2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <net/net_namespace.h>
99 #include <linux/rtnetlink.h>
100 #include <linux/stat.h>
102 #include <net/pkt_sched.h>
103 #include <net/checksum.h>
104 #include <net/xfrm.h>
105 #include <linux/highmem.h>
106 #include <linux/init.h>
107 #include <linux/module.h>
108 #include <linux/netpoll.h>
109 #include <linux/rcupdate.h>
110 #include <linux/delay.h>
111 #include <net/iw_handler.h>
112 #include <asm/current.h>
113 #include <linux/audit.h>
114 #include <linux/dmaengine.h>
115 #include <linux/err.h>
116 #include <linux/ctype.h>
117 #include <linux/if_arp.h>
118 #include <linux/if_vlan.h>
119 #include <linux/ip.h>
121 #include <linux/ipv6.h>
122 #include <linux/in.h>
123 #include <linux/jhash.h>
124 #include <linux/random.h>
125 #include <trace/events/napi.h>
126 #include <trace/events/net.h>
127 #include <trace/events/skb.h>
128 #include <linux/pci.h>
129 #include <linux/inetdevice.h>
130 #include <linux/cpu_rmap.h>
131 #include <linux/static_key.h>
132 #include <linux/hashtable.h>
133 #include <linux/vmalloc.h>
134 #include <linux/if_macvlan.h>
136 #include "net-sysfs.h"
138 /* Instead of increasing this, you should create a hash table. */
139 #define MAX_GRO_SKBS 8
141 /* This should be increased if a protocol with a bigger head is added. */
142 #define GRO_MAX_HEAD (MAX_HEADER + 128)
144 static DEFINE_SPINLOCK(ptype_lock
);
145 static DEFINE_SPINLOCK(offload_lock
);
146 struct list_head ptype_base
[PTYPE_HASH_SIZE
] __read_mostly
;
147 struct list_head ptype_all __read_mostly
; /* Taps */
148 static struct list_head offload_base __read_mostly
;
150 static int netif_rx_internal(struct sk_buff
*skb
);
151 static int call_netdevice_notifiers_info(unsigned long val
,
152 struct net_device
*dev
,
153 struct netdev_notifier_info
*info
);
156 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
159 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
161 * Writers must hold the rtnl semaphore while they loop through the
162 * dev_base_head list, and hold dev_base_lock for writing when they do the
163 * actual updates. This allows pure readers to access the list even
164 * while a writer is preparing to update it.
166 * To put it another way, dev_base_lock is held for writing only to
167 * protect against pure readers; the rtnl semaphore provides the
168 * protection against other writers.
170 * See, for example usages, register_netdevice() and
171 * unregister_netdevice(), which must be called with the rtnl
174 DEFINE_RWLOCK(dev_base_lock
);
175 EXPORT_SYMBOL(dev_base_lock
);
177 /* protects napi_hash addition/deletion and napi_gen_id */
178 static DEFINE_SPINLOCK(napi_hash_lock
);
180 static unsigned int napi_gen_id
;
181 static DEFINE_HASHTABLE(napi_hash
, 8);
183 static seqcount_t devnet_rename_seq
;
185 static inline void dev_base_seq_inc(struct net
*net
)
187 while (++net
->dev_base_seq
== 0);
190 static inline struct hlist_head
*dev_name_hash(struct net
*net
, const char *name
)
192 unsigned int hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
194 return &net
->dev_name_head
[hash_32(hash
, NETDEV_HASHBITS
)];
197 static inline struct hlist_head
*dev_index_hash(struct net
*net
, int ifindex
)
199 return &net
->dev_index_head
[ifindex
& (NETDEV_HASHENTRIES
- 1)];
202 static inline void rps_lock(struct softnet_data
*sd
)
205 spin_lock(&sd
->input_pkt_queue
.lock
);
209 static inline void rps_unlock(struct softnet_data
*sd
)
212 spin_unlock(&sd
->input_pkt_queue
.lock
);
216 /* Device list insertion */
217 static void list_netdevice(struct net_device
*dev
)
219 struct net
*net
= dev_net(dev
);
223 write_lock_bh(&dev_base_lock
);
224 list_add_tail_rcu(&dev
->dev_list
, &net
->dev_base_head
);
225 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
226 hlist_add_head_rcu(&dev
->index_hlist
,
227 dev_index_hash(net
, dev
->ifindex
));
228 write_unlock_bh(&dev_base_lock
);
230 dev_base_seq_inc(net
);
233 /* Device list removal
234 * caller must respect a RCU grace period before freeing/reusing dev
236 static void unlist_netdevice(struct net_device
*dev
)
240 /* Unlink dev from the device chain */
241 write_lock_bh(&dev_base_lock
);
242 list_del_rcu(&dev
->dev_list
);
243 hlist_del_rcu(&dev
->name_hlist
);
244 hlist_del_rcu(&dev
->index_hlist
);
245 write_unlock_bh(&dev_base_lock
);
247 dev_base_seq_inc(dev_net(dev
));
254 static RAW_NOTIFIER_HEAD(netdev_chain
);
257 * Device drivers call our routines to queue packets here. We empty the
258 * queue in the local softnet handler.
261 DEFINE_PER_CPU_ALIGNED(struct softnet_data
, softnet_data
);
262 EXPORT_PER_CPU_SYMBOL(softnet_data
);
264 #ifdef CONFIG_LOCKDEP
266 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
267 * according to dev->type
269 static const unsigned short netdev_lock_type
[] =
270 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
271 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
272 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
273 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
274 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
275 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
276 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
277 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
278 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
279 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
280 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
281 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
282 ARPHRD_FCFABRIC
, ARPHRD_IEEE80211
, ARPHRD_IEEE80211_PRISM
,
283 ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_PHONET
, ARPHRD_PHONET_PIPE
,
284 ARPHRD_IEEE802154
, ARPHRD_VOID
, ARPHRD_NONE
};
286 static const char *const netdev_lock_name
[] =
287 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
288 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
289 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
290 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
291 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
292 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
293 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
294 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
295 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
296 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
297 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
298 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
299 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
300 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
301 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
303 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
304 static struct lock_class_key netdev_addr_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
306 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
310 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
311 if (netdev_lock_type
[i
] == dev_type
)
313 /* the last key is used by default */
314 return ARRAY_SIZE(netdev_lock_type
) - 1;
317 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
318 unsigned short dev_type
)
322 i
= netdev_lock_pos(dev_type
);
323 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
324 netdev_lock_name
[i
]);
327 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
331 i
= netdev_lock_pos(dev
->type
);
332 lockdep_set_class_and_name(&dev
->addr_list_lock
,
333 &netdev_addr_lock_key
[i
],
334 netdev_lock_name
[i
]);
337 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
338 unsigned short dev_type
)
341 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
346 /*******************************************************************************
348 Protocol management and registration routines
350 *******************************************************************************/
353 * Add a protocol ID to the list. Now that the input handler is
354 * smarter we can dispense with all the messy stuff that used to be
357 * BEWARE!!! Protocol handlers, mangling input packets,
358 * MUST BE last in hash buckets and checking protocol handlers
359 * MUST start from promiscuous ptype_all chain in net_bh.
360 * It is true now, do not change it.
361 * Explanation follows: if protocol handler, mangling packet, will
362 * be the first on list, it is not able to sense, that packet
363 * is cloned and should be copied-on-write, so that it will
364 * change it and subsequent readers will get broken packet.
368 static inline struct list_head
*ptype_head(const struct packet_type
*pt
)
370 if (pt
->type
== htons(ETH_P_ALL
))
373 return &ptype_base
[ntohs(pt
->type
) & PTYPE_HASH_MASK
];
377 * dev_add_pack - add packet handler
378 * @pt: packet type declaration
380 * Add a protocol handler to the networking stack. The passed &packet_type
381 * is linked into kernel lists and may not be freed until it has been
382 * removed from the kernel lists.
384 * This call does not sleep therefore it can not
385 * guarantee all CPU's that are in middle of receiving packets
386 * will see the new packet type (until the next received packet).
389 void dev_add_pack(struct packet_type
*pt
)
391 struct list_head
*head
= ptype_head(pt
);
393 spin_lock(&ptype_lock
);
394 list_add_rcu(&pt
->list
, head
);
395 spin_unlock(&ptype_lock
);
397 EXPORT_SYMBOL(dev_add_pack
);
400 * __dev_remove_pack - remove packet handler
401 * @pt: packet type declaration
403 * Remove a protocol handler that was previously added to the kernel
404 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
405 * from the kernel lists and can be freed or reused once this function
408 * The packet type might still be in use by receivers
409 * and must not be freed until after all the CPU's have gone
410 * through a quiescent state.
412 void __dev_remove_pack(struct packet_type
*pt
)
414 struct list_head
*head
= ptype_head(pt
);
415 struct packet_type
*pt1
;
417 spin_lock(&ptype_lock
);
419 list_for_each_entry(pt1
, head
, list
) {
421 list_del_rcu(&pt
->list
);
426 pr_warn("dev_remove_pack: %p not found\n", pt
);
428 spin_unlock(&ptype_lock
);
430 EXPORT_SYMBOL(__dev_remove_pack
);
433 * dev_remove_pack - remove packet handler
434 * @pt: packet type declaration
436 * Remove a protocol handler that was previously added to the kernel
437 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
438 * from the kernel lists and can be freed or reused once this function
441 * This call sleeps to guarantee that no CPU is looking at the packet
444 void dev_remove_pack(struct packet_type
*pt
)
446 __dev_remove_pack(pt
);
450 EXPORT_SYMBOL(dev_remove_pack
);
454 * dev_add_offload - register offload handlers
455 * @po: protocol offload declaration
457 * Add protocol offload handlers to the networking stack. The passed
458 * &proto_offload is linked into kernel lists and may not be freed until
459 * it has been removed from the kernel lists.
461 * This call does not sleep therefore it can not
462 * guarantee all CPU's that are in middle of receiving packets
463 * will see the new offload handlers (until the next received packet).
465 void dev_add_offload(struct packet_offload
*po
)
467 struct list_head
*head
= &offload_base
;
469 spin_lock(&offload_lock
);
470 list_add_rcu(&po
->list
, head
);
471 spin_unlock(&offload_lock
);
473 EXPORT_SYMBOL(dev_add_offload
);
476 * __dev_remove_offload - remove offload handler
477 * @po: packet offload declaration
479 * Remove a protocol offload handler that was previously added to the
480 * kernel offload handlers by dev_add_offload(). The passed &offload_type
481 * is removed from the kernel lists and can be freed or reused once this
484 * The packet type might still be in use by receivers
485 * and must not be freed until after all the CPU's have gone
486 * through a quiescent state.
488 static void __dev_remove_offload(struct packet_offload
*po
)
490 struct list_head
*head
= &offload_base
;
491 struct packet_offload
*po1
;
493 spin_lock(&offload_lock
);
495 list_for_each_entry(po1
, head
, list
) {
497 list_del_rcu(&po
->list
);
502 pr_warn("dev_remove_offload: %p not found\n", po
);
504 spin_unlock(&offload_lock
);
508 * dev_remove_offload - remove packet offload handler
509 * @po: packet offload declaration
511 * Remove a packet offload handler that was previously added to the kernel
512 * offload handlers by dev_add_offload(). The passed &offload_type is
513 * removed from the kernel lists and can be freed or reused once this
516 * This call sleeps to guarantee that no CPU is looking at the packet
519 void dev_remove_offload(struct packet_offload
*po
)
521 __dev_remove_offload(po
);
525 EXPORT_SYMBOL(dev_remove_offload
);
527 /******************************************************************************
529 Device Boot-time Settings Routines
531 *******************************************************************************/
533 /* Boot time configuration table */
534 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
537 * netdev_boot_setup_add - add new setup entry
538 * @name: name of the device
539 * @map: configured settings for the device
541 * Adds new setup entry to the dev_boot_setup list. The function
542 * returns 0 on error and 1 on success. This is a generic routine to
545 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
547 struct netdev_boot_setup
*s
;
551 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
552 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
553 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
554 strlcpy(s
[i
].name
, name
, IFNAMSIZ
);
555 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
560 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
564 * netdev_boot_setup_check - check boot time settings
565 * @dev: the netdevice
567 * Check boot time settings for the device.
568 * The found settings are set for the device to be used
569 * later in the device probing.
570 * Returns 0 if no settings found, 1 if they are.
572 int netdev_boot_setup_check(struct net_device
*dev
)
574 struct netdev_boot_setup
*s
= dev_boot_setup
;
577 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
578 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
579 !strcmp(dev
->name
, s
[i
].name
)) {
580 dev
->irq
= s
[i
].map
.irq
;
581 dev
->base_addr
= s
[i
].map
.base_addr
;
582 dev
->mem_start
= s
[i
].map
.mem_start
;
583 dev
->mem_end
= s
[i
].map
.mem_end
;
589 EXPORT_SYMBOL(netdev_boot_setup_check
);
593 * netdev_boot_base - get address from boot time settings
594 * @prefix: prefix for network device
595 * @unit: id for network device
597 * Check boot time settings for the base address of device.
598 * The found settings are set for the device to be used
599 * later in the device probing.
600 * Returns 0 if no settings found.
602 unsigned long netdev_boot_base(const char *prefix
, int unit
)
604 const struct netdev_boot_setup
*s
= dev_boot_setup
;
608 sprintf(name
, "%s%d", prefix
, unit
);
611 * If device already registered then return base of 1
612 * to indicate not to probe for this interface
614 if (__dev_get_by_name(&init_net
, name
))
617 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
618 if (!strcmp(name
, s
[i
].name
))
619 return s
[i
].map
.base_addr
;
624 * Saves at boot time configured settings for any netdevice.
626 int __init
netdev_boot_setup(char *str
)
631 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
636 memset(&map
, 0, sizeof(map
));
640 map
.base_addr
= ints
[2];
642 map
.mem_start
= ints
[3];
644 map
.mem_end
= ints
[4];
646 /* Add new entry to the list */
647 return netdev_boot_setup_add(str
, &map
);
650 __setup("netdev=", netdev_boot_setup
);
652 /*******************************************************************************
654 Device Interface Subroutines
656 *******************************************************************************/
659 * __dev_get_by_name - find a device by its name
660 * @net: the applicable net namespace
661 * @name: name to find
663 * Find an interface by name. Must be called under RTNL semaphore
664 * or @dev_base_lock. If the name is found a pointer to the device
665 * is returned. If the name is not found then %NULL is returned. The
666 * reference counters are not incremented so the caller must be
667 * careful with locks.
670 struct net_device
*__dev_get_by_name(struct net
*net
, const char *name
)
672 struct net_device
*dev
;
673 struct hlist_head
*head
= dev_name_hash(net
, name
);
675 hlist_for_each_entry(dev
, head
, name_hlist
)
676 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
681 EXPORT_SYMBOL(__dev_get_by_name
);
684 * dev_get_by_name_rcu - find a device by its name
685 * @net: the applicable net namespace
686 * @name: name to find
688 * Find an interface by name.
689 * If the name is found a pointer to the device is returned.
690 * If the name is not found then %NULL is returned.
691 * The reference counters are not incremented so the caller must be
692 * careful with locks. The caller must hold RCU lock.
695 struct net_device
*dev_get_by_name_rcu(struct net
*net
, const char *name
)
697 struct net_device
*dev
;
698 struct hlist_head
*head
= dev_name_hash(net
, name
);
700 hlist_for_each_entry_rcu(dev
, head
, name_hlist
)
701 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
706 EXPORT_SYMBOL(dev_get_by_name_rcu
);
709 * dev_get_by_name - find a device by its name
710 * @net: the applicable net namespace
711 * @name: name to find
713 * Find an interface by name. This can be called from any
714 * context and does its own locking. The returned handle has
715 * the usage count incremented and the caller must use dev_put() to
716 * release it when it is no longer needed. %NULL is returned if no
717 * matching device is found.
720 struct net_device
*dev_get_by_name(struct net
*net
, const char *name
)
722 struct net_device
*dev
;
725 dev
= dev_get_by_name_rcu(net
, name
);
731 EXPORT_SYMBOL(dev_get_by_name
);
734 * __dev_get_by_index - find a device by its ifindex
735 * @net: the applicable net namespace
736 * @ifindex: index of device
738 * Search for an interface by index. Returns %NULL if the device
739 * is not found or a pointer to the device. The device has not
740 * had its reference counter increased so the caller must be careful
741 * about locking. The caller must hold either the RTNL semaphore
745 struct net_device
*__dev_get_by_index(struct net
*net
, int ifindex
)
747 struct net_device
*dev
;
748 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
750 hlist_for_each_entry(dev
, head
, index_hlist
)
751 if (dev
->ifindex
== ifindex
)
756 EXPORT_SYMBOL(__dev_get_by_index
);
759 * dev_get_by_index_rcu - find a device by its ifindex
760 * @net: the applicable net namespace
761 * @ifindex: index of device
763 * Search for an interface by index. Returns %NULL if the device
764 * is not found or a pointer to the device. The device has not
765 * had its reference counter increased so the caller must be careful
766 * about locking. The caller must hold RCU lock.
769 struct net_device
*dev_get_by_index_rcu(struct net
*net
, int ifindex
)
771 struct net_device
*dev
;
772 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
774 hlist_for_each_entry_rcu(dev
, head
, index_hlist
)
775 if (dev
->ifindex
== ifindex
)
780 EXPORT_SYMBOL(dev_get_by_index_rcu
);
784 * dev_get_by_index - find a device by its ifindex
785 * @net: the applicable net namespace
786 * @ifindex: index of device
788 * Search for an interface by index. Returns NULL if the device
789 * is not found or a pointer to the device. The device returned has
790 * had a reference added and the pointer is safe until the user calls
791 * dev_put to indicate they have finished with it.
794 struct net_device
*dev_get_by_index(struct net
*net
, int ifindex
)
796 struct net_device
*dev
;
799 dev
= dev_get_by_index_rcu(net
, ifindex
);
805 EXPORT_SYMBOL(dev_get_by_index
);
808 * netdev_get_name - get a netdevice name, knowing its ifindex.
809 * @net: network namespace
810 * @name: a pointer to the buffer where the name will be stored.
811 * @ifindex: the ifindex of the interface to get the name from.
813 * The use of raw_seqcount_begin() and cond_resched() before
814 * retrying is required as we want to give the writers a chance
815 * to complete when CONFIG_PREEMPT is not set.
817 int netdev_get_name(struct net
*net
, char *name
, int ifindex
)
819 struct net_device
*dev
;
823 seq
= raw_seqcount_begin(&devnet_rename_seq
);
825 dev
= dev_get_by_index_rcu(net
, ifindex
);
831 strcpy(name
, dev
->name
);
833 if (read_seqcount_retry(&devnet_rename_seq
, seq
)) {
842 * dev_getbyhwaddr_rcu - find a device by its hardware address
843 * @net: the applicable net namespace
844 * @type: media type of device
845 * @ha: hardware address
847 * Search for an interface by MAC address. Returns NULL if the device
848 * is not found or a pointer to the device.
849 * The caller must hold RCU or RTNL.
850 * The returned device has not had its ref count increased
851 * and the caller must therefore be careful about locking
855 struct net_device
*dev_getbyhwaddr_rcu(struct net
*net
, unsigned short type
,
858 struct net_device
*dev
;
860 for_each_netdev_rcu(net
, dev
)
861 if (dev
->type
== type
&&
862 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
867 EXPORT_SYMBOL(dev_getbyhwaddr_rcu
);
869 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
871 struct net_device
*dev
;
874 for_each_netdev(net
, dev
)
875 if (dev
->type
== type
)
880 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
882 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
884 struct net_device
*dev
, *ret
= NULL
;
887 for_each_netdev_rcu(net
, dev
)
888 if (dev
->type
== type
) {
896 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
899 * dev_get_by_flags_rcu - find any device with given flags
900 * @net: the applicable net namespace
901 * @if_flags: IFF_* values
902 * @mask: bitmask of bits in if_flags to check
904 * Search for any interface with the given flags. Returns NULL if a device
905 * is not found or a pointer to the device. Must be called inside
906 * rcu_read_lock(), and result refcount is unchanged.
909 struct net_device
*dev_get_by_flags_rcu(struct net
*net
, unsigned short if_flags
,
912 struct net_device
*dev
, *ret
;
915 for_each_netdev_rcu(net
, dev
) {
916 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
923 EXPORT_SYMBOL(dev_get_by_flags_rcu
);
926 * dev_valid_name - check if name is okay for network device
929 * Network device names need to be valid file names to
930 * to allow sysfs to work. We also disallow any kind of
933 bool dev_valid_name(const char *name
)
937 if (strlen(name
) >= IFNAMSIZ
)
939 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
943 if (*name
== '/' || isspace(*name
))
949 EXPORT_SYMBOL(dev_valid_name
);
952 * __dev_alloc_name - allocate a name for a device
953 * @net: network namespace to allocate the device name in
954 * @name: name format string
955 * @buf: scratch buffer and result name string
957 * Passed a format string - eg "lt%d" it will try and find a suitable
958 * id. It scans list of devices to build up a free map, then chooses
959 * the first empty slot. The caller must hold the dev_base or rtnl lock
960 * while allocating the name and adding the device in order to avoid
962 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
963 * Returns the number of the unit assigned or a negative errno code.
966 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
970 const int max_netdevices
= 8*PAGE_SIZE
;
971 unsigned long *inuse
;
972 struct net_device
*d
;
974 p
= strnchr(name
, IFNAMSIZ
-1, '%');
977 * Verify the string as this thing may have come from
978 * the user. There must be either one "%d" and no other "%"
981 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
984 /* Use one page as a bit array of possible slots */
985 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
989 for_each_netdev(net
, d
) {
990 if (!sscanf(d
->name
, name
, &i
))
992 if (i
< 0 || i
>= max_netdevices
)
995 /* avoid cases where sscanf is not exact inverse of printf */
996 snprintf(buf
, IFNAMSIZ
, name
, i
);
997 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
1001 i
= find_first_zero_bit(inuse
, max_netdevices
);
1002 free_page((unsigned long) inuse
);
1006 snprintf(buf
, IFNAMSIZ
, name
, i
);
1007 if (!__dev_get_by_name(net
, buf
))
1010 /* It is possible to run out of possible slots
1011 * when the name is long and there isn't enough space left
1012 * for the digits, or if all bits are used.
1018 * dev_alloc_name - allocate a name for a device
1020 * @name: name format string
1022 * Passed a format string - eg "lt%d" it will try and find a suitable
1023 * id. It scans list of devices to build up a free map, then chooses
1024 * the first empty slot. The caller must hold the dev_base or rtnl lock
1025 * while allocating the name and adding the device in order to avoid
1027 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1028 * Returns the number of the unit assigned or a negative errno code.
1031 int dev_alloc_name(struct net_device
*dev
, const char *name
)
1037 BUG_ON(!dev_net(dev
));
1039 ret
= __dev_alloc_name(net
, name
, buf
);
1041 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
1044 EXPORT_SYMBOL(dev_alloc_name
);
1046 static int dev_alloc_name_ns(struct net
*net
,
1047 struct net_device
*dev
,
1053 ret
= __dev_alloc_name(net
, name
, buf
);
1055 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
1059 static int dev_get_valid_name(struct net
*net
,
1060 struct net_device
*dev
,
1065 if (!dev_valid_name(name
))
1068 if (strchr(name
, '%'))
1069 return dev_alloc_name_ns(net
, dev
, name
);
1070 else if (__dev_get_by_name(net
, name
))
1072 else if (dev
->name
!= name
)
1073 strlcpy(dev
->name
, name
, IFNAMSIZ
);
1079 * dev_change_name - change name of a device
1081 * @newname: name (or format string) must be at least IFNAMSIZ
1083 * Change name of a device, can pass format strings "eth%d".
1086 int dev_change_name(struct net_device
*dev
, const char *newname
)
1088 char oldname
[IFNAMSIZ
];
1094 BUG_ON(!dev_net(dev
));
1097 if (dev
->flags
& IFF_UP
)
1100 write_seqcount_begin(&devnet_rename_seq
);
1102 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0) {
1103 write_seqcount_end(&devnet_rename_seq
);
1107 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
1109 err
= dev_get_valid_name(net
, dev
, newname
);
1111 write_seqcount_end(&devnet_rename_seq
);
1116 ret
= device_rename(&dev
->dev
, dev
->name
);
1118 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1119 write_seqcount_end(&devnet_rename_seq
);
1123 write_seqcount_end(&devnet_rename_seq
);
1125 netdev_adjacent_rename_links(dev
, oldname
);
1127 write_lock_bh(&dev_base_lock
);
1128 hlist_del_rcu(&dev
->name_hlist
);
1129 write_unlock_bh(&dev_base_lock
);
1133 write_lock_bh(&dev_base_lock
);
1134 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
1135 write_unlock_bh(&dev_base_lock
);
1137 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
1138 ret
= notifier_to_errno(ret
);
1141 /* err >= 0 after dev_alloc_name() or stores the first errno */
1144 write_seqcount_begin(&devnet_rename_seq
);
1145 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1146 memcpy(oldname
, newname
, IFNAMSIZ
);
1149 pr_err("%s: name change rollback failed: %d\n",
1158 * dev_set_alias - change ifalias of a device
1160 * @alias: name up to IFALIASZ
1161 * @len: limit of bytes to copy from info
1163 * Set ifalias for a device,
1165 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
1171 if (len
>= IFALIASZ
)
1175 kfree(dev
->ifalias
);
1176 dev
->ifalias
= NULL
;
1180 new_ifalias
= krealloc(dev
->ifalias
, len
+ 1, GFP_KERNEL
);
1183 dev
->ifalias
= new_ifalias
;
1185 strlcpy(dev
->ifalias
, alias
, len
+1);
1191 * netdev_features_change - device changes features
1192 * @dev: device to cause notification
1194 * Called to indicate a device has changed features.
1196 void netdev_features_change(struct net_device
*dev
)
1198 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1200 EXPORT_SYMBOL(netdev_features_change
);
1203 * netdev_state_change - device changes state
1204 * @dev: device to cause notification
1206 * Called to indicate a device has changed state. This function calls
1207 * the notifier chains for netdev_chain and sends a NEWLINK message
1208 * to the routing socket.
1210 void netdev_state_change(struct net_device
*dev
)
1212 if (dev
->flags
& IFF_UP
) {
1213 struct netdev_notifier_change_info change_info
;
1215 change_info
.flags_changed
= 0;
1216 call_netdevice_notifiers_info(NETDEV_CHANGE
, dev
,
1218 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0, GFP_KERNEL
);
1221 EXPORT_SYMBOL(netdev_state_change
);
1224 * netdev_notify_peers - notify network peers about existence of @dev
1225 * @dev: network device
1227 * Generate traffic such that interested network peers are aware of
1228 * @dev, such as by generating a gratuitous ARP. This may be used when
1229 * a device wants to inform the rest of the network about some sort of
1230 * reconfiguration such as a failover event or virtual machine
1233 void netdev_notify_peers(struct net_device
*dev
)
1236 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
, dev
);
1239 EXPORT_SYMBOL(netdev_notify_peers
);
1241 static int __dev_open(struct net_device
*dev
)
1243 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1248 if (!netif_device_present(dev
))
1251 /* Block netpoll from trying to do any rx path servicing.
1252 * If we don't do this there is a chance ndo_poll_controller
1253 * or ndo_poll may be running while we open the device
1255 netpoll_poll_disable(dev
);
1257 ret
= call_netdevice_notifiers(NETDEV_PRE_UP
, dev
);
1258 ret
= notifier_to_errno(ret
);
1262 set_bit(__LINK_STATE_START
, &dev
->state
);
1264 if (ops
->ndo_validate_addr
)
1265 ret
= ops
->ndo_validate_addr(dev
);
1267 if (!ret
&& ops
->ndo_open
)
1268 ret
= ops
->ndo_open(dev
);
1270 netpoll_poll_enable(dev
);
1273 clear_bit(__LINK_STATE_START
, &dev
->state
);
1275 dev
->flags
|= IFF_UP
;
1276 net_dmaengine_get();
1277 dev_set_rx_mode(dev
);
1279 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
1286 * dev_open - prepare an interface for use.
1287 * @dev: device to open
1289 * Takes a device from down to up state. The device's private open
1290 * function is invoked and then the multicast lists are loaded. Finally
1291 * the device is moved into the up state and a %NETDEV_UP message is
1292 * sent to the netdev notifier chain.
1294 * Calling this function on an active interface is a nop. On a failure
1295 * a negative errno code is returned.
1297 int dev_open(struct net_device
*dev
)
1301 if (dev
->flags
& IFF_UP
)
1304 ret
= __dev_open(dev
);
1308 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
, GFP_KERNEL
);
1309 call_netdevice_notifiers(NETDEV_UP
, dev
);
1313 EXPORT_SYMBOL(dev_open
);
1315 static int __dev_close_many(struct list_head
*head
)
1317 struct net_device
*dev
;
1322 list_for_each_entry(dev
, head
, close_list
) {
1323 /* Temporarily disable netpoll until the interface is down */
1324 netpoll_poll_disable(dev
);
1326 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1328 clear_bit(__LINK_STATE_START
, &dev
->state
);
1330 /* Synchronize to scheduled poll. We cannot touch poll list, it
1331 * can be even on different cpu. So just clear netif_running().
1333 * dev->stop() will invoke napi_disable() on all of it's
1334 * napi_struct instances on this device.
1336 smp_mb__after_atomic(); /* Commit netif_running(). */
1339 dev_deactivate_many(head
);
1341 list_for_each_entry(dev
, head
, close_list
) {
1342 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1345 * Call the device specific close. This cannot fail.
1346 * Only if device is UP
1348 * We allow it to be called even after a DETACH hot-plug
1354 dev
->flags
&= ~IFF_UP
;
1355 net_dmaengine_put();
1356 netpoll_poll_enable(dev
);
1362 static int __dev_close(struct net_device
*dev
)
1367 list_add(&dev
->close_list
, &single
);
1368 retval
= __dev_close_many(&single
);
1374 static int dev_close_many(struct list_head
*head
)
1376 struct net_device
*dev
, *tmp
;
1378 /* Remove the devices that don't need to be closed */
1379 list_for_each_entry_safe(dev
, tmp
, head
, close_list
)
1380 if (!(dev
->flags
& IFF_UP
))
1381 list_del_init(&dev
->close_list
);
1383 __dev_close_many(head
);
1385 list_for_each_entry_safe(dev
, tmp
, head
, close_list
) {
1386 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
, GFP_KERNEL
);
1387 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1388 list_del_init(&dev
->close_list
);
1395 * dev_close - shutdown an interface.
1396 * @dev: device to shutdown
1398 * This function moves an active device into down state. A
1399 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1400 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1403 int dev_close(struct net_device
*dev
)
1405 if (dev
->flags
& IFF_UP
) {
1408 list_add(&dev
->close_list
, &single
);
1409 dev_close_many(&single
);
1414 EXPORT_SYMBOL(dev_close
);
1418 * dev_disable_lro - disable Large Receive Offload on a device
1421 * Disable Large Receive Offload (LRO) on a net device. Must be
1422 * called under RTNL. This is needed if received packets may be
1423 * forwarded to another interface.
1425 void dev_disable_lro(struct net_device
*dev
)
1428 * If we're trying to disable lro on a vlan device
1429 * use the underlying physical device instead
1431 if (is_vlan_dev(dev
))
1432 dev
= vlan_dev_real_dev(dev
);
1434 /* the same for macvlan devices */
1435 if (netif_is_macvlan(dev
))
1436 dev
= macvlan_dev_real_dev(dev
);
1438 dev
->wanted_features
&= ~NETIF_F_LRO
;
1439 netdev_update_features(dev
);
1441 if (unlikely(dev
->features
& NETIF_F_LRO
))
1442 netdev_WARN(dev
, "failed to disable LRO!\n");
1444 EXPORT_SYMBOL(dev_disable_lro
);
1446 static int call_netdevice_notifier(struct notifier_block
*nb
, unsigned long val
,
1447 struct net_device
*dev
)
1449 struct netdev_notifier_info info
;
1451 netdev_notifier_info_init(&info
, dev
);
1452 return nb
->notifier_call(nb
, val
, &info
);
1455 static int dev_boot_phase
= 1;
1458 * register_netdevice_notifier - register a network notifier block
1461 * Register a notifier to be called when network device events occur.
1462 * The notifier passed is linked into the kernel structures and must
1463 * not be reused until it has been unregistered. A negative errno code
1464 * is returned on a failure.
1466 * When registered all registration and up events are replayed
1467 * to the new notifier to allow device to have a race free
1468 * view of the network device list.
1471 int register_netdevice_notifier(struct notifier_block
*nb
)
1473 struct net_device
*dev
;
1474 struct net_device
*last
;
1479 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1485 for_each_netdev(net
, dev
) {
1486 err
= call_netdevice_notifier(nb
, NETDEV_REGISTER
, dev
);
1487 err
= notifier_to_errno(err
);
1491 if (!(dev
->flags
& IFF_UP
))
1494 call_netdevice_notifier(nb
, NETDEV_UP
, dev
);
1505 for_each_netdev(net
, dev
) {
1509 if (dev
->flags
& IFF_UP
) {
1510 call_netdevice_notifier(nb
, NETDEV_GOING_DOWN
,
1512 call_netdevice_notifier(nb
, NETDEV_DOWN
, dev
);
1514 call_netdevice_notifier(nb
, NETDEV_UNREGISTER
, dev
);
1519 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1522 EXPORT_SYMBOL(register_netdevice_notifier
);
1525 * unregister_netdevice_notifier - unregister a network notifier block
1528 * Unregister a notifier previously registered by
1529 * register_netdevice_notifier(). The notifier is unlinked into the
1530 * kernel structures and may then be reused. A negative errno code
1531 * is returned on a failure.
1533 * After unregistering unregister and down device events are synthesized
1534 * for all devices on the device list to the removed notifier to remove
1535 * the need for special case cleanup code.
1538 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1540 struct net_device
*dev
;
1545 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1550 for_each_netdev(net
, dev
) {
1551 if (dev
->flags
& IFF_UP
) {
1552 call_netdevice_notifier(nb
, NETDEV_GOING_DOWN
,
1554 call_netdevice_notifier(nb
, NETDEV_DOWN
, dev
);
1556 call_netdevice_notifier(nb
, NETDEV_UNREGISTER
, dev
);
1563 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1566 * call_netdevice_notifiers_info - call all network notifier blocks
1567 * @val: value passed unmodified to notifier function
1568 * @dev: net_device pointer passed unmodified to notifier function
1569 * @info: notifier information data
1571 * Call all network notifier blocks. Parameters and return value
1572 * are as for raw_notifier_call_chain().
1575 static int call_netdevice_notifiers_info(unsigned long val
,
1576 struct net_device
*dev
,
1577 struct netdev_notifier_info
*info
)
1580 netdev_notifier_info_init(info
, dev
);
1581 return raw_notifier_call_chain(&netdev_chain
, val
, info
);
1585 * call_netdevice_notifiers - call all network notifier blocks
1586 * @val: value passed unmodified to notifier function
1587 * @dev: net_device pointer passed unmodified to notifier function
1589 * Call all network notifier blocks. Parameters and return value
1590 * are as for raw_notifier_call_chain().
1593 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1595 struct netdev_notifier_info info
;
1597 return call_netdevice_notifiers_info(val
, dev
, &info
);
1599 EXPORT_SYMBOL(call_netdevice_notifiers
);
1601 static struct static_key netstamp_needed __read_mostly
;
1602 #ifdef HAVE_JUMP_LABEL
1603 /* We are not allowed to call static_key_slow_dec() from irq context
1604 * If net_disable_timestamp() is called from irq context, defer the
1605 * static_key_slow_dec() calls.
1607 static atomic_t netstamp_needed_deferred
;
1610 void net_enable_timestamp(void)
1612 #ifdef HAVE_JUMP_LABEL
1613 int deferred
= atomic_xchg(&netstamp_needed_deferred
, 0);
1617 static_key_slow_dec(&netstamp_needed
);
1621 static_key_slow_inc(&netstamp_needed
);
1623 EXPORT_SYMBOL(net_enable_timestamp
);
1625 void net_disable_timestamp(void)
1627 #ifdef HAVE_JUMP_LABEL
1628 if (in_interrupt()) {
1629 atomic_inc(&netstamp_needed_deferred
);
1633 static_key_slow_dec(&netstamp_needed
);
1635 EXPORT_SYMBOL(net_disable_timestamp
);
1637 static inline void net_timestamp_set(struct sk_buff
*skb
)
1639 skb
->tstamp
.tv64
= 0;
1640 if (static_key_false(&netstamp_needed
))
1641 __net_timestamp(skb
);
1644 #define net_timestamp_check(COND, SKB) \
1645 if (static_key_false(&netstamp_needed)) { \
1646 if ((COND) && !(SKB)->tstamp.tv64) \
1647 __net_timestamp(SKB); \
1650 bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb)
1654 if (!(dev
->flags
& IFF_UP
))
1657 len
= dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
;
1658 if (skb
->len
<= len
)
1661 /* if TSO is enabled, we don't care about the length as the packet
1662 * could be forwarded without being segmented before
1664 if (skb_is_gso(skb
))
1669 EXPORT_SYMBOL_GPL(is_skb_forwardable
);
1671 int __dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1673 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
1674 if (skb_copy_ubufs(skb
, GFP_ATOMIC
)) {
1675 atomic_long_inc(&dev
->rx_dropped
);
1681 if (unlikely(!is_skb_forwardable(dev
, skb
))) {
1682 atomic_long_inc(&dev
->rx_dropped
);
1687 skb_scrub_packet(skb
, true);
1688 skb
->protocol
= eth_type_trans(skb
, dev
);
1692 EXPORT_SYMBOL_GPL(__dev_forward_skb
);
1695 * dev_forward_skb - loopback an skb to another netif
1697 * @dev: destination network device
1698 * @skb: buffer to forward
1701 * NET_RX_SUCCESS (no congestion)
1702 * NET_RX_DROP (packet was dropped, but freed)
1704 * dev_forward_skb can be used for injecting an skb from the
1705 * start_xmit function of one device into the receive queue
1706 * of another device.
1708 * The receiving device may be in another namespace, so
1709 * we have to clear all information in the skb that could
1710 * impact namespace isolation.
1712 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1714 return __dev_forward_skb(dev
, skb
) ?: netif_rx_internal(skb
);
1716 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1718 static inline int deliver_skb(struct sk_buff
*skb
,
1719 struct packet_type
*pt_prev
,
1720 struct net_device
*orig_dev
)
1722 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
1724 atomic_inc(&skb
->users
);
1725 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1728 static inline bool skb_loop_sk(struct packet_type
*ptype
, struct sk_buff
*skb
)
1730 if (!ptype
->af_packet_priv
|| !skb
->sk
)
1733 if (ptype
->id_match
)
1734 return ptype
->id_match(ptype
, skb
->sk
);
1735 else if ((struct sock
*)ptype
->af_packet_priv
== skb
->sk
)
1742 * Support routine. Sends outgoing frames to any network
1743 * taps currently in use.
1746 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1748 struct packet_type
*ptype
;
1749 struct sk_buff
*skb2
= NULL
;
1750 struct packet_type
*pt_prev
= NULL
;
1753 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1754 /* Never send packets back to the socket
1755 * they originated from - MvS (miquels@drinkel.ow.org)
1757 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1758 (!skb_loop_sk(ptype
, skb
))) {
1760 deliver_skb(skb2
, pt_prev
, skb
->dev
);
1765 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1769 net_timestamp_set(skb2
);
1771 /* skb->nh should be correctly
1772 set by sender, so that the second statement is
1773 just protection against buggy protocols.
1775 skb_reset_mac_header(skb2
);
1777 if (skb_network_header(skb2
) < skb2
->data
||
1778 skb_network_header(skb2
) > skb_tail_pointer(skb2
)) {
1779 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1780 ntohs(skb2
->protocol
),
1782 skb_reset_network_header(skb2
);
1785 skb2
->transport_header
= skb2
->network_header
;
1786 skb2
->pkt_type
= PACKET_OUTGOING
;
1791 pt_prev
->func(skb2
, skb
->dev
, pt_prev
, skb
->dev
);
1796 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1797 * @dev: Network device
1798 * @txq: number of queues available
1800 * If real_num_tx_queues is changed the tc mappings may no longer be
1801 * valid. To resolve this verify the tc mapping remains valid and if
1802 * not NULL the mapping. With no priorities mapping to this
1803 * offset/count pair it will no longer be used. In the worst case TC0
1804 * is invalid nothing can be done so disable priority mappings. If is
1805 * expected that drivers will fix this mapping if they can before
1806 * calling netif_set_real_num_tx_queues.
1808 static void netif_setup_tc(struct net_device
*dev
, unsigned int txq
)
1811 struct netdev_tc_txq
*tc
= &dev
->tc_to_txq
[0];
1813 /* If TC0 is invalidated disable TC mapping */
1814 if (tc
->offset
+ tc
->count
> txq
) {
1815 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1820 /* Invalidated prio to tc mappings set to TC0 */
1821 for (i
= 1; i
< TC_BITMASK
+ 1; i
++) {
1822 int q
= netdev_get_prio_tc_map(dev
, i
);
1824 tc
= &dev
->tc_to_txq
[q
];
1825 if (tc
->offset
+ tc
->count
> txq
) {
1826 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1828 netdev_set_prio_tc_map(dev
, i
, 0);
1834 static DEFINE_MUTEX(xps_map_mutex
);
1835 #define xmap_dereference(P) \
1836 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1838 static struct xps_map
*remove_xps_queue(struct xps_dev_maps
*dev_maps
,
1841 struct xps_map
*map
= NULL
;
1845 map
= xmap_dereference(dev_maps
->cpu_map
[cpu
]);
1847 for (pos
= 0; map
&& pos
< map
->len
; pos
++) {
1848 if (map
->queues
[pos
] == index
) {
1850 map
->queues
[pos
] = map
->queues
[--map
->len
];
1852 RCU_INIT_POINTER(dev_maps
->cpu_map
[cpu
], NULL
);
1853 kfree_rcu(map
, rcu
);
1863 static void netif_reset_xps_queues_gt(struct net_device
*dev
, u16 index
)
1865 struct xps_dev_maps
*dev_maps
;
1867 bool active
= false;
1869 mutex_lock(&xps_map_mutex
);
1870 dev_maps
= xmap_dereference(dev
->xps_maps
);
1875 for_each_possible_cpu(cpu
) {
1876 for (i
= index
; i
< dev
->num_tx_queues
; i
++) {
1877 if (!remove_xps_queue(dev_maps
, cpu
, i
))
1880 if (i
== dev
->num_tx_queues
)
1885 RCU_INIT_POINTER(dev
->xps_maps
, NULL
);
1886 kfree_rcu(dev_maps
, rcu
);
1889 for (i
= index
; i
< dev
->num_tx_queues
; i
++)
1890 netdev_queue_numa_node_write(netdev_get_tx_queue(dev
, i
),
1894 mutex_unlock(&xps_map_mutex
);
1897 static struct xps_map
*expand_xps_map(struct xps_map
*map
,
1900 struct xps_map
*new_map
;
1901 int alloc_len
= XPS_MIN_MAP_ALLOC
;
1904 for (pos
= 0; map
&& pos
< map
->len
; pos
++) {
1905 if (map
->queues
[pos
] != index
)
1910 /* Need to add queue to this CPU's existing map */
1912 if (pos
< map
->alloc_len
)
1915 alloc_len
= map
->alloc_len
* 2;
1918 /* Need to allocate new map to store queue on this CPU's map */
1919 new_map
= kzalloc_node(XPS_MAP_SIZE(alloc_len
), GFP_KERNEL
,
1924 for (i
= 0; i
< pos
; i
++)
1925 new_map
->queues
[i
] = map
->queues
[i
];
1926 new_map
->alloc_len
= alloc_len
;
1932 int netif_set_xps_queue(struct net_device
*dev
, const struct cpumask
*mask
,
1935 struct xps_dev_maps
*dev_maps
, *new_dev_maps
= NULL
;
1936 struct xps_map
*map
, *new_map
;
1937 int maps_sz
= max_t(unsigned int, XPS_DEV_MAPS_SIZE
, L1_CACHE_BYTES
);
1938 int cpu
, numa_node_id
= -2;
1939 bool active
= false;
1941 mutex_lock(&xps_map_mutex
);
1943 dev_maps
= xmap_dereference(dev
->xps_maps
);
1945 /* allocate memory for queue storage */
1946 for_each_online_cpu(cpu
) {
1947 if (!cpumask_test_cpu(cpu
, mask
))
1951 new_dev_maps
= kzalloc(maps_sz
, GFP_KERNEL
);
1952 if (!new_dev_maps
) {
1953 mutex_unlock(&xps_map_mutex
);
1957 map
= dev_maps
? xmap_dereference(dev_maps
->cpu_map
[cpu
]) :
1960 map
= expand_xps_map(map
, cpu
, index
);
1964 RCU_INIT_POINTER(new_dev_maps
->cpu_map
[cpu
], map
);
1968 goto out_no_new_maps
;
1970 for_each_possible_cpu(cpu
) {
1971 if (cpumask_test_cpu(cpu
, mask
) && cpu_online(cpu
)) {
1972 /* add queue to CPU maps */
1975 map
= xmap_dereference(new_dev_maps
->cpu_map
[cpu
]);
1976 while ((pos
< map
->len
) && (map
->queues
[pos
] != index
))
1979 if (pos
== map
->len
)
1980 map
->queues
[map
->len
++] = index
;
1982 if (numa_node_id
== -2)
1983 numa_node_id
= cpu_to_node(cpu
);
1984 else if (numa_node_id
!= cpu_to_node(cpu
))
1987 } else if (dev_maps
) {
1988 /* fill in the new device map from the old device map */
1989 map
= xmap_dereference(dev_maps
->cpu_map
[cpu
]);
1990 RCU_INIT_POINTER(new_dev_maps
->cpu_map
[cpu
], map
);
1995 rcu_assign_pointer(dev
->xps_maps
, new_dev_maps
);
1997 /* Cleanup old maps */
1999 for_each_possible_cpu(cpu
) {
2000 new_map
= xmap_dereference(new_dev_maps
->cpu_map
[cpu
]);
2001 map
= xmap_dereference(dev_maps
->cpu_map
[cpu
]);
2002 if (map
&& map
!= new_map
)
2003 kfree_rcu(map
, rcu
);
2006 kfree_rcu(dev_maps
, rcu
);
2009 dev_maps
= new_dev_maps
;
2013 /* update Tx queue numa node */
2014 netdev_queue_numa_node_write(netdev_get_tx_queue(dev
, index
),
2015 (numa_node_id
>= 0) ? numa_node_id
:
2021 /* removes queue from unused CPUs */
2022 for_each_possible_cpu(cpu
) {
2023 if (cpumask_test_cpu(cpu
, mask
) && cpu_online(cpu
))
2026 if (remove_xps_queue(dev_maps
, cpu
, index
))
2030 /* free map if not active */
2032 RCU_INIT_POINTER(dev
->xps_maps
, NULL
);
2033 kfree_rcu(dev_maps
, rcu
);
2037 mutex_unlock(&xps_map_mutex
);
2041 /* remove any maps that we added */
2042 for_each_possible_cpu(cpu
) {
2043 new_map
= xmap_dereference(new_dev_maps
->cpu_map
[cpu
]);
2044 map
= dev_maps
? xmap_dereference(dev_maps
->cpu_map
[cpu
]) :
2046 if (new_map
&& new_map
!= map
)
2050 mutex_unlock(&xps_map_mutex
);
2052 kfree(new_dev_maps
);
2055 EXPORT_SYMBOL(netif_set_xps_queue
);
2059 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2060 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2062 int netif_set_real_num_tx_queues(struct net_device
*dev
, unsigned int txq
)
2066 if (txq
< 1 || txq
> dev
->num_tx_queues
)
2069 if (dev
->reg_state
== NETREG_REGISTERED
||
2070 dev
->reg_state
== NETREG_UNREGISTERING
) {
2073 rc
= netdev_queue_update_kobjects(dev
, dev
->real_num_tx_queues
,
2079 netif_setup_tc(dev
, txq
);
2081 if (txq
< dev
->real_num_tx_queues
) {
2082 qdisc_reset_all_tx_gt(dev
, txq
);
2084 netif_reset_xps_queues_gt(dev
, txq
);
2089 dev
->real_num_tx_queues
= txq
;
2092 EXPORT_SYMBOL(netif_set_real_num_tx_queues
);
2096 * netif_set_real_num_rx_queues - set actual number of RX queues used
2097 * @dev: Network device
2098 * @rxq: Actual number of RX queues
2100 * This must be called either with the rtnl_lock held or before
2101 * registration of the net device. Returns 0 on success, or a
2102 * negative error code. If called before registration, it always
2105 int netif_set_real_num_rx_queues(struct net_device
*dev
, unsigned int rxq
)
2109 if (rxq
< 1 || rxq
> dev
->num_rx_queues
)
2112 if (dev
->reg_state
== NETREG_REGISTERED
) {
2115 rc
= net_rx_queue_update_kobjects(dev
, dev
->real_num_rx_queues
,
2121 dev
->real_num_rx_queues
= rxq
;
2124 EXPORT_SYMBOL(netif_set_real_num_rx_queues
);
2128 * netif_get_num_default_rss_queues - default number of RSS queues
2130 * This routine should set an upper limit on the number of RSS queues
2131 * used by default by multiqueue devices.
2133 int netif_get_num_default_rss_queues(void)
2135 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES
, num_online_cpus());
2137 EXPORT_SYMBOL(netif_get_num_default_rss_queues
);
2139 static inline void __netif_reschedule(struct Qdisc
*q
)
2141 struct softnet_data
*sd
;
2142 unsigned long flags
;
2144 local_irq_save(flags
);
2145 sd
= &__get_cpu_var(softnet_data
);
2146 q
->next_sched
= NULL
;
2147 *sd
->output_queue_tailp
= q
;
2148 sd
->output_queue_tailp
= &q
->next_sched
;
2149 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
2150 local_irq_restore(flags
);
2153 void __netif_schedule(struct Qdisc
*q
)
2155 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
2156 __netif_reschedule(q
);
2158 EXPORT_SYMBOL(__netif_schedule
);
2160 struct dev_kfree_skb_cb
{
2161 enum skb_free_reason reason
;
2164 static struct dev_kfree_skb_cb
*get_kfree_skb_cb(const struct sk_buff
*skb
)
2166 return (struct dev_kfree_skb_cb
*)skb
->cb
;
2169 void __dev_kfree_skb_irq(struct sk_buff
*skb
, enum skb_free_reason reason
)
2171 unsigned long flags
;
2173 if (likely(atomic_read(&skb
->users
) == 1)) {
2175 atomic_set(&skb
->users
, 0);
2176 } else if (likely(!atomic_dec_and_test(&skb
->users
))) {
2179 get_kfree_skb_cb(skb
)->reason
= reason
;
2180 local_irq_save(flags
);
2181 skb
->next
= __this_cpu_read(softnet_data
.completion_queue
);
2182 __this_cpu_write(softnet_data
.completion_queue
, skb
);
2183 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
2184 local_irq_restore(flags
);
2186 EXPORT_SYMBOL(__dev_kfree_skb_irq
);
2188 void __dev_kfree_skb_any(struct sk_buff
*skb
, enum skb_free_reason reason
)
2190 if (in_irq() || irqs_disabled())
2191 __dev_kfree_skb_irq(skb
, reason
);
2195 EXPORT_SYMBOL(__dev_kfree_skb_any
);
2199 * netif_device_detach - mark device as removed
2200 * @dev: network device
2202 * Mark device as removed from system and therefore no longer available.
2204 void netif_device_detach(struct net_device
*dev
)
2206 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
2207 netif_running(dev
)) {
2208 netif_tx_stop_all_queues(dev
);
2211 EXPORT_SYMBOL(netif_device_detach
);
2214 * netif_device_attach - mark device as attached
2215 * @dev: network device
2217 * Mark device as attached from system and restart if needed.
2219 void netif_device_attach(struct net_device
*dev
)
2221 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
2222 netif_running(dev
)) {
2223 netif_tx_wake_all_queues(dev
);
2224 __netdev_watchdog_up(dev
);
2227 EXPORT_SYMBOL(netif_device_attach
);
2229 static void skb_warn_bad_offload(const struct sk_buff
*skb
)
2231 static const netdev_features_t null_features
= 0;
2232 struct net_device
*dev
= skb
->dev
;
2233 const char *driver
= "";
2235 if (!net_ratelimit())
2238 if (dev
&& dev
->dev
.parent
)
2239 driver
= dev_driver_string(dev
->dev
.parent
);
2241 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2242 "gso_type=%d ip_summed=%d\n",
2243 driver
, dev
? &dev
->features
: &null_features
,
2244 skb
->sk
? &skb
->sk
->sk_route_caps
: &null_features
,
2245 skb
->len
, skb
->data_len
, skb_shinfo(skb
)->gso_size
,
2246 skb_shinfo(skb
)->gso_type
, skb
->ip_summed
);
2250 * Invalidate hardware checksum when packet is to be mangled, and
2251 * complete checksum manually on outgoing path.
2253 int skb_checksum_help(struct sk_buff
*skb
)
2256 int ret
= 0, offset
;
2258 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
2259 goto out_set_summed
;
2261 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
2262 skb_warn_bad_offload(skb
);
2266 /* Before computing a checksum, we should make sure no frag could
2267 * be modified by an external entity : checksum could be wrong.
2269 if (skb_has_shared_frag(skb
)) {
2270 ret
= __skb_linearize(skb
);
2275 offset
= skb_checksum_start_offset(skb
);
2276 BUG_ON(offset
>= skb_headlen(skb
));
2277 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
2279 offset
+= skb
->csum_offset
;
2280 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
2282 if (skb_cloned(skb
) &&
2283 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
2284 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2289 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
2291 skb
->ip_summed
= CHECKSUM_NONE
;
2295 EXPORT_SYMBOL(skb_checksum_help
);
2297 __be16
skb_network_protocol(struct sk_buff
*skb
, int *depth
)
2299 unsigned int vlan_depth
= skb
->mac_len
;
2300 __be16 type
= skb
->protocol
;
2302 /* Tunnel gso handlers can set protocol to ethernet. */
2303 if (type
== htons(ETH_P_TEB
)) {
2306 if (unlikely(!pskb_may_pull(skb
, sizeof(struct ethhdr
))))
2309 eth
= (struct ethhdr
*)skb_mac_header(skb
);
2310 type
= eth
->h_proto
;
2313 /* if skb->protocol is 802.1Q/AD then the header should already be
2314 * present at mac_len - VLAN_HLEN (if mac_len > 0), or at
2315 * ETH_HLEN otherwise
2317 if (type
== htons(ETH_P_8021Q
) || type
== htons(ETH_P_8021AD
)) {
2319 if (unlikely(WARN_ON(vlan_depth
< VLAN_HLEN
)))
2321 vlan_depth
-= VLAN_HLEN
;
2323 vlan_depth
= ETH_HLEN
;
2326 struct vlan_hdr
*vh
;
2328 if (unlikely(!pskb_may_pull(skb
,
2329 vlan_depth
+ VLAN_HLEN
)))
2332 vh
= (struct vlan_hdr
*)(skb
->data
+ vlan_depth
);
2333 type
= vh
->h_vlan_encapsulated_proto
;
2334 vlan_depth
+= VLAN_HLEN
;
2335 } while (type
== htons(ETH_P_8021Q
) ||
2336 type
== htons(ETH_P_8021AD
));
2339 *depth
= vlan_depth
;
2345 * skb_mac_gso_segment - mac layer segmentation handler.
2346 * @skb: buffer to segment
2347 * @features: features for the output path (see dev->features)
2349 struct sk_buff
*skb_mac_gso_segment(struct sk_buff
*skb
,
2350 netdev_features_t features
)
2352 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
2353 struct packet_offload
*ptype
;
2354 int vlan_depth
= skb
->mac_len
;
2355 __be16 type
= skb_network_protocol(skb
, &vlan_depth
);
2357 if (unlikely(!type
))
2358 return ERR_PTR(-EINVAL
);
2360 __skb_pull(skb
, vlan_depth
);
2363 list_for_each_entry_rcu(ptype
, &offload_base
, list
) {
2364 if (ptype
->type
== type
&& ptype
->callbacks
.gso_segment
) {
2365 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
2368 err
= ptype
->callbacks
.gso_send_check(skb
);
2369 segs
= ERR_PTR(err
);
2370 if (err
|| skb_gso_ok(skb
, features
))
2372 __skb_push(skb
, (skb
->data
-
2373 skb_network_header(skb
)));
2375 segs
= ptype
->callbacks
.gso_segment(skb
, features
);
2381 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
2385 EXPORT_SYMBOL(skb_mac_gso_segment
);
2388 /* openvswitch calls this on rx path, so we need a different check.
2390 static inline bool skb_needs_check(struct sk_buff
*skb
, bool tx_path
)
2393 return skb
->ip_summed
!= CHECKSUM_PARTIAL
;
2395 return skb
->ip_summed
== CHECKSUM_NONE
;
2399 * __skb_gso_segment - Perform segmentation on skb.
2400 * @skb: buffer to segment
2401 * @features: features for the output path (see dev->features)
2402 * @tx_path: whether it is called in TX path
2404 * This function segments the given skb and returns a list of segments.
2406 * It may return NULL if the skb requires no segmentation. This is
2407 * only possible when GSO is used for verifying header integrity.
2409 struct sk_buff
*__skb_gso_segment(struct sk_buff
*skb
,
2410 netdev_features_t features
, bool tx_path
)
2412 if (unlikely(skb_needs_check(skb
, tx_path
))) {
2415 skb_warn_bad_offload(skb
);
2417 if (skb_header_cloned(skb
) &&
2418 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
2419 return ERR_PTR(err
);
2422 SKB_GSO_CB(skb
)->mac_offset
= skb_headroom(skb
);
2423 SKB_GSO_CB(skb
)->encap_level
= 0;
2425 skb_reset_mac_header(skb
);
2426 skb_reset_mac_len(skb
);
2428 return skb_mac_gso_segment(skb
, features
);
2430 EXPORT_SYMBOL(__skb_gso_segment
);
2432 /* Take action when hardware reception checksum errors are detected. */
2434 void netdev_rx_csum_fault(struct net_device
*dev
)
2436 if (net_ratelimit()) {
2437 pr_err("%s: hw csum failure\n", dev
? dev
->name
: "<unknown>");
2441 EXPORT_SYMBOL(netdev_rx_csum_fault
);
2444 /* Actually, we should eliminate this check as soon as we know, that:
2445 * 1. IOMMU is present and allows to map all the memory.
2446 * 2. No high memory really exists on this machine.
2449 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
2451 #ifdef CONFIG_HIGHMEM
2453 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
2454 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2455 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2456 if (PageHighMem(skb_frag_page(frag
)))
2461 if (PCI_DMA_BUS_IS_PHYS
) {
2462 struct device
*pdev
= dev
->dev
.parent
;
2466 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2467 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2468 dma_addr_t addr
= page_to_phys(skb_frag_page(frag
));
2469 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
2478 void (*destructor
)(struct sk_buff
*skb
);
2481 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2483 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
2485 struct dev_gso_cb
*cb
;
2487 kfree_skb_list(skb
->next
);
2490 cb
= DEV_GSO_CB(skb
);
2492 cb
->destructor(skb
);
2496 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2497 * @skb: buffer to segment
2498 * @features: device features as applicable to this skb
2500 * This function segments the given skb and stores the list of segments
2503 static int dev_gso_segment(struct sk_buff
*skb
, netdev_features_t features
)
2505 struct sk_buff
*segs
;
2507 segs
= skb_gso_segment(skb
, features
);
2509 /* Verifying header integrity only. */
2514 return PTR_ERR(segs
);
2517 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
2518 skb
->destructor
= dev_gso_skb_destructor
;
2523 /* If MPLS offload request, verify we are testing hardware MPLS features
2524 * instead of standard features for the netdev.
2526 #ifdef CONFIG_NET_MPLS_GSO
2527 static netdev_features_t
net_mpls_features(struct sk_buff
*skb
,
2528 netdev_features_t features
,
2531 if (type
== htons(ETH_P_MPLS_UC
) || type
== htons(ETH_P_MPLS_MC
))
2532 features
&= skb
->dev
->mpls_features
;
2537 static netdev_features_t
net_mpls_features(struct sk_buff
*skb
,
2538 netdev_features_t features
,
2545 static netdev_features_t
harmonize_features(struct sk_buff
*skb
,
2546 netdev_features_t features
)
2551 type
= skb_network_protocol(skb
, &tmp
);
2552 features
= net_mpls_features(skb
, features
, type
);
2554 if (skb
->ip_summed
!= CHECKSUM_NONE
&&
2555 !can_checksum_protocol(features
, type
)) {
2556 features
&= ~NETIF_F_ALL_CSUM
;
2557 } else if (illegal_highdma(skb
->dev
, skb
)) {
2558 features
&= ~NETIF_F_SG
;
2564 netdev_features_t
netif_skb_features(struct sk_buff
*skb
)
2566 __be16 protocol
= skb
->protocol
;
2567 netdev_features_t features
= skb
->dev
->features
;
2569 if (skb_shinfo(skb
)->gso_segs
> skb
->dev
->gso_max_segs
)
2570 features
&= ~NETIF_F_GSO_MASK
;
2572 if (protocol
== htons(ETH_P_8021Q
) || protocol
== htons(ETH_P_8021AD
)) {
2573 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
2574 protocol
= veh
->h_vlan_encapsulated_proto
;
2575 } else if (!vlan_tx_tag_present(skb
)) {
2576 return harmonize_features(skb
, features
);
2579 features
&= (skb
->dev
->vlan_features
| NETIF_F_HW_VLAN_CTAG_TX
|
2580 NETIF_F_HW_VLAN_STAG_TX
);
2582 if (protocol
== htons(ETH_P_8021Q
) || protocol
== htons(ETH_P_8021AD
))
2583 features
&= NETIF_F_SG
| NETIF_F_HIGHDMA
| NETIF_F_FRAGLIST
|
2584 NETIF_F_GEN_CSUM
| NETIF_F_HW_VLAN_CTAG_TX
|
2585 NETIF_F_HW_VLAN_STAG_TX
;
2587 return harmonize_features(skb
, features
);
2589 EXPORT_SYMBOL(netif_skb_features
);
2591 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
2592 struct netdev_queue
*txq
)
2594 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2595 int rc
= NETDEV_TX_OK
;
2596 unsigned int skb_len
;
2598 if (likely(!skb
->next
)) {
2599 netdev_features_t features
;
2602 * If device doesn't need skb->dst, release it right now while
2603 * its hot in this cpu cache
2605 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2608 features
= netif_skb_features(skb
);
2610 if (vlan_tx_tag_present(skb
) &&
2611 !vlan_hw_offload_capable(features
, skb
->vlan_proto
)) {
2612 skb
= __vlan_put_tag(skb
, skb
->vlan_proto
,
2613 vlan_tx_tag_get(skb
));
2620 /* If encapsulation offload request, verify we are testing
2621 * hardware encapsulation features instead of standard
2622 * features for the netdev
2624 if (skb
->encapsulation
)
2625 features
&= dev
->hw_enc_features
;
2627 if (netif_needs_gso(skb
, features
)) {
2628 if (unlikely(dev_gso_segment(skb
, features
)))
2633 if (skb_needs_linearize(skb
, features
) &&
2634 __skb_linearize(skb
))
2637 /* If packet is not checksummed and device does not
2638 * support checksumming for this protocol, complete
2639 * checksumming here.
2641 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2642 if (skb
->encapsulation
)
2643 skb_set_inner_transport_header(skb
,
2644 skb_checksum_start_offset(skb
));
2646 skb_set_transport_header(skb
,
2647 skb_checksum_start_offset(skb
));
2648 if (!(features
& NETIF_F_ALL_CSUM
) &&
2649 skb_checksum_help(skb
))
2654 if (!list_empty(&ptype_all
))
2655 dev_queue_xmit_nit(skb
, dev
);
2658 trace_net_dev_start_xmit(skb
, dev
);
2659 rc
= ops
->ndo_start_xmit(skb
, dev
);
2660 trace_net_dev_xmit(skb
, rc
, dev
, skb_len
);
2661 if (rc
== NETDEV_TX_OK
)
2662 txq_trans_update(txq
);
2668 struct sk_buff
*nskb
= skb
->next
;
2670 skb
->next
= nskb
->next
;
2673 if (!list_empty(&ptype_all
))
2674 dev_queue_xmit_nit(nskb
, dev
);
2676 skb_len
= nskb
->len
;
2677 trace_net_dev_start_xmit(nskb
, dev
);
2678 rc
= ops
->ndo_start_xmit(nskb
, dev
);
2679 trace_net_dev_xmit(nskb
, rc
, dev
, skb_len
);
2680 if (unlikely(rc
!= NETDEV_TX_OK
)) {
2681 if (rc
& ~NETDEV_TX_MASK
)
2682 goto out_kfree_gso_skb
;
2683 nskb
->next
= skb
->next
;
2687 txq_trans_update(txq
);
2688 if (unlikely(netif_xmit_stopped(txq
) && skb
->next
))
2689 return NETDEV_TX_BUSY
;
2690 } while (skb
->next
);
2693 if (likely(skb
->next
== NULL
)) {
2694 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
2703 EXPORT_SYMBOL_GPL(dev_hard_start_xmit
);
2705 static void qdisc_pkt_len_init(struct sk_buff
*skb
)
2707 const struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
2709 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
2711 /* To get more precise estimation of bytes sent on wire,
2712 * we add to pkt_len the headers size of all segments
2714 if (shinfo
->gso_size
) {
2715 unsigned int hdr_len
;
2716 u16 gso_segs
= shinfo
->gso_segs
;
2718 /* mac layer + network layer */
2719 hdr_len
= skb_transport_header(skb
) - skb_mac_header(skb
);
2721 /* + transport layer */
2722 if (likely(shinfo
->gso_type
& (SKB_GSO_TCPV4
| SKB_GSO_TCPV6
)))
2723 hdr_len
+= tcp_hdrlen(skb
);
2725 hdr_len
+= sizeof(struct udphdr
);
2727 if (shinfo
->gso_type
& SKB_GSO_DODGY
)
2728 gso_segs
= DIV_ROUND_UP(skb
->len
- hdr_len
,
2731 qdisc_skb_cb(skb
)->pkt_len
+= (gso_segs
- 1) * hdr_len
;
2735 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2736 struct net_device
*dev
,
2737 struct netdev_queue
*txq
)
2739 spinlock_t
*root_lock
= qdisc_lock(q
);
2743 qdisc_pkt_len_init(skb
);
2744 qdisc_calculate_pkt_len(skb
, q
);
2746 * Heuristic to force contended enqueues to serialize on a
2747 * separate lock before trying to get qdisc main lock.
2748 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2749 * and dequeue packets faster.
2751 contended
= qdisc_is_running(q
);
2752 if (unlikely(contended
))
2753 spin_lock(&q
->busylock
);
2755 spin_lock(root_lock
);
2756 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2759 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2760 qdisc_run_begin(q
)) {
2762 * This is a work-conserving queue; there are no old skbs
2763 * waiting to be sent out; and the qdisc is not running -
2764 * xmit the skb directly.
2766 if (!(dev
->priv_flags
& IFF_XMIT_DST_RELEASE
))
2769 qdisc_bstats_update(q
, skb
);
2771 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
)) {
2772 if (unlikely(contended
)) {
2773 spin_unlock(&q
->busylock
);
2780 rc
= NET_XMIT_SUCCESS
;
2783 rc
= q
->enqueue(skb
, q
) & NET_XMIT_MASK
;
2784 if (qdisc_run_begin(q
)) {
2785 if (unlikely(contended
)) {
2786 spin_unlock(&q
->busylock
);
2792 spin_unlock(root_lock
);
2793 if (unlikely(contended
))
2794 spin_unlock(&q
->busylock
);
2798 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2799 static void skb_update_prio(struct sk_buff
*skb
)
2801 struct netprio_map
*map
= rcu_dereference_bh(skb
->dev
->priomap
);
2803 if (!skb
->priority
&& skb
->sk
&& map
) {
2804 unsigned int prioidx
= skb
->sk
->sk_cgrp_prioidx
;
2806 if (prioidx
< map
->priomap_len
)
2807 skb
->priority
= map
->priomap
[prioidx
];
2811 #define skb_update_prio(skb)
2814 static DEFINE_PER_CPU(int, xmit_recursion
);
2815 #define RECURSION_LIMIT 10
2818 * dev_loopback_xmit - loop back @skb
2819 * @skb: buffer to transmit
2821 int dev_loopback_xmit(struct sk_buff
*skb
)
2823 skb_reset_mac_header(skb
);
2824 __skb_pull(skb
, skb_network_offset(skb
));
2825 skb
->pkt_type
= PACKET_LOOPBACK
;
2826 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
2827 WARN_ON(!skb_dst(skb
));
2832 EXPORT_SYMBOL(dev_loopback_xmit
);
2835 * __dev_queue_xmit - transmit a buffer
2836 * @skb: buffer to transmit
2837 * @accel_priv: private data used for L2 forwarding offload
2839 * Queue a buffer for transmission to a network device. The caller must
2840 * have set the device and priority and built the buffer before calling
2841 * this function. The function can be called from an interrupt.
2843 * A negative errno code is returned on a failure. A success does not
2844 * guarantee the frame will be transmitted as it may be dropped due
2845 * to congestion or traffic shaping.
2847 * -----------------------------------------------------------------------------------
2848 * I notice this method can also return errors from the queue disciplines,
2849 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2852 * Regardless of the return value, the skb is consumed, so it is currently
2853 * difficult to retry a send to this method. (You can bump the ref count
2854 * before sending to hold a reference for retry if you are careful.)
2856 * When calling this method, interrupts MUST be enabled. This is because
2857 * the BH enable code must have IRQs enabled so that it will not deadlock.
2860 static int __dev_queue_xmit(struct sk_buff
*skb
, void *accel_priv
)
2862 struct net_device
*dev
= skb
->dev
;
2863 struct netdev_queue
*txq
;
2867 skb_reset_mac_header(skb
);
2869 /* Disable soft irqs for various locks below. Also
2870 * stops preemption for RCU.
2874 skb_update_prio(skb
);
2876 txq
= netdev_pick_tx(dev
, skb
, accel_priv
);
2877 q
= rcu_dereference_bh(txq
->qdisc
);
2879 #ifdef CONFIG_NET_CLS_ACT
2880 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2882 trace_net_dev_queue(skb
);
2884 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2888 /* The device has no queue. Common case for software devices:
2889 loopback, all the sorts of tunnels...
2891 Really, it is unlikely that netif_tx_lock protection is necessary
2892 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2894 However, it is possible, that they rely on protection
2897 Check this and shot the lock. It is not prone from deadlocks.
2898 Either shot noqueue qdisc, it is even simpler 8)
2900 if (dev
->flags
& IFF_UP
) {
2901 int cpu
= smp_processor_id(); /* ok because BHs are off */
2903 if (txq
->xmit_lock_owner
!= cpu
) {
2905 if (__this_cpu_read(xmit_recursion
) > RECURSION_LIMIT
)
2906 goto recursion_alert
;
2908 HARD_TX_LOCK(dev
, txq
, cpu
);
2910 if (!netif_xmit_stopped(txq
)) {
2911 __this_cpu_inc(xmit_recursion
);
2912 rc
= dev_hard_start_xmit(skb
, dev
, txq
);
2913 __this_cpu_dec(xmit_recursion
);
2914 if (dev_xmit_complete(rc
)) {
2915 HARD_TX_UNLOCK(dev
, txq
);
2919 HARD_TX_UNLOCK(dev
, txq
);
2920 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2923 /* Recursion is detected! It is possible,
2927 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2933 rcu_read_unlock_bh();
2935 atomic_long_inc(&dev
->tx_dropped
);
2939 rcu_read_unlock_bh();
2943 int dev_queue_xmit(struct sk_buff
*skb
)
2945 return __dev_queue_xmit(skb
, NULL
);
2947 EXPORT_SYMBOL(dev_queue_xmit
);
2949 int dev_queue_xmit_accel(struct sk_buff
*skb
, void *accel_priv
)
2951 return __dev_queue_xmit(skb
, accel_priv
);
2953 EXPORT_SYMBOL(dev_queue_xmit_accel
);
2956 /*=======================================================================
2958 =======================================================================*/
2960 int netdev_max_backlog __read_mostly
= 1000;
2961 EXPORT_SYMBOL(netdev_max_backlog
);
2963 int netdev_tstamp_prequeue __read_mostly
= 1;
2964 int netdev_budget __read_mostly
= 300;
2965 int weight_p __read_mostly
= 64; /* old backlog weight */
2967 /* Called with irq disabled */
2968 static inline void ____napi_schedule(struct softnet_data
*sd
,
2969 struct napi_struct
*napi
)
2971 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
2972 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2977 /* One global table that all flow-based protocols share. */
2978 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
2979 EXPORT_SYMBOL(rps_sock_flow_table
);
2981 struct static_key rps_needed __read_mostly
;
2983 static struct rps_dev_flow
*
2984 set_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2985 struct rps_dev_flow
*rflow
, u16 next_cpu
)
2987 if (next_cpu
!= RPS_NO_CPU
) {
2988 #ifdef CONFIG_RFS_ACCEL
2989 struct netdev_rx_queue
*rxqueue
;
2990 struct rps_dev_flow_table
*flow_table
;
2991 struct rps_dev_flow
*old_rflow
;
2996 /* Should we steer this flow to a different hardware queue? */
2997 if (!skb_rx_queue_recorded(skb
) || !dev
->rx_cpu_rmap
||
2998 !(dev
->features
& NETIF_F_NTUPLE
))
3000 rxq_index
= cpu_rmap_lookup_index(dev
->rx_cpu_rmap
, next_cpu
);
3001 if (rxq_index
== skb_get_rx_queue(skb
))
3004 rxqueue
= dev
->_rx
+ rxq_index
;
3005 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3008 flow_id
= skb_get_hash(skb
) & flow_table
->mask
;
3009 rc
= dev
->netdev_ops
->ndo_rx_flow_steer(dev
, skb
,
3010 rxq_index
, flow_id
);
3014 rflow
= &flow_table
->flows
[flow_id
];
3016 if (old_rflow
->filter
== rflow
->filter
)
3017 old_rflow
->filter
= RPS_NO_FILTER
;
3021 per_cpu(softnet_data
, next_cpu
).input_queue_head
;
3024 rflow
->cpu
= next_cpu
;
3029 * get_rps_cpu is called from netif_receive_skb and returns the target
3030 * CPU from the RPS map of the receiving queue for a given skb.
3031 * rcu_read_lock must be held on entry.
3033 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
3034 struct rps_dev_flow
**rflowp
)
3036 struct netdev_rx_queue
*rxqueue
;
3037 struct rps_map
*map
;
3038 struct rps_dev_flow_table
*flow_table
;
3039 struct rps_sock_flow_table
*sock_flow_table
;
3044 if (skb_rx_queue_recorded(skb
)) {
3045 u16 index
= skb_get_rx_queue(skb
);
3046 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
3047 WARN_ONCE(dev
->real_num_rx_queues
> 1,
3048 "%s received packet on queue %u, but number "
3049 "of RX queues is %u\n",
3050 dev
->name
, index
, dev
->real_num_rx_queues
);
3053 rxqueue
= dev
->_rx
+ index
;
3057 map
= rcu_dereference(rxqueue
->rps_map
);
3059 if (map
->len
== 1 &&
3060 !rcu_access_pointer(rxqueue
->rps_flow_table
)) {
3061 tcpu
= map
->cpus
[0];
3062 if (cpu_online(tcpu
))
3066 } else if (!rcu_access_pointer(rxqueue
->rps_flow_table
)) {
3070 skb_reset_network_header(skb
);
3071 hash
= skb_get_hash(skb
);
3075 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3076 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
3077 if (flow_table
&& sock_flow_table
) {
3079 struct rps_dev_flow
*rflow
;
3081 rflow
= &flow_table
->flows
[hash
& flow_table
->mask
];
3084 next_cpu
= sock_flow_table
->ents
[hash
& sock_flow_table
->mask
];
3087 * If the desired CPU (where last recvmsg was done) is
3088 * different from current CPU (one in the rx-queue flow
3089 * table entry), switch if one of the following holds:
3090 * - Current CPU is unset (equal to RPS_NO_CPU).
3091 * - Current CPU is offline.
3092 * - The current CPU's queue tail has advanced beyond the
3093 * last packet that was enqueued using this table entry.
3094 * This guarantees that all previous packets for the flow
3095 * have been dequeued, thus preserving in order delivery.
3097 if (unlikely(tcpu
!= next_cpu
) &&
3098 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
3099 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
3100 rflow
->last_qtail
)) >= 0)) {
3102 rflow
= set_rps_cpu(dev
, skb
, rflow
, next_cpu
);
3105 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
3113 tcpu
= map
->cpus
[((u64
) hash
* map
->len
) >> 32];
3115 if (cpu_online(tcpu
)) {
3125 #ifdef CONFIG_RFS_ACCEL
3128 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3129 * @dev: Device on which the filter was set
3130 * @rxq_index: RX queue index
3131 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3132 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3134 * Drivers that implement ndo_rx_flow_steer() should periodically call
3135 * this function for each installed filter and remove the filters for
3136 * which it returns %true.
3138 bool rps_may_expire_flow(struct net_device
*dev
, u16 rxq_index
,
3139 u32 flow_id
, u16 filter_id
)
3141 struct netdev_rx_queue
*rxqueue
= dev
->_rx
+ rxq_index
;
3142 struct rps_dev_flow_table
*flow_table
;
3143 struct rps_dev_flow
*rflow
;
3148 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3149 if (flow_table
&& flow_id
<= flow_table
->mask
) {
3150 rflow
= &flow_table
->flows
[flow_id
];
3151 cpu
= ACCESS_ONCE(rflow
->cpu
);
3152 if (rflow
->filter
== filter_id
&& cpu
!= RPS_NO_CPU
&&
3153 ((int)(per_cpu(softnet_data
, cpu
).input_queue_head
-
3154 rflow
->last_qtail
) <
3155 (int)(10 * flow_table
->mask
)))
3161 EXPORT_SYMBOL(rps_may_expire_flow
);
3163 #endif /* CONFIG_RFS_ACCEL */
3165 /* Called from hardirq (IPI) context */
3166 static void rps_trigger_softirq(void *data
)
3168 struct softnet_data
*sd
= data
;
3170 ____napi_schedule(sd
, &sd
->backlog
);
3174 #endif /* CONFIG_RPS */
3177 * Check if this softnet_data structure is another cpu one
3178 * If yes, queue it to our IPI list and return 1
3181 static int rps_ipi_queued(struct softnet_data
*sd
)
3184 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
3187 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
3188 mysd
->rps_ipi_list
= sd
;
3190 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3193 #endif /* CONFIG_RPS */
3197 #ifdef CONFIG_NET_FLOW_LIMIT
3198 int netdev_flow_limit_table_len __read_mostly
= (1 << 12);
3201 static bool skb_flow_limit(struct sk_buff
*skb
, unsigned int qlen
)
3203 #ifdef CONFIG_NET_FLOW_LIMIT
3204 struct sd_flow_limit
*fl
;
3205 struct softnet_data
*sd
;
3206 unsigned int old_flow
, new_flow
;
3208 if (qlen
< (netdev_max_backlog
>> 1))
3211 sd
= &__get_cpu_var(softnet_data
);
3214 fl
= rcu_dereference(sd
->flow_limit
);
3216 new_flow
= skb_get_hash(skb
) & (fl
->num_buckets
- 1);
3217 old_flow
= fl
->history
[fl
->history_head
];
3218 fl
->history
[fl
->history_head
] = new_flow
;
3221 fl
->history_head
&= FLOW_LIMIT_HISTORY
- 1;
3223 if (likely(fl
->buckets
[old_flow
]))
3224 fl
->buckets
[old_flow
]--;
3226 if (++fl
->buckets
[new_flow
] > (FLOW_LIMIT_HISTORY
>> 1)) {
3238 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3239 * queue (may be a remote CPU queue).
3241 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
3242 unsigned int *qtail
)
3244 struct softnet_data
*sd
;
3245 unsigned long flags
;
3248 sd
= &per_cpu(softnet_data
, cpu
);
3250 local_irq_save(flags
);
3253 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3254 if (qlen
<= netdev_max_backlog
&& !skb_flow_limit(skb
, qlen
)) {
3255 if (skb_queue_len(&sd
->input_pkt_queue
)) {
3257 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
3258 input_queue_tail_incr_save(sd
, qtail
);
3260 local_irq_restore(flags
);
3261 return NET_RX_SUCCESS
;
3264 /* Schedule NAPI for backlog device
3265 * We can use non atomic operation since we own the queue lock
3267 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
3268 if (!rps_ipi_queued(sd
))
3269 ____napi_schedule(sd
, &sd
->backlog
);
3277 local_irq_restore(flags
);
3279 atomic_long_inc(&skb
->dev
->rx_dropped
);
3284 static int netif_rx_internal(struct sk_buff
*skb
)
3288 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3290 trace_netif_rx(skb
);
3292 if (static_key_false(&rps_needed
)) {
3293 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3299 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3301 cpu
= smp_processor_id();
3303 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3311 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
3318 * netif_rx - post buffer to the network code
3319 * @skb: buffer to post
3321 * This function receives a packet from a device driver and queues it for
3322 * the upper (protocol) levels to process. It always succeeds. The buffer
3323 * may be dropped during processing for congestion control or by the
3327 * NET_RX_SUCCESS (no congestion)
3328 * NET_RX_DROP (packet was dropped)
3332 int netif_rx(struct sk_buff
*skb
)
3334 trace_netif_rx_entry(skb
);
3336 return netif_rx_internal(skb
);
3338 EXPORT_SYMBOL(netif_rx
);
3340 int netif_rx_ni(struct sk_buff
*skb
)
3344 trace_netif_rx_ni_entry(skb
);
3347 err
= netif_rx_internal(skb
);
3348 if (local_softirq_pending())
3354 EXPORT_SYMBOL(netif_rx_ni
);
3356 static void net_tx_action(struct softirq_action
*h
)
3358 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3360 if (sd
->completion_queue
) {
3361 struct sk_buff
*clist
;
3363 local_irq_disable();
3364 clist
= sd
->completion_queue
;
3365 sd
->completion_queue
= NULL
;
3369 struct sk_buff
*skb
= clist
;
3370 clist
= clist
->next
;
3372 WARN_ON(atomic_read(&skb
->users
));
3373 if (likely(get_kfree_skb_cb(skb
)->reason
== SKB_REASON_CONSUMED
))
3374 trace_consume_skb(skb
);
3376 trace_kfree_skb(skb
, net_tx_action
);
3381 if (sd
->output_queue
) {
3384 local_irq_disable();
3385 head
= sd
->output_queue
;
3386 sd
->output_queue
= NULL
;
3387 sd
->output_queue_tailp
= &sd
->output_queue
;
3391 struct Qdisc
*q
= head
;
3392 spinlock_t
*root_lock
;
3394 head
= head
->next_sched
;
3396 root_lock
= qdisc_lock(q
);
3397 if (spin_trylock(root_lock
)) {
3398 smp_mb__before_atomic();
3399 clear_bit(__QDISC_STATE_SCHED
,
3402 spin_unlock(root_lock
);
3404 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
3406 __netif_reschedule(q
);
3408 smp_mb__before_atomic();
3409 clear_bit(__QDISC_STATE_SCHED
,
3417 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3418 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3419 /* This hook is defined here for ATM LANE */
3420 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
3421 unsigned char *addr
) __read_mostly
;
3422 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
3425 #ifdef CONFIG_NET_CLS_ACT
3426 /* TODO: Maybe we should just force sch_ingress to be compiled in
3427 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3428 * a compare and 2 stores extra right now if we dont have it on
3429 * but have CONFIG_NET_CLS_ACT
3430 * NOTE: This doesn't stop any functionality; if you dont have
3431 * the ingress scheduler, you just can't add policies on ingress.
3434 static int ing_filter(struct sk_buff
*skb
, struct netdev_queue
*rxq
)
3436 struct net_device
*dev
= skb
->dev
;
3437 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
3438 int result
= TC_ACT_OK
;
3441 if (unlikely(MAX_RED_LOOP
< ttl
++)) {
3442 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3443 skb
->skb_iif
, dev
->ifindex
);
3447 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
3448 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
3451 if (q
!= &noop_qdisc
) {
3452 spin_lock(qdisc_lock(q
));
3453 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
3454 result
= qdisc_enqueue_root(skb
, q
);
3455 spin_unlock(qdisc_lock(q
));
3461 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
3462 struct packet_type
**pt_prev
,
3463 int *ret
, struct net_device
*orig_dev
)
3465 struct netdev_queue
*rxq
= rcu_dereference(skb
->dev
->ingress_queue
);
3467 if (!rxq
|| rxq
->qdisc
== &noop_qdisc
)
3471 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
3475 switch (ing_filter(skb
, rxq
)) {
3489 * netdev_rx_handler_register - register receive handler
3490 * @dev: device to register a handler for
3491 * @rx_handler: receive handler to register
3492 * @rx_handler_data: data pointer that is used by rx handler
3494 * Register a receive handler for a device. This handler will then be
3495 * called from __netif_receive_skb. A negative errno code is returned
3498 * The caller must hold the rtnl_mutex.
3500 * For a general description of rx_handler, see enum rx_handler_result.
3502 int netdev_rx_handler_register(struct net_device
*dev
,
3503 rx_handler_func_t
*rx_handler
,
3504 void *rx_handler_data
)
3508 if (dev
->rx_handler
)
3511 /* Note: rx_handler_data must be set before rx_handler */
3512 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
3513 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
3517 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
3520 * netdev_rx_handler_unregister - unregister receive handler
3521 * @dev: device to unregister a handler from
3523 * Unregister a receive handler from a device.
3525 * The caller must hold the rtnl_mutex.
3527 void netdev_rx_handler_unregister(struct net_device
*dev
)
3531 RCU_INIT_POINTER(dev
->rx_handler
, NULL
);
3532 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3533 * section has a guarantee to see a non NULL rx_handler_data
3537 RCU_INIT_POINTER(dev
->rx_handler_data
, NULL
);
3539 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
3542 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3543 * the special handling of PFMEMALLOC skbs.
3545 static bool skb_pfmemalloc_protocol(struct sk_buff
*skb
)
3547 switch (skb
->protocol
) {
3548 case htons(ETH_P_ARP
):
3549 case htons(ETH_P_IP
):
3550 case htons(ETH_P_IPV6
):
3551 case htons(ETH_P_8021Q
):
3552 case htons(ETH_P_8021AD
):
3559 static int __netif_receive_skb_core(struct sk_buff
*skb
, bool pfmemalloc
)
3561 struct packet_type
*ptype
, *pt_prev
;
3562 rx_handler_func_t
*rx_handler
;
3563 struct net_device
*orig_dev
;
3564 struct net_device
*null_or_dev
;
3565 bool deliver_exact
= false;
3566 int ret
= NET_RX_DROP
;
3569 net_timestamp_check(!netdev_tstamp_prequeue
, skb
);
3571 trace_netif_receive_skb(skb
);
3573 orig_dev
= skb
->dev
;
3575 skb_reset_network_header(skb
);
3576 if (!skb_transport_header_was_set(skb
))
3577 skb_reset_transport_header(skb
);
3578 skb_reset_mac_len(skb
);
3585 skb
->skb_iif
= skb
->dev
->ifindex
;
3587 __this_cpu_inc(softnet_data
.processed
);
3589 if (skb
->protocol
== cpu_to_be16(ETH_P_8021Q
) ||
3590 skb
->protocol
== cpu_to_be16(ETH_P_8021AD
)) {
3591 skb
= vlan_untag(skb
);
3596 #ifdef CONFIG_NET_CLS_ACT
3597 if (skb
->tc_verd
& TC_NCLS
) {
3598 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
3606 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
3607 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
3609 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3615 #ifdef CONFIG_NET_CLS_ACT
3616 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
3622 if (pfmemalloc
&& !skb_pfmemalloc_protocol(skb
))
3625 if (vlan_tx_tag_present(skb
)) {
3627 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3630 if (vlan_do_receive(&skb
))
3632 else if (unlikely(!skb
))
3636 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
3639 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3642 switch (rx_handler(&skb
)) {
3643 case RX_HANDLER_CONSUMED
:
3644 ret
= NET_RX_SUCCESS
;
3646 case RX_HANDLER_ANOTHER
:
3648 case RX_HANDLER_EXACT
:
3649 deliver_exact
= true;
3650 case RX_HANDLER_PASS
:
3657 if (unlikely(vlan_tx_tag_present(skb
))) {
3658 if (vlan_tx_tag_get_id(skb
))
3659 skb
->pkt_type
= PACKET_OTHERHOST
;
3660 /* Note: we might in the future use prio bits
3661 * and set skb->priority like in vlan_do_receive()
3662 * For the time being, just ignore Priority Code Point
3667 /* deliver only exact match when indicated */
3668 null_or_dev
= deliver_exact
? skb
->dev
: NULL
;
3670 type
= skb
->protocol
;
3671 list_for_each_entry_rcu(ptype
,
3672 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
3673 if (ptype
->type
== type
&&
3674 (ptype
->dev
== null_or_dev
|| ptype
->dev
== skb
->dev
||
3675 ptype
->dev
== orig_dev
)) {
3677 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3683 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
3686 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
3689 atomic_long_inc(&skb
->dev
->rx_dropped
);
3691 /* Jamal, now you will not able to escape explaining
3692 * me how you were going to use this. :-)
3702 static int __netif_receive_skb(struct sk_buff
*skb
)
3706 if (sk_memalloc_socks() && skb_pfmemalloc(skb
)) {
3707 unsigned long pflags
= current
->flags
;
3710 * PFMEMALLOC skbs are special, they should
3711 * - be delivered to SOCK_MEMALLOC sockets only
3712 * - stay away from userspace
3713 * - have bounded memory usage
3715 * Use PF_MEMALLOC as this saves us from propagating the allocation
3716 * context down to all allocation sites.
3718 current
->flags
|= PF_MEMALLOC
;
3719 ret
= __netif_receive_skb_core(skb
, true);
3720 tsk_restore_flags(current
, pflags
, PF_MEMALLOC
);
3722 ret
= __netif_receive_skb_core(skb
, false);
3727 static int netif_receive_skb_internal(struct sk_buff
*skb
)
3729 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3731 if (skb_defer_rx_timestamp(skb
))
3732 return NET_RX_SUCCESS
;
3735 if (static_key_false(&rps_needed
)) {
3736 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3741 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3744 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3751 return __netif_receive_skb(skb
);
3755 * netif_receive_skb - process receive buffer from network
3756 * @skb: buffer to process
3758 * netif_receive_skb() is the main receive data processing function.
3759 * It always succeeds. The buffer may be dropped during processing
3760 * for congestion control or by the protocol layers.
3762 * This function may only be called from softirq context and interrupts
3763 * should be enabled.
3765 * Return values (usually ignored):
3766 * NET_RX_SUCCESS: no congestion
3767 * NET_RX_DROP: packet was dropped
3769 int netif_receive_skb(struct sk_buff
*skb
)
3771 trace_netif_receive_skb_entry(skb
);
3773 return netif_receive_skb_internal(skb
);
3775 EXPORT_SYMBOL(netif_receive_skb
);
3777 /* Network device is going away, flush any packets still pending
3778 * Called with irqs disabled.
3780 static void flush_backlog(void *arg
)
3782 struct net_device
*dev
= arg
;
3783 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3784 struct sk_buff
*skb
, *tmp
;
3787 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3788 if (skb
->dev
== dev
) {
3789 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3791 input_queue_head_incr(sd
);
3796 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3797 if (skb
->dev
== dev
) {
3798 __skb_unlink(skb
, &sd
->process_queue
);
3800 input_queue_head_incr(sd
);
3805 static int napi_gro_complete(struct sk_buff
*skb
)
3807 struct packet_offload
*ptype
;
3808 __be16 type
= skb
->protocol
;
3809 struct list_head
*head
= &offload_base
;
3812 BUILD_BUG_ON(sizeof(struct napi_gro_cb
) > sizeof(skb
->cb
));
3814 if (NAPI_GRO_CB(skb
)->count
== 1) {
3815 skb_shinfo(skb
)->gso_size
= 0;
3820 list_for_each_entry_rcu(ptype
, head
, list
) {
3821 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_complete
)
3824 err
= ptype
->callbacks
.gro_complete(skb
, 0);
3830 WARN_ON(&ptype
->list
== head
);
3832 return NET_RX_SUCCESS
;
3836 return netif_receive_skb_internal(skb
);
3839 /* napi->gro_list contains packets ordered by age.
3840 * youngest packets at the head of it.
3841 * Complete skbs in reverse order to reduce latencies.
3843 void napi_gro_flush(struct napi_struct
*napi
, bool flush_old
)
3845 struct sk_buff
*skb
, *prev
= NULL
;
3847 /* scan list and build reverse chain */
3848 for (skb
= napi
->gro_list
; skb
!= NULL
; skb
= skb
->next
) {
3853 for (skb
= prev
; skb
; skb
= prev
) {
3856 if (flush_old
&& NAPI_GRO_CB(skb
)->age
== jiffies
)
3860 napi_gro_complete(skb
);
3864 napi
->gro_list
= NULL
;
3866 EXPORT_SYMBOL(napi_gro_flush
);
3868 static void gro_list_prepare(struct napi_struct
*napi
, struct sk_buff
*skb
)
3871 unsigned int maclen
= skb
->dev
->hard_header_len
;
3872 u32 hash
= skb_get_hash_raw(skb
);
3874 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3875 unsigned long diffs
;
3877 NAPI_GRO_CB(p
)->flush
= 0;
3879 if (hash
!= skb_get_hash_raw(p
)) {
3880 NAPI_GRO_CB(p
)->same_flow
= 0;
3884 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
3885 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
3886 if (maclen
== ETH_HLEN
)
3887 diffs
|= compare_ether_header(skb_mac_header(p
),
3888 skb_mac_header(skb
));
3890 diffs
= memcmp(skb_mac_header(p
),
3891 skb_mac_header(skb
),
3893 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
3897 static void skb_gro_reset_offset(struct sk_buff
*skb
)
3899 const struct skb_shared_info
*pinfo
= skb_shinfo(skb
);
3900 const skb_frag_t
*frag0
= &pinfo
->frags
[0];
3902 NAPI_GRO_CB(skb
)->data_offset
= 0;
3903 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3904 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3906 if (skb_mac_header(skb
) == skb_tail_pointer(skb
) &&
3908 !PageHighMem(skb_frag_page(frag0
))) {
3909 NAPI_GRO_CB(skb
)->frag0
= skb_frag_address(frag0
);
3910 NAPI_GRO_CB(skb
)->frag0_len
= skb_frag_size(frag0
);
3914 static void gro_pull_from_frag0(struct sk_buff
*skb
, int grow
)
3916 struct skb_shared_info
*pinfo
= skb_shinfo(skb
);
3918 BUG_ON(skb
->end
- skb
->tail
< grow
);
3920 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3922 skb
->data_len
-= grow
;
3925 pinfo
->frags
[0].page_offset
+= grow
;
3926 skb_frag_size_sub(&pinfo
->frags
[0], grow
);
3928 if (unlikely(!skb_frag_size(&pinfo
->frags
[0]))) {
3929 skb_frag_unref(skb
, 0);
3930 memmove(pinfo
->frags
, pinfo
->frags
+ 1,
3931 --pinfo
->nr_frags
* sizeof(pinfo
->frags
[0]));
3935 static enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3937 struct sk_buff
**pp
= NULL
;
3938 struct packet_offload
*ptype
;
3939 __be16 type
= skb
->protocol
;
3940 struct list_head
*head
= &offload_base
;
3942 enum gro_result ret
;
3945 if (!(skb
->dev
->features
& NETIF_F_GRO
))
3948 if (skb_is_gso(skb
) || skb_has_frag_list(skb
))
3951 gro_list_prepare(napi
, skb
);
3952 NAPI_GRO_CB(skb
)->csum
= skb
->csum
; /* Needed for CHECKSUM_COMPLETE */
3955 list_for_each_entry_rcu(ptype
, head
, list
) {
3956 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_receive
)
3959 skb_set_network_header(skb
, skb_gro_offset(skb
));
3960 skb_reset_mac_len(skb
);
3961 NAPI_GRO_CB(skb
)->same_flow
= 0;
3962 NAPI_GRO_CB(skb
)->flush
= 0;
3963 NAPI_GRO_CB(skb
)->free
= 0;
3964 NAPI_GRO_CB(skb
)->udp_mark
= 0;
3966 pp
= ptype
->callbacks
.gro_receive(&napi
->gro_list
, skb
);
3971 if (&ptype
->list
== head
)
3974 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3975 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3978 struct sk_buff
*nskb
= *pp
;
3982 napi_gro_complete(nskb
);
3989 if (NAPI_GRO_CB(skb
)->flush
)
3992 if (unlikely(napi
->gro_count
>= MAX_GRO_SKBS
)) {
3993 struct sk_buff
*nskb
= napi
->gro_list
;
3995 /* locate the end of the list to select the 'oldest' flow */
3996 while (nskb
->next
) {
4002 napi_gro_complete(nskb
);
4006 NAPI_GRO_CB(skb
)->count
= 1;
4007 NAPI_GRO_CB(skb
)->age
= jiffies
;
4008 NAPI_GRO_CB(skb
)->last
= skb
;
4009 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
4010 skb
->next
= napi
->gro_list
;
4011 napi
->gro_list
= skb
;
4015 grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
4017 gro_pull_from_frag0(skb
, grow
);
4026 struct packet_offload
*gro_find_receive_by_type(__be16 type
)
4028 struct list_head
*offload_head
= &offload_base
;
4029 struct packet_offload
*ptype
;
4031 list_for_each_entry_rcu(ptype
, offload_head
, list
) {
4032 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_receive
)
4038 EXPORT_SYMBOL(gro_find_receive_by_type
);
4040 struct packet_offload
*gro_find_complete_by_type(__be16 type
)
4042 struct list_head
*offload_head
= &offload_base
;
4043 struct packet_offload
*ptype
;
4045 list_for_each_entry_rcu(ptype
, offload_head
, list
) {
4046 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_complete
)
4052 EXPORT_SYMBOL(gro_find_complete_by_type
);
4054 static gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
4058 if (netif_receive_skb_internal(skb
))
4066 case GRO_MERGED_FREE
:
4067 if (NAPI_GRO_CB(skb
)->free
== NAPI_GRO_FREE_STOLEN_HEAD
)
4068 kmem_cache_free(skbuff_head_cache
, skb
);
4081 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
4083 trace_napi_gro_receive_entry(skb
);
4085 skb_gro_reset_offset(skb
);
4087 return napi_skb_finish(dev_gro_receive(napi
, skb
), skb
);
4089 EXPORT_SYMBOL(napi_gro_receive
);
4091 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
4093 __skb_pull(skb
, skb_headlen(skb
));
4094 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4095 skb_reserve(skb
, NET_SKB_PAD
+ NET_IP_ALIGN
- skb_headroom(skb
));
4097 skb
->dev
= napi
->dev
;
4099 skb
->truesize
= SKB_TRUESIZE(skb_end_offset(skb
));
4104 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
4106 struct sk_buff
*skb
= napi
->skb
;
4109 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
4114 EXPORT_SYMBOL(napi_get_frags
);
4116 static gro_result_t
napi_frags_finish(struct napi_struct
*napi
,
4117 struct sk_buff
*skb
,
4123 __skb_push(skb
, ETH_HLEN
);
4124 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
4125 if (ret
== GRO_NORMAL
&& netif_receive_skb_internal(skb
))
4130 case GRO_MERGED_FREE
:
4131 napi_reuse_skb(napi
, skb
);
4141 /* Upper GRO stack assumes network header starts at gro_offset=0
4142 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4143 * We copy ethernet header into skb->data to have a common layout.
4145 static struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
4147 struct sk_buff
*skb
= napi
->skb
;
4148 const struct ethhdr
*eth
;
4149 unsigned int hlen
= sizeof(*eth
);
4153 skb_reset_mac_header(skb
);
4154 skb_gro_reset_offset(skb
);
4156 eth
= skb_gro_header_fast(skb
, 0);
4157 if (unlikely(skb_gro_header_hard(skb
, hlen
))) {
4158 eth
= skb_gro_header_slow(skb
, hlen
, 0);
4159 if (unlikely(!eth
)) {
4160 napi_reuse_skb(napi
, skb
);
4164 gro_pull_from_frag0(skb
, hlen
);
4165 NAPI_GRO_CB(skb
)->frag0
+= hlen
;
4166 NAPI_GRO_CB(skb
)->frag0_len
-= hlen
;
4168 __skb_pull(skb
, hlen
);
4171 * This works because the only protocols we care about don't require
4173 * We'll fix it up properly in napi_frags_finish()
4175 skb
->protocol
= eth
->h_proto
;
4180 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
4182 struct sk_buff
*skb
= napi_frags_skb(napi
);
4187 trace_napi_gro_frags_entry(skb
);
4189 return napi_frags_finish(napi
, skb
, dev_gro_receive(napi
, skb
));
4191 EXPORT_SYMBOL(napi_gro_frags
);
4194 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4195 * Note: called with local irq disabled, but exits with local irq enabled.
4197 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
4200 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
4203 sd
->rps_ipi_list
= NULL
;
4207 /* Send pending IPI's to kick RPS processing on remote cpus. */
4209 struct softnet_data
*next
= remsd
->rps_ipi_next
;
4211 if (cpu_online(remsd
->cpu
))
4212 smp_call_function_single_async(remsd
->cpu
,
4221 static int process_backlog(struct napi_struct
*napi
, int quota
)
4224 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
4227 /* Check if we have pending ipi, its better to send them now,
4228 * not waiting net_rx_action() end.
4230 if (sd
->rps_ipi_list
) {
4231 local_irq_disable();
4232 net_rps_action_and_irq_enable(sd
);
4235 napi
->weight
= weight_p
;
4236 local_irq_disable();
4238 struct sk_buff
*skb
;
4240 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
4242 __netif_receive_skb(skb
);
4243 local_irq_disable();
4244 input_queue_head_incr(sd
);
4245 if (++work
>= quota
) {
4252 if (skb_queue_empty(&sd
->input_pkt_queue
)) {
4254 * Inline a custom version of __napi_complete().
4255 * only current cpu owns and manipulates this napi,
4256 * and NAPI_STATE_SCHED is the only possible flag set
4258 * We can use a plain write instead of clear_bit(),
4259 * and we dont need an smp_mb() memory barrier.
4261 list_del(&napi
->poll_list
);
4268 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
4269 &sd
->process_queue
);
4278 * __napi_schedule - schedule for receive
4279 * @n: entry to schedule
4281 * The entry's receive function will be scheduled to run
4283 void __napi_schedule(struct napi_struct
*n
)
4285 unsigned long flags
;
4287 local_irq_save(flags
);
4288 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
4289 local_irq_restore(flags
);
4291 EXPORT_SYMBOL(__napi_schedule
);
4293 void __napi_complete(struct napi_struct
*n
)
4295 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
4296 BUG_ON(n
->gro_list
);
4298 list_del(&n
->poll_list
);
4299 smp_mb__before_atomic();
4300 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
4302 EXPORT_SYMBOL(__napi_complete
);
4304 void napi_complete(struct napi_struct
*n
)
4306 unsigned long flags
;
4309 * don't let napi dequeue from the cpu poll list
4310 * just in case its running on a different cpu
4312 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
4315 napi_gro_flush(n
, false);
4316 local_irq_save(flags
);
4318 local_irq_restore(flags
);
4320 EXPORT_SYMBOL(napi_complete
);
4322 /* must be called under rcu_read_lock(), as we dont take a reference */
4323 struct napi_struct
*napi_by_id(unsigned int napi_id
)
4325 unsigned int hash
= napi_id
% HASH_SIZE(napi_hash
);
4326 struct napi_struct
*napi
;
4328 hlist_for_each_entry_rcu(napi
, &napi_hash
[hash
], napi_hash_node
)
4329 if (napi
->napi_id
== napi_id
)
4334 EXPORT_SYMBOL_GPL(napi_by_id
);
4336 void napi_hash_add(struct napi_struct
*napi
)
4338 if (!test_and_set_bit(NAPI_STATE_HASHED
, &napi
->state
)) {
4340 spin_lock(&napi_hash_lock
);
4342 /* 0 is not a valid id, we also skip an id that is taken
4343 * we expect both events to be extremely rare
4346 while (!napi
->napi_id
) {
4347 napi
->napi_id
= ++napi_gen_id
;
4348 if (napi_by_id(napi
->napi_id
))
4352 hlist_add_head_rcu(&napi
->napi_hash_node
,
4353 &napi_hash
[napi
->napi_id
% HASH_SIZE(napi_hash
)]);
4355 spin_unlock(&napi_hash_lock
);
4358 EXPORT_SYMBOL_GPL(napi_hash_add
);
4360 /* Warning : caller is responsible to make sure rcu grace period
4361 * is respected before freeing memory containing @napi
4363 void napi_hash_del(struct napi_struct
*napi
)
4365 spin_lock(&napi_hash_lock
);
4367 if (test_and_clear_bit(NAPI_STATE_HASHED
, &napi
->state
))
4368 hlist_del_rcu(&napi
->napi_hash_node
);
4370 spin_unlock(&napi_hash_lock
);
4372 EXPORT_SYMBOL_GPL(napi_hash_del
);
4374 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
4375 int (*poll
)(struct napi_struct
*, int), int weight
)
4377 INIT_LIST_HEAD(&napi
->poll_list
);
4378 napi
->gro_count
= 0;
4379 napi
->gro_list
= NULL
;
4382 if (weight
> NAPI_POLL_WEIGHT
)
4383 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4385 napi
->weight
= weight
;
4386 list_add(&napi
->dev_list
, &dev
->napi_list
);
4388 #ifdef CONFIG_NETPOLL
4389 spin_lock_init(&napi
->poll_lock
);
4390 napi
->poll_owner
= -1;
4392 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
4394 EXPORT_SYMBOL(netif_napi_add
);
4396 void netif_napi_del(struct napi_struct
*napi
)
4398 list_del_init(&napi
->dev_list
);
4399 napi_free_frags(napi
);
4401 kfree_skb_list(napi
->gro_list
);
4402 napi
->gro_list
= NULL
;
4403 napi
->gro_count
= 0;
4405 EXPORT_SYMBOL(netif_napi_del
);
4407 static void net_rx_action(struct softirq_action
*h
)
4409 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
4410 unsigned long time_limit
= jiffies
+ 2;
4411 int budget
= netdev_budget
;
4414 local_irq_disable();
4416 while (!list_empty(&sd
->poll_list
)) {
4417 struct napi_struct
*n
;
4420 /* If softirq window is exhuasted then punt.
4421 * Allow this to run for 2 jiffies since which will allow
4422 * an average latency of 1.5/HZ.
4424 if (unlikely(budget
<= 0 || time_after_eq(jiffies
, time_limit
)))
4429 /* Even though interrupts have been re-enabled, this
4430 * access is safe because interrupts can only add new
4431 * entries to the tail of this list, and only ->poll()
4432 * calls can remove this head entry from the list.
4434 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
4436 have
= netpoll_poll_lock(n
);
4440 /* This NAPI_STATE_SCHED test is for avoiding a race
4441 * with netpoll's poll_napi(). Only the entity which
4442 * obtains the lock and sees NAPI_STATE_SCHED set will
4443 * actually make the ->poll() call. Therefore we avoid
4444 * accidentally calling ->poll() when NAPI is not scheduled.
4447 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
4448 work
= n
->poll(n
, weight
);
4452 WARN_ON_ONCE(work
> weight
);
4456 local_irq_disable();
4458 /* Drivers must not modify the NAPI state if they
4459 * consume the entire weight. In such cases this code
4460 * still "owns" the NAPI instance and therefore can
4461 * move the instance around on the list at-will.
4463 if (unlikely(work
== weight
)) {
4464 if (unlikely(napi_disable_pending(n
))) {
4467 local_irq_disable();
4470 /* flush too old packets
4471 * If HZ < 1000, flush all packets.
4474 napi_gro_flush(n
, HZ
>= 1000);
4475 local_irq_disable();
4477 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
4481 netpoll_poll_unlock(have
);
4484 net_rps_action_and_irq_enable(sd
);
4486 #ifdef CONFIG_NET_DMA
4488 * There may not be any more sk_buffs coming right now, so push
4489 * any pending DMA copies to hardware
4491 dma_issue_pending_all();
4498 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
4502 struct netdev_adjacent
{
4503 struct net_device
*dev
;
4505 /* upper master flag, there can only be one master device per list */
4508 /* counter for the number of times this device was added to us */
4511 /* private field for the users */
4514 struct list_head list
;
4515 struct rcu_head rcu
;
4518 static struct netdev_adjacent
*__netdev_find_adj(struct net_device
*dev
,
4519 struct net_device
*adj_dev
,
4520 struct list_head
*adj_list
)
4522 struct netdev_adjacent
*adj
;
4524 list_for_each_entry(adj
, adj_list
, list
) {
4525 if (adj
->dev
== adj_dev
)
4532 * netdev_has_upper_dev - Check if device is linked to an upper device
4534 * @upper_dev: upper device to check
4536 * Find out if a device is linked to specified upper device and return true
4537 * in case it is. Note that this checks only immediate upper device,
4538 * not through a complete stack of devices. The caller must hold the RTNL lock.
4540 bool netdev_has_upper_dev(struct net_device
*dev
,
4541 struct net_device
*upper_dev
)
4545 return __netdev_find_adj(dev
, upper_dev
, &dev
->all_adj_list
.upper
);
4547 EXPORT_SYMBOL(netdev_has_upper_dev
);
4550 * netdev_has_any_upper_dev - Check if device is linked to some device
4553 * Find out if a device is linked to an upper device and return true in case
4554 * it is. The caller must hold the RTNL lock.
4556 static bool netdev_has_any_upper_dev(struct net_device
*dev
)
4560 return !list_empty(&dev
->all_adj_list
.upper
);
4564 * netdev_master_upper_dev_get - Get master upper device
4567 * Find a master upper device and return pointer to it or NULL in case
4568 * it's not there. The caller must hold the RTNL lock.
4570 struct net_device
*netdev_master_upper_dev_get(struct net_device
*dev
)
4572 struct netdev_adjacent
*upper
;
4576 if (list_empty(&dev
->adj_list
.upper
))
4579 upper
= list_first_entry(&dev
->adj_list
.upper
,
4580 struct netdev_adjacent
, list
);
4581 if (likely(upper
->master
))
4585 EXPORT_SYMBOL(netdev_master_upper_dev_get
);
4587 void *netdev_adjacent_get_private(struct list_head
*adj_list
)
4589 struct netdev_adjacent
*adj
;
4591 adj
= list_entry(adj_list
, struct netdev_adjacent
, list
);
4593 return adj
->private;
4595 EXPORT_SYMBOL(netdev_adjacent_get_private
);
4598 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
4600 * @iter: list_head ** of the current position
4602 * Gets the next device from the dev's upper list, starting from iter
4603 * position. The caller must hold RCU read lock.
4605 struct net_device
*netdev_upper_get_next_dev_rcu(struct net_device
*dev
,
4606 struct list_head
**iter
)
4608 struct netdev_adjacent
*upper
;
4610 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4612 upper
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
4614 if (&upper
->list
== &dev
->adj_list
.upper
)
4617 *iter
= &upper
->list
;
4621 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu
);
4624 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4626 * @iter: list_head ** of the current position
4628 * Gets the next device from the dev's upper list, starting from iter
4629 * position. The caller must hold RCU read lock.
4631 struct net_device
*netdev_all_upper_get_next_dev_rcu(struct net_device
*dev
,
4632 struct list_head
**iter
)
4634 struct netdev_adjacent
*upper
;
4636 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4638 upper
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
4640 if (&upper
->list
== &dev
->all_adj_list
.upper
)
4643 *iter
= &upper
->list
;
4647 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu
);
4650 * netdev_lower_get_next_private - Get the next ->private from the
4651 * lower neighbour list
4653 * @iter: list_head ** of the current position
4655 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4656 * list, starting from iter position. The caller must hold either hold the
4657 * RTNL lock or its own locking that guarantees that the neighbour lower
4658 * list will remain unchainged.
4660 void *netdev_lower_get_next_private(struct net_device
*dev
,
4661 struct list_head
**iter
)
4663 struct netdev_adjacent
*lower
;
4665 lower
= list_entry(*iter
, struct netdev_adjacent
, list
);
4667 if (&lower
->list
== &dev
->adj_list
.lower
)
4670 *iter
= lower
->list
.next
;
4672 return lower
->private;
4674 EXPORT_SYMBOL(netdev_lower_get_next_private
);
4677 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4678 * lower neighbour list, RCU
4681 * @iter: list_head ** of the current position
4683 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4684 * list, starting from iter position. The caller must hold RCU read lock.
4686 void *netdev_lower_get_next_private_rcu(struct net_device
*dev
,
4687 struct list_head
**iter
)
4689 struct netdev_adjacent
*lower
;
4691 WARN_ON_ONCE(!rcu_read_lock_held());
4693 lower
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
4695 if (&lower
->list
== &dev
->adj_list
.lower
)
4698 *iter
= &lower
->list
;
4700 return lower
->private;
4702 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu
);
4705 * netdev_lower_get_next - Get the next device from the lower neighbour
4708 * @iter: list_head ** of the current position
4710 * Gets the next netdev_adjacent from the dev's lower neighbour
4711 * list, starting from iter position. The caller must hold RTNL lock or
4712 * its own locking that guarantees that the neighbour lower
4713 * list will remain unchainged.
4715 void *netdev_lower_get_next(struct net_device
*dev
, struct list_head
**iter
)
4717 struct netdev_adjacent
*lower
;
4719 lower
= list_entry((*iter
)->next
, struct netdev_adjacent
, list
);
4721 if (&lower
->list
== &dev
->adj_list
.lower
)
4724 *iter
= &lower
->list
;
4728 EXPORT_SYMBOL(netdev_lower_get_next
);
4731 * netdev_lower_get_first_private_rcu - Get the first ->private from the
4732 * lower neighbour list, RCU
4736 * Gets the first netdev_adjacent->private from the dev's lower neighbour
4737 * list. The caller must hold RCU read lock.
4739 void *netdev_lower_get_first_private_rcu(struct net_device
*dev
)
4741 struct netdev_adjacent
*lower
;
4743 lower
= list_first_or_null_rcu(&dev
->adj_list
.lower
,
4744 struct netdev_adjacent
, list
);
4746 return lower
->private;
4749 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu
);
4752 * netdev_master_upper_dev_get_rcu - Get master upper device
4755 * Find a master upper device and return pointer to it or NULL in case
4756 * it's not there. The caller must hold the RCU read lock.
4758 struct net_device
*netdev_master_upper_dev_get_rcu(struct net_device
*dev
)
4760 struct netdev_adjacent
*upper
;
4762 upper
= list_first_or_null_rcu(&dev
->adj_list
.upper
,
4763 struct netdev_adjacent
, list
);
4764 if (upper
&& likely(upper
->master
))
4768 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu
);
4770 static int netdev_adjacent_sysfs_add(struct net_device
*dev
,
4771 struct net_device
*adj_dev
,
4772 struct list_head
*dev_list
)
4774 char linkname
[IFNAMSIZ
+7];
4775 sprintf(linkname
, dev_list
== &dev
->adj_list
.upper
?
4776 "upper_%s" : "lower_%s", adj_dev
->name
);
4777 return sysfs_create_link(&(dev
->dev
.kobj
), &(adj_dev
->dev
.kobj
),
4780 static void netdev_adjacent_sysfs_del(struct net_device
*dev
,
4782 struct list_head
*dev_list
)
4784 char linkname
[IFNAMSIZ
+7];
4785 sprintf(linkname
, dev_list
== &dev
->adj_list
.upper
?
4786 "upper_%s" : "lower_%s", name
);
4787 sysfs_remove_link(&(dev
->dev
.kobj
), linkname
);
4790 #define netdev_adjacent_is_neigh_list(dev, dev_list) \
4791 (dev_list == &dev->adj_list.upper || \
4792 dev_list == &dev->adj_list.lower)
4794 static int __netdev_adjacent_dev_insert(struct net_device
*dev
,
4795 struct net_device
*adj_dev
,
4796 struct list_head
*dev_list
,
4797 void *private, bool master
)
4799 struct netdev_adjacent
*adj
;
4802 adj
= __netdev_find_adj(dev
, adj_dev
, dev_list
);
4809 adj
= kmalloc(sizeof(*adj
), GFP_KERNEL
);
4814 adj
->master
= master
;
4816 adj
->private = private;
4819 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
4820 adj_dev
->name
, dev
->name
, adj_dev
->name
);
4822 if (netdev_adjacent_is_neigh_list(dev
, dev_list
)) {
4823 ret
= netdev_adjacent_sysfs_add(dev
, adj_dev
, dev_list
);
4828 /* Ensure that master link is always the first item in list. */
4830 ret
= sysfs_create_link(&(dev
->dev
.kobj
),
4831 &(adj_dev
->dev
.kobj
), "master");
4833 goto remove_symlinks
;
4835 list_add_rcu(&adj
->list
, dev_list
);
4837 list_add_tail_rcu(&adj
->list
, dev_list
);
4843 if (netdev_adjacent_is_neigh_list(dev
, dev_list
))
4844 netdev_adjacent_sysfs_del(dev
, adj_dev
->name
, dev_list
);
4852 static void __netdev_adjacent_dev_remove(struct net_device
*dev
,
4853 struct net_device
*adj_dev
,
4854 struct list_head
*dev_list
)
4856 struct netdev_adjacent
*adj
;
4858 adj
= __netdev_find_adj(dev
, adj_dev
, dev_list
);
4861 pr_err("tried to remove device %s from %s\n",
4862 dev
->name
, adj_dev
->name
);
4866 if (adj
->ref_nr
> 1) {
4867 pr_debug("%s to %s ref_nr-- = %d\n", dev
->name
, adj_dev
->name
,
4874 sysfs_remove_link(&(dev
->dev
.kobj
), "master");
4876 if (netdev_adjacent_is_neigh_list(dev
, dev_list
))
4877 netdev_adjacent_sysfs_del(dev
, adj_dev
->name
, dev_list
);
4879 list_del_rcu(&adj
->list
);
4880 pr_debug("dev_put for %s, because link removed from %s to %s\n",
4881 adj_dev
->name
, dev
->name
, adj_dev
->name
);
4883 kfree_rcu(adj
, rcu
);
4886 static int __netdev_adjacent_dev_link_lists(struct net_device
*dev
,
4887 struct net_device
*upper_dev
,
4888 struct list_head
*up_list
,
4889 struct list_head
*down_list
,
4890 void *private, bool master
)
4894 ret
= __netdev_adjacent_dev_insert(dev
, upper_dev
, up_list
, private,
4899 ret
= __netdev_adjacent_dev_insert(upper_dev
, dev
, down_list
, private,
4902 __netdev_adjacent_dev_remove(dev
, upper_dev
, up_list
);
4909 static int __netdev_adjacent_dev_link(struct net_device
*dev
,
4910 struct net_device
*upper_dev
)
4912 return __netdev_adjacent_dev_link_lists(dev
, upper_dev
,
4913 &dev
->all_adj_list
.upper
,
4914 &upper_dev
->all_adj_list
.lower
,
4918 static void __netdev_adjacent_dev_unlink_lists(struct net_device
*dev
,
4919 struct net_device
*upper_dev
,
4920 struct list_head
*up_list
,
4921 struct list_head
*down_list
)
4923 __netdev_adjacent_dev_remove(dev
, upper_dev
, up_list
);
4924 __netdev_adjacent_dev_remove(upper_dev
, dev
, down_list
);
4927 static void __netdev_adjacent_dev_unlink(struct net_device
*dev
,
4928 struct net_device
*upper_dev
)
4930 __netdev_adjacent_dev_unlink_lists(dev
, upper_dev
,
4931 &dev
->all_adj_list
.upper
,
4932 &upper_dev
->all_adj_list
.lower
);
4935 static int __netdev_adjacent_dev_link_neighbour(struct net_device
*dev
,
4936 struct net_device
*upper_dev
,
4937 void *private, bool master
)
4939 int ret
= __netdev_adjacent_dev_link(dev
, upper_dev
);
4944 ret
= __netdev_adjacent_dev_link_lists(dev
, upper_dev
,
4945 &dev
->adj_list
.upper
,
4946 &upper_dev
->adj_list
.lower
,
4949 __netdev_adjacent_dev_unlink(dev
, upper_dev
);
4956 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device
*dev
,
4957 struct net_device
*upper_dev
)
4959 __netdev_adjacent_dev_unlink(dev
, upper_dev
);
4960 __netdev_adjacent_dev_unlink_lists(dev
, upper_dev
,
4961 &dev
->adj_list
.upper
,
4962 &upper_dev
->adj_list
.lower
);
4965 static int __netdev_upper_dev_link(struct net_device
*dev
,
4966 struct net_device
*upper_dev
, bool master
,
4969 struct netdev_adjacent
*i
, *j
, *to_i
, *to_j
;
4974 if (dev
== upper_dev
)
4977 /* To prevent loops, check if dev is not upper device to upper_dev. */
4978 if (__netdev_find_adj(upper_dev
, dev
, &upper_dev
->all_adj_list
.upper
))
4981 if (__netdev_find_adj(dev
, upper_dev
, &dev
->all_adj_list
.upper
))
4984 if (master
&& netdev_master_upper_dev_get(dev
))
4987 ret
= __netdev_adjacent_dev_link_neighbour(dev
, upper_dev
, private,
4992 /* Now that we linked these devs, make all the upper_dev's
4993 * all_adj_list.upper visible to every dev's all_adj_list.lower an
4994 * versa, and don't forget the devices itself. All of these
4995 * links are non-neighbours.
4997 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
4998 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
) {
4999 pr_debug("Interlinking %s with %s, non-neighbour\n",
5000 i
->dev
->name
, j
->dev
->name
);
5001 ret
= __netdev_adjacent_dev_link(i
->dev
, j
->dev
);
5007 /* add dev to every upper_dev's upper device */
5008 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
) {
5009 pr_debug("linking %s's upper device %s with %s\n",
5010 upper_dev
->name
, i
->dev
->name
, dev
->name
);
5011 ret
= __netdev_adjacent_dev_link(dev
, i
->dev
);
5013 goto rollback_upper_mesh
;
5016 /* add upper_dev to every dev's lower device */
5017 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5018 pr_debug("linking %s's lower device %s with %s\n", dev
->name
,
5019 i
->dev
->name
, upper_dev
->name
);
5020 ret
= __netdev_adjacent_dev_link(i
->dev
, upper_dev
);
5022 goto rollback_lower_mesh
;
5025 call_netdevice_notifiers(NETDEV_CHANGEUPPER
, dev
);
5028 rollback_lower_mesh
:
5030 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5033 __netdev_adjacent_dev_unlink(i
->dev
, upper_dev
);
5038 rollback_upper_mesh
:
5040 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
) {
5043 __netdev_adjacent_dev_unlink(dev
, i
->dev
);
5051 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5052 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
) {
5053 if (i
== to_i
&& j
== to_j
)
5055 __netdev_adjacent_dev_unlink(i
->dev
, j
->dev
);
5061 __netdev_adjacent_dev_unlink_neighbour(dev
, upper_dev
);
5067 * netdev_upper_dev_link - Add a link to the upper device
5069 * @upper_dev: new upper device
5071 * Adds a link to device which is upper to this one. The caller must hold
5072 * the RTNL lock. On a failure a negative errno code is returned.
5073 * On success the reference counts are adjusted and the function
5076 int netdev_upper_dev_link(struct net_device
*dev
,
5077 struct net_device
*upper_dev
)
5079 return __netdev_upper_dev_link(dev
, upper_dev
, false, NULL
);
5081 EXPORT_SYMBOL(netdev_upper_dev_link
);
5084 * netdev_master_upper_dev_link - Add a master link to the upper device
5086 * @upper_dev: new upper device
5088 * Adds a link to device which is upper to this one. In this case, only
5089 * one master upper device can be linked, although other non-master devices
5090 * might be linked as well. The caller must hold the RTNL lock.
5091 * On a failure a negative errno code is returned. On success the reference
5092 * counts are adjusted and the function returns zero.
5094 int netdev_master_upper_dev_link(struct net_device
*dev
,
5095 struct net_device
*upper_dev
)
5097 return __netdev_upper_dev_link(dev
, upper_dev
, true, NULL
);
5099 EXPORT_SYMBOL(netdev_master_upper_dev_link
);
5101 int netdev_master_upper_dev_link_private(struct net_device
*dev
,
5102 struct net_device
*upper_dev
,
5105 return __netdev_upper_dev_link(dev
, upper_dev
, true, private);
5107 EXPORT_SYMBOL(netdev_master_upper_dev_link_private
);
5110 * netdev_upper_dev_unlink - Removes a link to upper device
5112 * @upper_dev: new upper device
5114 * Removes a link to device which is upper to this one. The caller must hold
5117 void netdev_upper_dev_unlink(struct net_device
*dev
,
5118 struct net_device
*upper_dev
)
5120 struct netdev_adjacent
*i
, *j
;
5123 __netdev_adjacent_dev_unlink_neighbour(dev
, upper_dev
);
5125 /* Here is the tricky part. We must remove all dev's lower
5126 * devices from all upper_dev's upper devices and vice
5127 * versa, to maintain the graph relationship.
5129 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
)
5130 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
)
5131 __netdev_adjacent_dev_unlink(i
->dev
, j
->dev
);
5133 /* remove also the devices itself from lower/upper device
5136 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
)
5137 __netdev_adjacent_dev_unlink(i
->dev
, upper_dev
);
5139 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
)
5140 __netdev_adjacent_dev_unlink(dev
, i
->dev
);
5142 call_netdevice_notifiers(NETDEV_CHANGEUPPER
, dev
);
5144 EXPORT_SYMBOL(netdev_upper_dev_unlink
);
5146 void netdev_adjacent_rename_links(struct net_device
*dev
, char *oldname
)
5148 struct netdev_adjacent
*iter
;
5150 list_for_each_entry(iter
, &dev
->adj_list
.upper
, list
) {
5151 netdev_adjacent_sysfs_del(iter
->dev
, oldname
,
5152 &iter
->dev
->adj_list
.lower
);
5153 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
5154 &iter
->dev
->adj_list
.lower
);
5157 list_for_each_entry(iter
, &dev
->adj_list
.lower
, list
) {
5158 netdev_adjacent_sysfs_del(iter
->dev
, oldname
,
5159 &iter
->dev
->adj_list
.upper
);
5160 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
5161 &iter
->dev
->adj_list
.upper
);
5165 void *netdev_lower_dev_get_private(struct net_device
*dev
,
5166 struct net_device
*lower_dev
)
5168 struct netdev_adjacent
*lower
;
5172 lower
= __netdev_find_adj(dev
, lower_dev
, &dev
->adj_list
.lower
);
5176 return lower
->private;
5178 EXPORT_SYMBOL(netdev_lower_dev_get_private
);
5181 int dev_get_nest_level(struct net_device
*dev
,
5182 bool (*type_check
)(struct net_device
*dev
))
5184 struct net_device
*lower
= NULL
;
5185 struct list_head
*iter
;
5191 netdev_for_each_lower_dev(dev
, lower
, iter
) {
5192 nest
= dev_get_nest_level(lower
, type_check
);
5193 if (max_nest
< nest
)
5197 if (type_check(dev
))
5202 EXPORT_SYMBOL(dev_get_nest_level
);
5204 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
5206 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5208 if (ops
->ndo_change_rx_flags
)
5209 ops
->ndo_change_rx_flags(dev
, flags
);
5212 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
, bool notify
)
5214 unsigned int old_flags
= dev
->flags
;
5220 dev
->flags
|= IFF_PROMISC
;
5221 dev
->promiscuity
+= inc
;
5222 if (dev
->promiscuity
== 0) {
5225 * If inc causes overflow, untouch promisc and return error.
5228 dev
->flags
&= ~IFF_PROMISC
;
5230 dev
->promiscuity
-= inc
;
5231 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5236 if (dev
->flags
!= old_flags
) {
5237 pr_info("device %s %s promiscuous mode\n",
5239 dev
->flags
& IFF_PROMISC
? "entered" : "left");
5240 if (audit_enabled
) {
5241 current_uid_gid(&uid
, &gid
);
5242 audit_log(current
->audit_context
, GFP_ATOMIC
,
5243 AUDIT_ANOM_PROMISCUOUS
,
5244 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5245 dev
->name
, (dev
->flags
& IFF_PROMISC
),
5246 (old_flags
& IFF_PROMISC
),
5247 from_kuid(&init_user_ns
, audit_get_loginuid(current
)),
5248 from_kuid(&init_user_ns
, uid
),
5249 from_kgid(&init_user_ns
, gid
),
5250 audit_get_sessionid(current
));
5253 dev_change_rx_flags(dev
, IFF_PROMISC
);
5256 __dev_notify_flags(dev
, old_flags
, IFF_PROMISC
);
5261 * dev_set_promiscuity - update promiscuity count on a device
5265 * Add or remove promiscuity from a device. While the count in the device
5266 * remains above zero the interface remains promiscuous. Once it hits zero
5267 * the device reverts back to normal filtering operation. A negative inc
5268 * value is used to drop promiscuity on the device.
5269 * Return 0 if successful or a negative errno code on error.
5271 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
5273 unsigned int old_flags
= dev
->flags
;
5276 err
= __dev_set_promiscuity(dev
, inc
, true);
5279 if (dev
->flags
!= old_flags
)
5280 dev_set_rx_mode(dev
);
5283 EXPORT_SYMBOL(dev_set_promiscuity
);
5285 static int __dev_set_allmulti(struct net_device
*dev
, int inc
, bool notify
)
5287 unsigned int old_flags
= dev
->flags
, old_gflags
= dev
->gflags
;
5291 dev
->flags
|= IFF_ALLMULTI
;
5292 dev
->allmulti
+= inc
;
5293 if (dev
->allmulti
== 0) {
5296 * If inc causes overflow, untouch allmulti and return error.
5299 dev
->flags
&= ~IFF_ALLMULTI
;
5301 dev
->allmulti
-= inc
;
5302 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5307 if (dev
->flags
^ old_flags
) {
5308 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
5309 dev_set_rx_mode(dev
);
5311 __dev_notify_flags(dev
, old_flags
,
5312 dev
->gflags
^ old_gflags
);
5318 * dev_set_allmulti - update allmulti count on a device
5322 * Add or remove reception of all multicast frames to a device. While the
5323 * count in the device remains above zero the interface remains listening
5324 * to all interfaces. Once it hits zero the device reverts back to normal
5325 * filtering operation. A negative @inc value is used to drop the counter
5326 * when releasing a resource needing all multicasts.
5327 * Return 0 if successful or a negative errno code on error.
5330 int dev_set_allmulti(struct net_device
*dev
, int inc
)
5332 return __dev_set_allmulti(dev
, inc
, true);
5334 EXPORT_SYMBOL(dev_set_allmulti
);
5337 * Upload unicast and multicast address lists to device and
5338 * configure RX filtering. When the device doesn't support unicast
5339 * filtering it is put in promiscuous mode while unicast addresses
5342 void __dev_set_rx_mode(struct net_device
*dev
)
5344 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5346 /* dev_open will call this function so the list will stay sane. */
5347 if (!(dev
->flags
&IFF_UP
))
5350 if (!netif_device_present(dev
))
5353 if (!(dev
->priv_flags
& IFF_UNICAST_FLT
)) {
5354 /* Unicast addresses changes may only happen under the rtnl,
5355 * therefore calling __dev_set_promiscuity here is safe.
5357 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
5358 __dev_set_promiscuity(dev
, 1, false);
5359 dev
->uc_promisc
= true;
5360 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
5361 __dev_set_promiscuity(dev
, -1, false);
5362 dev
->uc_promisc
= false;
5366 if (ops
->ndo_set_rx_mode
)
5367 ops
->ndo_set_rx_mode(dev
);
5370 void dev_set_rx_mode(struct net_device
*dev
)
5372 netif_addr_lock_bh(dev
);
5373 __dev_set_rx_mode(dev
);
5374 netif_addr_unlock_bh(dev
);
5378 * dev_get_flags - get flags reported to userspace
5381 * Get the combination of flag bits exported through APIs to userspace.
5383 unsigned int dev_get_flags(const struct net_device
*dev
)
5387 flags
= (dev
->flags
& ~(IFF_PROMISC
|
5392 (dev
->gflags
& (IFF_PROMISC
|
5395 if (netif_running(dev
)) {
5396 if (netif_oper_up(dev
))
5397 flags
|= IFF_RUNNING
;
5398 if (netif_carrier_ok(dev
))
5399 flags
|= IFF_LOWER_UP
;
5400 if (netif_dormant(dev
))
5401 flags
|= IFF_DORMANT
;
5406 EXPORT_SYMBOL(dev_get_flags
);
5408 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
5410 unsigned int old_flags
= dev
->flags
;
5416 * Set the flags on our device.
5419 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
5420 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
5422 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
5426 * Load in the correct multicast list now the flags have changed.
5429 if ((old_flags
^ flags
) & IFF_MULTICAST
)
5430 dev_change_rx_flags(dev
, IFF_MULTICAST
);
5432 dev_set_rx_mode(dev
);
5435 * Have we downed the interface. We handle IFF_UP ourselves
5436 * according to user attempts to set it, rather than blindly
5441 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
5442 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
5445 dev_set_rx_mode(dev
);
5448 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
5449 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
5450 unsigned int old_flags
= dev
->flags
;
5452 dev
->gflags
^= IFF_PROMISC
;
5454 if (__dev_set_promiscuity(dev
, inc
, false) >= 0)
5455 if (dev
->flags
!= old_flags
)
5456 dev_set_rx_mode(dev
);
5459 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5460 is important. Some (broken) drivers set IFF_PROMISC, when
5461 IFF_ALLMULTI is requested not asking us and not reporting.
5463 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
5464 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
5466 dev
->gflags
^= IFF_ALLMULTI
;
5467 __dev_set_allmulti(dev
, inc
, false);
5473 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
,
5474 unsigned int gchanges
)
5476 unsigned int changes
= dev
->flags
^ old_flags
;
5479 rtmsg_ifinfo(RTM_NEWLINK
, dev
, gchanges
, GFP_ATOMIC
);
5481 if (changes
& IFF_UP
) {
5482 if (dev
->flags
& IFF_UP
)
5483 call_netdevice_notifiers(NETDEV_UP
, dev
);
5485 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
5488 if (dev
->flags
& IFF_UP
&&
5489 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
))) {
5490 struct netdev_notifier_change_info change_info
;
5492 change_info
.flags_changed
= changes
;
5493 call_netdevice_notifiers_info(NETDEV_CHANGE
, dev
,
5499 * dev_change_flags - change device settings
5501 * @flags: device state flags
5503 * Change settings on device based state flags. The flags are
5504 * in the userspace exported format.
5506 int dev_change_flags(struct net_device
*dev
, unsigned int flags
)
5509 unsigned int changes
, old_flags
= dev
->flags
, old_gflags
= dev
->gflags
;
5511 ret
= __dev_change_flags(dev
, flags
);
5515 changes
= (old_flags
^ dev
->flags
) | (old_gflags
^ dev
->gflags
);
5516 __dev_notify_flags(dev
, old_flags
, changes
);
5519 EXPORT_SYMBOL(dev_change_flags
);
5521 static int __dev_set_mtu(struct net_device
*dev
, int new_mtu
)
5523 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5525 if (ops
->ndo_change_mtu
)
5526 return ops
->ndo_change_mtu(dev
, new_mtu
);
5533 * dev_set_mtu - Change maximum transfer unit
5535 * @new_mtu: new transfer unit
5537 * Change the maximum transfer size of the network device.
5539 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
5543 if (new_mtu
== dev
->mtu
)
5546 /* MTU must be positive. */
5550 if (!netif_device_present(dev
))
5553 err
= call_netdevice_notifiers(NETDEV_PRECHANGEMTU
, dev
);
5554 err
= notifier_to_errno(err
);
5558 orig_mtu
= dev
->mtu
;
5559 err
= __dev_set_mtu(dev
, new_mtu
);
5562 err
= call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
5563 err
= notifier_to_errno(err
);
5565 /* setting mtu back and notifying everyone again,
5566 * so that they have a chance to revert changes.
5568 __dev_set_mtu(dev
, orig_mtu
);
5569 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
5574 EXPORT_SYMBOL(dev_set_mtu
);
5577 * dev_set_group - Change group this device belongs to
5579 * @new_group: group this device should belong to
5581 void dev_set_group(struct net_device
*dev
, int new_group
)
5583 dev
->group
= new_group
;
5585 EXPORT_SYMBOL(dev_set_group
);
5588 * dev_set_mac_address - Change Media Access Control Address
5592 * Change the hardware (MAC) address of the device
5594 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
5596 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5599 if (!ops
->ndo_set_mac_address
)
5601 if (sa
->sa_family
!= dev
->type
)
5603 if (!netif_device_present(dev
))
5605 err
= ops
->ndo_set_mac_address(dev
, sa
);
5608 dev
->addr_assign_type
= NET_ADDR_SET
;
5609 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
5610 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
5613 EXPORT_SYMBOL(dev_set_mac_address
);
5616 * dev_change_carrier - Change device carrier
5618 * @new_carrier: new value
5620 * Change device carrier
5622 int dev_change_carrier(struct net_device
*dev
, bool new_carrier
)
5624 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5626 if (!ops
->ndo_change_carrier
)
5628 if (!netif_device_present(dev
))
5630 return ops
->ndo_change_carrier(dev
, new_carrier
);
5632 EXPORT_SYMBOL(dev_change_carrier
);
5635 * dev_get_phys_port_id - Get device physical port ID
5639 * Get device physical port ID
5641 int dev_get_phys_port_id(struct net_device
*dev
,
5642 struct netdev_phys_port_id
*ppid
)
5644 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5646 if (!ops
->ndo_get_phys_port_id
)
5648 return ops
->ndo_get_phys_port_id(dev
, ppid
);
5650 EXPORT_SYMBOL(dev_get_phys_port_id
);
5653 * dev_new_index - allocate an ifindex
5654 * @net: the applicable net namespace
5656 * Returns a suitable unique value for a new device interface
5657 * number. The caller must hold the rtnl semaphore or the
5658 * dev_base_lock to be sure it remains unique.
5660 static int dev_new_index(struct net
*net
)
5662 int ifindex
= net
->ifindex
;
5666 if (!__dev_get_by_index(net
, ifindex
))
5667 return net
->ifindex
= ifindex
;
5671 /* Delayed registration/unregisteration */
5672 static LIST_HEAD(net_todo_list
);
5673 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq
);
5675 static void net_set_todo(struct net_device
*dev
)
5677 list_add_tail(&dev
->todo_list
, &net_todo_list
);
5678 dev_net(dev
)->dev_unreg_count
++;
5681 static void rollback_registered_many(struct list_head
*head
)
5683 struct net_device
*dev
, *tmp
;
5684 LIST_HEAD(close_head
);
5686 BUG_ON(dev_boot_phase
);
5689 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
5690 /* Some devices call without registering
5691 * for initialization unwind. Remove those
5692 * devices and proceed with the remaining.
5694 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5695 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5699 list_del(&dev
->unreg_list
);
5702 dev
->dismantle
= true;
5703 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
5706 /* If device is running, close it first. */
5707 list_for_each_entry(dev
, head
, unreg_list
)
5708 list_add_tail(&dev
->close_list
, &close_head
);
5709 dev_close_many(&close_head
);
5711 list_for_each_entry(dev
, head
, unreg_list
) {
5712 /* And unlink it from device chain. */
5713 unlist_netdevice(dev
);
5715 dev
->reg_state
= NETREG_UNREGISTERING
;
5720 list_for_each_entry(dev
, head
, unreg_list
) {
5721 /* Shutdown queueing discipline. */
5725 /* Notify protocols, that we are about to destroy
5726 this device. They should clean all the things.
5728 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5731 * Flush the unicast and multicast chains
5736 if (dev
->netdev_ops
->ndo_uninit
)
5737 dev
->netdev_ops
->ndo_uninit(dev
);
5739 if (!dev
->rtnl_link_ops
||
5740 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5741 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U, GFP_KERNEL
);
5743 /* Notifier chain MUST detach us all upper devices. */
5744 WARN_ON(netdev_has_any_upper_dev(dev
));
5746 /* Remove entries from kobject tree */
5747 netdev_unregister_kobject(dev
);
5749 /* Remove XPS queueing entries */
5750 netif_reset_xps_queues_gt(dev
, 0);
5756 list_for_each_entry(dev
, head
, unreg_list
)
5760 static void rollback_registered(struct net_device
*dev
)
5764 list_add(&dev
->unreg_list
, &single
);
5765 rollback_registered_many(&single
);
5769 static netdev_features_t
netdev_fix_features(struct net_device
*dev
,
5770 netdev_features_t features
)
5772 /* Fix illegal checksum combinations */
5773 if ((features
& NETIF_F_HW_CSUM
) &&
5774 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5775 netdev_warn(dev
, "mixed HW and IP checksum settings.\n");
5776 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5779 /* TSO requires that SG is present as well. */
5780 if ((features
& NETIF_F_ALL_TSO
) && !(features
& NETIF_F_SG
)) {
5781 netdev_dbg(dev
, "Dropping TSO features since no SG feature.\n");
5782 features
&= ~NETIF_F_ALL_TSO
;
5785 if ((features
& NETIF_F_TSO
) && !(features
& NETIF_F_HW_CSUM
) &&
5786 !(features
& NETIF_F_IP_CSUM
)) {
5787 netdev_dbg(dev
, "Dropping TSO features since no CSUM feature.\n");
5788 features
&= ~NETIF_F_TSO
;
5789 features
&= ~NETIF_F_TSO_ECN
;
5792 if ((features
& NETIF_F_TSO6
) && !(features
& NETIF_F_HW_CSUM
) &&
5793 !(features
& NETIF_F_IPV6_CSUM
)) {
5794 netdev_dbg(dev
, "Dropping TSO6 features since no CSUM feature.\n");
5795 features
&= ~NETIF_F_TSO6
;
5798 /* TSO ECN requires that TSO is present as well. */
5799 if ((features
& NETIF_F_ALL_TSO
) == NETIF_F_TSO_ECN
)
5800 features
&= ~NETIF_F_TSO_ECN
;
5802 /* Software GSO depends on SG. */
5803 if ((features
& NETIF_F_GSO
) && !(features
& NETIF_F_SG
)) {
5804 netdev_dbg(dev
, "Dropping NETIF_F_GSO since no SG feature.\n");
5805 features
&= ~NETIF_F_GSO
;
5808 /* UFO needs SG and checksumming */
5809 if (features
& NETIF_F_UFO
) {
5810 /* maybe split UFO into V4 and V6? */
5811 if (!((features
& NETIF_F_GEN_CSUM
) ||
5812 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))
5813 == (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5815 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5816 features
&= ~NETIF_F_UFO
;
5819 if (!(features
& NETIF_F_SG
)) {
5821 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5822 features
&= ~NETIF_F_UFO
;
5826 #ifdef CONFIG_NET_RX_BUSY_POLL
5827 if (dev
->netdev_ops
->ndo_busy_poll
)
5828 features
|= NETIF_F_BUSY_POLL
;
5831 features
&= ~NETIF_F_BUSY_POLL
;
5836 int __netdev_update_features(struct net_device
*dev
)
5838 netdev_features_t features
;
5843 features
= netdev_get_wanted_features(dev
);
5845 if (dev
->netdev_ops
->ndo_fix_features
)
5846 features
= dev
->netdev_ops
->ndo_fix_features(dev
, features
);
5848 /* driver might be less strict about feature dependencies */
5849 features
= netdev_fix_features(dev
, features
);
5851 if (dev
->features
== features
)
5854 netdev_dbg(dev
, "Features changed: %pNF -> %pNF\n",
5855 &dev
->features
, &features
);
5857 if (dev
->netdev_ops
->ndo_set_features
)
5858 err
= dev
->netdev_ops
->ndo_set_features(dev
, features
);
5860 if (unlikely(err
< 0)) {
5862 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5863 err
, &features
, &dev
->features
);
5868 dev
->features
= features
;
5874 * netdev_update_features - recalculate device features
5875 * @dev: the device to check
5877 * Recalculate dev->features set and send notifications if it
5878 * has changed. Should be called after driver or hardware dependent
5879 * conditions might have changed that influence the features.
5881 void netdev_update_features(struct net_device
*dev
)
5883 if (__netdev_update_features(dev
))
5884 netdev_features_change(dev
);
5886 EXPORT_SYMBOL(netdev_update_features
);
5889 * netdev_change_features - recalculate device features
5890 * @dev: the device to check
5892 * Recalculate dev->features set and send notifications even
5893 * if they have not changed. Should be called instead of
5894 * netdev_update_features() if also dev->vlan_features might
5895 * have changed to allow the changes to be propagated to stacked
5898 void netdev_change_features(struct net_device
*dev
)
5900 __netdev_update_features(dev
);
5901 netdev_features_change(dev
);
5903 EXPORT_SYMBOL(netdev_change_features
);
5906 * netif_stacked_transfer_operstate - transfer operstate
5907 * @rootdev: the root or lower level device to transfer state from
5908 * @dev: the device to transfer operstate to
5910 * Transfer operational state from root to device. This is normally
5911 * called when a stacking relationship exists between the root
5912 * device and the device(a leaf device).
5914 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
5915 struct net_device
*dev
)
5917 if (rootdev
->operstate
== IF_OPER_DORMANT
)
5918 netif_dormant_on(dev
);
5920 netif_dormant_off(dev
);
5922 if (netif_carrier_ok(rootdev
)) {
5923 if (!netif_carrier_ok(dev
))
5924 netif_carrier_on(dev
);
5926 if (netif_carrier_ok(dev
))
5927 netif_carrier_off(dev
);
5930 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
5933 static int netif_alloc_rx_queues(struct net_device
*dev
)
5935 unsigned int i
, count
= dev
->num_rx_queues
;
5936 struct netdev_rx_queue
*rx
;
5940 rx
= kcalloc(count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5946 for (i
= 0; i
< count
; i
++)
5952 static void netdev_init_one_queue(struct net_device
*dev
,
5953 struct netdev_queue
*queue
, void *_unused
)
5955 /* Initialize queue lock */
5956 spin_lock_init(&queue
->_xmit_lock
);
5957 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
5958 queue
->xmit_lock_owner
= -1;
5959 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
5962 dql_init(&queue
->dql
, HZ
);
5966 static void netif_free_tx_queues(struct net_device
*dev
)
5971 static int netif_alloc_netdev_queues(struct net_device
*dev
)
5973 unsigned int count
= dev
->num_tx_queues
;
5974 struct netdev_queue
*tx
;
5975 size_t sz
= count
* sizeof(*tx
);
5977 BUG_ON(count
< 1 || count
> 0xffff);
5979 tx
= kzalloc(sz
, GFP_KERNEL
| __GFP_NOWARN
| __GFP_REPEAT
);
5987 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5988 spin_lock_init(&dev
->tx_global_lock
);
5994 * register_netdevice - register a network device
5995 * @dev: device to register
5997 * Take a completed network device structure and add it to the kernel
5998 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5999 * chain. 0 is returned on success. A negative errno code is returned
6000 * on a failure to set up the device, or if the name is a duplicate.
6002 * Callers must hold the rtnl semaphore. You may want
6003 * register_netdev() instead of this.
6006 * The locking appears insufficient to guarantee two parallel registers
6007 * will not get the same name.
6010 int register_netdevice(struct net_device
*dev
)
6013 struct net
*net
= dev_net(dev
);
6015 BUG_ON(dev_boot_phase
);
6020 /* When net_device's are persistent, this will be fatal. */
6021 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
6024 spin_lock_init(&dev
->addr_list_lock
);
6025 netdev_set_addr_lockdep_class(dev
);
6029 ret
= dev_get_valid_name(net
, dev
, dev
->name
);
6033 /* Init, if this function is available */
6034 if (dev
->netdev_ops
->ndo_init
) {
6035 ret
= dev
->netdev_ops
->ndo_init(dev
);
6043 if (((dev
->hw_features
| dev
->features
) &
6044 NETIF_F_HW_VLAN_CTAG_FILTER
) &&
6045 (!dev
->netdev_ops
->ndo_vlan_rx_add_vid
||
6046 !dev
->netdev_ops
->ndo_vlan_rx_kill_vid
)) {
6047 netdev_WARN(dev
, "Buggy VLAN acceleration in driver!\n");
6054 dev
->ifindex
= dev_new_index(net
);
6055 else if (__dev_get_by_index(net
, dev
->ifindex
))
6058 if (dev
->iflink
== -1)
6059 dev
->iflink
= dev
->ifindex
;
6061 /* Transfer changeable features to wanted_features and enable
6062 * software offloads (GSO and GRO).
6064 dev
->hw_features
|= NETIF_F_SOFT_FEATURES
;
6065 dev
->features
|= NETIF_F_SOFT_FEATURES
;
6066 dev
->wanted_features
= dev
->features
& dev
->hw_features
;
6068 if (!(dev
->flags
& IFF_LOOPBACK
)) {
6069 dev
->hw_features
|= NETIF_F_NOCACHE_COPY
;
6072 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6074 dev
->vlan_features
|= NETIF_F_HIGHDMA
;
6076 /* Make NETIF_F_SG inheritable to tunnel devices.
6078 dev
->hw_enc_features
|= NETIF_F_SG
;
6080 /* Make NETIF_F_SG inheritable to MPLS.
6082 dev
->mpls_features
|= NETIF_F_SG
;
6084 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
6085 ret
= notifier_to_errno(ret
);
6089 ret
= netdev_register_kobject(dev
);
6092 dev
->reg_state
= NETREG_REGISTERED
;
6094 __netdev_update_features(dev
);
6097 * Default initial state at registry is that the
6098 * device is present.
6101 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
6103 linkwatch_init_dev(dev
);
6105 dev_init_scheduler(dev
);
6107 list_netdevice(dev
);
6108 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
6110 /* If the device has permanent device address, driver should
6111 * set dev_addr and also addr_assign_type should be set to
6112 * NET_ADDR_PERM (default value).
6114 if (dev
->addr_assign_type
== NET_ADDR_PERM
)
6115 memcpy(dev
->perm_addr
, dev
->dev_addr
, dev
->addr_len
);
6117 /* Notify protocols, that a new device appeared. */
6118 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
6119 ret
= notifier_to_errno(ret
);
6121 rollback_registered(dev
);
6122 dev
->reg_state
= NETREG_UNREGISTERED
;
6125 * Prevent userspace races by waiting until the network
6126 * device is fully setup before sending notifications.
6128 if (!dev
->rtnl_link_ops
||
6129 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
6130 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U, GFP_KERNEL
);
6136 if (dev
->netdev_ops
->ndo_uninit
)
6137 dev
->netdev_ops
->ndo_uninit(dev
);
6140 EXPORT_SYMBOL(register_netdevice
);
6143 * init_dummy_netdev - init a dummy network device for NAPI
6144 * @dev: device to init
6146 * This takes a network device structure and initialize the minimum
6147 * amount of fields so it can be used to schedule NAPI polls without
6148 * registering a full blown interface. This is to be used by drivers
6149 * that need to tie several hardware interfaces to a single NAPI
6150 * poll scheduler due to HW limitations.
6152 int init_dummy_netdev(struct net_device
*dev
)
6154 /* Clear everything. Note we don't initialize spinlocks
6155 * are they aren't supposed to be taken by any of the
6156 * NAPI code and this dummy netdev is supposed to be
6157 * only ever used for NAPI polls
6159 memset(dev
, 0, sizeof(struct net_device
));
6161 /* make sure we BUG if trying to hit standard
6162 * register/unregister code path
6164 dev
->reg_state
= NETREG_DUMMY
;
6166 /* NAPI wants this */
6167 INIT_LIST_HEAD(&dev
->napi_list
);
6169 /* a dummy interface is started by default */
6170 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
6171 set_bit(__LINK_STATE_START
, &dev
->state
);
6173 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6174 * because users of this 'device' dont need to change
6180 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
6184 * register_netdev - register a network device
6185 * @dev: device to register
6187 * Take a completed network device structure and add it to the kernel
6188 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6189 * chain. 0 is returned on success. A negative errno code is returned
6190 * on a failure to set up the device, or if the name is a duplicate.
6192 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6193 * and expands the device name if you passed a format string to
6196 int register_netdev(struct net_device
*dev
)
6201 err
= register_netdevice(dev
);
6205 EXPORT_SYMBOL(register_netdev
);
6207 int netdev_refcnt_read(const struct net_device
*dev
)
6211 for_each_possible_cpu(i
)
6212 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
6215 EXPORT_SYMBOL(netdev_refcnt_read
);
6218 * netdev_wait_allrefs - wait until all references are gone.
6219 * @dev: target net_device
6221 * This is called when unregistering network devices.
6223 * Any protocol or device that holds a reference should register
6224 * for netdevice notification, and cleanup and put back the
6225 * reference if they receive an UNREGISTER event.
6226 * We can get stuck here if buggy protocols don't correctly
6229 static void netdev_wait_allrefs(struct net_device
*dev
)
6231 unsigned long rebroadcast_time
, warning_time
;
6234 linkwatch_forget_dev(dev
);
6236 rebroadcast_time
= warning_time
= jiffies
;
6237 refcnt
= netdev_refcnt_read(dev
);
6239 while (refcnt
!= 0) {
6240 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
6243 /* Rebroadcast unregister notification */
6244 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6250 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
6251 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
6253 /* We must not have linkwatch events
6254 * pending on unregister. If this
6255 * happens, we simply run the queue
6256 * unscheduled, resulting in a noop
6259 linkwatch_run_queue();
6264 rebroadcast_time
= jiffies
;
6269 refcnt
= netdev_refcnt_read(dev
);
6271 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
6272 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6274 warning_time
= jiffies
;
6283 * register_netdevice(x1);
6284 * register_netdevice(x2);
6286 * unregister_netdevice(y1);
6287 * unregister_netdevice(y2);
6293 * We are invoked by rtnl_unlock().
6294 * This allows us to deal with problems:
6295 * 1) We can delete sysfs objects which invoke hotplug
6296 * without deadlocking with linkwatch via keventd.
6297 * 2) Since we run with the RTNL semaphore not held, we can sleep
6298 * safely in order to wait for the netdev refcnt to drop to zero.
6300 * We must not return until all unregister events added during
6301 * the interval the lock was held have been completed.
6303 void netdev_run_todo(void)
6305 struct list_head list
;
6307 /* Snapshot list, allow later requests */
6308 list_replace_init(&net_todo_list
, &list
);
6313 /* Wait for rcu callbacks to finish before next phase */
6314 if (!list_empty(&list
))
6317 while (!list_empty(&list
)) {
6318 struct net_device
*dev
6319 = list_first_entry(&list
, struct net_device
, todo_list
);
6320 list_del(&dev
->todo_list
);
6323 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
6326 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
6327 pr_err("network todo '%s' but state %d\n",
6328 dev
->name
, dev
->reg_state
);
6333 dev
->reg_state
= NETREG_UNREGISTERED
;
6335 on_each_cpu(flush_backlog
, dev
, 1);
6337 netdev_wait_allrefs(dev
);
6340 BUG_ON(netdev_refcnt_read(dev
));
6341 WARN_ON(rcu_access_pointer(dev
->ip_ptr
));
6342 WARN_ON(rcu_access_pointer(dev
->ip6_ptr
));
6343 WARN_ON(dev
->dn_ptr
);
6345 if (dev
->destructor
)
6346 dev
->destructor(dev
);
6348 /* Report a network device has been unregistered */
6350 dev_net(dev
)->dev_unreg_count
--;
6352 wake_up(&netdev_unregistering_wq
);
6354 /* Free network device */
6355 kobject_put(&dev
->dev
.kobj
);
6359 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6360 * fields in the same order, with only the type differing.
6362 void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
6363 const struct net_device_stats
*netdev_stats
)
6365 #if BITS_PER_LONG == 64
6366 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
6367 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
6369 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
6370 const unsigned long *src
= (const unsigned long *)netdev_stats
;
6371 u64
*dst
= (u64
*)stats64
;
6373 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
6374 sizeof(*stats64
) / sizeof(u64
));
6375 for (i
= 0; i
< n
; i
++)
6379 EXPORT_SYMBOL(netdev_stats_to_stats64
);
6382 * dev_get_stats - get network device statistics
6383 * @dev: device to get statistics from
6384 * @storage: place to store stats
6386 * Get network statistics from device. Return @storage.
6387 * The device driver may provide its own method by setting
6388 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6389 * otherwise the internal statistics structure is used.
6391 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
6392 struct rtnl_link_stats64
*storage
)
6394 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6396 if (ops
->ndo_get_stats64
) {
6397 memset(storage
, 0, sizeof(*storage
));
6398 ops
->ndo_get_stats64(dev
, storage
);
6399 } else if (ops
->ndo_get_stats
) {
6400 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
6402 netdev_stats_to_stats64(storage
, &dev
->stats
);
6404 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
6405 storage
->tx_dropped
+= atomic_long_read(&dev
->tx_dropped
);
6408 EXPORT_SYMBOL(dev_get_stats
);
6410 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
6412 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
6414 #ifdef CONFIG_NET_CLS_ACT
6417 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
6420 netdev_init_one_queue(dev
, queue
, NULL
);
6421 queue
->qdisc
= &noop_qdisc
;
6422 queue
->qdisc_sleeping
= &noop_qdisc
;
6423 rcu_assign_pointer(dev
->ingress_queue
, queue
);
6428 static const struct ethtool_ops default_ethtool_ops
;
6430 void netdev_set_default_ethtool_ops(struct net_device
*dev
,
6431 const struct ethtool_ops
*ops
)
6433 if (dev
->ethtool_ops
== &default_ethtool_ops
)
6434 dev
->ethtool_ops
= ops
;
6436 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops
);
6438 void netdev_freemem(struct net_device
*dev
)
6440 char *addr
= (char *)dev
- dev
->padded
;
6446 * alloc_netdev_mqs - allocate network device
6447 * @sizeof_priv: size of private data to allocate space for
6448 * @name: device name format string
6449 * @setup: callback to initialize device
6450 * @txqs: the number of TX subqueues to allocate
6451 * @rxqs: the number of RX subqueues to allocate
6453 * Allocates a struct net_device with private data area for driver use
6454 * and performs basic initialization. Also allocates subqueue structs
6455 * for each queue on the device.
6457 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
6458 void (*setup
)(struct net_device
*),
6459 unsigned int txqs
, unsigned int rxqs
)
6461 struct net_device
*dev
;
6463 struct net_device
*p
;
6465 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
6468 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6474 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6479 alloc_size
= sizeof(struct net_device
);
6481 /* ensure 32-byte alignment of private area */
6482 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
6483 alloc_size
+= sizeof_priv
;
6485 /* ensure 32-byte alignment of whole construct */
6486 alloc_size
+= NETDEV_ALIGN
- 1;
6488 p
= kzalloc(alloc_size
, GFP_KERNEL
| __GFP_NOWARN
| __GFP_REPEAT
);
6490 p
= vzalloc(alloc_size
);
6494 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
6495 dev
->padded
= (char *)dev
- (char *)p
;
6497 dev
->pcpu_refcnt
= alloc_percpu(int);
6498 if (!dev
->pcpu_refcnt
)
6501 if (dev_addr_init(dev
))
6507 dev_net_set(dev
, &init_net
);
6509 dev
->gso_max_size
= GSO_MAX_SIZE
;
6510 dev
->gso_max_segs
= GSO_MAX_SEGS
;
6512 INIT_LIST_HEAD(&dev
->napi_list
);
6513 INIT_LIST_HEAD(&dev
->unreg_list
);
6514 INIT_LIST_HEAD(&dev
->close_list
);
6515 INIT_LIST_HEAD(&dev
->link_watch_list
);
6516 INIT_LIST_HEAD(&dev
->adj_list
.upper
);
6517 INIT_LIST_HEAD(&dev
->adj_list
.lower
);
6518 INIT_LIST_HEAD(&dev
->all_adj_list
.upper
);
6519 INIT_LIST_HEAD(&dev
->all_adj_list
.lower
);
6520 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
6523 dev
->num_tx_queues
= txqs
;
6524 dev
->real_num_tx_queues
= txqs
;
6525 if (netif_alloc_netdev_queues(dev
))
6529 dev
->num_rx_queues
= rxqs
;
6530 dev
->real_num_rx_queues
= rxqs
;
6531 if (netif_alloc_rx_queues(dev
))
6535 strcpy(dev
->name
, name
);
6536 dev
->group
= INIT_NETDEV_GROUP
;
6537 if (!dev
->ethtool_ops
)
6538 dev
->ethtool_ops
= &default_ethtool_ops
;
6546 free_percpu(dev
->pcpu_refcnt
);
6548 netdev_freemem(dev
);
6551 EXPORT_SYMBOL(alloc_netdev_mqs
);
6554 * free_netdev - free network device
6557 * This function does the last stage of destroying an allocated device
6558 * interface. The reference to the device object is released.
6559 * If this is the last reference then it will be freed.
6561 void free_netdev(struct net_device
*dev
)
6563 struct napi_struct
*p
, *n
;
6565 release_net(dev_net(dev
));
6567 netif_free_tx_queues(dev
);
6572 kfree(rcu_dereference_protected(dev
->ingress_queue
, 1));
6574 /* Flush device addresses */
6575 dev_addr_flush(dev
);
6577 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
6580 free_percpu(dev
->pcpu_refcnt
);
6581 dev
->pcpu_refcnt
= NULL
;
6583 /* Compatibility with error handling in drivers */
6584 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
6585 netdev_freemem(dev
);
6589 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
6590 dev
->reg_state
= NETREG_RELEASED
;
6592 /* will free via device release */
6593 put_device(&dev
->dev
);
6595 EXPORT_SYMBOL(free_netdev
);
6598 * synchronize_net - Synchronize with packet receive processing
6600 * Wait for packets currently being received to be done.
6601 * Does not block later packets from starting.
6603 void synchronize_net(void)
6606 if (rtnl_is_locked())
6607 synchronize_rcu_expedited();
6611 EXPORT_SYMBOL(synchronize_net
);
6614 * unregister_netdevice_queue - remove device from the kernel
6618 * This function shuts down a device interface and removes it
6619 * from the kernel tables.
6620 * If head not NULL, device is queued to be unregistered later.
6622 * Callers must hold the rtnl semaphore. You may want
6623 * unregister_netdev() instead of this.
6626 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
6631 list_move_tail(&dev
->unreg_list
, head
);
6633 rollback_registered(dev
);
6634 /* Finish processing unregister after unlock */
6638 EXPORT_SYMBOL(unregister_netdevice_queue
);
6641 * unregister_netdevice_many - unregister many devices
6642 * @head: list of devices
6644 * Note: As most callers use a stack allocated list_head,
6645 * we force a list_del() to make sure stack wont be corrupted later.
6647 void unregister_netdevice_many(struct list_head
*head
)
6649 struct net_device
*dev
;
6651 if (!list_empty(head
)) {
6652 rollback_registered_many(head
);
6653 list_for_each_entry(dev
, head
, unreg_list
)
6658 EXPORT_SYMBOL(unregister_netdevice_many
);
6661 * unregister_netdev - remove device from the kernel
6664 * This function shuts down a device interface and removes it
6665 * from the kernel tables.
6667 * This is just a wrapper for unregister_netdevice that takes
6668 * the rtnl semaphore. In general you want to use this and not
6669 * unregister_netdevice.
6671 void unregister_netdev(struct net_device
*dev
)
6674 unregister_netdevice(dev
);
6677 EXPORT_SYMBOL(unregister_netdev
);
6680 * dev_change_net_namespace - move device to different nethost namespace
6682 * @net: network namespace
6683 * @pat: If not NULL name pattern to try if the current device name
6684 * is already taken in the destination network namespace.
6686 * This function shuts down a device interface and moves it
6687 * to a new network namespace. On success 0 is returned, on
6688 * a failure a netagive errno code is returned.
6690 * Callers must hold the rtnl semaphore.
6693 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
6699 /* Don't allow namespace local devices to be moved. */
6701 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6704 /* Ensure the device has been registrered */
6705 if (dev
->reg_state
!= NETREG_REGISTERED
)
6708 /* Get out if there is nothing todo */
6710 if (net_eq(dev_net(dev
), net
))
6713 /* Pick the destination device name, and ensure
6714 * we can use it in the destination network namespace.
6717 if (__dev_get_by_name(net
, dev
->name
)) {
6718 /* We get here if we can't use the current device name */
6721 if (dev_get_valid_name(net
, dev
, pat
) < 0)
6726 * And now a mini version of register_netdevice unregister_netdevice.
6729 /* If device is running close it first. */
6732 /* And unlink it from device chain */
6734 unlist_netdevice(dev
);
6738 /* Shutdown queueing discipline. */
6741 /* Notify protocols, that we are about to destroy
6742 this device. They should clean all the things.
6744 Note that dev->reg_state stays at NETREG_REGISTERED.
6745 This is wanted because this way 8021q and macvlan know
6746 the device is just moving and can keep their slaves up.
6748 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6750 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
6751 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U, GFP_KERNEL
);
6754 * Flush the unicast and multicast chains
6759 /* Send a netdev-removed uevent to the old namespace */
6760 kobject_uevent(&dev
->dev
.kobj
, KOBJ_REMOVE
);
6762 /* Actually switch the network namespace */
6763 dev_net_set(dev
, net
);
6765 /* If there is an ifindex conflict assign a new one */
6766 if (__dev_get_by_index(net
, dev
->ifindex
)) {
6767 int iflink
= (dev
->iflink
== dev
->ifindex
);
6768 dev
->ifindex
= dev_new_index(net
);
6770 dev
->iflink
= dev
->ifindex
;
6773 /* Send a netdev-add uevent to the new namespace */
6774 kobject_uevent(&dev
->dev
.kobj
, KOBJ_ADD
);
6776 /* Fixup kobjects */
6777 err
= device_rename(&dev
->dev
, dev
->name
);
6780 /* Add the device back in the hashes */
6781 list_netdevice(dev
);
6783 /* Notify protocols, that a new device appeared. */
6784 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
6787 * Prevent userspace races by waiting until the network
6788 * device is fully setup before sending notifications.
6790 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U, GFP_KERNEL
);
6797 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
6799 static int dev_cpu_callback(struct notifier_block
*nfb
,
6800 unsigned long action
,
6803 struct sk_buff
**list_skb
;
6804 struct sk_buff
*skb
;
6805 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
6806 struct softnet_data
*sd
, *oldsd
;
6808 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
6811 local_irq_disable();
6812 cpu
= smp_processor_id();
6813 sd
= &per_cpu(softnet_data
, cpu
);
6814 oldsd
= &per_cpu(softnet_data
, oldcpu
);
6816 /* Find end of our completion_queue. */
6817 list_skb
= &sd
->completion_queue
;
6819 list_skb
= &(*list_skb
)->next
;
6820 /* Append completion queue from offline CPU. */
6821 *list_skb
= oldsd
->completion_queue
;
6822 oldsd
->completion_queue
= NULL
;
6824 /* Append output queue from offline CPU. */
6825 if (oldsd
->output_queue
) {
6826 *sd
->output_queue_tailp
= oldsd
->output_queue
;
6827 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
6828 oldsd
->output_queue
= NULL
;
6829 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
6831 /* Append NAPI poll list from offline CPU. */
6832 if (!list_empty(&oldsd
->poll_list
)) {
6833 list_splice_init(&oldsd
->poll_list
, &sd
->poll_list
);
6834 raise_softirq_irqoff(NET_RX_SOFTIRQ
);
6837 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
6840 /* Process offline CPU's input_pkt_queue */
6841 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
6842 netif_rx_internal(skb
);
6843 input_queue_head_incr(oldsd
);
6845 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
6846 netif_rx_internal(skb
);
6847 input_queue_head_incr(oldsd
);
6855 * netdev_increment_features - increment feature set by one
6856 * @all: current feature set
6857 * @one: new feature set
6858 * @mask: mask feature set
6860 * Computes a new feature set after adding a device with feature set
6861 * @one to the master device with current feature set @all. Will not
6862 * enable anything that is off in @mask. Returns the new feature set.
6864 netdev_features_t
netdev_increment_features(netdev_features_t all
,
6865 netdev_features_t one
, netdev_features_t mask
)
6867 if (mask
& NETIF_F_GEN_CSUM
)
6868 mask
|= NETIF_F_ALL_CSUM
;
6869 mask
|= NETIF_F_VLAN_CHALLENGED
;
6871 all
|= one
& (NETIF_F_ONE_FOR_ALL
|NETIF_F_ALL_CSUM
) & mask
;
6872 all
&= one
| ~NETIF_F_ALL_FOR_ALL
;
6874 /* If one device supports hw checksumming, set for all. */
6875 if (all
& NETIF_F_GEN_CSUM
)
6876 all
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_GEN_CSUM
);
6880 EXPORT_SYMBOL(netdev_increment_features
);
6882 static struct hlist_head
* __net_init
netdev_create_hash(void)
6885 struct hlist_head
*hash
;
6887 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
6889 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
6890 INIT_HLIST_HEAD(&hash
[i
]);
6895 /* Initialize per network namespace state */
6896 static int __net_init
netdev_init(struct net
*net
)
6898 if (net
!= &init_net
)
6899 INIT_LIST_HEAD(&net
->dev_base_head
);
6901 net
->dev_name_head
= netdev_create_hash();
6902 if (net
->dev_name_head
== NULL
)
6905 net
->dev_index_head
= netdev_create_hash();
6906 if (net
->dev_index_head
== NULL
)
6912 kfree(net
->dev_name_head
);
6918 * netdev_drivername - network driver for the device
6919 * @dev: network device
6921 * Determine network driver for device.
6923 const char *netdev_drivername(const struct net_device
*dev
)
6925 const struct device_driver
*driver
;
6926 const struct device
*parent
;
6927 const char *empty
= "";
6929 parent
= dev
->dev
.parent
;
6933 driver
= parent
->driver
;
6934 if (driver
&& driver
->name
)
6935 return driver
->name
;
6939 static int __netdev_printk(const char *level
, const struct net_device
*dev
,
6940 struct va_format
*vaf
)
6944 if (dev
&& dev
->dev
.parent
) {
6945 r
= dev_printk_emit(level
[1] - '0',
6948 dev_driver_string(dev
->dev
.parent
),
6949 dev_name(dev
->dev
.parent
),
6950 netdev_name(dev
), vaf
);
6952 r
= printk("%s%s: %pV", level
, netdev_name(dev
), vaf
);
6954 r
= printk("%s(NULL net_device): %pV", level
, vaf
);
6960 int netdev_printk(const char *level
, const struct net_device
*dev
,
6961 const char *format
, ...)
6963 struct va_format vaf
;
6967 va_start(args
, format
);
6972 r
= __netdev_printk(level
, dev
, &vaf
);
6978 EXPORT_SYMBOL(netdev_printk
);
6980 #define define_netdev_printk_level(func, level) \
6981 int func(const struct net_device *dev, const char *fmt, ...) \
6984 struct va_format vaf; \
6987 va_start(args, fmt); \
6992 r = __netdev_printk(level, dev, &vaf); \
6998 EXPORT_SYMBOL(func);
7000 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
7001 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
7002 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
7003 define_netdev_printk_level(netdev_err
, KERN_ERR
);
7004 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
7005 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
7006 define_netdev_printk_level(netdev_info
, KERN_INFO
);
7008 static void __net_exit
netdev_exit(struct net
*net
)
7010 kfree(net
->dev_name_head
);
7011 kfree(net
->dev_index_head
);
7014 static struct pernet_operations __net_initdata netdev_net_ops
= {
7015 .init
= netdev_init
,
7016 .exit
= netdev_exit
,
7019 static void __net_exit
default_device_exit(struct net
*net
)
7021 struct net_device
*dev
, *aux
;
7023 * Push all migratable network devices back to the
7024 * initial network namespace
7027 for_each_netdev_safe(net
, dev
, aux
) {
7029 char fb_name
[IFNAMSIZ
];
7031 /* Ignore unmoveable devices (i.e. loopback) */
7032 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
7035 /* Leave virtual devices for the generic cleanup */
7036 if (dev
->rtnl_link_ops
)
7039 /* Push remaining network devices to init_net */
7040 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
7041 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
7043 pr_emerg("%s: failed to move %s to init_net: %d\n",
7044 __func__
, dev
->name
, err
);
7051 static void __net_exit
rtnl_lock_unregistering(struct list_head
*net_list
)
7053 /* Return with the rtnl_lock held when there are no network
7054 * devices unregistering in any network namespace in net_list.
7061 prepare_to_wait(&netdev_unregistering_wq
, &wait
,
7062 TASK_UNINTERRUPTIBLE
);
7063 unregistering
= false;
7065 list_for_each_entry(net
, net_list
, exit_list
) {
7066 if (net
->dev_unreg_count
> 0) {
7067 unregistering
= true;
7076 finish_wait(&netdev_unregistering_wq
, &wait
);
7079 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
7081 /* At exit all network devices most be removed from a network
7082 * namespace. Do this in the reverse order of registration.
7083 * Do this across as many network namespaces as possible to
7084 * improve batching efficiency.
7086 struct net_device
*dev
;
7088 LIST_HEAD(dev_kill_list
);
7090 /* To prevent network device cleanup code from dereferencing
7091 * loopback devices or network devices that have been freed
7092 * wait here for all pending unregistrations to complete,
7093 * before unregistring the loopback device and allowing the
7094 * network namespace be freed.
7096 * The netdev todo list containing all network devices
7097 * unregistrations that happen in default_device_exit_batch
7098 * will run in the rtnl_unlock() at the end of
7099 * default_device_exit_batch.
7101 rtnl_lock_unregistering(net_list
);
7102 list_for_each_entry(net
, net_list
, exit_list
) {
7103 for_each_netdev_reverse(net
, dev
) {
7104 if (dev
->rtnl_link_ops
)
7105 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
7107 unregister_netdevice_queue(dev
, &dev_kill_list
);
7110 unregister_netdevice_many(&dev_kill_list
);
7114 static struct pernet_operations __net_initdata default_device_ops
= {
7115 .exit
= default_device_exit
,
7116 .exit_batch
= default_device_exit_batch
,
7120 * Initialize the DEV module. At boot time this walks the device list and
7121 * unhooks any devices that fail to initialise (normally hardware not
7122 * present) and leaves us with a valid list of present and active devices.
7127 * This is called single threaded during boot, so no need
7128 * to take the rtnl semaphore.
7130 static int __init
net_dev_init(void)
7132 int i
, rc
= -ENOMEM
;
7134 BUG_ON(!dev_boot_phase
);
7136 if (dev_proc_init())
7139 if (netdev_kobject_init())
7142 INIT_LIST_HEAD(&ptype_all
);
7143 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
7144 INIT_LIST_HEAD(&ptype_base
[i
]);
7146 INIT_LIST_HEAD(&offload_base
);
7148 if (register_pernet_subsys(&netdev_net_ops
))
7152 * Initialise the packet receive queues.
7155 for_each_possible_cpu(i
) {
7156 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
7158 skb_queue_head_init(&sd
->input_pkt_queue
);
7159 skb_queue_head_init(&sd
->process_queue
);
7160 INIT_LIST_HEAD(&sd
->poll_list
);
7161 sd
->output_queue_tailp
= &sd
->output_queue
;
7163 sd
->csd
.func
= rps_trigger_softirq
;
7168 sd
->backlog
.poll
= process_backlog
;
7169 sd
->backlog
.weight
= weight_p
;
7174 /* The loopback device is special if any other network devices
7175 * is present in a network namespace the loopback device must
7176 * be present. Since we now dynamically allocate and free the
7177 * loopback device ensure this invariant is maintained by
7178 * keeping the loopback device as the first device on the
7179 * list of network devices. Ensuring the loopback devices
7180 * is the first device that appears and the last network device
7183 if (register_pernet_device(&loopback_net_ops
))
7186 if (register_pernet_device(&default_device_ops
))
7189 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
7190 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
7192 hotcpu_notifier(dev_cpu_callback
, 0);
7199 subsys_initcall(net_dev_init
);