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
);
1181 * dev_open - prepare an interface for use.
1182 * @dev: device to open
1184 * Takes a device from down to up state. The device's private open
1185 * function is invoked and then the multicast lists are loaded. Finally
1186 * the device is moved into the up state and a %NETDEV_UP message is
1187 * sent to the netdev notifier chain.
1189 * Calling this function on an active interface is a nop. On a failure
1190 * a negative errno code is returned.
1192 int dev_open(struct net_device
*dev
)
1196 if (dev
->flags
& IFF_UP
)
1199 ret
= __dev_open(dev
);
1203 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1204 call_netdevice_notifiers(NETDEV_UP
, dev
);
1208 EXPORT_SYMBOL(dev_open
);
1210 static int __dev_close_many(struct list_head
*head
)
1212 struct net_device
*dev
;
1217 list_for_each_entry(dev
, head
, unreg_list
) {
1218 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1220 clear_bit(__LINK_STATE_START
, &dev
->state
);
1222 /* Synchronize to scheduled poll. We cannot touch poll list, it
1223 * can be even on different cpu. So just clear netif_running().
1225 * dev->stop() will invoke napi_disable() on all of it's
1226 * napi_struct instances on this device.
1228 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1231 dev_deactivate_many(head
);
1233 list_for_each_entry(dev
, head
, unreg_list
) {
1234 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1237 * Call the device specific close. This cannot fail.
1238 * Only if device is UP
1240 * We allow it to be called even after a DETACH hot-plug
1246 dev
->flags
&= ~IFF_UP
;
1247 net_dmaengine_put();
1253 static int __dev_close(struct net_device
*dev
)
1258 list_add(&dev
->unreg_list
, &single
);
1259 retval
= __dev_close_many(&single
);
1264 static int dev_close_many(struct list_head
*head
)
1266 struct net_device
*dev
, *tmp
;
1267 LIST_HEAD(tmp_list
);
1269 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
)
1270 if (!(dev
->flags
& IFF_UP
))
1271 list_move(&dev
->unreg_list
, &tmp_list
);
1273 __dev_close_many(head
);
1275 list_for_each_entry(dev
, head
, unreg_list
) {
1276 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1277 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1280 /* rollback_registered_many needs the complete original list */
1281 list_splice(&tmp_list
, head
);
1286 * dev_close - shutdown an interface.
1287 * @dev: device to shutdown
1289 * This function moves an active device into down state. A
1290 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1291 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1294 int dev_close(struct net_device
*dev
)
1296 if (dev
->flags
& IFF_UP
) {
1299 list_add(&dev
->unreg_list
, &single
);
1300 dev_close_many(&single
);
1305 EXPORT_SYMBOL(dev_close
);
1309 * dev_disable_lro - disable Large Receive Offload on a device
1312 * Disable Large Receive Offload (LRO) on a net device. Must be
1313 * called under RTNL. This is needed if received packets may be
1314 * forwarded to another interface.
1316 void dev_disable_lro(struct net_device
*dev
)
1319 * If we're trying to disable lro on a vlan device
1320 * use the underlying physical device instead
1322 if (is_vlan_dev(dev
))
1323 dev
= vlan_dev_real_dev(dev
);
1325 dev
->wanted_features
&= ~NETIF_F_LRO
;
1326 netdev_update_features(dev
);
1328 if (unlikely(dev
->features
& NETIF_F_LRO
))
1329 netdev_WARN(dev
, "failed to disable LRO!\n");
1331 EXPORT_SYMBOL(dev_disable_lro
);
1334 static int dev_boot_phase
= 1;
1337 * register_netdevice_notifier - register a network notifier block
1340 * Register a notifier to be called when network device events occur.
1341 * The notifier passed is linked into the kernel structures and must
1342 * not be reused until it has been unregistered. A negative errno code
1343 * is returned on a failure.
1345 * When registered all registration and up events are replayed
1346 * to the new notifier to allow device to have a race free
1347 * view of the network device list.
1350 int register_netdevice_notifier(struct notifier_block
*nb
)
1352 struct net_device
*dev
;
1353 struct net_device
*last
;
1358 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1364 for_each_netdev(net
, dev
) {
1365 err
= nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1366 err
= notifier_to_errno(err
);
1370 if (!(dev
->flags
& IFF_UP
))
1373 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1384 for_each_netdev(net
, dev
) {
1388 if (dev
->flags
& IFF_UP
) {
1389 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1390 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1392 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1393 nb
->notifier_call(nb
, NETDEV_UNREGISTER_BATCH
, dev
);
1398 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1401 EXPORT_SYMBOL(register_netdevice_notifier
);
1404 * unregister_netdevice_notifier - unregister a network notifier block
1407 * Unregister a notifier previously registered by
1408 * register_netdevice_notifier(). The notifier is unlinked into the
1409 * kernel structures and may then be reused. A negative errno code
1410 * is returned on a failure.
1412 * After unregistering unregister and down device events are synthesized
1413 * for all devices on the device list to the removed notifier to remove
1414 * the need for special case cleanup code.
1417 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1419 struct net_device
*dev
;
1424 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1429 for_each_netdev(net
, dev
) {
1430 if (dev
->flags
& IFF_UP
) {
1431 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1432 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1434 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1435 nb
->notifier_call(nb
, NETDEV_UNREGISTER_BATCH
, dev
);
1442 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1445 * call_netdevice_notifiers - call all network notifier blocks
1446 * @val: value passed unmodified to notifier function
1447 * @dev: net_device pointer passed unmodified to notifier function
1449 * Call all network notifier blocks. Parameters and return value
1450 * are as for raw_notifier_call_chain().
1453 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1456 return raw_notifier_call_chain(&netdev_chain
, val
, dev
);
1458 EXPORT_SYMBOL(call_netdevice_notifiers
);
1460 static struct static_key netstamp_needed __read_mostly
;
1461 #ifdef HAVE_JUMP_LABEL
1462 /* We are not allowed to call static_key_slow_dec() from irq context
1463 * If net_disable_timestamp() is called from irq context, defer the
1464 * static_key_slow_dec() calls.
1466 static atomic_t netstamp_needed_deferred
;
1469 void net_enable_timestamp(void)
1471 #ifdef HAVE_JUMP_LABEL
1472 int deferred
= atomic_xchg(&netstamp_needed_deferred
, 0);
1476 static_key_slow_dec(&netstamp_needed
);
1480 WARN_ON(in_interrupt());
1481 static_key_slow_inc(&netstamp_needed
);
1483 EXPORT_SYMBOL(net_enable_timestamp
);
1485 void net_disable_timestamp(void)
1487 #ifdef HAVE_JUMP_LABEL
1488 if (in_interrupt()) {
1489 atomic_inc(&netstamp_needed_deferred
);
1493 static_key_slow_dec(&netstamp_needed
);
1495 EXPORT_SYMBOL(net_disable_timestamp
);
1497 static inline void net_timestamp_set(struct sk_buff
*skb
)
1499 skb
->tstamp
.tv64
= 0;
1500 if (static_key_false(&netstamp_needed
))
1501 __net_timestamp(skb
);
1504 #define net_timestamp_check(COND, SKB) \
1505 if (static_key_false(&netstamp_needed)) { \
1506 if ((COND) && !(SKB)->tstamp.tv64) \
1507 __net_timestamp(SKB); \
1510 static int net_hwtstamp_validate(struct ifreq *ifr)
1512 struct hwtstamp_config cfg
;
1513 enum hwtstamp_tx_types tx_type
;
1514 enum hwtstamp_rx_filters rx_filter
;
1515 int tx_type_valid
= 0;
1516 int rx_filter_valid
= 0;
1518 if (copy_from_user(&cfg
, ifr
->ifr_data
, sizeof(cfg
)))
1521 if (cfg
.flags
) /* reserved for future extensions */
1524 tx_type
= cfg
.tx_type
;
1525 rx_filter
= cfg
.rx_filter
;
1528 case HWTSTAMP_TX_OFF
:
1529 case HWTSTAMP_TX_ON
:
1530 case HWTSTAMP_TX_ONESTEP_SYNC
:
1535 switch (rx_filter
) {
1536 case HWTSTAMP_FILTER_NONE
:
1537 case HWTSTAMP_FILTER_ALL
:
1538 case HWTSTAMP_FILTER_SOME
:
1539 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT
:
1540 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC
:
1541 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ
:
1542 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT
:
1543 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC
:
1544 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ
:
1545 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT
:
1546 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC
:
1547 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ
:
1548 case HWTSTAMP_FILTER_PTP_V2_EVENT
:
1549 case HWTSTAMP_FILTER_PTP_V2_SYNC
:
1550 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ
:
1551 rx_filter_valid
= 1;
1555 if (!tx_type_valid
|| !rx_filter_valid
)
1561 static inline bool is_skb_forwardable(struct net_device
*dev
,
1562 struct sk_buff
*skb
)
1566 if (!(dev
->flags
& IFF_UP
))
1569 len
= dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
;
1570 if (skb
->len
<= len
)
1573 /* if TSO is enabled, we don't care about the length as the packet
1574 * could be forwarded without being segmented before
1576 if (skb_is_gso(skb
))
1583 * dev_forward_skb - loopback an skb to another netif
1585 * @dev: destination network device
1586 * @skb: buffer to forward
1589 * NET_RX_SUCCESS (no congestion)
1590 * NET_RX_DROP (packet was dropped, but freed)
1592 * dev_forward_skb can be used for injecting an skb from the
1593 * start_xmit function of one device into the receive queue
1594 * of another device.
1596 * The receiving device may be in another namespace, so
1597 * we have to clear all information in the skb that could
1598 * impact namespace isolation.
1600 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1602 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
1603 if (skb_copy_ubufs(skb
, GFP_ATOMIC
)) {
1604 atomic_long_inc(&dev
->rx_dropped
);
1613 if (unlikely(!is_skb_forwardable(dev
, skb
))) {
1614 atomic_long_inc(&dev
->rx_dropped
);
1621 skb
->tstamp
.tv64
= 0;
1622 skb
->pkt_type
= PACKET_HOST
;
1623 skb
->protocol
= eth_type_trans(skb
, dev
);
1627 return netif_rx(skb
);
1629 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1631 static inline int deliver_skb(struct sk_buff
*skb
,
1632 struct packet_type
*pt_prev
,
1633 struct net_device
*orig_dev
)
1635 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
1637 atomic_inc(&skb
->users
);
1638 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1642 * Support routine. Sends outgoing frames to any network
1643 * taps currently in use.
1646 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1648 struct packet_type
*ptype
;
1649 struct sk_buff
*skb2
= NULL
;
1650 struct packet_type
*pt_prev
= NULL
;
1653 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1654 /* Never send packets back to the socket
1655 * they originated from - MvS (miquels@drinkel.ow.org)
1657 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1658 (ptype
->af_packet_priv
== NULL
||
1659 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1661 deliver_skb(skb2
, pt_prev
, skb
->dev
);
1666 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1670 net_timestamp_set(skb2
);
1672 /* skb->nh should be correctly
1673 set by sender, so that the second statement is
1674 just protection against buggy protocols.
1676 skb_reset_mac_header(skb2
);
1678 if (skb_network_header(skb2
) < skb2
->data
||
1679 skb2
->network_header
> skb2
->tail
) {
1680 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1681 ntohs(skb2
->protocol
),
1683 skb_reset_network_header(skb2
);
1686 skb2
->transport_header
= skb2
->network_header
;
1687 skb2
->pkt_type
= PACKET_OUTGOING
;
1692 pt_prev
->func(skb2
, skb
->dev
, pt_prev
, skb
->dev
);
1697 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1698 * @dev: Network device
1699 * @txq: number of queues available
1701 * If real_num_tx_queues is changed the tc mappings may no longer be
1702 * valid. To resolve this verify the tc mapping remains valid and if
1703 * not NULL the mapping. With no priorities mapping to this
1704 * offset/count pair it will no longer be used. In the worst case TC0
1705 * is invalid nothing can be done so disable priority mappings. If is
1706 * expected that drivers will fix this mapping if they can before
1707 * calling netif_set_real_num_tx_queues.
1709 static void netif_setup_tc(struct net_device
*dev
, unsigned int txq
)
1712 struct netdev_tc_txq
*tc
= &dev
->tc_to_txq
[0];
1714 /* If TC0 is invalidated disable TC mapping */
1715 if (tc
->offset
+ tc
->count
> txq
) {
1716 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1721 /* Invalidated prio to tc mappings set to TC0 */
1722 for (i
= 1; i
< TC_BITMASK
+ 1; i
++) {
1723 int q
= netdev_get_prio_tc_map(dev
, i
);
1725 tc
= &dev
->tc_to_txq
[q
];
1726 if (tc
->offset
+ tc
->count
> txq
) {
1727 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1729 netdev_set_prio_tc_map(dev
, i
, 0);
1735 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1736 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1738 int netif_set_real_num_tx_queues(struct net_device
*dev
, unsigned int txq
)
1742 if (txq
< 1 || txq
> dev
->num_tx_queues
)
1745 if (dev
->reg_state
== NETREG_REGISTERED
||
1746 dev
->reg_state
== NETREG_UNREGISTERING
) {
1749 rc
= netdev_queue_update_kobjects(dev
, dev
->real_num_tx_queues
,
1755 netif_setup_tc(dev
, txq
);
1757 if (txq
< dev
->real_num_tx_queues
)
1758 qdisc_reset_all_tx_gt(dev
, txq
);
1761 dev
->real_num_tx_queues
= txq
;
1764 EXPORT_SYMBOL(netif_set_real_num_tx_queues
);
1768 * netif_set_real_num_rx_queues - set actual number of RX queues used
1769 * @dev: Network device
1770 * @rxq: Actual number of RX queues
1772 * This must be called either with the rtnl_lock held or before
1773 * registration of the net device. Returns 0 on success, or a
1774 * negative error code. If called before registration, it always
1777 int netif_set_real_num_rx_queues(struct net_device
*dev
, unsigned int rxq
)
1781 if (rxq
< 1 || rxq
> dev
->num_rx_queues
)
1784 if (dev
->reg_state
== NETREG_REGISTERED
) {
1787 rc
= net_rx_queue_update_kobjects(dev
, dev
->real_num_rx_queues
,
1793 dev
->real_num_rx_queues
= rxq
;
1796 EXPORT_SYMBOL(netif_set_real_num_rx_queues
);
1800 * netif_get_num_default_rss_queues - default number of RSS queues
1802 * This routine should set an upper limit on the number of RSS queues
1803 * used by default by multiqueue devices.
1805 int netif_get_num_default_rss_queues(void)
1807 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES
, num_online_cpus());
1809 EXPORT_SYMBOL(netif_get_num_default_rss_queues
);
1811 static inline void __netif_reschedule(struct Qdisc
*q
)
1813 struct softnet_data
*sd
;
1814 unsigned long flags
;
1816 local_irq_save(flags
);
1817 sd
= &__get_cpu_var(softnet_data
);
1818 q
->next_sched
= NULL
;
1819 *sd
->output_queue_tailp
= q
;
1820 sd
->output_queue_tailp
= &q
->next_sched
;
1821 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1822 local_irq_restore(flags
);
1825 void __netif_schedule(struct Qdisc
*q
)
1827 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
1828 __netif_reschedule(q
);
1830 EXPORT_SYMBOL(__netif_schedule
);
1832 void dev_kfree_skb_irq(struct sk_buff
*skb
)
1834 if (atomic_dec_and_test(&skb
->users
)) {
1835 struct softnet_data
*sd
;
1836 unsigned long flags
;
1838 local_irq_save(flags
);
1839 sd
= &__get_cpu_var(softnet_data
);
1840 skb
->next
= sd
->completion_queue
;
1841 sd
->completion_queue
= skb
;
1842 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1843 local_irq_restore(flags
);
1846 EXPORT_SYMBOL(dev_kfree_skb_irq
);
1848 void dev_kfree_skb_any(struct sk_buff
*skb
)
1850 if (in_irq() || irqs_disabled())
1851 dev_kfree_skb_irq(skb
);
1855 EXPORT_SYMBOL(dev_kfree_skb_any
);
1859 * netif_device_detach - mark device as removed
1860 * @dev: network device
1862 * Mark device as removed from system and therefore no longer available.
1864 void netif_device_detach(struct net_device
*dev
)
1866 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1867 netif_running(dev
)) {
1868 netif_tx_stop_all_queues(dev
);
1871 EXPORT_SYMBOL(netif_device_detach
);
1874 * netif_device_attach - mark device as attached
1875 * @dev: network device
1877 * Mark device as attached from system and restart if needed.
1879 void netif_device_attach(struct net_device
*dev
)
1881 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1882 netif_running(dev
)) {
1883 netif_tx_wake_all_queues(dev
);
1884 __netdev_watchdog_up(dev
);
1887 EXPORT_SYMBOL(netif_device_attach
);
1889 static void skb_warn_bad_offload(const struct sk_buff
*skb
)
1891 static const netdev_features_t null_features
= 0;
1892 struct net_device
*dev
= skb
->dev
;
1893 const char *driver
= "";
1895 if (dev
&& dev
->dev
.parent
)
1896 driver
= dev_driver_string(dev
->dev
.parent
);
1898 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
1899 "gso_type=%d ip_summed=%d\n",
1900 driver
, dev
? &dev
->features
: &null_features
,
1901 skb
->sk
? &skb
->sk
->sk_route_caps
: &null_features
,
1902 skb
->len
, skb
->data_len
, skb_shinfo(skb
)->gso_size
,
1903 skb_shinfo(skb
)->gso_type
, skb
->ip_summed
);
1907 * Invalidate hardware checksum when packet is to be mangled, and
1908 * complete checksum manually on outgoing path.
1910 int skb_checksum_help(struct sk_buff
*skb
)
1913 int ret
= 0, offset
;
1915 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1916 goto out_set_summed
;
1918 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1919 skb_warn_bad_offload(skb
);
1923 offset
= skb_checksum_start_offset(skb
);
1924 BUG_ON(offset
>= skb_headlen(skb
));
1925 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1927 offset
+= skb
->csum_offset
;
1928 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
1930 if (skb_cloned(skb
) &&
1931 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
1932 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1937 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1939 skb
->ip_summed
= CHECKSUM_NONE
;
1943 EXPORT_SYMBOL(skb_checksum_help
);
1946 * skb_gso_segment - Perform segmentation on skb.
1947 * @skb: buffer to segment
1948 * @features: features for the output path (see dev->features)
1950 * This function segments the given skb and returns a list of segments.
1952 * It may return NULL if the skb requires no segmentation. This is
1953 * only possible when GSO is used for verifying header integrity.
1955 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
,
1956 netdev_features_t features
)
1958 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1959 struct packet_type
*ptype
;
1960 __be16 type
= skb
->protocol
;
1961 int vlan_depth
= ETH_HLEN
;
1964 while (type
== htons(ETH_P_8021Q
)) {
1965 struct vlan_hdr
*vh
;
1967 if (unlikely(!pskb_may_pull(skb
, vlan_depth
+ VLAN_HLEN
)))
1968 return ERR_PTR(-EINVAL
);
1970 vh
= (struct vlan_hdr
*)(skb
->data
+ vlan_depth
);
1971 type
= vh
->h_vlan_encapsulated_proto
;
1972 vlan_depth
+= VLAN_HLEN
;
1975 skb_reset_mac_header(skb
);
1976 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1977 __skb_pull(skb
, skb
->mac_len
);
1979 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1980 skb_warn_bad_offload(skb
);
1982 if (skb_header_cloned(skb
) &&
1983 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
1984 return ERR_PTR(err
);
1988 list_for_each_entry_rcu(ptype
,
1989 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
1990 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
1991 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1992 err
= ptype
->gso_send_check(skb
);
1993 segs
= ERR_PTR(err
);
1994 if (err
|| skb_gso_ok(skb
, features
))
1996 __skb_push(skb
, (skb
->data
-
1997 skb_network_header(skb
)));
1999 segs
= ptype
->gso_segment(skb
, features
);
2005 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
2009 EXPORT_SYMBOL(skb_gso_segment
);
2011 /* Take action when hardware reception checksum errors are detected. */
2013 void netdev_rx_csum_fault(struct net_device
*dev
)
2015 if (net_ratelimit()) {
2016 pr_err("%s: hw csum failure\n", dev
? dev
->name
: "<unknown>");
2020 EXPORT_SYMBOL(netdev_rx_csum_fault
);
2023 /* Actually, we should eliminate this check as soon as we know, that:
2024 * 1. IOMMU is present and allows to map all the memory.
2025 * 2. No high memory really exists on this machine.
2028 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
2030 #ifdef CONFIG_HIGHMEM
2032 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
2033 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2034 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2035 if (PageHighMem(skb_frag_page(frag
)))
2040 if (PCI_DMA_BUS_IS_PHYS
) {
2041 struct device
*pdev
= dev
->dev
.parent
;
2045 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2046 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2047 dma_addr_t addr
= page_to_phys(skb_frag_page(frag
));
2048 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
2057 void (*destructor
)(struct sk_buff
*skb
);
2060 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2062 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
2064 struct dev_gso_cb
*cb
;
2067 struct sk_buff
*nskb
= skb
->next
;
2069 skb
->next
= nskb
->next
;
2072 } while (skb
->next
);
2074 cb
= DEV_GSO_CB(skb
);
2076 cb
->destructor(skb
);
2080 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2081 * @skb: buffer to segment
2082 * @features: device features as applicable to this skb
2084 * This function segments the given skb and stores the list of segments
2087 static int dev_gso_segment(struct sk_buff
*skb
, netdev_features_t features
)
2089 struct sk_buff
*segs
;
2091 segs
= skb_gso_segment(skb
, features
);
2093 /* Verifying header integrity only. */
2098 return PTR_ERR(segs
);
2101 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
2102 skb
->destructor
= dev_gso_skb_destructor
;
2107 static bool can_checksum_protocol(netdev_features_t features
, __be16 protocol
)
2109 return ((features
& NETIF_F_GEN_CSUM
) ||
2110 ((features
& NETIF_F_V4_CSUM
) &&
2111 protocol
== htons(ETH_P_IP
)) ||
2112 ((features
& NETIF_F_V6_CSUM
) &&
2113 protocol
== htons(ETH_P_IPV6
)) ||
2114 ((features
& NETIF_F_FCOE_CRC
) &&
2115 protocol
== htons(ETH_P_FCOE
)));
2118 static netdev_features_t
harmonize_features(struct sk_buff
*skb
,
2119 __be16 protocol
, netdev_features_t features
)
2121 if (!can_checksum_protocol(features
, protocol
)) {
2122 features
&= ~NETIF_F_ALL_CSUM
;
2123 features
&= ~NETIF_F_SG
;
2124 } else if (illegal_highdma(skb
->dev
, skb
)) {
2125 features
&= ~NETIF_F_SG
;
2131 netdev_features_t
netif_skb_features(struct sk_buff
*skb
)
2133 __be16 protocol
= skb
->protocol
;
2134 netdev_features_t features
= skb
->dev
->features
;
2136 if (protocol
== htons(ETH_P_8021Q
)) {
2137 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
2138 protocol
= veh
->h_vlan_encapsulated_proto
;
2139 } else if (!vlan_tx_tag_present(skb
)) {
2140 return harmonize_features(skb
, protocol
, features
);
2143 features
&= (skb
->dev
->vlan_features
| NETIF_F_HW_VLAN_TX
);
2145 if (protocol
!= htons(ETH_P_8021Q
)) {
2146 return harmonize_features(skb
, protocol
, features
);
2148 features
&= NETIF_F_SG
| NETIF_F_HIGHDMA
| NETIF_F_FRAGLIST
|
2149 NETIF_F_GEN_CSUM
| NETIF_F_HW_VLAN_TX
;
2150 return harmonize_features(skb
, protocol
, features
);
2153 EXPORT_SYMBOL(netif_skb_features
);
2156 * Returns true if either:
2157 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2158 * 2. skb is fragmented and the device does not support SG, or if
2159 * at least one of fragments is in highmem and device does not
2160 * support DMA from it.
2162 static inline int skb_needs_linearize(struct sk_buff
*skb
,
2165 return skb_is_nonlinear(skb
) &&
2166 ((skb_has_frag_list(skb
) &&
2167 !(features
& NETIF_F_FRAGLIST
)) ||
2168 (skb_shinfo(skb
)->nr_frags
&&
2169 !(features
& NETIF_F_SG
)));
2172 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
2173 struct netdev_queue
*txq
)
2175 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2176 int rc
= NETDEV_TX_OK
;
2177 unsigned int skb_len
;
2179 if (likely(!skb
->next
)) {
2180 netdev_features_t features
;
2183 * If device doesn't need skb->dst, release it right now while
2184 * its hot in this cpu cache
2186 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2189 if (!list_empty(&ptype_all
))
2190 dev_queue_xmit_nit(skb
, dev
);
2192 features
= netif_skb_features(skb
);
2194 if (vlan_tx_tag_present(skb
) &&
2195 !(features
& NETIF_F_HW_VLAN_TX
)) {
2196 skb
= __vlan_put_tag(skb
, vlan_tx_tag_get(skb
));
2203 if (netif_needs_gso(skb
, features
)) {
2204 if (unlikely(dev_gso_segment(skb
, features
)))
2209 if (skb_needs_linearize(skb
, features
) &&
2210 __skb_linearize(skb
))
2213 /* If packet is not checksummed and device does not
2214 * support checksumming for this protocol, complete
2215 * checksumming here.
2217 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2218 skb_set_transport_header(skb
,
2219 skb_checksum_start_offset(skb
));
2220 if (!(features
& NETIF_F_ALL_CSUM
) &&
2221 skb_checksum_help(skb
))
2227 rc
= ops
->ndo_start_xmit(skb
, dev
);
2228 trace_net_dev_xmit(skb
, rc
, dev
, skb_len
);
2229 if (rc
== NETDEV_TX_OK
)
2230 txq_trans_update(txq
);
2236 struct sk_buff
*nskb
= skb
->next
;
2238 skb
->next
= nskb
->next
;
2242 * If device doesn't need nskb->dst, release it right now while
2243 * its hot in this cpu cache
2245 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2248 skb_len
= nskb
->len
;
2249 rc
= ops
->ndo_start_xmit(nskb
, dev
);
2250 trace_net_dev_xmit(nskb
, rc
, dev
, skb_len
);
2251 if (unlikely(rc
!= NETDEV_TX_OK
)) {
2252 if (rc
& ~NETDEV_TX_MASK
)
2253 goto out_kfree_gso_skb
;
2254 nskb
->next
= skb
->next
;
2258 txq_trans_update(txq
);
2259 if (unlikely(netif_xmit_stopped(txq
) && skb
->next
))
2260 return NETDEV_TX_BUSY
;
2261 } while (skb
->next
);
2264 if (likely(skb
->next
== NULL
))
2265 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
2272 static u32 hashrnd __read_mostly
;
2275 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2276 * to be used as a distribution range.
2278 u16
__skb_tx_hash(const struct net_device
*dev
, const struct sk_buff
*skb
,
2279 unsigned int num_tx_queues
)
2283 u16 qcount
= num_tx_queues
;
2285 if (skb_rx_queue_recorded(skb
)) {
2286 hash
= skb_get_rx_queue(skb
);
2287 while (unlikely(hash
>= num_tx_queues
))
2288 hash
-= num_tx_queues
;
2293 u8 tc
= netdev_get_prio_tc_map(dev
, skb
->priority
);
2294 qoffset
= dev
->tc_to_txq
[tc
].offset
;
2295 qcount
= dev
->tc_to_txq
[tc
].count
;
2298 if (skb
->sk
&& skb
->sk
->sk_hash
)
2299 hash
= skb
->sk
->sk_hash
;
2301 hash
= (__force u16
) skb
->protocol
;
2302 hash
= jhash_1word(hash
, hashrnd
);
2304 return (u16
) (((u64
) hash
* qcount
) >> 32) + qoffset
;
2306 EXPORT_SYMBOL(__skb_tx_hash
);
2308 static inline u16
dev_cap_txqueue(struct net_device
*dev
, u16 queue_index
)
2310 if (unlikely(queue_index
>= dev
->real_num_tx_queues
)) {
2311 net_warn_ratelimited("%s selects TX queue %d, but real number of TX queues is %d\n",
2312 dev
->name
, queue_index
,
2313 dev
->real_num_tx_queues
);
2319 static inline int get_xps_queue(struct net_device
*dev
, struct sk_buff
*skb
)
2322 struct xps_dev_maps
*dev_maps
;
2323 struct xps_map
*map
;
2324 int queue_index
= -1;
2327 dev_maps
= rcu_dereference(dev
->xps_maps
);
2329 map
= rcu_dereference(
2330 dev_maps
->cpu_map
[raw_smp_processor_id()]);
2333 queue_index
= map
->queues
[0];
2336 if (skb
->sk
&& skb
->sk
->sk_hash
)
2337 hash
= skb
->sk
->sk_hash
;
2339 hash
= (__force u16
) skb
->protocol
^
2341 hash
= jhash_1word(hash
, hashrnd
);
2342 queue_index
= map
->queues
[
2343 ((u64
)hash
* map
->len
) >> 32];
2345 if (unlikely(queue_index
>= dev
->real_num_tx_queues
))
2357 static struct netdev_queue
*dev_pick_tx(struct net_device
*dev
,
2358 struct sk_buff
*skb
)
2361 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2363 if (dev
->real_num_tx_queues
== 1)
2365 else if (ops
->ndo_select_queue
) {
2366 queue_index
= ops
->ndo_select_queue(dev
, skb
);
2367 queue_index
= dev_cap_txqueue(dev
, queue_index
);
2369 struct sock
*sk
= skb
->sk
;
2370 queue_index
= sk_tx_queue_get(sk
);
2372 if (queue_index
< 0 || skb
->ooo_okay
||
2373 queue_index
>= dev
->real_num_tx_queues
) {
2374 int old_index
= queue_index
;
2376 queue_index
= get_xps_queue(dev
, skb
);
2377 if (queue_index
< 0)
2378 queue_index
= skb_tx_hash(dev
, skb
);
2380 if (queue_index
!= old_index
&& sk
) {
2381 struct dst_entry
*dst
=
2382 rcu_dereference_check(sk
->sk_dst_cache
, 1);
2384 if (dst
&& skb_dst(skb
) == dst
)
2385 sk_tx_queue_set(sk
, queue_index
);
2390 skb_set_queue_mapping(skb
, queue_index
);
2391 return netdev_get_tx_queue(dev
, queue_index
);
2394 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2395 struct net_device
*dev
,
2396 struct netdev_queue
*txq
)
2398 spinlock_t
*root_lock
= qdisc_lock(q
);
2402 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
2403 qdisc_calculate_pkt_len(skb
, q
);
2405 * Heuristic to force contended enqueues to serialize on a
2406 * separate lock before trying to get qdisc main lock.
2407 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2408 * and dequeue packets faster.
2410 contended
= qdisc_is_running(q
);
2411 if (unlikely(contended
))
2412 spin_lock(&q
->busylock
);
2414 spin_lock(root_lock
);
2415 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2418 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2419 qdisc_run_begin(q
)) {
2421 * This is a work-conserving queue; there are no old skbs
2422 * waiting to be sent out; and the qdisc is not running -
2423 * xmit the skb directly.
2425 if (!(dev
->priv_flags
& IFF_XMIT_DST_RELEASE
))
2428 qdisc_bstats_update(q
, skb
);
2430 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
)) {
2431 if (unlikely(contended
)) {
2432 spin_unlock(&q
->busylock
);
2439 rc
= NET_XMIT_SUCCESS
;
2442 rc
= q
->enqueue(skb
, q
) & NET_XMIT_MASK
;
2443 if (qdisc_run_begin(q
)) {
2444 if (unlikely(contended
)) {
2445 spin_unlock(&q
->busylock
);
2451 spin_unlock(root_lock
);
2452 if (unlikely(contended
))
2453 spin_unlock(&q
->busylock
);
2457 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2458 static void skb_update_prio(struct sk_buff
*skb
)
2460 struct netprio_map
*map
= rcu_dereference_bh(skb
->dev
->priomap
);
2462 if (!skb
->priority
&& skb
->sk
&& map
) {
2463 unsigned int prioidx
= skb
->sk
->sk_cgrp_prioidx
;
2465 if (prioidx
< map
->priomap_len
)
2466 skb
->priority
= map
->priomap
[prioidx
];
2470 #define skb_update_prio(skb)
2473 static DEFINE_PER_CPU(int, xmit_recursion
);
2474 #define RECURSION_LIMIT 10
2477 * dev_loopback_xmit - loop back @skb
2478 * @skb: buffer to transmit
2480 int dev_loopback_xmit(struct sk_buff
*skb
)
2482 skb_reset_mac_header(skb
);
2483 __skb_pull(skb
, skb_network_offset(skb
));
2484 skb
->pkt_type
= PACKET_LOOPBACK
;
2485 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
2486 WARN_ON(!skb_dst(skb
));
2491 EXPORT_SYMBOL(dev_loopback_xmit
);
2494 * dev_queue_xmit - transmit a buffer
2495 * @skb: buffer to transmit
2497 * Queue a buffer for transmission to a network device. The caller must
2498 * have set the device and priority and built the buffer before calling
2499 * this function. The function can be called from an interrupt.
2501 * A negative errno code is returned on a failure. A success does not
2502 * guarantee the frame will be transmitted as it may be dropped due
2503 * to congestion or traffic shaping.
2505 * -----------------------------------------------------------------------------------
2506 * I notice this method can also return errors from the queue disciplines,
2507 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2510 * Regardless of the return value, the skb is consumed, so it is currently
2511 * difficult to retry a send to this method. (You can bump the ref count
2512 * before sending to hold a reference for retry if you are careful.)
2514 * When calling this method, interrupts MUST be enabled. This is because
2515 * the BH enable code must have IRQs enabled so that it will not deadlock.
2518 int dev_queue_xmit(struct sk_buff
*skb
)
2520 struct net_device
*dev
= skb
->dev
;
2521 struct netdev_queue
*txq
;
2525 /* Disable soft irqs for various locks below. Also
2526 * stops preemption for RCU.
2530 skb_update_prio(skb
);
2532 txq
= dev_pick_tx(dev
, skb
);
2533 q
= rcu_dereference_bh(txq
->qdisc
);
2535 #ifdef CONFIG_NET_CLS_ACT
2536 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2538 trace_net_dev_queue(skb
);
2540 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2544 /* The device has no queue. Common case for software devices:
2545 loopback, all the sorts of tunnels...
2547 Really, it is unlikely that netif_tx_lock protection is necessary
2548 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2550 However, it is possible, that they rely on protection
2553 Check this and shot the lock. It is not prone from deadlocks.
2554 Either shot noqueue qdisc, it is even simpler 8)
2556 if (dev
->flags
& IFF_UP
) {
2557 int cpu
= smp_processor_id(); /* ok because BHs are off */
2559 if (txq
->xmit_lock_owner
!= cpu
) {
2561 if (__this_cpu_read(xmit_recursion
) > RECURSION_LIMIT
)
2562 goto recursion_alert
;
2564 HARD_TX_LOCK(dev
, txq
, cpu
);
2566 if (!netif_xmit_stopped(txq
)) {
2567 __this_cpu_inc(xmit_recursion
);
2568 rc
= dev_hard_start_xmit(skb
, dev
, txq
);
2569 __this_cpu_dec(xmit_recursion
);
2570 if (dev_xmit_complete(rc
)) {
2571 HARD_TX_UNLOCK(dev
, txq
);
2575 HARD_TX_UNLOCK(dev
, txq
);
2576 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2579 /* Recursion is detected! It is possible,
2583 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2589 rcu_read_unlock_bh();
2594 rcu_read_unlock_bh();
2597 EXPORT_SYMBOL(dev_queue_xmit
);
2600 /*=======================================================================
2602 =======================================================================*/
2604 int netdev_max_backlog __read_mostly
= 1000;
2605 int netdev_tstamp_prequeue __read_mostly
= 1;
2606 int netdev_budget __read_mostly
= 300;
2607 int weight_p __read_mostly
= 64; /* old backlog weight */
2609 /* Called with irq disabled */
2610 static inline void ____napi_schedule(struct softnet_data
*sd
,
2611 struct napi_struct
*napi
)
2613 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
2614 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2618 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2619 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2620 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2621 * if hash is a canonical 4-tuple hash over transport ports.
2623 void __skb_get_rxhash(struct sk_buff
*skb
)
2625 struct flow_keys keys
;
2628 if (!skb_flow_dissect(skb
, &keys
))
2632 if ((__force u16
)keys
.port16
[1] < (__force u16
)keys
.port16
[0])
2633 swap(keys
.port16
[0], keys
.port16
[1]);
2637 /* get a consistent hash (same value on both flow directions) */
2638 if ((__force u32
)keys
.dst
< (__force u32
)keys
.src
)
2639 swap(keys
.dst
, keys
.src
);
2641 hash
= jhash_3words((__force u32
)keys
.dst
,
2642 (__force u32
)keys
.src
,
2643 (__force u32
)keys
.ports
, hashrnd
);
2649 EXPORT_SYMBOL(__skb_get_rxhash
);
2653 /* One global table that all flow-based protocols share. */
2654 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
2655 EXPORT_SYMBOL(rps_sock_flow_table
);
2657 struct static_key rps_needed __read_mostly
;
2659 static struct rps_dev_flow
*
2660 set_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2661 struct rps_dev_flow
*rflow
, u16 next_cpu
)
2663 if (next_cpu
!= RPS_NO_CPU
) {
2664 #ifdef CONFIG_RFS_ACCEL
2665 struct netdev_rx_queue
*rxqueue
;
2666 struct rps_dev_flow_table
*flow_table
;
2667 struct rps_dev_flow
*old_rflow
;
2672 /* Should we steer this flow to a different hardware queue? */
2673 if (!skb_rx_queue_recorded(skb
) || !dev
->rx_cpu_rmap
||
2674 !(dev
->features
& NETIF_F_NTUPLE
))
2676 rxq_index
= cpu_rmap_lookup_index(dev
->rx_cpu_rmap
, next_cpu
);
2677 if (rxq_index
== skb_get_rx_queue(skb
))
2680 rxqueue
= dev
->_rx
+ rxq_index
;
2681 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2684 flow_id
= skb
->rxhash
& flow_table
->mask
;
2685 rc
= dev
->netdev_ops
->ndo_rx_flow_steer(dev
, skb
,
2686 rxq_index
, flow_id
);
2690 rflow
= &flow_table
->flows
[flow_id
];
2692 if (old_rflow
->filter
== rflow
->filter
)
2693 old_rflow
->filter
= RPS_NO_FILTER
;
2697 per_cpu(softnet_data
, next_cpu
).input_queue_head
;
2700 rflow
->cpu
= next_cpu
;
2705 * get_rps_cpu is called from netif_receive_skb and returns the target
2706 * CPU from the RPS map of the receiving queue for a given skb.
2707 * rcu_read_lock must be held on entry.
2709 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2710 struct rps_dev_flow
**rflowp
)
2712 struct netdev_rx_queue
*rxqueue
;
2713 struct rps_map
*map
;
2714 struct rps_dev_flow_table
*flow_table
;
2715 struct rps_sock_flow_table
*sock_flow_table
;
2719 if (skb_rx_queue_recorded(skb
)) {
2720 u16 index
= skb_get_rx_queue(skb
);
2721 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
2722 WARN_ONCE(dev
->real_num_rx_queues
> 1,
2723 "%s received packet on queue %u, but number "
2724 "of RX queues is %u\n",
2725 dev
->name
, index
, dev
->real_num_rx_queues
);
2728 rxqueue
= dev
->_rx
+ index
;
2732 map
= rcu_dereference(rxqueue
->rps_map
);
2734 if (map
->len
== 1 &&
2735 !rcu_access_pointer(rxqueue
->rps_flow_table
)) {
2736 tcpu
= map
->cpus
[0];
2737 if (cpu_online(tcpu
))
2741 } else if (!rcu_access_pointer(rxqueue
->rps_flow_table
)) {
2745 skb_reset_network_header(skb
);
2746 if (!skb_get_rxhash(skb
))
2749 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2750 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
2751 if (flow_table
&& sock_flow_table
) {
2753 struct rps_dev_flow
*rflow
;
2755 rflow
= &flow_table
->flows
[skb
->rxhash
& flow_table
->mask
];
2758 next_cpu
= sock_flow_table
->ents
[skb
->rxhash
&
2759 sock_flow_table
->mask
];
2762 * If the desired CPU (where last recvmsg was done) is
2763 * different from current CPU (one in the rx-queue flow
2764 * table entry), switch if one of the following holds:
2765 * - Current CPU is unset (equal to RPS_NO_CPU).
2766 * - Current CPU is offline.
2767 * - The current CPU's queue tail has advanced beyond the
2768 * last packet that was enqueued using this table entry.
2769 * This guarantees that all previous packets for the flow
2770 * have been dequeued, thus preserving in order delivery.
2772 if (unlikely(tcpu
!= next_cpu
) &&
2773 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
2774 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
2775 rflow
->last_qtail
)) >= 0))
2776 rflow
= set_rps_cpu(dev
, skb
, rflow
, next_cpu
);
2778 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
2786 tcpu
= map
->cpus
[((u64
) skb
->rxhash
* map
->len
) >> 32];
2788 if (cpu_online(tcpu
)) {
2798 #ifdef CONFIG_RFS_ACCEL
2801 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2802 * @dev: Device on which the filter was set
2803 * @rxq_index: RX queue index
2804 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2805 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2807 * Drivers that implement ndo_rx_flow_steer() should periodically call
2808 * this function for each installed filter and remove the filters for
2809 * which it returns %true.
2811 bool rps_may_expire_flow(struct net_device
*dev
, u16 rxq_index
,
2812 u32 flow_id
, u16 filter_id
)
2814 struct netdev_rx_queue
*rxqueue
= dev
->_rx
+ rxq_index
;
2815 struct rps_dev_flow_table
*flow_table
;
2816 struct rps_dev_flow
*rflow
;
2821 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2822 if (flow_table
&& flow_id
<= flow_table
->mask
) {
2823 rflow
= &flow_table
->flows
[flow_id
];
2824 cpu
= ACCESS_ONCE(rflow
->cpu
);
2825 if (rflow
->filter
== filter_id
&& cpu
!= RPS_NO_CPU
&&
2826 ((int)(per_cpu(softnet_data
, cpu
).input_queue_head
-
2827 rflow
->last_qtail
) <
2828 (int)(10 * flow_table
->mask
)))
2834 EXPORT_SYMBOL(rps_may_expire_flow
);
2836 #endif /* CONFIG_RFS_ACCEL */
2838 /* Called from hardirq (IPI) context */
2839 static void rps_trigger_softirq(void *data
)
2841 struct softnet_data
*sd
= data
;
2843 ____napi_schedule(sd
, &sd
->backlog
);
2847 #endif /* CONFIG_RPS */
2850 * Check if this softnet_data structure is another cpu one
2851 * If yes, queue it to our IPI list and return 1
2854 static int rps_ipi_queued(struct softnet_data
*sd
)
2857 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
2860 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
2861 mysd
->rps_ipi_list
= sd
;
2863 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2866 #endif /* CONFIG_RPS */
2871 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2872 * queue (may be a remote CPU queue).
2874 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
2875 unsigned int *qtail
)
2877 struct softnet_data
*sd
;
2878 unsigned long flags
;
2880 sd
= &per_cpu(softnet_data
, cpu
);
2882 local_irq_save(flags
);
2885 if (skb_queue_len(&sd
->input_pkt_queue
) <= netdev_max_backlog
) {
2886 if (skb_queue_len(&sd
->input_pkt_queue
)) {
2888 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
2889 input_queue_tail_incr_save(sd
, qtail
);
2891 local_irq_restore(flags
);
2892 return NET_RX_SUCCESS
;
2895 /* Schedule NAPI for backlog device
2896 * We can use non atomic operation since we own the queue lock
2898 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
2899 if (!rps_ipi_queued(sd
))
2900 ____napi_schedule(sd
, &sd
->backlog
);
2908 local_irq_restore(flags
);
2910 atomic_long_inc(&skb
->dev
->rx_dropped
);
2916 * netif_rx - post buffer to the network code
2917 * @skb: buffer to post
2919 * This function receives a packet from a device driver and queues it for
2920 * the upper (protocol) levels to process. It always succeeds. The buffer
2921 * may be dropped during processing for congestion control or by the
2925 * NET_RX_SUCCESS (no congestion)
2926 * NET_RX_DROP (packet was dropped)
2930 int netif_rx(struct sk_buff
*skb
)
2934 /* if netpoll wants it, pretend we never saw it */
2935 if (netpoll_rx(skb
))
2938 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
2940 trace_netif_rx(skb
);
2942 if (static_key_false(&rps_needed
)) {
2943 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
2949 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
2951 cpu
= smp_processor_id();
2953 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
2961 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
2966 EXPORT_SYMBOL(netif_rx
);
2968 int netif_rx_ni(struct sk_buff
*skb
)
2973 err
= netif_rx(skb
);
2974 if (local_softirq_pending())
2980 EXPORT_SYMBOL(netif_rx_ni
);
2982 static void net_tx_action(struct softirq_action
*h
)
2984 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
2986 if (sd
->completion_queue
) {
2987 struct sk_buff
*clist
;
2989 local_irq_disable();
2990 clist
= sd
->completion_queue
;
2991 sd
->completion_queue
= NULL
;
2995 struct sk_buff
*skb
= clist
;
2996 clist
= clist
->next
;
2998 WARN_ON(atomic_read(&skb
->users
));
2999 trace_kfree_skb(skb
, net_tx_action
);
3004 if (sd
->output_queue
) {
3007 local_irq_disable();
3008 head
= sd
->output_queue
;
3009 sd
->output_queue
= NULL
;
3010 sd
->output_queue_tailp
= &sd
->output_queue
;
3014 struct Qdisc
*q
= head
;
3015 spinlock_t
*root_lock
;
3017 head
= head
->next_sched
;
3019 root_lock
= qdisc_lock(q
);
3020 if (spin_trylock(root_lock
)) {
3021 smp_mb__before_clear_bit();
3022 clear_bit(__QDISC_STATE_SCHED
,
3025 spin_unlock(root_lock
);
3027 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
3029 __netif_reschedule(q
);
3031 smp_mb__before_clear_bit();
3032 clear_bit(__QDISC_STATE_SCHED
,
3040 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3041 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3042 /* This hook is defined here for ATM LANE */
3043 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
3044 unsigned char *addr
) __read_mostly
;
3045 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
3048 #ifdef CONFIG_NET_CLS_ACT
3049 /* TODO: Maybe we should just force sch_ingress to be compiled in
3050 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3051 * a compare and 2 stores extra right now if we dont have it on
3052 * but have CONFIG_NET_CLS_ACT
3053 * NOTE: This doesn't stop any functionality; if you dont have
3054 * the ingress scheduler, you just can't add policies on ingress.
3057 static int ing_filter(struct sk_buff
*skb
, struct netdev_queue
*rxq
)
3059 struct net_device
*dev
= skb
->dev
;
3060 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
3061 int result
= TC_ACT_OK
;
3064 if (unlikely(MAX_RED_LOOP
< ttl
++)) {
3065 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3066 skb
->skb_iif
, dev
->ifindex
);
3070 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
3071 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
3074 if (q
!= &noop_qdisc
) {
3075 spin_lock(qdisc_lock(q
));
3076 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
3077 result
= qdisc_enqueue_root(skb
, q
);
3078 spin_unlock(qdisc_lock(q
));
3084 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
3085 struct packet_type
**pt_prev
,
3086 int *ret
, struct net_device
*orig_dev
)
3088 struct netdev_queue
*rxq
= rcu_dereference(skb
->dev
->ingress_queue
);
3090 if (!rxq
|| rxq
->qdisc
== &noop_qdisc
)
3094 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
3098 switch (ing_filter(skb
, rxq
)) {
3112 * netdev_rx_handler_register - register receive handler
3113 * @dev: device to register a handler for
3114 * @rx_handler: receive handler to register
3115 * @rx_handler_data: data pointer that is used by rx handler
3117 * Register a receive hander for a device. This handler will then be
3118 * called from __netif_receive_skb. A negative errno code is returned
3121 * The caller must hold the rtnl_mutex.
3123 * For a general description of rx_handler, see enum rx_handler_result.
3125 int netdev_rx_handler_register(struct net_device
*dev
,
3126 rx_handler_func_t
*rx_handler
,
3127 void *rx_handler_data
)
3131 if (dev
->rx_handler
)
3134 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
3135 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
3139 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
3142 * netdev_rx_handler_unregister - unregister receive handler
3143 * @dev: device to unregister a handler from
3145 * Unregister a receive hander from a device.
3147 * The caller must hold the rtnl_mutex.
3149 void netdev_rx_handler_unregister(struct net_device
*dev
)
3153 RCU_INIT_POINTER(dev
->rx_handler
, NULL
);
3154 RCU_INIT_POINTER(dev
->rx_handler_data
, NULL
);
3156 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
3158 static int __netif_receive_skb(struct sk_buff
*skb
)
3160 struct packet_type
*ptype
, *pt_prev
;
3161 rx_handler_func_t
*rx_handler
;
3162 struct net_device
*orig_dev
;
3163 struct net_device
*null_or_dev
;
3164 bool deliver_exact
= false;
3165 int ret
= NET_RX_DROP
;
3168 net_timestamp_check(!netdev_tstamp_prequeue
, skb
);
3170 trace_netif_receive_skb(skb
);
3172 /* if we've gotten here through NAPI, check netpoll */
3173 if (netpoll_receive_skb(skb
))
3176 orig_dev
= skb
->dev
;
3178 skb_reset_network_header(skb
);
3179 skb_reset_transport_header(skb
);
3180 skb_reset_mac_len(skb
);
3187 skb
->skb_iif
= skb
->dev
->ifindex
;
3189 __this_cpu_inc(softnet_data
.processed
);
3191 if (skb
->protocol
== cpu_to_be16(ETH_P_8021Q
)) {
3192 skb
= vlan_untag(skb
);
3197 #ifdef CONFIG_NET_CLS_ACT
3198 if (skb
->tc_verd
& TC_NCLS
) {
3199 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
3204 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
3205 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
3207 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3212 #ifdef CONFIG_NET_CLS_ACT
3213 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
3219 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
3220 if (vlan_tx_tag_present(skb
)) {
3222 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3225 if (vlan_do_receive(&skb
, !rx_handler
))
3227 else if (unlikely(!skb
))
3233 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3236 switch (rx_handler(&skb
)) {
3237 case RX_HANDLER_CONSUMED
:
3239 case RX_HANDLER_ANOTHER
:
3241 case RX_HANDLER_EXACT
:
3242 deliver_exact
= true;
3243 case RX_HANDLER_PASS
:
3250 /* deliver only exact match when indicated */
3251 null_or_dev
= deliver_exact
? skb
->dev
: NULL
;
3253 type
= skb
->protocol
;
3254 list_for_each_entry_rcu(ptype
,
3255 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
3256 if (ptype
->type
== type
&&
3257 (ptype
->dev
== null_or_dev
|| ptype
->dev
== skb
->dev
||
3258 ptype
->dev
== orig_dev
)) {
3260 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3266 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
3269 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
3271 atomic_long_inc(&skb
->dev
->rx_dropped
);
3273 /* Jamal, now you will not able to escape explaining
3274 * me how you were going to use this. :-)
3285 * netif_receive_skb - process receive buffer from network
3286 * @skb: buffer to process
3288 * netif_receive_skb() is the main receive data processing function.
3289 * It always succeeds. The buffer may be dropped during processing
3290 * for congestion control or by the protocol layers.
3292 * This function may only be called from softirq context and interrupts
3293 * should be enabled.
3295 * Return values (usually ignored):
3296 * NET_RX_SUCCESS: no congestion
3297 * NET_RX_DROP: packet was dropped
3299 int netif_receive_skb(struct sk_buff
*skb
)
3301 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3303 if (skb_defer_rx_timestamp(skb
))
3304 return NET_RX_SUCCESS
;
3307 if (static_key_false(&rps_needed
)) {
3308 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3313 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3316 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3323 return __netif_receive_skb(skb
);
3325 EXPORT_SYMBOL(netif_receive_skb
);
3327 /* Network device is going away, flush any packets still pending
3328 * Called with irqs disabled.
3330 static void flush_backlog(void *arg
)
3332 struct net_device
*dev
= arg
;
3333 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3334 struct sk_buff
*skb
, *tmp
;
3337 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3338 if (skb
->dev
== dev
) {
3339 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3341 input_queue_head_incr(sd
);
3346 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3347 if (skb
->dev
== dev
) {
3348 __skb_unlink(skb
, &sd
->process_queue
);
3350 input_queue_head_incr(sd
);
3355 static int napi_gro_complete(struct sk_buff
*skb
)
3357 struct packet_type
*ptype
;
3358 __be16 type
= skb
->protocol
;
3359 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3362 if (NAPI_GRO_CB(skb
)->count
== 1) {
3363 skb_shinfo(skb
)->gso_size
= 0;
3368 list_for_each_entry_rcu(ptype
, head
, list
) {
3369 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_complete
)
3372 err
= ptype
->gro_complete(skb
);
3378 WARN_ON(&ptype
->list
== head
);
3380 return NET_RX_SUCCESS
;
3384 return netif_receive_skb(skb
);
3387 inline void napi_gro_flush(struct napi_struct
*napi
)
3389 struct sk_buff
*skb
, *next
;
3391 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3394 napi_gro_complete(skb
);
3397 napi
->gro_count
= 0;
3398 napi
->gro_list
= NULL
;
3400 EXPORT_SYMBOL(napi_gro_flush
);
3402 enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3404 struct sk_buff
**pp
= NULL
;
3405 struct packet_type
*ptype
;
3406 __be16 type
= skb
->protocol
;
3407 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3410 enum gro_result ret
;
3412 if (!(skb
->dev
->features
& NETIF_F_GRO
) || netpoll_rx_on(skb
))
3415 if (skb_is_gso(skb
) || skb_has_frag_list(skb
))
3419 list_for_each_entry_rcu(ptype
, head
, list
) {
3420 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_receive
)
3423 skb_set_network_header(skb
, skb_gro_offset(skb
));
3424 mac_len
= skb
->network_header
- skb
->mac_header
;
3425 skb
->mac_len
= mac_len
;
3426 NAPI_GRO_CB(skb
)->same_flow
= 0;
3427 NAPI_GRO_CB(skb
)->flush
= 0;
3428 NAPI_GRO_CB(skb
)->free
= 0;
3430 pp
= ptype
->gro_receive(&napi
->gro_list
, skb
);
3435 if (&ptype
->list
== head
)
3438 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3439 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3442 struct sk_buff
*nskb
= *pp
;
3446 napi_gro_complete(nskb
);
3453 if (NAPI_GRO_CB(skb
)->flush
|| napi
->gro_count
>= MAX_GRO_SKBS
)
3457 NAPI_GRO_CB(skb
)->count
= 1;
3458 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
3459 skb
->next
= napi
->gro_list
;
3460 napi
->gro_list
= skb
;
3464 if (skb_headlen(skb
) < skb_gro_offset(skb
)) {
3465 int grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
3467 BUG_ON(skb
->end
- skb
->tail
< grow
);
3469 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3472 skb
->data_len
-= grow
;
3474 skb_shinfo(skb
)->frags
[0].page_offset
+= grow
;
3475 skb_frag_size_sub(&skb_shinfo(skb
)->frags
[0], grow
);
3477 if (unlikely(!skb_frag_size(&skb_shinfo(skb
)->frags
[0]))) {
3478 skb_frag_unref(skb
, 0);
3479 memmove(skb_shinfo(skb
)->frags
,
3480 skb_shinfo(skb
)->frags
+ 1,
3481 --skb_shinfo(skb
)->nr_frags
* sizeof(skb_frag_t
));
3492 EXPORT_SYMBOL(dev_gro_receive
);
3494 static inline gro_result_t
3495 __napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3498 unsigned int maclen
= skb
->dev
->hard_header_len
;
3500 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3501 unsigned long diffs
;
3503 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
3504 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
3505 if (maclen
== ETH_HLEN
)
3506 diffs
|= compare_ether_header(skb_mac_header(p
),
3507 skb_gro_mac_header(skb
));
3509 diffs
= memcmp(skb_mac_header(p
),
3510 skb_gro_mac_header(skb
),
3512 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
3513 NAPI_GRO_CB(p
)->flush
= 0;
3516 return dev_gro_receive(napi
, skb
);
3519 gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
3523 if (netif_receive_skb(skb
))
3531 case GRO_MERGED_FREE
:
3532 if (NAPI_GRO_CB(skb
)->free
== NAPI_GRO_FREE_STOLEN_HEAD
)
3533 kmem_cache_free(skbuff_head_cache
, skb
);
3545 EXPORT_SYMBOL(napi_skb_finish
);
3547 void skb_gro_reset_offset(struct sk_buff
*skb
)
3549 NAPI_GRO_CB(skb
)->data_offset
= 0;
3550 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3551 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3553 if (skb
->mac_header
== skb
->tail
&&
3554 !PageHighMem(skb_frag_page(&skb_shinfo(skb
)->frags
[0]))) {
3555 NAPI_GRO_CB(skb
)->frag0
=
3556 skb_frag_address(&skb_shinfo(skb
)->frags
[0]);
3557 NAPI_GRO_CB(skb
)->frag0_len
= skb_frag_size(&skb_shinfo(skb
)->frags
[0]);
3560 EXPORT_SYMBOL(skb_gro_reset_offset
);
3562 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3564 skb_gro_reset_offset(skb
);
3566 return napi_skb_finish(__napi_gro_receive(napi
, skb
), skb
);
3568 EXPORT_SYMBOL(napi_gro_receive
);
3570 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
3572 __skb_pull(skb
, skb_headlen(skb
));
3573 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3574 skb_reserve(skb
, NET_SKB_PAD
+ NET_IP_ALIGN
- skb_headroom(skb
));
3576 skb
->dev
= napi
->dev
;
3582 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
3584 struct sk_buff
*skb
= napi
->skb
;
3587 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
3593 EXPORT_SYMBOL(napi_get_frags
);
3595 gro_result_t
napi_frags_finish(struct napi_struct
*napi
, struct sk_buff
*skb
,
3601 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
3603 if (ret
== GRO_HELD
)
3604 skb_gro_pull(skb
, -ETH_HLEN
);
3605 else if (netif_receive_skb(skb
))
3610 case GRO_MERGED_FREE
:
3611 napi_reuse_skb(napi
, skb
);
3620 EXPORT_SYMBOL(napi_frags_finish
);
3622 static struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
3624 struct sk_buff
*skb
= napi
->skb
;
3631 skb_reset_mac_header(skb
);
3632 skb_gro_reset_offset(skb
);
3634 off
= skb_gro_offset(skb
);
3635 hlen
= off
+ sizeof(*eth
);
3636 eth
= skb_gro_header_fast(skb
, off
);
3637 if (skb_gro_header_hard(skb
, hlen
)) {
3638 eth
= skb_gro_header_slow(skb
, hlen
, off
);
3639 if (unlikely(!eth
)) {
3640 napi_reuse_skb(napi
, skb
);
3646 skb_gro_pull(skb
, sizeof(*eth
));
3649 * This works because the only protocols we care about don't require
3650 * special handling. We'll fix it up properly at the end.
3652 skb
->protocol
= eth
->h_proto
;
3658 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
3660 struct sk_buff
*skb
= napi_frags_skb(napi
);
3665 return napi_frags_finish(napi
, skb
, __napi_gro_receive(napi
, skb
));
3667 EXPORT_SYMBOL(napi_gro_frags
);
3670 * net_rps_action sends any pending IPI's for rps.
3671 * Note: called with local irq disabled, but exits with local irq enabled.
3673 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
3676 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
3679 sd
->rps_ipi_list
= NULL
;
3683 /* Send pending IPI's to kick RPS processing on remote cpus. */
3685 struct softnet_data
*next
= remsd
->rps_ipi_next
;
3687 if (cpu_online(remsd
->cpu
))
3688 __smp_call_function_single(remsd
->cpu
,
3697 static int process_backlog(struct napi_struct
*napi
, int quota
)
3700 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
3703 /* Check if we have pending ipi, its better to send them now,
3704 * not waiting net_rx_action() end.
3706 if (sd
->rps_ipi_list
) {
3707 local_irq_disable();
3708 net_rps_action_and_irq_enable(sd
);
3711 napi
->weight
= weight_p
;
3712 local_irq_disable();
3713 while (work
< quota
) {
3714 struct sk_buff
*skb
;
3717 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
3719 __netif_receive_skb(skb
);
3720 local_irq_disable();
3721 input_queue_head_incr(sd
);
3722 if (++work
>= quota
) {
3729 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3731 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
3732 &sd
->process_queue
);
3734 if (qlen
< quota
- work
) {
3736 * Inline a custom version of __napi_complete().
3737 * only current cpu owns and manipulates this napi,
3738 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3739 * we can use a plain write instead of clear_bit(),
3740 * and we dont need an smp_mb() memory barrier.
3742 list_del(&napi
->poll_list
);
3745 quota
= work
+ qlen
;
3755 * __napi_schedule - schedule for receive
3756 * @n: entry to schedule
3758 * The entry's receive function will be scheduled to run
3760 void __napi_schedule(struct napi_struct
*n
)
3762 unsigned long flags
;
3764 local_irq_save(flags
);
3765 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
3766 local_irq_restore(flags
);
3768 EXPORT_SYMBOL(__napi_schedule
);
3770 void __napi_complete(struct napi_struct
*n
)
3772 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
3773 BUG_ON(n
->gro_list
);
3775 list_del(&n
->poll_list
);
3776 smp_mb__before_clear_bit();
3777 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
3779 EXPORT_SYMBOL(__napi_complete
);
3781 void napi_complete(struct napi_struct
*n
)
3783 unsigned long flags
;
3786 * don't let napi dequeue from the cpu poll list
3787 * just in case its running on a different cpu
3789 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
3793 local_irq_save(flags
);
3795 local_irq_restore(flags
);
3797 EXPORT_SYMBOL(napi_complete
);
3799 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
3800 int (*poll
)(struct napi_struct
*, int), int weight
)
3802 INIT_LIST_HEAD(&napi
->poll_list
);
3803 napi
->gro_count
= 0;
3804 napi
->gro_list
= NULL
;
3807 napi
->weight
= weight
;
3808 list_add(&napi
->dev_list
, &dev
->napi_list
);
3810 #ifdef CONFIG_NETPOLL
3811 spin_lock_init(&napi
->poll_lock
);
3812 napi
->poll_owner
= -1;
3814 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
3816 EXPORT_SYMBOL(netif_napi_add
);
3818 void netif_napi_del(struct napi_struct
*napi
)
3820 struct sk_buff
*skb
, *next
;
3822 list_del_init(&napi
->dev_list
);
3823 napi_free_frags(napi
);
3825 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3831 napi
->gro_list
= NULL
;
3832 napi
->gro_count
= 0;
3834 EXPORT_SYMBOL(netif_napi_del
);
3836 static void net_rx_action(struct softirq_action
*h
)
3838 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3839 unsigned long time_limit
= jiffies
+ 2;
3840 int budget
= netdev_budget
;
3843 local_irq_disable();
3845 while (!list_empty(&sd
->poll_list
)) {
3846 struct napi_struct
*n
;
3849 /* If softirq window is exhuasted then punt.
3850 * Allow this to run for 2 jiffies since which will allow
3851 * an average latency of 1.5/HZ.
3853 if (unlikely(budget
<= 0 || time_after(jiffies
, time_limit
)))
3858 /* Even though interrupts have been re-enabled, this
3859 * access is safe because interrupts can only add new
3860 * entries to the tail of this list, and only ->poll()
3861 * calls can remove this head entry from the list.
3863 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
3865 have
= netpoll_poll_lock(n
);
3869 /* This NAPI_STATE_SCHED test is for avoiding a race
3870 * with netpoll's poll_napi(). Only the entity which
3871 * obtains the lock and sees NAPI_STATE_SCHED set will
3872 * actually make the ->poll() call. Therefore we avoid
3873 * accidentally calling ->poll() when NAPI is not scheduled.
3876 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
3877 work
= n
->poll(n
, weight
);
3881 WARN_ON_ONCE(work
> weight
);
3885 local_irq_disable();
3887 /* Drivers must not modify the NAPI state if they
3888 * consume the entire weight. In such cases this code
3889 * still "owns" the NAPI instance and therefore can
3890 * move the instance around on the list at-will.
3892 if (unlikely(work
== weight
)) {
3893 if (unlikely(napi_disable_pending(n
))) {
3896 local_irq_disable();
3898 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
3901 netpoll_poll_unlock(have
);
3904 net_rps_action_and_irq_enable(sd
);
3906 #ifdef CONFIG_NET_DMA
3908 * There may not be any more sk_buffs coming right now, so push
3909 * any pending DMA copies to hardware
3911 dma_issue_pending_all();
3918 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3922 static gifconf_func_t
*gifconf_list
[NPROTO
];
3925 * register_gifconf - register a SIOCGIF handler
3926 * @family: Address family
3927 * @gifconf: Function handler
3929 * Register protocol dependent address dumping routines. The handler
3930 * that is passed must not be freed or reused until it has been replaced
3931 * by another handler.
3933 int register_gifconf(unsigned int family
, gifconf_func_t
*gifconf
)
3935 if (family
>= NPROTO
)
3937 gifconf_list
[family
] = gifconf
;
3940 EXPORT_SYMBOL(register_gifconf
);
3944 * Map an interface index to its name (SIOCGIFNAME)
3948 * We need this ioctl for efficient implementation of the
3949 * if_indextoname() function required by the IPv6 API. Without
3950 * it, we would have to search all the interfaces to find a
3954 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
3956 struct net_device
*dev
;
3960 * Fetch the caller's info block.
3963 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
3967 dev
= dev_get_by_index_rcu(net
, ifr
.ifr_ifindex
);
3973 strcpy(ifr
.ifr_name
, dev
->name
);
3976 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
3982 * Perform a SIOCGIFCONF call. This structure will change
3983 * size eventually, and there is nothing I can do about it.
3984 * Thus we will need a 'compatibility mode'.
3987 static int dev_ifconf(struct net
*net
, char __user
*arg
)
3990 struct net_device
*dev
;
3997 * Fetch the caller's info block.
4000 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
4007 * Loop over the interfaces, and write an info block for each.
4011 for_each_netdev(net
, dev
) {
4012 for (i
= 0; i
< NPROTO
; i
++) {
4013 if (gifconf_list
[i
]) {
4016 done
= gifconf_list
[i
](dev
, NULL
, 0);
4018 done
= gifconf_list
[i
](dev
, pos
+ total
,
4028 * All done. Write the updated control block back to the caller.
4030 ifc
.ifc_len
= total
;
4033 * Both BSD and Solaris return 0 here, so we do too.
4035 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
4038 #ifdef CONFIG_PROC_FS
4040 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4042 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4043 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4044 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4046 static inline struct net_device
*dev_from_same_bucket(struct seq_file
*seq
, loff_t
*pos
)
4048 struct net
*net
= seq_file_net(seq
);
4049 struct net_device
*dev
;
4050 struct hlist_node
*p
;
4051 struct hlist_head
*h
;
4052 unsigned int count
= 0, offset
= get_offset(*pos
);
4054 h
= &net
->dev_name_head
[get_bucket(*pos
)];
4055 hlist_for_each_entry_rcu(dev
, p
, h
, name_hlist
) {
4056 if (++count
== offset
)
4063 static inline struct net_device
*dev_from_bucket(struct seq_file
*seq
, loff_t
*pos
)
4065 struct net_device
*dev
;
4066 unsigned int bucket
;
4069 dev
= dev_from_same_bucket(seq
, pos
);
4073 bucket
= get_bucket(*pos
) + 1;
4074 *pos
= set_bucket_offset(bucket
, 1);
4075 } while (bucket
< NETDEV_HASHENTRIES
);
4081 * This is invoked by the /proc filesystem handler to display a device
4084 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4089 return SEQ_START_TOKEN
;
4091 if (get_bucket(*pos
) >= NETDEV_HASHENTRIES
)
4094 return dev_from_bucket(seq
, pos
);
4097 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4100 return dev_from_bucket(seq
, pos
);
4103 void dev_seq_stop(struct seq_file
*seq
, void *v
)
4109 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
4111 struct rtnl_link_stats64 temp
;
4112 const struct rtnl_link_stats64
*stats
= dev_get_stats(dev
, &temp
);
4114 seq_printf(seq
, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4115 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4116 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
4118 stats
->rx_dropped
+ stats
->rx_missed_errors
,
4119 stats
->rx_fifo_errors
,
4120 stats
->rx_length_errors
+ stats
->rx_over_errors
+
4121 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
4122 stats
->rx_compressed
, stats
->multicast
,
4123 stats
->tx_bytes
, stats
->tx_packets
,
4124 stats
->tx_errors
, stats
->tx_dropped
,
4125 stats
->tx_fifo_errors
, stats
->collisions
,
4126 stats
->tx_carrier_errors
+
4127 stats
->tx_aborted_errors
+
4128 stats
->tx_window_errors
+
4129 stats
->tx_heartbeat_errors
,
4130 stats
->tx_compressed
);
4134 * Called from the PROCfs module. This now uses the new arbitrary sized
4135 * /proc/net interface to create /proc/net/dev
4137 static int dev_seq_show(struct seq_file
*seq
, void *v
)
4139 if (v
== SEQ_START_TOKEN
)
4140 seq_puts(seq
, "Inter-| Receive "
4142 " face |bytes packets errs drop fifo frame "
4143 "compressed multicast|bytes packets errs "
4144 "drop fifo colls carrier compressed\n");
4146 dev_seq_printf_stats(seq
, v
);
4150 static struct softnet_data
*softnet_get_online(loff_t
*pos
)
4152 struct softnet_data
*sd
= NULL
;
4154 while (*pos
< nr_cpu_ids
)
4155 if (cpu_online(*pos
)) {
4156 sd
= &per_cpu(softnet_data
, *pos
);
4163 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4165 return softnet_get_online(pos
);
4168 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4171 return softnet_get_online(pos
);
4174 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
4178 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
4180 struct softnet_data
*sd
= v
;
4182 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4183 sd
->processed
, sd
->dropped
, sd
->time_squeeze
, 0,
4184 0, 0, 0, 0, /* was fastroute */
4185 sd
->cpu_collision
, sd
->received_rps
);
4189 static const struct seq_operations dev_seq_ops
= {
4190 .start
= dev_seq_start
,
4191 .next
= dev_seq_next
,
4192 .stop
= dev_seq_stop
,
4193 .show
= dev_seq_show
,
4196 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
4198 return seq_open_net(inode
, file
, &dev_seq_ops
,
4199 sizeof(struct seq_net_private
));
4202 static const struct file_operations dev_seq_fops
= {
4203 .owner
= THIS_MODULE
,
4204 .open
= dev_seq_open
,
4206 .llseek
= seq_lseek
,
4207 .release
= seq_release_net
,
4210 static const struct seq_operations softnet_seq_ops
= {
4211 .start
= softnet_seq_start
,
4212 .next
= softnet_seq_next
,
4213 .stop
= softnet_seq_stop
,
4214 .show
= softnet_seq_show
,
4217 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
4219 return seq_open(file
, &softnet_seq_ops
);
4222 static const struct file_operations softnet_seq_fops
= {
4223 .owner
= THIS_MODULE
,
4224 .open
= softnet_seq_open
,
4226 .llseek
= seq_lseek
,
4227 .release
= seq_release
,
4230 static void *ptype_get_idx(loff_t pos
)
4232 struct packet_type
*pt
= NULL
;
4236 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
4242 for (t
= 0; t
< PTYPE_HASH_SIZE
; t
++) {
4243 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
4252 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4256 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
4259 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4261 struct packet_type
*pt
;
4262 struct list_head
*nxt
;
4266 if (v
== SEQ_START_TOKEN
)
4267 return ptype_get_idx(0);
4270 nxt
= pt
->list
.next
;
4271 if (pt
->type
== htons(ETH_P_ALL
)) {
4272 if (nxt
!= &ptype_all
)
4275 nxt
= ptype_base
[0].next
;
4277 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
4279 while (nxt
== &ptype_base
[hash
]) {
4280 if (++hash
>= PTYPE_HASH_SIZE
)
4282 nxt
= ptype_base
[hash
].next
;
4285 return list_entry(nxt
, struct packet_type
, list
);
4288 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
4294 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
4296 struct packet_type
*pt
= v
;
4298 if (v
== SEQ_START_TOKEN
)
4299 seq_puts(seq
, "Type Device Function\n");
4300 else if (pt
->dev
== NULL
|| dev_net(pt
->dev
) == seq_file_net(seq
)) {
4301 if (pt
->type
== htons(ETH_P_ALL
))
4302 seq_puts(seq
, "ALL ");
4304 seq_printf(seq
, "%04x", ntohs(pt
->type
));
4306 seq_printf(seq
, " %-8s %pF\n",
4307 pt
->dev
? pt
->dev
->name
: "", pt
->func
);
4313 static const struct seq_operations ptype_seq_ops
= {
4314 .start
= ptype_seq_start
,
4315 .next
= ptype_seq_next
,
4316 .stop
= ptype_seq_stop
,
4317 .show
= ptype_seq_show
,
4320 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
4322 return seq_open_net(inode
, file
, &ptype_seq_ops
,
4323 sizeof(struct seq_net_private
));
4326 static const struct file_operations ptype_seq_fops
= {
4327 .owner
= THIS_MODULE
,
4328 .open
= ptype_seq_open
,
4330 .llseek
= seq_lseek
,
4331 .release
= seq_release_net
,
4335 static int __net_init
dev_proc_net_init(struct net
*net
)
4339 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
4341 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
4343 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
4346 if (wext_proc_init(net
))
4352 proc_net_remove(net
, "ptype");
4354 proc_net_remove(net
, "softnet_stat");
4356 proc_net_remove(net
, "dev");
4360 static void __net_exit
dev_proc_net_exit(struct net
*net
)
4362 wext_proc_exit(net
);
4364 proc_net_remove(net
, "ptype");
4365 proc_net_remove(net
, "softnet_stat");
4366 proc_net_remove(net
, "dev");
4369 static struct pernet_operations __net_initdata dev_proc_ops
= {
4370 .init
= dev_proc_net_init
,
4371 .exit
= dev_proc_net_exit
,
4374 static int __init
dev_proc_init(void)
4376 return register_pernet_subsys(&dev_proc_ops
);
4379 #define dev_proc_init() 0
4380 #endif /* CONFIG_PROC_FS */
4384 * netdev_set_master - set up master pointer
4385 * @slave: slave device
4386 * @master: new master device
4388 * Changes the master device of the slave. Pass %NULL to break the
4389 * bonding. The caller must hold the RTNL semaphore. On a failure
4390 * a negative errno code is returned. On success the reference counts
4391 * are adjusted and the function returns zero.
4393 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
4395 struct net_device
*old
= slave
->master
;
4405 slave
->master
= master
;
4411 EXPORT_SYMBOL(netdev_set_master
);
4414 * netdev_set_bond_master - set up bonding master/slave pair
4415 * @slave: slave device
4416 * @master: new master device
4418 * Changes the master device of the slave. Pass %NULL to break the
4419 * bonding. The caller must hold the RTNL semaphore. On a failure
4420 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4421 * to the routing socket and the function returns zero.
4423 int netdev_set_bond_master(struct net_device
*slave
, struct net_device
*master
)
4429 err
= netdev_set_master(slave
, master
);
4433 slave
->flags
|= IFF_SLAVE
;
4435 slave
->flags
&= ~IFF_SLAVE
;
4437 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
4440 EXPORT_SYMBOL(netdev_set_bond_master
);
4442 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
4444 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4446 if ((dev
->flags
& IFF_UP
) && ops
->ndo_change_rx_flags
)
4447 ops
->ndo_change_rx_flags(dev
, flags
);
4450 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
4452 unsigned int old_flags
= dev
->flags
;
4458 dev
->flags
|= IFF_PROMISC
;
4459 dev
->promiscuity
+= inc
;
4460 if (dev
->promiscuity
== 0) {
4463 * If inc causes overflow, untouch promisc and return error.
4466 dev
->flags
&= ~IFF_PROMISC
;
4468 dev
->promiscuity
-= inc
;
4469 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4474 if (dev
->flags
!= old_flags
) {
4475 pr_info("device %s %s promiscuous mode\n",
4477 dev
->flags
& IFF_PROMISC
? "entered" : "left");
4478 if (audit_enabled
) {
4479 current_uid_gid(&uid
, &gid
);
4480 audit_log(current
->audit_context
, GFP_ATOMIC
,
4481 AUDIT_ANOM_PROMISCUOUS
,
4482 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4483 dev
->name
, (dev
->flags
& IFF_PROMISC
),
4484 (old_flags
& IFF_PROMISC
),
4485 audit_get_loginuid(current
),
4487 audit_get_sessionid(current
));
4490 dev_change_rx_flags(dev
, IFF_PROMISC
);
4496 * dev_set_promiscuity - update promiscuity count on a device
4500 * Add or remove promiscuity from a device. While the count in the device
4501 * remains above zero the interface remains promiscuous. Once it hits zero
4502 * the device reverts back to normal filtering operation. A negative inc
4503 * value is used to drop promiscuity on the device.
4504 * Return 0 if successful or a negative errno code on error.
4506 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
4508 unsigned int old_flags
= dev
->flags
;
4511 err
= __dev_set_promiscuity(dev
, inc
);
4514 if (dev
->flags
!= old_flags
)
4515 dev_set_rx_mode(dev
);
4518 EXPORT_SYMBOL(dev_set_promiscuity
);
4521 * dev_set_allmulti - update allmulti count on a device
4525 * Add or remove reception of all multicast frames to a device. While the
4526 * count in the device remains above zero the interface remains listening
4527 * to all interfaces. Once it hits zero the device reverts back to normal
4528 * filtering operation. A negative @inc value is used to drop the counter
4529 * when releasing a resource needing all multicasts.
4530 * Return 0 if successful or a negative errno code on error.
4533 int dev_set_allmulti(struct net_device
*dev
, int inc
)
4535 unsigned int old_flags
= dev
->flags
;
4539 dev
->flags
|= IFF_ALLMULTI
;
4540 dev
->allmulti
+= inc
;
4541 if (dev
->allmulti
== 0) {
4544 * If inc causes overflow, untouch allmulti and return error.
4547 dev
->flags
&= ~IFF_ALLMULTI
;
4549 dev
->allmulti
-= inc
;
4550 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4555 if (dev
->flags
^ old_flags
) {
4556 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
4557 dev_set_rx_mode(dev
);
4561 EXPORT_SYMBOL(dev_set_allmulti
);
4564 * Upload unicast and multicast address lists to device and
4565 * configure RX filtering. When the device doesn't support unicast
4566 * filtering it is put in promiscuous mode while unicast addresses
4569 void __dev_set_rx_mode(struct net_device
*dev
)
4571 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4573 /* dev_open will call this function so the list will stay sane. */
4574 if (!(dev
->flags
&IFF_UP
))
4577 if (!netif_device_present(dev
))
4580 if (!(dev
->priv_flags
& IFF_UNICAST_FLT
)) {
4581 /* Unicast addresses changes may only happen under the rtnl,
4582 * therefore calling __dev_set_promiscuity here is safe.
4584 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
4585 __dev_set_promiscuity(dev
, 1);
4586 dev
->uc_promisc
= true;
4587 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
4588 __dev_set_promiscuity(dev
, -1);
4589 dev
->uc_promisc
= false;
4593 if (ops
->ndo_set_rx_mode
)
4594 ops
->ndo_set_rx_mode(dev
);
4597 void dev_set_rx_mode(struct net_device
*dev
)
4599 netif_addr_lock_bh(dev
);
4600 __dev_set_rx_mode(dev
);
4601 netif_addr_unlock_bh(dev
);
4605 * dev_get_flags - get flags reported to userspace
4608 * Get the combination of flag bits exported through APIs to userspace.
4610 unsigned int dev_get_flags(const struct net_device
*dev
)
4614 flags
= (dev
->flags
& ~(IFF_PROMISC
|
4619 (dev
->gflags
& (IFF_PROMISC
|
4622 if (netif_running(dev
)) {
4623 if (netif_oper_up(dev
))
4624 flags
|= IFF_RUNNING
;
4625 if (netif_carrier_ok(dev
))
4626 flags
|= IFF_LOWER_UP
;
4627 if (netif_dormant(dev
))
4628 flags
|= IFF_DORMANT
;
4633 EXPORT_SYMBOL(dev_get_flags
);
4635 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4637 unsigned int old_flags
= dev
->flags
;
4643 * Set the flags on our device.
4646 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
4647 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
4649 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
4653 * Load in the correct multicast list now the flags have changed.
4656 if ((old_flags
^ flags
) & IFF_MULTICAST
)
4657 dev_change_rx_flags(dev
, IFF_MULTICAST
);
4659 dev_set_rx_mode(dev
);
4662 * Have we downed the interface. We handle IFF_UP ourselves
4663 * according to user attempts to set it, rather than blindly
4668 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
4669 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
4672 dev_set_rx_mode(dev
);
4675 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
4676 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
4678 dev
->gflags
^= IFF_PROMISC
;
4679 dev_set_promiscuity(dev
, inc
);
4682 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4683 is important. Some (broken) drivers set IFF_PROMISC, when
4684 IFF_ALLMULTI is requested not asking us and not reporting.
4686 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
4687 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
4689 dev
->gflags
^= IFF_ALLMULTI
;
4690 dev_set_allmulti(dev
, inc
);
4696 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
)
4698 unsigned int changes
= dev
->flags
^ old_flags
;
4700 if (changes
& IFF_UP
) {
4701 if (dev
->flags
& IFF_UP
)
4702 call_netdevice_notifiers(NETDEV_UP
, dev
);
4704 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
4707 if (dev
->flags
& IFF_UP
&&
4708 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
)))
4709 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
4713 * dev_change_flags - change device settings
4715 * @flags: device state flags
4717 * Change settings on device based state flags. The flags are
4718 * in the userspace exported format.
4720 int dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4723 unsigned int changes
, old_flags
= dev
->flags
;
4725 ret
= __dev_change_flags(dev
, flags
);
4729 changes
= old_flags
^ dev
->flags
;
4731 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
4733 __dev_notify_flags(dev
, old_flags
);
4736 EXPORT_SYMBOL(dev_change_flags
);
4739 * dev_set_mtu - Change maximum transfer unit
4741 * @new_mtu: new transfer unit
4743 * Change the maximum transfer size of the network device.
4745 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
4747 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4750 if (new_mtu
== dev
->mtu
)
4753 /* MTU must be positive. */
4757 if (!netif_device_present(dev
))
4761 if (ops
->ndo_change_mtu
)
4762 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
4766 if (!err
&& dev
->flags
& IFF_UP
)
4767 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
4770 EXPORT_SYMBOL(dev_set_mtu
);
4773 * dev_set_group - Change group this device belongs to
4775 * @new_group: group this device should belong to
4777 void dev_set_group(struct net_device
*dev
, int new_group
)
4779 dev
->group
= new_group
;
4781 EXPORT_SYMBOL(dev_set_group
);
4784 * dev_set_mac_address - Change Media Access Control Address
4788 * Change the hardware (MAC) address of the device
4790 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
4792 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4795 if (!ops
->ndo_set_mac_address
)
4797 if (sa
->sa_family
!= dev
->type
)
4799 if (!netif_device_present(dev
))
4801 err
= ops
->ndo_set_mac_address(dev
, sa
);
4803 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4806 EXPORT_SYMBOL(dev_set_mac_address
);
4809 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4811 static int dev_ifsioc_locked(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4814 struct net_device
*dev
= dev_get_by_name_rcu(net
, ifr
->ifr_name
);
4820 case SIOCGIFFLAGS
: /* Get interface flags */
4821 ifr
->ifr_flags
= (short) dev_get_flags(dev
);
4824 case SIOCGIFMETRIC
: /* Get the metric on the interface
4825 (currently unused) */
4826 ifr
->ifr_metric
= 0;
4829 case SIOCGIFMTU
: /* Get the MTU of a device */
4830 ifr
->ifr_mtu
= dev
->mtu
;
4835 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
4837 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
4838 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4839 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
4847 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
4848 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
4849 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
4850 ifr
->ifr_map
.irq
= dev
->irq
;
4851 ifr
->ifr_map
.dma
= dev
->dma
;
4852 ifr
->ifr_map
.port
= dev
->if_port
;
4856 ifr
->ifr_ifindex
= dev
->ifindex
;
4860 ifr
->ifr_qlen
= dev
->tx_queue_len
;
4864 /* dev_ioctl() should ensure this case
4876 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4878 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4881 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
4882 const struct net_device_ops
*ops
;
4887 ops
= dev
->netdev_ops
;
4890 case SIOCSIFFLAGS
: /* Set interface flags */
4891 return dev_change_flags(dev
, ifr
->ifr_flags
);
4893 case SIOCSIFMETRIC
: /* Set the metric on the interface
4894 (currently unused) */
4897 case SIOCSIFMTU
: /* Set the MTU of a device */
4898 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
4901 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
4903 case SIOCSIFHWBROADCAST
:
4904 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
4906 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
4907 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4908 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4912 if (ops
->ndo_set_config
) {
4913 if (!netif_device_present(dev
))
4915 return ops
->ndo_set_config(dev
, &ifr
->ifr_map
);
4920 if (!ops
->ndo_set_rx_mode
||
4921 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4923 if (!netif_device_present(dev
))
4925 return dev_mc_add_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4928 if (!ops
->ndo_set_rx_mode
||
4929 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4931 if (!netif_device_present(dev
))
4933 return dev_mc_del_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4936 if (ifr
->ifr_qlen
< 0)
4938 dev
->tx_queue_len
= ifr
->ifr_qlen
;
4942 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
4943 return dev_change_name(dev
, ifr
->ifr_newname
);
4946 err
= net_hwtstamp_validate(ifr
);
4952 * Unknown or private ioctl
4955 if ((cmd
>= SIOCDEVPRIVATE
&&
4956 cmd
<= SIOCDEVPRIVATE
+ 15) ||
4957 cmd
== SIOCBONDENSLAVE
||
4958 cmd
== SIOCBONDRELEASE
||
4959 cmd
== SIOCBONDSETHWADDR
||
4960 cmd
== SIOCBONDSLAVEINFOQUERY
||
4961 cmd
== SIOCBONDINFOQUERY
||
4962 cmd
== SIOCBONDCHANGEACTIVE
||
4963 cmd
== SIOCGMIIPHY
||
4964 cmd
== SIOCGMIIREG
||
4965 cmd
== SIOCSMIIREG
||
4966 cmd
== SIOCBRADDIF
||
4967 cmd
== SIOCBRDELIF
||
4968 cmd
== SIOCSHWTSTAMP
||
4969 cmd
== SIOCWANDEV
) {
4971 if (ops
->ndo_do_ioctl
) {
4972 if (netif_device_present(dev
))
4973 err
= ops
->ndo_do_ioctl(dev
, ifr
, cmd
);
4985 * This function handles all "interface"-type I/O control requests. The actual
4986 * 'doing' part of this is dev_ifsioc above.
4990 * dev_ioctl - network device ioctl
4991 * @net: the applicable net namespace
4992 * @cmd: command to issue
4993 * @arg: pointer to a struct ifreq in user space
4995 * Issue ioctl functions to devices. This is normally called by the
4996 * user space syscall interfaces but can sometimes be useful for
4997 * other purposes. The return value is the return from the syscall if
4998 * positive or a negative errno code on error.
5001 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
5007 /* One special case: SIOCGIFCONF takes ifconf argument
5008 and requires shared lock, because it sleeps writing
5012 if (cmd
== SIOCGIFCONF
) {
5014 ret
= dev_ifconf(net
, (char __user
*) arg
);
5018 if (cmd
== SIOCGIFNAME
)
5019 return dev_ifname(net
, (struct ifreq __user
*)arg
);
5021 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
5024 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
5026 colon
= strchr(ifr
.ifr_name
, ':');
5031 * See which interface the caller is talking about.
5036 * These ioctl calls:
5037 * - can be done by all.
5038 * - atomic and do not require locking.
5049 dev_load(net
, ifr
.ifr_name
);
5051 ret
= dev_ifsioc_locked(net
, &ifr
, cmd
);
5056 if (copy_to_user(arg
, &ifr
,
5057 sizeof(struct ifreq
)))
5063 dev_load(net
, ifr
.ifr_name
);
5065 ret
= dev_ethtool(net
, &ifr
);
5070 if (copy_to_user(arg
, &ifr
,
5071 sizeof(struct ifreq
)))
5077 * These ioctl calls:
5078 * - require superuser power.
5079 * - require strict serialization.
5085 if (!capable(CAP_NET_ADMIN
))
5087 dev_load(net
, ifr
.ifr_name
);
5089 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5094 if (copy_to_user(arg
, &ifr
,
5095 sizeof(struct ifreq
)))
5101 * These ioctl calls:
5102 * - require superuser power.
5103 * - require strict serialization.
5104 * - do not return a value
5114 case SIOCSIFHWBROADCAST
:
5117 case SIOCBONDENSLAVE
:
5118 case SIOCBONDRELEASE
:
5119 case SIOCBONDSETHWADDR
:
5120 case SIOCBONDCHANGEACTIVE
:
5124 if (!capable(CAP_NET_ADMIN
))
5127 case SIOCBONDSLAVEINFOQUERY
:
5128 case SIOCBONDINFOQUERY
:
5129 dev_load(net
, ifr
.ifr_name
);
5131 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5136 /* Get the per device memory space. We can add this but
5137 * currently do not support it */
5139 /* Set the per device memory buffer space.
5140 * Not applicable in our case */
5145 * Unknown or private ioctl.
5148 if (cmd
== SIOCWANDEV
||
5149 (cmd
>= SIOCDEVPRIVATE
&&
5150 cmd
<= SIOCDEVPRIVATE
+ 15)) {
5151 dev_load(net
, ifr
.ifr_name
);
5153 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5155 if (!ret
&& copy_to_user(arg
, &ifr
,
5156 sizeof(struct ifreq
)))
5160 /* Take care of Wireless Extensions */
5161 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
5162 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
5169 * dev_new_index - allocate an ifindex
5170 * @net: the applicable net namespace
5172 * Returns a suitable unique value for a new device interface
5173 * number. The caller must hold the rtnl semaphore or the
5174 * dev_base_lock to be sure it remains unique.
5176 static int dev_new_index(struct net
*net
)
5182 if (!__dev_get_by_index(net
, ifindex
))
5187 /* Delayed registration/unregisteration */
5188 static LIST_HEAD(net_todo_list
);
5190 static void net_set_todo(struct net_device
*dev
)
5192 list_add_tail(&dev
->todo_list
, &net_todo_list
);
5195 static void rollback_registered_many(struct list_head
*head
)
5197 struct net_device
*dev
, *tmp
;
5199 BUG_ON(dev_boot_phase
);
5202 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
5203 /* Some devices call without registering
5204 * for initialization unwind. Remove those
5205 * devices and proceed with the remaining.
5207 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5208 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5212 list_del(&dev
->unreg_list
);
5215 dev
->dismantle
= true;
5216 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
5219 /* If device is running, close it first. */
5220 dev_close_many(head
);
5222 list_for_each_entry(dev
, head
, unreg_list
) {
5223 /* And unlink it from device chain. */
5224 unlist_netdevice(dev
);
5226 dev
->reg_state
= NETREG_UNREGISTERING
;
5231 list_for_each_entry(dev
, head
, unreg_list
) {
5232 /* Shutdown queueing discipline. */
5236 /* Notify protocols, that we are about to destroy
5237 this device. They should clean all the things.
5239 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5241 if (!dev
->rtnl_link_ops
||
5242 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5243 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
5246 * Flush the unicast and multicast chains
5251 if (dev
->netdev_ops
->ndo_uninit
)
5252 dev
->netdev_ops
->ndo_uninit(dev
);
5254 /* Notifier chain MUST detach us from master device. */
5255 WARN_ON(dev
->master
);
5257 /* Remove entries from kobject tree */
5258 netdev_unregister_kobject(dev
);
5261 /* Process any work delayed until the end of the batch */
5262 dev
= list_first_entry(head
, struct net_device
, unreg_list
);
5263 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
5267 list_for_each_entry(dev
, head
, unreg_list
)
5271 static void rollback_registered(struct net_device
*dev
)
5275 list_add(&dev
->unreg_list
, &single
);
5276 rollback_registered_many(&single
);
5280 static netdev_features_t
netdev_fix_features(struct net_device
*dev
,
5281 netdev_features_t features
)
5283 /* Fix illegal checksum combinations */
5284 if ((features
& NETIF_F_HW_CSUM
) &&
5285 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5286 netdev_warn(dev
, "mixed HW and IP checksum settings.\n");
5287 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5290 /* Fix illegal SG+CSUM combinations. */
5291 if ((features
& NETIF_F_SG
) &&
5292 !(features
& NETIF_F_ALL_CSUM
)) {
5294 "Dropping NETIF_F_SG since no checksum feature.\n");
5295 features
&= ~NETIF_F_SG
;
5298 /* TSO requires that SG is present as well. */
5299 if ((features
& NETIF_F_ALL_TSO
) && !(features
& NETIF_F_SG
)) {
5300 netdev_dbg(dev
, "Dropping TSO features since no SG feature.\n");
5301 features
&= ~NETIF_F_ALL_TSO
;
5304 /* TSO ECN requires that TSO is present as well. */
5305 if ((features
& NETIF_F_ALL_TSO
) == NETIF_F_TSO_ECN
)
5306 features
&= ~NETIF_F_TSO_ECN
;
5308 /* Software GSO depends on SG. */
5309 if ((features
& NETIF_F_GSO
) && !(features
& NETIF_F_SG
)) {
5310 netdev_dbg(dev
, "Dropping NETIF_F_GSO since no SG feature.\n");
5311 features
&= ~NETIF_F_GSO
;
5314 /* UFO needs SG and checksumming */
5315 if (features
& NETIF_F_UFO
) {
5316 /* maybe split UFO into V4 and V6? */
5317 if (!((features
& NETIF_F_GEN_CSUM
) ||
5318 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))
5319 == (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5321 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5322 features
&= ~NETIF_F_UFO
;
5325 if (!(features
& NETIF_F_SG
)) {
5327 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5328 features
&= ~NETIF_F_UFO
;
5335 int __netdev_update_features(struct net_device
*dev
)
5337 netdev_features_t features
;
5342 features
= netdev_get_wanted_features(dev
);
5344 if (dev
->netdev_ops
->ndo_fix_features
)
5345 features
= dev
->netdev_ops
->ndo_fix_features(dev
, features
);
5347 /* driver might be less strict about feature dependencies */
5348 features
= netdev_fix_features(dev
, features
);
5350 if (dev
->features
== features
)
5353 netdev_dbg(dev
, "Features changed: %pNF -> %pNF\n",
5354 &dev
->features
, &features
);
5356 if (dev
->netdev_ops
->ndo_set_features
)
5357 err
= dev
->netdev_ops
->ndo_set_features(dev
, features
);
5359 if (unlikely(err
< 0)) {
5361 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5362 err
, &features
, &dev
->features
);
5367 dev
->features
= features
;
5373 * netdev_update_features - recalculate device features
5374 * @dev: the device to check
5376 * Recalculate dev->features set and send notifications if it
5377 * has changed. Should be called after driver or hardware dependent
5378 * conditions might have changed that influence the features.
5380 void netdev_update_features(struct net_device
*dev
)
5382 if (__netdev_update_features(dev
))
5383 netdev_features_change(dev
);
5385 EXPORT_SYMBOL(netdev_update_features
);
5388 * netdev_change_features - recalculate device features
5389 * @dev: the device to check
5391 * Recalculate dev->features set and send notifications even
5392 * if they have not changed. Should be called instead of
5393 * netdev_update_features() if also dev->vlan_features might
5394 * have changed to allow the changes to be propagated to stacked
5397 void netdev_change_features(struct net_device
*dev
)
5399 __netdev_update_features(dev
);
5400 netdev_features_change(dev
);
5402 EXPORT_SYMBOL(netdev_change_features
);
5405 * netif_stacked_transfer_operstate - transfer operstate
5406 * @rootdev: the root or lower level device to transfer state from
5407 * @dev: the device to transfer operstate to
5409 * Transfer operational state from root to device. This is normally
5410 * called when a stacking relationship exists between the root
5411 * device and the device(a leaf device).
5413 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
5414 struct net_device
*dev
)
5416 if (rootdev
->operstate
== IF_OPER_DORMANT
)
5417 netif_dormant_on(dev
);
5419 netif_dormant_off(dev
);
5421 if (netif_carrier_ok(rootdev
)) {
5422 if (!netif_carrier_ok(dev
))
5423 netif_carrier_on(dev
);
5425 if (netif_carrier_ok(dev
))
5426 netif_carrier_off(dev
);
5429 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
5432 static int netif_alloc_rx_queues(struct net_device
*dev
)
5434 unsigned int i
, count
= dev
->num_rx_queues
;
5435 struct netdev_rx_queue
*rx
;
5439 rx
= kcalloc(count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5441 pr_err("netdev: Unable to allocate %u rx queues\n", count
);
5446 for (i
= 0; i
< count
; i
++)
5452 static void netdev_init_one_queue(struct net_device
*dev
,
5453 struct netdev_queue
*queue
, void *_unused
)
5455 /* Initialize queue lock */
5456 spin_lock_init(&queue
->_xmit_lock
);
5457 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
5458 queue
->xmit_lock_owner
= -1;
5459 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
5462 dql_init(&queue
->dql
, HZ
);
5466 static int netif_alloc_netdev_queues(struct net_device
*dev
)
5468 unsigned int count
= dev
->num_tx_queues
;
5469 struct netdev_queue
*tx
;
5473 tx
= kcalloc(count
, sizeof(struct netdev_queue
), GFP_KERNEL
);
5475 pr_err("netdev: Unable to allocate %u tx queues\n", count
);
5480 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5481 spin_lock_init(&dev
->tx_global_lock
);
5487 * register_netdevice - register a network device
5488 * @dev: device to register
5490 * Take a completed network device structure and add it to the kernel
5491 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5492 * chain. 0 is returned on success. A negative errno code is returned
5493 * on a failure to set up the device, or if the name is a duplicate.
5495 * Callers must hold the rtnl semaphore. You may want
5496 * register_netdev() instead of this.
5499 * The locking appears insufficient to guarantee two parallel registers
5500 * will not get the same name.
5503 int register_netdevice(struct net_device
*dev
)
5506 struct net
*net
= dev_net(dev
);
5508 BUG_ON(dev_boot_phase
);
5513 /* When net_device's are persistent, this will be fatal. */
5514 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
5517 spin_lock_init(&dev
->addr_list_lock
);
5518 netdev_set_addr_lockdep_class(dev
);
5522 ret
= dev_get_valid_name(dev
, dev
->name
);
5526 /* Init, if this function is available */
5527 if (dev
->netdev_ops
->ndo_init
) {
5528 ret
= dev
->netdev_ops
->ndo_init(dev
);
5536 dev
->ifindex
= dev_new_index(net
);
5537 if (dev
->iflink
== -1)
5538 dev
->iflink
= dev
->ifindex
;
5540 /* Transfer changeable features to wanted_features and enable
5541 * software offloads (GSO and GRO).
5543 dev
->hw_features
|= NETIF_F_SOFT_FEATURES
;
5544 dev
->features
|= NETIF_F_SOFT_FEATURES
;
5545 dev
->wanted_features
= dev
->features
& dev
->hw_features
;
5547 /* Turn on no cache copy if HW is doing checksum */
5548 if (!(dev
->flags
& IFF_LOOPBACK
)) {
5549 dev
->hw_features
|= NETIF_F_NOCACHE_COPY
;
5550 if (dev
->features
& NETIF_F_ALL_CSUM
) {
5551 dev
->wanted_features
|= NETIF_F_NOCACHE_COPY
;
5552 dev
->features
|= NETIF_F_NOCACHE_COPY
;
5556 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5558 dev
->vlan_features
|= NETIF_F_HIGHDMA
;
5560 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
5561 ret
= notifier_to_errno(ret
);
5565 ret
= netdev_register_kobject(dev
);
5568 dev
->reg_state
= NETREG_REGISTERED
;
5570 __netdev_update_features(dev
);
5573 * Default initial state at registry is that the
5574 * device is present.
5577 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5579 dev_init_scheduler(dev
);
5581 list_netdevice(dev
);
5583 /* Notify protocols, that a new device appeared. */
5584 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5585 ret
= notifier_to_errno(ret
);
5587 rollback_registered(dev
);
5588 dev
->reg_state
= NETREG_UNREGISTERED
;
5591 * Prevent userspace races by waiting until the network
5592 * device is fully setup before sending notifications.
5594 if (!dev
->rtnl_link_ops
||
5595 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5596 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5602 if (dev
->netdev_ops
->ndo_uninit
)
5603 dev
->netdev_ops
->ndo_uninit(dev
);
5606 EXPORT_SYMBOL(register_netdevice
);
5609 * init_dummy_netdev - init a dummy network device for NAPI
5610 * @dev: device to init
5612 * This takes a network device structure and initialize the minimum
5613 * amount of fields so it can be used to schedule NAPI polls without
5614 * registering a full blown interface. This is to be used by drivers
5615 * that need to tie several hardware interfaces to a single NAPI
5616 * poll scheduler due to HW limitations.
5618 int init_dummy_netdev(struct net_device
*dev
)
5620 /* Clear everything. Note we don't initialize spinlocks
5621 * are they aren't supposed to be taken by any of the
5622 * NAPI code and this dummy netdev is supposed to be
5623 * only ever used for NAPI polls
5625 memset(dev
, 0, sizeof(struct net_device
));
5627 /* make sure we BUG if trying to hit standard
5628 * register/unregister code path
5630 dev
->reg_state
= NETREG_DUMMY
;
5632 /* NAPI wants this */
5633 INIT_LIST_HEAD(&dev
->napi_list
);
5635 /* a dummy interface is started by default */
5636 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5637 set_bit(__LINK_STATE_START
, &dev
->state
);
5639 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5640 * because users of this 'device' dont need to change
5646 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
5650 * register_netdev - register a network device
5651 * @dev: device to register
5653 * Take a completed network device structure and add it to the kernel
5654 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5655 * chain. 0 is returned on success. A negative errno code is returned
5656 * on a failure to set up the device, or if the name is a duplicate.
5658 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5659 * and expands the device name if you passed a format string to
5662 int register_netdev(struct net_device
*dev
)
5667 err
= register_netdevice(dev
);
5671 EXPORT_SYMBOL(register_netdev
);
5673 int netdev_refcnt_read(const struct net_device
*dev
)
5677 for_each_possible_cpu(i
)
5678 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
5681 EXPORT_SYMBOL(netdev_refcnt_read
);
5684 * netdev_wait_allrefs - wait until all references are gone.
5686 * This is called when unregistering network devices.
5688 * Any protocol or device that holds a reference should register
5689 * for netdevice notification, and cleanup and put back the
5690 * reference if they receive an UNREGISTER event.
5691 * We can get stuck here if buggy protocols don't correctly
5694 static void netdev_wait_allrefs(struct net_device
*dev
)
5696 unsigned long rebroadcast_time
, warning_time
;
5699 linkwatch_forget_dev(dev
);
5701 rebroadcast_time
= warning_time
= jiffies
;
5702 refcnt
= netdev_refcnt_read(dev
);
5704 while (refcnt
!= 0) {
5705 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
5708 /* Rebroadcast unregister notification */
5709 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5710 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5711 * should have already handle it the first time */
5713 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
5715 /* We must not have linkwatch events
5716 * pending on unregister. If this
5717 * happens, we simply run the queue
5718 * unscheduled, resulting in a noop
5721 linkwatch_run_queue();
5726 rebroadcast_time
= jiffies
;
5731 refcnt
= netdev_refcnt_read(dev
);
5733 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
5734 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5736 warning_time
= jiffies
;
5745 * register_netdevice(x1);
5746 * register_netdevice(x2);
5748 * unregister_netdevice(y1);
5749 * unregister_netdevice(y2);
5755 * We are invoked by rtnl_unlock().
5756 * This allows us to deal with problems:
5757 * 1) We can delete sysfs objects which invoke hotplug
5758 * without deadlocking with linkwatch via keventd.
5759 * 2) Since we run with the RTNL semaphore not held, we can sleep
5760 * safely in order to wait for the netdev refcnt to drop to zero.
5762 * We must not return until all unregister events added during
5763 * the interval the lock was held have been completed.
5765 void netdev_run_todo(void)
5767 struct list_head list
;
5769 /* Snapshot list, allow later requests */
5770 list_replace_init(&net_todo_list
, &list
);
5774 /* Wait for rcu callbacks to finish before attempting to drain
5775 * the device list. This usually avoids a 250ms wait.
5777 if (!list_empty(&list
))
5780 while (!list_empty(&list
)) {
5781 struct net_device
*dev
5782 = list_first_entry(&list
, struct net_device
, todo_list
);
5783 list_del(&dev
->todo_list
);
5785 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
5786 pr_err("network todo '%s' but state %d\n",
5787 dev
->name
, dev
->reg_state
);
5792 dev
->reg_state
= NETREG_UNREGISTERED
;
5794 on_each_cpu(flush_backlog
, dev
, 1);
5796 netdev_wait_allrefs(dev
);
5799 BUG_ON(netdev_refcnt_read(dev
));
5800 WARN_ON(rcu_access_pointer(dev
->ip_ptr
));
5801 WARN_ON(rcu_access_pointer(dev
->ip6_ptr
));
5802 WARN_ON(dev
->dn_ptr
);
5804 if (dev
->destructor
)
5805 dev
->destructor(dev
);
5807 /* Free network device */
5808 kobject_put(&dev
->dev
.kobj
);
5812 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5813 * fields in the same order, with only the type differing.
5815 void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
5816 const struct net_device_stats
*netdev_stats
)
5818 #if BITS_PER_LONG == 64
5819 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
5820 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
5822 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
5823 const unsigned long *src
= (const unsigned long *)netdev_stats
;
5824 u64
*dst
= (u64
*)stats64
;
5826 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
5827 sizeof(*stats64
) / sizeof(u64
));
5828 for (i
= 0; i
< n
; i
++)
5832 EXPORT_SYMBOL(netdev_stats_to_stats64
);
5835 * dev_get_stats - get network device statistics
5836 * @dev: device to get statistics from
5837 * @storage: place to store stats
5839 * Get network statistics from device. Return @storage.
5840 * The device driver may provide its own method by setting
5841 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5842 * otherwise the internal statistics structure is used.
5844 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
5845 struct rtnl_link_stats64
*storage
)
5847 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5849 if (ops
->ndo_get_stats64
) {
5850 memset(storage
, 0, sizeof(*storage
));
5851 ops
->ndo_get_stats64(dev
, storage
);
5852 } else if (ops
->ndo_get_stats
) {
5853 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
5855 netdev_stats_to_stats64(storage
, &dev
->stats
);
5857 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
5860 EXPORT_SYMBOL(dev_get_stats
);
5862 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
5864 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
5866 #ifdef CONFIG_NET_CLS_ACT
5869 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
5872 netdev_init_one_queue(dev
, queue
, NULL
);
5873 queue
->qdisc
= &noop_qdisc
;
5874 queue
->qdisc_sleeping
= &noop_qdisc
;
5875 rcu_assign_pointer(dev
->ingress_queue
, queue
);
5881 * alloc_netdev_mqs - allocate network device
5882 * @sizeof_priv: size of private data to allocate space for
5883 * @name: device name format string
5884 * @setup: callback to initialize device
5885 * @txqs: the number of TX subqueues to allocate
5886 * @rxqs: the number of RX subqueues to allocate
5888 * Allocates a struct net_device with private data area for driver use
5889 * and performs basic initialization. Also allocates subquue structs
5890 * for each queue on the device.
5892 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
5893 void (*setup
)(struct net_device
*),
5894 unsigned int txqs
, unsigned int rxqs
)
5896 struct net_device
*dev
;
5898 struct net_device
*p
;
5900 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
5903 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
5909 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
5914 alloc_size
= sizeof(struct net_device
);
5916 /* ensure 32-byte alignment of private area */
5917 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
5918 alloc_size
+= sizeof_priv
;
5920 /* ensure 32-byte alignment of whole construct */
5921 alloc_size
+= NETDEV_ALIGN
- 1;
5923 p
= kzalloc(alloc_size
, GFP_KERNEL
);
5925 pr_err("alloc_netdev: Unable to allocate device\n");
5929 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
5930 dev
->padded
= (char *)dev
- (char *)p
;
5932 dev
->pcpu_refcnt
= alloc_percpu(int);
5933 if (!dev
->pcpu_refcnt
)
5936 if (dev_addr_init(dev
))
5942 dev_net_set(dev
, &init_net
);
5944 dev
->gso_max_size
= GSO_MAX_SIZE
;
5946 INIT_LIST_HEAD(&dev
->napi_list
);
5947 INIT_LIST_HEAD(&dev
->unreg_list
);
5948 INIT_LIST_HEAD(&dev
->link_watch_list
);
5949 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
5952 dev
->num_tx_queues
= txqs
;
5953 dev
->real_num_tx_queues
= txqs
;
5954 if (netif_alloc_netdev_queues(dev
))
5958 dev
->num_rx_queues
= rxqs
;
5959 dev
->real_num_rx_queues
= rxqs
;
5960 if (netif_alloc_rx_queues(dev
))
5964 strcpy(dev
->name
, name
);
5965 dev
->group
= INIT_NETDEV_GROUP
;
5973 free_percpu(dev
->pcpu_refcnt
);
5983 EXPORT_SYMBOL(alloc_netdev_mqs
);
5986 * free_netdev - free network device
5989 * This function does the last stage of destroying an allocated device
5990 * interface. The reference to the device object is released.
5991 * If this is the last reference then it will be freed.
5993 void free_netdev(struct net_device
*dev
)
5995 struct napi_struct
*p
, *n
;
5997 release_net(dev_net(dev
));
6004 kfree(rcu_dereference_protected(dev
->ingress_queue
, 1));
6006 /* Flush device addresses */
6007 dev_addr_flush(dev
);
6009 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
6012 free_percpu(dev
->pcpu_refcnt
);
6013 dev
->pcpu_refcnt
= NULL
;
6015 /* Compatibility with error handling in drivers */
6016 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
6017 kfree((char *)dev
- dev
->padded
);
6021 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
6022 dev
->reg_state
= NETREG_RELEASED
;
6024 /* will free via device release */
6025 put_device(&dev
->dev
);
6027 EXPORT_SYMBOL(free_netdev
);
6030 * synchronize_net - Synchronize with packet receive processing
6032 * Wait for packets currently being received to be done.
6033 * Does not block later packets from starting.
6035 void synchronize_net(void)
6038 if (rtnl_is_locked())
6039 synchronize_rcu_expedited();
6043 EXPORT_SYMBOL(synchronize_net
);
6046 * unregister_netdevice_queue - remove device from the kernel
6050 * This function shuts down a device interface and removes it
6051 * from the kernel tables.
6052 * If head not NULL, device is queued to be unregistered later.
6054 * Callers must hold the rtnl semaphore. You may want
6055 * unregister_netdev() instead of this.
6058 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
6063 list_move_tail(&dev
->unreg_list
, head
);
6065 rollback_registered(dev
);
6066 /* Finish processing unregister after unlock */
6070 EXPORT_SYMBOL(unregister_netdevice_queue
);
6073 * unregister_netdevice_many - unregister many devices
6074 * @head: list of devices
6076 void unregister_netdevice_many(struct list_head
*head
)
6078 struct net_device
*dev
;
6080 if (!list_empty(head
)) {
6081 rollback_registered_many(head
);
6082 list_for_each_entry(dev
, head
, unreg_list
)
6086 EXPORT_SYMBOL(unregister_netdevice_many
);
6089 * unregister_netdev - remove device from the kernel
6092 * This function shuts down a device interface and removes it
6093 * from the kernel tables.
6095 * This is just a wrapper for unregister_netdevice that takes
6096 * the rtnl semaphore. In general you want to use this and not
6097 * unregister_netdevice.
6099 void unregister_netdev(struct net_device
*dev
)
6102 unregister_netdevice(dev
);
6105 EXPORT_SYMBOL(unregister_netdev
);
6108 * dev_change_net_namespace - move device to different nethost namespace
6110 * @net: network namespace
6111 * @pat: If not NULL name pattern to try if the current device name
6112 * is already taken in the destination network namespace.
6114 * This function shuts down a device interface and moves it
6115 * to a new network namespace. On success 0 is returned, on
6116 * a failure a netagive errno code is returned.
6118 * Callers must hold the rtnl semaphore.
6121 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
6127 /* Don't allow namespace local devices to be moved. */
6129 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6132 /* Ensure the device has been registrered */
6134 if (dev
->reg_state
!= NETREG_REGISTERED
)
6137 /* Get out if there is nothing todo */
6139 if (net_eq(dev_net(dev
), net
))
6142 /* Pick the destination device name, and ensure
6143 * we can use it in the destination network namespace.
6146 if (__dev_get_by_name(net
, dev
->name
)) {
6147 /* We get here if we can't use the current device name */
6150 if (dev_get_valid_name(dev
, pat
) < 0)
6155 * And now a mini version of register_netdevice unregister_netdevice.
6158 /* If device is running close it first. */
6161 /* And unlink it from device chain */
6163 unlist_netdevice(dev
);
6167 /* Shutdown queueing discipline. */
6170 /* Notify protocols, that we are about to destroy
6171 this device. They should clean all the things.
6173 Note that dev->reg_state stays at NETREG_REGISTERED.
6174 This is wanted because this way 8021q and macvlan know
6175 the device is just moving and can keep their slaves up.
6177 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6178 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
6179 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
6182 * Flush the unicast and multicast chains
6187 /* Actually switch the network namespace */
6188 dev_net_set(dev
, net
);
6190 /* If there is an ifindex conflict assign a new one */
6191 if (__dev_get_by_index(net
, dev
->ifindex
)) {
6192 int iflink
= (dev
->iflink
== dev
->ifindex
);
6193 dev
->ifindex
= dev_new_index(net
);
6195 dev
->iflink
= dev
->ifindex
;
6198 /* Fixup kobjects */
6199 err
= device_rename(&dev
->dev
, dev
->name
);
6202 /* Add the device back in the hashes */
6203 list_netdevice(dev
);
6205 /* Notify protocols, that a new device appeared. */
6206 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
6209 * Prevent userspace races by waiting until the network
6210 * device is fully setup before sending notifications.
6212 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
6219 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
6221 static int dev_cpu_callback(struct notifier_block
*nfb
,
6222 unsigned long action
,
6225 struct sk_buff
**list_skb
;
6226 struct sk_buff
*skb
;
6227 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
6228 struct softnet_data
*sd
, *oldsd
;
6230 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
6233 local_irq_disable();
6234 cpu
= smp_processor_id();
6235 sd
= &per_cpu(softnet_data
, cpu
);
6236 oldsd
= &per_cpu(softnet_data
, oldcpu
);
6238 /* Find end of our completion_queue. */
6239 list_skb
= &sd
->completion_queue
;
6241 list_skb
= &(*list_skb
)->next
;
6242 /* Append completion queue from offline CPU. */
6243 *list_skb
= oldsd
->completion_queue
;
6244 oldsd
->completion_queue
= NULL
;
6246 /* Append output queue from offline CPU. */
6247 if (oldsd
->output_queue
) {
6248 *sd
->output_queue_tailp
= oldsd
->output_queue
;
6249 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
6250 oldsd
->output_queue
= NULL
;
6251 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
6253 /* Append NAPI poll list from offline CPU. */
6254 if (!list_empty(&oldsd
->poll_list
)) {
6255 list_splice_init(&oldsd
->poll_list
, &sd
->poll_list
);
6256 raise_softirq_irqoff(NET_RX_SOFTIRQ
);
6259 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
6262 /* Process offline CPU's input_pkt_queue */
6263 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
6265 input_queue_head_incr(oldsd
);
6267 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
6269 input_queue_head_incr(oldsd
);
6277 * netdev_increment_features - increment feature set by one
6278 * @all: current feature set
6279 * @one: new feature set
6280 * @mask: mask feature set
6282 * Computes a new feature set after adding a device with feature set
6283 * @one to the master device with current feature set @all. Will not
6284 * enable anything that is off in @mask. Returns the new feature set.
6286 netdev_features_t
netdev_increment_features(netdev_features_t all
,
6287 netdev_features_t one
, netdev_features_t mask
)
6289 if (mask
& NETIF_F_GEN_CSUM
)
6290 mask
|= NETIF_F_ALL_CSUM
;
6291 mask
|= NETIF_F_VLAN_CHALLENGED
;
6293 all
|= one
& (NETIF_F_ONE_FOR_ALL
|NETIF_F_ALL_CSUM
) & mask
;
6294 all
&= one
| ~NETIF_F_ALL_FOR_ALL
;
6296 /* If one device supports hw checksumming, set for all. */
6297 if (all
& NETIF_F_GEN_CSUM
)
6298 all
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_GEN_CSUM
);
6302 EXPORT_SYMBOL(netdev_increment_features
);
6304 static struct hlist_head
*netdev_create_hash(void)
6307 struct hlist_head
*hash
;
6309 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
6311 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
6312 INIT_HLIST_HEAD(&hash
[i
]);
6317 /* Initialize per network namespace state */
6318 static int __net_init
netdev_init(struct net
*net
)
6320 if (net
!= &init_net
)
6321 INIT_LIST_HEAD(&net
->dev_base_head
);
6323 net
->dev_name_head
= netdev_create_hash();
6324 if (net
->dev_name_head
== NULL
)
6327 net
->dev_index_head
= netdev_create_hash();
6328 if (net
->dev_index_head
== NULL
)
6334 kfree(net
->dev_name_head
);
6340 * netdev_drivername - network driver for the device
6341 * @dev: network device
6343 * Determine network driver for device.
6345 const char *netdev_drivername(const struct net_device
*dev
)
6347 const struct device_driver
*driver
;
6348 const struct device
*parent
;
6349 const char *empty
= "";
6351 parent
= dev
->dev
.parent
;
6355 driver
= parent
->driver
;
6356 if (driver
&& driver
->name
)
6357 return driver
->name
;
6361 int __netdev_printk(const char *level
, const struct net_device
*dev
,
6362 struct va_format
*vaf
)
6366 if (dev
&& dev
->dev
.parent
)
6367 r
= dev_printk(level
, dev
->dev
.parent
, "%s: %pV",
6368 netdev_name(dev
), vaf
);
6370 r
= printk("%s%s: %pV", level
, netdev_name(dev
), vaf
);
6372 r
= printk("%s(NULL net_device): %pV", level
, vaf
);
6376 EXPORT_SYMBOL(__netdev_printk
);
6378 int netdev_printk(const char *level
, const struct net_device
*dev
,
6379 const char *format
, ...)
6381 struct va_format vaf
;
6385 va_start(args
, format
);
6390 r
= __netdev_printk(level
, dev
, &vaf
);
6395 EXPORT_SYMBOL(netdev_printk
);
6397 #define define_netdev_printk_level(func, level) \
6398 int func(const struct net_device *dev, const char *fmt, ...) \
6401 struct va_format vaf; \
6404 va_start(args, fmt); \
6409 r = __netdev_printk(level, dev, &vaf); \
6414 EXPORT_SYMBOL(func);
6416 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
6417 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
6418 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
6419 define_netdev_printk_level(netdev_err
, KERN_ERR
);
6420 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
6421 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
6422 define_netdev_printk_level(netdev_info
, KERN_INFO
);
6424 static void __net_exit
netdev_exit(struct net
*net
)
6426 kfree(net
->dev_name_head
);
6427 kfree(net
->dev_index_head
);
6430 static struct pernet_operations __net_initdata netdev_net_ops
= {
6431 .init
= netdev_init
,
6432 .exit
= netdev_exit
,
6435 static void __net_exit
default_device_exit(struct net
*net
)
6437 struct net_device
*dev
, *aux
;
6439 * Push all migratable network devices back to the
6440 * initial network namespace
6443 for_each_netdev_safe(net
, dev
, aux
) {
6445 char fb_name
[IFNAMSIZ
];
6447 /* Ignore unmoveable devices (i.e. loopback) */
6448 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6451 /* Leave virtual devices for the generic cleanup */
6452 if (dev
->rtnl_link_ops
)
6455 /* Push remaining network devices to init_net */
6456 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
6457 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
6459 pr_emerg("%s: failed to move %s to init_net: %d\n",
6460 __func__
, dev
->name
, err
);
6467 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
6469 /* At exit all network devices most be removed from a network
6470 * namespace. Do this in the reverse order of registration.
6471 * Do this across as many network namespaces as possible to
6472 * improve batching efficiency.
6474 struct net_device
*dev
;
6476 LIST_HEAD(dev_kill_list
);
6479 list_for_each_entry(net
, net_list
, exit_list
) {
6480 for_each_netdev_reverse(net
, dev
) {
6481 if (dev
->rtnl_link_ops
)
6482 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
6484 unregister_netdevice_queue(dev
, &dev_kill_list
);
6487 unregister_netdevice_many(&dev_kill_list
);
6488 list_del(&dev_kill_list
);
6492 static struct pernet_operations __net_initdata default_device_ops
= {
6493 .exit
= default_device_exit
,
6494 .exit_batch
= default_device_exit_batch
,
6498 * Initialize the DEV module. At boot time this walks the device list and
6499 * unhooks any devices that fail to initialise (normally hardware not
6500 * present) and leaves us with a valid list of present and active devices.
6505 * This is called single threaded during boot, so no need
6506 * to take the rtnl semaphore.
6508 static int __init
net_dev_init(void)
6510 int i
, rc
= -ENOMEM
;
6512 BUG_ON(!dev_boot_phase
);
6514 if (dev_proc_init())
6517 if (netdev_kobject_init())
6520 INIT_LIST_HEAD(&ptype_all
);
6521 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
6522 INIT_LIST_HEAD(&ptype_base
[i
]);
6524 if (register_pernet_subsys(&netdev_net_ops
))
6528 * Initialise the packet receive queues.
6531 for_each_possible_cpu(i
) {
6532 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
6534 memset(sd
, 0, sizeof(*sd
));
6535 skb_queue_head_init(&sd
->input_pkt_queue
);
6536 skb_queue_head_init(&sd
->process_queue
);
6537 sd
->completion_queue
= NULL
;
6538 INIT_LIST_HEAD(&sd
->poll_list
);
6539 sd
->output_queue
= NULL
;
6540 sd
->output_queue_tailp
= &sd
->output_queue
;
6542 sd
->csd
.func
= rps_trigger_softirq
;
6548 sd
->backlog
.poll
= process_backlog
;
6549 sd
->backlog
.weight
= weight_p
;
6550 sd
->backlog
.gro_list
= NULL
;
6551 sd
->backlog
.gro_count
= 0;
6556 /* The loopback device is special if any other network devices
6557 * is present in a network namespace the loopback device must
6558 * be present. Since we now dynamically allocate and free the
6559 * loopback device ensure this invariant is maintained by
6560 * keeping the loopback device as the first device on the
6561 * list of network devices. Ensuring the loopback devices
6562 * is the first device that appears and the last network device
6565 if (register_pernet_device(&loopback_net_ops
))
6568 if (register_pernet_device(&default_device_ops
))
6571 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
6572 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
6574 hotcpu_notifier(dev_cpu_callback
, 0);
6582 subsys_initcall(net_dev_init
);
6584 static int __init
initialize_hashrnd(void)
6586 get_random_bytes(&hashrnd
, sizeof(hashrnd
));
6590 late_initcall_sync(initialize_hashrnd
);