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
);
3159 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3160 * the special handling of PFMEMALLOC skbs.
3162 static bool skb_pfmemalloc_protocol(struct sk_buff
*skb
)
3164 switch (skb
->protocol
) {
3165 case __constant_htons(ETH_P_ARP
):
3166 case __constant_htons(ETH_P_IP
):
3167 case __constant_htons(ETH_P_IPV6
):
3168 case __constant_htons(ETH_P_8021Q
):
3175 static int __netif_receive_skb(struct sk_buff
*skb
)
3177 struct packet_type
*ptype
, *pt_prev
;
3178 rx_handler_func_t
*rx_handler
;
3179 struct net_device
*orig_dev
;
3180 struct net_device
*null_or_dev
;
3181 bool deliver_exact
= false;
3182 int ret
= NET_RX_DROP
;
3184 unsigned long pflags
= current
->flags
;
3186 net_timestamp_check(!netdev_tstamp_prequeue
, skb
);
3188 trace_netif_receive_skb(skb
);
3191 * PFMEMALLOC skbs are special, they should
3192 * - be delivered to SOCK_MEMALLOC sockets only
3193 * - stay away from userspace
3194 * - have bounded memory usage
3196 * Use PF_MEMALLOC as this saves us from propagating the allocation
3197 * context down to all allocation sites.
3199 if (sk_memalloc_socks() && skb_pfmemalloc(skb
))
3200 current
->flags
|= PF_MEMALLOC
;
3202 /* if we've gotten here through NAPI, check netpoll */
3203 if (netpoll_receive_skb(skb
))
3206 orig_dev
= skb
->dev
;
3208 skb_reset_network_header(skb
);
3209 skb_reset_transport_header(skb
);
3210 skb_reset_mac_len(skb
);
3217 skb
->skb_iif
= skb
->dev
->ifindex
;
3219 __this_cpu_inc(softnet_data
.processed
);
3221 if (skb
->protocol
== cpu_to_be16(ETH_P_8021Q
)) {
3222 skb
= vlan_untag(skb
);
3227 #ifdef CONFIG_NET_CLS_ACT
3228 if (skb
->tc_verd
& TC_NCLS
) {
3229 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
3234 if (sk_memalloc_socks() && skb_pfmemalloc(skb
))
3237 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
3238 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
3240 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3246 #ifdef CONFIG_NET_CLS_ACT
3247 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
3253 if (sk_memalloc_socks() && skb_pfmemalloc(skb
)
3254 && !skb_pfmemalloc_protocol(skb
))
3257 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
3258 if (vlan_tx_tag_present(skb
)) {
3260 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3263 if (vlan_do_receive(&skb
, !rx_handler
))
3265 else if (unlikely(!skb
))
3271 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3274 switch (rx_handler(&skb
)) {
3275 case RX_HANDLER_CONSUMED
:
3277 case RX_HANDLER_ANOTHER
:
3279 case RX_HANDLER_EXACT
:
3280 deliver_exact
= true;
3281 case RX_HANDLER_PASS
:
3288 /* deliver only exact match when indicated */
3289 null_or_dev
= deliver_exact
? skb
->dev
: NULL
;
3291 type
= skb
->protocol
;
3292 list_for_each_entry_rcu(ptype
,
3293 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
3294 if (ptype
->type
== type
&&
3295 (ptype
->dev
== null_or_dev
|| ptype
->dev
== skb
->dev
||
3296 ptype
->dev
== orig_dev
)) {
3298 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3304 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
3307 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
3310 atomic_long_inc(&skb
->dev
->rx_dropped
);
3312 /* Jamal, now you will not able to escape explaining
3313 * me how you were going to use this. :-)
3321 tsk_restore_flags(current
, pflags
, PF_MEMALLOC
);
3326 * netif_receive_skb - process receive buffer from network
3327 * @skb: buffer to process
3329 * netif_receive_skb() is the main receive data processing function.
3330 * It always succeeds. The buffer may be dropped during processing
3331 * for congestion control or by the protocol layers.
3333 * This function may only be called from softirq context and interrupts
3334 * should be enabled.
3336 * Return values (usually ignored):
3337 * NET_RX_SUCCESS: no congestion
3338 * NET_RX_DROP: packet was dropped
3340 int netif_receive_skb(struct sk_buff
*skb
)
3342 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3344 if (skb_defer_rx_timestamp(skb
))
3345 return NET_RX_SUCCESS
;
3348 if (static_key_false(&rps_needed
)) {
3349 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3354 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3357 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3364 return __netif_receive_skb(skb
);
3366 EXPORT_SYMBOL(netif_receive_skb
);
3368 /* Network device is going away, flush any packets still pending
3369 * Called with irqs disabled.
3371 static void flush_backlog(void *arg
)
3373 struct net_device
*dev
= arg
;
3374 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3375 struct sk_buff
*skb
, *tmp
;
3378 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3379 if (skb
->dev
== dev
) {
3380 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3382 input_queue_head_incr(sd
);
3387 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3388 if (skb
->dev
== dev
) {
3389 __skb_unlink(skb
, &sd
->process_queue
);
3391 input_queue_head_incr(sd
);
3396 static int napi_gro_complete(struct sk_buff
*skb
)
3398 struct packet_type
*ptype
;
3399 __be16 type
= skb
->protocol
;
3400 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3403 if (NAPI_GRO_CB(skb
)->count
== 1) {
3404 skb_shinfo(skb
)->gso_size
= 0;
3409 list_for_each_entry_rcu(ptype
, head
, list
) {
3410 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_complete
)
3413 err
= ptype
->gro_complete(skb
);
3419 WARN_ON(&ptype
->list
== head
);
3421 return NET_RX_SUCCESS
;
3425 return netif_receive_skb(skb
);
3428 inline void napi_gro_flush(struct napi_struct
*napi
)
3430 struct sk_buff
*skb
, *next
;
3432 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3435 napi_gro_complete(skb
);
3438 napi
->gro_count
= 0;
3439 napi
->gro_list
= NULL
;
3441 EXPORT_SYMBOL(napi_gro_flush
);
3443 enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3445 struct sk_buff
**pp
= NULL
;
3446 struct packet_type
*ptype
;
3447 __be16 type
= skb
->protocol
;
3448 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3451 enum gro_result ret
;
3453 if (!(skb
->dev
->features
& NETIF_F_GRO
) || netpoll_rx_on(skb
))
3456 if (skb_is_gso(skb
) || skb_has_frag_list(skb
))
3460 list_for_each_entry_rcu(ptype
, head
, list
) {
3461 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_receive
)
3464 skb_set_network_header(skb
, skb_gro_offset(skb
));
3465 mac_len
= skb
->network_header
- skb
->mac_header
;
3466 skb
->mac_len
= mac_len
;
3467 NAPI_GRO_CB(skb
)->same_flow
= 0;
3468 NAPI_GRO_CB(skb
)->flush
= 0;
3469 NAPI_GRO_CB(skb
)->free
= 0;
3471 pp
= ptype
->gro_receive(&napi
->gro_list
, skb
);
3476 if (&ptype
->list
== head
)
3479 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3480 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3483 struct sk_buff
*nskb
= *pp
;
3487 napi_gro_complete(nskb
);
3494 if (NAPI_GRO_CB(skb
)->flush
|| napi
->gro_count
>= MAX_GRO_SKBS
)
3498 NAPI_GRO_CB(skb
)->count
= 1;
3499 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
3500 skb
->next
= napi
->gro_list
;
3501 napi
->gro_list
= skb
;
3505 if (skb_headlen(skb
) < skb_gro_offset(skb
)) {
3506 int grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
3508 BUG_ON(skb
->end
- skb
->tail
< grow
);
3510 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3513 skb
->data_len
-= grow
;
3515 skb_shinfo(skb
)->frags
[0].page_offset
+= grow
;
3516 skb_frag_size_sub(&skb_shinfo(skb
)->frags
[0], grow
);
3518 if (unlikely(!skb_frag_size(&skb_shinfo(skb
)->frags
[0]))) {
3519 skb_frag_unref(skb
, 0);
3520 memmove(skb_shinfo(skb
)->frags
,
3521 skb_shinfo(skb
)->frags
+ 1,
3522 --skb_shinfo(skb
)->nr_frags
* sizeof(skb_frag_t
));
3533 EXPORT_SYMBOL(dev_gro_receive
);
3535 static inline gro_result_t
3536 __napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3539 unsigned int maclen
= skb
->dev
->hard_header_len
;
3541 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3542 unsigned long diffs
;
3544 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
3545 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
3546 if (maclen
== ETH_HLEN
)
3547 diffs
|= compare_ether_header(skb_mac_header(p
),
3548 skb_gro_mac_header(skb
));
3550 diffs
= memcmp(skb_mac_header(p
),
3551 skb_gro_mac_header(skb
),
3553 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
3554 NAPI_GRO_CB(p
)->flush
= 0;
3557 return dev_gro_receive(napi
, skb
);
3560 gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
3564 if (netif_receive_skb(skb
))
3572 case GRO_MERGED_FREE
:
3573 if (NAPI_GRO_CB(skb
)->free
== NAPI_GRO_FREE_STOLEN_HEAD
)
3574 kmem_cache_free(skbuff_head_cache
, skb
);
3586 EXPORT_SYMBOL(napi_skb_finish
);
3588 void skb_gro_reset_offset(struct sk_buff
*skb
)
3590 NAPI_GRO_CB(skb
)->data_offset
= 0;
3591 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3592 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3594 if (skb
->mac_header
== skb
->tail
&&
3595 !PageHighMem(skb_frag_page(&skb_shinfo(skb
)->frags
[0]))) {
3596 NAPI_GRO_CB(skb
)->frag0
=
3597 skb_frag_address(&skb_shinfo(skb
)->frags
[0]);
3598 NAPI_GRO_CB(skb
)->frag0_len
= skb_frag_size(&skb_shinfo(skb
)->frags
[0]);
3601 EXPORT_SYMBOL(skb_gro_reset_offset
);
3603 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3605 skb_gro_reset_offset(skb
);
3607 return napi_skb_finish(__napi_gro_receive(napi
, skb
), skb
);
3609 EXPORT_SYMBOL(napi_gro_receive
);
3611 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
3613 __skb_pull(skb
, skb_headlen(skb
));
3614 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3615 skb_reserve(skb
, NET_SKB_PAD
+ NET_IP_ALIGN
- skb_headroom(skb
));
3617 skb
->dev
= napi
->dev
;
3623 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
3625 struct sk_buff
*skb
= napi
->skb
;
3628 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
3634 EXPORT_SYMBOL(napi_get_frags
);
3636 gro_result_t
napi_frags_finish(struct napi_struct
*napi
, struct sk_buff
*skb
,
3642 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
3644 if (ret
== GRO_HELD
)
3645 skb_gro_pull(skb
, -ETH_HLEN
);
3646 else if (netif_receive_skb(skb
))
3651 case GRO_MERGED_FREE
:
3652 napi_reuse_skb(napi
, skb
);
3661 EXPORT_SYMBOL(napi_frags_finish
);
3663 static struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
3665 struct sk_buff
*skb
= napi
->skb
;
3672 skb_reset_mac_header(skb
);
3673 skb_gro_reset_offset(skb
);
3675 off
= skb_gro_offset(skb
);
3676 hlen
= off
+ sizeof(*eth
);
3677 eth
= skb_gro_header_fast(skb
, off
);
3678 if (skb_gro_header_hard(skb
, hlen
)) {
3679 eth
= skb_gro_header_slow(skb
, hlen
, off
);
3680 if (unlikely(!eth
)) {
3681 napi_reuse_skb(napi
, skb
);
3687 skb_gro_pull(skb
, sizeof(*eth
));
3690 * This works because the only protocols we care about don't require
3691 * special handling. We'll fix it up properly at the end.
3693 skb
->protocol
= eth
->h_proto
;
3699 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
3701 struct sk_buff
*skb
= napi_frags_skb(napi
);
3706 return napi_frags_finish(napi
, skb
, __napi_gro_receive(napi
, skb
));
3708 EXPORT_SYMBOL(napi_gro_frags
);
3711 * net_rps_action sends any pending IPI's for rps.
3712 * Note: called with local irq disabled, but exits with local irq enabled.
3714 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
3717 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
3720 sd
->rps_ipi_list
= NULL
;
3724 /* Send pending IPI's to kick RPS processing on remote cpus. */
3726 struct softnet_data
*next
= remsd
->rps_ipi_next
;
3728 if (cpu_online(remsd
->cpu
))
3729 __smp_call_function_single(remsd
->cpu
,
3738 static int process_backlog(struct napi_struct
*napi
, int quota
)
3741 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
3744 /* Check if we have pending ipi, its better to send them now,
3745 * not waiting net_rx_action() end.
3747 if (sd
->rps_ipi_list
) {
3748 local_irq_disable();
3749 net_rps_action_and_irq_enable(sd
);
3752 napi
->weight
= weight_p
;
3753 local_irq_disable();
3754 while (work
< quota
) {
3755 struct sk_buff
*skb
;
3758 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
3760 __netif_receive_skb(skb
);
3761 local_irq_disable();
3762 input_queue_head_incr(sd
);
3763 if (++work
>= quota
) {
3770 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3772 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
3773 &sd
->process_queue
);
3775 if (qlen
< quota
- work
) {
3777 * Inline a custom version of __napi_complete().
3778 * only current cpu owns and manipulates this napi,
3779 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3780 * we can use a plain write instead of clear_bit(),
3781 * and we dont need an smp_mb() memory barrier.
3783 list_del(&napi
->poll_list
);
3786 quota
= work
+ qlen
;
3796 * __napi_schedule - schedule for receive
3797 * @n: entry to schedule
3799 * The entry's receive function will be scheduled to run
3801 void __napi_schedule(struct napi_struct
*n
)
3803 unsigned long flags
;
3805 local_irq_save(flags
);
3806 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
3807 local_irq_restore(flags
);
3809 EXPORT_SYMBOL(__napi_schedule
);
3811 void __napi_complete(struct napi_struct
*n
)
3813 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
3814 BUG_ON(n
->gro_list
);
3816 list_del(&n
->poll_list
);
3817 smp_mb__before_clear_bit();
3818 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
3820 EXPORT_SYMBOL(__napi_complete
);
3822 void napi_complete(struct napi_struct
*n
)
3824 unsigned long flags
;
3827 * don't let napi dequeue from the cpu poll list
3828 * just in case its running on a different cpu
3830 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
3834 local_irq_save(flags
);
3836 local_irq_restore(flags
);
3838 EXPORT_SYMBOL(napi_complete
);
3840 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
3841 int (*poll
)(struct napi_struct
*, int), int weight
)
3843 INIT_LIST_HEAD(&napi
->poll_list
);
3844 napi
->gro_count
= 0;
3845 napi
->gro_list
= NULL
;
3848 napi
->weight
= weight
;
3849 list_add(&napi
->dev_list
, &dev
->napi_list
);
3851 #ifdef CONFIG_NETPOLL
3852 spin_lock_init(&napi
->poll_lock
);
3853 napi
->poll_owner
= -1;
3855 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
3857 EXPORT_SYMBOL(netif_napi_add
);
3859 void netif_napi_del(struct napi_struct
*napi
)
3861 struct sk_buff
*skb
, *next
;
3863 list_del_init(&napi
->dev_list
);
3864 napi_free_frags(napi
);
3866 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3872 napi
->gro_list
= NULL
;
3873 napi
->gro_count
= 0;
3875 EXPORT_SYMBOL(netif_napi_del
);
3877 static void net_rx_action(struct softirq_action
*h
)
3879 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3880 unsigned long time_limit
= jiffies
+ 2;
3881 int budget
= netdev_budget
;
3884 local_irq_disable();
3886 while (!list_empty(&sd
->poll_list
)) {
3887 struct napi_struct
*n
;
3890 /* If softirq window is exhuasted then punt.
3891 * Allow this to run for 2 jiffies since which will allow
3892 * an average latency of 1.5/HZ.
3894 if (unlikely(budget
<= 0 || time_after(jiffies
, time_limit
)))
3899 /* Even though interrupts have been re-enabled, this
3900 * access is safe because interrupts can only add new
3901 * entries to the tail of this list, and only ->poll()
3902 * calls can remove this head entry from the list.
3904 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
3906 have
= netpoll_poll_lock(n
);
3910 /* This NAPI_STATE_SCHED test is for avoiding a race
3911 * with netpoll's poll_napi(). Only the entity which
3912 * obtains the lock and sees NAPI_STATE_SCHED set will
3913 * actually make the ->poll() call. Therefore we avoid
3914 * accidentally calling ->poll() when NAPI is not scheduled.
3917 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
3918 work
= n
->poll(n
, weight
);
3922 WARN_ON_ONCE(work
> weight
);
3926 local_irq_disable();
3928 /* Drivers must not modify the NAPI state if they
3929 * consume the entire weight. In such cases this code
3930 * still "owns" the NAPI instance and therefore can
3931 * move the instance around on the list at-will.
3933 if (unlikely(work
== weight
)) {
3934 if (unlikely(napi_disable_pending(n
))) {
3937 local_irq_disable();
3939 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
3942 netpoll_poll_unlock(have
);
3945 net_rps_action_and_irq_enable(sd
);
3947 #ifdef CONFIG_NET_DMA
3949 * There may not be any more sk_buffs coming right now, so push
3950 * any pending DMA copies to hardware
3952 dma_issue_pending_all();
3959 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3963 static gifconf_func_t
*gifconf_list
[NPROTO
];
3966 * register_gifconf - register a SIOCGIF handler
3967 * @family: Address family
3968 * @gifconf: Function handler
3970 * Register protocol dependent address dumping routines. The handler
3971 * that is passed must not be freed or reused until it has been replaced
3972 * by another handler.
3974 int register_gifconf(unsigned int family
, gifconf_func_t
*gifconf
)
3976 if (family
>= NPROTO
)
3978 gifconf_list
[family
] = gifconf
;
3981 EXPORT_SYMBOL(register_gifconf
);
3985 * Map an interface index to its name (SIOCGIFNAME)
3989 * We need this ioctl for efficient implementation of the
3990 * if_indextoname() function required by the IPv6 API. Without
3991 * it, we would have to search all the interfaces to find a
3995 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
3997 struct net_device
*dev
;
4001 * Fetch the caller's info block.
4004 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
4008 dev
= dev_get_by_index_rcu(net
, ifr
.ifr_ifindex
);
4014 strcpy(ifr
.ifr_name
, dev
->name
);
4017 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
4023 * Perform a SIOCGIFCONF call. This structure will change
4024 * size eventually, and there is nothing I can do about it.
4025 * Thus we will need a 'compatibility mode'.
4028 static int dev_ifconf(struct net
*net
, char __user
*arg
)
4031 struct net_device
*dev
;
4038 * Fetch the caller's info block.
4041 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
4048 * Loop over the interfaces, and write an info block for each.
4052 for_each_netdev(net
, dev
) {
4053 for (i
= 0; i
< NPROTO
; i
++) {
4054 if (gifconf_list
[i
]) {
4057 done
= gifconf_list
[i
](dev
, NULL
, 0);
4059 done
= gifconf_list
[i
](dev
, pos
+ total
,
4069 * All done. Write the updated control block back to the caller.
4071 ifc
.ifc_len
= total
;
4074 * Both BSD and Solaris return 0 here, so we do too.
4076 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
4079 #ifdef CONFIG_PROC_FS
4081 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4083 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4084 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4085 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4087 static inline struct net_device
*dev_from_same_bucket(struct seq_file
*seq
, loff_t
*pos
)
4089 struct net
*net
= seq_file_net(seq
);
4090 struct net_device
*dev
;
4091 struct hlist_node
*p
;
4092 struct hlist_head
*h
;
4093 unsigned int count
= 0, offset
= get_offset(*pos
);
4095 h
= &net
->dev_name_head
[get_bucket(*pos
)];
4096 hlist_for_each_entry_rcu(dev
, p
, h
, name_hlist
) {
4097 if (++count
== offset
)
4104 static inline struct net_device
*dev_from_bucket(struct seq_file
*seq
, loff_t
*pos
)
4106 struct net_device
*dev
;
4107 unsigned int bucket
;
4110 dev
= dev_from_same_bucket(seq
, pos
);
4114 bucket
= get_bucket(*pos
) + 1;
4115 *pos
= set_bucket_offset(bucket
, 1);
4116 } while (bucket
< NETDEV_HASHENTRIES
);
4122 * This is invoked by the /proc filesystem handler to display a device
4125 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4130 return SEQ_START_TOKEN
;
4132 if (get_bucket(*pos
) >= NETDEV_HASHENTRIES
)
4135 return dev_from_bucket(seq
, pos
);
4138 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4141 return dev_from_bucket(seq
, pos
);
4144 void dev_seq_stop(struct seq_file
*seq
, void *v
)
4150 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
4152 struct rtnl_link_stats64 temp
;
4153 const struct rtnl_link_stats64
*stats
= dev_get_stats(dev
, &temp
);
4155 seq_printf(seq
, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4156 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4157 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
4159 stats
->rx_dropped
+ stats
->rx_missed_errors
,
4160 stats
->rx_fifo_errors
,
4161 stats
->rx_length_errors
+ stats
->rx_over_errors
+
4162 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
4163 stats
->rx_compressed
, stats
->multicast
,
4164 stats
->tx_bytes
, stats
->tx_packets
,
4165 stats
->tx_errors
, stats
->tx_dropped
,
4166 stats
->tx_fifo_errors
, stats
->collisions
,
4167 stats
->tx_carrier_errors
+
4168 stats
->tx_aborted_errors
+
4169 stats
->tx_window_errors
+
4170 stats
->tx_heartbeat_errors
,
4171 stats
->tx_compressed
);
4175 * Called from the PROCfs module. This now uses the new arbitrary sized
4176 * /proc/net interface to create /proc/net/dev
4178 static int dev_seq_show(struct seq_file
*seq
, void *v
)
4180 if (v
== SEQ_START_TOKEN
)
4181 seq_puts(seq
, "Inter-| Receive "
4183 " face |bytes packets errs drop fifo frame "
4184 "compressed multicast|bytes packets errs "
4185 "drop fifo colls carrier compressed\n");
4187 dev_seq_printf_stats(seq
, v
);
4191 static struct softnet_data
*softnet_get_online(loff_t
*pos
)
4193 struct softnet_data
*sd
= NULL
;
4195 while (*pos
< nr_cpu_ids
)
4196 if (cpu_online(*pos
)) {
4197 sd
= &per_cpu(softnet_data
, *pos
);
4204 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4206 return softnet_get_online(pos
);
4209 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4212 return softnet_get_online(pos
);
4215 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
4219 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
4221 struct softnet_data
*sd
= v
;
4223 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4224 sd
->processed
, sd
->dropped
, sd
->time_squeeze
, 0,
4225 0, 0, 0, 0, /* was fastroute */
4226 sd
->cpu_collision
, sd
->received_rps
);
4230 static const struct seq_operations dev_seq_ops
= {
4231 .start
= dev_seq_start
,
4232 .next
= dev_seq_next
,
4233 .stop
= dev_seq_stop
,
4234 .show
= dev_seq_show
,
4237 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
4239 return seq_open_net(inode
, file
, &dev_seq_ops
,
4240 sizeof(struct seq_net_private
));
4243 static const struct file_operations dev_seq_fops
= {
4244 .owner
= THIS_MODULE
,
4245 .open
= dev_seq_open
,
4247 .llseek
= seq_lseek
,
4248 .release
= seq_release_net
,
4251 static const struct seq_operations softnet_seq_ops
= {
4252 .start
= softnet_seq_start
,
4253 .next
= softnet_seq_next
,
4254 .stop
= softnet_seq_stop
,
4255 .show
= softnet_seq_show
,
4258 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
4260 return seq_open(file
, &softnet_seq_ops
);
4263 static const struct file_operations softnet_seq_fops
= {
4264 .owner
= THIS_MODULE
,
4265 .open
= softnet_seq_open
,
4267 .llseek
= seq_lseek
,
4268 .release
= seq_release
,
4271 static void *ptype_get_idx(loff_t pos
)
4273 struct packet_type
*pt
= NULL
;
4277 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
4283 for (t
= 0; t
< PTYPE_HASH_SIZE
; t
++) {
4284 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
4293 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4297 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
4300 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4302 struct packet_type
*pt
;
4303 struct list_head
*nxt
;
4307 if (v
== SEQ_START_TOKEN
)
4308 return ptype_get_idx(0);
4311 nxt
= pt
->list
.next
;
4312 if (pt
->type
== htons(ETH_P_ALL
)) {
4313 if (nxt
!= &ptype_all
)
4316 nxt
= ptype_base
[0].next
;
4318 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
4320 while (nxt
== &ptype_base
[hash
]) {
4321 if (++hash
>= PTYPE_HASH_SIZE
)
4323 nxt
= ptype_base
[hash
].next
;
4326 return list_entry(nxt
, struct packet_type
, list
);
4329 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
4335 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
4337 struct packet_type
*pt
= v
;
4339 if (v
== SEQ_START_TOKEN
)
4340 seq_puts(seq
, "Type Device Function\n");
4341 else if (pt
->dev
== NULL
|| dev_net(pt
->dev
) == seq_file_net(seq
)) {
4342 if (pt
->type
== htons(ETH_P_ALL
))
4343 seq_puts(seq
, "ALL ");
4345 seq_printf(seq
, "%04x", ntohs(pt
->type
));
4347 seq_printf(seq
, " %-8s %pF\n",
4348 pt
->dev
? pt
->dev
->name
: "", pt
->func
);
4354 static const struct seq_operations ptype_seq_ops
= {
4355 .start
= ptype_seq_start
,
4356 .next
= ptype_seq_next
,
4357 .stop
= ptype_seq_stop
,
4358 .show
= ptype_seq_show
,
4361 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
4363 return seq_open_net(inode
, file
, &ptype_seq_ops
,
4364 sizeof(struct seq_net_private
));
4367 static const struct file_operations ptype_seq_fops
= {
4368 .owner
= THIS_MODULE
,
4369 .open
= ptype_seq_open
,
4371 .llseek
= seq_lseek
,
4372 .release
= seq_release_net
,
4376 static int __net_init
dev_proc_net_init(struct net
*net
)
4380 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
4382 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
4384 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
4387 if (wext_proc_init(net
))
4393 proc_net_remove(net
, "ptype");
4395 proc_net_remove(net
, "softnet_stat");
4397 proc_net_remove(net
, "dev");
4401 static void __net_exit
dev_proc_net_exit(struct net
*net
)
4403 wext_proc_exit(net
);
4405 proc_net_remove(net
, "ptype");
4406 proc_net_remove(net
, "softnet_stat");
4407 proc_net_remove(net
, "dev");
4410 static struct pernet_operations __net_initdata dev_proc_ops
= {
4411 .init
= dev_proc_net_init
,
4412 .exit
= dev_proc_net_exit
,
4415 static int __init
dev_proc_init(void)
4417 return register_pernet_subsys(&dev_proc_ops
);
4420 #define dev_proc_init() 0
4421 #endif /* CONFIG_PROC_FS */
4425 * netdev_set_master - set up master pointer
4426 * @slave: slave device
4427 * @master: new master device
4429 * Changes the master device of the slave. Pass %NULL to break the
4430 * bonding. The caller must hold the RTNL semaphore. On a failure
4431 * a negative errno code is returned. On success the reference counts
4432 * are adjusted and the function returns zero.
4434 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
4436 struct net_device
*old
= slave
->master
;
4446 slave
->master
= master
;
4452 EXPORT_SYMBOL(netdev_set_master
);
4455 * netdev_set_bond_master - set up bonding master/slave pair
4456 * @slave: slave device
4457 * @master: new master device
4459 * Changes the master device of the slave. Pass %NULL to break the
4460 * bonding. The caller must hold the RTNL semaphore. On a failure
4461 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4462 * to the routing socket and the function returns zero.
4464 int netdev_set_bond_master(struct net_device
*slave
, struct net_device
*master
)
4470 err
= netdev_set_master(slave
, master
);
4474 slave
->flags
|= IFF_SLAVE
;
4476 slave
->flags
&= ~IFF_SLAVE
;
4478 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
4481 EXPORT_SYMBOL(netdev_set_bond_master
);
4483 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
4485 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4487 if ((dev
->flags
& IFF_UP
) && ops
->ndo_change_rx_flags
)
4488 ops
->ndo_change_rx_flags(dev
, flags
);
4491 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
4493 unsigned int old_flags
= dev
->flags
;
4499 dev
->flags
|= IFF_PROMISC
;
4500 dev
->promiscuity
+= inc
;
4501 if (dev
->promiscuity
== 0) {
4504 * If inc causes overflow, untouch promisc and return error.
4507 dev
->flags
&= ~IFF_PROMISC
;
4509 dev
->promiscuity
-= inc
;
4510 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4515 if (dev
->flags
!= old_flags
) {
4516 pr_info("device %s %s promiscuous mode\n",
4518 dev
->flags
& IFF_PROMISC
? "entered" : "left");
4519 if (audit_enabled
) {
4520 current_uid_gid(&uid
, &gid
);
4521 audit_log(current
->audit_context
, GFP_ATOMIC
,
4522 AUDIT_ANOM_PROMISCUOUS
,
4523 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4524 dev
->name
, (dev
->flags
& IFF_PROMISC
),
4525 (old_flags
& IFF_PROMISC
),
4526 audit_get_loginuid(current
),
4528 audit_get_sessionid(current
));
4531 dev_change_rx_flags(dev
, IFF_PROMISC
);
4537 * dev_set_promiscuity - update promiscuity count on a device
4541 * Add or remove promiscuity from a device. While the count in the device
4542 * remains above zero the interface remains promiscuous. Once it hits zero
4543 * the device reverts back to normal filtering operation. A negative inc
4544 * value is used to drop promiscuity on the device.
4545 * Return 0 if successful or a negative errno code on error.
4547 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
4549 unsigned int old_flags
= dev
->flags
;
4552 err
= __dev_set_promiscuity(dev
, inc
);
4555 if (dev
->flags
!= old_flags
)
4556 dev_set_rx_mode(dev
);
4559 EXPORT_SYMBOL(dev_set_promiscuity
);
4562 * dev_set_allmulti - update allmulti count on a device
4566 * Add or remove reception of all multicast frames to a device. While the
4567 * count in the device remains above zero the interface remains listening
4568 * to all interfaces. Once it hits zero the device reverts back to normal
4569 * filtering operation. A negative @inc value is used to drop the counter
4570 * when releasing a resource needing all multicasts.
4571 * Return 0 if successful or a negative errno code on error.
4574 int dev_set_allmulti(struct net_device
*dev
, int inc
)
4576 unsigned int old_flags
= dev
->flags
;
4580 dev
->flags
|= IFF_ALLMULTI
;
4581 dev
->allmulti
+= inc
;
4582 if (dev
->allmulti
== 0) {
4585 * If inc causes overflow, untouch allmulti and return error.
4588 dev
->flags
&= ~IFF_ALLMULTI
;
4590 dev
->allmulti
-= inc
;
4591 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4596 if (dev
->flags
^ old_flags
) {
4597 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
4598 dev_set_rx_mode(dev
);
4602 EXPORT_SYMBOL(dev_set_allmulti
);
4605 * Upload unicast and multicast address lists to device and
4606 * configure RX filtering. When the device doesn't support unicast
4607 * filtering it is put in promiscuous mode while unicast addresses
4610 void __dev_set_rx_mode(struct net_device
*dev
)
4612 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4614 /* dev_open will call this function so the list will stay sane. */
4615 if (!(dev
->flags
&IFF_UP
))
4618 if (!netif_device_present(dev
))
4621 if (!(dev
->priv_flags
& IFF_UNICAST_FLT
)) {
4622 /* Unicast addresses changes may only happen under the rtnl,
4623 * therefore calling __dev_set_promiscuity here is safe.
4625 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
4626 __dev_set_promiscuity(dev
, 1);
4627 dev
->uc_promisc
= true;
4628 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
4629 __dev_set_promiscuity(dev
, -1);
4630 dev
->uc_promisc
= false;
4634 if (ops
->ndo_set_rx_mode
)
4635 ops
->ndo_set_rx_mode(dev
);
4638 void dev_set_rx_mode(struct net_device
*dev
)
4640 netif_addr_lock_bh(dev
);
4641 __dev_set_rx_mode(dev
);
4642 netif_addr_unlock_bh(dev
);
4646 * dev_get_flags - get flags reported to userspace
4649 * Get the combination of flag bits exported through APIs to userspace.
4651 unsigned int dev_get_flags(const struct net_device
*dev
)
4655 flags
= (dev
->flags
& ~(IFF_PROMISC
|
4660 (dev
->gflags
& (IFF_PROMISC
|
4663 if (netif_running(dev
)) {
4664 if (netif_oper_up(dev
))
4665 flags
|= IFF_RUNNING
;
4666 if (netif_carrier_ok(dev
))
4667 flags
|= IFF_LOWER_UP
;
4668 if (netif_dormant(dev
))
4669 flags
|= IFF_DORMANT
;
4674 EXPORT_SYMBOL(dev_get_flags
);
4676 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4678 unsigned int old_flags
= dev
->flags
;
4684 * Set the flags on our device.
4687 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
4688 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
4690 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
4694 * Load in the correct multicast list now the flags have changed.
4697 if ((old_flags
^ flags
) & IFF_MULTICAST
)
4698 dev_change_rx_flags(dev
, IFF_MULTICAST
);
4700 dev_set_rx_mode(dev
);
4703 * Have we downed the interface. We handle IFF_UP ourselves
4704 * according to user attempts to set it, rather than blindly
4709 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
4710 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
4713 dev_set_rx_mode(dev
);
4716 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
4717 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
4719 dev
->gflags
^= IFF_PROMISC
;
4720 dev_set_promiscuity(dev
, inc
);
4723 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4724 is important. Some (broken) drivers set IFF_PROMISC, when
4725 IFF_ALLMULTI is requested not asking us and not reporting.
4727 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
4728 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
4730 dev
->gflags
^= IFF_ALLMULTI
;
4731 dev_set_allmulti(dev
, inc
);
4737 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
)
4739 unsigned int changes
= dev
->flags
^ old_flags
;
4741 if (changes
& IFF_UP
) {
4742 if (dev
->flags
& IFF_UP
)
4743 call_netdevice_notifiers(NETDEV_UP
, dev
);
4745 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
4748 if (dev
->flags
& IFF_UP
&&
4749 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
)))
4750 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
4754 * dev_change_flags - change device settings
4756 * @flags: device state flags
4758 * Change settings on device based state flags. The flags are
4759 * in the userspace exported format.
4761 int dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4764 unsigned int changes
, old_flags
= dev
->flags
;
4766 ret
= __dev_change_flags(dev
, flags
);
4770 changes
= old_flags
^ dev
->flags
;
4772 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
4774 __dev_notify_flags(dev
, old_flags
);
4777 EXPORT_SYMBOL(dev_change_flags
);
4780 * dev_set_mtu - Change maximum transfer unit
4782 * @new_mtu: new transfer unit
4784 * Change the maximum transfer size of the network device.
4786 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
4788 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4791 if (new_mtu
== dev
->mtu
)
4794 /* MTU must be positive. */
4798 if (!netif_device_present(dev
))
4802 if (ops
->ndo_change_mtu
)
4803 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
4807 if (!err
&& dev
->flags
& IFF_UP
)
4808 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
4811 EXPORT_SYMBOL(dev_set_mtu
);
4814 * dev_set_group - Change group this device belongs to
4816 * @new_group: group this device should belong to
4818 void dev_set_group(struct net_device
*dev
, int new_group
)
4820 dev
->group
= new_group
;
4822 EXPORT_SYMBOL(dev_set_group
);
4825 * dev_set_mac_address - Change Media Access Control Address
4829 * Change the hardware (MAC) address of the device
4831 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
4833 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4836 if (!ops
->ndo_set_mac_address
)
4838 if (sa
->sa_family
!= dev
->type
)
4840 if (!netif_device_present(dev
))
4842 err
= ops
->ndo_set_mac_address(dev
, sa
);
4844 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4847 EXPORT_SYMBOL(dev_set_mac_address
);
4850 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4852 static int dev_ifsioc_locked(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4855 struct net_device
*dev
= dev_get_by_name_rcu(net
, ifr
->ifr_name
);
4861 case SIOCGIFFLAGS
: /* Get interface flags */
4862 ifr
->ifr_flags
= (short) dev_get_flags(dev
);
4865 case SIOCGIFMETRIC
: /* Get the metric on the interface
4866 (currently unused) */
4867 ifr
->ifr_metric
= 0;
4870 case SIOCGIFMTU
: /* Get the MTU of a device */
4871 ifr
->ifr_mtu
= dev
->mtu
;
4876 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
4878 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
4879 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4880 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
4888 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
4889 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
4890 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
4891 ifr
->ifr_map
.irq
= dev
->irq
;
4892 ifr
->ifr_map
.dma
= dev
->dma
;
4893 ifr
->ifr_map
.port
= dev
->if_port
;
4897 ifr
->ifr_ifindex
= dev
->ifindex
;
4901 ifr
->ifr_qlen
= dev
->tx_queue_len
;
4905 /* dev_ioctl() should ensure this case
4917 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4919 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4922 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
4923 const struct net_device_ops
*ops
;
4928 ops
= dev
->netdev_ops
;
4931 case SIOCSIFFLAGS
: /* Set interface flags */
4932 return dev_change_flags(dev
, ifr
->ifr_flags
);
4934 case SIOCSIFMETRIC
: /* Set the metric on the interface
4935 (currently unused) */
4938 case SIOCSIFMTU
: /* Set the MTU of a device */
4939 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
4942 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
4944 case SIOCSIFHWBROADCAST
:
4945 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
4947 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
4948 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4949 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4953 if (ops
->ndo_set_config
) {
4954 if (!netif_device_present(dev
))
4956 return ops
->ndo_set_config(dev
, &ifr
->ifr_map
);
4961 if (!ops
->ndo_set_rx_mode
||
4962 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4964 if (!netif_device_present(dev
))
4966 return dev_mc_add_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4969 if (!ops
->ndo_set_rx_mode
||
4970 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4972 if (!netif_device_present(dev
))
4974 return dev_mc_del_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4977 if (ifr
->ifr_qlen
< 0)
4979 dev
->tx_queue_len
= ifr
->ifr_qlen
;
4983 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
4984 return dev_change_name(dev
, ifr
->ifr_newname
);
4987 err
= net_hwtstamp_validate(ifr
);
4993 * Unknown or private ioctl
4996 if ((cmd
>= SIOCDEVPRIVATE
&&
4997 cmd
<= SIOCDEVPRIVATE
+ 15) ||
4998 cmd
== SIOCBONDENSLAVE
||
4999 cmd
== SIOCBONDRELEASE
||
5000 cmd
== SIOCBONDSETHWADDR
||
5001 cmd
== SIOCBONDSLAVEINFOQUERY
||
5002 cmd
== SIOCBONDINFOQUERY
||
5003 cmd
== SIOCBONDCHANGEACTIVE
||
5004 cmd
== SIOCGMIIPHY
||
5005 cmd
== SIOCGMIIREG
||
5006 cmd
== SIOCSMIIREG
||
5007 cmd
== SIOCBRADDIF
||
5008 cmd
== SIOCBRDELIF
||
5009 cmd
== SIOCSHWTSTAMP
||
5010 cmd
== SIOCWANDEV
) {
5012 if (ops
->ndo_do_ioctl
) {
5013 if (netif_device_present(dev
))
5014 err
= ops
->ndo_do_ioctl(dev
, ifr
, cmd
);
5026 * This function handles all "interface"-type I/O control requests. The actual
5027 * 'doing' part of this is dev_ifsioc above.
5031 * dev_ioctl - network device ioctl
5032 * @net: the applicable net namespace
5033 * @cmd: command to issue
5034 * @arg: pointer to a struct ifreq in user space
5036 * Issue ioctl functions to devices. This is normally called by the
5037 * user space syscall interfaces but can sometimes be useful for
5038 * other purposes. The return value is the return from the syscall if
5039 * positive or a negative errno code on error.
5042 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
5048 /* One special case: SIOCGIFCONF takes ifconf argument
5049 and requires shared lock, because it sleeps writing
5053 if (cmd
== SIOCGIFCONF
) {
5055 ret
= dev_ifconf(net
, (char __user
*) arg
);
5059 if (cmd
== SIOCGIFNAME
)
5060 return dev_ifname(net
, (struct ifreq __user
*)arg
);
5062 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
5065 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
5067 colon
= strchr(ifr
.ifr_name
, ':');
5072 * See which interface the caller is talking about.
5077 * These ioctl calls:
5078 * - can be done by all.
5079 * - atomic and do not require locking.
5090 dev_load(net
, ifr
.ifr_name
);
5092 ret
= dev_ifsioc_locked(net
, &ifr
, cmd
);
5097 if (copy_to_user(arg
, &ifr
,
5098 sizeof(struct ifreq
)))
5104 dev_load(net
, ifr
.ifr_name
);
5106 ret
= dev_ethtool(net
, &ifr
);
5111 if (copy_to_user(arg
, &ifr
,
5112 sizeof(struct ifreq
)))
5118 * These ioctl calls:
5119 * - require superuser power.
5120 * - require strict serialization.
5126 if (!capable(CAP_NET_ADMIN
))
5128 dev_load(net
, ifr
.ifr_name
);
5130 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5135 if (copy_to_user(arg
, &ifr
,
5136 sizeof(struct ifreq
)))
5142 * These ioctl calls:
5143 * - require superuser power.
5144 * - require strict serialization.
5145 * - do not return a value
5155 case SIOCSIFHWBROADCAST
:
5158 case SIOCBONDENSLAVE
:
5159 case SIOCBONDRELEASE
:
5160 case SIOCBONDSETHWADDR
:
5161 case SIOCBONDCHANGEACTIVE
:
5165 if (!capable(CAP_NET_ADMIN
))
5168 case SIOCBONDSLAVEINFOQUERY
:
5169 case SIOCBONDINFOQUERY
:
5170 dev_load(net
, ifr
.ifr_name
);
5172 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5177 /* Get the per device memory space. We can add this but
5178 * currently do not support it */
5180 /* Set the per device memory buffer space.
5181 * Not applicable in our case */
5186 * Unknown or private ioctl.
5189 if (cmd
== SIOCWANDEV
||
5190 (cmd
>= SIOCDEVPRIVATE
&&
5191 cmd
<= SIOCDEVPRIVATE
+ 15)) {
5192 dev_load(net
, ifr
.ifr_name
);
5194 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5196 if (!ret
&& copy_to_user(arg
, &ifr
,
5197 sizeof(struct ifreq
)))
5201 /* Take care of Wireless Extensions */
5202 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
5203 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
5210 * dev_new_index - allocate an ifindex
5211 * @net: the applicable net namespace
5213 * Returns a suitable unique value for a new device interface
5214 * number. The caller must hold the rtnl semaphore or the
5215 * dev_base_lock to be sure it remains unique.
5217 static int dev_new_index(struct net
*net
)
5223 if (!__dev_get_by_index(net
, ifindex
))
5228 /* Delayed registration/unregisteration */
5229 static LIST_HEAD(net_todo_list
);
5231 static void net_set_todo(struct net_device
*dev
)
5233 list_add_tail(&dev
->todo_list
, &net_todo_list
);
5236 static void rollback_registered_many(struct list_head
*head
)
5238 struct net_device
*dev
, *tmp
;
5240 BUG_ON(dev_boot_phase
);
5243 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
5244 /* Some devices call without registering
5245 * for initialization unwind. Remove those
5246 * devices and proceed with the remaining.
5248 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5249 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5253 list_del(&dev
->unreg_list
);
5256 dev
->dismantle
= true;
5257 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
5260 /* If device is running, close it first. */
5261 dev_close_many(head
);
5263 list_for_each_entry(dev
, head
, unreg_list
) {
5264 /* And unlink it from device chain. */
5265 unlist_netdevice(dev
);
5267 dev
->reg_state
= NETREG_UNREGISTERING
;
5272 list_for_each_entry(dev
, head
, unreg_list
) {
5273 /* Shutdown queueing discipline. */
5277 /* Notify protocols, that we are about to destroy
5278 this device. They should clean all the things.
5280 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5282 if (!dev
->rtnl_link_ops
||
5283 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5284 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
5287 * Flush the unicast and multicast chains
5292 if (dev
->netdev_ops
->ndo_uninit
)
5293 dev
->netdev_ops
->ndo_uninit(dev
);
5295 /* Notifier chain MUST detach us from master device. */
5296 WARN_ON(dev
->master
);
5298 /* Remove entries from kobject tree */
5299 netdev_unregister_kobject(dev
);
5302 /* Process any work delayed until the end of the batch */
5303 dev
= list_first_entry(head
, struct net_device
, unreg_list
);
5304 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
5308 list_for_each_entry(dev
, head
, unreg_list
)
5312 static void rollback_registered(struct net_device
*dev
)
5316 list_add(&dev
->unreg_list
, &single
);
5317 rollback_registered_many(&single
);
5321 static netdev_features_t
netdev_fix_features(struct net_device
*dev
,
5322 netdev_features_t features
)
5324 /* Fix illegal checksum combinations */
5325 if ((features
& NETIF_F_HW_CSUM
) &&
5326 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5327 netdev_warn(dev
, "mixed HW and IP checksum settings.\n");
5328 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5331 /* Fix illegal SG+CSUM combinations. */
5332 if ((features
& NETIF_F_SG
) &&
5333 !(features
& NETIF_F_ALL_CSUM
)) {
5335 "Dropping NETIF_F_SG since no checksum feature.\n");
5336 features
&= ~NETIF_F_SG
;
5339 /* TSO requires that SG is present as well. */
5340 if ((features
& NETIF_F_ALL_TSO
) && !(features
& NETIF_F_SG
)) {
5341 netdev_dbg(dev
, "Dropping TSO features since no SG feature.\n");
5342 features
&= ~NETIF_F_ALL_TSO
;
5345 /* TSO ECN requires that TSO is present as well. */
5346 if ((features
& NETIF_F_ALL_TSO
) == NETIF_F_TSO_ECN
)
5347 features
&= ~NETIF_F_TSO_ECN
;
5349 /* Software GSO depends on SG. */
5350 if ((features
& NETIF_F_GSO
) && !(features
& NETIF_F_SG
)) {
5351 netdev_dbg(dev
, "Dropping NETIF_F_GSO since no SG feature.\n");
5352 features
&= ~NETIF_F_GSO
;
5355 /* UFO needs SG and checksumming */
5356 if (features
& NETIF_F_UFO
) {
5357 /* maybe split UFO into V4 and V6? */
5358 if (!((features
& NETIF_F_GEN_CSUM
) ||
5359 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))
5360 == (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5362 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5363 features
&= ~NETIF_F_UFO
;
5366 if (!(features
& NETIF_F_SG
)) {
5368 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5369 features
&= ~NETIF_F_UFO
;
5376 int __netdev_update_features(struct net_device
*dev
)
5378 netdev_features_t features
;
5383 features
= netdev_get_wanted_features(dev
);
5385 if (dev
->netdev_ops
->ndo_fix_features
)
5386 features
= dev
->netdev_ops
->ndo_fix_features(dev
, features
);
5388 /* driver might be less strict about feature dependencies */
5389 features
= netdev_fix_features(dev
, features
);
5391 if (dev
->features
== features
)
5394 netdev_dbg(dev
, "Features changed: %pNF -> %pNF\n",
5395 &dev
->features
, &features
);
5397 if (dev
->netdev_ops
->ndo_set_features
)
5398 err
= dev
->netdev_ops
->ndo_set_features(dev
, features
);
5400 if (unlikely(err
< 0)) {
5402 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5403 err
, &features
, &dev
->features
);
5408 dev
->features
= features
;
5414 * netdev_update_features - recalculate device features
5415 * @dev: the device to check
5417 * Recalculate dev->features set and send notifications if it
5418 * has changed. Should be called after driver or hardware dependent
5419 * conditions might have changed that influence the features.
5421 void netdev_update_features(struct net_device
*dev
)
5423 if (__netdev_update_features(dev
))
5424 netdev_features_change(dev
);
5426 EXPORT_SYMBOL(netdev_update_features
);
5429 * netdev_change_features - recalculate device features
5430 * @dev: the device to check
5432 * Recalculate dev->features set and send notifications even
5433 * if they have not changed. Should be called instead of
5434 * netdev_update_features() if also dev->vlan_features might
5435 * have changed to allow the changes to be propagated to stacked
5438 void netdev_change_features(struct net_device
*dev
)
5440 __netdev_update_features(dev
);
5441 netdev_features_change(dev
);
5443 EXPORT_SYMBOL(netdev_change_features
);
5446 * netif_stacked_transfer_operstate - transfer operstate
5447 * @rootdev: the root or lower level device to transfer state from
5448 * @dev: the device to transfer operstate to
5450 * Transfer operational state from root to device. This is normally
5451 * called when a stacking relationship exists between the root
5452 * device and the device(a leaf device).
5454 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
5455 struct net_device
*dev
)
5457 if (rootdev
->operstate
== IF_OPER_DORMANT
)
5458 netif_dormant_on(dev
);
5460 netif_dormant_off(dev
);
5462 if (netif_carrier_ok(rootdev
)) {
5463 if (!netif_carrier_ok(dev
))
5464 netif_carrier_on(dev
);
5466 if (netif_carrier_ok(dev
))
5467 netif_carrier_off(dev
);
5470 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
5473 static int netif_alloc_rx_queues(struct net_device
*dev
)
5475 unsigned int i
, count
= dev
->num_rx_queues
;
5476 struct netdev_rx_queue
*rx
;
5480 rx
= kcalloc(count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5482 pr_err("netdev: Unable to allocate %u rx queues\n", count
);
5487 for (i
= 0; i
< count
; i
++)
5493 static void netdev_init_one_queue(struct net_device
*dev
,
5494 struct netdev_queue
*queue
, void *_unused
)
5496 /* Initialize queue lock */
5497 spin_lock_init(&queue
->_xmit_lock
);
5498 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
5499 queue
->xmit_lock_owner
= -1;
5500 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
5503 dql_init(&queue
->dql
, HZ
);
5507 static int netif_alloc_netdev_queues(struct net_device
*dev
)
5509 unsigned int count
= dev
->num_tx_queues
;
5510 struct netdev_queue
*tx
;
5514 tx
= kcalloc(count
, sizeof(struct netdev_queue
), GFP_KERNEL
);
5516 pr_err("netdev: Unable to allocate %u tx queues\n", count
);
5521 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5522 spin_lock_init(&dev
->tx_global_lock
);
5528 * register_netdevice - register a network device
5529 * @dev: device to register
5531 * Take a completed network device structure and add it to the kernel
5532 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5533 * chain. 0 is returned on success. A negative errno code is returned
5534 * on a failure to set up the device, or if the name is a duplicate.
5536 * Callers must hold the rtnl semaphore. You may want
5537 * register_netdev() instead of this.
5540 * The locking appears insufficient to guarantee two parallel registers
5541 * will not get the same name.
5544 int register_netdevice(struct net_device
*dev
)
5547 struct net
*net
= dev_net(dev
);
5549 BUG_ON(dev_boot_phase
);
5554 /* When net_device's are persistent, this will be fatal. */
5555 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
5558 spin_lock_init(&dev
->addr_list_lock
);
5559 netdev_set_addr_lockdep_class(dev
);
5563 ret
= dev_get_valid_name(dev
, dev
->name
);
5567 /* Init, if this function is available */
5568 if (dev
->netdev_ops
->ndo_init
) {
5569 ret
= dev
->netdev_ops
->ndo_init(dev
);
5577 dev
->ifindex
= dev_new_index(net
);
5578 if (dev
->iflink
== -1)
5579 dev
->iflink
= dev
->ifindex
;
5581 /* Transfer changeable features to wanted_features and enable
5582 * software offloads (GSO and GRO).
5584 dev
->hw_features
|= NETIF_F_SOFT_FEATURES
;
5585 dev
->features
|= NETIF_F_SOFT_FEATURES
;
5586 dev
->wanted_features
= dev
->features
& dev
->hw_features
;
5588 /* Turn on no cache copy if HW is doing checksum */
5589 if (!(dev
->flags
& IFF_LOOPBACK
)) {
5590 dev
->hw_features
|= NETIF_F_NOCACHE_COPY
;
5591 if (dev
->features
& NETIF_F_ALL_CSUM
) {
5592 dev
->wanted_features
|= NETIF_F_NOCACHE_COPY
;
5593 dev
->features
|= NETIF_F_NOCACHE_COPY
;
5597 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5599 dev
->vlan_features
|= NETIF_F_HIGHDMA
;
5601 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
5602 ret
= notifier_to_errno(ret
);
5606 ret
= netdev_register_kobject(dev
);
5609 dev
->reg_state
= NETREG_REGISTERED
;
5611 __netdev_update_features(dev
);
5614 * Default initial state at registry is that the
5615 * device is present.
5618 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5620 dev_init_scheduler(dev
);
5622 list_netdevice(dev
);
5624 /* Notify protocols, that a new device appeared. */
5625 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5626 ret
= notifier_to_errno(ret
);
5628 rollback_registered(dev
);
5629 dev
->reg_state
= NETREG_UNREGISTERED
;
5632 * Prevent userspace races by waiting until the network
5633 * device is fully setup before sending notifications.
5635 if (!dev
->rtnl_link_ops
||
5636 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5637 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5643 if (dev
->netdev_ops
->ndo_uninit
)
5644 dev
->netdev_ops
->ndo_uninit(dev
);
5647 EXPORT_SYMBOL(register_netdevice
);
5650 * init_dummy_netdev - init a dummy network device for NAPI
5651 * @dev: device to init
5653 * This takes a network device structure and initialize the minimum
5654 * amount of fields so it can be used to schedule NAPI polls without
5655 * registering a full blown interface. This is to be used by drivers
5656 * that need to tie several hardware interfaces to a single NAPI
5657 * poll scheduler due to HW limitations.
5659 int init_dummy_netdev(struct net_device
*dev
)
5661 /* Clear everything. Note we don't initialize spinlocks
5662 * are they aren't supposed to be taken by any of the
5663 * NAPI code and this dummy netdev is supposed to be
5664 * only ever used for NAPI polls
5666 memset(dev
, 0, sizeof(struct net_device
));
5668 /* make sure we BUG if trying to hit standard
5669 * register/unregister code path
5671 dev
->reg_state
= NETREG_DUMMY
;
5673 /* NAPI wants this */
5674 INIT_LIST_HEAD(&dev
->napi_list
);
5676 /* a dummy interface is started by default */
5677 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5678 set_bit(__LINK_STATE_START
, &dev
->state
);
5680 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5681 * because users of this 'device' dont need to change
5687 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
5691 * register_netdev - register a network device
5692 * @dev: device to register
5694 * Take a completed network device structure and add it to the kernel
5695 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5696 * chain. 0 is returned on success. A negative errno code is returned
5697 * on a failure to set up the device, or if the name is a duplicate.
5699 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5700 * and expands the device name if you passed a format string to
5703 int register_netdev(struct net_device
*dev
)
5708 err
= register_netdevice(dev
);
5712 EXPORT_SYMBOL(register_netdev
);
5714 int netdev_refcnt_read(const struct net_device
*dev
)
5718 for_each_possible_cpu(i
)
5719 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
5722 EXPORT_SYMBOL(netdev_refcnt_read
);
5725 * netdev_wait_allrefs - wait until all references are gone.
5727 * This is called when unregistering network devices.
5729 * Any protocol or device that holds a reference should register
5730 * for netdevice notification, and cleanup and put back the
5731 * reference if they receive an UNREGISTER event.
5732 * We can get stuck here if buggy protocols don't correctly
5735 static void netdev_wait_allrefs(struct net_device
*dev
)
5737 unsigned long rebroadcast_time
, warning_time
;
5740 linkwatch_forget_dev(dev
);
5742 rebroadcast_time
= warning_time
= jiffies
;
5743 refcnt
= netdev_refcnt_read(dev
);
5745 while (refcnt
!= 0) {
5746 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
5749 /* Rebroadcast unregister notification */
5750 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5751 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5752 * should have already handle it the first time */
5754 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
5756 /* We must not have linkwatch events
5757 * pending on unregister. If this
5758 * happens, we simply run the queue
5759 * unscheduled, resulting in a noop
5762 linkwatch_run_queue();
5767 rebroadcast_time
= jiffies
;
5772 refcnt
= netdev_refcnt_read(dev
);
5774 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
5775 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5777 warning_time
= jiffies
;
5786 * register_netdevice(x1);
5787 * register_netdevice(x2);
5789 * unregister_netdevice(y1);
5790 * unregister_netdevice(y2);
5796 * We are invoked by rtnl_unlock().
5797 * This allows us to deal with problems:
5798 * 1) We can delete sysfs objects which invoke hotplug
5799 * without deadlocking with linkwatch via keventd.
5800 * 2) Since we run with the RTNL semaphore not held, we can sleep
5801 * safely in order to wait for the netdev refcnt to drop to zero.
5803 * We must not return until all unregister events added during
5804 * the interval the lock was held have been completed.
5806 void netdev_run_todo(void)
5808 struct list_head list
;
5810 /* Snapshot list, allow later requests */
5811 list_replace_init(&net_todo_list
, &list
);
5815 /* Wait for rcu callbacks to finish before attempting to drain
5816 * the device list. This usually avoids a 250ms wait.
5818 if (!list_empty(&list
))
5821 while (!list_empty(&list
)) {
5822 struct net_device
*dev
5823 = list_first_entry(&list
, struct net_device
, todo_list
);
5824 list_del(&dev
->todo_list
);
5826 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
5827 pr_err("network todo '%s' but state %d\n",
5828 dev
->name
, dev
->reg_state
);
5833 dev
->reg_state
= NETREG_UNREGISTERED
;
5835 on_each_cpu(flush_backlog
, dev
, 1);
5837 netdev_wait_allrefs(dev
);
5840 BUG_ON(netdev_refcnt_read(dev
));
5841 WARN_ON(rcu_access_pointer(dev
->ip_ptr
));
5842 WARN_ON(rcu_access_pointer(dev
->ip6_ptr
));
5843 WARN_ON(dev
->dn_ptr
);
5845 if (dev
->destructor
)
5846 dev
->destructor(dev
);
5848 /* Free network device */
5849 kobject_put(&dev
->dev
.kobj
);
5853 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5854 * fields in the same order, with only the type differing.
5856 void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
5857 const struct net_device_stats
*netdev_stats
)
5859 #if BITS_PER_LONG == 64
5860 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
5861 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
5863 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
5864 const unsigned long *src
= (const unsigned long *)netdev_stats
;
5865 u64
*dst
= (u64
*)stats64
;
5867 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
5868 sizeof(*stats64
) / sizeof(u64
));
5869 for (i
= 0; i
< n
; i
++)
5873 EXPORT_SYMBOL(netdev_stats_to_stats64
);
5876 * dev_get_stats - get network device statistics
5877 * @dev: device to get statistics from
5878 * @storage: place to store stats
5880 * Get network statistics from device. Return @storage.
5881 * The device driver may provide its own method by setting
5882 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5883 * otherwise the internal statistics structure is used.
5885 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
5886 struct rtnl_link_stats64
*storage
)
5888 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5890 if (ops
->ndo_get_stats64
) {
5891 memset(storage
, 0, sizeof(*storage
));
5892 ops
->ndo_get_stats64(dev
, storage
);
5893 } else if (ops
->ndo_get_stats
) {
5894 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
5896 netdev_stats_to_stats64(storage
, &dev
->stats
);
5898 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
5901 EXPORT_SYMBOL(dev_get_stats
);
5903 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
5905 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
5907 #ifdef CONFIG_NET_CLS_ACT
5910 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
5913 netdev_init_one_queue(dev
, queue
, NULL
);
5914 queue
->qdisc
= &noop_qdisc
;
5915 queue
->qdisc_sleeping
= &noop_qdisc
;
5916 rcu_assign_pointer(dev
->ingress_queue
, queue
);
5922 * alloc_netdev_mqs - allocate network device
5923 * @sizeof_priv: size of private data to allocate space for
5924 * @name: device name format string
5925 * @setup: callback to initialize device
5926 * @txqs: the number of TX subqueues to allocate
5927 * @rxqs: the number of RX subqueues to allocate
5929 * Allocates a struct net_device with private data area for driver use
5930 * and performs basic initialization. Also allocates subquue structs
5931 * for each queue on the device.
5933 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
5934 void (*setup
)(struct net_device
*),
5935 unsigned int txqs
, unsigned int rxqs
)
5937 struct net_device
*dev
;
5939 struct net_device
*p
;
5941 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
5944 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
5950 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
5955 alloc_size
= sizeof(struct net_device
);
5957 /* ensure 32-byte alignment of private area */
5958 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
5959 alloc_size
+= sizeof_priv
;
5961 /* ensure 32-byte alignment of whole construct */
5962 alloc_size
+= NETDEV_ALIGN
- 1;
5964 p
= kzalloc(alloc_size
, GFP_KERNEL
);
5966 pr_err("alloc_netdev: Unable to allocate device\n");
5970 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
5971 dev
->padded
= (char *)dev
- (char *)p
;
5973 dev
->pcpu_refcnt
= alloc_percpu(int);
5974 if (!dev
->pcpu_refcnt
)
5977 if (dev_addr_init(dev
))
5983 dev_net_set(dev
, &init_net
);
5985 dev
->gso_max_size
= GSO_MAX_SIZE
;
5987 INIT_LIST_HEAD(&dev
->napi_list
);
5988 INIT_LIST_HEAD(&dev
->unreg_list
);
5989 INIT_LIST_HEAD(&dev
->link_watch_list
);
5990 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
5993 dev
->num_tx_queues
= txqs
;
5994 dev
->real_num_tx_queues
= txqs
;
5995 if (netif_alloc_netdev_queues(dev
))
5999 dev
->num_rx_queues
= rxqs
;
6000 dev
->real_num_rx_queues
= rxqs
;
6001 if (netif_alloc_rx_queues(dev
))
6005 strcpy(dev
->name
, name
);
6006 dev
->group
= INIT_NETDEV_GROUP
;
6014 free_percpu(dev
->pcpu_refcnt
);
6024 EXPORT_SYMBOL(alloc_netdev_mqs
);
6027 * free_netdev - free network device
6030 * This function does the last stage of destroying an allocated device
6031 * interface. The reference to the device object is released.
6032 * If this is the last reference then it will be freed.
6034 void free_netdev(struct net_device
*dev
)
6036 struct napi_struct
*p
, *n
;
6038 release_net(dev_net(dev
));
6045 kfree(rcu_dereference_protected(dev
->ingress_queue
, 1));
6047 /* Flush device addresses */
6048 dev_addr_flush(dev
);
6050 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
6053 free_percpu(dev
->pcpu_refcnt
);
6054 dev
->pcpu_refcnt
= NULL
;
6056 /* Compatibility with error handling in drivers */
6057 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
6058 kfree((char *)dev
- dev
->padded
);
6062 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
6063 dev
->reg_state
= NETREG_RELEASED
;
6065 /* will free via device release */
6066 put_device(&dev
->dev
);
6068 EXPORT_SYMBOL(free_netdev
);
6071 * synchronize_net - Synchronize with packet receive processing
6073 * Wait for packets currently being received to be done.
6074 * Does not block later packets from starting.
6076 void synchronize_net(void)
6079 if (rtnl_is_locked())
6080 synchronize_rcu_expedited();
6084 EXPORT_SYMBOL(synchronize_net
);
6087 * unregister_netdevice_queue - remove device from the kernel
6091 * This function shuts down a device interface and removes it
6092 * from the kernel tables.
6093 * If head not NULL, device is queued to be unregistered later.
6095 * Callers must hold the rtnl semaphore. You may want
6096 * unregister_netdev() instead of this.
6099 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
6104 list_move_tail(&dev
->unreg_list
, head
);
6106 rollback_registered(dev
);
6107 /* Finish processing unregister after unlock */
6111 EXPORT_SYMBOL(unregister_netdevice_queue
);
6114 * unregister_netdevice_many - unregister many devices
6115 * @head: list of devices
6117 void unregister_netdevice_many(struct list_head
*head
)
6119 struct net_device
*dev
;
6121 if (!list_empty(head
)) {
6122 rollback_registered_many(head
);
6123 list_for_each_entry(dev
, head
, unreg_list
)
6127 EXPORT_SYMBOL(unregister_netdevice_many
);
6130 * unregister_netdev - remove device from the kernel
6133 * This function shuts down a device interface and removes it
6134 * from the kernel tables.
6136 * This is just a wrapper for unregister_netdevice that takes
6137 * the rtnl semaphore. In general you want to use this and not
6138 * unregister_netdevice.
6140 void unregister_netdev(struct net_device
*dev
)
6143 unregister_netdevice(dev
);
6146 EXPORT_SYMBOL(unregister_netdev
);
6149 * dev_change_net_namespace - move device to different nethost namespace
6151 * @net: network namespace
6152 * @pat: If not NULL name pattern to try if the current device name
6153 * is already taken in the destination network namespace.
6155 * This function shuts down a device interface and moves it
6156 * to a new network namespace. On success 0 is returned, on
6157 * a failure a netagive errno code is returned.
6159 * Callers must hold the rtnl semaphore.
6162 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
6168 /* Don't allow namespace local devices to be moved. */
6170 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6173 /* Ensure the device has been registrered */
6175 if (dev
->reg_state
!= NETREG_REGISTERED
)
6178 /* Get out if there is nothing todo */
6180 if (net_eq(dev_net(dev
), net
))
6183 /* Pick the destination device name, and ensure
6184 * we can use it in the destination network namespace.
6187 if (__dev_get_by_name(net
, dev
->name
)) {
6188 /* We get here if we can't use the current device name */
6191 if (dev_get_valid_name(dev
, pat
) < 0)
6196 * And now a mini version of register_netdevice unregister_netdevice.
6199 /* If device is running close it first. */
6202 /* And unlink it from device chain */
6204 unlist_netdevice(dev
);
6208 /* Shutdown queueing discipline. */
6211 /* Notify protocols, that we are about to destroy
6212 this device. They should clean all the things.
6214 Note that dev->reg_state stays at NETREG_REGISTERED.
6215 This is wanted because this way 8021q and macvlan know
6216 the device is just moving and can keep their slaves up.
6218 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6219 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
6220 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
6223 * Flush the unicast and multicast chains
6228 /* Actually switch the network namespace */
6229 dev_net_set(dev
, net
);
6231 /* If there is an ifindex conflict assign a new one */
6232 if (__dev_get_by_index(net
, dev
->ifindex
)) {
6233 int iflink
= (dev
->iflink
== dev
->ifindex
);
6234 dev
->ifindex
= dev_new_index(net
);
6236 dev
->iflink
= dev
->ifindex
;
6239 /* Fixup kobjects */
6240 err
= device_rename(&dev
->dev
, dev
->name
);
6243 /* Add the device back in the hashes */
6244 list_netdevice(dev
);
6246 /* Notify protocols, that a new device appeared. */
6247 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
6250 * Prevent userspace races by waiting until the network
6251 * device is fully setup before sending notifications.
6253 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
6260 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
6262 static int dev_cpu_callback(struct notifier_block
*nfb
,
6263 unsigned long action
,
6266 struct sk_buff
**list_skb
;
6267 struct sk_buff
*skb
;
6268 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
6269 struct softnet_data
*sd
, *oldsd
;
6271 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
6274 local_irq_disable();
6275 cpu
= smp_processor_id();
6276 sd
= &per_cpu(softnet_data
, cpu
);
6277 oldsd
= &per_cpu(softnet_data
, oldcpu
);
6279 /* Find end of our completion_queue. */
6280 list_skb
= &sd
->completion_queue
;
6282 list_skb
= &(*list_skb
)->next
;
6283 /* Append completion queue from offline CPU. */
6284 *list_skb
= oldsd
->completion_queue
;
6285 oldsd
->completion_queue
= NULL
;
6287 /* Append output queue from offline CPU. */
6288 if (oldsd
->output_queue
) {
6289 *sd
->output_queue_tailp
= oldsd
->output_queue
;
6290 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
6291 oldsd
->output_queue
= NULL
;
6292 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
6294 /* Append NAPI poll list from offline CPU. */
6295 if (!list_empty(&oldsd
->poll_list
)) {
6296 list_splice_init(&oldsd
->poll_list
, &sd
->poll_list
);
6297 raise_softirq_irqoff(NET_RX_SOFTIRQ
);
6300 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
6303 /* Process offline CPU's input_pkt_queue */
6304 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
6306 input_queue_head_incr(oldsd
);
6308 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
6310 input_queue_head_incr(oldsd
);
6318 * netdev_increment_features - increment feature set by one
6319 * @all: current feature set
6320 * @one: new feature set
6321 * @mask: mask feature set
6323 * Computes a new feature set after adding a device with feature set
6324 * @one to the master device with current feature set @all. Will not
6325 * enable anything that is off in @mask. Returns the new feature set.
6327 netdev_features_t
netdev_increment_features(netdev_features_t all
,
6328 netdev_features_t one
, netdev_features_t mask
)
6330 if (mask
& NETIF_F_GEN_CSUM
)
6331 mask
|= NETIF_F_ALL_CSUM
;
6332 mask
|= NETIF_F_VLAN_CHALLENGED
;
6334 all
|= one
& (NETIF_F_ONE_FOR_ALL
|NETIF_F_ALL_CSUM
) & mask
;
6335 all
&= one
| ~NETIF_F_ALL_FOR_ALL
;
6337 /* If one device supports hw checksumming, set for all. */
6338 if (all
& NETIF_F_GEN_CSUM
)
6339 all
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_GEN_CSUM
);
6343 EXPORT_SYMBOL(netdev_increment_features
);
6345 static struct hlist_head
*netdev_create_hash(void)
6348 struct hlist_head
*hash
;
6350 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
6352 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
6353 INIT_HLIST_HEAD(&hash
[i
]);
6358 /* Initialize per network namespace state */
6359 static int __net_init
netdev_init(struct net
*net
)
6361 if (net
!= &init_net
)
6362 INIT_LIST_HEAD(&net
->dev_base_head
);
6364 net
->dev_name_head
= netdev_create_hash();
6365 if (net
->dev_name_head
== NULL
)
6368 net
->dev_index_head
= netdev_create_hash();
6369 if (net
->dev_index_head
== NULL
)
6375 kfree(net
->dev_name_head
);
6381 * netdev_drivername - network driver for the device
6382 * @dev: network device
6384 * Determine network driver for device.
6386 const char *netdev_drivername(const struct net_device
*dev
)
6388 const struct device_driver
*driver
;
6389 const struct device
*parent
;
6390 const char *empty
= "";
6392 parent
= dev
->dev
.parent
;
6396 driver
= parent
->driver
;
6397 if (driver
&& driver
->name
)
6398 return driver
->name
;
6402 int __netdev_printk(const char *level
, const struct net_device
*dev
,
6403 struct va_format
*vaf
)
6407 if (dev
&& dev
->dev
.parent
)
6408 r
= dev_printk(level
, dev
->dev
.parent
, "%s: %pV",
6409 netdev_name(dev
), vaf
);
6411 r
= printk("%s%s: %pV", level
, netdev_name(dev
), vaf
);
6413 r
= printk("%s(NULL net_device): %pV", level
, vaf
);
6417 EXPORT_SYMBOL(__netdev_printk
);
6419 int netdev_printk(const char *level
, const struct net_device
*dev
,
6420 const char *format
, ...)
6422 struct va_format vaf
;
6426 va_start(args
, format
);
6431 r
= __netdev_printk(level
, dev
, &vaf
);
6436 EXPORT_SYMBOL(netdev_printk
);
6438 #define define_netdev_printk_level(func, level) \
6439 int func(const struct net_device *dev, const char *fmt, ...) \
6442 struct va_format vaf; \
6445 va_start(args, fmt); \
6450 r = __netdev_printk(level, dev, &vaf); \
6455 EXPORT_SYMBOL(func);
6457 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
6458 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
6459 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
6460 define_netdev_printk_level(netdev_err
, KERN_ERR
);
6461 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
6462 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
6463 define_netdev_printk_level(netdev_info
, KERN_INFO
);
6465 static void __net_exit
netdev_exit(struct net
*net
)
6467 kfree(net
->dev_name_head
);
6468 kfree(net
->dev_index_head
);
6471 static struct pernet_operations __net_initdata netdev_net_ops
= {
6472 .init
= netdev_init
,
6473 .exit
= netdev_exit
,
6476 static void __net_exit
default_device_exit(struct net
*net
)
6478 struct net_device
*dev
, *aux
;
6480 * Push all migratable network devices back to the
6481 * initial network namespace
6484 for_each_netdev_safe(net
, dev
, aux
) {
6486 char fb_name
[IFNAMSIZ
];
6488 /* Ignore unmoveable devices (i.e. loopback) */
6489 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6492 /* Leave virtual devices for the generic cleanup */
6493 if (dev
->rtnl_link_ops
)
6496 /* Push remaining network devices to init_net */
6497 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
6498 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
6500 pr_emerg("%s: failed to move %s to init_net: %d\n",
6501 __func__
, dev
->name
, err
);
6508 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
6510 /* At exit all network devices most be removed from a network
6511 * namespace. Do this in the reverse order of registration.
6512 * Do this across as many network namespaces as possible to
6513 * improve batching efficiency.
6515 struct net_device
*dev
;
6517 LIST_HEAD(dev_kill_list
);
6520 list_for_each_entry(net
, net_list
, exit_list
) {
6521 for_each_netdev_reverse(net
, dev
) {
6522 if (dev
->rtnl_link_ops
)
6523 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
6525 unregister_netdevice_queue(dev
, &dev_kill_list
);
6528 unregister_netdevice_many(&dev_kill_list
);
6529 list_del(&dev_kill_list
);
6533 static struct pernet_operations __net_initdata default_device_ops
= {
6534 .exit
= default_device_exit
,
6535 .exit_batch
= default_device_exit_batch
,
6539 * Initialize the DEV module. At boot time this walks the device list and
6540 * unhooks any devices that fail to initialise (normally hardware not
6541 * present) and leaves us with a valid list of present and active devices.
6546 * This is called single threaded during boot, so no need
6547 * to take the rtnl semaphore.
6549 static int __init
net_dev_init(void)
6551 int i
, rc
= -ENOMEM
;
6553 BUG_ON(!dev_boot_phase
);
6555 if (dev_proc_init())
6558 if (netdev_kobject_init())
6561 INIT_LIST_HEAD(&ptype_all
);
6562 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
6563 INIT_LIST_HEAD(&ptype_base
[i
]);
6565 if (register_pernet_subsys(&netdev_net_ops
))
6569 * Initialise the packet receive queues.
6572 for_each_possible_cpu(i
) {
6573 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
6575 memset(sd
, 0, sizeof(*sd
));
6576 skb_queue_head_init(&sd
->input_pkt_queue
);
6577 skb_queue_head_init(&sd
->process_queue
);
6578 sd
->completion_queue
= NULL
;
6579 INIT_LIST_HEAD(&sd
->poll_list
);
6580 sd
->output_queue
= NULL
;
6581 sd
->output_queue_tailp
= &sd
->output_queue
;
6583 sd
->csd
.func
= rps_trigger_softirq
;
6589 sd
->backlog
.poll
= process_backlog
;
6590 sd
->backlog
.weight
= weight_p
;
6591 sd
->backlog
.gro_list
= NULL
;
6592 sd
->backlog
.gro_count
= 0;
6597 /* The loopback device is special if any other network devices
6598 * is present in a network namespace the loopback device must
6599 * be present. Since we now dynamically allocate and free the
6600 * loopback device ensure this invariant is maintained by
6601 * keeping the loopback device as the first device on the
6602 * list of network devices. Ensuring the loopback devices
6603 * is the first device that appears and the last network device
6606 if (register_pernet_device(&loopback_net_ops
))
6609 if (register_pernet_device(&default_device_ops
))
6612 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
6613 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
6615 hotcpu_notifier(dev_cpu_callback
, 0);
6623 subsys_initcall(net_dev_init
);
6625 static int __init
initialize_hashrnd(void)
6627 get_random_bytes(&hashrnd
, sizeof(hashrnd
));
6631 late_initcall_sync(initialize_hashrnd
);