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
= this_cpu_ptr(&softnet_data
);
2157 q
->next_sched
= NULL
;
2158 *sd
->output_queue_tailp
= q
;
2159 sd
->output_queue_tailp
= &q
->next_sched
;
2160 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
2161 local_irq_restore(flags
);
2164 void __netif_schedule(struct Qdisc
*q
)
2166 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
2167 __netif_reschedule(q
);
2169 EXPORT_SYMBOL(__netif_schedule
);
2171 struct dev_kfree_skb_cb
{
2172 enum skb_free_reason reason
;
2175 static struct dev_kfree_skb_cb
*get_kfree_skb_cb(const struct sk_buff
*skb
)
2177 return (struct dev_kfree_skb_cb
*)skb
->cb
;
2180 void __dev_kfree_skb_irq(struct sk_buff
*skb
, enum skb_free_reason reason
)
2182 unsigned long flags
;
2184 if (likely(atomic_read(&skb
->users
) == 1)) {
2186 atomic_set(&skb
->users
, 0);
2187 } else if (likely(!atomic_dec_and_test(&skb
->users
))) {
2190 get_kfree_skb_cb(skb
)->reason
= reason
;
2191 local_irq_save(flags
);
2192 skb
->next
= __this_cpu_read(softnet_data
.completion_queue
);
2193 __this_cpu_write(softnet_data
.completion_queue
, skb
);
2194 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
2195 local_irq_restore(flags
);
2197 EXPORT_SYMBOL(__dev_kfree_skb_irq
);
2199 void __dev_kfree_skb_any(struct sk_buff
*skb
, enum skb_free_reason reason
)
2201 if (in_irq() || irqs_disabled())
2202 __dev_kfree_skb_irq(skb
, reason
);
2206 EXPORT_SYMBOL(__dev_kfree_skb_any
);
2210 * netif_device_detach - mark device as removed
2211 * @dev: network device
2213 * Mark device as removed from system and therefore no longer available.
2215 void netif_device_detach(struct net_device
*dev
)
2217 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
2218 netif_running(dev
)) {
2219 netif_tx_stop_all_queues(dev
);
2222 EXPORT_SYMBOL(netif_device_detach
);
2225 * netif_device_attach - mark device as attached
2226 * @dev: network device
2228 * Mark device as attached from system and restart if needed.
2230 void netif_device_attach(struct net_device
*dev
)
2232 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
2233 netif_running(dev
)) {
2234 netif_tx_wake_all_queues(dev
);
2235 __netdev_watchdog_up(dev
);
2238 EXPORT_SYMBOL(netif_device_attach
);
2240 static void skb_warn_bad_offload(const struct sk_buff
*skb
)
2242 static const netdev_features_t null_features
= 0;
2243 struct net_device
*dev
= skb
->dev
;
2244 const char *driver
= "";
2246 if (!net_ratelimit())
2249 if (dev
&& dev
->dev
.parent
)
2250 driver
= dev_driver_string(dev
->dev
.parent
);
2252 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2253 "gso_type=%d ip_summed=%d\n",
2254 driver
, dev
? &dev
->features
: &null_features
,
2255 skb
->sk
? &skb
->sk
->sk_route_caps
: &null_features
,
2256 skb
->len
, skb
->data_len
, skb_shinfo(skb
)->gso_size
,
2257 skb_shinfo(skb
)->gso_type
, skb
->ip_summed
);
2261 * Invalidate hardware checksum when packet is to be mangled, and
2262 * complete checksum manually on outgoing path.
2264 int skb_checksum_help(struct sk_buff
*skb
)
2267 int ret
= 0, offset
;
2269 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
2270 goto out_set_summed
;
2272 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
2273 skb_warn_bad_offload(skb
);
2277 /* Before computing a checksum, we should make sure no frag could
2278 * be modified by an external entity : checksum could be wrong.
2280 if (skb_has_shared_frag(skb
)) {
2281 ret
= __skb_linearize(skb
);
2286 offset
= skb_checksum_start_offset(skb
);
2287 BUG_ON(offset
>= skb_headlen(skb
));
2288 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
2290 offset
+= skb
->csum_offset
;
2291 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
2293 if (skb_cloned(skb
) &&
2294 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
2295 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2300 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
2302 skb
->ip_summed
= CHECKSUM_NONE
;
2306 EXPORT_SYMBOL(skb_checksum_help
);
2308 __be16
skb_network_protocol(struct sk_buff
*skb
, int *depth
)
2310 unsigned int vlan_depth
= skb
->mac_len
;
2311 __be16 type
= skb
->protocol
;
2313 /* Tunnel gso handlers can set protocol to ethernet. */
2314 if (type
== htons(ETH_P_TEB
)) {
2317 if (unlikely(!pskb_may_pull(skb
, sizeof(struct ethhdr
))))
2320 eth
= (struct ethhdr
*)skb_mac_header(skb
);
2321 type
= eth
->h_proto
;
2324 /* if skb->protocol is 802.1Q/AD then the header should already be
2325 * present at mac_len - VLAN_HLEN (if mac_len > 0), or at
2326 * ETH_HLEN otherwise
2328 if (type
== htons(ETH_P_8021Q
) || type
== htons(ETH_P_8021AD
)) {
2330 if (WARN_ON(vlan_depth
< VLAN_HLEN
))
2332 vlan_depth
-= VLAN_HLEN
;
2334 vlan_depth
= ETH_HLEN
;
2337 struct vlan_hdr
*vh
;
2339 if (unlikely(!pskb_may_pull(skb
,
2340 vlan_depth
+ VLAN_HLEN
)))
2343 vh
= (struct vlan_hdr
*)(skb
->data
+ vlan_depth
);
2344 type
= vh
->h_vlan_encapsulated_proto
;
2345 vlan_depth
+= VLAN_HLEN
;
2346 } while (type
== htons(ETH_P_8021Q
) ||
2347 type
== htons(ETH_P_8021AD
));
2350 *depth
= vlan_depth
;
2356 * skb_mac_gso_segment - mac layer segmentation handler.
2357 * @skb: buffer to segment
2358 * @features: features for the output path (see dev->features)
2360 struct sk_buff
*skb_mac_gso_segment(struct sk_buff
*skb
,
2361 netdev_features_t features
)
2363 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
2364 struct packet_offload
*ptype
;
2365 int vlan_depth
= skb
->mac_len
;
2366 __be16 type
= skb_network_protocol(skb
, &vlan_depth
);
2368 if (unlikely(!type
))
2369 return ERR_PTR(-EINVAL
);
2371 __skb_pull(skb
, vlan_depth
);
2374 list_for_each_entry_rcu(ptype
, &offload_base
, list
) {
2375 if (ptype
->type
== type
&& ptype
->callbacks
.gso_segment
) {
2376 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
2379 err
= ptype
->callbacks
.gso_send_check(skb
);
2380 segs
= ERR_PTR(err
);
2381 if (err
|| skb_gso_ok(skb
, features
))
2383 __skb_push(skb
, (skb
->data
-
2384 skb_network_header(skb
)));
2386 segs
= ptype
->callbacks
.gso_segment(skb
, features
);
2392 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
2396 EXPORT_SYMBOL(skb_mac_gso_segment
);
2399 /* openvswitch calls this on rx path, so we need a different check.
2401 static inline bool skb_needs_check(struct sk_buff
*skb
, bool tx_path
)
2404 return skb
->ip_summed
!= CHECKSUM_PARTIAL
;
2406 return skb
->ip_summed
== CHECKSUM_NONE
;
2410 * __skb_gso_segment - Perform segmentation on skb.
2411 * @skb: buffer to segment
2412 * @features: features for the output path (see dev->features)
2413 * @tx_path: whether it is called in TX path
2415 * This function segments the given skb and returns a list of segments.
2417 * It may return NULL if the skb requires no segmentation. This is
2418 * only possible when GSO is used for verifying header integrity.
2420 struct sk_buff
*__skb_gso_segment(struct sk_buff
*skb
,
2421 netdev_features_t features
, bool tx_path
)
2423 if (unlikely(skb_needs_check(skb
, tx_path
))) {
2426 skb_warn_bad_offload(skb
);
2428 err
= skb_cow_head(skb
, 0);
2430 return ERR_PTR(err
);
2433 SKB_GSO_CB(skb
)->mac_offset
= skb_headroom(skb
);
2434 SKB_GSO_CB(skb
)->encap_level
= 0;
2436 skb_reset_mac_header(skb
);
2437 skb_reset_mac_len(skb
);
2439 return skb_mac_gso_segment(skb
, features
);
2441 EXPORT_SYMBOL(__skb_gso_segment
);
2443 /* Take action when hardware reception checksum errors are detected. */
2445 void netdev_rx_csum_fault(struct net_device
*dev
)
2447 if (net_ratelimit()) {
2448 pr_err("%s: hw csum failure\n", dev
? dev
->name
: "<unknown>");
2452 EXPORT_SYMBOL(netdev_rx_csum_fault
);
2455 /* Actually, we should eliminate this check as soon as we know, that:
2456 * 1. IOMMU is present and allows to map all the memory.
2457 * 2. No high memory really exists on this machine.
2460 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
2462 #ifdef CONFIG_HIGHMEM
2464 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
2465 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2466 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2467 if (PageHighMem(skb_frag_page(frag
)))
2472 if (PCI_DMA_BUS_IS_PHYS
) {
2473 struct device
*pdev
= dev
->dev
.parent
;
2477 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2478 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2479 dma_addr_t addr
= page_to_phys(skb_frag_page(frag
));
2480 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
2489 void (*destructor
)(struct sk_buff
*skb
);
2492 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2494 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
2496 struct dev_gso_cb
*cb
;
2498 kfree_skb_list(skb
->next
);
2501 cb
= DEV_GSO_CB(skb
);
2503 cb
->destructor(skb
);
2507 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2508 * @skb: buffer to segment
2509 * @features: device features as applicable to this skb
2511 * This function segments the given skb and stores the list of segments
2514 static int dev_gso_segment(struct sk_buff
*skb
, netdev_features_t features
)
2516 struct sk_buff
*segs
;
2518 segs
= skb_gso_segment(skb
, features
);
2520 /* Verifying header integrity only. */
2525 return PTR_ERR(segs
);
2528 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
2529 skb
->destructor
= dev_gso_skb_destructor
;
2534 /* If MPLS offload request, verify we are testing hardware MPLS features
2535 * instead of standard features for the netdev.
2537 #ifdef CONFIG_NET_MPLS_GSO
2538 static netdev_features_t
net_mpls_features(struct sk_buff
*skb
,
2539 netdev_features_t features
,
2542 if (type
== htons(ETH_P_MPLS_UC
) || type
== htons(ETH_P_MPLS_MC
))
2543 features
&= skb
->dev
->mpls_features
;
2548 static netdev_features_t
net_mpls_features(struct sk_buff
*skb
,
2549 netdev_features_t features
,
2556 static netdev_features_t
harmonize_features(struct sk_buff
*skb
,
2557 netdev_features_t features
)
2562 type
= skb_network_protocol(skb
, &tmp
);
2563 features
= net_mpls_features(skb
, features
, type
);
2565 if (skb
->ip_summed
!= CHECKSUM_NONE
&&
2566 !can_checksum_protocol(features
, type
)) {
2567 features
&= ~NETIF_F_ALL_CSUM
;
2568 } else if (illegal_highdma(skb
->dev
, skb
)) {
2569 features
&= ~NETIF_F_SG
;
2575 netdev_features_t
netif_skb_features(struct sk_buff
*skb
)
2577 __be16 protocol
= skb
->protocol
;
2578 netdev_features_t features
= skb
->dev
->features
;
2580 if (skb_shinfo(skb
)->gso_segs
> skb
->dev
->gso_max_segs
)
2581 features
&= ~NETIF_F_GSO_MASK
;
2583 if (protocol
== htons(ETH_P_8021Q
) || protocol
== htons(ETH_P_8021AD
)) {
2584 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
2585 protocol
= veh
->h_vlan_encapsulated_proto
;
2586 } else if (!vlan_tx_tag_present(skb
)) {
2587 return harmonize_features(skb
, features
);
2590 features
&= (skb
->dev
->vlan_features
| NETIF_F_HW_VLAN_CTAG_TX
|
2591 NETIF_F_HW_VLAN_STAG_TX
);
2593 if (protocol
== htons(ETH_P_8021Q
) || protocol
== htons(ETH_P_8021AD
))
2594 features
&= NETIF_F_SG
| NETIF_F_HIGHDMA
| NETIF_F_FRAGLIST
|
2595 NETIF_F_GEN_CSUM
| NETIF_F_HW_VLAN_CTAG_TX
|
2596 NETIF_F_HW_VLAN_STAG_TX
;
2598 return harmonize_features(skb
, features
);
2600 EXPORT_SYMBOL(netif_skb_features
);
2602 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
2603 struct netdev_queue
*txq
)
2605 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2606 int rc
= NETDEV_TX_OK
;
2607 unsigned int skb_len
;
2609 if (likely(!skb
->next
)) {
2610 netdev_features_t features
;
2613 * If device doesn't need skb->dst, release it right now while
2614 * its hot in this cpu cache
2616 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2619 features
= netif_skb_features(skb
);
2621 if (vlan_tx_tag_present(skb
) &&
2622 !vlan_hw_offload_capable(features
, skb
->vlan_proto
)) {
2623 skb
= __vlan_put_tag(skb
, skb
->vlan_proto
,
2624 vlan_tx_tag_get(skb
));
2631 /* If encapsulation offload request, verify we are testing
2632 * hardware encapsulation features instead of standard
2633 * features for the netdev
2635 if (skb
->encapsulation
)
2636 features
&= dev
->hw_enc_features
;
2638 if (netif_needs_gso(skb
, features
)) {
2639 if (unlikely(dev_gso_segment(skb
, features
)))
2644 if (skb_needs_linearize(skb
, features
) &&
2645 __skb_linearize(skb
))
2648 /* If packet is not checksummed and device does not
2649 * support checksumming for this protocol, complete
2650 * checksumming here.
2652 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2653 if (skb
->encapsulation
)
2654 skb_set_inner_transport_header(skb
,
2655 skb_checksum_start_offset(skb
));
2657 skb_set_transport_header(skb
,
2658 skb_checksum_start_offset(skb
));
2659 if (!(features
& NETIF_F_ALL_CSUM
) &&
2660 skb_checksum_help(skb
))
2665 if (!list_empty(&ptype_all
))
2666 dev_queue_xmit_nit(skb
, dev
);
2669 trace_net_dev_start_xmit(skb
, dev
);
2670 rc
= ops
->ndo_start_xmit(skb
, dev
);
2671 trace_net_dev_xmit(skb
, rc
, dev
, skb_len
);
2672 if (rc
== NETDEV_TX_OK
)
2673 txq_trans_update(txq
);
2679 struct sk_buff
*nskb
= skb
->next
;
2681 skb
->next
= nskb
->next
;
2684 if (!list_empty(&ptype_all
))
2685 dev_queue_xmit_nit(nskb
, dev
);
2687 skb_len
= nskb
->len
;
2688 trace_net_dev_start_xmit(nskb
, dev
);
2689 rc
= ops
->ndo_start_xmit(nskb
, dev
);
2690 trace_net_dev_xmit(nskb
, rc
, dev
, skb_len
);
2691 if (unlikely(rc
!= NETDEV_TX_OK
)) {
2692 if (rc
& ~NETDEV_TX_MASK
)
2693 goto out_kfree_gso_skb
;
2694 nskb
->next
= skb
->next
;
2698 txq_trans_update(txq
);
2699 if (unlikely(netif_xmit_stopped(txq
) && skb
->next
))
2700 return NETDEV_TX_BUSY
;
2701 } while (skb
->next
);
2704 if (likely(skb
->next
== NULL
)) {
2705 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
2714 EXPORT_SYMBOL_GPL(dev_hard_start_xmit
);
2716 static void qdisc_pkt_len_init(struct sk_buff
*skb
)
2718 const struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
2720 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
2722 /* To get more precise estimation of bytes sent on wire,
2723 * we add to pkt_len the headers size of all segments
2725 if (shinfo
->gso_size
) {
2726 unsigned int hdr_len
;
2727 u16 gso_segs
= shinfo
->gso_segs
;
2729 /* mac layer + network layer */
2730 hdr_len
= skb_transport_header(skb
) - skb_mac_header(skb
);
2732 /* + transport layer */
2733 if (likely(shinfo
->gso_type
& (SKB_GSO_TCPV4
| SKB_GSO_TCPV6
)))
2734 hdr_len
+= tcp_hdrlen(skb
);
2736 hdr_len
+= sizeof(struct udphdr
);
2738 if (shinfo
->gso_type
& SKB_GSO_DODGY
)
2739 gso_segs
= DIV_ROUND_UP(skb
->len
- hdr_len
,
2742 qdisc_skb_cb(skb
)->pkt_len
+= (gso_segs
- 1) * hdr_len
;
2746 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2747 struct net_device
*dev
,
2748 struct netdev_queue
*txq
)
2750 spinlock_t
*root_lock
= qdisc_lock(q
);
2754 qdisc_pkt_len_init(skb
);
2755 qdisc_calculate_pkt_len(skb
, q
);
2757 * Heuristic to force contended enqueues to serialize on a
2758 * separate lock before trying to get qdisc main lock.
2759 * This permits __QDISC___STATE_RUNNING owner to get the lock more
2760 * often and dequeue packets faster.
2762 contended
= qdisc_is_running(q
);
2763 if (unlikely(contended
))
2764 spin_lock(&q
->busylock
);
2766 spin_lock(root_lock
);
2767 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2770 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2771 qdisc_run_begin(q
)) {
2773 * This is a work-conserving queue; there are no old skbs
2774 * waiting to be sent out; and the qdisc is not running -
2775 * xmit the skb directly.
2777 if (!(dev
->priv_flags
& IFF_XMIT_DST_RELEASE
))
2780 qdisc_bstats_update(q
, skb
);
2782 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
)) {
2783 if (unlikely(contended
)) {
2784 spin_unlock(&q
->busylock
);
2791 rc
= NET_XMIT_SUCCESS
;
2794 rc
= q
->enqueue(skb
, q
) & NET_XMIT_MASK
;
2795 if (qdisc_run_begin(q
)) {
2796 if (unlikely(contended
)) {
2797 spin_unlock(&q
->busylock
);
2803 spin_unlock(root_lock
);
2804 if (unlikely(contended
))
2805 spin_unlock(&q
->busylock
);
2809 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2810 static void skb_update_prio(struct sk_buff
*skb
)
2812 struct netprio_map
*map
= rcu_dereference_bh(skb
->dev
->priomap
);
2814 if (!skb
->priority
&& skb
->sk
&& map
) {
2815 unsigned int prioidx
= skb
->sk
->sk_cgrp_prioidx
;
2817 if (prioidx
< map
->priomap_len
)
2818 skb
->priority
= map
->priomap
[prioidx
];
2822 #define skb_update_prio(skb)
2825 static DEFINE_PER_CPU(int, xmit_recursion
);
2826 #define RECURSION_LIMIT 10
2829 * dev_loopback_xmit - loop back @skb
2830 * @skb: buffer to transmit
2832 int dev_loopback_xmit(struct sk_buff
*skb
)
2834 skb_reset_mac_header(skb
);
2835 __skb_pull(skb
, skb_network_offset(skb
));
2836 skb
->pkt_type
= PACKET_LOOPBACK
;
2837 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
2838 WARN_ON(!skb_dst(skb
));
2843 EXPORT_SYMBOL(dev_loopback_xmit
);
2846 * __dev_queue_xmit - transmit a buffer
2847 * @skb: buffer to transmit
2848 * @accel_priv: private data used for L2 forwarding offload
2850 * Queue a buffer for transmission to a network device. The caller must
2851 * have set the device and priority and built the buffer before calling
2852 * this function. The function can be called from an interrupt.
2854 * A negative errno code is returned on a failure. A success does not
2855 * guarantee the frame will be transmitted as it may be dropped due
2856 * to congestion or traffic shaping.
2858 * -----------------------------------------------------------------------------------
2859 * I notice this method can also return errors from the queue disciplines,
2860 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2863 * Regardless of the return value, the skb is consumed, so it is currently
2864 * difficult to retry a send to this method. (You can bump the ref count
2865 * before sending to hold a reference for retry if you are careful.)
2867 * When calling this method, interrupts MUST be enabled. This is because
2868 * the BH enable code must have IRQs enabled so that it will not deadlock.
2871 static int __dev_queue_xmit(struct sk_buff
*skb
, void *accel_priv
)
2873 struct net_device
*dev
= skb
->dev
;
2874 struct netdev_queue
*txq
;
2878 skb_reset_mac_header(skb
);
2880 if (unlikely(skb_shinfo(skb
)->tx_flags
& SKBTX_SCHED_TSTAMP
))
2881 __skb_tstamp_tx(skb
, NULL
, skb
->sk
, SCM_TSTAMP_SCHED
);
2883 /* Disable soft irqs for various locks below. Also
2884 * stops preemption for RCU.
2888 skb_update_prio(skb
);
2890 txq
= netdev_pick_tx(dev
, skb
, accel_priv
);
2891 q
= rcu_dereference_bh(txq
->qdisc
);
2893 #ifdef CONFIG_NET_CLS_ACT
2894 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2896 trace_net_dev_queue(skb
);
2898 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2902 /* The device has no queue. Common case for software devices:
2903 loopback, all the sorts of tunnels...
2905 Really, it is unlikely that netif_tx_lock protection is necessary
2906 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2908 However, it is possible, that they rely on protection
2911 Check this and shot the lock. It is not prone from deadlocks.
2912 Either shot noqueue qdisc, it is even simpler 8)
2914 if (dev
->flags
& IFF_UP
) {
2915 int cpu
= smp_processor_id(); /* ok because BHs are off */
2917 if (txq
->xmit_lock_owner
!= cpu
) {
2919 if (__this_cpu_read(xmit_recursion
) > RECURSION_LIMIT
)
2920 goto recursion_alert
;
2922 HARD_TX_LOCK(dev
, txq
, cpu
);
2924 if (!netif_xmit_stopped(txq
)) {
2925 __this_cpu_inc(xmit_recursion
);
2926 rc
= dev_hard_start_xmit(skb
, dev
, txq
);
2927 __this_cpu_dec(xmit_recursion
);
2928 if (dev_xmit_complete(rc
)) {
2929 HARD_TX_UNLOCK(dev
, txq
);
2933 HARD_TX_UNLOCK(dev
, txq
);
2934 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2937 /* Recursion is detected! It is possible,
2941 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2947 rcu_read_unlock_bh();
2949 atomic_long_inc(&dev
->tx_dropped
);
2953 rcu_read_unlock_bh();
2957 int dev_queue_xmit(struct sk_buff
*skb
)
2959 return __dev_queue_xmit(skb
, NULL
);
2961 EXPORT_SYMBOL(dev_queue_xmit
);
2963 int dev_queue_xmit_accel(struct sk_buff
*skb
, void *accel_priv
)
2965 return __dev_queue_xmit(skb
, accel_priv
);
2967 EXPORT_SYMBOL(dev_queue_xmit_accel
);
2970 /*=======================================================================
2972 =======================================================================*/
2974 int netdev_max_backlog __read_mostly
= 1000;
2975 EXPORT_SYMBOL(netdev_max_backlog
);
2977 int netdev_tstamp_prequeue __read_mostly
= 1;
2978 int netdev_budget __read_mostly
= 300;
2979 int weight_p __read_mostly
= 64; /* old backlog weight */
2981 /* Called with irq disabled */
2982 static inline void ____napi_schedule(struct softnet_data
*sd
,
2983 struct napi_struct
*napi
)
2985 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
2986 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2991 /* One global table that all flow-based protocols share. */
2992 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
2993 EXPORT_SYMBOL(rps_sock_flow_table
);
2995 struct static_key rps_needed __read_mostly
;
2997 static struct rps_dev_flow
*
2998 set_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2999 struct rps_dev_flow
*rflow
, u16 next_cpu
)
3001 if (next_cpu
!= RPS_NO_CPU
) {
3002 #ifdef CONFIG_RFS_ACCEL
3003 struct netdev_rx_queue
*rxqueue
;
3004 struct rps_dev_flow_table
*flow_table
;
3005 struct rps_dev_flow
*old_rflow
;
3010 /* Should we steer this flow to a different hardware queue? */
3011 if (!skb_rx_queue_recorded(skb
) || !dev
->rx_cpu_rmap
||
3012 !(dev
->features
& NETIF_F_NTUPLE
))
3014 rxq_index
= cpu_rmap_lookup_index(dev
->rx_cpu_rmap
, next_cpu
);
3015 if (rxq_index
== skb_get_rx_queue(skb
))
3018 rxqueue
= dev
->_rx
+ rxq_index
;
3019 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3022 flow_id
= skb_get_hash(skb
) & flow_table
->mask
;
3023 rc
= dev
->netdev_ops
->ndo_rx_flow_steer(dev
, skb
,
3024 rxq_index
, flow_id
);
3028 rflow
= &flow_table
->flows
[flow_id
];
3030 if (old_rflow
->filter
== rflow
->filter
)
3031 old_rflow
->filter
= RPS_NO_FILTER
;
3035 per_cpu(softnet_data
, next_cpu
).input_queue_head
;
3038 rflow
->cpu
= next_cpu
;
3043 * get_rps_cpu is called from netif_receive_skb and returns the target
3044 * CPU from the RPS map of the receiving queue for a given skb.
3045 * rcu_read_lock must be held on entry.
3047 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
3048 struct rps_dev_flow
**rflowp
)
3050 struct netdev_rx_queue
*rxqueue
;
3051 struct rps_map
*map
;
3052 struct rps_dev_flow_table
*flow_table
;
3053 struct rps_sock_flow_table
*sock_flow_table
;
3058 if (skb_rx_queue_recorded(skb
)) {
3059 u16 index
= skb_get_rx_queue(skb
);
3060 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
3061 WARN_ONCE(dev
->real_num_rx_queues
> 1,
3062 "%s received packet on queue %u, but number "
3063 "of RX queues is %u\n",
3064 dev
->name
, index
, dev
->real_num_rx_queues
);
3067 rxqueue
= dev
->_rx
+ index
;
3071 map
= rcu_dereference(rxqueue
->rps_map
);
3073 if (map
->len
== 1 &&
3074 !rcu_access_pointer(rxqueue
->rps_flow_table
)) {
3075 tcpu
= map
->cpus
[0];
3076 if (cpu_online(tcpu
))
3080 } else if (!rcu_access_pointer(rxqueue
->rps_flow_table
)) {
3084 skb_reset_network_header(skb
);
3085 hash
= skb_get_hash(skb
);
3089 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3090 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
3091 if (flow_table
&& sock_flow_table
) {
3093 struct rps_dev_flow
*rflow
;
3095 rflow
= &flow_table
->flows
[hash
& flow_table
->mask
];
3098 next_cpu
= sock_flow_table
->ents
[hash
& sock_flow_table
->mask
];
3101 * If the desired CPU (where last recvmsg was done) is
3102 * different from current CPU (one in the rx-queue flow
3103 * table entry), switch if one of the following holds:
3104 * - Current CPU is unset (equal to RPS_NO_CPU).
3105 * - Current CPU is offline.
3106 * - The current CPU's queue tail has advanced beyond the
3107 * last packet that was enqueued using this table entry.
3108 * This guarantees that all previous packets for the flow
3109 * have been dequeued, thus preserving in order delivery.
3111 if (unlikely(tcpu
!= next_cpu
) &&
3112 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
3113 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
3114 rflow
->last_qtail
)) >= 0)) {
3116 rflow
= set_rps_cpu(dev
, skb
, rflow
, next_cpu
);
3119 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
3127 tcpu
= map
->cpus
[((u64
) hash
* map
->len
) >> 32];
3129 if (cpu_online(tcpu
)) {
3139 #ifdef CONFIG_RFS_ACCEL
3142 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3143 * @dev: Device on which the filter was set
3144 * @rxq_index: RX queue index
3145 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3146 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3148 * Drivers that implement ndo_rx_flow_steer() should periodically call
3149 * this function for each installed filter and remove the filters for
3150 * which it returns %true.
3152 bool rps_may_expire_flow(struct net_device
*dev
, u16 rxq_index
,
3153 u32 flow_id
, u16 filter_id
)
3155 struct netdev_rx_queue
*rxqueue
= dev
->_rx
+ rxq_index
;
3156 struct rps_dev_flow_table
*flow_table
;
3157 struct rps_dev_flow
*rflow
;
3162 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3163 if (flow_table
&& flow_id
<= flow_table
->mask
) {
3164 rflow
= &flow_table
->flows
[flow_id
];
3165 cpu
= ACCESS_ONCE(rflow
->cpu
);
3166 if (rflow
->filter
== filter_id
&& cpu
!= RPS_NO_CPU
&&
3167 ((int)(per_cpu(softnet_data
, cpu
).input_queue_head
-
3168 rflow
->last_qtail
) <
3169 (int)(10 * flow_table
->mask
)))
3175 EXPORT_SYMBOL(rps_may_expire_flow
);
3177 #endif /* CONFIG_RFS_ACCEL */
3179 /* Called from hardirq (IPI) context */
3180 static void rps_trigger_softirq(void *data
)
3182 struct softnet_data
*sd
= data
;
3184 ____napi_schedule(sd
, &sd
->backlog
);
3188 #endif /* CONFIG_RPS */
3191 * Check if this softnet_data structure is another cpu one
3192 * If yes, queue it to our IPI list and return 1
3195 static int rps_ipi_queued(struct softnet_data
*sd
)
3198 struct softnet_data
*mysd
= this_cpu_ptr(&softnet_data
);
3201 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
3202 mysd
->rps_ipi_list
= sd
;
3204 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3207 #endif /* CONFIG_RPS */
3211 #ifdef CONFIG_NET_FLOW_LIMIT
3212 int netdev_flow_limit_table_len __read_mostly
= (1 << 12);
3215 static bool skb_flow_limit(struct sk_buff
*skb
, unsigned int qlen
)
3217 #ifdef CONFIG_NET_FLOW_LIMIT
3218 struct sd_flow_limit
*fl
;
3219 struct softnet_data
*sd
;
3220 unsigned int old_flow
, new_flow
;
3222 if (qlen
< (netdev_max_backlog
>> 1))
3225 sd
= this_cpu_ptr(&softnet_data
);
3228 fl
= rcu_dereference(sd
->flow_limit
);
3230 new_flow
= skb_get_hash(skb
) & (fl
->num_buckets
- 1);
3231 old_flow
= fl
->history
[fl
->history_head
];
3232 fl
->history
[fl
->history_head
] = new_flow
;
3235 fl
->history_head
&= FLOW_LIMIT_HISTORY
- 1;
3237 if (likely(fl
->buckets
[old_flow
]))
3238 fl
->buckets
[old_flow
]--;
3240 if (++fl
->buckets
[new_flow
] > (FLOW_LIMIT_HISTORY
>> 1)) {
3252 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3253 * queue (may be a remote CPU queue).
3255 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
3256 unsigned int *qtail
)
3258 struct softnet_data
*sd
;
3259 unsigned long flags
;
3262 sd
= &per_cpu(softnet_data
, cpu
);
3264 local_irq_save(flags
);
3267 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3268 if (qlen
<= netdev_max_backlog
&& !skb_flow_limit(skb
, qlen
)) {
3269 if (skb_queue_len(&sd
->input_pkt_queue
)) {
3271 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
3272 input_queue_tail_incr_save(sd
, qtail
);
3274 local_irq_restore(flags
);
3275 return NET_RX_SUCCESS
;
3278 /* Schedule NAPI for backlog device
3279 * We can use non atomic operation since we own the queue lock
3281 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
3282 if (!rps_ipi_queued(sd
))
3283 ____napi_schedule(sd
, &sd
->backlog
);
3291 local_irq_restore(flags
);
3293 atomic_long_inc(&skb
->dev
->rx_dropped
);
3298 static int netif_rx_internal(struct sk_buff
*skb
)
3302 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3304 trace_netif_rx(skb
);
3306 if (static_key_false(&rps_needed
)) {
3307 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3313 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3315 cpu
= smp_processor_id();
3317 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3325 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
3332 * netif_rx - post buffer to the network code
3333 * @skb: buffer to post
3335 * This function receives a packet from a device driver and queues it for
3336 * the upper (protocol) levels to process. It always succeeds. The buffer
3337 * may be dropped during processing for congestion control or by the
3341 * NET_RX_SUCCESS (no congestion)
3342 * NET_RX_DROP (packet was dropped)
3346 int netif_rx(struct sk_buff
*skb
)
3348 trace_netif_rx_entry(skb
);
3350 return netif_rx_internal(skb
);
3352 EXPORT_SYMBOL(netif_rx
);
3354 int netif_rx_ni(struct sk_buff
*skb
)
3358 trace_netif_rx_ni_entry(skb
);
3361 err
= netif_rx_internal(skb
);
3362 if (local_softirq_pending())
3368 EXPORT_SYMBOL(netif_rx_ni
);
3370 static void net_tx_action(struct softirq_action
*h
)
3372 struct softnet_data
*sd
= this_cpu_ptr(&softnet_data
);
3374 if (sd
->completion_queue
) {
3375 struct sk_buff
*clist
;
3377 local_irq_disable();
3378 clist
= sd
->completion_queue
;
3379 sd
->completion_queue
= NULL
;
3383 struct sk_buff
*skb
= clist
;
3384 clist
= clist
->next
;
3386 WARN_ON(atomic_read(&skb
->users
));
3387 if (likely(get_kfree_skb_cb(skb
)->reason
== SKB_REASON_CONSUMED
))
3388 trace_consume_skb(skb
);
3390 trace_kfree_skb(skb
, net_tx_action
);
3395 if (sd
->output_queue
) {
3398 local_irq_disable();
3399 head
= sd
->output_queue
;
3400 sd
->output_queue
= NULL
;
3401 sd
->output_queue_tailp
= &sd
->output_queue
;
3405 struct Qdisc
*q
= head
;
3406 spinlock_t
*root_lock
;
3408 head
= head
->next_sched
;
3410 root_lock
= qdisc_lock(q
);
3411 if (spin_trylock(root_lock
)) {
3412 smp_mb__before_atomic();
3413 clear_bit(__QDISC_STATE_SCHED
,
3416 spin_unlock(root_lock
);
3418 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
3420 __netif_reschedule(q
);
3422 smp_mb__before_atomic();
3423 clear_bit(__QDISC_STATE_SCHED
,
3431 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3432 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3433 /* This hook is defined here for ATM LANE */
3434 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
3435 unsigned char *addr
) __read_mostly
;
3436 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
3439 #ifdef CONFIG_NET_CLS_ACT
3440 /* TODO: Maybe we should just force sch_ingress to be compiled in
3441 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3442 * a compare and 2 stores extra right now if we dont have it on
3443 * but have CONFIG_NET_CLS_ACT
3444 * NOTE: This doesn't stop any functionality; if you dont have
3445 * the ingress scheduler, you just can't add policies on ingress.
3448 static int ing_filter(struct sk_buff
*skb
, struct netdev_queue
*rxq
)
3450 struct net_device
*dev
= skb
->dev
;
3451 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
3452 int result
= TC_ACT_OK
;
3455 if (unlikely(MAX_RED_LOOP
< ttl
++)) {
3456 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3457 skb
->skb_iif
, dev
->ifindex
);
3461 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
3462 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
3465 if (q
!= &noop_qdisc
) {
3466 spin_lock(qdisc_lock(q
));
3467 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
3468 result
= qdisc_enqueue_root(skb
, q
);
3469 spin_unlock(qdisc_lock(q
));
3475 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
3476 struct packet_type
**pt_prev
,
3477 int *ret
, struct net_device
*orig_dev
)
3479 struct netdev_queue
*rxq
= rcu_dereference(skb
->dev
->ingress_queue
);
3481 if (!rxq
|| rxq
->qdisc
== &noop_qdisc
)
3485 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
3489 switch (ing_filter(skb
, rxq
)) {
3503 * netdev_rx_handler_register - register receive handler
3504 * @dev: device to register a handler for
3505 * @rx_handler: receive handler to register
3506 * @rx_handler_data: data pointer that is used by rx handler
3508 * Register a receive handler for a device. This handler will then be
3509 * called from __netif_receive_skb. A negative errno code is returned
3512 * The caller must hold the rtnl_mutex.
3514 * For a general description of rx_handler, see enum rx_handler_result.
3516 int netdev_rx_handler_register(struct net_device
*dev
,
3517 rx_handler_func_t
*rx_handler
,
3518 void *rx_handler_data
)
3522 if (dev
->rx_handler
)
3525 /* Note: rx_handler_data must be set before rx_handler */
3526 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
3527 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
3531 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
3534 * netdev_rx_handler_unregister - unregister receive handler
3535 * @dev: device to unregister a handler from
3537 * Unregister a receive handler from a device.
3539 * The caller must hold the rtnl_mutex.
3541 void netdev_rx_handler_unregister(struct net_device
*dev
)
3545 RCU_INIT_POINTER(dev
->rx_handler
, NULL
);
3546 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3547 * section has a guarantee to see a non NULL rx_handler_data
3551 RCU_INIT_POINTER(dev
->rx_handler_data
, NULL
);
3553 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
3556 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3557 * the special handling of PFMEMALLOC skbs.
3559 static bool skb_pfmemalloc_protocol(struct sk_buff
*skb
)
3561 switch (skb
->protocol
) {
3562 case htons(ETH_P_ARP
):
3563 case htons(ETH_P_IP
):
3564 case htons(ETH_P_IPV6
):
3565 case htons(ETH_P_8021Q
):
3566 case htons(ETH_P_8021AD
):
3573 static int __netif_receive_skb_core(struct sk_buff
*skb
, bool pfmemalloc
)
3575 struct packet_type
*ptype
, *pt_prev
;
3576 rx_handler_func_t
*rx_handler
;
3577 struct net_device
*orig_dev
;
3578 struct net_device
*null_or_dev
;
3579 bool deliver_exact
= false;
3580 int ret
= NET_RX_DROP
;
3583 net_timestamp_check(!netdev_tstamp_prequeue
, skb
);
3585 trace_netif_receive_skb(skb
);
3587 orig_dev
= skb
->dev
;
3589 skb_reset_network_header(skb
);
3590 if (!skb_transport_header_was_set(skb
))
3591 skb_reset_transport_header(skb
);
3592 skb_reset_mac_len(skb
);
3599 skb
->skb_iif
= skb
->dev
->ifindex
;
3601 __this_cpu_inc(softnet_data
.processed
);
3603 if (skb
->protocol
== cpu_to_be16(ETH_P_8021Q
) ||
3604 skb
->protocol
== cpu_to_be16(ETH_P_8021AD
)) {
3605 skb
= skb_vlan_untag(skb
);
3610 #ifdef CONFIG_NET_CLS_ACT
3611 if (skb
->tc_verd
& TC_NCLS
) {
3612 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
3620 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
3621 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
3623 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3629 #ifdef CONFIG_NET_CLS_ACT
3630 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
3636 if (pfmemalloc
&& !skb_pfmemalloc_protocol(skb
))
3639 if (vlan_tx_tag_present(skb
)) {
3641 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3644 if (vlan_do_receive(&skb
))
3646 else if (unlikely(!skb
))
3650 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
3653 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3656 switch (rx_handler(&skb
)) {
3657 case RX_HANDLER_CONSUMED
:
3658 ret
= NET_RX_SUCCESS
;
3660 case RX_HANDLER_ANOTHER
:
3662 case RX_HANDLER_EXACT
:
3663 deliver_exact
= true;
3664 case RX_HANDLER_PASS
:
3671 if (unlikely(vlan_tx_tag_present(skb
))) {
3672 if (vlan_tx_tag_get_id(skb
))
3673 skb
->pkt_type
= PACKET_OTHERHOST
;
3674 /* Note: we might in the future use prio bits
3675 * and set skb->priority like in vlan_do_receive()
3676 * For the time being, just ignore Priority Code Point
3681 /* deliver only exact match when indicated */
3682 null_or_dev
= deliver_exact
? skb
->dev
: NULL
;
3684 type
= skb
->protocol
;
3685 list_for_each_entry_rcu(ptype
,
3686 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
3687 if (ptype
->type
== type
&&
3688 (ptype
->dev
== null_or_dev
|| ptype
->dev
== skb
->dev
||
3689 ptype
->dev
== orig_dev
)) {
3691 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3697 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
3700 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
3703 atomic_long_inc(&skb
->dev
->rx_dropped
);
3705 /* Jamal, now you will not able to escape explaining
3706 * me how you were going to use this. :-)
3716 static int __netif_receive_skb(struct sk_buff
*skb
)
3720 if (sk_memalloc_socks() && skb_pfmemalloc(skb
)) {
3721 unsigned long pflags
= current
->flags
;
3724 * PFMEMALLOC skbs are special, they should
3725 * - be delivered to SOCK_MEMALLOC sockets only
3726 * - stay away from userspace
3727 * - have bounded memory usage
3729 * Use PF_MEMALLOC as this saves us from propagating the allocation
3730 * context down to all allocation sites.
3732 current
->flags
|= PF_MEMALLOC
;
3733 ret
= __netif_receive_skb_core(skb
, true);
3734 tsk_restore_flags(current
, pflags
, PF_MEMALLOC
);
3736 ret
= __netif_receive_skb_core(skb
, false);
3741 static int netif_receive_skb_internal(struct sk_buff
*skb
)
3743 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3745 if (skb_defer_rx_timestamp(skb
))
3746 return NET_RX_SUCCESS
;
3749 if (static_key_false(&rps_needed
)) {
3750 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3755 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3758 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3765 return __netif_receive_skb(skb
);
3769 * netif_receive_skb - process receive buffer from network
3770 * @skb: buffer to process
3772 * netif_receive_skb() is the main receive data processing function.
3773 * It always succeeds. The buffer may be dropped during processing
3774 * for congestion control or by the protocol layers.
3776 * This function may only be called from softirq context and interrupts
3777 * should be enabled.
3779 * Return values (usually ignored):
3780 * NET_RX_SUCCESS: no congestion
3781 * NET_RX_DROP: packet was dropped
3783 int netif_receive_skb(struct sk_buff
*skb
)
3785 trace_netif_receive_skb_entry(skb
);
3787 return netif_receive_skb_internal(skb
);
3789 EXPORT_SYMBOL(netif_receive_skb
);
3791 /* Network device is going away, flush any packets still pending
3792 * Called with irqs disabled.
3794 static void flush_backlog(void *arg
)
3796 struct net_device
*dev
= arg
;
3797 struct softnet_data
*sd
= this_cpu_ptr(&softnet_data
);
3798 struct sk_buff
*skb
, *tmp
;
3801 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3802 if (skb
->dev
== dev
) {
3803 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3805 input_queue_head_incr(sd
);
3810 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3811 if (skb
->dev
== dev
) {
3812 __skb_unlink(skb
, &sd
->process_queue
);
3814 input_queue_head_incr(sd
);
3819 static int napi_gro_complete(struct sk_buff
*skb
)
3821 struct packet_offload
*ptype
;
3822 __be16 type
= skb
->protocol
;
3823 struct list_head
*head
= &offload_base
;
3826 BUILD_BUG_ON(sizeof(struct napi_gro_cb
) > sizeof(skb
->cb
));
3828 if (NAPI_GRO_CB(skb
)->count
== 1) {
3829 skb_shinfo(skb
)->gso_size
= 0;
3834 list_for_each_entry_rcu(ptype
, head
, list
) {
3835 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_complete
)
3838 err
= ptype
->callbacks
.gro_complete(skb
, 0);
3844 WARN_ON(&ptype
->list
== head
);
3846 return NET_RX_SUCCESS
;
3850 return netif_receive_skb_internal(skb
);
3853 /* napi->gro_list contains packets ordered by age.
3854 * youngest packets at the head of it.
3855 * Complete skbs in reverse order to reduce latencies.
3857 void napi_gro_flush(struct napi_struct
*napi
, bool flush_old
)
3859 struct sk_buff
*skb
, *prev
= NULL
;
3861 /* scan list and build reverse chain */
3862 for (skb
= napi
->gro_list
; skb
!= NULL
; skb
= skb
->next
) {
3867 for (skb
= prev
; skb
; skb
= prev
) {
3870 if (flush_old
&& NAPI_GRO_CB(skb
)->age
== jiffies
)
3874 napi_gro_complete(skb
);
3878 napi
->gro_list
= NULL
;
3880 EXPORT_SYMBOL(napi_gro_flush
);
3882 static void gro_list_prepare(struct napi_struct
*napi
, struct sk_buff
*skb
)
3885 unsigned int maclen
= skb
->dev
->hard_header_len
;
3886 u32 hash
= skb_get_hash_raw(skb
);
3888 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3889 unsigned long diffs
;
3891 NAPI_GRO_CB(p
)->flush
= 0;
3893 if (hash
!= skb_get_hash_raw(p
)) {
3894 NAPI_GRO_CB(p
)->same_flow
= 0;
3898 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
3899 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
3900 if (maclen
== ETH_HLEN
)
3901 diffs
|= compare_ether_header(skb_mac_header(p
),
3902 skb_mac_header(skb
));
3904 diffs
= memcmp(skb_mac_header(p
),
3905 skb_mac_header(skb
),
3907 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
3911 static void skb_gro_reset_offset(struct sk_buff
*skb
)
3913 const struct skb_shared_info
*pinfo
= skb_shinfo(skb
);
3914 const skb_frag_t
*frag0
= &pinfo
->frags
[0];
3916 NAPI_GRO_CB(skb
)->data_offset
= 0;
3917 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3918 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3920 if (skb_mac_header(skb
) == skb_tail_pointer(skb
) &&
3922 !PageHighMem(skb_frag_page(frag0
))) {
3923 NAPI_GRO_CB(skb
)->frag0
= skb_frag_address(frag0
);
3924 NAPI_GRO_CB(skb
)->frag0_len
= skb_frag_size(frag0
);
3928 static void gro_pull_from_frag0(struct sk_buff
*skb
, int grow
)
3930 struct skb_shared_info
*pinfo
= skb_shinfo(skb
);
3932 BUG_ON(skb
->end
- skb
->tail
< grow
);
3934 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3936 skb
->data_len
-= grow
;
3939 pinfo
->frags
[0].page_offset
+= grow
;
3940 skb_frag_size_sub(&pinfo
->frags
[0], grow
);
3942 if (unlikely(!skb_frag_size(&pinfo
->frags
[0]))) {
3943 skb_frag_unref(skb
, 0);
3944 memmove(pinfo
->frags
, pinfo
->frags
+ 1,
3945 --pinfo
->nr_frags
* sizeof(pinfo
->frags
[0]));
3949 static enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3951 struct sk_buff
**pp
= NULL
;
3952 struct packet_offload
*ptype
;
3953 __be16 type
= skb
->protocol
;
3954 struct list_head
*head
= &offload_base
;
3956 enum gro_result ret
;
3959 if (!(skb
->dev
->features
& NETIF_F_GRO
))
3962 if (skb_is_gso(skb
) || skb_has_frag_list(skb
))
3965 gro_list_prepare(napi
, skb
);
3966 NAPI_GRO_CB(skb
)->csum
= skb
->csum
; /* Needed for CHECKSUM_COMPLETE */
3969 list_for_each_entry_rcu(ptype
, head
, list
) {
3970 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_receive
)
3973 skb_set_network_header(skb
, skb_gro_offset(skb
));
3974 skb_reset_mac_len(skb
);
3975 NAPI_GRO_CB(skb
)->same_flow
= 0;
3976 NAPI_GRO_CB(skb
)->flush
= 0;
3977 NAPI_GRO_CB(skb
)->free
= 0;
3978 NAPI_GRO_CB(skb
)->udp_mark
= 0;
3980 pp
= ptype
->callbacks
.gro_receive(&napi
->gro_list
, skb
);
3985 if (&ptype
->list
== head
)
3988 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3989 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3992 struct sk_buff
*nskb
= *pp
;
3996 napi_gro_complete(nskb
);
4003 if (NAPI_GRO_CB(skb
)->flush
)
4006 if (unlikely(napi
->gro_count
>= MAX_GRO_SKBS
)) {
4007 struct sk_buff
*nskb
= napi
->gro_list
;
4009 /* locate the end of the list to select the 'oldest' flow */
4010 while (nskb
->next
) {
4016 napi_gro_complete(nskb
);
4020 NAPI_GRO_CB(skb
)->count
= 1;
4021 NAPI_GRO_CB(skb
)->age
= jiffies
;
4022 NAPI_GRO_CB(skb
)->last
= skb
;
4023 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
4024 skb
->next
= napi
->gro_list
;
4025 napi
->gro_list
= skb
;
4029 grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
4031 gro_pull_from_frag0(skb
, grow
);
4040 struct packet_offload
*gro_find_receive_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_receive
)
4052 EXPORT_SYMBOL(gro_find_receive_by_type
);
4054 struct packet_offload
*gro_find_complete_by_type(__be16 type
)
4056 struct list_head
*offload_head
= &offload_base
;
4057 struct packet_offload
*ptype
;
4059 list_for_each_entry_rcu(ptype
, offload_head
, list
) {
4060 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_complete
)
4066 EXPORT_SYMBOL(gro_find_complete_by_type
);
4068 static gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
4072 if (netif_receive_skb_internal(skb
))
4080 case GRO_MERGED_FREE
:
4081 if (NAPI_GRO_CB(skb
)->free
== NAPI_GRO_FREE_STOLEN_HEAD
)
4082 kmem_cache_free(skbuff_head_cache
, skb
);
4095 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
4097 trace_napi_gro_receive_entry(skb
);
4099 skb_gro_reset_offset(skb
);
4101 return napi_skb_finish(dev_gro_receive(napi
, skb
), skb
);
4103 EXPORT_SYMBOL(napi_gro_receive
);
4105 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
4107 __skb_pull(skb
, skb_headlen(skb
));
4108 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4109 skb_reserve(skb
, NET_SKB_PAD
+ NET_IP_ALIGN
- skb_headroom(skb
));
4111 skb
->dev
= napi
->dev
;
4113 skb
->encapsulation
= 0;
4114 skb_shinfo(skb
)->gso_type
= 0;
4115 skb
->truesize
= SKB_TRUESIZE(skb_end_offset(skb
));
4120 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
4122 struct sk_buff
*skb
= napi
->skb
;
4125 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
4130 EXPORT_SYMBOL(napi_get_frags
);
4132 static gro_result_t
napi_frags_finish(struct napi_struct
*napi
,
4133 struct sk_buff
*skb
,
4139 __skb_push(skb
, ETH_HLEN
);
4140 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
4141 if (ret
== GRO_NORMAL
&& netif_receive_skb_internal(skb
))
4146 case GRO_MERGED_FREE
:
4147 napi_reuse_skb(napi
, skb
);
4157 /* Upper GRO stack assumes network header starts at gro_offset=0
4158 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4159 * We copy ethernet header into skb->data to have a common layout.
4161 static struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
4163 struct sk_buff
*skb
= napi
->skb
;
4164 const struct ethhdr
*eth
;
4165 unsigned int hlen
= sizeof(*eth
);
4169 skb_reset_mac_header(skb
);
4170 skb_gro_reset_offset(skb
);
4172 eth
= skb_gro_header_fast(skb
, 0);
4173 if (unlikely(skb_gro_header_hard(skb
, hlen
))) {
4174 eth
= skb_gro_header_slow(skb
, hlen
, 0);
4175 if (unlikely(!eth
)) {
4176 napi_reuse_skb(napi
, skb
);
4180 gro_pull_from_frag0(skb
, hlen
);
4181 NAPI_GRO_CB(skb
)->frag0
+= hlen
;
4182 NAPI_GRO_CB(skb
)->frag0_len
-= hlen
;
4184 __skb_pull(skb
, hlen
);
4187 * This works because the only protocols we care about don't require
4189 * We'll fix it up properly in napi_frags_finish()
4191 skb
->protocol
= eth
->h_proto
;
4196 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
4198 struct sk_buff
*skb
= napi_frags_skb(napi
);
4203 trace_napi_gro_frags_entry(skb
);
4205 return napi_frags_finish(napi
, skb
, dev_gro_receive(napi
, skb
));
4207 EXPORT_SYMBOL(napi_gro_frags
);
4210 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4211 * Note: called with local irq disabled, but exits with local irq enabled.
4213 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
4216 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
4219 sd
->rps_ipi_list
= NULL
;
4223 /* Send pending IPI's to kick RPS processing on remote cpus. */
4225 struct softnet_data
*next
= remsd
->rps_ipi_next
;
4227 if (cpu_online(remsd
->cpu
))
4228 smp_call_function_single_async(remsd
->cpu
,
4237 static int process_backlog(struct napi_struct
*napi
, int quota
)
4240 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
4243 /* Check if we have pending ipi, its better to send them now,
4244 * not waiting net_rx_action() end.
4246 if (sd
->rps_ipi_list
) {
4247 local_irq_disable();
4248 net_rps_action_and_irq_enable(sd
);
4251 napi
->weight
= weight_p
;
4252 local_irq_disable();
4254 struct sk_buff
*skb
;
4256 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
4258 __netif_receive_skb(skb
);
4259 local_irq_disable();
4260 input_queue_head_incr(sd
);
4261 if (++work
>= quota
) {
4268 if (skb_queue_empty(&sd
->input_pkt_queue
)) {
4270 * Inline a custom version of __napi_complete().
4271 * only current cpu owns and manipulates this napi,
4272 * and NAPI_STATE_SCHED is the only possible flag set
4274 * We can use a plain write instead of clear_bit(),
4275 * and we dont need an smp_mb() memory barrier.
4277 list_del(&napi
->poll_list
);
4284 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
4285 &sd
->process_queue
);
4294 * __napi_schedule - schedule for receive
4295 * @n: entry to schedule
4297 * The entry's receive function will be scheduled to run
4299 void __napi_schedule(struct napi_struct
*n
)
4301 unsigned long flags
;
4303 local_irq_save(flags
);
4304 ____napi_schedule(this_cpu_ptr(&softnet_data
), n
);
4305 local_irq_restore(flags
);
4307 EXPORT_SYMBOL(__napi_schedule
);
4309 void __napi_complete(struct napi_struct
*n
)
4311 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
4312 BUG_ON(n
->gro_list
);
4314 list_del(&n
->poll_list
);
4315 smp_mb__before_atomic();
4316 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
4318 EXPORT_SYMBOL(__napi_complete
);
4320 void napi_complete(struct napi_struct
*n
)
4322 unsigned long flags
;
4325 * don't let napi dequeue from the cpu poll list
4326 * just in case its running on a different cpu
4328 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
4331 napi_gro_flush(n
, false);
4332 local_irq_save(flags
);
4334 local_irq_restore(flags
);
4336 EXPORT_SYMBOL(napi_complete
);
4338 /* must be called under rcu_read_lock(), as we dont take a reference */
4339 struct napi_struct
*napi_by_id(unsigned int napi_id
)
4341 unsigned int hash
= napi_id
% HASH_SIZE(napi_hash
);
4342 struct napi_struct
*napi
;
4344 hlist_for_each_entry_rcu(napi
, &napi_hash
[hash
], napi_hash_node
)
4345 if (napi
->napi_id
== napi_id
)
4350 EXPORT_SYMBOL_GPL(napi_by_id
);
4352 void napi_hash_add(struct napi_struct
*napi
)
4354 if (!test_and_set_bit(NAPI_STATE_HASHED
, &napi
->state
)) {
4356 spin_lock(&napi_hash_lock
);
4358 /* 0 is not a valid id, we also skip an id that is taken
4359 * we expect both events to be extremely rare
4362 while (!napi
->napi_id
) {
4363 napi
->napi_id
= ++napi_gen_id
;
4364 if (napi_by_id(napi
->napi_id
))
4368 hlist_add_head_rcu(&napi
->napi_hash_node
,
4369 &napi_hash
[napi
->napi_id
% HASH_SIZE(napi_hash
)]);
4371 spin_unlock(&napi_hash_lock
);
4374 EXPORT_SYMBOL_GPL(napi_hash_add
);
4376 /* Warning : caller is responsible to make sure rcu grace period
4377 * is respected before freeing memory containing @napi
4379 void napi_hash_del(struct napi_struct
*napi
)
4381 spin_lock(&napi_hash_lock
);
4383 if (test_and_clear_bit(NAPI_STATE_HASHED
, &napi
->state
))
4384 hlist_del_rcu(&napi
->napi_hash_node
);
4386 spin_unlock(&napi_hash_lock
);
4388 EXPORT_SYMBOL_GPL(napi_hash_del
);
4390 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
4391 int (*poll
)(struct napi_struct
*, int), int weight
)
4393 INIT_LIST_HEAD(&napi
->poll_list
);
4394 napi
->gro_count
= 0;
4395 napi
->gro_list
= NULL
;
4398 if (weight
> NAPI_POLL_WEIGHT
)
4399 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4401 napi
->weight
= weight
;
4402 list_add(&napi
->dev_list
, &dev
->napi_list
);
4404 #ifdef CONFIG_NETPOLL
4405 spin_lock_init(&napi
->poll_lock
);
4406 napi
->poll_owner
= -1;
4408 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
4410 EXPORT_SYMBOL(netif_napi_add
);
4412 void netif_napi_del(struct napi_struct
*napi
)
4414 list_del_init(&napi
->dev_list
);
4415 napi_free_frags(napi
);
4417 kfree_skb_list(napi
->gro_list
);
4418 napi
->gro_list
= NULL
;
4419 napi
->gro_count
= 0;
4421 EXPORT_SYMBOL(netif_napi_del
);
4423 static void net_rx_action(struct softirq_action
*h
)
4425 struct softnet_data
*sd
= this_cpu_ptr(&softnet_data
);
4426 unsigned long time_limit
= jiffies
+ 2;
4427 int budget
= netdev_budget
;
4430 local_irq_disable();
4432 while (!list_empty(&sd
->poll_list
)) {
4433 struct napi_struct
*n
;
4436 /* If softirq window is exhuasted then punt.
4437 * Allow this to run for 2 jiffies since which will allow
4438 * an average latency of 1.5/HZ.
4440 if (unlikely(budget
<= 0 || time_after_eq(jiffies
, time_limit
)))
4445 /* Even though interrupts have been re-enabled, this
4446 * access is safe because interrupts can only add new
4447 * entries to the tail of this list, and only ->poll()
4448 * calls can remove this head entry from the list.
4450 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
4452 have
= netpoll_poll_lock(n
);
4456 /* This NAPI_STATE_SCHED test is for avoiding a race
4457 * with netpoll's poll_napi(). Only the entity which
4458 * obtains the lock and sees NAPI_STATE_SCHED set will
4459 * actually make the ->poll() call. Therefore we avoid
4460 * accidentally calling ->poll() when NAPI is not scheduled.
4463 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
4464 work
= n
->poll(n
, weight
);
4468 WARN_ON_ONCE(work
> weight
);
4472 local_irq_disable();
4474 /* Drivers must not modify the NAPI state if they
4475 * consume the entire weight. In such cases this code
4476 * still "owns" the NAPI instance and therefore can
4477 * move the instance around on the list at-will.
4479 if (unlikely(work
== weight
)) {
4480 if (unlikely(napi_disable_pending(n
))) {
4483 local_irq_disable();
4486 /* flush too old packets
4487 * If HZ < 1000, flush all packets.
4490 napi_gro_flush(n
, HZ
>= 1000);
4491 local_irq_disable();
4493 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
4497 netpoll_poll_unlock(have
);
4500 net_rps_action_and_irq_enable(sd
);
4502 #ifdef CONFIG_NET_DMA
4504 * There may not be any more sk_buffs coming right now, so push
4505 * any pending DMA copies to hardware
4507 dma_issue_pending_all();
4514 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
4518 struct netdev_adjacent
{
4519 struct net_device
*dev
;
4521 /* upper master flag, there can only be one master device per list */
4524 /* counter for the number of times this device was added to us */
4527 /* private field for the users */
4530 struct list_head list
;
4531 struct rcu_head rcu
;
4534 static struct netdev_adjacent
*__netdev_find_adj(struct net_device
*dev
,
4535 struct net_device
*adj_dev
,
4536 struct list_head
*adj_list
)
4538 struct netdev_adjacent
*adj
;
4540 list_for_each_entry(adj
, adj_list
, list
) {
4541 if (adj
->dev
== adj_dev
)
4548 * netdev_has_upper_dev - Check if device is linked to an upper device
4550 * @upper_dev: upper device to check
4552 * Find out if a device is linked to specified upper device and return true
4553 * in case it is. Note that this checks only immediate upper device,
4554 * not through a complete stack of devices. The caller must hold the RTNL lock.
4556 bool netdev_has_upper_dev(struct net_device
*dev
,
4557 struct net_device
*upper_dev
)
4561 return __netdev_find_adj(dev
, upper_dev
, &dev
->all_adj_list
.upper
);
4563 EXPORT_SYMBOL(netdev_has_upper_dev
);
4566 * netdev_has_any_upper_dev - Check if device is linked to some device
4569 * Find out if a device is linked to an upper device and return true in case
4570 * it is. The caller must hold the RTNL lock.
4572 static bool netdev_has_any_upper_dev(struct net_device
*dev
)
4576 return !list_empty(&dev
->all_adj_list
.upper
);
4580 * netdev_master_upper_dev_get - Get master upper device
4583 * Find a master upper device and return pointer to it or NULL in case
4584 * it's not there. The caller must hold the RTNL lock.
4586 struct net_device
*netdev_master_upper_dev_get(struct net_device
*dev
)
4588 struct netdev_adjacent
*upper
;
4592 if (list_empty(&dev
->adj_list
.upper
))
4595 upper
= list_first_entry(&dev
->adj_list
.upper
,
4596 struct netdev_adjacent
, list
);
4597 if (likely(upper
->master
))
4601 EXPORT_SYMBOL(netdev_master_upper_dev_get
);
4603 void *netdev_adjacent_get_private(struct list_head
*adj_list
)
4605 struct netdev_adjacent
*adj
;
4607 adj
= list_entry(adj_list
, struct netdev_adjacent
, list
);
4609 return adj
->private;
4611 EXPORT_SYMBOL(netdev_adjacent_get_private
);
4614 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
4616 * @iter: list_head ** of the current position
4618 * Gets the next device from the dev's upper list, starting from iter
4619 * position. The caller must hold RCU read lock.
4621 struct net_device
*netdev_upper_get_next_dev_rcu(struct net_device
*dev
,
4622 struct list_head
**iter
)
4624 struct netdev_adjacent
*upper
;
4626 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4628 upper
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
4630 if (&upper
->list
== &dev
->adj_list
.upper
)
4633 *iter
= &upper
->list
;
4637 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu
);
4640 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4642 * @iter: list_head ** of the current position
4644 * Gets the next device from the dev's upper list, starting from iter
4645 * position. The caller must hold RCU read lock.
4647 struct net_device
*netdev_all_upper_get_next_dev_rcu(struct net_device
*dev
,
4648 struct list_head
**iter
)
4650 struct netdev_adjacent
*upper
;
4652 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4654 upper
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
4656 if (&upper
->list
== &dev
->all_adj_list
.upper
)
4659 *iter
= &upper
->list
;
4663 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu
);
4666 * netdev_lower_get_next_private - Get the next ->private from the
4667 * lower neighbour list
4669 * @iter: list_head ** of the current position
4671 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4672 * list, starting from iter position. The caller must hold either hold the
4673 * RTNL lock or its own locking that guarantees that the neighbour lower
4674 * list will remain unchainged.
4676 void *netdev_lower_get_next_private(struct net_device
*dev
,
4677 struct list_head
**iter
)
4679 struct netdev_adjacent
*lower
;
4681 lower
= list_entry(*iter
, struct netdev_adjacent
, list
);
4683 if (&lower
->list
== &dev
->adj_list
.lower
)
4686 *iter
= lower
->list
.next
;
4688 return lower
->private;
4690 EXPORT_SYMBOL(netdev_lower_get_next_private
);
4693 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4694 * lower neighbour list, RCU
4697 * @iter: list_head ** of the current position
4699 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4700 * list, starting from iter position. The caller must hold RCU read lock.
4702 void *netdev_lower_get_next_private_rcu(struct net_device
*dev
,
4703 struct list_head
**iter
)
4705 struct netdev_adjacent
*lower
;
4707 WARN_ON_ONCE(!rcu_read_lock_held());
4709 lower
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
4711 if (&lower
->list
== &dev
->adj_list
.lower
)
4714 *iter
= &lower
->list
;
4716 return lower
->private;
4718 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu
);
4721 * netdev_lower_get_next - Get the next device from the lower neighbour
4724 * @iter: list_head ** of the current position
4726 * Gets the next netdev_adjacent from the dev's lower neighbour
4727 * list, starting from iter position. The caller must hold RTNL lock or
4728 * its own locking that guarantees that the neighbour lower
4729 * list will remain unchainged.
4731 void *netdev_lower_get_next(struct net_device
*dev
, struct list_head
**iter
)
4733 struct netdev_adjacent
*lower
;
4735 lower
= list_entry((*iter
)->next
, struct netdev_adjacent
, list
);
4737 if (&lower
->list
== &dev
->adj_list
.lower
)
4740 *iter
= &lower
->list
;
4744 EXPORT_SYMBOL(netdev_lower_get_next
);
4747 * netdev_lower_get_first_private_rcu - Get the first ->private from the
4748 * lower neighbour list, RCU
4752 * Gets the first netdev_adjacent->private from the dev's lower neighbour
4753 * list. The caller must hold RCU read lock.
4755 void *netdev_lower_get_first_private_rcu(struct net_device
*dev
)
4757 struct netdev_adjacent
*lower
;
4759 lower
= list_first_or_null_rcu(&dev
->adj_list
.lower
,
4760 struct netdev_adjacent
, list
);
4762 return lower
->private;
4765 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu
);
4768 * netdev_master_upper_dev_get_rcu - Get master upper device
4771 * Find a master upper device and return pointer to it or NULL in case
4772 * it's not there. The caller must hold the RCU read lock.
4774 struct net_device
*netdev_master_upper_dev_get_rcu(struct net_device
*dev
)
4776 struct netdev_adjacent
*upper
;
4778 upper
= list_first_or_null_rcu(&dev
->adj_list
.upper
,
4779 struct netdev_adjacent
, list
);
4780 if (upper
&& likely(upper
->master
))
4784 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu
);
4786 static int netdev_adjacent_sysfs_add(struct net_device
*dev
,
4787 struct net_device
*adj_dev
,
4788 struct list_head
*dev_list
)
4790 char linkname
[IFNAMSIZ
+7];
4791 sprintf(linkname
, dev_list
== &dev
->adj_list
.upper
?
4792 "upper_%s" : "lower_%s", adj_dev
->name
);
4793 return sysfs_create_link(&(dev
->dev
.kobj
), &(adj_dev
->dev
.kobj
),
4796 static void netdev_adjacent_sysfs_del(struct net_device
*dev
,
4798 struct list_head
*dev_list
)
4800 char linkname
[IFNAMSIZ
+7];
4801 sprintf(linkname
, dev_list
== &dev
->adj_list
.upper
?
4802 "upper_%s" : "lower_%s", name
);
4803 sysfs_remove_link(&(dev
->dev
.kobj
), linkname
);
4806 #define netdev_adjacent_is_neigh_list(dev, dev_list) \
4807 (dev_list == &dev->adj_list.upper || \
4808 dev_list == &dev->adj_list.lower)
4810 static int __netdev_adjacent_dev_insert(struct net_device
*dev
,
4811 struct net_device
*adj_dev
,
4812 struct list_head
*dev_list
,
4813 void *private, bool master
)
4815 struct netdev_adjacent
*adj
;
4818 adj
= __netdev_find_adj(dev
, adj_dev
, dev_list
);
4825 adj
= kmalloc(sizeof(*adj
), GFP_KERNEL
);
4830 adj
->master
= master
;
4832 adj
->private = private;
4835 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
4836 adj_dev
->name
, dev
->name
, adj_dev
->name
);
4838 if (netdev_adjacent_is_neigh_list(dev
, dev_list
)) {
4839 ret
= netdev_adjacent_sysfs_add(dev
, adj_dev
, dev_list
);
4844 /* Ensure that master link is always the first item in list. */
4846 ret
= sysfs_create_link(&(dev
->dev
.kobj
),
4847 &(adj_dev
->dev
.kobj
), "master");
4849 goto remove_symlinks
;
4851 list_add_rcu(&adj
->list
, dev_list
);
4853 list_add_tail_rcu(&adj
->list
, dev_list
);
4859 if (netdev_adjacent_is_neigh_list(dev
, dev_list
))
4860 netdev_adjacent_sysfs_del(dev
, adj_dev
->name
, dev_list
);
4868 static void __netdev_adjacent_dev_remove(struct net_device
*dev
,
4869 struct net_device
*adj_dev
,
4870 struct list_head
*dev_list
)
4872 struct netdev_adjacent
*adj
;
4874 adj
= __netdev_find_adj(dev
, adj_dev
, dev_list
);
4877 pr_err("tried to remove device %s from %s\n",
4878 dev
->name
, adj_dev
->name
);
4882 if (adj
->ref_nr
> 1) {
4883 pr_debug("%s to %s ref_nr-- = %d\n", dev
->name
, adj_dev
->name
,
4890 sysfs_remove_link(&(dev
->dev
.kobj
), "master");
4892 if (netdev_adjacent_is_neigh_list(dev
, dev_list
))
4893 netdev_adjacent_sysfs_del(dev
, adj_dev
->name
, dev_list
);
4895 list_del_rcu(&adj
->list
);
4896 pr_debug("dev_put for %s, because link removed from %s to %s\n",
4897 adj_dev
->name
, dev
->name
, adj_dev
->name
);
4899 kfree_rcu(adj
, rcu
);
4902 static int __netdev_adjacent_dev_link_lists(struct net_device
*dev
,
4903 struct net_device
*upper_dev
,
4904 struct list_head
*up_list
,
4905 struct list_head
*down_list
,
4906 void *private, bool master
)
4910 ret
= __netdev_adjacent_dev_insert(dev
, upper_dev
, up_list
, private,
4915 ret
= __netdev_adjacent_dev_insert(upper_dev
, dev
, down_list
, private,
4918 __netdev_adjacent_dev_remove(dev
, upper_dev
, up_list
);
4925 static int __netdev_adjacent_dev_link(struct net_device
*dev
,
4926 struct net_device
*upper_dev
)
4928 return __netdev_adjacent_dev_link_lists(dev
, upper_dev
,
4929 &dev
->all_adj_list
.upper
,
4930 &upper_dev
->all_adj_list
.lower
,
4934 static void __netdev_adjacent_dev_unlink_lists(struct net_device
*dev
,
4935 struct net_device
*upper_dev
,
4936 struct list_head
*up_list
,
4937 struct list_head
*down_list
)
4939 __netdev_adjacent_dev_remove(dev
, upper_dev
, up_list
);
4940 __netdev_adjacent_dev_remove(upper_dev
, dev
, down_list
);
4943 static void __netdev_adjacent_dev_unlink(struct net_device
*dev
,
4944 struct net_device
*upper_dev
)
4946 __netdev_adjacent_dev_unlink_lists(dev
, upper_dev
,
4947 &dev
->all_adj_list
.upper
,
4948 &upper_dev
->all_adj_list
.lower
);
4951 static int __netdev_adjacent_dev_link_neighbour(struct net_device
*dev
,
4952 struct net_device
*upper_dev
,
4953 void *private, bool master
)
4955 int ret
= __netdev_adjacent_dev_link(dev
, upper_dev
);
4960 ret
= __netdev_adjacent_dev_link_lists(dev
, upper_dev
,
4961 &dev
->adj_list
.upper
,
4962 &upper_dev
->adj_list
.lower
,
4965 __netdev_adjacent_dev_unlink(dev
, upper_dev
);
4972 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device
*dev
,
4973 struct net_device
*upper_dev
)
4975 __netdev_adjacent_dev_unlink(dev
, upper_dev
);
4976 __netdev_adjacent_dev_unlink_lists(dev
, upper_dev
,
4977 &dev
->adj_list
.upper
,
4978 &upper_dev
->adj_list
.lower
);
4981 static int __netdev_upper_dev_link(struct net_device
*dev
,
4982 struct net_device
*upper_dev
, bool master
,
4985 struct netdev_adjacent
*i
, *j
, *to_i
, *to_j
;
4990 if (dev
== upper_dev
)
4993 /* To prevent loops, check if dev is not upper device to upper_dev. */
4994 if (__netdev_find_adj(upper_dev
, dev
, &upper_dev
->all_adj_list
.upper
))
4997 if (__netdev_find_adj(dev
, upper_dev
, &dev
->all_adj_list
.upper
))
5000 if (master
&& netdev_master_upper_dev_get(dev
))
5003 ret
= __netdev_adjacent_dev_link_neighbour(dev
, upper_dev
, private,
5008 /* Now that we linked these devs, make all the upper_dev's
5009 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5010 * versa, and don't forget the devices itself. All of these
5011 * links are non-neighbours.
5013 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5014 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
) {
5015 pr_debug("Interlinking %s with %s, non-neighbour\n",
5016 i
->dev
->name
, j
->dev
->name
);
5017 ret
= __netdev_adjacent_dev_link(i
->dev
, j
->dev
);
5023 /* add dev to every upper_dev's upper device */
5024 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
) {
5025 pr_debug("linking %s's upper device %s with %s\n",
5026 upper_dev
->name
, i
->dev
->name
, dev
->name
);
5027 ret
= __netdev_adjacent_dev_link(dev
, i
->dev
);
5029 goto rollback_upper_mesh
;
5032 /* add upper_dev to every dev's lower device */
5033 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5034 pr_debug("linking %s's lower device %s with %s\n", dev
->name
,
5035 i
->dev
->name
, upper_dev
->name
);
5036 ret
= __netdev_adjacent_dev_link(i
->dev
, upper_dev
);
5038 goto rollback_lower_mesh
;
5041 call_netdevice_notifiers(NETDEV_CHANGEUPPER
, dev
);
5044 rollback_lower_mesh
:
5046 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5049 __netdev_adjacent_dev_unlink(i
->dev
, upper_dev
);
5054 rollback_upper_mesh
:
5056 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
) {
5059 __netdev_adjacent_dev_unlink(dev
, i
->dev
);
5067 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5068 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
) {
5069 if (i
== to_i
&& j
== to_j
)
5071 __netdev_adjacent_dev_unlink(i
->dev
, j
->dev
);
5077 __netdev_adjacent_dev_unlink_neighbour(dev
, upper_dev
);
5083 * netdev_upper_dev_link - Add a link to the upper device
5085 * @upper_dev: new upper device
5087 * Adds a link to device which is upper to this one. The caller must hold
5088 * the RTNL lock. On a failure a negative errno code is returned.
5089 * On success the reference counts are adjusted and the function
5092 int netdev_upper_dev_link(struct net_device
*dev
,
5093 struct net_device
*upper_dev
)
5095 return __netdev_upper_dev_link(dev
, upper_dev
, false, NULL
);
5097 EXPORT_SYMBOL(netdev_upper_dev_link
);
5100 * netdev_master_upper_dev_link - Add a master link to the upper device
5102 * @upper_dev: new upper device
5104 * Adds a link to device which is upper to this one. In this case, only
5105 * one master upper device can be linked, although other non-master devices
5106 * might be linked as well. The caller must hold the RTNL lock.
5107 * On a failure a negative errno code is returned. On success the reference
5108 * counts are adjusted and the function returns zero.
5110 int netdev_master_upper_dev_link(struct net_device
*dev
,
5111 struct net_device
*upper_dev
)
5113 return __netdev_upper_dev_link(dev
, upper_dev
, true, NULL
);
5115 EXPORT_SYMBOL(netdev_master_upper_dev_link
);
5117 int netdev_master_upper_dev_link_private(struct net_device
*dev
,
5118 struct net_device
*upper_dev
,
5121 return __netdev_upper_dev_link(dev
, upper_dev
, true, private);
5123 EXPORT_SYMBOL(netdev_master_upper_dev_link_private
);
5126 * netdev_upper_dev_unlink - Removes a link to upper device
5128 * @upper_dev: new upper device
5130 * Removes a link to device which is upper to this one. The caller must hold
5133 void netdev_upper_dev_unlink(struct net_device
*dev
,
5134 struct net_device
*upper_dev
)
5136 struct netdev_adjacent
*i
, *j
;
5139 __netdev_adjacent_dev_unlink_neighbour(dev
, upper_dev
);
5141 /* Here is the tricky part. We must remove all dev's lower
5142 * devices from all upper_dev's upper devices and vice
5143 * versa, to maintain the graph relationship.
5145 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
)
5146 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
)
5147 __netdev_adjacent_dev_unlink(i
->dev
, j
->dev
);
5149 /* remove also the devices itself from lower/upper device
5152 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
)
5153 __netdev_adjacent_dev_unlink(i
->dev
, upper_dev
);
5155 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
)
5156 __netdev_adjacent_dev_unlink(dev
, i
->dev
);
5158 call_netdevice_notifiers(NETDEV_CHANGEUPPER
, dev
);
5160 EXPORT_SYMBOL(netdev_upper_dev_unlink
);
5162 void netdev_adjacent_rename_links(struct net_device
*dev
, char *oldname
)
5164 struct netdev_adjacent
*iter
;
5166 list_for_each_entry(iter
, &dev
->adj_list
.upper
, list
) {
5167 netdev_adjacent_sysfs_del(iter
->dev
, oldname
,
5168 &iter
->dev
->adj_list
.lower
);
5169 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
5170 &iter
->dev
->adj_list
.lower
);
5173 list_for_each_entry(iter
, &dev
->adj_list
.lower
, list
) {
5174 netdev_adjacent_sysfs_del(iter
->dev
, oldname
,
5175 &iter
->dev
->adj_list
.upper
);
5176 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
5177 &iter
->dev
->adj_list
.upper
);
5181 void *netdev_lower_dev_get_private(struct net_device
*dev
,
5182 struct net_device
*lower_dev
)
5184 struct netdev_adjacent
*lower
;
5188 lower
= __netdev_find_adj(dev
, lower_dev
, &dev
->adj_list
.lower
);
5192 return lower
->private;
5194 EXPORT_SYMBOL(netdev_lower_dev_get_private
);
5197 int dev_get_nest_level(struct net_device
*dev
,
5198 bool (*type_check
)(struct net_device
*dev
))
5200 struct net_device
*lower
= NULL
;
5201 struct list_head
*iter
;
5207 netdev_for_each_lower_dev(dev
, lower
, iter
) {
5208 nest
= dev_get_nest_level(lower
, type_check
);
5209 if (max_nest
< nest
)
5213 if (type_check(dev
))
5218 EXPORT_SYMBOL(dev_get_nest_level
);
5220 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
5222 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5224 if (ops
->ndo_change_rx_flags
)
5225 ops
->ndo_change_rx_flags(dev
, flags
);
5228 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
, bool notify
)
5230 unsigned int old_flags
= dev
->flags
;
5236 dev
->flags
|= IFF_PROMISC
;
5237 dev
->promiscuity
+= inc
;
5238 if (dev
->promiscuity
== 0) {
5241 * If inc causes overflow, untouch promisc and return error.
5244 dev
->flags
&= ~IFF_PROMISC
;
5246 dev
->promiscuity
-= inc
;
5247 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5252 if (dev
->flags
!= old_flags
) {
5253 pr_info("device %s %s promiscuous mode\n",
5255 dev
->flags
& IFF_PROMISC
? "entered" : "left");
5256 if (audit_enabled
) {
5257 current_uid_gid(&uid
, &gid
);
5258 audit_log(current
->audit_context
, GFP_ATOMIC
,
5259 AUDIT_ANOM_PROMISCUOUS
,
5260 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5261 dev
->name
, (dev
->flags
& IFF_PROMISC
),
5262 (old_flags
& IFF_PROMISC
),
5263 from_kuid(&init_user_ns
, audit_get_loginuid(current
)),
5264 from_kuid(&init_user_ns
, uid
),
5265 from_kgid(&init_user_ns
, gid
),
5266 audit_get_sessionid(current
));
5269 dev_change_rx_flags(dev
, IFF_PROMISC
);
5272 __dev_notify_flags(dev
, old_flags
, IFF_PROMISC
);
5277 * dev_set_promiscuity - update promiscuity count on a device
5281 * Add or remove promiscuity from a device. While the count in the device
5282 * remains above zero the interface remains promiscuous. Once it hits zero
5283 * the device reverts back to normal filtering operation. A negative inc
5284 * value is used to drop promiscuity on the device.
5285 * Return 0 if successful or a negative errno code on error.
5287 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
5289 unsigned int old_flags
= dev
->flags
;
5292 err
= __dev_set_promiscuity(dev
, inc
, true);
5295 if (dev
->flags
!= old_flags
)
5296 dev_set_rx_mode(dev
);
5299 EXPORT_SYMBOL(dev_set_promiscuity
);
5301 static int __dev_set_allmulti(struct net_device
*dev
, int inc
, bool notify
)
5303 unsigned int old_flags
= dev
->flags
, old_gflags
= dev
->gflags
;
5307 dev
->flags
|= IFF_ALLMULTI
;
5308 dev
->allmulti
+= inc
;
5309 if (dev
->allmulti
== 0) {
5312 * If inc causes overflow, untouch allmulti and return error.
5315 dev
->flags
&= ~IFF_ALLMULTI
;
5317 dev
->allmulti
-= inc
;
5318 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5323 if (dev
->flags
^ old_flags
) {
5324 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
5325 dev_set_rx_mode(dev
);
5327 __dev_notify_flags(dev
, old_flags
,
5328 dev
->gflags
^ old_gflags
);
5334 * dev_set_allmulti - update allmulti count on a device
5338 * Add or remove reception of all multicast frames to a device. While the
5339 * count in the device remains above zero the interface remains listening
5340 * to all interfaces. Once it hits zero the device reverts back to normal
5341 * filtering operation. A negative @inc value is used to drop the counter
5342 * when releasing a resource needing all multicasts.
5343 * Return 0 if successful or a negative errno code on error.
5346 int dev_set_allmulti(struct net_device
*dev
, int inc
)
5348 return __dev_set_allmulti(dev
, inc
, true);
5350 EXPORT_SYMBOL(dev_set_allmulti
);
5353 * Upload unicast and multicast address lists to device and
5354 * configure RX filtering. When the device doesn't support unicast
5355 * filtering it is put in promiscuous mode while unicast addresses
5358 void __dev_set_rx_mode(struct net_device
*dev
)
5360 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5362 /* dev_open will call this function so the list will stay sane. */
5363 if (!(dev
->flags
&IFF_UP
))
5366 if (!netif_device_present(dev
))
5369 if (!(dev
->priv_flags
& IFF_UNICAST_FLT
)) {
5370 /* Unicast addresses changes may only happen under the rtnl,
5371 * therefore calling __dev_set_promiscuity here is safe.
5373 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
5374 __dev_set_promiscuity(dev
, 1, false);
5375 dev
->uc_promisc
= true;
5376 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
5377 __dev_set_promiscuity(dev
, -1, false);
5378 dev
->uc_promisc
= false;
5382 if (ops
->ndo_set_rx_mode
)
5383 ops
->ndo_set_rx_mode(dev
);
5386 void dev_set_rx_mode(struct net_device
*dev
)
5388 netif_addr_lock_bh(dev
);
5389 __dev_set_rx_mode(dev
);
5390 netif_addr_unlock_bh(dev
);
5394 * dev_get_flags - get flags reported to userspace
5397 * Get the combination of flag bits exported through APIs to userspace.
5399 unsigned int dev_get_flags(const struct net_device
*dev
)
5403 flags
= (dev
->flags
& ~(IFF_PROMISC
|
5408 (dev
->gflags
& (IFF_PROMISC
|
5411 if (netif_running(dev
)) {
5412 if (netif_oper_up(dev
))
5413 flags
|= IFF_RUNNING
;
5414 if (netif_carrier_ok(dev
))
5415 flags
|= IFF_LOWER_UP
;
5416 if (netif_dormant(dev
))
5417 flags
|= IFF_DORMANT
;
5422 EXPORT_SYMBOL(dev_get_flags
);
5424 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
5426 unsigned int old_flags
= dev
->flags
;
5432 * Set the flags on our device.
5435 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
5436 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
5438 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
5442 * Load in the correct multicast list now the flags have changed.
5445 if ((old_flags
^ flags
) & IFF_MULTICAST
)
5446 dev_change_rx_flags(dev
, IFF_MULTICAST
);
5448 dev_set_rx_mode(dev
);
5451 * Have we downed the interface. We handle IFF_UP ourselves
5452 * according to user attempts to set it, rather than blindly
5457 if ((old_flags
^ flags
) & IFF_UP
)
5458 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
5460 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
5461 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
5462 unsigned int old_flags
= dev
->flags
;
5464 dev
->gflags
^= IFF_PROMISC
;
5466 if (__dev_set_promiscuity(dev
, inc
, false) >= 0)
5467 if (dev
->flags
!= old_flags
)
5468 dev_set_rx_mode(dev
);
5471 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5472 is important. Some (broken) drivers set IFF_PROMISC, when
5473 IFF_ALLMULTI is requested not asking us and not reporting.
5475 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
5476 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
5478 dev
->gflags
^= IFF_ALLMULTI
;
5479 __dev_set_allmulti(dev
, inc
, false);
5485 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
,
5486 unsigned int gchanges
)
5488 unsigned int changes
= dev
->flags
^ old_flags
;
5491 rtmsg_ifinfo(RTM_NEWLINK
, dev
, gchanges
, GFP_ATOMIC
);
5493 if (changes
& IFF_UP
) {
5494 if (dev
->flags
& IFF_UP
)
5495 call_netdevice_notifiers(NETDEV_UP
, dev
);
5497 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
5500 if (dev
->flags
& IFF_UP
&&
5501 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
))) {
5502 struct netdev_notifier_change_info change_info
;
5504 change_info
.flags_changed
= changes
;
5505 call_netdevice_notifiers_info(NETDEV_CHANGE
, dev
,
5511 * dev_change_flags - change device settings
5513 * @flags: device state flags
5515 * Change settings on device based state flags. The flags are
5516 * in the userspace exported format.
5518 int dev_change_flags(struct net_device
*dev
, unsigned int flags
)
5521 unsigned int changes
, old_flags
= dev
->flags
, old_gflags
= dev
->gflags
;
5523 ret
= __dev_change_flags(dev
, flags
);
5527 changes
= (old_flags
^ dev
->flags
) | (old_gflags
^ dev
->gflags
);
5528 __dev_notify_flags(dev
, old_flags
, changes
);
5531 EXPORT_SYMBOL(dev_change_flags
);
5533 static int __dev_set_mtu(struct net_device
*dev
, int new_mtu
)
5535 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5537 if (ops
->ndo_change_mtu
)
5538 return ops
->ndo_change_mtu(dev
, new_mtu
);
5545 * dev_set_mtu - Change maximum transfer unit
5547 * @new_mtu: new transfer unit
5549 * Change the maximum transfer size of the network device.
5551 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
5555 if (new_mtu
== dev
->mtu
)
5558 /* MTU must be positive. */
5562 if (!netif_device_present(dev
))
5565 err
= call_netdevice_notifiers(NETDEV_PRECHANGEMTU
, dev
);
5566 err
= notifier_to_errno(err
);
5570 orig_mtu
= dev
->mtu
;
5571 err
= __dev_set_mtu(dev
, new_mtu
);
5574 err
= call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
5575 err
= notifier_to_errno(err
);
5577 /* setting mtu back and notifying everyone again,
5578 * so that they have a chance to revert changes.
5580 __dev_set_mtu(dev
, orig_mtu
);
5581 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
5586 EXPORT_SYMBOL(dev_set_mtu
);
5589 * dev_set_group - Change group this device belongs to
5591 * @new_group: group this device should belong to
5593 void dev_set_group(struct net_device
*dev
, int new_group
)
5595 dev
->group
= new_group
;
5597 EXPORT_SYMBOL(dev_set_group
);
5600 * dev_set_mac_address - Change Media Access Control Address
5604 * Change the hardware (MAC) address of the device
5606 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
5608 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5611 if (!ops
->ndo_set_mac_address
)
5613 if (sa
->sa_family
!= dev
->type
)
5615 if (!netif_device_present(dev
))
5617 err
= ops
->ndo_set_mac_address(dev
, sa
);
5620 dev
->addr_assign_type
= NET_ADDR_SET
;
5621 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
5622 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
5625 EXPORT_SYMBOL(dev_set_mac_address
);
5628 * dev_change_carrier - Change device carrier
5630 * @new_carrier: new value
5632 * Change device carrier
5634 int dev_change_carrier(struct net_device
*dev
, bool new_carrier
)
5636 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5638 if (!ops
->ndo_change_carrier
)
5640 if (!netif_device_present(dev
))
5642 return ops
->ndo_change_carrier(dev
, new_carrier
);
5644 EXPORT_SYMBOL(dev_change_carrier
);
5647 * dev_get_phys_port_id - Get device physical port ID
5651 * Get device physical port ID
5653 int dev_get_phys_port_id(struct net_device
*dev
,
5654 struct netdev_phys_port_id
*ppid
)
5656 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5658 if (!ops
->ndo_get_phys_port_id
)
5660 return ops
->ndo_get_phys_port_id(dev
, ppid
);
5662 EXPORT_SYMBOL(dev_get_phys_port_id
);
5665 * dev_new_index - allocate an ifindex
5666 * @net: the applicable net namespace
5668 * Returns a suitable unique value for a new device interface
5669 * number. The caller must hold the rtnl semaphore or the
5670 * dev_base_lock to be sure it remains unique.
5672 static int dev_new_index(struct net
*net
)
5674 int ifindex
= net
->ifindex
;
5678 if (!__dev_get_by_index(net
, ifindex
))
5679 return net
->ifindex
= ifindex
;
5683 /* Delayed registration/unregisteration */
5684 static LIST_HEAD(net_todo_list
);
5685 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq
);
5687 static void net_set_todo(struct net_device
*dev
)
5689 list_add_tail(&dev
->todo_list
, &net_todo_list
);
5690 dev_net(dev
)->dev_unreg_count
++;
5693 static void rollback_registered_many(struct list_head
*head
)
5695 struct net_device
*dev
, *tmp
;
5696 LIST_HEAD(close_head
);
5698 BUG_ON(dev_boot_phase
);
5701 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
5702 /* Some devices call without registering
5703 * for initialization unwind. Remove those
5704 * devices and proceed with the remaining.
5706 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5707 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5711 list_del(&dev
->unreg_list
);
5714 dev
->dismantle
= true;
5715 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
5718 /* If device is running, close it first. */
5719 list_for_each_entry(dev
, head
, unreg_list
)
5720 list_add_tail(&dev
->close_list
, &close_head
);
5721 dev_close_many(&close_head
);
5723 list_for_each_entry(dev
, head
, unreg_list
) {
5724 /* And unlink it from device chain. */
5725 unlist_netdevice(dev
);
5727 dev
->reg_state
= NETREG_UNREGISTERING
;
5732 list_for_each_entry(dev
, head
, unreg_list
) {
5733 /* Shutdown queueing discipline. */
5737 /* Notify protocols, that we are about to destroy
5738 this device. They should clean all the things.
5740 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5743 * Flush the unicast and multicast chains
5748 if (dev
->netdev_ops
->ndo_uninit
)
5749 dev
->netdev_ops
->ndo_uninit(dev
);
5751 if (!dev
->rtnl_link_ops
||
5752 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5753 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U, GFP_KERNEL
);
5755 /* Notifier chain MUST detach us all upper devices. */
5756 WARN_ON(netdev_has_any_upper_dev(dev
));
5758 /* Remove entries from kobject tree */
5759 netdev_unregister_kobject(dev
);
5761 /* Remove XPS queueing entries */
5762 netif_reset_xps_queues_gt(dev
, 0);
5768 list_for_each_entry(dev
, head
, unreg_list
)
5772 static void rollback_registered(struct net_device
*dev
)
5776 list_add(&dev
->unreg_list
, &single
);
5777 rollback_registered_many(&single
);
5781 static netdev_features_t
netdev_fix_features(struct net_device
*dev
,
5782 netdev_features_t features
)
5784 /* Fix illegal checksum combinations */
5785 if ((features
& NETIF_F_HW_CSUM
) &&
5786 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5787 netdev_warn(dev
, "mixed HW and IP checksum settings.\n");
5788 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5791 /* TSO requires that SG is present as well. */
5792 if ((features
& NETIF_F_ALL_TSO
) && !(features
& NETIF_F_SG
)) {
5793 netdev_dbg(dev
, "Dropping TSO features since no SG feature.\n");
5794 features
&= ~NETIF_F_ALL_TSO
;
5797 if ((features
& NETIF_F_TSO
) && !(features
& NETIF_F_HW_CSUM
) &&
5798 !(features
& NETIF_F_IP_CSUM
)) {
5799 netdev_dbg(dev
, "Dropping TSO features since no CSUM feature.\n");
5800 features
&= ~NETIF_F_TSO
;
5801 features
&= ~NETIF_F_TSO_ECN
;
5804 if ((features
& NETIF_F_TSO6
) && !(features
& NETIF_F_HW_CSUM
) &&
5805 !(features
& NETIF_F_IPV6_CSUM
)) {
5806 netdev_dbg(dev
, "Dropping TSO6 features since no CSUM feature.\n");
5807 features
&= ~NETIF_F_TSO6
;
5810 /* TSO ECN requires that TSO is present as well. */
5811 if ((features
& NETIF_F_ALL_TSO
) == NETIF_F_TSO_ECN
)
5812 features
&= ~NETIF_F_TSO_ECN
;
5814 /* Software GSO depends on SG. */
5815 if ((features
& NETIF_F_GSO
) && !(features
& NETIF_F_SG
)) {
5816 netdev_dbg(dev
, "Dropping NETIF_F_GSO since no SG feature.\n");
5817 features
&= ~NETIF_F_GSO
;
5820 /* UFO needs SG and checksumming */
5821 if (features
& NETIF_F_UFO
) {
5822 /* maybe split UFO into V4 and V6? */
5823 if (!((features
& NETIF_F_GEN_CSUM
) ||
5824 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))
5825 == (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5827 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5828 features
&= ~NETIF_F_UFO
;
5831 if (!(features
& NETIF_F_SG
)) {
5833 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5834 features
&= ~NETIF_F_UFO
;
5838 #ifdef CONFIG_NET_RX_BUSY_POLL
5839 if (dev
->netdev_ops
->ndo_busy_poll
)
5840 features
|= NETIF_F_BUSY_POLL
;
5843 features
&= ~NETIF_F_BUSY_POLL
;
5848 int __netdev_update_features(struct net_device
*dev
)
5850 netdev_features_t features
;
5855 features
= netdev_get_wanted_features(dev
);
5857 if (dev
->netdev_ops
->ndo_fix_features
)
5858 features
= dev
->netdev_ops
->ndo_fix_features(dev
, features
);
5860 /* driver might be less strict about feature dependencies */
5861 features
= netdev_fix_features(dev
, features
);
5863 if (dev
->features
== features
)
5866 netdev_dbg(dev
, "Features changed: %pNF -> %pNF\n",
5867 &dev
->features
, &features
);
5869 if (dev
->netdev_ops
->ndo_set_features
)
5870 err
= dev
->netdev_ops
->ndo_set_features(dev
, features
);
5872 if (unlikely(err
< 0)) {
5874 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5875 err
, &features
, &dev
->features
);
5880 dev
->features
= features
;
5886 * netdev_update_features - recalculate device features
5887 * @dev: the device to check
5889 * Recalculate dev->features set and send notifications if it
5890 * has changed. Should be called after driver or hardware dependent
5891 * conditions might have changed that influence the features.
5893 void netdev_update_features(struct net_device
*dev
)
5895 if (__netdev_update_features(dev
))
5896 netdev_features_change(dev
);
5898 EXPORT_SYMBOL(netdev_update_features
);
5901 * netdev_change_features - recalculate device features
5902 * @dev: the device to check
5904 * Recalculate dev->features set and send notifications even
5905 * if they have not changed. Should be called instead of
5906 * netdev_update_features() if also dev->vlan_features might
5907 * have changed to allow the changes to be propagated to stacked
5910 void netdev_change_features(struct net_device
*dev
)
5912 __netdev_update_features(dev
);
5913 netdev_features_change(dev
);
5915 EXPORT_SYMBOL(netdev_change_features
);
5918 * netif_stacked_transfer_operstate - transfer operstate
5919 * @rootdev: the root or lower level device to transfer state from
5920 * @dev: the device to transfer operstate to
5922 * Transfer operational state from root to device. This is normally
5923 * called when a stacking relationship exists between the root
5924 * device and the device(a leaf device).
5926 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
5927 struct net_device
*dev
)
5929 if (rootdev
->operstate
== IF_OPER_DORMANT
)
5930 netif_dormant_on(dev
);
5932 netif_dormant_off(dev
);
5934 if (netif_carrier_ok(rootdev
)) {
5935 if (!netif_carrier_ok(dev
))
5936 netif_carrier_on(dev
);
5938 if (netif_carrier_ok(dev
))
5939 netif_carrier_off(dev
);
5942 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
5945 static int netif_alloc_rx_queues(struct net_device
*dev
)
5947 unsigned int i
, count
= dev
->num_rx_queues
;
5948 struct netdev_rx_queue
*rx
;
5952 rx
= kcalloc(count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5958 for (i
= 0; i
< count
; i
++)
5964 static void netdev_init_one_queue(struct net_device
*dev
,
5965 struct netdev_queue
*queue
, void *_unused
)
5967 /* Initialize queue lock */
5968 spin_lock_init(&queue
->_xmit_lock
);
5969 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
5970 queue
->xmit_lock_owner
= -1;
5971 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
5974 dql_init(&queue
->dql
, HZ
);
5978 static void netif_free_tx_queues(struct net_device
*dev
)
5983 static int netif_alloc_netdev_queues(struct net_device
*dev
)
5985 unsigned int count
= dev
->num_tx_queues
;
5986 struct netdev_queue
*tx
;
5987 size_t sz
= count
* sizeof(*tx
);
5989 BUG_ON(count
< 1 || count
> 0xffff);
5991 tx
= kzalloc(sz
, GFP_KERNEL
| __GFP_NOWARN
| __GFP_REPEAT
);
5999 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
6000 spin_lock_init(&dev
->tx_global_lock
);
6006 * register_netdevice - register a network device
6007 * @dev: device to register
6009 * Take a completed network device structure and add it to the kernel
6010 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6011 * chain. 0 is returned on success. A negative errno code is returned
6012 * on a failure to set up the device, or if the name is a duplicate.
6014 * Callers must hold the rtnl semaphore. You may want
6015 * register_netdev() instead of this.
6018 * The locking appears insufficient to guarantee two parallel registers
6019 * will not get the same name.
6022 int register_netdevice(struct net_device
*dev
)
6025 struct net
*net
= dev_net(dev
);
6027 BUG_ON(dev_boot_phase
);
6032 /* When net_device's are persistent, this will be fatal. */
6033 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
6036 spin_lock_init(&dev
->addr_list_lock
);
6037 netdev_set_addr_lockdep_class(dev
);
6041 ret
= dev_get_valid_name(net
, dev
, dev
->name
);
6045 /* Init, if this function is available */
6046 if (dev
->netdev_ops
->ndo_init
) {
6047 ret
= dev
->netdev_ops
->ndo_init(dev
);
6055 if (((dev
->hw_features
| dev
->features
) &
6056 NETIF_F_HW_VLAN_CTAG_FILTER
) &&
6057 (!dev
->netdev_ops
->ndo_vlan_rx_add_vid
||
6058 !dev
->netdev_ops
->ndo_vlan_rx_kill_vid
)) {
6059 netdev_WARN(dev
, "Buggy VLAN acceleration in driver!\n");
6066 dev
->ifindex
= dev_new_index(net
);
6067 else if (__dev_get_by_index(net
, dev
->ifindex
))
6070 if (dev
->iflink
== -1)
6071 dev
->iflink
= dev
->ifindex
;
6073 /* Transfer changeable features to wanted_features and enable
6074 * software offloads (GSO and GRO).
6076 dev
->hw_features
|= NETIF_F_SOFT_FEATURES
;
6077 dev
->features
|= NETIF_F_SOFT_FEATURES
;
6078 dev
->wanted_features
= dev
->features
& dev
->hw_features
;
6080 if (!(dev
->flags
& IFF_LOOPBACK
)) {
6081 dev
->hw_features
|= NETIF_F_NOCACHE_COPY
;
6084 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6086 dev
->vlan_features
|= NETIF_F_HIGHDMA
;
6088 /* Make NETIF_F_SG inheritable to tunnel devices.
6090 dev
->hw_enc_features
|= NETIF_F_SG
;
6092 /* Make NETIF_F_SG inheritable to MPLS.
6094 dev
->mpls_features
|= NETIF_F_SG
;
6096 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
6097 ret
= notifier_to_errno(ret
);
6101 ret
= netdev_register_kobject(dev
);
6104 dev
->reg_state
= NETREG_REGISTERED
;
6106 __netdev_update_features(dev
);
6109 * Default initial state at registry is that the
6110 * device is present.
6113 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
6115 linkwatch_init_dev(dev
);
6117 dev_init_scheduler(dev
);
6119 list_netdevice(dev
);
6120 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
6122 /* If the device has permanent device address, driver should
6123 * set dev_addr and also addr_assign_type should be set to
6124 * NET_ADDR_PERM (default value).
6126 if (dev
->addr_assign_type
== NET_ADDR_PERM
)
6127 memcpy(dev
->perm_addr
, dev
->dev_addr
, dev
->addr_len
);
6129 /* Notify protocols, that a new device appeared. */
6130 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
6131 ret
= notifier_to_errno(ret
);
6133 rollback_registered(dev
);
6134 dev
->reg_state
= NETREG_UNREGISTERED
;
6137 * Prevent userspace races by waiting until the network
6138 * device is fully setup before sending notifications.
6140 if (!dev
->rtnl_link_ops
||
6141 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
6142 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U, GFP_KERNEL
);
6148 if (dev
->netdev_ops
->ndo_uninit
)
6149 dev
->netdev_ops
->ndo_uninit(dev
);
6152 EXPORT_SYMBOL(register_netdevice
);
6155 * init_dummy_netdev - init a dummy network device for NAPI
6156 * @dev: device to init
6158 * This takes a network device structure and initialize the minimum
6159 * amount of fields so it can be used to schedule NAPI polls without
6160 * registering a full blown interface. This is to be used by drivers
6161 * that need to tie several hardware interfaces to a single NAPI
6162 * poll scheduler due to HW limitations.
6164 int init_dummy_netdev(struct net_device
*dev
)
6166 /* Clear everything. Note we don't initialize spinlocks
6167 * are they aren't supposed to be taken by any of the
6168 * NAPI code and this dummy netdev is supposed to be
6169 * only ever used for NAPI polls
6171 memset(dev
, 0, sizeof(struct net_device
));
6173 /* make sure we BUG if trying to hit standard
6174 * register/unregister code path
6176 dev
->reg_state
= NETREG_DUMMY
;
6178 /* NAPI wants this */
6179 INIT_LIST_HEAD(&dev
->napi_list
);
6181 /* a dummy interface is started by default */
6182 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
6183 set_bit(__LINK_STATE_START
, &dev
->state
);
6185 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6186 * because users of this 'device' dont need to change
6192 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
6196 * register_netdev - register a network device
6197 * @dev: device to register
6199 * Take a completed network device structure and add it to the kernel
6200 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6201 * chain. 0 is returned on success. A negative errno code is returned
6202 * on a failure to set up the device, or if the name is a duplicate.
6204 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6205 * and expands the device name if you passed a format string to
6208 int register_netdev(struct net_device
*dev
)
6213 err
= register_netdevice(dev
);
6217 EXPORT_SYMBOL(register_netdev
);
6219 int netdev_refcnt_read(const struct net_device
*dev
)
6223 for_each_possible_cpu(i
)
6224 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
6227 EXPORT_SYMBOL(netdev_refcnt_read
);
6230 * netdev_wait_allrefs - wait until all references are gone.
6231 * @dev: target net_device
6233 * This is called when unregistering network devices.
6235 * Any protocol or device that holds a reference should register
6236 * for netdevice notification, and cleanup and put back the
6237 * reference if they receive an UNREGISTER event.
6238 * We can get stuck here if buggy protocols don't correctly
6241 static void netdev_wait_allrefs(struct net_device
*dev
)
6243 unsigned long rebroadcast_time
, warning_time
;
6246 linkwatch_forget_dev(dev
);
6248 rebroadcast_time
= warning_time
= jiffies
;
6249 refcnt
= netdev_refcnt_read(dev
);
6251 while (refcnt
!= 0) {
6252 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
6255 /* Rebroadcast unregister notification */
6256 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6262 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
6263 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
6265 /* We must not have linkwatch events
6266 * pending on unregister. If this
6267 * happens, we simply run the queue
6268 * unscheduled, resulting in a noop
6271 linkwatch_run_queue();
6276 rebroadcast_time
= jiffies
;
6281 refcnt
= netdev_refcnt_read(dev
);
6283 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
6284 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6286 warning_time
= jiffies
;
6295 * register_netdevice(x1);
6296 * register_netdevice(x2);
6298 * unregister_netdevice(y1);
6299 * unregister_netdevice(y2);
6305 * We are invoked by rtnl_unlock().
6306 * This allows us to deal with problems:
6307 * 1) We can delete sysfs objects which invoke hotplug
6308 * without deadlocking with linkwatch via keventd.
6309 * 2) Since we run with the RTNL semaphore not held, we can sleep
6310 * safely in order to wait for the netdev refcnt to drop to zero.
6312 * We must not return until all unregister events added during
6313 * the interval the lock was held have been completed.
6315 void netdev_run_todo(void)
6317 struct list_head list
;
6319 /* Snapshot list, allow later requests */
6320 list_replace_init(&net_todo_list
, &list
);
6325 /* Wait for rcu callbacks to finish before next phase */
6326 if (!list_empty(&list
))
6329 while (!list_empty(&list
)) {
6330 struct net_device
*dev
6331 = list_first_entry(&list
, struct net_device
, todo_list
);
6332 list_del(&dev
->todo_list
);
6335 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
6338 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
6339 pr_err("network todo '%s' but state %d\n",
6340 dev
->name
, dev
->reg_state
);
6345 dev
->reg_state
= NETREG_UNREGISTERED
;
6347 on_each_cpu(flush_backlog
, dev
, 1);
6349 netdev_wait_allrefs(dev
);
6352 BUG_ON(netdev_refcnt_read(dev
));
6353 WARN_ON(rcu_access_pointer(dev
->ip_ptr
));
6354 WARN_ON(rcu_access_pointer(dev
->ip6_ptr
));
6355 WARN_ON(dev
->dn_ptr
);
6357 if (dev
->destructor
)
6358 dev
->destructor(dev
);
6360 /* Report a network device has been unregistered */
6362 dev_net(dev
)->dev_unreg_count
--;
6364 wake_up(&netdev_unregistering_wq
);
6366 /* Free network device */
6367 kobject_put(&dev
->dev
.kobj
);
6371 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6372 * fields in the same order, with only the type differing.
6374 void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
6375 const struct net_device_stats
*netdev_stats
)
6377 #if BITS_PER_LONG == 64
6378 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
6379 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
6381 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
6382 const unsigned long *src
= (const unsigned long *)netdev_stats
;
6383 u64
*dst
= (u64
*)stats64
;
6385 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
6386 sizeof(*stats64
) / sizeof(u64
));
6387 for (i
= 0; i
< n
; i
++)
6391 EXPORT_SYMBOL(netdev_stats_to_stats64
);
6394 * dev_get_stats - get network device statistics
6395 * @dev: device to get statistics from
6396 * @storage: place to store stats
6398 * Get network statistics from device. Return @storage.
6399 * The device driver may provide its own method by setting
6400 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6401 * otherwise the internal statistics structure is used.
6403 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
6404 struct rtnl_link_stats64
*storage
)
6406 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6408 if (ops
->ndo_get_stats64
) {
6409 memset(storage
, 0, sizeof(*storage
));
6410 ops
->ndo_get_stats64(dev
, storage
);
6411 } else if (ops
->ndo_get_stats
) {
6412 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
6414 netdev_stats_to_stats64(storage
, &dev
->stats
);
6416 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
6417 storage
->tx_dropped
+= atomic_long_read(&dev
->tx_dropped
);
6420 EXPORT_SYMBOL(dev_get_stats
);
6422 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
6424 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
6426 #ifdef CONFIG_NET_CLS_ACT
6429 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
6432 netdev_init_one_queue(dev
, queue
, NULL
);
6433 queue
->qdisc
= &noop_qdisc
;
6434 queue
->qdisc_sleeping
= &noop_qdisc
;
6435 rcu_assign_pointer(dev
->ingress_queue
, queue
);
6440 static const struct ethtool_ops default_ethtool_ops
;
6442 void netdev_set_default_ethtool_ops(struct net_device
*dev
,
6443 const struct ethtool_ops
*ops
)
6445 if (dev
->ethtool_ops
== &default_ethtool_ops
)
6446 dev
->ethtool_ops
= ops
;
6448 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops
);
6450 void netdev_freemem(struct net_device
*dev
)
6452 char *addr
= (char *)dev
- dev
->padded
;
6458 * alloc_netdev_mqs - allocate network device
6459 * @sizeof_priv: size of private data to allocate space for
6460 * @name: device name format string
6461 * @name_assign_type: origin of device name
6462 * @setup: callback to initialize device
6463 * @txqs: the number of TX subqueues to allocate
6464 * @rxqs: the number of RX subqueues to allocate
6466 * Allocates a struct net_device with private data area for driver use
6467 * and performs basic initialization. Also allocates subqueue structs
6468 * for each queue on the device.
6470 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
6471 unsigned char name_assign_type
,
6472 void (*setup
)(struct net_device
*),
6473 unsigned int txqs
, unsigned int rxqs
)
6475 struct net_device
*dev
;
6477 struct net_device
*p
;
6479 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
6482 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6488 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6493 alloc_size
= sizeof(struct net_device
);
6495 /* ensure 32-byte alignment of private area */
6496 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
6497 alloc_size
+= sizeof_priv
;
6499 /* ensure 32-byte alignment of whole construct */
6500 alloc_size
+= NETDEV_ALIGN
- 1;
6502 p
= kzalloc(alloc_size
, GFP_KERNEL
| __GFP_NOWARN
| __GFP_REPEAT
);
6504 p
= vzalloc(alloc_size
);
6508 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
6509 dev
->padded
= (char *)dev
- (char *)p
;
6511 dev
->pcpu_refcnt
= alloc_percpu(int);
6512 if (!dev
->pcpu_refcnt
)
6515 if (dev_addr_init(dev
))
6521 dev_net_set(dev
, &init_net
);
6523 dev
->gso_max_size
= GSO_MAX_SIZE
;
6524 dev
->gso_max_segs
= GSO_MAX_SEGS
;
6526 INIT_LIST_HEAD(&dev
->napi_list
);
6527 INIT_LIST_HEAD(&dev
->unreg_list
);
6528 INIT_LIST_HEAD(&dev
->close_list
);
6529 INIT_LIST_HEAD(&dev
->link_watch_list
);
6530 INIT_LIST_HEAD(&dev
->adj_list
.upper
);
6531 INIT_LIST_HEAD(&dev
->adj_list
.lower
);
6532 INIT_LIST_HEAD(&dev
->all_adj_list
.upper
);
6533 INIT_LIST_HEAD(&dev
->all_adj_list
.lower
);
6534 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
6537 dev
->num_tx_queues
= txqs
;
6538 dev
->real_num_tx_queues
= txqs
;
6539 if (netif_alloc_netdev_queues(dev
))
6543 dev
->num_rx_queues
= rxqs
;
6544 dev
->real_num_rx_queues
= rxqs
;
6545 if (netif_alloc_rx_queues(dev
))
6549 strcpy(dev
->name
, name
);
6550 dev
->name_assign_type
= name_assign_type
;
6551 dev
->group
= INIT_NETDEV_GROUP
;
6552 if (!dev
->ethtool_ops
)
6553 dev
->ethtool_ops
= &default_ethtool_ops
;
6561 free_percpu(dev
->pcpu_refcnt
);
6563 netdev_freemem(dev
);
6566 EXPORT_SYMBOL(alloc_netdev_mqs
);
6569 * free_netdev - free network device
6572 * This function does the last stage of destroying an allocated device
6573 * interface. The reference to the device object is released.
6574 * If this is the last reference then it will be freed.
6576 void free_netdev(struct net_device
*dev
)
6578 struct napi_struct
*p
, *n
;
6580 release_net(dev_net(dev
));
6582 netif_free_tx_queues(dev
);
6587 kfree(rcu_dereference_protected(dev
->ingress_queue
, 1));
6589 /* Flush device addresses */
6590 dev_addr_flush(dev
);
6592 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
6595 free_percpu(dev
->pcpu_refcnt
);
6596 dev
->pcpu_refcnt
= NULL
;
6598 /* Compatibility with error handling in drivers */
6599 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
6600 netdev_freemem(dev
);
6604 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
6605 dev
->reg_state
= NETREG_RELEASED
;
6607 /* will free via device release */
6608 put_device(&dev
->dev
);
6610 EXPORT_SYMBOL(free_netdev
);
6613 * synchronize_net - Synchronize with packet receive processing
6615 * Wait for packets currently being received to be done.
6616 * Does not block later packets from starting.
6618 void synchronize_net(void)
6621 if (rtnl_is_locked())
6622 synchronize_rcu_expedited();
6626 EXPORT_SYMBOL(synchronize_net
);
6629 * unregister_netdevice_queue - remove device from the kernel
6633 * This function shuts down a device interface and removes it
6634 * from the kernel tables.
6635 * If head not NULL, device is queued to be unregistered later.
6637 * Callers must hold the rtnl semaphore. You may want
6638 * unregister_netdev() instead of this.
6641 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
6646 list_move_tail(&dev
->unreg_list
, head
);
6648 rollback_registered(dev
);
6649 /* Finish processing unregister after unlock */
6653 EXPORT_SYMBOL(unregister_netdevice_queue
);
6656 * unregister_netdevice_many - unregister many devices
6657 * @head: list of devices
6659 * Note: As most callers use a stack allocated list_head,
6660 * we force a list_del() to make sure stack wont be corrupted later.
6662 void unregister_netdevice_many(struct list_head
*head
)
6664 struct net_device
*dev
;
6666 if (!list_empty(head
)) {
6667 rollback_registered_many(head
);
6668 list_for_each_entry(dev
, head
, unreg_list
)
6673 EXPORT_SYMBOL(unregister_netdevice_many
);
6676 * unregister_netdev - remove device from the kernel
6679 * This function shuts down a device interface and removes it
6680 * from the kernel tables.
6682 * This is just a wrapper for unregister_netdevice that takes
6683 * the rtnl semaphore. In general you want to use this and not
6684 * unregister_netdevice.
6686 void unregister_netdev(struct net_device
*dev
)
6689 unregister_netdevice(dev
);
6692 EXPORT_SYMBOL(unregister_netdev
);
6695 * dev_change_net_namespace - move device to different nethost namespace
6697 * @net: network namespace
6698 * @pat: If not NULL name pattern to try if the current device name
6699 * is already taken in the destination network namespace.
6701 * This function shuts down a device interface and moves it
6702 * to a new network namespace. On success 0 is returned, on
6703 * a failure a netagive errno code is returned.
6705 * Callers must hold the rtnl semaphore.
6708 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
6714 /* Don't allow namespace local devices to be moved. */
6716 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6719 /* Ensure the device has been registrered */
6720 if (dev
->reg_state
!= NETREG_REGISTERED
)
6723 /* Get out if there is nothing todo */
6725 if (net_eq(dev_net(dev
), net
))
6728 /* Pick the destination device name, and ensure
6729 * we can use it in the destination network namespace.
6732 if (__dev_get_by_name(net
, dev
->name
)) {
6733 /* We get here if we can't use the current device name */
6736 if (dev_get_valid_name(net
, dev
, pat
) < 0)
6741 * And now a mini version of register_netdevice unregister_netdevice.
6744 /* If device is running close it first. */
6747 /* And unlink it from device chain */
6749 unlist_netdevice(dev
);
6753 /* Shutdown queueing discipline. */
6756 /* Notify protocols, that we are about to destroy
6757 this device. They should clean all the things.
6759 Note that dev->reg_state stays at NETREG_REGISTERED.
6760 This is wanted because this way 8021q and macvlan know
6761 the device is just moving and can keep their slaves up.
6763 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6765 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
6766 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U, GFP_KERNEL
);
6769 * Flush the unicast and multicast chains
6774 /* Send a netdev-removed uevent to the old namespace */
6775 kobject_uevent(&dev
->dev
.kobj
, KOBJ_REMOVE
);
6777 /* Actually switch the network namespace */
6778 dev_net_set(dev
, net
);
6780 /* If there is an ifindex conflict assign a new one */
6781 if (__dev_get_by_index(net
, dev
->ifindex
)) {
6782 int iflink
= (dev
->iflink
== dev
->ifindex
);
6783 dev
->ifindex
= dev_new_index(net
);
6785 dev
->iflink
= dev
->ifindex
;
6788 /* Send a netdev-add uevent to the new namespace */
6789 kobject_uevent(&dev
->dev
.kobj
, KOBJ_ADD
);
6791 /* Fixup kobjects */
6792 err
= device_rename(&dev
->dev
, dev
->name
);
6795 /* Add the device back in the hashes */
6796 list_netdevice(dev
);
6798 /* Notify protocols, that a new device appeared. */
6799 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
6802 * Prevent userspace races by waiting until the network
6803 * device is fully setup before sending notifications.
6805 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U, GFP_KERNEL
);
6812 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
6814 static int dev_cpu_callback(struct notifier_block
*nfb
,
6815 unsigned long action
,
6818 struct sk_buff
**list_skb
;
6819 struct sk_buff
*skb
;
6820 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
6821 struct softnet_data
*sd
, *oldsd
;
6823 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
6826 local_irq_disable();
6827 cpu
= smp_processor_id();
6828 sd
= &per_cpu(softnet_data
, cpu
);
6829 oldsd
= &per_cpu(softnet_data
, oldcpu
);
6831 /* Find end of our completion_queue. */
6832 list_skb
= &sd
->completion_queue
;
6834 list_skb
= &(*list_skb
)->next
;
6835 /* Append completion queue from offline CPU. */
6836 *list_skb
= oldsd
->completion_queue
;
6837 oldsd
->completion_queue
= NULL
;
6839 /* Append output queue from offline CPU. */
6840 if (oldsd
->output_queue
) {
6841 *sd
->output_queue_tailp
= oldsd
->output_queue
;
6842 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
6843 oldsd
->output_queue
= NULL
;
6844 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
6846 /* Append NAPI poll list from offline CPU. */
6847 if (!list_empty(&oldsd
->poll_list
)) {
6848 list_splice_init(&oldsd
->poll_list
, &sd
->poll_list
);
6849 raise_softirq_irqoff(NET_RX_SOFTIRQ
);
6852 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
6855 /* Process offline CPU's input_pkt_queue */
6856 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
6857 netif_rx_internal(skb
);
6858 input_queue_head_incr(oldsd
);
6860 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
6861 netif_rx_internal(skb
);
6862 input_queue_head_incr(oldsd
);
6870 * netdev_increment_features - increment feature set by one
6871 * @all: current feature set
6872 * @one: new feature set
6873 * @mask: mask feature set
6875 * Computes a new feature set after adding a device with feature set
6876 * @one to the master device with current feature set @all. Will not
6877 * enable anything that is off in @mask. Returns the new feature set.
6879 netdev_features_t
netdev_increment_features(netdev_features_t all
,
6880 netdev_features_t one
, netdev_features_t mask
)
6882 if (mask
& NETIF_F_GEN_CSUM
)
6883 mask
|= NETIF_F_ALL_CSUM
;
6884 mask
|= NETIF_F_VLAN_CHALLENGED
;
6886 all
|= one
& (NETIF_F_ONE_FOR_ALL
|NETIF_F_ALL_CSUM
) & mask
;
6887 all
&= one
| ~NETIF_F_ALL_FOR_ALL
;
6889 /* If one device supports hw checksumming, set for all. */
6890 if (all
& NETIF_F_GEN_CSUM
)
6891 all
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_GEN_CSUM
);
6895 EXPORT_SYMBOL(netdev_increment_features
);
6897 static struct hlist_head
* __net_init
netdev_create_hash(void)
6900 struct hlist_head
*hash
;
6902 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
6904 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
6905 INIT_HLIST_HEAD(&hash
[i
]);
6910 /* Initialize per network namespace state */
6911 static int __net_init
netdev_init(struct net
*net
)
6913 if (net
!= &init_net
)
6914 INIT_LIST_HEAD(&net
->dev_base_head
);
6916 net
->dev_name_head
= netdev_create_hash();
6917 if (net
->dev_name_head
== NULL
)
6920 net
->dev_index_head
= netdev_create_hash();
6921 if (net
->dev_index_head
== NULL
)
6927 kfree(net
->dev_name_head
);
6933 * netdev_drivername - network driver for the device
6934 * @dev: network device
6936 * Determine network driver for device.
6938 const char *netdev_drivername(const struct net_device
*dev
)
6940 const struct device_driver
*driver
;
6941 const struct device
*parent
;
6942 const char *empty
= "";
6944 parent
= dev
->dev
.parent
;
6948 driver
= parent
->driver
;
6949 if (driver
&& driver
->name
)
6950 return driver
->name
;
6954 static int __netdev_printk(const char *level
, const struct net_device
*dev
,
6955 struct va_format
*vaf
)
6959 if (dev
&& dev
->dev
.parent
) {
6960 r
= dev_printk_emit(level
[1] - '0',
6963 dev_driver_string(dev
->dev
.parent
),
6964 dev_name(dev
->dev
.parent
),
6965 netdev_name(dev
), netdev_reg_state(dev
),
6968 r
= printk("%s%s%s: %pV", level
, netdev_name(dev
),
6969 netdev_reg_state(dev
), vaf
);
6971 r
= printk("%s(NULL net_device): %pV", level
, vaf
);
6977 int netdev_printk(const char *level
, const struct net_device
*dev
,
6978 const char *format
, ...)
6980 struct va_format vaf
;
6984 va_start(args
, format
);
6989 r
= __netdev_printk(level
, dev
, &vaf
);
6995 EXPORT_SYMBOL(netdev_printk
);
6997 #define define_netdev_printk_level(func, level) \
6998 int func(const struct net_device *dev, const char *fmt, ...) \
7001 struct va_format vaf; \
7004 va_start(args, fmt); \
7009 r = __netdev_printk(level, dev, &vaf); \
7015 EXPORT_SYMBOL(func);
7017 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
7018 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
7019 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
7020 define_netdev_printk_level(netdev_err
, KERN_ERR
);
7021 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
7022 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
7023 define_netdev_printk_level(netdev_info
, KERN_INFO
);
7025 static void __net_exit
netdev_exit(struct net
*net
)
7027 kfree(net
->dev_name_head
);
7028 kfree(net
->dev_index_head
);
7031 static struct pernet_operations __net_initdata netdev_net_ops
= {
7032 .init
= netdev_init
,
7033 .exit
= netdev_exit
,
7036 static void __net_exit
default_device_exit(struct net
*net
)
7038 struct net_device
*dev
, *aux
;
7040 * Push all migratable network devices back to the
7041 * initial network namespace
7044 for_each_netdev_safe(net
, dev
, aux
) {
7046 char fb_name
[IFNAMSIZ
];
7048 /* Ignore unmoveable devices (i.e. loopback) */
7049 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
7052 /* Leave virtual devices for the generic cleanup */
7053 if (dev
->rtnl_link_ops
)
7056 /* Push remaining network devices to init_net */
7057 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
7058 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
7060 pr_emerg("%s: failed to move %s to init_net: %d\n",
7061 __func__
, dev
->name
, err
);
7068 static void __net_exit
rtnl_lock_unregistering(struct list_head
*net_list
)
7070 /* Return with the rtnl_lock held when there are no network
7071 * devices unregistering in any network namespace in net_list.
7078 prepare_to_wait(&netdev_unregistering_wq
, &wait
,
7079 TASK_UNINTERRUPTIBLE
);
7080 unregistering
= false;
7082 list_for_each_entry(net
, net_list
, exit_list
) {
7083 if (net
->dev_unreg_count
> 0) {
7084 unregistering
= true;
7093 finish_wait(&netdev_unregistering_wq
, &wait
);
7096 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
7098 /* At exit all network devices most be removed from a network
7099 * namespace. Do this in the reverse order of registration.
7100 * Do this across as many network namespaces as possible to
7101 * improve batching efficiency.
7103 struct net_device
*dev
;
7105 LIST_HEAD(dev_kill_list
);
7107 /* To prevent network device cleanup code from dereferencing
7108 * loopback devices or network devices that have been freed
7109 * wait here for all pending unregistrations to complete,
7110 * before unregistring the loopback device and allowing the
7111 * network namespace be freed.
7113 * The netdev todo list containing all network devices
7114 * unregistrations that happen in default_device_exit_batch
7115 * will run in the rtnl_unlock() at the end of
7116 * default_device_exit_batch.
7118 rtnl_lock_unregistering(net_list
);
7119 list_for_each_entry(net
, net_list
, exit_list
) {
7120 for_each_netdev_reverse(net
, dev
) {
7121 if (dev
->rtnl_link_ops
&& dev
->rtnl_link_ops
->dellink
)
7122 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
7124 unregister_netdevice_queue(dev
, &dev_kill_list
);
7127 unregister_netdevice_many(&dev_kill_list
);
7131 static struct pernet_operations __net_initdata default_device_ops
= {
7132 .exit
= default_device_exit
,
7133 .exit_batch
= default_device_exit_batch
,
7137 * Initialize the DEV module. At boot time this walks the device list and
7138 * unhooks any devices that fail to initialise (normally hardware not
7139 * present) and leaves us with a valid list of present and active devices.
7144 * This is called single threaded during boot, so no need
7145 * to take the rtnl semaphore.
7147 static int __init
net_dev_init(void)
7149 int i
, rc
= -ENOMEM
;
7151 BUG_ON(!dev_boot_phase
);
7153 if (dev_proc_init())
7156 if (netdev_kobject_init())
7159 INIT_LIST_HEAD(&ptype_all
);
7160 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
7161 INIT_LIST_HEAD(&ptype_base
[i
]);
7163 INIT_LIST_HEAD(&offload_base
);
7165 if (register_pernet_subsys(&netdev_net_ops
))
7169 * Initialise the packet receive queues.
7172 for_each_possible_cpu(i
) {
7173 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
7175 skb_queue_head_init(&sd
->input_pkt_queue
);
7176 skb_queue_head_init(&sd
->process_queue
);
7177 INIT_LIST_HEAD(&sd
->poll_list
);
7178 sd
->output_queue_tailp
= &sd
->output_queue
;
7180 sd
->csd
.func
= rps_trigger_softirq
;
7185 sd
->backlog
.poll
= process_backlog
;
7186 sd
->backlog
.weight
= weight_p
;
7191 /* The loopback device is special if any other network devices
7192 * is present in a network namespace the loopback device must
7193 * be present. Since we now dynamically allocate and free the
7194 * loopback device ensure this invariant is maintained by
7195 * keeping the loopback device as the first device on the
7196 * list of network devices. Ensuring the loopback devices
7197 * is the first device that appears and the last network device
7200 if (register_pernet_device(&loopback_net_ops
))
7203 if (register_pernet_device(&default_device_ops
))
7206 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
7207 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
7209 hotcpu_notifier(dev_cpu_callback
, 0);
7216 subsys_initcall(net_dev_init
);