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
)
1062 if (len
>= IFALIASZ
)
1067 kfree(dev
->ifalias
);
1068 dev
->ifalias
= NULL
;
1073 new_ifalias
= krealloc(dev
->ifalias
, len
+ 1, GFP_KERNEL
);
1076 dev
->ifalias
= new_ifalias
;
1078 strlcpy(dev
->ifalias
, alias
, len
+1);
1084 * netdev_features_change - device changes features
1085 * @dev: device to cause notification
1087 * Called to indicate a device has changed features.
1089 void netdev_features_change(struct net_device
*dev
)
1091 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1093 EXPORT_SYMBOL(netdev_features_change
);
1096 * netdev_state_change - device changes state
1097 * @dev: device to cause notification
1099 * Called to indicate a device has changed state. This function calls
1100 * the notifier chains for netdev_chain and sends a NEWLINK message
1101 * to the routing socket.
1103 void netdev_state_change(struct net_device
*dev
)
1105 if (dev
->flags
& IFF_UP
) {
1106 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
1107 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
1110 EXPORT_SYMBOL(netdev_state_change
);
1113 * netdev_notify_peers - notify network peers about existence of @dev
1114 * @dev: network device
1116 * Generate traffic such that interested network peers are aware of
1117 * @dev, such as by generating a gratuitous ARP. This may be used when
1118 * a device wants to inform the rest of the network about some sort of
1119 * reconfiguration such as a failover event or virtual machine
1122 void netdev_notify_peers(struct net_device
*dev
)
1125 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
, dev
);
1128 EXPORT_SYMBOL(netdev_notify_peers
);
1131 * dev_load - load a network module
1132 * @net: the applicable net namespace
1133 * @name: name of interface
1135 * If a network interface is not present and the process has suitable
1136 * privileges this function loads the module. If module loading is not
1137 * available in this kernel then it becomes a nop.
1140 void dev_load(struct net
*net
, const char *name
)
1142 struct net_device
*dev
;
1146 dev
= dev_get_by_name_rcu(net
, name
);
1150 if (no_module
&& capable(CAP_NET_ADMIN
))
1151 no_module
= request_module("netdev-%s", name
);
1152 if (no_module
&& capable(CAP_SYS_MODULE
)) {
1153 if (!request_module("%s", name
))
1154 pr_warn("Loading kernel module for a network device with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s instead.\n",
1158 EXPORT_SYMBOL(dev_load
);
1160 static int __dev_open(struct net_device
*dev
)
1162 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1167 if (!netif_device_present(dev
))
1170 ret
= call_netdevice_notifiers(NETDEV_PRE_UP
, dev
);
1171 ret
= notifier_to_errno(ret
);
1175 set_bit(__LINK_STATE_START
, &dev
->state
);
1177 if (ops
->ndo_validate_addr
)
1178 ret
= ops
->ndo_validate_addr(dev
);
1180 if (!ret
&& ops
->ndo_open
)
1181 ret
= ops
->ndo_open(dev
);
1184 clear_bit(__LINK_STATE_START
, &dev
->state
);
1186 dev
->flags
|= IFF_UP
;
1187 net_dmaengine_get();
1188 dev_set_rx_mode(dev
);
1190 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
1197 * dev_open - prepare an interface for use.
1198 * @dev: device to open
1200 * Takes a device from down to up state. The device's private open
1201 * function is invoked and then the multicast lists are loaded. Finally
1202 * the device is moved into the up state and a %NETDEV_UP message is
1203 * sent to the netdev notifier chain.
1205 * Calling this function on an active interface is a nop. On a failure
1206 * a negative errno code is returned.
1208 int dev_open(struct net_device
*dev
)
1212 if (dev
->flags
& IFF_UP
)
1215 ret
= __dev_open(dev
);
1219 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1220 call_netdevice_notifiers(NETDEV_UP
, dev
);
1224 EXPORT_SYMBOL(dev_open
);
1226 static int __dev_close_many(struct list_head
*head
)
1228 struct net_device
*dev
;
1233 list_for_each_entry(dev
, head
, unreg_list
) {
1234 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1236 clear_bit(__LINK_STATE_START
, &dev
->state
);
1238 /* Synchronize to scheduled poll. We cannot touch poll list, it
1239 * can be even on different cpu. So just clear netif_running().
1241 * dev->stop() will invoke napi_disable() on all of it's
1242 * napi_struct instances on this device.
1244 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1247 dev_deactivate_many(head
);
1249 list_for_each_entry(dev
, head
, unreg_list
) {
1250 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1253 * Call the device specific close. This cannot fail.
1254 * Only if device is UP
1256 * We allow it to be called even after a DETACH hot-plug
1262 dev
->flags
&= ~IFF_UP
;
1263 net_dmaengine_put();
1269 static int __dev_close(struct net_device
*dev
)
1274 list_add(&dev
->unreg_list
, &single
);
1275 retval
= __dev_close_many(&single
);
1280 static int dev_close_many(struct list_head
*head
)
1282 struct net_device
*dev
, *tmp
;
1283 LIST_HEAD(tmp_list
);
1285 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
)
1286 if (!(dev
->flags
& IFF_UP
))
1287 list_move(&dev
->unreg_list
, &tmp_list
);
1289 __dev_close_many(head
);
1291 list_for_each_entry(dev
, head
, unreg_list
) {
1292 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1293 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1296 /* rollback_registered_many needs the complete original list */
1297 list_splice(&tmp_list
, head
);
1302 * dev_close - shutdown an interface.
1303 * @dev: device to shutdown
1305 * This function moves an active device into down state. A
1306 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1307 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1310 int dev_close(struct net_device
*dev
)
1312 if (dev
->flags
& IFF_UP
) {
1315 list_add(&dev
->unreg_list
, &single
);
1316 dev_close_many(&single
);
1321 EXPORT_SYMBOL(dev_close
);
1325 * dev_disable_lro - disable Large Receive Offload on a device
1328 * Disable Large Receive Offload (LRO) on a net device. Must be
1329 * called under RTNL. This is needed if received packets may be
1330 * forwarded to another interface.
1332 void dev_disable_lro(struct net_device
*dev
)
1335 * If we're trying to disable lro on a vlan device
1336 * use the underlying physical device instead
1338 if (is_vlan_dev(dev
))
1339 dev
= vlan_dev_real_dev(dev
);
1341 dev
->wanted_features
&= ~NETIF_F_LRO
;
1342 netdev_update_features(dev
);
1344 if (unlikely(dev
->features
& NETIF_F_LRO
))
1345 netdev_WARN(dev
, "failed to disable LRO!\n");
1347 EXPORT_SYMBOL(dev_disable_lro
);
1350 static int dev_boot_phase
= 1;
1353 * register_netdevice_notifier - register a network notifier block
1356 * Register a notifier to be called when network device events occur.
1357 * The notifier passed is linked into the kernel structures and must
1358 * not be reused until it has been unregistered. A negative errno code
1359 * is returned on a failure.
1361 * When registered all registration and up events are replayed
1362 * to the new notifier to allow device to have a race free
1363 * view of the network device list.
1366 int register_netdevice_notifier(struct notifier_block
*nb
)
1368 struct net_device
*dev
;
1369 struct net_device
*last
;
1374 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1380 for_each_netdev(net
, dev
) {
1381 err
= nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1382 err
= notifier_to_errno(err
);
1386 if (!(dev
->flags
& IFF_UP
))
1389 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1400 for_each_netdev(net
, dev
) {
1404 if (dev
->flags
& IFF_UP
) {
1405 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1406 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1408 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1413 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1416 EXPORT_SYMBOL(register_netdevice_notifier
);
1419 * unregister_netdevice_notifier - unregister a network notifier block
1422 * Unregister a notifier previously registered by
1423 * register_netdevice_notifier(). The notifier is unlinked into the
1424 * kernel structures and may then be reused. A negative errno code
1425 * is returned on a failure.
1427 * After unregistering unregister and down device events are synthesized
1428 * for all devices on the device list to the removed notifier to remove
1429 * the need for special case cleanup code.
1432 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1434 struct net_device
*dev
;
1439 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1444 for_each_netdev(net
, dev
) {
1445 if (dev
->flags
& IFF_UP
) {
1446 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1447 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1449 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1456 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1459 * call_netdevice_notifiers - call all network notifier blocks
1460 * @val: value passed unmodified to notifier function
1461 * @dev: net_device pointer passed unmodified to notifier function
1463 * Call all network notifier blocks. Parameters and return value
1464 * are as for raw_notifier_call_chain().
1467 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1469 if (val
!= NETDEV_UNREGISTER_FINAL
)
1471 return raw_notifier_call_chain(&netdev_chain
, val
, dev
);
1473 EXPORT_SYMBOL(call_netdevice_notifiers
);
1475 static struct static_key netstamp_needed __read_mostly
;
1476 #ifdef HAVE_JUMP_LABEL
1477 /* We are not allowed to call static_key_slow_dec() from irq context
1478 * If net_disable_timestamp() is called from irq context, defer the
1479 * static_key_slow_dec() calls.
1481 static atomic_t netstamp_needed_deferred
;
1484 void net_enable_timestamp(void)
1486 #ifdef HAVE_JUMP_LABEL
1487 int deferred
= atomic_xchg(&netstamp_needed_deferred
, 0);
1491 static_key_slow_dec(&netstamp_needed
);
1495 WARN_ON(in_interrupt());
1496 static_key_slow_inc(&netstamp_needed
);
1498 EXPORT_SYMBOL(net_enable_timestamp
);
1500 void net_disable_timestamp(void)
1502 #ifdef HAVE_JUMP_LABEL
1503 if (in_interrupt()) {
1504 atomic_inc(&netstamp_needed_deferred
);
1508 static_key_slow_dec(&netstamp_needed
);
1510 EXPORT_SYMBOL(net_disable_timestamp
);
1512 static inline void net_timestamp_set(struct sk_buff
*skb
)
1514 skb
->tstamp
.tv64
= 0;
1515 if (static_key_false(&netstamp_needed
))
1516 __net_timestamp(skb
);
1519 #define net_timestamp_check(COND, SKB) \
1520 if (static_key_false(&netstamp_needed)) { \
1521 if ((COND) && !(SKB)->tstamp.tv64) \
1522 __net_timestamp(SKB); \
1525 static int net_hwtstamp_validate(struct ifreq *ifr)
1527 struct hwtstamp_config cfg
;
1528 enum hwtstamp_tx_types tx_type
;
1529 enum hwtstamp_rx_filters rx_filter
;
1530 int tx_type_valid
= 0;
1531 int rx_filter_valid
= 0;
1533 if (copy_from_user(&cfg
, ifr
->ifr_data
, sizeof(cfg
)))
1536 if (cfg
.flags
) /* reserved for future extensions */
1539 tx_type
= cfg
.tx_type
;
1540 rx_filter
= cfg
.rx_filter
;
1543 case HWTSTAMP_TX_OFF
:
1544 case HWTSTAMP_TX_ON
:
1545 case HWTSTAMP_TX_ONESTEP_SYNC
:
1550 switch (rx_filter
) {
1551 case HWTSTAMP_FILTER_NONE
:
1552 case HWTSTAMP_FILTER_ALL
:
1553 case HWTSTAMP_FILTER_SOME
:
1554 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT
:
1555 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC
:
1556 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ
:
1557 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT
:
1558 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC
:
1559 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ
:
1560 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT
:
1561 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC
:
1562 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ
:
1563 case HWTSTAMP_FILTER_PTP_V2_EVENT
:
1564 case HWTSTAMP_FILTER_PTP_V2_SYNC
:
1565 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ
:
1566 rx_filter_valid
= 1;
1570 if (!tx_type_valid
|| !rx_filter_valid
)
1576 static inline bool is_skb_forwardable(struct net_device
*dev
,
1577 struct sk_buff
*skb
)
1581 if (!(dev
->flags
& IFF_UP
))
1584 len
= dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
;
1585 if (skb
->len
<= len
)
1588 /* if TSO is enabled, we don't care about the length as the packet
1589 * could be forwarded without being segmented before
1591 if (skb_is_gso(skb
))
1598 * dev_forward_skb - loopback an skb to another netif
1600 * @dev: destination network device
1601 * @skb: buffer to forward
1604 * NET_RX_SUCCESS (no congestion)
1605 * NET_RX_DROP (packet was dropped, but freed)
1607 * dev_forward_skb can be used for injecting an skb from the
1608 * start_xmit function of one device into the receive queue
1609 * of another device.
1611 * The receiving device may be in another namespace, so
1612 * we have to clear all information in the skb that could
1613 * impact namespace isolation.
1615 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1617 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
1618 if (skb_copy_ubufs(skb
, GFP_ATOMIC
)) {
1619 atomic_long_inc(&dev
->rx_dropped
);
1628 if (unlikely(!is_skb_forwardable(dev
, skb
))) {
1629 atomic_long_inc(&dev
->rx_dropped
);
1636 skb
->tstamp
.tv64
= 0;
1637 skb
->pkt_type
= PACKET_HOST
;
1638 skb
->protocol
= eth_type_trans(skb
, dev
);
1642 return netif_rx(skb
);
1644 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1646 static inline int deliver_skb(struct sk_buff
*skb
,
1647 struct packet_type
*pt_prev
,
1648 struct net_device
*orig_dev
)
1650 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
1652 atomic_inc(&skb
->users
);
1653 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1656 static inline bool skb_loop_sk(struct packet_type
*ptype
, struct sk_buff
*skb
)
1658 if (ptype
->af_packet_priv
== NULL
)
1661 if (ptype
->id_match
)
1662 return ptype
->id_match(ptype
, skb
->sk
);
1663 else if ((struct sock
*)ptype
->af_packet_priv
== skb
->sk
)
1670 * Support routine. Sends outgoing frames to any network
1671 * taps currently in use.
1674 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1676 struct packet_type
*ptype
;
1677 struct sk_buff
*skb2
= NULL
;
1678 struct packet_type
*pt_prev
= NULL
;
1681 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1682 /* Never send packets back to the socket
1683 * they originated from - MvS (miquels@drinkel.ow.org)
1685 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1686 (!skb_loop_sk(ptype
, skb
))) {
1688 deliver_skb(skb2
, pt_prev
, skb
->dev
);
1693 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1697 net_timestamp_set(skb2
);
1699 /* skb->nh should be correctly
1700 set by sender, so that the second statement is
1701 just protection against buggy protocols.
1703 skb_reset_mac_header(skb2
);
1705 if (skb_network_header(skb2
) < skb2
->data
||
1706 skb2
->network_header
> skb2
->tail
) {
1707 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1708 ntohs(skb2
->protocol
),
1710 skb_reset_network_header(skb2
);
1713 skb2
->transport_header
= skb2
->network_header
;
1714 skb2
->pkt_type
= PACKET_OUTGOING
;
1719 pt_prev
->func(skb2
, skb
->dev
, pt_prev
, skb
->dev
);
1724 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1725 * @dev: Network device
1726 * @txq: number of queues available
1728 * If real_num_tx_queues is changed the tc mappings may no longer be
1729 * valid. To resolve this verify the tc mapping remains valid and if
1730 * not NULL the mapping. With no priorities mapping to this
1731 * offset/count pair it will no longer be used. In the worst case TC0
1732 * is invalid nothing can be done so disable priority mappings. If is
1733 * expected that drivers will fix this mapping if they can before
1734 * calling netif_set_real_num_tx_queues.
1736 static void netif_setup_tc(struct net_device
*dev
, unsigned int txq
)
1739 struct netdev_tc_txq
*tc
= &dev
->tc_to_txq
[0];
1741 /* If TC0 is invalidated disable TC mapping */
1742 if (tc
->offset
+ tc
->count
> txq
) {
1743 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1748 /* Invalidated prio to tc mappings set to TC0 */
1749 for (i
= 1; i
< TC_BITMASK
+ 1; i
++) {
1750 int q
= netdev_get_prio_tc_map(dev
, i
);
1752 tc
= &dev
->tc_to_txq
[q
];
1753 if (tc
->offset
+ tc
->count
> txq
) {
1754 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1756 netdev_set_prio_tc_map(dev
, i
, 0);
1762 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1763 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1765 int netif_set_real_num_tx_queues(struct net_device
*dev
, unsigned int txq
)
1769 if (txq
< 1 || txq
> dev
->num_tx_queues
)
1772 if (dev
->reg_state
== NETREG_REGISTERED
||
1773 dev
->reg_state
== NETREG_UNREGISTERING
) {
1776 rc
= netdev_queue_update_kobjects(dev
, dev
->real_num_tx_queues
,
1782 netif_setup_tc(dev
, txq
);
1784 if (txq
< dev
->real_num_tx_queues
)
1785 qdisc_reset_all_tx_gt(dev
, txq
);
1788 dev
->real_num_tx_queues
= txq
;
1791 EXPORT_SYMBOL(netif_set_real_num_tx_queues
);
1795 * netif_set_real_num_rx_queues - set actual number of RX queues used
1796 * @dev: Network device
1797 * @rxq: Actual number of RX queues
1799 * This must be called either with the rtnl_lock held or before
1800 * registration of the net device. Returns 0 on success, or a
1801 * negative error code. If called before registration, it always
1804 int netif_set_real_num_rx_queues(struct net_device
*dev
, unsigned int rxq
)
1808 if (rxq
< 1 || rxq
> dev
->num_rx_queues
)
1811 if (dev
->reg_state
== NETREG_REGISTERED
) {
1814 rc
= net_rx_queue_update_kobjects(dev
, dev
->real_num_rx_queues
,
1820 dev
->real_num_rx_queues
= rxq
;
1823 EXPORT_SYMBOL(netif_set_real_num_rx_queues
);
1827 * netif_get_num_default_rss_queues - default number of RSS queues
1829 * This routine should set an upper limit on the number of RSS queues
1830 * used by default by multiqueue devices.
1832 int netif_get_num_default_rss_queues(void)
1834 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES
, num_online_cpus());
1836 EXPORT_SYMBOL(netif_get_num_default_rss_queues
);
1838 static inline void __netif_reschedule(struct Qdisc
*q
)
1840 struct softnet_data
*sd
;
1841 unsigned long flags
;
1843 local_irq_save(flags
);
1844 sd
= &__get_cpu_var(softnet_data
);
1845 q
->next_sched
= NULL
;
1846 *sd
->output_queue_tailp
= q
;
1847 sd
->output_queue_tailp
= &q
->next_sched
;
1848 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1849 local_irq_restore(flags
);
1852 void __netif_schedule(struct Qdisc
*q
)
1854 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
1855 __netif_reschedule(q
);
1857 EXPORT_SYMBOL(__netif_schedule
);
1859 void dev_kfree_skb_irq(struct sk_buff
*skb
)
1861 if (atomic_dec_and_test(&skb
->users
)) {
1862 struct softnet_data
*sd
;
1863 unsigned long flags
;
1865 local_irq_save(flags
);
1866 sd
= &__get_cpu_var(softnet_data
);
1867 skb
->next
= sd
->completion_queue
;
1868 sd
->completion_queue
= skb
;
1869 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1870 local_irq_restore(flags
);
1873 EXPORT_SYMBOL(dev_kfree_skb_irq
);
1875 void dev_kfree_skb_any(struct sk_buff
*skb
)
1877 if (in_irq() || irqs_disabled())
1878 dev_kfree_skb_irq(skb
);
1882 EXPORT_SYMBOL(dev_kfree_skb_any
);
1886 * netif_device_detach - mark device as removed
1887 * @dev: network device
1889 * Mark device as removed from system and therefore no longer available.
1891 void netif_device_detach(struct net_device
*dev
)
1893 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1894 netif_running(dev
)) {
1895 netif_tx_stop_all_queues(dev
);
1898 EXPORT_SYMBOL(netif_device_detach
);
1901 * netif_device_attach - mark device as attached
1902 * @dev: network device
1904 * Mark device as attached from system and restart if needed.
1906 void netif_device_attach(struct net_device
*dev
)
1908 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1909 netif_running(dev
)) {
1910 netif_tx_wake_all_queues(dev
);
1911 __netdev_watchdog_up(dev
);
1914 EXPORT_SYMBOL(netif_device_attach
);
1916 static void skb_warn_bad_offload(const struct sk_buff
*skb
)
1918 static const netdev_features_t null_features
= 0;
1919 struct net_device
*dev
= skb
->dev
;
1920 const char *driver
= "";
1922 if (dev
&& dev
->dev
.parent
)
1923 driver
= dev_driver_string(dev
->dev
.parent
);
1925 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
1926 "gso_type=%d ip_summed=%d\n",
1927 driver
, dev
? &dev
->features
: &null_features
,
1928 skb
->sk
? &skb
->sk
->sk_route_caps
: &null_features
,
1929 skb
->len
, skb
->data_len
, skb_shinfo(skb
)->gso_size
,
1930 skb_shinfo(skb
)->gso_type
, skb
->ip_summed
);
1934 * Invalidate hardware checksum when packet is to be mangled, and
1935 * complete checksum manually on outgoing path.
1937 int skb_checksum_help(struct sk_buff
*skb
)
1940 int ret
= 0, offset
;
1942 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1943 goto out_set_summed
;
1945 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1946 skb_warn_bad_offload(skb
);
1950 offset
= skb_checksum_start_offset(skb
);
1951 BUG_ON(offset
>= skb_headlen(skb
));
1952 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1954 offset
+= skb
->csum_offset
;
1955 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
1957 if (skb_cloned(skb
) &&
1958 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
1959 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1964 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1966 skb
->ip_summed
= CHECKSUM_NONE
;
1970 EXPORT_SYMBOL(skb_checksum_help
);
1973 * skb_gso_segment - Perform segmentation on skb.
1974 * @skb: buffer to segment
1975 * @features: features for the output path (see dev->features)
1977 * This function segments the given skb and returns a list of segments.
1979 * It may return NULL if the skb requires no segmentation. This is
1980 * only possible when GSO is used for verifying header integrity.
1982 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
,
1983 netdev_features_t features
)
1985 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1986 struct packet_type
*ptype
;
1987 __be16 type
= skb
->protocol
;
1988 int vlan_depth
= ETH_HLEN
;
1991 while (type
== htons(ETH_P_8021Q
)) {
1992 struct vlan_hdr
*vh
;
1994 if (unlikely(!pskb_may_pull(skb
, vlan_depth
+ VLAN_HLEN
)))
1995 return ERR_PTR(-EINVAL
);
1997 vh
= (struct vlan_hdr
*)(skb
->data
+ vlan_depth
);
1998 type
= vh
->h_vlan_encapsulated_proto
;
1999 vlan_depth
+= VLAN_HLEN
;
2002 skb_reset_mac_header(skb
);
2003 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
2004 __skb_pull(skb
, skb
->mac_len
);
2006 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
2007 skb_warn_bad_offload(skb
);
2009 if (skb_header_cloned(skb
) &&
2010 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
2011 return ERR_PTR(err
);
2015 list_for_each_entry_rcu(ptype
,
2016 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
2017 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
2018 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
2019 err
= ptype
->gso_send_check(skb
);
2020 segs
= ERR_PTR(err
);
2021 if (err
|| skb_gso_ok(skb
, features
))
2023 __skb_push(skb
, (skb
->data
-
2024 skb_network_header(skb
)));
2026 segs
= ptype
->gso_segment(skb
, features
);
2032 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
2036 EXPORT_SYMBOL(skb_gso_segment
);
2038 /* Take action when hardware reception checksum errors are detected. */
2040 void netdev_rx_csum_fault(struct net_device
*dev
)
2042 if (net_ratelimit()) {
2043 pr_err("%s: hw csum failure\n", dev
? dev
->name
: "<unknown>");
2047 EXPORT_SYMBOL(netdev_rx_csum_fault
);
2050 /* Actually, we should eliminate this check as soon as we know, that:
2051 * 1. IOMMU is present and allows to map all the memory.
2052 * 2. No high memory really exists on this machine.
2055 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
2057 #ifdef CONFIG_HIGHMEM
2059 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
2060 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2061 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2062 if (PageHighMem(skb_frag_page(frag
)))
2067 if (PCI_DMA_BUS_IS_PHYS
) {
2068 struct device
*pdev
= dev
->dev
.parent
;
2072 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2073 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2074 dma_addr_t addr
= page_to_phys(skb_frag_page(frag
));
2075 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
2084 void (*destructor
)(struct sk_buff
*skb
);
2087 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2089 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
2091 struct dev_gso_cb
*cb
;
2094 struct sk_buff
*nskb
= skb
->next
;
2096 skb
->next
= nskb
->next
;
2099 } while (skb
->next
);
2101 cb
= DEV_GSO_CB(skb
);
2103 cb
->destructor(skb
);
2107 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2108 * @skb: buffer to segment
2109 * @features: device features as applicable to this skb
2111 * This function segments the given skb and stores the list of segments
2114 static int dev_gso_segment(struct sk_buff
*skb
, netdev_features_t features
)
2116 struct sk_buff
*segs
;
2118 segs
= skb_gso_segment(skb
, features
);
2120 /* Verifying header integrity only. */
2125 return PTR_ERR(segs
);
2128 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
2129 skb
->destructor
= dev_gso_skb_destructor
;
2134 static bool can_checksum_protocol(netdev_features_t features
, __be16 protocol
)
2136 return ((features
& NETIF_F_GEN_CSUM
) ||
2137 ((features
& NETIF_F_V4_CSUM
) &&
2138 protocol
== htons(ETH_P_IP
)) ||
2139 ((features
& NETIF_F_V6_CSUM
) &&
2140 protocol
== htons(ETH_P_IPV6
)) ||
2141 ((features
& NETIF_F_FCOE_CRC
) &&
2142 protocol
== htons(ETH_P_FCOE
)));
2145 static netdev_features_t
harmonize_features(struct sk_buff
*skb
,
2146 __be16 protocol
, netdev_features_t features
)
2148 if (!can_checksum_protocol(features
, protocol
)) {
2149 features
&= ~NETIF_F_ALL_CSUM
;
2150 features
&= ~NETIF_F_SG
;
2151 } else if (illegal_highdma(skb
->dev
, skb
)) {
2152 features
&= ~NETIF_F_SG
;
2158 netdev_features_t
netif_skb_features(struct sk_buff
*skb
)
2160 __be16 protocol
= skb
->protocol
;
2161 netdev_features_t features
= skb
->dev
->features
;
2163 if (skb_shinfo(skb
)->gso_segs
> skb
->dev
->gso_max_segs
)
2164 features
&= ~NETIF_F_GSO_MASK
;
2166 if (protocol
== htons(ETH_P_8021Q
)) {
2167 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
2168 protocol
= veh
->h_vlan_encapsulated_proto
;
2169 } else if (!vlan_tx_tag_present(skb
)) {
2170 return harmonize_features(skb
, protocol
, features
);
2173 features
&= (skb
->dev
->vlan_features
| NETIF_F_HW_VLAN_TX
);
2175 if (protocol
!= htons(ETH_P_8021Q
)) {
2176 return harmonize_features(skb
, protocol
, features
);
2178 features
&= NETIF_F_SG
| NETIF_F_HIGHDMA
| NETIF_F_FRAGLIST
|
2179 NETIF_F_GEN_CSUM
| NETIF_F_HW_VLAN_TX
;
2180 return harmonize_features(skb
, protocol
, features
);
2183 EXPORT_SYMBOL(netif_skb_features
);
2186 * Returns true if either:
2187 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2188 * 2. skb is fragmented and the device does not support SG, or if
2189 * at least one of fragments is in highmem and device does not
2190 * support DMA from it.
2192 static inline int skb_needs_linearize(struct sk_buff
*skb
,
2195 return skb_is_nonlinear(skb
) &&
2196 ((skb_has_frag_list(skb
) &&
2197 !(features
& NETIF_F_FRAGLIST
)) ||
2198 (skb_shinfo(skb
)->nr_frags
&&
2199 !(features
& NETIF_F_SG
)));
2202 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
2203 struct netdev_queue
*txq
)
2205 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2206 int rc
= NETDEV_TX_OK
;
2207 unsigned int skb_len
;
2209 if (likely(!skb
->next
)) {
2210 netdev_features_t features
;
2213 * If device doesn't need skb->dst, release it right now while
2214 * its hot in this cpu cache
2216 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2219 if (!list_empty(&ptype_all
))
2220 dev_queue_xmit_nit(skb
, dev
);
2222 features
= netif_skb_features(skb
);
2224 if (vlan_tx_tag_present(skb
) &&
2225 !(features
& NETIF_F_HW_VLAN_TX
)) {
2226 skb
= __vlan_put_tag(skb
, vlan_tx_tag_get(skb
));
2233 if (netif_needs_gso(skb
, features
)) {
2234 if (unlikely(dev_gso_segment(skb
, features
)))
2239 if (skb_needs_linearize(skb
, features
) &&
2240 __skb_linearize(skb
))
2243 /* If packet is not checksummed and device does not
2244 * support checksumming for this protocol, complete
2245 * checksumming here.
2247 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2248 skb_set_transport_header(skb
,
2249 skb_checksum_start_offset(skb
));
2250 if (!(features
& NETIF_F_ALL_CSUM
) &&
2251 skb_checksum_help(skb
))
2257 rc
= ops
->ndo_start_xmit(skb
, dev
);
2258 trace_net_dev_xmit(skb
, rc
, dev
, skb_len
);
2259 if (rc
== NETDEV_TX_OK
)
2260 txq_trans_update(txq
);
2266 struct sk_buff
*nskb
= skb
->next
;
2268 skb
->next
= nskb
->next
;
2272 * If device doesn't need nskb->dst, release it right now while
2273 * its hot in this cpu cache
2275 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2278 skb_len
= nskb
->len
;
2279 rc
= ops
->ndo_start_xmit(nskb
, dev
);
2280 trace_net_dev_xmit(nskb
, rc
, dev
, skb_len
);
2281 if (unlikely(rc
!= NETDEV_TX_OK
)) {
2282 if (rc
& ~NETDEV_TX_MASK
)
2283 goto out_kfree_gso_skb
;
2284 nskb
->next
= skb
->next
;
2288 txq_trans_update(txq
);
2289 if (unlikely(netif_xmit_stopped(txq
) && skb
->next
))
2290 return NETDEV_TX_BUSY
;
2291 } while (skb
->next
);
2294 if (likely(skb
->next
== NULL
))
2295 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
2302 static u32 hashrnd __read_mostly
;
2305 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2306 * to be used as a distribution range.
2308 u16
__skb_tx_hash(const struct net_device
*dev
, const struct sk_buff
*skb
,
2309 unsigned int num_tx_queues
)
2313 u16 qcount
= num_tx_queues
;
2315 if (skb_rx_queue_recorded(skb
)) {
2316 hash
= skb_get_rx_queue(skb
);
2317 while (unlikely(hash
>= num_tx_queues
))
2318 hash
-= num_tx_queues
;
2323 u8 tc
= netdev_get_prio_tc_map(dev
, skb
->priority
);
2324 qoffset
= dev
->tc_to_txq
[tc
].offset
;
2325 qcount
= dev
->tc_to_txq
[tc
].count
;
2328 if (skb
->sk
&& skb
->sk
->sk_hash
)
2329 hash
= skb
->sk
->sk_hash
;
2331 hash
= (__force u16
) skb
->protocol
;
2332 hash
= jhash_1word(hash
, hashrnd
);
2334 return (u16
) (((u64
) hash
* qcount
) >> 32) + qoffset
;
2336 EXPORT_SYMBOL(__skb_tx_hash
);
2338 static inline u16
dev_cap_txqueue(struct net_device
*dev
, u16 queue_index
)
2340 if (unlikely(queue_index
>= dev
->real_num_tx_queues
)) {
2341 net_warn_ratelimited("%s selects TX queue %d, but real number of TX queues is %d\n",
2342 dev
->name
, queue_index
,
2343 dev
->real_num_tx_queues
);
2349 static inline int get_xps_queue(struct net_device
*dev
, struct sk_buff
*skb
)
2352 struct xps_dev_maps
*dev_maps
;
2353 struct xps_map
*map
;
2354 int queue_index
= -1;
2357 dev_maps
= rcu_dereference(dev
->xps_maps
);
2359 map
= rcu_dereference(
2360 dev_maps
->cpu_map
[raw_smp_processor_id()]);
2363 queue_index
= map
->queues
[0];
2366 if (skb
->sk
&& skb
->sk
->sk_hash
)
2367 hash
= skb
->sk
->sk_hash
;
2369 hash
= (__force u16
) skb
->protocol
^
2371 hash
= jhash_1word(hash
, hashrnd
);
2372 queue_index
= map
->queues
[
2373 ((u64
)hash
* map
->len
) >> 32];
2375 if (unlikely(queue_index
>= dev
->real_num_tx_queues
))
2387 static struct netdev_queue
*dev_pick_tx(struct net_device
*dev
,
2388 struct sk_buff
*skb
)
2391 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2393 if (dev
->real_num_tx_queues
== 1)
2395 else if (ops
->ndo_select_queue
) {
2396 queue_index
= ops
->ndo_select_queue(dev
, skb
);
2397 queue_index
= dev_cap_txqueue(dev
, queue_index
);
2399 struct sock
*sk
= skb
->sk
;
2400 queue_index
= sk_tx_queue_get(sk
);
2402 if (queue_index
< 0 || skb
->ooo_okay
||
2403 queue_index
>= dev
->real_num_tx_queues
) {
2404 int old_index
= queue_index
;
2406 queue_index
= get_xps_queue(dev
, skb
);
2407 if (queue_index
< 0)
2408 queue_index
= skb_tx_hash(dev
, skb
);
2410 if (queue_index
!= old_index
&& sk
) {
2411 struct dst_entry
*dst
=
2412 rcu_dereference_check(sk
->sk_dst_cache
, 1);
2414 if (dst
&& skb_dst(skb
) == dst
)
2415 sk_tx_queue_set(sk
, queue_index
);
2420 skb_set_queue_mapping(skb
, queue_index
);
2421 return netdev_get_tx_queue(dev
, queue_index
);
2424 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2425 struct net_device
*dev
,
2426 struct netdev_queue
*txq
)
2428 spinlock_t
*root_lock
= qdisc_lock(q
);
2432 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
2433 qdisc_calculate_pkt_len(skb
, q
);
2435 * Heuristic to force contended enqueues to serialize on a
2436 * separate lock before trying to get qdisc main lock.
2437 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2438 * and dequeue packets faster.
2440 contended
= qdisc_is_running(q
);
2441 if (unlikely(contended
))
2442 spin_lock(&q
->busylock
);
2444 spin_lock(root_lock
);
2445 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2448 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2449 qdisc_run_begin(q
)) {
2451 * This is a work-conserving queue; there are no old skbs
2452 * waiting to be sent out; and the qdisc is not running -
2453 * xmit the skb directly.
2455 if (!(dev
->priv_flags
& IFF_XMIT_DST_RELEASE
))
2458 qdisc_bstats_update(q
, skb
);
2460 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
)) {
2461 if (unlikely(contended
)) {
2462 spin_unlock(&q
->busylock
);
2469 rc
= NET_XMIT_SUCCESS
;
2472 rc
= q
->enqueue(skb
, q
) & NET_XMIT_MASK
;
2473 if (qdisc_run_begin(q
)) {
2474 if (unlikely(contended
)) {
2475 spin_unlock(&q
->busylock
);
2481 spin_unlock(root_lock
);
2482 if (unlikely(contended
))
2483 spin_unlock(&q
->busylock
);
2487 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2488 static void skb_update_prio(struct sk_buff
*skb
)
2490 struct netprio_map
*map
= rcu_dereference_bh(skb
->dev
->priomap
);
2492 if (!skb
->priority
&& skb
->sk
&& map
) {
2493 unsigned int prioidx
= skb
->sk
->sk_cgrp_prioidx
;
2495 if (prioidx
< map
->priomap_len
)
2496 skb
->priority
= map
->priomap
[prioidx
];
2500 #define skb_update_prio(skb)
2503 static DEFINE_PER_CPU(int, xmit_recursion
);
2504 #define RECURSION_LIMIT 10
2507 * dev_loopback_xmit - loop back @skb
2508 * @skb: buffer to transmit
2510 int dev_loopback_xmit(struct sk_buff
*skb
)
2512 skb_reset_mac_header(skb
);
2513 __skb_pull(skb
, skb_network_offset(skb
));
2514 skb
->pkt_type
= PACKET_LOOPBACK
;
2515 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
2516 WARN_ON(!skb_dst(skb
));
2521 EXPORT_SYMBOL(dev_loopback_xmit
);
2524 * dev_queue_xmit - transmit a buffer
2525 * @skb: buffer to transmit
2527 * Queue a buffer for transmission to a network device. The caller must
2528 * have set the device and priority and built the buffer before calling
2529 * this function. The function can be called from an interrupt.
2531 * A negative errno code is returned on a failure. A success does not
2532 * guarantee the frame will be transmitted as it may be dropped due
2533 * to congestion or traffic shaping.
2535 * -----------------------------------------------------------------------------------
2536 * I notice this method can also return errors from the queue disciplines,
2537 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2540 * Regardless of the return value, the skb is consumed, so it is currently
2541 * difficult to retry a send to this method. (You can bump the ref count
2542 * before sending to hold a reference for retry if you are careful.)
2544 * When calling this method, interrupts MUST be enabled. This is because
2545 * the BH enable code must have IRQs enabled so that it will not deadlock.
2548 int dev_queue_xmit(struct sk_buff
*skb
)
2550 struct net_device
*dev
= skb
->dev
;
2551 struct netdev_queue
*txq
;
2555 /* Disable soft irqs for various locks below. Also
2556 * stops preemption for RCU.
2560 skb_update_prio(skb
);
2562 txq
= dev_pick_tx(dev
, skb
);
2563 q
= rcu_dereference_bh(txq
->qdisc
);
2565 #ifdef CONFIG_NET_CLS_ACT
2566 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2568 trace_net_dev_queue(skb
);
2570 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2574 /* The device has no queue. Common case for software devices:
2575 loopback, all the sorts of tunnels...
2577 Really, it is unlikely that netif_tx_lock protection is necessary
2578 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2580 However, it is possible, that they rely on protection
2583 Check this and shot the lock. It is not prone from deadlocks.
2584 Either shot noqueue qdisc, it is even simpler 8)
2586 if (dev
->flags
& IFF_UP
) {
2587 int cpu
= smp_processor_id(); /* ok because BHs are off */
2589 if (txq
->xmit_lock_owner
!= cpu
) {
2591 if (__this_cpu_read(xmit_recursion
) > RECURSION_LIMIT
)
2592 goto recursion_alert
;
2594 HARD_TX_LOCK(dev
, txq
, cpu
);
2596 if (!netif_xmit_stopped(txq
)) {
2597 __this_cpu_inc(xmit_recursion
);
2598 rc
= dev_hard_start_xmit(skb
, dev
, txq
);
2599 __this_cpu_dec(xmit_recursion
);
2600 if (dev_xmit_complete(rc
)) {
2601 HARD_TX_UNLOCK(dev
, txq
);
2605 HARD_TX_UNLOCK(dev
, txq
);
2606 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2609 /* Recursion is detected! It is possible,
2613 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2619 rcu_read_unlock_bh();
2624 rcu_read_unlock_bh();
2627 EXPORT_SYMBOL(dev_queue_xmit
);
2630 /*=======================================================================
2632 =======================================================================*/
2634 int netdev_max_backlog __read_mostly
= 1000;
2635 int netdev_tstamp_prequeue __read_mostly
= 1;
2636 int netdev_budget __read_mostly
= 300;
2637 int weight_p __read_mostly
= 64; /* old backlog weight */
2639 /* Called with irq disabled */
2640 static inline void ____napi_schedule(struct softnet_data
*sd
,
2641 struct napi_struct
*napi
)
2643 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
2644 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2648 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2649 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2650 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2651 * if hash is a canonical 4-tuple hash over transport ports.
2653 void __skb_get_rxhash(struct sk_buff
*skb
)
2655 struct flow_keys keys
;
2658 if (!skb_flow_dissect(skb
, &keys
))
2662 if ((__force u16
)keys
.port16
[1] < (__force u16
)keys
.port16
[0])
2663 swap(keys
.port16
[0], keys
.port16
[1]);
2667 /* get a consistent hash (same value on both flow directions) */
2668 if ((__force u32
)keys
.dst
< (__force u32
)keys
.src
)
2669 swap(keys
.dst
, keys
.src
);
2671 hash
= jhash_3words((__force u32
)keys
.dst
,
2672 (__force u32
)keys
.src
,
2673 (__force u32
)keys
.ports
, hashrnd
);
2679 EXPORT_SYMBOL(__skb_get_rxhash
);
2683 /* One global table that all flow-based protocols share. */
2684 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
2685 EXPORT_SYMBOL(rps_sock_flow_table
);
2687 struct static_key rps_needed __read_mostly
;
2689 static struct rps_dev_flow
*
2690 set_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2691 struct rps_dev_flow
*rflow
, u16 next_cpu
)
2693 if (next_cpu
!= RPS_NO_CPU
) {
2694 #ifdef CONFIG_RFS_ACCEL
2695 struct netdev_rx_queue
*rxqueue
;
2696 struct rps_dev_flow_table
*flow_table
;
2697 struct rps_dev_flow
*old_rflow
;
2702 /* Should we steer this flow to a different hardware queue? */
2703 if (!skb_rx_queue_recorded(skb
) || !dev
->rx_cpu_rmap
||
2704 !(dev
->features
& NETIF_F_NTUPLE
))
2706 rxq_index
= cpu_rmap_lookup_index(dev
->rx_cpu_rmap
, next_cpu
);
2707 if (rxq_index
== skb_get_rx_queue(skb
))
2710 rxqueue
= dev
->_rx
+ rxq_index
;
2711 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2714 flow_id
= skb
->rxhash
& flow_table
->mask
;
2715 rc
= dev
->netdev_ops
->ndo_rx_flow_steer(dev
, skb
,
2716 rxq_index
, flow_id
);
2720 rflow
= &flow_table
->flows
[flow_id
];
2722 if (old_rflow
->filter
== rflow
->filter
)
2723 old_rflow
->filter
= RPS_NO_FILTER
;
2727 per_cpu(softnet_data
, next_cpu
).input_queue_head
;
2730 rflow
->cpu
= next_cpu
;
2735 * get_rps_cpu is called from netif_receive_skb and returns the target
2736 * CPU from the RPS map of the receiving queue for a given skb.
2737 * rcu_read_lock must be held on entry.
2739 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2740 struct rps_dev_flow
**rflowp
)
2742 struct netdev_rx_queue
*rxqueue
;
2743 struct rps_map
*map
;
2744 struct rps_dev_flow_table
*flow_table
;
2745 struct rps_sock_flow_table
*sock_flow_table
;
2749 if (skb_rx_queue_recorded(skb
)) {
2750 u16 index
= skb_get_rx_queue(skb
);
2751 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
2752 WARN_ONCE(dev
->real_num_rx_queues
> 1,
2753 "%s received packet on queue %u, but number "
2754 "of RX queues is %u\n",
2755 dev
->name
, index
, dev
->real_num_rx_queues
);
2758 rxqueue
= dev
->_rx
+ index
;
2762 map
= rcu_dereference(rxqueue
->rps_map
);
2764 if (map
->len
== 1 &&
2765 !rcu_access_pointer(rxqueue
->rps_flow_table
)) {
2766 tcpu
= map
->cpus
[0];
2767 if (cpu_online(tcpu
))
2771 } else if (!rcu_access_pointer(rxqueue
->rps_flow_table
)) {
2775 skb_reset_network_header(skb
);
2776 if (!skb_get_rxhash(skb
))
2779 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2780 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
2781 if (flow_table
&& sock_flow_table
) {
2783 struct rps_dev_flow
*rflow
;
2785 rflow
= &flow_table
->flows
[skb
->rxhash
& flow_table
->mask
];
2788 next_cpu
= sock_flow_table
->ents
[skb
->rxhash
&
2789 sock_flow_table
->mask
];
2792 * If the desired CPU (where last recvmsg was done) is
2793 * different from current CPU (one in the rx-queue flow
2794 * table entry), switch if one of the following holds:
2795 * - Current CPU is unset (equal to RPS_NO_CPU).
2796 * - Current CPU is offline.
2797 * - The current CPU's queue tail has advanced beyond the
2798 * last packet that was enqueued using this table entry.
2799 * This guarantees that all previous packets for the flow
2800 * have been dequeued, thus preserving in order delivery.
2802 if (unlikely(tcpu
!= next_cpu
) &&
2803 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
2804 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
2805 rflow
->last_qtail
)) >= 0))
2806 rflow
= set_rps_cpu(dev
, skb
, rflow
, next_cpu
);
2808 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
2816 tcpu
= map
->cpus
[((u64
) skb
->rxhash
* map
->len
) >> 32];
2818 if (cpu_online(tcpu
)) {
2828 #ifdef CONFIG_RFS_ACCEL
2831 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2832 * @dev: Device on which the filter was set
2833 * @rxq_index: RX queue index
2834 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2835 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2837 * Drivers that implement ndo_rx_flow_steer() should periodically call
2838 * this function for each installed filter and remove the filters for
2839 * which it returns %true.
2841 bool rps_may_expire_flow(struct net_device
*dev
, u16 rxq_index
,
2842 u32 flow_id
, u16 filter_id
)
2844 struct netdev_rx_queue
*rxqueue
= dev
->_rx
+ rxq_index
;
2845 struct rps_dev_flow_table
*flow_table
;
2846 struct rps_dev_flow
*rflow
;
2851 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2852 if (flow_table
&& flow_id
<= flow_table
->mask
) {
2853 rflow
= &flow_table
->flows
[flow_id
];
2854 cpu
= ACCESS_ONCE(rflow
->cpu
);
2855 if (rflow
->filter
== filter_id
&& cpu
!= RPS_NO_CPU
&&
2856 ((int)(per_cpu(softnet_data
, cpu
).input_queue_head
-
2857 rflow
->last_qtail
) <
2858 (int)(10 * flow_table
->mask
)))
2864 EXPORT_SYMBOL(rps_may_expire_flow
);
2866 #endif /* CONFIG_RFS_ACCEL */
2868 /* Called from hardirq (IPI) context */
2869 static void rps_trigger_softirq(void *data
)
2871 struct softnet_data
*sd
= data
;
2873 ____napi_schedule(sd
, &sd
->backlog
);
2877 #endif /* CONFIG_RPS */
2880 * Check if this softnet_data structure is another cpu one
2881 * If yes, queue it to our IPI list and return 1
2884 static int rps_ipi_queued(struct softnet_data
*sd
)
2887 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
2890 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
2891 mysd
->rps_ipi_list
= sd
;
2893 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2896 #endif /* CONFIG_RPS */
2901 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2902 * queue (may be a remote CPU queue).
2904 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
2905 unsigned int *qtail
)
2907 struct softnet_data
*sd
;
2908 unsigned long flags
;
2910 sd
= &per_cpu(softnet_data
, cpu
);
2912 local_irq_save(flags
);
2915 if (skb_queue_len(&sd
->input_pkt_queue
) <= netdev_max_backlog
) {
2916 if (skb_queue_len(&sd
->input_pkt_queue
)) {
2918 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
2919 input_queue_tail_incr_save(sd
, qtail
);
2921 local_irq_restore(flags
);
2922 return NET_RX_SUCCESS
;
2925 /* Schedule NAPI for backlog device
2926 * We can use non atomic operation since we own the queue lock
2928 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
2929 if (!rps_ipi_queued(sd
))
2930 ____napi_schedule(sd
, &sd
->backlog
);
2938 local_irq_restore(flags
);
2940 atomic_long_inc(&skb
->dev
->rx_dropped
);
2946 * netif_rx - post buffer to the network code
2947 * @skb: buffer to post
2949 * This function receives a packet from a device driver and queues it for
2950 * the upper (protocol) levels to process. It always succeeds. The buffer
2951 * may be dropped during processing for congestion control or by the
2955 * NET_RX_SUCCESS (no congestion)
2956 * NET_RX_DROP (packet was dropped)
2960 int netif_rx(struct sk_buff
*skb
)
2964 /* if netpoll wants it, pretend we never saw it */
2965 if (netpoll_rx(skb
))
2968 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
2970 trace_netif_rx(skb
);
2972 if (static_key_false(&rps_needed
)) {
2973 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
2979 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
2981 cpu
= smp_processor_id();
2983 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
2991 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
2996 EXPORT_SYMBOL(netif_rx
);
2998 int netif_rx_ni(struct sk_buff
*skb
)
3003 err
= netif_rx(skb
);
3004 if (local_softirq_pending())
3010 EXPORT_SYMBOL(netif_rx_ni
);
3012 static void net_tx_action(struct softirq_action
*h
)
3014 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3016 if (sd
->completion_queue
) {
3017 struct sk_buff
*clist
;
3019 local_irq_disable();
3020 clist
= sd
->completion_queue
;
3021 sd
->completion_queue
= NULL
;
3025 struct sk_buff
*skb
= clist
;
3026 clist
= clist
->next
;
3028 WARN_ON(atomic_read(&skb
->users
));
3029 trace_kfree_skb(skb
, net_tx_action
);
3034 if (sd
->output_queue
) {
3037 local_irq_disable();
3038 head
= sd
->output_queue
;
3039 sd
->output_queue
= NULL
;
3040 sd
->output_queue_tailp
= &sd
->output_queue
;
3044 struct Qdisc
*q
= head
;
3045 spinlock_t
*root_lock
;
3047 head
= head
->next_sched
;
3049 root_lock
= qdisc_lock(q
);
3050 if (spin_trylock(root_lock
)) {
3051 smp_mb__before_clear_bit();
3052 clear_bit(__QDISC_STATE_SCHED
,
3055 spin_unlock(root_lock
);
3057 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
3059 __netif_reschedule(q
);
3061 smp_mb__before_clear_bit();
3062 clear_bit(__QDISC_STATE_SCHED
,
3070 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3071 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3072 /* This hook is defined here for ATM LANE */
3073 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
3074 unsigned char *addr
) __read_mostly
;
3075 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
3078 #ifdef CONFIG_NET_CLS_ACT
3079 /* TODO: Maybe we should just force sch_ingress to be compiled in
3080 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3081 * a compare and 2 stores extra right now if we dont have it on
3082 * but have CONFIG_NET_CLS_ACT
3083 * NOTE: This doesn't stop any functionality; if you dont have
3084 * the ingress scheduler, you just can't add policies on ingress.
3087 static int ing_filter(struct sk_buff
*skb
, struct netdev_queue
*rxq
)
3089 struct net_device
*dev
= skb
->dev
;
3090 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
3091 int result
= TC_ACT_OK
;
3094 if (unlikely(MAX_RED_LOOP
< ttl
++)) {
3095 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3096 skb
->skb_iif
, dev
->ifindex
);
3100 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
3101 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
3104 if (q
!= &noop_qdisc
) {
3105 spin_lock(qdisc_lock(q
));
3106 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
3107 result
= qdisc_enqueue_root(skb
, q
);
3108 spin_unlock(qdisc_lock(q
));
3114 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
3115 struct packet_type
**pt_prev
,
3116 int *ret
, struct net_device
*orig_dev
)
3118 struct netdev_queue
*rxq
= rcu_dereference(skb
->dev
->ingress_queue
);
3120 if (!rxq
|| rxq
->qdisc
== &noop_qdisc
)
3124 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
3128 switch (ing_filter(skb
, rxq
)) {
3142 * netdev_rx_handler_register - register receive handler
3143 * @dev: device to register a handler for
3144 * @rx_handler: receive handler to register
3145 * @rx_handler_data: data pointer that is used by rx handler
3147 * Register a receive hander for a device. This handler will then be
3148 * called from __netif_receive_skb. A negative errno code is returned
3151 * The caller must hold the rtnl_mutex.
3153 * For a general description of rx_handler, see enum rx_handler_result.
3155 int netdev_rx_handler_register(struct net_device
*dev
,
3156 rx_handler_func_t
*rx_handler
,
3157 void *rx_handler_data
)
3161 if (dev
->rx_handler
)
3164 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
3165 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
3169 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
3172 * netdev_rx_handler_unregister - unregister receive handler
3173 * @dev: device to unregister a handler from
3175 * Unregister a receive hander from a device.
3177 * The caller must hold the rtnl_mutex.
3179 void netdev_rx_handler_unregister(struct net_device
*dev
)
3183 RCU_INIT_POINTER(dev
->rx_handler
, NULL
);
3184 RCU_INIT_POINTER(dev
->rx_handler_data
, NULL
);
3186 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
3189 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3190 * the special handling of PFMEMALLOC skbs.
3192 static bool skb_pfmemalloc_protocol(struct sk_buff
*skb
)
3194 switch (skb
->protocol
) {
3195 case __constant_htons(ETH_P_ARP
):
3196 case __constant_htons(ETH_P_IP
):
3197 case __constant_htons(ETH_P_IPV6
):
3198 case __constant_htons(ETH_P_8021Q
):
3205 static int __netif_receive_skb(struct sk_buff
*skb
)
3207 struct packet_type
*ptype
, *pt_prev
;
3208 rx_handler_func_t
*rx_handler
;
3209 struct net_device
*orig_dev
;
3210 struct net_device
*null_or_dev
;
3211 bool deliver_exact
= false;
3212 int ret
= NET_RX_DROP
;
3214 unsigned long pflags
= current
->flags
;
3216 net_timestamp_check(!netdev_tstamp_prequeue
, skb
);
3218 trace_netif_receive_skb(skb
);
3221 * PFMEMALLOC skbs are special, they should
3222 * - be delivered to SOCK_MEMALLOC sockets only
3223 * - stay away from userspace
3224 * - have bounded memory usage
3226 * Use PF_MEMALLOC as this saves us from propagating the allocation
3227 * context down to all allocation sites.
3229 if (sk_memalloc_socks() && skb_pfmemalloc(skb
))
3230 current
->flags
|= PF_MEMALLOC
;
3232 /* if we've gotten here through NAPI, check netpoll */
3233 if (netpoll_receive_skb(skb
))
3236 orig_dev
= skb
->dev
;
3238 skb_reset_network_header(skb
);
3239 skb_reset_transport_header(skb
);
3240 skb_reset_mac_len(skb
);
3247 skb
->skb_iif
= skb
->dev
->ifindex
;
3249 __this_cpu_inc(softnet_data
.processed
);
3251 if (skb
->protocol
== cpu_to_be16(ETH_P_8021Q
)) {
3252 skb
= vlan_untag(skb
);
3257 #ifdef CONFIG_NET_CLS_ACT
3258 if (skb
->tc_verd
& TC_NCLS
) {
3259 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
3264 if (sk_memalloc_socks() && skb_pfmemalloc(skb
))
3267 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
3268 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
3270 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3276 #ifdef CONFIG_NET_CLS_ACT
3277 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
3283 if (sk_memalloc_socks() && skb_pfmemalloc(skb
)
3284 && !skb_pfmemalloc_protocol(skb
))
3287 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
3288 if (vlan_tx_tag_present(skb
)) {
3290 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3293 if (vlan_do_receive(&skb
, !rx_handler
))
3295 else if (unlikely(!skb
))
3301 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3304 switch (rx_handler(&skb
)) {
3305 case RX_HANDLER_CONSUMED
:
3307 case RX_HANDLER_ANOTHER
:
3309 case RX_HANDLER_EXACT
:
3310 deliver_exact
= true;
3311 case RX_HANDLER_PASS
:
3318 /* deliver only exact match when indicated */
3319 null_or_dev
= deliver_exact
? skb
->dev
: NULL
;
3321 type
= skb
->protocol
;
3322 list_for_each_entry_rcu(ptype
,
3323 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
3324 if (ptype
->type
== type
&&
3325 (ptype
->dev
== null_or_dev
|| ptype
->dev
== skb
->dev
||
3326 ptype
->dev
== orig_dev
)) {
3328 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3334 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
3337 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
3340 atomic_long_inc(&skb
->dev
->rx_dropped
);
3342 /* Jamal, now you will not able to escape explaining
3343 * me how you were going to use this. :-)
3351 tsk_restore_flags(current
, pflags
, PF_MEMALLOC
);
3356 * netif_receive_skb - process receive buffer from network
3357 * @skb: buffer to process
3359 * netif_receive_skb() is the main receive data processing function.
3360 * It always succeeds. The buffer may be dropped during processing
3361 * for congestion control or by the protocol layers.
3363 * This function may only be called from softirq context and interrupts
3364 * should be enabled.
3366 * Return values (usually ignored):
3367 * NET_RX_SUCCESS: no congestion
3368 * NET_RX_DROP: packet was dropped
3370 int netif_receive_skb(struct sk_buff
*skb
)
3372 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3374 if (skb_defer_rx_timestamp(skb
))
3375 return NET_RX_SUCCESS
;
3378 if (static_key_false(&rps_needed
)) {
3379 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3384 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3387 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3394 return __netif_receive_skb(skb
);
3396 EXPORT_SYMBOL(netif_receive_skb
);
3398 /* Network device is going away, flush any packets still pending
3399 * Called with irqs disabled.
3401 static void flush_backlog(void *arg
)
3403 struct net_device
*dev
= arg
;
3404 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3405 struct sk_buff
*skb
, *tmp
;
3408 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3409 if (skb
->dev
== dev
) {
3410 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3412 input_queue_head_incr(sd
);
3417 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3418 if (skb
->dev
== dev
) {
3419 __skb_unlink(skb
, &sd
->process_queue
);
3421 input_queue_head_incr(sd
);
3426 static int napi_gro_complete(struct sk_buff
*skb
)
3428 struct packet_type
*ptype
;
3429 __be16 type
= skb
->protocol
;
3430 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3433 if (NAPI_GRO_CB(skb
)->count
== 1) {
3434 skb_shinfo(skb
)->gso_size
= 0;
3439 list_for_each_entry_rcu(ptype
, head
, list
) {
3440 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_complete
)
3443 err
= ptype
->gro_complete(skb
);
3449 WARN_ON(&ptype
->list
== head
);
3451 return NET_RX_SUCCESS
;
3455 return netif_receive_skb(skb
);
3458 inline void napi_gro_flush(struct napi_struct
*napi
)
3460 struct sk_buff
*skb
, *next
;
3462 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3465 napi_gro_complete(skb
);
3468 napi
->gro_count
= 0;
3469 napi
->gro_list
= NULL
;
3471 EXPORT_SYMBOL(napi_gro_flush
);
3473 enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3475 struct sk_buff
**pp
= NULL
;
3476 struct packet_type
*ptype
;
3477 __be16 type
= skb
->protocol
;
3478 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3481 enum gro_result ret
;
3483 if (!(skb
->dev
->features
& NETIF_F_GRO
) || netpoll_rx_on(skb
))
3486 if (skb_is_gso(skb
) || skb_has_frag_list(skb
))
3490 list_for_each_entry_rcu(ptype
, head
, list
) {
3491 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_receive
)
3494 skb_set_network_header(skb
, skb_gro_offset(skb
));
3495 mac_len
= skb
->network_header
- skb
->mac_header
;
3496 skb
->mac_len
= mac_len
;
3497 NAPI_GRO_CB(skb
)->same_flow
= 0;
3498 NAPI_GRO_CB(skb
)->flush
= 0;
3499 NAPI_GRO_CB(skb
)->free
= 0;
3501 pp
= ptype
->gro_receive(&napi
->gro_list
, skb
);
3506 if (&ptype
->list
== head
)
3509 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3510 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3513 struct sk_buff
*nskb
= *pp
;
3517 napi_gro_complete(nskb
);
3524 if (NAPI_GRO_CB(skb
)->flush
|| napi
->gro_count
>= MAX_GRO_SKBS
)
3528 NAPI_GRO_CB(skb
)->count
= 1;
3529 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
3530 skb
->next
= napi
->gro_list
;
3531 napi
->gro_list
= skb
;
3535 if (skb_headlen(skb
) < skb_gro_offset(skb
)) {
3536 int grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
3538 BUG_ON(skb
->end
- skb
->tail
< grow
);
3540 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3543 skb
->data_len
-= grow
;
3545 skb_shinfo(skb
)->frags
[0].page_offset
+= grow
;
3546 skb_frag_size_sub(&skb_shinfo(skb
)->frags
[0], grow
);
3548 if (unlikely(!skb_frag_size(&skb_shinfo(skb
)->frags
[0]))) {
3549 skb_frag_unref(skb
, 0);
3550 memmove(skb_shinfo(skb
)->frags
,
3551 skb_shinfo(skb
)->frags
+ 1,
3552 --skb_shinfo(skb
)->nr_frags
* sizeof(skb_frag_t
));
3563 EXPORT_SYMBOL(dev_gro_receive
);
3565 static inline gro_result_t
3566 __napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3569 unsigned int maclen
= skb
->dev
->hard_header_len
;
3571 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3572 unsigned long diffs
;
3574 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
3575 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
3576 if (maclen
== ETH_HLEN
)
3577 diffs
|= compare_ether_header(skb_mac_header(p
),
3578 skb_gro_mac_header(skb
));
3580 diffs
= memcmp(skb_mac_header(p
),
3581 skb_gro_mac_header(skb
),
3583 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
3584 NAPI_GRO_CB(p
)->flush
= 0;
3587 return dev_gro_receive(napi
, skb
);
3590 gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
3594 if (netif_receive_skb(skb
))
3602 case GRO_MERGED_FREE
:
3603 if (NAPI_GRO_CB(skb
)->free
== NAPI_GRO_FREE_STOLEN_HEAD
)
3604 kmem_cache_free(skbuff_head_cache
, skb
);
3616 EXPORT_SYMBOL(napi_skb_finish
);
3618 void skb_gro_reset_offset(struct sk_buff
*skb
)
3620 NAPI_GRO_CB(skb
)->data_offset
= 0;
3621 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3622 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3624 if (skb
->mac_header
== skb
->tail
&&
3625 !PageHighMem(skb_frag_page(&skb_shinfo(skb
)->frags
[0]))) {
3626 NAPI_GRO_CB(skb
)->frag0
=
3627 skb_frag_address(&skb_shinfo(skb
)->frags
[0]);
3628 NAPI_GRO_CB(skb
)->frag0_len
= skb_frag_size(&skb_shinfo(skb
)->frags
[0]);
3631 EXPORT_SYMBOL(skb_gro_reset_offset
);
3633 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3635 skb_gro_reset_offset(skb
);
3637 return napi_skb_finish(__napi_gro_receive(napi
, skb
), skb
);
3639 EXPORT_SYMBOL(napi_gro_receive
);
3641 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
3643 __skb_pull(skb
, skb_headlen(skb
));
3644 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3645 skb_reserve(skb
, NET_SKB_PAD
+ NET_IP_ALIGN
- skb_headroom(skb
));
3647 skb
->dev
= napi
->dev
;
3653 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
3655 struct sk_buff
*skb
= napi
->skb
;
3658 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
3664 EXPORT_SYMBOL(napi_get_frags
);
3666 gro_result_t
napi_frags_finish(struct napi_struct
*napi
, struct sk_buff
*skb
,
3672 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
3674 if (ret
== GRO_HELD
)
3675 skb_gro_pull(skb
, -ETH_HLEN
);
3676 else if (netif_receive_skb(skb
))
3681 case GRO_MERGED_FREE
:
3682 napi_reuse_skb(napi
, skb
);
3691 EXPORT_SYMBOL(napi_frags_finish
);
3693 static struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
3695 struct sk_buff
*skb
= napi
->skb
;
3702 skb_reset_mac_header(skb
);
3703 skb_gro_reset_offset(skb
);
3705 off
= skb_gro_offset(skb
);
3706 hlen
= off
+ sizeof(*eth
);
3707 eth
= skb_gro_header_fast(skb
, off
);
3708 if (skb_gro_header_hard(skb
, hlen
)) {
3709 eth
= skb_gro_header_slow(skb
, hlen
, off
);
3710 if (unlikely(!eth
)) {
3711 napi_reuse_skb(napi
, skb
);
3717 skb_gro_pull(skb
, sizeof(*eth
));
3720 * This works because the only protocols we care about don't require
3721 * special handling. We'll fix it up properly at the end.
3723 skb
->protocol
= eth
->h_proto
;
3729 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
3731 struct sk_buff
*skb
= napi_frags_skb(napi
);
3736 return napi_frags_finish(napi
, skb
, __napi_gro_receive(napi
, skb
));
3738 EXPORT_SYMBOL(napi_gro_frags
);
3741 * net_rps_action sends any pending IPI's for rps.
3742 * Note: called with local irq disabled, but exits with local irq enabled.
3744 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
3747 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
3750 sd
->rps_ipi_list
= NULL
;
3754 /* Send pending IPI's to kick RPS processing on remote cpus. */
3756 struct softnet_data
*next
= remsd
->rps_ipi_next
;
3758 if (cpu_online(remsd
->cpu
))
3759 __smp_call_function_single(remsd
->cpu
,
3768 static int process_backlog(struct napi_struct
*napi
, int quota
)
3771 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
3774 /* Check if we have pending ipi, its better to send them now,
3775 * not waiting net_rx_action() end.
3777 if (sd
->rps_ipi_list
) {
3778 local_irq_disable();
3779 net_rps_action_and_irq_enable(sd
);
3782 napi
->weight
= weight_p
;
3783 local_irq_disable();
3784 while (work
< quota
) {
3785 struct sk_buff
*skb
;
3788 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
3790 __netif_receive_skb(skb
);
3791 local_irq_disable();
3792 input_queue_head_incr(sd
);
3793 if (++work
>= quota
) {
3800 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3802 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
3803 &sd
->process_queue
);
3805 if (qlen
< quota
- work
) {
3807 * Inline a custom version of __napi_complete().
3808 * only current cpu owns and manipulates this napi,
3809 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3810 * we can use a plain write instead of clear_bit(),
3811 * and we dont need an smp_mb() memory barrier.
3813 list_del(&napi
->poll_list
);
3816 quota
= work
+ qlen
;
3826 * __napi_schedule - schedule for receive
3827 * @n: entry to schedule
3829 * The entry's receive function will be scheduled to run
3831 void __napi_schedule(struct napi_struct
*n
)
3833 unsigned long flags
;
3835 local_irq_save(flags
);
3836 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
3837 local_irq_restore(flags
);
3839 EXPORT_SYMBOL(__napi_schedule
);
3841 void __napi_complete(struct napi_struct
*n
)
3843 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
3844 BUG_ON(n
->gro_list
);
3846 list_del(&n
->poll_list
);
3847 smp_mb__before_clear_bit();
3848 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
3850 EXPORT_SYMBOL(__napi_complete
);
3852 void napi_complete(struct napi_struct
*n
)
3854 unsigned long flags
;
3857 * don't let napi dequeue from the cpu poll list
3858 * just in case its running on a different cpu
3860 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
3864 local_irq_save(flags
);
3866 local_irq_restore(flags
);
3868 EXPORT_SYMBOL(napi_complete
);
3870 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
3871 int (*poll
)(struct napi_struct
*, int), int weight
)
3873 INIT_LIST_HEAD(&napi
->poll_list
);
3874 napi
->gro_count
= 0;
3875 napi
->gro_list
= NULL
;
3878 napi
->weight
= weight
;
3879 list_add(&napi
->dev_list
, &dev
->napi_list
);
3881 #ifdef CONFIG_NETPOLL
3882 spin_lock_init(&napi
->poll_lock
);
3883 napi
->poll_owner
= -1;
3885 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
3887 EXPORT_SYMBOL(netif_napi_add
);
3889 void netif_napi_del(struct napi_struct
*napi
)
3891 struct sk_buff
*skb
, *next
;
3893 list_del_init(&napi
->dev_list
);
3894 napi_free_frags(napi
);
3896 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3902 napi
->gro_list
= NULL
;
3903 napi
->gro_count
= 0;
3905 EXPORT_SYMBOL(netif_napi_del
);
3907 static void net_rx_action(struct softirq_action
*h
)
3909 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3910 unsigned long time_limit
= jiffies
+ 2;
3911 int budget
= netdev_budget
;
3914 local_irq_disable();
3916 while (!list_empty(&sd
->poll_list
)) {
3917 struct napi_struct
*n
;
3920 /* If softirq window is exhuasted then punt.
3921 * Allow this to run for 2 jiffies since which will allow
3922 * an average latency of 1.5/HZ.
3924 if (unlikely(budget
<= 0 || time_after(jiffies
, time_limit
)))
3929 /* Even though interrupts have been re-enabled, this
3930 * access is safe because interrupts can only add new
3931 * entries to the tail of this list, and only ->poll()
3932 * calls can remove this head entry from the list.
3934 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
3936 have
= netpoll_poll_lock(n
);
3940 /* This NAPI_STATE_SCHED test is for avoiding a race
3941 * with netpoll's poll_napi(). Only the entity which
3942 * obtains the lock and sees NAPI_STATE_SCHED set will
3943 * actually make the ->poll() call. Therefore we avoid
3944 * accidentally calling ->poll() when NAPI is not scheduled.
3947 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
3948 work
= n
->poll(n
, weight
);
3952 WARN_ON_ONCE(work
> weight
);
3956 local_irq_disable();
3958 /* Drivers must not modify the NAPI state if they
3959 * consume the entire weight. In such cases this code
3960 * still "owns" the NAPI instance and therefore can
3961 * move the instance around on the list at-will.
3963 if (unlikely(work
== weight
)) {
3964 if (unlikely(napi_disable_pending(n
))) {
3967 local_irq_disable();
3969 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
3972 netpoll_poll_unlock(have
);
3975 net_rps_action_and_irq_enable(sd
);
3977 #ifdef CONFIG_NET_DMA
3979 * There may not be any more sk_buffs coming right now, so push
3980 * any pending DMA copies to hardware
3982 dma_issue_pending_all();
3989 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3993 static gifconf_func_t
*gifconf_list
[NPROTO
];
3996 * register_gifconf - register a SIOCGIF handler
3997 * @family: Address family
3998 * @gifconf: Function handler
4000 * Register protocol dependent address dumping routines. The handler
4001 * that is passed must not be freed or reused until it has been replaced
4002 * by another handler.
4004 int register_gifconf(unsigned int family
, gifconf_func_t
*gifconf
)
4006 if (family
>= NPROTO
)
4008 gifconf_list
[family
] = gifconf
;
4011 EXPORT_SYMBOL(register_gifconf
);
4015 * Map an interface index to its name (SIOCGIFNAME)
4019 * We need this ioctl for efficient implementation of the
4020 * if_indextoname() function required by the IPv6 API. Without
4021 * it, we would have to search all the interfaces to find a
4025 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
4027 struct net_device
*dev
;
4031 * Fetch the caller's info block.
4034 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
4038 dev
= dev_get_by_index_rcu(net
, ifr
.ifr_ifindex
);
4044 strcpy(ifr
.ifr_name
, dev
->name
);
4047 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
4053 * Perform a SIOCGIFCONF call. This structure will change
4054 * size eventually, and there is nothing I can do about it.
4055 * Thus we will need a 'compatibility mode'.
4058 static int dev_ifconf(struct net
*net
, char __user
*arg
)
4061 struct net_device
*dev
;
4068 * Fetch the caller's info block.
4071 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
4078 * Loop over the interfaces, and write an info block for each.
4082 for_each_netdev(net
, dev
) {
4083 for (i
= 0; i
< NPROTO
; i
++) {
4084 if (gifconf_list
[i
]) {
4087 done
= gifconf_list
[i
](dev
, NULL
, 0);
4089 done
= gifconf_list
[i
](dev
, pos
+ total
,
4099 * All done. Write the updated control block back to the caller.
4101 ifc
.ifc_len
= total
;
4104 * Both BSD and Solaris return 0 here, so we do too.
4106 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
4109 #ifdef CONFIG_PROC_FS
4111 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4113 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4114 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4115 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4117 static inline struct net_device
*dev_from_same_bucket(struct seq_file
*seq
, loff_t
*pos
)
4119 struct net
*net
= seq_file_net(seq
);
4120 struct net_device
*dev
;
4121 struct hlist_node
*p
;
4122 struct hlist_head
*h
;
4123 unsigned int count
= 0, offset
= get_offset(*pos
);
4125 h
= &net
->dev_name_head
[get_bucket(*pos
)];
4126 hlist_for_each_entry_rcu(dev
, p
, h
, name_hlist
) {
4127 if (++count
== offset
)
4134 static inline struct net_device
*dev_from_bucket(struct seq_file
*seq
, loff_t
*pos
)
4136 struct net_device
*dev
;
4137 unsigned int bucket
;
4140 dev
= dev_from_same_bucket(seq
, pos
);
4144 bucket
= get_bucket(*pos
) + 1;
4145 *pos
= set_bucket_offset(bucket
, 1);
4146 } while (bucket
< NETDEV_HASHENTRIES
);
4152 * This is invoked by the /proc filesystem handler to display a device
4155 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4160 return SEQ_START_TOKEN
;
4162 if (get_bucket(*pos
) >= NETDEV_HASHENTRIES
)
4165 return dev_from_bucket(seq
, pos
);
4168 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4171 return dev_from_bucket(seq
, pos
);
4174 void dev_seq_stop(struct seq_file
*seq
, void *v
)
4180 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
4182 struct rtnl_link_stats64 temp
;
4183 const struct rtnl_link_stats64
*stats
= dev_get_stats(dev
, &temp
);
4185 seq_printf(seq
, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4186 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4187 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
4189 stats
->rx_dropped
+ stats
->rx_missed_errors
,
4190 stats
->rx_fifo_errors
,
4191 stats
->rx_length_errors
+ stats
->rx_over_errors
+
4192 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
4193 stats
->rx_compressed
, stats
->multicast
,
4194 stats
->tx_bytes
, stats
->tx_packets
,
4195 stats
->tx_errors
, stats
->tx_dropped
,
4196 stats
->tx_fifo_errors
, stats
->collisions
,
4197 stats
->tx_carrier_errors
+
4198 stats
->tx_aborted_errors
+
4199 stats
->tx_window_errors
+
4200 stats
->tx_heartbeat_errors
,
4201 stats
->tx_compressed
);
4205 * Called from the PROCfs module. This now uses the new arbitrary sized
4206 * /proc/net interface to create /proc/net/dev
4208 static int dev_seq_show(struct seq_file
*seq
, void *v
)
4210 if (v
== SEQ_START_TOKEN
)
4211 seq_puts(seq
, "Inter-| Receive "
4213 " face |bytes packets errs drop fifo frame "
4214 "compressed multicast|bytes packets errs "
4215 "drop fifo colls carrier compressed\n");
4217 dev_seq_printf_stats(seq
, v
);
4221 static struct softnet_data
*softnet_get_online(loff_t
*pos
)
4223 struct softnet_data
*sd
= NULL
;
4225 while (*pos
< nr_cpu_ids
)
4226 if (cpu_online(*pos
)) {
4227 sd
= &per_cpu(softnet_data
, *pos
);
4234 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4236 return softnet_get_online(pos
);
4239 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4242 return softnet_get_online(pos
);
4245 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
4249 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
4251 struct softnet_data
*sd
= v
;
4253 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4254 sd
->processed
, sd
->dropped
, sd
->time_squeeze
, 0,
4255 0, 0, 0, 0, /* was fastroute */
4256 sd
->cpu_collision
, sd
->received_rps
);
4260 static const struct seq_operations dev_seq_ops
= {
4261 .start
= dev_seq_start
,
4262 .next
= dev_seq_next
,
4263 .stop
= dev_seq_stop
,
4264 .show
= dev_seq_show
,
4267 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
4269 return seq_open_net(inode
, file
, &dev_seq_ops
,
4270 sizeof(struct seq_net_private
));
4273 static const struct file_operations dev_seq_fops
= {
4274 .owner
= THIS_MODULE
,
4275 .open
= dev_seq_open
,
4277 .llseek
= seq_lseek
,
4278 .release
= seq_release_net
,
4281 static const struct seq_operations softnet_seq_ops
= {
4282 .start
= softnet_seq_start
,
4283 .next
= softnet_seq_next
,
4284 .stop
= softnet_seq_stop
,
4285 .show
= softnet_seq_show
,
4288 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
4290 return seq_open(file
, &softnet_seq_ops
);
4293 static const struct file_operations softnet_seq_fops
= {
4294 .owner
= THIS_MODULE
,
4295 .open
= softnet_seq_open
,
4297 .llseek
= seq_lseek
,
4298 .release
= seq_release
,
4301 static void *ptype_get_idx(loff_t pos
)
4303 struct packet_type
*pt
= NULL
;
4307 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
4313 for (t
= 0; t
< PTYPE_HASH_SIZE
; t
++) {
4314 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
4323 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4327 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
4330 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4332 struct packet_type
*pt
;
4333 struct list_head
*nxt
;
4337 if (v
== SEQ_START_TOKEN
)
4338 return ptype_get_idx(0);
4341 nxt
= pt
->list
.next
;
4342 if (pt
->type
== htons(ETH_P_ALL
)) {
4343 if (nxt
!= &ptype_all
)
4346 nxt
= ptype_base
[0].next
;
4348 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
4350 while (nxt
== &ptype_base
[hash
]) {
4351 if (++hash
>= PTYPE_HASH_SIZE
)
4353 nxt
= ptype_base
[hash
].next
;
4356 return list_entry(nxt
, struct packet_type
, list
);
4359 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
4365 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
4367 struct packet_type
*pt
= v
;
4369 if (v
== SEQ_START_TOKEN
)
4370 seq_puts(seq
, "Type Device Function\n");
4371 else if (pt
->dev
== NULL
|| dev_net(pt
->dev
) == seq_file_net(seq
)) {
4372 if (pt
->type
== htons(ETH_P_ALL
))
4373 seq_puts(seq
, "ALL ");
4375 seq_printf(seq
, "%04x", ntohs(pt
->type
));
4377 seq_printf(seq
, " %-8s %pF\n",
4378 pt
->dev
? pt
->dev
->name
: "", pt
->func
);
4384 static const struct seq_operations ptype_seq_ops
= {
4385 .start
= ptype_seq_start
,
4386 .next
= ptype_seq_next
,
4387 .stop
= ptype_seq_stop
,
4388 .show
= ptype_seq_show
,
4391 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
4393 return seq_open_net(inode
, file
, &ptype_seq_ops
,
4394 sizeof(struct seq_net_private
));
4397 static const struct file_operations ptype_seq_fops
= {
4398 .owner
= THIS_MODULE
,
4399 .open
= ptype_seq_open
,
4401 .llseek
= seq_lseek
,
4402 .release
= seq_release_net
,
4406 static int __net_init
dev_proc_net_init(struct net
*net
)
4410 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
4412 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
4414 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
4417 if (wext_proc_init(net
))
4423 proc_net_remove(net
, "ptype");
4425 proc_net_remove(net
, "softnet_stat");
4427 proc_net_remove(net
, "dev");
4431 static void __net_exit
dev_proc_net_exit(struct net
*net
)
4433 wext_proc_exit(net
);
4435 proc_net_remove(net
, "ptype");
4436 proc_net_remove(net
, "softnet_stat");
4437 proc_net_remove(net
, "dev");
4440 static struct pernet_operations __net_initdata dev_proc_ops
= {
4441 .init
= dev_proc_net_init
,
4442 .exit
= dev_proc_net_exit
,
4445 static int __init
dev_proc_init(void)
4447 return register_pernet_subsys(&dev_proc_ops
);
4450 #define dev_proc_init() 0
4451 #endif /* CONFIG_PROC_FS */
4455 * netdev_set_master - set up master pointer
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 the reference counts
4462 * are adjusted and the function returns zero.
4464 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
4466 struct net_device
*old
= slave
->master
;
4476 slave
->master
= master
;
4482 EXPORT_SYMBOL(netdev_set_master
);
4485 * netdev_set_bond_master - set up bonding master/slave pair
4486 * @slave: slave device
4487 * @master: new master device
4489 * Changes the master device of the slave. Pass %NULL to break the
4490 * bonding. The caller must hold the RTNL semaphore. On a failure
4491 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4492 * to the routing socket and the function returns zero.
4494 int netdev_set_bond_master(struct net_device
*slave
, struct net_device
*master
)
4500 err
= netdev_set_master(slave
, master
);
4504 slave
->flags
|= IFF_SLAVE
;
4506 slave
->flags
&= ~IFF_SLAVE
;
4508 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
4511 EXPORT_SYMBOL(netdev_set_bond_master
);
4513 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
4515 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4517 if ((dev
->flags
& IFF_UP
) && ops
->ndo_change_rx_flags
)
4518 ops
->ndo_change_rx_flags(dev
, flags
);
4521 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
4523 unsigned int old_flags
= dev
->flags
;
4529 dev
->flags
|= IFF_PROMISC
;
4530 dev
->promiscuity
+= inc
;
4531 if (dev
->promiscuity
== 0) {
4534 * If inc causes overflow, untouch promisc and return error.
4537 dev
->flags
&= ~IFF_PROMISC
;
4539 dev
->promiscuity
-= inc
;
4540 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4545 if (dev
->flags
!= old_flags
) {
4546 pr_info("device %s %s promiscuous mode\n",
4548 dev
->flags
& IFF_PROMISC
? "entered" : "left");
4549 if (audit_enabled
) {
4550 current_uid_gid(&uid
, &gid
);
4551 audit_log(current
->audit_context
, GFP_ATOMIC
,
4552 AUDIT_ANOM_PROMISCUOUS
,
4553 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4554 dev
->name
, (dev
->flags
& IFF_PROMISC
),
4555 (old_flags
& IFF_PROMISC
),
4556 audit_get_loginuid(current
),
4558 audit_get_sessionid(current
));
4561 dev_change_rx_flags(dev
, IFF_PROMISC
);
4567 * dev_set_promiscuity - update promiscuity count on a device
4571 * Add or remove promiscuity from a device. While the count in the device
4572 * remains above zero the interface remains promiscuous. Once it hits zero
4573 * the device reverts back to normal filtering operation. A negative inc
4574 * value is used to drop promiscuity on the device.
4575 * Return 0 if successful or a negative errno code on error.
4577 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
4579 unsigned int old_flags
= dev
->flags
;
4582 err
= __dev_set_promiscuity(dev
, inc
);
4585 if (dev
->flags
!= old_flags
)
4586 dev_set_rx_mode(dev
);
4589 EXPORT_SYMBOL(dev_set_promiscuity
);
4592 * dev_set_allmulti - update allmulti count on a device
4596 * Add or remove reception of all multicast frames to a device. While the
4597 * count in the device remains above zero the interface remains listening
4598 * to all interfaces. Once it hits zero the device reverts back to normal
4599 * filtering operation. A negative @inc value is used to drop the counter
4600 * when releasing a resource needing all multicasts.
4601 * Return 0 if successful or a negative errno code on error.
4604 int dev_set_allmulti(struct net_device
*dev
, int inc
)
4606 unsigned int old_flags
= dev
->flags
;
4610 dev
->flags
|= IFF_ALLMULTI
;
4611 dev
->allmulti
+= inc
;
4612 if (dev
->allmulti
== 0) {
4615 * If inc causes overflow, untouch allmulti and return error.
4618 dev
->flags
&= ~IFF_ALLMULTI
;
4620 dev
->allmulti
-= inc
;
4621 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4626 if (dev
->flags
^ old_flags
) {
4627 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
4628 dev_set_rx_mode(dev
);
4632 EXPORT_SYMBOL(dev_set_allmulti
);
4635 * Upload unicast and multicast address lists to device and
4636 * configure RX filtering. When the device doesn't support unicast
4637 * filtering it is put in promiscuous mode while unicast addresses
4640 void __dev_set_rx_mode(struct net_device
*dev
)
4642 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4644 /* dev_open will call this function so the list will stay sane. */
4645 if (!(dev
->flags
&IFF_UP
))
4648 if (!netif_device_present(dev
))
4651 if (!(dev
->priv_flags
& IFF_UNICAST_FLT
)) {
4652 /* Unicast addresses changes may only happen under the rtnl,
4653 * therefore calling __dev_set_promiscuity here is safe.
4655 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
4656 __dev_set_promiscuity(dev
, 1);
4657 dev
->uc_promisc
= true;
4658 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
4659 __dev_set_promiscuity(dev
, -1);
4660 dev
->uc_promisc
= false;
4664 if (ops
->ndo_set_rx_mode
)
4665 ops
->ndo_set_rx_mode(dev
);
4668 void dev_set_rx_mode(struct net_device
*dev
)
4670 netif_addr_lock_bh(dev
);
4671 __dev_set_rx_mode(dev
);
4672 netif_addr_unlock_bh(dev
);
4676 * dev_get_flags - get flags reported to userspace
4679 * Get the combination of flag bits exported through APIs to userspace.
4681 unsigned int dev_get_flags(const struct net_device
*dev
)
4685 flags
= (dev
->flags
& ~(IFF_PROMISC
|
4690 (dev
->gflags
& (IFF_PROMISC
|
4693 if (netif_running(dev
)) {
4694 if (netif_oper_up(dev
))
4695 flags
|= IFF_RUNNING
;
4696 if (netif_carrier_ok(dev
))
4697 flags
|= IFF_LOWER_UP
;
4698 if (netif_dormant(dev
))
4699 flags
|= IFF_DORMANT
;
4704 EXPORT_SYMBOL(dev_get_flags
);
4706 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4708 unsigned int old_flags
= dev
->flags
;
4714 * Set the flags on our device.
4717 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
4718 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
4720 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
4724 * Load in the correct multicast list now the flags have changed.
4727 if ((old_flags
^ flags
) & IFF_MULTICAST
)
4728 dev_change_rx_flags(dev
, IFF_MULTICAST
);
4730 dev_set_rx_mode(dev
);
4733 * Have we downed the interface. We handle IFF_UP ourselves
4734 * according to user attempts to set it, rather than blindly
4739 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
4740 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
4743 dev_set_rx_mode(dev
);
4746 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
4747 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
4749 dev
->gflags
^= IFF_PROMISC
;
4750 dev_set_promiscuity(dev
, inc
);
4753 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4754 is important. Some (broken) drivers set IFF_PROMISC, when
4755 IFF_ALLMULTI is requested not asking us and not reporting.
4757 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
4758 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
4760 dev
->gflags
^= IFF_ALLMULTI
;
4761 dev_set_allmulti(dev
, inc
);
4767 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
)
4769 unsigned int changes
= dev
->flags
^ old_flags
;
4771 if (changes
& IFF_UP
) {
4772 if (dev
->flags
& IFF_UP
)
4773 call_netdevice_notifiers(NETDEV_UP
, dev
);
4775 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
4778 if (dev
->flags
& IFF_UP
&&
4779 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
)))
4780 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
4784 * dev_change_flags - change device settings
4786 * @flags: device state flags
4788 * Change settings on device based state flags. The flags are
4789 * in the userspace exported format.
4791 int dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4794 unsigned int changes
, old_flags
= dev
->flags
;
4796 ret
= __dev_change_flags(dev
, flags
);
4800 changes
= old_flags
^ dev
->flags
;
4802 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
4804 __dev_notify_flags(dev
, old_flags
);
4807 EXPORT_SYMBOL(dev_change_flags
);
4810 * dev_set_mtu - Change maximum transfer unit
4812 * @new_mtu: new transfer unit
4814 * Change the maximum transfer size of the network device.
4816 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
4818 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4821 if (new_mtu
== dev
->mtu
)
4824 /* MTU must be positive. */
4828 if (!netif_device_present(dev
))
4832 if (ops
->ndo_change_mtu
)
4833 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
4837 if (!err
&& dev
->flags
& IFF_UP
)
4838 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
4841 EXPORT_SYMBOL(dev_set_mtu
);
4844 * dev_set_group - Change group this device belongs to
4846 * @new_group: group this device should belong to
4848 void dev_set_group(struct net_device
*dev
, int new_group
)
4850 dev
->group
= new_group
;
4852 EXPORT_SYMBOL(dev_set_group
);
4855 * dev_set_mac_address - Change Media Access Control Address
4859 * Change the hardware (MAC) address of the device
4861 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
4863 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4866 if (!ops
->ndo_set_mac_address
)
4868 if (sa
->sa_family
!= dev
->type
)
4870 if (!netif_device_present(dev
))
4872 err
= ops
->ndo_set_mac_address(dev
, sa
);
4874 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4875 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
4878 EXPORT_SYMBOL(dev_set_mac_address
);
4881 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4883 static int dev_ifsioc_locked(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4886 struct net_device
*dev
= dev_get_by_name_rcu(net
, ifr
->ifr_name
);
4892 case SIOCGIFFLAGS
: /* Get interface flags */
4893 ifr
->ifr_flags
= (short) dev_get_flags(dev
);
4896 case SIOCGIFMETRIC
: /* Get the metric on the interface
4897 (currently unused) */
4898 ifr
->ifr_metric
= 0;
4901 case SIOCGIFMTU
: /* Get the MTU of a device */
4902 ifr
->ifr_mtu
= dev
->mtu
;
4907 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
4909 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
4910 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4911 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
4919 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
4920 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
4921 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
4922 ifr
->ifr_map
.irq
= dev
->irq
;
4923 ifr
->ifr_map
.dma
= dev
->dma
;
4924 ifr
->ifr_map
.port
= dev
->if_port
;
4928 ifr
->ifr_ifindex
= dev
->ifindex
;
4932 ifr
->ifr_qlen
= dev
->tx_queue_len
;
4936 /* dev_ioctl() should ensure this case
4948 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4950 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4953 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
4954 const struct net_device_ops
*ops
;
4959 ops
= dev
->netdev_ops
;
4962 case SIOCSIFFLAGS
: /* Set interface flags */
4963 return dev_change_flags(dev
, ifr
->ifr_flags
);
4965 case SIOCSIFMETRIC
: /* Set the metric on the interface
4966 (currently unused) */
4969 case SIOCSIFMTU
: /* Set the MTU of a device */
4970 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
4973 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
4975 case SIOCSIFHWBROADCAST
:
4976 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
4978 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
4979 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4980 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4984 if (ops
->ndo_set_config
) {
4985 if (!netif_device_present(dev
))
4987 return ops
->ndo_set_config(dev
, &ifr
->ifr_map
);
4992 if (!ops
->ndo_set_rx_mode
||
4993 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4995 if (!netif_device_present(dev
))
4997 return dev_mc_add_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
5000 if (!ops
->ndo_set_rx_mode
||
5001 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
5003 if (!netif_device_present(dev
))
5005 return dev_mc_del_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
5008 if (ifr
->ifr_qlen
< 0)
5010 dev
->tx_queue_len
= ifr
->ifr_qlen
;
5014 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
5015 return dev_change_name(dev
, ifr
->ifr_newname
);
5018 err
= net_hwtstamp_validate(ifr
);
5024 * Unknown or private ioctl
5027 if ((cmd
>= SIOCDEVPRIVATE
&&
5028 cmd
<= SIOCDEVPRIVATE
+ 15) ||
5029 cmd
== SIOCBONDENSLAVE
||
5030 cmd
== SIOCBONDRELEASE
||
5031 cmd
== SIOCBONDSETHWADDR
||
5032 cmd
== SIOCBONDSLAVEINFOQUERY
||
5033 cmd
== SIOCBONDINFOQUERY
||
5034 cmd
== SIOCBONDCHANGEACTIVE
||
5035 cmd
== SIOCGMIIPHY
||
5036 cmd
== SIOCGMIIREG
||
5037 cmd
== SIOCSMIIREG
||
5038 cmd
== SIOCBRADDIF
||
5039 cmd
== SIOCBRDELIF
||
5040 cmd
== SIOCSHWTSTAMP
||
5041 cmd
== SIOCWANDEV
) {
5043 if (ops
->ndo_do_ioctl
) {
5044 if (netif_device_present(dev
))
5045 err
= ops
->ndo_do_ioctl(dev
, ifr
, cmd
);
5057 * This function handles all "interface"-type I/O control requests. The actual
5058 * 'doing' part of this is dev_ifsioc above.
5062 * dev_ioctl - network device ioctl
5063 * @net: the applicable net namespace
5064 * @cmd: command to issue
5065 * @arg: pointer to a struct ifreq in user space
5067 * Issue ioctl functions to devices. This is normally called by the
5068 * user space syscall interfaces but can sometimes be useful for
5069 * other purposes. The return value is the return from the syscall if
5070 * positive or a negative errno code on error.
5073 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
5079 /* One special case: SIOCGIFCONF takes ifconf argument
5080 and requires shared lock, because it sleeps writing
5084 if (cmd
== SIOCGIFCONF
) {
5086 ret
= dev_ifconf(net
, (char __user
*) arg
);
5090 if (cmd
== SIOCGIFNAME
)
5091 return dev_ifname(net
, (struct ifreq __user
*)arg
);
5093 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
5096 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
5098 colon
= strchr(ifr
.ifr_name
, ':');
5103 * See which interface the caller is talking about.
5108 * These ioctl calls:
5109 * - can be done by all.
5110 * - atomic and do not require locking.
5121 dev_load(net
, ifr
.ifr_name
);
5123 ret
= dev_ifsioc_locked(net
, &ifr
, cmd
);
5128 if (copy_to_user(arg
, &ifr
,
5129 sizeof(struct ifreq
)))
5135 dev_load(net
, ifr
.ifr_name
);
5137 ret
= dev_ethtool(net
, &ifr
);
5142 if (copy_to_user(arg
, &ifr
,
5143 sizeof(struct ifreq
)))
5149 * These ioctl calls:
5150 * - require superuser power.
5151 * - require strict serialization.
5157 if (!capable(CAP_NET_ADMIN
))
5159 dev_load(net
, ifr
.ifr_name
);
5161 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5166 if (copy_to_user(arg
, &ifr
,
5167 sizeof(struct ifreq
)))
5173 * These ioctl calls:
5174 * - require superuser power.
5175 * - require strict serialization.
5176 * - do not return a value
5186 case SIOCSIFHWBROADCAST
:
5189 case SIOCBONDENSLAVE
:
5190 case SIOCBONDRELEASE
:
5191 case SIOCBONDSETHWADDR
:
5192 case SIOCBONDCHANGEACTIVE
:
5196 if (!capable(CAP_NET_ADMIN
))
5199 case SIOCBONDSLAVEINFOQUERY
:
5200 case SIOCBONDINFOQUERY
:
5201 dev_load(net
, ifr
.ifr_name
);
5203 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5208 /* Get the per device memory space. We can add this but
5209 * currently do not support it */
5211 /* Set the per device memory buffer space.
5212 * Not applicable in our case */
5217 * Unknown or private ioctl.
5220 if (cmd
== SIOCWANDEV
||
5221 (cmd
>= SIOCDEVPRIVATE
&&
5222 cmd
<= SIOCDEVPRIVATE
+ 15)) {
5223 dev_load(net
, ifr
.ifr_name
);
5225 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5227 if (!ret
&& copy_to_user(arg
, &ifr
,
5228 sizeof(struct ifreq
)))
5232 /* Take care of Wireless Extensions */
5233 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
5234 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
5241 * dev_new_index - allocate an ifindex
5242 * @net: the applicable net namespace
5244 * Returns a suitable unique value for a new device interface
5245 * number. The caller must hold the rtnl semaphore or the
5246 * dev_base_lock to be sure it remains unique.
5248 static int dev_new_index(struct net
*net
)
5250 int ifindex
= net
->ifindex
;
5254 if (!__dev_get_by_index(net
, ifindex
))
5255 return net
->ifindex
= ifindex
;
5259 /* Delayed registration/unregisteration */
5260 static LIST_HEAD(net_todo_list
);
5262 static void net_set_todo(struct net_device
*dev
)
5264 list_add_tail(&dev
->todo_list
, &net_todo_list
);
5267 static void rollback_registered_many(struct list_head
*head
)
5269 struct net_device
*dev
, *tmp
;
5271 BUG_ON(dev_boot_phase
);
5274 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
5275 /* Some devices call without registering
5276 * for initialization unwind. Remove those
5277 * devices and proceed with the remaining.
5279 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5280 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5284 list_del(&dev
->unreg_list
);
5287 dev
->dismantle
= true;
5288 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
5291 /* If device is running, close it first. */
5292 dev_close_many(head
);
5294 list_for_each_entry(dev
, head
, unreg_list
) {
5295 /* And unlink it from device chain. */
5296 unlist_netdevice(dev
);
5298 dev
->reg_state
= NETREG_UNREGISTERING
;
5303 list_for_each_entry(dev
, head
, unreg_list
) {
5304 /* Shutdown queueing discipline. */
5308 /* Notify protocols, that we are about to destroy
5309 this device. They should clean all the things.
5311 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5313 if (!dev
->rtnl_link_ops
||
5314 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5315 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
5318 * Flush the unicast and multicast chains
5323 if (dev
->netdev_ops
->ndo_uninit
)
5324 dev
->netdev_ops
->ndo_uninit(dev
);
5326 /* Notifier chain MUST detach us from master device. */
5327 WARN_ON(dev
->master
);
5329 /* Remove entries from kobject tree */
5330 netdev_unregister_kobject(dev
);
5335 list_for_each_entry(dev
, head
, unreg_list
)
5339 static void rollback_registered(struct net_device
*dev
)
5343 list_add(&dev
->unreg_list
, &single
);
5344 rollback_registered_many(&single
);
5348 static netdev_features_t
netdev_fix_features(struct net_device
*dev
,
5349 netdev_features_t features
)
5351 /* Fix illegal checksum combinations */
5352 if ((features
& NETIF_F_HW_CSUM
) &&
5353 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5354 netdev_warn(dev
, "mixed HW and IP checksum settings.\n");
5355 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5358 /* Fix illegal SG+CSUM combinations. */
5359 if ((features
& NETIF_F_SG
) &&
5360 !(features
& NETIF_F_ALL_CSUM
)) {
5362 "Dropping NETIF_F_SG since no checksum feature.\n");
5363 features
&= ~NETIF_F_SG
;
5366 /* TSO requires that SG is present as well. */
5367 if ((features
& NETIF_F_ALL_TSO
) && !(features
& NETIF_F_SG
)) {
5368 netdev_dbg(dev
, "Dropping TSO features since no SG feature.\n");
5369 features
&= ~NETIF_F_ALL_TSO
;
5372 /* TSO ECN requires that TSO is present as well. */
5373 if ((features
& NETIF_F_ALL_TSO
) == NETIF_F_TSO_ECN
)
5374 features
&= ~NETIF_F_TSO_ECN
;
5376 /* Software GSO depends on SG. */
5377 if ((features
& NETIF_F_GSO
) && !(features
& NETIF_F_SG
)) {
5378 netdev_dbg(dev
, "Dropping NETIF_F_GSO since no SG feature.\n");
5379 features
&= ~NETIF_F_GSO
;
5382 /* UFO needs SG and checksumming */
5383 if (features
& NETIF_F_UFO
) {
5384 /* maybe split UFO into V4 and V6? */
5385 if (!((features
& NETIF_F_GEN_CSUM
) ||
5386 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))
5387 == (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5389 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5390 features
&= ~NETIF_F_UFO
;
5393 if (!(features
& NETIF_F_SG
)) {
5395 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5396 features
&= ~NETIF_F_UFO
;
5403 int __netdev_update_features(struct net_device
*dev
)
5405 netdev_features_t features
;
5410 features
= netdev_get_wanted_features(dev
);
5412 if (dev
->netdev_ops
->ndo_fix_features
)
5413 features
= dev
->netdev_ops
->ndo_fix_features(dev
, features
);
5415 /* driver might be less strict about feature dependencies */
5416 features
= netdev_fix_features(dev
, features
);
5418 if (dev
->features
== features
)
5421 netdev_dbg(dev
, "Features changed: %pNF -> %pNF\n",
5422 &dev
->features
, &features
);
5424 if (dev
->netdev_ops
->ndo_set_features
)
5425 err
= dev
->netdev_ops
->ndo_set_features(dev
, features
);
5427 if (unlikely(err
< 0)) {
5429 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5430 err
, &features
, &dev
->features
);
5435 dev
->features
= features
;
5441 * netdev_update_features - recalculate device features
5442 * @dev: the device to check
5444 * Recalculate dev->features set and send notifications if it
5445 * has changed. Should be called after driver or hardware dependent
5446 * conditions might have changed that influence the features.
5448 void netdev_update_features(struct net_device
*dev
)
5450 if (__netdev_update_features(dev
))
5451 netdev_features_change(dev
);
5453 EXPORT_SYMBOL(netdev_update_features
);
5456 * netdev_change_features - recalculate device features
5457 * @dev: the device to check
5459 * Recalculate dev->features set and send notifications even
5460 * if they have not changed. Should be called instead of
5461 * netdev_update_features() if also dev->vlan_features might
5462 * have changed to allow the changes to be propagated to stacked
5465 void netdev_change_features(struct net_device
*dev
)
5467 __netdev_update_features(dev
);
5468 netdev_features_change(dev
);
5470 EXPORT_SYMBOL(netdev_change_features
);
5473 * netif_stacked_transfer_operstate - transfer operstate
5474 * @rootdev: the root or lower level device to transfer state from
5475 * @dev: the device to transfer operstate to
5477 * Transfer operational state from root to device. This is normally
5478 * called when a stacking relationship exists between the root
5479 * device and the device(a leaf device).
5481 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
5482 struct net_device
*dev
)
5484 if (rootdev
->operstate
== IF_OPER_DORMANT
)
5485 netif_dormant_on(dev
);
5487 netif_dormant_off(dev
);
5489 if (netif_carrier_ok(rootdev
)) {
5490 if (!netif_carrier_ok(dev
))
5491 netif_carrier_on(dev
);
5493 if (netif_carrier_ok(dev
))
5494 netif_carrier_off(dev
);
5497 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
5500 static int netif_alloc_rx_queues(struct net_device
*dev
)
5502 unsigned int i
, count
= dev
->num_rx_queues
;
5503 struct netdev_rx_queue
*rx
;
5507 rx
= kcalloc(count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5509 pr_err("netdev: Unable to allocate %u rx queues\n", count
);
5514 for (i
= 0; i
< count
; i
++)
5520 static void netdev_init_one_queue(struct net_device
*dev
,
5521 struct netdev_queue
*queue
, void *_unused
)
5523 /* Initialize queue lock */
5524 spin_lock_init(&queue
->_xmit_lock
);
5525 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
5526 queue
->xmit_lock_owner
= -1;
5527 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
5530 dql_init(&queue
->dql
, HZ
);
5534 static int netif_alloc_netdev_queues(struct net_device
*dev
)
5536 unsigned int count
= dev
->num_tx_queues
;
5537 struct netdev_queue
*tx
;
5541 tx
= kcalloc(count
, sizeof(struct netdev_queue
), GFP_KERNEL
);
5543 pr_err("netdev: Unable to allocate %u tx queues\n", count
);
5548 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5549 spin_lock_init(&dev
->tx_global_lock
);
5555 * register_netdevice - register a network device
5556 * @dev: device to register
5558 * Take a completed network device structure and add it to the kernel
5559 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5560 * chain. 0 is returned on success. A negative errno code is returned
5561 * on a failure to set up the device, or if the name is a duplicate.
5563 * Callers must hold the rtnl semaphore. You may want
5564 * register_netdev() instead of this.
5567 * The locking appears insufficient to guarantee two parallel registers
5568 * will not get the same name.
5571 int register_netdevice(struct net_device
*dev
)
5574 struct net
*net
= dev_net(dev
);
5576 BUG_ON(dev_boot_phase
);
5581 /* When net_device's are persistent, this will be fatal. */
5582 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
5585 spin_lock_init(&dev
->addr_list_lock
);
5586 netdev_set_addr_lockdep_class(dev
);
5590 ret
= dev_get_valid_name(dev
, dev
->name
);
5594 /* Init, if this function is available */
5595 if (dev
->netdev_ops
->ndo_init
) {
5596 ret
= dev
->netdev_ops
->ndo_init(dev
);
5606 dev
->ifindex
= dev_new_index(net
);
5607 else if (__dev_get_by_index(net
, dev
->ifindex
))
5610 if (dev
->iflink
== -1)
5611 dev
->iflink
= dev
->ifindex
;
5613 /* Transfer changeable features to wanted_features and enable
5614 * software offloads (GSO and GRO).
5616 dev
->hw_features
|= NETIF_F_SOFT_FEATURES
;
5617 dev
->features
|= NETIF_F_SOFT_FEATURES
;
5618 dev
->wanted_features
= dev
->features
& dev
->hw_features
;
5620 /* Turn on no cache copy if HW is doing checksum */
5621 if (!(dev
->flags
& IFF_LOOPBACK
)) {
5622 dev
->hw_features
|= NETIF_F_NOCACHE_COPY
;
5623 if (dev
->features
& NETIF_F_ALL_CSUM
) {
5624 dev
->wanted_features
|= NETIF_F_NOCACHE_COPY
;
5625 dev
->features
|= NETIF_F_NOCACHE_COPY
;
5629 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5631 dev
->vlan_features
|= NETIF_F_HIGHDMA
;
5633 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
5634 ret
= notifier_to_errno(ret
);
5638 ret
= netdev_register_kobject(dev
);
5641 dev
->reg_state
= NETREG_REGISTERED
;
5643 __netdev_update_features(dev
);
5646 * Default initial state at registry is that the
5647 * device is present.
5650 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5652 dev_init_scheduler(dev
);
5654 list_netdevice(dev
);
5655 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
5657 /* Notify protocols, that a new device appeared. */
5658 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5659 ret
= notifier_to_errno(ret
);
5661 rollback_registered(dev
);
5662 dev
->reg_state
= NETREG_UNREGISTERED
;
5665 * Prevent userspace races by waiting until the network
5666 * device is fully setup before sending notifications.
5668 if (!dev
->rtnl_link_ops
||
5669 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5670 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5676 if (dev
->netdev_ops
->ndo_uninit
)
5677 dev
->netdev_ops
->ndo_uninit(dev
);
5680 EXPORT_SYMBOL(register_netdevice
);
5683 * init_dummy_netdev - init a dummy network device for NAPI
5684 * @dev: device to init
5686 * This takes a network device structure and initialize the minimum
5687 * amount of fields so it can be used to schedule NAPI polls without
5688 * registering a full blown interface. This is to be used by drivers
5689 * that need to tie several hardware interfaces to a single NAPI
5690 * poll scheduler due to HW limitations.
5692 int init_dummy_netdev(struct net_device
*dev
)
5694 /* Clear everything. Note we don't initialize spinlocks
5695 * are they aren't supposed to be taken by any of the
5696 * NAPI code and this dummy netdev is supposed to be
5697 * only ever used for NAPI polls
5699 memset(dev
, 0, sizeof(struct net_device
));
5701 /* make sure we BUG if trying to hit standard
5702 * register/unregister code path
5704 dev
->reg_state
= NETREG_DUMMY
;
5706 /* NAPI wants this */
5707 INIT_LIST_HEAD(&dev
->napi_list
);
5709 /* a dummy interface is started by default */
5710 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5711 set_bit(__LINK_STATE_START
, &dev
->state
);
5713 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5714 * because users of this 'device' dont need to change
5720 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
5724 * register_netdev - register a network device
5725 * @dev: device to register
5727 * Take a completed network device structure and add it to the kernel
5728 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5729 * chain. 0 is returned on success. A negative errno code is returned
5730 * on a failure to set up the device, or if the name is a duplicate.
5732 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5733 * and expands the device name if you passed a format string to
5736 int register_netdev(struct net_device
*dev
)
5741 err
= register_netdevice(dev
);
5745 EXPORT_SYMBOL(register_netdev
);
5747 int netdev_refcnt_read(const struct net_device
*dev
)
5751 for_each_possible_cpu(i
)
5752 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
5755 EXPORT_SYMBOL(netdev_refcnt_read
);
5758 * netdev_wait_allrefs - wait until all references are gone.
5759 * @dev: target net_device
5761 * This is called when unregistering network devices.
5763 * Any protocol or device that holds a reference should register
5764 * for netdevice notification, and cleanup and put back the
5765 * reference if they receive an UNREGISTER event.
5766 * We can get stuck here if buggy protocols don't correctly
5769 static void netdev_wait_allrefs(struct net_device
*dev
)
5771 unsigned long rebroadcast_time
, warning_time
;
5774 linkwatch_forget_dev(dev
);
5776 rebroadcast_time
= warning_time
= jiffies
;
5777 refcnt
= netdev_refcnt_read(dev
);
5779 while (refcnt
!= 0) {
5780 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
5783 /* Rebroadcast unregister notification */
5784 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5786 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
5787 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
5789 /* We must not have linkwatch events
5790 * pending on unregister. If this
5791 * happens, we simply run the queue
5792 * unscheduled, resulting in a noop
5795 linkwatch_run_queue();
5800 rebroadcast_time
= jiffies
;
5805 refcnt
= netdev_refcnt_read(dev
);
5807 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
5808 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5810 warning_time
= jiffies
;
5819 * register_netdevice(x1);
5820 * register_netdevice(x2);
5822 * unregister_netdevice(y1);
5823 * unregister_netdevice(y2);
5829 * We are invoked by rtnl_unlock().
5830 * This allows us to deal with problems:
5831 * 1) We can delete sysfs objects which invoke hotplug
5832 * without deadlocking with linkwatch via keventd.
5833 * 2) Since we run with the RTNL semaphore not held, we can sleep
5834 * safely in order to wait for the netdev refcnt to drop to zero.
5836 * We must not return until all unregister events added during
5837 * the interval the lock was held have been completed.
5839 void netdev_run_todo(void)
5841 struct list_head list
;
5843 /* Snapshot list, allow later requests */
5844 list_replace_init(&net_todo_list
, &list
);
5849 /* Wait for rcu callbacks to finish before next phase */
5850 if (!list_empty(&list
))
5853 while (!list_empty(&list
)) {
5854 struct net_device
*dev
5855 = list_first_entry(&list
, struct net_device
, todo_list
);
5856 list_del(&dev
->todo_list
);
5858 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
5860 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
5861 pr_err("network todo '%s' but state %d\n",
5862 dev
->name
, dev
->reg_state
);
5867 dev
->reg_state
= NETREG_UNREGISTERED
;
5869 on_each_cpu(flush_backlog
, dev
, 1);
5871 netdev_wait_allrefs(dev
);
5874 BUG_ON(netdev_refcnt_read(dev
));
5875 WARN_ON(rcu_access_pointer(dev
->ip_ptr
));
5876 WARN_ON(rcu_access_pointer(dev
->ip6_ptr
));
5877 WARN_ON(dev
->dn_ptr
);
5879 if (dev
->destructor
)
5880 dev
->destructor(dev
);
5882 /* Free network device */
5883 kobject_put(&dev
->dev
.kobj
);
5887 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5888 * fields in the same order, with only the type differing.
5890 void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
5891 const struct net_device_stats
*netdev_stats
)
5893 #if BITS_PER_LONG == 64
5894 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
5895 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
5897 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
5898 const unsigned long *src
= (const unsigned long *)netdev_stats
;
5899 u64
*dst
= (u64
*)stats64
;
5901 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
5902 sizeof(*stats64
) / sizeof(u64
));
5903 for (i
= 0; i
< n
; i
++)
5907 EXPORT_SYMBOL(netdev_stats_to_stats64
);
5910 * dev_get_stats - get network device statistics
5911 * @dev: device to get statistics from
5912 * @storage: place to store stats
5914 * Get network statistics from device. Return @storage.
5915 * The device driver may provide its own method by setting
5916 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5917 * otherwise the internal statistics structure is used.
5919 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
5920 struct rtnl_link_stats64
*storage
)
5922 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5924 if (ops
->ndo_get_stats64
) {
5925 memset(storage
, 0, sizeof(*storage
));
5926 ops
->ndo_get_stats64(dev
, storage
);
5927 } else if (ops
->ndo_get_stats
) {
5928 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
5930 netdev_stats_to_stats64(storage
, &dev
->stats
);
5932 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
5935 EXPORT_SYMBOL(dev_get_stats
);
5937 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
5939 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
5941 #ifdef CONFIG_NET_CLS_ACT
5944 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
5947 netdev_init_one_queue(dev
, queue
, NULL
);
5948 queue
->qdisc
= &noop_qdisc
;
5949 queue
->qdisc_sleeping
= &noop_qdisc
;
5950 rcu_assign_pointer(dev
->ingress_queue
, queue
);
5956 * alloc_netdev_mqs - allocate network device
5957 * @sizeof_priv: size of private data to allocate space for
5958 * @name: device name format string
5959 * @setup: callback to initialize device
5960 * @txqs: the number of TX subqueues to allocate
5961 * @rxqs: the number of RX subqueues to allocate
5963 * Allocates a struct net_device with private data area for driver use
5964 * and performs basic initialization. Also allocates subquue structs
5965 * for each queue on the device.
5967 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
5968 void (*setup
)(struct net_device
*),
5969 unsigned int txqs
, unsigned int rxqs
)
5971 struct net_device
*dev
;
5973 struct net_device
*p
;
5975 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
5978 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
5984 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
5989 alloc_size
= sizeof(struct net_device
);
5991 /* ensure 32-byte alignment of private area */
5992 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
5993 alloc_size
+= sizeof_priv
;
5995 /* ensure 32-byte alignment of whole construct */
5996 alloc_size
+= NETDEV_ALIGN
- 1;
5998 p
= kzalloc(alloc_size
, GFP_KERNEL
);
6000 pr_err("alloc_netdev: Unable to allocate device\n");
6004 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
6005 dev
->padded
= (char *)dev
- (char *)p
;
6007 dev
->pcpu_refcnt
= alloc_percpu(int);
6008 if (!dev
->pcpu_refcnt
)
6011 if (dev_addr_init(dev
))
6017 dev_net_set(dev
, &init_net
);
6019 dev
->gso_max_size
= GSO_MAX_SIZE
;
6020 dev
->gso_max_segs
= GSO_MAX_SEGS
;
6022 INIT_LIST_HEAD(&dev
->napi_list
);
6023 INIT_LIST_HEAD(&dev
->unreg_list
);
6024 INIT_LIST_HEAD(&dev
->link_watch_list
);
6025 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
6028 dev
->num_tx_queues
= txqs
;
6029 dev
->real_num_tx_queues
= txqs
;
6030 if (netif_alloc_netdev_queues(dev
))
6034 dev
->num_rx_queues
= rxqs
;
6035 dev
->real_num_rx_queues
= rxqs
;
6036 if (netif_alloc_rx_queues(dev
))
6040 strcpy(dev
->name
, name
);
6041 dev
->group
= INIT_NETDEV_GROUP
;
6049 free_percpu(dev
->pcpu_refcnt
);
6059 EXPORT_SYMBOL(alloc_netdev_mqs
);
6062 * free_netdev - free network device
6065 * This function does the last stage of destroying an allocated device
6066 * interface. The reference to the device object is released.
6067 * If this is the last reference then it will be freed.
6069 void free_netdev(struct net_device
*dev
)
6071 struct napi_struct
*p
, *n
;
6073 release_net(dev_net(dev
));
6080 kfree(rcu_dereference_protected(dev
->ingress_queue
, 1));
6082 /* Flush device addresses */
6083 dev_addr_flush(dev
);
6085 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
6088 free_percpu(dev
->pcpu_refcnt
);
6089 dev
->pcpu_refcnt
= NULL
;
6091 /* Compatibility with error handling in drivers */
6092 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
6093 kfree((char *)dev
- dev
->padded
);
6097 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
6098 dev
->reg_state
= NETREG_RELEASED
;
6100 /* will free via device release */
6101 put_device(&dev
->dev
);
6103 EXPORT_SYMBOL(free_netdev
);
6106 * synchronize_net - Synchronize with packet receive processing
6108 * Wait for packets currently being received to be done.
6109 * Does not block later packets from starting.
6111 void synchronize_net(void)
6114 if (rtnl_is_locked())
6115 synchronize_rcu_expedited();
6119 EXPORT_SYMBOL(synchronize_net
);
6122 * unregister_netdevice_queue - remove device from the kernel
6126 * This function shuts down a device interface and removes it
6127 * from the kernel tables.
6128 * If head not NULL, device is queued to be unregistered later.
6130 * Callers must hold the rtnl semaphore. You may want
6131 * unregister_netdev() instead of this.
6134 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
6139 list_move_tail(&dev
->unreg_list
, head
);
6141 rollback_registered(dev
);
6142 /* Finish processing unregister after unlock */
6146 EXPORT_SYMBOL(unregister_netdevice_queue
);
6149 * unregister_netdevice_many - unregister many devices
6150 * @head: list of devices
6152 void unregister_netdevice_many(struct list_head
*head
)
6154 struct net_device
*dev
;
6156 if (!list_empty(head
)) {
6157 rollback_registered_many(head
);
6158 list_for_each_entry(dev
, head
, unreg_list
)
6162 EXPORT_SYMBOL(unregister_netdevice_many
);
6165 * unregister_netdev - remove device from the kernel
6168 * This function shuts down a device interface and removes it
6169 * from the kernel tables.
6171 * This is just a wrapper for unregister_netdevice that takes
6172 * the rtnl semaphore. In general you want to use this and not
6173 * unregister_netdevice.
6175 void unregister_netdev(struct net_device
*dev
)
6178 unregister_netdevice(dev
);
6181 EXPORT_SYMBOL(unregister_netdev
);
6184 * dev_change_net_namespace - move device to different nethost namespace
6186 * @net: network namespace
6187 * @pat: If not NULL name pattern to try if the current device name
6188 * is already taken in the destination network namespace.
6190 * This function shuts down a device interface and moves it
6191 * to a new network namespace. On success 0 is returned, on
6192 * a failure a netagive errno code is returned.
6194 * Callers must hold the rtnl semaphore.
6197 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
6203 /* Don't allow namespace local devices to be moved. */
6205 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6208 /* Ensure the device has been registrered */
6210 if (dev
->reg_state
!= NETREG_REGISTERED
)
6213 /* Get out if there is nothing todo */
6215 if (net_eq(dev_net(dev
), net
))
6218 /* Pick the destination device name, and ensure
6219 * we can use it in the destination network namespace.
6222 if (__dev_get_by_name(net
, dev
->name
)) {
6223 /* We get here if we can't use the current device name */
6226 if (dev_get_valid_name(dev
, pat
) < 0)
6231 * And now a mini version of register_netdevice unregister_netdevice.
6234 /* If device is running close it first. */
6237 /* And unlink it from device chain */
6239 unlist_netdevice(dev
);
6243 /* Shutdown queueing discipline. */
6246 /* Notify protocols, that we are about to destroy
6247 this device. They should clean all the things.
6249 Note that dev->reg_state stays at NETREG_REGISTERED.
6250 This is wanted because this way 8021q and macvlan know
6251 the device is just moving and can keep their slaves up.
6253 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6254 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
6257 * Flush the unicast and multicast chains
6262 /* Actually switch the network namespace */
6263 dev_net_set(dev
, net
);
6265 /* If there is an ifindex conflict assign a new one */
6266 if (__dev_get_by_index(net
, dev
->ifindex
)) {
6267 int iflink
= (dev
->iflink
== dev
->ifindex
);
6268 dev
->ifindex
= dev_new_index(net
);
6270 dev
->iflink
= dev
->ifindex
;
6273 /* Fixup kobjects */
6274 err
= device_rename(&dev
->dev
, dev
->name
);
6277 /* Add the device back in the hashes */
6278 list_netdevice(dev
);
6280 /* Notify protocols, that a new device appeared. */
6281 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
6284 * Prevent userspace races by waiting until the network
6285 * device is fully setup before sending notifications.
6287 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
6294 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
6296 static int dev_cpu_callback(struct notifier_block
*nfb
,
6297 unsigned long action
,
6300 struct sk_buff
**list_skb
;
6301 struct sk_buff
*skb
;
6302 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
6303 struct softnet_data
*sd
, *oldsd
;
6305 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
6308 local_irq_disable();
6309 cpu
= smp_processor_id();
6310 sd
= &per_cpu(softnet_data
, cpu
);
6311 oldsd
= &per_cpu(softnet_data
, oldcpu
);
6313 /* Find end of our completion_queue. */
6314 list_skb
= &sd
->completion_queue
;
6316 list_skb
= &(*list_skb
)->next
;
6317 /* Append completion queue from offline CPU. */
6318 *list_skb
= oldsd
->completion_queue
;
6319 oldsd
->completion_queue
= NULL
;
6321 /* Append output queue from offline CPU. */
6322 if (oldsd
->output_queue
) {
6323 *sd
->output_queue_tailp
= oldsd
->output_queue
;
6324 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
6325 oldsd
->output_queue
= NULL
;
6326 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
6328 /* Append NAPI poll list from offline CPU. */
6329 if (!list_empty(&oldsd
->poll_list
)) {
6330 list_splice_init(&oldsd
->poll_list
, &sd
->poll_list
);
6331 raise_softirq_irqoff(NET_RX_SOFTIRQ
);
6334 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
6337 /* Process offline CPU's input_pkt_queue */
6338 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
6340 input_queue_head_incr(oldsd
);
6342 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
6344 input_queue_head_incr(oldsd
);
6352 * netdev_increment_features - increment feature set by one
6353 * @all: current feature set
6354 * @one: new feature set
6355 * @mask: mask feature set
6357 * Computes a new feature set after adding a device with feature set
6358 * @one to the master device with current feature set @all. Will not
6359 * enable anything that is off in @mask. Returns the new feature set.
6361 netdev_features_t
netdev_increment_features(netdev_features_t all
,
6362 netdev_features_t one
, netdev_features_t mask
)
6364 if (mask
& NETIF_F_GEN_CSUM
)
6365 mask
|= NETIF_F_ALL_CSUM
;
6366 mask
|= NETIF_F_VLAN_CHALLENGED
;
6368 all
|= one
& (NETIF_F_ONE_FOR_ALL
|NETIF_F_ALL_CSUM
) & mask
;
6369 all
&= one
| ~NETIF_F_ALL_FOR_ALL
;
6371 /* If one device supports hw checksumming, set for all. */
6372 if (all
& NETIF_F_GEN_CSUM
)
6373 all
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_GEN_CSUM
);
6377 EXPORT_SYMBOL(netdev_increment_features
);
6379 static struct hlist_head
*netdev_create_hash(void)
6382 struct hlist_head
*hash
;
6384 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
6386 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
6387 INIT_HLIST_HEAD(&hash
[i
]);
6392 /* Initialize per network namespace state */
6393 static int __net_init
netdev_init(struct net
*net
)
6395 if (net
!= &init_net
)
6396 INIT_LIST_HEAD(&net
->dev_base_head
);
6398 net
->dev_name_head
= netdev_create_hash();
6399 if (net
->dev_name_head
== NULL
)
6402 net
->dev_index_head
= netdev_create_hash();
6403 if (net
->dev_index_head
== NULL
)
6409 kfree(net
->dev_name_head
);
6415 * netdev_drivername - network driver for the device
6416 * @dev: network device
6418 * Determine network driver for device.
6420 const char *netdev_drivername(const struct net_device
*dev
)
6422 const struct device_driver
*driver
;
6423 const struct device
*parent
;
6424 const char *empty
= "";
6426 parent
= dev
->dev
.parent
;
6430 driver
= parent
->driver
;
6431 if (driver
&& driver
->name
)
6432 return driver
->name
;
6436 int __netdev_printk(const char *level
, const struct net_device
*dev
,
6437 struct va_format
*vaf
)
6441 if (dev
&& dev
->dev
.parent
)
6442 r
= dev_printk(level
, dev
->dev
.parent
, "%s: %pV",
6443 netdev_name(dev
), vaf
);
6445 r
= printk("%s%s: %pV", level
, netdev_name(dev
), vaf
);
6447 r
= printk("%s(NULL net_device): %pV", level
, vaf
);
6451 EXPORT_SYMBOL(__netdev_printk
);
6453 int netdev_printk(const char *level
, const struct net_device
*dev
,
6454 const char *format
, ...)
6456 struct va_format vaf
;
6460 va_start(args
, format
);
6465 r
= __netdev_printk(level
, dev
, &vaf
);
6470 EXPORT_SYMBOL(netdev_printk
);
6472 #define define_netdev_printk_level(func, level) \
6473 int func(const struct net_device *dev, const char *fmt, ...) \
6476 struct va_format vaf; \
6479 va_start(args, fmt); \
6484 r = __netdev_printk(level, dev, &vaf); \
6489 EXPORT_SYMBOL(func);
6491 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
6492 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
6493 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
6494 define_netdev_printk_level(netdev_err
, KERN_ERR
);
6495 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
6496 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
6497 define_netdev_printk_level(netdev_info
, KERN_INFO
);
6499 static void __net_exit
netdev_exit(struct net
*net
)
6501 kfree(net
->dev_name_head
);
6502 kfree(net
->dev_index_head
);
6505 static struct pernet_operations __net_initdata netdev_net_ops
= {
6506 .init
= netdev_init
,
6507 .exit
= netdev_exit
,
6510 static void __net_exit
default_device_exit(struct net
*net
)
6512 struct net_device
*dev
, *aux
;
6514 * Push all migratable network devices back to the
6515 * initial network namespace
6518 for_each_netdev_safe(net
, dev
, aux
) {
6520 char fb_name
[IFNAMSIZ
];
6522 /* Ignore unmoveable devices (i.e. loopback) */
6523 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6526 /* Leave virtual devices for the generic cleanup */
6527 if (dev
->rtnl_link_ops
)
6530 /* Push remaining network devices to init_net */
6531 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
6532 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
6534 pr_emerg("%s: failed to move %s to init_net: %d\n",
6535 __func__
, dev
->name
, err
);
6542 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
6544 /* At exit all network devices most be removed from a network
6545 * namespace. Do this in the reverse order of registration.
6546 * Do this across as many network namespaces as possible to
6547 * improve batching efficiency.
6549 struct net_device
*dev
;
6551 LIST_HEAD(dev_kill_list
);
6554 list_for_each_entry(net
, net_list
, exit_list
) {
6555 for_each_netdev_reverse(net
, dev
) {
6556 if (dev
->rtnl_link_ops
)
6557 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
6559 unregister_netdevice_queue(dev
, &dev_kill_list
);
6562 unregister_netdevice_many(&dev_kill_list
);
6563 list_del(&dev_kill_list
);
6567 static struct pernet_operations __net_initdata default_device_ops
= {
6568 .exit
= default_device_exit
,
6569 .exit_batch
= default_device_exit_batch
,
6573 * Initialize the DEV module. At boot time this walks the device list and
6574 * unhooks any devices that fail to initialise (normally hardware not
6575 * present) and leaves us with a valid list of present and active devices.
6580 * This is called single threaded during boot, so no need
6581 * to take the rtnl semaphore.
6583 static int __init
net_dev_init(void)
6585 int i
, rc
= -ENOMEM
;
6587 BUG_ON(!dev_boot_phase
);
6589 if (dev_proc_init())
6592 if (netdev_kobject_init())
6595 INIT_LIST_HEAD(&ptype_all
);
6596 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
6597 INIT_LIST_HEAD(&ptype_base
[i
]);
6599 if (register_pernet_subsys(&netdev_net_ops
))
6603 * Initialise the packet receive queues.
6606 for_each_possible_cpu(i
) {
6607 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
6609 memset(sd
, 0, sizeof(*sd
));
6610 skb_queue_head_init(&sd
->input_pkt_queue
);
6611 skb_queue_head_init(&sd
->process_queue
);
6612 sd
->completion_queue
= NULL
;
6613 INIT_LIST_HEAD(&sd
->poll_list
);
6614 sd
->output_queue
= NULL
;
6615 sd
->output_queue_tailp
= &sd
->output_queue
;
6617 sd
->csd
.func
= rps_trigger_softirq
;
6623 sd
->backlog
.poll
= process_backlog
;
6624 sd
->backlog
.weight
= weight_p
;
6625 sd
->backlog
.gro_list
= NULL
;
6626 sd
->backlog
.gro_count
= 0;
6631 /* The loopback device is special if any other network devices
6632 * is present in a network namespace the loopback device must
6633 * be present. Since we now dynamically allocate and free the
6634 * loopback device ensure this invariant is maintained by
6635 * keeping the loopback device as the first device on the
6636 * list of network devices. Ensuring the loopback devices
6637 * is the first device that appears and the last network device
6640 if (register_pernet_device(&loopback_net_ops
))
6643 if (register_pernet_device(&default_device_ops
))
6646 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
6647 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
6649 hotcpu_notifier(dev_cpu_callback
, 0);
6657 subsys_initcall(net_dev_init
);
6659 static int __init
initialize_hashrnd(void)
6661 get_random_bytes(&hashrnd
, sizeof(hashrnd
));
6665 late_initcall_sync(initialize_hashrnd
);