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 <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/hash.h>
83 #include <linux/slab.h>
84 #include <linux/sched.h>
85 #include <linux/mutex.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/notifier.h>
97 #include <linux/skbuff.h>
98 #include <net/net_namespace.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/stat.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include <linux/highmem.h>
109 #include <linux/init.h>
110 #include <linux/kmod.h>
111 #include <linux/module.h>
112 #include <linux/netpoll.h>
113 #include <linux/rcupdate.h>
114 #include <linux/delay.h>
115 #include <net/wext.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
126 #include <linux/ipv6.h>
127 #include <linux/in.h>
128 #include <linux/jhash.h>
129 #include <linux/random.h>
130 #include <trace/events/napi.h>
131 #include <linux/pci.h>
132 #include <linux/inetdevice.h>
134 #include "net-sysfs.h"
136 /* Instead of increasing this, you should create a hash table. */
137 #define MAX_GRO_SKBS 8
139 /* This should be increased if a protocol with a bigger head is added. */
140 #define GRO_MAX_HEAD (MAX_HEADER + 128)
143 * The list of packet types we will receive (as opposed to discard)
144 * and the routines to invoke.
146 * Why 16. Because with 16 the only overlap we get on a hash of the
147 * low nibble of the protocol value is RARP/SNAP/X.25.
149 * NOTE: That is no longer true with the addition of VLAN tags. Not
150 * sure which should go first, but I bet it won't make much
151 * difference if we are running VLANs. The good news is that
152 * this protocol won't be in the list unless compiled in, so
153 * the average user (w/out VLANs) will not be adversely affected.
170 #define PTYPE_HASH_SIZE (16)
171 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
173 static DEFINE_SPINLOCK(ptype_lock);
174 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
175 static struct list_head ptype_all __read_mostly; /* Taps */
178 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
181 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
183 * Writers must hold the rtnl semaphore while they loop through the
184 * dev_base_head list, and hold dev_base_lock for writing when they do the
185 * actual updates. This allows pure readers to access the list even
186 * while a writer is preparing to update it.
188 * To put it another way, dev_base_lock is held for writing only to
189 * protect against pure readers; the rtnl semaphore provides the
190 * protection against other writers.
192 * See, for example usages, register_netdevice() and
193 * unregister_netdevice(), which must be called with the rtnl
196 DEFINE_RWLOCK(dev_base_lock);
197 EXPORT_SYMBOL(dev_base_lock);
199 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
201 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
202 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
205 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
207 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
210 static inline void rps_lock(struct softnet_data *sd)
213 spin_lock(&sd->input_pkt_queue.lock);
217 static inline void rps_unlock(struct softnet_data *sd)
220 spin_unlock(&sd->input_pkt_queue.lock);
224 /* Device list insertion */
225 static int list_netdevice(struct net_device *dev)
227 struct net *net = dev_net(dev);
231 write_lock_bh(&dev_base_lock);
232 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
233 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
234 hlist_add_head_rcu(&dev->index_hlist,
235 dev_index_hash(net, dev->ifindex));
236 write_unlock_bh(&dev_base_lock);
240 /* Device list removal
241 * caller must respect a RCU grace period before freeing/reusing dev
243 static void unlist_netdevice(struct net_device *dev)
247 /* Unlink dev from the device chain */
248 write_lock_bh(&dev_base_lock);
249 list_del_rcu(&dev->dev_list);
250 hlist_del_rcu(&dev->name_hlist);
251 hlist_del_rcu(&dev->index_hlist);
252 write_unlock_bh(&dev_base_lock);
259 static RAW_NOTIFIER_HEAD(netdev_chain);
262 * Device drivers call our routines to queue packets here. We empty the
263 * queue in the local softnet handler.
266 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
267 EXPORT_PER_CPU_SYMBOL(softnet_data);
269 #ifdef CONFIG_LOCKDEP
271 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
272 * according to dev->type
274 static const unsigned short netdev_lock_type[] =
275 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
276 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
277 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
278 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
279 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
280 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
281 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
282 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
283 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
284 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
285 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
286 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
287 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
288 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
289 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
290 ARPHRD_VOID, ARPHRD_NONE};
292 static const char *const netdev_lock_name[] =
293 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
294 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
295 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
296 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
297 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
298 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
299 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
300 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
301 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
302 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
303 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
304 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
305 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
306 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
307 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
308 "_xmit_VOID", "_xmit_NONE"};
310 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
311 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
313 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
317 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
318 if (netdev_lock_type[i] == dev_type)
320 /* the last key is used by default */
321 return ARRAY_SIZE(netdev_lock_type) - 1;
324 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
325 unsigned short dev_type)
329 i = netdev_lock_pos(dev_type);
330 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
331 netdev_lock_name[i]);
334 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
338 i = netdev_lock_pos(dev->type);
339 lockdep_set_class_and_name(&dev->addr_list_lock,
340 &netdev_addr_lock_key[i],
341 netdev_lock_name[i]);
344 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
345 unsigned short dev_type)
348 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
353 /*******************************************************************************
355 Protocol management and registration routines
357 *******************************************************************************/
360 * Add a protocol ID to the list. Now that the input handler is
361 * smarter we can dispense with all the messy stuff that used to be
364 * BEWARE!!! Protocol handlers, mangling input packets,
365 * MUST BE last in hash buckets and checking protocol handlers
366 * MUST start from promiscuous ptype_all chain in net_bh.
367 * It is true now, do not change it.
368 * Explanation follows: if protocol handler, mangling packet, will
369 * be the first on list, it is not able to sense, that packet
370 * is cloned and should be copied-on-write, so that it will
371 * change it and subsequent readers will get broken packet.
375 static inline struct list_head *ptype_head(const struct packet_type *pt)
377 if (pt->type == htons(ETH_P_ALL))
380 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
384 * dev_add_pack - add packet handler
385 * @pt: packet type declaration
387 * Add a protocol handler to the networking stack. The passed &packet_type
388 * is linked into kernel lists and may not be freed until it has been
389 * removed from the kernel lists.
391 * This call does not sleep therefore it can not
392 * guarantee all CPU's that are in middle of receiving packets
393 * will see the new packet type (until the next received packet).
396 void dev_add_pack(struct packet_type *pt)
398 struct list_head *head = ptype_head(pt);
400 spin_lock(&ptype_lock);
401 list_add_rcu(&pt->list, head);
402 spin_unlock(&ptype_lock);
404 EXPORT_SYMBOL(dev_add_pack);
407 * __dev_remove_pack - remove packet handler
408 * @pt: packet type declaration
410 * Remove a protocol handler that was previously added to the kernel
411 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
412 * from the kernel lists and can be freed or reused once this function
415 * The packet type might still be in use by receivers
416 * and must not be freed until after all the CPU's have gone
417 * through a quiescent state.
419 void __dev_remove_pack(struct packet_type *pt)
421 struct list_head *head = ptype_head(pt);
422 struct packet_type *pt1;
424 spin_lock(&ptype_lock);
426 list_for_each_entry(pt1, head, list) {
428 list_del_rcu(&pt->list);
433 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
435 spin_unlock(&ptype_lock);
437 EXPORT_SYMBOL(__dev_remove_pack);
440 * dev_remove_pack - remove packet handler
441 * @pt: packet type declaration
443 * Remove a protocol handler that was previously added to the kernel
444 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
445 * from the kernel lists and can be freed or reused once this function
448 * This call sleeps to guarantee that no CPU is looking at the packet
451 void dev_remove_pack(struct packet_type *pt)
453 __dev_remove_pack(pt);
457 EXPORT_SYMBOL(dev_remove_pack);
459 /******************************************************************************
461 Device Boot-time Settings Routines
463 *******************************************************************************/
465 /* Boot time configuration table */
466 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
469 * netdev_boot_setup_add - add new setup entry
470 * @name: name of the device
471 * @map: configured settings for the device
473 * Adds new setup entry to the dev_boot_setup list. The function
474 * returns 0 on error and 1 on success. This is a generic routine to
477 static int netdev_boot_setup_add(char *name, struct ifmap *map)
479 struct netdev_boot_setup *s;
483 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
484 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
485 memset(s[i].name, 0, sizeof(s[i].name));
486 strlcpy(s[i].name, name, IFNAMSIZ);
487 memcpy(&s[i].map, map, sizeof(s[i].map));
492 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
496 * netdev_boot_setup_check - check boot time settings
497 * @dev: the netdevice
499 * Check boot time settings for the device.
500 * The found settings are set for the device to be used
501 * later in the device probing.
502 * Returns 0 if no settings found, 1 if they are.
504 int netdev_boot_setup_check(struct net_device *dev)
506 struct netdev_boot_setup *s = dev_boot_setup;
509 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
510 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
511 !strcmp(dev->name, s[i].name)) {
512 dev->irq = s[i].map.irq;
513 dev->base_addr = s[i].map.base_addr;
514 dev->mem_start = s[i].map.mem_start;
515 dev->mem_end = s[i].map.mem_end;
521 EXPORT_SYMBOL(netdev_boot_setup_check);
525 * netdev_boot_base - get address from boot time settings
526 * @prefix: prefix for network device
527 * @unit: id for network device
529 * Check boot time settings for the base address of device.
530 * The found settings are set for the device to be used
531 * later in the device probing.
532 * Returns 0 if no settings found.
534 unsigned long netdev_boot_base(const char *prefix, int unit)
536 const struct netdev_boot_setup *s = dev_boot_setup;
540 sprintf(name, "%s%d", prefix, unit);
543 * If device already registered then return base of 1
544 * to indicate not to probe for this interface
546 if (__dev_get_by_name(&init_net, name))
549 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
550 if (!strcmp(name, s[i].name))
551 return s[i].map.base_addr;
556 * Saves at boot time configured settings for any netdevice.
558 int __init netdev_boot_setup(char *str)
563 str = get_options(str, ARRAY_SIZE(ints), ints);
568 memset(&map, 0, sizeof(map));
572 map.base_addr = ints[2];
574 map.mem_start = ints[3];
576 map.mem_end = ints[4];
578 /* Add new entry to the list */
579 return netdev_boot_setup_add(str, &map);
582 __setup("netdev=", netdev_boot_setup);
584 /*******************************************************************************
586 Device Interface Subroutines
588 *******************************************************************************/
591 * __dev_get_by_name - find a device by its name
592 * @net: the applicable net namespace
593 * @name: name to find
595 * Find an interface by name. Must be called under RTNL semaphore
596 * or @dev_base_lock. If the name is found a pointer to the device
597 * is returned. If the name is not found then %NULL is returned. The
598 * reference counters are not incremented so the caller must be
599 * careful with locks.
602 struct net_device *__dev_get_by_name(struct net *net, const char *name)
604 struct hlist_node *p;
605 struct net_device *dev;
606 struct hlist_head *head = dev_name_hash(net, name);
608 hlist_for_each_entry(dev, p, head, name_hlist)
609 if (!strncmp(dev->name, name, IFNAMSIZ))
614 EXPORT_SYMBOL(__dev_get_by_name);
617 * dev_get_by_name_rcu - find a device by its name
618 * @net: the applicable net namespace
619 * @name: name to find
621 * Find an interface by name.
622 * If the name is found a pointer to the device is returned.
623 * If the name is not found then %NULL is returned.
624 * The reference counters are not incremented so the caller must be
625 * careful with locks. The caller must hold RCU lock.
628 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
630 struct hlist_node *p;
631 struct net_device *dev;
632 struct hlist_head *head = dev_name_hash(net, name);
634 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
635 if (!strncmp(dev->name, name, IFNAMSIZ))
640 EXPORT_SYMBOL(dev_get_by_name_rcu);
643 * dev_get_by_name - find a device by its name
644 * @net: the applicable net namespace
645 * @name: name to find
647 * Find an interface by name. This can be called from any
648 * context and does its own locking. The returned handle has
649 * the usage count incremented and the caller must use dev_put() to
650 * release it when it is no longer needed. %NULL is returned if no
651 * matching device is found.
654 struct net_device *dev_get_by_name(struct net *net, const char *name)
656 struct net_device *dev;
659 dev = dev_get_by_name_rcu(net, name);
665 EXPORT_SYMBOL(dev_get_by_name);
668 * __dev_get_by_index - find a device by its ifindex
669 * @net: the applicable net namespace
670 * @ifindex: index of device
672 * Search for an interface by index. Returns %NULL if the device
673 * is not found or a pointer to the device. The device has not
674 * had its reference counter increased so the caller must be careful
675 * about locking. The caller must hold either the RTNL semaphore
679 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
681 struct hlist_node *p;
682 struct net_device *dev;
683 struct hlist_head *head = dev_index_hash(net, ifindex);
685 hlist_for_each_entry(dev, p, head, index_hlist)
686 if (dev->ifindex == ifindex)
691 EXPORT_SYMBOL(__dev_get_by_index);
694 * dev_get_by_index_rcu - find a device by its ifindex
695 * @net: the applicable net namespace
696 * @ifindex: index of device
698 * Search for an interface by index. Returns %NULL if the device
699 * is not found or a pointer to the device. The device has not
700 * had its reference counter increased so the caller must be careful
701 * about locking. The caller must hold RCU lock.
704 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
706 struct hlist_node *p;
707 struct net_device *dev;
708 struct hlist_head *head = dev_index_hash(net, ifindex);
710 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
711 if (dev->ifindex == ifindex)
716 EXPORT_SYMBOL(dev_get_by_index_rcu);
720 * dev_get_by_index - find a device by its ifindex
721 * @net: the applicable net namespace
722 * @ifindex: index of device
724 * Search for an interface by index. Returns NULL if the device
725 * is not found or a pointer to the device. The device returned has
726 * had a reference added and the pointer is safe until the user calls
727 * dev_put to indicate they have finished with it.
730 struct net_device *dev_get_by_index(struct net *net, int ifindex)
732 struct net_device *dev;
735 dev = dev_get_by_index_rcu(net, ifindex);
741 EXPORT_SYMBOL(dev_get_by_index);
744 * dev_getbyhwaddr - find a device by its hardware address
745 * @net: the applicable net namespace
746 * @type: media type of device
747 * @ha: hardware address
749 * Search for an interface by MAC address. Returns NULL if the device
750 * is not found or a pointer to the device. The caller must hold the
751 * rtnl semaphore. The returned device has not had its ref count increased
752 * and the caller must therefore be careful about locking
755 * If the API was consistent this would be __dev_get_by_hwaddr
758 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
760 struct net_device *dev;
764 for_each_netdev(net, dev)
765 if (dev->type == type &&
766 !memcmp(dev->dev_addr, ha, dev->addr_len))
771 EXPORT_SYMBOL(dev_getbyhwaddr);
773 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
775 struct net_device *dev;
778 for_each_netdev(net, dev)
779 if (dev->type == type)
784 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
786 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
788 struct net_device *dev, *ret = NULL;
791 for_each_netdev_rcu(net, dev)
792 if (dev->type == type) {
800 EXPORT_SYMBOL(dev_getfirstbyhwtype);
803 * dev_get_by_flags_rcu - find any device with given flags
804 * @net: the applicable net namespace
805 * @if_flags: IFF_* values
806 * @mask: bitmask of bits in if_flags to check
808 * Search for any interface with the given flags. Returns NULL if a device
809 * is not found or a pointer to the device. Must be called inside
810 * rcu_read_lock(), and result refcount is unchanged.
813 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
816 struct net_device *dev, *ret;
819 for_each_netdev_rcu(net, dev) {
820 if (((dev->flags ^ if_flags) & mask) == 0) {
827 EXPORT_SYMBOL(dev_get_by_flags_rcu);
830 * dev_valid_name - check if name is okay for network device
833 * Network device names need to be valid file names to
834 * to allow sysfs to work. We also disallow any kind of
837 int dev_valid_name(const char *name)
841 if (strlen(name) >= IFNAMSIZ)
843 if (!strcmp(name, ".") || !strcmp(name, ".."))
847 if (*name == '/' || isspace(*name))
853 EXPORT_SYMBOL(dev_valid_name);
856 * __dev_alloc_name - allocate a name for a device
857 * @net: network namespace to allocate the device name in
858 * @name: name format string
859 * @buf: scratch buffer and result name string
861 * Passed a format string - eg "lt%d" it will try and find a suitable
862 * id. It scans list of devices to build up a free map, then chooses
863 * the first empty slot. The caller must hold the dev_base or rtnl lock
864 * while allocating the name and adding the device in order to avoid
866 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
867 * Returns the number of the unit assigned or a negative errno code.
870 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
874 const int max_netdevices = 8*PAGE_SIZE;
875 unsigned long *inuse;
876 struct net_device *d;
878 p = strnchr(name, IFNAMSIZ-1, '%');
881 * Verify the string as this thing may have come from
882 * the user. There must be either one "%d" and no other "%"
885 if (p[1] != 'd' || strchr(p + 2, '%'))
888 /* Use one page as a bit array of possible slots */
889 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
893 for_each_netdev(net, d) {
894 if (!sscanf(d->name, name, &i))
896 if (i < 0 || i >= max_netdevices)
899 /* avoid cases where sscanf is not exact inverse of printf */
900 snprintf(buf, IFNAMSIZ, name, i);
901 if (!strncmp(buf, d->name, IFNAMSIZ))
905 i = find_first_zero_bit(inuse, max_netdevices);
906 free_page((unsigned long) inuse);
910 snprintf(buf, IFNAMSIZ, name, i);
911 if (!__dev_get_by_name(net, buf))
914 /* It is possible to run out of possible slots
915 * when the name is long and there isn't enough space left
916 * for the digits, or if all bits are used.
922 * dev_alloc_name - allocate a name for a device
924 * @name: name format string
926 * Passed a format string - eg "lt%d" it will try and find a suitable
927 * id. It scans list of devices to build up a free map, then chooses
928 * the first empty slot. The caller must hold the dev_base or rtnl lock
929 * while allocating the name and adding the device in order to avoid
931 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
932 * Returns the number of the unit assigned or a negative errno code.
935 int dev_alloc_name(struct net_device *dev, const char *name)
941 BUG_ON(!dev_net(dev));
943 ret = __dev_alloc_name(net, name, buf);
945 strlcpy(dev->name, buf, IFNAMSIZ);
948 EXPORT_SYMBOL(dev_alloc_name);
950 static int dev_get_valid_name(struct net_device *dev, const char *name, bool fmt)
954 BUG_ON(!dev_net(dev));
957 if (!dev_valid_name(name))
960 if (fmt && strchr(name, '%'))
961 return dev_alloc_name(dev, name);
962 else if (__dev_get_by_name(net, name))
964 else if (dev->name != name)
965 strlcpy(dev->name, name, IFNAMSIZ);
971 * dev_change_name - change name of a device
973 * @newname: name (or format string) must be at least IFNAMSIZ
975 * Change name of a device, can pass format strings "eth%d".
978 int dev_change_name(struct net_device *dev, const char *newname)
980 char oldname[IFNAMSIZ];
986 BUG_ON(!dev_net(dev));
989 if (dev->flags & IFF_UP)
992 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
995 memcpy(oldname, dev->name, IFNAMSIZ);
997 err = dev_get_valid_name(dev, newname, 1);
1002 ret = device_rename(&dev->dev, dev->name);
1004 memcpy(dev->name, oldname, IFNAMSIZ);
1008 write_lock_bh(&dev_base_lock);
1009 hlist_del(&dev->name_hlist);
1010 write_unlock_bh(&dev_base_lock);
1014 write_lock_bh(&dev_base_lock);
1015 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1016 write_unlock_bh(&dev_base_lock);
1018 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1019 ret = notifier_to_errno(ret);
1022 /* err >= 0 after dev_alloc_name() or stores the first errno */
1025 memcpy(dev->name, oldname, IFNAMSIZ);
1029 "%s: name change rollback failed: %d.\n",
1038 * dev_set_alias - change ifalias of a device
1040 * @alias: name up to IFALIASZ
1041 * @len: limit of bytes to copy from info
1043 * Set ifalias for a device,
1045 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1049 if (len >= IFALIASZ)
1054 kfree(dev->ifalias);
1055 dev->ifalias = NULL;
1060 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1064 strlcpy(dev->ifalias, alias, len+1);
1070 * netdev_features_change - device changes features
1071 * @dev: device to cause notification
1073 * Called to indicate a device has changed features.
1075 void netdev_features_change(struct net_device *dev)
1077 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1079 EXPORT_SYMBOL(netdev_features_change);
1082 * netdev_state_change - device changes state
1083 * @dev: device to cause notification
1085 * Called to indicate a device has changed state. This function calls
1086 * the notifier chains for netdev_chain and sends a NEWLINK message
1087 * to the routing socket.
1089 void netdev_state_change(struct net_device *dev)
1091 if (dev->flags & IFF_UP) {
1092 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1093 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1096 EXPORT_SYMBOL(netdev_state_change);
1098 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1100 return call_netdevice_notifiers(event, dev);
1102 EXPORT_SYMBOL(netdev_bonding_change);
1105 * dev_load - load a network module
1106 * @net: the applicable net namespace
1107 * @name: name of interface
1109 * If a network interface is not present and the process has suitable
1110 * privileges this function loads the module. If module loading is not
1111 * available in this kernel then it becomes a nop.
1114 void dev_load(struct net *net, const char *name)
1116 struct net_device *dev;
1119 dev = dev_get_by_name_rcu(net, name);
1122 if (!dev && capable(CAP_NET_ADMIN))
1123 request_module("%s", name);
1125 EXPORT_SYMBOL(dev_load);
1127 static int __dev_open(struct net_device *dev)
1129 const struct net_device_ops *ops = dev->netdev_ops;
1135 * Is it even present?
1137 if (!netif_device_present(dev))
1140 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1141 ret = notifier_to_errno(ret);
1146 * Call device private open method
1148 set_bit(__LINK_STATE_START, &dev->state);
1150 if (ops->ndo_validate_addr)
1151 ret = ops->ndo_validate_addr(dev);
1153 if (!ret && ops->ndo_open)
1154 ret = ops->ndo_open(dev);
1157 * If it went open OK then:
1161 clear_bit(__LINK_STATE_START, &dev->state);
1166 dev->flags |= IFF_UP;
1171 net_dmaengine_get();
1174 * Initialize multicasting status
1176 dev_set_rx_mode(dev);
1179 * Wakeup transmit queue engine
1188 * dev_open - prepare an interface for use.
1189 * @dev: device to open
1191 * Takes a device from down to up state. The device's private open
1192 * function is invoked and then the multicast lists are loaded. Finally
1193 * the device is moved into the up state and a %NETDEV_UP message is
1194 * sent to the netdev notifier chain.
1196 * Calling this function on an active interface is a nop. On a failure
1197 * a negative errno code is returned.
1199 int dev_open(struct net_device *dev)
1206 if (dev->flags & IFF_UP)
1212 ret = __dev_open(dev);
1217 * ... and announce new interface.
1219 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1220 call_netdevice_notifiers(NETDEV_UP, dev);
1224 EXPORT_SYMBOL(dev_open);
1226 static int __dev_close(struct net_device *dev)
1228 const struct net_device_ops *ops = dev->netdev_ops;
1234 * Tell people we are going down, so that they can
1235 * prepare to death, when device is still operating.
1237 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1239 clear_bit(__LINK_STATE_START, &dev->state);
1241 /* Synchronize to scheduled poll. We cannot touch poll list,
1242 * it can be even on different cpu. So just clear netif_running().
1244 * dev->stop() will invoke napi_disable() on all of it's
1245 * napi_struct instances on this device.
1247 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1249 dev_deactivate(dev);
1252 * Call the device specific close. This cannot fail.
1253 * Only if device is UP
1255 * We allow it to be called even after a DETACH hot-plug
1262 * Device is now down.
1265 dev->flags &= ~IFF_UP;
1270 net_dmaengine_put();
1276 * dev_close - shutdown an interface.
1277 * @dev: device to shutdown
1279 * This function moves an active device into down state. A
1280 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1281 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1284 int dev_close(struct net_device *dev)
1286 if (!(dev->flags & IFF_UP))
1292 * Tell people we are down
1294 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1295 call_netdevice_notifiers(NETDEV_DOWN, dev);
1299 EXPORT_SYMBOL(dev_close);
1303 * dev_disable_lro - disable Large Receive Offload on a device
1306 * Disable Large Receive Offload (LRO) on a net device. Must be
1307 * called under RTNL. This is needed if received packets may be
1308 * forwarded to another interface.
1310 void dev_disable_lro(struct net_device *dev)
1312 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1313 dev->ethtool_ops->set_flags) {
1314 u32 flags = dev->ethtool_ops->get_flags(dev);
1315 if (flags & ETH_FLAG_LRO) {
1316 flags &= ~ETH_FLAG_LRO;
1317 dev->ethtool_ops->set_flags(dev, flags);
1320 WARN_ON(dev->features & NETIF_F_LRO);
1322 EXPORT_SYMBOL(dev_disable_lro);
1325 static int dev_boot_phase = 1;
1328 * Device change register/unregister. These are not inline or static
1329 * as we export them to the world.
1333 * register_netdevice_notifier - register a network notifier block
1336 * Register a notifier to be called when network device events occur.
1337 * The notifier passed is linked into the kernel structures and must
1338 * not be reused until it has been unregistered. A negative errno code
1339 * is returned on a failure.
1341 * When registered all registration and up events are replayed
1342 * to the new notifier to allow device to have a race free
1343 * view of the network device list.
1346 int register_netdevice_notifier(struct notifier_block *nb)
1348 struct net_device *dev;
1349 struct net_device *last;
1354 err = raw_notifier_chain_register(&netdev_chain, nb);
1360 for_each_netdev(net, dev) {
1361 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1362 err = notifier_to_errno(err);
1366 if (!(dev->flags & IFF_UP))
1369 nb->notifier_call(nb, NETDEV_UP, dev);
1380 for_each_netdev(net, dev) {
1384 if (dev->flags & IFF_UP) {
1385 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1386 nb->notifier_call(nb, NETDEV_DOWN, dev);
1388 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1389 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1393 raw_notifier_chain_unregister(&netdev_chain, nb);
1396 EXPORT_SYMBOL(register_netdevice_notifier);
1399 * unregister_netdevice_notifier - unregister a network notifier block
1402 * Unregister a notifier previously registered by
1403 * register_netdevice_notifier(). The notifier is unlinked into the
1404 * kernel structures and may then be reused. A negative errno code
1405 * is returned on a failure.
1408 int unregister_netdevice_notifier(struct notifier_block *nb)
1413 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1417 EXPORT_SYMBOL(unregister_netdevice_notifier);
1420 * call_netdevice_notifiers - call all network notifier blocks
1421 * @val: value passed unmodified to notifier function
1422 * @dev: net_device pointer passed unmodified to notifier function
1424 * Call all network notifier blocks. Parameters and return value
1425 * are as for raw_notifier_call_chain().
1428 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1431 return raw_notifier_call_chain(&netdev_chain, val, dev);
1434 /* When > 0 there are consumers of rx skb time stamps */
1435 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1437 void net_enable_timestamp(void)
1439 atomic_inc(&netstamp_needed);
1441 EXPORT_SYMBOL(net_enable_timestamp);
1443 void net_disable_timestamp(void)
1445 atomic_dec(&netstamp_needed);
1447 EXPORT_SYMBOL(net_disable_timestamp);
1449 static inline void net_timestamp_set(struct sk_buff *skb)
1451 if (atomic_read(&netstamp_needed))
1452 __net_timestamp(skb);
1454 skb->tstamp.tv64 = 0;
1457 static inline void net_timestamp_check(struct sk_buff *skb)
1459 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1460 __net_timestamp(skb);
1464 * dev_forward_skb - loopback an skb to another netif
1466 * @dev: destination network device
1467 * @skb: buffer to forward
1470 * NET_RX_SUCCESS (no congestion)
1471 * NET_RX_DROP (packet was dropped, but freed)
1473 * dev_forward_skb can be used for injecting an skb from the
1474 * start_xmit function of one device into the receive queue
1475 * of another device.
1477 * The receiving device may be in another namespace, so
1478 * we have to clear all information in the skb that could
1479 * impact namespace isolation.
1481 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1486 if (unlikely(!(dev->flags & IFF_UP) ||
1487 (skb->len > (dev->mtu + dev->hard_header_len)))) {
1488 atomic_long_inc(&dev->rx_dropped);
1492 skb_set_dev(skb, dev);
1493 skb->tstamp.tv64 = 0;
1494 skb->pkt_type = PACKET_HOST;
1495 skb->protocol = eth_type_trans(skb, dev);
1496 return netif_rx(skb);
1498 EXPORT_SYMBOL_GPL(dev_forward_skb);
1501 * Support routine. Sends outgoing frames to any network
1502 * taps currently in use.
1505 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1507 struct packet_type *ptype;
1509 #ifdef CONFIG_NET_CLS_ACT
1510 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1511 net_timestamp_set(skb);
1513 net_timestamp_set(skb);
1517 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1518 /* Never send packets back to the socket
1519 * they originated from - MvS (miquels@drinkel.ow.org)
1521 if ((ptype->dev == dev || !ptype->dev) &&
1522 (ptype->af_packet_priv == NULL ||
1523 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1524 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1528 /* skb->nh should be correctly
1529 set by sender, so that the second statement is
1530 just protection against buggy protocols.
1532 skb_reset_mac_header(skb2);
1534 if (skb_network_header(skb2) < skb2->data ||
1535 skb2->network_header > skb2->tail) {
1536 if (net_ratelimit())
1537 printk(KERN_CRIT "protocol %04x is "
1539 ntohs(skb2->protocol),
1541 skb_reset_network_header(skb2);
1544 skb2->transport_header = skb2->network_header;
1545 skb2->pkt_type = PACKET_OUTGOING;
1546 ptype->func(skb2, skb->dev, ptype, skb->dev);
1553 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1554 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1556 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1558 if (txq < 1 || txq > dev->num_tx_queues)
1561 if (dev->reg_state == NETREG_REGISTERED) {
1564 if (txq < dev->real_num_tx_queues)
1565 qdisc_reset_all_tx_gt(dev, txq);
1568 dev->real_num_tx_queues = txq;
1571 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1575 * netif_set_real_num_rx_queues - set actual number of RX queues used
1576 * @dev: Network device
1577 * @rxq: Actual number of RX queues
1579 * This must be called either with the rtnl_lock held or before
1580 * registration of the net device. Returns 0 on success, or a
1581 * negative error code. If called before registration, it always
1584 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1588 if (rxq < 1 || rxq > dev->num_rx_queues)
1591 if (dev->reg_state == NETREG_REGISTERED) {
1594 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1600 dev->real_num_rx_queues = rxq;
1603 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1606 static inline void __netif_reschedule(struct Qdisc *q)
1608 struct softnet_data *sd;
1609 unsigned long flags;
1611 local_irq_save(flags);
1612 sd = &__get_cpu_var(softnet_data);
1613 q->next_sched = NULL;
1614 *sd->output_queue_tailp = q;
1615 sd->output_queue_tailp = &q->next_sched;
1616 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1617 local_irq_restore(flags);
1620 void __netif_schedule(struct Qdisc *q)
1622 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1623 __netif_reschedule(q);
1625 EXPORT_SYMBOL(__netif_schedule);
1627 void dev_kfree_skb_irq(struct sk_buff *skb)
1629 if (atomic_dec_and_test(&skb->users)) {
1630 struct softnet_data *sd;
1631 unsigned long flags;
1633 local_irq_save(flags);
1634 sd = &__get_cpu_var(softnet_data);
1635 skb->next = sd->completion_queue;
1636 sd->completion_queue = skb;
1637 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1638 local_irq_restore(flags);
1641 EXPORT_SYMBOL(dev_kfree_skb_irq);
1643 void dev_kfree_skb_any(struct sk_buff *skb)
1645 if (in_irq() || irqs_disabled())
1646 dev_kfree_skb_irq(skb);
1650 EXPORT_SYMBOL(dev_kfree_skb_any);
1654 * netif_device_detach - mark device as removed
1655 * @dev: network device
1657 * Mark device as removed from system and therefore no longer available.
1659 void netif_device_detach(struct net_device *dev)
1661 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1662 netif_running(dev)) {
1663 netif_tx_stop_all_queues(dev);
1666 EXPORT_SYMBOL(netif_device_detach);
1669 * netif_device_attach - mark device as attached
1670 * @dev: network device
1672 * Mark device as attached from system and restart if needed.
1674 void netif_device_attach(struct net_device *dev)
1676 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1677 netif_running(dev)) {
1678 netif_tx_wake_all_queues(dev);
1679 __netdev_watchdog_up(dev);
1682 EXPORT_SYMBOL(netif_device_attach);
1684 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1686 return ((features & NETIF_F_GEN_CSUM) ||
1687 ((features & NETIF_F_IP_CSUM) &&
1688 protocol == htons(ETH_P_IP)) ||
1689 ((features & NETIF_F_IPV6_CSUM) &&
1690 protocol == htons(ETH_P_IPV6)) ||
1691 ((features & NETIF_F_FCOE_CRC) &&
1692 protocol == htons(ETH_P_FCOE)));
1695 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1697 if (can_checksum_protocol(dev->features, skb->protocol))
1700 if (skb->protocol == htons(ETH_P_8021Q)) {
1701 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1702 if (can_checksum_protocol(dev->features & dev->vlan_features,
1703 veh->h_vlan_encapsulated_proto))
1711 * skb_dev_set -- assign a new device to a buffer
1712 * @skb: buffer for the new device
1713 * @dev: network device
1715 * If an skb is owned by a device already, we have to reset
1716 * all data private to the namespace a device belongs to
1717 * before assigning it a new device.
1719 #ifdef CONFIG_NET_NS
1720 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1723 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1726 skb_init_secmark(skb);
1730 skb->ipvs_property = 0;
1731 #ifdef CONFIG_NET_SCHED
1737 EXPORT_SYMBOL(skb_set_dev);
1738 #endif /* CONFIG_NET_NS */
1741 * Invalidate hardware checksum when packet is to be mangled, and
1742 * complete checksum manually on outgoing path.
1744 int skb_checksum_help(struct sk_buff *skb)
1747 int ret = 0, offset;
1749 if (skb->ip_summed == CHECKSUM_COMPLETE)
1750 goto out_set_summed;
1752 if (unlikely(skb_shinfo(skb)->gso_size)) {
1753 /* Let GSO fix up the checksum. */
1754 goto out_set_summed;
1757 offset = skb->csum_start - skb_headroom(skb);
1758 BUG_ON(offset >= skb_headlen(skb));
1759 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1761 offset += skb->csum_offset;
1762 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1764 if (skb_cloned(skb) &&
1765 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1766 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1771 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1773 skb->ip_summed = CHECKSUM_NONE;
1777 EXPORT_SYMBOL(skb_checksum_help);
1780 * skb_gso_segment - Perform segmentation on skb.
1781 * @skb: buffer to segment
1782 * @features: features for the output path (see dev->features)
1784 * This function segments the given skb and returns a list of segments.
1786 * It may return NULL if the skb requires no segmentation. This is
1787 * only possible when GSO is used for verifying header integrity.
1789 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1791 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1792 struct packet_type *ptype;
1793 __be16 type = skb->protocol;
1796 skb_reset_mac_header(skb);
1797 skb->mac_len = skb->network_header - skb->mac_header;
1798 __skb_pull(skb, skb->mac_len);
1800 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1801 struct net_device *dev = skb->dev;
1802 struct ethtool_drvinfo info = {};
1804 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1805 dev->ethtool_ops->get_drvinfo(dev, &info);
1807 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1809 info.driver, dev ? dev->features : 0L,
1810 skb->sk ? skb->sk->sk_route_caps : 0L,
1811 skb->len, skb->data_len, skb->ip_summed);
1813 if (skb_header_cloned(skb) &&
1814 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1815 return ERR_PTR(err);
1819 list_for_each_entry_rcu(ptype,
1820 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1821 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1822 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1823 err = ptype->gso_send_check(skb);
1824 segs = ERR_PTR(err);
1825 if (err || skb_gso_ok(skb, features))
1827 __skb_push(skb, (skb->data -
1828 skb_network_header(skb)));
1830 segs = ptype->gso_segment(skb, features);
1836 __skb_push(skb, skb->data - skb_mac_header(skb));
1840 EXPORT_SYMBOL(skb_gso_segment);
1842 /* Take action when hardware reception checksum errors are detected. */
1844 void netdev_rx_csum_fault(struct net_device *dev)
1846 if (net_ratelimit()) {
1847 printk(KERN_ERR "%s: hw csum failure.\n",
1848 dev ? dev->name : "<unknown>");
1852 EXPORT_SYMBOL(netdev_rx_csum_fault);
1855 /* Actually, we should eliminate this check as soon as we know, that:
1856 * 1. IOMMU is present and allows to map all the memory.
1857 * 2. No high memory really exists on this machine.
1860 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1862 #ifdef CONFIG_HIGHMEM
1864 if (!(dev->features & NETIF_F_HIGHDMA)) {
1865 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1866 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1870 if (PCI_DMA_BUS_IS_PHYS) {
1871 struct device *pdev = dev->dev.parent;
1875 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1876 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1877 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1886 void (*destructor)(struct sk_buff *skb);
1889 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1891 static void dev_gso_skb_destructor(struct sk_buff *skb)
1893 struct dev_gso_cb *cb;
1896 struct sk_buff *nskb = skb->next;
1898 skb->next = nskb->next;
1901 } while (skb->next);
1903 cb = DEV_GSO_CB(skb);
1905 cb->destructor(skb);
1909 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1910 * @skb: buffer to segment
1912 * This function segments the given skb and stores the list of segments
1915 static int dev_gso_segment(struct sk_buff *skb)
1917 struct net_device *dev = skb->dev;
1918 struct sk_buff *segs;
1919 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1922 segs = skb_gso_segment(skb, features);
1924 /* Verifying header integrity only. */
1929 return PTR_ERR(segs);
1932 DEV_GSO_CB(skb)->destructor = skb->destructor;
1933 skb->destructor = dev_gso_skb_destructor;
1939 * Try to orphan skb early, right before transmission by the device.
1940 * We cannot orphan skb if tx timestamp is requested or the sk-reference
1941 * is needed on driver level for other reasons, e.g. see net/can/raw.c
1943 static inline void skb_orphan_try(struct sk_buff *skb)
1945 struct sock *sk = skb->sk;
1947 if (sk && !skb_shinfo(skb)->tx_flags) {
1948 /* skb_tx_hash() wont be able to get sk.
1949 * We copy sk_hash into skb->rxhash
1952 skb->rxhash = sk->sk_hash;
1958 * Returns true if either:
1959 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
1960 * 2. skb is fragmented and the device does not support SG, or if
1961 * at least one of fragments is in highmem and device does not
1962 * support DMA from it.
1964 static inline int skb_needs_linearize(struct sk_buff *skb,
1965 struct net_device *dev)
1967 return skb_is_nonlinear(skb) &&
1968 ((skb_has_frag_list(skb) && !(dev->features & NETIF_F_FRAGLIST)) ||
1969 (skb_shinfo(skb)->nr_frags && (!(dev->features & NETIF_F_SG) ||
1970 illegal_highdma(dev, skb))));
1973 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1974 struct netdev_queue *txq)
1976 const struct net_device_ops *ops = dev->netdev_ops;
1977 int rc = NETDEV_TX_OK;
1979 if (likely(!skb->next)) {
1980 if (!list_empty(&ptype_all))
1981 dev_queue_xmit_nit(skb, dev);
1984 * If device doesnt need skb->dst, release it right now while
1985 * its hot in this cpu cache
1987 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1990 skb_orphan_try(skb);
1992 if (netif_needs_gso(dev, skb)) {
1993 if (unlikely(dev_gso_segment(skb)))
1998 if (skb_needs_linearize(skb, dev) &&
1999 __skb_linearize(skb))
2002 /* If packet is not checksummed and device does not
2003 * support checksumming for this protocol, complete
2004 * checksumming here.
2006 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2007 skb_set_transport_header(skb, skb->csum_start -
2009 if (!dev_can_checksum(dev, skb) &&
2010 skb_checksum_help(skb))
2015 rc = ops->ndo_start_xmit(skb, dev);
2016 if (rc == NETDEV_TX_OK)
2017 txq_trans_update(txq);
2023 struct sk_buff *nskb = skb->next;
2025 skb->next = nskb->next;
2029 * If device doesnt need nskb->dst, release it right now while
2030 * its hot in this cpu cache
2032 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2035 rc = ops->ndo_start_xmit(nskb, dev);
2036 if (unlikely(rc != NETDEV_TX_OK)) {
2037 if (rc & ~NETDEV_TX_MASK)
2038 goto out_kfree_gso_skb;
2039 nskb->next = skb->next;
2043 txq_trans_update(txq);
2044 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2045 return NETDEV_TX_BUSY;
2046 } while (skb->next);
2049 if (likely(skb->next == NULL))
2050 skb->destructor = DEV_GSO_CB(skb)->destructor;
2056 static u32 hashrnd __read_mostly;
2058 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
2062 if (skb_rx_queue_recorded(skb)) {
2063 hash = skb_get_rx_queue(skb);
2064 while (unlikely(hash >= dev->real_num_tx_queues))
2065 hash -= dev->real_num_tx_queues;
2069 if (skb->sk && skb->sk->sk_hash)
2070 hash = skb->sk->sk_hash;
2072 hash = (__force u16) skb->protocol ^ skb->rxhash;
2073 hash = jhash_1word(hash, hashrnd);
2075 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
2077 EXPORT_SYMBOL(skb_tx_hash);
2079 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2081 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2082 if (net_ratelimit()) {
2083 pr_warning("%s selects TX queue %d, but "
2084 "real number of TX queues is %d\n",
2085 dev->name, queue_index, dev->real_num_tx_queues);
2092 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2093 struct sk_buff *skb)
2096 const struct net_device_ops *ops = dev->netdev_ops;
2098 if (ops->ndo_select_queue) {
2099 queue_index = ops->ndo_select_queue(dev, skb);
2100 queue_index = dev_cap_txqueue(dev, queue_index);
2102 struct sock *sk = skb->sk;
2103 queue_index = sk_tx_queue_get(sk);
2104 if (queue_index < 0) {
2107 if (dev->real_num_tx_queues > 1)
2108 queue_index = skb_tx_hash(dev, skb);
2111 struct dst_entry *dst = rcu_dereference_check(sk->sk_dst_cache, 1);
2113 if (dst && skb_dst(skb) == dst)
2114 sk_tx_queue_set(sk, queue_index);
2119 skb_set_queue_mapping(skb, queue_index);
2120 return netdev_get_tx_queue(dev, queue_index);
2123 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2124 struct net_device *dev,
2125 struct netdev_queue *txq)
2127 spinlock_t *root_lock = qdisc_lock(q);
2128 bool contended = qdisc_is_running(q);
2132 * Heuristic to force contended enqueues to serialize on a
2133 * separate lock before trying to get qdisc main lock.
2134 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2135 * and dequeue packets faster.
2137 if (unlikely(contended))
2138 spin_lock(&q->busylock);
2140 spin_lock(root_lock);
2141 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2144 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2145 qdisc_run_begin(q)) {
2147 * This is a work-conserving queue; there are no old skbs
2148 * waiting to be sent out; and the qdisc is not running -
2149 * xmit the skb directly.
2151 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2153 __qdisc_update_bstats(q, skb->len);
2154 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2155 if (unlikely(contended)) {
2156 spin_unlock(&q->busylock);
2163 rc = NET_XMIT_SUCCESS;
2166 rc = qdisc_enqueue_root(skb, q);
2167 if (qdisc_run_begin(q)) {
2168 if (unlikely(contended)) {
2169 spin_unlock(&q->busylock);
2175 spin_unlock(root_lock);
2176 if (unlikely(contended))
2177 spin_unlock(&q->busylock);
2181 static DEFINE_PER_CPU(int, xmit_recursion);
2182 #define RECURSION_LIMIT 3
2185 * dev_queue_xmit - transmit a buffer
2186 * @skb: buffer to transmit
2188 * Queue a buffer for transmission to a network device. The caller must
2189 * have set the device and priority and built the buffer before calling
2190 * this function. The function can be called from an interrupt.
2192 * A negative errno code is returned on a failure. A success does not
2193 * guarantee the frame will be transmitted as it may be dropped due
2194 * to congestion or traffic shaping.
2196 * -----------------------------------------------------------------------------------
2197 * I notice this method can also return errors from the queue disciplines,
2198 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2201 * Regardless of the return value, the skb is consumed, so it is currently
2202 * difficult to retry a send to this method. (You can bump the ref count
2203 * before sending to hold a reference for retry if you are careful.)
2205 * When calling this method, interrupts MUST be enabled. This is because
2206 * the BH enable code must have IRQs enabled so that it will not deadlock.
2209 int dev_queue_xmit(struct sk_buff *skb)
2211 struct net_device *dev = skb->dev;
2212 struct netdev_queue *txq;
2216 /* Disable soft irqs for various locks below. Also
2217 * stops preemption for RCU.
2221 txq = dev_pick_tx(dev, skb);
2222 q = rcu_dereference_bh(txq->qdisc);
2224 #ifdef CONFIG_NET_CLS_ACT
2225 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2228 rc = __dev_xmit_skb(skb, q, dev, txq);
2232 /* The device has no queue. Common case for software devices:
2233 loopback, all the sorts of tunnels...
2235 Really, it is unlikely that netif_tx_lock protection is necessary
2236 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2238 However, it is possible, that they rely on protection
2241 Check this and shot the lock. It is not prone from deadlocks.
2242 Either shot noqueue qdisc, it is even simpler 8)
2244 if (dev->flags & IFF_UP) {
2245 int cpu = smp_processor_id(); /* ok because BHs are off */
2247 if (txq->xmit_lock_owner != cpu) {
2249 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2250 goto recursion_alert;
2252 HARD_TX_LOCK(dev, txq, cpu);
2254 if (!netif_tx_queue_stopped(txq)) {
2255 __this_cpu_inc(xmit_recursion);
2256 rc = dev_hard_start_xmit(skb, dev, txq);
2257 __this_cpu_dec(xmit_recursion);
2258 if (dev_xmit_complete(rc)) {
2259 HARD_TX_UNLOCK(dev, txq);
2263 HARD_TX_UNLOCK(dev, txq);
2264 if (net_ratelimit())
2265 printk(KERN_CRIT "Virtual device %s asks to "
2266 "queue packet!\n", dev->name);
2268 /* Recursion is detected! It is possible,
2272 if (net_ratelimit())
2273 printk(KERN_CRIT "Dead loop on virtual device "
2274 "%s, fix it urgently!\n", dev->name);
2279 rcu_read_unlock_bh();
2284 rcu_read_unlock_bh();
2287 EXPORT_SYMBOL(dev_queue_xmit);
2290 /*=======================================================================
2292 =======================================================================*/
2294 int netdev_max_backlog __read_mostly = 1000;
2295 int netdev_tstamp_prequeue __read_mostly = 1;
2296 int netdev_budget __read_mostly = 300;
2297 int weight_p __read_mostly = 64; /* old backlog weight */
2299 /* Called with irq disabled */
2300 static inline void ____napi_schedule(struct softnet_data *sd,
2301 struct napi_struct *napi)
2303 list_add_tail(&napi->poll_list, &sd->poll_list);
2304 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2308 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2309 * and src/dst port numbers. Returns a non-zero hash number on success
2312 __u32 __skb_get_rxhash(struct sk_buff *skb)
2314 int nhoff, hash = 0, poff;
2315 struct ipv6hdr *ip6;
2318 u32 addr1, addr2, ihl;
2324 nhoff = skb_network_offset(skb);
2326 switch (skb->protocol) {
2327 case __constant_htons(ETH_P_IP):
2328 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2331 ip = (struct iphdr *) (skb->data + nhoff);
2332 if (ip->frag_off & htons(IP_MF | IP_OFFSET))
2335 ip_proto = ip->protocol;
2336 addr1 = (__force u32) ip->saddr;
2337 addr2 = (__force u32) ip->daddr;
2340 case __constant_htons(ETH_P_IPV6):
2341 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2344 ip6 = (struct ipv6hdr *) (skb->data + nhoff);
2345 ip_proto = ip6->nexthdr;
2346 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2347 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2355 poff = proto_ports_offset(ip_proto);
2357 nhoff += ihl * 4 + poff;
2358 if (pskb_may_pull(skb, nhoff + 4)) {
2359 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2360 if (ports.v16[1] < ports.v16[0])
2361 swap(ports.v16[0], ports.v16[1]);
2365 /* get a consistent hash (same value on both flow directions) */
2369 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2376 EXPORT_SYMBOL(__skb_get_rxhash);
2380 /* One global table that all flow-based protocols share. */
2381 struct rps_sock_flow_table *rps_sock_flow_table __read_mostly;
2382 EXPORT_SYMBOL(rps_sock_flow_table);
2385 * get_rps_cpu is called from netif_receive_skb and returns the target
2386 * CPU from the RPS map of the receiving queue for a given skb.
2387 * rcu_read_lock must be held on entry.
2389 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2390 struct rps_dev_flow **rflowp)
2392 struct netdev_rx_queue *rxqueue;
2393 struct rps_map *map = NULL;
2394 struct rps_dev_flow_table *flow_table;
2395 struct rps_sock_flow_table *sock_flow_table;
2399 if (skb_rx_queue_recorded(skb)) {
2400 u16 index = skb_get_rx_queue(skb);
2401 if (unlikely(index >= dev->real_num_rx_queues)) {
2402 WARN_ONCE(dev->real_num_rx_queues > 1,
2403 "%s received packet on queue %u, but number "
2404 "of RX queues is %u\n",
2405 dev->name, index, dev->real_num_rx_queues);
2408 rxqueue = dev->_rx + index;
2412 if (rxqueue->rps_map) {
2413 map = rcu_dereference(rxqueue->rps_map);
2414 if (map && map->len == 1) {
2415 tcpu = map->cpus[0];
2416 if (cpu_online(tcpu))
2420 } else if (!rxqueue->rps_flow_table) {
2424 skb_reset_network_header(skb);
2425 if (!skb_get_rxhash(skb))
2428 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2429 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2430 if (flow_table && sock_flow_table) {
2432 struct rps_dev_flow *rflow;
2434 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2437 next_cpu = sock_flow_table->ents[skb->rxhash &
2438 sock_flow_table->mask];
2441 * If the desired CPU (where last recvmsg was done) is
2442 * different from current CPU (one in the rx-queue flow
2443 * table entry), switch if one of the following holds:
2444 * - Current CPU is unset (equal to RPS_NO_CPU).
2445 * - Current CPU is offline.
2446 * - The current CPU's queue tail has advanced beyond the
2447 * last packet that was enqueued using this table entry.
2448 * This guarantees that all previous packets for the flow
2449 * have been dequeued, thus preserving in order delivery.
2451 if (unlikely(tcpu != next_cpu) &&
2452 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2453 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2454 rflow->last_qtail)) >= 0)) {
2455 tcpu = rflow->cpu = next_cpu;
2456 if (tcpu != RPS_NO_CPU)
2457 rflow->last_qtail = per_cpu(softnet_data,
2458 tcpu).input_queue_head;
2460 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2468 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2470 if (cpu_online(tcpu)) {
2480 /* Called from hardirq (IPI) context */
2481 static void rps_trigger_softirq(void *data)
2483 struct softnet_data *sd = data;
2485 ____napi_schedule(sd, &sd->backlog);
2489 #endif /* CONFIG_RPS */
2492 * Check if this softnet_data structure is another cpu one
2493 * If yes, queue it to our IPI list and return 1
2496 static int rps_ipi_queued(struct softnet_data *sd)
2499 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2502 sd->rps_ipi_next = mysd->rps_ipi_list;
2503 mysd->rps_ipi_list = sd;
2505 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2508 #endif /* CONFIG_RPS */
2513 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2514 * queue (may be a remote CPU queue).
2516 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2517 unsigned int *qtail)
2519 struct softnet_data *sd;
2520 unsigned long flags;
2522 sd = &per_cpu(softnet_data, cpu);
2524 local_irq_save(flags);
2527 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2528 if (skb_queue_len(&sd->input_pkt_queue)) {
2530 __skb_queue_tail(&sd->input_pkt_queue, skb);
2531 input_queue_tail_incr_save(sd, qtail);
2533 local_irq_restore(flags);
2534 return NET_RX_SUCCESS;
2537 /* Schedule NAPI for backlog device
2538 * We can use non atomic operation since we own the queue lock
2540 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2541 if (!rps_ipi_queued(sd))
2542 ____napi_schedule(sd, &sd->backlog);
2550 local_irq_restore(flags);
2552 atomic_long_inc(&skb->dev->rx_dropped);
2558 * netif_rx - post buffer to the network code
2559 * @skb: buffer to post
2561 * This function receives a packet from a device driver and queues it for
2562 * the upper (protocol) levels to process. It always succeeds. The buffer
2563 * may be dropped during processing for congestion control or by the
2567 * NET_RX_SUCCESS (no congestion)
2568 * NET_RX_DROP (packet was dropped)
2572 int netif_rx(struct sk_buff *skb)
2576 /* if netpoll wants it, pretend we never saw it */
2577 if (netpoll_rx(skb))
2580 if (netdev_tstamp_prequeue)
2581 net_timestamp_check(skb);
2585 struct rps_dev_flow voidflow, *rflow = &voidflow;
2591 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2593 cpu = smp_processor_id();
2595 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2603 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2609 EXPORT_SYMBOL(netif_rx);
2611 int netif_rx_ni(struct sk_buff *skb)
2616 err = netif_rx(skb);
2617 if (local_softirq_pending())
2623 EXPORT_SYMBOL(netif_rx_ni);
2625 static void net_tx_action(struct softirq_action *h)
2627 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2629 if (sd->completion_queue) {
2630 struct sk_buff *clist;
2632 local_irq_disable();
2633 clist = sd->completion_queue;
2634 sd->completion_queue = NULL;
2638 struct sk_buff *skb = clist;
2639 clist = clist->next;
2641 WARN_ON(atomic_read(&skb->users));
2646 if (sd->output_queue) {
2649 local_irq_disable();
2650 head = sd->output_queue;
2651 sd->output_queue = NULL;
2652 sd->output_queue_tailp = &sd->output_queue;
2656 struct Qdisc *q = head;
2657 spinlock_t *root_lock;
2659 head = head->next_sched;
2661 root_lock = qdisc_lock(q);
2662 if (spin_trylock(root_lock)) {
2663 smp_mb__before_clear_bit();
2664 clear_bit(__QDISC_STATE_SCHED,
2667 spin_unlock(root_lock);
2669 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2671 __netif_reschedule(q);
2673 smp_mb__before_clear_bit();
2674 clear_bit(__QDISC_STATE_SCHED,
2682 static inline int deliver_skb(struct sk_buff *skb,
2683 struct packet_type *pt_prev,
2684 struct net_device *orig_dev)
2686 atomic_inc(&skb->users);
2687 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2690 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2691 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2692 /* This hook is defined here for ATM LANE */
2693 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2694 unsigned char *addr) __read_mostly;
2695 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2698 #ifdef CONFIG_NET_CLS_ACT
2699 /* TODO: Maybe we should just force sch_ingress to be compiled in
2700 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2701 * a compare and 2 stores extra right now if we dont have it on
2702 * but have CONFIG_NET_CLS_ACT
2703 * NOTE: This doesnt stop any functionality; if you dont have
2704 * the ingress scheduler, you just cant add policies on ingress.
2707 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
2709 struct net_device *dev = skb->dev;
2710 u32 ttl = G_TC_RTTL(skb->tc_verd);
2711 int result = TC_ACT_OK;
2714 if (unlikely(MAX_RED_LOOP < ttl++)) {
2715 if (net_ratelimit())
2716 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
2717 skb->skb_iif, dev->ifindex);
2721 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2722 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2725 if (q != &noop_qdisc) {
2726 spin_lock(qdisc_lock(q));
2727 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2728 result = qdisc_enqueue_root(skb, q);
2729 spin_unlock(qdisc_lock(q));
2735 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2736 struct packet_type **pt_prev,
2737 int *ret, struct net_device *orig_dev)
2739 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
2741 if (!rxq || rxq->qdisc == &noop_qdisc)
2745 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2749 switch (ing_filter(skb, rxq)) {
2763 * netif_nit_deliver - deliver received packets to network taps
2766 * This function is used to deliver incoming packets to network
2767 * taps. It should be used when the normal netif_receive_skb path
2768 * is bypassed, for example because of VLAN acceleration.
2770 void netif_nit_deliver(struct sk_buff *skb)
2772 struct packet_type *ptype;
2774 if (list_empty(&ptype_all))
2777 skb_reset_network_header(skb);
2778 skb_reset_transport_header(skb);
2779 skb->mac_len = skb->network_header - skb->mac_header;
2782 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2783 if (!ptype->dev || ptype->dev == skb->dev)
2784 deliver_skb(skb, ptype, skb->dev);
2790 * netdev_rx_handler_register - register receive handler
2791 * @dev: device to register a handler for
2792 * @rx_handler: receive handler to register
2793 * @rx_handler_data: data pointer that is used by rx handler
2795 * Register a receive hander for a device. This handler will then be
2796 * called from __netif_receive_skb. A negative errno code is returned
2799 * The caller must hold the rtnl_mutex.
2801 int netdev_rx_handler_register(struct net_device *dev,
2802 rx_handler_func_t *rx_handler,
2803 void *rx_handler_data)
2807 if (dev->rx_handler)
2810 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
2811 rcu_assign_pointer(dev->rx_handler, rx_handler);
2815 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
2818 * netdev_rx_handler_unregister - unregister receive handler
2819 * @dev: device to unregister a handler from
2821 * Unregister a receive hander from a device.
2823 * The caller must hold the rtnl_mutex.
2825 void netdev_rx_handler_unregister(struct net_device *dev)
2829 rcu_assign_pointer(dev->rx_handler, NULL);
2830 rcu_assign_pointer(dev->rx_handler_data, NULL);
2832 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
2834 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
2835 struct net_device *master)
2837 if (skb->pkt_type == PACKET_HOST) {
2838 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
2840 memcpy(dest, master->dev_addr, ETH_ALEN);
2844 /* On bonding slaves other than the currently active slave, suppress
2845 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2846 * ARP on active-backup slaves with arp_validate enabled.
2848 int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master)
2850 struct net_device *dev = skb->dev;
2852 if (master->priv_flags & IFF_MASTER_ARPMON)
2853 dev->last_rx = jiffies;
2855 if ((master->priv_flags & IFF_MASTER_ALB) &&
2856 (master->priv_flags & IFF_BRIDGE_PORT)) {
2857 /* Do address unmangle. The local destination address
2858 * will be always the one master has. Provides the right
2859 * functionality in a bridge.
2861 skb_bond_set_mac_by_master(skb, master);
2864 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
2865 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
2866 skb->protocol == __cpu_to_be16(ETH_P_ARP))
2869 if (master->priv_flags & IFF_MASTER_ALB) {
2870 if (skb->pkt_type != PACKET_BROADCAST &&
2871 skb->pkt_type != PACKET_MULTICAST)
2874 if (master->priv_flags & IFF_MASTER_8023AD &&
2875 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
2882 EXPORT_SYMBOL(__skb_bond_should_drop);
2884 static int __netif_receive_skb(struct sk_buff *skb)
2886 struct packet_type *ptype, *pt_prev;
2887 rx_handler_func_t *rx_handler;
2888 struct net_device *orig_dev;
2889 struct net_device *master;
2890 struct net_device *null_or_orig;
2891 struct net_device *orig_or_bond;
2892 int ret = NET_RX_DROP;
2895 if (!netdev_tstamp_prequeue)
2896 net_timestamp_check(skb);
2898 if (vlan_tx_tag_present(skb))
2899 vlan_hwaccel_do_receive(skb);
2901 /* if we've gotten here through NAPI, check netpoll */
2902 if (netpoll_receive_skb(skb))
2906 skb->skb_iif = skb->dev->ifindex;
2909 * bonding note: skbs received on inactive slaves should only
2910 * be delivered to pkt handlers that are exact matches. Also
2911 * the deliver_no_wcard flag will be set. If packet handlers
2912 * are sensitive to duplicate packets these skbs will need to
2913 * be dropped at the handler. The vlan accel path may have
2914 * already set the deliver_no_wcard flag.
2916 null_or_orig = NULL;
2917 orig_dev = skb->dev;
2918 master = ACCESS_ONCE(orig_dev->master);
2919 if (skb->deliver_no_wcard)
2920 null_or_orig = orig_dev;
2922 if (skb_bond_should_drop(skb, master)) {
2923 skb->deliver_no_wcard = 1;
2924 null_or_orig = orig_dev; /* deliver only exact match */
2929 __this_cpu_inc(softnet_data.processed);
2930 skb_reset_network_header(skb);
2931 skb_reset_transport_header(skb);
2932 skb->mac_len = skb->network_header - skb->mac_header;
2938 #ifdef CONFIG_NET_CLS_ACT
2939 if (skb->tc_verd & TC_NCLS) {
2940 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2945 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2946 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2947 ptype->dev == orig_dev) {
2949 ret = deliver_skb(skb, pt_prev, orig_dev);
2954 #ifdef CONFIG_NET_CLS_ACT
2955 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2961 /* Handle special case of bridge or macvlan */
2962 rx_handler = rcu_dereference(skb->dev->rx_handler);
2965 ret = deliver_skb(skb, pt_prev, orig_dev);
2968 skb = rx_handler(skb);
2974 * Make sure frames received on VLAN interfaces stacked on
2975 * bonding interfaces still make their way to any base bonding
2976 * device that may have registered for a specific ptype. The
2977 * handler may have to adjust skb->dev and orig_dev.
2979 orig_or_bond = orig_dev;
2980 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
2981 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
2982 orig_or_bond = vlan_dev_real_dev(skb->dev);
2985 type = skb->protocol;
2986 list_for_each_entry_rcu(ptype,
2987 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2988 if (ptype->type == type && (ptype->dev == null_or_orig ||
2989 ptype->dev == skb->dev || ptype->dev == orig_dev ||
2990 ptype->dev == orig_or_bond)) {
2992 ret = deliver_skb(skb, pt_prev, orig_dev);
2998 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3000 atomic_long_inc(&skb->dev->rx_dropped);
3002 /* Jamal, now you will not able to escape explaining
3003 * me how you were going to use this. :-)
3014 * netif_receive_skb - process receive buffer from network
3015 * @skb: buffer to process
3017 * netif_receive_skb() is the main receive data processing function.
3018 * It always succeeds. The buffer may be dropped during processing
3019 * for congestion control or by the protocol layers.
3021 * This function may only be called from softirq context and interrupts
3022 * should be enabled.
3024 * Return values (usually ignored):
3025 * NET_RX_SUCCESS: no congestion
3026 * NET_RX_DROP: packet was dropped
3028 int netif_receive_skb(struct sk_buff *skb)
3030 if (netdev_tstamp_prequeue)
3031 net_timestamp_check(skb);
3033 if (skb_defer_rx_timestamp(skb))
3034 return NET_RX_SUCCESS;
3038 struct rps_dev_flow voidflow, *rflow = &voidflow;
3043 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3046 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3050 ret = __netif_receive_skb(skb);
3056 return __netif_receive_skb(skb);
3059 EXPORT_SYMBOL(netif_receive_skb);
3061 /* Network device is going away, flush any packets still pending
3062 * Called with irqs disabled.
3064 static void flush_backlog(void *arg)
3066 struct net_device *dev = arg;
3067 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3068 struct sk_buff *skb, *tmp;
3071 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3072 if (skb->dev == dev) {
3073 __skb_unlink(skb, &sd->input_pkt_queue);
3075 input_queue_head_incr(sd);
3080 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3081 if (skb->dev == dev) {
3082 __skb_unlink(skb, &sd->process_queue);
3084 input_queue_head_incr(sd);
3089 static int napi_gro_complete(struct sk_buff *skb)
3091 struct packet_type *ptype;
3092 __be16 type = skb->protocol;
3093 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3096 if (NAPI_GRO_CB(skb)->count == 1) {
3097 skb_shinfo(skb)->gso_size = 0;
3102 list_for_each_entry_rcu(ptype, head, list) {
3103 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3106 err = ptype->gro_complete(skb);
3112 WARN_ON(&ptype->list == head);
3114 return NET_RX_SUCCESS;
3118 return netif_receive_skb(skb);
3121 inline void napi_gro_flush(struct napi_struct *napi)
3123 struct sk_buff *skb, *next;
3125 for (skb = napi->gro_list; skb; skb = next) {
3128 napi_gro_complete(skb);
3131 napi->gro_count = 0;
3132 napi->gro_list = NULL;
3134 EXPORT_SYMBOL(napi_gro_flush);
3136 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3138 struct sk_buff **pp = NULL;
3139 struct packet_type *ptype;
3140 __be16 type = skb->protocol;
3141 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3144 enum gro_result ret;
3146 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3149 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3153 list_for_each_entry_rcu(ptype, head, list) {
3154 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3157 skb_set_network_header(skb, skb_gro_offset(skb));
3158 mac_len = skb->network_header - skb->mac_header;
3159 skb->mac_len = mac_len;
3160 NAPI_GRO_CB(skb)->same_flow = 0;
3161 NAPI_GRO_CB(skb)->flush = 0;
3162 NAPI_GRO_CB(skb)->free = 0;
3164 pp = ptype->gro_receive(&napi->gro_list, skb);
3169 if (&ptype->list == head)
3172 same_flow = NAPI_GRO_CB(skb)->same_flow;
3173 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3176 struct sk_buff *nskb = *pp;
3180 napi_gro_complete(nskb);
3187 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3191 NAPI_GRO_CB(skb)->count = 1;
3192 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3193 skb->next = napi->gro_list;
3194 napi->gro_list = skb;
3198 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3199 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3201 BUG_ON(skb->end - skb->tail < grow);
3203 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3206 skb->data_len -= grow;
3208 skb_shinfo(skb)->frags[0].page_offset += grow;
3209 skb_shinfo(skb)->frags[0].size -= grow;
3211 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3212 put_page(skb_shinfo(skb)->frags[0].page);
3213 memmove(skb_shinfo(skb)->frags,
3214 skb_shinfo(skb)->frags + 1,
3215 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3226 EXPORT_SYMBOL(dev_gro_receive);
3228 static inline gro_result_t
3229 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3233 for (p = napi->gro_list; p; p = p->next) {
3234 unsigned long diffs;
3236 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3237 diffs |= compare_ether_header(skb_mac_header(p),
3238 skb_gro_mac_header(skb));
3239 NAPI_GRO_CB(p)->same_flow = !diffs;
3240 NAPI_GRO_CB(p)->flush = 0;
3243 return dev_gro_receive(napi, skb);
3246 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3250 if (netif_receive_skb(skb))
3255 case GRO_MERGED_FREE:
3266 EXPORT_SYMBOL(napi_skb_finish);
3268 void skb_gro_reset_offset(struct sk_buff *skb)
3270 NAPI_GRO_CB(skb)->data_offset = 0;
3271 NAPI_GRO_CB(skb)->frag0 = NULL;
3272 NAPI_GRO_CB(skb)->frag0_len = 0;
3274 if (skb->mac_header == skb->tail &&
3275 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3276 NAPI_GRO_CB(skb)->frag0 =
3277 page_address(skb_shinfo(skb)->frags[0].page) +
3278 skb_shinfo(skb)->frags[0].page_offset;
3279 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3282 EXPORT_SYMBOL(skb_gro_reset_offset);
3284 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3286 skb_gro_reset_offset(skb);
3288 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3290 EXPORT_SYMBOL(napi_gro_receive);
3292 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3294 __skb_pull(skb, skb_headlen(skb));
3295 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3299 EXPORT_SYMBOL(napi_reuse_skb);
3301 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3303 struct sk_buff *skb = napi->skb;
3306 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3312 EXPORT_SYMBOL(napi_get_frags);
3314 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3320 skb->protocol = eth_type_trans(skb, skb->dev);
3322 if (ret == GRO_HELD)
3323 skb_gro_pull(skb, -ETH_HLEN);
3324 else if (netif_receive_skb(skb))
3329 case GRO_MERGED_FREE:
3330 napi_reuse_skb(napi, skb);
3339 EXPORT_SYMBOL(napi_frags_finish);
3341 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3343 struct sk_buff *skb = napi->skb;
3350 skb_reset_mac_header(skb);
3351 skb_gro_reset_offset(skb);
3353 off = skb_gro_offset(skb);
3354 hlen = off + sizeof(*eth);
3355 eth = skb_gro_header_fast(skb, off);
3356 if (skb_gro_header_hard(skb, hlen)) {
3357 eth = skb_gro_header_slow(skb, hlen, off);
3358 if (unlikely(!eth)) {
3359 napi_reuse_skb(napi, skb);
3365 skb_gro_pull(skb, sizeof(*eth));
3368 * This works because the only protocols we care about don't require
3369 * special handling. We'll fix it up properly at the end.
3371 skb->protocol = eth->h_proto;
3376 EXPORT_SYMBOL(napi_frags_skb);
3378 gro_result_t napi_gro_frags(struct napi_struct *napi)
3380 struct sk_buff *skb = napi_frags_skb(napi);
3385 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3387 EXPORT_SYMBOL(napi_gro_frags);
3390 * net_rps_action sends any pending IPI's for rps.
3391 * Note: called with local irq disabled, but exits with local irq enabled.
3393 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3396 struct softnet_data *remsd = sd->rps_ipi_list;
3399 sd->rps_ipi_list = NULL;
3403 /* Send pending IPI's to kick RPS processing on remote cpus. */
3405 struct softnet_data *next = remsd->rps_ipi_next;
3407 if (cpu_online(remsd->cpu))
3408 __smp_call_function_single(remsd->cpu,
3417 static int process_backlog(struct napi_struct *napi, int quota)
3420 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3423 /* Check if we have pending ipi, its better to send them now,
3424 * not waiting net_rx_action() end.
3426 if (sd->rps_ipi_list) {
3427 local_irq_disable();
3428 net_rps_action_and_irq_enable(sd);
3431 napi->weight = weight_p;
3432 local_irq_disable();
3433 while (work < quota) {
3434 struct sk_buff *skb;
3437 while ((skb = __skb_dequeue(&sd->process_queue))) {
3439 __netif_receive_skb(skb);
3440 local_irq_disable();
3441 input_queue_head_incr(sd);
3442 if (++work >= quota) {
3449 qlen = skb_queue_len(&sd->input_pkt_queue);
3451 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3452 &sd->process_queue);
3454 if (qlen < quota - work) {
3456 * Inline a custom version of __napi_complete().
3457 * only current cpu owns and manipulates this napi,
3458 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3459 * we can use a plain write instead of clear_bit(),
3460 * and we dont need an smp_mb() memory barrier.
3462 list_del(&napi->poll_list);
3465 quota = work + qlen;
3475 * __napi_schedule - schedule for receive
3476 * @n: entry to schedule
3478 * The entry's receive function will be scheduled to run
3480 void __napi_schedule(struct napi_struct *n)
3482 unsigned long flags;
3484 local_irq_save(flags);
3485 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3486 local_irq_restore(flags);
3488 EXPORT_SYMBOL(__napi_schedule);
3490 void __napi_complete(struct napi_struct *n)
3492 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3493 BUG_ON(n->gro_list);
3495 list_del(&n->poll_list);
3496 smp_mb__before_clear_bit();
3497 clear_bit(NAPI_STATE_SCHED, &n->state);
3499 EXPORT_SYMBOL(__napi_complete);
3501 void napi_complete(struct napi_struct *n)
3503 unsigned long flags;
3506 * don't let napi dequeue from the cpu poll list
3507 * just in case its running on a different cpu
3509 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3513 local_irq_save(flags);
3515 local_irq_restore(flags);
3517 EXPORT_SYMBOL(napi_complete);
3519 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3520 int (*poll)(struct napi_struct *, int), int weight)
3522 INIT_LIST_HEAD(&napi->poll_list);
3523 napi->gro_count = 0;
3524 napi->gro_list = NULL;
3527 napi->weight = weight;
3528 list_add(&napi->dev_list, &dev->napi_list);
3530 #ifdef CONFIG_NETPOLL
3531 spin_lock_init(&napi->poll_lock);
3532 napi->poll_owner = -1;
3534 set_bit(NAPI_STATE_SCHED, &napi->state);
3536 EXPORT_SYMBOL(netif_napi_add);
3538 void netif_napi_del(struct napi_struct *napi)
3540 struct sk_buff *skb, *next;
3542 list_del_init(&napi->dev_list);
3543 napi_free_frags(napi);
3545 for (skb = napi->gro_list; skb; skb = next) {
3551 napi->gro_list = NULL;
3552 napi->gro_count = 0;
3554 EXPORT_SYMBOL(netif_napi_del);
3556 static void net_rx_action(struct softirq_action *h)
3558 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3559 unsigned long time_limit = jiffies + 2;
3560 int budget = netdev_budget;
3563 local_irq_disable();
3565 while (!list_empty(&sd->poll_list)) {
3566 struct napi_struct *n;
3569 /* If softirq window is exhuasted then punt.
3570 * Allow this to run for 2 jiffies since which will allow
3571 * an average latency of 1.5/HZ.
3573 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3578 /* Even though interrupts have been re-enabled, this
3579 * access is safe because interrupts can only add new
3580 * entries to the tail of this list, and only ->poll()
3581 * calls can remove this head entry from the list.
3583 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3585 have = netpoll_poll_lock(n);
3589 /* This NAPI_STATE_SCHED test is for avoiding a race
3590 * with netpoll's poll_napi(). Only the entity which
3591 * obtains the lock and sees NAPI_STATE_SCHED set will
3592 * actually make the ->poll() call. Therefore we avoid
3593 * accidently calling ->poll() when NAPI is not scheduled.
3596 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3597 work = n->poll(n, weight);
3601 WARN_ON_ONCE(work > weight);
3605 local_irq_disable();
3607 /* Drivers must not modify the NAPI state if they
3608 * consume the entire weight. In such cases this code
3609 * still "owns" the NAPI instance and therefore can
3610 * move the instance around on the list at-will.
3612 if (unlikely(work == weight)) {
3613 if (unlikely(napi_disable_pending(n))) {
3616 local_irq_disable();
3618 list_move_tail(&n->poll_list, &sd->poll_list);
3621 netpoll_poll_unlock(have);
3624 net_rps_action_and_irq_enable(sd);
3626 #ifdef CONFIG_NET_DMA
3628 * There may not be any more sk_buffs coming right now, so push
3629 * any pending DMA copies to hardware
3631 dma_issue_pending_all();
3638 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3642 static gifconf_func_t *gifconf_list[NPROTO];
3645 * register_gifconf - register a SIOCGIF handler
3646 * @family: Address family
3647 * @gifconf: Function handler
3649 * Register protocol dependent address dumping routines. The handler
3650 * that is passed must not be freed or reused until it has been replaced
3651 * by another handler.
3653 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3655 if (family >= NPROTO)
3657 gifconf_list[family] = gifconf;
3660 EXPORT_SYMBOL(register_gifconf);
3664 * Map an interface index to its name (SIOCGIFNAME)
3668 * We need this ioctl for efficient implementation of the
3669 * if_indextoname() function required by the IPv6 API. Without
3670 * it, we would have to search all the interfaces to find a
3674 static int dev_ifname(struct