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>
132 #include <linux/hashtable.h>
133 #include <linux/vmalloc.h>
134 #include <linux/if_macvlan.h>
135 #include <linux/errqueue.h>
137 #include "net-sysfs.h"
139 /* Instead of increasing this, you should create a hash table. */
140 #define MAX_GRO_SKBS 8
142 /* This should be increased if a protocol with a bigger head is added. */
143 #define GRO_MAX_HEAD (MAX_HEADER + 128)
145 static DEFINE_SPINLOCK(ptype_lock
);
146 static DEFINE_SPINLOCK(offload_lock
);
147 struct list_head ptype_base
[PTYPE_HASH_SIZE
] __read_mostly
;
148 struct list_head ptype_all __read_mostly
; /* Taps */
149 static struct list_head offload_base __read_mostly
;
151 static int netif_rx_internal(struct sk_buff
*skb
);
152 static int call_netdevice_notifiers_info(unsigned long val
,
153 struct net_device
*dev
,
154 struct netdev_notifier_info
*info
);
157 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
160 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
162 * Writers must hold the rtnl semaphore while they loop through the
163 * dev_base_head list, and hold dev_base_lock for writing when they do the
164 * actual updates. This allows pure readers to access the list even
165 * while a writer is preparing to update it.
167 * To put it another way, dev_base_lock is held for writing only to
168 * protect against pure readers; the rtnl semaphore provides the
169 * protection against other writers.
171 * See, for example usages, register_netdevice() and
172 * unregister_netdevice(), which must be called with the rtnl
175 DEFINE_RWLOCK(dev_base_lock
);
176 EXPORT_SYMBOL(dev_base_lock
);
178 /* protects napi_hash addition/deletion and napi_gen_id */
179 static DEFINE_SPINLOCK(napi_hash_lock
);
181 static unsigned int napi_gen_id
;
182 static DEFINE_HASHTABLE(napi_hash
, 8);
184 static seqcount_t devnet_rename_seq
;
186 static inline void dev_base_seq_inc(struct net
*net
)
188 while (++net
->dev_base_seq
== 0);
191 static inline struct hlist_head
*dev_name_hash(struct net
*net
, const char *name
)
193 unsigned int hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
195 return &net
->dev_name_head
[hash_32(hash
, NETDEV_HASHBITS
)];
198 static inline struct hlist_head
*dev_index_hash(struct net
*net
, int ifindex
)
200 return &net
->dev_index_head
[ifindex
& (NETDEV_HASHENTRIES
- 1)];
203 static inline void rps_lock(struct softnet_data
*sd
)
206 spin_lock(&sd
->input_pkt_queue
.lock
);
210 static inline void rps_unlock(struct softnet_data
*sd
)
213 spin_unlock(&sd
->input_pkt_queue
.lock
);
217 /* Device list insertion */
218 static void list_netdevice(struct net_device
*dev
)
220 struct net
*net
= dev_net(dev
);
224 write_lock_bh(&dev_base_lock
);
225 list_add_tail_rcu(&dev
->dev_list
, &net
->dev_base_head
);
226 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
227 hlist_add_head_rcu(&dev
->index_hlist
,
228 dev_index_hash(net
, dev
->ifindex
));
229 write_unlock_bh(&dev_base_lock
);
231 dev_base_seq_inc(net
);
234 /* Device list removal
235 * caller must respect a RCU grace period before freeing/reusing dev
237 static void unlist_netdevice(struct net_device
*dev
)
241 /* Unlink dev from the device chain */
242 write_lock_bh(&dev_base_lock
);
243 list_del_rcu(&dev
->dev_list
);
244 hlist_del_rcu(&dev
->name_hlist
);
245 hlist_del_rcu(&dev
->index_hlist
);
246 write_unlock_bh(&dev_base_lock
);
248 dev_base_seq_inc(dev_net(dev
));
255 static RAW_NOTIFIER_HEAD(netdev_chain
);
258 * Device drivers call our routines to queue packets here. We empty the
259 * queue in the local softnet handler.
262 DEFINE_PER_CPU_ALIGNED(struct softnet_data
, softnet_data
);
263 EXPORT_PER_CPU_SYMBOL(softnet_data
);
265 #ifdef CONFIG_LOCKDEP
267 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
268 * according to dev->type
270 static const unsigned short netdev_lock_type
[] =
271 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
272 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
273 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
274 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
275 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
276 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
277 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
278 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
279 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
280 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
281 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
282 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
283 ARPHRD_FCFABRIC
, ARPHRD_IEEE80211
, ARPHRD_IEEE80211_PRISM
,
284 ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_PHONET
, ARPHRD_PHONET_PIPE
,
285 ARPHRD_IEEE802154
, ARPHRD_VOID
, ARPHRD_NONE
};
287 static const char *const netdev_lock_name
[] =
288 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
289 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
290 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
291 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
292 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
293 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
294 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
295 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
296 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
297 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
298 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
299 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
300 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
301 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
302 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
304 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
305 static struct lock_class_key netdev_addr_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
307 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
311 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
312 if (netdev_lock_type
[i
] == dev_type
)
314 /* the last key is used by default */
315 return ARRAY_SIZE(netdev_lock_type
) - 1;
318 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
319 unsigned short dev_type
)
323 i
= netdev_lock_pos(dev_type
);
324 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
325 netdev_lock_name
[i
]);
328 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
332 i
= netdev_lock_pos(dev
->type
);
333 lockdep_set_class_and_name(&dev
->addr_list_lock
,
334 &netdev_addr_lock_key
[i
],
335 netdev_lock_name
[i
]);
338 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
339 unsigned short dev_type
)
342 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
347 /*******************************************************************************
349 Protocol management and registration routines
351 *******************************************************************************/
354 * Add a protocol ID to the list. Now that the input handler is
355 * smarter we can dispense with all the messy stuff that used to be
358 * BEWARE!!! Protocol handlers, mangling input packets,
359 * MUST BE last in hash buckets and checking protocol handlers
360 * MUST start from promiscuous ptype_all chain in net_bh.
361 * It is true now, do not change it.
362 * Explanation follows: if protocol handler, mangling packet, will
363 * be the first on list, it is not able to sense, that packet
364 * is cloned and should be copied-on-write, so that it will
365 * change it and subsequent readers will get broken packet.
369 static inline struct list_head
*ptype_head(const struct packet_type
*pt
)
371 if (pt
->type
== htons(ETH_P_ALL
))
374 return &ptype_base
[ntohs(pt
->type
) & PTYPE_HASH_MASK
];
378 * dev_add_pack - add packet handler
379 * @pt: packet type declaration
381 * Add a protocol handler to the networking stack. The passed &packet_type
382 * is linked into kernel lists and may not be freed until it has been
383 * removed from the kernel lists.
385 * This call does not sleep therefore it can not
386 * guarantee all CPU's that are in middle of receiving packets
387 * will see the new packet type (until the next received packet).
390 void dev_add_pack(struct packet_type
*pt
)
392 struct list_head
*head
= ptype_head(pt
);
394 spin_lock(&ptype_lock
);
395 list_add_rcu(&pt
->list
, head
);
396 spin_unlock(&ptype_lock
);
398 EXPORT_SYMBOL(dev_add_pack
);
401 * __dev_remove_pack - remove packet handler
402 * @pt: packet type declaration
404 * Remove a protocol handler that was previously added to the kernel
405 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
406 * from the kernel lists and can be freed or reused once this function
409 * The packet type might still be in use by receivers
410 * and must not be freed until after all the CPU's have gone
411 * through a quiescent state.
413 void __dev_remove_pack(struct packet_type
*pt
)
415 struct list_head
*head
= ptype_head(pt
);
416 struct packet_type
*pt1
;
418 spin_lock(&ptype_lock
);
420 list_for_each_entry(pt1
, head
, list
) {
422 list_del_rcu(&pt
->list
);
427 pr_warn("dev_remove_pack: %p not found\n", pt
);
429 spin_unlock(&ptype_lock
);
431 EXPORT_SYMBOL(__dev_remove_pack
);
434 * dev_remove_pack - remove packet handler
435 * @pt: packet type declaration
437 * Remove a protocol handler that was previously added to the kernel
438 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
439 * from the kernel lists and can be freed or reused once this function
442 * This call sleeps to guarantee that no CPU is looking at the packet
445 void dev_remove_pack(struct packet_type
*pt
)
447 __dev_remove_pack(pt
);
451 EXPORT_SYMBOL(dev_remove_pack
);
455 * dev_add_offload - register offload handlers
456 * @po: protocol offload declaration
458 * Add protocol offload handlers to the networking stack. The passed
459 * &proto_offload is linked into kernel lists and may not be freed until
460 * it has been removed from the kernel lists.
462 * This call does not sleep therefore it can not
463 * guarantee all CPU's that are in middle of receiving packets
464 * will see the new offload handlers (until the next received packet).
466 void dev_add_offload(struct packet_offload
*po
)
468 struct list_head
*head
= &offload_base
;
470 spin_lock(&offload_lock
);
471 list_add_rcu(&po
->list
, head
);
472 spin_unlock(&offload_lock
);
474 EXPORT_SYMBOL(dev_add_offload
);
477 * __dev_remove_offload - remove offload handler
478 * @po: packet offload declaration
480 * Remove a protocol offload handler that was previously added to the
481 * kernel offload handlers by dev_add_offload(). The passed &offload_type
482 * is removed from the kernel lists and can be freed or reused once this
485 * The packet type might still be in use by receivers
486 * and must not be freed until after all the CPU's have gone
487 * through a quiescent state.
489 static void __dev_remove_offload(struct packet_offload
*po
)
491 struct list_head
*head
= &offload_base
;
492 struct packet_offload
*po1
;
494 spin_lock(&offload_lock
);
496 list_for_each_entry(po1
, head
, list
) {
498 list_del_rcu(&po
->list
);
503 pr_warn("dev_remove_offload: %p not found\n", po
);
505 spin_unlock(&offload_lock
);
509 * dev_remove_offload - remove packet offload handler
510 * @po: packet offload declaration
512 * Remove a packet offload handler that was previously added to the kernel
513 * offload handlers by dev_add_offload(). The passed &offload_type is
514 * removed from the kernel lists and can be freed or reused once this
517 * This call sleeps to guarantee that no CPU is looking at the packet
520 void dev_remove_offload(struct packet_offload
*po
)
522 __dev_remove_offload(po
);
526 EXPORT_SYMBOL(dev_remove_offload
);
528 /******************************************************************************
530 Device Boot-time Settings Routines
532 *******************************************************************************/
534 /* Boot time configuration table */
535 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
538 * netdev_boot_setup_add - add new setup entry
539 * @name: name of the device
540 * @map: configured settings for the device
542 * Adds new setup entry to the dev_boot_setup list. The function
543 * returns 0 on error and 1 on success. This is a generic routine to
546 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
548 struct netdev_boot_setup
*s
;
552 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
553 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
554 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
555 strlcpy(s
[i
].name
, name
, IFNAMSIZ
);
556 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
561 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
565 * netdev_boot_setup_check - check boot time settings
566 * @dev: the netdevice
568 * Check boot time settings for the device.
569 * The found settings are set for the device to be used
570 * later in the device probing.
571 * Returns 0 if no settings found, 1 if they are.
573 int netdev_boot_setup_check(struct net_device
*dev
)
575 struct netdev_boot_setup
*s
= dev_boot_setup
;
578 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
579 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
580 !strcmp(dev
->name
, s
[i
].name
)) {
581 dev
->irq
= s
[i
].map
.irq
;
582 dev
->base_addr
= s
[i
].map
.base_addr
;
583 dev
->mem_start
= s
[i
].map
.mem_start
;
584 dev
->mem_end
= s
[i
].map
.mem_end
;
590 EXPORT_SYMBOL(netdev_boot_setup_check
);
594 * netdev_boot_base - get address from boot time settings
595 * @prefix: prefix for network device
596 * @unit: id for network device
598 * Check boot time settings for the base address of device.
599 * The found settings are set for the device to be used
600 * later in the device probing.
601 * Returns 0 if no settings found.
603 unsigned long netdev_boot_base(const char *prefix
, int unit
)
605 const struct netdev_boot_setup
*s
= dev_boot_setup
;
609 sprintf(name
, "%s%d", prefix
, unit
);
612 * If device already registered then return base of 1
613 * to indicate not to probe for this interface
615 if (__dev_get_by_name(&init_net
, name
))
618 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
619 if (!strcmp(name
, s
[i
].name
))
620 return s
[i
].map
.base_addr
;
625 * Saves at boot time configured settings for any netdevice.
627 int __init
netdev_boot_setup(char *str
)
632 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
637 memset(&map
, 0, sizeof(map
));
641 map
.base_addr
= ints
[2];
643 map
.mem_start
= ints
[3];
645 map
.mem_end
= ints
[4];
647 /* Add new entry to the list */
648 return netdev_boot_setup_add(str
, &map
);
651 __setup("netdev=", netdev_boot_setup
);
653 /*******************************************************************************
655 Device Interface Subroutines
657 *******************************************************************************/
660 * __dev_get_by_name - find a device by its name
661 * @net: the applicable net namespace
662 * @name: name to find
664 * Find an interface by name. Must be called under RTNL semaphore
665 * or @dev_base_lock. If the name is found a pointer to the device
666 * is returned. If the name is not found then %NULL is returned. The
667 * reference counters are not incremented so the caller must be
668 * careful with locks.
671 struct net_device
*__dev_get_by_name(struct net
*net
, const char *name
)
673 struct net_device
*dev
;
674 struct hlist_head
*head
= dev_name_hash(net
, name
);
676 hlist_for_each_entry(dev
, head
, name_hlist
)
677 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
682 EXPORT_SYMBOL(__dev_get_by_name
);
685 * dev_get_by_name_rcu - find a device by its name
686 * @net: the applicable net namespace
687 * @name: name to find
689 * Find an interface by name.
690 * If the name is found a pointer to the device is returned.
691 * If the name is not found then %NULL is returned.
692 * The reference counters are not incremented so the caller must be
693 * careful with locks. The caller must hold RCU lock.
696 struct net_device
*dev_get_by_name_rcu(struct net
*net
, const char *name
)
698 struct net_device
*dev
;
699 struct hlist_head
*head
= dev_name_hash(net
, name
);
701 hlist_for_each_entry_rcu(dev
, head
, name_hlist
)
702 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
707 EXPORT_SYMBOL(dev_get_by_name_rcu
);
710 * dev_get_by_name - find a device by its name
711 * @net: the applicable net namespace
712 * @name: name to find
714 * Find an interface by name. This can be called from any
715 * context and does its own locking. The returned handle has
716 * the usage count incremented and the caller must use dev_put() to
717 * release it when it is no longer needed. %NULL is returned if no
718 * matching device is found.
721 struct net_device
*dev_get_by_name(struct net
*net
, const char *name
)
723 struct net_device
*dev
;
726 dev
= dev_get_by_name_rcu(net
, name
);
732 EXPORT_SYMBOL(dev_get_by_name
);
735 * __dev_get_by_index - find a device by its ifindex
736 * @net: the applicable net namespace
737 * @ifindex: index of device
739 * Search for an interface by index. Returns %NULL if the device
740 * is not found or a pointer to the device. The device has not
741 * had its reference counter increased so the caller must be careful
742 * about locking. The caller must hold either the RTNL semaphore
746 struct net_device
*__dev_get_by_index(struct net
*net
, int ifindex
)
748 struct net_device
*dev
;
749 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
751 hlist_for_each_entry(dev
, head
, index_hlist
)
752 if (dev
->ifindex
== ifindex
)
757 EXPORT_SYMBOL(__dev_get_by_index
);
760 * dev_get_by_index_rcu - find a device by its ifindex
761 * @net: the applicable net namespace
762 * @ifindex: index of device
764 * Search for an interface by index. Returns %NULL if the device
765 * is not found or a pointer to the device. The device has not
766 * had its reference counter increased so the caller must be careful
767 * about locking. The caller must hold RCU lock.
770 struct net_device
*dev_get_by_index_rcu(struct net
*net
, int ifindex
)
772 struct net_device
*dev
;
773 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
775 hlist_for_each_entry_rcu(dev
, head
, index_hlist
)
776 if (dev
->ifindex
== ifindex
)
781 EXPORT_SYMBOL(dev_get_by_index_rcu
);
785 * dev_get_by_index - find a device by its ifindex
786 * @net: the applicable net namespace
787 * @ifindex: index of device
789 * Search for an interface by index. Returns NULL if the device
790 * is not found or a pointer to the device. The device returned has
791 * had a reference added and the pointer is safe until the user calls
792 * dev_put to indicate they have finished with it.
795 struct net_device
*dev_get_by_index(struct net
*net
, int ifindex
)
797 struct net_device
*dev
;
800 dev
= dev_get_by_index_rcu(net
, ifindex
);
806 EXPORT_SYMBOL(dev_get_by_index
);
809 * netdev_get_name - get a netdevice name, knowing its ifindex.
810 * @net: network namespace
811 * @name: a pointer to the buffer where the name will be stored.
812 * @ifindex: the ifindex of the interface to get the name from.
814 * The use of raw_seqcount_begin() and cond_resched() before
815 * retrying is required as we want to give the writers a chance
816 * to complete when CONFIG_PREEMPT is not set.
818 int netdev_get_name(struct net
*net
, char *name
, int ifindex
)
820 struct net_device
*dev
;
824 seq
= raw_seqcount_begin(&devnet_rename_seq
);
826 dev
= dev_get_by_index_rcu(net
, ifindex
);
832 strcpy(name
, dev
->name
);
834 if (read_seqcount_retry(&devnet_rename_seq
, seq
)) {
843 * dev_getbyhwaddr_rcu - find a device by its hardware address
844 * @net: the applicable net namespace
845 * @type: media type of device
846 * @ha: hardware address
848 * Search for an interface by MAC address. Returns NULL if the device
849 * is not found or a pointer to the device.
850 * The caller must hold RCU or RTNL.
851 * The returned device has not had its ref count increased
852 * and the caller must therefore be careful about locking
856 struct net_device
*dev_getbyhwaddr_rcu(struct net
*net
, unsigned short type
,
859 struct net_device
*dev
;
861 for_each_netdev_rcu(net
, dev
)
862 if (dev
->type
== type
&&
863 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
868 EXPORT_SYMBOL(dev_getbyhwaddr_rcu
);
870 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
872 struct net_device
*dev
;
875 for_each_netdev(net
, dev
)
876 if (dev
->type
== type
)
881 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
883 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
885 struct net_device
*dev
, *ret
= NULL
;
888 for_each_netdev_rcu(net
, dev
)
889 if (dev
->type
== type
) {
897 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
900 * dev_get_by_flags_rcu - find any device with given flags
901 * @net: the applicable net namespace
902 * @if_flags: IFF_* values
903 * @mask: bitmask of bits in if_flags to check
905 * Search for any interface with the given flags. Returns NULL if a device
906 * is not found or a pointer to the device. Must be called inside
907 * rcu_read_lock(), and result refcount is unchanged.
910 struct net_device
*dev_get_by_flags_rcu(struct net
*net
, unsigned short if_flags
,
913 struct net_device
*dev
, *ret
;
916 for_each_netdev_rcu(net
, dev
) {
917 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
924 EXPORT_SYMBOL(dev_get_by_flags_rcu
);
927 * dev_valid_name - check if name is okay for network device
930 * Network device names need to be valid file names to
931 * to allow sysfs to work. We also disallow any kind of
934 bool dev_valid_name(const char *name
)
938 if (strlen(name
) >= IFNAMSIZ
)
940 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
944 if (*name
== '/' || isspace(*name
))
950 EXPORT_SYMBOL(dev_valid_name
);
953 * __dev_alloc_name - allocate a name for a device
954 * @net: network namespace to allocate the device name in
955 * @name: name format string
956 * @buf: scratch buffer and result name string
958 * Passed a format string - eg "lt%d" it will try and find a suitable
959 * id. It scans list of devices to build up a free map, then chooses
960 * the first empty slot. The caller must hold the dev_base or rtnl lock
961 * while allocating the name and adding the device in order to avoid
963 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
964 * Returns the number of the unit assigned or a negative errno code.
967 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
971 const int max_netdevices
= 8*PAGE_SIZE
;
972 unsigned long *inuse
;
973 struct net_device
*d
;
975 p
= strnchr(name
, IFNAMSIZ
-1, '%');
978 * Verify the string as this thing may have come from
979 * the user. There must be either one "%d" and no other "%"
982 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
985 /* Use one page as a bit array of possible slots */
986 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
990 for_each_netdev(net
, d
) {
991 if (!sscanf(d
->name
, name
, &i
))
993 if (i
< 0 || i
>= max_netdevices
)
996 /* avoid cases where sscanf is not exact inverse of printf */
997 snprintf(buf
, IFNAMSIZ
, name
, i
);
998 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
1002 i
= find_first_zero_bit(inuse
, max_netdevices
);
1003 free_page((unsigned long) inuse
);
1007 snprintf(buf
, IFNAMSIZ
, name
, i
);
1008 if (!__dev_get_by_name(net
, buf
))
1011 /* It is possible to run out of possible slots
1012 * when the name is long and there isn't enough space left
1013 * for the digits, or if all bits are used.
1019 * dev_alloc_name - allocate a name for a device
1021 * @name: name format string
1023 * Passed a format string - eg "lt%d" it will try and find a suitable
1024 * id. It scans list of devices to build up a free map, then chooses
1025 * the first empty slot. The caller must hold the dev_base or rtnl lock
1026 * while allocating the name and adding the device in order to avoid
1028 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1029 * Returns the number of the unit assigned or a negative errno code.
1032 int dev_alloc_name(struct net_device
*dev
, const char *name
)
1038 BUG_ON(!dev_net(dev
));
1040 ret
= __dev_alloc_name(net
, name
, buf
);
1042 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
1045 EXPORT_SYMBOL(dev_alloc_name
);
1047 static int dev_alloc_name_ns(struct net
*net
,
1048 struct net_device
*dev
,
1054 ret
= __dev_alloc_name(net
, name
, buf
);
1056 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
1060 static int dev_get_valid_name(struct net
*net
,
1061 struct net_device
*dev
,
1066 if (!dev_valid_name(name
))
1069 if (strchr(name
, '%'))
1070 return dev_alloc_name_ns(net
, dev
, name
);
1071 else if (__dev_get_by_name(net
, name
))
1073 else if (dev
->name
!= name
)
1074 strlcpy(dev
->name
, name
, IFNAMSIZ
);
1080 * dev_change_name - change name of a device
1082 * @newname: name (or format string) must be at least IFNAMSIZ
1084 * Change name of a device, can pass format strings "eth%d".
1087 int dev_change_name(struct net_device
*dev
, const char *newname
)
1089 unsigned char old_assign_type
;
1090 char oldname
[IFNAMSIZ
];
1096 BUG_ON(!dev_net(dev
));
1099 if (dev
->flags
& IFF_UP
)
1102 write_seqcount_begin(&devnet_rename_seq
);
1104 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0) {
1105 write_seqcount_end(&devnet_rename_seq
);
1109 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
1111 err
= dev_get_valid_name(net
, dev
, newname
);
1113 write_seqcount_end(&devnet_rename_seq
);
1117 if (oldname
[0] && !strchr(oldname
, '%'))
1118 netdev_info(dev
, "renamed from %s\n", oldname
);
1120 old_assign_type
= dev
->name_assign_type
;
1121 dev
->name_assign_type
= NET_NAME_RENAMED
;
1124 ret
= device_rename(&dev
->dev
, dev
->name
);
1126 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1127 dev
->name_assign_type
= old_assign_type
;
1128 write_seqcount_end(&devnet_rename_seq
);
1132 write_seqcount_end(&devnet_rename_seq
);
1134 netdev_adjacent_rename_links(dev
, oldname
);
1136 write_lock_bh(&dev_base_lock
);
1137 hlist_del_rcu(&dev
->name_hlist
);
1138 write_unlock_bh(&dev_base_lock
);
1142 write_lock_bh(&dev_base_lock
);
1143 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
1144 write_unlock_bh(&dev_base_lock
);
1146 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
1147 ret
= notifier_to_errno(ret
);
1150 /* err >= 0 after dev_alloc_name() or stores the first errno */
1153 write_seqcount_begin(&devnet_rename_seq
);
1154 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1155 memcpy(oldname
, newname
, IFNAMSIZ
);
1156 dev
->name_assign_type
= old_assign_type
;
1157 old_assign_type
= NET_NAME_RENAMED
;
1160 pr_err("%s: name change rollback failed: %d\n",
1169 * dev_set_alias - change ifalias of a device
1171 * @alias: name up to IFALIASZ
1172 * @len: limit of bytes to copy from info
1174 * Set ifalias for a device,
1176 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
1182 if (len
>= IFALIASZ
)
1186 kfree(dev
->ifalias
);
1187 dev
->ifalias
= NULL
;
1191 new_ifalias
= krealloc(dev
->ifalias
, len
+ 1, GFP_KERNEL
);
1194 dev
->ifalias
= new_ifalias
;
1196 strlcpy(dev
->ifalias
, alias
, len
+1);
1202 * netdev_features_change - device changes features
1203 * @dev: device to cause notification
1205 * Called to indicate a device has changed features.
1207 void netdev_features_change(struct net_device
*dev
)
1209 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1211 EXPORT_SYMBOL(netdev_features_change
);
1214 * netdev_state_change - device changes state
1215 * @dev: device to cause notification
1217 * Called to indicate a device has changed state. This function calls
1218 * the notifier chains for netdev_chain and sends a NEWLINK message
1219 * to the routing socket.
1221 void netdev_state_change(struct net_device
*dev
)
1223 if (dev
->flags
& IFF_UP
) {
1224 struct netdev_notifier_change_info change_info
;
1226 change_info
.flags_changed
= 0;
1227 call_netdevice_notifiers_info(NETDEV_CHANGE
, dev
,
1229 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0, GFP_KERNEL
);
1232 EXPORT_SYMBOL(netdev_state_change
);
1235 * netdev_notify_peers - notify network peers about existence of @dev
1236 * @dev: network device
1238 * Generate traffic such that interested network peers are aware of
1239 * @dev, such as by generating a gratuitous ARP. This may be used when
1240 * a device wants to inform the rest of the network about some sort of
1241 * reconfiguration such as a failover event or virtual machine
1244 void netdev_notify_peers(struct net_device
*dev
)
1247 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
, dev
);
1250 EXPORT_SYMBOL(netdev_notify_peers
);
1252 static int __dev_open(struct net_device
*dev
)
1254 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1259 if (!netif_device_present(dev
))
1262 /* Block netpoll from trying to do any rx path servicing.
1263 * If we don't do this there is a chance ndo_poll_controller
1264 * or ndo_poll may be running while we open the device
1266 netpoll_poll_disable(dev
);
1268 ret
= call_netdevice_notifiers(NETDEV_PRE_UP
, dev
);
1269 ret
= notifier_to_errno(ret
);
1273 set_bit(__LINK_STATE_START
, &dev
->state
);
1275 if (ops
->ndo_validate_addr
)
1276 ret
= ops
->ndo_validate_addr(dev
);
1278 if (!ret
&& ops
->ndo_open
)
1279 ret
= ops
->ndo_open(dev
);
1281 netpoll_poll_enable(dev
);
1284 clear_bit(__LINK_STATE_START
, &dev
->state
);
1286 dev
->flags
|= IFF_UP
;
1287 net_dmaengine_get();
1288 dev_set_rx_mode(dev
);
1290 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
1297 * dev_open - prepare an interface for use.
1298 * @dev: device to open
1300 * Takes a device from down to up state. The device's private open
1301 * function is invoked and then the multicast lists are loaded. Finally
1302 * the device is moved into the up state and a %NETDEV_UP message is
1303 * sent to the netdev notifier chain.
1305 * Calling this function on an active interface is a nop. On a failure
1306 * a negative errno code is returned.
1308 int dev_open(struct net_device
*dev
)
1312 if (dev
->flags
& IFF_UP
)
1315 ret
= __dev_open(dev
);
1319 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
, GFP_KERNEL
);
1320 call_netdevice_notifiers(NETDEV_UP
, dev
);
1324 EXPORT_SYMBOL(dev_open
);
1326 static int __dev_close_many(struct list_head
*head
)
1328 struct net_device
*dev
;
1333 list_for_each_entry(dev
, head
, close_list
) {
1334 /* Temporarily disable netpoll until the interface is down */
1335 netpoll_poll_disable(dev
);
1337 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1339 clear_bit(__LINK_STATE_START
, &dev
->state
);
1341 /* Synchronize to scheduled poll. We cannot touch poll list, it
1342 * can be even on different cpu. So just clear netif_running().
1344 * dev->stop() will invoke napi_disable() on all of it's
1345 * napi_struct instances on this device.
1347 smp_mb__after_atomic(); /* Commit netif_running(). */
1350 dev_deactivate_many(head
);
1352 list_for_each_entry(dev
, head
, close_list
) {
1353 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1356 * Call the device specific close. This cannot fail.
1357 * Only if device is UP
1359 * We allow it to be called even after a DETACH hot-plug
1365 dev
->flags
&= ~IFF_UP
;
1366 net_dmaengine_put();
1367 netpoll_poll_enable(dev
);
1373 static int __dev_close(struct net_device
*dev
)
1378 list_add(&dev
->close_list
, &single
);
1379 retval
= __dev_close_many(&single
);
1385 static int dev_close_many(struct list_head
*head
)
1387 struct net_device
*dev
, *tmp
;
1389 /* Remove the devices that don't need to be closed */
1390 list_for_each_entry_safe(dev
, tmp
, head
, close_list
)
1391 if (!(dev
->flags
& IFF_UP
))
1392 list_del_init(&dev
->close_list
);
1394 __dev_close_many(head
);
1396 list_for_each_entry_safe(dev
, tmp
, head
, close_list
) {
1397 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
, GFP_KERNEL
);
1398 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1399 list_del_init(&dev
->close_list
);
1406 * dev_close - shutdown an interface.
1407 * @dev: device to shutdown
1409 * This function moves an active device into down state. A
1410 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1411 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1414 int dev_close(struct net_device
*dev
)
1416 if (dev
->flags
& IFF_UP
) {
1419 list_add(&dev
->close_list
, &single
);
1420 dev_close_many(&single
);
1425 EXPORT_SYMBOL(dev_close
);
1429 * dev_disable_lro - disable Large Receive Offload on a device
1432 * Disable Large Receive Offload (LRO) on a net device. Must be
1433 * called under RTNL. This is needed if received packets may be
1434 * forwarded to another interface.
1436 void dev_disable_lro(struct net_device
*dev
)
1439 * If we're trying to disable lro on a vlan device
1440 * use the underlying physical device instead
1442 if (is_vlan_dev(dev
))
1443 dev
= vlan_dev_real_dev(dev
);
1445 /* the same for macvlan devices */
1446 if (netif_is_macvlan(dev
))
1447 dev
= macvlan_dev_real_dev(dev
);
1449 dev
->wanted_features
&= ~NETIF_F_LRO
;
1450 netdev_update_features(dev
);
1452 if (unlikely(dev
->features
& NETIF_F_LRO
))
1453 netdev_WARN(dev
, "failed to disable LRO!\n");
1455 EXPORT_SYMBOL(dev_disable_lro
);
1457 static int call_netdevice_notifier(struct notifier_block
*nb
, unsigned long val
,
1458 struct net_device
*dev
)
1460 struct netdev_notifier_info info
;
1462 netdev_notifier_info_init(&info
, dev
);
1463 return nb
->notifier_call(nb
, val
, &info
);
1466 static int dev_boot_phase
= 1;
1469 * register_netdevice_notifier - register a network notifier block
1472 * Register a notifier to be called when network device events occur.
1473 * The notifier passed is linked into the kernel structures and must
1474 * not be reused until it has been unregistered. A negative errno code
1475 * is returned on a failure.
1477 * When registered all registration and up events are replayed
1478 * to the new notifier to allow device to have a race free
1479 * view of the network device list.
1482 int register_netdevice_notifier(struct notifier_block
*nb
)
1484 struct net_device
*dev
;
1485 struct net_device
*last
;
1490 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1496 for_each_netdev(net
, dev
) {
1497 err
= call_netdevice_notifier(nb
, NETDEV_REGISTER
, dev
);
1498 err
= notifier_to_errno(err
);
1502 if (!(dev
->flags
& IFF_UP
))
1505 call_netdevice_notifier(nb
, NETDEV_UP
, dev
);
1516 for_each_netdev(net
, dev
) {
1520 if (dev
->flags
& IFF_UP
) {
1521 call_netdevice_notifier(nb
, NETDEV_GOING_DOWN
,
1523 call_netdevice_notifier(nb
, NETDEV_DOWN
, dev
);
1525 call_netdevice_notifier(nb
, NETDEV_UNREGISTER
, dev
);
1530 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1533 EXPORT_SYMBOL(register_netdevice_notifier
);
1536 * unregister_netdevice_notifier - unregister a network notifier block
1539 * Unregister a notifier previously registered by
1540 * register_netdevice_notifier(). The notifier is unlinked into the
1541 * kernel structures and may then be reused. A negative errno code
1542 * is returned on a failure.
1544 * After unregistering unregister and down device events are synthesized
1545 * for all devices on the device list to the removed notifier to remove
1546 * the need for special case cleanup code.
1549 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1551 struct net_device
*dev
;
1556 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1561 for_each_netdev(net
, dev
) {
1562 if (dev
->flags
& IFF_UP
) {
1563 call_netdevice_notifier(nb
, NETDEV_GOING_DOWN
,
1565 call_netdevice_notifier(nb
, NETDEV_DOWN
, dev
);
1567 call_netdevice_notifier(nb
, NETDEV_UNREGISTER
, dev
);
1574 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1577 * call_netdevice_notifiers_info - call all network notifier blocks
1578 * @val: value passed unmodified to notifier function
1579 * @dev: net_device pointer passed unmodified to notifier function
1580 * @info: notifier information data
1582 * Call all network notifier blocks. Parameters and return value
1583 * are as for raw_notifier_call_chain().
1586 static int call_netdevice_notifiers_info(unsigned long val
,
1587 struct net_device
*dev
,
1588 struct netdev_notifier_info
*info
)
1591 netdev_notifier_info_init(info
, dev
);
1592 return raw_notifier_call_chain(&netdev_chain
, val
, info
);
1596 * call_netdevice_notifiers - call all network notifier blocks
1597 * @val: value passed unmodified to notifier function
1598 * @dev: net_device pointer passed unmodified to notifier function
1600 * Call all network notifier blocks. Parameters and return value
1601 * are as for raw_notifier_call_chain().
1604 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1606 struct netdev_notifier_info info
;
1608 return call_netdevice_notifiers_info(val
, dev
, &info
);
1610 EXPORT_SYMBOL(call_netdevice_notifiers
);
1612 static struct static_key netstamp_needed __read_mostly
;
1613 #ifdef HAVE_JUMP_LABEL
1614 /* We are not allowed to call static_key_slow_dec() from irq context
1615 * If net_disable_timestamp() is called from irq context, defer the
1616 * static_key_slow_dec() calls.
1618 static atomic_t netstamp_needed_deferred
;
1621 void net_enable_timestamp(void)
1623 #ifdef HAVE_JUMP_LABEL
1624 int deferred
= atomic_xchg(&netstamp_needed_deferred
, 0);
1628 static_key_slow_dec(&netstamp_needed
);
1632 static_key_slow_inc(&netstamp_needed
);
1634 EXPORT_SYMBOL(net_enable_timestamp
);
1636 void net_disable_timestamp(void)
1638 #ifdef HAVE_JUMP_LABEL
1639 if (in_interrupt()) {
1640 atomic_inc(&netstamp_needed_deferred
);
1644 static_key_slow_dec(&netstamp_needed
);
1646 EXPORT_SYMBOL(net_disable_timestamp
);
1648 static inline void net_timestamp_set(struct sk_buff
*skb
)
1650 skb
->tstamp
.tv64
= 0;
1651 if (static_key_false(&netstamp_needed
))
1652 __net_timestamp(skb
);
1655 #define net_timestamp_check(COND, SKB) \
1656 if (static_key_false(&netstamp_needed)) { \
1657 if ((COND) && !(SKB)->tstamp.tv64) \
1658 __net_timestamp(SKB); \
1661 bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb)
1665 if (!(dev
->flags
& IFF_UP
))
1668 len
= dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
;
1669 if (skb
->len
<= len
)
1672 /* if TSO is enabled, we don't care about the length as the packet
1673 * could be forwarded without being segmented before
1675 if (skb_is_gso(skb
))
1680 EXPORT_SYMBOL_GPL(is_skb_forwardable
);
1682 int __dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1684 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
1685 if (skb_copy_ubufs(skb
, GFP_ATOMIC
)) {
1686 atomic_long_inc(&dev
->rx_dropped
);
1692 if (unlikely(!is_skb_forwardable(dev
, skb
))) {
1693 atomic_long_inc(&dev
->rx_dropped
);
1698 skb_scrub_packet(skb
, true);
1699 skb
->protocol
= eth_type_trans(skb
, dev
);
1703 EXPORT_SYMBOL_GPL(__dev_forward_skb
);
1706 * dev_forward_skb - loopback an skb to another netif
1708 * @dev: destination network device
1709 * @skb: buffer to forward
1712 * NET_RX_SUCCESS (no congestion)
1713 * NET_RX_DROP (packet was dropped, but freed)
1715 * dev_forward_skb can be used for injecting an skb from the
1716 * start_xmit function of one device into the receive queue
1717 * of another device.
1719 * The receiving device may be in another namespace, so
1720 * we have to clear all information in the skb that could
1721 * impact namespace isolation.
1723 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1725 return __dev_forward_skb(dev
, skb
) ?: netif_rx_internal(skb
);
1727 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1729 static inline int deliver_skb(struct sk_buff
*skb
,
1730 struct packet_type
*pt_prev
,
1731 struct net_device
*orig_dev
)
1733 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
1735 atomic_inc(&skb
->users
);
1736 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1739 static inline bool skb_loop_sk(struct packet_type
*ptype
, struct sk_buff
*skb
)
1741 if (!ptype
->af_packet_priv
|| !skb
->sk
)
1744 if (ptype
->id_match
)
1745 return ptype
->id_match(ptype
, skb
->sk
);
1746 else if ((struct sock
*)ptype
->af_packet_priv
== skb
->sk
)
1753 * Support routine. Sends outgoing frames to any network
1754 * taps currently in use.
1757 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1759 struct packet_type
*ptype
;
1760 struct sk_buff
*skb2
= NULL
;
1761 struct packet_type
*pt_prev
= NULL
;
1764 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1765 /* Never send packets back to the socket
1766 * they originated from - MvS (miquels@drinkel.ow.org)
1768 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1769 (!skb_loop_sk(ptype
, skb
))) {
1771 deliver_skb(skb2
, pt_prev
, skb
->dev
);
1776 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1780 net_timestamp_set(skb2
);
1782 /* skb->nh should be correctly
1783 set by sender, so that the second statement is
1784 just protection against buggy protocols.
1786 skb_reset_mac_header(skb2
);
1788 if (skb_network_header(skb2
) < skb2
->data
||
1789 skb_network_header(skb2
) > skb_tail_pointer(skb2
)) {
1790 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1791 ntohs(skb2
->protocol
),
1793 skb_reset_network_header(skb2
);
1796 skb2
->transport_header
= skb2
->network_header
;
1797 skb2
->pkt_type
= PACKET_OUTGOING
;
1802 pt_prev
->func(skb2
, skb
->dev
, pt_prev
, skb
->dev
);
1807 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1808 * @dev: Network device
1809 * @txq: number of queues available
1811 * If real_num_tx_queues is changed the tc mappings may no longer be
1812 * valid. To resolve this verify the tc mapping remains valid and if
1813 * not NULL the mapping. With no priorities mapping to this
1814 * offset/count pair it will no longer be used. In the worst case TC0
1815 * is invalid nothing can be done so disable priority mappings. If is
1816 * expected that drivers will fix this mapping if they can before
1817 * calling netif_set_real_num_tx_queues.
1819 static void netif_setup_tc(struct net_device
*dev
, unsigned int txq
)
1822 struct netdev_tc_txq
*tc
= &dev
->tc_to_txq
[0];
1824 /* If TC0 is invalidated disable TC mapping */
1825 if (tc
->offset
+ tc
->count
> txq
) {
1826 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1831 /* Invalidated prio to tc mappings set to TC0 */
1832 for (i
= 1; i
< TC_BITMASK
+ 1; i
++) {
1833 int q
= netdev_get_prio_tc_map(dev
, i
);
1835 tc
= &dev
->tc_to_txq
[q
];
1836 if (tc
->offset
+ tc
->count
> txq
) {
1837 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1839 netdev_set_prio_tc_map(dev
, i
, 0);
1845 static DEFINE_MUTEX(xps_map_mutex
);
1846 #define xmap_dereference(P) \
1847 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1849 static struct xps_map
*remove_xps_queue(struct xps_dev_maps
*dev_maps
,
1852 struct xps_map
*map
= NULL
;
1856 map
= xmap_dereference(dev_maps
->cpu_map
[cpu
]);
1858 for (pos
= 0; map
&& pos
< map
->len
; pos
++) {
1859 if (map
->queues
[pos
] == index
) {
1861 map
->queues
[pos
] = map
->queues
[--map
->len
];
1863 RCU_INIT_POINTER(dev_maps
->cpu_map
[cpu
], NULL
);
1864 kfree_rcu(map
, rcu
);
1874 static void netif_reset_xps_queues_gt(struct net_device
*dev
, u16 index
)
1876 struct xps_dev_maps
*dev_maps
;
1878 bool active
= false;
1880 mutex_lock(&xps_map_mutex
);
1881 dev_maps
= xmap_dereference(dev
->xps_maps
);
1886 for_each_possible_cpu(cpu
) {
1887 for (i
= index
; i
< dev
->num_tx_queues
; i
++) {
1888 if (!remove_xps_queue(dev_maps
, cpu
, i
))
1891 if (i
== dev
->num_tx_queues
)
1896 RCU_INIT_POINTER(dev
->xps_maps
, NULL
);
1897 kfree_rcu(dev_maps
, rcu
);
1900 for (i
= index
; i
< dev
->num_tx_queues
; i
++)
1901 netdev_queue_numa_node_write(netdev_get_tx_queue(dev
, i
),
1905 mutex_unlock(&xps_map_mutex
);
1908 static struct xps_map
*expand_xps_map(struct xps_map
*map
,
1911 struct xps_map
*new_map
;
1912 int alloc_len
= XPS_MIN_MAP_ALLOC
;
1915 for (pos
= 0; map
&& pos
< map
->len
; pos
++) {
1916 if (map
->queues
[pos
] != index
)
1921 /* Need to add queue to this CPU's existing map */
1923 if (pos
< map
->alloc_len
)
1926 alloc_len
= map
->alloc_len
* 2;
1929 /* Need to allocate new map to store queue on this CPU's map */
1930 new_map
= kzalloc_node(XPS_MAP_SIZE(alloc_len
), GFP_KERNEL
,
1935 for (i
= 0; i
< pos
; i
++)
1936 new_map
->queues
[i
] = map
->queues
[i
];
1937 new_map
->alloc_len
= alloc_len
;
1943 int netif_set_xps_queue(struct net_device
*dev
, const struct cpumask
*mask
,
1946 struct xps_dev_maps
*dev_maps
, *new_dev_maps
= NULL
;
1947 struct xps_map
*map
, *new_map
;
1948 int maps_sz
= max_t(unsigned int, XPS_DEV_MAPS_SIZE
, L1_CACHE_BYTES
);
1949 int cpu
, numa_node_id
= -2;
1950 bool active
= false;
1952 mutex_lock(&xps_map_mutex
);
1954 dev_maps
= xmap_dereference(dev
->xps_maps
);
1956 /* allocate memory for queue storage */
1957 for_each_online_cpu(cpu
) {
1958 if (!cpumask_test_cpu(cpu
, mask
))
1962 new_dev_maps
= kzalloc(maps_sz
, GFP_KERNEL
);
1963 if (!new_dev_maps
) {
1964 mutex_unlock(&xps_map_mutex
);
1968 map
= dev_maps
? xmap_dereference(dev_maps
->cpu_map
[cpu
]) :
1971 map
= expand_xps_map(map
, cpu
, index
);
1975 RCU_INIT_POINTER(new_dev_maps
->cpu_map
[cpu
], map
);
1979 goto out_no_new_maps
;
1981 for_each_possible_cpu(cpu
) {
1982 if (cpumask_test_cpu(cpu
, mask
) && cpu_online(cpu
)) {
1983 /* add queue to CPU maps */
1986 map
= xmap_dereference(new_dev_maps
->cpu_map
[cpu
]);
1987 while ((pos
< map
->len
) && (map
->queues
[pos
] != index
))
1990 if (pos
== map
->len
)
1991 map
->queues
[map
->len
++] = index
;
1993 if (numa_node_id
== -2)
1994 numa_node_id
= cpu_to_node(cpu
);
1995 else if (numa_node_id
!= cpu_to_node(cpu
))
1998 } else if (dev_maps
) {
1999 /* fill in the new device map from the old device map */
2000 map
= xmap_dereference(dev_maps
->cpu_map
[cpu
]);
2001 RCU_INIT_POINTER(new_dev_maps
->cpu_map
[cpu
], map
);
2006 rcu_assign_pointer(dev
->xps_maps
, new_dev_maps
);
2008 /* Cleanup old maps */
2010 for_each_possible_cpu(cpu
) {
2011 new_map
= xmap_dereference(new_dev_maps
->cpu_map
[cpu
]);
2012 map
= xmap_dereference(dev_maps
->cpu_map
[cpu
]);
2013 if (map
&& map
!= new_map
)
2014 kfree_rcu(map
, rcu
);
2017 kfree_rcu(dev_maps
, rcu
);
2020 dev_maps
= new_dev_maps
;
2024 /* update Tx queue numa node */
2025 netdev_queue_numa_node_write(netdev_get_tx_queue(dev
, index
),
2026 (numa_node_id
>= 0) ? numa_node_id
:
2032 /* removes queue from unused CPUs */
2033 for_each_possible_cpu(cpu
) {
2034 if (cpumask_test_cpu(cpu
, mask
) && cpu_online(cpu
))
2037 if (remove_xps_queue(dev_maps
, cpu
, index
))
2041 /* free map if not active */
2043 RCU_INIT_POINTER(dev
->xps_maps
, NULL
);
2044 kfree_rcu(dev_maps
, rcu
);
2048 mutex_unlock(&xps_map_mutex
);
2052 /* remove any maps that we added */
2053 for_each_possible_cpu(cpu
) {
2054 new_map
= xmap_dereference(new_dev_maps
->cpu_map
[cpu
]);
2055 map
= dev_maps
? xmap_dereference(dev_maps
->cpu_map
[cpu
]) :
2057 if (new_map
&& new_map
!= map
)
2061 mutex_unlock(&xps_map_mutex
);
2063 kfree(new_dev_maps
);
2066 EXPORT_SYMBOL(netif_set_xps_queue
);
2070 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2071 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2073 int netif_set_real_num_tx_queues(struct net_device
*dev
, unsigned int txq
)
2077 if (txq
< 1 || txq
> dev
->num_tx_queues
)
2080 if (dev
->reg_state
== NETREG_REGISTERED
||
2081 dev
->reg_state
== NETREG_UNREGISTERING
) {
2084 rc
= netdev_queue_update_kobjects(dev
, dev
->real_num_tx_queues
,
2090 netif_setup_tc(dev
, txq
);
2092 if (txq
< dev
->real_num_tx_queues
) {
2093 qdisc_reset_all_tx_gt(dev
, txq
);
2095 netif_reset_xps_queues_gt(dev
, txq
);
2100 dev
->real_num_tx_queues
= txq
;
2103 EXPORT_SYMBOL(netif_set_real_num_tx_queues
);
2107 * netif_set_real_num_rx_queues - set actual number of RX queues used
2108 * @dev: Network device
2109 * @rxq: Actual number of RX queues
2111 * This must be called either with the rtnl_lock held or before
2112 * registration of the net device. Returns 0 on success, or a
2113 * negative error code. If called before registration, it always
2116 int netif_set_real_num_rx_queues(struct net_device
*dev
, unsigned int rxq
)
2120 if (rxq
< 1 || rxq
> dev
->num_rx_queues
)
2123 if (dev
->reg_state
== NETREG_REGISTERED
) {
2126 rc
= net_rx_queue_update_kobjects(dev
, dev
->real_num_rx_queues
,
2132 dev
->real_num_rx_queues
= rxq
;
2135 EXPORT_SYMBOL(netif_set_real_num_rx_queues
);
2139 * netif_get_num_default_rss_queues - default number of RSS queues
2141 * This routine should set an upper limit on the number of RSS queues
2142 * used by default by multiqueue devices.
2144 int netif_get_num_default_rss_queues(void)
2146 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES
, num_online_cpus());
2148 EXPORT_SYMBOL(netif_get_num_default_rss_queues
);
2150 static inline void __netif_reschedule(struct Qdisc
*q
)
2152 struct softnet_data
*sd
;
2153 unsigned long flags
;
2155 local_irq_save(flags
);
2156 sd
= &__get_cpu_var(softnet_data
);
2157 q
->next_sched
= NULL
;
2158 *sd
->output_queue_tailp
= q
;
2159 sd
->output_queue_tailp
= &q
->next_sched
;
2160 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
2161 local_irq_restore(flags
);
2164 void __netif_schedule(struct Qdisc
*q
)
2166 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
2167 __netif_reschedule(q
);
2169 EXPORT_SYMBOL(__netif_schedule
);
2171 struct dev_kfree_skb_cb
{
2172 enum skb_free_reason reason
;
2175 static struct dev_kfree_skb_cb
*get_kfree_skb_cb(const struct sk_buff
*skb
)
2177 return (struct dev_kfree_skb_cb
*)skb
->cb
;
2180 void netif_schedule_queue(struct netdev_queue
*txq
)
2183 if (!(txq
->state
& QUEUE_STATE_ANY_XOFF
)) {
2184 struct Qdisc
*q
= rcu_dereference(txq
->qdisc
);
2186 __netif_schedule(q
);
2190 EXPORT_SYMBOL(netif_schedule_queue
);
2193 * netif_wake_subqueue - allow sending packets on subqueue
2194 * @dev: network device
2195 * @queue_index: sub queue index
2197 * Resume individual transmit queue of a device with multiple transmit queues.
2199 void netif_wake_subqueue(struct net_device
*dev
, u16 queue_index
)
2201 struct netdev_queue
*txq
= netdev_get_tx_queue(dev
, queue_index
);
2203 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF
, &txq
->state
)) {
2207 q
= rcu_dereference(txq
->qdisc
);
2208 __netif_schedule(q
);
2212 EXPORT_SYMBOL(netif_wake_subqueue
);
2214 void netif_tx_wake_queue(struct netdev_queue
*dev_queue
)
2216 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF
, &dev_queue
->state
)) {
2220 q
= rcu_dereference(dev_queue
->qdisc
);
2221 __netif_schedule(q
);
2225 EXPORT_SYMBOL(netif_tx_wake_queue
);
2227 void __dev_kfree_skb_irq(struct sk_buff
*skb
, enum skb_free_reason reason
)
2229 unsigned long flags
;
2231 if (likely(atomic_read(&skb
->users
) == 1)) {
2233 atomic_set(&skb
->users
, 0);
2234 } else if (likely(!atomic_dec_and_test(&skb
->users
))) {
2237 get_kfree_skb_cb(skb
)->reason
= reason
;
2238 local_irq_save(flags
);
2239 skb
->next
= __this_cpu_read(softnet_data
.completion_queue
);
2240 __this_cpu_write(softnet_data
.completion_queue
, skb
);
2241 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
2242 local_irq_restore(flags
);
2244 EXPORT_SYMBOL(__dev_kfree_skb_irq
);
2246 void __dev_kfree_skb_any(struct sk_buff
*skb
, enum skb_free_reason reason
)
2248 if (in_irq() || irqs_disabled())
2249 __dev_kfree_skb_irq(skb
, reason
);
2253 EXPORT_SYMBOL(__dev_kfree_skb_any
);
2257 * netif_device_detach - mark device as removed
2258 * @dev: network device
2260 * Mark device as removed from system and therefore no longer available.
2262 void netif_device_detach(struct net_device
*dev
)
2264 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
2265 netif_running(dev
)) {
2266 netif_tx_stop_all_queues(dev
);
2269 EXPORT_SYMBOL(netif_device_detach
);
2272 * netif_device_attach - mark device as attached
2273 * @dev: network device
2275 * Mark device as attached from system and restart if needed.
2277 void netif_device_attach(struct net_device
*dev
)
2279 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
2280 netif_running(dev
)) {
2281 netif_tx_wake_all_queues(dev
);
2282 __netdev_watchdog_up(dev
);
2285 EXPORT_SYMBOL(netif_device_attach
);
2287 static void skb_warn_bad_offload(const struct sk_buff
*skb
)
2289 static const netdev_features_t null_features
= 0;
2290 struct net_device
*dev
= skb
->dev
;
2291 const char *driver
= "";
2293 if (!net_ratelimit())
2296 if (dev
&& dev
->dev
.parent
)
2297 driver
= dev_driver_string(dev
->dev
.parent
);
2299 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2300 "gso_type=%d ip_summed=%d\n",
2301 driver
, dev
? &dev
->features
: &null_features
,
2302 skb
->sk
? &skb
->sk
->sk_route_caps
: &null_features
,
2303 skb
->len
, skb
->data_len
, skb_shinfo(skb
)->gso_size
,
2304 skb_shinfo(skb
)->gso_type
, skb
->ip_summed
);
2308 * Invalidate hardware checksum when packet is to be mangled, and
2309 * complete checksum manually on outgoing path.
2311 int skb_checksum_help(struct sk_buff
*skb
)
2314 int ret
= 0, offset
;
2316 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
2317 goto out_set_summed
;
2319 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
2320 skb_warn_bad_offload(skb
);
2324 /* Before computing a checksum, we should make sure no frag could
2325 * be modified by an external entity : checksum could be wrong.
2327 if (skb_has_shared_frag(skb
)) {
2328 ret
= __skb_linearize(skb
);
2333 offset
= skb_checksum_start_offset(skb
);
2334 BUG_ON(offset
>= skb_headlen(skb
));
2335 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
2337 offset
+= skb
->csum_offset
;
2338 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
2340 if (skb_cloned(skb
) &&
2341 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
2342 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2347 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
2349 skb
->ip_summed
= CHECKSUM_NONE
;
2353 EXPORT_SYMBOL(skb_checksum_help
);
2355 __be16
skb_network_protocol(struct sk_buff
*skb
, int *depth
)
2357 unsigned int vlan_depth
= skb
->mac_len
;
2358 __be16 type
= skb
->protocol
;
2360 /* Tunnel gso handlers can set protocol to ethernet. */
2361 if (type
== htons(ETH_P_TEB
)) {
2364 if (unlikely(!pskb_may_pull(skb
, sizeof(struct ethhdr
))))
2367 eth
= (struct ethhdr
*)skb_mac_header(skb
);
2368 type
= eth
->h_proto
;
2371 /* if skb->protocol is 802.1Q/AD then the header should already be
2372 * present at mac_len - VLAN_HLEN (if mac_len > 0), or at
2373 * ETH_HLEN otherwise
2375 if (type
== htons(ETH_P_8021Q
) || type
== htons(ETH_P_8021AD
)) {
2377 if (WARN_ON(vlan_depth
< VLAN_HLEN
))
2379 vlan_depth
-= VLAN_HLEN
;
2381 vlan_depth
= ETH_HLEN
;
2384 struct vlan_hdr
*vh
;
2386 if (unlikely(!pskb_may_pull(skb
,
2387 vlan_depth
+ VLAN_HLEN
)))
2390 vh
= (struct vlan_hdr
*)(skb
->data
+ vlan_depth
);
2391 type
= vh
->h_vlan_encapsulated_proto
;
2392 vlan_depth
+= VLAN_HLEN
;
2393 } while (type
== htons(ETH_P_8021Q
) ||
2394 type
== htons(ETH_P_8021AD
));
2397 *depth
= vlan_depth
;
2403 * skb_mac_gso_segment - mac layer segmentation handler.
2404 * @skb: buffer to segment
2405 * @features: features for the output path (see dev->features)
2407 struct sk_buff
*skb_mac_gso_segment(struct sk_buff
*skb
,
2408 netdev_features_t features
)
2410 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
2411 struct packet_offload
*ptype
;
2412 int vlan_depth
= skb
->mac_len
;
2413 __be16 type
= skb_network_protocol(skb
, &vlan_depth
);
2415 if (unlikely(!type
))
2416 return ERR_PTR(-EINVAL
);
2418 __skb_pull(skb
, vlan_depth
);
2421 list_for_each_entry_rcu(ptype
, &offload_base
, list
) {
2422 if (ptype
->type
== type
&& ptype
->callbacks
.gso_segment
) {
2423 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
2426 err
= ptype
->callbacks
.gso_send_check(skb
);
2427 segs
= ERR_PTR(err
);
2428 if (err
|| skb_gso_ok(skb
, features
))
2430 __skb_push(skb
, (skb
->data
-
2431 skb_network_header(skb
)));
2433 segs
= ptype
->callbacks
.gso_segment(skb
, features
);
2439 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
2443 EXPORT_SYMBOL(skb_mac_gso_segment
);
2446 /* openvswitch calls this on rx path, so we need a different check.
2448 static inline bool skb_needs_check(struct sk_buff
*skb
, bool tx_path
)
2451 return skb
->ip_summed
!= CHECKSUM_PARTIAL
;
2453 return skb
->ip_summed
== CHECKSUM_NONE
;
2457 * __skb_gso_segment - Perform segmentation on skb.
2458 * @skb: buffer to segment
2459 * @features: features for the output path (see dev->features)
2460 * @tx_path: whether it is called in TX path
2462 * This function segments the given skb and returns a list of segments.
2464 * It may return NULL if the skb requires no segmentation. This is
2465 * only possible when GSO is used for verifying header integrity.
2467 struct sk_buff
*__skb_gso_segment(struct sk_buff
*skb
,
2468 netdev_features_t features
, bool tx_path
)
2470 if (unlikely(skb_needs_check(skb
, tx_path
))) {
2473 skb_warn_bad_offload(skb
);
2475 err
= skb_cow_head(skb
, 0);
2477 return ERR_PTR(err
);
2480 SKB_GSO_CB(skb
)->mac_offset
= skb_headroom(skb
);
2481 SKB_GSO_CB(skb
)->encap_level
= 0;
2483 skb_reset_mac_header(skb
);
2484 skb_reset_mac_len(skb
);
2486 return skb_mac_gso_segment(skb
, features
);
2488 EXPORT_SYMBOL(__skb_gso_segment
);
2490 /* Take action when hardware reception checksum errors are detected. */
2492 void netdev_rx_csum_fault(struct net_device
*dev
)
2494 if (net_ratelimit()) {
2495 pr_err("%s: hw csum failure\n", dev
? dev
->name
: "<unknown>");
2499 EXPORT_SYMBOL(netdev_rx_csum_fault
);
2502 /* Actually, we should eliminate this check as soon as we know, that:
2503 * 1. IOMMU is present and allows to map all the memory.
2504 * 2. No high memory really exists on this machine.
2507 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
2509 #ifdef CONFIG_HIGHMEM
2511 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
2512 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2513 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2514 if (PageHighMem(skb_frag_page(frag
)))
2519 if (PCI_DMA_BUS_IS_PHYS
) {
2520 struct device
*pdev
= dev
->dev
.parent
;
2524 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2525 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2526 dma_addr_t addr
= page_to_phys(skb_frag_page(frag
));
2527 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
2535 /* If MPLS offload request, verify we are testing hardware MPLS features
2536 * instead of standard features for the netdev.
2538 #ifdef CONFIG_NET_MPLS_GSO
2539 static netdev_features_t
net_mpls_features(struct sk_buff
*skb
,
2540 netdev_features_t features
,
2543 if (type
== htons(ETH_P_MPLS_UC
) || type
== htons(ETH_P_MPLS_MC
))
2544 features
&= skb
->dev
->mpls_features
;
2549 static netdev_features_t
net_mpls_features(struct sk_buff
*skb
,
2550 netdev_features_t features
,
2557 static netdev_features_t
harmonize_features(struct sk_buff
*skb
,
2558 netdev_features_t features
)
2563 type
= skb_network_protocol(skb
, &tmp
);
2564 features
= net_mpls_features(skb
, features
, type
);
2566 if (skb
->ip_summed
!= CHECKSUM_NONE
&&
2567 !can_checksum_protocol(features
, type
)) {
2568 features
&= ~NETIF_F_ALL_CSUM
;
2569 } else if (illegal_highdma(skb
->dev
, skb
)) {
2570 features
&= ~NETIF_F_SG
;
2576 netdev_features_t
netif_skb_features(struct sk_buff
*skb
)
2578 __be16 protocol
= skb
->protocol
;
2579 netdev_features_t features
= skb
->dev
->features
;
2581 if (skb_shinfo(skb
)->gso_segs
> skb
->dev
->gso_max_segs
)
2582 features
&= ~NETIF_F_GSO_MASK
;
2584 if (protocol
== htons(ETH_P_8021Q
) || protocol
== htons(ETH_P_8021AD
)) {
2585 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
2586 protocol
= veh
->h_vlan_encapsulated_proto
;
2587 } else if (!vlan_tx_tag_present(skb
)) {
2588 return harmonize_features(skb
, features
);
2591 features
= netdev_intersect_features(features
,
2592 skb
->dev
->vlan_features
|
2593 NETIF_F_HW_VLAN_CTAG_TX
|
2594 NETIF_F_HW_VLAN_STAG_TX
);
2596 if (protocol
== htons(ETH_P_8021Q
) || protocol
== htons(ETH_P_8021AD
))
2597 features
= netdev_intersect_features(features
,
2602 NETIF_F_HW_VLAN_CTAG_TX
|
2603 NETIF_F_HW_VLAN_STAG_TX
);
2605 return harmonize_features(skb
, features
);
2607 EXPORT_SYMBOL(netif_skb_features
);
2609 static int xmit_one(struct sk_buff
*skb
, struct net_device
*dev
,
2610 struct netdev_queue
*txq
, bool more
)
2615 if (!list_empty(&ptype_all
))
2616 dev_queue_xmit_nit(skb
, dev
);
2619 trace_net_dev_start_xmit(skb
, dev
);
2620 rc
= netdev_start_xmit(skb
, dev
, txq
, more
);
2621 trace_net_dev_xmit(skb
, rc
, dev
, len
);
2626 struct sk_buff
*dev_hard_start_xmit(struct sk_buff
*first
, struct net_device
*dev
,
2627 struct netdev_queue
*txq
, int *ret
)
2629 struct sk_buff
*skb
= first
;
2630 int rc
= NETDEV_TX_OK
;
2633 struct sk_buff
*next
= skb
->next
;
2636 rc
= xmit_one(skb
, dev
, txq
, next
!= NULL
);
2637 if (unlikely(!dev_xmit_complete(rc
))) {
2643 if (netif_xmit_stopped(txq
) && skb
) {
2644 rc
= NETDEV_TX_BUSY
;
2654 struct sk_buff
*validate_xmit_vlan(struct sk_buff
*skb
, netdev_features_t features
)
2656 if (vlan_tx_tag_present(skb
) &&
2657 !vlan_hw_offload_capable(features
, skb
->vlan_proto
)) {
2658 skb
= __vlan_put_tag(skb
, skb
->vlan_proto
,
2659 vlan_tx_tag_get(skb
));
2666 struct sk_buff
*validate_xmit_skb(struct sk_buff
*skb
, struct net_device
*dev
)
2668 netdev_features_t features
;
2673 /* If device doesn't need skb->dst, release it right now while
2674 * its hot in this cpu cache
2676 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2679 features
= netif_skb_features(skb
);
2680 skb
= validate_xmit_vlan(skb
, features
);
2684 /* If encapsulation offload request, verify we are testing
2685 * hardware encapsulation features instead of standard
2686 * features for the netdev
2688 if (skb
->encapsulation
)
2689 features
&= dev
->hw_enc_features
;
2691 if (netif_needs_gso(skb
, features
)) {
2692 struct sk_buff
*segs
;
2694 segs
= skb_gso_segment(skb
, features
);
2700 if (skb_needs_linearize(skb
, features
) &&
2701 __skb_linearize(skb
))
2704 /* If packet is not checksummed and device does not
2705 * support checksumming for this protocol, complete
2706 * checksumming here.
2708 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2709 if (skb
->encapsulation
)
2710 skb_set_inner_transport_header(skb
,
2711 skb_checksum_start_offset(skb
));
2713 skb_set_transport_header(skb
,
2714 skb_checksum_start_offset(skb
));
2715 if (!(features
& NETIF_F_ALL_CSUM
) &&
2716 skb_checksum_help(skb
))
2729 static void qdisc_pkt_len_init(struct sk_buff
*skb
)
2731 const struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
2733 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
2735 /* To get more precise estimation of bytes sent on wire,
2736 * we add to pkt_len the headers size of all segments
2738 if (shinfo
->gso_size
) {
2739 unsigned int hdr_len
;
2740 u16 gso_segs
= shinfo
->gso_segs
;
2742 /* mac layer + network layer */
2743 hdr_len
= skb_transport_header(skb
) - skb_mac_header(skb
);
2745 /* + transport layer */
2746 if (likely(shinfo
->gso_type
& (SKB_GSO_TCPV4
| SKB_GSO_TCPV6
)))
2747 hdr_len
+= tcp_hdrlen(skb
);
2749 hdr_len
+= sizeof(struct udphdr
);
2751 if (shinfo
->gso_type
& SKB_GSO_DODGY
)
2752 gso_segs
= DIV_ROUND_UP(skb
->len
- hdr_len
,
2755 qdisc_skb_cb(skb
)->pkt_len
+= (gso_segs
- 1) * hdr_len
;
2759 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2760 struct net_device
*dev
,
2761 struct netdev_queue
*txq
)
2763 spinlock_t
*root_lock
= qdisc_lock(q
);
2767 qdisc_pkt_len_init(skb
);
2768 qdisc_calculate_pkt_len(skb
, q
);
2770 * Heuristic to force contended enqueues to serialize on a
2771 * separate lock before trying to get qdisc main lock.
2772 * This permits __QDISC___STATE_RUNNING owner to get the lock more
2773 * often and dequeue packets faster.
2775 contended
= qdisc_is_running(q
);
2776 if (unlikely(contended
))
2777 spin_lock(&q
->busylock
);
2779 spin_lock(root_lock
);
2780 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2783 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2784 qdisc_run_begin(q
)) {
2786 * This is a work-conserving queue; there are no old skbs
2787 * waiting to be sent out; and the qdisc is not running -
2788 * xmit the skb directly.
2790 if (!(dev
->priv_flags
& IFF_XMIT_DST_RELEASE
))
2793 qdisc_bstats_update(q
, skb
);
2795 skb
= validate_xmit_skb(skb
, dev
);
2796 if (skb
&& sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
)) {
2797 if (unlikely(contended
)) {
2798 spin_unlock(&q
->busylock
);
2805 rc
= NET_XMIT_SUCCESS
;
2808 rc
= q
->enqueue(skb
, q
) & NET_XMIT_MASK
;
2809 if (qdisc_run_begin(q
)) {
2810 if (unlikely(contended
)) {
2811 spin_unlock(&q
->busylock
);
2817 spin_unlock(root_lock
);
2818 if (unlikely(contended
))
2819 spin_unlock(&q
->busylock
);
2823 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2824 static void skb_update_prio(struct sk_buff
*skb
)
2826 struct netprio_map
*map
= rcu_dereference_bh(skb
->dev
->priomap
);
2828 if (!skb
->priority
&& skb
->sk
&& map
) {
2829 unsigned int prioidx
= skb
->sk
->sk_cgrp_prioidx
;
2831 if (prioidx
< map
->priomap_len
)
2832 skb
->priority
= map
->priomap
[prioidx
];
2836 #define skb_update_prio(skb)
2839 static DEFINE_PER_CPU(int, xmit_recursion
);
2840 #define RECURSION_LIMIT 10
2843 * dev_loopback_xmit - loop back @skb
2844 * @skb: buffer to transmit
2846 int dev_loopback_xmit(struct sk_buff
*skb
)
2848 skb_reset_mac_header(skb
);
2849 __skb_pull(skb
, skb_network_offset(skb
));
2850 skb
->pkt_type
= PACKET_LOOPBACK
;
2851 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
2852 WARN_ON(!skb_dst(skb
));
2857 EXPORT_SYMBOL(dev_loopback_xmit
);
2860 * __dev_queue_xmit - transmit a buffer
2861 * @skb: buffer to transmit
2862 * @accel_priv: private data used for L2 forwarding offload
2864 * Queue a buffer for transmission to a network device. The caller must
2865 * have set the device and priority and built the buffer before calling
2866 * this function. The function can be called from an interrupt.
2868 * A negative errno code is returned on a failure. A success does not
2869 * guarantee the frame will be transmitted as it may be dropped due
2870 * to congestion or traffic shaping.
2872 * -----------------------------------------------------------------------------------
2873 * I notice this method can also return errors from the queue disciplines,
2874 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2877 * Regardless of the return value, the skb is consumed, so it is currently
2878 * difficult to retry a send to this method. (You can bump the ref count
2879 * before sending to hold a reference for retry if you are careful.)
2881 * When calling this method, interrupts MUST be enabled. This is because
2882 * the BH enable code must have IRQs enabled so that it will not deadlock.
2885 static int __dev_queue_xmit(struct sk_buff
*skb
, void *accel_priv
)
2887 struct net_device
*dev
= skb
->dev
;
2888 struct netdev_queue
*txq
;
2892 skb_reset_mac_header(skb
);
2894 if (unlikely(skb_shinfo(skb
)->tx_flags
& SKBTX_SCHED_TSTAMP
))
2895 __skb_tstamp_tx(skb
, NULL
, skb
->sk
, SCM_TSTAMP_SCHED
);
2897 /* Disable soft irqs for various locks below. Also
2898 * stops preemption for RCU.
2902 skb_update_prio(skb
);
2904 txq
= netdev_pick_tx(dev
, skb
, accel_priv
);
2905 q
= rcu_dereference_bh(txq
->qdisc
);
2907 #ifdef CONFIG_NET_CLS_ACT
2908 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2910 trace_net_dev_queue(skb
);
2912 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2916 /* The device has no queue. Common case for software devices:
2917 loopback, all the sorts of tunnels...
2919 Really, it is unlikely that netif_tx_lock protection is necessary
2920 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2922 However, it is possible, that they rely on protection
2925 Check this and shot the lock. It is not prone from deadlocks.
2926 Either shot noqueue qdisc, it is even simpler 8)
2928 if (dev
->flags
& IFF_UP
) {
2929 int cpu
= smp_processor_id(); /* ok because BHs are off */
2931 if (txq
->xmit_lock_owner
!= cpu
) {
2933 if (__this_cpu_read(xmit_recursion
) > RECURSION_LIMIT
)
2934 goto recursion_alert
;
2936 skb
= validate_xmit_skb(skb
, dev
);
2940 HARD_TX_LOCK(dev
, txq
, cpu
);
2942 if (!netif_xmit_stopped(txq
)) {
2943 __this_cpu_inc(xmit_recursion
);
2944 skb
= dev_hard_start_xmit(skb
, dev
, txq
, &rc
);
2945 __this_cpu_dec(xmit_recursion
);
2946 if (dev_xmit_complete(rc
)) {
2947 HARD_TX_UNLOCK(dev
, txq
);
2951 HARD_TX_UNLOCK(dev
, txq
);
2952 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2955 /* Recursion is detected! It is possible,
2959 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2966 rcu_read_unlock_bh();
2968 atomic_long_inc(&dev
->tx_dropped
);
2969 kfree_skb_list(skb
);
2972 rcu_read_unlock_bh();
2976 int dev_queue_xmit(struct sk_buff
*skb
)
2978 return __dev_queue_xmit(skb
, NULL
);
2980 EXPORT_SYMBOL(dev_queue_xmit
);
2982 int dev_queue_xmit_accel(struct sk_buff
*skb
, void *accel_priv
)
2984 return __dev_queue_xmit(skb
, accel_priv
);
2986 EXPORT_SYMBOL(dev_queue_xmit_accel
);
2989 /*=======================================================================
2991 =======================================================================*/
2993 int netdev_max_backlog __read_mostly
= 1000;
2994 EXPORT_SYMBOL(netdev_max_backlog
);
2996 int netdev_tstamp_prequeue __read_mostly
= 1;
2997 int netdev_budget __read_mostly
= 300;
2998 int weight_p __read_mostly
= 64; /* old backlog weight */
3000 /* Called with irq disabled */
3001 static inline void ____napi_schedule(struct softnet_data
*sd
,
3002 struct napi_struct
*napi
)
3004 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
3005 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3010 /* One global table that all flow-based protocols share. */
3011 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
3012 EXPORT_SYMBOL(rps_sock_flow_table
);
3014 struct static_key rps_needed __read_mostly
;
3016 static struct rps_dev_flow
*
3017 set_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
3018 struct rps_dev_flow
*rflow
, u16 next_cpu
)
3020 if (next_cpu
!= RPS_NO_CPU
) {
3021 #ifdef CONFIG_RFS_ACCEL
3022 struct netdev_rx_queue
*rxqueue
;
3023 struct rps_dev_flow_table
*flow_table
;
3024 struct rps_dev_flow
*old_rflow
;
3029 /* Should we steer this flow to a different hardware queue? */
3030 if (!skb_rx_queue_recorded(skb
) || !dev
->rx_cpu_rmap
||
3031 !(dev
->features
& NETIF_F_NTUPLE
))
3033 rxq_index
= cpu_rmap_lookup_index(dev
->rx_cpu_rmap
, next_cpu
);
3034 if (rxq_index
== skb_get_rx_queue(skb
))
3037 rxqueue
= dev
->_rx
+ rxq_index
;
3038 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3041 flow_id
= skb_get_hash(skb
) & flow_table
->mask
;
3042 rc
= dev
->netdev_ops
->ndo_rx_flow_steer(dev
, skb
,
3043 rxq_index
, flow_id
);
3047 rflow
= &flow_table
->flows
[flow_id
];
3049 if (old_rflow
->filter
== rflow
->filter
)
3050 old_rflow
->filter
= RPS_NO_FILTER
;
3054 per_cpu(softnet_data
, next_cpu
).input_queue_head
;
3057 rflow
->cpu
= next_cpu
;
3062 * get_rps_cpu is called from netif_receive_skb and returns the target
3063 * CPU from the RPS map of the receiving queue for a given skb.
3064 * rcu_read_lock must be held on entry.
3066 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
3067 struct rps_dev_flow
**rflowp
)
3069 struct netdev_rx_queue
*rxqueue
;
3070 struct rps_map
*map
;
3071 struct rps_dev_flow_table
*flow_table
;
3072 struct rps_sock_flow_table
*sock_flow_table
;
3077 if (skb_rx_queue_recorded(skb
)) {
3078 u16 index
= skb_get_rx_queue(skb
);
3079 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
3080 WARN_ONCE(dev
->real_num_rx_queues
> 1,
3081 "%s received packet on queue %u, but number "
3082 "of RX queues is %u\n",
3083 dev
->name
, index
, dev
->real_num_rx_queues
);
3086 rxqueue
= dev
->_rx
+ index
;
3090 map
= rcu_dereference(rxqueue
->rps_map
);
3092 if (map
->len
== 1 &&
3093 !rcu_access_pointer(rxqueue
->rps_flow_table
)) {
3094 tcpu
= map
->cpus
[0];
3095 if (cpu_online(tcpu
))
3099 } else if (!rcu_access_pointer(rxqueue
->rps_flow_table
)) {
3103 skb_reset_network_header(skb
);
3104 hash
= skb_get_hash(skb
);
3108 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3109 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
3110 if (flow_table
&& sock_flow_table
) {
3112 struct rps_dev_flow
*rflow
;
3114 rflow
= &flow_table
->flows
[hash
& flow_table
->mask
];
3117 next_cpu
= sock_flow_table
->ents
[hash
& sock_flow_table
->mask
];
3120 * If the desired CPU (where last recvmsg was done) is
3121 * different from current CPU (one in the rx-queue flow
3122 * table entry), switch if one of the following holds:
3123 * - Current CPU is unset (equal to RPS_NO_CPU).
3124 * - Current CPU is offline.
3125 * - The current CPU's queue tail has advanced beyond the
3126 * last packet that was enqueued using this table entry.
3127 * This guarantees that all previous packets for the flow
3128 * have been dequeued, thus preserving in order delivery.
3130 if (unlikely(tcpu
!= next_cpu
) &&
3131 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
3132 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
3133 rflow
->last_qtail
)) >= 0)) {
3135 rflow
= set_rps_cpu(dev
, skb
, rflow
, next_cpu
);
3138 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
3146 tcpu
= map
->cpus
[reciprocal_scale(hash
, map
->len
)];
3147 if (cpu_online(tcpu
)) {
3157 #ifdef CONFIG_RFS_ACCEL
3160 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3161 * @dev: Device on which the filter was set
3162 * @rxq_index: RX queue index
3163 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3164 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3166 * Drivers that implement ndo_rx_flow_steer() should periodically call
3167 * this function for each installed filter and remove the filters for
3168 * which it returns %true.
3170 bool rps_may_expire_flow(struct net_device
*dev
, u16 rxq_index
,
3171 u32 flow_id
, u16 filter_id
)
3173 struct netdev_rx_queue
*rxqueue
= dev
->_rx
+ rxq_index
;
3174 struct rps_dev_flow_table
*flow_table
;
3175 struct rps_dev_flow
*rflow
;
3180 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3181 if (flow_table
&& flow_id
<= flow_table
->mask
) {
3182 rflow
= &flow_table
->flows
[flow_id
];
3183 cpu
= ACCESS_ONCE(rflow
->cpu
);
3184 if (rflow
->filter
== filter_id
&& cpu
!= RPS_NO_CPU
&&
3185 ((int)(per_cpu(softnet_data
, cpu
).input_queue_head
-
3186 rflow
->last_qtail
) <
3187 (int)(10 * flow_table
->mask
)))
3193 EXPORT_SYMBOL(rps_may_expire_flow
);
3195 #endif /* CONFIG_RFS_ACCEL */
3197 /* Called from hardirq (IPI) context */
3198 static void rps_trigger_softirq(void *data
)
3200 struct softnet_data
*sd
= data
;
3202 ____napi_schedule(sd
, &sd
->backlog
);
3206 #endif /* CONFIG_RPS */
3209 * Check if this softnet_data structure is another cpu one
3210 * If yes, queue it to our IPI list and return 1
3213 static int rps_ipi_queued(struct softnet_data
*sd
)
3216 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
3219 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
3220 mysd
->rps_ipi_list
= sd
;
3222 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3225 #endif /* CONFIG_RPS */
3229 #ifdef CONFIG_NET_FLOW_LIMIT
3230 int netdev_flow_limit_table_len __read_mostly
= (1 << 12);
3233 static bool skb_flow_limit(struct sk_buff
*skb
, unsigned int qlen
)
3235 #ifdef CONFIG_NET_FLOW_LIMIT
3236 struct sd_flow_limit
*fl
;
3237 struct softnet_data
*sd
;
3238 unsigned int old_flow
, new_flow
;
3240 if (qlen
< (netdev_max_backlog
>> 1))
3243 sd
= &__get_cpu_var(softnet_data
);
3246 fl
= rcu_dereference(sd
->flow_limit
);
3248 new_flow
= skb_get_hash(skb
) & (fl
->num_buckets
- 1);
3249 old_flow
= fl
->history
[fl
->history_head
];
3250 fl
->history
[fl
->history_head
] = new_flow
;
3253 fl
->history_head
&= FLOW_LIMIT_HISTORY
- 1;
3255 if (likely(fl
->buckets
[old_flow
]))
3256 fl
->buckets
[old_flow
]--;
3258 if (++fl
->buckets
[new_flow
] > (FLOW_LIMIT_HISTORY
>> 1)) {
3270 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3271 * queue (may be a remote CPU queue).
3273 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
3274 unsigned int *qtail
)
3276 struct softnet_data
*sd
;
3277 unsigned long flags
;
3280 sd
= &per_cpu(softnet_data
, cpu
);
3282 local_irq_save(flags
);
3285 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3286 if (qlen
<= netdev_max_backlog
&& !skb_flow_limit(skb
, qlen
)) {
3287 if (skb_queue_len(&sd
->input_pkt_queue
)) {
3289 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
3290 input_queue_tail_incr_save(sd
, qtail
);
3292 local_irq_restore(flags
);
3293 return NET_RX_SUCCESS
;
3296 /* Schedule NAPI for backlog device
3297 * We can use non atomic operation since we own the queue lock
3299 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
3300 if (!rps_ipi_queued(sd
))
3301 ____napi_schedule(sd
, &sd
->backlog
);
3309 local_irq_restore(flags
);
3311 atomic_long_inc(&skb
->dev
->rx_dropped
);
3316 static int netif_rx_internal(struct sk_buff
*skb
)
3320 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3322 trace_netif_rx(skb
);
3324 if (static_key_false(&rps_needed
)) {
3325 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3331 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3333 cpu
= smp_processor_id();
3335 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3343 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
3350 * netif_rx - post buffer to the network code
3351 * @skb: buffer to post
3353 * This function receives a packet from a device driver and queues it for
3354 * the upper (protocol) levels to process. It always succeeds. The buffer
3355 * may be dropped during processing for congestion control or by the
3359 * NET_RX_SUCCESS (no congestion)
3360 * NET_RX_DROP (packet was dropped)
3364 int netif_rx(struct sk_buff
*skb
)
3366 trace_netif_rx_entry(skb
);
3368 return netif_rx_internal(skb
);
3370 EXPORT_SYMBOL(netif_rx
);
3372 int netif_rx_ni(struct sk_buff
*skb
)
3376 trace_netif_rx_ni_entry(skb
);
3379 err
= netif_rx_internal(skb
);
3380 if (local_softirq_pending())
3386 EXPORT_SYMBOL(netif_rx_ni
);
3388 static void net_tx_action(struct softirq_action
*h
)
3390 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3392 if (sd
->completion_queue
) {
3393 struct sk_buff
*clist
;
3395 local_irq_disable();
3396 clist
= sd
->completion_queue
;
3397 sd
->completion_queue
= NULL
;
3401 struct sk_buff
*skb
= clist
;
3402 clist
= clist
->next
;
3404 WARN_ON(atomic_read(&skb
->users
));
3405 if (likely(get_kfree_skb_cb(skb
)->reason
== SKB_REASON_CONSUMED
))
3406 trace_consume_skb(skb
);
3408 trace_kfree_skb(skb
, net_tx_action
);
3413 if (sd
->output_queue
) {
3416 local_irq_disable();
3417 head
= sd
->output_queue
;
3418 sd
->output_queue
= NULL
;
3419 sd
->output_queue_tailp
= &sd
->output_queue
;
3423 struct Qdisc
*q
= head
;
3424 spinlock_t
*root_lock
;
3426 head
= head
->next_sched
;
3428 root_lock
= qdisc_lock(q
);
3429 if (spin_trylock(root_lock
)) {
3430 smp_mb__before_atomic();
3431 clear_bit(__QDISC_STATE_SCHED
,
3434 spin_unlock(root_lock
);
3436 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
3438 __netif_reschedule(q
);
3440 smp_mb__before_atomic();
3441 clear_bit(__QDISC_STATE_SCHED
,
3449 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3450 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3451 /* This hook is defined here for ATM LANE */
3452 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
3453 unsigned char *addr
) __read_mostly
;
3454 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
3457 #ifdef CONFIG_NET_CLS_ACT
3458 /* TODO: Maybe we should just force sch_ingress to be compiled in
3459 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3460 * a compare and 2 stores extra right now if we dont have it on
3461 * but have CONFIG_NET_CLS_ACT
3462 * NOTE: This doesn't stop any functionality; if you dont have
3463 * the ingress scheduler, you just can't add policies on ingress.
3466 static int ing_filter(struct sk_buff
*skb
, struct netdev_queue
*rxq
)
3468 struct net_device
*dev
= skb
->dev
;
3469 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
3470 int result
= TC_ACT_OK
;
3473 if (unlikely(MAX_RED_LOOP
< ttl
++)) {
3474 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3475 skb
->skb_iif
, dev
->ifindex
);
3479 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
3480 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
3482 q
= rcu_dereference(rxq
->qdisc
);
3483 if (q
!= &noop_qdisc
) {
3484 spin_lock(qdisc_lock(q
));
3485 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
3486 result
= qdisc_enqueue_root(skb
, q
);
3487 spin_unlock(qdisc_lock(q
));
3493 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
3494 struct packet_type
**pt_prev
,
3495 int *ret
, struct net_device
*orig_dev
)
3497 struct netdev_queue
*rxq
= rcu_dereference(skb
->dev
->ingress_queue
);
3499 if (!rxq
|| rcu_access_pointer(rxq
->qdisc
) == &noop_qdisc
)
3503 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
3507 switch (ing_filter(skb
, rxq
)) {
3521 * netdev_rx_handler_register - register receive handler
3522 * @dev: device to register a handler for
3523 * @rx_handler: receive handler to register
3524 * @rx_handler_data: data pointer that is used by rx handler
3526 * Register a receive handler for a device. This handler will then be
3527 * called from __netif_receive_skb. A negative errno code is returned
3530 * The caller must hold the rtnl_mutex.
3532 * For a general description of rx_handler, see enum rx_handler_result.
3534 int netdev_rx_handler_register(struct net_device
*dev
,
3535 rx_handler_func_t
*rx_handler
,
3536 void *rx_handler_data
)
3540 if (dev
->rx_handler
)
3543 /* Note: rx_handler_data must be set before rx_handler */
3544 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
3545 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
3549 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
3552 * netdev_rx_handler_unregister - unregister receive handler
3553 * @dev: device to unregister a handler from
3555 * Unregister a receive handler from a device.
3557 * The caller must hold the rtnl_mutex.
3559 void netdev_rx_handler_unregister(struct net_device
*dev
)
3563 RCU_INIT_POINTER(dev
->rx_handler
, NULL
);
3564 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3565 * section has a guarantee to see a non NULL rx_handler_data
3569 RCU_INIT_POINTER(dev
->rx_handler_data
, NULL
);
3571 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
3574 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3575 * the special handling of PFMEMALLOC skbs.
3577 static bool skb_pfmemalloc_protocol(struct sk_buff
*skb
)
3579 switch (skb
->protocol
) {
3580 case htons(ETH_P_ARP
):
3581 case htons(ETH_P_IP
):
3582 case htons(ETH_P_IPV6
):
3583 case htons(ETH_P_8021Q
):
3584 case htons(ETH_P_8021AD
):
3591 static int __netif_receive_skb_core(struct sk_buff
*skb
, bool pfmemalloc
)
3593 struct packet_type
*ptype
, *pt_prev
;
3594 rx_handler_func_t
*rx_handler
;
3595 struct net_device
*orig_dev
;
3596 struct net_device
*null_or_dev
;
3597 bool deliver_exact
= false;
3598 int ret
= NET_RX_DROP
;
3601 net_timestamp_check(!netdev_tstamp_prequeue
, skb
);
3603 trace_netif_receive_skb(skb
);
3605 orig_dev
= skb
->dev
;
3607 skb_reset_network_header(skb
);
3608 if (!skb_transport_header_was_set(skb
))
3609 skb_reset_transport_header(skb
);
3610 skb_reset_mac_len(skb
);
3617 skb
->skb_iif
= skb
->dev
->ifindex
;
3619 __this_cpu_inc(softnet_data
.processed
);
3621 if (skb
->protocol
== cpu_to_be16(ETH_P_8021Q
) ||
3622 skb
->protocol
== cpu_to_be16(ETH_P_8021AD
)) {
3623 skb
= skb_vlan_untag(skb
);
3628 #ifdef CONFIG_NET_CLS_ACT
3629 if (skb
->tc_verd
& TC_NCLS
) {
3630 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
3638 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
3639 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
3641 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3647 #ifdef CONFIG_NET_CLS_ACT
3648 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
3654 if (pfmemalloc
&& !skb_pfmemalloc_protocol(skb
))
3657 if (vlan_tx_tag_present(skb
)) {
3659 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3662 if (vlan_do_receive(&skb
))
3664 else if (unlikely(!skb
))
3668 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
3671 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3674 switch (rx_handler(&skb
)) {
3675 case RX_HANDLER_CONSUMED
:
3676 ret
= NET_RX_SUCCESS
;
3678 case RX_HANDLER_ANOTHER
:
3680 case RX_HANDLER_EXACT
:
3681 deliver_exact
= true;
3682 case RX_HANDLER_PASS
:
3689 if (unlikely(vlan_tx_tag_present(skb
))) {
3690 if (vlan_tx_tag_get_id(skb
))
3691 skb
->pkt_type
= PACKET_OTHERHOST
;
3692 /* Note: we might in the future use prio bits
3693 * and set skb->priority like in vlan_do_receive()
3694 * For the time being, just ignore Priority Code Point
3699 /* deliver only exact match when indicated */
3700 null_or_dev
= deliver_exact
? skb
->dev
: NULL
;
3702 type
= skb
->protocol
;
3703 list_for_each_entry_rcu(ptype
,
3704 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
3705 if (ptype
->type
== type
&&
3706 (ptype
->dev
== null_or_dev
|| ptype
->dev
== skb
->dev
||
3707 ptype
->dev
== orig_dev
)) {
3709 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3715 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
3718 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
3721 atomic_long_inc(&skb
->dev
->rx_dropped
);
3723 /* Jamal, now you will not able to escape explaining
3724 * me how you were going to use this. :-)
3734 static int __netif_receive_skb(struct sk_buff
*skb
)
3738 if (sk_memalloc_socks() && skb_pfmemalloc(skb
)) {
3739 unsigned long pflags
= current
->flags
;
3742 * PFMEMALLOC skbs are special, they should
3743 * - be delivered to SOCK_MEMALLOC sockets only
3744 * - stay away from userspace
3745 * - have bounded memory usage
3747 * Use PF_MEMALLOC as this saves us from propagating the allocation
3748 * context down to all allocation sites.
3750 current
->flags
|= PF_MEMALLOC
;
3751 ret
= __netif_receive_skb_core(skb
, true);
3752 tsk_restore_flags(current
, pflags
, PF_MEMALLOC
);
3754 ret
= __netif_receive_skb_core(skb
, false);
3759 static int netif_receive_skb_internal(struct sk_buff
*skb
)
3761 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3763 if (skb_defer_rx_timestamp(skb
))
3764 return NET_RX_SUCCESS
;
3767 if (static_key_false(&rps_needed
)) {
3768 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3773 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3776 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3783 return __netif_receive_skb(skb
);
3787 * netif_receive_skb - process receive buffer from network
3788 * @skb: buffer to process
3790 * netif_receive_skb() is the main receive data processing function.
3791 * It always succeeds. The buffer may be dropped during processing
3792 * for congestion control or by the protocol layers.
3794 * This function may only be called from softirq context and interrupts
3795 * should be enabled.
3797 * Return values (usually ignored):
3798 * NET_RX_SUCCESS: no congestion
3799 * NET_RX_DROP: packet was dropped
3801 int netif_receive_skb(struct sk_buff
*skb
)
3803 trace_netif_receive_skb_entry(skb
);
3805 return netif_receive_skb_internal(skb
);
3807 EXPORT_SYMBOL(netif_receive_skb
);
3809 /* Network device is going away, flush any packets still pending
3810 * Called with irqs disabled.
3812 static void flush_backlog(void *arg
)
3814 struct net_device
*dev
= arg
;
3815 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3816 struct sk_buff
*skb
, *tmp
;
3819 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3820 if (skb
->dev
== dev
) {
3821 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3823 input_queue_head_incr(sd
);
3828 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3829 if (skb
->dev
== dev
) {
3830 __skb_unlink(skb
, &sd
->process_queue
);
3832 input_queue_head_incr(sd
);
3837 static int napi_gro_complete(struct sk_buff
*skb
)
3839 struct packet_offload
*ptype
;
3840 __be16 type
= skb
->protocol
;
3841 struct list_head
*head
= &offload_base
;
3844 BUILD_BUG_ON(sizeof(struct napi_gro_cb
) > sizeof(skb
->cb
));
3846 if (NAPI_GRO_CB(skb
)->count
== 1) {
3847 skb_shinfo(skb
)->gso_size
= 0;
3852 list_for_each_entry_rcu(ptype
, head
, list
) {
3853 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_complete
)
3856 err
= ptype
->callbacks
.gro_complete(skb
, 0);
3862 WARN_ON(&ptype
->list
== head
);
3864 return NET_RX_SUCCESS
;
3868 return netif_receive_skb_internal(skb
);
3871 /* napi->gro_list contains packets ordered by age.
3872 * youngest packets at the head of it.
3873 * Complete skbs in reverse order to reduce latencies.
3875 void napi_gro_flush(struct napi_struct
*napi
, bool flush_old
)
3877 struct sk_buff
*skb
, *prev
= NULL
;
3879 /* scan list and build reverse chain */
3880 for (skb
= napi
->gro_list
; skb
!= NULL
; skb
= skb
->next
) {
3885 for (skb
= prev
; skb
; skb
= prev
) {
3888 if (flush_old
&& NAPI_GRO_CB(skb
)->age
== jiffies
)
3892 napi_gro_complete(skb
);
3896 napi
->gro_list
= NULL
;
3898 EXPORT_SYMBOL(napi_gro_flush
);
3900 static void gro_list_prepare(struct napi_struct
*napi
, struct sk_buff
*skb
)
3903 unsigned int maclen
= skb
->dev
->hard_header_len
;
3904 u32 hash
= skb_get_hash_raw(skb
);
3906 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3907 unsigned long diffs
;
3909 NAPI_GRO_CB(p
)->flush
= 0;
3911 if (hash
!= skb_get_hash_raw(p
)) {
3912 NAPI_GRO_CB(p
)->same_flow
= 0;
3916 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
3917 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
3918 if (maclen
== ETH_HLEN
)
3919 diffs
|= compare_ether_header(skb_mac_header(p
),
3920 skb_mac_header(skb
));
3922 diffs
= memcmp(skb_mac_header(p
),
3923 skb_mac_header(skb
),
3925 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
3929 static void skb_gro_reset_offset(struct sk_buff
*skb
)
3931 const struct skb_shared_info
*pinfo
= skb_shinfo(skb
);
3932 const skb_frag_t
*frag0
= &pinfo
->frags
[0];
3934 NAPI_GRO_CB(skb
)->data_offset
= 0;
3935 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3936 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3938 if (skb_mac_header(skb
) == skb_tail_pointer(skb
) &&
3940 !PageHighMem(skb_frag_page(frag0
))) {
3941 NAPI_GRO_CB(skb
)->frag0
= skb_frag_address(frag0
);
3942 NAPI_GRO_CB(skb
)->frag0_len
= skb_frag_size(frag0
);
3946 static void gro_pull_from_frag0(struct sk_buff
*skb
, int grow
)
3948 struct skb_shared_info
*pinfo
= skb_shinfo(skb
);
3950 BUG_ON(skb
->end
- skb
->tail
< grow
);
3952 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3954 skb
->data_len
-= grow
;
3957 pinfo
->frags
[0].page_offset
+= grow
;
3958 skb_frag_size_sub(&pinfo
->frags
[0], grow
);
3960 if (unlikely(!skb_frag_size(&pinfo
->frags
[0]))) {
3961 skb_frag_unref(skb
, 0);
3962 memmove(pinfo
->frags
, pinfo
->frags
+ 1,
3963 --pinfo
->nr_frags
* sizeof(pinfo
->frags
[0]));
3967 static enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3969 struct sk_buff
**pp
= NULL
;
3970 struct packet_offload
*ptype
;
3971 __be16 type
= skb
->protocol
;
3972 struct list_head
*head
= &offload_base
;
3974 enum gro_result ret
;
3977 if (!(skb
->dev
->features
& NETIF_F_GRO
))
3980 if (skb_is_gso(skb
) || skb_has_frag_list(skb
) || skb
->csum_bad
)
3983 gro_list_prepare(napi
, skb
);
3986 list_for_each_entry_rcu(ptype
, head
, list
) {
3987 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_receive
)
3990 skb_set_network_header(skb
, skb_gro_offset(skb
));
3991 skb_reset_mac_len(skb
);
3992 NAPI_GRO_CB(skb
)->same_flow
= 0;
3993 NAPI_GRO_CB(skb
)->flush
= 0;
3994 NAPI_GRO_CB(skb
)->free
= 0;
3995 NAPI_GRO_CB(skb
)->udp_mark
= 0;
3997 /* Setup for GRO checksum validation */
3998 switch (skb
->ip_summed
) {
3999 case CHECKSUM_COMPLETE
:
4000 NAPI_GRO_CB(skb
)->csum
= skb
->csum
;
4001 NAPI_GRO_CB(skb
)->csum_valid
= 1;
4002 NAPI_GRO_CB(skb
)->csum_cnt
= 0;
4004 case CHECKSUM_UNNECESSARY
:
4005 NAPI_GRO_CB(skb
)->csum_cnt
= skb
->csum_level
+ 1;
4006 NAPI_GRO_CB(skb
)->csum_valid
= 0;
4009 NAPI_GRO_CB(skb
)->csum_cnt
= 0;
4010 NAPI_GRO_CB(skb
)->csum_valid
= 0;
4013 pp
= ptype
->callbacks
.gro_receive(&napi
->gro_list
, skb
);
4018 if (&ptype
->list
== head
)
4021 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
4022 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
4025 struct sk_buff
*nskb
= *pp
;
4029 napi_gro_complete(nskb
);
4036 if (NAPI_GRO_CB(skb
)->flush
)
4039 if (unlikely(napi
->gro_count
>= MAX_GRO_SKBS
)) {
4040 struct sk_buff
*nskb
= napi
->gro_list
;
4042 /* locate the end of the list to select the 'oldest' flow */
4043 while (nskb
->next
) {
4049 napi_gro_complete(nskb
);
4053 NAPI_GRO_CB(skb
)->count
= 1;
4054 NAPI_GRO_CB(skb
)->age
= jiffies
;
4055 NAPI_GRO_CB(skb
)->last
= skb
;
4056 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
4057 skb
->next
= napi
->gro_list
;
4058 napi
->gro_list
= skb
;
4062 grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
4064 gro_pull_from_frag0(skb
, grow
);
4073 struct packet_offload
*gro_find_receive_by_type(__be16 type
)
4075 struct list_head
*offload_head
= &offload_base
;
4076 struct packet_offload
*ptype
;
4078 list_for_each_entry_rcu(ptype
, offload_head
, list
) {
4079 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_receive
)
4085 EXPORT_SYMBOL(gro_find_receive_by_type
);
4087 struct packet_offload
*gro_find_complete_by_type(__be16 type
)
4089 struct list_head
*offload_head
= &offload_base
;
4090 struct packet_offload
*ptype
;
4092 list_for_each_entry_rcu(ptype
, offload_head
, list
) {
4093 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_complete
)
4099 EXPORT_SYMBOL(gro_find_complete_by_type
);
4101 static gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
4105 if (netif_receive_skb_internal(skb
))
4113 case GRO_MERGED_FREE
:
4114 if (NAPI_GRO_CB(skb
)->free
== NAPI_GRO_FREE_STOLEN_HEAD
)
4115 kmem_cache_free(skbuff_head_cache
, skb
);
4128 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
4130 trace_napi_gro_receive_entry(skb
);
4132 skb_gro_reset_offset(skb
);
4134 return napi_skb_finish(dev_gro_receive(napi
, skb
), skb
);
4136 EXPORT_SYMBOL(napi_gro_receive
);
4138 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
4140 __skb_pull(skb
, skb_headlen(skb
));
4141 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4142 skb_reserve(skb
, NET_SKB_PAD
+ NET_IP_ALIGN
- skb_headroom(skb
));
4144 skb
->dev
= napi
->dev
;
4146 skb
->encapsulation
= 0;
4147 skb_shinfo(skb
)->gso_type
= 0;
4148 skb
->truesize
= SKB_TRUESIZE(skb_end_offset(skb
));
4153 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
4155 struct sk_buff
*skb
= napi
->skb
;
4158 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
4163 EXPORT_SYMBOL(napi_get_frags
);
4165 static gro_result_t
napi_frags_finish(struct napi_struct
*napi
,
4166 struct sk_buff
*skb
,
4172 __skb_push(skb
, ETH_HLEN
);
4173 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
4174 if (ret
== GRO_NORMAL
&& netif_receive_skb_internal(skb
))
4179 case GRO_MERGED_FREE
:
4180 napi_reuse_skb(napi
, skb
);
4190 /* Upper GRO stack assumes network header starts at gro_offset=0
4191 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4192 * We copy ethernet header into skb->data to have a common layout.
4194 static struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
4196 struct sk_buff
*skb
= napi
->skb
;
4197 const struct ethhdr
*eth
;
4198 unsigned int hlen
= sizeof(*eth
);
4202 skb_reset_mac_header(skb
);
4203 skb_gro_reset_offset(skb
);
4205 eth
= skb_gro_header_fast(skb
, 0);
4206 if (unlikely(skb_gro_header_hard(skb
, hlen
))) {
4207 eth
= skb_gro_header_slow(skb
, hlen
, 0);
4208 if (unlikely(!eth
)) {
4209 napi_reuse_skb(napi
, skb
);
4213 gro_pull_from_frag0(skb
, hlen
);
4214 NAPI_GRO_CB(skb
)->frag0
+= hlen
;
4215 NAPI_GRO_CB(skb
)->frag0_len
-= hlen
;
4217 __skb_pull(skb
, hlen
);
4220 * This works because the only protocols we care about don't require
4222 * We'll fix it up properly in napi_frags_finish()
4224 skb
->protocol
= eth
->h_proto
;
4229 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
4231 struct sk_buff
*skb
= napi_frags_skb(napi
);
4236 trace_napi_gro_frags_entry(skb
);
4238 return napi_frags_finish(napi
, skb
, dev_gro_receive(napi
, skb
));
4240 EXPORT_SYMBOL(napi_gro_frags
);
4242 /* Compute the checksum from gro_offset and return the folded value
4243 * after adding in any pseudo checksum.
4245 __sum16
__skb_gro_checksum_complete(struct sk_buff
*skb
)
4250 wsum
= skb_checksum(skb
, skb_gro_offset(skb
), skb_gro_len(skb
), 0);
4252 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4253 sum
= csum_fold(csum_add(NAPI_GRO_CB(skb
)->csum
, wsum
));
4255 if (unlikely(skb
->ip_summed
== CHECKSUM_COMPLETE
) &&
4256 !skb
->csum_complete_sw
)
4257 netdev_rx_csum_fault(skb
->dev
);
4260 NAPI_GRO_CB(skb
)->csum
= wsum
;
4261 NAPI_GRO_CB(skb
)->csum_valid
= 1;
4265 EXPORT_SYMBOL(__skb_gro_checksum_complete
);
4268 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4269 * Note: called with local irq disabled, but exits with local irq enabled.
4271 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
4274 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
4277 sd
->rps_ipi_list
= NULL
;
4281 /* Send pending IPI's to kick RPS processing on remote cpus. */
4283 struct softnet_data
*next
= remsd
->rps_ipi_next
;
4285 if (cpu_online(remsd
->cpu
))
4286 smp_call_function_single_async(remsd
->cpu
,
4295 static int process_backlog(struct napi_struct
*napi
, int quota
)
4298 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
4301 /* Check if we have pending ipi, its better to send them now,
4302 * not waiting net_rx_action() end.
4304 if (sd
->rps_ipi_list
) {
4305 local_irq_disable();
4306 net_rps_action_and_irq_enable(sd
);
4309 napi
->weight
= weight_p
;
4310 local_irq_disable();
4312 struct sk_buff
*skb
;
4314 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
4316 __netif_receive_skb(skb
);
4317 local_irq_disable();
4318 input_queue_head_incr(sd
);
4319 if (++work
>= quota
) {
4326 if (skb_queue_empty(&sd
->input_pkt_queue
)) {
4328 * Inline a custom version of __napi_complete().
4329 * only current cpu owns and manipulates this napi,
4330 * and NAPI_STATE_SCHED is the only possible flag set
4332 * We can use a plain write instead of clear_bit(),
4333 * and we dont need an smp_mb() memory barrier.
4335 list_del(&napi
->poll_list
);
4342 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
4343 &sd
->process_queue
);
4352 * __napi_schedule - schedule for receive
4353 * @n: entry to schedule
4355 * The entry's receive function will be scheduled to run
4357 void __napi_schedule(struct napi_struct
*n
)
4359 unsigned long flags
;
4361 local_irq_save(flags
);
4362 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
4363 local_irq_restore(flags
);
4365 EXPORT_SYMBOL(__napi_schedule
);
4367 void __napi_complete(struct napi_struct
*n
)
4369 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
4370 BUG_ON(n
->gro_list
);
4372 list_del(&n
->poll_list
);
4373 smp_mb__before_atomic();
4374 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
4376 EXPORT_SYMBOL(__napi_complete
);
4378 void napi_complete(struct napi_struct
*n
)
4380 unsigned long flags
;
4383 * don't let napi dequeue from the cpu poll list
4384 * just in case its running on a different cpu
4386 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
4389 napi_gro_flush(n
, false);
4390 local_irq_save(flags
);
4392 local_irq_restore(flags
);
4394 EXPORT_SYMBOL(napi_complete
);
4396 /* must be called under rcu_read_lock(), as we dont take a reference */
4397 struct napi_struct
*napi_by_id(unsigned int napi_id
)
4399 unsigned int hash
= napi_id
% HASH_SIZE(napi_hash
);
4400 struct napi_struct
*napi
;
4402 hlist_for_each_entry_rcu(napi
, &napi_hash
[hash
], napi_hash_node
)
4403 if (napi
->napi_id
== napi_id
)
4408 EXPORT_SYMBOL_GPL(napi_by_id
);
4410 void napi_hash_add(struct napi_struct
*napi
)
4412 if (!test_and_set_bit(NAPI_STATE_HASHED
, &napi
->state
)) {
4414 spin_lock(&napi_hash_lock
);
4416 /* 0 is not a valid id, we also skip an id that is taken
4417 * we expect both events to be extremely rare
4420 while (!napi
->napi_id
) {
4421 napi
->napi_id
= ++napi_gen_id
;
4422 if (napi_by_id(napi
->napi_id
))
4426 hlist_add_head_rcu(&napi
->napi_hash_node
,
4427 &napi_hash
[napi
->napi_id
% HASH_SIZE(napi_hash
)]);
4429 spin_unlock(&napi_hash_lock
);
4432 EXPORT_SYMBOL_GPL(napi_hash_add
);
4434 /* Warning : caller is responsible to make sure rcu grace period
4435 * is respected before freeing memory containing @napi
4437 void napi_hash_del(struct napi_struct
*napi
)
4439 spin_lock(&napi_hash_lock
);
4441 if (test_and_clear_bit(NAPI_STATE_HASHED
, &napi
->state
))
4442 hlist_del_rcu(&napi
->napi_hash_node
);
4444 spin_unlock(&napi_hash_lock
);
4446 EXPORT_SYMBOL_GPL(napi_hash_del
);
4448 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
4449 int (*poll
)(struct napi_struct
*, int), int weight
)
4451 INIT_LIST_HEAD(&napi
->poll_list
);
4452 napi
->gro_count
= 0;
4453 napi
->gro_list
= NULL
;
4456 if (weight
> NAPI_POLL_WEIGHT
)
4457 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4459 napi
->weight
= weight
;
4460 list_add(&napi
->dev_list
, &dev
->napi_list
);
4462 #ifdef CONFIG_NETPOLL
4463 spin_lock_init(&napi
->poll_lock
);
4464 napi
->poll_owner
= -1;
4466 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
4468 EXPORT_SYMBOL(netif_napi_add
);
4470 void netif_napi_del(struct napi_struct
*napi
)
4472 list_del_init(&napi
->dev_list
);
4473 napi_free_frags(napi
);
4475 kfree_skb_list(napi
->gro_list
);
4476 napi
->gro_list
= NULL
;
4477 napi
->gro_count
= 0;
4479 EXPORT_SYMBOL(netif_napi_del
);
4481 static void net_rx_action(struct softirq_action
*h
)
4483 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
4484 unsigned long time_limit
= jiffies
+ 2;
4485 int budget
= netdev_budget
;
4488 local_irq_disable();
4490 while (!list_empty(&sd
->poll_list
)) {
4491 struct napi_struct
*n
;
4494 /* If softirq window is exhuasted then punt.
4495 * Allow this to run for 2 jiffies since which will allow
4496 * an average latency of 1.5/HZ.
4498 if (unlikely(budget
<= 0 || time_after_eq(jiffies
, time_limit
)))
4503 /* Even though interrupts have been re-enabled, this
4504 * access is safe because interrupts can only add new
4505 * entries to the tail of this list, and only ->poll()
4506 * calls can remove this head entry from the list.
4508 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
4510 have
= netpoll_poll_lock(n
);
4514 /* This NAPI_STATE_SCHED test is for avoiding a race
4515 * with netpoll's poll_napi(). Only the entity which
4516 * obtains the lock and sees NAPI_STATE_SCHED set will
4517 * actually make the ->poll() call. Therefore we avoid
4518 * accidentally calling ->poll() when NAPI is not scheduled.
4521 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
4522 work
= n
->poll(n
, weight
);
4526 WARN_ON_ONCE(work
> weight
);
4530 local_irq_disable();
4532 /* Drivers must not modify the NAPI state if they
4533 * consume the entire weight. In such cases this code
4534 * still "owns" the NAPI instance and therefore can
4535 * move the instance around on the list at-will.
4537 if (unlikely(work
== weight
)) {
4538 if (unlikely(napi_disable_pending(n
))) {
4541 local_irq_disable();
4544 /* flush too old packets
4545 * If HZ < 1000, flush all packets.
4548 napi_gro_flush(n
, HZ
>= 1000);
4549 local_irq_disable();
4551 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
4555 netpoll_poll_unlock(have
);
4558 net_rps_action_and_irq_enable(sd
);
4560 #ifdef CONFIG_NET_DMA
4562 * There may not be any more sk_buffs coming right now, so push
4563 * any pending DMA copies to hardware
4565 dma_issue_pending_all();
4572 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
4576 struct netdev_adjacent
{
4577 struct net_device
*dev
;
4579 /* upper master flag, there can only be one master device per list */
4582 /* counter for the number of times this device was added to us */
4585 /* private field for the users */
4588 struct list_head list
;
4589 struct rcu_head rcu
;
4592 static struct netdev_adjacent
*__netdev_find_adj(struct net_device
*dev
,
4593 struct net_device
*adj_dev
,
4594 struct list_head
*adj_list
)
4596 struct netdev_adjacent
*adj
;
4598 list_for_each_entry(adj
, adj_list
, list
) {
4599 if (adj
->dev
== adj_dev
)
4606 * netdev_has_upper_dev - Check if device is linked to an upper device
4608 * @upper_dev: upper device to check
4610 * Find out if a device is linked to specified upper device and return true
4611 * in case it is. Note that this checks only immediate upper device,
4612 * not through a complete stack of devices. The caller must hold the RTNL lock.
4614 bool netdev_has_upper_dev(struct net_device
*dev
,
4615 struct net_device
*upper_dev
)
4619 return __netdev_find_adj(dev
, upper_dev
, &dev
->all_adj_list
.upper
);
4621 EXPORT_SYMBOL(netdev_has_upper_dev
);
4624 * netdev_has_any_upper_dev - Check if device is linked to some device
4627 * Find out if a device is linked to an upper device and return true in case
4628 * it is. The caller must hold the RTNL lock.
4630 static bool netdev_has_any_upper_dev(struct net_device
*dev
)
4634 return !list_empty(&dev
->all_adj_list
.upper
);
4638 * netdev_master_upper_dev_get - Get master upper device
4641 * Find a master upper device and return pointer to it or NULL in case
4642 * it's not there. The caller must hold the RTNL lock.
4644 struct net_device
*netdev_master_upper_dev_get(struct net_device
*dev
)
4646 struct netdev_adjacent
*upper
;
4650 if (list_empty(&dev
->adj_list
.upper
))
4653 upper
= list_first_entry(&dev
->adj_list
.upper
,
4654 struct netdev_adjacent
, list
);
4655 if (likely(upper
->master
))
4659 EXPORT_SYMBOL(netdev_master_upper_dev_get
);
4661 void *netdev_adjacent_get_private(struct list_head
*adj_list
)
4663 struct netdev_adjacent
*adj
;
4665 adj
= list_entry(adj_list
, struct netdev_adjacent
, list
);
4667 return adj
->private;
4669 EXPORT_SYMBOL(netdev_adjacent_get_private
);
4672 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
4674 * @iter: list_head ** of the current position
4676 * Gets the next device from the dev's upper list, starting from iter
4677 * position. The caller must hold RCU read lock.
4679 struct net_device
*netdev_upper_get_next_dev_rcu(struct net_device
*dev
,
4680 struct list_head
**iter
)
4682 struct netdev_adjacent
*upper
;
4684 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4686 upper
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
4688 if (&upper
->list
== &dev
->adj_list
.upper
)
4691 *iter
= &upper
->list
;
4695 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu
);
4698 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4700 * @iter: list_head ** of the current position
4702 * Gets the next device from the dev's upper list, starting from iter
4703 * position. The caller must hold RCU read lock.
4705 struct net_device
*netdev_all_upper_get_next_dev_rcu(struct net_device
*dev
,
4706 struct list_head
**iter
)
4708 struct netdev_adjacent
*upper
;
4710 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4712 upper
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
4714 if (&upper
->list
== &dev
->all_adj_list
.upper
)
4717 *iter
= &upper
->list
;
4721 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu
);
4724 * netdev_lower_get_next_private - Get the next ->private from the
4725 * lower neighbour list
4727 * @iter: list_head ** of the current position
4729 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4730 * list, starting from iter position. The caller must hold either hold the
4731 * RTNL lock or its own locking that guarantees that the neighbour lower
4732 * list will remain unchainged.
4734 void *netdev_lower_get_next_private(struct net_device
*dev
,
4735 struct list_head
**iter
)
4737 struct netdev_adjacent
*lower
;
4739 lower
= list_entry(*iter
, struct netdev_adjacent
, list
);
4741 if (&lower
->list
== &dev
->adj_list
.lower
)
4744 *iter
= lower
->list
.next
;
4746 return lower
->private;
4748 EXPORT_SYMBOL(netdev_lower_get_next_private
);
4751 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4752 * lower neighbour list, RCU
4755 * @iter: list_head ** of the current position
4757 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4758 * list, starting from iter position. The caller must hold RCU read lock.
4760 void *netdev_lower_get_next_private_rcu(struct net_device
*dev
,
4761 struct list_head
**iter
)
4763 struct netdev_adjacent
*lower
;
4765 WARN_ON_ONCE(!rcu_read_lock_held());
4767 lower
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
4769 if (&lower
->list
== &dev
->adj_list
.lower
)
4772 *iter
= &lower
->list
;
4774 return lower
->private;
4776 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu
);
4779 * netdev_lower_get_next - Get the next device from the lower neighbour
4782 * @iter: list_head ** of the current position
4784 * Gets the next netdev_adjacent from the dev's lower neighbour
4785 * list, starting from iter position. The caller must hold RTNL lock or
4786 * its own locking that guarantees that the neighbour lower
4787 * list will remain unchainged.
4789 void *netdev_lower_get_next(struct net_device
*dev
, struct list_head
**iter
)
4791 struct netdev_adjacent
*lower
;
4793 lower
= list_entry((*iter
)->next
, struct netdev_adjacent
, list
);
4795 if (&lower
->list
== &dev
->adj_list
.lower
)
4798 *iter
= &lower
->list
;
4802 EXPORT_SYMBOL(netdev_lower_get_next
);
4805 * netdev_lower_get_first_private_rcu - Get the first ->private from the
4806 * lower neighbour list, RCU
4810 * Gets the first netdev_adjacent->private from the dev's lower neighbour
4811 * list. The caller must hold RCU read lock.
4813 void *netdev_lower_get_first_private_rcu(struct net_device
*dev
)
4815 struct netdev_adjacent
*lower
;
4817 lower
= list_first_or_null_rcu(&dev
->adj_list
.lower
,
4818 struct netdev_adjacent
, list
);
4820 return lower
->private;
4823 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu
);
4826 * netdev_master_upper_dev_get_rcu - Get master upper device
4829 * Find a master upper device and return pointer to it or NULL in case
4830 * it's not there. The caller must hold the RCU read lock.
4832 struct net_device
*netdev_master_upper_dev_get_rcu(struct net_device
*dev
)
4834 struct netdev_adjacent
*upper
;
4836 upper
= list_first_or_null_rcu(&dev
->adj_list
.upper
,
4837 struct netdev_adjacent
, list
);
4838 if (upper
&& likely(upper
->master
))
4842 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu
);
4844 static int netdev_adjacent_sysfs_add(struct net_device
*dev
,
4845 struct net_device
*adj_dev
,
4846 struct list_head
*dev_list
)
4848 char linkname
[IFNAMSIZ
+7];
4849 sprintf(linkname
, dev_list
== &dev
->adj_list
.upper
?
4850 "upper_%s" : "lower_%s", adj_dev
->name
);
4851 return sysfs_create_link(&(dev
->dev
.kobj
), &(adj_dev
->dev
.kobj
),
4854 static void netdev_adjacent_sysfs_del(struct net_device
*dev
,
4856 struct list_head
*dev_list
)
4858 char linkname
[IFNAMSIZ
+7];
4859 sprintf(linkname
, dev_list
== &dev
->adj_list
.upper
?
4860 "upper_%s" : "lower_%s", name
);
4861 sysfs_remove_link(&(dev
->dev
.kobj
), linkname
);
4864 #define netdev_adjacent_is_neigh_list(dev, dev_list) \
4865 (dev_list == &dev->adj_list.upper || \
4866 dev_list == &dev->adj_list.lower)
4868 static int __netdev_adjacent_dev_insert(struct net_device
*dev
,
4869 struct net_device
*adj_dev
,
4870 struct list_head
*dev_list
,
4871 void *private, bool master
)
4873 struct netdev_adjacent
*adj
;
4876 adj
= __netdev_find_adj(dev
, adj_dev
, dev_list
);
4883 adj
= kmalloc(sizeof(*adj
), GFP_KERNEL
);
4888 adj
->master
= master
;
4890 adj
->private = private;
4893 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
4894 adj_dev
->name
, dev
->name
, adj_dev
->name
);
4896 if (netdev_adjacent_is_neigh_list(dev
, dev_list
)) {
4897 ret
= netdev_adjacent_sysfs_add(dev
, adj_dev
, dev_list
);
4902 /* Ensure that master link is always the first item in list. */
4904 ret
= sysfs_create_link(&(dev
->dev
.kobj
),
4905 &(adj_dev
->dev
.kobj
), "master");
4907 goto remove_symlinks
;
4909 list_add_rcu(&adj
->list
, dev_list
);
4911 list_add_tail_rcu(&adj
->list
, dev_list
);
4917 if (netdev_adjacent_is_neigh_list(dev
, dev_list
))
4918 netdev_adjacent_sysfs_del(dev
, adj_dev
->name
, dev_list
);
4926 static void __netdev_adjacent_dev_remove(struct net_device
*dev
,
4927 struct net_device
*adj_dev
,
4928 struct list_head
*dev_list
)
4930 struct netdev_adjacent
*adj
;
4932 adj
= __netdev_find_adj(dev
, adj_dev
, dev_list
);
4935 pr_err("tried to remove device %s from %s\n",
4936 dev
->name
, adj_dev
->name
);
4940 if (adj
->ref_nr
> 1) {
4941 pr_debug("%s to %s ref_nr-- = %d\n", dev
->name
, adj_dev
->name
,
4948 sysfs_remove_link(&(dev
->dev
.kobj
), "master");
4950 if (netdev_adjacent_is_neigh_list(dev
, dev_list
) &&
4951 net_eq(dev_net(dev
),dev_net(adj_dev
)))
4952 netdev_adjacent_sysfs_del(dev
, adj_dev
->name
, dev_list
);
4954 list_del_rcu(&adj
->list
);
4955 pr_debug("dev_put for %s, because link removed from %s to %s\n",
4956 adj_dev
->name
, dev
->name
, adj_dev
->name
);
4958 kfree_rcu(adj
, rcu
);
4961 static int __netdev_adjacent_dev_link_lists(struct net_device
*dev
,
4962 struct net_device
*upper_dev
,
4963 struct list_head
*up_list
,
4964 struct list_head
*down_list
,
4965 void *private, bool master
)
4969 ret
= __netdev_adjacent_dev_insert(dev
, upper_dev
, up_list
, private,
4974 ret
= __netdev_adjacent_dev_insert(upper_dev
, dev
, down_list
, private,
4977 __netdev_adjacent_dev_remove(dev
, upper_dev
, up_list
);
4984 static int __netdev_adjacent_dev_link(struct net_device
*dev
,
4985 struct net_device
*upper_dev
)
4987 return __netdev_adjacent_dev_link_lists(dev
, upper_dev
,
4988 &dev
->all_adj_list
.upper
,
4989 &upper_dev
->all_adj_list
.lower
,
4993 static void __netdev_adjacent_dev_unlink_lists(struct net_device
*dev
,
4994 struct net_device
*upper_dev
,
4995 struct list_head
*up_list
,
4996 struct list_head
*down_list
)
4998 __netdev_adjacent_dev_remove(dev
, upper_dev
, up_list
);
4999 __netdev_adjacent_dev_remove(upper_dev
, dev
, down_list
);
5002 static void __netdev_adjacent_dev_unlink(struct net_device
*dev
,
5003 struct net_device
*upper_dev
)
5005 __netdev_adjacent_dev_unlink_lists(dev
, upper_dev
,
5006 &dev
->all_adj_list
.upper
,
5007 &upper_dev
->all_adj_list
.lower
);
5010 static int __netdev_adjacent_dev_link_neighbour(struct net_device
*dev
,
5011 struct net_device
*upper_dev
,
5012 void *private, bool master
)
5014 int ret
= __netdev_adjacent_dev_link(dev
, upper_dev
);
5019 ret
= __netdev_adjacent_dev_link_lists(dev
, upper_dev
,
5020 &dev
->adj_list
.upper
,
5021 &upper_dev
->adj_list
.lower
,
5024 __netdev_adjacent_dev_unlink(dev
, upper_dev
);
5031 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device
*dev
,
5032 struct net_device
*upper_dev
)
5034 __netdev_adjacent_dev_unlink(dev
, upper_dev
);
5035 __netdev_adjacent_dev_unlink_lists(dev
, upper_dev
,
5036 &dev
->adj_list
.upper
,
5037 &upper_dev
->adj_list
.lower
);
5040 static int __netdev_upper_dev_link(struct net_device
*dev
,
5041 struct net_device
*upper_dev
, bool master
,
5044 struct netdev_adjacent
*i
, *j
, *to_i
, *to_j
;
5049 if (dev
== upper_dev
)
5052 /* To prevent loops, check if dev is not upper device to upper_dev. */
5053 if (__netdev_find_adj(upper_dev
, dev
, &upper_dev
->all_adj_list
.upper
))
5056 if (__netdev_find_adj(dev
, upper_dev
, &dev
->all_adj_list
.upper
))
5059 if (master
&& netdev_master_upper_dev_get(dev
))
5062 ret
= __netdev_adjacent_dev_link_neighbour(dev
, upper_dev
, private,
5067 /* Now that we linked these devs, make all the upper_dev's
5068 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5069 * versa, and don't forget the devices itself. All of these
5070 * links are non-neighbours.
5072 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5073 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
) {
5074 pr_debug("Interlinking %s with %s, non-neighbour\n",
5075 i
->dev
->name
, j
->dev
->name
);
5076 ret
= __netdev_adjacent_dev_link(i
->dev
, j
->dev
);
5082 /* add dev to every upper_dev's upper device */
5083 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
) {
5084 pr_debug("linking %s's upper device %s with %s\n",
5085 upper_dev
->name
, i
->dev
->name
, dev
->name
);
5086 ret
= __netdev_adjacent_dev_link(dev
, i
->dev
);
5088 goto rollback_upper_mesh
;
5091 /* add upper_dev to every dev's lower device */
5092 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5093 pr_debug("linking %s's lower device %s with %s\n", dev
->name
,
5094 i
->dev
->name
, upper_dev
->name
);
5095 ret
= __netdev_adjacent_dev_link(i
->dev
, upper_dev
);
5097 goto rollback_lower_mesh
;
5100 call_netdevice_notifiers(NETDEV_CHANGEUPPER
, dev
);
5103 rollback_lower_mesh
:
5105 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5108 __netdev_adjacent_dev_unlink(i
->dev
, upper_dev
);
5113 rollback_upper_mesh
:
5115 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
) {
5118 __netdev_adjacent_dev_unlink(dev
, i
->dev
);
5126 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5127 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
) {
5128 if (i
== to_i
&& j
== to_j
)
5130 __netdev_adjacent_dev_unlink(i
->dev
, j
->dev
);
5136 __netdev_adjacent_dev_unlink_neighbour(dev
, upper_dev
);
5142 * netdev_upper_dev_link - Add a link to the upper device
5144 * @upper_dev: new upper device
5146 * Adds a link to device which is upper to this one. The caller must hold
5147 * the RTNL lock. On a failure a negative errno code is returned.
5148 * On success the reference counts are adjusted and the function
5151 int netdev_upper_dev_link(struct net_device
*dev
,
5152 struct net_device
*upper_dev
)
5154 return __netdev_upper_dev_link(dev
, upper_dev
, false, NULL
);
5156 EXPORT_SYMBOL(netdev_upper_dev_link
);
5159 * netdev_master_upper_dev_link - Add a master link to the upper device
5161 * @upper_dev: new upper device
5163 * Adds a link to device which is upper to this one. In this case, only
5164 * one master upper device can be linked, although other non-master devices
5165 * might be linked as well. The caller must hold the RTNL lock.
5166 * On a failure a negative errno code is returned. On success the reference
5167 * counts are adjusted and the function returns zero.
5169 int netdev_master_upper_dev_link(struct net_device
*dev
,
5170 struct net_device
*upper_dev
)
5172 return __netdev_upper_dev_link(dev
, upper_dev
, true, NULL
);
5174 EXPORT_SYMBOL(netdev_master_upper_dev_link
);
5176 int netdev_master_upper_dev_link_private(struct net_device
*dev
,
5177 struct net_device
*upper_dev
,
5180 return __netdev_upper_dev_link(dev
, upper_dev
, true, private);
5182 EXPORT_SYMBOL(netdev_master_upper_dev_link_private
);
5185 * netdev_upper_dev_unlink - Removes a link to upper device
5187 * @upper_dev: new upper device
5189 * Removes a link to device which is upper to this one. The caller must hold
5192 void netdev_upper_dev_unlink(struct net_device
*dev
,
5193 struct net_device
*upper_dev
)
5195 struct netdev_adjacent
*i
, *j
;
5198 __netdev_adjacent_dev_unlink_neighbour(dev
, upper_dev
);
5200 /* Here is the tricky part. We must remove all dev's lower
5201 * devices from all upper_dev's upper devices and vice
5202 * versa, to maintain the graph relationship.
5204 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
)
5205 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
)
5206 __netdev_adjacent_dev_unlink(i
->dev
, j
->dev
);
5208 /* remove also the devices itself from lower/upper device
5211 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
)
5212 __netdev_adjacent_dev_unlink(i
->dev
, upper_dev
);
5214 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
)
5215 __netdev_adjacent_dev_unlink(dev
, i
->dev
);
5217 call_netdevice_notifiers(NETDEV_CHANGEUPPER
, dev
);
5219 EXPORT_SYMBOL(netdev_upper_dev_unlink
);
5221 void netdev_adjacent_add_links(struct net_device
*dev
)
5223 struct netdev_adjacent
*iter
;
5225 struct net
*net
= dev_net(dev
);
5227 list_for_each_entry(iter
, &dev
->adj_list
.upper
, list
) {
5228 if (!net_eq(net
,dev_net(iter
->dev
)))
5230 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
5231 &iter
->dev
->adj_list
.lower
);
5232 netdev_adjacent_sysfs_add(dev
, iter
->dev
,
5233 &dev
->adj_list
.upper
);
5236 list_for_each_entry(iter
, &dev
->adj_list
.lower
, list
) {
5237 if (!net_eq(net
,dev_net(iter
->dev
)))
5239 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
5240 &iter
->dev
->adj_list
.upper
);
5241 netdev_adjacent_sysfs_add(dev
, iter
->dev
,
5242 &dev
->adj_list
.lower
);
5246 void netdev_adjacent_del_links(struct net_device
*dev
)
5248 struct netdev_adjacent
*iter
;
5250 struct net
*net
= dev_net(dev
);
5252 list_for_each_entry(iter
, &dev
->adj_list
.upper
, list
) {
5253 if (!net_eq(net
,dev_net(iter
->dev
)))
5255 netdev_adjacent_sysfs_del(iter
->dev
, dev
->name
,
5256 &iter
->dev
->adj_list
.lower
);
5257 netdev_adjacent_sysfs_del(dev
, iter
->dev
->name
,
5258 &dev
->adj_list
.upper
);
5261 list_for_each_entry(iter
, &dev
->adj_list
.lower
, list
) {
5262 if (!net_eq(net
,dev_net(iter
->dev
)))
5264 netdev_adjacent_sysfs_del(iter
->dev
, dev
->name
,
5265 &iter
->dev
->adj_list
.upper
);
5266 netdev_adjacent_sysfs_del(dev
, iter
->dev
->name
,
5267 &dev
->adj_list
.lower
);
5271 void netdev_adjacent_rename_links(struct net_device
*dev
, char *oldname
)
5273 struct netdev_adjacent
*iter
;
5275 struct net
*net
= dev_net(dev
);
5277 list_for_each_entry(iter
, &dev
->adj_list
.upper
, list
) {
5278 if (!net_eq(net
,dev_net(iter
->dev
)))
5280 netdev_adjacent_sysfs_del(iter
->dev
, oldname
,
5281 &iter
->dev
->adj_list
.lower
);
5282 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
5283 &iter
->dev
->adj_list
.lower
);
5286 list_for_each_entry(iter
, &dev
->adj_list
.lower
, list
) {
5287 if (!net_eq(net
,dev_net(iter
->dev
)))
5289 netdev_adjacent_sysfs_del(iter
->dev
, oldname
,
5290 &iter
->dev
->adj_list
.upper
);
5291 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
5292 &iter
->dev
->adj_list
.upper
);
5296 void *netdev_lower_dev_get_private(struct net_device
*dev
,
5297 struct net_device
*lower_dev
)
5299 struct netdev_adjacent
*lower
;
5303 lower
= __netdev_find_adj(dev
, lower_dev
, &dev
->adj_list
.lower
);
5307 return lower
->private;
5309 EXPORT_SYMBOL(netdev_lower_dev_get_private
);
5312 int dev_get_nest_level(struct net_device
*dev
,
5313 bool (*type_check
)(struct net_device
*dev
))
5315 struct net_device
*lower
= NULL
;
5316 struct list_head
*iter
;
5322 netdev_for_each_lower_dev(dev
, lower
, iter
) {
5323 nest
= dev_get_nest_level(lower
, type_check
);
5324 if (max_nest
< nest
)
5328 if (type_check(dev
))
5333 EXPORT_SYMBOL(dev_get_nest_level
);
5335 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
5337 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5339 if (ops
->ndo_change_rx_flags
)
5340 ops
->ndo_change_rx_flags(dev
, flags
);
5343 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
, bool notify
)
5345 unsigned int old_flags
= dev
->flags
;
5351 dev
->flags
|= IFF_PROMISC
;
5352 dev
->promiscuity
+= inc
;
5353 if (dev
->promiscuity
== 0) {
5356 * If inc causes overflow, untouch promisc and return error.
5359 dev
->flags
&= ~IFF_PROMISC
;
5361 dev
->promiscuity
-= inc
;
5362 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5367 if (dev
->flags
!= old_flags
) {
5368 pr_info("device %s %s promiscuous mode\n",
5370 dev
->flags
& IFF_PROMISC
? "entered" : "left");
5371 if (audit_enabled
) {
5372 current_uid_gid(&uid
, &gid
);
5373 audit_log(current
->audit_context
, GFP_ATOMIC
,
5374 AUDIT_ANOM_PROMISCUOUS
,
5375 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5376 dev
->name
, (dev
->flags
& IFF_PROMISC
),
5377 (old_flags
& IFF_PROMISC
),
5378 from_kuid(&init_user_ns
, audit_get_loginuid(current
)),
5379 from_kuid(&init_user_ns
, uid
),
5380 from_kgid(&init_user_ns
, gid
),
5381 audit_get_sessionid(current
));
5384 dev_change_rx_flags(dev
, IFF_PROMISC
);
5387 __dev_notify_flags(dev
, old_flags
, IFF_PROMISC
);
5392 * dev_set_promiscuity - update promiscuity count on a device
5396 * Add or remove promiscuity from a device. While the count in the device
5397 * remains above zero the interface remains promiscuous. Once it hits zero
5398 * the device reverts back to normal filtering operation. A negative inc
5399 * value is used to drop promiscuity on the device.
5400 * Return 0 if successful or a negative errno code on error.
5402 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
5404 unsigned int old_flags
= dev
->flags
;
5407 err
= __dev_set_promiscuity(dev
, inc
, true);
5410 if (dev
->flags
!= old_flags
)
5411 dev_set_rx_mode(dev
);
5414 EXPORT_SYMBOL(dev_set_promiscuity
);
5416 static int __dev_set_allmulti(struct net_device
*dev
, int inc
, bool notify
)
5418 unsigned int old_flags
= dev
->flags
, old_gflags
= dev
->gflags
;
5422 dev
->flags
|= IFF_ALLMULTI
;
5423 dev
->allmulti
+= inc
;
5424 if (dev
->allmulti
== 0) {
5427 * If inc causes overflow, untouch allmulti and return error.
5430 dev
->flags
&= ~IFF_ALLMULTI
;
5432 dev
->allmulti
-= inc
;
5433 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5438 if (dev
->flags
^ old_flags
) {
5439 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
5440 dev_set_rx_mode(dev
);
5442 __dev_notify_flags(dev
, old_flags
,
5443 dev
->gflags
^ old_gflags
);
5449 * dev_set_allmulti - update allmulti count on a device
5453 * Add or remove reception of all multicast frames to a device. While the
5454 * count in the device remains above zero the interface remains listening
5455 * to all interfaces. Once it hits zero the device reverts back to normal
5456 * filtering operation. A negative @inc value is used to drop the counter
5457 * when releasing a resource needing all multicasts.
5458 * Return 0 if successful or a negative errno code on error.
5461 int dev_set_allmulti(struct net_device
*dev
, int inc
)
5463 return __dev_set_allmulti(dev
, inc
, true);
5465 EXPORT_SYMBOL(dev_set_allmulti
);
5468 * Upload unicast and multicast address lists to device and
5469 * configure RX filtering. When the device doesn't support unicast
5470 * filtering it is put in promiscuous mode while unicast addresses
5473 void __dev_set_rx_mode(struct net_device
*dev
)
5475 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5477 /* dev_open will call this function so the list will stay sane. */
5478 if (!(dev
->flags
&IFF_UP
))
5481 if (!netif_device_present(dev
))
5484 if (!(dev
->priv_flags
& IFF_UNICAST_FLT
)) {
5485 /* Unicast addresses changes may only happen under the rtnl,
5486 * therefore calling __dev_set_promiscuity here is safe.
5488 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
5489 __dev_set_promiscuity(dev
, 1, false);
5490 dev
->uc_promisc
= true;
5491 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
5492 __dev_set_promiscuity(dev
, -1, false);
5493 dev
->uc_promisc
= false;
5497 if (ops
->ndo_set_rx_mode
)
5498 ops
->ndo_set_rx_mode(dev
);
5501 void dev_set_rx_mode(struct net_device
*dev
)
5503 netif_addr_lock_bh(dev
);
5504 __dev_set_rx_mode(dev
);
5505 netif_addr_unlock_bh(dev
);
5509 * dev_get_flags - get flags reported to userspace
5512 * Get the combination of flag bits exported through APIs to userspace.
5514 unsigned int dev_get_flags(const struct net_device
*dev
)
5518 flags
= (dev
->flags
& ~(IFF_PROMISC
|
5523 (dev
->gflags
& (IFF_PROMISC
|
5526 if (netif_running(dev
)) {
5527 if (netif_oper_up(dev
))
5528 flags
|= IFF_RUNNING
;
5529 if (netif_carrier_ok(dev
))
5530 flags
|= IFF_LOWER_UP
;
5531 if (netif_dormant(dev
))
5532 flags
|= IFF_DORMANT
;
5537 EXPORT_SYMBOL(dev_get_flags
);
5539 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
5541 unsigned int old_flags
= dev
->flags
;
5547 * Set the flags on our device.
5550 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
5551 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
5553 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
5557 * Load in the correct multicast list now the flags have changed.
5560 if ((old_flags
^ flags
) & IFF_MULTICAST
)
5561 dev_change_rx_flags(dev
, IFF_MULTICAST
);
5563 dev_set_rx_mode(dev
);
5566 * Have we downed the interface. We handle IFF_UP ourselves
5567 * according to user attempts to set it, rather than blindly
5572 if ((old_flags
^ flags
) & IFF_UP
)
5573 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
5575 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
5576 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
5577 unsigned int old_flags
= dev
->flags
;
5579 dev
->gflags
^= IFF_PROMISC
;
5581 if (__dev_set_promiscuity(dev
, inc
, false) >= 0)
5582 if (dev
->flags
!= old_flags
)
5583 dev_set_rx_mode(dev
);
5586 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5587 is important. Some (broken) drivers set IFF_PROMISC, when
5588 IFF_ALLMULTI is requested not asking us and not reporting.
5590 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
5591 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
5593 dev
->gflags
^= IFF_ALLMULTI
;
5594 __dev_set_allmulti(dev
, inc
, false);
5600 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
,
5601 unsigned int gchanges
)
5603 unsigned int changes
= dev
->flags
^ old_flags
;
5606 rtmsg_ifinfo(RTM_NEWLINK
, dev
, gchanges
, GFP_ATOMIC
);
5608 if (changes
& IFF_UP
) {
5609 if (dev
->flags
& IFF_UP
)
5610 call_netdevice_notifiers(NETDEV_UP
, dev
);
5612 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
5615 if (dev
->flags
& IFF_UP
&&
5616 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
))) {
5617 struct netdev_notifier_change_info change_info
;
5619 change_info
.flags_changed
= changes
;
5620 call_netdevice_notifiers_info(NETDEV_CHANGE
, dev
,
5626 * dev_change_flags - change device settings
5628 * @flags: device state flags
5630 * Change settings on device based state flags. The flags are
5631 * in the userspace exported format.
5633 int dev_change_flags(struct net_device
*dev
, unsigned int flags
)
5636 unsigned int changes
, old_flags
= dev
->flags
, old_gflags
= dev
->gflags
;
5638 ret
= __dev_change_flags(dev
, flags
);
5642 changes
= (old_flags
^ dev
->flags
) | (old_gflags
^ dev
->gflags
);
5643 __dev_notify_flags(dev
, old_flags
, changes
);
5646 EXPORT_SYMBOL(dev_change_flags
);
5648 static int __dev_set_mtu(struct net_device
*dev
, int new_mtu
)
5650 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5652 if (ops
->ndo_change_mtu
)
5653 return ops
->ndo_change_mtu(dev
, new_mtu
);
5660 * dev_set_mtu - Change maximum transfer unit
5662 * @new_mtu: new transfer unit
5664 * Change the maximum transfer size of the network device.
5666 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
5670 if (new_mtu
== dev
->mtu
)
5673 /* MTU must be positive. */
5677 if (!netif_device_present(dev
))
5680 err
= call_netdevice_notifiers(NETDEV_PRECHANGEMTU
, dev
);
5681 err
= notifier_to_errno(err
);
5685 orig_mtu
= dev
->mtu
;
5686 err
= __dev_set_mtu(dev
, new_mtu
);
5689 err
= call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
5690 err
= notifier_to_errno(err
);
5692 /* setting mtu back and notifying everyone again,
5693 * so that they have a chance to revert changes.
5695 __dev_set_mtu(dev
, orig_mtu
);
5696 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
5701 EXPORT_SYMBOL(dev_set_mtu
);
5704 * dev_set_group - Change group this device belongs to
5706 * @new_group: group this device should belong to
5708 void dev_set_group(struct net_device
*dev
, int new_group
)
5710 dev
->group
= new_group
;
5712 EXPORT_SYMBOL(dev_set_group
);
5715 * dev_set_mac_address - Change Media Access Control Address
5719 * Change the hardware (MAC) address of the device
5721 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
5723 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5726 if (!ops
->ndo_set_mac_address
)
5728 if (sa
->sa_family
!= dev
->type
)
5730 if (!netif_device_present(dev
))
5732 err
= ops
->ndo_set_mac_address(dev
, sa
);
5735 dev
->addr_assign_type
= NET_ADDR_SET
;
5736 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
5737 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
5740 EXPORT_SYMBOL(dev_set_mac_address
);
5743 * dev_change_carrier - Change device carrier
5745 * @new_carrier: new value
5747 * Change device carrier
5749 int dev_change_carrier(struct net_device
*dev
, bool new_carrier
)
5751 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5753 if (!ops
->ndo_change_carrier
)
5755 if (!netif_device_present(dev
))
5757 return ops
->ndo_change_carrier(dev
, new_carrier
);
5759 EXPORT_SYMBOL(dev_change_carrier
);
5762 * dev_get_phys_port_id - Get device physical port ID
5766 * Get device physical port ID
5768 int dev_get_phys_port_id(struct net_device
*dev
,
5769 struct netdev_phys_port_id
*ppid
)
5771 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5773 if (!ops
->ndo_get_phys_port_id
)
5775 return ops
->ndo_get_phys_port_id(dev
, ppid
);
5777 EXPORT_SYMBOL(dev_get_phys_port_id
);
5780 * dev_new_index - allocate an ifindex
5781 * @net: the applicable net namespace
5783 * Returns a suitable unique value for a new device interface
5784 * number. The caller must hold the rtnl semaphore or the
5785 * dev_base_lock to be sure it remains unique.
5787 static int dev_new_index(struct net
*net
)
5789 int ifindex
= net
->ifindex
;
5793 if (!__dev_get_by_index(net
, ifindex
))
5794 return net
->ifindex
= ifindex
;
5798 /* Delayed registration/unregisteration */
5799 static LIST_HEAD(net_todo_list
);
5800 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq
);
5802 static void net_set_todo(struct net_device
*dev
)
5804 list_add_tail(&dev
->todo_list
, &net_todo_list
);
5805 dev_net(dev
)->dev_unreg_count
++;
5808 static void rollback_registered_many(struct list_head
*head
)
5810 struct net_device
*dev
, *tmp
;
5811 LIST_HEAD(close_head
);
5813 BUG_ON(dev_boot_phase
);
5816 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
5817 /* Some devices call without registering
5818 * for initialization unwind. Remove those
5819 * devices and proceed with the remaining.
5821 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5822 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5826 list_del(&dev
->unreg_list
);
5829 dev
->dismantle
= true;
5830 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
5833 /* If device is running, close it first. */
5834 list_for_each_entry(dev
, head
, unreg_list
)
5835 list_add_tail(&dev
->close_list
, &close_head
);
5836 dev_close_many(&close_head
);
5838 list_for_each_entry(dev
, head
, unreg_list
) {
5839 /* And unlink it from device chain. */
5840 unlist_netdevice(dev
);
5842 dev
->reg_state
= NETREG_UNREGISTERING
;
5847 list_for_each_entry(dev
, head
, unreg_list
) {
5848 /* Shutdown queueing discipline. */
5852 /* Notify protocols, that we are about to destroy
5853 this device. They should clean all the things.
5855 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5858 * Flush the unicast and multicast chains
5863 if (dev
->netdev_ops
->ndo_uninit
)
5864 dev
->netdev_ops
->ndo_uninit(dev
);
5866 if (!dev
->rtnl_link_ops
||
5867 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5868 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U, GFP_KERNEL
);
5870 /* Notifier chain MUST detach us all upper devices. */
5871 WARN_ON(netdev_has_any_upper_dev(dev
));
5873 /* Remove entries from kobject tree */
5874 netdev_unregister_kobject(dev
);
5876 /* Remove XPS queueing entries */
5877 netif_reset_xps_queues_gt(dev
, 0);
5883 list_for_each_entry(dev
, head
, unreg_list
)
5887 static void rollback_registered(struct net_device
*dev
)
5891 list_add(&dev
->unreg_list
, &single
);
5892 rollback_registered_many(&single
);
5896 static netdev_features_t
netdev_fix_features(struct net_device
*dev
,
5897 netdev_features_t features
)
5899 /* Fix illegal checksum combinations */
5900 if ((features
& NETIF_F_HW_CSUM
) &&
5901 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5902 netdev_warn(dev
, "mixed HW and IP checksum settings.\n");
5903 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5906 /* TSO requires that SG is present as well. */
5907 if ((features
& NETIF_F_ALL_TSO
) && !(features
& NETIF_F_SG
)) {
5908 netdev_dbg(dev
, "Dropping TSO features since no SG feature.\n");
5909 features
&= ~NETIF_F_ALL_TSO
;
5912 if ((features
& NETIF_F_TSO
) && !(features
& NETIF_F_HW_CSUM
) &&
5913 !(features
& NETIF_F_IP_CSUM
)) {
5914 netdev_dbg(dev
, "Dropping TSO features since no CSUM feature.\n");
5915 features
&= ~NETIF_F_TSO
;
5916 features
&= ~NETIF_F_TSO_ECN
;
5919 if ((features
& NETIF_F_TSO6
) && !(features
& NETIF_F_HW_CSUM
) &&
5920 !(features
& NETIF_F_IPV6_CSUM
)) {
5921 netdev_dbg(dev
, "Dropping TSO6 features since no CSUM feature.\n");
5922 features
&= ~NETIF_F_TSO6
;
5925 /* TSO ECN requires that TSO is present as well. */
5926 if ((features
& NETIF_F_ALL_TSO
) == NETIF_F_TSO_ECN
)
5927 features
&= ~NETIF_F_TSO_ECN
;
5929 /* Software GSO depends on SG. */
5930 if ((features
& NETIF_F_GSO
) && !(features
& NETIF_F_SG
)) {
5931 netdev_dbg(dev
, "Dropping NETIF_F_GSO since no SG feature.\n");
5932 features
&= ~NETIF_F_GSO
;
5935 /* UFO needs SG and checksumming */
5936 if (features
& NETIF_F_UFO
) {
5937 /* maybe split UFO into V4 and V6? */
5938 if (!((features
& NETIF_F_GEN_CSUM
) ||
5939 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))
5940 == (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5942 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5943 features
&= ~NETIF_F_UFO
;
5946 if (!(features
& NETIF_F_SG
)) {
5948 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5949 features
&= ~NETIF_F_UFO
;
5953 #ifdef CONFIG_NET_RX_BUSY_POLL
5954 if (dev
->netdev_ops
->ndo_busy_poll
)
5955 features
|= NETIF_F_BUSY_POLL
;
5958 features
&= ~NETIF_F_BUSY_POLL
;
5963 int __netdev_update_features(struct net_device
*dev
)
5965 netdev_features_t features
;
5970 features
= netdev_get_wanted_features(dev
);
5972 if (dev
->netdev_ops
->ndo_fix_features
)
5973 features
= dev
->netdev_ops
->ndo_fix_features(dev
, features
);
5975 /* driver might be less strict about feature dependencies */
5976 features
= netdev_fix_features(dev
, features
);
5978 if (dev
->features
== features
)
5981 netdev_dbg(dev
, "Features changed: %pNF -> %pNF\n",
5982 &dev
->features
, &features
);
5984 if (dev
->netdev_ops
->ndo_set_features
)
5985 err
= dev
->netdev_ops
->ndo_set_features(dev
, features
);
5987 if (unlikely(err
< 0)) {
5989 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5990 err
, &features
, &dev
->features
);
5995 dev
->features
= features
;
6001 * netdev_update_features - recalculate device features
6002 * @dev: the device to check
6004 * Recalculate dev->features set and send notifications if it
6005 * has changed. Should be called after driver or hardware dependent
6006 * conditions might have changed that influence the features.
6008 void netdev_update_features(struct net_device
*dev
)
6010 if (__netdev_update_features(dev
))
6011 netdev_features_change(dev
);
6013 EXPORT_SYMBOL(netdev_update_features
);
6016 * netdev_change_features - recalculate device features
6017 * @dev: the device to check
6019 * Recalculate dev->features set and send notifications even
6020 * if they have not changed. Should be called instead of
6021 * netdev_update_features() if also dev->vlan_features might
6022 * have changed to allow the changes to be propagated to stacked
6025 void netdev_change_features(struct net_device
*dev
)
6027 __netdev_update_features(dev
);
6028 netdev_features_change(dev
);
6030 EXPORT_SYMBOL(netdev_change_features
);
6033 * netif_stacked_transfer_operstate - transfer operstate
6034 * @rootdev: the root or lower level device to transfer state from
6035 * @dev: the device to transfer operstate to
6037 * Transfer operational state from root to device. This is normally
6038 * called when a stacking relationship exists between the root
6039 * device and the device(a leaf device).
6041 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
6042 struct net_device
*dev
)
6044 if (rootdev
->operstate
== IF_OPER_DORMANT
)
6045 netif_dormant_on(dev
);
6047 netif_dormant_off(dev
);
6049 if (netif_carrier_ok(rootdev
)) {
6050 if (!netif_carrier_ok(dev
))
6051 netif_carrier_on(dev
);
6053 if (netif_carrier_ok(dev
))
6054 netif_carrier_off(dev
);
6057 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
6060 static int netif_alloc_rx_queues(struct net_device
*dev
)
6062 unsigned int i
, count
= dev
->num_rx_queues
;
6063 struct netdev_rx_queue
*rx
;
6067 rx
= kcalloc(count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
6073 for (i
= 0; i
< count
; i
++)
6079 static void netdev_init_one_queue(struct net_device
*dev
,
6080 struct netdev_queue
*queue
, void *_unused
)
6082 /* Initialize queue lock */
6083 spin_lock_init(&queue
->_xmit_lock
);
6084 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
6085 queue
->xmit_lock_owner
= -1;
6086 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
6089 dql_init(&queue
->dql
, HZ
);
6093 static void netif_free_tx_queues(struct net_device
*dev
)
6098 static int netif_alloc_netdev_queues(struct net_device
*dev
)
6100 unsigned int count
= dev
->num_tx_queues
;
6101 struct netdev_queue
*tx
;
6102 size_t sz
= count
* sizeof(*tx
);
6104 BUG_ON(count
< 1 || count
> 0xffff);
6106 tx
= kzalloc(sz
, GFP_KERNEL
| __GFP_NOWARN
| __GFP_REPEAT
);
6114 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
6115 spin_lock_init(&dev
->tx_global_lock
);
6121 * register_netdevice - register a network device
6122 * @dev: device to register
6124 * Take a completed network device structure and add it to the kernel
6125 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6126 * chain. 0 is returned on success. A negative errno code is returned
6127 * on a failure to set up the device, or if the name is a duplicate.
6129 * Callers must hold the rtnl semaphore. You may want
6130 * register_netdev() instead of this.
6133 * The locking appears insufficient to guarantee two parallel registers
6134 * will not get the same name.
6137 int register_netdevice(struct net_device
*dev
)
6140 struct net
*net
= dev_net(dev
);
6142 BUG_ON(dev_boot_phase
);
6147 /* When net_device's are persistent, this will be fatal. */
6148 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
6151 spin_lock_init(&dev
->addr_list_lock
);
6152 netdev_set_addr_lockdep_class(dev
);
6156 ret
= dev_get_valid_name(net
, dev
, dev
->name
);
6160 /* Init, if this function is available */
6161 if (dev
->netdev_ops
->ndo_init
) {
6162 ret
= dev
->netdev_ops
->ndo_init(dev
);
6170 if (((dev
->hw_features
| dev
->features
) &
6171 NETIF_F_HW_VLAN_CTAG_FILTER
) &&
6172 (!dev
->netdev_ops
->ndo_vlan_rx_add_vid
||
6173 !dev
->netdev_ops
->ndo_vlan_rx_kill_vid
)) {
6174 netdev_WARN(dev
, "Buggy VLAN acceleration in driver!\n");
6181 dev
->ifindex
= dev_new_index(net
);
6182 else if (__dev_get_by_index(net
, dev
->ifindex
))
6185 if (dev
->iflink
== -1)
6186 dev
->iflink
= dev
->ifindex
;
6188 /* Transfer changeable features to wanted_features and enable
6189 * software offloads (GSO and GRO).
6191 dev
->hw_features
|= NETIF_F_SOFT_FEATURES
;
6192 dev
->features
|= NETIF_F_SOFT_FEATURES
;
6193 dev
->wanted_features
= dev
->features
& dev
->hw_features
;
6195 if (!(dev
->flags
& IFF_LOOPBACK
)) {
6196 dev
->hw_features
|= NETIF_F_NOCACHE_COPY
;
6199 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6201 dev
->vlan_features
|= NETIF_F_HIGHDMA
;
6203 /* Make NETIF_F_SG inheritable to tunnel devices.
6205 dev
->hw_enc_features
|= NETIF_F_SG
;
6207 /* Make NETIF_F_SG inheritable to MPLS.
6209 dev
->mpls_features
|= NETIF_F_SG
;
6211 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
6212 ret
= notifier_to_errno(ret
);
6216 ret
= netdev_register_kobject(dev
);
6219 dev
->reg_state
= NETREG_REGISTERED
;
6221 __netdev_update_features(dev
);
6224 * Default initial state at registry is that the
6225 * device is present.
6228 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
6230 linkwatch_init_dev(dev
);
6232 dev_init_scheduler(dev
);
6234 list_netdevice(dev
);
6235 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
6237 /* If the device has permanent device address, driver should
6238 * set dev_addr and also addr_assign_type should be set to
6239 * NET_ADDR_PERM (default value).
6241 if (dev
->addr_assign_type
== NET_ADDR_PERM
)
6242 memcpy(dev
->perm_addr
, dev
->dev_addr
, dev
->addr_len
);
6244 /* Notify protocols, that a new device appeared. */
6245 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
6246 ret
= notifier_to_errno(ret
);
6248 rollback_registered(dev
);
6249 dev
->reg_state
= NETREG_UNREGISTERED
;
6252 * Prevent userspace races by waiting until the network
6253 * device is fully setup before sending notifications.
6255 if (!dev
->rtnl_link_ops
||
6256 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
6257 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U, GFP_KERNEL
);
6263 if (dev
->netdev_ops
->ndo_uninit
)
6264 dev
->netdev_ops
->ndo_uninit(dev
);
6267 EXPORT_SYMBOL(register_netdevice
);
6270 * init_dummy_netdev - init a dummy network device for NAPI
6271 * @dev: device to init
6273 * This takes a network device structure and initialize the minimum
6274 * amount of fields so it can be used to schedule NAPI polls without
6275 * registering a full blown interface. This is to be used by drivers
6276 * that need to tie several hardware interfaces to a single NAPI
6277 * poll scheduler due to HW limitations.
6279 int init_dummy_netdev(struct net_device
*dev
)
6281 /* Clear everything. Note we don't initialize spinlocks
6282 * are they aren't supposed to be taken by any of the
6283 * NAPI code and this dummy netdev is supposed to be
6284 * only ever used for NAPI polls
6286 memset(dev
, 0, sizeof(struct net_device
));
6288 /* make sure we BUG if trying to hit standard
6289 * register/unregister code path
6291 dev
->reg_state
= NETREG_DUMMY
;
6293 /* NAPI wants this */
6294 INIT_LIST_HEAD(&dev
->napi_list
);
6296 /* a dummy interface is started by default */
6297 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
6298 set_bit(__LINK_STATE_START
, &dev
->state
);
6300 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6301 * because users of this 'device' dont need to change
6307 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
6311 * register_netdev - register a network device
6312 * @dev: device to register
6314 * Take a completed network device structure and add it to the kernel
6315 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6316 * chain. 0 is returned on success. A negative errno code is returned
6317 * on a failure to set up the device, or if the name is a duplicate.
6319 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6320 * and expands the device name if you passed a format string to
6323 int register_netdev(struct net_device
*dev
)
6328 err
= register_netdevice(dev
);
6332 EXPORT_SYMBOL(register_netdev
);
6334 int netdev_refcnt_read(const struct net_device
*dev
)
6338 for_each_possible_cpu(i
)
6339 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
6342 EXPORT_SYMBOL(netdev_refcnt_read
);
6345 * netdev_wait_allrefs - wait until all references are gone.
6346 * @dev: target net_device
6348 * This is called when unregistering network devices.
6350 * Any protocol or device that holds a reference should register
6351 * for netdevice notification, and cleanup and put back the
6352 * reference if they receive an UNREGISTER event.
6353 * We can get stuck here if buggy protocols don't correctly
6356 static void netdev_wait_allrefs(struct net_device
*dev
)
6358 unsigned long rebroadcast_time
, warning_time
;
6361 linkwatch_forget_dev(dev
);
6363 rebroadcast_time
= warning_time
= jiffies
;
6364 refcnt
= netdev_refcnt_read(dev
);
6366 while (refcnt
!= 0) {
6367 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
6370 /* Rebroadcast unregister notification */
6371 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6377 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
6378 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
6380 /* We must not have linkwatch events
6381 * pending on unregister. If this
6382 * happens, we simply run the queue
6383 * unscheduled, resulting in a noop
6386 linkwatch_run_queue();
6391 rebroadcast_time
= jiffies
;
6396 refcnt
= netdev_refcnt_read(dev
);
6398 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
6399 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6401 warning_time
= jiffies
;
6410 * register_netdevice(x1);
6411 * register_netdevice(x2);
6413 * unregister_netdevice(y1);
6414 * unregister_netdevice(y2);
6420 * We are invoked by rtnl_unlock().
6421 * This allows us to deal with problems:
6422 * 1) We can delete sysfs objects which invoke hotplug
6423 * without deadlocking with linkwatch via keventd.
6424 * 2) Since we run with the RTNL semaphore not held, we can sleep
6425 * safely in order to wait for the netdev refcnt to drop to zero.
6427 * We must not return until all unregister events added during
6428 * the interval the lock was held have been completed.
6430 void netdev_run_todo(void)
6432 struct list_head list
;
6434 /* Snapshot list, allow later requests */
6435 list_replace_init(&net_todo_list
, &list
);
6440 /* Wait for rcu callbacks to finish before next phase */
6441 if (!list_empty(&list
))
6444 while (!list_empty(&list
)) {
6445 struct net_device
*dev
6446 = list_first_entry(&list
, struct net_device
, todo_list
);
6447 list_del(&dev
->todo_list
);
6450 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
6453 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
6454 pr_err("network todo '%s' but state %d\n",
6455 dev
->name
, dev
->reg_state
);
6460 dev
->reg_state
= NETREG_UNREGISTERED
;
6462 on_each_cpu(flush_backlog
, dev
, 1);
6464 netdev_wait_allrefs(dev
);
6467 BUG_ON(netdev_refcnt_read(dev
));
6468 WARN_ON(rcu_access_pointer(dev
->ip_ptr
));
6469 WARN_ON(rcu_access_pointer(dev
->ip6_ptr
));
6470 WARN_ON(dev
->dn_ptr
);
6472 if (dev
->destructor
)
6473 dev
->destructor(dev
);
6475 /* Report a network device has been unregistered */
6477 dev_net(dev
)->dev_unreg_count
--;
6479 wake_up(&netdev_unregistering_wq
);
6481 /* Free network device */
6482 kobject_put(&dev
->dev
.kobj
);
6486 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6487 * fields in the same order, with only the type differing.
6489 void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
6490 const struct net_device_stats
*netdev_stats
)
6492 #if BITS_PER_LONG == 64
6493 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
6494 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
6496 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
6497 const unsigned long *src
= (const unsigned long *)netdev_stats
;
6498 u64
*dst
= (u64
*)stats64
;
6500 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
6501 sizeof(*stats64
) / sizeof(u64
));
6502 for (i
= 0; i
< n
; i
++)
6506 EXPORT_SYMBOL(netdev_stats_to_stats64
);
6509 * dev_get_stats - get network device statistics
6510 * @dev: device to get statistics from
6511 * @storage: place to store stats
6513 * Get network statistics from device. Return @storage.
6514 * The device driver may provide its own method by setting
6515 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6516 * otherwise the internal statistics structure is used.
6518 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
6519 struct rtnl_link_stats64
*storage
)
6521 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6523 if (ops
->ndo_get_stats64
) {
6524 memset(storage
, 0, sizeof(*storage
));
6525 ops
->ndo_get_stats64(dev
, storage
);
6526 } else if (ops
->ndo_get_stats
) {
6527 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
6529 netdev_stats_to_stats64(storage
, &dev
->stats
);
6531 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
6532 storage
->tx_dropped
+= atomic_long_read(&dev
->tx_dropped
);
6535 EXPORT_SYMBOL(dev_get_stats
);
6537 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
6539 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
6541 #ifdef CONFIG_NET_CLS_ACT
6544 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
6547 netdev_init_one_queue(dev
, queue
, NULL
);
6548 queue
->qdisc
= &noop_qdisc
;
6549 queue
->qdisc_sleeping
= &noop_qdisc
;
6550 rcu_assign_pointer(dev
->ingress_queue
, queue
);
6555 static const struct ethtool_ops default_ethtool_ops
;
6557 void netdev_set_default_ethtool_ops(struct net_device
*dev
,
6558 const struct ethtool_ops
*ops
)
6560 if (dev
->ethtool_ops
== &default_ethtool_ops
)
6561 dev
->ethtool_ops
= ops
;
6563 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops
);
6565 void netdev_freemem(struct net_device
*dev
)
6567 char *addr
= (char *)dev
- dev
->padded
;
6573 * alloc_netdev_mqs - allocate network device
6574 * @sizeof_priv: size of private data to allocate space for
6575 * @name: device name format string
6576 * @name_assign_type: origin of device name
6577 * @setup: callback to initialize device
6578 * @txqs: the number of TX subqueues to allocate
6579 * @rxqs: the number of RX subqueues to allocate
6581 * Allocates a struct net_device with private data area for driver use
6582 * and performs basic initialization. Also allocates subqueue structs
6583 * for each queue on the device.
6585 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
6586 unsigned char name_assign_type
,
6587 void (*setup
)(struct net_device
*),
6588 unsigned int txqs
, unsigned int rxqs
)
6590 struct net_device
*dev
;
6592 struct net_device
*p
;
6594 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
6597 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6603 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6608 alloc_size
= sizeof(struct net_device
);
6610 /* ensure 32-byte alignment of private area */
6611 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
6612 alloc_size
+= sizeof_priv
;
6614 /* ensure 32-byte alignment of whole construct */
6615 alloc_size
+= NETDEV_ALIGN
- 1;
6617 p
= kzalloc(alloc_size
, GFP_KERNEL
| __GFP_NOWARN
| __GFP_REPEAT
);
6619 p
= vzalloc(alloc_size
);
6623 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
6624 dev
->padded
= (char *)dev
- (char *)p
;
6626 dev
->pcpu_refcnt
= alloc_percpu(int);
6627 if (!dev
->pcpu_refcnt
)
6630 if (dev_addr_init(dev
))
6636 dev_net_set(dev
, &init_net
);
6638 dev
->gso_max_size
= GSO_MAX_SIZE
;
6639 dev
->gso_max_segs
= GSO_MAX_SEGS
;
6641 INIT_LIST_HEAD(&dev
->napi_list
);
6642 INIT_LIST_HEAD(&dev
->unreg_list
);
6643 INIT_LIST_HEAD(&dev
->close_list
);
6644 INIT_LIST_HEAD(&dev
->link_watch_list
);
6645 INIT_LIST_HEAD(&dev
->adj_list
.upper
);
6646 INIT_LIST_HEAD(&dev
->adj_list
.lower
);
6647 INIT_LIST_HEAD(&dev
->all_adj_list
.upper
);
6648 INIT_LIST_HEAD(&dev
->all_adj_list
.lower
);
6649 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
6652 dev
->num_tx_queues
= txqs
;
6653 dev
->real_num_tx_queues
= txqs
;
6654 if (netif_alloc_netdev_queues(dev
))
6658 dev
->num_rx_queues
= rxqs
;
6659 dev
->real_num_rx_queues
= rxqs
;
6660 if (netif_alloc_rx_queues(dev
))
6664 strcpy(dev
->name
, name
);
6665 dev
->name_assign_type
= name_assign_type
;
6666 dev
->group
= INIT_NETDEV_GROUP
;
6667 if (!dev
->ethtool_ops
)
6668 dev
->ethtool_ops
= &default_ethtool_ops
;
6676 free_percpu(dev
->pcpu_refcnt
);
6678 netdev_freemem(dev
);
6681 EXPORT_SYMBOL(alloc_netdev_mqs
);
6684 * free_netdev - free network device
6687 * This function does the last stage of destroying an allocated device
6688 * interface. The reference to the device object is released.
6689 * If this is the last reference then it will be freed.
6691 void free_netdev(struct net_device
*dev
)
6693 struct napi_struct
*p
, *n
;
6695 release_net(dev_net(dev
));
6697 netif_free_tx_queues(dev
);
6702 kfree(rcu_dereference_protected(dev
->ingress_queue
, 1));
6704 /* Flush device addresses */
6705 dev_addr_flush(dev
);
6707 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
6710 free_percpu(dev
->pcpu_refcnt
);
6711 dev
->pcpu_refcnt
= NULL
;
6713 /* Compatibility with error handling in drivers */
6714 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
6715 netdev_freemem(dev
);
6719 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
6720 dev
->reg_state
= NETREG_RELEASED
;
6722 /* will free via device release */
6723 put_device(&dev
->dev
);
6725 EXPORT_SYMBOL(free_netdev
);
6728 * synchronize_net - Synchronize with packet receive processing
6730 * Wait for packets currently being received to be done.
6731 * Does not block later packets from starting.
6733 void synchronize_net(void)
6736 if (rtnl_is_locked())
6737 synchronize_rcu_expedited();
6741 EXPORT_SYMBOL(synchronize_net
);
6744 * unregister_netdevice_queue - remove device from the kernel
6748 * This function shuts down a device interface and removes it
6749 * from the kernel tables.
6750 * If head not NULL, device is queued to be unregistered later.
6752 * Callers must hold the rtnl semaphore. You may want
6753 * unregister_netdev() instead of this.
6756 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
6761 list_move_tail(&dev
->unreg_list
, head
);
6763 rollback_registered(dev
);
6764 /* Finish processing unregister after unlock */
6768 EXPORT_SYMBOL(unregister_netdevice_queue
);
6771 * unregister_netdevice_many - unregister many devices
6772 * @head: list of devices
6774 * Note: As most callers use a stack allocated list_head,
6775 * we force a list_del() to make sure stack wont be corrupted later.
6777 void unregister_netdevice_many(struct list_head
*head
)
6779 struct net_device
*dev
;
6781 if (!list_empty(head
)) {
6782 rollback_registered_many(head
);
6783 list_for_each_entry(dev
, head
, unreg_list
)
6788 EXPORT_SYMBOL(unregister_netdevice_many
);
6791 * unregister_netdev - remove device from the kernel
6794 * This function shuts down a device interface and removes it
6795 * from the kernel tables.
6797 * This is just a wrapper for unregister_netdevice that takes
6798 * the rtnl semaphore. In general you want to use this and not
6799 * unregister_netdevice.
6801 void unregister_netdev(struct net_device
*dev
)
6804 unregister_netdevice(dev
);
6807 EXPORT_SYMBOL(unregister_netdev
);
6810 * dev_change_net_namespace - move device to different nethost namespace
6812 * @net: network namespace
6813 * @pat: If not NULL name pattern to try if the current device name
6814 * is already taken in the destination network namespace.
6816 * This function shuts down a device interface and moves it
6817 * to a new network namespace. On success 0 is returned, on
6818 * a failure a netagive errno code is returned.
6820 * Callers must hold the rtnl semaphore.
6823 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
6829 /* Don't allow namespace local devices to be moved. */
6831 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6834 /* Ensure the device has been registrered */
6835 if (dev
->reg_state
!= NETREG_REGISTERED
)
6838 /* Get out if there is nothing todo */
6840 if (net_eq(dev_net(dev
), net
))
6843 /* Pick the destination device name, and ensure
6844 * we can use it in the destination network namespace.
6847 if (__dev_get_by_name(net
, dev
->name
)) {
6848 /* We get here if we can't use the current device name */
6851 if (dev_get_valid_name(net
, dev
, pat
) < 0)
6856 * And now a mini version of register_netdevice unregister_netdevice.
6859 /* If device is running close it first. */
6862 /* And unlink it from device chain */
6864 unlist_netdevice(dev
);
6868 /* Shutdown queueing discipline. */
6871 /* Notify protocols, that we are about to destroy
6872 this device. They should clean all the things.
6874 Note that dev->reg_state stays at NETREG_REGISTERED.
6875 This is wanted because this way 8021q and macvlan know
6876 the device is just moving and can keep their slaves up.
6878 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6880 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
6881 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U, GFP_KERNEL
);
6884 * Flush the unicast and multicast chains
6889 /* Send a netdev-removed uevent to the old namespace */
6890 kobject_uevent(&dev
->dev
.kobj
, KOBJ_REMOVE
);
6891 netdev_adjacent_del_links(dev
);
6893 /* Actually switch the network namespace */
6894 dev_net_set(dev
, net
);
6896 /* If there is an ifindex conflict assign a new one */
6897 if (__dev_get_by_index(net
, dev
->ifindex
)) {
6898 int iflink
= (dev
->iflink
== dev
->ifindex
);
6899 dev
->ifindex
= dev_new_index(net
);
6901 dev
->iflink
= dev
->ifindex
;
6904 /* Send a netdev-add uevent to the new namespace */
6905 kobject_uevent(&dev
->dev
.kobj
, KOBJ_ADD
);
6906 netdev_adjacent_add_links(dev
);
6908 /* Fixup kobjects */
6909 err
= device_rename(&dev
->dev
, dev
->name
);
6912 /* Add the device back in the hashes */
6913 list_netdevice(dev
);
6915 /* Notify protocols, that a new device appeared. */
6916 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
6919 * Prevent userspace races by waiting until the network
6920 * device is fully setup before sending notifications.
6922 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U, GFP_KERNEL
);
6929 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
6931 static int dev_cpu_callback(struct notifier_block
*nfb
,
6932 unsigned long action
,
6935 struct sk_buff
**list_skb
;
6936 struct sk_buff
*skb
;
6937 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
6938 struct softnet_data
*sd
, *oldsd
;
6940 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
6943 local_irq_disable();
6944 cpu
= smp_processor_id();
6945 sd
= &per_cpu(softnet_data
, cpu
);
6946 oldsd
= &per_cpu(softnet_data
, oldcpu
);
6948 /* Find end of our completion_queue. */
6949 list_skb
= &sd
->completion_queue
;
6951 list_skb
= &(*list_skb
)->next
;
6952 /* Append completion queue from offline CPU. */
6953 *list_skb
= oldsd
->completion_queue
;
6954 oldsd
->completion_queue
= NULL
;
6956 /* Append output queue from offline CPU. */
6957 if (oldsd
->output_queue
) {
6958 *sd
->output_queue_tailp
= oldsd
->output_queue
;
6959 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
6960 oldsd
->output_queue
= NULL
;
6961 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
6963 /* Append NAPI poll list from offline CPU. */
6964 if (!list_empty(&oldsd
->poll_list
)) {
6965 list_splice_init(&oldsd
->poll_list
, &sd
->poll_list
);
6966 raise_softirq_irqoff(NET_RX_SOFTIRQ
);
6969 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
6972 /* Process offline CPU's input_pkt_queue */
6973 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
6974 netif_rx_internal(skb
);
6975 input_queue_head_incr(oldsd
);
6977 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
6978 netif_rx_internal(skb
);
6979 input_queue_head_incr(oldsd
);
6987 * netdev_increment_features - increment feature set by one
6988 * @all: current feature set
6989 * @one: new feature set
6990 * @mask: mask feature set
6992 * Computes a new feature set after adding a device with feature set
6993 * @one to the master device with current feature set @all. Will not
6994 * enable anything that is off in @mask. Returns the new feature set.
6996 netdev_features_t
netdev_increment_features(netdev_features_t all
,
6997 netdev_features_t one
, netdev_features_t mask
)
6999 if (mask
& NETIF_F_GEN_CSUM
)
7000 mask
|= NETIF_F_ALL_CSUM
;
7001 mask
|= NETIF_F_VLAN_CHALLENGED
;
7003 all
|= one
& (NETIF_F_ONE_FOR_ALL
|NETIF_F_ALL_CSUM
) & mask
;
7004 all
&= one
| ~NETIF_F_ALL_FOR_ALL
;
7006 /* If one device supports hw checksumming, set for all. */
7007 if (all
& NETIF_F_GEN_CSUM
)
7008 all
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_GEN_CSUM
);
7012 EXPORT_SYMBOL(netdev_increment_features
);
7014 static struct hlist_head
* __net_init
netdev_create_hash(void)
7017 struct hlist_head
*hash
;
7019 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
7021 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
7022 INIT_HLIST_HEAD(&hash
[i
]);
7027 /* Initialize per network namespace state */
7028 static int __net_init
netdev_init(struct net
*net
)
7030 if (net
!= &init_net
)
7031 INIT_LIST_HEAD(&net
->dev_base_head
);
7033 net
->dev_name_head
= netdev_create_hash();
7034 if (net
->dev_name_head
== NULL
)
7037 net
->dev_index_head
= netdev_create_hash();
7038 if (net
->dev_index_head
== NULL
)
7044 kfree(net
->dev_name_head
);
7050 * netdev_drivername - network driver for the device
7051 * @dev: network device
7053 * Determine network driver for device.
7055 const char *netdev_drivername(const struct net_device
*dev
)
7057 const struct device_driver
*driver
;
7058 const struct device
*parent
;
7059 const char *empty
= "";
7061 parent
= dev
->dev
.parent
;
7065 driver
= parent
->driver
;
7066 if (driver
&& driver
->name
)
7067 return driver
->name
;
7071 static int __netdev_printk(const char *level
, const struct net_device
*dev
,
7072 struct va_format
*vaf
)
7076 if (dev
&& dev
->dev
.parent
) {
7077 r
= dev_printk_emit(level
[1] - '0',
7080 dev_driver_string(dev
->dev
.parent
),
7081 dev_name(dev
->dev
.parent
),
7082 netdev_name(dev
), netdev_reg_state(dev
),
7085 r
= printk("%s%s%s: %pV", level
, netdev_name(dev
),
7086 netdev_reg_state(dev
), vaf
);
7088 r
= printk("%s(NULL net_device): %pV", level
, vaf
);
7094 int netdev_printk(const char *level
, const struct net_device
*dev
,
7095 const char *format
, ...)
7097 struct va_format vaf
;
7101 va_start(args
, format
);
7106 r
= __netdev_printk(level
, dev
, &vaf
);
7112 EXPORT_SYMBOL(netdev_printk
);
7114 #define define_netdev_printk_level(func, level) \
7115 int func(const struct net_device *dev, const char *fmt, ...) \
7118 struct va_format vaf; \
7121 va_start(args, fmt); \
7126 r = __netdev_printk(level, dev, &vaf); \
7132 EXPORT_SYMBOL(func);
7134 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
7135 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
7136 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
7137 define_netdev_printk_level(netdev_err
, KERN_ERR
);
7138 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
7139 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
7140 define_netdev_printk_level(netdev_info
, KERN_INFO
);
7142 static void __net_exit
netdev_exit(struct net
*net
)
7144 kfree(net
->dev_name_head
);
7145 kfree(net
->dev_index_head
);
7148 static struct pernet_operations __net_initdata netdev_net_ops
= {
7149 .init
= netdev_init
,
7150 .exit
= netdev_exit
,
7153 static void __net_exit
default_device_exit(struct net
*net
)
7155 struct net_device
*dev
, *aux
;
7157 * Push all migratable network devices back to the
7158 * initial network namespace
7161 for_each_netdev_safe(net
, dev
, aux
) {
7163 char fb_name
[IFNAMSIZ
];
7165 /* Ignore unmoveable devices (i.e. loopback) */
7166 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
7169 /* Leave virtual devices for the generic cleanup */
7170 if (dev
->rtnl_link_ops
)
7173 /* Push remaining network devices to init_net */
7174 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
7175 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
7177 pr_emerg("%s: failed to move %s to init_net: %d\n",
7178 __func__
, dev
->name
, err
);
7185 static void __net_exit
rtnl_lock_unregistering(struct list_head
*net_list
)
7187 /* Return with the rtnl_lock held when there are no network
7188 * devices unregistering in any network namespace in net_list.
7195 prepare_to_wait(&netdev_unregistering_wq
, &wait
,
7196 TASK_UNINTERRUPTIBLE
);
7197 unregistering
= false;
7199 list_for_each_entry(net
, net_list
, exit_list
) {
7200 if (net
->dev_unreg_count
> 0) {
7201 unregistering
= true;
7210 finish_wait(&netdev_unregistering_wq
, &wait
);
7213 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
7215 /* At exit all network devices most be removed from a network
7216 * namespace. Do this in the reverse order of registration.
7217 * Do this across as many network namespaces as possible to
7218 * improve batching efficiency.
7220 struct net_device
*dev
;
7222 LIST_HEAD(dev_kill_list
);
7224 /* To prevent network device cleanup code from dereferencing
7225 * loopback devices or network devices that have been freed
7226 * wait here for all pending unregistrations to complete,
7227 * before unregistring the loopback device and allowing the
7228 * network namespace be freed.
7230 * The netdev todo list containing all network devices
7231 * unregistrations that happen in default_device_exit_batch
7232 * will run in the rtnl_unlock() at the end of
7233 * default_device_exit_batch.
7235 rtnl_lock_unregistering(net_list
);
7236 list_for_each_entry(net
, net_list
, exit_list
) {
7237 for_each_netdev_reverse(net
, dev
) {
7238 if (dev
->rtnl_link_ops
&& dev
->rtnl_link_ops
->dellink
)
7239 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
7241 unregister_netdevice_queue(dev
, &dev_kill_list
);
7244 unregister_netdevice_many(&dev_kill_list
);
7248 static struct pernet_operations __net_initdata default_device_ops
= {
7249 .exit
= default_device_exit
,
7250 .exit_batch
= default_device_exit_batch
,
7254 * Initialize the DEV module. At boot time this walks the device list and
7255 * unhooks any devices that fail to initialise (normally hardware not
7256 * present) and leaves us with a valid list of present and active devices.
7261 * This is called single threaded during boot, so no need
7262 * to take the rtnl semaphore.
7264 static int __init
net_dev_init(void)
7266 int i
, rc
= -ENOMEM
;
7268 BUG_ON(!dev_boot_phase
);
7270 if (dev_proc_init())
7273 if (netdev_kobject_init())
7276 INIT_LIST_HEAD(&ptype_all
);
7277 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
7278 INIT_LIST_HEAD(&ptype_base
[i
]);
7280 INIT_LIST_HEAD(&offload_base
);
7282 if (register_pernet_subsys(&netdev_net_ops
))
7286 * Initialise the packet receive queues.
7289 for_each_possible_cpu(i
) {
7290 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
7292 skb_queue_head_init(&sd
->input_pkt_queue
);
7293 skb_queue_head_init(&sd
->process_queue
);
7294 INIT_LIST_HEAD(&sd
->poll_list
);
7295 sd
->output_queue_tailp
= &sd
->output_queue
;
7297 sd
->csd
.func
= rps_trigger_softirq
;
7302 sd
->backlog
.poll
= process_backlog
;
7303 sd
->backlog
.weight
= weight_p
;
7308 /* The loopback device is special if any other network devices
7309 * is present in a network namespace the loopback device must
7310 * be present. Since we now dynamically allocate and free the
7311 * loopback device ensure this invariant is maintained by
7312 * keeping the loopback device as the first device on the
7313 * list of network devices. Ensuring the loopback devices
7314 * is the first device that appears and the last network device
7317 if (register_pernet_device(&loopback_net_ops
))
7320 if (register_pernet_device(&default_device_ops
))
7323 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
7324 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
7326 hotcpu_notifier(dev_cpu_callback
, 0);
7333 subsys_initcall(net_dev_init
);