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_get_valid_name(struct net_device
*dev
, const char *name
)
966 BUG_ON(!dev_net(dev
));
969 if (!dev_valid_name(name
))
972 if (strchr(name
, '%'))
973 return dev_alloc_name(dev
, name
);
974 else if (__dev_get_by_name(net
, name
))
976 else if (dev
->name
!= name
)
977 strlcpy(dev
->name
, name
, IFNAMSIZ
);
983 * dev_change_name - change name of a device
985 * @newname: name (or format string) must be at least IFNAMSIZ
987 * Change name of a device, can pass format strings "eth%d".
990 int dev_change_name(struct net_device
*dev
, const char *newname
)
992 char oldname
[IFNAMSIZ
];
998 BUG_ON(!dev_net(dev
));
1001 if (dev
->flags
& IFF_UP
)
1004 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0)
1007 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
1009 err
= dev_get_valid_name(dev
, newname
);
1014 ret
= device_rename(&dev
->dev
, dev
->name
);
1016 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1020 write_lock_bh(&dev_base_lock
);
1021 hlist_del_rcu(&dev
->name_hlist
);
1022 write_unlock_bh(&dev_base_lock
);
1026 write_lock_bh(&dev_base_lock
);
1027 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
1028 write_unlock_bh(&dev_base_lock
);
1030 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
1031 ret
= notifier_to_errno(ret
);
1034 /* err >= 0 after dev_alloc_name() or stores the first errno */
1037 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1040 pr_err("%s: name change rollback failed: %d\n",
1049 * dev_set_alias - change ifalias of a device
1051 * @alias: name up to IFALIASZ
1052 * @len: limit of bytes to copy from info
1054 * Set ifalias for a device,
1056 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
1060 if (len
>= IFALIASZ
)
1065 kfree(dev
->ifalias
);
1066 dev
->ifalias
= NULL
;
1071 dev
->ifalias
= krealloc(dev
->ifalias
, len
+ 1, GFP_KERNEL
);
1075 strlcpy(dev
->ifalias
, alias
, len
+1);
1081 * netdev_features_change - device changes features
1082 * @dev: device to cause notification
1084 * Called to indicate a device has changed features.
1086 void netdev_features_change(struct net_device
*dev
)
1088 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1090 EXPORT_SYMBOL(netdev_features_change
);
1093 * netdev_state_change - device changes state
1094 * @dev: device to cause notification
1096 * Called to indicate a device has changed state. This function calls
1097 * the notifier chains for netdev_chain and sends a NEWLINK message
1098 * to the routing socket.
1100 void netdev_state_change(struct net_device
*dev
)
1102 if (dev
->flags
& IFF_UP
) {
1103 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
1104 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
1107 EXPORT_SYMBOL(netdev_state_change
);
1109 int netdev_bonding_change(struct net_device
*dev
, unsigned long event
)
1111 return call_netdevice_notifiers(event
, dev
);
1113 EXPORT_SYMBOL(netdev_bonding_change
);
1116 * dev_load - load a network module
1117 * @net: the applicable net namespace
1118 * @name: name of interface
1120 * If a network interface is not present and the process has suitable
1121 * privileges this function loads the module. If module loading is not
1122 * available in this kernel then it becomes a nop.
1125 void dev_load(struct net
*net
, const char *name
)
1127 struct net_device
*dev
;
1131 dev
= dev_get_by_name_rcu(net
, name
);
1135 if (no_module
&& capable(CAP_NET_ADMIN
))
1136 no_module
= request_module("netdev-%s", name
);
1137 if (no_module
&& capable(CAP_SYS_MODULE
)) {
1138 if (!request_module("%s", name
))
1139 pr_warn("Loading kernel module for a network device with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s instead.\n",
1143 EXPORT_SYMBOL(dev_load
);
1145 static int __dev_open(struct net_device
*dev
)
1147 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1152 if (!netif_device_present(dev
))
1155 ret
= call_netdevice_notifiers(NETDEV_PRE_UP
, dev
);
1156 ret
= notifier_to_errno(ret
);
1160 set_bit(__LINK_STATE_START
, &dev
->state
);
1162 if (ops
->ndo_validate_addr
)
1163 ret
= ops
->ndo_validate_addr(dev
);
1165 if (!ret
&& ops
->ndo_open
)
1166 ret
= ops
->ndo_open(dev
);
1169 clear_bit(__LINK_STATE_START
, &dev
->state
);
1171 dev
->flags
|= IFF_UP
;
1172 net_dmaengine_get();
1173 dev_set_rx_mode(dev
);
1175 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
1182 * dev_open - prepare an interface for use.
1183 * @dev: device to open
1185 * Takes a device from down to up state. The device's private open
1186 * function is invoked and then the multicast lists are loaded. Finally
1187 * the device is moved into the up state and a %NETDEV_UP message is
1188 * sent to the netdev notifier chain.
1190 * Calling this function on an active interface is a nop. On a failure
1191 * a negative errno code is returned.
1193 int dev_open(struct net_device
*dev
)
1197 if (dev
->flags
& IFF_UP
)
1200 ret
= __dev_open(dev
);
1204 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1205 call_netdevice_notifiers(NETDEV_UP
, dev
);
1209 EXPORT_SYMBOL(dev_open
);
1211 static int __dev_close_many(struct list_head
*head
)
1213 struct net_device
*dev
;
1218 list_for_each_entry(dev
, head
, unreg_list
) {
1219 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1221 clear_bit(__LINK_STATE_START
, &dev
->state
);
1223 /* Synchronize to scheduled poll. We cannot touch poll list, it
1224 * can be even on different cpu. So just clear netif_running().
1226 * dev->stop() will invoke napi_disable() on all of it's
1227 * napi_struct instances on this device.
1229 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1232 dev_deactivate_many(head
);
1234 list_for_each_entry(dev
, head
, unreg_list
) {
1235 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1238 * Call the device specific close. This cannot fail.
1239 * Only if device is UP
1241 * We allow it to be called even after a DETACH hot-plug
1247 dev
->flags
&= ~IFF_UP
;
1248 net_dmaengine_put();
1254 static int __dev_close(struct net_device
*dev
)
1259 list_add(&dev
->unreg_list
, &single
);
1260 retval
= __dev_close_many(&single
);
1265 static int dev_close_many(struct list_head
*head
)
1267 struct net_device
*dev
, *tmp
;
1268 LIST_HEAD(tmp_list
);
1270 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
)
1271 if (!(dev
->flags
& IFF_UP
))
1272 list_move(&dev
->unreg_list
, &tmp_list
);
1274 __dev_close_many(head
);
1276 list_for_each_entry(dev
, head
, unreg_list
) {
1277 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1278 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1281 /* rollback_registered_many needs the complete original list */
1282 list_splice(&tmp_list
, head
);
1287 * dev_close - shutdown an interface.
1288 * @dev: device to shutdown
1290 * This function moves an active device into down state. A
1291 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1292 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1295 int dev_close(struct net_device
*dev
)
1297 if (dev
->flags
& IFF_UP
) {
1300 list_add(&dev
->unreg_list
, &single
);
1301 dev_close_many(&single
);
1306 EXPORT_SYMBOL(dev_close
);
1310 * dev_disable_lro - disable Large Receive Offload on a device
1313 * Disable Large Receive Offload (LRO) on a net device. Must be
1314 * called under RTNL. This is needed if received packets may be
1315 * forwarded to another interface.
1317 void dev_disable_lro(struct net_device
*dev
)
1320 * If we're trying to disable lro on a vlan device
1321 * use the underlying physical device instead
1323 if (is_vlan_dev(dev
))
1324 dev
= vlan_dev_real_dev(dev
);
1326 dev
->wanted_features
&= ~NETIF_F_LRO
;
1327 netdev_update_features(dev
);
1329 if (unlikely(dev
->features
& NETIF_F_LRO
))
1330 netdev_WARN(dev
, "failed to disable LRO!\n");
1332 EXPORT_SYMBOL(dev_disable_lro
);
1335 static int dev_boot_phase
= 1;
1338 * register_netdevice_notifier - register a network notifier block
1341 * Register a notifier to be called when network device events occur.
1342 * The notifier passed is linked into the kernel structures and must
1343 * not be reused until it has been unregistered. A negative errno code
1344 * is returned on a failure.
1346 * When registered all registration and up events are replayed
1347 * to the new notifier to allow device to have a race free
1348 * view of the network device list.
1351 int register_netdevice_notifier(struct notifier_block
*nb
)
1353 struct net_device
*dev
;
1354 struct net_device
*last
;
1359 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1365 for_each_netdev(net
, dev
) {
1366 err
= nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1367 err
= notifier_to_errno(err
);
1371 if (!(dev
->flags
& IFF_UP
))
1374 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1385 for_each_netdev(net
, dev
) {
1389 if (dev
->flags
& IFF_UP
) {
1390 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1391 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1393 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1394 nb
->notifier_call(nb
, NETDEV_UNREGISTER_BATCH
, dev
);
1399 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1402 EXPORT_SYMBOL(register_netdevice_notifier
);
1405 * unregister_netdevice_notifier - unregister a network notifier block
1408 * Unregister a notifier previously registered by
1409 * register_netdevice_notifier(). The notifier is unlinked into the
1410 * kernel structures and may then be reused. A negative errno code
1411 * is returned on a failure.
1413 * After unregistering unregister and down device events are synthesized
1414 * for all devices on the device list to the removed notifier to remove
1415 * the need for special case cleanup code.
1418 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1420 struct net_device
*dev
;
1425 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1430 for_each_netdev(net
, dev
) {
1431 if (dev
->flags
& IFF_UP
) {
1432 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1433 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1435 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1436 nb
->notifier_call(nb
, NETDEV_UNREGISTER_BATCH
, dev
);
1443 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1446 * call_netdevice_notifiers - call all network notifier blocks
1447 * @val: value passed unmodified to notifier function
1448 * @dev: net_device pointer passed unmodified to notifier function
1450 * Call all network notifier blocks. Parameters and return value
1451 * are as for raw_notifier_call_chain().
1454 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1457 return raw_notifier_call_chain(&netdev_chain
, val
, dev
);
1459 EXPORT_SYMBOL(call_netdevice_notifiers
);
1461 static struct static_key netstamp_needed __read_mostly
;
1462 #ifdef HAVE_JUMP_LABEL
1463 /* We are not allowed to call static_key_slow_dec() from irq context
1464 * If net_disable_timestamp() is called from irq context, defer the
1465 * static_key_slow_dec() calls.
1467 static atomic_t netstamp_needed_deferred
;
1470 void net_enable_timestamp(void)
1472 #ifdef HAVE_JUMP_LABEL
1473 int deferred
= atomic_xchg(&netstamp_needed_deferred
, 0);
1477 static_key_slow_dec(&netstamp_needed
);
1481 WARN_ON(in_interrupt());
1482 static_key_slow_inc(&netstamp_needed
);
1484 EXPORT_SYMBOL(net_enable_timestamp
);
1486 void net_disable_timestamp(void)
1488 #ifdef HAVE_JUMP_LABEL
1489 if (in_interrupt()) {
1490 atomic_inc(&netstamp_needed_deferred
);
1494 static_key_slow_dec(&netstamp_needed
);
1496 EXPORT_SYMBOL(net_disable_timestamp
);
1498 static inline void net_timestamp_set(struct sk_buff
*skb
)
1500 skb
->tstamp
.tv64
= 0;
1501 if (static_key_false(&netstamp_needed
))
1502 __net_timestamp(skb
);
1505 #define net_timestamp_check(COND, SKB) \
1506 if (static_key_false(&netstamp_needed)) { \
1507 if ((COND) && !(SKB)->tstamp.tv64) \
1508 __net_timestamp(SKB); \
1511 static int net_hwtstamp_validate(struct ifreq *ifr)
1513 struct hwtstamp_config cfg
;
1514 enum hwtstamp_tx_types tx_type
;
1515 enum hwtstamp_rx_filters rx_filter
;
1516 int tx_type_valid
= 0;
1517 int rx_filter_valid
= 0;
1519 if (copy_from_user(&cfg
, ifr
->ifr_data
, sizeof(cfg
)))
1522 if (cfg
.flags
) /* reserved for future extensions */
1525 tx_type
= cfg
.tx_type
;
1526 rx_filter
= cfg
.rx_filter
;
1529 case HWTSTAMP_TX_OFF
:
1530 case HWTSTAMP_TX_ON
:
1531 case HWTSTAMP_TX_ONESTEP_SYNC
:
1536 switch (rx_filter
) {
1537 case HWTSTAMP_FILTER_NONE
:
1538 case HWTSTAMP_FILTER_ALL
:
1539 case HWTSTAMP_FILTER_SOME
:
1540 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT
:
1541 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC
:
1542 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ
:
1543 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT
:
1544 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC
:
1545 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ
:
1546 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT
:
1547 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC
:
1548 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ
:
1549 case HWTSTAMP_FILTER_PTP_V2_EVENT
:
1550 case HWTSTAMP_FILTER_PTP_V2_SYNC
:
1551 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ
:
1552 rx_filter_valid
= 1;
1556 if (!tx_type_valid
|| !rx_filter_valid
)
1562 static inline bool is_skb_forwardable(struct net_device
*dev
,
1563 struct sk_buff
*skb
)
1567 if (!(dev
->flags
& IFF_UP
))
1570 len
= dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
;
1571 if (skb
->len
<= len
)
1574 /* if TSO is enabled, we don't care about the length as the packet
1575 * could be forwarded without being segmented before
1577 if (skb_is_gso(skb
))
1584 * dev_forward_skb - loopback an skb to another netif
1586 * @dev: destination network device
1587 * @skb: buffer to forward
1590 * NET_RX_SUCCESS (no congestion)
1591 * NET_RX_DROP (packet was dropped, but freed)
1593 * dev_forward_skb can be used for injecting an skb from the
1594 * start_xmit function of one device into the receive queue
1595 * of another device.
1597 * The receiving device may be in another namespace, so
1598 * we have to clear all information in the skb that could
1599 * impact namespace isolation.
1601 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1603 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
1604 if (skb_copy_ubufs(skb
, GFP_ATOMIC
)) {
1605 atomic_long_inc(&dev
->rx_dropped
);
1614 if (unlikely(!is_skb_forwardable(dev
, skb
))) {
1615 atomic_long_inc(&dev
->rx_dropped
);
1622 skb
->tstamp
.tv64
= 0;
1623 skb
->pkt_type
= PACKET_HOST
;
1624 skb
->protocol
= eth_type_trans(skb
, dev
);
1628 return netif_rx(skb
);
1630 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1632 static inline int deliver_skb(struct sk_buff
*skb
,
1633 struct packet_type
*pt_prev
,
1634 struct net_device
*orig_dev
)
1636 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
1638 atomic_inc(&skb
->users
);
1639 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1643 * Support routine. Sends outgoing frames to any network
1644 * taps currently in use.
1647 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1649 struct packet_type
*ptype
;
1650 struct sk_buff
*skb2
= NULL
;
1651 struct packet_type
*pt_prev
= NULL
;
1654 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1655 /* Never send packets back to the socket
1656 * they originated from - MvS (miquels@drinkel.ow.org)
1658 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1659 (ptype
->af_packet_priv
== NULL
||
1660 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1662 deliver_skb(skb2
, pt_prev
, skb
->dev
);
1667 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1671 net_timestamp_set(skb2
);
1673 /* skb->nh should be correctly
1674 set by sender, so that the second statement is
1675 just protection against buggy protocols.
1677 skb_reset_mac_header(skb2
);
1679 if (skb_network_header(skb2
) < skb2
->data
||
1680 skb2
->network_header
> skb2
->tail
) {
1681 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1682 ntohs(skb2
->protocol
),
1684 skb_reset_network_header(skb2
);
1687 skb2
->transport_header
= skb2
->network_header
;
1688 skb2
->pkt_type
= PACKET_OUTGOING
;
1693 pt_prev
->func(skb2
, skb
->dev
, pt_prev
, skb
->dev
);
1698 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1699 * @dev: Network device
1700 * @txq: number of queues available
1702 * If real_num_tx_queues is changed the tc mappings may no longer be
1703 * valid. To resolve this verify the tc mapping remains valid and if
1704 * not NULL the mapping. With no priorities mapping to this
1705 * offset/count pair it will no longer be used. In the worst case TC0
1706 * is invalid nothing can be done so disable priority mappings. If is
1707 * expected that drivers will fix this mapping if they can before
1708 * calling netif_set_real_num_tx_queues.
1710 static void netif_setup_tc(struct net_device
*dev
, unsigned int txq
)
1713 struct netdev_tc_txq
*tc
= &dev
->tc_to_txq
[0];
1715 /* If TC0 is invalidated disable TC mapping */
1716 if (tc
->offset
+ tc
->count
> txq
) {
1717 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1722 /* Invalidated prio to tc mappings set to TC0 */
1723 for (i
= 1; i
< TC_BITMASK
+ 1; i
++) {
1724 int q
= netdev_get_prio_tc_map(dev
, i
);
1726 tc
= &dev
->tc_to_txq
[q
];
1727 if (tc
->offset
+ tc
->count
> txq
) {
1728 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1730 netdev_set_prio_tc_map(dev
, i
, 0);
1736 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1737 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1739 int netif_set_real_num_tx_queues(struct net_device
*dev
, unsigned int txq
)
1743 if (txq
< 1 || txq
> dev
->num_tx_queues
)
1746 if (dev
->reg_state
== NETREG_REGISTERED
||
1747 dev
->reg_state
== NETREG_UNREGISTERING
) {
1750 rc
= netdev_queue_update_kobjects(dev
, dev
->real_num_tx_queues
,
1756 netif_setup_tc(dev
, txq
);
1758 if (txq
< dev
->real_num_tx_queues
)
1759 qdisc_reset_all_tx_gt(dev
, txq
);
1762 dev
->real_num_tx_queues
= txq
;
1765 EXPORT_SYMBOL(netif_set_real_num_tx_queues
);
1769 * netif_set_real_num_rx_queues - set actual number of RX queues used
1770 * @dev: Network device
1771 * @rxq: Actual number of RX queues
1773 * This must be called either with the rtnl_lock held or before
1774 * registration of the net device. Returns 0 on success, or a
1775 * negative error code. If called before registration, it always
1778 int netif_set_real_num_rx_queues(struct net_device
*dev
, unsigned int rxq
)
1782 if (rxq
< 1 || rxq
> dev
->num_rx_queues
)
1785 if (dev
->reg_state
== NETREG_REGISTERED
) {
1788 rc
= net_rx_queue_update_kobjects(dev
, dev
->real_num_rx_queues
,
1794 dev
->real_num_rx_queues
= rxq
;
1797 EXPORT_SYMBOL(netif_set_real_num_rx_queues
);
1801 * netif_get_num_default_rss_queues - default number of RSS queues
1803 * This routine should set an upper limit on the number of RSS queues
1804 * used by default by multiqueue devices.
1806 int netif_get_num_default_rss_queues(void)
1808 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES
, num_online_cpus());
1810 EXPORT_SYMBOL(netif_get_num_default_rss_queues
);
1812 static inline void __netif_reschedule(struct Qdisc
*q
)
1814 struct softnet_data
*sd
;
1815 unsigned long flags
;
1817 local_irq_save(flags
);
1818 sd
= &__get_cpu_var(softnet_data
);
1819 q
->next_sched
= NULL
;
1820 *sd
->output_queue_tailp
= q
;
1821 sd
->output_queue_tailp
= &q
->next_sched
;
1822 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1823 local_irq_restore(flags
);
1826 void __netif_schedule(struct Qdisc
*q
)
1828 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
1829 __netif_reschedule(q
);
1831 EXPORT_SYMBOL(__netif_schedule
);
1833 void dev_kfree_skb_irq(struct sk_buff
*skb
)
1835 if (atomic_dec_and_test(&skb
->users
)) {
1836 struct softnet_data
*sd
;
1837 unsigned long flags
;
1839 local_irq_save(flags
);
1840 sd
= &__get_cpu_var(softnet_data
);
1841 skb
->next
= sd
->completion_queue
;
1842 sd
->completion_queue
= skb
;
1843 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1844 local_irq_restore(flags
);
1847 EXPORT_SYMBOL(dev_kfree_skb_irq
);
1849 void dev_kfree_skb_any(struct sk_buff
*skb
)
1851 if (in_irq() || irqs_disabled())
1852 dev_kfree_skb_irq(skb
);
1856 EXPORT_SYMBOL(dev_kfree_skb_any
);
1860 * netif_device_detach - mark device as removed
1861 * @dev: network device
1863 * Mark device as removed from system and therefore no longer available.
1865 void netif_device_detach(struct net_device
*dev
)
1867 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1868 netif_running(dev
)) {
1869 netif_tx_stop_all_queues(dev
);
1872 EXPORT_SYMBOL(netif_device_detach
);
1875 * netif_device_attach - mark device as attached
1876 * @dev: network device
1878 * Mark device as attached from system and restart if needed.
1880 void netif_device_attach(struct net_device
*dev
)
1882 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1883 netif_running(dev
)) {
1884 netif_tx_wake_all_queues(dev
);
1885 __netdev_watchdog_up(dev
);
1888 EXPORT_SYMBOL(netif_device_attach
);
1890 static void skb_warn_bad_offload(const struct sk_buff
*skb
)
1892 static const netdev_features_t null_features
= 0;
1893 struct net_device
*dev
= skb
->dev
;
1894 const char *driver
= "";
1896 if (dev
&& dev
->dev
.parent
)
1897 driver
= dev_driver_string(dev
->dev
.parent
);
1899 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
1900 "gso_type=%d ip_summed=%d\n",
1901 driver
, dev
? &dev
->features
: &null_features
,
1902 skb
->sk
? &skb
->sk
->sk_route_caps
: &null_features
,
1903 skb
->len
, skb
->data_len
, skb_shinfo(skb
)->gso_size
,
1904 skb_shinfo(skb
)->gso_type
, skb
->ip_summed
);
1908 * Invalidate hardware checksum when packet is to be mangled, and
1909 * complete checksum manually on outgoing path.
1911 int skb_checksum_help(struct sk_buff
*skb
)
1914 int ret
= 0, offset
;
1916 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1917 goto out_set_summed
;
1919 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1920 skb_warn_bad_offload(skb
);
1924 offset
= skb_checksum_start_offset(skb
);
1925 BUG_ON(offset
>= skb_headlen(skb
));
1926 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1928 offset
+= skb
->csum_offset
;
1929 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
1931 if (skb_cloned(skb
) &&
1932 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
1933 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1938 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1940 skb
->ip_summed
= CHECKSUM_NONE
;
1944 EXPORT_SYMBOL(skb_checksum_help
);
1947 * skb_gso_segment - Perform segmentation on skb.
1948 * @skb: buffer to segment
1949 * @features: features for the output path (see dev->features)
1951 * This function segments the given skb and returns a list of segments.
1953 * It may return NULL if the skb requires no segmentation. This is
1954 * only possible when GSO is used for verifying header integrity.
1956 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
,
1957 netdev_features_t features
)
1959 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1960 struct packet_type
*ptype
;
1961 __be16 type
= skb
->protocol
;
1962 int vlan_depth
= ETH_HLEN
;
1965 while (type
== htons(ETH_P_8021Q
)) {
1966 struct vlan_hdr
*vh
;
1968 if (unlikely(!pskb_may_pull(skb
, vlan_depth
+ VLAN_HLEN
)))
1969 return ERR_PTR(-EINVAL
);
1971 vh
= (struct vlan_hdr
*)(skb
->data
+ vlan_depth
);
1972 type
= vh
->h_vlan_encapsulated_proto
;
1973 vlan_depth
+= VLAN_HLEN
;
1976 skb_reset_mac_header(skb
);
1977 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1978 __skb_pull(skb
, skb
->mac_len
);
1980 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1981 skb_warn_bad_offload(skb
);
1983 if (skb_header_cloned(skb
) &&
1984 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
1985 return ERR_PTR(err
);
1989 list_for_each_entry_rcu(ptype
,
1990 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
1991 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
1992 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1993 err
= ptype
->gso_send_check(skb
);
1994 segs
= ERR_PTR(err
);
1995 if (err
|| skb_gso_ok(skb
, features
))
1997 __skb_push(skb
, (skb
->data
-
1998 skb_network_header(skb
)));
2000 segs
= ptype
->gso_segment(skb
, features
);
2006 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
2010 EXPORT_SYMBOL(skb_gso_segment
);
2012 /* Take action when hardware reception checksum errors are detected. */
2014 void netdev_rx_csum_fault(struct net_device
*dev
)
2016 if (net_ratelimit()) {
2017 pr_err("%s: hw csum failure\n", dev
? dev
->name
: "<unknown>");
2021 EXPORT_SYMBOL(netdev_rx_csum_fault
);
2024 /* Actually, we should eliminate this check as soon as we know, that:
2025 * 1. IOMMU is present and allows to map all the memory.
2026 * 2. No high memory really exists on this machine.
2029 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
2031 #ifdef CONFIG_HIGHMEM
2033 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
2034 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2035 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2036 if (PageHighMem(skb_frag_page(frag
)))
2041 if (PCI_DMA_BUS_IS_PHYS
) {
2042 struct device
*pdev
= dev
->dev
.parent
;
2046 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2047 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2048 dma_addr_t addr
= page_to_phys(skb_frag_page(frag
));
2049 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
2058 void (*destructor
)(struct sk_buff
*skb
);
2061 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2063 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
2065 struct dev_gso_cb
*cb
;
2068 struct sk_buff
*nskb
= skb
->next
;
2070 skb
->next
= nskb
->next
;
2073 } while (skb
->next
);
2075 cb
= DEV_GSO_CB(skb
);
2077 cb
->destructor(skb
);
2081 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2082 * @skb: buffer to segment
2083 * @features: device features as applicable to this skb
2085 * This function segments the given skb and stores the list of segments
2088 static int dev_gso_segment(struct sk_buff
*skb
, netdev_features_t features
)
2090 struct sk_buff
*segs
;
2092 segs
= skb_gso_segment(skb
, features
);
2094 /* Verifying header integrity only. */
2099 return PTR_ERR(segs
);
2102 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
2103 skb
->destructor
= dev_gso_skb_destructor
;
2108 static bool can_checksum_protocol(netdev_features_t features
, __be16 protocol
)
2110 return ((features
& NETIF_F_GEN_CSUM
) ||
2111 ((features
& NETIF_F_V4_CSUM
) &&
2112 protocol
== htons(ETH_P_IP
)) ||
2113 ((features
& NETIF_F_V6_CSUM
) &&
2114 protocol
== htons(ETH_P_IPV6
)) ||
2115 ((features
& NETIF_F_FCOE_CRC
) &&
2116 protocol
== htons(ETH_P_FCOE
)));
2119 static netdev_features_t
harmonize_features(struct sk_buff
*skb
,
2120 __be16 protocol
, netdev_features_t features
)
2122 if (!can_checksum_protocol(features
, protocol
)) {
2123 features
&= ~NETIF_F_ALL_CSUM
;
2124 features
&= ~NETIF_F_SG
;
2125 } else if (illegal_highdma(skb
->dev
, skb
)) {
2126 features
&= ~NETIF_F_SG
;
2132 netdev_features_t
netif_skb_features(struct sk_buff
*skb
)
2134 __be16 protocol
= skb
->protocol
;
2135 netdev_features_t features
= skb
->dev
->features
;
2137 if (skb_shinfo(skb
)->gso_segs
> skb
->dev
->gso_max_segs
)
2138 features
&= ~NETIF_F_GSO_MASK
;
2140 if (protocol
== htons(ETH_P_8021Q
)) {
2141 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
2142 protocol
= veh
->h_vlan_encapsulated_proto
;
2143 } else if (!vlan_tx_tag_present(skb
)) {
2144 return harmonize_features(skb
, protocol
, features
);
2147 features
&= (skb
->dev
->vlan_features
| NETIF_F_HW_VLAN_TX
);
2149 if (protocol
!= htons(ETH_P_8021Q
)) {
2150 return harmonize_features(skb
, protocol
, features
);
2152 features
&= NETIF_F_SG
| NETIF_F_HIGHDMA
| NETIF_F_FRAGLIST
|
2153 NETIF_F_GEN_CSUM
| NETIF_F_HW_VLAN_TX
;
2154 return harmonize_features(skb
, protocol
, features
);
2157 EXPORT_SYMBOL(netif_skb_features
);
2160 * Returns true if either:
2161 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2162 * 2. skb is fragmented and the device does not support SG, or if
2163 * at least one of fragments is in highmem and device does not
2164 * support DMA from it.
2166 static inline int skb_needs_linearize(struct sk_buff
*skb
,
2169 return skb_is_nonlinear(skb
) &&
2170 ((skb_has_frag_list(skb
) &&
2171 !(features
& NETIF_F_FRAGLIST
)) ||
2172 (skb_shinfo(skb
)->nr_frags
&&
2173 !(features
& NETIF_F_SG
)));
2176 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
2177 struct netdev_queue
*txq
)
2179 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2180 int rc
= NETDEV_TX_OK
;
2181 unsigned int skb_len
;
2183 if (likely(!skb
->next
)) {
2184 netdev_features_t features
;
2187 * If device doesn't need skb->dst, release it right now while
2188 * its hot in this cpu cache
2190 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2193 if (!list_empty(&ptype_all
))
2194 dev_queue_xmit_nit(skb
, dev
);
2196 features
= netif_skb_features(skb
);
2198 if (vlan_tx_tag_present(skb
) &&
2199 !(features
& NETIF_F_HW_VLAN_TX
)) {
2200 skb
= __vlan_put_tag(skb
, vlan_tx_tag_get(skb
));
2207 if (netif_needs_gso(skb
, features
)) {
2208 if (unlikely(dev_gso_segment(skb
, features
)))
2213 if (skb_needs_linearize(skb
, features
) &&
2214 __skb_linearize(skb
))
2217 /* If packet is not checksummed and device does not
2218 * support checksumming for this protocol, complete
2219 * checksumming here.
2221 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2222 skb_set_transport_header(skb
,
2223 skb_checksum_start_offset(skb
));
2224 if (!(features
& NETIF_F_ALL_CSUM
) &&
2225 skb_checksum_help(skb
))
2231 rc
= ops
->ndo_start_xmit(skb
, dev
);
2232 trace_net_dev_xmit(skb
, rc
, dev
, skb_len
);
2233 if (rc
== NETDEV_TX_OK
)
2234 txq_trans_update(txq
);
2240 struct sk_buff
*nskb
= skb
->next
;
2242 skb
->next
= nskb
->next
;
2246 * If device doesn't need nskb->dst, release it right now while
2247 * its hot in this cpu cache
2249 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2252 skb_len
= nskb
->len
;
2253 rc
= ops
->ndo_start_xmit(nskb
, dev
);
2254 trace_net_dev_xmit(nskb
, rc
, dev
, skb_len
);
2255 if (unlikely(rc
!= NETDEV_TX_OK
)) {
2256 if (rc
& ~NETDEV_TX_MASK
)
2257 goto out_kfree_gso_skb
;
2258 nskb
->next
= skb
->next
;
2262 txq_trans_update(txq
);
2263 if (unlikely(netif_xmit_stopped(txq
) && skb
->next
))
2264 return NETDEV_TX_BUSY
;
2265 } while (skb
->next
);
2268 if (likely(skb
->next
== NULL
))
2269 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
2276 static u32 hashrnd __read_mostly
;
2279 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2280 * to be used as a distribution range.
2282 u16
__skb_tx_hash(const struct net_device
*dev
, const struct sk_buff
*skb
,
2283 unsigned int num_tx_queues
)
2287 u16 qcount
= num_tx_queues
;
2289 if (skb_rx_queue_recorded(skb
)) {
2290 hash
= skb_get_rx_queue(skb
);
2291 while (unlikely(hash
>= num_tx_queues
))
2292 hash
-= num_tx_queues
;
2297 u8 tc
= netdev_get_prio_tc_map(dev
, skb
->priority
);
2298 qoffset
= dev
->tc_to_txq
[tc
].offset
;
2299 qcount
= dev
->tc_to_txq
[tc
].count
;
2302 if (skb
->sk
&& skb
->sk
->sk_hash
)
2303 hash
= skb
->sk
->sk_hash
;
2305 hash
= (__force u16
) skb
->protocol
;
2306 hash
= jhash_1word(hash
, hashrnd
);
2308 return (u16
) (((u64
) hash
* qcount
) >> 32) + qoffset
;
2310 EXPORT_SYMBOL(__skb_tx_hash
);
2312 static inline u16
dev_cap_txqueue(struct net_device
*dev
, u16 queue_index
)
2314 if (unlikely(queue_index
>= dev
->real_num_tx_queues
)) {
2315 net_warn_ratelimited("%s selects TX queue %d, but real number of TX queues is %d\n",
2316 dev
->name
, queue_index
,
2317 dev
->real_num_tx_queues
);
2323 static inline int get_xps_queue(struct net_device
*dev
, struct sk_buff
*skb
)
2326 struct xps_dev_maps
*dev_maps
;
2327 struct xps_map
*map
;
2328 int queue_index
= -1;
2331 dev_maps
= rcu_dereference(dev
->xps_maps
);
2333 map
= rcu_dereference(
2334 dev_maps
->cpu_map
[raw_smp_processor_id()]);
2337 queue_index
= map
->queues
[0];
2340 if (skb
->sk
&& skb
->sk
->sk_hash
)
2341 hash
= skb
->sk
->sk_hash
;
2343 hash
= (__force u16
) skb
->protocol
^
2345 hash
= jhash_1word(hash
, hashrnd
);
2346 queue_index
= map
->queues
[
2347 ((u64
)hash
* map
->len
) >> 32];
2349 if (unlikely(queue_index
>= dev
->real_num_tx_queues
))
2361 static struct netdev_queue
*dev_pick_tx(struct net_device
*dev
,
2362 struct sk_buff
*skb
)
2365 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2367 if (dev
->real_num_tx_queues
== 1)
2369 else if (ops
->ndo_select_queue
) {
2370 queue_index
= ops
->ndo_select_queue(dev
, skb
);
2371 queue_index
= dev_cap_txqueue(dev
, queue_index
);
2373 struct sock
*sk
= skb
->sk
;
2374 queue_index
= sk_tx_queue_get(sk
);
2376 if (queue_index
< 0 || skb
->ooo_okay
||
2377 queue_index
>= dev
->real_num_tx_queues
) {
2378 int old_index
= queue_index
;
2380 queue_index
= get_xps_queue(dev
, skb
);
2381 if (queue_index
< 0)
2382 queue_index
= skb_tx_hash(dev
, skb
);
2384 if (queue_index
!= old_index
&& sk
) {
2385 struct dst_entry
*dst
=
2386 rcu_dereference_check(sk
->sk_dst_cache
, 1);
2388 if (dst
&& skb_dst(skb
) == dst
)
2389 sk_tx_queue_set(sk
, queue_index
);
2394 skb_set_queue_mapping(skb
, queue_index
);
2395 return netdev_get_tx_queue(dev
, queue_index
);
2398 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2399 struct net_device
*dev
,
2400 struct netdev_queue
*txq
)
2402 spinlock_t
*root_lock
= qdisc_lock(q
);
2406 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
2407 qdisc_calculate_pkt_len(skb
, q
);
2409 * Heuristic to force contended enqueues to serialize on a
2410 * separate lock before trying to get qdisc main lock.
2411 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2412 * and dequeue packets faster.
2414 contended
= qdisc_is_running(q
);
2415 if (unlikely(contended
))
2416 spin_lock(&q
->busylock
);
2418 spin_lock(root_lock
);
2419 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2422 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2423 qdisc_run_begin(q
)) {
2425 * This is a work-conserving queue; there are no old skbs
2426 * waiting to be sent out; and the qdisc is not running -
2427 * xmit the skb directly.
2429 if (!(dev
->priv_flags
& IFF_XMIT_DST_RELEASE
))
2432 qdisc_bstats_update(q
, skb
);
2434 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
)) {
2435 if (unlikely(contended
)) {
2436 spin_unlock(&q
->busylock
);
2443 rc
= NET_XMIT_SUCCESS
;
2446 rc
= q
->enqueue(skb
, q
) & NET_XMIT_MASK
;
2447 if (qdisc_run_begin(q
)) {
2448 if (unlikely(contended
)) {
2449 spin_unlock(&q
->busylock
);
2455 spin_unlock(root_lock
);
2456 if (unlikely(contended
))
2457 spin_unlock(&q
->busylock
);
2461 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2462 static void skb_update_prio(struct sk_buff
*skb
)
2464 struct netprio_map
*map
= rcu_dereference_bh(skb
->dev
->priomap
);
2466 if (!skb
->priority
&& skb
->sk
&& map
) {
2467 unsigned int prioidx
= skb
->sk
->sk_cgrp_prioidx
;
2469 if (prioidx
< map
->priomap_len
)
2470 skb
->priority
= map
->priomap
[prioidx
];
2474 #define skb_update_prio(skb)
2477 static DEFINE_PER_CPU(int, xmit_recursion
);
2478 #define RECURSION_LIMIT 10
2481 * dev_loopback_xmit - loop back @skb
2482 * @skb: buffer to transmit
2484 int dev_loopback_xmit(struct sk_buff
*skb
)
2486 skb_reset_mac_header(skb
);
2487 __skb_pull(skb
, skb_network_offset(skb
));
2488 skb
->pkt_type
= PACKET_LOOPBACK
;
2489 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
2490 WARN_ON(!skb_dst(skb
));
2495 EXPORT_SYMBOL(dev_loopback_xmit
);
2498 * dev_queue_xmit - transmit a buffer
2499 * @skb: buffer to transmit
2501 * Queue a buffer for transmission to a network device. The caller must
2502 * have set the device and priority and built the buffer before calling
2503 * this function. The function can be called from an interrupt.
2505 * A negative errno code is returned on a failure. A success does not
2506 * guarantee the frame will be transmitted as it may be dropped due
2507 * to congestion or traffic shaping.
2509 * -----------------------------------------------------------------------------------
2510 * I notice this method can also return errors from the queue disciplines,
2511 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2514 * Regardless of the return value, the skb is consumed, so it is currently
2515 * difficult to retry a send to this method. (You can bump the ref count
2516 * before sending to hold a reference for retry if you are careful.)
2518 * When calling this method, interrupts MUST be enabled. This is because
2519 * the BH enable code must have IRQs enabled so that it will not deadlock.
2522 int dev_queue_xmit(struct sk_buff
*skb
)
2524 struct net_device
*dev
= skb
->dev
;
2525 struct netdev_queue
*txq
;
2529 /* Disable soft irqs for various locks below. Also
2530 * stops preemption for RCU.
2534 skb_update_prio(skb
);
2536 txq
= dev_pick_tx(dev
, skb
);
2537 q
= rcu_dereference_bh(txq
->qdisc
);
2539 #ifdef CONFIG_NET_CLS_ACT
2540 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2542 trace_net_dev_queue(skb
);
2544 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2548 /* The device has no queue. Common case for software devices:
2549 loopback, all the sorts of tunnels...
2551 Really, it is unlikely that netif_tx_lock protection is necessary
2552 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2554 However, it is possible, that they rely on protection
2557 Check this and shot the lock. It is not prone from deadlocks.
2558 Either shot noqueue qdisc, it is even simpler 8)
2560 if (dev
->flags
& IFF_UP
) {
2561 int cpu
= smp_processor_id(); /* ok because BHs are off */
2563 if (txq
->xmit_lock_owner
!= cpu
) {
2565 if (__this_cpu_read(xmit_recursion
) > RECURSION_LIMIT
)
2566 goto recursion_alert
;
2568 HARD_TX_LOCK(dev
, txq
, cpu
);
2570 if (!netif_xmit_stopped(txq
)) {
2571 __this_cpu_inc(xmit_recursion
);
2572 rc
= dev_hard_start_xmit(skb
, dev
, txq
);
2573 __this_cpu_dec(xmit_recursion
);
2574 if (dev_xmit_complete(rc
)) {
2575 HARD_TX_UNLOCK(dev
, txq
);
2579 HARD_TX_UNLOCK(dev
, txq
);
2580 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2583 /* Recursion is detected! It is possible,
2587 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2593 rcu_read_unlock_bh();
2598 rcu_read_unlock_bh();
2601 EXPORT_SYMBOL(dev_queue_xmit
);
2604 /*=======================================================================
2606 =======================================================================*/
2608 int netdev_max_backlog __read_mostly
= 1000;
2609 int netdev_tstamp_prequeue __read_mostly
= 1;
2610 int netdev_budget __read_mostly
= 300;
2611 int weight_p __read_mostly
= 64; /* old backlog weight */
2613 /* Called with irq disabled */
2614 static inline void ____napi_schedule(struct softnet_data
*sd
,
2615 struct napi_struct
*napi
)
2617 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
2618 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2622 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2623 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2624 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2625 * if hash is a canonical 4-tuple hash over transport ports.
2627 void __skb_get_rxhash(struct sk_buff
*skb
)
2629 struct flow_keys keys
;
2632 if (!skb_flow_dissect(skb
, &keys
))
2636 if ((__force u16
)keys
.port16
[1] < (__force u16
)keys
.port16
[0])
2637 swap(keys
.port16
[0], keys
.port16
[1]);
2641 /* get a consistent hash (same value on both flow directions) */
2642 if ((__force u32
)keys
.dst
< (__force u32
)keys
.src
)
2643 swap(keys
.dst
, keys
.src
);
2645 hash
= jhash_3words((__force u32
)keys
.dst
,
2646 (__force u32
)keys
.src
,
2647 (__force u32
)keys
.ports
, hashrnd
);
2653 EXPORT_SYMBOL(__skb_get_rxhash
);
2657 /* One global table that all flow-based protocols share. */
2658 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
2659 EXPORT_SYMBOL(rps_sock_flow_table
);
2661 struct static_key rps_needed __read_mostly
;
2663 static struct rps_dev_flow
*
2664 set_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2665 struct rps_dev_flow
*rflow
, u16 next_cpu
)
2667 if (next_cpu
!= RPS_NO_CPU
) {
2668 #ifdef CONFIG_RFS_ACCEL
2669 struct netdev_rx_queue
*rxqueue
;
2670 struct rps_dev_flow_table
*flow_table
;
2671 struct rps_dev_flow
*old_rflow
;
2676 /* Should we steer this flow to a different hardware queue? */
2677 if (!skb_rx_queue_recorded(skb
) || !dev
->rx_cpu_rmap
||
2678 !(dev
->features
& NETIF_F_NTUPLE
))
2680 rxq_index
= cpu_rmap_lookup_index(dev
->rx_cpu_rmap
, next_cpu
);
2681 if (rxq_index
== skb_get_rx_queue(skb
))
2684 rxqueue
= dev
->_rx
+ rxq_index
;
2685 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2688 flow_id
= skb
->rxhash
& flow_table
->mask
;
2689 rc
= dev
->netdev_ops
->ndo_rx_flow_steer(dev
, skb
,
2690 rxq_index
, flow_id
);
2694 rflow
= &flow_table
->flows
[flow_id
];
2696 if (old_rflow
->filter
== rflow
->filter
)
2697 old_rflow
->filter
= RPS_NO_FILTER
;
2701 per_cpu(softnet_data
, next_cpu
).input_queue_head
;
2704 rflow
->cpu
= next_cpu
;
2709 * get_rps_cpu is called from netif_receive_skb and returns the target
2710 * CPU from the RPS map of the receiving queue for a given skb.
2711 * rcu_read_lock must be held on entry.
2713 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2714 struct rps_dev_flow
**rflowp
)
2716 struct netdev_rx_queue
*rxqueue
;
2717 struct rps_map
*map
;
2718 struct rps_dev_flow_table
*flow_table
;
2719 struct rps_sock_flow_table
*sock_flow_table
;
2723 if (skb_rx_queue_recorded(skb
)) {
2724 u16 index
= skb_get_rx_queue(skb
);
2725 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
2726 WARN_ONCE(dev
->real_num_rx_queues
> 1,
2727 "%s received packet on queue %u, but number "
2728 "of RX queues is %u\n",
2729 dev
->name
, index
, dev
->real_num_rx_queues
);
2732 rxqueue
= dev
->_rx
+ index
;
2736 map
= rcu_dereference(rxqueue
->rps_map
);
2738 if (map
->len
== 1 &&
2739 !rcu_access_pointer(rxqueue
->rps_flow_table
)) {
2740 tcpu
= map
->cpus
[0];
2741 if (cpu_online(tcpu
))
2745 } else if (!rcu_access_pointer(rxqueue
->rps_flow_table
)) {
2749 skb_reset_network_header(skb
);
2750 if (!skb_get_rxhash(skb
))
2753 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2754 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
2755 if (flow_table
&& sock_flow_table
) {
2757 struct rps_dev_flow
*rflow
;
2759 rflow
= &flow_table
->flows
[skb
->rxhash
& flow_table
->mask
];
2762 next_cpu
= sock_flow_table
->ents
[skb
->rxhash
&
2763 sock_flow_table
->mask
];
2766 * If the desired CPU (where last recvmsg was done) is
2767 * different from current CPU (one in the rx-queue flow
2768 * table entry), switch if one of the following holds:
2769 * - Current CPU is unset (equal to RPS_NO_CPU).
2770 * - Current CPU is offline.
2771 * - The current CPU's queue tail has advanced beyond the
2772 * last packet that was enqueued using this table entry.
2773 * This guarantees that all previous packets for the flow
2774 * have been dequeued, thus preserving in order delivery.
2776 if (unlikely(tcpu
!= next_cpu
) &&
2777 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
2778 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
2779 rflow
->last_qtail
)) >= 0))
2780 rflow
= set_rps_cpu(dev
, skb
, rflow
, next_cpu
);
2782 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
2790 tcpu
= map
->cpus
[((u64
) skb
->rxhash
* map
->len
) >> 32];
2792 if (cpu_online(tcpu
)) {
2802 #ifdef CONFIG_RFS_ACCEL
2805 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2806 * @dev: Device on which the filter was set
2807 * @rxq_index: RX queue index
2808 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2809 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2811 * Drivers that implement ndo_rx_flow_steer() should periodically call
2812 * this function for each installed filter and remove the filters for
2813 * which it returns %true.
2815 bool rps_may_expire_flow(struct net_device
*dev
, u16 rxq_index
,
2816 u32 flow_id
, u16 filter_id
)
2818 struct netdev_rx_queue
*rxqueue
= dev
->_rx
+ rxq_index
;
2819 struct rps_dev_flow_table
*flow_table
;
2820 struct rps_dev_flow
*rflow
;
2825 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2826 if (flow_table
&& flow_id
<= flow_table
->mask
) {
2827 rflow
= &flow_table
->flows
[flow_id
];
2828 cpu
= ACCESS_ONCE(rflow
->cpu
);
2829 if (rflow
->filter
== filter_id
&& cpu
!= RPS_NO_CPU
&&
2830 ((int)(per_cpu(softnet_data
, cpu
).input_queue_head
-
2831 rflow
->last_qtail
) <
2832 (int)(10 * flow_table
->mask
)))
2838 EXPORT_SYMBOL(rps_may_expire_flow
);
2840 #endif /* CONFIG_RFS_ACCEL */
2842 /* Called from hardirq (IPI) context */
2843 static void rps_trigger_softirq(void *data
)
2845 struct softnet_data
*sd
= data
;
2847 ____napi_schedule(sd
, &sd
->backlog
);
2851 #endif /* CONFIG_RPS */
2854 * Check if this softnet_data structure is another cpu one
2855 * If yes, queue it to our IPI list and return 1
2858 static int rps_ipi_queued(struct softnet_data
*sd
)
2861 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
2864 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
2865 mysd
->rps_ipi_list
= sd
;
2867 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2870 #endif /* CONFIG_RPS */
2875 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2876 * queue (may be a remote CPU queue).
2878 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
2879 unsigned int *qtail
)
2881 struct softnet_data
*sd
;
2882 unsigned long flags
;
2884 sd
= &per_cpu(softnet_data
, cpu
);
2886 local_irq_save(flags
);
2889 if (skb_queue_len(&sd
->input_pkt_queue
) <= netdev_max_backlog
) {
2890 if (skb_queue_len(&sd
->input_pkt_queue
)) {
2892 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
2893 input_queue_tail_incr_save(sd
, qtail
);
2895 local_irq_restore(flags
);
2896 return NET_RX_SUCCESS
;
2899 /* Schedule NAPI for backlog device
2900 * We can use non atomic operation since we own the queue lock
2902 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
2903 if (!rps_ipi_queued(sd
))
2904 ____napi_schedule(sd
, &sd
->backlog
);
2912 local_irq_restore(flags
);
2914 atomic_long_inc(&skb
->dev
->rx_dropped
);
2920 * netif_rx - post buffer to the network code
2921 * @skb: buffer to post
2923 * This function receives a packet from a device driver and queues it for
2924 * the upper (protocol) levels to process. It always succeeds. The buffer
2925 * may be dropped during processing for congestion control or by the
2929 * NET_RX_SUCCESS (no congestion)
2930 * NET_RX_DROP (packet was dropped)
2934 int netif_rx(struct sk_buff
*skb
)
2938 /* if netpoll wants it, pretend we never saw it */
2939 if (netpoll_rx(skb
))
2942 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
2944 trace_netif_rx(skb
);
2946 if (static_key_false(&rps_needed
)) {
2947 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
2953 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
2955 cpu
= smp_processor_id();
2957 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
2965 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
2970 EXPORT_SYMBOL(netif_rx
);
2972 int netif_rx_ni(struct sk_buff
*skb
)
2977 err
= netif_rx(skb
);
2978 if (local_softirq_pending())
2984 EXPORT_SYMBOL(netif_rx_ni
);
2986 static void net_tx_action(struct softirq_action
*h
)
2988 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
2990 if (sd
->completion_queue
) {
2991 struct sk_buff
*clist
;
2993 local_irq_disable();
2994 clist
= sd
->completion_queue
;
2995 sd
->completion_queue
= NULL
;
2999 struct sk_buff
*skb
= clist
;
3000 clist
= clist
->next
;
3002 WARN_ON(atomic_read(&skb
->users
));
3003 trace_kfree_skb(skb
, net_tx_action
);
3008 if (sd
->output_queue
) {
3011 local_irq_disable();
3012 head
= sd
->output_queue
;
3013 sd
->output_queue
= NULL
;
3014 sd
->output_queue_tailp
= &sd
->output_queue
;
3018 struct Qdisc
*q
= head
;
3019 spinlock_t
*root_lock
;
3021 head
= head
->next_sched
;
3023 root_lock
= qdisc_lock(q
);
3024 if (spin_trylock(root_lock
)) {
3025 smp_mb__before_clear_bit();
3026 clear_bit(__QDISC_STATE_SCHED
,
3029 spin_unlock(root_lock
);
3031 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
3033 __netif_reschedule(q
);
3035 smp_mb__before_clear_bit();
3036 clear_bit(__QDISC_STATE_SCHED
,
3044 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3045 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3046 /* This hook is defined here for ATM LANE */
3047 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
3048 unsigned char *addr
) __read_mostly
;
3049 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
3052 #ifdef CONFIG_NET_CLS_ACT
3053 /* TODO: Maybe we should just force sch_ingress to be compiled in
3054 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3055 * a compare and 2 stores extra right now if we dont have it on
3056 * but have CONFIG_NET_CLS_ACT
3057 * NOTE: This doesn't stop any functionality; if you dont have
3058 * the ingress scheduler, you just can't add policies on ingress.
3061 static int ing_filter(struct sk_buff
*skb
, struct netdev_queue
*rxq
)
3063 struct net_device
*dev
= skb
->dev
;
3064 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
3065 int result
= TC_ACT_OK
;
3068 if (unlikely(MAX_RED_LOOP
< ttl
++)) {
3069 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3070 skb
->skb_iif
, dev
->ifindex
);
3074 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
3075 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
3078 if (q
!= &noop_qdisc
) {
3079 spin_lock(qdisc_lock(q
));
3080 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
3081 result
= qdisc_enqueue_root(skb
, q
);
3082 spin_unlock(qdisc_lock(q
));
3088 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
3089 struct packet_type
**pt_prev
,
3090 int *ret
, struct net_device
*orig_dev
)
3092 struct netdev_queue
*rxq
= rcu_dereference(skb
->dev
->ingress_queue
);
3094 if (!rxq
|| rxq
->qdisc
== &noop_qdisc
)
3098 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
3102 switch (ing_filter(skb
, rxq
)) {
3116 * netdev_rx_handler_register - register receive handler
3117 * @dev: device to register a handler for
3118 * @rx_handler: receive handler to register
3119 * @rx_handler_data: data pointer that is used by rx handler
3121 * Register a receive hander for a device. This handler will then be
3122 * called from __netif_receive_skb. A negative errno code is returned
3125 * The caller must hold the rtnl_mutex.
3127 * For a general description of rx_handler, see enum rx_handler_result.
3129 int netdev_rx_handler_register(struct net_device
*dev
,
3130 rx_handler_func_t
*rx_handler
,
3131 void *rx_handler_data
)
3135 if (dev
->rx_handler
)
3138 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
3139 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
3143 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
3146 * netdev_rx_handler_unregister - unregister receive handler
3147 * @dev: device to unregister a handler from
3149 * Unregister a receive hander from a device.
3151 * The caller must hold the rtnl_mutex.
3153 void netdev_rx_handler_unregister(struct net_device
*dev
)
3157 RCU_INIT_POINTER(dev
->rx_handler
, NULL
);
3158 RCU_INIT_POINTER(dev
->rx_handler_data
, NULL
);
3160 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
3163 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3164 * the special handling of PFMEMALLOC skbs.
3166 static bool skb_pfmemalloc_protocol(struct sk_buff
*skb
)
3168 switch (skb
->protocol
) {
3169 case __constant_htons(ETH_P_ARP
):
3170 case __constant_htons(ETH_P_IP
):
3171 case __constant_htons(ETH_P_IPV6
):
3172 case __constant_htons(ETH_P_8021Q
):
3179 static int __netif_receive_skb(struct sk_buff
*skb
)
3181 struct packet_type
*ptype
, *pt_prev
;
3182 rx_handler_func_t
*rx_handler
;
3183 struct net_device
*orig_dev
;
3184 struct net_device
*null_or_dev
;
3185 bool deliver_exact
= false;
3186 int ret
= NET_RX_DROP
;
3188 unsigned long pflags
= current
->flags
;
3190 net_timestamp_check(!netdev_tstamp_prequeue
, skb
);
3192 trace_netif_receive_skb(skb
);
3195 * PFMEMALLOC skbs are special, they should
3196 * - be delivered to SOCK_MEMALLOC sockets only
3197 * - stay away from userspace
3198 * - have bounded memory usage
3200 * Use PF_MEMALLOC as this saves us from propagating the allocation
3201 * context down to all allocation sites.
3203 if (sk_memalloc_socks() && skb_pfmemalloc(skb
))
3204 current
->flags
|= PF_MEMALLOC
;
3206 /* if we've gotten here through NAPI, check netpoll */
3207 if (netpoll_receive_skb(skb
))
3210 orig_dev
= skb
->dev
;
3212 skb_reset_network_header(skb
);
3213 skb_reset_transport_header(skb
);
3214 skb_reset_mac_len(skb
);
3221 skb
->skb_iif
= skb
->dev
->ifindex
;
3223 __this_cpu_inc(softnet_data
.processed
);
3225 if (skb
->protocol
== cpu_to_be16(ETH_P_8021Q
)) {
3226 skb
= vlan_untag(skb
);
3231 #ifdef CONFIG_NET_CLS_ACT
3232 if (skb
->tc_verd
& TC_NCLS
) {
3233 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
3238 if (sk_memalloc_socks() && skb_pfmemalloc(skb
))
3241 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
3242 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
3244 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3250 #ifdef CONFIG_NET_CLS_ACT
3251 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
3257 if (sk_memalloc_socks() && skb_pfmemalloc(skb
)
3258 && !skb_pfmemalloc_protocol(skb
))
3261 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
3262 if (vlan_tx_tag_present(skb
)) {
3264 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3267 if (vlan_do_receive(&skb
, !rx_handler
))
3269 else if (unlikely(!skb
))
3275 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3278 switch (rx_handler(&skb
)) {
3279 case RX_HANDLER_CONSUMED
:
3281 case RX_HANDLER_ANOTHER
:
3283 case RX_HANDLER_EXACT
:
3284 deliver_exact
= true;
3285 case RX_HANDLER_PASS
:
3292 /* deliver only exact match when indicated */
3293 null_or_dev
= deliver_exact
? skb
->dev
: NULL
;
3295 type
= skb
->protocol
;
3296 list_for_each_entry_rcu(ptype
,
3297 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
3298 if (ptype
->type
== type
&&
3299 (ptype
->dev
== null_or_dev
|| ptype
->dev
== skb
->dev
||
3300 ptype
->dev
== orig_dev
)) {
3302 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3308 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
3311 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
3314 atomic_long_inc(&skb
->dev
->rx_dropped
);
3316 /* Jamal, now you will not able to escape explaining
3317 * me how you were going to use this. :-)
3325 tsk_restore_flags(current
, pflags
, PF_MEMALLOC
);
3330 * netif_receive_skb - process receive buffer from network
3331 * @skb: buffer to process
3333 * netif_receive_skb() is the main receive data processing function.
3334 * It always succeeds. The buffer may be dropped during processing
3335 * for congestion control or by the protocol layers.
3337 * This function may only be called from softirq context and interrupts
3338 * should be enabled.
3340 * Return values (usually ignored):
3341 * NET_RX_SUCCESS: no congestion
3342 * NET_RX_DROP: packet was dropped
3344 int netif_receive_skb(struct sk_buff
*skb
)
3346 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3348 if (skb_defer_rx_timestamp(skb
))
3349 return NET_RX_SUCCESS
;
3352 if (static_key_false(&rps_needed
)) {
3353 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3358 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3361 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3368 return __netif_receive_skb(skb
);
3370 EXPORT_SYMBOL(netif_receive_skb
);
3372 /* Network device is going away, flush any packets still pending
3373 * Called with irqs disabled.
3375 static void flush_backlog(void *arg
)
3377 struct net_device
*dev
= arg
;
3378 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3379 struct sk_buff
*skb
, *tmp
;
3382 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3383 if (skb
->dev
== dev
) {
3384 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3386 input_queue_head_incr(sd
);
3391 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3392 if (skb
->dev
== dev
) {
3393 __skb_unlink(skb
, &sd
->process_queue
);
3395 input_queue_head_incr(sd
);
3400 static int napi_gro_complete(struct sk_buff
*skb
)
3402 struct packet_type
*ptype
;
3403 __be16 type
= skb
->protocol
;
3404 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3407 if (NAPI_GRO_CB(skb
)->count
== 1) {
3408 skb_shinfo(skb
)->gso_size
= 0;
3413 list_for_each_entry_rcu(ptype
, head
, list
) {
3414 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_complete
)
3417 err
= ptype
->gro_complete(skb
);
3423 WARN_ON(&ptype
->list
== head
);
3425 return NET_RX_SUCCESS
;
3429 return netif_receive_skb(skb
);
3432 inline void napi_gro_flush(struct napi_struct
*napi
)
3434 struct sk_buff
*skb
, *next
;
3436 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3439 napi_gro_complete(skb
);
3442 napi
->gro_count
= 0;
3443 napi
->gro_list
= NULL
;
3445 EXPORT_SYMBOL(napi_gro_flush
);
3447 enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3449 struct sk_buff
**pp
= NULL
;
3450 struct packet_type
*ptype
;
3451 __be16 type
= skb
->protocol
;
3452 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3455 enum gro_result ret
;
3457 if (!(skb
->dev
->features
& NETIF_F_GRO
) || netpoll_rx_on(skb
))
3460 if (skb_is_gso(skb
) || skb_has_frag_list(skb
))
3464 list_for_each_entry_rcu(ptype
, head
, list
) {
3465 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_receive
)
3468 skb_set_network_header(skb
, skb_gro_offset(skb
));
3469 mac_len
= skb
->network_header
- skb
->mac_header
;
3470 skb
->mac_len
= mac_len
;
3471 NAPI_GRO_CB(skb
)->same_flow
= 0;
3472 NAPI_GRO_CB(skb
)->flush
= 0;
3473 NAPI_GRO_CB(skb
)->free
= 0;
3475 pp
= ptype
->gro_receive(&napi
->gro_list
, skb
);
3480 if (&ptype
->list
== head
)
3483 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3484 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3487 struct sk_buff
*nskb
= *pp
;
3491 napi_gro_complete(nskb
);
3498 if (NAPI_GRO_CB(skb
)->flush
|| napi
->gro_count
>= MAX_GRO_SKBS
)
3502 NAPI_GRO_CB(skb
)->count
= 1;
3503 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
3504 skb
->next
= napi
->gro_list
;
3505 napi
->gro_list
= skb
;
3509 if (skb_headlen(skb
) < skb_gro_offset(skb
)) {
3510 int grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
3512 BUG_ON(skb
->end
- skb
->tail
< grow
);
3514 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3517 skb
->data_len
-= grow
;
3519 skb_shinfo(skb
)->frags
[0].page_offset
+= grow
;
3520 skb_frag_size_sub(&skb_shinfo(skb
)->frags
[0], grow
);
3522 if (unlikely(!skb_frag_size(&skb_shinfo(skb
)->frags
[0]))) {
3523 skb_frag_unref(skb
, 0);
3524 memmove(skb_shinfo(skb
)->frags
,
3525 skb_shinfo(skb
)->frags
+ 1,
3526 --skb_shinfo(skb
)->nr_frags
* sizeof(skb_frag_t
));
3537 EXPORT_SYMBOL(dev_gro_receive
);
3539 static inline gro_result_t
3540 __napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3543 unsigned int maclen
= skb
->dev
->hard_header_len
;
3545 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3546 unsigned long diffs
;
3548 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
3549 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
3550 if (maclen
== ETH_HLEN
)
3551 diffs
|= compare_ether_header(skb_mac_header(p
),
3552 skb_gro_mac_header(skb
));
3554 diffs
= memcmp(skb_mac_header(p
),
3555 skb_gro_mac_header(skb
),
3557 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
3558 NAPI_GRO_CB(p
)->flush
= 0;
3561 return dev_gro_receive(napi
, skb
);
3564 gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
3568 if (netif_receive_skb(skb
))
3576 case GRO_MERGED_FREE
:
3577 if (NAPI_GRO_CB(skb
)->free
== NAPI_GRO_FREE_STOLEN_HEAD
)
3578 kmem_cache_free(skbuff_head_cache
, skb
);
3590 EXPORT_SYMBOL(napi_skb_finish
);
3592 void skb_gro_reset_offset(struct sk_buff
*skb
)
3594 NAPI_GRO_CB(skb
)->data_offset
= 0;
3595 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3596 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3598 if (skb
->mac_header
== skb
->tail
&&
3599 !PageHighMem(skb_frag_page(&skb_shinfo(skb
)->frags
[0]))) {
3600 NAPI_GRO_CB(skb
)->frag0
=
3601 skb_frag_address(&skb_shinfo(skb
)->frags
[0]);
3602 NAPI_GRO_CB(skb
)->frag0_len
= skb_frag_size(&skb_shinfo(skb
)->frags
[0]);
3605 EXPORT_SYMBOL(skb_gro_reset_offset
);
3607 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3609 skb_gro_reset_offset(skb
);
3611 return napi_skb_finish(__napi_gro_receive(napi
, skb
), skb
);
3613 EXPORT_SYMBOL(napi_gro_receive
);
3615 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
3617 __skb_pull(skb
, skb_headlen(skb
));
3618 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3619 skb_reserve(skb
, NET_SKB_PAD
+ NET_IP_ALIGN
- skb_headroom(skb
));
3621 skb
->dev
= napi
->dev
;
3627 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
3629 struct sk_buff
*skb
= napi
->skb
;
3632 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
3638 EXPORT_SYMBOL(napi_get_frags
);
3640 gro_result_t
napi_frags_finish(struct napi_struct
*napi
, struct sk_buff
*skb
,
3646 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
3648 if (ret
== GRO_HELD
)
3649 skb_gro_pull(skb
, -ETH_HLEN
);
3650 else if (netif_receive_skb(skb
))
3655 case GRO_MERGED_FREE
:
3656 napi_reuse_skb(napi
, skb
);
3665 EXPORT_SYMBOL(napi_frags_finish
);
3667 static struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
3669 struct sk_buff
*skb
= napi
->skb
;
3676 skb_reset_mac_header(skb
);
3677 skb_gro_reset_offset(skb
);
3679 off
= skb_gro_offset(skb
);
3680 hlen
= off
+ sizeof(*eth
);
3681 eth
= skb_gro_header_fast(skb
, off
);
3682 if (skb_gro_header_hard(skb
, hlen
)) {
3683 eth
= skb_gro_header_slow(skb
, hlen
, off
);
3684 if (unlikely(!eth
)) {
3685 napi_reuse_skb(napi
, skb
);
3691 skb_gro_pull(skb
, sizeof(*eth
));
3694 * This works because the only protocols we care about don't require
3695 * special handling. We'll fix it up properly at the end.
3697 skb
->protocol
= eth
->h_proto
;
3703 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
3705 struct sk_buff
*skb
= napi_frags_skb(napi
);
3710 return napi_frags_finish(napi
, skb
, __napi_gro_receive(napi
, skb
));
3712 EXPORT_SYMBOL(napi_gro_frags
);
3715 * net_rps_action sends any pending IPI's for rps.
3716 * Note: called with local irq disabled, but exits with local irq enabled.
3718 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
3721 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
3724 sd
->rps_ipi_list
= NULL
;
3728 /* Send pending IPI's to kick RPS processing on remote cpus. */
3730 struct softnet_data
*next
= remsd
->rps_ipi_next
;
3732 if (cpu_online(remsd
->cpu
))
3733 __smp_call_function_single(remsd
->cpu
,
3742 static int process_backlog(struct napi_struct
*napi
, int quota
)
3745 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
3748 /* Check if we have pending ipi, its better to send them now,
3749 * not waiting net_rx_action() end.
3751 if (sd
->rps_ipi_list
) {
3752 local_irq_disable();
3753 net_rps_action_and_irq_enable(sd
);
3756 napi
->weight
= weight_p
;
3757 local_irq_disable();
3758 while (work
< quota
) {
3759 struct sk_buff
*skb
;
3762 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
3764 __netif_receive_skb(skb
);
3765 local_irq_disable();
3766 input_queue_head_incr(sd
);
3767 if (++work
>= quota
) {
3774 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3776 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
3777 &sd
->process_queue
);
3779 if (qlen
< quota
- work
) {
3781 * Inline a custom version of __napi_complete().
3782 * only current cpu owns and manipulates this napi,
3783 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3784 * we can use a plain write instead of clear_bit(),
3785 * and we dont need an smp_mb() memory barrier.
3787 list_del(&napi
->poll_list
);
3790 quota
= work
+ qlen
;
3800 * __napi_schedule - schedule for receive
3801 * @n: entry to schedule
3803 * The entry's receive function will be scheduled to run
3805 void __napi_schedule(struct napi_struct
*n
)
3807 unsigned long flags
;
3809 local_irq_save(flags
);
3810 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
3811 local_irq_restore(flags
);
3813 EXPORT_SYMBOL(__napi_schedule
);
3815 void __napi_complete(struct napi_struct
*n
)
3817 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
3818 BUG_ON(n
->gro_list
);
3820 list_del(&n
->poll_list
);
3821 smp_mb__before_clear_bit();
3822 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
3824 EXPORT_SYMBOL(__napi_complete
);
3826 void napi_complete(struct napi_struct
*n
)
3828 unsigned long flags
;
3831 * don't let napi dequeue from the cpu poll list
3832 * just in case its running on a different cpu
3834 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
3838 local_irq_save(flags
);
3840 local_irq_restore(flags
);
3842 EXPORT_SYMBOL(napi_complete
);
3844 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
3845 int (*poll
)(struct napi_struct
*, int), int weight
)
3847 INIT_LIST_HEAD(&napi
->poll_list
);
3848 napi
->gro_count
= 0;
3849 napi
->gro_list
= NULL
;
3852 napi
->weight
= weight
;
3853 list_add(&napi
->dev_list
, &dev
->napi_list
);
3855 #ifdef CONFIG_NETPOLL
3856 spin_lock_init(&napi
->poll_lock
);
3857 napi
->poll_owner
= -1;
3859 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
3861 EXPORT_SYMBOL(netif_napi_add
);
3863 void netif_napi_del(struct napi_struct
*napi
)
3865 struct sk_buff
*skb
, *next
;
3867 list_del_init(&napi
->dev_list
);
3868 napi_free_frags(napi
);
3870 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3876 napi
->gro_list
= NULL
;
3877 napi
->gro_count
= 0;
3879 EXPORT_SYMBOL(netif_napi_del
);
3881 static void net_rx_action(struct softirq_action
*h
)
3883 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3884 unsigned long time_limit
= jiffies
+ 2;
3885 int budget
= netdev_budget
;
3888 local_irq_disable();
3890 while (!list_empty(&sd
->poll_list
)) {
3891 struct napi_struct
*n
;
3894 /* If softirq window is exhuasted then punt.
3895 * Allow this to run for 2 jiffies since which will allow
3896 * an average latency of 1.5/HZ.
3898 if (unlikely(budget
<= 0 || time_after(jiffies
, time_limit
)))
3903 /* Even though interrupts have been re-enabled, this
3904 * access is safe because interrupts can only add new
3905 * entries to the tail of this list, and only ->poll()
3906 * calls can remove this head entry from the list.
3908 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
3910 have
= netpoll_poll_lock(n
);
3914 /* This NAPI_STATE_SCHED test is for avoiding a race
3915 * with netpoll's poll_napi(). Only the entity which
3916 * obtains the lock and sees NAPI_STATE_SCHED set will
3917 * actually make the ->poll() call. Therefore we avoid
3918 * accidentally calling ->poll() when NAPI is not scheduled.
3921 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
3922 work
= n
->poll(n
, weight
);
3926 WARN_ON_ONCE(work
> weight
);
3930 local_irq_disable();
3932 /* Drivers must not modify the NAPI state if they
3933 * consume the entire weight. In such cases this code
3934 * still "owns" the NAPI instance and therefore can
3935 * move the instance around on the list at-will.
3937 if (unlikely(work
== weight
)) {
3938 if (unlikely(napi_disable_pending(n
))) {
3941 local_irq_disable();
3943 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
3946 netpoll_poll_unlock(have
);
3949 net_rps_action_and_irq_enable(sd
);
3951 #ifdef CONFIG_NET_DMA
3953 * There may not be any more sk_buffs coming right now, so push
3954 * any pending DMA copies to hardware
3956 dma_issue_pending_all();
3963 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3967 static gifconf_func_t
*gifconf_list
[NPROTO
];
3970 * register_gifconf - register a SIOCGIF handler
3971 * @family: Address family
3972 * @gifconf: Function handler
3974 * Register protocol dependent address dumping routines. The handler
3975 * that is passed must not be freed or reused until it has been replaced
3976 * by another handler.
3978 int register_gifconf(unsigned int family
, gifconf_func_t
*gifconf
)
3980 if (family
>= NPROTO
)
3982 gifconf_list
[family
] = gifconf
;
3985 EXPORT_SYMBOL(register_gifconf
);
3989 * Map an interface index to its name (SIOCGIFNAME)
3993 * We need this ioctl for efficient implementation of the
3994 * if_indextoname() function required by the IPv6 API. Without
3995 * it, we would have to search all the interfaces to find a
3999 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
4001 struct net_device
*dev
;
4005 * Fetch the caller's info block.
4008 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
4012 dev
= dev_get_by_index_rcu(net
, ifr
.ifr_ifindex
);
4018 strcpy(ifr
.ifr_name
, dev
->name
);
4021 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
4027 * Perform a SIOCGIFCONF call. This structure will change
4028 * size eventually, and there is nothing I can do about it.
4029 * Thus we will need a 'compatibility mode'.
4032 static int dev_ifconf(struct net
*net
, char __user
*arg
)
4035 struct net_device
*dev
;
4042 * Fetch the caller's info block.
4045 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
4052 * Loop over the interfaces, and write an info block for each.
4056 for_each_netdev(net
, dev
) {
4057 for (i
= 0; i
< NPROTO
; i
++) {
4058 if (gifconf_list
[i
]) {
4061 done
= gifconf_list
[i
](dev
, NULL
, 0);
4063 done
= gifconf_list
[i
](dev
, pos
+ total
,
4073 * All done. Write the updated control block back to the caller.
4075 ifc
.ifc_len
= total
;
4078 * Both BSD and Solaris return 0 here, so we do too.
4080 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
4083 #ifdef CONFIG_PROC_FS
4085 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4087 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4088 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4089 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4091 static inline struct net_device
*dev_from_same_bucket(struct seq_file
*seq
, loff_t
*pos
)
4093 struct net
*net
= seq_file_net(seq
);
4094 struct net_device
*dev
;
4095 struct hlist_node
*p
;
4096 struct hlist_head
*h
;
4097 unsigned int count
= 0, offset
= get_offset(*pos
);
4099 h
= &net
->dev_name_head
[get_bucket(*pos
)];
4100 hlist_for_each_entry_rcu(dev
, p
, h
, name_hlist
) {
4101 if (++count
== offset
)
4108 static inline struct net_device
*dev_from_bucket(struct seq_file
*seq
, loff_t
*pos
)
4110 struct net_device
*dev
;
4111 unsigned int bucket
;
4114 dev
= dev_from_same_bucket(seq
, pos
);
4118 bucket
= get_bucket(*pos
) + 1;
4119 *pos
= set_bucket_offset(bucket
, 1);
4120 } while (bucket
< NETDEV_HASHENTRIES
);
4126 * This is invoked by the /proc filesystem handler to display a device
4129 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4134 return SEQ_START_TOKEN
;
4136 if (get_bucket(*pos
) >= NETDEV_HASHENTRIES
)
4139 return dev_from_bucket(seq
, pos
);
4142 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4145 return dev_from_bucket(seq
, pos
);
4148 void dev_seq_stop(struct seq_file
*seq
, void *v
)
4154 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
4156 struct rtnl_link_stats64 temp
;
4157 const struct rtnl_link_stats64
*stats
= dev_get_stats(dev
, &temp
);
4159 seq_printf(seq
, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4160 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4161 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
4163 stats
->rx_dropped
+ stats
->rx_missed_errors
,
4164 stats
->rx_fifo_errors
,
4165 stats
->rx_length_errors
+ stats
->rx_over_errors
+
4166 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
4167 stats
->rx_compressed
, stats
->multicast
,
4168 stats
->tx_bytes
, stats
->tx_packets
,
4169 stats
->tx_errors
, stats
->tx_dropped
,
4170 stats
->tx_fifo_errors
, stats
->collisions
,
4171 stats
->tx_carrier_errors
+
4172 stats
->tx_aborted_errors
+
4173 stats
->tx_window_errors
+
4174 stats
->tx_heartbeat_errors
,
4175 stats
->tx_compressed
);
4179 * Called from the PROCfs module. This now uses the new arbitrary sized
4180 * /proc/net interface to create /proc/net/dev
4182 static int dev_seq_show(struct seq_file
*seq
, void *v
)
4184 if (v
== SEQ_START_TOKEN
)
4185 seq_puts(seq
, "Inter-| Receive "
4187 " face |bytes packets errs drop fifo frame "
4188 "compressed multicast|bytes packets errs "
4189 "drop fifo colls carrier compressed\n");
4191 dev_seq_printf_stats(seq
, v
);
4195 static struct softnet_data
*softnet_get_online(loff_t
*pos
)
4197 struct softnet_data
*sd
= NULL
;
4199 while (*pos
< nr_cpu_ids
)
4200 if (cpu_online(*pos
)) {
4201 sd
= &per_cpu(softnet_data
, *pos
);
4208 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4210 return softnet_get_online(pos
);
4213 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4216 return softnet_get_online(pos
);
4219 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
4223 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
4225 struct softnet_data
*sd
= v
;
4227 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4228 sd
->processed
, sd
->dropped
, sd
->time_squeeze
, 0,
4229 0, 0, 0, 0, /* was fastroute */
4230 sd
->cpu_collision
, sd
->received_rps
);
4234 static const struct seq_operations dev_seq_ops
= {
4235 .start
= dev_seq_start
,
4236 .next
= dev_seq_next
,
4237 .stop
= dev_seq_stop
,
4238 .show
= dev_seq_show
,
4241 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
4243 return seq_open_net(inode
, file
, &dev_seq_ops
,
4244 sizeof(struct seq_net_private
));
4247 static const struct file_operations dev_seq_fops
= {
4248 .owner
= THIS_MODULE
,
4249 .open
= dev_seq_open
,
4251 .llseek
= seq_lseek
,
4252 .release
= seq_release_net
,
4255 static const struct seq_operations softnet_seq_ops
= {
4256 .start
= softnet_seq_start
,
4257 .next
= softnet_seq_next
,
4258 .stop
= softnet_seq_stop
,
4259 .show
= softnet_seq_show
,
4262 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
4264 return seq_open(file
, &softnet_seq_ops
);
4267 static const struct file_operations softnet_seq_fops
= {
4268 .owner
= THIS_MODULE
,
4269 .open
= softnet_seq_open
,
4271 .llseek
= seq_lseek
,
4272 .release
= seq_release
,
4275 static void *ptype_get_idx(loff_t pos
)
4277 struct packet_type
*pt
= NULL
;
4281 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
4287 for (t
= 0; t
< PTYPE_HASH_SIZE
; t
++) {
4288 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
4297 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4301 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
4304 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4306 struct packet_type
*pt
;
4307 struct list_head
*nxt
;
4311 if (v
== SEQ_START_TOKEN
)
4312 return ptype_get_idx(0);
4315 nxt
= pt
->list
.next
;
4316 if (pt
->type
== htons(ETH_P_ALL
)) {
4317 if (nxt
!= &ptype_all
)
4320 nxt
= ptype_base
[0].next
;
4322 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
4324 while (nxt
== &ptype_base
[hash
]) {
4325 if (++hash
>= PTYPE_HASH_SIZE
)
4327 nxt
= ptype_base
[hash
].next
;
4330 return list_entry(nxt
, struct packet_type
, list
);
4333 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
4339 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
4341 struct packet_type
*pt
= v
;
4343 if (v
== SEQ_START_TOKEN
)
4344 seq_puts(seq
, "Type Device Function\n");
4345 else if (pt
->dev
== NULL
|| dev_net(pt
->dev
) == seq_file_net(seq
)) {
4346 if (pt
->type
== htons(ETH_P_ALL
))
4347 seq_puts(seq
, "ALL ");
4349 seq_printf(seq
, "%04x", ntohs(pt
->type
));
4351 seq_printf(seq
, " %-8s %pF\n",
4352 pt
->dev
? pt
->dev
->name
: "", pt
->func
);
4358 static const struct seq_operations ptype_seq_ops
= {
4359 .start
= ptype_seq_start
,
4360 .next
= ptype_seq_next
,
4361 .stop
= ptype_seq_stop
,
4362 .show
= ptype_seq_show
,
4365 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
4367 return seq_open_net(inode
, file
, &ptype_seq_ops
,
4368 sizeof(struct seq_net_private
));
4371 static const struct file_operations ptype_seq_fops
= {
4372 .owner
= THIS_MODULE
,
4373 .open
= ptype_seq_open
,
4375 .llseek
= seq_lseek
,
4376 .release
= seq_release_net
,
4380 static int __net_init
dev_proc_net_init(struct net
*net
)
4384 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
4386 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
4388 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
4391 if (wext_proc_init(net
))
4397 proc_net_remove(net
, "ptype");
4399 proc_net_remove(net
, "softnet_stat");
4401 proc_net_remove(net
, "dev");
4405 static void __net_exit
dev_proc_net_exit(struct net
*net
)
4407 wext_proc_exit(net
);
4409 proc_net_remove(net
, "ptype");
4410 proc_net_remove(net
, "softnet_stat");
4411 proc_net_remove(net
, "dev");
4414 static struct pernet_operations __net_initdata dev_proc_ops
= {
4415 .init
= dev_proc_net_init
,
4416 .exit
= dev_proc_net_exit
,
4419 static int __init
dev_proc_init(void)
4421 return register_pernet_subsys(&dev_proc_ops
);
4424 #define dev_proc_init() 0
4425 #endif /* CONFIG_PROC_FS */
4429 * netdev_set_master - set up master pointer
4430 * @slave: slave device
4431 * @master: new master device
4433 * Changes the master device of the slave. Pass %NULL to break the
4434 * bonding. The caller must hold the RTNL semaphore. On a failure
4435 * a negative errno code is returned. On success the reference counts
4436 * are adjusted and the function returns zero.
4438 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
4440 struct net_device
*old
= slave
->master
;
4450 slave
->master
= master
;
4456 EXPORT_SYMBOL(netdev_set_master
);
4459 * netdev_set_bond_master - set up bonding master/slave pair
4460 * @slave: slave device
4461 * @master: new master device
4463 * Changes the master device of the slave. Pass %NULL to break the
4464 * bonding. The caller must hold the RTNL semaphore. On a failure
4465 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4466 * to the routing socket and the function returns zero.
4468 int netdev_set_bond_master(struct net_device
*slave
, struct net_device
*master
)
4474 err
= netdev_set_master(slave
, master
);
4478 slave
->flags
|= IFF_SLAVE
;
4480 slave
->flags
&= ~IFF_SLAVE
;
4482 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
4485 EXPORT_SYMBOL(netdev_set_bond_master
);
4487 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
4489 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4491 if ((dev
->flags
& IFF_UP
) && ops
->ndo_change_rx_flags
)
4492 ops
->ndo_change_rx_flags(dev
, flags
);
4495 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
4497 unsigned int old_flags
= dev
->flags
;
4503 dev
->flags
|= IFF_PROMISC
;
4504 dev
->promiscuity
+= inc
;
4505 if (dev
->promiscuity
== 0) {
4508 * If inc causes overflow, untouch promisc and return error.
4511 dev
->flags
&= ~IFF_PROMISC
;
4513 dev
->promiscuity
-= inc
;
4514 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4519 if (dev
->flags
!= old_flags
) {
4520 pr_info("device %s %s promiscuous mode\n",
4522 dev
->flags
& IFF_PROMISC
? "entered" : "left");
4523 if (audit_enabled
) {
4524 current_uid_gid(&uid
, &gid
);
4525 audit_log(current
->audit_context
, GFP_ATOMIC
,
4526 AUDIT_ANOM_PROMISCUOUS
,
4527 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4528 dev
->name
, (dev
->flags
& IFF_PROMISC
),
4529 (old_flags
& IFF_PROMISC
),
4530 audit_get_loginuid(current
),
4532 audit_get_sessionid(current
));
4535 dev_change_rx_flags(dev
, IFF_PROMISC
);
4541 * dev_set_promiscuity - update promiscuity count on a device
4545 * Add or remove promiscuity from a device. While the count in the device
4546 * remains above zero the interface remains promiscuous. Once it hits zero
4547 * the device reverts back to normal filtering operation. A negative inc
4548 * value is used to drop promiscuity on the device.
4549 * Return 0 if successful or a negative errno code on error.
4551 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
4553 unsigned int old_flags
= dev
->flags
;
4556 err
= __dev_set_promiscuity(dev
, inc
);
4559 if (dev
->flags
!= old_flags
)
4560 dev_set_rx_mode(dev
);
4563 EXPORT_SYMBOL(dev_set_promiscuity
);
4566 * dev_set_allmulti - update allmulti count on a device
4570 * Add or remove reception of all multicast frames to a device. While the
4571 * count in the device remains above zero the interface remains listening
4572 * to all interfaces. Once it hits zero the device reverts back to normal
4573 * filtering operation. A negative @inc value is used to drop the counter
4574 * when releasing a resource needing all multicasts.
4575 * Return 0 if successful or a negative errno code on error.
4578 int dev_set_allmulti(struct net_device
*dev
, int inc
)
4580 unsigned int old_flags
= dev
->flags
;
4584 dev
->flags
|= IFF_ALLMULTI
;
4585 dev
->allmulti
+= inc
;
4586 if (dev
->allmulti
== 0) {
4589 * If inc causes overflow, untouch allmulti and return error.
4592 dev
->flags
&= ~IFF_ALLMULTI
;
4594 dev
->allmulti
-= inc
;
4595 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4600 if (dev
->flags
^ old_flags
) {
4601 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
4602 dev_set_rx_mode(dev
);
4606 EXPORT_SYMBOL(dev_set_allmulti
);
4609 * Upload unicast and multicast address lists to device and
4610 * configure RX filtering. When the device doesn't support unicast
4611 * filtering it is put in promiscuous mode while unicast addresses
4614 void __dev_set_rx_mode(struct net_device
*dev
)
4616 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4618 /* dev_open will call this function so the list will stay sane. */
4619 if (!(dev
->flags
&IFF_UP
))
4622 if (!netif_device_present(dev
))
4625 if (!(dev
->priv_flags
& IFF_UNICAST_FLT
)) {
4626 /* Unicast addresses changes may only happen under the rtnl,
4627 * therefore calling __dev_set_promiscuity here is safe.
4629 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
4630 __dev_set_promiscuity(dev
, 1);
4631 dev
->uc_promisc
= true;
4632 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
4633 __dev_set_promiscuity(dev
, -1);
4634 dev
->uc_promisc
= false;
4638 if (ops
->ndo_set_rx_mode
)
4639 ops
->ndo_set_rx_mode(dev
);
4642 void dev_set_rx_mode(struct net_device
*dev
)
4644 netif_addr_lock_bh(dev
);
4645 __dev_set_rx_mode(dev
);
4646 netif_addr_unlock_bh(dev
);
4650 * dev_get_flags - get flags reported to userspace
4653 * Get the combination of flag bits exported through APIs to userspace.
4655 unsigned int dev_get_flags(const struct net_device
*dev
)
4659 flags
= (dev
->flags
& ~(IFF_PROMISC
|
4664 (dev
->gflags
& (IFF_PROMISC
|
4667 if (netif_running(dev
)) {
4668 if (netif_oper_up(dev
))
4669 flags
|= IFF_RUNNING
;
4670 if (netif_carrier_ok(dev
))
4671 flags
|= IFF_LOWER_UP
;
4672 if (netif_dormant(dev
))
4673 flags
|= IFF_DORMANT
;
4678 EXPORT_SYMBOL(dev_get_flags
);
4680 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4682 unsigned int old_flags
= dev
->flags
;
4688 * Set the flags on our device.
4691 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
4692 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
4694 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
4698 * Load in the correct multicast list now the flags have changed.
4701 if ((old_flags
^ flags
) & IFF_MULTICAST
)
4702 dev_change_rx_flags(dev
, IFF_MULTICAST
);
4704 dev_set_rx_mode(dev
);
4707 * Have we downed the interface. We handle IFF_UP ourselves
4708 * according to user attempts to set it, rather than blindly
4713 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
4714 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
4717 dev_set_rx_mode(dev
);
4720 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
4721 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
4723 dev
->gflags
^= IFF_PROMISC
;
4724 dev_set_promiscuity(dev
, inc
);
4727 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4728 is important. Some (broken) drivers set IFF_PROMISC, when
4729 IFF_ALLMULTI is requested not asking us and not reporting.
4731 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
4732 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
4734 dev
->gflags
^= IFF_ALLMULTI
;
4735 dev_set_allmulti(dev
, inc
);
4741 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
)
4743 unsigned int changes
= dev
->flags
^ old_flags
;
4745 if (changes
& IFF_UP
) {
4746 if (dev
->flags
& IFF_UP
)
4747 call_netdevice_notifiers(NETDEV_UP
, dev
);
4749 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
4752 if (dev
->flags
& IFF_UP
&&
4753 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
)))
4754 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
4758 * dev_change_flags - change device settings
4760 * @flags: device state flags
4762 * Change settings on device based state flags. The flags are
4763 * in the userspace exported format.
4765 int dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4768 unsigned int changes
, old_flags
= dev
->flags
;
4770 ret
= __dev_change_flags(dev
, flags
);
4774 changes
= old_flags
^ dev
->flags
;
4776 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
4778 __dev_notify_flags(dev
, old_flags
);
4781 EXPORT_SYMBOL(dev_change_flags
);
4784 * dev_set_mtu - Change maximum transfer unit
4786 * @new_mtu: new transfer unit
4788 * Change the maximum transfer size of the network device.
4790 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
4792 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4795 if (new_mtu
== dev
->mtu
)
4798 /* MTU must be positive. */
4802 if (!netif_device_present(dev
))
4806 if (ops
->ndo_change_mtu
)
4807 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
4811 if (!err
&& dev
->flags
& IFF_UP
)
4812 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
4815 EXPORT_SYMBOL(dev_set_mtu
);
4818 * dev_set_group - Change group this device belongs to
4820 * @new_group: group this device should belong to
4822 void dev_set_group(struct net_device
*dev
, int new_group
)
4824 dev
->group
= new_group
;
4826 EXPORT_SYMBOL(dev_set_group
);
4829 * dev_set_mac_address - Change Media Access Control Address
4833 * Change the hardware (MAC) address of the device
4835 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
4837 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4840 if (!ops
->ndo_set_mac_address
)
4842 if (sa
->sa_family
!= dev
->type
)
4844 if (!netif_device_present(dev
))
4846 err
= ops
->ndo_set_mac_address(dev
, sa
);
4848 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4849 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
4852 EXPORT_SYMBOL(dev_set_mac_address
);
4855 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4857 static int dev_ifsioc_locked(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4860 struct net_device
*dev
= dev_get_by_name_rcu(net
, ifr
->ifr_name
);
4866 case SIOCGIFFLAGS
: /* Get interface flags */
4867 ifr
->ifr_flags
= (short) dev_get_flags(dev
);
4870 case SIOCGIFMETRIC
: /* Get the metric on the interface
4871 (currently unused) */
4872 ifr
->ifr_metric
= 0;
4875 case SIOCGIFMTU
: /* Get the MTU of a device */
4876 ifr
->ifr_mtu
= dev
->mtu
;
4881 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
4883 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
4884 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4885 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
4893 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
4894 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
4895 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
4896 ifr
->ifr_map
.irq
= dev
->irq
;
4897 ifr
->ifr_map
.dma
= dev
->dma
;
4898 ifr
->ifr_map
.port
= dev
->if_port
;
4902 ifr
->ifr_ifindex
= dev
->ifindex
;
4906 ifr
->ifr_qlen
= dev
->tx_queue_len
;
4910 /* dev_ioctl() should ensure this case
4922 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4924 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4927 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
4928 const struct net_device_ops
*ops
;
4933 ops
= dev
->netdev_ops
;
4936 case SIOCSIFFLAGS
: /* Set interface flags */
4937 return dev_change_flags(dev
, ifr
->ifr_flags
);
4939 case SIOCSIFMETRIC
: /* Set the metric on the interface
4940 (currently unused) */
4943 case SIOCSIFMTU
: /* Set the MTU of a device */
4944 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
4947 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
4949 case SIOCSIFHWBROADCAST
:
4950 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
4952 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
4953 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4954 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4958 if (ops
->ndo_set_config
) {
4959 if (!netif_device_present(dev
))
4961 return ops
->ndo_set_config(dev
, &ifr
->ifr_map
);
4966 if (!ops
->ndo_set_rx_mode
||
4967 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4969 if (!netif_device_present(dev
))
4971 return dev_mc_add_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4974 if (!ops
->ndo_set_rx_mode
||
4975 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4977 if (!netif_device_present(dev
))
4979 return dev_mc_del_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4982 if (ifr
->ifr_qlen
< 0)
4984 dev
->tx_queue_len
= ifr
->ifr_qlen
;
4988 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
4989 return dev_change_name(dev
, ifr
->ifr_newname
);
4992 err
= net_hwtstamp_validate(ifr
);
4998 * Unknown or private ioctl
5001 if ((cmd
>= SIOCDEVPRIVATE
&&
5002 cmd
<= SIOCDEVPRIVATE
+ 15) ||
5003 cmd
== SIOCBONDENSLAVE
||
5004 cmd
== SIOCBONDRELEASE
||
5005 cmd
== SIOCBONDSETHWADDR
||
5006 cmd
== SIOCBONDSLAVEINFOQUERY
||
5007 cmd
== SIOCBONDINFOQUERY
||
5008 cmd
== SIOCBONDCHANGEACTIVE
||
5009 cmd
== SIOCGMIIPHY
||
5010 cmd
== SIOCGMIIREG
||
5011 cmd
== SIOCSMIIREG
||
5012 cmd
== SIOCBRADDIF
||
5013 cmd
== SIOCBRDELIF
||
5014 cmd
== SIOCSHWTSTAMP
||
5015 cmd
== SIOCWANDEV
) {
5017 if (ops
->ndo_do_ioctl
) {
5018 if (netif_device_present(dev
))
5019 err
= ops
->ndo_do_ioctl(dev
, ifr
, cmd
);
5031 * This function handles all "interface"-type I/O control requests. The actual
5032 * 'doing' part of this is dev_ifsioc above.
5036 * dev_ioctl - network device ioctl
5037 * @net: the applicable net namespace
5038 * @cmd: command to issue
5039 * @arg: pointer to a struct ifreq in user space
5041 * Issue ioctl functions to devices. This is normally called by the
5042 * user space syscall interfaces but can sometimes be useful for
5043 * other purposes. The return value is the return from the syscall if
5044 * positive or a negative errno code on error.
5047 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
5053 /* One special case: SIOCGIFCONF takes ifconf argument
5054 and requires shared lock, because it sleeps writing
5058 if (cmd
== SIOCGIFCONF
) {
5060 ret
= dev_ifconf(net
, (char __user
*) arg
);
5064 if (cmd
== SIOCGIFNAME
)
5065 return dev_ifname(net
, (struct ifreq __user
*)arg
);
5067 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
5070 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
5072 colon
= strchr(ifr
.ifr_name
, ':');
5077 * See which interface the caller is talking about.
5082 * These ioctl calls:
5083 * - can be done by all.
5084 * - atomic and do not require locking.
5095 dev_load(net
, ifr
.ifr_name
);
5097 ret
= dev_ifsioc_locked(net
, &ifr
, cmd
);
5102 if (copy_to_user(arg
, &ifr
,
5103 sizeof(struct ifreq
)))
5109 dev_load(net
, ifr
.ifr_name
);
5111 ret
= dev_ethtool(net
, &ifr
);
5116 if (copy_to_user(arg
, &ifr
,
5117 sizeof(struct ifreq
)))
5123 * These ioctl calls:
5124 * - require superuser power.
5125 * - require strict serialization.
5131 if (!capable(CAP_NET_ADMIN
))
5133 dev_load(net
, ifr
.ifr_name
);
5135 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5140 if (copy_to_user(arg
, &ifr
,
5141 sizeof(struct ifreq
)))
5147 * These ioctl calls:
5148 * - require superuser power.
5149 * - require strict serialization.
5150 * - do not return a value
5160 case SIOCSIFHWBROADCAST
:
5163 case SIOCBONDENSLAVE
:
5164 case SIOCBONDRELEASE
:
5165 case SIOCBONDSETHWADDR
:
5166 case SIOCBONDCHANGEACTIVE
:
5170 if (!capable(CAP_NET_ADMIN
))
5173 case SIOCBONDSLAVEINFOQUERY
:
5174 case SIOCBONDINFOQUERY
:
5175 dev_load(net
, ifr
.ifr_name
);
5177 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5182 /* Get the per device memory space. We can add this but
5183 * currently do not support it */
5185 /* Set the per device memory buffer space.
5186 * Not applicable in our case */
5191 * Unknown or private ioctl.
5194 if (cmd
== SIOCWANDEV
||
5195 (cmd
>= SIOCDEVPRIVATE
&&
5196 cmd
<= SIOCDEVPRIVATE
+ 15)) {
5197 dev_load(net
, ifr
.ifr_name
);
5199 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5201 if (!ret
&& copy_to_user(arg
, &ifr
,
5202 sizeof(struct ifreq
)))
5206 /* Take care of Wireless Extensions */
5207 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
5208 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
5215 * dev_new_index - allocate an ifindex
5216 * @net: the applicable net namespace
5218 * Returns a suitable unique value for a new device interface
5219 * number. The caller must hold the rtnl semaphore or the
5220 * dev_base_lock to be sure it remains unique.
5222 static int dev_new_index(struct net
*net
)
5228 if (!__dev_get_by_index(net
, ifindex
))
5233 /* Delayed registration/unregisteration */
5234 static LIST_HEAD(net_todo_list
);
5236 static void net_set_todo(struct net_device
*dev
)
5238 list_add_tail(&dev
->todo_list
, &net_todo_list
);
5241 static void rollback_registered_many(struct list_head
*head
)
5243 struct net_device
*dev
, *tmp
;
5245 BUG_ON(dev_boot_phase
);
5248 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
5249 /* Some devices call without registering
5250 * for initialization unwind. Remove those
5251 * devices and proceed with the remaining.
5253 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5254 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5258 list_del(&dev
->unreg_list
);
5261 dev
->dismantle
= true;
5262 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
5265 /* If device is running, close it first. */
5266 dev_close_many(head
);
5268 list_for_each_entry(dev
, head
, unreg_list
) {
5269 /* And unlink it from device chain. */
5270 unlist_netdevice(dev
);
5272 dev
->reg_state
= NETREG_UNREGISTERING
;
5277 list_for_each_entry(dev
, head
, unreg_list
) {
5278 /* Shutdown queueing discipline. */
5282 /* Notify protocols, that we are about to destroy
5283 this device. They should clean all the things.
5285 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5287 if (!dev
->rtnl_link_ops
||
5288 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5289 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
5292 * Flush the unicast and multicast chains
5297 if (dev
->netdev_ops
->ndo_uninit
)
5298 dev
->netdev_ops
->ndo_uninit(dev
);
5300 /* Notifier chain MUST detach us from master device. */
5301 WARN_ON(dev
->master
);
5303 /* Remove entries from kobject tree */
5304 netdev_unregister_kobject(dev
);
5307 /* Process any work delayed until the end of the batch */
5308 dev
= list_first_entry(head
, struct net_device
, unreg_list
);
5309 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
5313 list_for_each_entry(dev
, head
, unreg_list
)
5317 static void rollback_registered(struct net_device
*dev
)
5321 list_add(&dev
->unreg_list
, &single
);
5322 rollback_registered_many(&single
);
5326 static netdev_features_t
netdev_fix_features(struct net_device
*dev
,
5327 netdev_features_t features
)
5329 /* Fix illegal checksum combinations */
5330 if ((features
& NETIF_F_HW_CSUM
) &&
5331 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5332 netdev_warn(dev
, "mixed HW and IP checksum settings.\n");
5333 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5336 /* Fix illegal SG+CSUM combinations. */
5337 if ((features
& NETIF_F_SG
) &&
5338 !(features
& NETIF_F_ALL_CSUM
)) {
5340 "Dropping NETIF_F_SG since no checksum feature.\n");
5341 features
&= ~NETIF_F_SG
;
5344 /* TSO requires that SG is present as well. */
5345 if ((features
& NETIF_F_ALL_TSO
) && !(features
& NETIF_F_SG
)) {
5346 netdev_dbg(dev
, "Dropping TSO features since no SG feature.\n");
5347 features
&= ~NETIF_F_ALL_TSO
;
5350 /* TSO ECN requires that TSO is present as well. */
5351 if ((features
& NETIF_F_ALL_TSO
) == NETIF_F_TSO_ECN
)
5352 features
&= ~NETIF_F_TSO_ECN
;
5354 /* Software GSO depends on SG. */
5355 if ((features
& NETIF_F_GSO
) && !(features
& NETIF_F_SG
)) {
5356 netdev_dbg(dev
, "Dropping NETIF_F_GSO since no SG feature.\n");
5357 features
&= ~NETIF_F_GSO
;
5360 /* UFO needs SG and checksumming */
5361 if (features
& NETIF_F_UFO
) {
5362 /* maybe split UFO into V4 and V6? */
5363 if (!((features
& NETIF_F_GEN_CSUM
) ||
5364 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))
5365 == (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5367 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5368 features
&= ~NETIF_F_UFO
;
5371 if (!(features
& NETIF_F_SG
)) {
5373 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5374 features
&= ~NETIF_F_UFO
;
5381 int __netdev_update_features(struct net_device
*dev
)
5383 netdev_features_t features
;
5388 features
= netdev_get_wanted_features(dev
);
5390 if (dev
->netdev_ops
->ndo_fix_features
)
5391 features
= dev
->netdev_ops
->ndo_fix_features(dev
, features
);
5393 /* driver might be less strict about feature dependencies */
5394 features
= netdev_fix_features(dev
, features
);
5396 if (dev
->features
== features
)
5399 netdev_dbg(dev
, "Features changed: %pNF -> %pNF\n",
5400 &dev
->features
, &features
);
5402 if (dev
->netdev_ops
->ndo_set_features
)
5403 err
= dev
->netdev_ops
->ndo_set_features(dev
, features
);
5405 if (unlikely(err
< 0)) {
5407 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5408 err
, &features
, &dev
->features
);
5413 dev
->features
= features
;
5419 * netdev_update_features - recalculate device features
5420 * @dev: the device to check
5422 * Recalculate dev->features set and send notifications if it
5423 * has changed. Should be called after driver or hardware dependent
5424 * conditions might have changed that influence the features.
5426 void netdev_update_features(struct net_device
*dev
)
5428 if (__netdev_update_features(dev
))
5429 netdev_features_change(dev
);
5431 EXPORT_SYMBOL(netdev_update_features
);
5434 * netdev_change_features - recalculate device features
5435 * @dev: the device to check
5437 * Recalculate dev->features set and send notifications even
5438 * if they have not changed. Should be called instead of
5439 * netdev_update_features() if also dev->vlan_features might
5440 * have changed to allow the changes to be propagated to stacked
5443 void netdev_change_features(struct net_device
*dev
)
5445 __netdev_update_features(dev
);
5446 netdev_features_change(dev
);
5448 EXPORT_SYMBOL(netdev_change_features
);
5451 * netif_stacked_transfer_operstate - transfer operstate
5452 * @rootdev: the root or lower level device to transfer state from
5453 * @dev: the device to transfer operstate to
5455 * Transfer operational state from root to device. This is normally
5456 * called when a stacking relationship exists between the root
5457 * device and the device(a leaf device).
5459 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
5460 struct net_device
*dev
)
5462 if (rootdev
->operstate
== IF_OPER_DORMANT
)
5463 netif_dormant_on(dev
);
5465 netif_dormant_off(dev
);
5467 if (netif_carrier_ok(rootdev
)) {
5468 if (!netif_carrier_ok(dev
))
5469 netif_carrier_on(dev
);
5471 if (netif_carrier_ok(dev
))
5472 netif_carrier_off(dev
);
5475 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
5478 static int netif_alloc_rx_queues(struct net_device
*dev
)
5480 unsigned int i
, count
= dev
->num_rx_queues
;
5481 struct netdev_rx_queue
*rx
;
5485 rx
= kcalloc(count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5487 pr_err("netdev: Unable to allocate %u rx queues\n", count
);
5492 for (i
= 0; i
< count
; i
++)
5498 static void netdev_init_one_queue(struct net_device
*dev
,
5499 struct netdev_queue
*queue
, void *_unused
)
5501 /* Initialize queue lock */
5502 spin_lock_init(&queue
->_xmit_lock
);
5503 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
5504 queue
->xmit_lock_owner
= -1;
5505 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
5508 dql_init(&queue
->dql
, HZ
);
5512 static int netif_alloc_netdev_queues(struct net_device
*dev
)
5514 unsigned int count
= dev
->num_tx_queues
;
5515 struct netdev_queue
*tx
;
5519 tx
= kcalloc(count
, sizeof(struct netdev_queue
), GFP_KERNEL
);
5521 pr_err("netdev: Unable to allocate %u tx queues\n", count
);
5526 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5527 spin_lock_init(&dev
->tx_global_lock
);
5533 * register_netdevice - register a network device
5534 * @dev: device to register
5536 * Take a completed network device structure and add it to the kernel
5537 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5538 * chain. 0 is returned on success. A negative errno code is returned
5539 * on a failure to set up the device, or if the name is a duplicate.
5541 * Callers must hold the rtnl semaphore. You may want
5542 * register_netdev() instead of this.
5545 * The locking appears insufficient to guarantee two parallel registers
5546 * will not get the same name.
5549 int register_netdevice(struct net_device
*dev
)
5552 struct net
*net
= dev_net(dev
);
5554 BUG_ON(dev_boot_phase
);
5559 /* When net_device's are persistent, this will be fatal. */
5560 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
5563 spin_lock_init(&dev
->addr_list_lock
);
5564 netdev_set_addr_lockdep_class(dev
);
5568 ret
= dev_get_valid_name(dev
, dev
->name
);
5572 /* Init, if this function is available */
5573 if (dev
->netdev_ops
->ndo_init
) {
5574 ret
= dev
->netdev_ops
->ndo_init(dev
);
5582 dev
->ifindex
= dev_new_index(net
);
5583 if (dev
->iflink
== -1)
5584 dev
->iflink
= dev
->ifindex
;
5586 /* Transfer changeable features to wanted_features and enable
5587 * software offloads (GSO and GRO).
5589 dev
->hw_features
|= NETIF_F_SOFT_FEATURES
;
5590 dev
->features
|= NETIF_F_SOFT_FEATURES
;
5591 dev
->wanted_features
= dev
->features
& dev
->hw_features
;
5593 /* Turn on no cache copy if HW is doing checksum */
5594 if (!(dev
->flags
& IFF_LOOPBACK
)) {
5595 dev
->hw_features
|= NETIF_F_NOCACHE_COPY
;
5596 if (dev
->features
& NETIF_F_ALL_CSUM
) {
5597 dev
->wanted_features
|= NETIF_F_NOCACHE_COPY
;
5598 dev
->features
|= NETIF_F_NOCACHE_COPY
;
5602 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5604 dev
->vlan_features
|= NETIF_F_HIGHDMA
;
5606 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
5607 ret
= notifier_to_errno(ret
);
5611 ret
= netdev_register_kobject(dev
);
5614 dev
->reg_state
= NETREG_REGISTERED
;
5616 __netdev_update_features(dev
);
5619 * Default initial state at registry is that the
5620 * device is present.
5623 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5625 dev_init_scheduler(dev
);
5627 list_netdevice(dev
);
5628 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
5630 /* Notify protocols, that a new device appeared. */
5631 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5632 ret
= notifier_to_errno(ret
);
5634 rollback_registered(dev
);
5635 dev
->reg_state
= NETREG_UNREGISTERED
;
5638 * Prevent userspace races by waiting until the network
5639 * device is fully setup before sending notifications.
5641 if (!dev
->rtnl_link_ops
||
5642 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5643 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5649 if (dev
->netdev_ops
->ndo_uninit
)
5650 dev
->netdev_ops
->ndo_uninit(dev
);
5653 EXPORT_SYMBOL(register_netdevice
);
5656 * init_dummy_netdev - init a dummy network device for NAPI
5657 * @dev: device to init
5659 * This takes a network device structure and initialize the minimum
5660 * amount of fields so it can be used to schedule NAPI polls without
5661 * registering a full blown interface. This is to be used by drivers
5662 * that need to tie several hardware interfaces to a single NAPI
5663 * poll scheduler due to HW limitations.
5665 int init_dummy_netdev(struct net_device
*dev
)
5667 /* Clear everything. Note we don't initialize spinlocks
5668 * are they aren't supposed to be taken by any of the
5669 * NAPI code and this dummy netdev is supposed to be
5670 * only ever used for NAPI polls
5672 memset(dev
, 0, sizeof(struct net_device
));
5674 /* make sure we BUG if trying to hit standard
5675 * register/unregister code path
5677 dev
->reg_state
= NETREG_DUMMY
;
5679 /* NAPI wants this */
5680 INIT_LIST_HEAD(&dev
->napi_list
);
5682 /* a dummy interface is started by default */
5683 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5684 set_bit(__LINK_STATE_START
, &dev
->state
);
5686 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5687 * because users of this 'device' dont need to change
5693 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
5697 * register_netdev - register a network device
5698 * @dev: device to register
5700 * Take a completed network device structure and add it to the kernel
5701 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5702 * chain. 0 is returned on success. A negative errno code is returned
5703 * on a failure to set up the device, or if the name is a duplicate.
5705 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5706 * and expands the device name if you passed a format string to
5709 int register_netdev(struct net_device
*dev
)
5714 err
= register_netdevice(dev
);
5718 EXPORT_SYMBOL(register_netdev
);
5720 int netdev_refcnt_read(const struct net_device
*dev
)
5724 for_each_possible_cpu(i
)
5725 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
5728 EXPORT_SYMBOL(netdev_refcnt_read
);
5731 * netdev_wait_allrefs - wait until all references are gone.
5733 * This is called when unregistering network devices.
5735 * Any protocol or device that holds a reference should register
5736 * for netdevice notification, and cleanup and put back the
5737 * reference if they receive an UNREGISTER event.
5738 * We can get stuck here if buggy protocols don't correctly
5741 static void netdev_wait_allrefs(struct net_device
*dev
)
5743 unsigned long rebroadcast_time
, warning_time
;
5746 linkwatch_forget_dev(dev
);
5748 rebroadcast_time
= warning_time
= jiffies
;
5749 refcnt
= netdev_refcnt_read(dev
);
5751 while (refcnt
!= 0) {
5752 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
5755 /* Rebroadcast unregister notification */
5756 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5757 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5758 * should have already handle it the first time */
5760 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
5762 /* We must not have linkwatch events
5763 * pending on unregister. If this
5764 * happens, we simply run the queue
5765 * unscheduled, resulting in a noop
5768 linkwatch_run_queue();
5773 rebroadcast_time
= jiffies
;
5778 refcnt
= netdev_refcnt_read(dev
);
5780 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
5781 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5783 warning_time
= jiffies
;
5792 * register_netdevice(x1);
5793 * register_netdevice(x2);
5795 * unregister_netdevice(y1);
5796 * unregister_netdevice(y2);
5802 * We are invoked by rtnl_unlock().
5803 * This allows us to deal with problems:
5804 * 1) We can delete sysfs objects which invoke hotplug
5805 * without deadlocking with linkwatch via keventd.
5806 * 2) Since we run with the RTNL semaphore not held, we can sleep
5807 * safely in order to wait for the netdev refcnt to drop to zero.
5809 * We must not return until all unregister events added during
5810 * the interval the lock was held have been completed.
5812 void netdev_run_todo(void)
5814 struct list_head list
;
5816 /* Snapshot list, allow later requests */
5817 list_replace_init(&net_todo_list
, &list
);
5821 /* Wait for rcu callbacks to finish before attempting to drain
5822 * the device list. This usually avoids a 250ms wait.
5824 if (!list_empty(&list
))
5827 while (!list_empty(&list
)) {
5828 struct net_device
*dev
5829 = list_first_entry(&list
, struct net_device
, todo_list
);
5830 list_del(&dev
->todo_list
);
5832 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
5833 pr_err("network todo '%s' but state %d\n",
5834 dev
->name
, dev
->reg_state
);
5839 dev
->reg_state
= NETREG_UNREGISTERED
;
5841 on_each_cpu(flush_backlog
, dev
, 1);
5843 netdev_wait_allrefs(dev
);
5846 BUG_ON(netdev_refcnt_read(dev
));
5847 WARN_ON(rcu_access_pointer(dev
->ip_ptr
));
5848 WARN_ON(rcu_access_pointer(dev
->ip6_ptr
));
5849 WARN_ON(dev
->dn_ptr
);
5851 if (dev
->destructor
)
5852 dev
->destructor(dev
);
5854 /* Free network device */
5855 kobject_put(&dev
->dev
.kobj
);
5859 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5860 * fields in the same order, with only the type differing.
5862 void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
5863 const struct net_device_stats
*netdev_stats
)
5865 #if BITS_PER_LONG == 64
5866 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
5867 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
5869 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
5870 const unsigned long *src
= (const unsigned long *)netdev_stats
;
5871 u64
*dst
= (u64
*)stats64
;
5873 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
5874 sizeof(*stats64
) / sizeof(u64
));
5875 for (i
= 0; i
< n
; i
++)
5879 EXPORT_SYMBOL(netdev_stats_to_stats64
);
5882 * dev_get_stats - get network device statistics
5883 * @dev: device to get statistics from
5884 * @storage: place to store stats
5886 * Get network statistics from device. Return @storage.
5887 * The device driver may provide its own method by setting
5888 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5889 * otherwise the internal statistics structure is used.
5891 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
5892 struct rtnl_link_stats64
*storage
)
5894 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5896 if (ops
->ndo_get_stats64
) {
5897 memset(storage
, 0, sizeof(*storage
));
5898 ops
->ndo_get_stats64(dev
, storage
);
5899 } else if (ops
->ndo_get_stats
) {
5900 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
5902 netdev_stats_to_stats64(storage
, &dev
->stats
);
5904 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
5907 EXPORT_SYMBOL(dev_get_stats
);
5909 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
5911 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
5913 #ifdef CONFIG_NET_CLS_ACT
5916 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
5919 netdev_init_one_queue(dev
, queue
, NULL
);
5920 queue
->qdisc
= &noop_qdisc
;
5921 queue
->qdisc_sleeping
= &noop_qdisc
;
5922 rcu_assign_pointer(dev
->ingress_queue
, queue
);
5928 * alloc_netdev_mqs - allocate network device
5929 * @sizeof_priv: size of private data to allocate space for
5930 * @name: device name format string
5931 * @setup: callback to initialize device
5932 * @txqs: the number of TX subqueues to allocate
5933 * @rxqs: the number of RX subqueues to allocate
5935 * Allocates a struct net_device with private data area for driver use
5936 * and performs basic initialization. Also allocates subquue structs
5937 * for each queue on the device.
5939 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
5940 void (*setup
)(struct net_device
*),
5941 unsigned int txqs
, unsigned int rxqs
)
5943 struct net_device
*dev
;
5945 struct net_device
*p
;
5947 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
5950 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
5956 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
5961 alloc_size
= sizeof(struct net_device
);
5963 /* ensure 32-byte alignment of private area */
5964 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
5965 alloc_size
+= sizeof_priv
;
5967 /* ensure 32-byte alignment of whole construct */
5968 alloc_size
+= NETDEV_ALIGN
- 1;
5970 p
= kzalloc(alloc_size
, GFP_KERNEL
);
5972 pr_err("alloc_netdev: Unable to allocate device\n");
5976 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
5977 dev
->padded
= (char *)dev
- (char *)p
;
5979 dev
->pcpu_refcnt
= alloc_percpu(int);
5980 if (!dev
->pcpu_refcnt
)
5983 if (dev_addr_init(dev
))
5989 dev_net_set(dev
, &init_net
);
5991 dev
->gso_max_size
= GSO_MAX_SIZE
;
5992 dev
->gso_max_segs
= GSO_MAX_SEGS
;
5994 INIT_LIST_HEAD(&dev
->napi_list
);
5995 INIT_LIST_HEAD(&dev
->unreg_list
);
5996 INIT_LIST_HEAD(&dev
->link_watch_list
);
5997 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
6000 dev
->num_tx_queues
= txqs
;
6001 dev
->real_num_tx_queues
= txqs
;
6002 if (netif_alloc_netdev_queues(dev
))
6006 dev
->num_rx_queues
= rxqs
;
6007 dev
->real_num_rx_queues
= rxqs
;
6008 if (netif_alloc_rx_queues(dev
))
6012 strcpy(dev
->name
, name
);
6013 dev
->group
= INIT_NETDEV_GROUP
;
6021 free_percpu(dev
->pcpu_refcnt
);
6031 EXPORT_SYMBOL(alloc_netdev_mqs
);
6034 * free_netdev - free network device
6037 * This function does the last stage of destroying an allocated device
6038 * interface. The reference to the device object is released.
6039 * If this is the last reference then it will be freed.
6041 void free_netdev(struct net_device
*dev
)
6043 struct napi_struct
*p
, *n
;
6045 release_net(dev_net(dev
));
6052 kfree(rcu_dereference_protected(dev
->ingress_queue
, 1));
6054 /* Flush device addresses */
6055 dev_addr_flush(dev
);
6057 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
6060 free_percpu(dev
->pcpu_refcnt
);
6061 dev
->pcpu_refcnt
= NULL
;
6063 /* Compatibility with error handling in drivers */
6064 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
6065 kfree((char *)dev
- dev
->padded
);
6069 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
6070 dev
->reg_state
= NETREG_RELEASED
;
6072 /* will free via device release */
6073 put_device(&dev
->dev
);
6075 EXPORT_SYMBOL(free_netdev
);
6078 * synchronize_net - Synchronize with packet receive processing
6080 * Wait for packets currently being received to be done.
6081 * Does not block later packets from starting.
6083 void synchronize_net(void)
6086 if (rtnl_is_locked())
6087 synchronize_rcu_expedited();
6091 EXPORT_SYMBOL(synchronize_net
);
6094 * unregister_netdevice_queue - remove device from the kernel
6098 * This function shuts down a device interface and removes it
6099 * from the kernel tables.
6100 * If head not NULL, device is queued to be unregistered later.
6102 * Callers must hold the rtnl semaphore. You may want
6103 * unregister_netdev() instead of this.
6106 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
6111 list_move_tail(&dev
->unreg_list
, head
);
6113 rollback_registered(dev
);
6114 /* Finish processing unregister after unlock */
6118 EXPORT_SYMBOL(unregister_netdevice_queue
);
6121 * unregister_netdevice_many - unregister many devices
6122 * @head: list of devices
6124 void unregister_netdevice_many(struct list_head
*head
)
6126 struct net_device
*dev
;
6128 if (!list_empty(head
)) {
6129 rollback_registered_many(head
);
6130 list_for_each_entry(dev
, head
, unreg_list
)
6134 EXPORT_SYMBOL(unregister_netdevice_many
);
6137 * unregister_netdev - remove device from the kernel
6140 * This function shuts down a device interface and removes it
6141 * from the kernel tables.
6143 * This is just a wrapper for unregister_netdevice that takes
6144 * the rtnl semaphore. In general you want to use this and not
6145 * unregister_netdevice.
6147 void unregister_netdev(struct net_device
*dev
)
6150 unregister_netdevice(dev
);
6153 EXPORT_SYMBOL(unregister_netdev
);
6156 * dev_change_net_namespace - move device to different nethost namespace
6158 * @net: network namespace
6159 * @pat: If not NULL name pattern to try if the current device name
6160 * is already taken in the destination network namespace.
6162 * This function shuts down a device interface and moves it
6163 * to a new network namespace. On success 0 is returned, on
6164 * a failure a netagive errno code is returned.
6166 * Callers must hold the rtnl semaphore.
6169 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
6175 /* Don't allow namespace local devices to be moved. */
6177 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6180 /* Ensure the device has been registrered */
6182 if (dev
->reg_state
!= NETREG_REGISTERED
)
6185 /* Get out if there is nothing todo */
6187 if (net_eq(dev_net(dev
), net
))
6190 /* Pick the destination device name, and ensure
6191 * we can use it in the destination network namespace.
6194 if (__dev_get_by_name(net
, dev
->name
)) {
6195 /* We get here if we can't use the current device name */
6198 if (dev_get_valid_name(dev
, pat
) < 0)
6203 * And now a mini version of register_netdevice unregister_netdevice.
6206 /* If device is running close it first. */
6209 /* And unlink it from device chain */
6211 unlist_netdevice(dev
);
6215 /* Shutdown queueing discipline. */
6218 /* Notify protocols, that we are about to destroy
6219 this device. They should clean all the things.
6221 Note that dev->reg_state stays at NETREG_REGISTERED.
6222 This is wanted because this way 8021q and macvlan know
6223 the device is just moving and can keep their slaves up.
6225 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6226 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
6227 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
6230 * Flush the unicast and multicast chains
6235 /* Actually switch the network namespace */
6236 dev_net_set(dev
, net
);
6238 /* If there is an ifindex conflict assign a new one */
6239 if (__dev_get_by_index(net
, dev
->ifindex
)) {
6240 int iflink
= (dev
->iflink
== dev
->ifindex
);
6241 dev
->ifindex
= dev_new_index(net
);
6243 dev
->iflink
= dev
->ifindex
;
6246 /* Fixup kobjects */
6247 err
= device_rename(&dev
->dev
, dev
->name
);
6250 /* Add the device back in the hashes */
6251 list_netdevice(dev
);
6253 /* Notify protocols, that a new device appeared. */
6254 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
6257 * Prevent userspace races by waiting until the network
6258 * device is fully setup before sending notifications.
6260 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
6267 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
6269 static int dev_cpu_callback(struct notifier_block
*nfb
,
6270 unsigned long action
,
6273 struct sk_buff
**list_skb
;
6274 struct sk_buff
*skb
;
6275 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
6276 struct softnet_data
*sd
, *oldsd
;
6278 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
6281 local_irq_disable();
6282 cpu
= smp_processor_id();
6283 sd
= &per_cpu(softnet_data
, cpu
);
6284 oldsd
= &per_cpu(softnet_data
, oldcpu
);
6286 /* Find end of our completion_queue. */
6287 list_skb
= &sd
->completion_queue
;
6289 list_skb
= &(*list_skb
)->next
;
6290 /* Append completion queue from offline CPU. */
6291 *list_skb
= oldsd
->completion_queue
;
6292 oldsd
->completion_queue
= NULL
;
6294 /* Append output queue from offline CPU. */
6295 if (oldsd
->output_queue
) {
6296 *sd
->output_queue_tailp
= oldsd
->output_queue
;
6297 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
6298 oldsd
->output_queue
= NULL
;
6299 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
6301 /* Append NAPI poll list from offline CPU. */
6302 if (!list_empty(&oldsd
->poll_list
)) {
6303 list_splice_init(&oldsd
->poll_list
, &sd
->poll_list
);
6304 raise_softirq_irqoff(NET_RX_SOFTIRQ
);
6307 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
6310 /* Process offline CPU's input_pkt_queue */
6311 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
6313 input_queue_head_incr(oldsd
);
6315 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
6317 input_queue_head_incr(oldsd
);
6325 * netdev_increment_features - increment feature set by one
6326 * @all: current feature set
6327 * @one: new feature set
6328 * @mask: mask feature set
6330 * Computes a new feature set after adding a device with feature set
6331 * @one to the master device with current feature set @all. Will not
6332 * enable anything that is off in @mask. Returns the new feature set.
6334 netdev_features_t
netdev_increment_features(netdev_features_t all
,
6335 netdev_features_t one
, netdev_features_t mask
)
6337 if (mask
& NETIF_F_GEN_CSUM
)
6338 mask
|= NETIF_F_ALL_CSUM
;
6339 mask
|= NETIF_F_VLAN_CHALLENGED
;
6341 all
|= one
& (NETIF_F_ONE_FOR_ALL
|NETIF_F_ALL_CSUM
) & mask
;
6342 all
&= one
| ~NETIF_F_ALL_FOR_ALL
;
6344 /* If one device supports hw checksumming, set for all. */
6345 if (all
& NETIF_F_GEN_CSUM
)
6346 all
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_GEN_CSUM
);
6350 EXPORT_SYMBOL(netdev_increment_features
);
6352 static struct hlist_head
*netdev_create_hash(void)
6355 struct hlist_head
*hash
;
6357 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
6359 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
6360 INIT_HLIST_HEAD(&hash
[i
]);
6365 /* Initialize per network namespace state */
6366 static int __net_init
netdev_init(struct net
*net
)
6368 if (net
!= &init_net
)
6369 INIT_LIST_HEAD(&net
->dev_base_head
);
6371 net
->dev_name_head
= netdev_create_hash();
6372 if (net
->dev_name_head
== NULL
)
6375 net
->dev_index_head
= netdev_create_hash();
6376 if (net
->dev_index_head
== NULL
)
6382 kfree(net
->dev_name_head
);
6388 * netdev_drivername - network driver for the device
6389 * @dev: network device
6391 * Determine network driver for device.
6393 const char *netdev_drivername(const struct net_device
*dev
)
6395 const struct device_driver
*driver
;
6396 const struct device
*parent
;
6397 const char *empty
= "";
6399 parent
= dev
->dev
.parent
;
6403 driver
= parent
->driver
;
6404 if (driver
&& driver
->name
)
6405 return driver
->name
;
6409 int __netdev_printk(const char *level
, const struct net_device
*dev
,
6410 struct va_format
*vaf
)
6414 if (dev
&& dev
->dev
.parent
)
6415 r
= dev_printk(level
, dev
->dev
.parent
, "%s: %pV",
6416 netdev_name(dev
), vaf
);
6418 r
= printk("%s%s: %pV", level
, netdev_name(dev
), vaf
);
6420 r
= printk("%s(NULL net_device): %pV", level
, vaf
);
6424 EXPORT_SYMBOL(__netdev_printk
);
6426 int netdev_printk(const char *level
, const struct net_device
*dev
,
6427 const char *format
, ...)
6429 struct va_format vaf
;
6433 va_start(args
, format
);
6438 r
= __netdev_printk(level
, dev
, &vaf
);
6443 EXPORT_SYMBOL(netdev_printk
);
6445 #define define_netdev_printk_level(func, level) \
6446 int func(const struct net_device *dev, const char *fmt, ...) \
6449 struct va_format vaf; \
6452 va_start(args, fmt); \
6457 r = __netdev_printk(level, dev, &vaf); \
6462 EXPORT_SYMBOL(func);
6464 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
6465 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
6466 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
6467 define_netdev_printk_level(netdev_err
, KERN_ERR
);
6468 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
6469 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
6470 define_netdev_printk_level(netdev_info
, KERN_INFO
);
6472 static void __net_exit
netdev_exit(struct net
*net
)
6474 kfree(net
->dev_name_head
);
6475 kfree(net
->dev_index_head
);
6478 static struct pernet_operations __net_initdata netdev_net_ops
= {
6479 .init
= netdev_init
,
6480 .exit
= netdev_exit
,
6483 static void __net_exit
default_device_exit(struct net
*net
)
6485 struct net_device
*dev
, *aux
;
6487 * Push all migratable network devices back to the
6488 * initial network namespace
6491 for_each_netdev_safe(net
, dev
, aux
) {
6493 char fb_name
[IFNAMSIZ
];
6495 /* Ignore unmoveable devices (i.e. loopback) */
6496 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6499 /* Leave virtual devices for the generic cleanup */
6500 if (dev
->rtnl_link_ops
)
6503 /* Push remaining network devices to init_net */
6504 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
6505 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
6507 pr_emerg("%s: failed to move %s to init_net: %d\n",
6508 __func__
, dev
->name
, err
);
6515 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
6517 /* At exit all network devices most be removed from a network
6518 * namespace. Do this in the reverse order of registration.
6519 * Do this across as many network namespaces as possible to
6520 * improve batching efficiency.
6522 struct net_device
*dev
;
6524 LIST_HEAD(dev_kill_list
);
6527 list_for_each_entry(net
, net_list
, exit_list
) {
6528 for_each_netdev_reverse(net
, dev
) {
6529 if (dev
->rtnl_link_ops
)
6530 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
6532 unregister_netdevice_queue(dev
, &dev_kill_list
);
6535 unregister_netdevice_many(&dev_kill_list
);
6536 list_del(&dev_kill_list
);
6540 static struct pernet_operations __net_initdata default_device_ops
= {
6541 .exit
= default_device_exit
,
6542 .exit_batch
= default_device_exit_batch
,
6546 * Initialize the DEV module. At boot time this walks the device list and
6547 * unhooks any devices that fail to initialise (normally hardware not
6548 * present) and leaves us with a valid list of present and active devices.
6553 * This is called single threaded during boot, so no need
6554 * to take the rtnl semaphore.
6556 static int __init
net_dev_init(void)
6558 int i
, rc
= -ENOMEM
;
6560 BUG_ON(!dev_boot_phase
);
6562 if (dev_proc_init())
6565 if (netdev_kobject_init())
6568 INIT_LIST_HEAD(&ptype_all
);
6569 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
6570 INIT_LIST_HEAD(&ptype_base
[i
]);
6572 if (register_pernet_subsys(&netdev_net_ops
))
6576 * Initialise the packet receive queues.
6579 for_each_possible_cpu(i
) {
6580 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
6582 memset(sd
, 0, sizeof(*sd
));
6583 skb_queue_head_init(&sd
->input_pkt_queue
);
6584 skb_queue_head_init(&sd
->process_queue
);
6585 sd
->completion_queue
= NULL
;
6586 INIT_LIST_HEAD(&sd
->poll_list
);
6587 sd
->output_queue
= NULL
;
6588 sd
->output_queue_tailp
= &sd
->output_queue
;
6590 sd
->csd
.func
= rps_trigger_softirq
;
6596 sd
->backlog
.poll
= process_backlog
;
6597 sd
->backlog
.weight
= weight_p
;
6598 sd
->backlog
.gro_list
= NULL
;
6599 sd
->backlog
.gro_count
= 0;
6604 /* The loopback device is special if any other network devices
6605 * is present in a network namespace the loopback device must
6606 * be present. Since we now dynamically allocate and free the
6607 * loopback device ensure this invariant is maintained by
6608 * keeping the loopback device as the first device on the
6609 * list of network devices. Ensuring the loopback devices
6610 * is the first device that appears and the last network device
6613 if (register_pernet_device(&loopback_net_ops
))
6616 if (register_pernet_device(&default_device_ops
))
6619 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
6620 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
6622 hotcpu_notifier(dev_cpu_callback
, 0);
6630 subsys_initcall(net_dev_init
);
6632 static int __init
initialize_hashrnd(void)
6634 get_random_bytes(&hashrnd
, sizeof(hashrnd
));
6638 late_initcall_sync(initialize_hashrnd
);