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>
136 #include "net-sysfs.h"
138 /* Instead of increasing this, you should create a hash table. */
139 #define MAX_GRO_SKBS 8
141 /* This should be increased if a protocol with a bigger head is added. */
142 #define GRO_MAX_HEAD (MAX_HEADER + 128)
144 static DEFINE_SPINLOCK(ptype_lock
);
145 static DEFINE_SPINLOCK(offload_lock
);
146 struct list_head ptype_base
[PTYPE_HASH_SIZE
] __read_mostly
;
147 struct list_head ptype_all __read_mostly
; /* Taps */
148 static struct list_head offload_base __read_mostly
;
150 static int netif_rx_internal(struct sk_buff
*skb
);
153 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
156 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
158 * Writers must hold the rtnl semaphore while they loop through the
159 * dev_base_head list, and hold dev_base_lock for writing when they do the
160 * actual updates. This allows pure readers to access the list even
161 * while a writer is preparing to update it.
163 * To put it another way, dev_base_lock is held for writing only to
164 * protect against pure readers; the rtnl semaphore provides the
165 * protection against other writers.
167 * See, for example usages, register_netdevice() and
168 * unregister_netdevice(), which must be called with the rtnl
171 DEFINE_RWLOCK(dev_base_lock
);
172 EXPORT_SYMBOL(dev_base_lock
);
174 /* protects napi_hash addition/deletion and napi_gen_id */
175 static DEFINE_SPINLOCK(napi_hash_lock
);
177 static unsigned int napi_gen_id
;
178 static DEFINE_HASHTABLE(napi_hash
, 8);
180 static seqcount_t devnet_rename_seq
;
182 static inline void dev_base_seq_inc(struct net
*net
)
184 while (++net
->dev_base_seq
== 0);
187 static inline struct hlist_head
*dev_name_hash(struct net
*net
, const char *name
)
189 unsigned int hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
191 return &net
->dev_name_head
[hash_32(hash
, NETDEV_HASHBITS
)];
194 static inline struct hlist_head
*dev_index_hash(struct net
*net
, int ifindex
)
196 return &net
->dev_index_head
[ifindex
& (NETDEV_HASHENTRIES
- 1)];
199 static inline void rps_lock(struct softnet_data
*sd
)
202 spin_lock(&sd
->input_pkt_queue
.lock
);
206 static inline void rps_unlock(struct softnet_data
*sd
)
209 spin_unlock(&sd
->input_pkt_queue
.lock
);
213 /* Device list insertion */
214 static void list_netdevice(struct net_device
*dev
)
216 struct net
*net
= dev_net(dev
);
220 write_lock_bh(&dev_base_lock
);
221 list_add_tail_rcu(&dev
->dev_list
, &net
->dev_base_head
);
222 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
223 hlist_add_head_rcu(&dev
->index_hlist
,
224 dev_index_hash(net
, dev
->ifindex
));
225 write_unlock_bh(&dev_base_lock
);
227 dev_base_seq_inc(net
);
230 /* Device list removal
231 * caller must respect a RCU grace period before freeing/reusing dev
233 static void unlist_netdevice(struct net_device
*dev
)
237 /* Unlink dev from the device chain */
238 write_lock_bh(&dev_base_lock
);
239 list_del_rcu(&dev
->dev_list
);
240 hlist_del_rcu(&dev
->name_hlist
);
241 hlist_del_rcu(&dev
->index_hlist
);
242 write_unlock_bh(&dev_base_lock
);
244 dev_base_seq_inc(dev_net(dev
));
251 static RAW_NOTIFIER_HEAD(netdev_chain
);
254 * Device drivers call our routines to queue packets here. We empty the
255 * queue in the local softnet handler.
258 DEFINE_PER_CPU_ALIGNED(struct softnet_data
, softnet_data
);
259 EXPORT_PER_CPU_SYMBOL(softnet_data
);
261 #ifdef CONFIG_LOCKDEP
263 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
264 * according to dev->type
266 static const unsigned short netdev_lock_type
[] =
267 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
268 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
269 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
270 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
271 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
272 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
273 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
274 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
275 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
276 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
277 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
278 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
279 ARPHRD_FCFABRIC
, ARPHRD_IEEE80211
, ARPHRD_IEEE80211_PRISM
,
280 ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_PHONET
, ARPHRD_PHONET_PIPE
,
281 ARPHRD_IEEE802154
, ARPHRD_VOID
, ARPHRD_NONE
};
283 static const char *const netdev_lock_name
[] =
284 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
285 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
286 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
287 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
288 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
289 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
290 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
291 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
292 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
293 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
294 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
295 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
296 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
297 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
298 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
300 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
301 static struct lock_class_key netdev_addr_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
303 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
307 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
308 if (netdev_lock_type
[i
] == dev_type
)
310 /* the last key is used by default */
311 return ARRAY_SIZE(netdev_lock_type
) - 1;
314 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
315 unsigned short dev_type
)
319 i
= netdev_lock_pos(dev_type
);
320 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
321 netdev_lock_name
[i
]);
324 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
328 i
= netdev_lock_pos(dev
->type
);
329 lockdep_set_class_and_name(&dev
->addr_list_lock
,
330 &netdev_addr_lock_key
[i
],
331 netdev_lock_name
[i
]);
334 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
335 unsigned short dev_type
)
338 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
343 /*******************************************************************************
345 Protocol management and registration routines
347 *******************************************************************************/
350 * Add a protocol ID to the list. Now that the input handler is
351 * smarter we can dispense with all the messy stuff that used to be
354 * BEWARE!!! Protocol handlers, mangling input packets,
355 * MUST BE last in hash buckets and checking protocol handlers
356 * MUST start from promiscuous ptype_all chain in net_bh.
357 * It is true now, do not change it.
358 * Explanation follows: if protocol handler, mangling packet, will
359 * be the first on list, it is not able to sense, that packet
360 * is cloned and should be copied-on-write, so that it will
361 * change it and subsequent readers will get broken packet.
365 static inline struct list_head
*ptype_head(const struct packet_type
*pt
)
367 if (pt
->type
== htons(ETH_P_ALL
))
370 return &ptype_base
[ntohs(pt
->type
) & PTYPE_HASH_MASK
];
374 * dev_add_pack - add packet handler
375 * @pt: packet type declaration
377 * Add a protocol handler to the networking stack. The passed &packet_type
378 * is linked into kernel lists and may not be freed until it has been
379 * removed from the kernel lists.
381 * This call does not sleep therefore it can not
382 * guarantee all CPU's that are in middle of receiving packets
383 * will see the new packet type (until the next received packet).
386 void dev_add_pack(struct packet_type
*pt
)
388 struct list_head
*head
= ptype_head(pt
);
390 spin_lock(&ptype_lock
);
391 list_add_rcu(&pt
->list
, head
);
392 spin_unlock(&ptype_lock
);
394 EXPORT_SYMBOL(dev_add_pack
);
397 * __dev_remove_pack - remove packet handler
398 * @pt: packet type declaration
400 * Remove a protocol handler that was previously added to the kernel
401 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
402 * from the kernel lists and can be freed or reused once this function
405 * The packet type might still be in use by receivers
406 * and must not be freed until after all the CPU's have gone
407 * through a quiescent state.
409 void __dev_remove_pack(struct packet_type
*pt
)
411 struct list_head
*head
= ptype_head(pt
);
412 struct packet_type
*pt1
;
414 spin_lock(&ptype_lock
);
416 list_for_each_entry(pt1
, head
, list
) {
418 list_del_rcu(&pt
->list
);
423 pr_warn("dev_remove_pack: %p not found\n", pt
);
425 spin_unlock(&ptype_lock
);
427 EXPORT_SYMBOL(__dev_remove_pack
);
430 * dev_remove_pack - remove packet handler
431 * @pt: packet type declaration
433 * Remove a protocol handler that was previously added to the kernel
434 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
435 * from the kernel lists and can be freed or reused once this function
438 * This call sleeps to guarantee that no CPU is looking at the packet
441 void dev_remove_pack(struct packet_type
*pt
)
443 __dev_remove_pack(pt
);
447 EXPORT_SYMBOL(dev_remove_pack
);
451 * dev_add_offload - register offload handlers
452 * @po: protocol offload declaration
454 * Add protocol offload handlers to the networking stack. The passed
455 * &proto_offload is linked into kernel lists and may not be freed until
456 * it has been removed from the kernel lists.
458 * This call does not sleep therefore it can not
459 * guarantee all CPU's that are in middle of receiving packets
460 * will see the new offload handlers (until the next received packet).
462 void dev_add_offload(struct packet_offload
*po
)
464 struct list_head
*head
= &offload_base
;
466 spin_lock(&offload_lock
);
467 list_add_rcu(&po
->list
, head
);
468 spin_unlock(&offload_lock
);
470 EXPORT_SYMBOL(dev_add_offload
);
473 * __dev_remove_offload - remove offload handler
474 * @po: packet offload declaration
476 * Remove a protocol offload handler that was previously added to the
477 * kernel offload handlers by dev_add_offload(). The passed &offload_type
478 * is removed from the kernel lists and can be freed or reused once this
481 * The packet type might still be in use by receivers
482 * and must not be freed until after all the CPU's have gone
483 * through a quiescent state.
485 static void __dev_remove_offload(struct packet_offload
*po
)
487 struct list_head
*head
= &offload_base
;
488 struct packet_offload
*po1
;
490 spin_lock(&offload_lock
);
492 list_for_each_entry(po1
, head
, list
) {
494 list_del_rcu(&po
->list
);
499 pr_warn("dev_remove_offload: %p not found\n", po
);
501 spin_unlock(&offload_lock
);
505 * dev_remove_offload - remove packet offload handler
506 * @po: packet offload declaration
508 * Remove a packet offload handler that was previously added to the kernel
509 * offload handlers by dev_add_offload(). The passed &offload_type is
510 * removed from the kernel lists and can be freed or reused once this
513 * This call sleeps to guarantee that no CPU is looking at the packet
516 void dev_remove_offload(struct packet_offload
*po
)
518 __dev_remove_offload(po
);
522 EXPORT_SYMBOL(dev_remove_offload
);
524 /******************************************************************************
526 Device Boot-time Settings Routines
528 *******************************************************************************/
530 /* Boot time configuration table */
531 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
534 * netdev_boot_setup_add - add new setup entry
535 * @name: name of the device
536 * @map: configured settings for the device
538 * Adds new setup entry to the dev_boot_setup list. The function
539 * returns 0 on error and 1 on success. This is a generic routine to
542 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
544 struct netdev_boot_setup
*s
;
548 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
549 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
550 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
551 strlcpy(s
[i
].name
, name
, IFNAMSIZ
);
552 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
557 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
561 * netdev_boot_setup_check - check boot time settings
562 * @dev: the netdevice
564 * Check boot time settings for the device.
565 * The found settings are set for the device to be used
566 * later in the device probing.
567 * Returns 0 if no settings found, 1 if they are.
569 int netdev_boot_setup_check(struct net_device
*dev
)
571 struct netdev_boot_setup
*s
= dev_boot_setup
;
574 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
575 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
576 !strcmp(dev
->name
, s
[i
].name
)) {
577 dev
->irq
= s
[i
].map
.irq
;
578 dev
->base_addr
= s
[i
].map
.base_addr
;
579 dev
->mem_start
= s
[i
].map
.mem_start
;
580 dev
->mem_end
= s
[i
].map
.mem_end
;
586 EXPORT_SYMBOL(netdev_boot_setup_check
);
590 * netdev_boot_base - get address from boot time settings
591 * @prefix: prefix for network device
592 * @unit: id for network device
594 * Check boot time settings for the base address of device.
595 * The found settings are set for the device to be used
596 * later in the device probing.
597 * Returns 0 if no settings found.
599 unsigned long netdev_boot_base(const char *prefix
, int unit
)
601 const struct netdev_boot_setup
*s
= dev_boot_setup
;
605 sprintf(name
, "%s%d", prefix
, unit
);
608 * If device already registered then return base of 1
609 * to indicate not to probe for this interface
611 if (__dev_get_by_name(&init_net
, name
))
614 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
615 if (!strcmp(name
, s
[i
].name
))
616 return s
[i
].map
.base_addr
;
621 * Saves at boot time configured settings for any netdevice.
623 int __init
netdev_boot_setup(char *str
)
628 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
633 memset(&map
, 0, sizeof(map
));
637 map
.base_addr
= ints
[2];
639 map
.mem_start
= ints
[3];
641 map
.mem_end
= ints
[4];
643 /* Add new entry to the list */
644 return netdev_boot_setup_add(str
, &map
);
647 __setup("netdev=", netdev_boot_setup
);
649 /*******************************************************************************
651 Device Interface Subroutines
653 *******************************************************************************/
656 * __dev_get_by_name - find a device by its name
657 * @net: the applicable net namespace
658 * @name: name to find
660 * Find an interface by name. Must be called under RTNL semaphore
661 * or @dev_base_lock. If the name is found a pointer to the device
662 * is returned. If the name is not found then %NULL is returned. The
663 * reference counters are not incremented so the caller must be
664 * careful with locks.
667 struct net_device
*__dev_get_by_name(struct net
*net
, const char *name
)
669 struct net_device
*dev
;
670 struct hlist_head
*head
= dev_name_hash(net
, name
);
672 hlist_for_each_entry(dev
, head
, name_hlist
)
673 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
678 EXPORT_SYMBOL(__dev_get_by_name
);
681 * dev_get_by_name_rcu - find a device by its name
682 * @net: the applicable net namespace
683 * @name: name to find
685 * Find an interface by name.
686 * If the name is found a pointer to the device is returned.
687 * If the name is not found then %NULL is returned.
688 * The reference counters are not incremented so the caller must be
689 * careful with locks. The caller must hold RCU lock.
692 struct net_device
*dev_get_by_name_rcu(struct net
*net
, const char *name
)
694 struct net_device
*dev
;
695 struct hlist_head
*head
= dev_name_hash(net
, name
);
697 hlist_for_each_entry_rcu(dev
, head
, name_hlist
)
698 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
703 EXPORT_SYMBOL(dev_get_by_name_rcu
);
706 * dev_get_by_name - find a device by its name
707 * @net: the applicable net namespace
708 * @name: name to find
710 * Find an interface by name. This can be called from any
711 * context and does its own locking. The returned handle has
712 * the usage count incremented and the caller must use dev_put() to
713 * release it when it is no longer needed. %NULL is returned if no
714 * matching device is found.
717 struct net_device
*dev_get_by_name(struct net
*net
, const char *name
)
719 struct net_device
*dev
;
722 dev
= dev_get_by_name_rcu(net
, name
);
728 EXPORT_SYMBOL(dev_get_by_name
);
731 * __dev_get_by_index - find a device by its ifindex
732 * @net: the applicable net namespace
733 * @ifindex: index of device
735 * Search for an interface by index. Returns %NULL if the device
736 * is not found or a pointer to the device. The device has not
737 * had its reference counter increased so the caller must be careful
738 * about locking. The caller must hold either the RTNL semaphore
742 struct net_device
*__dev_get_by_index(struct net
*net
, int ifindex
)
744 struct net_device
*dev
;
745 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
747 hlist_for_each_entry(dev
, head
, index_hlist
)
748 if (dev
->ifindex
== ifindex
)
753 EXPORT_SYMBOL(__dev_get_by_index
);
756 * dev_get_by_index_rcu - find a device by its ifindex
757 * @net: the applicable net namespace
758 * @ifindex: index of device
760 * Search for an interface by index. Returns %NULL if the device
761 * is not found or a pointer to the device. The device has not
762 * had its reference counter increased so the caller must be careful
763 * about locking. The caller must hold RCU lock.
766 struct net_device
*dev_get_by_index_rcu(struct net
*net
, int ifindex
)
768 struct net_device
*dev
;
769 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
771 hlist_for_each_entry_rcu(dev
, head
, index_hlist
)
772 if (dev
->ifindex
== ifindex
)
777 EXPORT_SYMBOL(dev_get_by_index_rcu
);
781 * dev_get_by_index - find a device by its ifindex
782 * @net: the applicable net namespace
783 * @ifindex: index of device
785 * Search for an interface by index. Returns NULL if the device
786 * is not found or a pointer to the device. The device returned has
787 * had a reference added and the pointer is safe until the user calls
788 * dev_put to indicate they have finished with it.
791 struct net_device
*dev_get_by_index(struct net
*net
, int ifindex
)
793 struct net_device
*dev
;
796 dev
= dev_get_by_index_rcu(net
, ifindex
);
802 EXPORT_SYMBOL(dev_get_by_index
);
805 * netdev_get_name - get a netdevice name, knowing its ifindex.
806 * @net: network namespace
807 * @name: a pointer to the buffer where the name will be stored.
808 * @ifindex: the ifindex of the interface to get the name from.
810 * The use of raw_seqcount_begin() and cond_resched() before
811 * retrying is required as we want to give the writers a chance
812 * to complete when CONFIG_PREEMPT is not set.
814 int netdev_get_name(struct net
*net
, char *name
, int ifindex
)
816 struct net_device
*dev
;
820 seq
= raw_seqcount_begin(&devnet_rename_seq
);
822 dev
= dev_get_by_index_rcu(net
, ifindex
);
828 strcpy(name
, dev
->name
);
830 if (read_seqcount_retry(&devnet_rename_seq
, seq
)) {
839 * dev_getbyhwaddr_rcu - find a device by its hardware address
840 * @net: the applicable net namespace
841 * @type: media type of device
842 * @ha: hardware address
844 * Search for an interface by MAC address. Returns NULL if the device
845 * is not found or a pointer to the device.
846 * The caller must hold RCU or RTNL.
847 * The returned device has not had its ref count increased
848 * and the caller must therefore be careful about locking
852 struct net_device
*dev_getbyhwaddr_rcu(struct net
*net
, unsigned short type
,
855 struct net_device
*dev
;
857 for_each_netdev_rcu(net
, dev
)
858 if (dev
->type
== type
&&
859 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
864 EXPORT_SYMBOL(dev_getbyhwaddr_rcu
);
866 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
868 struct net_device
*dev
;
871 for_each_netdev(net
, dev
)
872 if (dev
->type
== type
)
877 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
879 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
881 struct net_device
*dev
, *ret
= NULL
;
884 for_each_netdev_rcu(net
, dev
)
885 if (dev
->type
== type
) {
893 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
896 * dev_get_by_flags_rcu - find any device with given flags
897 * @net: the applicable net namespace
898 * @if_flags: IFF_* values
899 * @mask: bitmask of bits in if_flags to check
901 * Search for any interface with the given flags. Returns NULL if a device
902 * is not found or a pointer to the device. Must be called inside
903 * rcu_read_lock(), and result refcount is unchanged.
906 struct net_device
*dev_get_by_flags_rcu(struct net
*net
, unsigned short if_flags
,
909 struct net_device
*dev
, *ret
;
912 for_each_netdev_rcu(net
, dev
) {
913 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
920 EXPORT_SYMBOL(dev_get_by_flags_rcu
);
923 * dev_valid_name - check if name is okay for network device
926 * Network device names need to be valid file names to
927 * to allow sysfs to work. We also disallow any kind of
930 bool dev_valid_name(const char *name
)
934 if (strlen(name
) >= IFNAMSIZ
)
936 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
940 if (*name
== '/' || isspace(*name
))
946 EXPORT_SYMBOL(dev_valid_name
);
949 * __dev_alloc_name - allocate a name for a device
950 * @net: network namespace to allocate the device name in
951 * @name: name format string
952 * @buf: scratch buffer and result name string
954 * Passed a format string - eg "lt%d" it will try and find a suitable
955 * id. It scans list of devices to build up a free map, then chooses
956 * the first empty slot. The caller must hold the dev_base or rtnl lock
957 * while allocating the name and adding the device in order to avoid
959 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
960 * Returns the number of the unit assigned or a negative errno code.
963 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
967 const int max_netdevices
= 8*PAGE_SIZE
;
968 unsigned long *inuse
;
969 struct net_device
*d
;
971 p
= strnchr(name
, IFNAMSIZ
-1, '%');
974 * Verify the string as this thing may have come from
975 * the user. There must be either one "%d" and no other "%"
978 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
981 /* Use one page as a bit array of possible slots */
982 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
986 for_each_netdev(net
, d
) {
987 if (!sscanf(d
->name
, name
, &i
))
989 if (i
< 0 || i
>= max_netdevices
)
992 /* avoid cases where sscanf is not exact inverse of printf */
993 snprintf(buf
, IFNAMSIZ
, name
, i
);
994 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
998 i
= find_first_zero_bit(inuse
, max_netdevices
);
999 free_page((unsigned long) inuse
);
1003 snprintf(buf
, IFNAMSIZ
, name
, i
);
1004 if (!__dev_get_by_name(net
, buf
))
1007 /* It is possible to run out of possible slots
1008 * when the name is long and there isn't enough space left
1009 * for the digits, or if all bits are used.
1015 * dev_alloc_name - allocate a name for a device
1017 * @name: name format string
1019 * Passed a format string - eg "lt%d" it will try and find a suitable
1020 * id. It scans list of devices to build up a free map, then chooses
1021 * the first empty slot. The caller must hold the dev_base or rtnl lock
1022 * while allocating the name and adding the device in order to avoid
1024 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1025 * Returns the number of the unit assigned or a negative errno code.
1028 int dev_alloc_name(struct net_device
*dev
, const char *name
)
1034 BUG_ON(!dev_net(dev
));
1036 ret
= __dev_alloc_name(net
, name
, buf
);
1038 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
1041 EXPORT_SYMBOL(dev_alloc_name
);
1043 static int dev_alloc_name_ns(struct net
*net
,
1044 struct net_device
*dev
,
1050 ret
= __dev_alloc_name(net
, name
, buf
);
1052 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
1056 static int dev_get_valid_name(struct net
*net
,
1057 struct net_device
*dev
,
1062 if (!dev_valid_name(name
))
1065 if (strchr(name
, '%'))
1066 return dev_alloc_name_ns(net
, dev
, name
);
1067 else if (__dev_get_by_name(net
, name
))
1069 else if (dev
->name
!= name
)
1070 strlcpy(dev
->name
, name
, IFNAMSIZ
);
1076 * dev_change_name - change name of a device
1078 * @newname: name (or format string) must be at least IFNAMSIZ
1080 * Change name of a device, can pass format strings "eth%d".
1083 int dev_change_name(struct net_device
*dev
, const char *newname
)
1085 char oldname
[IFNAMSIZ
];
1091 BUG_ON(!dev_net(dev
));
1094 if (dev
->flags
& IFF_UP
)
1097 write_seqcount_begin(&devnet_rename_seq
);
1099 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0) {
1100 write_seqcount_end(&devnet_rename_seq
);
1104 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
1106 err
= dev_get_valid_name(net
, dev
, newname
);
1108 write_seqcount_end(&devnet_rename_seq
);
1113 ret
= device_rename(&dev
->dev
, dev
->name
);
1115 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1116 write_seqcount_end(&devnet_rename_seq
);
1120 write_seqcount_end(&devnet_rename_seq
);
1122 netdev_adjacent_rename_links(dev
, oldname
);
1124 write_lock_bh(&dev_base_lock
);
1125 hlist_del_rcu(&dev
->name_hlist
);
1126 write_unlock_bh(&dev_base_lock
);
1130 write_lock_bh(&dev_base_lock
);
1131 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
1132 write_unlock_bh(&dev_base_lock
);
1134 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
1135 ret
= notifier_to_errno(ret
);
1138 /* err >= 0 after dev_alloc_name() or stores the first errno */
1141 write_seqcount_begin(&devnet_rename_seq
);
1142 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1143 memcpy(oldname
, newname
, IFNAMSIZ
);
1146 pr_err("%s: name change rollback failed: %d\n",
1155 * dev_set_alias - change ifalias of a device
1157 * @alias: name up to IFALIASZ
1158 * @len: limit of bytes to copy from info
1160 * Set ifalias for a device,
1162 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
1168 if (len
>= IFALIASZ
)
1172 kfree(dev
->ifalias
);
1173 dev
->ifalias
= NULL
;
1177 new_ifalias
= krealloc(dev
->ifalias
, len
+ 1, GFP_KERNEL
);
1180 dev
->ifalias
= new_ifalias
;
1182 strlcpy(dev
->ifalias
, alias
, len
+1);
1188 * netdev_features_change - device changes features
1189 * @dev: device to cause notification
1191 * Called to indicate a device has changed features.
1193 void netdev_features_change(struct net_device
*dev
)
1195 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1197 EXPORT_SYMBOL(netdev_features_change
);
1200 * netdev_state_change - device changes state
1201 * @dev: device to cause notification
1203 * Called to indicate a device has changed state. This function calls
1204 * the notifier chains for netdev_chain and sends a NEWLINK message
1205 * to the routing socket.
1207 void netdev_state_change(struct net_device
*dev
)
1209 if (dev
->flags
& IFF_UP
) {
1210 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
1211 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0, GFP_KERNEL
);
1214 EXPORT_SYMBOL(netdev_state_change
);
1217 * netdev_notify_peers - notify network peers about existence of @dev
1218 * @dev: network device
1220 * Generate traffic such that interested network peers are aware of
1221 * @dev, such as by generating a gratuitous ARP. This may be used when
1222 * a device wants to inform the rest of the network about some sort of
1223 * reconfiguration such as a failover event or virtual machine
1226 void netdev_notify_peers(struct net_device
*dev
)
1229 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
, dev
);
1232 EXPORT_SYMBOL(netdev_notify_peers
);
1234 static int __dev_open(struct net_device
*dev
)
1236 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1241 if (!netif_device_present(dev
))
1244 /* Block netpoll from trying to do any rx path servicing.
1245 * If we don't do this there is a chance ndo_poll_controller
1246 * or ndo_poll may be running while we open the device
1248 netpoll_poll_disable(dev
);
1250 ret
= call_netdevice_notifiers(NETDEV_PRE_UP
, dev
);
1251 ret
= notifier_to_errno(ret
);
1255 set_bit(__LINK_STATE_START
, &dev
->state
);
1257 if (ops
->ndo_validate_addr
)
1258 ret
= ops
->ndo_validate_addr(dev
);
1260 if (!ret
&& ops
->ndo_open
)
1261 ret
= ops
->ndo_open(dev
);
1263 netpoll_poll_enable(dev
);
1266 clear_bit(__LINK_STATE_START
, &dev
->state
);
1268 dev
->flags
|= IFF_UP
;
1269 net_dmaengine_get();
1270 dev_set_rx_mode(dev
);
1272 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
1279 * dev_open - prepare an interface for use.
1280 * @dev: device to open
1282 * Takes a device from down to up state. The device's private open
1283 * function is invoked and then the multicast lists are loaded. Finally
1284 * the device is moved into the up state and a %NETDEV_UP message is
1285 * sent to the netdev notifier chain.
1287 * Calling this function on an active interface is a nop. On a failure
1288 * a negative errno code is returned.
1290 int dev_open(struct net_device
*dev
)
1294 if (dev
->flags
& IFF_UP
)
1297 ret
= __dev_open(dev
);
1301 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
, GFP_KERNEL
);
1302 call_netdevice_notifiers(NETDEV_UP
, dev
);
1306 EXPORT_SYMBOL(dev_open
);
1308 static int __dev_close_many(struct list_head
*head
)
1310 struct net_device
*dev
;
1315 list_for_each_entry(dev
, head
, close_list
) {
1316 /* Temporarily disable netpoll until the interface is down */
1317 netpoll_poll_disable(dev
);
1319 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1321 clear_bit(__LINK_STATE_START
, &dev
->state
);
1323 /* Synchronize to scheduled poll. We cannot touch poll list, it
1324 * can be even on different cpu. So just clear netif_running().
1326 * dev->stop() will invoke napi_disable() on all of it's
1327 * napi_struct instances on this device.
1329 smp_mb__after_atomic(); /* Commit netif_running(). */
1332 dev_deactivate_many(head
);
1334 list_for_each_entry(dev
, head
, close_list
) {
1335 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1338 * Call the device specific close. This cannot fail.
1339 * Only if device is UP
1341 * We allow it to be called even after a DETACH hot-plug
1347 dev
->flags
&= ~IFF_UP
;
1348 net_dmaengine_put();
1349 netpoll_poll_enable(dev
);
1355 static int __dev_close(struct net_device
*dev
)
1360 list_add(&dev
->close_list
, &single
);
1361 retval
= __dev_close_many(&single
);
1367 static int dev_close_many(struct list_head
*head
)
1369 struct net_device
*dev
, *tmp
;
1371 /* Remove the devices that don't need to be closed */
1372 list_for_each_entry_safe(dev
, tmp
, head
, close_list
)
1373 if (!(dev
->flags
& IFF_UP
))
1374 list_del_init(&dev
->close_list
);
1376 __dev_close_many(head
);
1378 list_for_each_entry_safe(dev
, tmp
, head
, close_list
) {
1379 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
, GFP_KERNEL
);
1380 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1381 list_del_init(&dev
->close_list
);
1388 * dev_close - shutdown an interface.
1389 * @dev: device to shutdown
1391 * This function moves an active device into down state. A
1392 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1393 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1396 int dev_close(struct net_device
*dev
)
1398 if (dev
->flags
& IFF_UP
) {
1401 list_add(&dev
->close_list
, &single
);
1402 dev_close_many(&single
);
1407 EXPORT_SYMBOL(dev_close
);
1411 * dev_disable_lro - disable Large Receive Offload on a device
1414 * Disable Large Receive Offload (LRO) on a net device. Must be
1415 * called under RTNL. This is needed if received packets may be
1416 * forwarded to another interface.
1418 void dev_disable_lro(struct net_device
*dev
)
1421 * If we're trying to disable lro on a vlan device
1422 * use the underlying physical device instead
1424 if (is_vlan_dev(dev
))
1425 dev
= vlan_dev_real_dev(dev
);
1427 /* the same for macvlan devices */
1428 if (netif_is_macvlan(dev
))
1429 dev
= macvlan_dev_real_dev(dev
);
1431 dev
->wanted_features
&= ~NETIF_F_LRO
;
1432 netdev_update_features(dev
);
1434 if (unlikely(dev
->features
& NETIF_F_LRO
))
1435 netdev_WARN(dev
, "failed to disable LRO!\n");
1437 EXPORT_SYMBOL(dev_disable_lro
);
1439 static int call_netdevice_notifier(struct notifier_block
*nb
, unsigned long val
,
1440 struct net_device
*dev
)
1442 struct netdev_notifier_info info
;
1444 netdev_notifier_info_init(&info
, dev
);
1445 return nb
->notifier_call(nb
, val
, &info
);
1448 static int dev_boot_phase
= 1;
1451 * register_netdevice_notifier - register a network notifier block
1454 * Register a notifier to be called when network device events occur.
1455 * The notifier passed is linked into the kernel structures and must
1456 * not be reused until it has been unregistered. A negative errno code
1457 * is returned on a failure.
1459 * When registered all registration and up events are replayed
1460 * to the new notifier to allow device to have a race free
1461 * view of the network device list.
1464 int register_netdevice_notifier(struct notifier_block
*nb
)
1466 struct net_device
*dev
;
1467 struct net_device
*last
;
1472 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1478 for_each_netdev(net
, dev
) {
1479 err
= call_netdevice_notifier(nb
, NETDEV_REGISTER
, dev
);
1480 err
= notifier_to_errno(err
);
1484 if (!(dev
->flags
& IFF_UP
))
1487 call_netdevice_notifier(nb
, NETDEV_UP
, dev
);
1498 for_each_netdev(net
, dev
) {
1502 if (dev
->flags
& IFF_UP
) {
1503 call_netdevice_notifier(nb
, NETDEV_GOING_DOWN
,
1505 call_netdevice_notifier(nb
, NETDEV_DOWN
, dev
);
1507 call_netdevice_notifier(nb
, NETDEV_UNREGISTER
, dev
);
1512 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1515 EXPORT_SYMBOL(register_netdevice_notifier
);
1518 * unregister_netdevice_notifier - unregister a network notifier block
1521 * Unregister a notifier previously registered by
1522 * register_netdevice_notifier(). The notifier is unlinked into the
1523 * kernel structures and may then be reused. A negative errno code
1524 * is returned on a failure.
1526 * After unregistering unregister and down device events are synthesized
1527 * for all devices on the device list to the removed notifier to remove
1528 * the need for special case cleanup code.
1531 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1533 struct net_device
*dev
;
1538 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1543 for_each_netdev(net
, dev
) {
1544 if (dev
->flags
& IFF_UP
) {
1545 call_netdevice_notifier(nb
, NETDEV_GOING_DOWN
,
1547 call_netdevice_notifier(nb
, NETDEV_DOWN
, dev
);
1549 call_netdevice_notifier(nb
, NETDEV_UNREGISTER
, dev
);
1556 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1559 * call_netdevice_notifiers_info - call all network notifier blocks
1560 * @val: value passed unmodified to notifier function
1561 * @dev: net_device pointer passed unmodified to notifier function
1562 * @info: notifier information data
1564 * Call all network notifier blocks. Parameters and return value
1565 * are as for raw_notifier_call_chain().
1568 static int call_netdevice_notifiers_info(unsigned long val
,
1569 struct net_device
*dev
,
1570 struct netdev_notifier_info
*info
)
1573 netdev_notifier_info_init(info
, dev
);
1574 return raw_notifier_call_chain(&netdev_chain
, val
, info
);
1578 * call_netdevice_notifiers - call all network notifier blocks
1579 * @val: value passed unmodified to notifier function
1580 * @dev: net_device pointer passed unmodified to notifier function
1582 * Call all network notifier blocks. Parameters and return value
1583 * are as for raw_notifier_call_chain().
1586 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1588 struct netdev_notifier_info info
;
1590 return call_netdevice_notifiers_info(val
, dev
, &info
);
1592 EXPORT_SYMBOL(call_netdevice_notifiers
);
1594 static struct static_key netstamp_needed __read_mostly
;
1595 #ifdef HAVE_JUMP_LABEL
1596 /* We are not allowed to call static_key_slow_dec() from irq context
1597 * If net_disable_timestamp() is called from irq context, defer the
1598 * static_key_slow_dec() calls.
1600 static atomic_t netstamp_needed_deferred
;
1603 void net_enable_timestamp(void)
1605 #ifdef HAVE_JUMP_LABEL
1606 int deferred
= atomic_xchg(&netstamp_needed_deferred
, 0);
1610 static_key_slow_dec(&netstamp_needed
);
1614 static_key_slow_inc(&netstamp_needed
);
1616 EXPORT_SYMBOL(net_enable_timestamp
);
1618 void net_disable_timestamp(void)
1620 #ifdef HAVE_JUMP_LABEL
1621 if (in_interrupt()) {
1622 atomic_inc(&netstamp_needed_deferred
);
1626 static_key_slow_dec(&netstamp_needed
);
1628 EXPORT_SYMBOL(net_disable_timestamp
);
1630 static inline void net_timestamp_set(struct sk_buff
*skb
)
1632 skb
->tstamp
.tv64
= 0;
1633 if (static_key_false(&netstamp_needed
))
1634 __net_timestamp(skb
);
1637 #define net_timestamp_check(COND, SKB) \
1638 if (static_key_false(&netstamp_needed)) { \
1639 if ((COND) && !(SKB)->tstamp.tv64) \
1640 __net_timestamp(SKB); \
1643 bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb)
1647 if (!(dev
->flags
& IFF_UP
))
1650 len
= dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
;
1651 if (skb
->len
<= len
)
1654 /* if TSO is enabled, we don't care about the length as the packet
1655 * could be forwarded without being segmented before
1657 if (skb_is_gso(skb
))
1662 EXPORT_SYMBOL_GPL(is_skb_forwardable
);
1664 int __dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1666 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
1667 if (skb_copy_ubufs(skb
, GFP_ATOMIC
)) {
1668 atomic_long_inc(&dev
->rx_dropped
);
1674 if (unlikely(!is_skb_forwardable(dev
, skb
))) {
1675 atomic_long_inc(&dev
->rx_dropped
);
1680 skb_scrub_packet(skb
, true);
1681 skb
->protocol
= eth_type_trans(skb
, dev
);
1685 EXPORT_SYMBOL_GPL(__dev_forward_skb
);
1688 * dev_forward_skb - loopback an skb to another netif
1690 * @dev: destination network device
1691 * @skb: buffer to forward
1694 * NET_RX_SUCCESS (no congestion)
1695 * NET_RX_DROP (packet was dropped, but freed)
1697 * dev_forward_skb can be used for injecting an skb from the
1698 * start_xmit function of one device into the receive queue
1699 * of another device.
1701 * The receiving device may be in another namespace, so
1702 * we have to clear all information in the skb that could
1703 * impact namespace isolation.
1705 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1707 return __dev_forward_skb(dev
, skb
) ?: netif_rx_internal(skb
);
1709 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1711 static inline int deliver_skb(struct sk_buff
*skb
,
1712 struct packet_type
*pt_prev
,
1713 struct net_device
*orig_dev
)
1715 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
1717 atomic_inc(&skb
->users
);
1718 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1721 static inline bool skb_loop_sk(struct packet_type
*ptype
, struct sk_buff
*skb
)
1723 if (!ptype
->af_packet_priv
|| !skb
->sk
)
1726 if (ptype
->id_match
)
1727 return ptype
->id_match(ptype
, skb
->sk
);
1728 else if ((struct sock
*)ptype
->af_packet_priv
== skb
->sk
)
1735 * Support routine. Sends outgoing frames to any network
1736 * taps currently in use.
1739 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1741 struct packet_type
*ptype
;
1742 struct sk_buff
*skb2
= NULL
;
1743 struct packet_type
*pt_prev
= NULL
;
1746 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1747 /* Never send packets back to the socket
1748 * they originated from - MvS (miquels@drinkel.ow.org)
1750 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1751 (!skb_loop_sk(ptype
, skb
))) {
1753 deliver_skb(skb2
, pt_prev
, skb
->dev
);
1758 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1762 net_timestamp_set(skb2
);
1764 /* skb->nh should be correctly
1765 set by sender, so that the second statement is
1766 just protection against buggy protocols.
1768 skb_reset_mac_header(skb2
);
1770 if (skb_network_header(skb2
) < skb2
->data
||
1771 skb_network_header(skb2
) > skb_tail_pointer(skb2
)) {
1772 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1773 ntohs(skb2
->protocol
),
1775 skb_reset_network_header(skb2
);
1778 skb2
->transport_header
= skb2
->network_header
;
1779 skb2
->pkt_type
= PACKET_OUTGOING
;
1784 pt_prev
->func(skb2
, skb
->dev
, pt_prev
, skb
->dev
);
1789 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1790 * @dev: Network device
1791 * @txq: number of queues available
1793 * If real_num_tx_queues is changed the tc mappings may no longer be
1794 * valid. To resolve this verify the tc mapping remains valid and if
1795 * not NULL the mapping. With no priorities mapping to this
1796 * offset/count pair it will no longer be used. In the worst case TC0
1797 * is invalid nothing can be done so disable priority mappings. If is
1798 * expected that drivers will fix this mapping if they can before
1799 * calling netif_set_real_num_tx_queues.
1801 static void netif_setup_tc(struct net_device
*dev
, unsigned int txq
)
1804 struct netdev_tc_txq
*tc
= &dev
->tc_to_txq
[0];
1806 /* If TC0 is invalidated disable TC mapping */
1807 if (tc
->offset
+ tc
->count
> txq
) {
1808 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1813 /* Invalidated prio to tc mappings set to TC0 */
1814 for (i
= 1; i
< TC_BITMASK
+ 1; i
++) {
1815 int q
= netdev_get_prio_tc_map(dev
, i
);
1817 tc
= &dev
->tc_to_txq
[q
];
1818 if (tc
->offset
+ tc
->count
> txq
) {
1819 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1821 netdev_set_prio_tc_map(dev
, i
, 0);
1827 static DEFINE_MUTEX(xps_map_mutex
);
1828 #define xmap_dereference(P) \
1829 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1831 static struct xps_map
*remove_xps_queue(struct xps_dev_maps
*dev_maps
,
1834 struct xps_map
*map
= NULL
;
1838 map
= xmap_dereference(dev_maps
->cpu_map
[cpu
]);
1840 for (pos
= 0; map
&& pos
< map
->len
; pos
++) {
1841 if (map
->queues
[pos
] == index
) {
1843 map
->queues
[pos
] = map
->queues
[--map
->len
];
1845 RCU_INIT_POINTER(dev_maps
->cpu_map
[cpu
], NULL
);
1846 kfree_rcu(map
, rcu
);
1856 static void netif_reset_xps_queues_gt(struct net_device
*dev
, u16 index
)
1858 struct xps_dev_maps
*dev_maps
;
1860 bool active
= false;
1862 mutex_lock(&xps_map_mutex
);
1863 dev_maps
= xmap_dereference(dev
->xps_maps
);
1868 for_each_possible_cpu(cpu
) {
1869 for (i
= index
; i
< dev
->num_tx_queues
; i
++) {
1870 if (!remove_xps_queue(dev_maps
, cpu
, i
))
1873 if (i
== dev
->num_tx_queues
)
1878 RCU_INIT_POINTER(dev
->xps_maps
, NULL
);
1879 kfree_rcu(dev_maps
, rcu
);
1882 for (i
= index
; i
< dev
->num_tx_queues
; i
++)
1883 netdev_queue_numa_node_write(netdev_get_tx_queue(dev
, i
),
1887 mutex_unlock(&xps_map_mutex
);
1890 static struct xps_map
*expand_xps_map(struct xps_map
*map
,
1893 struct xps_map
*new_map
;
1894 int alloc_len
= XPS_MIN_MAP_ALLOC
;
1897 for (pos
= 0; map
&& pos
< map
->len
; pos
++) {
1898 if (map
->queues
[pos
] != index
)
1903 /* Need to add queue to this CPU's existing map */
1905 if (pos
< map
->alloc_len
)
1908 alloc_len
= map
->alloc_len
* 2;
1911 /* Need to allocate new map to store queue on this CPU's map */
1912 new_map
= kzalloc_node(XPS_MAP_SIZE(alloc_len
), GFP_KERNEL
,
1917 for (i
= 0; i
< pos
; i
++)
1918 new_map
->queues
[i
] = map
->queues
[i
];
1919 new_map
->alloc_len
= alloc_len
;
1925 int netif_set_xps_queue(struct net_device
*dev
, const struct cpumask
*mask
,
1928 struct xps_dev_maps
*dev_maps
, *new_dev_maps
= NULL
;
1929 struct xps_map
*map
, *new_map
;
1930 int maps_sz
= max_t(unsigned int, XPS_DEV_MAPS_SIZE
, L1_CACHE_BYTES
);
1931 int cpu
, numa_node_id
= -2;
1932 bool active
= false;
1934 mutex_lock(&xps_map_mutex
);
1936 dev_maps
= xmap_dereference(dev
->xps_maps
);
1938 /* allocate memory for queue storage */
1939 for_each_online_cpu(cpu
) {
1940 if (!cpumask_test_cpu(cpu
, mask
))
1944 new_dev_maps
= kzalloc(maps_sz
, GFP_KERNEL
);
1945 if (!new_dev_maps
) {
1946 mutex_unlock(&xps_map_mutex
);
1950 map
= dev_maps
? xmap_dereference(dev_maps
->cpu_map
[cpu
]) :
1953 map
= expand_xps_map(map
, cpu
, index
);
1957 RCU_INIT_POINTER(new_dev_maps
->cpu_map
[cpu
], map
);
1961 goto out_no_new_maps
;
1963 for_each_possible_cpu(cpu
) {
1964 if (cpumask_test_cpu(cpu
, mask
) && cpu_online(cpu
)) {
1965 /* add queue to CPU maps */
1968 map
= xmap_dereference(new_dev_maps
->cpu_map
[cpu
]);
1969 while ((pos
< map
->len
) && (map
->queues
[pos
] != index
))
1972 if (pos
== map
->len
)
1973 map
->queues
[map
->len
++] = index
;
1975 if (numa_node_id
== -2)
1976 numa_node_id
= cpu_to_node(cpu
);
1977 else if (numa_node_id
!= cpu_to_node(cpu
))
1980 } else if (dev_maps
) {
1981 /* fill in the new device map from the old device map */
1982 map
= xmap_dereference(dev_maps
->cpu_map
[cpu
]);
1983 RCU_INIT_POINTER(new_dev_maps
->cpu_map
[cpu
], map
);
1988 rcu_assign_pointer(dev
->xps_maps
, new_dev_maps
);
1990 /* Cleanup old maps */
1992 for_each_possible_cpu(cpu
) {
1993 new_map
= xmap_dereference(new_dev_maps
->cpu_map
[cpu
]);
1994 map
= xmap_dereference(dev_maps
->cpu_map
[cpu
]);
1995 if (map
&& map
!= new_map
)
1996 kfree_rcu(map
, rcu
);
1999 kfree_rcu(dev_maps
, rcu
);
2002 dev_maps
= new_dev_maps
;
2006 /* update Tx queue numa node */
2007 netdev_queue_numa_node_write(netdev_get_tx_queue(dev
, index
),
2008 (numa_node_id
>= 0) ? numa_node_id
:
2014 /* removes queue from unused CPUs */
2015 for_each_possible_cpu(cpu
) {
2016 if (cpumask_test_cpu(cpu
, mask
) && cpu_online(cpu
))
2019 if (remove_xps_queue(dev_maps
, cpu
, index
))
2023 /* free map if not active */
2025 RCU_INIT_POINTER(dev
->xps_maps
, NULL
);
2026 kfree_rcu(dev_maps
, rcu
);
2030 mutex_unlock(&xps_map_mutex
);
2034 /* remove any maps that we added */
2035 for_each_possible_cpu(cpu
) {
2036 new_map
= xmap_dereference(new_dev_maps
->cpu_map
[cpu
]);
2037 map
= dev_maps
? xmap_dereference(dev_maps
->cpu_map
[cpu
]) :
2039 if (new_map
&& new_map
!= map
)
2043 mutex_unlock(&xps_map_mutex
);
2045 kfree(new_dev_maps
);
2048 EXPORT_SYMBOL(netif_set_xps_queue
);
2052 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2053 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2055 int netif_set_real_num_tx_queues(struct net_device
*dev
, unsigned int txq
)
2059 if (txq
< 1 || txq
> dev
->num_tx_queues
)
2062 if (dev
->reg_state
== NETREG_REGISTERED
||
2063 dev
->reg_state
== NETREG_UNREGISTERING
) {
2066 rc
= netdev_queue_update_kobjects(dev
, dev
->real_num_tx_queues
,
2072 netif_setup_tc(dev
, txq
);
2074 if (txq
< dev
->real_num_tx_queues
) {
2075 qdisc_reset_all_tx_gt(dev
, txq
);
2077 netif_reset_xps_queues_gt(dev
, txq
);
2082 dev
->real_num_tx_queues
= txq
;
2085 EXPORT_SYMBOL(netif_set_real_num_tx_queues
);
2089 * netif_set_real_num_rx_queues - set actual number of RX queues used
2090 * @dev: Network device
2091 * @rxq: Actual number of RX queues
2093 * This must be called either with the rtnl_lock held or before
2094 * registration of the net device. Returns 0 on success, or a
2095 * negative error code. If called before registration, it always
2098 int netif_set_real_num_rx_queues(struct net_device
*dev
, unsigned int rxq
)
2102 if (rxq
< 1 || rxq
> dev
->num_rx_queues
)
2105 if (dev
->reg_state
== NETREG_REGISTERED
) {
2108 rc
= net_rx_queue_update_kobjects(dev
, dev
->real_num_rx_queues
,
2114 dev
->real_num_rx_queues
= rxq
;
2117 EXPORT_SYMBOL(netif_set_real_num_rx_queues
);
2121 * netif_get_num_default_rss_queues - default number of RSS queues
2123 * This routine should set an upper limit on the number of RSS queues
2124 * used by default by multiqueue devices.
2126 int netif_get_num_default_rss_queues(void)
2128 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES
, num_online_cpus());
2130 EXPORT_SYMBOL(netif_get_num_default_rss_queues
);
2132 static inline void __netif_reschedule(struct Qdisc
*q
)
2134 struct softnet_data
*sd
;
2135 unsigned long flags
;
2137 local_irq_save(flags
);
2138 sd
= &__get_cpu_var(softnet_data
);
2139 q
->next_sched
= NULL
;
2140 *sd
->output_queue_tailp
= q
;
2141 sd
->output_queue_tailp
= &q
->next_sched
;
2142 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
2143 local_irq_restore(flags
);
2146 void __netif_schedule(struct Qdisc
*q
)
2148 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
2149 __netif_reschedule(q
);
2151 EXPORT_SYMBOL(__netif_schedule
);
2153 struct dev_kfree_skb_cb
{
2154 enum skb_free_reason reason
;
2157 static struct dev_kfree_skb_cb
*get_kfree_skb_cb(const struct sk_buff
*skb
)
2159 return (struct dev_kfree_skb_cb
*)skb
->cb
;
2162 void __dev_kfree_skb_irq(struct sk_buff
*skb
, enum skb_free_reason reason
)
2164 unsigned long flags
;
2166 if (likely(atomic_read(&skb
->users
) == 1)) {
2168 atomic_set(&skb
->users
, 0);
2169 } else if (likely(!atomic_dec_and_test(&skb
->users
))) {
2172 get_kfree_skb_cb(skb
)->reason
= reason
;
2173 local_irq_save(flags
);
2174 skb
->next
= __this_cpu_read(softnet_data
.completion_queue
);
2175 __this_cpu_write(softnet_data
.completion_queue
, skb
);
2176 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
2177 local_irq_restore(flags
);
2179 EXPORT_SYMBOL(__dev_kfree_skb_irq
);
2181 void __dev_kfree_skb_any(struct sk_buff
*skb
, enum skb_free_reason reason
)
2183 if (in_irq() || irqs_disabled())
2184 __dev_kfree_skb_irq(skb
, reason
);
2188 EXPORT_SYMBOL(__dev_kfree_skb_any
);
2192 * netif_device_detach - mark device as removed
2193 * @dev: network device
2195 * Mark device as removed from system and therefore no longer available.
2197 void netif_device_detach(struct net_device
*dev
)
2199 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
2200 netif_running(dev
)) {
2201 netif_tx_stop_all_queues(dev
);
2204 EXPORT_SYMBOL(netif_device_detach
);
2207 * netif_device_attach - mark device as attached
2208 * @dev: network device
2210 * Mark device as attached from system and restart if needed.
2212 void netif_device_attach(struct net_device
*dev
)
2214 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
2215 netif_running(dev
)) {
2216 netif_tx_wake_all_queues(dev
);
2217 __netdev_watchdog_up(dev
);
2220 EXPORT_SYMBOL(netif_device_attach
);
2222 static void skb_warn_bad_offload(const struct sk_buff
*skb
)
2224 static const netdev_features_t null_features
= 0;
2225 struct net_device
*dev
= skb
->dev
;
2226 const char *driver
= "";
2228 if (!net_ratelimit())
2231 if (dev
&& dev
->dev
.parent
)
2232 driver
= dev_driver_string(dev
->dev
.parent
);
2234 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2235 "gso_type=%d ip_summed=%d\n",
2236 driver
, dev
? &dev
->features
: &null_features
,
2237 skb
->sk
? &skb
->sk
->sk_route_caps
: &null_features
,
2238 skb
->len
, skb
->data_len
, skb_shinfo(skb
)->gso_size
,
2239 skb_shinfo(skb
)->gso_type
, skb
->ip_summed
);
2243 * Invalidate hardware checksum when packet is to be mangled, and
2244 * complete checksum manually on outgoing path.
2246 int skb_checksum_help(struct sk_buff
*skb
)
2249 int ret
= 0, offset
;
2251 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
2252 goto out_set_summed
;
2254 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
2255 skb_warn_bad_offload(skb
);
2259 /* Before computing a checksum, we should make sure no frag could
2260 * be modified by an external entity : checksum could be wrong.
2262 if (skb_has_shared_frag(skb
)) {
2263 ret
= __skb_linearize(skb
);
2268 offset
= skb_checksum_start_offset(skb
);
2269 BUG_ON(offset
>= skb_headlen(skb
));
2270 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
2272 offset
+= skb
->csum_offset
;
2273 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
2275 if (skb_cloned(skb
) &&
2276 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
2277 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2282 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
2284 skb
->ip_summed
= CHECKSUM_NONE
;
2288 EXPORT_SYMBOL(skb_checksum_help
);
2290 __be16
skb_network_protocol(struct sk_buff
*skb
, int *depth
)
2292 unsigned int vlan_depth
= skb
->mac_len
;
2293 __be16 type
= skb
->protocol
;
2295 /* Tunnel gso handlers can set protocol to ethernet. */
2296 if (type
== htons(ETH_P_TEB
)) {
2299 if (unlikely(!pskb_may_pull(skb
, sizeof(struct ethhdr
))))
2302 eth
= (struct ethhdr
*)skb_mac_header(skb
);
2303 type
= eth
->h_proto
;
2306 /* if skb->protocol is 802.1Q/AD then the header should already be
2307 * present at mac_len - VLAN_HLEN (if mac_len > 0), or at
2308 * ETH_HLEN otherwise
2310 if (type
== htons(ETH_P_8021Q
) || type
== htons(ETH_P_8021AD
)) {
2312 if (unlikely(WARN_ON(vlan_depth
< VLAN_HLEN
)))
2314 vlan_depth
-= VLAN_HLEN
;
2316 vlan_depth
= ETH_HLEN
;
2319 struct vlan_hdr
*vh
;
2321 if (unlikely(!pskb_may_pull(skb
,
2322 vlan_depth
+ VLAN_HLEN
)))
2325 vh
= (struct vlan_hdr
*)(skb
->data
+ vlan_depth
);
2326 type
= vh
->h_vlan_encapsulated_proto
;
2327 vlan_depth
+= VLAN_HLEN
;
2328 } while (type
== htons(ETH_P_8021Q
) ||
2329 type
== htons(ETH_P_8021AD
));
2332 *depth
= vlan_depth
;
2338 * skb_mac_gso_segment - mac layer segmentation handler.
2339 * @skb: buffer to segment
2340 * @features: features for the output path (see dev->features)
2342 struct sk_buff
*skb_mac_gso_segment(struct sk_buff
*skb
,
2343 netdev_features_t features
)
2345 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
2346 struct packet_offload
*ptype
;
2347 int vlan_depth
= skb
->mac_len
;
2348 __be16 type
= skb_network_protocol(skb
, &vlan_depth
);
2350 if (unlikely(!type
))
2351 return ERR_PTR(-EINVAL
);
2353 __skb_pull(skb
, vlan_depth
);
2356 list_for_each_entry_rcu(ptype
, &offload_base
, list
) {
2357 if (ptype
->type
== type
&& ptype
->callbacks
.gso_segment
) {
2358 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
2361 err
= ptype
->callbacks
.gso_send_check(skb
);
2362 segs
= ERR_PTR(err
);
2363 if (err
|| skb_gso_ok(skb
, features
))
2365 __skb_push(skb
, (skb
->data
-
2366 skb_network_header(skb
)));
2368 segs
= ptype
->callbacks
.gso_segment(skb
, features
);
2374 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
2378 EXPORT_SYMBOL(skb_mac_gso_segment
);
2381 /* openvswitch calls this on rx path, so we need a different check.
2383 static inline bool skb_needs_check(struct sk_buff
*skb
, bool tx_path
)
2386 return skb
->ip_summed
!= CHECKSUM_PARTIAL
;
2388 return skb
->ip_summed
== CHECKSUM_NONE
;
2392 * __skb_gso_segment - Perform segmentation on skb.
2393 * @skb: buffer to segment
2394 * @features: features for the output path (see dev->features)
2395 * @tx_path: whether it is called in TX path
2397 * This function segments the given skb and returns a list of segments.
2399 * It may return NULL if the skb requires no segmentation. This is
2400 * only possible when GSO is used for verifying header integrity.
2402 struct sk_buff
*__skb_gso_segment(struct sk_buff
*skb
,
2403 netdev_features_t features
, bool tx_path
)
2405 if (unlikely(skb_needs_check(skb
, tx_path
))) {
2408 skb_warn_bad_offload(skb
);
2410 if (skb_header_cloned(skb
) &&
2411 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
2412 return ERR_PTR(err
);
2415 SKB_GSO_CB(skb
)->mac_offset
= skb_headroom(skb
);
2416 SKB_GSO_CB(skb
)->encap_level
= 0;
2418 skb_reset_mac_header(skb
);
2419 skb_reset_mac_len(skb
);
2421 return skb_mac_gso_segment(skb
, features
);
2423 EXPORT_SYMBOL(__skb_gso_segment
);
2425 /* Take action when hardware reception checksum errors are detected. */
2427 void netdev_rx_csum_fault(struct net_device
*dev
)
2429 if (net_ratelimit()) {
2430 pr_err("%s: hw csum failure\n", dev
? dev
->name
: "<unknown>");
2434 EXPORT_SYMBOL(netdev_rx_csum_fault
);
2437 /* Actually, we should eliminate this check as soon as we know, that:
2438 * 1. IOMMU is present and allows to map all the memory.
2439 * 2. No high memory really exists on this machine.
2442 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
2444 #ifdef CONFIG_HIGHMEM
2446 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
2447 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2448 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2449 if (PageHighMem(skb_frag_page(frag
)))
2454 if (PCI_DMA_BUS_IS_PHYS
) {
2455 struct device
*pdev
= dev
->dev
.parent
;
2459 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2460 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2461 dma_addr_t addr
= page_to_phys(skb_frag_page(frag
));
2462 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
2471 void (*destructor
)(struct sk_buff
*skb
);
2474 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2476 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
2478 struct dev_gso_cb
*cb
;
2480 kfree_skb_list(skb
->next
);
2483 cb
= DEV_GSO_CB(skb
);
2485 cb
->destructor(skb
);
2489 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2490 * @skb: buffer to segment
2491 * @features: device features as applicable to this skb
2493 * This function segments the given skb and stores the list of segments
2496 static int dev_gso_segment(struct sk_buff
*skb
, netdev_features_t features
)
2498 struct sk_buff
*segs
;
2500 segs
= skb_gso_segment(skb
, features
);
2502 /* Verifying header integrity only. */
2507 return PTR_ERR(segs
);
2510 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
2511 skb
->destructor
= dev_gso_skb_destructor
;
2516 /* If MPLS offload request, verify we are testing hardware MPLS features
2517 * instead of standard features for the netdev.
2519 #ifdef CONFIG_NET_MPLS_GSO
2520 static netdev_features_t
net_mpls_features(struct sk_buff
*skb
,
2521 netdev_features_t features
,
2524 if (type
== htons(ETH_P_MPLS_UC
) || type
== htons(ETH_P_MPLS_MC
))
2525 features
&= skb
->dev
->mpls_features
;
2530 static netdev_features_t
net_mpls_features(struct sk_buff
*skb
,
2531 netdev_features_t features
,
2538 static netdev_features_t
harmonize_features(struct sk_buff
*skb
,
2539 netdev_features_t features
)
2544 type
= skb_network_protocol(skb
, &tmp
);
2545 features
= net_mpls_features(skb
, features
, type
);
2547 if (skb
->ip_summed
!= CHECKSUM_NONE
&&
2548 !can_checksum_protocol(features
, type
)) {
2549 features
&= ~NETIF_F_ALL_CSUM
;
2550 } else if (illegal_highdma(skb
->dev
, skb
)) {
2551 features
&= ~NETIF_F_SG
;
2557 netdev_features_t
netif_skb_features(struct sk_buff
*skb
)
2559 __be16 protocol
= skb
->protocol
;
2560 netdev_features_t features
= skb
->dev
->features
;
2562 if (skb_shinfo(skb
)->gso_segs
> skb
->dev
->gso_max_segs
)
2563 features
&= ~NETIF_F_GSO_MASK
;
2565 if (protocol
== htons(ETH_P_8021Q
) || protocol
== htons(ETH_P_8021AD
)) {
2566 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
2567 protocol
= veh
->h_vlan_encapsulated_proto
;
2568 } else if (!vlan_tx_tag_present(skb
)) {
2569 return harmonize_features(skb
, features
);
2572 features
&= (skb
->dev
->vlan_features
| NETIF_F_HW_VLAN_CTAG_TX
|
2573 NETIF_F_HW_VLAN_STAG_TX
);
2575 if (protocol
== htons(ETH_P_8021Q
) || protocol
== htons(ETH_P_8021AD
))
2576 features
&= NETIF_F_SG
| NETIF_F_HIGHDMA
| NETIF_F_FRAGLIST
|
2577 NETIF_F_GEN_CSUM
| NETIF_F_HW_VLAN_CTAG_TX
|
2578 NETIF_F_HW_VLAN_STAG_TX
;
2580 return harmonize_features(skb
, features
);
2582 EXPORT_SYMBOL(netif_skb_features
);
2584 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
2585 struct netdev_queue
*txq
)
2587 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2588 int rc
= NETDEV_TX_OK
;
2589 unsigned int skb_len
;
2591 if (likely(!skb
->next
)) {
2592 netdev_features_t features
;
2595 * If device doesn't need skb->dst, release it right now while
2596 * its hot in this cpu cache
2598 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2601 features
= netif_skb_features(skb
);
2603 if (vlan_tx_tag_present(skb
) &&
2604 !vlan_hw_offload_capable(features
, skb
->vlan_proto
)) {
2605 skb
= __vlan_put_tag(skb
, skb
->vlan_proto
,
2606 vlan_tx_tag_get(skb
));
2613 /* If encapsulation offload request, verify we are testing
2614 * hardware encapsulation features instead of standard
2615 * features for the netdev
2617 if (skb
->encapsulation
)
2618 features
&= dev
->hw_enc_features
;
2620 if (netif_needs_gso(skb
, features
)) {
2621 if (unlikely(dev_gso_segment(skb
, features
)))
2626 if (skb_needs_linearize(skb
, features
) &&
2627 __skb_linearize(skb
))
2630 /* If packet is not checksummed and device does not
2631 * support checksumming for this protocol, complete
2632 * checksumming here.
2634 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2635 if (skb
->encapsulation
)
2636 skb_set_inner_transport_header(skb
,
2637 skb_checksum_start_offset(skb
));
2639 skb_set_transport_header(skb
,
2640 skb_checksum_start_offset(skb
));
2641 if (!(features
& NETIF_F_ALL_CSUM
) &&
2642 skb_checksum_help(skb
))
2647 if (!list_empty(&ptype_all
))
2648 dev_queue_xmit_nit(skb
, dev
);
2651 trace_net_dev_start_xmit(skb
, dev
);
2652 rc
= ops
->ndo_start_xmit(skb
, dev
);
2653 trace_net_dev_xmit(skb
, rc
, dev
, skb_len
);
2654 if (rc
== NETDEV_TX_OK
)
2655 txq_trans_update(txq
);
2661 struct sk_buff
*nskb
= skb
->next
;
2663 skb
->next
= nskb
->next
;
2666 if (!list_empty(&ptype_all
))
2667 dev_queue_xmit_nit(nskb
, dev
);
2669 skb_len
= nskb
->len
;
2670 trace_net_dev_start_xmit(nskb
, dev
);
2671 rc
= ops
->ndo_start_xmit(nskb
, dev
);
2672 trace_net_dev_xmit(nskb
, rc
, dev
, skb_len
);
2673 if (unlikely(rc
!= NETDEV_TX_OK
)) {
2674 if (rc
& ~NETDEV_TX_MASK
)
2675 goto out_kfree_gso_skb
;
2676 nskb
->next
= skb
->next
;
2680 txq_trans_update(txq
);
2681 if (unlikely(netif_xmit_stopped(txq
) && skb
->next
))
2682 return NETDEV_TX_BUSY
;
2683 } while (skb
->next
);
2686 if (likely(skb
->next
== NULL
)) {
2687 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
2696 EXPORT_SYMBOL_GPL(dev_hard_start_xmit
);
2698 static void qdisc_pkt_len_init(struct sk_buff
*skb
)
2700 const struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
2702 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
2704 /* To get more precise estimation of bytes sent on wire,
2705 * we add to pkt_len the headers size of all segments
2707 if (shinfo
->gso_size
) {
2708 unsigned int hdr_len
;
2709 u16 gso_segs
= shinfo
->gso_segs
;
2711 /* mac layer + network layer */
2712 hdr_len
= skb_transport_header(skb
) - skb_mac_header(skb
);
2714 /* + transport layer */
2715 if (likely(shinfo
->gso_type
& (SKB_GSO_TCPV4
| SKB_GSO_TCPV6
)))
2716 hdr_len
+= tcp_hdrlen(skb
);
2718 hdr_len
+= sizeof(struct udphdr
);
2720 if (shinfo
->gso_type
& SKB_GSO_DODGY
)
2721 gso_segs
= DIV_ROUND_UP(skb
->len
- hdr_len
,
2724 qdisc_skb_cb(skb
)->pkt_len
+= (gso_segs
- 1) * hdr_len
;
2728 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2729 struct net_device
*dev
,
2730 struct netdev_queue
*txq
)
2732 spinlock_t
*root_lock
= qdisc_lock(q
);
2736 qdisc_pkt_len_init(skb
);
2737 qdisc_calculate_pkt_len(skb
, q
);
2739 * Heuristic to force contended enqueues to serialize on a
2740 * separate lock before trying to get qdisc main lock.
2741 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2742 * and dequeue packets faster.
2744 contended
= qdisc_is_running(q
);
2745 if (unlikely(contended
))
2746 spin_lock(&q
->busylock
);
2748 spin_lock(root_lock
);
2749 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2752 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2753 qdisc_run_begin(q
)) {
2755 * This is a work-conserving queue; there are no old skbs
2756 * waiting to be sent out; and the qdisc is not running -
2757 * xmit the skb directly.
2759 if (!(dev
->priv_flags
& IFF_XMIT_DST_RELEASE
))
2762 qdisc_bstats_update(q
, skb
);
2764 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
)) {
2765 if (unlikely(contended
)) {
2766 spin_unlock(&q
->busylock
);
2773 rc
= NET_XMIT_SUCCESS
;
2776 rc
= q
->enqueue(skb
, q
) & NET_XMIT_MASK
;
2777 if (qdisc_run_begin(q
)) {
2778 if (unlikely(contended
)) {
2779 spin_unlock(&q
->busylock
);
2785 spin_unlock(root_lock
);
2786 if (unlikely(contended
))
2787 spin_unlock(&q
->busylock
);
2791 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2792 static void skb_update_prio(struct sk_buff
*skb
)
2794 struct netprio_map
*map
= rcu_dereference_bh(skb
->dev
->priomap
);
2796 if (!skb
->priority
&& skb
->sk
&& map
) {
2797 unsigned int prioidx
= skb
->sk
->sk_cgrp_prioidx
;
2799 if (prioidx
< map
->priomap_len
)
2800 skb
->priority
= map
->priomap
[prioidx
];
2804 #define skb_update_prio(skb)
2807 static DEFINE_PER_CPU(int, xmit_recursion
);
2808 #define RECURSION_LIMIT 10
2811 * dev_loopback_xmit - loop back @skb
2812 * @skb: buffer to transmit
2814 int dev_loopback_xmit(struct sk_buff
*skb
)
2816 skb_reset_mac_header(skb
);
2817 __skb_pull(skb
, skb_network_offset(skb
));
2818 skb
->pkt_type
= PACKET_LOOPBACK
;
2819 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
2820 WARN_ON(!skb_dst(skb
));
2825 EXPORT_SYMBOL(dev_loopback_xmit
);
2828 * __dev_queue_xmit - transmit a buffer
2829 * @skb: buffer to transmit
2830 * @accel_priv: private data used for L2 forwarding offload
2832 * Queue a buffer for transmission to a network device. The caller must
2833 * have set the device and priority and built the buffer before calling
2834 * this function. The function can be called from an interrupt.
2836 * A negative errno code is returned on a failure. A success does not
2837 * guarantee the frame will be transmitted as it may be dropped due
2838 * to congestion or traffic shaping.
2840 * -----------------------------------------------------------------------------------
2841 * I notice this method can also return errors from the queue disciplines,
2842 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2845 * Regardless of the return value, the skb is consumed, so it is currently
2846 * difficult to retry a send to this method. (You can bump the ref count
2847 * before sending to hold a reference for retry if you are careful.)
2849 * When calling this method, interrupts MUST be enabled. This is because
2850 * the BH enable code must have IRQs enabled so that it will not deadlock.
2853 static int __dev_queue_xmit(struct sk_buff
*skb
, void *accel_priv
)
2855 struct net_device
*dev
= skb
->dev
;
2856 struct netdev_queue
*txq
;
2860 skb_reset_mac_header(skb
);
2862 /* Disable soft irqs for various locks below. Also
2863 * stops preemption for RCU.
2867 skb_update_prio(skb
);
2869 txq
= netdev_pick_tx(dev
, skb
, accel_priv
);
2870 q
= rcu_dereference_bh(txq
->qdisc
);
2872 #ifdef CONFIG_NET_CLS_ACT
2873 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2875 trace_net_dev_queue(skb
);
2877 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2881 /* The device has no queue. Common case for software devices:
2882 loopback, all the sorts of tunnels...
2884 Really, it is unlikely that netif_tx_lock protection is necessary
2885 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2887 However, it is possible, that they rely on protection
2890 Check this and shot the lock. It is not prone from deadlocks.
2891 Either shot noqueue qdisc, it is even simpler 8)
2893 if (dev
->flags
& IFF_UP
) {
2894 int cpu
= smp_processor_id(); /* ok because BHs are off */
2896 if (txq
->xmit_lock_owner
!= cpu
) {
2898 if (__this_cpu_read(xmit_recursion
) > RECURSION_LIMIT
)
2899 goto recursion_alert
;
2901 HARD_TX_LOCK(dev
, txq
, cpu
);
2903 if (!netif_xmit_stopped(txq
)) {
2904 __this_cpu_inc(xmit_recursion
);
2905 rc
= dev_hard_start_xmit(skb
, dev
, txq
);
2906 __this_cpu_dec(xmit_recursion
);
2907 if (dev_xmit_complete(rc
)) {
2908 HARD_TX_UNLOCK(dev
, txq
);
2912 HARD_TX_UNLOCK(dev
, txq
);
2913 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2916 /* Recursion is detected! It is possible,
2920 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2926 rcu_read_unlock_bh();
2928 atomic_long_inc(&dev
->tx_dropped
);
2932 rcu_read_unlock_bh();
2936 int dev_queue_xmit(struct sk_buff
*skb
)
2938 return __dev_queue_xmit(skb
, NULL
);
2940 EXPORT_SYMBOL(dev_queue_xmit
);
2942 int dev_queue_xmit_accel(struct sk_buff
*skb
, void *accel_priv
)
2944 return __dev_queue_xmit(skb
, accel_priv
);
2946 EXPORT_SYMBOL(dev_queue_xmit_accel
);
2949 /*=======================================================================
2951 =======================================================================*/
2953 int netdev_max_backlog __read_mostly
= 1000;
2954 EXPORT_SYMBOL(netdev_max_backlog
);
2956 int netdev_tstamp_prequeue __read_mostly
= 1;
2957 int netdev_budget __read_mostly
= 300;
2958 int weight_p __read_mostly
= 64; /* old backlog weight */
2960 /* Called with irq disabled */
2961 static inline void ____napi_schedule(struct softnet_data
*sd
,
2962 struct napi_struct
*napi
)
2964 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
2965 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2970 /* One global table that all flow-based protocols share. */
2971 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
2972 EXPORT_SYMBOL(rps_sock_flow_table
);
2974 struct static_key rps_needed __read_mostly
;
2976 static struct rps_dev_flow
*
2977 set_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2978 struct rps_dev_flow
*rflow
, u16 next_cpu
)
2980 if (next_cpu
!= RPS_NO_CPU
) {
2981 #ifdef CONFIG_RFS_ACCEL
2982 struct netdev_rx_queue
*rxqueue
;
2983 struct rps_dev_flow_table
*flow_table
;
2984 struct rps_dev_flow
*old_rflow
;
2989 /* Should we steer this flow to a different hardware queue? */
2990 if (!skb_rx_queue_recorded(skb
) || !dev
->rx_cpu_rmap
||
2991 !(dev
->features
& NETIF_F_NTUPLE
))
2993 rxq_index
= cpu_rmap_lookup_index(dev
->rx_cpu_rmap
, next_cpu
);
2994 if (rxq_index
== skb_get_rx_queue(skb
))
2997 rxqueue
= dev
->_rx
+ rxq_index
;
2998 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3001 flow_id
= skb_get_hash(skb
) & flow_table
->mask
;
3002 rc
= dev
->netdev_ops
->ndo_rx_flow_steer(dev
, skb
,
3003 rxq_index
, flow_id
);
3007 rflow
= &flow_table
->flows
[flow_id
];
3009 if (old_rflow
->filter
== rflow
->filter
)
3010 old_rflow
->filter
= RPS_NO_FILTER
;
3014 per_cpu(softnet_data
, next_cpu
).input_queue_head
;
3017 rflow
->cpu
= next_cpu
;
3022 * get_rps_cpu is called from netif_receive_skb and returns the target
3023 * CPU from the RPS map of the receiving queue for a given skb.
3024 * rcu_read_lock must be held on entry.
3026 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
3027 struct rps_dev_flow
**rflowp
)
3029 struct netdev_rx_queue
*rxqueue
;
3030 struct rps_map
*map
;
3031 struct rps_dev_flow_table
*flow_table
;
3032 struct rps_sock_flow_table
*sock_flow_table
;
3037 if (skb_rx_queue_recorded(skb
)) {
3038 u16 index
= skb_get_rx_queue(skb
);
3039 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
3040 WARN_ONCE(dev
->real_num_rx_queues
> 1,
3041 "%s received packet on queue %u, but number "
3042 "of RX queues is %u\n",
3043 dev
->name
, index
, dev
->real_num_rx_queues
);
3046 rxqueue
= dev
->_rx
+ index
;
3050 map
= rcu_dereference(rxqueue
->rps_map
);
3052 if (map
->len
== 1 &&
3053 !rcu_access_pointer(rxqueue
->rps_flow_table
)) {
3054 tcpu
= map
->cpus
[0];
3055 if (cpu_online(tcpu
))
3059 } else if (!rcu_access_pointer(rxqueue
->rps_flow_table
)) {
3063 skb_reset_network_header(skb
);
3064 hash
= skb_get_hash(skb
);
3068 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3069 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
3070 if (flow_table
&& sock_flow_table
) {
3072 struct rps_dev_flow
*rflow
;
3074 rflow
= &flow_table
->flows
[hash
& flow_table
->mask
];
3077 next_cpu
= sock_flow_table
->ents
[hash
& sock_flow_table
->mask
];
3080 * If the desired CPU (where last recvmsg was done) is
3081 * different from current CPU (one in the rx-queue flow
3082 * table entry), switch if one of the following holds:
3083 * - Current CPU is unset (equal to RPS_NO_CPU).
3084 * - Current CPU is offline.
3085 * - The current CPU's queue tail has advanced beyond the
3086 * last packet that was enqueued using this table entry.
3087 * This guarantees that all previous packets for the flow
3088 * have been dequeued, thus preserving in order delivery.
3090 if (unlikely(tcpu
!= next_cpu
) &&
3091 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
3092 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
3093 rflow
->last_qtail
)) >= 0)) {
3095 rflow
= set_rps_cpu(dev
, skb
, rflow
, next_cpu
);
3098 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
3106 tcpu
= map
->cpus
[((u64
) hash
* map
->len
) >> 32];
3108 if (cpu_online(tcpu
)) {
3118 #ifdef CONFIG_RFS_ACCEL
3121 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3122 * @dev: Device on which the filter was set
3123 * @rxq_index: RX queue index
3124 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3125 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3127 * Drivers that implement ndo_rx_flow_steer() should periodically call
3128 * this function for each installed filter and remove the filters for
3129 * which it returns %true.
3131 bool rps_may_expire_flow(struct net_device
*dev
, u16 rxq_index
,
3132 u32 flow_id
, u16 filter_id
)
3134 struct netdev_rx_queue
*rxqueue
= dev
->_rx
+ rxq_index
;
3135 struct rps_dev_flow_table
*flow_table
;
3136 struct rps_dev_flow
*rflow
;
3141 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3142 if (flow_table
&& flow_id
<= flow_table
->mask
) {
3143 rflow
= &flow_table
->flows
[flow_id
];
3144 cpu
= ACCESS_ONCE(rflow
->cpu
);
3145 if (rflow
->filter
== filter_id
&& cpu
!= RPS_NO_CPU
&&
3146 ((int)(per_cpu(softnet_data
, cpu
).input_queue_head
-
3147 rflow
->last_qtail
) <
3148 (int)(10 * flow_table
->mask
)))
3154 EXPORT_SYMBOL(rps_may_expire_flow
);
3156 #endif /* CONFIG_RFS_ACCEL */
3158 /* Called from hardirq (IPI) context */
3159 static void rps_trigger_softirq(void *data
)
3161 struct softnet_data
*sd
= data
;
3163 ____napi_schedule(sd
, &sd
->backlog
);
3167 #endif /* CONFIG_RPS */
3170 * Check if this softnet_data structure is another cpu one
3171 * If yes, queue it to our IPI list and return 1
3174 static int rps_ipi_queued(struct softnet_data
*sd
)
3177 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
3180 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
3181 mysd
->rps_ipi_list
= sd
;
3183 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3186 #endif /* CONFIG_RPS */
3190 #ifdef CONFIG_NET_FLOW_LIMIT
3191 int netdev_flow_limit_table_len __read_mostly
= (1 << 12);
3194 static bool skb_flow_limit(struct sk_buff
*skb
, unsigned int qlen
)
3196 #ifdef CONFIG_NET_FLOW_LIMIT
3197 struct sd_flow_limit
*fl
;
3198 struct softnet_data
*sd
;
3199 unsigned int old_flow
, new_flow
;
3201 if (qlen
< (netdev_max_backlog
>> 1))
3204 sd
= &__get_cpu_var(softnet_data
);
3207 fl
= rcu_dereference(sd
->flow_limit
);
3209 new_flow
= skb_get_hash(skb
) & (fl
->num_buckets
- 1);
3210 old_flow
= fl
->history
[fl
->history_head
];
3211 fl
->history
[fl
->history_head
] = new_flow
;
3214 fl
->history_head
&= FLOW_LIMIT_HISTORY
- 1;
3216 if (likely(fl
->buckets
[old_flow
]))
3217 fl
->buckets
[old_flow
]--;
3219 if (++fl
->buckets
[new_flow
] > (FLOW_LIMIT_HISTORY
>> 1)) {
3231 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3232 * queue (may be a remote CPU queue).
3234 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
3235 unsigned int *qtail
)
3237 struct softnet_data
*sd
;
3238 unsigned long flags
;
3241 sd
= &per_cpu(softnet_data
, cpu
);
3243 local_irq_save(flags
);
3246 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3247 if (qlen
<= netdev_max_backlog
&& !skb_flow_limit(skb
, qlen
)) {
3248 if (skb_queue_len(&sd
->input_pkt_queue
)) {
3250 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
3251 input_queue_tail_incr_save(sd
, qtail
);
3253 local_irq_restore(flags
);
3254 return NET_RX_SUCCESS
;
3257 /* Schedule NAPI for backlog device
3258 * We can use non atomic operation since we own the queue lock
3260 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
3261 if (!rps_ipi_queued(sd
))
3262 ____napi_schedule(sd
, &sd
->backlog
);
3270 local_irq_restore(flags
);
3272 atomic_long_inc(&skb
->dev
->rx_dropped
);
3277 static int netif_rx_internal(struct sk_buff
*skb
)
3281 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3283 trace_netif_rx(skb
);
3285 if (static_key_false(&rps_needed
)) {
3286 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3292 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3294 cpu
= smp_processor_id();
3296 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3304 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
3311 * netif_rx - post buffer to the network code
3312 * @skb: buffer to post
3314 * This function receives a packet from a device driver and queues it for
3315 * the upper (protocol) levels to process. It always succeeds. The buffer
3316 * may be dropped during processing for congestion control or by the
3320 * NET_RX_SUCCESS (no congestion)
3321 * NET_RX_DROP (packet was dropped)
3325 int netif_rx(struct sk_buff
*skb
)
3327 trace_netif_rx_entry(skb
);
3329 return netif_rx_internal(skb
);
3331 EXPORT_SYMBOL(netif_rx
);
3333 int netif_rx_ni(struct sk_buff
*skb
)
3337 trace_netif_rx_ni_entry(skb
);
3340 err
= netif_rx_internal(skb
);
3341 if (local_softirq_pending())
3347 EXPORT_SYMBOL(netif_rx_ni
);
3349 static void net_tx_action(struct softirq_action
*h
)
3351 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3353 if (sd
->completion_queue
) {
3354 struct sk_buff
*clist
;
3356 local_irq_disable();
3357 clist
= sd
->completion_queue
;
3358 sd
->completion_queue
= NULL
;
3362 struct sk_buff
*skb
= clist
;
3363 clist
= clist
->next
;
3365 WARN_ON(atomic_read(&skb
->users
));
3366 if (likely(get_kfree_skb_cb(skb
)->reason
== SKB_REASON_CONSUMED
))
3367 trace_consume_skb(skb
);
3369 trace_kfree_skb(skb
, net_tx_action
);
3374 if (sd
->output_queue
) {
3377 local_irq_disable();
3378 head
= sd
->output_queue
;
3379 sd
->output_queue
= NULL
;
3380 sd
->output_queue_tailp
= &sd
->output_queue
;
3384 struct Qdisc
*q
= head
;
3385 spinlock_t
*root_lock
;
3387 head
= head
->next_sched
;
3389 root_lock
= qdisc_lock(q
);
3390 if (spin_trylock(root_lock
)) {
3391 smp_mb__before_atomic();
3392 clear_bit(__QDISC_STATE_SCHED
,
3395 spin_unlock(root_lock
);
3397 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
3399 __netif_reschedule(q
);
3401 smp_mb__before_atomic();
3402 clear_bit(__QDISC_STATE_SCHED
,
3410 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3411 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3412 /* This hook is defined here for ATM LANE */
3413 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
3414 unsigned char *addr
) __read_mostly
;
3415 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
3418 #ifdef CONFIG_NET_CLS_ACT
3419 /* TODO: Maybe we should just force sch_ingress to be compiled in
3420 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3421 * a compare and 2 stores extra right now if we dont have it on
3422 * but have CONFIG_NET_CLS_ACT
3423 * NOTE: This doesn't stop any functionality; if you dont have
3424 * the ingress scheduler, you just can't add policies on ingress.
3427 static int ing_filter(struct sk_buff
*skb
, struct netdev_queue
*rxq
)
3429 struct net_device
*dev
= skb
->dev
;
3430 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
3431 int result
= TC_ACT_OK
;
3434 if (unlikely(MAX_RED_LOOP
< ttl
++)) {
3435 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3436 skb
->skb_iif
, dev
->ifindex
);
3440 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
3441 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
3444 if (q
!= &noop_qdisc
) {
3445 spin_lock(qdisc_lock(q
));
3446 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
3447 result
= qdisc_enqueue_root(skb
, q
);
3448 spin_unlock(qdisc_lock(q
));
3454 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
3455 struct packet_type
**pt_prev
,
3456 int *ret
, struct net_device
*orig_dev
)
3458 struct netdev_queue
*rxq
= rcu_dereference(skb
->dev
->ingress_queue
);
3460 if (!rxq
|| rxq
->qdisc
== &noop_qdisc
)
3464 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
3468 switch (ing_filter(skb
, rxq
)) {
3482 * netdev_rx_handler_register - register receive handler
3483 * @dev: device to register a handler for
3484 * @rx_handler: receive handler to register
3485 * @rx_handler_data: data pointer that is used by rx handler
3487 * Register a receive handler for a device. This handler will then be
3488 * called from __netif_receive_skb. A negative errno code is returned
3491 * The caller must hold the rtnl_mutex.
3493 * For a general description of rx_handler, see enum rx_handler_result.
3495 int netdev_rx_handler_register(struct net_device
*dev
,
3496 rx_handler_func_t
*rx_handler
,
3497 void *rx_handler_data
)
3501 if (dev
->rx_handler
)
3504 /* Note: rx_handler_data must be set before rx_handler */
3505 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
3506 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
3510 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
3513 * netdev_rx_handler_unregister - unregister receive handler
3514 * @dev: device to unregister a handler from
3516 * Unregister a receive handler from a device.
3518 * The caller must hold the rtnl_mutex.
3520 void netdev_rx_handler_unregister(struct net_device
*dev
)
3524 RCU_INIT_POINTER(dev
->rx_handler
, NULL
);
3525 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3526 * section has a guarantee to see a non NULL rx_handler_data
3530 RCU_INIT_POINTER(dev
->rx_handler_data
, NULL
);
3532 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
3535 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3536 * the special handling of PFMEMALLOC skbs.
3538 static bool skb_pfmemalloc_protocol(struct sk_buff
*skb
)
3540 switch (skb
->protocol
) {
3541 case htons(ETH_P_ARP
):
3542 case htons(ETH_P_IP
):
3543 case htons(ETH_P_IPV6
):
3544 case htons(ETH_P_8021Q
):
3545 case htons(ETH_P_8021AD
):
3552 static int __netif_receive_skb_core(struct sk_buff
*skb
, bool pfmemalloc
)
3554 struct packet_type
*ptype
, *pt_prev
;
3555 rx_handler_func_t
*rx_handler
;
3556 struct net_device
*orig_dev
;
3557 struct net_device
*null_or_dev
;
3558 bool deliver_exact
= false;
3559 int ret
= NET_RX_DROP
;
3562 net_timestamp_check(!netdev_tstamp_prequeue
, skb
);
3564 trace_netif_receive_skb(skb
);
3566 orig_dev
= skb
->dev
;
3568 skb_reset_network_header(skb
);
3569 if (!skb_transport_header_was_set(skb
))
3570 skb_reset_transport_header(skb
);
3571 skb_reset_mac_len(skb
);
3578 skb
->skb_iif
= skb
->dev
->ifindex
;
3580 __this_cpu_inc(softnet_data
.processed
);
3582 if (skb
->protocol
== cpu_to_be16(ETH_P_8021Q
) ||
3583 skb
->protocol
== cpu_to_be16(ETH_P_8021AD
)) {
3584 skb
= vlan_untag(skb
);
3589 #ifdef CONFIG_NET_CLS_ACT
3590 if (skb
->tc_verd
& TC_NCLS
) {
3591 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
3599 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
3600 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
3602 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3608 #ifdef CONFIG_NET_CLS_ACT
3609 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
3615 if (pfmemalloc
&& !skb_pfmemalloc_protocol(skb
))
3618 if (vlan_tx_tag_present(skb
)) {
3620 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3623 if (vlan_do_receive(&skb
))
3625 else if (unlikely(!skb
))
3629 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
3632 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3635 switch (rx_handler(&skb
)) {
3636 case RX_HANDLER_CONSUMED
:
3637 ret
= NET_RX_SUCCESS
;
3639 case RX_HANDLER_ANOTHER
:
3641 case RX_HANDLER_EXACT
:
3642 deliver_exact
= true;
3643 case RX_HANDLER_PASS
:
3650 if (unlikely(vlan_tx_tag_present(skb
))) {
3651 if (vlan_tx_tag_get_id(skb
))
3652 skb
->pkt_type
= PACKET_OTHERHOST
;
3653 /* Note: we might in the future use prio bits
3654 * and set skb->priority like in vlan_do_receive()
3655 * For the time being, just ignore Priority Code Point
3660 /* deliver only exact match when indicated */
3661 null_or_dev
= deliver_exact
? skb
->dev
: NULL
;
3663 type
= skb
->protocol
;
3664 list_for_each_entry_rcu(ptype
,
3665 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
3666 if (ptype
->type
== type
&&
3667 (ptype
->dev
== null_or_dev
|| ptype
->dev
== skb
->dev
||
3668 ptype
->dev
== orig_dev
)) {
3670 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3676 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
3679 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
3682 atomic_long_inc(&skb
->dev
->rx_dropped
);
3684 /* Jamal, now you will not able to escape explaining
3685 * me how you were going to use this. :-)
3695 static int __netif_receive_skb(struct sk_buff
*skb
)
3699 if (sk_memalloc_socks() && skb_pfmemalloc(skb
)) {
3700 unsigned long pflags
= current
->flags
;
3703 * PFMEMALLOC skbs are special, they should
3704 * - be delivered to SOCK_MEMALLOC sockets only
3705 * - stay away from userspace
3706 * - have bounded memory usage
3708 * Use PF_MEMALLOC as this saves us from propagating the allocation
3709 * context down to all allocation sites.
3711 current
->flags
|= PF_MEMALLOC
;
3712 ret
= __netif_receive_skb_core(skb
, true);
3713 tsk_restore_flags(current
, pflags
, PF_MEMALLOC
);
3715 ret
= __netif_receive_skb_core(skb
, false);
3720 static int netif_receive_skb_internal(struct sk_buff
*skb
)
3722 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3724 if (skb_defer_rx_timestamp(skb
))
3725 return NET_RX_SUCCESS
;
3728 if (static_key_false(&rps_needed
)) {
3729 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3734 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3737 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3744 return __netif_receive_skb(skb
);
3748 * netif_receive_skb - process receive buffer from network
3749 * @skb: buffer to process
3751 * netif_receive_skb() is the main receive data processing function.
3752 * It always succeeds. The buffer may be dropped during processing
3753 * for congestion control or by the protocol layers.
3755 * This function may only be called from softirq context and interrupts
3756 * should be enabled.
3758 * Return values (usually ignored):
3759 * NET_RX_SUCCESS: no congestion
3760 * NET_RX_DROP: packet was dropped
3762 int netif_receive_skb(struct sk_buff
*skb
)
3764 trace_netif_receive_skb_entry(skb
);
3766 return netif_receive_skb_internal(skb
);
3768 EXPORT_SYMBOL(netif_receive_skb
);
3770 /* Network device is going away, flush any packets still pending
3771 * Called with irqs disabled.
3773 static void flush_backlog(void *arg
)
3775 struct net_device
*dev
= arg
;
3776 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3777 struct sk_buff
*skb
, *tmp
;
3780 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3781 if (skb
->dev
== dev
) {
3782 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3784 input_queue_head_incr(sd
);
3789 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3790 if (skb
->dev
== dev
) {
3791 __skb_unlink(skb
, &sd
->process_queue
);
3793 input_queue_head_incr(sd
);
3798 static int napi_gro_complete(struct sk_buff
*skb
)
3800 struct packet_offload
*ptype
;
3801 __be16 type
= skb
->protocol
;
3802 struct list_head
*head
= &offload_base
;
3805 BUILD_BUG_ON(sizeof(struct napi_gro_cb
) > sizeof(skb
->cb
));
3807 if (NAPI_GRO_CB(skb
)->count
== 1) {
3808 skb_shinfo(skb
)->gso_size
= 0;
3813 list_for_each_entry_rcu(ptype
, head
, list
) {
3814 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_complete
)
3817 err
= ptype
->callbacks
.gro_complete(skb
, 0);
3823 WARN_ON(&ptype
->list
== head
);
3825 return NET_RX_SUCCESS
;
3829 return netif_receive_skb_internal(skb
);
3832 /* napi->gro_list contains packets ordered by age.
3833 * youngest packets at the head of it.
3834 * Complete skbs in reverse order to reduce latencies.
3836 void napi_gro_flush(struct napi_struct
*napi
, bool flush_old
)
3838 struct sk_buff
*skb
, *prev
= NULL
;
3840 /* scan list and build reverse chain */
3841 for (skb
= napi
->gro_list
; skb
!= NULL
; skb
= skb
->next
) {
3846 for (skb
= prev
; skb
; skb
= prev
) {
3849 if (flush_old
&& NAPI_GRO_CB(skb
)->age
== jiffies
)
3853 napi_gro_complete(skb
);
3857 napi
->gro_list
= NULL
;
3859 EXPORT_SYMBOL(napi_gro_flush
);
3861 static void gro_list_prepare(struct napi_struct
*napi
, struct sk_buff
*skb
)
3864 unsigned int maclen
= skb
->dev
->hard_header_len
;
3865 u32 hash
= skb_get_hash_raw(skb
);
3867 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3868 unsigned long diffs
;
3870 NAPI_GRO_CB(p
)->flush
= 0;
3872 if (hash
!= skb_get_hash_raw(p
)) {
3873 NAPI_GRO_CB(p
)->same_flow
= 0;
3877 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
3878 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
3879 if (maclen
== ETH_HLEN
)
3880 diffs
|= compare_ether_header(skb_mac_header(p
),
3881 skb_mac_header(skb
));
3883 diffs
= memcmp(skb_mac_header(p
),
3884 skb_mac_header(skb
),
3886 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
3890 static void skb_gro_reset_offset(struct sk_buff
*skb
)
3892 const struct skb_shared_info
*pinfo
= skb_shinfo(skb
);
3893 const skb_frag_t
*frag0
= &pinfo
->frags
[0];
3895 NAPI_GRO_CB(skb
)->data_offset
= 0;
3896 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3897 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3899 if (skb_mac_header(skb
) == skb_tail_pointer(skb
) &&
3901 !PageHighMem(skb_frag_page(frag0
))) {
3902 NAPI_GRO_CB(skb
)->frag0
= skb_frag_address(frag0
);
3903 NAPI_GRO_CB(skb
)->frag0_len
= skb_frag_size(frag0
);
3907 static void gro_pull_from_frag0(struct sk_buff
*skb
, int grow
)
3909 struct skb_shared_info
*pinfo
= skb_shinfo(skb
);
3911 BUG_ON(skb
->end
- skb
->tail
< grow
);
3913 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3915 skb
->data_len
-= grow
;
3918 pinfo
->frags
[0].page_offset
+= grow
;
3919 skb_frag_size_sub(&pinfo
->frags
[0], grow
);
3921 if (unlikely(!skb_frag_size(&pinfo
->frags
[0]))) {
3922 skb_frag_unref(skb
, 0);
3923 memmove(pinfo
->frags
, pinfo
->frags
+ 1,
3924 --pinfo
->nr_frags
* sizeof(pinfo
->frags
[0]));
3928 static enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3930 struct sk_buff
**pp
= NULL
;
3931 struct packet_offload
*ptype
;
3932 __be16 type
= skb
->protocol
;
3933 struct list_head
*head
= &offload_base
;
3935 enum gro_result ret
;
3938 if (!(skb
->dev
->features
& NETIF_F_GRO
))
3941 if (skb_is_gso(skb
) || skb_has_frag_list(skb
))
3944 gro_list_prepare(napi
, skb
);
3945 NAPI_GRO_CB(skb
)->csum
= skb
->csum
; /* Needed for CHECKSUM_COMPLETE */
3948 list_for_each_entry_rcu(ptype
, head
, list
) {
3949 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_receive
)
3952 skb_set_network_header(skb
, skb_gro_offset(skb
));
3953 skb_reset_mac_len(skb
);
3954 NAPI_GRO_CB(skb
)->same_flow
= 0;
3955 NAPI_GRO_CB(skb
)->flush
= 0;
3956 NAPI_GRO_CB(skb
)->free
= 0;
3957 NAPI_GRO_CB(skb
)->udp_mark
= 0;
3959 pp
= ptype
->callbacks
.gro_receive(&napi
->gro_list
, skb
);
3964 if (&ptype
->list
== head
)
3967 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3968 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3971 struct sk_buff
*nskb
= *pp
;
3975 napi_gro_complete(nskb
);
3982 if (NAPI_GRO_CB(skb
)->flush
)
3985 if (unlikely(napi
->gro_count
>= MAX_GRO_SKBS
)) {
3986 struct sk_buff
*nskb
= napi
->gro_list
;
3988 /* locate the end of the list to select the 'oldest' flow */
3989 while (nskb
->next
) {
3995 napi_gro_complete(nskb
);
3999 NAPI_GRO_CB(skb
)->count
= 1;
4000 NAPI_GRO_CB(skb
)->age
= jiffies
;
4001 NAPI_GRO_CB(skb
)->last
= skb
;
4002 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
4003 skb
->next
= napi
->gro_list
;
4004 napi
->gro_list
= skb
;
4008 grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
4010 gro_pull_from_frag0(skb
, grow
);
4019 struct packet_offload
*gro_find_receive_by_type(__be16 type
)
4021 struct list_head
*offload_head
= &offload_base
;
4022 struct packet_offload
*ptype
;
4024 list_for_each_entry_rcu(ptype
, offload_head
, list
) {
4025 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_receive
)
4031 EXPORT_SYMBOL(gro_find_receive_by_type
);
4033 struct packet_offload
*gro_find_complete_by_type(__be16 type
)
4035 struct list_head
*offload_head
= &offload_base
;
4036 struct packet_offload
*ptype
;
4038 list_for_each_entry_rcu(ptype
, offload_head
, list
) {
4039 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_complete
)
4045 EXPORT_SYMBOL(gro_find_complete_by_type
);
4047 static gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
4051 if (netif_receive_skb_internal(skb
))
4059 case GRO_MERGED_FREE
:
4060 if (NAPI_GRO_CB(skb
)->free
== NAPI_GRO_FREE_STOLEN_HEAD
)
4061 kmem_cache_free(skbuff_head_cache
, skb
);
4074 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
4076 trace_napi_gro_receive_entry(skb
);
4078 skb_gro_reset_offset(skb
);
4080 return napi_skb_finish(dev_gro_receive(napi
, skb
), skb
);
4082 EXPORT_SYMBOL(napi_gro_receive
);
4084 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
4086 __skb_pull(skb
, skb_headlen(skb
));
4087 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4088 skb_reserve(skb
, NET_SKB_PAD
+ NET_IP_ALIGN
- skb_headroom(skb
));
4090 skb
->dev
= napi
->dev
;
4092 skb
->truesize
= SKB_TRUESIZE(skb_end_offset(skb
));
4097 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
4099 struct sk_buff
*skb
= napi
->skb
;
4102 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
4107 EXPORT_SYMBOL(napi_get_frags
);
4109 static gro_result_t
napi_frags_finish(struct napi_struct
*napi
,
4110 struct sk_buff
*skb
,
4116 __skb_push(skb
, ETH_HLEN
);
4117 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
4118 if (ret
== GRO_NORMAL
&& netif_receive_skb_internal(skb
))
4123 case GRO_MERGED_FREE
:
4124 napi_reuse_skb(napi
, skb
);
4134 /* Upper GRO stack assumes network header starts at gro_offset=0
4135 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4136 * We copy ethernet header into skb->data to have a common layout.
4138 static struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
4140 struct sk_buff
*skb
= napi
->skb
;
4141 const struct ethhdr
*eth
;
4142 unsigned int hlen
= sizeof(*eth
);
4146 skb_reset_mac_header(skb
);
4147 skb_gro_reset_offset(skb
);
4149 eth
= skb_gro_header_fast(skb
, 0);
4150 if (unlikely(skb_gro_header_hard(skb
, hlen
))) {
4151 eth
= skb_gro_header_slow(skb
, hlen
, 0);
4152 if (unlikely(!eth
)) {
4153 napi_reuse_skb(napi
, skb
);
4157 gro_pull_from_frag0(skb
, hlen
);
4158 NAPI_GRO_CB(skb
)->frag0
+= hlen
;
4159 NAPI_GRO_CB(skb
)->frag0_len
-= hlen
;
4161 __skb_pull(skb
, hlen
);
4164 * This works because the only protocols we care about don't require
4166 * We'll fix it up properly in napi_frags_finish()
4168 skb
->protocol
= eth
->h_proto
;
4173 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
4175 struct sk_buff
*skb
= napi_frags_skb(napi
);
4180 trace_napi_gro_frags_entry(skb
);
4182 return napi_frags_finish(napi
, skb
, dev_gro_receive(napi
, skb
));
4184 EXPORT_SYMBOL(napi_gro_frags
);
4187 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4188 * Note: called with local irq disabled, but exits with local irq enabled.
4190 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
4193 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
4196 sd
->rps_ipi_list
= NULL
;
4200 /* Send pending IPI's to kick RPS processing on remote cpus. */
4202 struct softnet_data
*next
= remsd
->rps_ipi_next
;
4204 if (cpu_online(remsd
->cpu
))
4205 smp_call_function_single_async(remsd
->cpu
,
4214 static int process_backlog(struct napi_struct
*napi
, int quota
)
4217 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
4220 /* Check if we have pending ipi, its better to send them now,
4221 * not waiting net_rx_action() end.
4223 if (sd
->rps_ipi_list
) {
4224 local_irq_disable();
4225 net_rps_action_and_irq_enable(sd
);
4228 napi
->weight
= weight_p
;
4229 local_irq_disable();
4231 struct sk_buff
*skb
;
4233 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
4235 __netif_receive_skb(skb
);
4236 local_irq_disable();
4237 input_queue_head_incr(sd
);
4238 if (++work
>= quota
) {
4245 if (skb_queue_empty(&sd
->input_pkt_queue
)) {
4247 * Inline a custom version of __napi_complete().
4248 * only current cpu owns and manipulates this napi,
4249 * and NAPI_STATE_SCHED is the only possible flag set
4251 * We can use a plain write instead of clear_bit(),
4252 * and we dont need an smp_mb() memory barrier.
4254 list_del(&napi
->poll_list
);
4261 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
4262 &sd
->process_queue
);
4271 * __napi_schedule - schedule for receive
4272 * @n: entry to schedule
4274 * The entry's receive function will be scheduled to run
4276 void __napi_schedule(struct napi_struct
*n
)
4278 unsigned long flags
;
4280 local_irq_save(flags
);
4281 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
4282 local_irq_restore(flags
);
4284 EXPORT_SYMBOL(__napi_schedule
);
4286 void __napi_complete(struct napi_struct
*n
)
4288 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
4289 BUG_ON(n
->gro_list
);
4291 list_del(&n
->poll_list
);
4292 smp_mb__before_atomic();
4293 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
4295 EXPORT_SYMBOL(__napi_complete
);
4297 void napi_complete(struct napi_struct
*n
)
4299 unsigned long flags
;
4302 * don't let napi dequeue from the cpu poll list
4303 * just in case its running on a different cpu
4305 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
4308 napi_gro_flush(n
, false);
4309 local_irq_save(flags
);
4311 local_irq_restore(flags
);
4313 EXPORT_SYMBOL(napi_complete
);
4315 /* must be called under rcu_read_lock(), as we dont take a reference */
4316 struct napi_struct
*napi_by_id(unsigned int napi_id
)
4318 unsigned int hash
= napi_id
% HASH_SIZE(napi_hash
);
4319 struct napi_struct
*napi
;
4321 hlist_for_each_entry_rcu(napi
, &napi_hash
[hash
], napi_hash_node
)
4322 if (napi
->napi_id
== napi_id
)
4327 EXPORT_SYMBOL_GPL(napi_by_id
);
4329 void napi_hash_add(struct napi_struct
*napi
)
4331 if (!test_and_set_bit(NAPI_STATE_HASHED
, &napi
->state
)) {
4333 spin_lock(&napi_hash_lock
);
4335 /* 0 is not a valid id, we also skip an id that is taken
4336 * we expect both events to be extremely rare
4339 while (!napi
->napi_id
) {
4340 napi
->napi_id
= ++napi_gen_id
;
4341 if (napi_by_id(napi
->napi_id
))
4345 hlist_add_head_rcu(&napi
->napi_hash_node
,
4346 &napi_hash
[napi
->napi_id
% HASH_SIZE(napi_hash
)]);
4348 spin_unlock(&napi_hash_lock
);
4351 EXPORT_SYMBOL_GPL(napi_hash_add
);
4353 /* Warning : caller is responsible to make sure rcu grace period
4354 * is respected before freeing memory containing @napi
4356 void napi_hash_del(struct napi_struct
*napi
)
4358 spin_lock(&napi_hash_lock
);
4360 if (test_and_clear_bit(NAPI_STATE_HASHED
, &napi
->state
))
4361 hlist_del_rcu(&napi
->napi_hash_node
);
4363 spin_unlock(&napi_hash_lock
);
4365 EXPORT_SYMBOL_GPL(napi_hash_del
);
4367 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
4368 int (*poll
)(struct napi_struct
*, int), int weight
)
4370 INIT_LIST_HEAD(&napi
->poll_list
);
4371 napi
->gro_count
= 0;
4372 napi
->gro_list
= NULL
;
4375 if (weight
> NAPI_POLL_WEIGHT
)
4376 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4378 napi
->weight
= weight
;
4379 list_add(&napi
->dev_list
, &dev
->napi_list
);
4381 #ifdef CONFIG_NETPOLL
4382 spin_lock_init(&napi
->poll_lock
);
4383 napi
->poll_owner
= -1;
4385 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
4387 EXPORT_SYMBOL(netif_napi_add
);
4389 void netif_napi_del(struct napi_struct
*napi
)
4391 list_del_init(&napi
->dev_list
);
4392 napi_free_frags(napi
);
4394 kfree_skb_list(napi
->gro_list
);
4395 napi
->gro_list
= NULL
;
4396 napi
->gro_count
= 0;
4398 EXPORT_SYMBOL(netif_napi_del
);
4400 static void net_rx_action(struct softirq_action
*h
)
4402 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
4403 unsigned long time_limit
= jiffies
+ 2;
4404 int budget
= netdev_budget
;
4407 local_irq_disable();
4409 while (!list_empty(&sd
->poll_list
)) {
4410 struct napi_struct
*n
;
4413 /* If softirq window is exhuasted then punt.
4414 * Allow this to run for 2 jiffies since which will allow
4415 * an average latency of 1.5/HZ.
4417 if (unlikely(budget
<= 0 || time_after_eq(jiffies
, time_limit
)))
4422 /* Even though interrupts have been re-enabled, this
4423 * access is safe because interrupts can only add new
4424 * entries to the tail of this list, and only ->poll()
4425 * calls can remove this head entry from the list.
4427 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
4429 have
= netpoll_poll_lock(n
);
4433 /* This NAPI_STATE_SCHED test is for avoiding a race
4434 * with netpoll's poll_napi(). Only the entity which
4435 * obtains the lock and sees NAPI_STATE_SCHED set will
4436 * actually make the ->poll() call. Therefore we avoid
4437 * accidentally calling ->poll() when NAPI is not scheduled.
4440 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
4441 work
= n
->poll(n
, weight
);
4445 WARN_ON_ONCE(work
> weight
);
4449 local_irq_disable();
4451 /* Drivers must not modify the NAPI state if they
4452 * consume the entire weight. In such cases this code
4453 * still "owns" the NAPI instance and therefore can
4454 * move the instance around on the list at-will.
4456 if (unlikely(work
== weight
)) {
4457 if (unlikely(napi_disable_pending(n
))) {
4460 local_irq_disable();
4463 /* flush too old packets
4464 * If HZ < 1000, flush all packets.
4467 napi_gro_flush(n
, HZ
>= 1000);
4468 local_irq_disable();
4470 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
4474 netpoll_poll_unlock(have
);
4477 net_rps_action_and_irq_enable(sd
);
4479 #ifdef CONFIG_NET_DMA
4481 * There may not be any more sk_buffs coming right now, so push
4482 * any pending DMA copies to hardware
4484 dma_issue_pending_all();
4491 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
4495 struct netdev_adjacent
{
4496 struct net_device
*dev
;
4498 /* upper master flag, there can only be one master device per list */
4501 /* counter for the number of times this device was added to us */
4504 /* private field for the users */
4507 struct list_head list
;
4508 struct rcu_head rcu
;
4511 static struct netdev_adjacent
*__netdev_find_adj(struct net_device
*dev
,
4512 struct net_device
*adj_dev
,
4513 struct list_head
*adj_list
)
4515 struct netdev_adjacent
*adj
;
4517 list_for_each_entry(adj
, adj_list
, list
) {
4518 if (adj
->dev
== adj_dev
)
4525 * netdev_has_upper_dev - Check if device is linked to an upper device
4527 * @upper_dev: upper device to check
4529 * Find out if a device is linked to specified upper device and return true
4530 * in case it is. Note that this checks only immediate upper device,
4531 * not through a complete stack of devices. The caller must hold the RTNL lock.
4533 bool netdev_has_upper_dev(struct net_device
*dev
,
4534 struct net_device
*upper_dev
)
4538 return __netdev_find_adj(dev
, upper_dev
, &dev
->all_adj_list
.upper
);
4540 EXPORT_SYMBOL(netdev_has_upper_dev
);
4543 * netdev_has_any_upper_dev - Check if device is linked to some device
4546 * Find out if a device is linked to an upper device and return true in case
4547 * it is. The caller must hold the RTNL lock.
4549 static bool netdev_has_any_upper_dev(struct net_device
*dev
)
4553 return !list_empty(&dev
->all_adj_list
.upper
);
4557 * netdev_master_upper_dev_get - Get master upper device
4560 * Find a master upper device and return pointer to it or NULL in case
4561 * it's not there. The caller must hold the RTNL lock.
4563 struct net_device
*netdev_master_upper_dev_get(struct net_device
*dev
)
4565 struct netdev_adjacent
*upper
;
4569 if (list_empty(&dev
->adj_list
.upper
))
4572 upper
= list_first_entry(&dev
->adj_list
.upper
,
4573 struct netdev_adjacent
, list
);
4574 if (likely(upper
->master
))
4578 EXPORT_SYMBOL(netdev_master_upper_dev_get
);
4580 void *netdev_adjacent_get_private(struct list_head
*adj_list
)
4582 struct netdev_adjacent
*adj
;
4584 adj
= list_entry(adj_list
, struct netdev_adjacent
, list
);
4586 return adj
->private;
4588 EXPORT_SYMBOL(netdev_adjacent_get_private
);
4591 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
4593 * @iter: list_head ** of the current position
4595 * Gets the next device from the dev's upper list, starting from iter
4596 * position. The caller must hold RCU read lock.
4598 struct net_device
*netdev_upper_get_next_dev_rcu(struct net_device
*dev
,
4599 struct list_head
**iter
)
4601 struct netdev_adjacent
*upper
;
4603 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4605 upper
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
4607 if (&upper
->list
== &dev
->adj_list
.upper
)
4610 *iter
= &upper
->list
;
4614 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu
);
4617 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4619 * @iter: list_head ** of the current position
4621 * Gets the next device from the dev's upper list, starting from iter
4622 * position. The caller must hold RCU read lock.
4624 struct net_device
*netdev_all_upper_get_next_dev_rcu(struct net_device
*dev
,
4625 struct list_head
**iter
)
4627 struct netdev_adjacent
*upper
;
4629 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4631 upper
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
4633 if (&upper
->list
== &dev
->all_adj_list
.upper
)
4636 *iter
= &upper
->list
;
4640 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu
);
4643 * netdev_lower_get_next_private - Get the next ->private from the
4644 * lower neighbour list
4646 * @iter: list_head ** of the current position
4648 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4649 * list, starting from iter position. The caller must hold either hold the
4650 * RTNL lock or its own locking that guarantees that the neighbour lower
4651 * list will remain unchainged.
4653 void *netdev_lower_get_next_private(struct net_device
*dev
,
4654 struct list_head
**iter
)
4656 struct netdev_adjacent
*lower
;
4658 lower
= list_entry(*iter
, struct netdev_adjacent
, list
);
4660 if (&lower
->list
== &dev
->adj_list
.lower
)
4663 *iter
= lower
->list
.next
;
4665 return lower
->private;
4667 EXPORT_SYMBOL(netdev_lower_get_next_private
);
4670 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4671 * lower neighbour list, RCU
4674 * @iter: list_head ** of the current position
4676 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4677 * list, starting from iter position. The caller must hold RCU read lock.
4679 void *netdev_lower_get_next_private_rcu(struct net_device
*dev
,
4680 struct list_head
**iter
)
4682 struct netdev_adjacent
*lower
;
4684 WARN_ON_ONCE(!rcu_read_lock_held());
4686 lower
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
4688 if (&lower
->list
== &dev
->adj_list
.lower
)
4691 *iter
= &lower
->list
;
4693 return lower
->private;
4695 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu
);
4698 * netdev_lower_get_next - Get the next device from the lower neighbour
4701 * @iter: list_head ** of the current position
4703 * Gets the next netdev_adjacent from the dev's lower neighbour
4704 * list, starting from iter position. The caller must hold RTNL lock or
4705 * its own locking that guarantees that the neighbour lower
4706 * list will remain unchainged.
4708 void *netdev_lower_get_next(struct net_device
*dev
, struct list_head
**iter
)
4710 struct netdev_adjacent
*lower
;
4712 lower
= list_entry((*iter
)->next
, struct netdev_adjacent
, list
);
4714 if (&lower
->list
== &dev
->adj_list
.lower
)
4717 *iter
= &lower
->list
;
4721 EXPORT_SYMBOL(netdev_lower_get_next
);
4724 * netdev_lower_get_first_private_rcu - Get the first ->private from the
4725 * lower neighbour list, RCU
4729 * Gets the first netdev_adjacent->private from the dev's lower neighbour
4730 * list. The caller must hold RCU read lock.
4732 void *netdev_lower_get_first_private_rcu(struct net_device
*dev
)
4734 struct netdev_adjacent
*lower
;
4736 lower
= list_first_or_null_rcu(&dev
->adj_list
.lower
,
4737 struct netdev_adjacent
, list
);
4739 return lower
->private;
4742 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu
);
4745 * netdev_master_upper_dev_get_rcu - Get master upper device
4748 * Find a master upper device and return pointer to it or NULL in case
4749 * it's not there. The caller must hold the RCU read lock.
4751 struct net_device
*netdev_master_upper_dev_get_rcu(struct net_device
*dev
)
4753 struct netdev_adjacent
*upper
;
4755 upper
= list_first_or_null_rcu(&dev
->adj_list
.upper
,
4756 struct netdev_adjacent
, list
);
4757 if (upper
&& likely(upper
->master
))
4761 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu
);
4763 static int netdev_adjacent_sysfs_add(struct net_device
*dev
,
4764 struct net_device
*adj_dev
,
4765 struct list_head
*dev_list
)
4767 char linkname
[IFNAMSIZ
+7];
4768 sprintf(linkname
, dev_list
== &dev
->adj_list
.upper
?
4769 "upper_%s" : "lower_%s", adj_dev
->name
);
4770 return sysfs_create_link(&(dev
->dev
.kobj
), &(adj_dev
->dev
.kobj
),
4773 static void netdev_adjacent_sysfs_del(struct net_device
*dev
,
4775 struct list_head
*dev_list
)
4777 char linkname
[IFNAMSIZ
+7];
4778 sprintf(linkname
, dev_list
== &dev
->adj_list
.upper
?
4779 "upper_%s" : "lower_%s", name
);
4780 sysfs_remove_link(&(dev
->dev
.kobj
), linkname
);
4783 #define netdev_adjacent_is_neigh_list(dev, dev_list) \
4784 (dev_list == &dev->adj_list.upper || \
4785 dev_list == &dev->adj_list.lower)
4787 static int __netdev_adjacent_dev_insert(struct net_device
*dev
,
4788 struct net_device
*adj_dev
,
4789 struct list_head
*dev_list
,
4790 void *private, bool master
)
4792 struct netdev_adjacent
*adj
;
4795 adj
= __netdev_find_adj(dev
, adj_dev
, dev_list
);
4802 adj
= kmalloc(sizeof(*adj
), GFP_KERNEL
);
4807 adj
->master
= master
;
4809 adj
->private = private;
4812 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
4813 adj_dev
->name
, dev
->name
, adj_dev
->name
);
4815 if (netdev_adjacent_is_neigh_list(dev
, dev_list
)) {
4816 ret
= netdev_adjacent_sysfs_add(dev
, adj_dev
, dev_list
);
4821 /* Ensure that master link is always the first item in list. */
4823 ret
= sysfs_create_link(&(dev
->dev
.kobj
),
4824 &(adj_dev
->dev
.kobj
), "master");
4826 goto remove_symlinks
;
4828 list_add_rcu(&adj
->list
, dev_list
);
4830 list_add_tail_rcu(&adj
->list
, dev_list
);
4836 if (netdev_adjacent_is_neigh_list(dev
, dev_list
))
4837 netdev_adjacent_sysfs_del(dev
, adj_dev
->name
, dev_list
);
4845 static void __netdev_adjacent_dev_remove(struct net_device
*dev
,
4846 struct net_device
*adj_dev
,
4847 struct list_head
*dev_list
)
4849 struct netdev_adjacent
*adj
;
4851 adj
= __netdev_find_adj(dev
, adj_dev
, dev_list
);
4854 pr_err("tried to remove device %s from %s\n",
4855 dev
->name
, adj_dev
->name
);
4859 if (adj
->ref_nr
> 1) {
4860 pr_debug("%s to %s ref_nr-- = %d\n", dev
->name
, adj_dev
->name
,
4867 sysfs_remove_link(&(dev
->dev
.kobj
), "master");
4869 if (netdev_adjacent_is_neigh_list(dev
, dev_list
))
4870 netdev_adjacent_sysfs_del(dev
, adj_dev
->name
, dev_list
);
4872 list_del_rcu(&adj
->list
);
4873 pr_debug("dev_put for %s, because link removed from %s to %s\n",
4874 adj_dev
->name
, dev
->name
, adj_dev
->name
);
4876 kfree_rcu(adj
, rcu
);
4879 static int __netdev_adjacent_dev_link_lists(struct net_device
*dev
,
4880 struct net_device
*upper_dev
,
4881 struct list_head
*up_list
,
4882 struct list_head
*down_list
,
4883 void *private, bool master
)
4887 ret
= __netdev_adjacent_dev_insert(dev
, upper_dev
, up_list
, private,
4892 ret
= __netdev_adjacent_dev_insert(upper_dev
, dev
, down_list
, private,
4895 __netdev_adjacent_dev_remove(dev
, upper_dev
, up_list
);
4902 static int __netdev_adjacent_dev_link(struct net_device
*dev
,
4903 struct net_device
*upper_dev
)
4905 return __netdev_adjacent_dev_link_lists(dev
, upper_dev
,
4906 &dev
->all_adj_list
.upper
,
4907 &upper_dev
->all_adj_list
.lower
,
4911 static void __netdev_adjacent_dev_unlink_lists(struct net_device
*dev
,
4912 struct net_device
*upper_dev
,
4913 struct list_head
*up_list
,
4914 struct list_head
*down_list
)
4916 __netdev_adjacent_dev_remove(dev
, upper_dev
, up_list
);
4917 __netdev_adjacent_dev_remove(upper_dev
, dev
, down_list
);
4920 static void __netdev_adjacent_dev_unlink(struct net_device
*dev
,
4921 struct net_device
*upper_dev
)
4923 __netdev_adjacent_dev_unlink_lists(dev
, upper_dev
,
4924 &dev
->all_adj_list
.upper
,
4925 &upper_dev
->all_adj_list
.lower
);
4928 static int __netdev_adjacent_dev_link_neighbour(struct net_device
*dev
,
4929 struct net_device
*upper_dev
,
4930 void *private, bool master
)
4932 int ret
= __netdev_adjacent_dev_link(dev
, upper_dev
);
4937 ret
= __netdev_adjacent_dev_link_lists(dev
, upper_dev
,
4938 &dev
->adj_list
.upper
,
4939 &upper_dev
->adj_list
.lower
,
4942 __netdev_adjacent_dev_unlink(dev
, upper_dev
);
4949 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device
*dev
,
4950 struct net_device
*upper_dev
)
4952 __netdev_adjacent_dev_unlink(dev
, upper_dev
);
4953 __netdev_adjacent_dev_unlink_lists(dev
, upper_dev
,
4954 &dev
->adj_list
.upper
,
4955 &upper_dev
->adj_list
.lower
);
4958 static int __netdev_upper_dev_link(struct net_device
*dev
,
4959 struct net_device
*upper_dev
, bool master
,
4962 struct netdev_adjacent
*i
, *j
, *to_i
, *to_j
;
4967 if (dev
== upper_dev
)
4970 /* To prevent loops, check if dev is not upper device to upper_dev. */
4971 if (__netdev_find_adj(upper_dev
, dev
, &upper_dev
->all_adj_list
.upper
))
4974 if (__netdev_find_adj(dev
, upper_dev
, &dev
->all_adj_list
.upper
))
4977 if (master
&& netdev_master_upper_dev_get(dev
))
4980 ret
= __netdev_adjacent_dev_link_neighbour(dev
, upper_dev
, private,
4985 /* Now that we linked these devs, make all the upper_dev's
4986 * all_adj_list.upper visible to every dev's all_adj_list.lower an
4987 * versa, and don't forget the devices itself. All of these
4988 * links are non-neighbours.
4990 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
4991 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
) {
4992 pr_debug("Interlinking %s with %s, non-neighbour\n",
4993 i
->dev
->name
, j
->dev
->name
);
4994 ret
= __netdev_adjacent_dev_link(i
->dev
, j
->dev
);
5000 /* add dev to every upper_dev's upper device */
5001 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
) {
5002 pr_debug("linking %s's upper device %s with %s\n",
5003 upper_dev
->name
, i
->dev
->name
, dev
->name
);
5004 ret
= __netdev_adjacent_dev_link(dev
, i
->dev
);
5006 goto rollback_upper_mesh
;
5009 /* add upper_dev to every dev's lower device */
5010 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5011 pr_debug("linking %s's lower device %s with %s\n", dev
->name
,
5012 i
->dev
->name
, upper_dev
->name
);
5013 ret
= __netdev_adjacent_dev_link(i
->dev
, upper_dev
);
5015 goto rollback_lower_mesh
;
5018 call_netdevice_notifiers(NETDEV_CHANGEUPPER
, dev
);
5021 rollback_lower_mesh
:
5023 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5026 __netdev_adjacent_dev_unlink(i
->dev
, upper_dev
);
5031 rollback_upper_mesh
:
5033 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
) {
5036 __netdev_adjacent_dev_unlink(dev
, i
->dev
);
5044 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
) {
5045 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
) {
5046 if (i
== to_i
&& j
== to_j
)
5048 __netdev_adjacent_dev_unlink(i
->dev
, j
->dev
);
5054 __netdev_adjacent_dev_unlink_neighbour(dev
, upper_dev
);
5060 * netdev_upper_dev_link - Add a link to the upper device
5062 * @upper_dev: new upper device
5064 * Adds a link to device which is upper to this one. The caller must hold
5065 * the RTNL lock. On a failure a negative errno code is returned.
5066 * On success the reference counts are adjusted and the function
5069 int netdev_upper_dev_link(struct net_device
*dev
,
5070 struct net_device
*upper_dev
)
5072 return __netdev_upper_dev_link(dev
, upper_dev
, false, NULL
);
5074 EXPORT_SYMBOL(netdev_upper_dev_link
);
5077 * netdev_master_upper_dev_link - Add a master link to the upper device
5079 * @upper_dev: new upper device
5081 * Adds a link to device which is upper to this one. In this case, only
5082 * one master upper device can be linked, although other non-master devices
5083 * might be linked as well. The caller must hold the RTNL lock.
5084 * On a failure a negative errno code is returned. On success the reference
5085 * counts are adjusted and the function returns zero.
5087 int netdev_master_upper_dev_link(struct net_device
*dev
,
5088 struct net_device
*upper_dev
)
5090 return __netdev_upper_dev_link(dev
, upper_dev
, true, NULL
);
5092 EXPORT_SYMBOL(netdev_master_upper_dev_link
);
5094 int netdev_master_upper_dev_link_private(struct net_device
*dev
,
5095 struct net_device
*upper_dev
,
5098 return __netdev_upper_dev_link(dev
, upper_dev
, true, private);
5100 EXPORT_SYMBOL(netdev_master_upper_dev_link_private
);
5103 * netdev_upper_dev_unlink - Removes a link to upper device
5105 * @upper_dev: new upper device
5107 * Removes a link to device which is upper to this one. The caller must hold
5110 void netdev_upper_dev_unlink(struct net_device
*dev
,
5111 struct net_device
*upper_dev
)
5113 struct netdev_adjacent
*i
, *j
;
5116 __netdev_adjacent_dev_unlink_neighbour(dev
, upper_dev
);
5118 /* Here is the tricky part. We must remove all dev's lower
5119 * devices from all upper_dev's upper devices and vice
5120 * versa, to maintain the graph relationship.
5122 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
)
5123 list_for_each_entry(j
, &upper_dev
->all_adj_list
.upper
, list
)
5124 __netdev_adjacent_dev_unlink(i
->dev
, j
->dev
);
5126 /* remove also the devices itself from lower/upper device
5129 list_for_each_entry(i
, &dev
->all_adj_list
.lower
, list
)
5130 __netdev_adjacent_dev_unlink(i
->dev
, upper_dev
);
5132 list_for_each_entry(i
, &upper_dev
->all_adj_list
.upper
, list
)
5133 __netdev_adjacent_dev_unlink(dev
, i
->dev
);
5135 call_netdevice_notifiers(NETDEV_CHANGEUPPER
, dev
);
5137 EXPORT_SYMBOL(netdev_upper_dev_unlink
);
5139 void netdev_adjacent_rename_links(struct net_device
*dev
, char *oldname
)
5141 struct netdev_adjacent
*iter
;
5143 list_for_each_entry(iter
, &dev
->adj_list
.upper
, list
) {
5144 netdev_adjacent_sysfs_del(iter
->dev
, oldname
,
5145 &iter
->dev
->adj_list
.lower
);
5146 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
5147 &iter
->dev
->adj_list
.lower
);
5150 list_for_each_entry(iter
, &dev
->adj_list
.lower
, list
) {
5151 netdev_adjacent_sysfs_del(iter
->dev
, oldname
,
5152 &iter
->dev
->adj_list
.upper
);
5153 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
5154 &iter
->dev
->adj_list
.upper
);
5158 void *netdev_lower_dev_get_private(struct net_device
*dev
,
5159 struct net_device
*lower_dev
)
5161 struct netdev_adjacent
*lower
;
5165 lower
= __netdev_find_adj(dev
, lower_dev
, &dev
->adj_list
.lower
);
5169 return lower
->private;
5171 EXPORT_SYMBOL(netdev_lower_dev_get_private
);
5174 int dev_get_nest_level(struct net_device
*dev
,
5175 bool (*type_check
)(struct net_device
*dev
))
5177 struct net_device
*lower
= NULL
;
5178 struct list_head
*iter
;
5184 netdev_for_each_lower_dev(dev
, lower
, iter
) {
5185 nest
= dev_get_nest_level(lower
, type_check
);
5186 if (max_nest
< nest
)
5190 if (type_check(dev
))
5195 EXPORT_SYMBOL(dev_get_nest_level
);
5197 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
5199 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5201 if (ops
->ndo_change_rx_flags
)
5202 ops
->ndo_change_rx_flags(dev
, flags
);
5205 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
, bool notify
)
5207 unsigned int old_flags
= dev
->flags
;
5213 dev
->flags
|= IFF_PROMISC
;
5214 dev
->promiscuity
+= inc
;
5215 if (dev
->promiscuity
== 0) {
5218 * If inc causes overflow, untouch promisc and return error.
5221 dev
->flags
&= ~IFF_PROMISC
;
5223 dev
->promiscuity
-= inc
;
5224 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5229 if (dev
->flags
!= old_flags
) {
5230 pr_info("device %s %s promiscuous mode\n",
5232 dev
->flags
& IFF_PROMISC
? "entered" : "left");
5233 if (audit_enabled
) {
5234 current_uid_gid(&uid
, &gid
);
5235 audit_log(current
->audit_context
, GFP_ATOMIC
,
5236 AUDIT_ANOM_PROMISCUOUS
,
5237 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5238 dev
->name
, (dev
->flags
& IFF_PROMISC
),
5239 (old_flags
& IFF_PROMISC
),
5240 from_kuid(&init_user_ns
, audit_get_loginuid(current
)),
5241 from_kuid(&init_user_ns
, uid
),
5242 from_kgid(&init_user_ns
, gid
),
5243 audit_get_sessionid(current
));
5246 dev_change_rx_flags(dev
, IFF_PROMISC
);
5249 __dev_notify_flags(dev
, old_flags
, IFF_PROMISC
);
5254 * dev_set_promiscuity - update promiscuity count on a device
5258 * Add or remove promiscuity from a device. While the count in the device
5259 * remains above zero the interface remains promiscuous. Once it hits zero
5260 * the device reverts back to normal filtering operation. A negative inc
5261 * value is used to drop promiscuity on the device.
5262 * Return 0 if successful or a negative errno code on error.
5264 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
5266 unsigned int old_flags
= dev
->flags
;
5269 err
= __dev_set_promiscuity(dev
, inc
, true);
5272 if (dev
->flags
!= old_flags
)
5273 dev_set_rx_mode(dev
);
5276 EXPORT_SYMBOL(dev_set_promiscuity
);
5278 static int __dev_set_allmulti(struct net_device
*dev
, int inc
, bool notify
)
5280 unsigned int old_flags
= dev
->flags
, old_gflags
= dev
->gflags
;
5284 dev
->flags
|= IFF_ALLMULTI
;
5285 dev
->allmulti
+= inc
;
5286 if (dev
->allmulti
== 0) {
5289 * If inc causes overflow, untouch allmulti and return error.
5292 dev
->flags
&= ~IFF_ALLMULTI
;
5294 dev
->allmulti
-= inc
;
5295 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5300 if (dev
->flags
^ old_flags
) {
5301 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
5302 dev_set_rx_mode(dev
);
5304 __dev_notify_flags(dev
, old_flags
,
5305 dev
->gflags
^ old_gflags
);
5311 * dev_set_allmulti - update allmulti count on a device
5315 * Add or remove reception of all multicast frames to a device. While the
5316 * count in the device remains above zero the interface remains listening
5317 * to all interfaces. Once it hits zero the device reverts back to normal
5318 * filtering operation. A negative @inc value is used to drop the counter
5319 * when releasing a resource needing all multicasts.
5320 * Return 0 if successful or a negative errno code on error.
5323 int dev_set_allmulti(struct net_device
*dev
, int inc
)
5325 return __dev_set_allmulti(dev
, inc
, true);
5327 EXPORT_SYMBOL(dev_set_allmulti
);
5330 * Upload unicast and multicast address lists to device and
5331 * configure RX filtering. When the device doesn't support unicast
5332 * filtering it is put in promiscuous mode while unicast addresses
5335 void __dev_set_rx_mode(struct net_device
*dev
)
5337 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5339 /* dev_open will call this function so the list will stay sane. */
5340 if (!(dev
->flags
&IFF_UP
))
5343 if (!netif_device_present(dev
))
5346 if (!(dev
->priv_flags
& IFF_UNICAST_FLT
)) {
5347 /* Unicast addresses changes may only happen under the rtnl,
5348 * therefore calling __dev_set_promiscuity here is safe.
5350 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
5351 __dev_set_promiscuity(dev
, 1, false);
5352 dev
->uc_promisc
= true;
5353 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
5354 __dev_set_promiscuity(dev
, -1, false);
5355 dev
->uc_promisc
= false;
5359 if (ops
->ndo_set_rx_mode
)
5360 ops
->ndo_set_rx_mode(dev
);
5363 void dev_set_rx_mode(struct net_device
*dev
)
5365 netif_addr_lock_bh(dev
);
5366 __dev_set_rx_mode(dev
);
5367 netif_addr_unlock_bh(dev
);
5371 * dev_get_flags - get flags reported to userspace
5374 * Get the combination of flag bits exported through APIs to userspace.
5376 unsigned int dev_get_flags(const struct net_device
*dev
)
5380 flags
= (dev
->flags
& ~(IFF_PROMISC
|
5385 (dev
->gflags
& (IFF_PROMISC
|
5388 if (netif_running(dev
)) {
5389 if (netif_oper_up(dev
))
5390 flags
|= IFF_RUNNING
;
5391 if (netif_carrier_ok(dev
))
5392 flags
|= IFF_LOWER_UP
;
5393 if (netif_dormant(dev
))
5394 flags
|= IFF_DORMANT
;
5399 EXPORT_SYMBOL(dev_get_flags
);
5401 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
5403 unsigned int old_flags
= dev
->flags
;
5409 * Set the flags on our device.
5412 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
5413 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
5415 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
5419 * Load in the correct multicast list now the flags have changed.
5422 if ((old_flags
^ flags
) & IFF_MULTICAST
)
5423 dev_change_rx_flags(dev
, IFF_MULTICAST
);
5425 dev_set_rx_mode(dev
);
5428 * Have we downed the interface. We handle IFF_UP ourselves
5429 * according to user attempts to set it, rather than blindly
5434 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
5435 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
5438 dev_set_rx_mode(dev
);
5441 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
5442 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
5443 unsigned int old_flags
= dev
->flags
;
5445 dev
->gflags
^= IFF_PROMISC
;
5447 if (__dev_set_promiscuity(dev
, inc
, false) >= 0)
5448 if (dev
->flags
!= old_flags
)
5449 dev_set_rx_mode(dev
);
5452 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5453 is important. Some (broken) drivers set IFF_PROMISC, when
5454 IFF_ALLMULTI is requested not asking us and not reporting.
5456 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
5457 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
5459 dev
->gflags
^= IFF_ALLMULTI
;
5460 __dev_set_allmulti(dev
, inc
, false);
5466 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
,
5467 unsigned int gchanges
)
5469 unsigned int changes
= dev
->flags
^ old_flags
;
5472 rtmsg_ifinfo(RTM_NEWLINK
, dev
, gchanges
, GFP_ATOMIC
);
5474 if (changes
& IFF_UP
) {
5475 if (dev
->flags
& IFF_UP
)
5476 call_netdevice_notifiers(NETDEV_UP
, dev
);
5478 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
5481 if (dev
->flags
& IFF_UP
&&
5482 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
))) {
5483 struct netdev_notifier_change_info change_info
;
5485 change_info
.flags_changed
= changes
;
5486 call_netdevice_notifiers_info(NETDEV_CHANGE
, dev
,
5492 * dev_change_flags - change device settings
5494 * @flags: device state flags
5496 * Change settings on device based state flags. The flags are
5497 * in the userspace exported format.
5499 int dev_change_flags(struct net_device
*dev
, unsigned int flags
)
5502 unsigned int changes
, old_flags
= dev
->flags
, old_gflags
= dev
->gflags
;
5504 ret
= __dev_change_flags(dev
, flags
);
5508 changes
= (old_flags
^ dev
->flags
) | (old_gflags
^ dev
->gflags
);
5509 __dev_notify_flags(dev
, old_flags
, changes
);
5512 EXPORT_SYMBOL(dev_change_flags
);
5514 static int __dev_set_mtu(struct net_device
*dev
, int new_mtu
)
5516 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5518 if (ops
->ndo_change_mtu
)
5519 return ops
->ndo_change_mtu(dev
, new_mtu
);
5526 * dev_set_mtu - Change maximum transfer unit
5528 * @new_mtu: new transfer unit
5530 * Change the maximum transfer size of the network device.
5532 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
5536 if (new_mtu
== dev
->mtu
)
5539 /* MTU must be positive. */
5543 if (!netif_device_present(dev
))
5546 err
= call_netdevice_notifiers(NETDEV_PRECHANGEMTU
, dev
);
5547 err
= notifier_to_errno(err
);
5551 orig_mtu
= dev
->mtu
;
5552 err
= __dev_set_mtu(dev
, new_mtu
);
5555 err
= call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
5556 err
= notifier_to_errno(err
);
5558 /* setting mtu back and notifying everyone again,
5559 * so that they have a chance to revert changes.
5561 __dev_set_mtu(dev
, orig_mtu
);
5562 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
5567 EXPORT_SYMBOL(dev_set_mtu
);
5570 * dev_set_group - Change group this device belongs to
5572 * @new_group: group this device should belong to
5574 void dev_set_group(struct net_device
*dev
, int new_group
)
5576 dev
->group
= new_group
;
5578 EXPORT_SYMBOL(dev_set_group
);
5581 * dev_set_mac_address - Change Media Access Control Address
5585 * Change the hardware (MAC) address of the device
5587 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
5589 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5592 if (!ops
->ndo_set_mac_address
)
5594 if (sa
->sa_family
!= dev
->type
)
5596 if (!netif_device_present(dev
))
5598 err
= ops
->ndo_set_mac_address(dev
, sa
);
5601 dev
->addr_assign_type
= NET_ADDR_SET
;
5602 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
5603 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
5606 EXPORT_SYMBOL(dev_set_mac_address
);
5609 * dev_change_carrier - Change device carrier
5611 * @new_carrier: new value
5613 * Change device carrier
5615 int dev_change_carrier(struct net_device
*dev
, bool new_carrier
)
5617 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5619 if (!ops
->ndo_change_carrier
)
5621 if (!netif_device_present(dev
))
5623 return ops
->ndo_change_carrier(dev
, new_carrier
);
5625 EXPORT_SYMBOL(dev_change_carrier
);
5628 * dev_get_phys_port_id - Get device physical port ID
5632 * Get device physical port ID
5634 int dev_get_phys_port_id(struct net_device
*dev
,
5635 struct netdev_phys_port_id
*ppid
)
5637 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5639 if (!ops
->ndo_get_phys_port_id
)
5641 return ops
->ndo_get_phys_port_id(dev
, ppid
);
5643 EXPORT_SYMBOL(dev_get_phys_port_id
);
5646 * dev_new_index - allocate an ifindex
5647 * @net: the applicable net namespace
5649 * Returns a suitable unique value for a new device interface
5650 * number. The caller must hold the rtnl semaphore or the
5651 * dev_base_lock to be sure it remains unique.
5653 static int dev_new_index(struct net
*net
)
5655 int ifindex
= net
->ifindex
;
5659 if (!__dev_get_by_index(net
, ifindex
))
5660 return net
->ifindex
= ifindex
;
5664 /* Delayed registration/unregisteration */
5665 static LIST_HEAD(net_todo_list
);
5666 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq
);
5668 static void net_set_todo(struct net_device
*dev
)
5670 list_add_tail(&dev
->todo_list
, &net_todo_list
);
5671 dev_net(dev
)->dev_unreg_count
++;
5674 static void rollback_registered_many(struct list_head
*head
)
5676 struct net_device
*dev
, *tmp
;
5677 LIST_HEAD(close_head
);
5679 BUG_ON(dev_boot_phase
);
5682 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
5683 /* Some devices call without registering
5684 * for initialization unwind. Remove those
5685 * devices and proceed with the remaining.
5687 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5688 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5692 list_del(&dev
->unreg_list
);
5695 dev
->dismantle
= true;
5696 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
5699 /* If device is running, close it first. */
5700 list_for_each_entry(dev
, head
, unreg_list
)
5701 list_add_tail(&dev
->close_list
, &close_head
);
5702 dev_close_many(&close_head
);
5704 list_for_each_entry(dev
, head
, unreg_list
) {
5705 /* And unlink it from device chain. */
5706 unlist_netdevice(dev
);
5708 dev
->reg_state
= NETREG_UNREGISTERING
;
5713 list_for_each_entry(dev
, head
, unreg_list
) {
5714 /* Shutdown queueing discipline. */
5718 /* Notify protocols, that we are about to destroy
5719 this device. They should clean all the things.
5721 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5724 * Flush the unicast and multicast chains
5729 if (dev
->netdev_ops
->ndo_uninit
)
5730 dev
->netdev_ops
->ndo_uninit(dev
);
5732 if (!dev
->rtnl_link_ops
||
5733 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5734 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U, GFP_KERNEL
);
5736 /* Notifier chain MUST detach us all upper devices. */
5737 WARN_ON(netdev_has_any_upper_dev(dev
));
5739 /* Remove entries from kobject tree */
5740 netdev_unregister_kobject(dev
);
5742 /* Remove XPS queueing entries */
5743 netif_reset_xps_queues_gt(dev
, 0);
5749 list_for_each_entry(dev
, head
, unreg_list
)
5753 static void rollback_registered(struct net_device
*dev
)
5757 list_add(&dev
->unreg_list
, &single
);
5758 rollback_registered_many(&single
);
5762 static netdev_features_t
netdev_fix_features(struct net_device
*dev
,
5763 netdev_features_t features
)
5765 /* Fix illegal checksum combinations */
5766 if ((features
& NETIF_F_HW_CSUM
) &&
5767 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5768 netdev_warn(dev
, "mixed HW and IP checksum settings.\n");
5769 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5772 /* TSO requires that SG is present as well. */
5773 if ((features
& NETIF_F_ALL_TSO
) && !(features
& NETIF_F_SG
)) {
5774 netdev_dbg(dev
, "Dropping TSO features since no SG feature.\n");
5775 features
&= ~NETIF_F_ALL_TSO
;
5778 if ((features
& NETIF_F_TSO
) && !(features
& NETIF_F_HW_CSUM
) &&
5779 !(features
& NETIF_F_IP_CSUM
)) {
5780 netdev_dbg(dev
, "Dropping TSO features since no CSUM feature.\n");
5781 features
&= ~NETIF_F_TSO
;
5782 features
&= ~NETIF_F_TSO_ECN
;
5785 if ((features
& NETIF_F_TSO6
) && !(features
& NETIF_F_HW_CSUM
) &&
5786 !(features
& NETIF_F_IPV6_CSUM
)) {
5787 netdev_dbg(dev
, "Dropping TSO6 features since no CSUM feature.\n");
5788 features
&= ~NETIF_F_TSO6
;
5791 /* TSO ECN requires that TSO is present as well. */
5792 if ((features
& NETIF_F_ALL_TSO
) == NETIF_F_TSO_ECN
)
5793 features
&= ~NETIF_F_TSO_ECN
;
5795 /* Software GSO depends on SG. */
5796 if ((features
& NETIF_F_GSO
) && !(features
& NETIF_F_SG
)) {
5797 netdev_dbg(dev
, "Dropping NETIF_F_GSO since no SG feature.\n");
5798 features
&= ~NETIF_F_GSO
;
5801 /* UFO needs SG and checksumming */
5802 if (features
& NETIF_F_UFO
) {
5803 /* maybe split UFO into V4 and V6? */
5804 if (!((features
& NETIF_F_GEN_CSUM
) ||
5805 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))
5806 == (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5808 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5809 features
&= ~NETIF_F_UFO
;
5812 if (!(features
& NETIF_F_SG
)) {
5814 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5815 features
&= ~NETIF_F_UFO
;
5819 #ifdef CONFIG_NET_RX_BUSY_POLL
5820 if (dev
->netdev_ops
->ndo_busy_poll
)
5821 features
|= NETIF_F_BUSY_POLL
;
5824 features
&= ~NETIF_F_BUSY_POLL
;
5829 int __netdev_update_features(struct net_device
*dev
)
5831 netdev_features_t features
;
5836 features
= netdev_get_wanted_features(dev
);
5838 if (dev
->netdev_ops
->ndo_fix_features
)
5839 features
= dev
->netdev_ops
->ndo_fix_features(dev
, features
);
5841 /* driver might be less strict about feature dependencies */
5842 features
= netdev_fix_features(dev
, features
);
5844 if (dev
->features
== features
)
5847 netdev_dbg(dev
, "Features changed: %pNF -> %pNF\n",
5848 &dev
->features
, &features
);
5850 if (dev
->netdev_ops
->ndo_set_features
)
5851 err
= dev
->netdev_ops
->ndo_set_features(dev
, features
);
5853 if (unlikely(err
< 0)) {
5855 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5856 err
, &features
, &dev
->features
);
5861 dev
->features
= features
;
5867 * netdev_update_features - recalculate device features
5868 * @dev: the device to check
5870 * Recalculate dev->features set and send notifications if it
5871 * has changed. Should be called after driver or hardware dependent
5872 * conditions might have changed that influence the features.
5874 void netdev_update_features(struct net_device
*dev
)
5876 if (__netdev_update_features(dev
))
5877 netdev_features_change(dev
);
5879 EXPORT_SYMBOL(netdev_update_features
);
5882 * netdev_change_features - recalculate device features
5883 * @dev: the device to check
5885 * Recalculate dev->features set and send notifications even
5886 * if they have not changed. Should be called instead of
5887 * netdev_update_features() if also dev->vlan_features might
5888 * have changed to allow the changes to be propagated to stacked
5891 void netdev_change_features(struct net_device
*dev
)
5893 __netdev_update_features(dev
);
5894 netdev_features_change(dev
);
5896 EXPORT_SYMBOL(netdev_change_features
);
5899 * netif_stacked_transfer_operstate - transfer operstate
5900 * @rootdev: the root or lower level device to transfer state from
5901 * @dev: the device to transfer operstate to
5903 * Transfer operational state from root to device. This is normally
5904 * called when a stacking relationship exists between the root
5905 * device and the device(a leaf device).
5907 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
5908 struct net_device
*dev
)
5910 if (rootdev
->operstate
== IF_OPER_DORMANT
)
5911 netif_dormant_on(dev
);
5913 netif_dormant_off(dev
);
5915 if (netif_carrier_ok(rootdev
)) {
5916 if (!netif_carrier_ok(dev
))
5917 netif_carrier_on(dev
);
5919 if (netif_carrier_ok(dev
))
5920 netif_carrier_off(dev
);
5923 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
5926 static int netif_alloc_rx_queues(struct net_device
*dev
)
5928 unsigned int i
, count
= dev
->num_rx_queues
;
5929 struct netdev_rx_queue
*rx
;
5933 rx
= kcalloc(count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5939 for (i
= 0; i
< count
; i
++)
5945 static void netdev_init_one_queue(struct net_device
*dev
,
5946 struct netdev_queue
*queue
, void *_unused
)
5948 /* Initialize queue lock */
5949 spin_lock_init(&queue
->_xmit_lock
);
5950 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
5951 queue
->xmit_lock_owner
= -1;
5952 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
5955 dql_init(&queue
->dql
, HZ
);
5959 static void netif_free_tx_queues(struct net_device
*dev
)
5964 static int netif_alloc_netdev_queues(struct net_device
*dev
)
5966 unsigned int count
= dev
->num_tx_queues
;
5967 struct netdev_queue
*tx
;
5968 size_t sz
= count
* sizeof(*tx
);
5970 BUG_ON(count
< 1 || count
> 0xffff);
5972 tx
= kzalloc(sz
, GFP_KERNEL
| __GFP_NOWARN
| __GFP_REPEAT
);
5980 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5981 spin_lock_init(&dev
->tx_global_lock
);
5987 * register_netdevice - register a network device
5988 * @dev: device to register
5990 * Take a completed network device structure and add it to the kernel
5991 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5992 * chain. 0 is returned on success. A negative errno code is returned
5993 * on a failure to set up the device, or if the name is a duplicate.
5995 * Callers must hold the rtnl semaphore. You may want
5996 * register_netdev() instead of this.
5999 * The locking appears insufficient to guarantee two parallel registers
6000 * will not get the same name.
6003 int register_netdevice(struct net_device
*dev
)
6006 struct net
*net
= dev_net(dev
);
6008 BUG_ON(dev_boot_phase
);
6013 /* When net_device's are persistent, this will be fatal. */
6014 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
6017 spin_lock_init(&dev
->addr_list_lock
);
6018 netdev_set_addr_lockdep_class(dev
);
6022 ret
= dev_get_valid_name(net
, dev
, dev
->name
);
6026 /* Init, if this function is available */
6027 if (dev
->netdev_ops
->ndo_init
) {
6028 ret
= dev
->netdev_ops
->ndo_init(dev
);
6036 if (((dev
->hw_features
| dev
->features
) &
6037 NETIF_F_HW_VLAN_CTAG_FILTER
) &&
6038 (!dev
->netdev_ops
->ndo_vlan_rx_add_vid
||
6039 !dev
->netdev_ops
->ndo_vlan_rx_kill_vid
)) {
6040 netdev_WARN(dev
, "Buggy VLAN acceleration in driver!\n");
6047 dev
->ifindex
= dev_new_index(net
);
6048 else if (__dev_get_by_index(net
, dev
->ifindex
))
6051 if (dev
->iflink
== -1)
6052 dev
->iflink
= dev
->ifindex
;
6054 /* Transfer changeable features to wanted_features and enable
6055 * software offloads (GSO and GRO).
6057 dev
->hw_features
|= NETIF_F_SOFT_FEATURES
;
6058 dev
->features
|= NETIF_F_SOFT_FEATURES
;
6059 dev
->wanted_features
= dev
->features
& dev
->hw_features
;
6061 if (!(dev
->flags
& IFF_LOOPBACK
)) {
6062 dev
->hw_features
|= NETIF_F_NOCACHE_COPY
;
6065 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6067 dev
->vlan_features
|= NETIF_F_HIGHDMA
;
6069 /* Make NETIF_F_SG inheritable to tunnel devices.
6071 dev
->hw_enc_features
|= NETIF_F_SG
;
6073 /* Make NETIF_F_SG inheritable to MPLS.
6075 dev
->mpls_features
|= NETIF_F_SG
;
6077 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
6078 ret
= notifier_to_errno(ret
);
6082 ret
= netdev_register_kobject(dev
);
6085 dev
->reg_state
= NETREG_REGISTERED
;
6087 __netdev_update_features(dev
);
6090 * Default initial state at registry is that the
6091 * device is present.
6094 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
6096 linkwatch_init_dev(dev
);
6098 dev_init_scheduler(dev
);
6100 list_netdevice(dev
);
6101 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
6103 /* If the device has permanent device address, driver should
6104 * set dev_addr and also addr_assign_type should be set to
6105 * NET_ADDR_PERM (default value).
6107 if (dev
->addr_assign_type
== NET_ADDR_PERM
)
6108 memcpy(dev
->perm_addr
, dev
->dev_addr
, dev
->addr_len
);
6110 /* Notify protocols, that a new device appeared. */
6111 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
6112 ret
= notifier_to_errno(ret
);
6114 rollback_registered(dev
);
6115 dev
->reg_state
= NETREG_UNREGISTERED
;
6118 * Prevent userspace races by waiting until the network
6119 * device is fully setup before sending notifications.
6121 if (!dev
->rtnl_link_ops
||
6122 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
6123 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U, GFP_KERNEL
);
6129 if (dev
->netdev_ops
->ndo_uninit
)
6130 dev
->netdev_ops
->ndo_uninit(dev
);
6133 EXPORT_SYMBOL(register_netdevice
);
6136 * init_dummy_netdev - init a dummy network device for NAPI
6137 * @dev: device to init
6139 * This takes a network device structure and initialize the minimum
6140 * amount of fields so it can be used to schedule NAPI polls without
6141 * registering a full blown interface. This is to be used by drivers
6142 * that need to tie several hardware interfaces to a single NAPI
6143 * poll scheduler due to HW limitations.
6145 int init_dummy_netdev(struct net_device
*dev
)
6147 /* Clear everything. Note we don't initialize spinlocks
6148 * are they aren't supposed to be taken by any of the
6149 * NAPI code and this dummy netdev is supposed to be
6150 * only ever used for NAPI polls
6152 memset(dev
, 0, sizeof(struct net_device
));
6154 /* make sure we BUG if trying to hit standard
6155 * register/unregister code path
6157 dev
->reg_state
= NETREG_DUMMY
;
6159 /* NAPI wants this */
6160 INIT_LIST_HEAD(&dev
->napi_list
);
6162 /* a dummy interface is started by default */
6163 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
6164 set_bit(__LINK_STATE_START
, &dev
->state
);
6166 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6167 * because users of this 'device' dont need to change
6173 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
6177 * register_netdev - register a network device
6178 * @dev: device to register
6180 * Take a completed network device structure and add it to the kernel
6181 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6182 * chain. 0 is returned on success. A negative errno code is returned
6183 * on a failure to set up the device, or if the name is a duplicate.
6185 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6186 * and expands the device name if you passed a format string to
6189 int register_netdev(struct net_device
*dev
)
6194 err
= register_netdevice(dev
);
6198 EXPORT_SYMBOL(register_netdev
);
6200 int netdev_refcnt_read(const struct net_device
*dev
)
6204 for_each_possible_cpu(i
)
6205 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
6208 EXPORT_SYMBOL(netdev_refcnt_read
);
6211 * netdev_wait_allrefs - wait until all references are gone.
6212 * @dev: target net_device
6214 * This is called when unregistering network devices.
6216 * Any protocol or device that holds a reference should register
6217 * for netdevice notification, and cleanup and put back the
6218 * reference if they receive an UNREGISTER event.
6219 * We can get stuck here if buggy protocols don't correctly
6222 static void netdev_wait_allrefs(struct net_device
*dev
)
6224 unsigned long rebroadcast_time
, warning_time
;
6227 linkwatch_forget_dev(dev
);
6229 rebroadcast_time
= warning_time
= jiffies
;
6230 refcnt
= netdev_refcnt_read(dev
);
6232 while (refcnt
!= 0) {
6233 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
6236 /* Rebroadcast unregister notification */
6237 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6243 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
6244 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
6246 /* We must not have linkwatch events
6247 * pending on unregister. If this
6248 * happens, we simply run the queue
6249 * unscheduled, resulting in a noop
6252 linkwatch_run_queue();
6257 rebroadcast_time
= jiffies
;
6262 refcnt
= netdev_refcnt_read(dev
);
6264 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
6265 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6267 warning_time
= jiffies
;
6276 * register_netdevice(x1);
6277 * register_netdevice(x2);
6279 * unregister_netdevice(y1);
6280 * unregister_netdevice(y2);
6286 * We are invoked by rtnl_unlock().
6287 * This allows us to deal with problems:
6288 * 1) We can delete sysfs objects which invoke hotplug
6289 * without deadlocking with linkwatch via keventd.
6290 * 2) Since we run with the RTNL semaphore not held, we can sleep
6291 * safely in order to wait for the netdev refcnt to drop to zero.
6293 * We must not return until all unregister events added during
6294 * the interval the lock was held have been completed.
6296 void netdev_run_todo(void)
6298 struct list_head list
;
6300 /* Snapshot list, allow later requests */
6301 list_replace_init(&net_todo_list
, &list
);
6306 /* Wait for rcu callbacks to finish before next phase */
6307 if (!list_empty(&list
))
6310 while (!list_empty(&list
)) {
6311 struct net_device
*dev
6312 = list_first_entry(&list
, struct net_device
, todo_list
);
6313 list_del(&dev
->todo_list
);
6316 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
6319 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
6320 pr_err("network todo '%s' but state %d\n",
6321 dev
->name
, dev
->reg_state
);
6326 dev
->reg_state
= NETREG_UNREGISTERED
;
6328 on_each_cpu(flush_backlog
, dev
, 1);
6330 netdev_wait_allrefs(dev
);
6333 BUG_ON(netdev_refcnt_read(dev
));
6334 WARN_ON(rcu_access_pointer(dev
->ip_ptr
));
6335 WARN_ON(rcu_access_pointer(dev
->ip6_ptr
));
6336 WARN_ON(dev
->dn_ptr
);
6338 if (dev
->destructor
)
6339 dev
->destructor(dev
);
6341 /* Report a network device has been unregistered */
6343 dev_net(dev
)->dev_unreg_count
--;
6345 wake_up(&netdev_unregistering_wq
);
6347 /* Free network device */
6348 kobject_put(&dev
->dev
.kobj
);
6352 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6353 * fields in the same order, with only the type differing.
6355 void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
6356 const struct net_device_stats
*netdev_stats
)
6358 #if BITS_PER_LONG == 64
6359 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
6360 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
6362 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
6363 const unsigned long *src
= (const unsigned long *)netdev_stats
;
6364 u64
*dst
= (u64
*)stats64
;
6366 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
6367 sizeof(*stats64
) / sizeof(u64
));
6368 for (i
= 0; i
< n
; i
++)
6372 EXPORT_SYMBOL(netdev_stats_to_stats64
);
6375 * dev_get_stats - get network device statistics
6376 * @dev: device to get statistics from
6377 * @storage: place to store stats
6379 * Get network statistics from device. Return @storage.
6380 * The device driver may provide its own method by setting
6381 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6382 * otherwise the internal statistics structure is used.
6384 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
6385 struct rtnl_link_stats64
*storage
)
6387 const struct net_device_ops
*ops
= dev
->netdev_ops
;
6389 if (ops
->ndo_get_stats64
) {
6390 memset(storage
, 0, sizeof(*storage
));
6391 ops
->ndo_get_stats64(dev
, storage
);
6392 } else if (ops
->ndo_get_stats
) {
6393 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
6395 netdev_stats_to_stats64(storage
, &dev
->stats
);
6397 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
6398 storage
->tx_dropped
+= atomic_long_read(&dev
->tx_dropped
);
6401 EXPORT_SYMBOL(dev_get_stats
);
6403 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
6405 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
6407 #ifdef CONFIG_NET_CLS_ACT
6410 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
6413 netdev_init_one_queue(dev
, queue
, NULL
);
6414 queue
->qdisc
= &noop_qdisc
;
6415 queue
->qdisc_sleeping
= &noop_qdisc
;
6416 rcu_assign_pointer(dev
->ingress_queue
, queue
);
6421 static const struct ethtool_ops default_ethtool_ops
;
6423 void netdev_set_default_ethtool_ops(struct net_device
*dev
,
6424 const struct ethtool_ops
*ops
)
6426 if (dev
->ethtool_ops
== &default_ethtool_ops
)
6427 dev
->ethtool_ops
= ops
;
6429 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops
);
6431 void netdev_freemem(struct net_device
*dev
)
6433 char *addr
= (char *)dev
- dev
->padded
;
6439 * alloc_netdev_mqs - allocate network device
6440 * @sizeof_priv: size of private data to allocate space for
6441 * @name: device name format string
6442 * @setup: callback to initialize device
6443 * @txqs: the number of TX subqueues to allocate
6444 * @rxqs: the number of RX subqueues to allocate
6446 * Allocates a struct net_device with private data area for driver use
6447 * and performs basic initialization. Also allocates subqueue structs
6448 * for each queue on the device.
6450 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
6451 void (*setup
)(struct net_device
*),
6452 unsigned int txqs
, unsigned int rxqs
)
6454 struct net_device
*dev
;
6456 struct net_device
*p
;
6458 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
6461 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6467 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6472 alloc_size
= sizeof(struct net_device
);
6474 /* ensure 32-byte alignment of private area */
6475 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
6476 alloc_size
+= sizeof_priv
;
6478 /* ensure 32-byte alignment of whole construct */
6479 alloc_size
+= NETDEV_ALIGN
- 1;
6481 p
= kzalloc(alloc_size
, GFP_KERNEL
| __GFP_NOWARN
| __GFP_REPEAT
);
6483 p
= vzalloc(alloc_size
);
6487 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
6488 dev
->padded
= (char *)dev
- (char *)p
;
6490 dev
->pcpu_refcnt
= alloc_percpu(int);
6491 if (!dev
->pcpu_refcnt
)
6494 if (dev_addr_init(dev
))
6500 dev_net_set(dev
, &init_net
);
6502 dev
->gso_max_size
= GSO_MAX_SIZE
;
6503 dev
->gso_max_segs
= GSO_MAX_SEGS
;
6505 INIT_LIST_HEAD(&dev
->napi_list
);
6506 INIT_LIST_HEAD(&dev
->unreg_list
);
6507 INIT_LIST_HEAD(&dev
->close_list
);
6508 INIT_LIST_HEAD(&dev
->link_watch_list
);
6509 INIT_LIST_HEAD(&dev
->adj_list
.upper
);
6510 INIT_LIST_HEAD(&dev
->adj_list
.lower
);
6511 INIT_LIST_HEAD(&dev
->all_adj_list
.upper
);
6512 INIT_LIST_HEAD(&dev
->all_adj_list
.lower
);
6513 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
6516 dev
->num_tx_queues
= txqs
;
6517 dev
->real_num_tx_queues
= txqs
;
6518 if (netif_alloc_netdev_queues(dev
))
6522 dev
->num_rx_queues
= rxqs
;
6523 dev
->real_num_rx_queues
= rxqs
;
6524 if (netif_alloc_rx_queues(dev
))
6528 strcpy(dev
->name
, name
);
6529 dev
->group
= INIT_NETDEV_GROUP
;
6530 if (!dev
->ethtool_ops
)
6531 dev
->ethtool_ops
= &default_ethtool_ops
;
6539 free_percpu(dev
->pcpu_refcnt
);
6541 netdev_freemem(dev
);
6544 EXPORT_SYMBOL(alloc_netdev_mqs
);
6547 * free_netdev - free network device
6550 * This function does the last stage of destroying an allocated device
6551 * interface. The reference to the device object is released.
6552 * If this is the last reference then it will be freed.
6554 void free_netdev(struct net_device
*dev
)
6556 struct napi_struct
*p
, *n
;
6558 release_net(dev_net(dev
));
6560 netif_free_tx_queues(dev
);
6565 kfree(rcu_dereference_protected(dev
->ingress_queue
, 1));
6567 /* Flush device addresses */
6568 dev_addr_flush(dev
);
6570 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
6573 free_percpu(dev
->pcpu_refcnt
);
6574 dev
->pcpu_refcnt
= NULL
;
6576 /* Compatibility with error handling in drivers */
6577 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
6578 netdev_freemem(dev
);
6582 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
6583 dev
->reg_state
= NETREG_RELEASED
;
6585 /* will free via device release */
6586 put_device(&dev
->dev
);
6588 EXPORT_SYMBOL(free_netdev
);
6591 * synchronize_net - Synchronize with packet receive processing
6593 * Wait for packets currently being received to be done.
6594 * Does not block later packets from starting.
6596 void synchronize_net(void)
6599 if (rtnl_is_locked())
6600 synchronize_rcu_expedited();
6604 EXPORT_SYMBOL(synchronize_net
);
6607 * unregister_netdevice_queue - remove device from the kernel
6611 * This function shuts down a device interface and removes it
6612 * from the kernel tables.
6613 * If head not NULL, device is queued to be unregistered later.
6615 * Callers must hold the rtnl semaphore. You may want
6616 * unregister_netdev() instead of this.
6619 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
6624 list_move_tail(&dev
->unreg_list
, head
);
6626 rollback_registered(dev
);
6627 /* Finish processing unregister after unlock */
6631 EXPORT_SYMBOL(unregister_netdevice_queue
);
6634 * unregister_netdevice_many - unregister many devices
6635 * @head: list of devices
6637 * Note: As most callers use a stack allocated list_head,
6638 * we force a list_del() to make sure stack wont be corrupted later.
6640 void unregister_netdevice_many(struct list_head
*head
)
6642 struct net_device
*dev
;
6644 if (!list_empty(head
)) {
6645 rollback_registered_many(head
);
6646 list_for_each_entry(dev
, head
, unreg_list
)
6651 EXPORT_SYMBOL(unregister_netdevice_many
);
6654 * unregister_netdev - remove device from the kernel
6657 * This function shuts down a device interface and removes it
6658 * from the kernel tables.
6660 * This is just a wrapper for unregister_netdevice that takes
6661 * the rtnl semaphore. In general you want to use this and not
6662 * unregister_netdevice.
6664 void unregister_netdev(struct net_device
*dev
)
6667 unregister_netdevice(dev
);
6670 EXPORT_SYMBOL(unregister_netdev
);
6673 * dev_change_net_namespace - move device to different nethost namespace
6675 * @net: network namespace
6676 * @pat: If not NULL name pattern to try if the current device name
6677 * is already taken in the destination network namespace.
6679 * This function shuts down a device interface and moves it
6680 * to a new network namespace. On success 0 is returned, on
6681 * a failure a netagive errno code is returned.
6683 * Callers must hold the rtnl semaphore.
6686 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
6692 /* Don't allow namespace local devices to be moved. */
6694 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6697 /* Ensure the device has been registrered */
6698 if (dev
->reg_state
!= NETREG_REGISTERED
)
6701 /* Get out if there is nothing todo */
6703 if (net_eq(dev_net(dev
), net
))
6706 /* Pick the destination device name, and ensure
6707 * we can use it in the destination network namespace.
6710 if (__dev_get_by_name(net
, dev
->name
)) {
6711 /* We get here if we can't use the current device name */
6714 if (dev_get_valid_name(net
, dev
, pat
) < 0)
6719 * And now a mini version of register_netdevice unregister_netdevice.
6722 /* If device is running close it first. */
6725 /* And unlink it from device chain */
6727 unlist_netdevice(dev
);
6731 /* Shutdown queueing discipline. */
6734 /* Notify protocols, that we are about to destroy
6735 this device. They should clean all the things.
6737 Note that dev->reg_state stays at NETREG_REGISTERED.
6738 This is wanted because this way 8021q and macvlan know
6739 the device is just moving and can keep their slaves up.
6741 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6743 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL
, dev
);
6744 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U, GFP_KERNEL
);
6747 * Flush the unicast and multicast chains
6752 /* Send a netdev-removed uevent to the old namespace */
6753 kobject_uevent(&dev
->dev
.kobj
, KOBJ_REMOVE
);
6755 /* Actually switch the network namespace */
6756 dev_net_set(dev
, net
);
6758 /* If there is an ifindex conflict assign a new one */
6759 if (__dev_get_by_index(net
, dev
->ifindex
)) {
6760 int iflink
= (dev
->iflink
== dev
->ifindex
);
6761 dev
->ifindex
= dev_new_index(net
);
6763 dev
->iflink
= dev
->ifindex
;
6766 /* Send a netdev-add uevent to the new namespace */
6767 kobject_uevent(&dev
->dev
.kobj
, KOBJ_ADD
);
6769 /* Fixup kobjects */
6770 err
= device_rename(&dev
->dev
, dev
->name
);
6773 /* Add the device back in the hashes */
6774 list_netdevice(dev
);
6776 /* Notify protocols, that a new device appeared. */
6777 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
6780 * Prevent userspace races by waiting until the network
6781 * device is fully setup before sending notifications.
6783 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U, GFP_KERNEL
);
6790 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
6792 static int dev_cpu_callback(struct notifier_block
*nfb
,
6793 unsigned long action
,
6796 struct sk_buff
**list_skb
;
6797 struct sk_buff
*skb
;
6798 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
6799 struct softnet_data
*sd
, *oldsd
;
6801 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
6804 local_irq_disable();
6805 cpu
= smp_processor_id();
6806 sd
= &per_cpu(softnet_data
, cpu
);
6807 oldsd
= &per_cpu(softnet_data
, oldcpu
);
6809 /* Find end of our completion_queue. */
6810 list_skb
= &sd
->completion_queue
;
6812 list_skb
= &(*list_skb
)->next
;
6813 /* Append completion queue from offline CPU. */
6814 *list_skb
= oldsd
->completion_queue
;
6815 oldsd
->completion_queue
= NULL
;
6817 /* Append output queue from offline CPU. */
6818 if (oldsd
->output_queue
) {
6819 *sd
->output_queue_tailp
= oldsd
->output_queue
;
6820 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
6821 oldsd
->output_queue
= NULL
;
6822 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
6824 /* Append NAPI poll list from offline CPU. */
6825 if (!list_empty(&oldsd
->poll_list
)) {
6826 list_splice_init(&oldsd
->poll_list
, &sd
->poll_list
);
6827 raise_softirq_irqoff(NET_RX_SOFTIRQ
);
6830 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
6833 /* Process offline CPU's input_pkt_queue */
6834 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
6835 netif_rx_internal(skb
);
6836 input_queue_head_incr(oldsd
);
6838 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
6839 netif_rx_internal(skb
);
6840 input_queue_head_incr(oldsd
);
6848 * netdev_increment_features - increment feature set by one
6849 * @all: current feature set
6850 * @one: new feature set
6851 * @mask: mask feature set
6853 * Computes a new feature set after adding a device with feature set
6854 * @one to the master device with current feature set @all. Will not
6855 * enable anything that is off in @mask. Returns the new feature set.
6857 netdev_features_t
netdev_increment_features(netdev_features_t all
,
6858 netdev_features_t one
, netdev_features_t mask
)
6860 if (mask
& NETIF_F_GEN_CSUM
)
6861 mask
|= NETIF_F_ALL_CSUM
;
6862 mask
|= NETIF_F_VLAN_CHALLENGED
;
6864 all
|= one
& (NETIF_F_ONE_FOR_ALL
|NETIF_F_ALL_CSUM
) & mask
;
6865 all
&= one
| ~NETIF_F_ALL_FOR_ALL
;
6867 /* If one device supports hw checksumming, set for all. */
6868 if (all
& NETIF_F_GEN_CSUM
)
6869 all
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_GEN_CSUM
);
6873 EXPORT_SYMBOL(netdev_increment_features
);
6875 static struct hlist_head
* __net_init
netdev_create_hash(void)
6878 struct hlist_head
*hash
;
6880 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
6882 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
6883 INIT_HLIST_HEAD(&hash
[i
]);
6888 /* Initialize per network namespace state */
6889 static int __net_init
netdev_init(struct net
*net
)
6891 if (net
!= &init_net
)
6892 INIT_LIST_HEAD(&net
->dev_base_head
);
6894 net
->dev_name_head
= netdev_create_hash();
6895 if (net
->dev_name_head
== NULL
)
6898 net
->dev_index_head
= netdev_create_hash();
6899 if (net
->dev_index_head
== NULL
)
6905 kfree(net
->dev_name_head
);
6911 * netdev_drivername - network driver for the device
6912 * @dev: network device
6914 * Determine network driver for device.
6916 const char *netdev_drivername(const struct net_device
*dev
)
6918 const struct device_driver
*driver
;
6919 const struct device
*parent
;
6920 const char *empty
= "";
6922 parent
= dev
->dev
.parent
;
6926 driver
= parent
->driver
;
6927 if (driver
&& driver
->name
)
6928 return driver
->name
;
6932 static int __netdev_printk(const char *level
, const struct net_device
*dev
,
6933 struct va_format
*vaf
)
6937 if (dev
&& dev
->dev
.parent
) {
6938 r
= dev_printk_emit(level
[1] - '0',
6941 dev_driver_string(dev
->dev
.parent
),
6942 dev_name(dev
->dev
.parent
),
6943 netdev_name(dev
), vaf
);
6945 r
= printk("%s%s: %pV", level
, netdev_name(dev
), vaf
);
6947 r
= printk("%s(NULL net_device): %pV", level
, vaf
);
6953 int netdev_printk(const char *level
, const struct net_device
*dev
,
6954 const char *format
, ...)
6956 struct va_format vaf
;
6960 va_start(args
, format
);
6965 r
= __netdev_printk(level
, dev
, &vaf
);
6971 EXPORT_SYMBOL(netdev_printk
);
6973 #define define_netdev_printk_level(func, level) \
6974 int func(const struct net_device *dev, const char *fmt, ...) \
6977 struct va_format vaf; \
6980 va_start(args, fmt); \
6985 r = __netdev_printk(level, dev, &vaf); \
6991 EXPORT_SYMBOL(func);
6993 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
6994 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
6995 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
6996 define_netdev_printk_level(netdev_err
, KERN_ERR
);
6997 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
6998 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
6999 define_netdev_printk_level(netdev_info
, KERN_INFO
);
7001 static void __net_exit
netdev_exit(struct net
*net
)
7003 kfree(net
->dev_name_head
);
7004 kfree(net
->dev_index_head
);
7007 static struct pernet_operations __net_initdata netdev_net_ops
= {
7008 .init
= netdev_init
,
7009 .exit
= netdev_exit
,
7012 static void __net_exit
default_device_exit(struct net
*net
)
7014 struct net_device
*dev
, *aux
;
7016 * Push all migratable network devices back to the
7017 * initial network namespace
7020 for_each_netdev_safe(net
, dev
, aux
) {
7022 char fb_name
[IFNAMSIZ
];
7024 /* Ignore unmoveable devices (i.e. loopback) */
7025 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
7028 /* Leave virtual devices for the generic cleanup */
7029 if (dev
->rtnl_link_ops
)
7032 /* Push remaining network devices to init_net */
7033 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
7034 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
7036 pr_emerg("%s: failed to move %s to init_net: %d\n",
7037 __func__
, dev
->name
, err
);
7044 static void __net_exit
rtnl_lock_unregistering(struct list_head
*net_list
)
7046 /* Return with the rtnl_lock held when there are no network
7047 * devices unregistering in any network namespace in net_list.
7054 prepare_to_wait(&netdev_unregistering_wq
, &wait
,
7055 TASK_UNINTERRUPTIBLE
);
7056 unregistering
= false;
7058 list_for_each_entry(net
, net_list
, exit_list
) {
7059 if (net
->dev_unreg_count
> 0) {
7060 unregistering
= true;
7069 finish_wait(&netdev_unregistering_wq
, &wait
);
7072 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
7074 /* At exit all network devices most be removed from a network
7075 * namespace. Do this in the reverse order of registration.
7076 * Do this across as many network namespaces as possible to
7077 * improve batching efficiency.
7079 struct net_device
*dev
;
7081 LIST_HEAD(dev_kill_list
);
7083 /* To prevent network device cleanup code from dereferencing
7084 * loopback devices or network devices that have been freed
7085 * wait here for all pending unregistrations to complete,
7086 * before unregistring the loopback device and allowing the
7087 * network namespace be freed.
7089 * The netdev todo list containing all network devices
7090 * unregistrations that happen in default_device_exit_batch
7091 * will run in the rtnl_unlock() at the end of
7092 * default_device_exit_batch.
7094 rtnl_lock_unregistering(net_list
);
7095 list_for_each_entry(net
, net_list
, exit_list
) {
7096 for_each_netdev_reverse(net
, dev
) {
7097 if (dev
->rtnl_link_ops
)
7098 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
7100 unregister_netdevice_queue(dev
, &dev_kill_list
);
7103 unregister_netdevice_many(&dev_kill_list
);
7107 static struct pernet_operations __net_initdata default_device_ops
= {
7108 .exit
= default_device_exit
,
7109 .exit_batch
= default_device_exit_batch
,
7113 * Initialize the DEV module. At boot time this walks the device list and
7114 * unhooks any devices that fail to initialise (normally hardware not
7115 * present) and leaves us with a valid list of present and active devices.
7120 * This is called single threaded during boot, so no need
7121 * to take the rtnl semaphore.
7123 static int __init
net_dev_init(void)
7125 int i
, rc
= -ENOMEM
;
7127 BUG_ON(!dev_boot_phase
);
7129 if (dev_proc_init())
7132 if (netdev_kobject_init())
7135 INIT_LIST_HEAD(&ptype_all
);
7136 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
7137 INIT_LIST_HEAD(&ptype_base
[i
]);
7139 INIT_LIST_HEAD(&offload_base
);
7141 if (register_pernet_subsys(&netdev_net_ops
))
7145 * Initialise the packet receive queues.
7148 for_each_possible_cpu(i
) {
7149 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
7151 skb_queue_head_init(&sd
->input_pkt_queue
);
7152 skb_queue_head_init(&sd
->process_queue
);
7153 INIT_LIST_HEAD(&sd
->poll_list
);
7154 sd
->output_queue_tailp
= &sd
->output_queue
;
7156 sd
->csd
.func
= rps_trigger_softirq
;
7161 sd
->backlog
.poll
= process_backlog
;
7162 sd
->backlog
.weight
= weight_p
;
7167 /* The loopback device is special if any other network devices
7168 * is present in a network namespace the loopback device must
7169 * be present. Since we now dynamically allocate and free the
7170 * loopback device ensure this invariant is maintained by
7171 * keeping the loopback device as the first device on the
7172 * list of network devices. Ensuring the loopback devices
7173 * is the first device that appears and the last network device
7176 if (register_pernet_device(&loopback_net_ops
))
7179 if (register_pernet_device(&default_device_ops
))
7182 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
7183 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
7185 hotcpu_notifier(dev_cpu_callback
, 0);
7192 subsys_initcall(net_dev_init
);