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/proc_fs.h>
101 #include <linux/seq_file.h>
102 #include <linux/stat.h>
104 #include <net/pkt_sched.h>
105 #include <net/checksum.h>
106 #include <net/xfrm.h>
107 #include <linux/highmem.h>
108 #include <linux/init.h>
109 #include <linux/kmod.h>
110 #include <linux/module.h>
111 #include <linux/netpoll.h>
112 #include <linux/rcupdate.h>
113 #include <linux/delay.h>
114 #include <net/wext.h>
115 #include <net/iw_handler.h>
116 #include <asm/current.h>
117 #include <linux/audit.h>
118 #include <linux/dmaengine.h>
119 #include <linux/err.h>
120 #include <linux/ctype.h>
121 #include <linux/if_arp.h>
122 #include <linux/if_vlan.h>
123 #include <linux/ip.h>
125 #include <linux/ipv6.h>
126 #include <linux/in.h>
127 #include <linux/jhash.h>
128 #include <linux/random.h>
129 #include <trace/events/napi.h>
130 #include <trace/events/net.h>
131 #include <trace/events/skb.h>
132 #include <linux/pci.h>
133 #include <linux/inetdevice.h>
134 #include <linux/cpu_rmap.h>
135 #include <linux/net_tstamp.h>
136 #include <linux/static_key.h>
137 #include <net/flow_keys.h>
139 #include "net-sysfs.h"
141 /* Instead of increasing this, you should create a hash table. */
142 #define MAX_GRO_SKBS 8
144 /* This should be increased if a protocol with a bigger head is added. */
145 #define GRO_MAX_HEAD (MAX_HEADER + 128)
148 * The list of packet types we will receive (as opposed to discard)
149 * and the routines to invoke.
151 * Why 16. Because with 16 the only overlap we get on a hash of the
152 * low nibble of the protocol value is RARP/SNAP/X.25.
154 * NOTE: That is no longer true with the addition of VLAN tags. Not
155 * sure which should go first, but I bet it won't make much
156 * difference if we are running VLANs. The good news is that
157 * this protocol won't be in the list unless compiled in, so
158 * the average user (w/out VLANs) will not be adversely affected.
175 #define PTYPE_HASH_SIZE (16)
176 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
178 static DEFINE_SPINLOCK(ptype_lock
);
179 static struct list_head ptype_base
[PTYPE_HASH_SIZE
] __read_mostly
;
180 static struct list_head ptype_all __read_mostly
; /* Taps */
183 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
186 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
188 * Writers must hold the rtnl semaphore while they loop through the
189 * dev_base_head list, and hold dev_base_lock for writing when they do the
190 * actual updates. This allows pure readers to access the list even
191 * while a writer is preparing to update it.
193 * To put it another way, dev_base_lock is held for writing only to
194 * protect against pure readers; the rtnl semaphore provides the
195 * protection against other writers.
197 * See, for example usages, register_netdevice() and
198 * unregister_netdevice(), which must be called with the rtnl
201 DEFINE_RWLOCK(dev_base_lock
);
202 EXPORT_SYMBOL(dev_base_lock
);
204 static inline void dev_base_seq_inc(struct net
*net
)
206 while (++net
->dev_base_seq
== 0);
209 static inline struct hlist_head
*dev_name_hash(struct net
*net
, const char *name
)
211 unsigned int hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
213 return &net
->dev_name_head
[hash_32(hash
, NETDEV_HASHBITS
)];
216 static inline struct hlist_head
*dev_index_hash(struct net
*net
, int ifindex
)
218 return &net
->dev_index_head
[ifindex
& (NETDEV_HASHENTRIES
- 1)];
221 static inline void rps_lock(struct softnet_data
*sd
)
224 spin_lock(&sd
->input_pkt_queue
.lock
);
228 static inline void rps_unlock(struct softnet_data
*sd
)
231 spin_unlock(&sd
->input_pkt_queue
.lock
);
235 /* Device list insertion */
236 static int list_netdevice(struct net_device
*dev
)
238 struct net
*net
= dev_net(dev
);
242 write_lock_bh(&dev_base_lock
);
243 list_add_tail_rcu(&dev
->dev_list
, &net
->dev_base_head
);
244 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
245 hlist_add_head_rcu(&dev
->index_hlist
,
246 dev_index_hash(net
, dev
->ifindex
));
247 write_unlock_bh(&dev_base_lock
);
249 dev_base_seq_inc(net
);
254 /* Device list removal
255 * caller must respect a RCU grace period before freeing/reusing dev
257 static void unlist_netdevice(struct net_device
*dev
)
261 /* Unlink dev from the device chain */
262 write_lock_bh(&dev_base_lock
);
263 list_del_rcu(&dev
->dev_list
);
264 hlist_del_rcu(&dev
->name_hlist
);
265 hlist_del_rcu(&dev
->index_hlist
);
266 write_unlock_bh(&dev_base_lock
);
268 dev_base_seq_inc(dev_net(dev
));
275 static RAW_NOTIFIER_HEAD(netdev_chain
);
278 * Device drivers call our routines to queue packets here. We empty the
279 * queue in the local softnet handler.
282 DEFINE_PER_CPU_ALIGNED(struct softnet_data
, softnet_data
);
283 EXPORT_PER_CPU_SYMBOL(softnet_data
);
285 #ifdef CONFIG_LOCKDEP
287 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
288 * according to dev->type
290 static const unsigned short netdev_lock_type
[] =
291 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
292 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
293 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
294 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
295 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
296 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
297 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
298 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
299 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
300 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
301 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
302 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
303 ARPHRD_FCFABRIC
, ARPHRD_IEEE80211
, ARPHRD_IEEE80211_PRISM
,
304 ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_PHONET
, ARPHRD_PHONET_PIPE
,
305 ARPHRD_IEEE802154
, ARPHRD_VOID
, ARPHRD_NONE
};
307 static const char *const netdev_lock_name
[] =
308 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
309 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
310 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
311 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
312 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
313 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
314 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
315 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
316 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
317 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
318 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
319 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
320 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
321 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
322 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
324 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
325 static struct lock_class_key netdev_addr_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
327 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
331 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
332 if (netdev_lock_type
[i
] == dev_type
)
334 /* the last key is used by default */
335 return ARRAY_SIZE(netdev_lock_type
) - 1;
338 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
339 unsigned short dev_type
)
343 i
= netdev_lock_pos(dev_type
);
344 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
345 netdev_lock_name
[i
]);
348 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
352 i
= netdev_lock_pos(dev
->type
);
353 lockdep_set_class_and_name(&dev
->addr_list_lock
,
354 &netdev_addr_lock_key
[i
],
355 netdev_lock_name
[i
]);
358 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
359 unsigned short dev_type
)
362 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
367 /*******************************************************************************
369 Protocol management and registration routines
371 *******************************************************************************/
374 * Add a protocol ID to the list. Now that the input handler is
375 * smarter we can dispense with all the messy stuff that used to be
378 * BEWARE!!! Protocol handlers, mangling input packets,
379 * MUST BE last in hash buckets and checking protocol handlers
380 * MUST start from promiscuous ptype_all chain in net_bh.
381 * It is true now, do not change it.
382 * Explanation follows: if protocol handler, mangling packet, will
383 * be the first on list, it is not able to sense, that packet
384 * is cloned and should be copied-on-write, so that it will
385 * change it and subsequent readers will get broken packet.
389 static inline struct list_head
*ptype_head(const struct packet_type
*pt
)
391 if (pt
->type
== htons(ETH_P_ALL
))
394 return &ptype_base
[ntohs(pt
->type
) & PTYPE_HASH_MASK
];
398 * dev_add_pack - add packet handler
399 * @pt: packet type declaration
401 * Add a protocol handler to the networking stack. The passed &packet_type
402 * is linked into kernel lists and may not be freed until it has been
403 * removed from the kernel lists.
405 * This call does not sleep therefore it can not
406 * guarantee all CPU's that are in middle of receiving packets
407 * will see the new packet type (until the next received packet).
410 void dev_add_pack(struct packet_type
*pt
)
412 struct list_head
*head
= ptype_head(pt
);
414 spin_lock(&ptype_lock
);
415 list_add_rcu(&pt
->list
, head
);
416 spin_unlock(&ptype_lock
);
418 EXPORT_SYMBOL(dev_add_pack
);
421 * __dev_remove_pack - remove packet handler
422 * @pt: packet type declaration
424 * Remove a protocol handler that was previously added to the kernel
425 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
426 * from the kernel lists and can be freed or reused once this function
429 * The packet type might still be in use by receivers
430 * and must not be freed until after all the CPU's have gone
431 * through a quiescent state.
433 void __dev_remove_pack(struct packet_type
*pt
)
435 struct list_head
*head
= ptype_head(pt
);
436 struct packet_type
*pt1
;
438 spin_lock(&ptype_lock
);
440 list_for_each_entry(pt1
, head
, list
) {
442 list_del_rcu(&pt
->list
);
447 pr_warn("dev_remove_pack: %p not found\n", pt
);
449 spin_unlock(&ptype_lock
);
451 EXPORT_SYMBOL(__dev_remove_pack
);
454 * dev_remove_pack - remove packet handler
455 * @pt: packet type declaration
457 * Remove a protocol handler that was previously added to the kernel
458 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
459 * from the kernel lists and can be freed or reused once this function
462 * This call sleeps to guarantee that no CPU is looking at the packet
465 void dev_remove_pack(struct packet_type
*pt
)
467 __dev_remove_pack(pt
);
471 EXPORT_SYMBOL(dev_remove_pack
);
473 /******************************************************************************
475 Device Boot-time Settings Routines
477 *******************************************************************************/
479 /* Boot time configuration table */
480 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
483 * netdev_boot_setup_add - add new setup entry
484 * @name: name of the device
485 * @map: configured settings for the device
487 * Adds new setup entry to the dev_boot_setup list. The function
488 * returns 0 on error and 1 on success. This is a generic routine to
491 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
493 struct netdev_boot_setup
*s
;
497 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
498 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
499 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
500 strlcpy(s
[i
].name
, name
, IFNAMSIZ
);
501 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
506 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
510 * netdev_boot_setup_check - check boot time settings
511 * @dev: the netdevice
513 * Check boot time settings for the device.
514 * The found settings are set for the device to be used
515 * later in the device probing.
516 * Returns 0 if no settings found, 1 if they are.
518 int netdev_boot_setup_check(struct net_device
*dev
)
520 struct netdev_boot_setup
*s
= dev_boot_setup
;
523 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
524 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
525 !strcmp(dev
->name
, s
[i
].name
)) {
526 dev
->irq
= s
[i
].map
.irq
;
527 dev
->base_addr
= s
[i
].map
.base_addr
;
528 dev
->mem_start
= s
[i
].map
.mem_start
;
529 dev
->mem_end
= s
[i
].map
.mem_end
;
535 EXPORT_SYMBOL(netdev_boot_setup_check
);
539 * netdev_boot_base - get address from boot time settings
540 * @prefix: prefix for network device
541 * @unit: id for network device
543 * Check boot time settings for the base address of device.
544 * The found settings are set for the device to be used
545 * later in the device probing.
546 * Returns 0 if no settings found.
548 unsigned long netdev_boot_base(const char *prefix
, int unit
)
550 const struct netdev_boot_setup
*s
= dev_boot_setup
;
554 sprintf(name
, "%s%d", prefix
, unit
);
557 * If device already registered then return base of 1
558 * to indicate not to probe for this interface
560 if (__dev_get_by_name(&init_net
, name
))
563 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
564 if (!strcmp(name
, s
[i
].name
))
565 return s
[i
].map
.base_addr
;
570 * Saves at boot time configured settings for any netdevice.
572 int __init
netdev_boot_setup(char *str
)
577 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
582 memset(&map
, 0, sizeof(map
));
586 map
.base_addr
= ints
[2];
588 map
.mem_start
= ints
[3];
590 map
.mem_end
= ints
[4];
592 /* Add new entry to the list */
593 return netdev_boot_setup_add(str
, &map
);
596 __setup("netdev=", netdev_boot_setup
);
598 /*******************************************************************************
600 Device Interface Subroutines
602 *******************************************************************************/
605 * __dev_get_by_name - find a device by its name
606 * @net: the applicable net namespace
607 * @name: name to find
609 * Find an interface by name. Must be called under RTNL semaphore
610 * or @dev_base_lock. If the name is found a pointer to the device
611 * is returned. If the name is not found then %NULL is returned. The
612 * reference counters are not incremented so the caller must be
613 * careful with locks.
616 struct net_device
*__dev_get_by_name(struct net
*net
, const char *name
)
618 struct hlist_node
*p
;
619 struct net_device
*dev
;
620 struct hlist_head
*head
= dev_name_hash(net
, name
);
622 hlist_for_each_entry(dev
, p
, head
, name_hlist
)
623 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
628 EXPORT_SYMBOL(__dev_get_by_name
);
631 * dev_get_by_name_rcu - find a device by its name
632 * @net: the applicable net namespace
633 * @name: name to find
635 * Find an interface by name.
636 * If the name is found a pointer to the device is returned.
637 * If the name is not found then %NULL is returned.
638 * The reference counters are not incremented so the caller must be
639 * careful with locks. The caller must hold RCU lock.
642 struct net_device
*dev_get_by_name_rcu(struct net
*net
, const char *name
)
644 struct hlist_node
*p
;
645 struct net_device
*dev
;
646 struct hlist_head
*head
= dev_name_hash(net
, name
);
648 hlist_for_each_entry_rcu(dev
, p
, head
, name_hlist
)
649 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
654 EXPORT_SYMBOL(dev_get_by_name_rcu
);
657 * dev_get_by_name - find a device by its name
658 * @net: the applicable net namespace
659 * @name: name to find
661 * Find an interface by name. This can be called from any
662 * context and does its own locking. The returned handle has
663 * the usage count incremented and the caller must use dev_put() to
664 * release it when it is no longer needed. %NULL is returned if no
665 * matching device is found.
668 struct net_device
*dev_get_by_name(struct net
*net
, const char *name
)
670 struct net_device
*dev
;
673 dev
= dev_get_by_name_rcu(net
, name
);
679 EXPORT_SYMBOL(dev_get_by_name
);
682 * __dev_get_by_index - find a device by its ifindex
683 * @net: the applicable net namespace
684 * @ifindex: index of device
686 * Search for an interface by index. Returns %NULL if the device
687 * is not found or a pointer to the device. The device has not
688 * had its reference counter increased so the caller must be careful
689 * about locking. The caller must hold either the RTNL semaphore
693 struct net_device
*__dev_get_by_index(struct net
*net
, int ifindex
)
695 struct hlist_node
*p
;
696 struct net_device
*dev
;
697 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
699 hlist_for_each_entry(dev
, p
, head
, index_hlist
)
700 if (dev
->ifindex
== ifindex
)
705 EXPORT_SYMBOL(__dev_get_by_index
);
708 * dev_get_by_index_rcu - find a device by its ifindex
709 * @net: the applicable net namespace
710 * @ifindex: index of device
712 * Search for an interface by index. Returns %NULL if the device
713 * is not found or a pointer to the device. The device has not
714 * had its reference counter increased so the caller must be careful
715 * about locking. The caller must hold RCU lock.
718 struct net_device
*dev_get_by_index_rcu(struct net
*net
, int ifindex
)
720 struct hlist_node
*p
;
721 struct net_device
*dev
;
722 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
724 hlist_for_each_entry_rcu(dev
, p
, head
, index_hlist
)
725 if (dev
->ifindex
== ifindex
)
730 EXPORT_SYMBOL(dev_get_by_index_rcu
);
734 * dev_get_by_index - find a device by its ifindex
735 * @net: the applicable net namespace
736 * @ifindex: index of device
738 * Search for an interface by index. Returns NULL if the device
739 * is not found or a pointer to the device. The device returned has
740 * had a reference added and the pointer is safe until the user calls
741 * dev_put to indicate they have finished with it.
744 struct net_device
*dev_get_by_index(struct net
*net
, int ifindex
)
746 struct net_device
*dev
;
749 dev
= dev_get_by_index_rcu(net
, ifindex
);
755 EXPORT_SYMBOL(dev_get_by_index
);
758 * dev_getbyhwaddr_rcu - find a device by its hardware address
759 * @net: the applicable net namespace
760 * @type: media type of device
761 * @ha: hardware address
763 * Search for an interface by MAC address. Returns NULL if the device
764 * is not found or a pointer to the device.
765 * The caller must hold RCU or RTNL.
766 * The returned device has not had its ref count increased
767 * and the caller must therefore be careful about locking
771 struct net_device
*dev_getbyhwaddr_rcu(struct net
*net
, unsigned short type
,
774 struct net_device
*dev
;
776 for_each_netdev_rcu(net
, dev
)
777 if (dev
->type
== type
&&
778 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
783 EXPORT_SYMBOL(dev_getbyhwaddr_rcu
);
785 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
787 struct net_device
*dev
;
790 for_each_netdev(net
, dev
)
791 if (dev
->type
== type
)
796 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
798 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
800 struct net_device
*dev
, *ret
= NULL
;
803 for_each_netdev_rcu(net
, dev
)
804 if (dev
->type
== type
) {
812 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
815 * dev_get_by_flags_rcu - find any device with given flags
816 * @net: the applicable net namespace
817 * @if_flags: IFF_* values
818 * @mask: bitmask of bits in if_flags to check
820 * Search for any interface with the given flags. Returns NULL if a device
821 * is not found or a pointer to the device. Must be called inside
822 * rcu_read_lock(), and result refcount is unchanged.
825 struct net_device
*dev_get_by_flags_rcu(struct net
*net
, unsigned short if_flags
,
828 struct net_device
*dev
, *ret
;
831 for_each_netdev_rcu(net
, dev
) {
832 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
839 EXPORT_SYMBOL(dev_get_by_flags_rcu
);
842 * dev_valid_name - check if name is okay for network device
845 * Network device names need to be valid file names to
846 * to allow sysfs to work. We also disallow any kind of
849 bool dev_valid_name(const char *name
)
853 if (strlen(name
) >= IFNAMSIZ
)
855 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
859 if (*name
== '/' || isspace(*name
))
865 EXPORT_SYMBOL(dev_valid_name
);
868 * __dev_alloc_name - allocate a name for a device
869 * @net: network namespace to allocate the device name in
870 * @name: name format string
871 * @buf: scratch buffer and result name string
873 * Passed a format string - eg "lt%d" it will try and find a suitable
874 * id. It scans list of devices to build up a free map, then chooses
875 * the first empty slot. The caller must hold the dev_base or rtnl lock
876 * while allocating the name and adding the device in order to avoid
878 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
879 * Returns the number of the unit assigned or a negative errno code.
882 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
886 const int max_netdevices
= 8*PAGE_SIZE
;
887 unsigned long *inuse
;
888 struct net_device
*d
;
890 p
= strnchr(name
, IFNAMSIZ
-1, '%');
893 * Verify the string as this thing may have come from
894 * the user. There must be either one "%d" and no other "%"
897 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
900 /* Use one page as a bit array of possible slots */
901 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
905 for_each_netdev(net
, d
) {
906 if (!sscanf(d
->name
, name
, &i
))
908 if (i
< 0 || i
>= max_netdevices
)
911 /* avoid cases where sscanf is not exact inverse of printf */
912 snprintf(buf
, IFNAMSIZ
, name
, i
);
913 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
917 i
= find_first_zero_bit(inuse
, max_netdevices
);
918 free_page((unsigned long) inuse
);
922 snprintf(buf
, IFNAMSIZ
, name
, i
);
923 if (!__dev_get_by_name(net
, buf
))
926 /* It is possible to run out of possible slots
927 * when the name is long and there isn't enough space left
928 * for the digits, or if all bits are used.
934 * dev_alloc_name - allocate a name for a device
936 * @name: name format string
938 * Passed a format string - eg "lt%d" it will try and find a suitable
939 * id. It scans list of devices to build up a free map, then chooses
940 * the first empty slot. The caller must hold the dev_base or rtnl lock
941 * while allocating the name and adding the device in order to avoid
943 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
944 * Returns the number of the unit assigned or a negative errno code.
947 int dev_alloc_name(struct net_device
*dev
, const char *name
)
953 BUG_ON(!dev_net(dev
));
955 ret
= __dev_alloc_name(net
, name
, buf
);
957 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
960 EXPORT_SYMBOL(dev_alloc_name
);
962 static int dev_alloc_name_ns(struct net
*net
,
963 struct net_device
*dev
,
969 ret
= __dev_alloc_name(net
, name
, buf
);
971 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
975 static int dev_get_valid_name(struct net
*net
,
976 struct net_device
*dev
,
981 if (!dev_valid_name(name
))
984 if (strchr(name
, '%'))
985 return dev_alloc_name_ns(net
, dev
, name
);
986 else if (__dev_get_by_name(net
, name
))
988 else if (dev
->name
!= name
)
989 strlcpy(dev
->name
, name
, IFNAMSIZ
);
995 * dev_change_name - change name of a device
997 * @newname: name (or format string) must be at least IFNAMSIZ
999 * Change name of a device, can pass format strings "eth%d".
1002 int dev_change_name(struct net_device
*dev
, const char *newname
)
1004 char oldname
[IFNAMSIZ
];
1010 BUG_ON(!dev_net(dev
));
1013 if (dev
->flags
& IFF_UP
)
1016 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0)
1019 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
1021 err
= dev_get_valid_name(net
, dev
, newname
);
1026 ret
= device_rename(&dev
->dev
, dev
->name
);
1028 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1032 write_lock_bh(&dev_base_lock
);
1033 hlist_del_rcu(&dev
->name_hlist
);
1034 write_unlock_bh(&dev_base_lock
);
1038 write_lock_bh(&dev_base_lock
);
1039 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
1040 write_unlock_bh(&dev_base_lock
);
1042 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
1043 ret
= notifier_to_errno(ret
);
1046 /* err >= 0 after dev_alloc_name() or stores the first errno */
1049 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1052 pr_err("%s: name change rollback failed: %d\n",
1061 * dev_set_alias - change ifalias of a device
1063 * @alias: name up to IFALIASZ
1064 * @len: limit of bytes to copy from info
1066 * Set ifalias for a device,
1068 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
1074 if (len
>= IFALIASZ
)
1079 kfree(dev
->ifalias
);
1080 dev
->ifalias
= NULL
;
1085 new_ifalias
= krealloc(dev
->ifalias
, len
+ 1, GFP_KERNEL
);
1088 dev
->ifalias
= new_ifalias
;
1090 strlcpy(dev
->ifalias
, alias
, len
+1);
1096 * netdev_features_change - device changes features
1097 * @dev: device to cause notification
1099 * Called to indicate a device has changed features.
1101 void netdev_features_change(struct net_device
*dev
)
1103 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1105 EXPORT_SYMBOL(netdev_features_change
);
1108 * netdev_state_change - device changes state
1109 * @dev: device to cause notification
1111 * Called to indicate a device has changed state. This function calls
1112 * the notifier chains for netdev_chain and sends a NEWLINK message
1113 * to the routing socket.
1115 void netdev_state_change(struct net_device
*dev
)
1117 if (dev
->flags
& IFF_UP
) {
1118 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
1119 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
1122 EXPORT_SYMBOL(netdev_state_change
);
1125 * netdev_notify_peers - notify network peers about existence of @dev
1126 * @dev: network device
1128 * Generate traffic such that interested network peers are aware of
1129 * @dev, such as by generating a gratuitous ARP. This may be used when
1130 * a device wants to inform the rest of the network about some sort of
1131 * reconfiguration such as a failover event or virtual machine
1134 void netdev_notify_peers(struct net_device
*dev
)
1137 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
, dev
);
1140 EXPORT_SYMBOL(netdev_notify_peers
);
1143 * dev_load - load a network module
1144 * @net: the applicable net namespace
1145 * @name: name of interface
1147 * If a network interface is not present and the process has suitable
1148 * privileges this function loads the module. If module loading is not
1149 * available in this kernel then it becomes a nop.
1152 void dev_load(struct net
*net
, const char *name
)
1154 struct net_device
*dev
;
1158 dev
= dev_get_by_name_rcu(net
, name
);
1162 if (no_module
&& capable(CAP_NET_ADMIN
))
1163 no_module
= request_module("netdev-%s", name
);
1164 if (no_module
&& capable(CAP_SYS_MODULE
)) {
1165 if (!request_module("%s", name
))
1166 pr_warn("Loading kernel module for a network device with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s instead.\n",
1170 EXPORT_SYMBOL(dev_load
);
1172 static int __dev_open(struct net_device
*dev
)
1174 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1179 if (!netif_device_present(dev
))
1182 ret
= call_netdevice_notifiers(NETDEV_PRE_UP
, dev
);
1183 ret
= notifier_to_errno(ret
);
1187 set_bit(__LINK_STATE_START
, &dev
->state
);
1189 if (ops
->ndo_validate_addr
)
1190 ret
= ops
->ndo_validate_addr(dev
);
1192 if (!ret
&& ops
->ndo_open
)
1193 ret
= ops
->ndo_open(dev
);
1196 clear_bit(__LINK_STATE_START
, &dev
->state
);
1198 dev
->flags
|= IFF_UP
;
1199 net_dmaengine_get();
1200 dev_set_rx_mode(dev
);
1202 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
1209 * dev_open - prepare an interface for use.
1210 * @dev: device to open
1212 * Takes a device from down to up state. The device's private open
1213 * function is invoked and then the multicast lists are loaded. Finally
1214 * the device is moved into the up state and a %NETDEV_UP message is
1215 * sent to the netdev notifier chain.
1217 * Calling this function on an active interface is a nop. On a failure
1218 * a negative errno code is returned.
1220 int dev_open(struct net_device
*dev
)
1224 if (dev
->flags
& IFF_UP
)
1227 ret
= __dev_open(dev
);
1231 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1232 call_netdevice_notifiers(NETDEV_UP
, dev
);
1236 EXPORT_SYMBOL(dev_open
);
1238 static int __dev_close_many(struct list_head
*head
)
1240 struct net_device
*dev
;
1245 list_for_each_entry(dev
, head
, unreg_list
) {
1246 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1248 clear_bit(__LINK_STATE_START
, &dev
->state
);
1250 /* Synchronize to scheduled poll. We cannot touch poll list, it
1251 * can be even on different cpu. So just clear netif_running().
1253 * dev->stop() will invoke napi_disable() on all of it's
1254 * napi_struct instances on this device.
1256 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1259 dev_deactivate_many(head
);
1261 list_for_each_entry(dev
, head
, unreg_list
) {
1262 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1265 * Call the device specific close. This cannot fail.
1266 * Only if device is UP
1268 * We allow it to be called even after a DETACH hot-plug
1274 dev
->flags
&= ~IFF_UP
;
1275 net_dmaengine_put();
1281 static int __dev_close(struct net_device
*dev
)
1286 list_add(&dev
->unreg_list
, &single
);
1287 retval
= __dev_close_many(&single
);
1292 static int dev_close_many(struct list_head
*head
)
1294 struct net_device
*dev
, *tmp
;
1295 LIST_HEAD(tmp_list
);
1297 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
)
1298 if (!(dev
->flags
& IFF_UP
))
1299 list_move(&dev
->unreg_list
, &tmp_list
);
1301 __dev_close_many(head
);
1303 list_for_each_entry(dev
, head
, unreg_list
) {
1304 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1305 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1308 /* rollback_registered_many needs the complete original list */
1309 list_splice(&tmp_list
, head
);
1314 * dev_close - shutdown an interface.
1315 * @dev: device to shutdown
1317 * This function moves an active device into down state. A
1318 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1319 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1322 int dev_close(struct net_device
*dev
)
1324 if (dev
->flags
& IFF_UP
) {
1327 list_add(&dev
->unreg_list
, &single
);
1328 dev_close_many(&single
);
1333 EXPORT_SYMBOL(dev_close
);
1337 * dev_disable_lro - disable Large Receive Offload on a device
1340 * Disable Large Receive Offload (LRO) on a net device. Must be
1341 * called under RTNL. This is needed if received packets may be
1342 * forwarded to another interface.
1344 void dev_disable_lro(struct net_device
*dev
)
1347 * If we're trying to disable lro on a vlan device
1348 * use the underlying physical device instead
1350 if (is_vlan_dev(dev
))
1351 dev
= vlan_dev_real_dev(dev
);
1353 dev
->wanted_features
&= ~NETIF_F_LRO
;
1354 netdev_update_features(dev
);
1356 if (unlikely(dev
->features
& NETIF_F_LRO
))
1357 netdev_WARN(dev
, "failed to disable LRO!\n");
1359 EXPORT_SYMBOL(dev_disable_lro
);
1362 static int dev_boot_phase
= 1;
1365 * register_netdevice_notifier - register a network notifier block
1368 * Register a notifier to be called when network device events occur.
1369 * The notifier passed is linked into the kernel structures and must
1370 * not be reused until it has been unregistered. A negative errno code
1371 * is returned on a failure.
1373 * When registered all registration and up events are replayed
1374 * to the new notifier to allow device to have a race free
1375 * view of the network device list.
1378 int register_netdevice_notifier(struct notifier_block
*nb
)
1380 struct net_device
*dev
;
1381 struct net_device
*last
;
1386 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1392 for_each_netdev(net
, dev
) {
1393 err
= nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1394 err
= notifier_to_errno(err
);
1398 if (!(dev
->flags
& IFF_UP
))
1401 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1412 for_each_netdev(net
, dev
) {
1416 if (dev
->flags
& IFF_UP
) {
1417 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1418 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1420 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1425 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1428 EXPORT_SYMBOL(register_netdevice_notifier
);
1431 * unregister_netdevice_notifier - unregister a network notifier block
1434 * Unregister a notifier previously registered by
1435 * register_netdevice_notifier(). The notifier is unlinked into the
1436 * kernel structures and may then be reused. A negative errno code
1437 * is returned on a failure.
1439 * After unregistering unregister and down device events are synthesized
1440 * for all devices on the device list to the removed notifier to remove
1441 * the need for special case cleanup code.
1444 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1446 struct net_device
*dev
;
1451 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1456 for_each_netdev(net
, dev
) {
1457 if (dev
->flags
& IFF_UP
) {
1458 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1459 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1461 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1468 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1471 * call_netdevice_notifiers - call all network notifier blocks
1472 * @val: value passed unmodified to notifier function
1473 * @dev: net_device pointer passed unmodified to notifier function
1475 * Call all network notifier blocks. Parameters and return value
1476 * are as for raw_notifier_call_chain().
1479 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1482 return raw_notifier_call_chain(&netdev_chain
, val
, dev
);
1484 EXPORT_SYMBOL(call_netdevice_notifiers
);
1486 static struct static_key netstamp_needed __read_mostly
;
1487 #ifdef HAVE_JUMP_LABEL
1488 /* We are not allowed to call static_key_slow_dec() from irq context
1489 * If net_disable_timestamp() is called from irq context, defer the
1490 * static_key_slow_dec() calls.
1492 static atomic_t netstamp_needed_deferred
;
1495 void net_enable_timestamp(void)
1497 #ifdef HAVE_JUMP_LABEL
1498 int deferred
= atomic_xchg(&netstamp_needed_deferred
, 0);
1502 static_key_slow_dec(&netstamp_needed
);
1506 WARN_ON(in_interrupt());
1507 static_key_slow_inc(&netstamp_needed
);
1509 EXPORT_SYMBOL(net_enable_timestamp
);
1511 void net_disable_timestamp(void)
1513 #ifdef HAVE_JUMP_LABEL
1514 if (in_interrupt()) {
1515 atomic_inc(&netstamp_needed_deferred
);
1519 static_key_slow_dec(&netstamp_needed
);
1521 EXPORT_SYMBOL(net_disable_timestamp
);
1523 static inline void net_timestamp_set(struct sk_buff
*skb
)
1525 skb
->tstamp
.tv64
= 0;
1526 if (static_key_false(&netstamp_needed
))
1527 __net_timestamp(skb
);
1530 #define net_timestamp_check(COND, SKB) \
1531 if (static_key_false(&netstamp_needed)) { \
1532 if ((COND) && !(SKB)->tstamp.tv64) \
1533 __net_timestamp(SKB); \
1536 static int net_hwtstamp_validate(struct ifreq *ifr)
1538 struct hwtstamp_config cfg
;
1539 enum hwtstamp_tx_types tx_type
;
1540 enum hwtstamp_rx_filters rx_filter
;
1541 int tx_type_valid
= 0;
1542 int rx_filter_valid
= 0;
1544 if (copy_from_user(&cfg
, ifr
->ifr_data
, sizeof(cfg
)))
1547 if (cfg
.flags
) /* reserved for future extensions */
1550 tx_type
= cfg
.tx_type
;
1551 rx_filter
= cfg
.rx_filter
;
1554 case HWTSTAMP_TX_OFF
:
1555 case HWTSTAMP_TX_ON
:
1556 case HWTSTAMP_TX_ONESTEP_SYNC
:
1561 switch (rx_filter
) {
1562 case HWTSTAMP_FILTER_NONE
:
1563 case HWTSTAMP_FILTER_ALL
:
1564 case HWTSTAMP_FILTER_SOME
:
1565 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT
:
1566 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC
:
1567 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ
:
1568 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT
:
1569 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC
:
1570 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ
:
1571 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT
:
1572 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC
:
1573 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ
:
1574 case HWTSTAMP_FILTER_PTP_V2_EVENT
:
1575 case HWTSTAMP_FILTER_PTP_V2_SYNC
:
1576 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ
:
1577 rx_filter_valid
= 1;
1581 if (!tx_type_valid
|| !rx_filter_valid
)
1587 static inline bool is_skb_forwardable(struct net_device
*dev
,
1588 struct sk_buff
*skb
)
1592 if (!(dev
->flags
& IFF_UP
))
1595 len
= dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
;
1596 if (skb
->len
<= len
)
1599 /* if TSO is enabled, we don't care about the length as the packet
1600 * could be forwarded without being segmented before
1602 if (skb_is_gso(skb
))
1609 * dev_forward_skb - loopback an skb to another netif
1611 * @dev: destination network device
1612 * @skb: buffer to forward
1615 * NET_RX_SUCCESS (no congestion)
1616 * NET_RX_DROP (packet was dropped, but freed)
1618 * dev_forward_skb can be used for injecting an skb from the
1619 * start_xmit function of one device into the receive queue
1620 * of another device.
1622 * The receiving device may be in another namespace, so
1623 * we have to clear all information in the skb that could
1624 * impact namespace isolation.
1626 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1628 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
1629 if (skb_copy_ubufs(skb
, GFP_ATOMIC
)) {
1630 atomic_long_inc(&dev
->rx_dropped
);
1639 if (unlikely(!is_skb_forwardable(dev
, skb
))) {
1640 atomic_long_inc(&dev
->rx_dropped
);
1647 skb
->tstamp
.tv64
= 0;
1648 skb
->pkt_type
= PACKET_HOST
;
1649 skb
->protocol
= eth_type_trans(skb
, dev
);
1653 return netif_rx(skb
);
1655 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1657 static inline int deliver_skb(struct sk_buff
*skb
,
1658 struct packet_type
*pt_prev
,
1659 struct net_device
*orig_dev
)
1661 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
1663 atomic_inc(&skb
->users
);
1664 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1667 static inline bool skb_loop_sk(struct packet_type
*ptype
, struct sk_buff
*skb
)
1669 if (!ptype
->af_packet_priv
|| !skb
->sk
)
1672 if (ptype
->id_match
)
1673 return ptype
->id_match(ptype
, skb
->sk
);
1674 else if ((struct sock
*)ptype
->af_packet_priv
== skb
->sk
)
1681 * Support routine. Sends outgoing frames to any network
1682 * taps currently in use.
1685 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1687 struct packet_type
*ptype
;
1688 struct sk_buff
*skb2
= NULL
;
1689 struct packet_type
*pt_prev
= NULL
;
1692 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1693 /* Never send packets back to the socket
1694 * they originated from - MvS (miquels@drinkel.ow.org)
1696 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1697 (!skb_loop_sk(ptype
, skb
))) {
1699 deliver_skb(skb2
, pt_prev
, skb
->dev
);
1704 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1708 net_timestamp_set(skb2
);
1710 /* skb->nh should be correctly
1711 set by sender, so that the second statement is
1712 just protection against buggy protocols.
1714 skb_reset_mac_header(skb2
);
1716 if (skb_network_header(skb2
) < skb2
->data
||
1717 skb2
->network_header
> skb2
->tail
) {
1718 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1719 ntohs(skb2
->protocol
),
1721 skb_reset_network_header(skb2
);
1724 skb2
->transport_header
= skb2
->network_header
;
1725 skb2
->pkt_type
= PACKET_OUTGOING
;
1730 pt_prev
->func(skb2
, skb
->dev
, pt_prev
, skb
->dev
);
1735 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1736 * @dev: Network device
1737 * @txq: number of queues available
1739 * If real_num_tx_queues is changed the tc mappings may no longer be
1740 * valid. To resolve this verify the tc mapping remains valid and if
1741 * not NULL the mapping. With no priorities mapping to this
1742 * offset/count pair it will no longer be used. In the worst case TC0
1743 * is invalid nothing can be done so disable priority mappings. If is
1744 * expected that drivers will fix this mapping if they can before
1745 * calling netif_set_real_num_tx_queues.
1747 static void netif_setup_tc(struct net_device
*dev
, unsigned int txq
)
1750 struct netdev_tc_txq
*tc
= &dev
->tc_to_txq
[0];
1752 /* If TC0 is invalidated disable TC mapping */
1753 if (tc
->offset
+ tc
->count
> txq
) {
1754 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1759 /* Invalidated prio to tc mappings set to TC0 */
1760 for (i
= 1; i
< TC_BITMASK
+ 1; i
++) {
1761 int q
= netdev_get_prio_tc_map(dev
, i
);
1763 tc
= &dev
->tc_to_txq
[q
];
1764 if (tc
->offset
+ tc
->count
> txq
) {
1765 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1767 netdev_set_prio_tc_map(dev
, i
, 0);
1773 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1774 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1776 int netif_set_real_num_tx_queues(struct net_device
*dev
, unsigned int txq
)
1780 if (txq
< 1 || txq
> dev
->num_tx_queues
)
1783 if (dev
->reg_state
== NETREG_REGISTERED
||
1784 dev
->reg_state
== NETREG_UNREGISTERING
) {
1787 rc
= netdev_queue_update_kobjects(dev
, dev
->real_num_tx_queues
,
1793 netif_setup_tc(dev
, txq
);
1795 if (txq
< dev
->real_num_tx_queues
)
1796 qdisc_reset_all_tx_gt(dev
, txq
);
1799 dev
->real_num_tx_queues
= txq
;
1802 EXPORT_SYMBOL(netif_set_real_num_tx_queues
);
1806 * netif_set_real_num_rx_queues - set actual number of RX queues used
1807 * @dev: Network device
1808 * @rxq: Actual number of RX queues
1810 * This must be called either with the rtnl_lock held or before
1811 * registration of the net device. Returns 0 on success, or a
1812 * negative error code. If called before registration, it always
1815 int netif_set_real_num_rx_queues(struct net_device
*dev
, unsigned int rxq
)
1819 if (rxq
< 1 || rxq
> dev
->num_rx_queues
)
1822 if (dev
->reg_state
== NETREG_REGISTERED
) {
1825 rc
= net_rx_queue_update_kobjects(dev
, dev
->real_num_rx_queues
,
1831 dev
->real_num_rx_queues
= rxq
;
1834 EXPORT_SYMBOL(netif_set_real_num_rx_queues
);
1838 * netif_get_num_default_rss_queues - default number of RSS queues
1840 * This routine should set an upper limit on the number of RSS queues
1841 * used by default by multiqueue devices.
1843 int netif_get_num_default_rss_queues(void)
1845 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES
, num_online_cpus());
1847 EXPORT_SYMBOL(netif_get_num_default_rss_queues
);
1849 static inline void __netif_reschedule(struct Qdisc
*q
)
1851 struct softnet_data
*sd
;
1852 unsigned long flags
;
1854 local_irq_save(flags
);
1855 sd
= &__get_cpu_var(softnet_data
);
1856 q
->next_sched
= NULL
;
1857 *sd
->output_queue_tailp
= q
;
1858 sd
->output_queue_tailp
= &q
->next_sched
;
1859 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1860 local_irq_restore(flags
);
1863 void __netif_schedule(struct Qdisc
*q
)
1865 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
1866 __netif_reschedule(q
);
1868 EXPORT_SYMBOL(__netif_schedule
);
1870 void dev_kfree_skb_irq(struct sk_buff
*skb
)
1872 if (atomic_dec_and_test(&skb
->users
)) {
1873 struct softnet_data
*sd
;
1874 unsigned long flags
;
1876 local_irq_save(flags
);
1877 sd
= &__get_cpu_var(softnet_data
);
1878 skb
->next
= sd
->completion_queue
;
1879 sd
->completion_queue
= skb
;
1880 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1881 local_irq_restore(flags
);
1884 EXPORT_SYMBOL(dev_kfree_skb_irq
);
1886 void dev_kfree_skb_any(struct sk_buff
*skb
)
1888 if (in_irq() || irqs_disabled())
1889 dev_kfree_skb_irq(skb
);
1893 EXPORT_SYMBOL(dev_kfree_skb_any
);
1897 * netif_device_detach - mark device as removed
1898 * @dev: network device
1900 * Mark device as removed from system and therefore no longer available.
1902 void netif_device_detach(struct net_device
*dev
)
1904 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1905 netif_running(dev
)) {
1906 netif_tx_stop_all_queues(dev
);
1909 EXPORT_SYMBOL(netif_device_detach
);
1912 * netif_device_attach - mark device as attached
1913 * @dev: network device
1915 * Mark device as attached from system and restart if needed.
1917 void netif_device_attach(struct net_device
*dev
)
1919 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1920 netif_running(dev
)) {
1921 netif_tx_wake_all_queues(dev
);
1922 __netdev_watchdog_up(dev
);
1925 EXPORT_SYMBOL(netif_device_attach
);
1927 static void skb_warn_bad_offload(const struct sk_buff
*skb
)
1929 static const netdev_features_t null_features
= 0;
1930 struct net_device
*dev
= skb
->dev
;
1931 const char *driver
= "";
1933 if (dev
&& dev
->dev
.parent
)
1934 driver
= dev_driver_string(dev
->dev
.parent
);
1936 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
1937 "gso_type=%d ip_summed=%d\n",
1938 driver
, dev
? &dev
->features
: &null_features
,
1939 skb
->sk
? &skb
->sk
->sk_route_caps
: &null_features
,
1940 skb
->len
, skb
->data_len
, skb_shinfo(skb
)->gso_size
,
1941 skb_shinfo(skb
)->gso_type
, skb
->ip_summed
);
1945 * Invalidate hardware checksum when packet is to be mangled, and
1946 * complete checksum manually on outgoing path.
1948 int skb_checksum_help(struct sk_buff
*skb
)
1951 int ret
= 0, offset
;
1953 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1954 goto out_set_summed
;
1956 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1957 skb_warn_bad_offload(skb
);
1961 offset
= skb_checksum_start_offset(skb
);
1962 BUG_ON(offset
>= skb_headlen(skb
));
1963 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1965 offset
+= skb
->csum_offset
;
1966 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
1968 if (skb_cloned(skb
) &&
1969 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
1970 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1975 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1977 skb
->ip_summed
= CHECKSUM_NONE
;
1981 EXPORT_SYMBOL(skb_checksum_help
);
1984 * skb_gso_segment - Perform segmentation on skb.
1985 * @skb: buffer to segment
1986 * @features: features for the output path (see dev->features)
1988 * This function segments the given skb and returns a list of segments.
1990 * It may return NULL if the skb requires no segmentation. This is
1991 * only possible when GSO is used for verifying header integrity.
1993 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
,
1994 netdev_features_t features
)
1996 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1997 struct packet_type
*ptype
;
1998 __be16 type
= skb
->protocol
;
1999 int vlan_depth
= ETH_HLEN
;
2002 while (type
== htons(ETH_P_8021Q
)) {
2003 struct vlan_hdr
*vh
;
2005 if (unlikely(!pskb_may_pull(skb
, vlan_depth
+ VLAN_HLEN
)))
2006 return ERR_PTR(-EINVAL
);
2008 vh
= (struct vlan_hdr
*)(skb
->data
+ vlan_depth
);
2009 type
= vh
->h_vlan_encapsulated_proto
;
2010 vlan_depth
+= VLAN_HLEN
;
2013 skb_reset_mac_header(skb
);
2014 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
2015 __skb_pull(skb
, skb
->mac_len
);
2017 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
2018 skb_warn_bad_offload(skb
);
2020 if (skb_header_cloned(skb
) &&
2021 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
2022 return ERR_PTR(err
);
2026 list_for_each_entry_rcu(ptype
,
2027 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
2028 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
2029 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
2030 err
= ptype
->gso_send_check(skb
);
2031 segs
= ERR_PTR(err
);
2032 if (err
|| skb_gso_ok(skb
, features
))
2034 __skb_push(skb
, (skb
->data
-
2035 skb_network_header(skb
)));
2037 segs
= ptype
->gso_segment(skb
, features
);
2043 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
2047 EXPORT_SYMBOL(skb_gso_segment
);
2049 /* Take action when hardware reception checksum errors are detected. */
2051 void netdev_rx_csum_fault(struct net_device
*dev
)
2053 if (net_ratelimit()) {
2054 pr_err("%s: hw csum failure\n", dev
? dev
->name
: "<unknown>");
2058 EXPORT_SYMBOL(netdev_rx_csum_fault
);
2061 /* Actually, we should eliminate this check as soon as we know, that:
2062 * 1. IOMMU is present and allows to map all the memory.
2063 * 2. No high memory really exists on this machine.
2066 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
2068 #ifdef CONFIG_HIGHMEM
2070 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
2071 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2072 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2073 if (PageHighMem(skb_frag_page(frag
)))
2078 if (PCI_DMA_BUS_IS_PHYS
) {
2079 struct device
*pdev
= dev
->dev
.parent
;
2083 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2084 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2085 dma_addr_t addr
= page_to_phys(skb_frag_page(frag
));
2086 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
2095 void (*destructor
)(struct sk_buff
*skb
);
2098 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2100 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
2102 struct dev_gso_cb
*cb
;
2105 struct sk_buff
*nskb
= skb
->next
;
2107 skb
->next
= nskb
->next
;
2110 } while (skb
->next
);
2112 cb
= DEV_GSO_CB(skb
);
2114 cb
->destructor(skb
);
2118 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2119 * @skb: buffer to segment
2120 * @features: device features as applicable to this skb
2122 * This function segments the given skb and stores the list of segments
2125 static int dev_gso_segment(struct sk_buff
*skb
, netdev_features_t features
)
2127 struct sk_buff
*segs
;
2129 segs
= skb_gso_segment(skb
, features
);
2131 /* Verifying header integrity only. */
2136 return PTR_ERR(segs
);
2139 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
2140 skb
->destructor
= dev_gso_skb_destructor
;
2145 static bool can_checksum_protocol(netdev_features_t features
, __be16 protocol
)
2147 return ((features
& NETIF_F_GEN_CSUM
) ||
2148 ((features
& NETIF_F_V4_CSUM
) &&
2149 protocol
== htons(ETH_P_IP
)) ||
2150 ((features
& NETIF_F_V6_CSUM
) &&
2151 protocol
== htons(ETH_P_IPV6
)) ||
2152 ((features
& NETIF_F_FCOE_CRC
) &&
2153 protocol
== htons(ETH_P_FCOE
)));
2156 static netdev_features_t
harmonize_features(struct sk_buff
*skb
,
2157 __be16 protocol
, netdev_features_t features
)
2159 if (skb
->ip_summed
!= CHECKSUM_NONE
&&
2160 !can_checksum_protocol(features
, protocol
)) {
2161 features
&= ~NETIF_F_ALL_CSUM
;
2162 features
&= ~NETIF_F_SG
;
2163 } else if (illegal_highdma(skb
->dev
, skb
)) {
2164 features
&= ~NETIF_F_SG
;
2170 netdev_features_t
netif_skb_features(struct sk_buff
*skb
)
2172 __be16 protocol
= skb
->protocol
;
2173 netdev_features_t features
= skb
->dev
->features
;
2175 if (skb_shinfo(skb
)->gso_segs
> skb
->dev
->gso_max_segs
)
2176 features
&= ~NETIF_F_GSO_MASK
;
2178 if (protocol
== htons(ETH_P_8021Q
)) {
2179 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
2180 protocol
= veh
->h_vlan_encapsulated_proto
;
2181 } else if (!vlan_tx_tag_present(skb
)) {
2182 return harmonize_features(skb
, protocol
, features
);
2185 features
&= (skb
->dev
->vlan_features
| NETIF_F_HW_VLAN_TX
);
2187 if (protocol
!= htons(ETH_P_8021Q
)) {
2188 return harmonize_features(skb
, protocol
, features
);
2190 features
&= NETIF_F_SG
| NETIF_F_HIGHDMA
| NETIF_F_FRAGLIST
|
2191 NETIF_F_GEN_CSUM
| NETIF_F_HW_VLAN_TX
;
2192 return harmonize_features(skb
, protocol
, features
);
2195 EXPORT_SYMBOL(netif_skb_features
);
2198 * Returns true if either:
2199 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2200 * 2. skb is fragmented and the device does not support SG.
2202 static inline int skb_needs_linearize(struct sk_buff
*skb
,
2205 return skb_is_nonlinear(skb
) &&
2206 ((skb_has_frag_list(skb
) &&
2207 !(features
& NETIF_F_FRAGLIST
)) ||
2208 (skb_shinfo(skb
)->nr_frags
&&
2209 !(features
& NETIF_F_SG
)));
2212 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
2213 struct netdev_queue
*txq
)
2215 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2216 int rc
= NETDEV_TX_OK
;
2217 unsigned int skb_len
;
2219 if (likely(!skb
->next
)) {
2220 netdev_features_t features
;
2223 * If device doesn't need skb->dst, release it right now while
2224 * its hot in this cpu cache
2226 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2229 features
= netif_skb_features(skb
);
2231 if (vlan_tx_tag_present(skb
) &&
2232 !(features
& NETIF_F_HW_VLAN_TX
)) {
2233 skb
= __vlan_put_tag(skb
, vlan_tx_tag_get(skb
));
2240 if (netif_needs_gso(skb
, features
)) {
2241 if (unlikely(dev_gso_segment(skb
, features
)))
2246 if (skb_needs_linearize(skb
, features
) &&
2247 __skb_linearize(skb
))
2250 /* If packet is not checksummed and device does not
2251 * support checksumming for this protocol, complete
2252 * checksumming here.
2254 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2255 skb_set_transport_header(skb
,
2256 skb_checksum_start_offset(skb
));
2257 if (!(features
& NETIF_F_ALL_CSUM
) &&
2258 skb_checksum_help(skb
))
2263 if (!list_empty(&ptype_all
))
2264 dev_queue_xmit_nit(skb
, dev
);
2267 rc
= ops
->ndo_start_xmit(skb
, dev
);
2268 trace_net_dev_xmit(skb
, rc
, dev
, skb_len
);
2269 if (rc
== NETDEV_TX_OK
)
2270 txq_trans_update(txq
);
2276 struct sk_buff
*nskb
= skb
->next
;
2278 skb
->next
= nskb
->next
;
2282 * If device doesn't need nskb->dst, release it right now while
2283 * its hot in this cpu cache
2285 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2288 if (!list_empty(&ptype_all
))
2289 dev_queue_xmit_nit(nskb
, dev
);
2291 skb_len
= nskb
->len
;
2292 rc
= ops
->ndo_start_xmit(nskb
, dev
);
2293 trace_net_dev_xmit(nskb
, rc
, dev
, skb_len
);
2294 if (unlikely(rc
!= NETDEV_TX_OK
)) {
2295 if (rc
& ~NETDEV_TX_MASK
)
2296 goto out_kfree_gso_skb
;
2297 nskb
->next
= skb
->next
;
2301 txq_trans_update(txq
);
2302 if (unlikely(netif_xmit_stopped(txq
) && skb
->next
))
2303 return NETDEV_TX_BUSY
;
2304 } while (skb
->next
);
2307 if (likely(skb
->next
== NULL
))
2308 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
2315 static u32 hashrnd __read_mostly
;
2318 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2319 * to be used as a distribution range.
2321 u16
__skb_tx_hash(const struct net_device
*dev
, const struct sk_buff
*skb
,
2322 unsigned int num_tx_queues
)
2326 u16 qcount
= num_tx_queues
;
2328 if (skb_rx_queue_recorded(skb
)) {
2329 hash
= skb_get_rx_queue(skb
);
2330 while (unlikely(hash
>= num_tx_queues
))
2331 hash
-= num_tx_queues
;
2336 u8 tc
= netdev_get_prio_tc_map(dev
, skb
->priority
);
2337 qoffset
= dev
->tc_to_txq
[tc
].offset
;
2338 qcount
= dev
->tc_to_txq
[tc
].count
;
2341 if (skb
->sk
&& skb
->sk
->sk_hash
)
2342 hash
= skb
->sk
->sk_hash
;
2344 hash
= (__force u16
) skb
->protocol
;
2345 hash
= jhash_1word(hash
, hashrnd
);
2347 return (u16
) (((u64
) hash
* qcount
) >> 32) + qoffset
;
2349 EXPORT_SYMBOL(__skb_tx_hash
);
2351 static inline u16
dev_cap_txqueue(struct net_device
*dev
, u16 queue_index
)
2353 if (unlikely(queue_index
>= dev
->real_num_tx_queues
)) {
2354 net_warn_ratelimited("%s selects TX queue %d, but real number of TX queues is %d\n",
2355 dev
->name
, queue_index
,
2356 dev
->real_num_tx_queues
);
2362 static inline int get_xps_queue(struct net_device
*dev
, struct sk_buff
*skb
)
2365 struct xps_dev_maps
*dev_maps
;
2366 struct xps_map
*map
;
2367 int queue_index
= -1;
2370 dev_maps
= rcu_dereference(dev
->xps_maps
);
2372 map
= rcu_dereference(
2373 dev_maps
->cpu_map
[raw_smp_processor_id()]);
2376 queue_index
= map
->queues
[0];
2379 if (skb
->sk
&& skb
->sk
->sk_hash
)
2380 hash
= skb
->sk
->sk_hash
;
2382 hash
= (__force u16
) skb
->protocol
^
2384 hash
= jhash_1word(hash
, hashrnd
);
2385 queue_index
= map
->queues
[
2386 ((u64
)hash
* map
->len
) >> 32];
2388 if (unlikely(queue_index
>= dev
->real_num_tx_queues
))
2400 struct netdev_queue
*netdev_pick_tx(struct net_device
*dev
,
2401 struct sk_buff
*skb
)
2404 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2406 if (dev
->real_num_tx_queues
== 1)
2408 else if (ops
->ndo_select_queue
) {
2409 queue_index
= ops
->ndo_select_queue(dev
, skb
);
2410 queue_index
= dev_cap_txqueue(dev
, queue_index
);
2412 struct sock
*sk
= skb
->sk
;
2413 queue_index
= sk_tx_queue_get(sk
);
2415 if (queue_index
< 0 || skb
->ooo_okay
||
2416 queue_index
>= dev
->real_num_tx_queues
) {
2417 int old_index
= queue_index
;
2419 queue_index
= get_xps_queue(dev
, skb
);
2420 if (queue_index
< 0)
2421 queue_index
= skb_tx_hash(dev
, skb
);
2423 if (queue_index
!= old_index
&& sk
) {
2424 struct dst_entry
*dst
=
2425 rcu_dereference_check(sk
->sk_dst_cache
, 1);
2427 if (dst
&& skb_dst(skb
) == dst
)
2428 sk_tx_queue_set(sk
, queue_index
);
2433 skb_set_queue_mapping(skb
, queue_index
);
2434 return netdev_get_tx_queue(dev
, queue_index
);
2437 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2438 struct net_device
*dev
,
2439 struct netdev_queue
*txq
)
2441 spinlock_t
*root_lock
= qdisc_lock(q
);
2445 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
2446 qdisc_calculate_pkt_len(skb
, q
);
2448 * Heuristic to force contended enqueues to serialize on a
2449 * separate lock before trying to get qdisc main lock.
2450 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2451 * and dequeue packets faster.
2453 contended
= qdisc_is_running(q
);
2454 if (unlikely(contended
))
2455 spin_lock(&q
->busylock
);
2457 spin_lock(root_lock
);
2458 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2461 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2462 qdisc_run_begin(q
)) {
2464 * This is a work-conserving queue; there are no old skbs
2465 * waiting to be sent out; and the qdisc is not running -
2466 * xmit the skb directly.
2468 if (!(dev
->priv_flags
& IFF_XMIT_DST_RELEASE
))
2471 qdisc_bstats_update(q
, skb
);
2473 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
)) {
2474 if (unlikely(contended
)) {
2475 spin_unlock(&q
->busylock
);
2482 rc
= NET_XMIT_SUCCESS
;
2485 rc
= q
->enqueue(skb
, q
) & NET_XMIT_MASK
;
2486 if (qdisc_run_begin(q
)) {
2487 if (unlikely(contended
)) {
2488 spin_unlock(&q
->busylock
);
2494 spin_unlock(root_lock
);
2495 if (unlikely(contended
))
2496 spin_unlock(&q
->busylock
);
2500 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2501 static void skb_update_prio(struct sk_buff
*skb
)
2503 struct netprio_map
*map
= rcu_dereference_bh(skb
->dev
->priomap
);
2505 if (!skb
->priority
&& skb
->sk
&& map
) {
2506 unsigned int prioidx
= skb
->sk
->sk_cgrp_prioidx
;
2508 if (prioidx
< map
->priomap_len
)
2509 skb
->priority
= map
->priomap
[prioidx
];
2513 #define skb_update_prio(skb)
2516 static DEFINE_PER_CPU(int, xmit_recursion
);
2517 #define RECURSION_LIMIT 10
2520 * dev_loopback_xmit - loop back @skb
2521 * @skb: buffer to transmit
2523 int dev_loopback_xmit(struct sk_buff
*skb
)
2525 skb_reset_mac_header(skb
);
2526 __skb_pull(skb
, skb_network_offset(skb
));
2527 skb
->pkt_type
= PACKET_LOOPBACK
;
2528 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
2529 WARN_ON(!skb_dst(skb
));
2534 EXPORT_SYMBOL(dev_loopback_xmit
);
2537 * dev_queue_xmit - transmit a buffer
2538 * @skb: buffer to transmit
2540 * Queue a buffer for transmission to a network device. The caller must
2541 * have set the device and priority and built the buffer before calling
2542 * this function. The function can be called from an interrupt.
2544 * A negative errno code is returned on a failure. A success does not
2545 * guarantee the frame will be transmitted as it may be dropped due
2546 * to congestion or traffic shaping.
2548 * -----------------------------------------------------------------------------------
2549 * I notice this method can also return errors from the queue disciplines,
2550 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2553 * Regardless of the return value, the skb is consumed, so it is currently
2554 * difficult to retry a send to this method. (You can bump the ref count
2555 * before sending to hold a reference for retry if you are careful.)
2557 * When calling this method, interrupts MUST be enabled. This is because
2558 * the BH enable code must have IRQs enabled so that it will not deadlock.
2561 int dev_queue_xmit(struct sk_buff
*skb
)
2563 struct net_device
*dev
= skb
->dev
;
2564 struct netdev_queue
*txq
;
2568 /* Disable soft irqs for various locks below. Also
2569 * stops preemption for RCU.
2573 skb_update_prio(skb
);
2575 txq
= netdev_pick_tx(dev
, skb
);
2576 q
= rcu_dereference_bh(txq
->qdisc
);
2578 #ifdef CONFIG_NET_CLS_ACT
2579 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2581 trace_net_dev_queue(skb
);
2583 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2587 /* The device has no queue. Common case for software devices:
2588 loopback, all the sorts of tunnels...
2590 Really, it is unlikely that netif_tx_lock protection is necessary
2591 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2593 However, it is possible, that they rely on protection
2596 Check this and shot the lock. It is not prone from deadlocks.
2597 Either shot noqueue qdisc, it is even simpler 8)
2599 if (dev
->flags
& IFF_UP
) {
2600 int cpu
= smp_processor_id(); /* ok because BHs are off */
2602 if (txq
->xmit_lock_owner
!= cpu
) {
2604 if (__this_cpu_read(xmit_recursion
) > RECURSION_LIMIT
)
2605 goto recursion_alert
;
2607 HARD_TX_LOCK(dev
, txq
, cpu
);
2609 if (!netif_xmit_stopped(txq
)) {
2610 __this_cpu_inc(xmit_recursion
);
2611 rc
= dev_hard_start_xmit(skb
, dev
, txq
);
2612 __this_cpu_dec(xmit_recursion
);
2613 if (dev_xmit_complete(rc
)) {
2614 HARD_TX_UNLOCK(dev
, txq
);
2618 HARD_TX_UNLOCK(dev
, txq
);
2619 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2622 /* Recursion is detected! It is possible,
2626 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2632 rcu_read_unlock_bh();
2637 rcu_read_unlock_bh();
2640 EXPORT_SYMBOL(dev_queue_xmit
);
2643 /*=======================================================================
2645 =======================================================================*/
2647 int netdev_max_backlog __read_mostly
= 1000;
2648 EXPORT_SYMBOL(netdev_max_backlog
);
2650 int netdev_tstamp_prequeue __read_mostly
= 1;
2651 int netdev_budget __read_mostly
= 300;
2652 int weight_p __read_mostly
= 64; /* old backlog weight */
2654 /* Called with irq disabled */
2655 static inline void ____napi_schedule(struct softnet_data
*sd
,
2656 struct napi_struct
*napi
)
2658 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
2659 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2663 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2664 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2665 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2666 * if hash is a canonical 4-tuple hash over transport ports.
2668 void __skb_get_rxhash(struct sk_buff
*skb
)
2670 struct flow_keys keys
;
2673 if (!skb_flow_dissect(skb
, &keys
))
2679 /* get a consistent hash (same value on both flow directions) */
2680 if (((__force u32
)keys
.dst
< (__force u32
)keys
.src
) ||
2681 (((__force u32
)keys
.dst
== (__force u32
)keys
.src
) &&
2682 ((__force u16
)keys
.port16
[1] < (__force u16
)keys
.port16
[0]))) {
2683 swap(keys
.dst
, keys
.src
);
2684 swap(keys
.port16
[0], keys
.port16
[1]);
2687 hash
= jhash_3words((__force u32
)keys
.dst
,
2688 (__force u32
)keys
.src
,
2689 (__force u32
)keys
.ports
, hashrnd
);
2695 EXPORT_SYMBOL(__skb_get_rxhash
);
2699 /* One global table that all flow-based protocols share. */
2700 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
2701 EXPORT_SYMBOL(rps_sock_flow_table
);
2703 struct static_key rps_needed __read_mostly
;
2705 static struct rps_dev_flow
*
2706 set_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2707 struct rps_dev_flow
*rflow
, u16 next_cpu
)
2709 if (next_cpu
!= RPS_NO_CPU
) {
2710 #ifdef CONFIG_RFS_ACCEL
2711 struct netdev_rx_queue
*rxqueue
;
2712 struct rps_dev_flow_table
*flow_table
;
2713 struct rps_dev_flow
*old_rflow
;
2718 /* Should we steer this flow to a different hardware queue? */
2719 if (!skb_rx_queue_recorded(skb
) || !dev
->rx_cpu_rmap
||
2720 !(dev
->features
& NETIF_F_NTUPLE
))
2722 rxq_index
= cpu_rmap_lookup_index(dev
->rx_cpu_rmap
, next_cpu
);
2723 if (rxq_index
== skb_get_rx_queue(skb
))
2726 rxqueue
= dev
->_rx
+ rxq_index
;
2727 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2730 flow_id
= skb
->rxhash
& flow_table
->mask
;
2731 rc
= dev
->netdev_ops
->ndo_rx_flow_steer(dev
, skb
,
2732 rxq_index
, flow_id
);
2736 rflow
= &flow_table
->flows
[flow_id
];
2738 if (old_rflow
->filter
== rflow
->filter
)
2739 old_rflow
->filter
= RPS_NO_FILTER
;
2743 per_cpu(softnet_data
, next_cpu
).input_queue_head
;
2746 rflow
->cpu
= next_cpu
;
2751 * get_rps_cpu is called from netif_receive_skb and returns the target
2752 * CPU from the RPS map of the receiving queue for a given skb.
2753 * rcu_read_lock must be held on entry.
2755 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2756 struct rps_dev_flow
**rflowp
)
2758 struct netdev_rx_queue
*rxqueue
;
2759 struct rps_map
*map
;
2760 struct rps_dev_flow_table
*flow_table
;
2761 struct rps_sock_flow_table
*sock_flow_table
;
2765 if (skb_rx_queue_recorded(skb
)) {
2766 u16 index
= skb_get_rx_queue(skb
);
2767 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
2768 WARN_ONCE(dev
->real_num_rx_queues
> 1,
2769 "%s received packet on queue %u, but number "
2770 "of RX queues is %u\n",
2771 dev
->name
, index
, dev
->real_num_rx_queues
);
2774 rxqueue
= dev
->_rx
+ index
;
2778 map
= rcu_dereference(rxqueue
->rps_map
);
2780 if (map
->len
== 1 &&
2781 !rcu_access_pointer(rxqueue
->rps_flow_table
)) {
2782 tcpu
= map
->cpus
[0];
2783 if (cpu_online(tcpu
))
2787 } else if (!rcu_access_pointer(rxqueue
->rps_flow_table
)) {
2791 skb_reset_network_header(skb
);
2792 if (!skb_get_rxhash(skb
))
2795 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2796 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
2797 if (flow_table
&& sock_flow_table
) {
2799 struct rps_dev_flow
*rflow
;
2801 rflow
= &flow_table
->flows
[skb
->rxhash
& flow_table
->mask
];
2804 next_cpu
= sock_flow_table
->ents
[skb
->rxhash
&
2805 sock_flow_table
->mask
];
2808 * If the desired CPU (where last recvmsg was done) is
2809 * different from current CPU (one in the rx-queue flow
2810 * table entry), switch if one of the following holds:
2811 * - Current CPU is unset (equal to RPS_NO_CPU).
2812 * - Current CPU is offline.
2813 * - The current CPU's queue tail has advanced beyond the
2814 * last packet that was enqueued using this table entry.
2815 * This guarantees that all previous packets for the flow
2816 * have been dequeued, thus preserving in order delivery.
2818 if (unlikely(tcpu
!= next_cpu
) &&
2819 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
2820 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
2821 rflow
->last_qtail
)) >= 0)) {
2823 rflow
= set_rps_cpu(dev
, skb
, rflow
, next_cpu
);
2826 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
2834 tcpu
= map
->cpus
[((u64
) skb
->rxhash
* map
->len
) >> 32];
2836 if (cpu_online(tcpu
)) {
2846 #ifdef CONFIG_RFS_ACCEL
2849 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2850 * @dev: Device on which the filter was set
2851 * @rxq_index: RX queue index
2852 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2853 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2855 * Drivers that implement ndo_rx_flow_steer() should periodically call
2856 * this function for each installed filter and remove the filters for
2857 * which it returns %true.
2859 bool rps_may_expire_flow(struct net_device
*dev
, u16 rxq_index
,
2860 u32 flow_id
, u16 filter_id
)
2862 struct netdev_rx_queue
*rxqueue
= dev
->_rx
+ rxq_index
;
2863 struct rps_dev_flow_table
*flow_table
;
2864 struct rps_dev_flow
*rflow
;
2869 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2870 if (flow_table
&& flow_id
<= flow_table
->mask
) {
2871 rflow
= &flow_table
->flows
[flow_id
];
2872 cpu
= ACCESS_ONCE(rflow
->cpu
);
2873 if (rflow
->filter
== filter_id
&& cpu
!= RPS_NO_CPU
&&
2874 ((int)(per_cpu(softnet_data
, cpu
).input_queue_head
-
2875 rflow
->last_qtail
) <
2876 (int)(10 * flow_table
->mask
)))
2882 EXPORT_SYMBOL(rps_may_expire_flow
);
2884 #endif /* CONFIG_RFS_ACCEL */
2886 /* Called from hardirq (IPI) context */
2887 static void rps_trigger_softirq(void *data
)
2889 struct softnet_data
*sd
= data
;
2891 ____napi_schedule(sd
, &sd
->backlog
);
2895 #endif /* CONFIG_RPS */
2898 * Check if this softnet_data structure is another cpu one
2899 * If yes, queue it to our IPI list and return 1
2902 static int rps_ipi_queued(struct softnet_data
*sd
)
2905 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
2908 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
2909 mysd
->rps_ipi_list
= sd
;
2911 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2914 #endif /* CONFIG_RPS */
2919 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2920 * queue (may be a remote CPU queue).
2922 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
2923 unsigned int *qtail
)
2925 struct softnet_data
*sd
;
2926 unsigned long flags
;
2928 sd
= &per_cpu(softnet_data
, cpu
);
2930 local_irq_save(flags
);
2933 if (skb_queue_len(&sd
->input_pkt_queue
) <= netdev_max_backlog
) {
2934 if (skb_queue_len(&sd
->input_pkt_queue
)) {
2936 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
2937 input_queue_tail_incr_save(sd
, qtail
);
2939 local_irq_restore(flags
);
2940 return NET_RX_SUCCESS
;
2943 /* Schedule NAPI for backlog device
2944 * We can use non atomic operation since we own the queue lock
2946 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
2947 if (!rps_ipi_queued(sd
))
2948 ____napi_schedule(sd
, &sd
->backlog
);
2956 local_irq_restore(flags
);
2958 atomic_long_inc(&skb
->dev
->rx_dropped
);
2964 * netif_rx - post buffer to the network code
2965 * @skb: buffer to post
2967 * This function receives a packet from a device driver and queues it for
2968 * the upper (protocol) levels to process. It always succeeds. The buffer
2969 * may be dropped during processing for congestion control or by the
2973 * NET_RX_SUCCESS (no congestion)
2974 * NET_RX_DROP (packet was dropped)
2978 int netif_rx(struct sk_buff
*skb
)
2982 /* if netpoll wants it, pretend we never saw it */
2983 if (netpoll_rx(skb
))
2986 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
2988 trace_netif_rx(skb
);
2990 if (static_key_false(&rps_needed
)) {
2991 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
2997 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
2999 cpu
= smp_processor_id();
3001 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3009 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
3014 EXPORT_SYMBOL(netif_rx
);
3016 int netif_rx_ni(struct sk_buff
*skb
)
3021 err
= netif_rx(skb
);
3022 if (local_softirq_pending())
3028 EXPORT_SYMBOL(netif_rx_ni
);
3030 static void net_tx_action(struct softirq_action
*h
)
3032 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3034 if (sd
->completion_queue
) {
3035 struct sk_buff
*clist
;
3037 local_irq_disable();
3038 clist
= sd
->completion_queue
;
3039 sd
->completion_queue
= NULL
;
3043 struct sk_buff
*skb
= clist
;
3044 clist
= clist
->next
;
3046 WARN_ON(atomic_read(&skb
->users
));
3047 trace_kfree_skb(skb
, net_tx_action
);
3052 if (sd
->output_queue
) {
3055 local_irq_disable();
3056 head
= sd
->output_queue
;
3057 sd
->output_queue
= NULL
;
3058 sd
->output_queue_tailp
= &sd
->output_queue
;
3062 struct Qdisc
*q
= head
;
3063 spinlock_t
*root_lock
;
3065 head
= head
->next_sched
;
3067 root_lock
= qdisc_lock(q
);
3068 if (spin_trylock(root_lock
)) {
3069 smp_mb__before_clear_bit();
3070 clear_bit(__QDISC_STATE_SCHED
,
3073 spin_unlock(root_lock
);
3075 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
3077 __netif_reschedule(q
);
3079 smp_mb__before_clear_bit();
3080 clear_bit(__QDISC_STATE_SCHED
,
3088 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3089 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3090 /* This hook is defined here for ATM LANE */
3091 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
3092 unsigned char *addr
) __read_mostly
;
3093 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
3096 #ifdef CONFIG_NET_CLS_ACT
3097 /* TODO: Maybe we should just force sch_ingress to be compiled in
3098 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3099 * a compare and 2 stores extra right now if we dont have it on
3100 * but have CONFIG_NET_CLS_ACT
3101 * NOTE: This doesn't stop any functionality; if you dont have
3102 * the ingress scheduler, you just can't add policies on ingress.
3105 static int ing_filter(struct sk_buff
*skb
, struct netdev_queue
*rxq
)
3107 struct net_device
*dev
= skb
->dev
;
3108 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
3109 int result
= TC_ACT_OK
;
3112 if (unlikely(MAX_RED_LOOP
< ttl
++)) {
3113 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3114 skb
->skb_iif
, dev
->ifindex
);
3118 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
3119 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
3122 if (q
!= &noop_qdisc
) {
3123 spin_lock(qdisc_lock(q
));
3124 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
3125 result
= qdisc_enqueue_root(skb
, q
);
3126 spin_unlock(qdisc_lock(q
));
3132 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
3133 struct packet_type
**pt_prev
,
3134 int *ret
, struct net_device
*orig_dev
)
3136 struct netdev_queue
*rxq
= rcu_dereference(skb
->dev
->ingress_queue
);
3138 if (!rxq
|| rxq
->qdisc
== &noop_qdisc
)
3142 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
3146 switch (ing_filter(skb
, rxq
)) {
3160 * netdev_rx_handler_register - register receive handler
3161 * @dev: device to register a handler for
3162 * @rx_handler: receive handler to register
3163 * @rx_handler_data: data pointer that is used by rx handler
3165 * Register a receive hander for a device. This handler will then be
3166 * called from __netif_receive_skb. A negative errno code is returned
3169 * The caller must hold the rtnl_mutex.
3171 * For a general description of rx_handler, see enum rx_handler_result.
3173 int netdev_rx_handler_register(struct net_device
*dev
,
3174 rx_handler_func_t
*rx_handler
,
3175 void *rx_handler_data
)
3179 if (dev
->rx_handler
)
3182 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
3183 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
3187 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
3190 * netdev_rx_handler_unregister - unregister receive handler
3191 * @dev: device to unregister a handler from
3193 * Unregister a receive hander from a device.
3195 * The caller must hold the rtnl_mutex.
3197 void netdev_rx_handler_unregister(struct net_device
*dev
)
3201 RCU_INIT_POINTER(dev
->rx_handler
, NULL
);
3202 RCU_INIT_POINTER(dev
->rx_handler_data
, NULL
);
3204 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
3207 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3208 * the special handling of PFMEMALLOC skbs.
3210 static bool skb_pfmemalloc_protocol(struct sk_buff
*skb
)
3212 switch (skb
->protocol
) {
3213 case __constant_htons(ETH_P_ARP
):
3214 case __constant_htons(ETH_P_IP
):
3215 case __constant_htons(ETH_P_IPV6
):
3216 case __constant_htons(ETH_P_8021Q
):
3223 static int __netif_receive_skb(struct sk_buff
*skb
)
3225 struct packet_type
*ptype
, *pt_prev
;
3226 rx_handler_func_t
*rx_handler
;
3227 struct net_device
*orig_dev
;
3228 struct net_device
*null_or_dev
;
3229 bool deliver_exact
= false;
3230 int ret
= NET_RX_DROP
;
3232 unsigned long pflags
= current
->flags
;
3234 net_timestamp_check(!netdev_tstamp_prequeue
, skb
);
3236 trace_netif_receive_skb(skb
);
3239 * PFMEMALLOC skbs are special, they should
3240 * - be delivered to SOCK_MEMALLOC sockets only
3241 * - stay away from userspace
3242 * - have bounded memory usage
3244 * Use PF_MEMALLOC as this saves us from propagating the allocation
3245 * context down to all allocation sites.
3247 if (sk_memalloc_socks() && skb_pfmemalloc(skb
))
3248 current
->flags
|= PF_MEMALLOC
;
3250 /* if we've gotten here through NAPI, check netpoll */
3251 if (netpoll_receive_skb(skb
))
3254 orig_dev
= skb
->dev
;
3256 skb_reset_network_header(skb
);
3257 skb_reset_transport_header(skb
);
3258 skb_reset_mac_len(skb
);
3265 skb
->skb_iif
= skb
->dev
->ifindex
;
3267 __this_cpu_inc(softnet_data
.processed
);
3269 if (skb
->protocol
== cpu_to_be16(ETH_P_8021Q
)) {
3270 skb
= vlan_untag(skb
);
3275 #ifdef CONFIG_NET_CLS_ACT
3276 if (skb
->tc_verd
& TC_NCLS
) {
3277 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
3282 if (sk_memalloc_socks() && skb_pfmemalloc(skb
))
3285 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
3286 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
3288 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3294 #ifdef CONFIG_NET_CLS_ACT
3295 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
3301 if (sk_memalloc_socks() && skb_pfmemalloc(skb
)
3302 && !skb_pfmemalloc_protocol(skb
))
3305 if (vlan_tx_tag_present(skb
)) {
3307 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3310 if (vlan_do_receive(&skb
))
3312 else if (unlikely(!skb
))
3316 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
3319 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3322 switch (rx_handler(&skb
)) {
3323 case RX_HANDLER_CONSUMED
:
3325 case RX_HANDLER_ANOTHER
:
3327 case RX_HANDLER_EXACT
:
3328 deliver_exact
= true;
3329 case RX_HANDLER_PASS
:
3336 if (vlan_tx_nonzero_tag_present(skb
))
3337 skb
->pkt_type
= PACKET_OTHERHOST
;
3339 /* deliver only exact match when indicated */
3340 null_or_dev
= deliver_exact
? skb
->dev
: NULL
;
3342 type
= skb
->protocol
;
3343 list_for_each_entry_rcu(ptype
,
3344 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
3345 if (ptype
->type
== type
&&
3346 (ptype
->dev
== null_or_dev
|| ptype
->dev
== skb
->dev
||
3347 ptype
->dev
== orig_dev
)) {
3349 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3355 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
3358 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
3361 atomic_long_inc(&skb
->dev
->rx_dropped
);
3363 /* Jamal, now you will not able to escape explaining
3364 * me how you were going to use this. :-)
3372 tsk_restore_flags(current
, pflags
, PF_MEMALLOC
);
3377 * netif_receive_skb - process receive buffer from network
3378 * @skb: buffer to process
3380 * netif_receive_skb() is the main receive data processing function.
3381 * It always succeeds. The buffer may be dropped during processing
3382 * for congestion control or by the protocol layers.
3384 * This function may only be called from softirq context and interrupts
3385 * should be enabled.
3387 * Return values (usually ignored):
3388 * NET_RX_SUCCESS: no congestion
3389 * NET_RX_DROP: packet was dropped
3391 int netif_receive_skb(struct sk_buff
*skb
)
3393 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3395 if (skb_defer_rx_timestamp(skb
))
3396 return NET_RX_SUCCESS
;
3399 if (static_key_false(&rps_needed
)) {
3400 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3405 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3408 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3415 return __netif_receive_skb(skb
);
3417 EXPORT_SYMBOL(netif_receive_skb
);
3419 /* Network device is going away, flush any packets still pending
3420 * Called with irqs disabled.
3422 static void flush_backlog(void *arg
)
3424 struct net_device
*dev
= arg
;
3425 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3426 struct sk_buff
*skb
, *tmp
;
3429 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3430 if (skb
->dev
== dev
) {
3431 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3433 input_queue_head_incr(sd
);
3438 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3439 if (skb
->dev
== dev
) {
3440 __skb_unlink(skb
, &sd
->process_queue
);
3442 input_queue_head_incr(sd
);
3447 static int napi_gro_complete(struct sk_buff
*skb
)
3449 struct packet_type
*ptype
;
3450 __be16 type
= skb
->protocol
;
3451 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3454 BUILD_BUG_ON(sizeof(struct napi_gro_cb
) > sizeof(skb
->cb
));
3456 if (NAPI_GRO_CB(skb
)->count
== 1) {
3457 skb_shinfo(skb
)->gso_size
= 0;
3462 list_for_each_entry_rcu(ptype
, head
, list
) {
3463 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_complete
)
3466 err
= ptype
->gro_complete(skb
);
3472 WARN_ON(&ptype
->list
== head
);
3474 return NET_RX_SUCCESS
;
3478 return netif_receive_skb(skb
);
3481 /* napi->gro_list contains packets ordered by age.
3482 * youngest packets at the head of it.
3483 * Complete skbs in reverse order to reduce latencies.
3485 void napi_gro_flush(struct napi_struct
*napi
, bool flush_old
)
3487 struct sk_buff
*skb
, *prev
= NULL
;
3489 /* scan list and build reverse chain */
3490 for (skb
= napi
->gro_list
; skb
!= NULL
; skb
= skb
->next
) {
3495 for (skb
= prev
; skb
; skb
= prev
) {
3498 if (flush_old
&& NAPI_GRO_CB(skb
)->age
== jiffies
)
3502 napi_gro_complete(skb
);
3506 napi
->gro_list
= NULL
;
3508 EXPORT_SYMBOL(napi_gro_flush
);
3510 enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3512 struct sk_buff
**pp
= NULL
;
3513 struct packet_type
*ptype
;
3514 __be16 type
= skb
->protocol
;
3515 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3518 enum gro_result ret
;
3520 if (!(skb
->dev
->features
& NETIF_F_GRO
) || netpoll_rx_on(skb
))
3523 if (skb_is_gso(skb
) || skb_has_frag_list(skb
))
3527 list_for_each_entry_rcu(ptype
, head
, list
) {
3528 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_receive
)
3531 skb_set_network_header(skb
, skb_gro_offset(skb
));
3532 mac_len
= skb
->network_header
- skb
->mac_header
;
3533 skb
->mac_len
= mac_len
;
3534 NAPI_GRO_CB(skb
)->same_flow
= 0;
3535 NAPI_GRO_CB(skb
)->flush
= 0;
3536 NAPI_GRO_CB(skb
)->free
= 0;
3538 pp
= ptype
->gro_receive(&napi
->gro_list
, skb
);
3543 if (&ptype
->list
== head
)
3546 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3547 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3550 struct sk_buff
*nskb
= *pp
;
3554 napi_gro_complete(nskb
);
3561 if (NAPI_GRO_CB(skb
)->flush
|| napi
->gro_count
>= MAX_GRO_SKBS
)
3565 NAPI_GRO_CB(skb
)->count
= 1;
3566 NAPI_GRO_CB(skb
)->age
= jiffies
;
3567 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
3568 skb
->next
= napi
->gro_list
;
3569 napi
->gro_list
= skb
;
3573 if (skb_headlen(skb
) < skb_gro_offset(skb
)) {
3574 int grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
3576 BUG_ON(skb
->end
- skb
->tail
< grow
);
3578 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3581 skb
->data_len
-= grow
;
3583 skb_shinfo(skb
)->frags
[0].page_offset
+= grow
;
3584 skb_frag_size_sub(&skb_shinfo(skb
)->frags
[0], grow
);
3586 if (unlikely(!skb_frag_size(&skb_shinfo(skb
)->frags
[0]))) {
3587 skb_frag_unref(skb
, 0);
3588 memmove(skb_shinfo(skb
)->frags
,
3589 skb_shinfo(skb
)->frags
+ 1,
3590 --skb_shinfo(skb
)->nr_frags
* sizeof(skb_frag_t
));
3601 EXPORT_SYMBOL(dev_gro_receive
);
3603 static inline gro_result_t
3604 __napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3607 unsigned int maclen
= skb
->dev
->hard_header_len
;
3609 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3610 unsigned long diffs
;
3612 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
3613 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
3614 if (maclen
== ETH_HLEN
)
3615 diffs
|= compare_ether_header(skb_mac_header(p
),
3616 skb_gro_mac_header(skb
));
3618 diffs
= memcmp(skb_mac_header(p
),
3619 skb_gro_mac_header(skb
),
3621 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
3622 NAPI_GRO_CB(p
)->flush
= 0;
3625 return dev_gro_receive(napi
, skb
);
3628 gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
3632 if (netif_receive_skb(skb
))
3640 case GRO_MERGED_FREE
:
3641 if (NAPI_GRO_CB(skb
)->free
== NAPI_GRO_FREE_STOLEN_HEAD
)
3642 kmem_cache_free(skbuff_head_cache
, skb
);
3654 EXPORT_SYMBOL(napi_skb_finish
);
3656 static void skb_gro_reset_offset(struct sk_buff
*skb
)
3658 const struct skb_shared_info
*pinfo
= skb_shinfo(skb
);
3659 const skb_frag_t
*frag0
= &pinfo
->frags
[0];
3661 NAPI_GRO_CB(skb
)->data_offset
= 0;
3662 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3663 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3665 if (skb
->mac_header
== skb
->tail
&&
3667 !PageHighMem(skb_frag_page(frag0
))) {
3668 NAPI_GRO_CB(skb
)->frag0
= skb_frag_address(frag0
);
3669 NAPI_GRO_CB(skb
)->frag0_len
= skb_frag_size(frag0
);
3673 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3675 skb_gro_reset_offset(skb
);
3677 return napi_skb_finish(__napi_gro_receive(napi
, skb
), skb
);
3679 EXPORT_SYMBOL(napi_gro_receive
);
3681 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
3683 __skb_pull(skb
, skb_headlen(skb
));
3684 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3685 skb_reserve(skb
, NET_SKB_PAD
+ NET_IP_ALIGN
- skb_headroom(skb
));
3687 skb
->dev
= napi
->dev
;
3693 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
3695 struct sk_buff
*skb
= napi
->skb
;
3698 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
3704 EXPORT_SYMBOL(napi_get_frags
);
3706 gro_result_t
napi_frags_finish(struct napi_struct
*napi
, struct sk_buff
*skb
,
3712 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
3714 if (ret
== GRO_HELD
)
3715 skb_gro_pull(skb
, -ETH_HLEN
);
3716 else if (netif_receive_skb(skb
))
3721 case GRO_MERGED_FREE
:
3722 napi_reuse_skb(napi
, skb
);
3731 EXPORT_SYMBOL(napi_frags_finish
);
3733 static struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
3735 struct sk_buff
*skb
= napi
->skb
;
3742 skb_reset_mac_header(skb
);
3743 skb_gro_reset_offset(skb
);
3745 off
= skb_gro_offset(skb
);
3746 hlen
= off
+ sizeof(*eth
);
3747 eth
= skb_gro_header_fast(skb
, off
);
3748 if (skb_gro_header_hard(skb
, hlen
)) {
3749 eth
= skb_gro_header_slow(skb
, hlen
, off
);
3750 if (unlikely(!eth
)) {
3751 napi_reuse_skb(napi
, skb
);
3757 skb_gro_pull(skb
, sizeof(*eth
));
3760 * This works because the only protocols we care about don't require
3761 * special handling. We'll fix it up properly at the end.
3763 skb
->protocol
= eth
->h_proto
;
3769 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
3771 struct sk_buff
*skb
= napi_frags_skb(napi
);
3776 return napi_frags_finish(napi
, skb
, __napi_gro_receive(napi
, skb
));
3778 EXPORT_SYMBOL(napi_gro_frags
);
3781 * net_rps_action sends any pending IPI's for rps.
3782 * Note: called with local irq disabled, but exits with local irq enabled.
3784 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
3787 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
3790 sd
->rps_ipi_list
= NULL
;
3794 /* Send pending IPI's to kick RPS processing on remote cpus. */
3796 struct softnet_data
*next
= remsd
->rps_ipi_next
;
3798 if (cpu_online(remsd
->cpu
))
3799 __smp_call_function_single(remsd
->cpu
,
3808 static int process_backlog(struct napi_struct
*napi
, int quota
)
3811 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
3814 /* Check if we have pending ipi, its better to send them now,
3815 * not waiting net_rx_action() end.
3817 if (sd
->rps_ipi_list
) {
3818 local_irq_disable();
3819 net_rps_action_and_irq_enable(sd
);
3822 napi
->weight
= weight_p
;
3823 local_irq_disable();
3824 while (work
< quota
) {
3825 struct sk_buff
*skb
;
3828 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
3830 __netif_receive_skb(skb
);
3831 local_irq_disable();
3832 input_queue_head_incr(sd
);
3833 if (++work
>= quota
) {
3840 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3842 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
3843 &sd
->process_queue
);
3845 if (qlen
< quota
- work
) {
3847 * Inline a custom version of __napi_complete().
3848 * only current cpu owns and manipulates this napi,
3849 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3850 * we can use a plain write instead of clear_bit(),
3851 * and we dont need an smp_mb() memory barrier.
3853 list_del(&napi
->poll_list
);
3856 quota
= work
+ qlen
;
3866 * __napi_schedule - schedule for receive
3867 * @n: entry to schedule
3869 * The entry's receive function will be scheduled to run
3871 void __napi_schedule(struct napi_struct
*n
)
3873 unsigned long flags
;
3875 local_irq_save(flags
);
3876 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
3877 local_irq_restore(flags
);
3879 EXPORT_SYMBOL(__napi_schedule
);
3881 void __napi_complete(struct napi_struct
*n
)
3883 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
3884 BUG_ON(n
->gro_list
);
3886 list_del(&n
->poll_list
);
3887 smp_mb__before_clear_bit();
3888 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
3890 EXPORT_SYMBOL(__napi_complete
);
3892 void napi_complete(struct napi_struct
*n
)
3894 unsigned long flags
;
3897 * don't let napi dequeue from the cpu poll list
3898 * just in case its running on a different cpu
3900 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
3903 napi_gro_flush(n
, false);
3904 local_irq_save(flags
);
3906 local_irq_restore(flags
);
3908 EXPORT_SYMBOL(napi_complete
);
3910 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
3911 int (*poll
)(struct napi_struct
*, int), int weight
)
3913 INIT_LIST_HEAD(&napi
->poll_list
);
3914 napi
->gro_count
= 0;
3915 napi
->gro_list
= NULL
;
3918 napi
->weight
= weight
;
3919 list_add(&napi
->dev_list
, &dev
->napi_list
);
3921 #ifdef CONFIG_NETPOLL
3922 spin_lock_init(&napi
->poll_lock
);
3923 napi
->poll_owner
= -1;
3925 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
3927 EXPORT_SYMBOL(netif_napi_add
);
3929 void netif_napi_del(struct napi_struct
*napi
)
3931 struct sk_buff
*skb
, *next
;
3933 list_del_init(&napi
->dev_list
);
3934 napi_free_frags(napi
);
3936 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3942 napi
->gro_list
= NULL
;
3943 napi
->gro_count
= 0;
3945 EXPORT_SYMBOL(netif_napi_del
);
3947 static void net_rx_action(struct softirq_action
*h
)
3949 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3950 unsigned long time_limit
= jiffies
+ 2;
3951 int budget
= netdev_budget
;
3954 local_irq_disable();
3956 while (!list_empty(&sd
->poll_list
)) {
3957 struct napi_struct
*n
;
3960 /* If softirq window is exhuasted then punt.
3961 * Allow this to run for 2 jiffies since which will allow
3962 * an average latency of 1.5/HZ.
3964 if (unlikely(budget
<= 0 || time_after(jiffies
, time_limit
)))
3969 /* Even though interrupts have been re-enabled, this
3970 * access is safe because interrupts can only add new
3971 * entries to the tail of this list, and only ->poll()
3972 * calls can remove this head entry from the list.
3974 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
3976 have
= netpoll_poll_lock(n
);
3980 /* This NAPI_STATE_SCHED test is for avoiding a race
3981 * with netpoll's poll_napi(). Only the entity which
3982 * obtains the lock and sees NAPI_STATE_SCHED set will
3983 * actually make the ->poll() call. Therefore we avoid
3984 * accidentally calling ->poll() when NAPI is not scheduled.
3987 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
3988 work
= n
->poll(n
, weight
);
3992 WARN_ON_ONCE(work
> weight
);
3996 local_irq_disable();
3998 /* Drivers must not modify the NAPI state if they
3999 * consume the entire weight. In such cases this code
4000 * still "owns" the NAPI instance and therefore can
4001 * move the instance around on the list at-will.
4003 if (unlikely(work
== weight
)) {
4004 if (unlikely(napi_disable_pending(n
))) {
4007 local_irq_disable();
4010 /* flush too old packets
4011 * If HZ < 1000, flush all packets.
4014 napi_gro_flush(n
, HZ
>= 1000);
4015 local_irq_disable();
4017 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
4021 netpoll_poll_unlock(have
);
4024 net_rps_action_and_irq_enable(sd
);
4026 #ifdef CONFIG_NET_DMA
4028 * There may not be any more sk_buffs coming right now, so push
4029 * any pending DMA copies to hardware
4031 dma_issue_pending_all();
4038 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
4042 static gifconf_func_t
*gifconf_list
[NPROTO
];
4045 * register_gifconf - register a SIOCGIF handler
4046 * @family: Address family
4047 * @gifconf: Function handler
4049 * Register protocol dependent address dumping routines. The handler
4050 * that is passed must not be freed or reused until it has been replaced
4051 * by another handler.
4053 int register_gifconf(unsigned int family
, gifconf_func_t
*gifconf
)
4055 if (family
>= NPROTO
)
4057 gifconf_list
[family
] = gifconf
;
4060 EXPORT_SYMBOL(register_gifconf
);
4064 * Map an interface index to its name (SIOCGIFNAME)
4068 * We need this ioctl for efficient implementation of the
4069 * if_indextoname() function required by the IPv6 API. Without
4070 * it, we would have to search all the interfaces to find a
4074 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
4076 struct net_device
*dev
;
4080 * Fetch the caller's info block.
4083 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
4087 dev
= dev_get_by_index_rcu(net
, ifr
.ifr_ifindex
);
4093 strcpy(ifr
.ifr_name
, dev
->name
);
4096 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
4102 * Perform a SIOCGIFCONF call. This structure will change
4103 * size eventually, and there is nothing I can do about it.
4104 * Thus we will need a 'compatibility mode'.
4107 static int dev_ifconf(struct net
*net
, char __user
*arg
)
4110 struct net_device
*dev
;
4117 * Fetch the caller's info block.
4120 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
4127 * Loop over the interfaces, and write an info block for each.
4131 for_each_netdev(net
, dev
) {
4132 for (i
= 0; i
< NPROTO
; i
++) {
4133 if (gifconf_list
[i
]) {
4136 done
= gifconf_list
[i
](dev
, NULL
, 0);
4138 done
= gifconf_list
[i
](dev
, pos
+ total
,
4148 * All done. Write the updated control block back to the caller.
4150 ifc
.ifc_len
= total
;
4153 * Both BSD and Solaris return 0 here, so we do too.
4155 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
4158 #ifdef CONFIG_PROC_FS
4160 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4162 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4163 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4164 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4166 static inline struct net_device
*dev_from_same_bucket(struct seq_file
*seq
, loff_t
*pos
)
4168 struct net
*net
= seq_file_net(seq
);
4169 struct net_device
*dev
;
4170 struct hlist_node
*p
;
4171 struct hlist_head
*h
;
4172 unsigned int count
= 0, offset
= get_offset(*pos
);
4174 h
= &net
->dev_name_head
[get_bucket(*pos
)];
4175 hlist_for_each_entry_rcu(dev
, p
, h
, name_hlist
) {
4176 if (++count
== offset
)
4183 static inline struct net_device
*dev_from_bucket(struct seq_file
*seq
, loff_t
*pos
)
4185 struct net_device
*dev
;
4186 unsigned int bucket
;
4189 dev
= dev_from_same_bucket(seq
, pos
);
4193 bucket
= get_bucket(*pos
) + 1;
4194 *pos
= set_bucket_offset(bucket
, 1);
4195 } while (bucket
< NETDEV_HASHENTRIES
);
4201 * This is invoked by the /proc filesystem handler to display a device
4204 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4209 return SEQ_START_TOKEN
;
4211 if (get_bucket(*pos
) >= NETDEV_HASHENTRIES
)
4214 return dev_from_bucket(seq
, pos
);
4217 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4220 return dev_from_bucket(seq
, pos
);
4223 void dev_seq_stop(struct seq_file
*seq
, void *v
)
4229 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
4231 struct rtnl_link_stats64 temp
;
4232 const struct rtnl_link_stats64
*stats
= dev_get_stats(dev
, &temp
);
4234 seq_printf(seq
, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4235 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4236 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
4238 stats
->rx_dropped
+ stats
->rx_missed_errors
,
4239 stats
->rx_fifo_errors
,
4240 stats
->rx_length_errors
+ stats
->rx_over_errors
+
4241 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
4242 stats
->rx_compressed
, stats
->multicast
,
4243 stats
->tx_bytes
, stats
->tx_packets
,
4244 stats
->tx_errors
, stats
->tx_dropped
,
4245 stats
->tx_fifo_errors
, stats
->collisions
,
4246 stats
->tx_carrier_errors
+
4247 stats
->tx_aborted_errors
+
4248 stats
->tx_window_errors
+
4249 stats
->tx_heartbeat_errors
,
4250 stats
->tx_compressed
);
4254 * Called from the PROCfs module. This now uses the new arbitrary sized
4255 * /proc/net interface to create /proc/net/dev
4257 static int dev_seq_show(struct seq_file
*seq
, void *v
)
4259 if (v
== SEQ_START_TOKEN
)
4260 seq_puts(seq
, "Inter-| Receive "
4262 " face |bytes packets errs drop fifo frame "
4263 "compressed multicast|bytes packets errs "
4264 "drop fifo colls carrier compressed\n");
4266 dev_seq_printf_stats(seq
, v
);
4270 static struct softnet_data
*softnet_get_online(loff_t
*pos
)
4272 struct softnet_data
*sd
= NULL
;
4274 while (*pos
< nr_cpu_ids
)
4275 if (cpu_online(*pos
)) {
4276 sd
= &per_cpu(softnet_data
, *pos
);
4283 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4285 return softnet_get_online(pos
);
4288 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4291 return softnet_get_online(pos
);
4294 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
4298 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
4300 struct softnet_data
*sd
= v
;
4302 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4303 sd
->processed
, sd
->dropped
, sd
->time_squeeze
, 0,
4304 0, 0, 0, 0, /* was fastroute */
4305 sd
->cpu_collision
, sd
->received_rps
);
4309 static const struct seq_operations dev_seq_ops
= {
4310 .start
= dev_seq_start
,
4311 .next
= dev_seq_next
,
4312 .stop
= dev_seq_stop
,
4313 .show
= dev_seq_show
,
4316 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
4318 return seq_open_net(inode
, file
, &dev_seq_ops
,
4319 sizeof(struct seq_net_private
));
4322 static const struct file_operations dev_seq_fops
= {
4323 .owner
= THIS_MODULE
,
4324 .open
= dev_seq_open
,
4326 .llseek
= seq_lseek
,
4327 .release
= seq_release_net
,
4330 static const struct seq_operations softnet_seq_ops
= {
4331 .start
= softnet_seq_start
,
4332 .next
= softnet_seq_next
,
4333 .stop
= softnet_seq_stop
,
4334 .show
= softnet_seq_show
,
4337 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
4339 return seq_open(file
, &softnet_seq_ops
);
4342 static const struct file_operations softnet_seq_fops
= {
4343 .owner
= THIS_MODULE
,
4344 .open
= softnet_seq_open
,
4346 .llseek
= seq_lseek
,
4347 .release
= seq_release
,
4350 static void *ptype_get_idx(loff_t pos
)
4352 struct packet_type
*pt
= NULL
;
4356 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
4362 for (t
= 0; t
< PTYPE_HASH_SIZE
; t
++) {
4363 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
4372 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4376 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
4379 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4381 struct packet_type
*pt
;
4382 struct list_head
*nxt
;
4386 if (v
== SEQ_START_TOKEN
)
4387 return ptype_get_idx(0);
4390 nxt
= pt
->list
.next
;
4391 if (pt
->type
== htons(ETH_P_ALL
)) {
4392 if (nxt
!= &ptype_all
)
4395 nxt
= ptype_base
[0].next
;
4397 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
4399 while (nxt
== &ptype_base
[hash
]) {
4400 if (++hash
>= PTYPE_HASH_SIZE
)
4402 nxt
= ptype_base
[hash
].next
;
4405 return list_entry(nxt
, struct packet_type
, list
);
4408 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
4414 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
4416 struct packet_type
*pt
= v
;
4418 if (v
== SEQ_START_TOKEN
)
4419 seq_puts(seq
, "Type Device Function\n");
4420 else if (pt
->dev
== NULL
|| dev_net(pt
->dev
) == seq_file_net(seq
)) {
4421 if (pt
->type
== htons(ETH_P_ALL
))
4422 seq_puts(seq
, "ALL ");
4424 seq_printf(seq
, "%04x", ntohs(pt
->type
));
4426 seq_printf(seq
, " %-8s %pF\n",
4427 pt
->dev
? pt
->dev
->name
: "", pt
->func
);
4433 static const struct seq_operations ptype_seq_ops
= {
4434 .start
= ptype_seq_start
,
4435 .next
= ptype_seq_next
,
4436 .stop
= ptype_seq_stop
,
4437 .show
= ptype_seq_show
,
4440 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
4442 return seq_open_net(inode
, file
, &ptype_seq_ops
,
4443 sizeof(struct seq_net_private
));
4446 static const struct file_operations ptype_seq_fops
= {
4447 .owner
= THIS_MODULE
,
4448 .open
= ptype_seq_open
,
4450 .llseek
= seq_lseek
,
4451 .release
= seq_release_net
,
4455 static int __net_init
dev_proc_net_init(struct net
*net
)
4459 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
4461 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
4463 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
4466 if (wext_proc_init(net
))
4472 proc_net_remove(net
, "ptype");
4474 proc_net_remove(net
, "softnet_stat");
4476 proc_net_remove(net
, "dev");
4480 static void __net_exit
dev_proc_net_exit(struct net
*net
)
4482 wext_proc_exit(net
);
4484 proc_net_remove(net
, "ptype");
4485 proc_net_remove(net
, "softnet_stat");
4486 proc_net_remove(net
, "dev");
4489 static struct pernet_operations __net_initdata dev_proc_ops
= {
4490 .init
= dev_proc_net_init
,
4491 .exit
= dev_proc_net_exit
,
4494 static int __init
dev_proc_init(void)
4496 return register_pernet_subsys(&dev_proc_ops
);
4499 #define dev_proc_init() 0
4500 #endif /* CONFIG_PROC_FS */
4504 * netdev_set_master - set up master pointer
4505 * @slave: slave device
4506 * @master: new master device
4508 * Changes the master device of the slave. Pass %NULL to break the
4509 * bonding. The caller must hold the RTNL semaphore. On a failure
4510 * a negative errno code is returned. On success the reference counts
4511 * are adjusted and the function returns zero.
4513 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
4515 struct net_device
*old
= slave
->master
;
4525 slave
->master
= master
;
4531 EXPORT_SYMBOL(netdev_set_master
);
4534 * netdev_set_bond_master - set up bonding master/slave pair
4535 * @slave: slave device
4536 * @master: new master device
4538 * Changes the master device of the slave. Pass %NULL to break the
4539 * bonding. The caller must hold the RTNL semaphore. On a failure
4540 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4541 * to the routing socket and the function returns zero.
4543 int netdev_set_bond_master(struct net_device
*slave
, struct net_device
*master
)
4549 err
= netdev_set_master(slave
, master
);
4553 slave
->flags
|= IFF_SLAVE
;
4555 slave
->flags
&= ~IFF_SLAVE
;
4557 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
4560 EXPORT_SYMBOL(netdev_set_bond_master
);
4562 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
4564 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4566 if ((dev
->flags
& IFF_UP
) && ops
->ndo_change_rx_flags
)
4567 ops
->ndo_change_rx_flags(dev
, flags
);
4570 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
4572 unsigned int old_flags
= dev
->flags
;
4578 dev
->flags
|= IFF_PROMISC
;
4579 dev
->promiscuity
+= inc
;
4580 if (dev
->promiscuity
== 0) {
4583 * If inc causes overflow, untouch promisc and return error.
4586 dev
->flags
&= ~IFF_PROMISC
;
4588 dev
->promiscuity
-= inc
;
4589 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4594 if (dev
->flags
!= old_flags
) {
4595 pr_info("device %s %s promiscuous mode\n",
4597 dev
->flags
& IFF_PROMISC
? "entered" : "left");
4598 if (audit_enabled
) {
4599 current_uid_gid(&uid
, &gid
);
4600 audit_log(current
->audit_context
, GFP_ATOMIC
,
4601 AUDIT_ANOM_PROMISCUOUS
,
4602 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4603 dev
->name
, (dev
->flags
& IFF_PROMISC
),
4604 (old_flags
& IFF_PROMISC
),
4605 from_kuid(&init_user_ns
, audit_get_loginuid(current
)),
4606 from_kuid(&init_user_ns
, uid
),
4607 from_kgid(&init_user_ns
, gid
),
4608 audit_get_sessionid(current
));
4611 dev_change_rx_flags(dev
, IFF_PROMISC
);
4617 * dev_set_promiscuity - update promiscuity count on a device
4621 * Add or remove promiscuity from a device. While the count in the device
4622 * remains above zero the interface remains promiscuous. Once it hits zero
4623 * the device reverts back to normal filtering operation. A negative inc
4624 * value is used to drop promiscuity on the device.
4625 * Return 0 if successful or a negative errno code on error.
4627 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
4629 unsigned int old_flags
= dev
->flags
;
4632 err
= __dev_set_promiscuity(dev
, inc
);
4635 if (dev
->flags
!= old_flags
)
4636 dev_set_rx_mode(dev
);
4639 EXPORT_SYMBOL(dev_set_promiscuity
);
4642 * dev_set_allmulti - update allmulti count on a device
4646 * Add or remove reception of all multicast frames to a device. While the
4647 * count in the device remains above zero the interface remains listening
4648 * to all interfaces. Once it hits zero the device reverts back to normal
4649 * filtering operation. A negative @inc value is used to drop the counter
4650 * when releasing a resource needing all multicasts.
4651 * Return 0 if successful or a negative errno code on error.
4654 int dev_set_allmulti(struct net_device
*dev
, int inc
)
4656 unsigned int old_flags
= dev
->flags
;
4660 dev
->flags
|= IFF_ALLMULTI
;
4661 dev
->allmulti
+= inc
;
4662 if (dev
->allmulti
== 0) {
4665 * If inc causes overflow, untouch allmulti and return error.
4668 dev
->flags
&= ~IFF_ALLMULTI
;
4670 dev
->allmulti
-= inc
;
4671 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4676 if (dev
->flags
^ old_flags
) {
4677 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
4678 dev_set_rx_mode(dev
);
4682 EXPORT_SYMBOL(dev_set_allmulti
);
4685 * Upload unicast and multicast address lists to device and
4686 * configure RX filtering. When the device doesn't support unicast
4687 * filtering it is put in promiscuous mode while unicast addresses
4690 void __dev_set_rx_mode(struct net_device
*dev
)
4692 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4694 /* dev_open will call this function so the list will stay sane. */
4695 if (!(dev
->flags
&IFF_UP
))
4698 if (!netif_device_present(dev
))
4701 if (!(dev
->priv_flags
& IFF_UNICAST_FLT
)) {
4702 /* Unicast addresses changes may only happen under the rtnl,
4703 * therefore calling __dev_set_promiscuity here is safe.
4705 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
4706 __dev_set_promiscuity(dev
, 1);
4707 dev
->uc_promisc
= true;
4708 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
4709 __dev_set_promiscuity(dev
, -1);
4710 dev
->uc_promisc
= false;
4714 if (ops
->ndo_set_rx_mode
)
4715 ops
->ndo_set_rx_mode(dev
);
4718 void dev_set_rx_mode(struct net_device
*dev
)
4720 netif_addr_lock_bh(dev
);
4721 __dev_set_rx_mode(dev
);
4722 netif_addr_unlock_bh(dev
);
4726 * dev_get_flags - get flags reported to userspace
4729 * Get the combination of flag bits exported through APIs to userspace.
4731 unsigned int dev_get_flags(const struct net_device
*dev
)
4735 flags
= (dev
->flags
& ~(IFF_PROMISC
|
4740 (dev
->gflags
& (IFF_PROMISC
|
4743 if (netif_running(dev
)) {
4744 if (netif_oper_up(dev
))
4745 flags
|= IFF_RUNNING
;
4746 if (netif_carrier_ok(dev
))
4747 flags
|= IFF_LOWER_UP
;
4748 if (netif_dormant(dev
))
4749 flags
|= IFF_DORMANT
;
4754 EXPORT_SYMBOL(dev_get_flags
);
4756 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4758 unsigned int old_flags
= dev
->flags
;
4764 * Set the flags on our device.
4767 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
4768 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
4770 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
4774 * Load in the correct multicast list now the flags have changed.
4777 if ((old_flags
^ flags
) & IFF_MULTICAST
)
4778 dev_change_rx_flags(dev
, IFF_MULTICAST
);
4780 dev_set_rx_mode(dev
);
4783 * Have we downed the interface. We handle IFF_UP ourselves
4784 * according to user attempts to set it, rather than blindly
4789 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
4790 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
4793 dev_set_rx_mode(dev
);
4796 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
4797 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
4799 dev
->gflags
^= IFF_PROMISC
;
4800 dev_set_promiscuity(dev
, inc
);
4803 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4804 is important. Some (broken) drivers set IFF_PROMISC, when
4805 IFF_ALLMULTI is requested not asking us and not reporting.
4807 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
4808 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
4810 dev
->gflags
^= IFF_ALLMULTI
;
4811 dev_set_allmulti(dev
, inc
);
4817 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
)
4819 unsigned int changes
= dev
->flags
^ old_flags
;
4821 if (changes
& IFF_UP
) {
4822 if (dev
->flags
& IFF_UP
)
4823 call_netdevice_notifiers(NETDEV_UP
, dev
);
4825 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
4828 if (dev
->flags
& IFF_UP
&&
4829 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
)))
4830 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
4834 * dev_change_flags - change device settings
4836 * @flags: device state flags
4838 * Change settings on device based state flags. The flags are
4839 * in the userspace exported format.
4841 int dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4844 unsigned int changes
, old_flags
= dev
->flags
;
4846 ret
= __dev_change_flags(dev
, flags
);
4850 changes
= old_flags
^ dev
->flags
;
4852 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
4854 __dev_notify_flags(dev
, old_flags
);
4857 EXPORT_SYMBOL(dev_change_flags
);
4860 * dev_set_mtu - Change maximum transfer unit
4862 * @new_mtu: new transfer unit
4864 * Change the maximum transfer size of the network device.
4866 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
4868 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4871 if (new_mtu
== dev
->mtu
)
4874 /* MTU must be positive. */
4878 if (!netif_device_present(dev
))
4882 if (ops
->ndo_change_mtu
)
4883 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
4887 if (!err
&& dev
->flags
& IFF_UP
)
4888 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
4891 EXPORT_SYMBOL(dev_set_mtu
);
4894 * dev_set_group - Change group this device belongs to
4896 * @new_group: group this device should belong to
4898 void dev_set_group(struct net_device
*dev
, int new_group
)
4900 dev
->group
= new_group
;
4902 EXPORT_SYMBOL(dev_set_group
);
4905 * dev_set_mac_address - Change Media Access Control Address
4909 * Change the hardware (MAC) address of the device
4911 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
4913 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4916 if (!ops
->ndo_set_mac_address
)
4918 if (sa
->sa_family
!= dev
->type
)
4920 if (!netif_device_present(dev
))
4922 err
= ops
->ndo_set_mac_address(dev
, sa
);
4924 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4925 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
4928 EXPORT_SYMBOL(dev_set_mac_address
);
4931 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4933 static int dev_ifsioc_locked(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4936 struct net_device
*dev
= dev_get_by_name_rcu(net
, ifr
->ifr_name
);
4942 case SIOCGIFFLAGS
: /* Get interface flags */
4943 ifr
->ifr_flags
= (short) dev_get_flags(dev
);
4946 case SIOCGIFMETRIC
: /* Get the metric on the interface
4947 (currently unused) */
4948 ifr
->ifr_metric
= 0;
4951 case SIOCGIFMTU
: /* Get the MTU of a device */
4952 ifr
->ifr_mtu
= dev
->mtu
;
4957 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
4959 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
4960 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4961 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
4969 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
4970 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
4971 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
4972 ifr
->ifr_map
.irq
= dev
->irq
;
4973 ifr
->ifr_map
.dma
= dev
->dma
;
4974 ifr
->ifr_map
.port
= dev
->if_port
;
4978 ifr
->ifr_ifindex
= dev
->ifindex
;
4982 ifr
->ifr_qlen
= dev
->tx_queue_len
;
4986 /* dev_ioctl() should ensure this case
4998 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
5000 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
5003 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
5004 const struct net_device_ops
*ops
;
5009 ops
= dev
->netdev_ops
;
5012 case SIOCSIFFLAGS
: /* Set interface flags */
5013 return dev_change_flags(dev
, ifr
->ifr_flags
);
5015 case SIOCSIFMETRIC
: /* Set the metric on the interface
5016 (currently unused) */
5019 case SIOCSIFMTU
: /* Set the MTU of a device */
5020 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
5023 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
5025 case SIOCSIFHWBROADCAST
:
5026 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
5028 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
5029 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
5030 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
5034 if (ops
->ndo_set_config
) {
5035 if (!netif_device_present(dev
))
5037 return ops
->ndo_set_config(dev
, &ifr
->ifr_map
);
5042 if (!ops
->ndo_set_rx_mode
||
5043 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
5045 if (!netif_device_present(dev
))
5047 return dev_mc_add_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
5050 if (!ops
->ndo_set_rx_mode
||
5051 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
5053 if (!netif_device_present(dev
))
5055 return dev_mc_del_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
5058 if (ifr
->ifr_qlen
< 0)
5060 dev
->tx_queue_len
= ifr
->ifr_qlen
;
5064 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
5065 return dev_change_name(dev
, ifr
->ifr_newname
);
5068 err
= net_hwtstamp_validate(ifr
);
5074 * Unknown or private ioctl
5077 if ((cmd
>= SIOCDEVPRIVATE
&&
5078 cmd
<= SIOCDEVPRIVATE
+ 15) ||
5079 cmd
== SIOCBONDENSLAVE
||
5080 cmd
== SIOCBONDRELEASE
||
5081 cmd
== SIOCBONDSETHWADDR
||
5082 cmd
== SIOCBONDSLAVEINFOQUERY
||
5083 cmd
== SIOCBONDINFOQUERY
||
5084 cmd
== SIOCBONDCHANGEACTIVE
||
5085 cmd
== SIOCGMIIPHY
||
5086 cmd
== SIOCGMIIREG
||
5087 cmd
== SIOCSMIIREG
||
5088 cmd
== SIOCBRADDIF
||
5089 cmd
== SIOCBRDELIF
||
5090 cmd
== SIOCSHWTSTAMP
||
5091 cmd
== SIOCWANDEV
) {
5093 if (ops
->ndo_do_ioctl
) {
5094 if (netif_device_present(dev
))
5095 err
= ops
->ndo_do_ioctl(dev
, ifr
, cmd
);
5107 * This function handles all "interface"-type I/O control requests. The actual
5108 * 'doing' part of this is dev_ifsioc above.
5112 * dev_ioctl - network device ioctl
5113 * @net: the applicable net namespace
5114 * @cmd: command to issue
5115 * @arg: pointer to a struct ifreq in user space
5117 * Issue ioctl functions to devices. This is normally called by the
5118 * user space syscall interfaces but can sometimes be useful for
5119 * other purposes. The return value is the return from the syscall if
5120 * positive or a negative errno code on error.
5123 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
5129 /* One special case: SIOCGIFCONF takes ifconf argument
5130 and requires shared lock, because it sleeps writing
5134 if (cmd
== SIOCGIFCONF
) {
5136 ret
= dev_ifconf(net
, (char __user
*) arg
);
5140 if (cmd
== SIOCGIFNAME
)
5141 return dev_ifname(net
, (struct ifreq __user
*)arg
);
5143 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
5146 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
5148 colon
= strchr(ifr
.ifr_name
, ':');
5153 * See which interface the caller is talking about.
5158 * These ioctl calls:
5159 * - can be done by all.
5160 * - atomic and do not require locking.
5171 dev_load(net
, ifr
.ifr_name
);
5173 ret
= dev_ifsioc_locked(net
, &ifr
, cmd
);
5178 if (copy_to_user(arg
, &ifr
,
5179 sizeof(struct ifreq
)))
5185 dev_load(net
, ifr
.ifr_name
);
5187 ret
= dev_ethtool(net
, &ifr
);
5192 if (copy_to_user(arg
, &ifr
,
5193 sizeof(struct ifreq
)))
5199 * These ioctl calls:
5200 * - require superuser power.
5201 * - require strict serialization.
5207 if (!capable(CAP_NET_ADMIN
))
5209 dev_load(net
, ifr
.ifr_name
);
5211 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5216 if (copy_to_user(arg
, &ifr
,
5217 sizeof(struct ifreq
)))
5223 * These ioctl calls:
5224 * - require superuser power.
5225 * - require strict serialization.
5226 * - do not return a value
5236 case SIOCSIFHWBROADCAST
:
5239 case SIOCBONDENSLAVE
:
5240 case SIOCBONDRELEASE
:
5241 case SIOCBONDSETHWADDR
:
5242 case SIOCBONDCHANGEACTIVE
:
5246 if (!capable(CAP_NET_ADMIN
))
5249 case SIOCBONDSLAVEINFOQUERY
:
5250 case SIOCBONDINFOQUERY
:
5251 dev_load(net
, ifr
.ifr_name
);
5253 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5258 /* Get the per device memory space. We can add this but
5259 * currently do not support it */
5261 /* Set the per device memory buffer space.
5262 * Not applicable in our case */
5267 * Unknown or private ioctl.
5270 if (cmd
== SIOCWANDEV
||
5271 (cmd
>= SIOCDEVPRIVATE
&&
5272 cmd
<= SIOCDEVPRIVATE
+ 15)) {
5273 dev_load(net
, ifr
.ifr_name
);
5275 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5277 if (!ret
&& copy_to_user(arg
, &ifr
,
5278 sizeof(struct ifreq
)))
5282 /* Take care of Wireless Extensions */
5283 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
5284 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
5291 * dev_new_index - allocate an ifindex
5292 * @net: the applicable net namespace
5294 * Returns a suitable unique value for a new device interface
5295 * number. The caller must hold the rtnl semaphore or the
5296 * dev_base_lock to be sure it remains unique.
5298 static int dev_new_index(struct net
*net
)
5300 int ifindex
= net
->ifindex
;
5304 if (!__dev_get_by_index(net
, ifindex
))
5305 return net
->ifindex
= ifindex
;
5309 /* Delayed registration/unregisteration */
5310 static LIST_HEAD(net_todo_list
);
5312 static void net_set_todo(struct net_device
*dev
)
5314 list_add_tail(&dev
->todo_list
, &net_todo_list
);
5317 static void rollback_registered_many(struct list_head
*head
)
5319 struct net_device
*dev
, *tmp
;
5321 BUG_ON(dev_boot_phase
);
5324 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
5325 /* Some devices call without registering
5326 * for initialization unwind. Remove those
5327 * devices and proceed with the remaining.
5329 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5330 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5334 list_del(&dev
->unreg_list
);
5337 dev
->dismantle
= true;
5338 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
5341 /* If device is running, close it first. */
5342 dev_close_many(head
);
5344 list_for_each_entry(dev
, head
, unreg_list
) {
5345 /* And unlink it from device chain. */
5346 unlist_netdevice(dev
);
5348 dev
->reg_state
= NETREG_UNREGISTERING
;
5353 list_for_each_entry(dev
, head
, unreg_list
) {
5354 /* Shutdown queueing discipline. */
5358 /* Notify protocols, that we are about to destroy
5359 this device. They should clean all the things.
5361 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5363 if (!dev
->rtnl_link_ops
||
5364 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5365 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
5368 * Flush the unicast and multicast chains
5373 if (dev
->netdev_ops
->ndo_uninit
)
5374 dev
->netdev_ops
->ndo_uninit(dev
);
5376 /* Notifier chain MUST detach us from master device. */
5377 WARN_ON(dev
->master
);
5379 /* Remove entries from kobject tree */
5380 netdev_unregister_kobject(dev
);
5385 list_for_each_entry(dev
, head
, unreg_list
)
5389 static void rollback_registered(struct net_device
*dev
)
5393 list_add(&dev
->unreg_list
, &single
);
5394 rollback_registered_many(&single
);
5398 static netdev_features_t
netdev_fix_features(struct net_device
*dev
,
5399 netdev_features_t features
)
5401 /* Fix illegal checksum combinations */
5402 if ((features
& NETIF_F_HW_CSUM
) &&
5403 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5404 netdev_warn(dev
, "mixed HW and IP checksum settings.\n");
5405 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5408 /* Fix illegal SG+CSUM combinations. */
5409 if ((features
& NETIF_F_SG
) &&
5410 !(features
& NETIF_F_ALL_CSUM
)) {
5412 "Dropping NETIF_F_SG since no checksum feature.\n");
5413 features
&= ~NETIF_F_SG
;
5416 /* TSO requires that SG is present as well. */
5417 if ((features
& NETIF_F_ALL_TSO
) && !(features
& NETIF_F_SG
)) {
5418 netdev_dbg(dev
, "Dropping TSO features since no SG feature.\n");
5419 features
&= ~NETIF_F_ALL_TSO
;
5422 /* TSO ECN requires that TSO is present as well. */
5423 if ((features
& NETIF_F_ALL_TSO
) == NETIF_F_TSO_ECN
)
5424 features
&= ~NETIF_F_TSO_ECN
;
5426 /* Software GSO depends on SG. */
5427 if ((features
& NETIF_F_GSO
) && !(features
& NETIF_F_SG
)) {
5428 netdev_dbg(dev
, "Dropping NETIF_F_GSO since no SG feature.\n");
5429 features
&= ~NETIF_F_GSO
;
5432 /* UFO needs SG and checksumming */
5433 if (features
& NETIF_F_UFO
) {
5434 /* maybe split UFO into V4 and V6? */
5435 if (!((features
& NETIF_F_GEN_CSUM
) ||
5436 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))
5437 == (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5439 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5440 features
&= ~NETIF_F_UFO
;
5443 if (!(features
& NETIF_F_SG
)) {
5445 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5446 features
&= ~NETIF_F_UFO
;
5453 int __netdev_update_features(struct net_device
*dev
)
5455 netdev_features_t features
;
5460 features
= netdev_get_wanted_features(dev
);
5462 if (dev
->netdev_ops
->ndo_fix_features
)
5463 features
= dev
->netdev_ops
->ndo_fix_features(dev
, features
);
5465 /* driver might be less strict about feature dependencies */
5466 features
= netdev_fix_features(dev
, features
);
5468 if (dev
->features
== features
)
5471 netdev_dbg(dev
, "Features changed: %pNF -> %pNF\n",
5472 &dev
->features
, &features
);
5474 if (dev
->netdev_ops
->ndo_set_features
)
5475 err
= dev
->netdev_ops
->ndo_set_features(dev
, features
);
5477 if (unlikely(err
< 0)) {
5479 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5480 err
, &features
, &dev
->features
);
5485 dev
->features
= features
;
5491 * netdev_update_features - recalculate device features
5492 * @dev: the device to check
5494 * Recalculate dev->features set and send notifications if it
5495 * has changed. Should be called after driver or hardware dependent
5496 * conditions might have changed that influence the features.
5498 void netdev_update_features(struct net_device
*dev
)
5500 if (__netdev_update_features(dev
))
5501 netdev_features_change(dev
);
5503 EXPORT_SYMBOL(netdev_update_features
);
5506 * netdev_change_features - recalculate device features
5507 * @dev: the device to check
5509 * Recalculate dev->features set and send notifications even
5510 * if they have not changed. Should be called instead of
5511 * netdev_update_features() if also dev->vlan_features might
5512 * have changed to allow the changes to be propagated to stacked
5515 void netdev_change_features(struct net_device
*dev
)
5517 __netdev_update_features(dev
);
5518 netdev_features_change(dev
);
5520 EXPORT_SYMBOL(netdev_change_features
);
5523 * netif_stacked_transfer_operstate - transfer operstate
5524 * @rootdev: the root or lower level device to transfer state from
5525 * @dev: the device to transfer operstate to
5527 * Transfer operational state from root to device. This is normally
5528 * called when a stacking relationship exists between the root
5529 * device and the device(a leaf device).
5531 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
5532 struct net_device
*dev
)
5534 if (rootdev
->operstate
== IF_OPER_DORMANT
)
5535 netif_dormant_on(dev
);
5537 netif_dormant_off(dev
);
5539 if (netif_carrier_ok(rootdev
)) {
5540 if (!netif_carrier_ok(dev
))
5541 netif_carrier_on(dev
);
5543 if (netif_carrier_ok(dev
))
5544 netif_carrier_off(dev
);
5547 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
5550 static int netif_alloc_rx_queues(struct net_device
*dev
)
5552 unsigned int i
, count
= dev
->num_rx_queues
;
5553 struct netdev_rx_queue
*rx
;
5557 rx
= kcalloc(count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5559 pr_err("netdev: Unable to allocate %u rx queues\n", count
);
5564 for (i
= 0; i
< count
; i
++)
5570 static void netdev_init_one_queue(struct net_device
*dev
,
5571 struct netdev_queue
*queue
, void *_unused
)
5573 /* Initialize queue lock */
5574 spin_lock_init(&queue
->_xmit_lock
);
5575 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
5576 queue
->xmit_lock_owner
= -1;
5577 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
5580 dql_init(&queue
->dql
, HZ
);
5584 static int netif_alloc_netdev_queues(struct net_device
*dev
)
5586 unsigned int count
= dev
->num_tx_queues
;
5587 struct netdev_queue
*tx
;
5591 tx
= kcalloc(count
, sizeof(struct netdev_queue
), GFP_KERNEL
);
5593 pr_err("netdev: Unable to allocate %u tx queues\n", count
);
5598 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5599 spin_lock_init(&dev
->tx_global_lock
);
5605 * register_netdevice - register a network device
5606 * @dev: device to register
5608 * Take a completed network device structure and add it to the kernel
5609 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5610 * chain. 0 is returned on success. A negative errno code is returned
5611 * on a failure to set up the device, or if the name is a duplicate.
5613 * Callers must hold the rtnl semaphore. You may want
5614 * register_netdev() instead of this.
5617 * The locking appears insufficient to guarantee two parallel registers
5618 * will not get the same name.
5621 int register_netdevice(struct net_device
*dev
)
5624 struct net
*net
= dev_net(dev
);
5626 BUG_ON(dev_boot_phase
);
5631 /* When net_device's are persistent, this will be fatal. */
5632 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
5635 spin_lock_init(&dev
->addr_list_lock
);
5636 netdev_set_addr_lockdep_class(dev
);
5640 ret
= dev_get_valid_name(net
, dev
, dev
->name
);
5644 /* Init, if this function is available */
5645 if (dev
->netdev_ops
->ndo_init
) {
5646 ret
= dev
->netdev_ops
->ndo_init(dev
);
5656 dev
->ifindex
= dev_new_index(net
);
5657 else if (__dev_get_by_index(net
, dev
->ifindex
))
5660 if (dev
->iflink
== -1)
5661 dev
->iflink
= dev
->ifindex
;
5663 /* Transfer changeable features to wanted_features and enable
5664 * software offloads (GSO and GRO).
5666 dev
->hw_features
|= NETIF_F_SOFT_FEATURES
;
5667 dev
->features
|= NETIF_F_SOFT_FEATURES
;
5668 dev
->wanted_features
= dev
->features
& dev
->hw_features
;
5670 /* Turn on no cache copy if HW is doing checksum */
5671 if (!(dev
->flags
& IFF_LOOPBACK
)) {
5672 dev
->hw_features
|= NETIF_F_NOCACHE_COPY
;
5673 if (dev
->features
& NETIF_F_ALL_CSUM
) {
5674 dev
->wanted_features
|= NETIF_F_NOCACHE_COPY
;
5675 dev
->features
|= NETIF_F_NOCACHE_COPY
;
5679 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5681 dev
->vlan_features
|= NETIF_F_HIGHDMA
;
5683 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
5684 ret
= notifier_to_errno(ret
);
5688 ret
= netdev_register_kobject(dev
);
5691 dev
->reg_state
= NETREG_REGISTERED
;
5693 __netdev_update_features(dev
);
5696 * Default initial state at registry is that the
5697 * device is present.
5700 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5702 linkwatch_init_dev(dev
);
5704 dev_init_scheduler(dev
);
5706 list_netdevice(dev
);
5707 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
5709 /* Notify protocols, that a new device appeared. */
5710 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5711 ret
= notifier_to_errno(ret
);
5713 rollback_registered(dev
);
5714 dev
->reg_state
= NETREG_UNREGISTERED
;
5717 * Prevent userspace races by waiting until the network
5718 * device is fully setup before sending notifications.
5720 if (!dev
->rtnl_link_ops
||
5721 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5722 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5728 if (dev
->netdev_ops
->ndo_uninit
)
5729 dev
->netdev_ops
->ndo_uninit(dev
);
5732 EXPORT_SYMBOL(register_netdevice
);
5735 * init_dummy_netdev - init a dummy network device for NAPI
5736 * @dev: device to init
5738 * This takes a network device structure and initialize the minimum
5739 * amount of fields so it can be used to schedule NAPI polls without
5740 * registering a full blown interface. This is to be used by drivers
5741 * that need to tie several hardware interfaces to a single NAPI
5742 * poll scheduler due to HW limitations.
5744 int init_dummy_netdev(struct net_device
*dev
)
5746 /* Clear everything. Note we don't initialize spinlocks
5747 * are they aren't supposed to be taken by any of the
5748 * NAPI code and this dummy netdev is supposed to be
5749 * only ever used for NAPI polls
5751 memset(dev
, 0, sizeof(struct net_device
));
5753 /* make sure we BUG if trying to hit standard
5754 * register/unregister code path
5756 dev
->reg_state
= NETREG_DUMMY
;
5758 /* NAPI wants this */
5759 INIT_LIST_HEAD(&dev
->napi_list
);
5761 /* a dummy interface is started by default */
5762 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5763 set_bit(__LINK_STATE_START
, &dev
->state
);
5765 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5766 * because users of this 'device' dont need to change
5772 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
5776 * register_netdev - register a network device
5777 * @dev: device to register
5779 * Take a completed network device structure and add it to the kernel
5780 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5781 * chain. 0 is returned on success. A negative errno code is returned
5782 * on a failure to set up the device, or if the name is a duplicate.
5784 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5785 * and expands the device name if you passed a format string to
5788 int register_netdev(struct net_device
*dev
)
5793 err
= register_netdevice(dev
);
5797 EXPORT_SYMBOL(register_netdev
);
5799 int netdev_refcnt_read(const struct net_device
*dev
)
5803 for_each_possible_cpu(i
)
5804 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
5807 EXPORT_SYMBOL(netdev_refcnt_read
);
5810 * netdev_wait_allrefs - wait until all references are gone.
5811 * @dev: target net_device
5813 * This is called when unregistering network devices.
5815 * Any protocol or device that holds a reference should register
5816 * for netdevice notification, and cleanup and put back the
5817 * reference if they receive an UNREGISTER event.
5818 * We can get stuck here if buggy protocols don't correctly
5821 static void netdev_wait_allrefs(struct net_device
*dev
)
5823 unsigned long rebroadcast_time
, warning_time
;
5826 linkwatch_forget_dev(dev
);
5828 rebroadcast_time
= warning_time
= jiffies
;
5829 refcnt
= netdev_refcnt_read(dev
);
5831 while (refcnt
!= 0) {
5832 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
5835 /* Rebroadcast unregister notification */
5836 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5842 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
5843 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
5845 /* We must not have linkwatch events
5846 * pending on unregister. If this
5847 * happens, we simply run the queue
5848 * unscheduled, resulting in a noop
5851 linkwatch_run_queue();
5856 rebroadcast_time
= jiffies
;
5861 refcnt
= netdev_refcnt_read(dev
);
5863 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
5864 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5866 warning_time
= jiffies
;
5875 * register_netdevice(x1);
5876 * register_netdevice(x2);
5878 * unregister_netdevice(y1);
5879 * unregister_netdevice(y2);
5885 * We are invoked by rtnl_unlock().
5886 * This allows us to deal with problems:
5887 * 1) We can delete sysfs objects which invoke hotplug
5888 * without deadlocking with linkwatch via keventd.
5889 * 2) Since we run with the RTNL semaphore not held, we can sleep
5890 * safely in order to wait for the netdev refcnt to drop to zero.
5892 * We must not return until all unregister events added during
5893 * the interval the lock was held have been completed.
5895 void netdev_run_todo(void)
5897 struct list_head list
;
5899 /* Snapshot list, allow later requests */
5900 list_replace_init(&net_todo_list
, &list
);
5905 /* Wait for rcu callbacks to finish before next phase */
5906 if (!list_empty(&list
))
5909 while (!list_empty(&list
)) {
5910 struct net_device
*dev
5911 = list_first_entry(&list
, struct net_device
, todo_list
);
5912 list_del(&dev
->todo_list
);
5915 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
5918 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
5919 pr_err("network todo '%s' but state %d\n",
5920 dev
->name
, dev
->reg_state
);
5925 dev
->reg_state
= NETREG_UNREGISTERED
;
5927 on_each_cpu(flush_backlog
, dev
, 1);
5929 netdev_wait_allrefs(dev
);
5932 BUG_ON(netdev_refcnt_read(dev
));
5933 WARN_ON(rcu_access_pointer(dev
->ip_ptr
));
5934 WARN_ON(rcu_access_pointer(dev
->ip6_ptr
));
5935 WARN_ON(dev
->dn_ptr
);
5937 if (dev
->destructor
)
5938 dev
->destructor(dev
);
5940 /* Free network device */
5941 kobject_put(&dev
->dev
.kobj
);
5945 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5946 * fields in the same order, with only the type differing.
5948 void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
5949 const struct net_device_stats
*netdev_stats
)
5951 #if BITS_PER_LONG == 64
5952 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
5953 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
5955 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
5956 const unsigned long *src
= (const unsigned long *)netdev_stats
;
5957 u64
*dst
= (u64
*)stats64
;
5959 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
5960 sizeof(*stats64
) / sizeof(u64
));
5961 for (i
= 0; i
< n
; i
++)
5965 EXPORT_SYMBOL(netdev_stats_to_stats64
);
5968 * dev_get_stats - get network device statistics
5969 * @dev: device to get statistics from
5970 * @storage: place to store stats
5972 * Get network statistics from device. Return @storage.
5973 * The device driver may provide its own method by setting
5974 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5975 * otherwise the internal statistics structure is used.
5977 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
5978 struct rtnl_link_stats64
*storage
)
5980 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5982 if (ops
->ndo_get_stats64
) {
5983 memset(storage
, 0, sizeof(*storage
));
5984 ops
->ndo_get_stats64(dev
, storage
);
5985 } else if (ops
->ndo_get_stats
) {
5986 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
5988 netdev_stats_to_stats64(storage
, &dev
->stats
);
5990 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
5993 EXPORT_SYMBOL(dev_get_stats
);
5995 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
5997 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
5999 #ifdef CONFIG_NET_CLS_ACT
6002 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
6005 netdev_init_one_queue(dev
, queue
, NULL
);
6006 queue
->qdisc
= &noop_qdisc
;
6007 queue
->qdisc_sleeping
= &noop_qdisc
;
6008 rcu_assign_pointer(dev
->ingress_queue
, queue
);
6013 static const struct ethtool_ops default_ethtool_ops
;
6016 * alloc_netdev_mqs - allocate network device
6017 * @sizeof_priv: size of private data to allocate space for
6018 * @name: device name format string
6019 * @setup: callback to initialize device
6020 * @txqs: the number of TX subqueues to allocate
6021 * @rxqs: the number of RX subqueues to allocate
6023 * Allocates a struct net_device with private data area for driver use
6024 * and performs basic initialization. Also allocates subquue structs
6025 * for each queue on the device.
6027 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
6028 void (*setup
)(struct net_device
*),
6029 unsigned int txqs
, unsigned int rxqs
)
6031 struct net_device
*dev
;
6033 struct net_device
*p
;
6035 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
6038 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6044 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6049 alloc_size
= sizeof(struct net_device
);
6051 /* ensure 32-byte alignment of private area */
6052 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
6053 alloc_size
+= sizeof_priv
;
6055 /* ensure 32-byte alignment of whole construct */
6056 alloc_size
+= NETDEV_ALIGN
- 1;
6058 p
= kzalloc(alloc_size
, GFP_KERNEL
);
6060 pr_err("alloc_netdev: Unable to allocate device\n");
6064 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
6065 dev
->padded
= (char *)dev
- (char *)p
;
6067 dev
->pcpu_refcnt
= alloc_percpu(int);
6068 if (!dev
->pcpu_refcnt
)
6071 if (dev_addr_init(dev
))
6077 dev_net_set(dev
, &init_net
);
6079 dev
->gso_max_size
= GSO_MAX_SIZE
;
6080 dev
->gso_max_segs
= GSO_MAX_SEGS
;
6082 INIT_LIST_HEAD(&dev
->napi_list
);
6083 INIT_LIST_HEAD(&dev
->unreg_list
);
6084 INIT_LIST_HEAD(&dev
->link_watch_list
);
6085 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
6088 dev
->num_tx_queues
= txqs
;
6089 dev
->real_num_tx_queues
= txqs
;
6090 if (netif_alloc_netdev_queues(dev
))
6094 dev
->num_rx_queues
= rxqs
;
6095 dev
->real_num_rx_queues
= rxqs
;
6096 if (netif_alloc_rx_queues(dev
))
6100 strcpy(dev
->name
, name
);
6101 dev
->group
= INIT_NETDEV_GROUP
;
6102 if (!dev
->ethtool_ops
)
6103 dev
->ethtool_ops
= &default_ethtool_ops
;
6111 free_percpu(dev
->pcpu_refcnt
);
6121 EXPORT_SYMBOL(alloc_netdev_mqs
);
6124 * free_netdev - free network device
6127 * This function does the last stage of destroying an allocated device
6128 * interface. The reference to the device object is released.
6129 * If this is the last reference then it will be freed.
6131 void free_netdev(struct net_device
*dev
)
6133 struct napi_struct
*p
, *n
;
6135 release_net(dev_net(dev
));
6142 kfree(rcu_dereference_protected(dev
->ingress_queue
, 1));
6144 /* Flush device addresses */
6145 dev_addr_flush(dev
);
6147 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
6150 free_percpu(dev
->pcpu_refcnt
);
6151 dev
->pcpu_refcnt
= NULL
;
6153 /* Compatibility with error handling in drivers */
6154 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
6155 kfree((char *)dev
- dev
->padded
);
6159 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
6160 dev
->reg_state
= NETREG_RELEASED
;
6162 /* will free via device release */
6163 put_device(&dev
->dev
);
6165 EXPORT_SYMBOL(free_netdev
);
6168 * synchronize_net - Synchronize with packet receive processing
6170 * Wait for packets currently being received to be done.
6171 * Does not block later packets from starting.
6173 void synchronize_net(void)
6176 if (rtnl_is_locked())
6177 synchronize_rcu_expedited();
6181 EXPORT_SYMBOL(synchronize_net
);
6184 * unregister_netdevice_queue - remove device from the kernel
6188 * This function shuts down a device interface and removes it
6189 * from the kernel tables.
6190 * If head not NULL, device is queued to be unregistered later.
6192 * Callers must hold the rtnl semaphore. You may want
6193 * unregister_netdev() instead of this.
6196 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
6201 list_move_tail(&dev
->unreg_list
, head
);
6203 rollback_registered(dev
);
6204 /* Finish processing unregister after unlock */
6208 EXPORT_SYMBOL(unregister_netdevice_queue
);
6211 * unregister_netdevice_many - unregister many devices
6212 * @head: list of devices
6214 void unregister_netdevice_many(struct list_head
*head
)
6216 struct net_device
*dev
;
6218 if (!list_empty(head
)) {
6219 rollback_registered_many(head
);
6220 list_for_each_entry(dev
, head
, unreg_list
)
6224 EXPORT_SYMBOL(unregister_netdevice_many
);
6227 * unregister_netdev - remove device from the kernel
6230 * This function shuts down a device interface and removes it
6231 * from the kernel tables.
6233 * This is just a wrapper for unregister_netdevice that takes
6234 * the rtnl semaphore. In general you want to use this and not
6235 * unregister_netdevice.
6237 void unregister_netdev(struct net_device
*dev
)
6240 unregister_netdevice(dev
);
6243 EXPORT_SYMBOL(unregister_netdev
);
6246 * dev_change_net_namespace - move device to different nethost namespace
6248 * @net: network namespace
6249 * @pat: If not NULL name pattern to try if the current device name
6250 * is already taken in the destination network namespace.
6252 * This function shuts down a device interface and moves it
6253 * to a new network namespace. On success 0 is returned, on
6254 * a failure a netagive errno code is returned.
6256 * Callers must hold the rtnl semaphore.
6259 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
6265 /* Don't allow namespace local devices to be moved. */
6267 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6270 /* Ensure the device has been registrered */
6272 if (dev
->reg_state
!= NETREG_REGISTERED
)
6275 /* Get out if there is nothing todo */
6277 if (net_eq(dev_net(dev
), net
))
6280 /* Pick the destination device name, and ensure
6281 * we can use it in the destination network namespace.
6284 if (__dev_get_by_name(net
, dev
->name
)) {
6285 /* We get here if we can't use the current device name */
6288 if (dev_get_valid_name(net
, dev
, pat
) < 0)
6293 * And now a mini version of register_netdevice unregister_netdevice.
6296 /* If device is running close it first. */
6299 /* And unlink it from device chain */
6301 unlist_netdevice(dev
);
6305 /* Shutdown queueing discipline. */
6308 /* Notify protocols, that we are about to destroy
6309 this device. They should clean all the things.
6311 Note that dev->reg_state stays at NETREG_REGISTERED.
6312 This is wanted because this way 8021q and macvlan know
6313 the device is just moving and can keep their slaves up.
6315 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6317 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
6318 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
6321 * Flush the unicast and multicast chains
6326 /* Actually switch the network namespace */
6327 dev_net_set(dev
, net
);
6329 /* If there is an ifindex conflict assign a new one */
6330 if (__dev_get_by_index(net
, dev
->ifindex
)) {
6331 int iflink
= (dev
->iflink
== dev
->ifindex
);
6332 dev
->ifindex
= dev_new_index(net
);
6334 dev
->iflink
= dev
->ifindex
;
6337 /* Fixup kobjects */
6338 err
= device_rename(&dev
->dev
, dev
->name
);
6341 /* Add the device back in the hashes */
6342 list_netdevice(dev
);
6344 /* Notify protocols, that a new device appeared. */
6345 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
6348 * Prevent userspace races by waiting until the network
6349 * device is fully setup before sending notifications.
6351 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
6358 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
6360 static int dev_cpu_callback(struct notifier_block
*nfb
,
6361 unsigned long action
,
6364 struct sk_buff
**list_skb
;
6365 struct sk_buff
*skb
;
6366 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
6367 struct softnet_data
*sd
, *oldsd
;
6369 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
6372 local_irq_disable();
6373 cpu
= smp_processor_id();
6374 sd
= &per_cpu(softnet_data
, cpu
);
6375 oldsd
= &per_cpu(softnet_data
, oldcpu
);
6377 /* Find end of our completion_queue. */
6378 list_skb
= &sd
->completion_queue
;
6380 list_skb
= &(*list_skb
)->next
;
6381 /* Append completion queue from offline CPU. */
6382 *list_skb
= oldsd
->completion_queue
;
6383 oldsd
->completion_queue
= NULL
;
6385 /* Append output queue from offline CPU. */
6386 if (oldsd
->output_queue
) {
6387 *sd
->output_queue_tailp
= oldsd
->output_queue
;
6388 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
6389 oldsd
->output_queue
= NULL
;
6390 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
6392 /* Append NAPI poll list from offline CPU. */
6393 if (!list_empty(&oldsd
->poll_list
)) {
6394 list_splice_init(&oldsd
->poll_list
, &sd
->poll_list
);
6395 raise_softirq_irqoff(NET_RX_SOFTIRQ
);
6398 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
6401 /* Process offline CPU's input_pkt_queue */
6402 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
6404 input_queue_head_incr(oldsd
);
6406 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
6408 input_queue_head_incr(oldsd
);
6416 * netdev_increment_features - increment feature set by one
6417 * @all: current feature set
6418 * @one: new feature set
6419 * @mask: mask feature set
6421 * Computes a new feature set after adding a device with feature set
6422 * @one to the master device with current feature set @all. Will not
6423 * enable anything that is off in @mask. Returns the new feature set.
6425 netdev_features_t
netdev_increment_features(netdev_features_t all
,
6426 netdev_features_t one
, netdev_features_t mask
)
6428 if (mask
& NETIF_F_GEN_CSUM
)
6429 mask
|= NETIF_F_ALL_CSUM
;
6430 mask
|= NETIF_F_VLAN_CHALLENGED
;
6432 all
|= one
& (NETIF_F_ONE_FOR_ALL
|NETIF_F_ALL_CSUM
) & mask
;
6433 all
&= one
| ~NETIF_F_ALL_FOR_ALL
;
6435 /* If one device supports hw checksumming, set for all. */
6436 if (all
& NETIF_F_GEN_CSUM
)
6437 all
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_GEN_CSUM
);
6441 EXPORT_SYMBOL(netdev_increment_features
);
6443 static struct hlist_head
*netdev_create_hash(void)
6446 struct hlist_head
*hash
;
6448 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
6450 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
6451 INIT_HLIST_HEAD(&hash
[i
]);
6456 /* Initialize per network namespace state */
6457 static int __net_init
netdev_init(struct net
*net
)
6459 if (net
!= &init_net
)
6460 INIT_LIST_HEAD(&net
->dev_base_head
);
6462 net
->dev_name_head
= netdev_create_hash();
6463 if (net
->dev_name_head
== NULL
)
6466 net
->dev_index_head
= netdev_create_hash();
6467 if (net
->dev_index_head
== NULL
)
6473 kfree(net
->dev_name_head
);
6479 * netdev_drivername - network driver for the device
6480 * @dev: network device
6482 * Determine network driver for device.
6484 const char *netdev_drivername(const struct net_device
*dev
)
6486 const struct device_driver
*driver
;
6487 const struct device
*parent
;
6488 const char *empty
= "";
6490 parent
= dev
->dev
.parent
;
6494 driver
= parent
->driver
;
6495 if (driver
&& driver
->name
)
6496 return driver
->name
;
6500 static int __netdev_printk(const char *level
, const struct net_device
*dev
,
6501 struct va_format
*vaf
)
6505 if (dev
&& dev
->dev
.parent
) {
6506 r
= dev_printk_emit(level
[1] - '0',
6509 dev_driver_string(dev
->dev
.parent
),
6510 dev_name(dev
->dev
.parent
),
6511 netdev_name(dev
), vaf
);
6513 r
= printk("%s%s: %pV", level
, netdev_name(dev
), vaf
);
6515 r
= printk("%s(NULL net_device): %pV", level
, vaf
);
6521 int netdev_printk(const char *level
, const struct net_device
*dev
,
6522 const char *format
, ...)
6524 struct va_format vaf
;
6528 va_start(args
, format
);
6533 r
= __netdev_printk(level
, dev
, &vaf
);
6539 EXPORT_SYMBOL(netdev_printk
);
6541 #define define_netdev_printk_level(func, level) \
6542 int func(const struct net_device *dev, const char *fmt, ...) \
6545 struct va_format vaf; \
6548 va_start(args, fmt); \
6553 r = __netdev_printk(level, dev, &vaf); \
6559 EXPORT_SYMBOL(func);
6561 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
6562 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
6563 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
6564 define_netdev_printk_level(netdev_err
, KERN_ERR
);
6565 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
6566 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
6567 define_netdev_printk_level(netdev_info
, KERN_INFO
);
6569 static void __net_exit
netdev_exit(struct net
*net
)
6571 kfree(net
->dev_name_head
);
6572 kfree(net
->dev_index_head
);
6575 static struct pernet_operations __net_initdata netdev_net_ops
= {
6576 .init
= netdev_init
,
6577 .exit
= netdev_exit
,
6580 static void __net_exit
default_device_exit(struct net
*net
)
6582 struct net_device
*dev
, *aux
;
6584 * Push all migratable network devices back to the
6585 * initial network namespace
6588 for_each_netdev_safe(net
, dev
, aux
) {
6590 char fb_name
[IFNAMSIZ
];
6592 /* Ignore unmoveable devices (i.e. loopback) */
6593 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6596 /* Leave virtual devices for the generic cleanup */
6597 if (dev
->rtnl_link_ops
)
6600 /* Push remaining network devices to init_net */
6601 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
6602 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
6604 pr_emerg("%s: failed to move %s to init_net: %d\n",
6605 __func__
, dev
->name
, err
);
6612 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
6614 /* At exit all network devices most be removed from a network
6615 * namespace. Do this in the reverse order of registration.
6616 * Do this across as many network namespaces as possible to
6617 * improve batching efficiency.
6619 struct net_device
*dev
;
6621 LIST_HEAD(dev_kill_list
);
6624 list_for_each_entry(net
, net_list
, exit_list
) {
6625 for_each_netdev_reverse(net
, dev
) {
6626 if (dev
->rtnl_link_ops
)
6627 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
6629 unregister_netdevice_queue(dev
, &dev_kill_list
);
6632 unregister_netdevice_many(&dev_kill_list
);
6633 list_del(&dev_kill_list
);
6637 static struct pernet_operations __net_initdata default_device_ops
= {
6638 .exit
= default_device_exit
,
6639 .exit_batch
= default_device_exit_batch
,
6643 * Initialize the DEV module. At boot time this walks the device list and
6644 * unhooks any devices that fail to initialise (normally hardware not
6645 * present) and leaves us with a valid list of present and active devices.
6650 * This is called single threaded during boot, so no need
6651 * to take the rtnl semaphore.
6653 static int __init
net_dev_init(void)
6655 int i
, rc
= -ENOMEM
;
6657 BUG_ON(!dev_boot_phase
);
6659 if (dev_proc_init())
6662 if (netdev_kobject_init())
6665 INIT_LIST_HEAD(&ptype_all
);
6666 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
6667 INIT_LIST_HEAD(&ptype_base
[i
]);
6669 if (register_pernet_subsys(&netdev_net_ops
))
6673 * Initialise the packet receive queues.
6676 for_each_possible_cpu(i
) {
6677 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
6679 memset(sd
, 0, sizeof(*sd
));
6680 skb_queue_head_init(&sd
->input_pkt_queue
);
6681 skb_queue_head_init(&sd
->process_queue
);
6682 sd
->completion_queue
= NULL
;
6683 INIT_LIST_HEAD(&sd
->poll_list
);
6684 sd
->output_queue
= NULL
;
6685 sd
->output_queue_tailp
= &sd
->output_queue
;
6687 sd
->csd
.func
= rps_trigger_softirq
;
6693 sd
->backlog
.poll
= process_backlog
;
6694 sd
->backlog
.weight
= weight_p
;
6695 sd
->backlog
.gro_list
= NULL
;
6696 sd
->backlog
.gro_count
= 0;
6701 /* The loopback device is special if any other network devices
6702 * is present in a network namespace the loopback device must
6703 * be present. Since we now dynamically allocate and free the
6704 * loopback device ensure this invariant is maintained by
6705 * keeping the loopback device as the first device on the
6706 * list of network devices. Ensuring the loopback devices
6707 * is the first device that appears and the last network device
6710 if (register_pernet_device(&loopback_net_ops
))
6713 if (register_pernet_device(&default_device_ops
))
6716 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
6717 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
6719 hotcpu_notifier(dev_cpu_callback
, 0);
6727 subsys_initcall(net_dev_init
);
6729 static int __init
initialize_hashrnd(void)
6731 get_random_bytes(&hashrnd
, sizeof(hashrnd
));
6735 late_initcall_sync(initialize_hashrnd
);