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/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/skbuff.h>
97 #include <linux/bpf.h>
98 #include <linux/bpf_trace.h>
99 #include <net/net_namespace.h>
100 #include <net/sock.h>
101 #include <net/busy_poll.h>
102 #include <linux/rtnetlink.h>
103 #include <linux/stat.h>
105 #include <net/dst_metadata.h>
106 #include <net/pkt_sched.h>
107 #include <net/pkt_cls.h>
108 #include <net/checksum.h>
109 #include <net/xfrm.h>
110 #include <linux/highmem.h>
111 #include <linux/init.h>
112 #include <linux/module.h>
113 #include <linux/netpoll.h>
114 #include <linux/rcupdate.h>
115 #include <linux/delay.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
126 #include <net/mpls.h>
127 #include <linux/ipv6.h>
128 #include <linux/in.h>
129 #include <linux/jhash.h>
130 #include <linux/random.h>
131 #include <trace/events/napi.h>
132 #include <trace/events/net.h>
133 #include <trace/events/skb.h>
134 #include <linux/pci.h>
135 #include <linux/inetdevice.h>
136 #include <linux/cpu_rmap.h>
137 #include <linux/static_key.h>
138 #include <linux/hashtable.h>
139 #include <linux/vmalloc.h>
140 #include <linux/if_macvlan.h>
141 #include <linux/errqueue.h>
142 #include <linux/hrtimer.h>
143 #include <linux/netfilter_ingress.h>
144 #include <linux/crash_dump.h>
145 #include <linux/sctp.h>
146 #include <net/udp_tunnel.h>
147 #include <linux/net_namespace.h>
148 #include <linux/indirect_call_wrapper.h>
150 #include "net-sysfs.h"
152 #define MAX_GRO_SKBS 8
154 /* This should be increased if a protocol with a bigger head is added. */
155 #define GRO_MAX_HEAD (MAX_HEADER + 128)
157 static DEFINE_SPINLOCK(ptype_lock);
158 static DEFINE_SPINLOCK(offload_lock);
159 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
160 struct list_head ptype_all __read_mostly; /* Taps */
161 static struct list_head offload_base __read_mostly;
163 static int netif_rx_internal(struct sk_buff *skb);
164 static int call_netdevice_notifiers_info(unsigned long val,
165 struct netdev_notifier_info *info);
166 static int call_netdevice_notifiers_extack(unsigned long val,
167 struct net_device *dev,
168 struct netlink_ext_ack *extack);
169 static struct napi_struct *napi_by_id(unsigned int napi_id);
172 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
175 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
177 * Writers must hold the rtnl semaphore while they loop through the
178 * dev_base_head list, and hold dev_base_lock for writing when they do the
179 * actual updates. This allows pure readers to access the list even
180 * while a writer is preparing to update it.
182 * To put it another way, dev_base_lock is held for writing only to
183 * protect against pure readers; the rtnl semaphore provides the
184 * protection against other writers.
186 * See, for example usages, register_netdevice() and
187 * unregister_netdevice(), which must be called with the rtnl
190 DEFINE_RWLOCK(dev_base_lock);
191 EXPORT_SYMBOL(dev_base_lock);
193 static DEFINE_MUTEX(ifalias_mutex);
195 /* protects napi_hash addition/deletion and napi_gen_id */
196 static DEFINE_SPINLOCK(napi_hash_lock);
198 static unsigned int napi_gen_id = NR_CPUS;
199 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
201 static seqcount_t devnet_rename_seq;
203 static inline void dev_base_seq_inc(struct net *net)
205 while (++net->dev_base_seq == 0)
209 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
211 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
213 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
216 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
218 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
221 static inline void rps_lock(struct softnet_data *sd)
224 spin_lock(&sd->input_pkt_queue.lock);
228 static inline void rps_unlock(struct softnet_data *sd)
231 spin_unlock(&sd->input_pkt_queue.lock);
235 /* Device list insertion */
236 static void list_netdevice(struct net_device *dev)
238 struct net *net = dev_net(dev);
242 write_lock_bh(&dev_base_lock);
243 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
244 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
245 hlist_add_head_rcu(&dev->index_hlist,
246 dev_index_hash(net, dev->ifindex));
247 write_unlock_bh(&dev_base_lock);
249 dev_base_seq_inc(net);
252 /* Device list removal
253 * caller must respect a RCU grace period before freeing/reusing dev
255 static void unlist_netdevice(struct net_device *dev)
259 /* Unlink dev from the device chain */
260 write_lock_bh(&dev_base_lock);
261 list_del_rcu(&dev->dev_list);
262 hlist_del_rcu(&dev->name_hlist);
263 hlist_del_rcu(&dev->index_hlist);
264 write_unlock_bh(&dev_base_lock);
266 dev_base_seq_inc(dev_net(dev));
273 static RAW_NOTIFIER_HEAD(netdev_chain);
276 * Device drivers call our routines to queue packets here. We empty the
277 * queue in the local softnet handler.
280 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
281 EXPORT_PER_CPU_SYMBOL(softnet_data);
283 #ifdef CONFIG_LOCKDEP
285 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
286 * according to dev->type
288 static const unsigned short netdev_lock_type[] = {
289 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
290 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
291 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
292 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
293 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
294 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
295 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
296 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
297 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
298 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
299 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
300 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
301 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
302 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
303 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
305 static const char *const netdev_lock_name[] = {
306 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
307 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
308 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
309 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
310 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
311 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
312 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
313 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
314 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
315 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
316 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
317 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
318 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
319 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
320 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
322 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
323 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
325 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
329 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
330 if (netdev_lock_type[i] == dev_type)
332 /* the last key is used by default */
333 return ARRAY_SIZE(netdev_lock_type) - 1;
336 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
337 unsigned short dev_type)
341 i = netdev_lock_pos(dev_type);
342 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
343 netdev_lock_name[i]);
346 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
350 i = netdev_lock_pos(dev->type);
351 lockdep_set_class_and_name(&dev->addr_list_lock,
352 &netdev_addr_lock_key[i],
353 netdev_lock_name[i]);
356 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
357 unsigned short dev_type)
360 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
365 /*******************************************************************************
367 * Protocol management and registration routines
369 *******************************************************************************/
373 * Add a protocol ID to the list. Now that the input handler is
374 * smarter we can dispense with all the messy stuff that used to be
377 * BEWARE!!! Protocol handlers, mangling input packets,
378 * MUST BE last in hash buckets and checking protocol handlers
379 * MUST start from promiscuous ptype_all chain in net_bh.
380 * It is true now, do not change it.
381 * Explanation follows: if protocol handler, mangling packet, will
382 * be the first on list, it is not able to sense, that packet
383 * is cloned and should be copied-on-write, so that it will
384 * change it and subsequent readers will get broken packet.
388 static inline struct list_head *ptype_head(const struct packet_type *pt)
390 if (pt->type == htons(ETH_P_ALL))
391 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
393 return pt->dev ? &pt->dev->ptype_specific :
394 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
398 * dev_add_pack - add packet handler
399 * @pt: packet type declaration
401 * Add a protocol handler to the networking stack. The passed &packet_type
402 * is linked into kernel lists and may not be freed until it has been
403 * removed from the kernel lists.
405 * This call does not sleep therefore it can not
406 * guarantee all CPU's that are in middle of receiving packets
407 * will see the new packet type (until the next received packet).
410 void dev_add_pack(struct packet_type *pt)
412 struct list_head *head = ptype_head(pt);
414 spin_lock(&ptype_lock);
415 list_add_rcu(&pt->list, head);
416 spin_unlock(&ptype_lock);
418 EXPORT_SYMBOL(dev_add_pack);
421 * __dev_remove_pack - remove packet handler
422 * @pt: packet type declaration
424 * Remove a protocol handler that was previously added to the kernel
425 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
426 * from the kernel lists and can be freed or reused once this function
429 * The packet type might still be in use by receivers
430 * and must not be freed until after all the CPU's have gone
431 * through a quiescent state.
433 void __dev_remove_pack(struct packet_type *pt)
435 struct list_head *head = ptype_head(pt);
436 struct packet_type *pt1;
438 spin_lock(&ptype_lock);
440 list_for_each_entry(pt1, head, list) {
442 list_del_rcu(&pt->list);
447 pr_warn("dev_remove_pack: %p not found\n", pt);
449 spin_unlock(&ptype_lock);
451 EXPORT_SYMBOL(__dev_remove_pack);
454 * dev_remove_pack - remove packet handler
455 * @pt: packet type declaration
457 * Remove a protocol handler that was previously added to the kernel
458 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
459 * from the kernel lists and can be freed or reused once this function
462 * This call sleeps to guarantee that no CPU is looking at the packet
465 void dev_remove_pack(struct packet_type *pt)
467 __dev_remove_pack(pt);
471 EXPORT_SYMBOL(dev_remove_pack);
475 * dev_add_offload - register offload handlers
476 * @po: protocol offload declaration
478 * Add protocol offload handlers to the networking stack. The passed
479 * &proto_offload is linked into kernel lists and may not be freed until
480 * it has been removed from the kernel lists.
482 * This call does not sleep therefore it can not
483 * guarantee all CPU's that are in middle of receiving packets
484 * will see the new offload handlers (until the next received packet).
486 void dev_add_offload(struct packet_offload *po)
488 struct packet_offload *elem;
490 spin_lock(&offload_lock);
491 list_for_each_entry(elem, &offload_base, list) {
492 if (po->priority < elem->priority)
495 list_add_rcu(&po->list, elem->list.prev);
496 spin_unlock(&offload_lock);
498 EXPORT_SYMBOL(dev_add_offload);
501 * __dev_remove_offload - remove offload handler
502 * @po: packet offload declaration
504 * Remove a protocol offload handler that was previously added to the
505 * kernel offload handlers by dev_add_offload(). The passed &offload_type
506 * is removed from the kernel lists and can be freed or reused once this
509 * The packet type might still be in use by receivers
510 * and must not be freed until after all the CPU's have gone
511 * through a quiescent state.
513 static void __dev_remove_offload(struct packet_offload *po)
515 struct list_head *head = &offload_base;
516 struct packet_offload *po1;
518 spin_lock(&offload_lock);
520 list_for_each_entry(po1, head, list) {
522 list_del_rcu(&po->list);
527 pr_warn("dev_remove_offload: %p not found\n", po);
529 spin_unlock(&offload_lock);
533 * dev_remove_offload - remove packet offload handler
534 * @po: packet offload declaration
536 * Remove a packet offload handler that was previously added to the kernel
537 * offload handlers by dev_add_offload(). The passed &offload_type is
538 * removed from the kernel lists and can be freed or reused once this
541 * This call sleeps to guarantee that no CPU is looking at the packet
544 void dev_remove_offload(struct packet_offload *po)
546 __dev_remove_offload(po);
550 EXPORT_SYMBOL(dev_remove_offload);
552 /******************************************************************************
554 * Device Boot-time Settings Routines
556 ******************************************************************************/
558 /* Boot time configuration table */
559 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
562 * netdev_boot_setup_add - add new setup entry
563 * @name: name of the device
564 * @map: configured settings for the device
566 * Adds new setup entry to the dev_boot_setup list. The function
567 * returns 0 on error and 1 on success. This is a generic routine to
570 static int netdev_boot_setup_add(char *name, struct ifmap *map)
572 struct netdev_boot_setup *s;
576 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
577 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
578 memset(s[i].name, 0, sizeof(s[i].name));
579 strlcpy(s[i].name, name, IFNAMSIZ);
580 memcpy(&s[i].map, map, sizeof(s[i].map));
585 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
589 * netdev_boot_setup_check - check boot time settings
590 * @dev: the netdevice
592 * Check boot time settings for the device.
593 * The found settings are set for the device to be used
594 * later in the device probing.
595 * Returns 0 if no settings found, 1 if they are.
597 int netdev_boot_setup_check(struct net_device *dev)
599 struct netdev_boot_setup *s = dev_boot_setup;
602 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
603 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
604 !strcmp(dev->name, s[i].name)) {
605 dev->irq = s[i].map.irq;
606 dev->base_addr = s[i].map.base_addr;
607 dev->mem_start = s[i].map.mem_start;
608 dev->mem_end = s[i].map.mem_end;
614 EXPORT_SYMBOL(netdev_boot_setup_check);
618 * netdev_boot_base - get address from boot time settings
619 * @prefix: prefix for network device
620 * @unit: id for network device
622 * Check boot time settings for the base address of device.
623 * The found settings are set for the device to be used
624 * later in the device probing.
625 * Returns 0 if no settings found.
627 unsigned long netdev_boot_base(const char *prefix, int unit)
629 const struct netdev_boot_setup *s = dev_boot_setup;
633 sprintf(name, "%s%d", prefix, unit);
636 * If device already registered then return base of 1
637 * to indicate not to probe for this interface
639 if (__dev_get_by_name(&init_net, name))
642 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
643 if (!strcmp(name, s[i].name))
644 return s[i].map.base_addr;
649 * Saves at boot time configured settings for any netdevice.
651 int __init netdev_boot_setup(char *str)
656 str = get_options(str, ARRAY_SIZE(ints), ints);
661 memset(&map, 0, sizeof(map));
665 map.base_addr = ints[2];
667 map.mem_start = ints[3];
669 map.mem_end = ints[4];
671 /* Add new entry to the list */
672 return netdev_boot_setup_add(str, &map);
675 __setup("netdev=", netdev_boot_setup);
677 /*******************************************************************************
679 * Device Interface Subroutines
681 *******************************************************************************/
684 * dev_get_iflink - get 'iflink' value of a interface
685 * @dev: targeted interface
687 * Indicates the ifindex the interface is linked to.
688 * Physical interfaces have the same 'ifindex' and 'iflink' values.
691 int dev_get_iflink(const struct net_device *dev)
693 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
694 return dev->netdev_ops->ndo_get_iflink(dev);
698 EXPORT_SYMBOL(dev_get_iflink);
701 * dev_fill_metadata_dst - Retrieve tunnel egress information.
702 * @dev: targeted interface
705 * For better visibility of tunnel traffic OVS needs to retrieve
706 * egress tunnel information for a packet. Following API allows
707 * user to get this info.
709 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
711 struct ip_tunnel_info *info;
713 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
716 info = skb_tunnel_info_unclone(skb);
719 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
722 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
724 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
727 * __dev_get_by_name - find a device by its name
728 * @net: the applicable net namespace
729 * @name: name to find
731 * Find an interface by name. Must be called under RTNL semaphore
732 * or @dev_base_lock. If the name is found a pointer to the device
733 * is returned. If the name is not found then %NULL is returned. The
734 * reference counters are not incremented so the caller must be
735 * careful with locks.
738 struct net_device *__dev_get_by_name(struct net *net, const char *name)
740 struct net_device *dev;
741 struct hlist_head *head = dev_name_hash(net, name);
743 hlist_for_each_entry(dev, head, name_hlist)
744 if (!strncmp(dev->name, name, IFNAMSIZ))
749 EXPORT_SYMBOL(__dev_get_by_name);
752 * dev_get_by_name_rcu - find a device by its name
753 * @net: the applicable net namespace
754 * @name: name to find
756 * Find an interface by name.
757 * If the name is found a pointer to the device is returned.
758 * If the name is not found then %NULL is returned.
759 * The reference counters are not incremented so the caller must be
760 * careful with locks. The caller must hold RCU lock.
763 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
765 struct net_device *dev;
766 struct hlist_head *head = dev_name_hash(net, name);
768 hlist_for_each_entry_rcu(dev, head, name_hlist)
769 if (!strncmp(dev->name, name, IFNAMSIZ))
774 EXPORT_SYMBOL(dev_get_by_name_rcu);
777 * dev_get_by_name - find a device by its name
778 * @net: the applicable net namespace
779 * @name: name to find
781 * Find an interface by name. This can be called from any
782 * context and does its own locking. The returned handle has
783 * the usage count incremented and the caller must use dev_put() to
784 * release it when it is no longer needed. %NULL is returned if no
785 * matching device is found.
788 struct net_device *dev_get_by_name(struct net *net, const char *name)
790 struct net_device *dev;
793 dev = dev_get_by_name_rcu(net, name);
799 EXPORT_SYMBOL(dev_get_by_name);
802 * __dev_get_by_index - find a device by its ifindex
803 * @net: the applicable net namespace
804 * @ifindex: index of device
806 * Search for an interface by index. Returns %NULL if the device
807 * is not found or a pointer to the device. The device has not
808 * had its reference counter increased so the caller must be careful
809 * about locking. The caller must hold either the RTNL semaphore
813 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
815 struct net_device *dev;
816 struct hlist_head *head = dev_index_hash(net, ifindex);
818 hlist_for_each_entry(dev, head, index_hlist)
819 if (dev->ifindex == ifindex)
824 EXPORT_SYMBOL(__dev_get_by_index);
827 * dev_get_by_index_rcu - find a device by its ifindex
828 * @net: the applicable net namespace
829 * @ifindex: index of device
831 * Search for an interface by index. Returns %NULL if the device
832 * is not found or a pointer to the device. The device has not
833 * had its reference counter increased so the caller must be careful
834 * about locking. The caller must hold RCU lock.
837 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
839 struct net_device *dev;
840 struct hlist_head *head = dev_index_hash(net, ifindex);
842 hlist_for_each_entry_rcu(dev, head, index_hlist)
843 if (dev->ifindex == ifindex)
848 EXPORT_SYMBOL(dev_get_by_index_rcu);
852 * dev_get_by_index - find a device by its ifindex
853 * @net: the applicable net namespace
854 * @ifindex: index of device
856 * Search for an interface by index. Returns NULL if the device
857 * is not found or a pointer to the device. The device returned has
858 * had a reference added and the pointer is safe until the user calls
859 * dev_put to indicate they have finished with it.
862 struct net_device *dev_get_by_index(struct net *net, int ifindex)
864 struct net_device *dev;
867 dev = dev_get_by_index_rcu(net, ifindex);
873 EXPORT_SYMBOL(dev_get_by_index);
876 * dev_get_by_napi_id - find a device by napi_id
877 * @napi_id: ID of the NAPI struct
879 * Search for an interface by NAPI ID. Returns %NULL if the device
880 * is not found or a pointer to the device. The device has not had
881 * its reference counter increased so the caller must be careful
882 * about locking. The caller must hold RCU lock.
885 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
887 struct napi_struct *napi;
889 WARN_ON_ONCE(!rcu_read_lock_held());
891 if (napi_id < MIN_NAPI_ID)
894 napi = napi_by_id(napi_id);
896 return napi ? napi->dev : NULL;
898 EXPORT_SYMBOL(dev_get_by_napi_id);
901 * netdev_get_name - get a netdevice name, knowing its ifindex.
902 * @net: network namespace
903 * @name: a pointer to the buffer where the name will be stored.
904 * @ifindex: the ifindex of the interface to get the name from.
906 * The use of raw_seqcount_begin() and cond_resched() before
907 * retrying is required as we want to give the writers a chance
908 * to complete when CONFIG_PREEMPT is not set.
910 int netdev_get_name(struct net *net, char *name, int ifindex)
912 struct net_device *dev;
916 seq = raw_seqcount_begin(&devnet_rename_seq);
918 dev = dev_get_by_index_rcu(net, ifindex);
924 strcpy(name, dev->name);
926 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
935 * dev_getbyhwaddr_rcu - find a device by its hardware address
936 * @net: the applicable net namespace
937 * @type: media type of device
938 * @ha: hardware address
940 * Search for an interface by MAC address. Returns NULL if the device
941 * is not found or a pointer to the device.
942 * The caller must hold RCU or RTNL.
943 * The returned device has not had its ref count increased
944 * and the caller must therefore be careful about locking
948 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
951 struct net_device *dev;
953 for_each_netdev_rcu(net, dev)
954 if (dev->type == type &&
955 !memcmp(dev->dev_addr, ha, dev->addr_len))
960 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
962 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
964 struct net_device *dev;
967 for_each_netdev(net, dev)
968 if (dev->type == type)
973 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
975 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
977 struct net_device *dev, *ret = NULL;
980 for_each_netdev_rcu(net, dev)
981 if (dev->type == type) {
989 EXPORT_SYMBOL(dev_getfirstbyhwtype);
992 * __dev_get_by_flags - find any device with given flags
993 * @net: the applicable net namespace
994 * @if_flags: IFF_* values
995 * @mask: bitmask of bits in if_flags to check
997 * Search for any interface with the given flags. Returns NULL if a device
998 * is not found or a pointer to the device. Must be called inside
999 * rtnl_lock(), and result refcount is unchanged.
1002 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
1003 unsigned short mask)
1005 struct net_device *dev, *ret;
1010 for_each_netdev(net, dev) {
1011 if (((dev->flags ^ if_flags) & mask) == 0) {
1018 EXPORT_SYMBOL(__dev_get_by_flags);
1021 * dev_valid_name - check if name is okay for network device
1022 * @name: name string
1024 * Network device names need to be valid file names to
1025 * to allow sysfs to work. We also disallow any kind of
1028 bool dev_valid_name(const char *name)
1032 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1034 if (!strcmp(name, ".") || !strcmp(name, ".."))
1038 if (*name == '/' || *name == ':' || isspace(*name))
1044 EXPORT_SYMBOL(dev_valid_name);
1047 * __dev_alloc_name - allocate a name for a device
1048 * @net: network namespace to allocate the device name in
1049 * @name: name format string
1050 * @buf: scratch buffer and result name string
1052 * Passed a format string - eg "lt%d" it will try and find a suitable
1053 * id. It scans list of devices to build up a free map, then chooses
1054 * the first empty slot. The caller must hold the dev_base or rtnl lock
1055 * while allocating the name and adding the device in order to avoid
1057 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1058 * Returns the number of the unit assigned or a negative errno code.
1061 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1065 const int max_netdevices = 8*PAGE_SIZE;
1066 unsigned long *inuse;
1067 struct net_device *d;
1069 if (!dev_valid_name(name))
1072 p = strchr(name, '%');
1075 * Verify the string as this thing may have come from
1076 * the user. There must be either one "%d" and no other "%"
1079 if (p[1] != 'd' || strchr(p + 2, '%'))
1082 /* Use one page as a bit array of possible slots */
1083 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1087 for_each_netdev(net, d) {
1088 if (!sscanf(d->name, name, &i))
1090 if (i < 0 || i >= max_netdevices)
1093 /* avoid cases where sscanf is not exact inverse of printf */
1094 snprintf(buf, IFNAMSIZ, name, i);
1095 if (!strncmp(buf, d->name, IFNAMSIZ))
1099 i = find_first_zero_bit(inuse, max_netdevices);
1100 free_page((unsigned long) inuse);
1103 snprintf(buf, IFNAMSIZ, name, i);
1104 if (!__dev_get_by_name(net, buf))
1107 /* It is possible to run out of possible slots
1108 * when the name is long and there isn't enough space left
1109 * for the digits, or if all bits are used.
1114 static int dev_alloc_name_ns(struct net *net,
1115 struct net_device *dev,
1122 ret = __dev_alloc_name(net, name, buf);
1124 strlcpy(dev->name, buf, IFNAMSIZ);
1129 * dev_alloc_name - allocate a name for a device
1131 * @name: name format string
1133 * Passed a format string - eg "lt%d" it will try and find a suitable
1134 * id. It scans list of devices to build up a free map, then chooses
1135 * the first empty slot. The caller must hold the dev_base or rtnl lock
1136 * while allocating the name and adding the device in order to avoid
1138 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1139 * Returns the number of the unit assigned or a negative errno code.
1142 int dev_alloc_name(struct net_device *dev, const char *name)
1144 return dev_alloc_name_ns(dev_net(dev), dev, name);
1146 EXPORT_SYMBOL(dev_alloc_name);
1148 int dev_get_valid_name(struct net *net, struct net_device *dev,
1153 if (!dev_valid_name(name))
1156 if (strchr(name, '%'))
1157 return dev_alloc_name_ns(net, dev, name);
1158 else if (__dev_get_by_name(net, name))
1160 else if (dev->name != name)
1161 strlcpy(dev->name, name, IFNAMSIZ);
1165 EXPORT_SYMBOL(dev_get_valid_name);
1168 * dev_change_name - change name of a device
1170 * @newname: name (or format string) must be at least IFNAMSIZ
1172 * Change name of a device, can pass format strings "eth%d".
1175 int dev_change_name(struct net_device *dev, const char *newname)
1177 unsigned char old_assign_type;
1178 char oldname[IFNAMSIZ];
1184 BUG_ON(!dev_net(dev));
1187 if (dev->flags & IFF_UP)
1190 write_seqcount_begin(&devnet_rename_seq);
1192 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1193 write_seqcount_end(&devnet_rename_seq);
1197 memcpy(oldname, dev->name, IFNAMSIZ);
1199 err = dev_get_valid_name(net, dev, newname);
1201 write_seqcount_end(&devnet_rename_seq);
1205 if (oldname[0] && !strchr(oldname, '%'))
1206 netdev_info(dev, "renamed from %s\n", oldname);
1208 old_assign_type = dev->name_assign_type;
1209 dev->name_assign_type = NET_NAME_RENAMED;
1212 ret = device_rename(&dev->dev, dev->name);
1214 memcpy(dev->name, oldname, IFNAMSIZ);
1215 dev->name_assign_type = old_assign_type;
1216 write_seqcount_end(&devnet_rename_seq);
1220 write_seqcount_end(&devnet_rename_seq);
1222 netdev_adjacent_rename_links(dev, oldname);
1224 write_lock_bh(&dev_base_lock);
1225 hlist_del_rcu(&dev->name_hlist);
1226 write_unlock_bh(&dev_base_lock);
1230 write_lock_bh(&dev_base_lock);
1231 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1232 write_unlock_bh(&dev_base_lock);
1234 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1235 ret = notifier_to_errno(ret);
1238 /* err >= 0 after dev_alloc_name() or stores the first errno */
1241 write_seqcount_begin(&devnet_rename_seq);
1242 memcpy(dev->name, oldname, IFNAMSIZ);
1243 memcpy(oldname, newname, IFNAMSIZ);
1244 dev->name_assign_type = old_assign_type;
1245 old_assign_type = NET_NAME_RENAMED;
1248 pr_err("%s: name change rollback failed: %d\n",
1257 * dev_set_alias - change ifalias of a device
1259 * @alias: name up to IFALIASZ
1260 * @len: limit of bytes to copy from info
1262 * Set ifalias for a device,
1264 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1266 struct dev_ifalias *new_alias = NULL;
1268 if (len >= IFALIASZ)
1272 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1276 memcpy(new_alias->ifalias, alias, len);
1277 new_alias->ifalias[len] = 0;
1280 mutex_lock(&ifalias_mutex);
1281 rcu_swap_protected(dev->ifalias, new_alias,
1282 mutex_is_locked(&ifalias_mutex));
1283 mutex_unlock(&ifalias_mutex);
1286 kfree_rcu(new_alias, rcuhead);
1290 EXPORT_SYMBOL(dev_set_alias);
1293 * dev_get_alias - get ifalias of a device
1295 * @name: buffer to store name of ifalias
1296 * @len: size of buffer
1298 * get ifalias for a device. Caller must make sure dev cannot go
1299 * away, e.g. rcu read lock or own a reference count to device.
1301 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1303 const struct dev_ifalias *alias;
1307 alias = rcu_dereference(dev->ifalias);
1309 ret = snprintf(name, len, "%s", alias->ifalias);
1316 * netdev_features_change - device changes features
1317 * @dev: device to cause notification
1319 * Called to indicate a device has changed features.
1321 void netdev_features_change(struct net_device *dev)
1323 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1325 EXPORT_SYMBOL(netdev_features_change);
1328 * netdev_state_change - device changes state
1329 * @dev: device to cause notification
1331 * Called to indicate a device has changed state. This function calls
1332 * the notifier chains for netdev_chain and sends a NEWLINK message
1333 * to the routing socket.
1335 void netdev_state_change(struct net_device *dev)
1337 if (dev->flags & IFF_UP) {
1338 struct netdev_notifier_change_info change_info = {
1342 call_netdevice_notifiers_info(NETDEV_CHANGE,
1344 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1347 EXPORT_SYMBOL(netdev_state_change);
1350 * netdev_notify_peers - notify network peers about existence of @dev
1351 * @dev: network device
1353 * Generate traffic such that interested network peers are aware of
1354 * @dev, such as by generating a gratuitous ARP. This may be used when
1355 * a device wants to inform the rest of the network about some sort of
1356 * reconfiguration such as a failover event or virtual machine
1359 void netdev_notify_peers(struct net_device *dev)
1362 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1363 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1366 EXPORT_SYMBOL(netdev_notify_peers);
1368 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1370 const struct net_device_ops *ops = dev->netdev_ops;
1375 if (!netif_device_present(dev))
1378 /* Block netpoll from trying to do any rx path servicing.
1379 * If we don't do this there is a chance ndo_poll_controller
1380 * or ndo_poll may be running while we open the device
1382 netpoll_poll_disable(dev);
1384 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack);
1385 ret = notifier_to_errno(ret);
1389 set_bit(__LINK_STATE_START, &dev->state);
1391 if (ops->ndo_validate_addr)
1392 ret = ops->ndo_validate_addr(dev);
1394 if (!ret && ops->ndo_open)
1395 ret = ops->ndo_open(dev);
1397 netpoll_poll_enable(dev);
1400 clear_bit(__LINK_STATE_START, &dev->state);
1402 dev->flags |= IFF_UP;
1403 dev_set_rx_mode(dev);
1405 add_device_randomness(dev->dev_addr, dev->addr_len);
1412 * dev_open - prepare an interface for use.
1413 * @dev: device to open
1414 * @extack: netlink extended ack
1416 * Takes a device from down to up state. The device's private open
1417 * function is invoked and then the multicast lists are loaded. Finally
1418 * the device is moved into the up state and a %NETDEV_UP message is
1419 * sent to the netdev notifier chain.
1421 * Calling this function on an active interface is a nop. On a failure
1422 * a negative errno code is returned.
1424 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1428 if (dev->flags & IFF_UP)
1431 ret = __dev_open(dev, extack);
1435 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1436 call_netdevice_notifiers(NETDEV_UP, dev);
1440 EXPORT_SYMBOL(dev_open);
1442 static void __dev_close_many(struct list_head *head)
1444 struct net_device *dev;
1449 list_for_each_entry(dev, head, close_list) {
1450 /* Temporarily disable netpoll until the interface is down */
1451 netpoll_poll_disable(dev);
1453 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1455 clear_bit(__LINK_STATE_START, &dev->state);
1457 /* Synchronize to scheduled poll. We cannot touch poll list, it
1458 * can be even on different cpu. So just clear netif_running().
1460 * dev->stop() will invoke napi_disable() on all of it's
1461 * napi_struct instances on this device.
1463 smp_mb__after_atomic(); /* Commit netif_running(). */
1466 dev_deactivate_many(head);
1468 list_for_each_entry(dev, head, close_list) {
1469 const struct net_device_ops *ops = dev->netdev_ops;
1472 * Call the device specific close. This cannot fail.
1473 * Only if device is UP
1475 * We allow it to be called even after a DETACH hot-plug
1481 dev->flags &= ~IFF_UP;
1482 netpoll_poll_enable(dev);
1486 static void __dev_close(struct net_device *dev)
1490 list_add(&dev->close_list, &single);
1491 __dev_close_many(&single);
1495 void dev_close_many(struct list_head *head, bool unlink)
1497 struct net_device *dev, *tmp;
1499 /* Remove the devices that don't need to be closed */
1500 list_for_each_entry_safe(dev, tmp, head, close_list)
1501 if (!(dev->flags & IFF_UP))
1502 list_del_init(&dev->close_list);
1504 __dev_close_many(head);
1506 list_for_each_entry_safe(dev, tmp, head, close_list) {
1507 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1508 call_netdevice_notifiers(NETDEV_DOWN, dev);
1510 list_del_init(&dev->close_list);
1513 EXPORT_SYMBOL(dev_close_many);
1516 * dev_close - shutdown an interface.
1517 * @dev: device to shutdown
1519 * This function moves an active device into down state. A
1520 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1521 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1524 void dev_close(struct net_device *dev)
1526 if (dev->flags & IFF_UP) {
1529 list_add(&dev->close_list, &single);
1530 dev_close_many(&single, true);
1534 EXPORT_SYMBOL(dev_close);
1538 * dev_disable_lro - disable Large Receive Offload on a device
1541 * Disable Large Receive Offload (LRO) on a net device. Must be
1542 * called under RTNL. This is needed if received packets may be
1543 * forwarded to another interface.
1545 void dev_disable_lro(struct net_device *dev)
1547 struct net_device *lower_dev;
1548 struct list_head *iter;
1550 dev->wanted_features &= ~NETIF_F_LRO;
1551 netdev_update_features(dev);
1553 if (unlikely(dev->features & NETIF_F_LRO))
1554 netdev_WARN(dev, "failed to disable LRO!\n");
1556 netdev_for_each_lower_dev(dev, lower_dev, iter)
1557 dev_disable_lro(lower_dev);
1559 EXPORT_SYMBOL(dev_disable_lro);
1562 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1565 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be
1566 * called under RTNL. This is needed if Generic XDP is installed on
1569 static void dev_disable_gro_hw(struct net_device *dev)
1571 dev->wanted_features &= ~NETIF_F_GRO_HW;
1572 netdev_update_features(dev);
1574 if (unlikely(dev->features & NETIF_F_GRO_HW))
1575 netdev_WARN(dev, "failed to disable GRO_HW!\n");
1578 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1581 case NETDEV_##val: \
1582 return "NETDEV_" __stringify(val);
1584 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1585 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1586 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1587 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER)
1588 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO)
1589 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO)
1590 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1591 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1592 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1596 return "UNKNOWN_NETDEV_EVENT";
1598 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1600 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1601 struct net_device *dev)
1603 struct netdev_notifier_info info = {
1607 return nb->notifier_call(nb, val, &info);
1610 static int dev_boot_phase = 1;
1613 * register_netdevice_notifier - register a network notifier block
1616 * Register a notifier to be called when network device events occur.
1617 * The notifier passed is linked into the kernel structures and must
1618 * not be reused until it has been unregistered. A negative errno code
1619 * is returned on a failure.
1621 * When registered all registration and up events are replayed
1622 * to the new notifier to allow device to have a race free
1623 * view of the network device list.
1626 int register_netdevice_notifier(struct notifier_block *nb)
1628 struct net_device *dev;
1629 struct net_device *last;
1633 /* Close race with setup_net() and cleanup_net() */
1634 down_write(&pernet_ops_rwsem);
1636 err = raw_notifier_chain_register(&netdev_chain, nb);
1642 for_each_netdev(net, dev) {
1643 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1644 err = notifier_to_errno(err);
1648 if (!(dev->flags & IFF_UP))
1651 call_netdevice_notifier(nb, NETDEV_UP, dev);
1657 up_write(&pernet_ops_rwsem);
1663 for_each_netdev(net, dev) {
1667 if (dev->flags & IFF_UP) {
1668 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1670 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1672 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1677 raw_notifier_chain_unregister(&netdev_chain, nb);
1680 EXPORT_SYMBOL(register_netdevice_notifier);
1683 * unregister_netdevice_notifier - unregister a network notifier block
1686 * Unregister a notifier previously registered by
1687 * register_netdevice_notifier(). The notifier is unlinked into the
1688 * kernel structures and may then be reused. A negative errno code
1689 * is returned on a failure.
1691 * After unregistering unregister and down device events are synthesized
1692 * for all devices on the device list to the removed notifier to remove
1693 * the need for special case cleanup code.
1696 int unregister_netdevice_notifier(struct notifier_block *nb)
1698 struct net_device *dev;
1702 /* Close race with setup_net() and cleanup_net() */
1703 down_write(&pernet_ops_rwsem);
1705 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1710 for_each_netdev(net, dev) {
1711 if (dev->flags & IFF_UP) {
1712 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1714 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1716 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1721 up_write(&pernet_ops_rwsem);
1724 EXPORT_SYMBOL(unregister_netdevice_notifier);
1727 * call_netdevice_notifiers_info - call all network notifier blocks
1728 * @val: value passed unmodified to notifier function
1729 * @info: notifier information data
1731 * Call all network notifier blocks. Parameters and return value
1732 * are as for raw_notifier_call_chain().
1735 static int call_netdevice_notifiers_info(unsigned long val,
1736 struct netdev_notifier_info *info)
1739 return raw_notifier_call_chain(&netdev_chain, val, info);
1742 static int call_netdevice_notifiers_extack(unsigned long val,
1743 struct net_device *dev,
1744 struct netlink_ext_ack *extack)
1746 struct netdev_notifier_info info = {
1751 return call_netdevice_notifiers_info(val, &info);
1755 * call_netdevice_notifiers - call all network notifier blocks
1756 * @val: value passed unmodified to notifier function
1757 * @dev: net_device pointer passed unmodified to notifier function
1759 * Call all network notifier blocks. Parameters and return value
1760 * are as for raw_notifier_call_chain().
1763 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1765 return call_netdevice_notifiers_extack(val, dev, NULL);
1767 EXPORT_SYMBOL(call_netdevice_notifiers);
1770 * call_netdevice_notifiers_mtu - call all network notifier blocks
1771 * @val: value passed unmodified to notifier function
1772 * @dev: net_device pointer passed unmodified to notifier function
1773 * @arg: additional u32 argument passed to the notifier function
1775 * Call all network notifier blocks. Parameters and return value
1776 * are as for raw_notifier_call_chain().
1778 static int call_netdevice_notifiers_mtu(unsigned long val,
1779 struct net_device *dev, u32 arg)
1781 struct netdev_notifier_info_ext info = {
1786 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
1788 return call_netdevice_notifiers_info(val, &info.info);
1791 #ifdef CONFIG_NET_INGRESS
1792 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
1794 void net_inc_ingress_queue(void)
1796 static_branch_inc(&ingress_needed_key);
1798 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
1800 void net_dec_ingress_queue(void)
1802 static_branch_dec(&ingress_needed_key);
1804 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
1807 #ifdef CONFIG_NET_EGRESS
1808 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
1810 void net_inc_egress_queue(void)
1812 static_branch_inc(&egress_needed_key);
1814 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
1816 void net_dec_egress_queue(void)
1818 static_branch_dec(&egress_needed_key);
1820 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
1823 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
1824 #ifdef CONFIG_JUMP_LABEL
1825 static atomic_t netstamp_needed_deferred;
1826 static atomic_t netstamp_wanted;
1827 static void netstamp_clear(struct work_struct *work)
1829 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1832 wanted = atomic_add_return(deferred, &netstamp_wanted);
1834 static_branch_enable(&netstamp_needed_key);
1836 static_branch_disable(&netstamp_needed_key);
1838 static DECLARE_WORK(netstamp_work, netstamp_clear);
1841 void net_enable_timestamp(void)
1843 #ifdef CONFIG_JUMP_LABEL
1847 wanted = atomic_read(&netstamp_wanted);
1850 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted)
1853 atomic_inc(&netstamp_needed_deferred);
1854 schedule_work(&netstamp_work);
1856 static_branch_inc(&netstamp_needed_key);
1859 EXPORT_SYMBOL(net_enable_timestamp);
1861 void net_disable_timestamp(void)
1863 #ifdef CONFIG_JUMP_LABEL
1867 wanted = atomic_read(&netstamp_wanted);
1870 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted)
1873 atomic_dec(&netstamp_needed_deferred);
1874 schedule_work(&netstamp_work);
1876 static_branch_dec(&netstamp_needed_key);
1879 EXPORT_SYMBOL(net_disable_timestamp);
1881 static inline void net_timestamp_set(struct sk_buff *skb)
1884 if (static_branch_unlikely(&netstamp_needed_key))
1885 __net_timestamp(skb);
1888 #define net_timestamp_check(COND, SKB) \
1889 if (static_branch_unlikely(&netstamp_needed_key)) { \
1890 if ((COND) && !(SKB)->tstamp) \
1891 __net_timestamp(SKB); \
1894 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
1898 if (!(dev->flags & IFF_UP))
1901 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1902 if (skb->len <= len)
1905 /* if TSO is enabled, we don't care about the length as the packet
1906 * could be forwarded without being segmented before
1908 if (skb_is_gso(skb))
1913 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1915 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1917 int ret = ____dev_forward_skb(dev, skb);
1920 skb->protocol = eth_type_trans(skb, dev);
1921 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1926 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1929 * dev_forward_skb - loopback an skb to another netif
1931 * @dev: destination network device
1932 * @skb: buffer to forward
1935 * NET_RX_SUCCESS (no congestion)
1936 * NET_RX_DROP (packet was dropped, but freed)
1938 * dev_forward_skb can be used for injecting an skb from the
1939 * start_xmit function of one device into the receive queue
1940 * of another device.
1942 * The receiving device may be in another namespace, so
1943 * we have to clear all information in the skb that could
1944 * impact namespace isolation.
1946 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1948 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1950 EXPORT_SYMBOL_GPL(dev_forward_skb);
1952 static inline int deliver_skb(struct sk_buff *skb,
1953 struct packet_type *pt_prev,
1954 struct net_device *orig_dev)
1956 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1958 refcount_inc(&skb->users);
1959 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1962 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1963 struct packet_type **pt,
1964 struct net_device *orig_dev,
1966 struct list_head *ptype_list)
1968 struct packet_type *ptype, *pt_prev = *pt;
1970 list_for_each_entry_rcu(ptype, ptype_list, list) {
1971 if (ptype->type != type)
1974 deliver_skb(skb, pt_prev, orig_dev);
1980 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1982 if (!ptype->af_packet_priv || !skb->sk)
1985 if (ptype->id_match)
1986 return ptype->id_match(ptype, skb->sk);
1987 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1994 * dev_nit_active - return true if any network interface taps are in use
1996 * @dev: network device to check for the presence of taps
1998 bool dev_nit_active(struct net_device *dev)
2000 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all);
2002 EXPORT_SYMBOL_GPL(dev_nit_active);
2005 * Support routine. Sends outgoing frames to any network
2006 * taps currently in use.
2009 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
2011 struct packet_type *ptype;
2012 struct sk_buff *skb2 = NULL;
2013 struct packet_type *pt_prev = NULL;
2014 struct list_head *ptype_list = &ptype_all;
2018 list_for_each_entry_rcu(ptype, ptype_list, list) {
2019 if (ptype->ignore_outgoing)
2022 /* Never send packets back to the socket
2023 * they originated from - MvS (miquels@drinkel.ow.org)
2025 if (skb_loop_sk(ptype, skb))
2029 deliver_skb(skb2, pt_prev, skb->dev);
2034 /* need to clone skb, done only once */
2035 skb2 = skb_clone(skb, GFP_ATOMIC);
2039 net_timestamp_set(skb2);
2041 /* skb->nh should be correctly
2042 * set by sender, so that the second statement is
2043 * just protection against buggy protocols.
2045 skb_reset_mac_header(skb2);
2047 if (skb_network_header(skb2) < skb2->data ||
2048 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2049 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2050 ntohs(skb2->protocol),
2052 skb_reset_network_header(skb2);
2055 skb2->transport_header = skb2->network_header;
2056 skb2->pkt_type = PACKET_OUTGOING;
2060 if (ptype_list == &ptype_all) {
2061 ptype_list = &dev->ptype_all;
2066 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2067 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2073 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2076 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2077 * @dev: Network device
2078 * @txq: number of queues available
2080 * If real_num_tx_queues is changed the tc mappings may no longer be
2081 * valid. To resolve this verify the tc mapping remains valid and if
2082 * not NULL the mapping. With no priorities mapping to this
2083 * offset/count pair it will no longer be used. In the worst case TC0
2084 * is invalid nothing can be done so disable priority mappings. If is
2085 * expected that drivers will fix this mapping if they can before
2086 * calling netif_set_real_num_tx_queues.
2088 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2091 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2093 /* If TC0 is invalidated disable TC mapping */
2094 if (tc->offset + tc->count > txq) {
2095 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2100 /* Invalidated prio to tc mappings set to TC0 */
2101 for (i = 1; i < TC_BITMASK + 1; i++) {
2102 int q = netdev_get_prio_tc_map(dev, i);
2104 tc = &dev->tc_to_txq[q];
2105 if (tc->offset + tc->count > txq) {
2106 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
2108 netdev_set_prio_tc_map(dev, i, 0);
2113 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2116 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2119 /* walk through the TCs and see if it falls into any of them */
2120 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2121 if ((txq - tc->offset) < tc->count)
2125 /* didn't find it, just return -1 to indicate no match */
2131 EXPORT_SYMBOL(netdev_txq_to_tc);
2134 struct static_key xps_needed __read_mostly;
2135 EXPORT_SYMBOL(xps_needed);
2136 struct static_key xps_rxqs_needed __read_mostly;
2137 EXPORT_SYMBOL(xps_rxqs_needed);
2138 static DEFINE_MUTEX(xps_map_mutex);
2139 #define xmap_dereference(P) \
2140 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2142 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2145 struct xps_map *map = NULL;
2149 map = xmap_dereference(dev_maps->attr_map[tci]);
2153 for (pos = map->len; pos--;) {
2154 if (map->queues[pos] != index)
2158 map->queues[pos] = map->queues[--map->len];
2162 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2163 kfree_rcu(map, rcu);
2170 static bool remove_xps_queue_cpu(struct net_device *dev,
2171 struct xps_dev_maps *dev_maps,
2172 int cpu, u16 offset, u16 count)
2174 int num_tc = dev->num_tc ? : 1;
2175 bool active = false;
2178 for (tci = cpu * num_tc; num_tc--; tci++) {
2181 for (i = count, j = offset; i--; j++) {
2182 if (!remove_xps_queue(dev_maps, tci, j))
2192 static void reset_xps_maps(struct net_device *dev,
2193 struct xps_dev_maps *dev_maps,
2197 static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2198 RCU_INIT_POINTER(dev->xps_rxqs_map, NULL);
2200 RCU_INIT_POINTER(dev->xps_cpus_map, NULL);
2202 static_key_slow_dec_cpuslocked(&xps_needed);
2203 kfree_rcu(dev_maps, rcu);
2206 static void clean_xps_maps(struct net_device *dev, const unsigned long *mask,
2207 struct xps_dev_maps *dev_maps, unsigned int nr_ids,
2208 u16 offset, u16 count, bool is_rxqs_map)
2210 bool active = false;
2213 for (j = -1; j = netif_attrmask_next(j, mask, nr_ids),
2215 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset,
2218 reset_xps_maps(dev, dev_maps, is_rxqs_map);
2221 for (i = offset + (count - 1); count--; i--) {
2222 netdev_queue_numa_node_write(
2223 netdev_get_tx_queue(dev, i),
2229 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2232 const unsigned long *possible_mask = NULL;
2233 struct xps_dev_maps *dev_maps;
2234 unsigned int nr_ids;
2236 if (!static_key_false(&xps_needed))
2240 mutex_lock(&xps_map_mutex);
2242 if (static_key_false(&xps_rxqs_needed)) {
2243 dev_maps = xmap_dereference(dev->xps_rxqs_map);
2245 nr_ids = dev->num_rx_queues;
2246 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids,
2247 offset, count, true);
2251 dev_maps = xmap_dereference(dev->xps_cpus_map);
2255 if (num_possible_cpus() > 1)
2256 possible_mask = cpumask_bits(cpu_possible_mask);
2257 nr_ids = nr_cpu_ids;
2258 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, offset, count,
2262 mutex_unlock(&xps_map_mutex);
2266 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2268 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2271 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2272 u16 index, bool is_rxqs_map)
2274 struct xps_map *new_map;
2275 int alloc_len = XPS_MIN_MAP_ALLOC;
2278 for (pos = 0; map && pos < map->len; pos++) {
2279 if (map->queues[pos] != index)
2284 /* Need to add tx-queue to this CPU's/rx-queue's existing map */
2286 if (pos < map->alloc_len)
2289 alloc_len = map->alloc_len * 2;
2292 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2296 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2298 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2299 cpu_to_node(attr_index));
2303 for (i = 0; i < pos; i++)
2304 new_map->queues[i] = map->queues[i];
2305 new_map->alloc_len = alloc_len;
2311 /* Must be called under cpus_read_lock */
2312 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2313 u16 index, bool is_rxqs_map)
2315 const unsigned long *online_mask = NULL, *possible_mask = NULL;
2316 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2317 int i, j, tci, numa_node_id = -2;
2318 int maps_sz, num_tc = 1, tc = 0;
2319 struct xps_map *map, *new_map;
2320 bool active = false;
2321 unsigned int nr_ids;
2324 /* Do not allow XPS on subordinate device directly */
2325 num_tc = dev->num_tc;
2329 /* If queue belongs to subordinate dev use its map */
2330 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2332 tc = netdev_txq_to_tc(dev, index);
2337 mutex_lock(&xps_map_mutex);
2339 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2340 dev_maps = xmap_dereference(dev->xps_rxqs_map);
2341 nr_ids = dev->num_rx_queues;
2343 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2344 if (num_possible_cpus() > 1) {
2345 online_mask = cpumask_bits(cpu_online_mask);
2346 possible_mask = cpumask_bits(cpu_possible_mask);
2348 dev_maps = xmap_dereference(dev->xps_cpus_map);
2349 nr_ids = nr_cpu_ids;
2352 if (maps_sz < L1_CACHE_BYTES)
2353 maps_sz = L1_CACHE_BYTES;
2355 /* allocate memory for queue storage */
2356 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2359 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2360 if (!new_dev_maps) {
2361 mutex_unlock(&xps_map_mutex);
2365 tci = j * num_tc + tc;
2366 map = dev_maps ? xmap_dereference(dev_maps->attr_map[tci]) :
2369 map = expand_xps_map(map, j, index, is_rxqs_map);
2373 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2377 goto out_no_new_maps;
2380 /* Increment static keys at most once per type */
2381 static_key_slow_inc_cpuslocked(&xps_needed);
2383 static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2386 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2388 /* copy maps belonging to foreign traffic classes */
2389 for (i = tc, tci = j * num_tc; dev_maps && i--; tci++) {
2390 /* fill in the new device map from the old device map */
2391 map = xmap_dereference(dev_maps->attr_map[tci]);
2392 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2395 /* We need to explicitly update tci as prevous loop
2396 * could break out early if dev_maps is NULL.
2398 tci = j * num_tc + tc;
2400 if (netif_attr_test_mask(j, mask, nr_ids) &&
2401 netif_attr_test_online(j, online_mask, nr_ids)) {
2402 /* add tx-queue to CPU/rx-queue maps */
2405 map = xmap_dereference(new_dev_maps->attr_map[tci]);
2406 while ((pos < map->len) && (map->queues[pos] != index))
2409 if (pos == map->len)
2410 map->queues[map->len++] = index;
2413 if (numa_node_id == -2)
2414 numa_node_id = cpu_to_node(j);
2415 else if (numa_node_id != cpu_to_node(j))
2419 } else if (dev_maps) {
2420 /* fill in the new device map from the old device map */
2421 map = xmap_dereference(dev_maps->attr_map[tci]);
2422 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2425 /* copy maps belonging to foreign traffic classes */
2426 for (i = num_tc - tc, tci++; dev_maps && --i; tci++) {
2427 /* fill in the new device map from the old device map */
2428 map = xmap_dereference(dev_maps->attr_map[tci]);
2429 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2434 rcu_assign_pointer(dev->xps_rxqs_map, new_dev_maps);
2436 rcu_assign_pointer(dev->xps_cpus_map, new_dev_maps);
2438 /* Cleanup old maps */
2440 goto out_no_old_maps;
2442 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2444 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2445 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2446 map = xmap_dereference(dev_maps->attr_map[tci]);
2447 if (map && map != new_map)
2448 kfree_rcu(map, rcu);
2452 kfree_rcu(dev_maps, rcu);
2455 dev_maps = new_dev_maps;
2460 /* update Tx queue numa node */
2461 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2462 (numa_node_id >= 0) ?
2463 numa_node_id : NUMA_NO_NODE);
2469 /* removes tx-queue from unused CPUs/rx-queues */
2470 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2472 for (i = tc, tci = j * num_tc; i--; tci++)
2473 active |= remove_xps_queue(dev_maps, tci, index);
2474 if (!netif_attr_test_mask(j, mask, nr_ids) ||
2475 !netif_attr_test_online(j, online_mask, nr_ids))
2476 active |= remove_xps_queue(dev_maps, tci, index);
2477 for (i = num_tc - tc, tci++; --i; tci++)
2478 active |= remove_xps_queue(dev_maps, tci, index);
2481 /* free map if not active */
2483 reset_xps_maps(dev, dev_maps, is_rxqs_map);
2486 mutex_unlock(&xps_map_mutex);
2490 /* remove any maps that we added */
2491 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2493 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2494 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2496 xmap_dereference(dev_maps->attr_map[tci]) :
2498 if (new_map && new_map != map)
2503 mutex_unlock(&xps_map_mutex);
2505 kfree(new_dev_maps);
2508 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2510 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2516 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, false);
2521 EXPORT_SYMBOL(netif_set_xps_queue);
2524 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2526 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2528 /* Unbind any subordinate channels */
2529 while (txq-- != &dev->_tx[0]) {
2531 netdev_unbind_sb_channel(dev, txq->sb_dev);
2535 void netdev_reset_tc(struct net_device *dev)
2538 netif_reset_xps_queues_gt(dev, 0);
2540 netdev_unbind_all_sb_channels(dev);
2542 /* Reset TC configuration of device */
2544 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2545 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2547 EXPORT_SYMBOL(netdev_reset_tc);
2549 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2551 if (tc >= dev->num_tc)
2555 netif_reset_xps_queues(dev, offset, count);
2557 dev->tc_to_txq[tc].count = count;
2558 dev->tc_to_txq[tc].offset = offset;
2561 EXPORT_SYMBOL(netdev_set_tc_queue);
2563 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2565 if (num_tc > TC_MAX_QUEUE)
2569 netif_reset_xps_queues_gt(dev, 0);
2571 netdev_unbind_all_sb_channels(dev);
2573 dev->num_tc = num_tc;
2576 EXPORT_SYMBOL(netdev_set_num_tc);
2578 void netdev_unbind_sb_channel(struct net_device *dev,
2579 struct net_device *sb_dev)
2581 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2584 netif_reset_xps_queues_gt(sb_dev, 0);
2586 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2587 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2589 while (txq-- != &dev->_tx[0]) {
2590 if (txq->sb_dev == sb_dev)
2594 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2596 int netdev_bind_sb_channel_queue(struct net_device *dev,
2597 struct net_device *sb_dev,
2598 u8 tc, u16 count, u16 offset)
2600 /* Make certain the sb_dev and dev are already configured */
2601 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2604 /* We cannot hand out queues we don't have */
2605 if ((offset + count) > dev->real_num_tx_queues)
2608 /* Record the mapping */
2609 sb_dev->tc_to_txq[tc].count = count;
2610 sb_dev->tc_to_txq[tc].offset = offset;
2612 /* Provide a way for Tx queue to find the tc_to_txq map or
2613 * XPS map for itself.
2616 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
2620 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
2622 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
2624 /* Do not use a multiqueue device to represent a subordinate channel */
2625 if (netif_is_multiqueue(dev))
2628 /* We allow channels 1 - 32767 to be used for subordinate channels.
2629 * Channel 0 is meant to be "native" mode and used only to represent
2630 * the main root device. We allow writing 0 to reset the device back
2631 * to normal mode after being used as a subordinate channel.
2633 if (channel > S16_MAX)
2636 dev->num_tc = -channel;
2640 EXPORT_SYMBOL(netdev_set_sb_channel);
2643 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2644 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2646 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2651 disabling = txq < dev->real_num_tx_queues;
2653 if (txq < 1 || txq > dev->num_tx_queues)
2656 if (dev->reg_state == NETREG_REGISTERED ||
2657 dev->reg_state == NETREG_UNREGISTERING) {
2660 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2666 netif_setup_tc(dev, txq);
2668 dev->real_num_tx_queues = txq;
2672 qdisc_reset_all_tx_gt(dev, txq);
2674 netif_reset_xps_queues_gt(dev, txq);
2678 dev->real_num_tx_queues = txq;
2683 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2687 * netif_set_real_num_rx_queues - set actual number of RX queues used
2688 * @dev: Network device
2689 * @rxq: Actual number of RX queues
2691 * This must be called either with the rtnl_lock held or before
2692 * registration of the net device. Returns 0 on success, or a
2693 * negative error code. If called before registration, it always
2696 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2700 if (rxq < 1 || rxq > dev->num_rx_queues)
2703 if (dev->reg_state == NETREG_REGISTERED) {
2706 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2712 dev->real_num_rx_queues = rxq;
2715 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2719 * netif_get_num_default_rss_queues - default number of RSS queues
2721 * This routine should set an upper limit on the number of RSS queues
2722 * used by default by multiqueue devices.
2724 int netif_get_num_default_rss_queues(void)
2726 return is_kdump_kernel() ?
2727 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2729 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2731 static void __netif_reschedule(struct Qdisc *q)
2733 struct softnet_data *sd;
2734 unsigned long flags;
2736 local_irq_save(flags);
2737 sd = this_cpu_ptr(&softnet_data);
2738 q->next_sched = NULL;
2739 *sd->output_queue_tailp = q;
2740 sd->output_queue_tailp = &q->next_sched;
2741 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2742 local_irq_restore(flags);
2745 void __netif_schedule(struct Qdisc *q)
2747 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2748 __netif_reschedule(q);
2750 EXPORT_SYMBOL(__netif_schedule);
2752 struct dev_kfree_skb_cb {
2753 enum skb_free_reason reason;
2756 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2758 return (struct dev_kfree_skb_cb *)skb->cb;
2761 void netif_schedule_queue(struct netdev_queue *txq)
2764 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2765 struct Qdisc *q = rcu_dereference(txq->qdisc);
2767 __netif_schedule(q);
2771 EXPORT_SYMBOL(netif_schedule_queue);
2773 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2775 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2779 q = rcu_dereference(dev_queue->qdisc);
2780 __netif_schedule(q);
2784 EXPORT_SYMBOL(netif_tx_wake_queue);
2786 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2788 unsigned long flags;
2793 if (likely(refcount_read(&skb->users) == 1)) {
2795 refcount_set(&skb->users, 0);
2796 } else if (likely(!refcount_dec_and_test(&skb->users))) {
2799 get_kfree_skb_cb(skb)->reason = reason;
2800 local_irq_save(flags);
2801 skb->next = __this_cpu_read(softnet_data.completion_queue);
2802 __this_cpu_write(softnet_data.completion_queue, skb);
2803 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2804 local_irq_restore(flags);
2806 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2808 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2810 if (in_irq() || irqs_disabled())
2811 __dev_kfree_skb_irq(skb, reason);
2815 EXPORT_SYMBOL(__dev_kfree_skb_any);
2819 * netif_device_detach - mark device as removed
2820 * @dev: network device
2822 * Mark device as removed from system and therefore no longer available.
2824 void netif_device_detach(struct net_device *dev)
2826 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2827 netif_running(dev)) {
2828 netif_tx_stop_all_queues(dev);
2831 EXPORT_SYMBOL(netif_device_detach);
2834 * netif_device_attach - mark device as attached
2835 * @dev: network device
2837 * Mark device as attached from system and restart if needed.
2839 void netif_device_attach(struct net_device *dev)
2841 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2842 netif_running(dev)) {
2843 netif_tx_wake_all_queues(dev);
2844 __netdev_watchdog_up(dev);
2847 EXPORT_SYMBOL(netif_device_attach);
2850 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2851 * to be used as a distribution range.
2853 static u16 skb_tx_hash(const struct net_device *dev,
2854 const struct net_device *sb_dev,
2855 struct sk_buff *skb)
2859 u16 qcount = dev->real_num_tx_queues;
2862 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2864 qoffset = sb_dev->tc_to_txq[tc].offset;
2865 qcount = sb_dev->tc_to_txq[tc].count;
2868 if (skb_rx_queue_recorded(skb)) {
2869 hash = skb_get_rx_queue(skb);
2870 while (unlikely(hash >= qcount))
2872 return hash + qoffset;
2875 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
2878 static void skb_warn_bad_offload(const struct sk_buff *skb)
2880 static const netdev_features_t null_features;
2881 struct net_device *dev = skb->dev;
2882 const char *name = "";
2884 if (!net_ratelimit())
2888 if (dev->dev.parent)
2889 name = dev_driver_string(dev->dev.parent);
2891 name = netdev_name(dev);
2893 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2894 "gso_type=%d ip_summed=%d\n",
2895 name, dev ? &dev->features : &null_features,
2896 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2897 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2898 skb_shinfo(skb)->gso_type, skb->ip_summed);
2902 * Invalidate hardware checksum when packet is to be mangled, and
2903 * complete checksum manually on outgoing path.
2905 int skb_checksum_help(struct sk_buff *skb)
2908 int ret = 0, offset;
2910 if (skb->ip_summed == CHECKSUM_COMPLETE)
2911 goto out_set_summed;
2913 if (unlikely(skb_shinfo(skb)->gso_size)) {
2914 skb_warn_bad_offload(skb);
2918 /* Before computing a checksum, we should make sure no frag could
2919 * be modified by an external entity : checksum could be wrong.
2921 if (skb_has_shared_frag(skb)) {
2922 ret = __skb_linearize(skb);
2927 offset = skb_checksum_start_offset(skb);
2928 BUG_ON(offset >= skb_headlen(skb));
2929 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2931 offset += skb->csum_offset;
2932 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2934 if (skb_cloned(skb) &&
2935 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2936 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2941 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
2943 skb->ip_summed = CHECKSUM_NONE;
2947 EXPORT_SYMBOL(skb_checksum_help);
2949 int skb_crc32c_csum_help(struct sk_buff *skb)
2952 int ret = 0, offset, start;
2954 if (skb->ip_summed != CHECKSUM_PARTIAL)
2957 if (unlikely(skb_is_gso(skb)))
2960 /* Before computing a checksum, we should make sure no frag could
2961 * be modified by an external entity : checksum could be wrong.
2963 if (unlikely(skb_has_shared_frag(skb))) {
2964 ret = __skb_linearize(skb);
2968 start = skb_checksum_start_offset(skb);
2969 offset = start + offsetof(struct sctphdr, checksum);
2970 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
2974 if (skb_cloned(skb) &&
2975 !skb_clone_writable(skb, offset + sizeof(__le32))) {
2976 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2980 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
2981 skb->len - start, ~(__u32)0,
2983 *(__le32 *)(skb->data + offset) = crc32c_csum;
2984 skb->ip_summed = CHECKSUM_NONE;
2985 skb->csum_not_inet = 0;
2990 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2992 __be16 type = skb->protocol;
2994 /* Tunnel gso handlers can set protocol to ethernet. */
2995 if (type == htons(ETH_P_TEB)) {
2998 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
3001 eth = (struct ethhdr *)skb->data;
3002 type = eth->h_proto;
3005 return __vlan_get_protocol(skb, type, depth);
3009 * skb_mac_gso_segment - mac layer segmentation handler.
3010 * @skb: buffer to segment
3011 * @features: features for the output path (see dev->features)
3013 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
3014 netdev_features_t features)
3016 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
3017 struct packet_offload *ptype;
3018 int vlan_depth = skb->mac_len;
3019 __be16 type = skb_network_protocol(skb, &vlan_depth);
3021 if (unlikely(!type))
3022 return ERR_PTR(-EINVAL);
3024 __skb_pull(skb, vlan_depth);
3027 list_for_each_entry_rcu(ptype, &offload_base, list) {
3028 if (ptype->type == type && ptype->callbacks.gso_segment) {
3029 segs = ptype->callbacks.gso_segment(skb, features);
3035 __skb_push(skb, skb->data - skb_mac_header(skb));
3039 EXPORT_SYMBOL(skb_mac_gso_segment);
3042 /* openvswitch calls this on rx path, so we need a different check.
3044 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
3047 return skb->ip_summed != CHECKSUM_PARTIAL &&
3048 skb->ip_summed != CHECKSUM_UNNECESSARY;
3050 return skb->ip_summed == CHECKSUM_NONE;
3054 * __skb_gso_segment - Perform segmentation on skb.
3055 * @skb: buffer to segment
3056 * @features: features for the output path (see dev->features)
3057 * @tx_path: whether it is called in TX path
3059 * This function segments the given skb and returns a list of segments.
3061 * It may return NULL if the skb requires no segmentation. This is
3062 * only possible when GSO is used for verifying header integrity.
3064 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
3066 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
3067 netdev_features_t features, bool tx_path)
3069 struct sk_buff *segs;
3071 if (unlikely(skb_needs_check(skb, tx_path))) {
3074 /* We're going to init ->check field in TCP or UDP header */
3075 err = skb_cow_head(skb, 0);
3077 return ERR_PTR(err);
3080 /* Only report GSO partial support if it will enable us to
3081 * support segmentation on this frame without needing additional
3084 if (features & NETIF_F_GSO_PARTIAL) {
3085 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
3086 struct net_device *dev = skb->dev;
3088 partial_features |= dev->features & dev->gso_partial_features;
3089 if (!skb_gso_ok(skb, features | partial_features))
3090 features &= ~NETIF_F_GSO_PARTIAL;
3093 BUILD_BUG_ON(SKB_SGO_CB_OFFSET +
3094 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
3096 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
3097 SKB_GSO_CB(skb)->encap_level = 0;
3099 skb_reset_mac_header(skb);
3100 skb_reset_mac_len(skb);
3102 segs = skb_mac_gso_segment(skb, features);
3104 if (unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
3105 skb_warn_bad_offload(skb);
3109 EXPORT_SYMBOL(__skb_gso_segment);
3111 /* Take action when hardware reception checksum errors are detected. */
3113 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3115 if (net_ratelimit()) {
3116 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
3118 pr_err("dev features: %pNF\n", &dev->features);
3119 pr_err("skb len=%u data_len=%u pkt_type=%u gso_size=%u gso_type=%u nr_frags=%u ip_summed=%u csum=%x csum_complete_sw=%d csum_valid=%d csum_level=%u\n",
3120 skb->len, skb->data_len, skb->pkt_type,
3121 skb_shinfo(skb)->gso_size, skb_shinfo(skb)->gso_type,
3122 skb_shinfo(skb)->nr_frags, skb->ip_summed, skb->csum,
3123 skb->csum_complete_sw, skb->csum_valid, skb->csum_level);
3127 EXPORT_SYMBOL(netdev_rx_csum_fault);
3130 /* XXX: check that highmem exists at all on the given machine. */
3131 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3133 #ifdef CONFIG_HIGHMEM
3136 if (!(dev->features & NETIF_F_HIGHDMA)) {
3137 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3138 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3140 if (PageHighMem(skb_frag_page(frag)))
3148 /* If MPLS offload request, verify we are testing hardware MPLS features
3149 * instead of standard features for the netdev.
3151 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3152 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3153 netdev_features_t features,
3156 if (eth_p_mpls(type))
3157 features &= skb->dev->mpls_features;
3162 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3163 netdev_features_t features,
3170 static netdev_features_t harmonize_features(struct sk_buff *skb,
3171 netdev_features_t features)
3176 type = skb_network_protocol(skb, &tmp);
3177 features = net_mpls_features(skb, features, type);
3179 if (skb->ip_summed != CHECKSUM_NONE &&
3180 !can_checksum_protocol(features, type)) {
3181 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3183 if (illegal_highdma(skb->dev, skb))
3184 features &= ~NETIF_F_SG;
3189 netdev_features_t passthru_features_check(struct sk_buff *skb,
3190 struct net_device *dev,
3191 netdev_features_t features)
3195 EXPORT_SYMBOL(passthru_features_check);
3197 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3198 struct net_device *dev,
3199 netdev_features_t features)
3201 return vlan_features_check(skb, features);
3204 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3205 struct net_device *dev,
3206 netdev_features_t features)
3208 u16 gso_segs = skb_shinfo(skb)->gso_segs;
3210 if (gso_segs > dev->gso_max_segs)
3211 return features & ~NETIF_F_GSO_MASK;
3213 /* Support for GSO partial features requires software
3214 * intervention before we can actually process the packets
3215 * so we need to strip support for any partial features now
3216 * and we can pull them back in after we have partially
3217 * segmented the frame.
3219 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3220 features &= ~dev->gso_partial_features;
3222 /* Make sure to clear the IPv4 ID mangling feature if the
3223 * IPv4 header has the potential to be fragmented.
3225 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3226 struct iphdr *iph = skb->encapsulation ?
3227 inner_ip_hdr(skb) : ip_hdr(skb);
3229 if (!(iph->frag_off & htons(IP_DF)))
3230 features &= ~NETIF_F_TSO_MANGLEID;
3236 netdev_features_t netif_skb_features(struct sk_buff *skb)
3238 struct net_device *dev = skb->dev;
3239 netdev_features_t features = dev->features;
3241 if (skb_is_gso(skb))
3242 features = gso_features_check(skb, dev, features);
3244 /* If encapsulation offload request, verify we are testing
3245 * hardware encapsulation features instead of standard
3246 * features for the netdev
3248 if (skb->encapsulation)
3249 features &= dev->hw_enc_features;
3251 if (skb_vlan_tagged(skb))
3252 features = netdev_intersect_features(features,
3253 dev->vlan_features |
3254 NETIF_F_HW_VLAN_CTAG_TX |
3255 NETIF_F_HW_VLAN_STAG_TX);
3257 if (dev->netdev_ops->ndo_features_check)
3258 features &= dev->netdev_ops->ndo_features_check(skb, dev,
3261 features &= dflt_features_check(skb, dev, features);
3263 return harmonize_features(skb, features);
3265 EXPORT_SYMBOL(netif_skb_features);
3267 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3268 struct netdev_queue *txq, bool more)
3273 if (dev_nit_active(dev))
3274 dev_queue_xmit_nit(skb, dev);
3277 trace_net_dev_start_xmit(skb, dev);
3278 rc = netdev_start_xmit(skb, dev, txq, more);
3279 trace_net_dev_xmit(skb, rc, dev, len);
3284 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3285 struct netdev_queue *txq, int *ret)
3287 struct sk_buff *skb = first;
3288 int rc = NETDEV_TX_OK;
3291 struct sk_buff *next = skb->next;
3293 skb_mark_not_on_list(skb);
3294 rc = xmit_one(skb, dev, txq, next != NULL);
3295 if (unlikely(!dev_xmit_complete(rc))) {
3301 if (netif_tx_queue_stopped(txq) && skb) {
3302 rc = NETDEV_TX_BUSY;
3312 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3313 netdev_features_t features)
3315 if (skb_vlan_tag_present(skb) &&
3316 !vlan_hw_offload_capable(features, skb->vlan_proto))
3317 skb = __vlan_hwaccel_push_inside(skb);
3321 int skb_csum_hwoffload_help(struct sk_buff *skb,
3322 const netdev_features_t features)
3324 if (unlikely(skb->csum_not_inet))
3325 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3326 skb_crc32c_csum_help(skb);
3328 return !!(features & NETIF_F_CSUM_MASK) ? 0 : skb_checksum_help(skb);
3330 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3332 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3334 netdev_features_t features;
3336 features = netif_skb_features(skb);
3337 skb = validate_xmit_vlan(skb, features);
3341 skb = sk_validate_xmit_skb(skb, dev);
3345 if (netif_needs_gso(skb, features)) {
3346 struct sk_buff *segs;
3348 segs = skb_gso_segment(skb, features);
3356 if (skb_needs_linearize(skb, features) &&
3357 __skb_linearize(skb))
3360 /* If packet is not checksummed and device does not
3361 * support checksumming for this protocol, complete
3362 * checksumming here.
3364 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3365 if (skb->encapsulation)
3366 skb_set_inner_transport_header(skb,
3367 skb_checksum_start_offset(skb));
3369 skb_set_transport_header(skb,
3370 skb_checksum_start_offset(skb));
3371 if (skb_csum_hwoffload_help(skb, features))
3376 skb = validate_xmit_xfrm(skb, features, again);
3383 atomic_long_inc(&dev->tx_dropped);
3387 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3389 struct sk_buff *next, *head = NULL, *tail;
3391 for (; skb != NULL; skb = next) {
3393 skb_mark_not_on_list(skb);
3395 /* in case skb wont be segmented, point to itself */
3398 skb = validate_xmit_skb(skb, dev, again);
3406 /* If skb was segmented, skb->prev points to
3407 * the last segment. If not, it still contains skb.
3413 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3415 static void qdisc_pkt_len_init(struct sk_buff *skb)
3417 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3419 qdisc_skb_cb(skb)->pkt_len = skb->len;
3421 /* To get more precise estimation of bytes sent on wire,
3422 * we add to pkt_len the headers size of all segments
3424 if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3425 unsigned int hdr_len;
3426 u16 gso_segs = shinfo->gso_segs;
3428 /* mac layer + network layer */
3429 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3431 /* + transport layer */
3432 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3433 const struct tcphdr *th;
3434 struct tcphdr _tcphdr;
3436 th = skb_header_pointer(skb, skb_transport_offset(skb),
3437 sizeof(_tcphdr), &_tcphdr);
3439 hdr_len += __tcp_hdrlen(th);
3441 struct udphdr _udphdr;
3443 if (skb_header_pointer(skb, skb_transport_offset(skb),
3444 sizeof(_udphdr), &_udphdr))
3445 hdr_len += sizeof(struct udphdr);
3448 if (shinfo->gso_type & SKB_GSO_DODGY)
3449 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3452 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3456 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3457 struct net_device *dev,
3458 struct netdev_queue *txq)
3460 spinlock_t *root_lock = qdisc_lock(q);
3461 struct sk_buff *to_free = NULL;
3465 qdisc_calculate_pkt_len(skb, q);
3467 if (q->flags & TCQ_F_NOLOCK) {
3468 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3469 __qdisc_drop(skb, &to_free);
3472 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3476 if (unlikely(to_free))
3477 kfree_skb_list(to_free);
3482 * Heuristic to force contended enqueues to serialize on a
3483 * separate lock before trying to get qdisc main lock.
3484 * This permits qdisc->running owner to get the lock more
3485 * often and dequeue packets faster.
3487 contended = qdisc_is_running(q);
3488 if (unlikely(contended))
3489 spin_lock(&q->busylock);
3491 spin_lock(root_lock);
3492 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3493 __qdisc_drop(skb, &to_free);
3495 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3496 qdisc_run_begin(q)) {
3498 * This is a work-conserving queue; there are no old skbs
3499 * waiting to be sent out; and the qdisc is not running -
3500 * xmit the skb directly.
3503 qdisc_bstats_update(q, skb);
3505 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3506 if (unlikely(contended)) {
3507 spin_unlock(&q->busylock);
3514 rc = NET_XMIT_SUCCESS;
3516 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3517 if (qdisc_run_begin(q)) {
3518 if (unlikely(contended)) {
3519 spin_unlock(&q->busylock);
3526 spin_unlock(root_lock);
3527 if (unlikely(to_free))
3528 kfree_skb_list(to_free);
3529 if (unlikely(contended))
3530 spin_unlock(&q->busylock);
3534 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3535 static void skb_update_prio(struct sk_buff *skb)
3537 const struct netprio_map *map;
3538 const struct sock *sk;
3539 unsigned int prioidx;
3543 map = rcu_dereference_bh(skb->dev->priomap);
3546 sk = skb_to_full_sk(skb);
3550 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3552 if (prioidx < map->priomap_len)
3553 skb->priority = map->priomap[prioidx];
3556 #define skb_update_prio(skb)
3559 DEFINE_PER_CPU(int, xmit_recursion);
3560 EXPORT_SYMBOL(xmit_recursion);
3563 * dev_loopback_xmit - loop back @skb
3564 * @net: network namespace this loopback is happening in
3565 * @sk: sk needed to be a netfilter okfn
3566 * @skb: buffer to transmit
3568 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3570 skb_reset_mac_header(skb);
3571 __skb_pull(skb, skb_network_offset(skb));
3572 skb->pkt_type = PACKET_LOOPBACK;
3573 skb->ip_summed = CHECKSUM_UNNECESSARY;
3574 WARN_ON(!skb_dst(skb));
3579 EXPORT_SYMBOL(dev_loopback_xmit);
3581 #ifdef CONFIG_NET_EGRESS
3582 static struct sk_buff *
3583 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3585 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3586 struct tcf_result cl_res;
3591 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3592 mini_qdisc_bstats_cpu_update(miniq, skb);
3594 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
3596 case TC_ACT_RECLASSIFY:
3597 skb->tc_index = TC_H_MIN(cl_res.classid);
3600 mini_qdisc_qstats_cpu_drop(miniq);
3601 *ret = NET_XMIT_DROP;
3607 *ret = NET_XMIT_SUCCESS;
3610 case TC_ACT_REDIRECT:
3611 /* No need to push/pop skb's mac_header here on egress! */
3612 skb_do_redirect(skb);
3613 *ret = NET_XMIT_SUCCESS;
3621 #endif /* CONFIG_NET_EGRESS */
3624 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
3625 struct xps_dev_maps *dev_maps, unsigned int tci)
3627 struct xps_map *map;
3628 int queue_index = -1;
3632 tci += netdev_get_prio_tc_map(dev, skb->priority);
3635 map = rcu_dereference(dev_maps->attr_map[tci]);
3638 queue_index = map->queues[0];
3640 queue_index = map->queues[reciprocal_scale(
3641 skb_get_hash(skb), map->len)];
3642 if (unlikely(queue_index >= dev->real_num_tx_queues))
3649 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
3650 struct sk_buff *skb)
3653 struct xps_dev_maps *dev_maps;
3654 struct sock *sk = skb->sk;
3655 int queue_index = -1;
3657 if (!static_key_false(&xps_needed))
3661 if (!static_key_false(&xps_rxqs_needed))
3664 dev_maps = rcu_dereference(sb_dev->xps_rxqs_map);
3666 int tci = sk_rx_queue_get(sk);
3668 if (tci >= 0 && tci < dev->num_rx_queues)
3669 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
3674 if (queue_index < 0) {
3675 dev_maps = rcu_dereference(sb_dev->xps_cpus_map);
3677 unsigned int tci = skb->sender_cpu - 1;
3679 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
3691 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
3692 struct net_device *sb_dev)
3696 EXPORT_SYMBOL(dev_pick_tx_zero);
3698 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
3699 struct net_device *sb_dev)
3701 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
3703 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
3705 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
3706 struct net_device *sb_dev)
3708 struct sock *sk = skb->sk;
3709 int queue_index = sk_tx_queue_get(sk);
3711 sb_dev = sb_dev ? : dev;
3713 if (queue_index < 0 || skb->ooo_okay ||
3714 queue_index >= dev->real_num_tx_queues) {
3715 int new_index = get_xps_queue(dev, sb_dev, skb);
3718 new_index = skb_tx_hash(dev, sb_dev, skb);
3720 if (queue_index != new_index && sk &&
3722 rcu_access_pointer(sk->sk_dst_cache))
3723 sk_tx_queue_set(sk, new_index);
3725 queue_index = new_index;
3730 EXPORT_SYMBOL(netdev_pick_tx);
3732 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
3733 struct sk_buff *skb,
3734 struct net_device *sb_dev)
3736 int queue_index = 0;
3739 u32 sender_cpu = skb->sender_cpu - 1;
3741 if (sender_cpu >= (u32)NR_CPUS)
3742 skb->sender_cpu = raw_smp_processor_id() + 1;
3745 if (dev->real_num_tx_queues != 1) {
3746 const struct net_device_ops *ops = dev->netdev_ops;
3748 if (ops->ndo_select_queue)
3749 queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
3751 queue_index = netdev_pick_tx(dev, skb, sb_dev);
3753 queue_index = netdev_cap_txqueue(dev, queue_index);
3756 skb_set_queue_mapping(skb, queue_index);
3757 return netdev_get_tx_queue(dev, queue_index);
3761 * __dev_queue_xmit - transmit a buffer
3762 * @skb: buffer to transmit
3763 * @sb_dev: suboordinate device used for L2 forwarding offload
3765 * Queue a buffer for transmission to a network device. The caller must
3766 * have set the device and priority and built the buffer before calling
3767 * this function. The function can be called from an interrupt.
3769 * A negative errno code is returned on a failure. A success does not
3770 * guarantee the frame will be transmitted as it may be dropped due
3771 * to congestion or traffic shaping.
3773 * -----------------------------------------------------------------------------------
3774 * I notice this method can also return errors from the queue disciplines,
3775 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3778 * Regardless of the return value, the skb is consumed, so it is currently
3779 * difficult to retry a send to this method. (You can bump the ref count
3780 * before sending to hold a reference for retry if you are careful.)
3782 * When calling this method, interrupts MUST be enabled. This is because
3783 * the BH enable code must have IRQs enabled so that it will not deadlock.
3786 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
3788 struct net_device *dev = skb->dev;
3789 struct netdev_queue *txq;
3794 skb_reset_mac_header(skb);
3796 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3797 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3799 /* Disable soft irqs for various locks below. Also
3800 * stops preemption for RCU.
3804 skb_update_prio(skb);
3806 qdisc_pkt_len_init(skb);
3807 #ifdef CONFIG_NET_CLS_ACT
3808 skb->tc_at_ingress = 0;
3809 # ifdef CONFIG_NET_EGRESS
3810 if (static_branch_unlikely(&egress_needed_key)) {
3811 skb = sch_handle_egress(skb, &rc, dev);
3817 /* If device/qdisc don't need skb->dst, release it right now while
3818 * its hot in this cpu cache.
3820 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3825 txq = netdev_core_pick_tx(dev, skb, sb_dev);
3826 q = rcu_dereference_bh(txq->qdisc);
3828 trace_net_dev_queue(skb);
3830 rc = __dev_xmit_skb(skb, q, dev, txq);
3834 /* The device has no queue. Common case for software devices:
3835 * loopback, all the sorts of tunnels...
3837 * Really, it is unlikely that netif_tx_lock protection is necessary
3838 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
3840 * However, it is possible, that they rely on protection
3843 * Check this and shot the lock. It is not prone from deadlocks.
3844 *Either shot noqueue qdisc, it is even simpler 8)
3846 if (dev->flags & IFF_UP) {
3847 int cpu = smp_processor_id(); /* ok because BHs are off */
3849 if (txq->xmit_lock_owner != cpu) {
3850 if (unlikely(__this_cpu_read(xmit_recursion) >
3851 XMIT_RECURSION_LIMIT))
3852 goto recursion_alert;
3854 skb = validate_xmit_skb(skb, dev, &again);
3858 HARD_TX_LOCK(dev, txq, cpu);
3860 if (!netif_xmit_stopped(txq)) {
3861 __this_cpu_inc(xmit_recursion);
3862 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
3863 __this_cpu_dec(xmit_recursion);
3864 if (dev_xmit_complete(rc)) {
3865 HARD_TX_UNLOCK(dev, txq);
3869 HARD_TX_UNLOCK(dev, txq);
3870 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3873 /* Recursion is detected! It is possible,
3877 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3883 rcu_read_unlock_bh();
3885 atomic_long_inc(&dev->tx_dropped);
3886 kfree_skb_list(skb);
3889 rcu_read_unlock_bh();
3893 int dev_queue_xmit(struct sk_buff *skb)
3895 return __dev_queue_xmit(skb, NULL);
3897 EXPORT_SYMBOL(dev_queue_xmit);
3899 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev)
3901 return __dev_queue_xmit(skb, sb_dev);
3903 EXPORT_SYMBOL(dev_queue_xmit_accel);
3905 int dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
3907 struct net_device *dev = skb->dev;
3908 struct sk_buff *orig_skb = skb;
3909 struct netdev_queue *txq;
3910 int ret = NETDEV_TX_BUSY;
3913 if (unlikely(!netif_running(dev) ||
3914 !netif_carrier_ok(dev)))
3917 skb = validate_xmit_skb_list(skb, dev, &again);
3918 if (skb != orig_skb)
3921 skb_set_queue_mapping(skb, queue_id);
3922 txq = skb_get_tx_queue(dev, skb);
3926 HARD_TX_LOCK(dev, txq, smp_processor_id());
3927 if (!netif_xmit_frozen_or_drv_stopped(txq))
3928 ret = netdev_start_xmit(skb, dev, txq, false);
3929 HARD_TX_UNLOCK(dev, txq);
3933 if (!dev_xmit_complete(ret))
3938 atomic_long_inc(&dev->tx_dropped);
3939 kfree_skb_list(skb);
3940 return NET_XMIT_DROP;
3942 EXPORT_SYMBOL(dev_direct_xmit);
3944 /*************************************************************************
3946 *************************************************************************/
3948 int netdev_max_backlog __read_mostly = 1000;
3949 EXPORT_SYMBOL(netdev_max_backlog);
3951 int netdev_tstamp_prequeue __read_mostly = 1;
3952 int netdev_budget __read_mostly = 300;
3953 unsigned int __read_mostly netdev_budget_usecs = 2000;
3954 int weight_p __read_mostly = 64; /* old backlog weight */
3955 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
3956 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
3957 int dev_rx_weight __read_mostly = 64;
3958 int dev_tx_weight __read_mostly = 64;
3960 /* Called with irq disabled */
3961 static inline void ____napi_schedule(struct softnet_data *sd,
3962 struct napi_struct *napi)
3964 list_add_tail(&napi->poll_list, &sd->poll_list);
3965 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3970 /* One global table that all flow-based protocols share. */
3971 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3972 EXPORT_SYMBOL(rps_sock_flow_table);
3973 u32 rps_cpu_mask __read_mostly;
3974 EXPORT_SYMBOL(rps_cpu_mask);
3976 struct static_key rps_needed __read_mostly;
3977 EXPORT_SYMBOL(rps_needed);
3978 struct static_key rfs_needed __read_mostly;
3979 EXPORT_SYMBOL(rfs_needed);
3981 static struct rps_dev_flow *
3982 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3983 struct rps_dev_flow *rflow, u16 next_cpu)
3985 if (next_cpu < nr_cpu_ids) {
3986 #ifdef CONFIG_RFS_ACCEL
3987 struct netdev_rx_queue *rxqueue;
3988 struct rps_dev_flow_table *flow_table;
3989 struct rps_dev_flow *old_rflow;
3994 /* Should we steer this flow to a different hardware queue? */
3995 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3996 !(dev->features & NETIF_F_NTUPLE))
3998 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3999 if (rxq_index == skb_get_rx_queue(skb))
4002 rxqueue = dev->_rx + rxq_index;
4003 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4006 flow_id = skb_get_hash(skb) & flow_table->mask;
4007 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
4008 rxq_index, flow_id);
4012 rflow = &flow_table->flows[flow_id];
4014 if (old_rflow->filter == rflow->filter)
4015 old_rflow->filter = RPS_NO_FILTER;
4019 per_cpu(softnet_data, next_cpu).input_queue_head;
4022 rflow->cpu = next_cpu;
4027 * get_rps_cpu is called from netif_receive_skb and returns the target
4028 * CPU from the RPS map of the receiving queue for a given skb.
4029 * rcu_read_lock must be held on entry.
4031 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4032 struct rps_dev_flow **rflowp)
4034 const struct rps_sock_flow_table *sock_flow_table;
4035 struct netdev_rx_queue *rxqueue = dev->_rx;
4036 struct rps_dev_flow_table *flow_table;
4037 struct rps_map *map;
4042 if (skb_rx_queue_recorded(skb)) {
4043 u16 index = skb_get_rx_queue(skb);
4045 if (unlikely(index >= dev->real_num_rx_queues)) {
4046 WARN_ONCE(dev->real_num_rx_queues > 1,
4047 "%s received packet on queue %u, but number "
4048 "of RX queues is %u\n",
4049 dev->name, index, dev->real_num_rx_queues);
4055 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4057 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4058 map = rcu_dereference(rxqueue->rps_map);
4059 if (!flow_table && !map)
4062 skb_reset_network_header(skb);
4063 hash = skb_get_hash(skb);
4067 sock_flow_table = rcu_dereference(rps_sock_flow_table);
4068 if (flow_table && sock_flow_table) {
4069 struct rps_dev_flow *rflow;
4073 /* First check into global flow table if there is a match */
4074 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
4075 if ((ident ^ hash) & ~rps_cpu_mask)
4078 next_cpu = ident & rps_cpu_mask;
4080 /* OK, now we know there is a match,
4081 * we can look at the local (per receive queue) flow table
4083 rflow = &flow_table->flows[hash & flow_table->mask];
4087 * If the desired CPU (where last recvmsg was done) is
4088 * different from current CPU (one in the rx-queue flow
4089 * table entry), switch if one of the following holds:
4090 * - Current CPU is unset (>= nr_cpu_ids).
4091 * - Current CPU is offline.
4092 * - The current CPU's queue tail has advanced beyond the
4093 * last packet that was enqueued using this table entry.
4094 * This guarantees that all previous packets for the flow
4095 * have been dequeued, thus preserving in order delivery.
4097 if (unlikely(tcpu != next_cpu) &&
4098 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4099 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
4100 rflow->last_qtail)) >= 0)) {
4102 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4105 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4115 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4116 if (cpu_online(tcpu)) {
4126 #ifdef CONFIG_RFS_ACCEL
4129 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4130 * @dev: Device on which the filter was set
4131 * @rxq_index: RX queue index
4132 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4133 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4135 * Drivers that implement ndo_rx_flow_steer() should periodically call
4136 * this function for each installed filter and remove the filters for
4137 * which it returns %true.
4139 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4140 u32 flow_id, u16 filter_id)
4142 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4143 struct rps_dev_flow_table *flow_table;
4144 struct rps_dev_flow *rflow;
4149 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4150 if (flow_table && flow_id <= flow_table->mask) {
4151 rflow = &flow_table->flows[flow_id];
4152 cpu = READ_ONCE(rflow->cpu);
4153 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
4154 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
4155 rflow->last_qtail) <
4156 (int)(10 * flow_table->mask)))
4162 EXPORT_SYMBOL(rps_may_expire_flow);
4164 #endif /* CONFIG_RFS_ACCEL */
4166 /* Called from hardirq (IPI) context */
4167 static void rps_trigger_softirq(void *data)
4169 struct softnet_data *sd = data;
4171 ____napi_schedule(sd, &sd->backlog);
4175 #endif /* CONFIG_RPS */
4178 * Check if this softnet_data structure is another cpu one
4179 * If yes, queue it to our IPI list and return 1
4182 static int rps_ipi_queued(struct softnet_data *sd)
4185 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4188 sd->rps_ipi_next = mysd->rps_ipi_list;
4189 mysd->rps_ipi_list = sd;
4191 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4194 #endif /* CONFIG_RPS */
4198 #ifdef CONFIG_NET_FLOW_LIMIT
4199 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4202 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4204 #ifdef CONFIG_NET_FLOW_LIMIT
4205 struct sd_flow_limit *fl;
4206 struct softnet_data *sd;
4207 unsigned int old_flow, new_flow;
4209 if (qlen < (netdev_max_backlog >> 1))
4212 sd = this_cpu_ptr(&softnet_data);
4215 fl = rcu_dereference(sd->flow_limit);
4217 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4218 old_flow = fl->history[fl->history_head];
4219 fl->history[fl->history_head] = new_flow;
4222 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4224 if (likely(fl->buckets[old_flow]))
4225 fl->buckets[old_flow]--;
4227 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4239 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4240 * queue (may be a remote CPU queue).
4242 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4243 unsigned int *qtail)
4245 struct softnet_data *sd;
4246 unsigned long flags;
4249 sd = &per_cpu(softnet_data, cpu);
4251 local_irq_save(flags);
4254 if (!netif_running(skb->dev))
4256 qlen = skb_queue_len(&sd->input_pkt_queue);
4257 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
4260 __skb_queue_tail(&sd->input_pkt_queue, skb);
4261 input_queue_tail_incr_save(sd, qtail);
4263 local_irq_restore(flags);
4264 return NET_RX_SUCCESS;
4267 /* Schedule NAPI for backlog device
4268 * We can use non atomic operation since we own the queue lock
4270 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
4271 if (!rps_ipi_queued(sd))
4272 ____napi_schedule(sd, &sd->backlog);
4281 local_irq_restore(flags);
4283 atomic_long_inc(&skb->dev->rx_dropped);
4288 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4290 struct net_device *dev = skb->dev;
4291 struct netdev_rx_queue *rxqueue;
4295 if (skb_rx_queue_recorded(skb)) {
4296 u16 index = skb_get_rx_queue(skb);
4298 if (unlikely(index >= dev->real_num_rx_queues)) {
4299 WARN_ONCE(dev->real_num_rx_queues > 1,
4300 "%s received packet on queue %u, but number "
4301 "of RX queues is %u\n",
4302 dev->name, index, dev->real_num_rx_queues);
4304 return rxqueue; /* Return first rxqueue */
4311 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4312 struct xdp_buff *xdp,
4313 struct bpf_prog *xdp_prog)
4315 struct netdev_rx_queue *rxqueue;
4316 void *orig_data, *orig_data_end;
4317 u32 metalen, act = XDP_DROP;
4318 __be16 orig_eth_type;
4324 /* Reinjected packets coming from act_mirred or similar should
4325 * not get XDP generic processing.
4327 if (skb_cloned(skb) || skb_is_tc_redirected(skb))
4330 /* XDP packets must be linear and must have sufficient headroom
4331 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4332 * native XDP provides, thus we need to do it here as well.
4334 if (skb_is_nonlinear(skb) ||
4335 skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4336 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4337 int troom = skb->tail + skb->data_len - skb->end;
4339 /* In case we have to go down the path and also linearize,
4340 * then lets do the pskb_expand_head() work just once here.
4342 if (pskb_expand_head(skb,
4343 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4344 troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4346 if (skb_linearize(skb))
4350 /* The XDP program wants to see the packet starting at the MAC
4353 mac_len = skb->data - skb_mac_header(skb);
4354 hlen = skb_headlen(skb) + mac_len;
4355 xdp->data = skb->data - mac_len;
4356 xdp->data_meta = xdp->data;
4357 xdp->data_end = xdp->data + hlen;
4358 xdp->data_hard_start = skb->data - skb_headroom(skb);
4359 orig_data_end = xdp->data_end;
4360 orig_data = xdp->data;
4361 eth = (struct ethhdr *)xdp->data;
4362 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4363 orig_eth_type = eth->h_proto;
4365 rxqueue = netif_get_rxqueue(skb);
4366 xdp->rxq = &rxqueue->xdp_rxq;
4368 act = bpf_prog_run_xdp(xdp_prog, xdp);
4370 off = xdp->data - orig_data;
4372 __skb_pull(skb, off);
4374 __skb_push(skb, -off);
4375 skb->mac_header += off;
4377 /* check if bpf_xdp_adjust_tail was used. it can only "shrink"
4380 off = orig_data_end - xdp->data_end;
4382 skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4387 /* check if XDP changed eth hdr such SKB needs update */
4388 eth = (struct ethhdr *)xdp->data;
4389 if ((orig_eth_type != eth->h_proto) ||
4390 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4391 __skb_push(skb, ETH_HLEN);
4392 skb->protocol = eth_type_trans(skb, skb->dev);
4398 __skb_push(skb, mac_len);
4401 metalen = xdp->data - xdp->data_meta;
4403 skb_metadata_set(skb, metalen);
4406 bpf_warn_invalid_xdp_action(act);
4409 trace_xdp_exception(skb->dev, xdp_prog, act);
4420 /* When doing generic XDP we have to bypass the qdisc layer and the
4421 * network taps in order to match in-driver-XDP behavior.
4423 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4425 struct net_device *dev = skb->dev;
4426 struct netdev_queue *txq;
4427 bool free_skb = true;
4430 txq = netdev_core_pick_tx(dev, skb, NULL);
4431 cpu = smp_processor_id();
4432 HARD_TX_LOCK(dev, txq, cpu);
4433 if (!netif_xmit_stopped(txq)) {
4434 rc = netdev_start_xmit(skb, dev, txq, 0);
4435 if (dev_xmit_complete(rc))
4438 HARD_TX_UNLOCK(dev, txq);
4440 trace_xdp_exception(dev, xdp_prog, XDP_TX);
4444 EXPORT_SYMBOL_GPL(generic_xdp_tx);
4446 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4448 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4451 struct xdp_buff xdp;
4455 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4456 if (act != XDP_PASS) {
4459 err = xdp_do_generic_redirect(skb->dev, skb,
4465 generic_xdp_tx(skb, xdp_prog);
4476 EXPORT_SYMBOL_GPL(do_xdp_generic);
4478 static int netif_rx_internal(struct sk_buff *skb)
4482 net_timestamp_check(netdev_tstamp_prequeue, skb);
4484 trace_netif_rx(skb);
4486 if (static_branch_unlikely(&generic_xdp_needed_key)) {
4491 ret = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
4495 /* Consider XDP consuming the packet a success from
4496 * the netdev point of view we do not want to count
4499 if (ret != XDP_PASS)
4500 return NET_RX_SUCCESS;
4504 if (static_key_false(&rps_needed)) {
4505 struct rps_dev_flow voidflow, *rflow = &voidflow;
4511 cpu = get_rps_cpu(skb->dev, skb, &rflow);
4513 cpu = smp_processor_id();
4515 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4524 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
4531 * netif_rx - post buffer to the network code
4532 * @skb: buffer to post
4534 * This function receives a packet from a device driver and queues it for
4535 * the upper (protocol) levels to process. It always succeeds. The buffer
4536 * may be dropped during processing for congestion control or by the
4540 * NET_RX_SUCCESS (no congestion)
4541 * NET_RX_DROP (packet was dropped)
4545 int netif_rx(struct sk_buff *skb)
4549 trace_netif_rx_entry(skb);
4551 ret = netif_rx_internal(skb);
4552 trace_netif_rx_exit(ret);
4556 EXPORT_SYMBOL(netif_rx);
4558 int netif_rx_ni(struct sk_buff *skb)
4562 trace_netif_rx_ni_entry(skb);
4565 err = netif_rx_internal(skb);
4566 if (local_softirq_pending())
4569 trace_netif_rx_ni_exit(err);
4573 EXPORT_SYMBOL(netif_rx_ni);
4575 static __latent_entropy void net_tx_action(struct softirq_action *h)
4577 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4579 if (sd->completion_queue) {
4580 struct sk_buff *clist;
4582 local_irq_disable();
4583 clist = sd->completion_queue;
4584 sd->completion_queue = NULL;
4588 struct sk_buff *skb = clist;
4590 clist = clist->next;
4592 WARN_ON(refcount_read(&skb->users));
4593 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
4594 trace_consume_skb(skb);
4596 trace_kfree_skb(skb, net_tx_action);
4598 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
4601 __kfree_skb_defer(skb);
4604 __kfree_skb_flush();
4607 if (sd->output_queue) {
4610 local_irq_disable();
4611 head = sd->output_queue;
4612 sd->output_queue = NULL;
4613 sd->output_queue_tailp = &sd->output_queue;
4617 struct Qdisc *q = head;
4618 spinlock_t *root_lock = NULL;
4620 head = head->next_sched;
4622 if (!(q->flags & TCQ_F_NOLOCK)) {
4623 root_lock = qdisc_lock(q);
4624 spin_lock(root_lock);
4626 /* We need to make sure head->next_sched is read
4627 * before clearing __QDISC_STATE_SCHED
4629 smp_mb__before_atomic();
4630 clear_bit(__QDISC_STATE_SCHED, &q->state);
4633 spin_unlock(root_lock);
4637 xfrm_dev_backlog(sd);
4640 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
4641 /* This hook is defined here for ATM LANE */
4642 int (*br_fdb_test_addr_hook)(struct net_device *dev,
4643 unsigned char *addr) __read_mostly;
4644 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
4647 static inline struct sk_buff *
4648 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
4649 struct net_device *orig_dev)
4651 #ifdef CONFIG_NET_CLS_ACT
4652 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
4653 struct tcf_result cl_res;
4655 /* If there's at least one ingress present somewhere (so
4656 * we get here via enabled static key), remaining devices
4657 * that are not configured with an ingress qdisc will bail
4664 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4668 qdisc_skb_cb(skb)->pkt_len = skb->len;
4669 skb->tc_at_ingress = 1;
4670 mini_qdisc_bstats_cpu_update(miniq, skb);
4672 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
4674 case TC_ACT_RECLASSIFY:
4675 skb->tc_index = TC_H_MIN(cl_res.classid);
4678 mini_qdisc_qstats_cpu_drop(miniq);
4686 case TC_ACT_REDIRECT:
4687 /* skb_mac_header check was done by cls/act_bpf, so
4688 * we can safely push the L2 header back before
4689 * redirecting to another netdev
4691 __skb_push(skb, skb->mac_len);
4692 skb_do_redirect(skb);
4694 case TC_ACT_REINSERT:
4695 /* this does not scrub the packet, and updates stats on error */
4696 skb_tc_reinsert(skb, &cl_res);
4701 #endif /* CONFIG_NET_CLS_ACT */
4706 * netdev_is_rx_handler_busy - check if receive handler is registered
4707 * @dev: device to check
4709 * Check if a receive handler is already registered for a given device.
4710 * Return true if there one.
4712 * The caller must hold the rtnl_mutex.
4714 bool netdev_is_rx_handler_busy(struct net_device *dev)
4717 return dev && rtnl_dereference(dev->rx_handler);
4719 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
4722 * netdev_rx_handler_register - register receive handler
4723 * @dev: device to register a handler for
4724 * @rx_handler: receive handler to register
4725 * @rx_handler_data: data pointer that is used by rx handler
4727 * Register a receive handler for a device. This handler will then be
4728 * called from __netif_receive_skb. A negative errno code is returned
4731 * The caller must hold the rtnl_mutex.
4733 * For a general description of rx_handler, see enum rx_handler_result.
4735 int netdev_rx_handler_register(struct net_device *dev,
4736 rx_handler_func_t *rx_handler,
4737 void *rx_handler_data)
4739 if (netdev_is_rx_handler_busy(dev))
4742 if (dev->priv_flags & IFF_NO_RX_HANDLER)
4745 /* Note: rx_handler_data must be set before rx_handler */
4746 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
4747 rcu_assign_pointer(dev->rx_handler, rx_handler);
4751 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
4754 * netdev_rx_handler_unregister - unregister receive handler
4755 * @dev: device to unregister a handler from
4757 * Unregister a receive handler from a device.
4759 * The caller must hold the rtnl_mutex.
4761 void netdev_rx_handler_unregister(struct net_device *dev)
4765 RCU_INIT_POINTER(dev->rx_handler, NULL);
4766 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
4767 * section has a guarantee to see a non NULL rx_handler_data
4771 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
4773 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
4776 * Limit the use of PFMEMALLOC reserves to those protocols that implement
4777 * the special handling of PFMEMALLOC skbs.
4779 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
4781 switch (skb->protocol) {
4782 case htons(ETH_P_ARP):
4783 case htons(ETH_P_IP):
4784 case htons(ETH_P_IPV6):
4785 case htons(ETH_P_8021Q):
4786 case htons(ETH_P_8021AD):
4793 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
4794 int *ret, struct net_device *orig_dev)
4796 #ifdef CONFIG_NETFILTER_INGRESS
4797 if (nf_hook_ingress_active(skb)) {
4801 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4806 ingress_retval = nf_hook_ingress(skb);
4808 return ingress_retval;
4810 #endif /* CONFIG_NETFILTER_INGRESS */
4814 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc,
4815 struct packet_type **ppt_prev)
4817 struct packet_type *ptype, *pt_prev;
4818 rx_handler_func_t *rx_handler;
4819 struct net_device *orig_dev;
4820 bool deliver_exact = false;
4821 int ret = NET_RX_DROP;
4824 net_timestamp_check(!netdev_tstamp_prequeue, skb);
4826 trace_netif_receive_skb(skb);
4828 orig_dev = skb->dev;
4830 skb_reset_network_header(skb);
4831 if (!skb_transport_header_was_set(skb))
4832 skb_reset_transport_header(skb);
4833 skb_reset_mac_len(skb);
4838 skb->skb_iif = skb->dev->ifindex;
4840 __this_cpu_inc(softnet_data.processed);
4842 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
4843 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
4844 skb = skb_vlan_untag(skb);
4849 if (skb_skip_tc_classify(skb))
4855 list_for_each_entry_rcu(ptype, &ptype_all, list) {
4857 ret = deliver_skb(skb, pt_prev, orig_dev);
4861 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
4863 ret = deliver_skb(skb, pt_prev, orig_dev);
4868 #ifdef CONFIG_NET_INGRESS
4869 if (static_branch_unlikely(&ingress_needed_key)) {
4870 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev);
4874 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
4880 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
4883 if (skb_vlan_tag_present(skb)) {
4885 ret = deliver_skb(skb, pt_prev, orig_dev);
4888 if (vlan_do_receive(&skb))
4890 else if (unlikely(!skb))
4894 rx_handler = rcu_dereference(skb->dev->rx_handler);
4897 ret = deliver_skb(skb, pt_prev, orig_dev);
4900 switch (rx_handler(&skb)) {
4901 case RX_HANDLER_CONSUMED:
4902 ret = NET_RX_SUCCESS;
4904 case RX_HANDLER_ANOTHER:
4906 case RX_HANDLER_EXACT:
4907 deliver_exact = true;
4908 case RX_HANDLER_PASS:
4915 if (unlikely(skb_vlan_tag_present(skb))) {
4916 if (skb_vlan_tag_get_id(skb))
4917 skb->pkt_type = PACKET_OTHERHOST;
4918 /* Note: we might in the future use prio bits
4919 * and set skb->priority like in vlan_do_receive()
4920 * For the time being, just ignore Priority Code Point
4922 __vlan_hwaccel_clear_tag(skb);
4925 type = skb->protocol;
4927 /* deliver only exact match when indicated */
4928 if (likely(!deliver_exact)) {
4929 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4930 &ptype_base[ntohs(type) &
4934 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4935 &orig_dev->ptype_specific);
4937 if (unlikely(skb->dev != orig_dev)) {
4938 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4939 &skb->dev->ptype_specific);
4943 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
4945 *ppt_prev = pt_prev;
4949 atomic_long_inc(&skb->dev->rx_dropped);
4951 atomic_long_inc(&skb->dev->rx_nohandler);
4953 /* Jamal, now you will not able to escape explaining
4954 * me how you were going to use this. :-)
4963 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
4965 struct net_device *orig_dev = skb->dev;
4966 struct packet_type *pt_prev = NULL;
4969 ret = __netif_receive_skb_core(skb, pfmemalloc, &pt_prev);
4971 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
4976 * netif_receive_skb_core - special purpose version of netif_receive_skb
4977 * @skb: buffer to process
4979 * More direct receive version of netif_receive_skb(). It should
4980 * only be used by callers that have a need to skip RPS and Generic XDP.
4981 * Caller must also take care of handling if (page_is_)pfmemalloc.
4983 * This function may only be called from softirq context and interrupts
4984 * should be enabled.
4986 * Return values (usually ignored):
4987 * NET_RX_SUCCESS: no congestion
4988 * NET_RX_DROP: packet was dropped
4990 int netif_receive_skb_core(struct sk_buff *skb)
4995 ret = __netif_receive_skb_one_core(skb, false);
5000 EXPORT_SYMBOL(netif_receive_skb_core);
5002 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5003 struct packet_type *pt_prev,
5004 struct net_device *orig_dev)
5006 struct sk_buff *skb, *next;
5010 if (list_empty(head))
5012 if (pt_prev->list_func != NULL)
5013 pt_prev->list_func(head, pt_prev, orig_dev);
5015 list_for_each_entry_safe(skb, next, head, list)
5016 pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5019 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5021 /* Fast-path assumptions:
5022 * - There is no RX handler.
5023 * - Only one packet_type matches.
5024 * If either of these fails, we will end up doing some per-packet
5025 * processing in-line, then handling the 'last ptype' for the whole
5026 * sublist. This can't cause out-of-order delivery to any single ptype,
5027 * because the 'last ptype' must be constant across the sublist, and all
5028 * other ptypes are handled per-packet.
5030 /* Current (common) ptype of sublist */
5031 struct packet_type *pt_curr = NULL;
5032 /* Current (common) orig_dev of sublist */
5033 struct net_device *od_curr = NULL;
5034 struct list_head sublist;
5035 struct sk_buff *skb, *next;
5037 INIT_LIST_HEAD(&sublist);
5038 list_for_each_entry_safe(skb, next, head, list) {
5039 struct net_device *orig_dev = skb->dev;
5040 struct packet_type *pt_prev = NULL;
5042 skb_list_del_init(skb);
5043 __netif_receive_skb_core(skb, pfmemalloc, &pt_prev);
5046 if (pt_curr != pt_prev || od_curr != orig_dev) {
5047 /* dispatch old sublist */
5048 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5049 /* start new sublist */
5050 INIT_LIST_HEAD(&sublist);
5054 list_add_tail(&skb->list, &sublist);
5057 /* dispatch final sublist */
5058 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5061 static int __netif_receive_skb(struct sk_buff *skb)
5065 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5066 unsigned int noreclaim_flag;
5069 * PFMEMALLOC skbs are special, they should
5070 * - be delivered to SOCK_MEMALLOC sockets only
5071 * - stay away from userspace
5072 * - have bounded memory usage
5074 * Use PF_MEMALLOC as this saves us from propagating the allocation
5075 * context down to all allocation sites.
5077 noreclaim_flag = memalloc_noreclaim_save();
5078 ret = __netif_receive_skb_one_core(skb, true);
5079 memalloc_noreclaim_restore(noreclaim_flag);
5081 ret = __netif_receive_skb_one_core(skb, false);
5086 static void __netif_receive_skb_list(struct list_head *head)
5088 unsigned long noreclaim_flag = 0;
5089 struct sk_buff *skb, *next;
5090 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5092 list_for_each_entry_safe(skb, next, head, list) {
5093 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5094 struct list_head sublist;
5096 /* Handle the previous sublist */
5097 list_cut_before(&sublist, head, &skb->list);
5098 if (!list_empty(&sublist))
5099 __netif_receive_skb_list_core(&sublist, pfmemalloc);
5100 pfmemalloc = !pfmemalloc;
5101 /* See comments in __netif_receive_skb */
5103 noreclaim_flag = memalloc_noreclaim_save();
5105 memalloc_noreclaim_restore(noreclaim_flag);
5108 /* Handle the remaining sublist */
5109 if (!list_empty(head))
5110 __netif_receive_skb_list_core(head, pfmemalloc);
5111 /* Restore pflags */
5113 memalloc_noreclaim_restore(noreclaim_flag);
5116 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5118 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5119 struct bpf_prog *new = xdp->prog;
5122 switch (xdp->command) {
5123 case XDP_SETUP_PROG:
5124 rcu_assign_pointer(dev->xdp_prog, new);
5129 static_branch_dec(&generic_xdp_needed_key);
5130 } else if (new && !old) {
5131 static_branch_inc(&generic_xdp_needed_key);
5132 dev_disable_lro(dev);
5133 dev_disable_gro_hw(dev);
5137 case XDP_QUERY_PROG:
5138 xdp->prog_id = old ? old->aux->id : 0;
5149 static int netif_receive_skb_internal(struct sk_buff *skb)
5153 net_timestamp_check(netdev_tstamp_prequeue, skb);
5155 if (skb_defer_rx_timestamp(skb))
5156 return NET_RX_SUCCESS;
5158 if (static_branch_unlikely(&generic_xdp_needed_key)) {
5163 ret = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
5167 if (ret != XDP_PASS)
5173 if (static_key_false(&rps_needed)) {
5174 struct rps_dev_flow voidflow, *rflow = &voidflow;
5175 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5178 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5184 ret = __netif_receive_skb(skb);
5189 static void netif_receive_skb_list_internal(struct list_head *head)
5191 struct bpf_prog *xdp_prog = NULL;
5192 struct sk_buff *skb, *next;
5193 struct list_head sublist;
5195 INIT_LIST_HEAD(&sublist);
5196 list_for_each_entry_safe(skb, next, head, list) {
5197 net_timestamp_check(netdev_tstamp_prequeue, skb);
5198 skb_list_del_init(skb);
5199 if (!skb_defer_rx_timestamp(skb))
5200 list_add_tail(&skb->list, &sublist);
5202 list_splice_init(&sublist, head);
5204 if (static_branch_unlikely(&generic_xdp_needed_key)) {
5207 list_for_each_entry_safe(skb, next, head, list) {
5208 xdp_prog = rcu_dereference(skb->dev->xdp_prog);
5209 skb_list_del_init(skb);
5210 if (do_xdp_generic(xdp_prog, skb) == XDP_PASS)
5211 list_add_tail(&skb->list, &sublist);
5215 /* Put passed packets back on main list */
5216 list_splice_init(&sublist, head);
5221 if (static_key_false(&rps_needed)) {
5222 list_for_each_entry_safe(skb, next, head, list) {
5223 struct rps_dev_flow voidflow, *rflow = &voidflow;
5224 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5227 /* Will be handled, remove from list */
5228 skb_list_del_init(skb);
5229 enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5234 __netif_receive_skb_list(head);
5239 * netif_receive_skb - process receive buffer from network
5240 * @skb: buffer to process
5242 * netif_receive_skb() is the main receive data processing function.
5243 * It always succeeds. The buffer may be dropped during processing
5244 * for congestion control or by the protocol layers.
5246 * This function may only be called from softirq context and interrupts
5247 * should be enabled.
5249 * Return values (usually ignored):
5250 * NET_RX_SUCCESS: no congestion
5251 * NET_RX_DROP: packet was dropped
5253 int netif_receive_skb(struct sk_buff *skb)
5257 trace_netif_receive_skb_entry(skb);
5259 ret = netif_receive_skb_internal(skb);
5260 trace_netif_receive_skb_exit(ret);
5264 EXPORT_SYMBOL(netif_receive_skb);
5267 * netif_receive_skb_list - process many receive buffers from network
5268 * @head: list of skbs to process.
5270 * Since return value of netif_receive_skb() is normally ignored, and
5271 * wouldn't be meaningful for a list, this function returns void.
5273 * This function may only be called from softirq context and interrupts
5274 * should be enabled.
5276 void netif_receive_skb_list(struct list_head *head)
5278 struct sk_buff *skb;
5280 if (list_empty(head))
5282 if (trace_netif_receive_skb_list_entry_enabled()) {
5283 list_for_each_entry(skb, head, list)
5284 trace_netif_receive_skb_list_entry(skb);
5286 netif_receive_skb_list_internal(head);
5287 trace_netif_receive_skb_list_exit(0);
5289 EXPORT_SYMBOL(netif_receive_skb_list);
5291 DEFINE_PER_CPU(struct work_struct, flush_works);
5293 /* Network device is going away, flush any packets still pending */
5294 static void flush_backlog(struct work_struct *work)
5296 struct sk_buff *skb, *tmp;
5297 struct softnet_data *sd;
5300 sd = this_cpu_ptr(&softnet_data);
5302 local_irq_disable();
5304 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5305 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5306 __skb_unlink(skb, &sd->input_pkt_queue);
5308 input_queue_head_incr(sd);
5314 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5315 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5316 __skb_unlink(skb, &sd->process_queue);
5318 input_queue_head_incr(sd);
5324 static void flush_all_backlogs(void)
5330 for_each_online_cpu(cpu)
5331 queue_work_on(cpu, system_highpri_wq,
5332 per_cpu_ptr(&flush_works, cpu));
5334 for_each_online_cpu(cpu)
5335 flush_work(per_cpu_ptr(&flush_works, cpu));
5340 INDIRECT_CALLABLE_DECLARE(int inet_gro_complete(struct sk_buff *, int));
5341 INDIRECT_CALLABLE_DECLARE(int ipv6_gro_complete(struct sk_buff *, int));
5342 static int napi_gro_complete(struct sk_buff *skb)
5344 struct packet_offload *ptype;
5345 __be16 type = skb->protocol;
5346 struct list_head *head = &offload_base;
5349 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
5351 if (NAPI_GRO_CB(skb)->count == 1) {
5352 skb_shinfo(skb)->gso_size = 0;
5357 list_for_each_entry_rcu(ptype, head, list) {
5358 if (ptype->type != type || !ptype->callbacks.gro_complete)
5361 err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete,
5362 ipv6_gro_complete, inet_gro_complete,
5369 WARN_ON(&ptype->list == head);
5371 return NET_RX_SUCCESS;
5375 return netif_receive_skb_internal(skb);
5378 static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index,
5381 struct list_head *head = &napi->gro_hash[index].list;
5382 struct sk_buff *skb, *p;
5384 list_for_each_entry_safe_reverse(skb, p, head, list) {
5385 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
5387 skb_list_del_init(skb);
5388 napi_gro_complete(skb);
5389 napi->gro_hash[index].count--;
5392 if (!napi->gro_hash[index].count)
5393 __clear_bit(index, &napi->gro_bitmask);
5396 /* napi->gro_hash[].list contains packets ordered by age.
5397 * youngest packets at the head of it.
5398 * Complete skbs in reverse order to reduce latencies.
5400 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
5402 unsigned long bitmask = napi->gro_bitmask;
5403 unsigned int i, base = ~0U;
5405 while ((i = ffs(bitmask)) != 0) {
5408 __napi_gro_flush_chain(napi, base, flush_old);
5411 EXPORT_SYMBOL(napi_gro_flush);
5413 static struct list_head *gro_list_prepare(struct napi_struct *napi,
5414 struct sk_buff *skb)
5416 unsigned int maclen = skb->dev->hard_header_len;
5417 u32 hash = skb_get_hash_raw(skb);
5418 struct list_head *head;
5421 head = &napi->gro_hash[hash & (GRO_HASH_BUCKETS - 1)].list;
5422 list_for_each_entry(p, head, list) {
5423 unsigned long diffs;
5425 NAPI_GRO_CB(p)->flush = 0;
5427 if (hash != skb_get_hash_raw(p)) {
5428 NAPI_GRO_CB(p)->same_flow = 0;
5432 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
5433 diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb);
5434 if (skb_vlan_tag_present(p))
5435 diffs |= p->vlan_tci ^ skb->vlan_tci;
5436 diffs |= skb_metadata_dst_cmp(p, skb);
5437 diffs |= skb_metadata_differs(p, skb);
5438 if (maclen == ETH_HLEN)
5439 diffs |= compare_ether_header(skb_mac_header(p),
5440 skb_mac_header(skb));
5442 diffs = memcmp(skb_mac_header(p),
5443 skb_mac_header(skb),
5445 NAPI_GRO_CB(p)->same_flow = !diffs;
5451 static void skb_gro_reset_offset(struct sk_buff *skb)
5453 const struct skb_shared_info *pinfo = skb_shinfo(skb);
5454 const skb_frag_t *frag0 = &pinfo->frags[0];
5456 NAPI_GRO_CB(skb)->data_offset = 0;
5457 NAPI_GRO_CB(skb)->frag0 = NULL;
5458 NAPI_GRO_CB(skb)->frag0_len = 0;
5460 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
5462 !PageHighMem(skb_frag_page(frag0))) {
5463 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
5464 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
5465 skb_frag_size(frag0),
5466 skb->end - skb->tail);
5470 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
5472 struct skb_shared_info *pinfo = skb_shinfo(skb);
5474 BUG_ON(skb->end - skb->tail < grow);
5476 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
5478 skb->data_len -= grow;
5481 pinfo->frags[0].page_offset += grow;
5482 skb_frag_size_sub(&pinfo->frags[0], grow);
5484 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
5485 skb_frag_unref(skb, 0);
5486 memmove(pinfo->frags, pinfo->frags + 1,
5487 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
5491 static void gro_flush_oldest(struct list_head *head)
5493 struct sk_buff *oldest;
5495 oldest = list_last_entry(head, struct sk_buff, list);
5497 /* We are called with head length >= MAX_GRO_SKBS, so this is
5500 if (WARN_ON_ONCE(!oldest))
5503 /* Do not adjust napi->gro_hash[].count, caller is adding a new
5506 skb_list_del_init(oldest);
5507 napi_gro_complete(oldest);
5510 INDIRECT_CALLABLE_DECLARE(struct sk_buff *inet_gro_receive(struct list_head *,
5512 INDIRECT_CALLABLE_DECLARE(struct sk_buff *ipv6_gro_receive(struct list_head *,
5514 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
5516 u32 hash = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1);
5517 struct list_head *head = &offload_base;
5518 struct packet_offload *ptype;
5519 __be16 type = skb->protocol;
5520 struct list_head *gro_head;
5521 struct sk_buff *pp = NULL;
5522 enum gro_result ret;
5526 if (netif_elide_gro(skb->dev))
5529 gro_head = gro_list_prepare(napi, skb);
5532 list_for_each_entry_rcu(ptype, head, list) {
5533 if (ptype->type != type || !ptype->callbacks.gro_receive)
5536 skb_set_network_header(skb, skb_gro_offset(skb));
5537 skb_reset_mac_len(skb);
5538 NAPI_GRO_CB(skb)->same_flow = 0;
5539 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb);
5540 NAPI_GRO_CB(skb)->free = 0;
5541 NAPI_GRO_CB(skb)->encap_mark = 0;
5542 NAPI_GRO_CB(skb)->recursion_counter = 0;
5543 NAPI_GRO_CB(skb)->is_fou = 0;
5544 NAPI_GRO_CB(skb)->is_atomic = 1;
5545 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
5547 /* Setup for GRO checksum validation */
5548 switch (skb->ip_summed) {
5549 case CHECKSUM_COMPLETE:
5550 NAPI_GRO_CB(skb)->csum = skb->csum;
5551 NAPI_GRO_CB(skb)->csum_valid = 1;
5552 NAPI_GRO_CB(skb)->csum_cnt = 0;
5554 case CHECKSUM_UNNECESSARY:
5555 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
5556 NAPI_GRO_CB(skb)->csum_valid = 0;
5559 NAPI_GRO_CB(skb)->csum_cnt = 0;
5560 NAPI_GRO_CB(skb)->csum_valid = 0;
5563 pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive,
5564 ipv6_gro_receive, inet_gro_receive,
5570 if (&ptype->list == head)
5573 if (IS_ERR(pp) && PTR_ERR(pp) == -EINPROGRESS) {
5578 same_flow = NAPI_GRO_CB(skb)->same_flow;
5579 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
5582 skb_list_del_init(pp);
5583 napi_gro_complete(pp);
5584 napi->gro_hash[hash].count--;
5590 if (NAPI_GRO_CB(skb)->flush)
5593 if (unlikely(napi->gro_hash[hash].count >= MAX_GRO_SKBS)) {
5594 gro_flush_oldest(gro_head);
5596 napi->gro_hash[hash].count++;
5598 NAPI_GRO_CB(skb)->count = 1;
5599 NAPI_GRO_CB(skb)->age = jiffies;
5600 NAPI_GRO_CB(skb)->last = skb;
5601 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
5602 list_add(&skb->list, gro_head);
5606 grow = skb_gro_offset(skb) - skb_headlen(skb);
5608 gro_pull_from_frag0(skb, grow);
5610 if (napi->gro_hash[hash].count) {
5611 if (!test_bit(hash, &napi->gro_bitmask))
5612 __set_bit(hash, &napi->gro_bitmask);
5613 } else if (test_bit(hash, &napi->gro_bitmask)) {
5614 __clear_bit(hash, &napi->gro_bitmask);
5624 struct packet_offload *gro_find_receive_by_type(__be16 type)
5626 struct list_head *offload_head = &offload_base;
5627 struct packet_offload *ptype;
5629 list_for_each_entry_rcu(ptype, offload_head, list) {
5630 if (ptype->type != type || !ptype->callbacks.gro_receive)
5636 EXPORT_SYMBOL(gro_find_receive_by_type);
5638 struct packet_offload *gro_find_complete_by_type(__be16 type)
5640 struct list_head *offload_head = &offload_base;
5641 struct packet_offload *ptype;
5643 list_for_each_entry_rcu(ptype, offload_head, list) {
5644 if (ptype->type != type || !ptype->callbacks.gro_complete)
5650 EXPORT_SYMBOL(gro_find_complete_by_type);
5652 static void napi_skb_free_stolen_head(struct sk_buff *skb)
5656 kmem_cache_free(skbuff_head_cache, skb);
5659 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
5663 if (netif_receive_skb_internal(skb))
5671 case GRO_MERGED_FREE:
5672 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
5673 napi_skb_free_stolen_head(skb);
5687 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
5691 skb_mark_napi_id(skb, napi);
5692 trace_napi_gro_receive_entry(skb);
5694 skb_gro_reset_offset(skb);
5696 ret = napi_skb_finish(dev_gro_receive(napi, skb), skb);
5697 trace_napi_gro_receive_exit(ret);
5701 EXPORT_SYMBOL(napi_gro_receive);
5703 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
5705 if (unlikely(skb->pfmemalloc)) {
5709 __skb_pull(skb, skb_headlen(skb));
5710 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
5711 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
5712 __vlan_hwaccel_clear_tag(skb);
5713 skb->dev = napi->dev;
5716 /* eth_type_trans() assumes pkt_type is PACKET_HOST */
5717 skb->pkt_type = PACKET_HOST;
5719 skb->encapsulation = 0;
5720 skb_shinfo(skb)->gso_type = 0;
5721 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
5727 struct sk_buff *napi_get_frags(struct napi_struct *napi)
5729 struct sk_buff *skb = napi->skb;
5732 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
5735 skb_mark_napi_id(skb, napi);
5740 EXPORT_SYMBOL(napi_get_frags);
5742 static gro_result_t napi_frags_finish(struct napi_struct *napi,
5743 struct sk_buff *skb,
5749 __skb_push(skb, ETH_HLEN);
5750 skb->protocol = eth_type_trans(skb, skb->dev);
5751 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
5756 napi_reuse_skb(napi, skb);
5759 case GRO_MERGED_FREE:
5760 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
5761 napi_skb_free_stolen_head(skb);
5763 napi_reuse_skb(napi, skb);
5774 /* Upper GRO stack assumes network header starts at gro_offset=0
5775 * Drivers could call both napi_gro_frags() and napi_gro_receive()
5776 * We copy ethernet header into skb->data to have a common layout.
5778 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
5780 struct sk_buff *skb = napi->skb;
5781 const struct ethhdr *eth;
5782 unsigned int hlen = sizeof(*eth);
5786 skb_reset_mac_header(skb);
5787 skb_gro_reset_offset(skb);
5789 eth = skb_gro_header_fast(skb, 0);
5790 if (unlikely(skb_gro_header_hard(skb, hlen))) {
5791 eth = skb_gro_header_slow(skb, hlen, 0);
5792 if (unlikely(!eth)) {
5793 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
5794 __func__, napi->dev->name);
5795 napi_reuse_skb(napi, skb);
5799 gro_pull_from_frag0(skb, hlen);
5800 NAPI_GRO_CB(skb)->frag0 += hlen;
5801 NAPI_GRO_CB(skb)->frag0_len -= hlen;
5803 __skb_pull(skb, hlen);
5806 * This works because the only protocols we care about don't require
5808 * We'll fix it up properly in napi_frags_finish()
5810 skb->protocol = eth->h_proto;
5815 gro_result_t napi_gro_frags(struct napi_struct *napi)
5818 struct sk_buff *skb = napi_frags_skb(napi);
5823 trace_napi_gro_frags_entry(skb);
5825 ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
5826 trace_napi_gro_frags_exit(ret);
5830 EXPORT_SYMBOL(napi_gro_frags);
5832 /* Compute the checksum from gro_offset and return the folded value
5833 * after adding in any pseudo checksum.
5835 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
5840 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
5842 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
5843 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
5844 /* See comments in __skb_checksum_complete(). */
5846 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
5847 !skb->csum_complete_sw)
5848 netdev_rx_csum_fault(skb->dev, skb);
5851 NAPI_GRO_CB(skb)->csum = wsum;
5852 NAPI_GRO_CB(skb)->csum_valid = 1;
5856 EXPORT_SYMBOL(__skb_gro_checksum_complete);
5858 static void net_rps_send_ipi(struct softnet_data *remsd)
5862 struct softnet_data *next = remsd->rps_ipi_next;
5864 if (cpu_online(remsd->cpu))
5865 smp_call_function_single_async(remsd->cpu, &remsd->csd);
5872 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
5873 * Note: called with local irq disabled, but exits with local irq enabled.
5875 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
5878 struct softnet_data *remsd = sd->rps_ipi_list;
5881 sd->rps_ipi_list = NULL;
5885 /* Send pending IPI's to kick RPS processing on remote cpus. */
5886 net_rps_send_ipi(remsd);
5892 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
5895 return sd->rps_ipi_list != NULL;
5901 static int process_backlog(struct napi_struct *napi, int quota)
5903 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
5907 /* Check if we have pending ipi, its better to send them now,
5908 * not waiting net_rx_action() end.
5910 if (sd_has_rps_ipi_waiting(sd)) {
5911 local_irq_disable();
5912 net_rps_action_and_irq_enable(sd);
5915 napi->weight = dev_rx_weight;
5917 struct sk_buff *skb;
5919 while ((skb = __skb_dequeue(&sd->process_queue))) {
5921 __netif_receive_skb(skb);
5923 input_queue_head_incr(sd);
5924 if (++work >= quota)
5929 local_irq_disable();
5931 if (skb_queue_empty(&sd->input_pkt_queue)) {
5933 * Inline a custom version of __napi_complete().
5934 * only current cpu owns and manipulates this napi,
5935 * and NAPI_STATE_SCHED is the only possible flag set
5937 * We can use a plain write instead of clear_bit(),
5938 * and we dont need an smp_mb() memory barrier.
5943 skb_queue_splice_tail_init(&sd->input_pkt_queue,
5944 &sd->process_queue);
5954 * __napi_schedule - schedule for receive
5955 * @n: entry to schedule
5957 * The entry's receive function will be scheduled to run.
5958 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
5960 void __napi_schedule(struct napi_struct *n)
5962 unsigned long flags;
5964 local_irq_save(flags);
5965 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
5966 local_irq_restore(flags);
5968 EXPORT_SYMBOL(__napi_schedule);
5971 * napi_schedule_prep - check if napi can be scheduled
5974 * Test if NAPI routine is already running, and if not mark
5975 * it as running. This is used as a condition variable
5976 * insure only one NAPI poll instance runs. We also make
5977 * sure there is no pending NAPI disable.
5979 bool napi_schedule_prep(struct napi_struct *n)
5981 unsigned long val, new;
5984 val = READ_ONCE(n->state);
5985 if (unlikely(val & NAPIF_STATE_DISABLE))
5987 new = val | NAPIF_STATE_SCHED;
5989 /* Sets STATE_MISSED bit if STATE_SCHED was already set
5990 * This was suggested by Alexander Duyck, as compiler
5991 * emits better code than :
5992 * if (val & NAPIF_STATE_SCHED)
5993 * new |= NAPIF_STATE_MISSED;
5995 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
5997 } while (cmpxchg(&n->state, val, new) != val);
5999 return !(val & NAPIF_STATE_SCHED);
6001 EXPORT_SYMBOL(napi_schedule_prep);
6004 * __napi_schedule_irqoff - schedule for receive
6005 * @n: entry to schedule
6007 * Variant of __napi_schedule() assuming hard irqs are masked
6009 void __napi_schedule_irqoff(struct napi_struct *n)
6011 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6013 EXPORT_SYMBOL(__napi_schedule_irqoff);
6015 bool napi_complete_done(struct napi_struct *n, int work_done)
6017 unsigned long flags, val, new;
6020 * 1) Don't let napi dequeue from the cpu poll list
6021 * just in case its running on a different cpu.
6022 * 2) If we are busy polling, do nothing here, we have
6023 * the guarantee we will be called later.
6025 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6026 NAPIF_STATE_IN_BUSY_POLL)))
6029 if (n->gro_bitmask) {
6030 unsigned long timeout = 0;
6033 timeout = n->dev->gro_flush_timeout;
6035 /* When the NAPI instance uses a timeout and keeps postponing
6036 * it, we need to bound somehow the time packets are kept in
6039 napi_gro_flush(n, !!timeout);
6041 hrtimer_start(&n->timer, ns_to_ktime(timeout),
6042 HRTIMER_MODE_REL_PINNED);
6044 if (unlikely(!list_empty(&n->poll_list))) {
6045 /* If n->poll_list is not empty, we need to mask irqs */
6046 local_irq_save(flags);
6047 list_del_init(&n->poll_list);
6048 local_irq_restore(flags);
6052 val = READ_ONCE(n->state);
6054 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6056 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED);
6058 /* If STATE_MISSED was set, leave STATE_SCHED set,
6059 * because we will call napi->poll() one more time.
6060 * This C code was suggested by Alexander Duyck to help gcc.
6062 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6064 } while (cmpxchg(&n->state, val, new) != val);
6066 if (unlikely(val & NAPIF_STATE_MISSED)) {
6073 EXPORT_SYMBOL(napi_complete_done);
6075 /* must be called under rcu_read_lock(), as we dont take a reference */
6076 static struct napi_struct *napi_by_id(unsigned int napi_id)
6078 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6079 struct napi_struct *napi;
6081 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6082 if (napi->napi_id == napi_id)
6088 #if defined(CONFIG_NET_RX_BUSY_POLL)
6090 #define BUSY_POLL_BUDGET 8
6092 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock)
6096 /* Busy polling means there is a high chance device driver hard irq
6097 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6098 * set in napi_schedule_prep().
6099 * Since we are about to call napi->poll() once more, we can safely
6100 * clear NAPI_STATE_MISSED.
6102 * Note: x86 could use a single "lock and ..." instruction
6103 * to perform these two clear_bit()
6105 clear_bit(NAPI_STATE_MISSED, &napi->state);
6106 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6110 /* All we really want here is to re-enable device interrupts.
6111 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6113 rc = napi->poll(napi, BUSY_POLL_BUDGET);
6114 trace_napi_poll(napi, rc, BUSY_POLL_BUDGET);
6115 netpoll_poll_unlock(have_poll_lock);
6116 if (rc == BUSY_POLL_BUDGET)
6117 __napi_schedule(napi);
6121 void napi_busy_loop(unsigned int napi_id,
6122 bool (*loop_end)(void *, unsigned long),
6125 unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6126 int (*napi_poll)(struct napi_struct *napi, int budget);
6127 void *have_poll_lock = NULL;
6128 struct napi_struct *napi;
6135 napi = napi_by_id(napi_id);
6145 unsigned long val = READ_ONCE(napi->state);
6147 /* If multiple threads are competing for this napi,
6148 * we avoid dirtying napi->state as much as we can.
6150 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6151 NAPIF_STATE_IN_BUSY_POLL))
6153 if (cmpxchg(&napi->state, val,
6154 val | NAPIF_STATE_IN_BUSY_POLL |
6155 NAPIF_STATE_SCHED) != val)
6157 have_poll_lock = netpoll_poll_lock(napi);
6158 napi_poll = napi->poll;
6160 work = napi_poll(napi, BUSY_POLL_BUDGET);
6161 trace_napi_poll(napi, work, BUSY_POLL_BUDGET);
6164 __NET_ADD_STATS(dev_net(napi->dev),
6165 LINUX_MIB_BUSYPOLLRXPACKETS, work);
6168 if (!loop_end || loop_end(loop_end_arg, start_time))
6171 if (unlikely(need_resched())) {
6173 busy_poll_stop(napi, have_poll_lock);
6177 if (loop_end(loop_end_arg, start_time))
6184 busy_poll_stop(napi, have_poll_lock);
6189 EXPORT_SYMBOL(napi_busy_loop);
6191 #endif /* CONFIG_NET_RX_BUSY_POLL */
6193 static void napi_hash_add(struct napi_struct *napi)
6195 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) ||
6196 test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
6199 spin_lock(&napi_hash_lock);
6201 /* 0..NR_CPUS range is reserved for sender_cpu use */
6203 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6204 napi_gen_id = MIN_NAPI_ID;
6205 } while (napi_by_id(napi_gen_id));
6206 napi->napi_id = napi_gen_id;
6208 hlist_add_head_rcu(&napi->napi_hash_node,
6209 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6211 spin_unlock(&napi_hash_lock);
6214 /* Warning : caller is responsible to make sure rcu grace period
6215 * is respected before freeing memory containing @napi
6217 bool napi_hash_del(struct napi_struct *napi)
6219 bool rcu_sync_needed = false;
6221 spin_lock(&napi_hash_lock);
6223 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) {
6224 rcu_sync_needed = true;
6225 hlist_del_rcu(&napi->napi_hash_node);
6227 spin_unlock(&napi_hash_lock);
6228 return rcu_sync_needed;
6230 EXPORT_SYMBOL_GPL(napi_hash_del);
6232 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6234 struct napi_struct *napi;
6236 napi = container_of(timer, struct napi_struct, timer);
6238 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
6239 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6241 if (napi->gro_bitmask && !napi_disable_pending(napi) &&
6242 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state))
6243 __napi_schedule_irqoff(napi);
6245 return HRTIMER_NORESTART;
6248 static void init_gro_hash(struct napi_struct *napi)
6252 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6253 INIT_LIST_HEAD(&napi->gro_hash[i].list);
6254 napi->gro_hash[i].count = 0;
6256 napi->gro_bitmask = 0;
6259 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
6260 int (*poll)(struct napi_struct *, int), int weight)
6262 INIT_LIST_HEAD(&napi->poll_list);
6263 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6264 napi->timer.function = napi_watchdog;
6265 init_gro_hash(napi);
6268 if (weight > NAPI_POLL_WEIGHT)
6269 netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6271 napi->weight = weight;
6272 list_add(&napi->dev_list, &dev->napi_list);
6274 #ifdef CONFIG_NETPOLL
6275 napi->poll_owner = -1;
6277 set_bit(NAPI_STATE_SCHED, &napi->state);
6278 napi_hash_add(napi);
6280 EXPORT_SYMBOL(netif_napi_add);
6282 void napi_disable(struct napi_struct *n)
6285 set_bit(NAPI_STATE_DISABLE, &n->state);
6287 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
6289 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
6292 hrtimer_cancel(&n->timer);
6294 clear_bit(NAPI_STATE_DISABLE, &n->state);
6296 EXPORT_SYMBOL(napi_disable);
6298 static void flush_gro_hash(struct napi_struct *napi)
6302 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6303 struct sk_buff *skb, *n;
6305 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6307 napi->gro_hash[i].count = 0;
6311 /* Must be called in process context */
6312 void netif_napi_del(struct napi_struct *napi)
6315 if (napi_hash_del(napi))
6317 list_del_init(&napi->dev_list);
6318 napi_free_frags(napi);
6320 flush_gro_hash(napi);
6321 napi->gro_bitmask = 0;
6323 EXPORT_SYMBOL(netif_napi_del);
6325 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
6330 list_del_init(&n->poll_list);
6332 have = netpoll_poll_lock(n);
6336 /* This NAPI_STATE_SCHED test is for avoiding a race
6337 * with netpoll's poll_napi(). Only the entity which
6338 * obtains the lock and sees NAPI_STATE_SCHED set will
6339 * actually make the ->poll() call. Therefore we avoid
6340 * accidentally calling ->poll() when NAPI is not scheduled.
6343 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
6344 work = n->poll(n, weight);
6345 trace_napi_poll(n, work, weight);
6348 WARN_ON_ONCE(work > weight);
6350 if (likely(work < weight))
6353 /* Drivers must not modify the NAPI state if they
6354 * consume the entire weight. In such cases this code
6355 * still "owns" the NAPI instance and therefore can
6356 * move the instance around on the list at-will.
6358 if (unlikely(napi_disable_pending(n))) {
6363 if (n->gro_bitmask) {
6364 /* flush too old packets
6365 * If HZ < 1000, flush all packets.
6367 napi_gro_flush(n, HZ >= 1000);
6370 /* Some drivers may have called napi_schedule
6371 * prior to exhausting their budget.
6373 if (unlikely(!list_empty(&n->poll_list))) {
6374 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
6375 n->dev ? n->dev->name : "backlog");
6379 list_add_tail(&n->poll_list, repoll);
6382 netpoll_poll_unlock(have);
6387 static __latent_entropy void net_rx_action(struct softirq_action *h)
6389 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6390 unsigned long time_limit = jiffies +
6391 usecs_to_jiffies(netdev_budget_usecs);
6392 int budget = netdev_budget;
6396 local_irq_disable();
6397 list_splice_init(&sd->poll_list, &list);
6401 struct napi_struct *n;
6403 if (list_empty(&list)) {
6404 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
6409 n = list_first_entry(&list, struct napi_struct, poll_list);
6410 budget -= napi_poll(n, &repoll);
6412 /* If softirq window is exhausted then punt.
6413 * Allow this to run for 2 jiffies since which will allow
6414 * an average latency of 1.5/HZ.
6416 if (unlikely(budget <= 0 ||
6417 time_after_eq(jiffies, time_limit))) {
6423 local_irq_disable();
6425 list_splice_tail_init(&sd->poll_list, &list);
6426 list_splice_tail(&repoll, &list);
6427 list_splice(&list, &sd->poll_list);
6428 if (!list_empty(&sd->poll_list))
6429 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
6431 net_rps_action_and_irq_enable(sd);
6433 __kfree_skb_flush();
6436 struct netdev_adjacent {
6437 struct net_device *dev;
6439 /* upper master flag, there can only be one master device per list */
6442 /* counter for the number of times this device was added to us */
6445 /* private field for the users */
6448 struct list_head list;
6449 struct rcu_head rcu;
6452 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
6453 struct list_head *adj_list)
6455 struct netdev_adjacent *adj;
6457 list_for_each_entry(adj, adj_list, list) {
6458 if (adj->dev == adj_dev)
6464 static int __netdev_has_upper_dev(struct net_device *upper_dev, void *data)
6466 struct net_device *dev = data;
6468 return upper_dev == dev;
6472 * netdev_has_upper_dev - Check if device is linked to an upper device
6474 * @upper_dev: upper device to check
6476 * Find out if a device is linked to specified upper device and return true
6477 * in case it is. Note that this checks only immediate upper device,
6478 * not through a complete stack of devices. The caller must hold the RTNL lock.
6480 bool netdev_has_upper_dev(struct net_device *dev,
6481 struct net_device *upper_dev)
6485 return netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
6488 EXPORT_SYMBOL(netdev_has_upper_dev);
6491 * netdev_has_upper_dev_all - Check if device is linked to an upper device
6493 * @upper_dev: upper device to check
6495 * Find out if a device is linked to specified upper device and return true
6496 * in case it is. Note that this checks the entire upper device chain.
6497 * The caller must hold rcu lock.
6500 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
6501 struct net_device *upper_dev)
6503 return !!netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
6506 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
6509 * netdev_has_any_upper_dev - Check if device is linked to some device
6512 * Find out if a device is linked to an upper device and return true in case
6513 * it is. The caller must hold the RTNL lock.
6515 bool netdev_has_any_upper_dev(struct net_device *dev)
6519 return !list_empty(&dev->adj_list.upper);
6521 EXPORT_SYMBOL(netdev_has_any_upper_dev);
6524 * netdev_master_upper_dev_get - Get master upper device
6527 * Find a master upper device and return pointer to it or NULL in case
6528 * it's not there. The caller must hold the RTNL lock.
6530 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
6532 struct netdev_adjacent *upper;
6536 if (list_empty(&dev->adj_list.upper))
6539 upper = list_first_entry(&dev->adj_list.upper,
6540 struct netdev_adjacent, list);
6541 if (likely(upper->master))
6545 EXPORT_SYMBOL(netdev_master_upper_dev_get);
6548 * netdev_has_any_lower_dev - Check if device is linked to some device
6551 * Find out if a device is linked to a lower device and return true in case
6552 * it is. The caller must hold the RTNL lock.
6554 static bool netdev_has_any_lower_dev(struct net_device *dev)
6558 return !list_empty(&dev->adj_list.lower);
6561 void *netdev_adjacent_get_private(struct list_head *adj_list)
6563 struct netdev_adjacent *adj;
6565 adj = list_entry(adj_list, struct netdev_adjacent, list);
6567 return adj->private;
6569 EXPORT_SYMBOL(netdev_adjacent_get_private);
6572 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
6574 * @iter: list_head ** of the current position
6576 * Gets the next device from the dev's upper list, starting from iter
6577 * position. The caller must hold RCU read lock.
6579 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
6580 struct list_head **iter)
6582 struct netdev_adjacent *upper;
6584 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6586 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6588 if (&upper->list == &dev->adj_list.upper)
6591 *iter = &upper->list;
6595 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
6597 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
6598 struct list_head **iter)
6600 struct netdev_adjacent *upper;
6602 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6604 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6606 if (&upper->list == &dev->adj_list.upper)
6609 *iter = &upper->list;
6614 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
6615 int (*fn)(struct net_device *dev,
6619 struct net_device *udev;
6620 struct list_head *iter;
6623 for (iter = &dev->adj_list.upper,
6624 udev = netdev_next_upper_dev_rcu(dev, &iter);
6626 udev = netdev_next_upper_dev_rcu(dev, &iter)) {
6627 /* first is the upper device itself */
6628 ret = fn(udev, data);
6632 /* then look at all of its upper devices */
6633 ret = netdev_walk_all_upper_dev_rcu(udev, fn, data);
6640 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
6643 * netdev_lower_get_next_private - Get the next ->private from the
6644 * lower neighbour list
6646 * @iter: list_head ** of the current position
6648 * Gets the next netdev_adjacent->private from the dev's lower neighbour
6649 * list, starting from iter position. The caller must hold either hold the
6650 * RTNL lock or its own locking that guarantees that the neighbour lower
6651 * list will remain unchanged.
6653 void *netdev_lower_get_next_private(struct net_device *dev,
6654 struct list_head **iter)
6656 struct netdev_adjacent *lower;
6658 lower = list_entry(*iter, struct netdev_adjacent, list);
6660 if (&lower->list == &dev->adj_list.lower)
6663 *iter = lower->list.next;
6665 return lower->private;
6667 EXPORT_SYMBOL(netdev_lower_get_next_private);
6670 * netdev_lower_get_next_private_rcu - Get the next ->private from the
6671 * lower neighbour list, RCU
6674 * @iter: list_head ** of the current position
6676 * Gets the next netdev_adjacent->private from the dev's lower neighbour
6677 * list, starting from iter position. The caller must hold RCU read lock.
6679 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
6680 struct list_head **iter)
6682 struct netdev_adjacent *lower;
6684 WARN_ON_ONCE(!rcu_read_lock_held());
6686 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6688 if (&lower->list == &dev->adj_list.lower)
6691 *iter = &lower->list;
6693 return lower->private;
6695 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
6698 * netdev_lower_get_next - Get the next device from the lower neighbour
6701 * @iter: list_head ** of the current position
6703 * Gets the next netdev_adjacent from the dev's lower neighbour
6704 * list, starting from iter position. The caller must hold RTNL lock or
6705 * its own locking that guarantees that the neighbour lower
6706 * list will remain unchanged.
6708 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
6710 struct netdev_adjacent *lower;
6712 lower = list_entry(*iter, struct netdev_adjacent, list);
6714 if (&lower->list == &dev->adj_list.lower)
6717 *iter = lower->list.next;
6721 EXPORT_SYMBOL(netdev_lower_get_next);
6723 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
6724 struct list_head **iter)
6726 struct netdev_adjacent *lower;
6728 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
6730 if (&lower->list == &dev->adj_list.lower)
6733 *iter = &lower->list;
6738 int netdev_walk_all_lower_dev(struct net_device *dev,
6739 int (*fn)(struct net_device *dev,
6743 struct net_device *ldev;
6744 struct list_head *iter;
6747 for (iter = &dev->adj_list.lower,
6748 ldev = netdev_next_lower_dev(dev, &iter);
6750 ldev = netdev_next_lower_dev(dev, &iter)) {
6751 /* first is the lower device itself */
6752 ret = fn(ldev, data);
6756 /* then look at all of its lower devices */
6757 ret = netdev_walk_all_lower_dev(ldev, fn, data);
6764 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
6766 static struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
6767 struct list_head **iter)
6769 struct netdev_adjacent *lower;
6771 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6772 if (&lower->list == &dev->adj_list.lower)
6775 *iter = &lower->list;
6780 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
6781 int (*fn)(struct net_device *dev,
6785 struct net_device *ldev;
6786 struct list_head *iter;
6789 for (iter = &dev->adj_list.lower,
6790 ldev = netdev_next_lower_dev_rcu(dev, &iter);
6792 ldev = netdev_next_lower_dev_rcu(dev, &iter)) {
6793 /* first is the lower device itself */
6794 ret = fn(ldev, data);
6798 /* then look at all of its lower devices */
6799 ret = netdev_walk_all_lower_dev_rcu(ldev, fn, data);
6806 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
6809 * netdev_lower_get_first_private_rcu - Get the first ->private from the
6810 * lower neighbour list, RCU
6814 * Gets the first netdev_adjacent->private from the dev's lower neighbour
6815 * list. The caller must hold RCU read lock.
6817 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
6819 struct netdev_adjacent *lower;
6821 lower = list_first_or_null_rcu(&dev->adj_list.lower,
6822 struct netdev_adjacent, list);
6824 return lower->private;
6827 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
6830 * netdev_master_upper_dev_get_rcu - Get master upper device
6833 * Find a master upper device and return pointer to it or NULL in case
6834 * it's not there. The caller must hold the RCU read lock.
6836 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
6838 struct netdev_adjacent *upper;
6840 upper = list_first_or_null_rcu(&dev->adj_list.upper,
6841 struct netdev_adjacent, list);
6842 if (upper && likely(upper->master))
6846 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
6848 static int netdev_adjacent_sysfs_add(struct net_device *dev,
6849 struct net_device *adj_dev,
6850 struct list_head *dev_list)
6852 char linkname[IFNAMSIZ+7];
6854 sprintf(linkname, dev_list == &dev->adj_list.upper ?
6855 "upper_%s" : "lower_%s", adj_dev->name);
6856 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
6859 static void netdev_adjacent_sysfs_del(struct net_device *dev,
6861 struct list_head *dev_list)
6863 char linkname[IFNAMSIZ+7];
6865 sprintf(linkname, dev_list == &dev->adj_list.upper ?
6866 "upper_%s" : "lower_%s", name);
6867 sysfs_remove_link(&(dev->dev.kobj), linkname);
6870 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
6871 struct net_device *adj_dev,
6872 struct list_head *dev_list)
6874 return (dev_list == &dev->adj_list.upper ||
6875 dev_list == &dev->adj_list.lower) &&
6876 net_eq(dev_net(dev), dev_net(adj_dev));
6879 static int __netdev_adjacent_dev_insert(struct net_device *dev,
6880 struct net_device *adj_dev,
6881 struct list_head *dev_list,
6882 void *private, bool master)
6884 struct netdev_adjacent *adj;
6887 adj = __netdev_find_adj(adj_dev, dev_list);
6891 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
6892 dev->name, adj_dev->name, adj->ref_nr);
6897 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
6902 adj->master = master;
6904 adj->private = private;
6907 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
6908 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
6910 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
6911 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
6916 /* Ensure that master link is always the first item in list. */
6918 ret = sysfs_create_link(&(dev->dev.kobj),
6919 &(adj_dev->dev.kobj), "master");
6921 goto remove_symlinks;
6923 list_add_rcu(&adj->list, dev_list);
6925 list_add_tail_rcu(&adj->list, dev_list);
6931 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
6932 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
6940 static void __netdev_adjacent_dev_remove(struct net_device *dev,
6941 struct net_device *adj_dev,
6943 struct list_head *dev_list)
6945 struct netdev_adjacent *adj;
6947 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
6948 dev->name, adj_dev->name, ref_nr);
6950 adj = __netdev_find_adj(adj_dev, dev_list);
6953 pr_err("Adjacency does not exist for device %s from %s\n",
6954 dev->name, adj_dev->name);
6959 if (adj->ref_nr > ref_nr) {
6960 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
6961 dev->name, adj_dev->name, ref_nr,
6962 adj->ref_nr - ref_nr);
6963 adj->ref_nr -= ref_nr;
6968 sysfs_remove_link(&(dev->dev.kobj), "master");
6970 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
6971 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
6973 list_del_rcu(&adj->list);
6974 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
6975 adj_dev->name, dev->name, adj_dev->name);
6977 kfree_rcu(adj, rcu);
6980 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
6981 struct net_device *upper_dev,
6982 struct list_head *up_list,
6983 struct list_head *down_list,
6984 void *private, bool master)
6988 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
6993 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
6996 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
7003 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
7004 struct net_device *upper_dev,
7006 struct list_head *up_list,
7007 struct list_head *down_list)
7009 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
7010 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
7013 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
7014 struct net_device *upper_dev,
7015 void *private, bool master)
7017 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
7018 &dev->adj_list.upper,
7019 &upper_dev->adj_list.lower,
7023 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
7024 struct net_device *upper_dev)
7026 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
7027 &dev->adj_list.upper,
7028 &upper_dev->adj_list.lower);
7031 static int __netdev_upper_dev_link(struct net_device *dev,
7032 struct net_device *upper_dev, bool master,
7033 void *upper_priv, void *upper_info,
7034 struct netlink_ext_ack *extack)
7036 struct netdev_notifier_changeupper_info changeupper_info = {
7041 .upper_dev = upper_dev,
7044 .upper_info = upper_info,
7046 struct net_device *master_dev;
7051 if (dev == upper_dev)
7054 /* To prevent loops, check if dev is not upper device to upper_dev. */
7055 if (netdev_has_upper_dev(upper_dev, dev))
7059 if (netdev_has_upper_dev(dev, upper_dev))
7062 master_dev = netdev_master_upper_dev_get(dev);
7064 return master_dev == upper_dev ? -EEXIST : -EBUSY;
7067 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7068 &changeupper_info.info);
7069 ret = notifier_to_errno(ret);
7073 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
7078 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7079 &changeupper_info.info);
7080 ret = notifier_to_errno(ret);
7087 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7093 * netdev_upper_dev_link - Add a link to the upper device
7095 * @upper_dev: new upper device
7096 * @extack: netlink extended ack
7098 * Adds a link to device which is upper to this one. The caller must hold
7099 * the RTNL lock. On a failure a negative errno code is returned.
7100 * On success the reference counts are adjusted and the function
7103 int netdev_upper_dev_link(struct net_device *dev,
7104 struct net_device *upper_dev,
7105 struct netlink_ext_ack *extack)
7107 return __netdev_upper_dev_link(dev, upper_dev, false,
7108 NULL, NULL, extack);
7110 EXPORT_SYMBOL(netdev_upper_dev_link);
7113 * netdev_master_upper_dev_link - Add a master link to the upper device
7115 * @upper_dev: new upper device
7116 * @upper_priv: upper device private
7117 * @upper_info: upper info to be passed down via notifier
7118 * @extack: netlink extended ack
7120 * Adds a link to device which is upper to this one. In this case, only
7121 * one master upper device can be linked, although other non-master devices
7122 * might be linked as well. The caller must hold the RTNL lock.
7123 * On a failure a negative errno code is returned. On success the reference
7124 * counts are adjusted and the function returns zero.
7126 int netdev_master_upper_dev_link(struct net_device *dev,
7127 struct net_device *upper_dev,
7128 void *upper_priv, void *upper_info,
7129 struct netlink_ext_ack *extack)
7131 return __netdev_upper_dev_link(dev, upper_dev, true,
7132 upper_priv, upper_info, extack);
7134 EXPORT_SYMBOL(netdev_master_upper_dev_link);
7137 * netdev_upper_dev_unlink - Removes a link to upper device
7139 * @upper_dev: new upper device
7141 * Removes a link to device which is upper to this one. The caller must hold
7144 void netdev_upper_dev_unlink(struct net_device *dev,
7145 struct net_device *upper_dev)
7147 struct netdev_notifier_changeupper_info changeupper_info = {
7151 .upper_dev = upper_dev,
7157 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
7159 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7160 &changeupper_info.info);
7162 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7164 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7165 &changeupper_info.info);
7167 EXPORT_SYMBOL(netdev_upper_dev_unlink);
7170 * netdev_bonding_info_change - Dispatch event about slave change
7172 * @bonding_info: info to dispatch
7174 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
7175 * The caller must hold the RTNL lock.
7177 void netdev_bonding_info_change(struct net_device *dev,
7178 struct netdev_bonding_info *bonding_info)
7180 struct netdev_notifier_bonding_info info = {
7184 memcpy(&info.bonding_info, bonding_info,
7185 sizeof(struct netdev_bonding_info));
7186 call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
7189 EXPORT_SYMBOL(netdev_bonding_info_change);
7191 static void netdev_adjacent_add_links(struct net_device *dev)
7193 struct netdev_adjacent *iter;
7195 struct net *net = dev_net(dev);
7197 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7198 if (!net_eq(net, dev_net(iter->dev)))
7200 netdev_adjacent_sysfs_add(iter->dev, dev,
7201 &iter->dev->adj_list.lower);
7202 netdev_adjacent_sysfs_add(dev, iter->dev,
7203 &dev->adj_list.upper);
7206 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7207 if (!net_eq(net, dev_net(iter->dev)))
7209 netdev_adjacent_sysfs_add(iter->dev, dev,
7210 &iter->dev->adj_list.upper);
7211 netdev_adjacent_sysfs_add(dev, iter->dev,
7212 &dev->adj_list.lower);
7216 static void netdev_adjacent_del_links(struct net_device *dev)
7218 struct netdev_adjacent *iter;
7220 struct net *net = dev_net(dev);
7222 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7223 if (!net_eq(net, dev_net(iter->dev)))
7225 netdev_adjacent_sysfs_del(iter->dev, dev->name,
7226 &iter->dev->adj_list.lower);
7227 netdev_adjacent_sysfs_del(dev, iter->dev->name,
7228 &dev->adj_list.upper);
7231 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7232 if (!net_eq(net, dev_net(iter->dev)))
7234 netdev_adjacent_sysfs_del(iter->dev, dev->name,
7235 &iter->dev->adj_list.upper);
7236 netdev_adjacent_sysfs_del(dev, iter->dev->name,
7237 &dev->adj_list.lower);
7241 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
7243 struct netdev_adjacent *iter;
7245 struct net *net = dev_net(dev);
7247 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7248 if (!net_eq(net, dev_net(iter->dev)))
7250 netdev_adjacent_sysfs_del(iter->dev, oldname,
7251 &iter->dev->adj_list.lower);
7252 netdev_adjacent_sysfs_add(iter->dev, dev,
7253 &iter->dev->adj_list.lower);
7256 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7257 if (!net_eq(net, dev_net(iter->dev)))
7259 netdev_adjacent_sysfs_del(iter->dev, oldname,
7260 &iter->dev->adj_list.upper);
7261 netdev_adjacent_sysfs_add(iter->dev, dev,
7262 &iter->dev->adj_list.upper);
7266 void *netdev_lower_dev_get_private(struct net_device *dev,
7267 struct net_device *lower_dev)
7269 struct netdev_adjacent *lower;
7273 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
7277 return lower->private;
7279 EXPORT_SYMBOL(netdev_lower_dev_get_private);
7282 int dev_get_nest_level(struct net_device *dev)
7284 struct net_device *lower = NULL;
7285 struct list_head *iter;
7291 netdev_for_each_lower_dev(dev, lower, iter) {
7292 nest = dev_get_nest_level(lower);
7293 if (max_nest < nest)
7297 return max_nest + 1;
7299 EXPORT_SYMBOL(dev_get_nest_level);
7302 * netdev_lower_change - Dispatch event about lower device state change
7303 * @lower_dev: device
7304 * @lower_state_info: state to dispatch
7306 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
7307 * The caller must hold the RTNL lock.
7309 void netdev_lower_state_changed(struct net_device *lower_dev,
7310 void *lower_state_info)
7312 struct netdev_notifier_changelowerstate_info changelowerstate_info = {
7313 .info.dev = lower_dev,
7317 changelowerstate_info.lower_state_info = lower_state_info;
7318 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
7319 &changelowerstate_info.info);
7321 EXPORT_SYMBOL(netdev_lower_state_changed);
7323 static void dev_change_rx_flags(struct net_device *dev, int flags)
7325 const struct net_device_ops *ops = dev->netdev_ops;
7327 if (ops->ndo_change_rx_flags)
7328 ops->ndo_change_rx_flags(dev, flags);
7331 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
7333 unsigned int old_flags = dev->flags;
7339 dev->flags |= IFF_PROMISC;
7340 dev->promiscuity += inc;
7341 if (dev->promiscuity == 0) {
7344 * If inc causes overflow, untouch promisc and return error.
7347 dev->flags &= ~IFF_PROMISC;
7349 dev->promiscuity -= inc;
7350 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
7355 if (dev->flags != old_flags) {
7356 pr_info("device %s %s promiscuous mode\n",
7358 dev->flags & IFF_PROMISC ? "entered" : "left");
7359 if (audit_enabled) {
7360 current_uid_gid(&uid, &gid);
7361 audit_log(audit_context(), GFP_ATOMIC,
7362 AUDIT_ANOM_PROMISCUOUS,
7363 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
7364 dev->name, (dev->flags & IFF_PROMISC),
7365 (old_flags & IFF_PROMISC),
7366 from_kuid(&init_user_ns, audit_get_loginuid(current)),
7367 from_kuid(&init_user_ns, uid),
7368 from_kgid(&init_user_ns, gid),
7369 audit_get_sessionid(current));
7372 dev_change_rx_flags(dev, IFF_PROMISC);
7375 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
7380 * dev_set_promiscuity - update promiscuity count on a device
7384 * Add or remove promiscuity from a device. While the count in the device
7385 * remains above zero the interface remains promiscuous. Once it hits zero
7386 * the device reverts back to normal filtering operation. A negative inc
7387 * value is used to drop promiscuity on the device.
7388 * Return 0 if successful or a negative errno code on error.
7390 int dev_set_promiscuity(struct net_device *dev, int inc)
7392 unsigned int old_flags = dev->flags;
7395 err = __dev_set_promiscuity(dev, inc, true);
7398 if (dev->flags != old_flags)
7399 dev_set_rx_mode(dev);
7402 EXPORT_SYMBOL(dev_set_promiscuity);
7404 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
7406 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
7410 dev->flags |= IFF_ALLMULTI;
7411 dev->allmulti += inc;
7412 if (dev->allmulti == 0) {
7415 * If inc causes overflow, untouch allmulti and return error.
7418 dev->flags &= ~IFF_ALLMULTI;
7420 dev->allmulti -= inc;
7421 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
7426 if (dev->flags ^ old_flags) {
7427 dev_change_rx_flags(dev, IFF_ALLMULTI);
7428 dev_set_rx_mode(dev);
7430 __dev_notify_flags(dev, old_flags,
7431 dev->gflags ^ old_gflags);
7437 * dev_set_allmulti - update allmulti count on a device
7441 * Add or remove reception of all multicast frames to a device. While the
7442 * count in the device remains above zero the interface remains listening
7443 * to all interfaces. Once it hits zero the device reverts back to normal
7444 * filtering operation. A negative @inc value is used to drop the counter
7445 * when releasing a resource needing all multicasts.
7446 * Return 0 if successful or a negative errno code on error.
7449 int dev_set_allmulti(struct net_device *dev, int inc)
7451 return __dev_set_allmulti(dev, inc, true);
7453 EXPORT_SYMBOL(dev_set_allmulti);
7456 * Upload unicast and multicast address lists to device and
7457 * configure RX filtering. When the device doesn't support unicast
7458 * filtering it is put in promiscuous mode while unicast addresses
7461 void __dev_set_rx_mode(struct net_device *dev)
7463 const struct net_device_ops *ops = dev->netdev_ops;
7465 /* dev_open will call this function so the list will stay sane. */
7466 if (!(dev->flags&IFF_UP))
7469 if (!netif_device_present(dev))
7472 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
7473 /* Unicast addresses changes may only happen under the rtnl,
7474 * therefore calling __dev_set_promiscuity here is safe.
7476 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
7477 __dev_set_promiscuity(dev, 1, false);
7478 dev->uc_promisc = true;
7479 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
7480 __dev_set_promiscuity(dev, -1, false);
7481 dev->uc_promisc = false;
7485 if (ops->ndo_set_rx_mode)
7486 ops->ndo_set_rx_mode(dev);
7489 void dev_set_rx_mode(struct net_device *dev)
7491 netif_addr_lock_bh(dev);
7492 __dev_set_rx_mode(dev);
7493 netif_addr_unlock_bh(dev);
7497 * dev_get_flags - get flags reported to userspace
7500 * Get the combination of flag bits exported through APIs to userspace.
7502 unsigned int dev_get_flags(const struct net_device *dev)
7506 flags = (dev->flags & ~(IFF_PROMISC |
7511 (dev->gflags & (IFF_PROMISC |
7514 if (netif_running(dev)) {
7515 if (netif_oper_up(dev))
7516 flags |= IFF_RUNNING;
7517 if (netif_carrier_ok(dev))
7518 flags |= IFF_LOWER_UP;
7519 if (netif_dormant(dev))
7520 flags |= IFF_DORMANT;
7525 EXPORT_SYMBOL(dev_get_flags);
7527 int __dev_change_flags(struct net_device *dev, unsigned int flags,
7528 struct netlink_ext_ack *extack)
7530 unsigned int old_flags = dev->flags;
7536 * Set the flags on our device.
7539 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
7540 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
7542 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
7546 * Load in the correct multicast list now the flags have changed.
7549 if ((old_flags ^ flags) & IFF_MULTICAST)
7550 dev_change_rx_flags(dev, IFF_MULTICAST);
7552 dev_set_rx_mode(dev);
7555 * Have we downed the interface. We handle IFF_UP ourselves
7556 * according to user attempts to set it, rather than blindly
7561 if ((old_flags ^ flags) & IFF_UP) {
7562 if (old_flags & IFF_UP)
7565 ret = __dev_open(dev, extack);
7568 if ((flags ^ dev->gflags) & IFF_PROMISC) {
7569 int inc = (flags & IFF_PROMISC) ? 1 : -1;
7570 unsigned int old_flags = dev->flags;
7572 dev->gflags ^= IFF_PROMISC;
7574 if (__dev_set_promiscuity(dev, inc, false) >= 0)
7575 if (dev->flags != old_flags)
7576 dev_set_rx_mode(dev);
7579 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
7580 * is important. Some (broken) drivers set IFF_PROMISC, when
7581 * IFF_ALLMULTI is requested not asking us and not reporting.
7583 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
7584 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
7586 dev->gflags ^= IFF_ALLMULTI;
7587 __dev_set_allmulti(dev, inc, false);
7593 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
7594 unsigned int gchanges)
7596 unsigned int changes = dev->flags ^ old_flags;
7599 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
7601 if (changes & IFF_UP) {
7602 if (dev->flags & IFF_UP)
7603 call_netdevice_notifiers(NETDEV_UP, dev);
7605 call_netdevice_notifiers(NETDEV_DOWN, dev);
7608 if (dev->flags & IFF_UP &&
7609 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
7610 struct netdev_notifier_change_info change_info = {
7614 .flags_changed = changes,
7617 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
7622 * dev_change_flags - change device settings
7624 * @flags: device state flags
7625 * @extack: netlink extended ack
7627 * Change settings on device based state flags. The flags are
7628 * in the userspace exported format.
7630 int dev_change_flags(struct net_device *dev, unsigned int flags,
7631 struct netlink_ext_ack *extack)
7634 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
7636 ret = __dev_change_flags(dev, flags, extack);
7640 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
7641 __dev_notify_flags(dev, old_flags, changes);
7644 EXPORT_SYMBOL(dev_change_flags);
7646 int __dev_set_mtu(struct net_device *dev, int new_mtu)
7648 const struct net_device_ops *ops = dev->netdev_ops;
7650 if (ops->ndo_change_mtu)
7651 return ops->ndo_change_mtu(dev, new_mtu);
7656 EXPORT_SYMBOL(__dev_set_mtu);
7659 * dev_set_mtu_ext - Change maximum transfer unit
7661 * @new_mtu: new transfer unit
7662 * @extack: netlink extended ack
7664 * Change the maximum transfer size of the network device.
7666 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
7667 struct netlink_ext_ack *extack)
7671 if (new_mtu == dev->mtu)
7674 /* MTU must be positive, and in range */
7675 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
7676 NL_SET_ERR_MSG(extack, "mtu less than device minimum");
7680 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
7681 NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
7685 if (!netif_device_present(dev))
7688 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
7689 err = notifier_to_errno(err);
7693 orig_mtu = dev->mtu;
7694 err = __dev_set_mtu(dev, new_mtu);
7697 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
7699 err = notifier_to_errno(err);
7701 /* setting mtu back and notifying everyone again,
7702 * so that they have a chance to revert changes.
7704 __dev_set_mtu(dev, orig_mtu);
7705 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
7712 int dev_set_mtu(struct net_device *dev, int new_mtu)
7714 struct netlink_ext_ack extack;
7717 memset(&extack, 0, sizeof(extack));
7718 err = dev_set_mtu_ext(dev, new_mtu, &extack);
7719 if (err && extack._msg)
7720 net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
7723 EXPORT_SYMBOL(dev_set_mtu);
7726 * dev_change_tx_queue_len - Change TX queue length of a netdevice
7728 * @new_len: new tx queue length
7730 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
7732 unsigned int orig_len = dev->tx_queue_len;
7735 if (new_len != (unsigned int)new_len)
7738 if (new_len != orig_len) {
7739 dev->tx_queue_len = new_len;
7740 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
7741 res = notifier_to_errno(res);
7744 res = dev_qdisc_change_tx_queue_len(dev);
7752 netdev_err(dev, "refused to change device tx_queue_len\n");
7753 dev->tx_queue_len = orig_len;
7758 * dev_set_group - Change group this device belongs to
7760 * @new_group: group this device should belong to
7762 void dev_set_group(struct net_device *dev, int new_group)
7764 dev->group = new_group;
7766 EXPORT_SYMBOL(dev_set_group);
7769 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
7771 * @addr: new address
7772 * @extack: netlink extended ack
7774 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
7775 struct netlink_ext_ack *extack)
7777 struct netdev_notifier_pre_changeaddr_info info = {
7779 .info.extack = extack,
7784 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
7785 return notifier_to_errno(rc);
7787 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
7790 * dev_set_mac_address - Change Media Access Control Address
7793 * @extack: netlink extended ack
7795 * Change the hardware (MAC) address of the device
7797 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
7798 struct netlink_ext_ack *extack)
7800 const struct net_device_ops *ops = dev->netdev_ops;
7803 if (!ops->ndo_set_mac_address)
7805 if (sa->sa_family != dev->type)
7807 if (!netif_device_present(dev))
7809 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
7812 err = ops->ndo_set_mac_address(dev, sa);
7815 dev->addr_assign_type = NET_ADDR_SET;
7816 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
7817 add_device_randomness(dev->dev_addr, dev->addr_len);
7820 EXPORT_SYMBOL(dev_set_mac_address);
7823 * dev_change_carrier - Change device carrier
7825 * @new_carrier: new value
7827 * Change device carrier
7829 int dev_change_carrier(struct net_device *dev, bool new_carrier)
7831 const struct net_device_ops *ops = dev->netdev_ops;
7833 if (!ops->ndo_change_carrier)
7835 if (!netif_device_present(dev))
7837 return ops->ndo_change_carrier(dev, new_carrier);
7839 EXPORT_SYMBOL(dev_change_carrier);
7842 * dev_get_phys_port_id - Get device physical port ID
7846 * Get device physical port ID
7848 int dev_get_phys_port_id(struct net_device *dev,
7849 struct netdev_phys_item_id *ppid)
7851 const struct net_device_ops *ops = dev->netdev_ops;
7853 if (!ops->ndo_get_phys_port_id)
7855 return ops->ndo_get_phys_port_id(dev, ppid);
7857 EXPORT_SYMBOL(dev_get_phys_port_id);
7860 * dev_get_phys_port_name - Get device physical port name
7863 * @len: limit of bytes to copy to name
7865 * Get device physical port name
7867 int dev_get_phys_port_name(struct net_device *dev,
7868 char *name, size_t len)
7870 const struct net_device_ops *ops = dev->netdev_ops;
7872 if (!ops->ndo_get_phys_port_name)
7874 return ops->ndo_get_phys_port_name(dev, name, len);
7876 EXPORT_SYMBOL(dev_get_phys_port_name);
7879 * dev_get_port_parent_id - Get the device's port parent identifier
7880 * @dev: network device
7881 * @ppid: pointer to a storage for the port's parent identifier
7882 * @recurse: allow/disallow recursion to lower devices
7884 * Get the devices's port parent identifier
7886 int dev_get_port_parent_id(struct net_device *dev,
7887 struct netdev_phys_item_id *ppid,
7890 const struct net_device_ops *ops = dev->netdev_ops;
7891 struct netdev_phys_item_id first = { };
7892 struct net_device *lower_dev;
7893 struct list_head *iter;
7894 int err = -EOPNOTSUPP;
7896 if (ops->ndo_get_port_parent_id)
7897 return ops->ndo_get_port_parent_id(dev, ppid);
7902 netdev_for_each_lower_dev(dev, lower_dev, iter) {
7903 err = dev_get_port_parent_id(lower_dev, ppid, recurse);
7908 else if (memcmp(&first, ppid, sizeof(*ppid)))
7914 EXPORT_SYMBOL(dev_get_port_parent_id);
7917 * netdev_port_same_parent_id - Indicate if two network devices have
7918 * the same port parent identifier
7919 * @a: first network device
7920 * @b: second network device
7922 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
7924 struct netdev_phys_item_id a_id = { };
7925 struct netdev_phys_item_id b_id = { };
7927 if (dev_get_port_parent_id(a, &a_id, true) ||
7928 dev_get_port_parent_id(b, &b_id, true))
7931 return netdev_phys_item_id_same(&a_id, &b_id);
7933 EXPORT_SYMBOL(netdev_port_same_parent_id);
7936 * dev_change_proto_down - update protocol port state information
7938 * @proto_down: new value
7940 * This info can be used by switch drivers to set the phys state of the
7943 int dev_change_proto_down(struct net_device *dev, bool proto_down)
7945 const struct net_device_ops *ops = dev->netdev_ops;
7947 if (!ops->ndo_change_proto_down)
7949 if (!netif_device_present(dev))
7951 return ops->ndo_change_proto_down(dev, proto_down);
7953 EXPORT_SYMBOL(dev_change_proto_down);
7956 * dev_change_proto_down_generic - generic implementation for
7957 * ndo_change_proto_down that sets carrier according to
7961 * @proto_down: new value
7963 int dev_change_proto_down_generic(struct net_device *dev, bool proto_down)
7966 netif_carrier_off(dev);
7968 netif_carrier_on(dev);
7969 dev->proto_down = proto_down;
7972 EXPORT_SYMBOL(dev_change_proto_down_generic);
7974 u32 __dev_xdp_query(struct net_device *dev, bpf_op_t bpf_op,
7975 enum bpf_netdev_command cmd)
7977 struct netdev_bpf xdp;
7982 memset(&xdp, 0, sizeof(xdp));
7985 /* Query must always succeed. */
7986 WARN_ON(bpf_op(dev, &xdp) < 0 && cmd == XDP_QUERY_PROG);
7991 static int dev_xdp_install(struct net_device *dev, bpf_op_t bpf_op,
7992 struct netlink_ext_ack *extack, u32 flags,
7993 struct bpf_prog *prog)
7995 struct netdev_bpf xdp;
7997 memset(&xdp, 0, sizeof(xdp));
7998 if (flags & XDP_FLAGS_HW_MODE)
7999 xdp.command = XDP_SETUP_PROG_HW;
8001 xdp.command = XDP_SETUP_PROG;
8002 xdp.extack = extack;
8006 return bpf_op(dev, &xdp);
8009 static void dev_xdp_uninstall(struct net_device *dev)
8011 struct netdev_bpf xdp;
8014 /* Remove generic XDP */
8015 WARN_ON(dev_xdp_install(dev, generic_xdp_install, NULL, 0, NULL));
8017 /* Remove from the driver */
8018 ndo_bpf = dev->netdev_ops->ndo_bpf;
8022 memset(&xdp, 0, sizeof(xdp));
8023 xdp.command = XDP_QUERY_PROG;
8024 WARN_ON(ndo_bpf(dev, &xdp));
8026 WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags,
8029 /* Remove HW offload */
8030 memset(&xdp, 0, sizeof(xdp));
8031 xdp.command = XDP_QUERY_PROG_HW;
8032 if (!ndo_bpf(dev, &xdp) && xdp.prog_id)
8033 WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags,
8038 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
8040 * @extack: netlink extended ack
8041 * @fd: new program fd or negative value to clear
8042 * @flags: xdp-related flags
8044 * Set or clear a bpf program for a device
8046 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
8049 const struct net_device_ops *ops = dev->netdev_ops;
8050 enum bpf_netdev_command query;
8051 struct bpf_prog *prog = NULL;
8052 bpf_op_t bpf_op, bpf_chk;
8058 offload = flags & XDP_FLAGS_HW_MODE;
8059 query = offload ? XDP_QUERY_PROG_HW : XDP_QUERY_PROG;
8061 bpf_op = bpf_chk = ops->ndo_bpf;
8062 if (!bpf_op && (flags & (XDP_FLAGS_DRV_MODE | XDP_FLAGS_HW_MODE))) {
8063 NL_SET_ERR_MSG(extack, "underlying driver does not support XDP in native mode");
8066 if (!bpf_op || (flags & XDP_FLAGS_SKB_MODE))
8067 bpf_op = generic_xdp_install;
8068 if (bpf_op == bpf_chk)
8069 bpf_chk = generic_xdp_install;
8072 if (!offload && __dev_xdp_query(dev, bpf_chk, XDP_QUERY_PROG)) {
8073 NL_SET_ERR_MSG(extack, "native and generic XDP can't be active at the same time");
8076 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) &&
8077 __dev_xdp_query(dev, bpf_op, query)) {
8078 NL_SET_ERR_MSG(extack, "XDP program already attached");
8082 prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
8083 bpf_op == ops->ndo_bpf);
8085 return PTR_ERR(prog);
8087 if (!offload && bpf_prog_is_dev_bound(prog->aux)) {
8088 NL_SET_ERR_MSG(extack, "using device-bound program without HW_MODE flag is not supported");
8094 err = dev_xdp_install(dev, bpf_op, extack, flags, prog);
8095 if (err < 0 && prog)
8102 * dev_new_index - allocate an ifindex
8103 * @net: the applicable net namespace
8105 * Returns a suitable unique value for a new device interface
8106 * number. The caller must hold the rtnl semaphore or the
8107 * dev_base_lock to be sure it remains unique.
8109 static int dev_new_index(struct net *net)
8111 int ifindex = net->ifindex;
8116 if (!__dev_get_by_index(net, ifindex))
8117 return net->ifindex = ifindex;
8121 /* Delayed registration/unregisteration */
8122 static LIST_HEAD(net_todo_list);
8123 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
8125 static void net_set_todo(struct net_device *dev)
8127 list_add_tail(&dev->todo_list, &net_todo_list);
8128 dev_net(dev)->dev_unreg_count++;
8131 static void rollback_registered_many(struct list_head *head)
8133 struct net_device *dev, *tmp;
8134 LIST_HEAD(close_head);
8136 BUG_ON(dev_boot_phase);
8139 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
8140 /* Some devices call without registering
8141 * for initialization unwind. Remove those
8142 * devices and proceed with the remaining.
8144 if (dev->reg_state == NETREG_UNINITIALIZED) {
8145 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
8149 list_del(&dev->unreg_list);
8152 dev->dismantle = true;
8153 BUG_ON(dev->reg_state != NETREG_REGISTERED);
8156 /* If device is running, close it first. */
8157 list_for_each_entry(dev, head, unreg_list)
8158 list_add_tail(&dev->close_list, &close_head);
8159 dev_close_many(&close_head, true);
8161 list_for_each_entry(dev, head, unreg_list) {
8162 /* And unlink it from device chain. */
8163 unlist_netdevice(dev);
8165 dev->reg_state = NETREG_UNREGISTERING;
8167 flush_all_backlogs();
8171 list_for_each_entry(dev, head, unreg_list) {
8172 struct sk_buff *skb = NULL;
8174 /* Shutdown queueing discipline. */
8177 dev_xdp_uninstall(dev);
8179 /* Notify protocols, that we are about to destroy
8180 * this device. They should clean all the things.
8182 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
8184 if (!dev->rtnl_link_ops ||
8185 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
8186 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
8187 GFP_KERNEL, NULL, 0);
8190 * Flush the unicast and multicast chains
8195 if (dev->netdev_ops->ndo_uninit)
8196 dev->netdev_ops->ndo_uninit(dev);
8199 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
8201 /* Notifier chain MUST detach us all upper devices. */
8202 WARN_ON(netdev_has_any_upper_dev(dev));
8203 WARN_ON(netdev_has_any_lower_dev(dev));
8205 /* Remove entries from kobject tree */
8206 netdev_unregister_kobject(dev);
8208 /* Remove XPS queueing entries */
8209 netif_reset_xps_queues_gt(dev, 0);
8215 list_for_each_entry(dev, head, unreg_list)
8219 static void rollback_registered(struct net_device *dev)
8223 list_add(&dev->unreg_list, &single);
8224 rollback_registered_many(&single);
8228 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
8229 struct net_device *upper, netdev_features_t features)
8231 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
8232 netdev_features_t feature;
8235 for_each_netdev_feature(upper_disables, feature_bit) {
8236 feature = __NETIF_F_BIT(feature_bit);
8237 if (!(upper->wanted_features & feature)
8238 && (features & feature)) {
8239 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
8240 &feature, upper->name);
8241 features &= ~feature;
8248 static void netdev_sync_lower_features(struct net_device *upper,
8249 struct net_device *lower, netdev_features_t features)
8251 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
8252 netdev_features_t feature;
8255 for_each_netdev_feature(upper_disables, feature_bit) {
8256 feature = __NETIF_F_BIT(feature_bit);
8257 if (!(features & feature) && (lower->features & feature)) {
8258 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
8259 &feature, lower->name);
8260 lower->wanted_features &= ~feature;
8261 netdev_update_features(lower);
8263 if (unlikely(lower->features & feature))
8264 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
8265 &feature, lower->name);
8270 static netdev_features_t netdev_fix_features(struct net_device *dev,
8271 netdev_features_t features)
8273 /* Fix illegal checksum combinations */
8274 if ((features & NETIF_F_HW_CSUM) &&
8275 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
8276 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
8277 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
8280 /* TSO requires that SG is present as well. */
8281 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
8282 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
8283 features &= ~NETIF_F_ALL_TSO;
8286 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
8287 !(features & NETIF_F_IP_CSUM)) {
8288 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
8289 features &= ~NETIF_F_TSO;
8290 features &= ~NETIF_F_TSO_ECN;
8293 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
8294 !(features & NETIF_F_IPV6_CSUM)) {
8295 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
8296 features &= ~NETIF_F_TSO6;
8299 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
8300 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
8301 features &= ~NETIF_F_TSO_MANGLEID;
8303 /* TSO ECN requires that TSO is present as well. */
8304 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
8305 features &= ~NETIF_F_TSO_ECN;
8307 /* Software GSO depends on SG. */
8308 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
8309 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
8310 features &= ~NETIF_F_GSO;
8313 /* GSO partial features require GSO partial be set */
8314 if ((features & dev->gso_partial_features) &&
8315 !(features & NETIF_F_GSO_PARTIAL)) {
8317 "Dropping partially supported GSO features since no GSO partial.\n");
8318 features &= ~dev->gso_partial_features;
8321 if (!(features & NETIF_F_RXCSUM)) {
8322 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet
8323 * successfully merged by hardware must also have the
8324 * checksum verified by hardware. If the user does not
8325 * want to enable RXCSUM, logically, we should disable GRO_HW.
8327 if (features & NETIF_F_GRO_HW) {
8328 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
8329 features &= ~NETIF_F_GRO_HW;
8333 /* LRO/HW-GRO features cannot be combined with RX-FCS */
8334 if (features & NETIF_F_RXFCS) {
8335 if (features & NETIF_F_LRO) {
8336 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
8337 features &= ~NETIF_F_LRO;
8340 if (features & NETIF_F_GRO_HW) {
8341 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
8342 features &= ~NETIF_F_GRO_HW;
8349 int __netdev_update_features(struct net_device *dev)
8351 struct net_device *upper, *lower;
8352 netdev_features_t features;
8353 struct list_head *iter;
8358 features = netdev_get_wanted_features(dev);
8360 if (dev->netdev_ops->ndo_fix_features)
8361 features = dev->netdev_ops->ndo_fix_features(dev, features);
8363 /* driver might be less strict about feature dependencies */
8364 features = netdev_fix_features(dev, features);
8366 /* some features can't be enabled if they're off an an upper device */
8367 netdev_for_each_upper_dev_rcu(dev, upper, iter)
8368 features = netdev_sync_upper_features(dev, upper, features);
8370 if (dev->features == features)
8373 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
8374 &dev->features, &features);
8376 if (dev->netdev_ops->ndo_set_features)
8377 err = dev->netdev_ops->ndo_set_features(dev, features);
8381 if (unlikely(err < 0)) {
8383 "set_features() failed (%d); wanted %pNF, left %pNF\n",
8384 err, &features, &dev->features);
8385 /* return non-0 since some features might have changed and
8386 * it's better to fire a spurious notification than miss it
8392 /* some features must be disabled on lower devices when disabled
8393 * on an upper device (think: bonding master or bridge)
8395 netdev_for_each_lower_dev(dev, lower, iter)
8396 netdev_sync_lower_features(dev, lower, features);
8399 netdev_features_t diff = features ^ dev->features;
8401 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
8402 /* udp_tunnel_{get,drop}_rx_info both need
8403 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
8404 * device, or they won't do anything.
8405 * Thus we need to update dev->features
8406 * *before* calling udp_tunnel_get_rx_info,
8407 * but *after* calling udp_tunnel_drop_rx_info.
8409 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
8410 dev->features = features;
8411 udp_tunnel_get_rx_info(dev);
8413 udp_tunnel_drop_rx_info(dev);
8417 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
8418 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
8419 dev->features = features;
8420 err |= vlan_get_rx_ctag_filter_info(dev);
8422 vlan_drop_rx_ctag_filter_info(dev);
8426 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
8427 if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
8428 dev->features = features;
8429 err |= vlan_get_rx_stag_filter_info(dev);
8431 vlan_drop_rx_stag_filter_info(dev);
8435 dev->features = features;
8438 return err < 0 ? 0 : 1;
8442 * netdev_update_features - recalculate device features
8443 * @dev: the device to check
8445 * Recalculate dev->features set and send notifications if it
8446 * has changed. Should be called after driver or hardware dependent
8447 * conditions might have changed that influence the features.
8449 void netdev_update_features(struct net_device *dev)
8451 if (__netdev_update_features(dev))
8452 netdev_features_change(dev);
8454 EXPORT_SYMBOL(netdev_update_features);
8457 * netdev_change_features - recalculate device features
8458 * @dev: the device to check
8460 * Recalculate dev->features set and send notifications even
8461 * if they have not changed. Should be called instead of
8462 * netdev_update_features() if also dev->vlan_features might
8463 * have changed to allow the changes to be propagated to stacked
8466 void netdev_change_features(struct net_device *dev)
8468 __netdev_update_features(dev);
8469 netdev_features_change(dev);
8471 EXPORT_SYMBOL(netdev_change_features);
8474 * netif_stacked_transfer_operstate - transfer operstate
8475 * @rootdev: the root or lower level device to transfer state from
8476 * @dev: the device to transfer operstate to
8478 * Transfer operational state from root to device. This is normally
8479 * called when a stacking relationship exists between the root
8480 * device and the device(a leaf device).
8482 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
8483 struct net_device *dev)
8485 if (rootdev->operstate == IF_OPER_DORMANT)
8486 netif_dormant_on(dev);
8488 netif_dormant_off(dev);
8490 if (netif_carrier_ok(rootdev))
8491 netif_carrier_on(dev);
8493 netif_carrier_off(dev);
8495 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
8497 static int netif_alloc_rx_queues(struct net_device *dev)
8499 unsigned int i, count = dev->num_rx_queues;
8500 struct netdev_rx_queue *rx;
8501 size_t sz = count * sizeof(*rx);
8506 rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
8512 for (i = 0; i < count; i++) {
8515 /* XDP RX-queue setup */
8516 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i);
8523 /* Rollback successful reg's and free other resources */
8525 xdp_rxq_info_unreg(&rx[i].xdp_rxq);
8531 static void netif_free_rx_queues(struct net_device *dev)
8533 unsigned int i, count = dev->num_rx_queues;
8535 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
8539 for (i = 0; i < count; i++)
8540 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
8545 static void netdev_init_one_queue(struct net_device *dev,
8546 struct netdev_queue *queue, void *_unused)
8548 /* Initialize queue lock */
8549 spin_lock_init(&queue->_xmit_lock);
8550 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
8551 queue->xmit_lock_owner = -1;
8552 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
8555 dql_init(&queue->dql, HZ);
8559 static void netif_free_tx_queues(struct net_device *dev)
8564 static int netif_alloc_netdev_queues(struct net_device *dev)
8566 unsigned int count = dev->num_tx_queues;
8567 struct netdev_queue *tx;
8568 size_t sz = count * sizeof(*tx);
8570 if (count < 1 || count > 0xffff)
8573 tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
8579 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
8580 spin_lock_init(&dev->tx_global_lock);
8585 void netif_tx_stop_all_queues(struct net_device *dev)
8589 for (i = 0; i < dev->num_tx_queues; i++) {
8590 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
8592 netif_tx_stop_queue(txq);
8595 EXPORT_SYMBOL(netif_tx_stop_all_queues);
8598 * register_netdevice - register a network device
8599 * @dev: device to register
8601 * Take a completed network device structure and add it to the kernel
8602 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
8603 * chain. 0 is returned on success. A negative errno code is returned
8604 * on a failure to set up the device, or if the name is a duplicate.
8606 * Callers must hold the rtnl semaphore. You may want
8607 * register_netdev() instead of this.
8610 * The locking appears insufficient to guarantee two parallel registers
8611 * will not get the same name.
8614 int register_netdevice(struct net_device *dev)
8617 struct net *net = dev_net(dev);
8619 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
8620 NETDEV_FEATURE_COUNT);
8621 BUG_ON(dev_boot_phase);
8626 /* When net_device's are persistent, this will be fatal. */
8627 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
8630 spin_lock_init(&dev->addr_list_lock);
8631 netdev_set_addr_lockdep_class(dev);
8633 ret = dev_get_valid_name(net, dev, dev->name);
8637 /* Init, if this function is available */
8638 if (dev->netdev_ops->ndo_init) {
8639 ret = dev->netdev_ops->ndo_init(dev);
8647 if (((dev->hw_features | dev->features) &
8648 NETIF_F_HW_VLAN_CTAG_FILTER) &&
8649 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
8650 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
8651 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
8658 dev->ifindex = dev_new_index(net);
8659 else if (__dev_get_by_index(net, dev->ifindex))
8662 /* Transfer changeable features to wanted_features and enable
8663 * software offloads (GSO and GRO).
8665 dev->hw_features |= NETIF_F_SOFT_FEATURES;
8666 dev->features |= NETIF_F_SOFT_FEATURES;
8668 if (dev->netdev_ops->ndo_udp_tunnel_add) {
8669 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
8670 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
8673 dev->wanted_features = dev->features & dev->hw_features;
8675 if (!(dev->flags & IFF_LOOPBACK))
8676 dev->hw_features |= NETIF_F_NOCACHE_COPY;
8678 /* If IPv4 TCP segmentation offload is supported we should also
8679 * allow the device to enable segmenting the frame with the option
8680 * of ignoring a static IP ID value. This doesn't enable the
8681 * feature itself but allows the user to enable it later.
8683 if (dev->hw_features & NETIF_F_TSO)
8684 dev->hw_features |= NETIF_F_TSO_MANGLEID;
8685 if (dev->vlan_features & NETIF_F_TSO)
8686 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
8687 if (dev->mpls_features & NETIF_F_TSO)
8688 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
8689 if (dev->hw_enc_features & NETIF_F_TSO)
8690 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
8692 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
8694 dev->vlan_features |= NETIF_F_HIGHDMA;
8696 /* Make NETIF_F_SG inheritable to tunnel devices.
8698 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
8700 /* Make NETIF_F_SG inheritable to MPLS.
8702 dev->mpls_features |= NETIF_F_SG;
8704 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
8705 ret = notifier_to_errno(ret);
8709 ret = netdev_register_kobject(dev);
8712 dev->reg_state = NETREG_REGISTERED;
8714 __netdev_update_features(dev);
8717 * Default initial state at registry is that the
8718 * device is present.
8721 set_bit(__LINK_STATE_PRESENT, &dev->state);
8723 linkwatch_init_dev(dev);
8725 dev_init_scheduler(dev);
8727 list_netdevice(dev);
8728 add_device_randomness(dev->dev_addr, dev->addr_len);
8730 /* If the device has permanent device address, driver should
8731 * set dev_addr and also addr_assign_type should be set to
8732 * NET_ADDR_PERM (default value).
8734 if (dev->addr_assign_type == NET_ADDR_PERM)
8735 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
8737 /* Notify protocols, that a new device appeared. */
8738 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
8739 ret = notifier_to_errno(ret);
8741 rollback_registered(dev);
8742 dev->reg_state = NETREG_UNREGISTERED;
8745 * Prevent userspace races by waiting until the network
8746 * device is fully setup before sending notifications.
8748 if (!dev->rtnl_link_ops ||
8749 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
8750 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
8756 if (dev->netdev_ops->ndo_uninit)
8757 dev->netdev_ops->ndo_uninit(dev);
8758 if (dev->priv_destructor)
8759 dev->priv_destructor(dev);
8762 EXPORT_SYMBOL(register_netdevice);
8765 * init_dummy_netdev - init a dummy network device for NAPI
8766 * @dev: device to init
8768 * This takes a network device structure and initialize the minimum
8769 * amount of fields so it can be used to schedule NAPI polls without
8770 * registering a full blown interface. This is to be used by drivers
8771 * that need to tie several hardware interfaces to a single NAPI
8772 * poll scheduler due to HW limitations.
8774 int init_dummy_netdev(struct net_device *dev)
8776 /* Clear everything. Note we don't initialize spinlocks
8777 * are they aren't supposed to be taken by any of the
8778 * NAPI code and this dummy netdev is supposed to be
8779 * only ever used for NAPI polls
8781 memset(dev, 0, sizeof(struct net_device));
8783 /* make sure we BUG if trying to hit standard
8784 * register/unregister code path
8786 dev->reg_state = NETREG_DUMMY;
8788 /* NAPI wants this */
8789 INIT_LIST_HEAD(&dev->napi_list);
8791 /* a dummy interface is started by default */
8792 set_bit(__LINK_STATE_PRESENT, &dev->state);
8793 set_bit(__LINK_STATE_START, &dev->state);
8795 /* napi_busy_loop stats accounting wants this */
8796 dev_net_set(dev, &init_net);
8798 /* Note : We dont allocate pcpu_refcnt for dummy devices,
8799 * because users of this 'device' dont need to change
8805 EXPORT_SYMBOL_GPL(init_dummy_netdev);
8809 * register_netdev - register a network device
8810 * @dev: device to register
8812 * Take a completed network device structure and add it to the kernel
8813 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
8814 * chain. 0 is returned on success. A negative errno code is returned
8815 * on a failure to set up the device, or if the name is a duplicate.
8817 * This is a wrapper around register_netdevice that takes the rtnl semaphore
8818 * and expands the device name if you passed a format string to
8821 int register_netdev(struct net_device *dev)
8825 if (rtnl_lock_killable())
8827 err = register_netdevice(dev);
8831 EXPORT_SYMBOL(register_netdev);
8833 int netdev_refcnt_read(const struct net_device *dev)
8837 for_each_possible_cpu(i)
8838 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
8841 EXPORT_SYMBOL(netdev_refcnt_read);
8844 * netdev_wait_allrefs - wait until all references are gone.
8845 * @dev: target net_device
8847 * This is called when unregistering network devices.
8849 * Any protocol or device that holds a reference should register
8850 * for netdevice notification, and cleanup and put back the
8851 * reference if they receive an UNREGISTER event.
8852 * We can get stuck here if buggy protocols don't correctly
8855 static void netdev_wait_allrefs(struct net_device *dev)
8857 unsigned long rebroadcast_time, warning_time;
8860 linkwatch_forget_dev(dev);
8862 rebroadcast_time = warning_time = jiffies;
8863 refcnt = netdev_refcnt_read(dev);
8865 while (refcnt != 0) {
8866 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
8869 /* Rebroadcast unregister notification */
8870 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
8876 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
8878 /* We must not have linkwatch events
8879 * pending on unregister. If this
8880 * happens, we simply run the queue
8881 * unscheduled, resulting in a noop
8884 linkwatch_run_queue();
8889 rebroadcast_time = jiffies;
8894 refcnt = netdev_refcnt_read(dev);
8896 if (time_after(jiffies, warning_time + 10 * HZ)) {
8897 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
8899 warning_time = jiffies;
8908 * register_netdevice(x1);
8909 * register_netdevice(x2);
8911 * unregister_netdevice(y1);
8912 * unregister_netdevice(y2);
8918 * We are invoked by rtnl_unlock().
8919 * This allows us to deal with problems:
8920 * 1) We can delete sysfs objects which invoke hotplug
8921 * without deadlocking with linkwatch via keventd.
8922 * 2) Since we run with the RTNL semaphore not held, we can sleep
8923 * safely in order to wait for the netdev refcnt to drop to zero.
8925 * We must not return until all unregister events added during
8926 * the interval the lock was held have been completed.
8928 void netdev_run_todo(void)
8930 struct list_head list;
8932 /* Snapshot list, allow later requests */
8933 list_replace_init(&net_todo_list, &list);
8938 /* Wait for rcu callbacks to finish before next phase */
8939 if (!list_empty(&list))
8942 while (!list_empty(&list)) {
8943 struct net_device *dev
8944 = list_first_entry(&list, struct net_device, todo_list);
8945 list_del(&dev->todo_list);
8947 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
8948 pr_err("network todo '%s' but state %d\n",
8949 dev->name, dev->reg_state);
8954 dev->reg_state = NETREG_UNREGISTERED;
8956 netdev_wait_allrefs(dev);
8959 BUG_ON(netdev_refcnt_read(dev));
8960 BUG_ON(!list_empty(&dev->ptype_all));
8961 BUG_ON(!list_empty(&dev->ptype_specific));
8962 WARN_ON(rcu_access_pointer(dev->ip_ptr));
8963 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
8964 #if IS_ENABLED(CONFIG_DECNET)
8965 WARN_ON(dev->dn_ptr);
8967 if (dev->priv_destructor)
8968 dev->priv_destructor(dev);
8969 if (dev->needs_free_netdev)
8972 /* Report a network device has been unregistered */
8974 dev_net(dev)->dev_unreg_count--;
8976 wake_up(&netdev_unregistering_wq);
8978 /* Free network device */
8979 kobject_put(&dev->dev.kobj);
8983 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
8984 * all the same fields in the same order as net_device_stats, with only
8985 * the type differing, but rtnl_link_stats64 may have additional fields
8986 * at the end for newer counters.
8988 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
8989 const struct net_device_stats *netdev_stats)
8991 #if BITS_PER_LONG == 64
8992 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
8993 memcpy(stats64, netdev_stats, sizeof(*netdev_stats));
8994 /* zero out counters that only exist in rtnl_link_stats64 */
8995 memset((char *)stats64 + sizeof(*netdev_stats), 0,
8996 sizeof(*stats64) - sizeof(*netdev_stats));
8998 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
8999 const unsigned long *src = (const unsigned long *)netdev_stats;
9000 u64 *dst = (u64 *)stats64;
9002 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
9003 for (i = 0; i < n; i++)
9005 /* zero out counters that only exist in rtnl_link_stats64 */
9006 memset((char *)stats64 + n * sizeof(u64), 0,
9007 sizeof(*stats64) - n * sizeof(u64));
9010 EXPORT_SYMBOL(netdev_stats_to_stats64);
9013 * dev_get_stats - get network device statistics
9014 * @dev: device to get statistics from
9015 * @storage: place to store stats
9017 * Get network statistics from device. Return @storage.
9018 * The device driver may provide its own method by setting
9019 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
9020 * otherwise the internal statistics structure is used.
9022 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
9023 struct rtnl_link_stats64 *storage)
9025 const struct net_device_ops *ops = dev->netdev_ops;
9027 if (ops->ndo_get_stats64) {
9028 memset(storage, 0, sizeof(*storage));
9029 ops->ndo_get_stats64(dev, storage);
9030 } else if (ops->ndo_get_stats) {
9031 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
9033 netdev_stats_to_stats64(storage, &dev->stats);
9035 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped);
9036 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped);
9037 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler);
9040 EXPORT_SYMBOL(dev_get_stats);
9042 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
9044 struct netdev_queue *queue = dev_ingress_queue(dev);
9046 #ifdef CONFIG_NET_CLS_ACT
9049 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
9052 netdev_init_one_queue(dev, queue, NULL);
9053 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
9054 queue->qdisc_sleeping = &noop_qdisc;
9055 rcu_assign_pointer(dev->ingress_queue, queue);
9060 static const struct ethtool_ops default_ethtool_ops;
9062 void netdev_set_default_ethtool_ops(struct net_device *dev,
9063 const struct ethtool_ops *ops)
9065 if (dev->ethtool_ops == &default_ethtool_ops)
9066 dev->ethtool_ops = ops;
9068 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
9070 void netdev_freemem(struct net_device *dev)
9072 char *addr = (char *)dev - dev->padded;
9078 * alloc_netdev_mqs - allocate network device
9079 * @sizeof_priv: size of private data to allocate space for
9080 * @name: device name format string
9081 * @name_assign_type: origin of device name
9082 * @setup: callback to initialize device
9083 * @txqs: the number of TX subqueues to allocate
9084 * @rxqs: the number of RX subqueues to allocate
9086 * Allocates a struct net_device with private data area for driver use
9087 * and performs basic initialization. Also allocates subqueue structs
9088 * for each queue on the device.
9090 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
9091 unsigned char name_assign_type,
9092 void (*setup)(struct net_device *),
9093 unsigned int txqs, unsigned int rxqs)
9095 struct net_device *dev;
9096 unsigned int alloc_size;
9097 struct net_device *p;
9099 BUG_ON(strlen(name) >= sizeof(dev->name));
9102 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
9107 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
9111 alloc_size = sizeof(struct net_device);
9113 /* ensure 32-byte alignment of private area */
9114 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
9115 alloc_size += sizeof_priv;
9117 /* ensure 32-byte alignment of whole construct */
9118 alloc_size += NETDEV_ALIGN - 1;
9120 p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
9124 dev = PTR_ALIGN(p, NETDEV_ALIGN);
9125 dev->padded = (char *)dev - (char *)p;
9127 dev->pcpu_refcnt = alloc_percpu(int);
9128 if (!dev->pcpu_refcnt)
9131 if (dev_addr_init(dev))
9137 dev_net_set(dev, &init_net);
9139 dev->gso_max_size = GSO_MAX_SIZE;
9140 dev->gso_max_segs = GSO_MAX_SEGS;
9142 INIT_LIST_HEAD(&dev->napi_list);
9143 INIT_LIST_HEAD(&dev->unreg_list);
9144 INIT_LIST_HEAD(&dev->close_list);
9145 INIT_LIST_HEAD(&dev->link_watch_list);
9146 INIT_LIST_HEAD(&dev->adj_list.upper);
9147 INIT_LIST_HEAD(&dev->adj_list.lower);
9148 INIT_LIST_HEAD(&dev->ptype_all);
9149 INIT_LIST_HEAD(&dev->ptype_specific);
9150 #ifdef CONFIG_NET_SCHED
9151 hash_init(dev->qdisc_hash);
9153 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
9156 if (!dev->tx_queue_len) {
9157 dev->priv_flags |= IFF_NO_QUEUE;
9158 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
9161 dev->num_tx_queues = txqs;
9162 dev->real_num_tx_queues = txqs;
9163 if (netif_alloc_netdev_queues(dev))
9166 dev->num_rx_queues = rxqs;
9167 dev->real_num_rx_queues = rxqs;
9168 if (netif_alloc_rx_queues(dev))
9171 strcpy(dev->name, name);
9172 dev->name_assign_type = name_assign_type;
9173 dev->group = INIT_NETDEV_GROUP;
9174 if (!dev->ethtool_ops)
9175 dev->ethtool_ops = &default_ethtool_ops;
9177 nf_hook_ingress_init(dev);
9186 free_percpu(dev->pcpu_refcnt);
9188 netdev_freemem(dev);
9191 EXPORT_SYMBOL(alloc_netdev_mqs);
9194 * free_netdev - free network device
9197 * This function does the last stage of destroying an allocated device
9198 * interface. The reference to the device object is released. If this
9199 * is the last reference then it will be freed.Must be called in process
9202 void free_netdev(struct net_device *dev)
9204 struct napi_struct *p, *n;
9207 netif_free_tx_queues(dev);
9208 netif_free_rx_queues(dev);
9210 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
9212 /* Flush device addresses */
9213 dev_addr_flush(dev);
9215 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
9218 free_percpu(dev->pcpu_refcnt);
9219 dev->pcpu_refcnt = NULL;
9221 /* Compatibility with error handling in drivers */
9222 if (dev->reg_state == NETREG_UNINITIALIZED) {
9223 netdev_freemem(dev);
9227 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
9228 dev->reg_state = NETREG_RELEASED;
9230 /* will free via device release */
9231 put_device(&dev->dev);
9233 EXPORT_SYMBOL(free_netdev);
9236 * synchronize_net - Synchronize with packet receive processing
9238 * Wait for packets currently being received to be done.
9239 * Does not block later packets from starting.
9241 void synchronize_net(void)
9244 if (rtnl_is_locked())
9245 synchronize_rcu_expedited();
9249 EXPORT_SYMBOL(synchronize_net);
9252 * unregister_netdevice_queue - remove device from the kernel
9256 * This function shuts down a device interface and removes it
9257 * from the kernel tables.
9258 * If head not NULL, device is queued to be unregistered later.
9260 * Callers must hold the rtnl semaphore. You may want
9261 * unregister_netdev() instead of this.
9264 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
9269 list_move_tail(&dev->unreg_list, head);
9271 rollback_registered(dev);
9272 /* Finish processing unregister after unlock */
9276 EXPORT_SYMBOL(unregister_netdevice_queue);
9279 * unregister_netdevice_many - unregister many devices
9280 * @head: list of devices
9282 * Note: As most callers use a stack allocated list_head,
9283 * we force a list_del() to make sure stack wont be corrupted later.
9285 void unregister_netdevice_many(struct list_head *head)
9287 struct net_device *dev;
9289 if (!list_empty(head)) {
9290 rollback_registered_many(head);
9291 list_for_each_entry(dev, head, unreg_list)
9296 EXPORT_SYMBOL(unregister_netdevice_many);
9299 * unregister_netdev - remove device from the kernel
9302 * This function shuts down a device interface and removes it
9303 * from the kernel tables.
9305 * This is just a wrapper for unregister_netdevice that takes
9306 * the rtnl semaphore. In general you want to use this and not
9307 * unregister_netdevice.
9309 void unregister_netdev(struct net_device *dev)
9312 unregister_netdevice(dev);
9315 EXPORT_SYMBOL(unregister_netdev);
9318 * dev_change_net_namespace - move device to different nethost namespace
9320 * @net: network namespace
9321 * @pat: If not NULL name pattern to try if the current device name
9322 * is already taken in the destination network namespace.
9324 * This function shuts down a device interface and moves it
9325 * to a new network namespace. On success 0 is returned, on
9326 * a failure a netagive errno code is returned.
9328 * Callers must hold the rtnl semaphore.
9331 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
9333 int err, new_nsid, new_ifindex;
9337 /* Don't allow namespace local devices to be moved. */
9339 if (dev->features & NETIF_F_NETNS_LOCAL)
9342 /* Ensure the device has been registrered */
9343 if (dev->reg_state != NETREG_REGISTERED)
9346 /* Get out if there is nothing todo */
9348 if (net_eq(dev_net(dev), net))
9351 /* Pick the destination device name, and ensure
9352 * we can use it in the destination network namespace.
9355 if (__dev_get_by_name(net, dev->name)) {
9356 /* We get here if we can't use the current device name */
9359 err = dev_get_valid_name(net, dev, pat);
9365 * And now a mini version of register_netdevice unregister_netdevice.
9368 /* If device is running close it first. */
9371 /* And unlink it from device chain */
9372 unlist_netdevice(dev);
9376 /* Shutdown queueing discipline. */
9379 /* Notify protocols, that we are about to destroy
9380 * this device. They should clean all the things.
9382 * Note that dev->reg_state stays at NETREG_REGISTERED.
9383 * This is wanted because this way 8021q and macvlan know
9384 * the device is just moving and can keep their slaves up.
9386 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
9389 new_nsid = peernet2id_alloc(dev_net(dev), net);
9390 /* If there is an ifindex conflict assign a new one */
9391 if (__dev_get_by_index(net, dev->ifindex))
9392 new_ifindex = dev_new_index(net);
9394 new_ifindex = dev->ifindex;
9396 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
9400 * Flush the unicast and multicast chains
9405 /* Send a netdev-removed uevent to the old namespace */
9406 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
9407 netdev_adjacent_del_links(dev);
9409 /* Actually switch the network namespace */
9410 dev_net_set(dev, net);
9411 dev->ifindex = new_ifindex;
9413 /* Send a netdev-add uevent to the new namespace */
9414 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
9415 netdev_adjacent_add_links(dev);
9417 /* Fixup kobjects */
9418 err = device_rename(&dev->dev, dev->name);
9421 /* Add the device back in the hashes */
9422 list_netdevice(dev);
9424 /* Notify protocols, that a new device appeared. */
9425 call_netdevice_notifiers(NETDEV_REGISTER, dev);
9428 * Prevent userspace races by waiting until the network
9429 * device is fully setup before sending notifications.
9431 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
9438 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
9440 static int dev_cpu_dead(unsigned int oldcpu)
9442 struct sk_buff **list_skb;
9443 struct sk_buff *skb;
9445 struct softnet_data *sd, *oldsd, *remsd = NULL;
9447 local_irq_disable();
9448 cpu = smp_processor_id();
9449 sd = &per_cpu(softnet_data, cpu);
9450 oldsd = &per_cpu(softnet_data, oldcpu);
9452 /* Find end of our completion_queue. */
9453 list_skb = &sd->completion_queue;
9455 list_skb = &(*list_skb)->next;
9456 /* Append completion queue from offline CPU. */
9457 *list_skb = oldsd->completion_queue;
9458 oldsd->completion_queue = NULL;
9460 /* Append output queue from offline CPU. */
9461 if (oldsd->output_queue) {
9462 *sd->output_queue_tailp = oldsd->output_queue;
9463 sd->output_queue_tailp = oldsd->output_queue_tailp;
9464 oldsd->output_queue = NULL;
9465 oldsd->output_queue_tailp = &oldsd->output_queue;
9467 /* Append NAPI poll list from offline CPU, with one exception :
9468 * process_backlog() must be called by cpu owning percpu backlog.
9469 * We properly handle process_queue & input_pkt_queue later.
9471 while (!list_empty(&oldsd->poll_list)) {
9472 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
9476 list_del_init(&napi->poll_list);
9477 if (napi->poll == process_backlog)
9480 ____napi_schedule(sd, napi);
9483 raise_softirq_irqoff(NET_TX_SOFTIRQ);
9487 remsd = oldsd->rps_ipi_list;
9488 oldsd->rps_ipi_list = NULL;
9490 /* send out pending IPI's on offline CPU */
9491 net_rps_send_ipi(remsd);
9493 /* Process offline CPU's input_pkt_queue */
9494 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
9496 input_queue_head_incr(oldsd);
9498 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
9500 input_queue_head_incr(oldsd);
9507 * netdev_increment_features - increment feature set by one
9508 * @all: current feature set
9509 * @one: new feature set
9510 * @mask: mask feature set
9512 * Computes a new feature set after adding a device with feature set
9513 * @one to the master device with current feature set @all. Will not
9514 * enable anything that is off in @mask. Returns the new feature set.
9516 netdev_features_t netdev_increment_features(netdev_features_t all,
9517 netdev_features_t one, netdev_features_t mask)
9519 if (mask & NETIF_F_HW_CSUM)
9520 mask |= NETIF_F_CSUM_MASK;
9521 mask |= NETIF_F_VLAN_CHALLENGED;
9523 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
9524 all &= one | ~NETIF_F_ALL_FOR_ALL;
9526 /* If one device supports hw checksumming, set for all. */
9527 if (all & NETIF_F_HW_CSUM)
9528 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
9532 EXPORT_SYMBOL(netdev_increment_features);
9534 static struct hlist_head * __net_init netdev_create_hash(void)
9537 struct hlist_head *hash;
9539 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
9541 for (i = 0; i < NETDEV_HASHENTRIES; i++)
9542 INIT_HLIST_HEAD(&hash[i]);
9547 /* Initialize per network namespace state */
9548 static int __net_init netdev_init(struct net *net)
9550 BUILD_BUG_ON(GRO_HASH_BUCKETS >
9551 8 * FIELD_SIZEOF(struct napi_struct, gro_bitmask));
9553 if (net != &init_net)
9554 INIT_LIST_HEAD(&net->dev_base_head);
9556 net->dev_name_head = netdev_create_hash();
9557 if (net->dev_name_head == NULL)
9560 net->dev_index_head = netdev_create_hash();
9561 if (net->dev_index_head == NULL)
9567 kfree(net->dev_name_head);
9573 * netdev_drivername - network driver for the device
9574 * @dev: network device
9576 * Determine network driver for device.
9578 const char *netdev_drivername(const struct net_device *dev)
9580 const struct device_driver *driver;
9581 const struct device *parent;
9582 const char *empty = "";
9584 parent = dev->dev.parent;
9588 driver = parent->driver;
9589 if (driver && driver->name)
9590 return driver->name;
9594 static void __netdev_printk(const char *level, const struct net_device *dev,
9595 struct va_format *vaf)
9597 if (dev && dev->dev.parent) {
9598 dev_printk_emit(level[1] - '0',
9601 dev_driver_string(dev->dev.parent),
9602 dev_name(dev->dev.parent),
9603 netdev_name(dev), netdev_reg_state(dev),
9606 printk("%s%s%s: %pV",
9607 level, netdev_name(dev), netdev_reg_state(dev), vaf);
9609 printk("%s(NULL net_device): %pV", level, vaf);
9613 void netdev_printk(const char *level, const struct net_device *dev,
9614 const char *format, ...)
9616 struct va_format vaf;
9619 va_start(args, format);
9624 __netdev_printk(level, dev, &vaf);
9628 EXPORT_SYMBOL(netdev_printk);
9630 #define define_netdev_printk_level(func, level) \
9631 void func(const struct net_device *dev, const char *fmt, ...) \
9633 struct va_format vaf; \
9636 va_start(args, fmt); \
9641 __netdev_printk(level, dev, &vaf); \
9645 EXPORT_SYMBOL(func);
9647 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
9648 define_netdev_printk_level(netdev_alert, KERN_ALERT);
9649 define_netdev_printk_level(netdev_crit, KERN_CRIT);
9650 define_netdev_printk_level(netdev_err, KERN_ERR);
9651 define_netdev_printk_level(netdev_warn, KERN_WARNING);
9652 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
9653 define_netdev_printk_level(netdev_info, KERN_INFO);
9655 static void __net_exit netdev_exit(struct net *net)
9657 kfree(net->dev_name_head);
9658 kfree(net->dev_index_head);
9659 if (net != &init_net)
9660 WARN_ON_ONCE(!list_empty(&net->dev_base_head));
9663 static struct pernet_operations __net_initdata netdev_net_ops = {
9664 .init = netdev_init,
9665 .exit = netdev_exit,
9668 static void __net_exit default_device_exit(struct net *net)
9670 struct net_device *dev, *aux;
9672 * Push all migratable network devices back to the
9673 * initial network namespace
9676 for_each_netdev_safe(net, dev, aux) {
9678 char fb_name[IFNAMSIZ];
9680 /* Ignore unmoveable devices (i.e. loopback) */
9681 if (dev->features & NETIF_F_NETNS_LOCAL)
9684 /* Leave virtual devices for the generic cleanup */
9685 if (dev->rtnl_link_ops)
9688 /* Push remaining network devices to init_net */
9689 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
9690 err = dev_change_net_namespace(dev, &init_net, fb_name);
9692 pr_emerg("%s: failed to move %s to init_net: %d\n",
9693 __func__, dev->name, err);
9700 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
9702 /* Return with the rtnl_lock held when there are no network
9703 * devices unregistering in any network namespace in net_list.
9707 DEFINE_WAIT_FUNC(wait, woken_wake_function);
9709 add_wait_queue(&netdev_unregistering_wq, &wait);
9711 unregistering = false;
9713 list_for_each_entry(net, net_list, exit_list) {
9714 if (net->dev_unreg_count > 0) {
9715 unregistering = true;
9723 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
9725 remove_wait_queue(&netdev_unregistering_wq, &wait);
9728 static void __net_exit default_device_exit_batch(struct list_head *net_list)
9730 /* At exit all network devices most be removed from a network
9731 * namespace. Do this in the reverse order of registration.
9732 * Do this across as many network namespaces as possible to
9733 * improve batching efficiency.
9735 struct net_device *dev;
9737 LIST_HEAD(dev_kill_list);
9739 /* To prevent network device cleanup code from dereferencing
9740 * loopback devices or network devices that have been freed
9741 * wait here for all pending unregistrations to complete,
9742 * before unregistring the loopback device and allowing the
9743 * network namespace be freed.
9745 * The netdev todo list containing all network devices
9746 * unregistrations that happen in default_device_exit_batch
9747 * will run in the rtnl_unlock() at the end of
9748 * default_device_exit_batch.
9750 rtnl_lock_unregistering(net_list);
9751 list_for_each_entry(net, net_list, exit_list) {
9752 for_each_netdev_reverse(net, dev) {
9753 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
9754 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
9756 unregister_netdevice_queue(dev, &dev_kill_list);
9759 unregister_netdevice_many(&dev_kill_list);
9763 static struct pernet_operations __net_initdata default_device_ops = {
9764 .exit = default_device_exit,
9765 .exit_batch = default_device_exit_batch,
9769 * Initialize the DEV module. At boot time this walks the device list and
9770 * unhooks any devices that fail to initialise (normally hardware not
9771 * present) and leaves us with a valid list of present and active devices.
9776 * This is called single threaded during boot, so no need
9777 * to take the rtnl semaphore.
9779 static int __init net_dev_init(void)
9781 int i, rc = -ENOMEM;
9783 BUG_ON(!dev_boot_phase);
9785 if (dev_proc_init())
9788 if (netdev_kobject_init())
9791 INIT_LIST_HEAD(&ptype_all);
9792 for (i = 0; i < PTYPE_HASH_SIZE; i++)
9793 INIT_LIST_HEAD(&ptype_base[i]);
9795 INIT_LIST_HEAD(&offload_base);
9797 if (register_pernet_subsys(&netdev_net_ops))
9801 * Initialise the packet receive queues.
9804 for_each_possible_cpu(i) {
9805 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
9806 struct softnet_data *sd = &per_cpu(softnet_data, i);
9808 INIT_WORK(flush, flush_backlog);
9810 skb_queue_head_init(&sd->input_pkt_queue);
9811 skb_queue_head_init(&sd->process_queue);
9812 #ifdef CONFIG_XFRM_OFFLOAD
9813 skb_queue_head_init(&sd->xfrm_backlog);
9815 INIT_LIST_HEAD(&sd->poll_list);
9816 sd->output_queue_tailp = &sd->output_queue;
9818 sd->csd.func = rps_trigger_softirq;
9823 init_gro_hash(&sd->backlog);
9824 sd->backlog.poll = process_backlog;
9825 sd->backlog.weight = weight_p;
9830 /* The loopback device is special if any other network devices
9831 * is present in a network namespace the loopback device must
9832 * be present. Since we now dynamically allocate and free the
9833 * loopback device ensure this invariant is maintained by
9834 * keeping the loopback device as the first device on the
9835 * list of network devices. Ensuring the loopback devices
9836 * is the first device that appears and the last network device
9839 if (register_pernet_device(&loopback_net_ops))
9842 if (register_pernet_device(&default_device_ops))
9845 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
9846 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
9848 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
9849 NULL, dev_cpu_dead);
9856 subsys_initcall(net_dev_init);