2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <net/net_namespace.h>
99 #include <linux/rtnetlink.h>
100 #include <linux/stat.h>
102 #include <net/dst_metadata.h>
103 #include <net/pkt_sched.h>
104 #include <net/checksum.h>
105 #include <net/xfrm.h>
106 #include <linux/highmem.h>
107 #include <linux/init.h>
108 #include <linux/module.h>
109 #include <linux/netpoll.h>
110 #include <linux/rcupdate.h>
111 #include <linux/delay.h>
112 #include <net/iw_handler.h>
113 #include <asm/current.h>
114 #include <linux/audit.h>
115 #include <linux/dmaengine.h>
116 #include <linux/err.h>
117 #include <linux/ctype.h>
118 #include <linux/if_arp.h>
119 #include <linux/if_vlan.h>
120 #include <linux/ip.h>
122 #include <net/mpls.h>
123 #include <linux/ipv6.h>
124 #include <linux/in.h>
125 #include <linux/jhash.h>
126 #include <linux/random.h>
127 #include <trace/events/napi.h>
128 #include <trace/events/net.h>
129 #include <trace/events/skb.h>
130 #include <linux/pci.h>
131 #include <linux/inetdevice.h>
132 #include <linux/cpu_rmap.h>
133 #include <linux/static_key.h>
134 #include <linux/hashtable.h>
135 #include <linux/vmalloc.h>
136 #include <linux/if_macvlan.h>
137 #include <linux/errqueue.h>
138 #include <linux/hrtimer.h>
139 #include <linux/netfilter_ingress.h>
141 #include "net-sysfs.h"
143 /* Instead of increasing this, you should create a hash table. */
144 #define MAX_GRO_SKBS 8
146 /* This should be increased if a protocol with a bigger head is added. */
147 #define GRO_MAX_HEAD (MAX_HEADER + 128)
149 static DEFINE_SPINLOCK(ptype_lock);
150 static DEFINE_SPINLOCK(offload_lock);
151 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
152 struct list_head ptype_all __read_mostly; /* Taps */
153 static struct list_head offload_base __read_mostly;
155 static int netif_rx_internal(struct sk_buff *skb);
156 static int call_netdevice_notifiers_info(unsigned long val,
157 struct net_device *dev,
158 struct netdev_notifier_info *info);
161 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
164 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
166 * Writers must hold the rtnl semaphore while they loop through the
167 * dev_base_head list, and hold dev_base_lock for writing when they do the
168 * actual updates. This allows pure readers to access the list even
169 * while a writer is preparing to update it.
171 * To put it another way, dev_base_lock is held for writing only to
172 * protect against pure readers; the rtnl semaphore provides the
173 * protection against other writers.
175 * See, for example usages, register_netdevice() and
176 * unregister_netdevice(), which must be called with the rtnl
179 DEFINE_RWLOCK(dev_base_lock);
180 EXPORT_SYMBOL(dev_base_lock);
182 /* protects napi_hash addition/deletion and napi_gen_id */
183 static DEFINE_SPINLOCK(napi_hash_lock);
185 static unsigned int napi_gen_id;
186 static DEFINE_HASHTABLE(napi_hash, 8);
188 static seqcount_t devnet_rename_seq;
190 static inline void dev_base_seq_inc(struct net *net)
192 while (++net->dev_base_seq == 0);
195 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
197 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
199 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
202 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
204 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
207 static inline void rps_lock(struct softnet_data *sd)
210 spin_lock(&sd->input_pkt_queue.lock);
214 static inline void rps_unlock(struct softnet_data *sd)
217 spin_unlock(&sd->input_pkt_queue.lock);
221 /* Device list insertion */
222 static void list_netdevice(struct net_device *dev)
224 struct net *net = dev_net(dev);
228 write_lock_bh(&dev_base_lock);
229 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
230 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
231 hlist_add_head_rcu(&dev->index_hlist,
232 dev_index_hash(net, dev->ifindex));
233 write_unlock_bh(&dev_base_lock);
235 dev_base_seq_inc(net);
238 /* Device list removal
239 * caller must respect a RCU grace period before freeing/reusing dev
241 static void unlist_netdevice(struct net_device *dev)
245 /* Unlink dev from the device chain */
246 write_lock_bh(&dev_base_lock);
247 list_del_rcu(&dev->dev_list);
248 hlist_del_rcu(&dev->name_hlist);
249 hlist_del_rcu(&dev->index_hlist);
250 write_unlock_bh(&dev_base_lock);
252 dev_base_seq_inc(dev_net(dev));
259 static RAW_NOTIFIER_HEAD(netdev_chain);
262 * Device drivers call our routines to queue packets here. We empty the
263 * queue in the local softnet handler.
266 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
267 EXPORT_PER_CPU_SYMBOL(softnet_data);
269 #ifdef CONFIG_LOCKDEP
271 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
272 * according to dev->type
274 static const unsigned short netdev_lock_type[] =
275 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
276 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
277 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
278 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
279 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
280 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
281 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
282 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
283 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
284 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
285 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
286 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
287 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
288 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
289 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
291 static const char *const netdev_lock_name[] =
292 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
293 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
294 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
295 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
296 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
297 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
298 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
299 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
300 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
301 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
302 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
303 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
304 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
305 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
306 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
308 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
309 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
311 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
315 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
316 if (netdev_lock_type[i] == dev_type)
318 /* the last key is used by default */
319 return ARRAY_SIZE(netdev_lock_type) - 1;
322 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
323 unsigned short dev_type)
327 i = netdev_lock_pos(dev_type);
328 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
329 netdev_lock_name[i]);
332 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
336 i = netdev_lock_pos(dev->type);
337 lockdep_set_class_and_name(&dev->addr_list_lock,
338 &netdev_addr_lock_key[i],
339 netdev_lock_name[i]);
342 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
343 unsigned short dev_type)
346 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
351 /*******************************************************************************
353 Protocol management and registration routines
355 *******************************************************************************/
358 * Add a protocol ID to the list. Now that the input handler is
359 * smarter we can dispense with all the messy stuff that used to be
362 * BEWARE!!! Protocol handlers, mangling input packets,
363 * MUST BE last in hash buckets and checking protocol handlers
364 * MUST start from promiscuous ptype_all chain in net_bh.
365 * It is true now, do not change it.
366 * Explanation follows: if protocol handler, mangling packet, will
367 * be the first on list, it is not able to sense, that packet
368 * is cloned and should be copied-on-write, so that it will
369 * change it and subsequent readers will get broken packet.
373 static inline struct list_head *ptype_head(const struct packet_type *pt)
375 if (pt->type == htons(ETH_P_ALL))
376 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
378 return pt->dev ? &pt->dev->ptype_specific :
379 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
383 * dev_add_pack - add packet handler
384 * @pt: packet type declaration
386 * Add a protocol handler to the networking stack. The passed &packet_type
387 * is linked into kernel lists and may not be freed until it has been
388 * removed from the kernel lists.
390 * This call does not sleep therefore it can not
391 * guarantee all CPU's that are in middle of receiving packets
392 * will see the new packet type (until the next received packet).
395 void dev_add_pack(struct packet_type *pt)
397 struct list_head *head = ptype_head(pt);
399 spin_lock(&ptype_lock);
400 list_add_rcu(&pt->list, head);
401 spin_unlock(&ptype_lock);
403 EXPORT_SYMBOL(dev_add_pack);
406 * __dev_remove_pack - remove packet handler
407 * @pt: packet type declaration
409 * Remove a protocol handler that was previously added to the kernel
410 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
411 * from the kernel lists and can be freed or reused once this function
414 * The packet type might still be in use by receivers
415 * and must not be freed until after all the CPU's have gone
416 * through a quiescent state.
418 void __dev_remove_pack(struct packet_type *pt)
420 struct list_head *head = ptype_head(pt);
421 struct packet_type *pt1;
423 spin_lock(&ptype_lock);
425 list_for_each_entry(pt1, head, list) {
427 list_del_rcu(&pt->list);
432 pr_warn("dev_remove_pack: %p not found\n", pt);
434 spin_unlock(&ptype_lock);
436 EXPORT_SYMBOL(__dev_remove_pack);
439 * dev_remove_pack - remove packet handler
440 * @pt: packet type declaration
442 * Remove a protocol handler that was previously added to the kernel
443 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
444 * from the kernel lists and can be freed or reused once this function
447 * This call sleeps to guarantee that no CPU is looking at the packet
450 void dev_remove_pack(struct packet_type *pt)
452 __dev_remove_pack(pt);
456 EXPORT_SYMBOL(dev_remove_pack);
460 * dev_add_offload - register offload handlers
461 * @po: protocol offload declaration
463 * Add protocol offload handlers to the networking stack. The passed
464 * &proto_offload is linked into kernel lists and may not be freed until
465 * it has been removed from the kernel lists.
467 * This call does not sleep therefore it can not
468 * guarantee all CPU's that are in middle of receiving packets
469 * will see the new offload handlers (until the next received packet).
471 void dev_add_offload(struct packet_offload *po)
473 struct packet_offload *elem;
475 spin_lock(&offload_lock);
476 list_for_each_entry(elem, &offload_base, list) {
477 if (po->priority < elem->priority)
480 list_add_rcu(&po->list, elem->list.prev);
481 spin_unlock(&offload_lock);
483 EXPORT_SYMBOL(dev_add_offload);
486 * __dev_remove_offload - remove offload handler
487 * @po: packet offload declaration
489 * Remove a protocol offload handler that was previously added to the
490 * kernel offload handlers by dev_add_offload(). The passed &offload_type
491 * is removed from the kernel lists and can be freed or reused once this
494 * The packet type might still be in use by receivers
495 * and must not be freed until after all the CPU's have gone
496 * through a quiescent state.
498 static void __dev_remove_offload(struct packet_offload *po)
500 struct list_head *head = &offload_base;
501 struct packet_offload *po1;
503 spin_lock(&offload_lock);
505 list_for_each_entry(po1, head, list) {
507 list_del_rcu(&po->list);
512 pr_warn("dev_remove_offload: %p not found\n", po);
514 spin_unlock(&offload_lock);
518 * dev_remove_offload - remove packet offload handler
519 * @po: packet offload declaration
521 * Remove a packet offload handler that was previously added to the kernel
522 * offload handlers by dev_add_offload(). The passed &offload_type is
523 * removed from the kernel lists and can be freed or reused once this
526 * This call sleeps to guarantee that no CPU is looking at the packet
529 void dev_remove_offload(struct packet_offload *po)
531 __dev_remove_offload(po);
535 EXPORT_SYMBOL(dev_remove_offload);
537 /******************************************************************************
539 Device Boot-time Settings Routines
541 *******************************************************************************/
543 /* Boot time configuration table */
544 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
547 * netdev_boot_setup_add - add new setup entry
548 * @name: name of the device
549 * @map: configured settings for the device
551 * Adds new setup entry to the dev_boot_setup list. The function
552 * returns 0 on error and 1 on success. This is a generic routine to
555 static int netdev_boot_setup_add(char *name, struct ifmap *map)
557 struct netdev_boot_setup *s;
561 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
562 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
563 memset(s[i].name, 0, sizeof(s[i].name));
564 strlcpy(s[i].name, name, IFNAMSIZ);
565 memcpy(&s[i].map, map, sizeof(s[i].map));
570 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
574 * netdev_boot_setup_check - check boot time settings
575 * @dev: the netdevice
577 * Check boot time settings for the device.
578 * The found settings are set for the device to be used
579 * later in the device probing.
580 * Returns 0 if no settings found, 1 if they are.
582 int netdev_boot_setup_check(struct net_device *dev)
584 struct netdev_boot_setup *s = dev_boot_setup;
587 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
588 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
589 !strcmp(dev->name, s[i].name)) {
590 dev->irq = s[i].map.irq;
591 dev->base_addr = s[i].map.base_addr;
592 dev->mem_start = s[i].map.mem_start;
593 dev->mem_end = s[i].map.mem_end;
599 EXPORT_SYMBOL(netdev_boot_setup_check);
603 * netdev_boot_base - get address from boot time settings
604 * @prefix: prefix for network device
605 * @unit: id for network device
607 * Check boot time settings for the base address of device.
608 * The found settings are set for the device to be used
609 * later in the device probing.
610 * Returns 0 if no settings found.
612 unsigned long netdev_boot_base(const char *prefix, int unit)
614 const struct netdev_boot_setup *s = dev_boot_setup;
618 sprintf(name, "%s%d", prefix, unit);
621 * If device already registered then return base of 1
622 * to indicate not to probe for this interface
624 if (__dev_get_by_name(&init_net, name))
627 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
628 if (!strcmp(name, s[i].name))
629 return s[i].map.base_addr;
634 * Saves at boot time configured settings for any netdevice.
636 int __init netdev_boot_setup(char *str)
641 str = get_options(str, ARRAY_SIZE(ints), ints);
646 memset(&map, 0, sizeof(map));
650 map.base_addr = ints[2];
652 map.mem_start = ints[3];
654 map.mem_end = ints[4];
656 /* Add new entry to the list */
657 return netdev_boot_setup_add(str, &map);
660 __setup("netdev=", netdev_boot_setup);
662 /*******************************************************************************
664 Device Interface Subroutines
666 *******************************************************************************/
669 * dev_get_iflink - get 'iflink' value of a interface
670 * @dev: targeted interface
672 * Indicates the ifindex the interface is linked to.
673 * Physical interfaces have the same 'ifindex' and 'iflink' values.
676 int dev_get_iflink(const struct net_device *dev)
678 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
679 return dev->netdev_ops->ndo_get_iflink(dev);
683 EXPORT_SYMBOL(dev_get_iflink);
686 * dev_fill_metadata_dst - Retrieve tunnel egress information.
687 * @dev: targeted interface
690 * For better visibility of tunnel traffic OVS needs to retrieve
691 * egress tunnel information for a packet. Following API allows
692 * user to get this info.
694 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
696 struct ip_tunnel_info *info;
698 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
701 info = skb_tunnel_info_unclone(skb);
704 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
707 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
709 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
712 * __dev_get_by_name - find a device by its name
713 * @net: the applicable net namespace
714 * @name: name to find
716 * Find an interface by name. Must be called under RTNL semaphore
717 * or @dev_base_lock. If the name is found a pointer to the device
718 * is returned. If the name is not found then %NULL is returned. The
719 * reference counters are not incremented so the caller must be
720 * careful with locks.
723 struct net_device *__dev_get_by_name(struct net *net, const char *name)
725 struct net_device *dev;
726 struct hlist_head *head = dev_name_hash(net, name);
728 hlist_for_each_entry(dev, head, name_hlist)
729 if (!strncmp(dev->name, name, IFNAMSIZ))
734 EXPORT_SYMBOL(__dev_get_by_name);
737 * dev_get_by_name_rcu - find a device by its name
738 * @net: the applicable net namespace
739 * @name: name to find
741 * Find an interface by name.
742 * If the name is found a pointer to the device is returned.
743 * If the name is not found then %NULL is returned.
744 * The reference counters are not incremented so the caller must be
745 * careful with locks. The caller must hold RCU lock.
748 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
750 struct net_device *dev;
751 struct hlist_head *head = dev_name_hash(net, name);
753 hlist_for_each_entry_rcu(dev, head, name_hlist)
754 if (!strncmp(dev->name, name, IFNAMSIZ))
759 EXPORT_SYMBOL(dev_get_by_name_rcu);
762 * dev_get_by_name - find a device by its name
763 * @net: the applicable net namespace
764 * @name: name to find
766 * Find an interface by name. This can be called from any
767 * context and does its own locking. The returned handle has
768 * the usage count incremented and the caller must use dev_put() to
769 * release it when it is no longer needed. %NULL is returned if no
770 * matching device is found.
773 struct net_device *dev_get_by_name(struct net *net, const char *name)
775 struct net_device *dev;
778 dev = dev_get_by_name_rcu(net, name);
784 EXPORT_SYMBOL(dev_get_by_name);
787 * __dev_get_by_index - find a device by its ifindex
788 * @net: the applicable net namespace
789 * @ifindex: index of device
791 * Search for an interface by index. Returns %NULL if the device
792 * is not found or a pointer to the device. The device has not
793 * had its reference counter increased so the caller must be careful
794 * about locking. The caller must hold either the RTNL semaphore
798 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
800 struct net_device *dev;
801 struct hlist_head *head = dev_index_hash(net, ifindex);
803 hlist_for_each_entry(dev, head, index_hlist)
804 if (dev->ifindex == ifindex)
809 EXPORT_SYMBOL(__dev_get_by_index);
812 * dev_get_by_index_rcu - find a device by its ifindex
813 * @net: the applicable net namespace
814 * @ifindex: index of device
816 * Search for an interface by index. Returns %NULL if the device
817 * is not found or a pointer to the device. The device has not
818 * had its reference counter increased so the caller must be careful
819 * about locking. The caller must hold RCU lock.
822 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
824 struct net_device *dev;
825 struct hlist_head *head = dev_index_hash(net, ifindex);
827 hlist_for_each_entry_rcu(dev, head, index_hlist)
828 if (dev->ifindex == ifindex)
833 EXPORT_SYMBOL(dev_get_by_index_rcu);
837 * dev_get_by_index - find a device by its ifindex
838 * @net: the applicable net namespace
839 * @ifindex: index of device
841 * Search for an interface by index. Returns NULL if the device
842 * is not found or a pointer to the device. The device returned has
843 * had a reference added and the pointer is safe until the user calls
844 * dev_put to indicate they have finished with it.
847 struct net_device *dev_get_by_index(struct net *net, int ifindex)
849 struct net_device *dev;
852 dev = dev_get_by_index_rcu(net, ifindex);
858 EXPORT_SYMBOL(dev_get_by_index);
861 * netdev_get_name - get a netdevice name, knowing its ifindex.
862 * @net: network namespace
863 * @name: a pointer to the buffer where the name will be stored.
864 * @ifindex: the ifindex of the interface to get the name from.
866 * The use of raw_seqcount_begin() and cond_resched() before
867 * retrying is required as we want to give the writers a chance
868 * to complete when CONFIG_PREEMPT is not set.
870 int netdev_get_name(struct net *net, char *name, int ifindex)
872 struct net_device *dev;
876 seq = raw_seqcount_begin(&devnet_rename_seq);
878 dev = dev_get_by_index_rcu(net, ifindex);
884 strcpy(name, dev->name);
886 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
895 * dev_getbyhwaddr_rcu - find a device by its hardware address
896 * @net: the applicable net namespace
897 * @type: media type of device
898 * @ha: hardware address
900 * Search for an interface by MAC address. Returns NULL if the device
901 * is not found or a pointer to the device.
902 * The caller must hold RCU or RTNL.
903 * The returned device has not had its ref count increased
904 * and the caller must therefore be careful about locking
908 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
911 struct net_device *dev;
913 for_each_netdev_rcu(net, dev)
914 if (dev->type == type &&
915 !memcmp(dev->dev_addr, ha, dev->addr_len))
920 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
922 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
924 struct net_device *dev;
927 for_each_netdev(net, dev)
928 if (dev->type == type)
933 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
935 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
937 struct net_device *dev, *ret = NULL;
940 for_each_netdev_rcu(net, dev)
941 if (dev->type == type) {
949 EXPORT_SYMBOL(dev_getfirstbyhwtype);
952 * __dev_get_by_flags - find any device with given flags
953 * @net: the applicable net namespace
954 * @if_flags: IFF_* values
955 * @mask: bitmask of bits in if_flags to check
957 * Search for any interface with the given flags. Returns NULL if a device
958 * is not found or a pointer to the device. Must be called inside
959 * rtnl_lock(), and result refcount is unchanged.
962 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
965 struct net_device *dev, *ret;
970 for_each_netdev(net, dev) {
971 if (((dev->flags ^ if_flags) & mask) == 0) {
978 EXPORT_SYMBOL(__dev_get_by_flags);
981 * dev_valid_name - check if name is okay for network device
984 * Network device names need to be valid file names to
985 * to allow sysfs to work. We also disallow any kind of
988 bool dev_valid_name(const char *name)
992 if (strlen(name) >= IFNAMSIZ)
994 if (!strcmp(name, ".") || !strcmp(name, ".."))
998 if (*name == '/' || *name == ':' || isspace(*name))
1004 EXPORT_SYMBOL(dev_valid_name);
1007 * __dev_alloc_name - allocate a name for a device
1008 * @net: network namespace to allocate the device name in
1009 * @name: name format string
1010 * @buf: scratch buffer and result name string
1012 * Passed a format string - eg "lt%d" it will try and find a suitable
1013 * id. It scans list of devices to build up a free map, then chooses
1014 * the first empty slot. The caller must hold the dev_base or rtnl lock
1015 * while allocating the name and adding the device in order to avoid
1017 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1018 * Returns the number of the unit assigned or a negative errno code.
1021 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1025 const int max_netdevices = 8*PAGE_SIZE;
1026 unsigned long *inuse;
1027 struct net_device *d;
1029 p = strnchr(name, IFNAMSIZ-1, '%');
1032 * Verify the string as this thing may have come from
1033 * the user. There must be either one "%d" and no other "%"
1036 if (p[1] != 'd' || strchr(p + 2, '%'))
1039 /* Use one page as a bit array of possible slots */
1040 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1044 for_each_netdev(net, d) {
1045 if (!sscanf(d->name, name, &i))
1047 if (i < 0 || i >= max_netdevices)
1050 /* avoid cases where sscanf is not exact inverse of printf */
1051 snprintf(buf, IFNAMSIZ, name, i);
1052 if (!strncmp(buf, d->name, IFNAMSIZ))
1056 i = find_first_zero_bit(inuse, max_netdevices);
1057 free_page((unsigned long) inuse);
1061 snprintf(buf, IFNAMSIZ, name, i);
1062 if (!__dev_get_by_name(net, buf))
1065 /* It is possible to run out of possible slots
1066 * when the name is long and there isn't enough space left
1067 * for the digits, or if all bits are used.
1073 * dev_alloc_name - allocate a name for a device
1075 * @name: name format string
1077 * Passed a format string - eg "lt%d" it will try and find a suitable
1078 * id. It scans list of devices to build up a free map, then chooses
1079 * the first empty slot. The caller must hold the dev_base or rtnl lock
1080 * while allocating the name and adding the device in order to avoid
1082 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1083 * Returns the number of the unit assigned or a negative errno code.
1086 int dev_alloc_name(struct net_device *dev, const char *name)
1092 BUG_ON(!dev_net(dev));
1094 ret = __dev_alloc_name(net, name, buf);
1096 strlcpy(dev->name, buf, IFNAMSIZ);
1099 EXPORT_SYMBOL(dev_alloc_name);
1101 static int dev_alloc_name_ns(struct net *net,
1102 struct net_device *dev,
1108 ret = __dev_alloc_name(net, name, buf);
1110 strlcpy(dev->name, buf, IFNAMSIZ);
1114 static int dev_get_valid_name(struct net *net,
1115 struct net_device *dev,
1120 if (!dev_valid_name(name))
1123 if (strchr(name, '%'))
1124 return dev_alloc_name_ns(net, dev, name);
1125 else if (__dev_get_by_name(net, name))
1127 else if (dev->name != name)
1128 strlcpy(dev->name, name, IFNAMSIZ);
1134 * dev_change_name - change name of a device
1136 * @newname: name (or format string) must be at least IFNAMSIZ
1138 * Change name of a device, can pass format strings "eth%d".
1141 int dev_change_name(struct net_device *dev, const char *newname)
1143 unsigned char old_assign_type;
1144 char oldname[IFNAMSIZ];
1150 BUG_ON(!dev_net(dev));
1153 if (dev->flags & IFF_UP)
1156 write_seqcount_begin(&devnet_rename_seq);
1158 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1159 write_seqcount_end(&devnet_rename_seq);
1163 memcpy(oldname, dev->name, IFNAMSIZ);
1165 err = dev_get_valid_name(net, dev, newname);
1167 write_seqcount_end(&devnet_rename_seq);
1171 if (oldname[0] && !strchr(oldname, '%'))
1172 netdev_info(dev, "renamed from %s\n", oldname);
1174 old_assign_type = dev->name_assign_type;
1175 dev->name_assign_type = NET_NAME_RENAMED;
1178 ret = device_rename(&dev->dev, dev->name);
1180 memcpy(dev->name, oldname, IFNAMSIZ);
1181 dev->name_assign_type = old_assign_type;
1182 write_seqcount_end(&devnet_rename_seq);
1186 write_seqcount_end(&devnet_rename_seq);
1188 netdev_adjacent_rename_links(dev, oldname);
1190 write_lock_bh(&dev_base_lock);
1191 hlist_del_rcu(&dev->name_hlist);
1192 write_unlock_bh(&dev_base_lock);
1196 write_lock_bh(&dev_base_lock);
1197 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1198 write_unlock_bh(&dev_base_lock);
1200 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1201 ret = notifier_to_errno(ret);
1204 /* err >= 0 after dev_alloc_name() or stores the first errno */
1207 write_seqcount_begin(&devnet_rename_seq);
1208 memcpy(dev->name, oldname, IFNAMSIZ);
1209 memcpy(oldname, newname, IFNAMSIZ);
1210 dev->name_assign_type = old_assign_type;
1211 old_assign_type = NET_NAME_RENAMED;
1214 pr_err("%s: name change rollback failed: %d\n",
1223 * dev_set_alias - change ifalias of a device
1225 * @alias: name up to IFALIASZ
1226 * @len: limit of bytes to copy from info
1228 * Set ifalias for a device,
1230 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1236 if (len >= IFALIASZ)
1240 kfree(dev->ifalias);
1241 dev->ifalias = NULL;
1245 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1248 dev->ifalias = new_ifalias;
1250 strlcpy(dev->ifalias, alias, len+1);
1256 * netdev_features_change - device changes features
1257 * @dev: device to cause notification
1259 * Called to indicate a device has changed features.
1261 void netdev_features_change(struct net_device *dev)
1263 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1265 EXPORT_SYMBOL(netdev_features_change);
1268 * netdev_state_change - device changes state
1269 * @dev: device to cause notification
1271 * Called to indicate a device has changed state. This function calls
1272 * the notifier chains for netdev_chain and sends a NEWLINK message
1273 * to the routing socket.
1275 void netdev_state_change(struct net_device *dev)
1277 if (dev->flags & IFF_UP) {
1278 struct netdev_notifier_change_info change_info;
1280 change_info.flags_changed = 0;
1281 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
1283 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1286 EXPORT_SYMBOL(netdev_state_change);
1289 * netdev_notify_peers - notify network peers about existence of @dev
1290 * @dev: network device
1292 * Generate traffic such that interested network peers are aware of
1293 * @dev, such as by generating a gratuitous ARP. This may be used when
1294 * a device wants to inform the rest of the network about some sort of
1295 * reconfiguration such as a failover event or virtual machine
1298 void netdev_notify_peers(struct net_device *dev)
1301 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1304 EXPORT_SYMBOL(netdev_notify_peers);
1306 static int __dev_open(struct net_device *dev)
1308 const struct net_device_ops *ops = dev->netdev_ops;
1313 if (!netif_device_present(dev))
1316 /* Block netpoll from trying to do any rx path servicing.
1317 * If we don't do this there is a chance ndo_poll_controller
1318 * or ndo_poll may be running while we open the device
1320 netpoll_poll_disable(dev);
1322 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1323 ret = notifier_to_errno(ret);
1327 set_bit(__LINK_STATE_START, &dev->state);
1329 if (ops->ndo_validate_addr)
1330 ret = ops->ndo_validate_addr(dev);
1332 if (!ret && ops->ndo_open)
1333 ret = ops->ndo_open(dev);
1335 netpoll_poll_enable(dev);
1338 clear_bit(__LINK_STATE_START, &dev->state);
1340 dev->flags |= IFF_UP;
1341 dev_set_rx_mode(dev);
1343 add_device_randomness(dev->dev_addr, dev->addr_len);
1350 * dev_open - prepare an interface for use.
1351 * @dev: device to open
1353 * Takes a device from down to up state. The device's private open
1354 * function is invoked and then the multicast lists are loaded. Finally
1355 * the device is moved into the up state and a %NETDEV_UP message is
1356 * sent to the netdev notifier chain.
1358 * Calling this function on an active interface is a nop. On a failure
1359 * a negative errno code is returned.
1361 int dev_open(struct net_device *dev)
1365 if (dev->flags & IFF_UP)
1368 ret = __dev_open(dev);
1372 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1373 call_netdevice_notifiers(NETDEV_UP, dev);
1377 EXPORT_SYMBOL(dev_open);
1379 static int __dev_close_many(struct list_head *head)
1381 struct net_device *dev;
1386 list_for_each_entry(dev, head, close_list) {
1387 /* Temporarily disable netpoll until the interface is down */
1388 netpoll_poll_disable(dev);
1390 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1392 clear_bit(__LINK_STATE_START, &dev->state);
1394 /* Synchronize to scheduled poll. We cannot touch poll list, it
1395 * can be even on different cpu. So just clear netif_running().
1397 * dev->stop() will invoke napi_disable() on all of it's
1398 * napi_struct instances on this device.
1400 smp_mb__after_atomic(); /* Commit netif_running(). */
1403 dev_deactivate_many(head);
1405 list_for_each_entry(dev, head, close_list) {
1406 const struct net_device_ops *ops = dev->netdev_ops;
1409 * Call the device specific close. This cannot fail.
1410 * Only if device is UP
1412 * We allow it to be called even after a DETACH hot-plug
1418 dev->flags &= ~IFF_UP;
1419 netpoll_poll_enable(dev);
1425 static int __dev_close(struct net_device *dev)
1430 list_add(&dev->close_list, &single);
1431 retval = __dev_close_many(&single);
1437 int dev_close_many(struct list_head *head, bool unlink)
1439 struct net_device *dev, *tmp;
1441 /* Remove the devices that don't need to be closed */
1442 list_for_each_entry_safe(dev, tmp, head, close_list)
1443 if (!(dev->flags & IFF_UP))
1444 list_del_init(&dev->close_list);
1446 __dev_close_many(head);
1448 list_for_each_entry_safe(dev, tmp, head, close_list) {
1449 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1450 call_netdevice_notifiers(NETDEV_DOWN, dev);
1452 list_del_init(&dev->close_list);
1457 EXPORT_SYMBOL(dev_close_many);
1460 * dev_close - shutdown an interface.
1461 * @dev: device to shutdown
1463 * This function moves an active device into down state. A
1464 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1465 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1468 int dev_close(struct net_device *dev)
1470 if (dev->flags & IFF_UP) {
1473 list_add(&dev->close_list, &single);
1474 dev_close_many(&single, true);
1479 EXPORT_SYMBOL(dev_close);
1483 * dev_disable_lro - disable Large Receive Offload on a device
1486 * Disable Large Receive Offload (LRO) on a net device. Must be
1487 * called under RTNL. This is needed if received packets may be
1488 * forwarded to another interface.
1490 void dev_disable_lro(struct net_device *dev)
1492 struct net_device *lower_dev;
1493 struct list_head *iter;
1495 dev->wanted_features &= ~NETIF_F_LRO;
1496 netdev_update_features(dev);
1498 if (unlikely(dev->features & NETIF_F_LRO))
1499 netdev_WARN(dev, "failed to disable LRO!\n");
1501 netdev_for_each_lower_dev(dev, lower_dev, iter)
1502 dev_disable_lro(lower_dev);
1504 EXPORT_SYMBOL(dev_disable_lro);
1506 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1507 struct net_device *dev)
1509 struct netdev_notifier_info info;
1511 netdev_notifier_info_init(&info, dev);
1512 return nb->notifier_call(nb, val, &info);
1515 static int dev_boot_phase = 1;
1518 * register_netdevice_notifier - register a network notifier block
1521 * Register a notifier to be called when network device events occur.
1522 * The notifier passed is linked into the kernel structures and must
1523 * not be reused until it has been unregistered. A negative errno code
1524 * is returned on a failure.
1526 * When registered all registration and up events are replayed
1527 * to the new notifier to allow device to have a race free
1528 * view of the network device list.
1531 int register_netdevice_notifier(struct notifier_block *nb)
1533 struct net_device *dev;
1534 struct net_device *last;
1539 err = raw_notifier_chain_register(&netdev_chain, nb);
1545 for_each_netdev(net, dev) {
1546 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1547 err = notifier_to_errno(err);
1551 if (!(dev->flags & IFF_UP))
1554 call_netdevice_notifier(nb, NETDEV_UP, dev);
1565 for_each_netdev(net, dev) {
1569 if (dev->flags & IFF_UP) {
1570 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1572 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1574 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1579 raw_notifier_chain_unregister(&netdev_chain, nb);
1582 EXPORT_SYMBOL(register_netdevice_notifier);
1585 * unregister_netdevice_notifier - unregister a network notifier block
1588 * Unregister a notifier previously registered by
1589 * register_netdevice_notifier(). The notifier is unlinked into the
1590 * kernel structures and may then be reused. A negative errno code
1591 * is returned on a failure.
1593 * After unregistering unregister and down device events are synthesized
1594 * for all devices on the device list to the removed notifier to remove
1595 * the need for special case cleanup code.
1598 int unregister_netdevice_notifier(struct notifier_block *nb)
1600 struct net_device *dev;
1605 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1610 for_each_netdev(net, dev) {
1611 if (dev->flags & IFF_UP) {
1612 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1614 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1616 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1623 EXPORT_SYMBOL(unregister_netdevice_notifier);
1626 * call_netdevice_notifiers_info - call all network notifier blocks
1627 * @val: value passed unmodified to notifier function
1628 * @dev: net_device pointer passed unmodified to notifier function
1629 * @info: notifier information data
1631 * Call all network notifier blocks. Parameters and return value
1632 * are as for raw_notifier_call_chain().
1635 static int call_netdevice_notifiers_info(unsigned long val,
1636 struct net_device *dev,
1637 struct netdev_notifier_info *info)
1640 netdev_notifier_info_init(info, dev);
1641 return raw_notifier_call_chain(&netdev_chain, val, info);
1645 * call_netdevice_notifiers - call all network notifier blocks
1646 * @val: value passed unmodified to notifier function
1647 * @dev: net_device pointer passed unmodified to notifier function
1649 * Call all network notifier blocks. Parameters and return value
1650 * are as for raw_notifier_call_chain().
1653 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1655 struct netdev_notifier_info info;
1657 return call_netdevice_notifiers_info(val, dev, &info);
1659 EXPORT_SYMBOL(call_netdevice_notifiers);
1661 #ifdef CONFIG_NET_INGRESS
1662 static struct static_key ingress_needed __read_mostly;
1664 void net_inc_ingress_queue(void)
1666 static_key_slow_inc(&ingress_needed);
1668 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
1670 void net_dec_ingress_queue(void)
1672 static_key_slow_dec(&ingress_needed);
1674 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
1677 static struct static_key netstamp_needed __read_mostly;
1678 #ifdef HAVE_JUMP_LABEL
1679 /* We are not allowed to call static_key_slow_dec() from irq context
1680 * If net_disable_timestamp() is called from irq context, defer the
1681 * static_key_slow_dec() calls.
1683 static atomic_t netstamp_needed_deferred;
1686 void net_enable_timestamp(void)
1688 #ifdef HAVE_JUMP_LABEL
1689 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1693 static_key_slow_dec(&netstamp_needed);
1697 static_key_slow_inc(&netstamp_needed);
1699 EXPORT_SYMBOL(net_enable_timestamp);
1701 void net_disable_timestamp(void)
1703 #ifdef HAVE_JUMP_LABEL
1704 if (in_interrupt()) {
1705 atomic_inc(&netstamp_needed_deferred);
1709 static_key_slow_dec(&netstamp_needed);
1711 EXPORT_SYMBOL(net_disable_timestamp);
1713 static inline void net_timestamp_set(struct sk_buff *skb)
1715 skb->tstamp.tv64 = 0;
1716 if (static_key_false(&netstamp_needed))
1717 __net_timestamp(skb);
1720 #define net_timestamp_check(COND, SKB) \
1721 if (static_key_false(&netstamp_needed)) { \
1722 if ((COND) && !(SKB)->tstamp.tv64) \
1723 __net_timestamp(SKB); \
1726 bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb)
1730 if (!(dev->flags & IFF_UP))
1733 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1734 if (skb->len <= len)
1737 /* if TSO is enabled, we don't care about the length as the packet
1738 * could be forwarded without being segmented before
1740 if (skb_is_gso(skb))
1745 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1747 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1749 if (skb_orphan_frags(skb, GFP_ATOMIC) ||
1750 unlikely(!is_skb_forwardable(dev, skb))) {
1751 atomic_long_inc(&dev->rx_dropped);
1756 skb_scrub_packet(skb, true);
1758 skb->protocol = eth_type_trans(skb, dev);
1759 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1763 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1766 * dev_forward_skb - loopback an skb to another netif
1768 * @dev: destination network device
1769 * @skb: buffer to forward
1772 * NET_RX_SUCCESS (no congestion)
1773 * NET_RX_DROP (packet was dropped, but freed)
1775 * dev_forward_skb can be used for injecting an skb from the
1776 * start_xmit function of one device into the receive queue
1777 * of another device.
1779 * The receiving device may be in another namespace, so
1780 * we have to clear all information in the skb that could
1781 * impact namespace isolation.
1783 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1785 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1787 EXPORT_SYMBOL_GPL(dev_forward_skb);
1789 static inline int deliver_skb(struct sk_buff *skb,
1790 struct packet_type *pt_prev,
1791 struct net_device *orig_dev)
1793 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1795 atomic_inc(&skb->users);
1796 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1799 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1800 struct packet_type **pt,
1801 struct net_device *orig_dev,
1803 struct list_head *ptype_list)
1805 struct packet_type *ptype, *pt_prev = *pt;
1807 list_for_each_entry_rcu(ptype, ptype_list, list) {
1808 if (ptype->type != type)
1811 deliver_skb(skb, pt_prev, orig_dev);
1817 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1819 if (!ptype->af_packet_priv || !skb->sk)
1822 if (ptype->id_match)
1823 return ptype->id_match(ptype, skb->sk);
1824 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1831 * Support routine. Sends outgoing frames to any network
1832 * taps currently in use.
1835 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1837 struct packet_type *ptype;
1838 struct sk_buff *skb2 = NULL;
1839 struct packet_type *pt_prev = NULL;
1840 struct list_head *ptype_list = &ptype_all;
1844 list_for_each_entry_rcu(ptype, ptype_list, list) {
1845 /* Never send packets back to the socket
1846 * they originated from - MvS (miquels@drinkel.ow.org)
1848 if (skb_loop_sk(ptype, skb))
1852 deliver_skb(skb2, pt_prev, skb->dev);
1857 /* need to clone skb, done only once */
1858 skb2 = skb_clone(skb, GFP_ATOMIC);
1862 net_timestamp_set(skb2);
1864 /* skb->nh should be correctly
1865 * set by sender, so that the second statement is
1866 * just protection against buggy protocols.
1868 skb_reset_mac_header(skb2);
1870 if (skb_network_header(skb2) < skb2->data ||
1871 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1872 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1873 ntohs(skb2->protocol),
1875 skb_reset_network_header(skb2);
1878 skb2->transport_header = skb2->network_header;
1879 skb2->pkt_type = PACKET_OUTGOING;
1883 if (ptype_list == &ptype_all) {
1884 ptype_list = &dev->ptype_all;
1889 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1894 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1895 * @dev: Network device
1896 * @txq: number of queues available
1898 * If real_num_tx_queues is changed the tc mappings may no longer be
1899 * valid. To resolve this verify the tc mapping remains valid and if
1900 * not NULL the mapping. With no priorities mapping to this
1901 * offset/count pair it will no longer be used. In the worst case TC0
1902 * is invalid nothing can be done so disable priority mappings. If is
1903 * expected that drivers will fix this mapping if they can before
1904 * calling netif_set_real_num_tx_queues.
1906 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1909 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1911 /* If TC0 is invalidated disable TC mapping */
1912 if (tc->offset + tc->count > txq) {
1913 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1918 /* Invalidated prio to tc mappings set to TC0 */
1919 for (i = 1; i < TC_BITMASK + 1; i++) {
1920 int q = netdev_get_prio_tc_map(dev, i);
1922 tc = &dev->tc_to_txq[q];
1923 if (tc->offset + tc->count > txq) {
1924 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1926 netdev_set_prio_tc_map(dev, i, 0);
1932 static DEFINE_MUTEX(xps_map_mutex);
1933 #define xmap_dereference(P) \
1934 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1936 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1939 struct xps_map *map = NULL;
1943 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1945 for (pos = 0; map && pos < map->len; pos++) {
1946 if (map->queues[pos] == index) {
1948 map->queues[pos] = map->queues[--map->len];
1950 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1951 kfree_rcu(map, rcu);
1961 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1963 struct xps_dev_maps *dev_maps;
1965 bool active = false;
1967 mutex_lock(&xps_map_mutex);
1968 dev_maps = xmap_dereference(dev->xps_maps);
1973 for_each_possible_cpu(cpu) {
1974 for (i = index; i < dev->num_tx_queues; i++) {
1975 if (!remove_xps_queue(dev_maps, cpu, i))
1978 if (i == dev->num_tx_queues)
1983 RCU_INIT_POINTER(dev->xps_maps, NULL);
1984 kfree_rcu(dev_maps, rcu);
1987 for (i = index; i < dev->num_tx_queues; i++)
1988 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1992 mutex_unlock(&xps_map_mutex);
1995 static struct xps_map *expand_xps_map(struct xps_map *map,
1998 struct xps_map *new_map;
1999 int alloc_len = XPS_MIN_MAP_ALLOC;
2002 for (pos = 0; map && pos < map->len; pos++) {
2003 if (map->queues[pos] != index)
2008 /* Need to add queue to this CPU's existing map */
2010 if (pos < map->alloc_len)
2013 alloc_len = map->alloc_len * 2;
2016 /* Need to allocate new map to store queue on this CPU's map */
2017 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2022 for (i = 0; i < pos; i++)
2023 new_map->queues[i] = map->queues[i];
2024 new_map->alloc_len = alloc_len;
2030 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2033 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2034 struct xps_map *map, *new_map;
2035 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
2036 int cpu, numa_node_id = -2;
2037 bool active = false;
2039 mutex_lock(&xps_map_mutex);
2041 dev_maps = xmap_dereference(dev->xps_maps);
2043 /* allocate memory for queue storage */
2044 for_each_online_cpu(cpu) {
2045 if (!cpumask_test_cpu(cpu, mask))
2049 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2050 if (!new_dev_maps) {
2051 mutex_unlock(&xps_map_mutex);
2055 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2058 map = expand_xps_map(map, cpu, index);
2062 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2066 goto out_no_new_maps;
2068 for_each_possible_cpu(cpu) {
2069 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
2070 /* add queue to CPU maps */
2073 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2074 while ((pos < map->len) && (map->queues[pos] != index))
2077 if (pos == map->len)
2078 map->queues[map->len++] = index;
2080 if (numa_node_id == -2)
2081 numa_node_id = cpu_to_node(cpu);
2082 else if (numa_node_id != cpu_to_node(cpu))
2085 } else if (dev_maps) {
2086 /* fill in the new device map from the old device map */
2087 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2088 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2093 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2095 /* Cleanup old maps */
2097 for_each_possible_cpu(cpu) {
2098 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2099 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2100 if (map && map != new_map)
2101 kfree_rcu(map, rcu);
2104 kfree_rcu(dev_maps, rcu);
2107 dev_maps = new_dev_maps;
2111 /* update Tx queue numa node */
2112 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2113 (numa_node_id >= 0) ? numa_node_id :
2119 /* removes queue from unused CPUs */
2120 for_each_possible_cpu(cpu) {
2121 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2124 if (remove_xps_queue(dev_maps, cpu, index))
2128 /* free map if not active */
2130 RCU_INIT_POINTER(dev->xps_maps, NULL);
2131 kfree_rcu(dev_maps, rcu);
2135 mutex_unlock(&xps_map_mutex);
2139 /* remove any maps that we added */
2140 for_each_possible_cpu(cpu) {
2141 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2142 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2144 if (new_map && new_map != map)
2148 mutex_unlock(&xps_map_mutex);
2150 kfree(new_dev_maps);
2153 EXPORT_SYMBOL(netif_set_xps_queue);
2157 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2158 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2160 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2164 if (txq < 1 || txq > dev->num_tx_queues)
2167 if (dev->reg_state == NETREG_REGISTERED ||
2168 dev->reg_state == NETREG_UNREGISTERING) {
2171 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2177 netif_setup_tc(dev, txq);
2179 if (txq < dev->real_num_tx_queues) {
2180 qdisc_reset_all_tx_gt(dev, txq);
2182 netif_reset_xps_queues_gt(dev, txq);
2187 dev->real_num_tx_queues = txq;
2190 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2194 * netif_set_real_num_rx_queues - set actual number of RX queues used
2195 * @dev: Network device
2196 * @rxq: Actual number of RX queues
2198 * This must be called either with the rtnl_lock held or before
2199 * registration of the net device. Returns 0 on success, or a
2200 * negative error code. If called before registration, it always
2203 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2207 if (rxq < 1 || rxq > dev->num_rx_queues)
2210 if (dev->reg_state == NETREG_REGISTERED) {
2213 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2219 dev->real_num_rx_queues = rxq;
2222 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2226 * netif_get_num_default_rss_queues - default number of RSS queues
2228 * This routine should set an upper limit on the number of RSS queues
2229 * used by default by multiqueue devices.
2231 int netif_get_num_default_rss_queues(void)
2233 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2235 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2237 static inline void __netif_reschedule(struct Qdisc *q)
2239 struct softnet_data *sd;
2240 unsigned long flags;
2242 local_irq_save(flags);
2243 sd = this_cpu_ptr(&softnet_data);
2244 q->next_sched = NULL;
2245 *sd->output_queue_tailp = q;
2246 sd->output_queue_tailp = &q->next_sched;
2247 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2248 local_irq_restore(flags);
2251 void __netif_schedule(struct Qdisc *q)
2253 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2254 __netif_reschedule(q);
2256 EXPORT_SYMBOL(__netif_schedule);
2258 struct dev_kfree_skb_cb {
2259 enum skb_free_reason reason;
2262 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2264 return (struct dev_kfree_skb_cb *)skb->cb;
2267 void netif_schedule_queue(struct netdev_queue *txq)
2270 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2271 struct Qdisc *q = rcu_dereference(txq->qdisc);
2273 __netif_schedule(q);
2277 EXPORT_SYMBOL(netif_schedule_queue);
2280 * netif_wake_subqueue - allow sending packets on subqueue
2281 * @dev: network device
2282 * @queue_index: sub queue index
2284 * Resume individual transmit queue of a device with multiple transmit queues.
2286 void netif_wake_subqueue(struct net_device *dev, u16 queue_index)
2288 struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index);
2290 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) {
2294 q = rcu_dereference(txq->qdisc);
2295 __netif_schedule(q);
2299 EXPORT_SYMBOL(netif_wake_subqueue);
2301 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2303 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2307 q = rcu_dereference(dev_queue->qdisc);
2308 __netif_schedule(q);
2312 EXPORT_SYMBOL(netif_tx_wake_queue);
2314 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2316 unsigned long flags;
2318 if (likely(atomic_read(&skb->users) == 1)) {
2320 atomic_set(&skb->users, 0);
2321 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2324 get_kfree_skb_cb(skb)->reason = reason;
2325 local_irq_save(flags);
2326 skb->next = __this_cpu_read(softnet_data.completion_queue);
2327 __this_cpu_write(softnet_data.completion_queue, skb);
2328 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2329 local_irq_restore(flags);
2331 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2333 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2335 if (in_irq() || irqs_disabled())
2336 __dev_kfree_skb_irq(skb, reason);
2340 EXPORT_SYMBOL(__dev_kfree_skb_any);
2344 * netif_device_detach - mark device as removed
2345 * @dev: network device
2347 * Mark device as removed from system and therefore no longer available.
2349 void netif_device_detach(struct net_device *dev)
2351 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2352 netif_running(dev)) {
2353 netif_tx_stop_all_queues(dev);
2356 EXPORT_SYMBOL(netif_device_detach);
2359 * netif_device_attach - mark device as attached
2360 * @dev: network device
2362 * Mark device as attached from system and restart if needed.
2364 void netif_device_attach(struct net_device *dev)
2366 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2367 netif_running(dev)) {
2368 netif_tx_wake_all_queues(dev);
2369 __netdev_watchdog_up(dev);
2372 EXPORT_SYMBOL(netif_device_attach);
2375 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2376 * to be used as a distribution range.
2378 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
2379 unsigned int num_tx_queues)
2383 u16 qcount = num_tx_queues;
2385 if (skb_rx_queue_recorded(skb)) {
2386 hash = skb_get_rx_queue(skb);
2387 while (unlikely(hash >= num_tx_queues))
2388 hash -= num_tx_queues;
2393 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2394 qoffset = dev->tc_to_txq[tc].offset;
2395 qcount = dev->tc_to_txq[tc].count;
2398 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
2400 EXPORT_SYMBOL(__skb_tx_hash);
2402 static void skb_warn_bad_offload(const struct sk_buff *skb)
2404 static const netdev_features_t null_features = 0;
2405 struct net_device *dev = skb->dev;
2406 const char *name = "";
2408 if (!net_ratelimit())
2412 if (dev->dev.parent)
2413 name = dev_driver_string(dev->dev.parent);
2415 name = netdev_name(dev);
2417 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2418 "gso_type=%d ip_summed=%d\n",
2419 name, dev ? &dev->features : &null_features,
2420 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2421 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2422 skb_shinfo(skb)->gso_type, skb->ip_summed);
2426 * Invalidate hardware checksum when packet is to be mangled, and
2427 * complete checksum manually on outgoing path.
2429 int skb_checksum_help(struct sk_buff *skb)
2432 int ret = 0, offset;
2434 if (skb->ip_summed == CHECKSUM_COMPLETE)
2435 goto out_set_summed;
2437 if (unlikely(skb_shinfo(skb)->gso_size)) {
2438 skb_warn_bad_offload(skb);
2442 /* Before computing a checksum, we should make sure no frag could
2443 * be modified by an external entity : checksum could be wrong.
2445 if (skb_has_shared_frag(skb)) {
2446 ret = __skb_linearize(skb);
2451 offset = skb_checksum_start_offset(skb);
2452 BUG_ON(offset >= skb_headlen(skb));
2453 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2455 offset += skb->csum_offset;
2456 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2458 if (skb_cloned(skb) &&
2459 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2460 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2465 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2467 skb->ip_summed = CHECKSUM_NONE;
2471 EXPORT_SYMBOL(skb_checksum_help);
2473 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2475 __be16 type = skb->protocol;
2477 /* Tunnel gso handlers can set protocol to ethernet. */
2478 if (type == htons(ETH_P_TEB)) {
2481 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2484 eth = (struct ethhdr *)skb_mac_header(skb);
2485 type = eth->h_proto;
2488 return __vlan_get_protocol(skb, type, depth);
2492 * skb_mac_gso_segment - mac layer segmentation handler.
2493 * @skb: buffer to segment
2494 * @features: features for the output path (see dev->features)
2496 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2497 netdev_features_t features)
2499 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2500 struct packet_offload *ptype;
2501 int vlan_depth = skb->mac_len;
2502 __be16 type = skb_network_protocol(skb, &vlan_depth);
2504 if (unlikely(!type))
2505 return ERR_PTR(-EINVAL);
2507 __skb_pull(skb, vlan_depth);
2510 list_for_each_entry_rcu(ptype, &offload_base, list) {
2511 if (ptype->type == type && ptype->callbacks.gso_segment) {
2512 segs = ptype->callbacks.gso_segment(skb, features);
2518 __skb_push(skb, skb->data - skb_mac_header(skb));
2522 EXPORT_SYMBOL(skb_mac_gso_segment);
2525 /* openvswitch calls this on rx path, so we need a different check.
2527 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2530 return skb->ip_summed != CHECKSUM_PARTIAL;
2532 return skb->ip_summed == CHECKSUM_NONE;
2536 * __skb_gso_segment - Perform segmentation on skb.
2537 * @skb: buffer to segment
2538 * @features: features for the output path (see dev->features)
2539 * @tx_path: whether it is called in TX path
2541 * This function segments the given skb and returns a list of segments.
2543 * It may return NULL if the skb requires no segmentation. This is
2544 * only possible when GSO is used for verifying header integrity.
2546 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
2548 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2549 netdev_features_t features, bool tx_path)
2551 if (unlikely(skb_needs_check(skb, tx_path))) {
2554 skb_warn_bad_offload(skb);
2556 err = skb_cow_head(skb, 0);
2558 return ERR_PTR(err);
2561 BUILD_BUG_ON(SKB_SGO_CB_OFFSET +
2562 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
2564 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2565 SKB_GSO_CB(skb)->encap_level = 0;
2567 skb_reset_mac_header(skb);
2568 skb_reset_mac_len(skb);
2570 return skb_mac_gso_segment(skb, features);
2572 EXPORT_SYMBOL(__skb_gso_segment);
2574 /* Take action when hardware reception checksum errors are detected. */
2576 void netdev_rx_csum_fault(struct net_device *dev)
2578 if (net_ratelimit()) {
2579 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2583 EXPORT_SYMBOL(netdev_rx_csum_fault);
2586 /* Actually, we should eliminate this check as soon as we know, that:
2587 * 1. IOMMU is present and allows to map all the memory.
2588 * 2. No high memory really exists on this machine.
2591 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2593 #ifdef CONFIG_HIGHMEM
2595 if (!(dev->features & NETIF_F_HIGHDMA)) {
2596 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2597 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2598 if (PageHighMem(skb_frag_page(frag)))
2603 if (PCI_DMA_BUS_IS_PHYS) {
2604 struct device *pdev = dev->dev.parent;
2608 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2609 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2610 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2611 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2619 /* If MPLS offload request, verify we are testing hardware MPLS features
2620 * instead of standard features for the netdev.
2622 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
2623 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2624 netdev_features_t features,
2627 if (eth_p_mpls(type))
2628 features &= skb->dev->mpls_features;
2633 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2634 netdev_features_t features,
2641 static netdev_features_t harmonize_features(struct sk_buff *skb,
2642 netdev_features_t features)
2647 type = skb_network_protocol(skb, &tmp);
2648 features = net_mpls_features(skb, features, type);
2650 if (skb->ip_summed != CHECKSUM_NONE &&
2651 !can_checksum_protocol(features, type)) {
2652 features &= ~NETIF_F_ALL_CSUM;
2653 } else if (illegal_highdma(skb->dev, skb)) {
2654 features &= ~NETIF_F_SG;
2660 netdev_features_t passthru_features_check(struct sk_buff *skb,
2661 struct net_device *dev,
2662 netdev_features_t features)
2666 EXPORT_SYMBOL(passthru_features_check);
2668 static netdev_features_t dflt_features_check(const struct sk_buff *skb,
2669 struct net_device *dev,
2670 netdev_features_t features)
2672 return vlan_features_check(skb, features);
2675 netdev_features_t netif_skb_features(struct sk_buff *skb)
2677 struct net_device *dev = skb->dev;
2678 netdev_features_t features = dev->features;
2679 u16 gso_segs = skb_shinfo(skb)->gso_segs;
2681 if (gso_segs > dev->gso_max_segs || gso_segs < dev->gso_min_segs)
2682 features &= ~NETIF_F_GSO_MASK;
2684 /* If encapsulation offload request, verify we are testing
2685 * hardware encapsulation features instead of standard
2686 * features for the netdev
2688 if (skb->encapsulation)
2689 features &= dev->hw_enc_features;
2691 if (skb_vlan_tagged(skb))
2692 features = netdev_intersect_features(features,
2693 dev->vlan_features |
2694 NETIF_F_HW_VLAN_CTAG_TX |
2695 NETIF_F_HW_VLAN_STAG_TX);
2697 if (dev->netdev_ops->ndo_features_check)
2698 features &= dev->netdev_ops->ndo_features_check(skb, dev,
2701 features &= dflt_features_check(skb, dev, features);
2703 return harmonize_features(skb, features);
2705 EXPORT_SYMBOL(netif_skb_features);
2707 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
2708 struct netdev_queue *txq, bool more)
2713 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
2714 dev_queue_xmit_nit(skb, dev);
2717 trace_net_dev_start_xmit(skb, dev);
2718 rc = netdev_start_xmit(skb, dev, txq, more);
2719 trace_net_dev_xmit(skb, rc, dev, len);
2724 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
2725 struct netdev_queue *txq, int *ret)
2727 struct sk_buff *skb = first;
2728 int rc = NETDEV_TX_OK;
2731 struct sk_buff *next = skb->next;
2734 rc = xmit_one(skb, dev, txq, next != NULL);
2735 if (unlikely(!dev_xmit_complete(rc))) {
2741 if (netif_xmit_stopped(txq) && skb) {
2742 rc = NETDEV_TX_BUSY;
2752 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
2753 netdev_features_t features)
2755 if (skb_vlan_tag_present(skb) &&
2756 !vlan_hw_offload_capable(features, skb->vlan_proto))
2757 skb = __vlan_hwaccel_push_inside(skb);
2761 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
2763 netdev_features_t features;
2768 features = netif_skb_features(skb);
2769 skb = validate_xmit_vlan(skb, features);
2773 if (netif_needs_gso(skb, features)) {
2774 struct sk_buff *segs;
2776 segs = skb_gso_segment(skb, features);
2784 if (skb_needs_linearize(skb, features) &&
2785 __skb_linearize(skb))
2788 /* If packet is not checksummed and device does not
2789 * support checksumming for this protocol, complete
2790 * checksumming here.
2792 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2793 if (skb->encapsulation)
2794 skb_set_inner_transport_header(skb,
2795 skb_checksum_start_offset(skb));
2797 skb_set_transport_header(skb,
2798 skb_checksum_start_offset(skb));
2799 if (!(features & NETIF_F_ALL_CSUM) &&
2800 skb_checksum_help(skb))
2813 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev)
2815 struct sk_buff *next, *head = NULL, *tail;
2817 for (; skb != NULL; skb = next) {
2821 /* in case skb wont be segmented, point to itself */
2824 skb = validate_xmit_skb(skb, dev);
2832 /* If skb was segmented, skb->prev points to
2833 * the last segment. If not, it still contains skb.
2840 static void qdisc_pkt_len_init(struct sk_buff *skb)
2842 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2844 qdisc_skb_cb(skb)->pkt_len = skb->len;
2846 /* To get more precise estimation of bytes sent on wire,
2847 * we add to pkt_len the headers size of all segments
2849 if (shinfo->gso_size) {
2850 unsigned int hdr_len;
2851 u16 gso_segs = shinfo->gso_segs;
2853 /* mac layer + network layer */
2854 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2856 /* + transport layer */
2857 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2858 hdr_len += tcp_hdrlen(skb);
2860 hdr_len += sizeof(struct udphdr);
2862 if (shinfo->gso_type & SKB_GSO_DODGY)
2863 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2866 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2870 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2871 struct net_device *dev,
2872 struct netdev_queue *txq)
2874 spinlock_t *root_lock = qdisc_lock(q);
2878 qdisc_pkt_len_init(skb);
2879 qdisc_calculate_pkt_len(skb, q);
2881 * Heuristic to force contended enqueues to serialize on a
2882 * separate lock before trying to get qdisc main lock.
2883 * This permits __QDISC___STATE_RUNNING owner to get the lock more
2884 * often and dequeue packets faster.
2886 contended = qdisc_is_running(q);
2887 if (unlikely(contended))
2888 spin_lock(&q->busylock);
2890 spin_lock(root_lock);
2891 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2894 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2895 qdisc_run_begin(q)) {
2897 * This is a work-conserving queue; there are no old skbs
2898 * waiting to be sent out; and the qdisc is not running -
2899 * xmit the skb directly.
2902 qdisc_bstats_update(q, skb);
2904 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
2905 if (unlikely(contended)) {
2906 spin_unlock(&q->busylock);
2913 rc = NET_XMIT_SUCCESS;
2915 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2916 if (qdisc_run_begin(q)) {
2917 if (unlikely(contended)) {
2918 spin_unlock(&q->busylock);
2924 spin_unlock(root_lock);
2925 if (unlikely(contended))
2926 spin_unlock(&q->busylock);
2930 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2931 static void skb_update_prio(struct sk_buff *skb)
2933 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2935 if (!skb->priority && skb->sk && map) {
2936 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2938 if (prioidx < map->priomap_len)
2939 skb->priority = map->priomap[prioidx];
2943 #define skb_update_prio(skb)
2946 DEFINE_PER_CPU(int, xmit_recursion);
2947 EXPORT_SYMBOL(xmit_recursion);
2949 #define RECURSION_LIMIT 10
2952 * dev_loopback_xmit - loop back @skb
2953 * @net: network namespace this loopback is happening in
2954 * @sk: sk needed to be a netfilter okfn
2955 * @skb: buffer to transmit
2957 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
2959 skb_reset_mac_header(skb);
2960 __skb_pull(skb, skb_network_offset(skb));
2961 skb->pkt_type = PACKET_LOOPBACK;
2962 skb->ip_summed = CHECKSUM_UNNECESSARY;
2963 WARN_ON(!skb_dst(skb));
2968 EXPORT_SYMBOL(dev_loopback_xmit);
2970 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2973 struct xps_dev_maps *dev_maps;
2974 struct xps_map *map;
2975 int queue_index = -1;
2978 dev_maps = rcu_dereference(dev->xps_maps);
2980 map = rcu_dereference(
2981 dev_maps->cpu_map[skb->sender_cpu - 1]);
2984 queue_index = map->queues[0];
2986 queue_index = map->queues[reciprocal_scale(skb_get_hash(skb),
2988 if (unlikely(queue_index >= dev->real_num_tx_queues))
3000 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb)
3002 struct sock *sk = skb->sk;
3003 int queue_index = sk_tx_queue_get(sk);
3005 if (queue_index < 0 || skb->ooo_okay ||
3006 queue_index >= dev->real_num_tx_queues) {
3007 int new_index = get_xps_queue(dev, skb);
3009 new_index = skb_tx_hash(dev, skb);
3011 if (queue_index != new_index && sk &&
3013 rcu_access_pointer(sk->sk_dst_cache))
3014 sk_tx_queue_set(sk, new_index);
3016 queue_index = new_index;
3022 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
3023 struct sk_buff *skb,
3026 int queue_index = 0;
3029 if (skb->sender_cpu == 0)
3030 skb->sender_cpu = raw_smp_processor_id() + 1;
3033 if (dev->real_num_tx_queues != 1) {
3034 const struct net_device_ops *ops = dev->netdev_ops;
3035 if (ops->ndo_select_queue)
3036 queue_index = ops->ndo_select_queue(dev, skb, accel_priv,
3039 queue_index = __netdev_pick_tx(dev, skb);
3042 queue_index = netdev_cap_txqueue(dev, queue_index);
3045 skb_set_queue_mapping(skb, queue_index);
3046 return netdev_get_tx_queue(dev, queue_index);
3050 * __dev_queue_xmit - transmit a buffer
3051 * @skb: buffer to transmit
3052 * @accel_priv: private data used for L2 forwarding offload
3054 * Queue a buffer for transmission to a network device. The caller must
3055 * have set the device and priority and built the buffer before calling
3056 * this function. The function can be called from an interrupt.
3058 * A negative errno code is returned on a failure. A success does not
3059 * guarantee the frame will be transmitted as it may be dropped due
3060 * to congestion or traffic shaping.
3062 * -----------------------------------------------------------------------------------
3063 * I notice this method can also return errors from the queue disciplines,
3064 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3067 * Regardless of the return value, the skb is consumed, so it is currently
3068 * difficult to retry a send to this method. (You can bump the ref count
3069 * before sending to hold a reference for retry if you are careful.)
3071 * When calling this method, interrupts MUST be enabled. This is because
3072 * the BH enable code must have IRQs enabled so that it will not deadlock.
3075 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
3077 struct net_device *dev = skb->dev;
3078 struct netdev_queue *txq;
3082 skb_reset_mac_header(skb);
3084 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3085 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3087 /* Disable soft irqs for various locks below. Also
3088 * stops preemption for RCU.
3092 skb_update_prio(skb);
3094 /* If device/qdisc don't need skb->dst, release it right now while
3095 * its hot in this cpu cache.
3097 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3102 #ifdef CONFIG_NET_SWITCHDEV
3103 /* Don't forward if offload device already forwarded */
3104 if (skb->offload_fwd_mark &&
3105 skb->offload_fwd_mark == dev->offload_fwd_mark) {
3107 rc = NET_XMIT_SUCCESS;
3112 txq = netdev_pick_tx(dev, skb, accel_priv);
3113 q = rcu_dereference_bh(txq->qdisc);
3115 #ifdef CONFIG_NET_CLS_ACT
3116 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
3118 trace_net_dev_queue(skb);
3120 rc = __dev_xmit_skb(skb, q, dev, txq);
3124 /* The device has no queue. Common case for software devices:
3125 loopback, all the sorts of tunnels...
3127 Really, it is unlikely that netif_tx_lock protection is necessary
3128 here. (f.e. loopback and IP tunnels are clean ignoring statistics
3130 However, it is possible, that they rely on protection
3133 Check this and shot the lock. It is not prone from deadlocks.
3134 Either shot noqueue qdisc, it is even simpler 8)
3136 if (dev->flags & IFF_UP) {
3137 int cpu = smp_processor_id(); /* ok because BHs are off */
3139 if (txq->xmit_lock_owner != cpu) {
3141 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
3142 goto recursion_alert;
3144 skb = validate_xmit_skb(skb, dev);
3148 HARD_TX_LOCK(dev, txq, cpu);
3150 if (!netif_xmit_stopped(txq)) {
3151 __this_cpu_inc(xmit_recursion);
3152 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
3153 __this_cpu_dec(xmit_recursion);
3154 if (dev_xmit_complete(rc)) {
3155 HARD_TX_UNLOCK(dev, txq);
3159 HARD_TX_UNLOCK(dev, txq);
3160 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3163 /* Recursion is detected! It is possible,
3167 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3174 rcu_read_unlock_bh();
3176 atomic_long_inc(&dev->tx_dropped);
3177 kfree_skb_list(skb);
3180 rcu_read_unlock_bh();
3184 int dev_queue_xmit(struct sk_buff *skb)
3186 return __dev_queue_xmit(skb, NULL);
3188 EXPORT_SYMBOL(dev_queue_xmit);
3190 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
3192 return __dev_queue_xmit(skb, accel_priv);
3194 EXPORT_SYMBOL(dev_queue_xmit_accel);
3197 /*=======================================================================
3199 =======================================================================*/
3201 int netdev_max_backlog __read_mostly = 1000;
3202 EXPORT_SYMBOL(netdev_max_backlog);
3204 int netdev_tstamp_prequeue __read_mostly = 1;
3205 int netdev_budget __read_mostly = 300;
3206 int weight_p __read_mostly = 64; /* old backlog weight */
3208 /* Called with irq disabled */
3209 static inline void ____napi_schedule(struct softnet_data *sd,
3210 struct napi_struct *napi)
3212 list_add_tail(&napi->poll_list, &sd->poll_list);
3213 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3218 /* One global table that all flow-based protocols share. */
3219 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3220 EXPORT_SYMBOL(rps_sock_flow_table);
3221 u32 rps_cpu_mask __read_mostly;
3222 EXPORT_SYMBOL(rps_cpu_mask);
3224 struct static_key rps_needed __read_mostly;
3226 static struct rps_dev_flow *
3227 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3228 struct rps_dev_flow *rflow, u16 next_cpu)
3230 if (next_cpu < nr_cpu_ids) {
3231 #ifdef CONFIG_RFS_ACCEL
3232 struct netdev_rx_queue *rxqueue;
3233 struct rps_dev_flow_table *flow_table;
3234 struct rps_dev_flow *old_rflow;
3239 /* Should we steer this flow to a different hardware queue? */
3240 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3241 !(dev->features & NETIF_F_NTUPLE))
3243 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3244 if (rxq_index == skb_get_rx_queue(skb))
3247 rxqueue = dev->_rx + rxq_index;
3248 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3251 flow_id = skb_get_hash(skb) & flow_table->mask;
3252 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3253 rxq_index, flow_id);
3257 rflow = &flow_table->flows[flow_id];
3259 if (old_rflow->filter == rflow->filter)
3260 old_rflow->filter = RPS_NO_FILTER;
3264 per_cpu(softnet_data, next_cpu).input_queue_head;
3267 rflow->cpu = next_cpu;
3272 * get_rps_cpu is called from netif_receive_skb and returns the target
3273 * CPU from the RPS map of the receiving queue for a given skb.
3274 * rcu_read_lock must be held on entry.
3276 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3277 struct rps_dev_flow **rflowp)
3279 const struct rps_sock_flow_table *sock_flow_table;
3280 struct netdev_rx_queue *rxqueue = dev->_rx;
3281 struct rps_dev_flow_table *flow_table;
3282 struct rps_map *map;
3287 if (skb_rx_queue_recorded(skb)) {
3288 u16 index = skb_get_rx_queue(skb);
3290 if (unlikely(index >= dev->real_num_rx_queues)) {
3291 WARN_ONCE(dev->real_num_rx_queues > 1,
3292 "%s received packet on queue %u, but number "
3293 "of RX queues is %u\n",
3294 dev->name, index, dev->real_num_rx_queues);
3300 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3302 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3303 map = rcu_dereference(rxqueue->rps_map);
3304 if (!flow_table && !map)
3307 skb_reset_network_header(skb);
3308 hash = skb_get_hash(skb);
3312 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3313 if (flow_table && sock_flow_table) {
3314 struct rps_dev_flow *rflow;
3318 /* First check into global flow table if there is a match */
3319 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
3320 if ((ident ^ hash) & ~rps_cpu_mask)
3323 next_cpu = ident & rps_cpu_mask;
3325 /* OK, now we know there is a match,
3326 * we can look at the local (per receive queue) flow table
3328 rflow = &flow_table->flows[hash & flow_table->mask];
3332 * If the desired CPU (where last recvmsg was done) is
3333 * different from current CPU (one in the rx-queue flow
3334 * table entry), switch if one of the following holds:
3335 * - Current CPU is unset (>= nr_cpu_ids).
3336 * - Current CPU is offline.
3337 * - The current CPU's queue tail has advanced beyond the
3338 * last packet that was enqueued using this table entry.
3339 * This guarantees that all previous packets for the flow
3340 * have been dequeued, thus preserving in order delivery.
3342 if (unlikely(tcpu != next_cpu) &&
3343 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
3344 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3345 rflow->last_qtail)) >= 0)) {
3347 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3350 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
3360 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3361 if (cpu_online(tcpu)) {
3371 #ifdef CONFIG_RFS_ACCEL
3374 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3375 * @dev: Device on which the filter was set
3376 * @rxq_index: RX queue index
3377 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3378 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3380 * Drivers that implement ndo_rx_flow_steer() should periodically call
3381 * this function for each installed filter and remove the filters for
3382 * which it returns %true.
3384 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3385 u32 flow_id, u16 filter_id)
3387 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3388 struct rps_dev_flow_table *flow_table;
3389 struct rps_dev_flow *rflow;
3394 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3395 if (flow_table && flow_id <= flow_table->mask) {
3396 rflow = &flow_table->flows[flow_id];
3397 cpu = ACCESS_ONCE(rflow->cpu);
3398 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
3399 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3400 rflow->last_qtail) <
3401 (int)(10 * flow_table->mask)))
3407 EXPORT_SYMBOL(rps_may_expire_flow);
3409 #endif /* CONFIG_RFS_ACCEL */
3411 /* Called from hardirq (IPI) context */
3412 static void rps_trigger_softirq(void *data)
3414 struct softnet_data *sd = data;
3416 ____napi_schedule(sd, &sd->backlog);
3420 #endif /* CONFIG_RPS */
3423 * Check if this softnet_data structure is another cpu one
3424 * If yes, queue it to our IPI list and return 1
3427 static int rps_ipi_queued(struct softnet_data *sd)
3430 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3433 sd->rps_ipi_next = mysd->rps_ipi_list;
3434 mysd->rps_ipi_list = sd;
3436 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3439 #endif /* CONFIG_RPS */
3443 #ifdef CONFIG_NET_FLOW_LIMIT
3444 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3447 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3449 #ifdef CONFIG_NET_FLOW_LIMIT
3450 struct sd_flow_limit *fl;
3451 struct softnet_data *sd;
3452 unsigned int old_flow, new_flow;
3454 if (qlen < (netdev_max_backlog >> 1))
3457 sd = this_cpu_ptr(&softnet_data);
3460 fl = rcu_dereference(sd->flow_limit);
3462 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3463 old_flow = fl->history[fl->history_head];
3464 fl->history[fl->history_head] = new_flow;
3467 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3469 if (likely(fl->buckets[old_flow]))
3470 fl->buckets[old_flow]--;
3472 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3484 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3485 * queue (may be a remote CPU queue).
3487 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3488 unsigned int *qtail)
3490 struct softnet_data *sd;
3491 unsigned long flags;
3494 sd = &per_cpu(softnet_data, cpu);
3496 local_irq_save(flags);
3499 if (!netif_running(skb->dev))
3501 qlen = skb_queue_len(&sd->input_pkt_queue);
3502 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3505 __skb_queue_tail(&sd->input_pkt_queue, skb);
3506 input_queue_tail_incr_save(sd, qtail);
3508 local_irq_restore(flags);
3509 return NET_RX_SUCCESS;
3512 /* Schedule NAPI for backlog device
3513 * We can use non atomic operation since we own the queue lock
3515 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3516 if (!rps_ipi_queued(sd))
3517 ____napi_schedule(sd, &sd->backlog);
3526 local_irq_restore(flags);
3528 atomic_long_inc(&skb->dev->rx_dropped);
3533 static int netif_rx_internal(struct sk_buff *skb)
3537 net_timestamp_check(netdev_tstamp_prequeue, skb);
3539 trace_netif_rx(skb);
3541 if (static_key_false(&rps_needed)) {
3542 struct rps_dev_flow voidflow, *rflow = &voidflow;
3548 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3550 cpu = smp_processor_id();
3552 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3560 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3567 * netif_rx - post buffer to the network code
3568 * @skb: buffer to post
3570 * This function receives a packet from a device driver and queues it for
3571 * the upper (protocol) levels to process. It always succeeds. The buffer
3572 * may be dropped during processing for congestion control or by the
3576 * NET_RX_SUCCESS (no congestion)
3577 * NET_RX_DROP (packet was dropped)
3581 int netif_rx(struct sk_buff *skb)
3583 trace_netif_rx_entry(skb);
3585 return netif_rx_internal(skb);
3587 EXPORT_SYMBOL(netif_rx);
3589 int netif_rx_ni(struct sk_buff *skb)
3593 trace_netif_rx_ni_entry(skb);
3596 err = netif_rx_internal(skb);
3597 if (local_softirq_pending())
3603 EXPORT_SYMBOL(netif_rx_ni);
3605 static void net_tx_action(struct softirq_action *h)
3607 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3609 if (sd->completion_queue) {
3610 struct sk_buff *clist;
3612 local_irq_disable();
3613 clist = sd->completion_queue;
3614 sd->completion_queue = NULL;
3618 struct sk_buff *skb = clist;
3619 clist = clist->next;
3621 WARN_ON(atomic_read(&skb->users));
3622 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3623 trace_consume_skb(skb);
3625 trace_kfree_skb(skb, net_tx_action);
3630 if (sd->output_queue) {
3633 local_irq_disable();
3634 head = sd->output_queue;
3635 sd->output_queue = NULL;
3636 sd->output_queue_tailp = &sd->output_queue;
3640 struct Qdisc *q = head;
3641 spinlock_t *root_lock;
3643 head = head->next_sched;
3645 root_lock = qdisc_lock(q);
3646 if (spin_trylock(root_lock)) {
3647 smp_mb__before_atomic();
3648 clear_bit(__QDISC_STATE_SCHED,
3651 spin_unlock(root_lock);
3653 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3655 __netif_reschedule(q);
3657 smp_mb__before_atomic();
3658 clear_bit(__QDISC_STATE_SCHED,
3666 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3667 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3668 /* This hook is defined here for ATM LANE */
3669 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3670 unsigned char *addr) __read_mostly;
3671 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3674 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3675 struct packet_type **pt_prev,
3676 int *ret, struct net_device *orig_dev)
3678 #ifdef CONFIG_NET_CLS_ACT
3679 struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list);
3680 struct tcf_result cl_res;
3682 /* If there's at least one ingress present somewhere (so
3683 * we get here via enabled static key), remaining devices
3684 * that are not configured with an ingress qdisc will bail
3690 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3694 qdisc_skb_cb(skb)->pkt_len = skb->len;
3695 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3696 qdisc_bstats_cpu_update(cl->q, skb);
3698 switch (tc_classify(skb, cl, &cl_res, false)) {
3700 case TC_ACT_RECLASSIFY:
3701 skb->tc_index = TC_H_MIN(cl_res.classid);
3704 qdisc_qstats_cpu_drop(cl->q);
3709 case TC_ACT_REDIRECT:
3710 /* skb_mac_header check was done by cls/act_bpf, so
3711 * we can safely push the L2 header back before
3712 * redirecting to another netdev
3714 __skb_push(skb, skb->mac_len);
3715 skb_do_redirect(skb);
3720 #endif /* CONFIG_NET_CLS_ACT */
3725 * netdev_rx_handler_register - register receive handler
3726 * @dev: device to register a handler for
3727 * @rx_handler: receive handler to register
3728 * @rx_handler_data: data pointer that is used by rx handler
3730 * Register a receive handler for a device. This handler will then be
3731 * called from __netif_receive_skb. A negative errno code is returned
3734 * The caller must hold the rtnl_mutex.
3736 * For a general description of rx_handler, see enum rx_handler_result.
3738 int netdev_rx_handler_register(struct net_device *dev,
3739 rx_handler_func_t *rx_handler,
3740 void *rx_handler_data)
3744 if (dev->rx_handler)
3747 /* Note: rx_handler_data must be set before rx_handler */
3748 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3749 rcu_assign_pointer(dev->rx_handler, rx_handler);
3753 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3756 * netdev_rx_handler_unregister - unregister receive handler
3757 * @dev: device to unregister a handler from
3759 * Unregister a receive handler from a device.
3761 * The caller must hold the rtnl_mutex.
3763 void netdev_rx_handler_unregister(struct net_device *dev)
3767 RCU_INIT_POINTER(dev->rx_handler, NULL);
3768 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3769 * section has a guarantee to see a non NULL rx_handler_data
3773 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3775 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3778 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3779 * the special handling of PFMEMALLOC skbs.
3781 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3783 switch (skb->protocol) {
3784 case htons(ETH_P_ARP):
3785 case htons(ETH_P_IP):
3786 case htons(ETH_P_IPV6):
3787 case htons(ETH_P_8021Q):
3788 case htons(ETH_P_8021AD):
3795 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
3796 int *ret, struct net_device *orig_dev)
3798 #ifdef CONFIG_NETFILTER_INGRESS
3799 if (nf_hook_ingress_active(skb)) {
3801 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3805 return nf_hook_ingress(skb);
3807 #endif /* CONFIG_NETFILTER_INGRESS */
3811 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3813 struct packet_type *ptype, *pt_prev;
3814 rx_handler_func_t *rx_handler;
3815 struct net_device *orig_dev;
3816 bool deliver_exact = false;
3817 int ret = NET_RX_DROP;
3820 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3822 trace_netif_receive_skb(skb);
3824 orig_dev = skb->dev;
3826 skb_reset_network_header(skb);
3827 if (!skb_transport_header_was_set(skb))
3828 skb_reset_transport_header(skb);
3829 skb_reset_mac_len(skb);
3834 skb->skb_iif = skb->dev->ifindex;
3836 __this_cpu_inc(softnet_data.processed);
3838 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3839 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3840 skb = skb_vlan_untag(skb);
3845 #ifdef CONFIG_NET_CLS_ACT
3846 if (skb->tc_verd & TC_NCLS) {
3847 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3855 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3857 ret = deliver_skb(skb, pt_prev, orig_dev);
3861 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
3863 ret = deliver_skb(skb, pt_prev, orig_dev);
3868 #ifdef CONFIG_NET_INGRESS
3869 if (static_key_false(&ingress_needed)) {
3870 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3874 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
3878 #ifdef CONFIG_NET_CLS_ACT
3882 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3885 if (skb_vlan_tag_present(skb)) {
3887 ret = deliver_skb(skb, pt_prev, orig_dev);
3890 if (vlan_do_receive(&skb))
3892 else if (unlikely(!skb))
3896 rx_handler = rcu_dereference(skb->dev->rx_handler);
3899 ret = deliver_skb(skb, pt_prev, orig_dev);
3902 switch (rx_handler(&skb)) {
3903 case RX_HANDLER_CONSUMED:
3904 ret = NET_RX_SUCCESS;
3906 case RX_HANDLER_ANOTHER:
3908 case RX_HANDLER_EXACT:
3909 deliver_exact = true;
3910 case RX_HANDLER_PASS:
3917 if (unlikely(skb_vlan_tag_present(skb))) {
3918 if (skb_vlan_tag_get_id(skb))
3919 skb->pkt_type = PACKET_OTHERHOST;
3920 /* Note: we might in the future use prio bits
3921 * and set skb->priority like in vlan_do_receive()
3922 * For the time being, just ignore Priority Code Point
3927 type = skb->protocol;
3929 /* deliver only exact match when indicated */
3930 if (likely(!deliver_exact)) {
3931 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3932 &ptype_base[ntohs(type) &
3936 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3937 &orig_dev->ptype_specific);
3939 if (unlikely(skb->dev != orig_dev)) {
3940 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3941 &skb->dev->ptype_specific);
3945 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3948 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3951 atomic_long_inc(&skb->dev->rx_dropped);
3953 /* Jamal, now you will not able to escape explaining
3954 * me how you were going to use this. :-)
3963 static int __netif_receive_skb(struct sk_buff *skb)
3967 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3968 unsigned long pflags = current->flags;
3971 * PFMEMALLOC skbs are special, they should
3972 * - be delivered to SOCK_MEMALLOC sockets only
3973 * - stay away from userspace
3974 * - have bounded memory usage
3976 * Use PF_MEMALLOC as this saves us from propagating the allocation
3977 * context down to all allocation sites.
3979 current->flags |= PF_MEMALLOC;
3980 ret = __netif_receive_skb_core(skb, true);
3981 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3983 ret = __netif_receive_skb_core(skb, false);
3988 static int netif_receive_skb_internal(struct sk_buff *skb)
3992 net_timestamp_check(netdev_tstamp_prequeue, skb);
3994 if (skb_defer_rx_timestamp(skb))
3995 return NET_RX_SUCCESS;
4000 if (static_key_false(&rps_needed)) {
4001 struct rps_dev_flow voidflow, *rflow = &voidflow;
4002 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
4005 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4011 ret = __netif_receive_skb(skb);
4017 * netif_receive_skb - process receive buffer from network
4018 * @skb: buffer to process
4020 * netif_receive_skb() is the main receive data processing function.
4021 * It always succeeds. The buffer may be dropped during processing
4022 * for congestion control or by the protocol layers.
4024 * This function may only be called from softirq context and interrupts
4025 * should be enabled.
4027 * Return values (usually ignored):
4028 * NET_RX_SUCCESS: no congestion
4029 * NET_RX_DROP: packet was dropped
4031 int netif_receive_skb(struct sk_buff *skb)
4033 trace_netif_receive_skb_entry(skb);
4035 return netif_receive_skb_internal(skb);
4037 EXPORT_SYMBOL(netif_receive_skb);
4039 /* Network device is going away, flush any packets still pending
4040 * Called with irqs disabled.
4042 static void flush_backlog(void *arg)
4044 struct net_device *dev = arg;
4045 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4046 struct sk_buff *skb, *tmp;
4049 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
4050 if (skb->dev == dev) {
4051 __skb_unlink(skb, &sd->input_pkt_queue);
4053 input_queue_head_incr(sd);
4058 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
4059 if (skb->dev == dev) {
4060 __skb_unlink(skb, &sd->process_queue);
4062 input_queue_head_incr(sd);
4067 static int napi_gro_complete(struct sk_buff *skb)
4069 struct packet_offload *ptype;
4070 __be16 type = skb->protocol;
4071 struct list_head *head = &offload_base;
4074 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
4076 if (NAPI_GRO_CB(skb)->count == 1) {
4077 skb_shinfo(skb)->gso_size = 0;
4082 list_for_each_entry_rcu(ptype, head, list) {
4083 if (ptype->type != type || !ptype->callbacks.gro_complete)
4086 err = ptype->callbacks.gro_complete(skb, 0);
4092 WARN_ON(&ptype->list == head);
4094 return NET_RX_SUCCESS;
4098 return netif_receive_skb_internal(skb);
4101 /* napi->gro_list contains packets ordered by age.
4102 * youngest packets at the head of it.
4103 * Complete skbs in reverse order to reduce latencies.
4105 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
4107 struct sk_buff *skb, *prev = NULL;
4109 /* scan list and build reverse chain */
4110 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
4115 for (skb = prev; skb; skb = prev) {
4118 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
4122 napi_gro_complete(skb);
4126 napi->gro_list = NULL;
4128 EXPORT_SYMBOL(napi_gro_flush);
4130 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
4133 unsigned int maclen = skb->dev->hard_header_len;
4134 u32 hash = skb_get_hash_raw(skb);
4136 for (p = napi->gro_list; p; p = p->next) {
4137 unsigned long diffs;
4139 NAPI_GRO_CB(p)->flush = 0;
4141 if (hash != skb_get_hash_raw(p)) {
4142 NAPI_GRO_CB(p)->same_flow = 0;
4146 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
4147 diffs |= p->vlan_tci ^ skb->vlan_tci;
4148 diffs |= skb_metadata_dst_cmp(p, skb);
4149 if (maclen == ETH_HLEN)
4150 diffs |= compare_ether_header(skb_mac_header(p),
4151 skb_mac_header(skb));
4153 diffs = memcmp(skb_mac_header(p),
4154 skb_mac_header(skb),
4156 NAPI_GRO_CB(p)->same_flow = !diffs;
4160 static void skb_gro_reset_offset(struct sk_buff *skb)
4162 const struct skb_shared_info *pinfo = skb_shinfo(skb);
4163 const skb_frag_t *frag0 = &pinfo->frags[0];
4165 NAPI_GRO_CB(skb)->data_offset = 0;
4166 NAPI_GRO_CB(skb)->frag0 = NULL;
4167 NAPI_GRO_CB(skb)->frag0_len = 0;
4169 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
4171 !PageHighMem(skb_frag_page(frag0))) {
4172 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
4173 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
4177 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
4179 struct skb_shared_info *pinfo = skb_shinfo(skb);
4181 BUG_ON(skb->end - skb->tail < grow);
4183 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
4185 skb->data_len -= grow;
4188 pinfo->frags[0].page_offset += grow;
4189 skb_frag_size_sub(&pinfo->frags[0], grow);
4191 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
4192 skb_frag_unref(skb, 0);
4193 memmove(pinfo->frags, pinfo->frags + 1,
4194 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
4198 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4200 struct sk_buff **pp = NULL;
4201 struct packet_offload *ptype;
4202 __be16 type = skb->protocol;
4203 struct list_head *head = &offload_base;
4205 enum gro_result ret;
4208 if (!(skb->dev->features & NETIF_F_GRO))
4211 if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad)
4214 gro_list_prepare(napi, skb);
4217 list_for_each_entry_rcu(ptype, head, list) {
4218 if (ptype->type != type || !ptype->callbacks.gro_receive)
4221 skb_set_network_header(skb, skb_gro_offset(skb));
4222 skb_reset_mac_len(skb);
4223 NAPI_GRO_CB(skb)->same_flow = 0;
4224 NAPI_GRO_CB(skb)->flush = 0;
4225 NAPI_GRO_CB(skb)->free = 0;
4226 NAPI_GRO_CB(skb)->udp_mark = 0;
4227 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
4229 /* Setup for GRO checksum validation */
4230 switch (skb->ip_summed) {
4231 case CHECKSUM_COMPLETE:
4232 NAPI_GRO_CB(skb)->csum = skb->csum;
4233 NAPI_GRO_CB(skb)->csum_valid = 1;
4234 NAPI_GRO_CB(skb)->csum_cnt = 0;
4236 case CHECKSUM_UNNECESSARY:
4237 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4238 NAPI_GRO_CB(skb)->csum_valid = 0;
4241 NAPI_GRO_CB(skb)->csum_cnt = 0;
4242 NAPI_GRO_CB(skb)->csum_valid = 0;
4245 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4250 if (&ptype->list == head)
4253 same_flow = NAPI_GRO_CB(skb)->same_flow;
4254 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4257 struct sk_buff *nskb = *pp;
4261 napi_gro_complete(nskb);
4268 if (NAPI_GRO_CB(skb)->flush)
4271 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4272 struct sk_buff *nskb = napi->gro_list;
4274 /* locate the end of the list to select the 'oldest' flow */
4275 while (nskb->next) {
4281 napi_gro_complete(nskb);
4285 NAPI_GRO_CB(skb)->count = 1;
4286 NAPI_GRO_CB(skb)->age = jiffies;
4287 NAPI_GRO_CB(skb)->last = skb;
4288 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4289 skb->next = napi->gro_list;
4290 napi->gro_list = skb;
4294 grow = skb_gro_offset(skb) - skb_headlen(skb);
4296 gro_pull_from_frag0(skb, grow);
4305 struct packet_offload *gro_find_receive_by_type(__be16 type)
4307 struct list_head *offload_head = &offload_base;
4308 struct packet_offload *ptype;
4310 list_for_each_entry_rcu(ptype, offload_head, list) {
4311 if (ptype->type != type || !ptype->callbacks.gro_receive)
4317 EXPORT_SYMBOL(gro_find_receive_by_type);
4319 struct packet_offload *gro_find_complete_by_type(__be16 type)
4321 struct list_head *offload_head = &offload_base;
4322 struct packet_offload *ptype;
4324 list_for_each_entry_rcu(ptype, offload_head, list) {
4325 if (ptype->type != type || !ptype->callbacks.gro_complete)
4331 EXPORT_SYMBOL(gro_find_complete_by_type);
4333 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4337 if (netif_receive_skb_internal(skb))
4345 case GRO_MERGED_FREE:
4346 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) {
4348 kmem_cache_free(skbuff_head_cache, skb);
4362 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4364 trace_napi_gro_receive_entry(skb);
4366 skb_gro_reset_offset(skb);
4368 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4370 EXPORT_SYMBOL(napi_gro_receive);
4372 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4374 if (unlikely(skb->pfmemalloc)) {
4378 __skb_pull(skb, skb_headlen(skb));
4379 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4380 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4382 skb->dev = napi->dev;
4384 skb->encapsulation = 0;
4385 skb_shinfo(skb)->gso_type = 0;
4386 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4391 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4393 struct sk_buff *skb = napi->skb;
4396 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
4401 EXPORT_SYMBOL(napi_get_frags);
4403 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4404 struct sk_buff *skb,
4410 __skb_push(skb, ETH_HLEN);
4411 skb->protocol = eth_type_trans(skb, skb->dev);
4412 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4417 case GRO_MERGED_FREE:
4418 napi_reuse_skb(napi, skb);
4428 /* Upper GRO stack assumes network header starts at gro_offset=0
4429 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4430 * We copy ethernet header into skb->data to have a common layout.
4432 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4434 struct sk_buff *skb = napi->skb;
4435 const struct ethhdr *eth;
4436 unsigned int hlen = sizeof(*eth);
4440 skb_reset_mac_header(skb);
4441 skb_gro_reset_offset(skb);
4443 eth = skb_gro_header_fast(skb, 0);
4444 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4445 eth = skb_gro_header_slow(skb, hlen, 0);
4446 if (unlikely(!eth)) {
4447 napi_reuse_skb(napi, skb);
4451 gro_pull_from_frag0(skb, hlen);
4452 NAPI_GRO_CB(skb)->frag0 += hlen;
4453 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4455 __skb_pull(skb, hlen);
4458 * This works because the only protocols we care about don't require
4460 * We'll fix it up properly in napi_frags_finish()
4462 skb->protocol = eth->h_proto;
4467 gro_result_t napi_gro_frags(struct napi_struct *napi)
4469 struct sk_buff *skb = napi_frags_skb(napi);
4474 trace_napi_gro_frags_entry(skb);
4476 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4478 EXPORT_SYMBOL(napi_gro_frags);
4480 /* Compute the checksum from gro_offset and return the folded value
4481 * after adding in any pseudo checksum.
4483 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
4488 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
4490 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4491 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
4493 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
4494 !skb->csum_complete_sw)
4495 netdev_rx_csum_fault(skb->dev);
4498 NAPI_GRO_CB(skb)->csum = wsum;
4499 NAPI_GRO_CB(skb)->csum_valid = 1;
4503 EXPORT_SYMBOL(__skb_gro_checksum_complete);
4506 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4507 * Note: called with local irq disabled, but exits with local irq enabled.
4509 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4512 struct softnet_data *remsd = sd->rps_ipi_list;
4515 sd->rps_ipi_list = NULL;
4519 /* Send pending IPI's to kick RPS processing on remote cpus. */
4521 struct softnet_data *next = remsd->rps_ipi_next;
4523 if (cpu_online(remsd->cpu))
4524 smp_call_function_single_async(remsd->cpu,
4533 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
4536 return sd->rps_ipi_list != NULL;
4542 static int process_backlog(struct napi_struct *napi, int quota)
4545 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4547 /* Check if we have pending ipi, its better to send them now,
4548 * not waiting net_rx_action() end.
4550 if (sd_has_rps_ipi_waiting(sd)) {
4551 local_irq_disable();
4552 net_rps_action_and_irq_enable(sd);
4555 napi->weight = weight_p;
4556 local_irq_disable();
4558 struct sk_buff *skb;
4560 while ((skb = __skb_dequeue(&sd->process_queue))) {
4563 __netif_receive_skb(skb);
4565 local_irq_disable();
4566 input_queue_head_incr(sd);
4567 if (++work >= quota) {
4574 if (skb_queue_empty(&sd->input_pkt_queue)) {
4576 * Inline a custom version of __napi_complete().
4577 * only current cpu owns and manipulates this napi,
4578 * and NAPI_STATE_SCHED is the only possible flag set
4580 * We can use a plain write instead of clear_bit(),
4581 * and we dont need an smp_mb() memory barrier.
4589 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4590 &sd->process_queue);
4599 * __napi_schedule - schedule for receive
4600 * @n: entry to schedule
4602 * The entry's receive function will be scheduled to run.
4603 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
4605 void __napi_schedule(struct napi_struct *n)
4607 unsigned long flags;
4609 local_irq_save(flags);
4610 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4611 local_irq_restore(flags);
4613 EXPORT_SYMBOL(__napi_schedule);
4616 * __napi_schedule_irqoff - schedule for receive
4617 * @n: entry to schedule
4619 * Variant of __napi_schedule() assuming hard irqs are masked
4621 void __napi_schedule_irqoff(struct napi_struct *n)
4623 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4625 EXPORT_SYMBOL(__napi_schedule_irqoff);
4627 void __napi_complete(struct napi_struct *n)
4629 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4631 list_del_init(&n->poll_list);
4632 smp_mb__before_atomic();
4633 clear_bit(NAPI_STATE_SCHED, &n->state);
4635 EXPORT_SYMBOL(__napi_complete);
4637 void napi_complete_done(struct napi_struct *n, int work_done)
4639 unsigned long flags;
4642 * don't let napi dequeue from the cpu poll list
4643 * just in case its running on a different cpu
4645 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4649 unsigned long timeout = 0;
4652 timeout = n->dev->gro_flush_timeout;
4655 hrtimer_start(&n->timer, ns_to_ktime(timeout),
4656 HRTIMER_MODE_REL_PINNED);
4658 napi_gro_flush(n, false);
4660 if (likely(list_empty(&n->poll_list))) {
4661 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state));
4663 /* If n->poll_list is not empty, we need to mask irqs */
4664 local_irq_save(flags);
4666 local_irq_restore(flags);
4669 EXPORT_SYMBOL(napi_complete_done);
4671 /* must be called under rcu_read_lock(), as we dont take a reference */
4672 struct napi_struct *napi_by_id(unsigned int napi_id)
4674 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4675 struct napi_struct *napi;
4677 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4678 if (napi->napi_id == napi_id)
4683 EXPORT_SYMBOL_GPL(napi_by_id);
4685 void napi_hash_add(struct napi_struct *napi)
4687 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4689 spin_lock(&napi_hash_lock);
4691 /* 0 is not a valid id, we also skip an id that is taken
4692 * we expect both events to be extremely rare
4695 while (!napi->napi_id) {
4696 napi->napi_id = ++napi_gen_id;
4697 if (napi_by_id(napi->napi_id))
4701 hlist_add_head_rcu(&napi->napi_hash_node,
4702 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4704 spin_unlock(&napi_hash_lock);
4707 EXPORT_SYMBOL_GPL(napi_hash_add);
4709 /* Warning : caller is responsible to make sure rcu grace period
4710 * is respected before freeing memory containing @napi
4712 void napi_hash_del(struct napi_struct *napi)
4714 spin_lock(&napi_hash_lock);
4716 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4717 hlist_del_rcu(&napi->napi_hash_node);
4719 spin_unlock(&napi_hash_lock);
4721 EXPORT_SYMBOL_GPL(napi_hash_del);
4723 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
4725 struct napi_struct *napi;
4727 napi = container_of(timer, struct napi_struct, timer);
4729 napi_schedule(napi);
4731 return HRTIMER_NORESTART;
4734 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4735 int (*poll)(struct napi_struct *, int), int weight)
4737 INIT_LIST_HEAD(&napi->poll_list);
4738 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
4739 napi->timer.function = napi_watchdog;
4740 napi->gro_count = 0;
4741 napi->gro_list = NULL;
4744 if (weight > NAPI_POLL_WEIGHT)
4745 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4747 napi->weight = weight;
4748 list_add(&napi->dev_list, &dev->napi_list);
4750 #ifdef CONFIG_NETPOLL
4751 spin_lock_init(&napi->poll_lock);
4752 napi->poll_owner = -1;
4754 set_bit(NAPI_STATE_SCHED, &napi->state);
4756 EXPORT_SYMBOL(netif_napi_add);
4758 void napi_disable(struct napi_struct *n)
4761 set_bit(NAPI_STATE_DISABLE, &n->state);
4763 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
4765 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
4768 hrtimer_cancel(&n->timer);
4770 clear_bit(NAPI_STATE_DISABLE, &n->state);
4772 EXPORT_SYMBOL(napi_disable);
4774 void netif_napi_del(struct napi_struct *napi)
4776 list_del_init(&napi->dev_list);
4777 napi_free_frags(napi);
4779 kfree_skb_list(napi->gro_list);
4780 napi->gro_list = NULL;
4781 napi->gro_count = 0;
4783 EXPORT_SYMBOL(netif_napi_del);
4785 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
4790 list_del_init(&n->poll_list);
4792 have = netpoll_poll_lock(n);
4796 /* This NAPI_STATE_SCHED test is for avoiding a race
4797 * with netpoll's poll_napi(). Only the entity which
4798 * obtains the lock and sees NAPI_STATE_SCHED set will
4799 * actually make the ->poll() call. Therefore we avoid
4800 * accidentally calling ->poll() when NAPI is not scheduled.
4803 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4804 work = n->poll(n, weight);
4808 WARN_ON_ONCE(work > weight);
4810 if (likely(work < weight))
4813 /* Drivers must not modify the NAPI state if they
4814 * consume the entire weight. In such cases this code
4815 * still "owns" the NAPI instance and therefore can
4816 * move the instance around on the list at-will.
4818 if (unlikely(napi_disable_pending(n))) {
4824 /* flush too old packets
4825 * If HZ < 1000, flush all packets.
4827 napi_gro_flush(n, HZ >= 1000);
4830 /* Some drivers may have called napi_schedule
4831 * prior to exhausting their budget.
4833 if (unlikely(!list_empty(&n->poll_list))) {
4834 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
4835 n->dev ? n->dev->name : "backlog");
4839 list_add_tail(&n->poll_list, repoll);
4842 netpoll_poll_unlock(have);
4847 static void net_rx_action(struct softirq_action *h)
4849 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4850 unsigned long time_limit = jiffies + 2;
4851 int budget = netdev_budget;
4855 local_irq_disable();
4856 list_splice_init(&sd->poll_list, &list);
4860 struct napi_struct *n;
4862 if (list_empty(&list)) {
4863 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
4868 n = list_first_entry(&list, struct napi_struct, poll_list);
4869 budget -= napi_poll(n, &repoll);
4871 /* If softirq window is exhausted then punt.
4872 * Allow this to run for 2 jiffies since which will allow
4873 * an average latency of 1.5/HZ.
4875 if (unlikely(budget <= 0 ||
4876 time_after_eq(jiffies, time_limit))) {
4882 local_irq_disable();
4884 list_splice_tail_init(&sd->poll_list, &list);
4885 list_splice_tail(&repoll, &list);
4886 list_splice(&list, &sd->poll_list);
4887 if (!list_empty(&sd->poll_list))
4888 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4890 net_rps_action_and_irq_enable(sd);
4893 struct netdev_adjacent {
4894 struct net_device *dev;
4896 /* upper master flag, there can only be one master device per list */
4899 /* counter for the number of times this device was added to us */
4902 /* private field for the users */
4905 struct list_head list;
4906 struct rcu_head rcu;
4909 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
4910 struct list_head *adj_list)
4912 struct netdev_adjacent *adj;
4914 list_for_each_entry(adj, adj_list, list) {
4915 if (adj->dev == adj_dev)
4922 * netdev_has_upper_dev - Check if device is linked to an upper device
4924 * @upper_dev: upper device to check
4926 * Find out if a device is linked to specified upper device and return true
4927 * in case it is. Note that this checks only immediate upper device,
4928 * not through a complete stack of devices. The caller must hold the RTNL lock.
4930 bool netdev_has_upper_dev(struct net_device *dev,
4931 struct net_device *upper_dev)
4935 return __netdev_find_adj(upper_dev, &dev->all_adj_list.upper);
4937 EXPORT_SYMBOL(netdev_has_upper_dev);
4940 * netdev_has_any_upper_dev - Check if device is linked to some device
4943 * Find out if a device is linked to an upper device and return true in case
4944 * it is. The caller must hold the RTNL lock.
4946 static bool netdev_has_any_upper_dev(struct net_device *dev)
4950 return !list_empty(&dev->all_adj_list.upper);
4954 * netdev_master_upper_dev_get - Get master upper device
4957 * Find a master upper device and return pointer to it or NULL in case
4958 * it's not there. The caller must hold the RTNL lock.
4960 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4962 struct netdev_adjacent *upper;
4966 if (list_empty(&dev->adj_list.upper))
4969 upper = list_first_entry(&dev->adj_list.upper,
4970 struct netdev_adjacent, list);
4971 if (likely(upper->master))
4975 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4977 void *netdev_adjacent_get_private(struct list_head *adj_list)
4979 struct netdev_adjacent *adj;
4981 adj = list_entry(adj_list, struct netdev_adjacent, list);
4983 return adj->private;
4985 EXPORT_SYMBOL(netdev_adjacent_get_private);
4988 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
4990 * @iter: list_head ** of the current position
4992 * Gets the next device from the dev's upper list, starting from iter
4993 * position. The caller must hold RCU read lock.
4995 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
4996 struct list_head **iter)
4998 struct netdev_adjacent *upper;
5000 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5002 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5004 if (&upper->list == &dev->adj_list.upper)
5007 *iter = &upper->list;
5011 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
5014 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
5016 * @iter: list_head ** of the current position
5018 * Gets the next device from the dev's upper list, starting from iter
5019 * position. The caller must hold RCU read lock.
5021 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
5022 struct list_head **iter)
5024 struct netdev_adjacent *upper;
5026 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5028 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5030 if (&upper->list == &dev->all_adj_list.upper)
5033 *iter = &upper->list;
5037 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
5040 * netdev_lower_get_next_private - Get the next ->private from the
5041 * lower neighbour list
5043 * @iter: list_head ** of the current position
5045 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5046 * list, starting from iter position. The caller must hold either hold the
5047 * RTNL lock or its own locking that guarantees that the neighbour lower
5048 * list will remain unchanged.
5050 void *netdev_lower_get_next_private(struct net_device *dev,
5051 struct list_head **iter)
5053 struct netdev_adjacent *lower;
5055 lower = list_entry(*iter, struct netdev_adjacent, list);
5057 if (&lower->list == &dev->adj_list.lower)
5060 *iter = lower->list.next;
5062 return lower->private;
5064 EXPORT_SYMBOL(netdev_lower_get_next_private);
5067 * netdev_lower_get_next_private_rcu - Get the next ->private from the
5068 * lower neighbour list, RCU
5071 * @iter: list_head ** of the current position
5073 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5074 * list, starting from iter position. The caller must hold RCU read lock.
5076 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
5077 struct list_head **iter)
5079 struct netdev_adjacent *lower;
5081 WARN_ON_ONCE(!rcu_read_lock_held());
5083 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5085 if (&lower->list == &dev->adj_list.lower)
5088 *iter = &lower->list;
5090 return lower->private;
5092 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
5095 * netdev_lower_get_next - Get the next device from the lower neighbour
5098 * @iter: list_head ** of the current position
5100 * Gets the next netdev_adjacent from the dev's lower neighbour
5101 * list, starting from iter position. The caller must hold RTNL lock or
5102 * its own locking that guarantees that the neighbour lower
5103 * list will remain unchanged.
5105 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
5107 struct netdev_adjacent *lower;
5109 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
5111 if (&lower->list == &dev->adj_list.lower)
5114 *iter = &lower->list;
5118 EXPORT_SYMBOL(netdev_lower_get_next);
5121 * netdev_lower_get_first_private_rcu - Get the first ->private from the
5122 * lower neighbour list, RCU
5126 * Gets the first netdev_adjacent->private from the dev's lower neighbour
5127 * list. The caller must hold RCU read lock.
5129 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
5131 struct netdev_adjacent *lower;
5133 lower = list_first_or_null_rcu(&dev->adj_list.lower,
5134 struct netdev_adjacent, list);
5136 return lower->private;
5139 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
5142 * netdev_master_upper_dev_get_rcu - Get master upper device
5145 * Find a master upper device and return pointer to it or NULL in case
5146 * it's not there. The caller must hold the RCU read lock.
5148 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
5150 struct netdev_adjacent *upper;
5152 upper = list_first_or_null_rcu(&dev->adj_list.upper,
5153 struct netdev_adjacent, list);
5154 if (upper && likely(upper->master))
5158 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
5160 static int netdev_adjacent_sysfs_add(struct net_device *dev,
5161 struct net_device *adj_dev,
5162 struct list_head *dev_list)
5164 char linkname[IFNAMSIZ+7];
5165 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5166 "upper_%s" : "lower_%s", adj_dev->name);
5167 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
5170 static void netdev_adjacent_sysfs_del(struct net_device *dev,
5172 struct list_head *dev_list)
5174 char linkname[IFNAMSIZ+7];
5175 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5176 "upper_%s" : "lower_%s", name);
5177 sysfs_remove_link(&(dev->dev.kobj), linkname);
5180 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
5181 struct net_device *adj_dev,
5182 struct list_head *dev_list)
5184 return (dev_list == &dev->adj_list.upper ||
5185 dev_list == &dev->adj_list.lower) &&
5186 net_eq(dev_net(dev), dev_net(adj_dev));
5189 static int __netdev_adjacent_dev_insert(struct net_device *dev,
5190 struct net_device *adj_dev,
5191 struct list_head *dev_list,
5192 void *private, bool master)
5194 struct netdev_adjacent *adj;
5197 adj = __netdev_find_adj(adj_dev, dev_list);
5204 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
5209 adj->master = master;
5211 adj->private = private;
5214 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
5215 adj_dev->name, dev->name, adj_dev->name);
5217 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
5218 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
5223 /* Ensure that master link is always the first item in list. */
5225 ret = sysfs_create_link(&(dev->dev.kobj),
5226 &(adj_dev->dev.kobj), "master");
5228 goto remove_symlinks;
5230 list_add_rcu(&adj->list, dev_list);
5232 list_add_tail_rcu(&adj->list, dev_list);
5238 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5239 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5247 static void __netdev_adjacent_dev_remove(struct net_device *dev,
5248 struct net_device *adj_dev,
5249 struct list_head *dev_list)
5251 struct netdev_adjacent *adj;
5253 adj = __netdev_find_adj(adj_dev, dev_list);
5256 pr_err("tried to remove device %s from %s\n",
5257 dev->name, adj_dev->name);
5261 if (adj->ref_nr > 1) {
5262 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
5269 sysfs_remove_link(&(dev->dev.kobj), "master");
5271 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5272 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5274 list_del_rcu(&adj->list);
5275 pr_debug("dev_put for %s, because link removed from %s to %s\n",
5276 adj_dev->name, dev->name, adj_dev->name);
5278 kfree_rcu(adj, rcu);
5281 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
5282 struct net_device *upper_dev,
5283 struct list_head *up_list,
5284 struct list_head *down_list,
5285 void *private, bool master)
5289 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
5294 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
5297 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5304 static int __netdev_adjacent_dev_link(struct net_device *dev,
5305 struct net_device *upper_dev)
5307 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
5308 &dev->all_adj_list.upper,
5309 &upper_dev->all_adj_list.lower,
5313 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
5314 struct net_device *upper_dev,
5315 struct list_head *up_list,
5316 struct list_head *down_list)
5318 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5319 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
5322 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
5323 struct net_device *upper_dev)
5325 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5326 &dev->all_adj_list.upper,
5327 &upper_dev->all_adj_list.lower);
5330 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
5331 struct net_device *upper_dev,
5332 void *private, bool master)
5334 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
5339 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
5340 &dev->adj_list.upper,
5341 &upper_dev->adj_list.lower,
5344 __netdev_adjacent_dev_unlink(dev, upper_dev);
5351 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
5352 struct net_device *upper_dev)
5354 __netdev_adjacent_dev_unlink(dev, upper_dev);
5355 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5356 &dev->adj_list.upper,
5357 &upper_dev->adj_list.lower);
5360 static int __netdev_upper_dev_link(struct net_device *dev,
5361 struct net_device *upper_dev, bool master,
5364 struct netdev_notifier_changeupper_info changeupper_info;
5365 struct netdev_adjacent *i, *j, *to_i, *to_j;
5370 if (dev == upper_dev)
5373 /* To prevent loops, check if dev is not upper device to upper_dev. */
5374 if (__netdev_find_adj(dev, &upper_dev->all_adj_list.upper))
5377 if (__netdev_find_adj(upper_dev, &dev->adj_list.upper))
5380 if (master && netdev_master_upper_dev_get(dev))
5383 changeupper_info.upper_dev = upper_dev;
5384 changeupper_info.master = master;
5385 changeupper_info.linking = true;
5387 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
5388 &changeupper_info.info);
5389 ret = notifier_to_errno(ret);
5393 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
5398 /* Now that we linked these devs, make all the upper_dev's
5399 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5400 * versa, and don't forget the devices itself. All of these
5401 * links are non-neighbours.
5403 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5404 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5405 pr_debug("Interlinking %s with %s, non-neighbour\n",
5406 i->dev->name, j->dev->name);
5407 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
5413 /* add dev to every upper_dev's upper device */
5414 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5415 pr_debug("linking %s's upper device %s with %s\n",
5416 upper_dev->name, i->dev->name, dev->name);
5417 ret = __netdev_adjacent_dev_link(dev, i->dev);
5419 goto rollback_upper_mesh;
5422 /* add upper_dev to every dev's lower device */
5423 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5424 pr_debug("linking %s's lower device %s with %s\n", dev->name,
5425 i->dev->name, upper_dev->name);
5426 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
5428 goto rollback_lower_mesh;
5431 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
5432 &changeupper_info.info);
5435 rollback_lower_mesh:
5437 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5440 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5445 rollback_upper_mesh:
5447 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5450 __netdev_adjacent_dev_unlink(dev, i->dev);
5458 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5459 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5460 if (i == to_i && j == to_j)
5462 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5468 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5474 * netdev_upper_dev_link - Add a link to the upper device
5476 * @upper_dev: new upper device
5478 * Adds a link to device which is upper to this one. The caller must hold
5479 * the RTNL lock. On a failure a negative errno code is returned.
5480 * On success the reference counts are adjusted and the function
5483 int netdev_upper_dev_link(struct net_device *dev,
5484 struct net_device *upper_dev)
5486 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
5488 EXPORT_SYMBOL(netdev_upper_dev_link);
5491 * netdev_master_upper_dev_link - Add a master link to the upper device
5493 * @upper_dev: new upper device
5495 * Adds a link to device which is upper to this one. In this case, only
5496 * one master upper device can be linked, although other non-master devices
5497 * might be linked as well. The caller must hold the RTNL lock.
5498 * On a failure a negative errno code is returned. On success the reference
5499 * counts are adjusted and the function returns zero.
5501 int netdev_master_upper_dev_link(struct net_device *dev,
5502 struct net_device *upper_dev)
5504 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
5506 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5508 int netdev_master_upper_dev_link_private(struct net_device *dev,
5509 struct net_device *upper_dev,
5512 return __netdev_upper_dev_link(dev, upper_dev, true, private);
5514 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
5517 * netdev_upper_dev_unlink - Removes a link to upper device
5519 * @upper_dev: new upper device
5521 * Removes a link to device which is upper to this one. The caller must hold
5524 void netdev_upper_dev_unlink(struct net_device *dev,
5525 struct net_device *upper_dev)
5527 struct netdev_notifier_changeupper_info changeupper_info;
5528 struct netdev_adjacent *i, *j;
5531 changeupper_info.upper_dev = upper_dev;
5532 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
5533 changeupper_info.linking = false;
5535 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
5536 &changeupper_info.info);
5538 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5540 /* Here is the tricky part. We must remove all dev's lower
5541 * devices from all upper_dev's upper devices and vice
5542 * versa, to maintain the graph relationship.
5544 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5545 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
5546 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5548 /* remove also the devices itself from lower/upper device
5551 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5552 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5554 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5555 __netdev_adjacent_dev_unlink(dev, i->dev);
5557 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
5558 &changeupper_info.info);
5560 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5563 * netdev_bonding_info_change - Dispatch event about slave change
5565 * @bonding_info: info to dispatch
5567 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
5568 * The caller must hold the RTNL lock.
5570 void netdev_bonding_info_change(struct net_device *dev,
5571 struct netdev_bonding_info *bonding_info)
5573 struct netdev_notifier_bonding_info info;
5575 memcpy(&info.bonding_info, bonding_info,
5576 sizeof(struct netdev_bonding_info));
5577 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev,
5580 EXPORT_SYMBOL(netdev_bonding_info_change);
5582 static void netdev_adjacent_add_links(struct net_device *dev)
5584 struct netdev_adjacent *iter;
5586 struct net *net = dev_net(dev);
5588 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5589 if (!net_eq(net,dev_net(iter->dev)))
5591 netdev_adjacent_sysfs_add(iter->dev, dev,
5592 &iter->dev->adj_list.lower);
5593 netdev_adjacent_sysfs_add(dev, iter->dev,
5594 &dev->adj_list.upper);
5597 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5598 if (!net_eq(net,dev_net(iter->dev)))
5600 netdev_adjacent_sysfs_add(iter->dev, dev,
5601 &iter->dev->adj_list.upper);
5602 netdev_adjacent_sysfs_add(dev, iter->dev,
5603 &dev->adj_list.lower);
5607 static void netdev_adjacent_del_links(struct net_device *dev)
5609 struct netdev_adjacent *iter;
5611 struct net *net = dev_net(dev);
5613 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5614 if (!net_eq(net,dev_net(iter->dev)))
5616 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5617 &iter->dev->adj_list.lower);
5618 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5619 &dev->adj_list.upper);
5622 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5623 if (!net_eq(net,dev_net(iter->dev)))
5625 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5626 &iter->dev->adj_list.upper);
5627 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5628 &dev->adj_list.lower);
5632 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5634 struct netdev_adjacent *iter;
5636 struct net *net = dev_net(dev);
5638 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5639 if (!net_eq(net,dev_net(iter->dev)))
5641 netdev_adjacent_sysfs_del(iter->dev, oldname,
5642 &iter->dev->adj_list.lower);
5643 netdev_adjacent_sysfs_add(iter->dev, dev,
5644 &iter->dev->adj_list.lower);
5647 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5648 if (!net_eq(net,dev_net(iter->dev)))
5650 netdev_adjacent_sysfs_del(iter->dev, oldname,
5651 &iter->dev->adj_list.upper);
5652 netdev_adjacent_sysfs_add(iter->dev, dev,
5653 &iter->dev->adj_list.upper);
5657 void *netdev_lower_dev_get_private(struct net_device *dev,
5658 struct net_device *lower_dev)
5660 struct netdev_adjacent *lower;
5664 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
5668 return lower->private;
5670 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5673 int dev_get_nest_level(struct net_device *dev,
5674 bool (*type_check)(struct net_device *dev))
5676 struct net_device *lower = NULL;
5677 struct list_head *iter;
5683 netdev_for_each_lower_dev(dev, lower, iter) {
5684 nest = dev_get_nest_level(lower, type_check);
5685 if (max_nest < nest)
5689 if (type_check(dev))
5694 EXPORT_SYMBOL(dev_get_nest_level);
5696 static void dev_change_rx_flags(struct net_device *dev, int flags)
5698 const struct net_device_ops *ops = dev->netdev_ops;
5700 if (ops->ndo_change_rx_flags)
5701 ops->ndo_change_rx_flags(dev, flags);
5704 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5706 unsigned int old_flags = dev->flags;
5712 dev->flags |= IFF_PROMISC;
5713 dev->promiscuity += inc;
5714 if (dev->promiscuity == 0) {
5717 * If inc causes overflow, untouch promisc and return error.
5720 dev->flags &= ~IFF_PROMISC;
5722 dev->promiscuity -= inc;
5723 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5728 if (dev->flags != old_flags) {
5729 pr_info("device %s %s promiscuous mode\n",
5731 dev->flags & IFF_PROMISC ? "entered" : "left");
5732 if (audit_enabled) {
5733 current_uid_gid(&uid, &gid);
5734 audit_log(current->audit_context, GFP_ATOMIC,
5735 AUDIT_ANOM_PROMISCUOUS,
5736 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5737 dev->name, (dev->flags & IFF_PROMISC),
5738 (old_flags & IFF_PROMISC),
5739 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5740 from_kuid(&init_user_ns, uid),
5741 from_kgid(&init_user_ns, gid),
5742 audit_get_sessionid(current));
5745 dev_change_rx_flags(dev, IFF_PROMISC);
5748 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5753 * dev_set_promiscuity - update promiscuity count on a device
5757 * Add or remove promiscuity from a device. While the count in the device
5758 * remains above zero the interface remains promiscuous. Once it hits zero
5759 * the device reverts back to normal filtering operation. A negative inc
5760 * value is used to drop promiscuity on the device.
5761 * Return 0 if successful or a negative errno code on error.
5763 int dev_set_promiscuity(struct net_device *dev, int inc)
5765 unsigned int old_flags = dev->flags;
5768 err = __dev_set_promiscuity(dev, inc, true);
5771 if (dev->flags != old_flags)
5772 dev_set_rx_mode(dev);
5775 EXPORT_SYMBOL(dev_set_promiscuity);
5777 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5779 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5783 dev->flags |= IFF_ALLMULTI;
5784 dev->allmulti += inc;
5785 if (dev->allmulti == 0) {
5788 * If inc causes overflow, untouch allmulti and return error.
5791 dev->flags &= ~IFF_ALLMULTI;
5793 dev->allmulti -= inc;
5794 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5799 if (dev->flags ^ old_flags) {
5800 dev_change_rx_flags(dev, IFF_ALLMULTI);
5801 dev_set_rx_mode(dev);
5803 __dev_notify_flags(dev, old_flags,
5804 dev->gflags ^ old_gflags);
5810 * dev_set_allmulti - update allmulti count on a device
5814 * Add or remove reception of all multicast frames to a device. While the
5815 * count in the device remains above zero the interface remains listening
5816 * to all interfaces. Once it hits zero the device reverts back to normal
5817 * filtering operation. A negative @inc value is used to drop the counter
5818 * when releasing a resource needing all multicasts.
5819 * Return 0 if successful or a negative errno code on error.
5822 int dev_set_allmulti(struct net_device *dev, int inc)
5824 return __dev_set_allmulti(dev, inc, true);
5826 EXPORT_SYMBOL(dev_set_allmulti);
5829 * Upload unicast and multicast address lists to device and
5830 * configure RX filtering. When the device doesn't support unicast
5831 * filtering it is put in promiscuous mode while unicast addresses
5834 void __dev_set_rx_mode(struct net_device *dev)
5836 const struct net_device_ops *ops = dev->netdev_ops;
5838 /* dev_open will call this function so the list will stay sane. */
5839 if (!(dev->flags&IFF_UP))
5842 if (!netif_device_present(dev))
5845 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5846 /* Unicast addresses changes may only happen under the rtnl,
5847 * therefore calling __dev_set_promiscuity here is safe.
5849 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5850 __dev_set_promiscuity(dev, 1, false);
5851 dev->uc_promisc = true;
5852 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5853 __dev_set_promiscuity(dev, -1, false);
5854 dev->uc_promisc = false;
5858 if (ops->ndo_set_rx_mode)
5859 ops->ndo_set_rx_mode(dev);
5862 void dev_set_rx_mode(struct net_device *dev)
5864 netif_addr_lock_bh(dev);
5865 __dev_set_rx_mode(dev);
5866 netif_addr_unlock_bh(dev);
5870 * dev_get_flags - get flags reported to userspace
5873 * Get the combination of flag bits exported through APIs to userspace.
5875 unsigned int dev_get_flags(const struct net_device *dev)
5879 flags = (dev->flags & ~(IFF_PROMISC |
5884 (dev->gflags & (IFF_PROMISC |
5887 if (netif_running(dev)) {
5888 if (netif_oper_up(dev))
5889 flags |= IFF_RUNNING;
5890 if (netif_carrier_ok(dev))
5891 flags |= IFF_LOWER_UP;
5892 if (netif_dormant(dev))
5893 flags |= IFF_DORMANT;
5898 EXPORT_SYMBOL(dev_get_flags);
5900 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5902 unsigned int old_flags = dev->flags;
5908 * Set the flags on our device.
5911 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5912 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5914 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5918 * Load in the correct multicast list now the flags have changed.
5921 if ((old_flags ^ flags) & IFF_MULTICAST)
5922 dev_change_rx_flags(dev, IFF_MULTICAST);
5924 dev_set_rx_mode(dev);
5927 * Have we downed the interface. We handle IFF_UP ourselves
5928 * according to user attempts to set it, rather than blindly
5933 if ((old_flags ^ flags) & IFF_UP)
5934 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5936 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5937 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5938 unsigned int old_flags = dev->flags;
5940 dev->gflags ^= IFF_PROMISC;
5942 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5943 if (dev->flags != old_flags)
5944 dev_set_rx_mode(dev);
5947 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5948 is important. Some (broken) drivers set IFF_PROMISC, when
5949 IFF_ALLMULTI is requested not asking us and not reporting.
5951 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5952 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5954 dev->gflags ^= IFF_ALLMULTI;
5955 __dev_set_allmulti(dev, inc, false);
5961 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5962 unsigned int gchanges)
5964 unsigned int changes = dev->flags ^ old_flags;
5967 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
5969 if (changes & IFF_UP) {
5970 if (dev->flags & IFF_UP)
5971 call_netdevice_notifiers(NETDEV_UP, dev);
5973 call_netdevice_notifiers(NETDEV_DOWN, dev);
5976 if (dev->flags & IFF_UP &&
5977 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5978 struct netdev_notifier_change_info change_info;
5980 change_info.flags_changed = changes;
5981 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
5987 * dev_change_flags - change device settings
5989 * @flags: device state flags
5991 * Change settings on device based state flags. The flags are
5992 * in the userspace exported format.
5994 int dev_change_flags(struct net_device *dev, unsigned int flags)
5997 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
5999 ret = __dev_change_flags(dev, flags);
6003 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
6004 __dev_notify_flags(dev, old_flags, changes);
6007 EXPORT_SYMBOL(dev_change_flags);
6009 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
6011 const struct net_device_ops *ops = dev->netdev_ops;
6013 if (ops->ndo_change_mtu)
6014 return ops->ndo_change_mtu(dev, new_mtu);
6021 * dev_set_mtu - Change maximum transfer unit
6023 * @new_mtu: new transfer unit
6025 * Change the maximum transfer size of the network device.
6027 int dev_set_mtu(struct net_device *dev, int new_mtu)
6031 if (new_mtu == dev->mtu)
6034 /* MTU must be positive. */
6038 if (!netif_device_present(dev))
6041 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
6042 err = notifier_to_errno(err);
6046 orig_mtu = dev->mtu;
6047 err = __dev_set_mtu(dev, new_mtu);
6050 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6051 err = notifier_to_errno(err);
6053 /* setting mtu back and notifying everyone again,
6054 * so that they have a chance to revert changes.
6056 __dev_set_mtu(dev, orig_mtu);
6057 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6062 EXPORT_SYMBOL(dev_set_mtu);
6065 * dev_set_group - Change group this device belongs to
6067 * @new_group: group this device should belong to
6069 void dev_set_group(struct net_device *dev, int new_group)
6071 dev->group = new_group;
6073 EXPORT_SYMBOL(dev_set_group);
6076 * dev_set_mac_address - Change Media Access Control Address
6080 * Change the hardware (MAC) address of the device
6082 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
6084 const struct net_device_ops *ops = dev->netdev_ops;
6087 if (!ops->ndo_set_mac_address)
6089 if (sa->sa_family != dev->type)
6091 if (!netif_device_present(dev))
6093 err = ops->ndo_set_mac_address(dev, sa);
6096 dev->addr_assign_type = NET_ADDR_SET;
6097 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
6098 add_device_randomness(dev->dev_addr, dev->addr_len);
6101 EXPORT_SYMBOL(dev_set_mac_address);
6104 * dev_change_carrier - Change device carrier
6106 * @new_carrier: new value
6108 * Change device carrier
6110 int dev_change_carrier(struct net_device *dev, bool new_carrier)
6112 const struct net_device_ops *ops = dev->netdev_ops;
6114 if (!ops->ndo_change_carrier)
6116 if (!netif_device_present(dev))
6118 return ops->ndo_change_carrier(dev, new_carrier);
6120 EXPORT_SYMBOL(dev_change_carrier);
6123 * dev_get_phys_port_id - Get device physical port ID
6127 * Get device physical port ID
6129 int dev_get_phys_port_id(struct net_device *dev,
6130 struct netdev_phys_item_id *ppid)
6132 const struct net_device_ops *ops = dev->netdev_ops;
6134 if (!ops->ndo_get_phys_port_id)
6136 return ops->ndo_get_phys_port_id(dev, ppid);
6138 EXPORT_SYMBOL(dev_get_phys_port_id);
6141 * dev_get_phys_port_name - Get device physical port name
6145 * Get device physical port name
6147 int dev_get_phys_port_name(struct net_device *dev,
6148 char *name, size_t len)
6150 const struct net_device_ops *ops = dev->netdev_ops;
6152 if (!ops->ndo_get_phys_port_name)
6154 return ops->ndo_get_phys_port_name(dev, name, len);
6156 EXPORT_SYMBOL(dev_get_phys_port_name);
6159 * dev_change_proto_down - update protocol port state information
6161 * @proto_down: new value
6163 * This info can be used by switch drivers to set the phys state of the
6166 int dev_change_proto_down(struct net_device *dev, bool proto_down)
6168 const struct net_device_ops *ops = dev->netdev_ops;
6170 if (!ops->ndo_change_proto_down)
6172 if (!netif_device_present(dev))
6174 return ops->ndo_change_proto_down(dev, proto_down);
6176 EXPORT_SYMBOL(dev_change_proto_down);
6179 * dev_new_index - allocate an ifindex
6180 * @net: the applicable net namespace
6182 * Returns a suitable unique value for a new device interface
6183 * number. The caller must hold the rtnl semaphore or the
6184 * dev_base_lock to be sure it remains unique.
6186 static int dev_new_index(struct net *net)
6188 int ifindex = net->ifindex;
6192 if (!__dev_get_by_index(net, ifindex))
6193 return net->ifindex = ifindex;
6197 /* Delayed registration/unregisteration */
6198 static LIST_HEAD(net_todo_list);
6199 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
6201 static void net_set_todo(struct net_device *dev)
6203 list_add_tail(&dev->todo_list, &net_todo_list);
6204 dev_net(dev)->dev_unreg_count++;
6207 static void rollback_registered_many(struct list_head *head)
6209 struct net_device *dev, *tmp;
6210 LIST_HEAD(close_head);
6212 BUG_ON(dev_boot_phase);
6215 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
6216 /* Some devices call without registering
6217 * for initialization unwind. Remove those
6218 * devices and proceed with the remaining.
6220 if (dev->reg_state == NETREG_UNINITIALIZED) {
6221 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
6225 list_del(&dev->unreg_list);
6228 dev->dismantle = true;
6229 BUG_ON(dev->reg_state != NETREG_REGISTERED);
6232 /* If device is running, close it first. */
6233 list_for_each_entry(dev, head, unreg_list)
6234 list_add_tail(&dev->close_list, &close_head);
6235 dev_close_many(&close_head, true);
6237 list_for_each_entry(dev, head, unreg_list) {
6238 /* And unlink it from device chain. */
6239 unlist_netdevice(dev);
6241 dev->reg_state = NETREG_UNREGISTERING;
6242 on_each_cpu(flush_backlog, dev, 1);
6247 list_for_each_entry(dev, head, unreg_list) {
6248 struct sk_buff *skb = NULL;
6250 /* Shutdown queueing discipline. */
6254 /* Notify protocols, that we are about to destroy
6255 this device. They should clean all the things.
6257 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6259 if (!dev->rtnl_link_ops ||
6260 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6261 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U,
6265 * Flush the unicast and multicast chains
6270 if (dev->netdev_ops->ndo_uninit)
6271 dev->netdev_ops->ndo_uninit(dev);
6274 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
6276 /* Notifier chain MUST detach us all upper devices. */
6277 WARN_ON(netdev_has_any_upper_dev(dev));
6279 /* Remove entries from kobject tree */
6280 netdev_unregister_kobject(dev);
6282 /* Remove XPS queueing entries */
6283 netif_reset_xps_queues_gt(dev, 0);
6289 list_for_each_entry(dev, head, unreg_list)
6293 static void rollback_registered(struct net_device *dev)
6297 list_add(&dev->unreg_list, &single);
6298 rollback_registered_many(&single);
6302 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
6303 struct net_device *upper, netdev_features_t features)
6305 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6306 netdev_features_t feature;
6309 for_each_netdev_feature(&upper_disables, feature_bit) {
6310 feature = __NETIF_F_BIT(feature_bit);
6311 if (!(upper->wanted_features & feature)
6312 && (features & feature)) {
6313 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
6314 &feature, upper->name);
6315 features &= ~feature;
6322 static void netdev_sync_lower_features(struct net_device *upper,
6323 struct net_device *lower, netdev_features_t features)
6325 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6326 netdev_features_t feature;
6329 for_each_netdev_feature(&upper_disables, feature_bit) {
6330 feature = __NETIF_F_BIT(feature_bit);
6331 if (!(features & feature) && (lower->features & feature)) {
6332 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
6333 &feature, lower->name);
6334 lower->wanted_features &= ~feature;
6335 netdev_update_features(lower);
6337 if (unlikely(lower->features & feature))
6338 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
6339 &feature, lower->name);
6344 static netdev_features_t netdev_fix_features(struct net_device *dev,
6345 netdev_features_t features)
6347 /* Fix illegal checksum combinations */
6348 if ((features & NETIF_F_HW_CSUM) &&
6349 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6350 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
6351 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
6354 /* TSO requires that SG is present as well. */
6355 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
6356 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
6357 features &= ~NETIF_F_ALL_TSO;
6360 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
6361 !(features & NETIF_F_IP_CSUM)) {
6362 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
6363 features &= ~NETIF_F_TSO;
6364 features &= ~NETIF_F_TSO_ECN;
6367 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
6368 !(features & NETIF_F_IPV6_CSUM)) {
6369 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
6370 features &= ~NETIF_F_TSO6;
6373 /* TSO ECN requires that TSO is present as well. */
6374 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
6375 features &= ~NETIF_F_TSO_ECN;
6377 /* Software GSO depends on SG. */
6378 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
6379 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
6380 features &= ~NETIF_F_GSO;
6383 /* UFO needs SG and checksumming */
6384 if (features & NETIF_F_UFO) {
6385 /* maybe split UFO into V4 and V6? */
6386 if (!((features & NETIF_F_GEN_CSUM) ||
6387 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
6388 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6390 "Dropping NETIF_F_UFO since no checksum offload features.\n");
6391 features &= ~NETIF_F_UFO;
6394 if (!(features & NETIF_F_SG)) {
6396 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
6397 features &= ~NETIF_F_UFO;
6401 #ifdef CONFIG_NET_RX_BUSY_POLL
6402 if (dev->netdev_ops->ndo_busy_poll)
6403 features |= NETIF_F_BUSY_POLL;
6406 features &= ~NETIF_F_BUSY_POLL;
6411 int __netdev_update_features(struct net_device *dev)
6413 struct net_device *upper, *lower;
6414 netdev_features_t features;
6415 struct list_head *iter;
6420 features = netdev_get_wanted_features(dev);
6422 if (dev->netdev_ops->ndo_fix_features)
6423 features = dev->netdev_ops->ndo_fix_features(dev, features);
6425 /* driver might be less strict about feature dependencies */
6426 features = netdev_fix_features(dev, features);
6428 /* some features can't be enabled if they're off an an upper device */
6429 netdev_for_each_upper_dev_rcu(dev, upper, iter)
6430 features = netdev_sync_upper_features(dev, upper, features);
6432 if (dev->features == features)
6435 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
6436 &dev->features, &features);
6438 if (dev->netdev_ops->ndo_set_features)
6439 err = dev->netdev_ops->ndo_set_features(dev, features);
6443 if (unlikely(err < 0)) {
6445 "set_features() failed (%d); wanted %pNF, left %pNF\n",
6446 err, &features, &dev->features);
6447 /* return non-0 since some features might have changed and
6448 * it's better to fire a spurious notification than miss it
6454 /* some features must be disabled on lower devices when disabled
6455 * on an upper device (think: bonding master or bridge)
6457 netdev_for_each_lower_dev(dev, lower, iter)
6458 netdev_sync_lower_features(dev, lower, features);
6461 dev->features = features;
6463 return err < 0 ? 0 : 1;
6467 * netdev_update_features - recalculate device features
6468 * @dev: the device to check
6470 * Recalculate dev->features set and send notifications if it
6471 * has changed. Should be called after driver or hardware dependent
6472 * conditions might have changed that influence the features.
6474 void netdev_update_features(struct net_device *dev)
6476 if (__netdev_update_features(dev))
6477 netdev_features_change(dev);
6479 EXPORT_SYMBOL(netdev_update_features);
6482 * netdev_change_features - recalculate device features
6483 * @dev: the device to check
6485 * Recalculate dev->features set and send notifications even
6486 * if they have not changed. Should be called instead of
6487 * netdev_update_features() if also dev->vlan_features might
6488 * have changed to allow the changes to be propagated to stacked
6491 void netdev_change_features(struct net_device *dev)
6493 __netdev_update_features(dev);
6494 netdev_features_change(dev);
6496 EXPORT_SYMBOL(netdev_change_features);
6499 * netif_stacked_transfer_operstate - transfer operstate
6500 * @rootdev: the root or lower level device to transfer state from
6501 * @dev: the device to transfer operstate to
6503 * Transfer operational state from root to device. This is normally
6504 * called when a stacking relationship exists between the root
6505 * device and the device(a leaf device).
6507 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
6508 struct net_device *dev)
6510 if (rootdev->operstate == IF_OPER_DORMANT)
6511 netif_dormant_on(dev);
6513 netif_dormant_off(dev);
6515 if (netif_carrier_ok(rootdev)) {
6516 if (!netif_carrier_ok(dev))
6517 netif_carrier_on(dev);
6519 if (netif_carrier_ok(dev))
6520 netif_carrier_off(dev);
6523 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
6526 static int netif_alloc_rx_queues(struct net_device *dev)
6528 unsigned int i, count = dev->num_rx_queues;
6529 struct netdev_rx_queue *rx;
6530 size_t sz = count * sizeof(*rx);
6534 rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6542 for (i = 0; i < count; i++)
6548 static void netdev_init_one_queue(struct net_device *dev,
6549 struct netdev_queue *queue, void *_unused)
6551 /* Initialize queue lock */
6552 spin_lock_init(&queue->_xmit_lock);
6553 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
6554 queue->xmit_lock_owner = -1;
6555 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
6558 dql_init(&queue->dql, HZ);
6562 static void netif_free_tx_queues(struct net_device *dev)
6567 static int netif_alloc_netdev_queues(struct net_device *dev)
6569 unsigned int count = dev->num_tx_queues;
6570 struct netdev_queue *tx;
6571 size_t sz = count * sizeof(*tx);
6573 if (count < 1 || count > 0xffff)
6576 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6584 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
6585 spin_lock_init(&dev->tx_global_lock);
6590 void netif_tx_stop_all_queues(struct net_device *dev)
6594 for (i = 0; i < dev->num_tx_queues; i++) {
6595 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
6596 netif_tx_stop_queue(txq);
6599 EXPORT_SYMBOL(netif_tx_stop_all_queues);
6602 * register_netdevice - register a network device
6603 * @dev: device to register
6605 * Take a completed network device structure and add it to the kernel
6606 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6607 * chain. 0 is returned on success. A negative errno code is returned
6608 * on a failure to set up the device, or if the name is a duplicate.
6610 * Callers must hold the rtnl semaphore. You may want
6611 * register_netdev() instead of this.
6614 * The locking appears insufficient to guarantee two parallel registers
6615 * will not get the same name.
6618 int register_netdevice(struct net_device *dev)
6621 struct net *net = dev_net(dev);
6623 BUG_ON(dev_boot_phase);
6628 /* When net_device's are persistent, this will be fatal. */
6629 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
6632 spin_lock_init(&dev->addr_list_lock);
6633 netdev_set_addr_lockdep_class(dev);
6635 ret = dev_get_valid_name(net, dev, dev->name);
6639 /* Init, if this function is available */
6640 if (dev->netdev_ops->ndo_init) {
6641 ret = dev->netdev_ops->ndo_init(dev);
6649 if (((dev->hw_features | dev->features) &
6650 NETIF_F_HW_VLAN_CTAG_FILTER) &&
6651 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
6652 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
6653 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
6660 dev->ifindex = dev_new_index(net);
6661 else if (__dev_get_by_index(net, dev->ifindex))
6664 /* Transfer changeable features to wanted_features and enable
6665 * software offloads (GSO and GRO).
6667 dev->hw_features |= NETIF_F_SOFT_FEATURES;
6668 dev->features |= NETIF_F_SOFT_FEATURES;
6669 dev->wanted_features = dev->features & dev->hw_features;
6671 if (!(dev->flags & IFF_LOOPBACK)) {
6672 dev->hw_features |= NETIF_F_NOCACHE_COPY;
6675 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6677 dev->vlan_features |= NETIF_F_HIGHDMA;
6679 /* Make NETIF_F_SG inheritable to tunnel devices.
6681 dev->hw_enc_features |= NETIF_F_SG;
6683 /* Make NETIF_F_SG inheritable to MPLS.
6685 dev->mpls_features |= NETIF_F_SG;
6687 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
6688 ret = notifier_to_errno(ret);
6692 ret = netdev_register_kobject(dev);
6695 dev->reg_state = NETREG_REGISTERED;
6697 __netdev_update_features(dev);
6700 * Default initial state at registry is that the
6701 * device is present.
6704 set_bit(__LINK_STATE_PRESENT, &dev->state);
6706 linkwatch_init_dev(dev);
6708 dev_init_scheduler(dev);
6710 list_netdevice(dev);
6711 add_device_randomness(dev->dev_addr, dev->addr_len);
6713 /* If the device has permanent device address, driver should
6714 * set dev_addr and also addr_assign_type should be set to
6715 * NET_ADDR_PERM (default value).
6717 if (dev->addr_assign_type == NET_ADDR_PERM)
6718 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
6720 /* Notify protocols, that a new device appeared. */
6721 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
6722 ret = notifier_to_errno(ret);
6724 rollback_registered(dev);
6725 dev->reg_state = NETREG_UNREGISTERED;
6728 * Prevent userspace races by waiting until the network
6729 * device is fully setup before sending notifications.
6731 if (!dev->rtnl_link_ops ||
6732 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6733 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6739 if (dev->netdev_ops->ndo_uninit)
6740 dev->netdev_ops->ndo_uninit(dev);
6743 EXPORT_SYMBOL(register_netdevice);
6746 * init_dummy_netdev - init a dummy network device for NAPI
6747 * @dev: device to init
6749 * This takes a network device structure and initialize the minimum
6750 * amount of fields so it can be used to schedule NAPI polls without
6751 * registering a full blown interface. This is to be used by drivers
6752 * that need to tie several hardware interfaces to a single NAPI
6753 * poll scheduler due to HW limitations.
6755 int init_dummy_netdev(struct net_device *dev)
6757 /* Clear everything. Note we don't initialize spinlocks
6758 * are they aren't supposed to be taken by any of the
6759 * NAPI code and this dummy netdev is supposed to be
6760 * only ever used for NAPI polls
6762 memset(dev, 0, sizeof(struct net_device));
6764 /* make sure we BUG if trying to hit standard
6765 * register/unregister code path
6767 dev->reg_state = NETREG_DUMMY;
6769 /* NAPI wants this */
6770 INIT_LIST_HEAD(&dev->napi_list);
6772 /* a dummy interface is started by default */
6773 set_bit(__LINK_STATE_PRESENT, &dev->state);
6774 set_bit(__LINK_STATE_START, &dev->state);
6776 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6777 * because users of this 'device' dont need to change
6783 EXPORT_SYMBOL_GPL(init_dummy_netdev);
6787 * register_netdev - register a network device
6788 * @dev: device to register
6790 * Take a completed network device structure and add it to the kernel
6791 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6792 * chain. 0 is returned on success. A negative errno code is returned
6793 * on a failure to set up the device, or if the name is a duplicate.
6795 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6796 * and expands the device name if you passed a format string to
6799 int register_netdev(struct net_device *dev)
6804 err = register_netdevice(dev);
6808 EXPORT_SYMBOL(register_netdev);
6810 int netdev_refcnt_read(const struct net_device *dev)
6814 for_each_possible_cpu(i)
6815 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
6818 EXPORT_SYMBOL(netdev_refcnt_read);
6821 * netdev_wait_allrefs - wait until all references are gone.
6822 * @dev: target net_device
6824 * This is called when unregistering network devices.
6826 * Any protocol or device that holds a reference should register
6827 * for netdevice notification, and cleanup and put back the
6828 * reference if they receive an UNREGISTER event.
6829 * We can get stuck here if buggy protocols don't correctly
6832 static void netdev_wait_allrefs(struct net_device *dev)
6834 unsigned long rebroadcast_time, warning_time;
6837 linkwatch_forget_dev(dev);
6839 rebroadcast_time = warning_time = jiffies;
6840 refcnt = netdev_refcnt_read(dev);
6842 while (refcnt != 0) {
6843 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6846 /* Rebroadcast unregister notification */
6847 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6853 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6854 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6856 /* We must not have linkwatch events
6857 * pending on unregister. If this
6858 * happens, we simply run the queue
6859 * unscheduled, resulting in a noop
6862 linkwatch_run_queue();
6867 rebroadcast_time = jiffies;
6872 refcnt = netdev_refcnt_read(dev);
6874 if (time_after(jiffies, warning_time + 10 * HZ)) {
6875 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6877 warning_time = jiffies;
6886 * register_netdevice(x1);
6887 * register_netdevice(x2);
6889 * unregister_netdevice(y1);
6890 * unregister_netdevice(y2);
6896 * We are invoked by rtnl_unlock().
6897 * This allows us to deal with problems:
6898 * 1) We can delete sysfs objects which invoke hotplug
6899 * without deadlocking with linkwatch via keventd.
6900 * 2) Since we run with the RTNL semaphore not held, we can sleep
6901 * safely in order to wait for the netdev refcnt to drop to zero.
6903 * We must not return until all unregister events added during
6904 * the interval the lock was held have been completed.
6906 void netdev_run_todo(void)
6908 struct list_head list;
6910 /* Snapshot list, allow later requests */
6911 list_replace_init(&net_todo_list, &list);
6916 /* Wait for rcu callbacks to finish before next phase */
6917 if (!list_empty(&list))
6920 while (!list_empty(&list)) {
6921 struct net_device *dev
6922 = list_first_entry(&list, struct net_device, todo_list);
6923 list_del(&dev->todo_list);
6926 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6929 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6930 pr_err("network todo '%s' but state %d\n",
6931 dev->name, dev->reg_state);
6936 dev->reg_state = NETREG_UNREGISTERED;
6938 netdev_wait_allrefs(dev);
6941 BUG_ON(netdev_refcnt_read(dev));
6942 BUG_ON(!list_empty(&dev->ptype_all));
6943 BUG_ON(!list_empty(&dev->ptype_specific));
6944 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6945 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6946 WARN_ON(dev->dn_ptr);
6948 if (dev->destructor)
6949 dev->destructor(dev);
6951 /* Report a network device has been unregistered */
6953 dev_net(dev)->dev_unreg_count--;
6955 wake_up(&netdev_unregistering_wq);
6957 /* Free network device */
6958 kobject_put(&dev->dev.kobj);
6962 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6963 * fields in the same order, with only the type differing.
6965 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6966 const struct net_device_stats *netdev_stats)
6968 #if BITS_PER_LONG == 64
6969 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6970 memcpy(stats64, netdev_stats, sizeof(*stats64));
6972 size_t i, n = sizeof(*stats64) / sizeof(u64);
6973 const unsigned long *src = (const unsigned long *)netdev_stats;
6974 u64 *dst = (u64 *)stats64;
6976 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6977 sizeof(*stats64) / sizeof(u64));
6978 for (i = 0; i < n; i++)
6982 EXPORT_SYMBOL(netdev_stats_to_stats64);
6985 * dev_get_stats - get network device statistics
6986 * @dev: device to get statistics from
6987 * @storage: place to store stats
6989 * Get network statistics from device. Return @storage.
6990 * The device driver may provide its own method by setting
6991 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6992 * otherwise the internal statistics structure is used.
6994 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6995 struct rtnl_link_stats64 *storage)
6997 const struct net_device_ops *ops = dev->netdev_ops;
6999 if (ops->ndo_get_stats64) {
7000 memset(storage, 0, sizeof(*storage));
7001 ops->ndo_get_stats64(dev, storage);
7002 } else if (ops->ndo_get_stats) {
7003 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
7005 netdev_stats_to_stats64(storage, &dev->stats);
7007 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
7008 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
7011 EXPORT_SYMBOL(dev_get_stats);
7013 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
7015 struct netdev_queue *queue = dev_ingress_queue(dev);
7017 #ifdef CONFIG_NET_CLS_ACT
7020 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
7023 netdev_init_one_queue(dev, queue, NULL);
7024 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
7025 queue->qdisc_sleeping = &noop_qdisc;
7026 rcu_assign_pointer(dev->ingress_queue, queue);
7031 static const struct ethtool_ops default_ethtool_ops;
7033 void netdev_set_default_ethtool_ops(struct net_device *dev,
7034 const struct ethtool_ops *ops)
7036 if (dev->ethtool_ops == &default_ethtool_ops)
7037 dev->ethtool_ops = ops;
7039 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
7041 void netdev_freemem(struct net_device *dev)
7043 char *addr = (char *)dev - dev->padded;
7049 * alloc_netdev_mqs - allocate network device
7050 * @sizeof_priv: size of private data to allocate space for
7051 * @name: device name format string
7052 * @name_assign_type: origin of device name
7053 * @setup: callback to initialize device
7054 * @txqs: the number of TX subqueues to allocate
7055 * @rxqs: the number of RX subqueues to allocate
7057 * Allocates a struct net_device with private data area for driver use
7058 * and performs basic initialization. Also allocates subqueue structs
7059 * for each queue on the device.
7061 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
7062 unsigned char name_assign_type,
7063 void (*setup)(struct net_device *),
7064 unsigned int txqs, unsigned int rxqs)
7066 struct net_device *dev;
7068 struct net_device *p;
7070 BUG_ON(strlen(name) >= sizeof(dev->name));
7073 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
7079 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
7084 alloc_size = sizeof(struct net_device);
7086 /* ensure 32-byte alignment of private area */
7087 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
7088 alloc_size += sizeof_priv;
7090 /* ensure 32-byte alignment of whole construct */
7091 alloc_size += NETDEV_ALIGN - 1;
7093 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
7095 p = vzalloc(alloc_size);
7099 dev = PTR_ALIGN(p, NETDEV_ALIGN);
7100 dev->padded = (char *)dev - (char *)p;
7102 dev->pcpu_refcnt = alloc_percpu(int);
7103 if (!dev->pcpu_refcnt)
7106 if (dev_addr_init(dev))
7112 dev_net_set(dev, &init_net);
7114 dev->gso_max_size = GSO_MAX_SIZE;
7115 dev->gso_max_segs = GSO_MAX_SEGS;
7116 dev->gso_min_segs = 0;
7118 INIT_LIST_HEAD(&dev->napi_list);
7119 INIT_LIST_HEAD(&dev->unreg_list);
7120 INIT_LIST_HEAD(&dev->close_list);
7121 INIT_LIST_HEAD(&dev->link_watch_list);
7122 INIT_LIST_HEAD(&dev->adj_list.upper);
7123 INIT_LIST_HEAD(&dev->adj_list.lower);
7124 INIT_LIST_HEAD(&dev->all_adj_list.upper);
7125 INIT_LIST_HEAD(&dev->all_adj_list.lower);
7126 INIT_LIST_HEAD(&dev->ptype_all);
7127 INIT_LIST_HEAD(&dev->ptype_specific);
7128 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
7131 if (!dev->tx_queue_len) {
7132 dev->priv_flags |= IFF_NO_QUEUE;
7133 dev->tx_queue_len = 1;
7136 dev->num_tx_queues = txqs;
7137 dev->real_num_tx_queues = txqs;
7138 if (netif_alloc_netdev_queues(dev))
7142 dev->num_rx_queues = rxqs;
7143 dev->real_num_rx_queues = rxqs;
7144 if (netif_alloc_rx_queues(dev))
7148 strcpy(dev->name, name);
7149 dev->name_assign_type = name_assign_type;
7150 dev->group = INIT_NETDEV_GROUP;
7151 if (!dev->ethtool_ops)
7152 dev->ethtool_ops = &default_ethtool_ops;
7154 nf_hook_ingress_init(dev);
7163 free_percpu(dev->pcpu_refcnt);
7165 netdev_freemem(dev);
7168 EXPORT_SYMBOL(alloc_netdev_mqs);
7171 * free_netdev - free network device
7174 * This function does the last stage of destroying an allocated device
7175 * interface. The reference to the device object is released.
7176 * If this is the last reference then it will be freed.
7178 void free_netdev(struct net_device *dev)
7180 struct napi_struct *p, *n;
7182 netif_free_tx_queues(dev);
7187 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
7189 /* Flush device addresses */
7190 dev_addr_flush(dev);
7192 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
7195 free_percpu(dev->pcpu_refcnt);
7196 dev->pcpu_refcnt = NULL;
7198 /* Compatibility with error handling in drivers */
7199 if (dev->reg_state == NETREG_UNINITIALIZED) {
7200 netdev_freemem(dev);
7204 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
7205 dev->reg_state = NETREG_RELEASED;
7207 /* will free via device release */
7208 put_device(&dev->dev);
7210 EXPORT_SYMBOL(free_netdev);
7213 * synchronize_net - Synchronize with packet receive processing
7215 * Wait for packets currently being received to be done.
7216 * Does not block later packets from starting.
7218 void synchronize_net(void)
7221 if (rtnl_is_locked())
7222 synchronize_rcu_expedited();
7226 EXPORT_SYMBOL(synchronize_net);
7229 * unregister_netdevice_queue - remove device from the kernel
7233 * This function shuts down a device interface and removes it
7234 * from the kernel tables.
7235 * If head not NULL, device is queued to be unregistered later.
7237 * Callers must hold the rtnl semaphore. You may want
7238 * unregister_netdev() instead of this.
7241 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
7246 list_move_tail(&dev->unreg_list, head);
7248 rollback_registered(dev);
7249 /* Finish processing unregister after unlock */
7253 EXPORT_SYMBOL(unregister_netdevice_queue);
7256 * unregister_netdevice_many - unregister many devices
7257 * @head: list of devices
7259 * Note: As most callers use a stack allocated list_head,
7260 * we force a list_del() to make sure stack wont be corrupted later.
7262 void unregister_netdevice_many(struct list_head *head)
7264 struct net_device *dev;
7266 if (!list_empty(head)) {
7267 rollback_registered_many(head);
7268 list_for_each_entry(dev, head, unreg_list)
7273 EXPORT_SYMBOL(unregister_netdevice_many);
7276 * unregister_netdev - remove device from the kernel
7279 * This function shuts down a device interface and removes it
7280 * from the kernel tables.
7282 * This is just a wrapper for unregister_netdevice that takes
7283 * the rtnl semaphore. In general you want to use this and not
7284 * unregister_netdevice.
7286 void unregister_netdev(struct net_device *dev)
7289 unregister_netdevice(dev);
7292 EXPORT_SYMBOL(unregister_netdev);
7295 * dev_change_net_namespace - move device to different nethost namespace
7297 * @net: network namespace
7298 * @pat: If not NULL name pattern to try if the current device name
7299 * is already taken in the destination network namespace.
7301 * This function shuts down a device interface and moves it
7302 * to a new network namespace. On success 0 is returned, on
7303 * a failure a netagive errno code is returned.
7305 * Callers must hold the rtnl semaphore.
7308 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
7314 /* Don't allow namespace local devices to be moved. */
7316 if (dev->features & NETIF_F_NETNS_LOCAL)
7319 /* Ensure the device has been registrered */
7320 if (dev->reg_state != NETREG_REGISTERED)
7323 /* Get out if there is nothing todo */
7325 if (net_eq(dev_net(dev), net))
7328 /* Pick the destination device name, and ensure
7329 * we can use it in the destination network namespace.
7332 if (__dev_get_by_name(net, dev->name)) {
7333 /* We get here if we can't use the current device name */
7336 if (dev_get_valid_name(net, dev, pat) < 0)
7341 * And now a mini version of register_netdevice unregister_netdevice.
7344 /* If device is running close it first. */
7347 /* And unlink it from device chain */
7349 unlist_netdevice(dev);
7353 /* Shutdown queueing discipline. */
7356 /* Notify protocols, that we are about to destroy
7357 this device. They should clean all the things.
7359 Note that dev->reg_state stays at NETREG_REGISTERED.
7360 This is wanted because this way 8021q and macvlan know
7361 the device is just moving and can keep their slaves up.
7363 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7365 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7366 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
7369 * Flush the unicast and multicast chains
7374 /* Send a netdev-removed uevent to the old namespace */
7375 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
7376 netdev_adjacent_del_links(dev);
7378 /* Actually switch the network namespace */
7379 dev_net_set(dev, net);
7381 /* If there is an ifindex conflict assign a new one */
7382 if (__dev_get_by_index(net, dev->ifindex))
7383 dev->ifindex = dev_new_index(net);
7385 /* Send a netdev-add uevent to the new namespace */
7386 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
7387 netdev_adjacent_add_links(dev);
7389 /* Fixup kobjects */
7390 err = device_rename(&dev->dev, dev->name);
7393 /* Add the device back in the hashes */
7394 list_netdevice(dev);
7396 /* Notify protocols, that a new device appeared. */
7397 call_netdevice_notifiers(NETDEV_REGISTER, dev);
7400 * Prevent userspace races by waiting until the network
7401 * device is fully setup before sending notifications.
7403 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7410 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
7412 static int dev_cpu_callback(struct notifier_block *nfb,
7413 unsigned long action,
7416 struct sk_buff **list_skb;
7417 struct sk_buff *skb;
7418 unsigned int cpu, oldcpu = (unsigned long)ocpu;
7419 struct softnet_data *sd, *oldsd;
7421 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
7424 local_irq_disable();
7425 cpu = smp_processor_id();
7426 sd = &per_cpu(softnet_data, cpu);
7427 oldsd = &per_cpu(softnet_data, oldcpu);
7429 /* Find end of our completion_queue. */
7430 list_skb = &sd->completion_queue;
7432 list_skb = &(*list_skb)->next;
7433 /* Append completion queue from offline CPU. */
7434 *list_skb = oldsd->completion_queue;
7435 oldsd->completion_queue = NULL;
7437 /* Append output queue from offline CPU. */
7438 if (oldsd->output_queue) {
7439 *sd->output_queue_tailp = oldsd->output_queue;
7440 sd->output_queue_tailp = oldsd->output_queue_tailp;
7441 oldsd->output_queue = NULL;
7442 oldsd->output_queue_tailp = &oldsd->output_queue;
7444 /* Append NAPI poll list from offline CPU, with one exception :
7445 * process_backlog() must be called by cpu owning percpu backlog.
7446 * We properly handle process_queue & input_pkt_queue later.
7448 while (!list_empty(&oldsd->poll_list)) {
7449 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
7453 list_del_init(&napi->poll_list);
7454 if (napi->poll == process_backlog)
7457 ____napi_schedule(sd, napi);
7460 raise_softirq_irqoff(NET_TX_SOFTIRQ);
7463 /* Process offline CPU's input_pkt_queue */
7464 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
7466 input_queue_head_incr(oldsd);
7468 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
7470 input_queue_head_incr(oldsd);
7478 * netdev_increment_features - increment feature set by one
7479 * @all: current feature set
7480 * @one: new feature set
7481 * @mask: mask feature set
7483 * Computes a new feature set after adding a device with feature set
7484 * @one to the master device with current feature set @all. Will not
7485 * enable anything that is off in @mask. Returns the new feature set.
7487 netdev_features_t netdev_increment_features(netdev_features_t all,
7488 netdev_features_t one, netdev_features_t mask)
7490 if (mask & NETIF_F_GEN_CSUM)
7491 mask |= NETIF_F_ALL_CSUM;
7492 mask |= NETIF_F_VLAN_CHALLENGED;
7494 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
7495 all &= one | ~NETIF_F_ALL_FOR_ALL;
7497 /* If one device supports hw checksumming, set for all. */
7498 if (all & NETIF_F_GEN_CSUM)
7499 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
7503 EXPORT_SYMBOL(netdev_increment_features);
7505 static struct hlist_head * __net_init netdev_create_hash(void)
7508 struct hlist_head *hash;
7510 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
7512 for (i = 0; i < NETDEV_HASHENTRIES; i++)
7513 INIT_HLIST_HEAD(&hash[i]);
7518 /* Initialize per network namespace state */
7519 static int __net_init netdev_init(struct net *net)
7521 if (net != &init_net)
7522 INIT_LIST_HEAD(&net->dev_base_head);
7524 net->dev_name_head = netdev_create_hash();
7525 if (net->dev_name_head == NULL)
7528 net->dev_index_head = netdev_create_hash();
7529 if (net->dev_index_head == NULL)
7535 kfree(net->dev_name_head);
7541 * netdev_drivername - network driver for the device
7542 * @dev: network device
7544 * Determine network driver for device.
7546 const char *netdev_drivername(const struct net_device *dev)
7548 const struct device_driver *driver;
7549 const struct device *parent;
7550 const char *empty = "";
7552 parent = dev->dev.parent;
7556 driver = parent->driver;
7557 if (driver && driver->name)
7558 return driver->name;
7562 static void __netdev_printk(const char *level, const struct net_device *dev,
7563 struct va_format *vaf)
7565 if (dev && dev->dev.parent) {
7566 dev_printk_emit(level[1] - '0',
7569 dev_driver_string(dev->dev.parent),
7570 dev_name(dev->dev.parent),
7571 netdev_name(dev), netdev_reg_state(dev),
7574 printk("%s%s%s: %pV",
7575 level, netdev_name(dev), netdev_reg_state(dev), vaf);
7577 printk("%s(NULL net_device): %pV", level, vaf);
7581 void netdev_printk(const char *level, const struct net_device *dev,
7582 const char *format, ...)
7584 struct va_format vaf;
7587 va_start(args, format);
7592 __netdev_printk(level, dev, &vaf);
7596 EXPORT_SYMBOL(netdev_printk);
7598 #define define_netdev_printk_level(func, level) \
7599 void func(const struct net_device *dev, const char *fmt, ...) \
7601 struct va_format vaf; \
7604 va_start(args, fmt); \
7609 __netdev_printk(level, dev, &vaf); \
7613 EXPORT_SYMBOL(func);
7615 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
7616 define_netdev_printk_level(netdev_alert, KERN_ALERT);
7617 define_netdev_printk_level(netdev_crit, KERN_CRIT);
7618 define_netdev_printk_level(netdev_err, KERN_ERR);
7619 define_netdev_printk_level(netdev_warn, KERN_WARNING);
7620 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
7621 define_netdev_printk_level(netdev_info, KERN_INFO);
7623 static void __net_exit netdev_exit(struct net *net)
7625 kfree(net->dev_name_head);
7626 kfree(net->dev_index_head);
7629 static struct pernet_operations __net_initdata netdev_net_ops = {
7630 .init = netdev_init,
7631 .exit = netdev_exit,
7634 static void __net_exit default_device_exit(struct net *net)
7636 struct net_device *dev, *aux;
7638 * Push all migratable network devices back to the
7639 * initial network namespace
7642 for_each_netdev_safe(net, dev, aux) {
7644 char fb_name[IFNAMSIZ];
7646 /* Ignore unmoveable devices (i.e. loopback) */
7647 if (dev->features & NETIF_F_NETNS_LOCAL)
7650 /* Leave virtual devices for the generic cleanup */
7651 if (dev->rtnl_link_ops)
7654 /* Push remaining network devices to init_net */
7655 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
7656 err = dev_change_net_namespace(dev, &init_net, fb_name);
7658 pr_emerg("%s: failed to move %s to init_net: %d\n",
7659 __func__, dev->name, err);
7666 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
7668 /* Return with the rtnl_lock held when there are no network
7669 * devices unregistering in any network namespace in net_list.
7673 DEFINE_WAIT_FUNC(wait, woken_wake_function);
7675 add_wait_queue(&netdev_unregistering_wq, &wait);
7677 unregistering = false;
7679 list_for_each_entry(net, net_list, exit_list) {
7680 if (net->dev_unreg_count > 0) {
7681 unregistering = true;
7689 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
7691 remove_wait_queue(&netdev_unregistering_wq, &wait);
7694 static void __net_exit default_device_exit_batch(struct list_head *net_list)
7696 /* At exit all network devices most be removed from a network
7697 * namespace. Do this in the reverse order of registration.
7698 * Do this across as many network namespaces as possible to
7699 * improve batching efficiency.
7701 struct net_device *dev;
7703 LIST_HEAD(dev_kill_list);
7705 /* To prevent network device cleanup code from dereferencing
7706 * loopback devices or network devices that have been freed
7707 * wait here for all pending unregistrations to complete,
7708 * before unregistring the loopback device and allowing the
7709 * network namespace be freed.
7711 * The netdev todo list containing all network devices
7712 * unregistrations that happen in default_device_exit_batch
7713 * will run in the rtnl_unlock() at the end of
7714 * default_device_exit_batch.
7716 rtnl_lock_unregistering(net_list);
7717 list_for_each_entry(net, net_list, exit_list) {
7718 for_each_netdev_reverse(net, dev) {
7719 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
7720 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
7722 unregister_netdevice_queue(dev, &dev_kill_list);
7725 unregister_netdevice_many(&dev_kill_list);
7729 static struct pernet_operations __net_initdata default_device_ops = {
7730 .exit = default_device_exit,
7731 .exit_batch = default_device_exit_batch,
7735 * Initialize the DEV module. At boot time this walks the device list and
7736 * unhooks any devices that fail to initialise (normally hardware not
7737 * present) and leaves us with a valid list of present and active devices.
7742 * This is called single threaded during boot, so no need
7743 * to take the rtnl semaphore.
7745 static int __init net_dev_init(void)
7747 int i, rc = -ENOMEM;
7749 BUG_ON(!dev_boot_phase);
7751 if (dev_proc_init())
7754 if (netdev_kobject_init())
7757 INIT_LIST_HEAD(&ptype_all);
7758 for (i = 0; i < PTYPE_HASH_SIZE; i++)
7759 INIT_LIST_HEAD(&ptype_base[i]);
7761 INIT_LIST_HEAD(&offload_base);
7763 if (register_pernet_subsys(&netdev_net_ops))
7767 * Initialise the packet receive queues.
7770 for_each_possible_cpu(i) {
7771 struct softnet_data *sd = &per_cpu(softnet_data, i);
7773 skb_queue_head_init(&sd->input_pkt_queue);
7774 skb_queue_head_init(&sd->process_queue);
7775 INIT_LIST_HEAD(&sd->poll_list);
7776 sd->output_queue_tailp = &sd->output_queue;
7778 sd->csd.func = rps_trigger_softirq;
7783 sd->backlog.poll = process_backlog;
7784 sd->backlog.weight = weight_p;
7789 /* The loopback device is special if any other network devices
7790 * is present in a network namespace the loopback device must
7791 * be present. Since we now dynamically allocate and free the
7792 * loopback device ensure this invariant is maintained by
7793 * keeping the loopback device as the first device on the
7794 * list of network devices. Ensuring the loopback devices
7795 * is the first device that appears and the last network device
7798 if (register_pernet_device(&loopback_net_ops))
7801 if (register_pernet_device(&default_device_ops))
7804 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
7805 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
7807 hotcpu_notifier(dev_cpu_callback, 0);
7814 subsys_initcall(net_dev_init);