2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <linux/bpf.h>
98 #include <net/net_namespace.h>
100 #include <net/busy_poll.h>
101 #include <linux/rtnetlink.h>
102 #include <linux/stat.h>
104 #include <net/dst_metadata.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include <linux/highmem.h>
109 #include <linux/init.h>
110 #include <linux/module.h>
111 #include <linux/netpoll.h>
112 #include <linux/rcupdate.h>
113 #include <linux/delay.h>
114 #include <net/iw_handler.h>
115 #include <asm/current.h>
116 #include <linux/audit.h>
117 #include <linux/dmaengine.h>
118 #include <linux/err.h>
119 #include <linux/ctype.h>
120 #include <linux/if_arp.h>
121 #include <linux/if_vlan.h>
122 #include <linux/ip.h>
124 #include <net/mpls.h>
125 #include <linux/ipv6.h>
126 #include <linux/in.h>
127 #include <linux/jhash.h>
128 #include <linux/random.h>
129 #include <trace/events/napi.h>
130 #include <trace/events/net.h>
131 #include <trace/events/skb.h>
132 #include <linux/pci.h>
133 #include <linux/inetdevice.h>
134 #include <linux/cpu_rmap.h>
135 #include <linux/static_key.h>
136 #include <linux/hashtable.h>
137 #include <linux/vmalloc.h>
138 #include <linux/if_macvlan.h>
139 #include <linux/errqueue.h>
140 #include <linux/hrtimer.h>
141 #include <linux/netfilter_ingress.h>
142 #include <linux/crash_dump.h>
144 #include "net-sysfs.h"
146 /* Instead of increasing this, you should create a hash table. */
147 #define MAX_GRO_SKBS 8
149 /* This should be increased if a protocol with a bigger head is added. */
150 #define GRO_MAX_HEAD (MAX_HEADER + 128)
152 static DEFINE_SPINLOCK(ptype_lock);
153 static DEFINE_SPINLOCK(offload_lock);
154 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
155 struct list_head ptype_all __read_mostly; /* Taps */
156 static struct list_head offload_base __read_mostly;
158 static int netif_rx_internal(struct sk_buff *skb);
159 static int call_netdevice_notifiers_info(unsigned long val,
160 struct net_device *dev,
161 struct netdev_notifier_info *info);
164 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
167 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
169 * Writers must hold the rtnl semaphore while they loop through the
170 * dev_base_head list, and hold dev_base_lock for writing when they do the
171 * actual updates. This allows pure readers to access the list even
172 * while a writer is preparing to update it.
174 * To put it another way, dev_base_lock is held for writing only to
175 * protect against pure readers; the rtnl semaphore provides the
176 * protection against other writers.
178 * See, for example usages, register_netdevice() and
179 * unregister_netdevice(), which must be called with the rtnl
182 DEFINE_RWLOCK(dev_base_lock);
183 EXPORT_SYMBOL(dev_base_lock);
185 /* protects napi_hash addition/deletion and napi_gen_id */
186 static DEFINE_SPINLOCK(napi_hash_lock);
188 static unsigned int napi_gen_id = NR_CPUS;
189 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
191 static seqcount_t devnet_rename_seq;
193 static inline void dev_base_seq_inc(struct net *net)
195 while (++net->dev_base_seq == 0);
198 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
200 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
202 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
205 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
207 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
210 static inline void rps_lock(struct softnet_data *sd)
213 spin_lock(&sd->input_pkt_queue.lock);
217 static inline void rps_unlock(struct softnet_data *sd)
220 spin_unlock(&sd->input_pkt_queue.lock);
224 /* Device list insertion */
225 static void list_netdevice(struct net_device *dev)
227 struct net *net = dev_net(dev);
231 write_lock_bh(&dev_base_lock);
232 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
233 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
234 hlist_add_head_rcu(&dev->index_hlist,
235 dev_index_hash(net, dev->ifindex));
236 write_unlock_bh(&dev_base_lock);
238 dev_base_seq_inc(net);
241 /* Device list removal
242 * caller must respect a RCU grace period before freeing/reusing dev
244 static void unlist_netdevice(struct net_device *dev)
248 /* Unlink dev from the device chain */
249 write_lock_bh(&dev_base_lock);
250 list_del_rcu(&dev->dev_list);
251 hlist_del_rcu(&dev->name_hlist);
252 hlist_del_rcu(&dev->index_hlist);
253 write_unlock_bh(&dev_base_lock);
255 dev_base_seq_inc(dev_net(dev));
262 static RAW_NOTIFIER_HEAD(netdev_chain);
265 * Device drivers call our routines to queue packets here. We empty the
266 * queue in the local softnet handler.
269 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
270 EXPORT_PER_CPU_SYMBOL(softnet_data);
272 #ifdef CONFIG_LOCKDEP
274 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
275 * according to dev->type
277 static const unsigned short netdev_lock_type[] =
278 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
279 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
280 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
281 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
282 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
283 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
284 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
285 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
286 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
287 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
288 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
289 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
290 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
291 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
292 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
294 static const char *const netdev_lock_name[] =
295 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
296 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
297 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
298 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
299 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
300 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
301 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
302 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
303 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
304 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
305 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
306 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
307 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
308 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
309 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
311 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
312 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
314 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
318 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
319 if (netdev_lock_type[i] == dev_type)
321 /* the last key is used by default */
322 return ARRAY_SIZE(netdev_lock_type) - 1;
325 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
326 unsigned short dev_type)
330 i = netdev_lock_pos(dev_type);
331 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
332 netdev_lock_name[i]);
335 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
339 i = netdev_lock_pos(dev->type);
340 lockdep_set_class_and_name(&dev->addr_list_lock,
341 &netdev_addr_lock_key[i],
342 netdev_lock_name[i]);
345 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
346 unsigned short dev_type)
349 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
354 /*******************************************************************************
356 Protocol management and registration routines
358 *******************************************************************************/
361 * Add a protocol ID to the list. Now that the input handler is
362 * smarter we can dispense with all the messy stuff that used to be
365 * BEWARE!!! Protocol handlers, mangling input packets,
366 * MUST BE last in hash buckets and checking protocol handlers
367 * MUST start from promiscuous ptype_all chain in net_bh.
368 * It is true now, do not change it.
369 * Explanation follows: if protocol handler, mangling packet, will
370 * be the first on list, it is not able to sense, that packet
371 * is cloned and should be copied-on-write, so that it will
372 * change it and subsequent readers will get broken packet.
376 static inline struct list_head *ptype_head(const struct packet_type *pt)
378 if (pt->type == htons(ETH_P_ALL))
379 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
381 return pt->dev ? &pt->dev->ptype_specific :
382 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
386 * dev_add_pack - add packet handler
387 * @pt: packet type declaration
389 * Add a protocol handler to the networking stack. The passed &packet_type
390 * is linked into kernel lists and may not be freed until it has been
391 * removed from the kernel lists.
393 * This call does not sleep therefore it can not
394 * guarantee all CPU's that are in middle of receiving packets
395 * will see the new packet type (until the next received packet).
398 void dev_add_pack(struct packet_type *pt)
400 struct list_head *head = ptype_head(pt);
402 spin_lock(&ptype_lock);
403 list_add_rcu(&pt->list, head);
404 spin_unlock(&ptype_lock);
406 EXPORT_SYMBOL(dev_add_pack);
409 * __dev_remove_pack - remove packet handler
410 * @pt: packet type declaration
412 * Remove a protocol handler that was previously added to the kernel
413 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
414 * from the kernel lists and can be freed or reused once this function
417 * The packet type might still be in use by receivers
418 * and must not be freed until after all the CPU's have gone
419 * through a quiescent state.
421 void __dev_remove_pack(struct packet_type *pt)
423 struct list_head *head = ptype_head(pt);
424 struct packet_type *pt1;
426 spin_lock(&ptype_lock);
428 list_for_each_entry(pt1, head, list) {
430 list_del_rcu(&pt->list);
435 pr_warn("dev_remove_pack: %p not found\n", pt);
437 spin_unlock(&ptype_lock);
439 EXPORT_SYMBOL(__dev_remove_pack);
442 * dev_remove_pack - remove packet handler
443 * @pt: packet type declaration
445 * Remove a protocol handler that was previously added to the kernel
446 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
447 * from the kernel lists and can be freed or reused once this function
450 * This call sleeps to guarantee that no CPU is looking at the packet
453 void dev_remove_pack(struct packet_type *pt)
455 __dev_remove_pack(pt);
459 EXPORT_SYMBOL(dev_remove_pack);
463 * dev_add_offload - register offload handlers
464 * @po: protocol offload declaration
466 * Add protocol offload handlers to the networking stack. The passed
467 * &proto_offload is linked into kernel lists and may not be freed until
468 * it has been removed from the kernel lists.
470 * This call does not sleep therefore it can not
471 * guarantee all CPU's that are in middle of receiving packets
472 * will see the new offload handlers (until the next received packet).
474 void dev_add_offload(struct packet_offload *po)
476 struct packet_offload *elem;
478 spin_lock(&offload_lock);
479 list_for_each_entry(elem, &offload_base, list) {
480 if (po->priority < elem->priority)
483 list_add_rcu(&po->list, elem->list.prev);
484 spin_unlock(&offload_lock);
486 EXPORT_SYMBOL(dev_add_offload);
489 * __dev_remove_offload - remove offload handler
490 * @po: packet offload declaration
492 * Remove a protocol offload handler that was previously added to the
493 * kernel offload handlers by dev_add_offload(). The passed &offload_type
494 * is removed from the kernel lists and can be freed or reused once this
497 * The packet type might still be in use by receivers
498 * and must not be freed until after all the CPU's have gone
499 * through a quiescent state.
501 static void __dev_remove_offload(struct packet_offload *po)
503 struct list_head *head = &offload_base;
504 struct packet_offload *po1;
506 spin_lock(&offload_lock);
508 list_for_each_entry(po1, head, list) {
510 list_del_rcu(&po->list);
515 pr_warn("dev_remove_offload: %p not found\n", po);
517 spin_unlock(&offload_lock);
521 * dev_remove_offload - remove packet offload handler
522 * @po: packet offload declaration
524 * Remove a packet offload handler that was previously added to the kernel
525 * offload handlers by dev_add_offload(). The passed &offload_type is
526 * removed from the kernel lists and can be freed or reused once this
529 * This call sleeps to guarantee that no CPU is looking at the packet
532 void dev_remove_offload(struct packet_offload *po)
534 __dev_remove_offload(po);
538 EXPORT_SYMBOL(dev_remove_offload);
540 /******************************************************************************
542 Device Boot-time Settings Routines
544 *******************************************************************************/
546 /* Boot time configuration table */
547 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
550 * netdev_boot_setup_add - add new setup entry
551 * @name: name of the device
552 * @map: configured settings for the device
554 * Adds new setup entry to the dev_boot_setup list. The function
555 * returns 0 on error and 1 on success. This is a generic routine to
558 static int netdev_boot_setup_add(char *name, struct ifmap *map)
560 struct netdev_boot_setup *s;
564 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
565 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
566 memset(s[i].name, 0, sizeof(s[i].name));
567 strlcpy(s[i].name, name, IFNAMSIZ);
568 memcpy(&s[i].map, map, sizeof(s[i].map));
573 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
577 * netdev_boot_setup_check - check boot time settings
578 * @dev: the netdevice
580 * Check boot time settings for the device.
581 * The found settings are set for the device to be used
582 * later in the device probing.
583 * Returns 0 if no settings found, 1 if they are.
585 int netdev_boot_setup_check(struct net_device *dev)
587 struct netdev_boot_setup *s = dev_boot_setup;
590 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
591 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
592 !strcmp(dev->name, s[i].name)) {
593 dev->irq = s[i].map.irq;
594 dev->base_addr = s[i].map.base_addr;
595 dev->mem_start = s[i].map.mem_start;
596 dev->mem_end = s[i].map.mem_end;
602 EXPORT_SYMBOL(netdev_boot_setup_check);
606 * netdev_boot_base - get address from boot time settings
607 * @prefix: prefix for network device
608 * @unit: id for network device
610 * Check boot time settings for the base address of device.
611 * The found settings are set for the device to be used
612 * later in the device probing.
613 * Returns 0 if no settings found.
615 unsigned long netdev_boot_base(const char *prefix, int unit)
617 const struct netdev_boot_setup *s = dev_boot_setup;
621 sprintf(name, "%s%d", prefix, unit);
624 * If device already registered then return base of 1
625 * to indicate not to probe for this interface
627 if (__dev_get_by_name(&init_net, name))
630 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
631 if (!strcmp(name, s[i].name))
632 return s[i].map.base_addr;
637 * Saves at boot time configured settings for any netdevice.
639 int __init netdev_boot_setup(char *str)
644 str = get_options(str, ARRAY_SIZE(ints), ints);
649 memset(&map, 0, sizeof(map));
653 map.base_addr = ints[2];
655 map.mem_start = ints[3];
657 map.mem_end = ints[4];
659 /* Add new entry to the list */
660 return netdev_boot_setup_add(str, &map);
663 __setup("netdev=", netdev_boot_setup);
665 /*******************************************************************************
667 Device Interface Subroutines
669 *******************************************************************************/
672 * dev_get_iflink - get 'iflink' value of a interface
673 * @dev: targeted interface
675 * Indicates the ifindex the interface is linked to.
676 * Physical interfaces have the same 'ifindex' and 'iflink' values.
679 int dev_get_iflink(const struct net_device *dev)
681 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
682 return dev->netdev_ops->ndo_get_iflink(dev);
686 EXPORT_SYMBOL(dev_get_iflink);
689 * dev_fill_metadata_dst - Retrieve tunnel egress information.
690 * @dev: targeted interface
693 * For better visibility of tunnel traffic OVS needs to retrieve
694 * egress tunnel information for a packet. Following API allows
695 * user to get this info.
697 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
699 struct ip_tunnel_info *info;
701 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
704 info = skb_tunnel_info_unclone(skb);
707 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
710 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
712 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
715 * __dev_get_by_name - find a device by its name
716 * @net: the applicable net namespace
717 * @name: name to find
719 * Find an interface by name. Must be called under RTNL semaphore
720 * or @dev_base_lock. If the name is found a pointer to the device
721 * is returned. If the name is not found then %NULL is returned. The
722 * reference counters are not incremented so the caller must be
723 * careful with locks.
726 struct net_device *__dev_get_by_name(struct net *net, const char *name)
728 struct net_device *dev;
729 struct hlist_head *head = dev_name_hash(net, name);
731 hlist_for_each_entry(dev, head, name_hlist)
732 if (!strncmp(dev->name, name, IFNAMSIZ))
737 EXPORT_SYMBOL(__dev_get_by_name);
740 * dev_get_by_name_rcu - find a device by its name
741 * @net: the applicable net namespace
742 * @name: name to find
744 * Find an interface by name.
745 * If the name is found a pointer to the device is returned.
746 * If the name is not found then %NULL is returned.
747 * The reference counters are not incremented so the caller must be
748 * careful with locks. The caller must hold RCU lock.
751 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
753 struct net_device *dev;
754 struct hlist_head *head = dev_name_hash(net, name);
756 hlist_for_each_entry_rcu(dev, head, name_hlist)
757 if (!strncmp(dev->name, name, IFNAMSIZ))
762 EXPORT_SYMBOL(dev_get_by_name_rcu);
765 * dev_get_by_name - find a device by its name
766 * @net: the applicable net namespace
767 * @name: name to find
769 * Find an interface by name. This can be called from any
770 * context and does its own locking. The returned handle has
771 * the usage count incremented and the caller must use dev_put() to
772 * release it when it is no longer needed. %NULL is returned if no
773 * matching device is found.
776 struct net_device *dev_get_by_name(struct net *net, const char *name)
778 struct net_device *dev;
781 dev = dev_get_by_name_rcu(net, name);
787 EXPORT_SYMBOL(dev_get_by_name);
790 * __dev_get_by_index - find a device by its ifindex
791 * @net: the applicable net namespace
792 * @ifindex: index of device
794 * Search for an interface by index. Returns %NULL if the device
795 * is not found or a pointer to the device. The device has not
796 * had its reference counter increased so the caller must be careful
797 * about locking. The caller must hold either the RTNL semaphore
801 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
803 struct net_device *dev;
804 struct hlist_head *head = dev_index_hash(net, ifindex);
806 hlist_for_each_entry(dev, head, index_hlist)
807 if (dev->ifindex == ifindex)
812 EXPORT_SYMBOL(__dev_get_by_index);
815 * dev_get_by_index_rcu - find a device by its ifindex
816 * @net: the applicable net namespace
817 * @ifindex: index of device
819 * Search for an interface by index. Returns %NULL if the device
820 * is not found or a pointer to the device. The device has not
821 * had its reference counter increased so the caller must be careful
822 * about locking. The caller must hold RCU lock.
825 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
827 struct net_device *dev;
828 struct hlist_head *head = dev_index_hash(net, ifindex);
830 hlist_for_each_entry_rcu(dev, head, index_hlist)
831 if (dev->ifindex == ifindex)
836 EXPORT_SYMBOL(dev_get_by_index_rcu);
840 * dev_get_by_index - find a device by its ifindex
841 * @net: the applicable net namespace
842 * @ifindex: index of device
844 * Search for an interface by index. Returns NULL if the device
845 * is not found or a pointer to the device. The device returned has
846 * had a reference added and the pointer is safe until the user calls
847 * dev_put to indicate they have finished with it.
850 struct net_device *dev_get_by_index(struct net *net, int ifindex)
852 struct net_device *dev;
855 dev = dev_get_by_index_rcu(net, ifindex);
861 EXPORT_SYMBOL(dev_get_by_index);
864 * netdev_get_name - get a netdevice name, knowing its ifindex.
865 * @net: network namespace
866 * @name: a pointer to the buffer where the name will be stored.
867 * @ifindex: the ifindex of the interface to get the name from.
869 * The use of raw_seqcount_begin() and cond_resched() before
870 * retrying is required as we want to give the writers a chance
871 * to complete when CONFIG_PREEMPT is not set.
873 int netdev_get_name(struct net *net, char *name, int ifindex)
875 struct net_device *dev;
879 seq = raw_seqcount_begin(&devnet_rename_seq);
881 dev = dev_get_by_index_rcu(net, ifindex);
887 strcpy(name, dev->name);
889 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
898 * dev_getbyhwaddr_rcu - find a device by its hardware address
899 * @net: the applicable net namespace
900 * @type: media type of device
901 * @ha: hardware address
903 * Search for an interface by MAC address. Returns NULL if the device
904 * is not found or a pointer to the device.
905 * The caller must hold RCU or RTNL.
906 * The returned device has not had its ref count increased
907 * and the caller must therefore be careful about locking
911 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
914 struct net_device *dev;
916 for_each_netdev_rcu(net, dev)
917 if (dev->type == type &&
918 !memcmp(dev->dev_addr, ha, dev->addr_len))
923 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
925 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
927 struct net_device *dev;
930 for_each_netdev(net, dev)
931 if (dev->type == type)
936 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
938 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
940 struct net_device *dev, *ret = NULL;
943 for_each_netdev_rcu(net, dev)
944 if (dev->type == type) {
952 EXPORT_SYMBOL(dev_getfirstbyhwtype);
955 * __dev_get_by_flags - find any device with given flags
956 * @net: the applicable net namespace
957 * @if_flags: IFF_* values
958 * @mask: bitmask of bits in if_flags to check
960 * Search for any interface with the given flags. Returns NULL if a device
961 * is not found or a pointer to the device. Must be called inside
962 * rtnl_lock(), and result refcount is unchanged.
965 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
968 struct net_device *dev, *ret;
973 for_each_netdev(net, dev) {
974 if (((dev->flags ^ if_flags) & mask) == 0) {
981 EXPORT_SYMBOL(__dev_get_by_flags);
984 * dev_valid_name - check if name is okay for network device
987 * Network device names need to be valid file names to
988 * to allow sysfs to work. We also disallow any kind of
991 bool dev_valid_name(const char *name)
995 if (strlen(name) >= IFNAMSIZ)
997 if (!strcmp(name, ".") || !strcmp(name, ".."))
1001 if (*name == '/' || *name == ':' || isspace(*name))
1007 EXPORT_SYMBOL(dev_valid_name);
1010 * __dev_alloc_name - allocate a name for a device
1011 * @net: network namespace to allocate the device name in
1012 * @name: name format string
1013 * @buf: scratch buffer and result name string
1015 * Passed a format string - eg "lt%d" it will try and find a suitable
1016 * id. It scans list of devices to build up a free map, then chooses
1017 * the first empty slot. The caller must hold the dev_base or rtnl lock
1018 * while allocating the name and adding the device in order to avoid
1020 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1021 * Returns the number of the unit assigned or a negative errno code.
1024 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1028 const int max_netdevices = 8*PAGE_SIZE;
1029 unsigned long *inuse;
1030 struct net_device *d;
1032 p = strnchr(name, IFNAMSIZ-1, '%');
1035 * Verify the string as this thing may have come from
1036 * the user. There must be either one "%d" and no other "%"
1039 if (p[1] != 'd' || strchr(p + 2, '%'))
1042 /* Use one page as a bit array of possible slots */
1043 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1047 for_each_netdev(net, d) {
1048 if (!sscanf(d->name, name, &i))
1050 if (i < 0 || i >= max_netdevices)
1053 /* avoid cases where sscanf is not exact inverse of printf */
1054 snprintf(buf, IFNAMSIZ, name, i);
1055 if (!strncmp(buf, d->name, IFNAMSIZ))
1059 i = find_first_zero_bit(inuse, max_netdevices);
1060 free_page((unsigned long) inuse);
1064 snprintf(buf, IFNAMSIZ, name, i);
1065 if (!__dev_get_by_name(net, buf))
1068 /* It is possible to run out of possible slots
1069 * when the name is long and there isn't enough space left
1070 * for the digits, or if all bits are used.
1076 * dev_alloc_name - allocate a name for a device
1078 * @name: name format string
1080 * Passed a format string - eg "lt%d" it will try and find a suitable
1081 * id. It scans list of devices to build up a free map, then chooses
1082 * the first empty slot. The caller must hold the dev_base or rtnl lock
1083 * while allocating the name and adding the device in order to avoid
1085 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1086 * Returns the number of the unit assigned or a negative errno code.
1089 int dev_alloc_name(struct net_device *dev, const char *name)
1095 BUG_ON(!dev_net(dev));
1097 ret = __dev_alloc_name(net, name, buf);
1099 strlcpy(dev->name, buf, IFNAMSIZ);
1102 EXPORT_SYMBOL(dev_alloc_name);
1104 static int dev_alloc_name_ns(struct net *net,
1105 struct net_device *dev,
1111 ret = __dev_alloc_name(net, name, buf);
1113 strlcpy(dev->name, buf, IFNAMSIZ);
1117 static int dev_get_valid_name(struct net *net,
1118 struct net_device *dev,
1123 if (!dev_valid_name(name))
1126 if (strchr(name, '%'))
1127 return dev_alloc_name_ns(net, dev, name);
1128 else if (__dev_get_by_name(net, name))
1130 else if (dev->name != name)
1131 strlcpy(dev->name, name, IFNAMSIZ);
1137 * dev_change_name - change name of a device
1139 * @newname: name (or format string) must be at least IFNAMSIZ
1141 * Change name of a device, can pass format strings "eth%d".
1144 int dev_change_name(struct net_device *dev, const char *newname)
1146 unsigned char old_assign_type;
1147 char oldname[IFNAMSIZ];
1153 BUG_ON(!dev_net(dev));
1156 if (dev->flags & IFF_UP)
1159 write_seqcount_begin(&devnet_rename_seq);
1161 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1162 write_seqcount_end(&devnet_rename_seq);
1166 memcpy(oldname, dev->name, IFNAMSIZ);
1168 err = dev_get_valid_name(net, dev, newname);
1170 write_seqcount_end(&devnet_rename_seq);
1174 if (oldname[0] && !strchr(oldname, '%'))
1175 netdev_info(dev, "renamed from %s\n", oldname);
1177 old_assign_type = dev->name_assign_type;
1178 dev->name_assign_type = NET_NAME_RENAMED;
1181 ret = device_rename(&dev->dev, dev->name);
1183 memcpy(dev->name, oldname, IFNAMSIZ);
1184 dev->name_assign_type = old_assign_type;
1185 write_seqcount_end(&devnet_rename_seq);
1189 write_seqcount_end(&devnet_rename_seq);
1191 netdev_adjacent_rename_links(dev, oldname);
1193 write_lock_bh(&dev_base_lock);
1194 hlist_del_rcu(&dev->name_hlist);
1195 write_unlock_bh(&dev_base_lock);
1199 write_lock_bh(&dev_base_lock);
1200 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1201 write_unlock_bh(&dev_base_lock);
1203 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1204 ret = notifier_to_errno(ret);
1207 /* err >= 0 after dev_alloc_name() or stores the first errno */
1210 write_seqcount_begin(&devnet_rename_seq);
1211 memcpy(dev->name, oldname, IFNAMSIZ);
1212 memcpy(oldname, newname, IFNAMSIZ);
1213 dev->name_assign_type = old_assign_type;
1214 old_assign_type = NET_NAME_RENAMED;
1217 pr_err("%s: name change rollback failed: %d\n",
1226 * dev_set_alias - change ifalias of a device
1228 * @alias: name up to IFALIASZ
1229 * @len: limit of bytes to copy from info
1231 * Set ifalias for a device,
1233 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1239 if (len >= IFALIASZ)
1243 kfree(dev->ifalias);
1244 dev->ifalias = NULL;
1248 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1251 dev->ifalias = new_ifalias;
1253 strlcpy(dev->ifalias, alias, len+1);
1259 * netdev_features_change - device changes features
1260 * @dev: device to cause notification
1262 * Called to indicate a device has changed features.
1264 void netdev_features_change(struct net_device *dev)
1266 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1268 EXPORT_SYMBOL(netdev_features_change);
1271 * netdev_state_change - device changes state
1272 * @dev: device to cause notification
1274 * Called to indicate a device has changed state. This function calls
1275 * the notifier chains for netdev_chain and sends a NEWLINK message
1276 * to the routing socket.
1278 void netdev_state_change(struct net_device *dev)
1280 if (dev->flags & IFF_UP) {
1281 struct netdev_notifier_change_info change_info;
1283 change_info.flags_changed = 0;
1284 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
1286 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1289 EXPORT_SYMBOL(netdev_state_change);
1292 * netdev_notify_peers - notify network peers about existence of @dev
1293 * @dev: network device
1295 * Generate traffic such that interested network peers are aware of
1296 * @dev, such as by generating a gratuitous ARP. This may be used when
1297 * a device wants to inform the rest of the network about some sort of
1298 * reconfiguration such as a failover event or virtual machine
1301 void netdev_notify_peers(struct net_device *dev)
1304 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1307 EXPORT_SYMBOL(netdev_notify_peers);
1309 static int __dev_open(struct net_device *dev)
1311 const struct net_device_ops *ops = dev->netdev_ops;
1316 if (!netif_device_present(dev))
1319 /* Block netpoll from trying to do any rx path servicing.
1320 * If we don't do this there is a chance ndo_poll_controller
1321 * or ndo_poll may be running while we open the device
1323 netpoll_poll_disable(dev);
1325 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1326 ret = notifier_to_errno(ret);
1330 set_bit(__LINK_STATE_START, &dev->state);
1332 if (ops->ndo_validate_addr)
1333 ret = ops->ndo_validate_addr(dev);
1335 if (!ret && ops->ndo_open)
1336 ret = ops->ndo_open(dev);
1338 netpoll_poll_enable(dev);
1341 clear_bit(__LINK_STATE_START, &dev->state);
1343 dev->flags |= IFF_UP;
1344 dev_set_rx_mode(dev);
1346 add_device_randomness(dev->dev_addr, dev->addr_len);
1353 * dev_open - prepare an interface for use.
1354 * @dev: device to open
1356 * Takes a device from down to up state. The device's private open
1357 * function is invoked and then the multicast lists are loaded. Finally
1358 * the device is moved into the up state and a %NETDEV_UP message is
1359 * sent to the netdev notifier chain.
1361 * Calling this function on an active interface is a nop. On a failure
1362 * a negative errno code is returned.
1364 int dev_open(struct net_device *dev)
1368 if (dev->flags & IFF_UP)
1371 ret = __dev_open(dev);
1375 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1376 call_netdevice_notifiers(NETDEV_UP, dev);
1380 EXPORT_SYMBOL(dev_open);
1382 static int __dev_close_many(struct list_head *head)
1384 struct net_device *dev;
1389 list_for_each_entry(dev, head, close_list) {
1390 /* Temporarily disable netpoll until the interface is down */
1391 netpoll_poll_disable(dev);
1393 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1395 clear_bit(__LINK_STATE_START, &dev->state);
1397 /* Synchronize to scheduled poll. We cannot touch poll list, it
1398 * can be even on different cpu. So just clear netif_running().
1400 * dev->stop() will invoke napi_disable() on all of it's
1401 * napi_struct instances on this device.
1403 smp_mb__after_atomic(); /* Commit netif_running(). */
1406 dev_deactivate_many(head);
1408 list_for_each_entry(dev, head, close_list) {
1409 const struct net_device_ops *ops = dev->netdev_ops;
1412 * Call the device specific close. This cannot fail.
1413 * Only if device is UP
1415 * We allow it to be called even after a DETACH hot-plug
1421 dev->flags &= ~IFF_UP;
1422 netpoll_poll_enable(dev);
1428 static int __dev_close(struct net_device *dev)
1433 list_add(&dev->close_list, &single);
1434 retval = __dev_close_many(&single);
1440 int dev_close_many(struct list_head *head, bool unlink)
1442 struct net_device *dev, *tmp;
1444 /* Remove the devices that don't need to be closed */
1445 list_for_each_entry_safe(dev, tmp, head, close_list)
1446 if (!(dev->flags & IFF_UP))
1447 list_del_init(&dev->close_list);
1449 __dev_close_many(head);
1451 list_for_each_entry_safe(dev, tmp, head, close_list) {
1452 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1453 call_netdevice_notifiers(NETDEV_DOWN, dev);
1455 list_del_init(&dev->close_list);
1460 EXPORT_SYMBOL(dev_close_many);
1463 * dev_close - shutdown an interface.
1464 * @dev: device to shutdown
1466 * This function moves an active device into down state. A
1467 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1468 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1471 int dev_close(struct net_device *dev)
1473 if (dev->flags & IFF_UP) {
1476 list_add(&dev->close_list, &single);
1477 dev_close_many(&single, true);
1482 EXPORT_SYMBOL(dev_close);
1486 * dev_disable_lro - disable Large Receive Offload on a device
1489 * Disable Large Receive Offload (LRO) on a net device. Must be
1490 * called under RTNL. This is needed if received packets may be
1491 * forwarded to another interface.
1493 void dev_disable_lro(struct net_device *dev)
1495 struct net_device *lower_dev;
1496 struct list_head *iter;
1498 dev->wanted_features &= ~NETIF_F_LRO;
1499 netdev_update_features(dev);
1501 if (unlikely(dev->features & NETIF_F_LRO))
1502 netdev_WARN(dev, "failed to disable LRO!\n");
1504 netdev_for_each_lower_dev(dev, lower_dev, iter)
1505 dev_disable_lro(lower_dev);
1507 EXPORT_SYMBOL(dev_disable_lro);
1509 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1510 struct net_device *dev)
1512 struct netdev_notifier_info info;
1514 netdev_notifier_info_init(&info, dev);
1515 return nb->notifier_call(nb, val, &info);
1518 static int dev_boot_phase = 1;
1521 * register_netdevice_notifier - register a network notifier block
1524 * Register a notifier to be called when network device events occur.
1525 * The notifier passed is linked into the kernel structures and must
1526 * not be reused until it has been unregistered. A negative errno code
1527 * is returned on a failure.
1529 * When registered all registration and up events are replayed
1530 * to the new notifier to allow device to have a race free
1531 * view of the network device list.
1534 int register_netdevice_notifier(struct notifier_block *nb)
1536 struct net_device *dev;
1537 struct net_device *last;
1542 err = raw_notifier_chain_register(&netdev_chain, nb);
1548 for_each_netdev(net, dev) {
1549 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1550 err = notifier_to_errno(err);
1554 if (!(dev->flags & IFF_UP))
1557 call_netdevice_notifier(nb, NETDEV_UP, dev);
1568 for_each_netdev(net, dev) {
1572 if (dev->flags & IFF_UP) {
1573 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1575 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1577 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1582 raw_notifier_chain_unregister(&netdev_chain, nb);
1585 EXPORT_SYMBOL(register_netdevice_notifier);
1588 * unregister_netdevice_notifier - unregister a network notifier block
1591 * Unregister a notifier previously registered by
1592 * register_netdevice_notifier(). The notifier is unlinked into the
1593 * kernel structures and may then be reused. A negative errno code
1594 * is returned on a failure.
1596 * After unregistering unregister and down device events are synthesized
1597 * for all devices on the device list to the removed notifier to remove
1598 * the need for special case cleanup code.
1601 int unregister_netdevice_notifier(struct notifier_block *nb)
1603 struct net_device *dev;
1608 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1613 for_each_netdev(net, dev) {
1614 if (dev->flags & IFF_UP) {
1615 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1617 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1619 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1626 EXPORT_SYMBOL(unregister_netdevice_notifier);
1629 * call_netdevice_notifiers_info - call all network notifier blocks
1630 * @val: value passed unmodified to notifier function
1631 * @dev: net_device pointer passed unmodified to notifier function
1632 * @info: notifier information data
1634 * Call all network notifier blocks. Parameters and return value
1635 * are as for raw_notifier_call_chain().
1638 static int call_netdevice_notifiers_info(unsigned long val,
1639 struct net_device *dev,
1640 struct netdev_notifier_info *info)
1643 netdev_notifier_info_init(info, dev);
1644 return raw_notifier_call_chain(&netdev_chain, val, info);
1648 * call_netdevice_notifiers - call all network notifier blocks
1649 * @val: value passed unmodified to notifier function
1650 * @dev: net_device pointer passed unmodified to notifier function
1652 * Call all network notifier blocks. Parameters and return value
1653 * are as for raw_notifier_call_chain().
1656 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1658 struct netdev_notifier_info info;
1660 return call_netdevice_notifiers_info(val, dev, &info);
1662 EXPORT_SYMBOL(call_netdevice_notifiers);
1664 #ifdef CONFIG_NET_INGRESS
1665 static struct static_key ingress_needed __read_mostly;
1667 void net_inc_ingress_queue(void)
1669 static_key_slow_inc(&ingress_needed);
1671 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
1673 void net_dec_ingress_queue(void)
1675 static_key_slow_dec(&ingress_needed);
1677 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
1680 #ifdef CONFIG_NET_EGRESS
1681 static struct static_key egress_needed __read_mostly;
1683 void net_inc_egress_queue(void)
1685 static_key_slow_inc(&egress_needed);
1687 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
1689 void net_dec_egress_queue(void)
1691 static_key_slow_dec(&egress_needed);
1693 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
1696 static struct static_key netstamp_needed __read_mostly;
1697 #ifdef HAVE_JUMP_LABEL
1698 /* We are not allowed to call static_key_slow_dec() from irq context
1699 * If net_disable_timestamp() is called from irq context, defer the
1700 * static_key_slow_dec() calls.
1702 static atomic_t netstamp_needed_deferred;
1705 void net_enable_timestamp(void)
1707 #ifdef HAVE_JUMP_LABEL
1708 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1712 static_key_slow_dec(&netstamp_needed);
1716 static_key_slow_inc(&netstamp_needed);
1718 EXPORT_SYMBOL(net_enable_timestamp);
1720 void net_disable_timestamp(void)
1722 #ifdef HAVE_JUMP_LABEL
1723 if (in_interrupt()) {
1724 atomic_inc(&netstamp_needed_deferred);
1728 static_key_slow_dec(&netstamp_needed);
1730 EXPORT_SYMBOL(net_disable_timestamp);
1732 static inline void net_timestamp_set(struct sk_buff *skb)
1734 skb->tstamp.tv64 = 0;
1735 if (static_key_false(&netstamp_needed))
1736 __net_timestamp(skb);
1739 #define net_timestamp_check(COND, SKB) \
1740 if (static_key_false(&netstamp_needed)) { \
1741 if ((COND) && !(SKB)->tstamp.tv64) \
1742 __net_timestamp(SKB); \
1745 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
1749 if (!(dev->flags & IFF_UP))
1752 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1753 if (skb->len <= len)
1756 /* if TSO is enabled, we don't care about the length as the packet
1757 * could be forwarded without being segmented before
1759 if (skb_is_gso(skb))
1764 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1766 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1768 if (skb_orphan_frags(skb, GFP_ATOMIC) ||
1769 unlikely(!is_skb_forwardable(dev, skb))) {
1770 atomic_long_inc(&dev->rx_dropped);
1775 skb_scrub_packet(skb, true);
1777 skb->protocol = eth_type_trans(skb, dev);
1778 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1782 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1785 * dev_forward_skb - loopback an skb to another netif
1787 * @dev: destination network device
1788 * @skb: buffer to forward
1791 * NET_RX_SUCCESS (no congestion)
1792 * NET_RX_DROP (packet was dropped, but freed)
1794 * dev_forward_skb can be used for injecting an skb from the
1795 * start_xmit function of one device into the receive queue
1796 * of another device.
1798 * The receiving device may be in another namespace, so
1799 * we have to clear all information in the skb that could
1800 * impact namespace isolation.
1802 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1804 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1806 EXPORT_SYMBOL_GPL(dev_forward_skb);
1808 static inline int deliver_skb(struct sk_buff *skb,
1809 struct packet_type *pt_prev,
1810 struct net_device *orig_dev)
1812 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1814 atomic_inc(&skb->users);
1815 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1818 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1819 struct packet_type **pt,
1820 struct net_device *orig_dev,
1822 struct list_head *ptype_list)
1824 struct packet_type *ptype, *pt_prev = *pt;
1826 list_for_each_entry_rcu(ptype, ptype_list, list) {
1827 if (ptype->type != type)
1830 deliver_skb(skb, pt_prev, orig_dev);
1836 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1838 if (!ptype->af_packet_priv || !skb->sk)
1841 if (ptype->id_match)
1842 return ptype->id_match(ptype, skb->sk);
1843 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1850 * Support routine. Sends outgoing frames to any network
1851 * taps currently in use.
1854 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1856 struct packet_type *ptype;
1857 struct sk_buff *skb2 = NULL;
1858 struct packet_type *pt_prev = NULL;
1859 struct list_head *ptype_list = &ptype_all;
1863 list_for_each_entry_rcu(ptype, ptype_list, list) {
1864 /* Never send packets back to the socket
1865 * they originated from - MvS (miquels@drinkel.ow.org)
1867 if (skb_loop_sk(ptype, skb))
1871 deliver_skb(skb2, pt_prev, skb->dev);
1876 /* need to clone skb, done only once */
1877 skb2 = skb_clone(skb, GFP_ATOMIC);
1881 net_timestamp_set(skb2);
1883 /* skb->nh should be correctly
1884 * set by sender, so that the second statement is
1885 * just protection against buggy protocols.
1887 skb_reset_mac_header(skb2);
1889 if (skb_network_header(skb2) < skb2->data ||
1890 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1891 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1892 ntohs(skb2->protocol),
1894 skb_reset_network_header(skb2);
1897 skb2->transport_header = skb2->network_header;
1898 skb2->pkt_type = PACKET_OUTGOING;
1902 if (ptype_list == &ptype_all) {
1903 ptype_list = &dev->ptype_all;
1908 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1911 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
1914 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1915 * @dev: Network device
1916 * @txq: number of queues available
1918 * If real_num_tx_queues is changed the tc mappings may no longer be
1919 * valid. To resolve this verify the tc mapping remains valid and if
1920 * not NULL the mapping. With no priorities mapping to this
1921 * offset/count pair it will no longer be used. In the worst case TC0
1922 * is invalid nothing can be done so disable priority mappings. If is
1923 * expected that drivers will fix this mapping if they can before
1924 * calling netif_set_real_num_tx_queues.
1926 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1929 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1931 /* If TC0 is invalidated disable TC mapping */
1932 if (tc->offset + tc->count > txq) {
1933 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1938 /* Invalidated prio to tc mappings set to TC0 */
1939 for (i = 1; i < TC_BITMASK + 1; i++) {
1940 int q = netdev_get_prio_tc_map(dev, i);
1942 tc = &dev->tc_to_txq[q];
1943 if (tc->offset + tc->count > txq) {
1944 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1946 netdev_set_prio_tc_map(dev, i, 0);
1952 static DEFINE_MUTEX(xps_map_mutex);
1953 #define xmap_dereference(P) \
1954 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1956 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1959 struct xps_map *map = NULL;
1963 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1965 for (pos = 0; map && pos < map->len; pos++) {
1966 if (map->queues[pos] == index) {
1968 map->queues[pos] = map->queues[--map->len];
1970 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1971 kfree_rcu(map, rcu);
1981 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1983 struct xps_dev_maps *dev_maps;
1985 bool active = false;
1987 mutex_lock(&xps_map_mutex);
1988 dev_maps = xmap_dereference(dev->xps_maps);
1993 for_each_possible_cpu(cpu) {
1994 for (i = index; i < dev->num_tx_queues; i++) {
1995 if (!remove_xps_queue(dev_maps, cpu, i))
1998 if (i == dev->num_tx_queues)
2003 RCU_INIT_POINTER(dev->xps_maps, NULL);
2004 kfree_rcu(dev_maps, rcu);
2007 for (i = index; i < dev->num_tx_queues; i++)
2008 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
2012 mutex_unlock(&xps_map_mutex);
2015 static struct xps_map *expand_xps_map(struct xps_map *map,
2018 struct xps_map *new_map;
2019 int alloc_len = XPS_MIN_MAP_ALLOC;
2022 for (pos = 0; map && pos < map->len; pos++) {
2023 if (map->queues[pos] != index)
2028 /* Need to add queue to this CPU's existing map */
2030 if (pos < map->alloc_len)
2033 alloc_len = map->alloc_len * 2;
2036 /* Need to allocate new map to store queue on this CPU's map */
2037 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2042 for (i = 0; i < pos; i++)
2043 new_map->queues[i] = map->queues[i];
2044 new_map->alloc_len = alloc_len;
2050 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2053 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2054 struct xps_map *map, *new_map;
2055 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
2056 int cpu, numa_node_id = -2;
2057 bool active = false;
2059 mutex_lock(&xps_map_mutex);
2061 dev_maps = xmap_dereference(dev->xps_maps);
2063 /* allocate memory for queue storage */
2064 for_each_online_cpu(cpu) {
2065 if (!cpumask_test_cpu(cpu, mask))
2069 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2070 if (!new_dev_maps) {
2071 mutex_unlock(&xps_map_mutex);
2075 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2078 map = expand_xps_map(map, cpu, index);
2082 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2086 goto out_no_new_maps;
2088 for_each_possible_cpu(cpu) {
2089 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
2090 /* add queue to CPU maps */
2093 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2094 while ((pos < map->len) && (map->queues[pos] != index))
2097 if (pos == map->len)
2098 map->queues[map->len++] = index;
2100 if (numa_node_id == -2)
2101 numa_node_id = cpu_to_node(cpu);
2102 else if (numa_node_id != cpu_to_node(cpu))
2105 } else if (dev_maps) {
2106 /* fill in the new device map from the old device map */
2107 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2108 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2113 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2115 /* Cleanup old maps */
2117 for_each_possible_cpu(cpu) {
2118 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2119 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2120 if (map && map != new_map)
2121 kfree_rcu(map, rcu);
2124 kfree_rcu(dev_maps, rcu);
2127 dev_maps = new_dev_maps;
2131 /* update Tx queue numa node */
2132 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2133 (numa_node_id >= 0) ? numa_node_id :
2139 /* removes queue from unused CPUs */
2140 for_each_possible_cpu(cpu) {
2141 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2144 if (remove_xps_queue(dev_maps, cpu, index))
2148 /* free map if not active */
2150 RCU_INIT_POINTER(dev->xps_maps, NULL);
2151 kfree_rcu(dev_maps, rcu);
2155 mutex_unlock(&xps_map_mutex);
2159 /* remove any maps that we added */
2160 for_each_possible_cpu(cpu) {
2161 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2162 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2164 if (new_map && new_map != map)
2168 mutex_unlock(&xps_map_mutex);
2170 kfree(new_dev_maps);
2173 EXPORT_SYMBOL(netif_set_xps_queue);
2177 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2178 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2180 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2184 if (txq < 1 || txq > dev->num_tx_queues)
2187 if (dev->reg_state == NETREG_REGISTERED ||
2188 dev->reg_state == NETREG_UNREGISTERING) {
2191 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2197 netif_setup_tc(dev, txq);
2199 if (txq < dev->real_num_tx_queues) {
2200 qdisc_reset_all_tx_gt(dev, txq);
2202 netif_reset_xps_queues_gt(dev, txq);
2207 dev->real_num_tx_queues = txq;
2210 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2214 * netif_set_real_num_rx_queues - set actual number of RX queues used
2215 * @dev: Network device
2216 * @rxq: Actual number of RX queues
2218 * This must be called either with the rtnl_lock held or before
2219 * registration of the net device. Returns 0 on success, or a
2220 * negative error code. If called before registration, it always
2223 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2227 if (rxq < 1 || rxq > dev->num_rx_queues)
2230 if (dev->reg_state == NETREG_REGISTERED) {
2233 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2239 dev->real_num_rx_queues = rxq;
2242 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2246 * netif_get_num_default_rss_queues - default number of RSS queues
2248 * This routine should set an upper limit on the number of RSS queues
2249 * used by default by multiqueue devices.
2251 int netif_get_num_default_rss_queues(void)
2253 return is_kdump_kernel() ?
2254 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2256 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2258 static void __netif_reschedule(struct Qdisc *q)
2260 struct softnet_data *sd;
2261 unsigned long flags;
2263 local_irq_save(flags);
2264 sd = this_cpu_ptr(&softnet_data);
2265 q->next_sched = NULL;
2266 *sd->output_queue_tailp = q;
2267 sd->output_queue_tailp = &q->next_sched;
2268 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2269 local_irq_restore(flags);
2272 void __netif_schedule(struct Qdisc *q)
2274 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2275 __netif_reschedule(q);
2277 EXPORT_SYMBOL(__netif_schedule);
2279 struct dev_kfree_skb_cb {
2280 enum skb_free_reason reason;
2283 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2285 return (struct dev_kfree_skb_cb *)skb->cb;
2288 void netif_schedule_queue(struct netdev_queue *txq)
2291 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2292 struct Qdisc *q = rcu_dereference(txq->qdisc);
2294 __netif_schedule(q);
2298 EXPORT_SYMBOL(netif_schedule_queue);
2301 * netif_wake_subqueue - allow sending packets on subqueue
2302 * @dev: network device
2303 * @queue_index: sub queue index
2305 * Resume individual transmit queue of a device with multiple transmit queues.
2307 void netif_wake_subqueue(struct net_device *dev, u16 queue_index)
2309 struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index);
2311 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) {
2315 q = rcu_dereference(txq->qdisc);
2316 __netif_schedule(q);
2320 EXPORT_SYMBOL(netif_wake_subqueue);
2322 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2324 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2328 q = rcu_dereference(dev_queue->qdisc);
2329 __netif_schedule(q);
2333 EXPORT_SYMBOL(netif_tx_wake_queue);
2335 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2337 unsigned long flags;
2339 if (likely(atomic_read(&skb->users) == 1)) {
2341 atomic_set(&skb->users, 0);
2342 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2345 get_kfree_skb_cb(skb)->reason = reason;
2346 local_irq_save(flags);
2347 skb->next = __this_cpu_read(softnet_data.completion_queue);
2348 __this_cpu_write(softnet_data.completion_queue, skb);
2349 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2350 local_irq_restore(flags);
2352 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2354 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2356 if (in_irq() || irqs_disabled())
2357 __dev_kfree_skb_irq(skb, reason);
2361 EXPORT_SYMBOL(__dev_kfree_skb_any);
2365 * netif_device_detach - mark device as removed
2366 * @dev: network device
2368 * Mark device as removed from system and therefore no longer available.
2370 void netif_device_detach(struct net_device *dev)
2372 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2373 netif_running(dev)) {
2374 netif_tx_stop_all_queues(dev);
2377 EXPORT_SYMBOL(netif_device_detach);
2380 * netif_device_attach - mark device as attached
2381 * @dev: network device
2383 * Mark device as attached from system and restart if needed.
2385 void netif_device_attach(struct net_device *dev)
2387 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2388 netif_running(dev)) {
2389 netif_tx_wake_all_queues(dev);
2390 __netdev_watchdog_up(dev);
2393 EXPORT_SYMBOL(netif_device_attach);
2396 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2397 * to be used as a distribution range.
2399 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
2400 unsigned int num_tx_queues)
2404 u16 qcount = num_tx_queues;
2406 if (skb_rx_queue_recorded(skb)) {
2407 hash = skb_get_rx_queue(skb);
2408 while (unlikely(hash >= num_tx_queues))
2409 hash -= num_tx_queues;
2414 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2415 qoffset = dev->tc_to_txq[tc].offset;
2416 qcount = dev->tc_to_txq[tc].count;
2419 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
2421 EXPORT_SYMBOL(__skb_tx_hash);
2423 static void skb_warn_bad_offload(const struct sk_buff *skb)
2425 static const netdev_features_t null_features;
2426 struct net_device *dev = skb->dev;
2427 const char *name = "";
2429 if (!net_ratelimit())
2433 if (dev->dev.parent)
2434 name = dev_driver_string(dev->dev.parent);
2436 name = netdev_name(dev);
2438 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2439 "gso_type=%d ip_summed=%d\n",
2440 name, dev ? &dev->features : &null_features,
2441 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2442 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2443 skb_shinfo(skb)->gso_type, skb->ip_summed);
2447 * Invalidate hardware checksum when packet is to be mangled, and
2448 * complete checksum manually on outgoing path.
2450 int skb_checksum_help(struct sk_buff *skb)
2453 int ret = 0, offset;
2455 if (skb->ip_summed == CHECKSUM_COMPLETE)
2456 goto out_set_summed;
2458 if (unlikely(skb_shinfo(skb)->gso_size)) {
2459 skb_warn_bad_offload(skb);
2463 /* Before computing a checksum, we should make sure no frag could
2464 * be modified by an external entity : checksum could be wrong.
2466 if (skb_has_shared_frag(skb)) {
2467 ret = __skb_linearize(skb);
2472 offset = skb_checksum_start_offset(skb);
2473 BUG_ON(offset >= skb_headlen(skb));
2474 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2476 offset += skb->csum_offset;
2477 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2479 if (skb_cloned(skb) &&
2480 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2481 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2486 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2488 skb->ip_summed = CHECKSUM_NONE;
2492 EXPORT_SYMBOL(skb_checksum_help);
2494 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2496 __be16 type = skb->protocol;
2498 /* Tunnel gso handlers can set protocol to ethernet. */
2499 if (type == htons(ETH_P_TEB)) {
2502 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2505 eth = (struct ethhdr *)skb_mac_header(skb);
2506 type = eth->h_proto;
2509 return __vlan_get_protocol(skb, type, depth);
2513 * skb_mac_gso_segment - mac layer segmentation handler.
2514 * @skb: buffer to segment
2515 * @features: features for the output path (see dev->features)
2517 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2518 netdev_features_t features)
2520 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2521 struct packet_offload *ptype;
2522 int vlan_depth = skb->mac_len;
2523 __be16 type = skb_network_protocol(skb, &vlan_depth);
2525 if (unlikely(!type))
2526 return ERR_PTR(-EINVAL);
2528 __skb_pull(skb, vlan_depth);
2531 list_for_each_entry_rcu(ptype, &offload_base, list) {
2532 if (ptype->type == type && ptype->callbacks.gso_segment) {
2533 segs = ptype->callbacks.gso_segment(skb, features);
2539 __skb_push(skb, skb->data - skb_mac_header(skb));
2543 EXPORT_SYMBOL(skb_mac_gso_segment);
2546 /* openvswitch calls this on rx path, so we need a different check.
2548 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2551 return skb->ip_summed != CHECKSUM_PARTIAL;
2553 return skb->ip_summed == CHECKSUM_NONE;
2557 * __skb_gso_segment - Perform segmentation on skb.
2558 * @skb: buffer to segment
2559 * @features: features for the output path (see dev->features)
2560 * @tx_path: whether it is called in TX path
2562 * This function segments the given skb and returns a list of segments.
2564 * It may return NULL if the skb requires no segmentation. This is
2565 * only possible when GSO is used for verifying header integrity.
2567 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
2569 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2570 netdev_features_t features, bool tx_path)
2572 if (unlikely(skb_needs_check(skb, tx_path))) {
2575 skb_warn_bad_offload(skb);
2577 err = skb_cow_head(skb, 0);
2579 return ERR_PTR(err);
2582 /* Only report GSO partial support if it will enable us to
2583 * support segmentation on this frame without needing additional
2586 if (features & NETIF_F_GSO_PARTIAL) {
2587 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
2588 struct net_device *dev = skb->dev;
2590 partial_features |= dev->features & dev->gso_partial_features;
2591 if (!skb_gso_ok(skb, features | partial_features))
2592 features &= ~NETIF_F_GSO_PARTIAL;
2595 BUILD_BUG_ON(SKB_SGO_CB_OFFSET +
2596 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
2598 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2599 SKB_GSO_CB(skb)->encap_level = 0;
2601 skb_reset_mac_header(skb);
2602 skb_reset_mac_len(skb);
2604 return skb_mac_gso_segment(skb, features);
2606 EXPORT_SYMBOL(__skb_gso_segment);
2608 /* Take action when hardware reception checksum errors are detected. */
2610 void netdev_rx_csum_fault(struct net_device *dev)
2612 if (net_ratelimit()) {
2613 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2617 EXPORT_SYMBOL(netdev_rx_csum_fault);
2620 /* Actually, we should eliminate this check as soon as we know, that:
2621 * 1. IOMMU is present and allows to map all the memory.
2622 * 2. No high memory really exists on this machine.
2625 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2627 #ifdef CONFIG_HIGHMEM
2629 if (!(dev->features & NETIF_F_HIGHDMA)) {
2630 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2631 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2632 if (PageHighMem(skb_frag_page(frag)))
2637 if (PCI_DMA_BUS_IS_PHYS) {
2638 struct device *pdev = dev->dev.parent;
2642 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2643 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2644 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2645 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2653 /* If MPLS offload request, verify we are testing hardware MPLS features
2654 * instead of standard features for the netdev.
2656 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
2657 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2658 netdev_features_t features,
2661 if (eth_p_mpls(type))
2662 features &= skb->dev->mpls_features;
2667 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2668 netdev_features_t features,
2675 static netdev_features_t harmonize_features(struct sk_buff *skb,
2676 netdev_features_t features)
2681 type = skb_network_protocol(skb, &tmp);
2682 features = net_mpls_features(skb, features, type);
2684 if (skb->ip_summed != CHECKSUM_NONE &&
2685 !can_checksum_protocol(features, type)) {
2686 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2687 } else if (illegal_highdma(skb->dev, skb)) {
2688 features &= ~NETIF_F_SG;
2694 netdev_features_t passthru_features_check(struct sk_buff *skb,
2695 struct net_device *dev,
2696 netdev_features_t features)
2700 EXPORT_SYMBOL(passthru_features_check);
2702 static netdev_features_t dflt_features_check(const struct sk_buff *skb,
2703 struct net_device *dev,
2704 netdev_features_t features)
2706 return vlan_features_check(skb, features);
2709 static netdev_features_t gso_features_check(const struct sk_buff *skb,
2710 struct net_device *dev,
2711 netdev_features_t features)
2713 u16 gso_segs = skb_shinfo(skb)->gso_segs;
2715 if (gso_segs > dev->gso_max_segs)
2716 return features & ~NETIF_F_GSO_MASK;
2718 /* Support for GSO partial features requires software
2719 * intervention before we can actually process the packets
2720 * so we need to strip support for any partial features now
2721 * and we can pull them back in after we have partially
2722 * segmented the frame.
2724 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
2725 features &= ~dev->gso_partial_features;
2727 /* Make sure to clear the IPv4 ID mangling feature if the
2728 * IPv4 header has the potential to be fragmented.
2730 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
2731 struct iphdr *iph = skb->encapsulation ?
2732 inner_ip_hdr(skb) : ip_hdr(skb);
2734 if (!(iph->frag_off & htons(IP_DF)))
2735 features &= ~NETIF_F_TSO_MANGLEID;
2741 netdev_features_t netif_skb_features(struct sk_buff *skb)
2743 struct net_device *dev = skb->dev;
2744 netdev_features_t features = dev->features;
2746 if (skb_is_gso(skb))
2747 features = gso_features_check(skb, dev, features);
2749 /* If encapsulation offload request, verify we are testing
2750 * hardware encapsulation features instead of standard
2751 * features for the netdev
2753 if (skb->encapsulation)
2754 features &= dev->hw_enc_features;
2756 if (skb_vlan_tagged(skb))
2757 features = netdev_intersect_features(features,
2758 dev->vlan_features |
2759 NETIF_F_HW_VLAN_CTAG_TX |
2760 NETIF_F_HW_VLAN_STAG_TX);
2762 if (dev->netdev_ops->ndo_features_check)
2763 features &= dev->netdev_ops->ndo_features_check(skb, dev,
2766 features &= dflt_features_check(skb, dev, features);
2768 return harmonize_features(skb, features);
2770 EXPORT_SYMBOL(netif_skb_features);
2772 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
2773 struct netdev_queue *txq, bool more)
2778 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
2779 dev_queue_xmit_nit(skb, dev);
2782 trace_net_dev_start_xmit(skb, dev);
2783 rc = netdev_start_xmit(skb, dev, txq, more);
2784 trace_net_dev_xmit(skb, rc, dev, len);
2789 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
2790 struct netdev_queue *txq, int *ret)
2792 struct sk_buff *skb = first;
2793 int rc = NETDEV_TX_OK;
2796 struct sk_buff *next = skb->next;
2799 rc = xmit_one(skb, dev, txq, next != NULL);
2800 if (unlikely(!dev_xmit_complete(rc))) {
2806 if (netif_xmit_stopped(txq) && skb) {
2807 rc = NETDEV_TX_BUSY;
2817 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
2818 netdev_features_t features)
2820 if (skb_vlan_tag_present(skb) &&
2821 !vlan_hw_offload_capable(features, skb->vlan_proto))
2822 skb = __vlan_hwaccel_push_inside(skb);
2826 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
2828 netdev_features_t features;
2830 features = netif_skb_features(skb);
2831 skb = validate_xmit_vlan(skb, features);
2835 if (netif_needs_gso(skb, features)) {
2836 struct sk_buff *segs;
2838 segs = skb_gso_segment(skb, features);
2846 if (skb_needs_linearize(skb, features) &&
2847 __skb_linearize(skb))
2850 /* If packet is not checksummed and device does not
2851 * support checksumming for this protocol, complete
2852 * checksumming here.
2854 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2855 if (skb->encapsulation)
2856 skb_set_inner_transport_header(skb,
2857 skb_checksum_start_offset(skb));
2859 skb_set_transport_header(skb,
2860 skb_checksum_start_offset(skb));
2861 if (!(features & NETIF_F_CSUM_MASK) &&
2862 skb_checksum_help(skb))
2872 atomic_long_inc(&dev->tx_dropped);
2876 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev)
2878 struct sk_buff *next, *head = NULL, *tail;
2880 for (; skb != NULL; skb = next) {
2884 /* in case skb wont be segmented, point to itself */
2887 skb = validate_xmit_skb(skb, dev);
2895 /* If skb was segmented, skb->prev points to
2896 * the last segment. If not, it still contains skb.
2903 static void qdisc_pkt_len_init(struct sk_buff *skb)
2905 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2907 qdisc_skb_cb(skb)->pkt_len = skb->len;
2909 /* To get more precise estimation of bytes sent on wire,
2910 * we add to pkt_len the headers size of all segments
2912 if (shinfo->gso_size) {
2913 unsigned int hdr_len;
2914 u16 gso_segs = shinfo->gso_segs;
2916 /* mac layer + network layer */
2917 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2919 /* + transport layer */
2920 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2921 hdr_len += tcp_hdrlen(skb);
2923 hdr_len += sizeof(struct udphdr);
2925 if (shinfo->gso_type & SKB_GSO_DODGY)
2926 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2929 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2933 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2934 struct net_device *dev,
2935 struct netdev_queue *txq)
2937 spinlock_t *root_lock = qdisc_lock(q);
2938 struct sk_buff *to_free = NULL;
2942 qdisc_calculate_pkt_len(skb, q);
2944 * Heuristic to force contended enqueues to serialize on a
2945 * separate lock before trying to get qdisc main lock.
2946 * This permits qdisc->running owner to get the lock more
2947 * often and dequeue packets faster.
2949 contended = qdisc_is_running(q);
2950 if (unlikely(contended))
2951 spin_lock(&q->busylock);
2953 spin_lock(root_lock);
2954 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2955 __qdisc_drop(skb, &to_free);
2957 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2958 qdisc_run_begin(q)) {
2960 * This is a work-conserving queue; there are no old skbs
2961 * waiting to be sent out; and the qdisc is not running -
2962 * xmit the skb directly.
2965 qdisc_bstats_update(q, skb);
2967 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
2968 if (unlikely(contended)) {
2969 spin_unlock(&q->busylock);
2976 rc = NET_XMIT_SUCCESS;
2978 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
2979 if (qdisc_run_begin(q)) {
2980 if (unlikely(contended)) {
2981 spin_unlock(&q->busylock);
2987 spin_unlock(root_lock);
2988 if (unlikely(to_free))
2989 kfree_skb_list(to_free);
2990 if (unlikely(contended))
2991 spin_unlock(&q->busylock);
2995 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2996 static void skb_update_prio(struct sk_buff *skb)
2998 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
3000 if (!skb->priority && skb->sk && map) {
3001 unsigned int prioidx =
3002 sock_cgroup_prioidx(&skb->sk->sk_cgrp_data);
3004 if (prioidx < map->priomap_len)
3005 skb->priority = map->priomap[prioidx];
3009 #define skb_update_prio(skb)
3012 DEFINE_PER_CPU(int, xmit_recursion);
3013 EXPORT_SYMBOL(xmit_recursion);
3016 * dev_loopback_xmit - loop back @skb
3017 * @net: network namespace this loopback is happening in
3018 * @sk: sk needed to be a netfilter okfn
3019 * @skb: buffer to transmit
3021 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3023 skb_reset_mac_header(skb);
3024 __skb_pull(skb, skb_network_offset(skb));
3025 skb->pkt_type = PACKET_LOOPBACK;
3026 skb->ip_summed = CHECKSUM_UNNECESSARY;
3027 WARN_ON(!skb_dst(skb));
3032 EXPORT_SYMBOL(dev_loopback_xmit);
3034 #ifdef CONFIG_NET_EGRESS
3035 static struct sk_buff *
3036 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3038 struct tcf_proto *cl = rcu_dereference_bh(dev->egress_cl_list);
3039 struct tcf_result cl_res;
3044 /* skb->tc_verd and qdisc_skb_cb(skb)->pkt_len were already set
3045 * earlier by the caller.
3047 qdisc_bstats_cpu_update(cl->q, skb);
3049 switch (tc_classify(skb, cl, &cl_res, false)) {
3051 case TC_ACT_RECLASSIFY:
3052 skb->tc_index = TC_H_MIN(cl_res.classid);
3055 qdisc_qstats_cpu_drop(cl->q);
3056 *ret = NET_XMIT_DROP;
3061 *ret = NET_XMIT_SUCCESS;
3064 case TC_ACT_REDIRECT:
3065 /* No need to push/pop skb's mac_header here on egress! */
3066 skb_do_redirect(skb);
3067 *ret = NET_XMIT_SUCCESS;
3075 #endif /* CONFIG_NET_EGRESS */
3077 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
3080 struct xps_dev_maps *dev_maps;
3081 struct xps_map *map;
3082 int queue_index = -1;
3085 dev_maps = rcu_dereference(dev->xps_maps);
3087 map = rcu_dereference(
3088 dev_maps->cpu_map[skb->sender_cpu - 1]);
3091 queue_index = map->queues[0];
3093 queue_index = map->queues[reciprocal_scale(skb_get_hash(skb),
3095 if (unlikely(queue_index >= dev->real_num_tx_queues))
3107 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb)
3109 struct sock *sk = skb->sk;
3110 int queue_index = sk_tx_queue_get(sk);
3112 if (queue_index < 0 || skb->ooo_okay ||
3113 queue_index >= dev->real_num_tx_queues) {
3114 int new_index = get_xps_queue(dev, skb);
3116 new_index = skb_tx_hash(dev, skb);
3118 if (queue_index != new_index && sk &&
3120 rcu_access_pointer(sk->sk_dst_cache))
3121 sk_tx_queue_set(sk, new_index);
3123 queue_index = new_index;
3129 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
3130 struct sk_buff *skb,
3133 int queue_index = 0;
3136 u32 sender_cpu = skb->sender_cpu - 1;
3138 if (sender_cpu >= (u32)NR_CPUS)
3139 skb->sender_cpu = raw_smp_processor_id() + 1;
3142 if (dev->real_num_tx_queues != 1) {
3143 const struct net_device_ops *ops = dev->netdev_ops;
3144 if (ops->ndo_select_queue)
3145 queue_index = ops->ndo_select_queue(dev, skb, accel_priv,
3148 queue_index = __netdev_pick_tx(dev, skb);
3151 queue_index = netdev_cap_txqueue(dev, queue_index);
3154 skb_set_queue_mapping(skb, queue_index);
3155 return netdev_get_tx_queue(dev, queue_index);
3159 * __dev_queue_xmit - transmit a buffer
3160 * @skb: buffer to transmit
3161 * @accel_priv: private data used for L2 forwarding offload
3163 * Queue a buffer for transmission to a network device. The caller must
3164 * have set the device and priority and built the buffer before calling
3165 * this function. The function can be called from an interrupt.
3167 * A negative errno code is returned on a failure. A success does not
3168 * guarantee the frame will be transmitted as it may be dropped due
3169 * to congestion or traffic shaping.
3171 * -----------------------------------------------------------------------------------
3172 * I notice this method can also return errors from the queue disciplines,
3173 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3176 * Regardless of the return value, the skb is consumed, so it is currently
3177 * difficult to retry a send to this method. (You can bump the ref count
3178 * before sending to hold a reference for retry if you are careful.)
3180 * When calling this method, interrupts MUST be enabled. This is because
3181 * the BH enable code must have IRQs enabled so that it will not deadlock.
3184 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
3186 struct net_device *dev = skb->dev;
3187 struct netdev_queue *txq;
3191 skb_reset_mac_header(skb);
3193 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3194 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3196 /* Disable soft irqs for various locks below. Also
3197 * stops preemption for RCU.
3201 skb_update_prio(skb);
3203 qdisc_pkt_len_init(skb);
3204 #ifdef CONFIG_NET_CLS_ACT
3205 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
3206 # ifdef CONFIG_NET_EGRESS
3207 if (static_key_false(&egress_needed)) {
3208 skb = sch_handle_egress(skb, &rc, dev);
3214 /* If device/qdisc don't need skb->dst, release it right now while
3215 * its hot in this cpu cache.
3217 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3222 txq = netdev_pick_tx(dev, skb, accel_priv);
3223 q = rcu_dereference_bh(txq->qdisc);
3225 trace_net_dev_queue(skb);
3227 rc = __dev_xmit_skb(skb, q, dev, txq);
3231 /* The device has no queue. Common case for software devices:
3232 loopback, all the sorts of tunnels...
3234 Really, it is unlikely that netif_tx_lock protection is necessary
3235 here. (f.e. loopback and IP tunnels are clean ignoring statistics
3237 However, it is possible, that they rely on protection
3240 Check this and shot the lock. It is not prone from deadlocks.
3241 Either shot noqueue qdisc, it is even simpler 8)
3243 if (dev->flags & IFF_UP) {
3244 int cpu = smp_processor_id(); /* ok because BHs are off */
3246 if (txq->xmit_lock_owner != cpu) {
3247 if (unlikely(__this_cpu_read(xmit_recursion) >
3248 XMIT_RECURSION_LIMIT))
3249 goto recursion_alert;
3251 skb = validate_xmit_skb(skb, dev);
3255 HARD_TX_LOCK(dev, txq, cpu);
3257 if (!netif_xmit_stopped(txq)) {
3258 __this_cpu_inc(xmit_recursion);
3259 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
3260 __this_cpu_dec(xmit_recursion);
3261 if (dev_xmit_complete(rc)) {
3262 HARD_TX_UNLOCK(dev, txq);
3266 HARD_TX_UNLOCK(dev, txq);
3267 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3270 /* Recursion is detected! It is possible,
3274 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3280 rcu_read_unlock_bh();
3282 atomic_long_inc(&dev->tx_dropped);
3283 kfree_skb_list(skb);
3286 rcu_read_unlock_bh();
3290 int dev_queue_xmit(struct sk_buff *skb)
3292 return __dev_queue_xmit(skb, NULL);
3294 EXPORT_SYMBOL(dev_queue_xmit);
3296 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
3298 return __dev_queue_xmit(skb, accel_priv);
3300 EXPORT_SYMBOL(dev_queue_xmit_accel);
3303 /*=======================================================================
3305 =======================================================================*/
3307 int netdev_max_backlog __read_mostly = 1000;
3308 EXPORT_SYMBOL(netdev_max_backlog);
3310 int netdev_tstamp_prequeue __read_mostly = 1;
3311 int netdev_budget __read_mostly = 300;
3312 int weight_p __read_mostly = 64; /* old backlog weight */
3314 /* Called with irq disabled */
3315 static inline void ____napi_schedule(struct softnet_data *sd,
3316 struct napi_struct *napi)
3318 list_add_tail(&napi->poll_list, &sd->poll_list);
3319 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3324 /* One global table that all flow-based protocols share. */
3325 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3326 EXPORT_SYMBOL(rps_sock_flow_table);
3327 u32 rps_cpu_mask __read_mostly;
3328 EXPORT_SYMBOL(rps_cpu_mask);
3330 struct static_key rps_needed __read_mostly;
3331 EXPORT_SYMBOL(rps_needed);
3333 static struct rps_dev_flow *
3334 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3335 struct rps_dev_flow *rflow, u16 next_cpu)
3337 if (next_cpu < nr_cpu_ids) {
3338 #ifdef CONFIG_RFS_ACCEL
3339 struct netdev_rx_queue *rxqueue;
3340 struct rps_dev_flow_table *flow_table;
3341 struct rps_dev_flow *old_rflow;
3346 /* Should we steer this flow to a different hardware queue? */
3347 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3348 !(dev->features & NETIF_F_NTUPLE))
3350 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3351 if (rxq_index == skb_get_rx_queue(skb))
3354 rxqueue = dev->_rx + rxq_index;
3355 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3358 flow_id = skb_get_hash(skb) & flow_table->mask;
3359 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3360 rxq_index, flow_id);
3364 rflow = &flow_table->flows[flow_id];
3366 if (old_rflow->filter == rflow->filter)
3367 old_rflow->filter = RPS_NO_FILTER;
3371 per_cpu(softnet_data, next_cpu).input_queue_head;
3374 rflow->cpu = next_cpu;
3379 * get_rps_cpu is called from netif_receive_skb and returns the target
3380 * CPU from the RPS map of the receiving queue for a given skb.
3381 * rcu_read_lock must be held on entry.
3383 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3384 struct rps_dev_flow **rflowp)
3386 const struct rps_sock_flow_table *sock_flow_table;
3387 struct netdev_rx_queue *rxqueue = dev->_rx;
3388 struct rps_dev_flow_table *flow_table;
3389 struct rps_map *map;
3394 if (skb_rx_queue_recorded(skb)) {
3395 u16 index = skb_get_rx_queue(skb);
3397 if (unlikely(index >= dev->real_num_rx_queues)) {
3398 WARN_ONCE(dev->real_num_rx_queues > 1,
3399 "%s received packet on queue %u, but number "
3400 "of RX queues is %u\n",
3401 dev->name, index, dev->real_num_rx_queues);
3407 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3409 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3410 map = rcu_dereference(rxqueue->rps_map);
3411 if (!flow_table && !map)
3414 skb_reset_network_header(skb);
3415 hash = skb_get_hash(skb);
3419 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3420 if (flow_table && sock_flow_table) {
3421 struct rps_dev_flow *rflow;
3425 /* First check into global flow table if there is a match */
3426 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
3427 if ((ident ^ hash) & ~rps_cpu_mask)
3430 next_cpu = ident & rps_cpu_mask;
3432 /* OK, now we know there is a match,
3433 * we can look at the local (per receive queue) flow table
3435 rflow = &flow_table->flows[hash & flow_table->mask];
3439 * If the desired CPU (where last recvmsg was done) is
3440 * different from current CPU (one in the rx-queue flow
3441 * table entry), switch if one of the following holds:
3442 * - Current CPU is unset (>= nr_cpu_ids).
3443 * - Current CPU is offline.
3444 * - The current CPU's queue tail has advanced beyond the
3445 * last packet that was enqueued using this table entry.
3446 * This guarantees that all previous packets for the flow
3447 * have been dequeued, thus preserving in order delivery.
3449 if (unlikely(tcpu != next_cpu) &&
3450 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
3451 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3452 rflow->last_qtail)) >= 0)) {
3454 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3457 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
3467 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3468 if (cpu_online(tcpu)) {
3478 #ifdef CONFIG_RFS_ACCEL
3481 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3482 * @dev: Device on which the filter was set
3483 * @rxq_index: RX queue index
3484 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3485 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3487 * Drivers that implement ndo_rx_flow_steer() should periodically call
3488 * this function for each installed filter and remove the filters for
3489 * which it returns %true.
3491 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3492 u32 flow_id, u16 filter_id)
3494 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3495 struct rps_dev_flow_table *flow_table;
3496 struct rps_dev_flow *rflow;
3501 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3502 if (flow_table && flow_id <= flow_table->mask) {
3503 rflow = &flow_table->flows[flow_id];
3504 cpu = ACCESS_ONCE(rflow->cpu);
3505 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
3506 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3507 rflow->last_qtail) <
3508 (int)(10 * flow_table->mask)))
3514 EXPORT_SYMBOL(rps_may_expire_flow);
3516 #endif /* CONFIG_RFS_ACCEL */
3518 /* Called from hardirq (IPI) context */
3519 static void rps_trigger_softirq(void *data)
3521 struct softnet_data *sd = data;
3523 ____napi_schedule(sd, &sd->backlog);
3527 #endif /* CONFIG_RPS */
3530 * Check if this softnet_data structure is another cpu one
3531 * If yes, queue it to our IPI list and return 1
3534 static int rps_ipi_queued(struct softnet_data *sd)
3537 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3540 sd->rps_ipi_next = mysd->rps_ipi_list;
3541 mysd->rps_ipi_list = sd;
3543 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3546 #endif /* CONFIG_RPS */
3550 #ifdef CONFIG_NET_FLOW_LIMIT
3551 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3554 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3556 #ifdef CONFIG_NET_FLOW_LIMIT
3557 struct sd_flow_limit *fl;
3558 struct softnet_data *sd;
3559 unsigned int old_flow, new_flow;
3561 if (qlen < (netdev_max_backlog >> 1))
3564 sd = this_cpu_ptr(&softnet_data);
3567 fl = rcu_dereference(sd->flow_limit);
3569 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3570 old_flow = fl->history[fl->history_head];
3571 fl->history[fl->history_head] = new_flow;
3574 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3576 if (likely(fl->buckets[old_flow]))
3577 fl->buckets[old_flow]--;
3579 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3591 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3592 * queue (may be a remote CPU queue).
3594 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3595 unsigned int *qtail)
3597 struct softnet_data *sd;
3598 unsigned long flags;
3601 sd = &per_cpu(softnet_data, cpu);
3603 local_irq_save(flags);
3606 if (!netif_running(skb->dev))
3608 qlen = skb_queue_len(&sd->input_pkt_queue);
3609 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3612 __skb_queue_tail(&sd->input_pkt_queue, skb);
3613 input_queue_tail_incr_save(sd, qtail);
3615 local_irq_restore(flags);
3616 return NET_RX_SUCCESS;
3619 /* Schedule NAPI for backlog device
3620 * We can use non atomic operation since we own the queue lock
3622 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3623 if (!rps_ipi_queued(sd))
3624 ____napi_schedule(sd, &sd->backlog);
3633 local_irq_restore(flags);
3635 atomic_long_inc(&skb->dev->rx_dropped);
3640 static int netif_rx_internal(struct sk_buff *skb)
3644 net_timestamp_check(netdev_tstamp_prequeue, skb);
3646 trace_netif_rx(skb);
3648 if (static_key_false(&rps_needed)) {
3649 struct rps_dev_flow voidflow, *rflow = &voidflow;
3655 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3657 cpu = smp_processor_id();
3659 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3667 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3674 * netif_rx - post buffer to the network code
3675 * @skb: buffer to post
3677 * This function receives a packet from a device driver and queues it for
3678 * the upper (protocol) levels to process. It always succeeds. The buffer
3679 * may be dropped during processing for congestion control or by the
3683 * NET_RX_SUCCESS (no congestion)
3684 * NET_RX_DROP (packet was dropped)
3688 int netif_rx(struct sk_buff *skb)
3690 trace_netif_rx_entry(skb);
3692 return netif_rx_internal(skb);
3694 EXPORT_SYMBOL(netif_rx);
3696 int netif_rx_ni(struct sk_buff *skb)
3700 trace_netif_rx_ni_entry(skb);
3703 err = netif_rx_internal(skb);
3704 if (local_softirq_pending())
3710 EXPORT_SYMBOL(netif_rx_ni);
3712 static void net_tx_action(struct softirq_action *h)
3714 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3716 if (sd->completion_queue) {
3717 struct sk_buff *clist;
3719 local_irq_disable();
3720 clist = sd->completion_queue;
3721 sd->completion_queue = NULL;
3725 struct sk_buff *skb = clist;
3726 clist = clist->next;
3728 WARN_ON(atomic_read(&skb->users));
3729 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3730 trace_consume_skb(skb);
3732 trace_kfree_skb(skb, net_tx_action);
3734 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
3737 __kfree_skb_defer(skb);
3740 __kfree_skb_flush();
3743 if (sd->output_queue) {
3746 local_irq_disable();
3747 head = sd->output_queue;
3748 sd->output_queue = NULL;
3749 sd->output_queue_tailp = &sd->output_queue;
3753 struct Qdisc *q = head;
3754 spinlock_t *root_lock;
3756 head = head->next_sched;
3758 root_lock = qdisc_lock(q);
3759 spin_lock(root_lock);
3760 /* We need to make sure head->next_sched is read
3761 * before clearing __QDISC_STATE_SCHED
3763 smp_mb__before_atomic();
3764 clear_bit(__QDISC_STATE_SCHED, &q->state);
3766 spin_unlock(root_lock);
3771 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
3772 /* This hook is defined here for ATM LANE */
3773 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3774 unsigned char *addr) __read_mostly;
3775 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3778 static inline struct sk_buff *
3779 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
3780 struct net_device *orig_dev)
3782 #ifdef CONFIG_NET_CLS_ACT
3783 struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list);
3784 struct tcf_result cl_res;
3786 /* If there's at least one ingress present somewhere (so
3787 * we get here via enabled static key), remaining devices
3788 * that are not configured with an ingress qdisc will bail
3794 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3798 qdisc_skb_cb(skb)->pkt_len = skb->len;
3799 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3800 qdisc_bstats_cpu_update(cl->q, skb);
3802 switch (tc_classify(skb, cl, &cl_res, false)) {
3804 case TC_ACT_RECLASSIFY:
3805 skb->tc_index = TC_H_MIN(cl_res.classid);
3808 qdisc_qstats_cpu_drop(cl->q);
3815 case TC_ACT_REDIRECT:
3816 /* skb_mac_header check was done by cls/act_bpf, so
3817 * we can safely push the L2 header back before
3818 * redirecting to another netdev
3820 __skb_push(skb, skb->mac_len);
3821 skb_do_redirect(skb);
3826 #endif /* CONFIG_NET_CLS_ACT */
3831 * netdev_is_rx_handler_busy - check if receive handler is registered
3832 * @dev: device to check
3834 * Check if a receive handler is already registered for a given device.
3835 * Return true if there one.
3837 * The caller must hold the rtnl_mutex.
3839 bool netdev_is_rx_handler_busy(struct net_device *dev)
3842 return dev && rtnl_dereference(dev->rx_handler);
3844 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
3847 * netdev_rx_handler_register - register receive handler
3848 * @dev: device to register a handler for
3849 * @rx_handler: receive handler to register
3850 * @rx_handler_data: data pointer that is used by rx handler
3852 * Register a receive handler for a device. This handler will then be
3853 * called from __netif_receive_skb. A negative errno code is returned
3856 * The caller must hold the rtnl_mutex.
3858 * For a general description of rx_handler, see enum rx_handler_result.
3860 int netdev_rx_handler_register(struct net_device *dev,
3861 rx_handler_func_t *rx_handler,
3862 void *rx_handler_data)
3866 if (dev->rx_handler)
3869 /* Note: rx_handler_data must be set before rx_handler */
3870 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3871 rcu_assign_pointer(dev->rx_handler, rx_handler);
3875 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3878 * netdev_rx_handler_unregister - unregister receive handler
3879 * @dev: device to unregister a handler from
3881 * Unregister a receive handler from a device.
3883 * The caller must hold the rtnl_mutex.
3885 void netdev_rx_handler_unregister(struct net_device *dev)
3889 RCU_INIT_POINTER(dev->rx_handler, NULL);
3890 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3891 * section has a guarantee to see a non NULL rx_handler_data
3895 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3897 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3900 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3901 * the special handling of PFMEMALLOC skbs.
3903 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3905 switch (skb->protocol) {
3906 case htons(ETH_P_ARP):
3907 case htons(ETH_P_IP):
3908 case htons(ETH_P_IPV6):
3909 case htons(ETH_P_8021Q):
3910 case htons(ETH_P_8021AD):
3917 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
3918 int *ret, struct net_device *orig_dev)
3920 #ifdef CONFIG_NETFILTER_INGRESS
3921 if (nf_hook_ingress_active(skb)) {
3925 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3930 ingress_retval = nf_hook_ingress(skb);
3932 return ingress_retval;
3934 #endif /* CONFIG_NETFILTER_INGRESS */
3938 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3940 struct packet_type *ptype, *pt_prev;
3941 rx_handler_func_t *rx_handler;
3942 struct net_device *orig_dev;
3943 bool deliver_exact = false;
3944 int ret = NET_RX_DROP;
3947 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3949 trace_netif_receive_skb(skb);
3951 orig_dev = skb->dev;
3953 skb_reset_network_header(skb);
3954 if (!skb_transport_header_was_set(skb))
3955 skb_reset_transport_header(skb);
3956 skb_reset_mac_len(skb);
3961 skb->skb_iif = skb->dev->ifindex;
3963 __this_cpu_inc(softnet_data.processed);
3965 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3966 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3967 skb = skb_vlan_untag(skb);
3972 #ifdef CONFIG_NET_CLS_ACT
3973 if (skb->tc_verd & TC_NCLS) {
3974 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3982 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3984 ret = deliver_skb(skb, pt_prev, orig_dev);
3988 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
3990 ret = deliver_skb(skb, pt_prev, orig_dev);
3995 #ifdef CONFIG_NET_INGRESS
3996 if (static_key_false(&ingress_needed)) {
3997 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev);
4001 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
4005 #ifdef CONFIG_NET_CLS_ACT
4009 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
4012 if (skb_vlan_tag_present(skb)) {
4014 ret = deliver_skb(skb, pt_prev, orig_dev);
4017 if (vlan_do_receive(&skb))
4019 else if (unlikely(!skb))
4023 rx_handler = rcu_dereference(skb->dev->rx_handler);
4026 ret = deliver_skb(skb, pt_prev, orig_dev);
4029 switch (rx_handler(&skb)) {
4030 case RX_HANDLER_CONSUMED:
4031 ret = NET_RX_SUCCESS;
4033 case RX_HANDLER_ANOTHER:
4035 case RX_HANDLER_EXACT:
4036 deliver_exact = true;
4037 case RX_HANDLER_PASS:
4044 if (unlikely(skb_vlan_tag_present(skb))) {
4045 if (skb_vlan_tag_get_id(skb))
4046 skb->pkt_type = PACKET_OTHERHOST;
4047 /* Note: we might in the future use prio bits
4048 * and set skb->priority like in vlan_do_receive()
4049 * For the time being, just ignore Priority Code Point
4054 type = skb->protocol;
4056 /* deliver only exact match when indicated */
4057 if (likely(!deliver_exact)) {
4058 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4059 &ptype_base[ntohs(type) &
4063 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4064 &orig_dev->ptype_specific);
4066 if (unlikely(skb->dev != orig_dev)) {
4067 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4068 &skb->dev->ptype_specific);
4072 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
4075 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
4079 atomic_long_inc(&skb->dev->rx_dropped);
4081 atomic_long_inc(&skb->dev->rx_nohandler);
4083 /* Jamal, now you will not able to escape explaining
4084 * me how you were going to use this. :-)
4093 static int __netif_receive_skb(struct sk_buff *skb)
4097 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
4098 unsigned long pflags = current->flags;
4101 * PFMEMALLOC skbs are special, they should
4102 * - be delivered to SOCK_MEMALLOC sockets only
4103 * - stay away from userspace
4104 * - have bounded memory usage
4106 * Use PF_MEMALLOC as this saves us from propagating the allocation
4107 * context down to all allocation sites.
4109 current->flags |= PF_MEMALLOC;
4110 ret = __netif_receive_skb_core(skb, true);
4111 tsk_restore_flags(current, pflags, PF_MEMALLOC);
4113 ret = __netif_receive_skb_core(skb, false);
4118 static int netif_receive_skb_internal(struct sk_buff *skb)
4122 net_timestamp_check(netdev_tstamp_prequeue, skb);
4124 if (skb_defer_rx_timestamp(skb))
4125 return NET_RX_SUCCESS;
4130 if (static_key_false(&rps_needed)) {
4131 struct rps_dev_flow voidflow, *rflow = &voidflow;
4132 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
4135 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4141 ret = __netif_receive_skb(skb);
4147 * netif_receive_skb - process receive buffer from network
4148 * @skb: buffer to process
4150 * netif_receive_skb() is the main receive data processing function.
4151 * It always succeeds. The buffer may be dropped during processing
4152 * for congestion control or by the protocol layers.
4154 * This function may only be called from softirq context and interrupts
4155 * should be enabled.
4157 * Return values (usually ignored):
4158 * NET_RX_SUCCESS: no congestion
4159 * NET_RX_DROP: packet was dropped
4161 int netif_receive_skb(struct sk_buff *skb)
4163 trace_netif_receive_skb_entry(skb);
4165 return netif_receive_skb_internal(skb);
4167 EXPORT_SYMBOL(netif_receive_skb);
4169 DEFINE_PER_CPU(struct work_struct, flush_works);
4171 /* Network device is going away, flush any packets still pending */
4172 static void flush_backlog(struct work_struct *work)
4174 struct sk_buff *skb, *tmp;
4175 struct softnet_data *sd;
4178 sd = this_cpu_ptr(&softnet_data);
4180 local_irq_disable();
4182 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
4183 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
4184 __skb_unlink(skb, &sd->input_pkt_queue);
4186 input_queue_head_incr(sd);
4192 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
4193 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
4194 __skb_unlink(skb, &sd->process_queue);
4196 input_queue_head_incr(sd);
4202 static void flush_all_backlogs(void)
4208 for_each_online_cpu(cpu)
4209 queue_work_on(cpu, system_highpri_wq,
4210 per_cpu_ptr(&flush_works, cpu));
4212 for_each_online_cpu(cpu)
4213 flush_work(per_cpu_ptr(&flush_works, cpu));
4218 static int napi_gro_complete(struct sk_buff *skb)
4220 struct packet_offload *ptype;
4221 __be16 type = skb->protocol;
4222 struct list_head *head = &offload_base;
4225 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
4227 if (NAPI_GRO_CB(skb)->count == 1) {
4228 skb_shinfo(skb)->gso_size = 0;
4233 list_for_each_entry_rcu(ptype, head, list) {
4234 if (ptype->type != type || !ptype->callbacks.gro_complete)
4237 err = ptype->callbacks.gro_complete(skb, 0);
4243 WARN_ON(&ptype->list == head);
4245 return NET_RX_SUCCESS;
4249 return netif_receive_skb_internal(skb);
4252 /* napi->gro_list contains packets ordered by age.
4253 * youngest packets at the head of it.
4254 * Complete skbs in reverse order to reduce latencies.
4256 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
4258 struct sk_buff *skb, *prev = NULL;
4260 /* scan list and build reverse chain */
4261 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
4266 for (skb = prev; skb; skb = prev) {
4269 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
4273 napi_gro_complete(skb);
4277 napi->gro_list = NULL;
4279 EXPORT_SYMBOL(napi_gro_flush);
4281 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
4284 unsigned int maclen = skb->dev->hard_header_len;
4285 u32 hash = skb_get_hash_raw(skb);
4287 for (p = napi->gro_list; p; p = p->next) {
4288 unsigned long diffs;
4290 NAPI_GRO_CB(p)->flush = 0;
4292 if (hash != skb_get_hash_raw(p)) {
4293 NAPI_GRO_CB(p)->same_flow = 0;
4297 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
4298 diffs |= p->vlan_tci ^ skb->vlan_tci;
4299 diffs |= skb_metadata_dst_cmp(p, skb);
4300 if (maclen == ETH_HLEN)
4301 diffs |= compare_ether_header(skb_mac_header(p),
4302 skb_mac_header(skb));
4304 diffs = memcmp(skb_mac_header(p),
4305 skb_mac_header(skb),
4307 NAPI_GRO_CB(p)->same_flow = !diffs;
4311 static void skb_gro_reset_offset(struct sk_buff *skb)
4313 const struct skb_shared_info *pinfo = skb_shinfo(skb);
4314 const skb_frag_t *frag0 = &pinfo->frags[0];
4316 NAPI_GRO_CB(skb)->data_offset = 0;
4317 NAPI_GRO_CB(skb)->frag0 = NULL;
4318 NAPI_GRO_CB(skb)->frag0_len = 0;
4320 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
4322 !PageHighMem(skb_frag_page(frag0))) {
4323 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
4324 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
4328 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
4330 struct skb_shared_info *pinfo = skb_shinfo(skb);
4332 BUG_ON(skb->end - skb->tail < grow);
4334 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
4336 skb->data_len -= grow;
4339 pinfo->frags[0].page_offset += grow;
4340 skb_frag_size_sub(&pinfo->frags[0], grow);
4342 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
4343 skb_frag_unref(skb, 0);
4344 memmove(pinfo->frags, pinfo->frags + 1,
4345 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
4349 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4351 struct sk_buff **pp = NULL;
4352 struct packet_offload *ptype;
4353 __be16 type = skb->protocol;
4354 struct list_head *head = &offload_base;
4356 enum gro_result ret;
4359 if (!(skb->dev->features & NETIF_F_GRO))
4362 if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad)
4365 gro_list_prepare(napi, skb);
4368 list_for_each_entry_rcu(ptype, head, list) {
4369 if (ptype->type != type || !ptype->callbacks.gro_receive)
4372 skb_set_network_header(skb, skb_gro_offset(skb));
4373 skb_reset_mac_len(skb);
4374 NAPI_GRO_CB(skb)->same_flow = 0;
4375 NAPI_GRO_CB(skb)->flush = 0;
4376 NAPI_GRO_CB(skb)->free = 0;
4377 NAPI_GRO_CB(skb)->encap_mark = 0;
4378 NAPI_GRO_CB(skb)->is_fou = 0;
4379 NAPI_GRO_CB(skb)->is_atomic = 1;
4380 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
4382 /* Setup for GRO checksum validation */
4383 switch (skb->ip_summed) {
4384 case CHECKSUM_COMPLETE:
4385 NAPI_GRO_CB(skb)->csum = skb->csum;
4386 NAPI_GRO_CB(skb)->csum_valid = 1;
4387 NAPI_GRO_CB(skb)->csum_cnt = 0;
4389 case CHECKSUM_UNNECESSARY:
4390 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4391 NAPI_GRO_CB(skb)->csum_valid = 0;
4394 NAPI_GRO_CB(skb)->csum_cnt = 0;
4395 NAPI_GRO_CB(skb)->csum_valid = 0;
4398 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4403 if (&ptype->list == head)
4406 same_flow = NAPI_GRO_CB(skb)->same_flow;
4407 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4410 struct sk_buff *nskb = *pp;
4414 napi_gro_complete(nskb);
4421 if (NAPI_GRO_CB(skb)->flush)
4424 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4425 struct sk_buff *nskb = napi->gro_list;
4427 /* locate the end of the list to select the 'oldest' flow */
4428 while (nskb->next) {
4434 napi_gro_complete(nskb);
4438 NAPI_GRO_CB(skb)->count = 1;
4439 NAPI_GRO_CB(skb)->age = jiffies;
4440 NAPI_GRO_CB(skb)->last = skb;
4441 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4442 skb->next = napi->gro_list;
4443 napi->gro_list = skb;
4447 grow = skb_gro_offset(skb) - skb_headlen(skb);
4449 gro_pull_from_frag0(skb, grow);
4458 struct packet_offload *gro_find_receive_by_type(__be16 type)
4460 struct list_head *offload_head = &offload_base;
4461 struct packet_offload *ptype;
4463 list_for_each_entry_rcu(ptype, offload_head, list) {
4464 if (ptype->type != type || !ptype->callbacks.gro_receive)
4470 EXPORT_SYMBOL(gro_find_receive_by_type);
4472 struct packet_offload *gro_find_complete_by_type(__be16 type)
4474 struct list_head *offload_head = &offload_base;
4475 struct packet_offload *ptype;
4477 list_for_each_entry_rcu(ptype, offload_head, list) {
4478 if (ptype->type != type || !ptype->callbacks.gro_complete)
4484 EXPORT_SYMBOL(gro_find_complete_by_type);
4486 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4490 if (netif_receive_skb_internal(skb))
4498 case GRO_MERGED_FREE:
4499 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) {
4501 kmem_cache_free(skbuff_head_cache, skb);
4515 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4517 skb_mark_napi_id(skb, napi);
4518 trace_napi_gro_receive_entry(skb);
4520 skb_gro_reset_offset(skb);
4522 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4524 EXPORT_SYMBOL(napi_gro_receive);
4526 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4528 if (unlikely(skb->pfmemalloc)) {
4532 __skb_pull(skb, skb_headlen(skb));
4533 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4534 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4536 skb->dev = napi->dev;
4538 skb->encapsulation = 0;
4539 skb_shinfo(skb)->gso_type = 0;
4540 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4545 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4547 struct sk_buff *skb = napi->skb;
4550 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
4553 skb_mark_napi_id(skb, napi);
4558 EXPORT_SYMBOL(napi_get_frags);
4560 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4561 struct sk_buff *skb,
4567 __skb_push(skb, ETH_HLEN);
4568 skb->protocol = eth_type_trans(skb, skb->dev);
4569 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4574 case GRO_MERGED_FREE:
4575 napi_reuse_skb(napi, skb);
4585 /* Upper GRO stack assumes network header starts at gro_offset=0
4586 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4587 * We copy ethernet header into skb->data to have a common layout.
4589 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4591 struct sk_buff *skb = napi->skb;
4592 const struct ethhdr *eth;
4593 unsigned int hlen = sizeof(*eth);
4597 skb_reset_mac_header(skb);
4598 skb_gro_reset_offset(skb);
4600 eth = skb_gro_header_fast(skb, 0);
4601 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4602 eth = skb_gro_header_slow(skb, hlen, 0);
4603 if (unlikely(!eth)) {
4604 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
4605 __func__, napi->dev->name);
4606 napi_reuse_skb(napi, skb);
4610 gro_pull_from_frag0(skb, hlen);
4611 NAPI_GRO_CB(skb)->frag0 += hlen;
4612 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4614 __skb_pull(skb, hlen);
4617 * This works because the only protocols we care about don't require
4619 * We'll fix it up properly in napi_frags_finish()
4621 skb->protocol = eth->h_proto;
4626 gro_result_t napi_gro_frags(struct napi_struct *napi)
4628 struct sk_buff *skb = napi_frags_skb(napi);
4633 trace_napi_gro_frags_entry(skb);
4635 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4637 EXPORT_SYMBOL(napi_gro_frags);
4639 /* Compute the checksum from gro_offset and return the folded value
4640 * after adding in any pseudo checksum.
4642 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
4647 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
4649 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4650 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
4652 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
4653 !skb->csum_complete_sw)
4654 netdev_rx_csum_fault(skb->dev);
4657 NAPI_GRO_CB(skb)->csum = wsum;
4658 NAPI_GRO_CB(skb)->csum_valid = 1;
4662 EXPORT_SYMBOL(__skb_gro_checksum_complete);
4665 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4666 * Note: called with local irq disabled, but exits with local irq enabled.
4668 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4671 struct softnet_data *remsd = sd->rps_ipi_list;
4674 sd->rps_ipi_list = NULL;
4678 /* Send pending IPI's to kick RPS processing on remote cpus. */
4680 struct softnet_data *next = remsd->rps_ipi_next;
4682 if (cpu_online(remsd->cpu))
4683 smp_call_function_single_async(remsd->cpu,
4692 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
4695 return sd->rps_ipi_list != NULL;
4701 static int process_backlog(struct napi_struct *napi, int quota)
4703 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4707 /* Check if we have pending ipi, its better to send them now,
4708 * not waiting net_rx_action() end.
4710 if (sd_has_rps_ipi_waiting(sd)) {
4711 local_irq_disable();
4712 net_rps_action_and_irq_enable(sd);
4715 napi->weight = weight_p;
4717 struct sk_buff *skb;
4719 while ((skb = __skb_dequeue(&sd->process_queue))) {
4721 __netif_receive_skb(skb);
4723 input_queue_head_incr(sd);
4724 if (++work >= quota)
4729 local_irq_disable();
4731 if (skb_queue_empty(&sd->input_pkt_queue)) {
4733 * Inline a custom version of __napi_complete().
4734 * only current cpu owns and manipulates this napi,
4735 * and NAPI_STATE_SCHED is the only possible flag set
4737 * We can use a plain write instead of clear_bit(),
4738 * and we dont need an smp_mb() memory barrier.
4743 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4744 &sd->process_queue);
4754 * __napi_schedule - schedule for receive
4755 * @n: entry to schedule
4757 * The entry's receive function will be scheduled to run.
4758 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
4760 void __napi_schedule(struct napi_struct *n)
4762 unsigned long flags;
4764 local_irq_save(flags);
4765 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4766 local_irq_restore(flags);
4768 EXPORT_SYMBOL(__napi_schedule);
4771 * __napi_schedule_irqoff - schedule for receive
4772 * @n: entry to schedule
4774 * Variant of __napi_schedule() assuming hard irqs are masked
4776 void __napi_schedule_irqoff(struct napi_struct *n)
4778 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4780 EXPORT_SYMBOL(__napi_schedule_irqoff);
4782 void __napi_complete(struct napi_struct *n)
4784 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4786 list_del_init(&n->poll_list);
4787 smp_mb__before_atomic();
4788 clear_bit(NAPI_STATE_SCHED, &n->state);
4790 EXPORT_SYMBOL(__napi_complete);
4792 void napi_complete_done(struct napi_struct *n, int work_done)
4794 unsigned long flags;
4797 * don't let napi dequeue from the cpu poll list
4798 * just in case its running on a different cpu
4800 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4804 unsigned long timeout = 0;
4807 timeout = n->dev->gro_flush_timeout;
4810 hrtimer_start(&n->timer, ns_to_ktime(timeout),
4811 HRTIMER_MODE_REL_PINNED);
4813 napi_gro_flush(n, false);
4815 if (likely(list_empty(&n->poll_list))) {
4816 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state));
4818 /* If n->poll_list is not empty, we need to mask irqs */
4819 local_irq_save(flags);
4821 local_irq_restore(flags);
4824 EXPORT_SYMBOL(napi_complete_done);
4826 /* must be called under rcu_read_lock(), as we dont take a reference */
4827 static struct napi_struct *napi_by_id(unsigned int napi_id)
4829 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4830 struct napi_struct *napi;
4832 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4833 if (napi->napi_id == napi_id)
4839 #if defined(CONFIG_NET_RX_BUSY_POLL)
4840 #define BUSY_POLL_BUDGET 8
4841 bool sk_busy_loop(struct sock *sk, int nonblock)
4843 unsigned long end_time = !nonblock ? sk_busy_loop_end_time(sk) : 0;
4844 int (*busy_poll)(struct napi_struct *dev);
4845 struct napi_struct *napi;
4850 napi = napi_by_id(sk->sk_napi_id);
4854 /* Note: ndo_busy_poll method is optional in linux-4.5 */
4855 busy_poll = napi->dev->netdev_ops->ndo_busy_poll;
4861 rc = busy_poll(napi);
4862 } else if (napi_schedule_prep(napi)) {
4863 void *have = netpoll_poll_lock(napi);
4865 if (test_bit(NAPI_STATE_SCHED, &napi->state)) {
4866 rc = napi->poll(napi, BUSY_POLL_BUDGET);
4867 trace_napi_poll(napi, rc, BUSY_POLL_BUDGET);
4868 if (rc == BUSY_POLL_BUDGET) {
4869 napi_complete_done(napi, rc);
4870 napi_schedule(napi);
4873 netpoll_poll_unlock(have);
4876 __NET_ADD_STATS(sock_net(sk),
4877 LINUX_MIB_BUSYPOLLRXPACKETS, rc);
4880 if (rc == LL_FLUSH_FAILED)
4881 break; /* permanent failure */
4884 } while (!nonblock && skb_queue_empty(&sk->sk_receive_queue) &&
4885 !need_resched() && !busy_loop_timeout(end_time));
4887 rc = !skb_queue_empty(&sk->sk_receive_queue);
4892 EXPORT_SYMBOL(sk_busy_loop);
4894 #endif /* CONFIG_NET_RX_BUSY_POLL */
4896 void napi_hash_add(struct napi_struct *napi)
4898 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) ||
4899 test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
4902 spin_lock(&napi_hash_lock);
4904 /* 0..NR_CPUS+1 range is reserved for sender_cpu use */
4906 if (unlikely(++napi_gen_id < NR_CPUS + 1))
4907 napi_gen_id = NR_CPUS + 1;
4908 } while (napi_by_id(napi_gen_id));
4909 napi->napi_id = napi_gen_id;
4911 hlist_add_head_rcu(&napi->napi_hash_node,
4912 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4914 spin_unlock(&napi_hash_lock);
4916 EXPORT_SYMBOL_GPL(napi_hash_add);
4918 /* Warning : caller is responsible to make sure rcu grace period
4919 * is respected before freeing memory containing @napi
4921 bool napi_hash_del(struct napi_struct *napi)
4923 bool rcu_sync_needed = false;
4925 spin_lock(&napi_hash_lock);
4927 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) {
4928 rcu_sync_needed = true;
4929 hlist_del_rcu(&napi->napi_hash_node);
4931 spin_unlock(&napi_hash_lock);
4932 return rcu_sync_needed;
4934 EXPORT_SYMBOL_GPL(napi_hash_del);
4936 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
4938 struct napi_struct *napi;
4940 napi = container_of(timer, struct napi_struct, timer);
4942 napi_schedule(napi);
4944 return HRTIMER_NORESTART;
4947 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4948 int (*poll)(struct napi_struct *, int), int weight)
4950 INIT_LIST_HEAD(&napi->poll_list);
4951 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
4952 napi->timer.function = napi_watchdog;
4953 napi->gro_count = 0;
4954 napi->gro_list = NULL;
4957 if (weight > NAPI_POLL_WEIGHT)
4958 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4960 napi->weight = weight;
4961 list_add(&napi->dev_list, &dev->napi_list);
4963 #ifdef CONFIG_NETPOLL
4964 spin_lock_init(&napi->poll_lock);
4965 napi->poll_owner = -1;
4967 set_bit(NAPI_STATE_SCHED, &napi->state);
4968 napi_hash_add(napi);
4970 EXPORT_SYMBOL(netif_napi_add);
4972 void napi_disable(struct napi_struct *n)
4975 set_bit(NAPI_STATE_DISABLE, &n->state);
4977 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
4979 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
4982 hrtimer_cancel(&n->timer);
4984 clear_bit(NAPI_STATE_DISABLE, &n->state);
4986 EXPORT_SYMBOL(napi_disable);
4988 /* Must be called in process context */
4989 void netif_napi_del(struct napi_struct *napi)
4992 if (napi_hash_del(napi))
4994 list_del_init(&napi->dev_list);
4995 napi_free_frags(napi);
4997 kfree_skb_list(napi->gro_list);
4998 napi->gro_list = NULL;
4999 napi->gro_count = 0;
5001 EXPORT_SYMBOL(netif_napi_del);
5003 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
5008 list_del_init(&n->poll_list);
5010 have = netpoll_poll_lock(n);
5014 /* This NAPI_STATE_SCHED test is for avoiding a race
5015 * with netpoll's poll_napi(). Only the entity which
5016 * obtains the lock and sees NAPI_STATE_SCHED set will
5017 * actually make the ->poll() call. Therefore we avoid
5018 * accidentally calling ->poll() when NAPI is not scheduled.
5021 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
5022 work = n->poll(n, weight);
5023 trace_napi_poll(n, work, weight);
5026 WARN_ON_ONCE(work > weight);
5028 if (likely(work < weight))
5031 /* Drivers must not modify the NAPI state if they
5032 * consume the entire weight. In such cases this code
5033 * still "owns" the NAPI instance and therefore can
5034 * move the instance around on the list at-will.
5036 if (unlikely(napi_disable_pending(n))) {
5042 /* flush too old packets
5043 * If HZ < 1000, flush all packets.
5045 napi_gro_flush(n, HZ >= 1000);
5048 /* Some drivers may have called napi_schedule
5049 * prior to exhausting their budget.
5051 if (unlikely(!list_empty(&n->poll_list))) {
5052 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
5053 n->dev ? n->dev->name : "backlog");
5057 list_add_tail(&n->poll_list, repoll);
5060 netpoll_poll_unlock(have);
5065 static void net_rx_action(struct softirq_action *h)
5067 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
5068 unsigned long time_limit = jiffies + 2;
5069 int budget = netdev_budget;
5073 local_irq_disable();
5074 list_splice_init(&sd->poll_list, &list);
5078 struct napi_struct *n;
5080 if (list_empty(&list)) {
5081 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
5086 n = list_first_entry(&list, struct napi_struct, poll_list);
5087 budget -= napi_poll(n, &repoll);
5089 /* If softirq window is exhausted then punt.
5090 * Allow this to run for 2 jiffies since which will allow
5091 * an average latency of 1.5/HZ.
5093 if (unlikely(budget <= 0 ||
5094 time_after_eq(jiffies, time_limit))) {
5100 __kfree_skb_flush();
5101 local_irq_disable();
5103 list_splice_tail_init(&sd->poll_list, &list);
5104 list_splice_tail(&repoll, &list);
5105 list_splice(&list, &sd->poll_list);
5106 if (!list_empty(&sd->poll_list))
5107 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
5109 net_rps_action_and_irq_enable(sd);
5112 struct netdev_adjacent {
5113 struct net_device *dev;
5115 /* upper master flag, there can only be one master device per list */
5118 /* counter for the number of times this device was added to us */
5121 /* private field for the users */
5124 struct list_head list;
5125 struct rcu_head rcu;
5128 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
5129 struct list_head *adj_list)
5131 struct netdev_adjacent *adj;
5133 list_for_each_entry(adj, adj_list, list) {
5134 if (adj->dev == adj_dev)
5140 static int __netdev_has_upper_dev(struct net_device *upper_dev, void *data)
5142 struct net_device *dev = data;
5144 return upper_dev == dev;
5148 * netdev_has_upper_dev - Check if device is linked to an upper device
5150 * @upper_dev: upper device to check
5152 * Find out if a device is linked to specified upper device and return true
5153 * in case it is. Note that this checks only immediate upper device,
5154 * not through a complete stack of devices. The caller must hold the RTNL lock.
5156 bool netdev_has_upper_dev(struct net_device *dev,
5157 struct net_device *upper_dev)
5161 return netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
5164 EXPORT_SYMBOL(netdev_has_upper_dev);
5167 * netdev_has_upper_dev_all - Check if device is linked to an upper device
5169 * @upper_dev: upper device to check
5171 * Find out if a device is linked to specified upper device and return true
5172 * in case it is. Note that this checks the entire upper device chain.
5173 * The caller must hold rcu lock.
5176 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
5177 struct net_device *upper_dev)
5179 return !!netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
5182 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
5185 * netdev_has_any_upper_dev - Check if device is linked to some device
5188 * Find out if a device is linked to an upper device and return true in case
5189 * it is. The caller must hold the RTNL lock.
5191 static bool netdev_has_any_upper_dev(struct net_device *dev)
5195 return !list_empty(&dev->adj_list.upper);
5199 * netdev_master_upper_dev_get - Get master upper device
5202 * Find a master upper device and return pointer to it or NULL in case
5203 * it's not there. The caller must hold the RTNL lock.
5205 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
5207 struct netdev_adjacent *upper;
5211 if (list_empty(&dev->adj_list.upper))
5214 upper = list_first_entry(&dev->adj_list.upper,
5215 struct netdev_adjacent, list);
5216 if (likely(upper->master))
5220 EXPORT_SYMBOL(netdev_master_upper_dev_get);
5223 * netdev_has_any_lower_dev - Check if device is linked to some device
5226 * Find out if a device is linked to a lower device and return true in case
5227 * it is. The caller must hold the RTNL lock.
5229 static bool netdev_has_any_lower_dev(struct net_device *dev)
5233 return !list_empty(&dev->adj_list.lower);
5236 void *netdev_adjacent_get_private(struct list_head *adj_list)
5238 struct netdev_adjacent *adj;
5240 adj = list_entry(adj_list, struct netdev_adjacent, list);
5242 return adj->private;
5244 EXPORT_SYMBOL(netdev_adjacent_get_private);
5247 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
5249 * @iter: list_head ** of the current position
5251 * Gets the next device from the dev's upper list, starting from iter
5252 * position. The caller must hold RCU read lock.
5254 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
5255 struct list_head **iter)
5257 struct netdev_adjacent *upper;
5259 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5261 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5263 if (&upper->list == &dev->adj_list.upper)
5266 *iter = &upper->list;
5270 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
5272 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
5273 struct list_head **iter)
5275 struct netdev_adjacent *upper;
5277 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5279 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5281 if (&upper->list == &dev->adj_list.upper)
5284 *iter = &upper->list;
5289 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
5290 int (*fn)(struct net_device *dev,
5294 struct net_device *udev;
5295 struct list_head *iter;
5298 for (iter = &dev->adj_list.upper,
5299 udev = netdev_next_upper_dev_rcu(dev, &iter);
5301 udev = netdev_next_upper_dev_rcu(dev, &iter)) {
5302 /* first is the upper device itself */
5303 ret = fn(udev, data);
5307 /* then look at all of its upper devices */
5308 ret = netdev_walk_all_upper_dev_rcu(udev, fn, data);
5315 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
5318 * netdev_lower_get_next_private - Get the next ->private from the
5319 * lower neighbour list
5321 * @iter: list_head ** of the current position
5323 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5324 * list, starting from iter position. The caller must hold either hold the
5325 * RTNL lock or its own locking that guarantees that the neighbour lower
5326 * list will remain unchanged.
5328 void *netdev_lower_get_next_private(struct net_device *dev,
5329 struct list_head **iter)
5331 struct netdev_adjacent *lower;
5333 lower = list_entry(*iter, struct netdev_adjacent, list);
5335 if (&lower->list == &dev->adj_list.lower)
5338 *iter = lower->list.next;
5340 return lower->private;
5342 EXPORT_SYMBOL(netdev_lower_get_next_private);
5345 * netdev_lower_get_next_private_rcu - Get the next ->private from the
5346 * lower neighbour list, RCU
5349 * @iter: list_head ** of the current position
5351 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5352 * list, starting from iter position. The caller must hold RCU read lock.
5354 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
5355 struct list_head **iter)
5357 struct netdev_adjacent *lower;
5359 WARN_ON_ONCE(!rcu_read_lock_held());
5361 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5363 if (&lower->list == &dev->adj_list.lower)
5366 *iter = &lower->list;
5368 return lower->private;
5370 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
5373 * netdev_lower_get_next - Get the next device from the lower neighbour
5376 * @iter: list_head ** of the current position
5378 * Gets the next netdev_adjacent from the dev's lower neighbour
5379 * list, starting from iter position. The caller must hold RTNL lock or
5380 * its own locking that guarantees that the neighbour lower
5381 * list will remain unchanged.
5383 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
5385 struct netdev_adjacent *lower;
5387 lower = list_entry(*iter, struct netdev_adjacent, list);
5389 if (&lower->list == &dev->adj_list.lower)
5392 *iter = lower->list.next;
5396 EXPORT_SYMBOL(netdev_lower_get_next);
5398 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
5399 struct list_head **iter)
5401 struct netdev_adjacent *lower;
5403 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
5405 if (&lower->list == &dev->adj_list.lower)
5408 *iter = &lower->list;
5413 int netdev_walk_all_lower_dev(struct net_device *dev,
5414 int (*fn)(struct net_device *dev,
5418 struct net_device *ldev;
5419 struct list_head *iter;
5422 for (iter = &dev->adj_list.lower,
5423 ldev = netdev_next_lower_dev(dev, &iter);
5425 ldev = netdev_next_lower_dev(dev, &iter)) {
5426 /* first is the lower device itself */
5427 ret = fn(ldev, data);
5431 /* then look at all of its lower devices */
5432 ret = netdev_walk_all_lower_dev(ldev, fn, data);
5439 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
5441 static struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
5442 struct list_head **iter)
5444 struct netdev_adjacent *lower;
5446 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5447 if (&lower->list == &dev->adj_list.lower)
5450 *iter = &lower->list;
5455 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
5456 int (*fn)(struct net_device *dev,
5460 struct net_device *ldev;
5461 struct list_head *iter;
5464 for (iter = &dev->adj_list.lower,
5465 ldev = netdev_next_lower_dev_rcu(dev, &iter);
5467 ldev = netdev_next_lower_dev_rcu(dev, &iter)) {
5468 /* first is the lower device itself */
5469 ret = fn(ldev, data);
5473 /* then look at all of its lower devices */
5474 ret = netdev_walk_all_lower_dev_rcu(ldev, fn, data);
5481 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
5484 * netdev_lower_get_first_private_rcu - Get the first ->private from the
5485 * lower neighbour list, RCU
5489 * Gets the first netdev_adjacent->private from the dev's lower neighbour
5490 * list. The caller must hold RCU read lock.
5492 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
5494 struct netdev_adjacent *lower;
5496 lower = list_first_or_null_rcu(&dev->adj_list.lower,
5497 struct netdev_adjacent, list);
5499 return lower->private;
5502 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
5505 * netdev_master_upper_dev_get_rcu - Get master upper device
5508 * Find a master upper device and return pointer to it or NULL in case
5509 * it's not there. The caller must hold the RCU read lock.
5511 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
5513 struct netdev_adjacent *upper;
5515 upper = list_first_or_null_rcu(&dev->adj_list.upper,
5516 struct netdev_adjacent, list);
5517 if (upper && likely(upper->master))
5521 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
5523 static int netdev_adjacent_sysfs_add(struct net_device *dev,
5524 struct net_device *adj_dev,
5525 struct list_head *dev_list)
5527 char linkname[IFNAMSIZ+7];
5528 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5529 "upper_%s" : "lower_%s", adj_dev->name);
5530 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
5533 static void netdev_adjacent_sysfs_del(struct net_device *dev,
5535 struct list_head *dev_list)
5537 char linkname[IFNAMSIZ+7];
5538 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5539 "upper_%s" : "lower_%s", name);
5540 sysfs_remove_link(&(dev->dev.kobj), linkname);
5543 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
5544 struct net_device *adj_dev,
5545 struct list_head *dev_list)
5547 return (dev_list == &dev->adj_list.upper ||
5548 dev_list == &dev->adj_list.lower) &&
5549 net_eq(dev_net(dev), dev_net(adj_dev));
5552 static int __netdev_adjacent_dev_insert(struct net_device *dev,
5553 struct net_device *adj_dev,
5554 struct list_head *dev_list,
5555 void *private, bool master)
5557 struct netdev_adjacent *adj;
5560 adj = __netdev_find_adj(adj_dev, dev_list);
5564 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
5565 dev->name, adj_dev->name, adj->ref_nr);
5570 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
5575 adj->master = master;
5577 adj->private = private;
5580 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
5581 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
5583 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
5584 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
5589 /* Ensure that master link is always the first item in list. */
5591 ret = sysfs_create_link(&(dev->dev.kobj),
5592 &(adj_dev->dev.kobj), "master");
5594 goto remove_symlinks;
5596 list_add_rcu(&adj->list, dev_list);
5598 list_add_tail_rcu(&adj->list, dev_list);
5604 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5605 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5613 static void __netdev_adjacent_dev_remove(struct net_device *dev,
5614 struct net_device *adj_dev,
5616 struct list_head *dev_list)
5618 struct netdev_adjacent *adj;
5620 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
5621 dev->name, adj_dev->name, ref_nr);
5623 adj = __netdev_find_adj(adj_dev, dev_list);
5626 pr_err("Adjacency does not exist for device %s from %s\n",
5627 dev->name, adj_dev->name);
5632 if (adj->ref_nr > ref_nr) {
5633 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
5634 dev->name, adj_dev->name, ref_nr,
5635 adj->ref_nr - ref_nr);
5636 adj->ref_nr -= ref_nr;
5641 sysfs_remove_link(&(dev->dev.kobj), "master");
5643 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5644 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5646 list_del_rcu(&adj->list);
5647 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
5648 adj_dev->name, dev->name, adj_dev->name);
5650 kfree_rcu(adj, rcu);
5653 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
5654 struct net_device *upper_dev,
5655 struct list_head *up_list,
5656 struct list_head *down_list,
5657 void *private, bool master)
5661 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
5666 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
5669 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
5676 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
5677 struct net_device *upper_dev,
5679 struct list_head *up_list,
5680 struct list_head *down_list)
5682 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
5683 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
5686 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
5687 struct net_device *upper_dev,
5688 void *private, bool master)
5690 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
5691 &dev->adj_list.upper,
5692 &upper_dev->adj_list.lower,
5696 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
5697 struct net_device *upper_dev)
5699 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
5700 &dev->adj_list.upper,
5701 &upper_dev->adj_list.lower);
5704 static int __netdev_upper_dev_link(struct net_device *dev,
5705 struct net_device *upper_dev, bool master,
5706 void *upper_priv, void *upper_info)
5708 struct netdev_notifier_changeupper_info changeupper_info;
5713 if (dev == upper_dev)
5716 /* To prevent loops, check if dev is not upper device to upper_dev. */
5717 if (netdev_has_upper_dev(upper_dev, dev))
5720 if (netdev_has_upper_dev(dev, upper_dev))
5723 if (master && netdev_master_upper_dev_get(dev))
5726 changeupper_info.upper_dev = upper_dev;
5727 changeupper_info.master = master;
5728 changeupper_info.linking = true;
5729 changeupper_info.upper_info = upper_info;
5731 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
5732 &changeupper_info.info);
5733 ret = notifier_to_errno(ret);
5737 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
5742 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
5743 &changeupper_info.info);
5744 ret = notifier_to_errno(ret);
5751 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5757 * netdev_upper_dev_link - Add a link to the upper device
5759 * @upper_dev: new upper device
5761 * Adds a link to device which is upper to this one. The caller must hold
5762 * the RTNL lock. On a failure a negative errno code is returned.
5763 * On success the reference counts are adjusted and the function
5766 int netdev_upper_dev_link(struct net_device *dev,
5767 struct net_device *upper_dev)
5769 return __netdev_upper_dev_link(dev, upper_dev, false, NULL, NULL);
5771 EXPORT_SYMBOL(netdev_upper_dev_link);
5774 * netdev_master_upper_dev_link - Add a master link to the upper device
5776 * @upper_dev: new upper device
5777 * @upper_priv: upper device private
5778 * @upper_info: upper info to be passed down via notifier
5780 * Adds a link to device which is upper to this one. In this case, only
5781 * one master upper device can be linked, although other non-master devices
5782 * might be linked as well. The caller must hold the RTNL lock.
5783 * On a failure a negative errno code is returned. On success the reference
5784 * counts are adjusted and the function returns zero.
5786 int netdev_master_upper_dev_link(struct net_device *dev,
5787 struct net_device *upper_dev,
5788 void *upper_priv, void *upper_info)
5790 return __netdev_upper_dev_link(dev, upper_dev, true,
5791 upper_priv, upper_info);
5793 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5796 * netdev_upper_dev_unlink - Removes a link to upper device
5798 * @upper_dev: new upper device
5800 * Removes a link to device which is upper to this one. The caller must hold
5803 void netdev_upper_dev_unlink(struct net_device *dev,
5804 struct net_device *upper_dev)
5806 struct netdev_notifier_changeupper_info changeupper_info;
5809 changeupper_info.upper_dev = upper_dev;
5810 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
5811 changeupper_info.linking = false;
5813 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
5814 &changeupper_info.info);
5816 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5818 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
5819 &changeupper_info.info);
5821 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5824 * netdev_bonding_info_change - Dispatch event about slave change
5826 * @bonding_info: info to dispatch
5828 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
5829 * The caller must hold the RTNL lock.
5831 void netdev_bonding_info_change(struct net_device *dev,
5832 struct netdev_bonding_info *bonding_info)
5834 struct netdev_notifier_bonding_info info;
5836 memcpy(&info.bonding_info, bonding_info,
5837 sizeof(struct netdev_bonding_info));
5838 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev,
5841 EXPORT_SYMBOL(netdev_bonding_info_change);
5843 static void netdev_adjacent_add_links(struct net_device *dev)
5845 struct netdev_adjacent *iter;
5847 struct net *net = dev_net(dev);
5849 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5850 if (!net_eq(net, dev_net(iter->dev)))
5852 netdev_adjacent_sysfs_add(iter->dev, dev,
5853 &iter->dev->adj_list.lower);
5854 netdev_adjacent_sysfs_add(dev, iter->dev,
5855 &dev->adj_list.upper);
5858 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5859 if (!net_eq(net, dev_net(iter->dev)))
5861 netdev_adjacent_sysfs_add(iter->dev, dev,
5862 &iter->dev->adj_list.upper);
5863 netdev_adjacent_sysfs_add(dev, iter->dev,
5864 &dev->adj_list.lower);
5868 static void netdev_adjacent_del_links(struct net_device *dev)
5870 struct netdev_adjacent *iter;
5872 struct net *net = dev_net(dev);
5874 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5875 if (!net_eq(net, dev_net(iter->dev)))
5877 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5878 &iter->dev->adj_list.lower);
5879 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5880 &dev->adj_list.upper);
5883 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5884 if (!net_eq(net, dev_net(iter->dev)))
5886 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5887 &iter->dev->adj_list.upper);
5888 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5889 &dev->adj_list.lower);
5893 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5895 struct netdev_adjacent *iter;
5897 struct net *net = dev_net(dev);
5899 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5900 if (!net_eq(net, dev_net(iter->dev)))
5902 netdev_adjacent_sysfs_del(iter->dev, oldname,
5903 &iter->dev->adj_list.lower);
5904 netdev_adjacent_sysfs_add(iter->dev, dev,
5905 &iter->dev->adj_list.lower);
5908 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5909 if (!net_eq(net, dev_net(iter->dev)))
5911 netdev_adjacent_sysfs_del(iter->dev, oldname,
5912 &iter->dev->adj_list.upper);
5913 netdev_adjacent_sysfs_add(iter->dev, dev,
5914 &iter->dev->adj_list.upper);
5918 void *netdev_lower_dev_get_private(struct net_device *dev,
5919 struct net_device *lower_dev)
5921 struct netdev_adjacent *lower;
5925 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
5929 return lower->private;
5931 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5934 int dev_get_nest_level(struct net_device *dev)
5936 struct net_device *lower = NULL;
5937 struct list_head *iter;
5943 netdev_for_each_lower_dev(dev, lower, iter) {
5944 nest = dev_get_nest_level(lower);
5945 if (max_nest < nest)
5949 return max_nest + 1;
5951 EXPORT_SYMBOL(dev_get_nest_level);
5954 * netdev_lower_change - Dispatch event about lower device state change
5955 * @lower_dev: device
5956 * @lower_state_info: state to dispatch
5958 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
5959 * The caller must hold the RTNL lock.
5961 void netdev_lower_state_changed(struct net_device *lower_dev,
5962 void *lower_state_info)
5964 struct netdev_notifier_changelowerstate_info changelowerstate_info;
5967 changelowerstate_info.lower_state_info = lower_state_info;
5968 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, lower_dev,
5969 &changelowerstate_info.info);
5971 EXPORT_SYMBOL(netdev_lower_state_changed);
5973 int netdev_default_l2upper_neigh_construct(struct net_device *dev,
5974 struct neighbour *n)
5976 struct net_device *lower_dev, *stop_dev;
5977 struct list_head *iter;
5980 netdev_for_each_lower_dev(dev, lower_dev, iter) {
5981 if (!lower_dev->netdev_ops->ndo_neigh_construct)
5983 err = lower_dev->netdev_ops->ndo_neigh_construct(lower_dev, n);
5985 stop_dev = lower_dev;
5992 netdev_for_each_lower_dev(dev, lower_dev, iter) {
5993 if (lower_dev == stop_dev)
5995 if (!lower_dev->netdev_ops->ndo_neigh_destroy)
5997 lower_dev->netdev_ops->ndo_neigh_destroy(lower_dev, n);
6001 EXPORT_SYMBOL_GPL(netdev_default_l2upper_neigh_construct);
6003 void netdev_default_l2upper_neigh_destroy(struct net_device *dev,
6004 struct neighbour *n)
6006 struct net_device *lower_dev;
6007 struct list_head *iter;
6009 netdev_for_each_lower_dev(dev, lower_dev, iter) {
6010 if (!lower_dev->netdev_ops->ndo_neigh_destroy)
6012 lower_dev->netdev_ops->ndo_neigh_destroy(lower_dev, n);
6015 EXPORT_SYMBOL_GPL(netdev_default_l2upper_neigh_destroy);
6017 static void dev_change_rx_flags(struct net_device *dev, int flags)
6019 const struct net_device_ops *ops = dev->netdev_ops;
6021 if (ops->ndo_change_rx_flags)
6022 ops->ndo_change_rx_flags(dev, flags);
6025 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
6027 unsigned int old_flags = dev->flags;
6033 dev->flags |= IFF_PROMISC;
6034 dev->promiscuity += inc;
6035 if (dev->promiscuity == 0) {
6038 * If inc causes overflow, untouch promisc and return error.
6041 dev->flags &= ~IFF_PROMISC;
6043 dev->promiscuity -= inc;
6044 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
6049 if (dev->flags != old_flags) {
6050 pr_info("device %s %s promiscuous mode\n",
6052 dev->flags & IFF_PROMISC ? "entered" : "left");
6053 if (audit_enabled) {
6054 current_uid_gid(&uid, &gid);
6055 audit_log(current->audit_context, GFP_ATOMIC,
6056 AUDIT_ANOM_PROMISCUOUS,
6057 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
6058 dev->name, (dev->flags & IFF_PROMISC),
6059 (old_flags & IFF_PROMISC),
6060 from_kuid(&init_user_ns, audit_get_loginuid(current)),
6061 from_kuid(&init_user_ns, uid),
6062 from_kgid(&init_user_ns, gid),
6063 audit_get_sessionid(current));
6066 dev_change_rx_flags(dev, IFF_PROMISC);
6069 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
6074 * dev_set_promiscuity - update promiscuity count on a device
6078 * Add or remove promiscuity from a device. While the count in the device
6079 * remains above zero the interface remains promiscuous. Once it hits zero
6080 * the device reverts back to normal filtering operation. A negative inc
6081 * value is used to drop promiscuity on the device.
6082 * Return 0 if successful or a negative errno code on error.
6084 int dev_set_promiscuity(struct net_device *dev, int inc)
6086 unsigned int old_flags = dev->flags;
6089 err = __dev_set_promiscuity(dev, inc, true);
6092 if (dev->flags != old_flags)
6093 dev_set_rx_mode(dev);
6096 EXPORT_SYMBOL(dev_set_promiscuity);
6098 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
6100 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
6104 dev->flags |= IFF_ALLMULTI;
6105 dev->allmulti += inc;
6106 if (dev->allmulti == 0) {
6109 * If inc causes overflow, untouch allmulti and return error.
6112 dev->flags &= ~IFF_ALLMULTI;
6114 dev->allmulti -= inc;
6115 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
6120 if (dev->flags ^ old_flags) {
6121 dev_change_rx_flags(dev, IFF_ALLMULTI);
6122 dev_set_rx_mode(dev);
6124 __dev_notify_flags(dev, old_flags,
6125 dev->gflags ^ old_gflags);
6131 * dev_set_allmulti - update allmulti count on a device
6135 * Add or remove reception of all multicast frames to a device. While the
6136 * count in the device remains above zero the interface remains listening
6137 * to all interfaces. Once it hits zero the device reverts back to normal
6138 * filtering operation. A negative @inc value is used to drop the counter
6139 * when releasing a resource needing all multicasts.
6140 * Return 0 if successful or a negative errno code on error.
6143 int dev_set_allmulti(struct net_device *dev, int inc)
6145 return __dev_set_allmulti(dev, inc, true);
6147 EXPORT_SYMBOL(dev_set_allmulti);
6150 * Upload unicast and multicast address lists to device and
6151 * configure RX filtering. When the device doesn't support unicast
6152 * filtering it is put in promiscuous mode while unicast addresses
6155 void __dev_set_rx_mode(struct net_device *dev)
6157 const struct net_device_ops *ops = dev->netdev_ops;
6159 /* dev_open will call this function so the list will stay sane. */
6160 if (!(dev->flags&IFF_UP))
6163 if (!netif_device_present(dev))
6166 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
6167 /* Unicast addresses changes may only happen under the rtnl,
6168 * therefore calling __dev_set_promiscuity here is safe.
6170 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
6171 __dev_set_promiscuity(dev, 1, false);
6172 dev->uc_promisc = true;
6173 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
6174 __dev_set_promiscuity(dev, -1, false);
6175 dev->uc_promisc = false;
6179 if (ops->ndo_set_rx_mode)
6180 ops->ndo_set_rx_mode(dev);
6183 void dev_set_rx_mode(struct net_device *dev)
6185 netif_addr_lock_bh(dev);
6186 __dev_set_rx_mode(dev);
6187 netif_addr_unlock_bh(dev);
6191 * dev_get_flags - get flags reported to userspace
6194 * Get the combination of flag bits exported through APIs to userspace.
6196 unsigned int dev_get_flags(const struct net_device *dev)
6200 flags = (dev->flags & ~(IFF_PROMISC |
6205 (dev->gflags & (IFF_PROMISC |
6208 if (netif_running(dev)) {
6209 if (netif_oper_up(dev))
6210 flags |= IFF_RUNNING;
6211 if (netif_carrier_ok(dev))
6212 flags |= IFF_LOWER_UP;
6213 if (netif_dormant(dev))
6214 flags |= IFF_DORMANT;
6219 EXPORT_SYMBOL(dev_get_flags);
6221 int __dev_change_flags(struct net_device *dev, unsigned int flags)
6223 unsigned int old_flags = dev->flags;
6229 * Set the flags on our device.
6232 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
6233 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
6235 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
6239 * Load in the correct multicast list now the flags have changed.
6242 if ((old_flags ^ flags) & IFF_MULTICAST)
6243 dev_change_rx_flags(dev, IFF_MULTICAST);
6245 dev_set_rx_mode(dev);
6248 * Have we downed the interface. We handle IFF_UP ourselves
6249 * according to user attempts to set it, rather than blindly
6254 if ((old_flags ^ flags) & IFF_UP)
6255 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
6257 if ((flags ^ dev->gflags) & IFF_PROMISC) {
6258 int inc = (flags & IFF_PROMISC) ? 1 : -1;
6259 unsigned int old_flags = dev->flags;
6261 dev->gflags ^= IFF_PROMISC;
6263 if (__dev_set_promiscuity(dev, inc, false) >= 0)
6264 if (dev->flags != old_flags)
6265 dev_set_rx_mode(dev);
6268 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
6269 is important. Some (broken) drivers set IFF_PROMISC, when
6270 IFF_ALLMULTI is requested not asking us and not reporting.
6272 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
6273 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
6275 dev->gflags ^= IFF_ALLMULTI;
6276 __dev_set_allmulti(dev, inc, false);
6282 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
6283 unsigned int gchanges)
6285 unsigned int changes = dev->flags ^ old_flags;
6288 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
6290 if (changes & IFF_UP) {
6291 if (dev->flags & IFF_UP)
6292 call_netdevice_notifiers(NETDEV_UP, dev);
6294 call_netdevice_notifiers(NETDEV_DOWN, dev);
6297 if (dev->flags & IFF_UP &&
6298 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
6299 struct netdev_notifier_change_info change_info;
6301 change_info.flags_changed = changes;
6302 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
6308 * dev_change_flags - change device settings
6310 * @flags: device state flags
6312 * Change settings on device based state flags. The flags are
6313 * in the userspace exported format.
6315 int dev_change_flags(struct net_device *dev, unsigned int flags)
6318 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
6320 ret = __dev_change_flags(dev, flags);
6324 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
6325 __dev_notify_flags(dev, old_flags, changes);
6328 EXPORT_SYMBOL(dev_change_flags);
6330 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
6332 const struct net_device_ops *ops = dev->netdev_ops;
6334 if (ops->ndo_change_mtu)
6335 return ops->ndo_change_mtu(dev, new_mtu);
6342 * dev_set_mtu - Change maximum transfer unit
6344 * @new_mtu: new transfer unit
6346 * Change the maximum transfer size of the network device.
6348 int dev_set_mtu(struct net_device *dev, int new_mtu)
6352 if (new_mtu == dev->mtu)
6355 /* MTU must be positive, and in range */
6356 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
6357 net_err_ratelimited("%s: Invalid MTU %d requested, hw min %d\n",
6358 dev->name, new_mtu, dev->min_mtu);
6362 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
6363 net_err_ratelimited("%s: Invalid MTU %d requested, hw max %d\n",
6364 dev->name, new_mtu, dev->max_mtu);
6368 if (!netif_device_present(dev))
6371 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
6372 err = notifier_to_errno(err);
6376 orig_mtu = dev->mtu;
6377 err = __dev_set_mtu(dev, new_mtu);
6380 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6381 err = notifier_to_errno(err);
6383 /* setting mtu back and notifying everyone again,
6384 * so that they have a chance to revert changes.
6386 __dev_set_mtu(dev, orig_mtu);
6387 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6392 EXPORT_SYMBOL(dev_set_mtu);
6395 * dev_set_group - Change group this device belongs to
6397 * @new_group: group this device should belong to
6399 void dev_set_group(struct net_device *dev, int new_group)
6401 dev->group = new_group;
6403 EXPORT_SYMBOL(dev_set_group);
6406 * dev_set_mac_address - Change Media Access Control Address
6410 * Change the hardware (MAC) address of the device
6412 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
6414 const struct net_device_ops *ops = dev->netdev_ops;
6417 if (!ops->ndo_set_mac_address)
6419 if (sa->sa_family != dev->type)
6421 if (!netif_device_present(dev))
6423 err = ops->ndo_set_mac_address(dev, sa);
6426 dev->addr_assign_type = NET_ADDR_SET;
6427 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
6428 add_device_randomness(dev->dev_addr, dev->addr_len);
6431 EXPORT_SYMBOL(dev_set_mac_address);
6434 * dev_change_carrier - Change device carrier
6436 * @new_carrier: new value
6438 * Change device carrier
6440 int dev_change_carrier(struct net_device *dev, bool new_carrier)
6442 const struct net_device_ops *ops = dev->netdev_ops;
6444 if (!ops->ndo_change_carrier)
6446 if (!netif_device_present(dev))
6448 return ops->ndo_change_carrier(dev, new_carrier);
6450 EXPORT_SYMBOL(dev_change_carrier);
6453 * dev_get_phys_port_id - Get device physical port ID
6457 * Get device physical port ID
6459 int dev_get_phys_port_id(struct net_device *dev,
6460 struct netdev_phys_item_id *ppid)
6462 const struct net_device_ops *ops = dev->netdev_ops;
6464 if (!ops->ndo_get_phys_port_id)
6466 return ops->ndo_get_phys_port_id(dev, ppid);
6468 EXPORT_SYMBOL(dev_get_phys_port_id);
6471 * dev_get_phys_port_name - Get device physical port name
6474 * @len: limit of bytes to copy to name
6476 * Get device physical port name
6478 int dev_get_phys_port_name(struct net_device *dev,
6479 char *name, size_t len)
6481 const struct net_device_ops *ops = dev->netdev_ops;
6483 if (!ops->ndo_get_phys_port_name)
6485 return ops->ndo_get_phys_port_name(dev, name, len);
6487 EXPORT_SYMBOL(dev_get_phys_port_name);
6490 * dev_change_proto_down - update protocol port state information
6492 * @proto_down: new value
6494 * This info can be used by switch drivers to set the phys state of the
6497 int dev_change_proto_down(struct net_device *dev, bool proto_down)
6499 const struct net_device_ops *ops = dev->netdev_ops;
6501 if (!ops->ndo_change_proto_down)
6503 if (!netif_device_present(dev))
6505 return ops->ndo_change_proto_down(dev, proto_down);
6507 EXPORT_SYMBOL(dev_change_proto_down);
6510 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
6512 * @fd: new program fd or negative value to clear
6514 * Set or clear a bpf program for a device
6516 int dev_change_xdp_fd(struct net_device *dev, int fd)
6518 const struct net_device_ops *ops = dev->netdev_ops;
6519 struct bpf_prog *prog = NULL;
6520 struct netdev_xdp xdp = {};
6526 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_XDP);
6528 return PTR_ERR(prog);
6531 xdp.command = XDP_SETUP_PROG;
6533 err = ops->ndo_xdp(dev, &xdp);
6534 if (err < 0 && prog)
6539 EXPORT_SYMBOL(dev_change_xdp_fd);
6542 * dev_new_index - allocate an ifindex
6543 * @net: the applicable net namespace
6545 * Returns a suitable unique value for a new device interface
6546 * number. The caller must hold the rtnl semaphore or the
6547 * dev_base_lock to be sure it remains unique.
6549 static int dev_new_index(struct net *net)
6551 int ifindex = net->ifindex;
6555 if (!__dev_get_by_index(net, ifindex))
6556 return net->ifindex = ifindex;
6560 /* Delayed registration/unregisteration */
6561 static LIST_HEAD(net_todo_list);
6562 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
6564 static void net_set_todo(struct net_device *dev)
6566 list_add_tail(&dev->todo_list, &net_todo_list);
6567 dev_net(dev)->dev_unreg_count++;
6570 static void rollback_registered_many(struct list_head *head)
6572 struct net_device *dev, *tmp;
6573 LIST_HEAD(close_head);
6575 BUG_ON(dev_boot_phase);
6578 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
6579 /* Some devices call without registering
6580 * for initialization unwind. Remove those
6581 * devices and proceed with the remaining.
6583 if (dev->reg_state == NETREG_UNINITIALIZED) {
6584 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
6588 list_del(&dev->unreg_list);
6591 dev->dismantle = true;
6592 BUG_ON(dev->reg_state != NETREG_REGISTERED);
6595 /* If device is running, close it first. */
6596 list_for_each_entry(dev, head, unreg_list)
6597 list_add_tail(&dev->close_list, &close_head);
6598 dev_close_many(&close_head, true);
6600 list_for_each_entry(dev, head, unreg_list) {
6601 /* And unlink it from device chain. */
6602 unlist_netdevice(dev);
6604 dev->reg_state = NETREG_UNREGISTERING;
6606 flush_all_backlogs();
6610 list_for_each_entry(dev, head, unreg_list) {
6611 struct sk_buff *skb = NULL;
6613 /* Shutdown queueing discipline. */
6617 /* Notify protocols, that we are about to destroy
6618 this device. They should clean all the things.
6620 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6622 if (!dev->rtnl_link_ops ||
6623 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6624 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U,
6628 * Flush the unicast and multicast chains
6633 if (dev->netdev_ops->ndo_uninit)
6634 dev->netdev_ops->ndo_uninit(dev);
6637 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
6639 /* Notifier chain MUST detach us all upper devices. */
6640 WARN_ON(netdev_has_any_upper_dev(dev));
6641 WARN_ON(netdev_has_any_lower_dev(dev));
6643 /* Remove entries from kobject tree */
6644 netdev_unregister_kobject(dev);
6646 /* Remove XPS queueing entries */
6647 netif_reset_xps_queues_gt(dev, 0);
6653 list_for_each_entry(dev, head, unreg_list)
6657 static void rollback_registered(struct net_device *dev)
6661 list_add(&dev->unreg_list, &single);
6662 rollback_registered_many(&single);
6666 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
6667 struct net_device *upper, netdev_features_t features)
6669 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6670 netdev_features_t feature;
6673 for_each_netdev_feature(&upper_disables, feature_bit) {
6674 feature = __NETIF_F_BIT(feature_bit);
6675 if (!(upper->wanted_features & feature)
6676 && (features & feature)) {
6677 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
6678 &feature, upper->name);
6679 features &= ~feature;
6686 static void netdev_sync_lower_features(struct net_device *upper,
6687 struct net_device *lower, netdev_features_t features)
6689 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6690 netdev_features_t feature;
6693 for_each_netdev_feature(&upper_disables, feature_bit) {
6694 feature = __NETIF_F_BIT(feature_bit);
6695 if (!(features & feature) && (lower->features & feature)) {
6696 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
6697 &feature, lower->name);
6698 lower->wanted_features &= ~feature;
6699 netdev_update_features(lower);
6701 if (unlikely(lower->features & feature))
6702 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
6703 &feature, lower->name);
6708 static netdev_features_t netdev_fix_features(struct net_device *dev,
6709 netdev_features_t features)
6711 /* Fix illegal checksum combinations */
6712 if ((features & NETIF_F_HW_CSUM) &&
6713 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6714 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
6715 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
6718 /* TSO requires that SG is present as well. */
6719 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
6720 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
6721 features &= ~NETIF_F_ALL_TSO;
6724 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
6725 !(features & NETIF_F_IP_CSUM)) {
6726 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
6727 features &= ~NETIF_F_TSO;
6728 features &= ~NETIF_F_TSO_ECN;
6731 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
6732 !(features & NETIF_F_IPV6_CSUM)) {
6733 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
6734 features &= ~NETIF_F_TSO6;
6737 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
6738 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
6739 features &= ~NETIF_F_TSO_MANGLEID;
6741 /* TSO ECN requires that TSO is present as well. */
6742 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
6743 features &= ~NETIF_F_TSO_ECN;
6745 /* Software GSO depends on SG. */
6746 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
6747 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
6748 features &= ~NETIF_F_GSO;
6751 /* UFO needs SG and checksumming */
6752 if (features & NETIF_F_UFO) {
6753 /* maybe split UFO into V4 and V6? */
6754 if (!(features & NETIF_F_HW_CSUM) &&
6755 ((features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) !=
6756 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))) {
6758 "Dropping NETIF_F_UFO since no checksum offload features.\n");
6759 features &= ~NETIF_F_UFO;
6762 if (!(features & NETIF_F_SG)) {
6764 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
6765 features &= ~NETIF_F_UFO;
6769 /* GSO partial features require GSO partial be set */
6770 if ((features & dev->gso_partial_features) &&
6771 !(features & NETIF_F_GSO_PARTIAL)) {
6773 "Dropping partially supported GSO features since no GSO partial.\n");
6774 features &= ~dev->gso_partial_features;
6777 #ifdef CONFIG_NET_RX_BUSY_POLL
6778 if (dev->netdev_ops->ndo_busy_poll)
6779 features |= NETIF_F_BUSY_POLL;
6782 features &= ~NETIF_F_BUSY_POLL;
6787 int __netdev_update_features(struct net_device *dev)
6789 struct net_device *upper, *lower;
6790 netdev_features_t features;
6791 struct list_head *iter;
6796 features = netdev_get_wanted_features(dev);
6798 if (dev->netdev_ops->ndo_fix_features)
6799 features = dev->netdev_ops->ndo_fix_features(dev, features);
6801 /* driver might be less strict about feature dependencies */
6802 features = netdev_fix_features(dev, features);
6804 /* some features can't be enabled if they're off an an upper device */
6805 netdev_for_each_upper_dev_rcu(dev, upper, iter)
6806 features = netdev_sync_upper_features(dev, upper, features);
6808 if (dev->features == features)
6811 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
6812 &dev->features, &features);
6814 if (dev->netdev_ops->ndo_set_features)
6815 err = dev->netdev_ops->ndo_set_features(dev, features);
6819 if (unlikely(err < 0)) {
6821 "set_features() failed (%d); wanted %pNF, left %pNF\n",
6822 err, &features, &dev->features);
6823 /* return non-0 since some features might have changed and
6824 * it's better to fire a spurious notification than miss it
6830 /* some features must be disabled on lower devices when disabled
6831 * on an upper device (think: bonding master or bridge)
6833 netdev_for_each_lower_dev(dev, lower, iter)
6834 netdev_sync_lower_features(dev, lower, features);
6837 dev->features = features;
6839 return err < 0 ? 0 : 1;
6843 * netdev_update_features - recalculate device features
6844 * @dev: the device to check
6846 * Recalculate dev->features set and send notifications if it
6847 * has changed. Should be called after driver or hardware dependent
6848 * conditions might have changed that influence the features.
6850 void netdev_update_features(struct net_device *dev)
6852 if (__netdev_update_features(dev))
6853 netdev_features_change(dev);
6855 EXPORT_SYMBOL(netdev_update_features);
6858 * netdev_change_features - recalculate device features
6859 * @dev: the device to check
6861 * Recalculate dev->features set and send notifications even
6862 * if they have not changed. Should be called instead of
6863 * netdev_update_features() if also dev->vlan_features might
6864 * have changed to allow the changes to be propagated to stacked
6867 void netdev_change_features(struct net_device *dev)
6869 __netdev_update_features(dev);
6870 netdev_features_change(dev);
6872 EXPORT_SYMBOL(netdev_change_features);
6875 * netif_stacked_transfer_operstate - transfer operstate
6876 * @rootdev: the root or lower level device to transfer state from
6877 * @dev: the device to transfer operstate to
6879 * Transfer operational state from root to device. This is normally
6880 * called when a stacking relationship exists between the root
6881 * device and the device(a leaf device).
6883 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
6884 struct net_device *dev)
6886 if (rootdev->operstate == IF_OPER_DORMANT)
6887 netif_dormant_on(dev);
6889 netif_dormant_off(dev);
6891 if (netif_carrier_ok(rootdev)) {
6892 if (!netif_carrier_ok(dev))
6893 netif_carrier_on(dev);
6895 if (netif_carrier_ok(dev))
6896 netif_carrier_off(dev);
6899 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
6902 static int netif_alloc_rx_queues(struct net_device *dev)
6904 unsigned int i, count = dev->num_rx_queues;
6905 struct netdev_rx_queue *rx;
6906 size_t sz = count * sizeof(*rx);
6910 rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6918 for (i = 0; i < count; i++)
6924 static void netdev_init_one_queue(struct net_device *dev,
6925 struct netdev_queue *queue, void *_unused)
6927 /* Initialize queue lock */
6928 spin_lock_init(&queue->_xmit_lock);
6929 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
6930 queue->xmit_lock_owner = -1;
6931 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
6934 dql_init(&queue->dql, HZ);
6938 static void netif_free_tx_queues(struct net_device *dev)
6943 static int netif_alloc_netdev_queues(struct net_device *dev)
6945 unsigned int count = dev->num_tx_queues;
6946 struct netdev_queue *tx;
6947 size_t sz = count * sizeof(*tx);
6949 if (count < 1 || count > 0xffff)
6952 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6960 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
6961 spin_lock_init(&dev->tx_global_lock);
6966 void netif_tx_stop_all_queues(struct net_device *dev)
6970 for (i = 0; i < dev->num_tx_queues; i++) {
6971 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
6972 netif_tx_stop_queue(txq);
6975 EXPORT_SYMBOL(netif_tx_stop_all_queues);
6978 * register_netdevice - register a network device
6979 * @dev: device to register
6981 * Take a completed network device structure and add it to the kernel
6982 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6983 * chain. 0 is returned on success. A negative errno code is returned
6984 * on a failure to set up the device, or if the name is a duplicate.
6986 * Callers must hold the rtnl semaphore. You may want
6987 * register_netdev() instead of this.
6990 * The locking appears insufficient to guarantee two parallel registers
6991 * will not get the same name.
6994 int register_netdevice(struct net_device *dev)
6997 struct net *net = dev_net(dev);
6999 BUG_ON(dev_boot_phase);
7004 /* When net_device's are persistent, this will be fatal. */
7005 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
7008 spin_lock_init(&dev->addr_list_lock);
7009 netdev_set_addr_lockdep_class(dev);
7011 ret = dev_get_valid_name(net, dev, dev->name);
7015 /* Init, if this function is available */
7016 if (dev->netdev_ops->ndo_init) {
7017 ret = dev->netdev_ops->ndo_init(dev);
7025 if (((dev->hw_features | dev->features) &
7026 NETIF_F_HW_VLAN_CTAG_FILTER) &&
7027 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
7028 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
7029 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
7036 dev->ifindex = dev_new_index(net);
7037 else if (__dev_get_by_index(net, dev->ifindex))
7040 /* Transfer changeable features to wanted_features and enable
7041 * software offloads (GSO and GRO).
7043 dev->hw_features |= NETIF_F_SOFT_FEATURES;
7044 dev->features |= NETIF_F_SOFT_FEATURES;
7045 dev->wanted_features = dev->features & dev->hw_features;
7047 if (!(dev->flags & IFF_LOOPBACK))
7048 dev->hw_features |= NETIF_F_NOCACHE_COPY;
7050 /* If IPv4 TCP segmentation offload is supported we should also
7051 * allow the device to enable segmenting the frame with the option
7052 * of ignoring a static IP ID value. This doesn't enable the
7053 * feature itself but allows the user to enable it later.
7055 if (dev->hw_features & NETIF_F_TSO)
7056 dev->hw_features |= NETIF_F_TSO_MANGLEID;
7057 if (dev->vlan_features & NETIF_F_TSO)
7058 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
7059 if (dev->mpls_features & NETIF_F_TSO)
7060 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
7061 if (dev->hw_enc_features & NETIF_F_TSO)
7062 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
7064 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
7066 dev->vlan_features |= NETIF_F_HIGHDMA;
7068 /* Make NETIF_F_SG inheritable to tunnel devices.
7070 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
7072 /* Make NETIF_F_SG inheritable to MPLS.
7074 dev->mpls_features |= NETIF_F_SG;
7076 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
7077 ret = notifier_to_errno(ret);
7081 ret = netdev_register_kobject(dev);
7084 dev->reg_state = NETREG_REGISTERED;
7086 __netdev_update_features(dev);
7089 * Default initial state at registry is that the
7090 * device is present.
7093 set_bit(__LINK_STATE_PRESENT, &dev->state);
7095 linkwatch_init_dev(dev);
7097 dev_init_scheduler(dev);
7099 list_netdevice(dev);
7100 add_device_randomness(dev->dev_addr, dev->addr_len);
7102 /* If the device has permanent device address, driver should
7103 * set dev_addr and also addr_assign_type should be set to
7104 * NET_ADDR_PERM (default value).
7106 if (dev->addr_assign_type == NET_ADDR_PERM)
7107 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
7109 /* Notify protocols, that a new device appeared. */
7110 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
7111 ret = notifier_to_errno(ret);
7113 rollback_registered(dev);
7114 dev->reg_state = NETREG_UNREGISTERED;
7117 * Prevent userspace races by waiting until the network
7118 * device is fully setup before sending notifications.
7120 if (!dev->rtnl_link_ops ||
7121 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
7122 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7128 if (dev->netdev_ops->ndo_uninit)
7129 dev->netdev_ops->ndo_uninit(dev);
7132 EXPORT_SYMBOL(register_netdevice);
7135 * init_dummy_netdev - init a dummy network device for NAPI
7136 * @dev: device to init
7138 * This takes a network device structure and initialize the minimum
7139 * amount of fields so it can be used to schedule NAPI polls without
7140 * registering a full blown interface. This is to be used by drivers
7141 * that need to tie several hardware interfaces to a single NAPI
7142 * poll scheduler due to HW limitations.
7144 int init_dummy_netdev(struct net_device *dev)
7146 /* Clear everything. Note we don't initialize spinlocks
7147 * are they aren't supposed to be taken by any of the
7148 * NAPI code and this dummy netdev is supposed to be
7149 * only ever used for NAPI polls
7151 memset(dev, 0, sizeof(struct net_device));
7153 /* make sure we BUG if trying to hit standard
7154 * register/unregister code path
7156 dev->reg_state = NETREG_DUMMY;
7158 /* NAPI wants this */
7159 INIT_LIST_HEAD(&dev->napi_list);
7161 /* a dummy interface is started by default */
7162 set_bit(__LINK_STATE_PRESENT, &dev->state);
7163 set_bit(__LINK_STATE_START, &dev->state);
7165 /* Note : We dont allocate pcpu_refcnt for dummy devices,
7166 * because users of this 'device' dont need to change
7172 EXPORT_SYMBOL_GPL(init_dummy_netdev);
7176 * register_netdev - register a network device
7177 * @dev: device to register
7179 * Take a completed network device structure and add it to the kernel
7180 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
7181 * chain. 0 is returned on success. A negative errno code is returned
7182 * on a failure to set up the device, or if the name is a duplicate.
7184 * This is a wrapper around register_netdevice that takes the rtnl semaphore
7185 * and expands the device name if you passed a format string to
7188 int register_netdev(struct net_device *dev)
7193 err = register_netdevice(dev);
7197 EXPORT_SYMBOL(register_netdev);
7199 int netdev_refcnt_read(const struct net_device *dev)
7203 for_each_possible_cpu(i)
7204 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
7207 EXPORT_SYMBOL(netdev_refcnt_read);
7210 * netdev_wait_allrefs - wait until all references are gone.
7211 * @dev: target net_device
7213 * This is called when unregistering network devices.
7215 * Any protocol or device that holds a reference should register
7216 * for netdevice notification, and cleanup and put back the
7217 * reference if they receive an UNREGISTER event.
7218 * We can get stuck here if buggy protocols don't correctly
7221 static void netdev_wait_allrefs(struct net_device *dev)
7223 unsigned long rebroadcast_time, warning_time;
7226 linkwatch_forget_dev(dev);
7228 rebroadcast_time = warning_time = jiffies;
7229 refcnt = netdev_refcnt_read(dev);
7231 while (refcnt != 0) {
7232 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
7235 /* Rebroadcast unregister notification */
7236 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7242 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7243 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
7245 /* We must not have linkwatch events
7246 * pending on unregister. If this
7247 * happens, we simply run the queue
7248 * unscheduled, resulting in a noop
7251 linkwatch_run_queue();
7256 rebroadcast_time = jiffies;
7261 refcnt = netdev_refcnt_read(dev);
7263 if (time_after(jiffies, warning_time + 10 * HZ)) {
7264 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
7266 warning_time = jiffies;
7275 * register_netdevice(x1);
7276 * register_netdevice(x2);
7278 * unregister_netdevice(y1);
7279 * unregister_netdevice(y2);
7285 * We are invoked by rtnl_unlock().
7286 * This allows us to deal with problems:
7287 * 1) We can delete sysfs objects which invoke hotplug
7288 * without deadlocking with linkwatch via keventd.
7289 * 2) Since we run with the RTNL semaphore not held, we can sleep
7290 * safely in order to wait for the netdev refcnt to drop to zero.
7292 * We must not return until all unregister events added during
7293 * the interval the lock was held have been completed.
7295 void netdev_run_todo(void)
7297 struct list_head list;
7299 /* Snapshot list, allow later requests */
7300 list_replace_init(&net_todo_list, &list);
7305 /* Wait for rcu callbacks to finish before next phase */
7306 if (!list_empty(&list))
7309 while (!list_empty(&list)) {
7310 struct net_device *dev
7311 = list_first_entry(&list, struct net_device, todo_list);
7312 list_del(&dev->todo_list);
7315 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7318 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
7319 pr_err("network todo '%s' but state %d\n",
7320 dev->name, dev->reg_state);
7325 dev->reg_state = NETREG_UNREGISTERED;
7327 netdev_wait_allrefs(dev);
7330 BUG_ON(netdev_refcnt_read(dev));
7331 BUG_ON(!list_empty(&dev->ptype_all));
7332 BUG_ON(!list_empty(&dev->ptype_specific));
7333 WARN_ON(rcu_access_pointer(dev->ip_ptr));
7334 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
7335 WARN_ON(dev->dn_ptr);
7337 if (dev->destructor)
7338 dev->destructor(dev);
7340 /* Report a network device has been unregistered */
7342 dev_net(dev)->dev_unreg_count--;
7344 wake_up(&netdev_unregistering_wq);
7346 /* Free network device */
7347 kobject_put(&dev->dev.kobj);
7351 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
7352 * all the same fields in the same order as net_device_stats, with only
7353 * the type differing, but rtnl_link_stats64 may have additional fields
7354 * at the end for newer counters.
7356 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
7357 const struct net_device_stats *netdev_stats)
7359 #if BITS_PER_LONG == 64
7360 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
7361 memcpy(stats64, netdev_stats, sizeof(*stats64));
7362 /* zero out counters that only exist in rtnl_link_stats64 */
7363 memset((char *)stats64 + sizeof(*netdev_stats), 0,
7364 sizeof(*stats64) - sizeof(*netdev_stats));
7366 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
7367 const unsigned long *src = (const unsigned long *)netdev_stats;
7368 u64 *dst = (u64 *)stats64;
7370 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
7371 for (i = 0; i < n; i++)
7373 /* zero out counters that only exist in rtnl_link_stats64 */
7374 memset((char *)stats64 + n * sizeof(u64), 0,
7375 sizeof(*stats64) - n * sizeof(u64));
7378 EXPORT_SYMBOL(netdev_stats_to_stats64);
7381 * dev_get_stats - get network device statistics
7382 * @dev: device to get statistics from
7383 * @storage: place to store stats
7385 * Get network statistics from device. Return @storage.
7386 * The device driver may provide its own method by setting
7387 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
7388 * otherwise the internal statistics structure is used.
7390 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
7391 struct rtnl_link_stats64 *storage)
7393 const struct net_device_ops *ops = dev->netdev_ops;
7395 if (ops->ndo_get_stats64) {
7396 memset(storage, 0, sizeof(*storage));
7397 ops->ndo_get_stats64(dev, storage);
7398 } else if (ops->ndo_get_stats) {
7399 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
7401 netdev_stats_to_stats64(storage, &dev->stats);
7403 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
7404 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
7405 storage->rx_nohandler += atomic_long_read(&dev->rx_nohandler);
7408 EXPORT_SYMBOL(dev_get_stats);
7410 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
7412 struct netdev_queue *queue = dev_ingress_queue(dev);
7414 #ifdef CONFIG_NET_CLS_ACT
7417 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
7420 netdev_init_one_queue(dev, queue, NULL);
7421 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
7422 queue->qdisc_sleeping = &noop_qdisc;
7423 rcu_assign_pointer(dev->ingress_queue, queue);
7428 static const struct ethtool_ops default_ethtool_ops;
7430 void netdev_set_default_ethtool_ops(struct net_device *dev,
7431 const struct ethtool_ops *ops)
7433 if (dev->ethtool_ops == &default_ethtool_ops)
7434 dev->ethtool_ops = ops;
7436 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
7438 void netdev_freemem(struct net_device *dev)
7440 char *addr = (char *)dev - dev->padded;
7446 * alloc_netdev_mqs - allocate network device
7447 * @sizeof_priv: size of private data to allocate space for
7448 * @name: device name format string
7449 * @name_assign_type: origin of device name
7450 * @setup: callback to initialize device
7451 * @txqs: the number of TX subqueues to allocate
7452 * @rxqs: the number of RX subqueues to allocate
7454 * Allocates a struct net_device with private data area for driver use
7455 * and performs basic initialization. Also allocates subqueue structs
7456 * for each queue on the device.
7458 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
7459 unsigned char name_assign_type,
7460 void (*setup)(struct net_device *),
7461 unsigned int txqs, unsigned int rxqs)
7463 struct net_device *dev;
7465 struct net_device *p;
7467 BUG_ON(strlen(name) >= sizeof(dev->name));
7470 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
7476 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
7481 alloc_size = sizeof(struct net_device);
7483 /* ensure 32-byte alignment of private area */
7484 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
7485 alloc_size += sizeof_priv;
7487 /* ensure 32-byte alignment of whole construct */
7488 alloc_size += NETDEV_ALIGN - 1;
7490 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
7492 p = vzalloc(alloc_size);
7496 dev = PTR_ALIGN(p, NETDEV_ALIGN);
7497 dev->padded = (char *)dev - (char *)p;
7499 dev->pcpu_refcnt = alloc_percpu(int);
7500 if (!dev->pcpu_refcnt)
7503 if (dev_addr_init(dev))
7509 dev_net_set(dev, &init_net);
7511 dev->gso_max_size = GSO_MAX_SIZE;
7512 dev->gso_max_segs = GSO_MAX_SEGS;
7514 INIT_LIST_HEAD(&dev->napi_list);
7515 INIT_LIST_HEAD(&dev->unreg_list);
7516 INIT_LIST_HEAD(&dev->close_list);
7517 INIT_LIST_HEAD(&dev->link_watch_list);
7518 INIT_LIST_HEAD(&dev->adj_list.upper);
7519 INIT_LIST_HEAD(&dev->adj_list.lower);
7520 INIT_LIST_HEAD(&dev->ptype_all);
7521 INIT_LIST_HEAD(&dev->ptype_specific);
7522 #ifdef CONFIG_NET_SCHED
7523 hash_init(dev->qdisc_hash);
7525 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
7528 if (!dev->tx_queue_len) {
7529 dev->priv_flags |= IFF_NO_QUEUE;
7530 dev->tx_queue_len = 1;
7533 dev->num_tx_queues = txqs;
7534 dev->real_num_tx_queues = txqs;
7535 if (netif_alloc_netdev_queues(dev))
7539 dev->num_rx_queues = rxqs;
7540 dev->real_num_rx_queues = rxqs;
7541 if (netif_alloc_rx_queues(dev))
7545 strcpy(dev->name, name);
7546 dev->name_assign_type = name_assign_type;
7547 dev->group = INIT_NETDEV_GROUP;
7548 if (!dev->ethtool_ops)
7549 dev->ethtool_ops = &default_ethtool_ops;
7551 nf_hook_ingress_init(dev);
7560 free_percpu(dev->pcpu_refcnt);
7562 netdev_freemem(dev);
7565 EXPORT_SYMBOL(alloc_netdev_mqs);
7568 * free_netdev - free network device
7571 * This function does the last stage of destroying an allocated device
7572 * interface. The reference to the device object is released.
7573 * If this is the last reference then it will be freed.
7574 * Must be called in process context.
7576 void free_netdev(struct net_device *dev)
7578 struct napi_struct *p, *n;
7581 netif_free_tx_queues(dev);
7586 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
7588 /* Flush device addresses */
7589 dev_addr_flush(dev);
7591 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
7594 free_percpu(dev->pcpu_refcnt);
7595 dev->pcpu_refcnt = NULL;
7597 /* Compatibility with error handling in drivers */
7598 if (dev->reg_state == NETREG_UNINITIALIZED) {
7599 netdev_freemem(dev);
7603 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
7604 dev->reg_state = NETREG_RELEASED;
7606 /* will free via device release */
7607 put_device(&dev->dev);
7609 EXPORT_SYMBOL(free_netdev);
7612 * synchronize_net - Synchronize with packet receive processing
7614 * Wait for packets currently being received to be done.
7615 * Does not block later packets from starting.
7617 void synchronize_net(void)
7620 if (rtnl_is_locked())
7621 synchronize_rcu_expedited();
7625 EXPORT_SYMBOL(synchronize_net);
7628 * unregister_netdevice_queue - remove device from the kernel
7632 * This function shuts down a device interface and removes it
7633 * from the kernel tables.
7634 * If head not NULL, device is queued to be unregistered later.
7636 * Callers must hold the rtnl semaphore. You may want
7637 * unregister_netdev() instead of this.
7640 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
7645 list_move_tail(&dev->unreg_list, head);
7647 rollback_registered(dev);
7648 /* Finish processing unregister after unlock */
7652 EXPORT_SYMBOL(unregister_netdevice_queue);
7655 * unregister_netdevice_many - unregister many devices
7656 * @head: list of devices
7658 * Note: As most callers use a stack allocated list_head,
7659 * we force a list_del() to make sure stack wont be corrupted later.
7661 void unregister_netdevice_many(struct list_head *head)
7663 struct net_device *dev;
7665 if (!list_empty(head)) {
7666 rollback_registered_many(head);
7667 list_for_each_entry(dev, head, unreg_list)
7672 EXPORT_SYMBOL(unregister_netdevice_many);
7675 * unregister_netdev - remove device from the kernel
7678 * This function shuts down a device interface and removes it
7679 * from the kernel tables.
7681 * This is just a wrapper for unregister_netdevice that takes
7682 * the rtnl semaphore. In general you want to use this and not
7683 * unregister_netdevice.
7685 void unregister_netdev(struct net_device *dev)
7688 unregister_netdevice(dev);
7691 EXPORT_SYMBOL(unregister_netdev);
7694 * dev_change_net_namespace - move device to different nethost namespace
7696 * @net: network namespace
7697 * @pat: If not NULL name pattern to try if the current device name
7698 * is already taken in the destination network namespace.
7700 * This function shuts down a device interface and moves it
7701 * to a new network namespace. On success 0 is returned, on
7702 * a failure a netagive errno code is returned.
7704 * Callers must hold the rtnl semaphore.
7707 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
7713 /* Don't allow namespace local devices to be moved. */
7715 if (dev->features & NETIF_F_NETNS_LOCAL)
7718 /* Ensure the device has been registrered */
7719 if (dev->reg_state != NETREG_REGISTERED)
7722 /* Get out if there is nothing todo */
7724 if (net_eq(dev_net(dev), net))
7727 /* Pick the destination device name, and ensure
7728 * we can use it in the destination network namespace.
7731 if (__dev_get_by_name(net, dev->name)) {
7732 /* We get here if we can't use the current device name */
7735 if (dev_get_valid_name(net, dev, pat) < 0)
7740 * And now a mini version of register_netdevice unregister_netdevice.
7743 /* If device is running close it first. */
7746 /* And unlink it from device chain */
7748 unlist_netdevice(dev);
7752 /* Shutdown queueing discipline. */
7755 /* Notify protocols, that we are about to destroy
7756 this device. They should clean all the things.
7758 Note that dev->reg_state stays at NETREG_REGISTERED.
7759 This is wanted because this way 8021q and macvlan know
7760 the device is just moving and can keep their slaves up.
7762 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7764 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7765 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
7768 * Flush the unicast and multicast chains
7773 /* Send a netdev-removed uevent to the old namespace */
7774 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
7775 netdev_adjacent_del_links(dev);
7777 /* Actually switch the network namespace */
7778 dev_net_set(dev, net);
7780 /* If there is an ifindex conflict assign a new one */
7781 if (__dev_get_by_index(net, dev->ifindex))
7782 dev->ifindex = dev_new_index(net);
7784 /* Send a netdev-add uevent to the new namespace */
7785 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
7786 netdev_adjacent_add_links(dev);
7788 /* Fixup kobjects */
7789 err = device_rename(&dev->dev, dev->name);
7792 /* Add the device back in the hashes */
7793 list_netdevice(dev);
7795 /* Notify protocols, that a new device appeared. */
7796 call_netdevice_notifiers(NETDEV_REGISTER, dev);
7799 * Prevent userspace races by waiting until the network
7800 * device is fully setup before sending notifications.
7802 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7809 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
7811 static int dev_cpu_callback(struct notifier_block *nfb,
7812 unsigned long action,
7815 struct sk_buff **list_skb;
7816 struct sk_buff *skb;
7817 unsigned int cpu, oldcpu = (unsigned long)ocpu;
7818 struct softnet_data *sd, *oldsd;
7820 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
7823 local_irq_disable();
7824 cpu = smp_processor_id();
7825 sd = &per_cpu(softnet_data, cpu);
7826 oldsd = &per_cpu(softnet_data, oldcpu);
7828 /* Find end of our completion_queue. */
7829 list_skb = &sd->completion_queue;
7831 list_skb = &(*list_skb)->next;
7832 /* Append completion queue from offline CPU. */
7833 *list_skb = oldsd->completion_queue;
7834 oldsd->completion_queue = NULL;
7836 /* Append output queue from offline CPU. */
7837 if (oldsd->output_queue) {
7838 *sd->output_queue_tailp = oldsd->output_queue;
7839 sd->output_queue_tailp = oldsd->output_queue_tailp;
7840 oldsd->output_queue = NULL;
7841 oldsd->output_queue_tailp = &oldsd->output_queue;
7843 /* Append NAPI poll list from offline CPU, with one exception :
7844 * process_backlog() must be called by cpu owning percpu backlog.
7845 * We properly handle process_queue & input_pkt_queue later.
7847 while (!list_empty(&oldsd->poll_list)) {
7848 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
7852 list_del_init(&napi->poll_list);
7853 if (napi->poll == process_backlog)
7856 ____napi_schedule(sd, napi);
7859 raise_softirq_irqoff(NET_TX_SOFTIRQ);
7862 /* Process offline CPU's input_pkt_queue */
7863 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
7865 input_queue_head_incr(oldsd);
7867 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
7869 input_queue_head_incr(oldsd);
7877 * netdev_increment_features - increment feature set by one
7878 * @all: current feature set
7879 * @one: new feature set
7880 * @mask: mask feature set
7882 * Computes a new feature set after adding a device with feature set
7883 * @one to the master device with current feature set @all. Will not
7884 * enable anything that is off in @mask. Returns the new feature set.
7886 netdev_features_t netdev_increment_features(netdev_features_t all,
7887 netdev_features_t one, netdev_features_t mask)
7889 if (mask & NETIF_F_HW_CSUM)
7890 mask |= NETIF_F_CSUM_MASK;
7891 mask |= NETIF_F_VLAN_CHALLENGED;
7893 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
7894 all &= one | ~NETIF_F_ALL_FOR_ALL;
7896 /* If one device supports hw checksumming, set for all. */
7897 if (all & NETIF_F_HW_CSUM)
7898 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
7902 EXPORT_SYMBOL(netdev_increment_features);
7904 static struct hlist_head * __net_init netdev_create_hash(void)
7907 struct hlist_head *hash;
7909 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
7911 for (i = 0; i < NETDEV_HASHENTRIES; i++)
7912 INIT_HLIST_HEAD(&hash[i]);
7917 /* Initialize per network namespace state */
7918 static int __net_init netdev_init(struct net *net)
7920 if (net != &init_net)
7921 INIT_LIST_HEAD(&net->dev_base_head);
7923 net->dev_name_head = netdev_create_hash();
7924 if (net->dev_name_head == NULL)
7927 net->dev_index_head = netdev_create_hash();
7928 if (net->dev_index_head == NULL)
7934 kfree(net->dev_name_head);
7940 * netdev_drivername - network driver for the device
7941 * @dev: network device
7943 * Determine network driver for device.
7945 const char *netdev_drivername(const struct net_device *dev)
7947 const struct device_driver *driver;
7948 const struct device *parent;
7949 const char *empty = "";
7951 parent = dev->dev.parent;
7955 driver = parent->driver;
7956 if (driver && driver->name)
7957 return driver->name;
7961 static void __netdev_printk(const char *level, const struct net_device *dev,
7962 struct va_format *vaf)
7964 if (dev && dev->dev.parent) {
7965 dev_printk_emit(level[1] - '0',
7968 dev_driver_string(dev->dev.parent),
7969 dev_name(dev->dev.parent),
7970 netdev_name(dev), netdev_reg_state(dev),
7973 printk("%s%s%s: %pV",
7974 level, netdev_name(dev), netdev_reg_state(dev), vaf);
7976 printk("%s(NULL net_device): %pV", level, vaf);
7980 void netdev_printk(const char *level, const struct net_device *dev,
7981 const char *format, ...)
7983 struct va_format vaf;
7986 va_start(args, format);
7991 __netdev_printk(level, dev, &vaf);
7995 EXPORT_SYMBOL(netdev_printk);
7997 #define define_netdev_printk_level(func, level) \
7998 void func(const struct net_device *dev, const char *fmt, ...) \
8000 struct va_format vaf; \
8003 va_start(args, fmt); \
8008 __netdev_printk(level, dev, &vaf); \
8012 EXPORT_SYMBOL(func);
8014 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
8015 define_netdev_printk_level(netdev_alert, KERN_ALERT);
8016 define_netdev_printk_level(netdev_crit, KERN_CRIT);
8017 define_netdev_printk_level(netdev_err, KERN_ERR);
8018 define_netdev_printk_level(netdev_warn, KERN_WARNING);
8019 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
8020 define_netdev_printk_level(netdev_info, KERN_INFO);
8022 static void __net_exit netdev_exit(struct net *net)
8024 kfree(net->dev_name_head);
8025 kfree(net->dev_index_head);
8028 static struct pernet_operations __net_initdata netdev_net_ops = {
8029 .init = netdev_init,
8030 .exit = netdev_exit,
8033 static void __net_exit default_device_exit(struct net *net)
8035 struct net_device *dev, *aux;
8037 * Push all migratable network devices back to the
8038 * initial network namespace
8041 for_each_netdev_safe(net, dev, aux) {
8043 char fb_name[IFNAMSIZ];
8045 /* Ignore unmoveable devices (i.e. loopback) */
8046 if (dev->features & NETIF_F_NETNS_LOCAL)
8049 /* Leave virtual devices for the generic cleanup */
8050 if (dev->rtnl_link_ops)
8053 /* Push remaining network devices to init_net */
8054 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
8055 err = dev_change_net_namespace(dev, &init_net, fb_name);
8057 pr_emerg("%s: failed to move %s to init_net: %d\n",
8058 __func__, dev->name, err);
8065 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
8067 /* Return with the rtnl_lock held when there are no network
8068 * devices unregistering in any network namespace in net_list.
8072 DEFINE_WAIT_FUNC(wait, woken_wake_function);
8074 add_wait_queue(&netdev_unregistering_wq, &wait);
8076 unregistering = false;
8078 list_for_each_entry(net, net_list, exit_list) {
8079 if (net->dev_unreg_count > 0) {
8080 unregistering = true;
8088 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
8090 remove_wait_queue(&netdev_unregistering_wq, &wait);
8093 static void __net_exit default_device_exit_batch(struct list_head *net_list)
8095 /* At exit all network devices most be removed from a network
8096 * namespace. Do this in the reverse order of registration.
8097 * Do this across as many network namespaces as possible to
8098 * improve batching efficiency.
8100 struct net_device *dev;
8102 LIST_HEAD(dev_kill_list);
8104 /* To prevent network device cleanup code from dereferencing
8105 * loopback devices or network devices that have been freed
8106 * wait here for all pending unregistrations to complete,
8107 * before unregistring the loopback device and allowing the
8108 * network namespace be freed.
8110 * The netdev todo list containing all network devices
8111 * unregistrations that happen in default_device_exit_batch
8112 * will run in the rtnl_unlock() at the end of
8113 * default_device_exit_batch.
8115 rtnl_lock_unregistering(net_list);
8116 list_for_each_entry(net, net_list, exit_list) {
8117 for_each_netdev_reverse(net, dev) {
8118 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
8119 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
8121 unregister_netdevice_queue(dev, &dev_kill_list);
8124 unregister_netdevice_many(&dev_kill_list);
8128 static struct pernet_operations __net_initdata default_device_ops = {
8129 .exit = default_device_exit,
8130 .exit_batch = default_device_exit_batch,
8134 * Initialize the DEV module. At boot time this walks the device list and
8135 * unhooks any devices that fail to initialise (normally hardware not
8136 * present) and leaves us with a valid list of present and active devices.
8141 * This is called single threaded during boot, so no need
8142 * to take the rtnl semaphore.
8144 static int __init net_dev_init(void)
8146 int i, rc = -ENOMEM;
8148 BUG_ON(!dev_boot_phase);
8150 if (dev_proc_init())
8153 if (netdev_kobject_init())
8156 INIT_LIST_HEAD(&ptype_all);
8157 for (i = 0; i < PTYPE_HASH_SIZE; i++)
8158 INIT_LIST_HEAD(&ptype_base[i]);
8160 INIT_LIST_HEAD(&offload_base);
8162 if (register_pernet_subsys(&netdev_net_ops))
8166 * Initialise the packet receive queues.
8169 for_each_possible_cpu(i) {
8170 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
8171 struct softnet_data *sd = &per_cpu(softnet_data, i);
8173 INIT_WORK(flush, flush_backlog);
8175 skb_queue_head_init(&sd->input_pkt_queue);
8176 skb_queue_head_init(&sd->process_queue);
8177 INIT_LIST_HEAD(&sd->poll_list);
8178 sd->output_queue_tailp = &sd->output_queue;
8180 sd->csd.func = rps_trigger_softirq;
8185 sd->backlog.poll = process_backlog;
8186 sd->backlog.weight = weight_p;
8191 /* The loopback device is special if any other network devices
8192 * is present in a network namespace the loopback device must
8193 * be present. Since we now dynamically allocate and free the
8194 * loopback device ensure this invariant is maintained by
8195 * keeping the loopback device as the first device on the
8196 * list of network devices. Ensuring the loopback devices
8197 * is the first device that appears and the last network device
8200 if (register_pernet_device(&loopback_net_ops))
8203 if (register_pernet_device(&default_device_ops))
8206 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
8207 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
8209 hotcpu_notifier(dev_cpu_callback, 0);
8216 subsys_initcall(net_dev_init);