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
)
2488 /* If MPLS offload request, verify we are testing hardware MPLS features
2489 * instead of standard features for the netdev.
2491 #ifdef CONFIG_NET_MPLS_GSO
2492 static netdev_features_t
net_mpls_features(struct sk_buff
*skb
,
2493 netdev_features_t features
,
2496 if (type
== htons(ETH_P_MPLS_UC
) || type
== htons(ETH_P_MPLS_MC
))
2497 features
&= skb
->dev
->mpls_features
;
2502 static netdev_features_t
net_mpls_features(struct sk_buff
*skb
,
2503 netdev_features_t features
,
2510 static netdev_features_t
harmonize_features(struct sk_buff
*skb
,
2511 netdev_features_t features
)
2516 type
= skb_network_protocol(skb
, &tmp
);
2517 features
= net_mpls_features(skb
, features
, type
);
2519 if (skb
->ip_summed
!= CHECKSUM_NONE
&&
2520 !can_checksum_protocol(features
, type
)) {
2521 features
&= ~NETIF_F_ALL_CSUM
;
2522 } else if (illegal_highdma(skb
->dev
, skb
)) {
2523 features
&= ~NETIF_F_SG
;
2529 netdev_features_t
netif_skb_features(struct sk_buff
*skb
)
2531 __be16 protocol
= skb
->protocol
;
2532 netdev_features_t features
= skb
->dev
->features
;
2534 if (skb_shinfo(skb
)->gso_segs
> skb
->dev
->gso_max_segs
)
2535 features
&= ~NETIF_F_GSO_MASK
;
2537 if (protocol
== htons(ETH_P_8021Q
) || protocol
== htons(ETH_P_8021AD
)) {
2538 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
2539 protocol
= veh
->h_vlan_encapsulated_proto
;
2540 } else if (!vlan_tx_tag_present(skb
)) {
2541 return harmonize_features(skb
, features
);
2544 features
= netdev_intersect_features(features
,
2545 skb
->dev
->vlan_features
|
2546 NETIF_F_HW_VLAN_CTAG_TX
|
2547 NETIF_F_HW_VLAN_STAG_TX
);
2549 if (protocol
== htons(ETH_P_8021Q
) || protocol
== htons(ETH_P_8021AD
))
2550 features
= netdev_intersect_features(features
,
2555 NETIF_F_HW_VLAN_CTAG_TX
|
2556 NETIF_F_HW_VLAN_STAG_TX
);
2558 return harmonize_features(skb
, features
);
2560 EXPORT_SYMBOL(netif_skb_features
);
2562 static int xmit_one(struct sk_buff
*skb
, struct net_device
*dev
,
2563 struct netdev_queue
*txq
, bool more
)
2568 if (!list_empty(&ptype_all
))
2569 dev_queue_xmit_nit(skb
, dev
);
2572 trace_net_dev_start_xmit(skb
, dev
);
2573 rc
= netdev_start_xmit(skb
, dev
, txq
, more
);
2574 trace_net_dev_xmit(skb
, rc
, dev
, len
);
2579 struct sk_buff
*dev_hard_start_xmit(struct sk_buff
*first
, struct net_device
*dev
,
2580 struct netdev_queue
*txq
, int *ret
)
2582 struct sk_buff
*skb
= first
;
2583 int rc
= NETDEV_TX_OK
;
2586 struct sk_buff
*next
= skb
->next
;
2589 rc
= xmit_one(skb
, dev
, txq
, next
!= NULL
);
2590 if (unlikely(!dev_xmit_complete(rc
))) {
2596 if (netif_xmit_stopped(txq
) && skb
) {
2597 rc
= NETDEV_TX_BUSY
;
2607 struct sk_buff
*validate_xmit_vlan(struct sk_buff
*skb
, netdev_features_t features
)
2609 if (vlan_tx_tag_present(skb
) &&
2610 !vlan_hw_offload_capable(features
, skb
->vlan_proto
)) {
2611 skb
= __vlan_put_tag(skb
, skb
->vlan_proto
,
2612 vlan_tx_tag_get(skb
));
2619 struct sk_buff
*validate_xmit_skb(struct sk_buff
*skb
, struct net_device
*dev
)
2621 netdev_features_t features
;
2626 /* If device doesn't need skb->dst, release it right now while
2627 * its hot in this cpu cache
2629 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2632 features
= netif_skb_features(skb
);
2633 skb
= validate_xmit_vlan(skb
, features
);
2637 /* If encapsulation offload request, verify we are testing
2638 * hardware encapsulation features instead of standard
2639 * features for the netdev
2641 if (skb
->encapsulation
)
2642 features
&= dev
->hw_enc_features
;
2644 if (netif_needs_gso(skb
, features
)) {
2645 struct sk_buff
*segs
;
2647 segs
= skb_gso_segment(skb
, features
);
2653 if (skb_needs_linearize(skb
, features
) &&
2654 __skb_linearize(skb
))
2657 /* If packet is not checksummed and device does not
2658 * support checksumming for this protocol, complete
2659 * checksumming here.
2661 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2662 if (skb
->encapsulation
)
2663 skb_set_inner_transport_header(skb
,
2664 skb_checksum_start_offset(skb
));
2666 skb_set_transport_header(skb
,
2667 skb_checksum_start_offset(skb
));
2668 if (!(features
& NETIF_F_ALL_CSUM
) &&
2669 skb_checksum_help(skb
))
2682 static void qdisc_pkt_len_init(struct sk_buff
*skb
)
2684 const struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
2686 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
2688 /* To get more precise estimation of bytes sent on wire,
2689 * we add to pkt_len the headers size of all segments
2691 if (shinfo
->gso_size
) {
2692 unsigned int hdr_len
;
2693 u16 gso_segs
= shinfo
->gso_segs
;
2695 /* mac layer + network layer */
2696 hdr_len
= skb_transport_header(skb
) - skb_mac_header(skb
);
2698 /* + transport layer */
2699 if (likely(shinfo
->gso_type
& (SKB_GSO_TCPV4
| SKB_GSO_TCPV6
)))
2700 hdr_len
+= tcp_hdrlen(skb
);
2702 hdr_len
+= sizeof(struct udphdr
);
2704 if (shinfo
->gso_type
& SKB_GSO_DODGY
)
2705 gso_segs
= DIV_ROUND_UP(skb
->len
- hdr_len
,
2708 qdisc_skb_cb(skb
)->pkt_len
+= (gso_segs
- 1) * hdr_len
;
2712 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2713 struct net_device
*dev
,
2714 struct netdev_queue
*txq
)
2716 spinlock_t
*root_lock
= qdisc_lock(q
);
2720 qdisc_pkt_len_init(skb
);
2721 qdisc_calculate_pkt_len(skb
, q
);
2723 * Heuristic to force contended enqueues to serialize on a
2724 * separate lock before trying to get qdisc main lock.
2725 * This permits __QDISC___STATE_RUNNING owner to get the lock more
2726 * often and dequeue packets faster.
2728 contended
= qdisc_is_running(q
);
2729 if (unlikely(contended
))
2730 spin_lock(&q
->busylock
);
2732 spin_lock(root_lock
);
2733 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2736 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2737 qdisc_run_begin(q
)) {
2739 * This is a work-conserving queue; there are no old skbs
2740 * waiting to be sent out; and the qdisc is not running -
2741 * xmit the skb directly.
2743 if (!(dev
->priv_flags
& IFF_XMIT_DST_RELEASE
))
2746 qdisc_bstats_update(q
, skb
);
2748 skb
= validate_xmit_skb(skb
, dev
);
2749 if (skb
&& sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
)) {
2750 if (unlikely(contended
)) {
2751 spin_unlock(&q
->busylock
);
2758 rc
= NET_XMIT_SUCCESS
;
2761 rc
= q
->enqueue(skb
, q
) & NET_XMIT_MASK
;
2762 if (qdisc_run_begin(q
)) {
2763 if (unlikely(contended
)) {
2764 spin_unlock(&q
->busylock
);
2770 spin_unlock(root_lock
);
2771 if (unlikely(contended
))
2772 spin_unlock(&q
->busylock
);
2776 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2777 static void skb_update_prio(struct sk_buff
*skb
)
2779 struct netprio_map
*map
= rcu_dereference_bh(skb
->dev
->priomap
);
2781 if (!skb
->priority
&& skb
->sk
&& map
) {
2782 unsigned int prioidx
= skb
->sk
->sk_cgrp_prioidx
;
2784 if (prioidx
< map
->priomap_len
)
2785 skb
->priority
= map
->priomap
[prioidx
];
2789 #define skb_update_prio(skb)
2792 static DEFINE_PER_CPU(int, xmit_recursion
);
2793 #define RECURSION_LIMIT 10
2796 * dev_loopback_xmit - loop back @skb
2797 * @skb: buffer to transmit
2799 int dev_loopback_xmit(struct sk_buff
*skb
)
2801 skb_reset_mac_header(skb
);
2802 __skb_pull(skb
, skb_network_offset(skb
));
2803 skb
->pkt_type
= PACKET_LOOPBACK
;
2804 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
2805 WARN_ON(!skb_dst(skb
));
2810 EXPORT_SYMBOL(dev_loopback_xmit
);
2813 * __dev_queue_xmit - transmit a buffer
2814 * @skb: buffer to transmit
2815 * @accel_priv: private data used for L2 forwarding offload
2817 * Queue a buffer for transmission to a network device. The caller must
2818 * have set the device and priority and built the buffer before calling
2819 * this function. The function can be called from an interrupt.
2821 * A negative errno code is returned on a failure. A success does not
2822 * guarantee the frame will be transmitted as it may be dropped due
2823 * to congestion or traffic shaping.
2825 * -----------------------------------------------------------------------------------
2826 * I notice this method can also return errors from the queue disciplines,
2827 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2830 * Regardless of the return value, the skb is consumed, so it is currently
2831 * difficult to retry a send to this method. (You can bump the ref count
2832 * before sending to hold a reference for retry if you are careful.)
2834 * When calling this method, interrupts MUST be enabled. This is because
2835 * the BH enable code must have IRQs enabled so that it will not deadlock.
2838 static int __dev_queue_xmit(struct sk_buff
*skb
, void *accel_priv
)
2840 struct net_device
*dev
= skb
->dev
;
2841 struct netdev_queue
*txq
;
2845 skb_reset_mac_header(skb
);
2847 if (unlikely(skb_shinfo(skb
)->tx_flags
& SKBTX_SCHED_TSTAMP
))
2848 __skb_tstamp_tx(skb
, NULL
, skb
->sk
, SCM_TSTAMP_SCHED
);
2850 /* Disable soft irqs for various locks below. Also
2851 * stops preemption for RCU.
2855 skb_update_prio(skb
);
2857 txq
= netdev_pick_tx(dev
, skb
, accel_priv
);
2858 q
= rcu_dereference_bh(txq
->qdisc
);
2860 #ifdef CONFIG_NET_CLS_ACT
2861 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2863 trace_net_dev_queue(skb
);
2865 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2869 /* The device has no queue. Common case for software devices:
2870 loopback, all the sorts of tunnels...
2872 Really, it is unlikely that netif_tx_lock protection is necessary
2873 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2875 However, it is possible, that they rely on protection
2878 Check this and shot the lock. It is not prone from deadlocks.
2879 Either shot noqueue qdisc, it is even simpler 8)
2881 if (dev
->flags
& IFF_UP
) {
2882 int cpu
= smp_processor_id(); /* ok because BHs are off */
2884 if (txq
->xmit_lock_owner
!= cpu
) {
2886 if (__this_cpu_read(xmit_recursion
) > RECURSION_LIMIT
)
2887 goto recursion_alert
;
2889 skb
= validate_xmit_skb(skb
, dev
);
2893 HARD_TX_LOCK(dev
, txq
, cpu
);
2895 if (!netif_xmit_stopped(txq
)) {
2896 __this_cpu_inc(xmit_recursion
);
2897 skb
= dev_hard_start_xmit(skb
, dev
, txq
, &rc
);
2898 __this_cpu_dec(xmit_recursion
);
2899 if (dev_xmit_complete(rc
)) {
2900 HARD_TX_UNLOCK(dev
, txq
);
2904 HARD_TX_UNLOCK(dev
, txq
);
2905 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2908 /* Recursion is detected! It is possible,
2912 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2919 rcu_read_unlock_bh();
2921 atomic_long_inc(&dev
->tx_dropped
);
2922 kfree_skb_list(skb
);
2925 rcu_read_unlock_bh();
2929 int dev_queue_xmit(struct sk_buff
*skb
)
2931 return __dev_queue_xmit(skb
, NULL
);
2933 EXPORT_SYMBOL(dev_queue_xmit
);
2935 int dev_queue_xmit_accel(struct sk_buff
*skb
, void *accel_priv
)
2937 return __dev_queue_xmit(skb
, accel_priv
);
2939 EXPORT_SYMBOL(dev_queue_xmit_accel
);
2942 /*=======================================================================
2944 =======================================================================*/
2946 int netdev_max_backlog __read_mostly
= 1000;
2947 EXPORT_SYMBOL(netdev_max_backlog
);
2949 int netdev_tstamp_prequeue __read_mostly
= 1;
2950 int netdev_budget __read_mostly
= 300;
2951 int weight_p __read_mostly
= 64; /* old backlog weight */
2953 /* Called with irq disabled */
2954 static inline void ____napi_schedule(struct softnet_data
*sd
,
2955 struct napi_struct
*napi
)
2957 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
2958 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2963 /* One global table that all flow-based protocols share. */
2964 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
2965 EXPORT_SYMBOL(rps_sock_flow_table
);
2967 struct static_key rps_needed __read_mostly
;
2969 static struct rps_dev_flow
*
2970 set_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2971 struct rps_dev_flow
*rflow
, u16 next_cpu
)
2973 if (next_cpu
!= RPS_NO_CPU
) {
2974 #ifdef CONFIG_RFS_ACCEL
2975 struct netdev_rx_queue
*rxqueue
;
2976 struct rps_dev_flow_table
*flow_table
;
2977 struct rps_dev_flow
*old_rflow
;
2982 /* Should we steer this flow to a different hardware queue? */
2983 if (!skb_rx_queue_recorded(skb
) || !dev
->rx_cpu_rmap
||
2984 !(dev
->features
& NETIF_F_NTUPLE
))
2986 rxq_index
= cpu_rmap_lookup_index(dev
->rx_cpu_rmap
, next_cpu
);
2987 if (rxq_index
== skb_get_rx_queue(skb
))
2990 rxqueue
= dev
->_rx
+ rxq_index
;
2991 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2994 flow_id
= skb_get_hash(skb
) & flow_table
->mask
;
2995 rc
= dev
->netdev_ops
->ndo_rx_flow_steer(dev
, skb
,
2996 rxq_index
, flow_id
);
3000 rflow
= &flow_table
->flows
[flow_id
];
3002 if (old_rflow
->filter
== rflow
->filter
)
3003 old_rflow
->filter
= RPS_NO_FILTER
;
3007 per_cpu(softnet_data
, next_cpu
).input_queue_head
;
3010 rflow
->cpu
= next_cpu
;
3015 * get_rps_cpu is called from netif_receive_skb and returns the target
3016 * CPU from the RPS map of the receiving queue for a given skb.
3017 * rcu_read_lock must be held on entry.
3019 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
3020 struct rps_dev_flow
**rflowp
)
3022 struct netdev_rx_queue
*rxqueue
;
3023 struct rps_map
*map
;
3024 struct rps_dev_flow_table
*flow_table
;
3025 struct rps_sock_flow_table
*sock_flow_table
;
3030 if (skb_rx_queue_recorded(skb
)) {
3031 u16 index
= skb_get_rx_queue(skb
);
3032 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
3033 WARN_ONCE(dev
->real_num_rx_queues
> 1,
3034 "%s received packet on queue %u, but number "
3035 "of RX queues is %u\n",
3036 dev
->name
, index
, dev
->real_num_rx_queues
);
3039 rxqueue
= dev
->_rx
+ index
;
3043 map
= rcu_dereference(rxqueue
->rps_map
);
3045 if (map
->len
== 1 &&
3046 !rcu_access_pointer(rxqueue
->rps_flow_table
)) {
3047 tcpu
= map
->cpus
[0];
3048 if (cpu_online(tcpu
))
3052 } else if (!rcu_access_pointer(rxqueue
->rps_flow_table
)) {
3056 skb_reset_network_header(skb
);
3057 hash
= skb_get_hash(skb
);
3061 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3062 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
3063 if (flow_table
&& sock_flow_table
) {
3065 struct rps_dev_flow
*rflow
;
3067 rflow
= &flow_table
->flows
[hash
& flow_table
->mask
];
3070 next_cpu
= sock_flow_table
->ents
[hash
& sock_flow_table
->mask
];
3073 * If the desired CPU (where last recvmsg was done) is
3074 * different from current CPU (one in the rx-queue flow
3075 * table entry), switch if one of the following holds:
3076 * - Current CPU is unset (equal to RPS_NO_CPU).
3077 * - Current CPU is offline.
3078 * - The current CPU's queue tail has advanced beyond the
3079 * last packet that was enqueued using this table entry.
3080 * This guarantees that all previous packets for the flow
3081 * have been dequeued, thus preserving in order delivery.
3083 if (unlikely(tcpu
!= next_cpu
) &&
3084 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
3085 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
3086 rflow
->last_qtail
)) >= 0)) {
3088 rflow
= set_rps_cpu(dev
, skb
, rflow
, next_cpu
);
3091 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
3099 tcpu
= map
->cpus
[reciprocal_scale(hash
, map
->len
)];
3100 if (cpu_online(tcpu
)) {
3110 #ifdef CONFIG_RFS_ACCEL
3113 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3114 * @dev: Device on which the filter was set
3115 * @rxq_index: RX queue index
3116 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3117 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3119 * Drivers that implement ndo_rx_flow_steer() should periodically call
3120 * this function for each installed filter and remove the filters for
3121 * which it returns %true.
3123 bool rps_may_expire_flow(struct net_device
*dev
, u16 rxq_index
,
3124 u32 flow_id
, u16 filter_id
)
3126 struct netdev_rx_queue
*rxqueue
= dev
->_rx
+ rxq_index
;
3127 struct rps_dev_flow_table
*flow_table
;
3128 struct rps_dev_flow
*rflow
;
3133 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3134 if (flow_table
&& flow_id
<= flow_table
->mask
) {
3135 rflow
= &flow_table
->flows
[flow_id
];
3136 cpu
= ACCESS_ONCE(rflow
->cpu
);
3137 if (rflow
->filter
== filter_id
&& cpu
!= RPS_NO_CPU
&&
3138 ((int)(per_cpu(softnet_data
, cpu
).input_queue_head
-
3139 rflow
->last_qtail
) <
3140 (int)(10 * flow_table
->mask
)))
3146 EXPORT_SYMBOL(rps_may_expire_flow
);
3148 #endif /* CONFIG_RFS_ACCEL */
3150 /* Called from hardirq (IPI) context */
3151 static void rps_trigger_softirq(void *data
)
3153 struct softnet_data
*sd
= data
;
3155 ____napi_schedule(sd
, &sd
->backlog
);
3159 #endif /* CONFIG_RPS */
3162 * Check if this softnet_data structure is another cpu one
3163 * If yes, queue it to our IPI list and return 1
3166 static int rps_ipi_queued(struct softnet_data
*sd
)
3169 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
3172 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
3173 mysd
->rps_ipi_list
= sd
;
3175 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3178 #endif /* CONFIG_RPS */
3182 #ifdef CONFIG_NET_FLOW_LIMIT
3183 int netdev_flow_limit_table_len __read_mostly
= (1 << 12);
3186 static bool skb_flow_limit(struct sk_buff
*skb
, unsigned int qlen
)
3188 #ifdef CONFIG_NET_FLOW_LIMIT
3189 struct sd_flow_limit
*fl
;
3190 struct softnet_data
*sd
;
3191 unsigned int old_flow
, new_flow
;
3193 if (qlen
< (netdev_max_backlog
>> 1))
3196 sd
= &__get_cpu_var(softnet_data
);
3199 fl
= rcu_dereference(sd
->flow_limit
);
3201 new_flow
= skb_get_hash(skb
) & (fl
->num_buckets
- 1);
3202 old_flow
= fl
->history
[fl
->history_head
];
3203 fl
->history
[fl
->history_head
] = new_flow
;
3206 fl
->history_head
&= FLOW_LIMIT_HISTORY
- 1;
3208 if (likely(fl
->buckets
[old_flow
]))
3209 fl
->buckets
[old_flow
]--;
3211 if (++fl
->buckets
[new_flow
] > (FLOW_LIMIT_HISTORY
>> 1)) {
3223 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3224 * queue (may be a remote CPU queue).
3226 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
3227 unsigned int *qtail
)
3229 struct softnet_data
*sd
;
3230 unsigned long flags
;
3233 sd
= &per_cpu(softnet_data
, cpu
);
3235 local_irq_save(flags
);
3238 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3239 if (qlen
<= netdev_max_backlog
&& !skb_flow_limit(skb
, qlen
)) {
3240 if (skb_queue_len(&sd
->input_pkt_queue
)) {
3242 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
3243 input_queue_tail_incr_save(sd
, qtail
);
3245 local_irq_restore(flags
);
3246 return NET_RX_SUCCESS
;
3249 /* Schedule NAPI for backlog device
3250 * We can use non atomic operation since we own the queue lock
3252 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
3253 if (!rps_ipi_queued(sd
))
3254 ____napi_schedule(sd
, &sd
->backlog
);
3262 local_irq_restore(flags
);
3264 atomic_long_inc(&skb
->dev
->rx_dropped
);
3269 static int netif_rx_internal(struct sk_buff
*skb
)
3273 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3275 trace_netif_rx(skb
);
3277 if (static_key_false(&rps_needed
)) {
3278 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3284 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3286 cpu
= smp_processor_id();
3288 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3296 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
3303 * netif_rx - post buffer to the network code
3304 * @skb: buffer to post
3306 * This function receives a packet from a device driver and queues it for
3307 * the upper (protocol) levels to process. It always succeeds. The buffer
3308 * may be dropped during processing for congestion control or by the
3312 * NET_RX_SUCCESS (no congestion)
3313 * NET_RX_DROP (packet was dropped)
3317 int netif_rx(struct sk_buff
*skb
)
3319 trace_netif_rx_entry(skb
);
3321 return netif_rx_internal(skb
);
3323 EXPORT_SYMBOL(netif_rx
);
3325 int netif_rx_ni(struct sk_buff
*skb
)
3329 trace_netif_rx_ni_entry(skb
);
3332 err
= netif_rx_internal(skb
);
3333 if (local_softirq_pending())
3339 EXPORT_SYMBOL(netif_rx_ni
);
3341 static void net_tx_action(struct softirq_action
*h
)
3343 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3345 if (sd
->completion_queue
) {
3346 struct sk_buff
*clist
;
3348 local_irq_disable();
3349 clist
= sd
->completion_queue
;
3350 sd
->completion_queue
= NULL
;
3354 struct sk_buff
*skb
= clist
;
3355 clist
= clist
->next
;
3357 WARN_ON(atomic_read(&skb
->users
));
3358 if (likely(get_kfree_skb_cb(skb
)->reason
== SKB_REASON_CONSUMED
))
3359 trace_consume_skb(skb
);
3361 trace_kfree_skb(skb
, net_tx_action
);
3366 if (sd
->output_queue
) {
3369 local_irq_disable();
3370 head
= sd
->output_queue
;
3371 sd
->output_queue
= NULL
;
3372 sd
->output_queue_tailp
= &sd
->output_queue
;
3376 struct Qdisc
*q
= head
;
3377 spinlock_t
*root_lock
;
3379 head
= head
->next_sched
;
3381 root_lock
= qdisc_lock(q
);
3382 if (spin_trylock(root_lock
)) {
3383 smp_mb__before_atomic();
3384 clear_bit(__QDISC_STATE_SCHED
,
3387 spin_unlock(root_lock
);
3389 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
3391 __netif_reschedule(q
);
3393 smp_mb__before_atomic();
3394 clear_bit(__QDISC_STATE_SCHED
,
3402 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3403 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3404 /* This hook is defined here for ATM LANE */
3405 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
3406 unsigned char *addr
) __read_mostly
;
3407 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
3410 #ifdef CONFIG_NET_CLS_ACT
3411 /* TODO: Maybe we should just force sch_ingress to be compiled in
3412 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3413 * a compare and 2 stores extra right now if we dont have it on
3414 * but have CONFIG_NET_CLS_ACT
3415 * NOTE: This doesn't stop any functionality; if you dont have
3416 * the ingress scheduler, you just can't add policies on ingress.
3419 static int ing_filter(struct sk_buff
*skb
, struct netdev_queue
*rxq
)
3421 struct net_device
*dev
= skb
->dev
;
3422 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
3423 int result
= TC_ACT_OK
;
3426 if (unlikely(MAX_RED_LOOP
< ttl
++)) {
3427 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3428 skb
->skb_iif
, dev
->ifindex
);
3432 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
3433 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
3436 if (q
!= &noop_qdisc
) {
3437 spin_lock(qdisc_lock(q
));
3438 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
3439 result
= qdisc_enqueue_root(skb
, q
);
3440 spin_unlock(qdisc_lock(q
));
3446 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
3447 struct packet_type
**pt_prev
,
3448 int *ret
, struct net_device
*orig_dev
)
3450 struct netdev_queue
*rxq
= rcu_dereference(skb
->dev
->ingress_queue
);
3452 if (!rxq
|| rxq
->qdisc
== &noop_qdisc
)
3456 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
3460 switch (ing_filter(skb
, rxq
)) {
3474 * netdev_rx_handler_register - register receive handler
3475 * @dev: device to register a handler for
3476 * @rx_handler: receive handler to register
3477 * @rx_handler_data: data pointer that is used by rx handler
3479 * Register a receive handler for a device. This handler will then be
3480 * called from __netif_receive_skb. A negative errno code is returned
3483 * The caller must hold the rtnl_mutex.
3485 * For a general description of rx_handler, see enum rx_handler_result.
3487 int netdev_rx_handler_register(struct net_device
*dev
,
3488 rx_handler_func_t
*rx_handler
,
3489 void *rx_handler_data
)
3493 if (dev
->rx_handler
)
3496 /* Note: rx_handler_data must be set before rx_handler */
3497 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
3498 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
3502 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
3505 * netdev_rx_handler_unregister - unregister receive handler
3506 * @dev: device to unregister a handler from
3508 * Unregister a receive handler from a device.
3510 * The caller must hold the rtnl_mutex.
3512 void netdev_rx_handler_unregister(struct net_device
*dev
)
3516 RCU_INIT_POINTER(dev
->rx_handler
, NULL
);
3517 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3518 * section has a guarantee to see a non NULL rx_handler_data
3522 RCU_INIT_POINTER(dev
->rx_handler_data
, NULL
);
3524 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
3527 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3528 * the special handling of PFMEMALLOC skbs.
3530 static bool skb_pfmemalloc_protocol(struct sk_buff
*skb
)
3532 switch (skb
->protocol
) {
3533 case htons(ETH_P_ARP
):
3534 case htons(ETH_P_IP
):
3535 case htons(ETH_P_IPV6
):
3536 case htons(ETH_P_8021Q
):
3537 case htons(ETH_P_8021AD
):
3544 static int __netif_receive_skb_core(struct sk_buff
*skb
, bool pfmemalloc
)
3546 struct packet_type
*ptype
, *pt_prev
;
3547 rx_handler_func_t
*rx_handler
;
3548 struct net_device
*orig_dev
;
3549 struct net_device
*null_or_dev
;
3550 bool deliver_exact
= false;
3551 int ret
= NET_RX_DROP
;
3554 net_timestamp_check(!netdev_tstamp_prequeue
, skb
);
3556 trace_netif_receive_skb(skb
);
3558 orig_dev
= skb
->dev
;
3560 skb_reset_network_header(skb
);
3561 if (!skb_transport_header_was_set(skb
))
3562 skb_reset_transport_header(skb
);
3563 skb_reset_mac_len(skb
);
3570 skb
->skb_iif
= skb
->dev
->ifindex
;
3572 __this_cpu_inc(softnet_data
.processed
);
3574 if (skb
->protocol
== cpu_to_be16(ETH_P_8021Q
) ||
3575 skb
->protocol
== cpu_to_be16(ETH_P_8021AD
)) {
3576 skb
= skb_vlan_untag(skb
);
3581 #ifdef CONFIG_NET_CLS_ACT
3582 if (skb
->tc_verd
& TC_NCLS
) {
3583 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
3591 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
3592 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
3594 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3600 #ifdef CONFIG_NET_CLS_ACT
3601 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
3607 if (pfmemalloc
&& !skb_pfmemalloc_protocol(skb
))
3610 if (vlan_tx_tag_present(skb
)) {
3612 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3615 if (vlan_do_receive(&skb
))
3617 else if (unlikely(!skb
))
3621 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
3624 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3627 switch (rx_handler(&skb
)) {
3628 case RX_HANDLER_CONSUMED
:
3629 ret
= NET_RX_SUCCESS
;
3631 case RX_HANDLER_ANOTHER
:
3633 case RX_HANDLER_EXACT
:
3634 deliver_exact
= true;
3635 case RX_HANDLER_PASS
:
3642 if (unlikely(vlan_tx_tag_present(skb
))) {
3643 if (vlan_tx_tag_get_id(skb
))
3644 skb
->pkt_type
= PACKET_OTHERHOST
;
3645 /* Note: we might in the future use prio bits
3646 * and set skb->priority like in vlan_do_receive()
3647 * For the time being, just ignore Priority Code Point
3652 /* deliver only exact match when indicated */
3653 null_or_dev
= deliver_exact
? skb
->dev
: NULL
;
3655 type
= skb
->protocol
;
3656 list_for_each_entry_rcu(ptype
,
3657 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
3658 if (ptype
->type
== type
&&
3659 (ptype
->dev
== null_or_dev
|| ptype
->dev
== skb
->dev
||
3660 ptype
->dev
== orig_dev
)) {
3662 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3668 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
3671 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
3674 atomic_long_inc(&skb
->dev
->rx_dropped
);
3676 /* Jamal, now you will not able to escape explaining
3677 * me how you were going to use this. :-)
3687 static int __netif_receive_skb(struct sk_buff
*skb
)
3691 if (sk_memalloc_socks() && skb_pfmemalloc(skb
)) {
3692 unsigned long pflags
= current
->flags
;
3695 * PFMEMALLOC skbs are special, they should
3696 * - be delivered to SOCK_MEMALLOC sockets only
3697 * - stay away from userspace
3698 * - have bounded memory usage
3700 * Use PF_MEMALLOC as this saves us from propagating the allocation
3701 * context down to all allocation sites.
3703 current
->flags
|= PF_MEMALLOC
;
3704 ret
= __netif_receive_skb_core(skb
, true);
3705 tsk_restore_flags(current
, pflags
, PF_MEMALLOC
);
3707 ret
= __netif_receive_skb_core(skb
, false);
3712 static int netif_receive_skb_internal(struct sk_buff
*skb
)
3714 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3716 if (skb_defer_rx_timestamp(skb
))
3717 return NET_RX_SUCCESS
;
3720 if (static_key_false(&rps_needed
)) {
3721 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3726 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3729 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3736 return __netif_receive_skb(skb
);
3740 * netif_receive_skb - process receive buffer from network
3741 * @skb: buffer to process
3743 * netif_receive_skb() is the main receive data processing function.
3744 * It always succeeds. The buffer may be dropped during processing
3745 * for congestion control or by the protocol layers.
3747 * This function may only be called from softirq context and interrupts
3748 * should be enabled.
3750 * Return values (usually ignored):
3751 * NET_RX_SUCCESS: no congestion
3752 * NET_RX_DROP: packet was dropped
3754 int netif_receive_skb(struct sk_buff
*skb
)
3756 trace_netif_receive_skb_entry(skb
);
3758 return netif_receive_skb_internal(skb
);
3760 EXPORT_SYMBOL(netif_receive_skb
);
3762 /* Network device is going away, flush any packets still pending
3763 * Called with irqs disabled.
3765 static void flush_backlog(void *arg
)
3767 struct net_device
*dev
= arg
;
3768 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3769 struct sk_buff
*skb
, *tmp
;
3772 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3773 if (skb
->dev
== dev
) {
3774 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3776 input_queue_head_incr(sd
);
3781 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3782 if (skb
->dev
== dev
) {
3783 __skb_unlink(skb
, &sd
->process_queue
);
3785 input_queue_head_incr(sd
);
3790 static int napi_gro_complete(struct sk_buff
*skb
)
3792 struct packet_offload
*ptype
;
3793 __be16 type
= skb
->protocol
;
3794 struct list_head
*head
= &offload_base
;
3797 BUILD_BUG_ON(sizeof(struct napi_gro_cb
) > sizeof(skb
->cb
));
3799 if (NAPI_GRO_CB(skb
)->count
== 1) {
3800 skb_shinfo(skb
)->gso_size
= 0;
3805 list_for_each_entry_rcu(ptype
, head
, list
) {
3806 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_complete
)
3809 err
= ptype
->callbacks
.gro_complete(skb
, 0);
3815 WARN_ON(&ptype
->list
== head
);
3817 return NET_RX_SUCCESS
;
3821 return netif_receive_skb_internal(skb
);
3824 /* napi->gro_list contains packets ordered by age.
3825 * youngest packets at the head of it.
3826 * Complete skbs in reverse order to reduce latencies.
3828 void napi_gro_flush(struct napi_struct
*napi
, bool flush_old
)
3830 struct sk_buff
*skb
, *prev
= NULL
;
3832 /* scan list and build reverse chain */
3833 for (skb
= napi
->gro_list
; skb
!= NULL
; skb
= skb
->next
) {
3838 for (skb
= prev
; skb
; skb
= prev
) {
3841 if (flush_old
&& NAPI_GRO_CB(skb
)->age
== jiffies
)
3845 napi_gro_complete(skb
);
3849 napi
->gro_list
= NULL
;
3851 EXPORT_SYMBOL(napi_gro_flush
);
3853 static void gro_list_prepare(struct napi_struct
*napi
, struct sk_buff
*skb
)
3856 unsigned int maclen
= skb
->dev
->hard_header_len
;
3857 u32 hash
= skb_get_hash_raw(skb
);
3859 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3860 unsigned long diffs
;
3862 NAPI_GRO_CB(p
)->flush
= 0;
3864 if (hash
!= skb_get_hash_raw(p
)) {
3865 NAPI_GRO_CB(p
)->same_flow
= 0;
3869 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
3870 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
3871 if (maclen
== ETH_HLEN
)
3872 diffs
|= compare_ether_header(skb_mac_header(p
),
3873 skb_mac_header(skb
));
3875 diffs
= memcmp(skb_mac_header(p
),
3876 skb_mac_header(skb
),
3878 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
3882 static void skb_gro_reset_offset(struct sk_buff
*skb
)
3884 const struct skb_shared_info
*pinfo
= skb_shinfo(skb
);
3885 const skb_frag_t
*frag0
= &pinfo
->frags
[0];
3887 NAPI_GRO_CB(skb
)->data_offset
= 0;
3888 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3889 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3891 if (skb_mac_header(skb
) == skb_tail_pointer(skb
) &&
3893 !PageHighMem(skb_frag_page(frag0
))) {
3894 NAPI_GRO_CB(skb
)->frag0
= skb_frag_address(frag0
);
3895 NAPI_GRO_CB(skb
)->frag0_len
= skb_frag_size(frag0
);
3899 static void gro_pull_from_frag0(struct sk_buff
*skb
, int grow
)
3901 struct skb_shared_info
*pinfo
= skb_shinfo(skb
);
3903 BUG_ON(skb
->end
- skb
->tail
< grow
);
3905 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3907 skb
->data_len
-= grow
;
3910 pinfo
->frags
[0].page_offset
+= grow
;
3911 skb_frag_size_sub(&pinfo
->frags
[0], grow
);
3913 if (unlikely(!skb_frag_size(&pinfo
->frags
[0]))) {
3914 skb_frag_unref(skb
, 0);
3915 memmove(pinfo
->frags
, pinfo
->frags
+ 1,
3916 --pinfo
->nr_frags
* sizeof(pinfo
->frags
[0]));
3920 static enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3922 struct sk_buff
**pp
= NULL
;
3923 struct packet_offload
*ptype
;
3924 __be16 type
= skb
->protocol
;
3925 struct list_head
*head
= &offload_base
;
3927 enum gro_result ret
;
3930 if (!(skb
->dev
->features
& NETIF_F_GRO
))
3933 if (skb_is_gso(skb
) || skb_has_frag_list(skb
) || skb
->csum_bad
)
3936 gro_list_prepare(napi
, skb
);
3939 list_for_each_entry_rcu(ptype
, head
, list
) {
3940 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_receive
)
3943 skb_set_network_header(skb
, skb_gro_offset(skb
));
3944 skb_reset_mac_len(skb
);
3945 NAPI_GRO_CB(skb
)->same_flow
= 0;
3946 NAPI_GRO_CB(skb
)->flush
= 0;
3947 NAPI_GRO_CB(skb
)->free
= 0;
3948 NAPI_GRO_CB(skb
)->udp_mark
= 0;
3950 /* Setup for GRO checksum validation */
3951 switch (skb
->ip_summed
) {
3952 case CHECKSUM_COMPLETE
:
3953 NAPI_GRO_CB(skb
)->csum
= skb
->csum
;
3954 NAPI_GRO_CB(skb
)->csum_valid
= 1;
3955 NAPI_GRO_CB(skb
)->csum_cnt
= 0;
3957 case CHECKSUM_UNNECESSARY
:
3958 NAPI_GRO_CB(skb
)->csum_cnt
= skb
->csum_level
+ 1;
3959 NAPI_GRO_CB(skb
)->csum_valid
= 0;
3962 NAPI_GRO_CB(skb
)->csum_cnt
= 0;
3963 NAPI_GRO_CB(skb
)->csum_valid
= 0;
3966 pp
= ptype
->callbacks
.gro_receive(&napi
->gro_list
, skb
);
3971 if (&ptype
->list
== head
)
3974 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3975 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3978 struct sk_buff
*nskb
= *pp
;
3982 napi_gro_complete(nskb
);
3989 if (NAPI_GRO_CB(skb
)->flush
)
3992 if (unlikely(napi
->gro_count
>= MAX_GRO_SKBS
)) {
3993 struct sk_buff
*nskb
= napi
->gro_list
;
3995 /* locate the end of the list to select the 'oldest' flow */
3996 while (nskb
->next
) {
4002 napi_gro_complete(nskb
);
4006 NAPI_GRO_CB(skb
)->count
= 1;
4007 NAPI_GRO_CB(skb
)->age
= jiffies
;
4008 NAPI_GRO_CB(skb
)->last
= skb
;
4009 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
4010 skb
->next
= napi
->gro_list
;
4011 napi
->gro_list
= skb
;
4015 grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
4017 gro_pull_from_frag0(skb
, grow
);
4026 struct packet_offload
*gro_find_receive_by_type(__be16 type
)
4028 struct list_head
*offload_head
= &offload_base
;
4029 struct packet_offload
*ptype
;
4031 list_for_each_entry_rcu(ptype
, offload_head
, list
) {
4032 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_receive
)
4038 EXPORT_SYMBOL(gro_find_receive_by_type
);
4040 struct packet_offload
*gro_find_complete_by_type(__be16 type
)
4042 struct list_head
*offload_head
= &offload_base
;
4043 struct packet_offload
*ptype
;
4045 list_for_each_entry_rcu(ptype
, offload_head
, list
) {
4046 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_complete
)
4052 EXPORT_SYMBOL(gro_find_complete_by_type
);
4054 static gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
4058 if (netif_receive_skb_internal(skb
))
4066 case GRO_MERGED_FREE
:
4067 if (NAPI_GRO_CB(skb
)->free
== NAPI_GRO_FREE_STOLEN_HEAD
)
4068 kmem_cache_free(skbuff_head_cache
, skb
);
4081 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
4083 trace_napi_gro_receive_entry(skb
);
4085 skb_gro_reset_offset(skb
);
4087 return napi_skb_finish(dev_gro_receive(napi
, skb
), skb
);
4089 EXPORT_SYMBOL(napi_gro_receive
);
4091 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
4093 __skb_pull(skb
, skb_headlen(skb
));
4094 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4095 skb_reserve(skb
, NET_SKB_PAD
+ NET_IP_ALIGN
- skb_headroom(skb
));
4097 skb
->dev
= napi
->dev
;
4099 skb
->encapsulation
= 0;
4100 skb_shinfo(skb
)->gso_type
= 0;
4101 skb
->truesize
= SKB_TRUESIZE(skb_end_offset(skb
));
4106 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
4108 struct sk_buff
*skb
= napi
->skb
;
4111 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
4116 EXPORT_SYMBOL(napi_get_frags
);
4118 static gro_result_t
napi_frags_finish(struct napi_struct
*napi
,
4119 struct sk_buff
*skb
,
4125 __skb_push(skb
, ETH_HLEN
);
4126 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
4127 if (ret
== GRO_NORMAL
&& netif_receive_skb_internal(skb
))
4132 case GRO_MERGED_FREE
:
4133 napi_reuse_skb(napi
, skb
);
4143 /* Upper GRO stack assumes network header starts at gro_offset=0
4144 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4145 * We copy ethernet header into skb->data to have a common layout.
4147 static struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
4149 struct sk_buff
*skb
= napi
->skb
;
4150 const struct ethhdr
*eth
;
4151 unsigned int hlen
= sizeof(*eth
);
4155 skb_reset_mac_header(skb
);
4156 skb_gro_reset_offset(skb
);
4158 eth
= skb_gro_header_fast(skb
, 0);
4159 if (unlikely(skb_gro_header_hard(skb
, hlen
))) {
4160 eth
= skb_gro_header_slow(skb
, hlen
, 0);
4161 if (unlikely(!eth
)) {
4162 napi_reuse_skb(napi
, skb
);
4166 gro_pull_from_frag0(skb
, hlen
);
4167 NAPI_GRO_CB(skb
)->frag0
+= hlen
;
4168 NAPI_GRO_CB(skb
)->frag0_len
-= hlen
;
4170 __skb_pull(skb
, hlen
);
4173 * This works because the only protocols we care about don't require
4175 * We'll fix it up properly in napi_frags_finish()
4177 skb
->protocol
= eth
->h_proto
;
4182 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
4184 struct sk_buff
*skb
= napi_frags_skb(napi
);
4189 trace_napi_gro_frags_entry(skb
);
4191 return napi_frags_finish(napi
, skb
, dev_gro_receive(napi
, skb
));
4193 EXPORT_SYMBOL(napi_gro_frags
);
4195 /* Compute the checksum from gro_offset and return the folded value
4196 * after adding in any pseudo checksum.
4198 __sum16
__skb_gro_checksum_complete(struct sk_buff
*skb
)
4203 wsum
= skb_checksum(skb
, skb_gro_offset(skb
), skb_gro_len(skb
), 0);
4205 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4206 sum
= csum_fold(csum_add(NAPI_GRO_CB(skb
)->csum
, wsum
));
4208 if (unlikely(skb
->ip_summed
== CHECKSUM_COMPLETE
) &&
4209 !skb
->csum_complete_sw
)
4210 netdev_rx_csum_fault(skb
->dev
);
4213 NAPI_GRO_CB(skb
)->csum
= wsum
;
4214 NAPI_GRO_CB(skb
)->csum_valid
= 1;
4218 EXPORT_SYMBOL(__skb_gro_checksum_complete
);
4221 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4222 * Note: called with local irq disabled, but exits with local irq enabled.
4224 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
4227 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
4230 sd
->rps_ipi_list
= NULL
;
4234 /* Send pending IPI's to kick RPS processing on remote cpus. */
4236 struct softnet_data
*next
= remsd
->rps_ipi_next
;
4238 if (cpu_online(remsd
->cpu
))
4239 smp_call_function_single_async(remsd
->cpu
,
4248 static int process_backlog(struct napi_struct
*napi
, int quota
)
4251 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
4254 /* Check if we have pending ipi, its better to send them now,
4255 * not waiting net_rx_action() end.
4257 if (sd
->rps_ipi_list
) {
4258 local_irq_disable();
4259 net_rps_action_and_irq_enable(sd
);
4262 napi
->weight
= weight_p
;
4263 local_irq_disable();
4265 struct sk_buff
*skb
;
4267 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
4269 __netif_receive_skb(skb
);
4270 local_irq_disable();
4271 input_queue_head_incr(sd
);
4272 if (++work
>= quota
) {
4279 if (skb_queue_empty(&sd
->input_pkt_queue
)) {
4281 * Inline a custom version of __napi_complete().
4282 * only current cpu owns and manipulates this napi,
4283 * and NAPI_STATE_SCHED is the only possible flag set
4285 * We can use a plain write instead of clear_bit(),
4286 * and we dont need an smp_mb() memory barrier.
4288 list_del(&napi
->poll_list
);
4295 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
4296 &sd
->process_queue
);
4305 * __napi_schedule - schedule for receive
4306 * @n: entry to schedule
4308 * The entry's receive function will be scheduled to run
4310 void __napi_schedule(struct napi_struct
*n
)
4312 unsigned long flags
;
4314 local_irq_save(flags
);
4315 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
4316 local_irq_restore(flags
);
4318 EXPORT_SYMBOL(__napi_schedule
);
4320 void __napi_complete(struct napi_struct
*n
)
4322 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
4323 BUG_ON(n
->gro_list
);
4325 list_del(&n
->poll_list
);
4326 smp_mb__before_atomic();
4327 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
4329 EXPORT_SYMBOL(__napi_complete
);
4331 void napi_complete(struct napi_struct
*n
)
4333 unsigned long flags
;
4336 * don't let napi dequeue from the cpu poll list
4337 * just in case its running on a different cpu
4339 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
4342 napi_gro_flush(n
, false);
4343 local_irq_save(flags
);
4345 local_irq_restore(flags
);
4347 EXPORT_SYMBOL(napi_complete
);
4349 /* must be called under rcu_read_lock(), as we dont take a reference */
4350 struct napi_struct
*napi_by_id(unsigned int napi_id
)
4352 unsigned int hash
= napi_id
% HASH_SIZE(napi_hash
);
4353 struct napi_struct
*napi
;
4355 hlist_for_each_entry_rcu(napi
, &napi_hash
[hash
], napi_hash_node
)
4356 if (napi
->napi_id
== napi_id
)
4361 EXPORT_SYMBOL_GPL(napi_by_id
);
4363 void napi_hash_add(struct napi_struct
*napi
)
4365 if (!test_and_set_bit(NAPI_STATE_HASHED
, &napi
->state
)) {
4367 spin_lock(&napi_hash_lock
);
4369 /* 0 is not a valid id, we also skip an id that is taken
4370 * we expect both events to be extremely rare
4373 while (!napi
->napi_id
) {
4374 napi
->napi_id
= ++napi_gen_id
;
4375 if (napi_by_id(napi
->napi_id
))
4379 hlist_add_head_rcu(&napi
->napi_hash_node
,
4380 &napi_hash
[napi
->napi_id
% HASH_SIZE(napi_hash
)]);
4382 spin_unlock(&napi_hash_lock
);
4385 EXPORT_SYMBOL_GPL(napi_hash_add
);
4387 /* Warning : caller is responsible to make sure rcu grace period
4388 * is respected before freeing memory containing @napi
4390 void napi_hash_del(struct napi_struct
*napi
)
4392 spin_lock(&napi_hash_lock
);
4394 if (test_and_clear_bit(NAPI_STATE_HASHED
, &napi
->state
))
4395 hlist_del_rcu(&napi
->napi_hash_node
);
4397 spin_unlock(&napi_hash_lock
);
4399 EXPORT_SYMBOL_GPL(napi_hash_del
);
4401 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
4402 int (*poll
)(struct napi_struct
*, int), int weight
)
4404 INIT_LIST_HEAD(&napi
->poll_list
);
4405 napi
->gro_count
= 0;
4406 napi
->gro_list
= NULL
;
4409 if (weight
> NAPI_POLL_WEIGHT
)
4410 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4412 napi
->weight
= weight
;
4413 list_add(&napi
->dev_list
, &dev
->napi_list
);
4415 #ifdef CONFIG_NETPOLL
4416 spin_lock_init(&napi
->poll_lock
);
4417 napi
->poll_owner
= -1;
4419 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
4421 EXPORT_SYMBOL(netif_napi_add
);
4423 void netif_napi_del(struct napi_struct
*napi
)
4425 list_del_init(&napi
->dev_list
);
4426 napi_free_frags(napi
);
4428 kfree_skb_list(napi
->gro_list
);
4429 napi
->gro_list
= NULL
;
4430 napi
->gro_count
= 0;
4432 EXPORT_SYMBOL(netif_napi_del
);
4434 static void net_rx_action(struct softirq_action
*h
)
4436 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
4437 unsigned long time_limit
= jiffies
+ 2;
4438 int budget
= netdev_budget
;
4441 local_irq_disable();
4443 while (!list_empty(&sd
->poll_list
)) {
4444 struct napi_struct
*n
;
4447 /* If softirq window is exhuasted then punt.
4448 * Allow this to run for 2 jiffies since which will allow
4449 * an average latency of 1.5/HZ.
4451 if (unlikely(budget
<= 0 || time_after_eq(jiffies
, time_limit
)))
4456 /* Even though interrupts have been re-enabled, this
4457 * access is safe because interrupts can only add new
4458 * entries to the tail of this list, and only ->poll()
4459 * calls can remove this head entry from the list.
4461 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
4463 have
= netpoll_poll_lock(n
);
4467 /* This NAPI_STATE_SCHED test is for avoiding a race
4468 * with netpoll's poll_napi(). Only the entity which
4469 * obtains the lock and sees NAPI_STATE_SCHED set will
4470 * actually make the ->poll() call. Therefore we avoid
4471 * accidentally calling ->poll() when NAPI is not scheduled.
4474 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
4475 work
= n
->poll(n
, weight
);
4479 WARN_ON_ONCE(work
> weight
);
4483 local_irq_disable();
4485 /* Drivers must not modify the NAPI state if they
4486 * consume the entire weight. In such cases this code
4487 * still "owns" the NAPI instance and therefore can
4488 * move the instance around on the list at-will.
4490 if (unlikely(work
== weight
)) {
4491 if (unlikely(napi_disable_pending(n
))) {
4494 local_irq_disable();
4497 /* flush too old packets
4498 * If HZ < 1000, flush all packets.
4501 napi_gro_flush(n
, HZ
>= 1000);
4502 local_irq_disable();
4504 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
4508 netpoll_poll_unlock(have
);
4511 net_rps_action_and_irq_enable(sd
);
4513 #ifdef CONFIG_NET_DMA
4515 * There may not be any more sk_buffs coming right now, so push
4516 * any pending DMA copies to hardware
4518 dma_issue_pending_all();
4525 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
4529 struct netdev_adjacent
{
4530 struct net_device
*dev
;
4532 /* upper master flag, there can only be one master device per list */
4535 /* counter for the number of times this device was added to us */
4538 /* private field for the users */
4541 struct list_head list
;
4542 struct rcu_head rcu
;
4545 static struct netdev_adjacent
*__netdev_find_adj(struct net_device
*dev
,
4546 struct net_device
*adj_dev
,
4547 struct list_head
*adj_list
)
4549 struct netdev_adjacent
*adj
;
4551 list_for_each_entry(adj
, adj_list
, list
) {
4552 if (adj
->dev
== adj_dev
)
4559 * netdev_has_upper_dev - Check if device is linked to an upper device
4561 * @upper_dev: upper device to check
4563 * Find out if a device is linked to specified upper device and return true
4564 * in case it is. Note that this checks only immediate upper device,
4565 * not through a complete stack of devices. The caller must hold the RTNL lock.
4567 bool netdev_has_upper_dev(struct net_device
*dev
,
4568 struct net_device
*upper_dev
)
4572 return __netdev_find_adj(dev
, upper_dev
, &dev
->all_adj_list
.upper
);
4574 EXPORT_SYMBOL(netdev_has_upper_dev
);
4577 * netdev_has_any_upper_dev - Check if device is linked to some device
4580 * Find out if a device is linked to an upper device and return true in case
4581 * it is. The caller must hold the RTNL lock.
4583 static bool netdev_has_any_upper_dev(struct net_device
*dev
)
4587 return !list_empty(&dev
->all_adj_list
.upper
);
4591 * netdev_master_upper_dev_get - Get master upper device
4594 * Find a master upper device and return pointer to it or NULL in case
4595 * it's not there. The caller must hold the RTNL lock.
4597 struct net_device
*netdev_master_upper_dev_get(struct net_device
*dev
)
4599 struct netdev_adjacent
*upper
;
4603 if (list_empty(&dev
->adj_list
.upper
))
4606 upper
= list_first_entry(&dev
->adj_list
.upper
,
4607 struct netdev_adjacent
, list
);
4608 if (likely(upper
->master
))
4612 EXPORT_SYMBOL(netdev_master_upper_dev_get
);
4614 void *netdev_adjacent_get_private(struct list_head
*adj_list
)
4616 struct netdev_adjacent
*adj
;
4618 adj
= list_entry(adj_list
, struct netdev_adjacent
, list
);
4620 return adj
->private;
4622 EXPORT_SYMBOL(netdev_adjacent_get_private
);
4625 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
4627 * @iter: list_head ** of the current position
4629 * Gets the next device from the dev's upper list, starting from iter
4630 * position. The caller must hold RCU read lock.
4632 struct net_device
*netdev_upper_get_next_dev_rcu(struct net_device
*dev
,
4633 struct list_head
**iter
)
4635 struct netdev_adjacent
*upper
;
4637 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4639 upper
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
4641 if (&upper
->list
== &dev
->adj_list
.upper
)
4644 *iter
= &upper
->list
;
4648 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu
);
4651 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4653 * @iter: list_head ** of the current position
4655 * Gets the next device from the dev's upper list, starting from iter
4656 * position. The caller must hold RCU read lock.
4658 struct net_device
*netdev_all_upper_get_next_dev_rcu(struct net_device
*dev
,
4659 struct list_head
**iter
)
4661 struct netdev_adjacent
*upper
;
4663 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4665 upper
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
4667 if (&upper
->list
== &dev
->all_adj_list
.upper
)
4670 *iter
= &upper
->list
;
4674 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu
);
4677 * netdev_lower_get_next_private - Get the next ->private from the
4678 * lower neighbour list
4680 * @iter: list_head ** of the current position
4682 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4683 * list, starting from iter position. The caller must hold either hold the
4684 * RTNL lock or its own locking that guarantees that the neighbour lower
4685 * list will remain unchainged.
4687 void *netdev_lower_get_next_private(struct net_device
*dev
,
4688 struct list_head
**iter
)
4690 struct netdev_adjacent
*lower
;
4692 lower
= list_entry(*iter
, struct netdev_adjacent
, list
);
4694 if (&lower
->list
== &dev
->adj_list
.lower
)
4697 *iter
= lower
->list
.next
;
4699 return lower
->private;
4701 EXPORT_SYMBOL(netdev_lower_get_next_private
);
4704 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4705 * lower neighbour list, RCU
4708 * @iter: list_head ** of the current position
4710 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4711 * list, starting from iter position. The caller must hold RCU read lock.
4713 void *netdev_lower_get_next_private_rcu(struct net_device
*dev
,
4714 struct list_head
**iter
)
4716 struct netdev_adjacent
*lower
;
4718 WARN_ON_ONCE(!rcu_read_lock_held());
4720 lower
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
4722 if (&lower
->list
== &dev
->adj_list
.lower
)
4725 *iter
= &lower
->list
;
4727 return lower
->private;
4729 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu
);
4732 * netdev_lower_get_next - Get the next device from the lower neighbour
4735 * @iter: list_head ** of the current position
4737 * Gets the next netdev_adjacent from the dev's lower neighbour
4738 * list, starting from iter position. The caller must hold RTNL lock or
4739 * its own locking that guarantees that the neighbour lower
4740 * list will remain unchainged.
4742 void *netdev_lower_get_next(struct net_device
*dev
, struct list_head
**iter
)
4744 struct netdev_adjacent
*lower
;
4746 lower
= list_entry((*iter
)->next
, struct netdev_adjacent
, list
);
4748 if (&lower
->list
== &dev
->adj_list
.lower
)
4751 *iter
= &lower
->list
;
4755 EXPORT_SYMBOL(netdev_lower_get_next
);
4758 * netdev_lower_get_first_private_rcu - Get the first ->private from the
4759 * lower neighbour list, RCU
4763 * Gets the first netdev_adjacent->private from the dev's lower neighbour
4764 * list. The caller must hold RCU read lock.
4766 void *netdev_lower_get_first_private_rcu(struct net_device
*dev
)
4768 struct netdev_adjacent
*lower
;
4770 lower
= list_first_or_null_rcu(&dev
->adj_list
.lower
,
4771 struct netdev_adjacent
, list
);
4773 return lower
->private;
4776 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu
);
4779 * netdev_master_upper_dev_get_rcu - Get master upper device
4782 * Find a master upper device and return pointer to it or NULL in case
4783 * it's not there. The caller must hold the RCU read lock.
4785 struct net_device
*netdev_master_upper_dev_get_rcu(struct net_device
*dev
)
4787 struct netdev_adjacent
*upper
;
4789 upper
= list_first_or_null_rcu(&dev
->adj_list
.upper
,
4790 struct netdev_adjacent
, list
);
4791 if (upper
&& likely(upper
->master
))
4795 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu
);
4797 static int netdev_adjacent_sysfs_add(struct net_device
*dev
,
4798 struct net_device
*adj_dev
,
4799 struct list_head
*dev_list
)
4801 char linkname
[IFNAMSIZ
+7];
4802 sprintf(linkname
, dev_list
== &dev
->adj_list
.upper
?
4803 "upper_%s" : "lower_%s", adj_dev
->name
);
4804 return sysfs_create_link(&(dev
->dev
.kobj
), &(adj_dev
->dev
.kobj
),
4807 static void netdev_adjacent_sysfs_del(struct net_device
*dev
,
4809 struct list_head
*dev_list
)
4811 char linkname
[IFNAMSIZ
+7];
4812 sprintf(linkname
, dev_list
== &dev
->adj_list
.upper
?
4813 "upper_%s" : "lower_%s", name
);
4814 sysfs_remove_link(&(dev
->dev
.kobj
), linkname
);
4817 #define netdev_adjacent_is_neigh_list(dev, dev_list) \
4818 (dev_list == &dev->adj_list.upper || \
4819 dev_list == &dev->adj_list.lower)
4821 static int __netdev_adjacent_dev_insert(struct net_device
*dev
,
4822 struct net_device
*adj_dev
,
4823 struct list_head
*dev_list
,
4824 void *private, bool master
)
4826 struct netdev_adjacent
*adj
;
4829 adj
= __netdev_find_adj(dev
, adj_dev
, dev_list
);
4836 adj
= kmalloc(sizeof(*adj
), GFP_KERNEL
);
4841 adj
->master
= master
;
4843 adj
->private = private;
4846 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
4847 adj_dev
->name
, dev
->name
, adj_dev
->name
);
4849 if (netdev_adjacent_is_neigh_list(dev
, dev_list
)) {
4850 ret
= netdev_adjacent_sysfs_add(dev
, adj_dev
, dev_list
);
4855 /* Ensure that master link is always the first item in list. */
4857 ret
= sysfs_create_link(&(dev
->dev
.kobj
),
4858 &(adj_dev
->dev
.kobj
), "master");
4860 goto remove_symlinks
;
4862 list_add_rcu(&adj
->list
, dev_list
);
4864 list_add_tail_rcu(&adj
->list
, dev_list
);
4870 if (netdev_adjacent_is_neigh_list(dev
, dev_list
))
4871 netdev_adjacent_sysfs_del(dev
, adj_dev
->name
, dev_list
);
4879 static void __netdev_adjacent_dev_remove(struct net_device
*dev
,
4880 struct net_device
*adj_dev
,
4881 struct list_head
*dev_list
)
4883 struct netdev_adjacent
*adj
;
4885 adj
= __netdev_find_adj(dev
, adj_dev
, dev_list
);
4888 pr_err("tried to remove device %s from %s\n",
4889 dev
->name
, adj_dev
->name
);
4893 if (adj
->ref_nr
> 1) {
4894 pr_debug("%s to %s ref_nr-- = %d\n", dev
->name
, adj_dev
->name
,
4901 sysfs_remove_link(&(dev
->dev
.kobj
), "master");
4903 if (netdev_adjacent_is_neigh_list(dev
, dev_list
) &&
4904 net_eq(dev_net(dev
),dev_net(adj_dev
)))
4905 netdev_adjacent_sysfs_del(dev
, adj_dev
->name
, dev_list
);
4907 list_del_rcu(&adj
->list
);
4908 pr_debug("dev_put for %s, because link removed from %s to %s\n",
4909 adj_dev
->name
, dev
->name
, adj_dev
->name
);
4911 kfree_rcu(adj
, rcu
);
4914 static int __netdev_adjacent_dev_link_lists(struct net_device
*dev
,
4915 struct net_device
*upper_dev
,
4916 struct list_head
*up_list
,
4917 struct list_head
*down_list
,
4918 void *private, bool master
)
4922 ret
= __netdev_adjacent_dev_insert(dev
, upper_dev
, up_list
, private,
4927 ret
= __netdev_adjacent_dev_insert(upper_dev
, dev
, down_list
, private,
4930 __netdev_adjacent_dev_remove(dev
, upper_dev
, up_list
);
4937 static int __netdev_adjacent_dev_link(struct net_device
*dev
,
4938 struct net_device
*upper_dev
)
4940 return __netdev_adjacent_dev_link_lists(dev
, upper_dev
,
4941 &dev
->all_adj_list
.upper
,
4942 &upper_dev
->all_adj_list
.lower
,
4946 static void __netdev_adjacent_dev_unlink_lists(struct net_device
*dev
,
4947 struct net_device
*upper_dev
,
4948 struct list_head
*up_list
,
4949 struct list_head
*down_list
)
4951 __netdev_adjacent_dev_remove(dev
, upper_dev
, up_list
);
4952 __netdev_adjacent_dev_remove(upper_dev
, dev
, down_list
);
4955 static void __netdev_adjacent_dev_unlink(struct net_device
*dev
,
4956 struct net_device
*upper_dev
)
4958 __netdev_adjacent_dev_unlink_lists(dev
, upper_dev
,
4959 &dev
->all_adj_list
.upper
,
4960 &upper_dev
->all_adj_list
.lower
);
4963 static int __netdev_adjacent_dev_link_neighbour(struct net_device
*dev
,
4964 struct net_device
*upper_dev
,
4965 void *private, bool master
)
4967 int ret
= __netdev_adjacent_dev_link(dev
, upper_dev
);
4972 ret
= __netdev_adjacent_dev_link_lists(dev
, upper_dev
,
4973 &dev
->adj_list
.upper
,
4974 &upper_dev
->adj_list
.lower
,
4977 __netdev_adjacent_dev_unlink(dev
, upper_dev
);
4984 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device
*dev
,
4985 struct net_device
*upper_dev
)
4987 __netdev_adjacent_dev_unlink(dev
, upper_dev
);
4988 __netdev_adjacent_dev_unlink_lists(dev
, upper_dev
,
4989 &dev
->adj_list
.upper
,
4990 &upper_dev
->adj_list
.lower
);
4993 static int __netdev_upper_dev_link(struct net_device
*dev
,
4994 struct net_device
*upper_dev
, bool master
,
4997 struct netdev_adjacent
*i
, *j
, *to_i
, *to_j
;
5002 if (dev
== upper_dev
)
5005 /* To prevent loops, check if dev is not upper device to upper_dev. */
5006 if (__netdev_find_adj(upper_dev
, dev
, &upper_dev
->all_adj_list
.upper
))
5009 if (__netdev_find_adj(dev
, upper_dev
, &dev
->all_adj_list
.upper
))
5012 if (master
&& netdev_master_upper_dev_get(dev
))
5015 ret
= __netdev_adjacent_dev_link_neighbour(dev
, upper_dev
, private,
5020 /* Now that we linked these devs, make all the upper_dev's
5021 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5022 * versa, and don't forget the devices itself. All of these
5023 * links are non-neighbours.
5025 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5026 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
) {
5027 pr_debug("Interlinking %s with %s, non-neighbour\n",
5028 i
->dev
->name
, j
->dev
->name
);
5029 ret
= __netdev_adjacent_dev_link(i
->dev
, j
->dev
);
5035 /* add dev to every upper_dev's upper device */
5036 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
) {
5037 pr_debug("linking %s's upper device %s with %s\n",
5038 upper_dev
->name
, i
->dev
->name
, dev
->name
);
5039 ret
= __netdev_adjacent_dev_link(dev
, i
->dev
);
5041 goto rollback_upper_mesh
;
5044 /* add upper_dev to every dev's lower device */
5045 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5046 pr_debug("linking %s's lower device %s with %s\n", dev
->name
,
5047 i
->dev
->name
, upper_dev
->name
);
5048 ret
= __netdev_adjacent_dev_link(i
->dev
, upper_dev
);
5050 goto rollback_lower_mesh
;
5053 call_netdevice_notifiers(NETDEV_CHANGEUPPER
, dev
);
5056 rollback_lower_mesh
:
5058 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5061 __netdev_adjacent_dev_unlink(i
->dev
, upper_dev
);
5066 rollback_upper_mesh
:
5068 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
) {
5071 __netdev_adjacent_dev_unlink(dev
, i
->dev
);
5079 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5080 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
) {
5081 if (i
== to_i
&& j
== to_j
)
5083 __netdev_adjacent_dev_unlink(i
->dev
, j
->dev
);
5089 __netdev_adjacent_dev_unlink_neighbour(dev
, upper_dev
);
5095 * netdev_upper_dev_link - Add a link to the upper device
5097 * @upper_dev: new upper device
5099 * Adds a link to device which is upper to this one. The caller must hold
5100 * the RTNL lock. On a failure a negative errno code is returned.
5101 * On success the reference counts are adjusted and the function
5104 int netdev_upper_dev_link(struct net_device
*dev
,
5105 struct net_device
*upper_dev
)
5107 return __netdev_upper_dev_link(dev
, upper_dev
, false, NULL
);
5109 EXPORT_SYMBOL(netdev_upper_dev_link
);
5112 * netdev_master_upper_dev_link - Add a master link to the upper device
5114 * @upper_dev: new upper device
5116 * Adds a link to device which is upper to this one. In this case, only
5117 * one master upper device can be linked, although other non-master devices
5118 * might be linked as well. The caller must hold the RTNL lock.
5119 * On a failure a negative errno code is returned. On success the reference
5120 * counts are adjusted and the function returns zero.
5122 int netdev_master_upper_dev_link(struct net_device
*dev
,
5123 struct net_device
*upper_dev
)
5125 return __netdev_upper_dev_link(dev
, upper_dev
, true, NULL
);
5127 EXPORT_SYMBOL(netdev_master_upper_dev_link
);
5129 int netdev_master_upper_dev_link_private(struct net_device
*dev
,
5130 struct net_device
*upper_dev
,
5133 return __netdev_upper_dev_link(dev
, upper_dev
, true, private);
5135 EXPORT_SYMBOL(netdev_master_upper_dev_link_private
);
5138 * netdev_upper_dev_unlink - Removes a link to upper device
5140 * @upper_dev: new upper device
5142 * Removes a link to device which is upper to this one. The caller must hold
5145 void netdev_upper_dev_unlink(struct net_device
*dev
,
5146 struct net_device
*upper_dev
)
5148 struct netdev_adjacent
*i
, *j
;
5151 __netdev_adjacent_dev_unlink_neighbour(dev
, upper_dev
);
5153 /* Here is the tricky part. We must remove all dev's lower
5154 * devices from all upper_dev's upper devices and vice
5155 * versa, to maintain the graph relationship.
5157 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
)
5158 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
)
5159 __netdev_adjacent_dev_unlink(i
->dev
, j
->dev
);
5161 /* remove also the devices itself from lower/upper device
5164 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
)
5165 __netdev_adjacent_dev_unlink(i
->dev
, upper_dev
);
5167 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
)
5168 __netdev_adjacent_dev_unlink(dev
, i
->dev
);
5170 call_netdevice_notifiers(NETDEV_CHANGEUPPER
, dev
);
5172 EXPORT_SYMBOL(netdev_upper_dev_unlink
);
5174 void netdev_adjacent_add_links(struct net_device
*dev
)
5176 struct netdev_adjacent
*iter
;
5178 struct net
*net
= dev_net(dev
);
5180 list_for_each_entry(iter
, &dev
->adj_list
.upper
, list
) {
5181 if (!net_eq(net
,dev_net(iter
->dev
)))
5183 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
5184 &iter
->dev
->adj_list
.lower
);
5185 netdev_adjacent_sysfs_add(dev
, iter
->dev
,
5186 &dev
->adj_list
.upper
);
5189 list_for_each_entry(iter
, &dev
->adj_list
.lower
, list
) {
5190 if (!net_eq(net
,dev_net(iter
->dev
)))
5192 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
5193 &iter
->dev
->adj_list
.upper
);
5194 netdev_adjacent_sysfs_add(dev
, iter
->dev
,
5195 &dev
->adj_list
.lower
);
5199 void netdev_adjacent_del_links(struct net_device
*dev
)
5201 struct netdev_adjacent
*iter
;
5203 struct net
*net
= dev_net(dev
);
5205 list_for_each_entry(iter
, &dev
->adj_list
.upper
, list
) {
5206 if (!net_eq(net
,dev_net(iter
->dev
)))
5208 netdev_adjacent_sysfs_del(iter
->dev
, dev
->name
,
5209 &iter
->dev
->adj_list
.lower
);
5210 netdev_adjacent_sysfs_del(dev
, iter
->dev
->name
,
5211 &dev
->adj_list
.upper
);
5214 list_for_each_entry(iter
, &dev
->adj_list
.lower
, list
) {
5215 if (!net_eq(net
,dev_net(iter
->dev
)))
5217 netdev_adjacent_sysfs_del(iter
->dev
, dev
->name
,
5218 &iter
->dev
->adj_list
.upper
);
5219 netdev_adjacent_sysfs_del(dev
, iter
->dev
->name
,
5220 &dev
->adj_list
.lower
);
5224 void netdev_adjacent_rename_links(struct net_device
*dev
, char *oldname
)
5226 struct netdev_adjacent
*iter
;
5228 struct net
*net
= dev_net(dev
);
5230 list_for_each_entry(iter
, &dev
->adj_list
.upper
, list
) {
5231 if (!net_eq(net
,dev_net(iter
->dev
)))
5233 netdev_adjacent_sysfs_del(iter
->dev
, oldname
,
5234 &iter
->dev
->adj_list
.lower
);
5235 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
5236 &iter
->dev
->adj_list
.lower
);
5239 list_for_each_entry(iter
, &dev
->adj_list
.lower
, list
) {
5240 if (!net_eq(net
,dev_net(iter
->dev
)))
5242 netdev_adjacent_sysfs_del(iter
->dev
, oldname
,
5243 &iter
->dev
->adj_list
.upper
);
5244 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
5245 &iter
->dev
->adj_list
.upper
);
5249 void *netdev_lower_dev_get_private(struct net_device
*dev
,
5250 struct net_device
*lower_dev
)
5252 struct netdev_adjacent
*lower
;
5256 lower
= __netdev_find_adj(dev
, lower_dev
, &dev
->adj_list
.lower
);
5260 return lower
->private;
5262 EXPORT_SYMBOL(netdev_lower_dev_get_private
);
5265 int dev_get_nest_level(struct net_device
*dev
,
5266 bool (*type_check
)(struct net_device
*dev
))
5268 struct net_device
*lower
= NULL
;
5269 struct list_head
*iter
;
5275 netdev_for_each_lower_dev(dev
, lower
, iter
) {
5276 nest
= dev_get_nest_level(lower
, type_check
);
5277 if (max_nest
< nest
)
5281 if (type_check(dev
))
5286 EXPORT_SYMBOL(dev_get_nest_level
);
5288 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
5290 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5292 if (ops
->ndo_change_rx_flags
)
5293 ops
->ndo_change_rx_flags(dev
, flags
);
5296 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
, bool notify
)
5298 unsigned int old_flags
= dev
->flags
;
5304 dev
->flags
|= IFF_PROMISC
;
5305 dev
->promiscuity
+= inc
;
5306 if (dev
->promiscuity
== 0) {
5309 * If inc causes overflow, untouch promisc and return error.
5312 dev
->flags
&= ~IFF_PROMISC
;
5314 dev
->promiscuity
-= inc
;
5315 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5320 if (dev
->flags
!= old_flags
) {
5321 pr_info("device %s %s promiscuous mode\n",
5323 dev
->flags
& IFF_PROMISC
? "entered" : "left");
5324 if (audit_enabled
) {
5325 current_uid_gid(&uid
, &gid
);
5326 audit_log(current
->audit_context
, GFP_ATOMIC
,
5327 AUDIT_ANOM_PROMISCUOUS
,
5328 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5329 dev
->name
, (dev
->flags
& IFF_PROMISC
),
5330 (old_flags
& IFF_PROMISC
),
5331 from_kuid(&init_user_ns
, audit_get_loginuid(current
)),
5332 from_kuid(&init_user_ns
, uid
),
5333 from_kgid(&init_user_ns
, gid
),
5334 audit_get_sessionid(current
));
5337 dev_change_rx_flags(dev
, IFF_PROMISC
);
5340 __dev_notify_flags(dev
, old_flags
, IFF_PROMISC
);
5345 * dev_set_promiscuity - update promiscuity count on a device
5349 * Add or remove promiscuity from a device. While the count in the device
5350 * remains above zero the interface remains promiscuous. Once it hits zero
5351 * the device reverts back to normal filtering operation. A negative inc
5352 * value is used to drop promiscuity on the device.
5353 * Return 0 if successful or a negative errno code on error.
5355 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
5357 unsigned int old_flags
= dev
->flags
;
5360 err
= __dev_set_promiscuity(dev
, inc
, true);
5363 if (dev
->flags
!= old_flags
)
5364 dev_set_rx_mode(dev
);
5367 EXPORT_SYMBOL(dev_set_promiscuity
);
5369 static int __dev_set_allmulti(struct net_device
*dev
, int inc
, bool notify
)
5371 unsigned int old_flags
= dev
->flags
, old_gflags
= dev
->gflags
;
5375 dev
->flags
|= IFF_ALLMULTI
;
5376 dev
->allmulti
+= inc
;
5377 if (dev
->allmulti
== 0) {
5380 * If inc causes overflow, untouch allmulti and return error.
5383 dev
->flags
&= ~IFF_ALLMULTI
;
5385 dev
->allmulti
-= inc
;
5386 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5391 if (dev
->flags
^ old_flags
) {
5392 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
5393 dev_set_rx_mode(dev
);
5395 __dev_notify_flags(dev
, old_flags
,
5396 dev
->gflags
^ old_gflags
);
5402 * dev_set_allmulti - update allmulti count on a device
5406 * Add or remove reception of all multicast frames to a device. While the
5407 * count in the device remains above zero the interface remains listening
5408 * to all interfaces. Once it hits zero the device reverts back to normal
5409 * filtering operation. A negative @inc value is used to drop the counter
5410 * when releasing a resource needing all multicasts.
5411 * Return 0 if successful or a negative errno code on error.
5414 int dev_set_allmulti(struct net_device
*dev
, int inc
)
5416 return __dev_set_allmulti(dev
, inc
, true);
5418 EXPORT_SYMBOL(dev_set_allmulti
);
5421 * Upload unicast and multicast address lists to device and
5422 * configure RX filtering. When the device doesn't support unicast
5423 * filtering it is put in promiscuous mode while unicast addresses
5426 void __dev_set_rx_mode(struct net_device
*dev
)
5428 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5430 /* dev_open will call this function so the list will stay sane. */
5431 if (!(dev
->flags
&IFF_UP
))
5434 if (!netif_device_present(dev
))
5437 if (!(dev
->priv_flags
& IFF_UNICAST_FLT
)) {
5438 /* Unicast addresses changes may only happen under the rtnl,
5439 * therefore calling __dev_set_promiscuity here is safe.
5441 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
5442 __dev_set_promiscuity(dev
, 1, false);
5443 dev
->uc_promisc
= true;
5444 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
5445 __dev_set_promiscuity(dev
, -1, false);
5446 dev
->uc_promisc
= false;
5450 if (ops
->ndo_set_rx_mode
)
5451 ops
->ndo_set_rx_mode(dev
);
5454 void dev_set_rx_mode(struct net_device
*dev
)
5456 netif_addr_lock_bh(dev
);
5457 __dev_set_rx_mode(dev
);
5458 netif_addr_unlock_bh(dev
);
5462 * dev_get_flags - get flags reported to userspace
5465 * Get the combination of flag bits exported through APIs to userspace.
5467 unsigned int dev_get_flags(const struct net_device
*dev
)
5471 flags
= (dev
->flags
& ~(IFF_PROMISC
|
5476 (dev
->gflags
& (IFF_PROMISC
|
5479 if (netif_running(dev
)) {
5480 if (netif_oper_up(dev
))
5481 flags
|= IFF_RUNNING
;
5482 if (netif_carrier_ok(dev
))
5483 flags
|= IFF_LOWER_UP
;
5484 if (netif_dormant(dev
))
5485 flags
|= IFF_DORMANT
;
5490 EXPORT_SYMBOL(dev_get_flags
);
5492 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
5494 unsigned int old_flags
= dev
->flags
;
5500 * Set the flags on our device.
5503 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
5504 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
5506 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
5510 * Load in the correct multicast list now the flags have changed.
5513 if ((old_flags
^ flags
) & IFF_MULTICAST
)
5514 dev_change_rx_flags(dev
, IFF_MULTICAST
);
5516 dev_set_rx_mode(dev
);
5519 * Have we downed the interface. We handle IFF_UP ourselves
5520 * according to user attempts to set it, rather than blindly
5525 if ((old_flags
^ flags
) & IFF_UP
)
5526 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
5528 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
5529 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
5530 unsigned int old_flags
= dev
->flags
;
5532 dev
->gflags
^= IFF_PROMISC
;
5534 if (__dev_set_promiscuity(dev
, inc
, false) >= 0)
5535 if (dev
->flags
!= old_flags
)
5536 dev_set_rx_mode(dev
);
5539 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5540 is important. Some (broken) drivers set IFF_PROMISC, when
5541 IFF_ALLMULTI is requested not asking us and not reporting.
5543 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
5544 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
5546 dev
->gflags
^= IFF_ALLMULTI
;
5547 __dev_set_allmulti(dev
, inc
, false);
5553 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
,
5554 unsigned int gchanges
)
5556 unsigned int changes
= dev
->flags
^ old_flags
;
5559 rtmsg_ifinfo(RTM_NEWLINK
, dev
, gchanges
, GFP_ATOMIC
);
5561 if (changes
& IFF_UP
) {
5562 if (dev
->flags
& IFF_UP
)
5563 call_netdevice_notifiers(NETDEV_UP
, dev
);
5565 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
5568 if (dev
->flags
& IFF_UP
&&
5569 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
))) {
5570 struct netdev_notifier_change_info change_info
;
5572 change_info
.flags_changed
= changes
;
5573 call_netdevice_notifiers_info(NETDEV_CHANGE
, dev
,
5579 * dev_change_flags - change device settings
5581 * @flags: device state flags
5583 * Change settings on device based state flags. The flags are
5584 * in the userspace exported format.
5586 int dev_change_flags(struct net_device
*dev
, unsigned int flags
)
5589 unsigned int changes
, old_flags
= dev
->flags
, old_gflags
= dev
->gflags
;
5591 ret
= __dev_change_flags(dev
, flags
);
5595 changes
= (old_flags
^ dev
->flags
) | (old_gflags
^ dev
->gflags
);
5596 __dev_notify_flags(dev
, old_flags
, changes
);
5599 EXPORT_SYMBOL(dev_change_flags
);
5601 static int __dev_set_mtu(struct net_device
*dev
, int new_mtu
)
5603 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5605 if (ops
->ndo_change_mtu
)
5606 return ops
->ndo_change_mtu(dev
, new_mtu
);
5613 * dev_set_mtu - Change maximum transfer unit
5615 * @new_mtu: new transfer unit
5617 * Change the maximum transfer size of the network device.
5619 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
5623 if (new_mtu
== dev
->mtu
)
5626 /* MTU must be positive. */
5630 if (!netif_device_present(dev
))
5633 err
= call_netdevice_notifiers(NETDEV_PRECHANGEMTU
, dev
);
5634 err
= notifier_to_errno(err
);
5638 orig_mtu
= dev
->mtu
;
5639 err
= __dev_set_mtu(dev
, new_mtu
);
5642 err
= call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
5643 err
= notifier_to_errno(err
);
5645 /* setting mtu back and notifying everyone again,
5646 * so that they have a chance to revert changes.
5648 __dev_set_mtu(dev
, orig_mtu
);
5649 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
5654 EXPORT_SYMBOL(dev_set_mtu
);
5657 * dev_set_group - Change group this device belongs to
5659 * @new_group: group this device should belong to
5661 void dev_set_group(struct net_device
*dev
, int new_group
)
5663 dev
->group
= new_group
;
5665 EXPORT_SYMBOL(dev_set_group
);
5668 * dev_set_mac_address - Change Media Access Control Address
5672 * Change the hardware (MAC) address of the device
5674 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
5676 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5679 if (!ops
->ndo_set_mac_address
)
5681 if (sa
->sa_family
!= dev
->type
)
5683 if (!netif_device_present(dev
))
5685 err
= ops
->ndo_set_mac_address(dev
, sa
);
5688 dev
->addr_assign_type
= NET_ADDR_SET
;
5689 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
5690 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
5693 EXPORT_SYMBOL(dev_set_mac_address
);
5696 * dev_change_carrier - Change device carrier
5698 * @new_carrier: new value
5700 * Change device carrier
5702 int dev_change_carrier(struct net_device
*dev
, bool new_carrier
)
5704 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5706 if (!ops
->ndo_change_carrier
)
5708 if (!netif_device_present(dev
))
5710 return ops
->ndo_change_carrier(dev
, new_carrier
);
5712 EXPORT_SYMBOL(dev_change_carrier
);
5715 * dev_get_phys_port_id - Get device physical port ID
5719 * Get device physical port ID
5721 int dev_get_phys_port_id(struct net_device
*dev
,
5722 struct netdev_phys_port_id
*ppid
)
5724 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5726 if (!ops
->ndo_get_phys_port_id
)
5728 return ops
->ndo_get_phys_port_id(dev
, ppid
);
5730 EXPORT_SYMBOL(dev_get_phys_port_id
);
5733 * dev_new_index - allocate an ifindex
5734 * @net: the applicable net namespace
5736 * Returns a suitable unique value for a new device interface
5737 * number. The caller must hold the rtnl semaphore or the
5738 * dev_base_lock to be sure it remains unique.
5740 static int dev_new_index(struct net
*net
)
5742 int ifindex
= net
->ifindex
;
5746 if (!__dev_get_by_index(net
, ifindex
))
5747 return net
->ifindex
= ifindex
;
5751 /* Delayed registration/unregisteration */
5752 static LIST_HEAD(net_todo_list
);
5753 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq
);
5755 static void net_set_todo(struct net_device
*dev
)
5757 list_add_tail(&dev
->todo_list
, &net_todo_list
);
5758 dev_net(dev
)->dev_unreg_count
++;
5761 static void rollback_registered_many(struct list_head
*head
)
5763 struct net_device
*dev
, *tmp
;
5764 LIST_HEAD(close_head
);
5766 BUG_ON(dev_boot_phase
);
5769 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
5770 /* Some devices call without registering
5771 * for initialization unwind. Remove those
5772 * devices and proceed with the remaining.
5774 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5775 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5779 list_del(&dev
->unreg_list
);
5782 dev
->dismantle
= true;
5783 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
5786 /* If device is running, close it first. */
5787 list_for_each_entry(dev
, head
, unreg_list
)
5788 list_add_tail(&dev
->close_list
, &close_head
);
5789 dev_close_many(&close_head
);
5791 list_for_each_entry(dev
, head
, unreg_list
) {
5792 /* And unlink it from device chain. */
5793 unlist_netdevice(dev
);
5795 dev
->reg_state
= NETREG_UNREGISTERING
;
5800 list_for_each_entry(dev
, head
, unreg_list
) {
5801 /* Shutdown queueing discipline. */
5805 /* Notify protocols, that we are about to destroy
5806 this device. They should clean all the things.
5808 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5811 * Flush the unicast and multicast chains
5816 if (dev
->netdev_ops
->ndo_uninit
)
5817 dev
->netdev_ops
->ndo_uninit(dev
);
5819 if (!dev
->rtnl_link_ops
||
5820 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5821 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U, GFP_KERNEL
);
5823 /* Notifier chain MUST detach us all upper devices. */
5824 WARN_ON(netdev_has_any_upper_dev(dev
));
5826 /* Remove entries from kobject tree */
5827 netdev_unregister_kobject(dev
);
5829 /* Remove XPS queueing entries */
5830 netif_reset_xps_queues_gt(dev
, 0);
5836 list_for_each_entry(dev
, head
, unreg_list
)
5840 static void rollback_registered(struct net_device
*dev
)
5844 list_add(&dev
->unreg_list
, &single
);
5845 rollback_registered_many(&single
);
5849 static netdev_features_t
netdev_fix_features(struct net_device
*dev
,
5850 netdev_features_t features
)
5852 /* Fix illegal checksum combinations */
5853 if ((features
& NETIF_F_HW_CSUM
) &&
5854 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5855 netdev_warn(dev
, "mixed HW and IP checksum settings.\n");
5856 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5859 /* TSO requires that SG is present as well. */
5860 if ((features
& NETIF_F_ALL_TSO
) && !(features
& NETIF_F_SG
)) {
5861 netdev_dbg(dev
, "Dropping TSO features since no SG feature.\n");
5862 features
&= ~NETIF_F_ALL_TSO
;
5865 if ((features
& NETIF_F_TSO
) && !(features
& NETIF_F_HW_CSUM
) &&
5866 !(features
& NETIF_F_IP_CSUM
)) {
5867 netdev_dbg(dev
, "Dropping TSO features since no CSUM feature.\n");
5868 features
&= ~NETIF_F_TSO
;
5869 features
&= ~NETIF_F_TSO_ECN
;
5872 if ((features
& NETIF_F_TSO6
) && !(features
& NETIF_F_HW_CSUM
) &&
5873 !(features
& NETIF_F_IPV6_CSUM
)) {
5874 netdev_dbg(dev
, "Dropping TSO6 features since no CSUM feature.\n");
5875 features
&= ~NETIF_F_TSO6
;
5878 /* TSO ECN requires that TSO is present as well. */
5879 if ((features
& NETIF_F_ALL_TSO
) == NETIF_F_TSO_ECN
)
5880 features
&= ~NETIF_F_TSO_ECN
;
5882 /* Software GSO depends on SG. */
5883 if ((features
& NETIF_F_GSO
) && !(features
& NETIF_F_SG
)) {
5884 netdev_dbg(dev
, "Dropping NETIF_F_GSO since no SG feature.\n");
5885 features
&= ~NETIF_F_GSO
;
5888 /* UFO needs SG and checksumming */
5889 if (features
& NETIF_F_UFO
) {
5890 /* maybe split UFO into V4 and V6? */
5891 if (!((features
& NETIF_F_GEN_CSUM
) ||
5892 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))
5893 == (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5895 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5896 features
&= ~NETIF_F_UFO
;
5899 if (!(features
& NETIF_F_SG
)) {
5901 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5902 features
&= ~NETIF_F_UFO
;
5906 #ifdef CONFIG_NET_RX_BUSY_POLL
5907 if (dev
->netdev_ops
->ndo_busy_poll
)
5908 features
|= NETIF_F_BUSY_POLL
;
5911 features
&= ~NETIF_F_BUSY_POLL
;
5916 int __netdev_update_features(struct net_device
*dev
)
5918 netdev_features_t features
;
5923 features
= netdev_get_wanted_features(dev
);
5925 if (dev
->netdev_ops
->ndo_fix_features
)
5926 features
= dev
->netdev_ops
->ndo_fix_features(dev
, features
);
5928 /* driver might be less strict about feature dependencies */
5929 features
= netdev_fix_features(dev
, features
);
5931 if (dev
->features
== features
)
5934 netdev_dbg(dev
, "Features changed: %pNF -> %pNF\n",
5935 &dev
->features
, &features
);
5937 if (dev
->netdev_ops
->ndo_set_features
)
5938 err
= dev
->netdev_ops
->ndo_set_features(dev
, features
);
5940 if (unlikely(err
< 0)) {
5942 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5943 err
, &features
, &dev
->features
);
5948 dev
->features
= features
;
5954 * netdev_update_features - recalculate device features
5955 * @dev: the device to check
5957 * Recalculate dev->features set and send notifications if it
5958 * has changed. Should be called after driver or hardware dependent
5959 * conditions might have changed that influence the features.
5961 void netdev_update_features(struct net_device
*dev
)
5963 if (__netdev_update_features(dev
))
5964 netdev_features_change(dev
);
5966 EXPORT_SYMBOL(netdev_update_features
);
5969 * netdev_change_features - recalculate device features
5970 * @dev: the device to check
5972 * Recalculate dev->features set and send notifications even
5973 * if they have not changed. Should be called instead of
5974 * netdev_update_features() if also dev->vlan_features might
5975 * have changed to allow the changes to be propagated to stacked
5978 void netdev_change_features(struct net_device
*dev
)
5980 __netdev_update_features(dev
);
5981 netdev_features_change(dev
);
5983 EXPORT_SYMBOL(netdev_change_features
);
5986 * netif_stacked_transfer_operstate - transfer operstate
5987 * @rootdev: the root or lower level device to transfer state from
5988 * @dev: the device to transfer operstate to
5990 * Transfer operational state from root to device. This is normally
5991 * called when a stacking relationship exists between the root
5992 * device and the device(a leaf device).
5994 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
5995 struct net_device
*dev
)
5997 if (rootdev
->operstate
== IF_OPER_DORMANT
)
5998 netif_dormant_on(dev
);
6000 netif_dormant_off(dev
);
6002 if (netif_carrier_ok(rootdev
)) {
6003 if (!netif_carrier_ok(dev
))
6004 netif_carrier_on(dev
);
6006 if (netif_carrier_ok(dev
))
6007 netif_carrier_off(dev
);
6010 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
6013 static int netif_alloc_rx_queues(struct net_device
*dev
)
6015 unsigned int i
, count
= dev
->num_rx_queues
;
6016 struct netdev_rx_queue
*rx
;
6020 rx
= kcalloc(count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
6026 for (i
= 0; i
< count
; i
++)
6032 static void netdev_init_one_queue(struct net_device
*dev
,
6033 struct netdev_queue
*queue
, void *_unused
)
6035 /* Initialize queue lock */
6036 spin_lock_init(&queue
->_xmit_lock
);
6037 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
6038 queue
->xmit_lock_owner
= -1;
6039 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
6042 dql_init(&queue
->dql
, HZ
);
6046 static void netif_free_tx_queues(struct net_device
*dev
)
6051 static int netif_alloc_netdev_queues(struct net_device
*dev
)
6053 unsigned int count
= dev
->num_tx_queues
;
6054 struct netdev_queue
*tx
;
6055 size_t sz
= count
* sizeof(*tx
);
6057 BUG_ON(count
< 1 || count
> 0xffff);
6059 tx
= kzalloc(sz
, GFP_KERNEL
| __GFP_NOWARN
| __GFP_REPEAT
);
6067 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
6068 spin_lock_init(&dev
->tx_global_lock
);
6074 * register_netdevice - register a network device
6075 * @dev: device to register
6077 * Take a completed network device structure and add it to the kernel
6078 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6079 * chain. 0 is returned on success. A negative errno code is returned
6080 * on a failure to set up the device, or if the name is a duplicate.
6082 * Callers must hold the rtnl semaphore. You may want
6083 * register_netdev() instead of this.
6086 * The locking appears insufficient to guarantee two parallel registers
6087 * will not get the same name.
6090 int register_netdevice(struct net_device
*dev
)
6093 struct net
*net
= dev_net(dev
);
6095 BUG_ON(dev_boot_phase
);
6100 /* When net_device's are persistent, this will be fatal. */
6101 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
6104 spin_lock_init(&dev
->addr_list_lock
);
6105 netdev_set_addr_lockdep_class(dev
);
6109 ret
= dev_get_valid_name(net
, dev
, dev
->name
);
6113 /* Init, if this function is available */
6114 if (dev
->netdev_ops
->ndo_init
) {
6115 ret
= dev
->netdev_ops
->ndo_init(dev
);
6123 if (((dev
->hw_features
| dev
->features
) &
6124 NETIF_F_HW_VLAN_CTAG_FILTER
) &&
6125 (!dev
->netdev_ops
->ndo_vlan_rx_add_vid
||
6126 !dev
->netdev_ops
->ndo_vlan_rx_kill_vid
)) {
6127 netdev_WARN(dev
, "Buggy VLAN acceleration in driver!\n");
6134 dev
->ifindex
= dev_new_index(net
);
6135 else if (__dev_get_by_index(net
, dev
->ifindex
))
6138 if (dev
->iflink
== -1)
6139 dev
->iflink
= dev
->ifindex
;
6141 /* Transfer changeable features to wanted_features and enable
6142 * software offloads (GSO and GRO).
6144 dev
->hw_features
|= NETIF_F_SOFT_FEATURES
;
6145 dev
->features
|= NETIF_F_SOFT_FEATURES
;
6146 dev
->wanted_features
= dev
->features
& dev
->hw_features
;
6148 if (!(dev
->flags
& IFF_LOOPBACK
)) {
6149 dev
->hw_features
|= NETIF_F_NOCACHE_COPY
;
6152 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6154 dev
->vlan_features
|= NETIF_F_HIGHDMA
;
6156 /* Make NETIF_F_SG inheritable to tunnel devices.
6158 dev
->hw_enc_features
|= NETIF_F_SG
;
6160 /* Make NETIF_F_SG inheritable to MPLS.
6162 dev
->mpls_features
|= NETIF_F_SG
;
6164 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
6165 ret
= notifier_to_errno(ret
);
6169 ret
= netdev_register_kobject(dev
);
6172 dev
->reg_state
= NETREG_REGISTERED
;
6174 __netdev_update_features(dev
);
6177 * Default initial state at registry is that the
6178 * device is present.
6181 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
6183 linkwatch_init_dev(dev
);
6185 dev_init_scheduler(dev
);
6187 list_netdevice(dev
);
6188 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
6190 /* If the device has permanent device address, driver should
6191 * set dev_addr and also addr_assign_type should be set to
6192 * NET_ADDR_PERM (default value).
6194 if (dev
->addr_assign_type
== NET_ADDR_PERM
)
6195 memcpy(dev
->perm_addr
, dev
->dev_addr
, dev
->addr_len
);
6197 /* Notify protocols, that a new device appeared. */
6198 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
6199 ret
= notifier_to_errno(ret
);
6201 rollback_registered(dev
);
6202 dev
->reg_state
= NETREG_UNREGISTERED
;
6205 * Prevent userspace races by waiting until the network
6206 * device is fully setup before sending notifications.
6208 if (!dev
->rtnl_link_ops
||
6209 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
6210 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U, GFP_KERNEL
);
6216 if (dev
->netdev_ops
->ndo_uninit
)
6217 dev
->netdev_ops
->ndo_uninit(dev
);
6220 EXPORT_SYMBOL(register_netdevice
);
6223 * init_dummy_netdev - init a dummy network device for NAPI
6224 * @dev: device to init
6226 * This takes a network device structure and initialize the minimum
6227 * amount of fields so it can be used to schedule NAPI polls without
6228 * registering a full blown interface. This is to be used by drivers
6229 * that need to tie several hardware interfaces to a single NAPI
6230 * poll scheduler due to HW limitations.
6232 int init_dummy_netdev(struct net_device
*dev
)
6234 /* Clear everything. Note we don't initialize spinlocks
6235 * are they aren't supposed to be taken by any of the
6236 * NAPI code and this dummy netdev is supposed to be
6237 * only ever used for NAPI polls
6239 memset(dev
, 0, sizeof(struct net_device
));
6241 /* make sure we BUG if trying to hit standard
6242 * register/unregister code path
6244 dev
->reg_state
= NETREG_DUMMY
;
6246 /* NAPI wants this */
6247 INIT_LIST_HEAD(&dev
->napi_list
);
6249 /* a dummy interface is started by default */
6250 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
6251 set_bit(__LINK_STATE_START
, &dev
->state
);
6253 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6254 * because users of this 'device' dont need to change
6260 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
6264 * register_netdev - register a network device
6265 * @dev: device to register
6267 * Take a completed network device structure and add it to the kernel
6268 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6269 * chain. 0 is returned on success. A negative errno code is returned
6270 * on a failure to set up the device, or if the name is a duplicate.
6272 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6273 * and expands the device name if you passed a format string to
6276 int register_netdev(struct net_device
*dev
)
6281 err
= register_netdevice(dev
);
6285 EXPORT_SYMBOL(register_netdev
);
6287 int netdev_refcnt_read(const struct net_device
*dev
)
6291 for_each_possible_cpu(i
)
6292 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
6295 EXPORT_SYMBOL(netdev_refcnt_read
);
6298 * netdev_wait_allrefs - wait until all references are gone.
6299 * @dev: target net_device
6301 * This is called when unregistering network devices.
6303 * Any protocol or device that holds a reference should register
6304 * for netdevice notification, and cleanup and put back the
6305 * reference if they receive an UNREGISTER event.
6306 * We can get stuck here if buggy protocols don't correctly
6309 static void netdev_wait_allrefs(struct net_device
*dev
)
6311 unsigned long rebroadcast_time
, warning_time
;
6314 linkwatch_forget_dev(dev
);
6316 rebroadcast_time
= warning_time
= jiffies
;
6317 refcnt
= netdev_refcnt_read(dev
);
6319 while (refcnt
!= 0) {
6320 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
6323 /* Rebroadcast unregister notification */
6324 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6330 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
6331 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
6333 /* We must not have linkwatch events
6334 * pending on unregister. If this
6335 * happens, we simply run the queue
6336 * unscheduled, resulting in a noop
6339 linkwatch_run_queue();
6344 rebroadcast_time
= jiffies
;
6349 refcnt
= netdev_refcnt_read(dev
);
6351 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
6352 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6354 warning_time
= jiffies
;
6363 * register_netdevice(x1);
6364 * register_netdevice(x2);
6366 * unregister_netdevice(y1);
6367 * unregister_netdevice(y2);
6373 * We are invoked by rtnl_unlock().
6374 * This allows us to deal with problems:
6375 * 1) We can delete sysfs objects which invoke hotplug
6376 * without deadlocking with linkwatch via keventd.
6377 * 2) Since we run with the RTNL semaphore not held, we can sleep
6378 * safely in order to wait for the netdev refcnt to drop to zero.
6380 * We must not return until all unregister events added during
6381 * the interval the lock was held have been completed.
6383 void netdev_run_todo(void)
6385 struct list_head list
;
6387 /* Snapshot list, allow later requests */
6388 list_replace_init(&net_todo_list
, &list
);
6393 /* Wait for rcu callbacks to finish before next phase */
6394 if (!list_empty(&list
))
6397 while (!list_empty(&list
)) {
6398 struct net_device
*dev
6399 = list_first_entry(&list
, struct net_device
, todo_list
);
6400 list_del(&dev
->todo_list
);
6403 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
6406 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
6407 pr_err("network todo '%s' but state %d\n",
6408 dev
->name
, dev
->reg_state
);
6413 dev
->reg_state
= NETREG_UNREGISTERED
;
6415 on_each_cpu(flush_backlog
, dev
, 1);
6417 netdev_wait_allrefs(dev
);
6420 BUG_ON(netdev_refcnt_read(dev
));
6421 WARN_ON(rcu_access_pointer(dev
->ip_ptr
));
6422 WARN_ON(rcu_access_pointer(dev
->ip6_ptr
));
6423 WARN_ON(dev
->dn_ptr
);
6425 if (dev
->destructor
)
6426 dev
->destructor(dev
);
6428 /* Report a network device has been unregistered */
6430 dev_net(dev
)->dev_unreg_count
--;
6432 wake_up(&netdev_unregistering_wq
);
6434 /* Free network device */
6435 kobject_put(&dev
->dev
.kobj
);
6439 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6440 * fields in the same order, with only the type differing.
6442 void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
6443 const struct net_device_stats
*netdev_stats
)
6445 #if BITS_PER_LONG == 64
6446 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
6447 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
6449 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
6450 const unsigned long *src
= (const unsigned long *)netdev_stats
;
6451 u64
*dst
= (u64
*)stats64
;
6453 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
6454 sizeof(*stats64
) / sizeof(u64
));
6455 for (i
= 0; i
< n
; i
++)
6459 EXPORT_SYMBOL(netdev_stats_to_stats64
);
6462 * dev_get_stats - get network device statistics
6463 * @dev: device to get statistics from
6464 * @storage: place to store stats
6466 * Get network statistics from device. Return @storage.
6467 * The device driver may provide its own method by setting
6468 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6469 * otherwise the internal statistics structure is used.
6471 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
6472 struct rtnl_link_stats64
*storage
)
6474 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6476 if (ops
->ndo_get_stats64
) {
6477 memset(storage
, 0, sizeof(*storage
));
6478 ops
->ndo_get_stats64(dev
, storage
);
6479 } else if (ops
->ndo_get_stats
) {
6480 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
6482 netdev_stats_to_stats64(storage
, &dev
->stats
);
6484 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
6485 storage
->tx_dropped
+= atomic_long_read(&dev
->tx_dropped
);
6488 EXPORT_SYMBOL(dev_get_stats
);
6490 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
6492 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
6494 #ifdef CONFIG_NET_CLS_ACT
6497 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
6500 netdev_init_one_queue(dev
, queue
, NULL
);
6501 queue
->qdisc
= &noop_qdisc
;
6502 queue
->qdisc_sleeping
= &noop_qdisc
;
6503 rcu_assign_pointer(dev
->ingress_queue
, queue
);
6508 static const struct ethtool_ops default_ethtool_ops
;
6510 void netdev_set_default_ethtool_ops(struct net_device
*dev
,
6511 const struct ethtool_ops
*ops
)
6513 if (dev
->ethtool_ops
== &default_ethtool_ops
)
6514 dev
->ethtool_ops
= ops
;
6516 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops
);
6518 void netdev_freemem(struct net_device
*dev
)
6520 char *addr
= (char *)dev
- dev
->padded
;
6526 * alloc_netdev_mqs - allocate network device
6527 * @sizeof_priv: size of private data to allocate space for
6528 * @name: device name format string
6529 * @name_assign_type: origin of device name
6530 * @setup: callback to initialize device
6531 * @txqs: the number of TX subqueues to allocate
6532 * @rxqs: the number of RX subqueues to allocate
6534 * Allocates a struct net_device with private data area for driver use
6535 * and performs basic initialization. Also allocates subqueue structs
6536 * for each queue on the device.
6538 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
6539 unsigned char name_assign_type
,
6540 void (*setup
)(struct net_device
*),
6541 unsigned int txqs
, unsigned int rxqs
)
6543 struct net_device
*dev
;
6545 struct net_device
*p
;
6547 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
6550 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6556 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6561 alloc_size
= sizeof(struct net_device
);
6563 /* ensure 32-byte alignment of private area */
6564 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
6565 alloc_size
+= sizeof_priv
;
6567 /* ensure 32-byte alignment of whole construct */
6568 alloc_size
+= NETDEV_ALIGN
- 1;
6570 p
= kzalloc(alloc_size
, GFP_KERNEL
| __GFP_NOWARN
| __GFP_REPEAT
);
6572 p
= vzalloc(alloc_size
);
6576 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
6577 dev
->padded
= (char *)dev
- (char *)p
;
6579 dev
->pcpu_refcnt
= alloc_percpu(int);
6580 if (!dev
->pcpu_refcnt
)
6583 if (dev_addr_init(dev
))
6589 dev_net_set(dev
, &init_net
);
6591 dev
->gso_max_size
= GSO_MAX_SIZE
;
6592 dev
->gso_max_segs
= GSO_MAX_SEGS
;
6594 INIT_LIST_HEAD(&dev
->napi_list
);
6595 INIT_LIST_HEAD(&dev
->unreg_list
);
6596 INIT_LIST_HEAD(&dev
->close_list
);
6597 INIT_LIST_HEAD(&dev
->link_watch_list
);
6598 INIT_LIST_HEAD(&dev
->adj_list
.upper
);
6599 INIT_LIST_HEAD(&dev
->adj_list
.lower
);
6600 INIT_LIST_HEAD(&dev
->all_adj_list
.upper
);
6601 INIT_LIST_HEAD(&dev
->all_adj_list
.lower
);
6602 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
6605 dev
->num_tx_queues
= txqs
;
6606 dev
->real_num_tx_queues
= txqs
;
6607 if (netif_alloc_netdev_queues(dev
))
6611 dev
->num_rx_queues
= rxqs
;
6612 dev
->real_num_rx_queues
= rxqs
;
6613 if (netif_alloc_rx_queues(dev
))
6617 strcpy(dev
->name
, name
);
6618 dev
->name_assign_type
= name_assign_type
;
6619 dev
->group
= INIT_NETDEV_GROUP
;
6620 if (!dev
->ethtool_ops
)
6621 dev
->ethtool_ops
= &default_ethtool_ops
;
6629 free_percpu(dev
->pcpu_refcnt
);
6631 netdev_freemem(dev
);
6634 EXPORT_SYMBOL(alloc_netdev_mqs
);
6637 * free_netdev - free network device
6640 * This function does the last stage of destroying an allocated device
6641 * interface. The reference to the device object is released.
6642 * If this is the last reference then it will be freed.
6644 void free_netdev(struct net_device
*dev
)
6646 struct napi_struct
*p
, *n
;
6648 release_net(dev_net(dev
));
6650 netif_free_tx_queues(dev
);
6655 kfree(rcu_dereference_protected(dev
->ingress_queue
, 1));
6657 /* Flush device addresses */
6658 dev_addr_flush(dev
);
6660 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
6663 free_percpu(dev
->pcpu_refcnt
);
6664 dev
->pcpu_refcnt
= NULL
;
6666 /* Compatibility with error handling in drivers */
6667 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
6668 netdev_freemem(dev
);
6672 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
6673 dev
->reg_state
= NETREG_RELEASED
;
6675 /* will free via device release */
6676 put_device(&dev
->dev
);
6678 EXPORT_SYMBOL(free_netdev
);
6681 * synchronize_net - Synchronize with packet receive processing
6683 * Wait for packets currently being received to be done.
6684 * Does not block later packets from starting.
6686 void synchronize_net(void)
6689 if (rtnl_is_locked())
6690 synchronize_rcu_expedited();
6694 EXPORT_SYMBOL(synchronize_net
);
6697 * unregister_netdevice_queue - remove device from the kernel
6701 * This function shuts down a device interface and removes it
6702 * from the kernel tables.
6703 * If head not NULL, device is queued to be unregistered later.
6705 * Callers must hold the rtnl semaphore. You may want
6706 * unregister_netdev() instead of this.
6709 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
6714 list_move_tail(&dev
->unreg_list
, head
);
6716 rollback_registered(dev
);
6717 /* Finish processing unregister after unlock */
6721 EXPORT_SYMBOL(unregister_netdevice_queue
);
6724 * unregister_netdevice_many - unregister many devices
6725 * @head: list of devices
6727 * Note: As most callers use a stack allocated list_head,
6728 * we force a list_del() to make sure stack wont be corrupted later.
6730 void unregister_netdevice_many(struct list_head
*head
)
6732 struct net_device
*dev
;
6734 if (!list_empty(head
)) {
6735 rollback_registered_many(head
);
6736 list_for_each_entry(dev
, head
, unreg_list
)
6741 EXPORT_SYMBOL(unregister_netdevice_many
);
6744 * unregister_netdev - remove device from the kernel
6747 * This function shuts down a device interface and removes it
6748 * from the kernel tables.
6750 * This is just a wrapper for unregister_netdevice that takes
6751 * the rtnl semaphore. In general you want to use this and not
6752 * unregister_netdevice.
6754 void unregister_netdev(struct net_device
*dev
)
6757 unregister_netdevice(dev
);
6760 EXPORT_SYMBOL(unregister_netdev
);
6763 * dev_change_net_namespace - move device to different nethost namespace
6765 * @net: network namespace
6766 * @pat: If not NULL name pattern to try if the current device name
6767 * is already taken in the destination network namespace.
6769 * This function shuts down a device interface and moves it
6770 * to a new network namespace. On success 0 is returned, on
6771 * a failure a netagive errno code is returned.
6773 * Callers must hold the rtnl semaphore.
6776 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
6782 /* Don't allow namespace local devices to be moved. */
6784 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6787 /* Ensure the device has been registrered */
6788 if (dev
->reg_state
!= NETREG_REGISTERED
)
6791 /* Get out if there is nothing todo */
6793 if (net_eq(dev_net(dev
), net
))
6796 /* Pick the destination device name, and ensure
6797 * we can use it in the destination network namespace.
6800 if (__dev_get_by_name(net
, dev
->name
)) {
6801 /* We get here if we can't use the current device name */
6804 if (dev_get_valid_name(net
, dev
, pat
) < 0)
6809 * And now a mini version of register_netdevice unregister_netdevice.
6812 /* If device is running close it first. */
6815 /* And unlink it from device chain */
6817 unlist_netdevice(dev
);
6821 /* Shutdown queueing discipline. */
6824 /* Notify protocols, that we are about to destroy
6825 this device. They should clean all the things.
6827 Note that dev->reg_state stays at NETREG_REGISTERED.
6828 This is wanted because this way 8021q and macvlan know
6829 the device is just moving and can keep their slaves up.
6831 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6833 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
6834 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U, GFP_KERNEL
);
6837 * Flush the unicast and multicast chains
6842 /* Send a netdev-removed uevent to the old namespace */
6843 kobject_uevent(&dev
->dev
.kobj
, KOBJ_REMOVE
);
6844 netdev_adjacent_del_links(dev
);
6846 /* Actually switch the network namespace */
6847 dev_net_set(dev
, net
);
6849 /* If there is an ifindex conflict assign a new one */
6850 if (__dev_get_by_index(net
, dev
->ifindex
)) {
6851 int iflink
= (dev
->iflink
== dev
->ifindex
);
6852 dev
->ifindex
= dev_new_index(net
);
6854 dev
->iflink
= dev
->ifindex
;
6857 /* Send a netdev-add uevent to the new namespace */
6858 kobject_uevent(&dev
->dev
.kobj
, KOBJ_ADD
);
6859 netdev_adjacent_add_links(dev
);
6861 /* Fixup kobjects */
6862 err
= device_rename(&dev
->dev
, dev
->name
);
6865 /* Add the device back in the hashes */
6866 list_netdevice(dev
);
6868 /* Notify protocols, that a new device appeared. */
6869 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
6872 * Prevent userspace races by waiting until the network
6873 * device is fully setup before sending notifications.
6875 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U, GFP_KERNEL
);
6882 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
6884 static int dev_cpu_callback(struct notifier_block
*nfb
,
6885 unsigned long action
,
6888 struct sk_buff
**list_skb
;
6889 struct sk_buff
*skb
;
6890 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
6891 struct softnet_data
*sd
, *oldsd
;
6893 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
6896 local_irq_disable();
6897 cpu
= smp_processor_id();
6898 sd
= &per_cpu(softnet_data
, cpu
);
6899 oldsd
= &per_cpu(softnet_data
, oldcpu
);
6901 /* Find end of our completion_queue. */
6902 list_skb
= &sd
->completion_queue
;
6904 list_skb
= &(*list_skb
)->next
;
6905 /* Append completion queue from offline CPU. */
6906 *list_skb
= oldsd
->completion_queue
;
6907 oldsd
->completion_queue
= NULL
;
6909 /* Append output queue from offline CPU. */
6910 if (oldsd
->output_queue
) {
6911 *sd
->output_queue_tailp
= oldsd
->output_queue
;
6912 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
6913 oldsd
->output_queue
= NULL
;
6914 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
6916 /* Append NAPI poll list from offline CPU. */
6917 if (!list_empty(&oldsd
->poll_list
)) {
6918 list_splice_init(&oldsd
->poll_list
, &sd
->poll_list
);
6919 raise_softirq_irqoff(NET_RX_SOFTIRQ
);
6922 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
6925 /* Process offline CPU's input_pkt_queue */
6926 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
6927 netif_rx_internal(skb
);
6928 input_queue_head_incr(oldsd
);
6930 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
6931 netif_rx_internal(skb
);
6932 input_queue_head_incr(oldsd
);
6940 * netdev_increment_features - increment feature set by one
6941 * @all: current feature set
6942 * @one: new feature set
6943 * @mask: mask feature set
6945 * Computes a new feature set after adding a device with feature set
6946 * @one to the master device with current feature set @all. Will not
6947 * enable anything that is off in @mask. Returns the new feature set.
6949 netdev_features_t
netdev_increment_features(netdev_features_t all
,
6950 netdev_features_t one
, netdev_features_t mask
)
6952 if (mask
& NETIF_F_GEN_CSUM
)
6953 mask
|= NETIF_F_ALL_CSUM
;
6954 mask
|= NETIF_F_VLAN_CHALLENGED
;
6956 all
|= one
& (NETIF_F_ONE_FOR_ALL
|NETIF_F_ALL_CSUM
) & mask
;
6957 all
&= one
| ~NETIF_F_ALL_FOR_ALL
;
6959 /* If one device supports hw checksumming, set for all. */
6960 if (all
& NETIF_F_GEN_CSUM
)
6961 all
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_GEN_CSUM
);
6965 EXPORT_SYMBOL(netdev_increment_features
);
6967 static struct hlist_head
* __net_init
netdev_create_hash(void)
6970 struct hlist_head
*hash
;
6972 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
6974 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
6975 INIT_HLIST_HEAD(&hash
[i
]);
6980 /* Initialize per network namespace state */
6981 static int __net_init
netdev_init(struct net
*net
)
6983 if (net
!= &init_net
)
6984 INIT_LIST_HEAD(&net
->dev_base_head
);
6986 net
->dev_name_head
= netdev_create_hash();
6987 if (net
->dev_name_head
== NULL
)
6990 net
->dev_index_head
= netdev_create_hash();
6991 if (net
->dev_index_head
== NULL
)
6997 kfree(net
->dev_name_head
);
7003 * netdev_drivername - network driver for the device
7004 * @dev: network device
7006 * Determine network driver for device.
7008 const char *netdev_drivername(const struct net_device
*dev
)
7010 const struct device_driver
*driver
;
7011 const struct device
*parent
;
7012 const char *empty
= "";
7014 parent
= dev
->dev
.parent
;
7018 driver
= parent
->driver
;
7019 if (driver
&& driver
->name
)
7020 return driver
->name
;
7024 static int __netdev_printk(const char *level
, const struct net_device
*dev
,
7025 struct va_format
*vaf
)
7029 if (dev
&& dev
->dev
.parent
) {
7030 r
= dev_printk_emit(level
[1] - '0',
7033 dev_driver_string(dev
->dev
.parent
),
7034 dev_name(dev
->dev
.parent
),
7035 netdev_name(dev
), netdev_reg_state(dev
),
7038 r
= printk("%s%s%s: %pV", level
, netdev_name(dev
),
7039 netdev_reg_state(dev
), vaf
);
7041 r
= printk("%s(NULL net_device): %pV", level
, vaf
);
7047 int netdev_printk(const char *level
, const struct net_device
*dev
,
7048 const char *format
, ...)
7050 struct va_format vaf
;
7054 va_start(args
, format
);
7059 r
= __netdev_printk(level
, dev
, &vaf
);
7065 EXPORT_SYMBOL(netdev_printk
);
7067 #define define_netdev_printk_level(func, level) \
7068 int func(const struct net_device *dev, const char *fmt, ...) \
7071 struct va_format vaf; \
7074 va_start(args, fmt); \
7079 r = __netdev_printk(level, dev, &vaf); \
7085 EXPORT_SYMBOL(func);
7087 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
7088 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
7089 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
7090 define_netdev_printk_level(netdev_err
, KERN_ERR
);
7091 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
7092 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
7093 define_netdev_printk_level(netdev_info
, KERN_INFO
);
7095 static void __net_exit
netdev_exit(struct net
*net
)
7097 kfree(net
->dev_name_head
);
7098 kfree(net
->dev_index_head
);
7101 static struct pernet_operations __net_initdata netdev_net_ops
= {
7102 .init
= netdev_init
,
7103 .exit
= netdev_exit
,
7106 static void __net_exit
default_device_exit(struct net
*net
)
7108 struct net_device
*dev
, *aux
;
7110 * Push all migratable network devices back to the
7111 * initial network namespace
7114 for_each_netdev_safe(net
, dev
, aux
) {
7116 char fb_name
[IFNAMSIZ
];
7118 /* Ignore unmoveable devices (i.e. loopback) */
7119 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
7122 /* Leave virtual devices for the generic cleanup */
7123 if (dev
->rtnl_link_ops
)
7126 /* Push remaining network devices to init_net */
7127 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
7128 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
7130 pr_emerg("%s: failed to move %s to init_net: %d\n",
7131 __func__
, dev
->name
, err
);
7138 static void __net_exit
rtnl_lock_unregistering(struct list_head
*net_list
)
7140 /* Return with the rtnl_lock held when there are no network
7141 * devices unregistering in any network namespace in net_list.
7148 prepare_to_wait(&netdev_unregistering_wq
, &wait
,
7149 TASK_UNINTERRUPTIBLE
);
7150 unregistering
= false;
7152 list_for_each_entry(net
, net_list
, exit_list
) {
7153 if (net
->dev_unreg_count
> 0) {
7154 unregistering
= true;
7163 finish_wait(&netdev_unregistering_wq
, &wait
);
7166 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
7168 /* At exit all network devices most be removed from a network
7169 * namespace. Do this in the reverse order of registration.
7170 * Do this across as many network namespaces as possible to
7171 * improve batching efficiency.
7173 struct net_device
*dev
;
7175 LIST_HEAD(dev_kill_list
);
7177 /* To prevent network device cleanup code from dereferencing
7178 * loopback devices or network devices that have been freed
7179 * wait here for all pending unregistrations to complete,
7180 * before unregistring the loopback device and allowing the
7181 * network namespace be freed.
7183 * The netdev todo list containing all network devices
7184 * unregistrations that happen in default_device_exit_batch
7185 * will run in the rtnl_unlock() at the end of
7186 * default_device_exit_batch.
7188 rtnl_lock_unregistering(net_list
);
7189 list_for_each_entry(net
, net_list
, exit_list
) {
7190 for_each_netdev_reverse(net
, dev
) {
7191 if (dev
->rtnl_link_ops
&& dev
->rtnl_link_ops
->dellink
)
7192 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
7194 unregister_netdevice_queue(dev
, &dev_kill_list
);
7197 unregister_netdevice_many(&dev_kill_list
);
7201 static struct pernet_operations __net_initdata default_device_ops
= {
7202 .exit
= default_device_exit
,
7203 .exit_batch
= default_device_exit_batch
,
7207 * Initialize the DEV module. At boot time this walks the device list and
7208 * unhooks any devices that fail to initialise (normally hardware not
7209 * present) and leaves us with a valid list of present and active devices.
7214 * This is called single threaded during boot, so no need
7215 * to take the rtnl semaphore.
7217 static int __init
net_dev_init(void)
7219 int i
, rc
= -ENOMEM
;
7221 BUG_ON(!dev_boot_phase
);
7223 if (dev_proc_init())
7226 if (netdev_kobject_init())
7229 INIT_LIST_HEAD(&ptype_all
);
7230 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
7231 INIT_LIST_HEAD(&ptype_base
[i
]);
7233 INIT_LIST_HEAD(&offload_base
);
7235 if (register_pernet_subsys(&netdev_net_ops
))
7239 * Initialise the packet receive queues.
7242 for_each_possible_cpu(i
) {
7243 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
7245 skb_queue_head_init(&sd
->input_pkt_queue
);
7246 skb_queue_head_init(&sd
->process_queue
);
7247 INIT_LIST_HEAD(&sd
->poll_list
);
7248 sd
->output_queue_tailp
= &sd
->output_queue
;
7250 sd
->csd
.func
= rps_trigger_softirq
;
7255 sd
->backlog
.poll
= process_backlog
;
7256 sd
->backlog
.weight
= weight_p
;
7261 /* The loopback device is special if any other network devices
7262 * is present in a network namespace the loopback device must
7263 * be present. Since we now dynamically allocate and free the
7264 * loopback device ensure this invariant is maintained by
7265 * keeping the loopback device as the first device on the
7266 * list of network devices. Ensuring the loopback devices
7267 * is the first device that appears and the last network device
7270 if (register_pernet_device(&loopback_net_ops
))
7273 if (register_pernet_device(&default_device_ops
))
7276 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
7277 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
7279 hotcpu_notifier(dev_cpu_callback
, 0);
7286 subsys_initcall(net_dev_init
);