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 <linux/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/sched/mm.h>
85 #include <linux/mutex.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/notifier.h>
97 #include <linux/skbuff.h>
98 #include <linux/bpf.h>
99 #include <linux/bpf_trace.h>
100 #include <net/net_namespace.h>
101 #include <net/sock.h>
102 #include <net/busy_poll.h>
103 #include <linux/rtnetlink.h>
104 #include <linux/stat.h>
106 #include <net/dst_metadata.h>
107 #include <net/pkt_sched.h>
108 #include <net/pkt_cls.h>
109 #include <net/checksum.h>
110 #include <net/xfrm.h>
111 #include <linux/highmem.h>
112 #include <linux/init.h>
113 #include <linux/module.h>
114 #include <linux/netpoll.h>
115 #include <linux/rcupdate.h>
116 #include <linux/delay.h>
117 #include <net/iw_handler.h>
118 #include <asm/current.h>
119 #include <linux/audit.h>
120 #include <linux/dmaengine.h>
121 #include <linux/err.h>
122 #include <linux/ctype.h>
123 #include <linux/if_arp.h>
124 #include <linux/if_vlan.h>
125 #include <linux/ip.h>
127 #include <net/mpls.h>
128 #include <linux/ipv6.h>
129 #include <linux/in.h>
130 #include <linux/jhash.h>
131 #include <linux/random.h>
132 #include <trace/events/napi.h>
133 #include <trace/events/net.h>
134 #include <trace/events/skb.h>
135 #include <linux/pci.h>
136 #include <linux/inetdevice.h>
137 #include <linux/cpu_rmap.h>
138 #include <linux/static_key.h>
139 #include <linux/hashtable.h>
140 #include <linux/vmalloc.h>
141 #include <linux/if_macvlan.h>
142 #include <linux/errqueue.h>
143 #include <linux/hrtimer.h>
144 #include <linux/netfilter_ingress.h>
145 #include <linux/crash_dump.h>
146 #include <linux/sctp.h>
147 #include <net/udp_tunnel.h>
148 #include <linux/net_namespace.h>
150 #include "net-sysfs.h"
152 #define MAX_GRO_SKBS 8
154 /* This should be increased if a protocol with a bigger head is added. */
155 #define GRO_MAX_HEAD (MAX_HEADER + 128)
157 static DEFINE_SPINLOCK(ptype_lock
);
158 static DEFINE_SPINLOCK(offload_lock
);
159 struct list_head ptype_base
[PTYPE_HASH_SIZE
] __read_mostly
;
160 struct list_head ptype_all __read_mostly
; /* Taps */
161 static struct list_head offload_base __read_mostly
;
163 static int netif_rx_internal(struct sk_buff
*skb
);
164 static int call_netdevice_notifiers_info(unsigned long val
,
165 struct netdev_notifier_info
*info
);
166 static struct napi_struct
*napi_by_id(unsigned int napi_id
);
169 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
172 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
174 * Writers must hold the rtnl semaphore while they loop through the
175 * dev_base_head list, and hold dev_base_lock for writing when they do the
176 * actual updates. This allows pure readers to access the list even
177 * while a writer is preparing to update it.
179 * To put it another way, dev_base_lock is held for writing only to
180 * protect against pure readers; the rtnl semaphore provides the
181 * protection against other writers.
183 * See, for example usages, register_netdevice() and
184 * unregister_netdevice(), which must be called with the rtnl
187 DEFINE_RWLOCK(dev_base_lock
);
188 EXPORT_SYMBOL(dev_base_lock
);
190 static DEFINE_MUTEX(ifalias_mutex
);
192 /* protects napi_hash addition/deletion and napi_gen_id */
193 static DEFINE_SPINLOCK(napi_hash_lock
);
195 static unsigned int napi_gen_id
= NR_CPUS
;
196 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash
, 8);
198 static seqcount_t devnet_rename_seq
;
200 static inline void dev_base_seq_inc(struct net
*net
)
202 while (++net
->dev_base_seq
== 0)
206 static inline struct hlist_head
*dev_name_hash(struct net
*net
, const char *name
)
208 unsigned int hash
= full_name_hash(net
, name
, strnlen(name
, IFNAMSIZ
));
210 return &net
->dev_name_head
[hash_32(hash
, NETDEV_HASHBITS
)];
213 static inline struct hlist_head
*dev_index_hash(struct net
*net
, int ifindex
)
215 return &net
->dev_index_head
[ifindex
& (NETDEV_HASHENTRIES
- 1)];
218 static inline void rps_lock(struct softnet_data
*sd
)
221 spin_lock(&sd
->input_pkt_queue
.lock
);
225 static inline void rps_unlock(struct softnet_data
*sd
)
228 spin_unlock(&sd
->input_pkt_queue
.lock
);
232 /* Device list insertion */
233 static void list_netdevice(struct net_device
*dev
)
235 struct net
*net
= dev_net(dev
);
239 write_lock_bh(&dev_base_lock
);
240 list_add_tail_rcu(&dev
->dev_list
, &net
->dev_base_head
);
241 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
242 hlist_add_head_rcu(&dev
->index_hlist
,
243 dev_index_hash(net
, dev
->ifindex
));
244 write_unlock_bh(&dev_base_lock
);
246 dev_base_seq_inc(net
);
249 /* Device list removal
250 * caller must respect a RCU grace period before freeing/reusing dev
252 static void unlist_netdevice(struct net_device
*dev
)
256 /* Unlink dev from the device chain */
257 write_lock_bh(&dev_base_lock
);
258 list_del_rcu(&dev
->dev_list
);
259 hlist_del_rcu(&dev
->name_hlist
);
260 hlist_del_rcu(&dev
->index_hlist
);
261 write_unlock_bh(&dev_base_lock
);
263 dev_base_seq_inc(dev_net(dev
));
270 static RAW_NOTIFIER_HEAD(netdev_chain
);
273 * Device drivers call our routines to queue packets here. We empty the
274 * queue in the local softnet handler.
277 DEFINE_PER_CPU_ALIGNED(struct softnet_data
, softnet_data
);
278 EXPORT_PER_CPU_SYMBOL(softnet_data
);
280 #ifdef CONFIG_LOCKDEP
282 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
283 * according to dev->type
285 static const unsigned short netdev_lock_type
[] = {
286 ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
287 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
288 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
289 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
290 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
291 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
292 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
293 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
294 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
295 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
296 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
297 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
298 ARPHRD_FCFABRIC
, ARPHRD_IEEE80211
, ARPHRD_IEEE80211_PRISM
,
299 ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_PHONET
, ARPHRD_PHONET_PIPE
,
300 ARPHRD_IEEE802154
, ARPHRD_VOID
, ARPHRD_NONE
};
302 static const char *const netdev_lock_name
[] = {
303 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
304 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
305 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
306 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
307 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
308 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
309 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
310 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
311 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
312 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
313 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
314 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
315 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
316 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
317 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
319 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
320 static struct lock_class_key netdev_addr_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
322 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
326 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
327 if (netdev_lock_type
[i
] == dev_type
)
329 /* the last key is used by default */
330 return ARRAY_SIZE(netdev_lock_type
) - 1;
333 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
334 unsigned short dev_type
)
338 i
= netdev_lock_pos(dev_type
);
339 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
340 netdev_lock_name
[i
]);
343 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
347 i
= netdev_lock_pos(dev
->type
);
348 lockdep_set_class_and_name(&dev
->addr_list_lock
,
349 &netdev_addr_lock_key
[i
],
350 netdev_lock_name
[i
]);
353 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
354 unsigned short dev_type
)
357 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
362 /*******************************************************************************
364 * Protocol management and registration routines
366 *******************************************************************************/
370 * Add a protocol ID to the list. Now that the input handler is
371 * smarter we can dispense with all the messy stuff that used to be
374 * BEWARE!!! Protocol handlers, mangling input packets,
375 * MUST BE last in hash buckets and checking protocol handlers
376 * MUST start from promiscuous ptype_all chain in net_bh.
377 * It is true now, do not change it.
378 * Explanation follows: if protocol handler, mangling packet, will
379 * be the first on list, it is not able to sense, that packet
380 * is cloned and should be copied-on-write, so that it will
381 * change it and subsequent readers will get broken packet.
385 static inline struct list_head
*ptype_head(const struct packet_type
*pt
)
387 if (pt
->type
== htons(ETH_P_ALL
))
388 return pt
->dev
? &pt
->dev
->ptype_all
: &ptype_all
;
390 return pt
->dev
? &pt
->dev
->ptype_specific
:
391 &ptype_base
[ntohs(pt
->type
) & PTYPE_HASH_MASK
];
395 * dev_add_pack - add packet handler
396 * @pt: packet type declaration
398 * Add a protocol handler to the networking stack. The passed &packet_type
399 * is linked into kernel lists and may not be freed until it has been
400 * removed from the kernel lists.
402 * This call does not sleep therefore it can not
403 * guarantee all CPU's that are in middle of receiving packets
404 * will see the new packet type (until the next received packet).
407 void dev_add_pack(struct packet_type
*pt
)
409 struct list_head
*head
= ptype_head(pt
);
411 spin_lock(&ptype_lock
);
412 list_add_rcu(&pt
->list
, head
);
413 spin_unlock(&ptype_lock
);
415 EXPORT_SYMBOL(dev_add_pack
);
418 * __dev_remove_pack - remove packet handler
419 * @pt: packet type declaration
421 * Remove a protocol handler that was previously added to the kernel
422 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
423 * from the kernel lists and can be freed or reused once this function
426 * The packet type might still be in use by receivers
427 * and must not be freed until after all the CPU's have gone
428 * through a quiescent state.
430 void __dev_remove_pack(struct packet_type
*pt
)
432 struct list_head
*head
= ptype_head(pt
);
433 struct packet_type
*pt1
;
435 spin_lock(&ptype_lock
);
437 list_for_each_entry(pt1
, head
, list
) {
439 list_del_rcu(&pt
->list
);
444 pr_warn("dev_remove_pack: %p not found\n", pt
);
446 spin_unlock(&ptype_lock
);
448 EXPORT_SYMBOL(__dev_remove_pack
);
451 * dev_remove_pack - remove packet handler
452 * @pt: packet type declaration
454 * Remove a protocol handler that was previously added to the kernel
455 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
456 * from the kernel lists and can be freed or reused once this function
459 * This call sleeps to guarantee that no CPU is looking at the packet
462 void dev_remove_pack(struct packet_type
*pt
)
464 __dev_remove_pack(pt
);
468 EXPORT_SYMBOL(dev_remove_pack
);
472 * dev_add_offload - register offload handlers
473 * @po: protocol offload declaration
475 * Add protocol offload handlers to the networking stack. The passed
476 * &proto_offload is linked into kernel lists and may not be freed until
477 * it has been removed from the kernel lists.
479 * This call does not sleep therefore it can not
480 * guarantee all CPU's that are in middle of receiving packets
481 * will see the new offload handlers (until the next received packet).
483 void dev_add_offload(struct packet_offload
*po
)
485 struct packet_offload
*elem
;
487 spin_lock(&offload_lock
);
488 list_for_each_entry(elem
, &offload_base
, list
) {
489 if (po
->priority
< elem
->priority
)
492 list_add_rcu(&po
->list
, elem
->list
.prev
);
493 spin_unlock(&offload_lock
);
495 EXPORT_SYMBOL(dev_add_offload
);
498 * __dev_remove_offload - remove offload handler
499 * @po: packet offload declaration
501 * Remove a protocol offload handler that was previously added to the
502 * kernel offload handlers by dev_add_offload(). The passed &offload_type
503 * is removed from the kernel lists and can be freed or reused once this
506 * The packet type might still be in use by receivers
507 * and must not be freed until after all the CPU's have gone
508 * through a quiescent state.
510 static void __dev_remove_offload(struct packet_offload
*po
)
512 struct list_head
*head
= &offload_base
;
513 struct packet_offload
*po1
;
515 spin_lock(&offload_lock
);
517 list_for_each_entry(po1
, head
, list
) {
519 list_del_rcu(&po
->list
);
524 pr_warn("dev_remove_offload: %p not found\n", po
);
526 spin_unlock(&offload_lock
);
530 * dev_remove_offload - remove packet offload handler
531 * @po: packet offload declaration
533 * Remove a packet offload handler that was previously added to the kernel
534 * offload handlers by dev_add_offload(). The passed &offload_type is
535 * removed from the kernel lists and can be freed or reused once this
538 * This call sleeps to guarantee that no CPU is looking at the packet
541 void dev_remove_offload(struct packet_offload
*po
)
543 __dev_remove_offload(po
);
547 EXPORT_SYMBOL(dev_remove_offload
);
549 /******************************************************************************
551 * Device Boot-time Settings Routines
553 ******************************************************************************/
555 /* Boot time configuration table */
556 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
559 * netdev_boot_setup_add - add new setup entry
560 * @name: name of the device
561 * @map: configured settings for the device
563 * Adds new setup entry to the dev_boot_setup list. The function
564 * returns 0 on error and 1 on success. This is a generic routine to
567 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
569 struct netdev_boot_setup
*s
;
573 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
574 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
575 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
576 strlcpy(s
[i
].name
, name
, IFNAMSIZ
);
577 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
582 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
586 * netdev_boot_setup_check - check boot time settings
587 * @dev: the netdevice
589 * Check boot time settings for the device.
590 * The found settings are set for the device to be used
591 * later in the device probing.
592 * Returns 0 if no settings found, 1 if they are.
594 int netdev_boot_setup_check(struct net_device
*dev
)
596 struct netdev_boot_setup
*s
= dev_boot_setup
;
599 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
600 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
601 !strcmp(dev
->name
, s
[i
].name
)) {
602 dev
->irq
= s
[i
].map
.irq
;
603 dev
->base_addr
= s
[i
].map
.base_addr
;
604 dev
->mem_start
= s
[i
].map
.mem_start
;
605 dev
->mem_end
= s
[i
].map
.mem_end
;
611 EXPORT_SYMBOL(netdev_boot_setup_check
);
615 * netdev_boot_base - get address from boot time settings
616 * @prefix: prefix for network device
617 * @unit: id for network device
619 * Check boot time settings for the base address of device.
620 * The found settings are set for the device to be used
621 * later in the device probing.
622 * Returns 0 if no settings found.
624 unsigned long netdev_boot_base(const char *prefix
, int unit
)
626 const struct netdev_boot_setup
*s
= dev_boot_setup
;
630 sprintf(name
, "%s%d", prefix
, unit
);
633 * If device already registered then return base of 1
634 * to indicate not to probe for this interface
636 if (__dev_get_by_name(&init_net
, name
))
639 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
640 if (!strcmp(name
, s
[i
].name
))
641 return s
[i
].map
.base_addr
;
646 * Saves at boot time configured settings for any netdevice.
648 int __init
netdev_boot_setup(char *str
)
653 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
658 memset(&map
, 0, sizeof(map
));
662 map
.base_addr
= ints
[2];
664 map
.mem_start
= ints
[3];
666 map
.mem_end
= ints
[4];
668 /* Add new entry to the list */
669 return netdev_boot_setup_add(str
, &map
);
672 __setup("netdev=", netdev_boot_setup
);
674 /*******************************************************************************
676 * Device Interface Subroutines
678 *******************************************************************************/
681 * dev_get_iflink - get 'iflink' value of a interface
682 * @dev: targeted interface
684 * Indicates the ifindex the interface is linked to.
685 * Physical interfaces have the same 'ifindex' and 'iflink' values.
688 int dev_get_iflink(const struct net_device
*dev
)
690 if (dev
->netdev_ops
&& dev
->netdev_ops
->ndo_get_iflink
)
691 return dev
->netdev_ops
->ndo_get_iflink(dev
);
695 EXPORT_SYMBOL(dev_get_iflink
);
698 * dev_fill_metadata_dst - Retrieve tunnel egress information.
699 * @dev: targeted interface
702 * For better visibility of tunnel traffic OVS needs to retrieve
703 * egress tunnel information for a packet. Following API allows
704 * user to get this info.
706 int dev_fill_metadata_dst(struct net_device
*dev
, struct sk_buff
*skb
)
708 struct ip_tunnel_info
*info
;
710 if (!dev
->netdev_ops
|| !dev
->netdev_ops
->ndo_fill_metadata_dst
)
713 info
= skb_tunnel_info_unclone(skb
);
716 if (unlikely(!(info
->mode
& IP_TUNNEL_INFO_TX
)))
719 return dev
->netdev_ops
->ndo_fill_metadata_dst(dev
, skb
);
721 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst
);
724 * __dev_get_by_name - find a device by its name
725 * @net: the applicable net namespace
726 * @name: name to find
728 * Find an interface by name. Must be called under RTNL semaphore
729 * or @dev_base_lock. If the name is found a pointer to the device
730 * is returned. If the name is not found then %NULL is returned. The
731 * reference counters are not incremented so the caller must be
732 * careful with locks.
735 struct net_device
*__dev_get_by_name(struct net
*net
, const char *name
)
737 struct net_device
*dev
;
738 struct hlist_head
*head
= dev_name_hash(net
, name
);
740 hlist_for_each_entry(dev
, head
, name_hlist
)
741 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
746 EXPORT_SYMBOL(__dev_get_by_name
);
749 * dev_get_by_name_rcu - find a device by its name
750 * @net: the applicable net namespace
751 * @name: name to find
753 * Find an interface by name.
754 * If the name is found a pointer to the device is returned.
755 * If the name is not found then %NULL is returned.
756 * The reference counters are not incremented so the caller must be
757 * careful with locks. The caller must hold RCU lock.
760 struct net_device
*dev_get_by_name_rcu(struct net
*net
, const char *name
)
762 struct net_device
*dev
;
763 struct hlist_head
*head
= dev_name_hash(net
, name
);
765 hlist_for_each_entry_rcu(dev
, head
, name_hlist
)
766 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
771 EXPORT_SYMBOL(dev_get_by_name_rcu
);
774 * dev_get_by_name - find a device by its name
775 * @net: the applicable net namespace
776 * @name: name to find
778 * Find an interface by name. This can be called from any
779 * context and does its own locking. The returned handle has
780 * the usage count incremented and the caller must use dev_put() to
781 * release it when it is no longer needed. %NULL is returned if no
782 * matching device is found.
785 struct net_device
*dev_get_by_name(struct net
*net
, const char *name
)
787 struct net_device
*dev
;
790 dev
= dev_get_by_name_rcu(net
, name
);
796 EXPORT_SYMBOL(dev_get_by_name
);
799 * __dev_get_by_index - find a device by its ifindex
800 * @net: the applicable net namespace
801 * @ifindex: index of device
803 * Search for an interface by index. Returns %NULL if the device
804 * is not found or a pointer to the device. The device has not
805 * had its reference counter increased so the caller must be careful
806 * about locking. The caller must hold either the RTNL semaphore
810 struct net_device
*__dev_get_by_index(struct net
*net
, int ifindex
)
812 struct net_device
*dev
;
813 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
815 hlist_for_each_entry(dev
, head
, index_hlist
)
816 if (dev
->ifindex
== ifindex
)
821 EXPORT_SYMBOL(__dev_get_by_index
);
824 * dev_get_by_index_rcu - find a device by its ifindex
825 * @net: the applicable net namespace
826 * @ifindex: index of device
828 * Search for an interface by index. Returns %NULL if the device
829 * is not found or a pointer to the device. The device has not
830 * had its reference counter increased so the caller must be careful
831 * about locking. The caller must hold RCU lock.
834 struct net_device
*dev_get_by_index_rcu(struct net
*net
, int ifindex
)
836 struct net_device
*dev
;
837 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
839 hlist_for_each_entry_rcu(dev
, head
, index_hlist
)
840 if (dev
->ifindex
== ifindex
)
845 EXPORT_SYMBOL(dev_get_by_index_rcu
);
849 * dev_get_by_index - find a device by its ifindex
850 * @net: the applicable net namespace
851 * @ifindex: index of device
853 * Search for an interface by index. Returns NULL if the device
854 * is not found or a pointer to the device. The device returned has
855 * had a reference added and the pointer is safe until the user calls
856 * dev_put to indicate they have finished with it.
859 struct net_device
*dev_get_by_index(struct net
*net
, int ifindex
)
861 struct net_device
*dev
;
864 dev
= dev_get_by_index_rcu(net
, ifindex
);
870 EXPORT_SYMBOL(dev_get_by_index
);
873 * dev_get_by_napi_id - find a device by napi_id
874 * @napi_id: ID of the NAPI struct
876 * Search for an interface by NAPI ID. Returns %NULL if the device
877 * is not found or a pointer to the device. The device has not had
878 * its reference counter increased so the caller must be careful
879 * about locking. The caller must hold RCU lock.
882 struct net_device
*dev_get_by_napi_id(unsigned int napi_id
)
884 struct napi_struct
*napi
;
886 WARN_ON_ONCE(!rcu_read_lock_held());
888 if (napi_id
< MIN_NAPI_ID
)
891 napi
= napi_by_id(napi_id
);
893 return napi
? napi
->dev
: NULL
;
895 EXPORT_SYMBOL(dev_get_by_napi_id
);
898 * netdev_get_name - get a netdevice name, knowing its ifindex.
899 * @net: network namespace
900 * @name: a pointer to the buffer where the name will be stored.
901 * @ifindex: the ifindex of the interface to get the name from.
903 * The use of raw_seqcount_begin() and cond_resched() before
904 * retrying is required as we want to give the writers a chance
905 * to complete when CONFIG_PREEMPT is not set.
907 int netdev_get_name(struct net
*net
, char *name
, int ifindex
)
909 struct net_device
*dev
;
913 seq
= raw_seqcount_begin(&devnet_rename_seq
);
915 dev
= dev_get_by_index_rcu(net
, ifindex
);
921 strcpy(name
, dev
->name
);
923 if (read_seqcount_retry(&devnet_rename_seq
, seq
)) {
932 * dev_getbyhwaddr_rcu - find a device by its hardware address
933 * @net: the applicable net namespace
934 * @type: media type of device
935 * @ha: hardware address
937 * Search for an interface by MAC address. Returns NULL if the device
938 * is not found or a pointer to the device.
939 * The caller must hold RCU or RTNL.
940 * The returned device has not had its ref count increased
941 * and the caller must therefore be careful about locking
945 struct net_device
*dev_getbyhwaddr_rcu(struct net
*net
, unsigned short type
,
948 struct net_device
*dev
;
950 for_each_netdev_rcu(net
, dev
)
951 if (dev
->type
== type
&&
952 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
957 EXPORT_SYMBOL(dev_getbyhwaddr_rcu
);
959 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
961 struct net_device
*dev
;
964 for_each_netdev(net
, dev
)
965 if (dev
->type
== type
)
970 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
972 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
974 struct net_device
*dev
, *ret
= NULL
;
977 for_each_netdev_rcu(net
, dev
)
978 if (dev
->type
== type
) {
986 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
989 * __dev_get_by_flags - find any device with given flags
990 * @net: the applicable net namespace
991 * @if_flags: IFF_* values
992 * @mask: bitmask of bits in if_flags to check
994 * Search for any interface with the given flags. Returns NULL if a device
995 * is not found or a pointer to the device. Must be called inside
996 * rtnl_lock(), and result refcount is unchanged.
999 struct net_device
*__dev_get_by_flags(struct net
*net
, unsigned short if_flags
,
1000 unsigned short mask
)
1002 struct net_device
*dev
, *ret
;
1007 for_each_netdev(net
, dev
) {
1008 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
1015 EXPORT_SYMBOL(__dev_get_by_flags
);
1018 * dev_valid_name - check if name is okay for network device
1019 * @name: name string
1021 * Network device names need to be valid file names to
1022 * to allow sysfs to work. We also disallow any kind of
1025 bool dev_valid_name(const char *name
)
1029 if (strnlen(name
, IFNAMSIZ
) == IFNAMSIZ
)
1031 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
1035 if (*name
== '/' || *name
== ':' || isspace(*name
))
1041 EXPORT_SYMBOL(dev_valid_name
);
1044 * __dev_alloc_name - allocate a name for a device
1045 * @net: network namespace to allocate the device name in
1046 * @name: name format string
1047 * @buf: scratch buffer and result name string
1049 * Passed a format string - eg "lt%d" it will try and find a suitable
1050 * id. It scans list of devices to build up a free map, then chooses
1051 * the first empty slot. The caller must hold the dev_base or rtnl lock
1052 * while allocating the name and adding the device in order to avoid
1054 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1055 * Returns the number of the unit assigned or a negative errno code.
1058 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
1062 const int max_netdevices
= 8*PAGE_SIZE
;
1063 unsigned long *inuse
;
1064 struct net_device
*d
;
1066 if (!dev_valid_name(name
))
1069 p
= strchr(name
, '%');
1072 * Verify the string as this thing may have come from
1073 * the user. There must be either one "%d" and no other "%"
1076 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
1079 /* Use one page as a bit array of possible slots */
1080 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
1084 for_each_netdev(net
, d
) {
1085 if (!sscanf(d
->name
, name
, &i
))
1087 if (i
< 0 || i
>= max_netdevices
)
1090 /* avoid cases where sscanf is not exact inverse of printf */
1091 snprintf(buf
, IFNAMSIZ
, name
, i
);
1092 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
1096 i
= find_first_zero_bit(inuse
, max_netdevices
);
1097 free_page((unsigned long) inuse
);
1100 snprintf(buf
, IFNAMSIZ
, name
, i
);
1101 if (!__dev_get_by_name(net
, buf
))
1104 /* It is possible to run out of possible slots
1105 * when the name is long and there isn't enough space left
1106 * for the digits, or if all bits are used.
1111 static int dev_alloc_name_ns(struct net
*net
,
1112 struct net_device
*dev
,
1119 ret
= __dev_alloc_name(net
, name
, buf
);
1121 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
1126 * dev_alloc_name - allocate a name for a device
1128 * @name: name format string
1130 * Passed a format string - eg "lt%d" it will try and find a suitable
1131 * id. It scans list of devices to build up a free map, then chooses
1132 * the first empty slot. The caller must hold the dev_base or rtnl lock
1133 * while allocating the name and adding the device in order to avoid
1135 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1136 * Returns the number of the unit assigned or a negative errno code.
1139 int dev_alloc_name(struct net_device
*dev
, const char *name
)
1141 return dev_alloc_name_ns(dev_net(dev
), dev
, name
);
1143 EXPORT_SYMBOL(dev_alloc_name
);
1145 int dev_get_valid_name(struct net
*net
, struct net_device
*dev
,
1150 if (!dev_valid_name(name
))
1153 if (strchr(name
, '%'))
1154 return dev_alloc_name_ns(net
, dev
, name
);
1155 else if (__dev_get_by_name(net
, name
))
1157 else if (dev
->name
!= name
)
1158 strlcpy(dev
->name
, name
, IFNAMSIZ
);
1162 EXPORT_SYMBOL(dev_get_valid_name
);
1165 * dev_change_name - change name of a device
1167 * @newname: name (or format string) must be at least IFNAMSIZ
1169 * Change name of a device, can pass format strings "eth%d".
1172 int dev_change_name(struct net_device
*dev
, const char *newname
)
1174 unsigned char old_assign_type
;
1175 char oldname
[IFNAMSIZ
];
1181 BUG_ON(!dev_net(dev
));
1184 if (dev
->flags
& IFF_UP
)
1187 write_seqcount_begin(&devnet_rename_seq
);
1189 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0) {
1190 write_seqcount_end(&devnet_rename_seq
);
1194 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
1196 err
= dev_get_valid_name(net
, dev
, newname
);
1198 write_seqcount_end(&devnet_rename_seq
);
1202 if (oldname
[0] && !strchr(oldname
, '%'))
1203 netdev_info(dev
, "renamed from %s\n", oldname
);
1205 old_assign_type
= dev
->name_assign_type
;
1206 dev
->name_assign_type
= NET_NAME_RENAMED
;
1209 ret
= device_rename(&dev
->dev
, dev
->name
);
1211 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1212 dev
->name_assign_type
= old_assign_type
;
1213 write_seqcount_end(&devnet_rename_seq
);
1217 write_seqcount_end(&devnet_rename_seq
);
1219 netdev_adjacent_rename_links(dev
, oldname
);
1221 write_lock_bh(&dev_base_lock
);
1222 hlist_del_rcu(&dev
->name_hlist
);
1223 write_unlock_bh(&dev_base_lock
);
1227 write_lock_bh(&dev_base_lock
);
1228 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
1229 write_unlock_bh(&dev_base_lock
);
1231 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
1232 ret
= notifier_to_errno(ret
);
1235 /* err >= 0 after dev_alloc_name() or stores the first errno */
1238 write_seqcount_begin(&devnet_rename_seq
);
1239 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1240 memcpy(oldname
, newname
, IFNAMSIZ
);
1241 dev
->name_assign_type
= old_assign_type
;
1242 old_assign_type
= NET_NAME_RENAMED
;
1245 pr_err("%s: name change rollback failed: %d\n",
1254 * dev_set_alias - change ifalias of a device
1256 * @alias: name up to IFALIASZ
1257 * @len: limit of bytes to copy from info
1259 * Set ifalias for a device,
1261 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
1263 struct dev_ifalias
*new_alias
= NULL
;
1265 if (len
>= IFALIASZ
)
1269 new_alias
= kmalloc(sizeof(*new_alias
) + len
+ 1, GFP_KERNEL
);
1273 memcpy(new_alias
->ifalias
, alias
, len
);
1274 new_alias
->ifalias
[len
] = 0;
1277 mutex_lock(&ifalias_mutex
);
1278 rcu_swap_protected(dev
->ifalias
, new_alias
,
1279 mutex_is_locked(&ifalias_mutex
));
1280 mutex_unlock(&ifalias_mutex
);
1283 kfree_rcu(new_alias
, rcuhead
);
1287 EXPORT_SYMBOL(dev_set_alias
);
1290 * dev_get_alias - get ifalias of a device
1292 * @name: buffer to store name of ifalias
1293 * @len: size of buffer
1295 * get ifalias for a device. Caller must make sure dev cannot go
1296 * away, e.g. rcu read lock or own a reference count to device.
1298 int dev_get_alias(const struct net_device
*dev
, char *name
, size_t len
)
1300 const struct dev_ifalias
*alias
;
1304 alias
= rcu_dereference(dev
->ifalias
);
1306 ret
= snprintf(name
, len
, "%s", alias
->ifalias
);
1313 * netdev_features_change - device changes features
1314 * @dev: device to cause notification
1316 * Called to indicate a device has changed features.
1318 void netdev_features_change(struct net_device
*dev
)
1320 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1322 EXPORT_SYMBOL(netdev_features_change
);
1325 * netdev_state_change - device changes state
1326 * @dev: device to cause notification
1328 * Called to indicate a device has changed state. This function calls
1329 * the notifier chains for netdev_chain and sends a NEWLINK message
1330 * to the routing socket.
1332 void netdev_state_change(struct net_device
*dev
)
1334 if (dev
->flags
& IFF_UP
) {
1335 struct netdev_notifier_change_info change_info
= {
1339 call_netdevice_notifiers_info(NETDEV_CHANGE
,
1341 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0, GFP_KERNEL
);
1344 EXPORT_SYMBOL(netdev_state_change
);
1347 * netdev_notify_peers - notify network peers about existence of @dev
1348 * @dev: network device
1350 * Generate traffic such that interested network peers are aware of
1351 * @dev, such as by generating a gratuitous ARP. This may be used when
1352 * a device wants to inform the rest of the network about some sort of
1353 * reconfiguration such as a failover event or virtual machine
1356 void netdev_notify_peers(struct net_device
*dev
)
1359 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
, dev
);
1360 call_netdevice_notifiers(NETDEV_RESEND_IGMP
, dev
);
1363 EXPORT_SYMBOL(netdev_notify_peers
);
1365 static int __dev_open(struct net_device
*dev
)
1367 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1372 if (!netif_device_present(dev
))
1375 /* Block netpoll from trying to do any rx path servicing.
1376 * If we don't do this there is a chance ndo_poll_controller
1377 * or ndo_poll may be running while we open the device
1379 netpoll_poll_disable(dev
);
1381 ret
= call_netdevice_notifiers(NETDEV_PRE_UP
, dev
);
1382 ret
= notifier_to_errno(ret
);
1386 set_bit(__LINK_STATE_START
, &dev
->state
);
1388 if (ops
->ndo_validate_addr
)
1389 ret
= ops
->ndo_validate_addr(dev
);
1391 if (!ret
&& ops
->ndo_open
)
1392 ret
= ops
->ndo_open(dev
);
1394 netpoll_poll_enable(dev
);
1397 clear_bit(__LINK_STATE_START
, &dev
->state
);
1399 dev
->flags
|= IFF_UP
;
1400 dev_set_rx_mode(dev
);
1402 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
1409 * dev_open - prepare an interface for use.
1410 * @dev: device to open
1412 * Takes a device from down to up state. The device's private open
1413 * function is invoked and then the multicast lists are loaded. Finally
1414 * the device is moved into the up state and a %NETDEV_UP message is
1415 * sent to the netdev notifier chain.
1417 * Calling this function on an active interface is a nop. On a failure
1418 * a negative errno code is returned.
1420 int dev_open(struct net_device
*dev
)
1424 if (dev
->flags
& IFF_UP
)
1427 ret
= __dev_open(dev
);
1431 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
, GFP_KERNEL
);
1432 call_netdevice_notifiers(NETDEV_UP
, dev
);
1436 EXPORT_SYMBOL(dev_open
);
1438 static void __dev_close_many(struct list_head
*head
)
1440 struct net_device
*dev
;
1445 list_for_each_entry(dev
, head
, close_list
) {
1446 /* Temporarily disable netpoll until the interface is down */
1447 netpoll_poll_disable(dev
);
1449 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1451 clear_bit(__LINK_STATE_START
, &dev
->state
);
1453 /* Synchronize to scheduled poll. We cannot touch poll list, it
1454 * can be even on different cpu. So just clear netif_running().
1456 * dev->stop() will invoke napi_disable() on all of it's
1457 * napi_struct instances on this device.
1459 smp_mb__after_atomic(); /* Commit netif_running(). */
1462 dev_deactivate_many(head
);
1464 list_for_each_entry(dev
, head
, close_list
) {
1465 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1468 * Call the device specific close. This cannot fail.
1469 * Only if device is UP
1471 * We allow it to be called even after a DETACH hot-plug
1477 dev
->flags
&= ~IFF_UP
;
1478 netpoll_poll_enable(dev
);
1482 static void __dev_close(struct net_device
*dev
)
1486 list_add(&dev
->close_list
, &single
);
1487 __dev_close_many(&single
);
1491 void dev_close_many(struct list_head
*head
, bool unlink
)
1493 struct net_device
*dev
, *tmp
;
1495 /* Remove the devices that don't need to be closed */
1496 list_for_each_entry_safe(dev
, tmp
, head
, close_list
)
1497 if (!(dev
->flags
& IFF_UP
))
1498 list_del_init(&dev
->close_list
);
1500 __dev_close_many(head
);
1502 list_for_each_entry_safe(dev
, tmp
, head
, close_list
) {
1503 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
, GFP_KERNEL
);
1504 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1506 list_del_init(&dev
->close_list
);
1509 EXPORT_SYMBOL(dev_close_many
);
1512 * dev_close - shutdown an interface.
1513 * @dev: device to shutdown
1515 * This function moves an active device into down state. A
1516 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1517 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1520 void dev_close(struct net_device
*dev
)
1522 if (dev
->flags
& IFF_UP
) {
1525 list_add(&dev
->close_list
, &single
);
1526 dev_close_many(&single
, true);
1530 EXPORT_SYMBOL(dev_close
);
1534 * dev_disable_lro - disable Large Receive Offload on a device
1537 * Disable Large Receive Offload (LRO) on a net device. Must be
1538 * called under RTNL. This is needed if received packets may be
1539 * forwarded to another interface.
1541 void dev_disable_lro(struct net_device
*dev
)
1543 struct net_device
*lower_dev
;
1544 struct list_head
*iter
;
1546 dev
->wanted_features
&= ~NETIF_F_LRO
;
1547 netdev_update_features(dev
);
1549 if (unlikely(dev
->features
& NETIF_F_LRO
))
1550 netdev_WARN(dev
, "failed to disable LRO!\n");
1552 netdev_for_each_lower_dev(dev
, lower_dev
, iter
)
1553 dev_disable_lro(lower_dev
);
1555 EXPORT_SYMBOL(dev_disable_lro
);
1558 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1561 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be
1562 * called under RTNL. This is needed if Generic XDP is installed on
1565 static void dev_disable_gro_hw(struct net_device
*dev
)
1567 dev
->wanted_features
&= ~NETIF_F_GRO_HW
;
1568 netdev_update_features(dev
);
1570 if (unlikely(dev
->features
& NETIF_F_GRO_HW
))
1571 netdev_WARN(dev
, "failed to disable GRO_HW!\n");
1574 const char *netdev_cmd_to_name(enum netdev_cmd cmd
)
1577 case NETDEV_##val: \
1578 return "NETDEV_" __stringify(val);
1580 N(UP
) N(DOWN
) N(REBOOT
) N(CHANGE
) N(REGISTER
) N(UNREGISTER
)
1581 N(CHANGEMTU
) N(CHANGEADDR
) N(GOING_DOWN
) N(CHANGENAME
) N(FEAT_CHANGE
)
1582 N(BONDING_FAILOVER
) N(PRE_UP
) N(PRE_TYPE_CHANGE
) N(POST_TYPE_CHANGE
)
1583 N(POST_INIT
) N(RELEASE
) N(NOTIFY_PEERS
) N(JOIN
) N(CHANGEUPPER
)
1584 N(RESEND_IGMP
) N(PRECHANGEMTU
) N(CHANGEINFODATA
) N(BONDING_INFO
)
1585 N(PRECHANGEUPPER
) N(CHANGELOWERSTATE
) N(UDP_TUNNEL_PUSH_INFO
)
1586 N(UDP_TUNNEL_DROP_INFO
) N(CHANGE_TX_QUEUE_LEN
)
1587 N(CVLAN_FILTER_PUSH_INFO
) N(CVLAN_FILTER_DROP_INFO
)
1588 N(SVLAN_FILTER_PUSH_INFO
) N(SVLAN_FILTER_DROP_INFO
)
1591 return "UNKNOWN_NETDEV_EVENT";
1593 EXPORT_SYMBOL_GPL(netdev_cmd_to_name
);
1595 static int call_netdevice_notifier(struct notifier_block
*nb
, unsigned long val
,
1596 struct net_device
*dev
)
1598 struct netdev_notifier_info info
= {
1602 return nb
->notifier_call(nb
, val
, &info
);
1605 static int dev_boot_phase
= 1;
1608 * register_netdevice_notifier - register a network notifier block
1611 * Register a notifier to be called when network device events occur.
1612 * The notifier passed is linked into the kernel structures and must
1613 * not be reused until it has been unregistered. A negative errno code
1614 * is returned on a failure.
1616 * When registered all registration and up events are replayed
1617 * to the new notifier to allow device to have a race free
1618 * view of the network device list.
1621 int register_netdevice_notifier(struct notifier_block
*nb
)
1623 struct net_device
*dev
;
1624 struct net_device
*last
;
1628 /* Close race with setup_net() and cleanup_net() */
1629 down_write(&pernet_ops_rwsem
);
1631 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1637 for_each_netdev(net
, dev
) {
1638 err
= call_netdevice_notifier(nb
, NETDEV_REGISTER
, dev
);
1639 err
= notifier_to_errno(err
);
1643 if (!(dev
->flags
& IFF_UP
))
1646 call_netdevice_notifier(nb
, NETDEV_UP
, dev
);
1652 up_write(&pernet_ops_rwsem
);
1658 for_each_netdev(net
, dev
) {
1662 if (dev
->flags
& IFF_UP
) {
1663 call_netdevice_notifier(nb
, NETDEV_GOING_DOWN
,
1665 call_netdevice_notifier(nb
, NETDEV_DOWN
, dev
);
1667 call_netdevice_notifier(nb
, NETDEV_UNREGISTER
, dev
);
1672 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1675 EXPORT_SYMBOL(register_netdevice_notifier
);
1678 * unregister_netdevice_notifier - unregister a network notifier block
1681 * Unregister a notifier previously registered by
1682 * register_netdevice_notifier(). The notifier is unlinked into the
1683 * kernel structures and may then be reused. A negative errno code
1684 * is returned on a failure.
1686 * After unregistering unregister and down device events are synthesized
1687 * for all devices on the device list to the removed notifier to remove
1688 * the need for special case cleanup code.
1691 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1693 struct net_device
*dev
;
1697 /* Close race with setup_net() and cleanup_net() */
1698 down_write(&pernet_ops_rwsem
);
1700 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1705 for_each_netdev(net
, dev
) {
1706 if (dev
->flags
& IFF_UP
) {
1707 call_netdevice_notifier(nb
, NETDEV_GOING_DOWN
,
1709 call_netdevice_notifier(nb
, NETDEV_DOWN
, dev
);
1711 call_netdevice_notifier(nb
, NETDEV_UNREGISTER
, dev
);
1716 up_write(&pernet_ops_rwsem
);
1719 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1722 * call_netdevice_notifiers_info - call all network notifier blocks
1723 * @val: value passed unmodified to notifier function
1724 * @info: notifier information data
1726 * Call all network notifier blocks. Parameters and return value
1727 * are as for raw_notifier_call_chain().
1730 static int call_netdevice_notifiers_info(unsigned long val
,
1731 struct netdev_notifier_info
*info
)
1734 return raw_notifier_call_chain(&netdev_chain
, val
, info
);
1738 * call_netdevice_notifiers - call all network notifier blocks
1739 * @val: value passed unmodified to notifier function
1740 * @dev: net_device pointer passed unmodified to notifier function
1742 * Call all network notifier blocks. Parameters and return value
1743 * are as for raw_notifier_call_chain().
1746 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1748 struct netdev_notifier_info info
= {
1752 return call_netdevice_notifiers_info(val
, &info
);
1754 EXPORT_SYMBOL(call_netdevice_notifiers
);
1756 #ifdef CONFIG_NET_INGRESS
1757 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key
);
1759 void net_inc_ingress_queue(void)
1761 static_branch_inc(&ingress_needed_key
);
1763 EXPORT_SYMBOL_GPL(net_inc_ingress_queue
);
1765 void net_dec_ingress_queue(void)
1767 static_branch_dec(&ingress_needed_key
);
1769 EXPORT_SYMBOL_GPL(net_dec_ingress_queue
);
1772 #ifdef CONFIG_NET_EGRESS
1773 static DEFINE_STATIC_KEY_FALSE(egress_needed_key
);
1775 void net_inc_egress_queue(void)
1777 static_branch_inc(&egress_needed_key
);
1779 EXPORT_SYMBOL_GPL(net_inc_egress_queue
);
1781 void net_dec_egress_queue(void)
1783 static_branch_dec(&egress_needed_key
);
1785 EXPORT_SYMBOL_GPL(net_dec_egress_queue
);
1788 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key
);
1789 #ifdef HAVE_JUMP_LABEL
1790 static atomic_t netstamp_needed_deferred
;
1791 static atomic_t netstamp_wanted
;
1792 static void netstamp_clear(struct work_struct
*work
)
1794 int deferred
= atomic_xchg(&netstamp_needed_deferred
, 0);
1797 wanted
= atomic_add_return(deferred
, &netstamp_wanted
);
1799 static_branch_enable(&netstamp_needed_key
);
1801 static_branch_disable(&netstamp_needed_key
);
1803 static DECLARE_WORK(netstamp_work
, netstamp_clear
);
1806 void net_enable_timestamp(void)
1808 #ifdef HAVE_JUMP_LABEL
1812 wanted
= atomic_read(&netstamp_wanted
);
1815 if (atomic_cmpxchg(&netstamp_wanted
, wanted
, wanted
+ 1) == wanted
)
1818 atomic_inc(&netstamp_needed_deferred
);
1819 schedule_work(&netstamp_work
);
1821 static_branch_inc(&netstamp_needed_key
);
1824 EXPORT_SYMBOL(net_enable_timestamp
);
1826 void net_disable_timestamp(void)
1828 #ifdef HAVE_JUMP_LABEL
1832 wanted
= atomic_read(&netstamp_wanted
);
1835 if (atomic_cmpxchg(&netstamp_wanted
, wanted
, wanted
- 1) == wanted
)
1838 atomic_dec(&netstamp_needed_deferred
);
1839 schedule_work(&netstamp_work
);
1841 static_branch_dec(&netstamp_needed_key
);
1844 EXPORT_SYMBOL(net_disable_timestamp
);
1846 static inline void net_timestamp_set(struct sk_buff
*skb
)
1849 if (static_branch_unlikely(&netstamp_needed_key
))
1850 __net_timestamp(skb
);
1853 #define net_timestamp_check(COND, SKB) \
1854 if (static_branch_unlikely(&netstamp_needed_key)) { \
1855 if ((COND) && !(SKB)->tstamp) \
1856 __net_timestamp(SKB); \
1859 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
1863 if (!(dev
->flags
& IFF_UP
))
1866 len
= dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
;
1867 if (skb
->len
<= len
)
1870 /* if TSO is enabled, we don't care about the length as the packet
1871 * could be forwarded without being segmented before
1873 if (skb_is_gso(skb
))
1878 EXPORT_SYMBOL_GPL(is_skb_forwardable
);
1880 int __dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1882 int ret
= ____dev_forward_skb(dev
, skb
);
1885 skb
->protocol
= eth_type_trans(skb
, dev
);
1886 skb_postpull_rcsum(skb
, eth_hdr(skb
), ETH_HLEN
);
1891 EXPORT_SYMBOL_GPL(__dev_forward_skb
);
1894 * dev_forward_skb - loopback an skb to another netif
1896 * @dev: destination network device
1897 * @skb: buffer to forward
1900 * NET_RX_SUCCESS (no congestion)
1901 * NET_RX_DROP (packet was dropped, but freed)
1903 * dev_forward_skb can be used for injecting an skb from the
1904 * start_xmit function of one device into the receive queue
1905 * of another device.
1907 * The receiving device may be in another namespace, so
1908 * we have to clear all information in the skb that could
1909 * impact namespace isolation.
1911 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1913 return __dev_forward_skb(dev
, skb
) ?: netif_rx_internal(skb
);
1915 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1917 static inline int deliver_skb(struct sk_buff
*skb
,
1918 struct packet_type
*pt_prev
,
1919 struct net_device
*orig_dev
)
1921 if (unlikely(skb_orphan_frags_rx(skb
, GFP_ATOMIC
)))
1923 refcount_inc(&skb
->users
);
1924 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1927 static inline void deliver_ptype_list_skb(struct sk_buff
*skb
,
1928 struct packet_type
**pt
,
1929 struct net_device
*orig_dev
,
1931 struct list_head
*ptype_list
)
1933 struct packet_type
*ptype
, *pt_prev
= *pt
;
1935 list_for_each_entry_rcu(ptype
, ptype_list
, list
) {
1936 if (ptype
->type
!= type
)
1939 deliver_skb(skb
, pt_prev
, orig_dev
);
1945 static inline bool skb_loop_sk(struct packet_type
*ptype
, struct sk_buff
*skb
)
1947 if (!ptype
->af_packet_priv
|| !skb
->sk
)
1950 if (ptype
->id_match
)
1951 return ptype
->id_match(ptype
, skb
->sk
);
1952 else if ((struct sock
*)ptype
->af_packet_priv
== skb
->sk
)
1959 * Support routine. Sends outgoing frames to any network
1960 * taps currently in use.
1963 void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1965 struct packet_type
*ptype
;
1966 struct sk_buff
*skb2
= NULL
;
1967 struct packet_type
*pt_prev
= NULL
;
1968 struct list_head
*ptype_list
= &ptype_all
;
1972 list_for_each_entry_rcu(ptype
, ptype_list
, list
) {
1973 /* Never send packets back to the socket
1974 * they originated from - MvS (miquels@drinkel.ow.org)
1976 if (skb_loop_sk(ptype
, skb
))
1980 deliver_skb(skb2
, pt_prev
, skb
->dev
);
1985 /* need to clone skb, done only once */
1986 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1990 net_timestamp_set(skb2
);
1992 /* skb->nh should be correctly
1993 * set by sender, so that the second statement is
1994 * just protection against buggy protocols.
1996 skb_reset_mac_header(skb2
);
1998 if (skb_network_header(skb2
) < skb2
->data
||
1999 skb_network_header(skb2
) > skb_tail_pointer(skb2
)) {
2000 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2001 ntohs(skb2
->protocol
),
2003 skb_reset_network_header(skb2
);
2006 skb2
->transport_header
= skb2
->network_header
;
2007 skb2
->pkt_type
= PACKET_OUTGOING
;
2011 if (ptype_list
== &ptype_all
) {
2012 ptype_list
= &dev
->ptype_all
;
2017 if (!skb_orphan_frags_rx(skb2
, GFP_ATOMIC
))
2018 pt_prev
->func(skb2
, skb
->dev
, pt_prev
, skb
->dev
);
2024 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit
);
2027 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2028 * @dev: Network device
2029 * @txq: number of queues available
2031 * If real_num_tx_queues is changed the tc mappings may no longer be
2032 * valid. To resolve this verify the tc mapping remains valid and if
2033 * not NULL the mapping. With no priorities mapping to this
2034 * offset/count pair it will no longer be used. In the worst case TC0
2035 * is invalid nothing can be done so disable priority mappings. If is
2036 * expected that drivers will fix this mapping if they can before
2037 * calling netif_set_real_num_tx_queues.
2039 static void netif_setup_tc(struct net_device
*dev
, unsigned int txq
)
2042 struct netdev_tc_txq
*tc
= &dev
->tc_to_txq
[0];
2044 /* If TC0 is invalidated disable TC mapping */
2045 if (tc
->offset
+ tc
->count
> txq
) {
2046 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2051 /* Invalidated prio to tc mappings set to TC0 */
2052 for (i
= 1; i
< TC_BITMASK
+ 1; i
++) {
2053 int q
= netdev_get_prio_tc_map(dev
, i
);
2055 tc
= &dev
->tc_to_txq
[q
];
2056 if (tc
->offset
+ tc
->count
> txq
) {
2057 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
2059 netdev_set_prio_tc_map(dev
, i
, 0);
2064 int netdev_txq_to_tc(struct net_device
*dev
, unsigned int txq
)
2067 struct netdev_tc_txq
*tc
= &dev
->tc_to_txq
[0];
2070 for (i
= 0; i
< TC_MAX_QUEUE
; i
++, tc
++) {
2071 if ((txq
- tc
->offset
) < tc
->count
)
2080 EXPORT_SYMBOL(netdev_txq_to_tc
);
2083 struct static_key xps_needed __read_mostly
;
2084 EXPORT_SYMBOL(xps_needed
);
2085 struct static_key xps_rxqs_needed __read_mostly
;
2086 EXPORT_SYMBOL(xps_rxqs_needed
);
2087 static DEFINE_MUTEX(xps_map_mutex
);
2088 #define xmap_dereference(P) \
2089 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2091 static bool remove_xps_queue(struct xps_dev_maps
*dev_maps
,
2094 struct xps_map
*map
= NULL
;
2098 map
= xmap_dereference(dev_maps
->attr_map
[tci
]);
2102 for (pos
= map
->len
; pos
--;) {
2103 if (map
->queues
[pos
] != index
)
2107 map
->queues
[pos
] = map
->queues
[--map
->len
];
2111 RCU_INIT_POINTER(dev_maps
->attr_map
[tci
], NULL
);
2112 kfree_rcu(map
, rcu
);
2119 static bool remove_xps_queue_cpu(struct net_device
*dev
,
2120 struct xps_dev_maps
*dev_maps
,
2121 int cpu
, u16 offset
, u16 count
)
2123 int num_tc
= dev
->num_tc
? : 1;
2124 bool active
= false;
2127 for (tci
= cpu
* num_tc
; num_tc
--; tci
++) {
2130 for (i
= count
, j
= offset
; i
--; j
++) {
2131 if (!remove_xps_queue(dev_maps
, tci
, j
))
2141 static void clean_xps_maps(struct net_device
*dev
, const unsigned long *mask
,
2142 struct xps_dev_maps
*dev_maps
, unsigned int nr_ids
,
2143 u16 offset
, u16 count
, bool is_rxqs_map
)
2145 bool active
= false;
2148 for (j
= -1; j
= netif_attrmask_next(j
, mask
, nr_ids
),
2150 active
|= remove_xps_queue_cpu(dev
, dev_maps
, j
, offset
,
2154 RCU_INIT_POINTER(dev
->xps_rxqs_map
, NULL
);
2156 RCU_INIT_POINTER(dev
->xps_cpus_map
, NULL
);
2158 for (i
= offset
+ (count
- 1); count
--; i
--)
2159 netdev_queue_numa_node_write(
2160 netdev_get_tx_queue(dev
, i
),
2163 kfree_rcu(dev_maps
, rcu
);
2167 static void netif_reset_xps_queues(struct net_device
*dev
, u16 offset
,
2170 const unsigned long *possible_mask
= NULL
;
2171 struct xps_dev_maps
*dev_maps
;
2172 unsigned int nr_ids
;
2174 if (!static_key_false(&xps_needed
))
2177 mutex_lock(&xps_map_mutex
);
2179 if (static_key_false(&xps_rxqs_needed
)) {
2180 dev_maps
= xmap_dereference(dev
->xps_rxqs_map
);
2182 nr_ids
= dev
->num_rx_queues
;
2183 clean_xps_maps(dev
, possible_mask
, dev_maps
, nr_ids
,
2184 offset
, count
, true);
2188 dev_maps
= xmap_dereference(dev
->xps_cpus_map
);
2192 if (num_possible_cpus() > 1)
2193 possible_mask
= cpumask_bits(cpu_possible_mask
);
2194 nr_ids
= nr_cpu_ids
;
2195 clean_xps_maps(dev
, possible_mask
, dev_maps
, nr_ids
, offset
, count
,
2199 if (static_key_enabled(&xps_rxqs_needed
))
2200 static_key_slow_dec(&xps_rxqs_needed
);
2202 static_key_slow_dec(&xps_needed
);
2203 mutex_unlock(&xps_map_mutex
);
2206 static void netif_reset_xps_queues_gt(struct net_device
*dev
, u16 index
)
2208 netif_reset_xps_queues(dev
, index
, dev
->num_tx_queues
- index
);
2211 static struct xps_map
*expand_xps_map(struct xps_map
*map
, int attr_index
,
2212 u16 index
, bool is_rxqs_map
)
2214 struct xps_map
*new_map
;
2215 int alloc_len
= XPS_MIN_MAP_ALLOC
;
2218 for (pos
= 0; map
&& pos
< map
->len
; pos
++) {
2219 if (map
->queues
[pos
] != index
)
2224 /* Need to add tx-queue to this CPU's/rx-queue's existing map */
2226 if (pos
< map
->alloc_len
)
2229 alloc_len
= map
->alloc_len
* 2;
2232 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2236 new_map
= kzalloc(XPS_MAP_SIZE(alloc_len
), GFP_KERNEL
);
2238 new_map
= kzalloc_node(XPS_MAP_SIZE(alloc_len
), GFP_KERNEL
,
2239 cpu_to_node(attr_index
));
2243 for (i
= 0; i
< pos
; i
++)
2244 new_map
->queues
[i
] = map
->queues
[i
];
2245 new_map
->alloc_len
= alloc_len
;
2251 int __netif_set_xps_queue(struct net_device
*dev
, const unsigned long *mask
,
2252 u16 index
, bool is_rxqs_map
)
2254 const unsigned long *online_mask
= NULL
, *possible_mask
= NULL
;
2255 struct xps_dev_maps
*dev_maps
, *new_dev_maps
= NULL
;
2256 int i
, j
, tci
, numa_node_id
= -2;
2257 int maps_sz
, num_tc
= 1, tc
= 0;
2258 struct xps_map
*map
, *new_map
;
2259 bool active
= false;
2260 unsigned int nr_ids
;
2263 num_tc
= dev
->num_tc
;
2264 tc
= netdev_txq_to_tc(dev
, index
);
2269 mutex_lock(&xps_map_mutex
);
2271 maps_sz
= XPS_RXQ_DEV_MAPS_SIZE(num_tc
, dev
->num_rx_queues
);
2272 dev_maps
= xmap_dereference(dev
->xps_rxqs_map
);
2273 nr_ids
= dev
->num_rx_queues
;
2275 maps_sz
= XPS_CPU_DEV_MAPS_SIZE(num_tc
);
2276 if (num_possible_cpus() > 1) {
2277 online_mask
= cpumask_bits(cpu_online_mask
);
2278 possible_mask
= cpumask_bits(cpu_possible_mask
);
2280 dev_maps
= xmap_dereference(dev
->xps_cpus_map
);
2281 nr_ids
= nr_cpu_ids
;
2284 if (maps_sz
< L1_CACHE_BYTES
)
2285 maps_sz
= L1_CACHE_BYTES
;
2287 /* allocate memory for queue storage */
2288 for (j
= -1; j
= netif_attrmask_next_and(j
, online_mask
, mask
, nr_ids
),
2291 new_dev_maps
= kzalloc(maps_sz
, GFP_KERNEL
);
2292 if (!new_dev_maps
) {
2293 mutex_unlock(&xps_map_mutex
);
2297 tci
= j
* num_tc
+ tc
;
2298 map
= dev_maps
? xmap_dereference(dev_maps
->attr_map
[tci
]) :
2301 map
= expand_xps_map(map
, j
, index
, is_rxqs_map
);
2305 RCU_INIT_POINTER(new_dev_maps
->attr_map
[tci
], map
);
2309 goto out_no_new_maps
;
2311 static_key_slow_inc(&xps_needed
);
2313 static_key_slow_inc(&xps_rxqs_needed
);
2315 for (j
= -1; j
= netif_attrmask_next(j
, possible_mask
, nr_ids
),
2317 /* copy maps belonging to foreign traffic classes */
2318 for (i
= tc
, tci
= j
* num_tc
; dev_maps
&& i
--; tci
++) {
2319 /* fill in the new device map from the old device map */
2320 map
= xmap_dereference(dev_maps
->attr_map
[tci
]);
2321 RCU_INIT_POINTER(new_dev_maps
->attr_map
[tci
], map
);
2324 /* We need to explicitly update tci as prevous loop
2325 * could break out early if dev_maps is NULL.
2327 tci
= j
* num_tc
+ tc
;
2329 if (netif_attr_test_mask(j
, mask
, nr_ids
) &&
2330 netif_attr_test_online(j
, online_mask
, nr_ids
)) {
2331 /* add tx-queue to CPU/rx-queue maps */
2334 map
= xmap_dereference(new_dev_maps
->attr_map
[tci
]);
2335 while ((pos
< map
->len
) && (map
->queues
[pos
] != index
))
2338 if (pos
== map
->len
)
2339 map
->queues
[map
->len
++] = index
;
2342 if (numa_node_id
== -2)
2343 numa_node_id
= cpu_to_node(j
);
2344 else if (numa_node_id
!= cpu_to_node(j
))
2348 } else if (dev_maps
) {
2349 /* fill in the new device map from the old device map */
2350 map
= xmap_dereference(dev_maps
->attr_map
[tci
]);
2351 RCU_INIT_POINTER(new_dev_maps
->attr_map
[tci
], map
);
2354 /* copy maps belonging to foreign traffic classes */
2355 for (i
= num_tc
- tc
, tci
++; dev_maps
&& --i
; tci
++) {
2356 /* fill in the new device map from the old device map */
2357 map
= xmap_dereference(dev_maps
->attr_map
[tci
]);
2358 RCU_INIT_POINTER(new_dev_maps
->attr_map
[tci
], map
);
2363 rcu_assign_pointer(dev
->xps_rxqs_map
, new_dev_maps
);
2365 rcu_assign_pointer(dev
->xps_cpus_map
, new_dev_maps
);
2367 /* Cleanup old maps */
2369 goto out_no_old_maps
;
2371 for (j
= -1; j
= netif_attrmask_next(j
, possible_mask
, nr_ids
),
2373 for (i
= num_tc
, tci
= j
* num_tc
; i
--; tci
++) {
2374 new_map
= xmap_dereference(new_dev_maps
->attr_map
[tci
]);
2375 map
= xmap_dereference(dev_maps
->attr_map
[tci
]);
2376 if (map
&& map
!= new_map
)
2377 kfree_rcu(map
, rcu
);
2381 kfree_rcu(dev_maps
, rcu
);
2384 dev_maps
= new_dev_maps
;
2389 /* update Tx queue numa node */
2390 netdev_queue_numa_node_write(netdev_get_tx_queue(dev
, index
),
2391 (numa_node_id
>= 0) ?
2392 numa_node_id
: NUMA_NO_NODE
);
2398 /* removes tx-queue from unused CPUs/rx-queues */
2399 for (j
= -1; j
= netif_attrmask_next(j
, possible_mask
, nr_ids
),
2401 for (i
= tc
, tci
= j
* num_tc
; i
--; tci
++)
2402 active
|= remove_xps_queue(dev_maps
, tci
, index
);
2403 if (!netif_attr_test_mask(j
, mask
, nr_ids
) ||
2404 !netif_attr_test_online(j
, online_mask
, nr_ids
))
2405 active
|= remove_xps_queue(dev_maps
, tci
, index
);
2406 for (i
= num_tc
- tc
, tci
++; --i
; tci
++)
2407 active
|= remove_xps_queue(dev_maps
, tci
, index
);
2410 /* free map if not active */
2413 RCU_INIT_POINTER(dev
->xps_rxqs_map
, NULL
);
2415 RCU_INIT_POINTER(dev
->xps_cpus_map
, NULL
);
2416 kfree_rcu(dev_maps
, rcu
);
2420 mutex_unlock(&xps_map_mutex
);
2424 /* remove any maps that we added */
2425 for (j
= -1; j
= netif_attrmask_next(j
, possible_mask
, nr_ids
),
2427 for (i
= num_tc
, tci
= j
* num_tc
; i
--; tci
++) {
2428 new_map
= xmap_dereference(new_dev_maps
->attr_map
[tci
]);
2430 xmap_dereference(dev_maps
->attr_map
[tci
]) :
2432 if (new_map
&& new_map
!= map
)
2437 mutex_unlock(&xps_map_mutex
);
2439 kfree(new_dev_maps
);
2443 int netif_set_xps_queue(struct net_device
*dev
, const struct cpumask
*mask
,
2446 return __netif_set_xps_queue(dev
, cpumask_bits(mask
), index
, false);
2448 EXPORT_SYMBOL(netif_set_xps_queue
);
2451 void netdev_reset_tc(struct net_device
*dev
)
2454 netif_reset_xps_queues_gt(dev
, 0);
2457 memset(dev
->tc_to_txq
, 0, sizeof(dev
->tc_to_txq
));
2458 memset(dev
->prio_tc_map
, 0, sizeof(dev
->prio_tc_map
));
2460 EXPORT_SYMBOL(netdev_reset_tc
);
2462 int netdev_set_tc_queue(struct net_device
*dev
, u8 tc
, u16 count
, u16 offset
)
2464 if (tc
>= dev
->num_tc
)
2468 netif_reset_xps_queues(dev
, offset
, count
);
2470 dev
->tc_to_txq
[tc
].count
= count
;
2471 dev
->tc_to_txq
[tc
].offset
= offset
;
2474 EXPORT_SYMBOL(netdev_set_tc_queue
);
2476 int netdev_set_num_tc(struct net_device
*dev
, u8 num_tc
)
2478 if (num_tc
> TC_MAX_QUEUE
)
2482 netif_reset_xps_queues_gt(dev
, 0);
2484 dev
->num_tc
= num_tc
;
2487 EXPORT_SYMBOL(netdev_set_num_tc
);
2490 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2491 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2493 int netif_set_real_num_tx_queues(struct net_device
*dev
, unsigned int txq
)
2498 disabling
= txq
< dev
->real_num_tx_queues
;
2500 if (txq
< 1 || txq
> dev
->num_tx_queues
)
2503 if (dev
->reg_state
== NETREG_REGISTERED
||
2504 dev
->reg_state
== NETREG_UNREGISTERING
) {
2507 rc
= netdev_queue_update_kobjects(dev
, dev
->real_num_tx_queues
,
2513 netif_setup_tc(dev
, txq
);
2515 dev
->real_num_tx_queues
= txq
;
2519 qdisc_reset_all_tx_gt(dev
, txq
);
2521 netif_reset_xps_queues_gt(dev
, txq
);
2525 dev
->real_num_tx_queues
= txq
;
2530 EXPORT_SYMBOL(netif_set_real_num_tx_queues
);
2534 * netif_set_real_num_rx_queues - set actual number of RX queues used
2535 * @dev: Network device
2536 * @rxq: Actual number of RX queues
2538 * This must be called either with the rtnl_lock held or before
2539 * registration of the net device. Returns 0 on success, or a
2540 * negative error code. If called before registration, it always
2543 int netif_set_real_num_rx_queues(struct net_device
*dev
, unsigned int rxq
)
2547 if (rxq
< 1 || rxq
> dev
->num_rx_queues
)
2550 if (dev
->reg_state
== NETREG_REGISTERED
) {
2553 rc
= net_rx_queue_update_kobjects(dev
, dev
->real_num_rx_queues
,
2559 dev
->real_num_rx_queues
= rxq
;
2562 EXPORT_SYMBOL(netif_set_real_num_rx_queues
);
2566 * netif_get_num_default_rss_queues - default number of RSS queues
2568 * This routine should set an upper limit on the number of RSS queues
2569 * used by default by multiqueue devices.
2571 int netif_get_num_default_rss_queues(void)
2573 return is_kdump_kernel() ?
2574 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES
, num_online_cpus());
2576 EXPORT_SYMBOL(netif_get_num_default_rss_queues
);
2578 static void __netif_reschedule(struct Qdisc
*q
)
2580 struct softnet_data
*sd
;
2581 unsigned long flags
;
2583 local_irq_save(flags
);
2584 sd
= this_cpu_ptr(&softnet_data
);
2585 q
->next_sched
= NULL
;
2586 *sd
->output_queue_tailp
= q
;
2587 sd
->output_queue_tailp
= &q
->next_sched
;
2588 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
2589 local_irq_restore(flags
);
2592 void __netif_schedule(struct Qdisc
*q
)
2594 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
2595 __netif_reschedule(q
);
2597 EXPORT_SYMBOL(__netif_schedule
);
2599 struct dev_kfree_skb_cb
{
2600 enum skb_free_reason reason
;
2603 static struct dev_kfree_skb_cb
*get_kfree_skb_cb(const struct sk_buff
*skb
)
2605 return (struct dev_kfree_skb_cb
*)skb
->cb
;
2608 void netif_schedule_queue(struct netdev_queue
*txq
)
2611 if (!(txq
->state
& QUEUE_STATE_ANY_XOFF
)) {
2612 struct Qdisc
*q
= rcu_dereference(txq
->qdisc
);
2614 __netif_schedule(q
);
2618 EXPORT_SYMBOL(netif_schedule_queue
);
2620 void netif_tx_wake_queue(struct netdev_queue
*dev_queue
)
2622 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF
, &dev_queue
->state
)) {
2626 q
= rcu_dereference(dev_queue
->qdisc
);
2627 __netif_schedule(q
);
2631 EXPORT_SYMBOL(netif_tx_wake_queue
);
2633 void __dev_kfree_skb_irq(struct sk_buff
*skb
, enum skb_free_reason reason
)
2635 unsigned long flags
;
2640 if (likely(refcount_read(&skb
->users
) == 1)) {
2642 refcount_set(&skb
->users
, 0);
2643 } else if (likely(!refcount_dec_and_test(&skb
->users
))) {
2646 get_kfree_skb_cb(skb
)->reason
= reason
;
2647 local_irq_save(flags
);
2648 skb
->next
= __this_cpu_read(softnet_data
.completion_queue
);
2649 __this_cpu_write(softnet_data
.completion_queue
, skb
);
2650 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
2651 local_irq_restore(flags
);
2653 EXPORT_SYMBOL(__dev_kfree_skb_irq
);
2655 void __dev_kfree_skb_any(struct sk_buff
*skb
, enum skb_free_reason reason
)
2657 if (in_irq() || irqs_disabled())
2658 __dev_kfree_skb_irq(skb
, reason
);
2662 EXPORT_SYMBOL(__dev_kfree_skb_any
);
2666 * netif_device_detach - mark device as removed
2667 * @dev: network device
2669 * Mark device as removed from system and therefore no longer available.
2671 void netif_device_detach(struct net_device
*dev
)
2673 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
2674 netif_running(dev
)) {
2675 netif_tx_stop_all_queues(dev
);
2678 EXPORT_SYMBOL(netif_device_detach
);
2681 * netif_device_attach - mark device as attached
2682 * @dev: network device
2684 * Mark device as attached from system and restart if needed.
2686 void netif_device_attach(struct net_device
*dev
)
2688 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
2689 netif_running(dev
)) {
2690 netif_tx_wake_all_queues(dev
);
2691 __netdev_watchdog_up(dev
);
2694 EXPORT_SYMBOL(netif_device_attach
);
2697 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2698 * to be used as a distribution range.
2700 static u16
skb_tx_hash(const struct net_device
*dev
, struct sk_buff
*skb
)
2704 u16 qcount
= dev
->real_num_tx_queues
;
2706 if (skb_rx_queue_recorded(skb
)) {
2707 hash
= skb_get_rx_queue(skb
);
2708 while (unlikely(hash
>= qcount
))
2714 u8 tc
= netdev_get_prio_tc_map(dev
, skb
->priority
);
2716 qoffset
= dev
->tc_to_txq
[tc
].offset
;
2717 qcount
= dev
->tc_to_txq
[tc
].count
;
2720 return (u16
) reciprocal_scale(skb_get_hash(skb
), qcount
) + qoffset
;
2723 static void skb_warn_bad_offload(const struct sk_buff
*skb
)
2725 static const netdev_features_t null_features
;
2726 struct net_device
*dev
= skb
->dev
;
2727 const char *name
= "";
2729 if (!net_ratelimit())
2733 if (dev
->dev
.parent
)
2734 name
= dev_driver_string(dev
->dev
.parent
);
2736 name
= netdev_name(dev
);
2738 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2739 "gso_type=%d ip_summed=%d\n",
2740 name
, dev
? &dev
->features
: &null_features
,
2741 skb
->sk
? &skb
->sk
->sk_route_caps
: &null_features
,
2742 skb
->len
, skb
->data_len
, skb_shinfo(skb
)->gso_size
,
2743 skb_shinfo(skb
)->gso_type
, skb
->ip_summed
);
2747 * Invalidate hardware checksum when packet is to be mangled, and
2748 * complete checksum manually on outgoing path.
2750 int skb_checksum_help(struct sk_buff
*skb
)
2753 int ret
= 0, offset
;
2755 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
2756 goto out_set_summed
;
2758 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
2759 skb_warn_bad_offload(skb
);
2763 /* Before computing a checksum, we should make sure no frag could
2764 * be modified by an external entity : checksum could be wrong.
2766 if (skb_has_shared_frag(skb
)) {
2767 ret
= __skb_linearize(skb
);
2772 offset
= skb_checksum_start_offset(skb
);
2773 BUG_ON(offset
>= skb_headlen(skb
));
2774 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
2776 offset
+= skb
->csum_offset
;
2777 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
2779 if (skb_cloned(skb
) &&
2780 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
2781 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2786 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
) ?: CSUM_MANGLED_0
;
2788 skb
->ip_summed
= CHECKSUM_NONE
;
2792 EXPORT_SYMBOL(skb_checksum_help
);
2794 int skb_crc32c_csum_help(struct sk_buff
*skb
)
2797 int ret
= 0, offset
, start
;
2799 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
)
2802 if (unlikely(skb_is_gso(skb
)))
2805 /* Before computing a checksum, we should make sure no frag could
2806 * be modified by an external entity : checksum could be wrong.
2808 if (unlikely(skb_has_shared_frag(skb
))) {
2809 ret
= __skb_linearize(skb
);
2813 start
= skb_checksum_start_offset(skb
);
2814 offset
= start
+ offsetof(struct sctphdr
, checksum
);
2815 if (WARN_ON_ONCE(offset
>= skb_headlen(skb
))) {
2819 if (skb_cloned(skb
) &&
2820 !skb_clone_writable(skb
, offset
+ sizeof(__le32
))) {
2821 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2825 crc32c_csum
= cpu_to_le32(~__skb_checksum(skb
, start
,
2826 skb
->len
- start
, ~(__u32
)0,
2828 *(__le32
*)(skb
->data
+ offset
) = crc32c_csum
;
2829 skb
->ip_summed
= CHECKSUM_NONE
;
2830 skb
->csum_not_inet
= 0;
2835 __be16
skb_network_protocol(struct sk_buff
*skb
, int *depth
)
2837 __be16 type
= skb
->protocol
;
2839 /* Tunnel gso handlers can set protocol to ethernet. */
2840 if (type
== htons(ETH_P_TEB
)) {
2843 if (unlikely(!pskb_may_pull(skb
, sizeof(struct ethhdr
))))
2846 eth
= (struct ethhdr
*)skb
->data
;
2847 type
= eth
->h_proto
;
2850 return __vlan_get_protocol(skb
, type
, depth
);
2854 * skb_mac_gso_segment - mac layer segmentation handler.
2855 * @skb: buffer to segment
2856 * @features: features for the output path (see dev->features)
2858 struct sk_buff
*skb_mac_gso_segment(struct sk_buff
*skb
,
2859 netdev_features_t features
)
2861 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
2862 struct packet_offload
*ptype
;
2863 int vlan_depth
= skb
->mac_len
;
2864 __be16 type
= skb_network_protocol(skb
, &vlan_depth
);
2866 if (unlikely(!type
))
2867 return ERR_PTR(-EINVAL
);
2869 __skb_pull(skb
, vlan_depth
);
2872 list_for_each_entry_rcu(ptype
, &offload_base
, list
) {
2873 if (ptype
->type
== type
&& ptype
->callbacks
.gso_segment
) {
2874 segs
= ptype
->callbacks
.gso_segment(skb
, features
);
2880 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
2884 EXPORT_SYMBOL(skb_mac_gso_segment
);
2887 /* openvswitch calls this on rx path, so we need a different check.
2889 static inline bool skb_needs_check(struct sk_buff
*skb
, bool tx_path
)
2892 return skb
->ip_summed
!= CHECKSUM_PARTIAL
&&
2893 skb
->ip_summed
!= CHECKSUM_UNNECESSARY
;
2895 return skb
->ip_summed
== CHECKSUM_NONE
;
2899 * __skb_gso_segment - Perform segmentation on skb.
2900 * @skb: buffer to segment
2901 * @features: features for the output path (see dev->features)
2902 * @tx_path: whether it is called in TX path
2904 * This function segments the given skb and returns a list of segments.
2906 * It may return NULL if the skb requires no segmentation. This is
2907 * only possible when GSO is used for verifying header integrity.
2909 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
2911 struct sk_buff
*__skb_gso_segment(struct sk_buff
*skb
,
2912 netdev_features_t features
, bool tx_path
)
2914 struct sk_buff
*segs
;
2916 if (unlikely(skb_needs_check(skb
, tx_path
))) {
2919 /* We're going to init ->check field in TCP or UDP header */
2920 err
= skb_cow_head(skb
, 0);
2922 return ERR_PTR(err
);
2925 /* Only report GSO partial support if it will enable us to
2926 * support segmentation on this frame without needing additional
2929 if (features
& NETIF_F_GSO_PARTIAL
) {
2930 netdev_features_t partial_features
= NETIF_F_GSO_ROBUST
;
2931 struct net_device
*dev
= skb
->dev
;
2933 partial_features
|= dev
->features
& dev
->gso_partial_features
;
2934 if (!skb_gso_ok(skb
, features
| partial_features
))
2935 features
&= ~NETIF_F_GSO_PARTIAL
;
2938 BUILD_BUG_ON(SKB_SGO_CB_OFFSET
+
2939 sizeof(*SKB_GSO_CB(skb
)) > sizeof(skb
->cb
));
2941 SKB_GSO_CB(skb
)->mac_offset
= skb_headroom(skb
);
2942 SKB_GSO_CB(skb
)->encap_level
= 0;
2944 skb_reset_mac_header(skb
);
2945 skb_reset_mac_len(skb
);
2947 segs
= skb_mac_gso_segment(skb
, features
);
2949 if (unlikely(skb_needs_check(skb
, tx_path
) && !IS_ERR(segs
)))
2950 skb_warn_bad_offload(skb
);
2954 EXPORT_SYMBOL(__skb_gso_segment
);
2956 /* Take action when hardware reception checksum errors are detected. */
2958 void netdev_rx_csum_fault(struct net_device
*dev
)
2960 if (net_ratelimit()) {
2961 pr_err("%s: hw csum failure\n", dev
? dev
->name
: "<unknown>");
2965 EXPORT_SYMBOL(netdev_rx_csum_fault
);
2968 /* XXX: check that highmem exists at all on the given machine. */
2969 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
2971 #ifdef CONFIG_HIGHMEM
2974 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
2975 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2976 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2978 if (PageHighMem(skb_frag_page(frag
)))
2986 /* If MPLS offload request, verify we are testing hardware MPLS features
2987 * instead of standard features for the netdev.
2989 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
2990 static netdev_features_t
net_mpls_features(struct sk_buff
*skb
,
2991 netdev_features_t features
,
2994 if (eth_p_mpls(type
))
2995 features
&= skb
->dev
->mpls_features
;
3000 static netdev_features_t
net_mpls_features(struct sk_buff
*skb
,
3001 netdev_features_t features
,
3008 static netdev_features_t
harmonize_features(struct sk_buff
*skb
,
3009 netdev_features_t features
)
3014 type
= skb_network_protocol(skb
, &tmp
);
3015 features
= net_mpls_features(skb
, features
, type
);
3017 if (skb
->ip_summed
!= CHECKSUM_NONE
&&
3018 !can_checksum_protocol(features
, type
)) {
3019 features
&= ~(NETIF_F_CSUM_MASK
| NETIF_F_GSO_MASK
);
3021 if (illegal_highdma(skb
->dev
, skb
))
3022 features
&= ~NETIF_F_SG
;
3027 netdev_features_t
passthru_features_check(struct sk_buff
*skb
,
3028 struct net_device
*dev
,
3029 netdev_features_t features
)
3033 EXPORT_SYMBOL(passthru_features_check
);
3035 static netdev_features_t
dflt_features_check(struct sk_buff
*skb
,
3036 struct net_device
*dev
,
3037 netdev_features_t features
)
3039 return vlan_features_check(skb
, features
);
3042 static netdev_features_t
gso_features_check(const struct sk_buff
*skb
,
3043 struct net_device
*dev
,
3044 netdev_features_t features
)
3046 u16 gso_segs
= skb_shinfo(skb
)->gso_segs
;
3048 if (gso_segs
> dev
->gso_max_segs
)
3049 return features
& ~NETIF_F_GSO_MASK
;
3051 /* Support for GSO partial features requires software
3052 * intervention before we can actually process the packets
3053 * so we need to strip support for any partial features now
3054 * and we can pull them back in after we have partially
3055 * segmented the frame.
3057 if (!(skb_shinfo(skb
)->gso_type
& SKB_GSO_PARTIAL
))
3058 features
&= ~dev
->gso_partial_features
;
3060 /* Make sure to clear the IPv4 ID mangling feature if the
3061 * IPv4 header has the potential to be fragmented.
3063 if (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCPV4
) {
3064 struct iphdr
*iph
= skb
->encapsulation
?
3065 inner_ip_hdr(skb
) : ip_hdr(skb
);
3067 if (!(iph
->frag_off
& htons(IP_DF
)))
3068 features
&= ~NETIF_F_TSO_MANGLEID
;
3074 netdev_features_t
netif_skb_features(struct sk_buff
*skb
)
3076 struct net_device
*dev
= skb
->dev
;
3077 netdev_features_t features
= dev
->features
;
3079 if (skb_is_gso(skb
))
3080 features
= gso_features_check(skb
, dev
, features
);
3082 /* If encapsulation offload request, verify we are testing
3083 * hardware encapsulation features instead of standard
3084 * features for the netdev
3086 if (skb
->encapsulation
)
3087 features
&= dev
->hw_enc_features
;
3089 if (skb_vlan_tagged(skb
))
3090 features
= netdev_intersect_features(features
,
3091 dev
->vlan_features
|
3092 NETIF_F_HW_VLAN_CTAG_TX
|
3093 NETIF_F_HW_VLAN_STAG_TX
);
3095 if (dev
->netdev_ops
->ndo_features_check
)
3096 features
&= dev
->netdev_ops
->ndo_features_check(skb
, dev
,
3099 features
&= dflt_features_check(skb
, dev
, features
);
3101 return harmonize_features(skb
, features
);
3103 EXPORT_SYMBOL(netif_skb_features
);
3105 static int xmit_one(struct sk_buff
*skb
, struct net_device
*dev
,
3106 struct netdev_queue
*txq
, bool more
)
3111 if (!list_empty(&ptype_all
) || !list_empty(&dev
->ptype_all
))
3112 dev_queue_xmit_nit(skb
, dev
);
3115 trace_net_dev_start_xmit(skb
, dev
);
3116 rc
= netdev_start_xmit(skb
, dev
, txq
, more
);
3117 trace_net_dev_xmit(skb
, rc
, dev
, len
);
3122 struct sk_buff
*dev_hard_start_xmit(struct sk_buff
*first
, struct net_device
*dev
,
3123 struct netdev_queue
*txq
, int *ret
)
3125 struct sk_buff
*skb
= first
;
3126 int rc
= NETDEV_TX_OK
;
3129 struct sk_buff
*next
= skb
->next
;
3132 rc
= xmit_one(skb
, dev
, txq
, next
!= NULL
);
3133 if (unlikely(!dev_xmit_complete(rc
))) {
3139 if (netif_xmit_stopped(txq
) && skb
) {
3140 rc
= NETDEV_TX_BUSY
;
3150 static struct sk_buff
*validate_xmit_vlan(struct sk_buff
*skb
,
3151 netdev_features_t features
)
3153 if (skb_vlan_tag_present(skb
) &&
3154 !vlan_hw_offload_capable(features
, skb
->vlan_proto
))
3155 skb
= __vlan_hwaccel_push_inside(skb
);
3159 int skb_csum_hwoffload_help(struct sk_buff
*skb
,
3160 const netdev_features_t features
)
3162 if (unlikely(skb
->csum_not_inet
))
3163 return !!(features
& NETIF_F_SCTP_CRC
) ? 0 :
3164 skb_crc32c_csum_help(skb
);
3166 return !!(features
& NETIF_F_CSUM_MASK
) ? 0 : skb_checksum_help(skb
);
3168 EXPORT_SYMBOL(skb_csum_hwoffload_help
);
3170 static struct sk_buff
*validate_xmit_skb(struct sk_buff
*skb
, struct net_device
*dev
, bool *again
)
3172 netdev_features_t features
;
3174 features
= netif_skb_features(skb
);
3175 skb
= validate_xmit_vlan(skb
, features
);
3179 skb
= sk_validate_xmit_skb(skb
, dev
);
3183 if (netif_needs_gso(skb
, features
)) {
3184 struct sk_buff
*segs
;
3186 segs
= skb_gso_segment(skb
, features
);
3194 if (skb_needs_linearize(skb
, features
) &&
3195 __skb_linearize(skb
))
3198 /* If packet is not checksummed and device does not
3199 * support checksumming for this protocol, complete
3200 * checksumming here.
3202 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
3203 if (skb
->encapsulation
)
3204 skb_set_inner_transport_header(skb
,
3205 skb_checksum_start_offset(skb
));
3207 skb_set_transport_header(skb
,
3208 skb_checksum_start_offset(skb
));
3209 if (skb_csum_hwoffload_help(skb
, features
))
3214 skb
= validate_xmit_xfrm(skb
, features
, again
);
3221 atomic_long_inc(&dev
->tx_dropped
);
3225 struct sk_buff
*validate_xmit_skb_list(struct sk_buff
*skb
, struct net_device
*dev
, bool *again
)
3227 struct sk_buff
*next
, *head
= NULL
, *tail
;
3229 for (; skb
!= NULL
; skb
= next
) {
3233 /* in case skb wont be segmented, point to itself */
3236 skb
= validate_xmit_skb(skb
, dev
, again
);
3244 /* If skb was segmented, skb->prev points to
3245 * the last segment. If not, it still contains skb.
3251 EXPORT_SYMBOL_GPL(validate_xmit_skb_list
);
3253 static void qdisc_pkt_len_init(struct sk_buff
*skb
)
3255 const struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
3257 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
3259 /* To get more precise estimation of bytes sent on wire,
3260 * we add to pkt_len the headers size of all segments
3262 if (shinfo
->gso_size
) {
3263 unsigned int hdr_len
;
3264 u16 gso_segs
= shinfo
->gso_segs
;
3266 /* mac layer + network layer */
3267 hdr_len
= skb_transport_header(skb
) - skb_mac_header(skb
);
3269 /* + transport layer */
3270 if (likely(shinfo
->gso_type
& (SKB_GSO_TCPV4
| SKB_GSO_TCPV6
))) {
3271 const struct tcphdr
*th
;
3272 struct tcphdr _tcphdr
;
3274 th
= skb_header_pointer(skb
, skb_transport_offset(skb
),
3275 sizeof(_tcphdr
), &_tcphdr
);
3277 hdr_len
+= __tcp_hdrlen(th
);
3279 struct udphdr _udphdr
;
3281 if (skb_header_pointer(skb
, skb_transport_offset(skb
),
3282 sizeof(_udphdr
), &_udphdr
))
3283 hdr_len
+= sizeof(struct udphdr
);
3286 if (shinfo
->gso_type
& SKB_GSO_DODGY
)
3287 gso_segs
= DIV_ROUND_UP(skb
->len
- hdr_len
,
3290 qdisc_skb_cb(skb
)->pkt_len
+= (gso_segs
- 1) * hdr_len
;
3294 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
3295 struct net_device
*dev
,
3296 struct netdev_queue
*txq
)
3298 spinlock_t
*root_lock
= qdisc_lock(q
);
3299 struct sk_buff
*to_free
= NULL
;
3303 qdisc_calculate_pkt_len(skb
, q
);
3305 if (q
->flags
& TCQ_F_NOLOCK
) {
3306 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
3307 __qdisc_drop(skb
, &to_free
);
3310 rc
= q
->enqueue(skb
, q
, &to_free
) & NET_XMIT_MASK
;
3314 if (unlikely(to_free
))
3315 kfree_skb_list(to_free
);
3320 * Heuristic to force contended enqueues to serialize on a
3321 * separate lock before trying to get qdisc main lock.
3322 * This permits qdisc->running owner to get the lock more
3323 * often and dequeue packets faster.
3325 contended
= qdisc_is_running(q
);
3326 if (unlikely(contended
))
3327 spin_lock(&q
->busylock
);
3329 spin_lock(root_lock
);
3330 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
3331 __qdisc_drop(skb
, &to_free
);
3333 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
3334 qdisc_run_begin(q
)) {
3336 * This is a work-conserving queue; there are no old skbs
3337 * waiting to be sent out; and the qdisc is not running -
3338 * xmit the skb directly.
3341 qdisc_bstats_update(q
, skb
);
3343 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
, true)) {
3344 if (unlikely(contended
)) {
3345 spin_unlock(&q
->busylock
);
3352 rc
= NET_XMIT_SUCCESS
;
3354 rc
= q
->enqueue(skb
, q
, &to_free
) & NET_XMIT_MASK
;
3355 if (qdisc_run_begin(q
)) {
3356 if (unlikely(contended
)) {
3357 spin_unlock(&q
->busylock
);
3364 spin_unlock(root_lock
);
3365 if (unlikely(to_free
))
3366 kfree_skb_list(to_free
);
3367 if (unlikely(contended
))
3368 spin_unlock(&q
->busylock
);
3372 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3373 static void skb_update_prio(struct sk_buff
*skb
)
3375 const struct netprio_map
*map
;
3376 const struct sock
*sk
;
3377 unsigned int prioidx
;
3381 map
= rcu_dereference_bh(skb
->dev
->priomap
);
3384 sk
= skb_to_full_sk(skb
);
3388 prioidx
= sock_cgroup_prioidx(&sk
->sk_cgrp_data
);
3390 if (prioidx
< map
->priomap_len
)
3391 skb
->priority
= map
->priomap
[prioidx
];
3394 #define skb_update_prio(skb)
3397 DEFINE_PER_CPU(int, xmit_recursion
);
3398 EXPORT_SYMBOL(xmit_recursion
);
3401 * dev_loopback_xmit - loop back @skb
3402 * @net: network namespace this loopback is happening in
3403 * @sk: sk needed to be a netfilter okfn
3404 * @skb: buffer to transmit
3406 int dev_loopback_xmit(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
3408 skb_reset_mac_header(skb
);
3409 __skb_pull(skb
, skb_network_offset(skb
));
3410 skb
->pkt_type
= PACKET_LOOPBACK
;
3411 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
3412 WARN_ON(!skb_dst(skb
));
3417 EXPORT_SYMBOL(dev_loopback_xmit
);
3419 #ifdef CONFIG_NET_EGRESS
3420 static struct sk_buff
*
3421 sch_handle_egress(struct sk_buff
*skb
, int *ret
, struct net_device
*dev
)
3423 struct mini_Qdisc
*miniq
= rcu_dereference_bh(dev
->miniq_egress
);
3424 struct tcf_result cl_res
;
3429 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3430 mini_qdisc_bstats_cpu_update(miniq
, skb
);
3432 switch (tcf_classify(skb
, miniq
->filter_list
, &cl_res
, false)) {
3434 case TC_ACT_RECLASSIFY
:
3435 skb
->tc_index
= TC_H_MIN(cl_res
.classid
);
3438 mini_qdisc_qstats_cpu_drop(miniq
);
3439 *ret
= NET_XMIT_DROP
;
3445 *ret
= NET_XMIT_SUCCESS
;
3448 case TC_ACT_REDIRECT
:
3449 /* No need to push/pop skb's mac_header here on egress! */
3450 skb_do_redirect(skb
);
3451 *ret
= NET_XMIT_SUCCESS
;
3459 #endif /* CONFIG_NET_EGRESS */
3462 static int __get_xps_queue_idx(struct net_device
*dev
, struct sk_buff
*skb
,
3463 struct xps_dev_maps
*dev_maps
, unsigned int tci
)
3465 struct xps_map
*map
;
3466 int queue_index
= -1;
3470 tci
+= netdev_get_prio_tc_map(dev
, skb
->priority
);
3473 map
= rcu_dereference(dev_maps
->attr_map
[tci
]);
3476 queue_index
= map
->queues
[0];
3478 queue_index
= map
->queues
[reciprocal_scale(
3479 skb_get_hash(skb
), map
->len
)];
3480 if (unlikely(queue_index
>= dev
->real_num_tx_queues
))
3487 static int get_xps_queue(struct net_device
*dev
, struct sk_buff
*skb
)
3490 struct xps_dev_maps
*dev_maps
;
3491 struct sock
*sk
= skb
->sk
;
3492 int queue_index
= -1;
3494 if (!static_key_false(&xps_needed
))
3498 if (!static_key_false(&xps_rxqs_needed
))
3501 dev_maps
= rcu_dereference(dev
->xps_rxqs_map
);
3503 int tci
= sk_rx_queue_get(sk
);
3505 if (tci
>= 0 && tci
< dev
->num_rx_queues
)
3506 queue_index
= __get_xps_queue_idx(dev
, skb
, dev_maps
,
3511 if (queue_index
< 0) {
3512 dev_maps
= rcu_dereference(dev
->xps_cpus_map
);
3514 unsigned int tci
= skb
->sender_cpu
- 1;
3516 queue_index
= __get_xps_queue_idx(dev
, skb
, dev_maps
,
3528 static u16
__netdev_pick_tx(struct net_device
*dev
, struct sk_buff
*skb
)
3530 struct sock
*sk
= skb
->sk
;
3531 int queue_index
= sk_tx_queue_get(sk
);
3533 if (queue_index
< 0 || skb
->ooo_okay
||
3534 queue_index
>= dev
->real_num_tx_queues
) {
3535 int new_index
= get_xps_queue(dev
, skb
);
3538 new_index
= skb_tx_hash(dev
, skb
);
3540 if (queue_index
!= new_index
&& sk
&&
3542 rcu_access_pointer(sk
->sk_dst_cache
))
3543 sk_tx_queue_set(sk
, new_index
);
3545 queue_index
= new_index
;
3551 struct netdev_queue
*netdev_pick_tx(struct net_device
*dev
,
3552 struct sk_buff
*skb
,
3555 int queue_index
= 0;
3558 u32 sender_cpu
= skb
->sender_cpu
- 1;
3560 if (sender_cpu
>= (u32
)NR_CPUS
)
3561 skb
->sender_cpu
= raw_smp_processor_id() + 1;
3564 if (dev
->real_num_tx_queues
!= 1) {
3565 const struct net_device_ops
*ops
= dev
->netdev_ops
;
3567 if (ops
->ndo_select_queue
)
3568 queue_index
= ops
->ndo_select_queue(dev
, skb
, accel_priv
,
3571 queue_index
= __netdev_pick_tx(dev
, skb
);
3573 queue_index
= netdev_cap_txqueue(dev
, queue_index
);
3576 skb_set_queue_mapping(skb
, queue_index
);
3577 return netdev_get_tx_queue(dev
, queue_index
);
3581 * __dev_queue_xmit - transmit a buffer
3582 * @skb: buffer to transmit
3583 * @accel_priv: private data used for L2 forwarding offload
3585 * Queue a buffer for transmission to a network device. The caller must
3586 * have set the device and priority and built the buffer before calling
3587 * this function. The function can be called from an interrupt.
3589 * A negative errno code is returned on a failure. A success does not
3590 * guarantee the frame will be transmitted as it may be dropped due
3591 * to congestion or traffic shaping.
3593 * -----------------------------------------------------------------------------------
3594 * I notice this method can also return errors from the queue disciplines,
3595 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3598 * Regardless of the return value, the skb is consumed, so it is currently
3599 * difficult to retry a send to this method. (You can bump the ref count
3600 * before sending to hold a reference for retry if you are careful.)
3602 * When calling this method, interrupts MUST be enabled. This is because
3603 * the BH enable code must have IRQs enabled so that it will not deadlock.
3606 static int __dev_queue_xmit(struct sk_buff
*skb
, void *accel_priv
)
3608 struct net_device
*dev
= skb
->dev
;
3609 struct netdev_queue
*txq
;
3614 skb_reset_mac_header(skb
);
3616 if (unlikely(skb_shinfo(skb
)->tx_flags
& SKBTX_SCHED_TSTAMP
))
3617 __skb_tstamp_tx(skb
, NULL
, skb
->sk
, SCM_TSTAMP_SCHED
);
3619 /* Disable soft irqs for various locks below. Also
3620 * stops preemption for RCU.
3624 skb_update_prio(skb
);
3626 qdisc_pkt_len_init(skb
);
3627 #ifdef CONFIG_NET_CLS_ACT
3628 skb
->tc_at_ingress
= 0;
3629 # ifdef CONFIG_NET_EGRESS
3630 if (static_branch_unlikely(&egress_needed_key
)) {
3631 skb
= sch_handle_egress(skb
, &rc
, dev
);
3637 /* If device/qdisc don't need skb->dst, release it right now while
3638 * its hot in this cpu cache.
3640 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
3645 txq
= netdev_pick_tx(dev
, skb
, accel_priv
);
3646 q
= rcu_dereference_bh(txq
->qdisc
);
3648 trace_net_dev_queue(skb
);
3650 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
3654 /* The device has no queue. Common case for software devices:
3655 * loopback, all the sorts of tunnels...
3657 * Really, it is unlikely that netif_tx_lock protection is necessary
3658 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
3660 * However, it is possible, that they rely on protection
3663 * Check this and shot the lock. It is not prone from deadlocks.
3664 *Either shot noqueue qdisc, it is even simpler 8)
3666 if (dev
->flags
& IFF_UP
) {
3667 int cpu
= smp_processor_id(); /* ok because BHs are off */
3669 if (txq
->xmit_lock_owner
!= cpu
) {
3670 if (unlikely(__this_cpu_read(xmit_recursion
) >
3671 XMIT_RECURSION_LIMIT
))
3672 goto recursion_alert
;
3674 skb
= validate_xmit_skb(skb
, dev
, &again
);
3678 HARD_TX_LOCK(dev
, txq
, cpu
);
3680 if (!netif_xmit_stopped(txq
)) {
3681 __this_cpu_inc(xmit_recursion
);
3682 skb
= dev_hard_start_xmit(skb
, dev
, txq
, &rc
);
3683 __this_cpu_dec(xmit_recursion
);
3684 if (dev_xmit_complete(rc
)) {
3685 HARD_TX_UNLOCK(dev
, txq
);
3689 HARD_TX_UNLOCK(dev
, txq
);
3690 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3693 /* Recursion is detected! It is possible,
3697 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3703 rcu_read_unlock_bh();
3705 atomic_long_inc(&dev
->tx_dropped
);
3706 kfree_skb_list(skb
);
3709 rcu_read_unlock_bh();
3713 int dev_queue_xmit(struct sk_buff
*skb
)
3715 return __dev_queue_xmit(skb
, NULL
);
3717 EXPORT_SYMBOL(dev_queue_xmit
);
3719 int dev_queue_xmit_accel(struct sk_buff
*skb
, void *accel_priv
)
3721 return __dev_queue_xmit(skb
, accel_priv
);
3723 EXPORT_SYMBOL(dev_queue_xmit_accel
);
3725 int dev_direct_xmit(struct sk_buff
*skb
, u16 queue_id
)
3727 struct net_device
*dev
= skb
->dev
;
3728 struct sk_buff
*orig_skb
= skb
;
3729 struct netdev_queue
*txq
;
3730 int ret
= NETDEV_TX_BUSY
;
3733 if (unlikely(!netif_running(dev
) ||
3734 !netif_carrier_ok(dev
)))
3737 skb
= validate_xmit_skb_list(skb
, dev
, &again
);
3738 if (skb
!= orig_skb
)
3741 skb_set_queue_mapping(skb
, queue_id
);
3742 txq
= skb_get_tx_queue(dev
, skb
);
3746 HARD_TX_LOCK(dev
, txq
, smp_processor_id());
3747 if (!netif_xmit_frozen_or_drv_stopped(txq
))
3748 ret
= netdev_start_xmit(skb
, dev
, txq
, false);
3749 HARD_TX_UNLOCK(dev
, txq
);
3753 if (!dev_xmit_complete(ret
))
3758 atomic_long_inc(&dev
->tx_dropped
);
3759 kfree_skb_list(skb
);
3760 return NET_XMIT_DROP
;
3762 EXPORT_SYMBOL(dev_direct_xmit
);
3764 /*************************************************************************
3766 *************************************************************************/
3768 int netdev_max_backlog __read_mostly
= 1000;
3769 EXPORT_SYMBOL(netdev_max_backlog
);
3771 int netdev_tstamp_prequeue __read_mostly
= 1;
3772 int netdev_budget __read_mostly
= 300;
3773 unsigned int __read_mostly netdev_budget_usecs
= 2000;
3774 int weight_p __read_mostly
= 64; /* old backlog weight */
3775 int dev_weight_rx_bias __read_mostly
= 1; /* bias for backlog weight */
3776 int dev_weight_tx_bias __read_mostly
= 1; /* bias for output_queue quota */
3777 int dev_rx_weight __read_mostly
= 64;
3778 int dev_tx_weight __read_mostly
= 64;
3780 /* Called with irq disabled */
3781 static inline void ____napi_schedule(struct softnet_data
*sd
,
3782 struct napi_struct
*napi
)
3784 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
3785 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3790 /* One global table that all flow-based protocols share. */
3791 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
3792 EXPORT_SYMBOL(rps_sock_flow_table
);
3793 u32 rps_cpu_mask __read_mostly
;
3794 EXPORT_SYMBOL(rps_cpu_mask
);
3796 struct static_key rps_needed __read_mostly
;
3797 EXPORT_SYMBOL(rps_needed
);
3798 struct static_key rfs_needed __read_mostly
;
3799 EXPORT_SYMBOL(rfs_needed
);
3801 static struct rps_dev_flow
*
3802 set_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
3803 struct rps_dev_flow
*rflow
, u16 next_cpu
)
3805 if (next_cpu
< nr_cpu_ids
) {
3806 #ifdef CONFIG_RFS_ACCEL
3807 struct netdev_rx_queue
*rxqueue
;
3808 struct rps_dev_flow_table
*flow_table
;
3809 struct rps_dev_flow
*old_rflow
;
3814 /* Should we steer this flow to a different hardware queue? */
3815 if (!skb_rx_queue_recorded(skb
) || !dev
->rx_cpu_rmap
||
3816 !(dev
->features
& NETIF_F_NTUPLE
))
3818 rxq_index
= cpu_rmap_lookup_index(dev
->rx_cpu_rmap
, next_cpu
);
3819 if (rxq_index
== skb_get_rx_queue(skb
))
3822 rxqueue
= dev
->_rx
+ rxq_index
;
3823 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3826 flow_id
= skb_get_hash(skb
) & flow_table
->mask
;
3827 rc
= dev
->netdev_ops
->ndo_rx_flow_steer(dev
, skb
,
3828 rxq_index
, flow_id
);
3832 rflow
= &flow_table
->flows
[flow_id
];
3834 if (old_rflow
->filter
== rflow
->filter
)
3835 old_rflow
->filter
= RPS_NO_FILTER
;
3839 per_cpu(softnet_data
, next_cpu
).input_queue_head
;
3842 rflow
->cpu
= next_cpu
;
3847 * get_rps_cpu is called from netif_receive_skb and returns the target
3848 * CPU from the RPS map of the receiving queue for a given skb.
3849 * rcu_read_lock must be held on entry.
3851 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
3852 struct rps_dev_flow
**rflowp
)
3854 const struct rps_sock_flow_table
*sock_flow_table
;
3855 struct netdev_rx_queue
*rxqueue
= dev
->_rx
;
3856 struct rps_dev_flow_table
*flow_table
;
3857 struct rps_map
*map
;
3862 if (skb_rx_queue_recorded(skb
)) {
3863 u16 index
= skb_get_rx_queue(skb
);
3865 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
3866 WARN_ONCE(dev
->real_num_rx_queues
> 1,
3867 "%s received packet on queue %u, but number "
3868 "of RX queues is %u\n",
3869 dev
->name
, index
, dev
->real_num_rx_queues
);
3875 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3877 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3878 map
= rcu_dereference(rxqueue
->rps_map
);
3879 if (!flow_table
&& !map
)
3882 skb_reset_network_header(skb
);
3883 hash
= skb_get_hash(skb
);
3887 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
3888 if (flow_table
&& sock_flow_table
) {
3889 struct rps_dev_flow
*rflow
;
3893 /* First check into global flow table if there is a match */
3894 ident
= sock_flow_table
->ents
[hash
& sock_flow_table
->mask
];
3895 if ((ident
^ hash
) & ~rps_cpu_mask
)
3898 next_cpu
= ident
& rps_cpu_mask
;
3900 /* OK, now we know there is a match,
3901 * we can look at the local (per receive queue) flow table
3903 rflow
= &flow_table
->flows
[hash
& flow_table
->mask
];
3907 * If the desired CPU (where last recvmsg was done) is
3908 * different from current CPU (one in the rx-queue flow
3909 * table entry), switch if one of the following holds:
3910 * - Current CPU is unset (>= nr_cpu_ids).
3911 * - Current CPU is offline.
3912 * - The current CPU's queue tail has advanced beyond the
3913 * last packet that was enqueued using this table entry.
3914 * This guarantees that all previous packets for the flow
3915 * have been dequeued, thus preserving in order delivery.
3917 if (unlikely(tcpu
!= next_cpu
) &&
3918 (tcpu
>= nr_cpu_ids
|| !cpu_online(tcpu
) ||
3919 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
3920 rflow
->last_qtail
)) >= 0)) {
3922 rflow
= set_rps_cpu(dev
, skb
, rflow
, next_cpu
);
3925 if (tcpu
< nr_cpu_ids
&& cpu_online(tcpu
)) {
3935 tcpu
= map
->cpus
[reciprocal_scale(hash
, map
->len
)];
3936 if (cpu_online(tcpu
)) {
3946 #ifdef CONFIG_RFS_ACCEL
3949 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3950 * @dev: Device on which the filter was set
3951 * @rxq_index: RX queue index
3952 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3953 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3955 * Drivers that implement ndo_rx_flow_steer() should periodically call
3956 * this function for each installed filter and remove the filters for
3957 * which it returns %true.
3959 bool rps_may_expire_flow(struct net_device
*dev
, u16 rxq_index
,
3960 u32 flow_id
, u16 filter_id
)
3962 struct netdev_rx_queue
*rxqueue
= dev
->_rx
+ rxq_index
;
3963 struct rps_dev_flow_table
*flow_table
;
3964 struct rps_dev_flow
*rflow
;
3969 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
3970 if (flow_table
&& flow_id
<= flow_table
->mask
) {
3971 rflow
= &flow_table
->flows
[flow_id
];
3972 cpu
= READ_ONCE(rflow
->cpu
);
3973 if (rflow
->filter
== filter_id
&& cpu
< nr_cpu_ids
&&
3974 ((int)(per_cpu(softnet_data
, cpu
).input_queue_head
-
3975 rflow
->last_qtail
) <
3976 (int)(10 * flow_table
->mask
)))
3982 EXPORT_SYMBOL(rps_may_expire_flow
);
3984 #endif /* CONFIG_RFS_ACCEL */
3986 /* Called from hardirq (IPI) context */
3987 static void rps_trigger_softirq(void *data
)
3989 struct softnet_data
*sd
= data
;
3991 ____napi_schedule(sd
, &sd
->backlog
);
3995 #endif /* CONFIG_RPS */
3998 * Check if this softnet_data structure is another cpu one
3999 * If yes, queue it to our IPI list and return 1
4002 static int rps_ipi_queued(struct softnet_data
*sd
)
4005 struct softnet_data
*mysd
= this_cpu_ptr(&softnet_data
);
4008 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
4009 mysd
->rps_ipi_list
= sd
;
4011 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
4014 #endif /* CONFIG_RPS */
4018 #ifdef CONFIG_NET_FLOW_LIMIT
4019 int netdev_flow_limit_table_len __read_mostly
= (1 << 12);
4022 static bool skb_flow_limit(struct sk_buff
*skb
, unsigned int qlen
)
4024 #ifdef CONFIG_NET_FLOW_LIMIT
4025 struct sd_flow_limit
*fl
;
4026 struct softnet_data
*sd
;
4027 unsigned int old_flow
, new_flow
;
4029 if (qlen
< (netdev_max_backlog
>> 1))
4032 sd
= this_cpu_ptr(&softnet_data
);
4035 fl
= rcu_dereference(sd
->flow_limit
);
4037 new_flow
= skb_get_hash(skb
) & (fl
->num_buckets
- 1);
4038 old_flow
= fl
->history
[fl
->history_head
];
4039 fl
->history
[fl
->history_head
] = new_flow
;
4042 fl
->history_head
&= FLOW_LIMIT_HISTORY
- 1;
4044 if (likely(fl
->buckets
[old_flow
]))
4045 fl
->buckets
[old_flow
]--;
4047 if (++fl
->buckets
[new_flow
] > (FLOW_LIMIT_HISTORY
>> 1)) {
4059 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4060 * queue (may be a remote CPU queue).
4062 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
4063 unsigned int *qtail
)
4065 struct softnet_data
*sd
;
4066 unsigned long flags
;
4069 sd
= &per_cpu(softnet_data
, cpu
);
4071 local_irq_save(flags
);
4074 if (!netif_running(skb
->dev
))
4076 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
4077 if (qlen
<= netdev_max_backlog
&& !skb_flow_limit(skb
, qlen
)) {
4080 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
4081 input_queue_tail_incr_save(sd
, qtail
);
4083 local_irq_restore(flags
);
4084 return NET_RX_SUCCESS
;
4087 /* Schedule NAPI for backlog device
4088 * We can use non atomic operation since we own the queue lock
4090 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
4091 if (!rps_ipi_queued(sd
))
4092 ____napi_schedule(sd
, &sd
->backlog
);
4101 local_irq_restore(flags
);
4103 atomic_long_inc(&skb
->dev
->rx_dropped
);
4108 static struct netdev_rx_queue
*netif_get_rxqueue(struct sk_buff
*skb
)
4110 struct net_device
*dev
= skb
->dev
;
4111 struct netdev_rx_queue
*rxqueue
;
4115 if (skb_rx_queue_recorded(skb
)) {
4116 u16 index
= skb_get_rx_queue(skb
);
4118 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
4119 WARN_ONCE(dev
->real_num_rx_queues
> 1,
4120 "%s received packet on queue %u, but number "
4121 "of RX queues is %u\n",
4122 dev
->name
, index
, dev
->real_num_rx_queues
);
4124 return rxqueue
; /* Return first rxqueue */
4131 static u32
netif_receive_generic_xdp(struct sk_buff
*skb
,
4132 struct xdp_buff
*xdp
,
4133 struct bpf_prog
*xdp_prog
)
4135 struct netdev_rx_queue
*rxqueue
;
4136 void *orig_data
, *orig_data_end
;
4137 u32 metalen
, act
= XDP_DROP
;
4141 /* Reinjected packets coming from act_mirred or similar should
4142 * not get XDP generic processing.
4144 if (skb_cloned(skb
))
4147 /* XDP packets must be linear and must have sufficient headroom
4148 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4149 * native XDP provides, thus we need to do it here as well.
4151 if (skb_is_nonlinear(skb
) ||
4152 skb_headroom(skb
) < XDP_PACKET_HEADROOM
) {
4153 int hroom
= XDP_PACKET_HEADROOM
- skb_headroom(skb
);
4154 int troom
= skb
->tail
+ skb
->data_len
- skb
->end
;
4156 /* In case we have to go down the path and also linearize,
4157 * then lets do the pskb_expand_head() work just once here.
4159 if (pskb_expand_head(skb
,
4160 hroom
> 0 ? ALIGN(hroom
, NET_SKB_PAD
) : 0,
4161 troom
> 0 ? troom
+ 128 : 0, GFP_ATOMIC
))
4163 if (skb_linearize(skb
))
4167 /* The XDP program wants to see the packet starting at the MAC
4170 mac_len
= skb
->data
- skb_mac_header(skb
);
4171 hlen
= skb_headlen(skb
) + mac_len
;
4172 xdp
->data
= skb
->data
- mac_len
;
4173 xdp
->data_meta
= xdp
->data
;
4174 xdp
->data_end
= xdp
->data
+ hlen
;
4175 xdp
->data_hard_start
= skb
->data
- skb_headroom(skb
);
4176 orig_data_end
= xdp
->data_end
;
4177 orig_data
= xdp
->data
;
4179 rxqueue
= netif_get_rxqueue(skb
);
4180 xdp
->rxq
= &rxqueue
->xdp_rxq
;
4182 act
= bpf_prog_run_xdp(xdp_prog
, xdp
);
4184 off
= xdp
->data
- orig_data
;
4186 __skb_pull(skb
, off
);
4188 __skb_push(skb
, -off
);
4189 skb
->mac_header
+= off
;
4191 /* check if bpf_xdp_adjust_tail was used. it can only "shrink"
4194 off
= orig_data_end
- xdp
->data_end
;
4196 skb_set_tail_pointer(skb
, xdp
->data_end
- xdp
->data
);
4204 __skb_push(skb
, mac_len
);
4207 metalen
= xdp
->data
- xdp
->data_meta
;
4209 skb_metadata_set(skb
, metalen
);
4212 bpf_warn_invalid_xdp_action(act
);
4215 trace_xdp_exception(skb
->dev
, xdp_prog
, act
);
4226 /* When doing generic XDP we have to bypass the qdisc layer and the
4227 * network taps in order to match in-driver-XDP behavior.
4229 void generic_xdp_tx(struct sk_buff
*skb
, struct bpf_prog
*xdp_prog
)
4231 struct net_device
*dev
= skb
->dev
;
4232 struct netdev_queue
*txq
;
4233 bool free_skb
= true;
4236 txq
= netdev_pick_tx(dev
, skb
, NULL
);
4237 cpu
= smp_processor_id();
4238 HARD_TX_LOCK(dev
, txq
, cpu
);
4239 if (!netif_xmit_stopped(txq
)) {
4240 rc
= netdev_start_xmit(skb
, dev
, txq
, 0);
4241 if (dev_xmit_complete(rc
))
4244 HARD_TX_UNLOCK(dev
, txq
);
4246 trace_xdp_exception(dev
, xdp_prog
, XDP_TX
);
4250 EXPORT_SYMBOL_GPL(generic_xdp_tx
);
4252 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key
);
4254 int do_xdp_generic(struct bpf_prog
*xdp_prog
, struct sk_buff
*skb
)
4257 struct xdp_buff xdp
;
4261 act
= netif_receive_generic_xdp(skb
, &xdp
, xdp_prog
);
4262 if (act
!= XDP_PASS
) {
4265 err
= xdp_do_generic_redirect(skb
->dev
, skb
,
4271 generic_xdp_tx(skb
, xdp_prog
);
4282 EXPORT_SYMBOL_GPL(do_xdp_generic
);
4284 static int netif_rx_internal(struct sk_buff
*skb
)
4288 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
4290 trace_netif_rx(skb
);
4292 if (static_branch_unlikely(&generic_xdp_needed_key
)) {
4297 ret
= do_xdp_generic(rcu_dereference(skb
->dev
->xdp_prog
), skb
);
4301 /* Consider XDP consuming the packet a success from
4302 * the netdev point of view we do not want to count
4305 if (ret
!= XDP_PASS
)
4306 return NET_RX_SUCCESS
;
4310 if (static_key_false(&rps_needed
)) {
4311 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
4317 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
4319 cpu
= smp_processor_id();
4321 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
4330 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
4337 * netif_rx - post buffer to the network code
4338 * @skb: buffer to post
4340 * This function receives a packet from a device driver and queues it for
4341 * the upper (protocol) levels to process. It always succeeds. The buffer
4342 * may be dropped during processing for congestion control or by the
4346 * NET_RX_SUCCESS (no congestion)
4347 * NET_RX_DROP (packet was dropped)
4351 int netif_rx(struct sk_buff
*skb
)
4353 trace_netif_rx_entry(skb
);
4355 return netif_rx_internal(skb
);
4357 EXPORT_SYMBOL(netif_rx
);
4359 int netif_rx_ni(struct sk_buff
*skb
)
4363 trace_netif_rx_ni_entry(skb
);
4366 err
= netif_rx_internal(skb
);
4367 if (local_softirq_pending())
4373 EXPORT_SYMBOL(netif_rx_ni
);
4375 static __latent_entropy
void net_tx_action(struct softirq_action
*h
)
4377 struct softnet_data
*sd
= this_cpu_ptr(&softnet_data
);
4379 if (sd
->completion_queue
) {
4380 struct sk_buff
*clist
;
4382 local_irq_disable();
4383 clist
= sd
->completion_queue
;
4384 sd
->completion_queue
= NULL
;
4388 struct sk_buff
*skb
= clist
;
4390 clist
= clist
->next
;
4392 WARN_ON(refcount_read(&skb
->users
));
4393 if (likely(get_kfree_skb_cb(skb
)->reason
== SKB_REASON_CONSUMED
))
4394 trace_consume_skb(skb
);
4396 trace_kfree_skb(skb
, net_tx_action
);
4398 if (skb
->fclone
!= SKB_FCLONE_UNAVAILABLE
)
4401 __kfree_skb_defer(skb
);
4404 __kfree_skb_flush();
4407 if (sd
->output_queue
) {
4410 local_irq_disable();
4411 head
= sd
->output_queue
;
4412 sd
->output_queue
= NULL
;
4413 sd
->output_queue_tailp
= &sd
->output_queue
;
4417 struct Qdisc
*q
= head
;
4418 spinlock_t
*root_lock
= NULL
;
4420 head
= head
->next_sched
;
4422 if (!(q
->flags
& TCQ_F_NOLOCK
)) {
4423 root_lock
= qdisc_lock(q
);
4424 spin_lock(root_lock
);
4426 /* We need to make sure head->next_sched is read
4427 * before clearing __QDISC_STATE_SCHED
4429 smp_mb__before_atomic();
4430 clear_bit(__QDISC_STATE_SCHED
, &q
->state
);
4433 spin_unlock(root_lock
);
4437 xfrm_dev_backlog(sd
);
4440 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
4441 /* This hook is defined here for ATM LANE */
4442 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
4443 unsigned char *addr
) __read_mostly
;
4444 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
4447 static inline struct sk_buff
*
4448 sch_handle_ingress(struct sk_buff
*skb
, struct packet_type
**pt_prev
, int *ret
,
4449 struct net_device
*orig_dev
)
4451 #ifdef CONFIG_NET_CLS_ACT
4452 struct mini_Qdisc
*miniq
= rcu_dereference_bh(skb
->dev
->miniq_ingress
);
4453 struct tcf_result cl_res
;
4455 /* If there's at least one ingress present somewhere (so
4456 * we get here via enabled static key), remaining devices
4457 * that are not configured with an ingress qdisc will bail
4464 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
4468 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
4469 skb
->tc_at_ingress
= 1;
4470 mini_qdisc_bstats_cpu_update(miniq
, skb
);
4472 switch (tcf_classify(skb
, miniq
->filter_list
, &cl_res
, false)) {
4474 case TC_ACT_RECLASSIFY
:
4475 skb
->tc_index
= TC_H_MIN(cl_res
.classid
);
4478 mini_qdisc_qstats_cpu_drop(miniq
);
4486 case TC_ACT_REDIRECT
:
4487 /* skb_mac_header check was done by cls/act_bpf, so
4488 * we can safely push the L2 header back before
4489 * redirecting to another netdev
4491 __skb_push(skb
, skb
->mac_len
);
4492 skb_do_redirect(skb
);
4497 #endif /* CONFIG_NET_CLS_ACT */
4502 * netdev_is_rx_handler_busy - check if receive handler is registered
4503 * @dev: device to check
4505 * Check if a receive handler is already registered for a given device.
4506 * Return true if there one.
4508 * The caller must hold the rtnl_mutex.
4510 bool netdev_is_rx_handler_busy(struct net_device
*dev
)
4513 return dev
&& rtnl_dereference(dev
->rx_handler
);
4515 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy
);
4518 * netdev_rx_handler_register - register receive handler
4519 * @dev: device to register a handler for
4520 * @rx_handler: receive handler to register
4521 * @rx_handler_data: data pointer that is used by rx handler
4523 * Register a receive handler for a device. This handler will then be
4524 * called from __netif_receive_skb. A negative errno code is returned
4527 * The caller must hold the rtnl_mutex.
4529 * For a general description of rx_handler, see enum rx_handler_result.
4531 int netdev_rx_handler_register(struct net_device
*dev
,
4532 rx_handler_func_t
*rx_handler
,
4533 void *rx_handler_data
)
4535 if (netdev_is_rx_handler_busy(dev
))
4538 if (dev
->priv_flags
& IFF_NO_RX_HANDLER
)
4541 /* Note: rx_handler_data must be set before rx_handler */
4542 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
4543 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
4547 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
4550 * netdev_rx_handler_unregister - unregister receive handler
4551 * @dev: device to unregister a handler from
4553 * Unregister a receive handler from a device.
4555 * The caller must hold the rtnl_mutex.
4557 void netdev_rx_handler_unregister(struct net_device
*dev
)
4561 RCU_INIT_POINTER(dev
->rx_handler
, NULL
);
4562 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
4563 * section has a guarantee to see a non NULL rx_handler_data
4567 RCU_INIT_POINTER(dev
->rx_handler_data
, NULL
);
4569 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
4572 * Limit the use of PFMEMALLOC reserves to those protocols that implement
4573 * the special handling of PFMEMALLOC skbs.
4575 static bool skb_pfmemalloc_protocol(struct sk_buff
*skb
)
4577 switch (skb
->protocol
) {
4578 case htons(ETH_P_ARP
):
4579 case htons(ETH_P_IP
):
4580 case htons(ETH_P_IPV6
):
4581 case htons(ETH_P_8021Q
):
4582 case htons(ETH_P_8021AD
):
4589 static inline int nf_ingress(struct sk_buff
*skb
, struct packet_type
**pt_prev
,
4590 int *ret
, struct net_device
*orig_dev
)
4592 #ifdef CONFIG_NETFILTER_INGRESS
4593 if (nf_hook_ingress_active(skb
)) {
4597 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
4602 ingress_retval
= nf_hook_ingress(skb
);
4604 return ingress_retval
;
4606 #endif /* CONFIG_NETFILTER_INGRESS */
4610 static int __netif_receive_skb_core(struct sk_buff
*skb
, bool pfmemalloc
,
4611 struct packet_type
**ppt_prev
)
4613 struct packet_type
*ptype
, *pt_prev
;
4614 rx_handler_func_t
*rx_handler
;
4615 struct net_device
*orig_dev
;
4616 bool deliver_exact
= false;
4617 int ret
= NET_RX_DROP
;
4620 net_timestamp_check(!netdev_tstamp_prequeue
, skb
);
4622 trace_netif_receive_skb(skb
);
4624 orig_dev
= skb
->dev
;
4626 skb_reset_network_header(skb
);
4627 if (!skb_transport_header_was_set(skb
))
4628 skb_reset_transport_header(skb
);
4629 skb_reset_mac_len(skb
);
4634 skb
->skb_iif
= skb
->dev
->ifindex
;
4636 __this_cpu_inc(softnet_data
.processed
);
4638 if (skb
->protocol
== cpu_to_be16(ETH_P_8021Q
) ||
4639 skb
->protocol
== cpu_to_be16(ETH_P_8021AD
)) {
4640 skb
= skb_vlan_untag(skb
);
4645 if (skb_skip_tc_classify(skb
))
4651 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
4653 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
4657 list_for_each_entry_rcu(ptype
, &skb
->dev
->ptype_all
, list
) {
4659 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
4664 #ifdef CONFIG_NET_INGRESS
4665 if (static_branch_unlikely(&ingress_needed_key
)) {
4666 skb
= sch_handle_ingress(skb
, &pt_prev
, &ret
, orig_dev
);
4670 if (nf_ingress(skb
, &pt_prev
, &ret
, orig_dev
) < 0)
4676 if (pfmemalloc
&& !skb_pfmemalloc_protocol(skb
))
4679 if (skb_vlan_tag_present(skb
)) {
4681 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
4684 if (vlan_do_receive(&skb
))
4686 else if (unlikely(!skb
))
4690 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
4693 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
4696 switch (rx_handler(&skb
)) {
4697 case RX_HANDLER_CONSUMED
:
4698 ret
= NET_RX_SUCCESS
;
4700 case RX_HANDLER_ANOTHER
:
4702 case RX_HANDLER_EXACT
:
4703 deliver_exact
= true;
4704 case RX_HANDLER_PASS
:
4711 if (unlikely(skb_vlan_tag_present(skb
))) {
4712 if (skb_vlan_tag_get_id(skb
))
4713 skb
->pkt_type
= PACKET_OTHERHOST
;
4714 /* Note: we might in the future use prio bits
4715 * and set skb->priority like in vlan_do_receive()
4716 * For the time being, just ignore Priority Code Point
4721 type
= skb
->protocol
;
4723 /* deliver only exact match when indicated */
4724 if (likely(!deliver_exact
)) {
4725 deliver_ptype_list_skb(skb
, &pt_prev
, orig_dev
, type
,
4726 &ptype_base
[ntohs(type
) &
4730 deliver_ptype_list_skb(skb
, &pt_prev
, orig_dev
, type
,
4731 &orig_dev
->ptype_specific
);
4733 if (unlikely(skb
->dev
!= orig_dev
)) {
4734 deliver_ptype_list_skb(skb
, &pt_prev
, orig_dev
, type
,
4735 &skb
->dev
->ptype_specific
);
4739 if (unlikely(skb_orphan_frags_rx(skb
, GFP_ATOMIC
)))
4741 *ppt_prev
= pt_prev
;
4745 atomic_long_inc(&skb
->dev
->rx_dropped
);
4747 atomic_long_inc(&skb
->dev
->rx_nohandler
);
4749 /* Jamal, now you will not able to escape explaining
4750 * me how you were going to use this. :-)
4759 static int __netif_receive_skb_one_core(struct sk_buff
*skb
, bool pfmemalloc
)
4761 struct net_device
*orig_dev
= skb
->dev
;
4762 struct packet_type
*pt_prev
= NULL
;
4765 ret
= __netif_receive_skb_core(skb
, pfmemalloc
, &pt_prev
);
4767 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
4772 * netif_receive_skb_core - special purpose version of netif_receive_skb
4773 * @skb: buffer to process
4775 * More direct receive version of netif_receive_skb(). It should
4776 * only be used by callers that have a need to skip RPS and Generic XDP.
4777 * Caller must also take care of handling if (page_is_)pfmemalloc.
4779 * This function may only be called from softirq context and interrupts
4780 * should be enabled.
4782 * Return values (usually ignored):
4783 * NET_RX_SUCCESS: no congestion
4784 * NET_RX_DROP: packet was dropped
4786 int netif_receive_skb_core(struct sk_buff
*skb
)
4791 ret
= __netif_receive_skb_one_core(skb
, false);
4796 EXPORT_SYMBOL(netif_receive_skb_core
);
4798 static inline void __netif_receive_skb_list_ptype(struct list_head
*head
,
4799 struct packet_type
*pt_prev
,
4800 struct net_device
*orig_dev
)
4802 struct sk_buff
*skb
, *next
;
4806 if (list_empty(head
))
4808 if (pt_prev
->list_func
!= NULL
)
4809 pt_prev
->list_func(head
, pt_prev
, orig_dev
);
4811 list_for_each_entry_safe(skb
, next
, head
, list
)
4812 pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
4815 static void __netif_receive_skb_list_core(struct list_head
*head
, bool pfmemalloc
)
4817 /* Fast-path assumptions:
4818 * - There is no RX handler.
4819 * - Only one packet_type matches.
4820 * If either of these fails, we will end up doing some per-packet
4821 * processing in-line, then handling the 'last ptype' for the whole
4822 * sublist. This can't cause out-of-order delivery to any single ptype,
4823 * because the 'last ptype' must be constant across the sublist, and all
4824 * other ptypes are handled per-packet.
4826 /* Current (common) ptype of sublist */
4827 struct packet_type
*pt_curr
= NULL
;
4828 /* Current (common) orig_dev of sublist */
4829 struct net_device
*od_curr
= NULL
;
4830 struct list_head sublist
;
4831 struct sk_buff
*skb
, *next
;
4833 list_for_each_entry_safe(skb
, next
, head
, list
) {
4834 struct net_device
*orig_dev
= skb
->dev
;
4835 struct packet_type
*pt_prev
= NULL
;
4837 __netif_receive_skb_core(skb
, pfmemalloc
, &pt_prev
);
4838 if (pt_curr
!= pt_prev
|| od_curr
!= orig_dev
) {
4839 /* dispatch old sublist */
4840 list_cut_before(&sublist
, head
, &skb
->list
);
4841 __netif_receive_skb_list_ptype(&sublist
, pt_curr
, od_curr
);
4842 /* start new sublist */
4848 /* dispatch final sublist */
4849 __netif_receive_skb_list_ptype(head
, pt_curr
, od_curr
);
4852 static int __netif_receive_skb(struct sk_buff
*skb
)
4856 if (sk_memalloc_socks() && skb_pfmemalloc(skb
)) {
4857 unsigned int noreclaim_flag
;
4860 * PFMEMALLOC skbs are special, they should
4861 * - be delivered to SOCK_MEMALLOC sockets only
4862 * - stay away from userspace
4863 * - have bounded memory usage
4865 * Use PF_MEMALLOC as this saves us from propagating the allocation
4866 * context down to all allocation sites.
4868 noreclaim_flag
= memalloc_noreclaim_save();
4869 ret
= __netif_receive_skb_one_core(skb
, true);
4870 memalloc_noreclaim_restore(noreclaim_flag
);
4872 ret
= __netif_receive_skb_one_core(skb
, false);
4877 static void __netif_receive_skb_list(struct list_head
*head
)
4879 unsigned long noreclaim_flag
= 0;
4880 struct sk_buff
*skb
, *next
;
4881 bool pfmemalloc
= false; /* Is current sublist PF_MEMALLOC? */
4883 list_for_each_entry_safe(skb
, next
, head
, list
) {
4884 if ((sk_memalloc_socks() && skb_pfmemalloc(skb
)) != pfmemalloc
) {
4885 struct list_head sublist
;
4887 /* Handle the previous sublist */
4888 list_cut_before(&sublist
, head
, &skb
->list
);
4889 if (!list_empty(&sublist
))
4890 __netif_receive_skb_list_core(&sublist
, pfmemalloc
);
4891 pfmemalloc
= !pfmemalloc
;
4892 /* See comments in __netif_receive_skb */
4894 noreclaim_flag
= memalloc_noreclaim_save();
4896 memalloc_noreclaim_restore(noreclaim_flag
);
4899 /* Handle the remaining sublist */
4900 if (!list_empty(head
))
4901 __netif_receive_skb_list_core(head
, pfmemalloc
);
4902 /* Restore pflags */
4904 memalloc_noreclaim_restore(noreclaim_flag
);
4907 static int generic_xdp_install(struct net_device
*dev
, struct netdev_bpf
*xdp
)
4909 struct bpf_prog
*old
= rtnl_dereference(dev
->xdp_prog
);
4910 struct bpf_prog
*new = xdp
->prog
;
4913 switch (xdp
->command
) {
4914 case XDP_SETUP_PROG
:
4915 rcu_assign_pointer(dev
->xdp_prog
, new);
4920 static_branch_dec(&generic_xdp_needed_key
);
4921 } else if (new && !old
) {
4922 static_branch_inc(&generic_xdp_needed_key
);
4923 dev_disable_lro(dev
);
4924 dev_disable_gro_hw(dev
);
4928 case XDP_QUERY_PROG
:
4929 xdp
->prog_attached
= !!old
;
4930 xdp
->prog_id
= old
? old
->aux
->id
: 0;
4941 static int netif_receive_skb_internal(struct sk_buff
*skb
)
4945 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
4947 if (skb_defer_rx_timestamp(skb
))
4948 return NET_RX_SUCCESS
;
4950 if (static_branch_unlikely(&generic_xdp_needed_key
)) {
4955 ret
= do_xdp_generic(rcu_dereference(skb
->dev
->xdp_prog
), skb
);
4959 if (ret
!= XDP_PASS
)
4965 if (static_key_false(&rps_needed
)) {
4966 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
4967 int cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
4970 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
4976 ret
= __netif_receive_skb(skb
);
4981 static void netif_receive_skb_list_internal(struct list_head
*head
)
4983 struct bpf_prog
*xdp_prog
= NULL
;
4984 struct sk_buff
*skb
, *next
;
4986 list_for_each_entry_safe(skb
, next
, head
, list
) {
4987 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
4988 if (skb_defer_rx_timestamp(skb
))
4989 /* Handled, remove from list */
4990 list_del(&skb
->list
);
4993 if (static_branch_unlikely(&generic_xdp_needed_key
)) {
4996 list_for_each_entry_safe(skb
, next
, head
, list
) {
4997 xdp_prog
= rcu_dereference(skb
->dev
->xdp_prog
);
4998 if (do_xdp_generic(xdp_prog
, skb
) != XDP_PASS
)
4999 /* Dropped, remove from list */
5000 list_del(&skb
->list
);
5008 if (static_key_false(&rps_needed
)) {
5009 list_for_each_entry_safe(skb
, next
, head
, list
) {
5010 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
5011 int cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
5014 enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
5015 /* Handled, remove from list */
5016 list_del(&skb
->list
);
5021 __netif_receive_skb_list(head
);
5026 * netif_receive_skb - process receive buffer from network
5027 * @skb: buffer to process
5029 * netif_receive_skb() is the main receive data processing function.
5030 * It always succeeds. The buffer may be dropped during processing
5031 * for congestion control or by the protocol layers.
5033 * This function may only be called from softirq context and interrupts
5034 * should be enabled.
5036 * Return values (usually ignored):
5037 * NET_RX_SUCCESS: no congestion
5038 * NET_RX_DROP: packet was dropped
5040 int netif_receive_skb(struct sk_buff
*skb
)
5042 trace_netif_receive_skb_entry(skb
);
5044 return netif_receive_skb_internal(skb
);
5046 EXPORT_SYMBOL(netif_receive_skb
);
5049 * netif_receive_skb_list - process many receive buffers from network
5050 * @head: list of skbs to process.
5052 * Since return value of netif_receive_skb() is normally ignored, and
5053 * wouldn't be meaningful for a list, this function returns void.
5055 * This function may only be called from softirq context and interrupts
5056 * should be enabled.
5058 void netif_receive_skb_list(struct list_head
*head
)
5060 struct sk_buff
*skb
;
5062 if (list_empty(head
))
5064 list_for_each_entry(skb
, head
, list
)
5065 trace_netif_receive_skb_list_entry(skb
);
5066 netif_receive_skb_list_internal(head
);
5068 EXPORT_SYMBOL(netif_receive_skb_list
);
5070 DEFINE_PER_CPU(struct work_struct
, flush_works
);
5072 /* Network device is going away, flush any packets still pending */
5073 static void flush_backlog(struct work_struct
*work
)
5075 struct sk_buff
*skb
, *tmp
;
5076 struct softnet_data
*sd
;
5079 sd
= this_cpu_ptr(&softnet_data
);
5081 local_irq_disable();
5083 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
5084 if (skb
->dev
->reg_state
== NETREG_UNREGISTERING
) {
5085 __skb_unlink(skb
, &sd
->input_pkt_queue
);
5087 input_queue_head_incr(sd
);
5093 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
5094 if (skb
->dev
->reg_state
== NETREG_UNREGISTERING
) {
5095 __skb_unlink(skb
, &sd
->process_queue
);
5097 input_queue_head_incr(sd
);
5103 static void flush_all_backlogs(void)
5109 for_each_online_cpu(cpu
)
5110 queue_work_on(cpu
, system_highpri_wq
,
5111 per_cpu_ptr(&flush_works
, cpu
));
5113 for_each_online_cpu(cpu
)
5114 flush_work(per_cpu_ptr(&flush_works
, cpu
));
5119 static int napi_gro_complete(struct sk_buff
*skb
)
5121 struct packet_offload
*ptype
;
5122 __be16 type
= skb
->protocol
;
5123 struct list_head
*head
= &offload_base
;
5126 BUILD_BUG_ON(sizeof(struct napi_gro_cb
) > sizeof(skb
->cb
));
5128 if (NAPI_GRO_CB(skb
)->count
== 1) {
5129 skb_shinfo(skb
)->gso_size
= 0;
5134 list_for_each_entry_rcu(ptype
, head
, list
) {
5135 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_complete
)
5138 err
= ptype
->callbacks
.gro_complete(skb
, 0);
5144 WARN_ON(&ptype
->list
== head
);
5146 return NET_RX_SUCCESS
;
5150 return netif_receive_skb_internal(skb
);
5153 static void __napi_gro_flush_chain(struct napi_struct
*napi
, u32 index
,
5156 struct list_head
*head
= &napi
->gro_hash
[index
].list
;
5157 struct sk_buff
*skb
, *p
;
5159 list_for_each_entry_safe_reverse(skb
, p
, head
, list
) {
5160 if (flush_old
&& NAPI_GRO_CB(skb
)->age
== jiffies
)
5162 list_del_init(&skb
->list
);
5163 napi_gro_complete(skb
);
5165 napi
->gro_hash
[index
].count
--;
5169 /* napi->gro_hash[].list contains packets ordered by age.
5170 * youngest packets at the head of it.
5171 * Complete skbs in reverse order to reduce latencies.
5173 void napi_gro_flush(struct napi_struct
*napi
, bool flush_old
)
5177 for (i
= 0; i
< GRO_HASH_BUCKETS
; i
++)
5178 __napi_gro_flush_chain(napi
, i
, flush_old
);
5180 EXPORT_SYMBOL(napi_gro_flush
);
5182 static struct list_head
*gro_list_prepare(struct napi_struct
*napi
,
5183 struct sk_buff
*skb
)
5185 unsigned int maclen
= skb
->dev
->hard_header_len
;
5186 u32 hash
= skb_get_hash_raw(skb
);
5187 struct list_head
*head
;
5190 head
= &napi
->gro_hash
[hash
& (GRO_HASH_BUCKETS
- 1)].list
;
5191 list_for_each_entry(p
, head
, list
) {
5192 unsigned long diffs
;
5194 NAPI_GRO_CB(p
)->flush
= 0;
5196 if (hash
!= skb_get_hash_raw(p
)) {
5197 NAPI_GRO_CB(p
)->same_flow
= 0;
5201 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
5202 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
5203 diffs
|= skb_metadata_dst_cmp(p
, skb
);
5204 diffs
|= skb_metadata_differs(p
, skb
);
5205 if (maclen
== ETH_HLEN
)
5206 diffs
|= compare_ether_header(skb_mac_header(p
),
5207 skb_mac_header(skb
));
5209 diffs
= memcmp(skb_mac_header(p
),
5210 skb_mac_header(skb
),
5212 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
5218 static void skb_gro_reset_offset(struct sk_buff
*skb
)
5220 const struct skb_shared_info
*pinfo
= skb_shinfo(skb
);
5221 const skb_frag_t
*frag0
= &pinfo
->frags
[0];
5223 NAPI_GRO_CB(skb
)->data_offset
= 0;
5224 NAPI_GRO_CB(skb
)->frag0
= NULL
;
5225 NAPI_GRO_CB(skb
)->frag0_len
= 0;
5227 if (skb_mac_header(skb
) == skb_tail_pointer(skb
) &&
5229 !PageHighMem(skb_frag_page(frag0
))) {
5230 NAPI_GRO_CB(skb
)->frag0
= skb_frag_address(frag0
);
5231 NAPI_GRO_CB(skb
)->frag0_len
= min_t(unsigned int,
5232 skb_frag_size(frag0
),
5233 skb
->end
- skb
->tail
);
5237 static void gro_pull_from_frag0(struct sk_buff
*skb
, int grow
)
5239 struct skb_shared_info
*pinfo
= skb_shinfo(skb
);
5241 BUG_ON(skb
->end
- skb
->tail
< grow
);
5243 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
5245 skb
->data_len
-= grow
;
5248 pinfo
->frags
[0].page_offset
+= grow
;
5249 skb_frag_size_sub(&pinfo
->frags
[0], grow
);
5251 if (unlikely(!skb_frag_size(&pinfo
->frags
[0]))) {
5252 skb_frag_unref(skb
, 0);
5253 memmove(pinfo
->frags
, pinfo
->frags
+ 1,
5254 --pinfo
->nr_frags
* sizeof(pinfo
->frags
[0]));
5258 static void gro_flush_oldest(struct list_head
*head
)
5260 struct sk_buff
*oldest
;
5262 oldest
= list_last_entry(head
, struct sk_buff
, list
);
5264 /* We are called with head length >= MAX_GRO_SKBS, so this is
5267 if (WARN_ON_ONCE(!oldest
))
5270 /* Do not adjust napi->gro_count, caller is adding a new SKB to
5273 list_del(&oldest
->list
);
5274 napi_gro_complete(oldest
);
5277 static enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
5279 u32 hash
= skb_get_hash_raw(skb
) & (GRO_HASH_BUCKETS
- 1);
5280 struct list_head
*head
= &offload_base
;
5281 struct packet_offload
*ptype
;
5282 __be16 type
= skb
->protocol
;
5283 struct list_head
*gro_head
;
5284 struct sk_buff
*pp
= NULL
;
5285 enum gro_result ret
;
5289 if (netif_elide_gro(skb
->dev
))
5292 gro_head
= gro_list_prepare(napi
, skb
);
5295 list_for_each_entry_rcu(ptype
, head
, list
) {
5296 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_receive
)
5299 skb_set_network_header(skb
, skb_gro_offset(skb
));
5300 skb_reset_mac_len(skb
);
5301 NAPI_GRO_CB(skb
)->same_flow
= 0;
5302 NAPI_GRO_CB(skb
)->flush
= skb_is_gso(skb
) || skb_has_frag_list(skb
);
5303 NAPI_GRO_CB(skb
)->free
= 0;
5304 NAPI_GRO_CB(skb
)->encap_mark
= 0;
5305 NAPI_GRO_CB(skb
)->recursion_counter
= 0;
5306 NAPI_GRO_CB(skb
)->is_fou
= 0;
5307 NAPI_GRO_CB(skb
)->is_atomic
= 1;
5308 NAPI_GRO_CB(skb
)->gro_remcsum_start
= 0;
5310 /* Setup for GRO checksum validation */
5311 switch (skb
->ip_summed
) {
5312 case CHECKSUM_COMPLETE
:
5313 NAPI_GRO_CB(skb
)->csum
= skb
->csum
;
5314 NAPI_GRO_CB(skb
)->csum_valid
= 1;
5315 NAPI_GRO_CB(skb
)->csum_cnt
= 0;
5317 case CHECKSUM_UNNECESSARY
:
5318 NAPI_GRO_CB(skb
)->csum_cnt
= skb
->csum_level
+ 1;
5319 NAPI_GRO_CB(skb
)->csum_valid
= 0;
5322 NAPI_GRO_CB(skb
)->csum_cnt
= 0;
5323 NAPI_GRO_CB(skb
)->csum_valid
= 0;
5326 pp
= ptype
->callbacks
.gro_receive(gro_head
, skb
);
5331 if (&ptype
->list
== head
)
5334 if (IS_ERR(pp
) && PTR_ERR(pp
) == -EINPROGRESS
) {
5339 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
5340 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
5343 list_del_init(&pp
->list
);
5344 napi_gro_complete(pp
);
5346 napi
->gro_hash
[hash
].count
--;
5352 if (NAPI_GRO_CB(skb
)->flush
)
5355 if (unlikely(napi
->gro_hash
[hash
].count
>= MAX_GRO_SKBS
)) {
5356 gro_flush_oldest(gro_head
);
5359 napi
->gro_hash
[hash
].count
++;
5361 NAPI_GRO_CB(skb
)->count
= 1;
5362 NAPI_GRO_CB(skb
)->age
= jiffies
;
5363 NAPI_GRO_CB(skb
)->last
= skb
;
5364 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
5365 list_add(&skb
->list
, gro_head
);
5369 grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
5371 gro_pull_from_frag0(skb
, grow
);
5380 struct packet_offload
*gro_find_receive_by_type(__be16 type
)
5382 struct list_head
*offload_head
= &offload_base
;
5383 struct packet_offload
*ptype
;
5385 list_for_each_entry_rcu(ptype
, offload_head
, list
) {
5386 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_receive
)
5392 EXPORT_SYMBOL(gro_find_receive_by_type
);
5394 struct packet_offload
*gro_find_complete_by_type(__be16 type
)
5396 struct list_head
*offload_head
= &offload_base
;
5397 struct packet_offload
*ptype
;
5399 list_for_each_entry_rcu(ptype
, offload_head
, list
) {
5400 if (ptype
->type
!= type
|| !ptype
->callbacks
.gro_complete
)
5406 EXPORT_SYMBOL(gro_find_complete_by_type
);
5408 static void napi_skb_free_stolen_head(struct sk_buff
*skb
)
5412 kmem_cache_free(skbuff_head_cache
, skb
);
5415 static gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
5419 if (netif_receive_skb_internal(skb
))
5427 case GRO_MERGED_FREE
:
5428 if (NAPI_GRO_CB(skb
)->free
== NAPI_GRO_FREE_STOLEN_HEAD
)
5429 napi_skb_free_stolen_head(skb
);
5443 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
5445 skb_mark_napi_id(skb
, napi
);
5446 trace_napi_gro_receive_entry(skb
);
5448 skb_gro_reset_offset(skb
);
5450 return napi_skb_finish(dev_gro_receive(napi
, skb
), skb
);
5452 EXPORT_SYMBOL(napi_gro_receive
);
5454 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
5456 if (unlikely(skb
->pfmemalloc
)) {
5460 __skb_pull(skb
, skb_headlen(skb
));
5461 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
5462 skb_reserve(skb
, NET_SKB_PAD
+ NET_IP_ALIGN
- skb_headroom(skb
));
5464 skb
->dev
= napi
->dev
;
5466 skb
->encapsulation
= 0;
5467 skb_shinfo(skb
)->gso_type
= 0;
5468 skb
->truesize
= SKB_TRUESIZE(skb_end_offset(skb
));
5474 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
5476 struct sk_buff
*skb
= napi
->skb
;
5479 skb
= napi_alloc_skb(napi
, GRO_MAX_HEAD
);
5482 skb_mark_napi_id(skb
, napi
);
5487 EXPORT_SYMBOL(napi_get_frags
);
5489 static gro_result_t
napi_frags_finish(struct napi_struct
*napi
,
5490 struct sk_buff
*skb
,
5496 __skb_push(skb
, ETH_HLEN
);
5497 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
5498 if (ret
== GRO_NORMAL
&& netif_receive_skb_internal(skb
))
5503 napi_reuse_skb(napi
, skb
);
5506 case GRO_MERGED_FREE
:
5507 if (NAPI_GRO_CB(skb
)->free
== NAPI_GRO_FREE_STOLEN_HEAD
)
5508 napi_skb_free_stolen_head(skb
);
5510 napi_reuse_skb(napi
, skb
);
5521 /* Upper GRO stack assumes network header starts at gro_offset=0
5522 * Drivers could call both napi_gro_frags() and napi_gro_receive()
5523 * We copy ethernet header into skb->data to have a common layout.
5525 static struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
5527 struct sk_buff
*skb
= napi
->skb
;
5528 const struct ethhdr
*eth
;
5529 unsigned int hlen
= sizeof(*eth
);
5533 skb_reset_mac_header(skb
);
5534 skb_gro_reset_offset(skb
);
5536 eth
= skb_gro_header_fast(skb
, 0);
5537 if (unlikely(skb_gro_header_hard(skb
, hlen
))) {
5538 eth
= skb_gro_header_slow(skb
, hlen
, 0);
5539 if (unlikely(!eth
)) {
5540 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
5541 __func__
, napi
->dev
->name
);
5542 napi_reuse_skb(napi
, skb
);
5546 gro_pull_from_frag0(skb
, hlen
);
5547 NAPI_GRO_CB(skb
)->frag0
+= hlen
;
5548 NAPI_GRO_CB(skb
)->frag0_len
-= hlen
;
5550 __skb_pull(skb
, hlen
);
5553 * This works because the only protocols we care about don't require
5555 * We'll fix it up properly in napi_frags_finish()
5557 skb
->protocol
= eth
->h_proto
;
5562 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
5564 struct sk_buff
*skb
= napi_frags_skb(napi
);
5569 trace_napi_gro_frags_entry(skb
);
5571 return napi_frags_finish(napi
, skb
, dev_gro_receive(napi
, skb
));
5573 EXPORT_SYMBOL(napi_gro_frags
);
5575 /* Compute the checksum from gro_offset and return the folded value
5576 * after adding in any pseudo checksum.
5578 __sum16
__skb_gro_checksum_complete(struct sk_buff
*skb
)
5583 wsum
= skb_checksum(skb
, skb_gro_offset(skb
), skb_gro_len(skb
), 0);
5585 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
5586 sum
= csum_fold(csum_add(NAPI_GRO_CB(skb
)->csum
, wsum
));
5588 if (unlikely(skb
->ip_summed
== CHECKSUM_COMPLETE
) &&
5589 !skb
->csum_complete_sw
)
5590 netdev_rx_csum_fault(skb
->dev
);
5593 NAPI_GRO_CB(skb
)->csum
= wsum
;
5594 NAPI_GRO_CB(skb
)->csum_valid
= 1;
5598 EXPORT_SYMBOL(__skb_gro_checksum_complete
);
5600 static void net_rps_send_ipi(struct softnet_data
*remsd
)
5604 struct softnet_data
*next
= remsd
->rps_ipi_next
;
5606 if (cpu_online(remsd
->cpu
))
5607 smp_call_function_single_async(remsd
->cpu
, &remsd
->csd
);
5614 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
5615 * Note: called with local irq disabled, but exits with local irq enabled.
5617 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
5620 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
5623 sd
->rps_ipi_list
= NULL
;
5627 /* Send pending IPI's to kick RPS processing on remote cpus. */
5628 net_rps_send_ipi(remsd
);
5634 static bool sd_has_rps_ipi_waiting(struct softnet_data
*sd
)
5637 return sd
->rps_ipi_list
!= NULL
;
5643 static int process_backlog(struct napi_struct
*napi
, int quota
)
5645 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
5649 /* Check if we have pending ipi, its better to send them now,
5650 * not waiting net_rx_action() end.
5652 if (sd_has_rps_ipi_waiting(sd
)) {
5653 local_irq_disable();
5654 net_rps_action_and_irq_enable(sd
);
5657 napi
->weight
= dev_rx_weight
;
5659 struct sk_buff
*skb
;
5661 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
5663 __netif_receive_skb(skb
);
5665 input_queue_head_incr(sd
);
5666 if (++work
>= quota
)
5671 local_irq_disable();
5673 if (skb_queue_empty(&sd
->input_pkt_queue
)) {
5675 * Inline a custom version of __napi_complete().
5676 * only current cpu owns and manipulates this napi,
5677 * and NAPI_STATE_SCHED is the only possible flag set
5679 * We can use a plain write instead of clear_bit(),
5680 * and we dont need an smp_mb() memory barrier.
5685 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
5686 &sd
->process_queue
);
5696 * __napi_schedule - schedule for receive
5697 * @n: entry to schedule
5699 * The entry's receive function will be scheduled to run.
5700 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
5702 void __napi_schedule(struct napi_struct
*n
)
5704 unsigned long flags
;
5706 local_irq_save(flags
);
5707 ____napi_schedule(this_cpu_ptr(&softnet_data
), n
);
5708 local_irq_restore(flags
);
5710 EXPORT_SYMBOL(__napi_schedule
);
5713 * napi_schedule_prep - check if napi can be scheduled
5716 * Test if NAPI routine is already running, and if not mark
5717 * it as running. This is used as a condition variable
5718 * insure only one NAPI poll instance runs. We also make
5719 * sure there is no pending NAPI disable.
5721 bool napi_schedule_prep(struct napi_struct
*n
)
5723 unsigned long val
, new;
5726 val
= READ_ONCE(n
->state
);
5727 if (unlikely(val
& NAPIF_STATE_DISABLE
))
5729 new = val
| NAPIF_STATE_SCHED
;
5731 /* Sets STATE_MISSED bit if STATE_SCHED was already set
5732 * This was suggested by Alexander Duyck, as compiler
5733 * emits better code than :
5734 * if (val & NAPIF_STATE_SCHED)
5735 * new |= NAPIF_STATE_MISSED;
5737 new |= (val
& NAPIF_STATE_SCHED
) / NAPIF_STATE_SCHED
*
5739 } while (cmpxchg(&n
->state
, val
, new) != val
);
5741 return !(val
& NAPIF_STATE_SCHED
);
5743 EXPORT_SYMBOL(napi_schedule_prep
);
5746 * __napi_schedule_irqoff - schedule for receive
5747 * @n: entry to schedule
5749 * Variant of __napi_schedule() assuming hard irqs are masked
5751 void __napi_schedule_irqoff(struct napi_struct
*n
)
5753 ____napi_schedule(this_cpu_ptr(&softnet_data
), n
);
5755 EXPORT_SYMBOL(__napi_schedule_irqoff
);
5757 bool napi_complete_done(struct napi_struct
*n
, int work_done
)
5759 unsigned long flags
, val
, new;
5762 * 1) Don't let napi dequeue from the cpu poll list
5763 * just in case its running on a different cpu.
5764 * 2) If we are busy polling, do nothing here, we have
5765 * the guarantee we will be called later.
5767 if (unlikely(n
->state
& (NAPIF_STATE_NPSVC
|
5768 NAPIF_STATE_IN_BUSY_POLL
)))
5772 unsigned long timeout
= 0;
5775 timeout
= n
->dev
->gro_flush_timeout
;
5778 hrtimer_start(&n
->timer
, ns_to_ktime(timeout
),
5779 HRTIMER_MODE_REL_PINNED
);
5781 napi_gro_flush(n
, false);
5783 if (unlikely(!list_empty(&n
->poll_list
))) {
5784 /* If n->poll_list is not empty, we need to mask irqs */
5785 local_irq_save(flags
);
5786 list_del_init(&n
->poll_list
);
5787 local_irq_restore(flags
);
5791 val
= READ_ONCE(n
->state
);
5793 WARN_ON_ONCE(!(val
& NAPIF_STATE_SCHED
));
5795 new = val
& ~(NAPIF_STATE_MISSED
| NAPIF_STATE_SCHED
);
5797 /* If STATE_MISSED was set, leave STATE_SCHED set,
5798 * because we will call napi->poll() one more time.
5799 * This C code was suggested by Alexander Duyck to help gcc.
5801 new |= (val
& NAPIF_STATE_MISSED
) / NAPIF_STATE_MISSED
*
5803 } while (cmpxchg(&n
->state
, val
, new) != val
);
5805 if (unlikely(val
& NAPIF_STATE_MISSED
)) {
5812 EXPORT_SYMBOL(napi_complete_done
);
5814 /* must be called under rcu_read_lock(), as we dont take a reference */
5815 static struct napi_struct
*napi_by_id(unsigned int napi_id
)
5817 unsigned int hash
= napi_id
% HASH_SIZE(napi_hash
);
5818 struct napi_struct
*napi
;
5820 hlist_for_each_entry_rcu(napi
, &napi_hash
[hash
], napi_hash_node
)
5821 if (napi
->napi_id
== napi_id
)
5827 #if defined(CONFIG_NET_RX_BUSY_POLL)
5829 #define BUSY_POLL_BUDGET 8
5831 static void busy_poll_stop(struct napi_struct
*napi
, void *have_poll_lock
)
5835 /* Busy polling means there is a high chance device driver hard irq
5836 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
5837 * set in napi_schedule_prep().
5838 * Since we are about to call napi->poll() once more, we can safely
5839 * clear NAPI_STATE_MISSED.
5841 * Note: x86 could use a single "lock and ..." instruction
5842 * to perform these two clear_bit()
5844 clear_bit(NAPI_STATE_MISSED
, &napi
->state
);
5845 clear_bit(NAPI_STATE_IN_BUSY_POLL
, &napi
->state
);
5849 /* All we really want here is to re-enable device interrupts.
5850 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
5852 rc
= napi
->poll(napi
, BUSY_POLL_BUDGET
);
5853 trace_napi_poll(napi
, rc
, BUSY_POLL_BUDGET
);
5854 netpoll_poll_unlock(have_poll_lock
);
5855 if (rc
== BUSY_POLL_BUDGET
)
5856 __napi_schedule(napi
);
5860 void napi_busy_loop(unsigned int napi_id
,
5861 bool (*loop_end
)(void *, unsigned long),
5864 unsigned long start_time
= loop_end
? busy_loop_current_time() : 0;
5865 int (*napi_poll
)(struct napi_struct
*napi
, int budget
);
5866 void *have_poll_lock
= NULL
;
5867 struct napi_struct
*napi
;
5874 napi
= napi_by_id(napi_id
);
5884 unsigned long val
= READ_ONCE(napi
->state
);
5886 /* If multiple threads are competing for this napi,
5887 * we avoid dirtying napi->state as much as we can.
5889 if (val
& (NAPIF_STATE_DISABLE
| NAPIF_STATE_SCHED
|
5890 NAPIF_STATE_IN_BUSY_POLL
))
5892 if (cmpxchg(&napi
->state
, val
,
5893 val
| NAPIF_STATE_IN_BUSY_POLL
|
5894 NAPIF_STATE_SCHED
) != val
)
5896 have_poll_lock
= netpoll_poll_lock(napi
);
5897 napi_poll
= napi
->poll
;
5899 work
= napi_poll(napi
, BUSY_POLL_BUDGET
);
5900 trace_napi_poll(napi
, work
, BUSY_POLL_BUDGET
);
5903 __NET_ADD_STATS(dev_net(napi
->dev
),
5904 LINUX_MIB_BUSYPOLLRXPACKETS
, work
);
5907 if (!loop_end
|| loop_end(loop_end_arg
, start_time
))
5910 if (unlikely(need_resched())) {
5912 busy_poll_stop(napi
, have_poll_lock
);
5916 if (loop_end(loop_end_arg
, start_time
))
5923 busy_poll_stop(napi
, have_poll_lock
);
5928 EXPORT_SYMBOL(napi_busy_loop
);
5930 #endif /* CONFIG_NET_RX_BUSY_POLL */
5932 static void napi_hash_add(struct napi_struct
*napi
)
5934 if (test_bit(NAPI_STATE_NO_BUSY_POLL
, &napi
->state
) ||
5935 test_and_set_bit(NAPI_STATE_HASHED
, &napi
->state
))
5938 spin_lock(&napi_hash_lock
);
5940 /* 0..NR_CPUS range is reserved for sender_cpu use */
5942 if (unlikely(++napi_gen_id
< MIN_NAPI_ID
))
5943 napi_gen_id
= MIN_NAPI_ID
;
5944 } while (napi_by_id(napi_gen_id
));
5945 napi
->napi_id
= napi_gen_id
;
5947 hlist_add_head_rcu(&napi
->napi_hash_node
,
5948 &napi_hash
[napi
->napi_id
% HASH_SIZE(napi_hash
)]);
5950 spin_unlock(&napi_hash_lock
);
5953 /* Warning : caller is responsible to make sure rcu grace period
5954 * is respected before freeing memory containing @napi
5956 bool napi_hash_del(struct napi_struct
*napi
)
5958 bool rcu_sync_needed
= false;
5960 spin_lock(&napi_hash_lock
);
5962 if (test_and_clear_bit(NAPI_STATE_HASHED
, &napi
->state
)) {
5963 rcu_sync_needed
= true;
5964 hlist_del_rcu(&napi
->napi_hash_node
);
5966 spin_unlock(&napi_hash_lock
);
5967 return rcu_sync_needed
;
5969 EXPORT_SYMBOL_GPL(napi_hash_del
);
5971 static enum hrtimer_restart
napi_watchdog(struct hrtimer
*timer
)
5973 struct napi_struct
*napi
;
5975 napi
= container_of(timer
, struct napi_struct
, timer
);
5977 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
5978 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
5980 if (napi
->gro_count
&& !napi_disable_pending(napi
) &&
5981 !test_and_set_bit(NAPI_STATE_SCHED
, &napi
->state
))
5982 __napi_schedule_irqoff(napi
);
5984 return HRTIMER_NORESTART
;
5987 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
5988 int (*poll
)(struct napi_struct
*, int), int weight
)
5992 INIT_LIST_HEAD(&napi
->poll_list
);
5993 hrtimer_init(&napi
->timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_REL_PINNED
);
5994 napi
->timer
.function
= napi_watchdog
;
5995 napi
->gro_count
= 0;
5996 for (i
= 0; i
< GRO_HASH_BUCKETS
; i
++) {
5997 INIT_LIST_HEAD(&napi
->gro_hash
[i
].list
);
5998 napi
->gro_hash
[i
].count
= 0;
6002 if (weight
> NAPI_POLL_WEIGHT
)
6003 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
6005 napi
->weight
= weight
;
6006 list_add(&napi
->dev_list
, &dev
->napi_list
);
6008 #ifdef CONFIG_NETPOLL
6009 napi
->poll_owner
= -1;
6011 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
6012 napi_hash_add(napi
);
6014 EXPORT_SYMBOL(netif_napi_add
);
6016 void napi_disable(struct napi_struct
*n
)
6019 set_bit(NAPI_STATE_DISABLE
, &n
->state
);
6021 while (test_and_set_bit(NAPI_STATE_SCHED
, &n
->state
))
6023 while (test_and_set_bit(NAPI_STATE_NPSVC
, &n
->state
))
6026 hrtimer_cancel(&n
->timer
);
6028 clear_bit(NAPI_STATE_DISABLE
, &n
->state
);
6030 EXPORT_SYMBOL(napi_disable
);
6032 static void flush_gro_hash(struct napi_struct
*napi
)
6036 for (i
= 0; i
< GRO_HASH_BUCKETS
; i
++) {
6037 struct sk_buff
*skb
, *n
;
6039 list_for_each_entry_safe(skb
, n
, &napi
->gro_hash
[i
].list
, list
)
6041 napi
->gro_hash
[i
].count
= 0;
6045 /* Must be called in process context */
6046 void netif_napi_del(struct napi_struct
*napi
)
6049 if (napi_hash_del(napi
))
6051 list_del_init(&napi
->dev_list
);
6052 napi_free_frags(napi
);
6054 flush_gro_hash(napi
);
6055 napi
->gro_count
= 0;
6057 EXPORT_SYMBOL(netif_napi_del
);
6059 static int napi_poll(struct napi_struct
*n
, struct list_head
*repoll
)
6064 list_del_init(&n
->poll_list
);
6066 have
= netpoll_poll_lock(n
);
6070 /* This NAPI_STATE_SCHED test is for avoiding a race
6071 * with netpoll's poll_napi(). Only the entity which
6072 * obtains the lock and sees NAPI_STATE_SCHED set will
6073 * actually make the ->poll() call. Therefore we avoid
6074 * accidentally calling ->poll() when NAPI is not scheduled.
6077 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
6078 work
= n
->poll(n
, weight
);
6079 trace_napi_poll(n
, work
, weight
);
6082 WARN_ON_ONCE(work
> weight
);
6084 if (likely(work
< weight
))
6087 /* Drivers must not modify the NAPI state if they
6088 * consume the entire weight. In such cases this code
6089 * still "owns" the NAPI instance and therefore can
6090 * move the instance around on the list at-will.
6092 if (unlikely(napi_disable_pending(n
))) {
6098 /* flush too old packets
6099 * If HZ < 1000, flush all packets.
6101 napi_gro_flush(n
, HZ
>= 1000);
6104 /* Some drivers may have called napi_schedule
6105 * prior to exhausting their budget.
6107 if (unlikely(!list_empty(&n
->poll_list
))) {
6108 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
6109 n
->dev
? n
->dev
->name
: "backlog");
6113 list_add_tail(&n
->poll_list
, repoll
);
6116 netpoll_poll_unlock(have
);
6121 static __latent_entropy
void net_rx_action(struct softirq_action
*h
)
6123 struct softnet_data
*sd
= this_cpu_ptr(&softnet_data
);
6124 unsigned long time_limit
= jiffies
+
6125 usecs_to_jiffies(netdev_budget_usecs
);
6126 int budget
= netdev_budget
;
6130 local_irq_disable();
6131 list_splice_init(&sd
->poll_list
, &list
);
6135 struct napi_struct
*n
;
6137 if (list_empty(&list
)) {
6138 if (!sd_has_rps_ipi_waiting(sd
) && list_empty(&repoll
))
6143 n
= list_first_entry(&list
, struct napi_struct
, poll_list
);
6144 budget
-= napi_poll(n
, &repoll
);
6146 /* If softirq window is exhausted then punt.
6147 * Allow this to run for 2 jiffies since which will allow
6148 * an average latency of 1.5/HZ.
6150 if (unlikely(budget
<= 0 ||
6151 time_after_eq(jiffies
, time_limit
))) {
6157 local_irq_disable();
6159 list_splice_tail_init(&sd
->poll_list
, &list
);
6160 list_splice_tail(&repoll
, &list
);
6161 list_splice(&list
, &sd
->poll_list
);
6162 if (!list_empty(&sd
->poll_list
))
6163 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
6165 net_rps_action_and_irq_enable(sd
);
6167 __kfree_skb_flush();
6170 struct netdev_adjacent
{
6171 struct net_device
*dev
;
6173 /* upper master flag, there can only be one master device per list */
6176 /* counter for the number of times this device was added to us */
6179 /* private field for the users */
6182 struct list_head list
;
6183 struct rcu_head rcu
;
6186 static struct netdev_adjacent
*__netdev_find_adj(struct net_device
*adj_dev
,
6187 struct list_head
*adj_list
)
6189 struct netdev_adjacent
*adj
;
6191 list_for_each_entry(adj
, adj_list
, list
) {
6192 if (adj
->dev
== adj_dev
)
6198 static int __netdev_has_upper_dev(struct net_device
*upper_dev
, void *data
)
6200 struct net_device
*dev
= data
;
6202 return upper_dev
== dev
;
6206 * netdev_has_upper_dev - Check if device is linked to an upper device
6208 * @upper_dev: upper device to check
6210 * Find out if a device is linked to specified upper device and return true
6211 * in case it is. Note that this checks only immediate upper device,
6212 * not through a complete stack of devices. The caller must hold the RTNL lock.
6214 bool netdev_has_upper_dev(struct net_device
*dev
,
6215 struct net_device
*upper_dev
)
6219 return netdev_walk_all_upper_dev_rcu(dev
, __netdev_has_upper_dev
,
6222 EXPORT_SYMBOL(netdev_has_upper_dev
);
6225 * netdev_has_upper_dev_all - Check if device is linked to an upper device
6227 * @upper_dev: upper device to check
6229 * Find out if a device is linked to specified upper device and return true
6230 * in case it is. Note that this checks the entire upper device chain.
6231 * The caller must hold rcu lock.
6234 bool netdev_has_upper_dev_all_rcu(struct net_device
*dev
,
6235 struct net_device
*upper_dev
)
6237 return !!netdev_walk_all_upper_dev_rcu(dev
, __netdev_has_upper_dev
,
6240 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu
);
6243 * netdev_has_any_upper_dev - Check if device is linked to some device
6246 * Find out if a device is linked to an upper device and return true in case
6247 * it is. The caller must hold the RTNL lock.
6249 bool netdev_has_any_upper_dev(struct net_device
*dev
)
6253 return !list_empty(&dev
->adj_list
.upper
);
6255 EXPORT_SYMBOL(netdev_has_any_upper_dev
);
6258 * netdev_master_upper_dev_get - Get master upper device
6261 * Find a master upper device and return pointer to it or NULL in case
6262 * it's not there. The caller must hold the RTNL lock.
6264 struct net_device
*netdev_master_upper_dev_get(struct net_device
*dev
)
6266 struct netdev_adjacent
*upper
;
6270 if (list_empty(&dev
->adj_list
.upper
))
6273 upper
= list_first_entry(&dev
->adj_list
.upper
,
6274 struct netdev_adjacent
, list
);
6275 if (likely(upper
->master
))
6279 EXPORT_SYMBOL(netdev_master_upper_dev_get
);
6282 * netdev_has_any_lower_dev - Check if device is linked to some device
6285 * Find out if a device is linked to a lower device and return true in case
6286 * it is. The caller must hold the RTNL lock.
6288 static bool netdev_has_any_lower_dev(struct net_device
*dev
)
6292 return !list_empty(&dev
->adj_list
.lower
);
6295 void *netdev_adjacent_get_private(struct list_head
*adj_list
)
6297 struct netdev_adjacent
*adj
;
6299 adj
= list_entry(adj_list
, struct netdev_adjacent
, list
);
6301 return adj
->private;
6303 EXPORT_SYMBOL(netdev_adjacent_get_private
);
6306 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
6308 * @iter: list_head ** of the current position
6310 * Gets the next device from the dev's upper list, starting from iter
6311 * position. The caller must hold RCU read lock.
6313 struct net_device
*netdev_upper_get_next_dev_rcu(struct net_device
*dev
,
6314 struct list_head
**iter
)
6316 struct netdev_adjacent
*upper
;
6318 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6320 upper
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
6322 if (&upper
->list
== &dev
->adj_list
.upper
)
6325 *iter
= &upper
->list
;
6329 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu
);
6331 static struct net_device
*netdev_next_upper_dev_rcu(struct net_device
*dev
,
6332 struct list_head
**iter
)
6334 struct netdev_adjacent
*upper
;
6336 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6338 upper
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
6340 if (&upper
->list
== &dev
->adj_list
.upper
)
6343 *iter
= &upper
->list
;
6348 int netdev_walk_all_upper_dev_rcu(struct net_device
*dev
,
6349 int (*fn
)(struct net_device
*dev
,
6353 struct net_device
*udev
;
6354 struct list_head
*iter
;
6357 for (iter
= &dev
->adj_list
.upper
,
6358 udev
= netdev_next_upper_dev_rcu(dev
, &iter
);
6360 udev
= netdev_next_upper_dev_rcu(dev
, &iter
)) {
6361 /* first is the upper device itself */
6362 ret
= fn(udev
, data
);
6366 /* then look at all of its upper devices */
6367 ret
= netdev_walk_all_upper_dev_rcu(udev
, fn
, data
);
6374 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu
);
6377 * netdev_lower_get_next_private - Get the next ->private from the
6378 * lower neighbour list
6380 * @iter: list_head ** of the current position
6382 * Gets the next netdev_adjacent->private from the dev's lower neighbour
6383 * list, starting from iter position. The caller must hold either hold the
6384 * RTNL lock or its own locking that guarantees that the neighbour lower
6385 * list will remain unchanged.
6387 void *netdev_lower_get_next_private(struct net_device
*dev
,
6388 struct list_head
**iter
)
6390 struct netdev_adjacent
*lower
;
6392 lower
= list_entry(*iter
, struct netdev_adjacent
, list
);
6394 if (&lower
->list
== &dev
->adj_list
.lower
)
6397 *iter
= lower
->list
.next
;
6399 return lower
->private;
6401 EXPORT_SYMBOL(netdev_lower_get_next_private
);
6404 * netdev_lower_get_next_private_rcu - Get the next ->private from the
6405 * lower neighbour list, RCU
6408 * @iter: list_head ** of the current position
6410 * Gets the next netdev_adjacent->private from the dev's lower neighbour
6411 * list, starting from iter position. The caller must hold RCU read lock.
6413 void *netdev_lower_get_next_private_rcu(struct net_device
*dev
,
6414 struct list_head
**iter
)
6416 struct netdev_adjacent
*lower
;
6418 WARN_ON_ONCE(!rcu_read_lock_held());
6420 lower
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
6422 if (&lower
->list
== &dev
->adj_list
.lower
)
6425 *iter
= &lower
->list
;
6427 return lower
->private;
6429 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu
);
6432 * netdev_lower_get_next - Get the next device from the lower neighbour
6435 * @iter: list_head ** of the current position
6437 * Gets the next netdev_adjacent from the dev's lower neighbour
6438 * list, starting from iter position. The caller must hold RTNL lock or
6439 * its own locking that guarantees that the neighbour lower
6440 * list will remain unchanged.
6442 void *netdev_lower_get_next(struct net_device
*dev
, struct list_head
**iter
)
6444 struct netdev_adjacent
*lower
;
6446 lower
= list_entry(*iter
, struct netdev_adjacent
, list
);
6448 if (&lower
->list
== &dev
->adj_list
.lower
)
6451 *iter
= lower
->list
.next
;
6455 EXPORT_SYMBOL(netdev_lower_get_next
);
6457 static struct net_device
*netdev_next_lower_dev(struct net_device
*dev
,
6458 struct list_head
**iter
)
6460 struct netdev_adjacent
*lower
;
6462 lower
= list_entry((*iter
)->next
, struct netdev_adjacent
, list
);
6464 if (&lower
->list
== &dev
->adj_list
.lower
)
6467 *iter
= &lower
->list
;
6472 int netdev_walk_all_lower_dev(struct net_device
*dev
,
6473 int (*fn
)(struct net_device
*dev
,
6477 struct net_device
*ldev
;
6478 struct list_head
*iter
;
6481 for (iter
= &dev
->adj_list
.lower
,
6482 ldev
= netdev_next_lower_dev(dev
, &iter
);
6484 ldev
= netdev_next_lower_dev(dev
, &iter
)) {
6485 /* first is the lower device itself */
6486 ret
= fn(ldev
, data
);
6490 /* then look at all of its lower devices */
6491 ret
= netdev_walk_all_lower_dev(ldev
, fn
, data
);
6498 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev
);
6500 static struct net_device
*netdev_next_lower_dev_rcu(struct net_device
*dev
,
6501 struct list_head
**iter
)
6503 struct netdev_adjacent
*lower
;
6505 lower
= list_entry_rcu((*iter
)->next
, struct netdev_adjacent
, list
);
6506 if (&lower
->list
== &dev
->adj_list
.lower
)
6509 *iter
= &lower
->list
;
6514 int netdev_walk_all_lower_dev_rcu(struct net_device
*dev
,
6515 int (*fn
)(struct net_device
*dev
,
6519 struct net_device
*ldev
;
6520 struct list_head
*iter
;
6523 for (iter
= &dev
->adj_list
.lower
,
6524 ldev
= netdev_next_lower_dev_rcu(dev
, &iter
);
6526 ldev
= netdev_next_lower_dev_rcu(dev
, &iter
)) {
6527 /* first is the lower device itself */
6528 ret
= fn(ldev
, data
);
6532 /* then look at all of its lower devices */
6533 ret
= netdev_walk_all_lower_dev_rcu(ldev
, fn
, data
);
6540 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu
);
6543 * netdev_lower_get_first_private_rcu - Get the first ->private from the
6544 * lower neighbour list, RCU
6548 * Gets the first netdev_adjacent->private from the dev's lower neighbour
6549 * list. The caller must hold RCU read lock.
6551 void *netdev_lower_get_first_private_rcu(struct net_device
*dev
)
6553 struct netdev_adjacent
*lower
;
6555 lower
= list_first_or_null_rcu(&dev
->adj_list
.lower
,
6556 struct netdev_adjacent
, list
);
6558 return lower
->private;
6561 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu
);
6564 * netdev_master_upper_dev_get_rcu - Get master upper device
6567 * Find a master upper device and return pointer to it or NULL in case
6568 * it's not there. The caller must hold the RCU read lock.
6570 struct net_device
*netdev_master_upper_dev_get_rcu(struct net_device
*dev
)
6572 struct netdev_adjacent
*upper
;
6574 upper
= list_first_or_null_rcu(&dev
->adj_list
.upper
,
6575 struct netdev_adjacent
, list
);
6576 if (upper
&& likely(upper
->master
))
6580 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu
);
6582 static int netdev_adjacent_sysfs_add(struct net_device
*dev
,
6583 struct net_device
*adj_dev
,
6584 struct list_head
*dev_list
)
6586 char linkname
[IFNAMSIZ
+7];
6588 sprintf(linkname
, dev_list
== &dev
->adj_list
.upper
?
6589 "upper_%s" : "lower_%s", adj_dev
->name
);
6590 return sysfs_create_link(&(dev
->dev
.kobj
), &(adj_dev
->dev
.kobj
),
6593 static void netdev_adjacent_sysfs_del(struct net_device
*dev
,
6595 struct list_head
*dev_list
)
6597 char linkname
[IFNAMSIZ
+7];
6599 sprintf(linkname
, dev_list
== &dev
->adj_list
.upper
?
6600 "upper_%s" : "lower_%s", name
);
6601 sysfs_remove_link(&(dev
->dev
.kobj
), linkname
);
6604 static inline bool netdev_adjacent_is_neigh_list(struct net_device
*dev
,
6605 struct net_device
*adj_dev
,
6606 struct list_head
*dev_list
)
6608 return (dev_list
== &dev
->adj_list
.upper
||
6609 dev_list
== &dev
->adj_list
.lower
) &&
6610 net_eq(dev_net(dev
), dev_net(adj_dev
));
6613 static int __netdev_adjacent_dev_insert(struct net_device
*dev
,
6614 struct net_device
*adj_dev
,
6615 struct list_head
*dev_list
,
6616 void *private, bool master
)
6618 struct netdev_adjacent
*adj
;
6621 adj
= __netdev_find_adj(adj_dev
, dev_list
);
6625 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
6626 dev
->name
, adj_dev
->name
, adj
->ref_nr
);
6631 adj
= kmalloc(sizeof(*adj
), GFP_KERNEL
);
6636 adj
->master
= master
;
6638 adj
->private = private;
6641 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
6642 dev
->name
, adj_dev
->name
, adj
->ref_nr
, adj_dev
->name
);
6644 if (netdev_adjacent_is_neigh_list(dev
, adj_dev
, dev_list
)) {
6645 ret
= netdev_adjacent_sysfs_add(dev
, adj_dev
, dev_list
);
6650 /* Ensure that master link is always the first item in list. */
6652 ret
= sysfs_create_link(&(dev
->dev
.kobj
),
6653 &(adj_dev
->dev
.kobj
), "master");
6655 goto remove_symlinks
;
6657 list_add_rcu(&adj
->list
, dev_list
);
6659 list_add_tail_rcu(&adj
->list
, dev_list
);
6665 if (netdev_adjacent_is_neigh_list(dev
, adj_dev
, dev_list
))
6666 netdev_adjacent_sysfs_del(dev
, adj_dev
->name
, dev_list
);
6674 static void __netdev_adjacent_dev_remove(struct net_device
*dev
,
6675 struct net_device
*adj_dev
,
6677 struct list_head
*dev_list
)
6679 struct netdev_adjacent
*adj
;
6681 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
6682 dev
->name
, adj_dev
->name
, ref_nr
);
6684 adj
= __netdev_find_adj(adj_dev
, dev_list
);
6687 pr_err("Adjacency does not exist for device %s from %s\n",
6688 dev
->name
, adj_dev
->name
);
6693 if (adj
->ref_nr
> ref_nr
) {
6694 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
6695 dev
->name
, adj_dev
->name
, ref_nr
,
6696 adj
->ref_nr
- ref_nr
);
6697 adj
->ref_nr
-= ref_nr
;
6702 sysfs_remove_link(&(dev
->dev
.kobj
), "master");
6704 if (netdev_adjacent_is_neigh_list(dev
, adj_dev
, dev_list
))
6705 netdev_adjacent_sysfs_del(dev
, adj_dev
->name
, dev_list
);
6707 list_del_rcu(&adj
->list
);
6708 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
6709 adj_dev
->name
, dev
->name
, adj_dev
->name
);
6711 kfree_rcu(adj
, rcu
);
6714 static int __netdev_adjacent_dev_link_lists(struct net_device
*dev
,
6715 struct net_device
*upper_dev
,
6716 struct list_head
*up_list
,
6717 struct list_head
*down_list
,
6718 void *private, bool master
)
6722 ret
= __netdev_adjacent_dev_insert(dev
, upper_dev
, up_list
,
6727 ret
= __netdev_adjacent_dev_insert(upper_dev
, dev
, down_list
,
6730 __netdev_adjacent_dev_remove(dev
, upper_dev
, 1, up_list
);
6737 static void __netdev_adjacent_dev_unlink_lists(struct net_device
*dev
,
6738 struct net_device
*upper_dev
,
6740 struct list_head
*up_list
,
6741 struct list_head
*down_list
)
6743 __netdev_adjacent_dev_remove(dev
, upper_dev
, ref_nr
, up_list
);
6744 __netdev_adjacent_dev_remove(upper_dev
, dev
, ref_nr
, down_list
);
6747 static int __netdev_adjacent_dev_link_neighbour(struct net_device
*dev
,
6748 struct net_device
*upper_dev
,
6749 void *private, bool master
)
6751 return __netdev_adjacent_dev_link_lists(dev
, upper_dev
,
6752 &dev
->adj_list
.upper
,
6753 &upper_dev
->adj_list
.lower
,
6757 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device
*dev
,
6758 struct net_device
*upper_dev
)
6760 __netdev_adjacent_dev_unlink_lists(dev
, upper_dev
, 1,
6761 &dev
->adj_list
.upper
,
6762 &upper_dev
->adj_list
.lower
);
6765 static int __netdev_upper_dev_link(struct net_device
*dev
,
6766 struct net_device
*upper_dev
, bool master
,
6767 void *upper_priv
, void *upper_info
,
6768 struct netlink_ext_ack
*extack
)
6770 struct netdev_notifier_changeupper_info changeupper_info
= {
6775 .upper_dev
= upper_dev
,
6778 .upper_info
= upper_info
,
6780 struct net_device
*master_dev
;
6785 if (dev
== upper_dev
)
6788 /* To prevent loops, check if dev is not upper device to upper_dev. */
6789 if (netdev_has_upper_dev(upper_dev
, dev
))
6793 if (netdev_has_upper_dev(dev
, upper_dev
))
6796 master_dev
= netdev_master_upper_dev_get(dev
);
6798 return master_dev
== upper_dev
? -EEXIST
: -EBUSY
;
6801 ret
= call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER
,
6802 &changeupper_info
.info
);
6803 ret
= notifier_to_errno(ret
);
6807 ret
= __netdev_adjacent_dev_link_neighbour(dev
, upper_dev
, upper_priv
,
6812 ret
= call_netdevice_notifiers_info(NETDEV_CHANGEUPPER
,
6813 &changeupper_info
.info
);
6814 ret
= notifier_to_errno(ret
);
6821 __netdev_adjacent_dev_unlink_neighbour(dev
, upper_dev
);
6827 * netdev_upper_dev_link - Add a link to the upper device
6829 * @upper_dev: new upper device
6830 * @extack: netlink extended ack
6832 * Adds a link to device which is upper to this one. The caller must hold
6833 * the RTNL lock. On a failure a negative errno code is returned.
6834 * On success the reference counts are adjusted and the function
6837 int netdev_upper_dev_link(struct net_device
*dev
,
6838 struct net_device
*upper_dev
,
6839 struct netlink_ext_ack
*extack
)
6841 return __netdev_upper_dev_link(dev
, upper_dev
, false,
6842 NULL
, NULL
, extack
);
6844 EXPORT_SYMBOL(netdev_upper_dev_link
);
6847 * netdev_master_upper_dev_link - Add a master link to the upper device
6849 * @upper_dev: new upper device
6850 * @upper_priv: upper device private
6851 * @upper_info: upper info to be passed down via notifier
6852 * @extack: netlink extended ack
6854 * Adds a link to device which is upper to this one. In this case, only
6855 * one master upper device can be linked, although other non-master devices
6856 * might be linked as well. The caller must hold the RTNL lock.
6857 * On a failure a negative errno code is returned. On success the reference
6858 * counts are adjusted and the function returns zero.
6860 int netdev_master_upper_dev_link(struct net_device
*dev
,
6861 struct net_device
*upper_dev
,
6862 void *upper_priv
, void *upper_info
,
6863 struct netlink_ext_ack
*extack
)
6865 return __netdev_upper_dev_link(dev
, upper_dev
, true,
6866 upper_priv
, upper_info
, extack
);
6868 EXPORT_SYMBOL(netdev_master_upper_dev_link
);
6871 * netdev_upper_dev_unlink - Removes a link to upper device
6873 * @upper_dev: new upper device
6875 * Removes a link to device which is upper to this one. The caller must hold
6878 void netdev_upper_dev_unlink(struct net_device
*dev
,
6879 struct net_device
*upper_dev
)
6881 struct netdev_notifier_changeupper_info changeupper_info
= {
6885 .upper_dev
= upper_dev
,
6891 changeupper_info
.master
= netdev_master_upper_dev_get(dev
) == upper_dev
;
6893 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER
,
6894 &changeupper_info
.info
);
6896 __netdev_adjacent_dev_unlink_neighbour(dev
, upper_dev
);
6898 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER
,
6899 &changeupper_info
.info
);
6901 EXPORT_SYMBOL(netdev_upper_dev_unlink
);
6904 * netdev_bonding_info_change - Dispatch event about slave change
6906 * @bonding_info: info to dispatch
6908 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
6909 * The caller must hold the RTNL lock.
6911 void netdev_bonding_info_change(struct net_device
*dev
,
6912 struct netdev_bonding_info
*bonding_info
)
6914 struct netdev_notifier_bonding_info info
= {
6918 memcpy(&info
.bonding_info
, bonding_info
,
6919 sizeof(struct netdev_bonding_info
));
6920 call_netdevice_notifiers_info(NETDEV_BONDING_INFO
,
6923 EXPORT_SYMBOL(netdev_bonding_info_change
);
6925 static void netdev_adjacent_add_links(struct net_device
*dev
)
6927 struct netdev_adjacent
*iter
;
6929 struct net
*net
= dev_net(dev
);
6931 list_for_each_entry(iter
, &dev
->adj_list
.upper
, list
) {
6932 if (!net_eq(net
, dev_net(iter
->dev
)))
6934 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
6935 &iter
->dev
->adj_list
.lower
);
6936 netdev_adjacent_sysfs_add(dev
, iter
->dev
,
6937 &dev
->adj_list
.upper
);
6940 list_for_each_entry(iter
, &dev
->adj_list
.lower
, list
) {
6941 if (!net_eq(net
, dev_net(iter
->dev
)))
6943 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
6944 &iter
->dev
->adj_list
.upper
);
6945 netdev_adjacent_sysfs_add(dev
, iter
->dev
,
6946 &dev
->adj_list
.lower
);
6950 static void netdev_adjacent_del_links(struct net_device
*dev
)
6952 struct netdev_adjacent
*iter
;
6954 struct net
*net
= dev_net(dev
);
6956 list_for_each_entry(iter
, &dev
->adj_list
.upper
, list
) {
6957 if (!net_eq(net
, dev_net(iter
->dev
)))
6959 netdev_adjacent_sysfs_del(iter
->dev
, dev
->name
,
6960 &iter
->dev
->adj_list
.lower
);
6961 netdev_adjacent_sysfs_del(dev
, iter
->dev
->name
,
6962 &dev
->adj_list
.upper
);
6965 list_for_each_entry(iter
, &dev
->adj_list
.lower
, list
) {
6966 if (!net_eq(net
, dev_net(iter
->dev
)))
6968 netdev_adjacent_sysfs_del(iter
->dev
, dev
->name
,
6969 &iter
->dev
->adj_list
.upper
);
6970 netdev_adjacent_sysfs_del(dev
, iter
->dev
->name
,
6971 &dev
->adj_list
.lower
);
6975 void netdev_adjacent_rename_links(struct net_device
*dev
, char *oldname
)
6977 struct netdev_adjacent
*iter
;
6979 struct net
*net
= dev_net(dev
);
6981 list_for_each_entry(iter
, &dev
->adj_list
.upper
, list
) {
6982 if (!net_eq(net
, dev_net(iter
->dev
)))
6984 netdev_adjacent_sysfs_del(iter
->dev
, oldname
,
6985 &iter
->dev
->adj_list
.lower
);
6986 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
6987 &iter
->dev
->adj_list
.lower
);
6990 list_for_each_entry(iter
, &dev
->adj_list
.lower
, list
) {
6991 if (!net_eq(net
, dev_net(iter
->dev
)))
6993 netdev_adjacent_sysfs_del(iter
->dev
, oldname
,
6994 &iter
->dev
->adj_list
.upper
);
6995 netdev_adjacent_sysfs_add(iter
->dev
, dev
,
6996 &iter
->dev
->adj_list
.upper
);
7000 void *netdev_lower_dev_get_private(struct net_device
*dev
,
7001 struct net_device
*lower_dev
)
7003 struct netdev_adjacent
*lower
;
7007 lower
= __netdev_find_adj(lower_dev
, &dev
->adj_list
.lower
);
7011 return lower
->private;
7013 EXPORT_SYMBOL(netdev_lower_dev_get_private
);
7016 int dev_get_nest_level(struct net_device
*dev
)
7018 struct net_device
*lower
= NULL
;
7019 struct list_head
*iter
;
7025 netdev_for_each_lower_dev(dev
, lower
, iter
) {
7026 nest
= dev_get_nest_level(lower
);
7027 if (max_nest
< nest
)
7031 return max_nest
+ 1;
7033 EXPORT_SYMBOL(dev_get_nest_level
);
7036 * netdev_lower_change - Dispatch event about lower device state change
7037 * @lower_dev: device
7038 * @lower_state_info: state to dispatch
7040 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
7041 * The caller must hold the RTNL lock.
7043 void netdev_lower_state_changed(struct net_device
*lower_dev
,
7044 void *lower_state_info
)
7046 struct netdev_notifier_changelowerstate_info changelowerstate_info
= {
7047 .info
.dev
= lower_dev
,
7051 changelowerstate_info
.lower_state_info
= lower_state_info
;
7052 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE
,
7053 &changelowerstate_info
.info
);
7055 EXPORT_SYMBOL(netdev_lower_state_changed
);
7057 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
7059 const struct net_device_ops
*ops
= dev
->netdev_ops
;
7061 if (ops
->ndo_change_rx_flags
)
7062 ops
->ndo_change_rx_flags(dev
, flags
);
7065 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
, bool notify
)
7067 unsigned int old_flags
= dev
->flags
;
7073 dev
->flags
|= IFF_PROMISC
;
7074 dev
->promiscuity
+= inc
;
7075 if (dev
->promiscuity
== 0) {
7078 * If inc causes overflow, untouch promisc and return error.
7081 dev
->flags
&= ~IFF_PROMISC
;
7083 dev
->promiscuity
-= inc
;
7084 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
7089 if (dev
->flags
!= old_flags
) {
7090 pr_info("device %s %s promiscuous mode\n",
7092 dev
->flags
& IFF_PROMISC
? "entered" : "left");
7093 if (audit_enabled
) {
7094 current_uid_gid(&uid
, &gid
);
7095 audit_log(audit_context(), GFP_ATOMIC
,
7096 AUDIT_ANOM_PROMISCUOUS
,
7097 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
7098 dev
->name
, (dev
->flags
& IFF_PROMISC
),
7099 (old_flags
& IFF_PROMISC
),
7100 from_kuid(&init_user_ns
, audit_get_loginuid(current
)),
7101 from_kuid(&init_user_ns
, uid
),
7102 from_kgid(&init_user_ns
, gid
),
7103 audit_get_sessionid(current
));
7106 dev_change_rx_flags(dev
, IFF_PROMISC
);
7109 __dev_notify_flags(dev
, old_flags
, IFF_PROMISC
);
7114 * dev_set_promiscuity - update promiscuity count on a device
7118 * Add or remove promiscuity from a device. While the count in the device
7119 * remains above zero the interface remains promiscuous. Once it hits zero
7120 * the device reverts back to normal filtering operation. A negative inc
7121 * value is used to drop promiscuity on the device.
7122 * Return 0 if successful or a negative errno code on error.
7124 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
7126 unsigned int old_flags
= dev
->flags
;
7129 err
= __dev_set_promiscuity(dev
, inc
, true);
7132 if (dev
->flags
!= old_flags
)
7133 dev_set_rx_mode(dev
);
7136 EXPORT_SYMBOL(dev_set_promiscuity
);
7138 static int __dev_set_allmulti(struct net_device
*dev
, int inc
, bool notify
)
7140 unsigned int old_flags
= dev
->flags
, old_gflags
= dev
->gflags
;
7144 dev
->flags
|= IFF_ALLMULTI
;
7145 dev
->allmulti
+= inc
;
7146 if (dev
->allmulti
== 0) {
7149 * If inc causes overflow, untouch allmulti and return error.
7152 dev
->flags
&= ~IFF_ALLMULTI
;
7154 dev
->allmulti
-= inc
;
7155 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
7160 if (dev
->flags
^ old_flags
) {
7161 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
7162 dev_set_rx_mode(dev
);
7164 __dev_notify_flags(dev
, old_flags
,
7165 dev
->gflags
^ old_gflags
);
7171 * dev_set_allmulti - update allmulti count on a device
7175 * Add or remove reception of all multicast frames to a device. While the
7176 * count in the device remains above zero the interface remains listening
7177 * to all interfaces. Once it hits zero the device reverts back to normal
7178 * filtering operation. A negative @inc value is used to drop the counter
7179 * when releasing a resource needing all multicasts.
7180 * Return 0 if successful or a negative errno code on error.
7183 int dev_set_allmulti(struct net_device
*dev
, int inc
)
7185 return __dev_set_allmulti(dev
, inc
, true);
7187 EXPORT_SYMBOL(dev_set_allmulti
);
7190 * Upload unicast and multicast address lists to device and
7191 * configure RX filtering. When the device doesn't support unicast
7192 * filtering it is put in promiscuous mode while unicast addresses
7195 void __dev_set_rx_mode(struct net_device
*dev
)
7197 const struct net_device_ops
*ops
= dev
->netdev_ops
;
7199 /* dev_open will call this function so the list will stay sane. */
7200 if (!(dev
->flags
&IFF_UP
))
7203 if (!netif_device_present(dev
))
7206 if (!(dev
->priv_flags
& IFF_UNICAST_FLT
)) {
7207 /* Unicast addresses changes may only happen under the rtnl,
7208 * therefore calling __dev_set_promiscuity here is safe.
7210 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
7211 __dev_set_promiscuity(dev
, 1, false);
7212 dev
->uc_promisc
= true;
7213 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
7214 __dev_set_promiscuity(dev
, -1, false);
7215 dev
->uc_promisc
= false;
7219 if (ops
->ndo_set_rx_mode
)
7220 ops
->ndo_set_rx_mode(dev
);
7223 void dev_set_rx_mode(struct net_device
*dev
)
7225 netif_addr_lock_bh(dev
);
7226 __dev_set_rx_mode(dev
);
7227 netif_addr_unlock_bh(dev
);
7231 * dev_get_flags - get flags reported to userspace
7234 * Get the combination of flag bits exported through APIs to userspace.
7236 unsigned int dev_get_flags(const struct net_device
*dev
)
7240 flags
= (dev
->flags
& ~(IFF_PROMISC
|
7245 (dev
->gflags
& (IFF_PROMISC
|
7248 if (netif_running(dev
)) {
7249 if (netif_oper_up(dev
))
7250 flags
|= IFF_RUNNING
;
7251 if (netif_carrier_ok(dev
))
7252 flags
|= IFF_LOWER_UP
;
7253 if (netif_dormant(dev
))
7254 flags
|= IFF_DORMANT
;
7259 EXPORT_SYMBOL(dev_get_flags
);
7261 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
7263 unsigned int old_flags
= dev
->flags
;
7269 * Set the flags on our device.
7272 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
7273 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
7275 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
7279 * Load in the correct multicast list now the flags have changed.
7282 if ((old_flags
^ flags
) & IFF_MULTICAST
)
7283 dev_change_rx_flags(dev
, IFF_MULTICAST
);
7285 dev_set_rx_mode(dev
);
7288 * Have we downed the interface. We handle IFF_UP ourselves
7289 * according to user attempts to set it, rather than blindly
7294 if ((old_flags
^ flags
) & IFF_UP
) {
7295 if (old_flags
& IFF_UP
)
7298 ret
= __dev_open(dev
);
7301 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
7302 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
7303 unsigned int old_flags
= dev
->flags
;
7305 dev
->gflags
^= IFF_PROMISC
;
7307 if (__dev_set_promiscuity(dev
, inc
, false) >= 0)
7308 if (dev
->flags
!= old_flags
)
7309 dev_set_rx_mode(dev
);
7312 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
7313 * is important. Some (broken) drivers set IFF_PROMISC, when
7314 * IFF_ALLMULTI is requested not asking us and not reporting.
7316 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
7317 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
7319 dev
->gflags
^= IFF_ALLMULTI
;
7320 __dev_set_allmulti(dev
, inc
, false);
7326 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
,
7327 unsigned int gchanges
)
7329 unsigned int changes
= dev
->flags
^ old_flags
;
7332 rtmsg_ifinfo(RTM_NEWLINK
, dev
, gchanges
, GFP_ATOMIC
);
7334 if (changes
& IFF_UP
) {
7335 if (dev
->flags
& IFF_UP
)
7336 call_netdevice_notifiers(NETDEV_UP
, dev
);
7338 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
7341 if (dev
->flags
& IFF_UP
&&
7342 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
))) {
7343 struct netdev_notifier_change_info change_info
= {
7347 .flags_changed
= changes
,
7350 call_netdevice_notifiers_info(NETDEV_CHANGE
, &change_info
.info
);
7355 * dev_change_flags - change device settings
7357 * @flags: device state flags
7359 * Change settings on device based state flags. The flags are
7360 * in the userspace exported format.
7362 int dev_change_flags(struct net_device
*dev
, unsigned int flags
)
7365 unsigned int changes
, old_flags
= dev
->flags
, old_gflags
= dev
->gflags
;
7367 ret
= __dev_change_flags(dev
, flags
);
7371 changes
= (old_flags
^ dev
->flags
) | (old_gflags
^ dev
->gflags
);
7372 __dev_notify_flags(dev
, old_flags
, changes
);
7375 EXPORT_SYMBOL(dev_change_flags
);
7377 int __dev_set_mtu(struct net_device
*dev
, int new_mtu
)
7379 const struct net_device_ops
*ops
= dev
->netdev_ops
;
7381 if (ops
->ndo_change_mtu
)
7382 return ops
->ndo_change_mtu(dev
, new_mtu
);
7387 EXPORT_SYMBOL(__dev_set_mtu
);
7390 * dev_set_mtu - Change maximum transfer unit
7392 * @new_mtu: new transfer unit
7394 * Change the maximum transfer size of the network device.
7396 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
7400 if (new_mtu
== dev
->mtu
)
7403 /* MTU must be positive, and in range */
7404 if (new_mtu
< 0 || new_mtu
< dev
->min_mtu
) {
7405 net_err_ratelimited("%s: Invalid MTU %d requested, hw min %d\n",
7406 dev
->name
, new_mtu
, dev
->min_mtu
);
7410 if (dev
->max_mtu
> 0 && new_mtu
> dev
->max_mtu
) {
7411 net_err_ratelimited("%s: Invalid MTU %d requested, hw max %d\n",
7412 dev
->name
, new_mtu
, dev
->max_mtu
);
7416 if (!netif_device_present(dev
))
7419 err
= call_netdevice_notifiers(NETDEV_PRECHANGEMTU
, dev
);
7420 err
= notifier_to_errno(err
);
7424 orig_mtu
= dev
->mtu
;
7425 err
= __dev_set_mtu(dev
, new_mtu
);
7428 err
= call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
7429 err
= notifier_to_errno(err
);
7431 /* setting mtu back and notifying everyone again,
7432 * so that they have a chance to revert changes.
7434 __dev_set_mtu(dev
, orig_mtu
);
7435 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
7440 EXPORT_SYMBOL(dev_set_mtu
);
7443 * dev_change_tx_queue_len - Change TX queue length of a netdevice
7445 * @new_len: new tx queue length
7447 int dev_change_tx_queue_len(struct net_device
*dev
, unsigned long new_len
)
7449 unsigned int orig_len
= dev
->tx_queue_len
;
7452 if (new_len
!= (unsigned int)new_len
)
7455 if (new_len
!= orig_len
) {
7456 dev
->tx_queue_len
= new_len
;
7457 res
= call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN
, dev
);
7458 res
= notifier_to_errno(res
);
7461 "refused to change device tx_queue_len\n");
7462 dev
->tx_queue_len
= orig_len
;
7465 return dev_qdisc_change_tx_queue_len(dev
);
7472 * dev_set_group - Change group this device belongs to
7474 * @new_group: group this device should belong to
7476 void dev_set_group(struct net_device
*dev
, int new_group
)
7478 dev
->group
= new_group
;
7480 EXPORT_SYMBOL(dev_set_group
);
7483 * dev_set_mac_address - Change Media Access Control Address
7487 * Change the hardware (MAC) address of the device
7489 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
7491 const struct net_device_ops
*ops
= dev
->netdev_ops
;
7494 if (!ops
->ndo_set_mac_address
)
7496 if (sa
->sa_family
!= dev
->type
)
7498 if (!netif_device_present(dev
))
7500 err
= ops
->ndo_set_mac_address(dev
, sa
);
7503 dev
->addr_assign_type
= NET_ADDR_SET
;
7504 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
7505 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
7508 EXPORT_SYMBOL(dev_set_mac_address
);
7511 * dev_change_carrier - Change device carrier
7513 * @new_carrier: new value
7515 * Change device carrier
7517 int dev_change_carrier(struct net_device
*dev
, bool new_carrier
)
7519 const struct net_device_ops
*ops
= dev
->netdev_ops
;
7521 if (!ops
->ndo_change_carrier
)
7523 if (!netif_device_present(dev
))
7525 return ops
->ndo_change_carrier(dev
, new_carrier
);
7527 EXPORT_SYMBOL(dev_change_carrier
);
7530 * dev_get_phys_port_id - Get device physical port ID
7534 * Get device physical port ID
7536 int dev_get_phys_port_id(struct net_device
*dev
,
7537 struct netdev_phys_item_id
*ppid
)
7539 const struct net_device_ops
*ops
= dev
->netdev_ops
;
7541 if (!ops
->ndo_get_phys_port_id
)
7543 return ops
->ndo_get_phys_port_id(dev
, ppid
);
7545 EXPORT_SYMBOL(dev_get_phys_port_id
);
7548 * dev_get_phys_port_name - Get device physical port name
7551 * @len: limit of bytes to copy to name
7553 * Get device physical port name
7555 int dev_get_phys_port_name(struct net_device
*dev
,
7556 char *name
, size_t len
)
7558 const struct net_device_ops
*ops
= dev
->netdev_ops
;
7560 if (!ops
->ndo_get_phys_port_name
)
7562 return ops
->ndo_get_phys_port_name(dev
, name
, len
);
7564 EXPORT_SYMBOL(dev_get_phys_port_name
);
7567 * dev_change_proto_down - update protocol port state information
7569 * @proto_down: new value
7571 * This info can be used by switch drivers to set the phys state of the
7574 int dev_change_proto_down(struct net_device
*dev
, bool proto_down
)
7576 const struct net_device_ops
*ops
= dev
->netdev_ops
;
7578 if (!ops
->ndo_change_proto_down
)
7580 if (!netif_device_present(dev
))
7582 return ops
->ndo_change_proto_down(dev
, proto_down
);
7584 EXPORT_SYMBOL(dev_change_proto_down
);
7586 void __dev_xdp_query(struct net_device
*dev
, bpf_op_t bpf_op
,
7587 struct netdev_bpf
*xdp
)
7589 memset(xdp
, 0, sizeof(*xdp
));
7590 xdp
->command
= XDP_QUERY_PROG
;
7592 /* Query must always succeed. */
7593 WARN_ON(bpf_op(dev
, xdp
) < 0);
7596 static u8
__dev_xdp_attached(struct net_device
*dev
, bpf_op_t bpf_op
)
7598 struct netdev_bpf xdp
;
7600 __dev_xdp_query(dev
, bpf_op
, &xdp
);
7602 return xdp
.prog_attached
;
7605 static int dev_xdp_install(struct net_device
*dev
, bpf_op_t bpf_op
,
7606 struct netlink_ext_ack
*extack
, u32 flags
,
7607 struct bpf_prog
*prog
)
7609 struct netdev_bpf xdp
;
7611 memset(&xdp
, 0, sizeof(xdp
));
7612 if (flags
& XDP_FLAGS_HW_MODE
)
7613 xdp
.command
= XDP_SETUP_PROG_HW
;
7615 xdp
.command
= XDP_SETUP_PROG
;
7616 xdp
.extack
= extack
;
7620 return bpf_op(dev
, &xdp
);
7623 static void dev_xdp_uninstall(struct net_device
*dev
)
7625 struct netdev_bpf xdp
;
7628 /* Remove generic XDP */
7629 WARN_ON(dev_xdp_install(dev
, generic_xdp_install
, NULL
, 0, NULL
));
7631 /* Remove from the driver */
7632 ndo_bpf
= dev
->netdev_ops
->ndo_bpf
;
7636 __dev_xdp_query(dev
, ndo_bpf
, &xdp
);
7637 if (xdp
.prog_attached
== XDP_ATTACHED_NONE
)
7640 /* Program removal should always succeed */
7641 WARN_ON(dev_xdp_install(dev
, ndo_bpf
, NULL
, xdp
.prog_flags
, NULL
));
7645 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
7647 * @extack: netlink extended ack
7648 * @fd: new program fd or negative value to clear
7649 * @flags: xdp-related flags
7651 * Set or clear a bpf program for a device
7653 int dev_change_xdp_fd(struct net_device
*dev
, struct netlink_ext_ack
*extack
,
7656 const struct net_device_ops
*ops
= dev
->netdev_ops
;
7657 struct bpf_prog
*prog
= NULL
;
7658 bpf_op_t bpf_op
, bpf_chk
;
7663 bpf_op
= bpf_chk
= ops
->ndo_bpf
;
7664 if (!bpf_op
&& (flags
& (XDP_FLAGS_DRV_MODE
| XDP_FLAGS_HW_MODE
)))
7666 if (!bpf_op
|| (flags
& XDP_FLAGS_SKB_MODE
))
7667 bpf_op
= generic_xdp_install
;
7668 if (bpf_op
== bpf_chk
)
7669 bpf_chk
= generic_xdp_install
;
7672 if (bpf_chk
&& __dev_xdp_attached(dev
, bpf_chk
))
7674 if ((flags
& XDP_FLAGS_UPDATE_IF_NOEXIST
) &&
7675 __dev_xdp_attached(dev
, bpf_op
))
7678 prog
= bpf_prog_get_type_dev(fd
, BPF_PROG_TYPE_XDP
,
7679 bpf_op
== ops
->ndo_bpf
);
7681 return PTR_ERR(prog
);
7683 if (!(flags
& XDP_FLAGS_HW_MODE
) &&
7684 bpf_prog_is_dev_bound(prog
->aux
)) {
7685 NL_SET_ERR_MSG(extack
, "using device-bound program without HW_MODE flag is not supported");
7691 err
= dev_xdp_install(dev
, bpf_op
, extack
, flags
, prog
);
7692 if (err
< 0 && prog
)
7699 * dev_new_index - allocate an ifindex
7700 * @net: the applicable net namespace
7702 * Returns a suitable unique value for a new device interface
7703 * number. The caller must hold the rtnl semaphore or the
7704 * dev_base_lock to be sure it remains unique.
7706 static int dev_new_index(struct net
*net
)
7708 int ifindex
= net
->ifindex
;
7713 if (!__dev_get_by_index(net
, ifindex
))
7714 return net
->ifindex
= ifindex
;
7718 /* Delayed registration/unregisteration */
7719 static LIST_HEAD(net_todo_list
);
7720 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq
);
7722 static void net_set_todo(struct net_device
*dev
)
7724 list_add_tail(&dev
->todo_list
, &net_todo_list
);
7725 dev_net(dev
)->dev_unreg_count
++;
7728 static void rollback_registered_many(struct list_head
*head
)
7730 struct net_device
*dev
, *tmp
;
7731 LIST_HEAD(close_head
);
7733 BUG_ON(dev_boot_phase
);
7736 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
7737 /* Some devices call without registering
7738 * for initialization unwind. Remove those
7739 * devices and proceed with the remaining.
7741 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
7742 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
7746 list_del(&dev
->unreg_list
);
7749 dev
->dismantle
= true;
7750 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
7753 /* If device is running, close it first. */
7754 list_for_each_entry(dev
, head
, unreg_list
)
7755 list_add_tail(&dev
->close_list
, &close_head
);
7756 dev_close_many(&close_head
, true);
7758 list_for_each_entry(dev
, head
, unreg_list
) {
7759 /* And unlink it from device chain. */
7760 unlist_netdevice(dev
);
7762 dev
->reg_state
= NETREG_UNREGISTERING
;
7764 flush_all_backlogs();
7768 list_for_each_entry(dev
, head
, unreg_list
) {
7769 struct sk_buff
*skb
= NULL
;
7771 /* Shutdown queueing discipline. */
7774 dev_xdp_uninstall(dev
);
7776 /* Notify protocols, that we are about to destroy
7777 * this device. They should clean all the things.
7779 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
7781 if (!dev
->rtnl_link_ops
||
7782 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
7783 skb
= rtmsg_ifinfo_build_skb(RTM_DELLINK
, dev
, ~0U, 0,
7784 GFP_KERNEL
, NULL
, 0);
7787 * Flush the unicast and multicast chains
7792 if (dev
->netdev_ops
->ndo_uninit
)
7793 dev
->netdev_ops
->ndo_uninit(dev
);
7796 rtmsg_ifinfo_send(skb
, dev
, GFP_KERNEL
);
7798 /* Notifier chain MUST detach us all upper devices. */
7799 WARN_ON(netdev_has_any_upper_dev(dev
));
7800 WARN_ON(netdev_has_any_lower_dev(dev
));
7802 /* Remove entries from kobject tree */
7803 netdev_unregister_kobject(dev
);
7805 /* Remove XPS queueing entries */
7806 netif_reset_xps_queues_gt(dev
, 0);
7812 list_for_each_entry(dev
, head
, unreg_list
)
7816 static void rollback_registered(struct net_device
*dev
)
7820 list_add(&dev
->unreg_list
, &single
);
7821 rollback_registered_many(&single
);
7825 static netdev_features_t
netdev_sync_upper_features(struct net_device
*lower
,
7826 struct net_device
*upper
, netdev_features_t features
)
7828 netdev_features_t upper_disables
= NETIF_F_UPPER_DISABLES
;
7829 netdev_features_t feature
;
7832 for_each_netdev_feature(&upper_disables
, feature_bit
) {
7833 feature
= __NETIF_F_BIT(feature_bit
);
7834 if (!(upper
->wanted_features
& feature
)
7835 && (features
& feature
)) {
7836 netdev_dbg(lower
, "Dropping feature %pNF, upper dev %s has it off.\n",
7837 &feature
, upper
->name
);
7838 features
&= ~feature
;
7845 static void netdev_sync_lower_features(struct net_device
*upper
,
7846 struct net_device
*lower
, netdev_features_t features
)
7848 netdev_features_t upper_disables
= NETIF_F_UPPER_DISABLES
;
7849 netdev_features_t feature
;
7852 for_each_netdev_feature(&upper_disables
, feature_bit
) {
7853 feature
= __NETIF_F_BIT(feature_bit
);
7854 if (!(features
& feature
) && (lower
->features
& feature
)) {
7855 netdev_dbg(upper
, "Disabling feature %pNF on lower dev %s.\n",
7856 &feature
, lower
->name
);
7857 lower
->wanted_features
&= ~feature
;
7858 netdev_update_features(lower
);
7860 if (unlikely(lower
->features
& feature
))
7861 netdev_WARN(upper
, "failed to disable %pNF on %s!\n",
7862 &feature
, lower
->name
);
7867 static netdev_features_t
netdev_fix_features(struct net_device
*dev
,
7868 netdev_features_t features
)
7870 /* Fix illegal checksum combinations */
7871 if ((features
& NETIF_F_HW_CSUM
) &&
7872 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
7873 netdev_warn(dev
, "mixed HW and IP checksum settings.\n");
7874 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
7877 /* TSO requires that SG is present as well. */
7878 if ((features
& NETIF_F_ALL_TSO
) && !(features
& NETIF_F_SG
)) {
7879 netdev_dbg(dev
, "Dropping TSO features since no SG feature.\n");
7880 features
&= ~NETIF_F_ALL_TSO
;
7883 if ((features
& NETIF_F_TSO
) && !(features
& NETIF_F_HW_CSUM
) &&
7884 !(features
& NETIF_F_IP_CSUM
)) {
7885 netdev_dbg(dev
, "Dropping TSO features since no CSUM feature.\n");
7886 features
&= ~NETIF_F_TSO
;
7887 features
&= ~NETIF_F_TSO_ECN
;
7890 if ((features
& NETIF_F_TSO6
) && !(features
& NETIF_F_HW_CSUM
) &&
7891 !(features
& NETIF_F_IPV6_CSUM
)) {
7892 netdev_dbg(dev
, "Dropping TSO6 features since no CSUM feature.\n");
7893 features
&= ~NETIF_F_TSO6
;
7896 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
7897 if ((features
& NETIF_F_TSO_MANGLEID
) && !(features
& NETIF_F_TSO
))
7898 features
&= ~NETIF_F_TSO_MANGLEID
;
7900 /* TSO ECN requires that TSO is present as well. */
7901 if ((features
& NETIF_F_ALL_TSO
) == NETIF_F_TSO_ECN
)
7902 features
&= ~NETIF_F_TSO_ECN
;
7904 /* Software GSO depends on SG. */
7905 if ((features
& NETIF_F_GSO
) && !(features
& NETIF_F_SG
)) {
7906 netdev_dbg(dev
, "Dropping NETIF_F_GSO since no SG feature.\n");
7907 features
&= ~NETIF_F_GSO
;
7910 /* GSO partial features require GSO partial be set */
7911 if ((features
& dev
->gso_partial_features
) &&
7912 !(features
& NETIF_F_GSO_PARTIAL
)) {
7914 "Dropping partially supported GSO features since no GSO partial.\n");
7915 features
&= ~dev
->gso_partial_features
;
7918 if (!(features
& NETIF_F_RXCSUM
)) {
7919 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet
7920 * successfully merged by hardware must also have the
7921 * checksum verified by hardware. If the user does not
7922 * want to enable RXCSUM, logically, we should disable GRO_HW.
7924 if (features
& NETIF_F_GRO_HW
) {
7925 netdev_dbg(dev
, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
7926 features
&= ~NETIF_F_GRO_HW
;
7930 /* LRO/HW-GRO features cannot be combined with RX-FCS */
7931 if (features
& NETIF_F_RXFCS
) {
7932 if (features
& NETIF_F_LRO
) {
7933 netdev_dbg(dev
, "Dropping LRO feature since RX-FCS is requested.\n");
7934 features
&= ~NETIF_F_LRO
;
7937 if (features
& NETIF_F_GRO_HW
) {
7938 netdev_dbg(dev
, "Dropping HW-GRO feature since RX-FCS is requested.\n");
7939 features
&= ~NETIF_F_GRO_HW
;
7946 int __netdev_update_features(struct net_device
*dev
)
7948 struct net_device
*upper
, *lower
;
7949 netdev_features_t features
;
7950 struct list_head
*iter
;
7955 features
= netdev_get_wanted_features(dev
);
7957 if (dev
->netdev_ops
->ndo_fix_features
)
7958 features
= dev
->netdev_ops
->ndo_fix_features(dev
, features
);
7960 /* driver might be less strict about feature dependencies */
7961 features
= netdev_fix_features(dev
, features
);
7963 /* some features can't be enabled if they're off an an upper device */
7964 netdev_for_each_upper_dev_rcu(dev
, upper
, iter
)
7965 features
= netdev_sync_upper_features(dev
, upper
, features
);
7967 if (dev
->features
== features
)
7970 netdev_dbg(dev
, "Features changed: %pNF -> %pNF\n",
7971 &dev
->features
, &features
);
7973 if (dev
->netdev_ops
->ndo_set_features
)
7974 err
= dev
->netdev_ops
->ndo_set_features(dev
, features
);
7978 if (unlikely(err
< 0)) {
7980 "set_features() failed (%d); wanted %pNF, left %pNF\n",
7981 err
, &features
, &dev
->features
);
7982 /* return non-0 since some features might have changed and
7983 * it's better to fire a spurious notification than miss it
7989 /* some features must be disabled on lower devices when disabled
7990 * on an upper device (think: bonding master or bridge)
7992 netdev_for_each_lower_dev(dev
, lower
, iter
)
7993 netdev_sync_lower_features(dev
, lower
, features
);
7996 netdev_features_t diff
= features
^ dev
->features
;
7998 if (diff
& NETIF_F_RX_UDP_TUNNEL_PORT
) {
7999 /* udp_tunnel_{get,drop}_rx_info both need
8000 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
8001 * device, or they won't do anything.
8002 * Thus we need to update dev->features
8003 * *before* calling udp_tunnel_get_rx_info,
8004 * but *after* calling udp_tunnel_drop_rx_info.
8006 if (features
& NETIF_F_RX_UDP_TUNNEL_PORT
) {
8007 dev
->features
= features
;
8008 udp_tunnel_get_rx_info(dev
);
8010 udp_tunnel_drop_rx_info(dev
);
8014 if (diff
& NETIF_F_HW_VLAN_CTAG_FILTER
) {
8015 if (features
& NETIF_F_HW_VLAN_CTAG_FILTER
) {
8016 dev
->features
= features
;
8017 err
|= vlan_get_rx_ctag_filter_info(dev
);
8019 vlan_drop_rx_ctag_filter_info(dev
);
8023 if (diff
& NETIF_F_HW_VLAN_STAG_FILTER
) {
8024 if (features
& NETIF_F_HW_VLAN_STAG_FILTER
) {
8025 dev
->features
= features
;
8026 err
|= vlan_get_rx_stag_filter_info(dev
);
8028 vlan_drop_rx_stag_filter_info(dev
);
8032 dev
->features
= features
;
8035 return err
< 0 ? 0 : 1;
8039 * netdev_update_features - recalculate device features
8040 * @dev: the device to check
8042 * Recalculate dev->features set and send notifications if it
8043 * has changed. Should be called after driver or hardware dependent
8044 * conditions might have changed that influence the features.
8046 void netdev_update_features(struct net_device
*dev
)
8048 if (__netdev_update_features(dev
))
8049 netdev_features_change(dev
);
8051 EXPORT_SYMBOL(netdev_update_features
);
8054 * netdev_change_features - recalculate device features
8055 * @dev: the device to check
8057 * Recalculate dev->features set and send notifications even
8058 * if they have not changed. Should be called instead of
8059 * netdev_update_features() if also dev->vlan_features might
8060 * have changed to allow the changes to be propagated to stacked
8063 void netdev_change_features(struct net_device
*dev
)
8065 __netdev_update_features(dev
);
8066 netdev_features_change(dev
);
8068 EXPORT_SYMBOL(netdev_change_features
);
8071 * netif_stacked_transfer_operstate - transfer operstate
8072 * @rootdev: the root or lower level device to transfer state from
8073 * @dev: the device to transfer operstate to
8075 * Transfer operational state from root to device. This is normally
8076 * called when a stacking relationship exists between the root
8077 * device and the device(a leaf device).
8079 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
8080 struct net_device
*dev
)
8082 if (rootdev
->operstate
== IF_OPER_DORMANT
)
8083 netif_dormant_on(dev
);
8085 netif_dormant_off(dev
);
8087 if (netif_carrier_ok(rootdev
))
8088 netif_carrier_on(dev
);
8090 netif_carrier_off(dev
);
8092 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
8094 static int netif_alloc_rx_queues(struct net_device
*dev
)
8096 unsigned int i
, count
= dev
->num_rx_queues
;
8097 struct netdev_rx_queue
*rx
;
8098 size_t sz
= count
* sizeof(*rx
);
8103 rx
= kvzalloc(sz
, GFP_KERNEL
| __GFP_RETRY_MAYFAIL
);
8109 for (i
= 0; i
< count
; i
++) {
8112 /* XDP RX-queue setup */
8113 err
= xdp_rxq_info_reg(&rx
[i
].xdp_rxq
, dev
, i
);
8120 /* Rollback successful reg's and free other resources */
8122 xdp_rxq_info_unreg(&rx
[i
].xdp_rxq
);
8128 static void netif_free_rx_queues(struct net_device
*dev
)
8130 unsigned int i
, count
= dev
->num_rx_queues
;
8132 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
8136 for (i
= 0; i
< count
; i
++)
8137 xdp_rxq_info_unreg(&dev
->_rx
[i
].xdp_rxq
);
8142 static void netdev_init_one_queue(struct net_device
*dev
,
8143 struct netdev_queue
*queue
, void *_unused
)
8145 /* Initialize queue lock */
8146 spin_lock_init(&queue
->_xmit_lock
);
8147 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
8148 queue
->xmit_lock_owner
= -1;
8149 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
8152 dql_init(&queue
->dql
, HZ
);
8156 static void netif_free_tx_queues(struct net_device
*dev
)
8161 static int netif_alloc_netdev_queues(struct net_device
*dev
)
8163 unsigned int count
= dev
->num_tx_queues
;
8164 struct netdev_queue
*tx
;
8165 size_t sz
= count
* sizeof(*tx
);
8167 if (count
< 1 || count
> 0xffff)
8170 tx
= kvzalloc(sz
, GFP_KERNEL
| __GFP_RETRY_MAYFAIL
);
8176 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
8177 spin_lock_init(&dev
->tx_global_lock
);
8182 void netif_tx_stop_all_queues(struct net_device
*dev
)
8186 for (i
= 0; i
< dev
->num_tx_queues
; i
++) {
8187 struct netdev_queue
*txq
= netdev_get_tx_queue(dev
, i
);
8189 netif_tx_stop_queue(txq
);
8192 EXPORT_SYMBOL(netif_tx_stop_all_queues
);
8195 * register_netdevice - register a network device
8196 * @dev: device to register
8198 * Take a completed network device structure and add it to the kernel
8199 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
8200 * chain. 0 is returned on success. A negative errno code is returned
8201 * on a failure to set up the device, or if the name is a duplicate.
8203 * Callers must hold the rtnl semaphore. You may want
8204 * register_netdev() instead of this.
8207 * The locking appears insufficient to guarantee two parallel registers
8208 * will not get the same name.
8211 int register_netdevice(struct net_device
*dev
)
8214 struct net
*net
= dev_net(dev
);
8216 BUILD_BUG_ON(sizeof(netdev_features_t
) * BITS_PER_BYTE
<
8217 NETDEV_FEATURE_COUNT
);
8218 BUG_ON(dev_boot_phase
);
8223 /* When net_device's are persistent, this will be fatal. */
8224 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
8227 spin_lock_init(&dev
->addr_list_lock
);
8228 netdev_set_addr_lockdep_class(dev
);
8230 ret
= dev_get_valid_name(net
, dev
, dev
->name
);
8234 /* Init, if this function is available */
8235 if (dev
->netdev_ops
->ndo_init
) {
8236 ret
= dev
->netdev_ops
->ndo_init(dev
);
8244 if (((dev
->hw_features
| dev
->features
) &
8245 NETIF_F_HW_VLAN_CTAG_FILTER
) &&
8246 (!dev
->netdev_ops
->ndo_vlan_rx_add_vid
||
8247 !dev
->netdev_ops
->ndo_vlan_rx_kill_vid
)) {
8248 netdev_WARN(dev
, "Buggy VLAN acceleration in driver!\n");
8255 dev
->ifindex
= dev_new_index(net
);
8256 else if (__dev_get_by_index(net
, dev
->ifindex
))
8259 /* Transfer changeable features to wanted_features and enable
8260 * software offloads (GSO and GRO).
8262 dev
->hw_features
|= NETIF_F_SOFT_FEATURES
;
8263 dev
->features
|= NETIF_F_SOFT_FEATURES
;
8265 if (dev
->netdev_ops
->ndo_udp_tunnel_add
) {
8266 dev
->features
|= NETIF_F_RX_UDP_TUNNEL_PORT
;
8267 dev
->hw_features
|= NETIF_F_RX_UDP_TUNNEL_PORT
;
8270 dev
->wanted_features
= dev
->features
& dev
->hw_features
;
8272 if (!(dev
->flags
& IFF_LOOPBACK
))
8273 dev
->hw_features
|= NETIF_F_NOCACHE_COPY
;
8275 /* If IPv4 TCP segmentation offload is supported we should also
8276 * allow the device to enable segmenting the frame with the option
8277 * of ignoring a static IP ID value. This doesn't enable the
8278 * feature itself but allows the user to enable it later.
8280 if (dev
->hw_features
& NETIF_F_TSO
)
8281 dev
->hw_features
|= NETIF_F_TSO_MANGLEID
;
8282 if (dev
->vlan_features
& NETIF_F_TSO
)
8283 dev
->vlan_features
|= NETIF_F_TSO_MANGLEID
;
8284 if (dev
->mpls_features
& NETIF_F_TSO
)
8285 dev
->mpls_features
|= NETIF_F_TSO_MANGLEID
;
8286 if (dev
->hw_enc_features
& NETIF_F_TSO
)
8287 dev
->hw_enc_features
|= NETIF_F_TSO_MANGLEID
;
8289 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
8291 dev
->vlan_features
|= NETIF_F_HIGHDMA
;
8293 /* Make NETIF_F_SG inheritable to tunnel devices.
8295 dev
->hw_enc_features
|= NETIF_F_SG
| NETIF_F_GSO_PARTIAL
;
8297 /* Make NETIF_F_SG inheritable to MPLS.
8299 dev
->mpls_features
|= NETIF_F_SG
;
8301 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
8302 ret
= notifier_to_errno(ret
);
8306 ret
= netdev_register_kobject(dev
);
8309 dev
->reg_state
= NETREG_REGISTERED
;
8311 __netdev_update_features(dev
);
8314 * Default initial state at registry is that the
8315 * device is present.
8318 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
8320 linkwatch_init_dev(dev
);
8322 dev_init_scheduler(dev
);
8324 list_netdevice(dev
);
8325 add_device_randomness(dev
->dev_addr
, dev
->addr_len
);
8327 /* If the device has permanent device address, driver should
8328 * set dev_addr and also addr_assign_type should be set to
8329 * NET_ADDR_PERM (default value).
8331 if (dev
->addr_assign_type
== NET_ADDR_PERM
)
8332 memcpy(dev
->perm_addr
, dev
->dev_addr
, dev
->addr_len
);
8334 /* Notify protocols, that a new device appeared. */
8335 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
8336 ret
= notifier_to_errno(ret
);
8338 rollback_registered(dev
);
8339 dev
->reg_state
= NETREG_UNREGISTERED
;
8342 * Prevent userspace races by waiting until the network
8343 * device is fully setup before sending notifications.
8345 if (!dev
->rtnl_link_ops
||
8346 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
8347 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U, GFP_KERNEL
);
8353 if (dev
->netdev_ops
->ndo_uninit
)
8354 dev
->netdev_ops
->ndo_uninit(dev
);
8355 if (dev
->priv_destructor
)
8356 dev
->priv_destructor(dev
);
8359 EXPORT_SYMBOL(register_netdevice
);
8362 * init_dummy_netdev - init a dummy network device for NAPI
8363 * @dev: device to init
8365 * This takes a network device structure and initialize the minimum
8366 * amount of fields so it can be used to schedule NAPI polls without
8367 * registering a full blown interface. This is to be used by drivers
8368 * that need to tie several hardware interfaces to a single NAPI
8369 * poll scheduler due to HW limitations.
8371 int init_dummy_netdev(struct net_device
*dev
)
8373 /* Clear everything. Note we don't initialize spinlocks
8374 * are they aren't supposed to be taken by any of the
8375 * NAPI code and this dummy netdev is supposed to be
8376 * only ever used for NAPI polls
8378 memset(dev
, 0, sizeof(struct net_device
));
8380 /* make sure we BUG if trying to hit standard
8381 * register/unregister code path
8383 dev
->reg_state
= NETREG_DUMMY
;
8385 /* NAPI wants this */
8386 INIT_LIST_HEAD(&dev
->napi_list
);
8388 /* a dummy interface is started by default */
8389 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
8390 set_bit(__LINK_STATE_START
, &dev
->state
);
8392 /* Note : We dont allocate pcpu_refcnt for dummy devices,
8393 * because users of this 'device' dont need to change
8399 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
8403 * register_netdev - register a network device
8404 * @dev: device to register
8406 * Take a completed network device structure and add it to the kernel
8407 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
8408 * chain. 0 is returned on success. A negative errno code is returned
8409 * on a failure to set up the device, or if the name is a duplicate.
8411 * This is a wrapper around register_netdevice that takes the rtnl semaphore
8412 * and expands the device name if you passed a format string to
8415 int register_netdev(struct net_device
*dev
)
8419 if (rtnl_lock_killable())
8421 err
= register_netdevice(dev
);
8425 EXPORT_SYMBOL(register_netdev
);
8427 int netdev_refcnt_read(const struct net_device
*dev
)
8431 for_each_possible_cpu(i
)
8432 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
8435 EXPORT_SYMBOL(netdev_refcnt_read
);
8438 * netdev_wait_allrefs - wait until all references are gone.
8439 * @dev: target net_device
8441 * This is called when unregistering network devices.
8443 * Any protocol or device that holds a reference should register
8444 * for netdevice notification, and cleanup and put back the
8445 * reference if they receive an UNREGISTER event.
8446 * We can get stuck here if buggy protocols don't correctly
8449 static void netdev_wait_allrefs(struct net_device
*dev
)
8451 unsigned long rebroadcast_time
, warning_time
;
8454 linkwatch_forget_dev(dev
);
8456 rebroadcast_time
= warning_time
= jiffies
;
8457 refcnt
= netdev_refcnt_read(dev
);
8459 while (refcnt
!= 0) {
8460 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
8463 /* Rebroadcast unregister notification */
8464 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
8470 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
8472 /* We must not have linkwatch events
8473 * pending on unregister. If this
8474 * happens, we simply run the queue
8475 * unscheduled, resulting in a noop
8478 linkwatch_run_queue();
8483 rebroadcast_time
= jiffies
;
8488 refcnt
= netdev_refcnt_read(dev
);
8490 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
8491 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
8493 warning_time
= jiffies
;
8502 * register_netdevice(x1);
8503 * register_netdevice(x2);
8505 * unregister_netdevice(y1);
8506 * unregister_netdevice(y2);
8512 * We are invoked by rtnl_unlock().
8513 * This allows us to deal with problems:
8514 * 1) We can delete sysfs objects which invoke hotplug
8515 * without deadlocking with linkwatch via keventd.
8516 * 2) Since we run with the RTNL semaphore not held, we can sleep
8517 * safely in order to wait for the netdev refcnt to drop to zero.
8519 * We must not return until all unregister events added during
8520 * the interval the lock was held have been completed.
8522 void netdev_run_todo(void)
8524 struct list_head list
;
8526 /* Snapshot list, allow later requests */
8527 list_replace_init(&net_todo_list
, &list
);
8532 /* Wait for rcu callbacks to finish before next phase */
8533 if (!list_empty(&list
))
8536 while (!list_empty(&list
)) {
8537 struct net_device
*dev
8538 = list_first_entry(&list
, struct net_device
, todo_list
);
8539 list_del(&dev
->todo_list
);
8541 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
8542 pr_err("network todo '%s' but state %d\n",
8543 dev
->name
, dev
->reg_state
);
8548 dev
->reg_state
= NETREG_UNREGISTERED
;
8550 netdev_wait_allrefs(dev
);
8553 BUG_ON(netdev_refcnt_read(dev
));
8554 BUG_ON(!list_empty(&dev
->ptype_all
));
8555 BUG_ON(!list_empty(&dev
->ptype_specific
));
8556 WARN_ON(rcu_access_pointer(dev
->ip_ptr
));
8557 WARN_ON(rcu_access_pointer(dev
->ip6_ptr
));
8558 #if IS_ENABLED(CONFIG_DECNET)
8559 WARN_ON(dev
->dn_ptr
);
8561 if (dev
->priv_destructor
)
8562 dev
->priv_destructor(dev
);
8563 if (dev
->needs_free_netdev
)
8566 /* Report a network device has been unregistered */
8568 dev_net(dev
)->dev_unreg_count
--;
8570 wake_up(&netdev_unregistering_wq
);
8572 /* Free network device */
8573 kobject_put(&dev
->dev
.kobj
);
8577 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
8578 * all the same fields in the same order as net_device_stats, with only
8579 * the type differing, but rtnl_link_stats64 may have additional fields
8580 * at the end for newer counters.
8582 void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
8583 const struct net_device_stats
*netdev_stats
)
8585 #if BITS_PER_LONG == 64
8586 BUILD_BUG_ON(sizeof(*stats64
) < sizeof(*netdev_stats
));
8587 memcpy(stats64
, netdev_stats
, sizeof(*netdev_stats
));
8588 /* zero out counters that only exist in rtnl_link_stats64 */
8589 memset((char *)stats64
+ sizeof(*netdev_stats
), 0,
8590 sizeof(*stats64
) - sizeof(*netdev_stats
));
8592 size_t i
, n
= sizeof(*netdev_stats
) / sizeof(unsigned long);
8593 const unsigned long *src
= (const unsigned long *)netdev_stats
;
8594 u64
*dst
= (u64
*)stats64
;
8596 BUILD_BUG_ON(n
> sizeof(*stats64
) / sizeof(u64
));
8597 for (i
= 0; i
< n
; i
++)
8599 /* zero out counters that only exist in rtnl_link_stats64 */
8600 memset((char *)stats64
+ n
* sizeof(u64
), 0,
8601 sizeof(*stats64
) - n
* sizeof(u64
));
8604 EXPORT_SYMBOL(netdev_stats_to_stats64
);
8607 * dev_get_stats - get network device statistics
8608 * @dev: device to get statistics from
8609 * @storage: place to store stats
8611 * Get network statistics from device. Return @storage.
8612 * The device driver may provide its own method by setting
8613 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
8614 * otherwise the internal statistics structure is used.
8616 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
8617 struct rtnl_link_stats64
*storage
)
8619 const struct net_device_ops
*ops
= dev
->netdev_ops
;
8621 if (ops
->ndo_get_stats64
) {
8622 memset(storage
, 0, sizeof(*storage
));
8623 ops
->ndo_get_stats64(dev
, storage
);
8624 } else if (ops
->ndo_get_stats
) {
8625 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
8627 netdev_stats_to_stats64(storage
, &dev
->stats
);
8629 storage
->rx_dropped
+= (unsigned long)atomic_long_read(&dev
->rx_dropped
);
8630 storage
->tx_dropped
+= (unsigned long)atomic_long_read(&dev
->tx_dropped
);
8631 storage
->rx_nohandler
+= (unsigned long)atomic_long_read(&dev
->rx_nohandler
);
8634 EXPORT_SYMBOL(dev_get_stats
);
8636 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
8638 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
8640 #ifdef CONFIG_NET_CLS_ACT
8643 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
8646 netdev_init_one_queue(dev
, queue
, NULL
);
8647 RCU_INIT_POINTER(queue
->qdisc
, &noop_qdisc
);
8648 queue
->qdisc_sleeping
= &noop_qdisc
;
8649 rcu_assign_pointer(dev
->ingress_queue
, queue
);
8654 static const struct ethtool_ops default_ethtool_ops
;
8656 void netdev_set_default_ethtool_ops(struct net_device
*dev
,
8657 const struct ethtool_ops
*ops
)
8659 if (dev
->ethtool_ops
== &default_ethtool_ops
)
8660 dev
->ethtool_ops
= ops
;
8662 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops
);
8664 void netdev_freemem(struct net_device
*dev
)
8666 char *addr
= (char *)dev
- dev
->padded
;
8672 * alloc_netdev_mqs - allocate network device
8673 * @sizeof_priv: size of private data to allocate space for
8674 * @name: device name format string
8675 * @name_assign_type: origin of device name
8676 * @setup: callback to initialize device
8677 * @txqs: the number of TX subqueues to allocate
8678 * @rxqs: the number of RX subqueues to allocate
8680 * Allocates a struct net_device with private data area for driver use
8681 * and performs basic initialization. Also allocates subqueue structs
8682 * for each queue on the device.
8684 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
8685 unsigned char name_assign_type
,
8686 void (*setup
)(struct net_device
*),
8687 unsigned int txqs
, unsigned int rxqs
)
8689 struct net_device
*dev
;
8690 unsigned int alloc_size
;
8691 struct net_device
*p
;
8693 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
8696 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
8701 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
8705 alloc_size
= sizeof(struct net_device
);
8707 /* ensure 32-byte alignment of private area */
8708 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
8709 alloc_size
+= sizeof_priv
;
8711 /* ensure 32-byte alignment of whole construct */
8712 alloc_size
+= NETDEV_ALIGN
- 1;
8714 p
= kvzalloc(alloc_size
, GFP_KERNEL
| __GFP_RETRY_MAYFAIL
);
8718 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
8719 dev
->padded
= (char *)dev
- (char *)p
;
8721 dev
->pcpu_refcnt
= alloc_percpu(int);
8722 if (!dev
->pcpu_refcnt
)
8725 if (dev_addr_init(dev
))
8731 dev_net_set(dev
, &init_net
);
8733 dev
->gso_max_size
= GSO_MAX_SIZE
;
8734 dev
->gso_max_segs
= GSO_MAX_SEGS
;
8736 INIT_LIST_HEAD(&dev
->napi_list
);
8737 INIT_LIST_HEAD(&dev
->unreg_list
);
8738 INIT_LIST_HEAD(&dev
->close_list
);
8739 INIT_LIST_HEAD(&dev
->link_watch_list
);
8740 INIT_LIST_HEAD(&dev
->adj_list
.upper
);
8741 INIT_LIST_HEAD(&dev
->adj_list
.lower
);
8742 INIT_LIST_HEAD(&dev
->ptype_all
);
8743 INIT_LIST_HEAD(&dev
->ptype_specific
);
8744 #ifdef CONFIG_NET_SCHED
8745 hash_init(dev
->qdisc_hash
);
8747 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
| IFF_XMIT_DST_RELEASE_PERM
;
8750 if (!dev
->tx_queue_len
) {
8751 dev
->priv_flags
|= IFF_NO_QUEUE
;
8752 dev
->tx_queue_len
= DEFAULT_TX_QUEUE_LEN
;
8755 dev
->num_tx_queues
= txqs
;
8756 dev
->real_num_tx_queues
= txqs
;
8757 if (netif_alloc_netdev_queues(dev
))
8760 dev
->num_rx_queues
= rxqs
;
8761 dev
->real_num_rx_queues
= rxqs
;
8762 if (netif_alloc_rx_queues(dev
))
8765 strcpy(dev
->name
, name
);
8766 dev
->name_assign_type
= name_assign_type
;
8767 dev
->group
= INIT_NETDEV_GROUP
;
8768 if (!dev
->ethtool_ops
)
8769 dev
->ethtool_ops
= &default_ethtool_ops
;
8771 nf_hook_ingress_init(dev
);
8780 free_percpu(dev
->pcpu_refcnt
);
8782 netdev_freemem(dev
);
8785 EXPORT_SYMBOL(alloc_netdev_mqs
);
8788 * free_netdev - free network device
8791 * This function does the last stage of destroying an allocated device
8792 * interface. The reference to the device object is released. If this
8793 * is the last reference then it will be freed.Must be called in process
8796 void free_netdev(struct net_device
*dev
)
8798 struct napi_struct
*p
, *n
;
8801 netif_free_tx_queues(dev
);
8802 netif_free_rx_queues(dev
);
8804 kfree(rcu_dereference_protected(dev
->ingress_queue
, 1));
8806 /* Flush device addresses */
8807 dev_addr_flush(dev
);
8809 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
8812 free_percpu(dev
->pcpu_refcnt
);
8813 dev
->pcpu_refcnt
= NULL
;
8815 /* Compatibility with error handling in drivers */
8816 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
8817 netdev_freemem(dev
);
8821 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
8822 dev
->reg_state
= NETREG_RELEASED
;
8824 /* will free via device release */
8825 put_device(&dev
->dev
);
8827 EXPORT_SYMBOL(free_netdev
);
8830 * synchronize_net - Synchronize with packet receive processing
8832 * Wait for packets currently being received to be done.
8833 * Does not block later packets from starting.
8835 void synchronize_net(void)
8838 if (rtnl_is_locked())
8839 synchronize_rcu_expedited();
8843 EXPORT_SYMBOL(synchronize_net
);
8846 * unregister_netdevice_queue - remove device from the kernel
8850 * This function shuts down a device interface and removes it
8851 * from the kernel tables.
8852 * If head not NULL, device is queued to be unregistered later.
8854 * Callers must hold the rtnl semaphore. You may want
8855 * unregister_netdev() instead of this.
8858 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
8863 list_move_tail(&dev
->unreg_list
, head
);
8865 rollback_registered(dev
);
8866 /* Finish processing unregister after unlock */
8870 EXPORT_SYMBOL(unregister_netdevice_queue
);
8873 * unregister_netdevice_many - unregister many devices
8874 * @head: list of devices
8876 * Note: As most callers use a stack allocated list_head,
8877 * we force a list_del() to make sure stack wont be corrupted later.
8879 void unregister_netdevice_many(struct list_head
*head
)
8881 struct net_device
*dev
;
8883 if (!list_empty(head
)) {
8884 rollback_registered_many(head
);
8885 list_for_each_entry(dev
, head
, unreg_list
)
8890 EXPORT_SYMBOL(unregister_netdevice_many
);
8893 * unregister_netdev - remove device from the kernel
8896 * This function shuts down a device interface and removes it
8897 * from the kernel tables.
8899 * This is just a wrapper for unregister_netdevice that takes
8900 * the rtnl semaphore. In general you want to use this and not
8901 * unregister_netdevice.
8903 void unregister_netdev(struct net_device
*dev
)
8906 unregister_netdevice(dev
);
8909 EXPORT_SYMBOL(unregister_netdev
);
8912 * dev_change_net_namespace - move device to different nethost namespace
8914 * @net: network namespace
8915 * @pat: If not NULL name pattern to try if the current device name
8916 * is already taken in the destination network namespace.
8918 * This function shuts down a device interface and moves it
8919 * to a new network namespace. On success 0 is returned, on
8920 * a failure a netagive errno code is returned.
8922 * Callers must hold the rtnl semaphore.
8925 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
8927 int err
, new_nsid
, new_ifindex
;
8931 /* Don't allow namespace local devices to be moved. */
8933 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
8936 /* Ensure the device has been registrered */
8937 if (dev
->reg_state
!= NETREG_REGISTERED
)
8940 /* Get out if there is nothing todo */
8942 if (net_eq(dev_net(dev
), net
))
8945 /* Pick the destination device name, and ensure
8946 * we can use it in the destination network namespace.
8949 if (__dev_get_by_name(net
, dev
->name
)) {
8950 /* We get here if we can't use the current device name */
8953 err
= dev_get_valid_name(net
, dev
, pat
);
8959 * And now a mini version of register_netdevice unregister_netdevice.
8962 /* If device is running close it first. */
8965 /* And unlink it from device chain */
8966 unlist_netdevice(dev
);
8970 /* Shutdown queueing discipline. */
8973 /* Notify protocols, that we are about to destroy
8974 * this device. They should clean all the things.
8976 * Note that dev->reg_state stays at NETREG_REGISTERED.
8977 * This is wanted because this way 8021q and macvlan know
8978 * the device is just moving and can keep their slaves up.
8980 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
8983 new_nsid
= peernet2id_alloc(dev_net(dev
), net
);
8984 /* If there is an ifindex conflict assign a new one */
8985 if (__dev_get_by_index(net
, dev
->ifindex
))
8986 new_ifindex
= dev_new_index(net
);
8988 new_ifindex
= dev
->ifindex
;
8990 rtmsg_ifinfo_newnet(RTM_DELLINK
, dev
, ~0U, GFP_KERNEL
, &new_nsid
,
8994 * Flush the unicast and multicast chains
8999 /* Send a netdev-removed uevent to the old namespace */
9000 kobject_uevent(&dev
->dev
.kobj
, KOBJ_REMOVE
);
9001 netdev_adjacent_del_links(dev
);
9003 /* Actually switch the network namespace */
9004 dev_net_set(dev
, net
);
9005 dev
->ifindex
= new_ifindex
;
9007 /* Send a netdev-add uevent to the new namespace */
9008 kobject_uevent(&dev
->dev
.kobj
, KOBJ_ADD
);
9009 netdev_adjacent_add_links(dev
);
9011 /* Fixup kobjects */
9012 err
= device_rename(&dev
->dev
, dev
->name
);
9015 /* Add the device back in the hashes */
9016 list_netdevice(dev
);
9018 /* Notify protocols, that a new device appeared. */
9019 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
9022 * Prevent userspace races by waiting until the network
9023 * device is fully setup before sending notifications.
9025 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U, GFP_KERNEL
);
9032 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
9034 static int dev_cpu_dead(unsigned int oldcpu
)
9036 struct sk_buff
**list_skb
;
9037 struct sk_buff
*skb
;
9039 struct softnet_data
*sd
, *oldsd
, *remsd
= NULL
;
9041 local_irq_disable();
9042 cpu
= smp_processor_id();
9043 sd
= &per_cpu(softnet_data
, cpu
);
9044 oldsd
= &per_cpu(softnet_data
, oldcpu
);
9046 /* Find end of our completion_queue. */
9047 list_skb
= &sd
->completion_queue
;
9049 list_skb
= &(*list_skb
)->next
;
9050 /* Append completion queue from offline CPU. */
9051 *list_skb
= oldsd
->completion_queue
;
9052 oldsd
->completion_queue
= NULL
;
9054 /* Append output queue from offline CPU. */
9055 if (oldsd
->output_queue
) {
9056 *sd
->output_queue_tailp
= oldsd
->output_queue
;
9057 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
9058 oldsd
->output_queue
= NULL
;
9059 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
9061 /* Append NAPI poll list from offline CPU, with one exception :
9062 * process_backlog() must be called by cpu owning percpu backlog.
9063 * We properly handle process_queue & input_pkt_queue later.
9065 while (!list_empty(&oldsd
->poll_list
)) {
9066 struct napi_struct
*napi
= list_first_entry(&oldsd
->poll_list
,
9070 list_del_init(&napi
->poll_list
);
9071 if (napi
->poll
== process_backlog
)
9074 ____napi_schedule(sd
, napi
);
9077 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
9081 remsd
= oldsd
->rps_ipi_list
;
9082 oldsd
->rps_ipi_list
= NULL
;
9084 /* send out pending IPI's on offline CPU */
9085 net_rps_send_ipi(remsd
);
9087 /* Process offline CPU's input_pkt_queue */
9088 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
9090 input_queue_head_incr(oldsd
);
9092 while ((skb
= skb_dequeue(&oldsd
->input_pkt_queue
))) {
9094 input_queue_head_incr(oldsd
);
9101 * netdev_increment_features - increment feature set by one
9102 * @all: current feature set
9103 * @one: new feature set
9104 * @mask: mask feature set
9106 * Computes a new feature set after adding a device with feature set
9107 * @one to the master device with current feature set @all. Will not
9108 * enable anything that is off in @mask. Returns the new feature set.
9110 netdev_features_t
netdev_increment_features(netdev_features_t all
,
9111 netdev_features_t one
, netdev_features_t mask
)
9113 if (mask
& NETIF_F_HW_CSUM
)
9114 mask
|= NETIF_F_CSUM_MASK
;
9115 mask
|= NETIF_F_VLAN_CHALLENGED
;
9117 all
|= one
& (NETIF_F_ONE_FOR_ALL
| NETIF_F_CSUM_MASK
) & mask
;
9118 all
&= one
| ~NETIF_F_ALL_FOR_ALL
;
9120 /* If one device supports hw checksumming, set for all. */
9121 if (all
& NETIF_F_HW_CSUM
)
9122 all
&= ~(NETIF_F_CSUM_MASK
& ~NETIF_F_HW_CSUM
);
9126 EXPORT_SYMBOL(netdev_increment_features
);
9128 static struct hlist_head
* __net_init
netdev_create_hash(void)
9131 struct hlist_head
*hash
;
9133 hash
= kmalloc_array(NETDEV_HASHENTRIES
, sizeof(*hash
), GFP_KERNEL
);
9135 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
9136 INIT_HLIST_HEAD(&hash
[i
]);
9141 /* Initialize per network namespace state */
9142 static int __net_init
netdev_init(struct net
*net
)
9144 if (net
!= &init_net
)
9145 INIT_LIST_HEAD(&net
->dev_base_head
);
9147 net
->dev_name_head
= netdev_create_hash();
9148 if (net
->dev_name_head
== NULL
)
9151 net
->dev_index_head
= netdev_create_hash();
9152 if (net
->dev_index_head
== NULL
)
9158 kfree(net
->dev_name_head
);
9164 * netdev_drivername - network driver for the device
9165 * @dev: network device
9167 * Determine network driver for device.
9169 const char *netdev_drivername(const struct net_device
*dev
)
9171 const struct device_driver
*driver
;
9172 const struct device
*parent
;
9173 const char *empty
= "";
9175 parent
= dev
->dev
.parent
;
9179 driver
= parent
->driver
;
9180 if (driver
&& driver
->name
)
9181 return driver
->name
;
9185 static void __netdev_printk(const char *level
, const struct net_device
*dev
,
9186 struct va_format
*vaf
)
9188 if (dev
&& dev
->dev
.parent
) {
9189 dev_printk_emit(level
[1] - '0',
9192 dev_driver_string(dev
->dev
.parent
),
9193 dev_name(dev
->dev
.parent
),
9194 netdev_name(dev
), netdev_reg_state(dev
),
9197 printk("%s%s%s: %pV",
9198 level
, netdev_name(dev
), netdev_reg_state(dev
), vaf
);
9200 printk("%s(NULL net_device): %pV", level
, vaf
);
9204 void netdev_printk(const char *level
, const struct net_device
*dev
,
9205 const char *format
, ...)
9207 struct va_format vaf
;
9210 va_start(args
, format
);
9215 __netdev_printk(level
, dev
, &vaf
);
9219 EXPORT_SYMBOL(netdev_printk
);
9221 #define define_netdev_printk_level(func, level) \
9222 void func(const struct net_device *dev, const char *fmt, ...) \
9224 struct va_format vaf; \
9227 va_start(args, fmt); \
9232 __netdev_printk(level, dev, &vaf); \
9236 EXPORT_SYMBOL(func);
9238 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
9239 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
9240 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
9241 define_netdev_printk_level(netdev_err
, KERN_ERR
);
9242 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
9243 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
9244 define_netdev_printk_level(netdev_info
, KERN_INFO
);
9246 static void __net_exit
netdev_exit(struct net
*net
)
9248 kfree(net
->dev_name_head
);
9249 kfree(net
->dev_index_head
);
9250 if (net
!= &init_net
)
9251 WARN_ON_ONCE(!list_empty(&net
->dev_base_head
));
9254 static struct pernet_operations __net_initdata netdev_net_ops
= {
9255 .init
= netdev_init
,
9256 .exit
= netdev_exit
,
9259 static void __net_exit
default_device_exit(struct net
*net
)
9261 struct net_device
*dev
, *aux
;
9263 * Push all migratable network devices back to the
9264 * initial network namespace
9267 for_each_netdev_safe(net
, dev
, aux
) {
9269 char fb_name
[IFNAMSIZ
];
9271 /* Ignore unmoveable devices (i.e. loopback) */
9272 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
9275 /* Leave virtual devices for the generic cleanup */
9276 if (dev
->rtnl_link_ops
)
9279 /* Push remaining network devices to init_net */
9280 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
9281 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
9283 pr_emerg("%s: failed to move %s to init_net: %d\n",
9284 __func__
, dev
->name
, err
);
9291 static void __net_exit
rtnl_lock_unregistering(struct list_head
*net_list
)
9293 /* Return with the rtnl_lock held when there are no network
9294 * devices unregistering in any network namespace in net_list.
9298 DEFINE_WAIT_FUNC(wait
, woken_wake_function
);
9300 add_wait_queue(&netdev_unregistering_wq
, &wait
);
9302 unregistering
= false;
9304 list_for_each_entry(net
, net_list
, exit_list
) {
9305 if (net
->dev_unreg_count
> 0) {
9306 unregistering
= true;
9314 wait_woken(&wait
, TASK_UNINTERRUPTIBLE
, MAX_SCHEDULE_TIMEOUT
);
9316 remove_wait_queue(&netdev_unregistering_wq
, &wait
);
9319 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
9321 /* At exit all network devices most be removed from a network
9322 * namespace. Do this in the reverse order of registration.
9323 * Do this across as many network namespaces as possible to
9324 * improve batching efficiency.
9326 struct net_device
*dev
;
9328 LIST_HEAD(dev_kill_list
);
9330 /* To prevent network device cleanup code from dereferencing
9331 * loopback devices or network devices that have been freed
9332 * wait here for all pending unregistrations to complete,
9333 * before unregistring the loopback device and allowing the
9334 * network namespace be freed.
9336 * The netdev todo list containing all network devices
9337 * unregistrations that happen in default_device_exit_batch
9338 * will run in the rtnl_unlock() at the end of
9339 * default_device_exit_batch.
9341 rtnl_lock_unregistering(net_list
);
9342 list_for_each_entry(net
, net_list
, exit_list
) {
9343 for_each_netdev_reverse(net
, dev
) {
9344 if (dev
->rtnl_link_ops
&& dev
->rtnl_link_ops
->dellink
)
9345 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
9347 unregister_netdevice_queue(dev
, &dev_kill_list
);
9350 unregister_netdevice_many(&dev_kill_list
);
9354 static struct pernet_operations __net_initdata default_device_ops
= {
9355 .exit
= default_device_exit
,
9356 .exit_batch
= default_device_exit_batch
,
9360 * Initialize the DEV module. At boot time this walks the device list and
9361 * unhooks any devices that fail to initialise (normally hardware not
9362 * present) and leaves us with a valid list of present and active devices.
9367 * This is called single threaded during boot, so no need
9368 * to take the rtnl semaphore.
9370 static int __init
net_dev_init(void)
9372 int i
, rc
= -ENOMEM
;
9374 BUG_ON(!dev_boot_phase
);
9376 if (dev_proc_init())
9379 if (netdev_kobject_init())
9382 INIT_LIST_HEAD(&ptype_all
);
9383 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
9384 INIT_LIST_HEAD(&ptype_base
[i
]);
9386 INIT_LIST_HEAD(&offload_base
);
9388 if (register_pernet_subsys(&netdev_net_ops
))
9392 * Initialise the packet receive queues.
9395 for_each_possible_cpu(i
) {
9396 struct work_struct
*flush
= per_cpu_ptr(&flush_works
, i
);
9397 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
9399 INIT_WORK(flush
, flush_backlog
);
9401 skb_queue_head_init(&sd
->input_pkt_queue
);
9402 skb_queue_head_init(&sd
->process_queue
);
9403 #ifdef CONFIG_XFRM_OFFLOAD
9404 skb_queue_head_init(&sd
->xfrm_backlog
);
9406 INIT_LIST_HEAD(&sd
->poll_list
);
9407 sd
->output_queue_tailp
= &sd
->output_queue
;
9409 sd
->csd
.func
= rps_trigger_softirq
;
9414 sd
->backlog
.poll
= process_backlog
;
9415 sd
->backlog
.weight
= weight_p
;
9420 /* The loopback device is special if any other network devices
9421 * is present in a network namespace the loopback device must
9422 * be present. Since we now dynamically allocate and free the
9423 * loopback device ensure this invariant is maintained by
9424 * keeping the loopback device as the first device on the
9425 * list of network devices. Ensuring the loopback devices
9426 * is the first device that appears and the last network device
9429 if (register_pernet_device(&loopback_net_ops
))
9432 if (register_pernet_device(&default_device_ops
))
9435 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
9436 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
9438 rc
= cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD
, "net/dev:dead",
9439 NULL
, dev_cpu_dead
);
9446 subsys_initcall(net_dev_init
);