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 <linux/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/sched/mm.h>
85 #include <linux/mutex.h>
86 #include <linux/rwsem.h>
87 #include <linux/string.h>
89 #include <linux/socket.h>
90 #include <linux/sockios.h>
91 #include <linux/errno.h>
92 #include <linux/interrupt.h>
93 #include <linux/if_ether.h>
94 #include <linux/netdevice.h>
95 #include <linux/etherdevice.h>
96 #include <linux/ethtool.h>
97 #include <linux/skbuff.h>
98 #include <linux/bpf.h>
99 #include <linux/bpf_trace.h>
100 #include <net/net_namespace.h>
101 #include <net/sock.h>
102 #include <net/busy_poll.h>
103 #include <linux/rtnetlink.h>
104 #include <linux/stat.h>
106 #include <net/dst_metadata.h>
107 #include <net/pkt_sched.h>
108 #include <net/pkt_cls.h>
109 #include <net/checksum.h>
110 #include <net/xfrm.h>
111 #include <linux/highmem.h>
112 #include <linux/init.h>
113 #include <linux/module.h>
114 #include <linux/netpoll.h>
115 #include <linux/rcupdate.h>
116 #include <linux/delay.h>
117 #include <net/iw_handler.h>
118 #include <asm/current.h>
119 #include <linux/audit.h>
120 #include <linux/dmaengine.h>
121 #include <linux/err.h>
122 #include <linux/ctype.h>
123 #include <linux/if_arp.h>
124 #include <linux/if_vlan.h>
125 #include <linux/ip.h>
127 #include <net/mpls.h>
128 #include <linux/ipv6.h>
129 #include <linux/in.h>
130 #include <linux/jhash.h>
131 #include <linux/random.h>
132 #include <trace/events/napi.h>
133 #include <trace/events/net.h>
134 #include <trace/events/skb.h>
135 #include <linux/pci.h>
136 #include <linux/inetdevice.h>
137 #include <linux/cpu_rmap.h>
138 #include <linux/static_key.h>
139 #include <linux/hashtable.h>
140 #include <linux/vmalloc.h>
141 #include <linux/if_macvlan.h>
142 #include <linux/errqueue.h>
143 #include <linux/hrtimer.h>
144 #include <linux/netfilter_ingress.h>
145 #include <linux/crash_dump.h>
146 #include <linux/sctp.h>
147 #include <net/udp_tunnel.h>
148 #include <linux/net_namespace.h>
150 #include "net-sysfs.h"
152 #define MAX_GRO_SKBS 8
153 #define MAX_NEST_DEV 8
155 /* This should be increased if a protocol with a bigger head is added. */
156 #define GRO_MAX_HEAD (MAX_HEADER + 128)
158 static DEFINE_SPINLOCK(ptype_lock);
159 static DEFINE_SPINLOCK(offload_lock);
160 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
161 struct list_head ptype_all __read_mostly; /* Taps */
162 static struct list_head offload_base __read_mostly;
164 static int netif_rx_internal(struct sk_buff *skb);
165 static int call_netdevice_notifiers_info(unsigned long val,
166 struct netdev_notifier_info *info);
167 static struct napi_struct *napi_by_id(unsigned int napi_id);
170 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
173 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
175 * Writers must hold the rtnl semaphore while they loop through the
176 * dev_base_head list, and hold dev_base_lock for writing when they do the
177 * actual updates. This allows pure readers to access the list even
178 * while a writer is preparing to update it.
180 * To put it another way, dev_base_lock is held for writing only to
181 * protect against pure readers; the rtnl semaphore provides the
182 * protection against other writers.
184 * See, for example usages, register_netdevice() and
185 * unregister_netdevice(), which must be called with the rtnl
188 DEFINE_RWLOCK(dev_base_lock);
189 EXPORT_SYMBOL(dev_base_lock);
191 static DEFINE_MUTEX(ifalias_mutex);
193 /* protects napi_hash addition/deletion and napi_gen_id */
194 static DEFINE_SPINLOCK(napi_hash_lock);
196 static unsigned int napi_gen_id = NR_CPUS;
197 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
199 static DECLARE_RWSEM(devnet_rename_sem);
201 static inline void dev_base_seq_inc(struct net *net)
203 while (++net->dev_base_seq == 0)
207 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
209 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
211 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
214 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
216 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
219 static inline void rps_lock(struct softnet_data *sd)
222 spin_lock(&sd->input_pkt_queue.lock);
226 static inline void rps_unlock(struct softnet_data *sd)
229 spin_unlock(&sd->input_pkt_queue.lock);
233 /* Device list insertion */
234 static void list_netdevice(struct net_device *dev)
236 struct net *net = dev_net(dev);
240 write_lock_bh(&dev_base_lock);
241 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
242 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
243 hlist_add_head_rcu(&dev->index_hlist,
244 dev_index_hash(net, dev->ifindex));
245 write_unlock_bh(&dev_base_lock);
247 dev_base_seq_inc(net);
250 /* Device list removal
251 * caller must respect a RCU grace period before freeing/reusing dev
253 static void unlist_netdevice(struct net_device *dev)
257 /* Unlink dev from the device chain */
258 write_lock_bh(&dev_base_lock);
259 list_del_rcu(&dev->dev_list);
260 hlist_del_rcu(&dev->name_hlist);
261 hlist_del_rcu(&dev->index_hlist);
262 write_unlock_bh(&dev_base_lock);
264 dev_base_seq_inc(dev_net(dev));
271 static RAW_NOTIFIER_HEAD(netdev_chain);
274 * Device drivers call our routines to queue packets here. We empty the
275 * queue in the local softnet handler.
278 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
279 EXPORT_PER_CPU_SYMBOL(softnet_data);
281 #ifdef CONFIG_LOCKDEP
283 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
284 * according to dev->type
286 static const unsigned short netdev_lock_type[] = {
287 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
288 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
289 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
290 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
291 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
292 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
293 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
294 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
295 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
296 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
297 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
298 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
299 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
300 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
301 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
303 static const char *const netdev_lock_name[] = {
304 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
305 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
306 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
307 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
308 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
309 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
310 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
311 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
312 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
313 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
314 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
315 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
316 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
317 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
318 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
320 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
321 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
323 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
327 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
328 if (netdev_lock_type[i] == dev_type)
330 /* the last key is used by default */
331 return ARRAY_SIZE(netdev_lock_type) - 1;
334 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
335 unsigned short dev_type)
339 i = netdev_lock_pos(dev_type);
340 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
341 netdev_lock_name[i]);
344 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
348 i = netdev_lock_pos(dev->type);
349 lockdep_set_class_and_name(&dev->addr_list_lock,
350 &netdev_addr_lock_key[i],
351 netdev_lock_name[i]);
354 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
355 unsigned short dev_type)
358 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
363 /*******************************************************************************
365 * Protocol management and registration routines
367 *******************************************************************************/
371 * Add a protocol ID to the list. Now that the input handler is
372 * smarter we can dispense with all the messy stuff that used to be
375 * BEWARE!!! Protocol handlers, mangling input packets,
376 * MUST BE last in hash buckets and checking protocol handlers
377 * MUST start from promiscuous ptype_all chain in net_bh.
378 * It is true now, do not change it.
379 * Explanation follows: if protocol handler, mangling packet, will
380 * be the first on list, it is not able to sense, that packet
381 * is cloned and should be copied-on-write, so that it will
382 * change it and subsequent readers will get broken packet.
386 static inline struct list_head *ptype_head(const struct packet_type *pt)
388 if (pt->type == htons(ETH_P_ALL))
389 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
391 return pt->dev ? &pt->dev->ptype_specific :
392 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
396 * dev_add_pack - add packet handler
397 * @pt: packet type declaration
399 * Add a protocol handler to the networking stack. The passed &packet_type
400 * is linked into kernel lists and may not be freed until it has been
401 * removed from the kernel lists.
403 * This call does not sleep therefore it can not
404 * guarantee all CPU's that are in middle of receiving packets
405 * will see the new packet type (until the next received packet).
408 void dev_add_pack(struct packet_type *pt)
410 struct list_head *head = ptype_head(pt);
412 spin_lock(&ptype_lock);
413 list_add_rcu(&pt->list, head);
414 spin_unlock(&ptype_lock);
416 EXPORT_SYMBOL(dev_add_pack);
419 * __dev_remove_pack - remove packet handler
420 * @pt: packet type declaration
422 * Remove a protocol handler that was previously added to the kernel
423 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
424 * from the kernel lists and can be freed or reused once this function
427 * The packet type might still be in use by receivers
428 * and must not be freed until after all the CPU's have gone
429 * through a quiescent state.
431 void __dev_remove_pack(struct packet_type *pt)
433 struct list_head *head = ptype_head(pt);
434 struct packet_type *pt1;
436 spin_lock(&ptype_lock);
438 list_for_each_entry(pt1, head, list) {
440 list_del_rcu(&pt->list);
445 pr_warn("dev_remove_pack: %p not found\n", pt);
447 spin_unlock(&ptype_lock);
449 EXPORT_SYMBOL(__dev_remove_pack);
452 * dev_remove_pack - remove packet handler
453 * @pt: packet type declaration
455 * Remove a protocol handler that was previously added to the kernel
456 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
457 * from the kernel lists and can be freed or reused once this function
460 * This call sleeps to guarantee that no CPU is looking at the packet
463 void dev_remove_pack(struct packet_type *pt)
465 __dev_remove_pack(pt);
469 EXPORT_SYMBOL(dev_remove_pack);
473 * dev_add_offload - register offload handlers
474 * @po: protocol offload declaration
476 * Add protocol offload handlers to the networking stack. The passed
477 * &proto_offload is linked into kernel lists and may not be freed until
478 * it has been removed from the kernel lists.
480 * This call does not sleep therefore it can not
481 * guarantee all CPU's that are in middle of receiving packets
482 * will see the new offload handlers (until the next received packet).
484 void dev_add_offload(struct packet_offload *po)
486 struct packet_offload *elem;
488 spin_lock(&offload_lock);
489 list_for_each_entry(elem, &offload_base, list) {
490 if (po->priority < elem->priority)
493 list_add_rcu(&po->list, elem->list.prev);
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 static void __dev_remove_offload(struct packet_offload *po)
513 struct list_head *head = &offload_base;
514 struct packet_offload *po1;
516 spin_lock(&offload_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(&offload_lock);
531 * dev_remove_offload - remove packet offload handler
532 * @po: packet offload declaration
534 * Remove a packet offload handler that was previously added to the kernel
535 * offload handlers by dev_add_offload(). The passed &offload_type is
536 * removed from the kernel lists and can be freed or reused once this
539 * This call sleeps to guarantee that no CPU is looking at the packet
542 void dev_remove_offload(struct packet_offload *po)
544 __dev_remove_offload(po);
548 EXPORT_SYMBOL(dev_remove_offload);
550 /******************************************************************************
552 * Device Boot-time Settings Routines
554 ******************************************************************************/
556 /* Boot time configuration table */
557 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
560 * netdev_boot_setup_add - add new setup entry
561 * @name: name of the device
562 * @map: configured settings for the device
564 * Adds new setup entry to the dev_boot_setup list. The function
565 * returns 0 on error and 1 on success. This is a generic routine to
568 static int netdev_boot_setup_add(char *name, struct ifmap *map)
570 struct netdev_boot_setup *s;
574 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
575 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
576 memset(s[i].name, 0, sizeof(s[i].name));
577 strlcpy(s[i].name, name, IFNAMSIZ);
578 memcpy(&s[i].map, map, sizeof(s[i].map));
583 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
587 * netdev_boot_setup_check - check boot time settings
588 * @dev: the netdevice
590 * Check boot time settings for the device.
591 * The found settings are set for the device to be used
592 * later in the device probing.
593 * Returns 0 if no settings found, 1 if they are.
595 int netdev_boot_setup_check(struct net_device *dev)
597 struct netdev_boot_setup *s = dev_boot_setup;
600 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
601 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
602 !strcmp(dev->name, s[i].name)) {
603 dev->irq = s[i].map.irq;
604 dev->base_addr = s[i].map.base_addr;
605 dev->mem_start = s[i].map.mem_start;
606 dev->mem_end = s[i].map.mem_end;
612 EXPORT_SYMBOL(netdev_boot_setup_check);
616 * netdev_boot_base - get address from boot time settings
617 * @prefix: prefix for network device
618 * @unit: id for network device
620 * Check boot time settings for the base address of device.
621 * The found settings are set for the device to be used
622 * later in the device probing.
623 * Returns 0 if no settings found.
625 unsigned long netdev_boot_base(const char *prefix, int unit)
627 const struct netdev_boot_setup *s = dev_boot_setup;
631 sprintf(name, "%s%d", prefix, unit);
634 * If device already registered then return base of 1
635 * to indicate not to probe for this interface
637 if (__dev_get_by_name(&init_net, name))
640 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
641 if (!strcmp(name, s[i].name))
642 return s[i].map.base_addr;
647 * Saves at boot time configured settings for any netdevice.
649 int __init netdev_boot_setup(char *str)
654 str = get_options(str, ARRAY_SIZE(ints), ints);
659 memset(&map, 0, sizeof(map));
663 map.base_addr = ints[2];
665 map.mem_start = ints[3];
667 map.mem_end = ints[4];
669 /* Add new entry to the list */
670 return netdev_boot_setup_add(str, &map);
673 __setup("netdev=", netdev_boot_setup);
675 /*******************************************************************************
677 * Device Interface Subroutines
679 *******************************************************************************/
682 * dev_get_iflink - get 'iflink' value of a interface
683 * @dev: targeted interface
685 * Indicates the ifindex the interface is linked to.
686 * Physical interfaces have the same 'ifindex' and 'iflink' values.
689 int dev_get_iflink(const struct net_device *dev)
691 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
692 return dev->netdev_ops->ndo_get_iflink(dev);
696 EXPORT_SYMBOL(dev_get_iflink);
699 * dev_fill_metadata_dst - Retrieve tunnel egress information.
700 * @dev: targeted interface
703 * For better visibility of tunnel traffic OVS needs to retrieve
704 * egress tunnel information for a packet. Following API allows
705 * user to get this info.
707 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
709 struct ip_tunnel_info *info;
711 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
714 info = skb_tunnel_info_unclone(skb);
717 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
720 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
722 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
725 * __dev_get_by_name - find a device by its name
726 * @net: the applicable net namespace
727 * @name: name to find
729 * Find an interface by name. Must be called under RTNL semaphore
730 * or @dev_base_lock. If the name is found a pointer to the device
731 * is returned. If the name is not found then %NULL is returned. The
732 * reference counters are not incremented so the caller must be
733 * careful with locks.
736 struct net_device *__dev_get_by_name(struct net *net, const char *name)
738 struct net_device *dev;
739 struct hlist_head *head = dev_name_hash(net, name);
741 hlist_for_each_entry(dev, head, name_hlist)
742 if (!strncmp(dev->name, name, IFNAMSIZ))
747 EXPORT_SYMBOL(__dev_get_by_name);
750 * dev_get_by_name_rcu - find a device by its name
751 * @net: the applicable net namespace
752 * @name: name to find
754 * Find an interface by name.
755 * If the name is found a pointer to the device is returned.
756 * If the name is not found then %NULL is returned.
757 * The reference counters are not incremented so the caller must be
758 * careful with locks. The caller must hold RCU lock.
761 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
763 struct net_device *dev;
764 struct hlist_head *head = dev_name_hash(net, name);
766 hlist_for_each_entry_rcu(dev, head, name_hlist)
767 if (!strncmp(dev->name, name, IFNAMSIZ))
772 EXPORT_SYMBOL(dev_get_by_name_rcu);
775 * dev_get_by_name - find a device by its name
776 * @net: the applicable net namespace
777 * @name: name to find
779 * Find an interface by name. This can be called from any
780 * context and does its own locking. The returned handle has
781 * the usage count incremented and the caller must use dev_put() to
782 * release it when it is no longer needed. %NULL is returned if no
783 * matching device is found.
786 struct net_device *dev_get_by_name(struct net *net, const char *name)
788 struct net_device *dev;
791 dev = dev_get_by_name_rcu(net, name);
797 EXPORT_SYMBOL(dev_get_by_name);
800 * __dev_get_by_index - find a device by its ifindex
801 * @net: the applicable net namespace
802 * @ifindex: index of device
804 * Search for an interface by index. Returns %NULL if the device
805 * is not found or a pointer to the device. The device has not
806 * had its reference counter increased so the caller must be careful
807 * about locking. The caller must hold either the RTNL semaphore
811 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
813 struct net_device *dev;
814 struct hlist_head *head = dev_index_hash(net, ifindex);
816 hlist_for_each_entry(dev, head, index_hlist)
817 if (dev->ifindex == ifindex)
822 EXPORT_SYMBOL(__dev_get_by_index);
825 * dev_get_by_index_rcu - find a device by its ifindex
826 * @net: the applicable net namespace
827 * @ifindex: index of device
829 * Search for an interface by index. Returns %NULL if the device
830 * is not found or a pointer to the device. The device has not
831 * had its reference counter increased so the caller must be careful
832 * about locking. The caller must hold RCU lock.
835 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
837 struct net_device *dev;
838 struct hlist_head *head = dev_index_hash(net, ifindex);
840 hlist_for_each_entry_rcu(dev, head, index_hlist)
841 if (dev->ifindex == ifindex)
846 EXPORT_SYMBOL(dev_get_by_index_rcu);
850 * dev_get_by_index - find a device by its ifindex
851 * @net: the applicable net namespace
852 * @ifindex: index of device
854 * Search for an interface by index. Returns NULL if the device
855 * is not found or a pointer to the device. The device returned has
856 * had a reference added and the pointer is safe until the user calls
857 * dev_put to indicate they have finished with it.
860 struct net_device *dev_get_by_index(struct net *net, int ifindex)
862 struct net_device *dev;
865 dev = dev_get_by_index_rcu(net, ifindex);
871 EXPORT_SYMBOL(dev_get_by_index);
874 * dev_get_by_napi_id - find a device by napi_id
875 * @napi_id: ID of the NAPI struct
877 * Search for an interface by NAPI ID. Returns %NULL if the device
878 * is not found or a pointer to the device. The device has not had
879 * its reference counter increased so the caller must be careful
880 * about locking. The caller must hold RCU lock.
883 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
885 struct napi_struct *napi;
887 WARN_ON_ONCE(!rcu_read_lock_held());
889 if (napi_id < MIN_NAPI_ID)
892 napi = napi_by_id(napi_id);
894 return napi ? napi->dev : NULL;
896 EXPORT_SYMBOL(dev_get_by_napi_id);
899 * netdev_get_name - get a netdevice name, knowing its ifindex.
900 * @net: network namespace
901 * @name: a pointer to the buffer where the name will be stored.
902 * @ifindex: the ifindex of the interface to get the name from.
904 int netdev_get_name(struct net *net, char *name, int ifindex)
906 struct net_device *dev;
909 down_read(&devnet_rename_sem);
912 dev = dev_get_by_index_rcu(net, ifindex);
918 strcpy(name, dev->name);
923 up_read(&devnet_rename_sem);
928 * dev_getbyhwaddr_rcu - find a device by its hardware address
929 * @net: the applicable net namespace
930 * @type: media type of device
931 * @ha: hardware address
933 * Search for an interface by MAC address. Returns NULL if the device
934 * is not found or a pointer to the device.
935 * The caller must hold RCU or RTNL.
936 * The returned device has not had its ref count increased
937 * and the caller must therefore be careful about locking
941 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
944 struct net_device *dev;
946 for_each_netdev_rcu(net, dev)
947 if (dev->type == type &&
948 !memcmp(dev->dev_addr, ha, dev->addr_len))
953 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
955 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
957 struct net_device *dev;
960 for_each_netdev(net, dev)
961 if (dev->type == type)
966 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
968 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
970 struct net_device *dev, *ret = NULL;
973 for_each_netdev_rcu(net, dev)
974 if (dev->type == type) {
982 EXPORT_SYMBOL(dev_getfirstbyhwtype);
985 * __dev_get_by_flags - find any device with given flags
986 * @net: the applicable net namespace
987 * @if_flags: IFF_* values
988 * @mask: bitmask of bits in if_flags to check
990 * Search for any interface with the given flags. Returns NULL if a device
991 * is not found or a pointer to the device. Must be called inside
992 * rtnl_lock(), and result refcount is unchanged.
995 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
998 struct net_device *dev, *ret;
1003 for_each_netdev(net, dev) {
1004 if (((dev->flags ^ if_flags) & mask) == 0) {
1011 EXPORT_SYMBOL(__dev_get_by_flags);
1014 * dev_valid_name - check if name is okay for network device
1015 * @name: name string
1017 * Network device names need to be valid file names to
1018 * to allow sysfs to work. We also disallow any kind of
1021 bool dev_valid_name(const char *name)
1025 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1027 if (!strcmp(name, ".") || !strcmp(name, ".."))
1031 if (*name == '/' || *name == ':' || isspace(*name))
1037 EXPORT_SYMBOL(dev_valid_name);
1040 * __dev_alloc_name - allocate a name for a device
1041 * @net: network namespace to allocate the device name in
1042 * @name: name format string
1043 * @buf: scratch buffer and result name string
1045 * Passed a format string - eg "lt%d" it will try and find a suitable
1046 * id. It scans list of devices to build up a free map, then chooses
1047 * the first empty slot. The caller must hold the dev_base or rtnl lock
1048 * while allocating the name and adding the device in order to avoid
1050 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1051 * Returns the number of the unit assigned or a negative errno code.
1054 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1058 const int max_netdevices = 8*PAGE_SIZE;
1059 unsigned long *inuse;
1060 struct net_device *d;
1062 if (!dev_valid_name(name))
1065 p = strchr(name, '%');
1068 * Verify the string as this thing may have come from
1069 * the user. There must be either one "%d" and no other "%"
1072 if (p[1] != 'd' || strchr(p + 2, '%'))
1075 /* Use one page as a bit array of possible slots */
1076 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1080 for_each_netdev(net, d) {
1081 if (!sscanf(d->name, name, &i))
1083 if (i < 0 || i >= max_netdevices)
1086 /* avoid cases where sscanf is not exact inverse of printf */
1087 snprintf(buf, IFNAMSIZ, name, i);
1088 if (!strncmp(buf, d->name, IFNAMSIZ))
1092 i = find_first_zero_bit(inuse, max_netdevices);
1093 free_page((unsigned long) inuse);
1096 snprintf(buf, IFNAMSIZ, name, i);
1097 if (!__dev_get_by_name(net, buf))
1100 /* It is possible to run out of possible slots
1101 * when the name is long and there isn't enough space left
1102 * for the digits, or if all bits are used.
1107 static int dev_alloc_name_ns(struct net *net,
1108 struct net_device *dev,
1115 ret = __dev_alloc_name(net, name, buf);
1117 strlcpy(dev->name, buf, IFNAMSIZ);
1122 * dev_alloc_name - allocate a name for a device
1124 * @name: name format string
1126 * Passed a format string - eg "lt%d" it will try and find a suitable
1127 * id. It scans list of devices to build up a free map, then chooses
1128 * the first empty slot. The caller must hold the dev_base or rtnl lock
1129 * while allocating the name and adding the device in order to avoid
1131 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1132 * Returns the number of the unit assigned or a negative errno code.
1135 int dev_alloc_name(struct net_device *dev, const char *name)
1137 return dev_alloc_name_ns(dev_net(dev), dev, name);
1139 EXPORT_SYMBOL(dev_alloc_name);
1141 int dev_get_valid_name(struct net *net, struct net_device *dev,
1146 if (!dev_valid_name(name))
1149 if (strchr(name, '%'))
1150 return dev_alloc_name_ns(net, dev, name);
1151 else if (__dev_get_by_name(net, name))
1153 else if (dev->name != name)
1154 strlcpy(dev->name, name, IFNAMSIZ);
1158 EXPORT_SYMBOL(dev_get_valid_name);
1161 * dev_change_name - change name of a device
1163 * @newname: name (or format string) must be at least IFNAMSIZ
1165 * Change name of a device, can pass format strings "eth%d".
1168 int dev_change_name(struct net_device *dev, const char *newname)
1170 unsigned char old_assign_type;
1171 char oldname[IFNAMSIZ];
1177 BUG_ON(!dev_net(dev));
1181 /* Some auto-enslaved devices e.g. failover slaves are
1182 * special, as userspace might rename the device after
1183 * the interface had been brought up and running since
1184 * the point kernel initiated auto-enslavement. Allow
1185 * live name change even when these slave devices are
1188 * Typically, users of these auto-enslaving devices
1189 * don't actually care about slave name change, as
1190 * they are supposed to operate on master interface
1193 if (dev->flags & IFF_UP &&
1194 likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK)))
1197 down_write(&devnet_rename_sem);
1199 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1200 up_write(&devnet_rename_sem);
1204 memcpy(oldname, dev->name, IFNAMSIZ);
1206 err = dev_get_valid_name(net, dev, newname);
1208 up_write(&devnet_rename_sem);
1212 if (oldname[0] && !strchr(oldname, '%'))
1213 netdev_info(dev, "renamed from %s\n", oldname);
1215 old_assign_type = dev->name_assign_type;
1216 dev->name_assign_type = NET_NAME_RENAMED;
1219 ret = device_rename(&dev->dev, dev->name);
1221 memcpy(dev->name, oldname, IFNAMSIZ);
1222 dev->name_assign_type = old_assign_type;
1223 up_write(&devnet_rename_sem);
1227 up_write(&devnet_rename_sem);
1229 netdev_adjacent_rename_links(dev, oldname);
1231 write_lock_bh(&dev_base_lock);
1232 hlist_del_rcu(&dev->name_hlist);
1233 write_unlock_bh(&dev_base_lock);
1237 write_lock_bh(&dev_base_lock);
1238 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1239 write_unlock_bh(&dev_base_lock);
1241 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1242 ret = notifier_to_errno(ret);
1245 /* err >= 0 after dev_alloc_name() or stores the first errno */
1248 down_write(&devnet_rename_sem);
1249 memcpy(dev->name, oldname, IFNAMSIZ);
1250 memcpy(oldname, newname, IFNAMSIZ);
1251 dev->name_assign_type = old_assign_type;
1252 old_assign_type = NET_NAME_RENAMED;
1255 pr_err("%s: name change rollback failed: %d\n",
1264 * dev_set_alias - change ifalias of a device
1266 * @alias: name up to IFALIASZ
1267 * @len: limit of bytes to copy from info
1269 * Set ifalias for a device,
1271 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1273 struct dev_ifalias *new_alias = NULL;
1275 if (len >= IFALIASZ)
1279 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1283 memcpy(new_alias->ifalias, alias, len);
1284 new_alias->ifalias[len] = 0;
1287 mutex_lock(&ifalias_mutex);
1288 rcu_swap_protected(dev->ifalias, new_alias,
1289 mutex_is_locked(&ifalias_mutex));
1290 mutex_unlock(&ifalias_mutex);
1293 kfree_rcu(new_alias, rcuhead);
1297 EXPORT_SYMBOL(dev_set_alias);
1300 * dev_get_alias - get ifalias of a device
1302 * @name: buffer to store name of ifalias
1303 * @len: size of buffer
1305 * get ifalias for a device. Caller must make sure dev cannot go
1306 * away, e.g. rcu read lock or own a reference count to device.
1308 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1310 const struct dev_ifalias *alias;
1314 alias = rcu_dereference(dev->ifalias);
1316 ret = snprintf(name, len, "%s", alias->ifalias);
1323 * netdev_features_change - device changes features
1324 * @dev: device to cause notification
1326 * Called to indicate a device has changed features.
1328 void netdev_features_change(struct net_device *dev)
1330 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1332 EXPORT_SYMBOL(netdev_features_change);
1335 * netdev_state_change - device changes state
1336 * @dev: device to cause notification
1338 * Called to indicate a device has changed state. This function calls
1339 * the notifier chains for netdev_chain and sends a NEWLINK message
1340 * to the routing socket.
1342 void netdev_state_change(struct net_device *dev)
1344 if (dev->flags & IFF_UP) {
1345 struct netdev_notifier_change_info change_info = {
1349 call_netdevice_notifiers_info(NETDEV_CHANGE,
1351 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1354 EXPORT_SYMBOL(netdev_state_change);
1357 * netdev_notify_peers - notify network peers about existence of @dev
1358 * @dev: network device
1360 * Generate traffic such that interested network peers are aware of
1361 * @dev, such as by generating a gratuitous ARP. This may be used when
1362 * a device wants to inform the rest of the network about some sort of
1363 * reconfiguration such as a failover event or virtual machine
1366 void netdev_notify_peers(struct net_device *dev)
1369 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1370 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1373 EXPORT_SYMBOL(netdev_notify_peers);
1375 static int __dev_open(struct net_device *dev)
1377 const struct net_device_ops *ops = dev->netdev_ops;
1382 if (!netif_device_present(dev))
1385 /* Block netpoll from trying to do any rx path servicing.
1386 * If we don't do this there is a chance ndo_poll_controller
1387 * or ndo_poll may be running while we open the device
1389 netpoll_poll_disable(dev);
1391 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1392 ret = notifier_to_errno(ret);
1396 set_bit(__LINK_STATE_START, &dev->state);
1398 if (ops->ndo_validate_addr)
1399 ret = ops->ndo_validate_addr(dev);
1401 if (!ret && ops->ndo_open)
1402 ret = ops->ndo_open(dev);
1404 netpoll_poll_enable(dev);
1407 clear_bit(__LINK_STATE_START, &dev->state);
1409 dev->flags |= IFF_UP;
1410 dev_set_rx_mode(dev);
1412 add_device_randomness(dev->dev_addr, dev->addr_len);
1419 * dev_open - prepare an interface for use.
1420 * @dev: device to open
1422 * Takes a device from down to up state. The device's private open
1423 * function is invoked and then the multicast lists are loaded. Finally
1424 * the device is moved into the up state and a %NETDEV_UP message is
1425 * sent to the netdev notifier chain.
1427 * Calling this function on an active interface is a nop. On a failure
1428 * a negative errno code is returned.
1430 int dev_open(struct net_device *dev)
1434 if (dev->flags & IFF_UP)
1437 ret = __dev_open(dev);
1441 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1442 call_netdevice_notifiers(NETDEV_UP, dev);
1446 EXPORT_SYMBOL(dev_open);
1448 static void __dev_close_many(struct list_head *head)
1450 struct net_device *dev;
1455 list_for_each_entry(dev, head, close_list) {
1456 /* Temporarily disable netpoll until the interface is down */
1457 netpoll_poll_disable(dev);
1459 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1461 clear_bit(__LINK_STATE_START, &dev->state);
1463 /* Synchronize to scheduled poll. We cannot touch poll list, it
1464 * can be even on different cpu. So just clear netif_running().
1466 * dev->stop() will invoke napi_disable() on all of it's
1467 * napi_struct instances on this device.
1469 smp_mb__after_atomic(); /* Commit netif_running(). */
1472 dev_deactivate_many(head);
1474 list_for_each_entry(dev, head, close_list) {
1475 const struct net_device_ops *ops = dev->netdev_ops;
1478 * Call the device specific close. This cannot fail.
1479 * Only if device is UP
1481 * We allow it to be called even after a DETACH hot-plug
1487 dev->flags &= ~IFF_UP;
1488 netpoll_poll_enable(dev);
1492 static void __dev_close(struct net_device *dev)
1496 list_add(&dev->close_list, &single);
1497 __dev_close_many(&single);
1501 void dev_close_many(struct list_head *head, bool unlink)
1503 struct net_device *dev, *tmp;
1505 /* Remove the devices that don't need to be closed */
1506 list_for_each_entry_safe(dev, tmp, head, close_list)
1507 if (!(dev->flags & IFF_UP))
1508 list_del_init(&dev->close_list);
1510 __dev_close_many(head);
1512 list_for_each_entry_safe(dev, tmp, head, close_list) {
1513 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1514 call_netdevice_notifiers(NETDEV_DOWN, dev);
1516 list_del_init(&dev->close_list);
1519 EXPORT_SYMBOL(dev_close_many);
1522 * dev_close - shutdown an interface.
1523 * @dev: device to shutdown
1525 * This function moves an active device into down state. A
1526 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1527 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1530 void dev_close(struct net_device *dev)
1532 if (dev->flags & IFF_UP) {
1535 list_add(&dev->close_list, &single);
1536 dev_close_many(&single, true);
1540 EXPORT_SYMBOL(dev_close);
1544 * dev_disable_lro - disable Large Receive Offload on a device
1547 * Disable Large Receive Offload (LRO) on a net device. Must be
1548 * called under RTNL. This is needed if received packets may be
1549 * forwarded to another interface.
1551 void dev_disable_lro(struct net_device *dev)
1553 struct net_device *lower_dev;
1554 struct list_head *iter;
1556 dev->wanted_features &= ~NETIF_F_LRO;
1557 netdev_update_features(dev);
1559 if (unlikely(dev->features & NETIF_F_LRO))
1560 netdev_WARN(dev, "failed to disable LRO!\n");
1562 netdev_for_each_lower_dev(dev, lower_dev, iter)
1563 dev_disable_lro(lower_dev);
1565 EXPORT_SYMBOL(dev_disable_lro);
1568 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1571 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be
1572 * called under RTNL. This is needed if Generic XDP is installed on
1575 static void dev_disable_gro_hw(struct net_device *dev)
1577 dev->wanted_features &= ~NETIF_F_GRO_HW;
1578 netdev_update_features(dev);
1580 if (unlikely(dev->features & NETIF_F_GRO_HW))
1581 netdev_WARN(dev, "failed to disable GRO_HW!\n");
1584 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1587 case NETDEV_##val: \
1588 return "NETDEV_" __stringify(val);
1590 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1591 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1592 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1593 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER)
1594 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO)
1595 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO)
1596 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1597 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1598 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1601 return "UNKNOWN_NETDEV_EVENT";
1603 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1605 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1606 struct net_device *dev)
1608 struct netdev_notifier_info info = {
1612 return nb->notifier_call(nb, val, &info);
1615 static int dev_boot_phase = 1;
1618 * register_netdevice_notifier - register a network notifier block
1621 * Register a notifier to be called when network device events occur.
1622 * The notifier passed is linked into the kernel structures and must
1623 * not be reused until it has been unregistered. A negative errno code
1624 * is returned on a failure.
1626 * When registered all registration and up events are replayed
1627 * to the new notifier to allow device to have a race free
1628 * view of the network device list.
1631 int register_netdevice_notifier(struct notifier_block *nb)
1633 struct net_device *dev;
1634 struct net_device *last;
1638 /* Close race with setup_net() and cleanup_net() */
1639 down_write(&pernet_ops_rwsem);
1641 err = raw_notifier_chain_register(&netdev_chain, nb);
1647 for_each_netdev(net, dev) {
1648 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1649 err = notifier_to_errno(err);
1653 if (!(dev->flags & IFF_UP))
1656 call_netdevice_notifier(nb, NETDEV_UP, dev);
1662 up_write(&pernet_ops_rwsem);
1668 for_each_netdev(net, dev) {
1672 if (dev->flags & IFF_UP) {
1673 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1675 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1677 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1682 raw_notifier_chain_unregister(&netdev_chain, nb);
1685 EXPORT_SYMBOL(register_netdevice_notifier);
1688 * unregister_netdevice_notifier - unregister a network notifier block
1691 * Unregister a notifier previously registered by
1692 * register_netdevice_notifier(). The notifier is unlinked into the
1693 * kernel structures and may then be reused. A negative errno code
1694 * is returned on a failure.
1696 * After unregistering unregister and down device events are synthesized
1697 * for all devices on the device list to the removed notifier to remove
1698 * the need for special case cleanup code.
1701 int unregister_netdevice_notifier(struct notifier_block *nb)
1703 struct net_device *dev;
1707 /* Close race with setup_net() and cleanup_net() */
1708 down_write(&pernet_ops_rwsem);
1710 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1715 for_each_netdev(net, dev) {
1716 if (dev->flags & IFF_UP) {
1717 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1719 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1721 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1726 up_write(&pernet_ops_rwsem);
1729 EXPORT_SYMBOL(unregister_netdevice_notifier);
1732 * call_netdevice_notifiers_info - call all network notifier blocks
1733 * @val: value passed unmodified to notifier function
1734 * @info: notifier information data
1736 * Call all network notifier blocks. Parameters and return value
1737 * are as for raw_notifier_call_chain().
1740 static int call_netdevice_notifiers_info(unsigned long val,
1741 struct netdev_notifier_info *info)
1744 return raw_notifier_call_chain(&netdev_chain, val, info);
1748 * call_netdevice_notifiers - call all network notifier blocks
1749 * @val: value passed unmodified to notifier function
1750 * @dev: net_device pointer passed unmodified to notifier function
1752 * Call all network notifier blocks. Parameters and return value
1753 * are as for raw_notifier_call_chain().
1756 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1758 struct netdev_notifier_info info = {
1762 return call_netdevice_notifiers_info(val, &info);
1764 EXPORT_SYMBOL(call_netdevice_notifiers);
1767 * call_netdevice_notifiers_mtu - call all network notifier blocks
1768 * @val: value passed unmodified to notifier function
1769 * @dev: net_device pointer passed unmodified to notifier function
1770 * @arg: additional u32 argument passed to the notifier function
1772 * Call all network notifier blocks. Parameters and return value
1773 * are as for raw_notifier_call_chain().
1775 static int call_netdevice_notifiers_mtu(unsigned long val,
1776 struct net_device *dev, u32 arg)
1778 struct netdev_notifier_info_ext info = {
1783 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
1785 return call_netdevice_notifiers_info(val, &info.info);
1788 #ifdef CONFIG_NET_INGRESS
1789 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
1791 void net_inc_ingress_queue(void)
1793 static_branch_inc(&ingress_needed_key);
1795 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
1797 void net_dec_ingress_queue(void)
1799 static_branch_dec(&ingress_needed_key);
1801 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
1804 #ifdef CONFIG_NET_EGRESS
1805 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
1807 void net_inc_egress_queue(void)
1809 static_branch_inc(&egress_needed_key);
1811 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
1813 void net_dec_egress_queue(void)
1815 static_branch_dec(&egress_needed_key);
1817 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
1820 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
1821 #ifdef CONFIG_JUMP_LABEL
1822 static atomic_t netstamp_needed_deferred;
1823 static atomic_t netstamp_wanted;
1824 static void netstamp_clear(struct work_struct *work)
1826 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1829 wanted = atomic_add_return(deferred, &netstamp_wanted);
1831 static_branch_enable(&netstamp_needed_key);
1833 static_branch_disable(&netstamp_needed_key);
1835 static DECLARE_WORK(netstamp_work, netstamp_clear);
1838 void net_enable_timestamp(void)
1840 #ifdef CONFIG_JUMP_LABEL
1844 wanted = atomic_read(&netstamp_wanted);
1847 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted)
1850 atomic_inc(&netstamp_needed_deferred);
1851 schedule_work(&netstamp_work);
1853 static_branch_inc(&netstamp_needed_key);
1856 EXPORT_SYMBOL(net_enable_timestamp);
1858 void net_disable_timestamp(void)
1860 #ifdef CONFIG_JUMP_LABEL
1864 wanted = atomic_read(&netstamp_wanted);
1867 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted)
1870 atomic_dec(&netstamp_needed_deferred);
1871 schedule_work(&netstamp_work);
1873 static_branch_dec(&netstamp_needed_key);
1876 EXPORT_SYMBOL(net_disable_timestamp);
1878 static inline void net_timestamp_set(struct sk_buff *skb)
1881 if (static_branch_unlikely(&netstamp_needed_key))
1882 __net_timestamp(skb);
1885 #define net_timestamp_check(COND, SKB) \
1886 if (static_branch_unlikely(&netstamp_needed_key)) { \
1887 if ((COND) && !(SKB)->tstamp) \
1888 __net_timestamp(SKB); \
1891 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
1895 if (!(dev->flags & IFF_UP))
1898 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1899 if (skb->len <= len)
1902 /* if TSO is enabled, we don't care about the length as the packet
1903 * could be forwarded without being segmented before
1905 if (skb_is_gso(skb))
1910 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1912 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1914 int ret = ____dev_forward_skb(dev, skb);
1917 skb->protocol = eth_type_trans(skb, dev);
1918 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1923 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1926 * dev_forward_skb - loopback an skb to another netif
1928 * @dev: destination network device
1929 * @skb: buffer to forward
1932 * NET_RX_SUCCESS (no congestion)
1933 * NET_RX_DROP (packet was dropped, but freed)
1935 * dev_forward_skb can be used for injecting an skb from the
1936 * start_xmit function of one device into the receive queue
1937 * of another device.
1939 * The receiving device may be in another namespace, so
1940 * we have to clear all information in the skb that could
1941 * impact namespace isolation.
1943 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1945 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1947 EXPORT_SYMBOL_GPL(dev_forward_skb);
1949 static inline int deliver_skb(struct sk_buff *skb,
1950 struct packet_type *pt_prev,
1951 struct net_device *orig_dev)
1953 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1955 refcount_inc(&skb->users);
1956 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1959 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1960 struct packet_type **pt,
1961 struct net_device *orig_dev,
1963 struct list_head *ptype_list)
1965 struct packet_type *ptype, *pt_prev = *pt;
1967 list_for_each_entry_rcu(ptype, ptype_list, list) {
1968 if (ptype->type != type)
1971 deliver_skb(skb, pt_prev, orig_dev);
1977 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1979 if (!ptype->af_packet_priv || !skb->sk)
1982 if (ptype->id_match)
1983 return ptype->id_match(ptype, skb->sk);
1984 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1991 * Support routine. Sends outgoing frames to any network
1992 * taps currently in use.
1995 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1997 struct packet_type *ptype;
1998 struct sk_buff *skb2 = NULL;
1999 struct packet_type *pt_prev = NULL;
2000 struct list_head *ptype_list = &ptype_all;
2004 list_for_each_entry_rcu(ptype, ptype_list, list) {
2005 /* Never send packets back to the socket
2006 * they originated from - MvS (miquels@drinkel.ow.org)
2008 if (skb_loop_sk(ptype, skb))
2012 deliver_skb(skb2, pt_prev, skb->dev);
2017 /* need to clone skb, done only once */
2018 skb2 = skb_clone(skb, GFP_ATOMIC);
2022 net_timestamp_set(skb2);
2024 /* skb->nh should be correctly
2025 * set by sender, so that the second statement is
2026 * just protection against buggy protocols.
2028 skb_reset_mac_header(skb2);
2030 if (skb_network_header(skb2) < skb2->data ||
2031 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2032 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2033 ntohs(skb2->protocol),
2035 skb_reset_network_header(skb2);
2038 skb2->transport_header = skb2->network_header;
2039 skb2->pkt_type = PACKET_OUTGOING;
2043 if (ptype_list == &ptype_all) {
2044 ptype_list = &dev->ptype_all;
2049 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2050 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2056 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2059 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2060 * @dev: Network device
2061 * @txq: number of queues available
2063 * If real_num_tx_queues is changed the tc mappings may no longer be
2064 * valid. To resolve this verify the tc mapping remains valid and if
2065 * not NULL the mapping. With no priorities mapping to this
2066 * offset/count pair it will no longer be used. In the worst case TC0
2067 * is invalid nothing can be done so disable priority mappings. If is
2068 * expected that drivers will fix this mapping if they can before
2069 * calling netif_set_real_num_tx_queues.
2071 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2074 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2076 /* If TC0 is invalidated disable TC mapping */
2077 if (tc->offset + tc->count > txq) {
2078 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2083 /* Invalidated prio to tc mappings set to TC0 */
2084 for (i = 1; i < TC_BITMASK + 1; i++) {
2085 int q = netdev_get_prio_tc_map(dev, i);
2087 tc = &dev->tc_to_txq[q];
2088 if (tc->offset + tc->count > txq) {
2089 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
2091 netdev_set_prio_tc_map(dev, i, 0);
2096 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2099 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2102 /* walk through the TCs and see if it falls into any of them */
2103 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2104 if ((txq - tc->offset) < tc->count)
2108 /* didn't find it, just return -1 to indicate no match */
2114 EXPORT_SYMBOL(netdev_txq_to_tc);
2117 struct static_key xps_needed __read_mostly;
2118 EXPORT_SYMBOL(xps_needed);
2119 struct static_key xps_rxqs_needed __read_mostly;
2120 EXPORT_SYMBOL(xps_rxqs_needed);
2121 static DEFINE_MUTEX(xps_map_mutex);
2122 #define xmap_dereference(P) \
2123 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2125 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2128 struct xps_map *map = NULL;
2132 map = xmap_dereference(dev_maps->attr_map[tci]);
2136 for (pos = map->len; pos--;) {
2137 if (map->queues[pos] != index)
2141 map->queues[pos] = map->queues[--map->len];
2145 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2146 kfree_rcu(map, rcu);
2153 static bool remove_xps_queue_cpu(struct net_device *dev,
2154 struct xps_dev_maps *dev_maps,
2155 int cpu, u16 offset, u16 count)
2157 int num_tc = dev->num_tc ? : 1;
2158 bool active = false;
2161 for (tci = cpu * num_tc; num_tc--; tci++) {
2164 for (i = count, j = offset; i--; j++) {
2165 if (!remove_xps_queue(dev_maps, tci, j))
2175 static void reset_xps_maps(struct net_device *dev,
2176 struct xps_dev_maps *dev_maps,
2180 static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2181 RCU_INIT_POINTER(dev->xps_rxqs_map, NULL);
2183 RCU_INIT_POINTER(dev->xps_cpus_map, NULL);
2185 static_key_slow_dec_cpuslocked(&xps_needed);
2186 kfree_rcu(dev_maps, rcu);
2189 static void clean_xps_maps(struct net_device *dev, const unsigned long *mask,
2190 struct xps_dev_maps *dev_maps, unsigned int nr_ids,
2191 u16 offset, u16 count, bool is_rxqs_map)
2193 bool active = false;
2196 for (j = -1; j = netif_attrmask_next(j, mask, nr_ids),
2198 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset,
2201 reset_xps_maps(dev, dev_maps, is_rxqs_map);
2204 for (i = offset + (count - 1); count--; i--) {
2205 netdev_queue_numa_node_write(
2206 netdev_get_tx_queue(dev, i),
2212 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2215 const unsigned long *possible_mask = NULL;
2216 struct xps_dev_maps *dev_maps;
2217 unsigned int nr_ids;
2219 if (!static_key_false(&xps_needed))
2223 mutex_lock(&xps_map_mutex);
2225 if (static_key_false(&xps_rxqs_needed)) {
2226 dev_maps = xmap_dereference(dev->xps_rxqs_map);
2228 nr_ids = dev->num_rx_queues;
2229 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids,
2230 offset, count, true);
2234 dev_maps = xmap_dereference(dev->xps_cpus_map);
2238 if (num_possible_cpus() > 1)
2239 possible_mask = cpumask_bits(cpu_possible_mask);
2240 nr_ids = nr_cpu_ids;
2241 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, offset, count,
2245 mutex_unlock(&xps_map_mutex);
2249 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2251 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2254 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2255 u16 index, bool is_rxqs_map)
2257 struct xps_map *new_map;
2258 int alloc_len = XPS_MIN_MAP_ALLOC;
2261 for (pos = 0; map && pos < map->len; pos++) {
2262 if (map->queues[pos] != index)
2267 /* Need to add tx-queue to this CPU's/rx-queue's existing map */
2269 if (pos < map->alloc_len)
2272 alloc_len = map->alloc_len * 2;
2275 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2279 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2281 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2282 cpu_to_node(attr_index));
2286 for (i = 0; i < pos; i++)
2287 new_map->queues[i] = map->queues[i];
2288 new_map->alloc_len = alloc_len;
2294 /* Must be called under cpus_read_lock */
2295 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2296 u16 index, bool is_rxqs_map)
2298 const unsigned long *online_mask = NULL, *possible_mask = NULL;
2299 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2300 int i, j, tci, numa_node_id = -2;
2301 int maps_sz, num_tc = 1, tc = 0;
2302 struct xps_map *map, *new_map;
2303 bool active = false;
2304 unsigned int nr_ids;
2307 /* Do not allow XPS on subordinate device directly */
2308 num_tc = dev->num_tc;
2312 /* If queue belongs to subordinate dev use its map */
2313 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2315 tc = netdev_txq_to_tc(dev, index);
2320 mutex_lock(&xps_map_mutex);
2322 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2323 dev_maps = xmap_dereference(dev->xps_rxqs_map);
2324 nr_ids = dev->num_rx_queues;
2326 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2327 if (num_possible_cpus() > 1) {
2328 online_mask = cpumask_bits(cpu_online_mask);
2329 possible_mask = cpumask_bits(cpu_possible_mask);
2331 dev_maps = xmap_dereference(dev->xps_cpus_map);
2332 nr_ids = nr_cpu_ids;
2335 if (maps_sz < L1_CACHE_BYTES)
2336 maps_sz = L1_CACHE_BYTES;
2338 /* allocate memory for queue storage */
2339 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2342 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2343 if (!new_dev_maps) {
2344 mutex_unlock(&xps_map_mutex);
2348 tci = j * num_tc + tc;
2349 map = dev_maps ? xmap_dereference(dev_maps->attr_map[tci]) :
2352 map = expand_xps_map(map, j, index, is_rxqs_map);
2356 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2360 goto out_no_new_maps;
2363 /* Increment static keys at most once per type */
2364 static_key_slow_inc_cpuslocked(&xps_needed);
2366 static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2369 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2371 /* copy maps belonging to foreign traffic classes */
2372 for (i = tc, tci = j * num_tc; dev_maps && i--; tci++) {
2373 /* fill in the new device map from the old device map */
2374 map = xmap_dereference(dev_maps->attr_map[tci]);
2375 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2378 /* We need to explicitly update tci as prevous loop
2379 * could break out early if dev_maps is NULL.
2381 tci = j * num_tc + tc;
2383 if (netif_attr_test_mask(j, mask, nr_ids) &&
2384 netif_attr_test_online(j, online_mask, nr_ids)) {
2385 /* add tx-queue to CPU/rx-queue maps */
2388 map = xmap_dereference(new_dev_maps->attr_map[tci]);
2389 while ((pos < map->len) && (map->queues[pos] != index))
2392 if (pos == map->len)
2393 map->queues[map->len++] = index;
2396 if (numa_node_id == -2)
2397 numa_node_id = cpu_to_node(j);
2398 else if (numa_node_id != cpu_to_node(j))
2402 } else if (dev_maps) {
2403 /* fill in the new device map from the old device map */
2404 map = xmap_dereference(dev_maps->attr_map[tci]);
2405 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2408 /* copy maps belonging to foreign traffic classes */
2409 for (i = num_tc - tc, tci++; dev_maps && --i; tci++) {
2410 /* fill in the new device map from the old device map */
2411 map = xmap_dereference(dev_maps->attr_map[tci]);
2412 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2417 rcu_assign_pointer(dev->xps_rxqs_map, new_dev_maps);
2419 rcu_assign_pointer(dev->xps_cpus_map, new_dev_maps);
2421 /* Cleanup old maps */
2423 goto out_no_old_maps;
2425 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2427 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2428 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2429 map = xmap_dereference(dev_maps->attr_map[tci]);
2430 if (map && map != new_map)
2431 kfree_rcu(map, rcu);
2435 kfree_rcu(dev_maps, rcu);
2438 dev_maps = new_dev_maps;
2443 /* update Tx queue numa node */
2444 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2445 (numa_node_id >= 0) ?
2446 numa_node_id : NUMA_NO_NODE);
2452 /* removes tx-queue from unused CPUs/rx-queues */
2453 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2455 for (i = tc, tci = j * num_tc; i--; tci++)
2456 active |= remove_xps_queue(dev_maps, tci, index);
2457 if (!netif_attr_test_mask(j, mask, nr_ids) ||
2458 !netif_attr_test_online(j, online_mask, nr_ids))
2459 active |= remove_xps_queue(dev_maps, tci, index);
2460 for (i = num_tc - tc, tci++; --i; tci++)
2461 active |= remove_xps_queue(dev_maps, tci, index);
2464 /* free map if not active */
2466 reset_xps_maps(dev, dev_maps, is_rxqs_map);
2469 mutex_unlock(&xps_map_mutex);
2473 /* remove any maps that we added */
2474 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2476 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2477 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2479 xmap_dereference(dev_maps->attr_map[tci]) :
2481 if (new_map && new_map != map)
2486 mutex_unlock(&xps_map_mutex);
2488 kfree(new_dev_maps);
2491 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2493 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2499 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, false);
2504 EXPORT_SYMBOL(netif_set_xps_queue);
2507 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2509 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2511 /* Unbind any subordinate channels */
2512 while (txq-- != &dev->_tx[0]) {
2514 netdev_unbind_sb_channel(dev, txq->sb_dev);
2518 void netdev_reset_tc(struct net_device *dev)
2521 netif_reset_xps_queues_gt(dev, 0);
2523 netdev_unbind_all_sb_channels(dev);
2525 /* Reset TC configuration of device */
2527 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2528 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2530 EXPORT_SYMBOL(netdev_reset_tc);
2532 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2534 if (tc >= dev->num_tc)
2538 netif_reset_xps_queues(dev, offset, count);
2540 dev->tc_to_txq[tc].count = count;
2541 dev->tc_to_txq[tc].offset = offset;
2544 EXPORT_SYMBOL(netdev_set_tc_queue);
2546 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2548 if (num_tc > TC_MAX_QUEUE)
2552 netif_reset_xps_queues_gt(dev, 0);
2554 netdev_unbind_all_sb_channels(dev);
2556 dev->num_tc = num_tc;
2559 EXPORT_SYMBOL(netdev_set_num_tc);
2561 void netdev_unbind_sb_channel(struct net_device *dev,
2562 struct net_device *sb_dev)
2564 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2567 netif_reset_xps_queues_gt(sb_dev, 0);
2569 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2570 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2572 while (txq-- != &dev->_tx[0]) {
2573 if (txq->sb_dev == sb_dev)
2577 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2579 int netdev_bind_sb_channel_queue(struct net_device *dev,
2580 struct net_device *sb_dev,
2581 u8 tc, u16 count, u16 offset)
2583 /* Make certain the sb_dev and dev are already configured */
2584 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2587 /* We cannot hand out queues we don't have */
2588 if ((offset + count) > dev->real_num_tx_queues)
2591 /* Record the mapping */
2592 sb_dev->tc_to_txq[tc].count = count;
2593 sb_dev->tc_to_txq[tc].offset = offset;
2595 /* Provide a way for Tx queue to find the tc_to_txq map or
2596 * XPS map for itself.
2599 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
2603 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
2605 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
2607 /* Do not use a multiqueue device to represent a subordinate channel */
2608 if (netif_is_multiqueue(dev))
2611 /* We allow channels 1 - 32767 to be used for subordinate channels.
2612 * Channel 0 is meant to be "native" mode and used only to represent
2613 * the main root device. We allow writing 0 to reset the device back
2614 * to normal mode after being used as a subordinate channel.
2616 if (channel > S16_MAX)
2619 dev->num_tc = -channel;
2623 EXPORT_SYMBOL(netdev_set_sb_channel);
2626 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2627 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2629 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2634 disabling = txq < dev->real_num_tx_queues;
2636 if (txq < 1 || txq > dev->num_tx_queues)
2639 if (dev->reg_state == NETREG_REGISTERED ||
2640 dev->reg_state == NETREG_UNREGISTERING) {
2643 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2649 netif_setup_tc(dev, txq);
2651 dev_qdisc_change_real_num_tx(dev, txq);
2653 dev->real_num_tx_queues = txq;
2657 qdisc_reset_all_tx_gt(dev, txq);
2659 netif_reset_xps_queues_gt(dev, txq);
2663 dev->real_num_tx_queues = txq;
2668 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2672 * netif_set_real_num_rx_queues - set actual number of RX queues used
2673 * @dev: Network device
2674 * @rxq: Actual number of RX queues
2676 * This must be called either with the rtnl_lock held or before
2677 * registration of the net device. Returns 0 on success, or a
2678 * negative error code. If called before registration, it always
2681 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2685 if (rxq < 1 || rxq > dev->num_rx_queues)
2688 if (dev->reg_state == NETREG_REGISTERED) {
2691 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2697 dev->real_num_rx_queues = rxq;
2700 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2704 * netif_get_num_default_rss_queues - default number of RSS queues
2706 * This routine should set an upper limit on the number of RSS queues
2707 * used by default by multiqueue devices.
2709 int netif_get_num_default_rss_queues(void)
2711 return is_kdump_kernel() ?
2712 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2714 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2716 static void __netif_reschedule(struct Qdisc *q)
2718 struct softnet_data *sd;
2719 unsigned long flags;
2721 local_irq_save(flags);
2722 sd = this_cpu_ptr(&softnet_data);
2723 q->next_sched = NULL;
2724 *sd->output_queue_tailp = q;
2725 sd->output_queue_tailp = &q->next_sched;
2726 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2727 local_irq_restore(flags);
2730 void __netif_schedule(struct Qdisc *q)
2732 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2733 __netif_reschedule(q);
2735 EXPORT_SYMBOL(__netif_schedule);
2737 struct dev_kfree_skb_cb {
2738 enum skb_free_reason reason;
2741 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2743 return (struct dev_kfree_skb_cb *)skb->cb;
2746 void netif_schedule_queue(struct netdev_queue *txq)
2749 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2750 struct Qdisc *q = rcu_dereference(txq->qdisc);
2752 __netif_schedule(q);
2756 EXPORT_SYMBOL(netif_schedule_queue);
2758 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2760 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2764 q = rcu_dereference(dev_queue->qdisc);
2765 __netif_schedule(q);
2769 EXPORT_SYMBOL(netif_tx_wake_queue);
2771 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2773 unsigned long flags;
2778 if (likely(refcount_read(&skb->users) == 1)) {
2780 refcount_set(&skb->users, 0);
2781 } else if (likely(!refcount_dec_and_test(&skb->users))) {
2784 get_kfree_skb_cb(skb)->reason = reason;
2785 local_irq_save(flags);
2786 skb->next = __this_cpu_read(softnet_data.completion_queue);
2787 __this_cpu_write(softnet_data.completion_queue, skb);
2788 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2789 local_irq_restore(flags);
2791 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2793 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2795 if (in_irq() || irqs_disabled())
2796 __dev_kfree_skb_irq(skb, reason);
2800 EXPORT_SYMBOL(__dev_kfree_skb_any);
2804 * netif_device_detach - mark device as removed
2805 * @dev: network device
2807 * Mark device as removed from system and therefore no longer available.
2809 void netif_device_detach(struct net_device *dev)
2811 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2812 netif_running(dev)) {
2813 netif_tx_stop_all_queues(dev);
2816 EXPORT_SYMBOL(netif_device_detach);
2819 * netif_device_attach - mark device as attached
2820 * @dev: network device
2822 * Mark device as attached from system and restart if needed.
2824 void netif_device_attach(struct net_device *dev)
2826 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2827 netif_running(dev)) {
2828 netif_tx_wake_all_queues(dev);
2829 __netdev_watchdog_up(dev);
2832 EXPORT_SYMBOL(netif_device_attach);
2835 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2836 * to be used as a distribution range.
2838 static u16 skb_tx_hash(const struct net_device *dev,
2839 const struct net_device *sb_dev,
2840 struct sk_buff *skb)
2844 u16 qcount = dev->real_num_tx_queues;
2847 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2849 qoffset = sb_dev->tc_to_txq[tc].offset;
2850 qcount = sb_dev->tc_to_txq[tc].count;
2851 if (unlikely(!qcount)) {
2852 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n",
2853 sb_dev->name, qoffset, tc);
2855 qcount = dev->real_num_tx_queues;
2859 if (skb_rx_queue_recorded(skb)) {
2860 hash = skb_get_rx_queue(skb);
2861 if (hash >= qoffset)
2863 while (unlikely(hash >= qcount))
2865 return hash + qoffset;
2868 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
2871 static void skb_warn_bad_offload(const struct sk_buff *skb)
2873 static const netdev_features_t null_features;
2874 struct net_device *dev = skb->dev;
2875 const char *name = "";
2877 if (!net_ratelimit())
2881 if (dev->dev.parent)
2882 name = dev_driver_string(dev->dev.parent);
2884 name = netdev_name(dev);
2886 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2887 "gso_type=%d ip_summed=%d\n",
2888 name, dev ? &dev->features : &null_features,
2889 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2890 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2891 skb_shinfo(skb)->gso_type, skb->ip_summed);
2895 * Invalidate hardware checksum when packet is to be mangled, and
2896 * complete checksum manually on outgoing path.
2898 int skb_checksum_help(struct sk_buff *skb)
2901 int ret = 0, offset;
2903 if (skb->ip_summed == CHECKSUM_COMPLETE)
2904 goto out_set_summed;
2906 if (unlikely(skb_shinfo(skb)->gso_size)) {
2907 skb_warn_bad_offload(skb);
2911 /* Before computing a checksum, we should make sure no frag could
2912 * be modified by an external entity : checksum could be wrong.
2914 if (skb_has_shared_frag(skb)) {
2915 ret = __skb_linearize(skb);
2920 offset = skb_checksum_start_offset(skb);
2921 BUG_ON(offset >= skb_headlen(skb));
2922 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2924 offset += skb->csum_offset;
2925 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2927 if (skb_cloned(skb) &&
2928 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2929 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2934 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
2936 skb->ip_summed = CHECKSUM_NONE;
2940 EXPORT_SYMBOL(skb_checksum_help);
2942 int skb_crc32c_csum_help(struct sk_buff *skb)
2945 int ret = 0, offset, start;
2947 if (skb->ip_summed != CHECKSUM_PARTIAL)
2950 if (unlikely(skb_is_gso(skb)))
2953 /* Before computing a checksum, we should make sure no frag could
2954 * be modified by an external entity : checksum could be wrong.
2956 if (unlikely(skb_has_shared_frag(skb))) {
2957 ret = __skb_linearize(skb);
2961 start = skb_checksum_start_offset(skb);
2962 offset = start + offsetof(struct sctphdr, checksum);
2963 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
2967 if (skb_cloned(skb) &&
2968 !skb_clone_writable(skb, offset + sizeof(__le32))) {
2969 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2973 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
2974 skb->len - start, ~(__u32)0,
2976 *(__le32 *)(skb->data + offset) = crc32c_csum;
2977 skb->ip_summed = CHECKSUM_NONE;
2978 skb->csum_not_inet = 0;
2983 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2985 __be16 type = skb->protocol;
2987 /* Tunnel gso handlers can set protocol to ethernet. */
2988 if (type == htons(ETH_P_TEB)) {
2991 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2994 eth = (struct ethhdr *)skb->data;
2995 type = eth->h_proto;
2998 return __vlan_get_protocol(skb, type, depth);
3002 * skb_mac_gso_segment - mac layer segmentation handler.
3003 * @skb: buffer to segment
3004 * @features: features for the output path (see dev->features)
3006 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
3007 netdev_features_t features)
3009 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
3010 struct packet_offload *ptype;
3011 int vlan_depth = skb->mac_len;
3012 __be16 type = skb_network_protocol(skb, &vlan_depth);
3014 if (unlikely(!type))
3015 return ERR_PTR(-EINVAL);
3017 __skb_pull(skb, vlan_depth);
3020 list_for_each_entry_rcu(ptype, &offload_base, list) {
3021 if (ptype->type == type && ptype->callbacks.gso_segment) {
3022 segs = ptype->callbacks.gso_segment(skb, features);
3028 __skb_push(skb, skb->data - skb_mac_header(skb));
3032 EXPORT_SYMBOL(skb_mac_gso_segment);
3035 /* openvswitch calls this on rx path, so we need a different check.
3037 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
3040 return skb->ip_summed != CHECKSUM_PARTIAL &&
3041 skb->ip_summed != CHECKSUM_UNNECESSARY;
3043 return skb->ip_summed == CHECKSUM_NONE;
3047 * __skb_gso_segment - Perform segmentation on skb.
3048 * @skb: buffer to segment
3049 * @features: features for the output path (see dev->features)
3050 * @tx_path: whether it is called in TX path
3052 * This function segments the given skb and returns a list of segments.
3054 * It may return NULL if the skb requires no segmentation. This is
3055 * only possible when GSO is used for verifying header integrity.
3057 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
3059 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
3060 netdev_features_t features, bool tx_path)
3062 struct sk_buff *segs;
3064 if (unlikely(skb_needs_check(skb, tx_path))) {
3067 /* We're going to init ->check field in TCP or UDP header */
3068 err = skb_cow_head(skb, 0);
3070 return ERR_PTR(err);
3073 /* Only report GSO partial support if it will enable us to
3074 * support segmentation on this frame without needing additional
3077 if (features & NETIF_F_GSO_PARTIAL) {
3078 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
3079 struct net_device *dev = skb->dev;
3081 partial_features |= dev->features & dev->gso_partial_features;
3082 if (!skb_gso_ok(skb, features | partial_features))
3083 features &= ~NETIF_F_GSO_PARTIAL;
3086 BUILD_BUG_ON(SKB_SGO_CB_OFFSET +
3087 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
3089 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
3090 SKB_GSO_CB(skb)->encap_level = 0;
3092 skb_reset_mac_header(skb);
3093 skb_reset_mac_len(skb);
3095 segs = skb_mac_gso_segment(skb, features);
3097 if (unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
3098 skb_warn_bad_offload(skb);
3102 EXPORT_SYMBOL(__skb_gso_segment);
3104 /* Take action when hardware reception checksum errors are detected. */
3106 void netdev_rx_csum_fault(struct net_device *dev)
3108 if (net_ratelimit()) {
3109 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
3113 EXPORT_SYMBOL(netdev_rx_csum_fault);
3116 /* XXX: check that highmem exists at all on the given machine. */
3117 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3119 #ifdef CONFIG_HIGHMEM
3122 if (!(dev->features & NETIF_F_HIGHDMA)) {
3123 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3124 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3126 if (PageHighMem(skb_frag_page(frag)))
3134 /* If MPLS offload request, verify we are testing hardware MPLS features
3135 * instead of standard features for the netdev.
3137 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3138 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3139 netdev_features_t features,
3142 if (eth_p_mpls(type))
3143 features &= skb->dev->mpls_features;
3148 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3149 netdev_features_t features,
3156 static netdev_features_t harmonize_features(struct sk_buff *skb,
3157 netdev_features_t features)
3162 type = skb_network_protocol(skb, &tmp);
3163 features = net_mpls_features(skb, features, type);
3165 if (skb->ip_summed != CHECKSUM_NONE &&
3166 !can_checksum_protocol(features, type)) {
3167 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3169 if (illegal_highdma(skb->dev, skb))
3170 features &= ~NETIF_F_SG;
3175 netdev_features_t passthru_features_check(struct sk_buff *skb,
3176 struct net_device *dev,
3177 netdev_features_t features)
3181 EXPORT_SYMBOL(passthru_features_check);
3183 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3184 struct net_device *dev,
3185 netdev_features_t features)
3187 return vlan_features_check(skb, features);
3190 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3191 struct net_device *dev,
3192 netdev_features_t features)
3194 u16 gso_segs = skb_shinfo(skb)->gso_segs;
3196 if (gso_segs > dev->gso_max_segs)
3197 return features & ~NETIF_F_GSO_MASK;
3199 /* Support for GSO partial features requires software
3200 * intervention before we can actually process the packets
3201 * so we need to strip support for any partial features now
3202 * and we can pull them back in after we have partially
3203 * segmented the frame.
3205 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3206 features &= ~dev->gso_partial_features;
3208 /* Make sure to clear the IPv4 ID mangling feature if the
3209 * IPv4 header has the potential to be fragmented.
3211 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3212 struct iphdr *iph = skb->encapsulation ?
3213 inner_ip_hdr(skb) : ip_hdr(skb);
3215 if (!(iph->frag_off & htons(IP_DF)))
3216 features &= ~NETIF_F_TSO_MANGLEID;
3222 netdev_features_t netif_skb_features(struct sk_buff *skb)
3224 struct net_device *dev = skb->dev;
3225 netdev_features_t features = dev->features;
3227 if (skb_is_gso(skb))
3228 features = gso_features_check(skb, dev, features);
3230 /* If encapsulation offload request, verify we are testing
3231 * hardware encapsulation features instead of standard
3232 * features for the netdev
3234 if (skb->encapsulation)
3235 features &= dev->hw_enc_features;
3237 if (skb_vlan_tagged(skb))
3238 features = netdev_intersect_features(features,
3239 dev->vlan_features |
3240 NETIF_F_HW_VLAN_CTAG_TX |
3241 NETIF_F_HW_VLAN_STAG_TX);
3243 if (dev->netdev_ops->ndo_features_check)
3244 features &= dev->netdev_ops->ndo_features_check(skb, dev,
3247 features &= dflt_features_check(skb, dev, features);
3249 return harmonize_features(skb, features);
3251 EXPORT_SYMBOL(netif_skb_features);
3253 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3254 struct netdev_queue *txq, bool more)
3259 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
3260 dev_queue_xmit_nit(skb, dev);
3263 trace_net_dev_start_xmit(skb, dev);
3264 rc = netdev_start_xmit(skb, dev, txq, more);
3265 trace_net_dev_xmit(skb, rc, dev, len);
3270 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3271 struct netdev_queue *txq, int *ret)
3273 struct sk_buff *skb = first;
3274 int rc = NETDEV_TX_OK;
3277 struct sk_buff *next = skb->next;
3280 rc = xmit_one(skb, dev, txq, next != NULL);
3281 if (unlikely(!dev_xmit_complete(rc))) {
3287 if (netif_tx_queue_stopped(txq) && skb) {
3288 rc = NETDEV_TX_BUSY;
3298 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3299 netdev_features_t features)
3301 if (skb_vlan_tag_present(skb) &&
3302 !vlan_hw_offload_capable(features, skb->vlan_proto))
3303 skb = __vlan_hwaccel_push_inside(skb);
3307 int skb_csum_hwoffload_help(struct sk_buff *skb,
3308 const netdev_features_t features)
3310 if (unlikely(skb->csum_not_inet))
3311 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3312 skb_crc32c_csum_help(skb);
3314 return !!(features & NETIF_F_CSUM_MASK) ? 0 : skb_checksum_help(skb);
3316 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3318 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3320 netdev_features_t features;
3322 features = netif_skb_features(skb);
3323 skb = validate_xmit_vlan(skb, features);
3327 skb = sk_validate_xmit_skb(skb, dev);
3331 if (netif_needs_gso(skb, features)) {
3332 struct sk_buff *segs;
3334 segs = skb_gso_segment(skb, features);
3342 if (skb_needs_linearize(skb, features) &&
3343 __skb_linearize(skb))
3346 /* If packet is not checksummed and device does not
3347 * support checksumming for this protocol, complete
3348 * checksumming here.
3350 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3351 if (skb->encapsulation)
3352 skb_set_inner_transport_header(skb,
3353 skb_checksum_start_offset(skb));
3355 skb_set_transport_header(skb,
3356 skb_checksum_start_offset(skb));
3357 if (skb_csum_hwoffload_help(skb, features))
3362 skb = validate_xmit_xfrm(skb, features, again);
3369 atomic_long_inc(&dev->tx_dropped);
3373 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3375 struct sk_buff *next, *head = NULL, *tail;
3377 for (; skb != NULL; skb = next) {
3381 /* in case skb wont be segmented, point to itself */
3384 skb = validate_xmit_skb(skb, dev, again);
3392 /* If skb was segmented, skb->prev points to
3393 * the last segment. If not, it still contains skb.
3399 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3401 static void qdisc_pkt_len_init(struct sk_buff *skb)
3403 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3405 qdisc_skb_cb(skb)->pkt_len = skb->len;
3407 /* To get more precise estimation of bytes sent on wire,
3408 * we add to pkt_len the headers size of all segments
3410 if (shinfo->gso_size) {
3411 unsigned int hdr_len;
3412 u16 gso_segs = shinfo->gso_segs;
3414 /* mac layer + network layer */
3415 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3417 /* + transport layer */
3418 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3419 const struct tcphdr *th;
3420 struct tcphdr _tcphdr;
3422 th = skb_header_pointer(skb, skb_transport_offset(skb),
3423 sizeof(_tcphdr), &_tcphdr);
3425 hdr_len += __tcp_hdrlen(th);
3427 struct udphdr _udphdr;
3429 if (skb_header_pointer(skb, skb_transport_offset(skb),
3430 sizeof(_udphdr), &_udphdr))
3431 hdr_len += sizeof(struct udphdr);
3434 if (shinfo->gso_type & SKB_GSO_DODGY)
3435 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3438 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3442 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3443 struct net_device *dev,
3444 struct netdev_queue *txq)
3446 spinlock_t *root_lock = qdisc_lock(q);
3447 struct sk_buff *to_free = NULL;
3451 qdisc_calculate_pkt_len(skb, q);
3453 if (q->flags & TCQ_F_NOLOCK) {
3454 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3455 __qdisc_drop(skb, &to_free);
3458 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3462 if (unlikely(to_free))
3463 kfree_skb_list(to_free);
3468 * Heuristic to force contended enqueues to serialize on a
3469 * separate lock before trying to get qdisc main lock.
3470 * This permits qdisc->running owner to get the lock more
3471 * often and dequeue packets faster.
3473 contended = qdisc_is_running(q);
3474 if (unlikely(contended))
3475 spin_lock(&q->busylock);
3477 spin_lock(root_lock);
3478 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3479 __qdisc_drop(skb, &to_free);
3481 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3482 qdisc_run_begin(q)) {
3484 * This is a work-conserving queue; there are no old skbs
3485 * waiting to be sent out; and the qdisc is not running -
3486 * xmit the skb directly.
3489 qdisc_bstats_update(q, skb);
3491 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3492 if (unlikely(contended)) {
3493 spin_unlock(&q->busylock);
3500 rc = NET_XMIT_SUCCESS;
3502 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3503 if (qdisc_run_begin(q)) {
3504 if (unlikely(contended)) {
3505 spin_unlock(&q->busylock);
3512 spin_unlock(root_lock);
3513 if (unlikely(to_free))
3514 kfree_skb_list(to_free);
3515 if (unlikely(contended))
3516 spin_unlock(&q->busylock);
3520 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3521 static void skb_update_prio(struct sk_buff *skb)
3523 const struct netprio_map *map;
3524 const struct sock *sk;
3525 unsigned int prioidx;
3529 map = rcu_dereference_bh(skb->dev->priomap);
3532 sk = skb_to_full_sk(skb);
3536 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3538 if (prioidx < map->priomap_len)
3539 skb->priority = map->priomap[prioidx];
3542 #define skb_update_prio(skb)
3546 * dev_loopback_xmit - loop back @skb
3547 * @net: network namespace this loopback is happening in
3548 * @sk: sk needed to be a netfilter okfn
3549 * @skb: buffer to transmit
3551 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3553 skb_reset_mac_header(skb);
3554 __skb_pull(skb, skb_network_offset(skb));
3555 skb->pkt_type = PACKET_LOOPBACK;
3556 skb->ip_summed = CHECKSUM_UNNECESSARY;
3557 WARN_ON(!skb_dst(skb));
3562 EXPORT_SYMBOL(dev_loopback_xmit);
3564 #ifdef CONFIG_NET_EGRESS
3565 static struct sk_buff *
3566 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3568 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3569 struct tcf_result cl_res;
3574 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3575 mini_qdisc_bstats_cpu_update(miniq, skb);
3577 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
3579 case TC_ACT_RECLASSIFY:
3580 skb->tc_index = TC_H_MIN(cl_res.classid);
3583 mini_qdisc_qstats_cpu_drop(miniq);
3584 *ret = NET_XMIT_DROP;
3590 *ret = NET_XMIT_SUCCESS;
3593 case TC_ACT_REDIRECT:
3594 /* No need to push/pop skb's mac_header here on egress! */
3595 skb_do_redirect(skb);
3596 *ret = NET_XMIT_SUCCESS;
3604 #endif /* CONFIG_NET_EGRESS */
3607 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
3608 struct xps_dev_maps *dev_maps, unsigned int tci)
3610 struct xps_map *map;
3611 int queue_index = -1;
3615 tci += netdev_get_prio_tc_map(dev, skb->priority);
3618 map = rcu_dereference(dev_maps->attr_map[tci]);
3621 queue_index = map->queues[0];
3623 queue_index = map->queues[reciprocal_scale(
3624 skb_get_hash(skb), map->len)];
3625 if (unlikely(queue_index >= dev->real_num_tx_queues))
3632 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
3633 struct sk_buff *skb)
3636 struct xps_dev_maps *dev_maps;
3637 struct sock *sk = skb->sk;
3638 int queue_index = -1;
3640 if (!static_key_false(&xps_needed))
3644 if (!static_key_false(&xps_rxqs_needed))
3647 dev_maps = rcu_dereference(sb_dev->xps_rxqs_map);
3649 int tci = sk_rx_queue_get(sk);
3651 if (tci >= 0 && tci < dev->num_rx_queues)
3652 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
3657 if (queue_index < 0) {
3658 dev_maps = rcu_dereference(sb_dev->xps_cpus_map);
3660 unsigned int tci = skb->sender_cpu - 1;
3662 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
3674 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
3675 struct net_device *sb_dev,
3676 select_queue_fallback_t fallback)
3680 EXPORT_SYMBOL(dev_pick_tx_zero);
3682 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
3683 struct net_device *sb_dev,
3684 select_queue_fallback_t fallback)
3686 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
3688 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
3690 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
3691 struct net_device *sb_dev)
3693 struct sock *sk = skb->sk;
3694 int queue_index = sk_tx_queue_get(sk);
3696 sb_dev = sb_dev ? : dev;
3698 if (queue_index < 0 || skb->ooo_okay ||
3699 queue_index >= dev->real_num_tx_queues) {
3700 int new_index = get_xps_queue(dev, sb_dev, skb);
3703 new_index = skb_tx_hash(dev, sb_dev, skb);
3705 if (queue_index != new_index && sk &&
3707 rcu_access_pointer(sk->sk_dst_cache))
3708 sk_tx_queue_set(sk, new_index);
3710 queue_index = new_index;
3716 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
3717 struct sk_buff *skb,
3718 struct net_device *sb_dev)
3720 int queue_index = 0;
3723 u32 sender_cpu = skb->sender_cpu - 1;
3725 if (sender_cpu >= (u32)NR_CPUS)
3726 skb->sender_cpu = raw_smp_processor_id() + 1;
3729 if (dev->real_num_tx_queues != 1) {
3730 const struct net_device_ops *ops = dev->netdev_ops;
3732 if (ops->ndo_select_queue)
3733 queue_index = ops->ndo_select_queue(dev, skb, sb_dev,
3736 queue_index = __netdev_pick_tx(dev, skb, sb_dev);
3738 queue_index = netdev_cap_txqueue(dev, queue_index);
3741 skb_set_queue_mapping(skb, queue_index);
3742 return netdev_get_tx_queue(dev, queue_index);
3746 * __dev_queue_xmit - transmit a buffer
3747 * @skb: buffer to transmit
3748 * @sb_dev: suboordinate device used for L2 forwarding offload
3750 * Queue a buffer for transmission to a network device. The caller must
3751 * have set the device and priority and built the buffer before calling
3752 * this function. The function can be called from an interrupt.
3754 * A negative errno code is returned on a failure. A success does not
3755 * guarantee the frame will be transmitted as it may be dropped due
3756 * to congestion or traffic shaping.
3758 * -----------------------------------------------------------------------------------
3759 * I notice this method can also return errors from the queue disciplines,
3760 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3763 * Regardless of the return value, the skb is consumed, so it is currently
3764 * difficult to retry a send to this method. (You can bump the ref count
3765 * before sending to hold a reference for retry if you are careful.)
3767 * When calling this method, interrupts MUST be enabled. This is because
3768 * the BH enable code must have IRQs enabled so that it will not deadlock.
3771 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
3773 struct net_device *dev = skb->dev;
3774 struct netdev_queue *txq;
3779 skb_reset_mac_header(skb);
3781 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3782 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3784 /* Disable soft irqs for various locks below. Also
3785 * stops preemption for RCU.
3789 skb_update_prio(skb);
3791 qdisc_pkt_len_init(skb);
3792 #ifdef CONFIG_NET_CLS_ACT
3793 skb->tc_at_ingress = 0;
3794 # ifdef CONFIG_NET_EGRESS
3795 if (static_branch_unlikely(&egress_needed_key)) {
3796 skb = sch_handle_egress(skb, &rc, dev);
3802 /* If device/qdisc don't need skb->dst, release it right now while
3803 * its hot in this cpu cache.
3805 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3810 txq = netdev_pick_tx(dev, skb, sb_dev);
3811 q = rcu_dereference_bh(txq->qdisc);
3813 trace_net_dev_queue(skb);
3815 rc = __dev_xmit_skb(skb, q, dev, txq);
3819 /* The device has no queue. Common case for software devices:
3820 * loopback, all the sorts of tunnels...
3822 * Really, it is unlikely that netif_tx_lock protection is necessary
3823 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
3825 * However, it is possible, that they rely on protection
3828 * Check this and shot the lock. It is not prone from deadlocks.
3829 *Either shot noqueue qdisc, it is even simpler 8)
3831 if (dev->flags & IFF_UP) {
3832 int cpu = smp_processor_id(); /* ok because BHs are off */
3834 /* Other cpus might concurrently change txq->xmit_lock_owner
3835 * to -1 or to their cpu id, but not to our id.
3837 if (READ_ONCE(txq->xmit_lock_owner) != cpu) {
3838 if (dev_xmit_recursion())
3839 goto recursion_alert;
3841 skb = validate_xmit_skb(skb, dev, &again);
3845 HARD_TX_LOCK(dev, txq, cpu);
3847 if (!netif_xmit_stopped(txq)) {
3848 dev_xmit_recursion_inc();
3849 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
3850 dev_xmit_recursion_dec();
3851 if (dev_xmit_complete(rc)) {
3852 HARD_TX_UNLOCK(dev, txq);
3856 HARD_TX_UNLOCK(dev, txq);
3857 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3860 /* Recursion is detected! It is possible,
3864 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3870 rcu_read_unlock_bh();
3872 atomic_long_inc(&dev->tx_dropped);
3873 kfree_skb_list(skb);
3876 rcu_read_unlock_bh();
3880 int dev_queue_xmit(struct sk_buff *skb)
3882 return __dev_queue_xmit(skb, NULL);
3884 EXPORT_SYMBOL(dev_queue_xmit);
3886 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev)
3888 return __dev_queue_xmit(skb, sb_dev);
3890 EXPORT_SYMBOL(dev_queue_xmit_accel);
3892 int dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
3894 struct net_device *dev = skb->dev;
3895 struct sk_buff *orig_skb = skb;
3896 struct netdev_queue *txq;
3897 int ret = NETDEV_TX_BUSY;
3900 if (unlikely(!netif_running(dev) ||
3901 !netif_carrier_ok(dev)))
3904 skb = validate_xmit_skb_list(skb, dev, &again);
3905 if (skb != orig_skb)
3908 skb_set_queue_mapping(skb, queue_id);
3909 txq = skb_get_tx_queue(dev, skb);
3913 dev_xmit_recursion_inc();
3914 HARD_TX_LOCK(dev, txq, smp_processor_id());
3915 if (!netif_xmit_frozen_or_drv_stopped(txq))
3916 ret = netdev_start_xmit(skb, dev, txq, false);
3917 HARD_TX_UNLOCK(dev, txq);
3918 dev_xmit_recursion_dec();
3922 if (!dev_xmit_complete(ret))
3927 atomic_long_inc(&dev->tx_dropped);
3928 kfree_skb_list(skb);
3929 return NET_XMIT_DROP;
3931 EXPORT_SYMBOL(dev_direct_xmit);
3933 /*************************************************************************
3935 *************************************************************************/
3937 int netdev_max_backlog __read_mostly = 1000;
3938 EXPORT_SYMBOL(netdev_max_backlog);
3940 int netdev_tstamp_prequeue __read_mostly = 1;
3941 int netdev_budget __read_mostly = 300;
3942 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */
3943 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ;
3944 int weight_p __read_mostly = 64; /* old backlog weight */
3945 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
3946 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
3947 int dev_rx_weight __read_mostly = 64;
3948 int dev_tx_weight __read_mostly = 64;
3950 /* Called with irq disabled */
3951 static inline void ____napi_schedule(struct softnet_data *sd,
3952 struct napi_struct *napi)
3954 list_add_tail(&napi->poll_list, &sd->poll_list);
3955 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3960 /* One global table that all flow-based protocols share. */
3961 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3962 EXPORT_SYMBOL(rps_sock_flow_table);
3963 u32 rps_cpu_mask __read_mostly;
3964 EXPORT_SYMBOL(rps_cpu_mask);
3966 struct static_key rps_needed __read_mostly;
3967 EXPORT_SYMBOL(rps_needed);
3968 struct static_key rfs_needed __read_mostly;
3969 EXPORT_SYMBOL(rfs_needed);
3971 static struct rps_dev_flow *
3972 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3973 struct rps_dev_flow *rflow, u16 next_cpu)
3975 if (next_cpu < nr_cpu_ids) {
3976 #ifdef CONFIG_RFS_ACCEL
3977 struct netdev_rx_queue *rxqueue;
3978 struct rps_dev_flow_table *flow_table;
3979 struct rps_dev_flow *old_rflow;
3984 /* Should we steer this flow to a different hardware queue? */
3985 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3986 !(dev->features & NETIF_F_NTUPLE))
3988 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3989 if (rxq_index == skb_get_rx_queue(skb))
3992 rxqueue = dev->_rx + rxq_index;
3993 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3996 flow_id = skb_get_hash(skb) & flow_table->mask;
3997 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3998 rxq_index, flow_id);
4002 rflow = &flow_table->flows[flow_id];
4004 if (old_rflow->filter == rflow->filter)
4005 old_rflow->filter = RPS_NO_FILTER;
4009 per_cpu(softnet_data, next_cpu).input_queue_head;
4012 rflow->cpu = next_cpu;
4017 * get_rps_cpu is called from netif_receive_skb and returns the target
4018 * CPU from the RPS map of the receiving queue for a given skb.
4019 * rcu_read_lock must be held on entry.
4021 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4022 struct rps_dev_flow **rflowp)
4024 const struct rps_sock_flow_table *sock_flow_table;
4025 struct netdev_rx_queue *rxqueue = dev->_rx;
4026 struct rps_dev_flow_table *flow_table;
4027 struct rps_map *map;
4032 if (skb_rx_queue_recorded(skb)) {
4033 u16 index = skb_get_rx_queue(skb);
4035 if (unlikely(index >= dev->real_num_rx_queues)) {
4036 WARN_ONCE(dev->real_num_rx_queues > 1,
4037 "%s received packet on queue %u, but number "
4038 "of RX queues is %u\n",
4039 dev->name, index, dev->real_num_rx_queues);
4045 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4047 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4048 map = rcu_dereference(rxqueue->rps_map);
4049 if (!flow_table && !map)
4052 skb_reset_network_header(skb);
4053 hash = skb_get_hash(skb);
4057 sock_flow_table = rcu_dereference(rps_sock_flow_table);
4058 if (flow_table && sock_flow_table) {
4059 struct rps_dev_flow *rflow;
4063 /* First check into global flow table if there is a match */
4064 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
4065 if ((ident ^ hash) & ~rps_cpu_mask)
4068 next_cpu = ident & rps_cpu_mask;
4070 /* OK, now we know there is a match,
4071 * we can look at the local (per receive queue) flow table
4073 rflow = &flow_table->flows[hash & flow_table->mask];
4077 * If the desired CPU (where last recvmsg was done) is
4078 * different from current CPU (one in the rx-queue flow
4079 * table entry), switch if one of the following holds:
4080 * - Current CPU is unset (>= nr_cpu_ids).
4081 * - Current CPU is offline.
4082 * - The current CPU's queue tail has advanced beyond the
4083 * last packet that was enqueued using this table entry.
4084 * This guarantees that all previous packets for the flow
4085 * have been dequeued, thus preserving in order delivery.
4087 if (unlikely(tcpu != next_cpu) &&
4088 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4089 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
4090 rflow->last_qtail)) >= 0)) {
4092 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4095 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4105 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4106 if (cpu_online(tcpu)) {
4116 #ifdef CONFIG_RFS_ACCEL
4119 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4120 * @dev: Device on which the filter was set
4121 * @rxq_index: RX queue index
4122 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4123 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4125 * Drivers that implement ndo_rx_flow_steer() should periodically call
4126 * this function for each installed filter and remove the filters for
4127 * which it returns %true.
4129 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4130 u32 flow_id, u16 filter_id)
4132 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4133 struct rps_dev_flow_table *flow_table;
4134 struct rps_dev_flow *rflow;
4139 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4140 if (flow_table && flow_id <= flow_table->mask) {
4141 rflow = &flow_table->flows[flow_id];
4142 cpu = READ_ONCE(rflow->cpu);
4143 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
4144 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
4145 rflow->last_qtail) <
4146 (int)(10 * flow_table->mask)))
4152 EXPORT_SYMBOL(rps_may_expire_flow);
4154 #endif /* CONFIG_RFS_ACCEL */
4156 /* Called from hardirq (IPI) context */
4157 static void rps_trigger_softirq(void *data)
4159 struct softnet_data *sd = data;
4161 ____napi_schedule(sd, &sd->backlog);
4165 #endif /* CONFIG_RPS */
4168 * Check if this softnet_data structure is another cpu one
4169 * If yes, queue it to our IPI list and return 1
4172 static int rps_ipi_queued(struct softnet_data *sd)
4175 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4178 sd->rps_ipi_next = mysd->rps_ipi_list;
4179 mysd->rps_ipi_list = sd;
4181 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4184 #endif /* CONFIG_RPS */
4188 #ifdef CONFIG_NET_FLOW_LIMIT
4189 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4192 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4194 #ifdef CONFIG_NET_FLOW_LIMIT
4195 struct sd_flow_limit *fl;
4196 struct softnet_data *sd;
4197 unsigned int old_flow, new_flow;
4199 if (qlen < (netdev_max_backlog >> 1))
4202 sd = this_cpu_ptr(&softnet_data);
4205 fl = rcu_dereference(sd->flow_limit);
4207 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4208 old_flow = fl->history[fl->history_head];
4209 fl->history[fl->history_head] = new_flow;
4212 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4214 if (likely(fl->buckets[old_flow]))
4215 fl->buckets[old_flow]--;
4217 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4229 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4230 * queue (may be a remote CPU queue).
4232 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4233 unsigned int *qtail)
4235 struct softnet_data *sd;
4236 unsigned long flags;
4239 sd = &per_cpu(softnet_data, cpu);
4241 local_irq_save(flags);
4244 if (!netif_running(skb->dev))
4246 qlen = skb_queue_len(&sd->input_pkt_queue);
4247 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
4250 __skb_queue_tail(&sd->input_pkt_queue, skb);
4251 input_queue_tail_incr_save(sd, qtail);
4253 local_irq_restore(flags);
4254 return NET_RX_SUCCESS;
4257 /* Schedule NAPI for backlog device
4258 * We can use non atomic operation since we own the queue lock
4260 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
4261 if (!rps_ipi_queued(sd))
4262 ____napi_schedule(sd, &sd->backlog);
4271 local_irq_restore(flags);
4273 atomic_long_inc(&skb->dev->rx_dropped);
4278 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4280 struct net_device *dev = skb->dev;
4281 struct netdev_rx_queue *rxqueue;
4285 if (skb_rx_queue_recorded(skb)) {
4286 u16 index = skb_get_rx_queue(skb);
4288 if (unlikely(index >= dev->real_num_rx_queues)) {
4289 WARN_ONCE(dev->real_num_rx_queues > 1,
4290 "%s received packet on queue %u, but number "
4291 "of RX queues is %u\n",
4292 dev->name, index, dev->real_num_rx_queues);
4294 return rxqueue; /* Return first rxqueue */
4301 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4302 struct xdp_buff *xdp,
4303 struct bpf_prog *xdp_prog)
4305 struct netdev_rx_queue *rxqueue;
4306 void *orig_data, *orig_data_end;
4307 u32 metalen, act = XDP_DROP;
4308 __be16 orig_eth_type;
4314 /* Reinjected packets coming from act_mirred or similar should
4315 * not get XDP generic processing.
4317 if (skb_is_tc_redirected(skb))
4320 /* XDP packets must be linear and must have sufficient headroom
4321 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4322 * native XDP provides, thus we need to do it here as well.
4324 if (skb_cloned(skb) || skb_is_nonlinear(skb) ||
4325 skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4326 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4327 int troom = skb->tail + skb->data_len - skb->end;
4329 /* In case we have to go down the path and also linearize,
4330 * then lets do the pskb_expand_head() work just once here.
4332 if (pskb_expand_head(skb,
4333 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4334 troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4336 if (skb_linearize(skb))
4340 /* The XDP program wants to see the packet starting at the MAC
4343 mac_len = skb->data - skb_mac_header(skb);
4344 hlen = skb_headlen(skb) + mac_len;
4345 xdp->data = skb->data - mac_len;
4346 xdp->data_meta = xdp->data;
4347 xdp->data_end = xdp->data + hlen;
4348 xdp->data_hard_start = skb->data - skb_headroom(skb);
4349 orig_data_end = xdp->data_end;
4350 orig_data = xdp->data;
4351 eth = (struct ethhdr *)xdp->data;
4352 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4353 orig_eth_type = eth->h_proto;
4355 rxqueue = netif_get_rxqueue(skb);
4356 xdp->rxq = &rxqueue->xdp_rxq;
4358 act = bpf_prog_run_xdp(xdp_prog, xdp);
4360 /* check if bpf_xdp_adjust_head was used */
4361 off = xdp->data - orig_data;
4364 __skb_pull(skb, off);
4366 __skb_push(skb, -off);
4368 skb->mac_header += off;
4369 skb_reset_network_header(skb);
4372 /* check if bpf_xdp_adjust_tail was used. it can only "shrink"
4375 off = orig_data_end - xdp->data_end;
4377 skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4382 /* check if XDP changed eth hdr such SKB needs update */
4383 eth = (struct ethhdr *)xdp->data;
4384 if ((orig_eth_type != eth->h_proto) ||
4385 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4386 __skb_push(skb, ETH_HLEN);
4387 skb->protocol = eth_type_trans(skb, skb->dev);
4393 __skb_push(skb, mac_len);
4396 metalen = xdp->data - xdp->data_meta;
4398 skb_metadata_set(skb, metalen);
4401 bpf_warn_invalid_xdp_action(act);
4404 trace_xdp_exception(skb->dev, xdp_prog, act);
4415 /* When doing generic XDP we have to bypass the qdisc layer and the
4416 * network taps in order to match in-driver-XDP behavior.
4418 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4420 struct net_device *dev = skb->dev;
4421 struct netdev_queue *txq;
4422 bool free_skb = true;
4425 txq = netdev_pick_tx(dev, skb, NULL);
4426 cpu = smp_processor_id();
4427 HARD_TX_LOCK(dev, txq, cpu);
4428 if (!netif_xmit_stopped(txq)) {
4429 rc = netdev_start_xmit(skb, dev, txq, 0);
4430 if (dev_xmit_complete(rc))
4433 HARD_TX_UNLOCK(dev, txq);
4435 trace_xdp_exception(dev, xdp_prog, XDP_TX);
4439 EXPORT_SYMBOL_GPL(generic_xdp_tx);
4441 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4443 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4446 struct xdp_buff xdp;
4450 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4451 if (act != XDP_PASS) {
4454 err = xdp_do_generic_redirect(skb->dev, skb,
4460 generic_xdp_tx(skb, xdp_prog);
4471 EXPORT_SYMBOL_GPL(do_xdp_generic);
4473 static int netif_rx_internal(struct sk_buff *skb)
4477 net_timestamp_check(netdev_tstamp_prequeue, skb);
4479 trace_netif_rx(skb);
4482 if (static_key_false(&rps_needed)) {
4483 struct rps_dev_flow voidflow, *rflow = &voidflow;
4489 cpu = get_rps_cpu(skb->dev, skb, &rflow);
4491 cpu = smp_processor_id();
4493 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4502 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
4509 * netif_rx - post buffer to the network code
4510 * @skb: buffer to post
4512 * This function receives a packet from a device driver and queues it for
4513 * the upper (protocol) levels to process. It always succeeds. The buffer
4514 * may be dropped during processing for congestion control or by the
4518 * NET_RX_SUCCESS (no congestion)
4519 * NET_RX_DROP (packet was dropped)
4523 int netif_rx(struct sk_buff *skb)
4525 trace_netif_rx_entry(skb);
4527 return netif_rx_internal(skb);
4529 EXPORT_SYMBOL(netif_rx);
4531 int netif_rx_ni(struct sk_buff *skb)
4535 trace_netif_rx_ni_entry(skb);
4538 err = netif_rx_internal(skb);
4539 if (local_softirq_pending())
4545 EXPORT_SYMBOL(netif_rx_ni);
4547 static __latent_entropy void net_tx_action(struct softirq_action *h)
4549 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4551 if (sd->completion_queue) {
4552 struct sk_buff *clist;
4554 local_irq_disable();
4555 clist = sd->completion_queue;
4556 sd->completion_queue = NULL;
4560 struct sk_buff *skb = clist;
4562 clist = clist->next;
4564 WARN_ON(refcount_read(&skb->users));
4565 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
4566 trace_consume_skb(skb);
4568 trace_kfree_skb(skb, net_tx_action);
4570 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
4573 __kfree_skb_defer(skb);
4576 __kfree_skb_flush();
4579 if (sd->output_queue) {
4582 local_irq_disable();
4583 head = sd->output_queue;
4584 sd->output_queue = NULL;
4585 sd->output_queue_tailp = &sd->output_queue;
4589 struct Qdisc *q = head;
4590 spinlock_t *root_lock = NULL;
4592 head = head->next_sched;
4594 if (!(q->flags & TCQ_F_NOLOCK)) {
4595 root_lock = qdisc_lock(q);
4596 spin_lock(root_lock);
4598 /* We need to make sure head->next_sched is read
4599 * before clearing __QDISC_STATE_SCHED
4601 smp_mb__before_atomic();
4602 clear_bit(__QDISC_STATE_SCHED, &q->state);
4605 spin_unlock(root_lock);
4609 xfrm_dev_backlog(sd);
4612 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
4613 /* This hook is defined here for ATM LANE */
4614 int (*br_fdb_test_addr_hook)(struct net_device *dev,
4615 unsigned char *addr) __read_mostly;
4616 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
4619 static inline struct sk_buff *
4620 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
4621 struct net_device *orig_dev)
4623 #ifdef CONFIG_NET_CLS_ACT
4624 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
4625 struct tcf_result cl_res;
4627 /* If there's at least one ingress present somewhere (so
4628 * we get here via enabled static key), remaining devices
4629 * that are not configured with an ingress qdisc will bail
4636 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4640 qdisc_skb_cb(skb)->pkt_len = skb->len;
4641 skb->tc_at_ingress = 1;
4642 mini_qdisc_bstats_cpu_update(miniq, skb);
4644 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
4646 case TC_ACT_RECLASSIFY:
4647 skb->tc_index = TC_H_MIN(cl_res.classid);
4650 mini_qdisc_qstats_cpu_drop(miniq);
4658 case TC_ACT_REDIRECT:
4659 /* skb_mac_header check was done by cls/act_bpf, so
4660 * we can safely push the L2 header back before
4661 * redirecting to another netdev
4663 __skb_push(skb, skb->mac_len);
4664 skb_do_redirect(skb);
4666 case TC_ACT_REINSERT:
4667 /* this does not scrub the packet, and updates stats on error */
4668 skb_tc_reinsert(skb, &cl_res);
4673 #endif /* CONFIG_NET_CLS_ACT */
4678 * netdev_is_rx_handler_busy - check if receive handler is registered
4679 * @dev: device to check
4681 * Check if a receive handler is already registered for a given device.
4682 * Return true if there one.
4684 * The caller must hold the rtnl_mutex.
4686 bool netdev_is_rx_handler_busy(struct net_device *dev)
4689 return dev && rtnl_dereference(dev->rx_handler);
4691 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
4694 * netdev_rx_handler_register - register receive handler
4695 * @dev: device to register a handler for
4696 * @rx_handler: receive handler to register
4697 * @rx_handler_data: data pointer that is used by rx handler
4699 * Register a receive handler for a device. This handler will then be
4700 * called from __netif_receive_skb. A negative errno code is returned
4703 * The caller must hold the rtnl_mutex.
4705 * For a general description of rx_handler, see enum rx_handler_result.
4707 int netdev_rx_handler_register(struct net_device *dev,
4708 rx_handler_func_t *rx_handler,
4709 void *rx_handler_data)
4711 if (netdev_is_rx_handler_busy(dev))
4714 if (dev->priv_flags & IFF_NO_RX_HANDLER)
4717 /* Note: rx_handler_data must be set before rx_handler */
4718 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
4719 rcu_assign_pointer(dev->rx_handler, rx_handler);
4723 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
4726 * netdev_rx_handler_unregister - unregister receive handler
4727 * @dev: device to unregister a handler from
4729 * Unregister a receive handler from a device.
4731 * The caller must hold the rtnl_mutex.
4733 void netdev_rx_handler_unregister(struct net_device *dev)
4737 RCU_INIT_POINTER(dev->rx_handler, NULL);
4738 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
4739 * section has a guarantee to see a non NULL rx_handler_data
4743 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
4745 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
4748 * Limit the use of PFMEMALLOC reserves to those protocols that implement
4749 * the special handling of PFMEMALLOC skbs.
4751 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
4753 switch (skb->protocol) {
4754 case htons(ETH_P_ARP):
4755 case htons(ETH_P_IP):
4756 case htons(ETH_P_IPV6):
4757 case htons(ETH_P_8021Q):
4758 case htons(ETH_P_8021AD):
4765 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
4766 int *ret, struct net_device *orig_dev)
4768 #ifdef CONFIG_NETFILTER_INGRESS
4769 if (nf_hook_ingress_active(skb)) {
4773 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4778 ingress_retval = nf_hook_ingress(skb);
4780 return ingress_retval;
4782 #endif /* CONFIG_NETFILTER_INGRESS */
4786 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
4787 struct packet_type **ppt_prev)
4789 struct packet_type *ptype, *pt_prev;
4790 rx_handler_func_t *rx_handler;
4791 struct sk_buff *skb = *pskb;
4792 struct net_device *orig_dev;
4793 bool deliver_exact = false;
4794 int ret = NET_RX_DROP;
4797 net_timestamp_check(!netdev_tstamp_prequeue, skb);
4799 trace_netif_receive_skb(skb);
4801 orig_dev = skb->dev;
4803 skb_reset_network_header(skb);
4804 if (!skb_transport_header_was_set(skb))
4805 skb_reset_transport_header(skb);
4806 skb_reset_mac_len(skb);
4811 skb->skb_iif = skb->dev->ifindex;
4813 __this_cpu_inc(softnet_data.processed);
4815 if (static_branch_unlikely(&generic_xdp_needed_key)) {
4819 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
4822 if (ret2 != XDP_PASS) {
4826 skb_reset_mac_len(skb);
4829 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
4830 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
4831 skb = skb_vlan_untag(skb);
4836 if (skb_skip_tc_classify(skb))
4842 list_for_each_entry_rcu(ptype, &ptype_all, list) {
4844 ret = deliver_skb(skb, pt_prev, orig_dev);
4848 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
4850 ret = deliver_skb(skb, pt_prev, orig_dev);
4855 #ifdef CONFIG_NET_INGRESS
4856 if (static_branch_unlikely(&ingress_needed_key)) {
4857 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev);
4861 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
4867 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
4870 if (skb_vlan_tag_present(skb)) {
4872 ret = deliver_skb(skb, pt_prev, orig_dev);
4875 if (vlan_do_receive(&skb))
4877 else if (unlikely(!skb))
4881 rx_handler = rcu_dereference(skb->dev->rx_handler);
4884 ret = deliver_skb(skb, pt_prev, orig_dev);
4887 switch (rx_handler(&skb)) {
4888 case RX_HANDLER_CONSUMED:
4889 ret = NET_RX_SUCCESS;
4891 case RX_HANDLER_ANOTHER:
4893 case RX_HANDLER_EXACT:
4894 deliver_exact = true;
4895 case RX_HANDLER_PASS:
4902 if (unlikely(skb_vlan_tag_present(skb))) {
4903 if (skb_vlan_tag_get_id(skb))
4904 skb->pkt_type = PACKET_OTHERHOST;
4905 /* Note: we might in the future use prio bits
4906 * and set skb->priority like in vlan_do_receive()
4907 * For the time being, just ignore Priority Code Point
4912 type = skb->protocol;
4914 /* deliver only exact match when indicated */
4915 if (likely(!deliver_exact)) {
4916 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4917 &ptype_base[ntohs(type) &
4921 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4922 &orig_dev->ptype_specific);
4924 if (unlikely(skb->dev != orig_dev)) {
4925 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4926 &skb->dev->ptype_specific);
4930 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
4932 *ppt_prev = pt_prev;
4936 atomic_long_inc(&skb->dev->rx_dropped);
4938 atomic_long_inc(&skb->dev->rx_nohandler);
4940 /* Jamal, now you will not able to escape explaining
4941 * me how you were going to use this. :-)
4947 /* The invariant here is that if *ppt_prev is not NULL
4948 * then skb should also be non-NULL.
4950 * Apparently *ppt_prev assignment above holds this invariant due to
4951 * skb dereferencing near it.
4957 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
4959 struct net_device *orig_dev = skb->dev;
4960 struct packet_type *pt_prev = NULL;
4963 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
4965 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
4970 * netif_receive_skb_core - special purpose version of netif_receive_skb
4971 * @skb: buffer to process
4973 * More direct receive version of netif_receive_skb(). It should
4974 * only be used by callers that have a need to skip RPS and Generic XDP.
4975 * Caller must also take care of handling if (page_is_)pfmemalloc.
4977 * This function may only be called from softirq context and interrupts
4978 * should be enabled.
4980 * Return values (usually ignored):
4981 * NET_RX_SUCCESS: no congestion
4982 * NET_RX_DROP: packet was dropped
4984 int netif_receive_skb_core(struct sk_buff *skb)
4989 ret = __netif_receive_skb_one_core(skb, false);
4994 EXPORT_SYMBOL(netif_receive_skb_core);
4996 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
4997 struct packet_type *pt_prev,
4998 struct net_device *orig_dev)
5000 struct sk_buff *skb, *next;
5004 if (list_empty(head))
5006 if (pt_prev->list_func != NULL)
5007 pt_prev->list_func(head, pt_prev, orig_dev);
5009 list_for_each_entry_safe(skb, next, head, list) {
5010 skb_list_del_init(skb);
5011 pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5015 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5017 /* Fast-path assumptions:
5018 * - There is no RX handler.
5019 * - Only one packet_type matches.
5020 * If either of these fails, we will end up doing some per-packet
5021 * processing in-line, then handling the 'last ptype' for the whole
5022 * sublist. This can't cause out-of-order delivery to any single ptype,
5023 * because the 'last ptype' must be constant across the sublist, and all
5024 * other ptypes are handled per-packet.
5026 /* Current (common) ptype of sublist */
5027 struct packet_type *pt_curr = NULL;
5028 /* Current (common) orig_dev of sublist */
5029 struct net_device *od_curr = NULL;
5030 struct list_head sublist;
5031 struct sk_buff *skb, *next;
5033 INIT_LIST_HEAD(&sublist);
5034 list_for_each_entry_safe(skb, next, head, list) {
5035 struct net_device *orig_dev = skb->dev;
5036 struct packet_type *pt_prev = NULL;
5038 skb_list_del_init(skb);
5039 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5042 if (pt_curr != pt_prev || od_curr != orig_dev) {
5043 /* dispatch old sublist */
5044 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5045 /* start new sublist */
5046 INIT_LIST_HEAD(&sublist);
5050 list_add_tail(&skb->list, &sublist);
5053 /* dispatch final sublist */
5054 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5057 static int __netif_receive_skb(struct sk_buff *skb)
5061 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5062 unsigned int noreclaim_flag;
5065 * PFMEMALLOC skbs are special, they should
5066 * - be delivered to SOCK_MEMALLOC sockets only
5067 * - stay away from userspace
5068 * - have bounded memory usage
5070 * Use PF_MEMALLOC as this saves us from propagating the allocation
5071 * context down to all allocation sites.
5073 noreclaim_flag = memalloc_noreclaim_save();
5074 ret = __netif_receive_skb_one_core(skb, true);
5075 memalloc_noreclaim_restore(noreclaim_flag);
5077 ret = __netif_receive_skb_one_core(skb, false);
5082 static void __netif_receive_skb_list(struct list_head *head)
5084 unsigned long noreclaim_flag = 0;
5085 struct sk_buff *skb, *next;
5086 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5088 list_for_each_entry_safe(skb, next, head, list) {
5089 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5090 struct list_head sublist;
5092 /* Handle the previous sublist */
5093 list_cut_before(&sublist, head, &skb->list);
5094 if (!list_empty(&sublist))
5095 __netif_receive_skb_list_core(&sublist, pfmemalloc);
5096 pfmemalloc = !pfmemalloc;
5097 /* See comments in __netif_receive_skb */
5099 noreclaim_flag = memalloc_noreclaim_save();
5101 memalloc_noreclaim_restore(noreclaim_flag);
5104 /* Handle the remaining sublist */
5105 if (!list_empty(head))
5106 __netif_receive_skb_list_core(head, pfmemalloc);
5107 /* Restore pflags */
5109 memalloc_noreclaim_restore(noreclaim_flag);
5112 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5114 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5115 struct bpf_prog *new = xdp->prog;
5118 switch (xdp->command) {
5119 case XDP_SETUP_PROG:
5120 rcu_assign_pointer(dev->xdp_prog, new);
5125 static_branch_dec(&generic_xdp_needed_key);
5126 } else if (new && !old) {
5127 static_branch_inc(&generic_xdp_needed_key);
5128 dev_disable_lro(dev);
5129 dev_disable_gro_hw(dev);
5133 case XDP_QUERY_PROG:
5134 xdp->prog_id = old ? old->aux->id : 0;
5145 static int netif_receive_skb_internal(struct sk_buff *skb)
5149 net_timestamp_check(netdev_tstamp_prequeue, skb);
5151 if (skb_defer_rx_timestamp(skb))
5152 return NET_RX_SUCCESS;
5156 if (static_key_false(&rps_needed)) {
5157 struct rps_dev_flow voidflow, *rflow = &voidflow;
5158 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5161 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5167 ret = __netif_receive_skb(skb);
5172 static void netif_receive_skb_list_internal(struct list_head *head)
5174 struct sk_buff *skb, *next;
5175 struct list_head sublist;
5177 INIT_LIST_HEAD(&sublist);
5178 list_for_each_entry_safe(skb, next, head, list) {
5179 net_timestamp_check(netdev_tstamp_prequeue, skb);
5180 skb_list_del_init(skb);
5181 if (!skb_defer_rx_timestamp(skb))
5182 list_add_tail(&skb->list, &sublist);
5184 list_splice_init(&sublist, head);
5188 if (static_key_false(&rps_needed)) {
5189 list_for_each_entry_safe(skb, next, head, list) {
5190 struct rps_dev_flow voidflow, *rflow = &voidflow;
5191 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5194 /* Will be handled, remove from list */
5195 skb_list_del_init(skb);
5196 enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5201 __netif_receive_skb_list(head);
5206 * netif_receive_skb - process receive buffer from network
5207 * @skb: buffer to process
5209 * netif_receive_skb() is the main receive data processing function.
5210 * It always succeeds. The buffer may be dropped during processing
5211 * for congestion control or by the protocol layers.
5213 * This function may only be called from softirq context and interrupts
5214 * should be enabled.
5216 * Return values (usually ignored):
5217 * NET_RX_SUCCESS: no congestion
5218 * NET_RX_DROP: packet was dropped
5220 int netif_receive_skb(struct sk_buff *skb)
5222 trace_netif_receive_skb_entry(skb);
5224 return netif_receive_skb_internal(skb);
5226 EXPORT_SYMBOL(netif_receive_skb);
5229 * netif_receive_skb_list - process many receive buffers from network
5230 * @head: list of skbs to process.
5232 * Since return value of netif_receive_skb() is normally ignored, and
5233 * wouldn't be meaningful for a list, this function returns void.
5235 * This function may only be called from softirq context and interrupts
5236 * should be enabled.
5238 void netif_receive_skb_list(struct list_head *head)
5240 struct sk_buff *skb;
5242 if (list_empty(head))
5244 list_for_each_entry(skb, head, list)
5245 trace_netif_receive_skb_list_entry(skb);
5246 netif_receive_skb_list_internal(head);
5248 EXPORT_SYMBOL(netif_receive_skb_list);
5250 DEFINE_PER_CPU(struct work_struct, flush_works);
5252 /* Network device is going away, flush any packets still pending */
5253 static void flush_backlog(struct work_struct *work)
5255 struct sk_buff *skb, *tmp;
5256 struct softnet_data *sd;
5259 sd = this_cpu_ptr(&softnet_data);
5261 local_irq_disable();
5263 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5264 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5265 __skb_unlink(skb, &sd->input_pkt_queue);
5266 dev_kfree_skb_irq(skb);
5267 input_queue_head_incr(sd);
5273 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5274 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5275 __skb_unlink(skb, &sd->process_queue);
5277 input_queue_head_incr(sd);
5283 static void flush_all_backlogs(void)
5289 for_each_online_cpu(cpu)
5290 queue_work_on(cpu, system_highpri_wq,
5291 per_cpu_ptr(&flush_works, cpu));
5293 for_each_online_cpu(cpu)
5294 flush_work(per_cpu_ptr(&flush_works, cpu));
5299 static int napi_gro_complete(struct sk_buff *skb)
5301 struct packet_offload *ptype;
5302 __be16 type = skb->protocol;
5303 struct list_head *head = &offload_base;
5306 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
5308 if (NAPI_GRO_CB(skb)->count == 1) {
5309 skb_shinfo(skb)->gso_size = 0;
5314 list_for_each_entry_rcu(ptype, head, list) {
5315 if (ptype->type != type || !ptype->callbacks.gro_complete)
5318 err = ptype->callbacks.gro_complete(skb, 0);
5324 WARN_ON(&ptype->list == head);
5326 return NET_RX_SUCCESS;
5330 return netif_receive_skb_internal(skb);
5333 static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index,
5336 struct list_head *head = &napi->gro_hash[index].list;
5337 struct sk_buff *skb, *p;
5339 list_for_each_entry_safe_reverse(skb, p, head, list) {
5340 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
5342 list_del(&skb->list);
5344 napi_gro_complete(skb);
5345 napi->gro_hash[index].count--;
5348 if (!napi->gro_hash[index].count)
5349 __clear_bit(index, &napi->gro_bitmask);
5352 /* napi->gro_hash[].list contains packets ordered by age.
5353 * youngest packets at the head of it.
5354 * Complete skbs in reverse order to reduce latencies.
5356 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
5360 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
5361 if (test_bit(i, &napi->gro_bitmask))
5362 __napi_gro_flush_chain(napi, i, flush_old);
5365 EXPORT_SYMBOL(napi_gro_flush);
5367 static struct list_head *gro_list_prepare(struct napi_struct *napi,
5368 struct sk_buff *skb)
5370 unsigned int maclen = skb->dev->hard_header_len;
5371 u32 hash = skb_get_hash_raw(skb);
5372 struct list_head *head;
5375 head = &napi->gro_hash[hash & (GRO_HASH_BUCKETS - 1)].list;
5376 list_for_each_entry(p, head, list) {
5377 unsigned long diffs;
5379 NAPI_GRO_CB(p)->flush = 0;
5381 if (hash != skb_get_hash_raw(p)) {
5382 NAPI_GRO_CB(p)->same_flow = 0;
5386 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
5387 diffs |= p->vlan_tci ^ skb->vlan_tci;
5388 diffs |= skb_metadata_dst_cmp(p, skb);
5389 diffs |= skb_metadata_differs(p, skb);
5390 if (maclen == ETH_HLEN)
5391 diffs |= compare_ether_header(skb_mac_header(p),
5392 skb_mac_header(skb));
5394 diffs = memcmp(skb_mac_header(p),
5395 skb_mac_header(skb),
5397 NAPI_GRO_CB(p)->same_flow = !diffs;
5403 static void skb_gro_reset_offset(struct sk_buff *skb)
5405 const struct skb_shared_info *pinfo = skb_shinfo(skb);
5406 const skb_frag_t *frag0 = &pinfo->frags[0];
5408 NAPI_GRO_CB(skb)->data_offset = 0;
5409 NAPI_GRO_CB(skb)->frag0 = NULL;
5410 NAPI_GRO_CB(skb)->frag0_len = 0;
5412 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
5414 !PageHighMem(skb_frag_page(frag0)) &&
5415 (!NET_IP_ALIGN || !(skb_frag_off(frag0) & 3))) {
5416 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
5417 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
5418 skb_frag_size(frag0),
5419 skb->end - skb->tail);
5423 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
5425 struct skb_shared_info *pinfo = skb_shinfo(skb);
5427 BUG_ON(skb->end - skb->tail < grow);
5429 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
5431 skb->data_len -= grow;
5434 pinfo->frags[0].page_offset += grow;
5435 skb_frag_size_sub(&pinfo->frags[0], grow);
5437 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
5438 skb_frag_unref(skb, 0);
5439 memmove(pinfo->frags, pinfo->frags + 1,
5440 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
5444 static void gro_flush_oldest(struct list_head *head)
5446 struct sk_buff *oldest;
5448 oldest = list_last_entry(head, struct sk_buff, list);
5450 /* We are called with head length >= MAX_GRO_SKBS, so this is
5453 if (WARN_ON_ONCE(!oldest))
5456 /* Do not adjust napi->gro_hash[].count, caller is adding a new
5459 list_del(&oldest->list);
5460 oldest->next = NULL;
5461 napi_gro_complete(oldest);
5464 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
5466 u32 hash = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1);
5467 struct list_head *head = &offload_base;
5468 struct packet_offload *ptype;
5469 __be16 type = skb->protocol;
5470 struct list_head *gro_head;
5471 struct sk_buff *pp = NULL;
5472 enum gro_result ret;
5476 if (netif_elide_gro(skb->dev))
5479 gro_head = gro_list_prepare(napi, skb);
5482 list_for_each_entry_rcu(ptype, head, list) {
5483 if (ptype->type != type || !ptype->callbacks.gro_receive)
5486 skb_set_network_header(skb, skb_gro_offset(skb));
5487 skb_reset_mac_len(skb);
5488 NAPI_GRO_CB(skb)->same_flow = 0;
5489 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb);
5490 NAPI_GRO_CB(skb)->free = 0;
5491 NAPI_GRO_CB(skb)->encap_mark = 0;
5492 NAPI_GRO_CB(skb)->recursion_counter = 0;
5493 NAPI_GRO_CB(skb)->is_fou = 0;
5494 NAPI_GRO_CB(skb)->is_atomic = 1;
5495 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
5497 /* Setup for GRO checksum validation */
5498 switch (skb->ip_summed) {
5499 case CHECKSUM_COMPLETE:
5500 NAPI_GRO_CB(skb)->csum = skb->csum;
5501 NAPI_GRO_CB(skb)->csum_valid = 1;
5502 NAPI_GRO_CB(skb)->csum_cnt = 0;
5504 case CHECKSUM_UNNECESSARY:
5505 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
5506 NAPI_GRO_CB(skb)->csum_valid = 0;
5509 NAPI_GRO_CB(skb)->csum_cnt = 0;
5510 NAPI_GRO_CB(skb)->csum_valid = 0;
5513 pp = ptype->callbacks.gro_receive(gro_head, skb);
5518 if (&ptype->list == head)
5521 if (IS_ERR(pp) && PTR_ERR(pp) == -EINPROGRESS) {
5526 same_flow = NAPI_GRO_CB(skb)->same_flow;
5527 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
5530 list_del(&pp->list);
5532 napi_gro_complete(pp);
5533 napi->gro_hash[hash].count--;
5539 if (NAPI_GRO_CB(skb)->flush)
5542 if (unlikely(napi->gro_hash[hash].count >= MAX_GRO_SKBS)) {
5543 gro_flush_oldest(gro_head);
5545 napi->gro_hash[hash].count++;
5547 NAPI_GRO_CB(skb)->count = 1;
5548 NAPI_GRO_CB(skb)->age = jiffies;
5549 NAPI_GRO_CB(skb)->last = skb;
5550 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
5551 list_add(&skb->list, gro_head);
5555 grow = skb_gro_offset(skb) - skb_headlen(skb);
5557 gro_pull_from_frag0(skb, grow);
5559 if (napi->gro_hash[hash].count) {
5560 if (!test_bit(hash, &napi->gro_bitmask))
5561 __set_bit(hash, &napi->gro_bitmask);
5562 } else if (test_bit(hash, &napi->gro_bitmask)) {
5563 __clear_bit(hash, &napi->gro_bitmask);
5573 struct packet_offload *gro_find_receive_by_type(__be16 type)
5575 struct list_head *offload_head = &offload_base;
5576 struct packet_offload *ptype;
5578 list_for_each_entry_rcu(ptype, offload_head, list) {
5579 if (ptype->type != type || !ptype->callbacks.gro_receive)
5585 EXPORT_SYMBOL(gro_find_receive_by_type);
5587 struct packet_offload *gro_find_complete_by_type(__be16 type)
5589 struct list_head *offload_head = &offload_base;
5590 struct packet_offload *ptype;
5592 list_for_each_entry_rcu(ptype, offload_head, list) {
5593 if (ptype->type != type || !ptype->callbacks.gro_complete)
5599 EXPORT_SYMBOL(gro_find_complete_by_type);
5601 static void napi_skb_free_stolen_head(struct sk_buff *skb)
5605 kmem_cache_free(skbuff_head_cache, skb);
5608 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
5612 if (netif_receive_skb_internal(skb))
5620 case GRO_MERGED_FREE:
5621 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
5622 napi_skb_free_stolen_head(skb);
5636 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
5638 skb_mark_napi_id(skb, napi);
5639 trace_napi_gro_receive_entry(skb);
5641 skb_gro_reset_offset(skb);
5643 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
5645 EXPORT_SYMBOL(napi_gro_receive);
5647 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
5649 if (unlikely(skb->pfmemalloc)) {
5653 __skb_pull(skb, skb_headlen(skb));
5654 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
5655 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
5657 skb->dev = napi->dev;
5660 /* eth_type_trans() assumes pkt_type is PACKET_HOST */
5661 skb->pkt_type = PACKET_HOST;
5663 skb->encapsulation = 0;
5664 skb_shinfo(skb)->gso_type = 0;
5665 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
5671 struct sk_buff *napi_get_frags(struct napi_struct *napi)
5673 struct sk_buff *skb = napi->skb;
5676 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
5679 skb_mark_napi_id(skb, napi);
5684 EXPORT_SYMBOL(napi_get_frags);
5686 static gro_result_t napi_frags_finish(struct napi_struct *napi,
5687 struct sk_buff *skb,
5693 __skb_push(skb, ETH_HLEN);
5694 skb->protocol = eth_type_trans(skb, skb->dev);
5695 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
5700 napi_reuse_skb(napi, skb);
5703 case GRO_MERGED_FREE:
5704 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
5705 napi_skb_free_stolen_head(skb);
5707 napi_reuse_skb(napi, skb);
5718 /* Upper GRO stack assumes network header starts at gro_offset=0
5719 * Drivers could call both napi_gro_frags() and napi_gro_receive()
5720 * We copy ethernet header into skb->data to have a common layout.
5722 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
5724 struct sk_buff *skb = napi->skb;
5725 const struct ethhdr *eth;
5726 unsigned int hlen = sizeof(*eth);
5730 skb_reset_mac_header(skb);
5731 skb_gro_reset_offset(skb);
5733 if (unlikely(skb_gro_header_hard(skb, hlen))) {
5734 eth = skb_gro_header_slow(skb, hlen, 0);
5735 if (unlikely(!eth)) {
5736 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
5737 __func__, napi->dev->name);
5738 napi_reuse_skb(napi, skb);
5742 eth = (const struct ethhdr *)skb->data;
5743 gro_pull_from_frag0(skb, hlen);
5744 NAPI_GRO_CB(skb)->frag0 += hlen;
5745 NAPI_GRO_CB(skb)->frag0_len -= hlen;
5747 __skb_pull(skb, hlen);
5750 * This works because the only protocols we care about don't require
5752 * We'll fix it up properly in napi_frags_finish()
5754 skb->protocol = eth->h_proto;
5759 gro_result_t napi_gro_frags(struct napi_struct *napi)
5761 struct sk_buff *skb = napi_frags_skb(napi);
5766 trace_napi_gro_frags_entry(skb);
5768 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
5770 EXPORT_SYMBOL(napi_gro_frags);
5772 /* Compute the checksum from gro_offset and return the folded value
5773 * after adding in any pseudo checksum.
5775 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
5780 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
5782 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
5783 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
5785 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
5786 !skb->csum_complete_sw)
5787 netdev_rx_csum_fault(skb->dev);
5790 NAPI_GRO_CB(skb)->csum = wsum;
5791 NAPI_GRO_CB(skb)->csum_valid = 1;
5795 EXPORT_SYMBOL(__skb_gro_checksum_complete);
5797 static void net_rps_send_ipi(struct softnet_data *remsd)
5801 struct softnet_data *next = remsd->rps_ipi_next;
5803 if (cpu_online(remsd->cpu))
5804 smp_call_function_single_async(remsd->cpu, &remsd->csd);
5811 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
5812 * Note: called with local irq disabled, but exits with local irq enabled.
5814 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
5817 struct softnet_data *remsd = sd->rps_ipi_list;
5820 sd->rps_ipi_list = NULL;
5824 /* Send pending IPI's to kick RPS processing on remote cpus. */
5825 net_rps_send_ipi(remsd);
5831 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
5834 return sd->rps_ipi_list != NULL;
5840 static int process_backlog(struct napi_struct *napi, int quota)
5842 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
5846 /* Check if we have pending ipi, its better to send them now,
5847 * not waiting net_rx_action() end.
5849 if (sd_has_rps_ipi_waiting(sd)) {
5850 local_irq_disable();
5851 net_rps_action_and_irq_enable(sd);
5854 napi->weight = dev_rx_weight;
5856 struct sk_buff *skb;
5858 while ((skb = __skb_dequeue(&sd->process_queue))) {
5860 __netif_receive_skb(skb);
5862 input_queue_head_incr(sd);
5863 if (++work >= quota)
5868 local_irq_disable();
5870 if (skb_queue_empty(&sd->input_pkt_queue)) {
5872 * Inline a custom version of __napi_complete().
5873 * only current cpu owns and manipulates this napi,
5874 * and NAPI_STATE_SCHED is the only possible flag set
5876 * We can use a plain write instead of clear_bit(),
5877 * and we dont need an smp_mb() memory barrier.
5882 skb_queue_splice_tail_init(&sd->input_pkt_queue,
5883 &sd->process_queue);
5893 * __napi_schedule - schedule for receive
5894 * @n: entry to schedule
5896 * The entry's receive function will be scheduled to run.
5897 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
5899 void __napi_schedule(struct napi_struct *n)
5901 unsigned long flags;
5903 local_irq_save(flags);
5904 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
5905 local_irq_restore(flags);
5907 EXPORT_SYMBOL(__napi_schedule);
5910 * napi_schedule_prep - check if napi can be scheduled
5913 * Test if NAPI routine is already running, and if not mark
5914 * it as running. This is used as a condition variable
5915 * insure only one NAPI poll instance runs. We also make
5916 * sure there is no pending NAPI disable.
5918 bool napi_schedule_prep(struct napi_struct *n)
5920 unsigned long val, new;
5923 val = READ_ONCE(n->state);
5924 if (unlikely(val & NAPIF_STATE_DISABLE))
5926 new = val | NAPIF_STATE_SCHED;
5928 /* Sets STATE_MISSED bit if STATE_SCHED was already set
5929 * This was suggested by Alexander Duyck, as compiler
5930 * emits better code than :
5931 * if (val & NAPIF_STATE_SCHED)
5932 * new |= NAPIF_STATE_MISSED;
5934 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
5936 } while (cmpxchg(&n->state, val, new) != val);
5938 return !(val & NAPIF_STATE_SCHED);
5940 EXPORT_SYMBOL(napi_schedule_prep);
5943 * __napi_schedule_irqoff - schedule for receive
5944 * @n: entry to schedule
5946 * Variant of __napi_schedule() assuming hard irqs are masked.
5948 * On PREEMPT_RT enabled kernels this maps to __napi_schedule()
5949 * because the interrupt disabled assumption might not be true
5950 * due to force-threaded interrupts and spinlock substitution.
5952 void __napi_schedule_irqoff(struct napi_struct *n)
5954 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
5955 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
5959 EXPORT_SYMBOL(__napi_schedule_irqoff);
5961 bool napi_complete_done(struct napi_struct *n, int work_done)
5963 unsigned long flags, val, new;
5966 * 1) Don't let napi dequeue from the cpu poll list
5967 * just in case its running on a different cpu.
5968 * 2) If we are busy polling, do nothing here, we have
5969 * the guarantee we will be called later.
5971 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
5972 NAPIF_STATE_IN_BUSY_POLL)))
5975 if (n->gro_bitmask) {
5976 unsigned long timeout = 0;
5979 timeout = n->dev->gro_flush_timeout;
5981 /* When the NAPI instance uses a timeout and keeps postponing
5982 * it, we need to bound somehow the time packets are kept in
5985 napi_gro_flush(n, !!timeout);
5987 hrtimer_start(&n->timer, ns_to_ktime(timeout),
5988 HRTIMER_MODE_REL_PINNED);
5990 if (unlikely(!list_empty(&n->poll_list))) {
5991 /* If n->poll_list is not empty, we need to mask irqs */
5992 local_irq_save(flags);
5993 list_del_init(&n->poll_list);
5994 local_irq_restore(flags);
5998 val = READ_ONCE(n->state);
6000 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6002 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED);
6004 /* If STATE_MISSED was set, leave STATE_SCHED set,
6005 * because we will call napi->poll() one more time.
6006 * This C code was suggested by Alexander Duyck to help gcc.
6008 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6010 } while (cmpxchg(&n->state, val, new) != val);
6012 if (unlikely(val & NAPIF_STATE_MISSED)) {
6019 EXPORT_SYMBOL(napi_complete_done);
6021 /* must be called under rcu_read_lock(), as we dont take a reference */
6022 static struct napi_struct *napi_by_id(unsigned int napi_id)
6024 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6025 struct napi_struct *napi;
6027 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6028 if (napi->napi_id == napi_id)
6034 #if defined(CONFIG_NET_RX_BUSY_POLL)
6036 #define BUSY_POLL_BUDGET 8
6038 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock)
6042 /* Busy polling means there is a high chance device driver hard irq
6043 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6044 * set in napi_schedule_prep().
6045 * Since we are about to call napi->poll() once more, we can safely
6046 * clear NAPI_STATE_MISSED.
6048 * Note: x86 could use a single "lock and ..." instruction
6049 * to perform these two clear_bit()
6051 clear_bit(NAPI_STATE_MISSED, &napi->state);
6052 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6056 /* All we really want here is to re-enable device interrupts.
6057 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6059 rc = napi->poll(napi, BUSY_POLL_BUDGET);
6060 trace_napi_poll(napi, rc, BUSY_POLL_BUDGET);
6061 netpoll_poll_unlock(have_poll_lock);
6062 if (rc == BUSY_POLL_BUDGET)
6063 __napi_schedule(napi);
6067 void napi_busy_loop(unsigned int napi_id,
6068 bool (*loop_end)(void *, unsigned long),
6071 unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6072 int (*napi_poll)(struct napi_struct *napi, int budget);
6073 void *have_poll_lock = NULL;
6074 struct napi_struct *napi;
6081 napi = napi_by_id(napi_id);
6091 unsigned long val = READ_ONCE(napi->state);
6093 /* If multiple threads are competing for this napi,
6094 * we avoid dirtying napi->state as much as we can.
6096 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6097 NAPIF_STATE_IN_BUSY_POLL))
6099 if (cmpxchg(&napi->state, val,
6100 val | NAPIF_STATE_IN_BUSY_POLL |
6101 NAPIF_STATE_SCHED) != val)
6103 have_poll_lock = netpoll_poll_lock(napi);
6104 napi_poll = napi->poll;
6106 work = napi_poll(napi, BUSY_POLL_BUDGET);
6107 trace_napi_poll(napi, work, BUSY_POLL_BUDGET);
6110 __NET_ADD_STATS(dev_net(napi->dev),
6111 LINUX_MIB_BUSYPOLLRXPACKETS, work);
6114 if (!loop_end || loop_end(loop_end_arg, start_time))
6117 if (unlikely(need_resched())) {
6119 busy_poll_stop(napi, have_poll_lock);
6123 if (loop_end(loop_end_arg, start_time))
6130 busy_poll_stop(napi, have_poll_lock);
6135 EXPORT_SYMBOL(napi_busy_loop);
6137 #endif /* CONFIG_NET_RX_BUSY_POLL */
6139 static void napi_hash_add(struct napi_struct *napi)
6141 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) ||
6142 test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
6145 spin_lock(&napi_hash_lock);
6147 /* 0..NR_CPUS range is reserved for sender_cpu use */
6149 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6150 napi_gen_id = MIN_NAPI_ID;
6151 } while (napi_by_id(napi_gen_id));
6152 napi->napi_id = napi_gen_id;
6154 hlist_add_head_rcu(&napi->napi_hash_node,
6155 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6157 spin_unlock(&napi_hash_lock);
6160 /* Warning : caller is responsible to make sure rcu grace period
6161 * is respected before freeing memory containing @napi
6163 bool napi_hash_del(struct napi_struct *napi)
6165 bool rcu_sync_needed = false;
6167 spin_lock(&napi_hash_lock);
6169 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) {
6170 rcu_sync_needed = true;
6171 hlist_del_rcu(&napi->napi_hash_node);
6173 spin_unlock(&napi_hash_lock);
6174 return rcu_sync_needed;
6176 EXPORT_SYMBOL_GPL(napi_hash_del);
6178 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6180 struct napi_struct *napi;
6182 napi = container_of(timer, struct napi_struct, timer);
6184 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
6185 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6187 if (napi->gro_bitmask && !napi_disable_pending(napi) &&
6188 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state))
6189 __napi_schedule_irqoff(napi);
6191 return HRTIMER_NORESTART;
6194 static void init_gro_hash(struct napi_struct *napi)
6198 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6199 INIT_LIST_HEAD(&napi->gro_hash[i].list);
6200 napi->gro_hash[i].count = 0;
6202 napi->gro_bitmask = 0;
6205 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
6206 int (*poll)(struct napi_struct *, int), int weight)
6208 INIT_LIST_HEAD(&napi->poll_list);
6209 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6210 napi->timer.function = napi_watchdog;
6211 init_gro_hash(napi);
6214 if (weight > NAPI_POLL_WEIGHT)
6215 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
6217 napi->weight = weight;
6219 #ifdef CONFIG_NETPOLL
6220 napi->poll_owner = -1;
6222 set_bit(NAPI_STATE_SCHED, &napi->state);
6223 set_bit(NAPI_STATE_NPSVC, &napi->state);
6224 list_add_rcu(&napi->dev_list, &dev->napi_list);
6225 napi_hash_add(napi);
6227 EXPORT_SYMBOL(netif_napi_add);
6229 void napi_disable(struct napi_struct *n)
6232 set_bit(NAPI_STATE_DISABLE, &n->state);
6234 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
6236 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
6239 hrtimer_cancel(&n->timer);
6241 clear_bit(NAPI_STATE_DISABLE, &n->state);
6243 EXPORT_SYMBOL(napi_disable);
6245 static void flush_gro_hash(struct napi_struct *napi)
6249 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6250 struct sk_buff *skb, *n;
6252 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6254 napi->gro_hash[i].count = 0;
6258 /* Must be called in process context */
6259 void netif_napi_del(struct napi_struct *napi)
6262 if (napi_hash_del(napi))
6264 list_del_init(&napi->dev_list);
6265 napi_free_frags(napi);
6267 flush_gro_hash(napi);
6268 napi->gro_bitmask = 0;
6270 EXPORT_SYMBOL(netif_napi_del);
6272 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
6277 list_del_init(&n->poll_list);
6279 have = netpoll_poll_lock(n);
6283 /* This NAPI_STATE_SCHED test is for avoiding a race
6284 * with netpoll's poll_napi(). Only the entity which
6285 * obtains the lock and sees NAPI_STATE_SCHED set will
6286 * actually make the ->poll() call. Therefore we avoid
6287 * accidentally calling ->poll() when NAPI is not scheduled.
6290 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
6291 work = n->poll(n, weight);
6292 trace_napi_poll(n, work, weight);
6295 WARN_ON_ONCE(work > weight);
6297 if (likely(work < weight))
6300 /* Drivers must not modify the NAPI state if they
6301 * consume the entire weight. In such cases this code
6302 * still "owns" the NAPI instance and therefore can
6303 * move the instance around on the list at-will.
6305 if (unlikely(napi_disable_pending(n))) {
6310 if (n->gro_bitmask) {
6311 /* flush too old packets
6312 * If HZ < 1000, flush all packets.
6314 napi_gro_flush(n, HZ >= 1000);
6317 /* Some drivers may have called napi_schedule
6318 * prior to exhausting their budget.
6320 if (unlikely(!list_empty(&n->poll_list))) {
6321 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
6322 n->dev ? n->dev->name : "backlog");
6326 list_add_tail(&n->poll_list, repoll);
6329 netpoll_poll_unlock(have);
6334 static __latent_entropy void net_rx_action(struct softirq_action *h)
6336 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6337 unsigned long time_limit = jiffies +
6338 usecs_to_jiffies(netdev_budget_usecs);
6339 int budget = netdev_budget;
6343 local_irq_disable();
6344 list_splice_init(&sd->poll_list, &list);
6348 struct napi_struct *n;
6350 if (list_empty(&list)) {
6351 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
6356 n = list_first_entry(&list, struct napi_struct, poll_list);
6357 budget -= napi_poll(n, &repoll);
6359 /* If softirq window is exhausted then punt.
6360 * Allow this to run for 2 jiffies since which will allow
6361 * an average latency of 1.5/HZ.
6363 if (unlikely(budget <= 0 ||
6364 time_after_eq(jiffies, time_limit))) {
6370 local_irq_disable();
6372 list_splice_tail_init(&sd->poll_list, &list);
6373 list_splice_tail(&repoll, &list);
6374 list_splice(&list, &sd->poll_list);
6375 if (!list_empty(&sd->poll_list))
6376 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
6378 net_rps_action_and_irq_enable(sd);
6380 __kfree_skb_flush();
6383 struct netdev_adjacent {
6384 struct net_device *dev;
6386 /* upper master flag, there can only be one master device per list */
6389 /* counter for the number of times this device was added to us */
6392 /* private field for the users */
6395 struct list_head list;
6396 struct rcu_head rcu;
6399 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
6400 struct list_head *adj_list)
6402 struct netdev_adjacent *adj;
6404 list_for_each_entry(adj, adj_list, list) {
6405 if (adj->dev == adj_dev)
6411 static int __netdev_has_upper_dev(struct net_device *upper_dev, void *data)
6413 struct net_device *dev = data;
6415 return upper_dev == dev;
6419 * netdev_has_upper_dev - Check if device is linked to an upper device
6421 * @upper_dev: upper device to check
6423 * Find out if a device is linked to specified upper device and return true
6424 * in case it is. Note that this checks only immediate upper device,
6425 * not through a complete stack of devices. The caller must hold the RTNL lock.
6427 bool netdev_has_upper_dev(struct net_device *dev,
6428 struct net_device *upper_dev)
6432 return netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
6435 EXPORT_SYMBOL(netdev_has_upper_dev);
6438 * netdev_has_upper_dev_all - Check if device is linked to an upper device
6440 * @upper_dev: upper device to check
6442 * Find out if a device is linked to specified upper device and return true
6443 * in case it is. Note that this checks the entire upper device chain.
6444 * The caller must hold rcu lock.
6447 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
6448 struct net_device *upper_dev)
6450 return !!netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
6453 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
6456 * netdev_has_any_upper_dev - Check if device is linked to some device
6459 * Find out if a device is linked to an upper device and return true in case
6460 * it is. The caller must hold the RTNL lock.
6462 bool netdev_has_any_upper_dev(struct net_device *dev)
6466 return !list_empty(&dev->adj_list.upper);
6468 EXPORT_SYMBOL(netdev_has_any_upper_dev);
6471 * netdev_master_upper_dev_get - Get master upper device
6474 * Find a master upper device and return pointer to it or NULL in case
6475 * it's not there. The caller must hold the RTNL lock.
6477 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
6479 struct netdev_adjacent *upper;
6483 if (list_empty(&dev->adj_list.upper))
6486 upper = list_first_entry(&dev->adj_list.upper,
6487 struct netdev_adjacent, list);
6488 if (likely(upper->master))
6492 EXPORT_SYMBOL(netdev_master_upper_dev_get);
6495 * netdev_has_any_lower_dev - Check if device is linked to some device
6498 * Find out if a device is linked to a lower device and return true in case
6499 * it is. The caller must hold the RTNL lock.
6501 static bool netdev_has_any_lower_dev(struct net_device *dev)
6505 return !list_empty(&dev->adj_list.lower);
6508 void *netdev_adjacent_get_private(struct list_head *adj_list)
6510 struct netdev_adjacent *adj;
6512 adj = list_entry(adj_list, struct netdev_adjacent, list);
6514 return adj->private;
6516 EXPORT_SYMBOL(netdev_adjacent_get_private);
6519 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
6521 * @iter: list_head ** of the current position
6523 * Gets the next device from the dev's upper list, starting from iter
6524 * position. The caller must hold RCU read lock.
6526 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
6527 struct list_head **iter)
6529 struct netdev_adjacent *upper;
6531 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6533 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6535 if (&upper->list == &dev->adj_list.upper)
6538 *iter = &upper->list;
6542 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
6544 static struct net_device *netdev_next_upper_dev(struct net_device *dev,
6545 struct list_head **iter)
6547 struct netdev_adjacent *upper;
6549 upper = list_entry((*iter)->next, struct netdev_adjacent, list);
6551 if (&upper->list == &dev->adj_list.upper)
6554 *iter = &upper->list;
6559 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
6560 struct list_head **iter)
6562 struct netdev_adjacent *upper;
6564 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6566 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6568 if (&upper->list == &dev->adj_list.upper)
6571 *iter = &upper->list;
6576 static int netdev_walk_all_upper_dev(struct net_device *dev,
6577 int (*fn)(struct net_device *dev,
6581 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6582 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6586 iter = &dev->adj_list.upper;
6590 ret = fn(now, data);
6597 udev = netdev_next_upper_dev(now, &iter);
6602 niter = &udev->adj_list.upper;
6603 dev_stack[cur] = now;
6604 iter_stack[cur++] = iter;
6611 next = dev_stack[--cur];
6612 niter = iter_stack[cur];
6622 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
6623 int (*fn)(struct net_device *dev,
6627 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6628 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6632 iter = &dev->adj_list.upper;
6636 ret = fn(now, data);
6643 udev = netdev_next_upper_dev_rcu(now, &iter);
6648 niter = &udev->adj_list.upper;
6649 dev_stack[cur] = now;
6650 iter_stack[cur++] = iter;
6657 next = dev_stack[--cur];
6658 niter = iter_stack[cur];
6667 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
6670 * netdev_lower_get_next_private - Get the next ->private from the
6671 * lower neighbour list
6673 * @iter: list_head ** of the current position
6675 * Gets the next netdev_adjacent->private from the dev's lower neighbour
6676 * list, starting from iter position. The caller must hold either hold the
6677 * RTNL lock or its own locking that guarantees that the neighbour lower
6678 * list will remain unchanged.
6680 void *netdev_lower_get_next_private(struct net_device *dev,
6681 struct list_head **iter)
6683 struct netdev_adjacent *lower;
6685 lower = list_entry(*iter, struct netdev_adjacent, list);
6687 if (&lower->list == &dev->adj_list.lower)
6690 *iter = lower->list.next;
6692 return lower->private;
6694 EXPORT_SYMBOL(netdev_lower_get_next_private);
6697 * netdev_lower_get_next_private_rcu - Get the next ->private from the
6698 * lower neighbour list, RCU
6701 * @iter: list_head ** of the current position
6703 * Gets the next netdev_adjacent->private from the dev's lower neighbour
6704 * list, starting from iter position. The caller must hold RCU read lock.
6706 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
6707 struct list_head **iter)
6709 struct netdev_adjacent *lower;
6711 WARN_ON_ONCE(!rcu_read_lock_held());
6713 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6715 if (&lower->list == &dev->adj_list.lower)
6718 *iter = &lower->list;
6720 return lower->private;
6722 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
6725 * netdev_lower_get_next - Get the next device from the lower neighbour
6728 * @iter: list_head ** of the current position
6730 * Gets the next netdev_adjacent from the dev's lower neighbour
6731 * list, starting from iter position. The caller must hold RTNL lock or
6732 * its own locking that guarantees that the neighbour lower
6733 * list will remain unchanged.
6735 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
6737 struct netdev_adjacent *lower;
6739 lower = list_entry(*iter, struct netdev_adjacent, list);
6741 if (&lower->list == &dev->adj_list.lower)
6744 *iter = lower->list.next;
6748 EXPORT_SYMBOL(netdev_lower_get_next);
6750 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
6751 struct list_head **iter)
6753 struct netdev_adjacent *lower;
6755 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
6757 if (&lower->list == &dev->adj_list.lower)
6760 *iter = &lower->list;
6765 int netdev_walk_all_lower_dev(struct net_device *dev,
6766 int (*fn)(struct net_device *dev,
6770 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6771 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6775 iter = &dev->adj_list.lower;
6779 ret = fn(now, data);
6786 ldev = netdev_next_lower_dev(now, &iter);
6791 niter = &ldev->adj_list.lower;
6792 dev_stack[cur] = now;
6793 iter_stack[cur++] = iter;
6800 next = dev_stack[--cur];
6801 niter = iter_stack[cur];
6810 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
6812 static struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
6813 struct list_head **iter)
6815 struct netdev_adjacent *lower;
6817 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6818 if (&lower->list == &dev->adj_list.lower)
6821 *iter = &lower->list;
6826 static u8 __netdev_upper_depth(struct net_device *dev)
6828 struct net_device *udev;
6829 struct list_head *iter;
6832 for (iter = &dev->adj_list.upper,
6833 udev = netdev_next_upper_dev(dev, &iter);
6835 udev = netdev_next_upper_dev(dev, &iter)) {
6836 if (max_depth < udev->upper_level)
6837 max_depth = udev->upper_level;
6843 static u8 __netdev_lower_depth(struct net_device *dev)
6845 struct net_device *ldev;
6846 struct list_head *iter;
6849 for (iter = &dev->adj_list.lower,
6850 ldev = netdev_next_lower_dev(dev, &iter);
6852 ldev = netdev_next_lower_dev(dev, &iter)) {
6853 if (max_depth < ldev->lower_level)
6854 max_depth = ldev->lower_level;
6860 static int __netdev_update_upper_level(struct net_device *dev, void *data)
6862 dev->upper_level = __netdev_upper_depth(dev) + 1;
6866 static int __netdev_update_lower_level(struct net_device *dev, void *data)
6868 dev->lower_level = __netdev_lower_depth(dev) + 1;
6872 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
6873 int (*fn)(struct net_device *dev,
6877 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6878 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6882 iter = &dev->adj_list.lower;
6886 ret = fn(now, data);
6893 ldev = netdev_next_lower_dev_rcu(now, &iter);
6898 niter = &ldev->adj_list.lower;
6899 dev_stack[cur] = now;
6900 iter_stack[cur++] = iter;
6907 next = dev_stack[--cur];
6908 niter = iter_stack[cur];
6917 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
6920 * netdev_lower_get_first_private_rcu - Get the first ->private from the
6921 * lower neighbour list, RCU
6925 * Gets the first netdev_adjacent->private from the dev's lower neighbour
6926 * list. The caller must hold RCU read lock.
6928 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
6930 struct netdev_adjacent *lower;
6932 lower = list_first_or_null_rcu(&dev->adj_list.lower,
6933 struct netdev_adjacent, list);
6935 return lower->private;
6938 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
6941 * netdev_master_upper_dev_get_rcu - Get master upper device
6944 * Find a master upper device and return pointer to it or NULL in case
6945 * it's not there. The caller must hold the RCU read lock.
6947 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
6949 struct netdev_adjacent *upper;
6951 upper = list_first_or_null_rcu(&dev->adj_list.upper,
6952 struct netdev_adjacent, list);
6953 if (upper && likely(upper->master))
6957 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
6959 static int netdev_adjacent_sysfs_add(struct net_device *dev,
6960 struct net_device *adj_dev,
6961 struct list_head *dev_list)
6963 char linkname[IFNAMSIZ+7];
6965 sprintf(linkname, dev_list == &dev->adj_list.upper ?
6966 "upper_%s" : "lower_%s", adj_dev->name);
6967 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
6970 static void netdev_adjacent_sysfs_del(struct net_device *dev,
6972 struct list_head *dev_list)
6974 char linkname[IFNAMSIZ+7];
6976 sprintf(linkname, dev_list == &dev->adj_list.upper ?
6977 "upper_%s" : "lower_%s", name);
6978 sysfs_remove_link(&(dev->dev.kobj), linkname);
6981 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
6982 struct net_device *adj_dev,
6983 struct list_head *dev_list)
6985 return (dev_list == &dev->adj_list.upper ||
6986 dev_list == &dev->adj_list.lower) &&
6987 net_eq(dev_net(dev), dev_net(adj_dev));
6990 static int __netdev_adjacent_dev_insert(struct net_device *dev,
6991 struct net_device *adj_dev,
6992 struct list_head *dev_list,
6993 void *private, bool master)
6995 struct netdev_adjacent *adj;
6998 adj = __netdev_find_adj(adj_dev, dev_list);
7002 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
7003 dev->name, adj_dev->name, adj->ref_nr);
7008 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
7013 adj->master = master;
7015 adj->private = private;
7018 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
7019 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
7021 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
7022 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
7027 /* Ensure that master link is always the first item in list. */
7029 ret = sysfs_create_link(&(dev->dev.kobj),
7030 &(adj_dev->dev.kobj), "master");
7032 goto remove_symlinks;
7034 list_add_rcu(&adj->list, dev_list);
7036 list_add_tail_rcu(&adj->list, dev_list);
7042 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7043 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7051 static void __netdev_adjacent_dev_remove(struct net_device *dev,
7052 struct net_device *adj_dev,
7054 struct list_head *dev_list)
7056 struct netdev_adjacent *adj;
7058 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
7059 dev->name, adj_dev->name, ref_nr);
7061 adj = __netdev_find_adj(adj_dev, dev_list);
7064 pr_err("Adjacency does not exist for device %s from %s\n",
7065 dev->name, adj_dev->name);
7070 if (adj->ref_nr > ref_nr) {
7071 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
7072 dev->name, adj_dev->name, ref_nr,
7073 adj->ref_nr - ref_nr);
7074 adj->ref_nr -= ref_nr;
7079 sysfs_remove_link(&(dev->dev.kobj), "master");
7081 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7082 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7084 list_del_rcu(&adj->list);
7085 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
7086 adj_dev->name, dev->name, adj_dev->name);
7088 kfree_rcu(adj, rcu);
7091 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
7092 struct net_device *upper_dev,
7093 struct list_head *up_list,
7094 struct list_head *down_list,
7095 void *private, bool master)
7099 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
7104 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
7107 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
7114 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
7115 struct net_device *upper_dev,
7117 struct list_head *up_list,
7118 struct list_head *down_list)
7120 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
7121 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
7124 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
7125 struct net_device *upper_dev,
7126 void *private, bool master)
7128 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
7129 &dev->adj_list.upper,
7130 &upper_dev->adj_list.lower,
7134 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
7135 struct net_device *upper_dev)
7137 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
7138 &dev->adj_list.upper,
7139 &upper_dev->adj_list.lower);
7142 static int __netdev_upper_dev_link(struct net_device *dev,
7143 struct net_device *upper_dev, bool master,
7144 void *upper_priv, void *upper_info,
7145 struct netlink_ext_ack *extack)
7147 struct netdev_notifier_changeupper_info changeupper_info = {
7152 .upper_dev = upper_dev,
7155 .upper_info = upper_info,
7157 struct net_device *master_dev;
7162 if (dev == upper_dev)
7165 /* To prevent loops, check if dev is not upper device to upper_dev. */
7166 if (netdev_has_upper_dev(upper_dev, dev))
7169 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
7173 if (netdev_has_upper_dev(dev, upper_dev))
7176 master_dev = netdev_master_upper_dev_get(dev);
7178 return master_dev == upper_dev ? -EEXIST : -EBUSY;
7181 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7182 &changeupper_info.info);
7183 ret = notifier_to_errno(ret);
7187 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
7192 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7193 &changeupper_info.info);
7194 ret = notifier_to_errno(ret);
7198 __netdev_update_upper_level(dev, NULL);
7199 netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7201 __netdev_update_lower_level(upper_dev, NULL);
7202 netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, NULL);
7207 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7213 * netdev_upper_dev_link - Add a link to the upper device
7215 * @upper_dev: new upper device
7216 * @extack: netlink extended ack
7218 * Adds a link to device which is upper to this one. The caller must hold
7219 * the RTNL lock. On a failure a negative errno code is returned.
7220 * On success the reference counts are adjusted and the function
7223 int netdev_upper_dev_link(struct net_device *dev,
7224 struct net_device *upper_dev,
7225 struct netlink_ext_ack *extack)
7227 return __netdev_upper_dev_link(dev, upper_dev, false,
7228 NULL, NULL, extack);
7230 EXPORT_SYMBOL(netdev_upper_dev_link);
7233 * netdev_master_upper_dev_link - Add a master link to the upper device
7235 * @upper_dev: new upper device
7236 * @upper_priv: upper device private
7237 * @upper_info: upper info to be passed down via notifier
7238 * @extack: netlink extended ack
7240 * Adds a link to device which is upper to this one. In this case, only
7241 * one master upper device can be linked, although other non-master devices
7242 * might be linked as well. The caller must hold the RTNL lock.
7243 * On a failure a negative errno code is returned. On success the reference
7244 * counts are adjusted and the function returns zero.
7246 int netdev_master_upper_dev_link(struct net_device *dev,
7247 struct net_device *upper_dev,
7248 void *upper_priv, void *upper_info,
7249 struct netlink_ext_ack *extack)
7251 return __netdev_upper_dev_link(dev, upper_dev, true,
7252 upper_priv, upper_info, extack);
7254 EXPORT_SYMBOL(netdev_master_upper_dev_link);
7257 * netdev_upper_dev_unlink - Removes a link to upper device
7259 * @upper_dev: new upper device
7261 * Removes a link to device which is upper to this one. The caller must hold
7264 void netdev_upper_dev_unlink(struct net_device *dev,
7265 struct net_device *upper_dev)
7267 struct netdev_notifier_changeupper_info changeupper_info = {
7271 .upper_dev = upper_dev,
7277 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
7279 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7280 &changeupper_info.info);
7282 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7284 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7285 &changeupper_info.info);
7287 __netdev_update_upper_level(dev, NULL);
7288 netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7290 __netdev_update_lower_level(upper_dev, NULL);
7291 netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, NULL);
7293 EXPORT_SYMBOL(netdev_upper_dev_unlink);
7296 * netdev_bonding_info_change - Dispatch event about slave change
7298 * @bonding_info: info to dispatch
7300 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
7301 * The caller must hold the RTNL lock.
7303 void netdev_bonding_info_change(struct net_device *dev,
7304 struct netdev_bonding_info *bonding_info)
7306 struct netdev_notifier_bonding_info info = {
7310 memcpy(&info.bonding_info, bonding_info,
7311 sizeof(struct netdev_bonding_info));
7312 call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
7315 EXPORT_SYMBOL(netdev_bonding_info_change);
7317 static void netdev_adjacent_add_links(struct net_device *dev)
7319 struct netdev_adjacent *iter;
7321 struct net *net = dev_net(dev);
7323 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7324 if (!net_eq(net, dev_net(iter->dev)))
7326 netdev_adjacent_sysfs_add(iter->dev, dev,
7327 &iter->dev->adj_list.lower);
7328 netdev_adjacent_sysfs_add(dev, iter->dev,
7329 &dev->adj_list.upper);
7332 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7333 if (!net_eq(net, dev_net(iter->dev)))
7335 netdev_adjacent_sysfs_add(iter->dev, dev,
7336 &iter->dev->adj_list.upper);
7337 netdev_adjacent_sysfs_add(dev, iter->dev,
7338 &dev->adj_list.lower);
7342 static void netdev_adjacent_del_links(struct net_device *dev)
7344 struct netdev_adjacent *iter;
7346 struct net *net = dev_net(dev);
7348 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7349 if (!net_eq(net, dev_net(iter->dev)))
7351 netdev_adjacent_sysfs_del(iter->dev, dev->name,
7352 &iter->dev->adj_list.lower);
7353 netdev_adjacent_sysfs_del(dev, iter->dev->name,
7354 &dev->adj_list.upper);
7357 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7358 if (!net_eq(net, dev_net(iter->dev)))
7360 netdev_adjacent_sysfs_del(iter->dev, dev->name,
7361 &iter->dev->adj_list.upper);
7362 netdev_adjacent_sysfs_del(dev, iter->dev->name,
7363 &dev->adj_list.lower);
7367 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
7369 struct netdev_adjacent *iter;
7371 struct net *net = dev_net(dev);
7373 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7374 if (!net_eq(net, dev_net(iter->dev)))
7376 netdev_adjacent_sysfs_del(iter->dev, oldname,
7377 &iter->dev->adj_list.lower);
7378 netdev_adjacent_sysfs_add(iter->dev, dev,
7379 &iter->dev->adj_list.lower);
7382 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7383 if (!net_eq(net, dev_net(iter->dev)))
7385 netdev_adjacent_sysfs_del(iter->dev, oldname,
7386 &iter->dev->adj_list.upper);
7387 netdev_adjacent_sysfs_add(iter->dev, dev,
7388 &iter->dev->adj_list.upper);
7392 void *netdev_lower_dev_get_private(struct net_device *dev,
7393 struct net_device *lower_dev)
7395 struct netdev_adjacent *lower;
7399 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
7403 return lower->private;
7405 EXPORT_SYMBOL(netdev_lower_dev_get_private);
7408 int dev_get_nest_level(struct net_device *dev)
7410 struct net_device *lower = NULL;
7411 struct list_head *iter;
7417 netdev_for_each_lower_dev(dev, lower, iter) {
7418 nest = dev_get_nest_level(lower);
7419 if (max_nest < nest)
7423 return max_nest + 1;
7425 EXPORT_SYMBOL(dev_get_nest_level);
7428 * netdev_lower_change - Dispatch event about lower device state change
7429 * @lower_dev: device
7430 * @lower_state_info: state to dispatch
7432 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
7433 * The caller must hold the RTNL lock.
7435 void netdev_lower_state_changed(struct net_device *lower_dev,
7436 void *lower_state_info)
7438 struct netdev_notifier_changelowerstate_info changelowerstate_info = {
7439 .info.dev = lower_dev,
7443 changelowerstate_info.lower_state_info = lower_state_info;
7444 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
7445 &changelowerstate_info.info);
7447 EXPORT_SYMBOL(netdev_lower_state_changed);
7449 static void dev_change_rx_flags(struct net_device *dev, int flags)
7451 const struct net_device_ops *ops = dev->netdev_ops;
7453 if (ops->ndo_change_rx_flags)
7454 ops->ndo_change_rx_flags(dev, flags);
7457 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
7459 unsigned int old_flags = dev->flags;
7465 dev->flags |= IFF_PROMISC;
7466 dev->promiscuity += inc;
7467 if (dev->promiscuity == 0) {
7470 * If inc causes overflow, untouch promisc and return error.
7473 dev->flags &= ~IFF_PROMISC;
7475 dev->promiscuity -= inc;
7476 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
7481 if (dev->flags != old_flags) {
7482 pr_info("device %s %s promiscuous mode\n",
7484 dev->flags & IFF_PROMISC ? "entered" : "left");
7485 if (audit_enabled) {
7486 current_uid_gid(&uid, &gid);
7487 audit_log(audit_context(), GFP_ATOMIC,
7488 AUDIT_ANOM_PROMISCUOUS,
7489 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
7490 dev->name, (dev->flags & IFF_PROMISC),
7491 (old_flags & IFF_PROMISC),
7492 from_kuid(&init_user_ns, audit_get_loginuid(current)),
7493 from_kuid(&init_user_ns, uid),
7494 from_kgid(&init_user_ns, gid),
7495 audit_get_sessionid(current));
7498 dev_change_rx_flags(dev, IFF_PROMISC);
7501 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
7506 * dev_set_promiscuity - update promiscuity count on a device
7510 * Add or remove promiscuity from a device. While the count in the device
7511 * remains above zero the interface remains promiscuous. Once it hits zero
7512 * the device reverts back to normal filtering operation. A negative inc
7513 * value is used to drop promiscuity on the device.
7514 * Return 0 if successful or a negative errno code on error.
7516 int dev_set_promiscuity(struct net_device *dev, int inc)
7518 unsigned int old_flags = dev->flags;
7521 err = __dev_set_promiscuity(dev, inc, true);
7524 if (dev->flags != old_flags)
7525 dev_set_rx_mode(dev);
7528 EXPORT_SYMBOL(dev_set_promiscuity);
7530 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
7532 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
7536 dev->flags |= IFF_ALLMULTI;
7537 dev->allmulti += inc;
7538 if (dev->allmulti == 0) {
7541 * If inc causes overflow, untouch allmulti and return error.
7544 dev->flags &= ~IFF_ALLMULTI;
7546 dev->allmulti -= inc;
7547 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
7552 if (dev->flags ^ old_flags) {
7553 dev_change_rx_flags(dev, IFF_ALLMULTI);
7554 dev_set_rx_mode(dev);
7556 __dev_notify_flags(dev, old_flags,
7557 dev->gflags ^ old_gflags);
7563 * dev_set_allmulti - update allmulti count on a device
7567 * Add or remove reception of all multicast frames to a device. While the
7568 * count in the device remains above zero the interface remains listening
7569 * to all interfaces. Once it hits zero the device reverts back to normal
7570 * filtering operation. A negative @inc value is used to drop the counter
7571 * when releasing a resource needing all multicasts.
7572 * Return 0 if successful or a negative errno code on error.
7575 int dev_set_allmulti(struct net_device *dev, int inc)
7577 return __dev_set_allmulti(dev, inc, true);
7579 EXPORT_SYMBOL(dev_set_allmulti);
7582 * Upload unicast and multicast address lists to device and
7583 * configure RX filtering. When the device doesn't support unicast
7584 * filtering it is put in promiscuous mode while unicast addresses
7587 void __dev_set_rx_mode(struct net_device *dev)
7589 const struct net_device_ops *ops = dev->netdev_ops;
7591 /* dev_open will call this function so the list will stay sane. */
7592 if (!(dev->flags&IFF_UP))
7595 if (!netif_device_present(dev))
7598 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
7599 /* Unicast addresses changes may only happen under the rtnl,
7600 * therefore calling __dev_set_promiscuity here is safe.
7602 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
7603 __dev_set_promiscuity(dev, 1, false);
7604 dev->uc_promisc = true;
7605 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
7606 __dev_set_promiscuity(dev, -1, false);
7607 dev->uc_promisc = false;
7611 if (ops->ndo_set_rx_mode)
7612 ops->ndo_set_rx_mode(dev);
7615 void dev_set_rx_mode(struct net_device *dev)
7617 netif_addr_lock_bh(dev);
7618 __dev_set_rx_mode(dev);
7619 netif_addr_unlock_bh(dev);
7623 * dev_get_flags - get flags reported to userspace
7626 * Get the combination of flag bits exported through APIs to userspace.
7628 unsigned int dev_get_flags(const struct net_device *dev)
7632 flags = (dev->flags & ~(IFF_PROMISC |
7637 (dev->gflags & (IFF_PROMISC |
7640 if (netif_running(dev)) {
7641 if (netif_oper_up(dev))
7642 flags |= IFF_RUNNING;
7643 if (netif_carrier_ok(dev))
7644 flags |= IFF_LOWER_UP;
7645 if (netif_dormant(dev))
7646 flags |= IFF_DORMANT;
7651 EXPORT_SYMBOL(dev_get_flags);
7653 int __dev_change_flags(struct net_device *dev, unsigned int flags)
7655 unsigned int old_flags = dev->flags;
7661 * Set the flags on our device.
7664 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
7665 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
7667 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
7671 * Load in the correct multicast list now the flags have changed.
7674 if ((old_flags ^ flags) & IFF_MULTICAST)
7675 dev_change_rx_flags(dev, IFF_MULTICAST);
7677 dev_set_rx_mode(dev);
7680 * Have we downed the interface. We handle IFF_UP ourselves
7681 * according to user attempts to set it, rather than blindly
7686 if ((old_flags ^ flags) & IFF_UP) {
7687 if (old_flags & IFF_UP)
7690 ret = __dev_open(dev);
7693 if ((flags ^ dev->gflags) & IFF_PROMISC) {
7694 int inc = (flags & IFF_PROMISC) ? 1 : -1;
7695 unsigned int old_flags = dev->flags;
7697 dev->gflags ^= IFF_PROMISC;
7699 if (__dev_set_promiscuity(dev, inc, false) >= 0)
7700 if (dev->flags != old_flags)
7701 dev_set_rx_mode(dev);
7704 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
7705 * is important. Some (broken) drivers set IFF_PROMISC, when
7706 * IFF_ALLMULTI is requested not asking us and not reporting.
7708 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
7709 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
7711 dev->gflags ^= IFF_ALLMULTI;
7712 __dev_set_allmulti(dev, inc, false);
7718 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
7719 unsigned int gchanges)
7721 unsigned int changes = dev->flags ^ old_flags;
7724 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
7726 if (changes & IFF_UP) {
7727 if (dev->flags & IFF_UP)
7728 call_netdevice_notifiers(NETDEV_UP, dev);
7730 call_netdevice_notifiers(NETDEV_DOWN, dev);
7733 if (dev->flags & IFF_UP &&
7734 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
7735 struct netdev_notifier_change_info change_info = {
7739 .flags_changed = changes,
7742 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
7747 * dev_change_flags - change device settings
7749 * @flags: device state flags
7751 * Change settings on device based state flags. The flags are
7752 * in the userspace exported format.
7754 int dev_change_flags(struct net_device *dev, unsigned int flags)
7757 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
7759 ret = __dev_change_flags(dev, flags);
7763 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
7764 __dev_notify_flags(dev, old_flags, changes);
7767 EXPORT_SYMBOL(dev_change_flags);
7769 int __dev_set_mtu(struct net_device *dev, int new_mtu)
7771 const struct net_device_ops *ops = dev->netdev_ops;
7773 if (ops->ndo_change_mtu)
7774 return ops->ndo_change_mtu(dev, new_mtu);
7776 /* Pairs with all the lockless reads of dev->mtu in the stack */
7777 WRITE_ONCE(dev->mtu, new_mtu);
7780 EXPORT_SYMBOL(__dev_set_mtu);
7782 int dev_validate_mtu(struct net_device *dev, int new_mtu,
7783 struct netlink_ext_ack *extack)
7785 /* MTU must be positive, and in range */
7786 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
7787 NL_SET_ERR_MSG(extack, "mtu less than device minimum");
7791 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
7792 NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
7799 * dev_set_mtu_ext - Change maximum transfer unit
7801 * @new_mtu: new transfer unit
7802 * @extack: netlink extended ack
7804 * Change the maximum transfer size of the network device.
7806 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
7807 struct netlink_ext_ack *extack)
7811 if (new_mtu == dev->mtu)
7814 err = dev_validate_mtu(dev, new_mtu, extack);
7818 if (!netif_device_present(dev))
7821 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
7822 err = notifier_to_errno(err);
7826 orig_mtu = dev->mtu;
7827 err = __dev_set_mtu(dev, new_mtu);
7830 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
7832 err = notifier_to_errno(err);
7834 /* setting mtu back and notifying everyone again,
7835 * so that they have a chance to revert changes.
7837 __dev_set_mtu(dev, orig_mtu);
7838 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
7845 int dev_set_mtu(struct net_device *dev, int new_mtu)
7847 struct netlink_ext_ack extack;
7850 memset(&extack, 0, sizeof(extack));
7851 err = dev_set_mtu_ext(dev, new_mtu, &extack);
7852 if (err && extack._msg)
7853 net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
7856 EXPORT_SYMBOL(dev_set_mtu);
7859 * dev_change_tx_queue_len - Change TX queue length of a netdevice
7861 * @new_len: new tx queue length
7863 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
7865 unsigned int orig_len = dev->tx_queue_len;
7868 if (new_len != (unsigned int)new_len)
7871 if (new_len != orig_len) {
7872 dev->tx_queue_len = new_len;
7873 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
7874 res = notifier_to_errno(res);
7877 res = dev_qdisc_change_tx_queue_len(dev);
7885 netdev_err(dev, "refused to change device tx_queue_len\n");
7886 dev->tx_queue_len = orig_len;
7891 * dev_set_group - Change group this device belongs to
7893 * @new_group: group this device should belong to
7895 void dev_set_group(struct net_device *dev, int new_group)
7897 dev->group = new_group;
7899 EXPORT_SYMBOL(dev_set_group);
7902 * dev_set_mac_address - Change Media Access Control Address
7906 * Change the hardware (MAC) address of the device
7908 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
7910 const struct net_device_ops *ops = dev->netdev_ops;
7913 if (!ops->ndo_set_mac_address)
7915 if (sa->sa_family != dev->type)
7917 if (!netif_device_present(dev))
7919 err = ops->ndo_set_mac_address(dev, sa);
7922 dev->addr_assign_type = NET_ADDR_SET;
7923 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
7924 add_device_randomness(dev->dev_addr, dev->addr_len);
7927 EXPORT_SYMBOL(dev_set_mac_address);
7930 * dev_change_carrier - Change device carrier
7932 * @new_carrier: new value
7934 * Change device carrier
7936 int dev_change_carrier(struct net_device *dev, bool new_carrier)
7938 const struct net_device_ops *ops = dev->netdev_ops;
7940 if (!ops->ndo_change_carrier)
7942 if (!netif_device_present(dev))
7944 return ops->ndo_change_carrier(dev, new_carrier);
7946 EXPORT_SYMBOL(dev_change_carrier);
7949 * dev_get_phys_port_id - Get device physical port ID
7953 * Get device physical port ID
7955 int dev_get_phys_port_id(struct net_device *dev,
7956 struct netdev_phys_item_id *ppid)
7958 const struct net_device_ops *ops = dev->netdev_ops;
7960 if (!ops->ndo_get_phys_port_id)
7962 return ops->ndo_get_phys_port_id(dev, ppid);
7964 EXPORT_SYMBOL(dev_get_phys_port_id);
7967 * dev_get_phys_port_name - Get device physical port name
7970 * @len: limit of bytes to copy to name
7972 * Get device physical port name
7974 int dev_get_phys_port_name(struct net_device *dev,
7975 char *name, size_t len)
7977 const struct net_device_ops *ops = dev->netdev_ops;
7979 if (!ops->ndo_get_phys_port_name)
7981 return ops->ndo_get_phys_port_name(dev, name, len);
7983 EXPORT_SYMBOL(dev_get_phys_port_name);
7986 * dev_change_proto_down - update protocol port state information
7988 * @proto_down: new value
7990 * This info can be used by switch drivers to set the phys state of the
7993 int dev_change_proto_down(struct net_device *dev, bool proto_down)
7995 const struct net_device_ops *ops = dev->netdev_ops;
7997 if (!ops->ndo_change_proto_down)
7999 if (!netif_device_present(dev))
8001 return ops->ndo_change_proto_down(dev, proto_down);
8003 EXPORT_SYMBOL(dev_change_proto_down);
8005 u32 __dev_xdp_query(struct net_device *dev, bpf_op_t bpf_op,
8006 enum bpf_netdev_command cmd)
8008 struct netdev_bpf xdp;
8013 memset(&xdp, 0, sizeof(xdp));
8016 /* Query must always succeed. */
8017 WARN_ON(bpf_op(dev, &xdp) < 0 && cmd == XDP_QUERY_PROG);
8022 static int dev_xdp_install(struct net_device *dev, bpf_op_t bpf_op,
8023 struct netlink_ext_ack *extack, u32 flags,
8024 struct bpf_prog *prog)
8026 struct netdev_bpf xdp;
8028 memset(&xdp, 0, sizeof(xdp));
8029 if (flags & XDP_FLAGS_HW_MODE)
8030 xdp.command = XDP_SETUP_PROG_HW;
8032 xdp.command = XDP_SETUP_PROG;
8033 xdp.extack = extack;
8037 return bpf_op(dev, &xdp);
8040 static void dev_xdp_uninstall(struct net_device *dev)
8042 struct netdev_bpf xdp;
8045 /* Remove generic XDP */
8046 WARN_ON(dev_xdp_install(dev, generic_xdp_install, NULL, 0, NULL));
8048 /* Remove from the driver */
8049 ndo_bpf = dev->netdev_ops->ndo_bpf;
8053 memset(&xdp, 0, sizeof(xdp));
8054 xdp.command = XDP_QUERY_PROG;
8055 WARN_ON(ndo_bpf(dev, &xdp));
8057 WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags,
8060 /* Remove HW offload */
8061 memset(&xdp, 0, sizeof(xdp));
8062 xdp.command = XDP_QUERY_PROG_HW;
8063 if (!ndo_bpf(dev, &xdp) && xdp.prog_id)
8064 WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags,
8069 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
8071 * @extack: netlink extended ack
8072 * @fd: new program fd or negative value to clear
8073 * @flags: xdp-related flags
8075 * Set or clear a bpf program for a device
8077 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
8080 const struct net_device_ops *ops = dev->netdev_ops;
8081 enum bpf_netdev_command query;
8082 struct bpf_prog *prog = NULL;
8083 bpf_op_t bpf_op, bpf_chk;
8088 query = flags & XDP_FLAGS_HW_MODE ? XDP_QUERY_PROG_HW : XDP_QUERY_PROG;
8090 bpf_op = bpf_chk = ops->ndo_bpf;
8091 if (!bpf_op && (flags & (XDP_FLAGS_DRV_MODE | XDP_FLAGS_HW_MODE)))
8093 if (!bpf_op || (flags & XDP_FLAGS_SKB_MODE))
8094 bpf_op = generic_xdp_install;
8095 if (bpf_op == bpf_chk)
8096 bpf_chk = generic_xdp_install;
8099 if (__dev_xdp_query(dev, bpf_chk, XDP_QUERY_PROG) ||
8100 __dev_xdp_query(dev, bpf_chk, XDP_QUERY_PROG_HW))
8102 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) &&
8103 __dev_xdp_query(dev, bpf_op, query))
8106 prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
8107 bpf_op == ops->ndo_bpf);
8109 return PTR_ERR(prog);
8111 if (!(flags & XDP_FLAGS_HW_MODE) &&
8112 bpf_prog_is_dev_bound(prog->aux)) {
8113 NL_SET_ERR_MSG(extack, "using device-bound program without HW_MODE flag is not supported");
8119 err = dev_xdp_install(dev, bpf_op, extack, flags, prog);
8120 if (err < 0 && prog)
8127 * dev_new_index - allocate an ifindex
8128 * @net: the applicable net namespace
8130 * Returns a suitable unique value for a new device interface
8131 * number. The caller must hold the rtnl semaphore or the
8132 * dev_base_lock to be sure it remains unique.
8134 static int dev_new_index(struct net *net)
8136 int ifindex = net->ifindex;
8141 if (!__dev_get_by_index(net, ifindex))
8142 return net->ifindex = ifindex;
8146 /* Delayed registration/unregisteration */
8147 static LIST_HEAD(net_todo_list);
8148 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
8150 static void net_set_todo(struct net_device *dev)
8152 list_add_tail(&dev->todo_list, &net_todo_list);
8153 dev_net(dev)->dev_unreg_count++;
8156 static void rollback_registered_many(struct list_head *head)
8158 struct net_device *dev, *tmp;
8159 LIST_HEAD(close_head);
8161 BUG_ON(dev_boot_phase);
8164 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
8165 /* Some devices call without registering
8166 * for initialization unwind. Remove those
8167 * devices and proceed with the remaining.
8169 if (dev->reg_state == NETREG_UNINITIALIZED) {
8170 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
8174 list_del(&dev->unreg_list);
8177 dev->dismantle = true;
8178 BUG_ON(dev->reg_state != NETREG_REGISTERED);
8181 /* If device is running, close it first. */
8182 list_for_each_entry(dev, head, unreg_list)
8183 list_add_tail(&dev->close_list, &close_head);
8184 dev_close_many(&close_head, true);
8186 list_for_each_entry(dev, head, unreg_list) {
8187 /* And unlink it from device chain. */
8188 unlist_netdevice(dev);
8190 dev->reg_state = NETREG_UNREGISTERING;
8192 flush_all_backlogs();
8196 list_for_each_entry(dev, head, unreg_list) {
8197 struct sk_buff *skb = NULL;
8199 /* Shutdown queueing discipline. */
8202 dev_xdp_uninstall(dev);
8204 /* Notify protocols, that we are about to destroy
8205 * this device. They should clean all the things.
8207 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
8209 if (!dev->rtnl_link_ops ||
8210 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
8211 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
8212 GFP_KERNEL, NULL, 0);
8215 * Flush the unicast and multicast chains
8220 if (dev->netdev_ops->ndo_uninit)
8221 dev->netdev_ops->ndo_uninit(dev);
8224 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
8226 /* Notifier chain MUST detach us all upper devices. */
8227 WARN_ON(netdev_has_any_upper_dev(dev));
8228 WARN_ON(netdev_has_any_lower_dev(dev));
8230 /* Remove entries from kobject tree */
8231 netdev_unregister_kobject(dev);
8233 /* Remove XPS queueing entries */
8234 netif_reset_xps_queues_gt(dev, 0);
8240 list_for_each_entry(dev, head, unreg_list)
8244 static void rollback_registered(struct net_device *dev)
8248 list_add(&dev->unreg_list, &single);
8249 rollback_registered_many(&single);
8253 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
8254 struct net_device *upper, netdev_features_t features)
8256 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
8257 netdev_features_t feature;
8260 for_each_netdev_feature(upper_disables, feature_bit) {
8261 feature = __NETIF_F_BIT(feature_bit);
8262 if (!(upper->wanted_features & feature)
8263 && (features & feature)) {
8264 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
8265 &feature, upper->name);
8266 features &= ~feature;
8273 static void netdev_sync_lower_features(struct net_device *upper,
8274 struct net_device *lower, netdev_features_t features)
8276 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
8277 netdev_features_t feature;
8280 for_each_netdev_feature(upper_disables, feature_bit) {
8281 feature = __NETIF_F_BIT(feature_bit);
8282 if (!(features & feature) && (lower->features & feature)) {
8283 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
8284 &feature, lower->name);
8285 lower->wanted_features &= ~feature;
8286 __netdev_update_features(lower);
8288 if (unlikely(lower->features & feature))
8289 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
8290 &feature, lower->name);
8292 netdev_features_change(lower);
8297 static netdev_features_t netdev_fix_features(struct net_device *dev,
8298 netdev_features_t features)
8300 /* Fix illegal checksum combinations */
8301 if ((features & NETIF_F_HW_CSUM) &&
8302 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
8303 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
8304 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
8307 /* TSO requires that SG is present as well. */
8308 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
8309 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
8310 features &= ~NETIF_F_ALL_TSO;
8313 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
8314 !(features & NETIF_F_IP_CSUM)) {
8315 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
8316 features &= ~NETIF_F_TSO;
8317 features &= ~NETIF_F_TSO_ECN;
8320 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
8321 !(features & NETIF_F_IPV6_CSUM)) {
8322 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
8323 features &= ~NETIF_F_TSO6;
8326 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
8327 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
8328 features &= ~NETIF_F_TSO_MANGLEID;
8330 /* TSO ECN requires that TSO is present as well. */
8331 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
8332 features &= ~NETIF_F_TSO_ECN;
8334 /* Software GSO depends on SG. */
8335 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
8336 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
8337 features &= ~NETIF_F_GSO;
8340 /* GSO partial features require GSO partial be set */
8341 if ((features & dev->gso_partial_features) &&
8342 !(features & NETIF_F_GSO_PARTIAL)) {
8344 "Dropping partially supported GSO features since no GSO partial.\n");
8345 features &= ~dev->gso_partial_features;
8348 if (!(features & NETIF_F_RXCSUM)) {
8349 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet
8350 * successfully merged by hardware must also have the
8351 * checksum verified by hardware. If the user does not
8352 * want to enable RXCSUM, logically, we should disable GRO_HW.
8354 if (features & NETIF_F_GRO_HW) {
8355 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
8356 features &= ~NETIF_F_GRO_HW;
8360 /* LRO/HW-GRO features cannot be combined with RX-FCS */
8361 if (features & NETIF_F_RXFCS) {
8362 if (features & NETIF_F_LRO) {
8363 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
8364 features &= ~NETIF_F_LRO;
8367 if (features & NETIF_F_GRO_HW) {
8368 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
8369 features &= ~NETIF_F_GRO_HW;
8373 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) {
8374 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n");
8375 features &= ~NETIF_F_HW_TLS_RX;
8381 int __netdev_update_features(struct net_device *dev)
8383 struct net_device *upper, *lower;
8384 netdev_features_t features;
8385 struct list_head *iter;
8390 features = netdev_get_wanted_features(dev);
8392 if (dev->netdev_ops->ndo_fix_features)
8393 features = dev->netdev_ops->ndo_fix_features(dev, features);
8395 /* driver might be less strict about feature dependencies */
8396 features = netdev_fix_features(dev, features);
8398 /* some features can't be enabled if they're off an an upper device */
8399 netdev_for_each_upper_dev_rcu(dev, upper, iter)
8400 features = netdev_sync_upper_features(dev, upper, features);
8402 if (dev->features == features)
8405 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
8406 &dev->features, &features);
8408 if (dev->netdev_ops->ndo_set_features)
8409 err = dev->netdev_ops->ndo_set_features(dev, features);
8413 if (unlikely(err < 0)) {
8415 "set_features() failed (%d); wanted %pNF, left %pNF\n",
8416 err, &features, &dev->features);
8417 /* return non-0 since some features might have changed and
8418 * it's better to fire a spurious notification than miss it
8424 /* some features must be disabled on lower devices when disabled
8425 * on an upper device (think: bonding master or bridge)
8427 netdev_for_each_lower_dev(dev, lower, iter)
8428 netdev_sync_lower_features(dev, lower, features);
8431 netdev_features_t diff = features ^ dev->features;
8433 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
8434 /* udp_tunnel_{get,drop}_rx_info both need
8435 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
8436 * device, or they won't do anything.
8437 * Thus we need to update dev->features
8438 * *before* calling udp_tunnel_get_rx_info,
8439 * but *after* calling udp_tunnel_drop_rx_info.
8441 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
8442 dev->features = features;
8443 udp_tunnel_get_rx_info(dev);
8445 udp_tunnel_drop_rx_info(dev);
8449 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
8450 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
8451 dev->features = features;
8452 err |= vlan_get_rx_ctag_filter_info(dev);
8454 vlan_drop_rx_ctag_filter_info(dev);
8458 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
8459 if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
8460 dev->features = features;
8461 err |= vlan_get_rx_stag_filter_info(dev);
8463 vlan_drop_rx_stag_filter_info(dev);
8467 dev->features = features;
8470 return err < 0 ? 0 : 1;
8474 * netdev_update_features - recalculate device features
8475 * @dev: the device to check
8477 * Recalculate dev->features set and send notifications if it
8478 * has changed. Should be called after driver or hardware dependent
8479 * conditions might have changed that influence the features.
8481 void netdev_update_features(struct net_device *dev)
8483 if (__netdev_update_features(dev))
8484 netdev_features_change(dev);
8486 EXPORT_SYMBOL(netdev_update_features);
8489 * netdev_change_features - recalculate device features
8490 * @dev: the device to check
8492 * Recalculate dev->features set and send notifications even
8493 * if they have not changed. Should be called instead of
8494 * netdev_update_features() if also dev->vlan_features might
8495 * have changed to allow the changes to be propagated to stacked
8498 void netdev_change_features(struct net_device *dev)
8500 __netdev_update_features(dev);
8501 netdev_features_change(dev);
8503 EXPORT_SYMBOL(netdev_change_features);
8506 * netif_stacked_transfer_operstate - transfer operstate
8507 * @rootdev: the root or lower level device to transfer state from
8508 * @dev: the device to transfer operstate to
8510 * Transfer operational state from root to device. This is normally
8511 * called when a stacking relationship exists between the root
8512 * device and the device(a leaf device).
8514 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
8515 struct net_device *dev)
8517 if (rootdev->operstate == IF_OPER_DORMANT)
8518 netif_dormant_on(dev);
8520 netif_dormant_off(dev);
8522 if (netif_carrier_ok(rootdev))
8523 netif_carrier_on(dev);
8525 netif_carrier_off(dev);
8527 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
8529 static int netif_alloc_rx_queues(struct net_device *dev)
8531 unsigned int i, count = dev->num_rx_queues;
8532 struct netdev_rx_queue *rx;
8533 size_t sz = count * sizeof(*rx);
8538 rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
8544 for (i = 0; i < count; i++) {
8547 /* XDP RX-queue setup */
8548 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i);
8555 /* Rollback successful reg's and free other resources */
8557 xdp_rxq_info_unreg(&rx[i].xdp_rxq);
8563 static void netif_free_rx_queues(struct net_device *dev)
8565 unsigned int i, count = dev->num_rx_queues;
8567 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
8571 for (i = 0; i < count; i++)
8572 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
8577 static void netdev_init_one_queue(struct net_device *dev,
8578 struct netdev_queue *queue, void *_unused)
8580 /* Initialize queue lock */
8581 spin_lock_init(&queue->_xmit_lock);
8582 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
8583 queue->xmit_lock_owner = -1;
8584 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
8587 dql_init(&queue->dql, HZ);
8591 static void netif_free_tx_queues(struct net_device *dev)
8596 static int netif_alloc_netdev_queues(struct net_device *dev)
8598 unsigned int count = dev->num_tx_queues;
8599 struct netdev_queue *tx;
8600 size_t sz = count * sizeof(*tx);
8602 if (count < 1 || count > 0xffff)
8605 tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
8611 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
8612 spin_lock_init(&dev->tx_global_lock);
8617 void netif_tx_stop_all_queues(struct net_device *dev)
8621 for (i = 0; i < dev->num_tx_queues; i++) {
8622 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
8624 netif_tx_stop_queue(txq);
8627 EXPORT_SYMBOL(netif_tx_stop_all_queues);
8630 * register_netdevice - register a network device
8631 * @dev: device to register
8633 * Take a completed network device structure and add it to the kernel
8634 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
8635 * chain. 0 is returned on success. A negative errno code is returned
8636 * on a failure to set up the device, or if the name is a duplicate.
8638 * Callers must hold the rtnl semaphore. You may want
8639 * register_netdev() instead of this.
8642 * The locking appears insufficient to guarantee two parallel registers
8643 * will not get the same name.
8646 int register_netdevice(struct net_device *dev)
8649 struct net *net = dev_net(dev);
8651 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
8652 NETDEV_FEATURE_COUNT);
8653 BUG_ON(dev_boot_phase);
8658 /* When net_device's are persistent, this will be fatal. */
8659 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
8662 spin_lock_init(&dev->addr_list_lock);
8663 netdev_set_addr_lockdep_class(dev);
8665 ret = dev_get_valid_name(net, dev, dev->name);
8669 /* Init, if this function is available */
8670 if (dev->netdev_ops->ndo_init) {
8671 ret = dev->netdev_ops->ndo_init(dev);
8679 if (((dev->hw_features | dev->features) &
8680 NETIF_F_HW_VLAN_CTAG_FILTER) &&
8681 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
8682 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
8683 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
8690 dev->ifindex = dev_new_index(net);
8691 else if (__dev_get_by_index(net, dev->ifindex))
8694 /* Transfer changeable features to wanted_features and enable
8695 * software offloads (GSO and GRO).
8697 dev->hw_features |= NETIF_F_SOFT_FEATURES;
8698 dev->features |= NETIF_F_SOFT_FEATURES;
8700 if (dev->netdev_ops->ndo_udp_tunnel_add) {
8701 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
8702 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
8705 dev->wanted_features = dev->features & dev->hw_features;
8707 if (!(dev->flags & IFF_LOOPBACK))
8708 dev->hw_features |= NETIF_F_NOCACHE_COPY;
8710 /* If IPv4 TCP segmentation offload is supported we should also
8711 * allow the device to enable segmenting the frame with the option
8712 * of ignoring a static IP ID value. This doesn't enable the
8713 * feature itself but allows the user to enable it later.
8715 if (dev->hw_features & NETIF_F_TSO)
8716 dev->hw_features |= NETIF_F_TSO_MANGLEID;
8717 if (dev->vlan_features & NETIF_F_TSO)
8718 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
8719 if (dev->mpls_features & NETIF_F_TSO)
8720 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
8721 if (dev->hw_enc_features & NETIF_F_TSO)
8722 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
8724 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
8726 dev->vlan_features |= NETIF_F_HIGHDMA;
8728 /* Make NETIF_F_SG inheritable to tunnel devices.
8730 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
8732 /* Make NETIF_F_SG inheritable to MPLS.
8734 dev->mpls_features |= NETIF_F_SG;
8736 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
8737 ret = notifier_to_errno(ret);
8741 ret = netdev_register_kobject(dev);
8743 dev->reg_state = NETREG_UNREGISTERED;
8746 dev->reg_state = NETREG_REGISTERED;
8748 __netdev_update_features(dev);
8751 * Default initial state at registry is that the
8752 * device is present.
8755 set_bit(__LINK_STATE_PRESENT, &dev->state);
8757 linkwatch_init_dev(dev);
8759 dev_init_scheduler(dev);
8761 list_netdevice(dev);
8762 add_device_randomness(dev->dev_addr, dev->addr_len);
8764 /* If the device has permanent device address, driver should
8765 * set dev_addr and also addr_assign_type should be set to
8766 * NET_ADDR_PERM (default value).
8768 if (dev->addr_assign_type == NET_ADDR_PERM)
8769 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
8771 /* Notify protocols, that a new device appeared. */
8772 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
8773 ret = notifier_to_errno(ret);
8775 rollback_registered(dev);
8778 dev->reg_state = NETREG_UNREGISTERED;
8779 /* We should put the kobject that hold in
8780 * netdev_unregister_kobject(), otherwise
8781 * the net device cannot be freed when
8782 * driver calls free_netdev(), because the
8783 * kobject is being hold.
8785 kobject_put(&dev->dev.kobj);
8788 * Prevent userspace races by waiting until the network
8789 * device is fully setup before sending notifications.
8791 if (!dev->rtnl_link_ops ||
8792 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
8793 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
8799 if (dev->netdev_ops->ndo_uninit)
8800 dev->netdev_ops->ndo_uninit(dev);
8801 if (dev->priv_destructor)
8802 dev->priv_destructor(dev);
8805 EXPORT_SYMBOL(register_netdevice);
8808 * init_dummy_netdev - init a dummy network device for NAPI
8809 * @dev: device to init
8811 * This takes a network device structure and initialize the minimum
8812 * amount of fields so it can be used to schedule NAPI polls without
8813 * registering a full blown interface. This is to be used by drivers
8814 * that need to tie several hardware interfaces to a single NAPI
8815 * poll scheduler due to HW limitations.
8817 int init_dummy_netdev(struct net_device *dev)
8819 /* Clear everything. Note we don't initialize spinlocks
8820 * are they aren't supposed to be taken by any of the
8821 * NAPI code and this dummy netdev is supposed to be
8822 * only ever used for NAPI polls
8824 memset(dev, 0, sizeof(struct net_device));
8826 /* make sure we BUG if trying to hit standard
8827 * register/unregister code path
8829 dev->reg_state = NETREG_DUMMY;
8831 /* NAPI wants this */
8832 INIT_LIST_HEAD(&dev->napi_list);
8834 /* a dummy interface is started by default */
8835 set_bit(__LINK_STATE_PRESENT, &dev->state);
8836 set_bit(__LINK_STATE_START, &dev->state);
8838 /* napi_busy_loop stats accounting wants this */
8839 dev_net_set(dev, &init_net);
8841 /* Note : We dont allocate pcpu_refcnt for dummy devices,
8842 * because users of this 'device' dont need to change
8848 EXPORT_SYMBOL_GPL(init_dummy_netdev);
8852 * register_netdev - register a network device
8853 * @dev: device to register
8855 * Take a completed network device structure and add it to the kernel
8856 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
8857 * chain. 0 is returned on success. A negative errno code is returned
8858 * on a failure to set up the device, or if the name is a duplicate.
8860 * This is a wrapper around register_netdevice that takes the rtnl semaphore
8861 * and expands the device name if you passed a format string to
8864 int register_netdev(struct net_device *dev)
8868 if (rtnl_lock_killable())
8870 err = register_netdevice(dev);
8874 EXPORT_SYMBOL(register_netdev);
8876 int netdev_refcnt_read(const struct net_device *dev)
8880 for_each_possible_cpu(i)
8881 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
8884 EXPORT_SYMBOL(netdev_refcnt_read);
8887 * netdev_wait_allrefs - wait until all references are gone.
8888 * @dev: target net_device
8890 * This is called when unregistering network devices.
8892 * Any protocol or device that holds a reference should register
8893 * for netdevice notification, and cleanup and put back the
8894 * reference if they receive an UNREGISTER event.
8895 * We can get stuck here if buggy protocols don't correctly
8898 static void netdev_wait_allrefs(struct net_device *dev)
8900 unsigned long rebroadcast_time, warning_time;
8903 linkwatch_forget_dev(dev);
8905 rebroadcast_time = warning_time = jiffies;
8906 refcnt = netdev_refcnt_read(dev);
8908 while (refcnt != 0) {
8909 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
8912 /* Rebroadcast unregister notification */
8913 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
8919 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
8921 /* We must not have linkwatch events
8922 * pending on unregister. If this
8923 * happens, we simply run the queue
8924 * unscheduled, resulting in a noop
8927 linkwatch_run_queue();
8932 rebroadcast_time = jiffies;
8937 refcnt = netdev_refcnt_read(dev);
8939 if (refcnt && time_after(jiffies, warning_time + 10 * HZ)) {
8940 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
8942 warning_time = jiffies;
8951 * register_netdevice(x1);
8952 * register_netdevice(x2);
8954 * unregister_netdevice(y1);
8955 * unregister_netdevice(y2);
8961 * We are invoked by rtnl_unlock().
8962 * This allows us to deal with problems:
8963 * 1) We can delete sysfs objects which invoke hotplug
8964 * without deadlocking with linkwatch via keventd.
8965 * 2) Since we run with the RTNL semaphore not held, we can sleep
8966 * safely in order to wait for the netdev refcnt to drop to zero.
8968 * We must not return until all unregister events added during
8969 * the interval the lock was held have been completed.
8971 void netdev_run_todo(void)
8973 struct list_head list;
8975 /* Snapshot list, allow later requests */
8976 list_replace_init(&net_todo_list, &list);
8981 /* Wait for rcu callbacks to finish before next phase */
8982 if (!list_empty(&list))
8985 while (!list_empty(&list)) {
8986 struct net_device *dev
8987 = list_first_entry(&list, struct net_device, todo_list);
8988 list_del(&dev->todo_list);
8990 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
8991 pr_err("network todo '%s' but state %d\n",
8992 dev->name, dev->reg_state);
8997 dev->reg_state = NETREG_UNREGISTERED;
8999 netdev_wait_allrefs(dev);
9002 BUG_ON(netdev_refcnt_read(dev));
9003 BUG_ON(!list_empty(&dev->ptype_all));
9004 BUG_ON(!list_empty(&dev->ptype_specific));
9005 WARN_ON(rcu_access_pointer(dev->ip_ptr));
9006 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
9007 #if IS_ENABLED(CONFIG_DECNET)
9008 WARN_ON(dev->dn_ptr);
9010 if (dev->priv_destructor)
9011 dev->priv_destructor(dev);
9012 if (dev->needs_free_netdev)
9015 /* Report a network device has been unregistered */
9017 dev_net(dev)->dev_unreg_count--;
9019 wake_up(&netdev_unregistering_wq);
9021 /* Free network device */
9022 kobject_put(&dev->dev.kobj);
9026 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
9027 * all the same fields in the same order as net_device_stats, with only
9028 * the type differing, but rtnl_link_stats64 may have additional fields
9029 * at the end for newer counters.
9031 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
9032 const struct net_device_stats *netdev_stats)
9034 #if BITS_PER_LONG == 64
9035 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
9036 memcpy(stats64, netdev_stats, sizeof(*netdev_stats));
9037 /* zero out counters that only exist in rtnl_link_stats64 */
9038 memset((char *)stats64 + sizeof(*netdev_stats), 0,
9039 sizeof(*stats64) - sizeof(*netdev_stats));
9041 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
9042 const unsigned long *src = (const unsigned long *)netdev_stats;
9043 u64 *dst = (u64 *)stats64;
9045 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
9046 for (i = 0; i < n; i++)
9048 /* zero out counters that only exist in rtnl_link_stats64 */
9049 memset((char *)stats64 + n * sizeof(u64), 0,
9050 sizeof(*stats64) - n * sizeof(u64));
9053 EXPORT_SYMBOL(netdev_stats_to_stats64);
9056 * dev_get_stats - get network device statistics
9057 * @dev: device to get statistics from
9058 * @storage: place to store stats
9060 * Get network statistics from device. Return @storage.
9061 * The device driver may provide its own method by setting
9062 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
9063 * otherwise the internal statistics structure is used.
9065 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
9066 struct rtnl_link_stats64 *storage)
9068 const struct net_device_ops *ops = dev->netdev_ops;
9070 if (ops->ndo_get_stats64) {
9071 memset(storage, 0, sizeof(*storage));
9072 ops->ndo_get_stats64(dev, storage);
9073 } else if (ops->ndo_get_stats) {
9074 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
9076 netdev_stats_to_stats64(storage, &dev->stats);
9078 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped);
9079 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped);
9080 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler);
9083 EXPORT_SYMBOL(dev_get_stats);
9085 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
9087 struct netdev_queue *queue = dev_ingress_queue(dev);
9089 #ifdef CONFIG_NET_CLS_ACT
9092 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
9095 netdev_init_one_queue(dev, queue, NULL);
9096 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
9097 queue->qdisc_sleeping = &noop_qdisc;
9098 rcu_assign_pointer(dev->ingress_queue, queue);
9103 static const struct ethtool_ops default_ethtool_ops;
9105 void netdev_set_default_ethtool_ops(struct net_device *dev,
9106 const struct ethtool_ops *ops)
9108 if (dev->ethtool_ops == &default_ethtool_ops)
9109 dev->ethtool_ops = ops;
9111 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
9113 void netdev_freemem(struct net_device *dev)
9115 char *addr = (char *)dev - dev->padded;
9121 * alloc_netdev_mqs - allocate network device
9122 * @sizeof_priv: size of private data to allocate space for
9123 * @name: device name format string
9124 * @name_assign_type: origin of device name
9125 * @setup: callback to initialize device
9126 * @txqs: the number of TX subqueues to allocate
9127 * @rxqs: the number of RX subqueues to allocate
9129 * Allocates a struct net_device with private data area for driver use
9130 * and performs basic initialization. Also allocates subqueue structs
9131 * for each queue on the device.
9133 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
9134 unsigned char name_assign_type,
9135 void (*setup)(struct net_device *),
9136 unsigned int txqs, unsigned int rxqs)
9138 struct net_device *dev;
9139 unsigned int alloc_size;
9140 struct net_device *p;
9142 BUG_ON(strlen(name) >= sizeof(dev->name));
9145 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
9150 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
9154 alloc_size = sizeof(struct net_device);
9156 /* ensure 32-byte alignment of private area */
9157 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
9158 alloc_size += sizeof_priv;
9160 /* ensure 32-byte alignment of whole construct */
9161 alloc_size += NETDEV_ALIGN - 1;
9163 p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
9167 dev = PTR_ALIGN(p, NETDEV_ALIGN);
9168 dev->padded = (char *)dev - (char *)p;
9170 dev->pcpu_refcnt = alloc_percpu(int);
9171 if (!dev->pcpu_refcnt)
9174 if (dev_addr_init(dev))
9180 dev_net_set(dev, &init_net);
9182 dev->gso_max_size = GSO_MAX_SIZE;
9183 dev->gso_max_segs = GSO_MAX_SEGS;
9184 dev->upper_level = 1;
9185 dev->lower_level = 1;
9187 INIT_LIST_HEAD(&dev->napi_list);
9188 INIT_LIST_HEAD(&dev->unreg_list);
9189 INIT_LIST_HEAD(&dev->close_list);
9190 INIT_LIST_HEAD(&dev->link_watch_list);
9191 INIT_LIST_HEAD(&dev->adj_list.upper);
9192 INIT_LIST_HEAD(&dev->adj_list.lower);
9193 INIT_LIST_HEAD(&dev->ptype_all);
9194 INIT_LIST_HEAD(&dev->ptype_specific);
9195 #ifdef CONFIG_NET_SCHED
9196 hash_init(dev->qdisc_hash);
9198 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
9201 if (!dev->tx_queue_len) {
9202 dev->priv_flags |= IFF_NO_QUEUE;
9203 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
9206 dev->num_tx_queues = txqs;
9207 dev->real_num_tx_queues = txqs;
9208 if (netif_alloc_netdev_queues(dev))
9211 dev->num_rx_queues = rxqs;
9212 dev->real_num_rx_queues = rxqs;
9213 if (netif_alloc_rx_queues(dev))
9216 strcpy(dev->name, name);
9217 dev->name_assign_type = name_assign_type;
9218 dev->group = INIT_NETDEV_GROUP;
9219 if (!dev->ethtool_ops)
9220 dev->ethtool_ops = &default_ethtool_ops;
9222 nf_hook_ingress_init(dev);
9231 free_percpu(dev->pcpu_refcnt);
9233 netdev_freemem(dev);
9236 EXPORT_SYMBOL(alloc_netdev_mqs);
9239 * free_netdev - free network device
9242 * This function does the last stage of destroying an allocated device
9243 * interface. The reference to the device object is released. If this
9244 * is the last reference then it will be freed.Must be called in process
9247 void free_netdev(struct net_device *dev)
9249 struct napi_struct *p, *n;
9252 netif_free_tx_queues(dev);
9253 netif_free_rx_queues(dev);
9255 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
9257 /* Flush device addresses */
9258 dev_addr_flush(dev);
9260 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
9263 free_percpu(dev->pcpu_refcnt);
9264 dev->pcpu_refcnt = NULL;
9266 /* Compatibility with error handling in drivers */
9267 if (dev->reg_state == NETREG_UNINITIALIZED) {
9268 netdev_freemem(dev);
9272 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
9273 dev->reg_state = NETREG_RELEASED;
9275 /* will free via device release */
9276 put_device(&dev->dev);
9278 EXPORT_SYMBOL(free_netdev);
9281 * synchronize_net - Synchronize with packet receive processing
9283 * Wait for packets currently being received to be done.
9284 * Does not block later packets from starting.
9286 void synchronize_net(void)
9289 if (rtnl_is_locked())
9290 synchronize_rcu_expedited();
9294 EXPORT_SYMBOL(synchronize_net);
9297 * unregister_netdevice_queue - remove device from the kernel
9301 * This function shuts down a device interface and removes it
9302 * from the kernel tables.
9303 * If head not NULL, device is queued to be unregistered later.
9305 * Callers must hold the rtnl semaphore. You may want
9306 * unregister_netdev() instead of this.
9309 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
9314 list_move_tail(&dev->unreg_list, head);
9316 rollback_registered(dev);
9317 /* Finish processing unregister after unlock */
9321 EXPORT_SYMBOL(unregister_netdevice_queue);
9324 * unregister_netdevice_many - unregister many devices
9325 * @head: list of devices
9327 * Note: As most callers use a stack allocated list_head,
9328 * we force a list_del() to make sure stack wont be corrupted later.
9330 void unregister_netdevice_many(struct list_head *head)
9332 struct net_device *dev;
9334 if (!list_empty(head)) {
9335 rollback_registered_many(head);
9336 list_for_each_entry(dev, head, unreg_list)
9341 EXPORT_SYMBOL(unregister_netdevice_many);
9344 * unregister_netdev - remove device from the kernel
9347 * This function shuts down a device interface and removes it
9348 * from the kernel tables.
9350 * This is just a wrapper for unregister_netdevice that takes
9351 * the rtnl semaphore. In general you want to use this and not
9352 * unregister_netdevice.
9354 void unregister_netdev(struct net_device *dev)
9357 unregister_netdevice(dev);
9360 EXPORT_SYMBOL(unregister_netdev);
9363 * dev_change_net_namespace - move device to different nethost namespace
9365 * @net: network namespace
9366 * @pat: If not NULL name pattern to try if the current device name
9367 * is already taken in the destination network namespace.
9369 * This function shuts down a device interface and moves it
9370 * to a new network namespace. On success 0 is returned, on
9371 * a failure a netagive errno code is returned.
9373 * Callers must hold the rtnl semaphore.
9376 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
9378 int err, new_nsid, new_ifindex;
9382 /* Don't allow namespace local devices to be moved. */
9384 if (dev->features & NETIF_F_NETNS_LOCAL)
9387 /* Ensure the device has been registrered */
9388 if (dev->reg_state != NETREG_REGISTERED)
9391 /* Get out if there is nothing todo */
9393 if (net_eq(dev_net(dev), net))
9396 /* Pick the destination device name, and ensure
9397 * we can use it in the destination network namespace.
9400 if (__dev_get_by_name(net, dev->name)) {
9401 /* We get here if we can't use the current device name */
9404 err = dev_get_valid_name(net, dev, pat);
9410 * And now a mini version of register_netdevice unregister_netdevice.
9413 /* If device is running close it first. */
9416 /* And unlink it from device chain */
9417 unlist_netdevice(dev);
9421 /* Shutdown queueing discipline. */
9424 /* Notify protocols, that we are about to destroy
9425 * this device. They should clean all the things.
9427 * Note that dev->reg_state stays at NETREG_REGISTERED.
9428 * This is wanted because this way 8021q and macvlan know
9429 * the device is just moving and can keep their slaves up.
9431 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
9434 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
9435 /* If there is an ifindex conflict assign a new one */
9436 if (__dev_get_by_index(net, dev->ifindex))
9437 new_ifindex = dev_new_index(net);
9439 new_ifindex = dev->ifindex;
9441 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
9445 * Flush the unicast and multicast chains
9450 /* Send a netdev-removed uevent to the old namespace */
9451 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
9452 netdev_adjacent_del_links(dev);
9454 /* Actually switch the network namespace */
9455 dev_net_set(dev, net);
9456 dev->ifindex = new_ifindex;
9458 /* Send a netdev-add uevent to the new namespace */
9459 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
9460 netdev_adjacent_add_links(dev);
9462 /* Fixup kobjects */
9463 err = device_rename(&dev->dev, dev->name);
9466 /* Add the device back in the hashes */
9467 list_netdevice(dev);
9469 /* Notify protocols, that a new device appeared. */
9470 call_netdevice_notifiers(NETDEV_REGISTER, dev);
9473 * Prevent userspace races by waiting until the network
9474 * device is fully setup before sending notifications.
9476 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
9483 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
9485 static int dev_cpu_dead(unsigned int oldcpu)
9487 struct sk_buff **list_skb;
9488 struct sk_buff *skb;
9490 struct softnet_data *sd, *oldsd, *remsd = NULL;
9492 local_irq_disable();
9493 cpu = smp_processor_id();
9494 sd = &per_cpu(softnet_data, cpu);
9495 oldsd = &per_cpu(softnet_data, oldcpu);
9497 /* Find end of our completion_queue. */
9498 list_skb = &sd->completion_queue;
9500 list_skb = &(*list_skb)->next;
9501 /* Append completion queue from offline CPU. */
9502 *list_skb = oldsd->completion_queue;
9503 oldsd->completion_queue = NULL;
9505 /* Append output queue from offline CPU. */
9506 if (oldsd->output_queue) {
9507 *sd->output_queue_tailp = oldsd->output_queue;
9508 sd->output_queue_tailp = oldsd->output_queue_tailp;
9509 oldsd->output_queue = NULL;
9510 oldsd->output_queue_tailp = &oldsd->output_queue;
9512 /* Append NAPI poll list from offline CPU, with one exception :
9513 * process_backlog() must be called by cpu owning percpu backlog.
9514 * We properly handle process_queue & input_pkt_queue later.
9516 while (!list_empty(&oldsd->poll_list)) {
9517 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
9521 list_del_init(&napi->poll_list);
9522 if (napi->poll == process_backlog)
9525 ____napi_schedule(sd, napi);
9528 raise_softirq_irqoff(NET_TX_SOFTIRQ);
9532 remsd = oldsd->rps_ipi_list;
9533 oldsd->rps_ipi_list = NULL;
9535 /* send out pending IPI's on offline CPU */
9536 net_rps_send_ipi(remsd);
9538 /* Process offline CPU's input_pkt_queue */
9539 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
9541 input_queue_head_incr(oldsd);
9543 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
9545 input_queue_head_incr(oldsd);
9552 * netdev_increment_features - increment feature set by one
9553 * @all: current feature set
9554 * @one: new feature set
9555 * @mask: mask feature set
9557 * Computes a new feature set after adding a device with feature set
9558 * @one to the master device with current feature set @all. Will not
9559 * enable anything that is off in @mask. Returns the new feature set.
9561 netdev_features_t netdev_increment_features(netdev_features_t all,
9562 netdev_features_t one, netdev_features_t mask)
9564 if (mask & NETIF_F_HW_CSUM)
9565 mask |= NETIF_F_CSUM_MASK;
9566 mask |= NETIF_F_VLAN_CHALLENGED;
9568 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
9569 all &= one | ~NETIF_F_ALL_FOR_ALL;
9571 /* If one device supports hw checksumming, set for all. */
9572 if (all & NETIF_F_HW_CSUM)
9573 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
9577 EXPORT_SYMBOL(netdev_increment_features);
9579 static struct hlist_head * __net_init netdev_create_hash(void)
9582 struct hlist_head *hash;
9584 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
9586 for (i = 0; i < NETDEV_HASHENTRIES; i++)
9587 INIT_HLIST_HEAD(&hash[i]);
9592 /* Initialize per network namespace state */
9593 static int __net_init netdev_init(struct net *net)
9595 BUILD_BUG_ON(GRO_HASH_BUCKETS >
9596 8 * FIELD_SIZEOF(struct napi_struct, gro_bitmask));
9598 if (net != &init_net)
9599 INIT_LIST_HEAD(&net->dev_base_head);
9601 net->dev_name_head = netdev_create_hash();
9602 if (net->dev_name_head == NULL)
9605 net->dev_index_head = netdev_create_hash();
9606 if (net->dev_index_head == NULL)
9612 kfree(net->dev_name_head);
9618 * netdev_drivername - network driver for the device
9619 * @dev: network device
9621 * Determine network driver for device.
9623 const char *netdev_drivername(const struct net_device *dev)
9625 const struct device_driver *driver;
9626 const struct device *parent;
9627 const char *empty = "";
9629 parent = dev->dev.parent;
9633 driver = parent->driver;
9634 if (driver && driver->name)
9635 return driver->name;
9639 static void __netdev_printk(const char *level, const struct net_device *dev,
9640 struct va_format *vaf)
9642 if (dev && dev->dev.parent) {
9643 dev_printk_emit(level[1] - '0',
9646 dev_driver_string(dev->dev.parent),
9647 dev_name(dev->dev.parent),
9648 netdev_name(dev), netdev_reg_state(dev),
9651 printk("%s%s%s: %pV",
9652 level, netdev_name(dev), netdev_reg_state(dev), vaf);
9654 printk("%s(NULL net_device): %pV", level, vaf);
9658 void netdev_printk(const char *level, const struct net_device *dev,
9659 const char *format, ...)
9661 struct va_format vaf;
9664 va_start(args, format);
9669 __netdev_printk(level, dev, &vaf);
9673 EXPORT_SYMBOL(netdev_printk);
9675 #define define_netdev_printk_level(func, level) \
9676 void func(const struct net_device *dev, const char *fmt, ...) \
9678 struct va_format vaf; \
9681 va_start(args, fmt); \
9686 __netdev_printk(level, dev, &vaf); \
9690 EXPORT_SYMBOL(func);
9692 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
9693 define_netdev_printk_level(netdev_alert, KERN_ALERT);
9694 define_netdev_printk_level(netdev_crit, KERN_CRIT);
9695 define_netdev_printk_level(netdev_err, KERN_ERR);
9696 define_netdev_printk_level(netdev_warn, KERN_WARNING);
9697 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
9698 define_netdev_printk_level(netdev_info, KERN_INFO);
9700 static void __net_exit netdev_exit(struct net *net)
9702 kfree(net->dev_name_head);
9703 kfree(net->dev_index_head);
9704 if (net != &init_net)
9705 WARN_ON_ONCE(!list_empty(&net->dev_base_head));
9708 static struct pernet_operations __net_initdata netdev_net_ops = {
9709 .init = netdev_init,
9710 .exit = netdev_exit,
9713 static void __net_exit default_device_exit(struct net *net)
9715 struct net_device *dev, *aux;
9717 * Push all migratable network devices back to the
9718 * initial network namespace
9721 for_each_netdev_safe(net, dev, aux) {
9723 char fb_name[IFNAMSIZ];
9725 /* Ignore unmoveable devices (i.e. loopback) */
9726 if (dev->features & NETIF_F_NETNS_LOCAL)
9729 /* Leave virtual devices for the generic cleanup */
9730 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund)
9733 /* Push remaining network devices to init_net */
9734 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
9735 if (__dev_get_by_name(&init_net, fb_name))
9736 snprintf(fb_name, IFNAMSIZ, "dev%%d");
9737 err = dev_change_net_namespace(dev, &init_net, fb_name);
9739 pr_emerg("%s: failed to move %s to init_net: %d\n",
9740 __func__, dev->name, err);
9747 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
9749 /* Return with the rtnl_lock held when there are no network
9750 * devices unregistering in any network namespace in net_list.
9754 DEFINE_WAIT_FUNC(wait, woken_wake_function);
9756 add_wait_queue(&netdev_unregistering_wq, &wait);
9758 unregistering = false;
9760 list_for_each_entry(net, net_list, exit_list) {
9761 if (net->dev_unreg_count > 0) {
9762 unregistering = true;
9770 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
9772 remove_wait_queue(&netdev_unregistering_wq, &wait);
9775 static void __net_exit default_device_exit_batch(struct list_head *net_list)
9777 /* At exit all network devices most be removed from a network
9778 * namespace. Do this in the reverse order of registration.
9779 * Do this across as many network namespaces as possible to
9780 * improve batching efficiency.
9782 struct net_device *dev;
9784 LIST_HEAD(dev_kill_list);
9786 /* To prevent network device cleanup code from dereferencing
9787 * loopback devices or network devices that have been freed
9788 * wait here for all pending unregistrations to complete,
9789 * before unregistring the loopback device and allowing the
9790 * network namespace be freed.
9792 * The netdev todo list containing all network devices
9793 * unregistrations that happen in default_device_exit_batch
9794 * will run in the rtnl_unlock() at the end of
9795 * default_device_exit_batch.
9797 rtnl_lock_unregistering(net_list);
9798 list_for_each_entry(net, net_list, exit_list) {
9799 for_each_netdev_reverse(net, dev) {
9800 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
9801 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
9803 unregister_netdevice_queue(dev, &dev_kill_list);
9806 unregister_netdevice_many(&dev_kill_list);
9810 static struct pernet_operations __net_initdata default_device_ops = {
9811 .exit = default_device_exit,
9812 .exit_batch = default_device_exit_batch,
9816 * Initialize the DEV module. At boot time this walks the device list and
9817 * unhooks any devices that fail to initialise (normally hardware not
9818 * present) and leaves us with a valid list of present and active devices.
9823 * This is called single threaded during boot, so no need
9824 * to take the rtnl semaphore.
9826 static int __init net_dev_init(void)
9828 int i, rc = -ENOMEM;
9830 BUG_ON(!dev_boot_phase);
9832 if (dev_proc_init())
9835 if (netdev_kobject_init())
9838 INIT_LIST_HEAD(&ptype_all);
9839 for (i = 0; i < PTYPE_HASH_SIZE; i++)
9840 INIT_LIST_HEAD(&ptype_base[i]);
9842 INIT_LIST_HEAD(&offload_base);
9844 if (register_pernet_subsys(&netdev_net_ops))
9848 * Initialise the packet receive queues.
9851 for_each_possible_cpu(i) {
9852 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
9853 struct softnet_data *sd = &per_cpu(softnet_data, i);
9855 INIT_WORK(flush, flush_backlog);
9857 skb_queue_head_init(&sd->input_pkt_queue);
9858 skb_queue_head_init(&sd->process_queue);
9859 #ifdef CONFIG_XFRM_OFFLOAD
9860 skb_queue_head_init(&sd->xfrm_backlog);
9862 INIT_LIST_HEAD(&sd->poll_list);
9863 sd->output_queue_tailp = &sd->output_queue;
9865 sd->csd.func = rps_trigger_softirq;
9870 init_gro_hash(&sd->backlog);
9871 sd->backlog.poll = process_backlog;
9872 sd->backlog.weight = weight_p;
9877 /* The loopback device is special if any other network devices
9878 * is present in a network namespace the loopback device must
9879 * be present. Since we now dynamically allocate and free the
9880 * loopback device ensure this invariant is maintained by
9881 * keeping the loopback device as the first device on the
9882 * list of network devices. Ensuring the loopback devices
9883 * is the first device that appears and the last network device
9886 if (register_pernet_device(&loopback_net_ops))
9889 if (register_pernet_device(&default_device_ops))
9892 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
9893 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
9895 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
9896 NULL, dev_cpu_dead);
9903 subsys_initcall(net_dev_init);