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>
133 #include "net-sysfs.h"
136 #include <linux/ftrace_event.h>
139 /* Instead of increasing this, you should create a hash table. */
140 #define MAX_GRO_SKBS 8
142 /* This should be increased if a protocol with a bigger head is added. */
143 #define GRO_MAX_HEAD (MAX_HEADER + 128)
145 static DEFINE_SPINLOCK(ptype_lock
);
146 static DEFINE_SPINLOCK(offload_lock
);
147 struct list_head ptype_base
[PTYPE_HASH_SIZE
] __read_mostly
;
148 struct list_head ptype_all __read_mostly
; /* Taps */
149 static struct list_head offload_base __read_mostly
;
152 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
155 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
157 * Writers must hold the rtnl semaphore while they loop through the
158 * dev_base_head list, and hold dev_base_lock for writing when they do the
159 * actual updates. This allows pure readers to access the list even
160 * while a writer is preparing to update it.
162 * To put it another way, dev_base_lock is held for writing only to
163 * protect against pure readers; the rtnl semaphore provides the
164 * protection against other writers.
166 * See, for example usages, register_netdevice() and
167 * unregister_netdevice(), which must be called with the rtnl
170 DEFINE_RWLOCK(dev_base_lock
);
171 EXPORT_SYMBOL(dev_base_lock
);
173 seqcount_t devnet_rename_seq
;
175 static inline void dev_base_seq_inc(struct net
*net
)
177 while (++net
->dev_base_seq
== 0);
180 static inline struct hlist_head
*dev_name_hash(struct net
*net
, const char *name
)
182 unsigned int hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
184 return &net
->dev_name_head
[hash_32(hash
, NETDEV_HASHBITS
)];
187 static inline struct hlist_head
*dev_index_hash(struct net
*net
, int ifindex
)
189 return &net
->dev_index_head
[ifindex
& (NETDEV_HASHENTRIES
- 1)];
192 static inline void rps_lock(struct softnet_data
*sd
)
195 spin_lock(&sd
->input_pkt_queue
.lock
);
199 static inline void rps_unlock(struct softnet_data
*sd
)
202 spin_unlock(&sd
->input_pkt_queue
.lock
);
206 /* Device list insertion */
207 static void list_netdevice(struct net_device
*dev
)
209 struct net
*net
= dev_net(dev
);
213 write_lock_bh(&dev_base_lock
);
214 list_add_tail_rcu(&dev
->dev_list
, &net
->dev_base_head
);
215 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
216 hlist_add_head_rcu(&dev
->index_hlist
,
217 dev_index_hash(net
, dev
->ifindex
));
218 write_unlock_bh(&dev_base_lock
);
220 dev_base_seq_inc(net
);
223 /* Device list removal
224 * caller must respect a RCU grace period before freeing/reusing dev
226 static void unlist_netdevice(struct net_device
*dev
)
230 /* Unlink dev from the device chain */
231 write_lock_bh(&dev_base_lock
);
232 list_del_rcu(&dev
->dev_list
);
233 hlist_del_rcu(&dev
->name_hlist
);
234 hlist_del_rcu(&dev
->index_hlist
);
235 write_unlock_bh(&dev_base_lock
);
237 dev_base_seq_inc(dev_net(dev
));
244 static RAW_NOTIFIER_HEAD(netdev_chain
);
247 * Device drivers call our routines to queue packets here. We empty the
248 * queue in the local softnet handler.
251 DEFINE_PER_CPU_ALIGNED(struct softnet_data
, softnet_data
);
252 EXPORT_PER_CPU_SYMBOL(softnet_data
);
254 #ifdef CONFIG_LOCKDEP
256 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
257 * according to dev->type
259 static const unsigned short netdev_lock_type
[] =
260 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
261 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
262 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
263 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
264 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
265 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
266 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
267 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
268 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
269 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
270 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
271 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
272 ARPHRD_FCFABRIC
, ARPHRD_IEEE80211
, ARPHRD_IEEE80211_PRISM
,
273 ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_PHONET
, ARPHRD_PHONET_PIPE
,
274 ARPHRD_IEEE802154
, ARPHRD_VOID
, ARPHRD_NONE
};
276 static const char *const netdev_lock_name
[] =
277 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
278 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
279 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
280 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
281 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
282 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
283 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
284 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
285 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
286 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
287 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
288 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
289 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
290 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
291 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
293 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
294 static struct lock_class_key netdev_addr_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
296 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
300 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
301 if (netdev_lock_type
[i
] == dev_type
)
303 /* the last key is used by default */
304 return ARRAY_SIZE(netdev_lock_type
) - 1;
307 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
308 unsigned short dev_type
)
312 i
= netdev_lock_pos(dev_type
);
313 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
314 netdev_lock_name
[i
]);
317 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
321 i
= netdev_lock_pos(dev
->type
);
322 lockdep_set_class_and_name(&dev
->addr_list_lock
,
323 &netdev_addr_lock_key
[i
],
324 netdev_lock_name
[i
]);
327 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
328 unsigned short dev_type
)
331 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
336 /*******************************************************************************
338 Protocol management and registration routines
340 *******************************************************************************/
343 * Add a protocol ID to the list. Now that the input handler is
344 * smarter we can dispense with all the messy stuff that used to be
347 * BEWARE!!! Protocol handlers, mangling input packets,
348 * MUST BE last in hash buckets and checking protocol handlers
349 * MUST start from promiscuous ptype_all chain in net_bh.
350 * It is true now, do not change it.
351 * Explanation follows: if protocol handler, mangling packet, will
352 * be the first on list, it is not able to sense, that packet
353 * is cloned and should be copied-on-write, so that it will
354 * change it and subsequent readers will get broken packet.
358 static inline struct list_head
*ptype_head(const struct packet_type
*pt
)
360 if (pt
->type
== htons(ETH_P_ALL
))
363 return &ptype_base
[ntohs(pt
->type
) & PTYPE_HASH_MASK
];
367 * dev_add_pack - add packet handler
368 * @pt: packet type declaration
370 * Add a protocol handler to the networking stack. The passed &packet_type
371 * is linked into kernel lists and may not be freed until it has been
372 * removed from the kernel lists.
374 * This call does not sleep therefore it can not
375 * guarantee all CPU's that are in middle of receiving packets
376 * will see the new packet type (until the next received packet).
379 void dev_add_pack(struct packet_type
*pt
)
381 struct list_head
*head
= ptype_head(pt
);
383 spin_lock(&ptype_lock
);
384 list_add_rcu(&pt
->list
, head
);
385 spin_unlock(&ptype_lock
);
387 EXPORT_SYMBOL(dev_add_pack
);
390 * __dev_remove_pack - remove packet handler
391 * @pt: packet type declaration
393 * Remove a protocol handler that was previously added to the kernel
394 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
395 * from the kernel lists and can be freed or reused once this function
398 * The packet type might still be in use by receivers
399 * and must not be freed until after all the CPU's have gone
400 * through a quiescent state.
402 void __dev_remove_pack(struct packet_type
*pt
)
404 struct list_head
*head
= ptype_head(pt
);
405 struct packet_type
*pt1
;
407 spin_lock(&ptype_lock
);
409 list_for_each_entry(pt1
, head
, list
) {
411 list_del_rcu(&pt
->list
);
416 pr_warn("dev_remove_pack: %p not found\n", pt
);
418 spin_unlock(&ptype_lock
);
420 EXPORT_SYMBOL(__dev_remove_pack
);
423 * dev_remove_pack - remove packet handler
424 * @pt: packet type declaration
426 * Remove a protocol handler that was previously added to the kernel
427 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
428 * from the kernel lists and can be freed or reused once this function
431 * This call sleeps to guarantee that no CPU is looking at the packet
434 void dev_remove_pack(struct packet_type
*pt
)
436 __dev_remove_pack(pt
);
440 EXPORT_SYMBOL(dev_remove_pack
);
444 * dev_add_offload - register offload handlers
445 * @po: protocol offload declaration
447 * Add protocol offload handlers to the networking stack. The passed
448 * &proto_offload is linked into kernel lists and may not be freed until
449 * it has been removed from the kernel lists.
451 * This call does not sleep therefore it can not
452 * guarantee all CPU's that are in middle of receiving packets
453 * will see the new offload handlers (until the next received packet).
455 void dev_add_offload(struct packet_offload
*po
)
457 struct list_head
*head
= &offload_base
;
459 spin_lock(&offload_lock
);
460 list_add_rcu(&po
->list
, head
);
461 spin_unlock(&offload_lock
);
463 EXPORT_SYMBOL(dev_add_offload
);
466 * __dev_remove_offload - remove offload handler
467 * @po: packet offload declaration
469 * Remove a protocol offload handler that was previously added to the
470 * kernel offload handlers by dev_add_offload(). The passed &offload_type
471 * is removed from the kernel lists and can be freed or reused once this
474 * The packet type might still be in use by receivers
475 * and must not be freed until after all the CPU's have gone
476 * through a quiescent state.
478 void __dev_remove_offload(struct packet_offload
*po
)
480 struct list_head
*head
= &offload_base
;
481 struct packet_offload
*po1
;
483 spin_lock(&offload_lock
);
485 list_for_each_entry(po1
, head
, list
) {
487 list_del_rcu(&po
->list
);
492 pr_warn("dev_remove_offload: %p not found\n", po
);
494 spin_unlock(&offload_lock
);
496 EXPORT_SYMBOL(__dev_remove_offload
);
499 * dev_remove_offload - remove packet offload handler
500 * @po: packet offload declaration
502 * Remove a packet offload handler that was previously added to the kernel
503 * offload handlers by dev_add_offload(). The passed &offload_type is
504 * removed from the kernel lists and can be freed or reused once this
507 * This call sleeps to guarantee that no CPU is looking at the packet
510 void dev_remove_offload(struct packet_offload
*po
)
512 __dev_remove_offload(po
);
516 EXPORT_SYMBOL(dev_remove_offload
);
518 /******************************************************************************
520 Device Boot-time Settings Routines
522 *******************************************************************************/
524 /* Boot time configuration table */
525 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
528 * netdev_boot_setup_add - add new setup entry
529 * @name: name of the device
530 * @map: configured settings for the device
532 * Adds new setup entry to the dev_boot_setup list. The function
533 * returns 0 on error and 1 on success. This is a generic routine to
536 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
538 struct netdev_boot_setup
*s
;
542 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
543 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
544 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
545 strlcpy(s
[i
].name
, name
, IFNAMSIZ
);
546 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
551 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
555 * netdev_boot_setup_check - check boot time settings
556 * @dev: the netdevice
558 * Check boot time settings for the device.
559 * The found settings are set for the device to be used
560 * later in the device probing.
561 * Returns 0 if no settings found, 1 if they are.
563 int netdev_boot_setup_check(struct net_device
*dev
)
565 struct netdev_boot_setup
*s
= dev_boot_setup
;
568 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
569 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
570 !strcmp(dev
->name
, s
[i
].name
)) {
571 dev
->irq
= s
[i
].map
.irq
;
572 dev
->base_addr
= s
[i
].map
.base_addr
;
573 dev
->mem_start
= s
[i
].map
.mem_start
;
574 dev
->mem_end
= s
[i
].map
.mem_end
;
580 EXPORT_SYMBOL(netdev_boot_setup_check
);
584 * netdev_boot_base - get address from boot time settings
585 * @prefix: prefix for network device
586 * @unit: id for network device
588 * Check boot time settings for the base address of device.
589 * The found settings are set for the device to be used
590 * later in the device probing.
591 * Returns 0 if no settings found.
593 unsigned long netdev_boot_base(const char *prefix
, int unit
)
595 const struct netdev_boot_setup
*s
= dev_boot_setup
;
599 sprintf(name
, "%s%d", prefix
, unit
);
602 * If device already registered then return base of 1
603 * to indicate not to probe for this interface
605 if (__dev_get_by_name(&init_net
, name
))
608 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
609 if (!strcmp(name
, s
[i
].name
))
610 return s
[i
].map
.base_addr
;
615 * Saves at boot time configured settings for any netdevice.
617 int __init
netdev_boot_setup(char *str
)
622 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
627 memset(&map
, 0, sizeof(map
));
631 map
.base_addr
= ints
[2];
633 map
.mem_start
= ints
[3];
635 map
.mem_end
= ints
[4];
637 /* Add new entry to the list */
638 return netdev_boot_setup_add(str
, &map
);
641 __setup("netdev=", netdev_boot_setup
);
643 /*******************************************************************************
645 Device Interface Subroutines
647 *******************************************************************************/
650 * __dev_get_by_name - find a device by its name
651 * @net: the applicable net namespace
652 * @name: name to find
654 * Find an interface by name. Must be called under RTNL semaphore
655 * or @dev_base_lock. If the name is found a pointer to the device
656 * is returned. If the name is not found then %NULL is returned. The
657 * reference counters are not incremented so the caller must be
658 * careful with locks.
661 struct net_device
*__dev_get_by_name(struct net
*net
, const char *name
)
663 struct net_device
*dev
;
664 struct hlist_head
*head
= dev_name_hash(net
, name
);
666 hlist_for_each_entry(dev
, head
, name_hlist
)
667 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
672 EXPORT_SYMBOL(__dev_get_by_name
);
675 * dev_get_by_name_rcu - find a device by its name
676 * @net: the applicable net namespace
677 * @name: name to find
679 * Find an interface by name.
680 * If the name is found a pointer to the device is returned.
681 * If the name is not found then %NULL is returned.
682 * The reference counters are not incremented so the caller must be
683 * careful with locks. The caller must hold RCU lock.
686 struct net_device
*dev_get_by_name_rcu(struct net
*net
, const char *name
)
688 struct net_device
*dev
;
689 struct hlist_head
*head
= dev_name_hash(net
, name
);
691 hlist_for_each_entry_rcu(dev
, head
, name_hlist
)
692 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
697 EXPORT_SYMBOL(dev_get_by_name_rcu
);
700 * dev_get_by_name - find a device by its name
701 * @net: the applicable net namespace
702 * @name: name to find
704 * Find an interface by name. This can be called from any
705 * context and does its own locking. The returned handle has
706 * the usage count incremented and the caller must use dev_put() to
707 * release it when it is no longer needed. %NULL is returned if no
708 * matching device is found.
711 struct net_device
*dev_get_by_name(struct net
*net
, const char *name
)
713 struct net_device
*dev
;
716 dev
= dev_get_by_name_rcu(net
, name
);
722 EXPORT_SYMBOL(dev_get_by_name
);
725 * __dev_get_by_index - find a device by its ifindex
726 * @net: the applicable net namespace
727 * @ifindex: index of device
729 * Search for an interface by index. Returns %NULL if the device
730 * is not found or a pointer to the device. The device has not
731 * had its reference counter increased so the caller must be careful
732 * about locking. The caller must hold either the RTNL semaphore
736 struct net_device
*__dev_get_by_index(struct net
*net
, int ifindex
)
738 struct net_device
*dev
;
739 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
741 hlist_for_each_entry(dev
, head
, index_hlist
)
742 if (dev
->ifindex
== ifindex
)
747 EXPORT_SYMBOL(__dev_get_by_index
);
750 * dev_get_by_index_rcu - find a device by its ifindex
751 * @net: the applicable net namespace
752 * @ifindex: index of device
754 * Search for an interface by index. Returns %NULL if the device
755 * is not found or a pointer to the device. The device has not
756 * had its reference counter increased so the caller must be careful
757 * about locking. The caller must hold RCU lock.
760 struct net_device
*dev_get_by_index_rcu(struct net
*net
, int ifindex
)
762 struct net_device
*dev
;
763 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
765 hlist_for_each_entry_rcu(dev
, head
, index_hlist
)
766 if (dev
->ifindex
== ifindex
)
771 EXPORT_SYMBOL(dev_get_by_index_rcu
);
775 * dev_get_by_index - find a device by its ifindex
776 * @net: the applicable net namespace
777 * @ifindex: index of device
779 * Search for an interface by index. Returns NULL if the device
780 * is not found or a pointer to the device. The device returned has
781 * had a reference added and the pointer is safe until the user calls
782 * dev_put to indicate they have finished with it.
785 struct net_device
*dev_get_by_index(struct net
*net
, int ifindex
)
787 struct net_device
*dev
;
790 dev
= dev_get_by_index_rcu(net
, ifindex
);
796 EXPORT_SYMBOL(dev_get_by_index
);
799 * netdev_get_name - get a netdevice name, knowing its ifindex.
800 * @net: network namespace
801 * @name: a pointer to the buffer where the name will be stored.
802 * @ifindex: the ifindex of the interface to get the name from.
804 * The use of raw_seqcount_begin() and cond_resched() before
805 * retrying is required as we want to give the writers a chance
806 * to complete when CONFIG_PREEMPT is not set.
808 int netdev_get_name(struct net
*net
, char *name
, int ifindex
)
810 struct net_device
*dev
;
814 seq
= raw_seqcount_begin(&devnet_rename_seq
);
816 dev
= dev_get_by_index_rcu(net
, ifindex
);
822 strcpy(name
, dev
->name
);
824 if (read_seqcount_retry(&devnet_rename_seq
, seq
)) {
833 * dev_getbyhwaddr_rcu - find a device by its hardware address
834 * @net: the applicable net namespace
835 * @type: media type of device
836 * @ha: hardware address
838 * Search for an interface by MAC address. Returns NULL if the device
839 * is not found or a pointer to the device.
840 * The caller must hold RCU or RTNL.
841 * The returned device has not had its ref count increased
842 * and the caller must therefore be careful about locking
846 struct net_device
*dev_getbyhwaddr_rcu(struct net
*net
, unsigned short type
,
849 struct net_device
*dev
;
851 for_each_netdev_rcu(net
, dev
)
852 if (dev
->type
== type
&&
853 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
858 EXPORT_SYMBOL(dev_getbyhwaddr_rcu
);
860 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
862 struct net_device
*dev
;
865 for_each_netdev(net
, dev
)
866 if (dev
->type
== type
)
871 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
873 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
875 struct net_device
*dev
, *ret
= NULL
;
878 for_each_netdev_rcu(net
, dev
)
879 if (dev
->type
== type
) {
887 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
890 * dev_get_by_flags_rcu - find any device with given flags
891 * @net: the applicable net namespace
892 * @if_flags: IFF_* values
893 * @mask: bitmask of bits in if_flags to check
895 * Search for any interface with the given flags. Returns NULL if a device
896 * is not found or a pointer to the device. Must be called inside
897 * rcu_read_lock(), and result refcount is unchanged.
900 struct net_device
*dev_get_by_flags_rcu(struct net
*net
, unsigned short if_flags
,
903 struct net_device
*dev
, *ret
;
906 for_each_netdev_rcu(net
, dev
) {
907 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
914 EXPORT_SYMBOL(dev_get_by_flags_rcu
);
917 * dev_valid_name - check if name is okay for network device
920 * Network device names need to be valid file names to
921 * to allow sysfs to work. We also disallow any kind of
924 bool dev_valid_name(const char *name
)
928 if (strlen(name
) >= IFNAMSIZ
)
930 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
934 if (*name
== '/' || *name
== ':' || isspace(*name
))
940 EXPORT_SYMBOL(dev_valid_name
);
943 * __dev_alloc_name - allocate a name for a device
944 * @net: network namespace to allocate the device name in
945 * @name: name format string
946 * @buf: scratch buffer and result name string
948 * Passed a format string - eg "lt%d" it will try and find a suitable
949 * id. It scans list of devices to build up a free map, then chooses
950 * the first empty slot. The caller must hold the dev_base or rtnl lock
951 * while allocating the name and adding the device in order to avoid
953 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
954 * Returns the number of the unit assigned or a negative errno code.
957 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
961 const int max_netdevices
= 8*PAGE_SIZE
;
962 unsigned long *inuse
;
963 struct net_device
*d
;
965 p
= strnchr(name
, IFNAMSIZ
-1, '%');
968 * Verify the string as this thing may have come from
969 * the user. There must be either one "%d" and no other "%"
972 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
975 /* Use one page as a bit array of possible slots */
976 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
980 for_each_netdev(net
, d
) {
981 if (!sscanf(d
->name
, name
, &i
))
983 if (i
< 0 || i
>= max_netdevices
)
986 /* avoid cases where sscanf is not exact inverse of printf */
987 snprintf(buf
, IFNAMSIZ
, name
, i
);
988 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
992 i
= find_first_zero_bit(inuse
, max_netdevices
);
993 free_page((unsigned long) inuse
);
997 snprintf(buf
, IFNAMSIZ
, name
, i
);
998 if (!__dev_get_by_name(net
, buf
))
1001 /* It is possible to run out of possible slots
1002 * when the name is long and there isn't enough space left
1003 * for the digits, or if all bits are used.
1009 * dev_alloc_name - allocate a name for a device
1011 * @name: name format string
1013 * Passed a format string - eg "lt%d" it will try and find a suitable
1014 * id. It scans list of devices to build up a free map, then chooses
1015 * the first empty slot. The caller must hold the dev_base or rtnl lock
1016 * while allocating the name and adding the device in order to avoid
1018 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1019 * Returns the number of the unit assigned or a negative errno code.
1022 int dev_alloc_name(struct net_device
*dev
, const char *name
)
1028 BUG_ON(!dev_net(dev
));
1030 ret
= __dev_alloc_name(net
, name
, buf
);
1032 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
1035 EXPORT_SYMBOL(dev_alloc_name
);
1037 static int dev_alloc_name_ns(struct net
*net
,
1038 struct net_device
*dev
,
1044 ret
= __dev_alloc_name(net
, name
, buf
);
1046 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
1050 static int dev_get_valid_name(struct net
*net
,
1051 struct net_device
*dev
,
1056 if (!dev_valid_name(name
))
1059 if (strchr(name
, '%'))
1060 return dev_alloc_name_ns(net
, dev
, name
);
1061 else if (__dev_get_by_name(net
, name
))
1063 else if (dev
->name
!= name
)
1064 strlcpy(dev
->name
, name
, IFNAMSIZ
);
1070 * dev_change_name - change name of a device
1072 * @newname: name (or format string) must be at least IFNAMSIZ
1074 * Change name of a device, can pass format strings "eth%d".
1077 int dev_change_name(struct net_device
*dev
, const char *newname
)
1079 char oldname
[IFNAMSIZ
];
1085 BUG_ON(!dev_net(dev
));
1088 if (dev
->flags
& IFF_UP
)
1091 write_seqcount_begin(&devnet_rename_seq
);
1093 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0) {
1094 write_seqcount_end(&devnet_rename_seq
);
1098 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
1100 err
= dev_get_valid_name(net
, dev
, newname
);
1102 write_seqcount_end(&devnet_rename_seq
);
1107 ret
= device_rename(&dev
->dev
, dev
->name
);
1109 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1110 write_seqcount_end(&devnet_rename_seq
);
1114 write_seqcount_end(&devnet_rename_seq
);
1116 write_lock_bh(&dev_base_lock
);
1117 hlist_del_rcu(&dev
->name_hlist
);
1118 write_unlock_bh(&dev_base_lock
);
1122 write_lock_bh(&dev_base_lock
);
1123 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
1124 write_unlock_bh(&dev_base_lock
);
1126 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
1127 ret
= notifier_to_errno(ret
);
1130 /* err >= 0 after dev_alloc_name() or stores the first errno */
1133 write_seqcount_begin(&devnet_rename_seq
);
1134 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1137 pr_err("%s: name change rollback failed: %d\n",
1146 * dev_set_alias - change ifalias of a device
1148 * @alias: name up to IFALIASZ
1149 * @len: limit of bytes to copy from info
1151 * Set ifalias for a device,
1153 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
1159 if (len
>= IFALIASZ
)
1163 kfree(dev
->ifalias
);
1164 dev
->ifalias
= NULL
;
1168 new_ifalias
= krealloc(dev
->ifalias
, len
+ 1, GFP_KERNEL
);
1171 dev
->ifalias
= new_ifalias
;
1173 strlcpy(dev
->ifalias
, alias
, len
+1);
1179 * netdev_features_change - device changes features
1180 * @dev: device to cause notification
1182 * Called to indicate a device has changed features.
1184 void netdev_features_change(struct net_device
*dev
)
1186 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1188 EXPORT_SYMBOL(netdev_features_change
);
1191 * netdev_state_change - device changes state
1192 * @dev: device to cause notification
1194 * Called to indicate a device has changed state. This function calls
1195 * the notifier chains for netdev_chain and sends a NEWLINK message
1196 * to the routing socket.
1198 void netdev_state_change(struct net_device
*dev
)
1200 if (dev
->flags
& IFF_UP
) {
1201 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
1202 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
1205 EXPORT_SYMBOL(netdev_state_change
);
1208 * netdev_notify_peers - notify network peers about existence of @dev
1209 * @dev: network device
1211 * Generate traffic such that interested network peers are aware of
1212 * @dev, such as by generating a gratuitous ARP. This may be used when
1213 * a device wants to inform the rest of the network about some sort of
1214 * reconfiguration such as a failover event or virtual machine
1217 void netdev_notify_peers(struct net_device
*dev
)
1220 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
, dev
);
1223 EXPORT_SYMBOL(netdev_notify_peers
);
1225 static int __dev_open(struct net_device
*dev
)
1227 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1232 if (!netif_device_present(dev
))
1235 /* Block netpoll from trying to do any rx path servicing.
1236 * If we don't do this there is a chance ndo_poll_controller
1237 * or ndo_poll may be running while we open the device
1239 ret
= netpoll_rx_disable(dev
);
1243 ret
= call_netdevice_notifiers(NETDEV_PRE_UP
, dev
);
1244 ret
= notifier_to_errno(ret
);
1248 set_bit(__LINK_STATE_START
, &dev
->state
);
1250 if (ops
->ndo_validate_addr
)
1251 ret
= ops
->ndo_validate_addr(dev
);
1253 if (!ret
&& ops
->ndo_open
)
1254 ret
= ops
->ndo_open(dev
);
1256 netpoll_rx_enable(dev
);
1259 clear_bit(__LINK_STATE_START
, &dev
->state
);
1261 dev
->flags
|= IFF_UP
;
1262 net_dmaengine_get();
1263 dev_set_rx_mode(dev
);
1265 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
1272 * dev_open - prepare an interface for use.
1273 * @dev: device to open
1275 * Takes a device from down to up state. The device's private open
1276 * function is invoked and then the multicast lists are loaded. Finally
1277 * the device is moved into the up state and a %NETDEV_UP message is
1278 * sent to the netdev notifier chain.
1280 * Calling this function on an active interface is a nop. On a failure
1281 * a negative errno code is returned.
1283 int dev_open(struct net_device
*dev
)
1287 if (dev
->flags
& IFF_UP
)
1290 ret
= __dev_open(dev
);
1294 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1295 call_netdevice_notifiers(NETDEV_UP
, dev
);
1299 EXPORT_SYMBOL(dev_open
);
1301 static int __dev_close_many(struct list_head
*head
)
1303 struct net_device
*dev
;
1308 list_for_each_entry(dev
, head
, unreg_list
) {
1309 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1311 clear_bit(__LINK_STATE_START
, &dev
->state
);
1313 /* Synchronize to scheduled poll. We cannot touch poll list, it
1314 * can be even on different cpu. So just clear netif_running().
1316 * dev->stop() will invoke napi_disable() on all of it's
1317 * napi_struct instances on this device.
1319 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1322 dev_deactivate_many(head
);
1324 list_for_each_entry(dev
, head
, unreg_list
) {
1325 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1328 * Call the device specific close. This cannot fail.
1329 * Only if device is UP
1331 * We allow it to be called even after a DETACH hot-plug
1337 dev
->flags
&= ~IFF_UP
;
1338 net_dmaengine_put();
1344 static int __dev_close(struct net_device
*dev
)
1349 /* Temporarily disable netpoll until the interface is down */
1350 retval
= netpoll_rx_disable(dev
);
1354 list_add(&dev
->unreg_list
, &single
);
1355 retval
= __dev_close_many(&single
);
1358 netpoll_rx_enable(dev
);
1362 static int dev_close_many(struct list_head
*head
)
1364 struct net_device
*dev
, *tmp
;
1365 LIST_HEAD(tmp_list
);
1367 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
)
1368 if (!(dev
->flags
& IFF_UP
))
1369 list_move(&dev
->unreg_list
, &tmp_list
);
1371 __dev_close_many(head
);
1373 list_for_each_entry(dev
, head
, unreg_list
) {
1374 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1375 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1378 /* rollback_registered_many needs the complete original list */
1379 list_splice(&tmp_list
, head
);
1384 * dev_close - shutdown an interface.
1385 * @dev: device to shutdown
1387 * This function moves an active device into down state. A
1388 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1389 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1392 int dev_close(struct net_device
*dev
)
1395 if (dev
->flags
& IFF_UP
) {
1398 /* Block netpoll rx while the interface is going down */
1399 ret
= netpoll_rx_disable(dev
);
1403 list_add(&dev
->unreg_list
, &single
);
1404 dev_close_many(&single
);
1407 netpoll_rx_enable(dev
);
1411 EXPORT_SYMBOL(dev_close
);
1415 * dev_disable_lro - disable Large Receive Offload on a device
1418 * Disable Large Receive Offload (LRO) on a net device. Must be
1419 * called under RTNL. This is needed if received packets may be
1420 * forwarded to another interface.
1422 void dev_disable_lro(struct net_device
*dev
)
1425 * If we're trying to disable lro on a vlan device
1426 * use the underlying physical device instead
1428 if (is_vlan_dev(dev
))
1429 dev
= vlan_dev_real_dev(dev
);
1431 dev
->wanted_features
&= ~NETIF_F_LRO
;
1432 netdev_update_features(dev
);
1434 if (unlikely(dev
->features
& NETIF_F_LRO
))
1435 netdev_WARN(dev
, "failed to disable LRO!\n");
1437 EXPORT_SYMBOL(dev_disable_lro
);
1440 static int dev_boot_phase
= 1;
1443 * register_netdevice_notifier - register a network notifier block
1446 * Register a notifier to be called when network device events occur.
1447 * The notifier passed is linked into the kernel structures and must
1448 * not be reused until it has been unregistered. A negative errno code
1449 * is returned on a failure.
1451 * When registered all registration and up events are replayed
1452 * to the new notifier to allow device to have a race free
1453 * view of the network device list.
1456 int register_netdevice_notifier(struct notifier_block
*nb
)
1458 struct net_device
*dev
;
1459 struct net_device
*last
;
1464 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1470 for_each_netdev(net
, dev
) {
1471 err
= nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1472 err
= notifier_to_errno(err
);
1476 if (!(dev
->flags
& IFF_UP
))
1479 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1490 for_each_netdev(net
, dev
) {
1494 if (dev
->flags
& IFF_UP
) {
1495 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1496 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1498 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1503 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1506 EXPORT_SYMBOL(register_netdevice_notifier
);
1509 * unregister_netdevice_notifier - unregister a network notifier block
1512 * Unregister a notifier previously registered by
1513 * register_netdevice_notifier(). The notifier is unlinked into the
1514 * kernel structures and may then be reused. A negative errno code
1515 * is returned on a failure.
1517 * After unregistering unregister and down device events are synthesized
1518 * for all devices on the device list to the removed notifier to remove
1519 * the need for special case cleanup code.
1522 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1524 struct net_device
*dev
;
1529 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1534 for_each_netdev(net
, dev
) {
1535 if (dev
->flags
& IFF_UP
) {
1536 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1537 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1539 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1546 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1549 * call_netdevice_notifiers - call all network notifier blocks
1550 * @val: value passed unmodified to notifier function
1551 * @dev: net_device pointer passed unmodified to notifier function
1553 * Call all network notifier blocks. Parameters and return value
1554 * are as for raw_notifier_call_chain().
1557 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1560 return raw_notifier_call_chain(&netdev_chain
, val
, dev
);
1562 EXPORT_SYMBOL(call_netdevice_notifiers
);
1564 static struct static_key netstamp_needed __read_mostly
;
1565 #ifdef HAVE_JUMP_LABEL
1566 /* We are not allowed to call static_key_slow_dec() from irq context
1567 * If net_disable_timestamp() is called from irq context, defer the
1568 * static_key_slow_dec() calls.
1570 static atomic_t netstamp_needed_deferred
;
1573 void net_enable_timestamp(void)
1575 #ifdef HAVE_JUMP_LABEL
1576 int deferred
= atomic_xchg(&netstamp_needed_deferred
, 0);
1580 static_key_slow_dec(&netstamp_needed
);
1584 static_key_slow_inc(&netstamp_needed
);
1586 EXPORT_SYMBOL(net_enable_timestamp
);
1588 void net_disable_timestamp(void)
1590 #ifdef HAVE_JUMP_LABEL
1591 if (in_interrupt()) {
1592 atomic_inc(&netstamp_needed_deferred
);
1596 static_key_slow_dec(&netstamp_needed
);
1598 EXPORT_SYMBOL(net_disable_timestamp
);
1600 static inline void net_timestamp_set(struct sk_buff
*skb
)
1602 skb
->tstamp
.tv64
= 0;
1603 if (static_key_false(&netstamp_needed
))
1604 __net_timestamp(skb
);
1607 #define net_timestamp_check(COND, SKB) \
1608 if (static_key_false(&netstamp_needed)) { \
1609 if ((COND) && !(SKB)->tstamp.tv64) \
1610 __net_timestamp(SKB); \
1613 static inline bool is_skb_forwardable(struct net_device *dev,
1614 struct sk_buff
*skb
)
1618 if (!(dev
->flags
& IFF_UP
))
1621 len
= dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
;
1622 if (skb
->len
<= len
)
1625 /* if TSO is enabled, we don't care about the length as the packet
1626 * could be forwarded without being segmented before
1628 if (skb_is_gso(skb
))
1635 * dev_forward_skb - loopback an skb to another netif
1637 * @dev: destination network device
1638 * @skb: buffer to forward
1641 * NET_RX_SUCCESS (no congestion)
1642 * NET_RX_DROP (packet was dropped, but freed)
1644 * dev_forward_skb can be used for injecting an skb from the
1645 * start_xmit function of one device into the receive queue
1646 * of another device.
1648 * The receiving device may be in another namespace, so
1649 * we have to clear all information in the skb that could
1650 * impact namespace isolation.
1652 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1654 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
1655 if (skb_copy_ubufs(skb
, GFP_ATOMIC
)) {
1656 atomic_long_inc(&dev
->rx_dropped
);
1664 if (unlikely(!is_skb_forwardable(dev
, skb
))) {
1665 atomic_long_inc(&dev
->rx_dropped
);
1672 skb
->tstamp
.tv64
= 0;
1673 skb
->pkt_type
= PACKET_HOST
;
1674 skb
->protocol
= eth_type_trans(skb
, dev
);
1678 nf_reset_trace(skb
);
1679 return netif_rx(skb
);
1681 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1683 static inline int deliver_skb(struct sk_buff
*skb
,
1684 struct packet_type
*pt_prev
,
1685 struct net_device
*orig_dev
)
1687 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
1689 atomic_inc(&skb
->users
);
1690 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1693 static inline bool skb_loop_sk(struct packet_type
*ptype
, struct sk_buff
*skb
)
1695 if (!ptype
->af_packet_priv
|| !skb
->sk
)
1698 if (ptype
->id_match
)
1699 return ptype
->id_match(ptype
, skb
->sk
);
1700 else if ((struct sock
*)ptype
->af_packet_priv
== skb
->sk
)
1707 * Support routine. Sends outgoing frames to any network
1708 * taps currently in use.
1711 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1713 struct packet_type
*ptype
;
1714 struct sk_buff
*skb2
= NULL
;
1715 struct packet_type
*pt_prev
= NULL
;
1718 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1719 /* Never send packets back to the socket
1720 * they originated from - MvS (miquels@drinkel.ow.org)
1722 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1723 (!skb_loop_sk(ptype
, skb
))) {
1725 deliver_skb(skb2
, pt_prev
, skb
->dev
);
1730 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1734 net_timestamp_set(skb2
);
1736 /* skb->nh should be correctly
1737 set by sender, so that the second statement is
1738 just protection against buggy protocols.
1740 skb_reset_mac_header(skb2
);
1742 if (skb_network_header(skb2
) < skb2
->data
||
1743 skb2
->network_header
> skb2
->tail
) {
1744 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1745 ntohs(skb2
->protocol
),
1747 skb_reset_network_header(skb2
);
1750 skb2
->transport_header
= skb2
->network_header
;
1751 skb2
->pkt_type
= PACKET_OUTGOING
;
1756 pt_prev
->func(skb2
, skb
->dev
, pt_prev
, skb
->dev
);
1761 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1762 * @dev: Network device
1763 * @txq: number of queues available
1765 * If real_num_tx_queues is changed the tc mappings may no longer be
1766 * valid. To resolve this verify the tc mapping remains valid and if
1767 * not NULL the mapping. With no priorities mapping to this
1768 * offset/count pair it will no longer be used. In the worst case TC0
1769 * is invalid nothing can be done so disable priority mappings. If is
1770 * expected that drivers will fix this mapping if they can before
1771 * calling netif_set_real_num_tx_queues.
1773 static void netif_setup_tc(struct net_device
*dev
, unsigned int txq
)
1776 struct netdev_tc_txq
*tc
= &dev
->tc_to_txq
[0];
1778 /* If TC0 is invalidated disable TC mapping */
1779 if (tc
->offset
+ tc
->count
> txq
) {
1780 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1785 /* Invalidated prio to tc mappings set to TC0 */
1786 for (i
= 1; i
< TC_BITMASK
+ 1; i
++) {
1787 int q
= netdev_get_prio_tc_map(dev
, i
);
1789 tc
= &dev
->tc_to_txq
[q
];
1790 if (tc
->offset
+ tc
->count
> txq
) {
1791 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1793 netdev_set_prio_tc_map(dev
, i
, 0);
1799 static DEFINE_MUTEX(xps_map_mutex
);
1800 #define xmap_dereference(P) \
1801 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1803 static struct xps_map
*remove_xps_queue(struct xps_dev_maps
*dev_maps
,
1806 struct xps_map
*map
= NULL
;
1810 map
= xmap_dereference(dev_maps
->cpu_map
[cpu
]);
1812 for (pos
= 0; map
&& pos
< map
->len
; pos
++) {
1813 if (map
->queues
[pos
] == index
) {
1815 map
->queues
[pos
] = map
->queues
[--map
->len
];
1817 RCU_INIT_POINTER(dev_maps
->cpu_map
[cpu
], NULL
);
1818 kfree_rcu(map
, rcu
);
1828 static void netif_reset_xps_queues_gt(struct net_device
*dev
, u16 index
)
1830 struct xps_dev_maps
*dev_maps
;
1832 bool active
= false;
1834 mutex_lock(&xps_map_mutex
);
1835 dev_maps
= xmap_dereference(dev
->xps_maps
);
1840 for_each_possible_cpu(cpu
) {
1841 for (i
= index
; i
< dev
->num_tx_queues
; i
++) {
1842 if (!remove_xps_queue(dev_maps
, cpu
, i
))
1845 if (i
== dev
->num_tx_queues
)
1850 RCU_INIT_POINTER(dev
->xps_maps
, NULL
);
1851 kfree_rcu(dev_maps
, rcu
);
1854 for (i
= index
; i
< dev
->num_tx_queues
; i
++)
1855 netdev_queue_numa_node_write(netdev_get_tx_queue(dev
, i
),
1859 mutex_unlock(&xps_map_mutex
);
1862 static struct xps_map
*expand_xps_map(struct xps_map
*map
,
1865 struct xps_map
*new_map
;
1866 int alloc_len
= XPS_MIN_MAP_ALLOC
;
1869 for (pos
= 0; map
&& pos
< map
->len
; pos
++) {
1870 if (map
->queues
[pos
] != index
)
1875 /* Need to add queue to this CPU's existing map */
1877 if (pos
< map
->alloc_len
)
1880 alloc_len
= map
->alloc_len
* 2;
1883 /* Need to allocate new map to store queue on this CPU's map */
1884 new_map
= kzalloc_node(XPS_MAP_SIZE(alloc_len
), GFP_KERNEL
,
1889 for (i
= 0; i
< pos
; i
++)
1890 new_map
->queues
[i
] = map
->queues
[i
];
1891 new_map
->alloc_len
= alloc_len
;
1897 int netif_set_xps_queue(struct net_device
*dev
, struct cpumask
*mask
, u16 index
)
1899 struct xps_dev_maps
*dev_maps
, *new_dev_maps
= NULL
;
1900 struct xps_map
*map
, *new_map
;
1901 int maps_sz
= max_t(unsigned int, XPS_DEV_MAPS_SIZE
, L1_CACHE_BYTES
);
1902 int cpu
, numa_node_id
= -2;
1903 bool active
= false;
1905 mutex_lock(&xps_map_mutex
);
1907 dev_maps
= xmap_dereference(dev
->xps_maps
);
1909 /* allocate memory for queue storage */
1910 for_each_online_cpu(cpu
) {
1911 if (!cpumask_test_cpu(cpu
, mask
))
1915 new_dev_maps
= kzalloc(maps_sz
, GFP_KERNEL
);
1916 if (!new_dev_maps
) {
1917 mutex_unlock(&xps_map_mutex
);
1921 map
= dev_maps
? xmap_dereference(dev_maps
->cpu_map
[cpu
]) :
1924 map
= expand_xps_map(map
, cpu
, index
);
1928 RCU_INIT_POINTER(new_dev_maps
->cpu_map
[cpu
], map
);
1932 goto out_no_new_maps
;
1934 for_each_possible_cpu(cpu
) {
1935 if (cpumask_test_cpu(cpu
, mask
) && cpu_online(cpu
)) {
1936 /* add queue to CPU maps */
1939 map
= xmap_dereference(new_dev_maps
->cpu_map
[cpu
]);
1940 while ((pos
< map
->len
) && (map
->queues
[pos
] != index
))
1943 if (pos
== map
->len
)
1944 map
->queues
[map
->len
++] = index
;
1946 if (numa_node_id
== -2)
1947 numa_node_id
= cpu_to_node(cpu
);
1948 else if (numa_node_id
!= cpu_to_node(cpu
))
1951 } else if (dev_maps
) {
1952 /* fill in the new device map from the old device map */
1953 map
= xmap_dereference(dev_maps
->cpu_map
[cpu
]);
1954 RCU_INIT_POINTER(new_dev_maps
->cpu_map
[cpu
], map
);
1959 rcu_assign_pointer(dev
->xps_maps
, new_dev_maps
);
1961 /* Cleanup old maps */
1963 for_each_possible_cpu(cpu
) {
1964 new_map
= xmap_dereference(new_dev_maps
->cpu_map
[cpu
]);
1965 map
= xmap_dereference(dev_maps
->cpu_map
[cpu
]);
1966 if (map
&& map
!= new_map
)
1967 kfree_rcu(map
, rcu
);
1970 kfree_rcu(dev_maps
, rcu
);
1973 dev_maps
= new_dev_maps
;
1977 /* update Tx queue numa node */
1978 netdev_queue_numa_node_write(netdev_get_tx_queue(dev
, index
),
1979 (numa_node_id
>= 0) ? numa_node_id
:
1985 /* removes queue from unused CPUs */
1986 for_each_possible_cpu(cpu
) {
1987 if (cpumask_test_cpu(cpu
, mask
) && cpu_online(cpu
))
1990 if (remove_xps_queue(dev_maps
, cpu
, index
))
1994 /* free map if not active */
1996 RCU_INIT_POINTER(dev
->xps_maps
, NULL
);
1997 kfree_rcu(dev_maps
, rcu
);
2001 mutex_unlock(&xps_map_mutex
);
2005 /* remove any maps that we added */
2006 for_each_possible_cpu(cpu
) {
2007 new_map
= xmap_dereference(new_dev_maps
->cpu_map
[cpu
]);
2008 map
= dev_maps
? xmap_dereference(dev_maps
->cpu_map
[cpu
]) :
2010 if (new_map
&& new_map
!= map
)
2014 mutex_unlock(&xps_map_mutex
);
2016 kfree(new_dev_maps
);
2019 EXPORT_SYMBOL(netif_set_xps_queue
);
2023 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2024 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2026 int netif_set_real_num_tx_queues(struct net_device
*dev
, unsigned int txq
)
2030 if (txq
< 1 || txq
> dev
->num_tx_queues
)
2033 if (dev
->reg_state
== NETREG_REGISTERED
||
2034 dev
->reg_state
== NETREG_UNREGISTERING
) {
2037 rc
= netdev_queue_update_kobjects(dev
, dev
->real_num_tx_queues
,
2043 netif_setup_tc(dev
, txq
);
2045 if (txq
< dev
->real_num_tx_queues
) {
2046 qdisc_reset_all_tx_gt(dev
, txq
);
2048 netif_reset_xps_queues_gt(dev
, txq
);
2053 dev
->real_num_tx_queues
= txq
;
2056 EXPORT_SYMBOL(netif_set_real_num_tx_queues
);
2060 * netif_set_real_num_rx_queues - set actual number of RX queues used
2061 * @dev: Network device
2062 * @rxq: Actual number of RX queues
2064 * This must be called either with the rtnl_lock held or before
2065 * registration of the net device. Returns 0 on success, or a
2066 * negative error code. If called before registration, it always
2069 int netif_set_real_num_rx_queues(struct net_device
*dev
, unsigned int rxq
)
2073 if (rxq
< 1 || rxq
> dev
->num_rx_queues
)
2076 if (dev
->reg_state
== NETREG_REGISTERED
) {
2079 rc
= net_rx_queue_update_kobjects(dev
, dev
->real_num_rx_queues
,
2085 dev
->real_num_rx_queues
= rxq
;
2088 EXPORT_SYMBOL(netif_set_real_num_rx_queues
);
2092 * netif_get_num_default_rss_queues - default number of RSS queues
2094 * This routine should set an upper limit on the number of RSS queues
2095 * used by default by multiqueue devices.
2097 int netif_get_num_default_rss_queues(void)
2099 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES
, num_online_cpus());
2101 EXPORT_SYMBOL(netif_get_num_default_rss_queues
);
2103 static inline void __netif_reschedule(struct Qdisc
*q
)
2105 struct softnet_data
*sd
;
2106 unsigned long flags
;
2108 local_irq_save(flags
);
2109 sd
= &__get_cpu_var(softnet_data
);
2110 q
->next_sched
= NULL
;
2111 *sd
->output_queue_tailp
= q
;
2112 sd
->output_queue_tailp
= &q
->next_sched
;
2113 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
2114 local_irq_restore(flags
);
2117 void __netif_schedule(struct Qdisc
*q
)
2119 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
2120 __netif_reschedule(q
);
2122 EXPORT_SYMBOL(__netif_schedule
);
2124 void dev_kfree_skb_irq(struct sk_buff
*skb
)
2126 if (atomic_dec_and_test(&skb
->users
)) {
2127 struct softnet_data
*sd
;
2128 unsigned long flags
;
2130 local_irq_save(flags
);
2131 sd
= &__get_cpu_var(softnet_data
);
2132 skb
->next
= sd
->completion_queue
;
2133 sd
->completion_queue
= skb
;
2134 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
2135 local_irq_restore(flags
);
2138 EXPORT_SYMBOL(dev_kfree_skb_irq
);
2140 void dev_kfree_skb_any(struct sk_buff
*skb
)
2142 if (in_irq() || irqs_disabled())
2143 dev_kfree_skb_irq(skb
);
2147 EXPORT_SYMBOL(dev_kfree_skb_any
);
2151 * netif_device_detach - mark device as removed
2152 * @dev: network device
2154 * Mark device as removed from system and therefore no longer available.
2156 void netif_device_detach(struct net_device
*dev
)
2158 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
2159 netif_running(dev
)) {
2160 netif_tx_stop_all_queues(dev
);
2163 EXPORT_SYMBOL(netif_device_detach
);
2166 * netif_device_attach - mark device as attached
2167 * @dev: network device
2169 * Mark device as attached from system and restart if needed.
2171 void netif_device_attach(struct net_device
*dev
)
2173 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
2174 netif_running(dev
)) {
2175 netif_tx_wake_all_queues(dev
);
2176 __netdev_watchdog_up(dev
);
2179 EXPORT_SYMBOL(netif_device_attach
);
2181 static void skb_warn_bad_offload(const struct sk_buff
*skb
)
2183 static const netdev_features_t null_features
= 0;
2184 struct net_device
*dev
= skb
->dev
;
2185 const char *driver
= "";
2187 if (!net_ratelimit())
2190 if (dev
&& dev
->dev
.parent
)
2191 driver
= dev_driver_string(dev
->dev
.parent
);
2193 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2194 "gso_type=%d ip_summed=%d\n",
2195 driver
, dev
? &dev
->features
: &null_features
,
2196 skb
->sk
? &skb
->sk
->sk_route_caps
: &null_features
,
2197 skb
->len
, skb
->data_len
, skb_shinfo(skb
)->gso_size
,
2198 skb_shinfo(skb
)->gso_type
, skb
->ip_summed
);
2202 * Invalidate hardware checksum when packet is to be mangled, and
2203 * complete checksum manually on outgoing path.
2205 int skb_checksum_help(struct sk_buff
*skb
)
2208 int ret
= 0, offset
;
2210 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
2211 goto out_set_summed
;
2213 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
2214 skb_warn_bad_offload(skb
);
2218 /* Before computing a checksum, we should make sure no frag could
2219 * be modified by an external entity : checksum could be wrong.
2221 if (skb_has_shared_frag(skb
)) {
2222 ret
= __skb_linearize(skb
);
2227 offset
= skb_checksum_start_offset(skb
);
2228 BUG_ON(offset
>= skb_headlen(skb
));
2229 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
2231 offset
+= skb
->csum_offset
;
2232 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
2234 if (skb_cloned(skb
) &&
2235 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
2236 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2241 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
2243 skb
->ip_summed
= CHECKSUM_NONE
;
2247 EXPORT_SYMBOL(skb_checksum_help
);
2249 __be16
skb_network_protocol(struct sk_buff
*skb
)
2251 __be16 type
= skb
->protocol
;
2252 int vlan_depth
= ETH_HLEN
;
2254 /* Tunnel gso handlers can set protocol to ethernet. */
2255 if (type
== htons(ETH_P_TEB
)) {
2258 if (unlikely(!pskb_may_pull(skb
, sizeof(struct ethhdr
))))
2261 eth
= (struct ethhdr
*)skb_mac_header(skb
);
2262 type
= eth
->h_proto
;
2265 while (type
== htons(ETH_P_8021Q
) || type
== htons(ETH_P_8021AD
)) {
2266 struct vlan_hdr
*vh
;
2268 if (unlikely(!pskb_may_pull(skb
, vlan_depth
+ VLAN_HLEN
)))
2271 vh
= (struct vlan_hdr
*)(skb
->data
+ vlan_depth
);
2272 type
= vh
->h_vlan_encapsulated_proto
;
2273 vlan_depth
+= VLAN_HLEN
;
2280 * skb_mac_gso_segment - mac layer segmentation handler.
2281 * @skb: buffer to segment
2282 * @features: features for the output path (see dev->features)
2284 struct sk_buff
*skb_mac_gso_segment(struct sk_buff
*skb
,
2285 netdev_features_t features
)
2287 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
2288 struct packet_offload
*ptype
;
2289 __be16 type
= skb_network_protocol(skb
);
2291 if (unlikely(!type
))
2292 return ERR_PTR(-EINVAL
);
2294 __skb_pull(skb
, skb
->mac_len
);
2297 list_for_each_entry_rcu(ptype
, &offload_base
, list
) {
2298 if (ptype
->type
== type
&& ptype
->callbacks
.gso_segment
) {
2299 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
2302 err
= ptype
->callbacks
.gso_send_check(skb
);
2303 segs
= ERR_PTR(err
);
2304 if (err
|| skb_gso_ok(skb
, features
))
2306 __skb_push(skb
, (skb
->data
-
2307 skb_network_header(skb
)));
2309 segs
= ptype
->callbacks
.gso_segment(skb
, features
);
2315 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
2319 EXPORT_SYMBOL(skb_mac_gso_segment
);
2322 /* openvswitch calls this on rx path, so we need a different check.
2324 static inline bool skb_needs_check(struct sk_buff
*skb
, bool tx_path
)
2327 return skb
->ip_summed
!= CHECKSUM_PARTIAL
;
2329 return skb
->ip_summed
== CHECKSUM_NONE
;
2333 * __skb_gso_segment - Perform segmentation on skb.
2334 * @skb: buffer to segment
2335 * @features: features for the output path (see dev->features)
2336 * @tx_path: whether it is called in TX path
2338 * This function segments the given skb and returns a list of segments.
2340 * It may return NULL if the skb requires no segmentation. This is
2341 * only possible when GSO is used for verifying header integrity.
2343 struct sk_buff
*__skb_gso_segment(struct sk_buff
*skb
,
2344 netdev_features_t features
, bool tx_path
)
2346 if (unlikely(skb_needs_check(skb
, tx_path
))) {
2349 skb_warn_bad_offload(skb
);
2351 if (skb_header_cloned(skb
) &&
2352 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
2353 return ERR_PTR(err
);
2356 SKB_GSO_CB(skb
)->mac_offset
= skb_headroom(skb
);
2357 skb_reset_mac_header(skb
);
2358 skb_reset_mac_len(skb
);
2360 return skb_mac_gso_segment(skb
, features
);
2362 EXPORT_SYMBOL(__skb_gso_segment
);
2364 /* Take action when hardware reception checksum errors are detected. */
2366 void netdev_rx_csum_fault(struct net_device
*dev
)
2368 if (net_ratelimit()) {
2369 pr_err("%s: hw csum failure\n", dev
? dev
->name
: "<unknown>");
2373 EXPORT_SYMBOL(netdev_rx_csum_fault
);
2376 /* Actually, we should eliminate this check as soon as we know, that:
2377 * 1. IOMMU is present and allows to map all the memory.
2378 * 2. No high memory really exists on this machine.
2381 static int illegal_highdma(const struct net_device
*dev
, struct sk_buff
*skb
)
2383 #ifdef CONFIG_HIGHMEM
2385 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
2386 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2387 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2388 if (PageHighMem(skb_frag_page(frag
)))
2393 if (PCI_DMA_BUS_IS_PHYS
) {
2394 struct device
*pdev
= dev
->dev
.parent
;
2398 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2399 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2400 dma_addr_t addr
= page_to_phys(skb_frag_page(frag
));
2401 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
2410 void (*destructor
)(struct sk_buff
*skb
);
2413 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2415 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
2417 struct dev_gso_cb
*cb
;
2420 struct sk_buff
*nskb
= skb
->next
;
2422 skb
->next
= nskb
->next
;
2425 } while (skb
->next
);
2427 cb
= DEV_GSO_CB(skb
);
2429 cb
->destructor(skb
);
2433 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2434 * @skb: buffer to segment
2435 * @features: device features as applicable to this skb
2437 * This function segments the given skb and stores the list of segments
2440 static int dev_gso_segment(struct sk_buff
*skb
, netdev_features_t features
)
2442 struct sk_buff
*segs
;
2444 segs
= skb_gso_segment(skb
, features
);
2446 /* Verifying header integrity only. */
2451 return PTR_ERR(segs
);
2454 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
2455 skb
->destructor
= dev_gso_skb_destructor
;
2460 static netdev_features_t
harmonize_features(struct sk_buff
*skb
,
2462 const struct net_device
*dev
,
2463 netdev_features_t features
)
2465 if (skb
->ip_summed
!= CHECKSUM_NONE
&&
2466 !can_checksum_protocol(features
, protocol
)) {
2467 features
&= ~NETIF_F_ALL_CSUM
;
2468 } else if (illegal_highdma(dev
, skb
)) {
2469 features
&= ~NETIF_F_SG
;
2475 netdev_features_t
netif_skb_dev_features(struct sk_buff
*skb
,
2476 const struct net_device
*dev
)
2478 __be16 protocol
= skb
->protocol
;
2479 netdev_features_t features
= dev
->features
;
2481 if (skb_shinfo(skb
)->gso_segs
> dev
->gso_max_segs
)
2482 features
&= ~NETIF_F_GSO_MASK
;
2484 if (protocol
== htons(ETH_P_8021Q
) || protocol
== htons(ETH_P_8021AD
)) {
2485 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
2486 protocol
= veh
->h_vlan_encapsulated_proto
;
2487 } else if (!vlan_tx_tag_present(skb
)) {
2488 return harmonize_features(skb
, protocol
, dev
, features
);
2491 features
&= (dev
->vlan_features
| NETIF_F_HW_VLAN_CTAG_TX
|
2492 NETIF_F_HW_VLAN_STAG_TX
);
2494 if (protocol
!= htons(ETH_P_8021Q
) && protocol
!= htons(ETH_P_8021AD
)) {
2495 return harmonize_features(skb
, protocol
, dev
, features
);
2497 features
&= NETIF_F_SG
| NETIF_F_HIGHDMA
| NETIF_F_FRAGLIST
|
2498 NETIF_F_GEN_CSUM
| NETIF_F_HW_VLAN_CTAG_TX
|
2499 NETIF_F_HW_VLAN_STAG_TX
;
2500 return harmonize_features(skb
, protocol
, dev
, features
);
2503 return harmonize_features(skb
, protocol
, dev
, features
);
2505 EXPORT_SYMBOL(netif_skb_dev_features
);
2508 * Returns true if either:
2509 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2510 * 2. skb is fragmented and the device does not support SG.
2512 static inline int skb_needs_linearize(struct sk_buff
*skb
,
2513 netdev_features_t features
)
2515 return skb_is_nonlinear(skb
) &&
2516 ((skb_has_frag_list(skb
) &&
2517 !(features
& NETIF_F_FRAGLIST
)) ||
2518 (skb_shinfo(skb
)->nr_frags
&&
2519 !(features
& NETIF_F_SG
)));
2522 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
2523 struct netdev_queue
*txq
)
2525 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2526 int rc
= NETDEV_TX_OK
;
2527 unsigned int skb_len
;
2529 if (likely(!skb
->next
)) {
2530 netdev_features_t features
;
2533 * If device doesn't need skb->dst, release it right now while
2534 * its hot in this cpu cache
2536 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2539 features
= netif_skb_features(skb
);
2541 if (vlan_tx_tag_present(skb
) &&
2542 !vlan_hw_offload_capable(features
, skb
->vlan_proto
)) {
2543 skb
= __vlan_put_tag(skb
, skb
->vlan_proto
,
2544 vlan_tx_tag_get(skb
));
2551 /* If encapsulation offload request, verify we are testing
2552 * hardware encapsulation features instead of standard
2553 * features for the netdev
2555 if (skb
->encapsulation
)
2556 features
&= dev
->hw_enc_features
;
2558 if (netif_needs_gso(skb
, features
)) {
2559 if (unlikely(dev_gso_segment(skb
, features
)))
2564 if (skb_needs_linearize(skb
, features
) &&
2565 __skb_linearize(skb
))
2568 /* If packet is not checksummed and device does not
2569 * support checksumming for this protocol, complete
2570 * checksumming here.
2572 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2573 if (skb
->encapsulation
)
2574 skb_set_inner_transport_header(skb
,
2575 skb_checksum_start_offset(skb
));
2577 skb_set_transport_header(skb
,
2578 skb_checksum_start_offset(skb
));
2579 if (!(features
& NETIF_F_ALL_CSUM
) &&
2580 skb_checksum_help(skb
))
2585 if (!list_empty(&ptype_all
))
2586 dev_queue_xmit_nit(skb
, dev
);
2589 rc
= ops
->ndo_start_xmit(skb
, dev
);
2590 trace_net_dev_xmit(skb
, rc
, dev
, skb_len
);
2591 if (rc
== NETDEV_TX_OK
)
2592 txq_trans_update(txq
);
2598 struct sk_buff
*nskb
= skb
->next
;
2600 skb
->next
= nskb
->next
;
2603 if (!list_empty(&ptype_all
))
2604 dev_queue_xmit_nit(nskb
, dev
);
2606 skb_len
= nskb
->len
;
2607 rc
= ops
->ndo_start_xmit(nskb
, dev
);
2608 trace_net_dev_xmit(nskb
, rc
, dev
, skb_len
);
2609 if (unlikely(rc
!= NETDEV_TX_OK
)) {
2610 if (rc
& ~NETDEV_TX_MASK
)
2611 goto out_kfree_gso_skb
;
2612 nskb
->next
= skb
->next
;
2616 txq_trans_update(txq
);
2617 if (unlikely(netif_xmit_stopped(txq
) && skb
->next
))
2618 return NETDEV_TX_BUSY
;
2619 } while (skb
->next
);
2622 if (likely(skb
->next
== NULL
)) {
2623 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
2633 static void qdisc_pkt_len_init(struct sk_buff
*skb
)
2635 const struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
2637 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
2639 /* To get more precise estimation of bytes sent on wire,
2640 * we add to pkt_len the headers size of all segments
2642 if (shinfo
->gso_size
) {
2643 unsigned int hdr_len
;
2644 u16 gso_segs
= shinfo
->gso_segs
;
2646 /* mac layer + network layer */
2647 hdr_len
= skb_transport_header(skb
) - skb_mac_header(skb
);
2649 /* + transport layer */
2650 if (likely(shinfo
->gso_type
& (SKB_GSO_TCPV4
| SKB_GSO_TCPV6
)))
2651 hdr_len
+= tcp_hdrlen(skb
);
2653 hdr_len
+= sizeof(struct udphdr
);
2655 if (shinfo
->gso_type
& SKB_GSO_DODGY
)
2656 gso_segs
= DIV_ROUND_UP(skb
->len
- hdr_len
,
2659 qdisc_skb_cb(skb
)->pkt_len
+= (gso_segs
- 1) * hdr_len
;
2663 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2664 struct net_device
*dev
,
2665 struct netdev_queue
*txq
)
2667 spinlock_t
*root_lock
= qdisc_lock(q
);
2671 qdisc_pkt_len_init(skb
);
2672 qdisc_calculate_pkt_len(skb
, q
);
2674 * Heuristic to force contended enqueues to serialize on a
2675 * separate lock before trying to get qdisc main lock.
2676 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2677 * and dequeue packets faster.
2679 contended
= qdisc_is_running(q
);
2680 if (unlikely(contended
))
2681 spin_lock(&q
->busylock
);
2683 spin_lock(root_lock
);
2684 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2685 printk(KERN_WARNING
"[mtk_net]__dev_xmit_skb drop skb_len = %d \n", skb
->len
);
2688 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2689 qdisc_run_begin(q
)) {
2691 * This is a work-conserving queue; there are no old skbs
2692 * waiting to be sent out; and the qdisc is not running -
2693 * xmit the skb directly.
2695 if (!(dev
->priv_flags
& IFF_XMIT_DST_RELEASE
))
2698 qdisc_bstats_update(q
, skb
);
2700 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
)) {
2701 if (unlikely(contended
)) {
2702 spin_unlock(&q
->busylock
);
2709 rc
= NET_XMIT_SUCCESS
;
2712 rc
= q
->enqueue(skb
, q
) & NET_XMIT_MASK
;
2713 if (qdisc_run_begin(q
)) {
2714 if (unlikely(contended
)) {
2715 spin_unlock(&q
->busylock
);
2721 spin_unlock(root_lock
);
2722 if (unlikely(contended
))
2723 spin_unlock(&q
->busylock
);
2727 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2728 static void skb_update_prio(struct sk_buff
*skb
)
2730 struct netprio_map
*map
= rcu_dereference_bh(skb
->dev
->priomap
);
2732 if (!skb
->priority
&& skb
->sk
&& map
) {
2733 unsigned int prioidx
= skb
->sk
->sk_cgrp_prioidx
;
2735 if (prioidx
< map
->priomap_len
)
2736 skb
->priority
= map
->priomap
[prioidx
];
2740 #define skb_update_prio(skb)
2743 static DEFINE_PER_CPU(int, xmit_recursion
);
2744 #define RECURSION_LIMIT 10
2747 * dev_loopback_xmit - loop back @skb
2748 * @skb: buffer to transmit
2750 int dev_loopback_xmit(struct sk_buff
*skb
)
2752 skb_reset_mac_header(skb
);
2753 __skb_pull(skb
, skb_network_offset(skb
));
2754 skb
->pkt_type
= PACKET_LOOPBACK
;
2755 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
2756 WARN_ON(!skb_dst(skb
));
2761 EXPORT_SYMBOL(dev_loopback_xmit
);
2764 * dev_queue_xmit - transmit a buffer
2765 * @skb: buffer to transmit
2767 * Queue a buffer for transmission to a network device. The caller must
2768 * have set the device and priority and built the buffer before calling
2769 * this function. The function can be called from an interrupt.
2771 * A negative errno code is returned on a failure. A success does not
2772 * guarantee the frame will be transmitted as it may be dropped due
2773 * to congestion or traffic shaping.
2775 * -----------------------------------------------------------------------------------
2776 * I notice this method can also return errors from the queue disciplines,
2777 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2780 * Regardless of the return value, the skb is consumed, so it is currently
2781 * difficult to retry a send to this method. (You can bump the ref count
2782 * before sending to hold a reference for retry if you are careful.)
2784 * When calling this method, interrupts MUST be enabled. This is because
2785 * the BH enable code must have IRQs enabled so that it will not deadlock.
2788 int dev_queue_xmit(struct sk_buff
*skb
)
2790 struct net_device
*dev
= skb
->dev
;
2791 struct netdev_queue
*txq
;
2795 skb_reset_mac_header(skb
);
2799 if (unlikely((sysctl_met_is_enable
== 1) && (sysctl_udp_met_port
> 0)
2800 && (ip_hdr(skb
)->protocol
== IPPROTO_UDP
) && skb
->sk
)) {
2802 if (sysctl_udp_met_port
== ntohs((inet_sk(skb
->sk
))->inet_sport
)) {
2803 struct udphdr
* udp_iphdr
= udp_hdr(skb
);
2804 if (udp_iphdr
&& (ntohs(udp_iphdr
->len
) >= 12)) {
2805 __u16
* seq_id
= (__u16
*)((char *)udp_iphdr
+ 10);
2806 udp_event_trace_printk("F|%d|%s|%d\n", current
->pid
, *seq_id
);
2813 /* Disable soft irqs for various locks below. Also
2814 * stops preemption for RCU.
2818 skb_update_prio(skb
);
2820 txq
= netdev_pick_tx(dev
, skb
);
2821 q
= rcu_dereference_bh(txq
->qdisc
);
2823 #ifdef CONFIG_NET_CLS_ACT
2824 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2826 trace_net_dev_queue(skb
);
2828 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2832 /* The device has no queue. Common case for software devices:
2833 loopback, all the sorts of tunnels...
2835 Really, it is unlikely that netif_tx_lock protection is necessary
2836 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2838 However, it is possible, that they rely on protection
2841 Check this and shot the lock. It is not prone from deadlocks.
2842 Either shot noqueue qdisc, it is even simpler 8)
2844 if (dev
->flags
& IFF_UP
) {
2845 int cpu
= smp_processor_id(); /* ok because BHs are off */
2847 if (txq
->xmit_lock_owner
!= cpu
) {
2849 if (__this_cpu_read(xmit_recursion
) > RECURSION_LIMIT
)
2850 goto recursion_alert
;
2852 HARD_TX_LOCK(dev
, txq
, cpu
);
2854 if (!netif_xmit_stopped(txq
)) {
2855 __this_cpu_inc(xmit_recursion
);
2856 rc
= dev_hard_start_xmit(skb
, dev
, txq
);
2857 __this_cpu_dec(xmit_recursion
);
2858 if (dev_xmit_complete(rc
)) {
2859 HARD_TX_UNLOCK(dev
, txq
);
2863 HARD_TX_UNLOCK(dev
, txq
);
2864 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2867 /* Recursion is detected! It is possible,
2871 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2877 rcu_read_unlock_bh();
2882 rcu_read_unlock_bh();
2885 EXPORT_SYMBOL(dev_queue_xmit
);
2888 /*=======================================================================
2890 =======================================================================*/
2892 int netdev_max_backlog __read_mostly
= 1000;
2893 EXPORT_SYMBOL(netdev_max_backlog
);
2895 int netdev_tstamp_prequeue __read_mostly
= 1;
2896 int netdev_budget __read_mostly
= 300;
2897 int weight_p __read_mostly
= 64; /* old backlog weight */
2899 /* Called with irq disabled */
2900 static inline void ____napi_schedule(struct softnet_data
*sd
,
2901 struct napi_struct
*napi
)
2903 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
2904 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2909 /* One global table that all flow-based protocols share. */
2910 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
2911 EXPORT_SYMBOL(rps_sock_flow_table
);
2913 struct static_key rps_needed __read_mostly
;
2915 static struct rps_dev_flow
*
2916 set_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2917 struct rps_dev_flow
*rflow
, u16 next_cpu
)
2919 if (next_cpu
!= RPS_NO_CPU
) {
2920 #ifdef CONFIG_RFS_ACCEL
2921 struct netdev_rx_queue
*rxqueue
;
2922 struct rps_dev_flow_table
*flow_table
;
2923 struct rps_dev_flow
*old_rflow
;
2928 /* Should we steer this flow to a different hardware queue? */
2929 if (!skb_rx_queue_recorded(skb
) || !dev
->rx_cpu_rmap
||
2930 !(dev
->features
& NETIF_F_NTUPLE
))
2932 rxq_index
= cpu_rmap_lookup_index(dev
->rx_cpu_rmap
, next_cpu
);
2933 if (rxq_index
== skb_get_rx_queue(skb
))
2936 rxqueue
= dev
->_rx
+ rxq_index
;
2937 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2940 flow_id
= skb
->rxhash
& flow_table
->mask
;
2941 rc
= dev
->netdev_ops
->ndo_rx_flow_steer(dev
, skb
,
2942 rxq_index
, flow_id
);
2946 rflow
= &flow_table
->flows
[flow_id
];
2948 if (old_rflow
->filter
== rflow
->filter
)
2949 old_rflow
->filter
= RPS_NO_FILTER
;
2953 per_cpu(softnet_data
, next_cpu
).input_queue_head
;
2956 rflow
->cpu
= next_cpu
;
2961 * get_rps_cpu is called from netif_receive_skb and returns the target
2962 * CPU from the RPS map of the receiving queue for a given skb.
2963 * rcu_read_lock must be held on entry.
2965 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2966 struct rps_dev_flow
**rflowp
)
2968 struct netdev_rx_queue
*rxqueue
;
2969 struct rps_map
*map
;
2970 struct rps_dev_flow_table
*flow_table
;
2971 struct rps_sock_flow_table
*sock_flow_table
;
2975 if (skb_rx_queue_recorded(skb
)) {
2976 u16 index
= skb_get_rx_queue(skb
);
2977 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
2978 WARN_ONCE(dev
->real_num_rx_queues
> 1,
2979 "%s received packet on queue %u, but number "
2980 "of RX queues is %u\n",
2981 dev
->name
, index
, dev
->real_num_rx_queues
);
2984 rxqueue
= dev
->_rx
+ index
;
2988 map
= rcu_dereference(rxqueue
->rps_map
);
2990 if (map
->len
== 1 &&
2991 !rcu_access_pointer(rxqueue
->rps_flow_table
)) {
2992 tcpu
= map
->cpus
[0];
2993 if (cpu_online(tcpu
))
2997 } else if (!rcu_access_pointer(rxqueue
->rps_flow_table
)) {
3001 skb_reset_network_header(skb
);
3002 if (!skb_get_rxhash(skb
))
3005 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3006 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
3007 if (flow_table
&& sock_flow_table
) {
3009 struct rps_dev_flow
*rflow
;
3011 rflow
= &flow_table
->flows
[skb
->rxhash
& flow_table
->mask
];
3014 next_cpu
= sock_flow_table
->ents
[skb
->rxhash
&
3015 sock_flow_table
->mask
];
3018 * If the desired CPU (where last recvmsg was done) is
3019 * different from current CPU (one in the rx-queue flow
3020 * table entry), switch if one of the following holds:
3021 * - Current CPU is unset (equal to RPS_NO_CPU).
3022 * - Current CPU is offline.
3023 * - The current CPU's queue tail has advanced beyond the
3024 * last packet that was enqueued using this table entry.
3025 * This guarantees that all previous packets for the flow
3026 * have been dequeued, thus preserving in order delivery.
3028 if (unlikely(tcpu
!= next_cpu
) &&
3029 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
3030 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
3031 rflow
->last_qtail
)) >= 0)) {
3033 rflow
= set_rps_cpu(dev
, skb
, rflow
, next_cpu
);
3036 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
3044 tcpu
= map
->cpus
[((u64
) skb
->rxhash
* map
->len
) >> 32];
3046 if (cpu_online(tcpu
)) {
3056 #ifdef CONFIG_RFS_ACCEL
3059 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3060 * @dev: Device on which the filter was set
3061 * @rxq_index: RX queue index
3062 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3063 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3065 * Drivers that implement ndo_rx_flow_steer() should periodically call
3066 * this function for each installed filter and remove the filters for
3067 * which it returns %true.
3069 bool rps_may_expire_flow(struct net_device
*dev
, u16 rxq_index
,
3070 u32 flow_id
, u16 filter_id
)
3072 struct netdev_rx_queue
*rxqueue
= dev
->_rx
+ rxq_index
;
3073 struct rps_dev_flow_table
*flow_table
;
3074 struct rps_dev_flow
*rflow
;
3079 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3080 if (flow_table
&& flow_id
<= flow_table
->mask
) {
3081 rflow
= &flow_table
->flows
[flow_id
];
3082 cpu
= ACCESS_ONCE(rflow
->cpu
);
3083 if (rflow
->filter
== filter_id
&& cpu
!= RPS_NO_CPU
&&
3084 ((int)(per_cpu(softnet_data
, cpu
).input_queue_head
-
3085 rflow
->last_qtail
) <
3086 (int)(10 * flow_table
->mask
)))
3092 EXPORT_SYMBOL(rps_may_expire_flow
);
3094 #endif /* CONFIG_RFS_ACCEL */
3096 /* Called from hardirq (IPI) context */
3097 static void rps_trigger_softirq(void *data
)
3099 struct softnet_data
*sd
= data
;
3101 ____napi_schedule(sd
, &sd
->backlog
);
3105 #endif /* CONFIG_RPS */
3108 * Check if this softnet_data structure is another cpu one
3109 * If yes, queue it to our IPI list and return 1
3112 static int rps_ipi_queued(struct softnet_data
*sd
)
3115 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
3118 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
3119 mysd
->rps_ipi_list
= sd
;
3121 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3124 #endif /* CONFIG_RPS */
3129 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3130 * queue (may be a remote CPU queue).
3132 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
3133 unsigned int *qtail
)
3135 struct softnet_data
*sd
;
3136 unsigned long flags
;
3138 sd
= &per_cpu(softnet_data
, cpu
);
3140 local_irq_save(flags
);
3143 if (skb_queue_len(&sd
->input_pkt_queue
) <= netdev_max_backlog
) {
3144 if (skb_queue_len(&sd
->input_pkt_queue
)) {
3146 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
3147 input_queue_tail_incr_save(sd
, qtail
);
3149 local_irq_restore(flags
);
3150 return NET_RX_SUCCESS
;
3153 /* Schedule NAPI for backlog device
3154 * We can use non atomic operation since we own the queue lock
3156 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
3157 if (!rps_ipi_queued(sd
))
3158 ____napi_schedule(sd
, &sd
->backlog
);
3166 local_irq_restore(flags
);
3168 atomic_long_inc(&skb
->dev
->rx_dropped
);
3174 * netif_rx - post buffer to the network code
3175 * @skb: buffer to post
3177 * This function receives a packet from a device driver and queues it for
3178 * the upper (protocol) levels to process. It always succeeds. The buffer
3179 * may be dropped during processing for congestion control or by the
3183 * NET_RX_SUCCESS (no congestion)
3184 * NET_RX_DROP (packet was dropped)
3188 int netif_rx(struct sk_buff
*skb
)
3192 /* if netpoll wants it, pretend we never saw it */
3193 if (netpoll_rx(skb
))
3196 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3198 trace_netif_rx(skb
);
3200 if (static_key_false(&rps_needed
)) {
3201 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3207 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3209 cpu
= smp_processor_id();
3211 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3219 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
3224 EXPORT_SYMBOL(netif_rx
);
3226 int netif_rx_ni(struct sk_buff
*skb
)
3231 err
= netif_rx(skb
);
3232 if (local_softirq_pending())
3238 EXPORT_SYMBOL(netif_rx_ni
);
3240 static void net_tx_action(struct softirq_action
*h
)
3242 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3244 if (sd
->completion_queue
) {
3245 struct sk_buff
*clist
;
3247 local_irq_disable();
3248 clist
= sd
->completion_queue
;
3249 sd
->completion_queue
= NULL
;
3253 struct sk_buff
*skb
= clist
;
3254 clist
= clist
->next
;
3256 WARN_ON(atomic_read(&skb
->users
));
3257 trace_kfree_skb(skb
, net_tx_action
);
3262 if (sd
->output_queue
) {
3265 local_irq_disable();
3266 head
= sd
->output_queue
;
3267 sd
->output_queue
= NULL
;
3268 sd
->output_queue_tailp
= &sd
->output_queue
;
3272 struct Qdisc
*q
= head
;
3273 spinlock_t
*root_lock
;
3275 head
= head
->next_sched
;
3277 root_lock
= qdisc_lock(q
);
3278 if (spin_trylock(root_lock
)) {
3279 smp_mb__before_clear_bit();
3280 clear_bit(__QDISC_STATE_SCHED
,
3283 spin_unlock(root_lock
);
3285 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
3287 __netif_reschedule(q
);
3289 smp_mb__before_clear_bit();
3290 clear_bit(__QDISC_STATE_SCHED
,
3298 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3299 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3300 /* This hook is defined here for ATM LANE */
3301 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
3302 unsigned char *addr
) __read_mostly
;
3303 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
3306 #ifdef CONFIG_NET_CLS_ACT
3307 /* TODO: Maybe we should just force sch_ingress to be compiled in
3308 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3309 * a compare and 2 stores extra right now if we dont have it on
3310 * but have CONFIG_NET_CLS_ACT
3311 * NOTE: This doesn't stop any functionality; if you dont have
3312 * the ingress scheduler, you just can't add policies on ingress.
3315 static int ing_filter(struct sk_buff
*skb
, struct netdev_queue
*rxq
)
3317 struct net_device
*dev
= skb
->dev
;
3318 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
3319 int result
= TC_ACT_OK
;
3322 if (unlikely(MAX_RED_LOOP
< ttl
++)) {
3323 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3324 skb
->skb_iif
, dev
->ifindex
);
3328 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
3329 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
3332 if (q
!= &noop_qdisc
) {
3333 spin_lock(qdisc_lock(q
));
3334 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
3335 result
= qdisc_enqueue_root(skb
, q
);
3336 spin_unlock(qdisc_lock(q
));
3342 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
3343 struct packet_type
**pt_prev
,
3344 int *ret
, struct net_device
*orig_dev
)
3346 struct netdev_queue
*rxq
= rcu_dereference(skb
->dev
->ingress_queue
);
3348 if (!rxq
|| rxq
->qdisc
== &noop_qdisc
)
3352 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
3356 switch (ing_filter(skb
, rxq
)) {
3370 * netdev_rx_handler_register - register receive handler
3371 * @dev: device to register a handler for
3372 * @rx_handler: receive handler to register
3373 * @rx_handler_data: data pointer that is used by rx handler
3375 * Register a receive hander for a device. This handler will then be
3376 * called from __netif_receive_skb. A negative errno code is returned
3379 * The caller must hold the rtnl_mutex.
3381 * For a general description of rx_handler, see enum rx_handler_result.
3383 int netdev_rx_handler_register(struct net_device
*dev
,
3384 rx_handler_func_t
*rx_handler
,
3385 void *rx_handler_data
)
3389 if (dev
->rx_handler
)
3392 /* Note: rx_handler_data must be set before rx_handler */
3393 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
3394 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
3398 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
3401 * netdev_rx_handler_unregister - unregister receive handler
3402 * @dev: device to unregister a handler from
3404 * Unregister a receive handler from a device.
3406 * The caller must hold the rtnl_mutex.
3408 void netdev_rx_handler_unregister(struct net_device
*dev
)
3412 RCU_INIT_POINTER(dev
->rx_handler
, NULL
);
3413 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3414 * section has a guarantee to see a non NULL rx_handler_data
3418 RCU_INIT_POINTER(dev
->rx_handler_data
, NULL
);
3420 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
3423 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3424 * the special handling of PFMEMALLOC skbs.
3426 static bool skb_pfmemalloc_protocol(struct sk_buff
*skb
)
3428 switch (skb
->protocol
) {
3429 case __constant_htons(ETH_P_ARP
):
3430 case __constant_htons(ETH_P_IP
):
3431 case __constant_htons(ETH_P_IPV6
):
3432 case __constant_htons(ETH_P_8021Q
):
3433 case __constant_htons(ETH_P_8021AD
):
3440 static int __netif_receive_skb_core(struct sk_buff
*skb
, bool pfmemalloc
)
3442 struct packet_type
*ptype
, *pt_prev
;
3443 rx_handler_func_t
*rx_handler
;
3444 struct net_device
*orig_dev
;
3445 struct net_device
*null_or_dev
;
3446 bool deliver_exact
= false;
3447 int ret
= NET_RX_DROP
;
3450 net_timestamp_check(!netdev_tstamp_prequeue
, skb
);
3452 trace_netif_receive_skb(skb
);
3454 /* if we've gotten here through NAPI, check netpoll */
3455 if (netpoll_receive_skb(skb
))
3458 orig_dev
= skb
->dev
;
3460 skb_reset_network_header(skb
);
3461 if (!skb_transport_header_was_set(skb
))
3462 skb_reset_transport_header(skb
);
3463 skb_reset_mac_len(skb
);
3468 skb
->skb_iif
= skb
->dev
->ifindex
;
3470 __this_cpu_inc(softnet_data
.processed
);
3472 if (skb
->protocol
== cpu_to_be16(ETH_P_8021Q
) ||
3473 skb
->protocol
== cpu_to_be16(ETH_P_8021AD
)) {
3474 skb
= vlan_untag(skb
);
3479 #ifdef CONFIG_NET_CLS_ACT
3480 if (skb
->tc_verd
& TC_NCLS
) {
3481 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
3489 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
3490 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
3492 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3498 #ifdef CONFIG_NET_CLS_ACT
3499 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
3505 if (pfmemalloc
&& !skb_pfmemalloc_protocol(skb
))
3508 if (vlan_tx_tag_present(skb
)) {
3510 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3513 if (vlan_do_receive(&skb
))
3515 else if (unlikely(!skb
))
3519 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
3522 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3525 switch (rx_handler(&skb
)) {
3526 case RX_HANDLER_CONSUMED
:
3527 ret
= NET_RX_SUCCESS
;
3529 case RX_HANDLER_ANOTHER
:
3531 case RX_HANDLER_EXACT
:
3532 deliver_exact
= true;
3533 case RX_HANDLER_PASS
:
3540 if (unlikely(vlan_tx_tag_present(skb
))) {
3541 if (vlan_tx_tag_get_id(skb
))
3542 skb
->pkt_type
= PACKET_OTHERHOST
;
3543 /* Note: we might in the future use prio bits
3544 * and set skb->priority like in vlan_do_receive()
3545 * For the time being, just ignore Priority Code Point
3550 /* deliver only exact match when indicated */
3551 null_or_dev
= deliver_exact
? skb
->dev
: NULL
;
3553 type
= skb
->protocol
;
3554 list_for_each_entry_rcu(ptype
,
3555 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
3556 if (ptype
->type
== type
&&
3557 (ptype
->dev
== null_or_dev
|| ptype
->dev
== skb
->dev
||
3558 ptype
->dev
== orig_dev
)) {
3560 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3566 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
3569 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
3572 atomic_long_inc(&skb
->dev
->rx_dropped
);
3574 /* Jamal, now you will not able to escape explaining
3575 * me how you were going to use this. :-)
3584 static int __netif_receive_skb(struct sk_buff
*skb
)
3588 if (sk_memalloc_socks() && skb_pfmemalloc(skb
)) {
3589 unsigned long pflags
= current
->flags
;
3592 * PFMEMALLOC skbs are special, they should
3593 * - be delivered to SOCK_MEMALLOC sockets only
3594 * - stay away from userspace
3595 * - have bounded memory usage
3597 * Use PF_MEMALLOC as this saves us from propagating the allocation
3598 * context down to all allocation sites.
3600 current
->flags
|= PF_MEMALLOC
;
3601 ret
= __netif_receive_skb_core(skb
, true);
3602 tsk_restore_flags(current
, pflags
, PF_MEMALLOC
);
3604 ret
= __netif_receive_skb_core(skb
, false);
3610 * netif_receive_skb - process receive buffer from network
3611 * @skb: buffer to process
3613 * netif_receive_skb() is the main receive data processing function.
3614 * It always succeeds. The buffer may be dropped during processing
3615 * for congestion control or by the protocol layers.
3617 * This function may only be called from softirq context and interrupts
3618 * should be enabled.
3620 * Return values (usually ignored):
3621 * NET_RX_SUCCESS: no congestion
3622 * NET_RX_DROP: packet was dropped
3624 int netif_receive_skb(struct sk_buff
*skb
)
3628 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3630 if (skb_defer_rx_timestamp(skb
))
3631 return NET_RX_SUCCESS
;
3636 if (static_key_false(&rps_needed
)) {
3637 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3638 int cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3641 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3647 ret
= __netif_receive_skb(skb
);
3651 EXPORT_SYMBOL(netif_receive_skb
);
3653 /* Network device is going away, flush any packets still pending
3654 * Called with irqs disabled.
3656 static void flush_backlog(void *arg
)
3658 struct net_device
*dev
= arg
;
3659 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3660 struct sk_buff
*skb
, *tmp
;
3663 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3664 if (skb
->dev
== dev
) {
3665 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3667 input_queue_head_incr(sd
);
3672 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3673 if (skb
->dev
== dev
) {
3674 __skb_unlink(skb
, &sd
->process_queue
);
3676 input_queue_head_incr(sd
);
3681 static int napi_gro_complete(struct sk_buff
*skb
)
3683 struct packet_offload
*ptype
;
3684 __be16 type
= skb
->protocol
;
3685 struct list_head
*head
= &offload_base
;
3688 BUILD_BUG_ON(sizeof(struct napi_gro_cb
) > sizeof(skb
->cb
));
3690 if (NAPI_GRO_CB(skb
)->count
== 1) {
3691 skb_shinfo(skb
)->gso_size
= 0;
3696 list_for_each_entry_rcu(ptype
, head
, list
) {
3697 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_complete
)
3700 err
= ptype
->callbacks
.gro_complete(skb
);
3706 WARN_ON(&ptype
->list
== head
);
3708 return NET_RX_SUCCESS
;
3712 return netif_receive_skb(skb
);
3715 /* napi->gro_list contains packets ordered by age.
3716 * youngest packets at the head of it.
3717 * Complete skbs in reverse order to reduce latencies.
3719 void napi_gro_flush(struct napi_struct
*napi
, bool flush_old
)
3721 struct sk_buff
*skb
, *prev
= NULL
;
3723 /* scan list and build reverse chain */
3724 for (skb
= napi
->gro_list
; skb
!= NULL
; skb
= skb
->next
) {
3729 for (skb
= prev
; skb
; skb
= prev
) {
3732 if (flush_old
&& NAPI_GRO_CB(skb
)->age
== jiffies
)
3736 napi_gro_complete(skb
);
3740 napi
->gro_list
= NULL
;
3742 EXPORT_SYMBOL(napi_gro_flush
);
3744 static void gro_list_prepare(struct napi_struct
*napi
, struct sk_buff
*skb
)
3747 unsigned int maclen
= skb
->dev
->hard_header_len
;
3749 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3750 unsigned long diffs
;
3752 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
3753 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
3754 if (maclen
== ETH_HLEN
)
3755 diffs
|= compare_ether_header(skb_mac_header(p
),
3756 skb_gro_mac_header(skb
));
3758 diffs
= memcmp(skb_mac_header(p
),
3759 skb_gro_mac_header(skb
),
3761 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
3762 NAPI_GRO_CB(p
)->flush
= 0;
3766 static enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3768 struct sk_buff
**pp
= NULL
;
3769 struct packet_offload
*ptype
;
3770 __be16 type
= skb
->protocol
;
3771 struct list_head
*head
= &offload_base
;
3773 enum gro_result ret
;
3775 if (!(skb
->dev
->features
& NETIF_F_GRO
) || netpoll_rx_on(skb
))
3778 if (skb_is_gso(skb
) || skb_has_frag_list(skb
))
3781 gro_list_prepare(napi
, skb
);
3784 list_for_each_entry_rcu(ptype
, head
, list
) {
3785 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_receive
)
3788 skb_set_network_header(skb
, skb_gro_offset(skb
));
3789 skb_reset_mac_len(skb
);
3790 NAPI_GRO_CB(skb
)->same_flow
= 0;
3791 NAPI_GRO_CB(skb
)->flush
= 0;
3792 NAPI_GRO_CB(skb
)->free
= 0;
3794 pp
= ptype
->callbacks
.gro_receive(&napi
->gro_list
, skb
);
3799 if (&ptype
->list
== head
)
3802 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3803 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3806 struct sk_buff
*nskb
= *pp
;
3810 napi_gro_complete(nskb
);
3817 if (NAPI_GRO_CB(skb
)->flush
|| napi
->gro_count
>= MAX_GRO_SKBS
)
3821 NAPI_GRO_CB(skb
)->count
= 1;
3822 NAPI_GRO_CB(skb
)->age
= jiffies
;
3823 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
3824 skb
->next
= napi
->gro_list
;
3825 napi
->gro_list
= skb
;
3829 if (skb_headlen(skb
) < skb_gro_offset(skb
)) {
3830 int grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
3832 BUG_ON(skb
->end
- skb
->tail
< grow
);
3834 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3837 skb
->data_len
-= grow
;
3839 skb_shinfo(skb
)->frags
[0].page_offset
+= grow
;
3840 skb_frag_size_sub(&skb_shinfo(skb
)->frags
[0], grow
);
3842 if (unlikely(!skb_frag_size(&skb_shinfo(skb
)->frags
[0]))) {
3843 skb_frag_unref(skb
, 0);
3844 memmove(skb_shinfo(skb
)->frags
,
3845 skb_shinfo(skb
)->frags
+ 1,
3846 --skb_shinfo(skb
)->nr_frags
* sizeof(skb_frag_t
));
3859 static gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
3863 if (netif_receive_skb(skb
))
3871 case GRO_MERGED_FREE
:
3872 if (NAPI_GRO_CB(skb
)->free
== NAPI_GRO_FREE_STOLEN_HEAD
)
3873 kmem_cache_free(skbuff_head_cache
, skb
);
3886 static void skb_gro_reset_offset(struct sk_buff
*skb
)
3888 const struct skb_shared_info
*pinfo
= skb_shinfo(skb
);
3889 const skb_frag_t
*frag0
= &pinfo
->frags
[0];
3891 NAPI_GRO_CB(skb
)->data_offset
= 0;
3892 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3893 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3895 if (skb
->mac_header
== skb
->tail
&&
3897 !PageHighMem(skb_frag_page(frag0
))) {
3898 NAPI_GRO_CB(skb
)->frag0
= skb_frag_address(frag0
);
3899 NAPI_GRO_CB(skb
)->frag0_len
= skb_frag_size(frag0
);
3903 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3905 skb_gro_reset_offset(skb
);
3907 return napi_skb_finish(dev_gro_receive(napi
, skb
), skb
);
3909 EXPORT_SYMBOL(napi_gro_receive
);
3911 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
3913 __skb_pull(skb
, skb_headlen(skb
));
3914 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3915 skb_reserve(skb
, NET_SKB_PAD
+ NET_IP_ALIGN
- skb_headroom(skb
));
3917 skb
->dev
= napi
->dev
;
3919 skb
->truesize
= SKB_TRUESIZE(skb_end_offset(skb
));
3924 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
3926 struct sk_buff
*skb
= napi
->skb
;
3929 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
3935 EXPORT_SYMBOL(napi_get_frags
);
3937 static gro_result_t
napi_frags_finish(struct napi_struct
*napi
, struct sk_buff
*skb
,
3943 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
3945 if (ret
== GRO_HELD
)
3946 skb_gro_pull(skb
, -ETH_HLEN
);
3947 else if (netif_receive_skb(skb
))
3952 case GRO_MERGED_FREE
:
3953 napi_reuse_skb(napi
, skb
);
3963 static struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
3965 struct sk_buff
*skb
= napi
->skb
;
3972 skb_reset_mac_header(skb
);
3973 skb_gro_reset_offset(skb
);
3975 off
= skb_gro_offset(skb
);
3976 hlen
= off
+ sizeof(*eth
);
3977 eth
= skb_gro_header_fast(skb
, off
);
3978 if (skb_gro_header_hard(skb
, hlen
)) {
3979 eth
= skb_gro_header_slow(skb
, hlen
, off
);
3980 if (unlikely(!eth
)) {
3981 napi_reuse_skb(napi
, skb
);
3987 skb_gro_pull(skb
, sizeof(*eth
));
3990 * This works because the only protocols we care about don't require
3991 * special handling. We'll fix it up properly at the end.
3993 skb
->protocol
= eth
->h_proto
;
3999 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
4001 struct sk_buff
*skb
= napi_frags_skb(napi
);
4006 return napi_frags_finish(napi
, skb
, dev_gro_receive(napi
, skb
));
4008 EXPORT_SYMBOL(napi_gro_frags
);
4011 * net_rps_action sends any pending IPI's for rps.
4012 * Note: called with local irq disabled, but exits with local irq enabled.
4014 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
4017 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
4020 sd
->rps_ipi_list
= NULL
;
4024 /* Send pending IPI's to kick RPS processing on remote cpus. */
4026 struct softnet_data
*next
= remsd
->rps_ipi_next
;
4028 if (cpu_online(remsd
->cpu
))
4029 __smp_call_function_single(remsd
->cpu
,
4038 static int process_backlog(struct napi_struct
*napi
, int quota
)
4041 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
4044 /* Check if we have pending ipi, its better to send them now,
4045 * not waiting net_rx_action() end.
4047 if (sd
->rps_ipi_list
) {
4048 local_irq_disable();
4049 net_rps_action_and_irq_enable(sd
);
4052 napi
->weight
= weight_p
;
4053 local_irq_disable();
4054 while (work
< quota
) {
4055 struct sk_buff
*skb
;
4058 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
4061 __netif_receive_skb(skb
);
4063 local_irq_disable();
4064 input_queue_head_incr(sd
);
4065 if (++work
>= quota
) {
4072 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
4074 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
4075 &sd
->process_queue
);
4077 if (qlen
< quota
- work
) {
4079 * Inline a custom version of __napi_complete().
4080 * only current cpu owns and manipulates this napi,
4081 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
4082 * we can use a plain write instead of clear_bit(),
4083 * and we dont need an smp_mb() memory barrier.
4085 list_del(&napi
->poll_list
);
4088 quota
= work
+ qlen
;
4098 * __napi_schedule - schedule for receive
4099 * @n: entry to schedule
4101 * The entry's receive function will be scheduled to run
4103 void __napi_schedule(struct napi_struct
*n
)
4105 unsigned long flags
;
4107 local_irq_save(flags
);
4108 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
4109 local_irq_restore(flags
);
4111 EXPORT_SYMBOL(__napi_schedule
);
4113 void __napi_complete(struct napi_struct
*n
)
4115 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
4116 BUG_ON(n
->gro_list
);
4118 list_del(&n
->poll_list
);
4119 smp_mb__before_clear_bit();
4120 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
4122 EXPORT_SYMBOL(__napi_complete
);
4124 void napi_complete(struct napi_struct
*n
)
4126 unsigned long flags
;
4129 * don't let napi dequeue from the cpu poll list
4130 * just in case its running on a different cpu
4132 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
4135 napi_gro_flush(n
, false);
4136 local_irq_save(flags
);
4138 local_irq_restore(flags
);
4140 EXPORT_SYMBOL(napi_complete
);
4142 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
4143 int (*poll
)(struct napi_struct
*, int), int weight
)
4145 INIT_LIST_HEAD(&napi
->poll_list
);
4146 napi
->gro_count
= 0;
4147 napi
->gro_list
= NULL
;
4150 if (weight
> NAPI_POLL_WEIGHT
)
4151 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4153 napi
->weight
= weight
;
4154 list_add(&napi
->dev_list
, &dev
->napi_list
);
4156 #ifdef CONFIG_NETPOLL
4157 spin_lock_init(&napi
->poll_lock
);
4158 napi
->poll_owner
= -1;
4160 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
4162 EXPORT_SYMBOL(netif_napi_add
);
4164 void netif_napi_del(struct napi_struct
*napi
)
4166 struct sk_buff
*skb
, *next
;
4168 list_del_init(&napi
->dev_list
);
4169 napi_free_frags(napi
);
4171 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
4177 napi
->gro_list
= NULL
;
4178 napi
->gro_count
= 0;
4180 EXPORT_SYMBOL(netif_napi_del
);
4182 static void net_rx_action(struct softirq_action
*h
)
4184 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
4185 unsigned long time_limit
= jiffies
+ 2;
4186 int budget
= netdev_budget
;
4189 local_irq_disable();
4191 while (!list_empty(&sd
->poll_list
)) {
4192 struct napi_struct
*n
;
4195 /* If softirq window is exhuasted then punt.
4196 * Allow this to run for 2 jiffies since which will allow
4197 * an average latency of 1.5/HZ.
4199 if (unlikely(budget
<= 0 || time_after_eq(jiffies
, time_limit
)))
4204 /* Even though interrupts have been re-enabled, this
4205 * access is safe because interrupts can only add new
4206 * entries to the tail of this list, and only ->poll()
4207 * calls can remove this head entry from the list.
4209 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
4211 have
= netpoll_poll_lock(n
);
4215 /* This NAPI_STATE_SCHED test is for avoiding a race
4216 * with netpoll's poll_napi(). Only the entity which
4217 * obtains the lock and sees NAPI_STATE_SCHED set will
4218 * actually make the ->poll() call. Therefore we avoid
4219 * accidentally calling ->poll() when NAPI is not scheduled.
4222 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
4223 work
= n
->poll(n
, weight
);
4227 WARN_ON_ONCE(work
> weight
);
4231 local_irq_disable();
4233 /* Drivers must not modify the NAPI state if they
4234 * consume the entire weight. In such cases this code
4235 * still "owns" the NAPI instance and therefore can
4236 * move the instance around on the list at-will.
4238 if (unlikely(work
== weight
)) {
4239 if (unlikely(napi_disable_pending(n
))) {
4242 local_irq_disable();
4245 /* flush too old packets
4246 * If HZ < 1000, flush all packets.
4249 napi_gro_flush(n
, HZ
>= 1000);
4250 local_irq_disable();
4252 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
4256 netpoll_poll_unlock(have
);
4259 net_rps_action_and_irq_enable(sd
);
4261 #ifdef CONFIG_NET_DMA
4263 * There may not be any more sk_buffs coming right now, so push
4264 * any pending DMA copies to hardware
4266 dma_issue_pending_all();
4273 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
4277 struct netdev_upper
{
4278 struct net_device
*dev
;
4280 struct list_head list
;
4281 struct rcu_head rcu
;
4282 struct list_head search_list
;
4285 static void __append_search_uppers(struct list_head
*search_list
,
4286 struct net_device
*dev
)
4288 struct netdev_upper
*upper
;
4290 list_for_each_entry(upper
, &dev
->upper_dev_list
, list
) {
4291 /* check if this upper is not already in search list */
4292 if (list_empty(&upper
->search_list
))
4293 list_add_tail(&upper
->search_list
, search_list
);
4297 static bool __netdev_search_upper_dev(struct net_device
*dev
,
4298 struct net_device
*upper_dev
)
4300 LIST_HEAD(search_list
);
4301 struct netdev_upper
*upper
;
4302 struct netdev_upper
*tmp
;
4305 __append_search_uppers(&search_list
, dev
);
4306 list_for_each_entry(upper
, &search_list
, search_list
) {
4307 if (upper
->dev
== upper_dev
) {
4311 __append_search_uppers(&search_list
, upper
->dev
);
4313 list_for_each_entry_safe(upper
, tmp
, &search_list
, search_list
)
4314 INIT_LIST_HEAD(&upper
->search_list
);
4318 static struct netdev_upper
*__netdev_find_upper(struct net_device
*dev
,
4319 struct net_device
*upper_dev
)
4321 struct netdev_upper
*upper
;
4323 list_for_each_entry(upper
, &dev
->upper_dev_list
, list
) {
4324 if (upper
->dev
== upper_dev
)
4331 * netdev_has_upper_dev - Check if device is linked to an upper device
4333 * @upper_dev: upper device to check
4335 * Find out if a device is linked to specified upper device and return true
4336 * in case it is. Note that this checks only immediate upper device,
4337 * not through a complete stack of devices. The caller must hold the RTNL lock.
4339 bool netdev_has_upper_dev(struct net_device
*dev
,
4340 struct net_device
*upper_dev
)
4344 return __netdev_find_upper(dev
, upper_dev
);
4346 EXPORT_SYMBOL(netdev_has_upper_dev
);
4349 * netdev_has_any_upper_dev - Check if device is linked to some device
4352 * Find out if a device is linked to an upper device and return true in case
4353 * it is. The caller must hold the RTNL lock.
4355 bool netdev_has_any_upper_dev(struct net_device
*dev
)
4359 return !list_empty(&dev
->upper_dev_list
);
4361 EXPORT_SYMBOL(netdev_has_any_upper_dev
);
4364 * netdev_master_upper_dev_get - Get master upper device
4367 * Find a master upper device and return pointer to it or NULL in case
4368 * it's not there. The caller must hold the RTNL lock.
4370 struct net_device
*netdev_master_upper_dev_get(struct net_device
*dev
)
4372 struct netdev_upper
*upper
;
4376 if (list_empty(&dev
->upper_dev_list
))
4379 upper
= list_first_entry(&dev
->upper_dev_list
,
4380 struct netdev_upper
, list
);
4381 if (likely(upper
->master
))
4385 EXPORT_SYMBOL(netdev_master_upper_dev_get
);
4388 * netdev_master_upper_dev_get_rcu - Get master upper device
4391 * Find a master upper device and return pointer to it or NULL in case
4392 * it's not there. The caller must hold the RCU read lock.
4394 struct net_device
*netdev_master_upper_dev_get_rcu(struct net_device
*dev
)
4396 struct netdev_upper
*upper
;
4398 upper
= list_first_or_null_rcu(&dev
->upper_dev_list
,
4399 struct netdev_upper
, list
);
4400 if (upper
&& likely(upper
->master
))
4404 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu
);
4406 static int __netdev_upper_dev_link(struct net_device
*dev
,
4407 struct net_device
*upper_dev
, bool master
)
4409 struct netdev_upper
*upper
;
4413 if (dev
== upper_dev
)
4416 /* To prevent loops, check if dev is not upper device to upper_dev. */
4417 if (__netdev_search_upper_dev(upper_dev
, dev
))
4420 if (__netdev_find_upper(dev
, upper_dev
))
4423 if (master
&& netdev_master_upper_dev_get(dev
))
4426 upper
= kmalloc(sizeof(*upper
), GFP_KERNEL
);
4430 upper
->dev
= upper_dev
;
4431 upper
->master
= master
;
4432 INIT_LIST_HEAD(&upper
->search_list
);
4434 /* Ensure that master upper link is always the first item in list. */
4436 list_add_rcu(&upper
->list
, &dev
->upper_dev_list
);
4438 list_add_tail_rcu(&upper
->list
, &dev
->upper_dev_list
);
4439 dev_hold(upper_dev
);
4445 * netdev_upper_dev_link - Add a link to the upper device
4447 * @upper_dev: new upper device
4449 * Adds a link to device which is upper to this one. The caller must hold
4450 * the RTNL lock. On a failure a negative errno code is returned.
4451 * On success the reference counts are adjusted and the function
4454 int netdev_upper_dev_link(struct net_device
*dev
,
4455 struct net_device
*upper_dev
)
4457 return __netdev_upper_dev_link(dev
, upper_dev
, false);
4459 EXPORT_SYMBOL(netdev_upper_dev_link
);
4462 * netdev_master_upper_dev_link - Add a master link to the upper device
4464 * @upper_dev: new upper device
4466 * Adds a link to device which is upper to this one. In this case, only
4467 * one master upper device can be linked, although other non-master devices
4468 * might be linked as well. The caller must hold the RTNL lock.
4469 * On a failure a negative errno code is returned. On success the reference
4470 * counts are adjusted and the function returns zero.
4472 int netdev_master_upper_dev_link(struct net_device
*dev
,
4473 struct net_device
*upper_dev
)
4475 return __netdev_upper_dev_link(dev
, upper_dev
, true);
4477 EXPORT_SYMBOL(netdev_master_upper_dev_link
);
4480 * netdev_upper_dev_unlink - Removes a link to upper device
4482 * @upper_dev: new upper device
4484 * Removes a link to device which is upper to this one. The caller must hold
4487 void netdev_upper_dev_unlink(struct net_device
*dev
,
4488 struct net_device
*upper_dev
)
4490 struct netdev_upper
*upper
;
4494 upper
= __netdev_find_upper(dev
, upper_dev
);
4497 list_del_rcu(&upper
->list
);
4499 kfree_rcu(upper
, rcu
);
4501 EXPORT_SYMBOL(netdev_upper_dev_unlink
);
4503 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
4505 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4507 if (ops
->ndo_change_rx_flags
)
4508 ops
->ndo_change_rx_flags(dev
, flags
);
4511 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
4513 unsigned int old_flags
= dev
->flags
;
4519 dev
->flags
|= IFF_PROMISC
;
4520 dev
->promiscuity
+= inc
;
4521 if (dev
->promiscuity
== 0) {
4524 * If inc causes overflow, untouch promisc and return error.
4527 dev
->flags
&= ~IFF_PROMISC
;
4529 dev
->promiscuity
-= inc
;
4530 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4535 if (dev
->flags
!= old_flags
) {
4536 pr_info("device %s %s promiscuous mode\n",
4538 dev
->flags
& IFF_PROMISC
? "entered" : "left");
4539 if (audit_enabled
) {
4540 current_uid_gid(&uid
, &gid
);
4541 audit_log(current
->audit_context
, GFP_ATOMIC
,
4542 AUDIT_ANOM_PROMISCUOUS
,
4543 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4544 dev
->name
, (dev
->flags
& IFF_PROMISC
),
4545 (old_flags
& IFF_PROMISC
),
4546 from_kuid(&init_user_ns
, audit_get_loginuid(current
)),
4547 from_kuid(&init_user_ns
, uid
),
4548 from_kgid(&init_user_ns
, gid
),
4549 audit_get_sessionid(current
));
4552 dev_change_rx_flags(dev
, IFF_PROMISC
);
4558 * dev_set_promiscuity - update promiscuity count on a device
4562 * Add or remove promiscuity from a device. While the count in the device
4563 * remains above zero the interface remains promiscuous. Once it hits zero
4564 * the device reverts back to normal filtering operation. A negative inc
4565 * value is used to drop promiscuity on the device.
4566 * Return 0 if successful or a negative errno code on error.
4568 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
4570 unsigned int old_flags
= dev
->flags
;
4573 err
= __dev_set_promiscuity(dev
, inc
);
4576 if (dev
->flags
!= old_flags
)
4577 dev_set_rx_mode(dev
);
4580 EXPORT_SYMBOL(dev_set_promiscuity
);
4583 * dev_set_allmulti - update allmulti count on a device
4587 * Add or remove reception of all multicast frames to a device. While the
4588 * count in the device remains above zero the interface remains listening
4589 * to all interfaces. Once it hits zero the device reverts back to normal
4590 * filtering operation. A negative @inc value is used to drop the counter
4591 * when releasing a resource needing all multicasts.
4592 * Return 0 if successful or a negative errno code on error.
4595 int dev_set_allmulti(struct net_device
*dev
, int inc
)
4597 unsigned int old_flags
= dev
->flags
;
4601 dev
->flags
|= IFF_ALLMULTI
;
4602 dev
->allmulti
+= inc
;
4603 if (dev
->allmulti
== 0) {
4606 * If inc causes overflow, untouch allmulti and return error.
4609 dev
->flags
&= ~IFF_ALLMULTI
;
4611 dev
->allmulti
-= inc
;
4612 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4617 if (dev
->flags
^ old_flags
) {
4618 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
4619 dev_set_rx_mode(dev
);
4623 EXPORT_SYMBOL(dev_set_allmulti
);
4626 * Upload unicast and multicast address lists to device and
4627 * configure RX filtering. When the device doesn't support unicast
4628 * filtering it is put in promiscuous mode while unicast addresses
4631 void __dev_set_rx_mode(struct net_device
*dev
)
4633 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4635 /* dev_open will call this function so the list will stay sane. */
4636 if (!(dev
->flags
&IFF_UP
))
4639 if (!netif_device_present(dev
))
4642 if (!(dev
->priv_flags
& IFF_UNICAST_FLT
)) {
4643 /* Unicast addresses changes may only happen under the rtnl,
4644 * therefore calling __dev_set_promiscuity here is safe.
4646 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
4647 __dev_set_promiscuity(dev
, 1);
4648 dev
->uc_promisc
= true;
4649 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
4650 __dev_set_promiscuity(dev
, -1);
4651 dev
->uc_promisc
= false;
4655 if (ops
->ndo_set_rx_mode
)
4656 ops
->ndo_set_rx_mode(dev
);
4658 EXPORT_SYMBOL(__dev_set_rx_mode
);
4660 void dev_set_rx_mode(struct net_device
*dev
)
4662 netif_addr_lock_bh(dev
);
4663 __dev_set_rx_mode(dev
);
4664 netif_addr_unlock_bh(dev
);
4668 * dev_get_flags - get flags reported to userspace
4671 * Get the combination of flag bits exported through APIs to userspace.
4673 unsigned int dev_get_flags(const struct net_device
*dev
)
4677 flags
= (dev
->flags
& ~(IFF_PROMISC
|
4682 (dev
->gflags
& (IFF_PROMISC
|
4685 if (netif_running(dev
)) {
4686 if (netif_oper_up(dev
))
4687 flags
|= IFF_RUNNING
;
4688 if (netif_carrier_ok(dev
))
4689 flags
|= IFF_LOWER_UP
;
4690 if (netif_dormant(dev
))
4691 flags
|= IFF_DORMANT
;
4696 EXPORT_SYMBOL(dev_get_flags
);
4698 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4700 unsigned int old_flags
= dev
->flags
;
4706 * Set the flags on our device.
4709 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
4710 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
4712 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
4716 * Load in the correct multicast list now the flags have changed.
4719 if ((old_flags
^ flags
) & IFF_MULTICAST
)
4720 dev_change_rx_flags(dev
, IFF_MULTICAST
);
4722 dev_set_rx_mode(dev
);
4725 * Have we downed the interface. We handle IFF_UP ourselves
4726 * according to user attempts to set it, rather than blindly
4731 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
4732 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
4735 dev_set_rx_mode(dev
);
4738 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
4739 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
4741 dev
->gflags
^= IFF_PROMISC
;
4742 dev_set_promiscuity(dev
, inc
);
4745 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4746 is important. Some (broken) drivers set IFF_PROMISC, when
4747 IFF_ALLMULTI is requested not asking us and not reporting.
4749 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
4750 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
4752 dev
->gflags
^= IFF_ALLMULTI
;
4753 dev_set_allmulti(dev
, inc
);
4759 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
)
4761 unsigned int changes
= dev
->flags
^ old_flags
;
4763 if (changes
& IFF_UP
) {
4764 if (dev
->flags
& IFF_UP
)
4765 call_netdevice_notifiers(NETDEV_UP
, dev
);
4767 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
4770 if (dev
->flags
& IFF_UP
&&
4771 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
)))
4772 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
4776 * dev_change_flags - change device settings
4778 * @flags: device state flags
4780 * Change settings on device based state flags. The flags are
4781 * in the userspace exported format.
4783 int dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4786 unsigned int changes
, old_flags
= dev
->flags
;
4788 ret
= __dev_change_flags(dev
, flags
);
4792 changes
= old_flags
^ dev
->flags
;
4794 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
4796 __dev_notify_flags(dev
, old_flags
);
4799 EXPORT_SYMBOL(dev_change_flags
);
4802 * dev_set_mtu - Change maximum transfer unit
4804 * @new_mtu: new transfer unit
4806 * Change the maximum transfer size of the network device.
4808 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
4810 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4813 if (new_mtu
== dev
->mtu
)
4816 /* MTU must be positive. */
4820 if (!netif_device_present(dev
))
4824 if (ops
->ndo_change_mtu
)
4825 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
4830 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
4833 EXPORT_SYMBOL(dev_set_mtu
);
4836 * dev_set_group - Change group this device belongs to
4838 * @new_group: group this device should belong to
4840 void dev_set_group(struct net_device
*dev
, int new_group
)
4842 dev
->group
= new_group
;
4844 EXPORT_SYMBOL(dev_set_group
);
4847 * dev_set_mac_address - Change Media Access Control Address
4851 * Change the hardware (MAC) address of the device
4853 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
4855 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4858 if (!ops
->ndo_set_mac_address
)
4860 if (sa
->sa_family
!= dev
->type
)
4862 if (!netif_device_present(dev
))
4864 err
= ops
->ndo_set_mac_address(dev
, sa
);
4867 dev
->addr_assign_type
= NET_ADDR_SET
;
4868 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4869 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
4872 EXPORT_SYMBOL(dev_set_mac_address
);
4875 * dev_change_carrier - Change device carrier
4877 * @new_carrier: new value
4879 * Change device carrier
4881 int dev_change_carrier(struct net_device
*dev
, bool new_carrier
)
4883 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4885 if (!ops
->ndo_change_carrier
)
4887 if (!netif_device_present(dev
))
4889 return ops
->ndo_change_carrier(dev
, new_carrier
);
4891 EXPORT_SYMBOL(dev_change_carrier
);
4894 * dev_new_index - allocate an ifindex
4895 * @net: the applicable net namespace
4897 * Returns a suitable unique value for a new device interface
4898 * number. The caller must hold the rtnl semaphore or the
4899 * dev_base_lock to be sure it remains unique.
4901 static int dev_new_index(struct net
*net
)
4903 int ifindex
= net
->ifindex
;
4907 if (!__dev_get_by_index(net
, ifindex
))
4908 return net
->ifindex
= ifindex
;
4912 /* Delayed registration/unregisteration */
4913 static LIST_HEAD(net_todo_list
);
4915 static void net_set_todo(struct net_device
*dev
)
4917 list_add_tail(&dev
->todo_list
, &net_todo_list
);
4920 static void rollback_registered_many(struct list_head
*head
)
4922 struct net_device
*dev
, *tmp
;
4924 BUG_ON(dev_boot_phase
);
4927 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
4928 /* Some devices call without registering
4929 * for initialization unwind. Remove those
4930 * devices and proceed with the remaining.
4932 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
4933 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
4937 list_del(&dev
->unreg_list
);
4940 dev
->dismantle
= true;
4941 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
4944 /* If device is running, close it first. */
4945 dev_close_many(head
);
4947 list_for_each_entry(dev
, head
, unreg_list
) {
4948 /* And unlink it from device chain. */
4949 unlist_netdevice(dev
);
4951 dev
->reg_state
= NETREG_UNREGISTERING
;
4956 list_for_each_entry(dev
, head
, unreg_list
) {
4957 /* Shutdown queueing discipline. */
4961 /* Notify protocols, that we are about to destroy
4962 this device. They should clean all the things.
4964 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
4966 if (!dev
->rtnl_link_ops
||
4967 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
4968 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
4971 * Flush the unicast and multicast chains
4976 if (dev
->netdev_ops
->ndo_uninit
)
4977 dev
->netdev_ops
->ndo_uninit(dev
);
4979 /* Notifier chain MUST detach us all upper devices. */
4980 WARN_ON(netdev_has_any_upper_dev(dev
));
4982 /* Remove entries from kobject tree */
4983 netdev_unregister_kobject(dev
);
4985 /* Remove XPS queueing entries */
4986 netif_reset_xps_queues_gt(dev
, 0);
4992 list_for_each_entry(dev
, head
, unreg_list
)
4996 static void rollback_registered(struct net_device
*dev
)
5000 list_add(&dev
->unreg_list
, &single
);
5001 rollback_registered_many(&single
);
5005 static netdev_features_t
netdev_fix_features(struct net_device
*dev
,
5006 netdev_features_t features
)
5008 /* Fix illegal checksum combinations */
5009 if ((features
& NETIF_F_HW_CSUM
) &&
5010 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5011 netdev_warn(dev
, "mixed HW and IP checksum settings.\n");
5012 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5015 /* TSO requires that SG is present as well. */
5016 if ((features
& NETIF_F_ALL_TSO
) && !(features
& NETIF_F_SG
)) {
5017 netdev_dbg(dev
, "Dropping TSO features since no SG feature.\n");
5018 features
&= ~NETIF_F_ALL_TSO
;
5021 if ((features
& NETIF_F_TSO
) && !(features
& NETIF_F_HW_CSUM
) &&
5022 !(features
& NETIF_F_IP_CSUM
)) {
5023 netdev_dbg(dev
, "Dropping TSO features since no CSUM feature.\n");
5024 features
&= ~NETIF_F_TSO
;
5025 features
&= ~NETIF_F_TSO_ECN
;
5028 if ((features
& NETIF_F_TSO6
) && !(features
& NETIF_F_HW_CSUM
) &&
5029 !(features
& NETIF_F_IPV6_CSUM
)) {
5030 netdev_dbg(dev
, "Dropping TSO6 features since no CSUM feature.\n");
5031 features
&= ~NETIF_F_TSO6
;
5034 /* TSO ECN requires that TSO is present as well. */
5035 if ((features
& NETIF_F_ALL_TSO
) == NETIF_F_TSO_ECN
)
5036 features
&= ~NETIF_F_TSO_ECN
;
5038 /* Software GSO depends on SG. */
5039 if ((features
& NETIF_F_GSO
) && !(features
& NETIF_F_SG
)) {
5040 netdev_dbg(dev
, "Dropping NETIF_F_GSO since no SG feature.\n");
5041 features
&= ~NETIF_F_GSO
;
5044 /* UFO needs SG and checksumming */
5045 if (features
& NETIF_F_UFO
) {
5046 /* maybe split UFO into V4 and V6? */
5047 if (!((features
& NETIF_F_GEN_CSUM
) ||
5048 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))
5049 == (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5051 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5052 features
&= ~NETIF_F_UFO
;
5055 if (!(features
& NETIF_F_SG
)) {
5057 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5058 features
&= ~NETIF_F_UFO
;
5065 int __netdev_update_features(struct net_device
*dev
)
5067 netdev_features_t features
;
5072 features
= netdev_get_wanted_features(dev
);
5074 if (dev
->netdev_ops
->ndo_fix_features
)
5075 features
= dev
->netdev_ops
->ndo_fix_features(dev
, features
);
5077 /* driver might be less strict about feature dependencies */
5078 features
= netdev_fix_features(dev
, features
);
5080 if (dev
->features
== features
)
5083 netdev_dbg(dev
, "Features changed: %pNF -> %pNF\n",
5084 &dev
->features
, &features
);
5086 if (dev
->netdev_ops
->ndo_set_features
)
5087 err
= dev
->netdev_ops
->ndo_set_features(dev
, features
);
5089 if (unlikely(err
< 0)) {
5091 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5092 err
, &features
, &dev
->features
);
5097 dev
->features
= features
;
5103 * netdev_update_features - recalculate device features
5104 * @dev: the device to check
5106 * Recalculate dev->features set and send notifications if it
5107 * has changed. Should be called after driver or hardware dependent
5108 * conditions might have changed that influence the features.
5110 void netdev_update_features(struct net_device
*dev
)
5112 if (__netdev_update_features(dev
))
5113 netdev_features_change(dev
);
5115 EXPORT_SYMBOL(netdev_update_features
);
5118 * netdev_change_features - recalculate device features
5119 * @dev: the device to check
5121 * Recalculate dev->features set and send notifications even
5122 * if they have not changed. Should be called instead of
5123 * netdev_update_features() if also dev->vlan_features might
5124 * have changed to allow the changes to be propagated to stacked
5127 void netdev_change_features(struct net_device
*dev
)
5129 __netdev_update_features(dev
);
5130 netdev_features_change(dev
);
5132 EXPORT_SYMBOL(netdev_change_features
);
5135 * netif_stacked_transfer_operstate - transfer operstate
5136 * @rootdev: the root or lower level device to transfer state from
5137 * @dev: the device to transfer operstate to
5139 * Transfer operational state from root to device. This is normally
5140 * called when a stacking relationship exists between the root
5141 * device and the device(a leaf device).
5143 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
5144 struct net_device
*dev
)
5146 if (rootdev
->operstate
== IF_OPER_DORMANT
)
5147 netif_dormant_on(dev
);
5149 netif_dormant_off(dev
);
5151 if (netif_carrier_ok(rootdev
)) {
5152 if (!netif_carrier_ok(dev
))
5153 netif_carrier_on(dev
);
5155 if (netif_carrier_ok(dev
))
5156 netif_carrier_off(dev
);
5159 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
5162 static int netif_alloc_rx_queues(struct net_device
*dev
)
5164 unsigned int i
, count
= dev
->num_rx_queues
;
5165 struct netdev_rx_queue
*rx
;
5169 rx
= kcalloc(count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5175 for (i
= 0; i
< count
; i
++)
5181 static void netdev_init_one_queue(struct net_device
*dev
,
5182 struct netdev_queue
*queue
, void *_unused
)
5184 /* Initialize queue lock */
5185 spin_lock_init(&queue
->_xmit_lock
);
5186 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
5187 queue
->xmit_lock_owner
= -1;
5188 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
5191 dql_init(&queue
->dql
, HZ
);
5195 static int netif_alloc_netdev_queues(struct net_device
*dev
)
5197 unsigned int count
= dev
->num_tx_queues
;
5198 struct netdev_queue
*tx
;
5202 tx
= kcalloc(count
, sizeof(struct netdev_queue
), GFP_KERNEL
);
5208 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5209 spin_lock_init(&dev
->tx_global_lock
);
5215 * register_netdevice - register a network device
5216 * @dev: device to register
5218 * Take a completed network device structure and add it to the kernel
5219 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5220 * chain. 0 is returned on success. A negative errno code is returned
5221 * on a failure to set up the device, or if the name is a duplicate.
5223 * Callers must hold the rtnl semaphore. You may want
5224 * register_netdev() instead of this.
5227 * The locking appears insufficient to guarantee two parallel registers
5228 * will not get the same name.
5231 int register_netdevice(struct net_device
*dev
)
5234 struct net
*net
= dev_net(dev
);
5236 BUG_ON(dev_boot_phase
);
5241 /* When net_device's are persistent, this will be fatal. */
5242 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
5245 spin_lock_init(&dev
->addr_list_lock
);
5246 netdev_set_addr_lockdep_class(dev
);
5250 ret
= dev_get_valid_name(net
, dev
, dev
->name
);
5254 /* Init, if this function is available */
5255 if (dev
->netdev_ops
->ndo_init
) {
5256 ret
= dev
->netdev_ops
->ndo_init(dev
);
5264 if (((dev
->hw_features
| dev
->features
) &
5265 NETIF_F_HW_VLAN_CTAG_FILTER
) &&
5266 (!dev
->netdev_ops
->ndo_vlan_rx_add_vid
||
5267 !dev
->netdev_ops
->ndo_vlan_rx_kill_vid
)) {
5268 netdev_WARN(dev
, "Buggy VLAN acceleration in driver!\n");
5275 dev
->ifindex
= dev_new_index(net
);
5276 else if (__dev_get_by_index(net
, dev
->ifindex
))
5279 if (dev
->iflink
== -1)
5280 dev
->iflink
= dev
->ifindex
;
5282 /* Transfer changeable features to wanted_features and enable
5283 * software offloads (GSO and GRO).
5285 dev
->hw_features
|= NETIF_F_SOFT_FEATURES
;
5286 dev
->features
|= NETIF_F_SOFT_FEATURES
;
5287 dev
->wanted_features
= dev
->features
& dev
->hw_features
;
5289 /* Turn on no cache copy if HW is doing checksum */
5290 if (!(dev
->flags
& IFF_LOOPBACK
)) {
5291 dev
->hw_features
|= NETIF_F_NOCACHE_COPY
;
5292 if (dev
->features
& NETIF_F_ALL_CSUM
) {
5293 dev
->wanted_features
|= NETIF_F_NOCACHE_COPY
;
5294 dev
->features
|= NETIF_F_NOCACHE_COPY
;
5298 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5300 dev
->vlan_features
|= NETIF_F_HIGHDMA
;
5302 /* Make NETIF_F_SG inheritable to tunnel devices.
5304 dev
->hw_enc_features
|= NETIF_F_SG
;
5306 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
5307 ret
= notifier_to_errno(ret
);
5311 ret
= netdev_register_kobject(dev
);
5314 dev
->reg_state
= NETREG_REGISTERED
;
5316 __netdev_update_features(dev
);
5319 * Default initial state at registry is that the
5320 * device is present.
5323 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5325 linkwatch_init_dev(dev
);
5327 dev_init_scheduler(dev
);
5329 list_netdevice(dev
);
5330 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
5332 /* If the device has permanent device address, driver should
5333 * set dev_addr and also addr_assign_type should be set to
5334 * NET_ADDR_PERM (default value).
5336 if (dev
->addr_assign_type
== NET_ADDR_PERM
)
5337 memcpy(dev
->perm_addr
, dev
->dev_addr
, dev
->addr_len
);
5339 /* Notify protocols, that a new device appeared. */
5340 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5341 ret
= notifier_to_errno(ret
);
5343 rollback_registered(dev
);
5344 dev
->reg_state
= NETREG_UNREGISTERED
;
5347 * Prevent userspace races by waiting until the network
5348 * device is fully setup before sending notifications.
5350 if (!dev
->rtnl_link_ops
||
5351 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5352 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5358 if (dev
->netdev_ops
->ndo_uninit
)
5359 dev
->netdev_ops
->ndo_uninit(dev
);
5362 EXPORT_SYMBOL(register_netdevice
);
5365 * init_dummy_netdev - init a dummy network device for NAPI
5366 * @dev: device to init
5368 * This takes a network device structure and initialize the minimum
5369 * amount of fields so it can be used to schedule NAPI polls without
5370 * registering a full blown interface. This is to be used by drivers
5371 * that need to tie several hardware interfaces to a single NAPI
5372 * poll scheduler due to HW limitations.
5374 int init_dummy_netdev(struct net_device
*dev
)
5376 /* Clear everything. Note we don't initialize spinlocks
5377 * are they aren't supposed to be taken by any of the
5378 * NAPI code and this dummy netdev is supposed to be
5379 * only ever used for NAPI polls
5381 memset(dev
, 0, sizeof(struct net_device
));
5383 /* make sure we BUG if trying to hit standard
5384 * register/unregister code path
5386 dev
->reg_state
= NETREG_DUMMY
;
5388 /* NAPI wants this */
5389 INIT_LIST_HEAD(&dev
->napi_list
);
5391 /* a dummy interface is started by default */
5392 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5393 set_bit(__LINK_STATE_START
, &dev
->state
);
5395 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5396 * because users of this 'device' dont need to change
5402 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
5406 * register_netdev - register a network device
5407 * @dev: device to register
5409 * Take a completed network device structure and add it to the kernel
5410 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5411 * chain. 0 is returned on success. A negative errno code is returned
5412 * on a failure to set up the device, or if the name is a duplicate.
5414 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5415 * and expands the device name if you passed a format string to
5418 int register_netdev(struct net_device
*dev
)
5423 err
= register_netdevice(dev
);
5427 EXPORT_SYMBOL(register_netdev
);
5429 int netdev_refcnt_read(const struct net_device
*dev
)
5433 for_each_possible_cpu(i
)
5434 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
5437 EXPORT_SYMBOL(netdev_refcnt_read
);
5440 * netdev_wait_allrefs - wait until all references are gone.
5441 * @dev: target net_device
5443 * This is called when unregistering network devices.
5445 * Any protocol or device that holds a reference should register
5446 * for netdevice notification, and cleanup and put back the
5447 * reference if they receive an UNREGISTER event.
5448 * We can get stuck here if buggy protocols don't correctly
5451 static void netdev_wait_allrefs(struct net_device
*dev
)
5453 unsigned long rebroadcast_time
, warning_time
;
5456 linkwatch_forget_dev(dev
);
5458 rebroadcast_time
= warning_time
= jiffies
;
5459 refcnt
= netdev_refcnt_read(dev
);
5461 while (refcnt
!= 0) {
5462 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
5465 /* Rebroadcast unregister notification */
5466 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5472 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
5473 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
5475 /* We must not have linkwatch events
5476 * pending on unregister. If this
5477 * happens, we simply run the queue
5478 * unscheduled, resulting in a noop
5481 linkwatch_run_queue();
5486 rebroadcast_time
= jiffies
;
5491 refcnt
= netdev_refcnt_read(dev
);
5493 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
5494 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5496 warning_time
= jiffies
;
5505 * register_netdevice(x1);
5506 * register_netdevice(x2);
5508 * unregister_netdevice(y1);
5509 * unregister_netdevice(y2);
5515 * We are invoked by rtnl_unlock().
5516 * This allows us to deal with problems:
5517 * 1) We can delete sysfs objects which invoke hotplug
5518 * without deadlocking with linkwatch via keventd.
5519 * 2) Since we run with the RTNL semaphore not held, we can sleep
5520 * safely in order to wait for the netdev refcnt to drop to zero.
5522 * We must not return until all unregister events added during
5523 * the interval the lock was held have been completed.
5525 void netdev_run_todo(void)
5527 struct list_head list
;
5529 /* Snapshot list, allow later requests */
5530 list_replace_init(&net_todo_list
, &list
);
5535 /* Wait for rcu callbacks to finish before next phase */
5536 if (!list_empty(&list
))
5539 while (!list_empty(&list
)) {
5540 struct net_device
*dev
5541 = list_first_entry(&list
, struct net_device
, todo_list
);
5542 list_del(&dev
->todo_list
);
5545 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
5548 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
5549 pr_err("network todo '%s' but state %d\n",
5550 dev
->name
, dev
->reg_state
);
5555 dev
->reg_state
= NETREG_UNREGISTERED
;
5557 on_each_cpu(flush_backlog
, dev
, 1);
5559 netdev_wait_allrefs(dev
);
5562 BUG_ON(netdev_refcnt_read(dev
));
5563 WARN_ON(rcu_access_pointer(dev
->ip_ptr
));
5564 WARN_ON(rcu_access_pointer(dev
->ip6_ptr
));
5565 WARN_ON(dev
->dn_ptr
);
5567 if (dev
->destructor
)
5568 dev
->destructor(dev
);
5570 /* Free network device */
5571 kobject_put(&dev
->dev
.kobj
);
5575 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5576 * fields in the same order, with only the type differing.
5578 void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
5579 const struct net_device_stats
*netdev_stats
)
5581 #if BITS_PER_LONG == 64
5582 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
5583 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
5585 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
5586 const unsigned long *src
= (const unsigned long *)netdev_stats
;
5587 u64
*dst
= (u64
*)stats64
;
5589 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
5590 sizeof(*stats64
) / sizeof(u64
));
5591 for (i
= 0; i
< n
; i
++)
5595 EXPORT_SYMBOL(netdev_stats_to_stats64
);
5598 * dev_get_stats - get network device statistics
5599 * @dev: device to get statistics from
5600 * @storage: place to store stats
5602 * Get network statistics from device. Return @storage.
5603 * The device driver may provide its own method by setting
5604 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5605 * otherwise the internal statistics structure is used.
5607 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
5608 struct rtnl_link_stats64
*storage
)
5610 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5612 if (ops
->ndo_get_stats64
) {
5613 memset(storage
, 0, sizeof(*storage
));
5614 ops
->ndo_get_stats64(dev
, storage
);
5615 } else if (ops
->ndo_get_stats
) {
5616 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
5618 netdev_stats_to_stats64(storage
, &dev
->stats
);
5620 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
5623 EXPORT_SYMBOL(dev_get_stats
);
5625 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
5627 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
5629 #ifdef CONFIG_NET_CLS_ACT
5632 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
5635 netdev_init_one_queue(dev
, queue
, NULL
);
5636 queue
->qdisc
= &noop_qdisc
;
5637 queue
->qdisc_sleeping
= &noop_qdisc
;
5638 rcu_assign_pointer(dev
->ingress_queue
, queue
);
5643 static const struct ethtool_ops default_ethtool_ops
;
5645 void netdev_set_default_ethtool_ops(struct net_device
*dev
,
5646 const struct ethtool_ops
*ops
)
5648 if (dev
->ethtool_ops
== &default_ethtool_ops
)
5649 dev
->ethtool_ops
= ops
;
5651 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops
);
5654 * alloc_netdev_mqs - allocate network device
5655 * @sizeof_priv: size of private data to allocate space for
5656 * @name: device name format string
5657 * @setup: callback to initialize device
5658 * @txqs: the number of TX subqueues to allocate
5659 * @rxqs: the number of RX subqueues to allocate
5661 * Allocates a struct net_device with private data area for driver use
5662 * and performs basic initialization. Also allocates subquue structs
5663 * for each queue on the device.
5665 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
5666 void (*setup
)(struct net_device
*),
5667 unsigned int txqs
, unsigned int rxqs
)
5669 struct net_device
*dev
;
5671 struct net_device
*p
;
5673 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
5676 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
5682 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
5687 alloc_size
= sizeof(struct net_device
);
5689 /* ensure 32-byte alignment of private area */
5690 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
5691 alloc_size
+= sizeof_priv
;
5693 /* ensure 32-byte alignment of whole construct */
5694 alloc_size
+= NETDEV_ALIGN
- 1;
5696 p
= kzalloc(alloc_size
, GFP_KERNEL
);
5700 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
5701 dev
->padded
= (char *)dev
- (char *)p
;
5703 dev
->pcpu_refcnt
= alloc_percpu(int);
5704 if (!dev
->pcpu_refcnt
)
5707 if (dev_addr_init(dev
))
5713 dev_net_set(dev
, &init_net
);
5715 dev
->gso_max_size
= GSO_MAX_SIZE
;
5716 dev
->gso_max_segs
= GSO_MAX_SEGS
;
5718 INIT_LIST_HEAD(&dev
->napi_list
);
5719 INIT_LIST_HEAD(&dev
->unreg_list
);
5720 INIT_LIST_HEAD(&dev
->link_watch_list
);
5721 INIT_LIST_HEAD(&dev
->upper_dev_list
);
5722 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
5725 dev
->num_tx_queues
= txqs
;
5726 dev
->real_num_tx_queues
= txqs
;
5727 if (netif_alloc_netdev_queues(dev
))
5731 dev
->num_rx_queues
= rxqs
;
5732 dev
->real_num_rx_queues
= rxqs
;
5733 if (netif_alloc_rx_queues(dev
))
5737 strcpy(dev
->name
, name
);
5738 dev
->group
= INIT_NETDEV_GROUP
;
5739 if (!dev
->ethtool_ops
)
5740 dev
->ethtool_ops
= &default_ethtool_ops
;
5748 free_percpu(dev
->pcpu_refcnt
);
5758 EXPORT_SYMBOL(alloc_netdev_mqs
);
5761 * free_netdev - free network device
5764 * This function does the last stage of destroying an allocated device
5765 * interface. The reference to the device object is released.
5766 * If this is the last reference then it will be freed.
5768 void free_netdev(struct net_device
*dev
)
5770 struct napi_struct
*p
, *n
;
5772 release_net(dev_net(dev
));
5779 kfree(rcu_dereference_protected(dev
->ingress_queue
, 1));
5781 /* Flush device addresses */
5782 dev_addr_flush(dev
);
5784 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
5787 free_percpu(dev
->pcpu_refcnt
);
5788 dev
->pcpu_refcnt
= NULL
;
5790 /* Compatibility with error handling in drivers */
5791 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5792 kfree((char *)dev
- dev
->padded
);
5796 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
5797 dev
->reg_state
= NETREG_RELEASED
;
5799 /* will free via device release */
5800 put_device(&dev
->dev
);
5802 EXPORT_SYMBOL(free_netdev
);
5805 * synchronize_net - Synchronize with packet receive processing
5807 * Wait for packets currently being received to be done.
5808 * Does not block later packets from starting.
5810 void synchronize_net(void)
5813 if (rtnl_is_locked())
5814 synchronize_rcu_expedited();
5818 EXPORT_SYMBOL(synchronize_net
);
5821 * unregister_netdevice_queue - remove device from the kernel
5825 * This function shuts down a device interface and removes it
5826 * from the kernel tables.
5827 * If head not NULL, device is queued to be unregistered later.
5829 * Callers must hold the rtnl semaphore. You may want
5830 * unregister_netdev() instead of this.
5833 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
5838 list_move_tail(&dev
->unreg_list
, head
);
5840 rollback_registered(dev
);
5841 /* Finish processing unregister after unlock */
5845 EXPORT_SYMBOL(unregister_netdevice_queue
);
5848 * unregister_netdevice_many - unregister many devices
5849 * @head: list of devices
5851 * Note: As most callers use a stack allocated list_head,
5852 * we force a list_del() to make sure stack wont be corrupted later.
5854 void unregister_netdevice_many(struct list_head
*head
)
5856 struct net_device
*dev
;
5858 if (!list_empty(head
)) {
5859 rollback_registered_many(head
);
5860 list_for_each_entry(dev
, head
, unreg_list
)
5865 EXPORT_SYMBOL(unregister_netdevice_many
);
5868 * unregister_netdev - remove device from the kernel
5871 * This function shuts down a device interface and removes it
5872 * from the kernel tables.
5874 * This is just a wrapper for unregister_netdevice that takes
5875 * the rtnl semaphore. In general you want to use this and not
5876 * unregister_netdevice.
5878 void unregister_netdev(struct net_device
*dev
)
5881 unregister_netdevice(dev
);
5884 EXPORT_SYMBOL(unregister_netdev
);
5887 * dev_change_net_namespace - move device to different nethost namespace
5889 * @net: network namespace
5890 * @pat: If not NULL name pattern to try if the current device name
5891 * is already taken in the destination network namespace.
5893 * This function shuts down a device interface and moves it
5894 * to a new network namespace. On success 0 is returned, on
5895 * a failure a netagive errno code is returned.
5897 * Callers must hold the rtnl semaphore.
5900 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
5906 /* Don't allow namespace local devices to be moved. */
5908 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
5911 /* Ensure the device has been registrered */
5912 if (dev
->reg_state
!= NETREG_REGISTERED
)
5915 /* Get out if there is nothing todo */
5917 if (net_eq(dev_net(dev
), net
))
5920 /* Pick the destination device name, and ensure
5921 * we can use it in the destination network namespace.
5924 if (__dev_get_by_name(net
, dev
->name
)) {
5925 /* We get here if we can't use the current device name */
5928 if (dev_get_valid_name(net
, dev
, pat
) < 0)
5933 * And now a mini version of register_netdevice unregister_netdevice.
5936 /* If device is running close it first. */
5939 /* And unlink it from device chain */
5941 unlist_netdevice(dev
);
5945 /* Shutdown queueing discipline. */
5948 /* Notify protocols, that we are about to destroy
5949 this device. They should clean all the things.
5951 Note that dev->reg_state stays at NETREG_REGISTERED.
5952 This is wanted because this way 8021q and macvlan know
5953 the device is just moving and can keep their slaves up.
5955 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5957 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
5958 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
5961 * Flush the unicast and multicast chains
5966 /* Send a netdev-removed uevent to the old namespace */
5967 kobject_uevent(&dev
->dev
.kobj
, KOBJ_REMOVE
);
5969 /* Actually switch the network namespace */
5970 dev_net_set(dev
, net
);
5972 /* If there is an ifindex conflict assign a new one */
5973 if (__dev_get_by_index(net
, dev
->ifindex
)) {
5974 int iflink
= (dev
->iflink
== dev
->ifindex
);
5975 dev
->ifindex
= dev_new_index(net
);
5977 dev
->iflink
= dev
->ifindex
;
5980 /* Send a netdev-add uevent to the new namespace */
5981 kobject_uevent(&dev
->dev
.kobj
, KOBJ_ADD
);
5983 /* Fixup kobjects */
5984 err
= device_rename(&dev
->dev
, dev
->name
);
5987 /* Add the device back in the hashes */
5988 list_netdevice(dev
);
5990 /* Notify protocols, that a new device appeared. */
5991 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5994 * Prevent userspace races by waiting until the network
5995 * device is fully setup before sending notifications.
5997 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
6004 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
6006 static int dev_cpu_callback(struct notifier_block
*nfb
,
6007 unsigned long action
,
6010 struct sk_buff
**list_skb
;
6011 struct sk_buff
*skb
;
6012 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
6013 struct softnet_data
*sd
, *oldsd
;
6015 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
6018 local_irq_disable();
6019 cpu
= smp_processor_id();
6020 sd
= &per_cpu(softnet_data
, cpu
);
6021 oldsd
= &per_cpu(softnet_data
, oldcpu
);
6023 /* Find end of our completion_queue. */
6024 list_skb
= &sd
->completion_queue
;
6026 list_skb
= &(*list_skb
)->next
;
6027 /* Append completion queue from offline CPU. */
6028 *list_skb
= oldsd
->completion_queue
;
6029 oldsd
->completion_queue
= NULL
;
6031 /* Append output queue from offline CPU. */
6032 if (oldsd
->output_queue
) {
6033 *sd
->output_queue_tailp
= oldsd
->output_queue
;
6034 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
6035 oldsd
->output_queue
= NULL
;
6036 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
6038 /* Append NAPI poll list from offline CPU, with one exception :
6039 * process_backlog() must be called by cpu owning percpu backlog.
6040 * We properly handle process_queue & input_pkt_queue later.
6042 while (!list_empty(&oldsd
->poll_list
)) {
6043 struct napi_struct
*napi
= list_first_entry(&oldsd
->poll_list
,
6047 list_del_init(&napi
->poll_list
);
6048 if (napi
->poll
== process_backlog
)
6051 ____napi_schedule(sd
, napi
);
6054 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
6057 /* Process offline CPU's input_pkt_queue */
6058 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
6060 input_queue_head_incr(oldsd
);
6062 while ((skb
= skb_dequeue(&oldsd
->input_pkt_queue
))) {
6064 input_queue_head_incr(oldsd
);
6072 * netdev_increment_features - increment feature set by one
6073 * @all: current feature set
6074 * @one: new feature set
6075 * @mask: mask feature set
6077 * Computes a new feature set after adding a device with feature set
6078 * @one to the master device with current feature set @all. Will not
6079 * enable anything that is off in @mask. Returns the new feature set.
6081 netdev_features_t
netdev_increment_features(netdev_features_t all
,
6082 netdev_features_t one
, netdev_features_t mask
)
6084 if (mask
& NETIF_F_GEN_CSUM
)
6085 mask
|= NETIF_F_ALL_CSUM
;
6086 mask
|= NETIF_F_VLAN_CHALLENGED
;
6088 all
|= one
& (NETIF_F_ONE_FOR_ALL
|NETIF_F_ALL_CSUM
) & mask
;
6089 all
&= one
| ~NETIF_F_ALL_FOR_ALL
;
6091 /* If one device supports hw checksumming, set for all. */
6092 if (all
& NETIF_F_GEN_CSUM
)
6093 all
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_GEN_CSUM
);
6097 EXPORT_SYMBOL(netdev_increment_features
);
6099 static struct hlist_head
*netdev_create_hash(void)
6102 struct hlist_head
*hash
;
6104 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
6106 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
6107 INIT_HLIST_HEAD(&hash
[i
]);
6112 /* Initialize per network namespace state */
6113 static int __net_init
netdev_init(struct net
*net
)
6115 if (net
!= &init_net
)
6116 INIT_LIST_HEAD(&net
->dev_base_head
);
6118 net
->dev_name_head
= netdev_create_hash();
6119 if (net
->dev_name_head
== NULL
)
6122 net
->dev_index_head
= netdev_create_hash();
6123 if (net
->dev_index_head
== NULL
)
6129 kfree(net
->dev_name_head
);
6135 * netdev_drivername - network driver for the device
6136 * @dev: network device
6138 * Determine network driver for device.
6140 const char *netdev_drivername(const struct net_device
*dev
)
6142 const struct device_driver
*driver
;
6143 const struct device
*parent
;
6144 const char *empty
= "";
6146 parent
= dev
->dev
.parent
;
6150 driver
= parent
->driver
;
6151 if (driver
&& driver
->name
)
6152 return driver
->name
;
6156 static int __netdev_printk(const char *level
, const struct net_device
*dev
,
6157 struct va_format
*vaf
)
6161 if (dev
&& dev
->dev
.parent
) {
6162 r
= dev_printk_emit(level
[1] - '0',
6165 dev_driver_string(dev
->dev
.parent
),
6166 dev_name(dev
->dev
.parent
),
6167 netdev_name(dev
), vaf
);
6169 r
= printk("%s%s: %pV", level
, netdev_name(dev
), vaf
);
6171 r
= printk("%s(NULL net_device): %pV", level
, vaf
);
6177 int netdev_printk(const char *level
, const struct net_device
*dev
,
6178 const char *format
, ...)
6180 struct va_format vaf
;
6184 va_start(args
, format
);
6189 r
= __netdev_printk(level
, dev
, &vaf
);
6195 EXPORT_SYMBOL(netdev_printk
);
6197 #define define_netdev_printk_level(func, level) \
6198 int func(const struct net_device *dev, const char *fmt, ...) \
6201 struct va_format vaf; \
6204 va_start(args, fmt); \
6209 r = __netdev_printk(level, dev, &vaf); \
6215 EXPORT_SYMBOL(func);
6217 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
6218 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
6219 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
6220 define_netdev_printk_level(netdev_err
, KERN_ERR
);
6221 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
6222 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
6223 define_netdev_printk_level(netdev_info
, KERN_INFO
);
6225 static void __net_exit
netdev_exit(struct net
*net
)
6227 kfree(net
->dev_name_head
);
6228 kfree(net
->dev_index_head
);
6231 static struct pernet_operations __net_initdata netdev_net_ops
= {
6232 .init
= netdev_init
,
6233 .exit
= netdev_exit
,
6236 static void __net_exit
default_device_exit(struct net
*net
)
6238 struct net_device
*dev
, *aux
;
6240 * Push all migratable network devices back to the
6241 * initial network namespace
6244 for_each_netdev_safe(net
, dev
, aux
) {
6246 char fb_name
[IFNAMSIZ
];
6248 /* Ignore unmoveable devices (i.e. loopback) */
6249 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6252 /* Leave virtual devices for the generic cleanup */
6253 if (dev
->rtnl_link_ops
)
6256 /* Push remaining network devices to init_net */
6257 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
6258 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
6260 pr_emerg("%s: failed to move %s to init_net: %d\n",
6261 __func__
, dev
->name
, err
);
6268 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
6270 /* At exit all network devices most be removed from a network
6271 * namespace. Do this in the reverse order of registration.
6272 * Do this across as many network namespaces as possible to
6273 * improve batching efficiency.
6275 struct net_device
*dev
;
6277 LIST_HEAD(dev_kill_list
);
6280 list_for_each_entry(net
, net_list
, exit_list
) {
6281 for_each_netdev_reverse(net
, dev
) {
6282 if (dev
->rtnl_link_ops
)
6283 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
6285 unregister_netdevice_queue(dev
, &dev_kill_list
);
6288 unregister_netdevice_many(&dev_kill_list
);
6292 static struct pernet_operations __net_initdata default_device_ops
= {
6293 .exit
= default_device_exit
,
6294 .exit_batch
= default_device_exit_batch
,
6298 * Initialize the DEV module. At boot time this walks the device list and
6299 * unhooks any devices that fail to initialise (normally hardware not
6300 * present) and leaves us with a valid list of present and active devices.
6305 * This is called single threaded during boot, so no need
6306 * to take the rtnl semaphore.
6308 static int __init
net_dev_init(void)
6310 int i
, rc
= -ENOMEM
;
6312 BUG_ON(!dev_boot_phase
);
6314 if (dev_proc_init())
6317 if (netdev_kobject_init())
6320 INIT_LIST_HEAD(&ptype_all
);
6321 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
6322 INIT_LIST_HEAD(&ptype_base
[i
]);
6324 INIT_LIST_HEAD(&offload_base
);
6326 if (register_pernet_subsys(&netdev_net_ops
))
6330 * Initialise the packet receive queues.
6333 for_each_possible_cpu(i
) {
6334 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
6336 memset(sd
, 0, sizeof(*sd
));
6337 skb_queue_head_init(&sd
->input_pkt_queue
);
6338 skb_queue_head_init(&sd
->process_queue
);
6339 sd
->completion_queue
= NULL
;
6340 INIT_LIST_HEAD(&sd
->poll_list
);
6341 sd
->output_queue
= NULL
;
6342 sd
->output_queue_tailp
= &sd
->output_queue
;
6344 sd
->csd
.func
= rps_trigger_softirq
;
6350 sd
->backlog
.poll
= process_backlog
;
6351 sd
->backlog
.weight
= weight_p
;
6352 sd
->backlog
.gro_list
= NULL
;
6353 sd
->backlog
.gro_count
= 0;
6358 /* The loopback device is special if any other network devices
6359 * is present in a network namespace the loopback device must
6360 * be present. Since we now dynamically allocate and free the
6361 * loopback device ensure this invariant is maintained by
6362 * keeping the loopback device as the first device on the
6363 * list of network devices. Ensuring the loopback devices
6364 * is the first device that appears and the last network device
6367 if (register_pernet_device(&loopback_net_ops
))
6370 if (register_pernet_device(&default_device_ops
))
6373 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
6374 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
6376 hotcpu_notifier(dev_cpu_callback
, 0);
6383 subsys_initcall(net_dev_init
);