll scheduler due to HW limitations.
5918 int init_dummy_netdev(struct net_device
*dev
)
5920 /* Clear everything. Note we don't initialize spinlocks
5921 * are they aren't supposed to be taken by any of the
5922 * NAPI code and this dummy netdev is supposed to be
5923 * only ever used for NAPI polls
5925 memset(dev
, 0, sizeof(struct net_device
));
5927 /* make sure we BUG if trying to hit standard
5928 * register/unregister code path
5930 dev
->reg_state
= NETREG_DUMMY
;
5932 /* NAPI wants this */
5933 INIT_LIST_HEAD(&dev
->napi_list
);
5935 /* a dummy interface is started by default */
5936 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5937 set_bit(__LINK_STATE_START
, &dev
->state
);
5939 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5940 * because users of this 'device' dont need to change
5946 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
5950 * register_netdev - register a network device
5951 * @dev: device to register
5953 * Take a completed network device structure and add it to the kernel
5954 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5955 * chain. 0 is returned on success. A negative errno code is returned
5956 * on a failure to set up the device, or if the name is a duplicate.
5958 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5959 * and expands the device name if you passed a format string to
5962 int register_netdev(struct net_device
*dev
)
5967 err
= register_netdevice(dev
);
5971 EXPORT_SYMBOL(register_netdev
);
5973 int netdev_refcnt_read(const struct net_device
*dev
)
5977 for_each_possible_cpu(i
)
5978 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
5981 EXPORT_SYMBOL(netdev_refcnt_read
);
5984 * netdev_wait_allrefs - wait until all references are gone.
5985 * @dev: target net_device
5987 * This is called when unregistering network devices.
5989 * Any protocol or device that holds a reference should register
5990 * for netdevice notification, and cleanup and put back the
5991 * reference if they receive an UNREGISTER event.
5992 * We can get stuck here if buggy protocols don't correctly
5995 static void netdev_wait_allrefs(struct net_device
*dev
)
5997 unsigned long rebroadcast_time
, warning_time
;
6000 linkwatch_forget_dev(dev
);
6002 rebroadcast_time
= warning_time
= jiffies
;
6003 refcnt
= netdev_refcnt_read(dev
);
6005 while (refcnt
!= 0) {
6006 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
6009 /* Rebroadcast unregister notification */
6010 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6016 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
6017 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
6019 /* We must not have linkwatch events
6020 * pending on unregister. If this
6021 * happens, we simply run the queue
6022 * unscheduled, resulting in a noop
6025 linkwatch_run_queue();
6030 rebroadcast_time
= jiffies
;
6035 refcnt
= netdev_refcnt_read(dev
);
6037 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
6038 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6040 warning_time
= jiffies
;
6049 * register_netdevice(x1);
6050 * register_netdevice(x2);
6052 * unregister_netdevice(y1);
6053 * unregister_netdevice(y2);
6059 * We are invoked by rtnl_unlock().
6060 * This allows us to deal with problems:
6061 * 1) We can delete sysfs objects which invoke hotplug
6062 * without deadlocking with linkwatch via keventd.
6063 * 2) Since we run with the RTNL semaphore not held, we can sleep
6064 * safely in order to wait for the netdev refcnt to drop to zero.
6066 * We must not return until all unregister events added during
6067 * the interval the lock was held have been completed.
6069 void netdev_run_todo(void)
6071 struct list_head list
;
6073 /* Snapshot list, allow later requests */
6074 list_replace_init(&net_todo_list
, &list
);
6079 /* Wait for rcu callbacks to finish before next phase */
6080 if (!list_empty(&list
))
6083 while (!list_empty(&list
)) {
6084 struct net_device
*dev
6085 = list_first_entry(&list
, struct net_device
, todo_list
);
6086 list_del(&dev
->todo_list
);
6089 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
6092 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
6093 pr_err("network todo '%s' but state %d\n",
6094 dev
->name
, dev
->reg_state
);
6099 dev
->reg_state
= NETREG_UNREGISTERED
;
6101 on_each_cpu(flush_backlog
, dev
, 1);
6103 netdev_wait_allrefs(dev
);
6106 BUG_ON(netdev_refcnt_read(dev
));
6107 WARN_ON(rcu_access_pointer(dev
->ip_ptr
));
6108 WARN_ON(rcu_access_pointer(dev
->ip6_ptr
));
6109 WARN_ON(dev
->dn_ptr
);
6111 if (dev
->destructor
)
6112 dev
->destructor(dev
);
6114 /* Report a network device has been unregistered */
6116 dev_net(dev
)->dev_unreg_count
--;
6118 wake_up(&netdev_unregistering_wq
);
6120 /* Free network device */
6121 kobject_put(&dev
->dev
.kobj
);
6125 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6126 * fields in the same order, with only the type differing.
6128 void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
6129 const struct net_device_stats
*netdev_stats
)
6131 #if BITS_PER_LONG == 64
6132 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
6133 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
6135 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
6136 const unsigned long *src
= (const unsigned long *)netdev_stats
;
6137 u64
*dst
= (u64
*)stats64
;
6139 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
6140 sizeof(*stats64
) / sizeof(u64
));
6141 for (i
= 0; i
< n
; i
++)
6145 EXPORT_SYMBOL(netdev_stats_to_stats64
);
6148 * dev_get_stats - get network device statistics
6149 * @dev: device to get statistics from
6150 * @storage: place to store stats
6152 * Get network statistics from device. Return @storage.
6153 * The device driver may provide its own method by setting
6154 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6155 * otherwise the internal statistics structure is used.
6157 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
6158 struct rtnl_link_stats64
*storage
)
6160 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6162 if (ops
->ndo_get_stats64
) {
6163 memset(storage
, 0, sizeof(*storage
));
6164 ops
->ndo_get_stats64(dev
, storage
);
6165 } else if (ops
->ndo_get_stats
) {
6166 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
6168 netdev_stats_to_stats64(storage
, &dev
->stats
);
6170 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
6173 EXPORT_SYMBOL(dev_get_stats
);
6175 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
6177 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
6179 #ifdef CONFIG_NET_CLS_ACT
6182 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
6185 netdev_init_one_queue(dev
, queue
, NULL
);
6186 queue
->qdisc
= &noop_qdisc
;
6187 queue
->qdisc_sleeping
= &noop_qdisc
;
6188 rcu_assign_pointer(dev
->ingress_queue
, queue
);
6193 static const struct ethtool_ops default_ethtool_ops
;
6195 void netdev_set_default_ethtool_ops(struct net_device
*dev
,
6196 const struct ethtool_ops
*ops
)
6198 if (dev
->ethtool_ops
== &default_ethtool_ops
)
6199 dev
->ethtool_ops
= ops
;
6201 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops
);
6203 void netdev_freemem(struct net_device
*dev
)
6205 char *addr
= (char *)dev
- dev
->padded
;
6207 if (is_vmalloc_addr(addr
))
6214 * alloc_netdev_mqs - allocate network device
6215 * @sizeof_priv: size of private data to allocate space for
6216 * @name: device name format string
6217 * @setup: callback to initialize device
6218 * @txqs: the number of TX subqueues to allocate
6219 * @rxqs: the number of RX subqueues to allocate
6221 * Allocates a struct net_device with private data area for driver use
6222 * and performs basic initialization. Also allocates subquue structs
6223 * for each queue on the device.
6225 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
6226 void (*setup
)(struct net_device
*),
6227 unsigned int txqs
, unsigned int rxqs
)
6229 struct net_device
*dev
;
6231 struct net_device
*p
;
6233 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
6236 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6242 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6247 alloc_size
= sizeof(struct net_device
);
6249 /* ensure 32-byte alignment of private area */
6250 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
6251 alloc_size
+= sizeof_priv
;
6253 /* ensure 32-byte alignment of whole construct */
6254 alloc_size
+= NETDEV_ALIGN
- 1;
6256 p
= kzalloc(alloc_size
, GFP_KERNEL
| __GFP_NOWARN
| __GFP_REPEAT
);
6258 p
= vzalloc(alloc_size
);
6262 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
6263 dev
->padded
= (char *)dev
- (char *)p
;
6265 dev
->pcpu_refcnt
= alloc_percpu(int);
6266 if (!dev
->pcpu_refcnt
)
6269 if (dev_addr_init(dev
))
6275 dev_net_set(dev
, &init_net
);
6277 dev
->gso_max_size
= GSO_MAX_SIZE
;
6278 dev
->gso_max_segs
= GSO_MAX_SEGS
;
6280 INIT_LIST_HEAD(&dev
->napi_list
);
6281 INIT_LIST_HEAD(&dev
->unreg_list
);
6282 INIT_LIST_HEAD(&dev
->close_list
);
6283 INIT_LIST_HEAD(&dev
->link_watch_list
);
6284 INIT_LIST_HEAD(&dev
->adj_list
.upper
);
6285 INIT_LIST_HEAD(&dev
->adj_list
.lower
);
6286 INIT_LIST_HEAD(&dev
->all_adj_list
.upper
);
6287 INIT_LIST_HEAD(&dev
->all_adj_list
.lower
);
6288 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
6291 dev
->num_tx_queues
= txqs
;
6292 dev
->real_num_tx_queues
= txqs
;
6293 if (netif_alloc_netdev_queues(dev
))
6297 dev
->num_rx_queues
= rxqs
;
6298 dev
->real_num_rx_queues
= rxqs
;
6299 if (netif_alloc_rx_queues(dev
))
6303 strcpy(dev
->name
, name
);
6304 dev
->group
= INIT_NETDEV_GROUP
;
6305 if (!dev
->ethtool_ops
)
6306 dev
->ethtool_ops
= &default_ethtool_ops
;
6314 free_percpu(dev
->pcpu_refcnt
);
6315 netif_free_tx_queues(dev
);
6321 netdev_freemem(dev
);
6324 EXPORT_SYMBOL(alloc_netdev_mqs
);
6327 * free_netdev - free network device
6330 * This function does the last stage of destroying an allocated device
6331 * interface. The reference to the device object is released.
6332 * If this is the last reference then it will be freed.
6334 void free_netdev(struct net_device
*dev
)
6336 struct napi_struct
*p
, *n
;
6338 release_net(dev_net(dev
));
6340 netif_free_tx_queues(dev
);
6345 kfree(rcu_dereference_protected(dev
->ingress_queue
, 1));
6347 /* Flush device addresses */
6348 dev_addr_flush(dev
);
6350 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
6353 free_percpu(dev
->pcpu_refcnt
);
6354 dev
->pcpu_refcnt
= NULL
;
6356 /* Compatibility with error handling in drivers */
6357 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
6358 netdev_freemem(dev
);
6362 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
6363 dev
->reg_state
= NETREG_RELEASED
;
6365 /* will free via device release */
6366 put_device(&dev
->dev
);
6368 EXPORT_SYMBOL(free_netdev
);
6371 * synchronize_net - Synchronize with packet receive processing
6373 * Wait for packets currently being received to be done.
6374 * Does not block later packets from starting.
6376 void synchronize_net(void)
6379 if (rtnl_is_locked())
6380 synchronize_rcu_expedited();
6384 EXPORT_SYMBOL(synchronize_net
);
6387 * unregister_netdevice_queue - remove device from the kernel
6391 * This function shuts down a device interface and removes it
6392 * from the kernel tables.
6393 * If head not NULL, device is queued to be unregistered later.
6395 * Callers must hold the rtnl semaphore. You may want
6396 * unregister_netdev() instead of this.
6399 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
6404 list_move_tail(&dev
->unreg_list
, head
);
6406 rollback_registered(dev
);
6407 /* Finish processing unregister after unlock */
6411 EXPORT_SYMBOL(unregister_netdevice_queue
);
6414 * unregister_netdevice_many - unregister many devices
6415 * @head: list of devices
6417 void unregister_netdevice_many(struct list_head
*head
)
6419 struct net_device
*dev
;
6421 if (!list_empty(head
)) {
6422 rollback_registered_many(head
);
6423 list_for_each_entry(dev
, head
, unreg_list
)
6427 EXPORT_SYMBOL(unregister_netdevice_many
);
6430 * unregister_netdev - remove device from the kernel
6433 * This function shuts down a device interface and removes it
6434 * from the kernel tables.
6436 * This is just a wrapper for unregister_netdevice that takes
6437 * the rtnl semaphore. In general you want to use this and not
6438 * unregister_netdevice.
6440 void unregister_netdev(struct net_device
*dev
)
6443 unregister_netdevice(dev
);
6446 EXPORT_SYMBOL(unregister_netdev
);
6449 * dev_change_net_namespace - move device to different nethost namespace
6451 * @net: network namespace
6452 * @pat: If not NULL name pattern to try if the current device name
6453 * is already taken in the destination network namespace.
6455 * This function shuts down a device interface and moves it
6456 * to a new network namespace. On success 0 is returned, on
6457 * a failure a netagive errno code is returned.
6459 * Callers must hold the rtnl semaphore.
6462 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
6468 /* Don't allow namespace local devices to be moved. */
6470 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6473 /* Ensure the device has been registrered */
6474 if (dev
->reg_state
!= NETREG_REGISTERED
)
6477 /* Get out if there is nothing todo */
6479 if (net_eq(dev_net(dev
), net
))
6482 /* Pick the destination device name, and ensure
6483 * we can use it in the destination network namespace.
6486 if (__dev_get_by_name(net
, dev
->name
)) {
6487 /* We get here if we can't use the current device name */
6490 if (dev_get_valid_name(net
, dev
, pat
) < 0)
6495 * And now a mini version of register_netdevice unregister_netdevice.
6498 /* If device is running close it first. */
6501 /* And unlink it from device chain */
6503 unlist_netdevice(dev
);
6507 /* Shutdown queueing discipline. */
6510 /* Notify protocols, that we are about to destroy
6511 this device. They should clean all the things.
6513 Note that dev->reg_state stays at NETREG_REGISTERED.
6514 This is wanted because this way 8021q and macvlan know
6515 the device is just moving and can keep their slaves up.
6517 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6519 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
6520 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U, GFP_KERNEL
);
6523 * Flush the unicast and multicast chains
6528 /* Send a netdev-removed uevent to the old namespace */
6529 kobject_uevent(&dev
->dev
.kobj
, KOBJ_REMOVE
);
6531 /* Actually switch the network namespace */
6532 dev_net_set(dev
, net
);
6534 /* If there is an ifindex conflict assign a new one */
6535 if (__dev_get_by_index(net
, dev
->ifindex
)) {
6536 int iflink
= (dev
->iflink
== dev
->ifindex
);
6537 dev
->ifindex
= dev_new_index(net
);
6539 dev
->iflink
= dev
->ifindex
;
6542 /* Send a netdev-add uevent to the new namespace */
6543 kobject_uevent(&dev
->dev
.kobj
, KOBJ_ADD
);
6545 /* Fixup kobjects */
6546 err
= device_rename(&dev
->dev
, dev
->name
);
6549 /* Add the device back in the hashes */
6550 list_netdevice(dev
);
6552 /* Notify protocols, that a new device appeared. */
6553 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
6556 * Prevent userspace races by waiting until the network
6557 * device is fully setup before sending notifications.
6559 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U, GFP_KERNEL
);
6566 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
6568 static int dev_cpu_callback(struct notifier_block
*nfb
,
6569 unsigned long action
,
6572 struct sk_buff
**list_skb
;
6573 struct sk_buff
*skb
;
6574 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
6575 struct softnet_data
*sd
, *oldsd
;
6577 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
6580 local_irq_disable();
6581 cpu
= smp_processor_id();
6582 sd
= &per_cpu(softnet_data
, cpu
);
6583 oldsd
= &per_cpu(softnet_data
, oldcpu
);
6585 /* Find end of our completion_queue. */
6586 list_skb
= &sd
->completion_queue
;
6588 list_skb
= &(*list_skb
)->next
;
6589 /* Append completion queue from offline CPU. */
6590 *list_skb
= oldsd
->completion_queue
;
6591 oldsd
->completion_queue
= NULL
;
6593 /* Append output queue from offline CPU. */
6594 if (oldsd
->output_queue
) {
6595 *sd
->output_queue_tailp
= oldsd
->output_queue
;
6596 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
6597 oldsd
->output_queue
= NULL
;
6598 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
6600 /* Append NAPI poll list from offline CPU. */
6601 if (!list_empty(&oldsd
->poll_list
)) {
6602 list_splice_init(&oldsd
->poll_list
, &sd
->poll_list
);
6603 raise_softirq_irqoff(NET_RX_SOFTIRQ
);
6606 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
6609 /* Process offline CPU's input_pkt_queue */
6610 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
6612 input_queue_head_incr(oldsd
);
6614 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
6616 input_queue_head_incr(oldsd
);
6624 * netdev_increment_features - increment feature set by one
6625 * @all: current feature set
6626 * @one: new feature set
6627 * @mask: mask feature set
6629 * Computes a new feature set after adding a device with feature set
6630 * @one to the master device with current feature set @all. Will not
6631 * enable anything that is off in @mask. Returns the new feature set.
6633 netdev_features_t
netdev_increment_features(netdev_features_t all
,
6634 netdev_features_t one
, netdev_features_t mask
)
6636 if (mask
& NETIF_F_GEN_CSUM
)
6637 mask
|= NETIF_F_ALL_CSUM
;
6638 mask
|= NETIF_F_VLAN_CHALLENGED
;
6640 all
|= one
& (NETIF_F_ONE_FOR_ALL
|NETIF_F_ALL_CSUM
) & mask
;
6641 all
&= one
| ~NETIF_F_ALL_FOR_ALL
;
6643 /* If one device supports hw checksumming, set for all. */
6644 if (all
& NETIF_F_GEN_CSUM
)
6645 all
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_GEN_CSUM
);
6649 EXPORT_SYMBOL(netdev_increment_features
);
6651 static struct hlist_head
* __net_init
netdev_create_hash(void)
6654 struct hlist_head
*hash
;
6656 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
6658 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
6659 INIT_HLIST_HEAD(&hash
[i
]);
6664 /* Initialize per network namespace state */
6665 static int __net_init
netdev_init(struct net
*net
)
6667 if (net
!= &init_net
)
6668 INIT_LIST_HEAD(&net
->dev_base_head
);
6670 net
->dev_name_head
= netdev_create_hash();
6671 if (net
->dev_name_head
== NULL
)
6674 net
->dev_index_head
= netdev_create_hash();
6675 if (net
->dev_index_head
== NULL
)
6681 kfree(net
->dev_name_head
);
6687 * netdev_drivername - network driver for the device
6688 * @dev: network device
6690 * Determine network driver for device.
6692 const char *netdev_drivername(const struct net_device
*dev
)
6694 const struct device_driver
*driver
;
6695 const struct device
*parent
;
6696 const char *empty
= "";
6698 parent
= dev
->dev
.parent
;
6702 driver
= parent
->driver
;
6703 if (driver
&& driver
->name
)
6704 return driver
->name
;
6708 static int __netdev_printk(const char *level
, const struct net_device
*dev
,
6709 struct va_format
*vaf
)
6713 if (dev
&& dev
->dev
.parent
) {
6714 r
= dev_printk_emit(level
[1] - '0',
6717 dev_driver_string(dev
->dev
.parent
),
6718 dev_name(dev
->dev
.parent
),
6719 netdev_name(dev
), vaf
);
6721 r
= printk("%s%s: %pV", level
, netdev_name(dev
), vaf
);
6723 r
= printk("%s(NULL net_device): %pV", level
, vaf
);
6729 int netdev_printk(const char *level
, const struct net_device
*dev
,
6730 const char *format
, ...)
6732 struct va_format vaf
;
6736 va_start(args
, format
);
6741 r
= __netdev_printk(level
, dev
, &vaf
);
6747 EXPORT_SYMBOL(netdev_printk
);
6749 #define define_netdev_printk_level(func, level) \
6750 int func(const struct net_device *dev, const char *fmt, ...) \
6753 struct va_format vaf; \
6756 va_start(args, fmt); \
6761 r = __netdev_printk(level, dev, &vaf); \
6767 EXPORT_SYMBOL(func);
6769 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
6770 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
6771 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
6772 define_netdev_printk_level(netdev_err
, KERN_ERR
);
6773 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
6774 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
6775 define_netdev_printk_level(netdev_info
, KERN_INFO
);
6777 static void __net_exit
netdev_exit(struct net
*net
)
6779 kfree(net
->dev_name_head
);
6780 kfree(net
->dev_index_head
);
6783 static struct pernet_operations __net_initdata netdev_net_ops
= {
6784 .init
= netdev_init
,
6785 .exit
= netdev_exit
,
6788 static void __net_exit
default_device_exit(struct net
*net
)
6790 struct net_device
*dev
, *aux
;
6792 * Push all migratable network devices back to the
6793 * initial network namespace
6796 for_each_netdev_safe(net
, dev
, aux
) {
6798 char fb_name
[IFNAMSIZ
];
6800 /* Ignore unmoveable devices (i.e. loopback) */
6801 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6804 /* Leave virtual devices for the generic cleanup */
6805 if (dev
->rtnl_link_ops
)
6808 /* Push remaining network devices to init_net */
6809 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
6810 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
6812 pr_emerg("%s: failed to move %s to init_net: %d\n",
6813 __func__
, dev
->name
, err
);
6820 static void __net_exit
rtnl_lock_unregistering(struct list_head
*net_list
)
6822 /* Return with the rtnl_lock held when there are no network
6823 * devices unregistering in any network namespace in net_list.
6830 prepare_to_wait(&netdev_unregistering_wq
, &wait
,
6831 TASK_UNINTERRUPTIBLE
);
6832 unregistering
= false;
6834 list_for_each_entry(net
, net_list
, exit_list
) {
6835 if (net
->dev_unreg_count
> 0) {
6836 unregistering
= true;
6845 finish_wait(&netdev_unregistering_wq
, &wait
);
6848 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
6850 /* At exit all network devices most be removed from a network
6851 * namespace. Do this in the reverse order of registration.
6852 * Do this across as many network namespaces as possible to
6853 * improve batching efficiency.
6855 struct net_device
*dev
;
6857 LIST_HEAD(dev_kill_list
);
6859 /* To prevent network device cleanup code from dereferencing
6860 * loopback devices or network devices that have been freed
6861 * wait here for all pending unregistrations to complete,
6862 * before unregistring the loopback device and allowing the
6863 * network namespace be freed.
6865 * The netdev todo list containing all network devices
6866 * unregistrations that happen in default_device_exit_batch
6867 * will run in the rtnl_unlock() at the end of
6868 * default_device_exit_batch.
6870 rtnl_lock_unregistering(net_list
);
6871 list_for_each_entry(net
, net_list
, exit_list
) {
6872 for_each_netdev_reverse(net
, dev
) {
6873 if (dev
->rtnl_link_ops
)
6874 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
6876 unregister_netdevice_queue(dev
, &dev_kill_list
);
6879 unregister_netdevice_many(&dev_kill_list
);
6880 list_del(&dev_kill_list
);
6884 static struct pernet_operations __net_initdata default_device_ops
= {
6885 .exit
= default_device_exit
,
6886 .exit_batch
= default_device_exit_batch
,
6890 * Initialize the DEV module. At boot time this walks the device list and
6891 * unhooks any devices that fail to initialise (normally hardware not
6892 * present) and leaves us with a valid list of present and active devices.
6897 * This is called single threaded during boot, so no need
6898 * to take the rtnl semaphore.
6900 static int __init
net_dev_init(void)
6902 int i
, rc
= -ENOMEM
;
6904 BUG_ON(!dev_boot_phase
);
6906 if (dev_proc_init())
6909 if (netdev_kobject_init())
6912 INIT_LIST_HEAD(&ptype_all
);
6913 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
6914 INIT_LIST_HEAD(&ptype_base
[i
]);
6916 INIT_LIST_HEAD(&offload_base
);
6918 if (register_pernet_subsys(&netdev_net_ops
))
6922 * Initialise the packet receive queues.
6925 for_each_possible_cpu(i
) {
6926 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
6928 memset(sd
, 0, sizeof(*sd
));
6929 skb_queue_head_init(&sd
->input_pkt_queue
);
6930 skb_queue_head_init(&sd
->process_queue
);
6931 sd
->completion_queue
= NULL
;
6932 INIT_LIST_HEAD(&sd
->poll_list
);
6933 sd
->output_queue
= NULL
;
6934 sd
->output_queue_tailp
= &sd
->output_queue
;
6936 sd
->csd
.func
= rps_trigger_softirq
;
6942 sd
->backlog
.poll
= process_backlog
;
6943 sd
->backlog
.weight
= weight_p
;
6944 sd
->backlog
.gro_list
= NULL
;
6945 sd
->backlog
.gro_count
= 0;
6947 #ifdef CONFIG_NET_FLOW_LIMIT
6948 sd
->flow_limit
= NULL
;
6954 /* The loopback device is special if any other network devices
6955 * is present in a network namespace the loopback device must
6956 * be present. Since we now dynamically allocate and free the
6957 * loopback device ensure this invariant is maintained by
6958 * keeping the loopback device as the first device on the
6959 * list of network devices. Ensuring the loopback devices
6960 * is the first device that appears and the last network device
6963 if (register_pernet_device(&loopback_net_ops
))
6966 if (register_pernet_device(&default_device_ops
))
6969 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
6970 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
6972 hotcpu_notifier(dev_cpu_callback
, 0);
6979 subsys_initcall(net_dev_init
);