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 __dev_kfree_skb_irq(struct sk_buff
*skb
, enum skb_free_reason reason
)
2182 unsigned long flags
;
2184 if (likely(atomic_read(&skb
->users
) == 1)) {
2186 atomic_set(&skb
->users
, 0);
2187 } else if (likely(!atomic_dec_and_test(&skb
->users
))) {
2190 get_kfree_skb_cb(skb
)->reason
= reason
;
2191 local_irq_save(flags
);
2192 skb
->next
= __this_cpu_read(softnet_data
.completion_queue
);
2193 __this_cpu_write(softnet_data
.completion_queue
, skb
);
2194 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
2195 local_irq_restore(flags
);
2197 EXPORT_SYMBOL(__dev_kfree_skb_irq
);
2199 void __dev_kfree_skb_any(struct sk_buff
*skb
, enum skb_free_reason reason
)
2201 if (in_irq() || irqs_disabled())
2202 __dev_kfree_skb_irq(skb
, reason
);
2206 EXPORT_SYMBOL(__dev_kfree_skb_any
);
2210 * netif_device_detach - mark device as removed
2211 * @dev: network device
2213 * Mark device as removed from system and therefore no longer available.
2215 void netif_device_detach(struct net_device
*dev
)
2217 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
2218 netif_running(dev
)) {
2219 netif_tx_stop_all_queues(dev
);
2222 EXPORT_SYMBOL(netif_device_detach
);
2225 * netif_device_attach - mark device as attached
2226 * @dev: network device
2228 * Mark device as attached from system and restart if needed.
2230 void netif_device_attach(struct net_device
*dev
)
2232 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
2233 netif_running(dev
)) {
2234 netif_tx_wake_all_queues(dev
);
2235 __netdev_watchdog_up(dev
);
2238 EXPORT_SYMBOL(netif_device_attach
);
2240 static void skb_warn_bad_offload(const struct sk_buff
*skb
)
2242 static const netdev_features_t null_features
= 0;
2243 struct net_device
*dev
= skb
->dev
;
2244 const char *driver
= "";
2246 if (!net_ratelimit())
2249 if (dev
&& dev
->dev
.parent
)
2250 driver
= dev_driver_string(dev
->dev
.parent
);
2252 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2253 "gso_type=%d ip_summed=%d\n",
2254 driver
, dev
? &dev
->features
: &null_features
,
2255 skb
->sk
? &skb
->sk
->sk_route_caps
: &null_features
,
2256 skb
->len
, skb
->data_len
, skb_shinfo(skb
)->gso_size
,
2257 skb_shinfo(skb
)->gso_type
, skb
->ip_summed
);
2261 * Invalidate hardware checksum when packet is to be mangled, and
2262 * complete checksum manually on outgoing path.
2264 int skb_checksum_help(struct sk_buff
*skb
)
2267 int ret
= 0, offset
;
2269 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
2270 goto out_set_summed
;
2272 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
2273 skb_warn_bad_offload(skb
);
2277 /* Before computing a checksum, we should make sure no frag could
2278 * be modified by an external entity : checksum could be wrong.
2280 if (skb_has_shared_frag(skb
)) {
2281 ret
= __skb_linearize(skb
);
2286 offset
= skb_checksum_start_offset(skb
);
2287 BUG_ON(offset
>= skb_headlen(skb
));
2288 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
2290 offset
+= skb
->csum_offset
;
2291 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
2293 if (skb_cloned(skb
) &&
2294 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
2295 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2300 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
2302 skb
->ip_summed
= CHECKSUM_NONE
;
2306 EXPORT_SYMBOL(skb_checksum_help
);
2308 __be16
skb_network_protocol(struct sk_buff
*skb
, int *depth
)
2310 unsigned int vlan_depth
= skb
->mac_len
;
2311 __be16 type
= skb
->protocol
;
2313 /* Tunnel gso handlers can set protocol to ethernet. */
2314 if (type
== htons(ETH_P_TEB
)) {
2317 if (unlikely(!pskb_may_pull(skb
, sizeof(struct ethhdr
))))
2320 eth
= (struct ethhdr
*)skb_mac_header(skb
);
2321 type
= eth
->h_proto
;
2324 /* if skb->protocol is 802.1Q/AD then the header should already be
2325 * present at mac_len - VLAN_HLEN (if mac_len > 0), or at
2326 * ETH_HLEN otherwise
2328 if (type
== htons(ETH_P_8021Q
) || type
== htons(ETH_P_8021AD
)) {
2330 if (WARN_ON(vlan_depth
< VLAN_HLEN
))
2332 vlan_depth
-= VLAN_HLEN
;
2334 vlan_depth
= ETH_HLEN
;
2337 struct vlan_hdr
*vh
;
2339 if (unlikely(!pskb_may_pull(skb
,
2340 vlan_depth
+ VLAN_HLEN
)))
2343 vh
= (struct vlan_hdr
*)(skb
->data
+ vlan_depth
);
2344 type
= vh
->h_vlan_encapsulated_proto
;
2345 vlan_depth
+= VLAN_HLEN
;
2346 } while (type
== htons(ETH_P_8021Q
) ||
2347 type
== htons(ETH_P_8021AD
));
2350 *depth
= vlan_depth
;
2356 * skb_mac_gso_segment - mac layer segmentation handler.
2357 * @skb: buffer to segment
2358 * @features: features for the output path (see dev->features)
2360 struct sk_buff
*skb_mac_gso_segment(struct sk_buff
*skb
,
2361 netdev_features_t features
)
2363 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
2364 struct packet_offload
*ptype
;
2365 int vlan_depth
= skb
->mac_len
;
2366 __be16 type
= skb_network_protocol(skb
, &vlan_depth
);
2368 if (unlikely(!type
))
2369 return ERR_PTR(-EINVAL
);
2371 __skb_pull(skb
, vlan_depth
);
2374 list_for_each_entry_rcu(ptype
, &offload_base
, list
) {
2375 if (ptype
->type
== type
&& ptype
->callbacks
.gso_segment
) {
2376 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
2379 err
= ptype
->callbacks
.gso_send_check(skb
);
2380 segs
= ERR_PTR(err
);
2381 if (err
|| skb_gso_ok(skb
, features
))
2383 __skb_push(skb
, (skb
->data
-
2384 skb_network_header(skb
)));
2386 segs
= ptype
->callbacks
.gso_segment(skb
, features
);
2392 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
2396 EXPORT_SYMBOL(skb_mac_gso_segment
);
2399 /* openvswitch calls this on rx path, so we need a different check.
2401 static inline bool skb_needs_check(struct sk_buff
*skb
, bool tx_path
)
2404 return skb
->ip_summed
!= CHECKSUM_PARTIAL
;
2406 return skb
->ip_summed
== CHECKSUM_NONE
;
2410 * __skb_gso_segment - Perform segmentation on skb.
2411 * @skb: buffer to segment
2412 * @features: features for the output path (see dev->features)
2413 * @tx_path: whether it is called in TX path
2415 * This function segments the given skb and returns a list of segments.
2417 * It may return NULL if the skb requires no segmentation. This is
2418 * only possible when GSO is used for verifying header integrity.
2420 struct sk_buff
*__skb_gso_segment(struct sk_buff
*skb
,
2421 netdev_features_t features
, bool tx_path
)
2423 if (unlikely(skb_needs_check(skb
, tx_path
))) {
2426 skb_warn_bad_offload(skb
);
2428 err
= skb_cow_head(skb
, 0);
2430 return ERR_PTR(err
);
2433 SKB_GSO_CB(skb
)->mac_offset
= skb_headroom(skb
);
2434 SKB_GSO_CB(skb
)->encap_level
= 0;
2436 skb_reset_mac_header(skb
);
2437 skb_reset_mac_len(skb
);
2439 return skb_mac_gso_segment(skb
, features
);
2441 EXPORT_SYMBOL(__skb_gso_segment
);
2443 /* Take action when hardware reception checksum errors are detected. */
2445 void netdev_rx_csum_fault(struct net_device
*dev
)
2447 if (net_ratelimit()) {
2448 pr_err("%s: hw csum failure\n", dev
? dev
->name
: "<unknown>");
2452 EXPORT_SYMBOL(netdev_rx_csum_fault
);
2455 /* Actually, we should eliminate this check as soon as we know, that:
2456 * 1. IOMMU is present and allows to map all the memory.
2457 * 2. No high memory really exists on this machine.
2460 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
2462 #ifdef CONFIG_HIGHMEM
2464 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
2465 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2466 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2467 if (PageHighMem(skb_frag_page(frag
)))
2472 if (PCI_DMA_BUS_IS_PHYS
) {
2473 struct device
*pdev
= dev
->dev
.parent
;
2477 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2478 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2479 dma_addr_t addr
= page_to_phys(skb_frag_page(frag
));
2480 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
2489 void (*destructor
)(struct sk_buff
*skb
);
2492 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2494 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
2496 struct dev_gso_cb
*cb
;
2498 kfree_skb_list(skb
->next
);
2501 cb
= DEV_GSO_CB(skb
);
2503 cb
->destructor(skb
);
2507 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2508 * @skb: buffer to segment
2509 * @features: device features as applicable to this skb
2511 * This function segments the given skb and stores the list of segments
2514 static int dev_gso_segment(struct sk_buff
*skb
, netdev_features_t features
)
2516 struct sk_buff
*segs
;
2518 segs
= skb_gso_segment(skb
, features
);
2520 /* Verifying header integrity only. */
2525 return PTR_ERR(segs
);
2528 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
2529 skb
->destructor
= dev_gso_skb_destructor
;
2534 /* If MPLS offload request, verify we are testing hardware MPLS features
2535 * instead of standard features for the netdev.
2537 #ifdef CONFIG_NET_MPLS_GSO
2538 static netdev_features_t
net_mpls_features(struct sk_buff
*skb
,
2539 netdev_features_t features
,
2542 if (type
== htons(ETH_P_MPLS_UC
) || type
== htons(ETH_P_MPLS_MC
))
2543 features
&= skb
->dev
->mpls_features
;
2548 static netdev_features_t
net_mpls_features(struct sk_buff
*skb
,
2549 netdev_features_t features
,
2556 static netdev_features_t
harmonize_features(struct sk_buff
*skb
,
2557 netdev_features_t features
)
2562 type
= skb_network_protocol(skb
, &tmp
);
2563 features
= net_mpls_features(skb
, features
, type
);
2565 if (skb
->ip_summed
!= CHECKSUM_NONE
&&
2566 !can_checksum_protocol(features
, type
)) {
2567 features
&= ~NETIF_F_ALL_CSUM
;
2568 } else if (illegal_highdma(skb
->dev
, skb
)) {
2569 features
&= ~NETIF_F_SG
;
2575 netdev_features_t
netif_skb_features(struct sk_buff
*skb
)
2577 __be16 protocol
= skb
->protocol
;
2578 netdev_features_t features
= skb
->dev
->features
;
2580 if (skb_shinfo(skb
)->gso_segs
> skb
->dev
->gso_max_segs
)
2581 features
&= ~NETIF_F_GSO_MASK
;
2583 if (protocol
== htons(ETH_P_8021Q
) || protocol
== htons(ETH_P_8021AD
)) {
2584 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
2585 protocol
= veh
->h_vlan_encapsulated_proto
;
2586 } else if (!vlan_tx_tag_present(skb
)) {
2587 return harmonize_features(skb
, features
);
2590 features
&= (skb
->dev
->vlan_features
| NETIF_F_HW_VLAN_CTAG_TX
|
2591 NETIF_F_HW_VLAN_STAG_TX
);
2593 if (protocol
== htons(ETH_P_8021Q
) || protocol
== htons(ETH_P_8021AD
))
2594 features
&= NETIF_F_SG
| NETIF_F_HIGHDMA
| NETIF_F_FRAGLIST
|
2595 NETIF_F_GEN_CSUM
| NETIF_F_HW_VLAN_CTAG_TX
|
2596 NETIF_F_HW_VLAN_STAG_TX
;
2598 return harmonize_features(skb
, features
);
2600 EXPORT_SYMBOL(netif_skb_features
);
2602 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
2603 struct netdev_queue
*txq
)
2605 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2606 int rc
= NETDEV_TX_OK
;
2607 unsigned int skb_len
;
2609 if (likely(!skb
->next
)) {
2610 netdev_features_t features
;
2613 * If device doesn't need skb->dst, release it right now while
2614 * its hot in this cpu cache
2616 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2619 features
= netif_skb_features(skb
);
2621 if (vlan_tx_tag_present(skb
) &&
2622 !vlan_hw_offload_capable(features
, skb
->vlan_proto
)) {
2623 skb
= __vlan_put_tag(skb
, skb
->vlan_proto
,
2624 vlan_tx_tag_get(skb
));
2631 /* If encapsulation offload request, verify we are testing
2632 * hardware encapsulation features instead of standard
2633 * features for the netdev
2635 if (skb
->encapsulation
)
2636 features
&= dev
->hw_enc_features
;
2638 if (netif_needs_gso(skb
, features
)) {
2639 if (unlikely(dev_gso_segment(skb
, features
)))
2644 if (skb_needs_linearize(skb
, features
) &&
2645 __skb_linearize(skb
))
2648 /* If packet is not checksummed and device does not
2649 * support checksumming for this protocol, complete
2650 * checksumming here.
2652 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2653 if (skb
->encapsulation
)
2654 skb_set_inner_transport_header(skb
,
2655 skb_checksum_start_offset(skb
));
2657 skb_set_transport_header(skb
,
2658 skb_checksum_start_offset(skb
));
2659 if (!(features
& NETIF_F_ALL_CSUM
) &&
2660 skb_checksum_help(skb
))
2665 if (!list_empty(&ptype_all
))
2666 dev_queue_xmit_nit(skb
, dev
);
2669 trace_net_dev_start_xmit(skb
, dev
);
2670 rc
= ops
->ndo_start_xmit(skb
, dev
);
2671 trace_net_dev_xmit(skb
, rc
, dev
, skb_len
);
2672 if (rc
== NETDEV_TX_OK
)
2673 txq_trans_update(txq
);
2679 struct sk_buff
*nskb
= skb
->next
;
2681 skb
->next
= nskb
->next
;
2684 if (!list_empty(&ptype_all
))
2685 dev_queue_xmit_nit(nskb
, dev
);
2687 skb_len
= nskb
->len
;
2688 trace_net_dev_start_xmit(nskb
, dev
);
2689 rc
= ops
->ndo_start_xmit(nskb
, dev
);
2690 trace_net_dev_xmit(nskb
, rc
, dev
, skb_len
);
2691 if (unlikely(rc
!= NETDEV_TX_OK
)) {
2692 if (rc
& ~NETDEV_TX_MASK
)
2693 goto out_kfree_gso_skb
;
2694 nskb
->next
= skb
->next
;
2698 txq_trans_update(txq
);
2699 if (unlikely(netif_xmit_stopped(txq
) && skb
->next
))
2700 return NETDEV_TX_BUSY
;
2701 } while (skb
->next
);
2704 if (likely(skb
->next
== NULL
)) {
2705 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
2714 EXPORT_SYMBOL_GPL(dev_hard_start_xmit
);
2716 static void qdisc_pkt_len_init(struct sk_buff
*skb
)
2718 const struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
2720 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
2722 /* To get more precise estimation of bytes sent on wire,
2723 * we add to pkt_len the headers size of all segments
2725 if (shinfo
->gso_size
) {
2726 unsigned int hdr_len
;
2727 u16 gso_segs
= shinfo
->gso_segs
;
2729 /* mac layer + network layer */
2730 hdr_len
= skb_transport_header(skb
) - skb_mac_header(skb
);
2732 /* + transport layer */
2733 if (likely(shinfo
->gso_type
& (SKB_GSO_TCPV4
| SKB_GSO_TCPV6
)))
2734 hdr_len
+= tcp_hdrlen(skb
);
2736 hdr_len
+= sizeof(struct udphdr
);
2738 if (shinfo
->gso_type
& SKB_GSO_DODGY
)
2739 gso_segs
= DIV_ROUND_UP(skb
->len
- hdr_len
,
2742 qdisc_skb_cb(skb
)->pkt_len
+= (gso_segs
- 1) * hdr_len
;
2746 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2747 struct net_device
*dev
,
2748 struct netdev_queue
*txq
)
2750 spinlock_t
*root_lock
= qdisc_lock(q
);
2754 qdisc_pkt_len_init(skb
);
2755 qdisc_calculate_pkt_len(skb
, q
);
2757 * Heuristic to force contended enqueues to serialize on a
2758 * separate lock before trying to get qdisc main lock.
2759 * This permits __QDISC___STATE_RUNNING owner to get the lock more
2760 * often and dequeue packets faster.
2762 contended
= qdisc_is_running(q
);
2763 if (unlikely(contended
))
2764 spin_lock(&q
->busylock
);
2766 spin_lock(root_lock
);
2767 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2770 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2771 qdisc_run_begin(q
)) {
2773 * This is a work-conserving queue; there are no old skbs
2774 * waiting to be sent out; and the qdisc is not running -
2775 * xmit the skb directly.
2777 if (!(dev
->priv_flags
& IFF_XMIT_DST_RELEASE
))
2780 qdisc_bstats_update(q
, skb
);
2782 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
)) {
2783 if (unlikely(contended
)) {
2784 spin_unlock(&q
->busylock
);
2791 rc
= NET_XMIT_SUCCESS
;
2794 rc
= q
->enqueue(skb
, q
) & NET_XMIT_MASK
;
2795 if (qdisc_run_begin(q
)) {
2796 if (unlikely(contended
)) {
2797 spin_unlock(&q
->busylock
);
2803 spin_unlock(root_lock
);
2804 if (unlikely(contended
))
2805 spin_unlock(&q
->busylock
);
2809 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2810 static void skb_update_prio(struct sk_buff
*skb
)
2812 struct netprio_map
*map
= rcu_dereference_bh(skb
->dev
->priomap
);
2814 if (!skb
->priority
&& skb
->sk
&& map
) {
2815 unsigned int prioidx
= skb
->sk
->sk_cgrp_prioidx
;
2817 if (prioidx
< map
->priomap_len
)
2818 skb
->priority
= map
->priomap
[prioidx
];
2822 #define skb_update_prio(skb)
2825 static DEFINE_PER_CPU(int, xmit_recursion
);
2826 #define RECURSION_LIMIT 10
2829 * dev_loopback_xmit - loop back @skb
2830 * @skb: buffer to transmit
2832 int dev_loopback_xmit(struct sk_buff
*skb
)
2834 skb_reset_mac_header(skb
);
2835 __skb_pull(skb
, skb_network_offset(skb
));
2836 skb
->pkt_type
= PACKET_LOOPBACK
;
2837 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
2838 WARN_ON(!skb_dst(skb
));
2843 EXPORT_SYMBOL(dev_loopback_xmit
);
2846 * __dev_queue_xmit - transmit a buffer
2847 * @skb: buffer to transmit
2848 * @accel_priv: private data used for L2 forwarding offload
2850 * Queue a buffer for transmission to a network device. The caller must
2851 * have set the device and priority and built the buffer before calling
2852 * this function. The function can be called from an interrupt.
2854 * A negative errno code is returned on a failure. A success does not
2855 * guarantee the frame will be transmitted as it may be dropped due
2856 * to congestion or traffic shaping.
2858 * -----------------------------------------------------------------------------------
2859 * I notice this method can also return errors from the queue disciplines,
2860 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2863 * Regardless of the return value, the skb is consumed, so it is currently
2864 * difficult to retry a send to this method. (You can bump the ref count
2865 * before sending to hold a reference for retry if you are careful.)
2867 * When calling this method, interrupts MUST be enabled. This is because
2868 * the BH enable code must have IRQs enabled so that it will not deadlock.
2871 static int __dev_queue_xmit(struct sk_buff
*skb
, void *accel_priv
)
2873 struct net_device
*dev
= skb
->dev
;
2874 struct netdev_queue
*txq
;
2878 skb_reset_mac_header(skb
);
2880 if (unlikely(skb_shinfo(skb
)->tx_flags
& SKBTX_SCHED_TSTAMP
))
2881 __skb_tstamp_tx(skb
, NULL
, skb
->sk
, SCM_TSTAMP_SCHED
);
2883 /* Disable soft irqs for various locks below. Also
2884 * stops preemption for RCU.
2888 skb_update_prio(skb
);
2890 txq
= netdev_pick_tx(dev
, skb
, accel_priv
);
2891 q
= rcu_dereference_bh(txq
->qdisc
);
2893 #ifdef CONFIG_NET_CLS_ACT
2894 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2896 trace_net_dev_queue(skb
);
2898 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2902 /* The device has no queue. Common case for software devices:
2903 loopback, all the sorts of tunnels...
2905 Really, it is unlikely that netif_tx_lock protection is necessary
2906 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2908 However, it is possible, that they rely on protection
2911 Check this and shot the lock. It is not prone from deadlocks.
2912 Either shot noqueue qdisc, it is even simpler 8)
2914 if (dev
->flags
& IFF_UP
) {
2915 int cpu
= smp_processor_id(); /* ok because BHs are off */
2917 if (txq
->xmit_lock_owner
!= cpu
) {
2919 if (__this_cpu_read(xmit_recursion
) > RECURSION_LIMIT
)
2920 goto recursion_alert
;
2922 HARD_TX_LOCK(dev
, txq
, cpu
);
2924 if (!netif_xmit_stopped(txq
)) {
2925 __this_cpu_inc(xmit_recursion
);
2926 rc
= dev_hard_start_xmit(skb
, dev
, txq
);
2927 __this_cpu_dec(xmit_recursion
);
2928 if (dev_xmit_complete(rc
)) {
2929 HARD_TX_UNLOCK(dev
, txq
);
2933 HARD_TX_UNLOCK(dev
, txq
);
2934 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2937 /* Recursion is detected! It is possible,
2941 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2947 rcu_read_unlock_bh();
2949 atomic_long_inc(&dev
->tx_dropped
);
2953 rcu_read_unlock_bh();
2957 int dev_queue_xmit(struct sk_buff
*skb
)
2959 return __dev_queue_xmit(skb
, NULL
);
2961 EXPORT_SYMBOL(dev_queue_xmit
);
2963 int dev_queue_xmit_accel(struct sk_buff
*skb
, void *accel_priv
)
2965 return __dev_queue_xmit(skb
, accel_priv
);
2967 EXPORT_SYMBOL(dev_queue_xmit_accel
);
2970 /*=======================================================================
2972 =======================================================================*/
2974 int netdev_max_backlog __read_mostly
= 1000;
2975 EXPORT_SYMBOL(netdev_max_backlog
);
2977 int netdev_tstamp_prequeue __read_mostly
= 1;
2978 int netdev_budget __read_mostly
= 300;
2979 int weight_p __read_mostly
= 64; /* old backlog weight */
2981 /* Called with irq disabled */
2982 static inline void ____napi_schedule(struct softnet_data
*sd
,
2983 struct napi_struct
*napi
)
2985 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
2986 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2991 /* One global table that all flow-based protocols share. */
2992 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
2993 EXPORT_SYMBOL(rps_sock_flow_table
);
2995 struct static_key rps_needed __read_mostly
;
2997 static struct rps_dev_flow
*
2998 set_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2999 struct rps_dev_flow
*rflow
, u16 next_cpu
)
3001 if (next_cpu
!= RPS_NO_CPU
) {
3002 #ifdef CONFIG_RFS_ACCEL
3003 struct netdev_rx_queue
*rxqueue
;
3004 struct rps_dev_flow_table
*flow_table
;
3005 struct rps_dev_flow
*old_rflow
;
3010 /* Should we steer this flow to a different hardware queue? */
3011 if (!skb_rx_queue_recorded(skb
) || !dev
->rx_cpu_rmap
||
3012 !(dev
->features
& NETIF_F_NTUPLE
))
3014 rxq_index
= cpu_rmap_lookup_index(dev
->rx_cpu_rmap
, next_cpu
);
3015 if (rxq_index
== skb_get_rx_queue(skb
))
3018 rxqueue
= dev
->_rx
+ rxq_index
;
3019 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3022 flow_id
= skb_get_hash(skb
) & flow_table
->mask
;
3023 rc
= dev
->netdev_ops
->ndo_rx_flow_steer(dev
, skb
,
3024 rxq_index
, flow_id
);
3028 rflow
= &flow_table
->flows
[flow_id
];
3030 if (old_rflow
->filter
== rflow
->filter
)
3031 old_rflow
->filter
= RPS_NO_FILTER
;
3035 per_cpu(softnet_data
, next_cpu
).input_queue_head
;
3038 rflow
->cpu
= next_cpu
;
3043 * get_rps_cpu is called from netif_receive_skb and returns the target
3044 * CPU from the RPS map of the receiving queue for a given skb.
3045 * rcu_read_lock must be held on entry.
3047 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
3048 struct rps_dev_flow
**rflowp
)
3050 struct netdev_rx_queue
*rxqueue
;
3051 struct rps_map
*map
;
3052 struct rps_dev_flow_table
*flow_table
;
3053 struct rps_sock_flow_table
*sock_flow_table
;
3058 if (skb_rx_queue_recorded(skb
)) {
3059 u16 index
= skb_get_rx_queue(skb
);
3060 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
3061 WARN_ONCE(dev
->real_num_rx_queues
> 1,
3062 "%s received packet on queue %u, but number "
3063 "of RX queues is %u\n",
3064 dev
->name
, index
, dev
->real_num_rx_queues
);
3067 rxqueue
= dev
->_rx
+ index
;
3071 map
= rcu_dereference(rxqueue
->rps_map
);
3073 if (map
->len
== 1 &&
3074 !rcu_access_pointer(rxqueue
->rps_flow_table
)) {
3075 tcpu
= map
->cpus
[0];
3076 if (cpu_online(tcpu
))
3080 } else if (!rcu_access_pointer(rxqueue
->rps_flow_table
)) {
3084 skb_reset_network_header(skb
);
3085 hash
= skb_get_hash(skb
);
3089 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3090 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
3091 if (flow_table
&& sock_flow_table
) {
3093 struct rps_dev_flow
*rflow
;
3095 rflow
= &flow_table
->flows
[hash
& flow_table
->mask
];
3098 next_cpu
= sock_flow_table
->ents
[hash
& sock_flow_table
->mask
];
3101 * If the desired CPU (where last recvmsg was done) is
3102 * different from current CPU (one in the rx-queue flow
3103 * table entry), switch if one of the following holds:
3104 * - Current CPU is unset (equal to RPS_NO_CPU).
3105 * - Current CPU is offline.
3106 * - The current CPU's queue tail has advanced beyond the
3107 * last packet that was enqueued using this table entry.
3108 * This guarantees that all previous packets for the flow
3109 * have been dequeued, thus preserving in order delivery.
3111 if (unlikely(tcpu
!= next_cpu
) &&
3112 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
3113 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
3114 rflow
->last_qtail
)) >= 0)) {
3116 rflow
= set_rps_cpu(dev
, skb
, rflow
, next_cpu
);
3119 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
3127 tcpu
= map
->cpus
[reciprocal_scale(hash
, map
->len
)];
3128 if (cpu_online(tcpu
)) {
3138 #ifdef CONFIG_RFS_ACCEL
3141 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3142 * @dev: Device on which the filter was set
3143 * @rxq_index: RX queue index
3144 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3145 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3147 * Drivers that implement ndo_rx_flow_steer() should periodically call
3148 * this function for each installed filter and remove the filters for
3149 * which it returns %true.
3151 bool rps_may_expire_flow(struct net_device
*dev
, u16 rxq_index
,
3152 u32 flow_id
, u16 filter_id
)
3154 struct netdev_rx_queue
*rxqueue
= dev
->_rx
+ rxq_index
;
3155 struct rps_dev_flow_table
*flow_table
;
3156 struct rps_dev_flow
*rflow
;
3161 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3162 if (flow_table
&& flow_id
<= flow_table
->mask
) {
3163 rflow
= &flow_table
->flows
[flow_id
];
3164 cpu
= ACCESS_ONCE(rflow
->cpu
);
3165 if (rflow
->filter
== filter_id
&& cpu
!= RPS_NO_CPU
&&
3166 ((int)(per_cpu(softnet_data
, cpu
).input_queue_head
-
3167 rflow
->last_qtail
) <
3168 (int)(10 * flow_table
->mask
)))
3174 EXPORT_SYMBOL(rps_may_expire_flow
);
3176 #endif /* CONFIG_RFS_ACCEL */
3178 /* Called from hardirq (IPI) context */
3179 static void rps_trigger_softirq(void *data
)
3181 struct softnet_data
*sd
= data
;
3183 ____napi_schedule(sd
, &sd
->backlog
);
3187 #endif /* CONFIG_RPS */
3190 * Check if this softnet_data structure is another cpu one
3191 * If yes, queue it to our IPI list and return 1
3194 static int rps_ipi_queued(struct softnet_data
*sd
)
3197 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
3200 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
3201 mysd
->rps_ipi_list
= sd
;
3203 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3206 #endif /* CONFIG_RPS */
3210 #ifdef CONFIG_NET_FLOW_LIMIT
3211 int netdev_flow_limit_table_len __read_mostly
= (1 << 12);
3214 static bool skb_flow_limit(struct sk_buff
*skb
, unsigned int qlen
)
3216 #ifdef CONFIG_NET_FLOW_LIMIT
3217 struct sd_flow_limit
*fl
;
3218 struct softnet_data
*sd
;
3219 unsigned int old_flow
, new_flow
;
3221 if (qlen
< (netdev_max_backlog
>> 1))
3224 sd
= &__get_cpu_var(softnet_data
);
3227 fl
= rcu_dereference(sd
->flow_limit
);
3229 new_flow
= skb_get_hash(skb
) & (fl
->num_buckets
- 1);
3230 old_flow
= fl
->history
[fl
->history_head
];
3231 fl
->history
[fl
->history_head
] = new_flow
;
3234 fl
->history_head
&= FLOW_LIMIT_HISTORY
- 1;
3236 if (likely(fl
->buckets
[old_flow
]))
3237 fl
->buckets
[old_flow
]--;
3239 if (++fl
->buckets
[new_flow
] > (FLOW_LIMIT_HISTORY
>> 1)) {
3251 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3252 * queue (may be a remote CPU queue).
3254 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
3255 unsigned int *qtail
)
3257 struct softnet_data
*sd
;
3258 unsigned long flags
;
3261 sd
= &per_cpu(softnet_data
, cpu
);
3263 local_irq_save(flags
);
3266 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3267 if (qlen
<= netdev_max_backlog
&& !skb_flow_limit(skb
, qlen
)) {
3268 if (skb_queue_len(&sd
->input_pkt_queue
)) {
3270 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
3271 input_queue_tail_incr_save(sd
, qtail
);
3273 local_irq_restore(flags
);
3274 return NET_RX_SUCCESS
;
3277 /* Schedule NAPI for backlog device
3278 * We can use non atomic operation since we own the queue lock
3280 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
3281 if (!rps_ipi_queued(sd
))
3282 ____napi_schedule(sd
, &sd
->backlog
);
3290 local_irq_restore(flags
);
3292 atomic_long_inc(&skb
->dev
->rx_dropped
);
3297 static int netif_rx_internal(struct sk_buff
*skb
)
3301 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3303 trace_netif_rx(skb
);
3305 if (static_key_false(&rps_needed
)) {
3306 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3312 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3314 cpu
= smp_processor_id();
3316 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3324 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
3331 * netif_rx - post buffer to the network code
3332 * @skb: buffer to post
3334 * This function receives a packet from a device driver and queues it for
3335 * the upper (protocol) levels to process. It always succeeds. The buffer
3336 * may be dropped during processing for congestion control or by the
3340 * NET_RX_SUCCESS (no congestion)
3341 * NET_RX_DROP (packet was dropped)
3345 int netif_rx(struct sk_buff
*skb
)
3347 trace_netif_rx_entry(skb
);
3349 return netif_rx_internal(skb
);
3351 EXPORT_SYMBOL(netif_rx
);
3353 int netif_rx_ni(struct sk_buff
*skb
)
3357 trace_netif_rx_ni_entry(skb
);
3360 err
= netif_rx_internal(skb
);
3361 if (local_softirq_pending())
3367 EXPORT_SYMBOL(netif_rx_ni
);
3369 static void net_tx_action(struct softirq_action
*h
)
3371 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3373 if (sd
->completion_queue
) {
3374 struct sk_buff
*clist
;
3376 local_irq_disable();
3377 clist
= sd
->completion_queue
;
3378 sd
->completion_queue
= NULL
;
3382 struct sk_buff
*skb
= clist
;
3383 clist
= clist
->next
;
3385 WARN_ON(atomic_read(&skb
->users
));
3386 if (likely(get_kfree_skb_cb(skb
)->reason
== SKB_REASON_CONSUMED
))
3387 trace_consume_skb(skb
);
3389 trace_kfree_skb(skb
, net_tx_action
);
3394 if (sd
->output_queue
) {
3397 local_irq_disable();
3398 head
= sd
->output_queue
;
3399 sd
->output_queue
= NULL
;
3400 sd
->output_queue_tailp
= &sd
->output_queue
;
3404 struct Qdisc
*q
= head
;
3405 spinlock_t
*root_lock
;
3407 head
= head
->next_sched
;
3409 root_lock
= qdisc_lock(q
);
3410 if (spin_trylock(root_lock
)) {
3411 smp_mb__before_atomic();
3412 clear_bit(__QDISC_STATE_SCHED
,
3415 spin_unlock(root_lock
);
3417 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
3419 __netif_reschedule(q
);
3421 smp_mb__before_atomic();
3422 clear_bit(__QDISC_STATE_SCHED
,
3430 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3431 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3432 /* This hook is defined here for ATM LANE */
3433 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
3434 unsigned char *addr
) __read_mostly
;
3435 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
3438 #ifdef CONFIG_NET_CLS_ACT
3439 /* TODO: Maybe we should just force sch_ingress to be compiled in
3440 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3441 * a compare and 2 stores extra right now if we dont have it on
3442 * but have CONFIG_NET_CLS_ACT
3443 * NOTE: This doesn't stop any functionality; if you dont have
3444 * the ingress scheduler, you just can't add policies on ingress.
3447 static int ing_filter(struct sk_buff
*skb
, struct netdev_queue
*rxq
)
3449 struct net_device
*dev
= skb
->dev
;
3450 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
3451 int result
= TC_ACT_OK
;
3454 if (unlikely(MAX_RED_LOOP
< ttl
++)) {
3455 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3456 skb
->skb_iif
, dev
->ifindex
);
3460 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
3461 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
3464 if (q
!= &noop_qdisc
) {
3465 spin_lock(qdisc_lock(q
));
3466 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
3467 result
= qdisc_enqueue_root(skb
, q
);
3468 spin_unlock(qdisc_lock(q
));
3474 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
3475 struct packet_type
**pt_prev
,
3476 int *ret
, struct net_device
*orig_dev
)
3478 struct netdev_queue
*rxq
= rcu_dereference(skb
->dev
->ingress_queue
);
3480 if (!rxq
|| rxq
->qdisc
== &noop_qdisc
)
3484 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
3488 switch (ing_filter(skb
, rxq
)) {
3502 * netdev_rx_handler_register - register receive handler
3503 * @dev: device to register a handler for
3504 * @rx_handler: receive handler to register
3505 * @rx_handler_data: data pointer that is used by rx handler
3507 * Register a receive handler for a device. This handler will then be
3508 * called from __netif_receive_skb. A negative errno code is returned
3511 * The caller must hold the rtnl_mutex.
3513 * For a general description of rx_handler, see enum rx_handler_result.
3515 int netdev_rx_handler_register(struct net_device
*dev
,
3516 rx_handler_func_t
*rx_handler
,
3517 void *rx_handler_data
)
3521 if (dev
->rx_handler
)
3524 /* Note: rx_handler_data must be set before rx_handler */
3525 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
3526 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
3530 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
3533 * netdev_rx_handler_unregister - unregister receive handler
3534 * @dev: device to unregister a handler from
3536 * Unregister a receive handler from a device.
3538 * The caller must hold the rtnl_mutex.
3540 void netdev_rx_handler_unregister(struct net_device
*dev
)
3544 RCU_INIT_POINTER(dev
->rx_handler
, NULL
);
3545 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3546 * section has a guarantee to see a non NULL rx_handler_data
3550 RCU_INIT_POINTER(dev
->rx_handler_data
, NULL
);
3552 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
3555 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3556 * the special handling of PFMEMALLOC skbs.
3558 static bool skb_pfmemalloc_protocol(struct sk_buff
*skb
)
3560 switch (skb
->protocol
) {
3561 case htons(ETH_P_ARP
):
3562 case htons(ETH_P_IP
):
3563 case htons(ETH_P_IPV6
):
3564 case htons(ETH_P_8021Q
):
3565 case htons(ETH_P_8021AD
):
3572 static int __netif_receive_skb_core(struct sk_buff
*skb
, bool pfmemalloc
)
3574 struct packet_type
*ptype
, *pt_prev
;
3575 rx_handler_func_t
*rx_handler
;
3576 struct net_device
*orig_dev
;
3577 struct net_device
*null_or_dev
;
3578 bool deliver_exact
= false;
3579 int ret
= NET_RX_DROP
;
3582 net_timestamp_check(!netdev_tstamp_prequeue
, skb
);
3584 trace_netif_receive_skb(skb
);
3586 orig_dev
= skb
->dev
;
3588 skb_reset_network_header(skb
);
3589 if (!skb_transport_header_was_set(skb
))
3590 skb_reset_transport_header(skb
);
3591 skb_reset_mac_len(skb
);
3598 skb
->skb_iif
= skb
->dev
->ifindex
;
3600 __this_cpu_inc(softnet_data
.processed
);
3602 if (skb
->protocol
== cpu_to_be16(ETH_P_8021Q
) ||
3603 skb
->protocol
== cpu_to_be16(ETH_P_8021AD
)) {
3604 skb
= skb_vlan_untag(skb
);
3609 #ifdef CONFIG_NET_CLS_ACT
3610 if (skb
->tc_verd
& TC_NCLS
) {
3611 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
3619 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
3620 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
3622 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3628 #ifdef CONFIG_NET_CLS_ACT
3629 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
3635 if (pfmemalloc
&& !skb_pfmemalloc_protocol(skb
))
3638 if (vlan_tx_tag_present(skb
)) {
3640 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3643 if (vlan_do_receive(&skb
))
3645 else if (unlikely(!skb
))
3649 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
3652 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3655 switch (rx_handler(&skb
)) {
3656 case RX_HANDLER_CONSUMED
:
3657 ret
= NET_RX_SUCCESS
;
3659 case RX_HANDLER_ANOTHER
:
3661 case RX_HANDLER_EXACT
:
3662 deliver_exact
= true;
3663 case RX_HANDLER_PASS
:
3670 if (unlikely(vlan_tx_tag_present(skb
))) {
3671 if (vlan_tx_tag_get_id(skb
))
3672 skb
->pkt_type
= PACKET_OTHERHOST
;
3673 /* Note: we might in the future use prio bits
3674 * and set skb->priority like in vlan_do_receive()
3675 * For the time being, just ignore Priority Code Point
3680 /* deliver only exact match when indicated */
3681 null_or_dev
= deliver_exact
? skb
->dev
: NULL
;
3683 type
= skb
->protocol
;
3684 list_for_each_entry_rcu(ptype
,
3685 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
3686 if (ptype
->type
== type
&&
3687 (ptype
->dev
== null_or_dev
|| ptype
->dev
== skb
->dev
||
3688 ptype
->dev
== orig_dev
)) {
3690 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3696 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
3699 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
3702 atomic_long_inc(&skb
->dev
->rx_dropped
);
3704 /* Jamal, now you will not able to escape explaining
3705 * me how you were going to use this. :-)
3715 static int __netif_receive_skb(struct sk_buff
*skb
)
3719 if (sk_memalloc_socks() && skb_pfmemalloc(skb
)) {
3720 unsigned long pflags
= current
->flags
;
3723 * PFMEMALLOC skbs are special, they should
3724 * - be delivered to SOCK_MEMALLOC sockets only
3725 * - stay away from userspace
3726 * - have bounded memory usage
3728 * Use PF_MEMALLOC as this saves us from propagating the allocation
3729 * context down to all allocation sites.
3731 current
->flags
|= PF_MEMALLOC
;
3732 ret
= __netif_receive_skb_core(skb
, true);
3733 tsk_restore_flags(current
, pflags
, PF_MEMALLOC
);
3735 ret
= __netif_receive_skb_core(skb
, false);
3740 static int netif_receive_skb_internal(struct sk_buff
*skb
)
3742 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3744 if (skb_defer_rx_timestamp(skb
))
3745 return NET_RX_SUCCESS
;
3748 if (static_key_false(&rps_needed
)) {
3749 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3754 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3757 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3764 return __netif_receive_skb(skb
);
3768 * netif_receive_skb - process receive buffer from network
3769 * @skb: buffer to process
3771 * netif_receive_skb() is the main receive data processing function.
3772 * It always succeeds. The buffer may be dropped during processing
3773 * for congestion control or by the protocol layers.
3775 * This function may only be called from softirq context and interrupts
3776 * should be enabled.
3778 * Return values (usually ignored):
3779 * NET_RX_SUCCESS: no congestion
3780 * NET_RX_DROP: packet was dropped
3782 int netif_receive_skb(struct sk_buff
*skb
)
3784 trace_netif_receive_skb_entry(skb
);
3786 return netif_receive_skb_internal(skb
);
3788 EXPORT_SYMBOL(netif_receive_skb
);
3790 /* Network device is going away, flush any packets still pending
3791 * Called with irqs disabled.
3793 static void flush_backlog(void *arg
)
3795 struct net_device
*dev
= arg
;
3796 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3797 struct sk_buff
*skb
, *tmp
;
3800 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3801 if (skb
->dev
== dev
) {
3802 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3804 input_queue_head_incr(sd
);
3809 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3810 if (skb
->dev
== dev
) {
3811 __skb_unlink(skb
, &sd
->process_queue
);
3813 input_queue_head_incr(sd
);
3818 static int napi_gro_complete(struct sk_buff
*skb
)
3820 struct packet_offload
*ptype
;
3821 __be16 type
= skb
->protocol
;
3822 struct list_head
*head
= &offload_base
;
3825 BUILD_BUG_ON(sizeof(struct napi_gro_cb
) > sizeof(skb
->cb
));
3827 if (NAPI_GRO_CB(skb
)->count
== 1) {
3828 skb_shinfo(skb
)->gso_size
= 0;
3833 list_for_each_entry_rcu(ptype
, head
, list
) {
3834 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_complete
)
3837 err
= ptype
->callbacks
.gro_complete(skb
, 0);
3843 WARN_ON(&ptype
->list
== head
);
3845 return NET_RX_SUCCESS
;
3849 return netif_receive_skb_internal(skb
);
3852 /* napi->gro_list contains packets ordered by age.
3853 * youngest packets at the head of it.
3854 * Complete skbs in reverse order to reduce latencies.
3856 void napi_gro_flush(struct napi_struct
*napi
, bool flush_old
)
3858 struct sk_buff
*skb
, *prev
= NULL
;
3860 /* scan list and build reverse chain */
3861 for (skb
= napi
->gro_list
; skb
!= NULL
; skb
= skb
->next
) {
3866 for (skb
= prev
; skb
; skb
= prev
) {
3869 if (flush_old
&& NAPI_GRO_CB(skb
)->age
== jiffies
)
3873 napi_gro_complete(skb
);
3877 napi
->gro_list
= NULL
;
3879 EXPORT_SYMBOL(napi_gro_flush
);
3881 static void gro_list_prepare(struct napi_struct
*napi
, struct sk_buff
*skb
)
3884 unsigned int maclen
= skb
->dev
->hard_header_len
;
3885 u32 hash
= skb_get_hash_raw(skb
);
3887 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3888 unsigned long diffs
;
3890 NAPI_GRO_CB(p
)->flush
= 0;
3892 if (hash
!= skb_get_hash_raw(p
)) {
3893 NAPI_GRO_CB(p
)->same_flow
= 0;
3897 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
3898 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
3899 if (maclen
== ETH_HLEN
)
3900 diffs
|= compare_ether_header(skb_mac_header(p
),
3901 skb_mac_header(skb
));
3903 diffs
= memcmp(skb_mac_header(p
),
3904 skb_mac_header(skb
),
3906 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
3910 static void skb_gro_reset_offset(struct sk_buff
*skb
)
3912 const struct skb_shared_info
*pinfo
= skb_shinfo(skb
);
3913 const skb_frag_t
*frag0
= &pinfo
->frags
[0];
3915 NAPI_GRO_CB(skb
)->data_offset
= 0;
3916 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3917 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3919 if (skb_mac_header(skb
) == skb_tail_pointer(skb
) &&
3921 !PageHighMem(skb_frag_page(frag0
))) {
3922 NAPI_GRO_CB(skb
)->frag0
= skb_frag_address(frag0
);
3923 NAPI_GRO_CB(skb
)->frag0_len
= skb_frag_size(frag0
);
3927 static void gro_pull_from_frag0(struct sk_buff
*skb
, int grow
)
3929 struct skb_shared_info
*pinfo
= skb_shinfo(skb
);
3931 BUG_ON(skb
->end
- skb
->tail
< grow
);
3933 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3935 skb
->data_len
-= grow
;
3938 pinfo
->frags
[0].page_offset
+= grow
;
3939 skb_frag_size_sub(&pinfo
->frags
[0], grow
);
3941 if (unlikely(!skb_frag_size(&pinfo
->frags
[0]))) {
3942 skb_frag_unref(skb
, 0);
3943 memmove(pinfo
->frags
, pinfo
->frags
+ 1,
3944 --pinfo
->nr_frags
* sizeof(pinfo
->frags
[0]));
3948 static enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3950 struct sk_buff
**pp
= NULL
;
3951 struct packet_offload
*ptype
;
3952 __be16 type
= skb
->protocol
;
3953 struct list_head
*head
= &offload_base
;
3955 enum gro_result ret
;
3958 if (!(skb
->dev
->features
& NETIF_F_GRO
))
3961 if (skb_is_gso(skb
) || skb_has_frag_list(skb
))
3964 gro_list_prepare(napi
, skb
);
3966 if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
3967 NAPI_GRO_CB(skb
)->csum
= skb
->csum
;
3968 NAPI_GRO_CB(skb
)->csum_valid
= 1;
3970 NAPI_GRO_CB(skb
)->csum_valid
= 0;
3974 list_for_each_entry_rcu(ptype
, head
, list
) {
3975 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_receive
)
3978 skb_set_network_header(skb
, skb_gro_offset(skb
));
3979 skb_reset_mac_len(skb
);
3980 NAPI_GRO_CB(skb
)->same_flow
= 0;
3981 NAPI_GRO_CB(skb
)->flush
= 0;
3982 NAPI_GRO_CB(skb
)->free
= 0;
3983 NAPI_GRO_CB(skb
)->udp_mark
= 0;
3984 NAPI_GRO_CB(skb
)->encapsulation
= 0;
3986 pp
= ptype
->callbacks
.gro_receive(&napi
->gro_list
, skb
);
3991 if (&ptype
->list
== head
)
3994 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3995 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3998 struct sk_buff
*nskb
= *pp
;
4002 napi_gro_complete(nskb
);
4009 if (NAPI_GRO_CB(skb
)->flush
)
4012 if (unlikely(napi
->gro_count
>= MAX_GRO_SKBS
)) {
4013 struct sk_buff
*nskb
= napi
->gro_list
;
4015 /* locate the end of the list to select the 'oldest' flow */
4016 while (nskb
->next
) {
4022 napi_gro_complete(nskb
);
4026 NAPI_GRO_CB(skb
)->count
= 1;
4027 NAPI_GRO_CB(skb
)->age
= jiffies
;
4028 NAPI_GRO_CB(skb
)->last
= skb
;
4029 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
4030 skb
->next
= napi
->gro_list
;
4031 napi
->gro_list
= skb
;
4035 grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
4037 gro_pull_from_frag0(skb
, grow
);
4046 struct packet_offload
*gro_find_receive_by_type(__be16 type
)
4048 struct list_head
*offload_head
= &offload_base
;
4049 struct packet_offload
*ptype
;
4051 list_for_each_entry_rcu(ptype
, offload_head
, list
) {
4052 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_receive
)
4058 EXPORT_SYMBOL(gro_find_receive_by_type
);
4060 struct packet_offload
*gro_find_complete_by_type(__be16 type
)
4062 struct list_head
*offload_head
= &offload_base
;
4063 struct packet_offload
*ptype
;
4065 list_for_each_entry_rcu(ptype
, offload_head
, list
) {
4066 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_complete
)
4072 EXPORT_SYMBOL(gro_find_complete_by_type
);
4074 static gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
4078 if (netif_receive_skb_internal(skb
))
4086 case GRO_MERGED_FREE
:
4087 if (NAPI_GRO_CB(skb
)->free
== NAPI_GRO_FREE_STOLEN_HEAD
)
4088 kmem_cache_free(skbuff_head_cache
, skb
);
4101 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
4103 trace_napi_gro_receive_entry(skb
);
4105 skb_gro_reset_offset(skb
);
4107 return napi_skb_finish(dev_gro_receive(napi
, skb
), skb
);
4109 EXPORT_SYMBOL(napi_gro_receive
);
4111 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
4113 __skb_pull(skb
, skb_headlen(skb
));
4114 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4115 skb_reserve(skb
, NET_SKB_PAD
+ NET_IP_ALIGN
- skb_headroom(skb
));
4117 skb
->dev
= napi
->dev
;
4119 skb
->encapsulation
= 0;
4120 skb_shinfo(skb
)->gso_type
= 0;
4121 skb
->truesize
= SKB_TRUESIZE(skb_end_offset(skb
));
4126 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
4128 struct sk_buff
*skb
= napi
->skb
;
4131 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
4136 EXPORT_SYMBOL(napi_get_frags
);
4138 static gro_result_t
napi_frags_finish(struct napi_struct
*napi
,
4139 struct sk_buff
*skb
,
4145 __skb_push(skb
, ETH_HLEN
);
4146 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
4147 if (ret
== GRO_NORMAL
&& netif_receive_skb_internal(skb
))
4152 case GRO_MERGED_FREE
:
4153 napi_reuse_skb(napi
, skb
);
4163 /* Upper GRO stack assumes network header starts at gro_offset=0
4164 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4165 * We copy ethernet header into skb->data to have a common layout.
4167 static struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
4169 struct sk_buff
*skb
= napi
->skb
;
4170 const struct ethhdr
*eth
;
4171 unsigned int hlen
= sizeof(*eth
);
4175 skb_reset_mac_header(skb
);
4176 skb_gro_reset_offset(skb
);
4178 eth
= skb_gro_header_fast(skb
, 0);
4179 if (unlikely(skb_gro_header_hard(skb
, hlen
))) {
4180 eth
= skb_gro_header_slow(skb
, hlen
, 0);
4181 if (unlikely(!eth
)) {
4182 napi_reuse_skb(napi
, skb
);
4186 gro_pull_from_frag0(skb
, hlen
);
4187 NAPI_GRO_CB(skb
)->frag0
+= hlen
;
4188 NAPI_GRO_CB(skb
)->frag0_len
-= hlen
;
4190 __skb_pull(skb
, hlen
);
4193 * This works because the only protocols we care about don't require
4195 * We'll fix it up properly in napi_frags_finish()
4197 skb
->protocol
= eth
->h_proto
;
4202 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
4204 struct sk_buff
*skb
= napi_frags_skb(napi
);
4209 trace_napi_gro_frags_entry(skb
);
4211 return napi_frags_finish(napi
, skb
, dev_gro_receive(napi
, skb
));
4213 EXPORT_SYMBOL(napi_gro_frags
);
4215 /* Compute the checksum from gro_offset and return the folded value
4216 * after adding in any pseudo checksum.
4218 __sum16
__skb_gro_checksum_complete(struct sk_buff
*skb
)
4223 wsum
= skb_checksum(skb
, skb_gro_offset(skb
), skb_gro_len(skb
), 0);
4225 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4226 sum
= csum_fold(csum_add(NAPI_GRO_CB(skb
)->csum
, wsum
));
4228 if (unlikely(skb
->ip_summed
== CHECKSUM_COMPLETE
) &&
4229 !skb
->csum_complete_sw
)
4230 netdev_rx_csum_fault(skb
->dev
);
4233 NAPI_GRO_CB(skb
)->csum
= wsum
;
4234 NAPI_GRO_CB(skb
)->csum_valid
= 1;
4238 EXPORT_SYMBOL(__skb_gro_checksum_complete
);
4241 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4242 * Note: called with local irq disabled, but exits with local irq enabled.
4244 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
4247 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
4250 sd
->rps_ipi_list
= NULL
;
4254 /* Send pending IPI's to kick RPS processing on remote cpus. */
4256 struct softnet_data
*next
= remsd
->rps_ipi_next
;
4258 if (cpu_online(remsd
->cpu
))
4259 smp_call_function_single_async(remsd
->cpu
,
4268 static int process_backlog(struct napi_struct
*napi
, int quota
)
4271 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
4274 /* Check if we have pending ipi, its better to send them now,
4275 * not waiting net_rx_action() end.
4277 if (sd
->rps_ipi_list
) {
4278 local_irq_disable();
4279 net_rps_action_and_irq_enable(sd
);
4282 napi
->weight
= weight_p
;
4283 local_irq_disable();
4285 struct sk_buff
*skb
;
4287 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
4289 __netif_receive_skb(skb
);
4290 local_irq_disable();
4291 input_queue_head_incr(sd
);
4292 if (++work
>= quota
) {
4299 if (skb_queue_empty(&sd
->input_pkt_queue
)) {
4301 * Inline a custom version of __napi_complete().
4302 * only current cpu owns and manipulates this napi,
4303 * and NAPI_STATE_SCHED is the only possible flag set
4305 * We can use a plain write instead of clear_bit(),
4306 * and we dont need an smp_mb() memory barrier.
4308 list_del(&napi
->poll_list
);
4315 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
4316 &sd
->process_queue
);
4325 * __napi_schedule - schedule for receive
4326 * @n: entry to schedule
4328 * The entry's receive function will be scheduled to run
4330 void __napi_schedule(struct napi_struct
*n
)
4332 unsigned long flags
;
4334 local_irq_save(flags
);
4335 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
4336 local_irq_restore(flags
);
4338 EXPORT_SYMBOL(__napi_schedule
);
4340 void __napi_complete(struct napi_struct
*n
)
4342 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
4343 BUG_ON(n
->gro_list
);
4345 list_del(&n
->poll_list
);
4346 smp_mb__before_atomic();
4347 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
4349 EXPORT_SYMBOL(__napi_complete
);
4351 void napi_complete(struct napi_struct
*n
)
4353 unsigned long flags
;
4356 * don't let napi dequeue from the cpu poll list
4357 * just in case its running on a different cpu
4359 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
4362 napi_gro_flush(n
, false);
4363 local_irq_save(flags
);
4365 local_irq_restore(flags
);
4367 EXPORT_SYMBOL(napi_complete
);
4369 /* must be called under rcu_read_lock(), as we dont take a reference */
4370 struct napi_struct
*napi_by_id(unsigned int napi_id
)
4372 unsigned int hash
= napi_id
% HASH_SIZE(napi_hash
);
4373 struct napi_struct
*napi
;
4375 hlist_for_each_entry_rcu(napi
, &napi_hash
[hash
], napi_hash_node
)
4376 if (napi
->napi_id
== napi_id
)
4381 EXPORT_SYMBOL_GPL(napi_by_id
);
4383 void napi_hash_add(struct napi_struct
*napi
)
4385 if (!test_and_set_bit(NAPI_STATE_HASHED
, &napi
->state
)) {
4387 spin_lock(&napi_hash_lock
);
4389 /* 0 is not a valid id, we also skip an id that is taken
4390 * we expect both events to be extremely rare
4393 while (!napi
->napi_id
) {
4394 napi
->napi_id
= ++napi_gen_id
;
4395 if (napi_by_id(napi
->napi_id
))
4399 hlist_add_head_rcu(&napi
->napi_hash_node
,
4400 &napi_hash
[napi
->napi_id
% HASH_SIZE(napi_hash
)]);
4402 spin_unlock(&napi_hash_lock
);
4405 EXPORT_SYMBOL_GPL(napi_hash_add
);
4407 /* Warning : caller is responsible to make sure rcu grace period
4408 * is respected before freeing memory containing @napi
4410 void napi_hash_del(struct napi_struct
*napi
)
4412 spin_lock(&napi_hash_lock
);
4414 if (test_and_clear_bit(NAPI_STATE_HASHED
, &napi
->state
))
4415 hlist_del_rcu(&napi
->napi_hash_node
);
4417 spin_unlock(&napi_hash_lock
);
4419 EXPORT_SYMBOL_GPL(napi_hash_del
);
4421 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
4422 int (*poll
)(struct napi_struct
*, int), int weight
)
4424 INIT_LIST_HEAD(&napi
->poll_list
);
4425 napi
->gro_count
= 0;
4426 napi
->gro_list
= NULL
;
4429 if (weight
> NAPI_POLL_WEIGHT
)
4430 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4432 napi
->weight
= weight
;
4433 list_add(&napi
->dev_list
, &dev
->napi_list
);
4435 #ifdef CONFIG_NETPOLL
4436 spin_lock_init(&napi
->poll_lock
);
4437 napi
->poll_owner
= -1;
4439 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
4441 EXPORT_SYMBOL(netif_napi_add
);
4443 void netif_napi_del(struct napi_struct
*napi
)
4445 list_del_init(&napi
->dev_list
);
4446 napi_free_frags(napi
);
4448 kfree_skb_list(napi
->gro_list
);
4449 napi
->gro_list
= NULL
;
4450 napi
->gro_count
= 0;
4452 EXPORT_SYMBOL(netif_napi_del
);
4454 static void net_rx_action(struct softirq_action
*h
)
4456 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
4457 unsigned long time_limit
= jiffies
+ 2;
4458 int budget
= netdev_budget
;
4461 local_irq_disable();
4463 while (!list_empty(&sd
->poll_list
)) {
4464 struct napi_struct
*n
;
4467 /* If softirq window is exhuasted then punt.
4468 * Allow this to run for 2 jiffies since which will allow
4469 * an average latency of 1.5/HZ.
4471 if (unlikely(budget
<= 0 || time_after_eq(jiffies
, time_limit
)))
4476 /* Even though interrupts have been re-enabled, this
4477 * access is safe because interrupts can only add new
4478 * entries to the tail of this list, and only ->poll()
4479 * calls can remove this head entry from the list.
4481 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
4483 have
= netpoll_poll_lock(n
);
4487 /* This NAPI_STATE_SCHED test is for avoiding a race
4488 * with netpoll's poll_napi(). Only the entity which
4489 * obtains the lock and sees NAPI_STATE_SCHED set will
4490 * actually make the ->poll() call. Therefore we avoid
4491 * accidentally calling ->poll() when NAPI is not scheduled.
4494 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
4495 work
= n
->poll(n
, weight
);
4499 WARN_ON_ONCE(work
> weight
);
4503 local_irq_disable();
4505 /* Drivers must not modify the NAPI state if they
4506 * consume the entire weight. In such cases this code
4507 * still "owns" the NAPI instance and therefore can
4508 * move the instance around on the list at-will.
4510 if (unlikely(work
== weight
)) {
4511 if (unlikely(napi_disable_pending(n
))) {
4514 local_irq_disable();
4517 /* flush too old packets
4518 * If HZ < 1000, flush all packets.
4521 napi_gro_flush(n
, HZ
>= 1000);
4522 local_irq_disable();
4524 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
4528 netpoll_poll_unlock(have
);
4531 net_rps_action_and_irq_enable(sd
);
4533 #ifdef CONFIG_NET_DMA
4535 * There may not be any more sk_buffs coming right now, so push
4536 * any pending DMA copies to hardware
4538 dma_issue_pending_all();
4545 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
4549 struct netdev_adjacent
{
4550 struct net_device
*dev
;
4552 /* upper master flag, there can only be one master device per list */
4555 /* counter for the number of times this device was added to us */
4558 /* private field for the users */
4561 struct list_head list
;
4562 struct rcu_head rcu
;
4565 static struct netdev_adjacent
*__netdev_find_adj(struct net_device
*dev
,
4566 struct net_device
*adj_dev
,
4567 struct list_head
*adj_list
)
4569 struct netdev_adjacent
*adj
;
4571 list_for_each_entry(adj
, adj_list
, list
) {
4572 if (adj
->dev
== adj_dev
)
4579 * netdev_has_upper_dev - Check if device is linked to an upper device
4581 * @upper_dev: upper device to check
4583 * Find out if a device is linked to specified upper device and return true
4584 * in case it is. Note that this checks only immediate upper device,
4585 * not through a complete stack of devices. The caller must hold the RTNL lock.
4587 bool netdev_has_upper_dev(struct net_device
*dev
,
4588 struct net_device
*upper_dev
)
4592 return __netdev_find_adj(dev
, upper_dev
, &dev
->all_adj_list
.upper
);
4594 EXPORT_SYMBOL(netdev_has_upper_dev
);
4597 * netdev_has_any_upper_dev - Check if device is linked to some device
4600 * Find out if a device is linked to an upper device and return true in case
4601 * it is. The caller must hold the RTNL lock.
4603 static bool netdev_has_any_upper_dev(struct net_device
*dev
)
4607 return !list_empty(&dev
->all_adj_list
.upper
);
4611 * netdev_master_upper_dev_get - Get master upper device
4614 * Find a master upper device and return pointer to it or NULL in case
4615 * it's not there. The caller must hold the RTNL lock.
4617 struct net_device
*netdev_master_upper_dev_get(struct net_device
*dev
)
4619 struct netdev_adjacent
*upper
;
4623 if (list_empty(&dev
->adj_list
.upper
))
4626 upper
= list_first_entry(&dev
->adj_list
.upper
,
4627 struct netdev_adjacent
, list
);
4628 if (likely(upper
->master
))
4632 EXPORT_SYMBOL(netdev_master_upper_dev_get
);
4634 void *netdev_adjacent_get_private(struct list_head
*adj_list
)
4636 struct netdev_adjacent
*adj
;
4638 adj
= list_entry(adj_list
, struct netdev_adjacent
, list
);
4640 return adj
->private;
4642 EXPORT_SYMBOL(netdev_adjacent_get_private
);
4645 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
4647 * @iter: list_head ** of the current position
4649 * Gets the next device from the dev's upper list, starting from iter
4650 * position. The caller must hold RCU read lock.
4652 struct net_device
*netdev_upper_get_next_dev_rcu(struct net_device
*dev
,
4653 struct list_head
**iter
)
4655 struct netdev_adjacent
*upper
;
4657 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4659 upper
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
4661 if (&upper
->list
== &dev
->adj_list
.upper
)
4664 *iter
= &upper
->list
;
4668 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu
);
4671 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4673 * @iter: list_head ** of the current position
4675 * Gets the next device from the dev's upper list, starting from iter
4676 * position. The caller must hold RCU read lock.
4678 struct net_device
*netdev_all_upper_get_next_dev_rcu(struct net_device
*dev
,
4679 struct list_head
**iter
)
4681 struct netdev_adjacent
*upper
;
4683 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4685 upper
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
4687 if (&upper
->list
== &dev
->all_adj_list
.upper
)
4690 *iter
= &upper
->list
;
4694 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu
);
4697 * netdev_lower_get_next_private - Get the next ->private from the
4698 * lower neighbour list
4700 * @iter: list_head ** of the current position
4702 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4703 * list, starting from iter position. The caller must hold either hold the
4704 * RTNL lock or its own locking that guarantees that the neighbour lower
4705 * list will remain unchainged.
4707 void *netdev_lower_get_next_private(struct net_device
*dev
,
4708 struct list_head
**iter
)
4710 struct netdev_adjacent
*lower
;
4712 lower
= list_entry(*iter
, struct netdev_adjacent
, list
);
4714 if (&lower
->list
== &dev
->adj_list
.lower
)
4717 *iter
= lower
->list
.next
;
4719 return lower
->private;
4721 EXPORT_SYMBOL(netdev_lower_get_next_private
);
4724 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4725 * lower neighbour list, RCU
4728 * @iter: list_head ** of the current position
4730 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4731 * list, starting from iter position. The caller must hold RCU read lock.
4733 void *netdev_lower_get_next_private_rcu(struct net_device
*dev
,
4734 struct list_head
**iter
)
4736 struct netdev_adjacent
*lower
;
4738 WARN_ON_ONCE(!rcu_read_lock_held());
4740 lower
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
4742 if (&lower
->list
== &dev
->adj_list
.lower
)
4745 *iter
= &lower
->list
;
4747 return lower
->private;
4749 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu
);
4752 * netdev_lower_get_next - Get the next device from the lower neighbour
4755 * @iter: list_head ** of the current position
4757 * Gets the next netdev_adjacent from the dev's lower neighbour
4758 * list, starting from iter position. The caller must hold RTNL lock or
4759 * its own locking that guarantees that the neighbour lower
4760 * list will remain unchainged.
4762 void *netdev_lower_get_next(struct net_device
*dev
, struct list_head
**iter
)
4764 struct netdev_adjacent
*lower
;
4766 lower
= list_entry((*iter
)->next
, struct netdev_adjacent
, list
);
4768 if (&lower
->list
== &dev
->adj_list
.lower
)
4771 *iter
= &lower
->list
;
4775 EXPORT_SYMBOL(netdev_lower_get_next
);
4778 * netdev_lower_get_first_private_rcu - Get the first ->private from the
4779 * lower neighbour list, RCU
4783 * Gets the first netdev_adjacent->private from the dev's lower neighbour
4784 * list. The caller must hold RCU read lock.
4786 void *netdev_lower_get_first_private_rcu(struct net_device
*dev
)
4788 struct netdev_adjacent
*lower
;
4790 lower
= list_first_or_null_rcu(&dev
->adj_list
.lower
,
4791 struct netdev_adjacent
, list
);
4793 return lower
->private;
4796 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu
);
4799 * netdev_master_upper_dev_get_rcu - Get master upper device
4802 * Find a master upper device and return pointer to it or NULL in case
4803 * it's not there. The caller must hold the RCU read lock.
4805 struct net_device
*netdev_master_upper_dev_get_rcu(struct net_device
*dev
)
4807 struct netdev_adjacent
*upper
;
4809 upper
= list_first_or_null_rcu(&dev
->adj_list
.upper
,
4810 struct netdev_adjacent
, list
);
4811 if (upper
&& likely(upper
->master
))
4815 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu
);
4817 static int netdev_adjacent_sysfs_add(struct net_device
*dev
,
4818 struct net_device
*adj_dev
,
4819 struct list_head
*dev_list
)
4821 char linkname
[IFNAMSIZ
+7];
4822 sprintf(linkname
, dev_list
== &dev
->adj_list
.upper
?
4823 "upper_%s" : "lower_%s", adj_dev
->name
);
4824 return sysfs_create_link(&(dev
->dev
.kobj
), &(adj_dev
->dev
.kobj
),
4827 static void netdev_adjacent_sysfs_del(struct net_device
*dev
,
4829 struct list_head
*dev_list
)
4831 char linkname
[IFNAMSIZ
+7];
4832 sprintf(linkname
, dev_list
== &dev
->adj_list
.upper
?
4833 "upper_%s" : "lower_%s", name
);
4834 sysfs_remove_link(&(dev
->dev
.kobj
), linkname
);
4837 #define netdev_adjacent_is_neigh_list(dev, dev_list) \
4838 (dev_list == &dev->adj_list.upper || \
4839 dev_list == &dev->adj_list.lower)
4841 static int __netdev_adjacent_dev_insert(struct net_device
*dev
,
4842 struct net_device
*adj_dev
,
4843 struct list_head
*dev_list
,
4844 void *private, bool master
)
4846 struct netdev_adjacent
*adj
;
4849 adj
= __netdev_find_adj(dev
, adj_dev
, dev_list
);
4856 adj
= kmalloc(sizeof(*adj
), GFP_KERNEL
);
4861 adj
->master
= master
;
4863 adj
->private = private;
4866 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
4867 adj_dev
->name
, dev
->name
, adj_dev
->name
);
4869 if (netdev_adjacent_is_neigh_list(dev
, dev_list
)) {
4870 ret
= netdev_adjacent_sysfs_add(dev
, adj_dev
, dev_list
);
4875 /* Ensure that master link is always the first item in list. */
4877 ret
= sysfs_create_link(&(dev
->dev
.kobj
),
4878 &(adj_dev
->dev
.kobj
), "master");
4880 goto remove_symlinks
;
4882 list_add_rcu(&adj
->list
, dev_list
);
4884 list_add_tail_rcu(&adj
->list
, dev_list
);
4890 if (netdev_adjacent_is_neigh_list(dev
, dev_list
))
4891 netdev_adjacent_sysfs_del(dev
, adj_dev
->name
, dev_list
);
4899 static void __netdev_adjacent_dev_remove(struct net_device
*dev
,
4900 struct net_device
*adj_dev
,
4901 struct list_head
*dev_list
)
4903 struct netdev_adjacent
*adj
;
4905 adj
= __netdev_find_adj(dev
, adj_dev
, dev_list
);
4908 pr_err("tried to remove device %s from %s\n",
4909 dev
->name
, adj_dev
->name
);
4913 if (adj
->ref_nr
> 1) {
4914 pr_debug("%s to %s ref_nr-- = %d\n", dev
->name
, adj_dev
->name
,
4921 sysfs_remove_link(&(dev
->dev
.kobj
), "master");
4923 if (netdev_adjacent_is_neigh_list(dev
, dev_list
))
4924 netdev_adjacent_sysfs_del(dev
, adj_dev
->name
, dev_list
);
4926 list_del_rcu(&adj
->list
);
4927 pr_debug("dev_put for %s, because link removed from %s to %s\n",
4928 adj_dev
->name
, dev
->name
, adj_dev
->name
);
4930 kfree_rcu(adj
, rcu
);
4933 static int __netdev_adjacent_dev_link_lists(struct net_device
*dev
,
4934 struct net_device
*upper_dev
,
4935 struct list_head
*up_list
,
4936 struct list_head
*down_list
,
4937 void *private, bool master
)
4941 ret
= __netdev_adjacent_dev_insert(dev
, upper_dev
, up_list
, private,
4946 ret
= __netdev_adjacent_dev_insert(upper_dev
, dev
, down_list
, private,
4949 __netdev_adjacent_dev_remove(dev
, upper_dev
, up_list
);
4956 static int __netdev_adjacent_dev_link(struct net_device
*dev
,
4957 struct net_device
*upper_dev
)
4959 return __netdev_adjacent_dev_link_lists(dev
, upper_dev
,
4960 &dev
->all_adj_list
.upper
,
4961 &upper_dev
->all_adj_list
.lower
,
4965 static void __netdev_adjacent_dev_unlink_lists(struct net_device
*dev
,
4966 struct net_device
*upper_dev
,
4967 struct list_head
*up_list
,
4968 struct list_head
*down_list
)
4970 __netdev_adjacent_dev_remove(dev
, upper_dev
, up_list
);
4971 __netdev_adjacent_dev_remove(upper_dev
, dev
, down_list
);
4974 static void __netdev_adjacent_dev_unlink(struct net_device
*dev
,
4975 struct net_device
*upper_dev
)
4977 __netdev_adjacent_dev_unlink_lists(dev
, upper_dev
,
4978 &dev
->all_adj_list
.upper
,
4979 &upper_dev
->all_adj_list
.lower
);
4982 static int __netdev_adjacent_dev_link_neighbour(struct net_device
*dev
,
4983 struct net_device
*upper_dev
,
4984 void *private, bool master
)
4986 int ret
= __netdev_adjacent_dev_link(dev
, upper_dev
);
4991 ret
= __netdev_adjacent_dev_link_lists(dev
, upper_dev
,
4992 &dev
->adj_list
.upper
,
4993 &upper_dev
->adj_list
.lower
,
4996 __netdev_adjacent_dev_unlink(dev
, upper_dev
);
5003 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device
*dev
,
5004 struct net_device
*upper_dev
)
5006 __netdev_adjacent_dev_unlink(dev
, upper_dev
);
5007 __netdev_adjacent_dev_unlink_lists(dev
, upper_dev
,
5008 &dev
->adj_list
.upper
,
5009 &upper_dev
->adj_list
.lower
);
5012 static int __netdev_upper_dev_link(struct net_device
*dev
,
5013 struct net_device
*upper_dev
, bool master
,
5016 struct netdev_adjacent
*i
, *j
, *to_i
, *to_j
;
5021 if (dev
== upper_dev
)
5024 /* To prevent loops, check if dev is not upper device to upper_dev. */
5025 if (__netdev_find_adj(upper_dev
, dev
, &upper_dev
->all_adj_list
.upper
))
5028 if (__netdev_find_adj(dev
, upper_dev
, &dev
->all_adj_list
.upper
))
5031 if (master
&& netdev_master_upper_dev_get(dev
))
5034 ret
= __netdev_adjacent_dev_link_neighbour(dev
, upper_dev
, private,
5039 /* Now that we linked these devs, make all the upper_dev's
5040 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5041 * versa, and don't forget the devices itself. All of these
5042 * links are non-neighbours.
5044 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5045 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
) {
5046 pr_debug("Interlinking %s with %s, non-neighbour\n",
5047 i
->dev
->name
, j
->dev
->name
);
5048 ret
= __netdev_adjacent_dev_link(i
->dev
, j
->dev
);
5054 /* add dev to every upper_dev's upper device */
5055 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
) {
5056 pr_debug("linking %s's upper device %s with %s\n",
5057 upper_dev
->name
, i
->dev
->name
, dev
->name
);
5058 ret
= __netdev_adjacent_dev_link(dev
, i
->dev
);
5060 goto rollback_upper_mesh
;
5063 /* add upper_dev to every dev's lower device */
5064 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5065 pr_debug("linking %s's lower device %s with %s\n", dev
->name
,
5066 i
->dev
->name
, upper_dev
->name
);
5067 ret
= __netdev_adjacent_dev_link(i
->dev
, upper_dev
);
5069 goto rollback_lower_mesh
;
5072 call_netdevice_notifiers(NETDEV_CHANGEUPPER
, dev
);
5075 rollback_lower_mesh
:
5077 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5080 __netdev_adjacent_dev_unlink(i
->dev
, upper_dev
);
5085 rollback_upper_mesh
:
5087 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
) {
5090 __netdev_adjacent_dev_unlink(dev
, i
->dev
);
5098 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5099 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
) {
5100 if (i
== to_i
&& j
== to_j
)
5102 __netdev_adjacent_dev_unlink(i
->dev
, j
->dev
);
5108 __netdev_adjacent_dev_unlink_neighbour(dev
, upper_dev
);
5114 * netdev_upper_dev_link - Add a link to the upper device
5116 * @upper_dev: new upper device
5118 * Adds a link to device which is upper to this one. The caller must hold
5119 * the RTNL lock. On a failure a negative errno code is returned.
5120 * On success the reference counts are adjusted and the function
5123 int netdev_upper_dev_link(struct net_device
*dev
,
5124 struct net_device
*upper_dev
)
5126 return __netdev_upper_dev_link(dev
, upper_dev
, false, NULL
);
5128 EXPORT_SYMBOL(netdev_upper_dev_link
);
5131 * netdev_master_upper_dev_link - Add a master link to the upper device
5133 * @upper_dev: new upper device
5135 * Adds a link to device which is upper to this one. In this case, only
5136 * one master upper device can be linked, although other non-master devices
5137 * might be linked as well. The caller must hold the RTNL lock.
5138 * On a failure a negative errno code is returned. On success the reference
5139 * counts are adjusted and the function returns zero.
5141 int netdev_master_upper_dev_link(struct net_device
*dev
,
5142 struct net_device
*upper_dev
)
5144 return __netdev_upper_dev_link(dev
, upper_dev
, true, NULL
);
5146 EXPORT_SYMBOL(netdev_master_upper_dev_link
);
5148 int netdev_master_upper_dev_link_private(struct net_device
*dev
,
5149 struct net_device
*upper_dev
,
5152 return __netdev_upper_dev_link(dev
, upper_dev
, true, private);
5154 EXPORT_SYMBOL(netdev_master_upper_dev_link_private
);
5157 * netdev_upper_dev_unlink - Removes a link to upper device
5159 * @upper_dev: new upper device
5161 * Removes a link to device which is upper to this one. The caller must hold
5164 void netdev_upper_dev_unlink(struct net_device
*dev
,
5165 struct net_device
*upper_dev
)
5167 struct netdev_adjacent
*i
, *j
;
5170 __netdev_adjacent_dev_unlink_neighbour(dev
, upper_dev
);
5172 /* Here is the tricky part. We must remove all dev's lower
5173 * devices from all upper_dev's upper devices and vice
5174 * versa, to maintain the graph relationship.
5176 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
)
5177 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
)
5178 __netdev_adjacent_dev_unlink(i
->dev
, j
->dev
);
5180 /* remove also the devices itself from lower/upper device
5183 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
)
5184 __netdev_adjacent_dev_unlink(i
->dev
, upper_dev
);
5186 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
)
5187 __netdev_adjacent_dev_unlink(dev
, i
->dev
);
5189 call_netdevice_notifiers(NETDEV_CHANGEUPPER
, dev
);
5191 EXPORT_SYMBOL(netdev_upper_dev_unlink
);
5193 void netdev_adjacent_rename_links(struct net_device
*dev
, char *oldname
)
5195 struct netdev_adjacent
*iter
;
5197 list_for_each_entry(iter
, &dev
->adj_list
.upper
, list
) {
5198 netdev_adjacent_sysfs_del(iter
->dev
, oldname
,
5199 &iter
->dev
->adj_list
.lower
);
5200 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
5201 &iter
->dev
->adj_list
.lower
);
5204 list_for_each_entry(iter
, &dev
->adj_list
.lower
, list
) {
5205 netdev_adjacent_sysfs_del(iter
->dev
, oldname
,
5206 &iter
->dev
->adj_list
.upper
);
5207 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
5208 &iter
->dev
->adj_list
.upper
);
5212 void *netdev_lower_dev_get_private(struct net_device
*dev
,
5213 struct net_device
*lower_dev
)
5215 struct netdev_adjacent
*lower
;
5219 lower
= __netdev_find_adj(dev
, lower_dev
, &dev
->adj_list
.lower
);
5223 return lower
->private;
5225 EXPORT_SYMBOL(netdev_lower_dev_get_private
);
5228 int dev_get_nest_level(struct net_device
*dev
,
5229 bool (*type_check
)(struct net_device
*dev
))
5231 struct net_device
*lower
= NULL
;
5232 struct list_head
*iter
;
5238 netdev_for_each_lower_dev(dev
, lower
, iter
) {
5239 nest
= dev_get_nest_level(lower
, type_check
);
5240 if (max_nest
< nest
)
5244 if (type_check(dev
))
5249 EXPORT_SYMBOL(dev_get_nest_level
);
5251 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
5253 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5255 if (ops
->ndo_change_rx_flags
)
5256 ops
->ndo_change_rx_flags(dev
, flags
);
5259 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
, bool notify
)
5261 unsigned int old_flags
= dev
->flags
;
5267 dev
->flags
|= IFF_PROMISC
;
5268 dev
->promiscuity
+= inc
;
5269 if (dev
->promiscuity
== 0) {
5272 * If inc causes overflow, untouch promisc and return error.
5275 dev
->flags
&= ~IFF_PROMISC
;
5277 dev
->promiscuity
-= inc
;
5278 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5283 if (dev
->flags
!= old_flags
) {
5284 pr_info("device %s %s promiscuous mode\n",
5286 dev
->flags
& IFF_PROMISC
? "entered" : "left");
5287 if (audit_enabled
) {
5288 current_uid_gid(&uid
, &gid
);
5289 audit_log(current
->audit_context
, GFP_ATOMIC
,
5290 AUDIT_ANOM_PROMISCUOUS
,
5291 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5292 dev
->name
, (dev
->flags
& IFF_PROMISC
),
5293 (old_flags
& IFF_PROMISC
),
5294 from_kuid(&init_user_ns
, audit_get_loginuid(current
)),
5295 from_kuid(&init_user_ns
, uid
),
5296 from_kgid(&init_user_ns
, gid
),
5297 audit_get_sessionid(current
));
5300 dev_change_rx_flags(dev
, IFF_PROMISC
);
5303 __dev_notify_flags(dev
, old_flags
, IFF_PROMISC
);
5308 * dev_set_promiscuity - update promiscuity count on a device
5312 * Add or remove promiscuity from a device. While the count in the device
5313 * remains above zero the interface remains promiscuous. Once it hits zero
5314 * the device reverts back to normal filtering operation. A negative inc
5315 * value is used to drop promiscuity on the device.
5316 * Return 0 if successful or a negative errno code on error.
5318 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
5320 unsigned int old_flags
= dev
->flags
;
5323 err
= __dev_set_promiscuity(dev
, inc
, true);
5326 if (dev
->flags
!= old_flags
)
5327 dev_set_rx_mode(dev
);
5330 EXPORT_SYMBOL(dev_set_promiscuity
);
5332 static int __dev_set_allmulti(struct net_device
*dev
, int inc
, bool notify
)
5334 unsigned int old_flags
= dev
->flags
, old_gflags
= dev
->gflags
;
5338 dev
->flags
|= IFF_ALLMULTI
;
5339 dev
->allmulti
+= inc
;
5340 if (dev
->allmulti
== 0) {
5343 * If inc causes overflow, untouch allmulti and return error.
5346 dev
->flags
&= ~IFF_ALLMULTI
;
5348 dev
->allmulti
-= inc
;
5349 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5354 if (dev
->flags
^ old_flags
) {
5355 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
5356 dev_set_rx_mode(dev
);
5358 __dev_notify_flags(dev
, old_flags
,
5359 dev
->gflags
^ old_gflags
);
5365 * dev_set_allmulti - update allmulti count on a device
5369 * Add or remove reception of all multicast frames to a device. While the
5370 * count in the device remains above zero the interface remains listening
5371 * to all interfaces. Once it hits zero the device reverts back to normal
5372 * filtering operation. A negative @inc value is used to drop the counter
5373 * when releasing a resource needing all multicasts.
5374 * Return 0 if successful or a negative errno code on error.
5377 int dev_set_allmulti(struct net_device
*dev
, int inc
)
5379 return __dev_set_allmulti(dev
, inc
, true);
5381 EXPORT_SYMBOL(dev_set_allmulti
);
5384 * Upload unicast and multicast address lists to device and
5385 * configure RX filtering. When the device doesn't support unicast
5386 * filtering it is put in promiscuous mode while unicast addresses
5389 void __dev_set_rx_mode(struct net_device
*dev
)
5391 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5393 /* dev_open will call this function so the list will stay sane. */
5394 if (!(dev
->flags
&IFF_UP
))
5397 if (!netif_device_present(dev
))
5400 if (!(dev
->priv_flags
& IFF_UNICAST_FLT
)) {
5401 /* Unicast addresses changes may only happen under the rtnl,
5402 * therefore calling __dev_set_promiscuity here is safe.
5404 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
5405 __dev_set_promiscuity(dev
, 1, false);
5406 dev
->uc_promisc
= true;
5407 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
5408 __dev_set_promiscuity(dev
, -1, false);
5409 dev
->uc_promisc
= false;
5413 if (ops
->ndo_set_rx_mode
)
5414 ops
->ndo_set_rx_mode(dev
);
5417 void dev_set_rx_mode(struct net_device
*dev
)
5419 netif_addr_lock_bh(dev
);
5420 __dev_set_rx_mode(dev
);
5421 netif_addr_unlock_bh(dev
);
5425 * dev_get_flags - get flags reported to userspace
5428 * Get the combination of flag bits exported through APIs to userspace.
5430 unsigned int dev_get_flags(const struct net_device
*dev
)
5434 flags
= (dev
->flags
& ~(IFF_PROMISC
|
5439 (dev
->gflags
& (IFF_PROMISC
|
5442 if (netif_running(dev
)) {
5443 if (netif_oper_up(dev
))
5444 flags
|= IFF_RUNNING
;
5445 if (netif_carrier_ok(dev
))
5446 flags
|= IFF_LOWER_UP
;
5447 if (netif_dormant(dev
))
5448 flags
|= IFF_DORMANT
;
5453 EXPORT_SYMBOL(dev_get_flags
);
5455 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
5457 unsigned int old_flags
= dev
->flags
;
5463 * Set the flags on our device.
5466 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
5467 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
5469 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
5473 * Load in the correct multicast list now the flags have changed.
5476 if ((old_flags
^ flags
) & IFF_MULTICAST
)
5477 dev_change_rx_flags(dev
, IFF_MULTICAST
);
5479 dev_set_rx_mode(dev
);
5482 * Have we downed the interface. We handle IFF_UP ourselves
5483 * according to user attempts to set it, rather than blindly
5488 if ((old_flags
^ flags
) & IFF_UP
)
5489 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
5491 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
5492 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
5493 unsigned int old_flags
= dev
->flags
;
5495 dev
->gflags
^= IFF_PROMISC
;
5497 if (__dev_set_promiscuity(dev
, inc
, false) >= 0)
5498 if (dev
->flags
!= old_flags
)
5499 dev_set_rx_mode(dev
);
5502 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5503 is important. Some (broken) drivers set IFF_PROMISC, when
5504 IFF_ALLMULTI is requested not asking us and not reporting.
5506 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
5507 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
5509 dev
->gflags
^= IFF_ALLMULTI
;
5510 __dev_set_allmulti(dev
, inc
, false);
5516 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
,
5517 unsigned int gchanges
)
5519 unsigned int changes
= dev
->flags
^ old_flags
;
5522 rtmsg_ifinfo(RTM_NEWLINK
, dev
, gchanges
, GFP_ATOMIC
);
5524 if (changes
& IFF_UP
) {
5525 if (dev
->flags
& IFF_UP
)
5526 call_netdevice_notifiers(NETDEV_UP
, dev
);
5528 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
5531 if (dev
->flags
& IFF_UP
&&
5532 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
))) {
5533 struct netdev_notifier_change_info change_info
;
5535 change_info
.flags_changed
= changes
;
5536 call_netdevice_notifiers_info(NETDEV_CHANGE
, dev
,
5542 * dev_change_flags - change device settings
5544 * @flags: device state flags
5546 * Change settings on device based state flags. The flags are
5547 * in the userspace exported format.
5549 int dev_change_flags(struct net_device
*dev
, unsigned int flags
)
5552 unsigned int changes
, old_flags
= dev
->flags
, old_gflags
= dev
->gflags
;
5554 ret
= __dev_change_flags(dev
, flags
);
5558 changes
= (old_flags
^ dev
->flags
) | (old_gflags
^ dev
->gflags
);
5559 __dev_notify_flags(dev
, old_flags
, changes
);
5562 EXPORT_SYMBOL(dev_change_flags
);
5564 static int __dev_set_mtu(struct net_device
*dev
, int new_mtu
)
5566 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5568 if (ops
->ndo_change_mtu
)
5569 return ops
->ndo_change_mtu(dev
, new_mtu
);
5576 * dev_set_mtu - Change maximum transfer unit
5578 * @new_mtu: new transfer unit
5580 * Change the maximum transfer size of the network device.
5582 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
5586 if (new_mtu
== dev
->mtu
)
5589 /* MTU must be positive. */
5593 if (!netif_device_present(dev
))
5596 err
= call_netdevice_notifiers(NETDEV_PRECHANGEMTU
, dev
);
5597 err
= notifier_to_errno(err
);
5601 orig_mtu
= dev
->mtu
;
5602 err
= __dev_set_mtu(dev
, new_mtu
);
5605 err
= call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
5606 err
= notifier_to_errno(err
);
5608 /* setting mtu back and notifying everyone again,
5609 * so that they have a chance to revert changes.
5611 __dev_set_mtu(dev
, orig_mtu
);
5612 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
5617 EXPORT_SYMBOL(dev_set_mtu
);
5620 * dev_set_group - Change group this device belongs to
5622 * @new_group: group this device should belong to
5624 void dev_set_group(struct net_device
*dev
, int new_group
)
5626 dev
->group
= new_group
;
5628 EXPORT_SYMBOL(dev_set_group
);
5631 * dev_set_mac_address - Change Media Access Control Address
5635 * Change the hardware (MAC) address of the device
5637 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
5639 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5642 if (!ops
->ndo_set_mac_address
)
5644 if (sa
->sa_family
!= dev
->type
)
5646 if (!netif_device_present(dev
))
5648 err
= ops
->ndo_set_mac_address(dev
, sa
);
5651 dev
->addr_assign_type
= NET_ADDR_SET
;
5652 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
5653 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
5656 EXPORT_SYMBOL(dev_set_mac_address
);
5659 * dev_change_carrier - Change device carrier
5661 * @new_carrier: new value
5663 * Change device carrier
5665 int dev_change_carrier(struct net_device
*dev
, bool new_carrier
)
5667 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5669 if (!ops
->ndo_change_carrier
)
5671 if (!netif_device_present(dev
))
5673 return ops
->ndo_change_carrier(dev
, new_carrier
);
5675 EXPORT_SYMBOL(dev_change_carrier
);
5678 * dev_get_phys_port_id - Get device physical port ID
5682 * Get device physical port ID
5684 int dev_get_phys_port_id(struct net_device
*dev
,
5685 struct netdev_phys_port_id
*ppid
)
5687 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5689 if (!ops
->ndo_get_phys_port_id
)
5691 return ops
->ndo_get_phys_port_id(dev
, ppid
);
5693 EXPORT_SYMBOL(dev_get_phys_port_id
);
5696 * dev_new_index - allocate an ifindex
5697 * @net: the applicable net namespace
5699 * Returns a suitable unique value for a new device interface
5700 * number. The caller must hold the rtnl semaphore or the
5701 * dev_base_lock to be sure it remains unique.
5703 static int dev_new_index(struct net
*net
)
5705 int ifindex
= net
->ifindex
;
5709 if (!__dev_get_by_index(net
, ifindex
))
5710 return net
->ifindex
= ifindex
;
5714 /* Delayed registration/unregisteration */
5715 static LIST_HEAD(net_todo_list
);
5716 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq
);
5718 static void net_set_todo(struct net_device
*dev
)
5720 list_add_tail(&dev
->todo_list
, &net_todo_list
);
5721 dev_net(dev
)->dev_unreg_count
++;
5724 static void rollback_registered_many(struct list_head
*head
)
5726 struct net_device
*dev
, *tmp
;
5727 LIST_HEAD(close_head
);
5729 BUG_ON(dev_boot_phase
);
5732 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
5733 /* Some devices call without registering
5734 * for initialization unwind. Remove those
5735 * devices and proceed with the remaining.
5737 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5738 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5742 list_del(&dev
->unreg_list
);
5745 dev
->dismantle
= true;
5746 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
5749 /* If device is running, close it first. */
5750 list_for_each_entry(dev
, head
, unreg_list
)
5751 list_add_tail(&dev
->close_list
, &close_head
);
5752 dev_close_many(&close_head
);
5754 list_for_each_entry(dev
, head
, unreg_list
) {
5755 /* And unlink it from device chain. */
5756 unlist_netdevice(dev
);
5758 dev
->reg_state
= NETREG_UNREGISTERING
;
5763 list_for_each_entry(dev
, head
, unreg_list
) {
5764 /* Shutdown queueing discipline. */
5768 /* Notify protocols, that we are about to destroy
5769 this device. They should clean all the things.
5771 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5774 * Flush the unicast and multicast chains
5779 if (dev
->netdev_ops
->ndo_uninit
)
5780 dev
->netdev_ops
->ndo_uninit(dev
);
5782 if (!dev
->rtnl_link_ops
||
5783 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5784 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U, GFP_KERNEL
);
5786 /* Notifier chain MUST detach us all upper devices. */
5787 WARN_ON(netdev_has_any_upper_dev(dev
));
5789 /* Remove entries from kobject tree */
5790 netdev_unregister_kobject(dev
);
5792 /* Remove XPS queueing entries */
5793 netif_reset_xps_queues_gt(dev
, 0);
5799 list_for_each_entry(dev
, head
, unreg_list
)
5803 static void rollback_registered(struct net_device
*dev
)
5807 list_add(&dev
->unreg_list
, &single
);
5808 rollback_registered_many(&single
);
5812 static netdev_features_t
netdev_fix_features(struct net_device
*dev
,
5813 netdev_features_t features
)
5815 /* Fix illegal checksum combinations */
5816 if ((features
& NETIF_F_HW_CSUM
) &&
5817 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5818 netdev_warn(dev
, "mixed HW and IP checksum settings.\n");
5819 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5822 /* TSO requires that SG is present as well. */
5823 if ((features
& NETIF_F_ALL_TSO
) && !(features
& NETIF_F_SG
)) {
5824 netdev_dbg(dev
, "Dropping TSO features since no SG feature.\n");
5825 features
&= ~NETIF_F_ALL_TSO
;
5828 if ((features
& NETIF_F_TSO
) && !(features
& NETIF_F_HW_CSUM
) &&
5829 !(features
& NETIF_F_IP_CSUM
)) {
5830 netdev_dbg(dev
, "Dropping TSO features since no CSUM feature.\n");
5831 features
&= ~NETIF_F_TSO
;
5832 features
&= ~NETIF_F_TSO_ECN
;
5835 if ((features
& NETIF_F_TSO6
) && !(features
& NETIF_F_HW_CSUM
) &&
5836 !(features
& NETIF_F_IPV6_CSUM
)) {
5837 netdev_dbg(dev
, "Dropping TSO6 features since no CSUM feature.\n");
5838 features
&= ~NETIF_F_TSO6
;
5841 /* TSO ECN requires that TSO is present as well. */
5842 if ((features
& NETIF_F_ALL_TSO
) == NETIF_F_TSO_ECN
)
5843 features
&= ~NETIF_F_TSO_ECN
;
5845 /* Software GSO depends on SG. */
5846 if ((features
& NETIF_F_GSO
) && !(features
& NETIF_F_SG
)) {
5847 netdev_dbg(dev
, "Dropping NETIF_F_GSO since no SG feature.\n");
5848 features
&= ~NETIF_F_GSO
;
5851 /* UFO needs SG and checksumming */
5852 if (features
& NETIF_F_UFO
) {
5853 /* maybe split UFO into V4 and V6? */
5854 if (!((features
& NETIF_F_GEN_CSUM
) ||
5855 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))
5856 == (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5858 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5859 features
&= ~NETIF_F_UFO
;
5862 if (!(features
& NETIF_F_SG
)) {
5864 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5865 features
&= ~NETIF_F_UFO
;
5869 #ifdef CONFIG_NET_RX_BUSY_POLL
5870 if (dev
->netdev_ops
->ndo_busy_poll
)
5871 features
|= NETIF_F_BUSY_POLL
;
5874 features
&= ~NETIF_F_BUSY_POLL
;
5879 int __netdev_update_features(struct net_device
*dev
)
5881 netdev_features_t features
;
5886 features
= netdev_get_wanted_features(dev
);
5888 if (dev
->netdev_ops
->ndo_fix_features
)
5889 features
= dev
->netdev_ops
->ndo_fix_features(dev
, features
);
5891 /* driver might be less strict about feature dependencies */
5892 features
= netdev_fix_features(dev
, features
);
5894 if (dev
->features
== features
)
5897 netdev_dbg(dev
, "Features changed: %pNF -> %pNF\n",
5898 &dev
->features
, &features
);
5900 if (dev
->netdev_ops
->ndo_set_features
)
5901 err
= dev
->netdev_ops
->ndo_set_features(dev
, features
);
5903 if (unlikely(err
< 0)) {
5905 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5906 err
, &features
, &dev
->features
);
5911 dev
->features
= features
;
5917 * netdev_update_features - recalculate device features
5918 * @dev: the device to check
5920 * Recalculate dev->features set and send notifications if it
5921 * has changed. Should be called after driver or hardware dependent
5922 * conditions might have changed that influence the features.
5924 void netdev_update_features(struct net_device
*dev
)
5926 if (__netdev_update_features(dev
))
5927 netdev_features_change(dev
);
5929 EXPORT_SYMBOL(netdev_update_features
);
5932 * netdev_change_features - recalculate device features
5933 * @dev: the device to check
5935 * Recalculate dev->features set and send notifications even
5936 * if they have not changed. Should be called instead of
5937 * netdev_update_features() if also dev->vlan_features might
5938 * have changed to allow the changes to be propagated to stacked
5941 void netdev_change_features(struct net_device
*dev
)
5943 __netdev_update_features(dev
);
5944 netdev_features_change(dev
);
5946 EXPORT_SYMBOL(netdev_change_features
);
5949 * netif_stacked_transfer_operstate - transfer operstate
5950 * @rootdev: the root or lower level device to transfer state from
5951 * @dev: the device to transfer operstate to
5953 * Transfer operational state from root to device. This is normally
5954 * called when a stacking relationship exists between the root
5955 * device and the device(a leaf device).
5957 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
5958 struct net_device
*dev
)
5960 if (rootdev
->operstate
== IF_OPER_DORMANT
)
5961 netif_dormant_on(dev
);
5963 netif_dormant_off(dev
);
5965 if (netif_carrier_ok(rootdev
)) {
5966 if (!netif_carrier_ok(dev
))
5967 netif_carrier_on(dev
);
5969 if (netif_carrier_ok(dev
))
5970 netif_carrier_off(dev
);
5973 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
5976 static int netif_alloc_rx_queues(struct net_device
*dev
)
5978 unsigned int i
, count
= dev
->num_rx_queues
;
5979 struct netdev_rx_queue
*rx
;
5983 rx
= kcalloc(count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5989 for (i
= 0; i
< count
; i
++)
5995 static void netdev_init_one_queue(struct net_device
*dev
,
5996 struct netdev_queue
*queue
, void *_unused
)
5998 /* Initialize queue lock */
5999 spin_lock_init(&queue
->_xmit_lock
);
6000 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
6001 queue
->xmit_lock_owner
= -1;
6002 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
6005 dql_init(&queue
->dql
, HZ
);
6009 static void netif_free_tx_queues(struct net_device
*dev
)
6014 static int netif_alloc_netdev_queues(struct net_device
*dev
)
6016 unsigned int count
= dev
->num_tx_queues
;
6017 struct netdev_queue
*tx
;
6018 size_t sz
= count
* sizeof(*tx
);
6020 BUG_ON(count
< 1 || count
> 0xffff);
6022 tx
= kzalloc(sz
, GFP_KERNEL
| __GFP_NOWARN
| __GFP_REPEAT
);
6030 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
6031 spin_lock_init(&dev
->tx_global_lock
);
6037 * register_netdevice - register a network device
6038 * @dev: device to register
6040 * Take a completed network device structure and add it to the kernel
6041 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6042 * chain. 0 is returned on success. A negative errno code is returned
6043 * on a failure to set up the device, or if the name is a duplicate.
6045 * Callers must hold the rtnl semaphore. You may want
6046 * register_netdev() instead of this.
6049 * The locking appears insufficient to guarantee two parallel registers
6050 * will not get the same name.
6053 int register_netdevice(struct net_device
*dev
)
6056 struct net
*net
= dev_net(dev
);
6058 BUG_ON(dev_boot_phase
);
6063 /* When net_device's are persistent, this will be fatal. */
6064 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
6067 spin_lock_init(&dev
->addr_list_lock
);
6068 netdev_set_addr_lockdep_class(dev
);
6072 ret
= dev_get_valid_name(net
, dev
, dev
->name
);
6076 /* Init, if this function is available */
6077 if (dev
->netdev_ops
->ndo_init
) {
6078 ret
= dev
->netdev_ops
->ndo_init(dev
);
6086 if (((dev
->hw_features
| dev
->features
) &
6087 NETIF_F_HW_VLAN_CTAG_FILTER
) &&
6088 (!dev
->netdev_ops
->ndo_vlan_rx_add_vid
||
6089 !dev
->netdev_ops
->ndo_vlan_rx_kill_vid
)) {
6090 netdev_WARN(dev
, "Buggy VLAN acceleration in driver!\n");
6097 dev
->ifindex
= dev_new_index(net
);
6098 else if (__dev_get_by_index(net
, dev
->ifindex
))
6101 if (dev
->iflink
== -1)
6102 dev
->iflink
= dev
->ifindex
;
6104 /* Transfer changeable features to wanted_features and enable
6105 * software offloads (GSO and GRO).
6107 dev
->hw_features
|= NETIF_F_SOFT_FEATURES
;
6108 dev
->features
|= NETIF_F_SOFT_FEATURES
;
6109 dev
->wanted_features
= dev
->features
& dev
->hw_features
;
6111 if (!(dev
->flags
& IFF_LOOPBACK
)) {
6112 dev
->hw_features
|= NETIF_F_NOCACHE_COPY
;
6115 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6117 dev
->vlan_features
|= NETIF_F_HIGHDMA
;
6119 /* Make NETIF_F_SG inheritable to tunnel devices.
6121 dev
->hw_enc_features
|= NETIF_F_SG
;
6123 /* Make NETIF_F_SG inheritable to MPLS.
6125 dev
->mpls_features
|= NETIF_F_SG
;
6127 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
6128 ret
= notifier_to_errno(ret
);
6132 ret
= netdev_register_kobject(dev
);
6135 dev
->reg_state
= NETREG_REGISTERED
;
6137 __netdev_update_features(dev
);
6140 * Default initial state at registry is that the
6141 * device is present.
6144 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
6146 linkwatch_init_dev(dev
);
6148 dev_init_scheduler(dev
);
6150 list_netdevice(dev
);
6151 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
6153 /* If the device has permanent device address, driver should
6154 * set dev_addr and also addr_assign_type should be set to
6155 * NET_ADDR_PERM (default value).
6157 if (dev
->addr_assign_type
== NET_ADDR_PERM
)
6158 memcpy(dev
->perm_addr
, dev
->dev_addr
, dev
->addr_len
);
6160 /* Notify protocols, that a new device appeared. */
6161 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
6162 ret
= notifier_to_errno(ret
);
6164 rollback_registered(dev
);
6165 dev
->reg_state
= NETREG_UNREGISTERED
;
6168 * Prevent userspace races by waiting until the network
6169 * device is fully setup before sending notifications.
6171 if (!dev
->rtnl_link_ops
||
6172 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
6173 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U, GFP_KERNEL
);
6179 if (dev
->netdev_ops
->ndo_uninit
)
6180 dev
->netdev_ops
->ndo_uninit(dev
);
6183 EXPORT_SYMBOL(register_netdevice
);
6186 * init_dummy_netdev - init a dummy network device for NAPI
6187 * @dev: device to init
6189 * This takes a network device structure and initialize the minimum
6190 * amount of fields so it can be used to schedule NAPI polls without
6191 * registering a full blown interface. This is to be used by drivers
6192 * that need to tie several hardware interfaces to a single NAPI
6193 * poll scheduler due to HW limitations.
6195 int init_dummy_netdev(struct net_device
*dev
)
6197 /* Clear everything. Note we don't initialize spinlocks
6198 * are they aren't supposed to be taken by any of the
6199 * NAPI code and this dummy netdev is supposed to be
6200 * only ever used for NAPI polls
6202 memset(dev
, 0, sizeof(struct net_device
));
6204 /* make sure we BUG if trying to hit standard
6205 * register/unregister code path
6207 dev
->reg_state
= NETREG_DUMMY
;
6209 /* NAPI wants this */
6210 INIT_LIST_HEAD(&dev
->napi_list
);
6212 /* a dummy interface is started by default */
6213 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
6214 set_bit(__LINK_STATE_START
, &dev
->state
);
6216 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6217 * because users of this 'device' dont need to change
6223 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
6227 * register_netdev - register a network device
6228 * @dev: device to register
6230 * Take a completed network device structure and add it to the kernel
6231 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6232 * chain. 0 is returned on success. A negative errno code is returned
6233 * on a failure to set up the device, or if the name is a duplicate.
6235 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6236 * and expands the device name if you passed a format string to
6239 int register_netdev(struct net_device
*dev
)
6244 err
= register_netdevice(dev
);
6248 EXPORT_SYMBOL(register_netdev
);
6250 int netdev_refcnt_read(const struct net_device
*dev
)
6254 for_each_possible_cpu(i
)
6255 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
6258 EXPORT_SYMBOL(netdev_refcnt_read
);
6261 * netdev_wait_allrefs - wait until all references are gone.
6262 * @dev: target net_device
6264 * This is called when unregistering network devices.
6266 * Any protocol or device that holds a reference should register
6267 * for netdevice notification, and cleanup and put back the
6268 * reference if they receive an UNREGISTER event.
6269 * We can get stuck here if buggy protocols don't correctly
6272 static void netdev_wait_allrefs(struct net_device
*dev
)
6274 unsigned long rebroadcast_time
, warning_time
;
6277 linkwatch_forget_dev(dev
);
6279 rebroadcast_time
= warning_time
= jiffies
;
6280 refcnt
= netdev_refcnt_read(dev
);
6282 while (refcnt
!= 0) {
6283 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
6286 /* Rebroadcast unregister notification */
6287 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6293 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
6294 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
6296 /* We must not have linkwatch events
6297 * pending on unregister. If this
6298 * happens, we simply run the queue
6299 * unscheduled, resulting in a noop
6302 linkwatch_run_queue();
6307 rebroadcast_time
= jiffies
;
6312 refcnt
= netdev_refcnt_read(dev
);
6314 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
6315 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6317 warning_time
= jiffies
;
6326 * register_netdevice(x1);
6327 * register_netdevice(x2);
6329 * unregister_netdevice(y1);
6330 * unregister_netdevice(y2);
6336 * We are invoked by rtnl_unlock().
6337 * This allows us to deal with problems:
6338 * 1) We can delete sysfs objects which invoke hotplug
6339 * without deadlocking with linkwatch via keventd.
6340 * 2) Since we run with the RTNL semaphore not held, we can sleep
6341 * safely in order to wait for the netdev refcnt to drop to zero.
6343 * We must not return until all unregister events added during
6344 * the interval the lock was held have been completed.
6346 void netdev_run_todo(void)
6348 struct list_head list
;
6350 /* Snapshot list, allow later requests */
6351 list_replace_init(&net_todo_list
, &list
);
6356 /* Wait for rcu callbacks to finish before next phase */
6357 if (!list_empty(&list
))
6360 while (!list_empty(&list
)) {
6361 struct net_device
*dev
6362 = list_first_entry(&list
, struct net_device
, todo_list
);
6363 list_del(&dev
->todo_list
);
6366 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
6369 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
6370 pr_err("network todo '%s' but state %d\n",
6371 dev
->name
, dev
->reg_state
);
6376 dev
->reg_state
= NETREG_UNREGISTERED
;
6378 on_each_cpu(flush_backlog
, dev
, 1);
6380 netdev_wait_allrefs(dev
);
6383 BUG_ON(netdev_refcnt_read(dev
));
6384 WARN_ON(rcu_access_pointer(dev
->ip_ptr
));
6385 WARN_ON(rcu_access_pointer(dev
->ip6_ptr
));
6386 WARN_ON(dev
->dn_ptr
);
6388 if (dev
->destructor
)
6389 dev
->destructor(dev
);
6391 /* Report a network device has been unregistered */
6393 dev_net(dev
)->dev_unreg_count
--;
6395 wake_up(&netdev_unregistering_wq
);
6397 /* Free network device */
6398 kobject_put(&dev
->dev
.kobj
);
6402 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6403 * fields in the same order, with only the type differing.
6405 void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
6406 const struct net_device_stats
*netdev_stats
)
6408 #if BITS_PER_LONG == 64
6409 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
6410 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
6412 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
6413 const unsigned long *src
= (const unsigned long *)netdev_stats
;
6414 u64
*dst
= (u64
*)stats64
;
6416 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
6417 sizeof(*stats64
) / sizeof(u64
));
6418 for (i
= 0; i
< n
; i
++)
6422 EXPORT_SYMBOL(netdev_stats_to_stats64
);
6425 * dev_get_stats - get network device statistics
6426 * @dev: device to get statistics from
6427 * @storage: place to store stats
6429 * Get network statistics from device. Return @storage.
6430 * The device driver may provide its own method by setting
6431 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6432 * otherwise the internal statistics structure is used.
6434 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
6435 struct rtnl_link_stats64
*storage
)
6437 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6439 if (ops
->ndo_get_stats64
) {
6440 memset(storage
, 0, sizeof(*storage
));
6441 ops
->ndo_get_stats64(dev
, storage
);
6442 } else if (ops
->ndo_get_stats
) {
6443 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
6445 netdev_stats_to_stats64(storage
, &dev
->stats
);
6447 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
6448 storage
->tx_dropped
+= atomic_long_read(&dev
->tx_dropped
);
6451 EXPORT_SYMBOL(dev_get_stats
);
6453 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
6455 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
6457 #ifdef CONFIG_NET_CLS_ACT
6460 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
6463 netdev_init_one_queue(dev
, queue
, NULL
);
6464 queue
->qdisc
= &noop_qdisc
;
6465 queue
->qdisc_sleeping
= &noop_qdisc
;
6466 rcu_assign_pointer(dev
->ingress_queue
, queue
);
6471 static const struct ethtool_ops default_ethtool_ops
;
6473 void netdev_set_default_ethtool_ops(struct net_device
*dev
,
6474 const struct ethtool_ops
*ops
)
6476 if (dev
->ethtool_ops
== &default_ethtool_ops
)
6477 dev
->ethtool_ops
= ops
;
6479 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops
);
6481 void netdev_freemem(struct net_device
*dev
)
6483 char *addr
= (char *)dev
- dev
->padded
;
6489 * alloc_netdev_mqs - allocate network device
6490 * @sizeof_priv: size of private data to allocate space for
6491 * @name: device name format string
6492 * @name_assign_type: origin of device name
6493 * @setup: callback to initialize device
6494 * @txqs: the number of TX subqueues to allocate
6495 * @rxqs: the number of RX subqueues to allocate
6497 * Allocates a struct net_device with private data area for driver use
6498 * and performs basic initialization. Also allocates subqueue structs
6499 * for each queue on the device.
6501 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
6502 unsigned char name_assign_type
,
6503 void (*setup
)(struct net_device
*),
6504 unsigned int txqs
, unsigned int rxqs
)
6506 struct net_device
*dev
;
6508 struct net_device
*p
;
6510 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
6513 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6519 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6524 alloc_size
= sizeof(struct net_device
);
6526 /* ensure 32-byte alignment of private area */
6527 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
6528 alloc_size
+= sizeof_priv
;
6530 /* ensure 32-byte alignment of whole construct */
6531 alloc_size
+= NETDEV_ALIGN
- 1;
6533 p
= kzalloc(alloc_size
, GFP_KERNEL
| __GFP_NOWARN
| __GFP_REPEAT
);
6535 p
= vzalloc(alloc_size
);
6539 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
6540 dev
->padded
= (char *)dev
- (char *)p
;
6542 dev
->pcpu_refcnt
= alloc_percpu(int);
6543 if (!dev
->pcpu_refcnt
)
6546 if (dev_addr_init(dev
))
6552 dev_net_set(dev
, &init_net
);
6554 dev
->gso_max_size
= GSO_MAX_SIZE
;
6555 dev
->gso_max_segs
= GSO_MAX_SEGS
;
6557 INIT_LIST_HEAD(&dev
->napi_list
);
6558 INIT_LIST_HEAD(&dev
->unreg_list
);
6559 INIT_LIST_HEAD(&dev
->close_list
);
6560 INIT_LIST_HEAD(&dev
->link_watch_list
);
6561 INIT_LIST_HEAD(&dev
->adj_list
.upper
);
6562 INIT_LIST_HEAD(&dev
->adj_list
.lower
);
6563 INIT_LIST_HEAD(&dev
->all_adj_list
.upper
);
6564 INIT_LIST_HEAD(&dev
->all_adj_list
.lower
);
6565 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
6568 dev
->num_tx_queues
= txqs
;
6569 dev
->real_num_tx_queues
= txqs
;
6570 if (netif_alloc_netdev_queues(dev
))
6574 dev
->num_rx_queues
= rxqs
;
6575 dev
->real_num_rx_queues
= rxqs
;
6576 if (netif_alloc_rx_queues(dev
))
6580 strcpy(dev
->name
, name
);
6581 dev
->name_assign_type
= name_assign_type
;
6582 dev
->group
= INIT_NETDEV_GROUP
;
6583 if (!dev
->ethtool_ops
)
6584 dev
->ethtool_ops
= &default_ethtool_ops
;
6592 free_percpu(dev
->pcpu_refcnt
);
6594 netdev_freemem(dev
);
6597 EXPORT_SYMBOL(alloc_netdev_mqs
);
6600 * free_netdev - free network device
6603 * This function does the last stage of destroying an allocated device
6604 * interface. The reference to the device object is released.
6605 * If this is the last reference then it will be freed.
6607 void free_netdev(struct net_device
*dev
)
6609 struct napi_struct
*p
, *n
;
6611 release_net(dev_net(dev
));
6613 netif_free_tx_queues(dev
);
6618 kfree(rcu_dereference_protected(dev
->ingress_queue
, 1));
6620 /* Flush device addresses */
6621 dev_addr_flush(dev
);
6623 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
6626 free_percpu(dev
->pcpu_refcnt
);
6627 dev
->pcpu_refcnt
= NULL
;
6629 /* Compatibility with error handling in drivers */
6630 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
6631 netdev_freemem(dev
);
6635 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
6636 dev
->reg_state
= NETREG_RELEASED
;
6638 /* will free via device release */
6639 put_device(&dev
->dev
);
6641 EXPORT_SYMBOL(free_netdev
);
6644 * synchronize_net - Synchronize with packet receive processing
6646 * Wait for packets currently being received to be done.
6647 * Does not block later packets from starting.
6649 void synchronize_net(void)
6652 if (rtnl_is_locked())
6653 synchronize_rcu_expedited();
6657 EXPORT_SYMBOL(synchronize_net
);
6660 * unregister_netdevice_queue - remove device from the kernel
6664 * This function shuts down a device interface and removes it
6665 * from the kernel tables.
6666 * If head not NULL, device is queued to be unregistered later.
6668 * Callers must hold the rtnl semaphore. You may want
6669 * unregister_netdev() instead of this.
6672 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
6677 list_move_tail(&dev
->unreg_list
, head
);
6679 rollback_registered(dev
);
6680 /* Finish processing unregister after unlock */
6684 EXPORT_SYMBOL(unregister_netdevice_queue
);
6687 * unregister_netdevice_many - unregister many devices
6688 * @head: list of devices
6690 * Note: As most callers use a stack allocated list_head,
6691 * we force a list_del() to make sure stack wont be corrupted later.
6693 void unregister_netdevice_many(struct list_head
*head
)
6695 struct net_device
*dev
;
6697 if (!list_empty(head
)) {
6698 rollback_registered_many(head
);
6699 list_for_each_entry(dev
, head
, unreg_list
)
6704 EXPORT_SYMBOL(unregister_netdevice_many
);
6707 * unregister_netdev - remove device from the kernel
6710 * This function shuts down a device interface and removes it
6711 * from the kernel tables.
6713 * This is just a wrapper for unregister_netdevice that takes
6714 * the rtnl semaphore. In general you want to use this and not
6715 * unregister_netdevice.
6717 void unregister_netdev(struct net_device
*dev
)
6720 unregister_netdevice(dev
);
6723 EXPORT_SYMBOL(unregister_netdev
);
6726 * dev_change_net_namespace - move device to different nethost namespace
6728 * @net: network namespace
6729 * @pat: If not NULL name pattern to try if the current device name
6730 * is already taken in the destination network namespace.
6732 * This function shuts down a device interface and moves it
6733 * to a new network namespace. On success 0 is returned, on
6734 * a failure a netagive errno code is returned.
6736 * Callers must hold the rtnl semaphore.
6739 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
6745 /* Don't allow namespace local devices to be moved. */
6747 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6750 /* Ensure the device has been registrered */
6751 if (dev
->reg_state
!= NETREG_REGISTERED
)
6754 /* Get out if there is nothing todo */
6756 if (net_eq(dev_net(dev
), net
))
6759 /* Pick the destination device name, and ensure
6760 * we can use it in the destination network namespace.
6763 if (__dev_get_by_name(net
, dev
->name
)) {
6764 /* We get here if we can't use the current device name */
6767 if (dev_get_valid_name(net
, dev
, pat
) < 0)
6772 * And now a mini version of register_netdevice unregister_netdevice.
6775 /* If device is running close it first. */
6778 /* And unlink it from device chain */
6780 unlist_netdevice(dev
);
6784 /* Shutdown queueing discipline. */
6787 /* Notify protocols, that we are about to destroy
6788 this device. They should clean all the things.
6790 Note that dev->reg_state stays at NETREG_REGISTERED.
6791 This is wanted because this way 8021q and macvlan know
6792 the device is just moving and can keep their slaves up.
6794 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6796 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
6797 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U, GFP_KERNEL
);
6800 * Flush the unicast and multicast chains
6805 /* Send a netdev-removed uevent to the old namespace */
6806 kobject_uevent(&dev
->dev
.kobj
, KOBJ_REMOVE
);
6808 /* Actually switch the network namespace */
6809 dev_net_set(dev
, net
);
6811 /* If there is an ifindex conflict assign a new one */
6812 if (__dev_get_by_index(net
, dev
->ifindex
)) {
6813 int iflink
= (dev
->iflink
== dev
->ifindex
);
6814 dev
->ifindex
= dev_new_index(net
);
6816 dev
->iflink
= dev
->ifindex
;
6819 /* Send a netdev-add uevent to the new namespace */
6820 kobject_uevent(&dev
->dev
.kobj
, KOBJ_ADD
);
6822 /* Fixup kobjects */
6823 err
= device_rename(&dev
->dev
, dev
->name
);
6826 /* Add the device back in the hashes */
6827 list_netdevice(dev
);
6829 /* Notify protocols, that a new device appeared. */
6830 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
6833 * Prevent userspace races by waiting until the network
6834 * device is fully setup before sending notifications.
6836 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U, GFP_KERNEL
);
6843 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
6845 static int dev_cpu_callback(struct notifier_block
*nfb
,
6846 unsigned long action
,
6849 struct sk_buff
**list_skb
;
6850 struct sk_buff
*skb
;
6851 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
6852 struct softnet_data
*sd
, *oldsd
;
6854 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
6857 local_irq_disable();
6858 cpu
= smp_processor_id();
6859 sd
= &per_cpu(softnet_data
, cpu
);
6860 oldsd
= &per_cpu(softnet_data
, oldcpu
);
6862 /* Find end of our completion_queue. */
6863 list_skb
= &sd
->completion_queue
;
6865 list_skb
= &(*list_skb
)->next
;
6866 /* Append completion queue from offline CPU. */
6867 *list_skb
= oldsd
->completion_queue
;
6868 oldsd
->completion_queue
= NULL
;
6870 /* Append output queue from offline CPU. */
6871 if (oldsd
->output_queue
) {
6872 *sd
->output_queue_tailp
= oldsd
->output_queue
;
6873 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
6874 oldsd
->output_queue
= NULL
;
6875 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
6877 /* Append NAPI poll list from offline CPU. */
6878 if (!list_empty(&oldsd
->poll_list
)) {
6879 list_splice_init(&oldsd
->poll_list
, &sd
->poll_list
);
6880 raise_softirq_irqoff(NET_RX_SOFTIRQ
);
6883 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
6886 /* Process offline CPU's input_pkt_queue */
6887 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
6888 netif_rx_internal(skb
);
6889 input_queue_head_incr(oldsd
);
6891 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
6892 netif_rx_internal(skb
);
6893 input_queue_head_incr(oldsd
);
6901 * netdev_increment_features - increment feature set by one
6902 * @all: current feature set
6903 * @one: new feature set
6904 * @mask: mask feature set
6906 * Computes a new feature set after adding a device with feature set
6907 * @one to the master device with current feature set @all. Will not
6908 * enable anything that is off in @mask. Returns the new feature set.
6910 netdev_features_t
netdev_increment_features(netdev_features_t all
,
6911 netdev_features_t one
, netdev_features_t mask
)
6913 if (mask
& NETIF_F_GEN_CSUM
)
6914 mask
|= NETIF_F_ALL_CSUM
;
6915 mask
|= NETIF_F_VLAN_CHALLENGED
;
6917 all
|= one
& (NETIF_F_ONE_FOR_ALL
|NETIF_F_ALL_CSUM
) & mask
;
6918 all
&= one
| ~NETIF_F_ALL_FOR_ALL
;
6920 /* If one device supports hw checksumming, set for all. */
6921 if (all
& NETIF_F_GEN_CSUM
)
6922 all
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_GEN_CSUM
);
6926 EXPORT_SYMBOL(netdev_increment_features
);
6928 static struct hlist_head
* __net_init
netdev_create_hash(void)
6931 struct hlist_head
*hash
;
6933 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
6935 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
6936 INIT_HLIST_HEAD(&hash
[i
]);
6941 /* Initialize per network namespace state */
6942 static int __net_init
netdev_init(struct net
*net
)
6944 if (net
!= &init_net
)
6945 INIT_LIST_HEAD(&net
->dev_base_head
);
6947 net
->dev_name_head
= netdev_create_hash();
6948 if (net
->dev_name_head
== NULL
)
6951 net
->dev_index_head
= netdev_create_hash();
6952 if (net
->dev_index_head
== NULL
)
6958 kfree(net
->dev_name_head
);
6964 * netdev_drivername - network driver for the device
6965 * @dev: network device
6967 * Determine network driver for device.
6969 const char *netdev_drivername(const struct net_device
*dev
)
6971 const struct device_driver
*driver
;
6972 const struct device
*parent
;
6973 const char *empty
= "";
6975 parent
= dev
->dev
.parent
;
6979 driver
= parent
->driver
;
6980 if (driver
&& driver
->name
)
6981 return driver
->name
;
6985 static int __netdev_printk(const char *level
, const struct net_device
*dev
,
6986 struct va_format
*vaf
)
6990 if (dev
&& dev
->dev
.parent
) {
6991 r
= dev_printk_emit(level
[1] - '0',
6994 dev_driver_string(dev
->dev
.parent
),
6995 dev_name(dev
->dev
.parent
),
6996 netdev_name(dev
), netdev_reg_state(dev
),
6999 r
= printk("%s%s%s: %pV", level
, netdev_name(dev
),
7000 netdev_reg_state(dev
), vaf
);
7002 r
= printk("%s(NULL net_device): %pV", level
, vaf
);
7008 int netdev_printk(const char *level
, const struct net_device
*dev
,
7009 const char *format
, ...)
7011 struct va_format vaf
;
7015 va_start(args
, format
);
7020 r
= __netdev_printk(level
, dev
, &vaf
);
7026 EXPORT_SYMBOL(netdev_printk
);
7028 #define define_netdev_printk_level(func, level) \
7029 int func(const struct net_device *dev, const char *fmt, ...) \
7032 struct va_format vaf; \
7035 va_start(args, fmt); \
7040 r = __netdev_printk(level, dev, &vaf); \
7046 EXPORT_SYMBOL(func);
7048 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
7049 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
7050 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
7051 define_netdev_printk_level(netdev_err
, KERN_ERR
);
7052 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
7053 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
7054 define_netdev_printk_level(netdev_info
, KERN_INFO
);
7056 static void __net_exit
netdev_exit(struct net
*net
)
7058 kfree(net
->dev_name_head
);
7059 kfree(net
->dev_index_head
);
7062 static struct pernet_operations __net_initdata netdev_net_ops
= {
7063 .init
= netdev_init
,
7064 .exit
= netdev_exit
,
7067 static void __net_exit
default_device_exit(struct net
*net
)
7069 struct net_device
*dev
, *aux
;
7071 * Push all migratable network devices back to the
7072 * initial network namespace
7075 for_each_netdev_safe(net
, dev
, aux
) {
7077 char fb_name
[IFNAMSIZ
];
7079 /* Ignore unmoveable devices (i.e. loopback) */
7080 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
7083 /* Leave virtual devices for the generic cleanup */
7084 if (dev
->rtnl_link_ops
)
7087 /* Push remaining network devices to init_net */
7088 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
7089 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
7091 pr_emerg("%s: failed to move %s to init_net: %d\n",
7092 __func__
, dev
->name
, err
);
7099 static void __net_exit
rtnl_lock_unregistering(struct list_head
*net_list
)
7101 /* Return with the rtnl_lock held when there are no network
7102 * devices unregistering in any network namespace in net_list.
7109 prepare_to_wait(&netdev_unregistering_wq
, &wait
,
7110 TASK_UNINTERRUPTIBLE
);
7111 unregistering
= false;
7113 list_for_each_entry(net
, net_list
, exit_list
) {
7114 if (net
->dev_unreg_count
> 0) {
7115 unregistering
= true;
7124 finish_wait(&netdev_unregistering_wq
, &wait
);
7127 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
7129 /* At exit all network devices most be removed from a network
7130 * namespace. Do this in the reverse order of registration.
7131 * Do this across as many network namespaces as possible to
7132 * improve batching efficiency.
7134 struct net_device
*dev
;
7136 LIST_HEAD(dev_kill_list
);
7138 /* To prevent network device cleanup code from dereferencing
7139 * loopback devices or network devices that have been freed
7140 * wait here for all pending unregistrations to complete,
7141 * before unregistring the loopback device and allowing the
7142 * network namespace be freed.
7144 * The netdev todo list containing all network devices
7145 * unregistrations that happen in default_device_exit_batch
7146 * will run in the rtnl_unlock() at the end of
7147 * default_device_exit_batch.
7149 rtnl_lock_unregistering(net_list
);
7150 list_for_each_entry(net
, net_list
, exit_list
) {
7151 for_each_netdev_reverse(net
, dev
) {
7152 if (dev
->rtnl_link_ops
&& dev
->rtnl_link_ops
->dellink
)
7153 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
7155 unregister_netdevice_queue(dev
, &dev_kill_list
);
7158 unregister_netdevice_many(&dev_kill_list
);
7162 static struct pernet_operations __net_initdata default_device_ops
= {
7163 .exit
= default_device_exit
,
7164 .exit_batch
= default_device_exit_batch
,
7168 * Initialize the DEV module. At boot time this walks the device list and
7169 * unhooks any devices that fail to initialise (normally hardware not
7170 * present) and leaves us with a valid list of present and active devices.
7175 * This is called single threaded during boot, so no need
7176 * to take the rtnl semaphore.
7178 static int __init
net_dev_init(void)
7180 int i
, rc
= -ENOMEM
;
7182 BUG_ON(!dev_boot_phase
);
7184 if (dev_proc_init())
7187 if (netdev_kobject_init())
7190 INIT_LIST_HEAD(&ptype_all
);
7191 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
7192 INIT_LIST_HEAD(&ptype_base
[i
]);
7194 INIT_LIST_HEAD(&offload_base
);
7196 if (register_pernet_subsys(&netdev_net_ops
))
7200 * Initialise the packet receive queues.
7203 for_each_possible_cpu(i
) {
7204 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
7206 skb_queue_head_init(&sd
->input_pkt_queue
);
7207 skb_queue_head_init(&sd
->process_queue
);
7208 INIT_LIST_HEAD(&sd
->poll_list
);
7209 sd
->output_queue_tailp
= &sd
->output_queue
;
7211 sd
->csd
.func
= rps_trigger_softirq
;
7216 sd
->backlog
.poll
= process_backlog
;
7217 sd
->backlog
.weight
= weight_p
;
7222 /* The loopback device is special if any other network devices
7223 * is present in a network namespace the loopback device must
7224 * be present. Since we now dynamically allocate and free the
7225 * loopback device ensure this invariant is maintained by
7226 * keeping the loopback device as the first device on the
7227 * list of network devices. Ensuring the loopback devices
7228 * is the first device that appears and the last network device
7231 if (register_pernet_device(&loopback_net_ops
))
7234 if (register_pernet_device(&default_device_ops
))
7237 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
7238 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
7240 hotcpu_notifier(dev_cpu_callback
, 0);
7247 subsys_initcall(net_dev_init
);