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)
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;
2320 * skb_mac_gso_segment - mac layer segmentation handler.
2321 * @skb: buffer to segment
2322 * @features: features for the output path (see dev->features)
2324 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2325 netdev_features_t features)
2327 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2328 struct packet_offload *ptype;
2329 __be16 type = skb_network_protocol(skb);
2331 if (unlikely(!type))
2332 return ERR_PTR(-EINVAL);
2334 __skb_pull(skb, skb->mac_len);
2337 list_for_each_entry_rcu(ptype, &offload_base, list) {
2338 if (ptype->type == type && ptype->callbacks.gso_segment) {
2339 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2342 err = ptype->callbacks.gso_send_check(skb);
2343 segs = ERR_PTR(err);
2344 if (err || skb_gso_ok(skb, features))
2346 __skb_push(skb, (skb->data -
2347 skb_network_header(skb)));
2349 segs = ptype->callbacks.gso_segment(skb, features);
2355 __skb_push(skb, skb->data - skb_mac_header(skb));
2359 EXPORT_SYMBOL(skb_mac_gso_segment);
2362 /* openvswitch calls this on rx path, so we need a different check.
2364 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2367 return skb->ip_summed != CHECKSUM_PARTIAL;
2369 return skb->ip_summed == CHECKSUM_NONE;
2373 * __skb_gso_segment - Perform segmentation on skb.
2374 * @skb: buffer to segment
2375 * @features: features for the output path (see dev->features)
2376 * @tx_path: whether it is called in TX path
2378 * This function segments the given skb and returns a list of segments.
2380 * It may return NULL if the skb requires no segmentation. This is
2381 * only possible when GSO is used for verifying header integrity.
2383 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2384 netdev_features_t features, bool tx_path)
2386 if (unlikely(skb_needs_check(skb, tx_path))) {
2389 skb_warn_bad_offload(skb);
2391 if (skb_header_cloned(skb) &&
2392 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
2393 return ERR_PTR(err);
2396 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2397 SKB_GSO_CB(skb)->encap_level = 0;
2399 skb_reset_mac_header(skb);
2400 skb_reset_mac_len(skb);
2402 return skb_mac_gso_segment(skb, features);
2404 EXPORT_SYMBOL(__skb_gso_segment);
2406 /* Take action when hardware reception checksum errors are detected. */
2408 void netdev_rx_csum_fault(struct net_device *dev)
2410 if (net_ratelimit()) {
2411 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2415 EXPORT_SYMBOL(netdev_rx_csum_fault);
2418 /* Actually, we should eliminate this check as soon as we know, that:
2419 * 1. IOMMU is present and allows to map all the memory.
2420 * 2. No high memory really exists on this machine.
2423 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2425 #ifdef CONFIG_HIGHMEM
2427 if (!(dev->features & NETIF_F_HIGHDMA)) {
2428 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2429 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2430 if (PageHighMem(skb_frag_page(frag)))
2435 if (PCI_DMA_BUS_IS_PHYS) {
2436 struct device *pdev = dev->dev.parent;
2440 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2441 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2442 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2443 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2452 void (*destructor)(struct sk_buff *skb);
2455 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2457 static void dev_gso_skb_destructor(struct sk_buff *skb)
2459 struct dev_gso_cb *cb;
2461 kfree_skb_list(skb->next);
2464 cb = DEV_GSO_CB(skb);
2466 cb->destructor(skb);
2470 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2471 * @skb: buffer to segment
2472 * @features: device features as applicable to this skb
2474 * This function segments the given skb and stores the list of segments
2477 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2479 struct sk_buff *segs;
2481 segs = skb_gso_segment(skb, features);
2483 /* Verifying header integrity only. */
2488 return PTR_ERR(segs);
2491 DEV_GSO_CB(skb)->destructor = skb->destructor;
2492 skb->destructor = dev_gso_skb_destructor;
2497 static netdev_features_t harmonize_features(struct sk_buff *skb,
2498 netdev_features_t features)
2500 if (skb->ip_summed != CHECKSUM_NONE &&
2501 !can_checksum_protocol(features, skb_network_protocol(skb))) {
2502 features &= ~NETIF_F_ALL_CSUM;
2503 } else if (illegal_highdma(skb->dev, skb)) {
2504 features &= ~NETIF_F_SG;
2510 netdev_features_t netif_skb_features(struct sk_buff *skb)
2512 __be16 protocol = skb->protocol;
2513 netdev_features_t features = skb->dev->features;
2515 if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2516 features &= ~NETIF_F_GSO_MASK;
2518 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) {
2519 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2520 protocol = veh->h_vlan_encapsulated_proto;
2521 } else if (!vlan_tx_tag_present(skb)) {
2522 return harmonize_features(skb, features);
2525 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_CTAG_TX |
2526 NETIF_F_HW_VLAN_STAG_TX);
2528 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD))
2529 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2530 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_CTAG_TX |
2531 NETIF_F_HW_VLAN_STAG_TX;
2533 return harmonize_features(skb, features);
2535 EXPORT_SYMBOL(netif_skb_features);
2537 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2538 struct netdev_queue *txq)
2540 const struct net_device_ops *ops = dev->netdev_ops;
2541 int rc = NETDEV_TX_OK;
2542 unsigned int skb_len;
2544 if (likely(!skb->next)) {
2545 netdev_features_t features;
2548 * If device doesn't need skb->dst, release it right now while
2549 * its hot in this cpu cache
2551 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2554 features = netif_skb_features(skb);
2556 if (vlan_tx_tag_present(skb) &&
2557 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
2558 skb = __vlan_put_tag(skb, skb->vlan_proto,
2559 vlan_tx_tag_get(skb));
2566 /* If encapsulation offload request, verify we are testing
2567 * hardware encapsulation features instead of standard
2568 * features for the netdev
2570 if (skb->encapsulation)
2571 features &= dev->hw_enc_features;
2573 if (netif_needs_gso(skb, features)) {
2574 if (unlikely(dev_gso_segment(skb, features)))
2579 if (skb_needs_linearize(skb, features) &&
2580 __skb_linearize(skb))
2583 /* If packet is not checksummed and device does not
2584 * support checksumming for this protocol, complete
2585 * checksumming here.
2587 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2588 if (skb->encapsulation)
2589 skb_set_inner_transport_header(skb,
2590 skb_checksum_start_offset(skb));
2592 skb_set_transport_header(skb,
2593 skb_checksum_start_offset(skb));
2594 if (!(features & NETIF_F_ALL_CSUM) &&
2595 skb_checksum_help(skb))
2600 if (!list_empty(&ptype_all))
2601 dev_queue_xmit_nit(skb, dev);
2604 trace_net_dev_start_xmit(skb, dev);
2605 rc = ops->ndo_start_xmit(skb, dev);
2606 trace_net_dev_xmit(skb, rc, dev, skb_len);
2607 if (rc == NETDEV_TX_OK)
2608 txq_trans_update(txq);
2614 struct sk_buff *nskb = skb->next;
2616 skb->next = nskb->next;
2619 if (!list_empty(&ptype_all))
2620 dev_queue_xmit_nit(nskb, dev);
2622 skb_len = nskb->len;
2623 trace_net_dev_start_xmit(nskb, dev);
2624 rc = ops->ndo_start_xmit(nskb, dev);
2625 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2626 if (unlikely(rc != NETDEV_TX_OK)) {
2627 if (rc & ~NETDEV_TX_MASK)
2628 goto out_kfree_gso_skb;
2629 nskb->next = skb->next;
2633 txq_trans_update(txq);
2634 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2635 return NETDEV_TX_BUSY;
2636 } while (skb->next);
2639 if (likely(skb->next == NULL)) {
2640 skb->destructor = DEV_GSO_CB(skb)->destructor;
2649 EXPORT_SYMBOL_GPL(dev_hard_start_xmit);
2651 static void qdisc_pkt_len_init(struct sk_buff *skb)
2653 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2655 qdisc_skb_cb(skb)->pkt_len = skb->len;
2657 /* To get more precise estimation of bytes sent on wire,
2658 * we add to pkt_len the headers size of all segments
2660 if (shinfo->gso_size) {
2661 unsigned int hdr_len;
2662 u16 gso_segs = shinfo->gso_segs;
2664 /* mac layer + network layer */
2665 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2667 /* + transport layer */
2668 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2669 hdr_len += tcp_hdrlen(skb);
2671 hdr_len += sizeof(struct udphdr);
2673 if (shinfo->gso_type & SKB_GSO_DODGY)
2674 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2677 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2681 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2682 struct net_device *dev,
2683 struct netdev_queue *txq)
2685 spinlock_t *root_lock = qdisc_lock(q);
2689 qdisc_pkt_len_init(skb);
2690 qdisc_calculate_pkt_len(skb, q);
2692 * Heuristic to force contended enqueues to serialize on a
2693 * separate lock before trying to get qdisc main lock.
2694 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2695 * and dequeue packets faster.
2697 contended = qdisc_is_running(q);
2698 if (unlikely(contended))
2699 spin_lock(&q->busylock);
2701 spin_lock(root_lock);
2702 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2705 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2706 qdisc_run_begin(q)) {
2708 * This is a work-conserving queue; there are no old skbs
2709 * waiting to be sent out; and the qdisc is not running -
2710 * xmit the skb directly.
2712 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2715 qdisc_bstats_update(q, skb);
2717 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2718 if (unlikely(contended)) {
2719 spin_unlock(&q->busylock);
2726 rc = NET_XMIT_SUCCESS;
2729 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2730 if (qdisc_run_begin(q)) {
2731 if (unlikely(contended)) {
2732 spin_unlock(&q->busylock);
2738 spin_unlock(root_lock);
2739 if (unlikely(contended))
2740 spin_unlock(&q->busylock);
2744 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2745 static void skb_update_prio(struct sk_buff *skb)
2747 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2749 if (!skb->priority && skb->sk && map) {
2750 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2752 if (prioidx < map->priomap_len)
2753 skb->priority = map->priomap[prioidx];
2757 #define skb_update_prio(skb)
2760 static DEFINE_PER_CPU(int, xmit_recursion);
2761 #define RECURSION_LIMIT 10
2764 * dev_loopback_xmit - loop back @skb
2765 * @skb: buffer to transmit
2767 int dev_loopback_xmit(struct sk_buff *skb)
2769 skb_reset_mac_header(skb);
2770 __skb_pull(skb, skb_network_offset(skb));
2771 skb->pkt_type = PACKET_LOOPBACK;
2772 skb->ip_summed = CHECKSUM_UNNECESSARY;
2773 WARN_ON(!skb_dst(skb));
2778 EXPORT_SYMBOL(dev_loopback_xmit);
2781 * __dev_queue_xmit - transmit a buffer
2782 * @skb: buffer to transmit
2783 * @accel_priv: private data used for L2 forwarding offload
2785 * Queue a buffer for transmission to a network device. The caller must
2786 * have set the device and priority and built the buffer before calling
2787 * this function. The function can be called from an interrupt.
2789 * A negative errno code is returned on a failure. A success does not
2790 * guarantee the frame will be transmitted as it may be dropped due
2791 * to congestion or traffic shaping.
2793 * -----------------------------------------------------------------------------------
2794 * I notice this method can also return errors from the queue disciplines,
2795 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2798 * Regardless of the return value, the skb is consumed, so it is currently
2799 * difficult to retry a send to this method. (You can bump the ref count
2800 * before sending to hold a reference for retry if you are careful.)
2802 * When calling this method, interrupts MUST be enabled. This is because
2803 * the BH enable code must have IRQs enabled so that it will not deadlock.
2806 int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
2808 struct net_device *dev = skb->dev;
2809 struct netdev_queue *txq;
2813 skb_reset_mac_header(skb);
2815 /* Disable soft irqs for various locks below. Also
2816 * stops preemption for RCU.
2820 skb_update_prio(skb);
2822 txq = netdev_pick_tx(dev, skb, accel_priv);
2823 q = rcu_dereference_bh(txq->qdisc);
2825 #ifdef CONFIG_NET_CLS_ACT
2826 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2828 trace_net_dev_queue(skb);
2830 rc = __dev_xmit_skb(skb, q, dev, txq);
2834 /* The device has no queue. Common case for software devices:
2835 loopback, all the sorts of tunnels...
2837 Really, it is unlikely that netif_tx_lock protection is necessary
2838 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2840 However, it is possible, that they rely on protection
2843 Check this and shot the lock. It is not prone from deadlocks.
2844 Either shot noqueue qdisc, it is even simpler 8)
2846 if (dev->flags & IFF_UP) {
2847 int cpu = smp_processor_id(); /* ok because BHs are off */
2849 if (txq->xmit_lock_owner != cpu) {
2851 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2852 goto recursion_alert;
2854 HARD_TX_LOCK(dev, txq, cpu);
2856 if (!netif_xmit_stopped(txq)) {
2857 __this_cpu_inc(xmit_recursion);
2858 rc = dev_hard_start_xmit(skb, dev, txq);
2859 __this_cpu_dec(xmit_recursion);
2860 if (dev_xmit_complete(rc)) {
2861 HARD_TX_UNLOCK(dev, txq);
2865 HARD_TX_UNLOCK(dev, txq);
2866 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2869 /* Recursion is detected! It is possible,
2873 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2879 rcu_read_unlock_bh();
2884 rcu_read_unlock_bh();
2888 int dev_queue_xmit(struct sk_buff *skb)
2890 return __dev_queue_xmit(skb, NULL);
2892 EXPORT_SYMBOL(dev_queue_xmit);
2894 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
2896 return __dev_queue_xmit(skb, accel_priv);
2898 EXPORT_SYMBOL(dev_queue_xmit_accel);
2901 /*=======================================================================
2903 =======================================================================*/
2905 int netdev_max_backlog __read_mostly = 1000;
2906 EXPORT_SYMBOL(netdev_max_backlog);
2908 int netdev_tstamp_prequeue __read_mostly = 1;
2909 int netdev_budget __read_mostly = 300;
2910 int weight_p __read_mostly = 64; /* old backlog weight */
2912 /* Called with irq disabled */
2913 static inline void ____napi_schedule(struct softnet_data *sd,
2914 struct napi_struct *napi)
2916 list_add_tail(&napi->poll_list, &sd->poll_list);
2917 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2922 /* One global table that all flow-based protocols share. */
2923 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2924 EXPORT_SYMBOL(rps_sock_flow_table);
2926 struct static_key rps_needed __read_mostly;
2928 static struct rps_dev_flow *
2929 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2930 struct rps_dev_flow *rflow, u16 next_cpu)
2932 if (next_cpu != RPS_NO_CPU) {
2933 #ifdef CONFIG_RFS_ACCEL
2934 struct netdev_rx_queue *rxqueue;
2935 struct rps_dev_flow_table *flow_table;
2936 struct rps_dev_flow *old_rflow;
2941 /* Should we steer this flow to a different hardware queue? */
2942 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2943 !(dev->features & NETIF_F_NTUPLE))
2945 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2946 if (rxq_index == skb_get_rx_queue(skb))
2949 rxqueue = dev->_rx + rxq_index;
2950 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2953 flow_id = skb->rxhash & flow_table->mask;
2954 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2955 rxq_index, flow_id);
2959 rflow = &flow_table->flows[flow_id];
2961 if (old_rflow->filter == rflow->filter)
2962 old_rflow->filter = RPS_NO_FILTER;
2966 per_cpu(softnet_data, next_cpu).input_queue_head;
2969 rflow->cpu = next_cpu;
2974 * get_rps_cpu is called from netif_receive_skb and returns the target
2975 * CPU from the RPS map of the receiving queue for a given skb.
2976 * rcu_read_lock must be held on entry.
2978 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2979 struct rps_dev_flow **rflowp)
2981 struct netdev_rx_queue *rxqueue;
2982 struct rps_map *map;
2983 struct rps_dev_flow_table *flow_table;
2984 struct rps_sock_flow_table *sock_flow_table;
2988 if (skb_rx_queue_recorded(skb)) {
2989 u16 index = skb_get_rx_queue(skb);
2990 if (unlikely(index >= dev->real_num_rx_queues)) {
2991 WARN_ONCE(dev->real_num_rx_queues > 1,
2992 "%s received packet on queue %u, but number "
2993 "of RX queues is %u\n",
2994 dev->name, index, dev->real_num_rx_queues);
2997 rxqueue = dev->_rx + index;
3001 map = rcu_dereference(rxqueue->rps_map);
3003 if (map->len == 1 &&
3004 !rcu_access_pointer(rxqueue->rps_flow_table)) {
3005 tcpu = map->cpus[0];
3006 if (cpu_online(tcpu))
3010 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
3014 skb_reset_network_header(skb);
3015 if (!skb_get_hash(skb))
3018 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3019 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3020 if (flow_table && sock_flow_table) {
3022 struct rps_dev_flow *rflow;
3024 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
3027 next_cpu = sock_flow_table->ents[skb->rxhash &
3028 sock_flow_table->mask];
3031 * If the desired CPU (where last recvmsg was done) is
3032 * different from current CPU (one in the rx-queue flow
3033 * table entry), switch if one of the following holds:
3034 * - Current CPU is unset (equal to RPS_NO_CPU).
3035 * - Current CPU is offline.
3036 * - The current CPU's queue tail has advanced beyond the
3037 * last packet that was enqueued using this table entry.
3038 * This guarantees that all previous packets for the flow
3039 * have been dequeued, thus preserving in order delivery.
3041 if (unlikely(tcpu != next_cpu) &&
3042 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
3043 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3044 rflow->last_qtail)) >= 0)) {
3046 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3049 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
3057 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
3059 if (cpu_online(tcpu)) {
3069 #ifdef CONFIG_RFS_ACCEL
3072 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3073 * @dev: Device on which the filter was set
3074 * @rxq_index: RX queue index
3075 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3076 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3078 * Drivers that implement ndo_rx_flow_steer() should periodically call
3079 * this function for each installed filter and remove the filters for
3080 * which it returns %true.
3082 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3083 u32 flow_id, u16 filter_id)
3085 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3086 struct rps_dev_flow_table *flow_table;
3087 struct rps_dev_flow *rflow;
3092 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3093 if (flow_table && flow_id <= flow_table->mask) {
3094 rflow = &flow_table->flows[flow_id];
3095 cpu = ACCESS_ONCE(rflow->cpu);
3096 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3097 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3098 rflow->last_qtail) <
3099 (int)(10 * flow_table->mask)))
3105 EXPORT_SYMBOL(rps_may_expire_flow);
3107 #endif /* CONFIG_RFS_ACCEL */
3109 /* Called from hardirq (IPI) context */
3110 static void rps_trigger_softirq(void *data)
3112 struct softnet_data *sd = data;
3114 ____napi_schedule(sd, &sd->backlog);
3118 #endif /* CONFIG_RPS */
3121 * Check if this softnet_data structure is another cpu one
3122 * If yes, queue it to our IPI list and return 1
3125 static int rps_ipi_queued(struct softnet_data *sd)
3128 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
3131 sd->rps_ipi_next = mysd->rps_ipi_list;
3132 mysd->rps_ipi_list = sd;
3134 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3137 #endif /* CONFIG_RPS */
3141 #ifdef CONFIG_NET_FLOW_LIMIT
3142 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3145 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3147 #ifdef CONFIG_NET_FLOW_LIMIT
3148 struct sd_flow_limit *fl;
3149 struct softnet_data *sd;
3150 unsigned int old_flow, new_flow;
3152 if (qlen < (netdev_max_backlog >> 1))
3155 sd = &__get_cpu_var(softnet_data);
3158 fl = rcu_dereference(sd->flow_limit);
3160 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3161 old_flow = fl->history[fl->history_head];
3162 fl->history[fl->history_head] = new_flow;
3165 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3167 if (likely(fl->buckets[old_flow]))
3168 fl->buckets[old_flow]--;
3170 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3182 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3183 * queue (may be a remote CPU queue).
3185 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3186 unsigned int *qtail)
3188 struct softnet_data *sd;
3189 unsigned long flags;
3192 sd = &per_cpu(softnet_data, cpu);
3194 local_irq_save(flags);
3197 qlen = skb_queue_len(&sd->input_pkt_queue);
3198 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3199 if (skb_queue_len(&sd->input_pkt_queue)) {
3201 __skb_queue_tail(&sd->input_pkt_queue, skb);
3202 input_queue_tail_incr_save(sd, qtail);
3204 local_irq_restore(flags);
3205 return NET_RX_SUCCESS;
3208 /* Schedule NAPI for backlog device
3209 * We can use non atomic operation since we own the queue lock
3211 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3212 if (!rps_ipi_queued(sd))
3213 ____napi_schedule(sd, &sd->backlog);
3221 local_irq_restore(flags);
3223 atomic_long_inc(&skb->dev->rx_dropped);
3228 static int netif_rx_internal(struct sk_buff *skb)
3232 /* if netpoll wants it, pretend we never saw it */
3233 if (netpoll_rx(skb))
3236 net_timestamp_check(netdev_tstamp_prequeue, skb);
3238 trace_netif_rx(skb);
3240 if (static_key_false(&rps_needed)) {
3241 struct rps_dev_flow voidflow, *rflow = &voidflow;
3247 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3249 cpu = smp_processor_id();
3251 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3259 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3266 * netif_rx - post buffer to the network code
3267 * @skb: buffer to post
3269 * This function receives a packet from a device driver and queues it for
3270 * the upper (protocol) levels to process. It always succeeds. The buffer
3271 * may be dropped during processing for congestion control or by the
3275 * NET_RX_SUCCESS (no congestion)
3276 * NET_RX_DROP (packet was dropped)
3280 int netif_rx(struct sk_buff *skb)
3282 trace_netif_rx_entry(skb);
3284 return netif_rx_internal(skb);
3286 EXPORT_SYMBOL(netif_rx);
3288 int netif_rx_ni(struct sk_buff *skb)
3292 trace_netif_rx_ni_entry(skb);
3295 err = netif_rx_internal(skb);
3296 if (local_softirq_pending())
3302 EXPORT_SYMBOL(netif_rx_ni);
3304 static void net_tx_action(struct softirq_action *h)
3306 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3308 if (sd->completion_queue) {
3309 struct sk_buff *clist;
3311 local_irq_disable();
3312 clist = sd->completion_queue;
3313 sd->completion_queue = NULL;
3317 struct sk_buff *skb = clist;
3318 clist = clist->next;
3320 WARN_ON(atomic_read(&skb->users));
3321 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3322 trace_consume_skb(skb);
3324 trace_kfree_skb(skb, net_tx_action);
3329 if (sd->output_queue) {
3332 local_irq_disable();
3333 head = sd->output_queue;
3334 sd->output_queue = NULL;
3335 sd->output_queue_tailp = &sd->output_queue;
3339 struct Qdisc *q = head;
3340 spinlock_t *root_lock;
3342 head = head->next_sched;
3344 root_lock = qdisc_lock(q);
3345 if (spin_trylock(root_lock)) {
3346 smp_mb__before_clear_bit();
3347 clear_bit(__QDISC_STATE_SCHED,
3350 spin_unlock(root_lock);
3352 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3354 __netif_reschedule(q);
3356 smp_mb__before_clear_bit();
3357 clear_bit(__QDISC_STATE_SCHED,
3365 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3366 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3367 /* This hook is defined here for ATM LANE */
3368 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3369 unsigned char *addr) __read_mostly;
3370 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3373 #ifdef CONFIG_NET_CLS_ACT
3374 /* TODO: Maybe we should just force sch_ingress to be compiled in
3375 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3376 * a compare and 2 stores extra right now if we dont have it on
3377 * but have CONFIG_NET_CLS_ACT
3378 * NOTE: This doesn't stop any functionality; if you dont have
3379 * the ingress scheduler, you just can't add policies on ingress.
3382 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3384 struct net_device *dev = skb->dev;
3385 u32 ttl = G_TC_RTTL(skb->tc_verd);
3386 int result = TC_ACT_OK;
3389 if (unlikely(MAX_RED_LOOP < ttl++)) {
3390 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3391 skb->skb_iif, dev->ifindex);
3395 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3396 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3399 if (q != &noop_qdisc) {
3400 spin_lock(qdisc_lock(q));
3401 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3402 result = qdisc_enqueue_root(skb, q);
3403 spin_unlock(qdisc_lock(q));
3409 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3410 struct packet_type **pt_prev,
3411 int *ret, struct net_device *orig_dev)
3413 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3415 if (!rxq || rxq->qdisc == &noop_qdisc)
3419 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3423 switch (ing_filter(skb, rxq)) {
3437 * netdev_rx_handler_register - register receive handler
3438 * @dev: device to register a handler for
3439 * @rx_handler: receive handler to register
3440 * @rx_handler_data: data pointer that is used by rx handler
3442 * Register a receive hander for a device. This handler will then be
3443 * called from __netif_receive_skb. A negative errno code is returned
3446 * The caller must hold the rtnl_mutex.
3448 * For a general description of rx_handler, see enum rx_handler_result.
3450 int netdev_rx_handler_register(struct net_device *dev,
3451 rx_handler_func_t *rx_handler,
3452 void *rx_handler_data)
3456 if (dev->rx_handler)
3459 /* Note: rx_handler_data must be set before rx_handler */
3460 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3461 rcu_assign_pointer(dev->rx_handler, rx_handler);
3465 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3468 * netdev_rx_handler_unregister - unregister receive handler
3469 * @dev: device to unregister a handler from
3471 * Unregister a receive handler from a device.
3473 * The caller must hold the rtnl_mutex.
3475 void netdev_rx_handler_unregister(struct net_device *dev)
3479 RCU_INIT_POINTER(dev->rx_handler, NULL);
3480 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3481 * section has a guarantee to see a non NULL rx_handler_data
3485 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3487 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3490 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3491 * the special handling of PFMEMALLOC skbs.
3493 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3495 switch (skb->protocol) {
3496 case __constant_htons(ETH_P_ARP):
3497 case __constant_htons(ETH_P_IP):
3498 case __constant_htons(ETH_P_IPV6):
3499 case __constant_htons(ETH_P_8021Q):
3500 case __constant_htons(ETH_P_8021AD):
3507 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3509 struct packet_type *ptype, *pt_prev;
3510 rx_handler_func_t *rx_handler;
3511 struct net_device *orig_dev;
3512 struct net_device *null_or_dev;
3513 bool deliver_exact = false;
3514 int ret = NET_RX_DROP;
3517 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3519 trace_netif_receive_skb(skb);
3521 /* if we've gotten here through NAPI, check netpoll */
3522 if (netpoll_receive_skb(skb))
3525 orig_dev = skb->dev;
3527 skb_reset_network_header(skb);
3528 if (!skb_transport_header_was_set(skb))
3529 skb_reset_transport_header(skb);
3530 skb_reset_mac_len(skb);
3537 skb->skb_iif = skb->dev->ifindex;
3539 __this_cpu_inc(softnet_data.processed);
3541 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3542 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3543 skb = vlan_untag(skb);
3548 #ifdef CONFIG_NET_CLS_ACT
3549 if (skb->tc_verd & TC_NCLS) {
3550 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3558 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3559 if (!ptype->dev || ptype->dev == skb->dev) {
3561 ret = deliver_skb(skb, pt_prev, orig_dev);
3567 #ifdef CONFIG_NET_CLS_ACT
3568 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3574 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3577 if (vlan_tx_tag_present(skb)) {
3579 ret = deliver_skb(skb, pt_prev, orig_dev);
3582 if (vlan_do_receive(&skb))
3584 else if (unlikely(!skb))
3588 rx_handler = rcu_dereference(skb->dev->rx_handler);
3591 ret = deliver_skb(skb, pt_prev, orig_dev);
3594 switch (rx_handler(&skb)) {
3595 case RX_HANDLER_CONSUMED:
3596 ret = NET_RX_SUCCESS;
3598 case RX_HANDLER_ANOTHER:
3600 case RX_HANDLER_EXACT:
3601 deliver_exact = true;
3602 case RX_HANDLER_PASS:
3609 if (unlikely(vlan_tx_tag_present(skb))) {
3610 if (vlan_tx_tag_get_id(skb))
3611 skb->pkt_type = PACKET_OTHERHOST;
3612 /* Note: we might in the future use prio bits
3613 * and set skb->priority like in vlan_do_receive()
3614 * For the time being, just ignore Priority Code Point
3619 /* deliver only exact match when indicated */
3620 null_or_dev = deliver_exact ? skb->dev : NULL;
3622 type = skb->protocol;
3623 list_for_each_entry_rcu(ptype,
3624 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3625 if (ptype->type == type &&
3626 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3627 ptype->dev == orig_dev)) {
3629 ret = deliver_skb(skb, pt_prev, orig_dev);
3635 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3638 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3641 atomic_long_inc(&skb->dev->rx_dropped);
3643 /* Jamal, now you will not able to escape explaining
3644 * me how you were going to use this. :-)
3655 static int __netif_receive_skb(struct sk_buff *skb)
3659 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3660 unsigned long pflags = current->flags;
3663 * PFMEMALLOC skbs are special, they should
3664 * - be delivered to SOCK_MEMALLOC sockets only
3665 * - stay away from userspace
3666 * - have bounded memory usage
3668 * Use PF_MEMALLOC as this saves us from propagating the allocation
3669 * context down to all allocation sites.
3671 current->flags |= PF_MEMALLOC;
3672 ret = __netif_receive_skb_core(skb, true);
3673 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3675 ret = __netif_receive_skb_core(skb, false);
3680 static int netif_receive_skb_internal(struct sk_buff *skb)
3682 net_timestamp_check(netdev_tstamp_prequeue, skb);
3684 if (skb_defer_rx_timestamp(skb))
3685 return NET_RX_SUCCESS;
3688 if (static_key_false(&rps_needed)) {
3689 struct rps_dev_flow voidflow, *rflow = &voidflow;
3694 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3697 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3704 return __netif_receive_skb(skb);
3708 * netif_receive_skb - process receive buffer from network
3709 * @skb: buffer to process
3711 * netif_receive_skb() is the main receive data processing function.
3712 * It always succeeds. The buffer may be dropped during processing
3713 * for congestion control or by the protocol layers.
3715 * This function may only be called from softirq context and interrupts
3716 * should be enabled.
3718 * Return values (usually ignored):
3719 * NET_RX_SUCCESS: no congestion
3720 * NET_RX_DROP: packet was dropped
3722 int netif_receive_skb(struct sk_buff *skb)
3724 trace_netif_receive_skb_entry(skb);
3726 return netif_receive_skb_internal(skb);
3728 EXPORT_SYMBOL(netif_receive_skb);
3730 /* Network device is going away, flush any packets still pending
3731 * Called with irqs disabled.
3733 static void flush_backlog(void *arg)
3735 struct net_device *dev = arg;
3736 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3737 struct sk_buff *skb, *tmp;
3740 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3741 if (skb->dev == dev) {
3742 __skb_unlink(skb, &sd->input_pkt_queue);
3744 input_queue_head_incr(sd);
3749 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3750 if (skb->dev == dev) {
3751 __skb_unlink(skb, &sd->process_queue);
3753 input_queue_head_incr(sd);
3758 static int napi_gro_complete(struct sk_buff *skb)
3760 struct packet_offload *ptype;
3761 __be16 type = skb->protocol;
3762 struct list_head *head = &offload_base;
3765 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3767 if (NAPI_GRO_CB(skb)->count == 1) {
3768 skb_shinfo(skb)->gso_size = 0;
3773 list_for_each_entry_rcu(ptype, head, list) {
3774 if (ptype->type != type || !ptype->callbacks.gro_complete)
3777 err = ptype->callbacks.gro_complete(skb, 0);
3783 WARN_ON(&ptype->list == head);
3785 return NET_RX_SUCCESS;
3789 return netif_receive_skb_internal(skb);
3792 /* napi->gro_list contains packets ordered by age.
3793 * youngest packets at the head of it.
3794 * Complete skbs in reverse order to reduce latencies.
3796 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3798 struct sk_buff *skb, *prev = NULL;
3800 /* scan list and build reverse chain */
3801 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3806 for (skb = prev; skb; skb = prev) {
3809 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3813 napi_gro_complete(skb);
3817 napi->gro_list = NULL;
3819 EXPORT_SYMBOL(napi_gro_flush);
3821 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3824 unsigned int maclen = skb->dev->hard_header_len;
3825 u32 hash = skb_get_hash_raw(skb);
3827 for (p = napi->gro_list; p; p = p->next) {
3828 unsigned long diffs;
3830 NAPI_GRO_CB(p)->flush = 0;
3832 if (hash != skb_get_hash_raw(p)) {
3833 NAPI_GRO_CB(p)->same_flow = 0;
3837 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3838 diffs |= p->vlan_tci ^ skb->vlan_tci;
3839 if (maclen == ETH_HLEN)
3840 diffs |= compare_ether_header(skb_mac_header(p),
3841 skb_gro_mac_header(skb));
3843 diffs = memcmp(skb_mac_header(p),
3844 skb_gro_mac_header(skb),
3846 NAPI_GRO_CB(p)->same_flow = !diffs;
3850 static void skb_gro_reset_offset(struct sk_buff *skb)
3852 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3853 const skb_frag_t *frag0 = &pinfo->frags[0];
3855 NAPI_GRO_CB(skb)->data_offset = 0;
3856 NAPI_GRO_CB(skb)->frag0 = NULL;
3857 NAPI_GRO_CB(skb)->frag0_len = 0;
3859 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
3861 !PageHighMem(skb_frag_page(frag0))) {
3862 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3863 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3867 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3869 struct sk_buff **pp = NULL;
3870 struct packet_offload *ptype;
3871 __be16 type = skb->protocol;
3872 struct list_head *head = &offload_base;
3874 enum gro_result ret;
3876 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3879 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3882 skb_gro_reset_offset(skb);
3883 gro_list_prepare(napi, skb);
3884 NAPI_GRO_CB(skb)->csum = skb->csum; /* Needed for CHECKSUM_COMPLETE */
3887 list_for_each_entry_rcu(ptype, head, list) {
3888 if (ptype->type != type || !ptype->callbacks.gro_receive)
3891 skb_set_network_header(skb, skb_gro_offset(skb));
3892 skb_reset_mac_len(skb);
3893 NAPI_GRO_CB(skb)->same_flow = 0;
3894 NAPI_GRO_CB(skb)->flush = 0;
3895 NAPI_GRO_CB(skb)->free = 0;
3897 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
3902 if (&ptype->list == head)
3905 same_flow = NAPI_GRO_CB(skb)->same_flow;
3906 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3909 struct sk_buff *nskb = *pp;
3913 napi_gro_complete(nskb);
3920 if (NAPI_GRO_CB(skb)->flush)
3923 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
3924 struct sk_buff *nskb = napi->gro_list;
3926 /* locate the end of the list to select the 'oldest' flow */
3927 while (nskb->next) {
3933 napi_gro_complete(nskb);
3937 NAPI_GRO_CB(skb)->count = 1;
3938 NAPI_GRO_CB(skb)->age = jiffies;
3939 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3940 skb->next = napi->gro_list;
3941 napi->gro_list = skb;
3945 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3946 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3948 BUG_ON(skb->end - skb->tail < grow);
3950 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3953 skb->data_len -= grow;
3955 skb_shinfo(skb)->frags[0].page_offset += grow;
3956 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3958 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3959 skb_frag_unref(skb, 0);
3960 memmove(skb_shinfo(skb)->frags,
3961 skb_shinfo(skb)->frags + 1,
3962 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3974 struct packet_offload *gro_find_receive_by_type(__be16 type)
3976 struct list_head *offload_head = &offload_base;
3977 struct packet_offload *ptype;
3979 list_for_each_entry_rcu(ptype, offload_head, list) {
3980 if (ptype->type != type || !ptype->callbacks.gro_receive)
3987 struct packet_offload *gro_find_complete_by_type(__be16 type)
3989 struct list_head *offload_head = &offload_base;
3990 struct packet_offload *ptype;
3992 list_for_each_entry_rcu(ptype, offload_head, list) {
3993 if (ptype->type != type || !ptype->callbacks.gro_complete)
4000 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4004 if (netif_receive_skb_internal(skb))
4012 case GRO_MERGED_FREE:
4013 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4014 kmem_cache_free(skbuff_head_cache, skb);
4027 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4029 trace_napi_gro_receive_entry(skb);
4031 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4033 EXPORT_SYMBOL(napi_gro_receive);
4035 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4037 __skb_pull(skb, skb_headlen(skb));
4038 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4039 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4041 skb->dev = napi->dev;
4047 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4049 struct sk_buff *skb = napi->skb;
4052 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
4057 EXPORT_SYMBOL(napi_get_frags);
4059 static gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
4064 if (netif_receive_skb_internal(skb))
4069 case GRO_MERGED_FREE:
4070 napi_reuse_skb(napi, skb);
4081 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4083 struct sk_buff *skb = napi->skb;
4087 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) {
4088 napi_reuse_skb(napi, skb);
4091 skb->protocol = eth_type_trans(skb, skb->dev);
4096 gro_result_t napi_gro_frags(struct napi_struct *napi)
4098 struct sk_buff *skb = napi_frags_skb(napi);
4103 trace_napi_gro_frags_entry(skb);
4105 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4107 EXPORT_SYMBOL(napi_gro_frags);
4110 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4111 * Note: called with local irq disabled, but exits with local irq enabled.
4113 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4116 struct softnet_data *remsd = sd->rps_ipi_list;
4119 sd->rps_ipi_list = NULL;
4123 /* Send pending IPI's to kick RPS processing on remote cpus. */
4125 struct softnet_data *next = remsd->rps_ipi_next;
4127 if (cpu_online(remsd->cpu))
4128 __smp_call_function_single(remsd->cpu,
4137 static int process_backlog(struct napi_struct *napi, int quota)
4140 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4143 /* Check if we have pending ipi, its better to send them now,
4144 * not waiting net_rx_action() end.
4146 if (sd->rps_ipi_list) {
4147 local_irq_disable();
4148 net_rps_action_and_irq_enable(sd);
4151 napi->weight = weight_p;
4152 local_irq_disable();
4153 while (work < quota) {
4154 struct sk_buff *skb;
4157 while ((skb = __skb_dequeue(&sd->process_queue))) {
4159 __netif_receive_skb(skb);
4160 local_irq_disable();
4161 input_queue_head_incr(sd);
4162 if (++work >= quota) {
4169 qlen = skb_queue_len(&sd->input_pkt_queue);
4171 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4172 &sd->process_queue);
4174 if (qlen < quota - work) {
4176 * Inline a custom version of __napi_complete().
4177 * only current cpu owns and manipulates this napi,
4178 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
4179 * we can use a plain write instead of clear_bit(),
4180 * and we dont need an smp_mb() memory barrier.
4182 list_del(&napi->poll_list);
4185 quota = work + qlen;
4195 * __napi_schedule - schedule for receive
4196 * @n: entry to schedule
4198 * The entry's receive function will be scheduled to run
4200 void __napi_schedule(struct napi_struct *n)
4202 unsigned long flags;
4204 local_irq_save(flags);
4205 ____napi_schedule(&__get_cpu_var(softnet_data), n);
4206 local_irq_restore(flags);
4208 EXPORT_SYMBOL(__napi_schedule);
4210 void __napi_complete(struct napi_struct *n)
4212 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4213 BUG_ON(n->gro_list);
4215 list_del(&n->poll_list);
4216 smp_mb__before_clear_bit();
4217 clear_bit(NAPI_STATE_SCHED, &n->state);
4219 EXPORT_SYMBOL(__napi_complete);
4221 void napi_complete(struct napi_struct *n)
4223 unsigned long flags;
4226 * don't let napi dequeue from the cpu poll list
4227 * just in case its running on a different cpu
4229 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4232 napi_gro_flush(n, false);
4233 local_irq_save(flags);
4235 local_irq_restore(flags);
4237 EXPORT_SYMBOL(napi_complete);
4239 /* must be called under rcu_read_lock(), as we dont take a reference */
4240 struct napi_struct *napi_by_id(unsigned int napi_id)
4242 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4243 struct napi_struct *napi;
4245 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4246 if (napi->napi_id == napi_id)
4251 EXPORT_SYMBOL_GPL(napi_by_id);
4253 void napi_hash_add(struct napi_struct *napi)
4255 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4257 spin_lock(&napi_hash_lock);
4259 /* 0 is not a valid id, we also skip an id that is taken
4260 * we expect both events to be extremely rare
4263 while (!napi->napi_id) {
4264 napi->napi_id = ++napi_gen_id;
4265 if (napi_by_id(napi->napi_id))
4269 hlist_add_head_rcu(&napi->napi_hash_node,
4270 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4272 spin_unlock(&napi_hash_lock);
4275 EXPORT_SYMBOL_GPL(napi_hash_add);
4277 /* Warning : caller is responsible to make sure rcu grace period
4278 * is respected before freeing memory containing @napi
4280 void napi_hash_del(struct napi_struct *napi)
4282 spin_lock(&napi_hash_lock);
4284 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4285 hlist_del_rcu(&napi->napi_hash_node);
4287 spin_unlock(&napi_hash_lock);
4289 EXPORT_SYMBOL_GPL(napi_hash_del);
4291 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4292 int (*poll)(struct napi_struct *, int), int weight)
4294 INIT_LIST_HEAD(&napi->poll_list);
4295 napi->gro_count = 0;
4296 napi->gro_list = NULL;
4299 if (weight > NAPI_POLL_WEIGHT)
4300 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4302 napi->weight = weight;
4303 list_add(&napi->dev_list, &dev->napi_list);
4305 #ifdef CONFIG_NETPOLL
4306 spin_lock_init(&napi->poll_lock);
4307 napi->poll_owner = -1;
4309 set_bit(NAPI_STATE_SCHED, &napi->state);
4311 EXPORT_SYMBOL(netif_napi_add);
4313 void netif_napi_del(struct napi_struct *napi)
4315 list_del_init(&napi->dev_list);
4316 napi_free_frags(napi);
4318 kfree_skb_list(napi->gro_list);
4319 napi->gro_list = NULL;
4320 napi->gro_count = 0;
4322 EXPORT_SYMBOL(netif_napi_del);
4324 static void net_rx_action(struct softirq_action *h)
4326 struct softnet_data *sd = &__get_cpu_var(softnet_data);
4327 unsigned long time_limit = jiffies + 2;
4328 int budget = netdev_budget;
4331 local_irq_disable();
4333 while (!list_empty(&sd->poll_list)) {
4334 struct napi_struct *n;
4337 /* If softirq window is exhuasted then punt.
4338 * Allow this to run for 2 jiffies since which will allow
4339 * an average latency of 1.5/HZ.
4341 if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit)))
4346 /* Even though interrupts have been re-enabled, this
4347 * access is safe because interrupts can only add new
4348 * entries to the tail of this list, and only ->poll()
4349 * calls can remove this head entry from the list.
4351 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
4353 have = netpoll_poll_lock(n);
4357 /* This NAPI_STATE_SCHED test is for avoiding a race
4358 * with netpoll's poll_napi(). Only the entity which
4359 * obtains the lock and sees NAPI_STATE_SCHED set will
4360 * actually make the ->poll() call. Therefore we avoid
4361 * accidentally calling ->poll() when NAPI is not scheduled.
4364 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4365 work = n->poll(n, weight);
4369 WARN_ON_ONCE(work > weight);
4373 local_irq_disable();
4375 /* Drivers must not modify the NAPI state if they
4376 * consume the entire weight. In such cases this code
4377 * still "owns" the NAPI instance and therefore can
4378 * move the instance around on the list at-will.
4380 if (unlikely(work == weight)) {
4381 if (unlikely(napi_disable_pending(n))) {
4384 local_irq_disable();
4387 /* flush too old packets
4388 * If HZ < 1000, flush all packets.
4391 napi_gro_flush(n, HZ >= 1000);
4392 local_irq_disable();
4394 list_move_tail(&n->poll_list, &sd->poll_list);
4398 netpoll_poll_unlock(have);
4401 net_rps_action_and_irq_enable(sd);
4403 #ifdef CONFIG_NET_DMA
4405 * There may not be any more sk_buffs coming right now, so push
4406 * any pending DMA copies to hardware
4408 dma_issue_pending_all();
4415 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4419 struct netdev_adjacent {
4420 struct net_device *dev;
4422 /* upper master flag, there can only be one master device per list */
4425 /* counter for the number of times this device was added to us */
4428 /* private field for the users */
4431 struct list_head list;
4432 struct rcu_head rcu;
4435 static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev,
4436 struct net_device *adj_dev,
4437 struct list_head *adj_list)
4439 struct netdev_adjacent *adj;
4441 list_for_each_entry(adj, adj_list, list) {
4442 if (adj->dev == adj_dev)
4449 * netdev_has_upper_dev - Check if device is linked to an upper device
4451 * @upper_dev: upper device to check
4453 * Find out if a device is linked to specified upper device and return true
4454 * in case it is. Note that this checks only immediate upper device,
4455 * not through a complete stack of devices. The caller must hold the RTNL lock.
4457 bool netdev_has_upper_dev(struct net_device *dev,
4458 struct net_device *upper_dev)
4462 return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper);
4464 EXPORT_SYMBOL(netdev_has_upper_dev);
4467 * netdev_has_any_upper_dev - Check if device is linked to some device
4470 * Find out if a device is linked to an upper device and return true in case
4471 * it is. The caller must hold the RTNL lock.
4473 static bool netdev_has_any_upper_dev(struct net_device *dev)
4477 return !list_empty(&dev->all_adj_list.upper);
4481 * netdev_master_upper_dev_get - Get master upper device
4484 * Find a master upper device and return pointer to it or NULL in case
4485 * it's not there. The caller must hold the RTNL lock.
4487 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4489 struct netdev_adjacent *upper;
4493 if (list_empty(&dev->adj_list.upper))
4496 upper = list_first_entry(&dev->adj_list.upper,
4497 struct netdev_adjacent, list);
4498 if (likely(upper->master))
4502 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4504 void *netdev_adjacent_get_private(struct list_head *adj_list)
4506 struct netdev_adjacent *adj;
4508 adj = list_entry(adj_list, struct netdev_adjacent, list);
4510 return adj->private;
4512 EXPORT_SYMBOL(netdev_adjacent_get_private);
4515 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4517 * @iter: list_head ** of the current position
4519 * Gets the next device from the dev's upper list, starting from iter
4520 * position. The caller must hold RCU read lock.
4522 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
4523 struct list_head **iter)
4525 struct netdev_adjacent *upper;
4527 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4529 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4531 if (&upper->list == &dev->all_adj_list.upper)
4534 *iter = &upper->list;
4538 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
4541 * netdev_lower_get_next_private - Get the next ->private from the
4542 * lower neighbour list
4544 * @iter: list_head ** of the current position
4546 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4547 * list, starting from iter position. The caller must hold either hold the
4548 * RTNL lock or its own locking that guarantees that the neighbour lower
4549 * list will remain unchainged.
4551 void *netdev_lower_get_next_private(struct net_device *dev,
4552 struct list_head **iter)
4554 struct netdev_adjacent *lower;
4556 lower = list_entry(*iter, struct netdev_adjacent, list);
4558 if (&lower->list == &dev->adj_list.lower)
4562 *iter = lower->list.next;
4564 return lower->private;
4566 EXPORT_SYMBOL(netdev_lower_get_next_private);
4569 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4570 * lower neighbour list, RCU
4573 * @iter: list_head ** of the current position
4575 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4576 * list, starting from iter position. The caller must hold RCU read lock.
4578 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
4579 struct list_head **iter)
4581 struct netdev_adjacent *lower;
4583 WARN_ON_ONCE(!rcu_read_lock_held());
4585 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4587 if (&lower->list == &dev->adj_list.lower)
4591 *iter = &lower->list;
4593 return lower->private;
4595 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
4598 * netdev_lower_get_first_private_rcu - Get the first ->private from the
4599 * lower neighbour list, RCU
4603 * Gets the first netdev_adjacent->private from the dev's lower neighbour
4604 * list. The caller must hold RCU read lock.
4606 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
4608 struct netdev_adjacent *lower;
4610 lower = list_first_or_null_rcu(&dev->adj_list.lower,
4611 struct netdev_adjacent, list);
4613 return lower->private;
4616 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
4619 * netdev_master_upper_dev_get_rcu - Get master upper device
4622 * Find a master upper device and return pointer to it or NULL in case
4623 * it's not there. The caller must hold the RCU read lock.
4625 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4627 struct netdev_adjacent *upper;
4629 upper = list_first_or_null_rcu(&dev->adj_list.upper,
4630 struct netdev_adjacent, list);
4631 if (upper && likely(upper->master))
4635 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4637 int netdev_adjacent_sysfs_add(struct net_device *dev,
4638 struct net_device *adj_dev,
4639 struct list_head *dev_list)
4641 char linkname[IFNAMSIZ+7];
4642 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4643 "upper_%s" : "lower_%s", adj_dev->name);
4644 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
4647 void netdev_adjacent_sysfs_del(struct net_device *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", name);
4654 sysfs_remove_link(&(dev->dev.kobj), linkname);
4657 #define netdev_adjacent_is_neigh_list(dev, dev_list) \
4658 (dev_list == &dev->adj_list.upper || \
4659 dev_list == &dev->adj_list.lower)
4661 static int __netdev_adjacent_dev_insert(struct net_device *dev,
4662 struct net_device *adj_dev,
4663 struct list_head *dev_list,
4664 void *private, bool master)
4666 struct netdev_adjacent *adj;
4669 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4676 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
4681 adj->master = master;
4683 adj->private = private;
4686 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
4687 adj_dev->name, dev->name, adj_dev->name);
4689 if (netdev_adjacent_is_neigh_list(dev, dev_list)) {
4690 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
4695 /* Ensure that master link is always the first item in list. */
4697 ret = sysfs_create_link(&(dev->dev.kobj),
4698 &(adj_dev->dev.kobj), "master");
4700 goto remove_symlinks;
4702 list_add_rcu(&adj->list, dev_list);
4704 list_add_tail_rcu(&adj->list, dev_list);
4710 if (netdev_adjacent_is_neigh_list(dev, dev_list))
4711 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
4719 static void __netdev_adjacent_dev_remove(struct net_device *dev,
4720 struct net_device *adj_dev,
4721 struct list_head *dev_list)
4723 struct netdev_adjacent *adj;
4725 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4728 pr_err("tried to remove device %s from %s\n",
4729 dev->name, adj_dev->name);
4733 if (adj->ref_nr > 1) {
4734 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
4741 sysfs_remove_link(&(dev->dev.kobj), "master");
4743 if (netdev_adjacent_is_neigh_list(dev, dev_list))
4744 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
4746 list_del_rcu(&adj->list);
4747 pr_debug("dev_put for %s, because link removed from %s to %s\n",
4748 adj_dev->name, dev->name, adj_dev->name);
4750 kfree_rcu(adj, rcu);
4753 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
4754 struct net_device *upper_dev,
4755 struct list_head *up_list,
4756 struct list_head *down_list,
4757 void *private, bool master)
4761 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
4766 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
4769 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4776 static int __netdev_adjacent_dev_link(struct net_device *dev,
4777 struct net_device *upper_dev)
4779 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
4780 &dev->all_adj_list.upper,
4781 &upper_dev->all_adj_list.lower,
4785 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
4786 struct net_device *upper_dev,
4787 struct list_head *up_list,
4788 struct list_head *down_list)
4790 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4791 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
4794 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
4795 struct net_device *upper_dev)
4797 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
4798 &dev->all_adj_list.upper,
4799 &upper_dev->all_adj_list.lower);
4802 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
4803 struct net_device *upper_dev,
4804 void *private, bool master)
4806 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
4811 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
4812 &dev->adj_list.upper,
4813 &upper_dev->adj_list.lower,
4816 __netdev_adjacent_dev_unlink(dev, upper_dev);
4823 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
4824 struct net_device *upper_dev)
4826 __netdev_adjacent_dev_unlink(dev, upper_dev);
4827 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
4828 &dev->adj_list.upper,
4829 &upper_dev->adj_list.lower);
4832 static int __netdev_upper_dev_link(struct net_device *dev,
4833 struct net_device *upper_dev, bool master,
4836 struct netdev_adjacent *i, *j, *to_i, *to_j;
4841 if (dev == upper_dev)
4844 /* To prevent loops, check if dev is not upper device to upper_dev. */
4845 if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper))
4848 if (__netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper))
4851 if (master && netdev_master_upper_dev_get(dev))
4854 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
4859 /* Now that we linked these devs, make all the upper_dev's
4860 * all_adj_list.upper visible to every dev's all_adj_list.lower an
4861 * versa, and don't forget the devices itself. All of these
4862 * links are non-neighbours.
4864 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4865 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
4866 pr_debug("Interlinking %s with %s, non-neighbour\n",
4867 i->dev->name, j->dev->name);
4868 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
4874 /* add dev to every upper_dev's upper device */
4875 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
4876 pr_debug("linking %s's upper device %s with %s\n",
4877 upper_dev->name, i->dev->name, dev->name);
4878 ret = __netdev_adjacent_dev_link(dev, i->dev);
4880 goto rollback_upper_mesh;
4883 /* add upper_dev to every dev's lower device */
4884 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4885 pr_debug("linking %s's lower device %s with %s\n", dev->name,
4886 i->dev->name, upper_dev->name);
4887 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
4889 goto rollback_lower_mesh;
4892 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
4895 rollback_lower_mesh:
4897 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4900 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
4905 rollback_upper_mesh:
4907 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
4910 __netdev_adjacent_dev_unlink(dev, i->dev);
4918 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4919 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
4920 if (i == to_i && j == to_j)
4922 __netdev_adjacent_dev_unlink(i->dev, j->dev);
4928 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
4934 * netdev_upper_dev_link - Add a link to the upper device
4936 * @upper_dev: new upper device
4938 * Adds a link to device which is upper to this one. The caller must hold
4939 * the RTNL lock. On a failure a negative errno code is returned.
4940 * On success the reference counts are adjusted and the function
4943 int netdev_upper_dev_link(struct net_device *dev,
4944 struct net_device *upper_dev)
4946 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
4948 EXPORT_SYMBOL(netdev_upper_dev_link);
4951 * netdev_master_upper_dev_link - Add a master link to the upper device
4953 * @upper_dev: new upper device
4955 * Adds a link to device which is upper to this one. In this case, only
4956 * one master upper device can be linked, although other non-master devices
4957 * might be linked as well. The caller must hold the RTNL lock.
4958 * On a failure a negative errno code is returned. On success the reference
4959 * counts are adjusted and the function returns zero.
4961 int netdev_master_upper_dev_link(struct net_device *dev,
4962 struct net_device *upper_dev)
4964 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
4966 EXPORT_SYMBOL(netdev_master_upper_dev_link);
4968 int netdev_master_upper_dev_link_private(struct net_device *dev,
4969 struct net_device *upper_dev,
4972 return __netdev_upper_dev_link(dev, upper_dev, true, private);
4974 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
4977 * netdev_upper_dev_unlink - Removes a link to upper device
4979 * @upper_dev: new upper device
4981 * Removes a link to device which is upper to this one. The caller must hold
4984 void netdev_upper_dev_unlink(struct net_device *dev,
4985 struct net_device *upper_dev)
4987 struct netdev_adjacent *i, *j;
4990 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
4992 /* Here is the tricky part. We must remove all dev's lower
4993 * devices from all upper_dev's upper devices and vice
4994 * versa, to maintain the graph relationship.
4996 list_for_each_entry(i, &dev->all_adj_list.lower, list)
4997 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
4998 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5000 /* remove also the devices itself from lower/upper device
5003 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5004 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5006 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5007 __netdev_adjacent_dev_unlink(dev, i->dev);
5009 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5011 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5013 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5015 struct netdev_adjacent *iter;
5017 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5018 netdev_adjacent_sysfs_del(iter->dev, oldname,
5019 &iter->dev->adj_list.lower);
5020 netdev_adjacent_sysfs_add(iter->dev, dev,
5021 &iter->dev->adj_list.lower);
5024 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5025 netdev_adjacent_sysfs_del(iter->dev, oldname,
5026 &iter->dev->adj_list.upper);
5027 netdev_adjacent_sysfs_add(iter->dev, dev,
5028 &iter->dev->adj_list.upper);
5032 void *netdev_lower_dev_get_private(struct net_device *dev,
5033 struct net_device *lower_dev)
5035 struct netdev_adjacent *lower;
5039 lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower);
5043 return lower->private;
5045 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5047 static void dev_change_rx_flags(struct net_device *dev, int flags)
5049 const struct net_device_ops *ops = dev->netdev_ops;
5051 if (ops->ndo_change_rx_flags)
5052 ops->ndo_change_rx_flags(dev, flags);
5055 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5057 unsigned int old_flags = dev->flags;
5063 dev->flags |= IFF_PROMISC;
5064 dev->promiscuity += inc;
5065 if (dev->promiscuity == 0) {
5068 * If inc causes overflow, untouch promisc and return error.
5071 dev->flags &= ~IFF_PROMISC;
5073 dev->promiscuity -= inc;
5074 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5079 if (dev->flags != old_flags) {
5080 pr_info("device %s %s promiscuous mode\n",
5082 dev->flags & IFF_PROMISC ? "entered" : "left");
5083 if (audit_enabled) {
5084 current_uid_gid(&uid, &gid);
5085 audit_log(current->audit_context, GFP_ATOMIC,
5086 AUDIT_ANOM_PROMISCUOUS,
5087 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5088 dev->name, (dev->flags & IFF_PROMISC),
5089 (old_flags & IFF_PROMISC),
5090 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5091 from_kuid(&init_user_ns, uid),
5092 from_kgid(&init_user_ns, gid),
5093 audit_get_sessionid(current));
5096 dev_change_rx_flags(dev, IFF_PROMISC);
5099 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5104 * dev_set_promiscuity - update promiscuity count on a device
5108 * Add or remove promiscuity from a device. While the count in the device
5109 * remains above zero the interface remains promiscuous. Once it hits zero
5110 * the device reverts back to normal filtering operation. A negative inc
5111 * value is used to drop promiscuity on the device.
5112 * Return 0 if successful or a negative errno code on error.
5114 int dev_set_promiscuity(struct net_device *dev, int inc)
5116 unsigned int old_flags = dev->flags;
5119 err = __dev_set_promiscuity(dev, inc, true);
5122 if (dev->flags != old_flags)
5123 dev_set_rx_mode(dev);
5126 EXPORT_SYMBOL(dev_set_promiscuity);
5128 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5130 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5134 dev->flags |= IFF_ALLMULTI;
5135 dev->allmulti += inc;
5136 if (dev->allmulti == 0) {
5139 * If inc causes overflow, untouch allmulti and return error.
5142 dev->flags &= ~IFF_ALLMULTI;
5144 dev->allmulti -= inc;
5145 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5150 if (dev->flags ^ old_flags) {
5151 dev_change_rx_flags(dev, IFF_ALLMULTI);
5152 dev_set_rx_mode(dev);
5154 __dev_notify_flags(dev, old_flags,
5155 dev->gflags ^ old_gflags);
5161 * dev_set_allmulti - update allmulti count on a device
5165 * Add or remove reception of all multicast frames to a device. While the
5166 * count in the device remains above zero the interface remains listening
5167 * to all interfaces. Once it hits zero the device reverts back to normal
5168 * filtering operation. A negative @inc value is used to drop the counter
5169 * when releasing a resource needing all multicasts.
5170 * Return 0 if successful or a negative errno code on error.
5173 int dev_set_allmulti(struct net_device *dev, int inc)
5175 return __dev_set_allmulti(dev, inc, true);
5177 EXPORT_SYMBOL(dev_set_allmulti);
5180 * Upload unicast and multicast address lists to device and
5181 * configure RX filtering. When the device doesn't support unicast
5182 * filtering it is put in promiscuous mode while unicast addresses
5185 void __dev_set_rx_mode(struct net_device *dev)
5187 const struct net_device_ops *ops = dev->netdev_ops;
5189 /* dev_open will call this function so the list will stay sane. */
5190 if (!(dev->flags&IFF_UP))
5193 if (!netif_device_present(dev))
5196 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5197 /* Unicast addresses changes may only happen under the rtnl,
5198 * therefore calling __dev_set_promiscuity here is safe.
5200 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5201 __dev_set_promiscuity(dev, 1, false);
5202 dev->uc_promisc = true;
5203 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5204 __dev_set_promiscuity(dev, -1, false);
5205 dev->uc_promisc = false;
5209 if (ops->ndo_set_rx_mode)
5210 ops->ndo_set_rx_mode(dev);
5213 void dev_set_rx_mode(struct net_device *dev)
5215 netif_addr_lock_bh(dev);
5216 __dev_set_rx_mode(dev);
5217 netif_addr_unlock_bh(dev);
5221 * dev_get_flags - get flags reported to userspace
5224 * Get the combination of flag bits exported through APIs to userspace.
5226 unsigned int dev_get_flags(const struct net_device *dev)
5230 flags = (dev->flags & ~(IFF_PROMISC |
5235 (dev->gflags & (IFF_PROMISC |
5238 if (netif_running(dev)) {
5239 if (netif_oper_up(dev))
5240 flags |= IFF_RUNNING;
5241 if (netif_carrier_ok(dev))
5242 flags |= IFF_LOWER_UP;
5243 if (netif_dormant(dev))
5244 flags |= IFF_DORMANT;
5249 EXPORT_SYMBOL(dev_get_flags);
5251 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5253 unsigned int old_flags = dev->flags;
5259 * Set the flags on our device.
5262 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5263 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5265 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5269 * Load in the correct multicast list now the flags have changed.
5272 if ((old_flags ^ flags) & IFF_MULTICAST)
5273 dev_change_rx_flags(dev, IFF_MULTICAST);
5275 dev_set_rx_mode(dev);
5278 * Have we downed the interface. We handle IFF_UP ourselves
5279 * according to user attempts to set it, rather than blindly
5284 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
5285 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5288 dev_set_rx_mode(dev);
5291 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5292 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5293 unsigned int old_flags = dev->flags;
5295 dev->gflags ^= IFF_PROMISC;
5297 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5298 if (dev->flags != old_flags)
5299 dev_set_rx_mode(dev);
5302 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5303 is important. Some (broken) drivers set IFF_PROMISC, when
5304 IFF_ALLMULTI is requested not asking us and not reporting.
5306 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5307 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5309 dev->gflags ^= IFF_ALLMULTI;
5310 __dev_set_allmulti(dev, inc, false);
5316 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5317 unsigned int gchanges)
5319 unsigned int changes = dev->flags ^ old_flags;
5322 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
5324 if (changes & IFF_UP) {
5325 if (dev->flags & IFF_UP)
5326 call_netdevice_notifiers(NETDEV_UP, dev);
5328 call_netdevice_notifiers(NETDEV_DOWN, dev);
5331 if (dev->flags & IFF_UP &&
5332 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5333 struct netdev_notifier_change_info change_info;
5335 change_info.flags_changed = changes;
5336 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
5342 * dev_change_flags - change device settings
5344 * @flags: device state flags
5346 * Change settings on device based state flags. The flags are
5347 * in the userspace exported format.
5349 int dev_change_flags(struct net_device *dev, unsigned int flags)
5352 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
5354 ret = __dev_change_flags(dev, flags);
5358 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
5359 __dev_notify_flags(dev, old_flags, changes);
5362 EXPORT_SYMBOL(dev_change_flags);
5364 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
5366 const struct net_device_ops *ops = dev->netdev_ops;
5368 if (ops->ndo_change_mtu)
5369 return ops->ndo_change_mtu(dev, new_mtu);
5376 * dev_set_mtu - Change maximum transfer unit
5378 * @new_mtu: new transfer unit
5380 * Change the maximum transfer size of the network device.
5382 int dev_set_mtu(struct net_device *dev, int new_mtu)
5386 if (new_mtu == dev->mtu)
5389 /* MTU must be positive. */
5393 if (!netif_device_present(dev))
5396 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
5397 err = notifier_to_errno(err);
5401 orig_mtu = dev->mtu;
5402 err = __dev_set_mtu(dev, new_mtu);
5405 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5406 err = notifier_to_errno(err);
5408 /* setting mtu back and notifying everyone again,
5409 * so that they have a chance to revert changes.
5411 __dev_set_mtu(dev, orig_mtu);
5412 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5417 EXPORT_SYMBOL(dev_set_mtu);
5420 * dev_set_group - Change group this device belongs to
5422 * @new_group: group this device should belong to
5424 void dev_set_group(struct net_device *dev, int new_group)
5426 dev->group = new_group;
5428 EXPORT_SYMBOL(dev_set_group);
5431 * dev_set_mac_address - Change Media Access Control Address
5435 * Change the hardware (MAC) address of the device
5437 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
5439 const struct net_device_ops *ops = dev->netdev_ops;
5442 if (!ops->ndo_set_mac_address)
5444 if (sa->sa_family != dev->type)
5446 if (!netif_device_present(dev))
5448 err = ops->ndo_set_mac_address(dev, sa);
5451 dev->addr_assign_type = NET_ADDR_SET;
5452 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5453 add_device_randomness(dev->dev_addr, dev->addr_len);
5456 EXPORT_SYMBOL(dev_set_mac_address);
5459 * dev_change_carrier - Change device carrier
5461 * @new_carrier: new value
5463 * Change device carrier
5465 int dev_change_carrier(struct net_device *dev, bool new_carrier)
5467 const struct net_device_ops *ops = dev->netdev_ops;
5469 if (!ops->ndo_change_carrier)
5471 if (!netif_device_present(dev))
5473 return ops->ndo_change_carrier(dev, new_carrier);
5475 EXPORT_SYMBOL(dev_change_carrier);
5478 * dev_get_phys_port_id - Get device physical port ID
5482 * Get device physical port ID
5484 int dev_get_phys_port_id(struct net_device *dev,
5485 struct netdev_phys_port_id *ppid)
5487 const struct net_device_ops *ops = dev->netdev_ops;
5489 if (!ops->ndo_get_phys_port_id)
5491 return ops->ndo_get_phys_port_id(dev, ppid);
5493 EXPORT_SYMBOL(dev_get_phys_port_id);
5496 * dev_new_index - allocate an ifindex
5497 * @net: the applicable net namespace
5499 * Returns a suitable unique value for a new device interface
5500 * number. The caller must hold the rtnl semaphore or the
5501 * dev_base_lock to be sure it remains unique.
5503 static int dev_new_index(struct net *net)
5505 int ifindex = net->ifindex;
5509 if (!__dev_get_by_index(net, ifindex))
5510 return net->ifindex = ifindex;
5514 /* Delayed registration/unregisteration */
5515 static LIST_HEAD(net_todo_list);
5516 static DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
5518 static void net_set_todo(struct net_device *dev)
5520 list_add_tail(&dev->todo_list, &net_todo_list);
5521 dev_net(dev)->dev_unreg_count++;
5524 static void rollback_registered_many(struct list_head *head)
5526 struct net_device *dev, *tmp;
5527 LIST_HEAD(close_head);
5529 BUG_ON(dev_boot_phase);
5532 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5533 /* Some devices call without registering
5534 * for initialization unwind. Remove those
5535 * devices and proceed with the remaining.
5537 if (dev->reg_state == NETREG_UNINITIALIZED) {
5538 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5542 list_del(&dev->unreg_list);
5545 dev->dismantle = true;
5546 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5549 /* If device is running, close it first. */
5550 list_for_each_entry(dev, head, unreg_list)
5551 list_add_tail(&dev->close_list, &close_head);
5552 dev_close_many(&close_head);
5554 list_for_each_entry(dev, head, unreg_list) {
5555 /* And unlink it from device chain. */
5556 unlist_netdevice(dev);
5558 dev->reg_state = NETREG_UNREGISTERING;
5563 list_for_each_entry(dev, head, unreg_list) {
5564 /* Shutdown queueing discipline. */
5568 /* Notify protocols, that we are about to destroy
5569 this device. They should clean all the things.
5571 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5573 if (!dev->rtnl_link_ops ||
5574 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5575 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
5578 * Flush the unicast and multicast chains
5583 if (dev->netdev_ops->ndo_uninit)
5584 dev->netdev_ops->ndo_uninit(dev);
5586 /* Notifier chain MUST detach us all upper devices. */
5587 WARN_ON(netdev_has_any_upper_dev(dev));
5589 /* Remove entries from kobject tree */
5590 netdev_unregister_kobject(dev);
5592 /* Remove XPS queueing entries */
5593 netif_reset_xps_queues_gt(dev, 0);
5599 list_for_each_entry(dev, head, unreg_list)
5603 static void rollback_registered(struct net_device *dev)
5607 list_add(&dev->unreg_list, &single);
5608 rollback_registered_many(&single);
5612 static netdev_features_t netdev_fix_features(struct net_device *dev,
5613 netdev_features_t features)
5615 /* Fix illegal checksum combinations */
5616 if ((features & NETIF_F_HW_CSUM) &&
5617 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5618 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5619 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5622 /* TSO requires that SG is present as well. */
5623 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5624 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5625 features &= ~NETIF_F_ALL_TSO;
5628 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
5629 !(features & NETIF_F_IP_CSUM)) {
5630 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
5631 features &= ~NETIF_F_TSO;
5632 features &= ~NETIF_F_TSO_ECN;
5635 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
5636 !(features & NETIF_F_IPV6_CSUM)) {
5637 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
5638 features &= ~NETIF_F_TSO6;
5641 /* TSO ECN requires that TSO is present as well. */
5642 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5643 features &= ~NETIF_F_TSO_ECN;
5645 /* Software GSO depends on SG. */
5646 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5647 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5648 features &= ~NETIF_F_GSO;
5651 /* UFO needs SG and checksumming */
5652 if (features & NETIF_F_UFO) {
5653 /* maybe split UFO into V4 and V6? */
5654 if (!((features & NETIF_F_GEN_CSUM) ||
5655 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5656 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5658 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5659 features &= ~NETIF_F_UFO;
5662 if (!(features & NETIF_F_SG)) {
5664 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5665 features &= ~NETIF_F_UFO;
5672 int __netdev_update_features(struct net_device *dev)
5674 netdev_features_t features;
5679 features = netdev_get_wanted_features(dev);
5681 if (dev->netdev_ops->ndo_fix_features)
5682 features = dev->netdev_ops->ndo_fix_features(dev, features);
5684 /* driver might be less strict about feature dependencies */
5685 features = netdev_fix_features(dev, features);
5687 if (dev->features == features)
5690 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5691 &dev->features, &features);
5693 if (dev->netdev_ops->ndo_set_features)
5694 err = dev->netdev_ops->ndo_set_features(dev, features);
5696 if (unlikely(err < 0)) {
5698 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5699 err, &features, &dev->features);
5704 dev->features = features;
5710 * netdev_update_features - recalculate device features
5711 * @dev: the device to check
5713 * Recalculate dev->features set and send notifications if it
5714 * has changed. Should be called after driver or hardware dependent
5715 * conditions might have changed that influence the features.
5717 void netdev_update_features(struct net_device *dev)
5719 if (__netdev_update_features(dev))
5720 netdev_features_change(dev);
5722 EXPORT_SYMBOL(netdev_update_features);
5725 * netdev_change_features - recalculate device features
5726 * @dev: the device to check
5728 * Recalculate dev->features set and send notifications even
5729 * if they have not changed. Should be called instead of
5730 * netdev_update_features() if also dev->vlan_features might
5731 * have changed to allow the changes to be propagated to stacked
5734 void netdev_change_features(struct net_device *dev)
5736 __netdev_update_features(dev);
5737 netdev_features_change(dev);
5739 EXPORT_SYMBOL(netdev_change_features);
5742 * netif_stacked_transfer_operstate - transfer operstate
5743 * @rootdev: the root or lower level device to transfer state from
5744 * @dev: the device to transfer operstate to
5746 * Transfer operational state from root to device. This is normally
5747 * called when a stacking relationship exists between the root
5748 * device and the device(a leaf device).
5750 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5751 struct net_device *dev)
5753 if (rootdev->operstate == IF_OPER_DORMANT)
5754 netif_dormant_on(dev);
5756 netif_dormant_off(dev);
5758 if (netif_carrier_ok(rootdev)) {
5759 if (!netif_carrier_ok(dev))
5760 netif_carrier_on(dev);
5762 if (netif_carrier_ok(dev))
5763 netif_carrier_off(dev);
5766 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5769 static int netif_alloc_rx_queues(struct net_device *dev)
5771 unsigned int i, count = dev->num_rx_queues;
5772 struct netdev_rx_queue *rx;
5776 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5782 for (i = 0; i < count; i++)
5788 static void netdev_init_one_queue(struct net_device *dev,
5789 struct netdev_queue *queue, void *_unused)
5791 /* Initialize queue lock */
5792 spin_lock_init(&queue->_xmit_lock);
5793 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5794 queue->xmit_lock_owner = -1;
5795 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5798 dql_init(&queue->dql, HZ);
5802 static void netif_free_tx_queues(struct net_device *dev)
5804 if (is_vmalloc_addr(dev->_tx))
5810 static int netif_alloc_netdev_queues(struct net_device *dev)
5812 unsigned int count = dev->num_tx_queues;
5813 struct netdev_queue *tx;
5814 size_t sz = count * sizeof(*tx);
5816 BUG_ON(count < 1 || count > 0xffff);
5818 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
5826 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5827 spin_lock_init(&dev->tx_global_lock);
5833 * register_netdevice - register a network device
5834 * @dev: device to register
5836 * Take a completed network device structure and add it to the kernel
5837 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5838 * chain. 0 is returned on success. A negative errno code is returned
5839 * on a failure to set up the device, or if the name is a duplicate.
5841 * Callers must hold the rtnl semaphore. You may want
5842 * register_netdev() instead of this.
5845 * The locking appears insufficient to guarantee two parallel registers
5846 * will not get the same name.
5849 int register_netdevice(struct net_device *dev)
5852 struct net *net = dev_net(dev);
5854 BUG_ON(dev_boot_phase);
5859 /* When net_device's are persistent, this will be fatal. */
5860 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5863 spin_lock_init(&dev->addr_list_lock);
5864 netdev_set_addr_lockdep_class(dev);
5868 ret = dev_get_valid_name(net, dev, dev->name);
5872 /* Init, if this function is available */
5873 if (dev->netdev_ops->ndo_init) {
5874 ret = dev->netdev_ops->ndo_init(dev);
5882 if (((dev->hw_features | dev->features) &
5883 NETIF_F_HW_VLAN_CTAG_FILTER) &&
5884 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
5885 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
5886 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
5893 dev->ifindex = dev_new_index(net);
5894 else if (__dev_get_by_index(net, dev->ifindex))
5897 if (dev->iflink == -1)
5898 dev->iflink = dev->ifindex;
5900 /* Transfer changeable features to wanted_features and enable
5901 * software offloads (GSO and GRO).
5903 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5904 dev->features |= NETIF_F_SOFT_FEATURES;
5905 dev->wanted_features = dev->features & dev->hw_features;
5907 if (!(dev->flags & IFF_LOOPBACK)) {
5908 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5911 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5913 dev->vlan_features |= NETIF_F_HIGHDMA;
5915 /* Make NETIF_F_SG inheritable to tunnel devices.
5917 dev->hw_enc_features |= NETIF_F_SG;
5919 /* Make NETIF_F_SG inheritable to MPLS.
5921 dev->mpls_features |= NETIF_F_SG;
5923 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5924 ret = notifier_to_errno(ret);
5928 ret = netdev_register_kobject(dev);
5931 dev->reg_state = NETREG_REGISTERED;
5933 __netdev_update_features(dev);
5936 * Default initial state at registry is that the
5937 * device is present.
5940 set_bit(__LINK_STATE_PRESENT, &dev->state);
5942 linkwatch_init_dev(dev);
5944 dev_init_scheduler(dev);
5946 list_netdevice(dev);
5947 add_device_randomness(dev->dev_addr, dev->addr_len);
5949 /* If the device has permanent device address, driver should
5950 * set dev_addr and also addr_assign_type should be set to
5951 * NET_ADDR_PERM (default value).
5953 if (dev->addr_assign_type == NET_ADDR_PERM)
5954 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
5956 /* Notify protocols, that a new device appeared. */
5957 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5958 ret = notifier_to_errno(ret);
5960 rollback_registered(dev);
5961 dev->reg_state = NETREG_UNREGISTERED;
5964 * Prevent userspace races by waiting until the network
5965 * device is fully setup before sending notifications.
5967 if (!dev->rtnl_link_ops ||
5968 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5969 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
5975 if (dev->netdev_ops->ndo_uninit)
5976 dev->netdev_ops->ndo_uninit(dev);
5979 EXPORT_SYMBOL(register_netdevice);
5982 * init_dummy_netdev - init a dummy network device for NAPI
5983 * @dev: device to init
5985 * This takes a network device structure and initialize the minimum
5986 * amount of fields so it can be used to schedule NAPI polls without
5987 * registering a full blown interface. This is to be used by drivers
5988 * that need to tie several hardware interfaces to a single NAPI
5989 * poll scheduler due to HW limitations.
5991 int init_dummy_netdev(struct net_device *dev)
5993 /* Clear everything. Note we don't initialize spinlocks
5994 * are they aren't supposed to be taken by any of the
5995 * NAPI code and this dummy netdev is supposed to be
5996 * only ever used for NAPI polls
5998 memset(dev, 0, sizeof(struct net_device));
6000 /* make sure we BUG if trying to hit standard
6001 * register/unregister code path
6003 dev->reg_state = NETREG_DUMMY;
6005 /* NAPI wants this */
6006 INIT_LIST_HEAD(&dev->napi_list);
6008 /* a dummy interface is started by default */
6009 set_bit(__LINK_STATE_PRESENT, &dev->state);
6010 set_bit(__LINK_STATE_START, &dev->state);
6012 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6013 * because users of this 'device' dont need to change
6019 EXPORT_SYMBOL_GPL(init_dummy_netdev);
6023 * register_netdev - register a network device
6024 * @dev: device to register
6026 * Take a completed network device structure and add it to the kernel
6027 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6028 * chain. 0 is returned on success. A negative errno code is returned
6029 * on a failure to set up the device, or if the name is a duplicate.
6031 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6032 * and expands the device name if you passed a format string to
6035 int register_netdev(struct net_device *dev)
6040 err = register_netdevice(dev);
6044 EXPORT_SYMBOL(register_netdev);
6046 int netdev_refcnt_read(const struct net_device *dev)
6050 for_each_possible_cpu(i)
6051 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
6054 EXPORT_SYMBOL(netdev_refcnt_read);
6057 * netdev_wait_allrefs - wait until all references are gone.
6058 * @dev: target net_device
6060 * This is called when unregistering network devices.
6062 * Any protocol or device that holds a reference should register
6063 * for netdevice notification, and cleanup and put back the
6064 * reference if they receive an UNREGISTER event.
6065 * We can get stuck here if buggy protocols don't correctly
6068 static void netdev_wait_allrefs(struct net_device *dev)
6070 unsigned long rebroadcast_time, warning_time;
6073 linkwatch_forget_dev(dev);
6075 rebroadcast_time = warning_time = jiffies;
6076 refcnt = netdev_refcnt_read(dev);
6078 while (refcnt != 0) {
6079 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6082 /* Rebroadcast unregister notification */
6083 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6089 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6090 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6092 /* We must not have linkwatch events
6093 * pending on unregister. If this
6094 * happens, we simply run the queue
6095 * unscheduled, resulting in a noop
6098 linkwatch_run_queue();
6103 rebroadcast_time = jiffies;
6108 refcnt = netdev_refcnt_read(dev);
6110 if (time_after(jiffies, warning_time + 10 * HZ)) {
6111 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6113 warning_time = jiffies;
6122 * register_netdevice(x1);
6123 * register_netdevice(x2);
6125 * unregister_netdevice(y1);
6126 * unregister_netdevice(y2);
6132 * We are invoked by rtnl_unlock().
6133 * This allows us to deal with problems:
6134 * 1) We can delete sysfs objects which invoke hotplug
6135 * without deadlocking with linkwatch via keventd.
6136 * 2) Since we run with the RTNL semaphore not held, we can sleep
6137 * safely in order to wait for the netdev refcnt to drop to zero.
6139 * We must not return until all unregister events added during
6140 * the interval the lock was held have been completed.
6142 void netdev_run_todo(void)
6144 struct list_head list;
6146 /* Snapshot list, allow later requests */
6147 list_replace_init(&net_todo_list, &list);
6152 /* Wait for rcu callbacks to finish before next phase */
6153 if (!list_empty(&list))
6156 while (!list_empty(&list)) {
6157 struct net_device *dev
6158 = list_first_entry(&list, struct net_device, todo_list);
6159 list_del(&dev->todo_list);
6162 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6165 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6166 pr_err("network todo '%s' but state %d\n",
6167 dev->name, dev->reg_state);
6172 dev->reg_state = NETREG_UNREGISTERED;
6174 on_each_cpu(flush_backlog, dev, 1);
6176 netdev_wait_allrefs(dev);
6179 BUG_ON(netdev_refcnt_read(dev));
6180 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6181 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6182 WARN_ON(dev->dn_ptr);
6184 if (dev->destructor)
6185 dev->destructor(dev);
6187 /* Report a network device has been unregistered */
6189 dev_net(dev)->dev_unreg_count--;
6191 wake_up(&netdev_unregistering_wq);
6193 /* Free network device */
6194 kobject_put(&dev->dev.kobj);
6198 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6199 * fields in the same order, with only the type differing.
6201 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6202 const struct net_device_stats *netdev_stats)
6204 #if BITS_PER_LONG == 64
6205 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6206 memcpy(stats64, netdev_stats, sizeof(*stats64));
6208 size_t i, n = sizeof(*stats64) / sizeof(u64);
6209 const unsigned long *src = (const unsigned long *)netdev_stats;
6210 u64 *dst = (u64 *)stats64;
6212 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6213 sizeof(*stats64) / sizeof(u64));
6214 for (i = 0; i < n; i++)
6218 EXPORT_SYMBOL(netdev_stats_to_stats64);
6221 * dev_get_stats - get network device statistics
6222 * @dev: device to get statistics from
6223 * @storage: place to store stats
6225 * Get network statistics from device. Return @storage.
6226 * The device driver may provide its own method by setting
6227 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6228 * otherwise the internal statistics structure is used.
6230 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6231 struct rtnl_link_stats64 *storage)
6233 const struct net_device_ops *ops = dev->netdev_ops;
6235 if (ops->ndo_get_stats64) {
6236 memset(storage, 0, sizeof(*storage));
6237 ops->ndo_get_stats64(dev, storage);
6238 } else if (ops->ndo_get_stats) {
6239 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6241 netdev_stats_to_stats64(storage, &dev->stats);
6243 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6246 EXPORT_SYMBOL(dev_get_stats);
6248 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
6250 struct netdev_queue *queue = dev_ingress_queue(dev);
6252 #ifdef CONFIG_NET_CLS_ACT
6255 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
6258 netdev_init_one_queue(dev, queue, NULL);
6259 queue->qdisc = &noop_qdisc;
6260 queue->qdisc_sleeping = &noop_qdisc;
6261 rcu_assign_pointer(dev->ingress_queue, queue);
6266 static const struct ethtool_ops default_ethtool_ops;
6268 void netdev_set_default_ethtool_ops(struct net_device *dev,
6269 const struct ethtool_ops *ops)
6271 if (dev->ethtool_ops == &default_ethtool_ops)
6272 dev->ethtool_ops = ops;
6274 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
6276 void netdev_freemem(struct net_device *dev)
6278 char *addr = (char *)dev - dev->padded;
6280 if (is_vmalloc_addr(addr))
6287 * alloc_netdev_mqs - allocate network device
6288 * @sizeof_priv: size of private data to allocate space for
6289 * @name: device name format string
6290 * @setup: callback to initialize device
6291 * @txqs: the number of TX subqueues to allocate
6292 * @rxqs: the number of RX subqueues to allocate
6294 * Allocates a struct net_device with private data area for driver use
6295 * and performs basic initialization. Also allocates subquue structs
6296 * for each queue on the device.
6298 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
6299 void (*setup)(struct net_device *),
6300 unsigned int txqs, unsigned int rxqs)
6302 struct net_device *dev;
6304 struct net_device *p;
6306 BUG_ON(strlen(name) >= sizeof(dev->name));
6309 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6315 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6320 alloc_size = sizeof(struct net_device);
6322 /* ensure 32-byte alignment of private area */
6323 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6324 alloc_size += sizeof_priv;
6326 /* ensure 32-byte alignment of whole construct */
6327 alloc_size += NETDEV_ALIGN - 1;
6329 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6331 p = vzalloc(alloc_size);
6335 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6336 dev->padded = (char *)dev - (char *)p;
6338 dev->pcpu_refcnt = alloc_percpu(int);
6339 if (!dev->pcpu_refcnt)
6342 if (dev_addr_init(dev))
6348 dev_net_set(dev, &init_net);
6350 dev->gso_max_size = GSO_MAX_SIZE;
6351 dev->gso_max_segs = GSO_MAX_SEGS;
6353 INIT_LIST_HEAD(&dev->napi_list);
6354 INIT_LIST_HEAD(&dev->unreg_list);
6355 INIT_LIST_HEAD(&dev->close_list);
6356 INIT_LIST_HEAD(&dev->link_watch_list);
6357 INIT_LIST_HEAD(&dev->adj_list.upper);
6358 INIT_LIST_HEAD(&dev->adj_list.lower);
6359 INIT_LIST_HEAD(&dev->all_adj_list.upper);
6360 INIT_LIST_HEAD(&dev->all_adj_list.lower);
6361 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6364 dev->num_tx_queues = txqs;
6365 dev->real_num_tx_queues = txqs;
6366 if (netif_alloc_netdev_queues(dev))
6370 dev->num_rx_queues = rxqs;
6371 dev->real_num_rx_queues = rxqs;
6372 if (netif_alloc_rx_queues(dev))
6376 strcpy(dev->name, name);
6377 dev->group = INIT_NETDEV_GROUP;
6378 if (!dev->ethtool_ops)
6379 dev->ethtool_ops = &default_ethtool_ops;
6387 free_percpu(dev->pcpu_refcnt);
6388 netif_free_tx_queues(dev);
6394 netdev_freemem(dev);
6397 EXPORT_SYMBOL(alloc_netdev_mqs);
6400 * free_netdev - free network device
6403 * This function does the last stage of destroying an allocated device
6404 * interface. The reference to the device object is released.
6405 * If this is the last reference then it will be freed.
6407 void free_netdev(struct net_device *dev)
6409 struct napi_struct *p, *n;
6411 release_net(dev_net(dev));
6413 netif_free_tx_queues(dev);
6418 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6420 /* Flush device addresses */
6421 dev_addr_flush(dev);
6423 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6426 free_percpu(dev->pcpu_refcnt);
6427 dev->pcpu_refcnt = NULL;
6429 /* Compatibility with error handling in drivers */
6430 if (dev->reg_state == NETREG_UNINITIALIZED) {
6431 netdev_freemem(dev);
6435 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6436 dev->reg_state = NETREG_RELEASED;
6438 /* will free via device release */
6439 put_device(&dev->dev);
6441 EXPORT_SYMBOL(free_netdev);
6444 * synchronize_net - Synchronize with packet receive processing
6446 * Wait for packets currently being received to be done.
6447 * Does not block later packets from starting.
6449 void synchronize_net(void)
6452 if (rtnl_is_locked())
6453 synchronize_rcu_expedited();
6457 EXPORT_SYMBOL(synchronize_net);
6460 * unregister_netdevice_queue - remove device from the kernel
6464 * This function shuts down a device interface and removes it
6465 * from the kernel tables.
6466 * If head not NULL, device is queued to be unregistered later.
6468 * Callers must hold the rtnl semaphore. You may want
6469 * unregister_netdev() instead of this.
6472 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6477 list_move_tail(&dev->unreg_list, head);
6479 rollback_registered(dev);
6480 /* Finish processing unregister after unlock */
6484 EXPORT_SYMBOL(unregister_netdevice_queue);
6487 * unregister_netdevice_many - unregister many devices
6488 * @head: list of devices
6490 void unregister_netdevice_many(struct list_head *head)
6492 struct net_device *dev;
6494 if (!list_empty(head)) {
6495 rollback_registered_many(head);
6496 list_for_each_entry(dev, head, unreg_list)
6500 EXPORT_SYMBOL(unregister_netdevice_many);
6503 * unregister_netdev - remove device from the kernel
6506 * This function shuts down a device interface and removes it
6507 * from the kernel tables.
6509 * This is just a wrapper for unregister_netdevice that takes
6510 * the rtnl semaphore. In general you want to use this and not
6511 * unregister_netdevice.
6513 void unregister_netdev(struct net_device *dev)
6516 unregister_netdevice(dev);
6519 EXPORT_SYMBOL(unregister_netdev);
6522 * dev_change_net_namespace - move device to different nethost namespace
6524 * @net: network namespace
6525 * @pat: If not NULL name pattern to try if the current device name
6526 * is already taken in the destination network namespace.
6528 * This function shuts down a device interface and moves it
6529 * to a new network namespace. On success 0 is returned, on
6530 * a failure a netagive errno code is returned.
6532 * Callers must hold the rtnl semaphore.
6535 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6541 /* Don't allow namespace local devices to be moved. */
6543 if (dev->features & NETIF_F_NETNS_LOCAL)
6546 /* Ensure the device has been registrered */
6547 if (dev->reg_state != NETREG_REGISTERED)
6550 /* Get out if there is nothing todo */
6552 if (net_eq(dev_net(dev), net))
6555 /* Pick the destination device name, and ensure
6556 * we can use it in the destination network namespace.
6559 if (__dev_get_by_name(net, dev->name)) {
6560 /* We get here if we can't use the current device name */
6563 if (dev_get_valid_name(net, dev, pat) < 0)
6568 * And now a mini version of register_netdevice unregister_netdevice.
6571 /* If device is running close it first. */
6574 /* And unlink it from device chain */
6576 unlist_netdevice(dev);
6580 /* Shutdown queueing discipline. */
6583 /* Notify protocols, that we are about to destroy
6584 this device. They should clean all the things.
6586 Note that dev->reg_state stays at NETREG_REGISTERED.
6587 This is wanted because this way 8021q and macvlan know
6588 the device is just moving and can keep their slaves up.
6590 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6592 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6593 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
6596 * Flush the unicast and multicast chains
6601 /* Send a netdev-removed uevent to the old namespace */
6602 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
6604 /* Actually switch the network namespace */
6605 dev_net_set(dev, net);
6607 /* If there is an ifindex conflict assign a new one */
6608 if (__dev_get_by_index(net, dev->ifindex)) {
6609 int iflink = (dev->iflink == dev->ifindex);
6610 dev->ifindex = dev_new_index(net);
6612 dev->iflink = dev->ifindex;
6615 /* Send a netdev-add uevent to the new namespace */
6616 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
6618 /* Fixup kobjects */
6619 err = device_rename(&dev->dev, dev->name);
6622 /* Add the device back in the hashes */
6623 list_netdevice(dev);
6625 /* Notify protocols, that a new device appeared. */
6626 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6629 * Prevent userspace races by waiting until the network
6630 * device is fully setup before sending notifications.
6632 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6639 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6641 static int dev_cpu_callback(struct notifier_block *nfb,
6642 unsigned long action,
6645 struct sk_buff **list_skb;
6646 struct sk_buff *skb;
6647 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6648 struct softnet_data *sd, *oldsd;
6650 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6653 local_irq_disable();
6654 cpu = smp_processor_id();
6655 sd = &per_cpu(softnet_data, cpu);
6656 oldsd = &per_cpu(softnet_data, oldcpu);
6658 /* Find end of our completion_queue. */
6659 list_skb = &sd->completion_queue;
6661 list_skb = &(*list_skb)->next;
6662 /* Append completion queue from offline CPU. */
6663 *list_skb = oldsd->completion_queue;
6664 oldsd->completion_queue = NULL;
6666 /* Append output queue from offline CPU. */
6667 if (oldsd->output_queue) {
6668 *sd->output_queue_tailp = oldsd->output_queue;
6669 sd->output_queue_tailp = oldsd->output_queue_tailp;
6670 oldsd->output_queue = NULL;
6671 oldsd->output_queue_tailp = &oldsd->output_queue;
6673 /* Append NAPI poll list from offline CPU. */
6674 if (!list_empty(&oldsd->poll_list)) {
6675 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6676 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6679 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6682 /* Process offline CPU's input_pkt_queue */
6683 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6684 netif_rx_internal(skb);
6685 input_queue_head_incr(oldsd);
6687 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6688 netif_rx_internal(skb);
6689 input_queue_head_incr(oldsd);
6697 * netdev_increment_features - increment feature set by one
6698 * @all: current feature set
6699 * @one: new feature set
6700 * @mask: mask feature set
6702 * Computes a new feature set after adding a device with feature set
6703 * @one to the master device with current feature set @all. Will not
6704 * enable anything that is off in @mask. Returns the new feature set.
6706 netdev_features_t netdev_increment_features(netdev_features_t all,
6707 netdev_features_t one, netdev_features_t mask)
6709 if (mask & NETIF_F_GEN_CSUM)
6710 mask |= NETIF_F_ALL_CSUM;
6711 mask |= NETIF_F_VLAN_CHALLENGED;
6713 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6714 all &= one | ~NETIF_F_ALL_FOR_ALL;
6716 /* If one device supports hw checksumming, set for all. */
6717 if (all & NETIF_F_GEN_CSUM)
6718 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6722 EXPORT_SYMBOL(netdev_increment_features);
6724 static struct hlist_head * __net_init netdev_create_hash(void)
6727 struct hlist_head *hash;
6729 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6731 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6732 INIT_HLIST_HEAD(&hash[i]);
6737 /* Initialize per network namespace state */
6738 static int __net_init netdev_init(struct net *net)
6740 if (net != &init_net)
6741 INIT_LIST_HEAD(&net->dev_base_head);
6743 net->dev_name_head = netdev_create_hash();
6744 if (net->dev_name_head == NULL)
6747 net->dev_index_head = netdev_create_hash();
6748 if (net->dev_index_head == NULL)
6754 kfree(net->dev_name_head);
6760 * netdev_drivername - network driver for the device
6761 * @dev: network device
6763 * Determine network driver for device.
6765 const char *netdev_drivername(const struct net_device *dev)
6767 const struct device_driver *driver;
6768 const struct device *parent;
6769 const char *empty = "";
6771 parent = dev->dev.parent;
6775 driver = parent->driver;
6776 if (driver && driver->name)
6777 return driver->name;
6781 static int __netdev_printk(const char *level, const struct net_device *dev,
6782 struct va_format *vaf)
6786 if (dev && dev->dev.parent) {
6787 r = dev_printk_emit(level[1] - '0',
6790 dev_driver_string(dev->dev.parent),
6791 dev_name(dev->dev.parent),
6792 netdev_name(dev), vaf);
6794 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6796 r = printk("%s(NULL net_device): %pV", level, vaf);
6802 int netdev_printk(const char *level, const struct net_device *dev,
6803 const char *format, ...)
6805 struct va_format vaf;
6809 va_start(args, format);
6814 r = __netdev_printk(level, dev, &vaf);
6820 EXPORT_SYMBOL(netdev_printk);
6822 #define define_netdev_printk_level(func, level) \
6823 int func(const struct net_device *dev, const char *fmt, ...) \
6826 struct va_format vaf; \
6829 va_start(args, fmt); \
6834 r = __netdev_printk(level, dev, &vaf); \
6840 EXPORT_SYMBOL(func);
6842 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6843 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6844 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6845 define_netdev_printk_level(netdev_err, KERN_ERR);
6846 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6847 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6848 define_netdev_printk_level(netdev_info, KERN_INFO);
6850 static void __net_exit netdev_exit(struct net *net)
6852 kfree(net->dev_name_head);
6853 kfree(net->dev_index_head);
6856 static struct pernet_operations __net_initdata netdev_net_ops = {
6857 .init = netdev_init,
6858 .exit = netdev_exit,
6861 static void __net_exit default_device_exit(struct net *net)
6863 struct net_device *dev, *aux;
6865 * Push all migratable network devices back to the
6866 * initial network namespace
6869 for_each_netdev_safe(net, dev, aux) {
6871 char fb_name[IFNAMSIZ];
6873 /* Ignore unmoveable devices (i.e. loopback) */
6874 if (dev->features & NETIF_F_NETNS_LOCAL)
6877 /* Leave virtual devices for the generic cleanup */
6878 if (dev->rtnl_link_ops)
6881 /* Push remaining network devices to init_net */
6882 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6883 err = dev_change_net_namespace(dev, &init_net, fb_name);
6885 pr_emerg("%s: failed to move %s to init_net: %d\n",
6886 __func__, dev->name, err);
6893 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
6895 /* Return with the rtnl_lock held when there are no network
6896 * devices unregistering in any network namespace in net_list.
6903 prepare_to_wait(&netdev_unregistering_wq, &wait,
6904 TASK_UNINTERRUPTIBLE);
6905 unregistering = false;
6907 list_for_each_entry(net, net_list, exit_list) {
6908 if (net->dev_unreg_count > 0) {
6909 unregistering = true;
6918 finish_wait(&netdev_unregistering_wq, &wait);
6921 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6923 /* At exit all network devices most be removed from a network
6924 * namespace. Do this in the reverse order of registration.
6925 * Do this across as many network namespaces as possible to
6926 * improve batching efficiency.
6928 struct net_device *dev;
6930 LIST_HEAD(dev_kill_list);
6932 /* To prevent network device cleanup code from dereferencing
6933 * loopback devices or network devices that have been freed
6934 * wait here for all pending unregistrations to complete,
6935 * before unregistring the loopback device and allowing the
6936 * network namespace be freed.
6938 * The netdev todo list containing all network devices
6939 * unregistrations that happen in default_device_exit_batch
6940 * will run in the rtnl_unlock() at the end of
6941 * default_device_exit_batch.
6943 rtnl_lock_unregistering(net_list);
6944 list_for_each_entry(net, net_list, exit_list) {
6945 for_each_netdev_reverse(net, dev) {
6946 if (dev->rtnl_link_ops)
6947 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6949 unregister_netdevice_queue(dev, &dev_kill_list);
6952 unregister_netdevice_many(&dev_kill_list);
6953 list_del(&dev_kill_list);
6957 static struct pernet_operations __net_initdata default_device_ops = {
6958 .exit = default_device_exit,
6959 .exit_batch = default_device_exit_batch,
6963 * Initialize the DEV module. At boot time this walks the device list and
6964 * unhooks any devices that fail to initialise (normally hardware not
6965 * present) and leaves us with a valid list of present and active devices.
6970 * This is called single threaded during boot, so no need
6971 * to take the rtnl semaphore.
6973 static int __init net_dev_init(void)
6975 int i, rc = -ENOMEM;
6977 BUG_ON(!dev_boot_phase);
6979 if (dev_proc_init())
6982 if (netdev_kobject_init())
6985 INIT_LIST_HEAD(&ptype_all);
6986 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6987 INIT_LIST_HEAD(&ptype_base[i]);
6989 INIT_LIST_HEAD(&offload_base);
6991 if (register_pernet_subsys(&netdev_net_ops))
6995 * Initialise the packet receive queues.
6998 for_each_possible_cpu(i) {
6999 struct softnet_data *sd = &per_cpu(softnet_data, i);
7001 memset(sd, 0, sizeof(*sd));
7002 skb_queue_head_init(&sd->input_pkt_queue);
7003 skb_queue_head_init(&sd->process_queue);
7004 sd->completion_queue = NULL;
7005 INIT_LIST_HEAD(&sd->poll_list);
7006 sd->output_queue = NULL;
7007 sd->output_queue_tailp = &sd->output_queue;
7009 sd->csd.func = rps_trigger_softirq;
7015 sd->backlog.poll = process_backlog;
7016 sd->backlog.weight = weight_p;
7017 sd->backlog.gro_list = NULL;
7018 sd->backlog.gro_count = 0;
7020 #ifdef CONFIG_NET_FLOW_LIMIT
7021 sd->flow_limit = NULL;
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);