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 <net/mpls.h>
122 #include <linux/ipv6.h>
123 #include <linux/in.h>
124 #include <linux/jhash.h>
125 #include <linux/random.h>
126 #include <trace/events/napi.h>
127 #include <trace/events/net.h>
128 #include <trace/events/skb.h>
129 #include <linux/pci.h>
130 #include <linux/inetdevice.h>
131 #include <linux/cpu_rmap.h>
132 #include <linux/static_key.h>
133 #include <linux/hashtable.h>
134 #include <linux/vmalloc.h>
135 #include <linux/if_macvlan.h>
136 #include <linux/errqueue.h>
137 #include <linux/hrtimer.h>
139 #include "net-sysfs.h"
141 /* Instead of increasing this, you should create a hash table. */
142 #define MAX_GRO_SKBS 8
144 /* This should be increased if a protocol with a bigger head is added. */
145 #define GRO_MAX_HEAD (MAX_HEADER + 128)
147 static DEFINE_SPINLOCK(ptype_lock);
148 static DEFINE_SPINLOCK(offload_lock);
149 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
150 struct list_head ptype_all __read_mostly; /* Taps */
151 static struct list_head offload_base __read_mostly;
153 static int netif_rx_internal(struct sk_buff *skb);
154 static int call_netdevice_notifiers_info(unsigned long val,
155 struct net_device *dev,
156 struct netdev_notifier_info *info);
159 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
162 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
164 * Writers must hold the rtnl semaphore while they loop through the
165 * dev_base_head list, and hold dev_base_lock for writing when they do the
166 * actual updates. This allows pure readers to access the list even
167 * while a writer is preparing to update it.
169 * To put it another way, dev_base_lock is held for writing only to
170 * protect against pure readers; the rtnl semaphore provides the
171 * protection against other writers.
173 * See, for example usages, register_netdevice() and
174 * unregister_netdevice(), which must be called with the rtnl
177 DEFINE_RWLOCK(dev_base_lock);
178 EXPORT_SYMBOL(dev_base_lock);
180 /* protects napi_hash addition/deletion and napi_gen_id */
181 static DEFINE_SPINLOCK(napi_hash_lock);
183 static unsigned int napi_gen_id;
184 static DEFINE_HASHTABLE(napi_hash, 8);
186 static seqcount_t devnet_rename_seq;
188 static inline void dev_base_seq_inc(struct net *net)
190 while (++net->dev_base_seq == 0);
193 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
195 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
197 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
200 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
202 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
205 static inline void rps_lock(struct softnet_data *sd)
208 spin_lock(&sd->input_pkt_queue.lock);
212 static inline void rps_unlock(struct softnet_data *sd)
215 spin_unlock(&sd->input_pkt_queue.lock);
219 /* Device list insertion */
220 static void list_netdevice(struct net_device *dev)
222 struct net *net = dev_net(dev);
226 write_lock_bh(&dev_base_lock);
227 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
228 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
229 hlist_add_head_rcu(&dev->index_hlist,
230 dev_index_hash(net, dev->ifindex));
231 write_unlock_bh(&dev_base_lock);
233 dev_base_seq_inc(net);
236 /* Device list removal
237 * caller must respect a RCU grace period before freeing/reusing dev
239 static void unlist_netdevice(struct net_device *dev)
243 /* Unlink dev from the device chain */
244 write_lock_bh(&dev_base_lock);
245 list_del_rcu(&dev->dev_list);
246 hlist_del_rcu(&dev->name_hlist);
247 hlist_del_rcu(&dev->index_hlist);
248 write_unlock_bh(&dev_base_lock);
250 dev_base_seq_inc(dev_net(dev));
257 static RAW_NOTIFIER_HEAD(netdev_chain);
260 * Device drivers call our routines to queue packets here. We empty the
261 * queue in the local softnet handler.
264 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
265 EXPORT_PER_CPU_SYMBOL(softnet_data);
267 #ifdef CONFIG_LOCKDEP
269 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
270 * according to dev->type
272 static const unsigned short netdev_lock_type[] =
273 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
274 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
275 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
276 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
277 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
278 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
279 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
280 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
281 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
282 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
283 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
284 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
285 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
286 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
287 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
289 static const char *const netdev_lock_name[] =
290 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
291 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
292 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
293 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
294 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
295 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
296 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
297 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
298 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
299 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
300 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
301 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
302 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
303 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
304 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
306 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
307 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
309 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
313 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
314 if (netdev_lock_type[i] == dev_type)
316 /* the last key is used by default */
317 return ARRAY_SIZE(netdev_lock_type) - 1;
320 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
321 unsigned short dev_type)
325 i = netdev_lock_pos(dev_type);
326 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
327 netdev_lock_name[i]);
330 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
334 i = netdev_lock_pos(dev->type);
335 lockdep_set_class_and_name(&dev->addr_list_lock,
336 &netdev_addr_lock_key[i],
337 netdev_lock_name[i]);
340 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
341 unsigned short dev_type)
344 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
349 /*******************************************************************************
351 Protocol management and registration routines
353 *******************************************************************************/
356 * Add a protocol ID to the list. Now that the input handler is
357 * smarter we can dispense with all the messy stuff that used to be
360 * BEWARE!!! Protocol handlers, mangling input packets,
361 * MUST BE last in hash buckets and checking protocol handlers
362 * MUST start from promiscuous ptype_all chain in net_bh.
363 * It is true now, do not change it.
364 * Explanation follows: if protocol handler, mangling packet, will
365 * be the first on list, it is not able to sense, that packet
366 * is cloned and should be copied-on-write, so that it will
367 * change it and subsequent readers will get broken packet.
371 static inline struct list_head *ptype_head(const struct packet_type *pt)
373 if (pt->type == htons(ETH_P_ALL))
376 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
380 * dev_add_pack - add packet handler
381 * @pt: packet type declaration
383 * Add a protocol handler to the networking stack. The passed &packet_type
384 * is linked into kernel lists and may not be freed until it has been
385 * removed from the kernel lists.
387 * This call does not sleep therefore it can not
388 * guarantee all CPU's that are in middle of receiving packets
389 * will see the new packet type (until the next received packet).
392 void dev_add_pack(struct packet_type *pt)
394 struct list_head *head = ptype_head(pt);
396 spin_lock(&ptype_lock);
397 list_add_rcu(&pt->list, head);
398 spin_unlock(&ptype_lock);
400 EXPORT_SYMBOL(dev_add_pack);
403 * __dev_remove_pack - remove packet handler
404 * @pt: packet type declaration
406 * Remove a protocol handler that was previously added to the kernel
407 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
408 * from the kernel lists and can be freed or reused once this function
411 * The packet type might still be in use by receivers
412 * and must not be freed until after all the CPU's have gone
413 * through a quiescent state.
415 void __dev_remove_pack(struct packet_type *pt)
417 struct list_head *head = ptype_head(pt);
418 struct packet_type *pt1;
420 spin_lock(&ptype_lock);
422 list_for_each_entry(pt1, head, list) {
424 list_del_rcu(&pt->list);
429 pr_warn("dev_remove_pack: %p not found\n", pt);
431 spin_unlock(&ptype_lock);
433 EXPORT_SYMBOL(__dev_remove_pack);
436 * dev_remove_pack - remove packet handler
437 * @pt: packet type declaration
439 * Remove a protocol handler that was previously added to the kernel
440 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
441 * from the kernel lists and can be freed or reused once this function
444 * This call sleeps to guarantee that no CPU is looking at the packet
447 void dev_remove_pack(struct packet_type *pt)
449 __dev_remove_pack(pt);
453 EXPORT_SYMBOL(dev_remove_pack);
457 * dev_add_offload - register offload handlers
458 * @po: protocol offload declaration
460 * Add protocol offload handlers to the networking stack. The passed
461 * &proto_offload is linked into kernel lists and may not be freed until
462 * it has been removed from the kernel lists.
464 * This call does not sleep therefore it can not
465 * guarantee all CPU's that are in middle of receiving packets
466 * will see the new offload handlers (until the next received packet).
468 void dev_add_offload(struct packet_offload *po)
470 struct list_head *head = &offload_base;
472 spin_lock(&offload_lock);
473 list_add_rcu(&po->list, head);
474 spin_unlock(&offload_lock);
476 EXPORT_SYMBOL(dev_add_offload);
479 * __dev_remove_offload - remove offload handler
480 * @po: packet offload declaration
482 * Remove a protocol offload handler that was previously added to the
483 * kernel offload handlers by dev_add_offload(). The passed &offload_type
484 * is removed from the kernel lists and can be freed or reused once this
487 * The packet type might still be in use by receivers
488 * and must not be freed until after all the CPU's have gone
489 * through a quiescent state.
491 static void __dev_remove_offload(struct packet_offload *po)
493 struct list_head *head = &offload_base;
494 struct packet_offload *po1;
496 spin_lock(&offload_lock);
498 list_for_each_entry(po1, head, list) {
500 list_del_rcu(&po->list);
505 pr_warn("dev_remove_offload: %p not found\n", po);
507 spin_unlock(&offload_lock);
511 * dev_remove_offload - remove packet offload handler
512 * @po: packet offload declaration
514 * Remove a packet offload handler that was previously added to the kernel
515 * offload handlers by dev_add_offload(). The passed &offload_type is
516 * removed from the kernel lists and can be freed or reused once this
519 * This call sleeps to guarantee that no CPU is looking at the packet
522 void dev_remove_offload(struct packet_offload *po)
524 __dev_remove_offload(po);
528 EXPORT_SYMBOL(dev_remove_offload);
530 /******************************************************************************
532 Device Boot-time Settings Routines
534 *******************************************************************************/
536 /* Boot time configuration table */
537 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
540 * netdev_boot_setup_add - add new setup entry
541 * @name: name of the device
542 * @map: configured settings for the device
544 * Adds new setup entry to the dev_boot_setup list. The function
545 * returns 0 on error and 1 on success. This is a generic routine to
548 static int netdev_boot_setup_add(char *name, struct ifmap *map)
550 struct netdev_boot_setup *s;
554 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
555 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
556 memset(s[i].name, 0, sizeof(s[i].name));
557 strlcpy(s[i].name, name, IFNAMSIZ);
558 memcpy(&s[i].map, map, sizeof(s[i].map));
563 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
567 * netdev_boot_setup_check - check boot time settings
568 * @dev: the netdevice
570 * Check boot time settings for the device.
571 * The found settings are set for the device to be used
572 * later in the device probing.
573 * Returns 0 if no settings found, 1 if they are.
575 int netdev_boot_setup_check(struct net_device *dev)
577 struct netdev_boot_setup *s = dev_boot_setup;
580 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
581 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
582 !strcmp(dev->name, s[i].name)) {
583 dev->irq = s[i].map.irq;
584 dev->base_addr = s[i].map.base_addr;
585 dev->mem_start = s[i].map.mem_start;
586 dev->mem_end = s[i].map.mem_end;
592 EXPORT_SYMBOL(netdev_boot_setup_check);
596 * netdev_boot_base - get address from boot time settings
597 * @prefix: prefix for network device
598 * @unit: id for network device
600 * Check boot time settings for the base address of device.
601 * The found settings are set for the device to be used
602 * later in the device probing.
603 * Returns 0 if no settings found.
605 unsigned long netdev_boot_base(const char *prefix, int unit)
607 const struct netdev_boot_setup *s = dev_boot_setup;
611 sprintf(name, "%s%d", prefix, unit);
614 * If device already registered then return base of 1
615 * to indicate not to probe for this interface
617 if (__dev_get_by_name(&init_net, name))
620 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
621 if (!strcmp(name, s[i].name))
622 return s[i].map.base_addr;
627 * Saves at boot time configured settings for any netdevice.
629 int __init netdev_boot_setup(char *str)
634 str = get_options(str, ARRAY_SIZE(ints), ints);
639 memset(&map, 0, sizeof(map));
643 map.base_addr = ints[2];
645 map.mem_start = ints[3];
647 map.mem_end = ints[4];
649 /* Add new entry to the list */
650 return netdev_boot_setup_add(str, &map);
653 __setup("netdev=", netdev_boot_setup);
655 /*******************************************************************************
657 Device Interface Subroutines
659 *******************************************************************************/
662 * __dev_get_by_name - find a device by its name
663 * @net: the applicable net namespace
664 * @name: name to find
666 * Find an interface by name. Must be called under RTNL semaphore
667 * or @dev_base_lock. If the name is found a pointer to the device
668 * is returned. If the name is not found then %NULL is returned. The
669 * reference counters are not incremented so the caller must be
670 * careful with locks.
673 struct net_device *__dev_get_by_name(struct net *net, const char *name)
675 struct net_device *dev;
676 struct hlist_head *head = dev_name_hash(net, name);
678 hlist_for_each_entry(dev, head, name_hlist)
679 if (!strncmp(dev->name, name, IFNAMSIZ))
684 EXPORT_SYMBOL(__dev_get_by_name);
687 * dev_get_by_name_rcu - find a device by its name
688 * @net: the applicable net namespace
689 * @name: name to find
691 * Find an interface by name.
692 * If the name is found a pointer to the device is returned.
693 * If the name is not found then %NULL is returned.
694 * The reference counters are not incremented so the caller must be
695 * careful with locks. The caller must hold RCU lock.
698 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
700 struct net_device *dev;
701 struct hlist_head *head = dev_name_hash(net, name);
703 hlist_for_each_entry_rcu(dev, head, name_hlist)
704 if (!strncmp(dev->name, name, IFNAMSIZ))
709 EXPORT_SYMBOL(dev_get_by_name_rcu);
712 * dev_get_by_name - find a device by its name
713 * @net: the applicable net namespace
714 * @name: name to find
716 * Find an interface by name. This can be called from any
717 * context and does its own locking. The returned handle has
718 * the usage count incremented and the caller must use dev_put() to
719 * release it when it is no longer needed. %NULL is returned if no
720 * matching device is found.
723 struct net_device *dev_get_by_name(struct net *net, const char *name)
725 struct net_device *dev;
728 dev = dev_get_by_name_rcu(net, name);
734 EXPORT_SYMBOL(dev_get_by_name);
737 * __dev_get_by_index - find a device by its ifindex
738 * @net: the applicable net namespace
739 * @ifindex: index of device
741 * Search for an interface by index. Returns %NULL if the device
742 * is not found or a pointer to the device. The device has not
743 * had its reference counter increased so the caller must be careful
744 * about locking. The caller must hold either the RTNL semaphore
748 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
750 struct net_device *dev;
751 struct hlist_head *head = dev_index_hash(net, ifindex);
753 hlist_for_each_entry(dev, head, index_hlist)
754 if (dev->ifindex == ifindex)
759 EXPORT_SYMBOL(__dev_get_by_index);
762 * dev_get_by_index_rcu - find a device by its ifindex
763 * @net: the applicable net namespace
764 * @ifindex: index of device
766 * Search for an interface by index. Returns %NULL if the device
767 * is not found or a pointer to the device. The device has not
768 * had its reference counter increased so the caller must be careful
769 * about locking. The caller must hold RCU lock.
772 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
774 struct net_device *dev;
775 struct hlist_head *head = dev_index_hash(net, ifindex);
777 hlist_for_each_entry_rcu(dev, head, index_hlist)
778 if (dev->ifindex == ifindex)
783 EXPORT_SYMBOL(dev_get_by_index_rcu);
787 * dev_get_by_index - find a device by its ifindex
788 * @net: the applicable net namespace
789 * @ifindex: index of device
791 * Search for an interface by index. Returns NULL if the device
792 * is not found or a pointer to the device. The device returned has
793 * had a reference added and the pointer is safe until the user calls
794 * dev_put to indicate they have finished with it.
797 struct net_device *dev_get_by_index(struct net *net, int ifindex)
799 struct net_device *dev;
802 dev = dev_get_by_index_rcu(net, ifindex);
808 EXPORT_SYMBOL(dev_get_by_index);
811 * netdev_get_name - get a netdevice name, knowing its ifindex.
812 * @net: network namespace
813 * @name: a pointer to the buffer where the name will be stored.
814 * @ifindex: the ifindex of the interface to get the name from.
816 * The use of raw_seqcount_begin() and cond_resched() before
817 * retrying is required as we want to give the writers a chance
818 * to complete when CONFIG_PREEMPT is not set.
820 int netdev_get_name(struct net *net, char *name, int ifindex)
822 struct net_device *dev;
826 seq = raw_seqcount_begin(&devnet_rename_seq);
828 dev = dev_get_by_index_rcu(net, ifindex);
834 strcpy(name, dev->name);
836 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
845 * dev_getbyhwaddr_rcu - find a device by its hardware address
846 * @net: the applicable net namespace
847 * @type: media type of device
848 * @ha: hardware address
850 * Search for an interface by MAC address. Returns NULL if the device
851 * is not found or a pointer to the device.
852 * The caller must hold RCU or RTNL.
853 * The returned device has not had its ref count increased
854 * and the caller must therefore be careful about locking
858 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
861 struct net_device *dev;
863 for_each_netdev_rcu(net, dev)
864 if (dev->type == type &&
865 !memcmp(dev->dev_addr, ha, dev->addr_len))
870 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
872 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
874 struct net_device *dev;
877 for_each_netdev(net, dev)
878 if (dev->type == type)
883 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
885 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
887 struct net_device *dev, *ret = NULL;
890 for_each_netdev_rcu(net, dev)
891 if (dev->type == type) {
899 EXPORT_SYMBOL(dev_getfirstbyhwtype);
902 * __dev_get_by_flags - find any device with given flags
903 * @net: the applicable net namespace
904 * @if_flags: IFF_* values
905 * @mask: bitmask of bits in if_flags to check
907 * Search for any interface with the given flags. Returns NULL if a device
908 * is not found or a pointer to the device. Must be called inside
909 * rtnl_lock(), and result refcount is unchanged.
912 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
915 struct net_device *dev, *ret;
920 for_each_netdev(net, dev) {
921 if (((dev->flags ^ if_flags) & mask) == 0) {
928 EXPORT_SYMBOL(__dev_get_by_flags);
931 * dev_valid_name - check if name is okay for network device
934 * Network device names need to be valid file names to
935 * to allow sysfs to work. We also disallow any kind of
938 bool dev_valid_name(const char *name)
942 if (strlen(name) >= IFNAMSIZ)
944 if (!strcmp(name, ".") || !strcmp(name, ".."))
948 if (*name == '/' || isspace(*name))
954 EXPORT_SYMBOL(dev_valid_name);
957 * __dev_alloc_name - allocate a name for a device
958 * @net: network namespace to allocate the device name in
959 * @name: name format string
960 * @buf: scratch buffer and result name string
962 * Passed a format string - eg "lt%d" it will try and find a suitable
963 * id. It scans list of devices to build up a free map, then chooses
964 * the first empty slot. The caller must hold the dev_base or rtnl lock
965 * while allocating the name and adding the device in order to avoid
967 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
968 * Returns the number of the unit assigned or a negative errno code.
971 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
975 const int max_netdevices = 8*PAGE_SIZE;
976 unsigned long *inuse;
977 struct net_device *d;
979 p = strnchr(name, IFNAMSIZ-1, '%');
982 * Verify the string as this thing may have come from
983 * the user. There must be either one "%d" and no other "%"
986 if (p[1] != 'd' || strchr(p + 2, '%'))
989 /* Use one page as a bit array of possible slots */
990 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
994 for_each_netdev(net, d) {
995 if (!sscanf(d->name, name, &i))
997 if (i < 0 || i >= max_netdevices)
1000 /* avoid cases where sscanf is not exact inverse of printf */
1001 snprintf(buf, IFNAMSIZ, name, i);
1002 if (!strncmp(buf, d->name, IFNAMSIZ))
1006 i = find_first_zero_bit(inuse, max_netdevices);
1007 free_page((unsigned long) inuse);
1011 snprintf(buf, IFNAMSIZ, name, i);
1012 if (!__dev_get_by_name(net, buf))
1015 /* It is possible to run out of possible slots
1016 * when the name is long and there isn't enough space left
1017 * for the digits, or if all bits are used.
1023 * dev_alloc_name - allocate a name for a device
1025 * @name: name format string
1027 * Passed a format string - eg "lt%d" it will try and find a suitable
1028 * id. It scans list of devices to build up a free map, then chooses
1029 * the first empty slot. The caller must hold the dev_base or rtnl lock
1030 * while allocating the name and adding the device in order to avoid
1032 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1033 * Returns the number of the unit assigned or a negative errno code.
1036 int dev_alloc_name(struct net_device *dev, const char *name)
1042 BUG_ON(!dev_net(dev));
1044 ret = __dev_alloc_name(net, name, buf);
1046 strlcpy(dev->name, buf, IFNAMSIZ);
1049 EXPORT_SYMBOL(dev_alloc_name);
1051 static int dev_alloc_name_ns(struct net *net,
1052 struct net_device *dev,
1058 ret = __dev_alloc_name(net, name, buf);
1060 strlcpy(dev->name, buf, IFNAMSIZ);
1064 static int dev_get_valid_name(struct net *net,
1065 struct net_device *dev,
1070 if (!dev_valid_name(name))
1073 if (strchr(name, '%'))
1074 return dev_alloc_name_ns(net, dev, name);
1075 else if (__dev_get_by_name(net, name))
1077 else if (dev->name != name)
1078 strlcpy(dev->name, name, IFNAMSIZ);
1084 * dev_change_name - change name of a device
1086 * @newname: name (or format string) must be at least IFNAMSIZ
1088 * Change name of a device, can pass format strings "eth%d".
1091 int dev_change_name(struct net_device *dev, const char *newname)
1093 unsigned char old_assign_type;
1094 char oldname[IFNAMSIZ];
1100 BUG_ON(!dev_net(dev));
1103 if (dev->flags & IFF_UP)
1106 write_seqcount_begin(&devnet_rename_seq);
1108 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1109 write_seqcount_end(&devnet_rename_seq);
1113 memcpy(oldname, dev->name, IFNAMSIZ);
1115 err = dev_get_valid_name(net, dev, newname);
1117 write_seqcount_end(&devnet_rename_seq);
1121 if (oldname[0] && !strchr(oldname, '%'))
1122 netdev_info(dev, "renamed from %s\n", oldname);
1124 old_assign_type = dev->name_assign_type;
1125 dev->name_assign_type = NET_NAME_RENAMED;
1128 ret = device_rename(&dev->dev, dev->name);
1130 memcpy(dev->name, oldname, IFNAMSIZ);
1131 dev->name_assign_type = old_assign_type;
1132 write_seqcount_end(&devnet_rename_seq);
1136 write_seqcount_end(&devnet_rename_seq);
1138 netdev_adjacent_rename_links(dev, oldname);
1140 write_lock_bh(&dev_base_lock);
1141 hlist_del_rcu(&dev->name_hlist);
1142 write_unlock_bh(&dev_base_lock);
1146 write_lock_bh(&dev_base_lock);
1147 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1148 write_unlock_bh(&dev_base_lock);
1150 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1151 ret = notifier_to_errno(ret);
1154 /* err >= 0 after dev_alloc_name() or stores the first errno */
1157 write_seqcount_begin(&devnet_rename_seq);
1158 memcpy(dev->name, oldname, IFNAMSIZ);
1159 memcpy(oldname, newname, IFNAMSIZ);
1160 dev->name_assign_type = old_assign_type;
1161 old_assign_type = NET_NAME_RENAMED;
1164 pr_err("%s: name change rollback failed: %d\n",
1173 * dev_set_alias - change ifalias of a device
1175 * @alias: name up to IFALIASZ
1176 * @len: limit of bytes to copy from info
1178 * Set ifalias for a device,
1180 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1186 if (len >= IFALIASZ)
1190 kfree(dev->ifalias);
1191 dev->ifalias = NULL;
1195 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1198 dev->ifalias = new_ifalias;
1200 strlcpy(dev->ifalias, alias, len+1);
1206 * netdev_features_change - device changes features
1207 * @dev: device to cause notification
1209 * Called to indicate a device has changed features.
1211 void netdev_features_change(struct net_device *dev)
1213 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1215 EXPORT_SYMBOL(netdev_features_change);
1218 * netdev_state_change - device changes state
1219 * @dev: device to cause notification
1221 * Called to indicate a device has changed state. This function calls
1222 * the notifier chains for netdev_chain and sends a NEWLINK message
1223 * to the routing socket.
1225 void netdev_state_change(struct net_device *dev)
1227 if (dev->flags & IFF_UP) {
1228 struct netdev_notifier_change_info change_info;
1230 change_info.flags_changed = 0;
1231 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
1233 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1236 EXPORT_SYMBOL(netdev_state_change);
1239 * netdev_notify_peers - notify network peers about existence of @dev
1240 * @dev: network device
1242 * Generate traffic such that interested network peers are aware of
1243 * @dev, such as by generating a gratuitous ARP. This may be used when
1244 * a device wants to inform the rest of the network about some sort of
1245 * reconfiguration such as a failover event or virtual machine
1248 void netdev_notify_peers(struct net_device *dev)
1251 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1254 EXPORT_SYMBOL(netdev_notify_peers);
1256 static int __dev_open(struct net_device *dev)
1258 const struct net_device_ops *ops = dev->netdev_ops;
1263 if (!netif_device_present(dev))
1266 /* Block netpoll from trying to do any rx path servicing.
1267 * If we don't do this there is a chance ndo_poll_controller
1268 * or ndo_poll may be running while we open the device
1270 netpoll_poll_disable(dev);
1272 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1273 ret = notifier_to_errno(ret);
1277 set_bit(__LINK_STATE_START, &dev->state);
1279 if (ops->ndo_validate_addr)
1280 ret = ops->ndo_validate_addr(dev);
1282 if (!ret && ops->ndo_open)
1283 ret = ops->ndo_open(dev);
1285 netpoll_poll_enable(dev);
1288 clear_bit(__LINK_STATE_START, &dev->state);
1290 dev->flags |= IFF_UP;
1291 dev_set_rx_mode(dev);
1293 add_device_randomness(dev->dev_addr, dev->addr_len);
1300 * dev_open - prepare an interface for use.
1301 * @dev: device to open
1303 * Takes a device from down to up state. The device's private open
1304 * function is invoked and then the multicast lists are loaded. Finally
1305 * the device is moved into the up state and a %NETDEV_UP message is
1306 * sent to the netdev notifier chain.
1308 * Calling this function on an active interface is a nop. On a failure
1309 * a negative errno code is returned.
1311 int dev_open(struct net_device *dev)
1315 if (dev->flags & IFF_UP)
1318 ret = __dev_open(dev);
1322 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1323 call_netdevice_notifiers(NETDEV_UP, dev);
1327 EXPORT_SYMBOL(dev_open);
1329 static int __dev_close_many(struct list_head *head)
1331 struct net_device *dev;
1336 list_for_each_entry(dev, head, close_list) {
1337 /* Temporarily disable netpoll until the interface is down */
1338 netpoll_poll_disable(dev);
1340 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1342 clear_bit(__LINK_STATE_START, &dev->state);
1344 /* Synchronize to scheduled poll. We cannot touch poll list, it
1345 * can be even on different cpu. So just clear netif_running().
1347 * dev->stop() will invoke napi_disable() on all of it's
1348 * napi_struct instances on this device.
1350 smp_mb__after_atomic(); /* Commit netif_running(). */
1353 dev_deactivate_many(head);
1355 list_for_each_entry(dev, head, close_list) {
1356 const struct net_device_ops *ops = dev->netdev_ops;
1359 * Call the device specific close. This cannot fail.
1360 * Only if device is UP
1362 * We allow it to be called even after a DETACH hot-plug
1368 dev->flags &= ~IFF_UP;
1369 netpoll_poll_enable(dev);
1375 static int __dev_close(struct net_device *dev)
1380 list_add(&dev->close_list, &single);
1381 retval = __dev_close_many(&single);
1387 static int dev_close_many(struct list_head *head)
1389 struct net_device *dev, *tmp;
1391 /* Remove the devices that don't need to be closed */
1392 list_for_each_entry_safe(dev, tmp, head, close_list)
1393 if (!(dev->flags & IFF_UP))
1394 list_del_init(&dev->close_list);
1396 __dev_close_many(head);
1398 list_for_each_entry_safe(dev, tmp, head, close_list) {
1399 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1400 call_netdevice_notifiers(NETDEV_DOWN, dev);
1401 list_del_init(&dev->close_list);
1408 * dev_close - shutdown an interface.
1409 * @dev: device to shutdown
1411 * This function moves an active device into down state. A
1412 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1413 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1416 int dev_close(struct net_device *dev)
1418 if (dev->flags & IFF_UP) {
1421 list_add(&dev->close_list, &single);
1422 dev_close_many(&single);
1427 EXPORT_SYMBOL(dev_close);
1431 * dev_disable_lro - disable Large Receive Offload on a device
1434 * Disable Large Receive Offload (LRO) on a net device. Must be
1435 * called under RTNL. This is needed if received packets may be
1436 * forwarded to another interface.
1438 void dev_disable_lro(struct net_device *dev)
1440 struct net_device *lower_dev;
1441 struct list_head *iter;
1443 dev->wanted_features &= ~NETIF_F_LRO;
1444 netdev_update_features(dev);
1446 if (unlikely(dev->features & NETIF_F_LRO))
1447 netdev_WARN(dev, "failed to disable LRO!\n");
1449 netdev_for_each_lower_dev(dev, lower_dev, iter)
1450 dev_disable_lro(lower_dev);
1452 EXPORT_SYMBOL(dev_disable_lro);
1454 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1455 struct net_device *dev)
1457 struct netdev_notifier_info info;
1459 netdev_notifier_info_init(&info, dev);
1460 return nb->notifier_call(nb, val, &info);
1463 static int dev_boot_phase = 1;
1466 * register_netdevice_notifier - register a network notifier block
1469 * Register a notifier to be called when network device events occur.
1470 * The notifier passed is linked into the kernel structures and must
1471 * not be reused until it has been unregistered. A negative errno code
1472 * is returned on a failure.
1474 * When registered all registration and up events are replayed
1475 * to the new notifier to allow device to have a race free
1476 * view of the network device list.
1479 int register_netdevice_notifier(struct notifier_block *nb)
1481 struct net_device *dev;
1482 struct net_device *last;
1487 err = raw_notifier_chain_register(&netdev_chain, nb);
1493 for_each_netdev(net, dev) {
1494 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1495 err = notifier_to_errno(err);
1499 if (!(dev->flags & IFF_UP))
1502 call_netdevice_notifier(nb, NETDEV_UP, dev);
1513 for_each_netdev(net, dev) {
1517 if (dev->flags & IFF_UP) {
1518 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1520 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1522 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1527 raw_notifier_chain_unregister(&netdev_chain, nb);
1530 EXPORT_SYMBOL(register_netdevice_notifier);
1533 * unregister_netdevice_notifier - unregister a network notifier block
1536 * Unregister a notifier previously registered by
1537 * register_netdevice_notifier(). The notifier is unlinked into the
1538 * kernel structures and may then be reused. A negative errno code
1539 * is returned on a failure.
1541 * After unregistering unregister and down device events are synthesized
1542 * for all devices on the device list to the removed notifier to remove
1543 * the need for special case cleanup code.
1546 int unregister_netdevice_notifier(struct notifier_block *nb)
1548 struct net_device *dev;
1553 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1558 for_each_netdev(net, dev) {
1559 if (dev->flags & IFF_UP) {
1560 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1562 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1564 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1571 EXPORT_SYMBOL(unregister_netdevice_notifier);
1574 * call_netdevice_notifiers_info - call all network notifier blocks
1575 * @val: value passed unmodified to notifier function
1576 * @dev: net_device pointer passed unmodified to notifier function
1577 * @info: notifier information data
1579 * Call all network notifier blocks. Parameters and return value
1580 * are as for raw_notifier_call_chain().
1583 static int call_netdevice_notifiers_info(unsigned long val,
1584 struct net_device *dev,
1585 struct netdev_notifier_info *info)
1588 netdev_notifier_info_init(info, dev);
1589 return raw_notifier_call_chain(&netdev_chain, val, info);
1593 * call_netdevice_notifiers - call all network notifier blocks
1594 * @val: value passed unmodified to notifier function
1595 * @dev: net_device pointer passed unmodified to notifier function
1597 * Call all network notifier blocks. Parameters and return value
1598 * are as for raw_notifier_call_chain().
1601 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1603 struct netdev_notifier_info info;
1605 return call_netdevice_notifiers_info(val, dev, &info);
1607 EXPORT_SYMBOL(call_netdevice_notifiers);
1609 static struct static_key netstamp_needed __read_mostly;
1610 #ifdef HAVE_JUMP_LABEL
1611 /* We are not allowed to call static_key_slow_dec() from irq context
1612 * If net_disable_timestamp() is called from irq context, defer the
1613 * static_key_slow_dec() calls.
1615 static atomic_t netstamp_needed_deferred;
1618 void net_enable_timestamp(void)
1620 #ifdef HAVE_JUMP_LABEL
1621 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1625 static_key_slow_dec(&netstamp_needed);
1629 static_key_slow_inc(&netstamp_needed);
1631 EXPORT_SYMBOL(net_enable_timestamp);
1633 void net_disable_timestamp(void)
1635 #ifdef HAVE_JUMP_LABEL
1636 if (in_interrupt()) {
1637 atomic_inc(&netstamp_needed_deferred);
1641 static_key_slow_dec(&netstamp_needed);
1643 EXPORT_SYMBOL(net_disable_timestamp);
1645 static inline void net_timestamp_set(struct sk_buff *skb)
1647 skb->tstamp.tv64 = 0;
1648 if (static_key_false(&netstamp_needed))
1649 __net_timestamp(skb);
1652 #define net_timestamp_check(COND, SKB) \
1653 if (static_key_false(&netstamp_needed)) { \
1654 if ((COND) && !(SKB)->tstamp.tv64) \
1655 __net_timestamp(SKB); \
1658 bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb)
1662 if (!(dev->flags & IFF_UP))
1665 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1666 if (skb->len <= len)
1669 /* if TSO is enabled, we don't care about the length as the packet
1670 * could be forwarded without being segmented before
1672 if (skb_is_gso(skb))
1677 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1679 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1681 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1682 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1683 atomic_long_inc(&dev->rx_dropped);
1689 if (unlikely(!is_skb_forwardable(dev, skb))) {
1690 atomic_long_inc(&dev->rx_dropped);
1695 skb_scrub_packet(skb, true);
1696 skb->protocol = eth_type_trans(skb, dev);
1700 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1703 * dev_forward_skb - loopback an skb to another netif
1705 * @dev: destination network device
1706 * @skb: buffer to forward
1709 * NET_RX_SUCCESS (no congestion)
1710 * NET_RX_DROP (packet was dropped, but freed)
1712 * dev_forward_skb can be used for injecting an skb from the
1713 * start_xmit function of one device into the receive queue
1714 * of another device.
1716 * The receiving device may be in another namespace, so
1717 * we have to clear all information in the skb that could
1718 * impact namespace isolation.
1720 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1722 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1724 EXPORT_SYMBOL_GPL(dev_forward_skb);
1726 static inline int deliver_skb(struct sk_buff *skb,
1727 struct packet_type *pt_prev,
1728 struct net_device *orig_dev)
1730 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1732 atomic_inc(&skb->users);
1733 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1736 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1738 if (!ptype->af_packet_priv || !skb->sk)
1741 if (ptype->id_match)
1742 return ptype->id_match(ptype, skb->sk);
1743 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1750 * Support routine. Sends outgoing frames to any network
1751 * taps currently in use.
1754 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1756 struct packet_type *ptype;
1757 struct sk_buff *skb2 = NULL;
1758 struct packet_type *pt_prev = NULL;
1761 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1762 /* Never send packets back to the socket
1763 * they originated from - MvS (miquels@drinkel.ow.org)
1765 if ((ptype->dev == dev || !ptype->dev) &&
1766 (!skb_loop_sk(ptype, skb))) {
1768 deliver_skb(skb2, pt_prev, skb->dev);
1773 skb2 = skb_clone(skb, GFP_ATOMIC);
1777 net_timestamp_set(skb2);
1779 /* skb->nh should be correctly
1780 set by sender, so that the second statement is
1781 just protection against buggy protocols.
1783 skb_reset_mac_header(skb2);
1785 if (skb_network_header(skb2) < skb2->data ||
1786 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1787 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1788 ntohs(skb2->protocol),
1790 skb_reset_network_header(skb2);
1793 skb2->transport_header = skb2->network_header;
1794 skb2->pkt_type = PACKET_OUTGOING;
1799 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1804 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1805 * @dev: Network device
1806 * @txq: number of queues available
1808 * If real_num_tx_queues is changed the tc mappings may no longer be
1809 * valid. To resolve this verify the tc mapping remains valid and if
1810 * not NULL the mapping. With no priorities mapping to this
1811 * offset/count pair it will no longer be used. In the worst case TC0
1812 * is invalid nothing can be done so disable priority mappings. If is
1813 * expected that drivers will fix this mapping if they can before
1814 * calling netif_set_real_num_tx_queues.
1816 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1819 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1821 /* If TC0 is invalidated disable TC mapping */
1822 if (tc->offset + tc->count > txq) {
1823 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1828 /* Invalidated prio to tc mappings set to TC0 */
1829 for (i = 1; i < TC_BITMASK + 1; i++) {
1830 int q = netdev_get_prio_tc_map(dev, i);
1832 tc = &dev->tc_to_txq[q];
1833 if (tc->offset + tc->count > txq) {
1834 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1836 netdev_set_prio_tc_map(dev, i, 0);
1842 static DEFINE_MUTEX(xps_map_mutex);
1843 #define xmap_dereference(P) \
1844 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1846 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1849 struct xps_map *map = NULL;
1853 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1855 for (pos = 0; map && pos < map->len; pos++) {
1856 if (map->queues[pos] == index) {
1858 map->queues[pos] = map->queues[--map->len];
1860 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1861 kfree_rcu(map, rcu);
1871 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1873 struct xps_dev_maps *dev_maps;
1875 bool active = false;
1877 mutex_lock(&xps_map_mutex);
1878 dev_maps = xmap_dereference(dev->xps_maps);
1883 for_each_possible_cpu(cpu) {
1884 for (i = index; i < dev->num_tx_queues; i++) {
1885 if (!remove_xps_queue(dev_maps, cpu, i))
1888 if (i == dev->num_tx_queues)
1893 RCU_INIT_POINTER(dev->xps_maps, NULL);
1894 kfree_rcu(dev_maps, rcu);
1897 for (i = index; i < dev->num_tx_queues; i++)
1898 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1902 mutex_unlock(&xps_map_mutex);
1905 static struct xps_map *expand_xps_map(struct xps_map *map,
1908 struct xps_map *new_map;
1909 int alloc_len = XPS_MIN_MAP_ALLOC;
1912 for (pos = 0; map && pos < map->len; pos++) {
1913 if (map->queues[pos] != index)
1918 /* Need to add queue to this CPU's existing map */
1920 if (pos < map->alloc_len)
1923 alloc_len = map->alloc_len * 2;
1926 /* Need to allocate new map to store queue on this CPU's map */
1927 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
1932 for (i = 0; i < pos; i++)
1933 new_map->queues[i] = map->queues[i];
1934 new_map->alloc_len = alloc_len;
1940 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
1943 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
1944 struct xps_map *map, *new_map;
1945 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
1946 int cpu, numa_node_id = -2;
1947 bool active = false;
1949 mutex_lock(&xps_map_mutex);
1951 dev_maps = xmap_dereference(dev->xps_maps);
1953 /* allocate memory for queue storage */
1954 for_each_online_cpu(cpu) {
1955 if (!cpumask_test_cpu(cpu, mask))
1959 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
1960 if (!new_dev_maps) {
1961 mutex_unlock(&xps_map_mutex);
1965 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
1968 map = expand_xps_map(map, cpu, index);
1972 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1976 goto out_no_new_maps;
1978 for_each_possible_cpu(cpu) {
1979 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
1980 /* add queue to CPU maps */
1983 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1984 while ((pos < map->len) && (map->queues[pos] != index))
1987 if (pos == map->len)
1988 map->queues[map->len++] = index;
1990 if (numa_node_id == -2)
1991 numa_node_id = cpu_to_node(cpu);
1992 else if (numa_node_id != cpu_to_node(cpu))
1995 } else if (dev_maps) {
1996 /* fill in the new device map from the old device map */
1997 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1998 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2003 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2005 /* Cleanup old maps */
2007 for_each_possible_cpu(cpu) {
2008 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2009 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2010 if (map && map != new_map)
2011 kfree_rcu(map, rcu);
2014 kfree_rcu(dev_maps, rcu);
2017 dev_maps = new_dev_maps;
2021 /* update Tx queue numa node */
2022 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2023 (numa_node_id >= 0) ? numa_node_id :
2029 /* removes queue from unused CPUs */
2030 for_each_possible_cpu(cpu) {
2031 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2034 if (remove_xps_queue(dev_maps, cpu, index))
2038 /* free map if not active */
2040 RCU_INIT_POINTER(dev->xps_maps, NULL);
2041 kfree_rcu(dev_maps, rcu);
2045 mutex_unlock(&xps_map_mutex);
2049 /* remove any maps that we added */
2050 for_each_possible_cpu(cpu) {
2051 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2052 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2054 if (new_map && new_map != map)
2058 mutex_unlock(&xps_map_mutex);
2060 kfree(new_dev_maps);
2063 EXPORT_SYMBOL(netif_set_xps_queue);
2067 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2068 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2070 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2074 if (txq < 1 || txq > dev->num_tx_queues)
2077 if (dev->reg_state == NETREG_REGISTERED ||
2078 dev->reg_state == NETREG_UNREGISTERING) {
2081 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2087 netif_setup_tc(dev, txq);
2089 if (txq < dev->real_num_tx_queues) {
2090 qdisc_reset_all_tx_gt(dev, txq);
2092 netif_reset_xps_queues_gt(dev, txq);
2097 dev->real_num_tx_queues = txq;
2100 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2104 * netif_set_real_num_rx_queues - set actual number of RX queues used
2105 * @dev: Network device
2106 * @rxq: Actual number of RX queues
2108 * This must be called either with the rtnl_lock held or before
2109 * registration of the net device. Returns 0 on success, or a
2110 * negative error code. If called before registration, it always
2113 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2117 if (rxq < 1 || rxq > dev->num_rx_queues)
2120 if (dev->reg_state == NETREG_REGISTERED) {
2123 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2129 dev->real_num_rx_queues = rxq;
2132 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2136 * netif_get_num_default_rss_queues - default number of RSS queues
2138 * This routine should set an upper limit on the number of RSS queues
2139 * used by default by multiqueue devices.
2141 int netif_get_num_default_rss_queues(void)
2143 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2145 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2147 static inline void __netif_reschedule(struct Qdisc *q)
2149 struct softnet_data *sd;
2150 unsigned long flags;
2152 local_irq_save(flags);
2153 sd = this_cpu_ptr(&softnet_data);
2154 q->next_sched = NULL;
2155 *sd->output_queue_tailp = q;
2156 sd->output_queue_tailp = &q->next_sched;
2157 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2158 local_irq_restore(flags);
2161 void __netif_schedule(struct Qdisc *q)
2163 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2164 __netif_reschedule(q);
2166 EXPORT_SYMBOL(__netif_schedule);
2168 struct dev_kfree_skb_cb {
2169 enum skb_free_reason reason;
2172 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2174 return (struct dev_kfree_skb_cb *)skb->cb;
2177 void netif_schedule_queue(struct netdev_queue *txq)
2180 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2181 struct Qdisc *q = rcu_dereference(txq->qdisc);
2183 __netif_schedule(q);
2187 EXPORT_SYMBOL(netif_schedule_queue);
2190 * netif_wake_subqueue - allow sending packets on subqueue
2191 * @dev: network device
2192 * @queue_index: sub queue index
2194 * Resume individual transmit queue of a device with multiple transmit queues.
2196 void netif_wake_subqueue(struct net_device *dev, u16 queue_index)
2198 struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index);
2200 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) {
2204 q = rcu_dereference(txq->qdisc);
2205 __netif_schedule(q);
2209 EXPORT_SYMBOL(netif_wake_subqueue);
2211 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2213 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2217 q = rcu_dereference(dev_queue->qdisc);
2218 __netif_schedule(q);
2222 EXPORT_SYMBOL(netif_tx_wake_queue);
2224 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2226 unsigned long flags;
2228 if (likely(atomic_read(&skb->users) == 1)) {
2230 atomic_set(&skb->users, 0);
2231 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2234 get_kfree_skb_cb(skb)->reason = reason;
2235 local_irq_save(flags);
2236 skb->next = __this_cpu_read(softnet_data.completion_queue);
2237 __this_cpu_write(softnet_data.completion_queue, skb);
2238 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2239 local_irq_restore(flags);
2241 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2243 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2245 if (in_irq() || irqs_disabled())
2246 __dev_kfree_skb_irq(skb, reason);
2250 EXPORT_SYMBOL(__dev_kfree_skb_any);
2254 * netif_device_detach - mark device as removed
2255 * @dev: network device
2257 * Mark device as removed from system and therefore no longer available.
2259 void netif_device_detach(struct net_device *dev)
2261 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2262 netif_running(dev)) {
2263 netif_tx_stop_all_queues(dev);
2266 EXPORT_SYMBOL(netif_device_detach);
2269 * netif_device_attach - mark device as attached
2270 * @dev: network device
2272 * Mark device as attached from system and restart if needed.
2274 void netif_device_attach(struct net_device *dev)
2276 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2277 netif_running(dev)) {
2278 netif_tx_wake_all_queues(dev);
2279 __netdev_watchdog_up(dev);
2282 EXPORT_SYMBOL(netif_device_attach);
2284 static void skb_warn_bad_offload(const struct sk_buff *skb)
2286 static const netdev_features_t null_features = 0;
2287 struct net_device *dev = skb->dev;
2288 const char *driver = "";
2290 if (!net_ratelimit())
2293 if (dev && dev->dev.parent)
2294 driver = dev_driver_string(dev->dev.parent);
2296 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2297 "gso_type=%d ip_summed=%d\n",
2298 driver, dev ? &dev->features : &null_features,
2299 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2300 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2301 skb_shinfo(skb)->gso_type, skb->ip_summed);
2305 * Invalidate hardware checksum when packet is to be mangled, and
2306 * complete checksum manually on outgoing path.
2308 int skb_checksum_help(struct sk_buff *skb)
2311 int ret = 0, offset;
2313 if (skb->ip_summed == CHECKSUM_COMPLETE)
2314 goto out_set_summed;
2316 if (unlikely(skb_shinfo(skb)->gso_size)) {
2317 skb_warn_bad_offload(skb);
2321 /* Before computing a checksum, we should make sure no frag could
2322 * be modified by an external entity : checksum could be wrong.
2324 if (skb_has_shared_frag(skb)) {
2325 ret = __skb_linearize(skb);
2330 offset = skb_checksum_start_offset(skb);
2331 BUG_ON(offset >= skb_headlen(skb));
2332 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2334 offset += skb->csum_offset;
2335 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2337 if (skb_cloned(skb) &&
2338 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2339 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2344 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2346 skb->ip_summed = CHECKSUM_NONE;
2350 EXPORT_SYMBOL(skb_checksum_help);
2352 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2354 unsigned int vlan_depth = skb->mac_len;
2355 __be16 type = skb->protocol;
2357 /* Tunnel gso handlers can set protocol to ethernet. */
2358 if (type == htons(ETH_P_TEB)) {
2361 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2364 eth = (struct ethhdr *)skb_mac_header(skb);
2365 type = eth->h_proto;
2368 /* if skb->protocol is 802.1Q/AD then the header should already be
2369 * present at mac_len - VLAN_HLEN (if mac_len > 0), or at
2370 * ETH_HLEN otherwise
2372 if (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) {
2374 if (WARN_ON(vlan_depth < VLAN_HLEN))
2376 vlan_depth -= VLAN_HLEN;
2378 vlan_depth = ETH_HLEN;
2381 struct vlan_hdr *vh;
2383 if (unlikely(!pskb_may_pull(skb,
2384 vlan_depth + VLAN_HLEN)))
2387 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
2388 type = vh->h_vlan_encapsulated_proto;
2389 vlan_depth += VLAN_HLEN;
2390 } while (type == htons(ETH_P_8021Q) ||
2391 type == htons(ETH_P_8021AD));
2394 *depth = vlan_depth;
2400 * skb_mac_gso_segment - mac layer segmentation handler.
2401 * @skb: buffer to segment
2402 * @features: features for the output path (see dev->features)
2404 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2405 netdev_features_t features)
2407 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2408 struct packet_offload *ptype;
2409 int vlan_depth = skb->mac_len;
2410 __be16 type = skb_network_protocol(skb, &vlan_depth);
2412 if (unlikely(!type))
2413 return ERR_PTR(-EINVAL);
2415 __skb_pull(skb, vlan_depth);
2418 list_for_each_entry_rcu(ptype, &offload_base, list) {
2419 if (ptype->type == type && ptype->callbacks.gso_segment) {
2420 segs = ptype->callbacks.gso_segment(skb, features);
2426 __skb_push(skb, skb->data - skb_mac_header(skb));
2430 EXPORT_SYMBOL(skb_mac_gso_segment);
2433 /* openvswitch calls this on rx path, so we need a different check.
2435 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2438 return skb->ip_summed != CHECKSUM_PARTIAL;
2440 return skb->ip_summed == CHECKSUM_NONE;
2444 * __skb_gso_segment - Perform segmentation on skb.
2445 * @skb: buffer to segment
2446 * @features: features for the output path (see dev->features)
2447 * @tx_path: whether it is called in TX path
2449 * This function segments the given skb and returns a list of segments.
2451 * It may return NULL if the skb requires no segmentation. This is
2452 * only possible when GSO is used for verifying header integrity.
2454 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2455 netdev_features_t features, bool tx_path)
2457 if (unlikely(skb_needs_check(skb, tx_path))) {
2460 skb_warn_bad_offload(skb);
2462 err = skb_cow_head(skb, 0);
2464 return ERR_PTR(err);
2467 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2468 SKB_GSO_CB(skb)->encap_level = 0;
2470 skb_reset_mac_header(skb);
2471 skb_reset_mac_len(skb);
2473 return skb_mac_gso_segment(skb, features);
2475 EXPORT_SYMBOL(__skb_gso_segment);
2477 /* Take action when hardware reception checksum errors are detected. */
2479 void netdev_rx_csum_fault(struct net_device *dev)
2481 if (net_ratelimit()) {
2482 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2486 EXPORT_SYMBOL(netdev_rx_csum_fault);
2489 /* Actually, we should eliminate this check as soon as we know, that:
2490 * 1. IOMMU is present and allows to map all the memory.
2491 * 2. No high memory really exists on this machine.
2494 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2496 #ifdef CONFIG_HIGHMEM
2498 if (!(dev->features & NETIF_F_HIGHDMA)) {
2499 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2500 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2501 if (PageHighMem(skb_frag_page(frag)))
2506 if (PCI_DMA_BUS_IS_PHYS) {
2507 struct device *pdev = dev->dev.parent;
2511 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2512 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2513 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2514 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2522 /* If MPLS offload request, verify we are testing hardware MPLS features
2523 * instead of standard features for the netdev.
2525 #ifdef CONFIG_NET_MPLS_GSO
2526 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2527 netdev_features_t features,
2530 if (eth_p_mpls(type))
2531 features &= skb->dev->mpls_features;
2536 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2537 netdev_features_t features,
2544 static netdev_features_t harmonize_features(struct sk_buff *skb,
2545 netdev_features_t features)
2550 type = skb_network_protocol(skb, &tmp);
2551 features = net_mpls_features(skb, features, type);
2553 if (skb->ip_summed != CHECKSUM_NONE &&
2554 !can_checksum_protocol(features, type)) {
2555 features &= ~NETIF_F_ALL_CSUM;
2556 } else if (illegal_highdma(skb->dev, skb)) {
2557 features &= ~NETIF_F_SG;
2563 netdev_features_t netif_skb_features(struct sk_buff *skb)
2565 const struct net_device *dev = skb->dev;
2566 netdev_features_t features = dev->features;
2567 u16 gso_segs = skb_shinfo(skb)->gso_segs;
2568 __be16 protocol = skb->protocol;
2570 if (gso_segs > dev->gso_max_segs || gso_segs < dev->gso_min_segs)
2571 features &= ~NETIF_F_GSO_MASK;
2573 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) {
2574 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2575 protocol = veh->h_vlan_encapsulated_proto;
2576 } else if (!vlan_tx_tag_present(skb)) {
2577 return harmonize_features(skb, features);
2580 features = netdev_intersect_features(features,
2581 dev->vlan_features |
2582 NETIF_F_HW_VLAN_CTAG_TX |
2583 NETIF_F_HW_VLAN_STAG_TX);
2585 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD))
2586 features = netdev_intersect_features(features,
2591 NETIF_F_HW_VLAN_CTAG_TX |
2592 NETIF_F_HW_VLAN_STAG_TX);
2594 return harmonize_features(skb, features);
2596 EXPORT_SYMBOL(netif_skb_features);
2598 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
2599 struct netdev_queue *txq, bool more)
2604 if (!list_empty(&ptype_all))
2605 dev_queue_xmit_nit(skb, dev);
2608 trace_net_dev_start_xmit(skb, dev);
2609 rc = netdev_start_xmit(skb, dev, txq, more);
2610 trace_net_dev_xmit(skb, rc, dev, len);
2615 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
2616 struct netdev_queue *txq, int *ret)
2618 struct sk_buff *skb = first;
2619 int rc = NETDEV_TX_OK;
2622 struct sk_buff *next = skb->next;
2625 rc = xmit_one(skb, dev, txq, next != NULL);
2626 if (unlikely(!dev_xmit_complete(rc))) {
2632 if (netif_xmit_stopped(txq) && skb) {
2633 rc = NETDEV_TX_BUSY;
2643 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
2644 netdev_features_t features)
2646 if (vlan_tx_tag_present(skb) &&
2647 !vlan_hw_offload_capable(features, skb->vlan_proto))
2648 skb = __vlan_hwaccel_push_inside(skb);
2652 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
2654 netdev_features_t features;
2659 features = netif_skb_features(skb);
2660 skb = validate_xmit_vlan(skb, features);
2664 /* If encapsulation offload request, verify we are testing
2665 * hardware encapsulation features instead of standard
2666 * features for the netdev
2668 if (skb->encapsulation)
2669 features &= dev->hw_enc_features;
2671 if (netif_needs_gso(dev, skb, features)) {
2672 struct sk_buff *segs;
2674 segs = skb_gso_segment(skb, features);
2682 if (skb_needs_linearize(skb, features) &&
2683 __skb_linearize(skb))
2686 /* If packet is not checksummed and device does not
2687 * support checksumming for this protocol, complete
2688 * checksumming here.
2690 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2691 if (skb->encapsulation)
2692 skb_set_inner_transport_header(skb,
2693 skb_checksum_start_offset(skb));
2695 skb_set_transport_header(skb,
2696 skb_checksum_start_offset(skb));
2697 if (!(features & NETIF_F_ALL_CSUM) &&
2698 skb_checksum_help(skb))
2711 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev)
2713 struct sk_buff *next, *head = NULL, *tail;
2715 for (; skb != NULL; skb = next) {
2719 /* in case skb wont be segmented, point to itself */
2722 skb = validate_xmit_skb(skb, dev);
2730 /* If skb was segmented, skb->prev points to
2731 * the last segment. If not, it still contains skb.
2738 static void qdisc_pkt_len_init(struct sk_buff *skb)
2740 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2742 qdisc_skb_cb(skb)->pkt_len = skb->len;
2744 /* To get more precise estimation of bytes sent on wire,
2745 * we add to pkt_len the headers size of all segments
2747 if (shinfo->gso_size) {
2748 unsigned int hdr_len;
2749 u16 gso_segs = shinfo->gso_segs;
2751 /* mac layer + network layer */
2752 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2754 /* + transport layer */
2755 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2756 hdr_len += tcp_hdrlen(skb);
2758 hdr_len += sizeof(struct udphdr);
2760 if (shinfo->gso_type & SKB_GSO_DODGY)
2761 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2764 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2768 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2769 struct net_device *dev,
2770 struct netdev_queue *txq)
2772 spinlock_t *root_lock = qdisc_lock(q);
2776 qdisc_pkt_len_init(skb);
2777 qdisc_calculate_pkt_len(skb, q);
2779 * Heuristic to force contended enqueues to serialize on a
2780 * separate lock before trying to get qdisc main lock.
2781 * This permits __QDISC___STATE_RUNNING owner to get the lock more
2782 * often and dequeue packets faster.
2784 contended = qdisc_is_running(q);
2785 if (unlikely(contended))
2786 spin_lock(&q->busylock);
2788 spin_lock(root_lock);
2789 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2792 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2793 qdisc_run_begin(q)) {
2795 * This is a work-conserving queue; there are no old skbs
2796 * waiting to be sent out; and the qdisc is not running -
2797 * xmit the skb directly.
2800 qdisc_bstats_update(q, skb);
2802 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
2803 if (unlikely(contended)) {
2804 spin_unlock(&q->busylock);
2811 rc = NET_XMIT_SUCCESS;
2813 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2814 if (qdisc_run_begin(q)) {
2815 if (unlikely(contended)) {
2816 spin_unlock(&q->busylock);
2822 spin_unlock(root_lock);
2823 if (unlikely(contended))
2824 spin_unlock(&q->busylock);
2828 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2829 static void skb_update_prio(struct sk_buff *skb)
2831 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2833 if (!skb->priority && skb->sk && map) {
2834 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2836 if (prioidx < map->priomap_len)
2837 skb->priority = map->priomap[prioidx];
2841 #define skb_update_prio(skb)
2844 static DEFINE_PER_CPU(int, xmit_recursion);
2845 #define RECURSION_LIMIT 10
2848 * dev_loopback_xmit - loop back @skb
2849 * @skb: buffer to transmit
2851 int dev_loopback_xmit(struct sk_buff *skb)
2853 skb_reset_mac_header(skb);
2854 __skb_pull(skb, skb_network_offset(skb));
2855 skb->pkt_type = PACKET_LOOPBACK;
2856 skb->ip_summed = CHECKSUM_UNNECESSARY;
2857 WARN_ON(!skb_dst(skb));
2862 EXPORT_SYMBOL(dev_loopback_xmit);
2865 * __dev_queue_xmit - transmit a buffer
2866 * @skb: buffer to transmit
2867 * @accel_priv: private data used for L2 forwarding offload
2869 * Queue a buffer for transmission to a network device. The caller must
2870 * have set the device and priority and built the buffer before calling
2871 * this function. The function can be called from an interrupt.
2873 * A negative errno code is returned on a failure. A success does not
2874 * guarantee the frame will be transmitted as it may be dropped due
2875 * to congestion or traffic shaping.
2877 * -----------------------------------------------------------------------------------
2878 * I notice this method can also return errors from the queue disciplines,
2879 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2882 * Regardless of the return value, the skb is consumed, so it is currently
2883 * difficult to retry a send to this method. (You can bump the ref count
2884 * before sending to hold a reference for retry if you are careful.)
2886 * When calling this method, interrupts MUST be enabled. This is because
2887 * the BH enable code must have IRQs enabled so that it will not deadlock.
2890 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
2892 struct net_device *dev = skb->dev;
2893 struct netdev_queue *txq;
2897 skb_reset_mac_header(skb);
2899 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
2900 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
2902 /* Disable soft irqs for various locks below. Also
2903 * stops preemption for RCU.
2907 skb_update_prio(skb);
2909 /* If device/qdisc don't need skb->dst, release it right now while
2910 * its hot in this cpu cache.
2912 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2917 txq = netdev_pick_tx(dev, skb, accel_priv);
2918 q = rcu_dereference_bh(txq->qdisc);
2920 #ifdef CONFIG_NET_CLS_ACT
2921 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2923 trace_net_dev_queue(skb);
2925 rc = __dev_xmit_skb(skb, q, dev, txq);
2929 /* The device has no queue. Common case for software devices:
2930 loopback, all the sorts of tunnels...
2932 Really, it is unlikely that netif_tx_lock protection is necessary
2933 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2935 However, it is possible, that they rely on protection
2938 Check this and shot the lock. It is not prone from deadlocks.
2939 Either shot noqueue qdisc, it is even simpler 8)
2941 if (dev->flags & IFF_UP) {
2942 int cpu = smp_processor_id(); /* ok because BHs are off */
2944 if (txq->xmit_lock_owner != cpu) {
2946 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2947 goto recursion_alert;
2949 skb = validate_xmit_skb(skb, dev);
2953 HARD_TX_LOCK(dev, txq, cpu);
2955 if (!netif_xmit_stopped(txq)) {
2956 __this_cpu_inc(xmit_recursion);
2957 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
2958 __this_cpu_dec(xmit_recursion);
2959 if (dev_xmit_complete(rc)) {
2960 HARD_TX_UNLOCK(dev, txq);
2964 HARD_TX_UNLOCK(dev, txq);
2965 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2968 /* Recursion is detected! It is possible,
2972 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2979 rcu_read_unlock_bh();
2981 atomic_long_inc(&dev->tx_dropped);
2982 kfree_skb_list(skb);
2985 rcu_read_unlock_bh();
2989 int dev_queue_xmit(struct sk_buff *skb)
2991 return __dev_queue_xmit(skb, NULL);
2993 EXPORT_SYMBOL(dev_queue_xmit);
2995 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
2997 return __dev_queue_xmit(skb, accel_priv);
2999 EXPORT_SYMBOL(dev_queue_xmit_accel);
3002 /*=======================================================================
3004 =======================================================================*/
3006 int netdev_max_backlog __read_mostly = 1000;
3007 EXPORT_SYMBOL(netdev_max_backlog);
3009 int netdev_tstamp_prequeue __read_mostly = 1;
3010 int netdev_budget __read_mostly = 300;
3011 int weight_p __read_mostly = 64; /* old backlog weight */
3013 /* Called with irq disabled */
3014 static inline void ____napi_schedule(struct softnet_data *sd,
3015 struct napi_struct *napi)
3017 list_add_tail(&napi->poll_list, &sd->poll_list);
3018 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3023 /* One global table that all flow-based protocols share. */
3024 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3025 EXPORT_SYMBOL(rps_sock_flow_table);
3027 struct static_key rps_needed __read_mostly;
3029 static struct rps_dev_flow *
3030 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3031 struct rps_dev_flow *rflow, u16 next_cpu)
3033 if (next_cpu != RPS_NO_CPU) {
3034 #ifdef CONFIG_RFS_ACCEL
3035 struct netdev_rx_queue *rxqueue;
3036 struct rps_dev_flow_table *flow_table;
3037 struct rps_dev_flow *old_rflow;
3042 /* Should we steer this flow to a different hardware queue? */
3043 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3044 !(dev->features & NETIF_F_NTUPLE))
3046 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3047 if (rxq_index == skb_get_rx_queue(skb))
3050 rxqueue = dev->_rx + rxq_index;
3051 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3054 flow_id = skb_get_hash(skb) & flow_table->mask;
3055 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3056 rxq_index, flow_id);
3060 rflow = &flow_table->flows[flow_id];
3062 if (old_rflow->filter == rflow->filter)
3063 old_rflow->filter = RPS_NO_FILTER;
3067 per_cpu(softnet_data, next_cpu).input_queue_head;
3070 rflow->cpu = next_cpu;
3075 * get_rps_cpu is called from netif_receive_skb and returns the target
3076 * CPU from the RPS map of the receiving queue for a given skb.
3077 * rcu_read_lock must be held on entry.
3079 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3080 struct rps_dev_flow **rflowp)
3082 struct netdev_rx_queue *rxqueue;
3083 struct rps_map *map;
3084 struct rps_dev_flow_table *flow_table;
3085 struct rps_sock_flow_table *sock_flow_table;
3090 if (skb_rx_queue_recorded(skb)) {
3091 u16 index = skb_get_rx_queue(skb);
3092 if (unlikely(index >= dev->real_num_rx_queues)) {
3093 WARN_ONCE(dev->real_num_rx_queues > 1,
3094 "%s received packet on queue %u, but number "
3095 "of RX queues is %u\n",
3096 dev->name, index, dev->real_num_rx_queues);
3099 rxqueue = dev->_rx + index;
3103 map = rcu_dereference(rxqueue->rps_map);
3105 if (map->len == 1 &&
3106 !rcu_access_pointer(rxqueue->rps_flow_table)) {
3107 tcpu = map->cpus[0];
3108 if (cpu_online(tcpu))
3112 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
3116 skb_reset_network_header(skb);
3117 hash = skb_get_hash(skb);
3121 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3122 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3123 if (flow_table && sock_flow_table) {
3125 struct rps_dev_flow *rflow;
3127 rflow = &flow_table->flows[hash & flow_table->mask];
3130 next_cpu = sock_flow_table->ents[hash & sock_flow_table->mask];
3133 * If the desired CPU (where last recvmsg was done) is
3134 * different from current CPU (one in the rx-queue flow
3135 * table entry), switch if one of the following holds:
3136 * - Current CPU is unset (equal to RPS_NO_CPU).
3137 * - Current CPU is offline.
3138 * - The current CPU's queue tail has advanced beyond the
3139 * last packet that was enqueued using this table entry.
3140 * This guarantees that all previous packets for the flow
3141 * have been dequeued, thus preserving in order delivery.
3143 if (unlikely(tcpu != next_cpu) &&
3144 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
3145 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3146 rflow->last_qtail)) >= 0)) {
3148 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3151 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
3159 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3160 if (cpu_online(tcpu)) {
3170 #ifdef CONFIG_RFS_ACCEL
3173 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3174 * @dev: Device on which the filter was set
3175 * @rxq_index: RX queue index
3176 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3177 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3179 * Drivers that implement ndo_rx_flow_steer() should periodically call
3180 * this function for each installed filter and remove the filters for
3181 * which it returns %true.
3183 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3184 u32 flow_id, u16 filter_id)
3186 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3187 struct rps_dev_flow_table *flow_table;
3188 struct rps_dev_flow *rflow;
3193 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3194 if (flow_table && flow_id <= flow_table->mask) {
3195 rflow = &flow_table->flows[flow_id];
3196 cpu = ACCESS_ONCE(rflow->cpu);
3197 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3198 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3199 rflow->last_qtail) <
3200 (int)(10 * flow_table->mask)))
3206 EXPORT_SYMBOL(rps_may_expire_flow);
3208 #endif /* CONFIG_RFS_ACCEL */
3210 /* Called from hardirq (IPI) context */
3211 static void rps_trigger_softirq(void *data)
3213 struct softnet_data *sd = data;
3215 ____napi_schedule(sd, &sd->backlog);
3219 #endif /* CONFIG_RPS */
3222 * Check if this softnet_data structure is another cpu one
3223 * If yes, queue it to our IPI list and return 1
3226 static int rps_ipi_queued(struct softnet_data *sd)
3229 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3232 sd->rps_ipi_next = mysd->rps_ipi_list;
3233 mysd->rps_ipi_list = sd;
3235 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3238 #endif /* CONFIG_RPS */
3242 #ifdef CONFIG_NET_FLOW_LIMIT
3243 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3246 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3248 #ifdef CONFIG_NET_FLOW_LIMIT
3249 struct sd_flow_limit *fl;
3250 struct softnet_data *sd;
3251 unsigned int old_flow, new_flow;
3253 if (qlen < (netdev_max_backlog >> 1))
3256 sd = this_cpu_ptr(&softnet_data);
3259 fl = rcu_dereference(sd->flow_limit);
3261 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3262 old_flow = fl->history[fl->history_head];
3263 fl->history[fl->history_head] = new_flow;
3266 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3268 if (likely(fl->buckets[old_flow]))
3269 fl->buckets[old_flow]--;
3271 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3283 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3284 * queue (may be a remote CPU queue).
3286 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3287 unsigned int *qtail)
3289 struct softnet_data *sd;
3290 unsigned long flags;
3293 sd = &per_cpu(softnet_data, cpu);
3295 local_irq_save(flags);
3298 qlen = skb_queue_len(&sd->input_pkt_queue);
3299 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3302 __skb_queue_tail(&sd->input_pkt_queue, skb);
3303 input_queue_tail_incr_save(sd, qtail);
3305 local_irq_restore(flags);
3306 return NET_RX_SUCCESS;
3309 /* Schedule NAPI for backlog device
3310 * We can use non atomic operation since we own the queue lock
3312 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3313 if (!rps_ipi_queued(sd))
3314 ____napi_schedule(sd, &sd->backlog);
3322 local_irq_restore(flags);
3324 atomic_long_inc(&skb->dev->rx_dropped);
3329 static int netif_rx_internal(struct sk_buff *skb)
3333 net_timestamp_check(netdev_tstamp_prequeue, skb);
3335 trace_netif_rx(skb);
3337 if (static_key_false(&rps_needed)) {
3338 struct rps_dev_flow voidflow, *rflow = &voidflow;
3344 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3346 cpu = smp_processor_id();
3348 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3356 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3363 * netif_rx - post buffer to the network code
3364 * @skb: buffer to post
3366 * This function receives a packet from a device driver and queues it for
3367 * the upper (protocol) levels to process. It always succeeds. The buffer
3368 * may be dropped during processing for congestion control or by the
3372 * NET_RX_SUCCESS (no congestion)
3373 * NET_RX_DROP (packet was dropped)
3377 int netif_rx(struct sk_buff *skb)
3379 trace_netif_rx_entry(skb);
3381 return netif_rx_internal(skb);
3383 EXPORT_SYMBOL(netif_rx);
3385 int netif_rx_ni(struct sk_buff *skb)
3389 trace_netif_rx_ni_entry(skb);
3392 err = netif_rx_internal(skb);
3393 if (local_softirq_pending())
3399 EXPORT_SYMBOL(netif_rx_ni);
3401 static void net_tx_action(struct softirq_action *h)
3403 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3405 if (sd->completion_queue) {
3406 struct sk_buff *clist;
3408 local_irq_disable();
3409 clist = sd->completion_queue;
3410 sd->completion_queue = NULL;
3414 struct sk_buff *skb = clist;
3415 clist = clist->next;
3417 WARN_ON(atomic_read(&skb->users));
3418 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3419 trace_consume_skb(skb);
3421 trace_kfree_skb(skb, net_tx_action);
3426 if (sd->output_queue) {
3429 local_irq_disable();
3430 head = sd->output_queue;
3431 sd->output_queue = NULL;
3432 sd->output_queue_tailp = &sd->output_queue;
3436 struct Qdisc *q = head;
3437 spinlock_t *root_lock;
3439 head = head->next_sched;
3441 root_lock = qdisc_lock(q);
3442 if (spin_trylock(root_lock)) {
3443 smp_mb__before_atomic();
3444 clear_bit(__QDISC_STATE_SCHED,
3447 spin_unlock(root_lock);
3449 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3451 __netif_reschedule(q);
3453 smp_mb__before_atomic();
3454 clear_bit(__QDISC_STATE_SCHED,
3462 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3463 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3464 /* This hook is defined here for ATM LANE */
3465 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3466 unsigned char *addr) __read_mostly;
3467 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3470 #ifdef CONFIG_NET_CLS_ACT
3471 /* TODO: Maybe we should just force sch_ingress to be compiled in
3472 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3473 * a compare and 2 stores extra right now if we dont have it on
3474 * but have CONFIG_NET_CLS_ACT
3475 * NOTE: This doesn't stop any functionality; if you dont have
3476 * the ingress scheduler, you just can't add policies on ingress.
3479 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3481 struct net_device *dev = skb->dev;
3482 u32 ttl = G_TC_RTTL(skb->tc_verd);
3483 int result = TC_ACT_OK;
3486 if (unlikely(MAX_RED_LOOP < ttl++)) {
3487 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3488 skb->skb_iif, dev->ifindex);
3492 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3493 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3495 q = rcu_dereference(rxq->qdisc);
3496 if (q != &noop_qdisc) {
3497 spin_lock(qdisc_lock(q));
3498 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3499 result = qdisc_enqueue_root(skb, q);
3500 spin_unlock(qdisc_lock(q));
3506 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3507 struct packet_type **pt_prev,
3508 int *ret, struct net_device *orig_dev)
3510 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3512 if (!rxq || rcu_access_pointer(rxq->qdisc) == &noop_qdisc)
3516 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3520 switch (ing_filter(skb, rxq)) {
3534 * netdev_rx_handler_register - register receive handler
3535 * @dev: device to register a handler for
3536 * @rx_handler: receive handler to register
3537 * @rx_handler_data: data pointer that is used by rx handler
3539 * Register a receive handler for a device. This handler will then be
3540 * called from __netif_receive_skb. A negative errno code is returned
3543 * The caller must hold the rtnl_mutex.
3545 * For a general description of rx_handler, see enum rx_handler_result.
3547 int netdev_rx_handler_register(struct net_device *dev,
3548 rx_handler_func_t *rx_handler,
3549 void *rx_handler_data)
3553 if (dev->rx_handler)
3556 /* Note: rx_handler_data must be set before rx_handler */
3557 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3558 rcu_assign_pointer(dev->rx_handler, rx_handler);
3562 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3565 * netdev_rx_handler_unregister - unregister receive handler
3566 * @dev: device to unregister a handler from
3568 * Unregister a receive handler from a device.
3570 * The caller must hold the rtnl_mutex.
3572 void netdev_rx_handler_unregister(struct net_device *dev)
3576 RCU_INIT_POINTER(dev->rx_handler, NULL);
3577 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3578 * section has a guarantee to see a non NULL rx_handler_data
3582 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3584 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3587 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3588 * the special handling of PFMEMALLOC skbs.
3590 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3592 switch (skb->protocol) {
3593 case htons(ETH_P_ARP):
3594 case htons(ETH_P_IP):
3595 case htons(ETH_P_IPV6):
3596 case htons(ETH_P_8021Q):
3597 case htons(ETH_P_8021AD):
3604 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3606 struct packet_type *ptype, *pt_prev;
3607 rx_handler_func_t *rx_handler;
3608 struct net_device *orig_dev;
3609 struct net_device *null_or_dev;
3610 bool deliver_exact = false;
3611 int ret = NET_RX_DROP;
3614 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3616 trace_netif_receive_skb(skb);
3618 orig_dev = skb->dev;
3620 skb_reset_network_header(skb);
3621 if (!skb_transport_header_was_set(skb))
3622 skb_reset_transport_header(skb);
3623 skb_reset_mac_len(skb);
3630 skb->skb_iif = skb->dev->ifindex;
3632 __this_cpu_inc(softnet_data.processed);
3634 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3635 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3636 skb = skb_vlan_untag(skb);
3641 #ifdef CONFIG_NET_CLS_ACT
3642 if (skb->tc_verd & TC_NCLS) {
3643 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3651 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3652 if (!ptype->dev || ptype->dev == skb->dev) {
3654 ret = deliver_skb(skb, pt_prev, orig_dev);
3660 #ifdef CONFIG_NET_CLS_ACT
3661 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3667 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3670 if (vlan_tx_tag_present(skb)) {
3672 ret = deliver_skb(skb, pt_prev, orig_dev);
3675 if (vlan_do_receive(&skb))
3677 else if (unlikely(!skb))
3681 rx_handler = rcu_dereference(skb->dev->rx_handler);
3684 ret = deliver_skb(skb, pt_prev, orig_dev);
3687 switch (rx_handler(&skb)) {
3688 case RX_HANDLER_CONSUMED:
3689 ret = NET_RX_SUCCESS;
3691 case RX_HANDLER_ANOTHER:
3693 case RX_HANDLER_EXACT:
3694 deliver_exact = true;
3695 case RX_HANDLER_PASS:
3702 if (unlikely(vlan_tx_tag_present(skb))) {
3703 if (vlan_tx_tag_get_id(skb))
3704 skb->pkt_type = PACKET_OTHERHOST;
3705 /* Note: we might in the future use prio bits
3706 * and set skb->priority like in vlan_do_receive()
3707 * For the time being, just ignore Priority Code Point
3712 /* deliver only exact match when indicated */
3713 null_or_dev = deliver_exact ? skb->dev : NULL;
3715 type = skb->protocol;
3716 list_for_each_entry_rcu(ptype,
3717 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3718 if (ptype->type == type &&
3719 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3720 ptype->dev == orig_dev)) {
3722 ret = deliver_skb(skb, pt_prev, orig_dev);
3728 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3731 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3734 atomic_long_inc(&skb->dev->rx_dropped);
3736 /* Jamal, now you will not able to escape explaining
3737 * me how you were going to use this. :-)
3747 static int __netif_receive_skb(struct sk_buff *skb)
3751 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3752 unsigned long pflags = current->flags;
3755 * PFMEMALLOC skbs are special, they should
3756 * - be delivered to SOCK_MEMALLOC sockets only
3757 * - stay away from userspace
3758 * - have bounded memory usage
3760 * Use PF_MEMALLOC as this saves us from propagating the allocation
3761 * context down to all allocation sites.
3763 current->flags |= PF_MEMALLOC;
3764 ret = __netif_receive_skb_core(skb, true);
3765 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3767 ret = __netif_receive_skb_core(skb, false);
3772 static int netif_receive_skb_internal(struct sk_buff *skb)
3774 net_timestamp_check(netdev_tstamp_prequeue, skb);
3776 if (skb_defer_rx_timestamp(skb))
3777 return NET_RX_SUCCESS;
3780 if (static_key_false(&rps_needed)) {
3781 struct rps_dev_flow voidflow, *rflow = &voidflow;
3786 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3789 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3796 return __netif_receive_skb(skb);
3800 * netif_receive_skb - process receive buffer from network
3801 * @skb: buffer to process
3803 * netif_receive_skb() is the main receive data processing function.
3804 * It always succeeds. The buffer may be dropped during processing
3805 * for congestion control or by the protocol layers.
3807 * This function may only be called from softirq context and interrupts
3808 * should be enabled.
3810 * Return values (usually ignored):
3811 * NET_RX_SUCCESS: no congestion
3812 * NET_RX_DROP: packet was dropped
3814 int netif_receive_skb(struct sk_buff *skb)
3816 trace_netif_receive_skb_entry(skb);
3818 return netif_receive_skb_internal(skb);
3820 EXPORT_SYMBOL(netif_receive_skb);
3822 /* Network device is going away, flush any packets still pending
3823 * Called with irqs disabled.
3825 static void flush_backlog(void *arg)
3827 struct net_device *dev = arg;
3828 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3829 struct sk_buff *skb, *tmp;
3832 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3833 if (skb->dev == dev) {
3834 __skb_unlink(skb, &sd->input_pkt_queue);
3836 input_queue_head_incr(sd);
3841 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3842 if (skb->dev == dev) {
3843 __skb_unlink(skb, &sd->process_queue);
3845 input_queue_head_incr(sd);
3850 static int napi_gro_complete(struct sk_buff *skb)
3852 struct packet_offload *ptype;
3853 __be16 type = skb->protocol;
3854 struct list_head *head = &offload_base;
3857 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3859 if (NAPI_GRO_CB(skb)->count == 1) {
3860 skb_shinfo(skb)->gso_size = 0;
3865 list_for_each_entry_rcu(ptype, head, list) {
3866 if (ptype->type != type || !ptype->callbacks.gro_complete)
3869 err = ptype->callbacks.gro_complete(skb, 0);
3875 WARN_ON(&ptype->list == head);
3877 return NET_RX_SUCCESS;
3881 return netif_receive_skb_internal(skb);
3884 /* napi->gro_list contains packets ordered by age.
3885 * youngest packets at the head of it.
3886 * Complete skbs in reverse order to reduce latencies.
3888 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3890 struct sk_buff *skb, *prev = NULL;
3892 /* scan list and build reverse chain */
3893 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3898 for (skb = prev; skb; skb = prev) {
3901 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3905 napi_gro_complete(skb);
3909 napi->gro_list = NULL;
3911 EXPORT_SYMBOL(napi_gro_flush);
3913 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3916 unsigned int maclen = skb->dev->hard_header_len;
3917 u32 hash = skb_get_hash_raw(skb);
3919 for (p = napi->gro_list; p; p = p->next) {
3920 unsigned long diffs;
3922 NAPI_GRO_CB(p)->flush = 0;
3924 if (hash != skb_get_hash_raw(p)) {
3925 NAPI_GRO_CB(p)->same_flow = 0;
3929 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3930 diffs |= p->vlan_tci ^ skb->vlan_tci;
3931 if (maclen == ETH_HLEN)
3932 diffs |= compare_ether_header(skb_mac_header(p),
3933 skb_mac_header(skb));
3935 diffs = memcmp(skb_mac_header(p),
3936 skb_mac_header(skb),
3938 NAPI_GRO_CB(p)->same_flow = !diffs;
3942 static void skb_gro_reset_offset(struct sk_buff *skb)
3944 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3945 const skb_frag_t *frag0 = &pinfo->frags[0];
3947 NAPI_GRO_CB(skb)->data_offset = 0;
3948 NAPI_GRO_CB(skb)->frag0 = NULL;
3949 NAPI_GRO_CB(skb)->frag0_len = 0;
3951 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
3953 !PageHighMem(skb_frag_page(frag0))) {
3954 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3955 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3959 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
3961 struct skb_shared_info *pinfo = skb_shinfo(skb);
3963 BUG_ON(skb->end - skb->tail < grow);
3965 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3967 skb->data_len -= grow;
3970 pinfo->frags[0].page_offset += grow;
3971 skb_frag_size_sub(&pinfo->frags[0], grow);
3973 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
3974 skb_frag_unref(skb, 0);
3975 memmove(pinfo->frags, pinfo->frags + 1,
3976 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
3980 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3982 struct sk_buff **pp = NULL;
3983 struct packet_offload *ptype;
3984 __be16 type = skb->protocol;
3985 struct list_head *head = &offload_base;
3987 enum gro_result ret;
3990 if (!(skb->dev->features & NETIF_F_GRO))
3993 if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad)
3996 gro_list_prepare(napi, skb);
3999 list_for_each_entry_rcu(ptype, head, list) {
4000 if (ptype->type != type || !ptype->callbacks.gro_receive)
4003 skb_set_network_header(skb, skb_gro_offset(skb));
4004 skb_reset_mac_len(skb);
4005 NAPI_GRO_CB(skb)->same_flow = 0;
4006 NAPI_GRO_CB(skb)->flush = 0;
4007 NAPI_GRO_CB(skb)->free = 0;
4008 NAPI_GRO_CB(skb)->udp_mark = 0;
4010 /* Setup for GRO checksum validation */
4011 switch (skb->ip_summed) {
4012 case CHECKSUM_COMPLETE:
4013 NAPI_GRO_CB(skb)->csum = skb->csum;
4014 NAPI_GRO_CB(skb)->csum_valid = 1;
4015 NAPI_GRO_CB(skb)->csum_cnt = 0;
4017 case CHECKSUM_UNNECESSARY:
4018 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4019 NAPI_GRO_CB(skb)->csum_valid = 0;
4022 NAPI_GRO_CB(skb)->csum_cnt = 0;
4023 NAPI_GRO_CB(skb)->csum_valid = 0;
4026 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4031 if (&ptype->list == head)
4034 same_flow = NAPI_GRO_CB(skb)->same_flow;
4035 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4038 struct sk_buff *nskb = *pp;
4042 napi_gro_complete(nskb);
4049 if (NAPI_GRO_CB(skb)->flush)
4052 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4053 struct sk_buff *nskb = napi->gro_list;
4055 /* locate the end of the list to select the 'oldest' flow */
4056 while (nskb->next) {
4062 napi_gro_complete(nskb);
4066 NAPI_GRO_CB(skb)->count = 1;
4067 NAPI_GRO_CB(skb)->age = jiffies;
4068 NAPI_GRO_CB(skb)->last = skb;
4069 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4070 skb->next = napi->gro_list;
4071 napi->gro_list = skb;
4075 grow = skb_gro_offset(skb) - skb_headlen(skb);
4077 gro_pull_from_frag0(skb, grow);
4086 struct packet_offload *gro_find_receive_by_type(__be16 type)
4088 struct list_head *offload_head = &offload_base;
4089 struct packet_offload *ptype;
4091 list_for_each_entry_rcu(ptype, offload_head, list) {
4092 if (ptype->type != type || !ptype->callbacks.gro_receive)
4098 EXPORT_SYMBOL(gro_find_receive_by_type);
4100 struct packet_offload *gro_find_complete_by_type(__be16 type)
4102 struct list_head *offload_head = &offload_base;
4103 struct packet_offload *ptype;
4105 list_for_each_entry_rcu(ptype, offload_head, list) {
4106 if (ptype->type != type || !ptype->callbacks.gro_complete)
4112 EXPORT_SYMBOL(gro_find_complete_by_type);
4114 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4118 if (netif_receive_skb_internal(skb))
4126 case GRO_MERGED_FREE:
4127 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4128 kmem_cache_free(skbuff_head_cache, skb);
4141 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4143 trace_napi_gro_receive_entry(skb);
4145 skb_gro_reset_offset(skb);
4147 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4149 EXPORT_SYMBOL(napi_gro_receive);
4151 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4153 if (unlikely(skb->pfmemalloc)) {
4157 __skb_pull(skb, skb_headlen(skb));
4158 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4159 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4161 skb->dev = napi->dev;
4163 skb->encapsulation = 0;
4164 skb_shinfo(skb)->gso_type = 0;
4165 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4170 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4172 struct sk_buff *skb = napi->skb;
4175 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
4180 EXPORT_SYMBOL(napi_get_frags);
4182 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4183 struct sk_buff *skb,
4189 __skb_push(skb, ETH_HLEN);
4190 skb->protocol = eth_type_trans(skb, skb->dev);
4191 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4196 case GRO_MERGED_FREE:
4197 napi_reuse_skb(napi, skb);
4207 /* Upper GRO stack assumes network header starts at gro_offset=0
4208 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4209 * We copy ethernet header into skb->data to have a common layout.
4211 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4213 struct sk_buff *skb = napi->skb;
4214 const struct ethhdr *eth;
4215 unsigned int hlen = sizeof(*eth);
4219 skb_reset_mac_header(skb);
4220 skb_gro_reset_offset(skb);
4222 eth = skb_gro_header_fast(skb, 0);
4223 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4224 eth = skb_gro_header_slow(skb, hlen, 0);
4225 if (unlikely(!eth)) {
4226 napi_reuse_skb(napi, skb);
4230 gro_pull_from_frag0(skb, hlen);
4231 NAPI_GRO_CB(skb)->frag0 += hlen;
4232 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4234 __skb_pull(skb, hlen);
4237 * This works because the only protocols we care about don't require
4239 * We'll fix it up properly in napi_frags_finish()
4241 skb->protocol = eth->h_proto;
4246 gro_result_t napi_gro_frags(struct napi_struct *napi)
4248 struct sk_buff *skb = napi_frags_skb(napi);
4253 trace_napi_gro_frags_entry(skb);
4255 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4257 EXPORT_SYMBOL(napi_gro_frags);
4259 /* Compute the checksum from gro_offset and return the folded value
4260 * after adding in any pseudo checksum.
4262 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
4267 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
4269 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4270 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
4272 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
4273 !skb->csum_complete_sw)
4274 netdev_rx_csum_fault(skb->dev);
4277 NAPI_GRO_CB(skb)->csum = wsum;
4278 NAPI_GRO_CB(skb)->csum_valid = 1;
4282 EXPORT_SYMBOL(__skb_gro_checksum_complete);
4285 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4286 * Note: called with local irq disabled, but exits with local irq enabled.
4288 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4291 struct softnet_data *remsd = sd->rps_ipi_list;
4294 sd->rps_ipi_list = NULL;
4298 /* Send pending IPI's to kick RPS processing on remote cpus. */
4300 struct softnet_data *next = remsd->rps_ipi_next;
4302 if (cpu_online(remsd->cpu))
4303 smp_call_function_single_async(remsd->cpu,
4312 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
4315 return sd->rps_ipi_list != NULL;
4321 static int process_backlog(struct napi_struct *napi, int quota)
4324 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4326 /* Check if we have pending ipi, its better to send them now,
4327 * not waiting net_rx_action() end.
4329 if (sd_has_rps_ipi_waiting(sd)) {
4330 local_irq_disable();
4331 net_rps_action_and_irq_enable(sd);
4334 napi->weight = weight_p;
4335 local_irq_disable();
4337 struct sk_buff *skb;
4339 while ((skb = __skb_dequeue(&sd->process_queue))) {
4341 __netif_receive_skb(skb);
4342 local_irq_disable();
4343 input_queue_head_incr(sd);
4344 if (++work >= quota) {
4351 if (skb_queue_empty(&sd->input_pkt_queue)) {
4353 * Inline a custom version of __napi_complete().
4354 * only current cpu owns and manipulates this napi,
4355 * and NAPI_STATE_SCHED is the only possible flag set
4357 * We can use a plain write instead of clear_bit(),
4358 * and we dont need an smp_mb() memory barrier.
4366 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4367 &sd->process_queue);
4376 * __napi_schedule - schedule for receive
4377 * @n: entry to schedule
4379 * The entry's receive function will be scheduled to run.
4380 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
4382 void __napi_schedule(struct napi_struct *n)
4384 unsigned long flags;
4386 local_irq_save(flags);
4387 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4388 local_irq_restore(flags);
4390 EXPORT_SYMBOL(__napi_schedule);
4393 * __napi_schedule_irqoff - schedule for receive
4394 * @n: entry to schedule
4396 * Variant of __napi_schedule() assuming hard irqs are masked
4398 void __napi_schedule_irqoff(struct napi_struct *n)
4400 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4402 EXPORT_SYMBOL(__napi_schedule_irqoff);
4404 void __napi_complete(struct napi_struct *n)
4406 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4408 list_del_init(&n->poll_list);
4409 smp_mb__before_atomic();
4410 clear_bit(NAPI_STATE_SCHED, &n->state);
4412 EXPORT_SYMBOL(__napi_complete);
4414 void napi_complete_done(struct napi_struct *n, int work_done)
4416 unsigned long flags;
4419 * don't let napi dequeue from the cpu poll list
4420 * just in case its running on a different cpu
4422 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4426 unsigned long timeout = 0;
4429 timeout = n->dev->gro_flush_timeout;
4432 hrtimer_start(&n->timer, ns_to_ktime(timeout),
4433 HRTIMER_MODE_REL_PINNED);
4435 napi_gro_flush(n, false);
4437 if (likely(list_empty(&n->poll_list))) {
4438 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state));
4440 /* If n->poll_list is not empty, we need to mask irqs */
4441 local_irq_save(flags);
4443 local_irq_restore(flags);
4446 EXPORT_SYMBOL(napi_complete_done);
4448 /* must be called under rcu_read_lock(), as we dont take a reference */
4449 struct napi_struct *napi_by_id(unsigned int napi_id)
4451 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4452 struct napi_struct *napi;
4454 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4455 if (napi->napi_id == napi_id)
4460 EXPORT_SYMBOL_GPL(napi_by_id);
4462 void napi_hash_add(struct napi_struct *napi)
4464 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4466 spin_lock(&napi_hash_lock);
4468 /* 0 is not a valid id, we also skip an id that is taken
4469 * we expect both events to be extremely rare
4472 while (!napi->napi_id) {
4473 napi->napi_id = ++napi_gen_id;
4474 if (napi_by_id(napi->napi_id))
4478 hlist_add_head_rcu(&napi->napi_hash_node,
4479 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4481 spin_unlock(&napi_hash_lock);
4484 EXPORT_SYMBOL_GPL(napi_hash_add);
4486 /* Warning : caller is responsible to make sure rcu grace period
4487 * is respected before freeing memory containing @napi
4489 void napi_hash_del(struct napi_struct *napi)
4491 spin_lock(&napi_hash_lock);
4493 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4494 hlist_del_rcu(&napi->napi_hash_node);
4496 spin_unlock(&napi_hash_lock);
4498 EXPORT_SYMBOL_GPL(napi_hash_del);
4500 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
4502 struct napi_struct *napi;
4504 napi = container_of(timer, struct napi_struct, timer);
4506 napi_schedule(napi);
4508 return HRTIMER_NORESTART;
4511 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4512 int (*poll)(struct napi_struct *, int), int weight)
4514 INIT_LIST_HEAD(&napi->poll_list);
4515 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
4516 napi->timer.function = napi_watchdog;
4517 napi->gro_count = 0;
4518 napi->gro_list = NULL;
4521 if (weight > NAPI_POLL_WEIGHT)
4522 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4524 napi->weight = weight;
4525 list_add(&napi->dev_list, &dev->napi_list);
4527 #ifdef CONFIG_NETPOLL
4528 spin_lock_init(&napi->poll_lock);
4529 napi->poll_owner = -1;
4531 set_bit(NAPI_STATE_SCHED, &napi->state);
4533 EXPORT_SYMBOL(netif_napi_add);
4535 void napi_disable(struct napi_struct *n)
4538 set_bit(NAPI_STATE_DISABLE, &n->state);
4540 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
4543 hrtimer_cancel(&n->timer);
4545 clear_bit(NAPI_STATE_DISABLE, &n->state);
4547 EXPORT_SYMBOL(napi_disable);
4549 void netif_napi_del(struct napi_struct *napi)
4551 list_del_init(&napi->dev_list);
4552 napi_free_frags(napi);
4554 kfree_skb_list(napi->gro_list);
4555 napi->gro_list = NULL;
4556 napi->gro_count = 0;
4558 EXPORT_SYMBOL(netif_napi_del);
4560 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
4565 list_del_init(&n->poll_list);
4567 have = netpoll_poll_lock(n);
4571 /* This NAPI_STATE_SCHED test is for avoiding a race
4572 * with netpoll's poll_napi(). Only the entity which
4573 * obtains the lock and sees NAPI_STATE_SCHED set will
4574 * actually make the ->poll() call. Therefore we avoid
4575 * accidentally calling ->poll() when NAPI is not scheduled.
4578 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4579 work = n->poll(n, weight);
4583 WARN_ON_ONCE(work > weight);
4585 if (likely(work < weight))
4588 /* Drivers must not modify the NAPI state if they
4589 * consume the entire weight. In such cases this code
4590 * still "owns" the NAPI instance and therefore can
4591 * move the instance around on the list at-will.
4593 if (unlikely(napi_disable_pending(n))) {
4599 /* flush too old packets
4600 * If HZ < 1000, flush all packets.
4602 napi_gro_flush(n, HZ >= 1000);
4605 /* Some drivers may have called napi_schedule
4606 * prior to exhausting their budget.
4608 if (unlikely(!list_empty(&n->poll_list))) {
4609 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
4610 n->dev ? n->dev->name : "backlog");
4614 list_add_tail(&n->poll_list, repoll);
4617 netpoll_poll_unlock(have);
4622 static void net_rx_action(struct softirq_action *h)
4624 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4625 unsigned long time_limit = jiffies + 2;
4626 int budget = netdev_budget;
4630 local_irq_disable();
4631 list_splice_init(&sd->poll_list, &list);
4634 while (!list_empty(&list)) {
4635 struct napi_struct *n;
4637 n = list_first_entry(&list, struct napi_struct, poll_list);
4638 budget -= napi_poll(n, &repoll);
4640 /* If softirq window is exhausted then punt.
4641 * Allow this to run for 2 jiffies since which will allow
4642 * an average latency of 1.5/HZ.
4644 if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit)))
4648 if (!sd_has_rps_ipi_waiting(sd) &&
4649 list_empty(&list) &&
4650 list_empty(&repoll))
4653 local_irq_disable();
4655 list_splice_tail_init(&sd->poll_list, &list);
4656 list_splice_tail(&repoll, &list);
4657 list_splice(&list, &sd->poll_list);
4658 if (!list_empty(&sd->poll_list))
4659 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4661 net_rps_action_and_irq_enable(sd);
4670 struct netdev_adjacent {
4671 struct net_device *dev;
4673 /* upper master flag, there can only be one master device per list */
4676 /* counter for the number of times this device was added to us */
4679 /* private field for the users */
4682 struct list_head list;
4683 struct rcu_head rcu;
4686 static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev,
4687 struct net_device *adj_dev,
4688 struct list_head *adj_list)
4690 struct netdev_adjacent *adj;
4692 list_for_each_entry(adj, adj_list, list) {
4693 if (adj->dev == adj_dev)
4700 * netdev_has_upper_dev - Check if device is linked to an upper device
4702 * @upper_dev: upper device to check
4704 * Find out if a device is linked to specified upper device and return true
4705 * in case it is. Note that this checks only immediate upper device,
4706 * not through a complete stack of devices. The caller must hold the RTNL lock.
4708 bool netdev_has_upper_dev(struct net_device *dev,
4709 struct net_device *upper_dev)
4713 return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper);
4715 EXPORT_SYMBOL(netdev_has_upper_dev);
4718 * netdev_has_any_upper_dev - Check if device is linked to some device
4721 * Find out if a device is linked to an upper device and return true in case
4722 * it is. The caller must hold the RTNL lock.
4724 static bool netdev_has_any_upper_dev(struct net_device *dev)
4728 return !list_empty(&dev->all_adj_list.upper);
4732 * netdev_master_upper_dev_get - Get master upper device
4735 * Find a master upper device and return pointer to it or NULL in case
4736 * it's not there. The caller must hold the RTNL lock.
4738 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4740 struct netdev_adjacent *upper;
4744 if (list_empty(&dev->adj_list.upper))
4747 upper = list_first_entry(&dev->adj_list.upper,
4748 struct netdev_adjacent, list);
4749 if (likely(upper->master))
4753 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4755 void *netdev_adjacent_get_private(struct list_head *adj_list)
4757 struct netdev_adjacent *adj;
4759 adj = list_entry(adj_list, struct netdev_adjacent, list);
4761 return adj->private;
4763 EXPORT_SYMBOL(netdev_adjacent_get_private);
4766 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
4768 * @iter: list_head ** of the current position
4770 * Gets the next device from the dev's upper list, starting from iter
4771 * position. The caller must hold RCU read lock.
4773 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
4774 struct list_head **iter)
4776 struct netdev_adjacent *upper;
4778 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4780 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4782 if (&upper->list == &dev->adj_list.upper)
4785 *iter = &upper->list;
4789 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
4792 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4794 * @iter: list_head ** of the current position
4796 * Gets the next device from the dev's upper list, starting from iter
4797 * position. The caller must hold RCU read lock.
4799 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
4800 struct list_head **iter)
4802 struct netdev_adjacent *upper;
4804 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4806 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4808 if (&upper->list == &dev->all_adj_list.upper)
4811 *iter = &upper->list;
4815 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
4818 * netdev_lower_get_next_private - Get the next ->private from the
4819 * lower neighbour list
4821 * @iter: list_head ** of the current position
4823 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4824 * list, starting from iter position. The caller must hold either hold the
4825 * RTNL lock or its own locking that guarantees that the neighbour lower
4826 * list will remain unchainged.
4828 void *netdev_lower_get_next_private(struct net_device *dev,
4829 struct list_head **iter)
4831 struct netdev_adjacent *lower;
4833 lower = list_entry(*iter, struct netdev_adjacent, list);
4835 if (&lower->list == &dev->adj_list.lower)
4838 *iter = lower->list.next;
4840 return lower->private;
4842 EXPORT_SYMBOL(netdev_lower_get_next_private);
4845 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4846 * lower neighbour list, RCU
4849 * @iter: list_head ** of the current position
4851 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4852 * list, starting from iter position. The caller must hold RCU read lock.
4854 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
4855 struct list_head **iter)
4857 struct netdev_adjacent *lower;
4859 WARN_ON_ONCE(!rcu_read_lock_held());
4861 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4863 if (&lower->list == &dev->adj_list.lower)
4866 *iter = &lower->list;
4868 return lower->private;
4870 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
4873 * netdev_lower_get_next - Get the next device from the lower neighbour
4876 * @iter: list_head ** of the current position
4878 * Gets the next netdev_adjacent from the dev's lower neighbour
4879 * list, starting from iter position. The caller must hold RTNL lock or
4880 * its own locking that guarantees that the neighbour lower
4881 * list will remain unchainged.
4883 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
4885 struct netdev_adjacent *lower;
4887 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
4889 if (&lower->list == &dev->adj_list.lower)
4892 *iter = &lower->list;
4896 EXPORT_SYMBOL(netdev_lower_get_next);
4899 * netdev_lower_get_first_private_rcu - Get the first ->private from the
4900 * lower neighbour list, RCU
4904 * Gets the first netdev_adjacent->private from the dev's lower neighbour
4905 * list. The caller must hold RCU read lock.
4907 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
4909 struct netdev_adjacent *lower;
4911 lower = list_first_or_null_rcu(&dev->adj_list.lower,
4912 struct netdev_adjacent, list);
4914 return lower->private;
4917 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
4920 * netdev_master_upper_dev_get_rcu - Get master upper device
4923 * Find a master upper device and return pointer to it or NULL in case
4924 * it's not there. The caller must hold the RCU read lock.
4926 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4928 struct netdev_adjacent *upper;
4930 upper = list_first_or_null_rcu(&dev->adj_list.upper,
4931 struct netdev_adjacent, list);
4932 if (upper && likely(upper->master))
4936 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4938 static int netdev_adjacent_sysfs_add(struct net_device *dev,
4939 struct net_device *adj_dev,
4940 struct list_head *dev_list)
4942 char linkname[IFNAMSIZ+7];
4943 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4944 "upper_%s" : "lower_%s", adj_dev->name);
4945 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
4948 static void netdev_adjacent_sysfs_del(struct net_device *dev,
4950 struct list_head *dev_list)
4952 char linkname[IFNAMSIZ+7];
4953 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4954 "upper_%s" : "lower_%s", name);
4955 sysfs_remove_link(&(dev->dev.kobj), linkname);
4958 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
4959 struct net_device *adj_dev,
4960 struct list_head *dev_list)
4962 return (dev_list == &dev->adj_list.upper ||
4963 dev_list == &dev->adj_list.lower) &&
4964 net_eq(dev_net(dev), dev_net(adj_dev));
4967 static int __netdev_adjacent_dev_insert(struct net_device *dev,
4968 struct net_device *adj_dev,
4969 struct list_head *dev_list,
4970 void *private, bool master)
4972 struct netdev_adjacent *adj;
4975 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4982 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
4987 adj->master = master;
4989 adj->private = private;
4992 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
4993 adj_dev->name, dev->name, adj_dev->name);
4995 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
4996 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
5001 /* Ensure that master link is always the first item in list. */
5003 ret = sysfs_create_link(&(dev->dev.kobj),
5004 &(adj_dev->dev.kobj), "master");
5006 goto remove_symlinks;
5008 list_add_rcu(&adj->list, dev_list);
5010 list_add_tail_rcu(&adj->list, dev_list);
5016 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5017 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5025 static void __netdev_adjacent_dev_remove(struct net_device *dev,
5026 struct net_device *adj_dev,
5027 struct list_head *dev_list)
5029 struct netdev_adjacent *adj;
5031 adj = __netdev_find_adj(dev, adj_dev, dev_list);
5034 pr_err("tried to remove device %s from %s\n",
5035 dev->name, adj_dev->name);
5039 if (adj->ref_nr > 1) {
5040 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
5047 sysfs_remove_link(&(dev->dev.kobj), "master");
5049 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5050 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5052 list_del_rcu(&adj->list);
5053 pr_debug("dev_put for %s, because link removed from %s to %s\n",
5054 adj_dev->name, dev->name, adj_dev->name);
5056 kfree_rcu(adj, rcu);
5059 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
5060 struct net_device *upper_dev,
5061 struct list_head *up_list,
5062 struct list_head *down_list,
5063 void *private, bool master)
5067 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
5072 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
5075 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5082 static int __netdev_adjacent_dev_link(struct net_device *dev,
5083 struct net_device *upper_dev)
5085 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
5086 &dev->all_adj_list.upper,
5087 &upper_dev->all_adj_list.lower,
5091 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
5092 struct net_device *upper_dev,
5093 struct list_head *up_list,
5094 struct list_head *down_list)
5096 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5097 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
5100 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
5101 struct net_device *upper_dev)
5103 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5104 &dev->all_adj_list.upper,
5105 &upper_dev->all_adj_list.lower);
5108 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
5109 struct net_device *upper_dev,
5110 void *private, bool master)
5112 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
5117 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
5118 &dev->adj_list.upper,
5119 &upper_dev->adj_list.lower,
5122 __netdev_adjacent_dev_unlink(dev, upper_dev);
5129 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
5130 struct net_device *upper_dev)
5132 __netdev_adjacent_dev_unlink(dev, upper_dev);
5133 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5134 &dev->adj_list.upper,
5135 &upper_dev->adj_list.lower);
5138 static int __netdev_upper_dev_link(struct net_device *dev,
5139 struct net_device *upper_dev, bool master,
5142 struct netdev_adjacent *i, *j, *to_i, *to_j;
5147 if (dev == upper_dev)
5150 /* To prevent loops, check if dev is not upper device to upper_dev. */
5151 if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper))
5154 if (__netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper))
5157 if (master && netdev_master_upper_dev_get(dev))
5160 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
5165 /* Now that we linked these devs, make all the upper_dev's
5166 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5167 * versa, and don't forget the devices itself. All of these
5168 * links are non-neighbours.
5170 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5171 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5172 pr_debug("Interlinking %s with %s, non-neighbour\n",
5173 i->dev->name, j->dev->name);
5174 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
5180 /* add dev to every upper_dev's upper device */
5181 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5182 pr_debug("linking %s's upper device %s with %s\n",
5183 upper_dev->name, i->dev->name, dev->name);
5184 ret = __netdev_adjacent_dev_link(dev, i->dev);
5186 goto rollback_upper_mesh;
5189 /* add upper_dev to every dev's lower device */
5190 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5191 pr_debug("linking %s's lower device %s with %s\n", dev->name,
5192 i->dev->name, upper_dev->name);
5193 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
5195 goto rollback_lower_mesh;
5198 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5201 rollback_lower_mesh:
5203 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5206 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5211 rollback_upper_mesh:
5213 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5216 __netdev_adjacent_dev_unlink(dev, i->dev);
5224 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5225 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5226 if (i == to_i && j == to_j)
5228 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5234 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5240 * netdev_upper_dev_link - Add a link to the upper device
5242 * @upper_dev: new upper device
5244 * Adds a link to device which is upper to this one. The caller must hold
5245 * the RTNL lock. On a failure a negative errno code is returned.
5246 * On success the reference counts are adjusted and the function
5249 int netdev_upper_dev_link(struct net_device *dev,
5250 struct net_device *upper_dev)
5252 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
5254 EXPORT_SYMBOL(netdev_upper_dev_link);
5257 * netdev_master_upper_dev_link - Add a master link to the upper device
5259 * @upper_dev: new upper device
5261 * Adds a link to device which is upper to this one. In this case, only
5262 * one master upper device can be linked, although other non-master devices
5263 * might be linked as well. The caller must hold the RTNL lock.
5264 * On a failure a negative errno code is returned. On success the reference
5265 * counts are adjusted and the function returns zero.
5267 int netdev_master_upper_dev_link(struct net_device *dev,
5268 struct net_device *upper_dev)
5270 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
5272 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5274 int netdev_master_upper_dev_link_private(struct net_device *dev,
5275 struct net_device *upper_dev,
5278 return __netdev_upper_dev_link(dev, upper_dev, true, private);
5280 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
5283 * netdev_upper_dev_unlink - Removes a link to upper device
5285 * @upper_dev: new upper device
5287 * Removes a link to device which is upper to this one. The caller must hold
5290 void netdev_upper_dev_unlink(struct net_device *dev,
5291 struct net_device *upper_dev)
5293 struct netdev_adjacent *i, *j;
5296 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5298 /* Here is the tricky part. We must remove all dev's lower
5299 * devices from all upper_dev's upper devices and vice
5300 * versa, to maintain the graph relationship.
5302 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5303 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
5304 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5306 /* remove also the devices itself from lower/upper device
5309 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5310 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5312 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5313 __netdev_adjacent_dev_unlink(dev, i->dev);
5315 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5317 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5319 void netdev_adjacent_add_links(struct net_device *dev)
5321 struct netdev_adjacent *iter;
5323 struct net *net = dev_net(dev);
5325 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5326 if (!net_eq(net,dev_net(iter->dev)))
5328 netdev_adjacent_sysfs_add(iter->dev, dev,
5329 &iter->dev->adj_list.lower);
5330 netdev_adjacent_sysfs_add(dev, iter->dev,
5331 &dev->adj_list.upper);
5334 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5335 if (!net_eq(net,dev_net(iter->dev)))
5337 netdev_adjacent_sysfs_add(iter->dev, dev,
5338 &iter->dev->adj_list.upper);
5339 netdev_adjacent_sysfs_add(dev, iter->dev,
5340 &dev->adj_list.lower);
5344 void netdev_adjacent_del_links(struct net_device *dev)
5346 struct netdev_adjacent *iter;
5348 struct net *net = dev_net(dev);
5350 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5351 if (!net_eq(net,dev_net(iter->dev)))
5353 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5354 &iter->dev->adj_list.lower);
5355 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5356 &dev->adj_list.upper);
5359 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5360 if (!net_eq(net,dev_net(iter->dev)))
5362 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5363 &iter->dev->adj_list.upper);
5364 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5365 &dev->adj_list.lower);
5369 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5371 struct netdev_adjacent *iter;
5373 struct net *net = dev_net(dev);
5375 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5376 if (!net_eq(net,dev_net(iter->dev)))
5378 netdev_adjacent_sysfs_del(iter->dev, oldname,
5379 &iter->dev->adj_list.lower);
5380 netdev_adjacent_sysfs_add(iter->dev, dev,
5381 &iter->dev->adj_list.lower);
5384 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5385 if (!net_eq(net,dev_net(iter->dev)))
5387 netdev_adjacent_sysfs_del(iter->dev, oldname,
5388 &iter->dev->adj_list.upper);
5389 netdev_adjacent_sysfs_add(iter->dev, dev,
5390 &iter->dev->adj_list.upper);
5394 void *netdev_lower_dev_get_private(struct net_device *dev,
5395 struct net_device *lower_dev)
5397 struct netdev_adjacent *lower;
5401 lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower);
5405 return lower->private;
5407 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5410 int dev_get_nest_level(struct net_device *dev,
5411 bool (*type_check)(struct net_device *dev))
5413 struct net_device *lower = NULL;
5414 struct list_head *iter;
5420 netdev_for_each_lower_dev(dev, lower, iter) {
5421 nest = dev_get_nest_level(lower, type_check);
5422 if (max_nest < nest)
5426 if (type_check(dev))
5431 EXPORT_SYMBOL(dev_get_nest_level);
5433 static void dev_change_rx_flags(struct net_device *dev, int flags)
5435 const struct net_device_ops *ops = dev->netdev_ops;
5437 if (ops->ndo_change_rx_flags)
5438 ops->ndo_change_rx_flags(dev, flags);
5441 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5443 unsigned int old_flags = dev->flags;
5449 dev->flags |= IFF_PROMISC;
5450 dev->promiscuity += inc;
5451 if (dev->promiscuity == 0) {
5454 * If inc causes overflow, untouch promisc and return error.
5457 dev->flags &= ~IFF_PROMISC;
5459 dev->promiscuity -= inc;
5460 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5465 if (dev->flags != old_flags) {
5466 pr_info("device %s %s promiscuous mode\n",
5468 dev->flags & IFF_PROMISC ? "entered" : "left");
5469 if (audit_enabled) {
5470 current_uid_gid(&uid, &gid);
5471 audit_log(current->audit_context, GFP_ATOMIC,
5472 AUDIT_ANOM_PROMISCUOUS,
5473 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5474 dev->name, (dev->flags & IFF_PROMISC),
5475 (old_flags & IFF_PROMISC),
5476 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5477 from_kuid(&init_user_ns, uid),
5478 from_kgid(&init_user_ns, gid),
5479 audit_get_sessionid(current));
5482 dev_change_rx_flags(dev, IFF_PROMISC);
5485 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5490 * dev_set_promiscuity - update promiscuity count on a device
5494 * Add or remove promiscuity from a device. While the count in the device
5495 * remains above zero the interface remains promiscuous. Once it hits zero
5496 * the device reverts back to normal filtering operation. A negative inc
5497 * value is used to drop promiscuity on the device.
5498 * Return 0 if successful or a negative errno code on error.
5500 int dev_set_promiscuity(struct net_device *dev, int inc)
5502 unsigned int old_flags = dev->flags;
5505 err = __dev_set_promiscuity(dev, inc, true);
5508 if (dev->flags != old_flags)
5509 dev_set_rx_mode(dev);
5512 EXPORT_SYMBOL(dev_set_promiscuity);
5514 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5516 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5520 dev->flags |= IFF_ALLMULTI;
5521 dev->allmulti += inc;
5522 if (dev->allmulti == 0) {
5525 * If inc causes overflow, untouch allmulti and return error.
5528 dev->flags &= ~IFF_ALLMULTI;
5530 dev->allmulti -= inc;
5531 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5536 if (dev->flags ^ old_flags) {
5537 dev_change_rx_flags(dev, IFF_ALLMULTI);
5538 dev_set_rx_mode(dev);
5540 __dev_notify_flags(dev, old_flags,
5541 dev->gflags ^ old_gflags);
5547 * dev_set_allmulti - update allmulti count on a device
5551 * Add or remove reception of all multicast frames to a device. While the
5552 * count in the device remains above zero the interface remains listening
5553 * to all interfaces. Once it hits zero the device reverts back to normal
5554 * filtering operation. A negative @inc value is used to drop the counter
5555 * when releasing a resource needing all multicasts.
5556 * Return 0 if successful or a negative errno code on error.
5559 int dev_set_allmulti(struct net_device *dev, int inc)
5561 return __dev_set_allmulti(dev, inc, true);
5563 EXPORT_SYMBOL(dev_set_allmulti);
5566 * Upload unicast and multicast address lists to device and
5567 * configure RX filtering. When the device doesn't support unicast
5568 * filtering it is put in promiscuous mode while unicast addresses
5571 void __dev_set_rx_mode(struct net_device *dev)
5573 const struct net_device_ops *ops = dev->netdev_ops;
5575 /* dev_open will call this function so the list will stay sane. */
5576 if (!(dev->flags&IFF_UP))
5579 if (!netif_device_present(dev))
5582 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5583 /* Unicast addresses changes may only happen under the rtnl,
5584 * therefore calling __dev_set_promiscuity here is safe.
5586 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5587 __dev_set_promiscuity(dev, 1, false);
5588 dev->uc_promisc = true;
5589 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5590 __dev_set_promiscuity(dev, -1, false);
5591 dev->uc_promisc = false;
5595 if (ops->ndo_set_rx_mode)
5596 ops->ndo_set_rx_mode(dev);
5599 void dev_set_rx_mode(struct net_device *dev)
5601 netif_addr_lock_bh(dev);
5602 __dev_set_rx_mode(dev);
5603 netif_addr_unlock_bh(dev);
5607 * dev_get_flags - get flags reported to userspace
5610 * Get the combination of flag bits exported through APIs to userspace.
5612 unsigned int dev_get_flags(const struct net_device *dev)
5616 flags = (dev->flags & ~(IFF_PROMISC |
5621 (dev->gflags & (IFF_PROMISC |
5624 if (netif_running(dev)) {
5625 if (netif_oper_up(dev))
5626 flags |= IFF_RUNNING;
5627 if (netif_carrier_ok(dev))
5628 flags |= IFF_LOWER_UP;
5629 if (netif_dormant(dev))
5630 flags |= IFF_DORMANT;
5635 EXPORT_SYMBOL(dev_get_flags);
5637 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5639 unsigned int old_flags = dev->flags;
5645 * Set the flags on our device.
5648 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5649 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5651 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5655 * Load in the correct multicast list now the flags have changed.
5658 if ((old_flags ^ flags) & IFF_MULTICAST)
5659 dev_change_rx_flags(dev, IFF_MULTICAST);
5661 dev_set_rx_mode(dev);
5664 * Have we downed the interface. We handle IFF_UP ourselves
5665 * according to user attempts to set it, rather than blindly
5670 if ((old_flags ^ flags) & IFF_UP)
5671 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5673 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5674 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5675 unsigned int old_flags = dev->flags;
5677 dev->gflags ^= IFF_PROMISC;
5679 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5680 if (dev->flags != old_flags)
5681 dev_set_rx_mode(dev);
5684 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5685 is important. Some (broken) drivers set IFF_PROMISC, when
5686 IFF_ALLMULTI is requested not asking us and not reporting.
5688 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5689 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5691 dev->gflags ^= IFF_ALLMULTI;
5692 __dev_set_allmulti(dev, inc, false);
5698 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5699 unsigned int gchanges)
5701 unsigned int changes = dev->flags ^ old_flags;
5704 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
5706 if (changes & IFF_UP) {
5707 if (dev->flags & IFF_UP)
5708 call_netdevice_notifiers(NETDEV_UP, dev);
5710 call_netdevice_notifiers(NETDEV_DOWN, dev);
5713 if (dev->flags & IFF_UP &&
5714 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5715 struct netdev_notifier_change_info change_info;
5717 change_info.flags_changed = changes;
5718 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
5724 * dev_change_flags - change device settings
5726 * @flags: device state flags
5728 * Change settings on device based state flags. The flags are
5729 * in the userspace exported format.
5731 int dev_change_flags(struct net_device *dev, unsigned int flags)
5734 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
5736 ret = __dev_change_flags(dev, flags);
5740 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
5741 __dev_notify_flags(dev, old_flags, changes);
5744 EXPORT_SYMBOL(dev_change_flags);
5746 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
5748 const struct net_device_ops *ops = dev->netdev_ops;
5750 if (ops->ndo_change_mtu)
5751 return ops->ndo_change_mtu(dev, new_mtu);
5758 * dev_set_mtu - Change maximum transfer unit
5760 * @new_mtu: new transfer unit
5762 * Change the maximum transfer size of the network device.
5764 int dev_set_mtu(struct net_device *dev, int new_mtu)
5768 if (new_mtu == dev->mtu)
5771 /* MTU must be positive. */
5775 if (!netif_device_present(dev))
5778 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
5779 err = notifier_to_errno(err);
5783 orig_mtu = dev->mtu;
5784 err = __dev_set_mtu(dev, new_mtu);
5787 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5788 err = notifier_to_errno(err);
5790 /* setting mtu back and notifying everyone again,
5791 * so that they have a chance to revert changes.
5793 __dev_set_mtu(dev, orig_mtu);
5794 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5799 EXPORT_SYMBOL(dev_set_mtu);
5802 * dev_set_group - Change group this device belongs to
5804 * @new_group: group this device should belong to
5806 void dev_set_group(struct net_device *dev, int new_group)
5808 dev->group = new_group;
5810 EXPORT_SYMBOL(dev_set_group);
5813 * dev_set_mac_address - Change Media Access Control Address
5817 * Change the hardware (MAC) address of the device
5819 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
5821 const struct net_device_ops *ops = dev->netdev_ops;
5824 if (!ops->ndo_set_mac_address)
5826 if (sa->sa_family != dev->type)
5828 if (!netif_device_present(dev))
5830 err = ops->ndo_set_mac_address(dev, sa);
5833 dev->addr_assign_type = NET_ADDR_SET;
5834 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5835 add_device_randomness(dev->dev_addr, dev->addr_len);
5838 EXPORT_SYMBOL(dev_set_mac_address);
5841 * dev_change_carrier - Change device carrier
5843 * @new_carrier: new value
5845 * Change device carrier
5847 int dev_change_carrier(struct net_device *dev, bool new_carrier)
5849 const struct net_device_ops *ops = dev->netdev_ops;
5851 if (!ops->ndo_change_carrier)
5853 if (!netif_device_present(dev))
5855 return ops->ndo_change_carrier(dev, new_carrier);
5857 EXPORT_SYMBOL(dev_change_carrier);
5860 * dev_get_phys_port_id - Get device physical port ID
5864 * Get device physical port ID
5866 int dev_get_phys_port_id(struct net_device *dev,
5867 struct netdev_phys_item_id *ppid)
5869 const struct net_device_ops *ops = dev->netdev_ops;
5871 if (!ops->ndo_get_phys_port_id)
5873 return ops->ndo_get_phys_port_id(dev, ppid);
5875 EXPORT_SYMBOL(dev_get_phys_port_id);
5878 * dev_new_index - allocate an ifindex
5879 * @net: the applicable net namespace
5881 * Returns a suitable unique value for a new device interface
5882 * number. The caller must hold the rtnl semaphore or the
5883 * dev_base_lock to be sure it remains unique.
5885 static int dev_new_index(struct net *net)
5887 int ifindex = net->ifindex;
5891 if (!__dev_get_by_index(net, ifindex))
5892 return net->ifindex = ifindex;
5896 /* Delayed registration/unregisteration */
5897 static LIST_HEAD(net_todo_list);
5898 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
5900 static void net_set_todo(struct net_device *dev)
5902 list_add_tail(&dev->todo_list, &net_todo_list);
5903 dev_net(dev)->dev_unreg_count++;
5906 static void rollback_registered_many(struct list_head *head)
5908 struct net_device *dev, *tmp;
5909 LIST_HEAD(close_head);
5911 BUG_ON(dev_boot_phase);
5914 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5915 /* Some devices call without registering
5916 * for initialization unwind. Remove those
5917 * devices and proceed with the remaining.
5919 if (dev->reg_state == NETREG_UNINITIALIZED) {
5920 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5924 list_del(&dev->unreg_list);
5927 dev->dismantle = true;
5928 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5931 /* If device is running, close it first. */
5932 list_for_each_entry(dev, head, unreg_list)
5933 list_add_tail(&dev->close_list, &close_head);
5934 dev_close_many(&close_head);
5936 list_for_each_entry(dev, head, unreg_list) {
5937 /* And unlink it from device chain. */
5938 unlist_netdevice(dev);
5940 dev->reg_state = NETREG_UNREGISTERING;
5945 list_for_each_entry(dev, head, unreg_list) {
5946 struct sk_buff *skb = NULL;
5948 /* Shutdown queueing discipline. */
5952 /* Notify protocols, that we are about to destroy
5953 this device. They should clean all the things.
5955 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5957 if (!dev->rtnl_link_ops ||
5958 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5959 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U,
5963 * Flush the unicast and multicast chains
5968 if (dev->netdev_ops->ndo_uninit)
5969 dev->netdev_ops->ndo_uninit(dev);
5972 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
5974 /* Notifier chain MUST detach us all upper devices. */
5975 WARN_ON(netdev_has_any_upper_dev(dev));
5977 /* Remove entries from kobject tree */
5978 netdev_unregister_kobject(dev);
5980 /* Remove XPS queueing entries */
5981 netif_reset_xps_queues_gt(dev, 0);
5987 list_for_each_entry(dev, head, unreg_list)
5991 static void rollback_registered(struct net_device *dev)
5995 list_add(&dev->unreg_list, &single);
5996 rollback_registered_many(&single);
6000 static netdev_features_t netdev_fix_features(struct net_device *dev,
6001 netdev_features_t features)
6003 /* Fix illegal checksum combinations */
6004 if ((features & NETIF_F_HW_CSUM) &&
6005 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6006 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
6007 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
6010 /* TSO requires that SG is present as well. */
6011 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
6012 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
6013 features &= ~NETIF_F_ALL_TSO;
6016 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
6017 !(features & NETIF_F_IP_CSUM)) {
6018 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
6019 features &= ~NETIF_F_TSO;
6020 features &= ~NETIF_F_TSO_ECN;
6023 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
6024 !(features & NETIF_F_IPV6_CSUM)) {
6025 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
6026 features &= ~NETIF_F_TSO6;
6029 /* TSO ECN requires that TSO is present as well. */
6030 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
6031 features &= ~NETIF_F_TSO_ECN;
6033 /* Software GSO depends on SG. */
6034 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
6035 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
6036 features &= ~NETIF_F_GSO;
6039 /* UFO needs SG and checksumming */
6040 if (features & NETIF_F_UFO) {
6041 /* maybe split UFO into V4 and V6? */
6042 if (!((features & NETIF_F_GEN_CSUM) ||
6043 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
6044 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6046 "Dropping NETIF_F_UFO since no checksum offload features.\n");
6047 features &= ~NETIF_F_UFO;
6050 if (!(features & NETIF_F_SG)) {
6052 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
6053 features &= ~NETIF_F_UFO;
6057 #ifdef CONFIG_NET_RX_BUSY_POLL
6058 if (dev->netdev_ops->ndo_busy_poll)
6059 features |= NETIF_F_BUSY_POLL;
6062 features &= ~NETIF_F_BUSY_POLL;
6067 int __netdev_update_features(struct net_device *dev)
6069 netdev_features_t features;
6074 features = netdev_get_wanted_features(dev);
6076 if (dev->netdev_ops->ndo_fix_features)
6077 features = dev->netdev_ops->ndo_fix_features(dev, features);
6079 /* driver might be less strict about feature dependencies */
6080 features = netdev_fix_features(dev, features);
6082 if (dev->features == features)
6085 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
6086 &dev->features, &features);
6088 if (dev->netdev_ops->ndo_set_features)
6089 err = dev->netdev_ops->ndo_set_features(dev, features);
6091 if (unlikely(err < 0)) {
6093 "set_features() failed (%d); wanted %pNF, left %pNF\n",
6094 err, &features, &dev->features);
6099 dev->features = features;
6105 * netdev_update_features - recalculate device features
6106 * @dev: the device to check
6108 * Recalculate dev->features set and send notifications if it
6109 * has changed. Should be called after driver or hardware dependent
6110 * conditions might have changed that influence the features.
6112 void netdev_update_features(struct net_device *dev)
6114 if (__netdev_update_features(dev))
6115 netdev_features_change(dev);
6117 EXPORT_SYMBOL(netdev_update_features);
6120 * netdev_change_features - recalculate device features
6121 * @dev: the device to check
6123 * Recalculate dev->features set and send notifications even
6124 * if they have not changed. Should be called instead of
6125 * netdev_update_features() if also dev->vlan_features might
6126 * have changed to allow the changes to be propagated to stacked
6129 void netdev_change_features(struct net_device *dev)
6131 __netdev_update_features(dev);
6132 netdev_features_change(dev);
6134 EXPORT_SYMBOL(netdev_change_features);
6137 * netif_stacked_transfer_operstate - transfer operstate
6138 * @rootdev: the root or lower level device to transfer state from
6139 * @dev: the device to transfer operstate to
6141 * Transfer operational state from root to device. This is normally
6142 * called when a stacking relationship exists between the root
6143 * device and the device(a leaf device).
6145 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
6146 struct net_device *dev)
6148 if (rootdev->operstate == IF_OPER_DORMANT)
6149 netif_dormant_on(dev);
6151 netif_dormant_off(dev);
6153 if (netif_carrier_ok(rootdev)) {
6154 if (!netif_carrier_ok(dev))
6155 netif_carrier_on(dev);
6157 if (netif_carrier_ok(dev))
6158 netif_carrier_off(dev);
6161 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
6164 static int netif_alloc_rx_queues(struct net_device *dev)
6166 unsigned int i, count = dev->num_rx_queues;
6167 struct netdev_rx_queue *rx;
6171 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
6177 for (i = 0; i < count; i++)
6183 static void netdev_init_one_queue(struct net_device *dev,
6184 struct netdev_queue *queue, void *_unused)
6186 /* Initialize queue lock */
6187 spin_lock_init(&queue->_xmit_lock);
6188 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
6189 queue->xmit_lock_owner = -1;
6190 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
6193 dql_init(&queue->dql, HZ);
6197 static void netif_free_tx_queues(struct net_device *dev)
6202 static int netif_alloc_netdev_queues(struct net_device *dev)
6204 unsigned int count = dev->num_tx_queues;
6205 struct netdev_queue *tx;
6206 size_t sz = count * sizeof(*tx);
6208 BUG_ON(count < 1 || count > 0xffff);
6210 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6218 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
6219 spin_lock_init(&dev->tx_global_lock);
6225 * register_netdevice - register a network device
6226 * @dev: device to register
6228 * Take a completed network device structure and add it to the kernel
6229 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6230 * chain. 0 is returned on success. A negative errno code is returned
6231 * on a failure to set up the device, or if the name is a duplicate.
6233 * Callers must hold the rtnl semaphore. You may want
6234 * register_netdev() instead of this.
6237 * The locking appears insufficient to guarantee two parallel registers
6238 * will not get the same name.
6241 int register_netdevice(struct net_device *dev)
6244 struct net *net = dev_net(dev);
6246 BUG_ON(dev_boot_phase);
6251 /* When net_device's are persistent, this will be fatal. */
6252 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
6255 spin_lock_init(&dev->addr_list_lock);
6256 netdev_set_addr_lockdep_class(dev);
6260 ret = dev_get_valid_name(net, dev, dev->name);
6264 /* Init, if this function is available */
6265 if (dev->netdev_ops->ndo_init) {
6266 ret = dev->netdev_ops->ndo_init(dev);
6274 if (((dev->hw_features | dev->features) &
6275 NETIF_F_HW_VLAN_CTAG_FILTER) &&
6276 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
6277 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
6278 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
6285 dev->ifindex = dev_new_index(net);
6286 else if (__dev_get_by_index(net, dev->ifindex))
6289 if (dev->iflink == -1)
6290 dev->iflink = dev->ifindex;
6292 /* Transfer changeable features to wanted_features and enable
6293 * software offloads (GSO and GRO).
6295 dev->hw_features |= NETIF_F_SOFT_FEATURES;
6296 dev->features |= NETIF_F_SOFT_FEATURES;
6297 dev->wanted_features = dev->features & dev->hw_features;
6299 if (!(dev->flags & IFF_LOOPBACK)) {
6300 dev->hw_features |= NETIF_F_NOCACHE_COPY;
6303 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6305 dev->vlan_features |= NETIF_F_HIGHDMA;
6307 /* Make NETIF_F_SG inheritable to tunnel devices.
6309 dev->hw_enc_features |= NETIF_F_SG;
6311 /* Make NETIF_F_SG inheritable to MPLS.
6313 dev->mpls_features |= NETIF_F_SG;
6315 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
6316 ret = notifier_to_errno(ret);
6320 ret = netdev_register_kobject(dev);
6323 dev->reg_state = NETREG_REGISTERED;
6325 __netdev_update_features(dev);
6328 * Default initial state at registry is that the
6329 * device is present.
6332 set_bit(__LINK_STATE_PRESENT, &dev->state);
6334 linkwatch_init_dev(dev);
6336 dev_init_scheduler(dev);
6338 list_netdevice(dev);
6339 add_device_randomness(dev->dev_addr, dev->addr_len);
6341 /* If the device has permanent device address, driver should
6342 * set dev_addr and also addr_assign_type should be set to
6343 * NET_ADDR_PERM (default value).
6345 if (dev->addr_assign_type == NET_ADDR_PERM)
6346 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
6348 /* Notify protocols, that a new device appeared. */
6349 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
6350 ret = notifier_to_errno(ret);
6352 rollback_registered(dev);
6353 dev->reg_state = NETREG_UNREGISTERED;
6356 * Prevent userspace races by waiting until the network
6357 * device is fully setup before sending notifications.
6359 if (!dev->rtnl_link_ops ||
6360 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6361 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6367 if (dev->netdev_ops->ndo_uninit)
6368 dev->netdev_ops->ndo_uninit(dev);
6371 EXPORT_SYMBOL(register_netdevice);
6374 * init_dummy_netdev - init a dummy network device for NAPI
6375 * @dev: device to init
6377 * This takes a network device structure and initialize the minimum
6378 * amount of fields so it can be used to schedule NAPI polls without
6379 * registering a full blown interface. This is to be used by drivers
6380 * that need to tie several hardware interfaces to a single NAPI
6381 * poll scheduler due to HW limitations.
6383 int init_dummy_netdev(struct net_device *dev)
6385 /* Clear everything. Note we don't initialize spinlocks
6386 * are they aren't supposed to be taken by any of the
6387 * NAPI code and this dummy netdev is supposed to be
6388 * only ever used for NAPI polls
6390 memset(dev, 0, sizeof(struct net_device));
6392 /* make sure we BUG if trying to hit standard
6393 * register/unregister code path
6395 dev->reg_state = NETREG_DUMMY;
6397 /* NAPI wants this */
6398 INIT_LIST_HEAD(&dev->napi_list);
6400 /* a dummy interface is started by default */
6401 set_bit(__LINK_STATE_PRESENT, &dev->state);
6402 set_bit(__LINK_STATE_START, &dev->state);
6404 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6405 * because users of this 'device' dont need to change
6411 EXPORT_SYMBOL_GPL(init_dummy_netdev);
6415 * register_netdev - register a network device
6416 * @dev: device to register
6418 * Take a completed network device structure and add it to the kernel
6419 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6420 * chain. 0 is returned on success. A negative errno code is returned
6421 * on a failure to set up the device, or if the name is a duplicate.
6423 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6424 * and expands the device name if you passed a format string to
6427 int register_netdev(struct net_device *dev)
6432 err = register_netdevice(dev);
6436 EXPORT_SYMBOL(register_netdev);
6438 int netdev_refcnt_read(const struct net_device *dev)
6442 for_each_possible_cpu(i)
6443 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
6446 EXPORT_SYMBOL(netdev_refcnt_read);
6449 * netdev_wait_allrefs - wait until all references are gone.
6450 * @dev: target net_device
6452 * This is called when unregistering network devices.
6454 * Any protocol or device that holds a reference should register
6455 * for netdevice notification, and cleanup and put back the
6456 * reference if they receive an UNREGISTER event.
6457 * We can get stuck here if buggy protocols don't correctly
6460 static void netdev_wait_allrefs(struct net_device *dev)
6462 unsigned long rebroadcast_time, warning_time;
6465 linkwatch_forget_dev(dev);
6467 rebroadcast_time = warning_time = jiffies;
6468 refcnt = netdev_refcnt_read(dev);
6470 while (refcnt != 0) {
6471 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6474 /* Rebroadcast unregister notification */
6475 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6481 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6482 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6484 /* We must not have linkwatch events
6485 * pending on unregister. If this
6486 * happens, we simply run the queue
6487 * unscheduled, resulting in a noop
6490 linkwatch_run_queue();
6495 rebroadcast_time = jiffies;
6500 refcnt = netdev_refcnt_read(dev);
6502 if (time_after(jiffies, warning_time + 10 * HZ)) {
6503 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6505 warning_time = jiffies;
6514 * register_netdevice(x1);
6515 * register_netdevice(x2);
6517 * unregister_netdevice(y1);
6518 * unregister_netdevice(y2);
6524 * We are invoked by rtnl_unlock().
6525 * This allows us to deal with problems:
6526 * 1) We can delete sysfs objects which invoke hotplug
6527 * without deadlocking with linkwatch via keventd.
6528 * 2) Since we run with the RTNL semaphore not held, we can sleep
6529 * safely in order to wait for the netdev refcnt to drop to zero.
6531 * We must not return until all unregister events added during
6532 * the interval the lock was held have been completed.
6534 void netdev_run_todo(void)
6536 struct list_head list;
6538 /* Snapshot list, allow later requests */
6539 list_replace_init(&net_todo_list, &list);
6544 /* Wait for rcu callbacks to finish before next phase */
6545 if (!list_empty(&list))
6548 while (!list_empty(&list)) {
6549 struct net_device *dev
6550 = list_first_entry(&list, struct net_device, todo_list);
6551 list_del(&dev->todo_list);
6554 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6557 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6558 pr_err("network todo '%s' but state %d\n",
6559 dev->name, dev->reg_state);
6564 dev->reg_state = NETREG_UNREGISTERED;
6566 on_each_cpu(flush_backlog, dev, 1);
6568 netdev_wait_allrefs(dev);
6571 BUG_ON(netdev_refcnt_read(dev));
6572 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6573 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6574 WARN_ON(dev->dn_ptr);
6576 if (dev->destructor)
6577 dev->destructor(dev);
6579 /* Report a network device has been unregistered */
6581 dev_net(dev)->dev_unreg_count--;
6583 wake_up(&netdev_unregistering_wq);
6585 /* Free network device */
6586 kobject_put(&dev->dev.kobj);
6590 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6591 * fields in the same order, with only the type differing.
6593 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6594 const struct net_device_stats *netdev_stats)
6596 #if BITS_PER_LONG == 64
6597 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6598 memcpy(stats64, netdev_stats, sizeof(*stats64));
6600 size_t i, n = sizeof(*stats64) / sizeof(u64);
6601 const unsigned long *src = (const unsigned long *)netdev_stats;
6602 u64 *dst = (u64 *)stats64;
6604 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6605 sizeof(*stats64) / sizeof(u64));
6606 for (i = 0; i < n; i++)
6610 EXPORT_SYMBOL(netdev_stats_to_stats64);
6613 * dev_get_stats - get network device statistics
6614 * @dev: device to get statistics from
6615 * @storage: place to store stats
6617 * Get network statistics from device. Return @storage.
6618 * The device driver may provide its own method by setting
6619 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6620 * otherwise the internal statistics structure is used.
6622 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6623 struct rtnl_link_stats64 *storage)
6625 const struct net_device_ops *ops = dev->netdev_ops;
6627 if (ops->ndo_get_stats64) {
6628 memset(storage, 0, sizeof(*storage));
6629 ops->ndo_get_stats64(dev, storage);
6630 } else if (ops->ndo_get_stats) {
6631 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6633 netdev_stats_to_stats64(storage, &dev->stats);
6635 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6636 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
6639 EXPORT_SYMBOL(dev_get_stats);
6641 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
6643 struct netdev_queue *queue = dev_ingress_queue(dev);
6645 #ifdef CONFIG_NET_CLS_ACT
6648 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
6651 netdev_init_one_queue(dev, queue, NULL);
6652 queue->qdisc = &noop_qdisc;
6653 queue->qdisc_sleeping = &noop_qdisc;
6654 rcu_assign_pointer(dev->ingress_queue, queue);
6659 static const struct ethtool_ops default_ethtool_ops;
6661 void netdev_set_default_ethtool_ops(struct net_device *dev,
6662 const struct ethtool_ops *ops)
6664 if (dev->ethtool_ops == &default_ethtool_ops)
6665 dev->ethtool_ops = ops;
6667 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
6669 void netdev_freemem(struct net_device *dev)
6671 char *addr = (char *)dev - dev->padded;
6677 * alloc_netdev_mqs - allocate network device
6678 * @sizeof_priv: size of private data to allocate space for
6679 * @name: device name format string
6680 * @name_assign_type: origin of device name
6681 * @setup: callback to initialize device
6682 * @txqs: the number of TX subqueues to allocate
6683 * @rxqs: the number of RX subqueues to allocate
6685 * Allocates a struct net_device with private data area for driver use
6686 * and performs basic initialization. Also allocates subqueue structs
6687 * for each queue on the device.
6689 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
6690 unsigned char name_assign_type,
6691 void (*setup)(struct net_device *),
6692 unsigned int txqs, unsigned int rxqs)
6694 struct net_device *dev;
6696 struct net_device *p;
6698 BUG_ON(strlen(name) >= sizeof(dev->name));
6701 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6707 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6712 alloc_size = sizeof(struct net_device);
6714 /* ensure 32-byte alignment of private area */
6715 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6716 alloc_size += sizeof_priv;
6718 /* ensure 32-byte alignment of whole construct */
6719 alloc_size += NETDEV_ALIGN - 1;
6721 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6723 p = vzalloc(alloc_size);
6727 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6728 dev->padded = (char *)dev - (char *)p;
6730 dev->pcpu_refcnt = alloc_percpu(int);
6731 if (!dev->pcpu_refcnt)
6734 if (dev_addr_init(dev))
6740 dev_net_set(dev, &init_net);
6742 dev->gso_max_size = GSO_MAX_SIZE;
6743 dev->gso_max_segs = GSO_MAX_SEGS;
6744 dev->gso_min_segs = 0;
6746 INIT_LIST_HEAD(&dev->napi_list);
6747 INIT_LIST_HEAD(&dev->unreg_list);
6748 INIT_LIST_HEAD(&dev->close_list);
6749 INIT_LIST_HEAD(&dev->link_watch_list);
6750 INIT_LIST_HEAD(&dev->adj_list.upper);
6751 INIT_LIST_HEAD(&dev->adj_list.lower);
6752 INIT_LIST_HEAD(&dev->all_adj_list.upper);
6753 INIT_LIST_HEAD(&dev->all_adj_list.lower);
6754 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
6757 dev->num_tx_queues = txqs;
6758 dev->real_num_tx_queues = txqs;
6759 if (netif_alloc_netdev_queues(dev))
6763 dev->num_rx_queues = rxqs;
6764 dev->real_num_rx_queues = rxqs;
6765 if (netif_alloc_rx_queues(dev))
6769 strcpy(dev->name, name);
6770 dev->name_assign_type = name_assign_type;
6771 dev->group = INIT_NETDEV_GROUP;
6772 if (!dev->ethtool_ops)
6773 dev->ethtool_ops = &default_ethtool_ops;
6781 free_percpu(dev->pcpu_refcnt);
6783 netdev_freemem(dev);
6786 EXPORT_SYMBOL(alloc_netdev_mqs);
6789 * free_netdev - free network device
6792 * This function does the last stage of destroying an allocated device
6793 * interface. The reference to the device object is released.
6794 * If this is the last reference then it will be freed.
6796 void free_netdev(struct net_device *dev)
6798 struct napi_struct *p, *n;
6800 release_net(dev_net(dev));
6802 netif_free_tx_queues(dev);
6807 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6809 /* Flush device addresses */
6810 dev_addr_flush(dev);
6812 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6815 free_percpu(dev->pcpu_refcnt);
6816 dev->pcpu_refcnt = NULL;
6818 /* Compatibility with error handling in drivers */
6819 if (dev->reg_state == NETREG_UNINITIALIZED) {
6820 netdev_freemem(dev);
6824 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6825 dev->reg_state = NETREG_RELEASED;
6827 /* will free via device release */
6828 put_device(&dev->dev);
6830 EXPORT_SYMBOL(free_netdev);
6833 * synchronize_net - Synchronize with packet receive processing
6835 * Wait for packets currently being received to be done.
6836 * Does not block later packets from starting.
6838 void synchronize_net(void)
6841 if (rtnl_is_locked())
6842 synchronize_rcu_expedited();
6846 EXPORT_SYMBOL(synchronize_net);
6849 * unregister_netdevice_queue - remove device from the kernel
6853 * This function shuts down a device interface and removes it
6854 * from the kernel tables.
6855 * If head not NULL, device is queued to be unregistered later.
6857 * Callers must hold the rtnl semaphore. You may want
6858 * unregister_netdev() instead of this.
6861 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6866 list_move_tail(&dev->unreg_list, head);
6868 rollback_registered(dev);
6869 /* Finish processing unregister after unlock */
6873 EXPORT_SYMBOL(unregister_netdevice_queue);
6876 * unregister_netdevice_many - unregister many devices
6877 * @head: list of devices
6879 * Note: As most callers use a stack allocated list_head,
6880 * we force a list_del() to make sure stack wont be corrupted later.
6882 void unregister_netdevice_many(struct list_head *head)
6884 struct net_device *dev;
6886 if (!list_empty(head)) {
6887 rollback_registered_many(head);
6888 list_for_each_entry(dev, head, unreg_list)
6893 EXPORT_SYMBOL(unregister_netdevice_many);
6896 * unregister_netdev - remove device from the kernel
6899 * This function shuts down a device interface and removes it
6900 * from the kernel tables.
6902 * This is just a wrapper for unregister_netdevice that takes
6903 * the rtnl semaphore. In general you want to use this and not
6904 * unregister_netdevice.
6906 void unregister_netdev(struct net_device *dev)
6909 unregister_netdevice(dev);
6912 EXPORT_SYMBOL(unregister_netdev);
6915 * dev_change_net_namespace - move device to different nethost namespace
6917 * @net: network namespace
6918 * @pat: If not NULL name pattern to try if the current device name
6919 * is already taken in the destination network namespace.
6921 * This function shuts down a device interface and moves it
6922 * to a new network namespace. On success 0 is returned, on
6923 * a failure a netagive errno code is returned.
6925 * Callers must hold the rtnl semaphore.
6928 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6934 /* Don't allow namespace local devices to be moved. */
6936 if (dev->features & NETIF_F_NETNS_LOCAL)
6939 /* Ensure the device has been registrered */
6940 if (dev->reg_state != NETREG_REGISTERED)
6943 /* Get out if there is nothing todo */
6945 if (net_eq(dev_net(dev), net))
6948 /* Pick the destination device name, and ensure
6949 * we can use it in the destination network namespace.
6952 if (__dev_get_by_name(net, dev->name)) {
6953 /* We get here if we can't use the current device name */
6956 if (dev_get_valid_name(net, dev, pat) < 0)
6961 * And now a mini version of register_netdevice unregister_netdevice.
6964 /* If device is running close it first. */
6967 /* And unlink it from device chain */
6969 unlist_netdevice(dev);
6973 /* Shutdown queueing discipline. */
6976 /* Notify protocols, that we are about to destroy
6977 this device. They should clean all the things.
6979 Note that dev->reg_state stays at NETREG_REGISTERED.
6980 This is wanted because this way 8021q and macvlan know
6981 the device is just moving and can keep their slaves up.
6983 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6985 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6986 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
6989 * Flush the unicast and multicast chains
6994 /* Send a netdev-removed uevent to the old namespace */
6995 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
6996 netdev_adjacent_del_links(dev);
6998 /* Actually switch the network namespace */
6999 dev_net_set(dev, net);
7001 /* If there is an ifindex conflict assign a new one */
7002 if (__dev_get_by_index(net, dev->ifindex)) {
7003 int iflink = (dev->iflink == dev->ifindex);
7004 dev->ifindex = dev_new_index(net);
7006 dev->iflink = dev->ifindex;
7009 /* Send a netdev-add uevent to the new namespace */
7010 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
7011 netdev_adjacent_add_links(dev);
7013 /* Fixup kobjects */
7014 err = device_rename(&dev->dev, dev->name);
7017 /* Add the device back in the hashes */
7018 list_netdevice(dev);
7020 /* Notify protocols, that a new device appeared. */
7021 call_netdevice_notifiers(NETDEV_REGISTER, dev);
7024 * Prevent userspace races by waiting until the network
7025 * device is fully setup before sending notifications.
7027 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7034 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
7036 static int dev_cpu_callback(struct notifier_block *nfb,
7037 unsigned long action,
7040 struct sk_buff **list_skb;
7041 struct sk_buff *skb;
7042 unsigned int cpu, oldcpu = (unsigned long)ocpu;
7043 struct softnet_data *sd, *oldsd;
7045 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
7048 local_irq_disable();
7049 cpu = smp_processor_id();
7050 sd = &per_cpu(softnet_data, cpu);
7051 oldsd = &per_cpu(softnet_data, oldcpu);
7053 /* Find end of our completion_queue. */
7054 list_skb = &sd->completion_queue;
7056 list_skb = &(*list_skb)->next;
7057 /* Append completion queue from offline CPU. */
7058 *list_skb = oldsd->completion_queue;
7059 oldsd->completion_queue = NULL;
7061 /* Append output queue from offline CPU. */
7062 if (oldsd->output_queue) {
7063 *sd->output_queue_tailp = oldsd->output_queue;
7064 sd->output_queue_tailp = oldsd->output_queue_tailp;
7065 oldsd->output_queue = NULL;
7066 oldsd->output_queue_tailp = &oldsd->output_queue;
7068 /* Append NAPI poll list from offline CPU. */
7069 if (!list_empty(&oldsd->poll_list)) {
7070 list_splice_init(&oldsd->poll_list, &sd->poll_list);
7071 raise_softirq_irqoff(NET_RX_SOFTIRQ);
7074 raise_softirq_irqoff(NET_TX_SOFTIRQ);
7077 /* Process offline CPU's input_pkt_queue */
7078 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
7079 netif_rx_internal(skb);
7080 input_queue_head_incr(oldsd);
7082 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
7083 netif_rx_internal(skb);
7084 input_queue_head_incr(oldsd);
7092 * netdev_increment_features - increment feature set by one
7093 * @all: current feature set
7094 * @one: new feature set
7095 * @mask: mask feature set
7097 * Computes a new feature set after adding a device with feature set
7098 * @one to the master device with current feature set @all. Will not
7099 * enable anything that is off in @mask. Returns the new feature set.
7101 netdev_features_t netdev_increment_features(netdev_features_t all,
7102 netdev_features_t one, netdev_features_t mask)
7104 if (mask & NETIF_F_GEN_CSUM)
7105 mask |= NETIF_F_ALL_CSUM;
7106 mask |= NETIF_F_VLAN_CHALLENGED;
7108 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
7109 all &= one | ~NETIF_F_ALL_FOR_ALL;
7111 /* If one device supports hw checksumming, set for all. */
7112 if (all & NETIF_F_GEN_CSUM)
7113 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
7117 EXPORT_SYMBOL(netdev_increment_features);
7119 static struct hlist_head * __net_init netdev_create_hash(void)
7122 struct hlist_head *hash;
7124 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
7126 for (i = 0; i < NETDEV_HASHENTRIES; i++)
7127 INIT_HLIST_HEAD(&hash[i]);
7132 /* Initialize per network namespace state */
7133 static int __net_init netdev_init(struct net *net)
7135 if (net != &init_net)
7136 INIT_LIST_HEAD(&net->dev_base_head);
7138 net->dev_name_head = netdev_create_hash();
7139 if (net->dev_name_head == NULL)
7142 net->dev_index_head = netdev_create_hash();
7143 if (net->dev_index_head == NULL)
7149 kfree(net->dev_name_head);
7155 * netdev_drivername - network driver for the device
7156 * @dev: network device
7158 * Determine network driver for device.
7160 const char *netdev_drivername(const struct net_device *dev)
7162 const struct device_driver *driver;
7163 const struct device *parent;
7164 const char *empty = "";
7166 parent = dev->dev.parent;
7170 driver = parent->driver;
7171 if (driver && driver->name)
7172 return driver->name;
7176 static void __netdev_printk(const char *level, const struct net_device *dev,
7177 struct va_format *vaf)
7179 if (dev && dev->dev.parent) {
7180 dev_printk_emit(level[1] - '0',
7183 dev_driver_string(dev->dev.parent),
7184 dev_name(dev->dev.parent),
7185 netdev_name(dev), netdev_reg_state(dev),
7188 printk("%s%s%s: %pV",
7189 level, netdev_name(dev), netdev_reg_state(dev), vaf);
7191 printk("%s(NULL net_device): %pV", level, vaf);
7195 void netdev_printk(const char *level, const struct net_device *dev,
7196 const char *format, ...)
7198 struct va_format vaf;
7201 va_start(args, format);
7206 __netdev_printk(level, dev, &vaf);
7210 EXPORT_SYMBOL(netdev_printk);
7212 #define define_netdev_printk_level(func, level) \
7213 void func(const struct net_device *dev, const char *fmt, ...) \
7215 struct va_format vaf; \
7218 va_start(args, fmt); \
7223 __netdev_printk(level, dev, &vaf); \
7227 EXPORT_SYMBOL(func);
7229 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
7230 define_netdev_printk_level(netdev_alert, KERN_ALERT);
7231 define_netdev_printk_level(netdev_crit, KERN_CRIT);
7232 define_netdev_printk_level(netdev_err, KERN_ERR);
7233 define_netdev_printk_level(netdev_warn, KERN_WARNING);
7234 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
7235 define_netdev_printk_level(netdev_info, KERN_INFO);
7237 static void __net_exit netdev_exit(struct net *net)
7239 kfree(net->dev_name_head);
7240 kfree(net->dev_index_head);
7243 static struct pernet_operations __net_initdata netdev_net_ops = {
7244 .init = netdev_init,
7245 .exit = netdev_exit,
7248 static void __net_exit default_device_exit(struct net *net)
7250 struct net_device *dev, *aux;
7252 * Push all migratable network devices back to the
7253 * initial network namespace
7256 for_each_netdev_safe(net, dev, aux) {
7258 char fb_name[IFNAMSIZ];
7260 /* Ignore unmoveable devices (i.e. loopback) */
7261 if (dev->features & NETIF_F_NETNS_LOCAL)
7264 /* Leave virtual devices for the generic cleanup */
7265 if (dev->rtnl_link_ops)
7268 /* Push remaining network devices to init_net */
7269 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
7270 err = dev_change_net_namespace(dev, &init_net, fb_name);
7272 pr_emerg("%s: failed to move %s to init_net: %d\n",
7273 __func__, dev->name, err);
7280 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
7282 /* Return with the rtnl_lock held when there are no network
7283 * devices unregistering in any network namespace in net_list.
7287 DEFINE_WAIT_FUNC(wait, woken_wake_function);
7289 add_wait_queue(&netdev_unregistering_wq, &wait);
7291 unregistering = false;
7293 list_for_each_entry(net, net_list, exit_list) {
7294 if (net->dev_unreg_count > 0) {
7295 unregistering = true;
7303 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
7305 remove_wait_queue(&netdev_unregistering_wq, &wait);
7308 static void __net_exit default_device_exit_batch(struct list_head *net_list)
7310 /* At exit all network devices most be removed from a network
7311 * namespace. Do this in the reverse order of registration.
7312 * Do this across as many network namespaces as possible to
7313 * improve batching efficiency.
7315 struct net_device *dev;
7317 LIST_HEAD(dev_kill_list);
7319 /* To prevent network device cleanup code from dereferencing
7320 * loopback devices or network devices that have been freed
7321 * wait here for all pending unregistrations to complete,
7322 * before unregistring the loopback device and allowing the
7323 * network namespace be freed.
7325 * The netdev todo list containing all network devices
7326 * unregistrations that happen in default_device_exit_batch
7327 * will run in the rtnl_unlock() at the end of
7328 * default_device_exit_batch.
7330 rtnl_lock_unregistering(net_list);
7331 list_for_each_entry(net, net_list, exit_list) {
7332 for_each_netdev_reverse(net, dev) {
7333 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
7334 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
7336 unregister_netdevice_queue(dev, &dev_kill_list);
7339 unregister_netdevice_many(&dev_kill_list);
7343 static struct pernet_operations __net_initdata default_device_ops = {
7344 .exit = default_device_exit,
7345 .exit_batch = default_device_exit_batch,
7349 * Initialize the DEV module. At boot time this walks the device list and
7350 * unhooks any devices that fail to initialise (normally hardware not
7351 * present) and leaves us with a valid list of present and active devices.
7356 * This is called single threaded during boot, so no need
7357 * to take the rtnl semaphore.
7359 static int __init net_dev_init(void)
7361 int i, rc = -ENOMEM;
7363 BUG_ON(!dev_boot_phase);
7365 if (dev_proc_init())
7368 if (netdev_kobject_init())
7371 INIT_LIST_HEAD(&ptype_all);
7372 for (i = 0; i < PTYPE_HASH_SIZE; i++)
7373 INIT_LIST_HEAD(&ptype_base[i]);
7375 INIT_LIST_HEAD(&offload_base);
7377 if (register_pernet_subsys(&netdev_net_ops))
7381 * Initialise the packet receive queues.
7384 for_each_possible_cpu(i) {
7385 struct softnet_data *sd = &per_cpu(softnet_data, i);
7387 skb_queue_head_init(&sd->input_pkt_queue);
7388 skb_queue_head_init(&sd->process_queue);
7389 INIT_LIST_HEAD(&sd->poll_list);
7390 sd->output_queue_tailp = &sd->output_queue;
7392 sd->csd.func = rps_trigger_softirq;
7397 sd->backlog.poll = process_backlog;
7398 sd->backlog.weight = weight_p;
7403 /* The loopback device is special if any other network devices
7404 * is present in a network namespace the loopback device must
7405 * be present. Since we now dynamically allocate and free the
7406 * loopback device ensure this invariant is maintained by
7407 * keeping the loopback device as the first device on the
7408 * list of network devices. Ensuring the loopback devices
7409 * is the first device that appears and the last network device
7412 if (register_pernet_device(&loopback_net_ops))
7415 if (register_pernet_device(&default_device_ops))
7418 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
7419 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
7421 hotcpu_notifier(dev_cpu_callback, 0);
7428 subsys_initcall(net_dev_init);