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/proc_fs.h>
101 #include <linux/seq_file.h>
102 #include <linux/stat.h>
104 #include <net/pkt_sched.h>
105 #include <net/checksum.h>
106 #include <net/xfrm.h>
107 #include <linux/highmem.h>
108 #include <linux/init.h>
109 #include <linux/kmod.h>
110 #include <linux/module.h>
111 #include <linux/netpoll.h>
112 #include <linux/rcupdate.h>
113 #include <linux/delay.h>
114 #include <net/wext.h>
115 #include <net/iw_handler.h>
116 #include <asm/current.h>
117 #include <linux/audit.h>
118 #include <linux/dmaengine.h>
119 #include <linux/err.h>
120 #include <linux/ctype.h>
121 #include <linux/if_arp.h>
122 #include <linux/if_vlan.h>
123 #include <linux/ip.h>
125 #include <linux/ipv6.h>
126 #include <linux/in.h>
127 #include <linux/jhash.h>
128 #include <linux/random.h>
129 #include <trace/events/napi.h>
130 #include <trace/events/net.h>
131 #include <trace/events/skb.h>
132 #include <linux/pci.h>
133 #include <linux/inetdevice.h>
134 #include <linux/cpu_rmap.h>
135 #include <linux/net_tstamp.h>
136 #include <linux/static_key.h>
137 #include <net/flow_keys.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)
148 * The list of packet types we will receive (as opposed to discard)
149 * and the routines to invoke.
151 * Why 16. Because with 16 the only overlap we get on a hash of the
152 * low nibble of the protocol value is RARP/SNAP/X.25.
154 * NOTE: That is no longer true with the addition of VLAN tags. Not
155 * sure which should go first, but I bet it won't make much
156 * difference if we are running VLANs. The good news is that
157 * this protocol won't be in the list unless compiled in, so
158 * the average user (w/out VLANs) will not be adversely affected.
175 #define PTYPE_HASH_SIZE (16)
176 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
178 static DEFINE_SPINLOCK(ptype_lock);
179 static DEFINE_SPINLOCK(offload_lock);
180 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
181 static struct list_head ptype_all __read_mostly; /* Taps */
182 static struct list_head offload_base __read_mostly;
185 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
188 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
190 * Writers must hold the rtnl semaphore while they loop through the
191 * dev_base_head list, and hold dev_base_lock for writing when they do the
192 * actual updates. This allows pure readers to access the list even
193 * while a writer is preparing to update it.
195 * To put it another way, dev_base_lock is held for writing only to
196 * protect against pure readers; the rtnl semaphore provides the
197 * protection against other writers.
199 * See, for example usages, register_netdevice() and
200 * unregister_netdevice(), which must be called with the rtnl
203 DEFINE_RWLOCK(dev_base_lock);
204 EXPORT_SYMBOL(dev_base_lock);
206 static inline void dev_base_seq_inc(struct net *net)
208 while (++net->dev_base_seq == 0);
211 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
213 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
215 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
218 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
220 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
223 static inline void rps_lock(struct softnet_data *sd)
226 spin_lock(&sd->input_pkt_queue.lock);
230 static inline void rps_unlock(struct softnet_data *sd)
233 spin_unlock(&sd->input_pkt_queue.lock);
237 /* Device list insertion */
238 static int list_netdevice(struct net_device *dev)
240 struct net *net = dev_net(dev);
244 write_lock_bh(&dev_base_lock);
245 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
246 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
247 hlist_add_head_rcu(&dev->index_hlist,
248 dev_index_hash(net, dev->ifindex));
249 write_unlock_bh(&dev_base_lock);
251 dev_base_seq_inc(net);
256 /* Device list removal
257 * caller must respect a RCU grace period before freeing/reusing dev
259 static void unlist_netdevice(struct net_device *dev)
263 /* Unlink dev from the device chain */
264 write_lock_bh(&dev_base_lock);
265 list_del_rcu(&dev->dev_list);
266 hlist_del_rcu(&dev->name_hlist);
267 hlist_del_rcu(&dev->index_hlist);
268 write_unlock_bh(&dev_base_lock);
270 dev_base_seq_inc(dev_net(dev));
277 static RAW_NOTIFIER_HEAD(netdev_chain);
280 * Device drivers call our routines to queue packets here. We empty the
281 * queue in the local softnet handler.
284 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
285 EXPORT_PER_CPU_SYMBOL(softnet_data);
287 #ifdef CONFIG_LOCKDEP
289 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
290 * according to dev->type
292 static const unsigned short netdev_lock_type[] =
293 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
294 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
295 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
296 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
297 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
298 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
299 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
300 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
301 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
302 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
303 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
304 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
305 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
306 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
307 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
309 static const char *const netdev_lock_name[] =
310 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
311 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
312 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
313 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
314 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
315 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
316 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
317 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
318 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
319 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
320 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
321 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
322 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
323 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
324 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
326 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
327 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
329 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
333 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
334 if (netdev_lock_type[i] == dev_type)
336 /* the last key is used by default */
337 return ARRAY_SIZE(netdev_lock_type) - 1;
340 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
341 unsigned short dev_type)
345 i = netdev_lock_pos(dev_type);
346 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
347 netdev_lock_name[i]);
350 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
354 i = netdev_lock_pos(dev->type);
355 lockdep_set_class_and_name(&dev->addr_list_lock,
356 &netdev_addr_lock_key[i],
357 netdev_lock_name[i]);
360 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
361 unsigned short dev_type)
364 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
369 /*******************************************************************************
371 Protocol management and registration routines
373 *******************************************************************************/
376 * Add a protocol ID to the list. Now that the input handler is
377 * smarter we can dispense with all the messy stuff that used to be
380 * BEWARE!!! Protocol handlers, mangling input packets,
381 * MUST BE last in hash buckets and checking protocol handlers
382 * MUST start from promiscuous ptype_all chain in net_bh.
383 * It is true now, do not change it.
384 * Explanation follows: if protocol handler, mangling packet, will
385 * be the first on list, it is not able to sense, that packet
386 * is cloned and should be copied-on-write, so that it will
387 * change it and subsequent readers will get broken packet.
391 static inline struct list_head *ptype_head(const struct packet_type *pt)
393 if (pt->type == htons(ETH_P_ALL))
396 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
400 * dev_add_pack - add packet handler
401 * @pt: packet type declaration
403 * Add a protocol handler to the networking stack. The passed &packet_type
404 * is linked into kernel lists and may not be freed until it has been
405 * removed from the kernel lists.
407 * This call does not sleep therefore it can not
408 * guarantee all CPU's that are in middle of receiving packets
409 * will see the new packet type (until the next received packet).
412 void dev_add_pack(struct packet_type *pt)
414 struct list_head *head = ptype_head(pt);
416 spin_lock(&ptype_lock);
417 list_add_rcu(&pt->list, head);
418 spin_unlock(&ptype_lock);
420 EXPORT_SYMBOL(dev_add_pack);
423 * __dev_remove_pack - remove packet handler
424 * @pt: packet type declaration
426 * Remove a protocol handler that was previously added to the kernel
427 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
428 * from the kernel lists and can be freed or reused once this function
431 * The packet type might still be in use by receivers
432 * and must not be freed until after all the CPU's have gone
433 * through a quiescent state.
435 void __dev_remove_pack(struct packet_type *pt)
437 struct list_head *head = ptype_head(pt);
438 struct packet_type *pt1;
440 spin_lock(&ptype_lock);
442 list_for_each_entry(pt1, head, list) {
444 list_del_rcu(&pt->list);
449 pr_warn("dev_remove_pack: %p not found\n", pt);
451 spin_unlock(&ptype_lock);
453 EXPORT_SYMBOL(__dev_remove_pack);
456 * dev_remove_pack - remove packet handler
457 * @pt: packet type declaration
459 * Remove a protocol handler that was previously added to the kernel
460 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
461 * from the kernel lists and can be freed or reused once this function
464 * This call sleeps to guarantee that no CPU is looking at the packet
467 void dev_remove_pack(struct packet_type *pt)
469 __dev_remove_pack(pt);
473 EXPORT_SYMBOL(dev_remove_pack);
477 * dev_add_offload - register offload handlers
478 * @po: protocol offload declaration
480 * Add protocol offload handlers to the networking stack. The passed
481 * &proto_offload is linked into kernel lists and may not be freed until
482 * it has been removed from the kernel lists.
484 * This call does not sleep therefore it can not
485 * guarantee all CPU's that are in middle of receiving packets
486 * will see the new offload handlers (until the next received packet).
488 void dev_add_offload(struct packet_offload *po)
490 struct list_head *head = &offload_base;
492 spin_lock(&offload_lock);
493 list_add_rcu(&po->list, head);
494 spin_unlock(&offload_lock);
496 EXPORT_SYMBOL(dev_add_offload);
499 * __dev_remove_offload - remove offload handler
500 * @po: packet offload declaration
502 * Remove a protocol offload handler that was previously added to the
503 * kernel offload handlers by dev_add_offload(). The passed &offload_type
504 * is removed from the kernel lists and can be freed or reused once this
507 * The packet type might still be in use by receivers
508 * and must not be freed until after all the CPU's have gone
509 * through a quiescent state.
511 void __dev_remove_offload(struct packet_offload *po)
513 struct list_head *head = &offload_base;
514 struct packet_offload *po1;
516 spin_lock(&ptype_lock);
518 list_for_each_entry(po1, head, list) {
520 list_del_rcu(&po->list);
525 pr_warn("dev_remove_offload: %p not found\n", po);
527 spin_unlock(&ptype_lock);
529 EXPORT_SYMBOL(__dev_remove_offload);
532 * dev_remove_offload - remove packet offload handler
533 * @po: packet offload declaration
535 * Remove a packet offload handler that was previously added to the kernel
536 * offload handlers by dev_add_offload(). The passed &offload_type is
537 * removed from the kernel lists and can be freed or reused once this
540 * This call sleeps to guarantee that no CPU is looking at the packet
543 void dev_remove_offload(struct packet_offload *po)
545 __dev_remove_offload(po);
549 EXPORT_SYMBOL(dev_remove_offload);
551 /******************************************************************************
553 Device Boot-time Settings Routines
555 *******************************************************************************/
557 /* Boot time configuration table */
558 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
561 * netdev_boot_setup_add - add new setup entry
562 * @name: name of the device
563 * @map: configured settings for the device
565 * Adds new setup entry to the dev_boot_setup list. The function
566 * returns 0 on error and 1 on success. This is a generic routine to
569 static int netdev_boot_setup_add(char *name, struct ifmap *map)
571 struct netdev_boot_setup *s;
575 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
576 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
577 memset(s[i].name, 0, sizeof(s[i].name));
578 strlcpy(s[i].name, name, IFNAMSIZ);
579 memcpy(&s[i].map, map, sizeof(s[i].map));
584 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
588 * netdev_boot_setup_check - check boot time settings
589 * @dev: the netdevice
591 * Check boot time settings for the device.
592 * The found settings are set for the device to be used
593 * later in the device probing.
594 * Returns 0 if no settings found, 1 if they are.
596 int netdev_boot_setup_check(struct net_device *dev)
598 struct netdev_boot_setup *s = dev_boot_setup;
601 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
602 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
603 !strcmp(dev->name, s[i].name)) {
604 dev->irq = s[i].map.irq;
605 dev->base_addr = s[i].map.base_addr;
606 dev->mem_start = s[i].map.mem_start;
607 dev->mem_end = s[i].map.mem_end;
613 EXPORT_SYMBOL(netdev_boot_setup_check);
617 * netdev_boot_base - get address from boot time settings
618 * @prefix: prefix for network device
619 * @unit: id for network device
621 * Check boot time settings for the base address of device.
622 * The found settings are set for the device to be used
623 * later in the device probing.
624 * Returns 0 if no settings found.
626 unsigned long netdev_boot_base(const char *prefix, int unit)
628 const struct netdev_boot_setup *s = dev_boot_setup;
632 sprintf(name, "%s%d", prefix, unit);
635 * If device already registered then return base of 1
636 * to indicate not to probe for this interface
638 if (__dev_get_by_name(&init_net, name))
641 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
642 if (!strcmp(name, s[i].name))
643 return s[i].map.base_addr;
648 * Saves at boot time configured settings for any netdevice.
650 int __init netdev_boot_setup(char *str)
655 str = get_options(str, ARRAY_SIZE(ints), ints);
660 memset(&map, 0, sizeof(map));
664 map.base_addr = ints[2];
666 map.mem_start = ints[3];
668 map.mem_end = ints[4];
670 /* Add new entry to the list */
671 return netdev_boot_setup_add(str, &map);
674 __setup("netdev=", netdev_boot_setup);
676 /*******************************************************************************
678 Device Interface Subroutines
680 *******************************************************************************/
683 * __dev_get_by_name - find a device by its name
684 * @net: the applicable net namespace
685 * @name: name to find
687 * Find an interface by name. Must be called under RTNL semaphore
688 * or @dev_base_lock. If the name is found a pointer to the device
689 * is returned. If the name is not found then %NULL is returned. The
690 * reference counters are not incremented so the caller must be
691 * careful with locks.
694 struct net_device *__dev_get_by_name(struct net *net, const char *name)
696 struct hlist_node *p;
697 struct net_device *dev;
698 struct hlist_head *head = dev_name_hash(net, name);
700 hlist_for_each_entry(dev, p, head, name_hlist)
701 if (!strncmp(dev->name, name, IFNAMSIZ))
706 EXPORT_SYMBOL(__dev_get_by_name);
709 * dev_get_by_name_rcu - find a device by its name
710 * @net: the applicable net namespace
711 * @name: name to find
713 * Find an interface by name.
714 * If the name is found a pointer to the device is returned.
715 * If the name is not found then %NULL is returned.
716 * The reference counters are not incremented so the caller must be
717 * careful with locks. The caller must hold RCU lock.
720 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
722 struct hlist_node *p;
723 struct net_device *dev;
724 struct hlist_head *head = dev_name_hash(net, name);
726 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
727 if (!strncmp(dev->name, name, IFNAMSIZ))
732 EXPORT_SYMBOL(dev_get_by_name_rcu);
735 * dev_get_by_name - find a device by its name
736 * @net: the applicable net namespace
737 * @name: name to find
739 * Find an interface by name. This can be called from any
740 * context and does its own locking. The returned handle has
741 * the usage count incremented and the caller must use dev_put() to
742 * release it when it is no longer needed. %NULL is returned if no
743 * matching device is found.
746 struct net_device *dev_get_by_name(struct net *net, const char *name)
748 struct net_device *dev;
751 dev = dev_get_by_name_rcu(net, name);
757 EXPORT_SYMBOL(dev_get_by_name);
760 * __dev_get_by_index - find a device by its ifindex
761 * @net: the applicable net namespace
762 * @ifindex: index of device
764 * Search for an interface by index. Returns %NULL if the device
765 * is not found or a pointer to the device. The device has not
766 * had its reference counter increased so the caller must be careful
767 * about locking. The caller must hold either the RTNL semaphore
771 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
773 struct hlist_node *p;
774 struct net_device *dev;
775 struct hlist_head *head = dev_index_hash(net, ifindex);
777 hlist_for_each_entry(dev, p, head, index_hlist)
778 if (dev->ifindex == ifindex)
783 EXPORT_SYMBOL(__dev_get_by_index);
786 * dev_get_by_index_rcu - find a device by its ifindex
787 * @net: the applicable net namespace
788 * @ifindex: index of device
790 * Search for an interface by index. Returns %NULL if the device
791 * is not found or a pointer to the device. The device has not
792 * had its reference counter increased so the caller must be careful
793 * about locking. The caller must hold RCU lock.
796 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
798 struct hlist_node *p;
799 struct net_device *dev;
800 struct hlist_head *head = dev_index_hash(net, ifindex);
802 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
803 if (dev->ifindex == ifindex)
808 EXPORT_SYMBOL(dev_get_by_index_rcu);
812 * dev_get_by_index - find a device by its ifindex
813 * @net: the applicable net namespace
814 * @ifindex: index of device
816 * Search for an interface by index. Returns NULL if the device
817 * is not found or a pointer to the device. The device returned has
818 * had a reference added and the pointer is safe until the user calls
819 * dev_put to indicate they have finished with it.
822 struct net_device *dev_get_by_index(struct net *net, int ifindex)
824 struct net_device *dev;
827 dev = dev_get_by_index_rcu(net, ifindex);
833 EXPORT_SYMBOL(dev_get_by_index);
836 * dev_getbyhwaddr_rcu - find a device by its hardware address
837 * @net: the applicable net namespace
838 * @type: media type of device
839 * @ha: hardware address
841 * Search for an interface by MAC address. Returns NULL if the device
842 * is not found or a pointer to the device.
843 * The caller must hold RCU or RTNL.
844 * The returned device has not had its ref count increased
845 * and the caller must therefore be careful about locking
849 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
852 struct net_device *dev;
854 for_each_netdev_rcu(net, dev)
855 if (dev->type == type &&
856 !memcmp(dev->dev_addr, ha, dev->addr_len))
861 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
863 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
865 struct net_device *dev;
868 for_each_netdev(net, dev)
869 if (dev->type == type)
874 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
876 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
878 struct net_device *dev, *ret = NULL;
881 for_each_netdev_rcu(net, dev)
882 if (dev->type == type) {
890 EXPORT_SYMBOL(dev_getfirstbyhwtype);
893 * dev_get_by_flags_rcu - find any device with given flags
894 * @net: the applicable net namespace
895 * @if_flags: IFF_* values
896 * @mask: bitmask of bits in if_flags to check
898 * Search for any interface with the given flags. Returns NULL if a device
899 * is not found or a pointer to the device. Must be called inside
900 * rcu_read_lock(), and result refcount is unchanged.
903 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
906 struct net_device *dev, *ret;
909 for_each_netdev_rcu(net, dev) {
910 if (((dev->flags ^ if_flags) & mask) == 0) {
917 EXPORT_SYMBOL(dev_get_by_flags_rcu);
920 * dev_valid_name - check if name is okay for network device
923 * Network device names need to be valid file names to
924 * to allow sysfs to work. We also disallow any kind of
927 bool dev_valid_name(const char *name)
931 if (strlen(name) >= IFNAMSIZ)
933 if (!strcmp(name, ".") || !strcmp(name, ".."))
937 if (*name == '/' || isspace(*name))
943 EXPORT_SYMBOL(dev_valid_name);
946 * __dev_alloc_name - allocate a name for a device
947 * @net: network namespace to allocate the device name in
948 * @name: name format string
949 * @buf: scratch buffer and result name string
951 * Passed a format string - eg "lt%d" it will try and find a suitable
952 * id. It scans list of devices to build up a free map, then chooses
953 * the first empty slot. The caller must hold the dev_base or rtnl lock
954 * while allocating the name and adding the device in order to avoid
956 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
957 * Returns the number of the unit assigned or a negative errno code.
960 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
964 const int max_netdevices = 8*PAGE_SIZE;
965 unsigned long *inuse;
966 struct net_device *d;
968 p = strnchr(name, IFNAMSIZ-1, '%');
971 * Verify the string as this thing may have come from
972 * the user. There must be either one "%d" and no other "%"
975 if (p[1] != 'd' || strchr(p + 2, '%'))
978 /* Use one page as a bit array of possible slots */
979 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
983 for_each_netdev(net, d) {
984 if (!sscanf(d->name, name, &i))
986 if (i < 0 || i >= max_netdevices)
989 /* avoid cases where sscanf is not exact inverse of printf */
990 snprintf(buf, IFNAMSIZ, name, i);
991 if (!strncmp(buf, d->name, IFNAMSIZ))
995 i = find_first_zero_bit(inuse, max_netdevices);
996 free_page((unsigned long) inuse);
1000 snprintf(buf, IFNAMSIZ, name, i);
1001 if (!__dev_get_by_name(net, buf))
1004 /* It is possible to run out of possible slots
1005 * when the name is long and there isn't enough space left
1006 * for the digits, or if all bits are used.
1012 * dev_alloc_name - allocate a name for a device
1014 * @name: name format string
1016 * Passed a format string - eg "lt%d" it will try and find a suitable
1017 * id. It scans list of devices to build up a free map, then chooses
1018 * the first empty slot. The caller must hold the dev_base or rtnl lock
1019 * while allocating the name and adding the device in order to avoid
1021 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1022 * Returns the number of the unit assigned or a negative errno code.
1025 int dev_alloc_name(struct net_device *dev, const char *name)
1031 BUG_ON(!dev_net(dev));
1033 ret = __dev_alloc_name(net, name, buf);
1035 strlcpy(dev->name, buf, IFNAMSIZ);
1038 EXPORT_SYMBOL(dev_alloc_name);
1040 static int dev_alloc_name_ns(struct net *net,
1041 struct net_device *dev,
1047 ret = __dev_alloc_name(net, name, buf);
1049 strlcpy(dev->name, buf, IFNAMSIZ);
1053 static int dev_get_valid_name(struct net *net,
1054 struct net_device *dev,
1059 if (!dev_valid_name(name))
1062 if (strchr(name, '%'))
1063 return dev_alloc_name_ns(net, dev, name);
1064 else if (__dev_get_by_name(net, name))
1066 else if (dev->name != name)
1067 strlcpy(dev->name, name, IFNAMSIZ);
1073 * dev_change_name - change name of a device
1075 * @newname: name (or format string) must be at least IFNAMSIZ
1077 * Change name of a device, can pass format strings "eth%d".
1080 int dev_change_name(struct net_device *dev, const char *newname)
1082 char oldname[IFNAMSIZ];
1088 BUG_ON(!dev_net(dev));
1091 if (dev->flags & IFF_UP)
1094 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
1097 memcpy(oldname, dev->name, IFNAMSIZ);
1099 err = dev_get_valid_name(net, dev, newname);
1104 ret = device_rename(&dev->dev, dev->name);
1106 memcpy(dev->name, oldname, IFNAMSIZ);
1110 write_lock_bh(&dev_base_lock);
1111 hlist_del_rcu(&dev->name_hlist);
1112 write_unlock_bh(&dev_base_lock);
1116 write_lock_bh(&dev_base_lock);
1117 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1118 write_unlock_bh(&dev_base_lock);
1120 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1121 ret = notifier_to_errno(ret);
1124 /* err >= 0 after dev_alloc_name() or stores the first errno */
1127 memcpy(dev->name, oldname, IFNAMSIZ);
1130 pr_err("%s: name change rollback failed: %d\n",
1139 * dev_set_alias - change ifalias of a device
1141 * @alias: name up to IFALIASZ
1142 * @len: limit of bytes to copy from info
1144 * Set ifalias for a device,
1146 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1152 if (len >= IFALIASZ)
1157 kfree(dev->ifalias);
1158 dev->ifalias = NULL;
1163 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1166 dev->ifalias = new_ifalias;
1168 strlcpy(dev->ifalias, alias, len+1);
1174 * netdev_features_change - device changes features
1175 * @dev: device to cause notification
1177 * Called to indicate a device has changed features.
1179 void netdev_features_change(struct net_device *dev)
1181 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1183 EXPORT_SYMBOL(netdev_features_change);
1186 * netdev_state_change - device changes state
1187 * @dev: device to cause notification
1189 * Called to indicate a device has changed state. This function calls
1190 * the notifier chains for netdev_chain and sends a NEWLINK message
1191 * to the routing socket.
1193 void netdev_state_change(struct net_device *dev)
1195 if (dev->flags & IFF_UP) {
1196 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1197 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1200 EXPORT_SYMBOL(netdev_state_change);
1203 * netdev_notify_peers - notify network peers about existence of @dev
1204 * @dev: network device
1206 * Generate traffic such that interested network peers are aware of
1207 * @dev, such as by generating a gratuitous ARP. This may be used when
1208 * a device wants to inform the rest of the network about some sort of
1209 * reconfiguration such as a failover event or virtual machine
1212 void netdev_notify_peers(struct net_device *dev)
1215 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1218 EXPORT_SYMBOL(netdev_notify_peers);
1221 * dev_load - load a network module
1222 * @net: the applicable net namespace
1223 * @name: name of interface
1225 * If a network interface is not present and the process has suitable
1226 * privileges this function loads the module. If module loading is not
1227 * available in this kernel then it becomes a nop.
1230 void dev_load(struct net *net, const char *name)
1232 struct net_device *dev;
1236 dev = dev_get_by_name_rcu(net, name);
1240 if (no_module && capable(CAP_NET_ADMIN))
1241 no_module = request_module("netdev-%s", name);
1242 if (no_module && capable(CAP_SYS_MODULE)) {
1243 if (!request_module("%s", name))
1244 pr_warn("Loading kernel module for a network device with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s instead.\n",
1248 EXPORT_SYMBOL(dev_load);
1250 static int __dev_open(struct net_device *dev)
1252 const struct net_device_ops *ops = dev->netdev_ops;
1257 if (!netif_device_present(dev))
1260 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1261 ret = notifier_to_errno(ret);
1265 set_bit(__LINK_STATE_START, &dev->state);
1267 if (ops->ndo_validate_addr)
1268 ret = ops->ndo_validate_addr(dev);
1270 if (!ret && ops->ndo_open)
1271 ret = ops->ndo_open(dev);
1274 clear_bit(__LINK_STATE_START, &dev->state);
1276 dev->flags |= IFF_UP;
1277 net_dmaengine_get();
1278 dev_set_rx_mode(dev);
1280 add_device_randomness(dev->dev_addr, dev->addr_len);
1287 * dev_open - prepare an interface for use.
1288 * @dev: device to open
1290 * Takes a device from down to up state. The device's private open
1291 * function is invoked and then the multicast lists are loaded. Finally
1292 * the device is moved into the up state and a %NETDEV_UP message is
1293 * sent to the netdev notifier chain.
1295 * Calling this function on an active interface is a nop. On a failure
1296 * a negative errno code is returned.
1298 int dev_open(struct net_device *dev)
1302 if (dev->flags & IFF_UP)
1305 ret = __dev_open(dev);
1309 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1310 call_netdevice_notifiers(NETDEV_UP, dev);
1314 EXPORT_SYMBOL(dev_open);
1316 static int __dev_close_many(struct list_head *head)
1318 struct net_device *dev;
1323 list_for_each_entry(dev, head, unreg_list) {
1324 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1326 clear_bit(__LINK_STATE_START, &dev->state);
1328 /* Synchronize to scheduled poll. We cannot touch poll list, it
1329 * can be even on different cpu. So just clear netif_running().
1331 * dev->stop() will invoke napi_disable() on all of it's
1332 * napi_struct instances on this device.
1334 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1337 dev_deactivate_many(head);
1339 list_for_each_entry(dev, head, unreg_list) {
1340 const struct net_device_ops *ops = dev->netdev_ops;
1343 * Call the device specific close. This cannot fail.
1344 * Only if device is UP
1346 * We allow it to be called even after a DETACH hot-plug
1352 dev->flags &= ~IFF_UP;
1353 net_dmaengine_put();
1359 static int __dev_close(struct net_device *dev)
1364 list_add(&dev->unreg_list, &single);
1365 retval = __dev_close_many(&single);
1370 static int dev_close_many(struct list_head *head)
1372 struct net_device *dev, *tmp;
1373 LIST_HEAD(tmp_list);
1375 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1376 if (!(dev->flags & IFF_UP))
1377 list_move(&dev->unreg_list, &tmp_list);
1379 __dev_close_many(head);
1381 list_for_each_entry(dev, head, unreg_list) {
1382 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1383 call_netdevice_notifiers(NETDEV_DOWN, dev);
1386 /* rollback_registered_many needs the complete original list */
1387 list_splice(&tmp_list, head);
1392 * dev_close - shutdown an interface.
1393 * @dev: device to shutdown
1395 * This function moves an active device into down state. A
1396 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1397 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1400 int dev_close(struct net_device *dev)
1402 if (dev->flags & IFF_UP) {
1405 list_add(&dev->unreg_list, &single);
1406 dev_close_many(&single);
1411 EXPORT_SYMBOL(dev_close);
1415 * dev_disable_lro - disable Large Receive Offload on a device
1418 * Disable Large Receive Offload (LRO) on a net device. Must be
1419 * called under RTNL. This is needed if received packets may be
1420 * forwarded to another interface.
1422 void dev_disable_lro(struct net_device *dev)
1425 * If we're trying to disable lro on a vlan device
1426 * use the underlying physical device instead
1428 if (is_vlan_dev(dev))
1429 dev = vlan_dev_real_dev(dev);
1431 dev->wanted_features &= ~NETIF_F_LRO;
1432 netdev_update_features(dev);
1434 if (unlikely(dev->features & NETIF_F_LRO))
1435 netdev_WARN(dev, "failed to disable LRO!\n");
1437 EXPORT_SYMBOL(dev_disable_lro);
1440 static int dev_boot_phase = 1;
1443 * register_netdevice_notifier - register a network notifier block
1446 * Register a notifier to be called when network device events occur.
1447 * The notifier passed is linked into the kernel structures and must
1448 * not be reused until it has been unregistered. A negative errno code
1449 * is returned on a failure.
1451 * When registered all registration and up events are replayed
1452 * to the new notifier to allow device to have a race free
1453 * view of the network device list.
1456 int register_netdevice_notifier(struct notifier_block *nb)
1458 struct net_device *dev;
1459 struct net_device *last;
1464 err = raw_notifier_chain_register(&netdev_chain, nb);
1470 for_each_netdev(net, dev) {
1471 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1472 err = notifier_to_errno(err);
1476 if (!(dev->flags & IFF_UP))
1479 nb->notifier_call(nb, NETDEV_UP, dev);
1490 for_each_netdev(net, dev) {
1494 if (dev->flags & IFF_UP) {
1495 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1496 nb->notifier_call(nb, NETDEV_DOWN, dev);
1498 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1503 raw_notifier_chain_unregister(&netdev_chain, nb);
1506 EXPORT_SYMBOL(register_netdevice_notifier);
1509 * unregister_netdevice_notifier - unregister a network notifier block
1512 * Unregister a notifier previously registered by
1513 * register_netdevice_notifier(). The notifier is unlinked into the
1514 * kernel structures and may then be reused. A negative errno code
1515 * is returned on a failure.
1517 * After unregistering unregister and down device events are synthesized
1518 * for all devices on the device list to the removed notifier to remove
1519 * the need for special case cleanup code.
1522 int unregister_netdevice_notifier(struct notifier_block *nb)
1524 struct net_device *dev;
1529 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1534 for_each_netdev(net, dev) {
1535 if (dev->flags & IFF_UP) {
1536 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1537 nb->notifier_call(nb, NETDEV_DOWN, dev);
1539 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1546 EXPORT_SYMBOL(unregister_netdevice_notifier);
1549 * call_netdevice_notifiers - call all network notifier blocks
1550 * @val: value passed unmodified to notifier function
1551 * @dev: net_device pointer passed unmodified to notifier function
1553 * Call all network notifier blocks. Parameters and return value
1554 * are as for raw_notifier_call_chain().
1557 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1560 return raw_notifier_call_chain(&netdev_chain, val, dev);
1562 EXPORT_SYMBOL(call_netdevice_notifiers);
1564 static struct static_key netstamp_needed __read_mostly;
1565 #ifdef HAVE_JUMP_LABEL
1566 /* We are not allowed to call static_key_slow_dec() from irq context
1567 * If net_disable_timestamp() is called from irq context, defer the
1568 * static_key_slow_dec() calls.
1570 static atomic_t netstamp_needed_deferred;
1573 void net_enable_timestamp(void)
1575 #ifdef HAVE_JUMP_LABEL
1576 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1580 static_key_slow_dec(&netstamp_needed);
1584 WARN_ON(in_interrupt());
1585 static_key_slow_inc(&netstamp_needed);
1587 EXPORT_SYMBOL(net_enable_timestamp);
1589 void net_disable_timestamp(void)
1591 #ifdef HAVE_JUMP_LABEL
1592 if (in_interrupt()) {
1593 atomic_inc(&netstamp_needed_deferred);
1597 static_key_slow_dec(&netstamp_needed);
1599 EXPORT_SYMBOL(net_disable_timestamp);
1601 static inline void net_timestamp_set(struct sk_buff *skb)
1603 skb->tstamp.tv64 = 0;
1604 if (static_key_false(&netstamp_needed))
1605 __net_timestamp(skb);
1608 #define net_timestamp_check(COND, SKB) \
1609 if (static_key_false(&netstamp_needed)) { \
1610 if ((COND) && !(SKB)->tstamp.tv64) \
1611 __net_timestamp(SKB); \
1614 static int net_hwtstamp_validate(struct ifreq *ifr)
1616 struct hwtstamp_config cfg;
1617 enum hwtstamp_tx_types tx_type;
1618 enum hwtstamp_rx_filters rx_filter;
1619 int tx_type_valid = 0;
1620 int rx_filter_valid = 0;
1622 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
1625 if (cfg.flags) /* reserved for future extensions */
1628 tx_type = cfg.tx_type;
1629 rx_filter = cfg.rx_filter;
1632 case HWTSTAMP_TX_OFF:
1633 case HWTSTAMP_TX_ON:
1634 case HWTSTAMP_TX_ONESTEP_SYNC:
1639 switch (rx_filter) {
1640 case HWTSTAMP_FILTER_NONE:
1641 case HWTSTAMP_FILTER_ALL:
1642 case HWTSTAMP_FILTER_SOME:
1643 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1644 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1645 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1646 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1647 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1648 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1649 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1650 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1651 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1652 case HWTSTAMP_FILTER_PTP_V2_EVENT:
1653 case HWTSTAMP_FILTER_PTP_V2_SYNC:
1654 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1655 rx_filter_valid = 1;
1659 if (!tx_type_valid || !rx_filter_valid)
1665 static inline bool is_skb_forwardable(struct net_device *dev,
1666 struct sk_buff *skb)
1670 if (!(dev->flags & IFF_UP))
1673 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1674 if (skb->len <= len)
1677 /* if TSO is enabled, we don't care about the length as the packet
1678 * could be forwarded without being segmented before
1680 if (skb_is_gso(skb))
1687 * dev_forward_skb - loopback an skb to another netif
1689 * @dev: destination network device
1690 * @skb: buffer to forward
1693 * NET_RX_SUCCESS (no congestion)
1694 * NET_RX_DROP (packet was dropped, but freed)
1696 * dev_forward_skb can be used for injecting an skb from the
1697 * start_xmit function of one device into the receive queue
1698 * of another device.
1700 * The receiving device may be in another namespace, so
1701 * we have to clear all information in the skb that could
1702 * impact namespace isolation.
1704 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1706 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1707 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1708 atomic_long_inc(&dev->rx_dropped);
1717 if (unlikely(!is_skb_forwardable(dev, skb))) {
1718 atomic_long_inc(&dev->rx_dropped);
1725 skb->tstamp.tv64 = 0;
1726 skb->pkt_type = PACKET_HOST;
1727 skb->protocol = eth_type_trans(skb, dev);
1731 return netif_rx(skb);
1733 EXPORT_SYMBOL_GPL(dev_forward_skb);
1735 static inline int deliver_skb(struct sk_buff *skb,
1736 struct packet_type *pt_prev,
1737 struct net_device *orig_dev)
1739 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1741 atomic_inc(&skb->users);
1742 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1745 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1747 if (!ptype->af_packet_priv || !skb->sk)
1750 if (ptype->id_match)
1751 return ptype->id_match(ptype, skb->sk);
1752 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1759 * Support routine. Sends outgoing frames to any network
1760 * taps currently in use.
1763 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1765 struct packet_type *ptype;
1766 struct sk_buff *skb2 = NULL;
1767 struct packet_type *pt_prev = NULL;
1770 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1771 /* Never send packets back to the socket
1772 * they originated from - MvS (miquels@drinkel.ow.org)
1774 if ((ptype->dev == dev || !ptype->dev) &&
1775 (!skb_loop_sk(ptype, skb))) {
1777 deliver_skb(skb2, pt_prev, skb->dev);
1782 skb2 = skb_clone(skb, GFP_ATOMIC);
1786 net_timestamp_set(skb2);
1788 /* skb->nh should be correctly
1789 set by sender, so that the second statement is
1790 just protection against buggy protocols.
1792 skb_reset_mac_header(skb2);
1794 if (skb_network_header(skb2) < skb2->data ||
1795 skb2->network_header > skb2->tail) {
1796 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1797 ntohs(skb2->protocol),
1799 skb_reset_network_header(skb2);
1802 skb2->transport_header = skb2->network_header;
1803 skb2->pkt_type = PACKET_OUTGOING;
1808 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1813 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1814 * @dev: Network device
1815 * @txq: number of queues available
1817 * If real_num_tx_queues is changed the tc mappings may no longer be
1818 * valid. To resolve this verify the tc mapping remains valid and if
1819 * not NULL the mapping. With no priorities mapping to this
1820 * offset/count pair it will no longer be used. In the worst case TC0
1821 * is invalid nothing can be done so disable priority mappings. If is
1822 * expected that drivers will fix this mapping if they can before
1823 * calling netif_set_real_num_tx_queues.
1825 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1828 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1830 /* If TC0 is invalidated disable TC mapping */
1831 if (tc->offset + tc->count > txq) {
1832 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1837 /* Invalidated prio to tc mappings set to TC0 */
1838 for (i = 1; i < TC_BITMASK + 1; i++) {
1839 int q = netdev_get_prio_tc_map(dev, i);
1841 tc = &dev->tc_to_txq[q];
1842 if (tc->offset + tc->count > txq) {
1843 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1845 netdev_set_prio_tc_map(dev, i, 0);
1851 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1852 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1854 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1858 if (txq < 1 || txq > dev->num_tx_queues)
1861 if (dev->reg_state == NETREG_REGISTERED ||
1862 dev->reg_state == NETREG_UNREGISTERING) {
1865 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1871 netif_setup_tc(dev, txq);
1873 if (txq < dev->real_num_tx_queues)
1874 qdisc_reset_all_tx_gt(dev, txq);
1877 dev->real_num_tx_queues = txq;
1880 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1884 * netif_set_real_num_rx_queues - set actual number of RX queues used
1885 * @dev: Network device
1886 * @rxq: Actual number of RX queues
1888 * This must be called either with the rtnl_lock held or before
1889 * registration of the net device. Returns 0 on success, or a
1890 * negative error code. If called before registration, it always
1893 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1897 if (rxq < 1 || rxq > dev->num_rx_queues)
1900 if (dev->reg_state == NETREG_REGISTERED) {
1903 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1909 dev->real_num_rx_queues = rxq;
1912 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1916 * netif_get_num_default_rss_queues - default number of RSS queues
1918 * This routine should set an upper limit on the number of RSS queues
1919 * used by default by multiqueue devices.
1921 int netif_get_num_default_rss_queues(void)
1923 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
1925 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
1927 static inline void __netif_reschedule(struct Qdisc *q)
1929 struct softnet_data *sd;
1930 unsigned long flags;
1932 local_irq_save(flags);
1933 sd = &__get_cpu_var(softnet_data);
1934 q->next_sched = NULL;
1935 *sd->output_queue_tailp = q;
1936 sd->output_queue_tailp = &q->next_sched;
1937 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1938 local_irq_restore(flags);
1941 void __netif_schedule(struct Qdisc *q)
1943 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1944 __netif_reschedule(q);
1946 EXPORT_SYMBOL(__netif_schedule);
1948 void dev_kfree_skb_irq(struct sk_buff *skb)
1950 if (atomic_dec_and_test(&skb->users)) {
1951 struct softnet_data *sd;
1952 unsigned long flags;
1954 local_irq_save(flags);
1955 sd = &__get_cpu_var(softnet_data);
1956 skb->next = sd->completion_queue;
1957 sd->completion_queue = skb;
1958 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1959 local_irq_restore(flags);
1962 EXPORT_SYMBOL(dev_kfree_skb_irq);
1964 void dev_kfree_skb_any(struct sk_buff *skb)
1966 if (in_irq() || irqs_disabled())
1967 dev_kfree_skb_irq(skb);
1971 EXPORT_SYMBOL(dev_kfree_skb_any);
1975 * netif_device_detach - mark device as removed
1976 * @dev: network device
1978 * Mark device as removed from system and therefore no longer available.
1980 void netif_device_detach(struct net_device *dev)
1982 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1983 netif_running(dev)) {
1984 netif_tx_stop_all_queues(dev);
1987 EXPORT_SYMBOL(netif_device_detach);
1990 * netif_device_attach - mark device as attached
1991 * @dev: network device
1993 * Mark device as attached from system and restart if needed.
1995 void netif_device_attach(struct net_device *dev)
1997 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1998 netif_running(dev)) {
1999 netif_tx_wake_all_queues(dev);
2000 __netdev_watchdog_up(dev);
2003 EXPORT_SYMBOL(netif_device_attach);
2005 static void skb_warn_bad_offload(const struct sk_buff *skb)
2007 static const netdev_features_t null_features = 0;
2008 struct net_device *dev = skb->dev;
2009 const char *driver = "";
2011 if (dev && dev->dev.parent)
2012 driver = dev_driver_string(dev->dev.parent);
2014 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2015 "gso_type=%d ip_summed=%d\n",
2016 driver, dev ? &dev->features : &null_features,
2017 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2018 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2019 skb_shinfo(skb)->gso_type, skb->ip_summed);
2023 * Invalidate hardware checksum when packet is to be mangled, and
2024 * complete checksum manually on outgoing path.
2026 int skb_checksum_help(struct sk_buff *skb)
2029 int ret = 0, offset;
2031 if (skb->ip_summed == CHECKSUM_COMPLETE)
2032 goto out_set_summed;
2034 if (unlikely(skb_shinfo(skb)->gso_size)) {
2035 skb_warn_bad_offload(skb);
2039 offset = skb_checksum_start_offset(skb);
2040 BUG_ON(offset >= skb_headlen(skb));
2041 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2043 offset += skb->csum_offset;
2044 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2046 if (skb_cloned(skb) &&
2047 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2048 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2053 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2055 skb->ip_summed = CHECKSUM_NONE;
2059 EXPORT_SYMBOL(skb_checksum_help);
2062 * skb_gso_segment - Perform segmentation on skb.
2063 * @skb: buffer to segment
2064 * @features: features for the output path (see dev->features)
2066 * This function segments the given skb and returns a list of segments.
2068 * It may return NULL if the skb requires no segmentation. This is
2069 * only possible when GSO is used for verifying header integrity.
2071 struct sk_buff *skb_gso_segment(struct sk_buff *skb,
2072 netdev_features_t features)
2074 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2075 struct packet_type *ptype;
2076 __be16 type = skb->protocol;
2077 int vlan_depth = ETH_HLEN;
2080 while (type == htons(ETH_P_8021Q)) {
2081 struct vlan_hdr *vh;
2083 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
2084 return ERR_PTR(-EINVAL);
2086 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
2087 type = vh->h_vlan_encapsulated_proto;
2088 vlan_depth += VLAN_HLEN;
2091 skb_reset_mac_header(skb);
2092 skb->mac_len = skb->network_header - skb->mac_header;
2093 __skb_pull(skb, skb->mac_len);
2095 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2096 skb_warn_bad_offload(skb);
2098 if (skb_header_cloned(skb) &&
2099 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
2100 return ERR_PTR(err);
2104 list_for_each_entry_rcu(ptype,
2105 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2106 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
2107 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2108 err = ptype->gso_send_check(skb);
2109 segs = ERR_PTR(err);
2110 if (err || skb_gso_ok(skb, features))
2112 __skb_push(skb, (skb->data -
2113 skb_network_header(skb)));
2115 segs = ptype->gso_segment(skb, features);
2121 __skb_push(skb, skb->data - skb_mac_header(skb));
2125 EXPORT_SYMBOL(skb_gso_segment);
2127 /* Take action when hardware reception checksum errors are detected. */
2129 void netdev_rx_csum_fault(struct net_device *dev)
2131 if (net_ratelimit()) {
2132 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2136 EXPORT_SYMBOL(netdev_rx_csum_fault);
2139 /* Actually, we should eliminate this check as soon as we know, that:
2140 * 1. IOMMU is present and allows to map all the memory.
2141 * 2. No high memory really exists on this machine.
2144 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2146 #ifdef CONFIG_HIGHMEM
2148 if (!(dev->features & NETIF_F_HIGHDMA)) {
2149 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2150 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2151 if (PageHighMem(skb_frag_page(frag)))
2156 if (PCI_DMA_BUS_IS_PHYS) {
2157 struct device *pdev = dev->dev.parent;
2161 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2162 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2163 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2164 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2173 void (*destructor)(struct sk_buff *skb);
2176 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2178 static void dev_gso_skb_destructor(struct sk_buff *skb)
2180 struct dev_gso_cb *cb;
2183 struct sk_buff *nskb = skb->next;
2185 skb->next = nskb->next;
2188 } while (skb->next);
2190 cb = DEV_GSO_CB(skb);
2192 cb->destructor(skb);
2196 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2197 * @skb: buffer to segment
2198 * @features: device features as applicable to this skb
2200 * This function segments the given skb and stores the list of segments
2203 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2205 struct sk_buff *segs;
2207 segs = skb_gso_segment(skb, features);
2209 /* Verifying header integrity only. */
2214 return PTR_ERR(segs);
2217 DEV_GSO_CB(skb)->destructor = skb->destructor;
2218 skb->destructor = dev_gso_skb_destructor;
2223 static bool can_checksum_protocol(netdev_features_t features, __be16 protocol)
2225 return ((features & NETIF_F_GEN_CSUM) ||
2226 ((features & NETIF_F_V4_CSUM) &&
2227 protocol == htons(ETH_P_IP)) ||
2228 ((features & NETIF_F_V6_CSUM) &&
2229 protocol == htons(ETH_P_IPV6)) ||
2230 ((features & NETIF_F_FCOE_CRC) &&
2231 protocol == htons(ETH_P_FCOE)));
2234 static netdev_features_t harmonize_features(struct sk_buff *skb,
2235 __be16 protocol, netdev_features_t features)
2237 if (skb->ip_summed != CHECKSUM_NONE &&
2238 !can_checksum_protocol(features, protocol)) {
2239 features &= ~NETIF_F_ALL_CSUM;
2240 features &= ~NETIF_F_SG;
2241 } else if (illegal_highdma(skb->dev, skb)) {
2242 features &= ~NETIF_F_SG;
2248 netdev_features_t netif_skb_features(struct sk_buff *skb)
2250 __be16 protocol = skb->protocol;
2251 netdev_features_t features = skb->dev->features;
2253 if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2254 features &= ~NETIF_F_GSO_MASK;
2256 if (protocol == htons(ETH_P_8021Q)) {
2257 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2258 protocol = veh->h_vlan_encapsulated_proto;
2259 } else if (!vlan_tx_tag_present(skb)) {
2260 return harmonize_features(skb, protocol, features);
2263 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2265 if (protocol != htons(ETH_P_8021Q)) {
2266 return harmonize_features(skb, protocol, features);
2268 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2269 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2270 return harmonize_features(skb, protocol, features);
2273 EXPORT_SYMBOL(netif_skb_features);
2276 * Returns true if either:
2277 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2278 * 2. skb is fragmented and the device does not support SG.
2280 static inline int skb_needs_linearize(struct sk_buff *skb,
2283 return skb_is_nonlinear(skb) &&
2284 ((skb_has_frag_list(skb) &&
2285 !(features & NETIF_F_FRAGLIST)) ||
2286 (skb_shinfo(skb)->nr_frags &&
2287 !(features & NETIF_F_SG)));
2290 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2291 struct netdev_queue *txq)
2293 const struct net_device_ops *ops = dev->netdev_ops;
2294 int rc = NETDEV_TX_OK;
2295 unsigned int skb_len;
2297 if (likely(!skb->next)) {
2298 netdev_features_t features;
2301 * If device doesn't need skb->dst, release it right now while
2302 * its hot in this cpu cache
2304 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2307 features = netif_skb_features(skb);
2309 if (vlan_tx_tag_present(skb) &&
2310 !(features & NETIF_F_HW_VLAN_TX)) {
2311 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2318 if (netif_needs_gso(skb, features)) {
2319 if (unlikely(dev_gso_segment(skb, features)))
2324 if (skb_needs_linearize(skb, features) &&
2325 __skb_linearize(skb))
2328 /* If packet is not checksummed and device does not
2329 * support checksumming for this protocol, complete
2330 * checksumming here.
2332 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2333 skb_set_transport_header(skb,
2334 skb_checksum_start_offset(skb));
2335 if (!(features & NETIF_F_ALL_CSUM) &&
2336 skb_checksum_help(skb))
2341 if (!list_empty(&ptype_all))
2342 dev_queue_xmit_nit(skb, dev);
2345 rc = ops->ndo_start_xmit(skb, dev);
2346 trace_net_dev_xmit(skb, rc, dev, skb_len);
2347 if (rc == NETDEV_TX_OK)
2348 txq_trans_update(txq);
2354 struct sk_buff *nskb = skb->next;
2356 skb->next = nskb->next;
2360 * If device doesn't need nskb->dst, release it right now while
2361 * its hot in this cpu cache
2363 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2366 if (!list_empty(&ptype_all))
2367 dev_queue_xmit_nit(nskb, dev);
2369 skb_len = nskb->len;
2370 rc = ops->ndo_start_xmit(nskb, dev);
2371 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2372 if (unlikely(rc != NETDEV_TX_OK)) {
2373 if (rc & ~NETDEV_TX_MASK)
2374 goto out_kfree_gso_skb;
2375 nskb->next = skb->next;
2379 txq_trans_update(txq);
2380 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2381 return NETDEV_TX_BUSY;
2382 } while (skb->next);
2385 if (likely(skb->next == NULL))
2386 skb->destructor = DEV_GSO_CB(skb)->destructor;
2393 static u32 hashrnd __read_mostly;
2396 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2397 * to be used as a distribution range.
2399 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2400 unsigned int num_tx_queues)
2404 u16 qcount = num_tx_queues;
2406 if (skb_rx_queue_recorded(skb)) {
2407 hash = skb_get_rx_queue(skb);
2408 while (unlikely(hash >= num_tx_queues))
2409 hash -= num_tx_queues;
2414 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2415 qoffset = dev->tc_to_txq[tc].offset;
2416 qcount = dev->tc_to_txq[tc].count;
2419 if (skb->sk && skb->sk->sk_hash)
2420 hash = skb->sk->sk_hash;
2422 hash = (__force u16) skb->protocol;
2423 hash = jhash_1word(hash, hashrnd);
2425 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2427 EXPORT_SYMBOL(__skb_tx_hash);
2429 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2431 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2432 net_warn_ratelimited("%s selects TX queue %d, but real number of TX queues is %d\n",
2433 dev->name, queue_index,
2434 dev->real_num_tx_queues);
2440 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2443 struct xps_dev_maps *dev_maps;
2444 struct xps_map *map;
2445 int queue_index = -1;
2448 dev_maps = rcu_dereference(dev->xps_maps);
2450 map = rcu_dereference(
2451 dev_maps->cpu_map[raw_smp_processor_id()]);
2454 queue_index = map->queues[0];
2457 if (skb->sk && skb->sk->sk_hash)
2458 hash = skb->sk->sk_hash;
2460 hash = (__force u16) skb->protocol ^
2462 hash = jhash_1word(hash, hashrnd);
2463 queue_index = map->queues[
2464 ((u64)hash * map->len) >> 32];
2466 if (unlikely(queue_index >= dev->real_num_tx_queues))
2478 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
2479 struct sk_buff *skb)
2482 const struct net_device_ops *ops = dev->netdev_ops;
2484 if (dev->real_num_tx_queues == 1)
2486 else if (ops->ndo_select_queue) {
2487 queue_index = ops->ndo_select_queue(dev, skb);
2488 queue_index = dev_cap_txqueue(dev, queue_index);
2490 struct sock *sk = skb->sk;
2491 queue_index = sk_tx_queue_get(sk);
2493 if (queue_index < 0 || skb->ooo_okay ||
2494 queue_index >= dev->real_num_tx_queues) {
2495 int old_index = queue_index;
2497 queue_index = get_xps_queue(dev, skb);
2498 if (queue_index < 0)
2499 queue_index = skb_tx_hash(dev, skb);
2501 if (queue_index != old_index && sk) {
2502 struct dst_entry *dst =
2503 rcu_dereference_check(sk->sk_dst_cache, 1);
2505 if (dst && skb_dst(skb) == dst)
2506 sk_tx_queue_set(sk, queue_index);
2511 skb_set_queue_mapping(skb, queue_index);
2512 return netdev_get_tx_queue(dev, queue_index);
2515 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2516 struct net_device *dev,
2517 struct netdev_queue *txq)
2519 spinlock_t *root_lock = qdisc_lock(q);
2523 qdisc_skb_cb(skb)->pkt_len = skb->len;
2524 qdisc_calculate_pkt_len(skb, q);
2526 * Heuristic to force contended enqueues to serialize on a
2527 * separate lock before trying to get qdisc main lock.
2528 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2529 * and dequeue packets faster.
2531 contended = qdisc_is_running(q);
2532 if (unlikely(contended))
2533 spin_lock(&q->busylock);
2535 spin_lock(root_lock);
2536 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2539 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2540 qdisc_run_begin(q)) {
2542 * This is a work-conserving queue; there are no old skbs
2543 * waiting to be sent out; and the qdisc is not running -
2544 * xmit the skb directly.
2546 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2549 qdisc_bstats_update(q, skb);
2551 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2552 if (unlikely(contended)) {
2553 spin_unlock(&q->busylock);
2560 rc = NET_XMIT_SUCCESS;
2563 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2564 if (qdisc_run_begin(q)) {
2565 if (unlikely(contended)) {
2566 spin_unlock(&q->busylock);
2572 spin_unlock(root_lock);
2573 if (unlikely(contended))
2574 spin_unlock(&q->busylock);
2578 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2579 static void skb_update_prio(struct sk_buff *skb)
2581 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2583 if (!skb->priority && skb->sk && map) {
2584 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2586 if (prioidx < map->priomap_len)
2587 skb->priority = map->priomap[prioidx];
2591 #define skb_update_prio(skb)
2594 static DEFINE_PER_CPU(int, xmit_recursion);
2595 #define RECURSION_LIMIT 10
2598 * dev_loopback_xmit - loop back @skb
2599 * @skb: buffer to transmit
2601 int dev_loopback_xmit(struct sk_buff *skb)
2603 skb_reset_mac_header(skb);
2604 __skb_pull(skb, skb_network_offset(skb));
2605 skb->pkt_type = PACKET_LOOPBACK;
2606 skb->ip_summed = CHECKSUM_UNNECESSARY;
2607 WARN_ON(!skb_dst(skb));
2612 EXPORT_SYMBOL(dev_loopback_xmit);
2615 * dev_queue_xmit - transmit a buffer
2616 * @skb: buffer to transmit
2618 * Queue a buffer for transmission to a network device. The caller must
2619 * have set the device and priority and built the buffer before calling
2620 * this function. The function can be called from an interrupt.
2622 * A negative errno code is returned on a failure. A success does not
2623 * guarantee the frame will be transmitted as it may be dropped due
2624 * to congestion or traffic shaping.
2626 * -----------------------------------------------------------------------------------
2627 * I notice this method can also return errors from the queue disciplines,
2628 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2631 * Regardless of the return value, the skb is consumed, so it is currently
2632 * difficult to retry a send to this method. (You can bump the ref count
2633 * before sending to hold a reference for retry if you are careful.)
2635 * When calling this method, interrupts MUST be enabled. This is because
2636 * the BH enable code must have IRQs enabled so that it will not deadlock.
2639 int dev_queue_xmit(struct sk_buff *skb)
2641 struct net_device *dev = skb->dev;
2642 struct netdev_queue *txq;
2646 /* Disable soft irqs for various locks below. Also
2647 * stops preemption for RCU.
2651 skb_update_prio(skb);
2653 txq = netdev_pick_tx(dev, skb);
2654 q = rcu_dereference_bh(txq->qdisc);
2656 #ifdef CONFIG_NET_CLS_ACT
2657 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2659 trace_net_dev_queue(skb);
2661 rc = __dev_xmit_skb(skb, q, dev, txq);
2665 /* The device has no queue. Common case for software devices:
2666 loopback, all the sorts of tunnels...
2668 Really, it is unlikely that netif_tx_lock protection is necessary
2669 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2671 However, it is possible, that they rely on protection
2674 Check this and shot the lock. It is not prone from deadlocks.
2675 Either shot noqueue qdisc, it is even simpler 8)
2677 if (dev->flags & IFF_UP) {
2678 int cpu = smp_processor_id(); /* ok because BHs are off */
2680 if (txq->xmit_lock_owner != cpu) {
2682 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2683 goto recursion_alert;
2685 HARD_TX_LOCK(dev, txq, cpu);
2687 if (!netif_xmit_stopped(txq)) {
2688 __this_cpu_inc(xmit_recursion);
2689 rc = dev_hard_start_xmit(skb, dev, txq);
2690 __this_cpu_dec(xmit_recursion);
2691 if (dev_xmit_complete(rc)) {
2692 HARD_TX_UNLOCK(dev, txq);
2696 HARD_TX_UNLOCK(dev, txq);
2697 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2700 /* Recursion is detected! It is possible,
2704 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2710 rcu_read_unlock_bh();
2715 rcu_read_unlock_bh();
2718 EXPORT_SYMBOL(dev_queue_xmit);
2721 /*=======================================================================
2723 =======================================================================*/
2725 int netdev_max_backlog __read_mostly = 1000;
2726 EXPORT_SYMBOL(netdev_max_backlog);
2728 int netdev_tstamp_prequeue __read_mostly = 1;
2729 int netdev_budget __read_mostly = 300;
2730 int weight_p __read_mostly = 64; /* old backlog weight */
2732 /* Called with irq disabled */
2733 static inline void ____napi_schedule(struct softnet_data *sd,
2734 struct napi_struct *napi)
2736 list_add_tail(&napi->poll_list, &sd->poll_list);
2737 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2741 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2742 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2743 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2744 * if hash is a canonical 4-tuple hash over transport ports.
2746 void __skb_get_rxhash(struct sk_buff *skb)
2748 struct flow_keys keys;
2751 if (!skb_flow_dissect(skb, &keys))
2757 /* get a consistent hash (same value on both flow directions) */
2758 if (((__force u32)keys.dst < (__force u32)keys.src) ||
2759 (((__force u32)keys.dst == (__force u32)keys.src) &&
2760 ((__force u16)keys.port16[1] < (__force u16)keys.port16[0]))) {
2761 swap(keys.dst, keys.src);
2762 swap(keys.port16[0], keys.port16[1]);
2765 hash = jhash_3words((__force u32)keys.dst,
2766 (__force u32)keys.src,
2767 (__force u32)keys.ports, hashrnd);
2773 EXPORT_SYMBOL(__skb_get_rxhash);
2777 /* One global table that all flow-based protocols share. */
2778 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2779 EXPORT_SYMBOL(rps_sock_flow_table);
2781 struct static_key rps_needed __read_mostly;
2783 static struct rps_dev_flow *
2784 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2785 struct rps_dev_flow *rflow, u16 next_cpu)
2787 if (next_cpu != RPS_NO_CPU) {
2788 #ifdef CONFIG_RFS_ACCEL
2789 struct netdev_rx_queue *rxqueue;
2790 struct rps_dev_flow_table *flow_table;
2791 struct rps_dev_flow *old_rflow;
2796 /* Should we steer this flow to a different hardware queue? */
2797 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2798 !(dev->features & NETIF_F_NTUPLE))
2800 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2801 if (rxq_index == skb_get_rx_queue(skb))
2804 rxqueue = dev->_rx + rxq_index;
2805 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2808 flow_id = skb->rxhash & flow_table->mask;
2809 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2810 rxq_index, flow_id);
2814 rflow = &flow_table->flows[flow_id];
2816 if (old_rflow->filter == rflow->filter)
2817 old_rflow->filter = RPS_NO_FILTER;
2821 per_cpu(softnet_data, next_cpu).input_queue_head;
2824 rflow->cpu = next_cpu;
2829 * get_rps_cpu is called from netif_receive_skb and returns the target
2830 * CPU from the RPS map of the receiving queue for a given skb.
2831 * rcu_read_lock must be held on entry.
2833 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2834 struct rps_dev_flow **rflowp)
2836 struct netdev_rx_queue *rxqueue;
2837 struct rps_map *map;
2838 struct rps_dev_flow_table *flow_table;
2839 struct rps_sock_flow_table *sock_flow_table;
2843 if (skb_rx_queue_recorded(skb)) {
2844 u16 index = skb_get_rx_queue(skb);
2845 if (unlikely(index >= dev->real_num_rx_queues)) {
2846 WARN_ONCE(dev->real_num_rx_queues > 1,
2847 "%s received packet on queue %u, but number "
2848 "of RX queues is %u\n",
2849 dev->name, index, dev->real_num_rx_queues);
2852 rxqueue = dev->_rx + index;
2856 map = rcu_dereference(rxqueue->rps_map);
2858 if (map->len == 1 &&
2859 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2860 tcpu = map->cpus[0];
2861 if (cpu_online(tcpu))
2865 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2869 skb_reset_network_header(skb);
2870 if (!skb_get_rxhash(skb))
2873 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2874 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2875 if (flow_table && sock_flow_table) {
2877 struct rps_dev_flow *rflow;
2879 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2882 next_cpu = sock_flow_table->ents[skb->rxhash &
2883 sock_flow_table->mask];
2886 * If the desired CPU (where last recvmsg was done) is
2887 * different from current CPU (one in the rx-queue flow
2888 * table entry), switch if one of the following holds:
2889 * - Current CPU is unset (equal to RPS_NO_CPU).
2890 * - Current CPU is offline.
2891 * - The current CPU's queue tail has advanced beyond the
2892 * last packet that was enqueued using this table entry.
2893 * This guarantees that all previous packets for the flow
2894 * have been dequeued, thus preserving in order delivery.
2896 if (unlikely(tcpu != next_cpu) &&
2897 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2898 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2899 rflow->last_qtail)) >= 0))
2900 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2902 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2910 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2912 if (cpu_online(tcpu)) {
2922 #ifdef CONFIG_RFS_ACCEL
2925 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2926 * @dev: Device on which the filter was set
2927 * @rxq_index: RX queue index
2928 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2929 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2931 * Drivers that implement ndo_rx_flow_steer() should periodically call
2932 * this function for each installed filter and remove the filters for
2933 * which it returns %true.
2935 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2936 u32 flow_id, u16 filter_id)
2938 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2939 struct rps_dev_flow_table *flow_table;
2940 struct rps_dev_flow *rflow;
2945 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2946 if (flow_table && flow_id <= flow_table->mask) {
2947 rflow = &flow_table->flows[flow_id];
2948 cpu = ACCESS_ONCE(rflow->cpu);
2949 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2950 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2951 rflow->last_qtail) <
2952 (int)(10 * flow_table->mask)))
2958 EXPORT_SYMBOL(rps_may_expire_flow);
2960 #endif /* CONFIG_RFS_ACCEL */
2962 /* Called from hardirq (IPI) context */
2963 static void rps_trigger_softirq(void *data)
2965 struct softnet_data *sd = data;
2967 ____napi_schedule(sd, &sd->backlog);
2971 #endif /* CONFIG_RPS */
2974 * Check if this softnet_data structure is another cpu one
2975 * If yes, queue it to our IPI list and return 1
2978 static int rps_ipi_queued(struct softnet_data *sd)
2981 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2984 sd->rps_ipi_next = mysd->rps_ipi_list;
2985 mysd->rps_ipi_list = sd;
2987 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2990 #endif /* CONFIG_RPS */
2995 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2996 * queue (may be a remote CPU queue).
2998 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2999 unsigned int *qtail)
3001 struct softnet_data *sd;
3002 unsigned long flags;
3004 sd = &per_cpu(softnet_data, cpu);
3006 local_irq_save(flags);
3009 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
3010 if (skb_queue_len(&sd->input_pkt_queue)) {
3012 __skb_queue_tail(&sd->input_pkt_queue, skb);
3013 input_queue_tail_incr_save(sd, qtail);
3015 local_irq_restore(flags);
3016 return NET_RX_SUCCESS;
3019 /* Schedule NAPI for backlog device
3020 * We can use non atomic operation since we own the queue lock
3022 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3023 if (!rps_ipi_queued(sd))
3024 ____napi_schedule(sd, &sd->backlog);
3032 local_irq_restore(flags);
3034 atomic_long_inc(&skb->dev->rx_dropped);
3040 * netif_rx - post buffer to the network code
3041 * @skb: buffer to post
3043 * This function receives a packet from a device driver and queues it for
3044 * the upper (protocol) levels to process. It always succeeds. The buffer
3045 * may be dropped during processing for congestion control or by the
3049 * NET_RX_SUCCESS (no congestion)
3050 * NET_RX_DROP (packet was dropped)
3054 int netif_rx(struct sk_buff *skb)
3058 /* if netpoll wants it, pretend we never saw it */
3059 if (netpoll_rx(skb))
3062 net_timestamp_check(netdev_tstamp_prequeue, skb);
3064 trace_netif_rx(skb);
3066 if (static_key_false(&rps_needed)) {
3067 struct rps_dev_flow voidflow, *rflow = &voidflow;
3073 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3075 cpu = smp_processor_id();
3077 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3085 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3090 EXPORT_SYMBOL(netif_rx);
3092 int netif_rx_ni(struct sk_buff *skb)
3097 err = netif_rx(skb);
3098 if (local_softirq_pending())
3104 EXPORT_SYMBOL(netif_rx_ni);
3106 static void net_tx_action(struct softirq_action *h)
3108 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3110 if (sd->completion_queue) {
3111 struct sk_buff *clist;
3113 local_irq_disable();
3114 clist = sd->completion_queue;
3115 sd->completion_queue = NULL;
3119 struct sk_buff *skb = clist;
3120 clist = clist->next;
3122 WARN_ON(atomic_read(&skb->users));
3123 trace_kfree_skb(skb, net_tx_action);
3128 if (sd->output_queue) {
3131 local_irq_disable();
3132 head = sd->output_queue;
3133 sd->output_queue = NULL;
3134 sd->output_queue_tailp = &sd->output_queue;
3138 struct Qdisc *q = head;
3139 spinlock_t *root_lock;
3141 head = head->next_sched;
3143 root_lock = qdisc_lock(q);
3144 if (spin_trylock(root_lock)) {
3145 smp_mb__before_clear_bit();
3146 clear_bit(__QDISC_STATE_SCHED,
3149 spin_unlock(root_lock);
3151 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3153 __netif_reschedule(q);
3155 smp_mb__before_clear_bit();
3156 clear_bit(__QDISC_STATE_SCHED,
3164 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3165 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3166 /* This hook is defined here for ATM LANE */
3167 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3168 unsigned char *addr) __read_mostly;
3169 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3172 #ifdef CONFIG_NET_CLS_ACT
3173 /* TODO: Maybe we should just force sch_ingress to be compiled in
3174 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3175 * a compare and 2 stores extra right now if we dont have it on
3176 * but have CONFIG_NET_CLS_ACT
3177 * NOTE: This doesn't stop any functionality; if you dont have
3178 * the ingress scheduler, you just can't add policies on ingress.
3181 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3183 struct net_device *dev = skb->dev;
3184 u32 ttl = G_TC_RTTL(skb->tc_verd);
3185 int result = TC_ACT_OK;
3188 if (unlikely(MAX_RED_LOOP < ttl++)) {
3189 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3190 skb->skb_iif, dev->ifindex);
3194 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3195 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3198 if (q != &noop_qdisc) {
3199 spin_lock(qdisc_lock(q));
3200 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3201 result = qdisc_enqueue_root(skb, q);
3202 spin_unlock(qdisc_lock(q));
3208 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3209 struct packet_type **pt_prev,
3210 int *ret, struct net_device *orig_dev)
3212 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3214 if (!rxq || rxq->qdisc == &noop_qdisc)
3218 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3222 switch (ing_filter(skb, rxq)) {
3236 * netdev_rx_handler_register - register receive handler
3237 * @dev: device to register a handler for
3238 * @rx_handler: receive handler to register
3239 * @rx_handler_data: data pointer that is used by rx handler
3241 * Register a receive hander for a device. This handler will then be
3242 * called from __netif_receive_skb. A negative errno code is returned
3245 * The caller must hold the rtnl_mutex.
3247 * For a general description of rx_handler, see enum rx_handler_result.
3249 int netdev_rx_handler_register(struct net_device *dev,
3250 rx_handler_func_t *rx_handler,
3251 void *rx_handler_data)
3255 if (dev->rx_handler)
3258 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3259 rcu_assign_pointer(dev->rx_handler, rx_handler);
3263 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3266 * netdev_rx_handler_unregister - unregister receive handler
3267 * @dev: device to unregister a handler from
3269 * Unregister a receive hander from a device.
3271 * The caller must hold the rtnl_mutex.
3273 void netdev_rx_handler_unregister(struct net_device *dev)
3277 RCU_INIT_POINTER(dev->rx_handler, NULL);
3278 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3280 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3283 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3284 * the special handling of PFMEMALLOC skbs.
3286 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3288 switch (skb->protocol) {
3289 case __constant_htons(ETH_P_ARP):
3290 case __constant_htons(ETH_P_IP):
3291 case __constant_htons(ETH_P_IPV6):
3292 case __constant_htons(ETH_P_8021Q):
3299 static int __netif_receive_skb(struct sk_buff *skb)
3301 struct packet_type *ptype, *pt_prev;
3302 rx_handler_func_t *rx_handler;
3303 struct net_device *orig_dev;
3304 struct net_device *null_or_dev;
3305 bool deliver_exact = false;
3306 int ret = NET_RX_DROP;
3308 unsigned long pflags = current->flags;
3310 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3312 trace_netif_receive_skb(skb);
3315 * PFMEMALLOC skbs are special, they should
3316 * - be delivered to SOCK_MEMALLOC sockets only
3317 * - stay away from userspace
3318 * - have bounded memory usage
3320 * Use PF_MEMALLOC as this saves us from propagating the allocation
3321 * context down to all allocation sites.
3323 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
3324 current->flags |= PF_MEMALLOC;
3326 /* if we've gotten here through NAPI, check netpoll */
3327 if (netpoll_receive_skb(skb))
3330 orig_dev = skb->dev;
3332 skb_reset_network_header(skb);
3333 skb_reset_transport_header(skb);
3334 skb_reset_mac_len(skb);
3341 skb->skb_iif = skb->dev->ifindex;
3343 __this_cpu_inc(softnet_data.processed);
3345 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3346 skb = vlan_untag(skb);
3351 #ifdef CONFIG_NET_CLS_ACT
3352 if (skb->tc_verd & TC_NCLS) {
3353 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3358 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
3361 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3362 if (!ptype->dev || ptype->dev == skb->dev) {
3364 ret = deliver_skb(skb, pt_prev, orig_dev);
3370 #ifdef CONFIG_NET_CLS_ACT
3371 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3377 if (sk_memalloc_socks() && skb_pfmemalloc(skb)
3378 && !skb_pfmemalloc_protocol(skb))
3381 if (vlan_tx_tag_present(skb)) {
3383 ret = deliver_skb(skb, pt_prev, orig_dev);
3386 if (vlan_do_receive(&skb))
3388 else if (unlikely(!skb))
3392 rx_handler = rcu_dereference(skb->dev->rx_handler);
3395 ret = deliver_skb(skb, pt_prev, orig_dev);
3398 switch (rx_handler(&skb)) {
3399 case RX_HANDLER_CONSUMED:
3401 case RX_HANDLER_ANOTHER:
3403 case RX_HANDLER_EXACT:
3404 deliver_exact = true;
3405 case RX_HANDLER_PASS:
3412 if (vlan_tx_nonzero_tag_present(skb))
3413 skb->pkt_type = PACKET_OTHERHOST;
3415 /* deliver only exact match when indicated */
3416 null_or_dev = deliver_exact ? skb->dev : NULL;
3418 type = skb->protocol;
3419 list_for_each_entry_rcu(ptype,
3420 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3421 if (ptype->type == type &&
3422 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3423 ptype->dev == orig_dev)) {
3425 ret = deliver_skb(skb, pt_prev, orig_dev);
3431 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3434 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3437 atomic_long_inc(&skb->dev->rx_dropped);
3439 /* Jamal, now you will not able to escape explaining
3440 * me how you were going to use this. :-)
3448 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3453 * netif_receive_skb - process receive buffer from network
3454 * @skb: buffer to process
3456 * netif_receive_skb() is the main receive data processing function.
3457 * It always succeeds. The buffer may be dropped during processing
3458 * for congestion control or by the protocol layers.
3460 * This function may only be called from softirq context and interrupts
3461 * should be enabled.
3463 * Return values (usually ignored):
3464 * NET_RX_SUCCESS: no congestion
3465 * NET_RX_DROP: packet was dropped
3467 int netif_receive_skb(struct sk_buff *skb)
3469 net_timestamp_check(netdev_tstamp_prequeue, skb);
3471 if (skb_defer_rx_timestamp(skb))
3472 return NET_RX_SUCCESS;
3475 if (static_key_false(&rps_needed)) {
3476 struct rps_dev_flow voidflow, *rflow = &voidflow;
3481 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3484 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3491 return __netif_receive_skb(skb);
3493 EXPORT_SYMBOL(netif_receive_skb);
3495 /* Network device is going away, flush any packets still pending
3496 * Called with irqs disabled.
3498 static void flush_backlog(void *arg)
3500 struct net_device *dev = arg;
3501 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3502 struct sk_buff *skb, *tmp;
3505 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3506 if (skb->dev == dev) {
3507 __skb_unlink(skb, &sd->input_pkt_queue);
3509 input_queue_head_incr(sd);
3514 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3515 if (skb->dev == dev) {
3516 __skb_unlink(skb, &sd->process_queue);
3518 input_queue_head_incr(sd);
3523 static int napi_gro_complete(struct sk_buff *skb)
3525 struct packet_type *ptype;
3526 __be16 type = skb->protocol;
3527 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3530 if (NAPI_GRO_CB(skb)->count == 1) {
3531 skb_shinfo(skb)->gso_size = 0;
3536 list_for_each_entry_rcu(ptype, head, list) {
3537 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3540 err = ptype->gro_complete(skb);
3546 WARN_ON(&ptype->list == head);
3548 return NET_RX_SUCCESS;
3552 return netif_receive_skb(skb);
3555 /* napi->gro_list contains packets ordered by age.
3556 * youngest packets at the head of it.
3557 * Complete skbs in reverse order to reduce latencies.
3559 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3561 struct sk_buff *skb, *prev = NULL;
3563 /* scan list and build reverse chain */
3564 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3569 for (skb = prev; skb; skb = prev) {
3572 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3576 napi_gro_complete(skb);
3580 napi->gro_list = NULL;
3582 EXPORT_SYMBOL(napi_gro_flush);
3584 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3586 struct sk_buff **pp = NULL;
3587 struct packet_type *ptype;
3588 __be16 type = skb->protocol;
3589 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3592 enum gro_result ret;
3594 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3597 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3601 list_for_each_entry_rcu(ptype, head, list) {
3602 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3605 skb_set_network_header(skb, skb_gro_offset(skb));
3606 mac_len = skb->network_header - skb->mac_header;
3607 skb->mac_len = mac_len;
3608 NAPI_GRO_CB(skb)->same_flow = 0;
3609 NAPI_GRO_CB(skb)->flush = 0;
3610 NAPI_GRO_CB(skb)->free = 0;
3612 pp = ptype->gro_receive(&napi->gro_list, skb);
3617 if (&ptype->list == head)
3620 same_flow = NAPI_GRO_CB(skb)->same_flow;
3621 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3624 struct sk_buff *nskb = *pp;
3628 napi_gro_complete(nskb);
3635 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3639 NAPI_GRO_CB(skb)->count = 1;
3640 NAPI_GRO_CB(skb)->age = jiffies;
3641 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3642 skb->next = napi->gro_list;
3643 napi->gro_list = skb;
3647 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3648 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3650 BUG_ON(skb->end - skb->tail < grow);
3652 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3655 skb->data_len -= grow;
3657 skb_shinfo(skb)->frags[0].page_offset += grow;
3658 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3660 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3661 skb_frag_unref(skb, 0);
3662 memmove(skb_shinfo(skb)->frags,
3663 skb_shinfo(skb)->frags + 1,
3664 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3675 EXPORT_SYMBOL(dev_gro_receive);
3677 static inline gro_result_t
3678 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3681 unsigned int maclen = skb->dev->hard_header_len;
3683 for (p = napi->gro_list; p; p = p->next) {
3684 unsigned long diffs;
3686 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3687 diffs |= p->vlan_tci ^ skb->vlan_tci;
3688 if (maclen == ETH_HLEN)
3689 diffs |= compare_ether_header(skb_mac_header(p),
3690 skb_gro_mac_header(skb));
3692 diffs = memcmp(skb_mac_header(p),
3693 skb_gro_mac_header(skb),
3695 NAPI_GRO_CB(p)->same_flow = !diffs;
3696 NAPI_GRO_CB(p)->flush = 0;
3699 return dev_gro_receive(napi, skb);
3702 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3706 if (netif_receive_skb(skb))
3714 case GRO_MERGED_FREE:
3715 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
3716 kmem_cache_free(skbuff_head_cache, skb);
3728 EXPORT_SYMBOL(napi_skb_finish);
3730 static void skb_gro_reset_offset(struct sk_buff *skb)
3732 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3733 const skb_frag_t *frag0 = &pinfo->frags[0];
3735 NAPI_GRO_CB(skb)->data_offset = 0;
3736 NAPI_GRO_CB(skb)->frag0 = NULL;
3737 NAPI_GRO_CB(skb)->frag0_len = 0;
3739 if (skb->mac_header == skb->tail &&
3741 !PageHighMem(skb_frag_page(frag0))) {
3742 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3743 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3747 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3749 skb_gro_reset_offset(skb);
3751 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3753 EXPORT_SYMBOL(napi_gro_receive);
3755 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3757 __skb_pull(skb, skb_headlen(skb));
3758 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3759 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3761 skb->dev = napi->dev;
3767 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3769 struct sk_buff *skb = napi->skb;
3772 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3778 EXPORT_SYMBOL(napi_get_frags);
3780 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3786 skb->protocol = eth_type_trans(skb, skb->dev);
3788 if (ret == GRO_HELD)
3789 skb_gro_pull(skb, -ETH_HLEN);
3790 else if (netif_receive_skb(skb))
3795 case GRO_MERGED_FREE:
3796 napi_reuse_skb(napi, skb);
3805 EXPORT_SYMBOL(napi_frags_finish);
3807 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3809 struct sk_buff *skb = napi->skb;
3816 skb_reset_mac_header(skb);
3817 skb_gro_reset_offset(skb);
3819 off = skb_gro_offset(skb);
3820 hlen = off + sizeof(*eth);
3821 eth = skb_gro_header_fast(skb, off);
3822 if (skb_gro_header_hard(skb, hlen)) {
3823 eth = skb_gro_header_slow(skb, hlen, off);
3824 if (unlikely(!eth)) {
3825 napi_reuse_skb(napi, skb);
3831 skb_gro_pull(skb, sizeof(*eth));
3834 * This works because the only protocols we care about don't require
3835 * special handling. We'll fix it up properly at the end.
3837 skb->protocol = eth->h_proto;
3843 gro_result_t napi_gro_frags(struct napi_struct *napi)
3845 struct sk_buff *skb = napi_frags_skb(napi);
3850 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3852 EXPORT_SYMBOL(napi_gro_frags);
3855 * net_rps_action sends any pending IPI's for rps.
3856 * Note: called with local irq disabled, but exits with local irq enabled.
3858 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3861 struct softnet_data *remsd = sd->rps_ipi_list;
3864 sd->rps_ipi_list = NULL;
3868 /* Send pending IPI's to kick RPS processing on remote cpus. */
3870 struct softnet_data *next = remsd->rps_ipi_next;
3872 if (cpu_online(remsd->cpu))
3873 __smp_call_function_single(remsd->cpu,
3882 static int process_backlog(struct napi_struct *napi, int quota)
3885 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3888 /* Check if we have pending ipi, its better to send them now,
3889 * not waiting net_rx_action() end.
3891 if (sd->rps_ipi_list) {
3892 local_irq_disable();
3893 net_rps_action_and_irq_enable(sd);
3896 napi->weight = weight_p;
3897 local_irq_disable();
3898 while (work < quota) {
3899 struct sk_buff *skb;
3902 while ((skb = __skb_dequeue(&sd->process_queue))) {
3904 __netif_receive_skb(skb);
3905 local_irq_disable();
3906 input_queue_head_incr(sd);
3907 if (++work >= quota) {
3914 qlen = skb_queue_len(&sd->input_pkt_queue);
3916 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3917 &sd->process_queue);
3919 if (qlen < quota - work) {
3921 * Inline a custom version of __napi_complete().
3922 * only current cpu owns and manipulates this napi,
3923 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3924 * we can use a plain write instead of clear_bit(),
3925 * and we dont need an smp_mb() memory barrier.
3927 list_del(&napi->poll_list);
3930 quota = work + qlen;
3940 * __napi_schedule - schedule for receive
3941 * @n: entry to schedule
3943 * The entry's receive function will be scheduled to run
3945 void __napi_schedule(struct napi_struct *n)
3947 unsigned long flags;
3949 local_irq_save(flags);
3950 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3951 local_irq_restore(flags);
3953 EXPORT_SYMBOL(__napi_schedule);
3955 void __napi_complete(struct napi_struct *n)
3957 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3958 BUG_ON(n->gro_list);
3960 list_del(&n->poll_list);
3961 smp_mb__before_clear_bit();
3962 clear_bit(NAPI_STATE_SCHED, &n->state);
3964 EXPORT_SYMBOL(__napi_complete);
3966 void napi_complete(struct napi_struct *n)
3968 unsigned long flags;
3971 * don't let napi dequeue from the cpu poll list
3972 * just in case its running on a different cpu
3974 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3977 napi_gro_flush(n, false);
3978 local_irq_save(flags);
3980 local_irq_restore(flags);
3982 EXPORT_SYMBOL(napi_complete);
3984 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3985 int (*poll)(struct napi_struct *, int), int weight)
3987 INIT_LIST_HEAD(&napi->poll_list);
3988 napi->gro_count = 0;
3989 napi->gro_list = NULL;
3992 napi->weight = weight;
3993 list_add(&napi->dev_list, &dev->napi_list);
3995 #ifdef CONFIG_NETPOLL
3996 spin_lock_init(&napi->poll_lock);
3997 napi->poll_owner = -1;
3999 set_bit(NAPI_STATE_SCHED, &napi->state);
4001 EXPORT_SYMBOL(netif_napi_add);
4003 void netif_napi_del(struct napi_struct *napi)
4005 struct sk_buff *skb, *next;
4007 list_del_init(&napi->dev_list);
4008 napi_free_frags(napi);
4010 for (skb = napi->gro_list; skb; skb = next) {
4016 napi->gro_list = NULL;
4017 napi->gro_count = 0;
4019 EXPORT_SYMBOL(netif_napi_del);
4021 static void net_rx_action(struct softirq_action *h)
4023 struct softnet_data *sd = &__get_cpu_var(softnet_data);
4024 unsigned long time_limit = jiffies + 2;
4025 int budget = netdev_budget;
4028 local_irq_disable();
4030 while (!list_empty(&sd->poll_list)) {
4031 struct napi_struct *n;
4034 /* If softirq window is exhuasted then punt.
4035 * Allow this to run for 2 jiffies since which will allow
4036 * an average latency of 1.5/HZ.
4038 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
4043 /* Even though interrupts have been re-enabled, this
4044 * access is safe because interrupts can only add new
4045 * entries to the tail of this list, and only ->poll()
4046 * calls can remove this head entry from the list.
4048 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
4050 have = netpoll_poll_lock(n);
4054 /* This NAPI_STATE_SCHED test is for avoiding a race
4055 * with netpoll's poll_napi(). Only the entity which
4056 * obtains the lock and sees NAPI_STATE_SCHED set will
4057 * actually make the ->poll() call. Therefore we avoid
4058 * accidentally calling ->poll() when NAPI is not scheduled.
4061 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4062 work = n->poll(n, weight);
4066 WARN_ON_ONCE(work > weight);
4070 local_irq_disable();
4072 /* Drivers must not modify the NAPI state if they
4073 * consume the entire weight. In such cases this code
4074 * still "owns" the NAPI instance and therefore can
4075 * move the instance around on the list at-will.
4077 if (unlikely(work == weight)) {
4078 if (unlikely(napi_disable_pending(n))) {
4081 local_irq_disable();
4084 /* flush too old packets
4085 * If HZ < 1000, flush all packets.
4088 napi_gro_flush(n, HZ >= 1000);
4089 local_irq_disable();
4091 list_move_tail(&n->poll_list, &sd->poll_list);
4095 netpoll_poll_unlock(have);
4098 net_rps_action_and_irq_enable(sd);
4100 #ifdef CONFIG_NET_DMA
4102 * There may not be any more sk_buffs coming right now, so push
4103 * any pending DMA copies to hardware
4105 dma_issue_pending_all();
4112 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4116 static gifconf_func_t *gifconf_list[NPROTO];
4119 * register_gifconf - register a SIOCGIF handler
4120 * @family: Address family
4121 * @gifconf: Function handler
4123 * Register protocol dependent address dumping routines. The handler
4124 * that is passed must not be freed or reused until it has been replaced
4125 * by another handler.
4127 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
4129 if (family >= NPROTO)
4131 gifconf_list[family] = gifconf;
4134 EXPORT_SYMBOL(register_gifconf);
4138 * Map an interface index to its name (SIOCGIFNAME)
4142 * We need this ioctl for efficient implementation of the
4143 * if_indextoname() function required by the IPv6 API. Without
4144 * it, we would have to search all the interfaces to find a
4148 static int dev_ifname(struct net *net, struct ifreq __user *arg)
4150 struct net_device *dev;
4154 * Fetch the caller's info block.
4157 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4161 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
4167 strcpy(ifr.ifr_name, dev->name);
4170 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
4176 * Perform a SIOCGIFCONF call. This structure will change
4177 * size eventually, and there is nothing I can do about it.
4178 * Thus we will need a 'compatibility mode'.
4181 static int dev_ifconf(struct net *net, char __user *arg)
4184 struct net_device *dev;
4191 * Fetch the caller's info block.
4194 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
4201 * Loop over the interfaces, and write an info block for each.
4205 for_each_netdev(net, dev) {
4206 for (i = 0; i < NPROTO; i++) {
4207 if (gifconf_list[i]) {
4210 done = gifconf_list[i](dev, NULL, 0);
4212 done = gifconf_list[i](dev, pos + total,
4222 * All done. Write the updated control block back to the caller.
4224 ifc.ifc_len = total;
4227 * Both BSD and Solaris return 0 here, so we do too.
4229 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4232 #ifdef CONFIG_PROC_FS
4234 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4236 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4237 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4238 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4240 static inline struct net_device *dev_from_same_bucket(struct seq_file *seq, loff_t *pos)
4242 struct net *net = seq_file_net(seq);
4243 struct net_device *dev;
4244 struct hlist_node *p;
4245 struct hlist_head *h;
4246 unsigned int count = 0, offset = get_offset(*pos);
4248 h = &net->dev_name_head[get_bucket(*pos)];
4249 hlist_for_each_entry_rcu(dev, p, h, name_hlist) {
4250 if (++count == offset)
4257 static inline struct net_device *dev_from_bucket(struct seq_file *seq, loff_t *pos)
4259 struct net_device *dev;
4260 unsigned int bucket;
4263 dev = dev_from_same_bucket(seq, pos);
4267 bucket = get_bucket(*pos) + 1;
4268 *pos = set_bucket_offset(bucket, 1);
4269 } while (bucket < NETDEV_HASHENTRIES);
4275 * This is invoked by the /proc filesystem handler to display a device
4278 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4283 return SEQ_START_TOKEN;
4285 if (get_bucket(*pos) >= NETDEV_HASHENTRIES)
4288 return dev_from_bucket(seq, pos);
4291 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4294 return dev_from_bucket(seq, pos);
4297 void dev_seq_stop(struct seq_file *seq, void *v)
4303 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4305 struct rtnl_link_stats64 temp;
4306 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4308 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4309 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4310 dev->name, stats->rx_bytes, stats->rx_packets,
4312 stats->rx_dropped + stats->rx_missed_errors,
4313 stats->rx_fifo_errors,
4314 stats->rx_length_errors + stats->rx_over_errors +
4315 stats->rx_crc_errors + stats->rx_frame_errors,
4316 stats->rx_compressed, stats->multicast,
4317 stats->tx_bytes, stats->tx_packets,
4318 stats->tx_errors, stats->tx_dropped,
4319 stats->tx_fifo_errors, stats->collisions,
4320 stats->tx_carrier_errors +
4321 stats->tx_aborted_errors +
4322 stats->tx_window_errors +
4323 stats->tx_heartbeat_errors,
4324 stats->tx_compressed);
4328 * Called from the PROCfs module. This now uses the new arbitrary sized
4329 * /proc/net interface to create /proc/net/dev
4331 static int dev_seq_show(struct seq_file *seq, void *v)
4333 if (v == SEQ_START_TOKEN)
4334 seq_puts(seq, "Inter-| Receive "
4336 " face |bytes packets errs drop fifo frame "
4337 "compressed multicast|bytes packets errs "
4338 "drop fifo colls carrier compressed\n");
4340 dev_seq_printf_stats(seq, v);
4344 static struct softnet_data *softnet_get_online(loff_t *pos)
4346 struct softnet_data *sd = NULL;
4348 while (*pos < nr_cpu_ids)
4349 if (cpu_online(*pos)) {
4350 sd = &per_cpu(softnet_data, *pos);
4357 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4359 return softnet_get_online(pos);
4362 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4365 return softnet_get_online(pos);
4368 static void softnet_seq_stop(struct seq_file *seq, void *v)
4372 static int softnet_seq_show(struct seq_file *seq, void *v)
4374 struct softnet_data *sd = v;
4376 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4377 sd->processed, sd->dropped, sd->time_squeeze, 0,
4378 0, 0, 0, 0, /* was fastroute */
4379 sd->cpu_collision, sd->received_rps);
4383 static const struct seq_operations dev_seq_ops = {
4384 .start = dev_seq_start,
4385 .next = dev_seq_next,
4386 .stop = dev_seq_stop,
4387 .show = dev_seq_show,
4390 static int dev_seq_open(struct inode *inode, struct file *file)
4392 return seq_open_net(inode, file, &dev_seq_ops,
4393 sizeof(struct seq_net_private));
4396 static const struct file_operations dev_seq_fops = {
4397 .owner = THIS_MODULE,
4398 .open = dev_seq_open,
4400 .llseek = seq_lseek,
4401 .release = seq_release_net,
4404 static const struct seq_operations softnet_seq_ops = {
4405 .start = softnet_seq_start,
4406 .next = softnet_seq_next,
4407 .stop = softnet_seq_stop,
4408 .show = softnet_seq_show,
4411 static int softnet_seq_open(struct inode *inode, struct file *file)
4413 return seq_open(file, &softnet_seq_ops);
4416 static const struct file_operations softnet_seq_fops = {
4417 .owner = THIS_MODULE,
4418 .open = softnet_seq_open,
4420 .llseek = seq_lseek,
4421 .release = seq_release,
4424 static void *ptype_get_idx(loff_t pos)
4426 struct packet_type *pt = NULL;
4430 list_for_each_entry_rcu(pt, &ptype_all, list) {
4436 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4437 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4446 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4450 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4453 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4455 struct packet_type *pt;
4456 struct list_head *nxt;
4460 if (v == SEQ_START_TOKEN)
4461 return ptype_get_idx(0);
4464 nxt = pt->list.next;
4465 if (pt->type == htons(ETH_P_ALL)) {
4466 if (nxt != &ptype_all)
4469 nxt = ptype_base[0].next;
4471 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4473 while (nxt == &ptype_base[hash]) {
4474 if (++hash >= PTYPE_HASH_SIZE)
4476 nxt = ptype_base[hash].next;
4479 return list_entry(nxt, struct packet_type, list);
4482 static void ptype_seq_stop(struct seq_file *seq, void *v)
4488 static int ptype_seq_show(struct seq_file *seq, void *v)
4490 struct packet_type *pt = v;
4492 if (v == SEQ_START_TOKEN)
4493 seq_puts(seq, "Type Device Function\n");
4494 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4495 if (pt->type == htons(ETH_P_ALL))
4496 seq_puts(seq, "ALL ");
4498 seq_printf(seq, "%04x", ntohs(pt->type));
4500 seq_printf(seq, " %-8s %pF\n",
4501 pt->dev ? pt->dev->name : "", pt->func);
4507 static const struct seq_operations ptype_seq_ops = {
4508 .start = ptype_seq_start,
4509 .next = ptype_seq_next,
4510 .stop = ptype_seq_stop,
4511 .show = ptype_seq_show,
4514 static int ptype_seq_open(struct inode *inode, struct file *file)
4516 return seq_open_net(inode, file, &ptype_seq_ops,
4517 sizeof(struct seq_net_private));
4520 static const struct file_operations ptype_seq_fops = {
4521 .owner = THIS_MODULE,
4522 .open = ptype_seq_open,
4524 .llseek = seq_lseek,
4525 .release = seq_release_net,
4529 static int __net_init dev_proc_net_init(struct net *net)
4533 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4535 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4537 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4540 if (wext_proc_init(net))
4546 proc_net_remove(net, "ptype");
4548 proc_net_remove(net, "softnet_stat");
4550 proc_net_remove(net, "dev");
4554 static void __net_exit dev_proc_net_exit(struct net *net)
4556 wext_proc_exit(net);
4558 proc_net_remove(net, "ptype");
4559 proc_net_remove(net, "softnet_stat");
4560 proc_net_remove(net, "dev");
4563 static struct pernet_operations __net_initdata dev_proc_ops = {
4564 .init = dev_proc_net_init,
4565 .exit = dev_proc_net_exit,
4568 static int __init dev_proc_init(void)
4570 return register_pernet_subsys(&dev_proc_ops);
4573 #define dev_proc_init() 0
4574 #endif /* CONFIG_PROC_FS */
4578 * netdev_set_master - set up master pointer
4579 * @slave: slave device
4580 * @master: new master device
4582 * Changes the master device of the slave. Pass %NULL to break the
4583 * bonding. The caller must hold the RTNL semaphore. On a failure
4584 * a negative errno code is returned. On success the reference counts
4585 * are adjusted and the function returns zero.
4587 int netdev_set_master(struct net_device *slave, struct net_device *master)
4589 struct net_device *old = slave->master;
4599 slave->master = master;
4605 EXPORT_SYMBOL(netdev_set_master);
4608 * netdev_set_bond_master - set up bonding master/slave pair
4609 * @slave: slave device
4610 * @master: new master device
4612 * Changes the master device of the slave. Pass %NULL to break the
4613 * bonding. The caller must hold the RTNL semaphore. On a failure
4614 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4615 * to the routing socket and the function returns zero.
4617 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4623 err = netdev_set_master(slave, master);
4627 slave->flags |= IFF_SLAVE;
4629 slave->flags &= ~IFF_SLAVE;
4631 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4634 EXPORT_SYMBOL(netdev_set_bond_master);
4636 static void dev_change_rx_flags(struct net_device *dev, int flags)
4638 const struct net_device_ops *ops = dev->netdev_ops;
4640 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4641 ops->ndo_change_rx_flags(dev, flags);
4644 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4646 unsigned int old_flags = dev->flags;
4652 dev->flags |= IFF_PROMISC;
4653 dev->promiscuity += inc;
4654 if (dev->promiscuity == 0) {
4657 * If inc causes overflow, untouch promisc and return error.
4660 dev->flags &= ~IFF_PROMISC;
4662 dev->promiscuity -= inc;
4663 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4668 if (dev->flags != old_flags) {
4669 pr_info("device %s %s promiscuous mode\n",
4671 dev->flags & IFF_PROMISC ? "entered" : "left");
4672 if (audit_enabled) {
4673 current_uid_gid(&uid, &gid);
4674 audit_log(current->audit_context, GFP_ATOMIC,
4675 AUDIT_ANOM_PROMISCUOUS,
4676 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4677 dev->name, (dev->flags & IFF_PROMISC),
4678 (old_flags & IFF_PROMISC),
4679 from_kuid(&init_user_ns, audit_get_loginuid(current)),
4680 from_kuid(&init_user_ns, uid),
4681 from_kgid(&init_user_ns, gid),
4682 audit_get_sessionid(current));
4685 dev_change_rx_flags(dev, IFF_PROMISC);
4691 * dev_set_promiscuity - update promiscuity count on a device
4695 * Add or remove promiscuity from a device. While the count in the device
4696 * remains above zero the interface remains promiscuous. Once it hits zero
4697 * the device reverts back to normal filtering operation. A negative inc
4698 * value is used to drop promiscuity on the device.
4699 * Return 0 if successful or a negative errno code on error.
4701 int dev_set_promiscuity(struct net_device *dev, int inc)
4703 unsigned int old_flags = dev->flags;
4706 err = __dev_set_promiscuity(dev, inc);
4709 if (dev->flags != old_flags)
4710 dev_set_rx_mode(dev);
4713 EXPORT_SYMBOL(dev_set_promiscuity);
4716 * dev_set_allmulti - update allmulti count on a device
4720 * Add or remove reception of all multicast frames to a device. While the
4721 * count in the device remains above zero the interface remains listening
4722 * to all interfaces. Once it hits zero the device reverts back to normal
4723 * filtering operation. A negative @inc value is used to drop the counter
4724 * when releasing a resource needing all multicasts.
4725 * Return 0 if successful or a negative errno code on error.
4728 int dev_set_allmulti(struct net_device *dev, int inc)
4730 unsigned int old_flags = dev->flags;
4734 dev->flags |= IFF_ALLMULTI;
4735 dev->allmulti += inc;
4736 if (dev->allmulti == 0) {
4739 * If inc causes overflow, untouch allmulti and return error.
4742 dev->flags &= ~IFF_ALLMULTI;
4744 dev->allmulti -= inc;
4745 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4750 if (dev->flags ^ old_flags) {
4751 dev_change_rx_flags(dev, IFF_ALLMULTI);
4752 dev_set_rx_mode(dev);
4756 EXPORT_SYMBOL(dev_set_allmulti);
4759 * Upload unicast and multicast address lists to device and
4760 * configure RX filtering. When the device doesn't support unicast
4761 * filtering it is put in promiscuous mode while unicast addresses
4764 void __dev_set_rx_mode(struct net_device *dev)
4766 const struct net_device_ops *ops = dev->netdev_ops;
4768 /* dev_open will call this function so the list will stay sane. */
4769 if (!(dev->flags&IFF_UP))
4772 if (!netif_device_present(dev))
4775 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4776 /* Unicast addresses changes may only happen under the rtnl,
4777 * therefore calling __dev_set_promiscuity here is safe.
4779 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4780 __dev_set_promiscuity(dev, 1);
4781 dev->uc_promisc = true;
4782 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4783 __dev_set_promiscuity(dev, -1);
4784 dev->uc_promisc = false;
4788 if (ops->ndo_set_rx_mode)
4789 ops->ndo_set_rx_mode(dev);
4792 void dev_set_rx_mode(struct net_device *dev)
4794 netif_addr_lock_bh(dev);
4795 __dev_set_rx_mode(dev);
4796 netif_addr_unlock_bh(dev);
4800 * dev_get_flags - get flags reported to userspace
4803 * Get the combination of flag bits exported through APIs to userspace.
4805 unsigned int dev_get_flags(const struct net_device *dev)
4809 flags = (dev->flags & ~(IFF_PROMISC |
4814 (dev->gflags & (IFF_PROMISC |
4817 if (netif_running(dev)) {
4818 if (netif_oper_up(dev))
4819 flags |= IFF_RUNNING;
4820 if (netif_carrier_ok(dev))
4821 flags |= IFF_LOWER_UP;
4822 if (netif_dormant(dev))
4823 flags |= IFF_DORMANT;
4828 EXPORT_SYMBOL(dev_get_flags);
4830 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4832 unsigned int old_flags = dev->flags;
4838 * Set the flags on our device.
4841 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4842 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4844 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4848 * Load in the correct multicast list now the flags have changed.
4851 if ((old_flags ^ flags) & IFF_MULTICAST)
4852 dev_change_rx_flags(dev, IFF_MULTICAST);
4854 dev_set_rx_mode(dev);
4857 * Have we downed the interface. We handle IFF_UP ourselves
4858 * according to user attempts to set it, rather than blindly
4863 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4864 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4867 dev_set_rx_mode(dev);
4870 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4871 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4873 dev->gflags ^= IFF_PROMISC;
4874 dev_set_promiscuity(dev, inc);
4877 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4878 is important. Some (broken) drivers set IFF_PROMISC, when
4879 IFF_ALLMULTI is requested not asking us and not reporting.
4881 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4882 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4884 dev->gflags ^= IFF_ALLMULTI;
4885 dev_set_allmulti(dev, inc);
4891 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4893 unsigned int changes = dev->flags ^ old_flags;
4895 if (changes & IFF_UP) {
4896 if (dev->flags & IFF_UP)
4897 call_netdevice_notifiers(NETDEV_UP, dev);
4899 call_netdevice_notifiers(NETDEV_DOWN, dev);
4902 if (dev->flags & IFF_UP &&
4903 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4904 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4908 * dev_change_flags - change device settings
4910 * @flags: device state flags
4912 * Change settings on device based state flags. The flags are
4913 * in the userspace exported format.
4915 int dev_change_flags(struct net_device *dev, unsigned int flags)
4918 unsigned int changes, old_flags = dev->flags;
4920 ret = __dev_change_flags(dev, flags);
4924 changes = old_flags ^ dev->flags;
4926 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4928 __dev_notify_flags(dev, old_flags);
4931 EXPORT_SYMBOL(dev_change_flags);
4934 * dev_set_mtu - Change maximum transfer unit
4936 * @new_mtu: new transfer unit
4938 * Change the maximum transfer size of the network device.
4940 int dev_set_mtu(struct net_device *dev, int new_mtu)
4942 const struct net_device_ops *ops = dev->netdev_ops;
4945 if (new_mtu == dev->mtu)
4948 /* MTU must be positive. */
4952 if (!netif_device_present(dev))
4956 if (ops->ndo_change_mtu)
4957 err = ops->ndo_change_mtu(dev, new_mtu);
4961 if (!err && dev->flags & IFF_UP)
4962 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4965 EXPORT_SYMBOL(dev_set_mtu);
4968 * dev_set_group - Change group this device belongs to
4970 * @new_group: group this device should belong to
4972 void dev_set_group(struct net_device *dev, int new_group)
4974 dev->group = new_group;
4976 EXPORT_SYMBOL(dev_set_group);
4979 * dev_set_mac_address - Change Media Access Control Address
4983 * Change the hardware (MAC) address of the device
4985 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4987 const struct net_device_ops *ops = dev->netdev_ops;
4990 if (!ops->ndo_set_mac_address)
4992 if (sa->sa_family != dev->type)
4994 if (!netif_device_present(dev))
4996 err = ops->ndo_set_mac_address(dev, sa);
4998 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4999 add_device_randomness(dev->dev_addr, dev->addr_len);
5002 EXPORT_SYMBOL(dev_set_mac_address);
5005 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
5007 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
5010 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
5016 case SIOCGIFFLAGS: /* Get interface flags */
5017 ifr->ifr_flags = (short) dev_get_flags(dev);
5020 case SIOCGIFMETRIC: /* Get the metric on the interface
5021 (currently unused) */
5022 ifr->ifr_metric = 0;
5025 case SIOCGIFMTU: /* Get the MTU of a device */
5026 ifr->ifr_mtu = dev->mtu;
5031 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
5033 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
5034 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
5035 ifr->ifr_hwaddr.sa_family = dev->type;
5043 ifr->ifr_map.mem_start = dev->mem_start;
5044 ifr->ifr_map.mem_end = dev->mem_end;
5045 ifr->ifr_map.base_addr = dev->base_addr;
5046 ifr->ifr_map.irq = dev->irq;
5047 ifr->ifr_map.dma = dev->dma;
5048 ifr->ifr_map.port = dev->if_port;
5052 ifr->ifr_ifindex = dev->ifindex;
5056 ifr->ifr_qlen = dev->tx_queue_len;
5060 /* dev_ioctl() should ensure this case
5072 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
5074 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
5077 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
5078 const struct net_device_ops *ops;
5083 ops = dev->netdev_ops;
5086 case SIOCSIFFLAGS: /* Set interface flags */
5087 return dev_change_flags(dev, ifr->ifr_flags);
5089 case SIOCSIFMETRIC: /* Set the metric on the interface
5090 (currently unused) */
5093 case SIOCSIFMTU: /* Set the MTU of a device */
5094 return dev_set_mtu(dev, ifr->ifr_mtu);
5097 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
5099 case SIOCSIFHWBROADCAST:
5100 if (ifr->ifr_hwaddr.sa_family != dev->type)
5102 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
5103 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
5104 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5108 if (ops->ndo_set_config) {
5109 if (!netif_device_present(dev))
5111 return ops->ndo_set_config(dev, &ifr->ifr_map);
5116 if (!ops->ndo_set_rx_mode ||
5117 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
5119 if (!netif_device_present(dev))
5121 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
5124 if (!ops->ndo_set_rx_mode ||
5125 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
5127 if (!netif_device_present(dev))
5129 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
5132 if (ifr->ifr_qlen < 0)
5134 dev->tx_queue_len = ifr->ifr_qlen;
5138 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
5139 return dev_change_name(dev, ifr->ifr_newname);
5142 err = net_hwtstamp_validate(ifr);
5148 * Unknown or private ioctl
5151 if ((cmd >= SIOCDEVPRIVATE &&
5152 cmd <= SIOCDEVPRIVATE + 15) ||
5153 cmd == SIOCBONDENSLAVE ||
5154 cmd == SIOCBONDRELEASE ||
5155 cmd == SIOCBONDSETHWADDR ||
5156 cmd == SIOCBONDSLAVEINFOQUERY ||
5157 cmd == SIOCBONDINFOQUERY ||
5158 cmd == SIOCBONDCHANGEACTIVE ||
5159 cmd == SIOCGMIIPHY ||
5160 cmd == SIOCGMIIREG ||
5161 cmd == SIOCSMIIREG ||
5162 cmd == SIOCBRADDIF ||
5163 cmd == SIOCBRDELIF ||
5164 cmd == SIOCSHWTSTAMP ||
5165 cmd == SIOCWANDEV) {
5167 if (ops->ndo_do_ioctl) {
5168 if (netif_device_present(dev))
5169 err = ops->ndo_do_ioctl(dev, ifr, cmd);
5181 * This function handles all "interface"-type I/O control requests. The actual
5182 * 'doing' part of this is dev_ifsioc above.
5186 * dev_ioctl - network device ioctl
5187 * @net: the applicable net namespace
5188 * @cmd: command to issue
5189 * @arg: pointer to a struct ifreq in user space
5191 * Issue ioctl functions to devices. This is normally called by the
5192 * user space syscall interfaces but can sometimes be useful for
5193 * other purposes. The return value is the return from the syscall if
5194 * positive or a negative errno code on error.
5197 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
5203 /* One special case: SIOCGIFCONF takes ifconf argument
5204 and requires shared lock, because it sleeps writing
5208 if (cmd == SIOCGIFCONF) {
5210 ret = dev_ifconf(net, (char __user *) arg);
5214 if (cmd == SIOCGIFNAME)
5215 return dev_ifname(net, (struct ifreq __user *)arg);
5217 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
5220 ifr.ifr_name[IFNAMSIZ-1] = 0;
5222 colon = strchr(ifr.ifr_name, ':');
5227 * See which interface the caller is talking about.
5232 * These ioctl calls:
5233 * - can be done by all.
5234 * - atomic and do not require locking.
5245 dev_load(net, ifr.ifr_name);
5247 ret = dev_ifsioc_locked(net, &ifr, cmd);
5252 if (copy_to_user(arg, &ifr,
5253 sizeof(struct ifreq)))
5259 dev_load(net, ifr.ifr_name);
5261 ret = dev_ethtool(net, &ifr);
5266 if (copy_to_user(arg, &ifr,
5267 sizeof(struct ifreq)))
5273 * These ioctl calls:
5274 * - require superuser power.
5275 * - require strict serialization.
5281 if (!capable(CAP_NET_ADMIN))
5283 dev_load(net, ifr.ifr_name);
5285 ret = dev_ifsioc(net, &ifr, cmd);
5290 if (copy_to_user(arg, &ifr,
5291 sizeof(struct ifreq)))
5297 * These ioctl calls:
5298 * - require superuser power.
5299 * - require strict serialization.
5300 * - do not return a value
5310 case SIOCSIFHWBROADCAST:
5313 case SIOCBONDENSLAVE:
5314 case SIOCBONDRELEASE:
5315 case SIOCBONDSETHWADDR:
5316 case SIOCBONDCHANGEACTIVE:
5320 if (!capable(CAP_NET_ADMIN))
5323 case SIOCBONDSLAVEINFOQUERY:
5324 case SIOCBONDINFOQUERY:
5325 dev_load(net, ifr.ifr_name);
5327 ret = dev_ifsioc(net, &ifr, cmd);
5332 /* Get the per device memory space. We can add this but
5333 * currently do not support it */
5335 /* Set the per device memory buffer space.
5336 * Not applicable in our case */
5341 * Unknown or private ioctl.
5344 if (cmd == SIOCWANDEV ||
5345 (cmd >= SIOCDEVPRIVATE &&
5346 cmd <= SIOCDEVPRIVATE + 15)) {
5347 dev_load(net, ifr.ifr_name);
5349 ret = dev_ifsioc(net, &ifr, cmd);
5351 if (!ret && copy_to_user(arg, &ifr,
5352 sizeof(struct ifreq)))
5356 /* Take care of Wireless Extensions */
5357 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5358 return wext_handle_ioctl(net, &ifr, cmd, arg);
5365 * dev_new_index - allocate an ifindex
5366 * @net: the applicable net namespace
5368 * Returns a suitable unique value for a new device interface
5369 * number. The caller must hold the rtnl semaphore or the
5370 * dev_base_lock to be sure it remains unique.
5372 static int dev_new_index(struct net *net)
5374 int ifindex = net->ifindex;
5378 if (!__dev_get_by_index(net, ifindex))
5379 return net->ifindex = ifindex;
5383 /* Delayed registration/unregisteration */
5384 static LIST_HEAD(net_todo_list);
5386 static void net_set_todo(struct net_device *dev)
5388 list_add_tail(&dev->todo_list, &net_todo_list);
5391 static void rollback_registered_many(struct list_head *head)
5393 struct net_device *dev, *tmp;
5395 BUG_ON(dev_boot_phase);
5398 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5399 /* Some devices call without registering
5400 * for initialization unwind. Remove those
5401 * devices and proceed with the remaining.
5403 if (dev->reg_state == NETREG_UNINITIALIZED) {
5404 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5408 list_del(&dev->unreg_list);
5411 dev->dismantle = true;
5412 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5415 /* If device is running, close it first. */
5416 dev_close_many(head);
5418 list_for_each_entry(dev, head, unreg_list) {
5419 /* And unlink it from device chain. */
5420 unlist_netdevice(dev);
5422 dev->reg_state = NETREG_UNREGISTERING;
5427 list_for_each_entry(dev, head, unreg_list) {
5428 /* Shutdown queueing discipline. */
5432 /* Notify protocols, that we are about to destroy
5433 this device. They should clean all the things.
5435 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5437 if (!dev->rtnl_link_ops ||
5438 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5439 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5442 * Flush the unicast and multicast chains
5447 if (dev->netdev_ops->ndo_uninit)
5448 dev->netdev_ops->ndo_uninit(dev);
5450 /* Notifier chain MUST detach us from master device. */
5451 WARN_ON(dev->master);
5453 /* Remove entries from kobject tree */
5454 netdev_unregister_kobject(dev);
5459 list_for_each_entry(dev, head, unreg_list)
5463 static void rollback_registered(struct net_device *dev)
5467 list_add(&dev->unreg_list, &single);
5468 rollback_registered_many(&single);
5472 static netdev_features_t netdev_fix_features(struct net_device *dev,
5473 netdev_features_t features)
5475 /* Fix illegal checksum combinations */
5476 if ((features & NETIF_F_HW_CSUM) &&
5477 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5478 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5479 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5482 /* Fix illegal SG+CSUM combinations. */
5483 if ((features & NETIF_F_SG) &&
5484 !(features & NETIF_F_ALL_CSUM)) {
5486 "Dropping NETIF_F_SG since no checksum feature.\n");
5487 features &= ~NETIF_F_SG;
5490 /* TSO requires that SG is present as well. */
5491 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5492 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5493 features &= ~NETIF_F_ALL_TSO;
5496 /* TSO ECN requires that TSO is present as well. */
5497 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5498 features &= ~NETIF_F_TSO_ECN;
5500 /* Software GSO depends on SG. */
5501 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5502 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5503 features &= ~NETIF_F_GSO;
5506 /* UFO needs SG and checksumming */
5507 if (features & NETIF_F_UFO) {
5508 /* maybe split UFO into V4 and V6? */
5509 if (!((features & NETIF_F_GEN_CSUM) ||
5510 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5511 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5513 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5514 features &= ~NETIF_F_UFO;
5517 if (!(features & NETIF_F_SG)) {
5519 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5520 features &= ~NETIF_F_UFO;
5527 int __netdev_update_features(struct net_device *dev)
5529 netdev_features_t features;
5534 features = netdev_get_wanted_features(dev);
5536 if (dev->netdev_ops->ndo_fix_features)
5537 features = dev->netdev_ops->ndo_fix_features(dev, features);
5539 /* driver might be less strict about feature dependencies */
5540 features = netdev_fix_features(dev, features);
5542 if (dev->features == features)
5545 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5546 &dev->features, &features);
5548 if (dev->netdev_ops->ndo_set_features)
5549 err = dev->netdev_ops->ndo_set_features(dev, features);
5551 if (unlikely(err < 0)) {
5553 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5554 err, &features, &dev->features);
5559 dev->features = features;
5565 * netdev_update_features - recalculate device features
5566 * @dev: the device to check
5568 * Recalculate dev->features set and send notifications if it
5569 * has changed. Should be called after driver or hardware dependent
5570 * conditions might have changed that influence the features.
5572 void netdev_update_features(struct net_device *dev)
5574 if (__netdev_update_features(dev))
5575 netdev_features_change(dev);
5577 EXPORT_SYMBOL(netdev_update_features);
5580 * netdev_change_features - recalculate device features
5581 * @dev: the device to check
5583 * Recalculate dev->features set and send notifications even
5584 * if they have not changed. Should be called instead of
5585 * netdev_update_features() if also dev->vlan_features might
5586 * have changed to allow the changes to be propagated to stacked
5589 void netdev_change_features(struct net_device *dev)
5591 __netdev_update_features(dev);
5592 netdev_features_change(dev);
5594 EXPORT_SYMBOL(netdev_change_features);
5597 * netif_stacked_transfer_operstate - transfer operstate
5598 * @rootdev: the root or lower level device to transfer state from
5599 * @dev: the device to transfer operstate to
5601 * Transfer operational state from root to device. This is normally
5602 * called when a stacking relationship exists between the root
5603 * device and the device(a leaf device).
5605 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5606 struct net_device *dev)
5608 if (rootdev->operstate == IF_OPER_DORMANT)
5609 netif_dormant_on(dev);
5611 netif_dormant_off(dev);
5613 if (netif_carrier_ok(rootdev)) {
5614 if (!netif_carrier_ok(dev))
5615 netif_carrier_on(dev);
5617 if (netif_carrier_ok(dev))
5618 netif_carrier_off(dev);
5621 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5624 static int netif_alloc_rx_queues(struct net_device *dev)
5626 unsigned int i, count = dev->num_rx_queues;
5627 struct netdev_rx_queue *rx;
5631 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5633 pr_err("netdev: Unable to allocate %u rx queues\n", count);
5638 for (i = 0; i < count; i++)
5644 static void netdev_init_one_queue(struct net_device *dev,
5645 struct netdev_queue *queue, void *_unused)
5647 /* Initialize queue lock */
5648 spin_lock_init(&queue->_xmit_lock);
5649 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5650 queue->xmit_lock_owner = -1;
5651 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5654 dql_init(&queue->dql, HZ);
5658 static int netif_alloc_netdev_queues(struct net_device *dev)
5660 unsigned int count = dev->num_tx_queues;
5661 struct netdev_queue *tx;
5665 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5667 pr_err("netdev: Unable to allocate %u tx queues\n", count);
5672 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5673 spin_lock_init(&dev->tx_global_lock);
5679 * register_netdevice - register a network device
5680 * @dev: device to register
5682 * Take a completed network device structure and add it to the kernel
5683 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5684 * chain. 0 is returned on success. A negative errno code is returned
5685 * on a failure to set up the device, or if the name is a duplicate.
5687 * Callers must hold the rtnl semaphore. You may want
5688 * register_netdev() instead of this.
5691 * The locking appears insufficient to guarantee two parallel registers
5692 * will not get the same name.
5695 int register_netdevice(struct net_device *dev)
5698 struct net *net = dev_net(dev);
5700 BUG_ON(dev_boot_phase);
5705 /* When net_device's are persistent, this will be fatal. */
5706 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5709 spin_lock_init(&dev->addr_list_lock);
5710 netdev_set_addr_lockdep_class(dev);
5714 ret = dev_get_valid_name(net, dev, dev->name);
5718 /* Init, if this function is available */
5719 if (dev->netdev_ops->ndo_init) {
5720 ret = dev->netdev_ops->ndo_init(dev);
5730 dev->ifindex = dev_new_index(net);
5731 else if (__dev_get_by_index(net, dev->ifindex))
5734 if (dev->iflink == -1)
5735 dev->iflink = dev->ifindex;
5737 /* Transfer changeable features to wanted_features and enable
5738 * software offloads (GSO and GRO).
5740 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5741 dev->features |= NETIF_F_SOFT_FEATURES;
5742 dev->wanted_features = dev->features & dev->hw_features;
5744 /* Turn on no cache copy if HW is doing checksum */
5745 if (!(dev->flags & IFF_LOOPBACK)) {
5746 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5747 if (dev->features & NETIF_F_ALL_CSUM) {
5748 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5749 dev->features |= NETIF_F_NOCACHE_COPY;
5753 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5755 dev->vlan_features |= NETIF_F_HIGHDMA;
5757 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5758 ret = notifier_to_errno(ret);
5762 ret = netdev_register_kobject(dev);
5765 dev->reg_state = NETREG_REGISTERED;
5767 __netdev_update_features(dev);
5770 * Default initial state at registry is that the
5771 * device is present.
5774 set_bit(__LINK_STATE_PRESENT, &dev->state);
5776 linkwatch_init_dev(dev);
5778 dev_init_scheduler(dev);
5780 list_netdevice(dev);
5781 add_device_randomness(dev->dev_addr, dev->addr_len);
5783 /* Notify protocols, that a new device appeared. */
5784 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5785 ret = notifier_to_errno(ret);
5787 rollback_registered(dev);
5788 dev->reg_state = NETREG_UNREGISTERED;
5791 * Prevent userspace races by waiting until the network
5792 * device is fully setup before sending notifications.
5794 if (!dev->rtnl_link_ops ||
5795 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5796 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5802 if (dev->netdev_ops->ndo_uninit)
5803 dev->netdev_ops->ndo_uninit(dev);
5806 EXPORT_SYMBOL(register_netdevice);
5809 * init_dummy_netdev - init a dummy network device for NAPI
5810 * @dev: device to init
5812 * This takes a network device structure and initialize the minimum
5813 * amount of fields so it can be used to schedule NAPI polls without
5814 * registering a full blown interface. This is to be used by drivers
5815 * that need to tie several hardware interfaces to a single NAPI
5816 * poll scheduler due to HW limitations.
5818 int init_dummy_netdev(struct net_device *dev)
5820 /* Clear everything. Note we don't initialize spinlocks
5821 * are they aren't supposed to be taken by any of the
5822 * NAPI code and this dummy netdev is supposed to be
5823 * only ever used for NAPI polls
5825 memset(dev, 0, sizeof(struct net_device));
5827 /* make sure we BUG if trying to hit standard
5828 * register/unregister code path
5830 dev->reg_state = NETREG_DUMMY;
5832 /* NAPI wants this */
5833 INIT_LIST_HEAD(&dev->napi_list);
5835 /* a dummy interface is started by default */
5836 set_bit(__LINK_STATE_PRESENT, &dev->state);
5837 set_bit(__LINK_STATE_START, &dev->state);
5839 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5840 * because users of this 'device' dont need to change
5846 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5850 * register_netdev - register a network device
5851 * @dev: device to register
5853 * Take a completed network device structure and add it to the kernel
5854 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5855 * chain. 0 is returned on success. A negative errno code is returned
5856 * on a failure to set up the device, or if the name is a duplicate.
5858 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5859 * and expands the device name if you passed a format string to
5862 int register_netdev(struct net_device *dev)
5867 err = register_netdevice(dev);
5871 EXPORT_SYMBOL(register_netdev);
5873 int netdev_refcnt_read(const struct net_device *dev)
5877 for_each_possible_cpu(i)
5878 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5881 EXPORT_SYMBOL(netdev_refcnt_read);
5884 * netdev_wait_allrefs - wait until all references are gone.
5885 * @dev: target net_device
5887 * This is called when unregistering network devices.
5889 * Any protocol or device that holds a reference should register
5890 * for netdevice notification, and cleanup and put back the
5891 * reference if they receive an UNREGISTER event.
5892 * We can get stuck here if buggy protocols don't correctly
5895 static void netdev_wait_allrefs(struct net_device *dev)
5897 unsigned long rebroadcast_time, warning_time;
5900 linkwatch_forget_dev(dev);
5902 rebroadcast_time = warning_time = jiffies;
5903 refcnt = netdev_refcnt_read(dev);
5905 while (refcnt != 0) {
5906 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5909 /* Rebroadcast unregister notification */
5910 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5916 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
5917 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5919 /* We must not have linkwatch events
5920 * pending on unregister. If this
5921 * happens, we simply run the queue
5922 * unscheduled, resulting in a noop
5925 linkwatch_run_queue();
5930 rebroadcast_time = jiffies;
5935 refcnt = netdev_refcnt_read(dev);
5937 if (time_after(jiffies, warning_time + 10 * HZ)) {
5938 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5940 warning_time = jiffies;
5949 * register_netdevice(x1);
5950 * register_netdevice(x2);
5952 * unregister_netdevice(y1);
5953 * unregister_netdevice(y2);
5959 * We are invoked by rtnl_unlock().
5960 * This allows us to deal with problems:
5961 * 1) We can delete sysfs objects which invoke hotplug
5962 * without deadlocking with linkwatch via keventd.
5963 * 2) Since we run with the RTNL semaphore not held, we can sleep
5964 * safely in order to wait for the netdev refcnt to drop to zero.
5966 * We must not return until all unregister events added during
5967 * the interval the lock was held have been completed.
5969 void netdev_run_todo(void)
5971 struct list_head list;
5973 /* Snapshot list, allow later requests */
5974 list_replace_init(&net_todo_list, &list);
5979 /* Wait for rcu callbacks to finish before next phase */
5980 if (!list_empty(&list))
5983 while (!list_empty(&list)) {
5984 struct net_device *dev
5985 = list_first_entry(&list, struct net_device, todo_list);
5986 list_del(&dev->todo_list);
5989 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
5992 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5993 pr_err("network todo '%s' but state %d\n",
5994 dev->name, dev->reg_state);
5999 dev->reg_state = NETREG_UNREGISTERED;
6001 on_each_cpu(flush_backlog, dev, 1);
6003 netdev_wait_allrefs(dev);
6006 BUG_ON(netdev_refcnt_read(dev));
6007 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6008 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6009 WARN_ON(dev->dn_ptr);
6011 if (dev->destructor)
6012 dev->destructor(dev);
6014 /* Free network device */
6015 kobject_put(&dev->dev.kobj);
6019 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6020 * fields in the same order, with only the type differing.
6022 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6023 const struct net_device_stats *netdev_stats)
6025 #if BITS_PER_LONG == 64
6026 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6027 memcpy(stats64, netdev_stats, sizeof(*stats64));
6029 size_t i, n = sizeof(*stats64) / sizeof(u64);
6030 const unsigned long *src = (const unsigned long *)netdev_stats;
6031 u64 *dst = (u64 *)stats64;
6033 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6034 sizeof(*stats64) / sizeof(u64));
6035 for (i = 0; i < n; i++)
6039 EXPORT_SYMBOL(netdev_stats_to_stats64);
6042 * dev_get_stats - get network device statistics
6043 * @dev: device to get statistics from
6044 * @storage: place to store stats
6046 * Get network statistics from device. Return @storage.
6047 * The device driver may provide its own method by setting
6048 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6049 * otherwise the internal statistics structure is used.
6051 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6052 struct rtnl_link_stats64 *storage)
6054 const struct net_device_ops *ops = dev->netdev_ops;
6056 if (ops->ndo_get_stats64) {
6057 memset(storage, 0, sizeof(*storage));
6058 ops->ndo_get_stats64(dev, storage);
6059 } else if (ops->ndo_get_stats) {
6060 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6062 netdev_stats_to_stats64(storage, &dev->stats);
6064 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6067 EXPORT_SYMBOL(dev_get_stats);
6069 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
6071 struct netdev_queue *queue = dev_ingress_queue(dev);
6073 #ifdef CONFIG_NET_CLS_ACT
6076 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
6079 netdev_init_one_queue(dev, queue, NULL);
6080 queue->qdisc = &noop_qdisc;
6081 queue->qdisc_sleeping = &noop_qdisc;
6082 rcu_assign_pointer(dev->ingress_queue, queue);
6087 static const struct ethtool_ops default_ethtool_ops;
6090 * alloc_netdev_mqs - allocate network device
6091 * @sizeof_priv: size of private data to allocate space for
6092 * @name: device name format string
6093 * @setup: callback to initialize device
6094 * @txqs: the number of TX subqueues to allocate
6095 * @rxqs: the number of RX subqueues to allocate
6097 * Allocates a struct net_device with private data area for driver use
6098 * and performs basic initialization. Also allocates subquue structs
6099 * for each queue on the device.
6101 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
6102 void (*setup)(struct net_device *),
6103 unsigned int txqs, unsigned int rxqs)
6105 struct net_device *dev;
6107 struct net_device *p;
6109 BUG_ON(strlen(name) >= sizeof(dev->name));
6112 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6118 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6123 alloc_size = sizeof(struct net_device);
6125 /* ensure 32-byte alignment of private area */
6126 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6127 alloc_size += sizeof_priv;
6129 /* ensure 32-byte alignment of whole construct */
6130 alloc_size += NETDEV_ALIGN - 1;
6132 p = kzalloc(alloc_size, GFP_KERNEL);
6134 pr_err("alloc_netdev: Unable to allocate device\n");
6138 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6139 dev->padded = (char *)dev - (char *)p;
6141 dev->pcpu_refcnt = alloc_percpu(int);
6142 if (!dev->pcpu_refcnt)
6145 if (dev_addr_init(dev))
6151 dev_net_set(dev, &init_net);
6153 dev->gso_max_size = GSO_MAX_SIZE;
6154 dev->gso_max_segs = GSO_MAX_SEGS;
6156 INIT_LIST_HEAD(&dev->napi_list);
6157 INIT_LIST_HEAD(&dev->unreg_list);
6158 INIT_LIST_HEAD(&dev->link_watch_list);
6159 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6162 dev->num_tx_queues = txqs;
6163 dev->real_num_tx_queues = txqs;
6164 if (netif_alloc_netdev_queues(dev))
6168 dev->num_rx_queues = rxqs;
6169 dev->real_num_rx_queues = rxqs;
6170 if (netif_alloc_rx_queues(dev))
6174 strcpy(dev->name, name);
6175 dev->group = INIT_NETDEV_GROUP;
6176 if (!dev->ethtool_ops)
6177 dev->ethtool_ops = &default_ethtool_ops;
6185 free_percpu(dev->pcpu_refcnt);
6195 EXPORT_SYMBOL(alloc_netdev_mqs);
6198 * free_netdev - free network device
6201 * This function does the last stage of destroying an allocated device
6202 * interface. The reference to the device object is released.
6203 * If this is the last reference then it will be freed.
6205 void free_netdev(struct net_device *dev)
6207 struct napi_struct *p, *n;
6209 release_net(dev_net(dev));
6216 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6218 /* Flush device addresses */
6219 dev_addr_flush(dev);
6221 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6224 free_percpu(dev->pcpu_refcnt);
6225 dev->pcpu_refcnt = NULL;
6227 /* Compatibility with error handling in drivers */
6228 if (dev->reg_state == NETREG_UNINITIALIZED) {
6229 kfree((char *)dev - dev->padded);
6233 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6234 dev->reg_state = NETREG_RELEASED;
6236 /* will free via device release */
6237 put_device(&dev->dev);
6239 EXPORT_SYMBOL(free_netdev);
6242 * synchronize_net - Synchronize with packet receive processing
6244 * Wait for packets currently being received to be done.
6245 * Does not block later packets from starting.
6247 void synchronize_net(void)
6250 if (rtnl_is_locked())
6251 synchronize_rcu_expedited();
6255 EXPORT_SYMBOL(synchronize_net);
6258 * unregister_netdevice_queue - remove device from the kernel
6262 * This function shuts down a device interface and removes it
6263 * from the kernel tables.
6264 * If head not NULL, device is queued to be unregistered later.
6266 * Callers must hold the rtnl semaphore. You may want
6267 * unregister_netdev() instead of this.
6270 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6275 list_move_tail(&dev->unreg_list, head);
6277 rollback_registered(dev);
6278 /* Finish processing unregister after unlock */
6282 EXPORT_SYMBOL(unregister_netdevice_queue);
6285 * unregister_netdevice_many - unregister many devices
6286 * @head: list of devices
6288 void unregister_netdevice_many(struct list_head *head)
6290 struct net_device *dev;
6292 if (!list_empty(head)) {
6293 rollback_registered_many(head);
6294 list_for_each_entry(dev, head, unreg_list)
6298 EXPORT_SYMBOL(unregister_netdevice_many);
6301 * unregister_netdev - remove device from the kernel
6304 * This function shuts down a device interface and removes it
6305 * from the kernel tables.
6307 * This is just a wrapper for unregister_netdevice that takes
6308 * the rtnl semaphore. In general you want to use this and not
6309 * unregister_netdevice.
6311 void unregister_netdev(struct net_device *dev)
6314 unregister_netdevice(dev);
6317 EXPORT_SYMBOL(unregister_netdev);
6320 * dev_change_net_namespace - move device to different nethost namespace
6322 * @net: network namespace
6323 * @pat: If not NULL name pattern to try if the current device name
6324 * is already taken in the destination network namespace.
6326 * This function shuts down a device interface and moves it
6327 * to a new network namespace. On success 0 is returned, on
6328 * a failure a netagive errno code is returned.
6330 * Callers must hold the rtnl semaphore.
6333 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6339 /* Don't allow namespace local devices to be moved. */
6341 if (dev->features & NETIF_F_NETNS_LOCAL)
6344 /* Ensure the device has been registrered */
6345 if (dev->reg_state != NETREG_REGISTERED)
6348 /* Get out if there is nothing todo */
6350 if (net_eq(dev_net(dev), net))
6353 /* Pick the destination device name, and ensure
6354 * we can use it in the destination network namespace.
6357 if (__dev_get_by_name(net, dev->name)) {
6358 /* We get here if we can't use the current device name */
6361 if (dev_get_valid_name(net, dev, pat) < 0)
6366 * And now a mini version of register_netdevice unregister_netdevice.
6369 /* If device is running close it first. */
6372 /* And unlink it from device chain */
6374 unlist_netdevice(dev);
6378 /* Shutdown queueing discipline. */
6381 /* Notify protocols, that we are about to destroy
6382 this device. They should clean all the things.
6384 Note that dev->reg_state stays at NETREG_REGISTERED.
6385 This is wanted because this way 8021q and macvlan know
6386 the device is just moving and can keep their slaves up.
6388 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6390 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6391 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6394 * Flush the unicast and multicast chains
6399 /* Actually switch the network namespace */
6400 dev_net_set(dev, net);
6402 /* If there is an ifindex conflict assign a new one */
6403 if (__dev_get_by_index(net, dev->ifindex)) {
6404 int iflink = (dev->iflink == dev->ifindex);
6405 dev->ifindex = dev_new_index(net);
6407 dev->iflink = dev->ifindex;
6410 /* Fixup kobjects */
6411 err = device_rename(&dev->dev, dev->name);
6414 /* Add the device back in the hashes */
6415 list_netdevice(dev);
6417 /* Notify protocols, that a new device appeared. */
6418 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6421 * Prevent userspace races by waiting until the network
6422 * device is fully setup before sending notifications.
6424 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6431 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6433 static int dev_cpu_callback(struct notifier_block *nfb,
6434 unsigned long action,
6437 struct sk_buff **list_skb;
6438 struct sk_buff *skb;
6439 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6440 struct softnet_data *sd, *oldsd;
6442 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6445 local_irq_disable();
6446 cpu = smp_processor_id();
6447 sd = &per_cpu(softnet_data, cpu);
6448 oldsd = &per_cpu(softnet_data, oldcpu);
6450 /* Find end of our completion_queue. */
6451 list_skb = &sd->completion_queue;
6453 list_skb = &(*list_skb)->next;
6454 /* Append completion queue from offline CPU. */
6455 *list_skb = oldsd->completion_queue;
6456 oldsd->completion_queue = NULL;
6458 /* Append output queue from offline CPU. */
6459 if (oldsd->output_queue) {
6460 *sd->output_queue_tailp = oldsd->output_queue;
6461 sd->output_queue_tailp = oldsd->output_queue_tailp;
6462 oldsd->output_queue = NULL;
6463 oldsd->output_queue_tailp = &oldsd->output_queue;
6465 /* Append NAPI poll list from offline CPU. */
6466 if (!list_empty(&oldsd->poll_list)) {
6467 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6468 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6471 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6474 /* Process offline CPU's input_pkt_queue */
6475 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6477 input_queue_head_incr(oldsd);
6479 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6481 input_queue_head_incr(oldsd);
6489 * netdev_increment_features - increment feature set by one
6490 * @all: current feature set
6491 * @one: new feature set
6492 * @mask: mask feature set
6494 * Computes a new feature set after adding a device with feature set
6495 * @one to the master device with current feature set @all. Will not
6496 * enable anything that is off in @mask. Returns the new feature set.
6498 netdev_features_t netdev_increment_features(netdev_features_t all,
6499 netdev_features_t one, netdev_features_t mask)
6501 if (mask & NETIF_F_GEN_CSUM)
6502 mask |= NETIF_F_ALL_CSUM;
6503 mask |= NETIF_F_VLAN_CHALLENGED;
6505 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6506 all &= one | ~NETIF_F_ALL_FOR_ALL;
6508 /* If one device supports hw checksumming, set for all. */
6509 if (all & NETIF_F_GEN_CSUM)
6510 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6514 EXPORT_SYMBOL(netdev_increment_features);
6516 static struct hlist_head *netdev_create_hash(void)
6519 struct hlist_head *hash;
6521 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6523 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6524 INIT_HLIST_HEAD(&hash[i]);
6529 /* Initialize per network namespace state */
6530 static int __net_init netdev_init(struct net *net)
6532 if (net != &init_net)
6533 INIT_LIST_HEAD(&net->dev_base_head);
6535 net->dev_name_head = netdev_create_hash();
6536 if (net->dev_name_head == NULL)
6539 net->dev_index_head = netdev_create_hash();
6540 if (net->dev_index_head == NULL)
6546 kfree(net->dev_name_head);
6552 * netdev_drivername - network driver for the device
6553 * @dev: network device
6555 * Determine network driver for device.
6557 const char *netdev_drivername(const struct net_device *dev)
6559 const struct device_driver *driver;
6560 const struct device *parent;
6561 const char *empty = "";
6563 parent = dev->dev.parent;
6567 driver = parent->driver;
6568 if (driver && driver->name)
6569 return driver->name;
6573 static int __netdev_printk(const char *level, const struct net_device *dev,
6574 struct va_format *vaf)
6578 if (dev && dev->dev.parent) {
6579 r = dev_printk_emit(level[1] - '0',
6582 dev_driver_string(dev->dev.parent),
6583 dev_name(dev->dev.parent),
6584 netdev_name(dev), vaf);
6586 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6588 r = printk("%s(NULL net_device): %pV", level, vaf);
6594 int netdev_printk(const char *level, const struct net_device *dev,
6595 const char *format, ...)
6597 struct va_format vaf;
6601 va_start(args, format);
6606 r = __netdev_printk(level, dev, &vaf);
6612 EXPORT_SYMBOL(netdev_printk);
6614 #define define_netdev_printk_level(func, level) \
6615 int func(const struct net_device *dev, const char *fmt, ...) \
6618 struct va_format vaf; \
6621 va_start(args, fmt); \
6626 r = __netdev_printk(level, dev, &vaf); \
6632 EXPORT_SYMBOL(func);
6634 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6635 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6636 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6637 define_netdev_printk_level(netdev_err, KERN_ERR);
6638 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6639 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6640 define_netdev_printk_level(netdev_info, KERN_INFO);
6642 static void __net_exit netdev_exit(struct net *net)
6644 kfree(net->dev_name_head);
6645 kfree(net->dev_index_head);
6648 static struct pernet_operations __net_initdata netdev_net_ops = {
6649 .init = netdev_init,
6650 .exit = netdev_exit,
6653 static void __net_exit default_device_exit(struct net *net)
6655 struct net_device *dev, *aux;
6657 * Push all migratable network devices back to the
6658 * initial network namespace
6661 for_each_netdev_safe(net, dev, aux) {
6663 char fb_name[IFNAMSIZ];
6665 /* Ignore unmoveable devices (i.e. loopback) */
6666 if (dev->features & NETIF_F_NETNS_LOCAL)
6669 /* Leave virtual devices for the generic cleanup */
6670 if (dev->rtnl_link_ops)
6673 /* Push remaining network devices to init_net */
6674 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6675 err = dev_change_net_namespace(dev, &init_net, fb_name);
6677 pr_emerg("%s: failed to move %s to init_net: %d\n",
6678 __func__, dev->name, err);
6685 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6687 /* At exit all network devices most be removed from a network
6688 * namespace. Do this in the reverse order of registration.
6689 * Do this across as many network namespaces as possible to
6690 * improve batching efficiency.
6692 struct net_device *dev;
6694 LIST_HEAD(dev_kill_list);
6697 list_for_each_entry(net, net_list, exit_list) {
6698 for_each_netdev_reverse(net, dev) {
6699 if (dev->rtnl_link_ops)
6700 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6702 unregister_netdevice_queue(dev, &dev_kill_list);
6705 unregister_netdevice_many(&dev_kill_list);
6706 list_del(&dev_kill_list);
6710 static struct pernet_operations __net_initdata default_device_ops = {
6711 .exit = default_device_exit,
6712 .exit_batch = default_device_exit_batch,
6716 * Initialize the DEV module. At boot time this walks the device list and
6717 * unhooks any devices that fail to initialise (normally hardware not
6718 * present) and leaves us with a valid list of present and active devices.
6723 * This is called single threaded during boot, so no need
6724 * to take the rtnl semaphore.
6726 static int __init net_dev_init(void)
6728 int i, rc = -ENOMEM;
6730 BUG_ON(!dev_boot_phase);
6732 if (dev_proc_init())
6735 if (netdev_kobject_init())
6738 INIT_LIST_HEAD(&ptype_all);
6739 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6740 INIT_LIST_HEAD(&ptype_base[i]);
6742 INIT_LIST_HEAD(&offload_base);
6744 if (register_pernet_subsys(&netdev_net_ops))
6748 * Initialise the packet receive queues.
6751 for_each_possible_cpu(i) {
6752 struct softnet_data *sd = &per_cpu(softnet_data, i);
6754 memset(sd, 0, sizeof(*sd));
6755 skb_queue_head_init(&sd->input_pkt_queue);
6756 skb_queue_head_init(&sd->process_queue);
6757 sd->completion_queue = NULL;
6758 INIT_LIST_HEAD(&sd->poll_list);
6759 sd->output_queue = NULL;
6760 sd->output_queue_tailp = &sd->output_queue;
6762 sd->csd.func = rps_trigger_softirq;
6768 sd->backlog.poll = process_backlog;
6769 sd->backlog.weight = weight_p;
6770 sd->backlog.gro_list = NULL;
6771 sd->backlog.gro_count = 0;
6776 /* The loopback device is special if any other network devices
6777 * is present in a network namespace the loopback device must
6778 * be present. Since we now dynamically allocate and free the
6779 * loopback device ensure this invariant is maintained by
6780 * keeping the loopback device as the first device on the
6781 * list of network devices. Ensuring the loopback devices
6782 * is the first device that appears and the last network device
6785 if (register_pernet_device(&loopback_net_ops))
6788 if (register_pernet_device(&default_device_ops))
6791 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6792 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6794 hotcpu_notifier(dev_cpu_callback, 0);
6802 subsys_initcall(net_dev_init);
6804 static int __init initialize_hashrnd(void)
6806 get_random_bytes(&hashrnd, sizeof(hashrnd));
6810 late_initcall_sync(initialize_hashrnd);