2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <net/net_namespace.h>
99 #include <linux/rtnetlink.h>
100 #include <linux/stat.h>
102 #include <net/pkt_sched.h>
103 #include <net/checksum.h>
104 #include <net/xfrm.h>
105 #include <linux/highmem.h>
106 #include <linux/init.h>
107 #include <linux/module.h>
108 #include <linux/netpoll.h>
109 #include <linux/rcupdate.h>
110 #include <linux/delay.h>
111 #include <net/iw_handler.h>
112 #include <asm/current.h>
113 #include <linux/audit.h>
114 #include <linux/dmaengine.h>
115 #include <linux/err.h>
116 #include <linux/ctype.h>
117 #include <linux/if_arp.h>
118 #include <linux/if_vlan.h>
119 #include <linux/ip.h>
121 #include <linux/ipv6.h>
122 #include <linux/in.h>
123 #include <linux/jhash.h>
124 #include <linux/random.h>
125 #include <trace/events/napi.h>
126 #include <trace/events/net.h>
127 #include <trace/events/skb.h>
128 #include <linux/pci.h>
129 #include <linux/inetdevice.h>
130 #include <linux/cpu_rmap.h>
131 #include <linux/static_key.h>
132 #include <linux/hashtable.h>
133 #include <linux/vmalloc.h>
134 #include <linux/if_macvlan.h>
136 #include "net-sysfs.h"
138 /* Instead of increasing this, you should create a hash table. */
139 #define MAX_GRO_SKBS 8
141 /* This should be increased if a protocol with a bigger head is added. */
142 #define GRO_MAX_HEAD (MAX_HEADER + 128)
144 static DEFINE_SPINLOCK(ptype_lock);
145 static DEFINE_SPINLOCK(offload_lock);
146 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
147 struct list_head ptype_all __read_mostly; /* Taps */
148 static struct list_head offload_base __read_mostly;
150 static int netif_rx_internal(struct sk_buff *skb);
153 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
156 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
158 * Writers must hold the rtnl semaphore while they loop through the
159 * dev_base_head list, and hold dev_base_lock for writing when they do the
160 * actual updates. This allows pure readers to access the list even
161 * while a writer is preparing to update it.
163 * To put it another way, dev_base_lock is held for writing only to
164 * protect against pure readers; the rtnl semaphore provides the
165 * protection against other writers.
167 * See, for example usages, register_netdevice() and
168 * unregister_netdevice(), which must be called with the rtnl
171 DEFINE_RWLOCK(dev_base_lock);
172 EXPORT_SYMBOL(dev_base_lock);
174 /* protects napi_hash addition/deletion and napi_gen_id */
175 static DEFINE_SPINLOCK(napi_hash_lock);
177 static unsigned int napi_gen_id;
178 static DEFINE_HASHTABLE(napi_hash, 8);
180 static seqcount_t devnet_rename_seq;
182 static inline void dev_base_seq_inc(struct net *net)
184 while (++net->dev_base_seq == 0);
187 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
189 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
191 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
194 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
196 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
199 static inline void rps_lock(struct softnet_data *sd)
202 spin_lock(&sd->input_pkt_queue.lock);
206 static inline void rps_unlock(struct softnet_data *sd)
209 spin_unlock(&sd->input_pkt_queue.lock);
213 /* Device list insertion */
214 static void list_netdevice(struct net_device *dev)
216 struct net *net = dev_net(dev);
220 write_lock_bh(&dev_base_lock);
221 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
222 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
223 hlist_add_head_rcu(&dev->index_hlist,
224 dev_index_hash(net, dev->ifindex));
225 write_unlock_bh(&dev_base_lock);
227 dev_base_seq_inc(net);
230 /* Device list removal
231 * caller must respect a RCU grace period before freeing/reusing dev
233 static void unlist_netdevice(struct net_device *dev)
237 /* Unlink dev from the device chain */
238 write_lock_bh(&dev_base_lock);
239 list_del_rcu(&dev->dev_list);
240 hlist_del_rcu(&dev->name_hlist);
241 hlist_del_rcu(&dev->index_hlist);
242 write_unlock_bh(&dev_base_lock);
244 dev_base_seq_inc(dev_net(dev));
251 static RAW_NOTIFIER_HEAD(netdev_chain);
254 * Device drivers call our routines to queue packets here. We empty the
255 * queue in the local softnet handler.
258 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
259 EXPORT_PER_CPU_SYMBOL(softnet_data);
261 #ifdef CONFIG_LOCKDEP
263 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
264 * according to dev->type
266 static const unsigned short netdev_lock_type[] =
267 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
268 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
269 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
270 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
271 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
272 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
273 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
274 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
275 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
276 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
277 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
278 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
279 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
280 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
281 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
283 static const char *const netdev_lock_name[] =
284 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
285 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
286 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
287 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
288 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
289 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
290 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
291 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
292 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
293 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
294 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
295 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
296 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
297 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
298 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
300 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
301 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
303 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
307 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
308 if (netdev_lock_type[i] == dev_type)
310 /* the last key is used by default */
311 return ARRAY_SIZE(netdev_lock_type) - 1;
314 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
315 unsigned short dev_type)
319 i = netdev_lock_pos(dev_type);
320 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
321 netdev_lock_name[i]);
324 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
328 i = netdev_lock_pos(dev->type);
329 lockdep_set_class_and_name(&dev->addr_list_lock,
330 &netdev_addr_lock_key[i],
331 netdev_lock_name[i]);
334 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
335 unsigned short dev_type)
338 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
343 /*******************************************************************************
345 Protocol management and registration routines
347 *******************************************************************************/
350 * Add a protocol ID to the list. Now that the input handler is
351 * smarter we can dispense with all the messy stuff that used to be
354 * BEWARE!!! Protocol handlers, mangling input packets,
355 * MUST BE last in hash buckets and checking protocol handlers
356 * MUST start from promiscuous ptype_all chain in net_bh.
357 * It is true now, do not change it.
358 * Explanation follows: if protocol handler, mangling packet, will
359 * be the first on list, it is not able to sense, that packet
360 * is cloned and should be copied-on-write, so that it will
361 * change it and subsequent readers will get broken packet.
365 static inline struct list_head *ptype_head(const struct packet_type *pt)
367 if (pt->type == htons(ETH_P_ALL))
370 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
374 * dev_add_pack - add packet handler
375 * @pt: packet type declaration
377 * Add a protocol handler to the networking stack. The passed &packet_type
378 * is linked into kernel lists and may not be freed until it has been
379 * removed from the kernel lists.
381 * This call does not sleep therefore it can not
382 * guarantee all CPU's that are in middle of receiving packets
383 * will see the new packet type (until the next received packet).
386 void dev_add_pack(struct packet_type *pt)
388 struct list_head *head = ptype_head(pt);
390 spin_lock(&ptype_lock);
391 list_add_rcu(&pt->list, head);
392 spin_unlock(&ptype_lock);
394 EXPORT_SYMBOL(dev_add_pack);
397 * __dev_remove_pack - remove packet handler
398 * @pt: packet type declaration
400 * Remove a protocol handler that was previously added to the kernel
401 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
402 * from the kernel lists and can be freed or reused once this function
405 * The packet type might still be in use by receivers
406 * and must not be freed until after all the CPU's have gone
407 * through a quiescent state.
409 void __dev_remove_pack(struct packet_type *pt)
411 struct list_head *head = ptype_head(pt);
412 struct packet_type *pt1;
414 spin_lock(&ptype_lock);
416 list_for_each_entry(pt1, head, list) {
418 list_del_rcu(&pt->list);
423 pr_warn("dev_remove_pack: %p not found\n", pt);
425 spin_unlock(&ptype_lock);
427 EXPORT_SYMBOL(__dev_remove_pack);
430 * dev_remove_pack - remove packet handler
431 * @pt: packet type declaration
433 * Remove a protocol handler that was previously added to the kernel
434 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
435 * from the kernel lists and can be freed or reused once this function
438 * This call sleeps to guarantee that no CPU is looking at the packet
441 void dev_remove_pack(struct packet_type *pt)
443 __dev_remove_pack(pt);
447 EXPORT_SYMBOL(dev_remove_pack);
451 * dev_add_offload - register offload handlers
452 * @po: protocol offload declaration
454 * Add protocol offload handlers to the networking stack. The passed
455 * &proto_offload is linked into kernel lists and may not be freed until
456 * it has been removed from the kernel lists.
458 * This call does not sleep therefore it can not
459 * guarantee all CPU's that are in middle of receiving packets
460 * will see the new offload handlers (until the next received packet).
462 void dev_add_offload(struct packet_offload *po)
464 struct list_head *head = &offload_base;
466 spin_lock(&offload_lock);
467 list_add_rcu(&po->list, head);
468 spin_unlock(&offload_lock);
470 EXPORT_SYMBOL(dev_add_offload);
473 * __dev_remove_offload - remove offload handler
474 * @po: packet offload declaration
476 * Remove a protocol offload handler that was previously added to the
477 * kernel offload handlers by dev_add_offload(). The passed &offload_type
478 * is removed from the kernel lists and can be freed or reused once this
481 * The packet type might still be in use by receivers
482 * and must not be freed until after all the CPU's have gone
483 * through a quiescent state.
485 static void __dev_remove_offload(struct packet_offload *po)
487 struct list_head *head = &offload_base;
488 struct packet_offload *po1;
490 spin_lock(&offload_lock);
492 list_for_each_entry(po1, head, list) {
494 list_del_rcu(&po->list);
499 pr_warn("dev_remove_offload: %p not found\n", po);
501 spin_unlock(&offload_lock);
505 * dev_remove_offload - remove packet offload handler
506 * @po: packet offload declaration
508 * Remove a packet offload handler that was previously added to the kernel
509 * offload handlers by dev_add_offload(). The passed &offload_type is
510 * removed from the kernel lists and can be freed or reused once this
513 * This call sleeps to guarantee that no CPU is looking at the packet
516 void dev_remove_offload(struct packet_offload *po)
518 __dev_remove_offload(po);
522 EXPORT_SYMBOL(dev_remove_offload);
524 /******************************************************************************
526 Device Boot-time Settings Routines
528 *******************************************************************************/
530 /* Boot time configuration table */
531 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
534 * netdev_boot_setup_add - add new setup entry
535 * @name: name of the device
536 * @map: configured settings for the device
538 * Adds new setup entry to the dev_boot_setup list. The function
539 * returns 0 on error and 1 on success. This is a generic routine to
542 static int netdev_boot_setup_add(char *name, struct ifmap *map)
544 struct netdev_boot_setup *s;
548 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
549 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
550 memset(s[i].name, 0, sizeof(s[i].name));
551 strlcpy(s[i].name, name, IFNAMSIZ);
552 memcpy(&s[i].map, map, sizeof(s[i].map));
557 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
561 * netdev_boot_setup_check - check boot time settings
562 * @dev: the netdevice
564 * Check boot time settings for the device.
565 * The found settings are set for the device to be used
566 * later in the device probing.
567 * Returns 0 if no settings found, 1 if they are.
569 int netdev_boot_setup_check(struct net_device *dev)
571 struct netdev_boot_setup *s = dev_boot_setup;
574 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
575 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
576 !strcmp(dev->name, s[i].name)) {
577 dev->irq = s[i].map.irq;
578 dev->base_addr = s[i].map.base_addr;
579 dev->mem_start = s[i].map.mem_start;
580 dev->mem_end = s[i].map.mem_end;
586 EXPORT_SYMBOL(netdev_boot_setup_check);
590 * netdev_boot_base - get address from boot time settings
591 * @prefix: prefix for network device
592 * @unit: id for network device
594 * Check boot time settings for the base address of device.
595 * The found settings are set for the device to be used
596 * later in the device probing.
597 * Returns 0 if no settings found.
599 unsigned long netdev_boot_base(const char *prefix, int unit)
601 const struct netdev_boot_setup *s = dev_boot_setup;
605 sprintf(name, "%s%d", prefix, unit);
608 * If device already registered then return base of 1
609 * to indicate not to probe for this interface
611 if (__dev_get_by_name(&init_net, name))
614 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
615 if (!strcmp(name, s[i].name))
616 return s[i].map.base_addr;
621 * Saves at boot time configured settings for any netdevice.
623 int __init netdev_boot_setup(char *str)
628 str = get_options(str, ARRAY_SIZE(ints), ints);
633 memset(&map, 0, sizeof(map));
637 map.base_addr = ints[2];
639 map.mem_start = ints[3];
641 map.mem_end = ints[4];
643 /* Add new entry to the list */
644 return netdev_boot_setup_add(str, &map);
647 __setup("netdev=", netdev_boot_setup);
649 /*******************************************************************************
651 Device Interface Subroutines
653 *******************************************************************************/
656 * __dev_get_by_name - find a device by its name
657 * @net: the applicable net namespace
658 * @name: name to find
660 * Find an interface by name. Must be called under RTNL semaphore
661 * or @dev_base_lock. If the name is found a pointer to the device
662 * is returned. If the name is not found then %NULL is returned. The
663 * reference counters are not incremented so the caller must be
664 * careful with locks.
667 struct net_device *__dev_get_by_name(struct net *net, const char *name)
669 struct net_device *dev;
670 struct hlist_head *head = dev_name_hash(net, name);
672 hlist_for_each_entry(dev, head, name_hlist)
673 if (!strncmp(dev->name, name, IFNAMSIZ))
678 EXPORT_SYMBOL(__dev_get_by_name);
681 * dev_get_by_name_rcu - find a device by its name
682 * @net: the applicable net namespace
683 * @name: name to find
685 * Find an interface by name.
686 * If the name is found a pointer to the device is returned.
687 * If the name is not found then %NULL is returned.
688 * The reference counters are not incremented so the caller must be
689 * careful with locks. The caller must hold RCU lock.
692 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
694 struct net_device *dev;
695 struct hlist_head *head = dev_name_hash(net, name);
697 hlist_for_each_entry_rcu(dev, head, name_hlist)
698 if (!strncmp(dev->name, name, IFNAMSIZ))
703 EXPORT_SYMBOL(dev_get_by_name_rcu);
706 * dev_get_by_name - find a device by its name
707 * @net: the applicable net namespace
708 * @name: name to find
710 * Find an interface by name. This can be called from any
711 * context and does its own locking. The returned handle has
712 * the usage count incremented and the caller must use dev_put() to
713 * release it when it is no longer needed. %NULL is returned if no
714 * matching device is found.
717 struct net_device *dev_get_by_name(struct net *net, const char *name)
719 struct net_device *dev;
722 dev = dev_get_by_name_rcu(net, name);
728 EXPORT_SYMBOL(dev_get_by_name);
731 * __dev_get_by_index - find a device by its ifindex
732 * @net: the applicable net namespace
733 * @ifindex: index of device
735 * Search for an interface by index. Returns %NULL if the device
736 * is not found or a pointer to the device. The device has not
737 * had its reference counter increased so the caller must be careful
738 * about locking. The caller must hold either the RTNL semaphore
742 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
744 struct net_device *dev;
745 struct hlist_head *head = dev_index_hash(net, ifindex);
747 hlist_for_each_entry(dev, head, index_hlist)
748 if (dev->ifindex == ifindex)
753 EXPORT_SYMBOL(__dev_get_by_index);
756 * dev_get_by_index_rcu - find a device by its ifindex
757 * @net: the applicable net namespace
758 * @ifindex: index of device
760 * Search for an interface by index. Returns %NULL if the device
761 * is not found or a pointer to the device. The device has not
762 * had its reference counter increased so the caller must be careful
763 * about locking. The caller must hold RCU lock.
766 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
768 struct net_device *dev;
769 struct hlist_head *head = dev_index_hash(net, ifindex);
771 hlist_for_each_entry_rcu(dev, head, index_hlist)
772 if (dev->ifindex == ifindex)
777 EXPORT_SYMBOL(dev_get_by_index_rcu);
781 * dev_get_by_index - find a device by its ifindex
782 * @net: the applicable net namespace
783 * @ifindex: index of device
785 * Search for an interface by index. Returns NULL if the device
786 * is not found or a pointer to the device. The device returned has
787 * had a reference added and the pointer is safe until the user calls
788 * dev_put to indicate they have finished with it.
791 struct net_device *dev_get_by_index(struct net *net, int ifindex)
793 struct net_device *dev;
796 dev = dev_get_by_index_rcu(net, ifindex);
802 EXPORT_SYMBOL(dev_get_by_index);
805 * netdev_get_name - get a netdevice name, knowing its ifindex.
806 * @net: network namespace
807 * @name: a pointer to the buffer where the name will be stored.
808 * @ifindex: the ifindex of the interface to get the name from.
810 * The use of raw_seqcount_begin() and cond_resched() before
811 * retrying is required as we want to give the writers a chance
812 * to complete when CONFIG_PREEMPT is not set.
814 int netdev_get_name(struct net *net, char *name, int ifindex)
816 struct net_device *dev;
820 seq = raw_seqcount_begin(&devnet_rename_seq);
822 dev = dev_get_by_index_rcu(net, ifindex);
828 strcpy(name, dev->name);
830 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
839 * dev_getbyhwaddr_rcu - find a device by its hardware address
840 * @net: the applicable net namespace
841 * @type: media type of device
842 * @ha: hardware address
844 * Search for an interface by MAC address. Returns NULL if the device
845 * is not found or a pointer to the device.
846 * The caller must hold RCU or RTNL.
847 * The returned device has not had its ref count increased
848 * and the caller must therefore be careful about locking
852 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
855 struct net_device *dev;
857 for_each_netdev_rcu(net, dev)
858 if (dev->type == type &&
859 !memcmp(dev->dev_addr, ha, dev->addr_len))
864 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
866 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
868 struct net_device *dev;
871 for_each_netdev(net, dev)
872 if (dev->type == type)
877 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
879 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
881 struct net_device *dev, *ret = NULL;
884 for_each_netdev_rcu(net, dev)
885 if (dev->type == type) {
893 EXPORT_SYMBOL(dev_getfirstbyhwtype);
896 * dev_get_by_flags_rcu - find any device with given flags
897 * @net: the applicable net namespace
898 * @if_flags: IFF_* values
899 * @mask: bitmask of bits in if_flags to check
901 * Search for any interface with the given flags. Returns NULL if a device
902 * is not found or a pointer to the device. Must be called inside
903 * rcu_read_lock(), and result refcount is unchanged.
906 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
909 struct net_device *dev, *ret;
912 for_each_netdev_rcu(net, dev) {
913 if (((dev->flags ^ if_flags) & mask) == 0) {
920 EXPORT_SYMBOL(dev_get_by_flags_rcu);
923 * dev_valid_name - check if name is okay for network device
926 * Network device names need to be valid file names to
927 * to allow sysfs to work. We also disallow any kind of
930 bool dev_valid_name(const char *name)
934 if (strlen(name) >= IFNAMSIZ)
936 if (!strcmp(name, ".") || !strcmp(name, ".."))
940 if (*name == '/' || isspace(*name))
946 EXPORT_SYMBOL(dev_valid_name);
949 * __dev_alloc_name - allocate a name for a device
950 * @net: network namespace to allocate the device name in
951 * @name: name format string
952 * @buf: scratch buffer and result name string
954 * Passed a format string - eg "lt%d" it will try and find a suitable
955 * id. It scans list of devices to build up a free map, then chooses
956 * the first empty slot. The caller must hold the dev_base or rtnl lock
957 * while allocating the name and adding the device in order to avoid
959 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
960 * Returns the number of the unit assigned or a negative errno code.
963 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
967 const int max_netdevices = 8*PAGE_SIZE;
968 unsigned long *inuse;
969 struct net_device *d;
971 p = strnchr(name, IFNAMSIZ-1, '%');
974 * Verify the string as this thing may have come from
975 * the user. There must be either one "%d" and no other "%"
978 if (p[1] != 'd' || strchr(p + 2, '%'))
981 /* Use one page as a bit array of possible slots */
982 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
986 for_each_netdev(net, d) {
987 if (!sscanf(d->name, name, &i))
989 if (i < 0 || i >= max_netdevices)
992 /* avoid cases where sscanf is not exact inverse of printf */
993 snprintf(buf, IFNAMSIZ, name, i);
994 if (!strncmp(buf, d->name, IFNAMSIZ))
998 i = find_first_zero_bit(inuse, max_netdevices);
999 free_page((unsigned long) inuse);
1003 snprintf(buf, IFNAMSIZ, name, i);
1004 if (!__dev_get_by_name(net, buf))
1007 /* It is possible to run out of possible slots
1008 * when the name is long and there isn't enough space left
1009 * for the digits, or if all bits are used.
1015 * dev_alloc_name - allocate a name for a device
1017 * @name: name format string
1019 * Passed a format string - eg "lt%d" it will try and find a suitable
1020 * id. It scans list of devices to build up a free map, then chooses
1021 * the first empty slot. The caller must hold the dev_base or rtnl lock
1022 * while allocating the name and adding the device in order to avoid
1024 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1025 * Returns the number of the unit assigned or a negative errno code.
1028 int dev_alloc_name(struct net_device *dev, const char *name)
1034 BUG_ON(!dev_net(dev));
1036 ret = __dev_alloc_name(net, name, buf);
1038 strlcpy(dev->name, buf, IFNAMSIZ);
1041 EXPORT_SYMBOL(dev_alloc_name);
1043 static int dev_alloc_name_ns(struct net *net,
1044 struct net_device *dev,
1050 ret = __dev_alloc_name(net, name, buf);
1052 strlcpy(dev->name, buf, IFNAMSIZ);
1056 static int dev_get_valid_name(struct net *net,
1057 struct net_device *dev,
1062 if (!dev_valid_name(name))
1065 if (strchr(name, '%'))
1066 return dev_alloc_name_ns(net, dev, name);
1067 else if (__dev_get_by_name(net, name))
1069 else if (dev->name != name)
1070 strlcpy(dev->name, name, IFNAMSIZ);
1076 * dev_change_name - change name of a device
1078 * @newname: name (or format string) must be at least IFNAMSIZ
1080 * Change name of a device, can pass format strings "eth%d".
1083 int dev_change_name(struct net_device *dev, const char *newname)
1085 char oldname[IFNAMSIZ];
1091 BUG_ON(!dev_net(dev));
1094 if (dev->flags & IFF_UP)
1097 write_seqcount_begin(&devnet_rename_seq);
1099 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1100 write_seqcount_end(&devnet_rename_seq);
1104 memcpy(oldname, dev->name, IFNAMSIZ);
1106 err = dev_get_valid_name(net, dev, newname);
1108 write_seqcount_end(&devnet_rename_seq);
1113 ret = device_rename(&dev->dev, dev->name);
1115 memcpy(dev->name, oldname, IFNAMSIZ);
1116 write_seqcount_end(&devnet_rename_seq);
1120 write_seqcount_end(&devnet_rename_seq);
1122 netdev_adjacent_rename_links(dev, oldname);
1124 write_lock_bh(&dev_base_lock);
1125 hlist_del_rcu(&dev->name_hlist);
1126 write_unlock_bh(&dev_base_lock);
1130 write_lock_bh(&dev_base_lock);
1131 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1132 write_unlock_bh(&dev_base_lock);
1134 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1135 ret = notifier_to_errno(ret);
1138 /* err >= 0 after dev_alloc_name() or stores the first errno */
1141 write_seqcount_begin(&devnet_rename_seq);
1142 memcpy(dev->name, oldname, IFNAMSIZ);
1143 memcpy(oldname, newname, IFNAMSIZ);
1146 pr_err("%s: name change rollback failed: %d\n",
1155 * dev_set_alias - change ifalias of a device
1157 * @alias: name up to IFALIASZ
1158 * @len: limit of bytes to copy from info
1160 * Set ifalias for a device,
1162 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1168 if (len >= IFALIASZ)
1172 kfree(dev->ifalias);
1173 dev->ifalias = NULL;
1177 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1180 dev->ifalias = new_ifalias;
1182 strlcpy(dev->ifalias, alias, len+1);
1188 * netdev_features_change - device changes features
1189 * @dev: device to cause notification
1191 * Called to indicate a device has changed features.
1193 void netdev_features_change(struct net_device *dev)
1195 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1197 EXPORT_SYMBOL(netdev_features_change);
1200 * netdev_state_change - device changes state
1201 * @dev: device to cause notification
1203 * Called to indicate a device has changed state. This function calls
1204 * the notifier chains for netdev_chain and sends a NEWLINK message
1205 * to the routing socket.
1207 void netdev_state_change(struct net_device *dev)
1209 if (dev->flags & IFF_UP) {
1210 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1211 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1214 EXPORT_SYMBOL(netdev_state_change);
1217 * netdev_notify_peers - notify network peers about existence of @dev
1218 * @dev: network device
1220 * Generate traffic such that interested network peers are aware of
1221 * @dev, such as by generating a gratuitous ARP. This may be used when
1222 * a device wants to inform the rest of the network about some sort of
1223 * reconfiguration such as a failover event or virtual machine
1226 void netdev_notify_peers(struct net_device *dev)
1229 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1232 EXPORT_SYMBOL(netdev_notify_peers);
1234 static int __dev_open(struct net_device *dev)
1236 const struct net_device_ops *ops = dev->netdev_ops;
1241 if (!netif_device_present(dev))
1244 /* Block netpoll from trying to do any rx path servicing.
1245 * If we don't do this there is a chance ndo_poll_controller
1246 * or ndo_poll may be running while we open the device
1248 netpoll_rx_disable(dev);
1250 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1251 ret = notifier_to_errno(ret);
1255 set_bit(__LINK_STATE_START, &dev->state);
1257 if (ops->ndo_validate_addr)
1258 ret = ops->ndo_validate_addr(dev);
1260 if (!ret && ops->ndo_open)
1261 ret = ops->ndo_open(dev);
1263 netpoll_rx_enable(dev);
1266 clear_bit(__LINK_STATE_START, &dev->state);
1268 dev->flags |= IFF_UP;
1269 net_dmaengine_get();
1270 dev_set_rx_mode(dev);
1272 add_device_randomness(dev->dev_addr, dev->addr_len);
1279 * dev_open - prepare an interface for use.
1280 * @dev: device to open
1282 * Takes a device from down to up state. The device's private open
1283 * function is invoked and then the multicast lists are loaded. Finally
1284 * the device is moved into the up state and a %NETDEV_UP message is
1285 * sent to the netdev notifier chain.
1287 * Calling this function on an active interface is a nop. On a failure
1288 * a negative errno code is returned.
1290 int dev_open(struct net_device *dev)
1294 if (dev->flags & IFF_UP)
1297 ret = __dev_open(dev);
1301 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1302 call_netdevice_notifiers(NETDEV_UP, dev);
1306 EXPORT_SYMBOL(dev_open);
1308 static int __dev_close_many(struct list_head *head)
1310 struct net_device *dev;
1315 list_for_each_entry(dev, head, close_list) {
1316 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1318 clear_bit(__LINK_STATE_START, &dev->state);
1320 /* Synchronize to scheduled poll. We cannot touch poll list, it
1321 * can be even on different cpu. So just clear netif_running().
1323 * dev->stop() will invoke napi_disable() on all of it's
1324 * napi_struct instances on this device.
1326 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1329 dev_deactivate_many(head);
1331 list_for_each_entry(dev, head, close_list) {
1332 const struct net_device_ops *ops = dev->netdev_ops;
1335 * Call the device specific close. This cannot fail.
1336 * Only if device is UP
1338 * We allow it to be called even after a DETACH hot-plug
1344 dev->flags &= ~IFF_UP;
1345 net_dmaengine_put();
1351 static int __dev_close(struct net_device *dev)
1356 /* Temporarily disable netpoll until the interface is down */
1357 netpoll_rx_disable(dev);
1359 list_add(&dev->close_list, &single);
1360 retval = __dev_close_many(&single);
1363 netpoll_rx_enable(dev);
1367 static int dev_close_many(struct list_head *head)
1369 struct net_device *dev, *tmp;
1371 /* Remove the devices that don't need to be closed */
1372 list_for_each_entry_safe(dev, tmp, head, close_list)
1373 if (!(dev->flags & IFF_UP))
1374 list_del_init(&dev->close_list);
1376 __dev_close_many(head);
1378 list_for_each_entry_safe(dev, tmp, head, close_list) {
1379 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1380 call_netdevice_notifiers(NETDEV_DOWN, dev);
1381 list_del_init(&dev->close_list);
1388 * dev_close - shutdown an interface.
1389 * @dev: device to shutdown
1391 * This function moves an active device into down state. A
1392 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1393 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1396 int dev_close(struct net_device *dev)
1398 if (dev->flags & IFF_UP) {
1401 /* Block netpoll rx while the interface is going down */
1402 netpoll_rx_disable(dev);
1404 list_add(&dev->close_list, &single);
1405 dev_close_many(&single);
1408 netpoll_rx_enable(dev);
1412 EXPORT_SYMBOL(dev_close);
1416 * dev_disable_lro - disable Large Receive Offload on a device
1419 * Disable Large Receive Offload (LRO) on a net device. Must be
1420 * called under RTNL. This is needed if received packets may be
1421 * forwarded to another interface.
1423 void dev_disable_lro(struct net_device *dev)
1426 * If we're trying to disable lro on a vlan device
1427 * use the underlying physical device instead
1429 if (is_vlan_dev(dev))
1430 dev = vlan_dev_real_dev(dev);
1432 /* the same for macvlan devices */
1433 if (netif_is_macvlan(dev))
1434 dev = macvlan_dev_real_dev(dev);
1436 dev->wanted_features &= ~NETIF_F_LRO;
1437 netdev_update_features(dev);
1439 if (unlikely(dev->features & NETIF_F_LRO))
1440 netdev_WARN(dev, "failed to disable LRO!\n");
1442 EXPORT_SYMBOL(dev_disable_lro);
1444 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1445 struct net_device *dev)
1447 struct netdev_notifier_info info;
1449 netdev_notifier_info_init(&info, dev);
1450 return nb->notifier_call(nb, val, &info);
1453 static int dev_boot_phase = 1;
1456 * register_netdevice_notifier - register a network notifier block
1459 * Register a notifier to be called when network device events occur.
1460 * The notifier passed is linked into the kernel structures and must
1461 * not be reused until it has been unregistered. A negative errno code
1462 * is returned on a failure.
1464 * When registered all registration and up events are replayed
1465 * to the new notifier to allow device to have a race free
1466 * view of the network device list.
1469 int register_netdevice_notifier(struct notifier_block *nb)
1471 struct net_device *dev;
1472 struct net_device *last;
1477 err = raw_notifier_chain_register(&netdev_chain, nb);
1483 for_each_netdev(net, dev) {
1484 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1485 err = notifier_to_errno(err);
1489 if (!(dev->flags & IFF_UP))
1492 call_netdevice_notifier(nb, NETDEV_UP, dev);
1503 for_each_netdev(net, dev) {
1507 if (dev->flags & IFF_UP) {
1508 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1510 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1512 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1517 raw_notifier_chain_unregister(&netdev_chain, nb);
1520 EXPORT_SYMBOL(register_netdevice_notifier);
1523 * unregister_netdevice_notifier - unregister a network notifier block
1526 * Unregister a notifier previously registered by
1527 * register_netdevice_notifier(). The notifier is unlinked into the
1528 * kernel structures and may then be reused. A negative errno code
1529 * is returned on a failure.
1531 * After unregistering unregister and down device events are synthesized
1532 * for all devices on the device list to the removed notifier to remove
1533 * the need for special case cleanup code.
1536 int unregister_netdevice_notifier(struct notifier_block *nb)
1538 struct net_device *dev;
1543 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1548 for_each_netdev(net, dev) {
1549 if (dev->flags & IFF_UP) {
1550 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1552 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1554 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1561 EXPORT_SYMBOL(unregister_netdevice_notifier);
1564 * call_netdevice_notifiers_info - call all network notifier blocks
1565 * @val: value passed unmodified to notifier function
1566 * @dev: net_device pointer passed unmodified to notifier function
1567 * @info: notifier information data
1569 * Call all network notifier blocks. Parameters and return value
1570 * are as for raw_notifier_call_chain().
1573 static int call_netdevice_notifiers_info(unsigned long val,
1574 struct net_device *dev,
1575 struct netdev_notifier_info *info)
1578 netdev_notifier_info_init(info, dev);
1579 return raw_notifier_call_chain(&netdev_chain, val, info);
1583 * call_netdevice_notifiers - call all network notifier blocks
1584 * @val: value passed unmodified to notifier function
1585 * @dev: net_device pointer passed unmodified to notifier function
1587 * Call all network notifier blocks. Parameters and return value
1588 * are as for raw_notifier_call_chain().
1591 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1593 struct netdev_notifier_info info;
1595 return call_netdevice_notifiers_info(val, dev, &info);
1597 EXPORT_SYMBOL(call_netdevice_notifiers);
1599 static struct static_key netstamp_needed __read_mostly;
1600 #ifdef HAVE_JUMP_LABEL
1601 /* We are not allowed to call static_key_slow_dec() from irq context
1602 * If net_disable_timestamp() is called from irq context, defer the
1603 * static_key_slow_dec() calls.
1605 static atomic_t netstamp_needed_deferred;
1608 void net_enable_timestamp(void)
1610 #ifdef HAVE_JUMP_LABEL
1611 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1615 static_key_slow_dec(&netstamp_needed);
1619 static_key_slow_inc(&netstamp_needed);
1621 EXPORT_SYMBOL(net_enable_timestamp);
1623 void net_disable_timestamp(void)
1625 #ifdef HAVE_JUMP_LABEL
1626 if (in_interrupt()) {
1627 atomic_inc(&netstamp_needed_deferred);
1631 static_key_slow_dec(&netstamp_needed);
1633 EXPORT_SYMBOL(net_disable_timestamp);
1635 static inline void net_timestamp_set(struct sk_buff *skb)
1637 skb->tstamp.tv64 = 0;
1638 if (static_key_false(&netstamp_needed))
1639 __net_timestamp(skb);
1642 #define net_timestamp_check(COND, SKB) \
1643 if (static_key_false(&netstamp_needed)) { \
1644 if ((COND) && !(SKB)->tstamp.tv64) \
1645 __net_timestamp(SKB); \
1648 static inline bool is_skb_forwardable(struct net_device *dev,
1649 struct sk_buff *skb)
1653 if (!(dev->flags & IFF_UP))
1656 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1657 if (skb->len <= len)
1660 /* if TSO is enabled, we don't care about the length as the packet
1661 * could be forwarded without being segmented before
1663 if (skb_is_gso(skb))
1670 * dev_forward_skb - loopback an skb to another netif
1672 * @dev: destination network device
1673 * @skb: buffer to forward
1676 * NET_RX_SUCCESS (no congestion)
1677 * NET_RX_DROP (packet was dropped, but freed)
1679 * dev_forward_skb can be used for injecting an skb from the
1680 * start_xmit function of one device into the receive queue
1681 * of another device.
1683 * The receiving device may be in another namespace, so
1684 * we have to clear all information in the skb that could
1685 * impact namespace isolation.
1687 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1689 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1690 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1691 atomic_long_inc(&dev->rx_dropped);
1697 if (unlikely(!is_skb_forwardable(dev, skb))) {
1698 atomic_long_inc(&dev->rx_dropped);
1703 skb_scrub_packet(skb, true);
1704 skb->protocol = eth_type_trans(skb, dev);
1706 return netif_rx_internal(skb);
1708 EXPORT_SYMBOL_GPL(dev_forward_skb);
1710 static inline int deliver_skb(struct sk_buff *skb,
1711 struct packet_type *pt_prev,
1712 struct net_device *orig_dev)
1714 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1716 atomic_inc(&skb->users);
1717 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1720 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1722 if (!ptype->af_packet_priv || !skb->sk)
1725 if (ptype->id_match)
1726 return ptype->id_match(ptype, skb->sk);
1727 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1734 * Support routine. Sends outgoing frames to any network
1735 * taps currently in use.
1738 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1740 struct packet_type *ptype;
1741 struct sk_buff *skb2 = NULL;
1742 struct packet_type *pt_prev = NULL;
1745 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1746 /* Never send packets back to the socket
1747 * they originated from - MvS (miquels@drinkel.ow.org)
1749 if ((ptype->dev == dev || !ptype->dev) &&
1750 (!skb_loop_sk(ptype, skb))) {
1752 deliver_skb(skb2, pt_prev, skb->dev);
1757 skb2 = skb_clone(skb, GFP_ATOMIC);
1761 net_timestamp_set(skb2);
1763 /* skb->nh should be correctly
1764 set by sender, so that the second statement is
1765 just protection against buggy protocols.
1767 skb_reset_mac_header(skb2);
1769 if (skb_network_header(skb2) < skb2->data ||
1770 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1771 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1772 ntohs(skb2->protocol),
1774 skb_reset_network_header(skb2);
1777 skb2->transport_header = skb2->network_header;
1778 skb2->pkt_type = PACKET_OUTGOING;
1783 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1788 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1789 * @dev: Network device
1790 * @txq: number of queues available
1792 * If real_num_tx_queues is changed the tc mappings may no longer be
1793 * valid. To resolve this verify the tc mapping remains valid and if
1794 * not NULL the mapping. With no priorities mapping to this
1795 * offset/count pair it will no longer be used. In the worst case TC0
1796 * is invalid nothing can be done so disable priority mappings. If is
1797 * expected that drivers will fix this mapping if they can before
1798 * calling netif_set_real_num_tx_queues.
1800 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1803 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1805 /* If TC0 is invalidated disable TC mapping */
1806 if (tc->offset + tc->count > txq) {
1807 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1812 /* Invalidated prio to tc mappings set to TC0 */
1813 for (i = 1; i < TC_BITMASK + 1; i++) {
1814 int q = netdev_get_prio_tc_map(dev, i);
1816 tc = &dev->tc_to_txq[q];
1817 if (tc->offset + tc->count > txq) {
1818 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1820 netdev_set_prio_tc_map(dev, i, 0);
1826 static DEFINE_MUTEX(xps_map_mutex);
1827 #define xmap_dereference(P) \
1828 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1830 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1833 struct xps_map *map = NULL;
1837 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1839 for (pos = 0; map && pos < map->len; pos++) {
1840 if (map->queues[pos] == index) {
1842 map->queues[pos] = map->queues[--map->len];
1844 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1845 kfree_rcu(map, rcu);
1855 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1857 struct xps_dev_maps *dev_maps;
1859 bool active = false;
1861 mutex_lock(&xps_map_mutex);
1862 dev_maps = xmap_dereference(dev->xps_maps);
1867 for_each_possible_cpu(cpu) {
1868 for (i = index; i < dev->num_tx_queues; i++) {
1869 if (!remove_xps_queue(dev_maps, cpu, i))
1872 if (i == dev->num_tx_queues)
1877 RCU_INIT_POINTER(dev->xps_maps, NULL);
1878 kfree_rcu(dev_maps, rcu);
1881 for (i = index; i < dev->num_tx_queues; i++)
1882 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1886 mutex_unlock(&xps_map_mutex);
1889 static struct xps_map *expand_xps_map(struct xps_map *map,
1892 struct xps_map *new_map;
1893 int alloc_len = XPS_MIN_MAP_ALLOC;
1896 for (pos = 0; map && pos < map->len; pos++) {
1897 if (map->queues[pos] != index)
1902 /* Need to add queue to this CPU's existing map */
1904 if (pos < map->alloc_len)
1907 alloc_len = map->alloc_len * 2;
1910 /* Need to allocate new map to store queue on this CPU's map */
1911 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
1916 for (i = 0; i < pos; i++)
1917 new_map->queues[i] = map->queues[i];
1918 new_map->alloc_len = alloc_len;
1924 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
1927 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
1928 struct xps_map *map, *new_map;
1929 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
1930 int cpu, numa_node_id = -2;
1931 bool active = false;
1933 mutex_lock(&xps_map_mutex);
1935 dev_maps = xmap_dereference(dev->xps_maps);
1937 /* allocate memory for queue storage */
1938 for_each_online_cpu(cpu) {
1939 if (!cpumask_test_cpu(cpu, mask))
1943 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
1944 if (!new_dev_maps) {
1945 mutex_unlock(&xps_map_mutex);
1949 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
1952 map = expand_xps_map(map, cpu, index);
1956 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1960 goto out_no_new_maps;
1962 for_each_possible_cpu(cpu) {
1963 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
1964 /* add queue to CPU maps */
1967 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1968 while ((pos < map->len) && (map->queues[pos] != index))
1971 if (pos == map->len)
1972 map->queues[map->len++] = index;
1974 if (numa_node_id == -2)
1975 numa_node_id = cpu_to_node(cpu);
1976 else if (numa_node_id != cpu_to_node(cpu))
1979 } else if (dev_maps) {
1980 /* fill in the new device map from the old device map */
1981 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1982 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1987 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
1989 /* Cleanup old maps */
1991 for_each_possible_cpu(cpu) {
1992 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1993 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1994 if (map && map != new_map)
1995 kfree_rcu(map, rcu);
1998 kfree_rcu(dev_maps, rcu);
2001 dev_maps = new_dev_maps;
2005 /* update Tx queue numa node */
2006 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2007 (numa_node_id >= 0) ? numa_node_id :
2013 /* removes queue from unused CPUs */
2014 for_each_possible_cpu(cpu) {
2015 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2018 if (remove_xps_queue(dev_maps, cpu, index))
2022 /* free map if not active */
2024 RCU_INIT_POINTER(dev->xps_maps, NULL);
2025 kfree_rcu(dev_maps, rcu);
2029 mutex_unlock(&xps_map_mutex);
2033 /* remove any maps that we added */
2034 for_each_possible_cpu(cpu) {
2035 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2036 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2038 if (new_map && new_map != map)
2042 mutex_unlock(&xps_map_mutex);
2044 kfree(new_dev_maps);
2047 EXPORT_SYMBOL(netif_set_xps_queue);
2051 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2052 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2054 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2058 if (txq < 1 || txq > dev->num_tx_queues)
2061 if (dev->reg_state == NETREG_REGISTERED ||
2062 dev->reg_state == NETREG_UNREGISTERING) {
2065 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2071 netif_setup_tc(dev, txq);
2073 if (txq < dev->real_num_tx_queues) {
2074 qdisc_reset_all_tx_gt(dev, txq);
2076 netif_reset_xps_queues_gt(dev, txq);
2081 dev->real_num_tx_queues = txq;
2084 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2088 * netif_set_real_num_rx_queues - set actual number of RX queues used
2089 * @dev: Network device
2090 * @rxq: Actual number of RX queues
2092 * This must be called either with the rtnl_lock held or before
2093 * registration of the net device. Returns 0 on success, or a
2094 * negative error code. If called before registration, it always
2097 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2101 if (rxq < 1 || rxq > dev->num_rx_queues)
2104 if (dev->reg_state == NETREG_REGISTERED) {
2107 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2113 dev->real_num_rx_queues = rxq;
2116 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2120 * netif_get_num_default_rss_queues - default number of RSS queues
2122 * This routine should set an upper limit on the number of RSS queues
2123 * used by default by multiqueue devices.
2125 int netif_get_num_default_rss_queues(void)
2127 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2129 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2131 static inline void __netif_reschedule(struct Qdisc *q)
2133 struct softnet_data *sd;
2134 unsigned long flags;
2136 local_irq_save(flags);
2137 sd = &__get_cpu_var(softnet_data);
2138 q->next_sched = NULL;
2139 *sd->output_queue_tailp = q;
2140 sd->output_queue_tailp = &q->next_sched;
2141 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2142 local_irq_restore(flags);
2145 void __netif_schedule(struct Qdisc *q)
2147 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2148 __netif_reschedule(q);
2150 EXPORT_SYMBOL(__netif_schedule);
2152 struct dev_kfree_skb_cb {
2153 enum skb_free_reason reason;
2156 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2158 return (struct dev_kfree_skb_cb *)skb->cb;
2161 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2163 unsigned long flags;
2165 if (likely(atomic_read(&skb->users) == 1)) {
2167 atomic_set(&skb->users, 0);
2168 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2171 get_kfree_skb_cb(skb)->reason = reason;
2172 local_irq_save(flags);
2173 skb->next = __this_cpu_read(softnet_data.completion_queue);
2174 __this_cpu_write(softnet_data.completion_queue, skb);
2175 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2176 local_irq_restore(flags);
2178 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2180 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2182 if (in_irq() || irqs_disabled())
2183 __dev_kfree_skb_irq(skb, reason);
2187 EXPORT_SYMBOL(__dev_kfree_skb_any);
2191 * netif_device_detach - mark device as removed
2192 * @dev: network device
2194 * Mark device as removed from system and therefore no longer available.
2196 void netif_device_detach(struct net_device *dev)
2198 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2199 netif_running(dev)) {
2200 netif_tx_stop_all_queues(dev);
2203 EXPORT_SYMBOL(netif_device_detach);
2206 * netif_device_attach - mark device as attached
2207 * @dev: network device
2209 * Mark device as attached from system and restart if needed.
2211 void netif_device_attach(struct net_device *dev)
2213 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2214 netif_running(dev)) {
2215 netif_tx_wake_all_queues(dev);
2216 __netdev_watchdog_up(dev);
2219 EXPORT_SYMBOL(netif_device_attach);
2221 static void skb_warn_bad_offload(const struct sk_buff *skb)
2223 static const netdev_features_t null_features = 0;
2224 struct net_device *dev = skb->dev;
2225 const char *driver = "";
2227 if (!net_ratelimit())
2230 if (dev && dev->dev.parent)
2231 driver = dev_driver_string(dev->dev.parent);
2233 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2234 "gso_type=%d ip_summed=%d\n",
2235 driver, dev ? &dev->features : &null_features,
2236 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2237 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2238 skb_shinfo(skb)->gso_type, skb->ip_summed);
2242 * Invalidate hardware checksum when packet is to be mangled, and
2243 * complete checksum manually on outgoing path.
2245 int skb_checksum_help(struct sk_buff *skb)
2248 int ret = 0, offset;
2250 if (skb->ip_summed == CHECKSUM_COMPLETE)
2251 goto out_set_summed;
2253 if (unlikely(skb_shinfo(skb)->gso_size)) {
2254 skb_warn_bad_offload(skb);
2258 /* Before computing a checksum, we should make sure no frag could
2259 * be modified by an external entity : checksum could be wrong.
2261 if (skb_has_shared_frag(skb)) {
2262 ret = __skb_linearize(skb);
2267 offset = skb_checksum_start_offset(skb);
2268 BUG_ON(offset >= skb_headlen(skb));
2269 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2271 offset += skb->csum_offset;
2272 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2274 if (skb_cloned(skb) &&
2275 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2276 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2281 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2283 skb->ip_summed = CHECKSUM_NONE;
2287 EXPORT_SYMBOL(skb_checksum_help);
2289 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2291 __be16 type = skb->protocol;
2292 int vlan_depth = ETH_HLEN;
2294 /* Tunnel gso handlers can set protocol to ethernet. */
2295 if (type == htons(ETH_P_TEB)) {
2298 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2301 eth = (struct ethhdr *)skb_mac_header(skb);
2302 type = eth->h_proto;
2305 while (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) {
2306 struct vlan_hdr *vh;
2308 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
2311 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
2312 type = vh->h_vlan_encapsulated_proto;
2313 vlan_depth += VLAN_HLEN;
2316 *depth = vlan_depth;
2322 * skb_mac_gso_segment - mac layer segmentation handler.
2323 * @skb: buffer to segment
2324 * @features: features for the output path (see dev->features)
2326 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2327 netdev_features_t features)
2329 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2330 struct packet_offload *ptype;
2331 int vlan_depth = skb->mac_len;
2332 __be16 type = skb_network_protocol(skb, &vlan_depth);
2334 if (unlikely(!type))
2335 return ERR_PTR(-EINVAL);
2337 __skb_pull(skb, vlan_depth);
2340 list_for_each_entry_rcu(ptype, &offload_base, list) {
2341 if (ptype->type == type && ptype->callbacks.gso_segment) {
2342 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2345 err = ptype->callbacks.gso_send_check(skb);
2346 segs = ERR_PTR(err);
2347 if (err || skb_gso_ok(skb, features))
2349 __skb_push(skb, (skb->data -
2350 skb_network_header(skb)));
2352 segs = ptype->callbacks.gso_segment(skb, features);
2358 __skb_push(skb, skb->data - skb_mac_header(skb));
2362 EXPORT_SYMBOL(skb_mac_gso_segment);
2365 /* openvswitch calls this on rx path, so we need a different check.
2367 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2370 return skb->ip_summed != CHECKSUM_PARTIAL;
2372 return skb->ip_summed == CHECKSUM_NONE;
2376 * __skb_gso_segment - Perform segmentation on skb.
2377 * @skb: buffer to segment
2378 * @features: features for the output path (see dev->features)
2379 * @tx_path: whether it is called in TX path
2381 * This function segments the given skb and returns a list of segments.
2383 * It may return NULL if the skb requires no segmentation. This is
2384 * only possible when GSO is used for verifying header integrity.
2386 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2387 netdev_features_t features, bool tx_path)
2389 if (unlikely(skb_needs_check(skb, tx_path))) {
2392 skb_warn_bad_offload(skb);
2394 if (skb_header_cloned(skb) &&
2395 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
2396 return ERR_PTR(err);
2399 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2400 SKB_GSO_CB(skb)->encap_level = 0;
2402 skb_reset_mac_header(skb);
2403 skb_reset_mac_len(skb);
2405 return skb_mac_gso_segment(skb, features);
2407 EXPORT_SYMBOL(__skb_gso_segment);
2409 /* Take action when hardware reception checksum errors are detected. */
2411 void netdev_rx_csum_fault(struct net_device *dev)
2413 if (net_ratelimit()) {
2414 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2418 EXPORT_SYMBOL(netdev_rx_csum_fault);
2421 /* Actually, we should eliminate this check as soon as we know, that:
2422 * 1. IOMMU is present and allows to map all the memory.
2423 * 2. No high memory really exists on this machine.
2426 static int illegal_highdma(const struct net_device *dev, struct sk_buff *skb)
2428 #ifdef CONFIG_HIGHMEM
2430 if (!(dev->features & NETIF_F_HIGHDMA)) {
2431 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2432 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2433 if (PageHighMem(skb_frag_page(frag)))
2438 if (PCI_DMA_BUS_IS_PHYS) {
2439 struct device *pdev = dev->dev.parent;
2443 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2444 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2445 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2446 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2455 void (*destructor)(struct sk_buff *skb);
2458 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2460 static void dev_gso_skb_destructor(struct sk_buff *skb)
2462 struct dev_gso_cb *cb;
2464 kfree_skb_list(skb->next);
2467 cb = DEV_GSO_CB(skb);
2469 cb->destructor(skb);
2473 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2474 * @skb: buffer to segment
2475 * @features: device features as applicable to this skb
2477 * This function segments the given skb and stores the list of segments
2480 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2482 struct sk_buff *segs;
2484 segs = skb_gso_segment(skb, features);
2486 /* Verifying header integrity only. */
2491 return PTR_ERR(segs);
2494 DEV_GSO_CB(skb)->destructor = skb->destructor;
2495 skb->destructor = dev_gso_skb_destructor;
2500 static netdev_features_t harmonize_features(struct sk_buff *skb,
2501 const struct net_device *dev,
2502 netdev_features_t features)
2506 if (skb->ip_summed != CHECKSUM_NONE &&
2507 !can_checksum_protocol(features, skb_network_protocol(skb, &tmp))) {
2508 features &= ~NETIF_F_ALL_CSUM;
2509 } else if (illegal_highdma(dev, skb)) {
2510 features &= ~NETIF_F_SG;
2516 netdev_features_t netif_skb_dev_features(struct sk_buff *skb,
2517 const struct net_device *dev)
2519 __be16 protocol = skb->protocol;
2520 netdev_features_t features = dev->features;
2522 if (skb_shinfo(skb)->gso_segs > dev->gso_max_segs)
2523 features &= ~NETIF_F_GSO_MASK;
2525 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) {
2526 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2527 protocol = veh->h_vlan_encapsulated_proto;
2528 } else if (!vlan_tx_tag_present(skb)) {
2529 return harmonize_features(skb, dev, features);
2532 features &= (dev->vlan_features | NETIF_F_HW_VLAN_CTAG_TX |
2533 NETIF_F_HW_VLAN_STAG_TX);
2535 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD))
2536 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2537 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_CTAG_TX |
2538 NETIF_F_HW_VLAN_STAG_TX;
2540 return harmonize_features(skb, dev, features);
2542 EXPORT_SYMBOL(netif_skb_dev_features);
2544 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2545 struct netdev_queue *txq)
2547 const struct net_device_ops *ops = dev->netdev_ops;
2548 int rc = NETDEV_TX_OK;
2549 unsigned int skb_len;
2551 if (likely(!skb->next)) {
2552 netdev_features_t features;
2555 * If device doesn't need skb->dst, release it right now while
2556 * its hot in this cpu cache
2558 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2561 features = netif_skb_features(skb);
2563 if (vlan_tx_tag_present(skb) &&
2564 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
2565 skb = __vlan_put_tag(skb, skb->vlan_proto,
2566 vlan_tx_tag_get(skb));
2573 /* If encapsulation offload request, verify we are testing
2574 * hardware encapsulation features instead of standard
2575 * features for the netdev
2577 if (skb->encapsulation)
2578 features &= dev->hw_enc_features;
2580 if (netif_needs_gso(skb, features)) {
2581 if (unlikely(dev_gso_segment(skb, features)))
2586 if (skb_needs_linearize(skb, features) &&
2587 __skb_linearize(skb))
2590 /* If packet is not checksummed and device does not
2591 * support checksumming for this protocol, complete
2592 * checksumming here.
2594 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2595 if (skb->encapsulation)
2596 skb_set_inner_transport_header(skb,
2597 skb_checksum_start_offset(skb));
2599 skb_set_transport_header(skb,
2600 skb_checksum_start_offset(skb));
2601 if (!(features & NETIF_F_ALL_CSUM) &&
2602 skb_checksum_help(skb))
2607 if (!list_empty(&ptype_all))
2608 dev_queue_xmit_nit(skb, dev);
2611 trace_net_dev_start_xmit(skb, dev);
2612 rc = ops->ndo_start_xmit(skb, dev);
2613 trace_net_dev_xmit(skb, rc, dev, skb_len);
2614 if (rc == NETDEV_TX_OK)
2615 txq_trans_update(txq);
2621 struct sk_buff *nskb = skb->next;
2623 skb->next = nskb->next;
2626 if (!list_empty(&ptype_all))
2627 dev_queue_xmit_nit(nskb, dev);
2629 skb_len = nskb->len;
2630 trace_net_dev_start_xmit(nskb, dev);
2631 rc = ops->ndo_start_xmit(nskb, dev);
2632 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2633 if (unlikely(rc != NETDEV_TX_OK)) {
2634 if (rc & ~NETDEV_TX_MASK)
2635 goto out_kfree_gso_skb;
2636 nskb->next = skb->next;
2640 txq_trans_update(txq);
2641 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2642 return NETDEV_TX_BUSY;
2643 } while (skb->next);
2646 if (likely(skb->next == NULL)) {
2647 skb->destructor = DEV_GSO_CB(skb)->destructor;
2656 EXPORT_SYMBOL_GPL(dev_hard_start_xmit);
2658 static void qdisc_pkt_len_init(struct sk_buff *skb)
2660 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2662 qdisc_skb_cb(skb)->pkt_len = skb->len;
2664 /* To get more precise estimation of bytes sent on wire,
2665 * we add to pkt_len the headers size of all segments
2667 if (shinfo->gso_size) {
2668 unsigned int hdr_len;
2669 u16 gso_segs = shinfo->gso_segs;
2671 /* mac layer + network layer */
2672 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2674 /* + transport layer */
2675 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2676 hdr_len += tcp_hdrlen(skb);
2678 hdr_len += sizeof(struct udphdr);
2680 if (shinfo->gso_type & SKB_GSO_DODGY)
2681 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2684 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2688 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2689 struct net_device *dev,
2690 struct netdev_queue *txq)
2692 spinlock_t *root_lock = qdisc_lock(q);
2696 qdisc_pkt_len_init(skb);
2697 qdisc_calculate_pkt_len(skb, q);
2699 * Heuristic to force contended enqueues to serialize on a
2700 * separate lock before trying to get qdisc main lock.
2701 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2702 * and dequeue packets faster.
2704 contended = qdisc_is_running(q);
2705 if (unlikely(contended))
2706 spin_lock(&q->busylock);
2708 spin_lock(root_lock);
2709 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2712 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2713 qdisc_run_begin(q)) {
2715 * This is a work-conserving queue; there are no old skbs
2716 * waiting to be sent out; and the qdisc is not running -
2717 * xmit the skb directly.
2719 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2722 qdisc_bstats_update(q, skb);
2724 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2725 if (unlikely(contended)) {
2726 spin_unlock(&q->busylock);
2733 rc = NET_XMIT_SUCCESS;
2736 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2737 if (qdisc_run_begin(q)) {
2738 if (unlikely(contended)) {
2739 spin_unlock(&q->busylock);
2745 spin_unlock(root_lock);
2746 if (unlikely(contended))
2747 spin_unlock(&q->busylock);
2751 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2752 static void skb_update_prio(struct sk_buff *skb)
2754 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2756 if (!skb->priority && skb->sk && map) {
2757 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2759 if (prioidx < map->priomap_len)
2760 skb->priority = map->priomap[prioidx];
2764 #define skb_update_prio(skb)
2767 static DEFINE_PER_CPU(int, xmit_recursion);
2768 #define RECURSION_LIMIT 10
2771 * dev_loopback_xmit - loop back @skb
2772 * @skb: buffer to transmit
2774 int dev_loopback_xmit(struct sk_buff *skb)
2776 skb_reset_mac_header(skb);
2777 __skb_pull(skb, skb_network_offset(skb));
2778 skb->pkt_type = PACKET_LOOPBACK;
2779 skb->ip_summed = CHECKSUM_UNNECESSARY;
2780 WARN_ON(!skb_dst(skb));
2785 EXPORT_SYMBOL(dev_loopback_xmit);
2788 * __dev_queue_xmit - transmit a buffer
2789 * @skb: buffer to transmit
2790 * @accel_priv: private data used for L2 forwarding offload
2792 * Queue a buffer for transmission to a network device. The caller must
2793 * have set the device and priority and built the buffer before calling
2794 * this function. The function can be called from an interrupt.
2796 * A negative errno code is returned on a failure. A success does not
2797 * guarantee the frame will be transmitted as it may be dropped due
2798 * to congestion or traffic shaping.
2800 * -----------------------------------------------------------------------------------
2801 * I notice this method can also return errors from the queue disciplines,
2802 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2805 * Regardless of the return value, the skb is consumed, so it is currently
2806 * difficult to retry a send to this method. (You can bump the ref count
2807 * before sending to hold a reference for retry if you are careful.)
2809 * When calling this method, interrupts MUST be enabled. This is because
2810 * the BH enable code must have IRQs enabled so that it will not deadlock.
2813 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
2815 struct net_device *dev = skb->dev;
2816 struct netdev_queue *txq;
2820 skb_reset_mac_header(skb);
2822 /* Disable soft irqs for various locks below. Also
2823 * stops preemption for RCU.
2827 skb_update_prio(skb);
2829 txq = netdev_pick_tx(dev, skb, accel_priv);
2830 q = rcu_dereference_bh(txq->qdisc);
2832 #ifdef CONFIG_NET_CLS_ACT
2833 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2835 trace_net_dev_queue(skb);
2837 rc = __dev_xmit_skb(skb, q, dev, txq);
2841 /* The device has no queue. Common case for software devices:
2842 loopback, all the sorts of tunnels...
2844 Really, it is unlikely that netif_tx_lock protection is necessary
2845 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2847 However, it is possible, that they rely on protection
2850 Check this and shot the lock. It is not prone from deadlocks.
2851 Either shot noqueue qdisc, it is even simpler 8)
2853 if (dev->flags & IFF_UP) {
2854 int cpu = smp_processor_id(); /* ok because BHs are off */
2856 if (txq->xmit_lock_owner != cpu) {
2858 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2859 goto recursion_alert;
2861 HARD_TX_LOCK(dev, txq, cpu);
2863 if (!netif_xmit_stopped(txq)) {
2864 __this_cpu_inc(xmit_recursion);
2865 rc = dev_hard_start_xmit(skb, dev, txq);
2866 __this_cpu_dec(xmit_recursion);
2867 if (dev_xmit_complete(rc)) {
2868 HARD_TX_UNLOCK(dev, txq);
2872 HARD_TX_UNLOCK(dev, txq);
2873 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2876 /* Recursion is detected! It is possible,
2880 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2886 rcu_read_unlock_bh();
2891 rcu_read_unlock_bh();
2895 int dev_queue_xmit(struct sk_buff *skb)
2897 return __dev_queue_xmit(skb, NULL);
2899 EXPORT_SYMBOL(dev_queue_xmit);
2901 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
2903 return __dev_queue_xmit(skb, accel_priv);
2905 EXPORT_SYMBOL(dev_queue_xmit_accel);
2908 /*=======================================================================
2910 =======================================================================*/
2912 int netdev_max_backlog __read_mostly = 1000;
2913 EXPORT_SYMBOL(netdev_max_backlog);
2915 int netdev_tstamp_prequeue __read_mostly = 1;
2916 int netdev_budget __read_mostly = 300;
2917 int weight_p __read_mostly = 64; /* old backlog weight */
2919 /* Called with irq disabled */
2920 static inline void ____napi_schedule(struct softnet_data *sd,
2921 struct napi_struct *napi)
2923 list_add_tail(&napi->poll_list, &sd->poll_list);
2924 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2929 /* One global table that all flow-based protocols share. */
2930 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2931 EXPORT_SYMBOL(rps_sock_flow_table);
2933 struct static_key rps_needed __read_mostly;
2935 static struct rps_dev_flow *
2936 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2937 struct rps_dev_flow *rflow, u16 next_cpu)
2939 if (next_cpu != RPS_NO_CPU) {
2940 #ifdef CONFIG_RFS_ACCEL
2941 struct netdev_rx_queue *rxqueue;
2942 struct rps_dev_flow_table *flow_table;
2943 struct rps_dev_flow *old_rflow;
2948 /* Should we steer this flow to a different hardware queue? */
2949 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2950 !(dev->features & NETIF_F_NTUPLE))
2952 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2953 if (rxq_index == skb_get_rx_queue(skb))
2956 rxqueue = dev->_rx + rxq_index;
2957 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2960 flow_id = skb->rxhash & flow_table->mask;
2961 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2962 rxq_index, flow_id);
2966 rflow = &flow_table->flows[flow_id];
2968 if (old_rflow->filter == rflow->filter)
2969 old_rflow->filter = RPS_NO_FILTER;
2973 per_cpu(softnet_data, next_cpu).input_queue_head;
2976 rflow->cpu = next_cpu;
2981 * get_rps_cpu is called from netif_receive_skb and returns the target
2982 * CPU from the RPS map of the receiving queue for a given skb.
2983 * rcu_read_lock must be held on entry.
2985 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2986 struct rps_dev_flow **rflowp)
2988 struct netdev_rx_queue *rxqueue;
2989 struct rps_map *map;
2990 struct rps_dev_flow_table *flow_table;
2991 struct rps_sock_flow_table *sock_flow_table;
2995 if (skb_rx_queue_recorded(skb)) {
2996 u16 index = skb_get_rx_queue(skb);
2997 if (unlikely(index >= dev->real_num_rx_queues)) {
2998 WARN_ONCE(dev->real_num_rx_queues > 1,
2999 "%s received packet on queue %u, but number "
3000 "of RX queues is %u\n",
3001 dev->name, index, dev->real_num_rx_queues);
3004 rxqueue = dev->_rx + index;
3008 map = rcu_dereference(rxqueue->rps_map);
3010 if (map->len == 1 &&
3011 !rcu_access_pointer(rxqueue->rps_flow_table)) {
3012 tcpu = map->cpus[0];
3013 if (cpu_online(tcpu))
3017 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
3021 skb_reset_network_header(skb);
3022 if (!skb_get_hash(skb))
3025 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3026 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3027 if (flow_table && sock_flow_table) {
3029 struct rps_dev_flow *rflow;
3031 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
3034 next_cpu = sock_flow_table->ents[skb->rxhash &
3035 sock_flow_table->mask];
3038 * If the desired CPU (where last recvmsg was done) is
3039 * different from current CPU (one in the rx-queue flow
3040 * table entry), switch if one of the following holds:
3041 * - Current CPU is unset (equal to RPS_NO_CPU).
3042 * - Current CPU is offline.
3043 * - The current CPU's queue tail has advanced beyond the
3044 * last packet that was enqueued using this table entry.
3045 * This guarantees that all previous packets for the flow
3046 * have been dequeued, thus preserving in order delivery.
3048 if (unlikely(tcpu != next_cpu) &&
3049 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
3050 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3051 rflow->last_qtail)) >= 0)) {
3053 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3056 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
3064 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
3066 if (cpu_online(tcpu)) {
3076 #ifdef CONFIG_RFS_ACCEL
3079 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3080 * @dev: Device on which the filter was set
3081 * @rxq_index: RX queue index
3082 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3083 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3085 * Drivers that implement ndo_rx_flow_steer() should periodically call
3086 * this function for each installed filter and remove the filters for
3087 * which it returns %true.
3089 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3090 u32 flow_id, u16 filter_id)
3092 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3093 struct rps_dev_flow_table *flow_table;
3094 struct rps_dev_flow *rflow;
3099 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3100 if (flow_table && flow_id <= flow_table->mask) {
3101 rflow = &flow_table->flows[flow_id];
3102 cpu = ACCESS_ONCE(rflow->cpu);
3103 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3104 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3105 rflow->last_qtail) <
3106 (int)(10 * flow_table->mask)))
3112 EXPORT_SYMBOL(rps_may_expire_flow);
3114 #endif /* CONFIG_RFS_ACCEL */
3116 /* Called from hardirq (IPI) context */
3117 static void rps_trigger_softirq(void *data)
3119 struct softnet_data *sd = data;
3121 ____napi_schedule(sd, &sd->backlog);
3125 #endif /* CONFIG_RPS */
3128 * Check if this softnet_data structure is another cpu one
3129 * If yes, queue it to our IPI list and return 1
3132 static int rps_ipi_queued(struct softnet_data *sd)
3135 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
3138 sd->rps_ipi_next = mysd->rps_ipi_list;
3139 mysd->rps_ipi_list = sd;
3141 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3144 #endif /* CONFIG_RPS */
3148 #ifdef CONFIG_NET_FLOW_LIMIT
3149 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3152 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3154 #ifdef CONFIG_NET_FLOW_LIMIT
3155 struct sd_flow_limit *fl;
3156 struct softnet_data *sd;
3157 unsigned int old_flow, new_flow;
3159 if (qlen < (netdev_max_backlog >> 1))
3162 sd = &__get_cpu_var(softnet_data);
3165 fl = rcu_dereference(sd->flow_limit);
3167 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3168 old_flow = fl->history[fl->history_head];
3169 fl->history[fl->history_head] = new_flow;
3172 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3174 if (likely(fl->buckets[old_flow]))
3175 fl->buckets[old_flow]--;
3177 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3189 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3190 * queue (may be a remote CPU queue).
3192 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3193 unsigned int *qtail)
3195 struct softnet_data *sd;
3196 unsigned long flags;
3199 sd = &per_cpu(softnet_data, cpu);
3201 local_irq_save(flags);
3204 qlen = skb_queue_len(&sd->input_pkt_queue);
3205 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3206 if (skb_queue_len(&sd->input_pkt_queue)) {
3208 __skb_queue_tail(&sd->input_pkt_queue, skb);
3209 input_queue_tail_incr_save(sd, qtail);
3211 local_irq_restore(flags);
3212 return NET_RX_SUCCESS;
3215 /* Schedule NAPI for backlog device
3216 * We can use non atomic operation since we own the queue lock
3218 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3219 if (!rps_ipi_queued(sd))
3220 ____napi_schedule(sd, &sd->backlog);
3228 local_irq_restore(flags);
3230 atomic_long_inc(&skb->dev->rx_dropped);
3235 static int netif_rx_internal(struct sk_buff *skb)
3239 /* if netpoll wants it, pretend we never saw it */
3240 if (netpoll_rx(skb))
3243 net_timestamp_check(netdev_tstamp_prequeue, skb);
3245 trace_netif_rx(skb);
3247 if (static_key_false(&rps_needed)) {
3248 struct rps_dev_flow voidflow, *rflow = &voidflow;
3254 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3256 cpu = smp_processor_id();
3258 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3266 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3273 * netif_rx - post buffer to the network code
3274 * @skb: buffer to post
3276 * This function receives a packet from a device driver and queues it for
3277 * the upper (protocol) levels to process. It always succeeds. The buffer
3278 * may be dropped during processing for congestion control or by the
3282 * NET_RX_SUCCESS (no congestion)
3283 * NET_RX_DROP (packet was dropped)
3287 int netif_rx(struct sk_buff *skb)
3289 trace_netif_rx_entry(skb);
3291 return netif_rx_internal(skb);
3293 EXPORT_SYMBOL(netif_rx);
3295 int netif_rx_ni(struct sk_buff *skb)
3299 trace_netif_rx_ni_entry(skb);
3302 err = netif_rx_internal(skb);
3303 if (local_softirq_pending())
3309 EXPORT_SYMBOL(netif_rx_ni);
3311 static void net_tx_action(struct softirq_action *h)
3313 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3315 if (sd->completion_queue) {
3316 struct sk_buff *clist;
3318 local_irq_disable();
3319 clist = sd->completion_queue;
3320 sd->completion_queue = NULL;
3324 struct sk_buff *skb = clist;
3325 clist = clist->next;
3327 WARN_ON(atomic_read(&skb->users));
3328 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3329 trace_consume_skb(skb);
3331 trace_kfree_skb(skb, net_tx_action);
3336 if (sd->output_queue) {
3339 local_irq_disable();
3340 head = sd->output_queue;
3341 sd->output_queue = NULL;
3342 sd->output_queue_tailp = &sd->output_queue;
3346 struct Qdisc *q = head;
3347 spinlock_t *root_lock;
3349 head = head->next_sched;
3351 root_lock = qdisc_lock(q);
3352 if (spin_trylock(root_lock)) {
3353 smp_mb__before_clear_bit();
3354 clear_bit(__QDISC_STATE_SCHED,
3357 spin_unlock(root_lock);
3359 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3361 __netif_reschedule(q);
3363 smp_mb__before_clear_bit();
3364 clear_bit(__QDISC_STATE_SCHED,
3372 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3373 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3374 /* This hook is defined here for ATM LANE */
3375 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3376 unsigned char *addr) __read_mostly;
3377 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3380 #ifdef CONFIG_NET_CLS_ACT
3381 /* TODO: Maybe we should just force sch_ingress to be compiled in
3382 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3383 * a compare and 2 stores extra right now if we dont have it on
3384 * but have CONFIG_NET_CLS_ACT
3385 * NOTE: This doesn't stop any functionality; if you dont have
3386 * the ingress scheduler, you just can't add policies on ingress.
3389 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3391 struct net_device *dev = skb->dev;
3392 u32 ttl = G_TC_RTTL(skb->tc_verd);
3393 int result = TC_ACT_OK;
3396 if (unlikely(MAX_RED_LOOP < ttl++)) {
3397 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3398 skb->skb_iif, dev->ifindex);
3402 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3403 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3406 if (q != &noop_qdisc) {
3407 spin_lock(qdisc_lock(q));
3408 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3409 result = qdisc_enqueue_root(skb, q);
3410 spin_unlock(qdisc_lock(q));
3416 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3417 struct packet_type **pt_prev,
3418 int *ret, struct net_device *orig_dev)
3420 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3422 if (!rxq || rxq->qdisc == &noop_qdisc)
3426 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3430 switch (ing_filter(skb, rxq)) {
3444 * netdev_rx_handler_register - register receive handler
3445 * @dev: device to register a handler for
3446 * @rx_handler: receive handler to register
3447 * @rx_handler_data: data pointer that is used by rx handler
3449 * Register a receive hander for a device. This handler will then be
3450 * called from __netif_receive_skb. A negative errno code is returned
3453 * The caller must hold the rtnl_mutex.
3455 * For a general description of rx_handler, see enum rx_handler_result.
3457 int netdev_rx_handler_register(struct net_device *dev,
3458 rx_handler_func_t *rx_handler,
3459 void *rx_handler_data)
3463 if (dev->rx_handler)
3466 /* Note: rx_handler_data must be set before rx_handler */
3467 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3468 rcu_assign_pointer(dev->rx_handler, rx_handler);
3472 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3475 * netdev_rx_handler_unregister - unregister receive handler
3476 * @dev: device to unregister a handler from
3478 * Unregister a receive handler from a device.
3480 * The caller must hold the rtnl_mutex.
3482 void netdev_rx_handler_unregister(struct net_device *dev)
3486 RCU_INIT_POINTER(dev->rx_handler, NULL);
3487 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3488 * section has a guarantee to see a non NULL rx_handler_data
3492 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3494 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3497 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3498 * the special handling of PFMEMALLOC skbs.
3500 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3502 switch (skb->protocol) {
3503 case __constant_htons(ETH_P_ARP):
3504 case __constant_htons(ETH_P_IP):
3505 case __constant_htons(ETH_P_IPV6):
3506 case __constant_htons(ETH_P_8021Q):
3507 case __constant_htons(ETH_P_8021AD):
3514 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3516 struct packet_type *ptype, *pt_prev;
3517 rx_handler_func_t *rx_handler;
3518 struct net_device *orig_dev;
3519 struct net_device *null_or_dev;
3520 bool deliver_exact = false;
3521 int ret = NET_RX_DROP;
3524 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3526 trace_netif_receive_skb(skb);
3528 /* if we've gotten here through NAPI, check netpoll */
3529 if (netpoll_receive_skb(skb))
3532 orig_dev = skb->dev;
3534 skb_reset_network_header(skb);
3535 if (!skb_transport_header_was_set(skb))
3536 skb_reset_transport_header(skb);
3537 skb_reset_mac_len(skb);
3544 skb->skb_iif = skb->dev->ifindex;
3546 __this_cpu_inc(softnet_data.processed);
3548 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3549 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3550 skb = vlan_untag(skb);
3555 #ifdef CONFIG_NET_CLS_ACT
3556 if (skb->tc_verd & TC_NCLS) {
3557 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3565 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3566 if (!ptype->dev || ptype->dev == skb->dev) {
3568 ret = deliver_skb(skb, pt_prev, orig_dev);
3574 #ifdef CONFIG_NET_CLS_ACT
3575 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3581 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3584 if (vlan_tx_tag_present(skb)) {
3586 ret = deliver_skb(skb, pt_prev, orig_dev);
3589 if (vlan_do_receive(&skb))
3591 else if (unlikely(!skb))
3595 rx_handler = rcu_dereference(skb->dev->rx_handler);
3598 ret = deliver_skb(skb, pt_prev, orig_dev);
3601 switch (rx_handler(&skb)) {
3602 case RX_HANDLER_CONSUMED:
3603 ret = NET_RX_SUCCESS;
3605 case RX_HANDLER_ANOTHER:
3607 case RX_HANDLER_EXACT:
3608 deliver_exact = true;
3609 case RX_HANDLER_PASS:
3616 if (unlikely(vlan_tx_tag_present(skb))) {
3617 if (vlan_tx_tag_get_id(skb))
3618 skb->pkt_type = PACKET_OTHERHOST;
3619 /* Note: we might in the future use prio bits
3620 * and set skb->priority like in vlan_do_receive()
3621 * For the time being, just ignore Priority Code Point
3626 /* deliver only exact match when indicated */
3627 null_or_dev = deliver_exact ? skb->dev : NULL;
3629 type = skb->protocol;
3630 list_for_each_entry_rcu(ptype,
3631 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3632 if (ptype->type == type &&
3633 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3634 ptype->dev == orig_dev)) {
3636 ret = deliver_skb(skb, pt_prev, orig_dev);
3642 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3645 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3648 atomic_long_inc(&skb->dev->rx_dropped);
3650 /* Jamal, now you will not able to escape explaining
3651 * me how you were going to use this. :-)
3662 static int __netif_receive_skb(struct sk_buff *skb)
3666 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3667 unsigned long pflags = current->flags;
3670 * PFMEMALLOC skbs are special, they should
3671 * - be delivered to SOCK_MEMALLOC sockets only
3672 * - stay away from userspace
3673 * - have bounded memory usage
3675 * Use PF_MEMALLOC as this saves us from propagating the allocation
3676 * context down to all allocation sites.
3678 current->flags |= PF_MEMALLOC;
3679 ret = __netif_receive_skb_core(skb, true);
3680 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3682 ret = __netif_receive_skb_core(skb, false);
3687 static int netif_receive_skb_internal(struct sk_buff *skb)
3689 net_timestamp_check(netdev_tstamp_prequeue, skb);
3691 if (skb_defer_rx_timestamp(skb))
3692 return NET_RX_SUCCESS;
3695 if (static_key_false(&rps_needed)) {
3696 struct rps_dev_flow voidflow, *rflow = &voidflow;
3701 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3704 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3711 return __netif_receive_skb(skb);
3715 * netif_receive_skb - process receive buffer from network
3716 * @skb: buffer to process
3718 * netif_receive_skb() is the main receive data processing function.
3719 * It always succeeds. The buffer may be dropped during processing
3720 * for congestion control or by the protocol layers.
3722 * This function may only be called from softirq context and interrupts
3723 * should be enabled.
3725 * Return values (usually ignored):
3726 * NET_RX_SUCCESS: no congestion
3727 * NET_RX_DROP: packet was dropped
3729 int netif_receive_skb(struct sk_buff *skb)
3731 trace_netif_receive_skb_entry(skb);
3733 return netif_receive_skb_internal(skb);
3735 EXPORT_SYMBOL(netif_receive_skb);
3737 /* Network device is going away, flush any packets still pending
3738 * Called with irqs disabled.
3740 static void flush_backlog(void *arg)
3742 struct net_device *dev = arg;
3743 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3744 struct sk_buff *skb, *tmp;
3747 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3748 if (skb->dev == dev) {
3749 __skb_unlink(skb, &sd->input_pkt_queue);
3751 input_queue_head_incr(sd);
3756 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3757 if (skb->dev == dev) {
3758 __skb_unlink(skb, &sd->process_queue);
3760 input_queue_head_incr(sd);
3765 static int napi_gro_complete(struct sk_buff *skb)
3767 struct packet_offload *ptype;
3768 __be16 type = skb->protocol;
3769 struct list_head *head = &offload_base;
3772 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3774 if (NAPI_GRO_CB(skb)->count == 1) {
3775 skb_shinfo(skb)->gso_size = 0;
3780 list_for_each_entry_rcu(ptype, head, list) {
3781 if (ptype->type != type || !ptype->callbacks.gro_complete)
3784 err = ptype->callbacks.gro_complete(skb, 0);
3790 WARN_ON(&ptype->list == head);
3792 return NET_RX_SUCCESS;
3796 return netif_receive_skb_internal(skb);
3799 /* napi->gro_list contains packets ordered by age.
3800 * youngest packets at the head of it.
3801 * Complete skbs in reverse order to reduce latencies.
3803 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3805 struct sk_buff *skb, *prev = NULL;
3807 /* scan list and build reverse chain */
3808 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3813 for (skb = prev; skb; skb = prev) {
3816 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3820 napi_gro_complete(skb);
3824 napi->gro_list = NULL;
3826 EXPORT_SYMBOL(napi_gro_flush);
3828 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3831 unsigned int maclen = skb->dev->hard_header_len;
3832 u32 hash = skb_get_hash_raw(skb);
3834 for (p = napi->gro_list; p; p = p->next) {
3835 unsigned long diffs;
3837 NAPI_GRO_CB(p)->flush = 0;
3839 if (hash != skb_get_hash_raw(p)) {
3840 NAPI_GRO_CB(p)->same_flow = 0;
3844 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3845 diffs |= p->vlan_tci ^ skb->vlan_tci;
3846 if (maclen == ETH_HLEN)
3847 diffs |= compare_ether_header(skb_mac_header(p),
3848 skb_gro_mac_header(skb));
3850 diffs = memcmp(skb_mac_header(p),
3851 skb_gro_mac_header(skb),
3853 NAPI_GRO_CB(p)->same_flow = !diffs;
3857 static void skb_gro_reset_offset(struct sk_buff *skb)
3859 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3860 const skb_frag_t *frag0 = &pinfo->frags[0];
3862 NAPI_GRO_CB(skb)->data_offset = 0;
3863 NAPI_GRO_CB(skb)->frag0 = NULL;
3864 NAPI_GRO_CB(skb)->frag0_len = 0;
3866 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
3868 !PageHighMem(skb_frag_page(frag0))) {
3869 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3870 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3874 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3876 struct sk_buff **pp = NULL;
3877 struct packet_offload *ptype;
3878 __be16 type = skb->protocol;
3879 struct list_head *head = &offload_base;
3881 enum gro_result ret;
3883 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3886 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3889 skb_gro_reset_offset(skb);
3890 gro_list_prepare(napi, skb);
3891 NAPI_GRO_CB(skb)->csum = skb->csum; /* Needed for CHECKSUM_COMPLETE */
3894 list_for_each_entry_rcu(ptype, head, list) {
3895 if (ptype->type != type || !ptype->callbacks.gro_receive)
3898 skb_set_network_header(skb, skb_gro_offset(skb));
3899 skb_reset_mac_len(skb);
3900 NAPI_GRO_CB(skb)->same_flow = 0;
3901 NAPI_GRO_CB(skb)->flush = 0;
3902 NAPI_GRO_CB(skb)->free = 0;
3903 NAPI_GRO_CB(skb)->udp_mark = 0;
3905 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
3910 if (&ptype->list == head)
3913 same_flow = NAPI_GRO_CB(skb)->same_flow;
3914 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3917 struct sk_buff *nskb = *pp;
3921 napi_gro_complete(nskb);
3928 if (NAPI_GRO_CB(skb)->flush)
3931 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
3932 struct sk_buff *nskb = napi->gro_list;
3934 /* locate the end of the list to select the 'oldest' flow */
3935 while (nskb->next) {
3941 napi_gro_complete(nskb);
3945 NAPI_GRO_CB(skb)->count = 1;
3946 NAPI_GRO_CB(skb)->age = jiffies;
3947 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3948 skb->next = napi->gro_list;
3949 napi->gro_list = skb;
3953 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3954 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3956 BUG_ON(skb->end - skb->tail < grow);
3958 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3961 skb->data_len -= grow;
3963 skb_shinfo(skb)->frags[0].page_offset += grow;
3964 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3966 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3967 skb_frag_unref(skb, 0);
3968 memmove(skb_shinfo(skb)->frags,
3969 skb_shinfo(skb)->frags + 1,
3970 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3982 struct packet_offload *gro_find_receive_by_type(__be16 type)
3984 struct list_head *offload_head = &offload_base;
3985 struct packet_offload *ptype;
3987 list_for_each_entry_rcu(ptype, offload_head, list) {
3988 if (ptype->type != type || !ptype->callbacks.gro_receive)
3994 EXPORT_SYMBOL(gro_find_receive_by_type);
3996 struct packet_offload *gro_find_complete_by_type(__be16 type)
3998 struct list_head *offload_head = &offload_base;
3999 struct packet_offload *ptype;
4001 list_for_each_entry_rcu(ptype, offload_head, list) {
4002 if (ptype->type != type || !ptype->callbacks.gro_complete)
4008 EXPORT_SYMBOL(gro_find_complete_by_type);
4010 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4014 if (netif_receive_skb_internal(skb))
4022 case GRO_MERGED_FREE:
4023 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4024 kmem_cache_free(skbuff_head_cache, skb);
4037 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4039 trace_napi_gro_receive_entry(skb);
4041 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4043 EXPORT_SYMBOL(napi_gro_receive);
4045 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4047 __skb_pull(skb, skb_headlen(skb));
4048 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4049 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4051 skb->dev = napi->dev;
4057 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4059 struct sk_buff *skb = napi->skb;
4062 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
4067 EXPORT_SYMBOL(napi_get_frags);
4069 static gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
4074 if (netif_receive_skb_internal(skb))
4079 case GRO_MERGED_FREE:
4080 napi_reuse_skb(napi, skb);
4091 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4093 struct sk_buff *skb = napi->skb;
4097 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) {
4098 napi_reuse_skb(napi, skb);
4101 skb->protocol = eth_type_trans(skb, skb->dev);
4106 gro_result_t napi_gro_frags(struct napi_struct *napi)
4108 struct sk_buff *skb = napi_frags_skb(napi);
4113 trace_napi_gro_frags_entry(skb);
4115 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4117 EXPORT_SYMBOL(napi_gro_frags);
4120 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4121 * Note: called with local irq disabled, but exits with local irq enabled.
4123 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4126 struct softnet_data *remsd = sd->rps_ipi_list;
4129 sd->rps_ipi_list = NULL;
4133 /* Send pending IPI's to kick RPS processing on remote cpus. */
4135 struct softnet_data *next = remsd->rps_ipi_next;
4137 if (cpu_online(remsd->cpu))
4138 __smp_call_function_single(remsd->cpu,
4147 static int process_backlog(struct napi_struct *napi, int quota)
4150 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4153 /* Check if we have pending ipi, its better to send them now,
4154 * not waiting net_rx_action() end.
4156 if (sd->rps_ipi_list) {
4157 local_irq_disable();
4158 net_rps_action_and_irq_enable(sd);
4161 napi->weight = weight_p;
4162 local_irq_disable();
4163 while (work < quota) {
4164 struct sk_buff *skb;
4167 while ((skb = __skb_dequeue(&sd->process_queue))) {
4169 __netif_receive_skb(skb);
4170 local_irq_disable();
4171 input_queue_head_incr(sd);
4172 if (++work >= quota) {
4179 qlen = skb_queue_len(&sd->input_pkt_queue);
4181 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4182 &sd->process_queue);
4184 if (qlen < quota - work) {
4186 * Inline a custom version of __napi_complete().
4187 * only current cpu owns and manipulates this napi,
4188 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
4189 * we can use a plain write instead of clear_bit(),
4190 * and we dont need an smp_mb() memory barrier.
4192 list_del(&napi->poll_list);
4195 quota = work + qlen;
4205 * __napi_schedule - schedule for receive
4206 * @n: entry to schedule
4208 * The entry's receive function will be scheduled to run
4210 void __napi_schedule(struct napi_struct *n)
4212 unsigned long flags;
4214 local_irq_save(flags);
4215 ____napi_schedule(&__get_cpu_var(softnet_data), n);
4216 local_irq_restore(flags);
4218 EXPORT_SYMBOL(__napi_schedule);
4220 void __napi_complete(struct napi_struct *n)
4222 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4223 BUG_ON(n->gro_list);
4225 list_del(&n->poll_list);
4226 smp_mb__before_clear_bit();
4227 clear_bit(NAPI_STATE_SCHED, &n->state);
4229 EXPORT_SYMBOL(__napi_complete);
4231 void napi_complete(struct napi_struct *n)
4233 unsigned long flags;
4236 * don't let napi dequeue from the cpu poll list
4237 * just in case its running on a different cpu
4239 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4242 napi_gro_flush(n, false);
4243 local_irq_save(flags);
4245 local_irq_restore(flags);
4247 EXPORT_SYMBOL(napi_complete);
4249 /* must be called under rcu_read_lock(), as we dont take a reference */
4250 struct napi_struct *napi_by_id(unsigned int napi_id)
4252 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4253 struct napi_struct *napi;
4255 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4256 if (napi->napi_id == napi_id)
4261 EXPORT_SYMBOL_GPL(napi_by_id);
4263 void napi_hash_add(struct napi_struct *napi)
4265 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4267 spin_lock(&napi_hash_lock);
4269 /* 0 is not a valid id, we also skip an id that is taken
4270 * we expect both events to be extremely rare
4273 while (!napi->napi_id) {
4274 napi->napi_id = ++napi_gen_id;
4275 if (napi_by_id(napi->napi_id))
4279 hlist_add_head_rcu(&napi->napi_hash_node,
4280 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4282 spin_unlock(&napi_hash_lock);
4285 EXPORT_SYMBOL_GPL(napi_hash_add);
4287 /* Warning : caller is responsible to make sure rcu grace period
4288 * is respected before freeing memory containing @napi
4290 void napi_hash_del(struct napi_struct *napi)
4292 spin_lock(&napi_hash_lock);
4294 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4295 hlist_del_rcu(&napi->napi_hash_node);
4297 spin_unlock(&napi_hash_lock);
4299 EXPORT_SYMBOL_GPL(napi_hash_del);
4301 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4302 int (*poll)(struct napi_struct *, int), int weight)
4304 INIT_LIST_HEAD(&napi->poll_list);
4305 napi->gro_count = 0;
4306 napi->gro_list = NULL;
4309 if (weight > NAPI_POLL_WEIGHT)
4310 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4312 napi->weight = weight;
4313 list_add(&napi->dev_list, &dev->napi_list);
4315 #ifdef CONFIG_NETPOLL
4316 spin_lock_init(&napi->poll_lock);
4317 napi->poll_owner = -1;
4319 set_bit(NAPI_STATE_SCHED, &napi->state);
4321 EXPORT_SYMBOL(netif_napi_add);
4323 void netif_napi_del(struct napi_struct *napi)
4325 list_del_init(&napi->dev_list);
4326 napi_free_frags(napi);
4328 kfree_skb_list(napi->gro_list);
4329 napi->gro_list = NULL;
4330 napi->gro_count = 0;
4332 EXPORT_SYMBOL(netif_napi_del);
4334 static void net_rx_action(struct softirq_action *h)
4336 struct softnet_data *sd = &__get_cpu_var(softnet_data);
4337 unsigned long time_limit = jiffies + 2;
4338 int budget = netdev_budget;
4341 local_irq_disable();
4343 while (!list_empty(&sd->poll_list)) {
4344 struct napi_struct *n;
4347 /* If softirq window is exhuasted then punt.
4348 * Allow this to run for 2 jiffies since which will allow
4349 * an average latency of 1.5/HZ.
4351 if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit)))
4356 /* Even though interrupts have been re-enabled, this
4357 * access is safe because interrupts can only add new
4358 * entries to the tail of this list, and only ->poll()
4359 * calls can remove this head entry from the list.
4361 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
4363 have = netpoll_poll_lock(n);
4367 /* This NAPI_STATE_SCHED test is for avoiding a race
4368 * with netpoll's poll_napi(). Only the entity which
4369 * obtains the lock and sees NAPI_STATE_SCHED set will
4370 * actually make the ->poll() call. Therefore we avoid
4371 * accidentally calling ->poll() when NAPI is not scheduled.
4374 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4375 work = n->poll(n, weight);
4379 WARN_ON_ONCE(work > weight);
4383 local_irq_disable();
4385 /* Drivers must not modify the NAPI state if they
4386 * consume the entire weight. In such cases this code
4387 * still "owns" the NAPI instance and therefore can
4388 * move the instance around on the list at-will.
4390 if (unlikely(work == weight)) {
4391 if (unlikely(napi_disable_pending(n))) {
4394 local_irq_disable();
4397 /* flush too old packets
4398 * If HZ < 1000, flush all packets.
4401 napi_gro_flush(n, HZ >= 1000);
4402 local_irq_disable();
4404 list_move_tail(&n->poll_list, &sd->poll_list);
4408 netpoll_poll_unlock(have);
4411 net_rps_action_and_irq_enable(sd);
4413 #ifdef CONFIG_NET_DMA
4415 * There may not be any more sk_buffs coming right now, so push
4416 * any pending DMA copies to hardware
4418 dma_issue_pending_all();
4425 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4429 struct netdev_adjacent {
4430 struct net_device *dev;
4432 /* upper master flag, there can only be one master device per list */
4435 /* counter for the number of times this device was added to us */
4438 /* private field for the users */
4441 struct list_head list;
4442 struct rcu_head rcu;
4445 static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev,
4446 struct net_device *adj_dev,
4447 struct list_head *adj_list)
4449 struct netdev_adjacent *adj;
4451 list_for_each_entry(adj, adj_list, list) {
4452 if (adj->dev == adj_dev)
4459 * netdev_has_upper_dev - Check if device is linked to an upper device
4461 * @upper_dev: upper device to check
4463 * Find out if a device is linked to specified upper device and return true
4464 * in case it is. Note that this checks only immediate upper device,
4465 * not through a complete stack of devices. The caller must hold the RTNL lock.
4467 bool netdev_has_upper_dev(struct net_device *dev,
4468 struct net_device *upper_dev)
4472 return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper);
4474 EXPORT_SYMBOL(netdev_has_upper_dev);
4477 * netdev_has_any_upper_dev - Check if device is linked to some device
4480 * Find out if a device is linked to an upper device and return true in case
4481 * it is. The caller must hold the RTNL lock.
4483 static bool netdev_has_any_upper_dev(struct net_device *dev)
4487 return !list_empty(&dev->all_adj_list.upper);
4491 * netdev_master_upper_dev_get - Get master upper device
4494 * Find a master upper device and return pointer to it or NULL in case
4495 * it's not there. The caller must hold the RTNL lock.
4497 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4499 struct netdev_adjacent *upper;
4503 if (list_empty(&dev->adj_list.upper))
4506 upper = list_first_entry(&dev->adj_list.upper,
4507 struct netdev_adjacent, list);
4508 if (likely(upper->master))
4512 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4514 void *netdev_adjacent_get_private(struct list_head *adj_list)
4516 struct netdev_adjacent *adj;
4518 adj = list_entry(adj_list, struct netdev_adjacent, list);
4520 return adj->private;
4522 EXPORT_SYMBOL(netdev_adjacent_get_private);
4525 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4527 * @iter: list_head ** of the current position
4529 * Gets the next device from the dev's upper list, starting from iter
4530 * position. The caller must hold RCU read lock.
4532 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
4533 struct list_head **iter)
4535 struct netdev_adjacent *upper;
4537 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4539 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4541 if (&upper->list == &dev->all_adj_list.upper)
4544 *iter = &upper->list;
4548 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
4551 * netdev_lower_get_next_private - Get the next ->private from the
4552 * lower neighbour list
4554 * @iter: list_head ** of the current position
4556 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4557 * list, starting from iter position. The caller must hold either hold the
4558 * RTNL lock or its own locking that guarantees that the neighbour lower
4559 * list will remain unchainged.
4561 void *netdev_lower_get_next_private(struct net_device *dev,
4562 struct list_head **iter)
4564 struct netdev_adjacent *lower;
4566 lower = list_entry(*iter, struct netdev_adjacent, list);
4568 if (&lower->list == &dev->adj_list.lower)
4572 *iter = lower->list.next;
4574 return lower->private;
4576 EXPORT_SYMBOL(netdev_lower_get_next_private);
4579 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4580 * lower neighbour list, RCU
4583 * @iter: list_head ** of the current position
4585 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4586 * list, starting from iter position. The caller must hold RCU read lock.
4588 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
4589 struct list_head **iter)
4591 struct netdev_adjacent *lower;
4593 WARN_ON_ONCE(!rcu_read_lock_held());
4595 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4597 if (&lower->list == &dev->adj_list.lower)
4601 *iter = &lower->list;
4603 return lower->private;
4605 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
4608 * netdev_lower_get_first_private_rcu - Get the first ->private from the
4609 * lower neighbour list, RCU
4613 * Gets the first netdev_adjacent->private from the dev's lower neighbour
4614 * list. The caller must hold RCU read lock.
4616 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
4618 struct netdev_adjacent *lower;
4620 lower = list_first_or_null_rcu(&dev->adj_list.lower,
4621 struct netdev_adjacent, list);
4623 return lower->private;
4626 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
4629 * netdev_master_upper_dev_get_rcu - Get master upper device
4632 * Find a master upper device and return pointer to it or NULL in case
4633 * it's not there. The caller must hold the RCU read lock.
4635 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4637 struct netdev_adjacent *upper;
4639 upper = list_first_or_null_rcu(&dev->adj_list.upper,
4640 struct netdev_adjacent, list);
4641 if (upper && likely(upper->master))
4645 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4647 static int netdev_adjacent_sysfs_add(struct net_device *dev,
4648 struct net_device *adj_dev,
4649 struct list_head *dev_list)
4651 char linkname[IFNAMSIZ+7];
4652 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4653 "upper_%s" : "lower_%s", adj_dev->name);
4654 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
4657 static void netdev_adjacent_sysfs_del(struct net_device *dev,
4659 struct list_head *dev_list)
4661 char linkname[IFNAMSIZ+7];
4662 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4663 "upper_%s" : "lower_%s", name);
4664 sysfs_remove_link(&(dev->dev.kobj), linkname);
4667 #define netdev_adjacent_is_neigh_list(dev, dev_list) \
4668 (dev_list == &dev->adj_list.upper || \
4669 dev_list == &dev->adj_list.lower)
4671 static int __netdev_adjacent_dev_insert(struct net_device *dev,
4672 struct net_device *adj_dev,
4673 struct list_head *dev_list,
4674 void *private, bool master)
4676 struct netdev_adjacent *adj;
4679 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4686 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
4691 adj->master = master;
4693 adj->private = private;
4696 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
4697 adj_dev->name, dev->name, adj_dev->name);
4699 if (netdev_adjacent_is_neigh_list(dev, dev_list)) {
4700 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
4705 /* Ensure that master link is always the first item in list. */
4707 ret = sysfs_create_link(&(dev->dev.kobj),
4708 &(adj_dev->dev.kobj), "master");
4710 goto remove_symlinks;
4712 list_add_rcu(&adj->list, dev_list);
4714 list_add_tail_rcu(&adj->list, dev_list);
4720 if (netdev_adjacent_is_neigh_list(dev, dev_list))
4721 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
4729 static void __netdev_adjacent_dev_remove(struct net_device *dev,
4730 struct net_device *adj_dev,
4731 struct list_head *dev_list)
4733 struct netdev_adjacent *adj;
4735 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4738 pr_err("tried to remove device %s from %s\n",
4739 dev->name, adj_dev->name);
4743 if (adj->ref_nr > 1) {
4744 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
4751 sysfs_remove_link(&(dev->dev.kobj), "master");
4753 if (netdev_adjacent_is_neigh_list(dev, dev_list))
4754 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
4756 list_del_rcu(&adj->list);
4757 pr_debug("dev_put for %s, because link removed from %s to %s\n",
4758 adj_dev->name, dev->name, adj_dev->name);
4760 kfree_rcu(adj, rcu);
4763 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
4764 struct net_device *upper_dev,
4765 struct list_head *up_list,
4766 struct list_head *down_list,
4767 void *private, bool master)
4771 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
4776 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
4779 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4786 static int __netdev_adjacent_dev_link(struct net_device *dev,
4787 struct net_device *upper_dev)
4789 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
4790 &dev->all_adj_list.upper,
4791 &upper_dev->all_adj_list.lower,
4795 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
4796 struct net_device *upper_dev,
4797 struct list_head *up_list,
4798 struct list_head *down_list)
4800 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4801 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
4804 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
4805 struct net_device *upper_dev)
4807 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
4808 &dev->all_adj_list.upper,
4809 &upper_dev->all_adj_list.lower);
4812 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
4813 struct net_device *upper_dev,
4814 void *private, bool master)
4816 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
4821 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
4822 &dev->adj_list.upper,
4823 &upper_dev->adj_list.lower,
4826 __netdev_adjacent_dev_unlink(dev, upper_dev);
4833 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
4834 struct net_device *upper_dev)
4836 __netdev_adjacent_dev_unlink(dev, upper_dev);
4837 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
4838 &dev->adj_list.upper,
4839 &upper_dev->adj_list.lower);
4842 static int __netdev_upper_dev_link(struct net_device *dev,
4843 struct net_device *upper_dev, bool master,
4846 struct netdev_adjacent *i, *j, *to_i, *to_j;
4851 if (dev == upper_dev)
4854 /* To prevent loops, check if dev is not upper device to upper_dev. */
4855 if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper))
4858 if (__netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper))
4861 if (master && netdev_master_upper_dev_get(dev))
4864 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
4869 /* Now that we linked these devs, make all the upper_dev's
4870 * all_adj_list.upper visible to every dev's all_adj_list.lower an
4871 * versa, and don't forget the devices itself. All of these
4872 * links are non-neighbours.
4874 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4875 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
4876 pr_debug("Interlinking %s with %s, non-neighbour\n",
4877 i->dev->name, j->dev->name);
4878 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
4884 /* add dev to every upper_dev's upper device */
4885 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
4886 pr_debug("linking %s's upper device %s with %s\n",
4887 upper_dev->name, i->dev->name, dev->name);
4888 ret = __netdev_adjacent_dev_link(dev, i->dev);
4890 goto rollback_upper_mesh;
4893 /* add upper_dev to every dev's lower device */
4894 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4895 pr_debug("linking %s's lower device %s with %s\n", dev->name,
4896 i->dev->name, upper_dev->name);
4897 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
4899 goto rollback_lower_mesh;
4902 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
4905 rollback_lower_mesh:
4907 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4910 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
4915 rollback_upper_mesh:
4917 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
4920 __netdev_adjacent_dev_unlink(dev, i->dev);
4928 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4929 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
4930 if (i == to_i && j == to_j)
4932 __netdev_adjacent_dev_unlink(i->dev, j->dev);
4938 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
4944 * netdev_upper_dev_link - Add a link to the upper device
4946 * @upper_dev: new upper device
4948 * Adds a link to device which is upper to this one. The caller must hold
4949 * the RTNL lock. On a failure a negative errno code is returned.
4950 * On success the reference counts are adjusted and the function
4953 int netdev_upper_dev_link(struct net_device *dev,
4954 struct net_device *upper_dev)
4956 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
4958 EXPORT_SYMBOL(netdev_upper_dev_link);
4961 * netdev_master_upper_dev_link - Add a master link to the upper device
4963 * @upper_dev: new upper device
4965 * Adds a link to device which is upper to this one. In this case, only
4966 * one master upper device can be linked, although other non-master devices
4967 * might be linked as well. The caller must hold the RTNL lock.
4968 * On a failure a negative errno code is returned. On success the reference
4969 * counts are adjusted and the function returns zero.
4971 int netdev_master_upper_dev_link(struct net_device *dev,
4972 struct net_device *upper_dev)
4974 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
4976 EXPORT_SYMBOL(netdev_master_upper_dev_link);
4978 int netdev_master_upper_dev_link_private(struct net_device *dev,
4979 struct net_device *upper_dev,
4982 return __netdev_upper_dev_link(dev, upper_dev, true, private);
4984 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
4987 * netdev_upper_dev_unlink - Removes a link to upper device
4989 * @upper_dev: new upper device
4991 * Removes a link to device which is upper to this one. The caller must hold
4994 void netdev_upper_dev_unlink(struct net_device *dev,
4995 struct net_device *upper_dev)
4997 struct netdev_adjacent *i, *j;
5000 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5002 /* Here is the tricky part. We must remove all dev's lower
5003 * devices from all upper_dev's upper devices and vice
5004 * versa, to maintain the graph relationship.
5006 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5007 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
5008 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5010 /* remove also the devices itself from lower/upper device
5013 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5014 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5016 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5017 __netdev_adjacent_dev_unlink(dev, i->dev);
5019 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5021 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5023 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5025 struct netdev_adjacent *iter;
5027 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5028 netdev_adjacent_sysfs_del(iter->dev, oldname,
5029 &iter->dev->adj_list.lower);
5030 netdev_adjacent_sysfs_add(iter->dev, dev,
5031 &iter->dev->adj_list.lower);
5034 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5035 netdev_adjacent_sysfs_del(iter->dev, oldname,
5036 &iter->dev->adj_list.upper);
5037 netdev_adjacent_sysfs_add(iter->dev, dev,
5038 &iter->dev->adj_list.upper);
5042 void *netdev_lower_dev_get_private(struct net_device *dev,
5043 struct net_device *lower_dev)
5045 struct netdev_adjacent *lower;
5049 lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower);
5053 return lower->private;
5055 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5057 static void dev_change_rx_flags(struct net_device *dev, int flags)
5059 const struct net_device_ops *ops = dev->netdev_ops;
5061 if (ops->ndo_change_rx_flags)
5062 ops->ndo_change_rx_flags(dev, flags);
5065 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5067 unsigned int old_flags = dev->flags;
5073 dev->flags |= IFF_PROMISC;
5074 dev->promiscuity += inc;
5075 if (dev->promiscuity == 0) {
5078 * If inc causes overflow, untouch promisc and return error.
5081 dev->flags &= ~IFF_PROMISC;
5083 dev->promiscuity -= inc;
5084 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5089 if (dev->flags != old_flags) {
5090 pr_info("device %s %s promiscuous mode\n",
5092 dev->flags & IFF_PROMISC ? "entered" : "left");
5093 if (audit_enabled) {
5094 current_uid_gid(&uid, &gid);
5095 audit_log(current->audit_context, GFP_ATOMIC,
5096 AUDIT_ANOM_PROMISCUOUS,
5097 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5098 dev->name, (dev->flags & IFF_PROMISC),
5099 (old_flags & IFF_PROMISC),
5100 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5101 from_kuid(&init_user_ns, uid),
5102 from_kgid(&init_user_ns, gid),
5103 audit_get_sessionid(current));
5106 dev_change_rx_flags(dev, IFF_PROMISC);
5109 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5114 * dev_set_promiscuity - update promiscuity count on a device
5118 * Add or remove promiscuity from a device. While the count in the device
5119 * remains above zero the interface remains promiscuous. Once it hits zero
5120 * the device reverts back to normal filtering operation. A negative inc
5121 * value is used to drop promiscuity on the device.
5122 * Return 0 if successful or a negative errno code on error.
5124 int dev_set_promiscuity(struct net_device *dev, int inc)
5126 unsigned int old_flags = dev->flags;
5129 err = __dev_set_promiscuity(dev, inc, true);
5132 if (dev->flags != old_flags)
5133 dev_set_rx_mode(dev);
5136 EXPORT_SYMBOL(dev_set_promiscuity);
5138 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5140 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5144 dev->flags |= IFF_ALLMULTI;
5145 dev->allmulti += inc;
5146 if (dev->allmulti == 0) {
5149 * If inc causes overflow, untouch allmulti and return error.
5152 dev->flags &= ~IFF_ALLMULTI;
5154 dev->allmulti -= inc;
5155 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5160 if (dev->flags ^ old_flags) {
5161 dev_change_rx_flags(dev, IFF_ALLMULTI);
5162 dev_set_rx_mode(dev);
5164 __dev_notify_flags(dev, old_flags,
5165 dev->gflags ^ old_gflags);
5171 * dev_set_allmulti - update allmulti count on a device
5175 * Add or remove reception of all multicast frames to a device. While the
5176 * count in the device remains above zero the interface remains listening
5177 * to all interfaces. Once it hits zero the device reverts back to normal
5178 * filtering operation. A negative @inc value is used to drop the counter
5179 * when releasing a resource needing all multicasts.
5180 * Return 0 if successful or a negative errno code on error.
5183 int dev_set_allmulti(struct net_device *dev, int inc)
5185 return __dev_set_allmulti(dev, inc, true);
5187 EXPORT_SYMBOL(dev_set_allmulti);
5190 * Upload unicast and multicast address lists to device and
5191 * configure RX filtering. When the device doesn't support unicast
5192 * filtering it is put in promiscuous mode while unicast addresses
5195 void __dev_set_rx_mode(struct net_device *dev)
5197 const struct net_device_ops *ops = dev->netdev_ops;
5199 /* dev_open will call this function so the list will stay sane. */
5200 if (!(dev->flags&IFF_UP))
5203 if (!netif_device_present(dev))
5206 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5207 /* Unicast addresses changes may only happen under the rtnl,
5208 * therefore calling __dev_set_promiscuity here is safe.
5210 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5211 __dev_set_promiscuity(dev, 1, false);
5212 dev->uc_promisc = true;
5213 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5214 __dev_set_promiscuity(dev, -1, false);
5215 dev->uc_promisc = false;
5219 if (ops->ndo_set_rx_mode)
5220 ops->ndo_set_rx_mode(dev);
5223 void dev_set_rx_mode(struct net_device *dev)
5225 netif_addr_lock_bh(dev);
5226 __dev_set_rx_mode(dev);
5227 netif_addr_unlock_bh(dev);
5231 * dev_get_flags - get flags reported to userspace
5234 * Get the combination of flag bits exported through APIs to userspace.
5236 unsigned int dev_get_flags(const struct net_device *dev)
5240 flags = (dev->flags & ~(IFF_PROMISC |
5245 (dev->gflags & (IFF_PROMISC |
5248 if (netif_running(dev)) {
5249 if (netif_oper_up(dev))
5250 flags |= IFF_RUNNING;
5251 if (netif_carrier_ok(dev))
5252 flags |= IFF_LOWER_UP;
5253 if (netif_dormant(dev))
5254 flags |= IFF_DORMANT;
5259 EXPORT_SYMBOL(dev_get_flags);
5261 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5263 unsigned int old_flags = dev->flags;
5269 * Set the flags on our device.
5272 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5273 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5275 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5279 * Load in the correct multicast list now the flags have changed.
5282 if ((old_flags ^ flags) & IFF_MULTICAST)
5283 dev_change_rx_flags(dev, IFF_MULTICAST);
5285 dev_set_rx_mode(dev);
5288 * Have we downed the interface. We handle IFF_UP ourselves
5289 * according to user attempts to set it, rather than blindly
5294 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
5295 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5298 dev_set_rx_mode(dev);
5301 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5302 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5303 unsigned int old_flags = dev->flags;
5305 dev->gflags ^= IFF_PROMISC;
5307 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5308 if (dev->flags != old_flags)
5309 dev_set_rx_mode(dev);
5312 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5313 is important. Some (broken) drivers set IFF_PROMISC, when
5314 IFF_ALLMULTI is requested not asking us and not reporting.
5316 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5317 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5319 dev->gflags ^= IFF_ALLMULTI;
5320 __dev_set_allmulti(dev, inc, false);
5326 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5327 unsigned int gchanges)
5329 unsigned int changes = dev->flags ^ old_flags;
5332 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
5334 if (changes & IFF_UP) {
5335 if (dev->flags & IFF_UP)
5336 call_netdevice_notifiers(NETDEV_UP, dev);
5338 call_netdevice_notifiers(NETDEV_DOWN, dev);
5341 if (dev->flags & IFF_UP &&
5342 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5343 struct netdev_notifier_change_info change_info;
5345 change_info.flags_changed = changes;
5346 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
5352 * dev_change_flags - change device settings
5354 * @flags: device state flags
5356 * Change settings on device based state flags. The flags are
5357 * in the userspace exported format.
5359 int dev_change_flags(struct net_device *dev, unsigned int flags)
5362 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
5364 ret = __dev_change_flags(dev, flags);
5368 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
5369 __dev_notify_flags(dev, old_flags, changes);
5372 EXPORT_SYMBOL(dev_change_flags);
5374 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
5376 const struct net_device_ops *ops = dev->netdev_ops;
5378 if (ops->ndo_change_mtu)
5379 return ops->ndo_change_mtu(dev, new_mtu);
5386 * dev_set_mtu - Change maximum transfer unit
5388 * @new_mtu: new transfer unit
5390 * Change the maximum transfer size of the network device.
5392 int dev_set_mtu(struct net_device *dev, int new_mtu)
5396 if (new_mtu == dev->mtu)
5399 /* MTU must be positive. */
5403 if (!netif_device_present(dev))
5406 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
5407 err = notifier_to_errno(err);
5411 orig_mtu = dev->mtu;
5412 err = __dev_set_mtu(dev, new_mtu);
5415 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5416 err = notifier_to_errno(err);
5418 /* setting mtu back and notifying everyone again,
5419 * so that they have a chance to revert changes.
5421 __dev_set_mtu(dev, orig_mtu);
5422 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5427 EXPORT_SYMBOL(dev_set_mtu);
5430 * dev_set_group - Change group this device belongs to
5432 * @new_group: group this device should belong to
5434 void dev_set_group(struct net_device *dev, int new_group)
5436 dev->group = new_group;
5438 EXPORT_SYMBOL(dev_set_group);
5441 * dev_set_mac_address - Change Media Access Control Address
5445 * Change the hardware (MAC) address of the device
5447 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
5449 const struct net_device_ops *ops = dev->netdev_ops;
5452 if (!ops->ndo_set_mac_address)
5454 if (sa->sa_family != dev->type)
5456 if (!netif_device_present(dev))
5458 err = ops->ndo_set_mac_address(dev, sa);
5461 dev->addr_assign_type = NET_ADDR_SET;
5462 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5463 add_device_randomness(dev->dev_addr, dev->addr_len);
5466 EXPORT_SYMBOL(dev_set_mac_address);
5469 * dev_change_carrier - Change device carrier
5471 * @new_carrier: new value
5473 * Change device carrier
5475 int dev_change_carrier(struct net_device *dev, bool new_carrier)
5477 const struct net_device_ops *ops = dev->netdev_ops;
5479 if (!ops->ndo_change_carrier)
5481 if (!netif_device_present(dev))
5483 return ops->ndo_change_carrier(dev, new_carrier);
5485 EXPORT_SYMBOL(dev_change_carrier);
5488 * dev_get_phys_port_id - Get device physical port ID
5492 * Get device physical port ID
5494 int dev_get_phys_port_id(struct net_device *dev,
5495 struct netdev_phys_port_id *ppid)
5497 const struct net_device_ops *ops = dev->netdev_ops;
5499 if (!ops->ndo_get_phys_port_id)
5501 return ops->ndo_get_phys_port_id(dev, ppid);
5503 EXPORT_SYMBOL(dev_get_phys_port_id);
5506 * dev_new_index - allocate an ifindex
5507 * @net: the applicable net namespace
5509 * Returns a suitable unique value for a new device interface
5510 * number. The caller must hold the rtnl semaphore or the
5511 * dev_base_lock to be sure it remains unique.
5513 static int dev_new_index(struct net *net)
5515 int ifindex = net->ifindex;
5519 if (!__dev_get_by_index(net, ifindex))
5520 return net->ifindex = ifindex;
5524 /* Delayed registration/unregisteration */
5525 static LIST_HEAD(net_todo_list);
5526 static DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
5528 static void net_set_todo(struct net_device *dev)
5530 list_add_tail(&dev->todo_list, &net_todo_list);
5531 dev_net(dev)->dev_unreg_count++;
5534 static void rollback_registered_many(struct list_head *head)
5536 struct net_device *dev, *tmp;
5537 LIST_HEAD(close_head);
5539 BUG_ON(dev_boot_phase);
5542 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5543 /* Some devices call without registering
5544 * for initialization unwind. Remove those
5545 * devices and proceed with the remaining.
5547 if (dev->reg_state == NETREG_UNINITIALIZED) {
5548 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5552 list_del(&dev->unreg_list);
5555 dev->dismantle = true;
5556 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5559 /* If device is running, close it first. */
5560 list_for_each_entry(dev, head, unreg_list)
5561 list_add_tail(&dev->close_list, &close_head);
5562 dev_close_many(&close_head);
5564 list_for_each_entry(dev, head, unreg_list) {
5565 /* And unlink it from device chain. */
5566 unlist_netdevice(dev);
5568 dev->reg_state = NETREG_UNREGISTERING;
5573 list_for_each_entry(dev, head, unreg_list) {
5574 /* Shutdown queueing discipline. */
5578 /* Notify protocols, that we are about to destroy
5579 this device. They should clean all the things.
5581 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5583 if (!dev->rtnl_link_ops ||
5584 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5585 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
5588 * Flush the unicast and multicast chains
5593 if (dev->netdev_ops->ndo_uninit)
5594 dev->netdev_ops->ndo_uninit(dev);
5596 /* Notifier chain MUST detach us all upper devices. */
5597 WARN_ON(netdev_has_any_upper_dev(dev));
5599 /* Remove entries from kobject tree */
5600 netdev_unregister_kobject(dev);
5602 /* Remove XPS queueing entries */
5603 netif_reset_xps_queues_gt(dev, 0);
5609 list_for_each_entry(dev, head, unreg_list)
5613 static void rollback_registered(struct net_device *dev)
5617 list_add(&dev->unreg_list, &single);
5618 rollback_registered_many(&single);
5622 static netdev_features_t netdev_fix_features(struct net_device *dev,
5623 netdev_features_t features)
5625 /* Fix illegal checksum combinations */
5626 if ((features & NETIF_F_HW_CSUM) &&
5627 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5628 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5629 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5632 /* TSO requires that SG is present as well. */
5633 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5634 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5635 features &= ~NETIF_F_ALL_TSO;
5638 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
5639 !(features & NETIF_F_IP_CSUM)) {
5640 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
5641 features &= ~NETIF_F_TSO;
5642 features &= ~NETIF_F_TSO_ECN;
5645 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
5646 !(features & NETIF_F_IPV6_CSUM)) {
5647 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
5648 features &= ~NETIF_F_TSO6;
5651 /* TSO ECN requires that TSO is present as well. */
5652 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5653 features &= ~NETIF_F_TSO_ECN;
5655 /* Software GSO depends on SG. */
5656 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5657 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5658 features &= ~NETIF_F_GSO;
5661 /* UFO needs SG and checksumming */
5662 if (features & NETIF_F_UFO) {
5663 /* maybe split UFO into V4 and V6? */
5664 if (!((features & NETIF_F_GEN_CSUM) ||
5665 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5666 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5668 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5669 features &= ~NETIF_F_UFO;
5672 if (!(features & NETIF_F_SG)) {
5674 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5675 features &= ~NETIF_F_UFO;
5682 int __netdev_update_features(struct net_device *dev)
5684 netdev_features_t features;
5689 features = netdev_get_wanted_features(dev);
5691 if (dev->netdev_ops->ndo_fix_features)
5692 features = dev->netdev_ops->ndo_fix_features(dev, features);
5694 /* driver might be less strict about feature dependencies */
5695 features = netdev_fix_features(dev, features);
5697 if (dev->features == features)
5700 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5701 &dev->features, &features);
5703 if (dev->netdev_ops->ndo_set_features)
5704 err = dev->netdev_ops->ndo_set_features(dev, features);
5706 if (unlikely(err < 0)) {
5708 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5709 err, &features, &dev->features);
5714 dev->features = features;
5720 * netdev_update_features - recalculate device features
5721 * @dev: the device to check
5723 * Recalculate dev->features set and send notifications if it
5724 * has changed. Should be called after driver or hardware dependent
5725 * conditions might have changed that influence the features.
5727 void netdev_update_features(struct net_device *dev)
5729 if (__netdev_update_features(dev))
5730 netdev_features_change(dev);
5732 EXPORT_SYMBOL(netdev_update_features);
5735 * netdev_change_features - recalculate device features
5736 * @dev: the device to check
5738 * Recalculate dev->features set and send notifications even
5739 * if they have not changed. Should be called instead of
5740 * netdev_update_features() if also dev->vlan_features might
5741 * have changed to allow the changes to be propagated to stacked
5744 void netdev_change_features(struct net_device *dev)
5746 __netdev_update_features(dev);
5747 netdev_features_change(dev);
5749 EXPORT_SYMBOL(netdev_change_features);
5752 * netif_stacked_transfer_operstate - transfer operstate
5753 * @rootdev: the root or lower level device to transfer state from
5754 * @dev: the device to transfer operstate to
5756 * Transfer operational state from root to device. This is normally
5757 * called when a stacking relationship exists between the root
5758 * device and the device(a leaf device).
5760 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5761 struct net_device *dev)
5763 if (rootdev->operstate == IF_OPER_DORMANT)
5764 netif_dormant_on(dev);
5766 netif_dormant_off(dev);
5768 if (netif_carrier_ok(rootdev)) {
5769 if (!netif_carrier_ok(dev))
5770 netif_carrier_on(dev);
5772 if (netif_carrier_ok(dev))
5773 netif_carrier_off(dev);
5776 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5779 static int netif_alloc_rx_queues(struct net_device *dev)
5781 unsigned int i, count = dev->num_rx_queues;
5782 struct netdev_rx_queue *rx;
5786 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5792 for (i = 0; i < count; i++)
5798 static void netdev_init_one_queue(struct net_device *dev,
5799 struct netdev_queue *queue, void *_unused)
5801 /* Initialize queue lock */
5802 spin_lock_init(&queue->_xmit_lock);
5803 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5804 queue->xmit_lock_owner = -1;
5805 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5808 dql_init(&queue->dql, HZ);
5812 static void netif_free_tx_queues(struct net_device *dev)
5814 if (is_vmalloc_addr(dev->_tx))
5820 static int netif_alloc_netdev_queues(struct net_device *dev)
5822 unsigned int count = dev->num_tx_queues;
5823 struct netdev_queue *tx;
5824 size_t sz = count * sizeof(*tx);
5826 BUG_ON(count < 1 || count > 0xffff);
5828 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
5836 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5837 spin_lock_init(&dev->tx_global_lock);
5843 * register_netdevice - register a network device
5844 * @dev: device to register
5846 * Take a completed network device structure and add it to the kernel
5847 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5848 * chain. 0 is returned on success. A negative errno code is returned
5849 * on a failure to set up the device, or if the name is a duplicate.
5851 * Callers must hold the rtnl semaphore. You may want
5852 * register_netdev() instead of this.
5855 * The locking appears insufficient to guarantee two parallel registers
5856 * will not get the same name.
5859 int register_netdevice(struct net_device *dev)
5862 struct net *net = dev_net(dev);
5864 BUG_ON(dev_boot_phase);
5869 /* When net_device's are persistent, this will be fatal. */
5870 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5873 spin_lock_init(&dev->addr_list_lock);
5874 netdev_set_addr_lockdep_class(dev);
5878 ret = dev_get_valid_name(net, dev, dev->name);
5882 /* Init, if this function is available */
5883 if (dev->netdev_ops->ndo_init) {
5884 ret = dev->netdev_ops->ndo_init(dev);
5892 if (((dev->hw_features | dev->features) &
5893 NETIF_F_HW_VLAN_CTAG_FILTER) &&
5894 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
5895 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
5896 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
5903 dev->ifindex = dev_new_index(net);
5904 else if (__dev_get_by_index(net, dev->ifindex))
5907 if (dev->iflink == -1)
5908 dev->iflink = dev->ifindex;
5910 /* Transfer changeable features to wanted_features and enable
5911 * software offloads (GSO and GRO).
5913 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5914 dev->features |= NETIF_F_SOFT_FEATURES;
5915 dev->wanted_features = dev->features & dev->hw_features;
5917 if (!(dev->flags & IFF_LOOPBACK)) {
5918 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5921 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5923 dev->vlan_features |= NETIF_F_HIGHDMA;
5925 /* Make NETIF_F_SG inheritable to tunnel devices.
5927 dev->hw_enc_features |= NETIF_F_SG;
5929 /* Make NETIF_F_SG inheritable to MPLS.
5931 dev->mpls_features |= NETIF_F_SG;
5933 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5934 ret = notifier_to_errno(ret);
5938 ret = netdev_register_kobject(dev);
5941 dev->reg_state = NETREG_REGISTERED;
5943 __netdev_update_features(dev);
5946 * Default initial state at registry is that the
5947 * device is present.
5950 set_bit(__LINK_STATE_PRESENT, &dev->state);
5952 linkwatch_init_dev(dev);
5954 dev_init_scheduler(dev);
5956 list_netdevice(dev);
5957 add_device_randomness(dev->dev_addr, dev->addr_len);
5959 /* If the device has permanent device address, driver should
5960 * set dev_addr and also addr_assign_type should be set to
5961 * NET_ADDR_PERM (default value).
5963 if (dev->addr_assign_type == NET_ADDR_PERM)
5964 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
5966 /* Notify protocols, that a new device appeared. */
5967 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5968 ret = notifier_to_errno(ret);
5970 rollback_registered(dev);
5971 dev->reg_state = NETREG_UNREGISTERED;
5974 * Prevent userspace races by waiting until the network
5975 * device is fully setup before sending notifications.
5977 if (!dev->rtnl_link_ops ||
5978 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5979 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
5985 if (dev->netdev_ops->ndo_uninit)
5986 dev->netdev_ops->ndo_uninit(dev);
5989 EXPORT_SYMBOL(register_netdevice);
5992 * init_dummy_netdev - init a dummy network device for NAPI
5993 * @dev: device to init
5995 * This takes a network device structure and initialize the minimum
5996 * amount of fields so it can be used to schedule NAPI polls without
5997 * registering a full blown interface. This is to be used by drivers
5998 * that need to tie several hardware interfaces to a single NAPI
5999 * poll scheduler due to HW limitations.
6001 int init_dummy_netdev(struct net_device *dev)
6003 /* Clear everything. Note we don't initialize spinlocks
6004 * are they aren't supposed to be taken by any of the
6005 * NAPI code and this dummy netdev is supposed to be
6006 * only ever used for NAPI polls
6008 memset(dev, 0, sizeof(struct net_device));
6010 /* make sure we BUG if trying to hit standard
6011 * register/unregister code path
6013 dev->reg_state = NETREG_DUMMY;
6015 /* NAPI wants this */
6016 INIT_LIST_HEAD(&dev->napi_list);
6018 /* a dummy interface is started by default */
6019 set_bit(__LINK_STATE_PRESENT, &dev->state);
6020 set_bit(__LINK_STATE_START, &dev->state);
6022 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6023 * because users of this 'device' dont need to change
6029 EXPORT_SYMBOL_GPL(init_dummy_netdev);
6033 * register_netdev - register a network device
6034 * @dev: device to register
6036 * Take a completed network device structure and add it to the kernel
6037 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6038 * chain. 0 is returned on success. A negative errno code is returned
6039 * on a failure to set up the device, or if the name is a duplicate.
6041 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6042 * and expands the device name if you passed a format string to
6045 int register_netdev(struct net_device *dev)
6050 err = register_netdevice(dev);
6054 EXPORT_SYMBOL(register_netdev);
6056 int netdev_refcnt_read(const struct net_device *dev)
6060 for_each_possible_cpu(i)
6061 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
6064 EXPORT_SYMBOL(netdev_refcnt_read);
6067 * netdev_wait_allrefs - wait until all references are gone.
6068 * @dev: target net_device
6070 * This is called when unregistering network devices.
6072 * Any protocol or device that holds a reference should register
6073 * for netdevice notification, and cleanup and put back the
6074 * reference if they receive an UNREGISTER event.
6075 * We can get stuck here if buggy protocols don't correctly
6078 static void netdev_wait_allrefs(struct net_device *dev)
6080 unsigned long rebroadcast_time, warning_time;
6083 linkwatch_forget_dev(dev);
6085 rebroadcast_time = warning_time = jiffies;
6086 refcnt = netdev_refcnt_read(dev);
6088 while (refcnt != 0) {
6089 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6092 /* Rebroadcast unregister notification */
6093 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6099 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6100 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6102 /* We must not have linkwatch events
6103 * pending on unregister. If this
6104 * happens, we simply run the queue
6105 * unscheduled, resulting in a noop
6108 linkwatch_run_queue();
6113 rebroadcast_time = jiffies;
6118 refcnt = netdev_refcnt_read(dev);
6120 if (time_after(jiffies, warning_time + 10 * HZ)) {
6121 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6123 warning_time = jiffies;
6132 * register_netdevice(x1);
6133 * register_netdevice(x2);
6135 * unregister_netdevice(y1);
6136 * unregister_netdevice(y2);
6142 * We are invoked by rtnl_unlock().
6143 * This allows us to deal with problems:
6144 * 1) We can delete sysfs objects which invoke hotplug
6145 * without deadlocking with linkwatch via keventd.
6146 * 2) Since we run with the RTNL semaphore not held, we can sleep
6147 * safely in order to wait for the netdev refcnt to drop to zero.
6149 * We must not return until all unregister events added during
6150 * the interval the lock was held have been completed.
6152 void netdev_run_todo(void)
6154 struct list_head list;
6156 /* Snapshot list, allow later requests */
6157 list_replace_init(&net_todo_list, &list);
6162 /* Wait for rcu callbacks to finish before next phase */
6163 if (!list_empty(&list))
6166 while (!list_empty(&list)) {
6167 struct net_device *dev
6168 = list_first_entry(&list, struct net_device, todo_list);
6169 list_del(&dev->todo_list);
6172 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6175 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6176 pr_err("network todo '%s' but state %d\n",
6177 dev->name, dev->reg_state);
6182 dev->reg_state = NETREG_UNREGISTERED;
6184 on_each_cpu(flush_backlog, dev, 1);
6186 netdev_wait_allrefs(dev);
6189 BUG_ON(netdev_refcnt_read(dev));
6190 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6191 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6192 WARN_ON(dev->dn_ptr);
6194 if (dev->destructor)
6195 dev->destructor(dev);
6197 /* Report a network device has been unregistered */
6199 dev_net(dev)->dev_unreg_count--;
6201 wake_up(&netdev_unregistering_wq);
6203 /* Free network device */
6204 kobject_put(&dev->dev.kobj);
6208 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6209 * fields in the same order, with only the type differing.
6211 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6212 const struct net_device_stats *netdev_stats)
6214 #if BITS_PER_LONG == 64
6215 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6216 memcpy(stats64, netdev_stats, sizeof(*stats64));
6218 size_t i, n = sizeof(*stats64) / sizeof(u64);
6219 const unsigned long *src = (const unsigned long *)netdev_stats;
6220 u64 *dst = (u64 *)stats64;
6222 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6223 sizeof(*stats64) / sizeof(u64));
6224 for (i = 0; i < n; i++)
6228 EXPORT_SYMBOL(netdev_stats_to_stats64);
6231 * dev_get_stats - get network device statistics
6232 * @dev: device to get statistics from
6233 * @storage: place to store stats
6235 * Get network statistics from device. Return @storage.
6236 * The device driver may provide its own method by setting
6237 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6238 * otherwise the internal statistics structure is used.
6240 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6241 struct rtnl_link_stats64 *storage)
6243 const struct net_device_ops *ops = dev->netdev_ops;
6245 if (ops->ndo_get_stats64) {
6246 memset(storage, 0, sizeof(*storage));
6247 ops->ndo_get_stats64(dev, storage);
6248 } else if (ops->ndo_get_stats) {
6249 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6251 netdev_stats_to_stats64(storage, &dev->stats);
6253 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6256 EXPORT_SYMBOL(dev_get_stats);
6258 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
6260 struct netdev_queue *queue = dev_ingress_queue(dev);
6262 #ifdef CONFIG_NET_CLS_ACT
6265 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
6268 netdev_init_one_queue(dev, queue, NULL);
6269 queue->qdisc = &noop_qdisc;
6270 queue->qdisc_sleeping = &noop_qdisc;
6271 rcu_assign_pointer(dev->ingress_queue, queue);
6276 static const struct ethtool_ops default_ethtool_ops;
6278 void netdev_set_default_ethtool_ops(struct net_device *dev,
6279 const struct ethtool_ops *ops)
6281 if (dev->ethtool_ops == &default_ethtool_ops)
6282 dev->ethtool_ops = ops;
6284 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
6286 void netdev_freemem(struct net_device *dev)
6288 char *addr = (char *)dev - dev->padded;
6290 if (is_vmalloc_addr(addr))
6297 * alloc_netdev_mqs - allocate network device
6298 * @sizeof_priv: size of private data to allocate space for
6299 * @name: device name format string
6300 * @setup: callback to initialize device
6301 * @txqs: the number of TX subqueues to allocate
6302 * @rxqs: the number of RX subqueues to allocate
6304 * Allocates a struct net_device with private data area for driver use
6305 * and performs basic initialization. Also allocates subqueue structs
6306 * for each queue on the device.
6308 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
6309 void (*setup)(struct net_device *),
6310 unsigned int txqs, unsigned int rxqs)
6312 struct net_device *dev;
6314 struct net_device *p;
6316 BUG_ON(strlen(name) >= sizeof(dev->name));
6319 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6325 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6330 alloc_size = sizeof(struct net_device);
6332 /* ensure 32-byte alignment of private area */
6333 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6334 alloc_size += sizeof_priv;
6336 /* ensure 32-byte alignment of whole construct */
6337 alloc_size += NETDEV_ALIGN - 1;
6339 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6341 p = vzalloc(alloc_size);
6345 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6346 dev->padded = (char *)dev - (char *)p;
6348 dev->pcpu_refcnt = alloc_percpu(int);
6349 if (!dev->pcpu_refcnt)
6352 if (dev_addr_init(dev))
6358 dev_net_set(dev, &init_net);
6360 dev->gso_max_size = GSO_MAX_SIZE;
6361 dev->gso_max_segs = GSO_MAX_SEGS;
6363 INIT_LIST_HEAD(&dev->napi_list);
6364 INIT_LIST_HEAD(&dev->unreg_list);
6365 INIT_LIST_HEAD(&dev->close_list);
6366 INIT_LIST_HEAD(&dev->link_watch_list);
6367 INIT_LIST_HEAD(&dev->adj_list.upper);
6368 INIT_LIST_HEAD(&dev->adj_list.lower);
6369 INIT_LIST_HEAD(&dev->all_adj_list.upper);
6370 INIT_LIST_HEAD(&dev->all_adj_list.lower);
6371 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6374 dev->num_tx_queues = txqs;
6375 dev->real_num_tx_queues = txqs;
6376 if (netif_alloc_netdev_queues(dev))
6380 dev->num_rx_queues = rxqs;
6381 dev->real_num_rx_queues = rxqs;
6382 if (netif_alloc_rx_queues(dev))
6386 strcpy(dev->name, name);
6387 dev->group = INIT_NETDEV_GROUP;
6388 if (!dev->ethtool_ops)
6389 dev->ethtool_ops = &default_ethtool_ops;
6397 free_percpu(dev->pcpu_refcnt);
6398 netif_free_tx_queues(dev);
6404 netdev_freemem(dev);
6407 EXPORT_SYMBOL(alloc_netdev_mqs);
6410 * free_netdev - free network device
6413 * This function does the last stage of destroying an allocated device
6414 * interface. The reference to the device object is released.
6415 * If this is the last reference then it will be freed.
6417 void free_netdev(struct net_device *dev)
6419 struct napi_struct *p, *n;
6421 release_net(dev_net(dev));
6423 netif_free_tx_queues(dev);
6428 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6430 /* Flush device addresses */
6431 dev_addr_flush(dev);
6433 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6436 free_percpu(dev->pcpu_refcnt);
6437 dev->pcpu_refcnt = NULL;
6439 /* Compatibility with error handling in drivers */
6440 if (dev->reg_state == NETREG_UNINITIALIZED) {
6441 netdev_freemem(dev);
6445 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6446 dev->reg_state = NETREG_RELEASED;
6448 /* will free via device release */
6449 put_device(&dev->dev);
6451 EXPORT_SYMBOL(free_netdev);
6454 * synchronize_net - Synchronize with packet receive processing
6456 * Wait for packets currently being received to be done.
6457 * Does not block later packets from starting.
6459 void synchronize_net(void)
6462 if (rtnl_is_locked())
6463 synchronize_rcu_expedited();
6467 EXPORT_SYMBOL(synchronize_net);
6470 * unregister_netdevice_queue - remove device from the kernel
6474 * This function shuts down a device interface and removes it
6475 * from the kernel tables.
6476 * If head not NULL, device is queued to be unregistered later.
6478 * Callers must hold the rtnl semaphore. You may want
6479 * unregister_netdev() instead of this.
6482 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6487 list_move_tail(&dev->unreg_list, head);
6489 rollback_registered(dev);
6490 /* Finish processing unregister after unlock */
6494 EXPORT_SYMBOL(unregister_netdevice_queue);
6497 * unregister_netdevice_many - unregister many devices
6498 * @head: list of devices
6500 void unregister_netdevice_many(struct list_head *head)
6502 struct net_device *dev;
6504 if (!list_empty(head)) {
6505 rollback_registered_many(head);
6506 list_for_each_entry(dev, head, unreg_list)
6510 EXPORT_SYMBOL(unregister_netdevice_many);
6513 * unregister_netdev - remove device from the kernel
6516 * This function shuts down a device interface and removes it
6517 * from the kernel tables.
6519 * This is just a wrapper for unregister_netdevice that takes
6520 * the rtnl semaphore. In general you want to use this and not
6521 * unregister_netdevice.
6523 void unregister_netdev(struct net_device *dev)
6526 unregister_netdevice(dev);
6529 EXPORT_SYMBOL(unregister_netdev);
6532 * dev_change_net_namespace - move device to different nethost namespace
6534 * @net: network namespace
6535 * @pat: If not NULL name pattern to try if the current device name
6536 * is already taken in the destination network namespace.
6538 * This function shuts down a device interface and moves it
6539 * to a new network namespace. On success 0 is returned, on
6540 * a failure a netagive errno code is returned.
6542 * Callers must hold the rtnl semaphore.
6545 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6551 /* Don't allow namespace local devices to be moved. */
6553 if (dev->features & NETIF_F_NETNS_LOCAL)
6556 /* Ensure the device has been registrered */
6557 if (dev->reg_state != NETREG_REGISTERED)
6560 /* Get out if there is nothing todo */
6562 if (net_eq(dev_net(dev), net))
6565 /* Pick the destination device name, and ensure
6566 * we can use it in the destination network namespace.
6569 if (__dev_get_by_name(net, dev->name)) {
6570 /* We get here if we can't use the current device name */
6573 if (dev_get_valid_name(net, dev, pat) < 0)
6578 * And now a mini version of register_netdevice unregister_netdevice.
6581 /* If device is running close it first. */
6584 /* And unlink it from device chain */
6586 unlist_netdevice(dev);
6590 /* Shutdown queueing discipline. */
6593 /* Notify protocols, that we are about to destroy
6594 this device. They should clean all the things.
6596 Note that dev->reg_state stays at NETREG_REGISTERED.
6597 This is wanted because this way 8021q and macvlan know
6598 the device is just moving and can keep their slaves up.
6600 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6602 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6603 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
6606 * Flush the unicast and multicast chains
6611 /* Send a netdev-removed uevent to the old namespace */
6612 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
6614 /* Actually switch the network namespace */
6615 dev_net_set(dev, net);
6617 /* If there is an ifindex conflict assign a new one */
6618 if (__dev_get_by_index(net, dev->ifindex)) {
6619 int iflink = (dev->iflink == dev->ifindex);
6620 dev->ifindex = dev_new_index(net);
6622 dev->iflink = dev->ifindex;
6625 /* Send a netdev-add uevent to the new namespace */
6626 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
6628 /* Fixup kobjects */
6629 err = device_rename(&dev->dev, dev->name);
6632 /* Add the device back in the hashes */
6633 list_netdevice(dev);
6635 /* Notify protocols, that a new device appeared. */
6636 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6639 * Prevent userspace races by waiting until the network
6640 * device is fully setup before sending notifications.
6642 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6649 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6651 static int dev_cpu_callback(struct notifier_block *nfb,
6652 unsigned long action,
6655 struct sk_buff **list_skb;
6656 struct sk_buff *skb;
6657 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6658 struct softnet_data *sd, *oldsd;
6660 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6663 local_irq_disable();
6664 cpu = smp_processor_id();
6665 sd = &per_cpu(softnet_data, cpu);
6666 oldsd = &per_cpu(softnet_data, oldcpu);
6668 /* Find end of our completion_queue. */
6669 list_skb = &sd->completion_queue;
6671 list_skb = &(*list_skb)->next;
6672 /* Append completion queue from offline CPU. */
6673 *list_skb = oldsd->completion_queue;
6674 oldsd->completion_queue = NULL;
6676 /* Append output queue from offline CPU. */
6677 if (oldsd->output_queue) {
6678 *sd->output_queue_tailp = oldsd->output_queue;
6679 sd->output_queue_tailp = oldsd->output_queue_tailp;
6680 oldsd->output_queue = NULL;
6681 oldsd->output_queue_tailp = &oldsd->output_queue;
6683 /* Append NAPI poll list from offline CPU. */
6684 if (!list_empty(&oldsd->poll_list)) {
6685 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6686 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6689 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6692 /* Process offline CPU's input_pkt_queue */
6693 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6694 netif_rx_internal(skb);
6695 input_queue_head_incr(oldsd);
6697 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6698 netif_rx_internal(skb);
6699 input_queue_head_incr(oldsd);
6707 * netdev_increment_features - increment feature set by one
6708 * @all: current feature set
6709 * @one: new feature set
6710 * @mask: mask feature set
6712 * Computes a new feature set after adding a device with feature set
6713 * @one to the master device with current feature set @all. Will not
6714 * enable anything that is off in @mask. Returns the new feature set.
6716 netdev_features_t netdev_increment_features(netdev_features_t all,
6717 netdev_features_t one, netdev_features_t mask)
6719 if (mask & NETIF_F_GEN_CSUM)
6720 mask |= NETIF_F_ALL_CSUM;
6721 mask |= NETIF_F_VLAN_CHALLENGED;
6723 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6724 all &= one | ~NETIF_F_ALL_FOR_ALL;
6726 /* If one device supports hw checksumming, set for all. */
6727 if (all & NETIF_F_GEN_CSUM)
6728 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6732 EXPORT_SYMBOL(netdev_increment_features);
6734 static struct hlist_head * __net_init netdev_create_hash(void)
6737 struct hlist_head *hash;
6739 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6741 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6742 INIT_HLIST_HEAD(&hash[i]);
6747 /* Initialize per network namespace state */
6748 static int __net_init netdev_init(struct net *net)
6750 if (net != &init_net)
6751 INIT_LIST_HEAD(&net->dev_base_head);
6753 net->dev_name_head = netdev_create_hash();
6754 if (net->dev_name_head == NULL)
6757 net->dev_index_head = netdev_create_hash();
6758 if (net->dev_index_head == NULL)
6764 kfree(net->dev_name_head);
6770 * netdev_drivername - network driver for the device
6771 * @dev: network device
6773 * Determine network driver for device.
6775 const char *netdev_drivername(const struct net_device *dev)
6777 const struct device_driver *driver;
6778 const struct device *parent;
6779 const char *empty = "";
6781 parent = dev->dev.parent;
6785 driver = parent->driver;
6786 if (driver && driver->name)
6787 return driver->name;
6791 static int __netdev_printk(const char *level, const struct net_device *dev,
6792 struct va_format *vaf)
6796 if (dev && dev->dev.parent) {
6797 r = dev_printk_emit(level[1] - '0',
6800 dev_driver_string(dev->dev.parent),
6801 dev_name(dev->dev.parent),
6802 netdev_name(dev), vaf);
6804 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6806 r = printk("%s(NULL net_device): %pV", level, vaf);
6812 int netdev_printk(const char *level, const struct net_device *dev,
6813 const char *format, ...)
6815 struct va_format vaf;
6819 va_start(args, format);
6824 r = __netdev_printk(level, dev, &vaf);
6830 EXPORT_SYMBOL(netdev_printk);
6832 #define define_netdev_printk_level(func, level) \
6833 int func(const struct net_device *dev, const char *fmt, ...) \
6836 struct va_format vaf; \
6839 va_start(args, fmt); \
6844 r = __netdev_printk(level, dev, &vaf); \
6850 EXPORT_SYMBOL(func);
6852 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6853 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6854 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6855 define_netdev_printk_level(netdev_err, KERN_ERR);
6856 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6857 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6858 define_netdev_printk_level(netdev_info, KERN_INFO);
6860 static void __net_exit netdev_exit(struct net *net)
6862 kfree(net->dev_name_head);
6863 kfree(net->dev_index_head);
6866 static struct pernet_operations __net_initdata netdev_net_ops = {
6867 .init = netdev_init,
6868 .exit = netdev_exit,
6871 static void __net_exit default_device_exit(struct net *net)
6873 struct net_device *dev, *aux;
6875 * Push all migratable network devices back to the
6876 * initial network namespace
6879 for_each_netdev_safe(net, dev, aux) {
6881 char fb_name[IFNAMSIZ];
6883 /* Ignore unmoveable devices (i.e. loopback) */
6884 if (dev->features & NETIF_F_NETNS_LOCAL)
6887 /* Leave virtual devices for the generic cleanup */
6888 if (dev->rtnl_link_ops)
6891 /* Push remaining network devices to init_net */
6892 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6893 err = dev_change_net_namespace(dev, &init_net, fb_name);
6895 pr_emerg("%s: failed to move %s to init_net: %d\n",
6896 __func__, dev->name, err);
6903 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
6905 /* Return with the rtnl_lock held when there are no network
6906 * devices unregistering in any network namespace in net_list.
6913 prepare_to_wait(&netdev_unregistering_wq, &wait,
6914 TASK_UNINTERRUPTIBLE);
6915 unregistering = false;
6917 list_for_each_entry(net, net_list, exit_list) {
6918 if (net->dev_unreg_count > 0) {
6919 unregistering = true;
6928 finish_wait(&netdev_unregistering_wq, &wait);
6931 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6933 /* At exit all network devices most be removed from a network
6934 * namespace. Do this in the reverse order of registration.
6935 * Do this across as many network namespaces as possible to
6936 * improve batching efficiency.
6938 struct net_device *dev;
6940 LIST_HEAD(dev_kill_list);
6942 /* To prevent network device cleanup code from dereferencing
6943 * loopback devices or network devices that have been freed
6944 * wait here for all pending unregistrations to complete,
6945 * before unregistring the loopback device and allowing the
6946 * network namespace be freed.
6948 * The netdev todo list containing all network devices
6949 * unregistrations that happen in default_device_exit_batch
6950 * will run in the rtnl_unlock() at the end of
6951 * default_device_exit_batch.
6953 rtnl_lock_unregistering(net_list);
6954 list_for_each_entry(net, net_list, exit_list) {
6955 for_each_netdev_reverse(net, dev) {
6956 if (dev->rtnl_link_ops)
6957 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6959 unregister_netdevice_queue(dev, &dev_kill_list);
6962 unregister_netdevice_many(&dev_kill_list);
6963 list_del(&dev_kill_list);
6967 static struct pernet_operations __net_initdata default_device_ops = {
6968 .exit = default_device_exit,
6969 .exit_batch = default_device_exit_batch,
6973 * Initialize the DEV module. At boot time this walks the device list and
6974 * unhooks any devices that fail to initialise (normally hardware not
6975 * present) and leaves us with a valid list of present and active devices.
6980 * This is called single threaded during boot, so no need
6981 * to take the rtnl semaphore.
6983 static int __init net_dev_init(void)
6985 int i, rc = -ENOMEM;
6987 BUG_ON(!dev_boot_phase);
6989 if (dev_proc_init())
6992 if (netdev_kobject_init())
6995 INIT_LIST_HEAD(&ptype_all);
6996 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6997 INIT_LIST_HEAD(&ptype_base[i]);
6999 INIT_LIST_HEAD(&offload_base);
7001 if (register_pernet_subsys(&netdev_net_ops))
7005 * Initialise the packet receive queues.
7008 for_each_possible_cpu(i) {
7009 struct softnet_data *sd = &per_cpu(softnet_data, i);
7011 skb_queue_head_init(&sd->input_pkt_queue);
7012 skb_queue_head_init(&sd->process_queue);
7013 INIT_LIST_HEAD(&sd->poll_list);
7014 sd->output_queue_tailp = &sd->output_queue;
7016 sd->csd.func = rps_trigger_softirq;
7021 sd->backlog.poll = process_backlog;
7022 sd->backlog.weight = weight_p;
7027 /* The loopback device is special if any other network devices
7028 * is present in a network namespace the loopback device must
7029 * be present. Since we now dynamically allocate and free the
7030 * loopback device ensure this invariant is maintained by
7031 * keeping the loopback device as the first device on the
7032 * list of network devices. Ensuring the loopback devices
7033 * is the first device that appears and the last network device
7036 if (register_pernet_device(&loopback_net_ops))
7039 if (register_pernet_device(&default_device_ops))
7042 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
7043 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
7045 hotcpu_notifier(dev_cpu_callback, 0);
7052 subsys_initcall(net_dev_init);