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 #if IS_ENABLED(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 (!vlan_tx_tag_present(skb)) {
2574 if (unlikely(protocol == htons(ETH_P_8021Q) ||
2575 protocol == htons(ETH_P_8021AD))) {
2576 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2577 protocol = veh->h_vlan_encapsulated_proto;
2579 return harmonize_features(skb, features);
2583 features = netdev_intersect_features(features,
2584 dev->vlan_features |
2585 NETIF_F_HW_VLAN_CTAG_TX |
2586 NETIF_F_HW_VLAN_STAG_TX);
2588 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD))
2589 features = netdev_intersect_features(features,
2594 NETIF_F_HW_VLAN_CTAG_TX |
2595 NETIF_F_HW_VLAN_STAG_TX);
2597 return harmonize_features(skb, features);
2599 EXPORT_SYMBOL(netif_skb_features);
2601 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
2602 struct netdev_queue *txq, bool more)
2607 if (!list_empty(&ptype_all))
2608 dev_queue_xmit_nit(skb, dev);
2611 trace_net_dev_start_xmit(skb, dev);
2612 rc = netdev_start_xmit(skb, dev, txq, more);
2613 trace_net_dev_xmit(skb, rc, dev, len);
2618 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
2619 struct netdev_queue *txq, int *ret)
2621 struct sk_buff *skb = first;
2622 int rc = NETDEV_TX_OK;
2625 struct sk_buff *next = skb->next;
2628 rc = xmit_one(skb, dev, txq, next != NULL);
2629 if (unlikely(!dev_xmit_complete(rc))) {
2635 if (netif_xmit_stopped(txq) && skb) {
2636 rc = NETDEV_TX_BUSY;
2646 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
2647 netdev_features_t features)
2649 if (vlan_tx_tag_present(skb) &&
2650 !vlan_hw_offload_capable(features, skb->vlan_proto))
2651 skb = __vlan_hwaccel_push_inside(skb);
2655 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
2657 netdev_features_t features;
2662 features = netif_skb_features(skb);
2663 skb = validate_xmit_vlan(skb, features);
2667 /* If encapsulation offload request, verify we are testing
2668 * hardware encapsulation features instead of standard
2669 * features for the netdev
2671 if (skb->encapsulation)
2672 features &= dev->hw_enc_features;
2674 if (netif_needs_gso(dev, skb, features)) {
2675 struct sk_buff *segs;
2677 segs = skb_gso_segment(skb, features);
2685 if (skb_needs_linearize(skb, features) &&
2686 __skb_linearize(skb))
2689 /* If packet is not checksummed and device does not
2690 * support checksumming for this protocol, complete
2691 * checksumming here.
2693 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2694 if (skb->encapsulation)
2695 skb_set_inner_transport_header(skb,
2696 skb_checksum_start_offset(skb));
2698 skb_set_transport_header(skb,
2699 skb_checksum_start_offset(skb));
2700 if (!(features & NETIF_F_ALL_CSUM) &&
2701 skb_checksum_help(skb))
2714 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev)
2716 struct sk_buff *next, *head = NULL, *tail;
2718 for (; skb != NULL; skb = next) {
2722 /* in case skb wont be segmented, point to itself */
2725 skb = validate_xmit_skb(skb, dev);
2733 /* If skb was segmented, skb->prev points to
2734 * the last segment. If not, it still contains skb.
2741 static void qdisc_pkt_len_init(struct sk_buff *skb)
2743 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2745 qdisc_skb_cb(skb)->pkt_len = skb->len;
2747 /* To get more precise estimation of bytes sent on wire,
2748 * we add to pkt_len the headers size of all segments
2750 if (shinfo->gso_size) {
2751 unsigned int hdr_len;
2752 u16 gso_segs = shinfo->gso_segs;
2754 /* mac layer + network layer */
2755 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2757 /* + transport layer */
2758 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2759 hdr_len += tcp_hdrlen(skb);
2761 hdr_len += sizeof(struct udphdr);
2763 if (shinfo->gso_type & SKB_GSO_DODGY)
2764 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2767 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2771 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2772 struct net_device *dev,
2773 struct netdev_queue *txq)
2775 spinlock_t *root_lock = qdisc_lock(q);
2779 qdisc_pkt_len_init(skb);
2780 qdisc_calculate_pkt_len(skb, q);
2782 * Heuristic to force contended enqueues to serialize on a
2783 * separate lock before trying to get qdisc main lock.
2784 * This permits __QDISC___STATE_RUNNING owner to get the lock more
2785 * often and dequeue packets faster.
2787 contended = qdisc_is_running(q);
2788 if (unlikely(contended))
2789 spin_lock(&q->busylock);
2791 spin_lock(root_lock);
2792 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2795 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2796 qdisc_run_begin(q)) {
2798 * This is a work-conserving queue; there are no old skbs
2799 * waiting to be sent out; and the qdisc is not running -
2800 * xmit the skb directly.
2803 qdisc_bstats_update(q, skb);
2805 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
2806 if (unlikely(contended)) {
2807 spin_unlock(&q->busylock);
2814 rc = NET_XMIT_SUCCESS;
2816 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2817 if (qdisc_run_begin(q)) {
2818 if (unlikely(contended)) {
2819 spin_unlock(&q->busylock);
2825 spin_unlock(root_lock);
2826 if (unlikely(contended))
2827 spin_unlock(&q->busylock);
2831 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2832 static void skb_update_prio(struct sk_buff *skb)
2834 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2836 if (!skb->priority && skb->sk && map) {
2837 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2839 if (prioidx < map->priomap_len)
2840 skb->priority = map->priomap[prioidx];
2844 #define skb_update_prio(skb)
2847 static DEFINE_PER_CPU(int, xmit_recursion);
2848 #define RECURSION_LIMIT 10
2851 * dev_loopback_xmit - loop back @skb
2852 * @skb: buffer to transmit
2854 int dev_loopback_xmit(struct sk_buff *skb)
2856 skb_reset_mac_header(skb);
2857 __skb_pull(skb, skb_network_offset(skb));
2858 skb->pkt_type = PACKET_LOOPBACK;
2859 skb->ip_summed = CHECKSUM_UNNECESSARY;
2860 WARN_ON(!skb_dst(skb));
2865 EXPORT_SYMBOL(dev_loopback_xmit);
2868 * __dev_queue_xmit - transmit a buffer
2869 * @skb: buffer to transmit
2870 * @accel_priv: private data used for L2 forwarding offload
2872 * Queue a buffer for transmission to a network device. The caller must
2873 * have set the device and priority and built the buffer before calling
2874 * this function. The function can be called from an interrupt.
2876 * A negative errno code is returned on a failure. A success does not
2877 * guarantee the frame will be transmitted as it may be dropped due
2878 * to congestion or traffic shaping.
2880 * -----------------------------------------------------------------------------------
2881 * I notice this method can also return errors from the queue disciplines,
2882 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2885 * Regardless of the return value, the skb is consumed, so it is currently
2886 * difficult to retry a send to this method. (You can bump the ref count
2887 * before sending to hold a reference for retry if you are careful.)
2889 * When calling this method, interrupts MUST be enabled. This is because
2890 * the BH enable code must have IRQs enabled so that it will not deadlock.
2893 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
2895 struct net_device *dev = skb->dev;
2896 struct netdev_queue *txq;
2900 skb_reset_mac_header(skb);
2902 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
2903 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
2905 /* Disable soft irqs for various locks below. Also
2906 * stops preemption for RCU.
2910 skb_update_prio(skb);
2912 /* If device/qdisc don't need skb->dst, release it right now while
2913 * its hot in this cpu cache.
2915 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2920 txq = netdev_pick_tx(dev, skb, accel_priv);
2921 q = rcu_dereference_bh(txq->qdisc);
2923 #ifdef CONFIG_NET_CLS_ACT
2924 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2926 trace_net_dev_queue(skb);
2928 rc = __dev_xmit_skb(skb, q, dev, txq);
2932 /* The device has no queue. Common case for software devices:
2933 loopback, all the sorts of tunnels...
2935 Really, it is unlikely that netif_tx_lock protection is necessary
2936 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2938 However, it is possible, that they rely on protection
2941 Check this and shot the lock. It is not prone from deadlocks.
2942 Either shot noqueue qdisc, it is even simpler 8)
2944 if (dev->flags & IFF_UP) {
2945 int cpu = smp_processor_id(); /* ok because BHs are off */
2947 if (txq->xmit_lock_owner != cpu) {
2949 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2950 goto recursion_alert;
2952 skb = validate_xmit_skb(skb, dev);
2956 HARD_TX_LOCK(dev, txq, cpu);
2958 if (!netif_xmit_stopped(txq)) {
2959 __this_cpu_inc(xmit_recursion);
2960 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
2961 __this_cpu_dec(xmit_recursion);
2962 if (dev_xmit_complete(rc)) {
2963 HARD_TX_UNLOCK(dev, txq);
2967 HARD_TX_UNLOCK(dev, txq);
2968 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2971 /* Recursion is detected! It is possible,
2975 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2982 rcu_read_unlock_bh();
2984 atomic_long_inc(&dev->tx_dropped);
2985 kfree_skb_list(skb);
2988 rcu_read_unlock_bh();
2992 int dev_queue_xmit(struct sk_buff *skb)
2994 return __dev_queue_xmit(skb, NULL);
2996 EXPORT_SYMBOL(dev_queue_xmit);
2998 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
3000 return __dev_queue_xmit(skb, accel_priv);
3002 EXPORT_SYMBOL(dev_queue_xmit_accel);
3005 /*=======================================================================
3007 =======================================================================*/
3009 int netdev_max_backlog __read_mostly = 1000;
3010 EXPORT_SYMBOL(netdev_max_backlog);
3012 int netdev_tstamp_prequeue __read_mostly = 1;
3013 int netdev_budget __read_mostly = 300;
3014 int weight_p __read_mostly = 64; /* old backlog weight */
3016 /* Called with irq disabled */
3017 static inline void ____napi_schedule(struct softnet_data *sd,
3018 struct napi_struct *napi)
3020 list_add_tail(&napi->poll_list, &sd->poll_list);
3021 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3026 /* One global table that all flow-based protocols share. */
3027 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3028 EXPORT_SYMBOL(rps_sock_flow_table);
3030 struct static_key rps_needed __read_mostly;
3032 static struct rps_dev_flow *
3033 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3034 struct rps_dev_flow *rflow, u16 next_cpu)
3036 if (next_cpu != RPS_NO_CPU) {
3037 #ifdef CONFIG_RFS_ACCEL
3038 struct netdev_rx_queue *rxqueue;
3039 struct rps_dev_flow_table *flow_table;
3040 struct rps_dev_flow *old_rflow;
3045 /* Should we steer this flow to a different hardware queue? */
3046 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3047 !(dev->features & NETIF_F_NTUPLE))
3049 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3050 if (rxq_index == skb_get_rx_queue(skb))
3053 rxqueue = dev->_rx + rxq_index;
3054 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3057 flow_id = skb_get_hash(skb) & flow_table->mask;
3058 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3059 rxq_index, flow_id);
3063 rflow = &flow_table->flows[flow_id];
3065 if (old_rflow->filter == rflow->filter)
3066 old_rflow->filter = RPS_NO_FILTER;
3070 per_cpu(softnet_data, next_cpu).input_queue_head;
3073 rflow->cpu = next_cpu;
3078 * get_rps_cpu is called from netif_receive_skb and returns the target
3079 * CPU from the RPS map of the receiving queue for a given skb.
3080 * rcu_read_lock must be held on entry.
3082 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3083 struct rps_dev_flow **rflowp)
3085 struct netdev_rx_queue *rxqueue;
3086 struct rps_map *map;
3087 struct rps_dev_flow_table *flow_table;
3088 struct rps_sock_flow_table *sock_flow_table;
3093 if (skb_rx_queue_recorded(skb)) {
3094 u16 index = skb_get_rx_queue(skb);
3095 if (unlikely(index >= dev->real_num_rx_queues)) {
3096 WARN_ONCE(dev->real_num_rx_queues > 1,
3097 "%s received packet on queue %u, but number "
3098 "of RX queues is %u\n",
3099 dev->name, index, dev->real_num_rx_queues);
3102 rxqueue = dev->_rx + index;
3106 map = rcu_dereference(rxqueue->rps_map);
3108 if (map->len == 1 &&
3109 !rcu_access_pointer(rxqueue->rps_flow_table)) {
3110 tcpu = map->cpus[0];
3111 if (cpu_online(tcpu))
3115 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
3119 skb_reset_network_header(skb);
3120 hash = skb_get_hash(skb);
3124 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3125 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3126 if (flow_table && sock_flow_table) {
3128 struct rps_dev_flow *rflow;
3130 rflow = &flow_table->flows[hash & flow_table->mask];
3133 next_cpu = sock_flow_table->ents[hash & sock_flow_table->mask];
3136 * If the desired CPU (where last recvmsg was done) is
3137 * different from current CPU (one in the rx-queue flow
3138 * table entry), switch if one of the following holds:
3139 * - Current CPU is unset (equal to RPS_NO_CPU).
3140 * - Current CPU is offline.
3141 * - The current CPU's queue tail has advanced beyond the
3142 * last packet that was enqueued using this table entry.
3143 * This guarantees that all previous packets for the flow
3144 * have been dequeued, thus preserving in order delivery.
3146 if (unlikely(tcpu != next_cpu) &&
3147 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
3148 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3149 rflow->last_qtail)) >= 0)) {
3151 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3154 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
3162 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3163 if (cpu_online(tcpu)) {
3173 #ifdef CONFIG_RFS_ACCEL
3176 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3177 * @dev: Device on which the filter was set
3178 * @rxq_index: RX queue index
3179 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3180 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3182 * Drivers that implement ndo_rx_flow_steer() should periodically call
3183 * this function for each installed filter and remove the filters for
3184 * which it returns %true.
3186 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3187 u32 flow_id, u16 filter_id)
3189 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3190 struct rps_dev_flow_table *flow_table;
3191 struct rps_dev_flow *rflow;
3196 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3197 if (flow_table && flow_id <= flow_table->mask) {
3198 rflow = &flow_table->flows[flow_id];
3199 cpu = ACCESS_ONCE(rflow->cpu);
3200 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3201 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3202 rflow->last_qtail) <
3203 (int)(10 * flow_table->mask)))
3209 EXPORT_SYMBOL(rps_may_expire_flow);
3211 #endif /* CONFIG_RFS_ACCEL */
3213 /* Called from hardirq (IPI) context */
3214 static void rps_trigger_softirq(void *data)
3216 struct softnet_data *sd = data;
3218 ____napi_schedule(sd, &sd->backlog);
3222 #endif /* CONFIG_RPS */
3225 * Check if this softnet_data structure is another cpu one
3226 * If yes, queue it to our IPI list and return 1
3229 static int rps_ipi_queued(struct softnet_data *sd)
3232 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3235 sd->rps_ipi_next = mysd->rps_ipi_list;
3236 mysd->rps_ipi_list = sd;
3238 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3241 #endif /* CONFIG_RPS */
3245 #ifdef CONFIG_NET_FLOW_LIMIT
3246 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3249 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3251 #ifdef CONFIG_NET_FLOW_LIMIT
3252 struct sd_flow_limit *fl;
3253 struct softnet_data *sd;
3254 unsigned int old_flow, new_flow;
3256 if (qlen < (netdev_max_backlog >> 1))
3259 sd = this_cpu_ptr(&softnet_data);
3262 fl = rcu_dereference(sd->flow_limit);
3264 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3265 old_flow = fl->history[fl->history_head];
3266 fl->history[fl->history_head] = new_flow;
3269 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3271 if (likely(fl->buckets[old_flow]))
3272 fl->buckets[old_flow]--;
3274 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3286 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3287 * queue (may be a remote CPU queue).
3289 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3290 unsigned int *qtail)
3292 struct softnet_data *sd;
3293 unsigned long flags;
3296 sd = &per_cpu(softnet_data, cpu);
3298 local_irq_save(flags);
3301 qlen = skb_queue_len(&sd->input_pkt_queue);
3302 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3305 __skb_queue_tail(&sd->input_pkt_queue, skb);
3306 input_queue_tail_incr_save(sd, qtail);
3308 local_irq_restore(flags);
3309 return NET_RX_SUCCESS;
3312 /* Schedule NAPI for backlog device
3313 * We can use non atomic operation since we own the queue lock
3315 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3316 if (!rps_ipi_queued(sd))
3317 ____napi_schedule(sd, &sd->backlog);
3325 local_irq_restore(flags);
3327 atomic_long_inc(&skb->dev->rx_dropped);
3332 static int netif_rx_internal(struct sk_buff *skb)
3336 net_timestamp_check(netdev_tstamp_prequeue, skb);
3338 trace_netif_rx(skb);
3340 if (static_key_false(&rps_needed)) {
3341 struct rps_dev_flow voidflow, *rflow = &voidflow;
3347 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3349 cpu = smp_processor_id();
3351 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3359 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3366 * netif_rx - post buffer to the network code
3367 * @skb: buffer to post
3369 * This function receives a packet from a device driver and queues it for
3370 * the upper (protocol) levels to process. It always succeeds. The buffer
3371 * may be dropped during processing for congestion control or by the
3375 * NET_RX_SUCCESS (no congestion)
3376 * NET_RX_DROP (packet was dropped)
3380 int netif_rx(struct sk_buff *skb)
3382 trace_netif_rx_entry(skb);
3384 return netif_rx_internal(skb);
3386 EXPORT_SYMBOL(netif_rx);
3388 int netif_rx_ni(struct sk_buff *skb)
3392 trace_netif_rx_ni_entry(skb);
3395 err = netif_rx_internal(skb);
3396 if (local_softirq_pending())
3402 EXPORT_SYMBOL(netif_rx_ni);
3404 static void net_tx_action(struct softirq_action *h)
3406 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3408 if (sd->completion_queue) {
3409 struct sk_buff *clist;
3411 local_irq_disable();
3412 clist = sd->completion_queue;
3413 sd->completion_queue = NULL;
3417 struct sk_buff *skb = clist;
3418 clist = clist->next;
3420 WARN_ON(atomic_read(&skb->users));
3421 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3422 trace_consume_skb(skb);
3424 trace_kfree_skb(skb, net_tx_action);
3429 if (sd->output_queue) {
3432 local_irq_disable();
3433 head = sd->output_queue;
3434 sd->output_queue = NULL;
3435 sd->output_queue_tailp = &sd->output_queue;
3439 struct Qdisc *q = head;
3440 spinlock_t *root_lock;
3442 head = head->next_sched;
3444 root_lock = qdisc_lock(q);
3445 if (spin_trylock(root_lock)) {
3446 smp_mb__before_atomic();
3447 clear_bit(__QDISC_STATE_SCHED,
3450 spin_unlock(root_lock);
3452 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3454 __netif_reschedule(q);
3456 smp_mb__before_atomic();
3457 clear_bit(__QDISC_STATE_SCHED,
3465 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3466 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3467 /* This hook is defined here for ATM LANE */
3468 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3469 unsigned char *addr) __read_mostly;
3470 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3473 #ifdef CONFIG_NET_CLS_ACT
3474 /* TODO: Maybe we should just force sch_ingress to be compiled in
3475 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3476 * a compare and 2 stores extra right now if we dont have it on
3477 * but have CONFIG_NET_CLS_ACT
3478 * NOTE: This doesn't stop any functionality; if you dont have
3479 * the ingress scheduler, you just can't add policies on ingress.
3482 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3484 struct net_device *dev = skb->dev;
3485 u32 ttl = G_TC_RTTL(skb->tc_verd);
3486 int result = TC_ACT_OK;
3489 if (unlikely(MAX_RED_LOOP < ttl++)) {
3490 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3491 skb->skb_iif, dev->ifindex);
3495 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3496 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3498 q = rcu_dereference(rxq->qdisc);
3499 if (q != &noop_qdisc) {
3500 spin_lock(qdisc_lock(q));
3501 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3502 result = qdisc_enqueue_root(skb, q);
3503 spin_unlock(qdisc_lock(q));
3509 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3510 struct packet_type **pt_prev,
3511 int *ret, struct net_device *orig_dev)
3513 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3515 if (!rxq || rcu_access_pointer(rxq->qdisc) == &noop_qdisc)
3519 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3523 switch (ing_filter(skb, rxq)) {
3537 * netdev_rx_handler_register - register receive handler
3538 * @dev: device to register a handler for
3539 * @rx_handler: receive handler to register
3540 * @rx_handler_data: data pointer that is used by rx handler
3542 * Register a receive handler for a device. This handler will then be
3543 * called from __netif_receive_skb. A negative errno code is returned
3546 * The caller must hold the rtnl_mutex.
3548 * For a general description of rx_handler, see enum rx_handler_result.
3550 int netdev_rx_handler_register(struct net_device *dev,
3551 rx_handler_func_t *rx_handler,
3552 void *rx_handler_data)
3556 if (dev->rx_handler)
3559 /* Note: rx_handler_data must be set before rx_handler */
3560 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3561 rcu_assign_pointer(dev->rx_handler, rx_handler);
3565 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3568 * netdev_rx_handler_unregister - unregister receive handler
3569 * @dev: device to unregister a handler from
3571 * Unregister a receive handler from a device.
3573 * The caller must hold the rtnl_mutex.
3575 void netdev_rx_handler_unregister(struct net_device *dev)
3579 RCU_INIT_POINTER(dev->rx_handler, NULL);
3580 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3581 * section has a guarantee to see a non NULL rx_handler_data
3585 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3587 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3590 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3591 * the special handling of PFMEMALLOC skbs.
3593 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3595 switch (skb->protocol) {
3596 case htons(ETH_P_ARP):
3597 case htons(ETH_P_IP):
3598 case htons(ETH_P_IPV6):
3599 case htons(ETH_P_8021Q):
3600 case htons(ETH_P_8021AD):
3607 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3609 struct packet_type *ptype, *pt_prev;
3610 rx_handler_func_t *rx_handler;
3611 struct net_device *orig_dev;
3612 struct net_device *null_or_dev;
3613 bool deliver_exact = false;
3614 int ret = NET_RX_DROP;
3617 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3619 trace_netif_receive_skb(skb);
3621 orig_dev = skb->dev;
3623 skb_reset_network_header(skb);
3624 if (!skb_transport_header_was_set(skb))
3625 skb_reset_transport_header(skb);
3626 skb_reset_mac_len(skb);
3633 skb->skb_iif = skb->dev->ifindex;
3635 __this_cpu_inc(softnet_data.processed);
3637 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3638 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3639 skb = skb_vlan_untag(skb);
3644 #ifdef CONFIG_NET_CLS_ACT
3645 if (skb->tc_verd & TC_NCLS) {
3646 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3654 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3655 if (!ptype->dev || ptype->dev == skb->dev) {
3657 ret = deliver_skb(skb, pt_prev, orig_dev);
3663 #ifdef CONFIG_NET_CLS_ACT
3664 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3670 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3673 if (vlan_tx_tag_present(skb)) {
3675 ret = deliver_skb(skb, pt_prev, orig_dev);
3678 if (vlan_do_receive(&skb))
3680 else if (unlikely(!skb))
3684 rx_handler = rcu_dereference(skb->dev->rx_handler);
3687 ret = deliver_skb(skb, pt_prev, orig_dev);
3690 switch (rx_handler(&skb)) {
3691 case RX_HANDLER_CONSUMED:
3692 ret = NET_RX_SUCCESS;
3694 case RX_HANDLER_ANOTHER:
3696 case RX_HANDLER_EXACT:
3697 deliver_exact = true;
3698 case RX_HANDLER_PASS:
3705 if (unlikely(vlan_tx_tag_present(skb))) {
3706 if (vlan_tx_tag_get_id(skb))
3707 skb->pkt_type = PACKET_OTHERHOST;
3708 /* Note: we might in the future use prio bits
3709 * and set skb->priority like in vlan_do_receive()
3710 * For the time being, just ignore Priority Code Point
3715 /* deliver only exact match when indicated */
3716 null_or_dev = deliver_exact ? skb->dev : NULL;
3718 type = skb->protocol;
3719 list_for_each_entry_rcu(ptype,
3720 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3721 if (ptype->type == type &&
3722 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3723 ptype->dev == orig_dev)) {
3725 ret = deliver_skb(skb, pt_prev, orig_dev);
3731 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3734 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3737 atomic_long_inc(&skb->dev->rx_dropped);
3739 /* Jamal, now you will not able to escape explaining
3740 * me how you were going to use this. :-)
3750 static int __netif_receive_skb(struct sk_buff *skb)
3754 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3755 unsigned long pflags = current->flags;
3758 * PFMEMALLOC skbs are special, they should
3759 * - be delivered to SOCK_MEMALLOC sockets only
3760 * - stay away from userspace
3761 * - have bounded memory usage
3763 * Use PF_MEMALLOC as this saves us from propagating the allocation
3764 * context down to all allocation sites.
3766 current->flags |= PF_MEMALLOC;
3767 ret = __netif_receive_skb_core(skb, true);
3768 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3770 ret = __netif_receive_skb_core(skb, false);
3775 static int netif_receive_skb_internal(struct sk_buff *skb)
3777 net_timestamp_check(netdev_tstamp_prequeue, skb);
3779 if (skb_defer_rx_timestamp(skb))
3780 return NET_RX_SUCCESS;
3783 if (static_key_false(&rps_needed)) {
3784 struct rps_dev_flow voidflow, *rflow = &voidflow;
3789 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3792 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3799 return __netif_receive_skb(skb);
3803 * netif_receive_skb - process receive buffer from network
3804 * @skb: buffer to process
3806 * netif_receive_skb() is the main receive data processing function.
3807 * It always succeeds. The buffer may be dropped during processing
3808 * for congestion control or by the protocol layers.
3810 * This function may only be called from softirq context and interrupts
3811 * should be enabled.
3813 * Return values (usually ignored):
3814 * NET_RX_SUCCESS: no congestion
3815 * NET_RX_DROP: packet was dropped
3817 int netif_receive_skb(struct sk_buff *skb)
3819 trace_netif_receive_skb_entry(skb);
3821 return netif_receive_skb_internal(skb);
3823 EXPORT_SYMBOL(netif_receive_skb);
3825 /* Network device is going away, flush any packets still pending
3826 * Called with irqs disabled.
3828 static void flush_backlog(void *arg)
3830 struct net_device *dev = arg;
3831 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3832 struct sk_buff *skb, *tmp;
3835 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3836 if (skb->dev == dev) {
3837 __skb_unlink(skb, &sd->input_pkt_queue);
3839 input_queue_head_incr(sd);
3844 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3845 if (skb->dev == dev) {
3846 __skb_unlink(skb, &sd->process_queue);
3848 input_queue_head_incr(sd);
3853 static int napi_gro_complete(struct sk_buff *skb)
3855 struct packet_offload *ptype;
3856 __be16 type = skb->protocol;
3857 struct list_head *head = &offload_base;
3860 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3862 if (NAPI_GRO_CB(skb)->count == 1) {
3863 skb_shinfo(skb)->gso_size = 0;
3868 list_for_each_entry_rcu(ptype, head, list) {
3869 if (ptype->type != type || !ptype->callbacks.gro_complete)
3872 err = ptype->callbacks.gro_complete(skb, 0);
3878 WARN_ON(&ptype->list == head);
3880 return NET_RX_SUCCESS;
3884 return netif_receive_skb_internal(skb);
3887 /* napi->gro_list contains packets ordered by age.
3888 * youngest packets at the head of it.
3889 * Complete skbs in reverse order to reduce latencies.
3891 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3893 struct sk_buff *skb, *prev = NULL;
3895 /* scan list and build reverse chain */
3896 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3901 for (skb = prev; skb; skb = prev) {
3904 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3908 napi_gro_complete(skb);
3912 napi->gro_list = NULL;
3914 EXPORT_SYMBOL(napi_gro_flush);
3916 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3919 unsigned int maclen = skb->dev->hard_header_len;
3920 u32 hash = skb_get_hash_raw(skb);
3922 for (p = napi->gro_list; p; p = p->next) {
3923 unsigned long diffs;
3925 NAPI_GRO_CB(p)->flush = 0;
3927 if (hash != skb_get_hash_raw(p)) {
3928 NAPI_GRO_CB(p)->same_flow = 0;
3932 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3933 diffs |= p->vlan_tci ^ skb->vlan_tci;
3934 if (maclen == ETH_HLEN)
3935 diffs |= compare_ether_header(skb_mac_header(p),
3936 skb_mac_header(skb));
3938 diffs = memcmp(skb_mac_header(p),
3939 skb_mac_header(skb),
3941 NAPI_GRO_CB(p)->same_flow = !diffs;
3945 static void skb_gro_reset_offset(struct sk_buff *skb)
3947 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3948 const skb_frag_t *frag0 = &pinfo->frags[0];
3950 NAPI_GRO_CB(skb)->data_offset = 0;
3951 NAPI_GRO_CB(skb)->frag0 = NULL;
3952 NAPI_GRO_CB(skb)->frag0_len = 0;
3954 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
3956 !PageHighMem(skb_frag_page(frag0))) {
3957 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3958 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3962 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
3964 struct skb_shared_info *pinfo = skb_shinfo(skb);
3966 BUG_ON(skb->end - skb->tail < grow);
3968 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3970 skb->data_len -= grow;
3973 pinfo->frags[0].page_offset += grow;
3974 skb_frag_size_sub(&pinfo->frags[0], grow);
3976 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
3977 skb_frag_unref(skb, 0);
3978 memmove(pinfo->frags, pinfo->frags + 1,
3979 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
3983 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3985 struct sk_buff **pp = NULL;
3986 struct packet_offload *ptype;
3987 __be16 type = skb->protocol;
3988 struct list_head *head = &offload_base;
3990 enum gro_result ret;
3993 if (!(skb->dev->features & NETIF_F_GRO))
3996 if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad)
3999 gro_list_prepare(napi, skb);
4002 list_for_each_entry_rcu(ptype, head, list) {
4003 if (ptype->type != type || !ptype->callbacks.gro_receive)
4006 skb_set_network_header(skb, skb_gro_offset(skb));
4007 skb_reset_mac_len(skb);
4008 NAPI_GRO_CB(skb)->same_flow = 0;
4009 NAPI_GRO_CB(skb)->flush = 0;
4010 NAPI_GRO_CB(skb)->free = 0;
4011 NAPI_GRO_CB(skb)->udp_mark = 0;
4013 /* Setup for GRO checksum validation */
4014 switch (skb->ip_summed) {
4015 case CHECKSUM_COMPLETE:
4016 NAPI_GRO_CB(skb)->csum = skb->csum;
4017 NAPI_GRO_CB(skb)->csum_valid = 1;
4018 NAPI_GRO_CB(skb)->csum_cnt = 0;
4020 case CHECKSUM_UNNECESSARY:
4021 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4022 NAPI_GRO_CB(skb)->csum_valid = 0;
4025 NAPI_GRO_CB(skb)->csum_cnt = 0;
4026 NAPI_GRO_CB(skb)->csum_valid = 0;
4029 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4034 if (&ptype->list == head)
4037 same_flow = NAPI_GRO_CB(skb)->same_flow;
4038 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4041 struct sk_buff *nskb = *pp;
4045 napi_gro_complete(nskb);
4052 if (NAPI_GRO_CB(skb)->flush)
4055 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4056 struct sk_buff *nskb = napi->gro_list;
4058 /* locate the end of the list to select the 'oldest' flow */
4059 while (nskb->next) {
4065 napi_gro_complete(nskb);
4069 NAPI_GRO_CB(skb)->count = 1;
4070 NAPI_GRO_CB(skb)->age = jiffies;
4071 NAPI_GRO_CB(skb)->last = skb;
4072 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4073 skb->next = napi->gro_list;
4074 napi->gro_list = skb;
4078 grow = skb_gro_offset(skb) - skb_headlen(skb);
4080 gro_pull_from_frag0(skb, grow);
4089 struct packet_offload *gro_find_receive_by_type(__be16 type)
4091 struct list_head *offload_head = &offload_base;
4092 struct packet_offload *ptype;
4094 list_for_each_entry_rcu(ptype, offload_head, list) {
4095 if (ptype->type != type || !ptype->callbacks.gro_receive)
4101 EXPORT_SYMBOL(gro_find_receive_by_type);
4103 struct packet_offload *gro_find_complete_by_type(__be16 type)
4105 struct list_head *offload_head = &offload_base;
4106 struct packet_offload *ptype;
4108 list_for_each_entry_rcu(ptype, offload_head, list) {
4109 if (ptype->type != type || !ptype->callbacks.gro_complete)
4115 EXPORT_SYMBOL(gro_find_complete_by_type);
4117 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4121 if (netif_receive_skb_internal(skb))
4129 case GRO_MERGED_FREE:
4130 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4131 kmem_cache_free(skbuff_head_cache, skb);
4144 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4146 trace_napi_gro_receive_entry(skb);
4148 skb_gro_reset_offset(skb);
4150 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4152 EXPORT_SYMBOL(napi_gro_receive);
4154 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4156 if (unlikely(skb->pfmemalloc)) {
4160 __skb_pull(skb, skb_headlen(skb));
4161 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4162 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4164 skb->dev = napi->dev;
4166 skb->encapsulation = 0;
4167 skb_shinfo(skb)->gso_type = 0;
4168 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4173 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4175 struct sk_buff *skb = napi->skb;
4178 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
4183 EXPORT_SYMBOL(napi_get_frags);
4185 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4186 struct sk_buff *skb,
4192 __skb_push(skb, ETH_HLEN);
4193 skb->protocol = eth_type_trans(skb, skb->dev);
4194 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4199 case GRO_MERGED_FREE:
4200 napi_reuse_skb(napi, skb);
4210 /* Upper GRO stack assumes network header starts at gro_offset=0
4211 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4212 * We copy ethernet header into skb->data to have a common layout.
4214 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4216 struct sk_buff *skb = napi->skb;
4217 const struct ethhdr *eth;
4218 unsigned int hlen = sizeof(*eth);
4222 skb_reset_mac_header(skb);
4223 skb_gro_reset_offset(skb);
4225 eth = skb_gro_header_fast(skb, 0);
4226 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4227 eth = skb_gro_header_slow(skb, hlen, 0);
4228 if (unlikely(!eth)) {
4229 napi_reuse_skb(napi, skb);
4233 gro_pull_from_frag0(skb, hlen);
4234 NAPI_GRO_CB(skb)->frag0 += hlen;
4235 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4237 __skb_pull(skb, hlen);
4240 * This works because the only protocols we care about don't require
4242 * We'll fix it up properly in napi_frags_finish()
4244 skb->protocol = eth->h_proto;
4249 gro_result_t napi_gro_frags(struct napi_struct *napi)
4251 struct sk_buff *skb = napi_frags_skb(napi);
4256 trace_napi_gro_frags_entry(skb);
4258 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4260 EXPORT_SYMBOL(napi_gro_frags);
4262 /* Compute the checksum from gro_offset and return the folded value
4263 * after adding in any pseudo checksum.
4265 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
4270 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
4272 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4273 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
4275 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
4276 !skb->csum_complete_sw)
4277 netdev_rx_csum_fault(skb->dev);
4280 NAPI_GRO_CB(skb)->csum = wsum;
4281 NAPI_GRO_CB(skb)->csum_valid = 1;
4285 EXPORT_SYMBOL(__skb_gro_checksum_complete);
4288 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4289 * Note: called with local irq disabled, but exits with local irq enabled.
4291 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4294 struct softnet_data *remsd = sd->rps_ipi_list;
4297 sd->rps_ipi_list = NULL;
4301 /* Send pending IPI's to kick RPS processing on remote cpus. */
4303 struct softnet_data *next = remsd->rps_ipi_next;
4305 if (cpu_online(remsd->cpu))
4306 smp_call_function_single_async(remsd->cpu,
4315 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
4318 return sd->rps_ipi_list != NULL;
4324 static int process_backlog(struct napi_struct *napi, int quota)
4327 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4329 /* Check if we have pending ipi, its better to send them now,
4330 * not waiting net_rx_action() end.
4332 if (sd_has_rps_ipi_waiting(sd)) {
4333 local_irq_disable();
4334 net_rps_action_and_irq_enable(sd);
4337 napi->weight = weight_p;
4338 local_irq_disable();
4340 struct sk_buff *skb;
4342 while ((skb = __skb_dequeue(&sd->process_queue))) {
4344 __netif_receive_skb(skb);
4345 local_irq_disable();
4346 input_queue_head_incr(sd);
4347 if (++work >= quota) {
4354 if (skb_queue_empty(&sd->input_pkt_queue)) {
4356 * Inline a custom version of __napi_complete().
4357 * only current cpu owns and manipulates this napi,
4358 * and NAPI_STATE_SCHED is the only possible flag set
4360 * We can use a plain write instead of clear_bit(),
4361 * and we dont need an smp_mb() memory barrier.
4369 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4370 &sd->process_queue);
4379 * __napi_schedule - schedule for receive
4380 * @n: entry to schedule
4382 * The entry's receive function will be scheduled to run.
4383 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
4385 void __napi_schedule(struct napi_struct *n)
4387 unsigned long flags;
4389 local_irq_save(flags);
4390 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4391 local_irq_restore(flags);
4393 EXPORT_SYMBOL(__napi_schedule);
4396 * __napi_schedule_irqoff - schedule for receive
4397 * @n: entry to schedule
4399 * Variant of __napi_schedule() assuming hard irqs are masked
4401 void __napi_schedule_irqoff(struct napi_struct *n)
4403 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4405 EXPORT_SYMBOL(__napi_schedule_irqoff);
4407 void __napi_complete(struct napi_struct *n)
4409 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4411 list_del_init(&n->poll_list);
4412 smp_mb__before_atomic();
4413 clear_bit(NAPI_STATE_SCHED, &n->state);
4415 EXPORT_SYMBOL(__napi_complete);
4417 void napi_complete_done(struct napi_struct *n, int work_done)
4419 unsigned long flags;
4422 * don't let napi dequeue from the cpu poll list
4423 * just in case its running on a different cpu
4425 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4429 unsigned long timeout = 0;
4432 timeout = n->dev->gro_flush_timeout;
4435 hrtimer_start(&n->timer, ns_to_ktime(timeout),
4436 HRTIMER_MODE_REL_PINNED);
4438 napi_gro_flush(n, false);
4440 if (likely(list_empty(&n->poll_list))) {
4441 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state));
4443 /* If n->poll_list is not empty, we need to mask irqs */
4444 local_irq_save(flags);
4446 local_irq_restore(flags);
4449 EXPORT_SYMBOL(napi_complete_done);
4451 /* must be called under rcu_read_lock(), as we dont take a reference */
4452 struct napi_struct *napi_by_id(unsigned int napi_id)
4454 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4455 struct napi_struct *napi;
4457 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4458 if (napi->napi_id == napi_id)
4463 EXPORT_SYMBOL_GPL(napi_by_id);
4465 void napi_hash_add(struct napi_struct *napi)
4467 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4469 spin_lock(&napi_hash_lock);
4471 /* 0 is not a valid id, we also skip an id that is taken
4472 * we expect both events to be extremely rare
4475 while (!napi->napi_id) {
4476 napi->napi_id = ++napi_gen_id;
4477 if (napi_by_id(napi->napi_id))
4481 hlist_add_head_rcu(&napi->napi_hash_node,
4482 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4484 spin_unlock(&napi_hash_lock);
4487 EXPORT_SYMBOL_GPL(napi_hash_add);
4489 /* Warning : caller is responsible to make sure rcu grace period
4490 * is respected before freeing memory containing @napi
4492 void napi_hash_del(struct napi_struct *napi)
4494 spin_lock(&napi_hash_lock);
4496 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4497 hlist_del_rcu(&napi->napi_hash_node);
4499 spin_unlock(&napi_hash_lock);
4501 EXPORT_SYMBOL_GPL(napi_hash_del);
4503 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
4505 struct napi_struct *napi;
4507 napi = container_of(timer, struct napi_struct, timer);
4509 napi_schedule(napi);
4511 return HRTIMER_NORESTART;
4514 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4515 int (*poll)(struct napi_struct *, int), int weight)
4517 INIT_LIST_HEAD(&napi->poll_list);
4518 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
4519 napi->timer.function = napi_watchdog;
4520 napi->gro_count = 0;
4521 napi->gro_list = NULL;
4524 if (weight > NAPI_POLL_WEIGHT)
4525 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4527 napi->weight = weight;
4528 list_add(&napi->dev_list, &dev->napi_list);
4530 #ifdef CONFIG_NETPOLL
4531 spin_lock_init(&napi->poll_lock);
4532 napi->poll_owner = -1;
4534 set_bit(NAPI_STATE_SCHED, &napi->state);
4536 EXPORT_SYMBOL(netif_napi_add);
4538 void napi_disable(struct napi_struct *n)
4541 set_bit(NAPI_STATE_DISABLE, &n->state);
4543 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
4546 hrtimer_cancel(&n->timer);
4548 clear_bit(NAPI_STATE_DISABLE, &n->state);
4550 EXPORT_SYMBOL(napi_disable);
4552 void netif_napi_del(struct napi_struct *napi)
4554 list_del_init(&napi->dev_list);
4555 napi_free_frags(napi);
4557 kfree_skb_list(napi->gro_list);
4558 napi->gro_list = NULL;
4559 napi->gro_count = 0;
4561 EXPORT_SYMBOL(netif_napi_del);
4563 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
4568 list_del_init(&n->poll_list);
4570 have = netpoll_poll_lock(n);
4574 /* This NAPI_STATE_SCHED test is for avoiding a race
4575 * with netpoll's poll_napi(). Only the entity which
4576 * obtains the lock and sees NAPI_STATE_SCHED set will
4577 * actually make the ->poll() call. Therefore we avoid
4578 * accidentally calling ->poll() when NAPI is not scheduled.
4581 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4582 work = n->poll(n, weight);
4586 WARN_ON_ONCE(work > weight);
4588 if (likely(work < weight))
4591 /* Drivers must not modify the NAPI state if they
4592 * consume the entire weight. In such cases this code
4593 * still "owns" the NAPI instance and therefore can
4594 * move the instance around on the list at-will.
4596 if (unlikely(napi_disable_pending(n))) {
4602 /* flush too old packets
4603 * If HZ < 1000, flush all packets.
4605 napi_gro_flush(n, HZ >= 1000);
4608 /* Some drivers may have called napi_schedule
4609 * prior to exhausting their budget.
4611 if (unlikely(!list_empty(&n->poll_list))) {
4612 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
4613 n->dev ? n->dev->name : "backlog");
4617 list_add_tail(&n->poll_list, repoll);
4620 netpoll_poll_unlock(have);
4625 static void net_rx_action(struct softirq_action *h)
4627 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4628 unsigned long time_limit = jiffies + 2;
4629 int budget = netdev_budget;
4633 local_irq_disable();
4634 list_splice_init(&sd->poll_list, &list);
4638 struct napi_struct *n;
4640 if (list_empty(&list)) {
4641 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
4646 n = list_first_entry(&list, struct napi_struct, poll_list);
4647 budget -= napi_poll(n, &repoll);
4649 /* If softirq window is exhausted then punt.
4650 * Allow this to run for 2 jiffies since which will allow
4651 * an average latency of 1.5/HZ.
4653 if (unlikely(budget <= 0 ||
4654 time_after_eq(jiffies, time_limit))) {
4660 local_irq_disable();
4662 list_splice_tail_init(&sd->poll_list, &list);
4663 list_splice_tail(&repoll, &list);
4664 list_splice(&list, &sd->poll_list);
4665 if (!list_empty(&sd->poll_list))
4666 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4668 net_rps_action_and_irq_enable(sd);
4671 struct netdev_adjacent {
4672 struct net_device *dev;
4674 /* upper master flag, there can only be one master device per list */
4677 /* counter for the number of times this device was added to us */
4680 /* private field for the users */
4683 struct list_head list;
4684 struct rcu_head rcu;
4687 static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev,
4688 struct net_device *adj_dev,
4689 struct list_head *adj_list)
4691 struct netdev_adjacent *adj;
4693 list_for_each_entry(adj, adj_list, list) {
4694 if (adj->dev == adj_dev)
4701 * netdev_has_upper_dev - Check if device is linked to an upper device
4703 * @upper_dev: upper device to check
4705 * Find out if a device is linked to specified upper device and return true
4706 * in case it is. Note that this checks only immediate upper device,
4707 * not through a complete stack of devices. The caller must hold the RTNL lock.
4709 bool netdev_has_upper_dev(struct net_device *dev,
4710 struct net_device *upper_dev)
4714 return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper);
4716 EXPORT_SYMBOL(netdev_has_upper_dev);
4719 * netdev_has_any_upper_dev - Check if device is linked to some device
4722 * Find out if a device is linked to an upper device and return true in case
4723 * it is. The caller must hold the RTNL lock.
4725 static bool netdev_has_any_upper_dev(struct net_device *dev)
4729 return !list_empty(&dev->all_adj_list.upper);
4733 * netdev_master_upper_dev_get - Get master upper device
4736 * Find a master upper device and return pointer to it or NULL in case
4737 * it's not there. The caller must hold the RTNL lock.
4739 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4741 struct netdev_adjacent *upper;
4745 if (list_empty(&dev->adj_list.upper))
4748 upper = list_first_entry(&dev->adj_list.upper,
4749 struct netdev_adjacent, list);
4750 if (likely(upper->master))
4754 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4756 void *netdev_adjacent_get_private(struct list_head *adj_list)
4758 struct netdev_adjacent *adj;
4760 adj = list_entry(adj_list, struct netdev_adjacent, list);
4762 return adj->private;
4764 EXPORT_SYMBOL(netdev_adjacent_get_private);
4767 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
4769 * @iter: list_head ** of the current position
4771 * Gets the next device from the dev's upper list, starting from iter
4772 * position. The caller must hold RCU read lock.
4774 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
4775 struct list_head **iter)
4777 struct netdev_adjacent *upper;
4779 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4781 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4783 if (&upper->list == &dev->adj_list.upper)
4786 *iter = &upper->list;
4790 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
4793 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4795 * @iter: list_head ** of the current position
4797 * Gets the next device from the dev's upper list, starting from iter
4798 * position. The caller must hold RCU read lock.
4800 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
4801 struct list_head **iter)
4803 struct netdev_adjacent *upper;
4805 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4807 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4809 if (&upper->list == &dev->all_adj_list.upper)
4812 *iter = &upper->list;
4816 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
4819 * netdev_lower_get_next_private - Get the next ->private from the
4820 * lower neighbour list
4822 * @iter: list_head ** of the current position
4824 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4825 * list, starting from iter position. The caller must hold either hold the
4826 * RTNL lock or its own locking that guarantees that the neighbour lower
4827 * list will remain unchainged.
4829 void *netdev_lower_get_next_private(struct net_device *dev,
4830 struct list_head **iter)
4832 struct netdev_adjacent *lower;
4834 lower = list_entry(*iter, struct netdev_adjacent, list);
4836 if (&lower->list == &dev->adj_list.lower)
4839 *iter = lower->list.next;
4841 return lower->private;
4843 EXPORT_SYMBOL(netdev_lower_get_next_private);
4846 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4847 * lower neighbour list, RCU
4850 * @iter: list_head ** of the current position
4852 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4853 * list, starting from iter position. The caller must hold RCU read lock.
4855 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
4856 struct list_head **iter)
4858 struct netdev_adjacent *lower;
4860 WARN_ON_ONCE(!rcu_read_lock_held());
4862 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4864 if (&lower->list == &dev->adj_list.lower)
4867 *iter = &lower->list;
4869 return lower->private;
4871 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
4874 * netdev_lower_get_next - Get the next device from the lower neighbour
4877 * @iter: list_head ** of the current position
4879 * Gets the next netdev_adjacent from the dev's lower neighbour
4880 * list, starting from iter position. The caller must hold RTNL lock or
4881 * its own locking that guarantees that the neighbour lower
4882 * list will remain unchainged.
4884 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
4886 struct netdev_adjacent *lower;
4888 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
4890 if (&lower->list == &dev->adj_list.lower)
4893 *iter = &lower->list;
4897 EXPORT_SYMBOL(netdev_lower_get_next);
4900 * netdev_lower_get_first_private_rcu - Get the first ->private from the
4901 * lower neighbour list, RCU
4905 * Gets the first netdev_adjacent->private from the dev's lower neighbour
4906 * list. The caller must hold RCU read lock.
4908 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
4910 struct netdev_adjacent *lower;
4912 lower = list_first_or_null_rcu(&dev->adj_list.lower,
4913 struct netdev_adjacent, list);
4915 return lower->private;
4918 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
4921 * netdev_master_upper_dev_get_rcu - Get master upper device
4924 * Find a master upper device and return pointer to it or NULL in case
4925 * it's not there. The caller must hold the RCU read lock.
4927 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4929 struct netdev_adjacent *upper;
4931 upper = list_first_or_null_rcu(&dev->adj_list.upper,
4932 struct netdev_adjacent, list);
4933 if (upper && likely(upper->master))
4937 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4939 static int netdev_adjacent_sysfs_add(struct net_device *dev,
4940 struct net_device *adj_dev,
4941 struct list_head *dev_list)
4943 char linkname[IFNAMSIZ+7];
4944 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4945 "upper_%s" : "lower_%s", adj_dev->name);
4946 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
4949 static void netdev_adjacent_sysfs_del(struct net_device *dev,
4951 struct list_head *dev_list)
4953 char linkname[IFNAMSIZ+7];
4954 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4955 "upper_%s" : "lower_%s", name);
4956 sysfs_remove_link(&(dev->dev.kobj), linkname);
4959 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
4960 struct net_device *adj_dev,
4961 struct list_head *dev_list)
4963 return (dev_list == &dev->adj_list.upper ||
4964 dev_list == &dev->adj_list.lower) &&
4965 net_eq(dev_net(dev), dev_net(adj_dev));
4968 static int __netdev_adjacent_dev_insert(struct net_device *dev,
4969 struct net_device *adj_dev,
4970 struct list_head *dev_list,
4971 void *private, bool master)
4973 struct netdev_adjacent *adj;
4976 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4983 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
4988 adj->master = master;
4990 adj->private = private;
4993 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
4994 adj_dev->name, dev->name, adj_dev->name);
4996 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
4997 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
5002 /* Ensure that master link is always the first item in list. */
5004 ret = sysfs_create_link(&(dev->dev.kobj),
5005 &(adj_dev->dev.kobj), "master");
5007 goto remove_symlinks;
5009 list_add_rcu(&adj->list, dev_list);
5011 list_add_tail_rcu(&adj->list, dev_list);
5017 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5018 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5026 static void __netdev_adjacent_dev_remove(struct net_device *dev,
5027 struct net_device *adj_dev,
5028 struct list_head *dev_list)
5030 struct netdev_adjacent *adj;
5032 adj = __netdev_find_adj(dev, adj_dev, dev_list);
5035 pr_err("tried to remove device %s from %s\n",
5036 dev->name, adj_dev->name);
5040 if (adj->ref_nr > 1) {
5041 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
5048 sysfs_remove_link(&(dev->dev.kobj), "master");
5050 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5051 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5053 list_del_rcu(&adj->list);
5054 pr_debug("dev_put for %s, because link removed from %s to %s\n",
5055 adj_dev->name, dev->name, adj_dev->name);
5057 kfree_rcu(adj, rcu);
5060 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
5061 struct net_device *upper_dev,
5062 struct list_head *up_list,
5063 struct list_head *down_list,
5064 void *private, bool master)
5068 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
5073 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
5076 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5083 static int __netdev_adjacent_dev_link(struct net_device *dev,
5084 struct net_device *upper_dev)
5086 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
5087 &dev->all_adj_list.upper,
5088 &upper_dev->all_adj_list.lower,
5092 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
5093 struct net_device *upper_dev,
5094 struct list_head *up_list,
5095 struct list_head *down_list)
5097 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5098 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
5101 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
5102 struct net_device *upper_dev)
5104 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5105 &dev->all_adj_list.upper,
5106 &upper_dev->all_adj_list.lower);
5109 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
5110 struct net_device *upper_dev,
5111 void *private, bool master)
5113 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
5118 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
5119 &dev->adj_list.upper,
5120 &upper_dev->adj_list.lower,
5123 __netdev_adjacent_dev_unlink(dev, upper_dev);
5130 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
5131 struct net_device *upper_dev)
5133 __netdev_adjacent_dev_unlink(dev, upper_dev);
5134 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5135 &dev->adj_list.upper,
5136 &upper_dev->adj_list.lower);
5139 static int __netdev_upper_dev_link(struct net_device *dev,
5140 struct net_device *upper_dev, bool master,
5143 struct netdev_adjacent *i, *j, *to_i, *to_j;
5148 if (dev == upper_dev)
5151 /* To prevent loops, check if dev is not upper device to upper_dev. */
5152 if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper))
5155 if (__netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper))
5158 if (master && netdev_master_upper_dev_get(dev))
5161 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
5166 /* Now that we linked these devs, make all the upper_dev's
5167 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5168 * versa, and don't forget the devices itself. All of these
5169 * links are non-neighbours.
5171 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5172 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5173 pr_debug("Interlinking %s with %s, non-neighbour\n",
5174 i->dev->name, j->dev->name);
5175 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
5181 /* add dev to every upper_dev's upper device */
5182 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5183 pr_debug("linking %s's upper device %s with %s\n",
5184 upper_dev->name, i->dev->name, dev->name);
5185 ret = __netdev_adjacent_dev_link(dev, i->dev);
5187 goto rollback_upper_mesh;
5190 /* add upper_dev to every dev's lower device */
5191 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5192 pr_debug("linking %s's lower device %s with %s\n", dev->name,
5193 i->dev->name, upper_dev->name);
5194 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
5196 goto rollback_lower_mesh;
5199 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5202 rollback_lower_mesh:
5204 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5207 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5212 rollback_upper_mesh:
5214 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5217 __netdev_adjacent_dev_unlink(dev, i->dev);
5225 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5226 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5227 if (i == to_i && j == to_j)
5229 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5235 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5241 * netdev_upper_dev_link - Add a link to the upper device
5243 * @upper_dev: new upper device
5245 * Adds a link to device which is upper to this one. The caller must hold
5246 * the RTNL lock. On a failure a negative errno code is returned.
5247 * On success the reference counts are adjusted and the function
5250 int netdev_upper_dev_link(struct net_device *dev,
5251 struct net_device *upper_dev)
5253 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
5255 EXPORT_SYMBOL(netdev_upper_dev_link);
5258 * netdev_master_upper_dev_link - Add a master link to the upper device
5260 * @upper_dev: new upper device
5262 * Adds a link to device which is upper to this one. In this case, only
5263 * one master upper device can be linked, although other non-master devices
5264 * might be linked as well. The caller must hold the RTNL lock.
5265 * On a failure a negative errno code is returned. On success the reference
5266 * counts are adjusted and the function returns zero.
5268 int netdev_master_upper_dev_link(struct net_device *dev,
5269 struct net_device *upper_dev)
5271 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
5273 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5275 int netdev_master_upper_dev_link_private(struct net_device *dev,
5276 struct net_device *upper_dev,
5279 return __netdev_upper_dev_link(dev, upper_dev, true, private);
5281 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
5284 * netdev_upper_dev_unlink - Removes a link to upper device
5286 * @upper_dev: new upper device
5288 * Removes a link to device which is upper to this one. The caller must hold
5291 void netdev_upper_dev_unlink(struct net_device *dev,
5292 struct net_device *upper_dev)
5294 struct netdev_adjacent *i, *j;
5297 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5299 /* Here is the tricky part. We must remove all dev's lower
5300 * devices from all upper_dev's upper devices and vice
5301 * versa, to maintain the graph relationship.
5303 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5304 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
5305 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5307 /* remove also the devices itself from lower/upper device
5310 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5311 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5313 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5314 __netdev_adjacent_dev_unlink(dev, i->dev);
5316 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5318 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5320 void netdev_adjacent_add_links(struct net_device *dev)
5322 struct netdev_adjacent *iter;
5324 struct net *net = dev_net(dev);
5326 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5327 if (!net_eq(net,dev_net(iter->dev)))
5329 netdev_adjacent_sysfs_add(iter->dev, dev,
5330 &iter->dev->adj_list.lower);
5331 netdev_adjacent_sysfs_add(dev, iter->dev,
5332 &dev->adj_list.upper);
5335 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5336 if (!net_eq(net,dev_net(iter->dev)))
5338 netdev_adjacent_sysfs_add(iter->dev, dev,
5339 &iter->dev->adj_list.upper);
5340 netdev_adjacent_sysfs_add(dev, iter->dev,
5341 &dev->adj_list.lower);
5345 void netdev_adjacent_del_links(struct net_device *dev)
5347 struct netdev_adjacent *iter;
5349 struct net *net = dev_net(dev);
5351 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5352 if (!net_eq(net,dev_net(iter->dev)))
5354 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5355 &iter->dev->adj_list.lower);
5356 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5357 &dev->adj_list.upper);
5360 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5361 if (!net_eq(net,dev_net(iter->dev)))
5363 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5364 &iter->dev->adj_list.upper);
5365 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5366 &dev->adj_list.lower);
5370 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5372 struct netdev_adjacent *iter;
5374 struct net *net = dev_net(dev);
5376 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5377 if (!net_eq(net,dev_net(iter->dev)))
5379 netdev_adjacent_sysfs_del(iter->dev, oldname,
5380 &iter->dev->adj_list.lower);
5381 netdev_adjacent_sysfs_add(iter->dev, dev,
5382 &iter->dev->adj_list.lower);
5385 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5386 if (!net_eq(net,dev_net(iter->dev)))
5388 netdev_adjacent_sysfs_del(iter->dev, oldname,
5389 &iter->dev->adj_list.upper);
5390 netdev_adjacent_sysfs_add(iter->dev, dev,
5391 &iter->dev->adj_list.upper);
5395 void *netdev_lower_dev_get_private(struct net_device *dev,
5396 struct net_device *lower_dev)
5398 struct netdev_adjacent *lower;
5402 lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower);
5406 return lower->private;
5408 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5411 int dev_get_nest_level(struct net_device *dev,
5412 bool (*type_check)(struct net_device *dev))
5414 struct net_device *lower = NULL;
5415 struct list_head *iter;
5421 netdev_for_each_lower_dev(dev, lower, iter) {
5422 nest = dev_get_nest_level(lower, type_check);
5423 if (max_nest < nest)
5427 if (type_check(dev))
5432 EXPORT_SYMBOL(dev_get_nest_level);
5434 static void dev_change_rx_flags(struct net_device *dev, int flags)
5436 const struct net_device_ops *ops = dev->netdev_ops;
5438 if (ops->ndo_change_rx_flags)
5439 ops->ndo_change_rx_flags(dev, flags);
5442 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5444 unsigned int old_flags = dev->flags;
5450 dev->flags |= IFF_PROMISC;
5451 dev->promiscuity += inc;
5452 if (dev->promiscuity == 0) {
5455 * If inc causes overflow, untouch promisc and return error.
5458 dev->flags &= ~IFF_PROMISC;
5460 dev->promiscuity -= inc;
5461 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5466 if (dev->flags != old_flags) {
5467 pr_info("device %s %s promiscuous mode\n",
5469 dev->flags & IFF_PROMISC ? "entered" : "left");
5470 if (audit_enabled) {
5471 current_uid_gid(&uid, &gid);
5472 audit_log(current->audit_context, GFP_ATOMIC,
5473 AUDIT_ANOM_PROMISCUOUS,
5474 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5475 dev->name, (dev->flags & IFF_PROMISC),
5476 (old_flags & IFF_PROMISC),
5477 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5478 from_kuid(&init_user_ns, uid),
5479 from_kgid(&init_user_ns, gid),
5480 audit_get_sessionid(current));
5483 dev_change_rx_flags(dev, IFF_PROMISC);
5486 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5491 * dev_set_promiscuity - update promiscuity count on a device
5495 * Add or remove promiscuity from a device. While the count in the device
5496 * remains above zero the interface remains promiscuous. Once it hits zero
5497 * the device reverts back to normal filtering operation. A negative inc
5498 * value is used to drop promiscuity on the device.
5499 * Return 0 if successful or a negative errno code on error.
5501 int dev_set_promiscuity(struct net_device *dev, int inc)
5503 unsigned int old_flags = dev->flags;
5506 err = __dev_set_promiscuity(dev, inc, true);
5509 if (dev->flags != old_flags)
5510 dev_set_rx_mode(dev);
5513 EXPORT_SYMBOL(dev_set_promiscuity);
5515 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5517 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5521 dev->flags |= IFF_ALLMULTI;
5522 dev->allmulti += inc;
5523 if (dev->allmulti == 0) {
5526 * If inc causes overflow, untouch allmulti and return error.
5529 dev->flags &= ~IFF_ALLMULTI;
5531 dev->allmulti -= inc;
5532 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5537 if (dev->flags ^ old_flags) {
5538 dev_change_rx_flags(dev, IFF_ALLMULTI);
5539 dev_set_rx_mode(dev);
5541 __dev_notify_flags(dev, old_flags,
5542 dev->gflags ^ old_gflags);
5548 * dev_set_allmulti - update allmulti count on a device
5552 * Add or remove reception of all multicast frames to a device. While the
5553 * count in the device remains above zero the interface remains listening
5554 * to all interfaces. Once it hits zero the device reverts back to normal
5555 * filtering operation. A negative @inc value is used to drop the counter
5556 * when releasing a resource needing all multicasts.
5557 * Return 0 if successful or a negative errno code on error.
5560 int dev_set_allmulti(struct net_device *dev, int inc)
5562 return __dev_set_allmulti(dev, inc, true);
5564 EXPORT_SYMBOL(dev_set_allmulti);
5567 * Upload unicast and multicast address lists to device and
5568 * configure RX filtering. When the device doesn't support unicast
5569 * filtering it is put in promiscuous mode while unicast addresses
5572 void __dev_set_rx_mode(struct net_device *dev)
5574 const struct net_device_ops *ops = dev->netdev_ops;
5576 /* dev_open will call this function so the list will stay sane. */
5577 if (!(dev->flags&IFF_UP))
5580 if (!netif_device_present(dev))
5583 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5584 /* Unicast addresses changes may only happen under the rtnl,
5585 * therefore calling __dev_set_promiscuity here is safe.
5587 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5588 __dev_set_promiscuity(dev, 1, false);
5589 dev->uc_promisc = true;
5590 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5591 __dev_set_promiscuity(dev, -1, false);
5592 dev->uc_promisc = false;
5596 if (ops->ndo_set_rx_mode)
5597 ops->ndo_set_rx_mode(dev);
5600 void dev_set_rx_mode(struct net_device *dev)
5602 netif_addr_lock_bh(dev);
5603 __dev_set_rx_mode(dev);
5604 netif_addr_unlock_bh(dev);
5608 * dev_get_flags - get flags reported to userspace
5611 * Get the combination of flag bits exported through APIs to userspace.
5613 unsigned int dev_get_flags(const struct net_device *dev)
5617 flags = (dev->flags & ~(IFF_PROMISC |
5622 (dev->gflags & (IFF_PROMISC |
5625 if (netif_running(dev)) {
5626 if (netif_oper_up(dev))
5627 flags |= IFF_RUNNING;
5628 if (netif_carrier_ok(dev))
5629 flags |= IFF_LOWER_UP;
5630 if (netif_dormant(dev))
5631 flags |= IFF_DORMANT;
5636 EXPORT_SYMBOL(dev_get_flags);
5638 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5640 unsigned int old_flags = dev->flags;
5646 * Set the flags on our device.
5649 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5650 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5652 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5656 * Load in the correct multicast list now the flags have changed.
5659 if ((old_flags ^ flags) & IFF_MULTICAST)
5660 dev_change_rx_flags(dev, IFF_MULTICAST);
5662 dev_set_rx_mode(dev);
5665 * Have we downed the interface. We handle IFF_UP ourselves
5666 * according to user attempts to set it, rather than blindly
5671 if ((old_flags ^ flags) & IFF_UP)
5672 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5674 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5675 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5676 unsigned int old_flags = dev->flags;
5678 dev->gflags ^= IFF_PROMISC;
5680 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5681 if (dev->flags != old_flags)
5682 dev_set_rx_mode(dev);
5685 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5686 is important. Some (broken) drivers set IFF_PROMISC, when
5687 IFF_ALLMULTI is requested not asking us and not reporting.
5689 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5690 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5692 dev->gflags ^= IFF_ALLMULTI;
5693 __dev_set_allmulti(dev, inc, false);
5699 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5700 unsigned int gchanges)
5702 unsigned int changes = dev->flags ^ old_flags;
5705 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
5707 if (changes & IFF_UP) {
5708 if (dev->flags & IFF_UP)
5709 call_netdevice_notifiers(NETDEV_UP, dev);
5711 call_netdevice_notifiers(NETDEV_DOWN, dev);
5714 if (dev->flags & IFF_UP &&
5715 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5716 struct netdev_notifier_change_info change_info;
5718 change_info.flags_changed = changes;
5719 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
5725 * dev_change_flags - change device settings
5727 * @flags: device state flags
5729 * Change settings on device based state flags. The flags are
5730 * in the userspace exported format.
5732 int dev_change_flags(struct net_device *dev, unsigned int flags)
5735 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
5737 ret = __dev_change_flags(dev, flags);
5741 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
5742 __dev_notify_flags(dev, old_flags, changes);
5745 EXPORT_SYMBOL(dev_change_flags);
5747 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
5749 const struct net_device_ops *ops = dev->netdev_ops;
5751 if (ops->ndo_change_mtu)
5752 return ops->ndo_change_mtu(dev, new_mtu);
5759 * dev_set_mtu - Change maximum transfer unit
5761 * @new_mtu: new transfer unit
5763 * Change the maximum transfer size of the network device.
5765 int dev_set_mtu(struct net_device *dev, int new_mtu)
5769 if (new_mtu == dev->mtu)
5772 /* MTU must be positive. */
5776 if (!netif_device_present(dev))
5779 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
5780 err = notifier_to_errno(err);
5784 orig_mtu = dev->mtu;
5785 err = __dev_set_mtu(dev, new_mtu);
5788 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5789 err = notifier_to_errno(err);
5791 /* setting mtu back and notifying everyone again,
5792 * so that they have a chance to revert changes.
5794 __dev_set_mtu(dev, orig_mtu);
5795 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5800 EXPORT_SYMBOL(dev_set_mtu);
5803 * dev_set_group - Change group this device belongs to
5805 * @new_group: group this device should belong to
5807 void dev_set_group(struct net_device *dev, int new_group)
5809 dev->group = new_group;
5811 EXPORT_SYMBOL(dev_set_group);
5814 * dev_set_mac_address - Change Media Access Control Address
5818 * Change the hardware (MAC) address of the device
5820 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
5822 const struct net_device_ops *ops = dev->netdev_ops;
5825 if (!ops->ndo_set_mac_address)
5827 if (sa->sa_family != dev->type)
5829 if (!netif_device_present(dev))
5831 err = ops->ndo_set_mac_address(dev, sa);
5834 dev->addr_assign_type = NET_ADDR_SET;
5835 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5836 add_device_randomness(dev->dev_addr, dev->addr_len);
5839 EXPORT_SYMBOL(dev_set_mac_address);
5842 * dev_change_carrier - Change device carrier
5844 * @new_carrier: new value
5846 * Change device carrier
5848 int dev_change_carrier(struct net_device *dev, bool new_carrier)
5850 const struct net_device_ops *ops = dev->netdev_ops;
5852 if (!ops->ndo_change_carrier)
5854 if (!netif_device_present(dev))
5856 return ops->ndo_change_carrier(dev, new_carrier);
5858 EXPORT_SYMBOL(dev_change_carrier);
5861 * dev_get_phys_port_id - Get device physical port ID
5865 * Get device physical port ID
5867 int dev_get_phys_port_id(struct net_device *dev,
5868 struct netdev_phys_item_id *ppid)
5870 const struct net_device_ops *ops = dev->netdev_ops;
5872 if (!ops->ndo_get_phys_port_id)
5874 return ops->ndo_get_phys_port_id(dev, ppid);
5876 EXPORT_SYMBOL(dev_get_phys_port_id);
5879 * dev_new_index - allocate an ifindex
5880 * @net: the applicable net namespace
5882 * Returns a suitable unique value for a new device interface
5883 * number. The caller must hold the rtnl semaphore or the
5884 * dev_base_lock to be sure it remains unique.
5886 static int dev_new_index(struct net *net)
5888 int ifindex = net->ifindex;
5892 if (!__dev_get_by_index(net, ifindex))
5893 return net->ifindex = ifindex;
5897 /* Delayed registration/unregisteration */
5898 static LIST_HEAD(net_todo_list);
5899 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
5901 static void net_set_todo(struct net_device *dev)
5903 list_add_tail(&dev->todo_list, &net_todo_list);
5904 dev_net(dev)->dev_unreg_count++;
5907 static void rollback_registered_many(struct list_head *head)
5909 struct net_device *dev, *tmp;
5910 LIST_HEAD(close_head);
5912 BUG_ON(dev_boot_phase);
5915 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5916 /* Some devices call without registering
5917 * for initialization unwind. Remove those
5918 * devices and proceed with the remaining.
5920 if (dev->reg_state == NETREG_UNINITIALIZED) {
5921 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5925 list_del(&dev->unreg_list);
5928 dev->dismantle = true;
5929 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5932 /* If device is running, close it first. */
5933 list_for_each_entry(dev, head, unreg_list)
5934 list_add_tail(&dev->close_list, &close_head);
5935 dev_close_many(&close_head);
5937 list_for_each_entry(dev, head, unreg_list) {
5938 /* And unlink it from device chain. */
5939 unlist_netdevice(dev);
5941 dev->reg_state = NETREG_UNREGISTERING;
5946 list_for_each_entry(dev, head, unreg_list) {
5947 struct sk_buff *skb = NULL;
5949 /* Shutdown queueing discipline. */
5953 /* Notify protocols, that we are about to destroy
5954 this device. They should clean all the things.
5956 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5958 if (!dev->rtnl_link_ops ||
5959 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5960 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U,
5964 * Flush the unicast and multicast chains
5969 if (dev->netdev_ops->ndo_uninit)
5970 dev->netdev_ops->ndo_uninit(dev);
5973 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
5975 /* Notifier chain MUST detach us all upper devices. */
5976 WARN_ON(netdev_has_any_upper_dev(dev));
5978 /* Remove entries from kobject tree */
5979 netdev_unregister_kobject(dev);
5981 /* Remove XPS queueing entries */
5982 netif_reset_xps_queues_gt(dev, 0);
5988 list_for_each_entry(dev, head, unreg_list)
5992 static void rollback_registered(struct net_device *dev)
5996 list_add(&dev->unreg_list, &single);
5997 rollback_registered_many(&single);
6001 static netdev_features_t netdev_fix_features(struct net_device *dev,
6002 netdev_features_t features)
6004 /* Fix illegal checksum combinations */
6005 if ((features & NETIF_F_HW_CSUM) &&
6006 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6007 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
6008 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
6011 /* TSO requires that SG is present as well. */
6012 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
6013 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
6014 features &= ~NETIF_F_ALL_TSO;
6017 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
6018 !(features & NETIF_F_IP_CSUM)) {
6019 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
6020 features &= ~NETIF_F_TSO;
6021 features &= ~NETIF_F_TSO_ECN;
6024 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
6025 !(features & NETIF_F_IPV6_CSUM)) {
6026 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
6027 features &= ~NETIF_F_TSO6;
6030 /* TSO ECN requires that TSO is present as well. */
6031 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
6032 features &= ~NETIF_F_TSO_ECN;
6034 /* Software GSO depends on SG. */
6035 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
6036 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
6037 features &= ~NETIF_F_GSO;
6040 /* UFO needs SG and checksumming */
6041 if (features & NETIF_F_UFO) {
6042 /* maybe split UFO into V4 and V6? */
6043 if (!((features & NETIF_F_GEN_CSUM) ||
6044 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
6045 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6047 "Dropping NETIF_F_UFO since no checksum offload features.\n");
6048 features &= ~NETIF_F_UFO;
6051 if (!(features & NETIF_F_SG)) {
6053 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
6054 features &= ~NETIF_F_UFO;
6058 #ifdef CONFIG_NET_RX_BUSY_POLL
6059 if (dev->netdev_ops->ndo_busy_poll)
6060 features |= NETIF_F_BUSY_POLL;
6063 features &= ~NETIF_F_BUSY_POLL;
6068 int __netdev_update_features(struct net_device *dev)
6070 netdev_features_t features;
6075 features = netdev_get_wanted_features(dev);
6077 if (dev->netdev_ops->ndo_fix_features)
6078 features = dev->netdev_ops->ndo_fix_features(dev, features);
6080 /* driver might be less strict about feature dependencies */
6081 features = netdev_fix_features(dev, features);
6083 if (dev->features == features)
6086 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
6087 &dev->features, &features);
6089 if (dev->netdev_ops->ndo_set_features)
6090 err = dev->netdev_ops->ndo_set_features(dev, features);
6092 if (unlikely(err < 0)) {
6094 "set_features() failed (%d); wanted %pNF, left %pNF\n",
6095 err, &features, &dev->features);
6100 dev->features = features;
6106 * netdev_update_features - recalculate device features
6107 * @dev: the device to check
6109 * Recalculate dev->features set and send notifications if it
6110 * has changed. Should be called after driver or hardware dependent
6111 * conditions might have changed that influence the features.
6113 void netdev_update_features(struct net_device *dev)
6115 if (__netdev_update_features(dev))
6116 netdev_features_change(dev);
6118 EXPORT_SYMBOL(netdev_update_features);
6121 * netdev_change_features - recalculate device features
6122 * @dev: the device to check
6124 * Recalculate dev->features set and send notifications even
6125 * if they have not changed. Should be called instead of
6126 * netdev_update_features() if also dev->vlan_features might
6127 * have changed to allow the changes to be propagated to stacked
6130 void netdev_change_features(struct net_device *dev)
6132 __netdev_update_features(dev);
6133 netdev_features_change(dev);
6135 EXPORT_SYMBOL(netdev_change_features);
6138 * netif_stacked_transfer_operstate - transfer operstate
6139 * @rootdev: the root or lower level device to transfer state from
6140 * @dev: the device to transfer operstate to
6142 * Transfer operational state from root to device. This is normally
6143 * called when a stacking relationship exists between the root
6144 * device and the device(a leaf device).
6146 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
6147 struct net_device *dev)
6149 if (rootdev->operstate == IF_OPER_DORMANT)
6150 netif_dormant_on(dev);
6152 netif_dormant_off(dev);
6154 if (netif_carrier_ok(rootdev)) {
6155 if (!netif_carrier_ok(dev))
6156 netif_carrier_on(dev);
6158 if (netif_carrier_ok(dev))
6159 netif_carrier_off(dev);
6162 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
6165 static int netif_alloc_rx_queues(struct net_device *dev)
6167 unsigned int i, count = dev->num_rx_queues;
6168 struct netdev_rx_queue *rx;
6172 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
6178 for (i = 0; i < count; i++)
6184 static void netdev_init_one_queue(struct net_device *dev,
6185 struct netdev_queue *queue, void *_unused)
6187 /* Initialize queue lock */
6188 spin_lock_init(&queue->_xmit_lock);
6189 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
6190 queue->xmit_lock_owner = -1;
6191 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
6194 dql_init(&queue->dql, HZ);
6198 static void netif_free_tx_queues(struct net_device *dev)
6203 static int netif_alloc_netdev_queues(struct net_device *dev)
6205 unsigned int count = dev->num_tx_queues;
6206 struct netdev_queue *tx;
6207 size_t sz = count * sizeof(*tx);
6209 BUG_ON(count < 1 || count > 0xffff);
6211 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6219 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
6220 spin_lock_init(&dev->tx_global_lock);
6226 * register_netdevice - register a network device
6227 * @dev: device to register
6229 * Take a completed network device structure and add it to the kernel
6230 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6231 * chain. 0 is returned on success. A negative errno code is returned
6232 * on a failure to set up the device, or if the name is a duplicate.
6234 * Callers must hold the rtnl semaphore. You may want
6235 * register_netdev() instead of this.
6238 * The locking appears insufficient to guarantee two parallel registers
6239 * will not get the same name.
6242 int register_netdevice(struct net_device *dev)
6245 struct net *net = dev_net(dev);
6247 BUG_ON(dev_boot_phase);
6252 /* When net_device's are persistent, this will be fatal. */
6253 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
6256 spin_lock_init(&dev->addr_list_lock);
6257 netdev_set_addr_lockdep_class(dev);
6261 ret = dev_get_valid_name(net, dev, dev->name);
6265 /* Init, if this function is available */
6266 if (dev->netdev_ops->ndo_init) {
6267 ret = dev->netdev_ops->ndo_init(dev);
6275 if (((dev->hw_features | dev->features) &
6276 NETIF_F_HW_VLAN_CTAG_FILTER) &&
6277 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
6278 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
6279 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
6286 dev->ifindex = dev_new_index(net);
6287 else if (__dev_get_by_index(net, dev->ifindex))
6290 if (dev->iflink == -1)
6291 dev->iflink = dev->ifindex;
6293 /* Transfer changeable features to wanted_features and enable
6294 * software offloads (GSO and GRO).
6296 dev->hw_features |= NETIF_F_SOFT_FEATURES;
6297 dev->features |= NETIF_F_SOFT_FEATURES;
6298 dev->wanted_features = dev->features & dev->hw_features;
6300 if (!(dev->flags & IFF_LOOPBACK)) {
6301 dev->hw_features |= NETIF_F_NOCACHE_COPY;
6304 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6306 dev->vlan_features |= NETIF_F_HIGHDMA;
6308 /* Make NETIF_F_SG inheritable to tunnel devices.
6310 dev->hw_enc_features |= NETIF_F_SG;
6312 /* Make NETIF_F_SG inheritable to MPLS.
6314 dev->mpls_features |= NETIF_F_SG;
6316 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
6317 ret = notifier_to_errno(ret);
6321 ret = netdev_register_kobject(dev);
6324 dev->reg_state = NETREG_REGISTERED;
6326 __netdev_update_features(dev);
6329 * Default initial state at registry is that the
6330 * device is present.
6333 set_bit(__LINK_STATE_PRESENT, &dev->state);
6335 linkwatch_init_dev(dev);
6337 dev_init_scheduler(dev);
6339 list_netdevice(dev);
6340 add_device_randomness(dev->dev_addr, dev->addr_len);
6342 /* If the device has permanent device address, driver should
6343 * set dev_addr and also addr_assign_type should be set to
6344 * NET_ADDR_PERM (default value).
6346 if (dev->addr_assign_type == NET_ADDR_PERM)
6347 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
6349 /* Notify protocols, that a new device appeared. */
6350 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
6351 ret = notifier_to_errno(ret);
6353 rollback_registered(dev);
6354 dev->reg_state = NETREG_UNREGISTERED;
6357 * Prevent userspace races by waiting until the network
6358 * device is fully setup before sending notifications.
6360 if (!dev->rtnl_link_ops ||
6361 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6362 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6368 if (dev->netdev_ops->ndo_uninit)
6369 dev->netdev_ops->ndo_uninit(dev);
6372 EXPORT_SYMBOL(register_netdevice);
6375 * init_dummy_netdev - init a dummy network device for NAPI
6376 * @dev: device to init
6378 * This takes a network device structure and initialize the minimum
6379 * amount of fields so it can be used to schedule NAPI polls without
6380 * registering a full blown interface. This is to be used by drivers
6381 * that need to tie several hardware interfaces to a single NAPI
6382 * poll scheduler due to HW limitations.
6384 int init_dummy_netdev(struct net_device *dev)
6386 /* Clear everything. Note we don't initialize spinlocks
6387 * are they aren't supposed to be taken by any of the
6388 * NAPI code and this dummy netdev is supposed to be
6389 * only ever used for NAPI polls
6391 memset(dev, 0, sizeof(struct net_device));
6393 /* make sure we BUG if trying to hit standard
6394 * register/unregister code path
6396 dev->reg_state = NETREG_DUMMY;
6398 /* NAPI wants this */
6399 INIT_LIST_HEAD(&dev->napi_list);
6401 /* a dummy interface is started by default */
6402 set_bit(__LINK_STATE_PRESENT, &dev->state);
6403 set_bit(__LINK_STATE_START, &dev->state);
6405 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6406 * because users of this 'device' dont need to change
6412 EXPORT_SYMBOL_GPL(init_dummy_netdev);
6416 * register_netdev - register a network device
6417 * @dev: device to register
6419 * Take a completed network device structure and add it to the kernel
6420 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6421 * chain. 0 is returned on success. A negative errno code is returned
6422 * on a failure to set up the device, or if the name is a duplicate.
6424 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6425 * and expands the device name if you passed a format string to
6428 int register_netdev(struct net_device *dev)
6433 err = register_netdevice(dev);
6437 EXPORT_SYMBOL(register_netdev);
6439 int netdev_refcnt_read(const struct net_device *dev)
6443 for_each_possible_cpu(i)
6444 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
6447 EXPORT_SYMBOL(netdev_refcnt_read);
6450 * netdev_wait_allrefs - wait until all references are gone.
6451 * @dev: target net_device
6453 * This is called when unregistering network devices.
6455 * Any protocol or device that holds a reference should register
6456 * for netdevice notification, and cleanup and put back the
6457 * reference if they receive an UNREGISTER event.
6458 * We can get stuck here if buggy protocols don't correctly
6461 static void netdev_wait_allrefs(struct net_device *dev)
6463 unsigned long rebroadcast_time, warning_time;
6466 linkwatch_forget_dev(dev);
6468 rebroadcast_time = warning_time = jiffies;
6469 refcnt = netdev_refcnt_read(dev);
6471 while (refcnt != 0) {
6472 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6475 /* Rebroadcast unregister notification */
6476 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6482 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6483 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6485 /* We must not have linkwatch events
6486 * pending on unregister. If this
6487 * happens, we simply run the queue
6488 * unscheduled, resulting in a noop
6491 linkwatch_run_queue();
6496 rebroadcast_time = jiffies;
6501 refcnt = netdev_refcnt_read(dev);
6503 if (time_after(jiffies, warning_time + 10 * HZ)) {
6504 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6506 warning_time = jiffies;
6515 * register_netdevice(x1);
6516 * register_netdevice(x2);
6518 * unregister_netdevice(y1);
6519 * unregister_netdevice(y2);
6525 * We are invoked by rtnl_unlock().
6526 * This allows us to deal with problems:
6527 * 1) We can delete sysfs objects which invoke hotplug
6528 * without deadlocking with linkwatch via keventd.
6529 * 2) Since we run with the RTNL semaphore not held, we can sleep
6530 * safely in order to wait for the netdev refcnt to drop to zero.
6532 * We must not return until all unregister events added during
6533 * the interval the lock was held have been completed.
6535 void netdev_run_todo(void)
6537 struct list_head list;
6539 /* Snapshot list, allow later requests */
6540 list_replace_init(&net_todo_list, &list);
6545 /* Wait for rcu callbacks to finish before next phase */
6546 if (!list_empty(&list))
6549 while (!list_empty(&list)) {
6550 struct net_device *dev
6551 = list_first_entry(&list, struct net_device, todo_list);
6552 list_del(&dev->todo_list);
6555 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6558 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6559 pr_err("network todo '%s' but state %d\n",
6560 dev->name, dev->reg_state);
6565 dev->reg_state = NETREG_UNREGISTERED;
6567 on_each_cpu(flush_backlog, dev, 1);
6569 netdev_wait_allrefs(dev);
6572 BUG_ON(netdev_refcnt_read(dev));
6573 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6574 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6575 WARN_ON(dev->dn_ptr);
6577 if (dev->destructor)
6578 dev->destructor(dev);
6580 /* Report a network device has been unregistered */
6582 dev_net(dev)->dev_unreg_count--;
6584 wake_up(&netdev_unregistering_wq);
6586 /* Free network device */
6587 kobject_put(&dev->dev.kobj);
6591 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6592 * fields in the same order, with only the type differing.
6594 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6595 const struct net_device_stats *netdev_stats)
6597 #if BITS_PER_LONG == 64
6598 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6599 memcpy(stats64, netdev_stats, sizeof(*stats64));
6601 size_t i, n = sizeof(*stats64) / sizeof(u64);
6602 const unsigned long *src = (const unsigned long *)netdev_stats;
6603 u64 *dst = (u64 *)stats64;
6605 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6606 sizeof(*stats64) / sizeof(u64));
6607 for (i = 0; i < n; i++)
6611 EXPORT_SYMBOL(netdev_stats_to_stats64);
6614 * dev_get_stats - get network device statistics
6615 * @dev: device to get statistics from
6616 * @storage: place to store stats
6618 * Get network statistics from device. Return @storage.
6619 * The device driver may provide its own method by setting
6620 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6621 * otherwise the internal statistics structure is used.
6623 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6624 struct rtnl_link_stats64 *storage)
6626 const struct net_device_ops *ops = dev->netdev_ops;
6628 if (ops->ndo_get_stats64) {
6629 memset(storage, 0, sizeof(*storage));
6630 ops->ndo_get_stats64(dev, storage);
6631 } else if (ops->ndo_get_stats) {
6632 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6634 netdev_stats_to_stats64(storage, &dev->stats);
6636 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6637 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
6640 EXPORT_SYMBOL(dev_get_stats);
6642 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
6644 struct netdev_queue *queue = dev_ingress_queue(dev);
6646 #ifdef CONFIG_NET_CLS_ACT
6649 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
6652 netdev_init_one_queue(dev, queue, NULL);
6653 queue->qdisc = &noop_qdisc;
6654 queue->qdisc_sleeping = &noop_qdisc;
6655 rcu_assign_pointer(dev->ingress_queue, queue);
6660 static const struct ethtool_ops default_ethtool_ops;
6662 void netdev_set_default_ethtool_ops(struct net_device *dev,
6663 const struct ethtool_ops *ops)
6665 if (dev->ethtool_ops == &default_ethtool_ops)
6666 dev->ethtool_ops = ops;
6668 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
6670 void netdev_freemem(struct net_device *dev)
6672 char *addr = (char *)dev - dev->padded;
6678 * alloc_netdev_mqs - allocate network device
6679 * @sizeof_priv: size of private data to allocate space for
6680 * @name: device name format string
6681 * @name_assign_type: origin of device name
6682 * @setup: callback to initialize device
6683 * @txqs: the number of TX subqueues to allocate
6684 * @rxqs: the number of RX subqueues to allocate
6686 * Allocates a struct net_device with private data area for driver use
6687 * and performs basic initialization. Also allocates subqueue structs
6688 * for each queue on the device.
6690 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
6691 unsigned char name_assign_type,
6692 void (*setup)(struct net_device *),
6693 unsigned int txqs, unsigned int rxqs)
6695 struct net_device *dev;
6697 struct net_device *p;
6699 BUG_ON(strlen(name) >= sizeof(dev->name));
6702 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6708 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6713 alloc_size = sizeof(struct net_device);
6715 /* ensure 32-byte alignment of private area */
6716 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6717 alloc_size += sizeof_priv;
6719 /* ensure 32-byte alignment of whole construct */
6720 alloc_size += NETDEV_ALIGN - 1;
6722 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6724 p = vzalloc(alloc_size);
6728 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6729 dev->padded = (char *)dev - (char *)p;
6731 dev->pcpu_refcnt = alloc_percpu(int);
6732 if (!dev->pcpu_refcnt)
6735 if (dev_addr_init(dev))
6741 dev_net_set(dev, &init_net);
6743 dev->gso_max_size = GSO_MAX_SIZE;
6744 dev->gso_max_segs = GSO_MAX_SEGS;
6745 dev->gso_min_segs = 0;
6747 INIT_LIST_HEAD(&dev->napi_list);
6748 INIT_LIST_HEAD(&dev->unreg_list);
6749 INIT_LIST_HEAD(&dev->close_list);
6750 INIT_LIST_HEAD(&dev->link_watch_list);
6751 INIT_LIST_HEAD(&dev->adj_list.upper);
6752 INIT_LIST_HEAD(&dev->adj_list.lower);
6753 INIT_LIST_HEAD(&dev->all_adj_list.upper);
6754 INIT_LIST_HEAD(&dev->all_adj_list.lower);
6755 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
6758 dev->num_tx_queues = txqs;
6759 dev->real_num_tx_queues = txqs;
6760 if (netif_alloc_netdev_queues(dev))
6764 dev->num_rx_queues = rxqs;
6765 dev->real_num_rx_queues = rxqs;
6766 if (netif_alloc_rx_queues(dev))
6770 strcpy(dev->name, name);
6771 dev->name_assign_type = name_assign_type;
6772 dev->group = INIT_NETDEV_GROUP;
6773 if (!dev->ethtool_ops)
6774 dev->ethtool_ops = &default_ethtool_ops;
6782 free_percpu(dev->pcpu_refcnt);
6784 netdev_freemem(dev);
6787 EXPORT_SYMBOL(alloc_netdev_mqs);
6790 * free_netdev - free network device
6793 * This function does the last stage of destroying an allocated device
6794 * interface. The reference to the device object is released.
6795 * If this is the last reference then it will be freed.
6797 void free_netdev(struct net_device *dev)
6799 struct napi_struct *p, *n;
6801 release_net(dev_net(dev));
6803 netif_free_tx_queues(dev);
6808 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6810 /* Flush device addresses */
6811 dev_addr_flush(dev);
6813 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6816 free_percpu(dev->pcpu_refcnt);
6817 dev->pcpu_refcnt = NULL;
6819 /* Compatibility with error handling in drivers */
6820 if (dev->reg_state == NETREG_UNINITIALIZED) {
6821 netdev_freemem(dev);
6825 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6826 dev->reg_state = NETREG_RELEASED;
6828 /* will free via device release */
6829 put_device(&dev->dev);
6831 EXPORT_SYMBOL(free_netdev);
6834 * synchronize_net - Synchronize with packet receive processing
6836 * Wait for packets currently being received to be done.
6837 * Does not block later packets from starting.
6839 void synchronize_net(void)
6842 if (rtnl_is_locked())
6843 synchronize_rcu_expedited();
6847 EXPORT_SYMBOL(synchronize_net);
6850 * unregister_netdevice_queue - remove device from the kernel
6854 * This function shuts down a device interface and removes it
6855 * from the kernel tables.
6856 * If head not NULL, device is queued to be unregistered later.
6858 * Callers must hold the rtnl semaphore. You may want
6859 * unregister_netdev() instead of this.
6862 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6867 list_move_tail(&dev->unreg_list, head);
6869 rollback_registered(dev);
6870 /* Finish processing unregister after unlock */
6874 EXPORT_SYMBOL(unregister_netdevice_queue);
6877 * unregister_netdevice_many - unregister many devices
6878 * @head: list of devices
6880 * Note: As most callers use a stack allocated list_head,
6881 * we force a list_del() to make sure stack wont be corrupted later.
6883 void unregister_netdevice_many(struct list_head *head)
6885 struct net_device *dev;
6887 if (!list_empty(head)) {
6888 rollback_registered_many(head);
6889 list_for_each_entry(dev, head, unreg_list)
6894 EXPORT_SYMBOL(unregister_netdevice_many);
6897 * unregister_netdev - remove device from the kernel
6900 * This function shuts down a device interface and removes it
6901 * from the kernel tables.
6903 * This is just a wrapper for unregister_netdevice that takes
6904 * the rtnl semaphore. In general you want to use this and not
6905 * unregister_netdevice.
6907 void unregister_netdev(struct net_device *dev)
6910 unregister_netdevice(dev);
6913 EXPORT_SYMBOL(unregister_netdev);
6916 * dev_change_net_namespace - move device to different nethost namespace
6918 * @net: network namespace
6919 * @pat: If not NULL name pattern to try if the current device name
6920 * is already taken in the destination network namespace.
6922 * This function shuts down a device interface and moves it
6923 * to a new network namespace. On success 0 is returned, on
6924 * a failure a netagive errno code is returned.
6926 * Callers must hold the rtnl semaphore.
6929 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6935 /* Don't allow namespace local devices to be moved. */
6937 if (dev->features & NETIF_F_NETNS_LOCAL)
6940 /* Ensure the device has been registrered */
6941 if (dev->reg_state != NETREG_REGISTERED)
6944 /* Get out if there is nothing todo */
6946 if (net_eq(dev_net(dev), net))
6949 /* Pick the destination device name, and ensure
6950 * we can use it in the destination network namespace.
6953 if (__dev_get_by_name(net, dev->name)) {
6954 /* We get here if we can't use the current device name */
6957 if (dev_get_valid_name(net, dev, pat) < 0)
6962 * And now a mini version of register_netdevice unregister_netdevice.
6965 /* If device is running close it first. */
6968 /* And unlink it from device chain */
6970 unlist_netdevice(dev);
6974 /* Shutdown queueing discipline. */
6977 /* Notify protocols, that we are about to destroy
6978 this device. They should clean all the things.
6980 Note that dev->reg_state stays at NETREG_REGISTERED.
6981 This is wanted because this way 8021q and macvlan know
6982 the device is just moving and can keep their slaves up.
6984 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6986 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6987 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
6990 * Flush the unicast and multicast chains
6995 /* Send a netdev-removed uevent to the old namespace */
6996 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
6997 netdev_adjacent_del_links(dev);
6999 /* Actually switch the network namespace */
7000 dev_net_set(dev, net);
7002 /* If there is an ifindex conflict assign a new one */
7003 if (__dev_get_by_index(net, dev->ifindex)) {
7004 int iflink = (dev->iflink == dev->ifindex);
7005 dev->ifindex = dev_new_index(net);
7007 dev->iflink = dev->ifindex;
7010 /* Send a netdev-add uevent to the new namespace */
7011 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
7012 netdev_adjacent_add_links(dev);
7014 /* Fixup kobjects */
7015 err = device_rename(&dev->dev, dev->name);
7018 /* Add the device back in the hashes */
7019 list_netdevice(dev);
7021 /* Notify protocols, that a new device appeared. */
7022 call_netdevice_notifiers(NETDEV_REGISTER, dev);
7025 * Prevent userspace races by waiting until the network
7026 * device is fully setup before sending notifications.
7028 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7035 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
7037 static int dev_cpu_callback(struct notifier_block *nfb,
7038 unsigned long action,
7041 struct sk_buff **list_skb;
7042 struct sk_buff *skb;
7043 unsigned int cpu, oldcpu = (unsigned long)ocpu;
7044 struct softnet_data *sd, *oldsd;
7046 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
7049 local_irq_disable();
7050 cpu = smp_processor_id();
7051 sd = &per_cpu(softnet_data, cpu);
7052 oldsd = &per_cpu(softnet_data, oldcpu);
7054 /* Find end of our completion_queue. */
7055 list_skb = &sd->completion_queue;
7057 list_skb = &(*list_skb)->next;
7058 /* Append completion queue from offline CPU. */
7059 *list_skb = oldsd->completion_queue;
7060 oldsd->completion_queue = NULL;
7062 /* Append output queue from offline CPU. */
7063 if (oldsd->output_queue) {
7064 *sd->output_queue_tailp = oldsd->output_queue;
7065 sd->output_queue_tailp = oldsd->output_queue_tailp;
7066 oldsd->output_queue = NULL;
7067 oldsd->output_queue_tailp = &oldsd->output_queue;
7069 /* Append NAPI poll list from offline CPU. */
7070 if (!list_empty(&oldsd->poll_list)) {
7071 list_splice_init(&oldsd->poll_list, &sd->poll_list);
7072 raise_softirq_irqoff(NET_RX_SOFTIRQ);
7075 raise_softirq_irqoff(NET_TX_SOFTIRQ);
7078 /* Process offline CPU's input_pkt_queue */
7079 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
7080 netif_rx_internal(skb);
7081 input_queue_head_incr(oldsd);
7083 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
7084 netif_rx_internal(skb);
7085 input_queue_head_incr(oldsd);
7093 * netdev_increment_features - increment feature set by one
7094 * @all: current feature set
7095 * @one: new feature set
7096 * @mask: mask feature set
7098 * Computes a new feature set after adding a device with feature set
7099 * @one to the master device with current feature set @all. Will not
7100 * enable anything that is off in @mask. Returns the new feature set.
7102 netdev_features_t netdev_increment_features(netdev_features_t all,
7103 netdev_features_t one, netdev_features_t mask)
7105 if (mask & NETIF_F_GEN_CSUM)
7106 mask |= NETIF_F_ALL_CSUM;
7107 mask |= NETIF_F_VLAN_CHALLENGED;
7109 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
7110 all &= one | ~NETIF_F_ALL_FOR_ALL;
7112 /* If one device supports hw checksumming, set for all. */
7113 if (all & NETIF_F_GEN_CSUM)
7114 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
7118 EXPORT_SYMBOL(netdev_increment_features);
7120 static struct hlist_head * __net_init netdev_create_hash(void)
7123 struct hlist_head *hash;
7125 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
7127 for (i = 0; i < NETDEV_HASHENTRIES; i++)
7128 INIT_HLIST_HEAD(&hash[i]);
7133 /* Initialize per network namespace state */
7134 static int __net_init netdev_init(struct net *net)
7136 if (net != &init_net)
7137 INIT_LIST_HEAD(&net->dev_base_head);
7139 net->dev_name_head = netdev_create_hash();
7140 if (net->dev_name_head == NULL)
7143 net->dev_index_head = netdev_create_hash();
7144 if (net->dev_index_head == NULL)
7150 kfree(net->dev_name_head);
7156 * netdev_drivername - network driver for the device
7157 * @dev: network device
7159 * Determine network driver for device.
7161 const char *netdev_drivername(const struct net_device *dev)
7163 const struct device_driver *driver;
7164 const struct device *parent;
7165 const char *empty = "";
7167 parent = dev->dev.parent;
7171 driver = parent->driver;
7172 if (driver && driver->name)
7173 return driver->name;
7177 static void __netdev_printk(const char *level, const struct net_device *dev,
7178 struct va_format *vaf)
7180 if (dev && dev->dev.parent) {
7181 dev_printk_emit(level[1] - '0',
7184 dev_driver_string(dev->dev.parent),
7185 dev_name(dev->dev.parent),
7186 netdev_name(dev), netdev_reg_state(dev),
7189 printk("%s%s%s: %pV",
7190 level, netdev_name(dev), netdev_reg_state(dev), vaf);
7192 printk("%s(NULL net_device): %pV", level, vaf);
7196 void netdev_printk(const char *level, const struct net_device *dev,
7197 const char *format, ...)
7199 struct va_format vaf;
7202 va_start(args, format);
7207 __netdev_printk(level, dev, &vaf);
7211 EXPORT_SYMBOL(netdev_printk);
7213 #define define_netdev_printk_level(func, level) \
7214 void func(const struct net_device *dev, const char *fmt, ...) \
7216 struct va_format vaf; \
7219 va_start(args, fmt); \
7224 __netdev_printk(level, dev, &vaf); \
7228 EXPORT_SYMBOL(func);
7230 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
7231 define_netdev_printk_level(netdev_alert, KERN_ALERT);
7232 define_netdev_printk_level(netdev_crit, KERN_CRIT);
7233 define_netdev_printk_level(netdev_err, KERN_ERR);
7234 define_netdev_printk_level(netdev_warn, KERN_WARNING);
7235 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
7236 define_netdev_printk_level(netdev_info, KERN_INFO);
7238 static void __net_exit netdev_exit(struct net *net)
7240 kfree(net->dev_name_head);
7241 kfree(net->dev_index_head);
7244 static struct pernet_operations __net_initdata netdev_net_ops = {
7245 .init = netdev_init,
7246 .exit = netdev_exit,
7249 static void __net_exit default_device_exit(struct net *net)
7251 struct net_device *dev, *aux;
7253 * Push all migratable network devices back to the
7254 * initial network namespace
7257 for_each_netdev_safe(net, dev, aux) {
7259 char fb_name[IFNAMSIZ];
7261 /* Ignore unmoveable devices (i.e. loopback) */
7262 if (dev->features & NETIF_F_NETNS_LOCAL)
7265 /* Leave virtual devices for the generic cleanup */
7266 if (dev->rtnl_link_ops)
7269 /* Push remaining network devices to init_net */
7270 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
7271 err = dev_change_net_namespace(dev, &init_net, fb_name);
7273 pr_emerg("%s: failed to move %s to init_net: %d\n",
7274 __func__, dev->name, err);
7281 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
7283 /* Return with the rtnl_lock held when there are no network
7284 * devices unregistering in any network namespace in net_list.
7288 DEFINE_WAIT_FUNC(wait, woken_wake_function);
7290 add_wait_queue(&netdev_unregistering_wq, &wait);
7292 unregistering = false;
7294 list_for_each_entry(net, net_list, exit_list) {
7295 if (net->dev_unreg_count > 0) {
7296 unregistering = true;
7304 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
7306 remove_wait_queue(&netdev_unregistering_wq, &wait);
7309 static void __net_exit default_device_exit_batch(struct list_head *net_list)
7311 /* At exit all network devices most be removed from a network
7312 * namespace. Do this in the reverse order of registration.
7313 * Do this across as many network namespaces as possible to
7314 * improve batching efficiency.
7316 struct net_device *dev;
7318 LIST_HEAD(dev_kill_list);
7320 /* To prevent network device cleanup code from dereferencing
7321 * loopback devices or network devices that have been freed
7322 * wait here for all pending unregistrations to complete,
7323 * before unregistring the loopback device and allowing the
7324 * network namespace be freed.
7326 * The netdev todo list containing all network devices
7327 * unregistrations that happen in default_device_exit_batch
7328 * will run in the rtnl_unlock() at the end of
7329 * default_device_exit_batch.
7331 rtnl_lock_unregistering(net_list);
7332 list_for_each_entry(net, net_list, exit_list) {
7333 for_each_netdev_reverse(net, dev) {
7334 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
7335 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
7337 unregister_netdevice_queue(dev, &dev_kill_list);
7340 unregister_netdevice_many(&dev_kill_list);
7344 static struct pernet_operations __net_initdata default_device_ops = {
7345 .exit = default_device_exit,
7346 .exit_batch = default_device_exit_batch,
7350 * Initialize the DEV module. At boot time this walks the device list and
7351 * unhooks any devices that fail to initialise (normally hardware not
7352 * present) and leaves us with a valid list of present and active devices.
7357 * This is called single threaded during boot, so no need
7358 * to take the rtnl semaphore.
7360 static int __init net_dev_init(void)
7362 int i, rc = -ENOMEM;
7364 BUG_ON(!dev_boot_phase);
7366 if (dev_proc_init())
7369 if (netdev_kobject_init())
7372 INIT_LIST_HEAD(&ptype_all);
7373 for (i = 0; i < PTYPE_HASH_SIZE; i++)
7374 INIT_LIST_HEAD(&ptype_base[i]);
7376 INIT_LIST_HEAD(&offload_base);
7378 if (register_pernet_subsys(&netdev_net_ops))
7382 * Initialise the packet receive queues.
7385 for_each_possible_cpu(i) {
7386 struct softnet_data *sd = &per_cpu(softnet_data, i);
7388 skb_queue_head_init(&sd->input_pkt_queue);
7389 skb_queue_head_init(&sd->process_queue);
7390 INIT_LIST_HEAD(&sd->poll_list);
7391 sd->output_queue_tailp = &sd->output_queue;
7393 sd->csd.func = rps_trigger_softirq;
7398 sd->backlog.poll = process_backlog;
7399 sd->backlog.weight = weight_p;
7404 /* The loopback device is special if any other network devices
7405 * is present in a network namespace the loopback device must
7406 * be present. Since we now dynamically allocate and free the
7407 * loopback device ensure this invariant is maintained by
7408 * keeping the loopback device as the first device on the
7409 * list of network devices. Ensuring the loopback devices
7410 * is the first device that appears and the last network device
7413 if (register_pernet_device(&loopback_net_ops))
7416 if (register_pernet_device(&default_device_ops))
7419 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
7420 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
7422 hotcpu_notifier(dev_cpu_callback, 0);
7429 subsys_initcall(net_dev_init);