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/inetdevice.h>
135 #include <linux/cpu_rmap.h>
136 #include <linux/static_key.h>
137 #include <linux/hashtable.h>
138 #include <linux/vmalloc.h>
139 #include <linux/if_macvlan.h>
140 #include <linux/errqueue.h>
141 #include <linux/hrtimer.h>
142 #include <linux/netfilter_ingress.h>
143 #include <linux/crash_dump.h>
144 #include <linux/sctp.h>
145 #include <net/udp_tunnel.h>
146 #include <linux/net_namespace.h>
147 #include <linux/indirect_call_wrapper.h>
148 #include <net/devlink.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));
1188 /* Some auto-enslaved devices e.g. failover slaves are
1189 * special, as userspace might rename the device after
1190 * the interface had been brought up and running since
1191 * the point kernel initiated auto-enslavement. Allow
1192 * live name change even when these slave devices are
1195 * Typically, users of these auto-enslaving devices
1196 * don't actually care about slave name change, as
1197 * they are supposed to operate on master interface
1200 if (dev->flags & IFF_UP &&
1201 likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK)))
1204 write_seqcount_begin(&devnet_rename_seq);
1206 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1207 write_seqcount_end(&devnet_rename_seq);
1211 memcpy(oldname, dev->name, IFNAMSIZ);
1213 err = dev_get_valid_name(net, dev, newname);
1215 write_seqcount_end(&devnet_rename_seq);
1219 if (oldname[0] && !strchr(oldname, '%'))
1220 netdev_info(dev, "renamed from %s\n", oldname);
1222 old_assign_type = dev->name_assign_type;
1223 dev->name_assign_type = NET_NAME_RENAMED;
1226 ret = device_rename(&dev->dev, dev->name);
1228 memcpy(dev->name, oldname, IFNAMSIZ);
1229 dev->name_assign_type = old_assign_type;
1230 write_seqcount_end(&devnet_rename_seq);
1234 write_seqcount_end(&devnet_rename_seq);
1236 netdev_adjacent_rename_links(dev, oldname);
1238 write_lock_bh(&dev_base_lock);
1239 hlist_del_rcu(&dev->name_hlist);
1240 write_unlock_bh(&dev_base_lock);
1244 write_lock_bh(&dev_base_lock);
1245 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1246 write_unlock_bh(&dev_base_lock);
1248 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1249 ret = notifier_to_errno(ret);
1252 /* err >= 0 after dev_alloc_name() or stores the first errno */
1255 write_seqcount_begin(&devnet_rename_seq);
1256 memcpy(dev->name, oldname, IFNAMSIZ);
1257 memcpy(oldname, newname, IFNAMSIZ);
1258 dev->name_assign_type = old_assign_type;
1259 old_assign_type = NET_NAME_RENAMED;
1262 pr_err("%s: name change rollback failed: %d\n",
1271 * dev_set_alias - change ifalias of a device
1273 * @alias: name up to IFALIASZ
1274 * @len: limit of bytes to copy from info
1276 * Set ifalias for a device,
1278 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1280 struct dev_ifalias *new_alias = NULL;
1282 if (len >= IFALIASZ)
1286 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1290 memcpy(new_alias->ifalias, alias, len);
1291 new_alias->ifalias[len] = 0;
1294 mutex_lock(&ifalias_mutex);
1295 rcu_swap_protected(dev->ifalias, new_alias,
1296 mutex_is_locked(&ifalias_mutex));
1297 mutex_unlock(&ifalias_mutex);
1300 kfree_rcu(new_alias, rcuhead);
1304 EXPORT_SYMBOL(dev_set_alias);
1307 * dev_get_alias - get ifalias of a device
1309 * @name: buffer to store name of ifalias
1310 * @len: size of buffer
1312 * get ifalias for a device. Caller must make sure dev cannot go
1313 * away, e.g. rcu read lock or own a reference count to device.
1315 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1317 const struct dev_ifalias *alias;
1321 alias = rcu_dereference(dev->ifalias);
1323 ret = snprintf(name, len, "%s", alias->ifalias);
1330 * netdev_features_change - device changes features
1331 * @dev: device to cause notification
1333 * Called to indicate a device has changed features.
1335 void netdev_features_change(struct net_device *dev)
1337 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1339 EXPORT_SYMBOL(netdev_features_change);
1342 * netdev_state_change - device changes state
1343 * @dev: device to cause notification
1345 * Called to indicate a device has changed state. This function calls
1346 * the notifier chains for netdev_chain and sends a NEWLINK message
1347 * to the routing socket.
1349 void netdev_state_change(struct net_device *dev)
1351 if (dev->flags & IFF_UP) {
1352 struct netdev_notifier_change_info change_info = {
1356 call_netdevice_notifiers_info(NETDEV_CHANGE,
1358 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1361 EXPORT_SYMBOL(netdev_state_change);
1364 * netdev_notify_peers - notify network peers about existence of @dev
1365 * @dev: network device
1367 * Generate traffic such that interested network peers are aware of
1368 * @dev, such as by generating a gratuitous ARP. This may be used when
1369 * a device wants to inform the rest of the network about some sort of
1370 * reconfiguration such as a failover event or virtual machine
1373 void netdev_notify_peers(struct net_device *dev)
1376 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1377 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1380 EXPORT_SYMBOL(netdev_notify_peers);
1382 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1384 const struct net_device_ops *ops = dev->netdev_ops;
1389 if (!netif_device_present(dev))
1392 /* Block netpoll from trying to do any rx path servicing.
1393 * If we don't do this there is a chance ndo_poll_controller
1394 * or ndo_poll may be running while we open the device
1396 netpoll_poll_disable(dev);
1398 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack);
1399 ret = notifier_to_errno(ret);
1403 set_bit(__LINK_STATE_START, &dev->state);
1405 if (ops->ndo_validate_addr)
1406 ret = ops->ndo_validate_addr(dev);
1408 if (!ret && ops->ndo_open)
1409 ret = ops->ndo_open(dev);
1411 netpoll_poll_enable(dev);
1414 clear_bit(__LINK_STATE_START, &dev->state);
1416 dev->flags |= IFF_UP;
1417 dev_set_rx_mode(dev);
1419 add_device_randomness(dev->dev_addr, dev->addr_len);
1426 * dev_open - prepare an interface for use.
1427 * @dev: device to open
1428 * @extack: netlink extended ack
1430 * Takes a device from down to up state. The device's private open
1431 * function is invoked and then the multicast lists are loaded. Finally
1432 * the device is moved into the up state and a %NETDEV_UP message is
1433 * sent to the netdev notifier chain.
1435 * Calling this function on an active interface is a nop. On a failure
1436 * a negative errno code is returned.
1438 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1442 if (dev->flags & IFF_UP)
1445 ret = __dev_open(dev, extack);
1449 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1450 call_netdevice_notifiers(NETDEV_UP, dev);
1454 EXPORT_SYMBOL(dev_open);
1456 static void __dev_close_many(struct list_head *head)
1458 struct net_device *dev;
1463 list_for_each_entry(dev, head, close_list) {
1464 /* Temporarily disable netpoll until the interface is down */
1465 netpoll_poll_disable(dev);
1467 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1469 clear_bit(__LINK_STATE_START, &dev->state);
1471 /* Synchronize to scheduled poll. We cannot touch poll list, it
1472 * can be even on different cpu. So just clear netif_running().
1474 * dev->stop() will invoke napi_disable() on all of it's
1475 * napi_struct instances on this device.
1477 smp_mb__after_atomic(); /* Commit netif_running(). */
1480 dev_deactivate_many(head);
1482 list_for_each_entry(dev, head, close_list) {
1483 const struct net_device_ops *ops = dev->netdev_ops;
1486 * Call the device specific close. This cannot fail.
1487 * Only if device is UP
1489 * We allow it to be called even after a DETACH hot-plug
1495 dev->flags &= ~IFF_UP;
1496 netpoll_poll_enable(dev);
1500 static void __dev_close(struct net_device *dev)
1504 list_add(&dev->close_list, &single);
1505 __dev_close_many(&single);
1509 void dev_close_many(struct list_head *head, bool unlink)
1511 struct net_device *dev, *tmp;
1513 /* Remove the devices that don't need to be closed */
1514 list_for_each_entry_safe(dev, tmp, head, close_list)
1515 if (!(dev->flags & IFF_UP))
1516 list_del_init(&dev->close_list);
1518 __dev_close_many(head);
1520 list_for_each_entry_safe(dev, tmp, head, close_list) {
1521 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1522 call_netdevice_notifiers(NETDEV_DOWN, dev);
1524 list_del_init(&dev->close_list);
1527 EXPORT_SYMBOL(dev_close_many);
1530 * dev_close - shutdown an interface.
1531 * @dev: device to shutdown
1533 * This function moves an active device into down state. A
1534 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1535 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1538 void dev_close(struct net_device *dev)
1540 if (dev->flags & IFF_UP) {
1543 list_add(&dev->close_list, &single);
1544 dev_close_many(&single, true);
1548 EXPORT_SYMBOL(dev_close);
1552 * dev_disable_lro - disable Large Receive Offload on a device
1555 * Disable Large Receive Offload (LRO) on a net device. Must be
1556 * called under RTNL. This is needed if received packets may be
1557 * forwarded to another interface.
1559 void dev_disable_lro(struct net_device *dev)
1561 struct net_device *lower_dev;
1562 struct list_head *iter;
1564 dev->wanted_features &= ~NETIF_F_LRO;
1565 netdev_update_features(dev);
1567 if (unlikely(dev->features & NETIF_F_LRO))
1568 netdev_WARN(dev, "failed to disable LRO!\n");
1570 netdev_for_each_lower_dev(dev, lower_dev, iter)
1571 dev_disable_lro(lower_dev);
1573 EXPORT_SYMBOL(dev_disable_lro);
1576 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1579 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be
1580 * called under RTNL. This is needed if Generic XDP is installed on
1583 static void dev_disable_gro_hw(struct net_device *dev)
1585 dev->wanted_features &= ~NETIF_F_GRO_HW;
1586 netdev_update_features(dev);
1588 if (unlikely(dev->features & NETIF_F_GRO_HW))
1589 netdev_WARN(dev, "failed to disable GRO_HW!\n");
1592 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1595 case NETDEV_##val: \
1596 return "NETDEV_" __stringify(val);
1598 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1599 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1600 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1601 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER)
1602 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO)
1603 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO)
1604 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1605 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1606 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1610 return "UNKNOWN_NETDEV_EVENT";
1612 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1614 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1615 struct net_device *dev)
1617 struct netdev_notifier_info info = {
1621 return nb->notifier_call(nb, val, &info);
1624 static int dev_boot_phase = 1;
1627 * register_netdevice_notifier - register a network notifier block
1630 * Register a notifier to be called when network device events occur.
1631 * The notifier passed is linked into the kernel structures and must
1632 * not be reused until it has been unregistered. A negative errno code
1633 * is returned on a failure.
1635 * When registered all registration and up events are replayed
1636 * to the new notifier to allow device to have a race free
1637 * view of the network device list.
1640 int register_netdevice_notifier(struct notifier_block *nb)
1642 struct net_device *dev;
1643 struct net_device *last;
1647 /* Close race with setup_net() and cleanup_net() */
1648 down_write(&pernet_ops_rwsem);
1650 err = raw_notifier_chain_register(&netdev_chain, nb);
1656 for_each_netdev(net, dev) {
1657 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1658 err = notifier_to_errno(err);
1662 if (!(dev->flags & IFF_UP))
1665 call_netdevice_notifier(nb, NETDEV_UP, dev);
1671 up_write(&pernet_ops_rwsem);
1677 for_each_netdev(net, dev) {
1681 if (dev->flags & IFF_UP) {
1682 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1684 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1686 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1691 raw_notifier_chain_unregister(&netdev_chain, nb);
1694 EXPORT_SYMBOL(register_netdevice_notifier);
1697 * unregister_netdevice_notifier - unregister a network notifier block
1700 * Unregister a notifier previously registered by
1701 * register_netdevice_notifier(). The notifier is unlinked into the
1702 * kernel structures and may then be reused. A negative errno code
1703 * is returned on a failure.
1705 * After unregistering unregister and down device events are synthesized
1706 * for all devices on the device list to the removed notifier to remove
1707 * the need for special case cleanup code.
1710 int unregister_netdevice_notifier(struct notifier_block *nb)
1712 struct net_device *dev;
1716 /* Close race with setup_net() and cleanup_net() */
1717 down_write(&pernet_ops_rwsem);
1719 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1724 for_each_netdev(net, dev) {
1725 if (dev->flags & IFF_UP) {
1726 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1728 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1730 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1735 up_write(&pernet_ops_rwsem);
1738 EXPORT_SYMBOL(unregister_netdevice_notifier);
1741 * call_netdevice_notifiers_info - call all network notifier blocks
1742 * @val: value passed unmodified to notifier function
1743 * @info: notifier information data
1745 * Call all network notifier blocks. Parameters and return value
1746 * are as for raw_notifier_call_chain().
1749 static int call_netdevice_notifiers_info(unsigned long val,
1750 struct netdev_notifier_info *info)
1753 return raw_notifier_call_chain(&netdev_chain, val, info);
1756 static int call_netdevice_notifiers_extack(unsigned long val,
1757 struct net_device *dev,
1758 struct netlink_ext_ack *extack)
1760 struct netdev_notifier_info info = {
1765 return call_netdevice_notifiers_info(val, &info);
1769 * call_netdevice_notifiers - call all network notifier blocks
1770 * @val: value passed unmodified to notifier function
1771 * @dev: net_device pointer passed unmodified to notifier function
1773 * Call all network notifier blocks. Parameters and return value
1774 * are as for raw_notifier_call_chain().
1777 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1779 return call_netdevice_notifiers_extack(val, dev, NULL);
1781 EXPORT_SYMBOL(call_netdevice_notifiers);
1784 * call_netdevice_notifiers_mtu - call all network notifier blocks
1785 * @val: value passed unmodified to notifier function
1786 * @dev: net_device pointer passed unmodified to notifier function
1787 * @arg: additional u32 argument passed to the notifier function
1789 * Call all network notifier blocks. Parameters and return value
1790 * are as for raw_notifier_call_chain().
1792 static int call_netdevice_notifiers_mtu(unsigned long val,
1793 struct net_device *dev, u32 arg)
1795 struct netdev_notifier_info_ext info = {
1800 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
1802 return call_netdevice_notifiers_info(val, &info.info);
1805 #ifdef CONFIG_NET_INGRESS
1806 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
1808 void net_inc_ingress_queue(void)
1810 static_branch_inc(&ingress_needed_key);
1812 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
1814 void net_dec_ingress_queue(void)
1816 static_branch_dec(&ingress_needed_key);
1818 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
1821 #ifdef CONFIG_NET_EGRESS
1822 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
1824 void net_inc_egress_queue(void)
1826 static_branch_inc(&egress_needed_key);
1828 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
1830 void net_dec_egress_queue(void)
1832 static_branch_dec(&egress_needed_key);
1834 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
1837 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
1838 #ifdef CONFIG_JUMP_LABEL
1839 static atomic_t netstamp_needed_deferred;
1840 static atomic_t netstamp_wanted;
1841 static void netstamp_clear(struct work_struct *work)
1843 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1846 wanted = atomic_add_return(deferred, &netstamp_wanted);
1848 static_branch_enable(&netstamp_needed_key);
1850 static_branch_disable(&netstamp_needed_key);
1852 static DECLARE_WORK(netstamp_work, netstamp_clear);
1855 void net_enable_timestamp(void)
1857 #ifdef CONFIG_JUMP_LABEL
1861 wanted = atomic_read(&netstamp_wanted);
1864 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted)
1867 atomic_inc(&netstamp_needed_deferred);
1868 schedule_work(&netstamp_work);
1870 static_branch_inc(&netstamp_needed_key);
1873 EXPORT_SYMBOL(net_enable_timestamp);
1875 void net_disable_timestamp(void)
1877 #ifdef CONFIG_JUMP_LABEL
1881 wanted = atomic_read(&netstamp_wanted);
1884 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted)
1887 atomic_dec(&netstamp_needed_deferred);
1888 schedule_work(&netstamp_work);
1890 static_branch_dec(&netstamp_needed_key);
1893 EXPORT_SYMBOL(net_disable_timestamp);
1895 static inline void net_timestamp_set(struct sk_buff *skb)
1898 if (static_branch_unlikely(&netstamp_needed_key))
1899 __net_timestamp(skb);
1902 #define net_timestamp_check(COND, SKB) \
1903 if (static_branch_unlikely(&netstamp_needed_key)) { \
1904 if ((COND) && !(SKB)->tstamp) \
1905 __net_timestamp(SKB); \
1908 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
1912 if (!(dev->flags & IFF_UP))
1915 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1916 if (skb->len <= len)
1919 /* if TSO is enabled, we don't care about the length as the packet
1920 * could be forwarded without being segmented before
1922 if (skb_is_gso(skb))
1927 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1929 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1931 int ret = ____dev_forward_skb(dev, skb);
1934 skb->protocol = eth_type_trans(skb, dev);
1935 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1940 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1943 * dev_forward_skb - loopback an skb to another netif
1945 * @dev: destination network device
1946 * @skb: buffer to forward
1949 * NET_RX_SUCCESS (no congestion)
1950 * NET_RX_DROP (packet was dropped, but freed)
1952 * dev_forward_skb can be used for injecting an skb from the
1953 * start_xmit function of one device into the receive queue
1954 * of another device.
1956 * The receiving device may be in another namespace, so
1957 * we have to clear all information in the skb that could
1958 * impact namespace isolation.
1960 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1962 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1964 EXPORT_SYMBOL_GPL(dev_forward_skb);
1966 static inline int deliver_skb(struct sk_buff *skb,
1967 struct packet_type *pt_prev,
1968 struct net_device *orig_dev)
1970 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1972 refcount_inc(&skb->users);
1973 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1976 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1977 struct packet_type **pt,
1978 struct net_device *orig_dev,
1980 struct list_head *ptype_list)
1982 struct packet_type *ptype, *pt_prev = *pt;
1984 list_for_each_entry_rcu(ptype, ptype_list, list) {
1985 if (ptype->type != type)
1988 deliver_skb(skb, pt_prev, orig_dev);
1994 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1996 if (!ptype->af_packet_priv || !skb->sk)
1999 if (ptype->id_match)
2000 return ptype->id_match(ptype, skb->sk);
2001 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
2008 * dev_nit_active - return true if any network interface taps are in use
2010 * @dev: network device to check for the presence of taps
2012 bool dev_nit_active(struct net_device *dev)
2014 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all);
2016 EXPORT_SYMBOL_GPL(dev_nit_active);
2019 * Support routine. Sends outgoing frames to any network
2020 * taps currently in use.
2023 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
2025 struct packet_type *ptype;
2026 struct sk_buff *skb2 = NULL;
2027 struct packet_type *pt_prev = NULL;
2028 struct list_head *ptype_list = &ptype_all;
2032 list_for_each_entry_rcu(ptype, ptype_list, list) {
2033 if (ptype->ignore_outgoing)
2036 /* Never send packets back to the socket
2037 * they originated from - MvS (miquels@drinkel.ow.org)
2039 if (skb_loop_sk(ptype, skb))
2043 deliver_skb(skb2, pt_prev, skb->dev);
2048 /* need to clone skb, done only once */
2049 skb2 = skb_clone(skb, GFP_ATOMIC);
2053 net_timestamp_set(skb2);
2055 /* skb->nh should be correctly
2056 * set by sender, so that the second statement is
2057 * just protection against buggy protocols.
2059 skb_reset_mac_header(skb2);
2061 if (skb_network_header(skb2) < skb2->data ||
2062 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2063 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2064 ntohs(skb2->protocol),
2066 skb_reset_network_header(skb2);
2069 skb2->transport_header = skb2->network_header;
2070 skb2->pkt_type = PACKET_OUTGOING;
2074 if (ptype_list == &ptype_all) {
2075 ptype_list = &dev->ptype_all;
2080 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2081 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2087 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2090 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2091 * @dev: Network device
2092 * @txq: number of queues available
2094 * If real_num_tx_queues is changed the tc mappings may no longer be
2095 * valid. To resolve this verify the tc mapping remains valid and if
2096 * not NULL the mapping. With no priorities mapping to this
2097 * offset/count pair it will no longer be used. In the worst case TC0
2098 * is invalid nothing can be done so disable priority mappings. If is
2099 * expected that drivers will fix this mapping if they can before
2100 * calling netif_set_real_num_tx_queues.
2102 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2105 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2107 /* If TC0 is invalidated disable TC mapping */
2108 if (tc->offset + tc->count > txq) {
2109 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2114 /* Invalidated prio to tc mappings set to TC0 */
2115 for (i = 1; i < TC_BITMASK + 1; i++) {
2116 int q = netdev_get_prio_tc_map(dev, i);
2118 tc = &dev->tc_to_txq[q];
2119 if (tc->offset + tc->count > txq) {
2120 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
2122 netdev_set_prio_tc_map(dev, i, 0);
2127 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2130 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2133 /* walk through the TCs and see if it falls into any of them */
2134 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2135 if ((txq - tc->offset) < tc->count)
2139 /* didn't find it, just return -1 to indicate no match */
2145 EXPORT_SYMBOL(netdev_txq_to_tc);
2148 struct static_key xps_needed __read_mostly;
2149 EXPORT_SYMBOL(xps_needed);
2150 struct static_key xps_rxqs_needed __read_mostly;
2151 EXPORT_SYMBOL(xps_rxqs_needed);
2152 static DEFINE_MUTEX(xps_map_mutex);
2153 #define xmap_dereference(P) \
2154 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2156 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2159 struct xps_map *map = NULL;
2163 map = xmap_dereference(dev_maps->attr_map[tci]);
2167 for (pos = map->len; pos--;) {
2168 if (map->queues[pos] != index)
2172 map->queues[pos] = map->queues[--map->len];
2176 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2177 kfree_rcu(map, rcu);
2184 static bool remove_xps_queue_cpu(struct net_device *dev,
2185 struct xps_dev_maps *dev_maps,
2186 int cpu, u16 offset, u16 count)
2188 int num_tc = dev->num_tc ? : 1;
2189 bool active = false;
2192 for (tci = cpu * num_tc; num_tc--; tci++) {
2195 for (i = count, j = offset; i--; j++) {
2196 if (!remove_xps_queue(dev_maps, tci, j))
2206 static void reset_xps_maps(struct net_device *dev,
2207 struct xps_dev_maps *dev_maps,
2211 static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2212 RCU_INIT_POINTER(dev->xps_rxqs_map, NULL);
2214 RCU_INIT_POINTER(dev->xps_cpus_map, NULL);
2216 static_key_slow_dec_cpuslocked(&xps_needed);
2217 kfree_rcu(dev_maps, rcu);
2220 static void clean_xps_maps(struct net_device *dev, const unsigned long *mask,
2221 struct xps_dev_maps *dev_maps, unsigned int nr_ids,
2222 u16 offset, u16 count, bool is_rxqs_map)
2224 bool active = false;
2227 for (j = -1; j = netif_attrmask_next(j, mask, nr_ids),
2229 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset,
2232 reset_xps_maps(dev, dev_maps, is_rxqs_map);
2235 for (i = offset + (count - 1); count--; i--) {
2236 netdev_queue_numa_node_write(
2237 netdev_get_tx_queue(dev, i),
2243 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2246 const unsigned long *possible_mask = NULL;
2247 struct xps_dev_maps *dev_maps;
2248 unsigned int nr_ids;
2250 if (!static_key_false(&xps_needed))
2254 mutex_lock(&xps_map_mutex);
2256 if (static_key_false(&xps_rxqs_needed)) {
2257 dev_maps = xmap_dereference(dev->xps_rxqs_map);
2259 nr_ids = dev->num_rx_queues;
2260 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids,
2261 offset, count, true);
2265 dev_maps = xmap_dereference(dev->xps_cpus_map);
2269 if (num_possible_cpus() > 1)
2270 possible_mask = cpumask_bits(cpu_possible_mask);
2271 nr_ids = nr_cpu_ids;
2272 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, offset, count,
2276 mutex_unlock(&xps_map_mutex);
2280 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2282 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2285 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2286 u16 index, bool is_rxqs_map)
2288 struct xps_map *new_map;
2289 int alloc_len = XPS_MIN_MAP_ALLOC;
2292 for (pos = 0; map && pos < map->len; pos++) {
2293 if (map->queues[pos] != index)
2298 /* Need to add tx-queue to this CPU's/rx-queue's existing map */
2300 if (pos < map->alloc_len)
2303 alloc_len = map->alloc_len * 2;
2306 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2310 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2312 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2313 cpu_to_node(attr_index));
2317 for (i = 0; i < pos; i++)
2318 new_map->queues[i] = map->queues[i];
2319 new_map->alloc_len = alloc_len;
2325 /* Must be called under cpus_read_lock */
2326 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2327 u16 index, bool is_rxqs_map)
2329 const unsigned long *online_mask = NULL, *possible_mask = NULL;
2330 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2331 int i, j, tci, numa_node_id = -2;
2332 int maps_sz, num_tc = 1, tc = 0;
2333 struct xps_map *map, *new_map;
2334 bool active = false;
2335 unsigned int nr_ids;
2338 /* Do not allow XPS on subordinate device directly */
2339 num_tc = dev->num_tc;
2343 /* If queue belongs to subordinate dev use its map */
2344 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2346 tc = netdev_txq_to_tc(dev, index);
2351 mutex_lock(&xps_map_mutex);
2353 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2354 dev_maps = xmap_dereference(dev->xps_rxqs_map);
2355 nr_ids = dev->num_rx_queues;
2357 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2358 if (num_possible_cpus() > 1) {
2359 online_mask = cpumask_bits(cpu_online_mask);
2360 possible_mask = cpumask_bits(cpu_possible_mask);
2362 dev_maps = xmap_dereference(dev->xps_cpus_map);
2363 nr_ids = nr_cpu_ids;
2366 if (maps_sz < L1_CACHE_BYTES)
2367 maps_sz = L1_CACHE_BYTES;
2369 /* allocate memory for queue storage */
2370 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2373 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2374 if (!new_dev_maps) {
2375 mutex_unlock(&xps_map_mutex);
2379 tci = j * num_tc + tc;
2380 map = dev_maps ? xmap_dereference(dev_maps->attr_map[tci]) :
2383 map = expand_xps_map(map, j, index, is_rxqs_map);
2387 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2391 goto out_no_new_maps;
2394 /* Increment static keys at most once per type */
2395 static_key_slow_inc_cpuslocked(&xps_needed);
2397 static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2400 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2402 /* copy maps belonging to foreign traffic classes */
2403 for (i = tc, tci = j * num_tc; dev_maps && i--; tci++) {
2404 /* fill in the new device map from the old device map */
2405 map = xmap_dereference(dev_maps->attr_map[tci]);
2406 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2409 /* We need to explicitly update tci as prevous loop
2410 * could break out early if dev_maps is NULL.
2412 tci = j * num_tc + tc;
2414 if (netif_attr_test_mask(j, mask, nr_ids) &&
2415 netif_attr_test_online(j, online_mask, nr_ids)) {
2416 /* add tx-queue to CPU/rx-queue maps */
2419 map = xmap_dereference(new_dev_maps->attr_map[tci]);
2420 while ((pos < map->len) && (map->queues[pos] != index))
2423 if (pos == map->len)
2424 map->queues[map->len++] = index;
2427 if (numa_node_id == -2)
2428 numa_node_id = cpu_to_node(j);
2429 else if (numa_node_id != cpu_to_node(j))
2433 } else if (dev_maps) {
2434 /* fill in the new device map from the old device map */
2435 map = xmap_dereference(dev_maps->attr_map[tci]);
2436 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2439 /* copy maps belonging to foreign traffic classes */
2440 for (i = num_tc - tc, tci++; dev_maps && --i; tci++) {
2441 /* fill in the new device map from the old device map */
2442 map = xmap_dereference(dev_maps->attr_map[tci]);
2443 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2448 rcu_assign_pointer(dev->xps_rxqs_map, new_dev_maps);
2450 rcu_assign_pointer(dev->xps_cpus_map, new_dev_maps);
2452 /* Cleanup old maps */
2454 goto out_no_old_maps;
2456 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2458 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2459 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2460 map = xmap_dereference(dev_maps->attr_map[tci]);
2461 if (map && map != new_map)
2462 kfree_rcu(map, rcu);
2466 kfree_rcu(dev_maps, rcu);
2469 dev_maps = new_dev_maps;
2474 /* update Tx queue numa node */
2475 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2476 (numa_node_id >= 0) ?
2477 numa_node_id : NUMA_NO_NODE);
2483 /* removes tx-queue from unused CPUs/rx-queues */
2484 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2486 for (i = tc, tci = j * num_tc; i--; tci++)
2487 active |= remove_xps_queue(dev_maps, tci, index);
2488 if (!netif_attr_test_mask(j, mask, nr_ids) ||
2489 !netif_attr_test_online(j, online_mask, nr_ids))
2490 active |= remove_xps_queue(dev_maps, tci, index);
2491 for (i = num_tc - tc, tci++; --i; tci++)
2492 active |= remove_xps_queue(dev_maps, tci, index);
2495 /* free map if not active */
2497 reset_xps_maps(dev, dev_maps, is_rxqs_map);
2500 mutex_unlock(&xps_map_mutex);
2504 /* remove any maps that we added */
2505 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2507 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2508 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2510 xmap_dereference(dev_maps->attr_map[tci]) :
2512 if (new_map && new_map != map)
2517 mutex_unlock(&xps_map_mutex);
2519 kfree(new_dev_maps);
2522 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2524 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2530 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, false);
2535 EXPORT_SYMBOL(netif_set_xps_queue);
2538 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2540 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2542 /* Unbind any subordinate channels */
2543 while (txq-- != &dev->_tx[0]) {
2545 netdev_unbind_sb_channel(dev, txq->sb_dev);
2549 void netdev_reset_tc(struct net_device *dev)
2552 netif_reset_xps_queues_gt(dev, 0);
2554 netdev_unbind_all_sb_channels(dev);
2556 /* Reset TC configuration of device */
2558 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2559 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2561 EXPORT_SYMBOL(netdev_reset_tc);
2563 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2565 if (tc >= dev->num_tc)
2569 netif_reset_xps_queues(dev, offset, count);
2571 dev->tc_to_txq[tc].count = count;
2572 dev->tc_to_txq[tc].offset = offset;
2575 EXPORT_SYMBOL(netdev_set_tc_queue);
2577 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2579 if (num_tc > TC_MAX_QUEUE)
2583 netif_reset_xps_queues_gt(dev, 0);
2585 netdev_unbind_all_sb_channels(dev);
2587 dev->num_tc = num_tc;
2590 EXPORT_SYMBOL(netdev_set_num_tc);
2592 void netdev_unbind_sb_channel(struct net_device *dev,
2593 struct net_device *sb_dev)
2595 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2598 netif_reset_xps_queues_gt(sb_dev, 0);
2600 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2601 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2603 while (txq-- != &dev->_tx[0]) {
2604 if (txq->sb_dev == sb_dev)
2608 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2610 int netdev_bind_sb_channel_queue(struct net_device *dev,
2611 struct net_device *sb_dev,
2612 u8 tc, u16 count, u16 offset)
2614 /* Make certain the sb_dev and dev are already configured */
2615 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2618 /* We cannot hand out queues we don't have */
2619 if ((offset + count) > dev->real_num_tx_queues)
2622 /* Record the mapping */
2623 sb_dev->tc_to_txq[tc].count = count;
2624 sb_dev->tc_to_txq[tc].offset = offset;
2626 /* Provide a way for Tx queue to find the tc_to_txq map or
2627 * XPS map for itself.
2630 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
2634 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
2636 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
2638 /* Do not use a multiqueue device to represent a subordinate channel */
2639 if (netif_is_multiqueue(dev))
2642 /* We allow channels 1 - 32767 to be used for subordinate channels.
2643 * Channel 0 is meant to be "native" mode and used only to represent
2644 * the main root device. We allow writing 0 to reset the device back
2645 * to normal mode after being used as a subordinate channel.
2647 if (channel > S16_MAX)
2650 dev->num_tc = -channel;
2654 EXPORT_SYMBOL(netdev_set_sb_channel);
2657 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2658 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2660 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2665 disabling = txq < dev->real_num_tx_queues;
2667 if (txq < 1 || txq > dev->num_tx_queues)
2670 if (dev->reg_state == NETREG_REGISTERED ||
2671 dev->reg_state == NETREG_UNREGISTERING) {
2674 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2680 netif_setup_tc(dev, txq);
2682 dev->real_num_tx_queues = txq;
2686 qdisc_reset_all_tx_gt(dev, txq);
2688 netif_reset_xps_queues_gt(dev, txq);
2692 dev->real_num_tx_queues = txq;
2697 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2701 * netif_set_real_num_rx_queues - set actual number of RX queues used
2702 * @dev: Network device
2703 * @rxq: Actual number of RX queues
2705 * This must be called either with the rtnl_lock held or before
2706 * registration of the net device. Returns 0 on success, or a
2707 * negative error code. If called before registration, it always
2710 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2714 if (rxq < 1 || rxq > dev->num_rx_queues)
2717 if (dev->reg_state == NETREG_REGISTERED) {
2720 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2726 dev->real_num_rx_queues = rxq;
2729 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2733 * netif_get_num_default_rss_queues - default number of RSS queues
2735 * This routine should set an upper limit on the number of RSS queues
2736 * used by default by multiqueue devices.
2738 int netif_get_num_default_rss_queues(void)
2740 return is_kdump_kernel() ?
2741 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2743 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2745 static void __netif_reschedule(struct Qdisc *q)
2747 struct softnet_data *sd;
2748 unsigned long flags;
2750 local_irq_save(flags);
2751 sd = this_cpu_ptr(&softnet_data);
2752 q->next_sched = NULL;
2753 *sd->output_queue_tailp = q;
2754 sd->output_queue_tailp = &q->next_sched;
2755 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2756 local_irq_restore(flags);
2759 void __netif_schedule(struct Qdisc *q)
2761 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2762 __netif_reschedule(q);
2764 EXPORT_SYMBOL(__netif_schedule);
2766 struct dev_kfree_skb_cb {
2767 enum skb_free_reason reason;
2770 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2772 return (struct dev_kfree_skb_cb *)skb->cb;
2775 void netif_schedule_queue(struct netdev_queue *txq)
2778 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2779 struct Qdisc *q = rcu_dereference(txq->qdisc);
2781 __netif_schedule(q);
2785 EXPORT_SYMBOL(netif_schedule_queue);
2787 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2789 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2793 q = rcu_dereference(dev_queue->qdisc);
2794 __netif_schedule(q);
2798 EXPORT_SYMBOL(netif_tx_wake_queue);
2800 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2802 unsigned long flags;
2807 if (likely(refcount_read(&skb->users) == 1)) {
2809 refcount_set(&skb->users, 0);
2810 } else if (likely(!refcount_dec_and_test(&skb->users))) {
2813 get_kfree_skb_cb(skb)->reason = reason;
2814 local_irq_save(flags);
2815 skb->next = __this_cpu_read(softnet_data.completion_queue);
2816 __this_cpu_write(softnet_data.completion_queue, skb);
2817 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2818 local_irq_restore(flags);
2820 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2822 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2824 if (in_irq() || irqs_disabled())
2825 __dev_kfree_skb_irq(skb, reason);
2829 EXPORT_SYMBOL(__dev_kfree_skb_any);
2833 * netif_device_detach - mark device as removed
2834 * @dev: network device
2836 * Mark device as removed from system and therefore no longer available.
2838 void netif_device_detach(struct net_device *dev)
2840 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2841 netif_running(dev)) {
2842 netif_tx_stop_all_queues(dev);
2845 EXPORT_SYMBOL(netif_device_detach);
2848 * netif_device_attach - mark device as attached
2849 * @dev: network device
2851 * Mark device as attached from system and restart if needed.
2853 void netif_device_attach(struct net_device *dev)
2855 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2856 netif_running(dev)) {
2857 netif_tx_wake_all_queues(dev);
2858 __netdev_watchdog_up(dev);
2861 EXPORT_SYMBOL(netif_device_attach);
2864 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2865 * to be used as a distribution range.
2867 static u16 skb_tx_hash(const struct net_device *dev,
2868 const struct net_device *sb_dev,
2869 struct sk_buff *skb)
2873 u16 qcount = dev->real_num_tx_queues;
2876 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2878 qoffset = sb_dev->tc_to_txq[tc].offset;
2879 qcount = sb_dev->tc_to_txq[tc].count;
2882 if (skb_rx_queue_recorded(skb)) {
2883 hash = skb_get_rx_queue(skb);
2884 while (unlikely(hash >= qcount))
2886 return hash + qoffset;
2889 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
2892 static void skb_warn_bad_offload(const struct sk_buff *skb)
2894 static const netdev_features_t null_features;
2895 struct net_device *dev = skb->dev;
2896 const char *name = "";
2898 if (!net_ratelimit())
2902 if (dev->dev.parent)
2903 name = dev_driver_string(dev->dev.parent);
2905 name = netdev_name(dev);
2907 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2908 "gso_type=%d ip_summed=%d\n",
2909 name, dev ? &dev->features : &null_features,
2910 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2911 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2912 skb_shinfo(skb)->gso_type, skb->ip_summed);
2916 * Invalidate hardware checksum when packet is to be mangled, and
2917 * complete checksum manually on outgoing path.
2919 int skb_checksum_help(struct sk_buff *skb)
2922 int ret = 0, offset;
2924 if (skb->ip_summed == CHECKSUM_COMPLETE)
2925 goto out_set_summed;
2927 if (unlikely(skb_shinfo(skb)->gso_size)) {
2928 skb_warn_bad_offload(skb);
2932 /* Before computing a checksum, we should make sure no frag could
2933 * be modified by an external entity : checksum could be wrong.
2935 if (skb_has_shared_frag(skb)) {
2936 ret = __skb_linearize(skb);
2941 offset = skb_checksum_start_offset(skb);
2942 BUG_ON(offset >= skb_headlen(skb));
2943 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2945 offset += skb->csum_offset;
2946 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2948 if (skb_cloned(skb) &&
2949 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2950 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2955 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
2957 skb->ip_summed = CHECKSUM_NONE;
2961 EXPORT_SYMBOL(skb_checksum_help);
2963 int skb_crc32c_csum_help(struct sk_buff *skb)
2966 int ret = 0, offset, start;
2968 if (skb->ip_summed != CHECKSUM_PARTIAL)
2971 if (unlikely(skb_is_gso(skb)))
2974 /* Before computing a checksum, we should make sure no frag could
2975 * be modified by an external entity : checksum could be wrong.
2977 if (unlikely(skb_has_shared_frag(skb))) {
2978 ret = __skb_linearize(skb);
2982 start = skb_checksum_start_offset(skb);
2983 offset = start + offsetof(struct sctphdr, checksum);
2984 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
2988 if (skb_cloned(skb) &&
2989 !skb_clone_writable(skb, offset + sizeof(__le32))) {
2990 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2994 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
2995 skb->len - start, ~(__u32)0,
2997 *(__le32 *)(skb->data + offset) = crc32c_csum;
2998 skb->ip_summed = CHECKSUM_NONE;
2999 skb->csum_not_inet = 0;
3004 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
3006 __be16 type = skb->protocol;
3008 /* Tunnel gso handlers can set protocol to ethernet. */
3009 if (type == htons(ETH_P_TEB)) {
3012 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
3015 eth = (struct ethhdr *)skb->data;
3016 type = eth->h_proto;
3019 return __vlan_get_protocol(skb, type, depth);
3023 * skb_mac_gso_segment - mac layer segmentation handler.
3024 * @skb: buffer to segment
3025 * @features: features for the output path (see dev->features)
3027 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
3028 netdev_features_t features)
3030 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
3031 struct packet_offload *ptype;
3032 int vlan_depth = skb->mac_len;
3033 __be16 type = skb_network_protocol(skb, &vlan_depth);
3035 if (unlikely(!type))
3036 return ERR_PTR(-EINVAL);
3038 __skb_pull(skb, vlan_depth);
3041 list_for_each_entry_rcu(ptype, &offload_base, list) {
3042 if (ptype->type == type && ptype->callbacks.gso_segment) {
3043 segs = ptype->callbacks.gso_segment(skb, features);
3049 __skb_push(skb, skb->data - skb_mac_header(skb));
3053 EXPORT_SYMBOL(skb_mac_gso_segment);
3056 /* openvswitch calls this on rx path, so we need a different check.
3058 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
3061 return skb->ip_summed != CHECKSUM_PARTIAL &&
3062 skb->ip_summed != CHECKSUM_UNNECESSARY;
3064 return skb->ip_summed == CHECKSUM_NONE;
3068 * __skb_gso_segment - Perform segmentation on skb.
3069 * @skb: buffer to segment
3070 * @features: features for the output path (see dev->features)
3071 * @tx_path: whether it is called in TX path
3073 * This function segments the given skb and returns a list of segments.
3075 * It may return NULL if the skb requires no segmentation. This is
3076 * only possible when GSO is used for verifying header integrity.
3078 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
3080 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
3081 netdev_features_t features, bool tx_path)
3083 struct sk_buff *segs;
3085 if (unlikely(skb_needs_check(skb, tx_path))) {
3088 /* We're going to init ->check field in TCP or UDP header */
3089 err = skb_cow_head(skb, 0);
3091 return ERR_PTR(err);
3094 /* Only report GSO partial support if it will enable us to
3095 * support segmentation on this frame without needing additional
3098 if (features & NETIF_F_GSO_PARTIAL) {
3099 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
3100 struct net_device *dev = skb->dev;
3102 partial_features |= dev->features & dev->gso_partial_features;
3103 if (!skb_gso_ok(skb, features | partial_features))
3104 features &= ~NETIF_F_GSO_PARTIAL;
3107 BUILD_BUG_ON(SKB_SGO_CB_OFFSET +
3108 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
3110 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
3111 SKB_GSO_CB(skb)->encap_level = 0;
3113 skb_reset_mac_header(skb);
3114 skb_reset_mac_len(skb);
3116 segs = skb_mac_gso_segment(skb, features);
3118 if (unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
3119 skb_warn_bad_offload(skb);
3123 EXPORT_SYMBOL(__skb_gso_segment);
3125 /* Take action when hardware reception checksum errors are detected. */
3127 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3129 if (net_ratelimit()) {
3130 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
3132 pr_err("dev features: %pNF\n", &dev->features);
3133 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",
3134 skb->len, skb->data_len, skb->pkt_type,
3135 skb_shinfo(skb)->gso_size, skb_shinfo(skb)->gso_type,
3136 skb_shinfo(skb)->nr_frags, skb->ip_summed, skb->csum,
3137 skb->csum_complete_sw, skb->csum_valid, skb->csum_level);
3141 EXPORT_SYMBOL(netdev_rx_csum_fault);
3144 /* XXX: check that highmem exists at all on the given machine. */
3145 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3147 #ifdef CONFIG_HIGHMEM
3150 if (!(dev->features & NETIF_F_HIGHDMA)) {
3151 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3152 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3154 if (PageHighMem(skb_frag_page(frag)))
3162 /* If MPLS offload request, verify we are testing hardware MPLS features
3163 * instead of standard features for the netdev.
3165 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3166 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3167 netdev_features_t features,
3170 if (eth_p_mpls(type))
3171 features &= skb->dev->mpls_features;
3176 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3177 netdev_features_t features,
3184 static netdev_features_t harmonize_features(struct sk_buff *skb,
3185 netdev_features_t features)
3190 type = skb_network_protocol(skb, &tmp);
3191 features = net_mpls_features(skb, features, type);
3193 if (skb->ip_summed != CHECKSUM_NONE &&
3194 !can_checksum_protocol(features, type)) {
3195 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3197 if (illegal_highdma(skb->dev, skb))
3198 features &= ~NETIF_F_SG;
3203 netdev_features_t passthru_features_check(struct sk_buff *skb,
3204 struct net_device *dev,
3205 netdev_features_t features)
3209 EXPORT_SYMBOL(passthru_features_check);
3211 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3212 struct net_device *dev,
3213 netdev_features_t features)
3215 return vlan_features_check(skb, features);
3218 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3219 struct net_device *dev,
3220 netdev_features_t features)
3222 u16 gso_segs = skb_shinfo(skb)->gso_segs;
3224 if (gso_segs > dev->gso_max_segs)
3225 return features & ~NETIF_F_GSO_MASK;
3227 /* Support for GSO partial features requires software
3228 * intervention before we can actually process the packets
3229 * so we need to strip support for any partial features now
3230 * and we can pull them back in after we have partially
3231 * segmented the frame.
3233 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3234 features &= ~dev->gso_partial_features;
3236 /* Make sure to clear the IPv4 ID mangling feature if the
3237 * IPv4 header has the potential to be fragmented.
3239 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3240 struct iphdr *iph = skb->encapsulation ?
3241 inner_ip_hdr(skb) : ip_hdr(skb);
3243 if (!(iph->frag_off & htons(IP_DF)))
3244 features &= ~NETIF_F_TSO_MANGLEID;
3250 netdev_features_t netif_skb_features(struct sk_buff *skb)
3252 struct net_device *dev = skb->dev;
3253 netdev_features_t features = dev->features;
3255 if (skb_is_gso(skb))
3256 features = gso_features_check(skb, dev, features);
3258 /* If encapsulation offload request, verify we are testing
3259 * hardware encapsulation features instead of standard
3260 * features for the netdev
3262 if (skb->encapsulation)
3263 features &= dev->hw_enc_features;
3265 if (skb_vlan_tagged(skb))
3266 features = netdev_intersect_features(features,
3267 dev->vlan_features |
3268 NETIF_F_HW_VLAN_CTAG_TX |
3269 NETIF_F_HW_VLAN_STAG_TX);
3271 if (dev->netdev_ops->ndo_features_check)
3272 features &= dev->netdev_ops->ndo_features_check(skb, dev,
3275 features &= dflt_features_check(skb, dev, features);
3277 return harmonize_features(skb, features);
3279 EXPORT_SYMBOL(netif_skb_features);
3281 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3282 struct netdev_queue *txq, bool more)
3287 if (dev_nit_active(dev))
3288 dev_queue_xmit_nit(skb, dev);
3291 trace_net_dev_start_xmit(skb, dev);
3292 rc = netdev_start_xmit(skb, dev, txq, more);
3293 trace_net_dev_xmit(skb, rc, dev, len);
3298 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3299 struct netdev_queue *txq, int *ret)
3301 struct sk_buff *skb = first;
3302 int rc = NETDEV_TX_OK;
3305 struct sk_buff *next = skb->next;
3307 skb_mark_not_on_list(skb);
3308 rc = xmit_one(skb, dev, txq, next != NULL);
3309 if (unlikely(!dev_xmit_complete(rc))) {
3315 if (netif_tx_queue_stopped(txq) && skb) {
3316 rc = NETDEV_TX_BUSY;
3326 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3327 netdev_features_t features)
3329 if (skb_vlan_tag_present(skb) &&
3330 !vlan_hw_offload_capable(features, skb->vlan_proto))
3331 skb = __vlan_hwaccel_push_inside(skb);
3335 int skb_csum_hwoffload_help(struct sk_buff *skb,
3336 const netdev_features_t features)
3338 if (unlikely(skb->csum_not_inet))
3339 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3340 skb_crc32c_csum_help(skb);
3342 return !!(features & NETIF_F_CSUM_MASK) ? 0 : skb_checksum_help(skb);
3344 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3346 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3348 netdev_features_t features;
3350 features = netif_skb_features(skb);
3351 skb = validate_xmit_vlan(skb, features);
3355 skb = sk_validate_xmit_skb(skb, dev);
3359 if (netif_needs_gso(skb, features)) {
3360 struct sk_buff *segs;
3362 segs = skb_gso_segment(skb, features);
3370 if (skb_needs_linearize(skb, features) &&
3371 __skb_linearize(skb))
3374 /* If packet is not checksummed and device does not
3375 * support checksumming for this protocol, complete
3376 * checksumming here.
3378 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3379 if (skb->encapsulation)
3380 skb_set_inner_transport_header(skb,
3381 skb_checksum_start_offset(skb));
3383 skb_set_transport_header(skb,
3384 skb_checksum_start_offset(skb));
3385 if (skb_csum_hwoffload_help(skb, features))
3390 skb = validate_xmit_xfrm(skb, features, again);
3397 atomic_long_inc(&dev->tx_dropped);
3401 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3403 struct sk_buff *next, *head = NULL, *tail;
3405 for (; skb != NULL; skb = next) {
3407 skb_mark_not_on_list(skb);
3409 /* in case skb wont be segmented, point to itself */
3412 skb = validate_xmit_skb(skb, dev, again);
3420 /* If skb was segmented, skb->prev points to
3421 * the last segment. If not, it still contains skb.
3427 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3429 static void qdisc_pkt_len_init(struct sk_buff *skb)
3431 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3433 qdisc_skb_cb(skb)->pkt_len = skb->len;
3435 /* To get more precise estimation of bytes sent on wire,
3436 * we add to pkt_len the headers size of all segments
3438 if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3439 unsigned int hdr_len;
3440 u16 gso_segs = shinfo->gso_segs;
3442 /* mac layer + network layer */
3443 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3445 /* + transport layer */
3446 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3447 const struct tcphdr *th;
3448 struct tcphdr _tcphdr;
3450 th = skb_header_pointer(skb, skb_transport_offset(skb),
3451 sizeof(_tcphdr), &_tcphdr);
3453 hdr_len += __tcp_hdrlen(th);
3455 struct udphdr _udphdr;
3457 if (skb_header_pointer(skb, skb_transport_offset(skb),
3458 sizeof(_udphdr), &_udphdr))
3459 hdr_len += sizeof(struct udphdr);
3462 if (shinfo->gso_type & SKB_GSO_DODGY)
3463 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3466 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3470 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3471 struct net_device *dev,
3472 struct netdev_queue *txq)
3474 spinlock_t *root_lock = qdisc_lock(q);
3475 struct sk_buff *to_free = NULL;
3479 qdisc_calculate_pkt_len(skb, q);
3481 if (q->flags & TCQ_F_NOLOCK) {
3482 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3483 __qdisc_drop(skb, &to_free);
3485 } else if ((q->flags & TCQ_F_CAN_BYPASS) && q->empty &&
3486 qdisc_run_begin(q)) {
3487 qdisc_bstats_cpu_update(q, skb);
3489 if (sch_direct_xmit(skb, q, dev, txq, NULL, true))
3493 rc = NET_XMIT_SUCCESS;
3495 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3499 if (unlikely(to_free))
3500 kfree_skb_list(to_free);
3505 * Heuristic to force contended enqueues to serialize on a
3506 * separate lock before trying to get qdisc main lock.
3507 * This permits qdisc->running owner to get the lock more
3508 * often and dequeue packets faster.
3510 contended = qdisc_is_running(q);
3511 if (unlikely(contended))
3512 spin_lock(&q->busylock);
3514 spin_lock(root_lock);
3515 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3516 __qdisc_drop(skb, &to_free);
3518 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3519 qdisc_run_begin(q)) {
3521 * This is a work-conserving queue; there are no old skbs
3522 * waiting to be sent out; and the qdisc is not running -
3523 * xmit the skb directly.
3526 qdisc_bstats_update(q, skb);
3528 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3529 if (unlikely(contended)) {
3530 spin_unlock(&q->busylock);
3537 rc = NET_XMIT_SUCCESS;
3539 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3540 if (qdisc_run_begin(q)) {
3541 if (unlikely(contended)) {
3542 spin_unlock(&q->busylock);
3549 spin_unlock(root_lock);
3550 if (unlikely(to_free))
3551 kfree_skb_list(to_free);
3552 if (unlikely(contended))
3553 spin_unlock(&q->busylock);
3557 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3558 static void skb_update_prio(struct sk_buff *skb)
3560 const struct netprio_map *map;
3561 const struct sock *sk;
3562 unsigned int prioidx;
3566 map = rcu_dereference_bh(skb->dev->priomap);
3569 sk = skb_to_full_sk(skb);
3573 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3575 if (prioidx < map->priomap_len)
3576 skb->priority = map->priomap[prioidx];
3579 #define skb_update_prio(skb)
3583 * dev_loopback_xmit - loop back @skb
3584 * @net: network namespace this loopback is happening in
3585 * @sk: sk needed to be a netfilter okfn
3586 * @skb: buffer to transmit
3588 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3590 skb_reset_mac_header(skb);
3591 __skb_pull(skb, skb_network_offset(skb));
3592 skb->pkt_type = PACKET_LOOPBACK;
3593 skb->ip_summed = CHECKSUM_UNNECESSARY;
3594 WARN_ON(!skb_dst(skb));
3599 EXPORT_SYMBOL(dev_loopback_xmit);
3601 #ifdef CONFIG_NET_EGRESS
3602 static struct sk_buff *
3603 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3605 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3606 struct tcf_result cl_res;
3611 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3612 mini_qdisc_bstats_cpu_update(miniq, skb);
3614 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
3616 case TC_ACT_RECLASSIFY:
3617 skb->tc_index = TC_H_MIN(cl_res.classid);
3620 mini_qdisc_qstats_cpu_drop(miniq);
3621 *ret = NET_XMIT_DROP;
3627 *ret = NET_XMIT_SUCCESS;
3630 case TC_ACT_REDIRECT:
3631 /* No need to push/pop skb's mac_header here on egress! */
3632 skb_do_redirect(skb);
3633 *ret = NET_XMIT_SUCCESS;
3641 #endif /* CONFIG_NET_EGRESS */
3644 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
3645 struct xps_dev_maps *dev_maps, unsigned int tci)
3647 struct xps_map *map;
3648 int queue_index = -1;
3652 tci += netdev_get_prio_tc_map(dev, skb->priority);
3655 map = rcu_dereference(dev_maps->attr_map[tci]);
3658 queue_index = map->queues[0];
3660 queue_index = map->queues[reciprocal_scale(
3661 skb_get_hash(skb), map->len)];
3662 if (unlikely(queue_index >= dev->real_num_tx_queues))
3669 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
3670 struct sk_buff *skb)
3673 struct xps_dev_maps *dev_maps;
3674 struct sock *sk = skb->sk;
3675 int queue_index = -1;
3677 if (!static_key_false(&xps_needed))
3681 if (!static_key_false(&xps_rxqs_needed))
3684 dev_maps = rcu_dereference(sb_dev->xps_rxqs_map);
3686 int tci = sk_rx_queue_get(sk);
3688 if (tci >= 0 && tci < dev->num_rx_queues)
3689 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
3694 if (queue_index < 0) {
3695 dev_maps = rcu_dereference(sb_dev->xps_cpus_map);
3697 unsigned int tci = skb->sender_cpu - 1;
3699 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
3711 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
3712 struct net_device *sb_dev)
3716 EXPORT_SYMBOL(dev_pick_tx_zero);
3718 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
3719 struct net_device *sb_dev)
3721 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
3723 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
3725 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
3726 struct net_device *sb_dev)
3728 struct sock *sk = skb->sk;
3729 int queue_index = sk_tx_queue_get(sk);
3731 sb_dev = sb_dev ? : dev;
3733 if (queue_index < 0 || skb->ooo_okay ||
3734 queue_index >= dev->real_num_tx_queues) {
3735 int new_index = get_xps_queue(dev, sb_dev, skb);
3738 new_index = skb_tx_hash(dev, sb_dev, skb);
3740 if (queue_index != new_index && sk &&
3742 rcu_access_pointer(sk->sk_dst_cache))
3743 sk_tx_queue_set(sk, new_index);
3745 queue_index = new_index;
3750 EXPORT_SYMBOL(netdev_pick_tx);
3752 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
3753 struct sk_buff *skb,
3754 struct net_device *sb_dev)
3756 int queue_index = 0;
3759 u32 sender_cpu = skb->sender_cpu - 1;
3761 if (sender_cpu >= (u32)NR_CPUS)
3762 skb->sender_cpu = raw_smp_processor_id() + 1;
3765 if (dev->real_num_tx_queues != 1) {
3766 const struct net_device_ops *ops = dev->netdev_ops;
3768 if (ops->ndo_select_queue)
3769 queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
3771 queue_index = netdev_pick_tx(dev, skb, sb_dev);
3773 queue_index = netdev_cap_txqueue(dev, queue_index);
3776 skb_set_queue_mapping(skb, queue_index);
3777 return netdev_get_tx_queue(dev, queue_index);
3781 * __dev_queue_xmit - transmit a buffer
3782 * @skb: buffer to transmit
3783 * @sb_dev: suboordinate device used for L2 forwarding offload
3785 * Queue a buffer for transmission to a network device. The caller must
3786 * have set the device and priority and built the buffer before calling
3787 * this function. The function can be called from an interrupt.
3789 * A negative errno code is returned on a failure. A success does not
3790 * guarantee the frame will be transmitted as it may be dropped due
3791 * to congestion or traffic shaping.
3793 * -----------------------------------------------------------------------------------
3794 * I notice this method can also return errors from the queue disciplines,
3795 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3798 * Regardless of the return value, the skb is consumed, so it is currently
3799 * difficult to retry a send to this method. (You can bump the ref count
3800 * before sending to hold a reference for retry if you are careful.)
3802 * When calling this method, interrupts MUST be enabled. This is because
3803 * the BH enable code must have IRQs enabled so that it will not deadlock.
3806 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
3808 struct net_device *dev = skb->dev;
3809 struct netdev_queue *txq;
3814 skb_reset_mac_header(skb);
3816 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3817 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3819 /* Disable soft irqs for various locks below. Also
3820 * stops preemption for RCU.
3824 skb_update_prio(skb);
3826 qdisc_pkt_len_init(skb);
3827 #ifdef CONFIG_NET_CLS_ACT
3828 skb->tc_at_ingress = 0;
3829 # ifdef CONFIG_NET_EGRESS
3830 if (static_branch_unlikely(&egress_needed_key)) {
3831 skb = sch_handle_egress(skb, &rc, dev);
3837 /* If device/qdisc don't need skb->dst, release it right now while
3838 * its hot in this cpu cache.
3840 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3845 txq = netdev_core_pick_tx(dev, skb, sb_dev);
3846 q = rcu_dereference_bh(txq->qdisc);
3848 trace_net_dev_queue(skb);
3850 rc = __dev_xmit_skb(skb, q, dev, txq);
3854 /* The device has no queue. Common case for software devices:
3855 * loopback, all the sorts of tunnels...
3857 * Really, it is unlikely that netif_tx_lock protection is necessary
3858 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
3860 * However, it is possible, that they rely on protection
3863 * Check this and shot the lock. It is not prone from deadlocks.
3864 *Either shot noqueue qdisc, it is even simpler 8)
3866 if (dev->flags & IFF_UP) {
3867 int cpu = smp_processor_id(); /* ok because BHs are off */
3869 if (txq->xmit_lock_owner != cpu) {
3870 if (dev_xmit_recursion())
3871 goto recursion_alert;
3873 skb = validate_xmit_skb(skb, dev, &again);
3877 HARD_TX_LOCK(dev, txq, cpu);
3879 if (!netif_xmit_stopped(txq)) {
3880 dev_xmit_recursion_inc();
3881 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
3882 dev_xmit_recursion_dec();
3883 if (dev_xmit_complete(rc)) {
3884 HARD_TX_UNLOCK(dev, txq);
3888 HARD_TX_UNLOCK(dev, txq);
3889 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3892 /* Recursion is detected! It is possible,
3896 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3902 rcu_read_unlock_bh();
3904 atomic_long_inc(&dev->tx_dropped);
3905 kfree_skb_list(skb);
3908 rcu_read_unlock_bh();
3912 int dev_queue_xmit(struct sk_buff *skb)
3914 return __dev_queue_xmit(skb, NULL);
3916 EXPORT_SYMBOL(dev_queue_xmit);
3918 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev)
3920 return __dev_queue_xmit(skb, sb_dev);
3922 EXPORT_SYMBOL(dev_queue_xmit_accel);
3924 int dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
3926 struct net_device *dev = skb->dev;
3927 struct sk_buff *orig_skb = skb;
3928 struct netdev_queue *txq;
3929 int ret = NETDEV_TX_BUSY;
3932 if (unlikely(!netif_running(dev) ||
3933 !netif_carrier_ok(dev)))
3936 skb = validate_xmit_skb_list(skb, dev, &again);
3937 if (skb != orig_skb)
3940 skb_set_queue_mapping(skb, queue_id);
3941 txq = skb_get_tx_queue(dev, skb);
3945 HARD_TX_LOCK(dev, txq, smp_processor_id());
3946 if (!netif_xmit_frozen_or_drv_stopped(txq))
3947 ret = netdev_start_xmit(skb, dev, txq, false);
3948 HARD_TX_UNLOCK(dev, txq);
3952 if (!dev_xmit_complete(ret))
3957 atomic_long_inc(&dev->tx_dropped);
3958 kfree_skb_list(skb);
3959 return NET_XMIT_DROP;
3961 EXPORT_SYMBOL(dev_direct_xmit);
3963 /*************************************************************************
3965 *************************************************************************/
3967 int netdev_max_backlog __read_mostly = 1000;
3968 EXPORT_SYMBOL(netdev_max_backlog);
3970 int netdev_tstamp_prequeue __read_mostly = 1;
3971 int netdev_budget __read_mostly = 300;
3972 unsigned int __read_mostly netdev_budget_usecs = 2000;
3973 int weight_p __read_mostly = 64; /* old backlog weight */
3974 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
3975 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
3976 int dev_rx_weight __read_mostly = 64;
3977 int dev_tx_weight __read_mostly = 64;
3979 /* Called with irq disabled */
3980 static inline void ____napi_schedule(struct softnet_data *sd,
3981 struct napi_struct *napi)
3983 list_add_tail(&napi->poll_list, &sd->poll_list);
3984 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3989 /* One global table that all flow-based protocols share. */
3990 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3991 EXPORT_SYMBOL(rps_sock_flow_table);
3992 u32 rps_cpu_mask __read_mostly;
3993 EXPORT_SYMBOL(rps_cpu_mask);
3995 struct static_key_false rps_needed __read_mostly;
3996 EXPORT_SYMBOL(rps_needed);
3997 struct static_key_false rfs_needed __read_mostly;
3998 EXPORT_SYMBOL(rfs_needed);
4000 static struct rps_dev_flow *
4001 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4002 struct rps_dev_flow *rflow, u16 next_cpu)
4004 if (next_cpu < nr_cpu_ids) {
4005 #ifdef CONFIG_RFS_ACCEL
4006 struct netdev_rx_queue *rxqueue;
4007 struct rps_dev_flow_table *flow_table;
4008 struct rps_dev_flow *old_rflow;
4013 /* Should we steer this flow to a different hardware queue? */
4014 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
4015 !(dev->features & NETIF_F_NTUPLE))
4017 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
4018 if (rxq_index == skb_get_rx_queue(skb))
4021 rxqueue = dev->_rx + rxq_index;
4022 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4025 flow_id = skb_get_hash(skb) & flow_table->mask;
4026 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
4027 rxq_index, flow_id);
4031 rflow = &flow_table->flows[flow_id];
4033 if (old_rflow->filter == rflow->filter)
4034 old_rflow->filter = RPS_NO_FILTER;
4038 per_cpu(softnet_data, next_cpu).input_queue_head;
4041 rflow->cpu = next_cpu;
4046 * get_rps_cpu is called from netif_receive_skb and returns the target
4047 * CPU from the RPS map of the receiving queue for a given skb.
4048 * rcu_read_lock must be held on entry.
4050 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4051 struct rps_dev_flow **rflowp)
4053 const struct rps_sock_flow_table *sock_flow_table;
4054 struct netdev_rx_queue *rxqueue = dev->_rx;
4055 struct rps_dev_flow_table *flow_table;
4056 struct rps_map *map;
4061 if (skb_rx_queue_recorded(skb)) {
4062 u16 index = skb_get_rx_queue(skb);
4064 if (unlikely(index >= dev->real_num_rx_queues)) {
4065 WARN_ONCE(dev->real_num_rx_queues > 1,
4066 "%s received packet on queue %u, but number "
4067 "of RX queues is %u\n",
4068 dev->name, index, dev->real_num_rx_queues);
4074 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4076 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4077 map = rcu_dereference(rxqueue->rps_map);
4078 if (!flow_table && !map)
4081 skb_reset_network_header(skb);
4082 hash = skb_get_hash(skb);
4086 sock_flow_table = rcu_dereference(rps_sock_flow_table);
4087 if (flow_table && sock_flow_table) {
4088 struct rps_dev_flow *rflow;
4092 /* First check into global flow table if there is a match */
4093 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
4094 if ((ident ^ hash) & ~rps_cpu_mask)
4097 next_cpu = ident & rps_cpu_mask;
4099 /* OK, now we know there is a match,
4100 * we can look at the local (per receive queue) flow table
4102 rflow = &flow_table->flows[hash & flow_table->mask];
4106 * If the desired CPU (where last recvmsg was done) is
4107 * different from current CPU (one in the rx-queue flow
4108 * table entry), switch if one of the following holds:
4109 * - Current CPU is unset (>= nr_cpu_ids).
4110 * - Current CPU is offline.
4111 * - The current CPU's queue tail has advanced beyond the
4112 * last packet that was enqueued using this table entry.
4113 * This guarantees that all previous packets for the flow
4114 * have been dequeued, thus preserving in order delivery.
4116 if (unlikely(tcpu != next_cpu) &&
4117 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4118 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
4119 rflow->last_qtail)) >= 0)) {
4121 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4124 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4134 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4135 if (cpu_online(tcpu)) {
4145 #ifdef CONFIG_RFS_ACCEL
4148 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4149 * @dev: Device on which the filter was set
4150 * @rxq_index: RX queue index
4151 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4152 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4154 * Drivers that implement ndo_rx_flow_steer() should periodically call
4155 * this function for each installed filter and remove the filters for
4156 * which it returns %true.
4158 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4159 u32 flow_id, u16 filter_id)
4161 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4162 struct rps_dev_flow_table *flow_table;
4163 struct rps_dev_flow *rflow;
4168 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4169 if (flow_table && flow_id <= flow_table->mask) {
4170 rflow = &flow_table->flows[flow_id];
4171 cpu = READ_ONCE(rflow->cpu);
4172 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
4173 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
4174 rflow->last_qtail) <
4175 (int)(10 * flow_table->mask)))
4181 EXPORT_SYMBOL(rps_may_expire_flow);
4183 #endif /* CONFIG_RFS_ACCEL */
4185 /* Called from hardirq (IPI) context */
4186 static void rps_trigger_softirq(void *data)
4188 struct softnet_data *sd = data;
4190 ____napi_schedule(sd, &sd->backlog);
4194 #endif /* CONFIG_RPS */
4197 * Check if this softnet_data structure is another cpu one
4198 * If yes, queue it to our IPI list and return 1
4201 static int rps_ipi_queued(struct softnet_data *sd)
4204 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4207 sd->rps_ipi_next = mysd->rps_ipi_list;
4208 mysd->rps_ipi_list = sd;
4210 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4213 #endif /* CONFIG_RPS */
4217 #ifdef CONFIG_NET_FLOW_LIMIT
4218 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4221 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4223 #ifdef CONFIG_NET_FLOW_LIMIT
4224 struct sd_flow_limit *fl;
4225 struct softnet_data *sd;
4226 unsigned int old_flow, new_flow;
4228 if (qlen < (netdev_max_backlog >> 1))
4231 sd = this_cpu_ptr(&softnet_data);
4234 fl = rcu_dereference(sd->flow_limit);
4236 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4237 old_flow = fl->history[fl->history_head];
4238 fl->history[fl->history_head] = new_flow;
4241 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4243 if (likely(fl->buckets[old_flow]))
4244 fl->buckets[old_flow]--;
4246 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4258 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4259 * queue (may be a remote CPU queue).
4261 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4262 unsigned int *qtail)
4264 struct softnet_data *sd;
4265 unsigned long flags;
4268 sd = &per_cpu(softnet_data, cpu);
4270 local_irq_save(flags);
4273 if (!netif_running(skb->dev))
4275 qlen = skb_queue_len(&sd->input_pkt_queue);
4276 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
4279 __skb_queue_tail(&sd->input_pkt_queue, skb);
4280 input_queue_tail_incr_save(sd, qtail);
4282 local_irq_restore(flags);
4283 return NET_RX_SUCCESS;
4286 /* Schedule NAPI for backlog device
4287 * We can use non atomic operation since we own the queue lock
4289 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
4290 if (!rps_ipi_queued(sd))
4291 ____napi_schedule(sd, &sd->backlog);
4300 local_irq_restore(flags);
4302 atomic_long_inc(&skb->dev->rx_dropped);
4307 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4309 struct net_device *dev = skb->dev;
4310 struct netdev_rx_queue *rxqueue;
4314 if (skb_rx_queue_recorded(skb)) {
4315 u16 index = skb_get_rx_queue(skb);
4317 if (unlikely(index >= dev->real_num_rx_queues)) {
4318 WARN_ONCE(dev->real_num_rx_queues > 1,
4319 "%s received packet on queue %u, but number "
4320 "of RX queues is %u\n",
4321 dev->name, index, dev->real_num_rx_queues);
4323 return rxqueue; /* Return first rxqueue */
4330 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4331 struct xdp_buff *xdp,
4332 struct bpf_prog *xdp_prog)
4334 struct netdev_rx_queue *rxqueue;
4335 void *orig_data, *orig_data_end;
4336 u32 metalen, act = XDP_DROP;
4337 __be16 orig_eth_type;
4343 /* Reinjected packets coming from act_mirred or similar should
4344 * not get XDP generic processing.
4346 if (skb_cloned(skb) || skb_is_tc_redirected(skb))
4349 /* XDP packets must be linear and must have sufficient headroom
4350 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4351 * native XDP provides, thus we need to do it here as well.
4353 if (skb_is_nonlinear(skb) ||
4354 skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4355 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4356 int troom = skb->tail + skb->data_len - skb->end;
4358 /* In case we have to go down the path and also linearize,
4359 * then lets do the pskb_expand_head() work just once here.
4361 if (pskb_expand_head(skb,
4362 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4363 troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4365 if (skb_linearize(skb))
4369 /* The XDP program wants to see the packet starting at the MAC
4372 mac_len = skb->data - skb_mac_header(skb);
4373 hlen = skb_headlen(skb) + mac_len;
4374 xdp->data = skb->data - mac_len;
4375 xdp->data_meta = xdp->data;
4376 xdp->data_end = xdp->data + hlen;
4377 xdp->data_hard_start = skb->data - skb_headroom(skb);
4378 orig_data_end = xdp->data_end;
4379 orig_data = xdp->data;
4380 eth = (struct ethhdr *)xdp->data;
4381 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4382 orig_eth_type = eth->h_proto;
4384 rxqueue = netif_get_rxqueue(skb);
4385 xdp->rxq = &rxqueue->xdp_rxq;
4387 act = bpf_prog_run_xdp(xdp_prog, xdp);
4389 off = xdp->data - orig_data;
4391 __skb_pull(skb, off);
4393 __skb_push(skb, -off);
4394 skb->mac_header += off;
4396 /* check if bpf_xdp_adjust_tail was used. it can only "shrink"
4399 off = orig_data_end - xdp->data_end;
4401 skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4406 /* check if XDP changed eth hdr such SKB needs update */
4407 eth = (struct ethhdr *)xdp->data;
4408 if ((orig_eth_type != eth->h_proto) ||
4409 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4410 __skb_push(skb, ETH_HLEN);
4411 skb->protocol = eth_type_trans(skb, skb->dev);
4417 __skb_push(skb, mac_len);
4420 metalen = xdp->data - xdp->data_meta;
4422 skb_metadata_set(skb, metalen);
4425 bpf_warn_invalid_xdp_action(act);
4428 trace_xdp_exception(skb->dev, xdp_prog, act);
4439 /* When doing generic XDP we have to bypass the qdisc layer and the
4440 * network taps in order to match in-driver-XDP behavior.
4442 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4444 struct net_device *dev = skb->dev;
4445 struct netdev_queue *txq;
4446 bool free_skb = true;
4449 txq = netdev_core_pick_tx(dev, skb, NULL);
4450 cpu = smp_processor_id();
4451 HARD_TX_LOCK(dev, txq, cpu);
4452 if (!netif_xmit_stopped(txq)) {
4453 rc = netdev_start_xmit(skb, dev, txq, 0);
4454 if (dev_xmit_complete(rc))
4457 HARD_TX_UNLOCK(dev, txq);
4459 trace_xdp_exception(dev, xdp_prog, XDP_TX);
4463 EXPORT_SYMBOL_GPL(generic_xdp_tx);
4465 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4467 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4470 struct xdp_buff xdp;
4474 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4475 if (act != XDP_PASS) {
4478 err = xdp_do_generic_redirect(skb->dev, skb,
4484 generic_xdp_tx(skb, xdp_prog);
4495 EXPORT_SYMBOL_GPL(do_xdp_generic);
4497 static int netif_rx_internal(struct sk_buff *skb)
4501 net_timestamp_check(netdev_tstamp_prequeue, skb);
4503 trace_netif_rx(skb);
4506 if (static_branch_unlikely(&rps_needed)) {
4507 struct rps_dev_flow voidflow, *rflow = &voidflow;
4513 cpu = get_rps_cpu(skb->dev, skb, &rflow);
4515 cpu = smp_processor_id();
4517 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4526 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
4533 * netif_rx - post buffer to the network code
4534 * @skb: buffer to post
4536 * This function receives a packet from a device driver and queues it for
4537 * the upper (protocol) levels to process. It always succeeds. The buffer
4538 * may be dropped during processing for congestion control or by the
4542 * NET_RX_SUCCESS (no congestion)
4543 * NET_RX_DROP (packet was dropped)
4547 int netif_rx(struct sk_buff *skb)
4551 trace_netif_rx_entry(skb);
4553 ret = netif_rx_internal(skb);
4554 trace_netif_rx_exit(ret);
4558 EXPORT_SYMBOL(netif_rx);
4560 int netif_rx_ni(struct sk_buff *skb)
4564 trace_netif_rx_ni_entry(skb);
4567 err = netif_rx_internal(skb);
4568 if (local_softirq_pending())
4571 trace_netif_rx_ni_exit(err);
4575 EXPORT_SYMBOL(netif_rx_ni);
4577 static __latent_entropy void net_tx_action(struct softirq_action *h)
4579 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4581 if (sd->completion_queue) {
4582 struct sk_buff *clist;
4584 local_irq_disable();
4585 clist = sd->completion_queue;
4586 sd->completion_queue = NULL;
4590 struct sk_buff *skb = clist;
4592 clist = clist->next;
4594 WARN_ON(refcount_read(&skb->users));
4595 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
4596 trace_consume_skb(skb);
4598 trace_kfree_skb(skb, net_tx_action);
4600 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
4603 __kfree_skb_defer(skb);
4606 __kfree_skb_flush();
4609 if (sd->output_queue) {
4612 local_irq_disable();
4613 head = sd->output_queue;
4614 sd->output_queue = NULL;
4615 sd->output_queue_tailp = &sd->output_queue;
4619 struct Qdisc *q = head;
4620 spinlock_t *root_lock = NULL;
4622 head = head->next_sched;
4624 if (!(q->flags & TCQ_F_NOLOCK)) {
4625 root_lock = qdisc_lock(q);
4626 spin_lock(root_lock);
4628 /* We need to make sure head->next_sched is read
4629 * before clearing __QDISC_STATE_SCHED
4631 smp_mb__before_atomic();
4632 clear_bit(__QDISC_STATE_SCHED, &q->state);
4635 spin_unlock(root_lock);
4639 xfrm_dev_backlog(sd);
4642 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
4643 /* This hook is defined here for ATM LANE */
4644 int (*br_fdb_test_addr_hook)(struct net_device *dev,
4645 unsigned char *addr) __read_mostly;
4646 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
4649 static inline struct sk_buff *
4650 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
4651 struct net_device *orig_dev)
4653 #ifdef CONFIG_NET_CLS_ACT
4654 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
4655 struct tcf_result cl_res;
4657 /* If there's at least one ingress present somewhere (so
4658 * we get here via enabled static key), remaining devices
4659 * that are not configured with an ingress qdisc will bail
4666 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4670 qdisc_skb_cb(skb)->pkt_len = skb->len;
4671 skb->tc_at_ingress = 1;
4672 mini_qdisc_bstats_cpu_update(miniq, skb);
4674 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
4676 case TC_ACT_RECLASSIFY:
4677 skb->tc_index = TC_H_MIN(cl_res.classid);
4680 mini_qdisc_qstats_cpu_drop(miniq);
4688 case TC_ACT_REDIRECT:
4689 /* skb_mac_header check was done by cls/act_bpf, so
4690 * we can safely push the L2 header back before
4691 * redirecting to another netdev
4693 __skb_push(skb, skb->mac_len);
4694 skb_do_redirect(skb);
4696 case TC_ACT_REINSERT:
4697 /* this does not scrub the packet, and updates stats on error */
4698 skb_tc_reinsert(skb, &cl_res);
4703 #endif /* CONFIG_NET_CLS_ACT */
4708 * netdev_is_rx_handler_busy - check if receive handler is registered
4709 * @dev: device to check
4711 * Check if a receive handler is already registered for a given device.
4712 * Return true if there one.
4714 * The caller must hold the rtnl_mutex.
4716 bool netdev_is_rx_handler_busy(struct net_device *dev)
4719 return dev && rtnl_dereference(dev->rx_handler);
4721 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
4724 * netdev_rx_handler_register - register receive handler
4725 * @dev: device to register a handler for
4726 * @rx_handler: receive handler to register
4727 * @rx_handler_data: data pointer that is used by rx handler
4729 * Register a receive handler for a device. This handler will then be
4730 * called from __netif_receive_skb. A negative errno code is returned
4733 * The caller must hold the rtnl_mutex.
4735 * For a general description of rx_handler, see enum rx_handler_result.
4737 int netdev_rx_handler_register(struct net_device *dev,
4738 rx_handler_func_t *rx_handler,
4739 void *rx_handler_data)
4741 if (netdev_is_rx_handler_busy(dev))
4744 if (dev->priv_flags & IFF_NO_RX_HANDLER)
4747 /* Note: rx_handler_data must be set before rx_handler */
4748 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
4749 rcu_assign_pointer(dev->rx_handler, rx_handler);
4753 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
4756 * netdev_rx_handler_unregister - unregister receive handler
4757 * @dev: device to unregister a handler from
4759 * Unregister a receive handler from a device.
4761 * The caller must hold the rtnl_mutex.
4763 void netdev_rx_handler_unregister(struct net_device *dev)
4767 RCU_INIT_POINTER(dev->rx_handler, NULL);
4768 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
4769 * section has a guarantee to see a non NULL rx_handler_data
4773 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
4775 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
4778 * Limit the use of PFMEMALLOC reserves to those protocols that implement
4779 * the special handling of PFMEMALLOC skbs.
4781 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
4783 switch (skb->protocol) {
4784 case htons(ETH_P_ARP):
4785 case htons(ETH_P_IP):
4786 case htons(ETH_P_IPV6):
4787 case htons(ETH_P_8021Q):
4788 case htons(ETH_P_8021AD):
4795 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
4796 int *ret, struct net_device *orig_dev)
4798 #ifdef CONFIG_NETFILTER_INGRESS
4799 if (nf_hook_ingress_active(skb)) {
4803 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4808 ingress_retval = nf_hook_ingress(skb);
4810 return ingress_retval;
4812 #endif /* CONFIG_NETFILTER_INGRESS */
4816 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc,
4817 struct packet_type **ppt_prev)
4819 struct packet_type *ptype, *pt_prev;
4820 rx_handler_func_t *rx_handler;
4821 struct net_device *orig_dev;
4822 bool deliver_exact = false;
4823 int ret = NET_RX_DROP;
4826 net_timestamp_check(!netdev_tstamp_prequeue, skb);
4828 trace_netif_receive_skb(skb);
4830 orig_dev = skb->dev;
4832 skb_reset_network_header(skb);
4833 if (!skb_transport_header_was_set(skb))
4834 skb_reset_transport_header(skb);
4835 skb_reset_mac_len(skb);
4840 skb->skb_iif = skb->dev->ifindex;
4842 __this_cpu_inc(softnet_data.processed);
4844 if (static_branch_unlikely(&generic_xdp_needed_key)) {
4848 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
4851 if (ret2 != XDP_PASS)
4853 skb_reset_mac_len(skb);
4856 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
4857 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
4858 skb = skb_vlan_untag(skb);
4863 if (skb_skip_tc_classify(skb))
4869 list_for_each_entry_rcu(ptype, &ptype_all, list) {
4871 ret = deliver_skb(skb, pt_prev, orig_dev);
4875 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
4877 ret = deliver_skb(skb, pt_prev, orig_dev);
4882 #ifdef CONFIG_NET_INGRESS
4883 if (static_branch_unlikely(&ingress_needed_key)) {
4884 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev);
4888 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
4894 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
4897 if (skb_vlan_tag_present(skb)) {
4899 ret = deliver_skb(skb, pt_prev, orig_dev);
4902 if (vlan_do_receive(&skb))
4904 else if (unlikely(!skb))
4908 rx_handler = rcu_dereference(skb->dev->rx_handler);
4911 ret = deliver_skb(skb, pt_prev, orig_dev);
4914 switch (rx_handler(&skb)) {
4915 case RX_HANDLER_CONSUMED:
4916 ret = NET_RX_SUCCESS;
4918 case RX_HANDLER_ANOTHER:
4920 case RX_HANDLER_EXACT:
4921 deliver_exact = true;
4922 case RX_HANDLER_PASS:
4929 if (unlikely(skb_vlan_tag_present(skb))) {
4930 if (skb_vlan_tag_get_id(skb))
4931 skb->pkt_type = PACKET_OTHERHOST;
4932 /* Note: we might in the future use prio bits
4933 * and set skb->priority like in vlan_do_receive()
4934 * For the time being, just ignore Priority Code Point
4936 __vlan_hwaccel_clear_tag(skb);
4939 type = skb->protocol;
4941 /* deliver only exact match when indicated */
4942 if (likely(!deliver_exact)) {
4943 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4944 &ptype_base[ntohs(type) &
4948 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4949 &orig_dev->ptype_specific);
4951 if (unlikely(skb->dev != orig_dev)) {
4952 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4953 &skb->dev->ptype_specific);
4957 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
4959 *ppt_prev = pt_prev;
4963 atomic_long_inc(&skb->dev->rx_dropped);
4965 atomic_long_inc(&skb->dev->rx_nohandler);
4967 /* Jamal, now you will not able to escape explaining
4968 * me how you were going to use this. :-)
4977 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
4979 struct net_device *orig_dev = skb->dev;
4980 struct packet_type *pt_prev = NULL;
4983 ret = __netif_receive_skb_core(skb, pfmemalloc, &pt_prev);
4985 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
4986 skb->dev, pt_prev, orig_dev);
4991 * netif_receive_skb_core - special purpose version of netif_receive_skb
4992 * @skb: buffer to process
4994 * More direct receive version of netif_receive_skb(). It should
4995 * only be used by callers that have a need to skip RPS and Generic XDP.
4996 * Caller must also take care of handling if (page_is_)pfmemalloc.
4998 * This function may only be called from softirq context and interrupts
4999 * should be enabled.
5001 * Return values (usually ignored):
5002 * NET_RX_SUCCESS: no congestion
5003 * NET_RX_DROP: packet was dropped
5005 int netif_receive_skb_core(struct sk_buff *skb)
5010 ret = __netif_receive_skb_one_core(skb, false);
5015 EXPORT_SYMBOL(netif_receive_skb_core);
5017 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5018 struct packet_type *pt_prev,
5019 struct net_device *orig_dev)
5021 struct sk_buff *skb, *next;
5025 if (list_empty(head))
5027 if (pt_prev->list_func != NULL)
5028 pt_prev->list_func(head, pt_prev, orig_dev);
5030 list_for_each_entry_safe(skb, next, head, list) {
5031 skb_list_del_init(skb);
5032 INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
5033 skb->dev, pt_prev, orig_dev);
5037 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5039 /* Fast-path assumptions:
5040 * - There is no RX handler.
5041 * - Only one packet_type matches.
5042 * If either of these fails, we will end up doing some per-packet
5043 * processing in-line, then handling the 'last ptype' for the whole
5044 * sublist. This can't cause out-of-order delivery to any single ptype,
5045 * because the 'last ptype' must be constant across the sublist, and all
5046 * other ptypes are handled per-packet.
5048 /* Current (common) ptype of sublist */
5049 struct packet_type *pt_curr = NULL;
5050 /* Current (common) orig_dev of sublist */
5051 struct net_device *od_curr = NULL;
5052 struct list_head sublist;
5053 struct sk_buff *skb, *next;
5055 INIT_LIST_HEAD(&sublist);
5056 list_for_each_entry_safe(skb, next, head, list) {
5057 struct net_device *orig_dev = skb->dev;
5058 struct packet_type *pt_prev = NULL;
5060 skb_list_del_init(skb);
5061 __netif_receive_skb_core(skb, pfmemalloc, &pt_prev);
5064 if (pt_curr != pt_prev || od_curr != orig_dev) {
5065 /* dispatch old sublist */
5066 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5067 /* start new sublist */
5068 INIT_LIST_HEAD(&sublist);
5072 list_add_tail(&skb->list, &sublist);
5075 /* dispatch final sublist */
5076 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5079 static int __netif_receive_skb(struct sk_buff *skb)
5083 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5084 unsigned int noreclaim_flag;
5087 * PFMEMALLOC skbs are special, they should
5088 * - be delivered to SOCK_MEMALLOC sockets only
5089 * - stay away from userspace
5090 * - have bounded memory usage
5092 * Use PF_MEMALLOC as this saves us from propagating the allocation
5093 * context down to all allocation sites.
5095 noreclaim_flag = memalloc_noreclaim_save();
5096 ret = __netif_receive_skb_one_core(skb, true);
5097 memalloc_noreclaim_restore(noreclaim_flag);
5099 ret = __netif_receive_skb_one_core(skb, false);
5104 static void __netif_receive_skb_list(struct list_head *head)
5106 unsigned long noreclaim_flag = 0;
5107 struct sk_buff *skb, *next;
5108 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5110 list_for_each_entry_safe(skb, next, head, list) {
5111 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5112 struct list_head sublist;
5114 /* Handle the previous sublist */
5115 list_cut_before(&sublist, head, &skb->list);
5116 if (!list_empty(&sublist))
5117 __netif_receive_skb_list_core(&sublist, pfmemalloc);
5118 pfmemalloc = !pfmemalloc;
5119 /* See comments in __netif_receive_skb */
5121 noreclaim_flag = memalloc_noreclaim_save();
5123 memalloc_noreclaim_restore(noreclaim_flag);
5126 /* Handle the remaining sublist */
5127 if (!list_empty(head))
5128 __netif_receive_skb_list_core(head, pfmemalloc);
5129 /* Restore pflags */
5131 memalloc_noreclaim_restore(noreclaim_flag);
5134 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5136 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5137 struct bpf_prog *new = xdp->prog;
5140 switch (xdp->command) {
5141 case XDP_SETUP_PROG:
5142 rcu_assign_pointer(dev->xdp_prog, new);
5147 static_branch_dec(&generic_xdp_needed_key);
5148 } else if (new && !old) {
5149 static_branch_inc(&generic_xdp_needed_key);
5150 dev_disable_lro(dev);
5151 dev_disable_gro_hw(dev);
5155 case XDP_QUERY_PROG:
5156 xdp->prog_id = old ? old->aux->id : 0;
5167 static int netif_receive_skb_internal(struct sk_buff *skb)
5171 net_timestamp_check(netdev_tstamp_prequeue, skb);
5173 if (skb_defer_rx_timestamp(skb))
5174 return NET_RX_SUCCESS;
5178 if (static_branch_unlikely(&rps_needed)) {
5179 struct rps_dev_flow voidflow, *rflow = &voidflow;
5180 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5183 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5189 ret = __netif_receive_skb(skb);
5194 static void netif_receive_skb_list_internal(struct list_head *head)
5196 struct sk_buff *skb, *next;
5197 struct list_head sublist;
5199 INIT_LIST_HEAD(&sublist);
5200 list_for_each_entry_safe(skb, next, head, list) {
5201 net_timestamp_check(netdev_tstamp_prequeue, skb);
5202 skb_list_del_init(skb);
5203 if (!skb_defer_rx_timestamp(skb))
5204 list_add_tail(&skb->list, &sublist);
5206 list_splice_init(&sublist, head);
5210 if (static_branch_unlikely(&rps_needed)) {
5211 list_for_each_entry_safe(skb, next, head, list) {
5212 struct rps_dev_flow voidflow, *rflow = &voidflow;
5213 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5216 /* Will be handled, remove from list */
5217 skb_list_del_init(skb);
5218 enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5223 __netif_receive_skb_list(head);
5228 * netif_receive_skb - process receive buffer from network
5229 * @skb: buffer to process
5231 * netif_receive_skb() is the main receive data processing function.
5232 * It always succeeds. The buffer may be dropped during processing
5233 * for congestion control or by the protocol layers.
5235 * This function may only be called from softirq context and interrupts
5236 * should be enabled.
5238 * Return values (usually ignored):
5239 * NET_RX_SUCCESS: no congestion
5240 * NET_RX_DROP: packet was dropped
5242 int netif_receive_skb(struct sk_buff *skb)
5246 trace_netif_receive_skb_entry(skb);
5248 ret = netif_receive_skb_internal(skb);
5249 trace_netif_receive_skb_exit(ret);
5253 EXPORT_SYMBOL(netif_receive_skb);
5256 * netif_receive_skb_list - process many receive buffers from network
5257 * @head: list of skbs to process.
5259 * Since return value of netif_receive_skb() is normally ignored, and
5260 * wouldn't be meaningful for a list, this function returns void.
5262 * This function may only be called from softirq context and interrupts
5263 * should be enabled.
5265 void netif_receive_skb_list(struct list_head *head)
5267 struct sk_buff *skb;
5269 if (list_empty(head))
5271 if (trace_netif_receive_skb_list_entry_enabled()) {
5272 list_for_each_entry(skb, head, list)
5273 trace_netif_receive_skb_list_entry(skb);
5275 netif_receive_skb_list_internal(head);
5276 trace_netif_receive_skb_list_exit(0);
5278 EXPORT_SYMBOL(netif_receive_skb_list);
5280 DEFINE_PER_CPU(struct work_struct, flush_works);
5282 /* Network device is going away, flush any packets still pending */
5283 static void flush_backlog(struct work_struct *work)
5285 struct sk_buff *skb, *tmp;
5286 struct softnet_data *sd;
5289 sd = this_cpu_ptr(&softnet_data);
5291 local_irq_disable();
5293 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5294 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5295 __skb_unlink(skb, &sd->input_pkt_queue);
5297 input_queue_head_incr(sd);
5303 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5304 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5305 __skb_unlink(skb, &sd->process_queue);
5307 input_queue_head_incr(sd);
5313 static void flush_all_backlogs(void)
5319 for_each_online_cpu(cpu)
5320 queue_work_on(cpu, system_highpri_wq,
5321 per_cpu_ptr(&flush_works, cpu));
5323 for_each_online_cpu(cpu)
5324 flush_work(per_cpu_ptr(&flush_works, cpu));
5329 INDIRECT_CALLABLE_DECLARE(int inet_gro_complete(struct sk_buff *, int));
5330 INDIRECT_CALLABLE_DECLARE(int ipv6_gro_complete(struct sk_buff *, int));
5331 static int napi_gro_complete(struct sk_buff *skb)
5333 struct packet_offload *ptype;
5334 __be16 type = skb->protocol;
5335 struct list_head *head = &offload_base;
5338 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
5340 if (NAPI_GRO_CB(skb)->count == 1) {
5341 skb_shinfo(skb)->gso_size = 0;
5346 list_for_each_entry_rcu(ptype, head, list) {
5347 if (ptype->type != type || !ptype->callbacks.gro_complete)
5350 err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete,
5351 ipv6_gro_complete, inet_gro_complete,
5358 WARN_ON(&ptype->list == head);
5360 return NET_RX_SUCCESS;
5364 return netif_receive_skb_internal(skb);
5367 static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index,
5370 struct list_head *head = &napi->gro_hash[index].list;
5371 struct sk_buff *skb, *p;
5373 list_for_each_entry_safe_reverse(skb, p, head, list) {
5374 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
5376 skb_list_del_init(skb);
5377 napi_gro_complete(skb);
5378 napi->gro_hash[index].count--;
5381 if (!napi->gro_hash[index].count)
5382 __clear_bit(index, &napi->gro_bitmask);
5385 /* napi->gro_hash[].list contains packets ordered by age.
5386 * youngest packets at the head of it.
5387 * Complete skbs in reverse order to reduce latencies.
5389 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
5391 unsigned long bitmask = napi->gro_bitmask;
5392 unsigned int i, base = ~0U;
5394 while ((i = ffs(bitmask)) != 0) {
5397 __napi_gro_flush_chain(napi, base, flush_old);
5400 EXPORT_SYMBOL(napi_gro_flush);
5402 static struct list_head *gro_list_prepare(struct napi_struct *napi,
5403 struct sk_buff *skb)
5405 unsigned int maclen = skb->dev->hard_header_len;
5406 u32 hash = skb_get_hash_raw(skb);
5407 struct list_head *head;
5410 head = &napi->gro_hash[hash & (GRO_HASH_BUCKETS - 1)].list;
5411 list_for_each_entry(p, head, list) {
5412 unsigned long diffs;
5414 NAPI_GRO_CB(p)->flush = 0;
5416 if (hash != skb_get_hash_raw(p)) {
5417 NAPI_GRO_CB(p)->same_flow = 0;
5421 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
5422 diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb);
5423 if (skb_vlan_tag_present(p))
5424 diffs |= p->vlan_tci ^ skb->vlan_tci;
5425 diffs |= skb_metadata_dst_cmp(p, skb);
5426 diffs |= skb_metadata_differs(p, skb);
5427 if (maclen == ETH_HLEN)
5428 diffs |= compare_ether_header(skb_mac_header(p),
5429 skb_mac_header(skb));
5431 diffs = memcmp(skb_mac_header(p),
5432 skb_mac_header(skb),
5434 NAPI_GRO_CB(p)->same_flow = !diffs;
5440 static void skb_gro_reset_offset(struct sk_buff *skb)
5442 const struct skb_shared_info *pinfo = skb_shinfo(skb);
5443 const skb_frag_t *frag0 = &pinfo->frags[0];
5445 NAPI_GRO_CB(skb)->data_offset = 0;
5446 NAPI_GRO_CB(skb)->frag0 = NULL;
5447 NAPI_GRO_CB(skb)->frag0_len = 0;
5449 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
5451 !PageHighMem(skb_frag_page(frag0))) {
5452 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
5453 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
5454 skb_frag_size(frag0),
5455 skb->end - skb->tail);
5459 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
5461 struct skb_shared_info *pinfo = skb_shinfo(skb);
5463 BUG_ON(skb->end - skb->tail < grow);
5465 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
5467 skb->data_len -= grow;
5470 pinfo->frags[0].page_offset += grow;
5471 skb_frag_size_sub(&pinfo->frags[0], grow);
5473 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
5474 skb_frag_unref(skb, 0);
5475 memmove(pinfo->frags, pinfo->frags + 1,
5476 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
5480 static void gro_flush_oldest(struct list_head *head)
5482 struct sk_buff *oldest;
5484 oldest = list_last_entry(head, struct sk_buff, list);
5486 /* We are called with head length >= MAX_GRO_SKBS, so this is
5489 if (WARN_ON_ONCE(!oldest))
5492 /* Do not adjust napi->gro_hash[].count, caller is adding a new
5495 skb_list_del_init(oldest);
5496 napi_gro_complete(oldest);
5499 INDIRECT_CALLABLE_DECLARE(struct sk_buff *inet_gro_receive(struct list_head *,
5501 INDIRECT_CALLABLE_DECLARE(struct sk_buff *ipv6_gro_receive(struct list_head *,
5503 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
5505 u32 hash = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1);
5506 struct list_head *head = &offload_base;
5507 struct packet_offload *ptype;
5508 __be16 type = skb->protocol;
5509 struct list_head *gro_head;
5510 struct sk_buff *pp = NULL;
5511 enum gro_result ret;
5515 if (netif_elide_gro(skb->dev))
5518 gro_head = gro_list_prepare(napi, skb);
5521 list_for_each_entry_rcu(ptype, head, list) {
5522 if (ptype->type != type || !ptype->callbacks.gro_receive)
5525 skb_set_network_header(skb, skb_gro_offset(skb));
5526 skb_reset_mac_len(skb);
5527 NAPI_GRO_CB(skb)->same_flow = 0;
5528 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb);
5529 NAPI_GRO_CB(skb)->free = 0;
5530 NAPI_GRO_CB(skb)->encap_mark = 0;
5531 NAPI_GRO_CB(skb)->recursion_counter = 0;
5532 NAPI_GRO_CB(skb)->is_fou = 0;
5533 NAPI_GRO_CB(skb)->is_atomic = 1;
5534 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
5536 /* Setup for GRO checksum validation */
5537 switch (skb->ip_summed) {
5538 case CHECKSUM_COMPLETE:
5539 NAPI_GRO_CB(skb)->csum = skb->csum;
5540 NAPI_GRO_CB(skb)->csum_valid = 1;
5541 NAPI_GRO_CB(skb)->csum_cnt = 0;
5543 case CHECKSUM_UNNECESSARY:
5544 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
5545 NAPI_GRO_CB(skb)->csum_valid = 0;
5548 NAPI_GRO_CB(skb)->csum_cnt = 0;
5549 NAPI_GRO_CB(skb)->csum_valid = 0;
5552 pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive,
5553 ipv6_gro_receive, inet_gro_receive,
5559 if (&ptype->list == head)
5562 if (IS_ERR(pp) && PTR_ERR(pp) == -EINPROGRESS) {
5567 same_flow = NAPI_GRO_CB(skb)->same_flow;
5568 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
5571 skb_list_del_init(pp);
5572 napi_gro_complete(pp);
5573 napi->gro_hash[hash].count--;
5579 if (NAPI_GRO_CB(skb)->flush)
5582 if (unlikely(napi->gro_hash[hash].count >= MAX_GRO_SKBS)) {
5583 gro_flush_oldest(gro_head);
5585 napi->gro_hash[hash].count++;
5587 NAPI_GRO_CB(skb)->count = 1;
5588 NAPI_GRO_CB(skb)->age = jiffies;
5589 NAPI_GRO_CB(skb)->last = skb;
5590 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
5591 list_add(&skb->list, gro_head);
5595 grow = skb_gro_offset(skb) - skb_headlen(skb);
5597 gro_pull_from_frag0(skb, grow);
5599 if (napi->gro_hash[hash].count) {
5600 if (!test_bit(hash, &napi->gro_bitmask))
5601 __set_bit(hash, &napi->gro_bitmask);
5602 } else if (test_bit(hash, &napi->gro_bitmask)) {
5603 __clear_bit(hash, &napi->gro_bitmask);
5613 struct packet_offload *gro_find_receive_by_type(__be16 type)
5615 struct list_head *offload_head = &offload_base;
5616 struct packet_offload *ptype;
5618 list_for_each_entry_rcu(ptype, offload_head, list) {
5619 if (ptype->type != type || !ptype->callbacks.gro_receive)
5625 EXPORT_SYMBOL(gro_find_receive_by_type);
5627 struct packet_offload *gro_find_complete_by_type(__be16 type)
5629 struct list_head *offload_head = &offload_base;
5630 struct packet_offload *ptype;
5632 list_for_each_entry_rcu(ptype, offload_head, list) {
5633 if (ptype->type != type || !ptype->callbacks.gro_complete)
5639 EXPORT_SYMBOL(gro_find_complete_by_type);
5641 static void napi_skb_free_stolen_head(struct sk_buff *skb)
5645 kmem_cache_free(skbuff_head_cache, skb);
5648 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
5652 if (netif_receive_skb_internal(skb))
5660 case GRO_MERGED_FREE:
5661 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
5662 napi_skb_free_stolen_head(skb);
5676 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
5680 skb_mark_napi_id(skb, napi);
5681 trace_napi_gro_receive_entry(skb);
5683 skb_gro_reset_offset(skb);
5685 ret = napi_skb_finish(dev_gro_receive(napi, skb), skb);
5686 trace_napi_gro_receive_exit(ret);
5690 EXPORT_SYMBOL(napi_gro_receive);
5692 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
5694 if (unlikely(skb->pfmemalloc)) {
5698 __skb_pull(skb, skb_headlen(skb));
5699 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
5700 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
5701 __vlan_hwaccel_clear_tag(skb);
5702 skb->dev = napi->dev;
5705 /* eth_type_trans() assumes pkt_type is PACKET_HOST */
5706 skb->pkt_type = PACKET_HOST;
5708 skb->encapsulation = 0;
5709 skb_shinfo(skb)->gso_type = 0;
5710 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
5716 struct sk_buff *napi_get_frags(struct napi_struct *napi)
5718 struct sk_buff *skb = napi->skb;
5721 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
5724 skb_mark_napi_id(skb, napi);
5729 EXPORT_SYMBOL(napi_get_frags);
5731 static gro_result_t napi_frags_finish(struct napi_struct *napi,
5732 struct sk_buff *skb,
5738 __skb_push(skb, ETH_HLEN);
5739 skb->protocol = eth_type_trans(skb, skb->dev);
5740 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
5745 napi_reuse_skb(napi, skb);
5748 case GRO_MERGED_FREE:
5749 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
5750 napi_skb_free_stolen_head(skb);
5752 napi_reuse_skb(napi, skb);
5763 /* Upper GRO stack assumes network header starts at gro_offset=0
5764 * Drivers could call both napi_gro_frags() and napi_gro_receive()
5765 * We copy ethernet header into skb->data to have a common layout.
5767 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
5769 struct sk_buff *skb = napi->skb;
5770 const struct ethhdr *eth;
5771 unsigned int hlen = sizeof(*eth);
5775 skb_reset_mac_header(skb);
5776 skb_gro_reset_offset(skb);
5778 if (unlikely(skb_gro_header_hard(skb, hlen))) {
5779 eth = skb_gro_header_slow(skb, hlen, 0);
5780 if (unlikely(!eth)) {
5781 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
5782 __func__, napi->dev->name);
5783 napi_reuse_skb(napi, skb);
5787 eth = (const struct ethhdr *)skb->data;
5788 gro_pull_from_frag0(skb, hlen);
5789 NAPI_GRO_CB(skb)->frag0 += hlen;
5790 NAPI_GRO_CB(skb)->frag0_len -= hlen;
5792 __skb_pull(skb, hlen);
5795 * This works because the only protocols we care about don't require
5797 * We'll fix it up properly in napi_frags_finish()
5799 skb->protocol = eth->h_proto;
5804 gro_result_t napi_gro_frags(struct napi_struct *napi)
5807 struct sk_buff *skb = napi_frags_skb(napi);
5812 trace_napi_gro_frags_entry(skb);
5814 ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
5815 trace_napi_gro_frags_exit(ret);
5819 EXPORT_SYMBOL(napi_gro_frags);
5821 /* Compute the checksum from gro_offset and return the folded value
5822 * after adding in any pseudo checksum.
5824 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
5829 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
5831 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
5832 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
5833 /* See comments in __skb_checksum_complete(). */
5835 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
5836 !skb->csum_complete_sw)
5837 netdev_rx_csum_fault(skb->dev, skb);
5840 NAPI_GRO_CB(skb)->csum = wsum;
5841 NAPI_GRO_CB(skb)->csum_valid = 1;
5845 EXPORT_SYMBOL(__skb_gro_checksum_complete);
5847 static void net_rps_send_ipi(struct softnet_data *remsd)
5851 struct softnet_data *next = remsd->rps_ipi_next;
5853 if (cpu_online(remsd->cpu))
5854 smp_call_function_single_async(remsd->cpu, &remsd->csd);
5861 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
5862 * Note: called with local irq disabled, but exits with local irq enabled.
5864 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
5867 struct softnet_data *remsd = sd->rps_ipi_list;
5870 sd->rps_ipi_list = NULL;
5874 /* Send pending IPI's to kick RPS processing on remote cpus. */
5875 net_rps_send_ipi(remsd);
5881 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
5884 return sd->rps_ipi_list != NULL;
5890 static int process_backlog(struct napi_struct *napi, int quota)
5892 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
5896 /* Check if we have pending ipi, its better to send them now,
5897 * not waiting net_rx_action() end.
5899 if (sd_has_rps_ipi_waiting(sd)) {
5900 local_irq_disable();
5901 net_rps_action_and_irq_enable(sd);
5904 napi->weight = dev_rx_weight;
5906 struct sk_buff *skb;
5908 while ((skb = __skb_dequeue(&sd->process_queue))) {
5910 __netif_receive_skb(skb);
5912 input_queue_head_incr(sd);
5913 if (++work >= quota)
5918 local_irq_disable();
5920 if (skb_queue_empty(&sd->input_pkt_queue)) {
5922 * Inline a custom version of __napi_complete().
5923 * only current cpu owns and manipulates this napi,
5924 * and NAPI_STATE_SCHED is the only possible flag set
5926 * We can use a plain write instead of clear_bit(),
5927 * and we dont need an smp_mb() memory barrier.
5932 skb_queue_splice_tail_init(&sd->input_pkt_queue,
5933 &sd->process_queue);
5943 * __napi_schedule - schedule for receive
5944 * @n: entry to schedule
5946 * The entry's receive function will be scheduled to run.
5947 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
5949 void __napi_schedule(struct napi_struct *n)
5951 unsigned long flags;
5953 local_irq_save(flags);
5954 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
5955 local_irq_restore(flags);
5957 EXPORT_SYMBOL(__napi_schedule);
5960 * napi_schedule_prep - check if napi can be scheduled
5963 * Test if NAPI routine is already running, and if not mark
5964 * it as running. This is used as a condition variable
5965 * insure only one NAPI poll instance runs. We also make
5966 * sure there is no pending NAPI disable.
5968 bool napi_schedule_prep(struct napi_struct *n)
5970 unsigned long val, new;
5973 val = READ_ONCE(n->state);
5974 if (unlikely(val & NAPIF_STATE_DISABLE))
5976 new = val | NAPIF_STATE_SCHED;
5978 /* Sets STATE_MISSED bit if STATE_SCHED was already set
5979 * This was suggested by Alexander Duyck, as compiler
5980 * emits better code than :
5981 * if (val & NAPIF_STATE_SCHED)
5982 * new |= NAPIF_STATE_MISSED;
5984 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
5986 } while (cmpxchg(&n->state, val, new) != val);
5988 return !(val & NAPIF_STATE_SCHED);
5990 EXPORT_SYMBOL(napi_schedule_prep);
5993 * __napi_schedule_irqoff - schedule for receive
5994 * @n: entry to schedule
5996 * Variant of __napi_schedule() assuming hard irqs are masked
5998 void __napi_schedule_irqoff(struct napi_struct *n)
6000 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6002 EXPORT_SYMBOL(__napi_schedule_irqoff);
6004 bool napi_complete_done(struct napi_struct *n, int work_done)
6006 unsigned long flags, val, new;
6009 * 1) Don't let napi dequeue from the cpu poll list
6010 * just in case its running on a different cpu.
6011 * 2) If we are busy polling, do nothing here, we have
6012 * the guarantee we will be called later.
6014 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6015 NAPIF_STATE_IN_BUSY_POLL)))
6018 if (n->gro_bitmask) {
6019 unsigned long timeout = 0;
6022 timeout = n->dev->gro_flush_timeout;
6024 /* When the NAPI instance uses a timeout and keeps postponing
6025 * it, we need to bound somehow the time packets are kept in
6028 napi_gro_flush(n, !!timeout);
6030 hrtimer_start(&n->timer, ns_to_ktime(timeout),
6031 HRTIMER_MODE_REL_PINNED);
6033 if (unlikely(!list_empty(&n->poll_list))) {
6034 /* If n->poll_list is not empty, we need to mask irqs */
6035 local_irq_save(flags);
6036 list_del_init(&n->poll_list);
6037 local_irq_restore(flags);
6041 val = READ_ONCE(n->state);
6043 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6045 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED);
6047 /* If STATE_MISSED was set, leave STATE_SCHED set,
6048 * because we will call napi->poll() one more time.
6049 * This C code was suggested by Alexander Duyck to help gcc.
6051 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6053 } while (cmpxchg(&n->state, val, new) != val);
6055 if (unlikely(val & NAPIF_STATE_MISSED)) {
6062 EXPORT_SYMBOL(napi_complete_done);
6064 /* must be called under rcu_read_lock(), as we dont take a reference */
6065 static struct napi_struct *napi_by_id(unsigned int napi_id)
6067 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6068 struct napi_struct *napi;
6070 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6071 if (napi->napi_id == napi_id)
6077 #if defined(CONFIG_NET_RX_BUSY_POLL)
6079 #define BUSY_POLL_BUDGET 8
6081 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock)
6085 /* Busy polling means there is a high chance device driver hard irq
6086 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6087 * set in napi_schedule_prep().
6088 * Since we are about to call napi->poll() once more, we can safely
6089 * clear NAPI_STATE_MISSED.
6091 * Note: x86 could use a single "lock and ..." instruction
6092 * to perform these two clear_bit()
6094 clear_bit(NAPI_STATE_MISSED, &napi->state);
6095 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6099 /* All we really want here is to re-enable device interrupts.
6100 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6102 rc = napi->poll(napi, BUSY_POLL_BUDGET);
6103 trace_napi_poll(napi, rc, BUSY_POLL_BUDGET);
6104 netpoll_poll_unlock(have_poll_lock);
6105 if (rc == BUSY_POLL_BUDGET)
6106 __napi_schedule(napi);
6110 void napi_busy_loop(unsigned int napi_id,
6111 bool (*loop_end)(void *, unsigned long),
6114 unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6115 int (*napi_poll)(struct napi_struct *napi, int budget);
6116 void *have_poll_lock = NULL;
6117 struct napi_struct *napi;
6124 napi = napi_by_id(napi_id);
6134 unsigned long val = READ_ONCE(napi->state);
6136 /* If multiple threads are competing for this napi,
6137 * we avoid dirtying napi->state as much as we can.
6139 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6140 NAPIF_STATE_IN_BUSY_POLL))
6142 if (cmpxchg(&napi->state, val,
6143 val | NAPIF_STATE_IN_BUSY_POLL |
6144 NAPIF_STATE_SCHED) != val)
6146 have_poll_lock = netpoll_poll_lock(napi);
6147 napi_poll = napi->poll;
6149 work = napi_poll(napi, BUSY_POLL_BUDGET);
6150 trace_napi_poll(napi, work, BUSY_POLL_BUDGET);
6153 __NET_ADD_STATS(dev_net(napi->dev),
6154 LINUX_MIB_BUSYPOLLRXPACKETS, work);
6157 if (!loop_end || loop_end(loop_end_arg, start_time))
6160 if (unlikely(need_resched())) {
6162 busy_poll_stop(napi, have_poll_lock);
6166 if (loop_end(loop_end_arg, start_time))
6173 busy_poll_stop(napi, have_poll_lock);
6178 EXPORT_SYMBOL(napi_busy_loop);
6180 #endif /* CONFIG_NET_RX_BUSY_POLL */
6182 static void napi_hash_add(struct napi_struct *napi)
6184 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) ||
6185 test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
6188 spin_lock(&napi_hash_lock);
6190 /* 0..NR_CPUS range is reserved for sender_cpu use */
6192 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6193 napi_gen_id = MIN_NAPI_ID;
6194 } while (napi_by_id(napi_gen_id));
6195 napi->napi_id = napi_gen_id;
6197 hlist_add_head_rcu(&napi->napi_hash_node,
6198 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6200 spin_unlock(&napi_hash_lock);
6203 /* Warning : caller is responsible to make sure rcu grace period
6204 * is respected before freeing memory containing @napi
6206 bool napi_hash_del(struct napi_struct *napi)
6208 bool rcu_sync_needed = false;
6210 spin_lock(&napi_hash_lock);
6212 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) {
6213 rcu_sync_needed = true;
6214 hlist_del_rcu(&napi->napi_hash_node);
6216 spin_unlock(&napi_hash_lock);
6217 return rcu_sync_needed;
6219 EXPORT_SYMBOL_GPL(napi_hash_del);
6221 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6223 struct napi_struct *napi;
6225 napi = container_of(timer, struct napi_struct, timer);
6227 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
6228 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6230 if (napi->gro_bitmask && !napi_disable_pending(napi) &&
6231 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state))
6232 __napi_schedule_irqoff(napi);
6234 return HRTIMER_NORESTART;
6237 static void init_gro_hash(struct napi_struct *napi)
6241 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6242 INIT_LIST_HEAD(&napi->gro_hash[i].list);
6243 napi->gro_hash[i].count = 0;
6245 napi->gro_bitmask = 0;
6248 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
6249 int (*poll)(struct napi_struct *, int), int weight)
6251 INIT_LIST_HEAD(&napi->poll_list);
6252 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6253 napi->timer.function = napi_watchdog;
6254 init_gro_hash(napi);
6257 if (weight > NAPI_POLL_WEIGHT)
6258 netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6260 napi->weight = weight;
6261 list_add(&napi->dev_list, &dev->napi_list);
6263 #ifdef CONFIG_NETPOLL
6264 napi->poll_owner = -1;
6266 set_bit(NAPI_STATE_SCHED, &napi->state);
6267 napi_hash_add(napi);
6269 EXPORT_SYMBOL(netif_napi_add);
6271 void napi_disable(struct napi_struct *n)
6274 set_bit(NAPI_STATE_DISABLE, &n->state);
6276 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
6278 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
6281 hrtimer_cancel(&n->timer);
6283 clear_bit(NAPI_STATE_DISABLE, &n->state);
6285 EXPORT_SYMBOL(napi_disable);
6287 static void flush_gro_hash(struct napi_struct *napi)
6291 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6292 struct sk_buff *skb, *n;
6294 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6296 napi->gro_hash[i].count = 0;
6300 /* Must be called in process context */
6301 void netif_napi_del(struct napi_struct *napi)
6304 if (napi_hash_del(napi))
6306 list_del_init(&napi->dev_list);
6307 napi_free_frags(napi);
6309 flush_gro_hash(napi);
6310 napi->gro_bitmask = 0;
6312 EXPORT_SYMBOL(netif_napi_del);
6314 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
6319 list_del_init(&n->poll_list);
6321 have = netpoll_poll_lock(n);
6325 /* This NAPI_STATE_SCHED test is for avoiding a race
6326 * with netpoll's poll_napi(). Only the entity which
6327 * obtains the lock and sees NAPI_STATE_SCHED set will
6328 * actually make the ->poll() call. Therefore we avoid
6329 * accidentally calling ->poll() when NAPI is not scheduled.
6332 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
6333 work = n->poll(n, weight);
6334 trace_napi_poll(n, work, weight);
6337 WARN_ON_ONCE(work > weight);
6339 if (likely(work < weight))
6342 /* Drivers must not modify the NAPI state if they
6343 * consume the entire weight. In such cases this code
6344 * still "owns" the NAPI instance and therefore can
6345 * move the instance around on the list at-will.
6347 if (unlikely(napi_disable_pending(n))) {
6352 if (n->gro_bitmask) {
6353 /* flush too old packets
6354 * If HZ < 1000, flush all packets.
6356 napi_gro_flush(n, HZ >= 1000);
6359 /* Some drivers may have called napi_schedule
6360 * prior to exhausting their budget.
6362 if (unlikely(!list_empty(&n->poll_list))) {
6363 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
6364 n->dev ? n->dev->name : "backlog");
6368 list_add_tail(&n->poll_list, repoll);
6371 netpoll_poll_unlock(have);
6376 static __latent_entropy void net_rx_action(struct softirq_action *h)
6378 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6379 unsigned long time_limit = jiffies +
6380 usecs_to_jiffies(netdev_budget_usecs);
6381 int budget = netdev_budget;
6385 local_irq_disable();
6386 list_splice_init(&sd->poll_list, &list);
6390 struct napi_struct *n;
6392 if (list_empty(&list)) {
6393 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
6398 n = list_first_entry(&list, struct napi_struct, poll_list);
6399 budget -= napi_poll(n, &repoll);
6401 /* If softirq window is exhausted then punt.
6402 * Allow this to run for 2 jiffies since which will allow
6403 * an average latency of 1.5/HZ.
6405 if (unlikely(budget <= 0 ||
6406 time_after_eq(jiffies, time_limit))) {
6412 local_irq_disable();
6414 list_splice_tail_init(&sd->poll_list, &list);
6415 list_splice_tail(&repoll, &list);
6416 list_splice(&list, &sd->poll_list);
6417 if (!list_empty(&sd->poll_list))
6418 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
6420 net_rps_action_and_irq_enable(sd);
6422 __kfree_skb_flush();
6425 struct netdev_adjacent {
6426 struct net_device *dev;
6428 /* upper master flag, there can only be one master device per list */
6431 /* counter for the number of times this device was added to us */
6434 /* private field for the users */
6437 struct list_head list;
6438 struct rcu_head rcu;
6441 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
6442 struct list_head *adj_list)
6444 struct netdev_adjacent *adj;
6446 list_for_each_entry(adj, adj_list, list) {
6447 if (adj->dev == adj_dev)
6453 static int __netdev_has_upper_dev(struct net_device *upper_dev, void *data)
6455 struct net_device *dev = data;
6457 return upper_dev == dev;
6461 * netdev_has_upper_dev - Check if device is linked to an upper device
6463 * @upper_dev: upper device to check
6465 * Find out if a device is linked to specified upper device and return true
6466 * in case it is. Note that this checks only immediate upper device,
6467 * not through a complete stack of devices. The caller must hold the RTNL lock.
6469 bool netdev_has_upper_dev(struct net_device *dev,
6470 struct net_device *upper_dev)
6474 return netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
6477 EXPORT_SYMBOL(netdev_has_upper_dev);
6480 * netdev_has_upper_dev_all - Check if device is linked to an upper device
6482 * @upper_dev: upper device to check
6484 * Find out if a device is linked to specified upper device and return true
6485 * in case it is. Note that this checks the entire upper device chain.
6486 * The caller must hold rcu lock.
6489 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
6490 struct net_device *upper_dev)
6492 return !!netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
6495 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
6498 * netdev_has_any_upper_dev - Check if device is linked to some device
6501 * Find out if a device is linked to an upper device and return true in case
6502 * it is. The caller must hold the RTNL lock.
6504 bool netdev_has_any_upper_dev(struct net_device *dev)
6508 return !list_empty(&dev->adj_list.upper);
6510 EXPORT_SYMBOL(netdev_has_any_upper_dev);
6513 * netdev_master_upper_dev_get - Get master upper device
6516 * Find a master upper device and return pointer to it or NULL in case
6517 * it's not there. The caller must hold the RTNL lock.
6519 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
6521 struct netdev_adjacent *upper;
6525 if (list_empty(&dev->adj_list.upper))
6528 upper = list_first_entry(&dev->adj_list.upper,
6529 struct netdev_adjacent, list);
6530 if (likely(upper->master))
6534 EXPORT_SYMBOL(netdev_master_upper_dev_get);
6537 * netdev_has_any_lower_dev - Check if device is linked to some device
6540 * Find out if a device is linked to a lower device and return true in case
6541 * it is. The caller must hold the RTNL lock.
6543 static bool netdev_has_any_lower_dev(struct net_device *dev)
6547 return !list_empty(&dev->adj_list.lower);
6550 void *netdev_adjacent_get_private(struct list_head *adj_list)
6552 struct netdev_adjacent *adj;
6554 adj = list_entry(adj_list, struct netdev_adjacent, list);
6556 return adj->private;
6558 EXPORT_SYMBOL(netdev_adjacent_get_private);
6561 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
6563 * @iter: list_head ** of the current position
6565 * Gets the next device from the dev's upper list, starting from iter
6566 * position. The caller must hold RCU read lock.
6568 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
6569 struct list_head **iter)
6571 struct netdev_adjacent *upper;
6573 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6575 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6577 if (&upper->list == &dev->adj_list.upper)
6580 *iter = &upper->list;
6584 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
6586 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
6587 struct list_head **iter)
6589 struct netdev_adjacent *upper;
6591 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6593 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6595 if (&upper->list == &dev->adj_list.upper)
6598 *iter = &upper->list;
6603 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
6604 int (*fn)(struct net_device *dev,
6608 struct net_device *udev;
6609 struct list_head *iter;
6612 for (iter = &dev->adj_list.upper,
6613 udev = netdev_next_upper_dev_rcu(dev, &iter);
6615 udev = netdev_next_upper_dev_rcu(dev, &iter)) {
6616 /* first is the upper device itself */
6617 ret = fn(udev, data);
6621 /* then look at all of its upper devices */
6622 ret = netdev_walk_all_upper_dev_rcu(udev, fn, data);
6629 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
6632 * netdev_lower_get_next_private - Get the next ->private from the
6633 * lower neighbour list
6635 * @iter: list_head ** of the current position
6637 * Gets the next netdev_adjacent->private from the dev's lower neighbour
6638 * list, starting from iter position. The caller must hold either hold the
6639 * RTNL lock or its own locking that guarantees that the neighbour lower
6640 * list will remain unchanged.
6642 void *netdev_lower_get_next_private(struct net_device *dev,
6643 struct list_head **iter)
6645 struct netdev_adjacent *lower;
6647 lower = list_entry(*iter, struct netdev_adjacent, list);
6649 if (&lower->list == &dev->adj_list.lower)
6652 *iter = lower->list.next;
6654 return lower->private;
6656 EXPORT_SYMBOL(netdev_lower_get_next_private);
6659 * netdev_lower_get_next_private_rcu - Get the next ->private from the
6660 * lower neighbour list, RCU
6663 * @iter: list_head ** of the current position
6665 * Gets the next netdev_adjacent->private from the dev's lower neighbour
6666 * list, starting from iter position. The caller must hold RCU read lock.
6668 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
6669 struct list_head **iter)
6671 struct netdev_adjacent *lower;
6673 WARN_ON_ONCE(!rcu_read_lock_held());
6675 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6677 if (&lower->list == &dev->adj_list.lower)
6680 *iter = &lower->list;
6682 return lower->private;
6684 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
6687 * netdev_lower_get_next - Get the next device from the lower neighbour
6690 * @iter: list_head ** of the current position
6692 * Gets the next netdev_adjacent from the dev's lower neighbour
6693 * list, starting from iter position. The caller must hold RTNL lock or
6694 * its own locking that guarantees that the neighbour lower
6695 * list will remain unchanged.
6697 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
6699 struct netdev_adjacent *lower;
6701 lower = list_entry(*iter, struct netdev_adjacent, list);
6703 if (&lower->list == &dev->adj_list.lower)
6706 *iter = lower->list.next;
6710 EXPORT_SYMBOL(netdev_lower_get_next);
6712 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
6713 struct list_head **iter)
6715 struct netdev_adjacent *lower;
6717 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
6719 if (&lower->list == &dev->adj_list.lower)
6722 *iter = &lower->list;
6727 int netdev_walk_all_lower_dev(struct net_device *dev,
6728 int (*fn)(struct net_device *dev,
6732 struct net_device *ldev;
6733 struct list_head *iter;
6736 for (iter = &dev->adj_list.lower,
6737 ldev = netdev_next_lower_dev(dev, &iter);
6739 ldev = netdev_next_lower_dev(dev, &iter)) {
6740 /* first is the lower device itself */
6741 ret = fn(ldev, data);
6745 /* then look at all of its lower devices */
6746 ret = netdev_walk_all_lower_dev(ldev, fn, data);
6753 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
6755 static struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
6756 struct list_head **iter)
6758 struct netdev_adjacent *lower;
6760 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6761 if (&lower->list == &dev->adj_list.lower)
6764 *iter = &lower->list;
6769 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
6770 int (*fn)(struct net_device *dev,
6774 struct net_device *ldev;
6775 struct list_head *iter;
6778 for (iter = &dev->adj_list.lower,
6779 ldev = netdev_next_lower_dev_rcu(dev, &iter);
6781 ldev = netdev_next_lower_dev_rcu(dev, &iter)) {
6782 /* first is the lower device itself */
6783 ret = fn(ldev, data);
6787 /* then look at all of its lower devices */
6788 ret = netdev_walk_all_lower_dev_rcu(ldev, fn, data);
6795 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
6798 * netdev_lower_get_first_private_rcu - Get the first ->private from the
6799 * lower neighbour list, RCU
6803 * Gets the first netdev_adjacent->private from the dev's lower neighbour
6804 * list. The caller must hold RCU read lock.
6806 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
6808 struct netdev_adjacent *lower;
6810 lower = list_first_or_null_rcu(&dev->adj_list.lower,
6811 struct netdev_adjacent, list);
6813 return lower->private;
6816 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
6819 * netdev_master_upper_dev_get_rcu - Get master upper device
6822 * Find a master upper device and return pointer to it or NULL in case
6823 * it's not there. The caller must hold the RCU read lock.
6825 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
6827 struct netdev_adjacent *upper;
6829 upper = list_first_or_null_rcu(&dev->adj_list.upper,
6830 struct netdev_adjacent, list);
6831 if (upper && likely(upper->master))
6835 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
6837 static int netdev_adjacent_sysfs_add(struct net_device *dev,
6838 struct net_device *adj_dev,
6839 struct list_head *dev_list)
6841 char linkname[IFNAMSIZ+7];
6843 sprintf(linkname, dev_list == &dev->adj_list.upper ?
6844 "upper_%s" : "lower_%s", adj_dev->name);
6845 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
6848 static void netdev_adjacent_sysfs_del(struct net_device *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", name);
6856 sysfs_remove_link(&(dev->dev.kobj), linkname);
6859 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
6860 struct net_device *adj_dev,
6861 struct list_head *dev_list)
6863 return (dev_list == &dev->adj_list.upper ||
6864 dev_list == &dev->adj_list.lower) &&
6865 net_eq(dev_net(dev), dev_net(adj_dev));
6868 static int __netdev_adjacent_dev_insert(struct net_device *dev,
6869 struct net_device *adj_dev,
6870 struct list_head *dev_list,
6871 void *private, bool master)
6873 struct netdev_adjacent *adj;
6876 adj = __netdev_find_adj(adj_dev, dev_list);
6880 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
6881 dev->name, adj_dev->name, adj->ref_nr);
6886 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
6891 adj->master = master;
6893 adj->private = private;
6896 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
6897 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
6899 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
6900 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
6905 /* Ensure that master link is always the first item in list. */
6907 ret = sysfs_create_link(&(dev->dev.kobj),
6908 &(adj_dev->dev.kobj), "master");
6910 goto remove_symlinks;
6912 list_add_rcu(&adj->list, dev_list);
6914 list_add_tail_rcu(&adj->list, dev_list);
6920 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
6921 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
6929 static void __netdev_adjacent_dev_remove(struct net_device *dev,
6930 struct net_device *adj_dev,
6932 struct list_head *dev_list)
6934 struct netdev_adjacent *adj;
6936 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
6937 dev->name, adj_dev->name, ref_nr);
6939 adj = __netdev_find_adj(adj_dev, dev_list);
6942 pr_err("Adjacency does not exist for device %s from %s\n",
6943 dev->name, adj_dev->name);
6948 if (adj->ref_nr > ref_nr) {
6949 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
6950 dev->name, adj_dev->name, ref_nr,
6951 adj->ref_nr - ref_nr);
6952 adj->ref_nr -= ref_nr;
6957 sysfs_remove_link(&(dev->dev.kobj), "master");
6959 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
6960 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
6962 list_del_rcu(&adj->list);
6963 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
6964 adj_dev->name, dev->name, adj_dev->name);
6966 kfree_rcu(adj, rcu);
6969 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
6970 struct net_device *upper_dev,
6971 struct list_head *up_list,
6972 struct list_head *down_list,
6973 void *private, bool master)
6977 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
6982 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
6985 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
6992 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
6993 struct net_device *upper_dev,
6995 struct list_head *up_list,
6996 struct list_head *down_list)
6998 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
6999 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
7002 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
7003 struct net_device *upper_dev,
7004 void *private, bool master)
7006 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
7007 &dev->adj_list.upper,
7008 &upper_dev->adj_list.lower,
7012 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
7013 struct net_device *upper_dev)
7015 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
7016 &dev->adj_list.upper,
7017 &upper_dev->adj_list.lower);
7020 static int __netdev_upper_dev_link(struct net_device *dev,
7021 struct net_device *upper_dev, bool master,
7022 void *upper_priv, void *upper_info,
7023 struct netlink_ext_ack *extack)
7025 struct netdev_notifier_changeupper_info changeupper_info = {
7030 .upper_dev = upper_dev,
7033 .upper_info = upper_info,
7035 struct net_device *master_dev;
7040 if (dev == upper_dev)
7043 /* To prevent loops, check if dev is not upper device to upper_dev. */
7044 if (netdev_has_upper_dev(upper_dev, dev))
7048 if (netdev_has_upper_dev(dev, upper_dev))
7051 master_dev = netdev_master_upper_dev_get(dev);
7053 return master_dev == upper_dev ? -EEXIST : -EBUSY;
7056 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7057 &changeupper_info.info);
7058 ret = notifier_to_errno(ret);
7062 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
7067 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7068 &changeupper_info.info);
7069 ret = notifier_to_errno(ret);
7076 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7082 * netdev_upper_dev_link - Add a link to the upper device
7084 * @upper_dev: new upper device
7085 * @extack: netlink extended ack
7087 * Adds a link to device which is upper to this one. The caller must hold
7088 * the RTNL lock. On a failure a negative errno code is returned.
7089 * On success the reference counts are adjusted and the function
7092 int netdev_upper_dev_link(struct net_device *dev,
7093 struct net_device *upper_dev,
7094 struct netlink_ext_ack *extack)
7096 return __netdev_upper_dev_link(dev, upper_dev, false,
7097 NULL, NULL, extack);
7099 EXPORT_SYMBOL(netdev_upper_dev_link);
7102 * netdev_master_upper_dev_link - Add a master link to the upper device
7104 * @upper_dev: new upper device
7105 * @upper_priv: upper device private
7106 * @upper_info: upper info to be passed down via notifier
7107 * @extack: netlink extended ack
7109 * Adds a link to device which is upper to this one. In this case, only
7110 * one master upper device can be linked, although other non-master devices
7111 * might be linked as well. The caller must hold the RTNL lock.
7112 * On a failure a negative errno code is returned. On success the reference
7113 * counts are adjusted and the function returns zero.
7115 int netdev_master_upper_dev_link(struct net_device *dev,
7116 struct net_device *upper_dev,
7117 void *upper_priv, void *upper_info,
7118 struct netlink_ext_ack *extack)
7120 return __netdev_upper_dev_link(dev, upper_dev, true,
7121 upper_priv, upper_info, extack);
7123 EXPORT_SYMBOL(netdev_master_upper_dev_link);
7126 * netdev_upper_dev_unlink - Removes a link to upper device
7128 * @upper_dev: new upper device
7130 * Removes a link to device which is upper to this one. The caller must hold
7133 void netdev_upper_dev_unlink(struct net_device *dev,
7134 struct net_device *upper_dev)
7136 struct netdev_notifier_changeupper_info changeupper_info = {
7140 .upper_dev = upper_dev,
7146 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
7148 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7149 &changeupper_info.info);
7151 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7153 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7154 &changeupper_info.info);
7156 EXPORT_SYMBOL(netdev_upper_dev_unlink);
7159 * netdev_bonding_info_change - Dispatch event about slave change
7161 * @bonding_info: info to dispatch
7163 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
7164 * The caller must hold the RTNL lock.
7166 void netdev_bonding_info_change(struct net_device *dev,
7167 struct netdev_bonding_info *bonding_info)
7169 struct netdev_notifier_bonding_info info = {
7173 memcpy(&info.bonding_info, bonding_info,
7174 sizeof(struct netdev_bonding_info));
7175 call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
7178 EXPORT_SYMBOL(netdev_bonding_info_change);
7180 static void netdev_adjacent_add_links(struct net_device *dev)
7182 struct netdev_adjacent *iter;
7184 struct net *net = dev_net(dev);
7186 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7187 if (!net_eq(net, dev_net(iter->dev)))
7189 netdev_adjacent_sysfs_add(iter->dev, dev,
7190 &iter->dev->adj_list.lower);
7191 netdev_adjacent_sysfs_add(dev, iter->dev,
7192 &dev->adj_list.upper);
7195 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7196 if (!net_eq(net, dev_net(iter->dev)))
7198 netdev_adjacent_sysfs_add(iter->dev, dev,
7199 &iter->dev->adj_list.upper);
7200 netdev_adjacent_sysfs_add(dev, iter->dev,
7201 &dev->adj_list.lower);
7205 static void netdev_adjacent_del_links(struct net_device *dev)
7207 struct netdev_adjacent *iter;
7209 struct net *net = dev_net(dev);
7211 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7212 if (!net_eq(net, dev_net(iter->dev)))
7214 netdev_adjacent_sysfs_del(iter->dev, dev->name,
7215 &iter->dev->adj_list.lower);
7216 netdev_adjacent_sysfs_del(dev, iter->dev->name,
7217 &dev->adj_list.upper);
7220 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7221 if (!net_eq(net, dev_net(iter->dev)))
7223 netdev_adjacent_sysfs_del(iter->dev, dev->name,
7224 &iter->dev->adj_list.upper);
7225 netdev_adjacent_sysfs_del(dev, iter->dev->name,
7226 &dev->adj_list.lower);
7230 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
7232 struct netdev_adjacent *iter;
7234 struct net *net = dev_net(dev);
7236 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7237 if (!net_eq(net, dev_net(iter->dev)))
7239 netdev_adjacent_sysfs_del(iter->dev, oldname,
7240 &iter->dev->adj_list.lower);
7241 netdev_adjacent_sysfs_add(iter->dev, dev,
7242 &iter->dev->adj_list.lower);
7245 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7246 if (!net_eq(net, dev_net(iter->dev)))
7248 netdev_adjacent_sysfs_del(iter->dev, oldname,
7249 &iter->dev->adj_list.upper);
7250 netdev_adjacent_sysfs_add(iter->dev, dev,
7251 &iter->dev->adj_list.upper);
7255 void *netdev_lower_dev_get_private(struct net_device *dev,
7256 struct net_device *lower_dev)
7258 struct netdev_adjacent *lower;
7262 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
7266 return lower->private;
7268 EXPORT_SYMBOL(netdev_lower_dev_get_private);
7271 int dev_get_nest_level(struct net_device *dev)
7273 struct net_device *lower = NULL;
7274 struct list_head *iter;
7280 netdev_for_each_lower_dev(dev, lower, iter) {
7281 nest = dev_get_nest_level(lower);
7282 if (max_nest < nest)
7286 return max_nest + 1;
7288 EXPORT_SYMBOL(dev_get_nest_level);
7291 * netdev_lower_change - Dispatch event about lower device state change
7292 * @lower_dev: device
7293 * @lower_state_info: state to dispatch
7295 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
7296 * The caller must hold the RTNL lock.
7298 void netdev_lower_state_changed(struct net_device *lower_dev,
7299 void *lower_state_info)
7301 struct netdev_notifier_changelowerstate_info changelowerstate_info = {
7302 .info.dev = lower_dev,
7306 changelowerstate_info.lower_state_info = lower_state_info;
7307 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
7308 &changelowerstate_info.info);
7310 EXPORT_SYMBOL(netdev_lower_state_changed);
7312 static void dev_change_rx_flags(struct net_device *dev, int flags)
7314 const struct net_device_ops *ops = dev->netdev_ops;
7316 if (ops->ndo_change_rx_flags)
7317 ops->ndo_change_rx_flags(dev, flags);
7320 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
7322 unsigned int old_flags = dev->flags;
7328 dev->flags |= IFF_PROMISC;
7329 dev->promiscuity += inc;
7330 if (dev->promiscuity == 0) {
7333 * If inc causes overflow, untouch promisc and return error.
7336 dev->flags &= ~IFF_PROMISC;
7338 dev->promiscuity -= inc;
7339 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
7344 if (dev->flags != old_flags) {
7345 pr_info("device %s %s promiscuous mode\n",
7347 dev->flags & IFF_PROMISC ? "entered" : "left");
7348 if (audit_enabled) {
7349 current_uid_gid(&uid, &gid);
7350 audit_log(audit_context(), GFP_ATOMIC,
7351 AUDIT_ANOM_PROMISCUOUS,
7352 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
7353 dev->name, (dev->flags & IFF_PROMISC),
7354 (old_flags & IFF_PROMISC),
7355 from_kuid(&init_user_ns, audit_get_loginuid(current)),
7356 from_kuid(&init_user_ns, uid),
7357 from_kgid(&init_user_ns, gid),
7358 audit_get_sessionid(current));
7361 dev_change_rx_flags(dev, IFF_PROMISC);
7364 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
7369 * dev_set_promiscuity - update promiscuity count on a device
7373 * Add or remove promiscuity from a device. While the count in the device
7374 * remains above zero the interface remains promiscuous. Once it hits zero
7375 * the device reverts back to normal filtering operation. A negative inc
7376 * value is used to drop promiscuity on the device.
7377 * Return 0 if successful or a negative errno code on error.
7379 int dev_set_promiscuity(struct net_device *dev, int inc)
7381 unsigned int old_flags = dev->flags;
7384 err = __dev_set_promiscuity(dev, inc, true);
7387 if (dev->flags != old_flags)
7388 dev_set_rx_mode(dev);
7391 EXPORT_SYMBOL(dev_set_promiscuity);
7393 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
7395 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
7399 dev->flags |= IFF_ALLMULTI;
7400 dev->allmulti += inc;
7401 if (dev->allmulti == 0) {
7404 * If inc causes overflow, untouch allmulti and return error.
7407 dev->flags &= ~IFF_ALLMULTI;
7409 dev->allmulti -= inc;
7410 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
7415 if (dev->flags ^ old_flags) {
7416 dev_change_rx_flags(dev, IFF_ALLMULTI);
7417 dev_set_rx_mode(dev);
7419 __dev_notify_flags(dev, old_flags,
7420 dev->gflags ^ old_gflags);
7426 * dev_set_allmulti - update allmulti count on a device
7430 * Add or remove reception of all multicast frames to a device. While the
7431 * count in the device remains above zero the interface remains listening
7432 * to all interfaces. Once it hits zero the device reverts back to normal
7433 * filtering operation. A negative @inc value is used to drop the counter
7434 * when releasing a resource needing all multicasts.
7435 * Return 0 if successful or a negative errno code on error.
7438 int dev_set_allmulti(struct net_device *dev, int inc)
7440 return __dev_set_allmulti(dev, inc, true);
7442 EXPORT_SYMBOL(dev_set_allmulti);
7445 * Upload unicast and multicast address lists to device and
7446 * configure RX filtering. When the device doesn't support unicast
7447 * filtering it is put in promiscuous mode while unicast addresses
7450 void __dev_set_rx_mode(struct net_device *dev)
7452 const struct net_device_ops *ops = dev->netdev_ops;
7454 /* dev_open will call this function so the list will stay sane. */
7455 if (!(dev->flags&IFF_UP))
7458 if (!netif_device_present(dev))
7461 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
7462 /* Unicast addresses changes may only happen under the rtnl,
7463 * therefore calling __dev_set_promiscuity here is safe.
7465 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
7466 __dev_set_promiscuity(dev, 1, false);
7467 dev->uc_promisc = true;
7468 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
7469 __dev_set_promiscuity(dev, -1, false);
7470 dev->uc_promisc = false;
7474 if (ops->ndo_set_rx_mode)
7475 ops->ndo_set_rx_mode(dev);
7478 void dev_set_rx_mode(struct net_device *dev)
7480 netif_addr_lock_bh(dev);
7481 __dev_set_rx_mode(dev);
7482 netif_addr_unlock_bh(dev);
7486 * dev_get_flags - get flags reported to userspace
7489 * Get the combination of flag bits exported through APIs to userspace.
7491 unsigned int dev_get_flags(const struct net_device *dev)
7495 flags = (dev->flags & ~(IFF_PROMISC |
7500 (dev->gflags & (IFF_PROMISC |
7503 if (netif_running(dev)) {
7504 if (netif_oper_up(dev))
7505 flags |= IFF_RUNNING;
7506 if (netif_carrier_ok(dev))
7507 flags |= IFF_LOWER_UP;
7508 if (netif_dormant(dev))
7509 flags |= IFF_DORMANT;
7514 EXPORT_SYMBOL(dev_get_flags);
7516 int __dev_change_flags(struct net_device *dev, unsigned int flags,
7517 struct netlink_ext_ack *extack)
7519 unsigned int old_flags = dev->flags;
7525 * Set the flags on our device.
7528 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
7529 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
7531 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
7535 * Load in the correct multicast list now the flags have changed.
7538 if ((old_flags ^ flags) & IFF_MULTICAST)
7539 dev_change_rx_flags(dev, IFF_MULTICAST);
7541 dev_set_rx_mode(dev);
7544 * Have we downed the interface. We handle IFF_UP ourselves
7545 * according to user attempts to set it, rather than blindly
7550 if ((old_flags ^ flags) & IFF_UP) {
7551 if (old_flags & IFF_UP)
7554 ret = __dev_open(dev, extack);
7557 if ((flags ^ dev->gflags) & IFF_PROMISC) {
7558 int inc = (flags & IFF_PROMISC) ? 1 : -1;
7559 unsigned int old_flags = dev->flags;
7561 dev->gflags ^= IFF_PROMISC;
7563 if (__dev_set_promiscuity(dev, inc, false) >= 0)
7564 if (dev->flags != old_flags)
7565 dev_set_rx_mode(dev);
7568 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
7569 * is important. Some (broken) drivers set IFF_PROMISC, when
7570 * IFF_ALLMULTI is requested not asking us and not reporting.
7572 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
7573 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
7575 dev->gflags ^= IFF_ALLMULTI;
7576 __dev_set_allmulti(dev, inc, false);
7582 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
7583 unsigned int gchanges)
7585 unsigned int changes = dev->flags ^ old_flags;
7588 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
7590 if (changes & IFF_UP) {
7591 if (dev->flags & IFF_UP)
7592 call_netdevice_notifiers(NETDEV_UP, dev);
7594 call_netdevice_notifiers(NETDEV_DOWN, dev);
7597 if (dev->flags & IFF_UP &&
7598 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
7599 struct netdev_notifier_change_info change_info = {
7603 .flags_changed = changes,
7606 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
7611 * dev_change_flags - change device settings
7613 * @flags: device state flags
7614 * @extack: netlink extended ack
7616 * Change settings on device based state flags. The flags are
7617 * in the userspace exported format.
7619 int dev_change_flags(struct net_device *dev, unsigned int flags,
7620 struct netlink_ext_ack *extack)
7623 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
7625 ret = __dev_change_flags(dev, flags, extack);
7629 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
7630 __dev_notify_flags(dev, old_flags, changes);
7633 EXPORT_SYMBOL(dev_change_flags);
7635 int __dev_set_mtu(struct net_device *dev, int new_mtu)
7637 const struct net_device_ops *ops = dev->netdev_ops;
7639 if (ops->ndo_change_mtu)
7640 return ops->ndo_change_mtu(dev, new_mtu);
7645 EXPORT_SYMBOL(__dev_set_mtu);
7648 * dev_set_mtu_ext - Change maximum transfer unit
7650 * @new_mtu: new transfer unit
7651 * @extack: netlink extended ack
7653 * Change the maximum transfer size of the network device.
7655 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
7656 struct netlink_ext_ack *extack)
7660 if (new_mtu == dev->mtu)
7663 /* MTU must be positive, and in range */
7664 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
7665 NL_SET_ERR_MSG(extack, "mtu less than device minimum");
7669 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
7670 NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
7674 if (!netif_device_present(dev))
7677 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
7678 err = notifier_to_errno(err);
7682 orig_mtu = dev->mtu;
7683 err = __dev_set_mtu(dev, new_mtu);
7686 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
7688 err = notifier_to_errno(err);
7690 /* setting mtu back and notifying everyone again,
7691 * so that they have a chance to revert changes.
7693 __dev_set_mtu(dev, orig_mtu);
7694 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
7701 int dev_set_mtu(struct net_device *dev, int new_mtu)
7703 struct netlink_ext_ack extack;
7706 memset(&extack, 0, sizeof(extack));
7707 err = dev_set_mtu_ext(dev, new_mtu, &extack);
7708 if (err && extack._msg)
7709 net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
7712 EXPORT_SYMBOL(dev_set_mtu);
7715 * dev_change_tx_queue_len - Change TX queue length of a netdevice
7717 * @new_len: new tx queue length
7719 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
7721 unsigned int orig_len = dev->tx_queue_len;
7724 if (new_len != (unsigned int)new_len)
7727 if (new_len != orig_len) {
7728 dev->tx_queue_len = new_len;
7729 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
7730 res = notifier_to_errno(res);
7733 res = dev_qdisc_change_tx_queue_len(dev);
7741 netdev_err(dev, "refused to change device tx_queue_len\n");
7742 dev->tx_queue_len = orig_len;
7747 * dev_set_group - Change group this device belongs to
7749 * @new_group: group this device should belong to
7751 void dev_set_group(struct net_device *dev, int new_group)
7753 dev->group = new_group;
7755 EXPORT_SYMBOL(dev_set_group);
7758 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
7760 * @addr: new address
7761 * @extack: netlink extended ack
7763 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
7764 struct netlink_ext_ack *extack)
7766 struct netdev_notifier_pre_changeaddr_info info = {
7768 .info.extack = extack,
7773 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
7774 return notifier_to_errno(rc);
7776 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
7779 * dev_set_mac_address - Change Media Access Control Address
7782 * @extack: netlink extended ack
7784 * Change the hardware (MAC) address of the device
7786 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
7787 struct netlink_ext_ack *extack)
7789 const struct net_device_ops *ops = dev->netdev_ops;
7792 if (!ops->ndo_set_mac_address)
7794 if (sa->sa_family != dev->type)
7796 if (!netif_device_present(dev))
7798 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
7801 err = ops->ndo_set_mac_address(dev, sa);
7804 dev->addr_assign_type = NET_ADDR_SET;
7805 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
7806 add_device_randomness(dev->dev_addr, dev->addr_len);
7809 EXPORT_SYMBOL(dev_set_mac_address);
7812 * dev_change_carrier - Change device carrier
7814 * @new_carrier: new value
7816 * Change device carrier
7818 int dev_change_carrier(struct net_device *dev, bool new_carrier)
7820 const struct net_device_ops *ops = dev->netdev_ops;
7822 if (!ops->ndo_change_carrier)
7824 if (!netif_device_present(dev))
7826 return ops->ndo_change_carrier(dev, new_carrier);
7828 EXPORT_SYMBOL(dev_change_carrier);
7831 * dev_get_phys_port_id - Get device physical port ID
7835 * Get device physical port ID
7837 int dev_get_phys_port_id(struct net_device *dev,
7838 struct netdev_phys_item_id *ppid)
7840 const struct net_device_ops *ops = dev->netdev_ops;
7842 if (!ops->ndo_get_phys_port_id)
7844 return ops->ndo_get_phys_port_id(dev, ppid);
7846 EXPORT_SYMBOL(dev_get_phys_port_id);
7849 * dev_get_phys_port_name - Get device physical port name
7852 * @len: limit of bytes to copy to name
7854 * Get device physical port name
7856 int dev_get_phys_port_name(struct net_device *dev,
7857 char *name, size_t len)
7859 const struct net_device_ops *ops = dev->netdev_ops;
7862 if (ops->ndo_get_phys_port_name) {
7863 err = ops->ndo_get_phys_port_name(dev, name, len);
7864 if (err != -EOPNOTSUPP)
7867 return devlink_compat_phys_port_name_get(dev, name, len);
7869 EXPORT_SYMBOL(dev_get_phys_port_name);
7872 * dev_get_port_parent_id - Get the device's port parent identifier
7873 * @dev: network device
7874 * @ppid: pointer to a storage for the port's parent identifier
7875 * @recurse: allow/disallow recursion to lower devices
7877 * Get the devices's port parent identifier
7879 int dev_get_port_parent_id(struct net_device *dev,
7880 struct netdev_phys_item_id *ppid,
7883 const struct net_device_ops *ops = dev->netdev_ops;
7884 struct netdev_phys_item_id first = { };
7885 struct net_device *lower_dev;
7886 struct list_head *iter;
7889 if (ops->ndo_get_port_parent_id) {
7890 err = ops->ndo_get_port_parent_id(dev, ppid);
7891 if (err != -EOPNOTSUPP)
7895 err = devlink_compat_switch_id_get(dev, ppid);
7896 if (!err || err != -EOPNOTSUPP)
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 (refcnt && 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);