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/stat.h>
102 #include <net/pkt_sched.h>
103 #include <net/checksum.h>
104 #include <net/xfrm.h>
105 #include <linux/highmem.h>
106 #include <linux/init.h>
107 #include <linux/module.h>
108 #include <linux/netpoll.h>
109 #include <linux/rcupdate.h>
110 #include <linux/delay.h>
111 #include <net/iw_handler.h>
112 #include <asm/current.h>
113 #include <linux/audit.h>
114 #include <linux/dmaengine.h>
115 #include <linux/err.h>
116 #include <linux/ctype.h>
117 #include <linux/if_arp.h>
118 #include <linux/if_vlan.h>
119 #include <linux/ip.h>
121 #include <linux/ipv6.h>
122 #include <linux/in.h>
123 #include <linux/jhash.h>
124 #include <linux/random.h>
125 #include <trace/events/napi.h>
126 #include <trace/events/net.h>
127 #include <trace/events/skb.h>
128 #include <linux/pci.h>
129 #include <linux/inetdevice.h>
130 #include <linux/cpu_rmap.h>
131 #include <linux/static_key.h>
133 #include "net-sysfs.h"
135 /* Instead of increasing this, you should create a hash table. */
136 #define MAX_GRO_SKBS 8
138 /* This should be increased if a protocol with a bigger head is added. */
139 #define GRO_MAX_HEAD (MAX_HEADER + 128)
141 static DEFINE_SPINLOCK(ptype_lock
);
142 static DEFINE_SPINLOCK(offload_lock
);
143 struct list_head ptype_base
[PTYPE_HASH_SIZE
] __read_mostly
;
144 struct list_head ptype_all __read_mostly
; /* Taps */
145 static struct list_head offload_base __read_mostly
;
148 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
151 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
153 * Writers must hold the rtnl semaphore while they loop through the
154 * dev_base_head list, and hold dev_base_lock for writing when they do the
155 * actual updates. This allows pure readers to access the list even
156 * while a writer is preparing to update it.
158 * To put it another way, dev_base_lock is held for writing only to
159 * protect against pure readers; the rtnl semaphore provides the
160 * protection against other writers.
162 * See, for example usages, register_netdevice() and
163 * unregister_netdevice(), which must be called with the rtnl
166 DEFINE_RWLOCK(dev_base_lock
);
167 EXPORT_SYMBOL(dev_base_lock
);
169 seqcount_t devnet_rename_seq
;
171 static inline void dev_base_seq_inc(struct net
*net
)
173 while (++net
->dev_base_seq
== 0);
176 static inline struct hlist_head
*dev_name_hash(struct net
*net
, const char *name
)
178 unsigned int hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
180 return &net
->dev_name_head
[hash_32(hash
, NETDEV_HASHBITS
)];
183 static inline struct hlist_head
*dev_index_hash(struct net
*net
, int ifindex
)
185 return &net
->dev_index_head
[ifindex
& (NETDEV_HASHENTRIES
- 1)];
188 static inline void rps_lock(struct softnet_data
*sd
)
191 spin_lock(&sd
->input_pkt_queue
.lock
);
195 static inline void rps_unlock(struct softnet_data
*sd
)
198 spin_unlock(&sd
->input_pkt_queue
.lock
);
202 /* Device list insertion */
203 static int list_netdevice(struct net_device
*dev
)
205 struct net
*net
= dev_net(dev
);
209 write_lock_bh(&dev_base_lock
);
210 list_add_tail_rcu(&dev
->dev_list
, &net
->dev_base_head
);
211 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
212 hlist_add_head_rcu(&dev
->index_hlist
,
213 dev_index_hash(net
, dev
->ifindex
));
214 write_unlock_bh(&dev_base_lock
);
216 dev_base_seq_inc(net
);
221 /* Device list removal
222 * caller must respect a RCU grace period before freeing/reusing dev
224 static void unlist_netdevice(struct net_device
*dev
)
228 /* Unlink dev from the device chain */
229 write_lock_bh(&dev_base_lock
);
230 list_del_rcu(&dev
->dev_list
);
231 hlist_del_rcu(&dev
->name_hlist
);
232 hlist_del_rcu(&dev
->index_hlist
);
233 write_unlock_bh(&dev_base_lock
);
235 dev_base_seq_inc(dev_net(dev
));
242 static RAW_NOTIFIER_HEAD(netdev_chain
);
245 * Device drivers call our routines to queue packets here. We empty the
246 * queue in the local softnet handler.
249 DEFINE_PER_CPU_ALIGNED(struct softnet_data
, softnet_data
);
250 EXPORT_PER_CPU_SYMBOL(softnet_data
);
252 #ifdef CONFIG_LOCKDEP
254 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
255 * according to dev->type
257 static const unsigned short netdev_lock_type
[] =
258 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
259 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
260 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
261 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
262 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
263 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
264 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
265 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
266 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
267 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
268 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
269 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
270 ARPHRD_FCFABRIC
, ARPHRD_IEEE80211
, ARPHRD_IEEE80211_PRISM
,
271 ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_PHONET
, ARPHRD_PHONET_PIPE
,
272 ARPHRD_IEEE802154
, ARPHRD_VOID
, ARPHRD_NONE
};
274 static const char *const netdev_lock_name
[] =
275 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
276 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
277 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
278 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
279 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
280 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
281 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
282 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
283 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
284 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
285 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
286 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
287 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
288 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
289 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
291 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
292 static struct lock_class_key netdev_addr_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
294 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
298 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
299 if (netdev_lock_type
[i
] == dev_type
)
301 /* the last key is used by default */
302 return ARRAY_SIZE(netdev_lock_type
) - 1;
305 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
306 unsigned short dev_type
)
310 i
= netdev_lock_pos(dev_type
);
311 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
312 netdev_lock_name
[i
]);
315 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
319 i
= netdev_lock_pos(dev
->type
);
320 lockdep_set_class_and_name(&dev
->addr_list_lock
,
321 &netdev_addr_lock_key
[i
],
322 netdev_lock_name
[i
]);
325 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
326 unsigned short dev_type
)
329 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
334 /*******************************************************************************
336 Protocol management and registration routines
338 *******************************************************************************/
341 * Add a protocol ID to the list. Now that the input handler is
342 * smarter we can dispense with all the messy stuff that used to be
345 * BEWARE!!! Protocol handlers, mangling input packets,
346 * MUST BE last in hash buckets and checking protocol handlers
347 * MUST start from promiscuous ptype_all chain in net_bh.
348 * It is true now, do not change it.
349 * Explanation follows: if protocol handler, mangling packet, will
350 * be the first on list, it is not able to sense, that packet
351 * is cloned and should be copied-on-write, so that it will
352 * change it and subsequent readers will get broken packet.
356 static inline struct list_head
*ptype_head(const struct packet_type
*pt
)
358 if (pt
->type
== htons(ETH_P_ALL
))
361 return &ptype_base
[ntohs(pt
->type
) & PTYPE_HASH_MASK
];
365 * dev_add_pack - add packet handler
366 * @pt: packet type declaration
368 * Add a protocol handler to the networking stack. The passed &packet_type
369 * is linked into kernel lists and may not be freed until it has been
370 * removed from the kernel lists.
372 * This call does not sleep therefore it can not
373 * guarantee all CPU's that are in middle of receiving packets
374 * will see the new packet type (until the next received packet).
377 void dev_add_pack(struct packet_type
*pt
)
379 struct list_head
*head
= ptype_head(pt
);
381 spin_lock(&ptype_lock
);
382 list_add_rcu(&pt
->list
, head
);
383 spin_unlock(&ptype_lock
);
385 EXPORT_SYMBOL(dev_add_pack
);
388 * __dev_remove_pack - remove packet handler
389 * @pt: packet type declaration
391 * Remove a protocol handler that was previously added to the kernel
392 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
393 * from the kernel lists and can be freed or reused once this function
396 * The packet type might still be in use by receivers
397 * and must not be freed until after all the CPU's have gone
398 * through a quiescent state.
400 void __dev_remove_pack(struct packet_type
*pt
)
402 struct list_head
*head
= ptype_head(pt
);
403 struct packet_type
*pt1
;
405 spin_lock(&ptype_lock
);
407 list_for_each_entry(pt1
, head
, list
) {
409 list_del_rcu(&pt
->list
);
414 pr_warn("dev_remove_pack: %p not found\n", pt
);
416 spin_unlock(&ptype_lock
);
418 EXPORT_SYMBOL(__dev_remove_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 * This call sleeps to guarantee that no CPU is looking at the packet
432 void dev_remove_pack(struct packet_type
*pt
)
434 __dev_remove_pack(pt
);
438 EXPORT_SYMBOL(dev_remove_pack
);
442 * dev_add_offload - register offload handlers
443 * @po: protocol offload declaration
445 * Add protocol offload handlers to the networking stack. The passed
446 * &proto_offload is linked into kernel lists and may not be freed until
447 * it has been removed from the kernel lists.
449 * This call does not sleep therefore it can not
450 * guarantee all CPU's that are in middle of receiving packets
451 * will see the new offload handlers (until the next received packet).
453 void dev_add_offload(struct packet_offload
*po
)
455 struct list_head
*head
= &offload_base
;
457 spin_lock(&offload_lock
);
458 list_add_rcu(&po
->list
, head
);
459 spin_unlock(&offload_lock
);
461 EXPORT_SYMBOL(dev_add_offload
);
464 * __dev_remove_offload - remove offload handler
465 * @po: packet offload declaration
467 * Remove a protocol offload handler that was previously added to the
468 * kernel offload handlers by dev_add_offload(). The passed &offload_type
469 * is removed from the kernel lists and can be freed or reused once this
472 * The packet type might still be in use by receivers
473 * and must not be freed until after all the CPU's have gone
474 * through a quiescent state.
476 void __dev_remove_offload(struct packet_offload
*po
)
478 struct list_head
*head
= &offload_base
;
479 struct packet_offload
*po1
;
481 spin_lock(&offload_lock
);
483 list_for_each_entry(po1
, head
, list
) {
485 list_del_rcu(&po
->list
);
490 pr_warn("dev_remove_offload: %p not found\n", po
);
492 spin_unlock(&offload_lock
);
494 EXPORT_SYMBOL(__dev_remove_offload
);
497 * dev_remove_offload - remove packet offload handler
498 * @po: packet offload declaration
500 * Remove a packet offload handler that was previously added to the kernel
501 * offload handlers by dev_add_offload(). The passed &offload_type is
502 * removed from the kernel lists and can be freed or reused once this
505 * This call sleeps to guarantee that no CPU is looking at the packet
508 void dev_remove_offload(struct packet_offload
*po
)
510 __dev_remove_offload(po
);
514 EXPORT_SYMBOL(dev_remove_offload
);
516 /******************************************************************************
518 Device Boot-time Settings Routines
520 *******************************************************************************/
522 /* Boot time configuration table */
523 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
526 * netdev_boot_setup_add - add new setup entry
527 * @name: name of the device
528 * @map: configured settings for the device
530 * Adds new setup entry to the dev_boot_setup list. The function
531 * returns 0 on error and 1 on success. This is a generic routine to
534 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
536 struct netdev_boot_setup
*s
;
540 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
541 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
542 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
543 strlcpy(s
[i
].name
, name
, IFNAMSIZ
);
544 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
549 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
553 * netdev_boot_setup_check - check boot time settings
554 * @dev: the netdevice
556 * Check boot time settings for the device.
557 * The found settings are set for the device to be used
558 * later in the device probing.
559 * Returns 0 if no settings found, 1 if they are.
561 int netdev_boot_setup_check(struct net_device
*dev
)
563 struct netdev_boot_setup
*s
= dev_boot_setup
;
566 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
567 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
568 !strcmp(dev
->name
, s
[i
].name
)) {
569 dev
->irq
= s
[i
].map
.irq
;
570 dev
->base_addr
= s
[i
].map
.base_addr
;
571 dev
->mem_start
= s
[i
].map
.mem_start
;
572 dev
->mem_end
= s
[i
].map
.mem_end
;
578 EXPORT_SYMBOL(netdev_boot_setup_check
);
582 * netdev_boot_base - get address from boot time settings
583 * @prefix: prefix for network device
584 * @unit: id for network device
586 * Check boot time settings for the base address of device.
587 * The found settings are set for the device to be used
588 * later in the device probing.
589 * Returns 0 if no settings found.
591 unsigned long netdev_boot_base(const char *prefix
, int unit
)
593 const struct netdev_boot_setup
*s
= dev_boot_setup
;
597 sprintf(name
, "%s%d", prefix
, unit
);
600 * If device already registered then return base of 1
601 * to indicate not to probe for this interface
603 if (__dev_get_by_name(&init_net
, name
))
606 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
607 if (!strcmp(name
, s
[i
].name
))
608 return s
[i
].map
.base_addr
;
613 * Saves at boot time configured settings for any netdevice.
615 int __init
netdev_boot_setup(char *str
)
620 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
625 memset(&map
, 0, sizeof(map
));
629 map
.base_addr
= ints
[2];
631 map
.mem_start
= ints
[3];
633 map
.mem_end
= ints
[4];
635 /* Add new entry to the list */
636 return netdev_boot_setup_add(str
, &map
);
639 __setup("netdev=", netdev_boot_setup
);
641 /*******************************************************************************
643 Device Interface Subroutines
645 *******************************************************************************/
648 * __dev_get_by_name - find a device by its name
649 * @net: the applicable net namespace
650 * @name: name to find
652 * Find an interface by name. Must be called under RTNL semaphore
653 * or @dev_base_lock. If the name is found a pointer to the device
654 * is returned. If the name is not found then %NULL is returned. The
655 * reference counters are not incremented so the caller must be
656 * careful with locks.
659 struct net_device
*__dev_get_by_name(struct net
*net
, const char *name
)
661 struct net_device
*dev
;
662 struct hlist_head
*head
= dev_name_hash(net
, name
);
664 hlist_for_each_entry(dev
, head
, name_hlist
)
665 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
670 EXPORT_SYMBOL(__dev_get_by_name
);
673 * dev_get_by_name_rcu - find a device by its name
674 * @net: the applicable net namespace
675 * @name: name to find
677 * Find an interface by name.
678 * If the name is found a pointer to the device is returned.
679 * If the name is not found then %NULL is returned.
680 * The reference counters are not incremented so the caller must be
681 * careful with locks. The caller must hold RCU lock.
684 struct net_device
*dev_get_by_name_rcu(struct net
*net
, const char *name
)
686 struct net_device
*dev
;
687 struct hlist_head
*head
= dev_name_hash(net
, name
);
689 hlist_for_each_entry_rcu(dev
, head
, name_hlist
)
690 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
695 EXPORT_SYMBOL(dev_get_by_name_rcu
);
698 * dev_get_by_name - find a device by its name
699 * @net: the applicable net namespace
700 * @name: name to find
702 * Find an interface by name. This can be called from any
703 * context and does its own locking. The returned handle has
704 * the usage count incremented and the caller must use dev_put() to
705 * release it when it is no longer needed. %NULL is returned if no
706 * matching device is found.
709 struct net_device
*dev_get_by_name(struct net
*net
, const char *name
)
711 struct net_device
*dev
;
714 dev
= dev_get_by_name_rcu(net
, name
);
720 EXPORT_SYMBOL(dev_get_by_name
);
723 * __dev_get_by_index - find a device by its ifindex
724 * @net: the applicable net namespace
725 * @ifindex: index of device
727 * Search for an interface by index. Returns %NULL if the device
728 * is not found or a pointer to the device. The device has not
729 * had its reference counter increased so the caller must be careful
730 * about locking. The caller must hold either the RTNL semaphore
734 struct net_device
*__dev_get_by_index(struct net
*net
, int ifindex
)
736 struct net_device
*dev
;
737 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
739 hlist_for_each_entry(dev
, head
, index_hlist
)
740 if (dev
->ifindex
== ifindex
)
745 EXPORT_SYMBOL(__dev_get_by_index
);
748 * dev_get_by_index_rcu - find a device by its ifindex
749 * @net: the applicable net namespace
750 * @ifindex: index of device
752 * Search for an interface by index. Returns %NULL if the device
753 * is not found or a pointer to the device. The device has not
754 * had its reference counter increased so the caller must be careful
755 * about locking. The caller must hold RCU lock.
758 struct net_device
*dev_get_by_index_rcu(struct net
*net
, int ifindex
)
760 struct net_device
*dev
;
761 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
763 hlist_for_each_entry_rcu(dev
, head
, index_hlist
)
764 if (dev
->ifindex
== ifindex
)
769 EXPORT_SYMBOL(dev_get_by_index_rcu
);
773 * dev_get_by_index - find a device by its ifindex
774 * @net: the applicable net namespace
775 * @ifindex: index of device
777 * Search for an interface by index. Returns NULL if the device
778 * is not found or a pointer to the device. The device returned has
779 * had a reference added and the pointer is safe until the user calls
780 * dev_put to indicate they have finished with it.
783 struct net_device
*dev_get_by_index(struct net
*net
, int ifindex
)
785 struct net_device
*dev
;
788 dev
= dev_get_by_index_rcu(net
, ifindex
);
794 EXPORT_SYMBOL(dev_get_by_index
);
797 * dev_getbyhwaddr_rcu - find a device by its hardware address
798 * @net: the applicable net namespace
799 * @type: media type of device
800 * @ha: hardware address
802 * Search for an interface by MAC address. Returns NULL if the device
803 * is not found or a pointer to the device.
804 * The caller must hold RCU or RTNL.
805 * The returned device has not had its ref count increased
806 * and the caller must therefore be careful about locking
810 struct net_device
*dev_getbyhwaddr_rcu(struct net
*net
, unsigned short type
,
813 struct net_device
*dev
;
815 for_each_netdev_rcu(net
, dev
)
816 if (dev
->type
== type
&&
817 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
822 EXPORT_SYMBOL(dev_getbyhwaddr_rcu
);
824 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
826 struct net_device
*dev
;
829 for_each_netdev(net
, dev
)
830 if (dev
->type
== type
)
835 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
837 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
839 struct net_device
*dev
, *ret
= NULL
;
842 for_each_netdev_rcu(net
, dev
)
843 if (dev
->type
== type
) {
851 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
854 * dev_get_by_flags_rcu - find any device with given flags
855 * @net: the applicable net namespace
856 * @if_flags: IFF_* values
857 * @mask: bitmask of bits in if_flags to check
859 * Search for any interface with the given flags. Returns NULL if a device
860 * is not found or a pointer to the device. Must be called inside
861 * rcu_read_lock(), and result refcount is unchanged.
864 struct net_device
*dev_get_by_flags_rcu(struct net
*net
, unsigned short if_flags
,
867 struct net_device
*dev
, *ret
;
870 for_each_netdev_rcu(net
, dev
) {
871 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
878 EXPORT_SYMBOL(dev_get_by_flags_rcu
);
881 * dev_valid_name - check if name is okay for network device
884 * Network device names need to be valid file names to
885 * to allow sysfs to work. We also disallow any kind of
888 bool dev_valid_name(const char *name
)
892 if (strlen(name
) >= IFNAMSIZ
)
894 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
898 if (*name
== '/' || isspace(*name
))
904 EXPORT_SYMBOL(dev_valid_name
);
907 * __dev_alloc_name - allocate a name for a device
908 * @net: network namespace to allocate the device name in
909 * @name: name format string
910 * @buf: scratch buffer and result name string
912 * Passed a format string - eg "lt%d" it will try and find a suitable
913 * id. It scans list of devices to build up a free map, then chooses
914 * the first empty slot. The caller must hold the dev_base or rtnl lock
915 * while allocating the name and adding the device in order to avoid
917 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
918 * Returns the number of the unit assigned or a negative errno code.
921 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
925 const int max_netdevices
= 8*PAGE_SIZE
;
926 unsigned long *inuse
;
927 struct net_device
*d
;
929 p
= strnchr(name
, IFNAMSIZ
-1, '%');
932 * Verify the string as this thing may have come from
933 * the user. There must be either one "%d" and no other "%"
936 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
939 /* Use one page as a bit array of possible slots */
940 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
944 for_each_netdev(net
, d
) {
945 if (!sscanf(d
->name
, name
, &i
))
947 if (i
< 0 || i
>= max_netdevices
)
950 /* avoid cases where sscanf is not exact inverse of printf */
951 snprintf(buf
, IFNAMSIZ
, name
, i
);
952 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
956 i
= find_first_zero_bit(inuse
, max_netdevices
);
957 free_page((unsigned long) inuse
);
961 snprintf(buf
, IFNAMSIZ
, name
, i
);
962 if (!__dev_get_by_name(net
, buf
))
965 /* It is possible to run out of possible slots
966 * when the name is long and there isn't enough space left
967 * for the digits, or if all bits are used.
973 * dev_alloc_name - allocate a name for a device
975 * @name: name format string
977 * Passed a format string - eg "lt%d" it will try and find a suitable
978 * id. It scans list of devices to build up a free map, then chooses
979 * the first empty slot. The caller must hold the dev_base or rtnl lock
980 * while allocating the name and adding the device in order to avoid
982 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
983 * Returns the number of the unit assigned or a negative errno code.
986 int dev_alloc_name(struct net_device
*dev
, const char *name
)
992 BUG_ON(!dev_net(dev
));
994 ret
= __dev_alloc_name(net
, name
, buf
);
996 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
999 EXPORT_SYMBOL(dev_alloc_name
);
1001 static int dev_alloc_name_ns(struct net
*net
,
1002 struct net_device
*dev
,
1008 ret
= __dev_alloc_name(net
, name
, buf
);
1010 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
1014 static int dev_get_valid_name(struct net
*net
,
1015 struct net_device
*dev
,
1020 if (!dev_valid_name(name
))
1023 if (strchr(name
, '%'))
1024 return dev_alloc_name_ns(net
, dev
, name
);
1025 else if (__dev_get_by_name(net
, name
))
1027 else if (dev
->name
!= name
)
1028 strlcpy(dev
->name
, name
, IFNAMSIZ
);
1034 * dev_change_name - change name of a device
1036 * @newname: name (or format string) must be at least IFNAMSIZ
1038 * Change name of a device, can pass format strings "eth%d".
1041 int dev_change_name(struct net_device
*dev
, const char *newname
)
1043 char oldname
[IFNAMSIZ
];
1049 BUG_ON(!dev_net(dev
));
1052 if (dev
->flags
& IFF_UP
)
1055 write_seqcount_begin(&devnet_rename_seq
);
1057 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0) {
1058 write_seqcount_end(&devnet_rename_seq
);
1062 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
1064 err
= dev_get_valid_name(net
, dev
, newname
);
1066 write_seqcount_end(&devnet_rename_seq
);
1071 ret
= device_rename(&dev
->dev
, dev
->name
);
1073 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1074 write_seqcount_end(&devnet_rename_seq
);
1078 write_seqcount_end(&devnet_rename_seq
);
1080 write_lock_bh(&dev_base_lock
);
1081 hlist_del_rcu(&dev
->name_hlist
);
1082 write_unlock_bh(&dev_base_lock
);
1086 write_lock_bh(&dev_base_lock
);
1087 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
1088 write_unlock_bh(&dev_base_lock
);
1090 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
1091 ret
= notifier_to_errno(ret
);
1094 /* err >= 0 after dev_alloc_name() or stores the first errno */
1097 write_seqcount_begin(&devnet_rename_seq
);
1098 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1101 pr_err("%s: name change rollback failed: %d\n",
1110 * dev_set_alias - change ifalias of a device
1112 * @alias: name up to IFALIASZ
1113 * @len: limit of bytes to copy from info
1115 * Set ifalias for a device,
1117 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
1123 if (len
>= IFALIASZ
)
1127 kfree(dev
->ifalias
);
1128 dev
->ifalias
= NULL
;
1132 new_ifalias
= krealloc(dev
->ifalias
, len
+ 1, GFP_KERNEL
);
1135 dev
->ifalias
= new_ifalias
;
1137 strlcpy(dev
->ifalias
, alias
, len
+1);
1143 * netdev_features_change - device changes features
1144 * @dev: device to cause notification
1146 * Called to indicate a device has changed features.
1148 void netdev_features_change(struct net_device
*dev
)
1150 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1152 EXPORT_SYMBOL(netdev_features_change
);
1155 * netdev_state_change - device changes state
1156 * @dev: device to cause notification
1158 * Called to indicate a device has changed state. This function calls
1159 * the notifier chains for netdev_chain and sends a NEWLINK message
1160 * to the routing socket.
1162 void netdev_state_change(struct net_device
*dev
)
1164 if (dev
->flags
& IFF_UP
) {
1165 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
1166 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
1169 EXPORT_SYMBOL(netdev_state_change
);
1172 * netdev_notify_peers - notify network peers about existence of @dev
1173 * @dev: network device
1175 * Generate traffic such that interested network peers are aware of
1176 * @dev, such as by generating a gratuitous ARP. This may be used when
1177 * a device wants to inform the rest of the network about some sort of
1178 * reconfiguration such as a failover event or virtual machine
1181 void netdev_notify_peers(struct net_device
*dev
)
1184 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
, dev
);
1187 EXPORT_SYMBOL(netdev_notify_peers
);
1189 static int __dev_open(struct net_device
*dev
)
1191 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1196 if (!netif_device_present(dev
))
1199 /* Block netpoll from trying to do any rx path servicing.
1200 * If we don't do this there is a chance ndo_poll_controller
1201 * or ndo_poll may be running while we open the device
1203 ret
= netpoll_rx_disable(dev
);
1207 ret
= call_netdevice_notifiers(NETDEV_PRE_UP
, dev
);
1208 ret
= notifier_to_errno(ret
);
1212 set_bit(__LINK_STATE_START
, &dev
->state
);
1214 if (ops
->ndo_validate_addr
)
1215 ret
= ops
->ndo_validate_addr(dev
);
1217 if (!ret
&& ops
->ndo_open
)
1218 ret
= ops
->ndo_open(dev
);
1220 netpoll_rx_enable(dev
);
1223 clear_bit(__LINK_STATE_START
, &dev
->state
);
1225 dev
->flags
|= IFF_UP
;
1226 net_dmaengine_get();
1227 dev_set_rx_mode(dev
);
1229 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
1236 * dev_open - prepare an interface for use.
1237 * @dev: device to open
1239 * Takes a device from down to up state. The device's private open
1240 * function is invoked and then the multicast lists are loaded. Finally
1241 * the device is moved into the up state and a %NETDEV_UP message is
1242 * sent to the netdev notifier chain.
1244 * Calling this function on an active interface is a nop. On a failure
1245 * a negative errno code is returned.
1247 int dev_open(struct net_device
*dev
)
1251 if (dev
->flags
& IFF_UP
)
1254 ret
= __dev_open(dev
);
1258 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1259 call_netdevice_notifiers(NETDEV_UP
, dev
);
1263 EXPORT_SYMBOL(dev_open
);
1265 static int __dev_close_many(struct list_head
*head
)
1267 struct net_device
*dev
;
1272 list_for_each_entry(dev
, head
, unreg_list
) {
1273 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1275 clear_bit(__LINK_STATE_START
, &dev
->state
);
1277 /* Synchronize to scheduled poll. We cannot touch poll list, it
1278 * can be even on different cpu. So just clear netif_running().
1280 * dev->stop() will invoke napi_disable() on all of it's
1281 * napi_struct instances on this device.
1283 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1286 dev_deactivate_many(head
);
1288 list_for_each_entry(dev
, head
, unreg_list
) {
1289 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1292 * Call the device specific close. This cannot fail.
1293 * Only if device is UP
1295 * We allow it to be called even after a DETACH hot-plug
1301 dev
->flags
&= ~IFF_UP
;
1302 net_dmaengine_put();
1308 static int __dev_close(struct net_device
*dev
)
1313 /* Temporarily disable netpoll until the interface is down */
1314 retval
= netpoll_rx_disable(dev
);
1318 list_add(&dev
->unreg_list
, &single
);
1319 retval
= __dev_close_many(&single
);
1322 netpoll_rx_enable(dev
);
1326 static int dev_close_many(struct list_head
*head
)
1328 struct net_device
*dev
, *tmp
;
1329 LIST_HEAD(tmp_list
);
1331 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
)
1332 if (!(dev
->flags
& IFF_UP
))
1333 list_move(&dev
->unreg_list
, &tmp_list
);
1335 __dev_close_many(head
);
1337 list_for_each_entry(dev
, head
, unreg_list
) {
1338 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1339 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1342 /* rollback_registered_many needs the complete original list */
1343 list_splice(&tmp_list
, head
);
1348 * dev_close - shutdown an interface.
1349 * @dev: device to shutdown
1351 * This function moves an active device into down state. A
1352 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1353 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1356 int dev_close(struct net_device
*dev
)
1359 if (dev
->flags
& IFF_UP
) {
1362 /* Block netpoll rx while the interface is going down */
1363 ret
= netpoll_rx_disable(dev
);
1367 list_add(&dev
->unreg_list
, &single
);
1368 dev_close_many(&single
);
1371 netpoll_rx_enable(dev
);
1375 EXPORT_SYMBOL(dev_close
);
1379 * dev_disable_lro - disable Large Receive Offload on a device
1382 * Disable Large Receive Offload (LRO) on a net device. Must be
1383 * called under RTNL. This is needed if received packets may be
1384 * forwarded to another interface.
1386 void dev_disable_lro(struct net_device
*dev
)
1389 * If we're trying to disable lro on a vlan device
1390 * use the underlying physical device instead
1392 if (is_vlan_dev(dev
))
1393 dev
= vlan_dev_real_dev(dev
);
1395 dev
->wanted_features
&= ~NETIF_F_LRO
;
1396 netdev_update_features(dev
);
1398 if (unlikely(dev
->features
& NETIF_F_LRO
))
1399 netdev_WARN(dev
, "failed to disable LRO!\n");
1401 EXPORT_SYMBOL(dev_disable_lro
);
1404 static int dev_boot_phase
= 1;
1407 * register_netdevice_notifier - register a network notifier block
1410 * Register a notifier to be called when network device events occur.
1411 * The notifier passed is linked into the kernel structures and must
1412 * not be reused until it has been unregistered. A negative errno code
1413 * is returned on a failure.
1415 * When registered all registration and up events are replayed
1416 * to the new notifier to allow device to have a race free
1417 * view of the network device list.
1420 int register_netdevice_notifier(struct notifier_block
*nb
)
1422 struct net_device
*dev
;
1423 struct net_device
*last
;
1428 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1434 for_each_netdev(net
, dev
) {
1435 err
= nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1436 err
= notifier_to_errno(err
);
1440 if (!(dev
->flags
& IFF_UP
))
1443 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1454 for_each_netdev(net
, dev
) {
1458 if (dev
->flags
& IFF_UP
) {
1459 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1460 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1462 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1467 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1470 EXPORT_SYMBOL(register_netdevice_notifier
);
1473 * unregister_netdevice_notifier - unregister a network notifier block
1476 * Unregister a notifier previously registered by
1477 * register_netdevice_notifier(). The notifier is unlinked into the
1478 * kernel structures and may then be reused. A negative errno code
1479 * is returned on a failure.
1481 * After unregistering unregister and down device events are synthesized
1482 * for all devices on the device list to the removed notifier to remove
1483 * the need for special case cleanup code.
1486 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1488 struct net_device
*dev
;
1493 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1498 for_each_netdev(net
, dev
) {
1499 if (dev
->flags
& IFF_UP
) {
1500 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1501 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1503 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1510 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1513 * call_netdevice_notifiers - call all network notifier blocks
1514 * @val: value passed unmodified to notifier function
1515 * @dev: net_device pointer passed unmodified to notifier function
1517 * Call all network notifier blocks. Parameters and return value
1518 * are as for raw_notifier_call_chain().
1521 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1524 return raw_notifier_call_chain(&netdev_chain
, val
, dev
);
1526 EXPORT_SYMBOL(call_netdevice_notifiers
);
1528 static struct static_key netstamp_needed __read_mostly
;
1529 #ifdef HAVE_JUMP_LABEL
1530 /* We are not allowed to call static_key_slow_dec() from irq context
1531 * If net_disable_timestamp() is called from irq context, defer the
1532 * static_key_slow_dec() calls.
1534 static atomic_t netstamp_needed_deferred
;
1537 void net_enable_timestamp(void)
1539 #ifdef HAVE_JUMP_LABEL
1540 int deferred
= atomic_xchg(&netstamp_needed_deferred
, 0);
1544 static_key_slow_dec(&netstamp_needed
);
1548 static_key_slow_inc(&netstamp_needed
);
1550 EXPORT_SYMBOL(net_enable_timestamp
);
1552 void net_disable_timestamp(void)
1554 #ifdef HAVE_JUMP_LABEL
1555 if (in_interrupt()) {
1556 atomic_inc(&netstamp_needed_deferred
);
1560 static_key_slow_dec(&netstamp_needed
);
1562 EXPORT_SYMBOL(net_disable_timestamp
);
1564 static inline void net_timestamp_set(struct sk_buff
*skb
)
1566 skb
->tstamp
.tv64
= 0;
1567 if (static_key_false(&netstamp_needed
))
1568 __net_timestamp(skb
);
1571 #define net_timestamp_check(COND, SKB) \
1572 if (static_key_false(&netstamp_needed)) { \
1573 if ((COND) && !(SKB)->tstamp.tv64) \
1574 __net_timestamp(SKB); \
1577 static inline bool is_skb_forwardable(struct net_device *dev,
1578 struct sk_buff
*skb
)
1582 if (!(dev
->flags
& IFF_UP
))
1585 len
= dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
;
1586 if (skb
->len
<= len
)
1589 /* if TSO is enabled, we don't care about the length as the packet
1590 * could be forwarded without being segmented before
1592 if (skb_is_gso(skb
))
1599 * dev_forward_skb - loopback an skb to another netif
1601 * @dev: destination network device
1602 * @skb: buffer to forward
1605 * NET_RX_SUCCESS (no congestion)
1606 * NET_RX_DROP (packet was dropped, but freed)
1608 * dev_forward_skb can be used for injecting an skb from the
1609 * start_xmit function of one device into the receive queue
1610 * of another device.
1612 * The receiving device may be in another namespace, so
1613 * we have to clear all information in the skb that could
1614 * impact namespace isolation.
1616 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1618 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
1619 if (skb_copy_ubufs(skb
, GFP_ATOMIC
)) {
1620 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
|| !skb
->sk
)
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 static DEFINE_MUTEX(xps_map_mutex
);
1763 #define xmap_dereference(P) \
1764 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1766 static struct xps_map
*remove_xps_queue(struct xps_dev_maps
*dev_maps
,
1769 struct xps_map
*map
= NULL
;
1773 map
= xmap_dereference(dev_maps
->cpu_map
[cpu
]);
1775 for (pos
= 0; map
&& pos
< map
->len
; pos
++) {
1776 if (map
->queues
[pos
] == index
) {
1778 map
->queues
[pos
] = map
->queues
[--map
->len
];
1780 RCU_INIT_POINTER(dev_maps
->cpu_map
[cpu
], NULL
);
1781 kfree_rcu(map
, rcu
);
1791 static void netif_reset_xps_queues_gt(struct net_device
*dev
, u16 index
)
1793 struct xps_dev_maps
*dev_maps
;
1795 bool active
= false;
1797 mutex_lock(&xps_map_mutex
);
1798 dev_maps
= xmap_dereference(dev
->xps_maps
);
1803 for_each_possible_cpu(cpu
) {
1804 for (i
= index
; i
< dev
->num_tx_queues
; i
++) {
1805 if (!remove_xps_queue(dev_maps
, cpu
, i
))
1808 if (i
== dev
->num_tx_queues
)
1813 RCU_INIT_POINTER(dev
->xps_maps
, NULL
);
1814 kfree_rcu(dev_maps
, rcu
);
1817 for (i
= index
; i
< dev
->num_tx_queues
; i
++)
1818 netdev_queue_numa_node_write(netdev_get_tx_queue(dev
, i
),
1822 mutex_unlock(&xps_map_mutex
);
1825 static struct xps_map
*expand_xps_map(struct xps_map
*map
,
1828 struct xps_map
*new_map
;
1829 int alloc_len
= XPS_MIN_MAP_ALLOC
;
1832 for (pos
= 0; map
&& pos
< map
->len
; pos
++) {
1833 if (map
->queues
[pos
] != index
)
1838 /* Need to add queue to this CPU's existing map */
1840 if (pos
< map
->alloc_len
)
1843 alloc_len
= map
->alloc_len
* 2;
1846 /* Need to allocate new map to store queue on this CPU's map */
1847 new_map
= kzalloc_node(XPS_MAP_SIZE(alloc_len
), GFP_KERNEL
,
1852 for (i
= 0; i
< pos
; i
++)
1853 new_map
->queues
[i
] = map
->queues
[i
];
1854 new_map
->alloc_len
= alloc_len
;
1860 int netif_set_xps_queue(struct net_device
*dev
, struct cpumask
*mask
, u16 index
)
1862 struct xps_dev_maps
*dev_maps
, *new_dev_maps
= NULL
;
1863 struct xps_map
*map
, *new_map
;
1864 int maps_sz
= max_t(unsigned int, XPS_DEV_MAPS_SIZE
, L1_CACHE_BYTES
);
1865 int cpu
, numa_node_id
= -2;
1866 bool active
= false;
1868 mutex_lock(&xps_map_mutex
);
1870 dev_maps
= xmap_dereference(dev
->xps_maps
);
1872 /* allocate memory for queue storage */
1873 for_each_online_cpu(cpu
) {
1874 if (!cpumask_test_cpu(cpu
, mask
))
1878 new_dev_maps
= kzalloc(maps_sz
, GFP_KERNEL
);
1879 if (!new_dev_maps
) {
1880 mutex_unlock(&xps_map_mutex
);
1884 map
= dev_maps
? xmap_dereference(dev_maps
->cpu_map
[cpu
]) :
1887 map
= expand_xps_map(map
, cpu
, index
);
1891 RCU_INIT_POINTER(new_dev_maps
->cpu_map
[cpu
], map
);
1895 goto out_no_new_maps
;
1897 for_each_possible_cpu(cpu
) {
1898 if (cpumask_test_cpu(cpu
, mask
) && cpu_online(cpu
)) {
1899 /* add queue to CPU maps */
1902 map
= xmap_dereference(new_dev_maps
->cpu_map
[cpu
]);
1903 while ((pos
< map
->len
) && (map
->queues
[pos
] != index
))
1906 if (pos
== map
->len
)
1907 map
->queues
[map
->len
++] = index
;
1909 if (numa_node_id
== -2)
1910 numa_node_id
= cpu_to_node(cpu
);
1911 else if (numa_node_id
!= cpu_to_node(cpu
))
1914 } else if (dev_maps
) {
1915 /* fill in the new device map from the old device map */
1916 map
= xmap_dereference(dev_maps
->cpu_map
[cpu
]);
1917 RCU_INIT_POINTER(new_dev_maps
->cpu_map
[cpu
], map
);
1922 rcu_assign_pointer(dev
->xps_maps
, new_dev_maps
);
1924 /* Cleanup old maps */
1926 for_each_possible_cpu(cpu
) {
1927 new_map
= xmap_dereference(new_dev_maps
->cpu_map
[cpu
]);
1928 map
= xmap_dereference(dev_maps
->cpu_map
[cpu
]);
1929 if (map
&& map
!= new_map
)
1930 kfree_rcu(map
, rcu
);
1933 kfree_rcu(dev_maps
, rcu
);
1936 dev_maps
= new_dev_maps
;
1940 /* update Tx queue numa node */
1941 netdev_queue_numa_node_write(netdev_get_tx_queue(dev
, index
),
1942 (numa_node_id
>= 0) ? numa_node_id
:
1948 /* removes queue from unused CPUs */
1949 for_each_possible_cpu(cpu
) {
1950 if (cpumask_test_cpu(cpu
, mask
) && cpu_online(cpu
))
1953 if (remove_xps_queue(dev_maps
, cpu
, index
))
1957 /* free map if not active */
1959 RCU_INIT_POINTER(dev
->xps_maps
, NULL
);
1960 kfree_rcu(dev_maps
, rcu
);
1964 mutex_unlock(&xps_map_mutex
);
1968 /* remove any maps that we added */
1969 for_each_possible_cpu(cpu
) {
1970 new_map
= xmap_dereference(new_dev_maps
->cpu_map
[cpu
]);
1971 map
= dev_maps
? xmap_dereference(dev_maps
->cpu_map
[cpu
]) :
1973 if (new_map
&& new_map
!= map
)
1977 mutex_unlock(&xps_map_mutex
);
1979 kfree(new_dev_maps
);
1982 EXPORT_SYMBOL(netif_set_xps_queue
);
1986 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1987 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1989 int netif_set_real_num_tx_queues(struct net_device
*dev
, unsigned int txq
)
1993 if (txq
< 1 || txq
> dev
->num_tx_queues
)
1996 if (dev
->reg_state
== NETREG_REGISTERED
||
1997 dev
->reg_state
== NETREG_UNREGISTERING
) {
2000 rc
= netdev_queue_update_kobjects(dev
, dev
->real_num_tx_queues
,
2006 netif_setup_tc(dev
, txq
);
2008 if (txq
< dev
->real_num_tx_queues
) {
2009 qdisc_reset_all_tx_gt(dev
, txq
);
2011 netif_reset_xps_queues_gt(dev
, txq
);
2016 dev
->real_num_tx_queues
= txq
;
2019 EXPORT_SYMBOL(netif_set_real_num_tx_queues
);
2023 * netif_set_real_num_rx_queues - set actual number of RX queues used
2024 * @dev: Network device
2025 * @rxq: Actual number of RX queues
2027 * This must be called either with the rtnl_lock held or before
2028 * registration of the net device. Returns 0 on success, or a
2029 * negative error code. If called before registration, it always
2032 int netif_set_real_num_rx_queues(struct net_device
*dev
, unsigned int rxq
)
2036 if (rxq
< 1 || rxq
> dev
->num_rx_queues
)
2039 if (dev
->reg_state
== NETREG_REGISTERED
) {
2042 rc
= net_rx_queue_update_kobjects(dev
, dev
->real_num_rx_queues
,
2048 dev
->real_num_rx_queues
= rxq
;
2051 EXPORT_SYMBOL(netif_set_real_num_rx_queues
);
2055 * netif_get_num_default_rss_queues - default number of RSS queues
2057 * This routine should set an upper limit on the number of RSS queues
2058 * used by default by multiqueue devices.
2060 int netif_get_num_default_rss_queues(void)
2062 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES
, num_online_cpus());
2064 EXPORT_SYMBOL(netif_get_num_default_rss_queues
);
2066 static inline void __netif_reschedule(struct Qdisc
*q
)
2068 struct softnet_data
*sd
;
2069 unsigned long flags
;
2071 local_irq_save(flags
);
2072 sd
= &__get_cpu_var(softnet_data
);
2073 q
->next_sched
= NULL
;
2074 *sd
->output_queue_tailp
= q
;
2075 sd
->output_queue_tailp
= &q
->next_sched
;
2076 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
2077 local_irq_restore(flags
);
2080 void __netif_schedule(struct Qdisc
*q
)
2082 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
2083 __netif_reschedule(q
);
2085 EXPORT_SYMBOL(__netif_schedule
);
2087 void dev_kfree_skb_irq(struct sk_buff
*skb
)
2089 if (atomic_dec_and_test(&skb
->users
)) {
2090 struct softnet_data
*sd
;
2091 unsigned long flags
;
2093 local_irq_save(flags
);
2094 sd
= &__get_cpu_var(softnet_data
);
2095 skb
->next
= sd
->completion_queue
;
2096 sd
->completion_queue
= skb
;
2097 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
2098 local_irq_restore(flags
);
2101 EXPORT_SYMBOL(dev_kfree_skb_irq
);
2103 void dev_kfree_skb_any(struct sk_buff
*skb
)
2105 if (in_irq() || irqs_disabled())
2106 dev_kfree_skb_irq(skb
);
2110 EXPORT_SYMBOL(dev_kfree_skb_any
);
2114 * netif_device_detach - mark device as removed
2115 * @dev: network device
2117 * Mark device as removed from system and therefore no longer available.
2119 void netif_device_detach(struct net_device
*dev
)
2121 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
2122 netif_running(dev
)) {
2123 netif_tx_stop_all_queues(dev
);
2126 EXPORT_SYMBOL(netif_device_detach
);
2129 * netif_device_attach - mark device as attached
2130 * @dev: network device
2132 * Mark device as attached from system and restart if needed.
2134 void netif_device_attach(struct net_device
*dev
)
2136 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
2137 netif_running(dev
)) {
2138 netif_tx_wake_all_queues(dev
);
2139 __netdev_watchdog_up(dev
);
2142 EXPORT_SYMBOL(netif_device_attach
);
2144 static void skb_warn_bad_offload(const struct sk_buff
*skb
)
2146 static const netdev_features_t null_features
= 0;
2147 struct net_device
*dev
= skb
->dev
;
2148 const char *driver
= "";
2150 if (dev
&& dev
->dev
.parent
)
2151 driver
= dev_driver_string(dev
->dev
.parent
);
2153 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2154 "gso_type=%d ip_summed=%d\n",
2155 driver
, dev
? &dev
->features
: &null_features
,
2156 skb
->sk
? &skb
->sk
->sk_route_caps
: &null_features
,
2157 skb
->len
, skb
->data_len
, skb_shinfo(skb
)->gso_size
,
2158 skb_shinfo(skb
)->gso_type
, skb
->ip_summed
);
2162 * Invalidate hardware checksum when packet is to be mangled, and
2163 * complete checksum manually on outgoing path.
2165 int skb_checksum_help(struct sk_buff
*skb
)
2168 int ret
= 0, offset
;
2170 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
2171 goto out_set_summed
;
2173 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
2174 skb_warn_bad_offload(skb
);
2178 /* Before computing a checksum, we should make sure no frag could
2179 * be modified by an external entity : checksum could be wrong.
2181 if (skb_has_shared_frag(skb
)) {
2182 ret
= __skb_linearize(skb
);
2187 offset
= skb_checksum_start_offset(skb
);
2188 BUG_ON(offset
>= skb_headlen(skb
));
2189 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
2191 offset
+= skb
->csum_offset
;
2192 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
2194 if (skb_cloned(skb
) &&
2195 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
2196 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2201 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
2203 skb
->ip_summed
= CHECKSUM_NONE
;
2207 EXPORT_SYMBOL(skb_checksum_help
);
2210 * skb_mac_gso_segment - mac layer segmentation handler.
2211 * @skb: buffer to segment
2212 * @features: features for the output path (see dev->features)
2214 struct sk_buff
*skb_mac_gso_segment(struct sk_buff
*skb
,
2215 netdev_features_t features
)
2217 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
2218 struct packet_offload
*ptype
;
2219 __be16 type
= skb
->protocol
;
2220 int vlan_depth
= ETH_HLEN
;
2222 while (type
== htons(ETH_P_8021Q
)) {
2223 struct vlan_hdr
*vh
;
2225 if (unlikely(!pskb_may_pull(skb
, vlan_depth
+ VLAN_HLEN
)))
2226 return ERR_PTR(-EINVAL
);
2228 vh
= (struct vlan_hdr
*)(skb
->data
+ vlan_depth
);
2229 type
= vh
->h_vlan_encapsulated_proto
;
2230 vlan_depth
+= VLAN_HLEN
;
2233 __skb_pull(skb
, skb
->mac_len
);
2236 list_for_each_entry_rcu(ptype
, &offload_base
, list
) {
2237 if (ptype
->type
== type
&& ptype
->callbacks
.gso_segment
) {
2238 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
2241 err
= ptype
->callbacks
.gso_send_check(skb
);
2242 segs
= ERR_PTR(err
);
2243 if (err
|| skb_gso_ok(skb
, features
))
2245 __skb_push(skb
, (skb
->data
-
2246 skb_network_header(skb
)));
2248 segs
= ptype
->callbacks
.gso_segment(skb
, features
);
2254 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
2258 EXPORT_SYMBOL(skb_mac_gso_segment
);
2261 /* openvswitch calls this on rx path, so we need a different check.
2263 static inline bool skb_needs_check(struct sk_buff
*skb
, bool tx_path
)
2266 return skb
->ip_summed
!= CHECKSUM_PARTIAL
;
2268 return skb
->ip_summed
== CHECKSUM_NONE
;
2272 * __skb_gso_segment - Perform segmentation on skb.
2273 * @skb: buffer to segment
2274 * @features: features for the output path (see dev->features)
2275 * @tx_path: whether it is called in TX path
2277 * This function segments the given skb and returns a list of segments.
2279 * It may return NULL if the skb requires no segmentation. This is
2280 * only possible when GSO is used for verifying header integrity.
2282 struct sk_buff
*__skb_gso_segment(struct sk_buff
*skb
,
2283 netdev_features_t features
, bool tx_path
)
2285 if (unlikely(skb_needs_check(skb
, tx_path
))) {
2288 skb_warn_bad_offload(skb
);
2290 if (skb_header_cloned(skb
) &&
2291 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
2292 return ERR_PTR(err
);
2295 SKB_GSO_CB(skb
)->mac_offset
= skb_headroom(skb
);
2296 skb_reset_mac_header(skb
);
2297 skb_reset_mac_len(skb
);
2299 return skb_mac_gso_segment(skb
, features
);
2301 EXPORT_SYMBOL(__skb_gso_segment
);
2303 /* Take action when hardware reception checksum errors are detected. */
2305 void netdev_rx_csum_fault(struct net_device
*dev
)
2307 if (net_ratelimit()) {
2308 pr_err("%s: hw csum failure\n", dev
? dev
->name
: "<unknown>");
2312 EXPORT_SYMBOL(netdev_rx_csum_fault
);
2315 /* Actually, we should eliminate this check as soon as we know, that:
2316 * 1. IOMMU is present and allows to map all the memory.
2317 * 2. No high memory really exists on this machine.
2320 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
2322 #ifdef CONFIG_HIGHMEM
2324 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
2325 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2326 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2327 if (PageHighMem(skb_frag_page(frag
)))
2332 if (PCI_DMA_BUS_IS_PHYS
) {
2333 struct device
*pdev
= dev
->dev
.parent
;
2337 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2338 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2339 dma_addr_t addr
= page_to_phys(skb_frag_page(frag
));
2340 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
2349 void (*destructor
)(struct sk_buff
*skb
);
2352 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2354 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
2356 struct dev_gso_cb
*cb
;
2359 struct sk_buff
*nskb
= skb
->next
;
2361 skb
->next
= nskb
->next
;
2364 } while (skb
->next
);
2366 cb
= DEV_GSO_CB(skb
);
2368 cb
->destructor(skb
);
2372 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2373 * @skb: buffer to segment
2374 * @features: device features as applicable to this skb
2376 * This function segments the given skb and stores the list of segments
2379 static int dev_gso_segment(struct sk_buff
*skb
, netdev_features_t features
)
2381 struct sk_buff
*segs
;
2383 segs
= skb_gso_segment(skb
, features
);
2385 /* Verifying header integrity only. */
2390 return PTR_ERR(segs
);
2393 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
2394 skb
->destructor
= dev_gso_skb_destructor
;
2399 static bool can_checksum_protocol(netdev_features_t features
, __be16 protocol
)
2401 return ((features
& NETIF_F_GEN_CSUM
) ||
2402 ((features
& NETIF_F_V4_CSUM
) &&
2403 protocol
== htons(ETH_P_IP
)) ||
2404 ((features
& NETIF_F_V6_CSUM
) &&
2405 protocol
== htons(ETH_P_IPV6
)) ||
2406 ((features
& NETIF_F_FCOE_CRC
) &&
2407 protocol
== htons(ETH_P_FCOE
)));
2410 static netdev_features_t
harmonize_features(struct sk_buff
*skb
,
2411 __be16 protocol
, netdev_features_t features
)
2413 if (skb
->ip_summed
!= CHECKSUM_NONE
&&
2414 !can_checksum_protocol(features
, protocol
)) {
2415 features
&= ~NETIF_F_ALL_CSUM
;
2416 features
&= ~NETIF_F_SG
;
2417 } else if (illegal_highdma(skb
->dev
, skb
)) {
2418 features
&= ~NETIF_F_SG
;
2424 netdev_features_t
netif_skb_features(struct sk_buff
*skb
)
2426 __be16 protocol
= skb
->protocol
;
2427 netdev_features_t features
= skb
->dev
->features
;
2429 if (skb_shinfo(skb
)->gso_segs
> skb
->dev
->gso_max_segs
)
2430 features
&= ~NETIF_F_GSO_MASK
;
2432 if (protocol
== htons(ETH_P_8021Q
)) {
2433 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
2434 protocol
= veh
->h_vlan_encapsulated_proto
;
2435 } else if (!vlan_tx_tag_present(skb
)) {
2436 return harmonize_features(skb
, protocol
, features
);
2439 features
&= (skb
->dev
->vlan_features
| NETIF_F_HW_VLAN_TX
);
2441 if (protocol
!= htons(ETH_P_8021Q
)) {
2442 return harmonize_features(skb
, protocol
, features
);
2444 features
&= NETIF_F_SG
| NETIF_F_HIGHDMA
| NETIF_F_FRAGLIST
|
2445 NETIF_F_GEN_CSUM
| NETIF_F_HW_VLAN_TX
;
2446 return harmonize_features(skb
, protocol
, features
);
2449 EXPORT_SYMBOL(netif_skb_features
);
2452 * Returns true if either:
2453 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2454 * 2. skb is fragmented and the device does not support SG.
2456 static inline int skb_needs_linearize(struct sk_buff
*skb
,
2459 return skb_is_nonlinear(skb
) &&
2460 ((skb_has_frag_list(skb
) &&
2461 !(features
& NETIF_F_FRAGLIST
)) ||
2462 (skb_shinfo(skb
)->nr_frags
&&
2463 !(features
& NETIF_F_SG
)));
2466 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
2467 struct netdev_queue
*txq
)
2469 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2470 int rc
= NETDEV_TX_OK
;
2471 unsigned int skb_len
;
2473 if (likely(!skb
->next
)) {
2474 netdev_features_t features
;
2477 * If device doesn't need skb->dst, release it right now while
2478 * its hot in this cpu cache
2480 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2483 features
= netif_skb_features(skb
);
2485 if (vlan_tx_tag_present(skb
) &&
2486 !(features
& NETIF_F_HW_VLAN_TX
)) {
2487 skb
= __vlan_put_tag(skb
, vlan_tx_tag_get(skb
));
2494 /* If encapsulation offload request, verify we are testing
2495 * hardware encapsulation features instead of standard
2496 * features for the netdev
2498 if (skb
->encapsulation
)
2499 features
&= dev
->hw_enc_features
;
2501 if (netif_needs_gso(skb
, features
)) {
2502 if (unlikely(dev_gso_segment(skb
, features
)))
2507 if (skb_needs_linearize(skb
, features
) &&
2508 __skb_linearize(skb
))
2511 /* If packet is not checksummed and device does not
2512 * support checksumming for this protocol, complete
2513 * checksumming here.
2515 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2516 if (skb
->encapsulation
)
2517 skb_set_inner_transport_header(skb
,
2518 skb_checksum_start_offset(skb
));
2520 skb_set_transport_header(skb
,
2521 skb_checksum_start_offset(skb
));
2522 if (!(features
& NETIF_F_ALL_CSUM
) &&
2523 skb_checksum_help(skb
))
2528 if (!list_empty(&ptype_all
))
2529 dev_queue_xmit_nit(skb
, dev
);
2532 rc
= ops
->ndo_start_xmit(skb
, dev
);
2533 trace_net_dev_xmit(skb
, rc
, dev
, skb_len
);
2534 if (rc
== NETDEV_TX_OK
)
2535 txq_trans_update(txq
);
2541 struct sk_buff
*nskb
= skb
->next
;
2543 skb
->next
= nskb
->next
;
2547 * If device doesn't need nskb->dst, release it right now while
2548 * its hot in this cpu cache
2550 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2553 if (!list_empty(&ptype_all
))
2554 dev_queue_xmit_nit(nskb
, dev
);
2556 skb_len
= nskb
->len
;
2557 rc
= ops
->ndo_start_xmit(nskb
, dev
);
2558 trace_net_dev_xmit(nskb
, rc
, dev
, skb_len
);
2559 if (unlikely(rc
!= NETDEV_TX_OK
)) {
2560 if (rc
& ~NETDEV_TX_MASK
)
2561 goto out_kfree_gso_skb
;
2562 nskb
->next
= skb
->next
;
2566 txq_trans_update(txq
);
2567 if (unlikely(netif_xmit_stopped(txq
) && skb
->next
))
2568 return NETDEV_TX_BUSY
;
2569 } while (skb
->next
);
2572 if (likely(skb
->next
== NULL
))
2573 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
2580 static void qdisc_pkt_len_init(struct sk_buff
*skb
)
2582 const struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
2584 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
2586 /* To get more precise estimation of bytes sent on wire,
2587 * we add to pkt_len the headers size of all segments
2589 if (shinfo
->gso_size
) {
2590 unsigned int hdr_len
;
2592 /* mac layer + network layer */
2593 hdr_len
= skb_transport_header(skb
) - skb_mac_header(skb
);
2595 /* + transport layer */
2596 if (likely(shinfo
->gso_type
& (SKB_GSO_TCPV4
| SKB_GSO_TCPV6
)))
2597 hdr_len
+= tcp_hdrlen(skb
);
2599 hdr_len
+= sizeof(struct udphdr
);
2600 qdisc_skb_cb(skb
)->pkt_len
+= (shinfo
->gso_segs
- 1) * hdr_len
;
2604 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2605 struct net_device
*dev
,
2606 struct netdev_queue
*txq
)
2608 spinlock_t
*root_lock
= qdisc_lock(q
);
2612 qdisc_pkt_len_init(skb
);
2613 qdisc_calculate_pkt_len(skb
, q
);
2615 * Heuristic to force contended enqueues to serialize on a
2616 * separate lock before trying to get qdisc main lock.
2617 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2618 * and dequeue packets faster.
2620 contended
= qdisc_is_running(q
);
2621 if (unlikely(contended
))
2622 spin_lock(&q
->busylock
);
2624 spin_lock(root_lock
);
2625 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2628 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2629 qdisc_run_begin(q
)) {
2631 * This is a work-conserving queue; there are no old skbs
2632 * waiting to be sent out; and the qdisc is not running -
2633 * xmit the skb directly.
2635 if (!(dev
->priv_flags
& IFF_XMIT_DST_RELEASE
))
2638 qdisc_bstats_update(q
, skb
);
2640 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
)) {
2641 if (unlikely(contended
)) {
2642 spin_unlock(&q
->busylock
);
2649 rc
= NET_XMIT_SUCCESS
;
2652 rc
= q
->enqueue(skb
, q
) & NET_XMIT_MASK
;
2653 if (qdisc_run_begin(q
)) {
2654 if (unlikely(contended
)) {
2655 spin_unlock(&q
->busylock
);
2661 spin_unlock(root_lock
);
2662 if (unlikely(contended
))
2663 spin_unlock(&q
->busylock
);
2667 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2668 static void skb_update_prio(struct sk_buff
*skb
)
2670 struct netprio_map
*map
= rcu_dereference_bh(skb
->dev
->priomap
);
2672 if (!skb
->priority
&& skb
->sk
&& map
) {
2673 unsigned int prioidx
= skb
->sk
->sk_cgrp_prioidx
;
2675 if (prioidx
< map
->priomap_len
)
2676 skb
->priority
= map
->priomap
[prioidx
];
2680 #define skb_update_prio(skb)
2683 static DEFINE_PER_CPU(int, xmit_recursion
);
2684 #define RECURSION_LIMIT 10
2687 * dev_loopback_xmit - loop back @skb
2688 * @skb: buffer to transmit
2690 int dev_loopback_xmit(struct sk_buff
*skb
)
2692 skb_reset_mac_header(skb
);
2693 __skb_pull(skb
, skb_network_offset(skb
));
2694 skb
->pkt_type
= PACKET_LOOPBACK
;
2695 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
2696 WARN_ON(!skb_dst(skb
));
2701 EXPORT_SYMBOL(dev_loopback_xmit
);
2704 * dev_queue_xmit - transmit a buffer
2705 * @skb: buffer to transmit
2707 * Queue a buffer for transmission to a network device. The caller must
2708 * have set the device and priority and built the buffer before calling
2709 * this function. The function can be called from an interrupt.
2711 * A negative errno code is returned on a failure. A success does not
2712 * guarantee the frame will be transmitted as it may be dropped due
2713 * to congestion or traffic shaping.
2715 * -----------------------------------------------------------------------------------
2716 * I notice this method can also return errors from the queue disciplines,
2717 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2720 * Regardless of the return value, the skb is consumed, so it is currently
2721 * difficult to retry a send to this method. (You can bump the ref count
2722 * before sending to hold a reference for retry if you are careful.)
2724 * When calling this method, interrupts MUST be enabled. This is because
2725 * the BH enable code must have IRQs enabled so that it will not deadlock.
2728 int dev_queue_xmit(struct sk_buff
*skb
)
2730 struct net_device
*dev
= skb
->dev
;
2731 struct netdev_queue
*txq
;
2735 skb_reset_mac_header(skb
);
2737 /* Disable soft irqs for various locks below. Also
2738 * stops preemption for RCU.
2742 skb_update_prio(skb
);
2744 txq
= netdev_pick_tx(dev
, skb
);
2745 q
= rcu_dereference_bh(txq
->qdisc
);
2747 #ifdef CONFIG_NET_CLS_ACT
2748 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2750 trace_net_dev_queue(skb
);
2752 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2756 /* The device has no queue. Common case for software devices:
2757 loopback, all the sorts of tunnels...
2759 Really, it is unlikely that netif_tx_lock protection is necessary
2760 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2762 However, it is possible, that they rely on protection
2765 Check this and shot the lock. It is not prone from deadlocks.
2766 Either shot noqueue qdisc, it is even simpler 8)
2768 if (dev
->flags
& IFF_UP
) {
2769 int cpu
= smp_processor_id(); /* ok because BHs are off */
2771 if (txq
->xmit_lock_owner
!= cpu
) {
2773 if (__this_cpu_read(xmit_recursion
) > RECURSION_LIMIT
)
2774 goto recursion_alert
;
2776 HARD_TX_LOCK(dev
, txq
, cpu
);
2778 if (!netif_xmit_stopped(txq
)) {
2779 __this_cpu_inc(xmit_recursion
);
2780 rc
= dev_hard_start_xmit(skb
, dev
, txq
);
2781 __this_cpu_dec(xmit_recursion
);
2782 if (dev_xmit_complete(rc
)) {
2783 HARD_TX_UNLOCK(dev
, txq
);
2787 HARD_TX_UNLOCK(dev
, txq
);
2788 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2791 /* Recursion is detected! It is possible,
2795 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2801 rcu_read_unlock_bh();
2806 rcu_read_unlock_bh();
2809 EXPORT_SYMBOL(dev_queue_xmit
);
2812 /*=======================================================================
2814 =======================================================================*/
2816 int netdev_max_backlog __read_mostly
= 1000;
2817 EXPORT_SYMBOL(netdev_max_backlog
);
2819 int netdev_tstamp_prequeue __read_mostly
= 1;
2820 int netdev_budget __read_mostly
= 300;
2821 int weight_p __read_mostly
= 64; /* old backlog weight */
2823 /* Called with irq disabled */
2824 static inline void ____napi_schedule(struct softnet_data
*sd
,
2825 struct napi_struct
*napi
)
2827 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
2828 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2833 /* One global table that all flow-based protocols share. */
2834 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
2835 EXPORT_SYMBOL(rps_sock_flow_table
);
2837 struct static_key rps_needed __read_mostly
;
2839 static struct rps_dev_flow
*
2840 set_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2841 struct rps_dev_flow
*rflow
, u16 next_cpu
)
2843 if (next_cpu
!= RPS_NO_CPU
) {
2844 #ifdef CONFIG_RFS_ACCEL
2845 struct netdev_rx_queue
*rxqueue
;
2846 struct rps_dev_flow_table
*flow_table
;
2847 struct rps_dev_flow
*old_rflow
;
2852 /* Should we steer this flow to a different hardware queue? */
2853 if (!skb_rx_queue_recorded(skb
) || !dev
->rx_cpu_rmap
||
2854 !(dev
->features
& NETIF_F_NTUPLE
))
2856 rxq_index
= cpu_rmap_lookup_index(dev
->rx_cpu_rmap
, next_cpu
);
2857 if (rxq_index
== skb_get_rx_queue(skb
))
2860 rxqueue
= dev
->_rx
+ rxq_index
;
2861 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2864 flow_id
= skb
->rxhash
& flow_table
->mask
;
2865 rc
= dev
->netdev_ops
->ndo_rx_flow_steer(dev
, skb
,
2866 rxq_index
, flow_id
);
2870 rflow
= &flow_table
->flows
[flow_id
];
2872 if (old_rflow
->filter
== rflow
->filter
)
2873 old_rflow
->filter
= RPS_NO_FILTER
;
2877 per_cpu(softnet_data
, next_cpu
).input_queue_head
;
2880 rflow
->cpu
= next_cpu
;
2885 * get_rps_cpu is called from netif_receive_skb and returns the target
2886 * CPU from the RPS map of the receiving queue for a given skb.
2887 * rcu_read_lock must be held on entry.
2889 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2890 struct rps_dev_flow
**rflowp
)
2892 struct netdev_rx_queue
*rxqueue
;
2893 struct rps_map
*map
;
2894 struct rps_dev_flow_table
*flow_table
;
2895 struct rps_sock_flow_table
*sock_flow_table
;
2899 if (skb_rx_queue_recorded(skb
)) {
2900 u16 index
= skb_get_rx_queue(skb
);
2901 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
2902 WARN_ONCE(dev
->real_num_rx_queues
> 1,
2903 "%s received packet on queue %u, but number "
2904 "of RX queues is %u\n",
2905 dev
->name
, index
, dev
->real_num_rx_queues
);
2908 rxqueue
= dev
->_rx
+ index
;
2912 map
= rcu_dereference(rxqueue
->rps_map
);
2914 if (map
->len
== 1 &&
2915 !rcu_access_pointer(rxqueue
->rps_flow_table
)) {
2916 tcpu
= map
->cpus
[0];
2917 if (cpu_online(tcpu
))
2921 } else if (!rcu_access_pointer(rxqueue
->rps_flow_table
)) {
2925 skb_reset_network_header(skb
);
2926 if (!skb_get_rxhash(skb
))
2929 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2930 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
2931 if (flow_table
&& sock_flow_table
) {
2933 struct rps_dev_flow
*rflow
;
2935 rflow
= &flow_table
->flows
[skb
->rxhash
& flow_table
->mask
];
2938 next_cpu
= sock_flow_table
->ents
[skb
->rxhash
&
2939 sock_flow_table
->mask
];
2942 * If the desired CPU (where last recvmsg was done) is
2943 * different from current CPU (one in the rx-queue flow
2944 * table entry), switch if one of the following holds:
2945 * - Current CPU is unset (equal to RPS_NO_CPU).
2946 * - Current CPU is offline.
2947 * - The current CPU's queue tail has advanced beyond the
2948 * last packet that was enqueued using this table entry.
2949 * This guarantees that all previous packets for the flow
2950 * have been dequeued, thus preserving in order delivery.
2952 if (unlikely(tcpu
!= next_cpu
) &&
2953 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
2954 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
2955 rflow
->last_qtail
)) >= 0)) {
2957 rflow
= set_rps_cpu(dev
, skb
, rflow
, next_cpu
);
2960 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
2968 tcpu
= map
->cpus
[((u64
) skb
->rxhash
* map
->len
) >> 32];
2970 if (cpu_online(tcpu
)) {
2980 #ifdef CONFIG_RFS_ACCEL
2983 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2984 * @dev: Device on which the filter was set
2985 * @rxq_index: RX queue index
2986 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2987 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2989 * Drivers that implement ndo_rx_flow_steer() should periodically call
2990 * this function for each installed filter and remove the filters for
2991 * which it returns %true.
2993 bool rps_may_expire_flow(struct net_device
*dev
, u16 rxq_index
,
2994 u32 flow_id
, u16 filter_id
)
2996 struct netdev_rx_queue
*rxqueue
= dev
->_rx
+ rxq_index
;
2997 struct rps_dev_flow_table
*flow_table
;
2998 struct rps_dev_flow
*rflow
;
3003 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3004 if (flow_table
&& flow_id
<= flow_table
->mask
) {
3005 rflow
= &flow_table
->flows
[flow_id
];
3006 cpu
= ACCESS_ONCE(rflow
->cpu
);
3007 if (rflow
->filter
== filter_id
&& cpu
!= RPS_NO_CPU
&&
3008 ((int)(per_cpu(softnet_data
, cpu
).input_queue_head
-
3009 rflow
->last_qtail
) <
3010 (int)(10 * flow_table
->mask
)))
3016 EXPORT_SYMBOL(rps_may_expire_flow
);
3018 #endif /* CONFIG_RFS_ACCEL */
3020 /* Called from hardirq (IPI) context */
3021 static void rps_trigger_softirq(void *data
)
3023 struct softnet_data
*sd
= data
;
3025 ____napi_schedule(sd
, &sd
->backlog
);
3029 #endif /* CONFIG_RPS */
3032 * Check if this softnet_data structure is another cpu one
3033 * If yes, queue it to our IPI list and return 1
3036 static int rps_ipi_queued(struct softnet_data
*sd
)
3039 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
3042 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
3043 mysd
->rps_ipi_list
= sd
;
3045 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3048 #endif /* CONFIG_RPS */
3053 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3054 * queue (may be a remote CPU queue).
3056 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
3057 unsigned int *qtail
)
3059 struct softnet_data
*sd
;
3060 unsigned long flags
;
3062 sd
= &per_cpu(softnet_data
, cpu
);
3064 local_irq_save(flags
);
3067 if (skb_queue_len(&sd
->input_pkt_queue
) <= netdev_max_backlog
) {
3068 if (skb_queue_len(&sd
->input_pkt_queue
)) {
3070 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
3071 input_queue_tail_incr_save(sd
, qtail
);
3073 local_irq_restore(flags
);
3074 return NET_RX_SUCCESS
;
3077 /* Schedule NAPI for backlog device
3078 * We can use non atomic operation since we own the queue lock
3080 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
3081 if (!rps_ipi_queued(sd
))
3082 ____napi_schedule(sd
, &sd
->backlog
);
3090 local_irq_restore(flags
);
3092 atomic_long_inc(&skb
->dev
->rx_dropped
);
3098 * netif_rx - post buffer to the network code
3099 * @skb: buffer to post
3101 * This function receives a packet from a device driver and queues it for
3102 * the upper (protocol) levels to process. It always succeeds. The buffer
3103 * may be dropped during processing for congestion control or by the
3107 * NET_RX_SUCCESS (no congestion)
3108 * NET_RX_DROP (packet was dropped)
3112 int netif_rx(struct sk_buff
*skb
)
3116 /* if netpoll wants it, pretend we never saw it */
3117 if (netpoll_rx(skb
))
3120 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3122 trace_netif_rx(skb
);
3124 if (static_key_false(&rps_needed
)) {
3125 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3131 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3133 cpu
= smp_processor_id();
3135 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3143 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
3148 EXPORT_SYMBOL(netif_rx
);
3150 int netif_rx_ni(struct sk_buff
*skb
)
3155 err
= netif_rx(skb
);
3156 if (local_softirq_pending())
3162 EXPORT_SYMBOL(netif_rx_ni
);
3164 static void net_tx_action(struct softirq_action
*h
)
3166 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3168 if (sd
->completion_queue
) {
3169 struct sk_buff
*clist
;
3171 local_irq_disable();
3172 clist
= sd
->completion_queue
;
3173 sd
->completion_queue
= NULL
;
3177 struct sk_buff
*skb
= clist
;
3178 clist
= clist
->next
;
3180 WARN_ON(atomic_read(&skb
->users
));
3181 trace_kfree_skb(skb
, net_tx_action
);
3186 if (sd
->output_queue
) {
3189 local_irq_disable();
3190 head
= sd
->output_queue
;
3191 sd
->output_queue
= NULL
;
3192 sd
->output_queue_tailp
= &sd
->output_queue
;
3196 struct Qdisc
*q
= head
;
3197 spinlock_t
*root_lock
;
3199 head
= head
->next_sched
;
3201 root_lock
= qdisc_lock(q
);
3202 if (spin_trylock(root_lock
)) {
3203 smp_mb__before_clear_bit();
3204 clear_bit(__QDISC_STATE_SCHED
,
3207 spin_unlock(root_lock
);
3209 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
3211 __netif_reschedule(q
);
3213 smp_mb__before_clear_bit();
3214 clear_bit(__QDISC_STATE_SCHED
,
3222 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3223 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3224 /* This hook is defined here for ATM LANE */
3225 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
3226 unsigned char *addr
) __read_mostly
;
3227 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
3230 #ifdef CONFIG_NET_CLS_ACT
3231 /* TODO: Maybe we should just force sch_ingress to be compiled in
3232 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3233 * a compare and 2 stores extra right now if we dont have it on
3234 * but have CONFIG_NET_CLS_ACT
3235 * NOTE: This doesn't stop any functionality; if you dont have
3236 * the ingress scheduler, you just can't add policies on ingress.
3239 static int ing_filter(struct sk_buff
*skb
, struct netdev_queue
*rxq
)
3241 struct net_device
*dev
= skb
->dev
;
3242 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
3243 int result
= TC_ACT_OK
;
3246 if (unlikely(MAX_RED_LOOP
< ttl
++)) {
3247 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3248 skb
->skb_iif
, dev
->ifindex
);
3252 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
3253 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
3256 if (q
!= &noop_qdisc
) {
3257 spin_lock(qdisc_lock(q
));
3258 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
3259 result
= qdisc_enqueue_root(skb
, q
);
3260 spin_unlock(qdisc_lock(q
));
3266 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
3267 struct packet_type
**pt_prev
,
3268 int *ret
, struct net_device
*orig_dev
)
3270 struct netdev_queue
*rxq
= rcu_dereference(skb
->dev
->ingress_queue
);
3272 if (!rxq
|| rxq
->qdisc
== &noop_qdisc
)
3276 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
3280 switch (ing_filter(skb
, rxq
)) {
3294 * netdev_rx_handler_register - register receive handler
3295 * @dev: device to register a handler for
3296 * @rx_handler: receive handler to register
3297 * @rx_handler_data: data pointer that is used by rx handler
3299 * Register a receive hander for a device. This handler will then be
3300 * called from __netif_receive_skb. A negative errno code is returned
3303 * The caller must hold the rtnl_mutex.
3305 * For a general description of rx_handler, see enum rx_handler_result.
3307 int netdev_rx_handler_register(struct net_device
*dev
,
3308 rx_handler_func_t
*rx_handler
,
3309 void *rx_handler_data
)
3313 if (dev
->rx_handler
)
3316 /* Note: rx_handler_data must be set before rx_handler */
3317 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
3318 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
3322 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
3325 * netdev_rx_handler_unregister - unregister receive handler
3326 * @dev: device to unregister a handler from
3328 * Unregister a receive hander from a device.
3330 * The caller must hold the rtnl_mutex.
3332 void netdev_rx_handler_unregister(struct net_device
*dev
)
3336 RCU_INIT_POINTER(dev
->rx_handler
, NULL
);
3337 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3338 * section has a guarantee to see a non NULL rx_handler_data
3342 RCU_INIT_POINTER(dev
->rx_handler_data
, NULL
);
3344 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
3347 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3348 * the special handling of PFMEMALLOC skbs.
3350 static bool skb_pfmemalloc_protocol(struct sk_buff
*skb
)
3352 switch (skb
->protocol
) {
3353 case __constant_htons(ETH_P_ARP
):
3354 case __constant_htons(ETH_P_IP
):
3355 case __constant_htons(ETH_P_IPV6
):
3356 case __constant_htons(ETH_P_8021Q
):
3363 static int __netif_receive_skb_core(struct sk_buff
*skb
, bool pfmemalloc
)
3365 struct packet_type
*ptype
, *pt_prev
;
3366 rx_handler_func_t
*rx_handler
;
3367 struct net_device
*orig_dev
;
3368 struct net_device
*null_or_dev
;
3369 bool deliver_exact
= false;
3370 int ret
= NET_RX_DROP
;
3373 net_timestamp_check(!netdev_tstamp_prequeue
, skb
);
3375 trace_netif_receive_skb(skb
);
3377 /* if we've gotten here through NAPI, check netpoll */
3378 if (netpoll_receive_skb(skb
))
3381 orig_dev
= skb
->dev
;
3383 skb_reset_network_header(skb
);
3384 if (!skb_transport_header_was_set(skb
))
3385 skb_reset_transport_header(skb
);
3386 skb_reset_mac_len(skb
);
3393 skb
->skb_iif
= skb
->dev
->ifindex
;
3395 __this_cpu_inc(softnet_data
.processed
);
3397 if (skb
->protocol
== cpu_to_be16(ETH_P_8021Q
)) {
3398 skb
= vlan_untag(skb
);
3403 #ifdef CONFIG_NET_CLS_ACT
3404 if (skb
->tc_verd
& TC_NCLS
) {
3405 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
3413 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
3414 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
3416 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3422 #ifdef CONFIG_NET_CLS_ACT
3423 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
3429 if (pfmemalloc
&& !skb_pfmemalloc_protocol(skb
))
3432 if (vlan_tx_tag_present(skb
)) {
3434 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3437 if (vlan_do_receive(&skb
))
3439 else if (unlikely(!skb
))
3443 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
3446 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3449 switch (rx_handler(&skb
)) {
3450 case RX_HANDLER_CONSUMED
:
3451 ret
= NET_RX_SUCCESS
;
3453 case RX_HANDLER_ANOTHER
:
3455 case RX_HANDLER_EXACT
:
3456 deliver_exact
= true;
3457 case RX_HANDLER_PASS
:
3464 if (vlan_tx_nonzero_tag_present(skb
))
3465 skb
->pkt_type
= PACKET_OTHERHOST
;
3467 /* deliver only exact match when indicated */
3468 null_or_dev
= deliver_exact
? skb
->dev
: NULL
;
3470 type
= skb
->protocol
;
3471 list_for_each_entry_rcu(ptype
,
3472 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
3473 if (ptype
->type
== type
&&
3474 (ptype
->dev
== null_or_dev
|| ptype
->dev
== skb
->dev
||
3475 ptype
->dev
== orig_dev
)) {
3477 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3483 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
3486 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
3489 atomic_long_inc(&skb
->dev
->rx_dropped
);
3491 /* Jamal, now you will not able to escape explaining
3492 * me how you were going to use this. :-)
3503 static int __netif_receive_skb(struct sk_buff
*skb
)
3507 if (sk_memalloc_socks() && skb_pfmemalloc(skb
)) {
3508 unsigned long pflags
= current
->flags
;
3511 * PFMEMALLOC skbs are special, they should
3512 * - be delivered to SOCK_MEMALLOC sockets only
3513 * - stay away from userspace
3514 * - have bounded memory usage
3516 * Use PF_MEMALLOC as this saves us from propagating the allocation
3517 * context down to all allocation sites.
3519 current
->flags
|= PF_MEMALLOC
;
3520 ret
= __netif_receive_skb_core(skb
, true);
3521 tsk_restore_flags(current
, pflags
, PF_MEMALLOC
);
3523 ret
= __netif_receive_skb_core(skb
, false);
3529 * netif_receive_skb - process receive buffer from network
3530 * @skb: buffer to process
3532 * netif_receive_skb() is the main receive data processing function.
3533 * It always succeeds. The buffer may be dropped during processing
3534 * for congestion control or by the protocol layers.
3536 * This function may only be called from softirq context and interrupts
3537 * should be enabled.
3539 * Return values (usually ignored):
3540 * NET_RX_SUCCESS: no congestion
3541 * NET_RX_DROP: packet was dropped
3543 int netif_receive_skb(struct sk_buff
*skb
)
3545 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3547 if (skb_defer_rx_timestamp(skb
))
3548 return NET_RX_SUCCESS
;
3551 if (static_key_false(&rps_needed
)) {
3552 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3557 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3560 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3567 return __netif_receive_skb(skb
);
3569 EXPORT_SYMBOL(netif_receive_skb
);
3571 /* Network device is going away, flush any packets still pending
3572 * Called with irqs disabled.
3574 static void flush_backlog(void *arg
)
3576 struct net_device
*dev
= arg
;
3577 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3578 struct sk_buff
*skb
, *tmp
;
3581 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3582 if (skb
->dev
== dev
) {
3583 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3585 input_queue_head_incr(sd
);
3590 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3591 if (skb
->dev
== dev
) {
3592 __skb_unlink(skb
, &sd
->process_queue
);
3594 input_queue_head_incr(sd
);
3599 static int napi_gro_complete(struct sk_buff
*skb
)
3601 struct packet_offload
*ptype
;
3602 __be16 type
= skb
->protocol
;
3603 struct list_head
*head
= &offload_base
;
3606 BUILD_BUG_ON(sizeof(struct napi_gro_cb
) > sizeof(skb
->cb
));
3608 if (NAPI_GRO_CB(skb
)->count
== 1) {
3609 skb_shinfo(skb
)->gso_size
= 0;
3614 list_for_each_entry_rcu(ptype
, head
, list
) {
3615 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_complete
)
3618 err
= ptype
->callbacks
.gro_complete(skb
);
3624 WARN_ON(&ptype
->list
== head
);
3626 return NET_RX_SUCCESS
;
3630 return netif_receive_skb(skb
);
3633 /* napi->gro_list contains packets ordered by age.
3634 * youngest packets at the head of it.
3635 * Complete skbs in reverse order to reduce latencies.
3637 void napi_gro_flush(struct napi_struct
*napi
, bool flush_old
)
3639 struct sk_buff
*skb
, *prev
= NULL
;
3641 /* scan list and build reverse chain */
3642 for (skb
= napi
->gro_list
; skb
!= NULL
; skb
= skb
->next
) {
3647 for (skb
= prev
; skb
; skb
= prev
) {
3650 if (flush_old
&& NAPI_GRO_CB(skb
)->age
== jiffies
)
3654 napi_gro_complete(skb
);
3658 napi
->gro_list
= NULL
;
3660 EXPORT_SYMBOL(napi_gro_flush
);
3662 static void gro_list_prepare(struct napi_struct
*napi
, struct sk_buff
*skb
)
3665 unsigned int maclen
= skb
->dev
->hard_header_len
;
3667 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3668 unsigned long diffs
;
3670 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
3671 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
3672 if (maclen
== ETH_HLEN
)
3673 diffs
|= compare_ether_header(skb_mac_header(p
),
3674 skb_gro_mac_header(skb
));
3676 diffs
= memcmp(skb_mac_header(p
),
3677 skb_gro_mac_header(skb
),
3679 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
3680 NAPI_GRO_CB(p
)->flush
= 0;
3684 static enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3686 struct sk_buff
**pp
= NULL
;
3687 struct packet_offload
*ptype
;
3688 __be16 type
= skb
->protocol
;
3689 struct list_head
*head
= &offload_base
;
3691 enum gro_result ret
;
3693 if (!(skb
->dev
->features
& NETIF_F_GRO
) || netpoll_rx_on(skb
))
3696 if (skb_is_gso(skb
) || skb_has_frag_list(skb
))
3699 gro_list_prepare(napi
, skb
);
3702 list_for_each_entry_rcu(ptype
, head
, list
) {
3703 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_receive
)
3706 skb_set_network_header(skb
, skb_gro_offset(skb
));
3707 skb_reset_mac_len(skb
);
3708 NAPI_GRO_CB(skb
)->same_flow
= 0;
3709 NAPI_GRO_CB(skb
)->flush
= 0;
3710 NAPI_GRO_CB(skb
)->free
= 0;
3712 pp
= ptype
->callbacks
.gro_receive(&napi
->gro_list
, skb
);
3717 if (&ptype
->list
== head
)
3720 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3721 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3724 struct sk_buff
*nskb
= *pp
;
3728 napi_gro_complete(nskb
);
3735 if (NAPI_GRO_CB(skb
)->flush
|| napi
->gro_count
>= MAX_GRO_SKBS
)
3739 NAPI_GRO_CB(skb
)->count
= 1;
3740 NAPI_GRO_CB(skb
)->age
= jiffies
;
3741 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
3742 skb
->next
= napi
->gro_list
;
3743 napi
->gro_list
= skb
;
3747 if (skb_headlen(skb
) < skb_gro_offset(skb
)) {
3748 int grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
3750 BUG_ON(skb
->end
- skb
->tail
< grow
);
3752 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3755 skb
->data_len
-= grow
;
3757 skb_shinfo(skb
)->frags
[0].page_offset
+= grow
;
3758 skb_frag_size_sub(&skb_shinfo(skb
)->frags
[0], grow
);
3760 if (unlikely(!skb_frag_size(&skb_shinfo(skb
)->frags
[0]))) {
3761 skb_frag_unref(skb
, 0);
3762 memmove(skb_shinfo(skb
)->frags
,
3763 skb_shinfo(skb
)->frags
+ 1,
3764 --skb_shinfo(skb
)->nr_frags
* sizeof(skb_frag_t
));
3777 static gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
3781 if (netif_receive_skb(skb
))
3789 case GRO_MERGED_FREE
:
3790 if (NAPI_GRO_CB(skb
)->free
== NAPI_GRO_FREE_STOLEN_HEAD
)
3791 kmem_cache_free(skbuff_head_cache
, skb
);
3804 static void skb_gro_reset_offset(struct sk_buff
*skb
)
3806 const struct skb_shared_info
*pinfo
= skb_shinfo(skb
);
3807 const skb_frag_t
*frag0
= &pinfo
->frags
[0];
3809 NAPI_GRO_CB(skb
)->data_offset
= 0;
3810 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3811 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3813 if (skb
->mac_header
== skb
->tail
&&
3815 !PageHighMem(skb_frag_page(frag0
))) {
3816 NAPI_GRO_CB(skb
)->frag0
= skb_frag_address(frag0
);
3817 NAPI_GRO_CB(skb
)->frag0_len
= skb_frag_size(frag0
);
3821 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3823 skb_gro_reset_offset(skb
);
3825 return napi_skb_finish(dev_gro_receive(napi
, skb
), skb
);
3827 EXPORT_SYMBOL(napi_gro_receive
);
3829 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
3831 __skb_pull(skb
, skb_headlen(skb
));
3832 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3833 skb_reserve(skb
, NET_SKB_PAD
+ NET_IP_ALIGN
- skb_headroom(skb
));
3835 skb
->dev
= napi
->dev
;
3841 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
3843 struct sk_buff
*skb
= napi
->skb
;
3846 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
3852 EXPORT_SYMBOL(napi_get_frags
);
3854 static gro_result_t
napi_frags_finish(struct napi_struct
*napi
, struct sk_buff
*skb
,
3860 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
3862 if (ret
== GRO_HELD
)
3863 skb_gro_pull(skb
, -ETH_HLEN
);
3864 else if (netif_receive_skb(skb
))
3869 case GRO_MERGED_FREE
:
3870 napi_reuse_skb(napi
, skb
);
3880 static struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
3882 struct sk_buff
*skb
= napi
->skb
;
3889 skb_reset_mac_header(skb
);
3890 skb_gro_reset_offset(skb
);
3892 off
= skb_gro_offset(skb
);
3893 hlen
= off
+ sizeof(*eth
);
3894 eth
= skb_gro_header_fast(skb
, off
);
3895 if (skb_gro_header_hard(skb
, hlen
)) {
3896 eth
= skb_gro_header_slow(skb
, hlen
, off
);
3897 if (unlikely(!eth
)) {
3898 napi_reuse_skb(napi
, skb
);
3904 skb_gro_pull(skb
, sizeof(*eth
));
3907 * This works because the only protocols we care about don't require
3908 * special handling. We'll fix it up properly at the end.
3910 skb
->protocol
= eth
->h_proto
;
3916 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
3918 struct sk_buff
*skb
= napi_frags_skb(napi
);
3923 return napi_frags_finish(napi
, skb
, dev_gro_receive(napi
, skb
));
3925 EXPORT_SYMBOL(napi_gro_frags
);
3928 * net_rps_action sends any pending IPI's for rps.
3929 * Note: called with local irq disabled, but exits with local irq enabled.
3931 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
3934 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
3937 sd
->rps_ipi_list
= NULL
;
3941 /* Send pending IPI's to kick RPS processing on remote cpus. */
3943 struct softnet_data
*next
= remsd
->rps_ipi_next
;
3945 if (cpu_online(remsd
->cpu
))
3946 __smp_call_function_single(remsd
->cpu
,
3955 static int process_backlog(struct napi_struct
*napi
, int quota
)
3958 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
3961 /* Check if we have pending ipi, its better to send them now,
3962 * not waiting net_rx_action() end.
3964 if (sd
->rps_ipi_list
) {
3965 local_irq_disable();
3966 net_rps_action_and_irq_enable(sd
);
3969 napi
->weight
= weight_p
;
3970 local_irq_disable();
3971 while (work
< quota
) {
3972 struct sk_buff
*skb
;
3975 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
3977 __netif_receive_skb(skb
);
3978 local_irq_disable();
3979 input_queue_head_incr(sd
);
3980 if (++work
>= quota
) {
3987 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3989 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
3990 &sd
->process_queue
);
3992 if (qlen
< quota
- work
) {
3994 * Inline a custom version of __napi_complete().
3995 * only current cpu owns and manipulates this napi,
3996 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3997 * we can use a plain write instead of clear_bit(),
3998 * and we dont need an smp_mb() memory barrier.
4000 list_del(&napi
->poll_list
);
4003 quota
= work
+ qlen
;
4013 * __napi_schedule - schedule for receive
4014 * @n: entry to schedule
4016 * The entry's receive function will be scheduled to run
4018 void __napi_schedule(struct napi_struct
*n
)
4020 unsigned long flags
;
4022 local_irq_save(flags
);
4023 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
4024 local_irq_restore(flags
);
4026 EXPORT_SYMBOL(__napi_schedule
);
4028 void __napi_complete(struct napi_struct
*n
)
4030 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
4031 BUG_ON(n
->gro_list
);
4033 list_del(&n
->poll_list
);
4034 smp_mb__before_clear_bit();
4035 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
4037 EXPORT_SYMBOL(__napi_complete
);
4039 void napi_complete(struct napi_struct
*n
)
4041 unsigned long flags
;
4044 * don't let napi dequeue from the cpu poll list
4045 * just in case its running on a different cpu
4047 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
4050 napi_gro_flush(n
, false);
4051 local_irq_save(flags
);
4053 local_irq_restore(flags
);
4055 EXPORT_SYMBOL(napi_complete
);
4057 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
4058 int (*poll
)(struct napi_struct
*, int), int weight
)
4060 INIT_LIST_HEAD(&napi
->poll_list
);
4061 napi
->gro_count
= 0;
4062 napi
->gro_list
= NULL
;
4065 napi
->weight
= weight
;
4066 list_add(&napi
->dev_list
, &dev
->napi_list
);
4068 #ifdef CONFIG_NETPOLL
4069 spin_lock_init(&napi
->poll_lock
);
4070 napi
->poll_owner
= -1;
4072 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
4074 EXPORT_SYMBOL(netif_napi_add
);
4076 void netif_napi_del(struct napi_struct
*napi
)
4078 struct sk_buff
*skb
, *next
;
4080 list_del_init(&napi
->dev_list
);
4081 napi_free_frags(napi
);
4083 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
4089 napi
->gro_list
= NULL
;
4090 napi
->gro_count
= 0;
4092 EXPORT_SYMBOL(netif_napi_del
);
4094 static void net_rx_action(struct softirq_action
*h
)
4096 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
4097 unsigned long time_limit
= jiffies
+ 2;
4098 int budget
= netdev_budget
;
4101 local_irq_disable();
4103 while (!list_empty(&sd
->poll_list
)) {
4104 struct napi_struct
*n
;
4107 /* If softirq window is exhuasted then punt.
4108 * Allow this to run for 2 jiffies since which will allow
4109 * an average latency of 1.5/HZ.
4111 if (unlikely(budget
<= 0 || time_after_eq(jiffies
, time_limit
)))
4116 /* Even though interrupts have been re-enabled, this
4117 * access is safe because interrupts can only add new
4118 * entries to the tail of this list, and only ->poll()
4119 * calls can remove this head entry from the list.
4121 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
4123 have
= netpoll_poll_lock(n
);
4127 /* This NAPI_STATE_SCHED test is for avoiding a race
4128 * with netpoll's poll_napi(). Only the entity which
4129 * obtains the lock and sees NAPI_STATE_SCHED set will
4130 * actually make the ->poll() call. Therefore we avoid
4131 * accidentally calling ->poll() when NAPI is not scheduled.
4134 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
4135 work
= n
->poll(n
, weight
);
4139 WARN_ON_ONCE(work
> weight
);
4143 local_irq_disable();
4145 /* Drivers must not modify the NAPI state if they
4146 * consume the entire weight. In such cases this code
4147 * still "owns" the NAPI instance and therefore can
4148 * move the instance around on the list at-will.
4150 if (unlikely(work
== weight
)) {
4151 if (unlikely(napi_disable_pending(n
))) {
4154 local_irq_disable();
4157 /* flush too old packets
4158 * If HZ < 1000, flush all packets.
4161 napi_gro_flush(n
, HZ
>= 1000);
4162 local_irq_disable();
4164 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
4168 netpoll_poll_unlock(have
);
4171 net_rps_action_and_irq_enable(sd
);
4173 #ifdef CONFIG_NET_DMA
4175 * There may not be any more sk_buffs coming right now, so push
4176 * any pending DMA copies to hardware
4178 dma_issue_pending_all();
4185 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
4189 struct netdev_upper
{
4190 struct net_device
*dev
;
4192 struct list_head list
;
4193 struct rcu_head rcu
;
4194 struct list_head search_list
;
4197 static void __append_search_uppers(struct list_head
*search_list
,
4198 struct net_device
*dev
)
4200 struct netdev_upper
*upper
;
4202 list_for_each_entry(upper
, &dev
->upper_dev_list
, list
) {
4203 /* check if this upper is not already in search list */
4204 if (list_empty(&upper
->search_list
))
4205 list_add_tail(&upper
->search_list
, search_list
);
4209 static bool __netdev_search_upper_dev(struct net_device
*dev
,
4210 struct net_device
*upper_dev
)
4212 LIST_HEAD(search_list
);
4213 struct netdev_upper
*upper
;
4214 struct netdev_upper
*tmp
;
4217 __append_search_uppers(&search_list
, dev
);
4218 list_for_each_entry(upper
, &search_list
, search_list
) {
4219 if (upper
->dev
== upper_dev
) {
4223 __append_search_uppers(&search_list
, upper
->dev
);
4225 list_for_each_entry_safe(upper
, tmp
, &search_list
, search_list
)
4226 INIT_LIST_HEAD(&upper
->search_list
);
4230 static struct netdev_upper
*__netdev_find_upper(struct net_device
*dev
,
4231 struct net_device
*upper_dev
)
4233 struct netdev_upper
*upper
;
4235 list_for_each_entry(upper
, &dev
->upper_dev_list
, list
) {
4236 if (upper
->dev
== upper_dev
)
4243 * netdev_has_upper_dev - Check if device is linked to an upper device
4245 * @upper_dev: upper device to check
4247 * Find out if a device is linked to specified upper device and return true
4248 * in case it is. Note that this checks only immediate upper device,
4249 * not through a complete stack of devices. The caller must hold the RTNL lock.
4251 bool netdev_has_upper_dev(struct net_device
*dev
,
4252 struct net_device
*upper_dev
)
4256 return __netdev_find_upper(dev
, upper_dev
);
4258 EXPORT_SYMBOL(netdev_has_upper_dev
);
4261 * netdev_has_any_upper_dev - Check if device is linked to some device
4264 * Find out if a device is linked to an upper device and return true in case
4265 * it is. The caller must hold the RTNL lock.
4267 bool netdev_has_any_upper_dev(struct net_device
*dev
)
4271 return !list_empty(&dev
->upper_dev_list
);
4273 EXPORT_SYMBOL(netdev_has_any_upper_dev
);
4276 * netdev_master_upper_dev_get - Get master upper device
4279 * Find a master upper device and return pointer to it or NULL in case
4280 * it's not there. The caller must hold the RTNL lock.
4282 struct net_device
*netdev_master_upper_dev_get(struct net_device
*dev
)
4284 struct netdev_upper
*upper
;
4288 if (list_empty(&dev
->upper_dev_list
))
4291 upper
= list_first_entry(&dev
->upper_dev_list
,
4292 struct netdev_upper
, list
);
4293 if (likely(upper
->master
))
4297 EXPORT_SYMBOL(netdev_master_upper_dev_get
);
4300 * netdev_master_upper_dev_get_rcu - Get master upper device
4303 * Find a master upper device and return pointer to it or NULL in case
4304 * it's not there. The caller must hold the RCU read lock.
4306 struct net_device
*netdev_master_upper_dev_get_rcu(struct net_device
*dev
)
4308 struct netdev_upper
*upper
;
4310 upper
= list_first_or_null_rcu(&dev
->upper_dev_list
,
4311 struct netdev_upper
, list
);
4312 if (upper
&& likely(upper
->master
))
4316 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu
);
4318 static int __netdev_upper_dev_link(struct net_device
*dev
,
4319 struct net_device
*upper_dev
, bool master
)
4321 struct netdev_upper
*upper
;
4325 if (dev
== upper_dev
)
4328 /* To prevent loops, check if dev is not upper device to upper_dev. */
4329 if (__netdev_search_upper_dev(upper_dev
, dev
))
4332 if (__netdev_find_upper(dev
, upper_dev
))
4335 if (master
&& netdev_master_upper_dev_get(dev
))
4338 upper
= kmalloc(sizeof(*upper
), GFP_KERNEL
);
4342 upper
->dev
= upper_dev
;
4343 upper
->master
= master
;
4344 INIT_LIST_HEAD(&upper
->search_list
);
4346 /* Ensure that master upper link is always the first item in list. */
4348 list_add_rcu(&upper
->list
, &dev
->upper_dev_list
);
4350 list_add_tail_rcu(&upper
->list
, &dev
->upper_dev_list
);
4351 dev_hold(upper_dev
);
4357 * netdev_upper_dev_link - Add a link to the upper device
4359 * @upper_dev: new upper device
4361 * Adds a link to device which is upper to this one. The caller must hold
4362 * the RTNL lock. On a failure a negative errno code is returned.
4363 * On success the reference counts are adjusted and the function
4366 int netdev_upper_dev_link(struct net_device
*dev
,
4367 struct net_device
*upper_dev
)
4369 return __netdev_upper_dev_link(dev
, upper_dev
, false);
4371 EXPORT_SYMBOL(netdev_upper_dev_link
);
4374 * netdev_master_upper_dev_link - Add a master link to the upper device
4376 * @upper_dev: new upper device
4378 * Adds a link to device which is upper to this one. In this case, only
4379 * one master upper device can be linked, although other non-master devices
4380 * might be linked as well. The caller must hold the RTNL lock.
4381 * On a failure a negative errno code is returned. On success the reference
4382 * counts are adjusted and the function returns zero.
4384 int netdev_master_upper_dev_link(struct net_device
*dev
,
4385 struct net_device
*upper_dev
)
4387 return __netdev_upper_dev_link(dev
, upper_dev
, true);
4389 EXPORT_SYMBOL(netdev_master_upper_dev_link
);
4392 * netdev_upper_dev_unlink - Removes a link to upper device
4394 * @upper_dev: new upper device
4396 * Removes a link to device which is upper to this one. The caller must hold
4399 void netdev_upper_dev_unlink(struct net_device
*dev
,
4400 struct net_device
*upper_dev
)
4402 struct netdev_upper
*upper
;
4406 upper
= __netdev_find_upper(dev
, upper_dev
);
4409 list_del_rcu(&upper
->list
);
4411 kfree_rcu(upper
, rcu
);
4413 EXPORT_SYMBOL(netdev_upper_dev_unlink
);
4415 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
4417 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4419 if ((dev
->flags
& IFF_UP
) && ops
->ndo_change_rx_flags
)
4420 ops
->ndo_change_rx_flags(dev
, flags
);
4423 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
4425 unsigned int old_flags
= dev
->flags
;
4431 dev
->flags
|= IFF_PROMISC
;
4432 dev
->promiscuity
+= inc
;
4433 if (dev
->promiscuity
== 0) {
4436 * If inc causes overflow, untouch promisc and return error.
4439 dev
->flags
&= ~IFF_PROMISC
;
4441 dev
->promiscuity
-= inc
;
4442 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4447 if (dev
->flags
!= old_flags
) {
4448 pr_info("device %s %s promiscuous mode\n",
4450 dev
->flags
& IFF_PROMISC
? "entered" : "left");
4451 if (audit_enabled
) {
4452 current_uid_gid(&uid
, &gid
);
4453 audit_log(current
->audit_context
, GFP_ATOMIC
,
4454 AUDIT_ANOM_PROMISCUOUS
,
4455 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4456 dev
->name
, (dev
->flags
& IFF_PROMISC
),
4457 (old_flags
& IFF_PROMISC
),
4458 from_kuid(&init_user_ns
, audit_get_loginuid(current
)),
4459 from_kuid(&init_user_ns
, uid
),
4460 from_kgid(&init_user_ns
, gid
),
4461 audit_get_sessionid(current
));
4464 dev_change_rx_flags(dev
, IFF_PROMISC
);
4470 * dev_set_promiscuity - update promiscuity count on a device
4474 * Add or remove promiscuity from a device. While the count in the device
4475 * remains above zero the interface remains promiscuous. Once it hits zero
4476 * the device reverts back to normal filtering operation. A negative inc
4477 * value is used to drop promiscuity on the device.
4478 * Return 0 if successful or a negative errno code on error.
4480 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
4482 unsigned int old_flags
= dev
->flags
;
4485 err
= __dev_set_promiscuity(dev
, inc
);
4488 if (dev
->flags
!= old_flags
)
4489 dev_set_rx_mode(dev
);
4492 EXPORT_SYMBOL(dev_set_promiscuity
);
4495 * dev_set_allmulti - update allmulti count on a device
4499 * Add or remove reception of all multicast frames to a device. While the
4500 * count in the device remains above zero the interface remains listening
4501 * to all interfaces. Once it hits zero the device reverts back to normal
4502 * filtering operation. A negative @inc value is used to drop the counter
4503 * when releasing a resource needing all multicasts.
4504 * Return 0 if successful or a negative errno code on error.
4507 int dev_set_allmulti(struct net_device
*dev
, int inc
)
4509 unsigned int old_flags
= dev
->flags
;
4513 dev
->flags
|= IFF_ALLMULTI
;
4514 dev
->allmulti
+= inc
;
4515 if (dev
->allmulti
== 0) {
4518 * If inc causes overflow, untouch allmulti and return error.
4521 dev
->flags
&= ~IFF_ALLMULTI
;
4523 dev
->allmulti
-= inc
;
4524 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4529 if (dev
->flags
^ old_flags
) {
4530 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
4531 dev_set_rx_mode(dev
);
4535 EXPORT_SYMBOL(dev_set_allmulti
);
4538 * Upload unicast and multicast address lists to device and
4539 * configure RX filtering. When the device doesn't support unicast
4540 * filtering it is put in promiscuous mode while unicast addresses
4543 void __dev_set_rx_mode(struct net_device
*dev
)
4545 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4547 /* dev_open will call this function so the list will stay sane. */
4548 if (!(dev
->flags
&IFF_UP
))
4551 if (!netif_device_present(dev
))
4554 if (!(dev
->priv_flags
& IFF_UNICAST_FLT
)) {
4555 /* Unicast addresses changes may only happen under the rtnl,
4556 * therefore calling __dev_set_promiscuity here is safe.
4558 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
4559 __dev_set_promiscuity(dev
, 1);
4560 dev
->uc_promisc
= true;
4561 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
4562 __dev_set_promiscuity(dev
, -1);
4563 dev
->uc_promisc
= false;
4567 if (ops
->ndo_set_rx_mode
)
4568 ops
->ndo_set_rx_mode(dev
);
4571 void dev_set_rx_mode(struct net_device
*dev
)
4573 netif_addr_lock_bh(dev
);
4574 __dev_set_rx_mode(dev
);
4575 netif_addr_unlock_bh(dev
);
4579 * dev_get_flags - get flags reported to userspace
4582 * Get the combination of flag bits exported through APIs to userspace.
4584 unsigned int dev_get_flags(const struct net_device
*dev
)
4588 flags
= (dev
->flags
& ~(IFF_PROMISC
|
4593 (dev
->gflags
& (IFF_PROMISC
|
4596 if (netif_running(dev
)) {
4597 if (netif_oper_up(dev
))
4598 flags
|= IFF_RUNNING
;
4599 if (netif_carrier_ok(dev
))
4600 flags
|= IFF_LOWER_UP
;
4601 if (netif_dormant(dev
))
4602 flags
|= IFF_DORMANT
;
4607 EXPORT_SYMBOL(dev_get_flags
);
4609 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4611 unsigned int old_flags
= dev
->flags
;
4617 * Set the flags on our device.
4620 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
4621 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
4623 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
4627 * Load in the correct multicast list now the flags have changed.
4630 if ((old_flags
^ flags
) & IFF_MULTICAST
)
4631 dev_change_rx_flags(dev
, IFF_MULTICAST
);
4633 dev_set_rx_mode(dev
);
4636 * Have we downed the interface. We handle IFF_UP ourselves
4637 * according to user attempts to set it, rather than blindly
4642 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
4643 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
4646 dev_set_rx_mode(dev
);
4649 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
4650 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
4652 dev
->gflags
^= IFF_PROMISC
;
4653 dev_set_promiscuity(dev
, inc
);
4656 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4657 is important. Some (broken) drivers set IFF_PROMISC, when
4658 IFF_ALLMULTI is requested not asking us and not reporting.
4660 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
4661 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
4663 dev
->gflags
^= IFF_ALLMULTI
;
4664 dev_set_allmulti(dev
, inc
);
4670 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
)
4672 unsigned int changes
= dev
->flags
^ old_flags
;
4674 if (changes
& IFF_UP
) {
4675 if (dev
->flags
& IFF_UP
)
4676 call_netdevice_notifiers(NETDEV_UP
, dev
);
4678 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
4681 if (dev
->flags
& IFF_UP
&&
4682 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
)))
4683 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
4687 * dev_change_flags - change device settings
4689 * @flags: device state flags
4691 * Change settings on device based state flags. The flags are
4692 * in the userspace exported format.
4694 int dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4697 unsigned int changes
, old_flags
= dev
->flags
;
4699 ret
= __dev_change_flags(dev
, flags
);
4703 changes
= old_flags
^ dev
->flags
;
4705 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
4707 __dev_notify_flags(dev
, old_flags
);
4710 EXPORT_SYMBOL(dev_change_flags
);
4713 * dev_set_mtu - Change maximum transfer unit
4715 * @new_mtu: new transfer unit
4717 * Change the maximum transfer size of the network device.
4719 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
4721 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4724 if (new_mtu
== dev
->mtu
)
4727 /* MTU must be positive. */
4731 if (!netif_device_present(dev
))
4735 if (ops
->ndo_change_mtu
)
4736 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
4741 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
4744 EXPORT_SYMBOL(dev_set_mtu
);
4747 * dev_set_group - Change group this device belongs to
4749 * @new_group: group this device should belong to
4751 void dev_set_group(struct net_device
*dev
, int new_group
)
4753 dev
->group
= new_group
;
4755 EXPORT_SYMBOL(dev_set_group
);
4758 * dev_set_mac_address - Change Media Access Control Address
4762 * Change the hardware (MAC) address of the device
4764 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
4766 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4769 if (!ops
->ndo_set_mac_address
)
4771 if (sa
->sa_family
!= dev
->type
)
4773 if (!netif_device_present(dev
))
4775 err
= ops
->ndo_set_mac_address(dev
, sa
);
4778 dev
->addr_assign_type
= NET_ADDR_SET
;
4779 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4780 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
4783 EXPORT_SYMBOL(dev_set_mac_address
);
4786 * dev_change_carrier - Change device carrier
4788 * @new_carrier: new value
4790 * Change device carrier
4792 int dev_change_carrier(struct net_device
*dev
, bool new_carrier
)
4794 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4796 if (!ops
->ndo_change_carrier
)
4798 if (!netif_device_present(dev
))
4800 return ops
->ndo_change_carrier(dev
, new_carrier
);
4802 EXPORT_SYMBOL(dev_change_carrier
);
4805 * dev_new_index - allocate an ifindex
4806 * @net: the applicable net namespace
4808 * Returns a suitable unique value for a new device interface
4809 * number. The caller must hold the rtnl semaphore or the
4810 * dev_base_lock to be sure it remains unique.
4812 static int dev_new_index(struct net
*net
)
4814 int ifindex
= net
->ifindex
;
4818 if (!__dev_get_by_index(net
, ifindex
))
4819 return net
->ifindex
= ifindex
;
4823 /* Delayed registration/unregisteration */
4824 static LIST_HEAD(net_todo_list
);
4826 static void net_set_todo(struct net_device
*dev
)
4828 list_add_tail(&dev
->todo_list
, &net_todo_list
);
4831 static void rollback_registered_many(struct list_head
*head
)
4833 struct net_device
*dev
, *tmp
;
4835 BUG_ON(dev_boot_phase
);
4838 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
4839 /* Some devices call without registering
4840 * for initialization unwind. Remove those
4841 * devices and proceed with the remaining.
4843 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
4844 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
4848 list_del(&dev
->unreg_list
);
4851 dev
->dismantle
= true;
4852 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
4855 /* If device is running, close it first. */
4856 dev_close_many(head
);
4858 list_for_each_entry(dev
, head
, unreg_list
) {
4859 /* And unlink it from device chain. */
4860 unlist_netdevice(dev
);
4862 dev
->reg_state
= NETREG_UNREGISTERING
;
4867 list_for_each_entry(dev
, head
, unreg_list
) {
4868 /* Shutdown queueing discipline. */
4872 /* Notify protocols, that we are about to destroy
4873 this device. They should clean all the things.
4875 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
4877 if (!dev
->rtnl_link_ops
||
4878 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
4879 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
4882 * Flush the unicast and multicast chains
4887 if (dev
->netdev_ops
->ndo_uninit
)
4888 dev
->netdev_ops
->ndo_uninit(dev
);
4890 /* Notifier chain MUST detach us all upper devices. */
4891 WARN_ON(netdev_has_any_upper_dev(dev
));
4893 /* Remove entries from kobject tree */
4894 netdev_unregister_kobject(dev
);
4896 /* Remove XPS queueing entries */
4897 netif_reset_xps_queues_gt(dev
, 0);
4903 list_for_each_entry(dev
, head
, unreg_list
)
4907 static void rollback_registered(struct net_device
*dev
)
4911 list_add(&dev
->unreg_list
, &single
);
4912 rollback_registered_many(&single
);
4916 static netdev_features_t
netdev_fix_features(struct net_device
*dev
,
4917 netdev_features_t features
)
4919 /* Fix illegal checksum combinations */
4920 if ((features
& NETIF_F_HW_CSUM
) &&
4921 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
4922 netdev_warn(dev
, "mixed HW and IP checksum settings.\n");
4923 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
4926 /* Fix illegal SG+CSUM combinations. */
4927 if ((features
& NETIF_F_SG
) &&
4928 !(features
& NETIF_F_ALL_CSUM
)) {
4930 "Dropping NETIF_F_SG since no checksum feature.\n");
4931 features
&= ~NETIF_F_SG
;
4934 /* TSO requires that SG is present as well. */
4935 if ((features
& NETIF_F_ALL_TSO
) && !(features
& NETIF_F_SG
)) {
4936 netdev_dbg(dev
, "Dropping TSO features since no SG feature.\n");
4937 features
&= ~NETIF_F_ALL_TSO
;
4940 /* TSO ECN requires that TSO is present as well. */
4941 if ((features
& NETIF_F_ALL_TSO
) == NETIF_F_TSO_ECN
)
4942 features
&= ~NETIF_F_TSO_ECN
;
4944 /* Software GSO depends on SG. */
4945 if ((features
& NETIF_F_GSO
) && !(features
& NETIF_F_SG
)) {
4946 netdev_dbg(dev
, "Dropping NETIF_F_GSO since no SG feature.\n");
4947 features
&= ~NETIF_F_GSO
;
4950 /* UFO needs SG and checksumming */
4951 if (features
& NETIF_F_UFO
) {
4952 /* maybe split UFO into V4 and V6? */
4953 if (!((features
& NETIF_F_GEN_CSUM
) ||
4954 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))
4955 == (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
4957 "Dropping NETIF_F_UFO since no checksum offload features.\n");
4958 features
&= ~NETIF_F_UFO
;
4961 if (!(features
& NETIF_F_SG
)) {
4963 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
4964 features
&= ~NETIF_F_UFO
;
4971 int __netdev_update_features(struct net_device
*dev
)
4973 netdev_features_t features
;
4978 features
= netdev_get_wanted_features(dev
);
4980 if (dev
->netdev_ops
->ndo_fix_features
)
4981 features
= dev
->netdev_ops
->ndo_fix_features(dev
, features
);
4983 /* driver might be less strict about feature dependencies */
4984 features
= netdev_fix_features(dev
, features
);
4986 if (dev
->features
== features
)
4989 netdev_dbg(dev
, "Features changed: %pNF -> %pNF\n",
4990 &dev
->features
, &features
);
4992 if (dev
->netdev_ops
->ndo_set_features
)
4993 err
= dev
->netdev_ops
->ndo_set_features(dev
, features
);
4995 if (unlikely(err
< 0)) {
4997 "set_features() failed (%d); wanted %pNF, left %pNF\n",
4998 err
, &features
, &dev
->features
);
5003 dev
->features
= features
;
5009 * netdev_update_features - recalculate device features
5010 * @dev: the device to check
5012 * Recalculate dev->features set and send notifications if it
5013 * has changed. Should be called after driver or hardware dependent
5014 * conditions might have changed that influence the features.
5016 void netdev_update_features(struct net_device
*dev
)
5018 if (__netdev_update_features(dev
))
5019 netdev_features_change(dev
);
5021 EXPORT_SYMBOL(netdev_update_features
);
5024 * netdev_change_features - recalculate device features
5025 * @dev: the device to check
5027 * Recalculate dev->features set and send notifications even
5028 * if they have not changed. Should be called instead of
5029 * netdev_update_features() if also dev->vlan_features might
5030 * have changed to allow the changes to be propagated to stacked
5033 void netdev_change_features(struct net_device
*dev
)
5035 __netdev_update_features(dev
);
5036 netdev_features_change(dev
);
5038 EXPORT_SYMBOL(netdev_change_features
);
5041 * netif_stacked_transfer_operstate - transfer operstate
5042 * @rootdev: the root or lower level device to transfer state from
5043 * @dev: the device to transfer operstate to
5045 * Transfer operational state from root to device. This is normally
5046 * called when a stacking relationship exists between the root
5047 * device and the device(a leaf device).
5049 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
5050 struct net_device
*dev
)
5052 if (rootdev
->operstate
== IF_OPER_DORMANT
)
5053 netif_dormant_on(dev
);
5055 netif_dormant_off(dev
);
5057 if (netif_carrier_ok(rootdev
)) {
5058 if (!netif_carrier_ok(dev
))
5059 netif_carrier_on(dev
);
5061 if (netif_carrier_ok(dev
))
5062 netif_carrier_off(dev
);
5065 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
5068 static int netif_alloc_rx_queues(struct net_device
*dev
)
5070 unsigned int i
, count
= dev
->num_rx_queues
;
5071 struct netdev_rx_queue
*rx
;
5075 rx
= kcalloc(count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5081 for (i
= 0; i
< count
; i
++)
5087 static void netdev_init_one_queue(struct net_device
*dev
,
5088 struct netdev_queue
*queue
, void *_unused
)
5090 /* Initialize queue lock */
5091 spin_lock_init(&queue
->_xmit_lock
);
5092 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
5093 queue
->xmit_lock_owner
= -1;
5094 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
5097 dql_init(&queue
->dql
, HZ
);
5101 static int netif_alloc_netdev_queues(struct net_device
*dev
)
5103 unsigned int count
= dev
->num_tx_queues
;
5104 struct netdev_queue
*tx
;
5108 tx
= kcalloc(count
, sizeof(struct netdev_queue
), GFP_KERNEL
);
5114 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5115 spin_lock_init(&dev
->tx_global_lock
);
5121 * register_netdevice - register a network device
5122 * @dev: device to register
5124 * Take a completed network device structure and add it to the kernel
5125 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5126 * chain. 0 is returned on success. A negative errno code is returned
5127 * on a failure to set up the device, or if the name is a duplicate.
5129 * Callers must hold the rtnl semaphore. You may want
5130 * register_netdev() instead of this.
5133 * The locking appears insufficient to guarantee two parallel registers
5134 * will not get the same name.
5137 int register_netdevice(struct net_device
*dev
)
5140 struct net
*net
= dev_net(dev
);
5142 BUG_ON(dev_boot_phase
);
5147 /* When net_device's are persistent, this will be fatal. */
5148 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
5151 spin_lock_init(&dev
->addr_list_lock
);
5152 netdev_set_addr_lockdep_class(dev
);
5156 ret
= dev_get_valid_name(net
, dev
, dev
->name
);
5160 /* Init, if this function is available */
5161 if (dev
->netdev_ops
->ndo_init
) {
5162 ret
= dev
->netdev_ops
->ndo_init(dev
);
5170 if (((dev
->hw_features
| dev
->features
) & NETIF_F_HW_VLAN_FILTER
) &&
5171 (!dev
->netdev_ops
->ndo_vlan_rx_add_vid
||
5172 !dev
->netdev_ops
->ndo_vlan_rx_kill_vid
)) {
5173 netdev_WARN(dev
, "Buggy VLAN acceleration in driver!\n");
5180 dev
->ifindex
= dev_new_index(net
);
5181 else if (__dev_get_by_index(net
, dev
->ifindex
))
5184 if (dev
->iflink
== -1)
5185 dev
->iflink
= dev
->ifindex
;
5187 /* Transfer changeable features to wanted_features and enable
5188 * software offloads (GSO and GRO).
5190 dev
->hw_features
|= NETIF_F_SOFT_FEATURES
;
5191 dev
->features
|= NETIF_F_SOFT_FEATURES
;
5192 dev
->wanted_features
= dev
->features
& dev
->hw_features
;
5194 /* Turn on no cache copy if HW is doing checksum */
5195 if (!(dev
->flags
& IFF_LOOPBACK
)) {
5196 dev
->hw_features
|= NETIF_F_NOCACHE_COPY
;
5197 if (dev
->features
& NETIF_F_ALL_CSUM
) {
5198 dev
->wanted_features
|= NETIF_F_NOCACHE_COPY
;
5199 dev
->features
|= NETIF_F_NOCACHE_COPY
;
5203 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5205 dev
->vlan_features
|= NETIF_F_HIGHDMA
;
5207 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
5208 ret
= notifier_to_errno(ret
);
5212 ret
= netdev_register_kobject(dev
);
5215 dev
->reg_state
= NETREG_REGISTERED
;
5217 __netdev_update_features(dev
);
5220 * Default initial state at registry is that the
5221 * device is present.
5224 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5226 linkwatch_init_dev(dev
);
5228 dev_init_scheduler(dev
);
5230 list_netdevice(dev
);
5231 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
5233 /* If the device has permanent device address, driver should
5234 * set dev_addr and also addr_assign_type should be set to
5235 * NET_ADDR_PERM (default value).
5237 if (dev
->addr_assign_type
== NET_ADDR_PERM
)
5238 memcpy(dev
->perm_addr
, dev
->dev_addr
, dev
->addr_len
);
5240 /* Notify protocols, that a new device appeared. */
5241 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5242 ret
= notifier_to_errno(ret
);
5244 rollback_registered(dev
);
5245 dev
->reg_state
= NETREG_UNREGISTERED
;
5248 * Prevent userspace races by waiting until the network
5249 * device is fully setup before sending notifications.
5251 if (!dev
->rtnl_link_ops
||
5252 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5253 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5259 if (dev
->netdev_ops
->ndo_uninit
)
5260 dev
->netdev_ops
->ndo_uninit(dev
);
5263 EXPORT_SYMBOL(register_netdevice
);
5266 * init_dummy_netdev - init a dummy network device for NAPI
5267 * @dev: device to init
5269 * This takes a network device structure and initialize the minimum
5270 * amount of fields so it can be used to schedule NAPI polls without
5271 * registering a full blown interface. This is to be used by drivers
5272 * that need to tie several hardware interfaces to a single NAPI
5273 * poll scheduler due to HW limitations.
5275 int init_dummy_netdev(struct net_device
*dev
)
5277 /* Clear everything. Note we don't initialize spinlocks
5278 * are they aren't supposed to be taken by any of the
5279 * NAPI code and this dummy netdev is supposed to be
5280 * only ever used for NAPI polls
5282 memset(dev
, 0, sizeof(struct net_device
));
5284 /* make sure we BUG if trying to hit standard
5285 * register/unregister code path
5287 dev
->reg_state
= NETREG_DUMMY
;
5289 /* NAPI wants this */
5290 INIT_LIST_HEAD(&dev
->napi_list
);
5292 /* a dummy interface is started by default */
5293 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5294 set_bit(__LINK_STATE_START
, &dev
->state
);
5296 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5297 * because users of this 'device' dont need to change
5303 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
5307 * register_netdev - register a network device
5308 * @dev: device to register
5310 * Take a completed network device structure and add it to the kernel
5311 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5312 * chain. 0 is returned on success. A negative errno code is returned
5313 * on a failure to set up the device, or if the name is a duplicate.
5315 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5316 * and expands the device name if you passed a format string to
5319 int register_netdev(struct net_device
*dev
)
5324 err
= register_netdevice(dev
);
5328 EXPORT_SYMBOL(register_netdev
);
5330 int netdev_refcnt_read(const struct net_device
*dev
)
5334 for_each_possible_cpu(i
)
5335 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
5338 EXPORT_SYMBOL(netdev_refcnt_read
);
5341 * netdev_wait_allrefs - wait until all references are gone.
5342 * @dev: target net_device
5344 * This is called when unregistering network devices.
5346 * Any protocol or device that holds a reference should register
5347 * for netdevice notification, and cleanup and put back the
5348 * reference if they receive an UNREGISTER event.
5349 * We can get stuck here if buggy protocols don't correctly
5352 static void netdev_wait_allrefs(struct net_device
*dev
)
5354 unsigned long rebroadcast_time
, warning_time
;
5357 linkwatch_forget_dev(dev
);
5359 rebroadcast_time
= warning_time
= jiffies
;
5360 refcnt
= netdev_refcnt_read(dev
);
5362 while (refcnt
!= 0) {
5363 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
5366 /* Rebroadcast unregister notification */
5367 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5373 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
5374 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
5376 /* We must not have linkwatch events
5377 * pending on unregister. If this
5378 * happens, we simply run the queue
5379 * unscheduled, resulting in a noop
5382 linkwatch_run_queue();
5387 rebroadcast_time
= jiffies
;
5392 refcnt
= netdev_refcnt_read(dev
);
5394 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
5395 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5397 warning_time
= jiffies
;
5406 * register_netdevice(x1);
5407 * register_netdevice(x2);
5409 * unregister_netdevice(y1);
5410 * unregister_netdevice(y2);
5416 * We are invoked by rtnl_unlock().
5417 * This allows us to deal with problems:
5418 * 1) We can delete sysfs objects which invoke hotplug
5419 * without deadlocking with linkwatch via keventd.
5420 * 2) Since we run with the RTNL semaphore not held, we can sleep
5421 * safely in order to wait for the netdev refcnt to drop to zero.
5423 * We must not return until all unregister events added during
5424 * the interval the lock was held have been completed.
5426 void netdev_run_todo(void)
5428 struct list_head list
;
5430 /* Snapshot list, allow later requests */
5431 list_replace_init(&net_todo_list
, &list
);
5436 /* Wait for rcu callbacks to finish before next phase */
5437 if (!list_empty(&list
))
5440 while (!list_empty(&list
)) {
5441 struct net_device
*dev
5442 = list_first_entry(&list
, struct net_device
, todo_list
);
5443 list_del(&dev
->todo_list
);
5446 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
5449 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
5450 pr_err("network todo '%s' but state %d\n",
5451 dev
->name
, dev
->reg_state
);
5456 dev
->reg_state
= NETREG_UNREGISTERED
;
5458 on_each_cpu(flush_backlog
, dev
, 1);
5460 netdev_wait_allrefs(dev
);
5463 BUG_ON(netdev_refcnt_read(dev
));
5464 WARN_ON(rcu_access_pointer(dev
->ip_ptr
));
5465 WARN_ON(rcu_access_pointer(dev
->ip6_ptr
));
5466 WARN_ON(dev
->dn_ptr
);
5468 if (dev
->destructor
)
5469 dev
->destructor(dev
);
5471 /* Free network device */
5472 kobject_put(&dev
->dev
.kobj
);
5476 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5477 * fields in the same order, with only the type differing.
5479 void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
5480 const struct net_device_stats
*netdev_stats
)
5482 #if BITS_PER_LONG == 64
5483 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
5484 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
5486 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
5487 const unsigned long *src
= (const unsigned long *)netdev_stats
;
5488 u64
*dst
= (u64
*)stats64
;
5490 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
5491 sizeof(*stats64
) / sizeof(u64
));
5492 for (i
= 0; i
< n
; i
++)
5496 EXPORT_SYMBOL(netdev_stats_to_stats64
);
5499 * dev_get_stats - get network device statistics
5500 * @dev: device to get statistics from
5501 * @storage: place to store stats
5503 * Get network statistics from device. Return @storage.
5504 * The device driver may provide its own method by setting
5505 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5506 * otherwise the internal statistics structure is used.
5508 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
5509 struct rtnl_link_stats64
*storage
)
5511 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5513 if (ops
->ndo_get_stats64
) {
5514 memset(storage
, 0, sizeof(*storage
));
5515 ops
->ndo_get_stats64(dev
, storage
);
5516 } else if (ops
->ndo_get_stats
) {
5517 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
5519 netdev_stats_to_stats64(storage
, &dev
->stats
);
5521 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
5524 EXPORT_SYMBOL(dev_get_stats
);
5526 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
5528 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
5530 #ifdef CONFIG_NET_CLS_ACT
5533 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
5536 netdev_init_one_queue(dev
, queue
, NULL
);
5537 queue
->qdisc
= &noop_qdisc
;
5538 queue
->qdisc_sleeping
= &noop_qdisc
;
5539 rcu_assign_pointer(dev
->ingress_queue
, queue
);
5544 static const struct ethtool_ops default_ethtool_ops
;
5546 void netdev_set_default_ethtool_ops(struct net_device
*dev
,
5547 const struct ethtool_ops
*ops
)
5549 if (dev
->ethtool_ops
== &default_ethtool_ops
)
5550 dev
->ethtool_ops
= ops
;
5552 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops
);
5555 * alloc_netdev_mqs - allocate network device
5556 * @sizeof_priv: size of private data to allocate space for
5557 * @name: device name format string
5558 * @setup: callback to initialize device
5559 * @txqs: the number of TX subqueues to allocate
5560 * @rxqs: the number of RX subqueues to allocate
5562 * Allocates a struct net_device with private data area for driver use
5563 * and performs basic initialization. Also allocates subquue structs
5564 * for each queue on the device.
5566 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
5567 void (*setup
)(struct net_device
*),
5568 unsigned int txqs
, unsigned int rxqs
)
5570 struct net_device
*dev
;
5572 struct net_device
*p
;
5574 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
5577 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
5583 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
5588 alloc_size
= sizeof(struct net_device
);
5590 /* ensure 32-byte alignment of private area */
5591 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
5592 alloc_size
+= sizeof_priv
;
5594 /* ensure 32-byte alignment of whole construct */
5595 alloc_size
+= NETDEV_ALIGN
- 1;
5597 p
= kzalloc(alloc_size
, GFP_KERNEL
);
5601 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
5602 dev
->padded
= (char *)dev
- (char *)p
;
5604 dev
->pcpu_refcnt
= alloc_percpu(int);
5605 if (!dev
->pcpu_refcnt
)
5608 if (dev_addr_init(dev
))
5614 dev_net_set(dev
, &init_net
);
5616 dev
->gso_max_size
= GSO_MAX_SIZE
;
5617 dev
->gso_max_segs
= GSO_MAX_SEGS
;
5619 INIT_LIST_HEAD(&dev
->napi_list
);
5620 INIT_LIST_HEAD(&dev
->unreg_list
);
5621 INIT_LIST_HEAD(&dev
->link_watch_list
);
5622 INIT_LIST_HEAD(&dev
->upper_dev_list
);
5623 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
5626 dev
->num_tx_queues
= txqs
;
5627 dev
->real_num_tx_queues
= txqs
;
5628 if (netif_alloc_netdev_queues(dev
))
5632 dev
->num_rx_queues
= rxqs
;
5633 dev
->real_num_rx_queues
= rxqs
;
5634 if (netif_alloc_rx_queues(dev
))
5638 strcpy(dev
->name
, name
);
5639 dev
->group
= INIT_NETDEV_GROUP
;
5640 if (!dev
->ethtool_ops
)
5641 dev
->ethtool_ops
= &default_ethtool_ops
;
5649 free_percpu(dev
->pcpu_refcnt
);
5659 EXPORT_SYMBOL(alloc_netdev_mqs
);
5662 * free_netdev - free network device
5665 * This function does the last stage of destroying an allocated device
5666 * interface. The reference to the device object is released.
5667 * If this is the last reference then it will be freed.
5669 void free_netdev(struct net_device
*dev
)
5671 struct napi_struct
*p
, *n
;
5673 release_net(dev_net(dev
));
5680 kfree(rcu_dereference_protected(dev
->ingress_queue
, 1));
5682 /* Flush device addresses */
5683 dev_addr_flush(dev
);
5685 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
5688 free_percpu(dev
->pcpu_refcnt
);
5689 dev
->pcpu_refcnt
= NULL
;
5691 /* Compatibility with error handling in drivers */
5692 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5693 kfree((char *)dev
- dev
->padded
);
5697 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
5698 dev
->reg_state
= NETREG_RELEASED
;
5700 /* will free via device release */
5701 put_device(&dev
->dev
);
5703 EXPORT_SYMBOL(free_netdev
);
5706 * synchronize_net - Synchronize with packet receive processing
5708 * Wait for packets currently being received to be done.
5709 * Does not block later packets from starting.
5711 void synchronize_net(void)
5714 if (rtnl_is_locked())
5715 synchronize_rcu_expedited();
5719 EXPORT_SYMBOL(synchronize_net
);
5722 * unregister_netdevice_queue - remove device from the kernel
5726 * This function shuts down a device interface and removes it
5727 * from the kernel tables.
5728 * If head not NULL, device is queued to be unregistered later.
5730 * Callers must hold the rtnl semaphore. You may want
5731 * unregister_netdev() instead of this.
5734 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
5739 list_move_tail(&dev
->unreg_list
, head
);
5741 rollback_registered(dev
);
5742 /* Finish processing unregister after unlock */
5746 EXPORT_SYMBOL(unregister_netdevice_queue
);
5749 * unregister_netdevice_many - unregister many devices
5750 * @head: list of devices
5752 void unregister_netdevice_many(struct list_head
*head
)
5754 struct net_device
*dev
;
5756 if (!list_empty(head
)) {
5757 rollback_registered_many(head
);
5758 list_for_each_entry(dev
, head
, unreg_list
)
5762 EXPORT_SYMBOL(unregister_netdevice_many
);
5765 * unregister_netdev - remove device from the kernel
5768 * This function shuts down a device interface and removes it
5769 * from the kernel tables.
5771 * This is just a wrapper for unregister_netdevice that takes
5772 * the rtnl semaphore. In general you want to use this and not
5773 * unregister_netdevice.
5775 void unregister_netdev(struct net_device
*dev
)
5778 unregister_netdevice(dev
);
5781 EXPORT_SYMBOL(unregister_netdev
);
5784 * dev_change_net_namespace - move device to different nethost namespace
5786 * @net: network namespace
5787 * @pat: If not NULL name pattern to try if the current device name
5788 * is already taken in the destination network namespace.
5790 * This function shuts down a device interface and moves it
5791 * to a new network namespace. On success 0 is returned, on
5792 * a failure a netagive errno code is returned.
5794 * Callers must hold the rtnl semaphore.
5797 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
5803 /* Don't allow namespace local devices to be moved. */
5805 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
5808 /* Ensure the device has been registrered */
5809 if (dev
->reg_state
!= NETREG_REGISTERED
)
5812 /* Get out if there is nothing todo */
5814 if (net_eq(dev_net(dev
), net
))
5817 /* Pick the destination device name, and ensure
5818 * we can use it in the destination network namespace.
5821 if (__dev_get_by_name(net
, dev
->name
)) {
5822 /* We get here if we can't use the current device name */
5825 if (dev_get_valid_name(net
, dev
, pat
) < 0)
5830 * And now a mini version of register_netdevice unregister_netdevice.
5833 /* If device is running close it first. */
5836 /* And unlink it from device chain */
5838 unlist_netdevice(dev
);
5842 /* Shutdown queueing discipline. */
5845 /* Notify protocols, that we are about to destroy
5846 this device. They should clean all the things.
5848 Note that dev->reg_state stays at NETREG_REGISTERED.
5849 This is wanted because this way 8021q and macvlan know
5850 the device is just moving and can keep their slaves up.
5852 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5854 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
5855 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
5858 * Flush the unicast and multicast chains
5863 /* Send a netdev-removed uevent to the old namespace */
5864 kobject_uevent(&dev
->dev
.kobj
, KOBJ_REMOVE
);
5866 /* Actually switch the network namespace */
5867 dev_net_set(dev
, net
);
5869 /* If there is an ifindex conflict assign a new one */
5870 if (__dev_get_by_index(net
, dev
->ifindex
)) {
5871 int iflink
= (dev
->iflink
== dev
->ifindex
);
5872 dev
->ifindex
= dev_new_index(net
);
5874 dev
->iflink
= dev
->ifindex
;
5877 /* Send a netdev-add uevent to the new namespace */
5878 kobject_uevent(&dev
->dev
.kobj
, KOBJ_ADD
);
5880 /* Fixup kobjects */
5881 err
= device_rename(&dev
->dev
, dev
->name
);
5884 /* Add the device back in the hashes */
5885 list_netdevice(dev
);
5887 /* Notify protocols, that a new device appeared. */
5888 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5891 * Prevent userspace races by waiting until the network
5892 * device is fully setup before sending notifications.
5894 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5901 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
5903 static int dev_cpu_callback(struct notifier_block
*nfb
,
5904 unsigned long action
,
5907 struct sk_buff
**list_skb
;
5908 struct sk_buff
*skb
;
5909 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
5910 struct softnet_data
*sd
, *oldsd
;
5912 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
5915 local_irq_disable();
5916 cpu
= smp_processor_id();
5917 sd
= &per_cpu(softnet_data
, cpu
);
5918 oldsd
= &per_cpu(softnet_data
, oldcpu
);
5920 /* Find end of our completion_queue. */
5921 list_skb
= &sd
->completion_queue
;
5923 list_skb
= &(*list_skb
)->next
;
5924 /* Append completion queue from offline CPU. */
5925 *list_skb
= oldsd
->completion_queue
;
5926 oldsd
->completion_queue
= NULL
;
5928 /* Append output queue from offline CPU. */
5929 if (oldsd
->output_queue
) {
5930 *sd
->output_queue_tailp
= oldsd
->output_queue
;
5931 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
5932 oldsd
->output_queue
= NULL
;
5933 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
5935 /* Append NAPI poll list from offline CPU. */
5936 if (!list_empty(&oldsd
->poll_list
)) {
5937 list_splice_init(&oldsd
->poll_list
, &sd
->poll_list
);
5938 raise_softirq_irqoff(NET_RX_SOFTIRQ
);
5941 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
5944 /* Process offline CPU's input_pkt_queue */
5945 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
5947 input_queue_head_incr(oldsd
);
5949 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
5951 input_queue_head_incr(oldsd
);
5959 * netdev_increment_features - increment feature set by one
5960 * @all: current feature set
5961 * @one: new feature set
5962 * @mask: mask feature set
5964 * Computes a new feature set after adding a device with feature set
5965 * @one to the master device with current feature set @all. Will not
5966 * enable anything that is off in @mask. Returns the new feature set.
5968 netdev_features_t
netdev_increment_features(netdev_features_t all
,
5969 netdev_features_t one
, netdev_features_t mask
)
5971 if (mask
& NETIF_F_GEN_CSUM
)
5972 mask
|= NETIF_F_ALL_CSUM
;
5973 mask
|= NETIF_F_VLAN_CHALLENGED
;
5975 all
|= one
& (NETIF_F_ONE_FOR_ALL
|NETIF_F_ALL_CSUM
) & mask
;
5976 all
&= one
| ~NETIF_F_ALL_FOR_ALL
;
5978 /* If one device supports hw checksumming, set for all. */
5979 if (all
& NETIF_F_GEN_CSUM
)
5980 all
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_GEN_CSUM
);
5984 EXPORT_SYMBOL(netdev_increment_features
);
5986 static struct hlist_head
*netdev_create_hash(void)
5989 struct hlist_head
*hash
;
5991 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
5993 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
5994 INIT_HLIST_HEAD(&hash
[i
]);
5999 /* Initialize per network namespace state */
6000 static int __net_init
netdev_init(struct net
*net
)
6002 if (net
!= &init_net
)
6003 INIT_LIST_HEAD(&net
->dev_base_head
);
6005 net
->dev_name_head
= netdev_create_hash();
6006 if (net
->dev_name_head
== NULL
)
6009 net
->dev_index_head
= netdev_create_hash();
6010 if (net
->dev_index_head
== NULL
)
6016 kfree(net
->dev_name_head
);
6022 * netdev_drivername - network driver for the device
6023 * @dev: network device
6025 * Determine network driver for device.
6027 const char *netdev_drivername(const struct net_device
*dev
)
6029 const struct device_driver
*driver
;
6030 const struct device
*parent
;
6031 const char *empty
= "";
6033 parent
= dev
->dev
.parent
;
6037 driver
= parent
->driver
;
6038 if (driver
&& driver
->name
)
6039 return driver
->name
;
6043 static int __netdev_printk(const char *level
, const struct net_device
*dev
,
6044 struct va_format
*vaf
)
6048 if (dev
&& dev
->dev
.parent
) {
6049 r
= dev_printk_emit(level
[1] - '0',
6052 dev_driver_string(dev
->dev
.parent
),
6053 dev_name(dev
->dev
.parent
),
6054 netdev_name(dev
), vaf
);
6056 r
= printk("%s%s: %pV", level
, netdev_name(dev
), vaf
);
6058 r
= printk("%s(NULL net_device): %pV", level
, vaf
);
6064 int netdev_printk(const char *level
, const struct net_device
*dev
,
6065 const char *format
, ...)
6067 struct va_format vaf
;
6071 va_start(args
, format
);
6076 r
= __netdev_printk(level
, dev
, &vaf
);
6082 EXPORT_SYMBOL(netdev_printk
);
6084 #define define_netdev_printk_level(func, level) \
6085 int func(const struct net_device *dev, const char *fmt, ...) \
6088 struct va_format vaf; \
6091 va_start(args, fmt); \
6096 r = __netdev_printk(level, dev, &vaf); \
6102 EXPORT_SYMBOL(func);
6104 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
6105 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
6106 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
6107 define_netdev_printk_level(netdev_err
, KERN_ERR
);
6108 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
6109 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
6110 define_netdev_printk_level(netdev_info
, KERN_INFO
);
6112 static void __net_exit
netdev_exit(struct net
*net
)
6114 kfree(net
->dev_name_head
);
6115 kfree(net
->dev_index_head
);
6118 static struct pernet_operations __net_initdata netdev_net_ops
= {
6119 .init
= netdev_init
,
6120 .exit
= netdev_exit
,
6123 static void __net_exit
default_device_exit(struct net
*net
)
6125 struct net_device
*dev
, *aux
;
6127 * Push all migratable network devices back to the
6128 * initial network namespace
6131 for_each_netdev_safe(net
, dev
, aux
) {
6133 char fb_name
[IFNAMSIZ
];
6135 /* Ignore unmoveable devices (i.e. loopback) */
6136 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6139 /* Leave virtual devices for the generic cleanup */
6140 if (dev
->rtnl_link_ops
)
6143 /* Push remaining network devices to init_net */
6144 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
6145 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
6147 pr_emerg("%s: failed to move %s to init_net: %d\n",
6148 __func__
, dev
->name
, err
);
6155 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
6157 /* At exit all network devices most be removed from a network
6158 * namespace. Do this in the reverse order of registration.
6159 * Do this across as many network namespaces as possible to
6160 * improve batching efficiency.
6162 struct net_device
*dev
;
6164 LIST_HEAD(dev_kill_list
);
6167 list_for_each_entry(net
, net_list
, exit_list
) {
6168 for_each_netdev_reverse(net
, dev
) {
6169 if (dev
->rtnl_link_ops
)
6170 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
6172 unregister_netdevice_queue(dev
, &dev_kill_list
);
6175 unregister_netdevice_many(&dev_kill_list
);
6176 list_del(&dev_kill_list
);
6180 static struct pernet_operations __net_initdata default_device_ops
= {
6181 .exit
= default_device_exit
,
6182 .exit_batch
= default_device_exit_batch
,
6186 * Initialize the DEV module. At boot time this walks the device list and
6187 * unhooks any devices that fail to initialise (normally hardware not
6188 * present) and leaves us with a valid list of present and active devices.
6193 * This is called single threaded during boot, so no need
6194 * to take the rtnl semaphore.
6196 static int __init
net_dev_init(void)
6198 int i
, rc
= -ENOMEM
;
6200 BUG_ON(!dev_boot_phase
);
6202 if (dev_proc_init())
6205 if (netdev_kobject_init())
6208 INIT_LIST_HEAD(&ptype_all
);
6209 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
6210 INIT_LIST_HEAD(&ptype_base
[i
]);
6212 INIT_LIST_HEAD(&offload_base
);
6214 if (register_pernet_subsys(&netdev_net_ops
))
6218 * Initialise the packet receive queues.
6221 for_each_possible_cpu(i
) {
6222 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
6224 memset(sd
, 0, sizeof(*sd
));
6225 skb_queue_head_init(&sd
->input_pkt_queue
);
6226 skb_queue_head_init(&sd
->process_queue
);
6227 sd
->completion_queue
= NULL
;
6228 INIT_LIST_HEAD(&sd
->poll_list
);
6229 sd
->output_queue
= NULL
;
6230 sd
->output_queue_tailp
= &sd
->output_queue
;
6232 sd
->csd
.func
= rps_trigger_softirq
;
6238 sd
->backlog
.poll
= process_backlog
;
6239 sd
->backlog
.weight
= weight_p
;
6240 sd
->backlog
.gro_list
= NULL
;
6241 sd
->backlog
.gro_count
= 0;
6246 /* The loopback device is special if any other network devices
6247 * is present in a network namespace the loopback device must
6248 * be present. Since we now dynamically allocate and free the
6249 * loopback device ensure this invariant is maintained by
6250 * keeping the loopback device as the first device on the
6251 * list of network devices. Ensuring the loopback devices
6252 * is the first device that appears and the last network device
6255 if (register_pernet_device(&loopback_net_ops
))
6258 if (register_pernet_device(&default_device_ops
))
6261 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
6262 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
6264 hotcpu_notifier(dev_cpu_callback
, 0);
6271 subsys_initcall(net_dev_init
);